Sonnenhauser, B. Yet There's Method In It. Semantics, Pragmatics and the Interpretation of the...

272
Barbara Sonnenhauser Yet There’s Method In It Semantics, Pragmatics and the Interpretation of the Russian Imperfective Aspect

Transcript of Sonnenhauser, B. Yet There's Method In It. Semantics, Pragmatics and the Interpretation of the...

Barbara Sonnenhauser

Yet There’s Method In It Semantics, Pragmatics and the Interpretation of the

Russian Imperfective Aspect

SL AV I ST IS CH E BE I TR Ä GE

H e r a u s g e g e b e n v o n P e t e r R e h d e r

B e i r a t : Tilman Berger · Walter Breu · Johanna Renate Döring-Smirnov

Walter Koschmal · Ulrich Schweier · Milos Sedmidubsky´ · Klaus Steinke

Band 449

VERLAG OTTO SAGNER MÜNCHEN 2006

Barbara Sonnenhauser

Yet There’s Method In It Semantics, Pragmatics and the Interpretation of the

Russian Imperfective Aspect

VERLAG OTTO SAGNER MÜNCHEN 2006

Bibliografische Information der Deutschen Bibliothek: Die Deutsche Bibliothek verzeichnet diese Publikation in der

Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind abrufbar im Internet über

http://dnb.ddb.de/

ISBN 978-3-87690-973-8 © Verlag Otto Sagner, München 2006 Abteilung der Firma Kubon & Sagner

D-80328 München Druck und Bindung:

Strauss Offsetdruck GmbH D-69509 Mörlenbach

Gedruckt auf alterungsbeständigem Papier

ACKNOWLEDGMENTS

The present thesis was carried out at the University of Leipzig, within the Graduiertenkolleg “Universalität und Diversität: Sprachliche Strukturen und Prozesse”, funded by the Deutsche Forschungsgemeinschaft. I would like to thank Prof. Gerhild Zybatow for offering me the opportunity of joining the Graduiertenkolleg from April 2002 to March 2005 and for supervising my thesis.

In writing this thesis, I was supported by a number of people. Most of all, I want to thank Nikolai Vazov from the University of Sofia who was never tired of helping me with my examples, discussing my ideas and commenting on parts of the thesis, no matter how unfinished they were. Moreover, he introduced me to intervals and aroused my interest in computational linguistics and natural language processing. My stay at the University of Sofia was an inspiring experience, which is the merit of Krasi Petrova and all other people from the Russian department there as well.

Thanks to Kristin Börjesson, Petra Augurzky and Susann Lingel for helping me putting the text into proper English. I am grateful to Kristin Börjesson for discussing some of the main issues with me, and to Petr Biskup, who also commented on parts of the thesis and who tried to convince me of trees.

There are several other people whom I need to thank a lot. Mersi and teşekkürler to Christian for our journey to Trabzon which ended on the most beautiful balcony in Sofia. Thanks also to Bärbel Lochner who made possible the impossible and to Hans-Joachim Jäger who kept the computers going.

I enjoyed my time in Leipzig, not only because of Barcelona and Besser Leben, but also because of my colleagues from the Graduiertenkolleg (who were in fact more than just colleagues).

Most importantly, however, I got to know some friends there, who must have been the underlying reason for my happening to show up in the vast plains of Leipzig: Petra, Andreas and Torsten. München, September 2006 Barbara Sonnenhauser

TABLE OF CONTENTS Abbreviations..................................................................................................................... 9 1 INTRODUCTION ......................................................................................................... 11 2 SETTING THE STAGE ................................................................................................ 19 2.1 Analysing Aspect ................................................................................................. 19 2.1.1 The Russian perspective............................................................................... 19 2.1.2 The cross-linguistic perspective ................................................................... 30 2.2 Basic distinctions ................................................................................................. 34 2.2.1 Terminativity, boundedness and aspect ....................................................... 35 2.2.2 Aktionsart and event structure...................................................................... 41 2.2.3 Aktionsart and aspect ................................................................................... 50 2.3 Conclusion............................................................................................................ 61 3 UTTERANCE INTERPRETATION ............................................................................. 63 3.1 Underspecification of ‘meaning’ ........................................................................ 63

3.1.1 Meaning: Encoding, decoding, inferring...................................................... 64 3.1.2 Underspecification ....................................................................................... 76

3.2 Interpreting underspecified utterances ............................................................ 87

3.2.1 Semantics and pragmatics ............................................................................ 87 3.2.2 From sentence meaning to utterance meaning ............................................ 94

3.3 Conclusion.......................................................................................................... 102 4 SEMANTICS ............................................................................................................... 105 4.1 Grammatical aspect cross-linguistically.......................................................... 106

4.1.1 The common denominator ......................................................................... 107 4.1.2 The ingredients ........................................................................................... 115

4.2 Russian ............................................................................................................... 127

4.2.1 Aspect, boundedness and terminativity...................................................... 127 4.2.2 Aspect semantics and aspect readings........................................................ 141

4.3 Bulgarian............................................................................................................ 157

4.3.1 Aspect semantics and aspect readings........................................................ 157 4.3.2 Excursus: Semi-perfectivity ....................................................................... 170

4.4 Turkish ............................................................................................................... 175 4.4.1 Aspect semantics and aspect readings........................................................ 175 4.4.2 Excursus: The ‘Aorist’ ............................................................................... 187

4.5 Conclusion.......................................................................................................... 192 5 PRAGMATICS ............................................................................................................ 201 5.1 The semantic basis............................................................................................. 202

5.1.2 Encoding: Semantic information................................................................ 203 5.2.2 Decoding: Ambiguity and sense-generality ............................................... 210

5.2 Pragmatic embedding ....................................................................................... 226

5.2.1 GCI and aspect ........................................................................................... 227 5.2.2 Default, deviation and mismatch................................................................ 239 5.2.3 Semantic invariants, pragmatic scales and cross-linguistic analysis.......... 248

5.3 Conclusion.......................................................................................................... 254 6 CONCLUDING REMARKS ....................................................................................... 257 6.1 Summary ............................................................................................................ 257 6.2 Outlook ............................................................................................................... 259 References ...................................................................................................................... 265

ABBREVIATIONS

The following abbreviations will be used in the glosses1 and in the text: 1/2/3SG,PL 1st, 2nd, 3rd person singular/plural A Aorist (Bulgarian) ABL ablative case ACC accusative case AUX auxiliary CVB converb DAT dative case DEF definite FNOM factive nominal FUT future tense GEN genitive case I Imperfect (Bulgarian) IMP imperative INF infinitive INS instrumental case ipf imperfective aspect (Russian, Bulgarian) ir Turkish aspect marker (‘Aorist’2) iyor Turkish aspect marker LOC locative case mekte Turkish aspect marker NEG negation neut neutral aspect (Bulgarian) P Perfect (Bulgarian) PART partitive case (Finnish) PASS passive pf perfective aspect (Russian, Bulgarian) PRS present tense POSS possessive PRF prefix PST past tense

1 The glosses follow basically the guidelines stated in the Leipzig Glossing Rules (available at:

http://www.eva.mpg.de/lingua/files/morpheme.html) 2 The Turkish ‘Aorist’, glossed by ‘ir’ is not to be confused with the Bulgarian Aorist, glossed by ‘A’.

10

PTCP participle Q question SUFF suffix ϕdyn dynamic phase (encoded part of the denoted event) ϕstat static phase (encoded part of the denoted event) τ boundary (encoded part of the denoted event) I(e) event time interval TO točka otsčeta (‘standpoint of the observer’) TSit situation time TT/I(TT) topic time / topic time interval TU time of utterance T-Val/I(T-Val) validation time / validation time interval Some further remarks on the notation: Turkish suffixes are glossed ignoring vocal harmony. That is, ‘i’ in the glosses of the

aspect markers comprises its variants ‘i’, ‘ı’, ‘ü’, ‘u’, and ‘e’ in the glosses comprises the variants ‘e’ and ‘a’.

The Turkish aspect markers -iyor, -mekte and -ir are not glossed in order to avoid misleading commitments that may arise with glosses like ‘pf’, ‘ipf’ (which are reserved for Russian and Bulgarian only) or ‘progressive’ (which is reserved for English).

Glosses in SMALL CAPITALS are used basically for pure ‘grammatical’ suffixes that are not discussed in this thesis; the semantically relevant aspect suffixes are glossed in lower case letters.

If a single object language element is rendered by several meta-language elements, these are separated by periods (‘.’) in case of semantic elements, and by colons (‘:’) in case of grammatical elements.

Object-language elements appear in italics, the respective translations in ‘single quotes’.

1 INTRODUCTION

This thesis is concerned with the interpretation of grammatical aspect (perfective and imperfective1) in Russian. And it is concerned with the general issue of how semantics and pragmatics interact in utterance interpretation. It thus addresses Slavists interested in a more general embedding of their traditional topic of aspect, as well as linguists interested in the application of the general topic of the semantics/pragmatics interface onto one specific case.

Aspect in Russian, especially the ipf, is profoundly underspecified and allows for a range of possible interpretations. An investigation of this phenomenon with respect to the division of labour between semantics and pragmatics promises interesting insights in how their interaction proceeds. Vice versa, the determination of the interaction of semantics and pragmatics in interpreting Russian aspect helps to clarify the issue of aspect semantics. Levinson (2000a: 168) puts this mutual impact as follows: “Semantics and pragmatics can interleave in ways that are probably controlled by the constructional types in the semantic representation.”

The reasons for why one might refrain from using pragmatic principles as part of a systematic account of aspect interpretation2 are largely based on misconceptions about what pragmatic theories are concerned with and what they have to achieve. One of the predominant reasons to reject pragmatic principles as part of linguistic theories consists in their alleged lack of systematicity and predictive power. For quite some time, pragmatics has been conceived of as ‘meaning minus truth conditions’ (e.g., Gazdar 1979), assuming that semantics is concerned with truth conditions and truth conditional content, while pragmatics deals with ‘the rest’. More recent approaches, however, treat pragmatics on an equal footing with semantics and presume both to work hand in hand in utterance interpretation. The prerequisite for this cooperation is a conception of semantics that allows for the application of pragmatic principles, and a conception of pragmatics in terms of systematic, predictable inferences that are independent of specific contexts and specific intentions. This design of the interaction between semantics and pragmatics discards assumptions of pragmatics being some hardly describable additional mechanism that applies after semantics has done its job.

Pragmatics plays a crucial role already at the subpropositional level and is necessary to establish truth-conditional propositions. As regards aspect interpretation, much of what traditionally has been assumed to be part of semantics is in fact better accounted for in pragmatic terms. This thesis can thus very roughly be located close to the research tradition of Radical Pragmatics: “Radical pragmatics is the hypothesis that many linguistic phenomena, which had previously been viewed as belonging to the semantic subsystem, in fact belong to the pragmatic subsystem.” (Cole 1981: xi)

The basic pragmatic principles will be argued to consist in default inferences based on heuristics concerning regularities in language use and interpretation. Levinson (1995; 2000a,b) proposes a third layer of meaning that is located between the meaning encoded by linguistic expressions and inferences to the intended speaker-meaning. This is the layer of utterance-type meaning which constitutes the basis for systematic inferences. These inferences are based on heuristics both speaker and hearer are mutually aware of. The principles operating at this layer are generalised conversational implicatures (GCI), that is, context-independent inferences helping to specify and enrich the encoded information to the proposition expressed. GCI are especially interesting if applied to paradigmatically opposed

1 Henceforth abbreviated as ‘pf’ (perfective) and ‘ipf’ (imperfective). 2 This is in fact a shortcut for ‘interpretation of utterances containing a verb specified for aspect’.

Chapter 1: Introduction

12

function words or morphemes, which are oftentimes notoriously underspecified and thus pose considerable problems for purely semantic analyses. The application of GCI to the interpretation of Russian ipf aspect facilitates the establishment of a reliable semantic basis for this form. If pragmatic inferences are not mistaken for semantics, the semantics of the Russian ipf aspect, of Russian aspect in general and the semantics of aspect in a cross-linguistic perspective can be stated in reliable terms. It is mistaken to assume that either semantics or pragmatics has to account for every detail of utterance interpretation. On the one hand, this would make semantics too specific to guarantee stability and reliability in interpretation, whereas on the other hand, if pragmatics is assumed to do everything, it loses its predictive power and is turned into a mere ad hoc mechanism. Both semantics and pragmatics need to supplement each other on an equal footing.

The predictive power of pragmatic theories crucially depends on the semantic analysis of the respective expressions, that is, a pragmatic theory needs to be related to a semantic theory (and vice versa). To a certain extent, pragmatics can be said to constrain shape and content of the underlying semantic representations. These representations need to be stated in a way that provides the basis and frame for the inferences to be drawn, and do not have to represent more than what is encoded by the respective expressions. Accordingly, the semantics of aspect has to be elaborated and represented in a way that allows for GCI to apply. ‘Semantics’ pertains to the stable, invariant information encoded by the aspect marker and by the verbal basis it applies to. Additional meaning components are accounted for in pragmatic terms. Such an approach establishes a basic semantic skeleton for grammatical aspect and thus allows for a systematic cross-linguistic analysis and comparison of different aspect categories.

Although aspect is one of the traditional topics of Slavic linguistics, a semantic analysis of this category still poses a considerable challenge. The basic problems for an account of Russian aspect are summarised in the following quotation: “Verbs in Russian belong to one of the two aspect categories, the perfective or the imperfective. Although there is some variation in their morphological expression, these categories can be described as morphologically encoded aspect. They are used to express a number of partially distinct semantic features such as durativity, iterativity, progressivity, completion, and the like. Each use of a particular aspect to express one of these semantic features defines a contextual variant of that aspect.” (Timberlake’s 1982: 305)

One of the many problems in dealing with grammatical aspect in Russian is that its two instantiations – pf and ipf aspect – are not necessarily morphologically encoded. In Russian, there is no consistent and unambiguous overt aspect marking. Furthermore, there is no consistent semantic feature, no prima facie semantic invariant meaning for pf and ipf aspect that could be taken as basis for their different possible interpretations. These different aspect readings, the contextual variants, are a further problem in dealing with aspect interpretation. They constitute one important centre of aspectological debate – the number of contextual variants and sub-variants proposed is still increasing, as well as the number of proposals how to name them. This is evident particularly with the ipf aspect, which is notorious for its range of possible interpretations and which is thus especially hard to capture in terms of one semantic basis. Notions such as iterativity, durativity, progressivity and completion used to state the semantics of the ipf aspect are in fact highly misleading as they actually pertain to the level of interpretation and are not part of the semantics of aspect. The questions that arise from Timberlake’s statement above are thus the following: First, how is it possible for every verb to be assigned one aspect value although there is no consistent overt marking? Second, how can the semantics of aspect be captured in a way that is stable and at the same time flexible enough to allow for the different contextual variants? Third, is there a way to systematically account for aspect interpretation?

The starting point for the account developed in this thesis are the various different readings that are possible for aspect in Russian. This is especially evident with the ipf aspect, which

Chapter 1: Introduction

13

will be in the centre of interest. Some of the readings that are traditionally postulated for it are illustrated in ( 1.1a-f): 1.1 a. ACTUAL-PROCESSUAL READING

Šar medlenno podnimal’sja. balloon slowly ascend:PST:ipf

‘The balloon was ascending slowly.’

b. INACTUAL READING Ran’še on rabotal v universitete. in.the.past he work:PST:ipf at university:LOC. ‘He used to work at university.’ (= ‘He was a teacher at university.’)

c. GENERAL-FACTUAL READING Da, ja čitala rasskaz Dama s sobačkoj. yes I read:PST:ipf novel Dama s sobačkoj Yes, I have (already) read the novel Dama s Sobačkoj. d. DURATIVE READING Ona dolgo smotrela na fotografii detej.

she long.time look:PST:ipf at photographs:ACC children:GEN ‘She looked at the children’s photos for a long time.’

e. HABITUAL READING Ja chorošo vspominajo o deduške: on guljal so I well remember:PRS:1Sg:ipf grandpa:LOC he take.a.walk:PST:ipf with

vnukami, s nimi igral v futbol, kuril trubku,... grandschildren, with them play:PST:ipf football, smoke:PAST:ipf pipe, ... ‘I remember grandpa very well: he used to take a walk with the grandchildren, to play football with them, to smoke a pipe, ...’

f. POTENTIAL READING Ja mnogo trenirovala i podnimala 100 kg. I much train:PST:ipf and lift:PST:ipf 100 kg ‘I trained a lot and could lift 100 kg.’

g. ATEMPORAL READING Železo tonet v vode.

iron sink:PST:ipf in water:LOC ‘Iron sinks in water.’

Quite commonly, the Russian ipf aspect is defined as expressing an incomplete action or an ongoing process. Facing the general-factual reading in ( 1.1c) or the potential and atemporal reading in ( 1.1f,g), it is evident that such definitions are not tenable – neither does the ipf aspect have to express an incomplete action or an ongoing process, nor does it always introduce an individualised event into the discourse.

A further problem is that these interpretations are defaults that can be overridden by the following context:

1.2 a. Moj otec govoril po-turecki.

my father speak:PST:ipf Turkish My father could speak Turkish.

b. Moj otec govoril po-turecki, kogda ja vošel v komnatu my father speak:PST:ipf Turkish, when I come.in:PST:pf into room:ACC My father was speaking Turkish, when I came into the room.

Chapter 1: Introduction

14

Such examples support the assumption that the various readings of the ipf aspect cannot be part of its semantics. Semantics is the stable part of the information conveyed by lexical items, that is, the explicitly encoded information. Since these readings are context-dependent and defeasible, they are not part of the encoded information, but arise due to information that has to be inferred. Thus, a semantic account of the ipf aspect on the one hand has to avoid making use of notions that pertain to interpretation, and on the other hand has to incorporate enough flexibility to licence the various interpretations. Accounting for the different interpretations is the task of pragmatics, not that of semantics.

The aim of this thesis is to develop an analysis of Russian aspect that on the one hand captures the invariant meanings, i.e. the basic semantics of pf and ipf, and on the other hand allows for a systematic account of the various readings that are possible for one aspectual form.3 Everybody looking for answers to the more or less ‘traditional’ questions concerning Slavic, or Russian, aspect might be disappointed. Questions concerning the morphology of aspect, the question of aspectual pairs and the delimitation of aspect from sposoby dejstvija (‘modes of action’) are not central here and are dealt with only as far as necessary for establishing the semantic and pragmatic framework.

Definitions of aspect semantics tend to mistake interpretations for semantics or are stated in rather metaphorical terms that do not get to the core, i.e., to the encoded information. The strict separation of semantics and interpretation on the one hand, and aspect and Aktionsart4 on the other hand, allows for a semantic analysis of aspect that is uniform for different languages, as will be demonstrated on the basis of the Russian, Bulgarian and Turkish aspect systems. The cross-linguistic differences in the interpretation of aspectual forms can be accounted for in terms of language-specific markedness relations and oppositions giving rise to different pragmatic inferences. Acknowledging the interaction of semantics and pragmatics accounts for the derivation of the various aspectual readings in a systematic and predictable way and avoids mere enumeration of possible interpretations.

Traditional accounts of aspect in Russian predominantly deal with morphological questions (derivation vs. inflection), questions concerning the status of specific verbs (aspect partners vs. mode of action verbs) or questions as to which verbs constitute ‘real’ aspectual pairs (pairs of verbs that differ only in their aspect feature). Another prominent domain is the question of how many and which readings can be attributed to the pf and the ipf aspect, and which of these interpretations are core and which are secondary. Syntactic issues, and the interaction of grammatical aspect with other verbal and nominal categories (e.g., tense or definiteness) are dealt with as well. What is still lacking, though, is an approach that equally incorporates semantics and pragmatics and accounts for semantics and interpretation of aspect. This involves not only elaborating the semantics of aspect but also stating the pragmatic factors that are involved in specifying and enriching this semantic basis in order to yield the respective interpretation. This rather complex can hardly be achieved within one single theoretic framework. As it is this overall picture that is at issue here, contributions from several distinct approaches will be taken into account and be combined with respect to that overall goal.

In order to state the semantics in an appropriate way, the contributions of aspect and verbal basis have to be separated. Since these contributions interact in the actual interpretation process, they need to be compatible in several respects. Verbal basis and grammatical aspect encode different kinds of information – conceptual and procedural (Blakemore 2002). The difference between the lexical entries of aspect and verbal basis consists in the fact that aspect lacks encyclopaedic information and delivers merely logical information. The logical

3 The notion of ‘aspectual form’ (also ‘aspectualised verb’) will be used to refer to the combination of aspect

and the predicate it applies to. 4 ‘Aktionsart’ refers to the classification of the verb/VP aspect is applied to and is not used in the traditional

Slavistic sense as referring to different modes of action (chapter 2).

Chapter 1: Introduction

15

information is invariant and stable across contexts and thus guarantees compositionality, whereas the flexibility in interpretations is to be attributed to the encyclopaedic information activated by the predicates5. Since the Aktionsart basis also provides logical information, the contributions of aspect and Aktionsart interact in terms of this logical information.

The semantics of aspect, i.e. the encoded information the respective marker contributes to the interpretation process, will be captured along the lines of Klein’s (1995, 2001) general account of aspect in time-relational terms, which will be supplemented by Bickel’s (1996, 2000) selection theoretic approach. The combination of both frameworks allows for a cross-linguistic definition of aspect in terms of intersecting intervals whose features (boundary characterisation and internal specification) are language-specifically modified in predictable ways. The thus specified semantics constitutes both basis and frame for possible interpretations.

The semantic analysis for aspect will be embedded in the pragmatic framework provided by Levinson’s (2000a) GCI theory, since this framework allows to systematically capture the inference processes started off by the semantic information. The main questions and aims of this thesis can be summarised as follows: (a) What is the common denominator that justifies the subsumption of certain linguistic

phenomena under the label of ‘grammatical aspect’? If certain phenomena are subsumed under one common heading, it is reasonable to assume that they have certain characteristics in common. With respect to grammatical aspect, the question arises whether there exists a semantic basis that is common to this category in whatever language it shows up. In chapter 4, a cross-linguistically applicable semantic characterisation of grammatical aspect will be established, based on distinctions that will be introduced in chapter 2. This general semantics of aspect will be argued to have different language-specific instantiations. It will be shown that the general semantics of aspect can be captured by a set of basic features and relations that are language-specifically modified in certain predictable and constrained ways. The analysis of grammatical aspect in Bulgarian and Turkish will justify the basic semantic components postulated for Russian, and illustrate the cross-linguistic validity of the general analysis proposed.

An invariant semantic basis can not only be established for aspect cross-linguistically, but also for the language-specific instantiations of aspect and the respective markers. As regards the ipf aspect in Russian, it is hard to state such an invariant meaning component. However, by taking into account the pragmatic notion of scalar implicature, the Russian ipf aspect can be assigned such an invariant meaning as well. Chapter 5 will illustrate the pragmatic reasoning involved in specifying this rather general ipf meaning to yield a specific interpretation. (b) How do aspect and verbal basis (Aktionsart) interact? Aspect is a grammatical category of the verb and is therefore not interpreted in isolation. Dealing with the interpretation of aspect thus actually means dealing with the interpretation of an ‘aspectualised verb’, or even more precisely, with the interpretation of an utterance containing an aspectualised verb. This is not to claim that the task of figuring out the semantics of aspect is redundant – quite the contrary. It is exactly because aspect and Aktionsart do not appear in isolation and are not interpreted separately that the semantic contributions of both needs to be specified. This proves crucial not only for a systematic account of aspect in Russian, but also with respect to a cross-linguistic analysis. If grammatical aspect is to be ascribed a cross-linguistically valid semantics, its contribution

5 This comes close to Sag’s (1981) view that the principle of composition remains free of pragmatics,

whereas the meaning of the terms combined is essentially pragmatic.

Chapter 1: Introduction

16

needs to be separated from that of the Aktionsart. Chapter 3 will set the stage for such an analysis of aspect and Aktionsart semantics by arguing that both encode different kinds of information – procedural (aspect) and conceptual (verb) – and by specifying the place of interaction and the common properties that make that interaction possible. This level of interaction will be important in the analysis of aspect semantics and aspect readings in chapter 4, and in accounting for the interpretation process in pragmatic terms in chapter 5. (c) Facing the variety of possible interpretations for the Russian ipf aspect – is it an instance

of multiple ambiguity or is there one invariant meaning basic to all of its uses? The ipf aspect in Russian is informationally less specific than the pf aspect and therefore allows for a considerable range of possible interpretations. Basically, one could assume one of two possibilities to account for this: postulate the ipf aspect to be multiply ambiguous between various core interpretations, or claim it to have one highly general meaning that needs to be specified to one particular interpretation in a given context. This thesis will argue for an account that, to a certain degree, combines both possibilities. This is feasible by making use of the concept of pragmatic ambiguity (Lascarides et al. 1996), which captures the intuition that ambiguity effects (e.g. zeugma effects, impossibility of crossed-readings) may be caused not only by one lexeme that can be ascribed multiple lexical entries. Conflicting default interpretations may just as well give rise to such effects for one single lexeme with one single entry. Applied to the Russian ipf aspect, this will amount to postulating one general semantic basis that is pragmatically ambiguous between three specifications. This indicates that pragmatic processes are already needed to specify the semantic basis, i.e. to establish the semantic representation that is the basis for the context specific interpretation. The specific semantic representation still underspecifies the final, contextually relevant interpretation. To arrive at the final interpretation, further pragmatic inference processes are necessary. Accordingly, the ipf aspect is underspecified in two respects: it is pragmatically ambiguous with respect to its semantic representation, and it is sense-general with respect to the final interpretation. In order to solve both instances of underspecification, an account in purely semantic terms is not sufficient. It rather has to be embedded in a pragmatic framework. The respective pragmatic processes need to be systematic and predictable from the semantic basis. This combination of semantic basis and pragmatic inferences explains the derivation of the different readings listed above. Chapter 4 elaborates the semantics of aspect, the pragmatic principles and their interaction with the semantics will be introduced in chapter 5. (d) What is the status of the various aspect readings? In traditional accounts of Russian aspect, the different readings that are possible with pf and ipf aspect play an important role. Both aspects seem to be defined in terms of their readings, or at least in terms of some alleged core interpretation(s). Both number and label of the suggested readings differ across different studies and different authors. Here, it will be argued that it is neither possible nor necessary to list all the interpretations that might be possible. These readings do not constitute different and discrete meanings that could be stated in form of a list from which the relevant one must be chosen, as is the case with ambiguous expressions. Ambiguity is involved in the interpretation of aspect, but it is encountered at the semantic level and not at the level of interpretation where the different readings are located. Instead of being selected from a predefined list, the readings are constructed in an utterance according to the semantic basis and according to context. Although a stock of readings will be used in the analysis, this is not to imply that they ‘exist’ in the sense of clearly distinguishable, separate meanings. These readings are used merely for the sake of the argument.

Chapter 1: Introduction

17

(e) Can the cross-linguistic differences in aspect use and interpretation be accounted for in a systematic, predictable way?

It will be shown that a systematic account for the cross-linguistic differences of grammatical aspect is possible. The semantic analysis of grammatical aspect in general will make use of a basic toolbox of features and relations. These features and relations are language-specifically instantiated in predictable and constrained ways and thus yield a language-specific characterisation of grammatical aspect based on and derived from the general picture. The suggested pragmatic principles are cross-linguistically valid as well, since the respective processes pertain to reasoning about what rational speakers and hearers would understand by putting things in a certain way. Structure and encoded meaning of ‘what is said’ gives rise to further, presumptive meanings (Levinson 2000a) based on the knowledge of observed regularities of how language is used and understood. Crucially, the pragmatic reasoning processes do not substitute the semantic basis, but provide the embedding for it.

The basic semantic and pragmatic framework will be argued to be cross-linguistically valid. It therefore serves as a fairly abstract skeleton that has to be fledged out by language-specific lexical and grammatical features. By relating aspect categories of different languages to this abstract skeleton, the language-specific peculiarities and, consequently, the cross-linguistic differences can be accounted for, and in fact can be derived in both a principled and systematic way. The semantics of the respective aspect operators is quite stable and invariant across languages, the peculiarities arise from different markedness relations and different possibilities of morphological encoding. These differences give rise to pragmatic inferences that are cross-linguistically valid but differ in their exact outcomes according to the language-specific specifications of the general semantic framework. The different outcomes of the pragmatic processes explain the differences in usage and interpretation of the respective forms. The questions discussed above pertain to the phenomenon of grammatical aspect and its interpretation. This specific issue is embedded in the general framework of linguistic underspecification and the contribution of semantics and pragmatics to resolving it. The basic question and thread of this thesis thus is the following: (f) How do semantics and pragmatics interact in the interpretation of underspecified

linguistic expressions? The main concern of this thesis is the elaboration of a systematic analysis of aspect interpretation. This elaboration crucially relies on a separation of encoded information (semantics) from inferred information (pragmatics) and on an account of how both interact. Chapter 3 will be concerned with the general question of how to distinguish semantics and pragmatics and of how to specify their interaction. In chapter 5, the semantic analysis of aspect (chapter 4) will be embedded in the pragmatic framework developed by Levinson (2000a), building on his basic insight that the predictive power of pragmatic theories crucially depends on the semantic analysis of the respective lexical items. The analysis of aspect interpretation will provide further evidence for the claim that semantics and pragmatics work hand in hand in utterance interpretation and that pragmatics plays a crucial role already at the subpropositional level in establishing the semantic representation and the proposition expressed. This thesis is structured as follows: Chapter 2 presents previous approaches to aspect in Russian and to aspect in a cross-linguistic perspective, pointing out the problems these approaches are confronted with. The basic distinctions for the semantic analysis of aspect will be outlined and it will be shown that aspect systems may differ with respect to which level of

Chapter 1: Introduction

18

the verbal basis the respective markers are sensitive to. Chapter 3 is concerned with the question of how underspecified linguistic expressions can be interpreted in a predictive and reliable way. This bears on the tradeoff between stability and flexibility involved in language use and interpretation. The basic assumptions on semantics and pragmatics and their respective role and division of labour in the interpretation process will be outlined. Moreover, the difference between conceptual and procedural information will be introduced, as well as that between decoding and inferring, the two modes of information gathering that play a role in utterance interpretation. Chapter 4 develops a cross-linguistic semantic analysis of grammatical aspect and applies it to Russian, the main focus being on the ipf aspect. Bulgarian and Turkish will serve to illustrate the validity of the claims made for Russian on the one hand, and the cross-linguistic applicability of the proposed analysis on the other hand. Chapter 5 focuses on the ipf aspect in Russian and illustrates the need for a pragmatic embedding of the semantic basis in order to account for its interpretation. This form is highly underspecified with respect to its semantics and its interpretation and thus is hard to capture in systematic terms. It will be shown that pragmatic reasoning plays an important role at every level of interpretation. Moreover, it will be illustrated how the consideration of pragmatics facilitates a cross-linguistic account of aspect.

A caveat at the beginning: Since the aim is to specify the interaction of semantics and pragmatics, a number of issues cannot be dealt with in sufficient detail. With respect to Russian, this pertains not only to morphological and syntactic issues (that are not dealt with at all in this thesis), but also to more specific semantic topics like the general-factual use of the ipf aspect, or the expression of iterativity/habituality. The interaction of aspect with other verbal or nominal categories, such as tense, mood, voice, definiteness, etc., are addressed only peripherally. The same holds for the analyses of the Bulgarian and Turkish aspect systems. Both languages would need and deserve a more extensive, more detailed treatment which cannot be delivered within the scope of this thesis. However, this does not have a bearing on the overall goal which is to provide a first outline of how semantics and pragmatics interact in the interpretation of aspect in general and of ipf aspect in Russian in particular.

To put it in a nutshell, this thesis will make a strong cause for semantics and pragmatics working in hand, interacting and constraining each other.

2 SETTING THE STAGE

This chapter will set the stage for an analysis of aspect that is able to capture the Russian as well as the cross-linguistic data.

In section 2.1, some problems connected with most traditional Slavic aspect definitions will be illustrated. These problems are basically related to the confusion of semantics and interpretation, to metaphoric definitions and to listing readings instead of fixing the semantic basis. Problems also arise with respect to cross-linguistic analyses of grammatical aspect if one of these ‘definitions’ of Slavic (or even more specifically, Russian) aspect is taken as the general definition of the category as such. This then amounts to postulating either more and more aspect categories (progressive, habitual, perfect, etc.) or to ignoring aspect oppositions in languages that do not conform to the Russian picture1.

Section 2.2 will clarify some basic terminological and notional issues. In section 2.2.1, terminativity, boundedness and aspect will be distinguished. Terminativity and boundedness will be argued to constitute the features that characterise the Aktionsart2 basis aspect is applied to. The make-up of Aktionsarten and their conceptual basis will be discussed in section 2.2.2. The need for a classification not only of VPs but also of verbs according to specific features will be supported in section 2.2.3, where it will be shown that aspect markers of different languages are sensitive to different levels of Aktionsart composition. That section will also emphasise the crucial distinction between prefixes as eventuality3 type modifiers and grammatical aspect as propositional operator.

2.1 Analysing Aspect

Section 2.1.1 illustrates some of the difficulties Russian aspect, especially the ipf, provides for a semantic analysis. Several previous analyses will briefly be outlined and each will be argued to suffer from different problems. Some approaches seem to confuse semantics and interpretation, others characterise the ipf aspect in simply negative terms. There are also proposals that make use of the Reichenbachian system of event time, speech time and reference time. Such approaches are equally inadequate, as is the attempt to define the ipf and pf aspect by listing their possible interpretations.

Section 2.1.2 adopts a cross-linguistic perspective, where one is confronted with the fact that aspect categories of different languages differ to a considerable degree but still have too much in common to be regarded as completely different categories. This affords an analysis of aspect that captures the language-specific data as well as the cross-linguistic facts.

2.1.1 The Russian perspective Among students and linguists, Russian is notorious for its grammatical category of aspect. Verbs in Russian are either pf or ipf. A number of verbs form ‘aspectual pairs’, i.e. pairs of verbs that differ only in their aspect feature, as illustrated in ( 2.1):4

1 Compare, e.g., the discussion on aspect in Turkish outlined in Johanson (1971: 47-49). 2 The notion of Aktionsart refers to the classification of predicates (section 2.2). 3 The term ‘eventuality’ subsumes all kinds of entities that may be denoted by a predicate: states, events,

processes, points, and the like. 4 There exists also a group of verbs which are only pf or ipf, and a group of verbs that are both pf and ipf.

Chapter 2: Setting the Stage

20

2.1 rešit’ (pf) – rešat’ (ipf): ‘to solve’; dat’ (pf) – davat’ (ipf): ‘to give’; delat’ (ipf) – sdelat’ (pf): ‘to do’; pisat’ (ipf) – napisat’ (pf): ‘to write’

Aspect assignment is independent of the existence of an aspect partner and the concept of aspectual pair. The analysis of aspect in Russian is further complicated by the fact that formal aspect marking is not consistent as is the case, e.g., in English: 2.2 a. delat’ – sdelat’

do:ipf:INF – PRF:do:pf:INF ‘to do’

b. otkryt’ – otkryvat’ open:pf:INF – open:SUFF:ipf:INF

‘to open’ 2.3 a. to do – to be doing

b. to open – to be opening In ( 2.2a) the pf verb is formally marked (prefix s-) as compared to the ipf simplex verb, whereas in ( 2.2b) it is the ipf verb that is formally marked (suffix -yva- ), while the pf verb remains unmarked. This contradicts assumptions that simplex verbs are ipf, with only few exceptions (cf. section 2.2.3). The English system (as well as the Turkish one, see below), however, consistently marks the progressive aspect by means of one unambiguous marker -ing.

In Russian, there is no unambiguous overt pf or ipf maker. Moreover, grammatical aspect does not even need to be overtly expressed: 2.4 a. rešit’

solve:pf:INF ‘to solve’

c. pisat’ write:ipf:INF ‘to write’

b. čitat’ read:ipf:INF ‘to read’

d. perepisat’ PRF:write:pf:INF ‘to copy’

e. perepisyvat’ PRF:write:SUFF:ipf:INF

‘to copy’ Neither ( 2.4a) nor ( 2.4b) are overtly marked for aspect. Still, both differ in their aspect, as becomes evident in ( 2.5) and ( 2.6). Combined with non-past morphology, rešit’ cannot refer to the present tense, compare ( 2.5a). Furthermore, it cannot be combined with phasal verbs, as shown in ( 2.5b). Both facts indicate that rešit’ is perfective. 2.5 a. Ja rešu zadaču.

I solve:PRS:pf exercise:ACC *‘I am solving the exercise.’5 ‘I will solve the exercise.’

5 Note that this example is not about the impossibility of a progressive interpretation with Russian pf aspect.

To refer to the actual, ongoing present tense, English has to use the progressive form as the simple form would denote habituality.

Chapter 2: Setting the Stage

21

b. *Ja načal rešit’ zadaču. I begin:PST:pf solve:pf:INF exercise:ACC ‘I began to solve the exercise.’

Both, reference to the present tense and combination with phasal verbs is possible with čitat’, compare ( 2.6a,b). This indicates, that čitat’ is imperfective. 2.6 a. Ja čitaju gazetu.

I read:PRS:ipf newspaper:ACC ‘I am reading the newspaper.’

b. Ja načal čitat’ gazetu. I begin:PST:pf read:ipf:INF newspaper:ACC ‘I began to read the newspaper.’

From the fact that pisat’ in ( 2.4c) is ipf and perepisat’ in ( 2.4d) is pf, it cannot be concluded that the prefix pere- is an overt marker of perfective aspect. If it were, the ipf perepisyvat’ in ( 2.4e) would not be possible. Furthermore, pere- is not the only prefix that may be attached to simplex ipf verbs. The problems of aspect assignment in the absence of consistent and overt marking and the status of prefixes will be dealt with in section 2.2.3.

With aspect in Russian, it is not only the formal side that causes problems for an analysis. Due to the multitude of possible interpretations, the semantic side is problematic as well, as it is hard to establish a characterisation of aspect in terms of one definite semantic meaning for pf and ipf aspect, respectively. When it comes to defining aspect in Russian, one is confronted with rather metaphorical definitions (for a critical analysis see, e.g., Klein 1995) that do not capture the data adequately, as will be illustrated with some examples. This is by no means intended as an exhaustive critical overview of different aspect theories, nor does it justice to the overall value of the theories mentioned.

It is very common to define the pf aspect in terms of notions such as ‘completedness’, ‘terminativity’, ‘totality’ or ‘resultativity’ of an action, or to define it as denoting finished actions. However, these definitions are in fact rather descriptions of specific interpretations. They do not arise with every application of the pf aspect and thus cannot be taken to constitute its semantics.

Not all actions denoted by a pf verb are finished in the sense that a final limit is described as being reached. Ingressive and inceptive verbs encode an initial boundary and not a final one, compare ( 2.7a,b). The same argument holds against the term celostnost’ (‘completedness’) that likewise puts too much emphasis on the final limit of an action. Predel’nost’ (‘limitedness’) corresponds to telicity/terminativity6 and this cannot be equalled with perfectivity, as atelic predications may also appear in the pf aspect ( 2.7c):

2.7 a. On zapel.

he PRF:sing:PST:pf ‘He started to sing.’

b. Vdrug, ja ego uvidel. suddendly, I him PRF:see:PST:pf ‘Suddenly, I caught sight of him.’

6 For the notion of terminativity see section 2.2.1, where it will be shown that perfectivity and

telicity/terminativity have to be kept apart.

Chapter 2: Setting the Stage

22

c. Ona prostojala na ugle celyj čas. she PRF.stand:PST:pf at corner:LOC whole hour ‘She stood at the corner for a whole hour.’ The pf aspect cannot be characterised in terms of resultativity either, because not all pf verbs denote a resultant state. A resultative interpretation may arise with the so-called ‘perfect reading’ of the pf aspect, which is possible only with verbs that lexically specify a resultant state. Resultativity is thus dependent on the lexical entry of the verb. As prefixation is one of the means to lexically specify a resultant state, and as such prefixed verbs are mostly pf, this might have led to the conflation of resultativity and perfectivity7.

In example ( 2.8a) a perfect reading is possible and the result – the broken arm – is asserted to hold at the time the utterance is made. This is different in ( 2.8b), where the consequent state (being at Voronež) is not asserted to hold at the time of utterance, but at some lexically specified reference time, namely ‘May 1962’. This time adverbial gives rise to a pluperfect interpretation.

2.8 a. Ja slomal ruku.

I break:PST:pf hand:ACC [Padučeva 1996: 294] ‘I broke my hand.’

b. V mae 1962 ja priechal v Voronež. in May 1962 I arrive:PST:pf in Voronež [ibd.]

‘In May 1962 I had arrived at Voronež.’ In cases where the perfect reading would be possible, it can be suppressed by the following context. Thus, prišel in ( 2.9) has an event reading and not a perfect reading: 2.9 Ivan prišel v 10 časov i srazu leg spat’.

Ivan arrive:PST:pf at 10 o’clock and at.once lay:PST:pf sleep:INF:ipf [Breu 1998: 58] ‘Ivan arrived at 10 o’clock and immediately went to bed.’

Resultativity thus cannot be taken as the semantics of the pf aspect. The Russian pf aspect does not focus solely on the resultant state but crucially includes the change into that resultant state. The mere assertion of a resultant state is restricted to imperfectivity – if the ipf aspect is applied to verbs that lexically encode a resultant state. This is the case with inchoative verbs that encode a preparation phase, a culmination point and a consequent state8 as in ( 2.10), or with inceptive verbs encoding an initial culmination point and a subsequent resultant state as in ( 2.11): 2.10 a. Olen’ sprjatalsja ot ochotnikov v vinogradnik.

dear hide:PST:pf from hunters in vineyard [Breu 1998: 69] ‘The dear hid from the hunters in the vineyard.’

b. (Ochotnik videl, kak) Olen’ prjatalsja. (hunter see:PST:ipf, how) dear hide:PST:ipf ‘(The hunter saw, how) The dear was hiding.’

c. (Nikto ego ne videl, potomu čto) On prjatalsja pod stolom. (Nobody him Neg see:Pst:ipf because) he hide:Pst:ipf under table:INS ‘(Nobody saw him, because) He hid under the table.’

7 The issue of prefixed verbs as well as the status of prefixes with respect to aspect will be dealt with in

section 2.2.3. 8 For these notions compare section 2.2.2.

Chapter 2: Setting the Stage

23

2.11 a. Ja ponjal na čto on namekal. I understand:PST:pf to what he allude:PST:ipf [Padučeva 1996: 156] ‘I got to understand what he was alluding to.’

b. Ja ponimala na čto on namekal. I understand:PST:ipf to what he allude:PST:ipf [ibd.] ‘I understood what he was alluding to.’ (= I was aware of what he was alluding to)

The pf aspect in ( 2.10a) focuses on the change-of-state from non-hiding to hiding, whereas the ipf aspect may either focus on the process of hiding as in ( 2.10b), or on the state of being hidden as in ( 2.10c). The difference between the perfect reading of the pf verb in ( 2.11a) and the interpretation of the ipf verb ( 2.11b) asserting the resultant state9 is the following: both the action and the transition from terminating that action to starting the consequent state is part of what is asserted by the pf aspect, whereas with the ipf aspect both components are presupposed (Padučeva 1996: 156), and what is asserted is the consequent state.

Thus, the above mentioned characterisations of the pf aspect do not seem to capture its semantics. Another definition is provided by C. Smith (1997) who characterises the pf aspect as ‘informationally closed’. However, this is difficult to accommodate with the perfect reading, which asserts the validity of the consequent/resultant state at the time of utterance. Comrie (1976) defines the pf aspect as ‘viewing the situation as completed whole’. The problem this definition shares with Smith’s is that both do not clearly distinguish between the event and the situation it is central to (cf. section 2.2.1).

Filip (1999: 184) defines the pf aspect in terms of the feature [+TOT] (totality):

2.12 [Perfective φ] denotes events represented as integrated wholes (i.e., in their totality, as single indivisible wholes).

The question arising from this definition is, where this totality-reading comes from and whether totality is really the constitutive feature of the semantics of pf. Compare example ( 2.13): whereas in ( 2.13a), the totality-interpretation for the pf aspect is rather straightforward, it is difficult to get with ( 2.13b): 2.13 a. On rešil zadaču.

he solve:PST:pf exercise:ACC ‘He solved the exercise.’

b. Ona posidela v parke. she PRF:sit:PST:pf in park:LOC ‘She sat in the park (for some while).’ In fact, ‘totality’ is a cover term for two distinct notions – finishing and completing – that have to be kept apart in aspectual analysis. If ‘totality’ is understood in this cover term sense, it is not as confusing as the other definitions of pf aspect. But still, it has to be accounted for where it comes from.

With respect to the ipf aspect, the picture is not better – quite the contrary. Again, the mostly metaphoric characterisations cannot account for the data and tend to mistake possible

9 Context may also force a process-reading (cf. Breu 1998: 63f): On ponimal problemu vse lučše i lučše (‘He

was understanding the problem better and better.’). As will be shown in chapter 4, section 4.4, Turkish has explicit means to distinguish between the two possibilities.

Chapter 2: Setting the Stage

24

interpretations for the semantics of ipf. Symptomatic in this respect is the characterisation of the ipf in terms of ‘ongoing activity’ or ‘non-completed action’. 2.14 a. Ja uže vyražal emu svoe neudovol’stvie.

I already express:PST:ipf him my displeasure [Padučeva 1996: 43] ‘I have already expressed to him my displeasure.’

b. Ran’še on rabotal v universitete. in.the.past he work:PST:ipf at university:LOC ‘He worked at university.’ (= He was a teacher there.) Although ( 2.14a) and ( 2.14b) are both ipf, they express neither an ongoing activity nor an incomplete action, but the mere statement of a fact in ( 2.14a), and a characterisation of on (‘he’) in ( 2.14b). Moreover, ( 2.14a) illustrates the fact that ipf cannot be characterised as denoting incomplete actions either. ( 2.14a) is an example of the general-factual use of ipf10 and conveys a statement of the fact as such. The action is not viewed as ongoing or incomplete, it is rather presented as a whole. Definitions of ipf aspect have to account for examples like these as well.

Analogously to the pf aspect, C. Smith (1997) characterises the ipf aspect as informationally open. Comrie (1976) defines it as referring to the internal structure of a situation, and Filip (1999: 187) in terms of the feature [+PART] (partiality):

2.15 [Imperfective φ] relates eventualities denoted by φ to their parts, where the notion of

‘part’ is understood in the sense of the week ordering relation ‘≤’. As Filip emphasises, [imperfective φ] is an improper part relation in Russian, but a proper part relation in English. Analogously to her definition of the pf aspect, the question arises where this partiality comes from and what exactly it pertains to. As will become evident in chapter 4, section 4.2., the improper part relation is not enough either to capture the semantics of Russian ipf aspect adequately.

Supposedly in order to avoid the difficulties mentioned with respect to the above definitions, the ipf aspect is also characterised in purely negative terms, as being ‘not perfective’. Such a negative characterisation is not wrong, but still not very helpful or satisfying – neither for an analysis of ipf aspect in Russian, nor in cross-linguistic studies. One example of such approaches is Borik (2002), who defines the ipf aspect as ‘not perfective’: 2.16 a. pf: S ∩ R = ∅ & E ⊆ R

b. ipf: ¬(S ∩ R = ∅ & E ⊆ R), namely S ∩ R ≠ ∅ ∨ E ⊄ R According to Borik, two conditions are necessary for the pf aspect,( 2.16a): the intersection of speech time S and reference time R has to be empty, and the event time E has to be included in R. Ipf aspect results, whenever one of the two conditions does not hold, ( 2.16b). This is not wrong, but being presented this way, “one still wonders what referential properties ipf actually has” (Grønn 2003: 107). Since the central aim of this thesis is a systematic and predictable account for the interpretation of Russian ipf aspect, a purely negative characterisation of the ipf is not sufficient and as unsatisfactory as the above mentioned misleading descriptive characterisations, which mistake possible interpretations for the semantics of the form. A positive characterisation is needed and will be developed in chapter 4. It is a semantic analysis that avoids the problems of the above mentioned definitions and at the same time recognises the

10 For a profound analysis of the general-factual ipf and its several subgroups see Grønn (2003).

Chapter 2: Setting the Stage

25

unmarked status of ipf aspect as compared to pf. This is possible by defining a definite, but highly general semantics for the ipf aspect, which defines a specific semantic range within which various interpretations are possible.

Borik’s analysis of aspect (cf. also Borik & Reinhart 2004) in terms of Reichenbach’s (1947) notions event time E, speech time S and reference time R is not the only attempt to define aspect along these lines. The Reichenbachian system is rather frequently used in aspectual analyses, although Reichenbach himself did not intend to provide an analysis for the aspectual differences between the English simple and progressive forms. He merely stated that in some tenses an indication is given concerning the time extension of the event, e.g. in case of the present participle in English (Reichenbach 1947: 290). Reichenbach’s representations of the English simple ( 2.17a) and progressive ( 2.17b) past in terms of the relations between R, E and S do not exhibit any difference:

2.17 a. I saw John.

R, E S

b. I was seeing John.

R, E S

c. I have been seeing John.

E S, R The only difference between ( 2.17a) and ( 2.17b) is the additional marking of what Reichenbach calls ‘extension’. The relation between E and R remains identical. Examples ( 2.17b) and ( 2.17c) have the same ‘extension’, but differ in the location of R, i.e. the relation between E and R. Furthermore, it is not clear as to what time the extension pertains to, i.e., either E includes R, or R includes E.

Timberlake (1985) points out that there is more to aspect than just the relation between R and E. He translates Reichenbach’s definitions of the simple and the progressive past into statements including intervals and inclusion relations. Accordingly, the simple past can be described as E ⊆ R < S, and the progressive past as E ⊃ R < S. Implicit in these definitions is the assumption of the “dual character” (C. Smith 1997: 103) of the reference time. On the one hand, this reference time is part of the system of temporal relations – that is, it is located before, after or at S and E – and on the other hand it specifies the ‘aspectual perspective’ (Nakhimovsky 1988), which is indicated by the inclusion-relation obtaining between E and R.

However, the mere relations between S, E and R are not sufficient to distinguish pf and ipf aspect in Russian. Timberlake (1985: 160ff) illustrates this with the durative reading of the ipf aspect in ( 2.18a) and the event reading of pf aspect in ( 2.18b): 2.18 a. Uznav (pf) o bolezni Aleksandra Andreeviča, ja poechala (pf) v Moskvu, i

ostalas’ (pf) tam nadolgo.

b. Uznav (pf) o bolezni Aleksandra Andreeviča, ja poechala (pf) v Moskvu, i ostavalas’ (ipf) do ego vyzdorovlenija.

‘On learning about the illness of Alexandr Andreevič, I went to Moscow and remained there until his recovery.’

Chapter 2: Setting the Stage

26

The decisive units are ostalas’ and ostavalas’. With the pf ostalas’ in ( 2.18a), the inception of the state of remaining happens at some time R, that follows the preceding pf events. With the ipf ostavalas’ in ( 2.18b), the state of remaining is presented11 as extending from R of the previous pf event until Aleksandr Andreevič’s recovery, that is, it is presented as a bounded state. Both ( 2.18a) and ( 2.18b) have the configuration E ⊆ R < S (Timberlake 1985: 161): 2.19 a. eventive pf: ostalas’

E, R S b. durative ipf: ostavalas’

E, R S

The difference between ( 2.18a) and ( 2.18b) has to do with some internal limit, that is reached at R with the event reading of the pf aspect, but not reached at R with the durative ipf. This causes Timberlake (1985: 162) to introduce an additional time, the P time denoting the time span it would take for the predicate to reach its inherent limit. The notion of inherent limit in terms of the relation between P and R is opposed to that of an actual limit which is expressed by the relation of E and R. Timberlake uses the three notions of E, R and P to differentiate the progressive imperfect, the durative imperfect and the eventive perfective:

ACTUAL LIMIT INHERENT LIMIT progressive imperfect R ⊂ E R ⊂ P durative imperfect R ⊇ E R ⊂ P eventive perfective R ⊇ E R ⊇ P

Table 2.1 The configurations presented in Table 2.1 express the intuition “that a perfective is a predicate-time configuration in which there is an actual temporal limit on the event which also happens to be the inherent limit […] If the relationship between E and R is used to characterize the actual limit, then this relationship cannot also be used to characterise the inherent limit.” (Timberlake 1985: 162)

Note that although Timberlake distinguishes actual and inherent limits, he defines the pf aspect as expressing predicate-time configurations where both kinds of limit coincide. As will become evident in section 2.2.1, the two limits do not necessarily coincide, and the pf aspect may also denote the reaching of actual limits without referring to inherent ones.

Padučeva (1996) interprets Reichenbach’s R in terms of the notion točka otsčeta (henceforth: TO; position of the observer, adopted from Glovinskaja 1982). She distinguishes two different types of TO: a retrospective and a synchronous TO12. Reichenbach’s R is located with respect to the speech time, whereas TO is located with respect to the event time13. Although the position of TO helps to distinguish pf and ipf aspect, it is still not

11 It is important to note that the remaining is presented as extending to the recovery, as it might well go

beyond that if indicated by the context (note, that this defeasibility is due to the pragmatic nature of the specific readings). This does not invalidate Timberlake’s claim, contrary to Borik’s (2002: 162) comment.

12 For the future reference of morphologically present pf forms, a prospective TO has to be assumed. Thus the main distinction could be restated as synchronous vs. not synchronous.

13 A more detailed discussion of TO is provided in chapter 4, section 4.2.1., where it will also be argued that the notion of event time is too coarse and needs to be refined.

Chapter 2: Setting the Stage

27

sufficient. TO is retrospective with both the pf aspect and the general-factual reading of the ipf aspect. What distinguishes both aspects is not the position of TO, but a difference in what part of the event is asserted to hold: pf asserts the boundary, i.e. the ‘inherent limit’ in Timberlake’s terms, whereas the general-factual reading of the ipf is indifferent with respect to an inherent limit, but rather expresses an actual limit.

To recap so far, metaphorical definitions do not yield satisfactory results. Defining the ipf aspect as the unmarked member of the Russian aspect opposition in negative terms is probably not wrong, but not satisfactory either. The two notions of P-time and TO will be used in the analysis of aspect in chapter 4 as well, albeit in slightly modified terms.

Another problem for the analysis of Russian aspect consists in the multitude of readings that are possible for both aspects. One attempt to account for aspect in Russian is to simply list all these possible readings. However, this will fail, since the interpretations do not constitute distinct and unambiguously enumerable meanings of aspect, but rather context-dependent specifications of its semantics arising from the interaction with the Aktionsart basis. These readings must not be confused with the semantics of aspect – interpretation and semantics need to be separated categorically. What is commonly regarded as ‘core meaning’ of pf and ipf aspect might actually be a preferred interpretation assigned to that form in a sentence uttered out of the blue or in a minimal context14.

In ( 2.20), some of the readings proposed for ipf and pf are listed. These are the readings that will be dealt with in this thesis (chapter 4): 2.20 These readings will be used merely for expository reasons; they do not constitute different meanings of pf and ipf aspect. It is not necessary, and in fact not possible to list them exhaustively, as they are constructed in a given context and not selected from some predefined list. Therefore, the discussions about aspect readings, about primary and secondary readings and the like are futile to a certain degree. What is required is the elaboration of the semantic, i.e. context-independent, basis of both aspects (cf. chapters 4 and 5).

In order to keep the semantics of aspect basic, simple and stable, and at the same time to account for the various possible interpretations and their derivation, one has to take into account pragmatic principles such as, for example, generalised conversational implicatures. This has been suggested in particular by Bickel (1996: 31): “The traditional danger of structuralist analyses is that, in their attempts to isolate pure semantics, the rich contextualised meanings are not explicitly accounted for and, specifically, no attempt is commonly made to relate semantic structure to generalised implicatures.“

14 I share Ruhl’s (1989) view that the interpretation of an utterance out of the blue in fact always involves the

(unconscious) construction of a prototypical/stereotypical context.

readings for the Russian ipf aspect a. actual-processual b. conative c. inactual / continuous d. durative e. general-factual f. iterative g. habitual h. potential i. permanent j. atemporal

readings for the Russian pf aspect a. event b. perfect c. pluperfect d. future e. delimitative /

perdurative f. iterative g. summary h. exemplifying i. potential

Chapter 2: Setting the Stage

28

The general question of how semantics and pragmatics interact in interpretation is dealt with in chapter 3 and applied to aspect in Russian in chapter 5. There it will be accounted for why it is not decisive how many specific readings are postulated, and why it is not even necessary to list them all. What is important is to capture the generalisations that are involved in deriving them. These generalisations help to classify all possible interpretations into clearly distinct groups related to one common semantic basis each. It is enough to figure out and state this specific basis, as it serves to constrain the range of possible interpretations.

However, in traditional accounts of aspect in Russian, the concept of different readings, and the discussion on which and how many of them exist, is one of the most central issues. The prevailing assumption is that there exist certain core or primary meanings/ interpretations15 that have to be supplemented by further information to yield peripheral readings. Analyses of aspect in Russian typically distinguish pervičnye (‘primary’) and vtoričnye značenija (‘secondary meanings’). Primary meanings are understood as prototypical interpretations, which are less complex than secondary meanings: “Inače govorja, vtoričnye značenija mogut byt’ predstavleny kak rezul’tat popolnenija sobstvennogo značenija grammemy NSV […] za sčet informacii, zaključennoj v kontekste. V ėtom semantičeskom vyvode mogut prinimat’ učastie obščie zakony pragmatiki (predstavlennye v rabote Grajs 1989 v vide postulatov rečevogo obščenija).“ (Padučeva 1996: 11)

[To put it differently, the secondary readings can be thought of as the result of supplementing the basic meanings of the ipf grammeme by means of information provided by the context. In this semantic conclusion general principles of pragmatics may participate (proposed in Grice’s work in the form of conversational maxims.)]

In fact, however, both primary and secondary meanings are based on a common semantic basis. The semantics has to be such that all these context-dependent interpretations may be derived from it. What Padučeva seems to refer to with ‘basic meanings’ is the semantics of aspect, that has to be supplemented by means of contextual information in any case.

The distinction of semantics of the form and possible interpretations crucially requires the establishment of some semantic invariant meaning that is basic to the variety of aspectual readings. There are different kinds of invariant meaning that play a role here, e.g., the invariant meaning that characterises the grammatical category of aspect cross-linguistically and the invariant meaning that is basic to the language-specific instantiations of grammatical aspect (cf. chapter 4). Furthermore, a clear definition of the basic aspect semantics is required for the proper use of the central notions ‘opposition’ and ‘markedness’. Neither can be dealt with without the notion of a semantic invariant.

Basically, the Russian aspect system could be conceived of as an equipollent opposition or as a privative one. The former conception would assume the two members pf and ipf aspect to be of equal status, the latter would claim for one member of the opposition to be semantically marked. The marked member of a privative opposition is characterised as having one feature more as compared to the unmarked member which is agnostic with respect this feature – explicitly denying it would render the opposition equipollent. In Russian, the ipf aspect is commonly assumed to constitute the unmarked member, because it has a broader range of application as compared to the marked pf aspect, and can be used just to instantiate the category, that is, express an almost neutral value (cf. Jakobson 1971). It is decisive to state the common basis for both members of an opposition, since it is only with respect to that basis that markedness in terms of an additional feature specification receives its relevance. It is not enough to ascribe the pf aspect the additional feature of, say, totality, which the ipf aspect lacks. This leaves open what should be the shared characteristics of pf and ipf. The explicit statement of a common basis has to be provided by a semantic analysis of aspect, and will also be of considerable importance to the pragmatic analysis developed in chapter 5.

15 In fact, it is not always clear, whether it is interpretations or meanings that are referred to as core/primary

or secondary.

Chapter 2: Setting the Stage

29

The assumption of the ipf aspect constituting the unmarked member of the Russian aspectual system will not be challenged here. However, it is important to clarify the notion of markedness. Markedness may refer to several quite distinct phenomena even with respect to one and the same lexical item16. Talking of the markedness status of the pf and ipf aspect therefore requires stating the exact domain that is ascribed a markedness value. This is captured by the distinction of formal and semantic (or distributional) marking (Lyons 1977: 305-307). As will become evident, both kinds of marking need to be kept apart with respect to Russian aspect. Formal marking pertains to the distinction of simplex verbs that are not overtly marked for aspect but are specified by default aspect (see section 2.2.3), and verbs that are morphologically marked for deviation from that default aspect assignment. There is no correlation between simplex verbs and a certain aspect feature – simplex verbs may be pf and ipf, depending on the terminativity feature of the basic Aktionsart. Consequently, both pf and ipf verbs may be morphologically marked if their aspect feature does not correspond to the default aspect as predicted by the verb.

This formal marking has to be distinguished from semantic marking. Semantic marking often correlates with, and explains, distributional marking. A semantically marked lexical item, such as the pf aspect, is more specific than the corresponding unmarked one, and is therefore more restricted in application and distribution. In certain contexts, the semantic contrast between marked and unmarked members is neutralised, whereas in other contexts the semantically unmarked member may express a more specific value, namely the negation of the marked item.

Pragmatic marking pertains to deviations from the default combination of Aktionsart and aspect. This deviation has to be ‘explained’ by a respective interpretation. As this requires more interpretational effort, one can speak of pragmatic marking. Deviations from the default are indicated by morphological marking, and therefore, pragmatic marking tends to correlate with morphological marking. Pragmatic marking and its effects on interpretation will be dealt with in more detail in chapter 5.

Table 2.2 summarises the three different markedness notions and their relevance for Russian pf and ipf aspect:

pf

otkryt’ ipf

otkryvat’ pf

sdelat’ ipf

delat’ morphological marking - + + - semantic marking + - + - pragmatic marking - + + -

Table 2.2 As has been pointed out, in Russian there is no correlation between aspect and overt marking, that is verbs may, but do not have to be overtly marked for aspect in order to be assigned either pf or ipf. This is different for semantic markedness: the Russian pf aspect is consistently semantically marked, the ipf consistently semantically unmarked. This semantic markedness is systematic in that it is present independently of formal or pragmatic marking. Morphological and pragmatic markedness is different with different verbs. The connection between morphological markedness and pragmatic markedness can be explained in terms of ‘default aspect’ (section 2.2.3). This is illustrated in Table 2.2, where the pf otkryt’ (‘open’) is formally and pragmatically unmarked, but semantically marked as it is the more specific member of the aspect opposition. The ipf otkryvat’ is morphologically marked, which

16 The different notions, or domains, of markedness are also important with respect to possible pragmatic

inferences (cf. chapter 5).

Chapter 2: Setting the Stage

30

indicates the deviation from default aspect assignment. Consequently, it is also pragmatically marked, but this does not effect its semantic unmarkedness. With the forms sdelat’ (pf; ‘do’) and delat’ (ipf) the formal and pragmatic markedness relations are converse as compared to otkryt’/otkryvat’, but the semantic markedness remains unchanged.

2.1.2 The cross-linguistic perspective As has been pointed out repeatedly, the aim of this thesis is to provide an analysis of aspect not only in Russian, but to develop an account that is able to capture grammatical aspect cross-linguistically. Being positioned within and biased by the Slavistic tradition and perspective, one might be inclined to mistake the pf/ipf-distinction of Slavic languages, or even more specifically, that of Russian, as being generally applicable. As a result, potential aspect categories of other languages that do not fit this schema might be not regarded as aspect at all. For this reason, Jespersen (1924/1992: 288) suggests to restrict the terms ‘perfective’ and ‘imperfective’ to Slavic: “I think it would be better to do without the terms perfective and imperfective except in dealing with the Slavic verb, where they have a definite sense and have long been in universal use. In other languages it will be well in each separate instance to examine carefully what is the meaning of the verbal expression concerned, and whether it is due to the verb itself, to its prefix or suffix, to its tense-form or to the context. Different things are comprised under the term perfective.”

It is undeniable, though, that alleged aspect categories of different languages share certain common characteristics. Therefore, what is needed is a general, unbiased definition, i.e. semantic characterisation, of grammatical aspect, which is general enough to provide the semantic basis for every language-specific instantiation of this category, and at the same time flexible enough to account for the language-specific modifications of that general basis.

One suggestion for a general characterisation of grammatical aspect is Comrie’s frequently quoted definition: “Aspects are different ways of viewing the internal temporal constituency of a situation.“ (Comrie 1976: 3)

The category of aspect may be instantiated by different members. The most prominent are probably perfectivity and imperfectivity, which Comrie defines as follows: “[P]erfectivity indicates the view of a situation as a single whole, without distinction of the various separate phases that make up that situation; while the imperfective pays essential attention to the internal structure of the situation.” (Comrie 1976: 16)

However, the notions ‘perfectivity’ and ‘imperfectivity’ are mostly used without specifying what they refer to, i.e., whether it is to the meta-categories PERFECTIVITY and IMPERFECTIVITY or to their language-specific manifestation like pf and ipf aspect in Russian (or the progressive and the simple form in English)17. If referring to the meta-categories, it is important to note that they can be brought about differently. To give an example, it is correct to refer to the English simple past as PERFECTIVE, but not as ‘perfective’, since this form is not semantically marked for perfectivity (cf. chapter 4, section 4.5).

In order to cope with this terminological problem, Johanson (1996, 2000) proposes ‘terminality categories’ as unbiased cross-linguistic and terminology-independent aspectual categories that are instantiated by the language-specific categories. The notion of ‘terminality’ derives from ‘terminus’, which Johanson regards as the decisive feature of verbal classification (cf. section 2.2.2). These terminality categories describe the way an event is envisaged, that is within its limits (intraterminality), in the attainment of its terminus (adterminality) or after the transgression of the salient terminus (postterminality). As will be

17 Throughout this thesis, pf and ipf refer to the aspect category of Russian (and Bulgarian), IMPERFECTIVE

and PERFECTIVE refer to the meta-categories.

Chapter 2: Setting the Stage

31

shown, these categories are related in different ways to PERFECTIVE and IMPERFECTIVE (Johanson 1996; see also chapter 4, section 4.5 and chapter 5, section 5.2.3):

2.21 a. [+INTRA] intraterminality

b. [+AD ] adterminality: “perfectivity proper”

c. [-INTRA] non-intraterminality: “pseudoperfectivity“

d. [-AD] non-adterminality: “imperfectivity proper”

e. [+POST] postterminality

f. [-POST] non-postterminality: “pseudoperfectivity ”

Accordingly, the Russian pf aspect constitutes an instance of adterminality, whereas the ipf aspect is non-adterminal, not in the sense of explicitly negating adterminality but in the sense of being agnostic with respect to it – it is best regarded as ‘im-perfective’. In contrast, alleged perfective readings of the Turkish marker -di derive from its unmarkedness compared to the intraterminality encoded by -iyordu, -mekteydi and -irdi (chapter 4, section 4.4). Contrary to the Russian aspect system, the Turkish one is marked for intraterminality, and the unmarked member -di is to be analysed as an instance of ‘pseudoperfectivity’. The English progressive also instantiates intraterminality, whereas the simple form is unmarked in this respect and therefore allows for different uses. The variety of interpretations of the simple forms have been ignored in some analyses of the English aspectual system that ascribe aspect status only to the progressive (for a criticism of this approach cf. Thelin 1990). IMPERFECTIVITY is instantiated differently in Russian and in Turkish, which explains the differences between these at the same time rather similar language-specific aspect categories. In chapter 4 it will be shown how Johanson’s terminality categories relate to PERFECTIVITY and IMPERFECTIVITY.

Taking interpretations for semantics results in a superficial analysis which does not arrive at the common basis for all the possible readings. Not getting to the common denominator of grammatical aspect in one language further complicates cross-linguistic analysis. Grammatical aspect might be neglected in languages different from Russian, as they do not seem to correspond to the scheme postulated for Russian. Johanson (1996: 243) points out the problem we are confronted with here: “[O]n the one hand, the individual categories differ too much from each other to constitute meaningful general types, on the other hand, they display evident similarities which must be accounted for.”

The ipf aspect in Russian, the English progressive and the Turkish -iyordu have certain characteristics in common, but they also differ in certain respects, as is shown in ( 2.22)-( 2.27).

The English progressive has to be translated by the ipf aspect in Russian, but this is only a one-way-relation: 2.22 a. He was reading a book.

b. On čital knigu. he read:PST:ipf book:ACC

2.23 a. Ona dolgo smotrela na fotografii detej.

she long look:PST:ipf at photos children:GEN b. She looked at the children’s photos for a long time. / *... was looking...

Furthermore, applying the English progressive onto stative verbs requires contextual support, whereas the ipf aspect is the only possibility for stative verbs in Russian.

Chapter 2: Setting the Stage

32

The situation in Turkish is similar: the alleged progressive-marker -iyordu has to be translated into Russian by the ipf aspect, compare ( 2.24), but again this does not hold vice versa, as is shown in ( 2.25), where the ipf aspect is translated by the past tense marker -di, and ( 2.26), where it is translated by -ir (the ‘Aorist’): 2.24 a. Ben odaya girdiğimde erkek kardeşim kanepede

I room:DAT enter:FNOM:1SG:LOC brother:1SG sofa:LOC oturmuş kitap okuyordu. sit:PTCP book read:iyor:PST ‘When I entered the room, my brother was sitting on the sofa and reading a

book.’

b. Kogda ja vošel, moj brat ležal na divane i when I enter:PST:pf, my brother lie:PST:ipf on the sofa:LOC and

čital knigu. read:PST:ipf a book:ACC. 2.25 a. Ty čitala ėtu knigu?

you read:PST:ipf this book:ACC ‘Did you read this book?’

b. Bu kitabı okudun mu? this book:ACC read:PST:2SG Q ‘Did you read this book?’ 2.26 a. Deduška chorošo igral v šachmaty. grandpa well play:PST:ipf chess ‘Grandpa could play chess very well.’ (= He was a good chess-player.)

b. Büyükbabam iyi satranç oynardı. grandpa:1SG well chess play:ir:PST ‘Grandpa could play chess very well.’ (= He was a good chess-player.)

Examples like these cannot be handled without taking into account the language-specific semantic markedness relations holding among the members of the aspect category and the pragmatic inferences arising from them (cf. chapter 5). The general-factual reading, i.e. the mere statement of a fact, is always conveyed by the unmarked members of the respective aspect systems, that is by the ipf aspect in Russian ( 2.27a), the simple tense in English ( 2.27b) and -di in Turkish ( 2.27c): 2.27 a. Ty pokazyval ej ėto pis’mo?

you show:PST:ipf her this letter b. Did you show her this letter?

c. Bu mektubu ona gösterdin mi? this letter:ACC her show:PST:2SG Q This supports the claim that different aspect systems are based on different markedness relations: English and Turkish are marked for intraterminality (-ing and -iyordu), whereas Russian is marked for adterminality (pf aspect). Furthermore, Turkish has a second aspectual

Chapter 2: Setting the Stage

33

opposition concerning postterminality with the marked element -miş (as opposed to the unmarked -di)18.

These are just a few examples to illustrate the difficulties arising in cross-linguistic research. The English progressive and the Russian ipf aspect overlap in certain uses, but differ in others. How then can we capture both in a uniform way – if that is possible at all? Questions like this have to be addressed by an aspectual analysis and will be dealt with in what follows.

Thus, the analysis to be developed should be able to account for two domains, namely the language internal derivation of aspectual readings, here: of Russian ipf aspect, and the cross-linguistic variation of aspectual markers, here: Russian, Turkish, Bulgarian and English. The goal is to develop a framework that captures the common characteristics, the common denominator of grammatical aspect, but which at the same time is flexible enough to account for the language-specific properties in a systematic way. This flexibility is best captured by language-specific markedness relations that are decisive in the pragmatic processes involved in the derivation of the interpretations. If it is not taken into account how PERFECTIVITY and IMPERFECTIVITY are brought about, important differences might be overlooked. The cross-linguistic applicability, which avoids terminological confusion, is one of the big advantages of Johanson’s system of terminality categories, ( 2.21), that classify aspect operators with respect to the way they envisage events.

Figure 2.1 illustrates the above outlined conception of grammatical aspect on the cross-linguistic and language-specific level and of the interaction between grammatical aspects and the verb it is applied to:

Figure 2.1

18 In chapter 4, section 4.4 it will be shown that -di is unmarked in two respects: as regards intraterminality

and as regards postterminality.

grammatical aspect Aktionsart

operation interface OPERATOR OPERANDUM

SEMANTICS

SEMANTICS

INTERPRETATION

SEMANTICS

INTRA AD POST

▪ oppositions ▪ markedness relations

cros

s-lin

guis

tic;

inva

riant

sem

antic

m

eani

ng o

f asp

ect

lang

uage

-spe

cific

; di

ffer

ent i

nsta

ntia

tion

of g

ener

al

aspe

ct se

man

tics

interaction

interaction

Chapter 2: Setting the Stage

34

Basic to a systematic analysis of aspect interpretation is the separation of aspect and the Aktionsart (the verbal basis, section 2.2.2) it is applied to. Both stand in an operator-operandum relationship that presupposes an interface where the interaction can take place. The assumption here is that the Aktionsart provides certain elements aspect is sensitive to and interacts with. The exact specification of how aspect interacts with which elements of the Aktionsart basis constitutes the semantics of aspect. The separation of aspect and Aktionsart, the basic units constituting the operation interface and the way in which aspect makes use of them are cross-linguistically valid assumptions that constitute the common denominator for grammatical aspect (chapter 4).

This basic semantics of grammatical aspect is uniform across languages but instantiated differently in different languages. Possible instantiations are constrained to the three terminality categories – intraterminality, adterminality and postterminality – which in turn may be expressed by different morphological markers that differ in their respective opposition- and markedness-relations. The terminativity categories relate to the concept of ‘event nucleus’ (section 2.2.2) which represents the internal make-up of a prototypical event. Since within these possible instantiations languages differ with respect to their morphological means to overtly encode the respective categories, they also differ in the oppositions established by the respective markedness relations. The language specific markers of grammatical aspect interact with the Aktionsart basis in specific ways, which might misleadingly give rise to the assumption of grammatical aspect being a rather heterogeneous category that cannot be ascribed one basic semantic meaning.

One of the main points here is that it is possible to assume one cross-linguistic semantics for aspect on the one hand, and to establish one invariant semantic meaning for the members of the language specific instantiations of grammatical aspect on the other. To account for the cross-linguistic and language-internal diversity and variety in usage and interpretation, pragmatic inferences have to be taken into account. Figure 2.1 sketches only the basic semantic assumptions on aspect and its interaction with the verbal basis. The pragmatic principles will be added in chapter 3, and applied to grammatical aspect in chapter 5. The overall approach taken in this thesis is the following: starting from the data aiming at a consistent account of the linguistic facts. Another possibility would be to take one certain model of grammar (HPSG, Minimalism, Construction Grammar, etc.) as basis into which the data are incorporated, thereby bearing out the assumptions of that respective model, or modifying and supplementing them

To sum up, it is important not to confuse the semantics of the form and its interpretation. If semantics is separated from interpretation, the search for the invariant meaning basic to the aspect cross-linguistically and language-specifically can be dealt with in a more straightforward way. The concept of markedness and its different applications proves crucial for adequately capturing that invariant meaning. The differences between the language-specific instantiations of grammatical aspect are due to the different terminality categories encoded by the different aspect systems, differences in morphological means to express grammatical aspect, the oppositions these forms enter into and the resulting markedness-relations. These latter factors give rise to pragmatic inferences that explain the cross-linguistic differences in usage and interpretation of grammatical aspect. The pragmatic inferences will be dealt with in chapter 5, but to furnish them with the necessary predictive power, the semantic basis of the respective expressions – aspect markers – needs to be established (chapter 4).

2.2 Basic distinctions

This section introduces the notions and distinctions that will be important for the semantic analysis of aspect outlined in chapter 4. Crucial is the distinction of lexical aspect or

Chapter 2: Setting the Stage

35

‘Aktionsart’ and grammatical aspect on the one hand, and that of terminativity and boundedness on the other. The distinction between Aktionsart and grammatical aspect has been emphasised in several approaches to aspect (e.g., C. Smith 1997). Aktionsart will be used in the sense of Egg (1994: 11f), namely as scheme classifying VPs and verbs according to certain features. This accounts for the fact that VPs may have Aktionsarten that differ from those of the respective verbs. At the verb level, the relevant features are phases and boundaries, at the VP level this is basically the feature terminativity. Boundedness may also be specified within the VP, but may as well come into play at a higher level. ‘Aktionsart’ thus refers to basic eventuality descriptions expressed by a verbal predicate whose morphological exponent is the aspectless verb stem (Filip 2005: 126). This points out the distinction between Aktionsart and aspect that will be important to the account developed here. Aktionsart pertains to the level of event semantics, it relates to properties of the denoted event without the contribution of aspect19.

Section 2.2.1 introduces the distinctions between terminativity, boundedness and aspect. In section 2.2.2, the question of how to classify verbs and VPs is dealt with. Section 2.2.3 emphasises and substantiates the need to distinguish between aspect and Aktionsart.

2.2.1 Terminativity, boundedness and aspect The three notions of terminativity, boundedness and grammatical aspect play a crucial role in defining the semantics of aspect: (non)terminativity relating to the presence of an inherent culmination point (terminus) of a predicate, (un)boundedness relating to the presence of actual (temporal) boundaries20 and grammatical aspect. Terminativity, boundedness and aspect need to be kept apart in order to account for the semantics of aspect categories in different languages. There is a crucial difference between terminativity and boundedness on the one hand, and aspect on the other: Terminativity and boundedness are properties of the Aktionsart that provides the basis for aspect to apply21. For an analysis of the semantics of grammatical aspect, it is crucial to distinguish between the semantic contributions of aspect and Aktionsart, the latter being specified by terminativity and boundedness (cf. Depraetere 1995; vanden Wyngaerd 2001; Capelle & Declerck 2005). As aspect may be sensitive to both boundedness and terminativity, it might easily be confused with these notions.

The conflation of aspect and terminativity/boundedness is especially tempting in languages with an aspect system based on adterminality. In such systems the inherent (as well as the actual) boundaries of the denoted event play a crucial role as aspect is sensitive to exactly these boundaries. Moreover, the distinction between grammatical aspect and VP-based terminativity (and boundedness) is not always easy to draw, since grammatical aspect is not always encoded by a separate morpheme22. Thus, it may be difficult to divide the aspectualised verb into the respective contributions of verb and aspect. This is probably one of the reason why prefixes tend to be mistaken for exponents of the pf aspect (cf. section 2.2.3). Aspect and Aktionsart interact very closely, and this interaction presupposes a specific interface where features provided by the verb constitute the basis for aspect to apply (cf. chapter 4).

19 If necessary, the specific level will be referred to – verb class or VP – otherwise, the cover term Aktionsart

will be used (compare also Bickel’s ‘Aspectual Uniformity Hypothesis’ below). 20 For simplification these notions will be referred to as ‘terminativity’ and ‘boundedness’, denoting the

respective unmarked or negative values as well. 21 More precisely, a certain kind of boundedness may be expressed within the VP, in which case it constitutes

an input factor for aspect, but it may as well be expressed above the VP, in which case it is not decisive for aspect. Compare chapter 4, section 4.2.1.

22 In some languages the combination of verb and aspect is easily segmentable. Turkish, for example, expresses aspect consistently by separate morphemes attached to the verb stem. The situation is more complicated in Russian, where aspect is not expressed consistently by special morphemes but is inherent to the verb stem (section 2.1).

Chapter 2: Setting the Stage

36

A short remark on terminology: throughout this thesis, ‘terminativity’ will be preferred to ‘telicity’. Terminativity refers to the notion of inherent boundary, or culmination point, respectively. Telicity is more restricted and covers only a subset of terminative VPs, namely those with a final inherent boundary. Replacing ‘telicity’ with the more general term ‘terminativity’ is justified by the existence of verbs and predicates that encode initial boundaries with a subsequent process or a subsequent state (ingressive and inceptive verbs; cf. Breu 1998), and verbs that encode an initial phase, a culmination point and a consequent state (inchoative verbs; cf. Breu 1998), that is, predicates with an inherent, but not necessarily final, boundary. Capturing characteristics of VPs in terms of terminativity instead of telicity is important in dealing with languages that cannot be adequately analysed in terms of the telic-atelic dichotomy, such as Turkish (Johanson 1971: 197). Terminative VPs denote actions which have a natural, inherent turning point that may be an initial, medial or final terminus – the culmination point at which a transformation takes place.

The distinction between terminativity and boundedness is even less established than that between terminativity and aspect23. The difference between terminativity and boundedness consists in the two kinds of boundaries mentioned above: inherent and actual boundaries. This has been emphasised in the traditional ‘Soviet’ accounts of aspect already:24 “Predel možet byt’ vnutrennim i vnešnim. V tex slučajach, kodga vremennaja granica ne zavisit ot charaktera samogo dejstvija i obuslovlivaetsja vnešnimi po otnošeniju k nemu faktorami [...], my imeem delo s vnešnimi predelami. [...] Esli že ograničenie protekanija dejstvija otražennym v značenijach glagola [...] to reč’ dolžna idti o vnutrennem predele.“ (Bondarko 1987: 46)

[The limit may be internal or external. In cases where the temporal limit does not depend on the character of the action itself and is dependent on factors external to it [...], we are dealing with external boundaries. [...] If, however, the boundedness of the course of an action is expressed within the meaning of the verb [...], one has to talk about an inherent boundary.]

The need to distinguish between terminativity and boundedness is emphasised also by Depraetere (1995), who defines these notions as follows (1995: 2f): “(A)telicity25 has to do with whether or not a situation is described as having an inherent or intended endpoint; (un)boundedness relates to whether or not a situation is described as having reached a temporal boundary“

Terminativity is an inherent characteristic of the predicate. It classifies VPs, although verbs are already specified for a terminativity value in their lexical entries. This value may be passed through to the VP level, or it may be changed there. Thus, terminativity is finally determined at the VP-level and depends not only on the verb but may be influenced by other factors as well (e.g., certain kinds of arguments, most prominently the incremental theme argument, cf. Dowty 1991; Filip 1999)26.

In contrast, boundedness is a characteristic of the asserted situation, or more specifically, of the time interval27 an assertion is made about. It pertains to a particular way of representing or interpreting a particular actualisation of an event, i.e. to whether the event represented by a clause is coming to an end (Cappelle & Declerck 2005: 893). Boundedness is not a feature of the VP but is specified by a finite clause. The term ‘bounded event’ therefore means “event whose actualisation is represented as bounded by a finite clause” (Cappelle & Declerck 2005:

23 One exception is Koenig & Muansuwan (2000), who define the distinction of terminativity and

boundedness in formal terms (cf. chapter 4, section 4.3). 24 These distinctions do not seem to play an important role in the ‘Anglo-American’ tradition. In this respect,

it is very revealing to see how Slavic Slavists working in the more formally oriented and still more Anglo-American coined paradigm, do not stop to emphasise these distinctions and differences (e.g. Filip 1999), as if reinventing something that has been common knowledge in the Soviet/Eastern European line of research.

25 Remember that here, the notion of telicity is replaced by that of terminativity. 26 Telicity/terminativity is a rather complex topic in event semantics and will not be dealt with here in detail.

For a thorough investigation of telicity and related topics see, e.g., Krifka (1998), Rothstein (2004). 27 Chapter 4, section 4.1 deals with intervals and their contribution to aspect semantics in more detail.

Chapter 2: Setting the Stage

37

894). Although there is a default association of terminative predicates with bounded situations and aterminative predicates with unbounded ones, boundedness is independent of the inherent characteristics of the predicate. In chapter 4, section 4.1 ‘boundedness’ will be argued to correspond to Klein’s (1995, 2001) notion of topic time. The distinction between temporally bounded and unbounded situations28 is grammaticalised in some languages29. In languages that do not encode boundedness grammatically, this feature can either be inferred as a default from aspect marking or from the terminativity feature of the Aktionsart, or it can be specified by lexical and contextual means. Boundedness is independent of terminativity, since terminative situations may be presented as unbounded and aterminative situations as bounded. Temporal boundaries can be imposed onto any denoted situation, that is also on static and momentaneous ones. And finally, boundedness is not identical to completion, as an event may terminate before it has reached its inherent end point.

Another distinguishing feature between terminativity and boundedness is that only the former implies a change of state and lexically encodes a resultant state (or an initial state; Demjjanow 1998). Boundedness of a situation does not entail a change of state and does not convey a lexically encoded resultant state (but may give rise to a resultant state in the sense of some experience that results from the event)30.

English provides evidence for the independence of boundedness and terminativity. Examples ( 2.28a, b) illustrate that atelic predicates may be presented as bounded (cf. Thelin 1990: 4-22, who argues for a strict distinction of grammatical aspect, terminativity and boundedness): 2.28 a. John pushed the cart for a while and then sat down on the green.

b. Mary swam for half an hour before she had breakfast. Here, the simple forms convey aterminative predications that are temporally bounded but cannot be treated in terms of change-of-state. Therefore, Thelin (1990: 9) concludes: “This possibility to totalise ‘activities’ [i.e. nonterminative predicates, B.S.] with the aid of simple forms [...] shows that aspect distinctions cannot primarily be understood as an expression of different verb classes or types of verb phrases, or, for that matter, of the presence or absence of an ‘aspectual’ operator BECOME.”

The main consequence of distinguishing boundedness and terminativity is that the use of the progressive form with a terminative predicate does not impose an aterminative reading of the respective event. Rather, it turns the event into an unbounded one, but does not affect its terminativity, as is shown in ( 2.29): 2.29 John was opening a parcel.

28 In chapter 4, section 4.1.2 it will be shown that the mere distinction of bounded and unbounded is too

coarse-grained. 29 One example is Bulgarian, where Aorist and Imperfect are used to express a specific kind of boundedness.

Whether this distinction is temporal or aspectual is central to discussions of the Bulgarian aspect system; here it will be conceived of as aspectual (cf. also Thelin 1978, 1990; Maslov 1959). Compare chapter 4, section 4.3.

30 These two notions of terminativity and boundedness are not related to one another, they sometimes coincide but do not necessarily have to, contrary to what is assumed, e.g. by Krifka (1998: 232): “I use ‘telicity’ here in the sense of ‘boundedness’ in Depraetere (1995), who reserves ‘telicity’ for events with a natural or intended endpoint [...]. I feel justified in doing so, as the nature of the endpoint does not affect the points to be made here.” While it might be right that a distinction between the two kinds of endpoints does not affect certain analyses, this does not justify conflating them. Such a conflation would suggest that there is no difference at all. Moreover, the difference may turn out to be crucial for other analyses. Furthermore, it is highly confusing to explicitly refer to terms defined in a previous analysis and then use them with the exact opposite sense, as is done by Krifka (1998). In this thesis, terminativity will be regarded as referring to predicate inherent boundaries, boundedness as referring to temporal, actual boundaries as defined by Depraetere (1995). The two notions will be kept apart as the difference between these two kinds of terminal points does matter.

Chapter 2: Setting the Stage

38

This is only a tendency, however, and does not entail that there is no distinction between aspect and boundedness. Both notions rather cross-cut each other (Cappelle & Declerck 2005: 895). Sentences containing the simple form can be temporally bounded as in ( 2.30a) or nonbounded, as in ( 2.30b). Furthermore, although progressive forms usually trigger unboundedness, they may be used in bounded sentences as well, compare ( 2.31b): 2.30 a. Barney drank a glass of beer.

b. Barney drank beer. 2.31 a. It is raining.

b. It has been raining: the cobbles are still wet. The fact that languages do not always overtly encode the distinctions presented here may be the explanation for why they have not always been kept apart properly. This is probably the main reason for conflating boundedness and terminativity, and for the more ‘traditional’ mixing of Aktionsart (lexical aspect) and grammatical aspect. Aspect definitions are sometimes stated in terms of boundedness or terminativity. C. Smith (1997), for example, defines the pf aspect as presenting a situation as informationally closed and the ipf aspect as presenting a situation as informationally open (cf. section 2.1.1): “Perfective viewpoints are closed informationally, in the sense that they present situations as complete with both endpoints. Imperfectives are open.” (C. Smith 1997: 65)

This definition seems to refer to actual boundaries only, suggesting that PERFECTIVE aspect expresses the temporal boundedness of an asserted situation. C. Smith exemplifies her definition of what she calls ‘pf aspect’ (the simple form) with the following examples (1997: 66-68): 2.32 a. Lily swam in the pond.

b. Mrs Ramsey wrote a letter.

c. Lily coughed.

d. Mr Ramsey reached the lighthouse. These examples show, however, that PERFECTIVE viewpoints, or more specifically the English simple form, may relate to both kinds of boundaries. Whereas ( 2.32a) and ( 2.32c) present terminated (finished) events, ( 2.32b) and ( 2.32d) convey completed ones. According to C. Smith these interpretations are due to the simple form, and the interpretation of termination or completion depend on the situation type: if a situation type has a natural endpoint, so does the ‘perfective’ sentence31 which then conveys completion and not termination. Situation types without natural endpoints terminate at an actual boundary, but cannot be said to be completed. In both cases English uses the unmarked simple form. A situation with an inherent boundary

31 C. Smith conceives of simple forms in English as expressing perfective viewpoint. This implies the

members of the English aspectual system to stand in an equipollent opposition. In fact, however, the progressive is the marked member expressing intraterminality, whereas the simple form is unmarked and does not necessarily convey perfective meaning. This is obvious with the statement of a fact, that is always conveyed by the respective unmarked aspect forms – in English, with the simple past. Nakhimovsky (1988) assumes English to have an imperfective/unmarked system, but Russian to have an imperfective/perfective system. This seems to misconceive the Russian aspect opposition as equipollent, but correctly points out the privative opposition for English, with the simple form as the unmarked member. See also Thelin (1990) for a further elaboration of that point.

Chapter 2: Setting the Stage

39

may also be presented as being finished but not being completed, i.e. as having stopped before the inherent boundary has been reached.

Taking into account the notion of boundedness as being different from terminativity and grammatical aspect explains the fact that Russian delimitative pf predicates behave in some respects as would be expected from ipf predicates. This is the case especially with respect to durative adverbials, which are supposed to test the terminativity of the respective predicates (cf. Demjjanow 1998: 45). The fact that delimitative and perdurative verbs are aterminative is illustrated by their combination with durative adverbials, illustrated in ( 2.33). Their behaviour with phasal verbs and the impossibility of present tense reference, illustrated in ( 2.34), shows that they are perfective: 2.33 a. On čertil dva časa.

he paint:PST:ipf two hours ‘He painted for two hours.’

b. On počertil dva časa. he paint:PST:pf two hours. ‘He painted for two hours.’

2.34 a. *On načal počertit’. he begin:Pst:pf paint:INF:pf ‘He began to paint.’32

b. *On sejčas počerti. he now paint:PRS:pf *‘He paints now.’

Such examples can be accounted for straightforwardly, if terminativity, boundedness and aspect are kept apart. As will be shown in more detail in chapter 4, section 4.2, the pf aspect applies to VPs that encode an inherent or an actual boundary. In case of delimitative and perdurative verbs, the required feature for pf to apply is provided by the temporal boundedness of the event time, which is indicated by the prefixes po- or pro-. In contexts that test for perfectivity, these predicates pattern with all other pf verbs in that they cannot be combined with phasal verbs like begin or stop, and do not have present tense reference ( 2.34), in contexts that test for telicity, these predicates pattern with aterminative predicates, as is shown by their compatibility with durative adverbials in ( 2.33).

The Russian examples show that aspect is independent of both terminativity and boundedness (cf., e.g., Filip 1996). Russian delimitative and perdurative verbs marked by the prefixes po- and pro- express temporal boundedness, i.e. boundedness of the run time of the encoded event33. Except for these two prefixes, Russian has to use lexical means, predominantly temporal adverbials, in order to express boundedness, or it is induced by the context, as is the case with the durative reading of ipf (see below). Aspect does not change the terminativity feature of the verbal basis (cf. section 2.2.3, where aspect and terminativity will be argued to work on different semantic levels), and thus cannot be defined in terms of terminativity. Lindstedt (1985: 155f) argues along the same lines, concluding that terminativity (or telicity, as he calls it) is different from aspect:

32 The unacceptability of the Russian example is not preserved in the English translation. 33 Borik (2002) argues that delimitedness is connected to the notion of reference time, that is, delimiting

expressions restrict the reference time interval (compare chapter 4, section 4.2.1).

Chapter 2: Setting the Stage

40

“Since telicity is defined as a certain kind of relationship between the interpretations of the perfective and imperfective34, it logically cannot be grammaticalised as an aspect distinction but must remain part of the aspectual character [of the verb, B.S.].” (1985: 156)

Thus, defining aspect as the “completion or non-completion of a situation” (Bennett et al. 1989: 248), or in terms of terminativity is misleading. As has been shown here, aspect, terminativity35 and boundedness are independent notions. In cases where these notions are morphologically distinguished, probably nobody would think of neglecting the need to distinguish between them. The interpretations of completion and termination do not follow from the semantics of the simple form in English or of the pf aspect in Russian – at least not in the straightforward manner suggested by C. Smith. This confusion is due to the fact that English does not morphologically distinguish between terminativity, boundedness and PERFECTIVITY. Other languages mark at least some of these distinctions overtly. Russian marks the reaching of inherent boundaries36, Bulgarian marks the distinction between the reaching of inherent and the reaching of actual boundaries. Furthermore, there are languages that systematically encode terminativity without having a system of grammatical aspect at all37.

Independently of potential endpoints, any situation can be described as having reached a temporal boundary. Guentchéva 1989 (cf. also Desclés 1993, Desclés & Guentchéva 1990a,b) divides dynamic situations (‘processes’) according to whether they are presented in the course of their development and are concomitant to the act of utterance, or whether they are located before the act of utterance. In the former case, she speaks of non-accomplished processes, in the latter of accomplished ones. This latter group is further divided into accomplished processes that are finished before they reach their inherent boundary, and into completed processes that have reached their inherent limit and are completed:

Figure 2.2 Languages then differ with respect to which oppositions they grammaticalise: “L’opposition achèvement/absence de signification d’achèvement est à la base de l’opposition aspectuelle perfectif/imperfectif des langues slaves ; l’opposition accompli/inaccompli se trouve à la base d’un grand nombre de langues, en particulier les langues sémitiques.” (Guentchéva 1989: 28)

As shall be seen (chapter 4, section 4.3), Bulgarian is one of the languages that grammaticalise the opposition accompli/inaccompli (in addition to achevé/non-achevé)38.

34 He notes that an ipf verb is telic if it denotes a process directed towards the completion of the same event

as denoted by the pf aspect, and a pf verb is telic if it has a telic ipf counterpart. This leaves open the possibility of pf verbs being atelic, like the delimitative/perdurative ones. These do not have ipf telic counterparts and are therefore not telic: posidet’ (‘sit for a while’) is pf and atelic, its ipf counterpart sidet’ (‘sit’) is atelic.

35 For the distinction of terminativity and aspect in Russian compare, e.g., Filip (1999) and Borik (2002).. 36 In case of delimitative and perdurative verbs, the reaching of actual boundaries is marked. 37 Hungarian is an example for a language that seems to systematically encode terminativity, but not

grammatical aspect (Csató 1994; compare also the remarks in Johanson 2000). 38 Other languages exhibiting this distinction are, e.g., Mandarin Chinese (C. Smith 1997), Hindi (Singh

1998) and Thai (Koenig & Muansuwan 2000); cf. chapter 4, section 4.3.2.

denoted event

ongoing at TU over at TU

finished (accompli)

completed (achevé)

Chapter 2: Setting the Stage

41

To sum up this section, terminativity and boundedness are independent of each other and do not always coincide39. Perfectivity as an instantiation of the respective grammatical category of aspect has to be described independently of both terminativity and boundedness (see section 2.2.3). With respect to Slavic languages, Filip (2005) puts it as follows: “In Slavic languages, the grammatical aspect, perfective and imperfective, is a property of the total sum of the morphological parts of a fully formed verb, excluding its generic and temporal suffixes”.

As will be shown in section 2.2.3, aspect takes eventuality descriptions – Aktionsarten – as input. Aspect makes a stable contribution to the interpretation process (to be developed in chapter 4), whereas its context, first of all the verbal basis, is variable and has to fulfil the input requirements imposed by aspect. The conditions under which such context variations occur and how they are interpreted will be the topic of chapter 5.

2.2.2 Aktionsart and event structure This section addresses the question of how predicates can be classified according to their interaction with grammatical aspect. Three proposals of classification will be outlined, all of which play a role in this thesis. It will be argued for a classification that is also backed by certain conceptual notions concerning the structure of events.

Natural language predicates encode and denote properties of events. One of the central questions in event semantics is what kind of knowledge about events is part of the lexical meaning of predicates denoting them (Nakhimovsky 1988). Different properties are decisive in different respects. One of the big questions in event semantics dealing with aspect pertains to which parts of those encoded by the predicate are relevant for grammatical aspect. The interaction of aspect and the verbal basis requires a classification of predicates according to the features that play a role for grammatical aspect. This section aims at figuring out the basic components that make up verb and VP classes and enter into the operational interface of aspect and verbal basis.

Classifications necessarily are a trade-off between generality and specificity, they need to focus on general common features and may ignore more specific ones: “Classifications are useful if they help sort out the data. They are potentially harmful if they drown us in detail and obscure the general picture.” (Singh 1998: 178)

39 One approach that takes the distinction between (a)terminativity and (un)boundedness as decisive is

developed by Egg (1994). He does not deal with grammatical aspect, but is mainly concerned with aterminative predicates that are bounded, i.e., with what he calls ‘intergressives’. This group of bounded aterminative predicates cannot be adequately classified into one of the existing classes (i.e., the Vendler-classes). They cannot be simply defined as change-of-state predicates. If all events (i.e. non-stative eventualities) were classified as bounded, the distinction of aterminative and terminative predicates would be lost. Thus, in order to capture this group of intergressive predicates, Egg proposes to view the features of boundedness and terminativity as distinct (1994: 38-50). This distinction is relevant for interval-based, that is, dynamic, predicates. The set of predicates is ordered in terms of three features: interval-based, bounded and telic, the latter two only being relevant in case of interval-based (i.e. non-stative) predicates. This makes non-interval-based states crucially distinct from the rest of predicates. The features are ordered along the following left-to-right hierarchy (note that here nothing is said about how these features are realised with a given predicate):

interval-based < bounded < telic Bounded predicates constitute a subset of interval-based predicates, telic predicates a subset of bounded predicates. Boundedness does not imply a change-of-state, and thus is distinct from telicity/terminativity. According to Egg, telic/terminative predicates are always bounded. Each of this features divides the set of predicates into two subsets yielding four non overlapping groups (these four groups divide predicates, not verbs):

State Process Intergressive Change This is one proposal of how VPs might be classified. The matter of classification is highly controversial, in terms of what is to be classified by which criteria (cf. section 2.2.2).

Chapter 2: Setting the Stage

42

There can never be something like ‘the’ classification of verbs or predicates, since a classification is always carried out with a certain purpose or goal. The kind of classification looked for here is one in terms of notions relevant for grammatical aspect. It cannot, and does not, claim to be a classification valid for all other possible purposes as well.

Central to this section is the notion of Aktionsart understood as aspectually relevant classification of verbs and verb phrases, i.e. predicates. The traditional Slavistic use of the term ‘Aktionsart’ refers to lexicalisation patterns of actions by means of derivational morphology, more precisely, by means of prefixes that make a specific semantic contribution to the meaning of the verb they attach to. These semantic specifications include notions that are related to phase, measure, distributivity, intensity and the like. This traditional use has been extended, and more recently the notion of ‘Aktionsart’ is used to refer to the classification of predicates encoding specific eventuality types. It is in this latter conception, that ‘Aktionsart’ will be used here. The term does not refer to ontological distinctions of eventuality types, but to predicates that are classified according to which features of the denoted eventualities they encode. Verbs and verb phrases provide certain features aspect is sensitive to. These features are used as criteria for classifying verbs into Aktionsarten. The respective Aktionsarten then allow for predictions and generalisations with respect to the interaction of aspect and verb/VP and the interpretation of the respective word forms. As will be shown in chapter 5, such a classification also makes predictions about possible recategorisations caused by aspect-Aktionsart mismatches.

The classification task is not as easy as it might seem at first glance. For example, how is a verb like write to be classified? As terminative or aterminative? As underspecified with respect to terminativity? Or should it be assigned two lexical entries – one terminative version, one aterminative version? Crucially, Aktionsarten are no permanent features attached to a verb or VP. Therefore, it is best to give a characterisation of Aktionsart in rather basic, meta-language terms, and then explain how a default association between a linguistic expression and an Aktionsart arises (cf. van Lambalgen & Hamm 2005: 85-90). This default connection may be changed in the course of interpretation40. This is the reason for why a classification of verbs is sometimes regarded as superfluous. Johanson (1971: 200), for example, argues that the affiliation of verbs to certain classes is not decisive, since aspect determines the whole actional phrase: “Die Frage der Klassenzugehörigkeit des einzelnen Lexems ist in der Tat sekundär, da der Aspekt die gesamte Aktionalphrase bestimmt und Lexeme selten ohne Ergänzungen auftreten.“ (Johanson 1971: 200)

[The question of class affiliation of a single lexeme is in fact secondary, since aspect determines the whole actional phrase (VP, B.S.) and lexemes rarely appear without complements.]

Here, Johanson points out one important level for aspect, namely that of verb phrases, where terminativity is determined. However, not all aspect categories are sensitive to that level. It is indeed important to figure out the constitutive features of verbs as well, since aspect categories may be sensitive to these components. In section 2.2.3 it will be shown that Russian aspect system is sensitive to VP-level terminativity41, the English system to the level of verb class (dynamic vs. static) and the Turkish system for the basic constituents that characterise the internal make-up of verbs (dynamic and static phases). As aspect markers in English, and more obviously in Turkish, are sensitive to a distinction that is not specified at the VP level, it is necessary to classify also verbs. Another reason for why it is important to classify verbs is that different verbal bases constitute different frames for pragmatic inferences like presupposition, accommodation and implicature. Furthermore, a classification

40 This change is commonly referred to as ‘coercion’ (compare chapter 5). 41 Mere sensitivity does not claim that terminativity equals pf aspect.

Chapter 2: Setting the Stage

43

is also needed with respect to applications in computer-guided natural language processing, where exact input data have to be stated.

Verbs are best classified according to a minimal repository of basic elements (compare the discussion in Bickel 1996: 16-20). Postulating verb classes a priori, e.g., in terms of activities, states, achievements and accomplishments obscures the basic units these classes consist of.

Below, three proposals concerning verb classification will be outlined, all of which are relevant for the reasoning here. Pulman’s (1997) classification will be argued to be inadequate in dealing with languages other than English, but it will prove useful in dealing with recategorisation phenomena (chapter 5). Johanson’s (2000) classification emphasises the important role of the internal make-up of the ‘actional content’ of predicates, which is decisive for the use of aspect operators. Bickel (1996) classifies verbs and VPs according to a minimal set of basic units – phases and boundaries – that are decisive for aspect to apply. (a) Pulman (1997) Pulman takes the following to be the basic kinds of eventualities: ‘point’, ‘state’ and ‘process’, based on which combinations to complex types are possible. The combination <point, state> corresponds to Vendler’s class of achievements, the combination <process, state> to that of accomplishments. Pulman relies on structures consisting of only two elements. He takes a tripartite structure (as proposed, e.g., by Moens 1987 and Moens & Steedman 1988) as plausible from a conceptual point of view, but as superfluous from a linguistic position as there seem to be no cases where all three components of such a structure may be accessed independently and simultaneously: “[W]hile there are phenomena in which both separate components of the <point, state> and the <process, state> combinations are modified separately, there are no clear examples where all three components of the more complex structure can be accessed independently and simultaneously.” (Pulman 1997: 7)

Such an impression may indeed arise if the analysis is restricted to English only. However, taking into account languages like Russian and Turkish reveals that in order to describe and account for aspectual phenomena in an adequate way, a tripartite structure is indeed necessary. Problematic with Pulman’s representations is the seemingly fluent transition from ‘process’ to ‘state’. It is, however, exactly this transition – or culmination point – that is focused upon by the Russian pf aspect (chapter 4, section 4.2). Moreover, Pulman’s conception contradicts Jackendoff’s (1987) criterion for well-formed structures on the temporal tier (compare Bickel’s proposal below): “The well-formedness constraint on this tier is that Ps [points, B.S.] and Rs [regions, B.S.] must alternate; that is, two Ps must be separated by a region, and two Rs can be adjoined only by designating a point in time that ends one and begins the next.” (Jackendoff 1987: 398)

In Pulman’s representation of <process, state>, two regions follow each other without being separated by a point. This is problematic from a conceptual point of view, because in order to distinguish a process (dynamic phase) from an adjoining state (static phase)42, there has to be some transition bound between the two. Moreover, Pulman’s conception is also inadequate from a linguistic point of view, as there are languages, e.g., Russian, whose aspect system is sensitive to exactly that transition bound. Furthermore, the second part of that structure does not always have to constitute a state, but may be a process as well. This is the case with ingressive verbs which are common in Turkish (see Johanson’s classification below).

Pulman assumes that verbs are lexically specified with respect to their default Aktionsart interpretation, ‘default’ pertaining to verbs with a singular definite argument without aspectual modifier (cf. also Moens 1987). For the default-assignment to Aktionsarten classes

42 The notions of dynamic and static phase will be introduced below, and defined in more detail in chapter 4,

section 4.1.2.

Chapter 2: Setting the Stage

44

he assumes similar criteria as those proposed by Kamp & Reyle (1993): temporal duration relative to a contextual ‘baseline’, the perceived salience of a consequent state, and whether such a consequent state is indicated at all (Pulman 1997: 9). At the same time, Pulman points out that one verb may belong to more than one class. Therefore, he argues for a probability distribution instead of rigid class affiliation: “The assumption that this matter is lexically determined does not preclude a verb being assigned to more than one category: a more robust treatment might actually be to assign a probability distribution over each verb for each category.” (Pulman 1997: 9)

(b) Johanson (2000) Johanson points out that not all elements of the semantic representation of a verb are equally important for aspect. The relevant parameters are those of the ‘internal phase structure’ (IPS) that constitute the aspect-sensitive actional elements basic to the use of viewpoint (aspect) operators: “Actional content parameters of particular relevance for viewpoint realisations are subsumed under the internal phase structure (IPS). They do not concern the perspective applied to an event, but constitute aspect-sensitive actional categories basic to the use of viewpoint operators as terminality categories.” (Johanson 2000: 58)

Every state of affairs consists of three main internal phases: beginning (terminus initialis), course (cursus) and end (terminus finalis). The decisive classificatory criterion is that of transformativity, i.e. the question, whether the encoded event has a natural turning point or not. A further distinction is made between initio- and finitransformativity on the one hand and nontransformativity on the other hand. Additional features are momentaneity [±mom] and dynamicity [±dyn]. This yields the following categorisation of predicates (Johanson 1996: 234, 2000: 58):

2.35 IPS Category The actional content is conceptualised

terminative [+t] as implying transformation finitrans as implying final transformation [+mom] without a salient cursus [-mom] with a salient cursus initiotrans as implying initial transformation non-terminative [-t ] without transformation [+dyn] as dynamic [-dyn] as static Johanson’s proposal is interesting in at least two respects: it distinguishes two kinds of boundaries, and it takes into account whether the boundary is initial or final, ( 2.36). What he calls ‘relevant limit’ corresponds to the actual boundary, his ‘crucial limit’ to the inherent boundary, i.e. to the culmination point at which a transformation from one state into another takes place. Nontransformatives do not have an inherent limit, the relevant limit is identical to the initial or final limit that may be specified by lexical means or pragmatically induced, i.e. inferred from world knowledge telling that every action normally has a beginning and an end (cf. also Breu 1985). This is summarised in ( 2.36): 2.36 [+tf] ------------⊗

[+ti] ⊗-----------

[-t] x------------

(x = relevant limit; ⊗ = crucial limit)

Chapter 2: Setting the Stage

45

The presence of a limit in the actional content of the predicate is decisive for the interpretation, as it may be focused upon by aspectual operators. With nontransformative VP-types the relevant limit, i.e. the actual boundary, is identical to the initial boundary of the action, with transformatives this is the decisive boundary at which the culmination, the change-of-state, takes place. This may be at the beginning (initiotransformativity) or at the end (finitransformativity) of the cursus (Johanson 2000: 59). The feature [±dyn] will prove crucial for aspect (chapter 4, section 4.1.2), as aspect markers may be sensitive to the distinction between dynamic and stative phases.

The inherent boundary divides the denoted state of affairs into a preparation phase and consequent state which aspect operators may be sensitive to as well (cf. chapter 4, section 4.4 on Turkish). Such a focussing is of aspectual nature and requires the application of aspect operators. The mere presence of a limit does not mean that it has to be focused on. Actional and aspectual notions therefore have to be strictly kept apart. Note that focusing does not mean that the non-focused parts of the actional content are cut off. Rather, they are backgrounded and constitute the basis for possible inferences. (c) Bickel (1996) Bickel’s approach searches for primitive basic elements grammatical aspect is sensitive to. Their respective combination then yields the set of possible Aktionsarten (cf. also chapter 4, section 4.1.2): “[G]iven the potential range of cross-linguistic variation in lexical structure, it is helpful to have Aktionsart structures generated by a small inventory of primitives rather than to postulate a predetermined set of Aktionsarten […]. Thus, I advocate a strictly compositional theory of Aktionsart rather than the reification of an a priori set with ‘totally static’, ‘inceptively static’, ‘activity’, ‘gradually terminative’ and ‘totally terminative’,43 or, with the Vendlerian classes ‘state’, ‘activity’, ‘achievement’ and ‘accomplishment’.” (Bickel 1996: 195)

The basic units Bickel assumes for a classification of verbs in aspectual terms are phases and boundaries. Boundaries can be defined minimally as changes from one situation into another, phases as temporally extended structures. Phases and boundaries may be specified in the lexical entry of the verb or induced by further processes – semantic, syntactic and pragmatic induction are treated as being of equal status. For further operations only their presence is relevant, and not the way they are induced (cf. the Aspectual Uniformity Hypothesis in ( 2.40)). Verbs are classified according to the phases (ϕ) and boundaries (τ) they encode, and their temporal order. These elements are located on a special tier44, the temporal tier, which is summarised in Bickel’s (1996: 198) temporal tier theorem: 2.37 TEMPORAL TIER THEOREM:

There is a cross-linguistically significant set of morphemes traditionally called aspect markers, which operate exclusively on temporal information (the ‘temporal tier’) in semantic structure.

These elements allow for five verb classes (Bickel 1996: 18; cf. also Michaelis 2004)45:

43 Such a classification is proposed, e.g., by Breu (1996; 2000). 44 Jackendoff (1987) distinguishes a temporal, an actional and a thematic tier, which are each responsible for

different specifications, but also interact with each other. Each layer specifies different elements decisive for the interpretation of a lexical item, in our case, verbs. The temporal layer is the one decisive for the interaction of Aktionsart and aspect. It is this tier where phases and boundaries are located on (cf. also Michaelis 2004).

45 In chapter 4, section 4.1.2 a set of possible Aktionsarten will be established according to more fine-grained criteria, taking into account the distinction between dynamic and static phases.

Chapter 2: Setting the Stage

46

2.38 [ϕ] durative

[τ ϕ] ingressive-phasal

[τ ϕ τ] delimitative

[ϕ τ] telic

[τ] momentaneous The durative class corresponds to Vendler’s states and activities that are both characterised by the absence of inherent limits. The differentiating factor is the feature of dynamicity which does not directly concern the universal aspect categories but further subdivides IMPERFECTIVE-markers (cf. chapter 4, section 4.4 on Turkish). The delimitative class may also pertain to Vendlerian states and activities as both can be bounded by external temporal boundaries. Bickel argues for the distinction between the durative and the delimitative class on the basis of Russian perdurative and delimitative verbs. The delimitative class, however, differs from the others in one important respect: its boundaries are actual boundaries and not inherent ones as is the case with the other Aktionsarten. For the further implementation of Bickel’s selectional theory (cf. chapter 4) this does not matter, but still, inherent and actual boundaries are different and may not be conflated for an adequate analysis of the semantics of aspect46. Bulgarian has verbs that specify these temporal boundaries in their lexical entries and do not need prefixes for that. A few examples (Lindstedt 1985: 180) are listed in ( 2.39): 2.39 prekaram (‘to spend’), spra (‘to stop’), ostana (‘to stay’), prodălža (‘to continue’),

zadărža (‘to keep’) The elements encoded by the verb may be supplemented by phases and boundaries provided by other constituents of the utterance that make up the VP47. For aspect to apply it does not matter whether these decisive units are specified by the verb or added by additional material – they all have the same representational format, i.e. phases and boundaries. This is captured by Bickel’s Aspectual Uniformity Hypothesis (1996: 204): 2.40 ASPECTUAL UNIFORMITY HYPOTHESIS

Aspect and Aktionsart representations have the same format, viz. configurations of ϕ and/or τ [phases and/or boundaries, B.S.], and this format is the same on all levels of meaning composition (lexical semantics, morphological derivation, syntactic composition, and pragmatic enhancement).

This hypothesis justifies the definition of Aktionsart presented at the beginning of this section, namely as referring to a classification scheme pertaining to verb classes or verb phrases. Both are defined according to the same features – phases and boundaries.

With respect to phases, a distinction can be drawn between dynamic and static phases (Bickel 2000). The phases that precede an inherent boundary are presumably always dynamic, whereas the phases following the inherent boundary may be both. Ingressive verbs consist of an

46 The matter of boundedness is rather complex. It will be dealt with in more detail in chapter 4, section 4.2.1. 47 Compare Babko-Malaya (1999: 59f) who distinguishes three different kinds of <ϕdyn, τ>-composition:

either the lexical entry of the verb specifies this structure (Ivan otkryl dver’ – ‘Ivan opened the door’), or a <ϕdyn>-predicate combines with a resultative prefix (Ivan vykopal klad – ‘Ivan dug out a treasure’), or a <ϕdyn>-predicate combines with a perfectivising prefix (Ivan pročital knigu – ‘Ivan read the book’). The differences between these structures are irrelevant for aspect application, but they result in different syntactic configurations.

Chapter 2: Setting the Stage

47

inherent boundary followed by a dynamic phase, inceptive verbs of an inherent boundary followed by a static phase (cf. Breu 1998).

(d) Event nucleus

The fact that the class affiliation of verbs my be different at the verb and the VP-level, and the existence of certain recategorisation phenomena suggest that a verb classification should also predict which changes may possibly occur. If certain verbs are said to belong to a certain class, then all members of that class should behave consistently with respect to all possible reinterpretation processes. Verbs belonging to one class are expected to show the same predictable behaviour when confronted with items that may induce changes in their inherent structure.

The basic verb meanings and their recategorisation potential can be brought together by the notion of event nucleus as proposed by Moens (1987) and Moens & Steedman (1988). This event nucleus delivers the conceptual basis for Aktionsart classification and recategorisation phenomena48. As will become evident below, Bickel’s classification of verbs and VPs in terms of phases and boundaries relates to the conceptual structure of events. Johanson’s terminality categories are also supported by the assumption of the event nucleus.

The shifting of verbs between different classes is not at random; the freedom of reinterpretation is neither unconstrained nor arbitrary: “[M]any of the permissible transitions between aspectual categories [...] appear to be related to a single elementary contingency-based event structure.” (Moens & Steedman 1988: 18)

Moens & Steedman call this event structure an ‘event nucleus’ defined as a structure comprising a preparation process, a culmination point and a consequent state. As Moens (1987) and Moens & Steedman (1988) show, all possible recategorisations refer back to, and are constrained by, a network based on an event nucleus. This network captures possible transitions between aspectual categories of verbs (Aktionsarten). The meaning of all these categories is associated with an event nucleus defined as follows: “[A] complex entity consisting of a preparatory process, which can lead up to (without necessarily reaching) a culmination point, and this in turn has certain consequences attached to it.” (Moens 1987: 47)

Schematically, the event nucleus can be represented as follows:

Figure 2.3 Such a nucleus is not just a temporal ordering of its constituents, but establishes a stronger link between them, namely contingency relations like, e.g., causality. The aspectual and temporal information a verbal expression conveys is related to the concept of a nucleus. An expression may be associated with different parts of the nucleus: “If the expression belongs to the culminated process category, it conjures up a complete nucleus, i.e. the point at which the event finished, the preparation leading up to this point and its consequences; a process expression only pertains to the (preparatory) process; culmination expressions focus on the culmination point and its

48 A similar approach is van Lambalgen & Hamm’s (2005) event calculus, which is based on the assumption

of human’s goal-driven conceptualisation of events.

preparatory process consequent state

culmination

Chapter 2: Setting the Stage

48

consequences; points refer to the culmination without considering the consequences; and expressions belonging to the category of consequent states focus exactly that part of a nucleus.” (Moens 1987: 48)

The nucleus represents the maximum of components an event may consist of. The parts of the event nucleus constitute the conceptual counterpart of the linguistically encoded elements grammatical aspect is sensitive to49. If world knowledge allows for it, every event may be completed to a whole nucleus. Grammatical aspect is one of the means to achieve such a completion (by means of it input requirements; cf. chapter 5, section 5.2.2). Verbs encode the elements of the nucleus in terms of phases and boundaries as illustrated in Figure 2.4. Aspect is sensitive to these elements in that language-specific aspect categories are marked with respect to exactly which part of the nucleus, or which unit of the Aktionsart, they interact with. Aspect markers make explicit the relation of a denoted event to the nucleus (chapter 4, section 4.1). Aktionsart thus can be understood as the intermediate level between the conceptual level of event structure and the grammatical level of aspect. It functions as mediator between the conceptual construal of the situation and aspect marking: “[T]hese notions of boundary and phase do not directly refer to our conceptualisation of situations but […] they are mediated by lexical Aktionsart and similar time structures on higher levels (such as a verb complex or even a whole clause).” (Bickel 1996: 194)

In languages that do not have grammatical aspect this relation is induced as a default from the Aktionsart or specified by lexical means. Grammatical aspect may confirm the default induced by the Aktionsart or override it by triggering an accommodation process (chapter 5), caused by the input requirement aspect imposes on the predicate it applies to.

Figure 2.4

An approach similar to the tripartite event structure proposed by Moens & Steedman is put forward by Passonneau (1987, 1988) and her conception of transition events. She regards such events as complex situations consisting of a process which culminates in a new state or process, coming into being as the result of the initial process. Her argument for the existence of these three components of transition events derives from the fact that time adverbials may modify one of the three distinct parts of the predicated event (1988: 47):

49 The question is how to deal with stative eventualities and how to relate them to the event nucleus. Moens

& Steedman (1988: 23) argue that states do not enter into contingency relations and therefore are not part of an event nucleus. They may however be coerced into events, e.g., by imposing temporal boundaries as is the case with prosuščestvovat’ (‘to exist for a while’). In this case, then, they enter into the contingency relations and become part of the event nucleus.

preparatory process consequent state

culmination

event structure conceptual knowledge

Aktionsart linguistic knowledge

ϕdyn τ ϕstat

GRAMMATICAL ASPECT

aspect grammatical knowledge

Chapter 2: Setting the Stage

49

2.41 a. It took 5 minutes for the pump to seize.

b. The pump seized precisely at 14:04:01.

c. The pump was seized for two hours. The durative adverbial in ( 2.41a) applies to the preparatory process in which the pump was in the process of seizing. The clock time in ( 2.41b) specifies the moment when the pump is said to have made a transition into the new state of being seized which again is measured by the interval two hours in ( 2.41c). The boundary between the two intervals – the dynamic preparatory process and the consequent state – is defined as the transition bound. Passonneau (1988: 48) takes this to be “a convenient abstraction for representing how transitions events are perceived and talked about”. In theory, there has to be a point in time before which the new situation does not exist and subsequent to which it does. The notion of transition bound / culmination point / boundary is crucial to the theory of aspect developed here.

An interesting proposal of how to classify the grammaticalisation of the possible ‘stages’ of an event is provided by Desclés (1993). These stages are located on a “continuum d’accomplissement croissant” (1993: 4) and can be related to the event nucleus as follows:

Figure 2.5

This continuum relates to the event nucleus insofar as each of the stages an event may consist of can be grammaticalised. Not all languages grammaticalise all of these stages. Russian, for example, grammaticalises the ‘achevé’-stage (by means of the pf aspect) whereas the other stages may be interpretations of the unmarked form. Turkish grammatically encodes the inaccompli- and the attributif-stage by means of -iyordu/-mekteydi and -miş, achevé being one of the possible interpretations of the unmarked form -di (chapter 4, section 4.4). Bulgarian is different from both Russian and Turkish as it grammatically encodes achevé vs. non-achevé in terms of pf and ipf aspect, and in addition to that also accompli vs. non-accompli by means of the Aorist/Imperfect distinction (chapter 4, section 4.3). The accompli-stage is expressed by Egg’s (1994) class of ‘intergressives’ (cf. footnote 39).

To sum up this section, it is reasonable to classify verbs into categories which reflect certain properties of the encoded event, namely those that are relevant for grammatical aspect – phases and boundaries. This is justified by the fact that there exist languages whose aspect categories are sensitive to this level of verb-class composition (more on that in section 2.2.3). Terminativity is a property that is finally specified on the VP-level. Certain verb classes are characterised by having a terminativity-value that may be different on the V- and VP-level (cf. Rothstein 2004: 29-35). Build a house and build houses are VPs that are both headed by an accomplishment verb but differ in their telicity-feature: the former is telic, the later atelic. The aspectual class of a verb also determines the shifts it may undergo. Such shifting operations do not completely change the Aktionsart basis, but operate on it in a way that

inaccompli = non accomplissement

accompli achevé attributif

preparatory process culmination

consequent state

Chapter 2: Setting the Stage

50

preserves the information conveyed by the original predicate. This is captured, e.g., by Michaelis’ principle of ‘Aktionsart preservation’ (2004: 16): “In aspectual mapping, whether implicit or explicit, input and output types must share some portions of their respective causal and/or temporal representations.”

Shift operations do not blur the aspectual distinctions of verbs, but make use of them. Such shift, or coercion, operations will be examined in more detail in chapter 5.

2.2.3 Aktionsart and aspect This section will substantiate the claim put forward in section 2.2.2, namely that it is reasonable to classify not only VPs but also verbs. It will be shown that the aspect systems of different languages may be sensitive to different levels – terminativity as specified at the VP level in the case of Russian50, the verb class level (static vs. dynamic) in the case of English and the level of verb class components (phases and boundaries) in the case of Turkish. Importantly, this sensitivity does not contradict the strict separation of Aktionsart and aspect. This section will also reveal another interesting difference between the Russian system on the one hand, and the Turkish and English systems on the other. In Russian, where there is no consistent overt aspect marking, aspect assignment and aspect interpretation rely on default aspect. The notion of default aspect explains why verbs that are not overtly marked for aspect are still assigned a specific aspect value. Systems like the English or Turkish ones do not need the assumption of default aspect, as there is consistent overt marking and non-marking.

In section 2.1 it has been pointed out that the notion ‘aspect marker’ does not suggest or require strict overt marking, and even in case of overt marking, it might still not be consistent with respect to the information encoded. Two of the difficulties connected with Russian aspect concern the fact that it is not always morphologically encoded, and that it lacks uniform formal means for its expression. But despite this lack of consistent overt marking, every verb is characterised as either ipf or pf. Simplex verbs may be pf or ipf, depending on the terminativity value of the Aktionsart basis. In cases of simplex verbs that are regularly ipf and of simplex verbs that are regularly perfective, the lacking overt grammatical form is contributed by a semantic default (Steube 1997: 218).

In chapter 4, section 4.1.1, grammatical aspect will be defined along the lines proposed by Klein (1995, 2001), namely as establishing a relation between a sentence base and a topic component, i.e. as constraining the assertion in a specific way. In the languages under discussion here (Russian, Bulgarian and Turkish), an assertion is made by finite forms and clauses. Although infinite forms of a verb do not mark assertion, each nonfinite verb and bare stem51 in Russian can be assigned one of the two aspects52. Therefore, and because of the lack of consistent overt marking, the aspect feature (perfective or imperfective) is assumed to be part of the lexical knowledge associated with the verb stem (cf. Klein 1995; Sell 1994; Filip 1999). But how can this be conciliated with the fact that aspect marks assertion and thus is part of the finite component, i.e. interpreted at the propositional level corresponding to the syntactic IP-level (Filip 2001, 2005)?

Klein (1995: 690) argues that each lexical entry of a Russian verb is specified for a feature [±p], which is to some extent derivable from the morphological form of this entry (cf. 2.42). This feature is merely a property of the lexical entry and is not to be confused with grammatical aspect itself. The aspectual differentiation comes into play only when the verb

50 Such aspectual systems are also interesting with respect to the expression of definiteness (for an analysis of

the interplay of aspect and definiteness cf. e.g., Leiss 2000; Filip 1997). Aspect systems that are not based on terminativity do not exhibit an interaction between aspect and definiteness.

51 Klein (2001) emphasises the distinction between marked and bare infinite forms. 52 This is different in English and Turkish, where bare stems do not contain an aspect feature.

Chapter 2: Setting the Stage

51

becomes part of a finite construction, i.e. with the linking of a sentence base to a topic component. Then the temporal scope of the assertion is differentiated according to whether the form is pf or ipf.

Filip (2004, 2005), following Carlson (2000), distinguishes two semantic levels: the level of event semantics and the level of propositional semantics. Aspect is interpreted on the latter level (see below). Klein’s feature specification in terms of [±p] pertains to the level of event semantics, that is, to the level of verb and verb phrase. What, then, are the criteria for the assignment of [+p] or [-p]? Is it idiosyncratic or does it follow certain predictable regularities? Klein (1995: 690) formulates the following rules for this specific kind of lexical knowledge: 2.42 a. Each lexical verb in Russian is either [+p] or [-p], unless it belongs to a limited

list of ‘ambiguous entries’.

b. Each lexical verb in Russian is [+p] unless: (i) it is morphologically simple and does not belong to a limited list of

exceptions, or (ii) it is marked by the suffix -iv/-yv [...]

According to ( 2.42), all simplex verbs are ipf unless they belong to some undefined, seemingly idiosyncratic list of exceptions. ‘Morphologically simple’ verbs express 1-state contents in Klein’s (1994, 1995) terminology, i.e. they are not lexically stative like 0-state contents, and do not encode an inherent limit like 2-state contents which specify a source state and a target state (‘preparation phase’ and ‘consequent state’ in the terminology used here). The ‘list of exceptions’ Klein refers to is not as limited as he suggests it to be, and not as idiosyncratic as it might seem at first sight. Simplex verbs may very well be pf, without being exceptional or marked. The aspectual feature associated with the lexical entry of simplex verbs depends on the terminativity-feature of that verb. That is, simplex aterminative verbs tend to be associated with the feature [-p], simplex terminative verbs with the feature [+p]. This correlation has been noticed, among others, by Schoorlemmer (1995), Johanson (2000) and Bohnemeyer & Swift (2001, 2004). The latter argue for a ‘default aspect’ in Russian based on the terminativity feature of the predicate, the respective correlation being the following: 2.43 a. terminative predicate → pf aspect

b. aterminative predicate → ipf aspect Bohnemeyer & Swift derive this correlation from cross-linguistic data and evidence from early child language that shows a preference to refer to terminative events with the pf aspect, to aterminative ones with the ipf aspect53. Further manifestations of the proposed correlation are morphological markedness relations and aspectual interpretation of predicates that are not overtly marked for aspect. This explains why aspect assignment is possible for Russian verbs, although there is no consistent morphological marking of grammatical aspect, except for the imperfectivising suffix -(y)va-54 and the perfectivising -nu-55. Bohnemeyer & Swift account

53 Compare also language-contact data, where terminative verbs of the dominant language are borrowed as

perfective, and aterminative verbs as imperfective (Breu 2003 finds this correlation for Molisean Slavic). 54 The notation -(y)va- is meant to represent the allomorphic for the different variations of this suffix. 55 Some prefixes could probably also be regarded as perfectivisers. These prefixes apply to process verbs that

may regularly change their terminativity feature on the VP level, e.g., in case of combination with a direct object. Compare po- as in stroit’ (ipf; ‘build’) – stroit’ dom (ipf; ‘build a house’) – postroit’ dom (pf, ‘build the house’), or s- as in delat’ – delat’ č-t. – sdelat’ č.-t. (‘do – do s.th.’). Compare also Babko-Malaya (1999: 50f) who distinguishes perfectivising prefixes that mark a process as completed without changing the selection restrictions

Chapter 2: Setting the Stage

52

for the correlation between the terminativity of the verb and its aspectual specification in terms of a notional aspect operator ‘DASP’ (default aspect), whose unified interpretation is ‘event realisation’. This operator interprets the lexically specified feature [±p]. DASP presents events as realised without providing further information about them. It provides the minimal aspectual interpretation of an event predicate that entails event realisation. Aterminative predicates merely require the ipf aspect to assert the realisation of the event, whereas terminative verbs require the pf aspect which is more complex, more specific in that it provides more information about the event, namely that the inherent boundary is reached – which is necessary for the realisation of events denoted by a terminative predicate (a detailed analysis of the semantics of pf and ipf will be provided in chapter 4): “DASP simply selects a topic time of an event such that it is realized at this topic time – nothing more (not completion in the case of atelic predicates) and nothing less (not ongoingness in the case of events denoted by telic predicates.” (Bohnemeyer & Swift 2001: 9)

The notional aspect operator DASP selects a minimal topic time for the assertion of the event predicate, such that the event e denoted by a predicate P is realised. For a denoted event to be realised at a certain time interval t, this interval has to contain the run time of a certain part e’ of that event. The decisive part differs with terminative and aterminative predicates. Bohnemeyer & Swift (2001: 7, 2004: 286) define event realisation as follows: 2.44 event realisation

∀P,e,t ⊆E[REALE(P,e,t) ↔ P(e) ∧ ∃e’[P(e’) ∧ e’≤E e ∧ tSIT(e’) ⊆ t]] This is incorporated in the definition of DASP as follows: 2.45 default aspect

DASP := λTTminλPλe[REALE(P,e,tTTmin) ∧ ∃rT∈{⊆, ⊃} [rT(tSIT(e), tTT)]] The default aspect operator has built in the condition of event realisation REALE(P,e,t). The time interval within which the event is realised is specified and replaced by the topic time interval tTTmin, yielding REALE(P,e,tTTmin). The temporal relation rT between tSIT(e) and tTT incorporates the general semantics of aspect – the relation between topic time and situation time – that will be defined in chapter 4, section 4.1.1. DASP asserts the denoted event to be realised at TT, and it introduces a specific temporal relation rT between the situation time tSIT (i.e. the run time of the event) and the topic time tTT. The topic time is minimal in the sense that only such parts of e fall in it that are entailed by REALE(P,e,tTTmin). As noted above, the relevant parts e’ of e depend on the terminativity of the predicate. For a terminative predicate to be true, the culmination point (boundary) τ56 must be included in the topic time (the temporal relation rT being specified to tSIT(τ) ⊆ tTT), whereas for aterminative predicates any part of e may be included in the topic time for the event to be realised, and some parts of e may be outside of TT. No culmination point is asserted within TT, because this would violate the minimality-condition specified by tTTmin.

To recap, a terminative predicate is realised if the culmination point is included in TT and the temporal relation is specified as tSIT(τ) ⊆ tTT. With aterminative predicates, any part of the event denotation may be included in TT. Bohnemeyer & Swift establish the relation tTT ⊂

of the verb, and resultative prefixes that indicate completion and a result, thereby changing the selection restrictions of the verb.

56 By specifying the situation time this way, I deviate from Bohnemeyer & Swift, who stick to Klein’s (1995) analysis and specify this relation as tSIT(e) ⊆ tTTmin.

Chapter 2: Setting the Stage

53

tSIT(e). This relation, however, is valid only for cases of marked imperfectivity like the English progressive, but it cannot be applied to Russian. The Russian ipf aspect does not express exactly that relation, but a general overlap relation (chapter 4, section 4.1.2) that needs to be specified further. For Russian, it is therefore more adequate to replace Bohnemeyer & Swift’s specific relation by the general relation tTT ○ tSIT(e).

The assignment of the pf aspect to a terminative predicate is a default, applying in case nothing else happens. This default may be overridden by imperfective triggers (Schoorlemmer 1995), specifically by means of the suffix -(y)va-, compare ( 2.46): 2.46 a. dat’ – davat’

give:INF:pf – give<(y)va>INF:ipf ‘to give’

b. otkryt’ – otkryvat’ open:INF:pf – open<(y)va>INF:ipf ‘to open’

The overriding of the default from terminative-pf to terminative-ipf has to be marked morphologically. This holds for an override in the other direction as well, i.e. for the conversion of aterminative ipf into aterminative pf, compare ( 2.47): 2.47 a. sidet’ – posidet’ sit:INF:ipf – PRF:sit:INF:pf ‘to sit’ – ‘to sit for a while’

b. suščestvovat’ – prosuščestvovat’ exist:INF:ipf – PRF:exist:INF:pf ‘to exist’ – ‘to exist for a while’ As will be seen below, the prefixes in ( 2.47) are different from lexical prefixes which derive new verbs that carry their own default aspect in their lexical entries, as is the case in ( 2.48): 2.48 a. pisat’ – perepisat’

wite:INF:ipf – PRF:write:INF:pf ‘to write’ – ‘to re-write’

b. dumat’ – zadumat’ think:INF:ipf – PRF:think:INF:pf ‘to think’ – ‘to plan’

The prefixes used in ( 2.48) do not encode grammatical aspect, but modify the basic event by making it terminative. This yields the default assignment ‘pf’, which again may be overridden by the suffix -(y)va- turning the terminative-pf default into terminative-ipf ( 2.49): 2.49 a. perepisat’ – perepisyvat’ rewrite:INF:pf – rewrite<yva>INF:ipf ‘to rewrite’

b. zadumat’ – zadumyvat’ plan:INF:pf – plan<yva>INF:ipf ‘to plan’

The correlation between aspect and terminativity in Russian can be summarised as follows (cf. Bohnemeyer & Swift 2004: 266):

Chapter 2: Setting the Stage

54

PREDICATE

aterminative terminative imperfective ∅ overtly expressed ASPECT perfective overtly expressed ∅

Table 2.3 In Russian, DASP is an entailment. Simplex terminative verbs in Russian are per default assigned the pf aspect, simplex aterminative verbs the ipf aspect. Deviations from this default aspect assignment are morphologically marked.

It is important to note that the notion of ‘default aspect’ does not pertain to all languages with grammatical aspect but only to those that predetermine the respective aspect feature as default already in the lexicon, like Russian. Furthermore, default aspect is restricted to languages where grammatical aspect is sensitive to the terminativity of the predicate, and where there is no consistent overt aspect marking. The notion of default aspect applies neither to English nor to Turkish. Both English and Turkish have consistent aspect marking and consistent non-marking. Therefore, there is no need for DASP – non-marked forms are interpreted according to their non-marked status, i.e. as being neutral towards the feature specification of the marked form57. In such cases, semantic marking corresponds to formal marking. The aspect categories of both languages are not sensitive to the terminativity of the predicate in the way Russian is.

English has one morpheme that unambiguously assigns aspect, namely the progressive marker -ing which consistently applies to dynamic events. The other member of the category, the simple form, may receive different interpretations depending on the lexical class of the verb and on the context in which it is used, but not on the terminativity of the predicate. The aspectual interpretation of the simple form is sensitive to the difference between process verbs and state of change verbs, but this sensitivity is merely an implicature and no entailment (Bohnemeyer & Swift 2004: 276f). Progressive marking in English depends on dynamicity, that is, on whether the event description entails a stage-of relation (chapter 4, section 4.1.2). It does not depend on terminativity, and neither does the interpretation of the simple forms. The perfective-like interpretation of the simple form is derived by Q-inference (chapter 5), but is not encoded in the semantics of this form. Dynamicity based aspect marking in the case of English is summarised in Table 2.4 (Bohnemeyer & Swift 204: 277):

VERB CLASS

stative dynamic IMPERFECTIVE58 ∅ overtly expressed ASPECT PERFECTIVE ∅ ∅

Table 2.4 Stative verbs that are not overtly marked for aspect by the progressive form can be interpreted as both PERFECTIVE or IMPERFECTIVE, whereas unmarked dynamic verbs are interpreted as PERFECTIVE. This interpretation is a default only, and it can be overridden by the following context as illustrated in ( 2.50), adopted from Žegarac (1993: 201):

57 In chapter 5 (cf. especially section 5.2.3) this will be accounted for in terms of Q-inference. 58 The terms PERFECTIVE and IMPERFECTIVE are used in a meta-categorical sense. Compare Table 2.3, where

the terms ‘perfective’ and ‘imperfective’ are used to refer to Russian.

Chapter 2: Setting the Stage

55

2.50 Q: How did Jane spend the morning?

A: Oh, she read her new book all morning. I suppose she is still at it. The PERFECTIVE-interpretation that arises from the first sentence of the answer is cancelled by the next sentence. This second sentence does not contradict the first one, and therefore the PERFECTIVE-reading of she read her new book is not semantically encoded. Simple forms in English are not marked for perfectivity in the same way as the Russian pf aspect. That aspect marking does not depend on the terminativity feature of the Aktionsart is illustrated with ( 2.51), where read her new book is terminative, sleep is aterminative, but both are marked with the progressive: 2.51 Q: How did Jane spend the morning?

A: Oh, when I saw her, she was reading her new book. Or maybe she was just sleeping. What is thus important for aspect in English is the presence of a dynamic phase provided by the Aktionsart.

Roughly the same holds for Turkish which has an aspect system that is morphologically more explicit than that of English (for a detailed analysis see chapter 4, section 4.4). Aspect marking in Turkish is neither sensitive to the terminativity feature of the VP, nor exclusively to verb classes in terms of stative vs. dynamic. This difference between lexically stative and dynamic verbs is important for Turkish insofar, as dynamic verbs may encode dynamic and stative phases (cf. Figure 2.6). And these phases constitute the elements Turkish aspect is sensitive to. The system of aspect marking in Turkish is shown in Table 2.5:

TEMPORAL TIER UNIT

ϕdyn ϕstat IMPERFECTIVE overtly expressed overtly expressed ASPECT PERFECTIVE ∅ ∅

Table 2.5 Note that Table 2.5 pertains only to lexically dynamic verbs, as they provide the input the respective aspect markers may apply to. Contrary to English, the static phase is aspectually relevant as well, since Turkish has one explicit marker for its selection (-miş). Lexically dynamic verbs that are used in the unmarked form are assigned a PERFECTIVE-value as default. They need to be marked if they are to receive an IMPERFECTIVE interpretation, either in terms of intraterminality or in terms of postterminality, compare ( 2.52) and ( 2.53): 2.52 a. Pencereyi açtı. window:ACC open:PST ‘He opened the window.’

b. Pencereyi açıyordu. window:Acc open:iyor:PST ‘He was opening the window.’

c. Pencereyi açmış. window:Acc open:miş ‘He has opened the window.’ (He must have opened it, because it is open now.)

Chapter 2: Setting the Stage

56

2.53 a. Vapur kalktı. boat depart:PST ‘The boat departed.’

b. Vapur kalkıyordu. boat depart:iyor:PST ‘The boat was departing.’

c. Vapur kalkmış. boat depart:miş ‘The boat has departed.’ (= It is gone.) Examples ( 2.52) and ( 2.53) show that verbs encoding a dynamic or a static phase get interpreted as PERFECTIVE if neither phase is selected, i.e. with the application of -di59. This is an implicature that may be overridden by the context. The dynamic and the static phase can both explicitly be selected by specific markers. Thus, aspect in Turkish is sensitive not only to the distinction of lexically stative and non-stative verbs, but also to the internal make-up of non-stative verbs, i.e. to the kind of phase they encode.

To sum up so far, aspect systems of different languages may interact with different elements provided by the Aktionsart basis. This sensitivity is exhibited by the marked members of the respective aspect system. In Russian, the pf aspect is sensitive to the boundaries specified at the VP level. These may be inherent or temporal boundaries as is the case with delimitative and perdurative verbs. If the VP does not specify an inherent boundary and the temporal boundaries are specified by adverbials which are located outside the VP, the pf aspect cannot apply (cf. chapter 4, section 4.2). The English aspect system interacts with dynamic phases that are explicitly selected by the progressive marker60. Thus, the difference between lexically dynamic and lexically stative verbs is decisive for English. For Turkish, this distinction is decisive only indirectly, what matters here is the distinction between dynamic and static phases. Grammatical aspect in Turkish interacts with both – Turkish has markers for the explicit selection of both the dynamic phase and the static phase.

Bare verb stems in Russian, but not in English and Turkish, are already specified for a pre-aspectual feature [±p]. This feature is assigned according to the terminativity of the verb61 and explains the default aspect assignment for verb stems that are not overtly marked for aspect. In English and Turkish, aspect assignment depends on overt morphological marking and is not as predictable from verb roots and stems as in Russian. English and Turkish differ in the levels the respective aspect categories are sensitive to. In English, it is the level of verb classes with the opposition of dynamic and stative Aktionsarten, in Turkish it is the temporal tier level with the distinction of dynamic and static phases.

Figure 2.6 summarises the different levels the aspect categories discussed here are sensitive to. It illustrates the need to classify not only VPs, but also verbs. Aspect in Russian is sensitive to VP-level telicity, aspect in English to verb-class dynamicity and Turkish to the features that constitute verb classes – dynamic and static phases (boundaries do not play a role in Turkish)62. Figure 2.6 pertains only to the (semantically) marked members of the respective aspect systems.

59 This will become more evident in chapter 5, section 5.2.3, with the notion of scalar implicature and the

different scales that induce such an implicature. 60 The perfect is not dealt with here. 61 There are also verbs that specify in their lexical entry that the terminativity value can be changed at the

VP-level. These verbs are assigned [±p], and the specification is delayed until the VP-level. Default aspect applies depending on the terminativity value specified there. It seems to be the case that this group of verbs enters into ‘real’ aspectual pairs, just like simplex pf verbs (cf. footnote 55).

62 Note that the underlying assumption here is the crucial difference between the stative verb class on the one hand, and static phases that may be part of lexically dynamic verbs on the other hand.

Chapter 2: Setting the Stage

57

The respective unmarked aspect markers in Russian, English and Turkish are not constrained to a specific input. Depending on the context, they may either express a neutral value or focus upon the parts that are not selected by the marked members. Thus, ipf aspect in Russian may focus on the dynamic phase (or the static phase if encoded) if the context supports that. The English simple form and the Turkish unmarked -di may focus on the boundary in case of an appropriate context.

Figure 2.6 Back to Russian. One of the big debates in Slavic aspectological research concerns the question which pairs of verbs in Russian are to be regarded as ‘true’ aspectual pairs and where to posit the demarcation line to ‘modes-of-action’-verbs (the Slavistic sposoby dejstvija). The notion of default aspect helps to settle this issue: truly aspectual pairs are verbs that denote one and the same eventuality, in one case with its default assertion time in the other case with a change of that default. Aterminative verbs modified by lexical prefixes changing them into terminative verbs constitute new ‘simplex’ verbs with their own default aspect that may again be overridden by suffixation. The discussion about aspectual pairs is not relevant here – every Russian verb is specified for pf or ipf, independently of whether it is part of such a pair or not.

The lack of consistent aspect marking in Russian, combined with the fact that prefixed verbs are predominantly pf, easily leds to the erroneous assumption that prefixes constitute markers of pf aspect. That this cannot be the case is illustrated by a number of facts. Verbs containing prefixes may again be prefixed, and may also be imperfectivised by the suffix -(y)va- (cf. ( 2.49) above; compare, e.g., Steube 1997; Filip 2003, 2005). If prefixes were perfectivity-markers, this would amount to recursive applications of markers from one and the same grammatical category. However, this is not possible, since the members of a grammatical category stand in complementary distribution such that one grammatical category may be applied only once. Steube (1997: 216) argues that the pf aspect requires a prefixed stem to apply. If a prefixed stem is required for aspect to apply, both prefix and aspect must be independent of each other and prefixes cannot be the formal means to express pf aspect. Filip (1999, 2003, 2005) rejects the view that prefixes are morphological exponents of perfectivity arguing that they do not exhibit any of the properties characteristic for members of a grammatical category. She argues against the conflation of prefixes and pf aspect, predominantly on the basis of Czech data, but the arguments are valid for Russian as well. Prefixes often encode argument-structure changing operations (Filip 2005; Spencer & Zaretskaya 1998) which sets them apart from the imperfective suffix -(y)va- and from other formal markers of grammatical aspect, like the perfectivising suffix -nu- and some special prefixes (cf. footnote 55).

VP

ϕstat

static dynamic

ϕdyn

PREDICATE

VERB CLASS

TEMPORAL TIER

Terminativity

Dynamicity

Phase

Russian

English

Turkish

Chapter 2: Setting the Stage

58

The presence of a prefix is neither necessary nor sufficient for perfectivity, as there are simplex pf verbs and there are verbs suffixed with -(y)va- that are ipf, no matter whether they contain also prefixes. If prefixes were markers of pf aspect, they could not appear together with the imperfectivising suffix -(y)va-, as members of one grammatical category cannot apply recursively.

The examples in ( 2.54) exhibit multiple occurrences of prefixes and suffixes: 2.54 a. B auditorii pootkryvali vse okna.

in lecture.hall:LOC PRF:open<yva>PST:ipf all windows ‘They had opened all the windows in the lecture hall.’

b. Obyknovenno ja napivajus’ tak odin raz v mesjac. usually I PRF:drink<yva>PRS:ipf:1SG so one time in month ‘I usually get drunk once a month.’ [Filip to appear]

Another argument in this respect is that prefixes do not consistently encode one single meaning component that could be identified with the meaning posited for the pf aspect. Prefixes are neither sufficient nor necessary to mark aspect – there are unprefixed pf verbs as in ( 2.55a), and prefixed ipf verbs as in ( 2.55b) (for the derivation compare Figure 2.7): 2.55 a. dat’ give:INF:pf ‘to give’

b. otdavat PRF:give:INF:ipf` ‘to hand over’ As shortly pointed out above, prefixes and aspectual markers operate on different levels of the semantic interpretation (Carlson 2000). Prefixes work on the level of eventuality description, possibly changing the argument structure of the verb. Grammatical aspect, just as all other contextually-dependent elements, is located at the level of propositional semantics where it is interpreted. This intuition is captured by Klein’s (1995, 2001) characterisation of aspect as part of the finite component.

Filip (2005) proposes the following schematic representation63: 2.56 [Tense [Aspect [ Eventuality-Modifier* [Eventuality Description]]]]] This schema consists of two distinct parts: the level of event semantics including eventuality descriptions and eventuality type modifiers, and the level of propositional semantics

63 Compare also de Swart’s (1998: 348) independently proposed structure which looks as follows:

[Tense [Aspect* [eventuality description]]] Contrary to Filip, however, de Swart considers grammatical aspect an eventuality type modifier. Therefore aspect may apply recursively with her proposal, instead of only once as it is required for an element of a grammatical category. Eventuality-type modifiers are not grammatical and thus may apply more than once. Both assumptions are crucial for analysing Russian aspect, and therefore Filip’s conception is preferred here.

event semantics: VP level

propositional semantics: IP level

Chapter 2: Setting the Stage

59

consisting of aspect and tense64. These two levels are proposed in Carlson (2000) and regarded as two independent but interacting components of semantic interpretation. According to Carlson, two fundamentally different types of information have to be distinguished in semantic interpretation: […] the event-semantics portion providing a kind of “core” input into the more usual propositional semantics, which is about the interpretation of entire sentences in context. (Carlson 2000: 8)

This assumption is basic to Carlson’s semantic motivation for the fact that weak indefinites, but no other types of NPs must stay within the VP to be interpretable. Event semantics is context free and serves as input to the standard context-sensitive propositional semantics which corresponds to the IP-level. Contextually dependent elements, that is, elements with open slots that need to be filled in by propositional information from the context like tense, modality and aspect, must be located at the IP-level in order to be interpretable. Weak indefinites must stay within the VP, that is, on the level of event semantics, because they can be interpreted without reference to the context, and because they conform to the structure of VP denotations. John fed dogs is redescribable as John fed animals, i.e. feed dogs describes an eventuality that is part of the eventuality described by feed animals. This is different for arguments with strong quantifiers, that do not preserve the structure of VP denotations: John fed every dog is not redescribable by John fed every animal.

Filip (2001, 2005) applies this argument in her analysis of prefixes and aspect operators. Prefixed verbs yield weak indefinite and predicative readings for the associated nominal arguments, strong readings are missing. Strong specific readings of nominal arguments are induced by the pf aspect. Thus, she argues, prefixes pertain to the level of event semantics and extend over the VP level. They function as modifiers of eventuality types and as such may affect the terminativity feature of the respective predicate. Aspect, on the other hand, is a propositional operator with functional correspondents at the syntactic IP-level. Aspect takes eventuality descriptions as input. The relation between eventuality type and aspect operators can be stated as follows: “[...] aspectual operators (i.e., perfective, imperfective) take as their input eventuality descriptions, basic or derived with modifiers, and they are interpreted in terms of separate conditions that operate on the eventuality description.” (Filip 2001: 4)

As eventuality type modifiers, prefixes change the basic eventuality descriptions. The combination of prefix and verb therefore results in a ‘new’ verb encoding an eventuality different from that encoded by the unprefixed verb. This new verb now specifies its own default aspect, which is pf if the prefixes add some kind of limit or boundary. This default can be changed by secondary imperfectivisation.

Figure 2.7 illustrates the hierarchical structure of Slavic prefixed verbs (cf. Filip 2005):

Figure 2.7

64 There are other operators located at this level as well, but these do not play a role here.

V0 [ipf]

V0 [pf] -yva-

PREF+ V0 [±p]

inflection: grammatical aspect

derivation: eventuality types

Chapter 2: Setting the Stage

60

Prefixes can be attached to [+p] and [-p] roots. As eventuality type modifiers they pertain to the domain of derivation. Prefixed verbs may provide the appropriate input for pf aspect to apply (cf. chapter 4). The perfective V0 (the perfective value results from default aspect assignment) may again be imperfectivised by adding the suffix -(y)va-. The aspect operators are inflectional and located at the level of propositional semantics. The structure presented in Figure 2.7 can be illustrated with example ( 2.55b) above: The terminative verb dat’ is specified for the feature [+p]. The prefix ot- is applied to this verb, which illustrates that prefixes cannot only be attached to [-p] roots. The operations up to here belong to the level of VP-composition, to the level of event semantics. The denotation of the prefix modifies the denotation of the verb. By default, the combination of this verb and the prefix results in a perfective interpretation. In ( 2.55b), however, this default is overridden by the application of the imperfectivising suffix -yva-. This suffix is an aspectual operator which takes the eventuality description consisting of verb and prefix as input and thus specifies the relation between the time of the eventuality description and the topic time in a way typical of the ipf aspect.

To sum up, with respect to grammatical aspect marking languages are sensitive to different levels. Russian is sensitive to the VP-level where terminativity is specified. Russian makes also use of default aspect, i.e. in case of no overt aspect marking the predicate gets assigned a default aspect according to the terminativity of the predicate. Terminative predicates will receive an pf interpretation, aterminative verbs an ipf interpretation. English is sensitive for the level of verb classes, namely for the difference between dynamic and stative verbs. Turkish is sensitive for the temporal tier information, i.e. for stative and dynamic phases, that are both overtly marked. The main difference between the aspect system of Russian (and Bulgarian) on the one hand and Turkish and English on the other, is that the latter do not pre-specify the aspect feature already in the verbal lexical entries, and are not sensitive to terminativity. This latter fact helps to explain why Turkish aspect does not interact with definiteness, although it lacks articles just as Russian (for the interaction of aspect and definiteness see, e.g., Leiss 2000).

In Russian, one rather small group of prefixes can be taken as markers of pf aspect (perfectivising prefixes, cf. Babko-Malaya 1999). Another group of prefixes delimits the situation temporally and imposes actual boundaries on it. Other prefixes change the semantic representation of the verb by adding certain components, namely inherent boundaries (resultative prefixes, cf. Babko-Malaya 1999). The composition with such prefixes yields new simplex verbs, and these new simplex verbs are characterised by a change in default aspect (with respect to the basic verb). The basic aterminative verb was ipf by default, the new terminative verb is pf. Contrary to prefixes, grammatical aspect works at the level of propositional semantics that corresponds to the IP-level, where contextually-dependent elements are located at to be interpretable (Carlson 2000; Filip 2005). The context-free event semantics serves as input to the context-sensitive propositional semantics that concerns the interpretation of sentences in context. One consequence of this assumption is that aspect cannot be regarded as an eventuality type modifier as seems to be suggested by de Swart (1998; cf. footnote 63). Aspect is a propositional operator affecting the relation between the topic time and the run time of the eventuality description. Prefixes are modifiers of eventuality descriptions, i.e. functions that map sets of eventualities of one type onto sets of eventualities of some possibly other type. Aspect imposes specific input requirements onto the eventuality description it takes as argument. If the eventuality description does not deliver the required input, it has to be modified according to the requirement imposed by the aspect marker. This modified basis then constitutes the input for aspect to operate on. Thus, the input requirement of aspect markers may trigger certain kinds of accommodation that apply at the level of event semantics. Chapter 5 will discuss such coercion phenomena in more detail.

Chapter 2: Setting the Stage

61

Although both Filip and Carlson emphasise the importance of the propositional level where grammatical aspect is located at in order to be interpretable, they do not mention the mechanisms involved in interpretation. These mechanisms will be topic of chapter 5, but first, the semantics of the aspect operators has to be fixed (chapter 4).

2.3 Conclusion

This chapter has provided a brief outline of the main problems that arise in dealing with grammatical aspect in Russian and with grammatical aspect cross-linguistically. Some basic distinctions have been introduced that will be taken as background assumptions for the semantic analysis of aspect in chapter 4: semantics vs. interpretation, aspect vs. Aktionsart (verb class and VP-level) and terminativity vs. boundedness vs. aspect. Furthermore, the criteria for classifying verbs have been argued to be elements of the inherent phase structure, namely phases and boundaries. Out of the event properties encoded by the verb, these are the ones that are relevant for grammatical aspect. The basic conceptual structure they relate to is the event nucleus consisting of three components.

What remains to be done is to specify the exact semantics of grammatical aspect, in general terms as well as with respect to specific languages. The main focus will be on Russian, but the aspect systems of Bulgarian and Turkish will also be analysed. Bulgarian and Turkish serve to substantiate the basic claims for Russian ipf aspect, and to show the cross-linguistic applicability of the semantics proposed. Besides the semantics of aspect, the exact interaction of aspect and Aktionsart has to be specified, since this interaction constitutes the basis for the context-specific interpretation. The semantics can be kept rather simple, as pragmatics is assigned an equal role in the interpretation process. For that purpose it is necessary to specify what is understood by semantics and pragmatics, and how they interact in utterance interpretation.

The remainder is structured as follows: Chapter 3 deals with the problem of how to interpret underspecified forms and utterances. The notions of semantics and pragmatics will be defined and their interaction will be sketched. Chapters 4 and 5 apply the general statements of chapter 3 to the interpretation of grammatical aspect. The semantics of grammatical aspect in Russian will be outlined in chapter 4. The validity of the claims made for Russian will be tested by applying the analysis to Bulgarian and Turkish. Bulgarian as another Slavic language shows that Slavic languages do not behave in exactly the same way with respect to grammatical aspect. Turkish belongs to a different language family but its aspect system can be analysed along the general lines established in chapter 4. This semantic analysis will be embedded in a pragmatic framework in chapter 5. The pragmatic framework will be Levinson’s (2000a) inferential heuristics that are the appropriate means to deal with the interpretation of grammatical categories, although their usefulness might be questioned with respect to other domains of interpretation.

3 UTTERANCE INTERPRETATION

In chapter 2 it has been emphasised that in order to account for aspect and its interpretation in a systematic way it is necessary to separate aspect and Aktionsart on the one hand, and semantics and interpretation on the other. The semantic contribution of both aspect and Aktionsart is provided by the stable and context-invariant, that is, encoded, information they deliver to the interpretation process. However, this encoded information does not supply all the information necessary for interpretation, but provides merely a template that guides and constrains the inference processes needed in order to arrive at a specific interpretation in a certain context.

The interpretation of aspectualised verbs in specific contexts exhibits two central properties of natural language which, at first sight, seem contradictory and incompatible: systematicity and flexibility. Systematicity is a property of a system of representations where each of the symbols it contains occurs with the same value as a constituent of many different hosts (Fodor 2001). With respect to language, systematicity pertains to the stable information encoded by lexical items. Flexibility has to do with the variation in interpretation that is possible for encountered strings of linguistic input. It pertains to the context-dependent meaning components associated with these items in specific contexts on specific occasions of use. However, this variation is not unlimited. Rather, it is constrained by the semantic contribution of the respective lexical items. To account for utterance interpretation, both the stable contribution as well as the mechanisms guiding the inferences need to be figured out. These two features will be dealt with in this chapter, setting the stage for the semantic analysis of aspect in chapter 4 and its embedding in a pragmatic framework in chapter 5.

The systematicity of interpretation is captured by the encoded information a lexical item contributes to the interpretation process. The flexibility of interpretation can be accounted for in terms of ‘underspecification’, that is, in terms of the assumption that the information encoded by natural language expressions delivers only templates that need further elaboration in the course of the interpretation process. The gap between these templates and the final interpretation is bridged by pragmatic mechanisms that closely interact with the explicitly encoded information.

This chapter is structured as follows: Section 3.1 deals with the different types of ‘meaning’ that play a role in interpretation, and illustrates in which respects semantic information may be underspecified. Section 3.2 focuses on the question of how – based on semantic information encoded by the lexical items used – the hearer arrives at a specific interpretation.

3.1 Underspecification of ‘meaning’

Underspecification is the central notion in the account of aspect interpretation developed in this thesis. ‘Underspecification’ presupposes some kind of ‘specification’ and this specification is delivered by the explicitly encoded information and supplied by the lexical entries of the respective lexical items. Section 3.1.1 proposes a make-up of lexical entries in a way that reveals the stable part entering into the interpretation process and at the same time allows to capture the flexibility involved. This structure of lexical entries will be basic to the further assumptions concerning the contributions of aspect and Aktionsart to the interpretation of aspectualised verbs. Section 3.1.2 illustrates different ways in which lexical items may be underspecified.

Chapter 3: Utterance Interpretation

64

3.1.1 Meaning: encoding, decoding, inferring The notion of ‘meaning’ is not as clear and unambiguous as it might seem at first glance. In fact, it is possible to speak of different kinds of meaning, depending on what kind of linguistic unit is taken as basis. Talking about meaning requires specifying the basis ‘meaning’ is assigned to. Among the different kinds of meaning that are relevant in linguistic analyses are word meaning, sentence meaning, utterance meaning, speaker’s meaning, hearer’s meaning and discourse or text meaning. This section deals with what is understood by these different kinds of ‘meaning’ and how meaning may be assigned. The two main processes of meaning assignment involved in interpretation are decoding and inferring.

Word meaning is assigned to a single lexical item in isolation. Lexical items may encode two different kinds of information: conceptual information delivered by the root and derivational morphemes on the one hand, and procedural information delivered by, e.g., grammatical categories1, discourse connectives or markers of speaker’s attitude (see below) on the other hand2. The word meaning assigned to a lexical item constitutes an instruction to access the concept activated by that item (conceptual information) as well as instructions of how to specify this concept and of how to embed it in the wider structural context (procedural information). This is illustrated in ( 3.1) with a rather simple example: 3.1 ‘walked’

word meaning: (i) conceptual information: activate concept <WALK> (ii) procedural information: put it prior to time of utterance

The root walk- delivers the conceptual information to activate the concept associated with it, the past tense morpheme -ed encodes the procedural information to put the concept in the past3. The internal structure of conceptual and procedural information will be dealt with in more detail below.

Sentence meaning pertains to an uninstantiated sentence, i.e. to a sentence not used in a specific context. It is its context-invariant meaning composed out of the context-invariant information provided by the lexical entries of the individual lexical items that are combined according to the respective instructions encoded by them. The way things are combined and put together is decisive at this level, and here it is possible to speak of compositionality in the strict sense. At this level, ambiguities are not resolved yet and time and place variables still have to be specified.

Utterance meaning pertains to a sentence used in a specific context. It constitutes the meaning of an instantiated sentence, i.e. a proposition. At this level, disambiguation has taken place, as well as reference resolution. Levinson (1995) argues that ‘utterance meaning’ is too coarse grained and needs to be separated into ‘utterance-type’ and ‘utterance-token’ meaning (cf. section 3.2.2). This distinction will turn out crucial to the pragmatic analysis of aspect developed in chapter 5.

Speaker’s meaning pertains to an utterance used in a specific context on a specific occasion by a specific speaker. It is sentence meaning plus the specific kind of concept activation the speaker had in mind using that sentence (speaker’s intention). Speaker’s meaning includes additional aspects of meaning based on utterance meaning, predominantly implicatures in the original Gricean sense, that is implicatures mapping propositions onto additional propositions.

1 These need not be overtly marked. 2 For the distinction of conceptual and procedural encoding compare Blakemore (2002). 3 The procedural information does not always need to be morphologically, i.e. overtly, encoded. This will

prove important with aspect, especially in Russian (cf. the remarks in chapter 2, section 2.1).

Chapter 3: Utterance Interpretation

65

Hearer’s meaning is the meaning assigned to the natural language string processed by the hearer, i.e. utterance meaning as interpreted by the hearer. Hearer’s meaning does not necessarily have to be absolutely identical with speaker’s meaning, in fact, it seems reasonable to claim that this is hardly ever the case. It is the reconstruction of the intended communicative content of the speaker’s utterance. This kind of meaning is referred to by the notion of ‘interpretation’ throughout this thesis. It constitutes its focus, as the overall question pertains to what processes are involved in interpreting an aspectual form.

An interpretation can be assigned to a single sentence used in a certain context, but also to larger units – to whole discourses or texts. At this level, again, one has to distinguish between speaker’s and hearer’s meaning. The notion of ‘text meaning’ is under vivid discussion (cf. ACL 2004), as it is anything but clear what exactly is to be understood by the term. It is neither clear which unit counts as ‘text’, nor whether text meaning is simply determined by the sum of its single smaller units, that is, sentences, or whether it is more than the sum of it. These questions will not be touched here, but it should at least be noted that bigger units than sentences may also be important for interpretation.

Figure 3.1 sketches the role these different kinds of meaning outlined above play in the communication process:

Figure 3.1

A lexical item contributes the information (conceptual and/or procedural) encoded in its lexical entry to the overall sentences structure. This sentence structure is combined according to the instructions specified by the individual lexical items and constitutes the common basis for speaker and hearer. In communication, a speaker usually aims at conveying a certain intention (e.g., information). This intention needs to be made public in order to address the hearer. To this end the speaker needs to pack his intention into lexical items. Since lexical items encode only part of what is to be conveyed, this intention is not completely encoded, but specified to a certain degree by its use in a specific context (‘utterance type’). Therefore, the speaker has to rely on certain mechanisms of completion that are available to the hearer as well. These mechanisms allow the hearer to enrich the utterance type meaning in order to arrive at an interpretation – his utterance token – that is as close to the speaker’s meaning as possible. Both speaker and hearer have a certain utterance token in mind, conveyed by and encountered as an utterance type. The utterance type meaning is built up compositionally from the (disambiguated, if necessary) instructions delivered by the lexical items used; the interpretation derived from this utterance type meaning is guided by inferential heuristics that give rise to defeasible, non-monotonic inferences and therefore do not adhere to the principle

utterance type:

heuristics

word meaning

sentence meaning HearerSpeaker

hearer‘s meaning utterance token interpretation

speaker’s meaning utterance token

intention

Chapter 3: Utterance Interpretation

66

of compositionality4. Note that the illustration in Figure 3.1 already incorporates Levinson’s (1995) distinction of utterance type and utterance token meaning that will be introduced in more detail in section 3.2.2.

The semantics of the lexical items (word meaning) and their combination (sentence meaning) is the stable part that assures compositionality. Compositionality thus holds at the level of encoded information, at the semantic level. As lexical items are assigned an interpretation only in the context of their use5, context and other factors like background assumptions and conceptual knowledge play a crucial role in interpreting lexical items. They do this by providing further information, certain aspects of which are inferred in order to enrich the encoded information. Interpretation therefore involves a considerable amount of non-monotonic inference, characterised by defeasible default assumptions (cf. section 3.2 and chapter 5). Thus, compositionality cannot be assumed to hold at the level of information inferring, i.e. at the pragmatic level.

With respect to aspect, a semantic analysis has to be able to capture its invariant meaning, that is, its stable, context-invariant contribution to the interpretation process. One of the central claims here is that it is possible, and necessary, to assign such an invariant contribution also to the unmarked Russian ipf aspect, and presumably to unmarked elements of privative oppositions in general. This approach is different from traditional structuralist approaches which define the ipf aspect as the unmarked member of the Russian aspect opposition and assign it a negative (‘im-perfective’) value. The unmarked status of Russian ipf aspect is unquestioned in this thesis, but defining it in solely negative terms is not satisfactory. Though such a characterisation captures the unmarked status of the ipf aspect as compared to the pf aspect, it is not very helpful for a systematic derivation of ipf interpretations. In her neo-structuralist approach, Stoll (2001) is aware of this problem and tries to avoid it by combining the structuralist analysis with neo-Gricean pragmatics. A similar approach is taken here: chapter 4 sketches the semantics of the ipf aspect, and chapter 5 provides the pragmatic framework and outlines the interaction of both in the interpretation process.

From what has been said above, the task of accounting for the interpretation of aspectualised verbs can be described as follows:

3.2 Aspect6: from semantics to interpretation (I)

word meaning (aspectualised verb) → hearer’s meaning (interpretation)

As mentioned above (cf. 3.1), word meaning is composed from the meaning of the word root and derivational suffixes (if present) on the one hand, and the contribution of grammatical morphemes (inflectional suffixes) on the other hand. Linguistic expressions encode two basic types of information: concepts and procedures for manipulating them (Wilson & Sperber 1993; Blakemore 2002). Conceptual information pertains to the concepts that are activated by lexical items and constitutes the representational kind of meaning. Procedural information specifies computational processes, that is, the way the activated concepts are addressed, manipulated and put together:

4 Cf. Sag (1981) who takes semantics to consist in the combinatorial principle by means of which pragmatic

‘meanings’ are combined. 5 It is doubtful whether interpretation, even of sentences uttered out of the blue, is ever carried out in isolation

from context. The more realistic assumption is that interpretation always involves a context – if it is not explicitly given then it will unconsciously be constructed by the interpreter (cf. Ruhl 1989). This fits with Relevance Theory’s assumption of contexts being variable (Sperber & Wilson 1995: 132-142).

6 This is to be understood in the sense of ‘aspectualised verb’, i.e. as the combination of aspect and verb as encountered in actual language use.

Chapter 3: Utterance Interpretation

67

“[I]n principle, linguistic forms could encode not only the constituents of the conceptual representations that enter into inferential computations, but also information which constrains the computations in which these computations are involved. In other words, it is possible for linguistic forms to encode either conceptual information or procedural information.” (Blakemore 2002:78)

Wilson & Sperber (1993: 2) put it as follows: “An utterance can [..] be expected to encode two basic types of information: representational and computational, or conceptual and procedural – that is, information about the representations to be manipulated, and information about how to manipulate them.”

Sperber & Wilson (1995: 89) state the difference between conceptual and procedural information in terms of the difference between representation (conceptual meaning) and computation (procedural meaning). The two are formally distinct and complementary processes, neither of which can exist without the other, and both of which are necessary for comprehension to take place. The two kinds of information specification have to interact in a certain way, as procedures are specified with respect to how they manipulate the respective concepts. To make this cooperation possible, there have to be certain points of interaction, certain features specified by the conceptual information which the procedural part may work on. The specification of this interaction will also play a crucial role for the interpretation of aspectualised verbs.

Procedural information encodes constraints on the inferential phase of comprehension. Blakemore (2002) is mainly concerned with discourse connectives like so or after all, that do not encode concepts but procedures for the identification of the cognitive effect the speaker intended to achieve. They deliver information about the inferential phase of communication (Blakemore 2002: 78f), and about how the inferential phase is to be constrained:

3.3 a. Tom can open Ben’s safe.

b. He knows the combination. The sequence in ( 3.3) may be interpreted in different ways: ( 3.3b) can be understood as evidence for ( 3.3a), or it may constitute the conclusion derived from ( 3.3a). These possible inferences as to how both sentences are related, are constrained by discourse markers:

3.4 a. Tom can open Ben’s safe. So he knows the combination.

b. Tom can open Ben’s safe. After all, he knows the combination. As indicated by so, the second sentence in ( 3.4a) is derived as a conclusion from the first one, whereas after all in ( 3.4b) indicates that the second sentence is the evidence for the first one.

The notion of procedural meaning can be extended from discourse markers to, e.g., mood indicators or illocutionary particles. Wilson & Sperber (1993) also analyse pronouns in terms of procedural encoding, arguing that pronouns guide the search for the intended referent that is part of the proposition expressed. Tense markers can also be analysed as serving to encode procedures for the manipulation of the activated concepts (Carston 1999), in the sense of specifying how these concepts are to be integrated in the sequence of events. Here it will be argued that grammatical aspect equally encodes procedural information.

Wilson & Sperber (1993) distinguish four possible types of meaning that can be characterised in terms of the distinctions procedural vs. conceptual and truth-conditional vs. non truth-conditional. The distinction between procedural and conceptual encoding refers to the type of information encoded by a linguistic expression, the distinction of truth-conditional vs. non truth-conditional pertains to whether or not an expression contributes to the truth conditional content of an utterance. Adopting this scheme, grammatical aspect can be classified as specifying truth-conditional procedural information:

Chapter 3: Utterance Interpretation

68

procedural conceptual

truth-conditional pronouns, tense aspect

content words manner adverbials

non truth-conditional discourse connectives (e.g. so, after all) sentence adverbials

Table 3.1 There is another distinction that cross-cuts these former two: the distinction between explicit and implicit information referring to the communicative level a lexical item contributes to (Carston 1998). Both conceptual and procedural information may be explicitly encoded or may need to be derived via inferences (cf. ( 3.10) for an illustration with respect to aspect). As will be discussed in chapter 5, grammatical aspect constitutes an instance of underspecified procedural information. The vast majority of verbal prefixes, which are sometimes – erroneously – considered as markers of grammatical aspect, encode conceptual information, which is presumably typical of derivational morphemes in general. Prefixes help to specify the concept activated by the respective verb, mostly resulting in changes in the argument structure and, consequently, in changes of the encoded eventuality7. This specified conceptual information in turn is worked on and embedded in the context by grammatical aspect in characteristic ways. The difference between prefixes specifying modes of action (the Slavic sposoby dejstvija) and grammatical aspect can thus be explained in terms of the difference between conceptual and procedural information.

Anticipating the semantic analysis of aspect in chapter 4, the relevant notions of meaning that have to be specified with respect to aspect and its interpretation are the following: 3.5 grammatical aspect (procedural information)

word meaning: general overlap (ipf) / inclusion (pf) relation between the event time interval I(e) and the topic time interval I(TT)

utterance meaning: specific relations between I(e) and I(TT)

hearer’s meaning: contextually relevant interpretation Word meaning in terms of the general I(e)-I(TT) relation constitutes the invariant semantic basis of aspect. This general meaning may be specified in three different ways yielding three different specific relations. Hearer’s meaning pertains to the contextually relevant interpretation of these specific relations, which is achieved by integrating additional information from different sources. This will be clarified in chapter 5, as well as the status of the specific relations that constitute utterance (type) meaning

Aspect is a category of the verb and thus never appears in isolation but always applies to a verbal basis, the Aktionsart. The Aktionsart basis provides conceptual information that is manipulated, or worked on, by aspect in the way specified in ( 3.5). Aspect and Aktionsart interact, and in order to determine the place of interaction, the make-up of the lexical entries has to be specified.

7 This is one of the arguments Filip (2003, 2005) provides for regarding prefixes on Russian verbs as

eventuality-type modifiers and not as exponents of grammatical aspect (cf. chapter 2, section 2.2.3).

Chapter 3: Utterance Interpretation

69

Sperber & Wilson (1995) separate the information stored in the entry of one lexical item into three different entries, i.e., subparts: a logical, an encyclopaedic and a lexical entry8.

The logical entry consists in a set of deductive rules specifying certain input conditions and yielding a corresponding output. These deductive rules describe a set of input and output assumptions and are stable across contexts and speakers.

The lexical entry contains information about the natural-language counterpart of the concept (word or phrase used to express it), i.e. syntactic and phonological information, such as syntactic category membership, co-occurrence possibilities, selectional restrictions and phonological structure.

The encyclopaedic entry contains information about the extension and/or denotation of the concept, i.e. about the objects, events and/or properties that instantiate it. It contains stereotypical assumptions and expectations about frequently encountered objects and events, as well as assumption schemas which an appropriate context may convert into full-fledged assumptions. Contrary to the logical and lexical entries, the encyclopaedic one varies across speakers and is open-ended, which means that there is no point at which such an entry can be said to be completed.

Utterance interpretation thus proceeds as follows: “Recovery of the content of an utterance involves the ability to identify the individual words it contains, to recover the associated concepts, and to apply the deductive rules attached to their logical entries.” (Sperber & Wilson 1995: 90)

This raises the question how these different kinds of information can be brought together in one lexical entry. In the Dynamic Syntax framework (Kempson et al. 2001; Cann et al. 2004), lexical entries are assumed to consist of two distinct kinds of information: a declarative unit and instructions of how to deal with this declarative unit, i.e. how to integrate which parts of it into the overall process of interpretation. The information encoded by a lexical item is therefore both declarative and instructional. It does not only declare the information that is to be activated, but delivers also instructions how this information is activated and combined in the process of interpretation (cf. Perrett 2000: 89; Kempson et al. 2001: 55-102). The lexical entry is characterised as a “conditional action package” (Kempson et al. 2001: 59) consisting of instructions in terms of IF-THEN-ELSE rules (cf. Marten 2002: 33): 3.6 <item> IF ?Ty(X)

THEN make (…), put (…), go (…)

ELSE action The IF-part gives the conditions under which the information provided by the lexical entry can be introduced into the interpretation process. To insert a type-<e> expression, for instance, there needs to be a requirement (indicated by ‘?’) for such an expression at the stage at which the lexical entry is scanned (this is the case in ( 3.7) below). This part thus specifies the input requirement the respective item imposes onto its context. The THEN-part specifies the actions (put, make, go9) that are performed in case the condition in the IF-part is met. These actions specify the exact contribution of the respective lexical item to the building up

8 These subparts will be referred to as ‘entries’. Note that then the notion of ‘lexical entry’ is used in a

twofold way: as referring to the overall information encoded by one lexical item, and as referring to a specific subpart part of that information. Context will make it clear in which sense the notion is used.

9 The technical details are not relevant here.

Chapter 3: Utterance Interpretation

70

of the interpretation. If the input requirements are not fulfilled, the ELSE-part specifies the actions to be performed in this case – it either aborts or initiates another action10.

To give a simple example, the lexical entry for ‘John’ is specified as follows: 3.7 <John> IF ?Ty(e)

THEN put(Fo(John), Ty(e))

ELSE abort The IF-part of the lexical entry for ‘John’ specifies an input requirement, which is a requirement for a type-<e> expression somewhere in the tree structure. The IF-part is the trigger that starts the action specified in the THEN-part, here the instruction to replace (‘put’) the requirement for a type-<e> expression by the formula value (Fo) ‘John’. This formula value is of type <e> and thus fulfils and cancels the requirement ‘?Ty(e)’. This deductive process drives the interpretation process with the overall goal to fulfil all requirements – an interpreted string does not contain any open requirements. This is, of course, a rather simple example but it suffices to illustrate the basic idea: the information provided by the lexicon is declarative, as it includes the information delivered by a lexical item, and it is instructional in that it encodes instructions of how to integrate this declarative content in the interpretation process. <John> serves as a pointer to a certain concept (the ‘John’-concept) and activates the knowledge associated with it. The explicitly encoded information, i.e. the context-invariant contribution to the interpretation process, consists in the IF-THEN-structure and the instruction to activate the concept associated with the respective lexical item. The exact content of this concept, the respective encyclopaedic knowledge, may vary across speakers and thus does not pertain to the stable information. This is one reason for the flexibility of interpretation.

What is left somewhat unclear by this approach is the notion of ‘declarative information’ provided by the respective lexical item. This can be specified by Sperber & Wilson’s (1995) above mentioned distinction of lexical, encyclopaedic and logical information stored in different entries that constitute the knowledge associated with one lexical item11. These entries are summarised in ( 3.8):

10 This is only a rather short description of the Dynamic Syntax view of the parsing process. A detailed

presentation is provided in Kempson et al. (2001); Cann et al. (2004); Perrett (2000); Marten (2002). Very roughly, Dynamic Syntax can be sketched as follows: The key notions are underspecification of natural language strings and the need for conceptual enrichment in order to arrive at an interpretation for the respective strings. The primary purpose of Dynamic Syntax is to model the structure-building process of the growth of interpretation from an initial specification given by the first expression in the string to be parsed. Similar to assumptions in Relevance Theory, the structures built up in Dynamic Syntax do not represent a direct mapping from linguistic form to model-theoretic interpretation, that is, syntax and semantics are not modelled as proceeding in tandem. What is modelled is how interpretation processes, with pragmatic inferences coming into play at every level of derivation. Dynamic Syntax puts special emphasis on the process of establishing the representation of linguistic structure, the structure-building process being defined as goal-driven and incremental. The progressive update of partial structures reflects central properties of natural language processing. It captures the dynamicity and the goal-directedness of the parsing process, the goal being the establishment of a representation of the propositional content for a natural language string. This goal is achieved by progressively accumulating the necessary information across the left-to-right sequence. This assumption of tree growth captures the fact that the specification of interpretation provided by a sequence of words may be less than the interpretation assigned to that sequence. Thus, interaction with pragmatic actions may be needed to update the lexical specifications provided. The fact that lexical information is defined as contribution to the process of establishing an interpretation for the overall utterance provides for the necessary amount of flexibility without losing the invariance that is needed for assuring compositionality.

11 Sperber & Wilson (1995) assume these kinds of information to be stored under one ‘conceptual address’. This notion will not be used here, as this would misleadingly suggest that items encoding procedural information

Chapter 3: Utterance Interpretation

71

3.8 a. lexical entry: phonological and syntactic information

b. logical entry: deductive rules; input and output conditions

c. encyclopaedic entry: information about denotation and/or extension of the lexical item12 ‘Declarative information’ subsumes these three kinds of information. The logical and lexical entries correspond to the IF – THEN – ELSE structures that deductively guide the building up of an interpretation for an utterance. Lexical items that encode conceptual information do not only have a logical and lexical entry, but also an encyclopaedic one which activates the knowledge associated with the respective item. Therefore, lexical items that encode conceptual information can be regarded as pointers, as instructions to activate concepts and the encyclopaedic information associated with them. Lexical items encoding procedural information do not have such an encyclopaedic entry. As will be shown with the case of Russian aspect, procedural information does not even have to specify a lexical entry, as it is not always overtly realised13.

Combining Sperber & Wilson’s (1995) assumption of three different entries specified by one lexical item and the Dynamic Syntax specification of lexical entries yields the following conception of information provided by one lexical item:

Figure 3.2 Figure 3.2 shows the specific entries that are specified with conceptual and procedural information. A logical and a lexical entry (indicated by logicalconc./proced. and lexicalconc./proced.) is specified in case of both procedural and conceptual information. An encyclopaedic entry, however, is specified only in case of conceptual information (encyclopaedicconc.).

also activate concepts. As has been argued above, they do not activate concepts but instructions on how to deal with concepts. Even though procedural information does not point to concepts, such items have a lexical and logical entry. This will prove crucial for the interpretation of aspect.

12 Carston (2004c) takes this to be general and particular knowledge associated with the respective lexical item; another term to refer to this would be ‘intension’.

13 Remember that Russian lacks consistent overt aspect marking. Nonetheless, every verb is specified for aspect, i.e. every verb form encodes the procedural information provided for by aspect.

<LEXICAL ITEMconc./proced> declarative information

Logicalconc./proced. Lexicalconc./proced. Encyclopaedicconc.

IF trigger THEN action ELSE action

Chapter 3: Utterance Interpretation

72

Figure 3.2 also illustrates that the encyclopaedic information is one of the addresses activated by a declarative unit heading the lexical entry. It differs from lexical and logical information in a crucial way: lexical and logical information is stable across speakers and constitutes the invariant part of meaning, whereas with respect to the encyclopaedic information only the activation-part is stable. The specific content of the activated concept may vary across speakers and thus allows for flexibility in interpretation. In different contexts and/or with different speakers/hearers different parts of its content may be more prominent than others14.

The lexical and the logical entries ensure compositionality; they work together in specifying which elements of the concept activated by a word are composed in which way. Note that ‘compositionality’ here is understood in the sense of the combination of instructions the lexical items deliver to the interpretation process. The stable contributions of lexical items consists in the instructions of when (the IF-part) and in which way (the THEN-part) they are to be integrated in the overall structure, and in the instructions to activate a specific concept. The content of this concept is not part of what is encoded and thus is not subject to compositionality. Lexical items project actions, and compositionality constrains the relation between these actions15.

This conception of lexical entries is interesting with respect to lexical semantics which is concerned with the meaning of separate lexical items. One of the central issues in lexical semantics is the question of where to draw the line between the lexicon and the encyclopaedia – if any can be drawn at all. Lexical semantics concerns the question of whether to „separate information about the meaning of words from information about the (supposed) reality associated with these words“ (Blutner 1998a: 143). Adopting this distinction between lexicalised and encyclopaedic information, two different mechanisms have to be assumed: one dealing with the combinatory aspects of meaning, and a second, supposedly pragmatic, mechanism that delivers a conceptual interpretation. This amounts to equalling the distinction between lexicon and encyclopaedia with the distinction between semantics and pragmatics, a move that is taken, e.g., by Blutner (1998a,b). However, the conception of the lexicon outlined above suggests that ‘the lexicon’ and ‘the encyclopaedia’ are not to be conceived of in terms of two completely distinct domains. The relevant distinction to be drawn is one with respect to the information provided by lexical entries, namely that between logical and lexical entries specified by both conceptual and procedural information on the one hand, and encyclopaedic entries specified exclusively by conceptual information on the other. Importantly, all three entries are encoded by the lexical item. With respect to the encyclopaedic entry this might seem contradictory at first sight. However, it is only the instruction to activate the knowledge associated with the lexical item that is encoded; the exact specification of this knowledge is not encoded; in fact, it cannot be, since it varies across speakers, across contexts and across times (i.e. in the diachronic dimension).

To recap, as can be seen from the make-up of lexical entries specified in Figure 3.2, every lexical item specifies a logical entry which is relevant for its integration into sentence meaning, as well as a lexical entry that specifies its phonological form and subcategorisation information. Lexical items that express conceptual information do not only have an lexical and logical entry, but also an encyclopaedic one, which differentiates them from lexical items

14 Encyclopaedic and logical/lexical entries also differ in the kind of inferences they give rise to. The former

constitute the basis for the derivation of context-dependent, particularised, non meta-linguistic inferences, whereas inferences derived from the latter are context-independent, generalised and meta-linguistic (I-inferences vs. Q/M-inferences, cf. chapter 5).

15 This comes close to the Dynamic Syntax view of compositionality, which Kempson et al. (2001: 99) regard as a ‘tree-growth concept of compositionality’: “Because […] words project actions, the compositionality principle constrains the relation between the actions induced by strings of words and those associated with the individual elements.” (Kempson et al. 2001: 261f)

Chapter 3: Utterance Interpretation

73

expressing procedural information. Such lexical items might not even specify phonological information, as is the case with aspect in Russian, which is not always marked by a special morpheme (chapter 2, section 2.3). The aspect marker provides procedural information that has to be combined with the conceptual information provided by the Aktionsart base. The place of interaction are the respective logical entries.

Accordingly, the way from semantics to meaning in the case of aspect interpretation can be further specified as follows:

3.9 Aspect: from semantics to interpretation (II)

conceptual information (Aktionsart)

word meaning interpretation procedural information (aspect)

In cases where conceptual and procedural information work hand in hand, the conceptual information has to provide the factors that are required by the input specification of the procedural logical entry. The concepts activated by verbs can be modified in different respects. Which part of the conceptual information the modification pertains to is specified by different procedural items that may apply. The actional, causal and temporal tier of a verb, for instance, encode different parts of the denoted eventuality and may be accessed and modified by different procedural markers encoding tense, mood, aspect, person, etc. In their logical entries, such procedural items specify input requirements and possible outputs. The respective input requirements have to be delivered by the logical entries of the conceptual basis. Grammatical aspect imposes input requirements onto its context, which first of all consists in the verb it applies to. These input requirements have to be fulfilled in order to yield the output specified in the logical entry of the aspect marker. In case of mismatch between the input requirement and the actual input, this context needs to be accommodated, the required features need to be induced (cf. ( 3.16) below; the accommodation account will be elaborated in chapter 5).

To sum up so far, for the interaction of procedural and conceptual information to work, the respective information has to provide matching units. In the case of grammatical aspect the respective units consist in the phases and boundaries encoded by the Aktionsart. These units provide the input for aspect and are operated on by it in specific ways (chapter 2, section 2.2.2). Languages that do not have grammatical aspect nonetheless convey the respective information. In such cases, there are no explicit procedural constraints on the interpretation process, and the defaults provided by the overall lexical entry of the predicate constitute the basis for interpretation (‘default aspect’ as implicature, cf. Bohnemeyer & Swift 2001, 2004; cf. chapter 2, section 2.2.3).

The specification of the lexical information delivered by aspect and Aktionsart in terms of lexical, logical and encyclopaedic entry is illustrated in Figure 3.3. Note that with the logical entry, only features that are relevant for aspect are mentioned (i.e. the elements specified at the temporal tier):

Chapter 3: Utterance Interpretation

74

Figure 3.3 The Aktionsart basis delivers conceptual information and thus specifies a lexical, logical and encyclopaedic entry. The encyclopaedic entry is responsible for the construction of the specific interpretation of an aspectualised verb in a certain context, as it activates the knowledge necessary for the process of concept formation. For the purpose of this thesis, which does not deal with concept formation, the logical entry is decisive. The aspectually relevant features specified at the logical entry are the temporal tier units, i.e., phases and boundaries.

Aspect encodes procedural information and thus does not specify an encyclopaedic entry. It may specify a lexical entry, but since Russian lacks consistent aspect marking, this lexical entry is not very interesting here. The decisive part is again the logical entry, which specifies input conditions that have to be met by the logical entry of the Aktionsart basis, and an output consisting in specific relations between the run time of the event I(e) and the topic time interval I(TT). This thesis is concerned with exactly this interaction of the logical entries of aspect and Aktionsart (chapter 4) and the inferences arising from the single logical entries and their interaction (chapter 5).

However, utterance interpretation does not only involve decoding of what is explicitly encoded by word and sentence meaning, but requires also a certain amount of inferences. As will be outlined in more detail in section 3.2, the semantics/pragmatics distinction is best conceived of in performance terms, i.e. in terms of decoding (semantics) and inferring (pragmatics). Since in utterance interpretation mere decoding is not enough, the hearer has to infer information. Linguistically decoded meaning stems from the information that is explicitly delivered by the forms used, whereas inferred meaning pertains to additional information that needs to be derived and integrated by pragmatic inferences. Since the explicitly encoded contribution, i.e. word and sentence meaning, delivers only partial specification of the proposition expressed, pragmatic inferences are necessary already at the subpropositional level. Since inferences are necessary to establish the truth-conditional proposition, the distinction between truth-conditional and non-truth-conditional content cannot be equalled with the semantics/pragmatics distinction (cf. section 3.2.1).

LEXICAL ENTRY:

AKTIONSART: OPERANDUM

LOGICAL ENTRY:

ENCYCLOPAEDIC ENTRY:

/phon/, subcategorisation, ...

temporal tier: phases & boundaries

general & particular knowledge about the concept activated

LEXICAL ENTRY:

ASPECT: OPERATOR

LOGICAL ENTRY:

ENCYCLOPAEDIC ENTRY:

/phon/ (optional)

input requirements: phases (dynamic/static), boundaries

output: relation I(e) – I(TT)

--------

OPE

RA

TIO

NA

L IN

TE

RFA

CE

Chapter 3: Utterance Interpretation

75

Taking into account the decoding/inferring distinction, the interpretation of aspectualised verbs can be further specified as follows:

3.10 Aspect: from semantics to interpretation (III)

As is shown in ( 3.10), both aspect and Aktionsart encode a certain amount of information, but leave a certain amount to be inferred. The information delivered is at least partially encoded, but to a certain degree also subject to inference processes. Since decoding is a semantic process and inferring a pragmatic one, this points out the close interconnection of semantics and pragmatics in the interpretation of aspectualised verbs. It is impossible to really strictly separate semantics and pragmatics in the actual derivation of an interpretation.

For this interaction to be possible, there have to be certain matching features that provide some identical format in the representation of aspect and Aktionsart. This identical format establishes the operational interface where procedural information provided by aspect, and conceptual information provided by Aktionsart meet and interact. As has been shown in chapter 2, section 2.2.2, the identical format of aspect and Aktionsart representations consists in the phases and boundaries that are delivered by the Aktionsart basis and selected by aspect markers.

The process from the semantics of an aspectualised verb to its interpretation can thus be further specified: 3.11 Aspect: from semantics to interpretation (IV)

word meaning

conceptual information (Aktionsart)

procedural information (aspect)

inferring

decoding

decoding

inferring

interpretation

WORD MEANING

conceptual information(Aktionsart)

procedural information(aspect)

INTERPRETATION

inferring

decoding

decoding

inferring

ϕdyn, ϕstat, τ

Chapter 3: Utterance Interpretation

76

This section has shown that two different kinds of information are encoded by lexical items – conceptual and procedural information. Lexical items encoding conceptual information differ from lexical items encoding procedural information in that they specify also an encyclopaedic entry. The encoded information – conceptual and procedural – is the stable word meaning of the lexical items that combines to sentence meaning. In the interpretation process, this sentence is decoded and has to be enriched by inferred material. Lexical entries specify only part of what is needed to arrive at the final interpretation – the information delivered underspecified in different respects. Underspecification is dealt with in the following section.

3.1.2 Underspecification The discussion above has illustrated that the information encoded by lexical items hardly ever delivers all the input required for interpretation. The coded information constitutes only part of that input and needs elaboration of different kind to be assigned a contextually relevant interpretation16. Consequently, word and sentence meaning may need to be specified at the various stages of the interpretation process: “Language processing as a task of establishing interpretation involves manipulating incomplete objects at every stage except at the very end.” (Kempson et al. 2001:3)

Underspecification follows from the fact that there is no one-to-one mapping between words and concepts, as there are far more concepts than words to activate them. Analoguosly, the procedural information about how to access these concepts may as well be underspecified to a certain degree. Procedural information serves to constrain the activation of a concept in a specific way. This is the case with grammatical morphemes that encode, e.g., tense, mood, case, number – and aspect. If grammatical morphemes are assumed to constrain the concepts activated by the lexical roots in specifying certain aspects of them, then they may as well underdetermine the exact way in which they execute their constraining function. Grammatical morphemes, or more generally grammatical operators, may thus encode a set of possible procedures, where the relevant one has to be figured out by means of context or conceptual knowledge.

There exist several theories on underspecification. In this thesis, however, no theoretical commitments are made in this respect; the notion of ‘underspecification’ is used in a rather intuitive way. This notion is intended to capture certain phenomena that share the common feature of not fully specifying their semantics and their interpretation. Hence, both word meaning (and, consequently, sentence meaning) and hearer’s meaning may be underspecified. Most approaches to underspecification deal with underspecification of conceptual information and leave aside the question of procedural information. Here, grammatical aspect will be analysed in terms of ambiguous procedural encoding (chapter 5, section 5.1.2)17.

Inferring information is therefore necessary for both conceptual and procedural information to be fully specified; it is necessary to establish a full proposition. Approaches differ with respect to the level of meaning that is assumed to be underspecified: linguistic meaning (word and sentence meaning), ‘what is said’ (roughly: hearer’s meaning) and ‘what

16 In saying that most natural language expressions do not fully determine their final interpretation, I do not

want to adopt a strong contextualist view, but a more moderate, quasi-contextualist one stating that pragmatic enrichment applies only if it is needed. This allows also for the possibility of not resolving all instances of underspecification in interpretation.

17 The distinction Poesio (1996a: 184) draws in outlining two respects in which interpretations may be underspecified resembles very much the conceptual/procedural distinction: “[The] initial interpretation may be ‘underspecified’ in two senses: the meaning associated to some utterances, and/or how these utterances fit together in larger utterances, may not be unambiguously be specified by the grammar, and may have to be determined by contextual inference.”

Chapter 3: Utterance Interpretation

77

is meant’ (roughly: hearer’s meaning including the recognition of speaker’s intentions). Carston (2002: 19) lists three possible underspecification theses:

3.12 a. Linguistic meaning underspecifies what is meant.

b. What is said underspecifies what is meant.

c. Linguistic meaning underspecifies what is said. ( 3.12a) is the most general underspecification thesis. It states that word/sentence meaning underspecifies the final interpretation and thus subsumes the other two, more specific, underspecification theses. ( 3.12b) states that a proposition (‘what is said’) does not fully specify what the speaker meant to convey by his utterance. The fact that propositions are not fully determined by the lexical items used in the respective utterance is stated in thesis ( 3.12c). Underspecification thesis ( 3.12b) is what Grice is concerned with, while underspecification thesis ( 3.12c) is central to Relevance Theory (Sperber & Wilson 1995; Carston 2002). Crucially, underspecification pertains also to the level of ‘what is said’ – and not only to ‘what is meant’. This is captured by Carston’s linguistic underdeterminacy hypothesis: “The meanings encoded in the linguistic expressions used, the relatively stable meanings in a linguistic system, meanings which are widely shared across a community of users of the system, underdetermine the proposition expressed (what is said).” (Carston 2002: 19f)

Accounts differ with respect to where they draw the line between ‘what is said’ and ‘what is meant’, and with respect to which phenomena they assume to pertain to these domains. Differences pertain also to the mechanisms proposed to resolve the underspecification at different levels of the interpretation process (cf. section 3.2.1). Most approaches agree in that underspecified lexical items deliver underspecified information to the interpretation process and therefore underspecify sentence meaning as well as the proposition expressed, i.e., hearer’s meaning. The notion of underspecification thus pertains to two domains: underspecification of the semantic representation of a form in a specific context, and underspecification of the semantic representation with respect to the proposition expressed. Both domains will play a role with the interpretation of Russian ipf aspect, where the encoded information underspecifies word meaning as well as ‘what is said’ (cf. chapter 5). Based on that, additional implicatures may be drawn, constituting ‘what is meant’. These additional implicatures of the Gricean type, that is implicatures mapping propositions onto propositions, will not be dealt with in this thesis.

As has been pointed out in section 3.1.1, word meaning pertains to the semantic representation of a lexical item in isolation, i.e. to the encoded information, which is stable across contexts. This stable contribution to the interpretation process may as well be a meta-variable capturing different possible realisations that may be activated in different contexts. If these different possible realisations are discrete specifications of the overall general contribution, the respective lexical item is ambiguous – but ambiguous only outside a context. Being part of an uninstantiated sentence, an ambiguous form may yield an ambiguous logical form of the sentence. To put it with Atlas (1989: 135), “what logical form is associated with a sentence can change with the context”. The ambiguity is resolved by using the respective expression in an utterance. Utterance meaning does not involve ambiguous expressions. However, even solving the ambiguity might still underspecify the contextually relevant interpretation intended by the speaker. That is, the semantic contribution of a word as instantiated in an utterance may again be underspecified and again deliver only part of the information that is necessary for interpretation.

Chapter 3: Utterance Interpretation

78

Instances of underspecification of encoded information do not only comprise ambiguity (mainly lexical), but also indexical reference, vagueness, fuzziness and generality of natural language expressions. What all these phenomena share is a certain parsing uncertainty, which allows for their subsumption under the general heading of ‘underspecification’. The individual phenomena differ, however, in the underlying reasons for the parsing uncertainty and in their respective solution strategies. Approaches dealing with underspecification agree in that the lexical meaning gives only a weak specification of the proposition that is to be assigned to the natural language string, and they agree in that there is a range of ways in which encoded linguistic meaning may underdetermine the intended proposition. Carston (2002: 28) provides the following taxonomy:

3.13 a. multiple encodings (i.e. ambiguities)

b. indexical references

c. missing constituents

d. unspecified scope of elements

e. underspecificity or weakness of encoded conceptual content

f. overspecificity or narrowness of encoded conceptual content

Note that Carston includes also ambiguity among underspecification, which is not always the case. Some approaches to underspecification exclude ambiguity and treat it as a separate phenomenon, resulting in an opposition of ambiguity and underspecification. Here, ‘underspecification’ is conceived of as referring to the general phenomena of insufficient specification of the interpretation to be assigned to the respective form and/or utterance. Consequently, ambiguity is regarded as a special instance of underspecification.

Zhang (1998) distinguishes fuzziness, vagueness, generality and ambiguity. All these four concepts are characterised by conveying imprecise, un- or underspecified information, but differ in which respects they are underspecified. Thus, a fuzzy expression is referentially opaque in that it has no clear-cut referential boundaries. This is what distinguishes it from the other three concepts whose underspecification has not so much to do with reference, but rather with meaning and interpretation; a general expression leaves certain details of its meaning underspecified, whereas a vague expression can be assigned more than one possible interpretations, which are semantically related. Ambiguity is opposed to the other forms of unspecified information. Ambiguous expressions are characterised by having more than one semantically unrelated meaning, each of them having a paraphrase that is not the paraphrase of one of the other meanings. These different meanings can be stated in form of a disjunction.

Importantly, expressions can be ambiguous, vague, general or fuzzy, depending on the perspective one takes: “One important point [...] is that we tend to examine an expression in a uni-dimensional manner, i.e. as either ambiguous or vague. In fact, an expression can be ambiguous, vague, general or fuzzy, depending on how we look at it.” (Zhang 1998: 29)

Therefore, a multi-dimensional approach to examining an expression with respect to these four categories is more promising than a uni-dimensional one, which takes an expression to be either ambiguous or vague or general or fuzzy. As for ipf aspect in Russian, it will be claimed (chapter 5) that at one level of interpretation, it is best captured in terms of a meta-variable ‘IPF’ that is ambiguous between certain values. These values are again sense-general, that is neutral with respect to the interpretation that may be constructed from them.

Ambiguity differs from the others kinds of underspecification in that ambiguous expressions give rise to a clearly defined number of discrete specifications. Zhang (1998)

Chapter 3: Utterance Interpretation

79

refers to ambiguity in terms of ‘overdetermination’ as opposed to the other cases of underspecification, that are ‘underdetermined’. This distinction is due to the fact that with ambiguous expressions the hearer has the choice between several clearly distinct interpretations, whereas with the other cases of underspecification he has to enrich the information provided by the respective expression to arrive at a certain interpretation. Therefore, ambiguity can be said to provide more information than necessary, whereas the other instances of underspecification do not provide enough information. Here, these other instances of underspecification will be subsumed by ‘sense-generality’ (Atlas 1989).

In case of sense-generality, as Atlas (1989: 130) emphasises, “there is nothing in the words themselves, independently of any collateral information in the context of utterance, by virtue of which an ordinary speaker [...] could claim that the sentence HAD to mean one or the other”.

Outside a context, an ordinary speaker would not say, according to Atlas (1989: 130), that the (sense-general) utterance does have to mean one or the other. This is because the utterance is not ambigous between different interpretations (i.e., two or more clearly distinguishable meanings standing in complementary distribution), but is neutral with respect to its possible interpretations. The crucial point in Atlas’ statement above is the ‘or’, which indicates that with sense-generality the possible interpretations cannot be captured by stating the meaning of the utterance in terms of a disjunction. This holds for instances ( 3.13b-f) above, but not for truely ambiguous cases, where a disjunctive analysis is possible18.

Thus, the main opposition within the different kinds of underspecification is that between ambiguity and sense-generality. This distinction is emphasised especially by Atlas (1989; 1997), but also in Dynamic Syntax (Kempson et al. 2001: 25f), where the content of an expression is defined as the contribution it makes to building up an interpretation. Some expressions may make two or more contributions to the building process and are therefore ambiguous. Other expressions may make one definite contribution, which is definite, but not fully determined, and therefore sense-general. These two kinds of underspecification share the superficial phenomena of parsing uncertainty, but the underlying phenomena and the respective solution strategies are different. An expression that makes two or more different contributions to the interpretation process has to be disambiguated by selecting the right specification, whereas an expression without fully determined contribution has to be enriched and specified in order to construct an interpretation for it.

Thus, in case of ambiguity, one or more of the possible understandings are eliminated by the context, whereas sense-generality is not subject to elimination of certain readings. Instead of being selected, these readings need to be constructed: “In interpreting the utterances of sense-general sentences, we are not selecting from the readings of an ambiguous sentence; we are CONSTRUCTING from a definite but general sense and from collateral information a specific content [emphasis in the original, B.S.].” (Atlas 1989: 31)

Atlas (1989: 26) exemplifies sense-generality with the word tokens the girl with the flowers, that may receive different specifications in different contexts: ‘the girl wearing flowers’, ‘the girl selling flowers’, ‘the girl carrying flowers’, ‘the girl strewing flowers’, and the like. These, and probably more, relations between the girl and the flowers are subsumed by with, but this does not mean that with is n-way ambiguous between exactly these relations. Rather, with determines a certain relation between the girl and the flowers, but the exact specification

18 Compare, however, Poesio (1996a,b) who argues against a disjunctive account of ambiguity, since,

according to him, this does not adequately reflect the way ambiguous utterances are processed. Ambiguous sentences are not necessarily perceived as ambiguous and therefore the meaning of an ambigous sentence (or lexical item) cannot be stated in terms of a disjunction. However, with respect to the Russian ipf aspect, it will prove useful to specify its general semantics in terms of a disjunction even though this might not reflect the actual processing.

Chapter 3: Utterance Interpretation

80

of this relation is constructed according to the respective context. The same holds for relations of the kind ‘X’s Y’ (Levinson 1995), where the genitive specifies a relation holding between X and Y, the exact nature of which is constructed at the specific occasion of use in a certain context. The meaning of sentences containing such expressions is neither the one nor the other interpretation. It is definite, but neutral between the different interpretations (Atlas 1989: 30). Sense-general expressions do not provide all the information needed for interpretation, but they constrain the range of possible interpretations and provide frame and base for further inferences (Atlas 1997: 223). The different specifications of sense-general meanings are not different word meanings of the respective item: “These are contextually relevant ‘specializations’ of the generalized, literal sense of the words, distinct and relevant fillings-in, so to speak, of the sense of the words, but not themselves distinct senses of those very words [emphasis in the original, B.S.].” (Atlas 1989: 26)

This is important in dealing with the different readings of aspectualised verbs. In chapter 5 it will be argued that these are not different word meanings that could be listed a priori, but specifications of one definite semantic configuration of aspect and Aktionsart. Hence, it is not necessary to specify all the interpretations that are possible with the pf or the ipf aspect.

In order to determine whether a given lexical item is ambiguous or sense-general, two standard tests are commonly applied: zeugma and impossibility of crossed readings. It is assumed that if one lexical item can be assigned multiple lexical entries then it displays these properties. However, ambiguity does not necessarily have to be licensed by two separate underlying lexical entries that happen to have the same surface realisation, but may as well be licensed by pragmatic principles. Thus, having two separate lexical entries is not the crucial point that sets ambiguity apart from the other kinds of underspecification. This has been argued for by Lascarides et al. (1996), who emphasise that ambiguity is also possible with single lexical entries that reveal their ambiguity only with regress to pragmatic reasoning (see below)19. The respective effects – zeugma and failure of crossed readings – then arise due to pragmatic inferences yielding a certain default interpretation, and due to inferences about discourse coherence. Thus, even though there is one single lexical entry for an item there may emerge an irresolvable conflict in the pragmatic component which causes the respective item to be pragmatically ambiguous.

Lascarides et al. (1996: 4) illustrate this claim with the following examples:

3.14 a. Nylon bristles and plastic handles are used to make cheap brushes.

b. # Rembrandt used a brush and so did our janitor.

c. # Rembrandt and our janitor used a brush.

In ( 3.14a), brushes is used in a rather general sense and if one wants to postulate separate lexicon entries for all its different uses, then there also has to be one such general entry that is neutral with respect to these readings. The assumption of one lexical entry conflicts with the zeugma effect in ( 3.14b) – ‘paint brush’ and ‘cleaning brush’ – and the impossibility of crossed readings in ( 3.14c). Both effects indicate ambiguity. Lascarides et al. provide an explanation in terms of the way pragmatics affects interpretation, namely by means of different default interpretations. This is the case with the conjuncts in ( 3.14b) and ( 3.14c), that each give rise to different default interpretations, as indicated in ( 3.15):

3.15 a. Rembrandt used a brush.

b. The janitor used a brush.

19 Russian ipf aspect can also be analysed as an instance of pragmatic ambiguity, cf. chapter 5.

Chapter 3: Utterance Interpretation

81

By default, that is, in the absence of information to the contrary, brush in ( 3.15a) will be interpreted as ‘paint brush’, whereas brush in ( 3.15b) will be interpreted as ‘cleaning brush’. In ( 3.14b), these two default interpretations conflict, and this irresolvable pragmatic conflict causes the zeugma effect and the impossibility of crossed readings in ( 3.14c) – brush turns out to be pragmatically ambiguous.

A further reason for the impossibility of crossed readings are constraints on discourse coherence imposed by the way the respective phrases are linked together. Interpretations yielding only weak coherence are dispreferred compared to interpretations yielding strong coherence. In case of ( 3.14b) the respective relation is ‘parallel’20, which requires a partial structural isomorphism between the constituents of the conjuncts and a common theme. In the case of ( 3.14b) this common theme would have to be something like ‘people doing something with a brush’. This would be a rather bad theme, however, as there is nothing to provide an explanation why the speaker intends to talk about people using brushes for rather different activities (Lascarides et al. 1996: 13)21. The crossed reading is therefore rejected, and the hearer is left with the problem which default inference is to be preferred and which one is to be dropped.

To sum up so far, the decisive notions in the preceding discussion have been ambiguity and sense-generality. Both constitute the two types of underspecification that are relevant to an analysis of aspect. Ambiguity is characterised by giving rise to clearly distinct meanings that may be listed as a disjunction. From this list the contextually relevant interpretation has to be selected by the hearer. Importantly, ambiguity is not merely a semantic phenomenon but also a pragmatic one: it may arise due conflicting default interpretations, which cannot be part of semantics, and it is resolved by pragmatic reasoning (cf. chapter 5, section 5.1.2). There are cases where there is one lexical item with a single lexical entry specifying one semantic contribution to the interpretation process, which nonetheless exhibits behaviour typical of ambiguous expressions. This can be explained by assuming the ambiguity to be caused by conflicting default interpretations and by discourse relations yielding only weak coherence. This kind of ambiguity is important with respect to the characterisation of the ipf aspect, specifically when it comes to assigning it a positive semantics. Sense-generality differs from ambiguity in that it does not give rise to clearly distinct meanings, but is characterised by one definite meaning which crucially needs further enrichment and specification to arrive at a contextually relevant interpretation. This interpretation is not selected, but – based on that definite but general encoded information – constructed in a given context by inferring additional information from different knowledge sources. Vagueness will be subsumed under the notion of generality; fuzziness, as referential underspecification, is outside the scope of this thesis.

The concept of underspecification accounts for the properties that make natural language a special means of communication: efficiency, flexibility, regularity and compositionality. Underspecification guarantees efficiency in that there is no need to really encode every aspect of what is to be conveyed. Flexibility is achieved by the possibility of enriching the underspecified lexical entries within the boundaries provided by the explicitly encoded information (procedural and conceptual). Regularity is due to the fact that the encoded meaning components of the lexical entries, that is, the lexical and logical information, remain invariant across speakers and contexts and thus can be taken for granted by the interlocutors in a communication setting. It guarantees compositionality in that the semantic specification encodes the way concepts are accessed and combined. This again relates to effeciency, as the information stored in the lexical entry is the information to access a certain concept, whereas

20 For such discourse relations compare, e.g., Lascarides & Asher (1993). 21 Two comments are in order here: First, this could, of course, give rise to an implicature based on the

violation of the maxim of relevance (which is beyond the scope of this thesis). Second, further context or discourse might provide an explanation for why this common theme should be relevant.

Chapter 3: Utterance Interpretation

82

the exact content of the concept does not need to be encoded. Flexibility is best captured within a pragmatic framework, as it is in most cases connected to nonmonotonic reasoning and default interpretations. Pragmatics enriches semantics with the flexibility needed, without giving up regularity and compositionality.

All types of underspecification mentioned so far leave the final interpretation undetermined to a certain degree. In all these instances, lexical meaning underspecifies ‘what is said’, i.e. what is encoded in the lexical items underspecifies the proposition expressed. There is another kind of underspecification that pertains to ‘what is meant’, capturing the fact that a hearer has to reason about the speaker’s intentions in uttering an expression (cf. ( 3.12b)). Interpretation involves the recognition of these intentions, and the question is how this is possible. The main reason is that – just as with the other cases of underspecification – speaker and hearer both have access to assumptions on how communication normally proceeds, based on experience from observed communicative behaviour. These assumptions are accessible for all participants in the communication process. They lead to certain generalisations about alledged principles of conversation. This gives rise to specific interpretational effects on the side of the hearer if he assumes one of these principles to have been violated. The speaker, on the other side, might very well be aware of that effect and therefore may deliberately make use of such violations in order to safe the effort of making the intention behind them explicit.

This is the realm of Gricean implicatures, which are primarily about how to reconstruct the speaker’s intentions/speaker’s meaning, and not so much about how the interpretation is carried out by the audience (Saul 2002). Grice was primarly interested in implicatures that arise from violation of the maxims. However, there are also implicatures that arise from obeying the maxims. Since all participants in the communicative setting unconcsiously know about the maxims (per definitionem), they know what it means to obey them and therefore can draw inferences from that. Harnish (1979/1998: 27522) points out the need to distinguish between these two kinds of implicatures: “Indirect [conversational] implicatures require that a maxim be flouted, whereas direct implicatures require that the highest-valued maxims are intended to be observed, at the level of what is said. Non-conventional, but non-conversational implicatures might be called ‘contextual’ [emphasis in the original, B.S.]”.

Direct implicatures that arise from obeying the maxims will play a crucial role in accounting for the interpretation of Russian ipf aspect, whereas indirect implicatures arising from violating the maxims will play a minor role. The latter need to be made explicit by the speaker in order for the hearer to recognise the intended violation of the maxims, whereas direct implicatures arise by the assumption of the maxims being obeyed. Since this is the default assumption, this kind of inference goes through per default.

The readiness of the hearer to infer additional information in order to enrich the underspecified encoded information can be accounted for by the assumption of cooperativity. Presuming some form of cooperativity on the part of speaker and hearer23, any utterance will

22 Page numbers refer to the reprint in Kasher (1998). 23 The notion of cooperativity is sometimes denied to play a role in interpretation. Compare, e.g., Parikh

(2000), who replaces cooperativity by the following assumptions: 1) A intends to convey p. 2) A utters φ. 3) B intends to interpret φ. 4) B receives and interprets φ.

These assumptions amount to simply stating that if agents act in the right way, communication can occur, if they do not, it cannot. According to Parikh, there is nothing, which forces them to cooperate. Parikh bases these assumptions on the rationality of A and B, which seems, however, a prerequisite as strong as cooperativity.

In this thesis, cooperativity is assumed to constitute the basic fundaments of communication, the prerequisite for the interpretation process to start. Why should one engage in interpreting a natural language string, if it was

Chapter 3: Utterance Interpretation

83

get interpreted – at least, the hearer aims at assigning an interpretation to a perceived utterance. It is the presumption of cooperativity that drives the interpretation process in a goal-directed way. Different utterances perceived on the basis of different knowledge resources may require different degrees of effort to be assigned an interpretation. The most important factor directing the interpretation is the search for a context that helps to make the assumption of a certain interpretation reasonable. This context does not necessarily have to be explicitly given, it may as well be created by different reasoning processes. Relevance Theory assumes that the search for the right context is driven by the goal to maximise relevance, more precisely, to maximise “the cumulative relevance of the inputs” that are processed (Sperber & Wilson 1995: 261). The hearer presumes the speaker to behave cooperatively, and therefore to produce some relevant string of expressions. This justifies the search for, and the selection of a context that allows this expectation to be fulfilled: „In verbal comprehension in particular, it is relevance which is treated as given, and context which is treated as variable“. (Sperber & Wilson 1995: 142)

Phenomena like recategorisation or coercion follow from the variability of context. Any utterance can be interpreted and judged acceptable, as long as a plausible context can be accommodated for it. This is interesting also with respect to the interaction of aspect and Aktionsart. The primary and most immediate context for aspect is the Aktionsart basis it applies to. As has been pointed out above, aspect and Aktionsart interact in their logical entries: the Aktionsart basis provides certain elements aspect markers are sensitive to. Aspect has been defined as the stable part of that combination, the part that imposes certain input requirements on its context, i.e. the Aktionsart basis. If these requirements are not fulfilled they have to be induced in order for aspect to apply.

Thus, the presumption of cooperativity accounts for the possibility of assigning an interpretation also to utterances that do not meet the standard requirements to be interpretable. This is the case, for instance, with utterances lacking the required presuppositions. These utterances can be interpreted if the necessary presuppositions are added to the context.

This strategy is captured by Lewis’ notion of accommodation, which is subject to the following rules: “[C]onversational score does tend to evolve in such a way as is required in order to make whatever occurs count as correct play.” (Lewis 1979/2002: 16824)

This statement can be taken as figurative way to express cooperativity. Cooperativity allows for contexts to be modified in order to make the conversation a correct play: “If at time t something is said that requires component sn of conversational score to have a value in the range r if what is said is to be true, or otherwise acceptable; and if sn does not have a value in the range r just before t; and if such-and-such further conditions hold, then at t the score-component sn takes some value in the range of r.” (Lewis 1979/2002: 168)

This can be applied to the issue of aspect interpretation as follows: sn are the required elements that have to be present in the logical entry of the Aktionsart, i.e. in the range r. If the required elements are not present, that is, in a case of mismatch between aspect and Aktionsart, they can be induced by changing the context with ‘such-and-such further conditions’, i.e. by means of adding information provided by further context, background assumptions or conceptual knowledge. Note that Lewis states this rule of accommodation as tendency.

Accommodation can be regarded as a kind of repair strategy, which repairs the context of an utterance such that it can be interpreted (van der Sandt 1992: 340f). In the case of aspect, the immediate context is the Aktionsart base it applies to, followed by the material included in

not presumed that the speaker behaves cooperatively? To find out that some utterance was not cooperative, it is still necessary to (try to) interpret it first.

24 Page numbers refer to the reprint in Portner & Partee (2002).

Chapter 3: Utterance Interpretation

84

the VP (objects, adjuncts, subjects). The relevant presuppositions are phases and boundaries, the elements aspect is sensitive to and operates on. If these elements are not explicitly provided for, the default option is to add them, i.e. to infer them, in order to make the utterance interpretable. Accommodation is an additional operation carried out early during the interpretation process, i.e., before the final interpretation is established: “[Accommodation] is thus best seen as a kind of pre-processing of an utterance in order to adjust contextual parameters so as to create an auxiliary content in which it can be interpreted. It is the context thus established which enters in the computation of the content, implicatures, and other pragmatic information. Contextual incrementation will then map the adjusted context into the next one.” (van der Sandt 1992: 341)

Here, van der Sandt indicates two important points. First, accommodation is not to be equated with implicatures, neither is it a process that makes the conception of implicatures superfluous. Rather, both processes work on different levels of interpretation and differ in their influence on the following discourse. Accommodation and contextual incrementation by means of implicatures are two different processes. Second, accommodation is a kind of pre-processing, that is, it applies before the utterance is assigned its final interpretation. With respect to the interpretation of aspectualised verbs, accommodation is carried out before aspect finally applies. In the preceding chapters it has repeatedly been pointed out that aspect itself does not change the basic eventuality. It requires a certain input and thus the basic eventuality may need to be adjusted to fulfil the input requirements. It is this adjusted basis, aspect applies to.

The process from semantics to interpretation for aspectualised verbs can now be further specified with respect to the Aktionsart basis: 3.16 Aspect: from semantics to interpretation (V)

The inferences that may be necessary with respect to the logical entry provided by the Aktionsart pertain to the provision of the required input. In case the input requirements for aspect to apply are not met, the necessary temporal tier unit (phase or boundary) has to be accommodated. The specification of the inferences that might be necessary with respect to the logical entry provided by aspect are a language-specific matter depending on the information encoded by the respective aspect marker. The inferences that may be necessary for the Russian ipf aspect will be dealt with in chapter 4, section 4.2.1.

WORD MEANING

conceptual information (Aktionsart)

procedural information (aspect)

INTERPRETATION

inferring: adjustment of required input (ϕdyn, ϕstat, τ)

decoding

decoding

inferring

ϕdyn, ϕstat, τ

Chapter 3: Utterance Interpretation

85

One of the problems in dealing with the interpretation of underspecified linguistic expressions is how to delimit alleged core interpretations from alleged peripheral interpretations. With respect to aspect this distinction has been referred to as, e.g., direct vs. augmented interpretations (C. Smith 1997), or primary (pervičnye) vs. secondary (vtoričnye) interpretations (Padučeva 1996). Distinctions like these are problematic insofar, as it is not always entirely clear, whether they are intended to refer to ‘meaning’ in the sense of word/sentence meaning pertaining semantics, or to ‘meaning’ in the sense of hearer’s meaning pertaining to interpretation. Furthermore, such a distinction is difficult to maintain, because different contexts may make different interpretations more or less ‘core’ or ‘peripheral’25. This will become clearer in analysing aspect and its interpretations in chapters 4 and 5, and is shortly illustrated by the following pair of examples:

3.17 a. Moj otec govoril po-turecki.

my father speak:PST:ipf Turkish ‘My father could speak Turkish.’

b. Moj otec govoril po-turecki, kogda ja vošla v komnatu. my father speak:PST:ipf Turkish, when I come.in:PST:ipf into room:ACC

‘My father was speaking Turkish, when I came into the room.’ In ( 3.17a), the interpretation requiring the least contextual assumptions is the potential one (could speak) which is backed by encyclopaedic knowledge that makes this interpretation the most probable one. This is different with ( 3.17b), where a change in context makes the actual-processual reading (was speaking) the most probable one. This illustrates that ‘core meanings’ are context-dependent. They are therefore best taken as default interpretations, which also accounts for the fact that they might be overridden by the context. A distinction between core/primary/direct and peripheral/secondary/augmented meanings or interpretations is thus not helpful. The semantics of the form cannot be conceived of as core or peripheral either, since the semantics is assumed to remain stable across contexts. What may vary is the context, that is first of all the Aktionsart basis. Therefore, a distinction of core and peripheral cases is more reasonably taken to pertain to matches of aspect and Aktionsart (core) and mismatches (peripheral) and not to context-specific interpretations. Mismatches require rearrangement/accommodation of the context and thus more additional assumptions and more interpretational effort. Once this arrangement has taken place, the contribution of the aspect operator is the same as it would be with a fitting context (chapter 5).

To recap, aspect operators impose specific input requirements onto the context provided by the logical entry of the Aktionsart. In case of a mismatch between operator and basis, the missing units have to be inferred and in this case one may reasonably speak of ‘peripheral’ cases. The mechanism behind this kind of inferences has been captured by the notion of accommodation, which is closely related, but not equal, to that of implicature (Thomason 1990). In chapter 5, section 5.2.2 it will be shown that both strategies play a role in the interpretation of Russian ipf forms. This is contra Levinson (2000: 60-63) who claims that accommodation is superfluous and can be replaced by GCI theory.

25 The question of how to distinguish core and peripheral cases is closely connected to the question of

grammaticality and acceptability of an utterance. Not all acceptable sentences are judged grammatical by linguists, but the problem is that such allegedly ungrammatical sentences may nonetheless be used and be accepted, that is, interpreted. Dynamic Syntax (Kempson et al. 2001) therefore makes use of a more liberal notion of grammaticality and refers to it in terms of ‘well-formedness’. All well-formed sentences have to reach the goal of establishing a type-<t> expression and to do so, all sub-goals have to be achieved by following the actions provided by the words, or their lexical entries, respectively, in the linear sequence in which they are presented. Well-formedness is given if some set of actions licensed by the system makes available at least one pairing of a string with an interpretation.

Chapter 3: Utterance Interpretation

86

The underspecification of linguistic expressions is best captured by a dynamic conception of the interpretation process, which is driven by the goal to establish a truth-conditional proposition. One framework with special emphasis on both properties is Dynamic Syntax (Kempson et al. 2001; Cann et al. 2004), which conceives the interpretation process as dynamic and goal-driven. The semantic content of a linguistic expression is defined as the contribution it makes to building up an interpretation. This contribution may be underspecified in different respects: one expression may deliver two or more clearly distinct contributions to the interpretation process or it may deliver only one clearly defined contribution that is not fully determined. As has been argued for above, these two instances of underspecified contribution characterise ambiguity and sense-generality. Underspecification can be captured by means of meta-variables that have to be replaced in the course of the interpretation process. Such meta-variables constrain the way they may be substituted in characteristic ways. Expressions that are highly context-dependent, e.g., pronominals, are assigned a definite word meaning which is explicitly marked as being underspecified and represented as meta-variable.

In Dynamic Syntax, the requirement for a type-<t> expression drives the interpretation process. In the course of this process, pragmatic considerations leading to the choice of interpretation can be made at any step in the derivation. Such pragmatic actions include, for example, the question of how to incorporate which kind of information: “Linguistic input projects instructions for building up logical forms, and [...] these instructions, in combination with information independently established in context, yield some completed propositional structure. Any one such logical form is then combined with some larger set of premises, the context, to derive context-specific inferences. [...] On this view, the computational/lexical system of a natural language systematically underdetermines the set of possible interpretations for any senence, and it is only interaction between pragmatic and system-internal actions which yields a context-particular representation of content as articulated in some logical form [emphasis in the original, B.S.].” (Kempson et al. 2001: 16)

Information independently established is selected and integrated by pragmatic actions that take into account information that is not part of the string currently being processed (cf. section 3.2.1). Pragmatic actions incorporate the context dependency of the interpretation process. They are needed, among other things, to solve the underspecification introduced by meta-variables (cf. Perrett 2000: 102). The notion of meta-variable will prove useful also to analyse aspect. Aspect is conceived of as contributing an instruction to the interpretation process in terms of which part(s) of the situation being talked about is the crucial one and how this part is related to the time an assertion is made about (chapter 4, section 4.1.1). This instuction is left unspecified in certain respects and is therefore captured in terms of a meta-variable that clearly defines this contribution while at the same time leaving it underspecified. Pragmatic actions do not apply only after the representation of the utterance has been built up, but may apply at every step towards fixing the sentence meaning26. Languages differ in which contributions they encode, and whether the encoded information is encoded unambiguously or in terms of meta-variables. Languages thus differ in the respective contributions of semantics and pragmatics to the interpretation process. This will be shown to be the case with grammatical aspect as well (chapters 4 and 5).

Dynamic Syntax is not concerned with the articulation of criteria that enable the parser to make the choices necessary for deriving the contextually-appropriate interpretation of a given string and the articulation of the further pragmatic effects such a choice gives rise to. This latter task will be central in sketching the interpretation of underspecified aspectual forms in Russian.

To conclude, understanding an utterance does not only involve decoding of what is explicitly given, but also dealing with what is left underspecified. This latter task requires selection of meanings from several possibilities delivered by an ambiguous string as well as

26 Marten (2002) is very explicit about the contribution of pragmatics at any step of interpretation.

Chapter 3: Utterance Interpretation

87

constructing an interpretation for a univocal, sense-general string. It involves ‘reading out’ as well as ‘reading in’ meanings (Atlas 1989: 146). One lexical item can be subject to both strategies, depending on what level of the interpretation process it is being looked at. Both strategies require the inferring of additional information, that is, information which is not encoded in the strings currently being processed.

3.2 Interpreting underspecified utterances

As has been pointed out, underspecification of natural language expressions is neither an exceptional phenomenon, nor a sign of imperfection of language. Although a semantically underspecified sentence meaning does not present all the information contained in the speaker’s meaning, it does provide for cues constraining the possible interpretations or the basis for further inference (Atlas 1997: 223).

It has also been pointed out that underspecified expressions nonetheless deliver, and crucially have to deliver, a fixed contribution to the interpretation process. This fixed contribution ensures the compositionality of semantics. In case of ambiguity, one expression delivers more than one contribution, but these are clearly distinguishable and separate specifications to choose from. With sense-general items, the contribution is also definite but rather general and thus requires enrichments of different kinds, enrichments that are constrained by the instructions encoded by the respective item.

The question that has to be addressed in this section is how the gap between explicitly encoded information and contextually relevant interpretation can be bridged. It will be argued that this is possible by taking into account pragmatic principles that interact with the semantic basis in a systematic and predictable way. Section 3.2.1 addresses the general question of how to delimit semantics and pragmatics and how to account for and specify their interaction. The pragmatic framework and the basic pragmatic principles are outlined in section 3.2.2.

3.2.1 Semantics and pragmatics This section argues for the incorporation of pragmatic mechanisms in order to account for the interpretation of underspecified forms. There exist two rather influential frameworks in pragmatics that are sometimes conceived of as competing theories: Relevance Theory and (neo-)Gricean pragmatics. As Saul (2002) points out, however, both theories in fact aim at explaining different phenomena: whereas Grice’s theory is mainly concerned with the speaker’s side27, Relevance Theory takes the hearer’s perspective. Criticism of one of these theories in most cases stems from not taking into account these rather different goals. The analysis of the pragmatic contribution to the interpretation of Russian ipf aspect in chapter 5 shows that both theories apply with different components involved in the interpretation process. They complement rather than compete with each other. Neo-Gricean theory, more specifically the framework proposed by Levinson (2000a), applies to the logical entries of aspect and Aktionsart, whereas relevance theoretic principles apply to the integration of encyclopaedic knowledge. Since this latter part will not be dealt with in this thesis, Levinson’s framework will be in the focus of attention.

Bridging the gap between linguistic meaning and what is said by referring to some notion of pragmatics brings about two main difficulties: First, the notion of pragmatics is not always clearly defined and moreover tends to be used in some ‘waste-basket’, last resort sense. This is obvious in statements of the sort ‘how this works, is left to pragmatics’, making pragmatics

27 This has been changed in further developments of Grice’s theory which incorporate also the hearer’s

perspective, cf. Levinson 2000a.

Chapter 3: Utterance Interpretation

88

resemble a deus ex machina that rescues linguists from being left without explanations. Second, the delimitation of semantics and pragmatics and the specification of their interaction is problematic. Semantics and pragmatics can only be delimited if there are clearly defined notions of both, and only then their interaction in utterance interpretation can be specified. Somewhat misleading in this respect is the assumption of an interface, which seems to imply a clear-cut demarcation line. As will be shown in this section and in chapter 5, this interface is not as clear-cut as the term suggests. In actual interpretation, semantic and pragmatic contributions may be rather hard to tease apart.

According to what has been said above, the semantics of lexical items consists in the information encoded in their lexical entries: instructions to access a certain domain of knowledge associated with one lexical item, as well as instructions concerning how and where to integrate that item into a sentences. The lexicon thus combines logical, lexical and – in case of conceptual information – encyclopaedic information. The exact content of the encyclopaedic knowledge activated by a lexical item can hardly be specified unambiguously and invariantly, but the instruction to activate it pertains to the information encoded in the lexicon. The information encoded by a lexical item constitutes the stable, invariant parts of ‘meaning’ which pertain to the knowledge of a language and are common to all its speakers. The fact that the information encoded in these lexical entries contains instructions to activate concepts in a specific way, and the fact that these instructions may be underspecified to a certain extent, explains why interpretation is more than just combining the different parts and pieces. It also explains the fact that the content intended by the speaker may very well differ from the one constructed by the hearer. The concepts activated by certain linguistic items may differ across speakers and intersect only partially. Moreover, lexical items do not even have to encode a full-fledged concept, as Sperber & Wilson (1998: 200) emphasise: “[I]t does not matter whether or not a word linguistically encodes a full-fledged concept, and, if so, whether it encodes the same concept for both speaker and hearer. Even if it does, comprehension is not guaranteed. Even if it does not, comprehension need not be impaired. Whether they encode concepts or pro-concepts, words are used as pointers to contextually intended senses.”

Via the encoded instruction to activate a certain concept, conceptual or encyclopaedic knowledge is connected to lexical semantics and therefore may be taken as part of the lexicon in a wider sense. Whether it is part of the lexicon in a wider sense or whether it is assumed to constitute a separate knowledge domain – according to the conception advocated here, it does not pertain to the realm of pragmatics. Pragmatics here is conceived of as the processes that guide the search for additional information as well as its combination with the semantically encoded information, whereas encyclopaedic knowledge supplies that additional information. This conception of pragmatics is different from conceptions like Dölling’s (2004) who takes pragmatics to be conceptual knowledge, or Blutner’s (1998) who equals the distinction between lexicon and encyclopaedia with that between semantics and pragmatics.

There are further distinctions that are sometimes taken to correspond to, or even to constitute the semantics/pragmatics distinction. The ‘classic’ conceptions of the semantics/pragmatics distinction would assume it to coincide with one of following distinctions (Bach 1999):

3.18 a. linguistic (conventional) meaning vs. use

b. truth-conditional vs. non-truth-conditional meaning

c. context independence vs. context dependence As has been argued above (cf. Table 3.1), distinction ( 3.18a) is not tenable since there are expressions whose semantics can only be specified by referring to their use. This is the case with discourse markers like so, after all or anyway, as discussed by Blakemore (2002). The same holds for personal pronouns like I, where the semantics specifies the speaker as the

Chapter 3: Utterance Interpretation

89

referent, but the specific referent is determined only in use28. There are also expressions whose semantics cannot be stated in truth-conditional terms, which is contra ( 3.18b). Speech act expressions are one example, as well as indexicals and demonstratives. The third distinction, ( 3.18c), cannot be taken to constitute the semantics/pragmatics distinction either, as context plays a role with both semantics and pragmatics. Bach (1999) distinguishes two sorts of contextual information: variables such as the identity of the participants in the communicative setting and time and place constitute context in a narrow sense, whereas anything the hearer has to take into account to come as close as possible to the interpretation intended by the speaker constitutes context in the broad sense. He conceives only of this latter kind of context as pragmatic. This assumption, however, is based on Bach’s conception of the distinction between ‘what is said’ and ‘what is meant’. There are different suggestions concerning this distinction, as will be illustrated below (cf. Table 3.2).

The picture is further complicated by the fact that many sentences are syntactically complete, but not semantically. To arrive at a proposition, pragmatic processes like completion, expansion or specification are needed, and these processes are highly dependent on context. This is a further argument against an exclusively truth-conditional conception of semantics and a conception of pragmatics as being somehow ‘above’ the semantic level. It also suggests that it might be reasonable to speak of two kinds of semantics and two kinds of pragmatics (cf. Figure 3.4). Pragmatic processes that are necessary to enrich the semantic representation of a sentence to yield a truth-conditional proposition are, among others, captured by the notions of ‘impliciture’ (a term coined by Bach) or the relevance-theoretic ‘explicature’ (cf. Table 3.3).

What is needed is a formulation of the semantics/pragmatics distinction that takes into account the distinctions mentioned above, ( 3.18), but does not rely too strongly on them. Such a formulation has to bring together the facts that only explicitly encoded information is semantically valid, and that there are expressions whose semantic contribution indicates their context-dependency. Such expressions are context-sensitive as a matter of explicitly encoded information; their semantic contribution is relevant to use rather than to truth-conditions. Thus, semantics cannot be simply defined in terms of truth-conditional relevance. Lexically encoded, semantic information may also pertain to how an utterance relates to a previous one, or to the topic of the utterance or to what the speaker is doing – it may pertain to use as well. Irrespective of the kind of information encoded by lexical items, it will deliver only part of the information needed to arrive at a specific proposition: “[The semantic representation] is a kind of template or schema for a range of possible propositions, rather than itself being a particular proposition. As a ‘schema’, it is a formula that contains slots to be filled; what may go into a given slot may be partially constrained by a procedure which specifies how to go about filling it. Such a formula necessarily requires [...] pragmatic inference to develop it into the proposition the speaker intended to express.” (Carston 2002: 57)

This incorporates the conception of procedural information sketched in section 3.1.1, and illustrates that semantics cannot be defined as being equal to conceptual information. The difference between semantics and pragmatics is best stated in performance terms, that is, in terms of the distinction between decoding and inferring, the two types of processes that are involved in utterance interpretation (Carston 2002: 56f). Both conceptual and procedural information may be encoded or may need to be inferred. With respect to aspect, the elements that may need to be inferred in order to deliver the necessary conceptual information have been mentioned in ( 3.16), those that need to be inferred to specify the procedural information will be stated in chapter 529.

28 The matter with ‘I’ is in fact much more complicated, but this discussion is not at issue here. 29 This kind of inferred information cannot be stated cross-linguistically, as it depends on the semantic

specification of the respective aspect marker.

Chapter 3: Utterance Interpretation

90

Remember that pragmatics is conceived as the processes that guide the inferring of additional information and its integration into the information provided by the lexical items used. The additional information is first of all being looked for in the immediate sentence context. Thus, one task of pragmatics can be regarded as “taking context into account in a principled way” (Bunt & Black 2000: 25). This raises three questions that any theory of natural language interpretation has to deal with: what is understood by ‘context’, how is it ‘taken into account’, and what is the ‘principled way’ in which this has to be done. The problem with the notion of context is that it is context-dependent itself – what counts as ‘context’ changes with the focus of investigation, i.e. the specific lexical item, the specific phrase or utterance. Applied to grammatical aspect, local context pertains to the predicate the aspect operator is applied to, whereas global context subsumes elements from the sentence level up to the paragraph or discourse level, as well as factors concerning use and setting. The ‘principled way’ and the ‘taking into account’ will be argued to consist in the default inferences derived at the third layer of meaning.

Perrett’s (2000: 102) conception of pragmatics incorporates the context dependency of context: “Information that is not contained in the string currently being processed is subject to pragmatic actions.”

This definition of pragmatics is developed within the framework of Dynamic Syntax. Like Bunt & Black’s, it incorporates the context-dependency of interpretation by defining as context everything that is not currently being processed. In addition to that, it provides a specification of what is to be understood by ‘context’. Context may be recently processed material, inferred material, or something which is part of the physical situation in which the utterance takes place. Crucially, ‘context’ is specified by the position of the parser – everything that is outside its scope counts as context.

In taking context into account, it is not sufficient just to point out the relevant contextual information. What needs to be specified as well is how hearers add this additional contextual information to what is explicitly given by the speaker’s utterance. This is incorporated in the following definition of pragmatics: “Pragmatics is the study of how hearers add contextual information to the semantic structure and how they draw inferences from what is said.” (Jaszczolt 2002b: 1)

Jaszczolt’s definition emphasises the need to focus on the processes involved in adding additional information, which is crucial to arrive at the level of ‘what is said’. Based on this level, further inferences may be drawn, inferences of the kind of Grice’s conversational implicatures.

Robinson (1997: 254) points out the crucial property of pragmatic theories, which can also be taken as their main task: “Pragmatic theories such as Grice (1989) and Sperber and Wilson (1986) [= Sperber & Wilson 1995. B.S.] […] describe and account for systematic regularities in how linguistic expressions are used and understood in context.”

What a pragmatic theory has to deliver is a systematic description and account of the use of linguistic expressions and of their interpretation in a specific context. Pragmatics consists in explaining the hearer’s task to interpret a certain linguistic input, i.e. the assignment of a unique propositional form to an utterance (Sperber & Wilson 1995: 179). To achieve this, a semantic representation has to be selected, completed and enriched in various ways to yield the propositional form expressed by the utterances – or, at least, what the hearer takes to be expressed by it. The selection of a semantic representation consists in resolving ambiguities, where the hearer has to choose among a clearly defined number of discrete interpretational possibilities. The other tasks are constructional in the sense that the contextually relevant reading is not selected from a certain list of possibilities but rather constructed in the context it appears according to that context.

Chapter 3: Utterance Interpretation

91

What all the above mentioned definitions of pragmatics share is the perspective they take – the perspective of the audience. Furthermore, all these definitions assume a close interconnection of pragmatics with the semantics of the respective forms. Without a principled account of the basic semantics, pragmatic theories are simply ad hoc and do not have any predictive power.

The above considerations seem to amount to a satisfying distinction of semantics and pragmatics. However, even the most satisfying distinction and definition of semantics and pragmatics does not always help in determining whether a specific phenomenon is to be accounted for in semantic or in pragmatic terms. Although the definitions in terms of encoded vs. inferred information seems to be rather straightforward, semantics and pragmatics are too closely intertwined in the interpretation process to be clearly separable in every case: “These broad features […] do not determine on which side of the semantics-pragmatics boundary particular linguistic phenomena fall. Whether a given phenomenon has a semantic or a pragmatic explanation or, as is often the case, some combination of both, must be settled on a case-by-case basis. Obviously it is one thing to formulate the semantics-pragmatics distinction and another to apply it.” (Bach 1999: 81)

Furthermore, languages differ in which side of the semantics/pragmatics boundary a certain linguistic phenomenon falls onto.

To sum up so far, pragmatics as conceived here deals with regularities in use and interpretation of linguistic expressions, focusing on the following questions: What kind of information can possibly be taken into account, which kind of information is the decisive one in a given context and how is this information integrated into what is encoded by the forms used.

Taking pragmatics into account as an additional mechanism in interpretation raises the question, whether a separate, encapsulated and autonomous module should be assumed for it. Under such an assumption, pragmatics would apply ‘after’ semantics. From the above discussion, however, it should have become evident that such a view is not tenable. Pragmatic actions apply at every level of interpretation and therefore cannot apply after the semantic process have been executed30. Pragmatic actions constitute a system which is not specific to language but is related to more general cognitive principles (Robinson 1997). It is different from grammar, but it is not encapsulated in the sense of being isolated from it (cf. Marten 2002). Rather, pragmatics interacts closely with the linguistic system, since the principles that guide the inference process are dependent on and constrained by the encoded information. In this way, pragmatic comprehension is fed by decoding of the lexically given input.

The above discussion should also have illustrated that semantics and pragmatics are not as clearly separable as the notion of ‘interface’ suggests since both interact in a rather complex way. There seems to be no clear-cut dividing line between semantics and pragmatics, even the notion of logical form or semantic representation does not fix this line. Pragmatic processes may as well be necessary to fix the semantic representation, which then can no longer be regarded as the output of semantics serving as input to pragmatics. This is the case, e.g., with ambiguous expressions that deliver two or more distinct contributions to the interpretation process. The right one has to be figured out in order to establish the semantic representation of the sentence. Since this figuring out of the right contribution in a given context requires additional information, this process is pragmatic. In this sense, pragmatics is needed in order to fix the semantic representation which in turn starts off further pragmatic processes (cf. chapter 5, especially section 5.2.2). This is the reason why Marten (2002) suggests to give up the notion of a logical form/semantic representation. However, facing the need to account for the systematicity of language and the reliability of interpretation, this step seems too radical.

30 Note that this does not contradict the assumption of compositionality being guaranteed by the semantic

contribution. In case the exact semantic representation needs to be specified by means of pragmatic reasoning, e.g., in case of disambiguation, it delivers the stable contribution to the interpretation process once it is fixed.

Chapter 3: Utterance Interpretation

92

Although there is no clear-cut boundary between the domains of semantics and pragmatics in interpretation, there is a domain where both closely interact. This domain of interaction neither pertains to semantics nor to pragmatics alone, and can neither be located at the level of word/sentence meaning nor at the level of utterance meaning, but seems to be located somewhere in between. This domain of interaction between semantics and pragmatics has been captured by different notions, like, for instance, Carston’s (2002) ‘explicature’, Bach’s ‘impliciture’, and the level of ‘utterance-type’ meaning giving rise to generalised conversational implicature (GCI) postulated by Levinson (1995; 2000a).

Explicature, impliciture and GCI all relate to Grice’s distinction between ‘what is said’ and ‘what is implicated’, but specify and locate the boundary between the two notions in slightly different ways. Table 3.2 shows some examples of the terminology that is used in this domain (cf. Levinson 2000a: 195): semantic

representation

deictic & reference resolution

minimal proposition

enriched proposition

additional propositions

Grice (1989) what is said implicature semantics explicature

Carston (2002) what is said

implicature

Bach (1994) what is said impliciture implicature what is said Levinson (2000a)

the coded implicature Table 3.2

Carston’s notion of explicature refers to the proposition expressed. This proposition constitutes a development of the logical form of a sentence, that is, an enriched semantic representation. It is derived via decoding (of what is provided by semantics) and inferring. An explicature is a propositional form communicated by an utterance which is pragmatically constructed on the basis of the template the utterance encodes (Carston 1998a). Carston extends Grice’s notion of ‘what is said’, as she assumes the enriched proposition also to be part of this level. Within her framework, two types of semantics are postulated, namely lexical and truth-conditional semantics. She also postulates two types of pragmatics: one that develops the logical form encoded by the lexical items to yield truth-evaluable propositions and thus immediately interacts with semantics (pragmatics1), and one which derives implicatures from that level and thus acts above the immediate structure of semantic representation (pragmatics2). Since explicatures are derived not only by decoding but are also in need of inferences, pragmatics plays a role already at the subpropositional level. The process of developing the semantically encoded information into an explicature is constrained by truth-evaluability as lower bound, and by relevance, i.e. the presumed speaker‘s meaning, as upper bound.

Since there does not seem to be a clear border-line between what is said and what is implicated, Bach (1994) proposes a middle ground between both. This middle ground constitutes what is implicit in what is said. The respective kind of inferences are implicitures (with an -i- to distinguish them from implicatures): “Implicitures go beyond what is said, but unlike implicatures, which are additional propositions external to what is said, implicitures are built out of what is said.” (Bach 1994: 139)

Implicitures thus differ from explicatures in that they are not part of what is said but located on a different level. What both notions share, however, is that they are context-based and context-dependent.

Chapter 3: Utterance Interpretation

93

Levinson leaves ‘what is said’ as specified by Grice. Contrary to Grice, however, he takes implicatures to play a considerable role also in specifying ‘what is said’. Thus, they take part in deictic and reference resolution, as well as in disambiguation of the coded information. Disambiguation may be mediated by Gricean procedures: the speaker must presume that the addressee will use the same salient stereotype (regardless of its factual probabilities) to resolve the ambiguity31. Levinson (1995; 2000a) proposes a middle ground between ‘what is said’ and ‘what is implicated’. This middle ground is the level of ‘utterance type meaning’ where Grice’s generalised conversational implicatures (CGI) are located at. Contrary to explicatures and implicitures, the inferences derived at this level are not context-dependent. This is the reason for their status as default inferences. Levinson’s notion of GCI will be introduced in more detail in section 3.2.2, where it is argued that this conception is the most appropriate one in dealing with the interpretation of closed class morphemes like grammatical aspect.

The three notions of explicature, impliciture and CGI can be characterised according to whether they are assumed to constitute an extra level of meaning and whether they are context-dependent:

separate layer context-dependent explicature - + impliciture + + default inferences (GCI) + -

Table 3.3 Despite these differences, the approaches presented above share one crucial insight, namely that in utterance interpretation, semantics and pragmatics work hand in hand. The overall picture arising from the above discussion can be summarised as follows: There are basically two kinds of semantics – lexical semantics and truth-conditional semantics, and two kinds of pragmatics – pragmatics1 contributing to truth-conditional semantics (i.e. to ‘what is said’), and pragmatics2 yielding additional implicatures in the original Gricean sense (‘what is meant’).

Figure 3.4 sketches the interaction of the two kinds of semantics and pragmatics and their contribution to utterance interpretation, located at three different levels: 1) the level of logical form/semantic representation of the explicitly encoded information (‘what is coded’), 2) the propositional/truth-conditional level (‘what is said’), and 3) the level of implicatures (‘what is meant’).

Figure 3.4 Level 1 is subject to decoding and is therefore autonomous with respect to pragmatics. Levels 2 and 3 both are supplied by pragmatic inference processes, but whereas at level 2 both

31 This will prove important in analysing the ipf aspect in Russian. In chapter 5, section 5.1.2 it will be

argued that ipf aspect is ambiguous and the disambiguation needs a certain amount of pragmatic reasoning.

Truth-conditional

Pragmatics1

Lexical

Pragmatics2

Semantics

Pragmatics3

2 Proposition

Implicatures

1 Sem. Repres.

Chapter 3: Utterance Interpretation

94

decoding and inference both play a role, at level 3 it is only inference, yielding Gricean implicatures. To avoid terminological commitment to one of the theories above, these two kinds of pragmatics will be referred to simply as pragmatics1 and pragmatics2 (primary vs. secondary pragmatic processes, cf. Turner 1997: 167). These three levels are not independent or sequential, but rather interact insofar as the pragmatic part poses certain constraints on the semantic part and vice versa. On the one hand, since pragmatics plays a role in utterance interpretation, the semantic representation does not have to supply all the information necessary for utterance interpretation. On the other hand, since pragmatic processes crucially rely on the semantically provided information, the semantic representation constrains the inferences it gives rise to.

The different accounts outlined above differ in what they regard as being coded or inferred, and in the assumptions about what is explicitly conveyed and what implicitly. This discussion mainly pertains to the interaction of truth-conditional semantics and pragmatics1 that both interact to yield a proposition (cf. the dark grey areas in Figure 3.4).

This section has argued for taking into account pragmatic mechanisms in order to bridge the gap between encoded information and propositions. The general picture of how utterance interpretation proceeds can be outlined as follows: “The content is represented by an utterance, a token of the sentence in a context, understood according to the rules of the grammar and to the conventions of interpretation shared by language-users of the language of the sentence.” (Atlas 1989: 28)

The task of pragmatics is to account for which parts of context are taken into account in which way in order to arrive at an interpretation. The ‘conventions of interpretation shared by language-users’ Atlas talks about in the quotation above are dealt with in the next section. It will be argued that these conventions are specified at the level of utterance-type meaning introduced by Levinson. They have the form of heuristics which give rise to default inferences and default interpretations.

3.2.2 From sentence meaning to utterance meaning The central question with respect to utterance interpretation is how it is possible for it to be flexible and reliable at the same time. The bottleneck in human communication consists in the relation between the slow transmission rate of human speech and the rather fast functioning of all other aspects of speech production and comprehension (Levinson 2000b). In order to make the use of language successful to a certain degree, the speaker has to rely on certain mechanisms that he can assume to be shared by the hearer. Otherwise, communication would not be as successful as it seems to be32. This section addresses the contribution of pragmatics to the interpretation of underspecified forms and it sketches the mechanisms that are important for the interpretation of Russian ipf aspect.

The theories outlined in the preceding section are all dealing with the question of how to distinguish ‘what is said’ from ‘what is meant’ and how to account for the processes that play a role in these two domains. To shortly recap, Carston (2002) proposes the notion of ‘explicature’ as constituting an enriched but still explicitly conveyed proposition, Bach (1994) proposes a middle ground of ‘impliciture’ between ‘what is said’ and ‘what is meant’. Levinson, however, sees “no consistent way in cutting up the semiotic pie such that ‘what is said’ excludes ‘what is implicated’ ” (2000a: 198). Based on this assumption he does not see any sense in the terminological efforts of delimiting explicatures or implicitures from implicatures: “Merely relabeling pragmatic inferences that play a role in fixing the intended proposition something other than ‘implicature’ hardly alters the nature of the problem.” (Levinson 2000a: 198)

32 Again, this is not to say, that the hearer interprets an utterance always and one-hundred-percent as intended

by the speaker. This is an idealisation that is hardly ever achieved with natural language discourses.

Chapter 3: Utterance Interpretation

95

Gricean inference is not constrained to operating on the propositional level, but the respective processes are also involved in deriving propositions, cf. Atlas (1989: 146): “Pragmatic inference assists in the construction of the logical form of (or propositions expressed by) an utterance. Gricean inference is not restricted to mapping propositions (‘what is said’) into propositions (‘what is implicated’). It also maps Lockean ‘abstract’ sentence meanings into propositions (utterance meanings).”

The decisive distinction in Levinson’s conception is that between decoding and inferring, the two different processes that both may contribute to what is said: “I shall presume that we want to define the types of content by the processes that yield them and the important semantical properties they have (e.g. default presumptions, defeasibility under distinct conditions, …).” (Levinson 2000a: 198)

Table 3.4 shows how Levinson conceives of the distinction between what is said and what is meant, and the respective processes:

what is said what is meant semantic

representation deictic/reference

resolution minimal

proposition enriched

proposition33 additional

propositions the coded DECODING

implicature INFERRING

Table 3.4 The basic distinction within this conception is that between decoding and inferring. Implicatures are derived via inferring and are necessary for both ‘what is said’ and ‘what is meant’. The fact that ‘what is said’ both determines and is determined by implicature is what Levinson calls ‘Grice’s Circle’ (2000a: 172-174; 186f): implicatures depend on ‘what is said’, but ‘what is said’ in turn depends on processes like disambiguation, indexical resolution, reference fixing, ellipsis unpacking and generality narrowing. These processes involve the same inferential mechanisms that also characterise Gricean pragmatics. Accordingly, there are implicatural contributions to ‘what is said’, and therefore Levinson’s only distinction is that between ‘the coded’ and ‘the implicated’. ‘What is said’ constitutes the truth-conditional content and depends on pragmatic inference, that is, pragmatic inference is a crucial input to semantic interpretation. Levinson thus distinguishes two kinds of semantics (cf. also section 3.2.1, Figure 3.4): Semantics1 extracts an impoverished level of semantic representation from surface structure and lexical content, whereas semantics2 operates over a richer level of semantic representation to which pragmatics has contributed in fundamental ways (by the pragmatic processes mentioned above). This level of semantics2 is essentially linguistic and cannot be equated with conceptual structure (Levinson 2000a: 166/fn. 6).

Although this seems to imply that all implicatures are of the same type, Levinson still draws one crucial distinction: context-dependent implicatures about speaker’s intentions based on the results of semantic representation on the one hand, and context-independent implicatures that do not hinge on the recovery of speaker’s intentions and contribute to semantic representation on the other hand (Levinson 1995: 93). These latter are based on default heuristics that derive from observed regularities in language use and interpretation. They rely on the coded information and on the form of the lexical items used and are thus available to all interlocutors in a communication. These implicatures are the generalised

33 The distinction of minimal and enriched proposition is repeated here from Figure 3.3, but will be given up

in the following discussion. This distinction is primarily necessary for theories that take inferences contributing to ‘what is said’ as different from those contributing to ‘what is meant’ (e.g. Relevance Theory). In Levinson’s framework there is no such distinction, and hence no need to distinguish minimal and enriched propositions.

Chapter 3: Utterance Interpretation

96

conversational implicatures that constitute the heart of Levinson’s GCI theory which will be introduced in this section and applied to the interpretation of Russian ipf aspect in chapter 5.

As has been outlined in the sections above, interpretation involves a considerable amount of inferences. The only way to constrain such inferences and make them predictable to a certain degree is for speaker and hearer to share simple, but powerful heuristics. These heuristics multiply the coded information by licensing inferential enrichment of what is encoded and at the same time constrain the enrichment so that the overall message is still recoverable. Levinson (1995; 2000a) provides a further development of the Gricean approach by relating Grice’s speaker-based principles to the hearer. In this way he brings together both parts of the communication process. Levinson conceives Grice’s maxims as heuristics that constitute the mechanisms to start enrichment and specification, guide it, stop it and make it predictable – they guarantee the reliability of interpretation. These maxims (Grice 1989: 26f) are listed in ( 3.19)-( 3.23): 3.19 The maxim of quantity, relating to the quantity of information to be provided:

a. Make your contribution as informative as is required (for the current purpose of exchange).

b. Do not make your contribution more informative as required. 3.20 The maxim of quality, requiring to make the contribution a true one; specifically:

a. Do not say what you believe to be false.

b. Do not say that for which you lack adequate evidence. 3.21 The maxim of relation:

Be relevant. 3.22 The maxim of manner:

a. Avoid obscurity of expression.

b. Avoid ambiguity.

c. Be brief (avoid unnecessary prolixity).

d. Be orderly. All these maxims follow from one general principles which the participants in a communication are assumed to follow: 3.23 The cooperative principle:

Make your conversational contribution such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which you are engaged.

As Robinson (1997) points out, the notions used to describe these maxims – relevance, manner, informativity – are interpretations of some observed behaviour. In order to apply these principles, it is necessary to have access to information determining whether a certain usage or interpretation is informative, relevant, orderly, etc. Such notions are culturally dependent, and may even vary from speaker to speaker, and from discourse setting to discourse setting. In trying to capture observed regularities of language use and interpretation,

Chapter 3: Utterance Interpretation

97

there is no problem in regarding these heuristics as the basic mechanisms that give rise to default interpretation, but they must not be taken for the explanation of that behaviour. The Gricean rules are descriptions of the relationship between a word’s meanings and its use, and as such they specify linguistic behaviour at the level of use and specify regularities in use and understanding. However, such a descriptive theory may not be turned into a cognitive one. Grice was not interested in the psychological reality of his maxims, thus objections to his principles that are based on the requirement of psychological reality will fail (Saul 2002). Robinson (1997) takes pragmatic principles like the Gricean maxims as tendencies that have to be taken as pointing to regularities in the operation of a cognitive system. The pragmatic principles can be considered as a number of behavioural tendencies, but the explanation for these tendencies has to be sought in a cognitive model: “Pragmatic theories are useful and necessary, but we must be aware of what they are and where they come from. The tendencies described in these theories should be taken as pointers to regularities in the operation of a cognitive model.” (Robinson 1997: 260).

The Gricean maxims are criticised mainly because of their vague nature. Frederking (1996)34, for instance, claims that Grice’s maxims are similar to the maxim ‘do the right thing’. He argues that they are to be treated as what they are – as maxims, and not as rules that have to be followed and are followed in communication. But as such, they are what they are intended to be – generalisations of observed behaviour. As generalisations, they merely capture the overall picture and therefore allow for exceptions as well as for more specific principles on a more fine-grained level of investigation. Thus, Frederking’s criticism of the Gricean maxims, namely, that there exist much more specific mechanisms that apply to much more specific phenomena, is in fact a mere description of how maxims work. Maxims have to remain general to describe the way organisms arrive at certain decisions, such as interpretations.

Levinson (2000a) argues that in order to make Grice’s maxims account for the recognition of an interlocutor’s intentions, they have to be thought of in a different way, namely in terms of heuristics: “Instead of thinking about them [the maxims, B.S.] as rules (or rules of thumb) or behavioral norms, it is useful to think of them as primarily inferential heuristics which then motivate the behavioral norm. [emphasis in the original, B.S.]” (Levinson 2000a: 35)

Just like the Gricean maxims, these heuristics do not constitute the basic mechanisms themselves, they rather describe certain regularities in language use and interpretation and do not create by themselves that regular behaviour.

Before turning to the heuristics Levinson proposes to apply in utterance interpretation, it is expedient to give a short remark on the general utility of heuristics – in particular in view of their usefulness and appropriateness in accounting for the interpretation process being exposed to criticism, mainly by relevance-theorists.

Minds constantly have to make decisions in a world where knowledge is limited and time is pressing. In order to account for that, Gigerenzer et al. (1999) propose simple heuristics as rules for making decisions with realistic mental recourse – fast, frugal and accurate heuristics35. By drawing inferences, fast and frugal heuristics are claimed to capture the way real minds make decision under the constraints mentioned. In handling tasks like interpretation, humans have to use approximate methods and simple strategies that guide the search for information complementing the coded information and stop mechanisms that determine when this search should end. What is also needed are decision rules that make use of the information found. Basic to such reasoning is the concept of ecological rationality, the study of which involves analysing the structure of environments, the structure of heuristics

34 Frederking argues mainly from the perspective of computational linguistics. 35 Wilson (2003) also emphasises the usefulness of heuristics in effective interpretation.

Chapter 3: Utterance Interpretation

98

and the match between them. Ecological rationality makes it possible to go beyond the assumption that basing decision making on more information and computation will always lead to more accurate inferences. Fast and frugal heuristics work that well because they make a trade-off between generality and specificity. Their simplicity allows them to be robust when confronted with a change in the environment and to generalise to new situations instead of having to be substituted by a new set of heuristics for every environmental change or every new situation. Thus these heuristics can be accurate without being too complex and thereby avoid being too closely tied to any particular environment. Fast and frugal heuristics pay attention to systematic informative cues while overlooking more variable and more uninformative ones.

Such heuristics work at Levinson’s third layer of meaning which is located between the encoded meaning of linguistic expressions and inferences to speaker meaning. Levinson (1995: 110) argues against a conflation of this third level of meaning with semantics or pragmatics: “Because this [third] level of meaning sits midway between grammar and lexicon on the one hand and speaker-meaning on the other, most analysts attempt to reduce it in one or the other direction. This is a mistake, because it is a level with distinct properties – default, defeasible inferences based on the comparison of alternative linguistic expressions and on the presumption of stereotypical situations, which interact in specific ways.”

Levinson introduces this level of utterance type meaning arguing that the distinction between sentence meaning and utterance meaning does not suffice. By sentence meaning he understands the overall meaning composed from the meanings of all constituents together with the meaning of the constructions in which they appear. Utterance meaning refers to the meaning of the same sentence when used in a particular context (section 3.1.1). This latter kind of meaning has to be divided into utterance type and utterance token meaning, the decisive part for the purposes here being the former one. The level of utterance type meaning is posited between the encoded sentence meaning and the intended speaker’s meaning and constitutes the basis on which systematic, i.e. contextually-independent, pragmatic inferences can be drawn. As these inferences (GCI) are default inferences that go through in the absence of evidence to the contrary, they are available to both speaker and hearer. They are one of the main reasons for the efficiency, speed and reliability of communication.

This third layer is thus based on general expectations about how language is normally used: „This third layer is a level of systematic pragmatic inference based not on direct computations about speaker-intentions, but rather on general expectations about how language is normally used. These expectations give rise to presumptions, default inferences, about both content and force; and it is at this level (if at all) that we can sensibly talk about speech acts, presuppositions, felicity conditions, conversational presequences, preference organisation and [...] generalised conversational implicatures [emphasis in original].“ (Levinson 1995: 93)

At this layer semantics and pragmatics meet and interact. This shows that any pragmatic analysis is highly dependent on the semantic analysis of the respective expressions, “to the extent that it can become problematical to decide which phenomena should be rendered unto semantic theory and which unto pragmatics” (Levinson 2000a: 23). Positing this third level acknowledges the lack of a sharp and clear cut interface and makes it possible to deal with this fact in a systematic way, without regarding it as something undesirable. At this third level, the phenomenon of pragmatic inference is given the systematicity that is indispensable for any account of how interpretations are derived. It is especially applications in natural language processing that can profit from such a level of default inferences. Levinson’s proposal of a layer of utterance type meaning is also central to the account of aspect interpretation.

At the level of utterance type meaning, heuristics give rise to default inferences that are generalised to a considerable degree, i.e. context-independent and systematic. Since these heuristics are generalised, they allow for expectations about inferences that ought to arise from particular expressions; they are about what speaker and hearer can be assumed to understand by a certain way of putting things. Thus, GCI are inferences from structure and

Chapter 3: Utterance Interpretation

99

meaning of the lexical items used to further presumptive meanings. They are therefore deeply interconnected with linguistic structure and word or sentence meaning. These inferences arise due to certain heuristics connected with the use of language. Levinson does not conceive pragmatics as a separate module, but he supposes that there is a certain body of knowledge associated with the use of language. This body involves, among other things, meta-linguistic knowledge about the structure of the lexicon, about the structuring of semantic fields and about the availability of alternate expressions. Crucial for GCI is the mutual awareness of the interlocutors of the underlying central heuristics. As Levinson emphasises, pragmatic reasoning of this sort must be organised in a way such that it can interact with the results of semantic processing at any stage of that processing. This yields a level of representation to which semantic contributes entailed conditions and default inferences contribute defeasible pragmatic content. This whole structure then – including the contributions of semantics and pragmatics – can be semantically interpreted and can be further enriched by additional pragmatic inferences based on the propositional content. The central contribution of GCI theory, according to Levinson, is that it suggests specific ways in which the possible interpretations of an underspecified semantic representation of the meaning of a sentence are not equipollent, but systematically biased even before specific contextual influence comes into play. That is, GCI theory allows for systematic and predictive inferences that are based on the encoded information provided by the lexical items used.

To warrant the predictability of the respective inferences, the semantics of the basic expressions needs to be specified in a way that allows for the inferences to apply. In this respect, GCI theory can be said to constrain the form of the basic semantic theory: “The more that can be said about default pragmatic inferences, the more we can guess about the nature of the semantic representations that must support them.” (Levinson 2000a: 377)

Attributing to pragmatics a more prominent role in the interpretation process and acknowledging its contribution in linguistic theory may furthermore lead to a shift in how semantics is conceived of: “The question for linguistic semantics is not what contribution an expression makes to truth conditions, but rather what kind of contribution it makes to pragmatic inference.” (Blakemore 2002: 78)

As regards the interpretation of aspect, the semantics that will be elaborated in chapter 4 will be specified in a way which enables the application of pragmatic principles. As pragmatics is assumed to play an important role in the interpretation process, the encoded information does not have to be as detailed as it would have to be without considering pragmatic interaction. The semantic analysis to be developed for aspect allows for systematic inferences based on both form and meaning of the items used, just as GCI theory predicts.

Levinson’s GCI theory is based on two of Grice’s maxims, namely quantity and manner. The heuristics are labelled after the maxim they derive from: Q-inferences are based on the first quantity maxim (compare ( 3.19a)) and license inference to the negation of a stronger interpretation, M-inferences stem from violations of the manner maxim (compare ( 3.22)), and license the inference from marked expressions to marked interpretations. I-inferences are based on the second quantity maxim (compare ( 3.19b)) and allow for inference to a stereotype.

The Q-principle is related to Grice’s first quantity maxim: make your contribution as informative as required. It is stated as follows (Levinson 2000a: 76): 3.24 Q-Principle

Speaker’s maxim: Do not provide a statement that is informationally weaker than your knowledge of the world allows. […] Specifically, select the informationally strongest paradigmatic alternate that is consistent with the facts.

Chapter 3: Utterance Interpretation

100

Recipient’s corollary: Take it that the speaker made the strongest statement consistent with what he knows.

Q-Heuristic: What isn’t said isn’t. The next principle, the M-Principle, derives from Grice’s manner-maxim and specifically pertains to the submaxims “avoid unnecessary prolixity” and “avoid obscurity of expression”. Levinson formulates the M-Principle as follows (2000a: 136): 3.25 M-Principle

Speaker’s maxim: Indicate an abnormal, nonstereotypical situation by using marked expressions that contrast with those you would use to describe the corresponding normal, stereotypical situation.

Recipient’s corollary: What is said in an abnormal way indicates an abnormal situation, or marked messages indicate marked situations.

M-Heuristic: What is said in an abnormal way isn’t normal.

The I-principle is based on Grice’s second quantity maxim saying that what is expressed simply is stereotypically exemplified. The I-principle is stated as follows (Levinson 2000a: 114): 3.26 I-Principle

Speaker’s maxim: The maxim of minimization. “Say as little as necessary”; that is, produce the minimal linguistic information sufficient to achieve your communicational ends (bearing Q in mind).

Recipient’s corollary: The enrichment rule. Amplify the informational content of the speaker’s utterance, by finding the most specific interpretation, up to what you judge the speaker’s […] intended point, unless the speaker has broken the maxim of minimization by using a marked or prolix interpretation.

I-Heuristic: What is expressed simply is stereotypically exemplified.

This principle requires the speaker to produce the minimal linguistic information that is sufficient to achieve his communicative goals. The hearer’s corollary is an enrichment rule, allowing to amplify the content of the perceived utterance to the most specific interpretation. This includes, e.g., the search for maximal cohesion in terms of temporal, spatial, causal and referential connectedness, and the presumption of stereotypical relations. All these different cases are based on the general instruction to enrich.

It is questionable, however, whether the third principle, the I-principle, really gives rise to general and context-independent heuristics as the Q- and M-principles do. Inferences evoked by Q- and M-heuristics do not need any context, they are based on linguistic alternates with essentially the same form and contrastive semantic content in the case of Q-inferences, and with contrasting forms but essentially the same inherent semantic content in the case of M-

Chapter 3: Utterance Interpretation

101

inferences. The I-heuristic, however, gives rise to inferences that arise from the interaction of semantic representation with context and conceptual knowledge. Therefore, Levinson’s (2000b: 4) general claim that the heuristics “have the requisite property of inducing default inferences, because they are based solely on properties of the form and content of the utterance, not on extra propositions about the context” seems at least questionable with respect to I-inferences. The only generalised instruction I-inferences give rise to is the instruction to enrich the decoded semantic information. The exact content of this enrichment is context-dependent. I-inferences demand access to conceptual knowledge to be applicable in a reasonable way.

To sum up so far, Levinson’s heuristics are not to be confused with cognitive explanations. Rather, they are descriptions of observed regularities in language use and interpretation. They are instances of fast and frugal heuristics giving rise to defeasible default inferences. Their usefulness in interpretation consists in their trade-off between generality and specificity, and in their mutual availability to both speaker and hearer. They are specific in that different lexically encoded phenomena license different heuristics, and they are general in that each heuristic may give rise to a certain range of interpretations. These heuristics constrain the search space for possible interpretations to a considerable degree. In this sense, they are the right means to deal with underspecification of linguistic expressions. Levinson takes the role of GCI in linguistic theory as an additional tool in the explanatory armoury. GCI theory is predictive, but precise predictions depend on the semantic analysis of the relevant expressions (Levinson 2000a: 368/fn. 2).

It is interesting to see how Levinson’s GCI relate to the original Gricean account. Grice (1989: 88-91) distinguishes four main forms of meaning specification (Grice 1989: 90f): 3.27 a. the specification of the timeless meaning for an utterance type

b. the specification of the applied timeless meaning of an utterance type that captures disambiguations

c. the specification of the utterance type occasion meaning capturing what the speaker means by using his words

d. the specification of the speaker’s occasion meaning, that captures the intentions of the speaker GCI are located between the second and the third specification, that is, between the ‘applied timeless meaning’ and the ‘occasion meaning’ of the utterance type (Levinson 2000a: 373). They can be conceived of as mediating between these two levels. What is crucial in this respect is the fact that GCI are associated with utterance types which can be accessed without considering the speaker’s occasion meaning. That is the reason for why GCI can be viewed from both perspectives, from the speaker’s and from the hearer’s. Levinson therefore posits corresponding hearer’s corollaries to the maxims cf. ( 3.24)-( 3.26). The independence of GCI from speaker’s intention also justifies their conception as defaults that are based on form and (coded) meaning of the respective expressions.

An analysis of the interpretation process in terms of generalised conversational implicatures applying to utterances types and constituting default interpretations may be applied to grammatical morphemes as well, as Levinson (1995:106) points out: “Exactly the same kind of analysis can now be applied to any grammatically closed class of morphemes, and should yield the same kind of harvest: invariant core meanings supplemented by preferred interpretations”

Such an implicature analysis of closed class morphemes “may yield many important insights and resolve many puzzles that arise from the apparently elusive and variable content of these expressions” (Levinson 2000a: 98). This is exactly the case for Russian aspect – an

Chapter 3: Utterance Interpretation

102

implicature analysis lightens the burden for semantics and accounts for the variability in use and interpretation, not only in Russian, but also cross-linguistically.

This section has argued for the conception of a third layer between utterance meaning and speaker’s meaning, namely for the level of utterance type meaning. This level is important as a special sort of generalised pragmatic principles applies here – GCI – that allow for a systematic account of how pragmatics interacts with what is encoded by the utterance type meaning. For these principles to be predictive, exactly that coded utterance type meaning has to be analysed and specified. Acknowledging the contribution GCI to the interpretation process lightens the burden of semantics to a considerable degree and consequently helps to shape the semantic representation of the linguistic expressions in question. The semantic analysis of aspect that will be sketched in chapter 4 is elaborated with respect to the application of GCI.

3.3 Conclusion

Starting from the supposedly contradictory characterisation of natural language interpretation as being stable and flexible at the same time, this chapter has delivered an account of how these two properties can be accommodated and integrated into an account of utterance interpretation. The stable contribution of lexical items to utterance interpretation is encoded in their respective lexical entries, the make-up of which has been sketched in terms of three different subentries – a lexical, a logical and an encyclopaedic one. Two basic kinds of encoded information have been distinguished: conceptual and procedural information. With respect to the interpretation of grammatical aspect, the Aktionsart basis encodes conceptual, the aspect marker procedural information. Both have to come together at an operational interface that is provided by matching units specified by the logical entries of both. As regards the Aktionsart, these are phases and boundaries located at the temporal tier, as regards aspect, this is the information which of these units are to be selected. The information that is encoded by the aspect marker will be dealt with in chapter 4.

Linguistic expressions deliver merely underspecified templates that need to be enriched and specified in the course of interpretation. In order to bridge the gap between encoded information and interpretation, it has been argued for the incorporation of pragmatic mechanisms. Pragmatics has been defined in terms of information-inferring and the processes guiding these inferences. There is no clear-cut semantics/pragmatics interface, rather pragmatics applies at every level of interpretation. The regularities of language use and interpretation in a specific context are best captured by inferential heuristics that are available to all participants in a communicative setting. These heuristics give rise to default inferences – generalised conversational implicatures (Q-, M- and I-implicatures). The interaction of semantics and pragmatics in the interpretation of aspect is summarised in Figure 3.5, that supplements Figure 1 from chapter 2 with Levinson’s Q-, M- and I-inferences. Figure 3.5 is still general and does not specifically pertain to aspect in Russian. It is based on the assumption of a cross-linguistic characterisation of aspect (to be delivered in chapter 4) and the cross-linguistic validity of the respective pragmatic principles. It shows the two kinds of semantics and pragmatics with the interpretation of grammatical aspect and what they consist of. Note that pragmatics2 will not be dealt with here. ‘Interpretation’ refers to the outcome of semantics2. It will be shown that with the interpretation of Russian ipf aspect, pragmatics is involved in arriving at semantics1 (chapter 5, section 5.1.2).

Figure 3.5 also reflects the basic difference between Q-/M-inferences on the one hand, and I-inferences on the other. While the former are based on the logical entry of aspect (Q) and the interaction of the logical entries of aspect and Aktionsart (M), respectively, I-inferences take into account also the encyclopaedic entry of the Aktionsart basis. The latter inferences are therefore neither meta-linguistic nor context-independent, and it is at this level of

Chapter 3: Utterance Interpretation

103

utterance interpretation that Relevance Theory is more appropriate than Levinson’s neo-Gricean pragmatics.

Figure 3.5 For the systematic derivation of ipf readings in Russian, the distinction between semantics of the form (word meaning, combined by the respective contributions of aspect and Aktionsart) and its interpretation (hearer’s meaning) is of special importance. Confusion of both has lead to definitions of ipf aspect as ‘incomplete action’ or ‘ongoing process’ and the like. These notions represent some of the possible interpretations of the form, the semantics of which cannot be captured by either one of these ‘definitions’. In order to systematically derive interpretations of certain lexical items, their semantics has to be made explicit and taken as the basis for all the possible interpretations, and all the possible combinations with certain

cros

s-lin

guis

tic;

inva

riant

sem

antic

m

eani

ng o

f asp

ect

lang

uage

-spe

cific

; di

ffer

ent i

nsta

ntia

tion

of g

ener

al

sem

antic

s and

pra

gmat

ics

grammatical aspect

Aktionsart operation interface

OPERATOR procedural information

OPERANDUM conceptual information

PROPOSITION (I)

encyclopaedic entry:

conceptual knowledge

SEMANTICS SEMANTICS

INTRA AD POST

▪ oppositions ▪ markedness relations

pragmatic inferences (Q)

pragmatic inferences (M)

interaction

logical entries(ϕdyn, τ, ϕstat)

event encoding

additional implicatures

Sem1

Prag1

Sem2

Prag2

Chapter 3: Utterance Interpretation

104

contextual factors. The additional contextual factors contributing to the interpretation, however, do not have to be explicitly encoded. They may as well pertain to background, world and conceptual knowledge and therefore do not belong to semantics proper but are activated by pragmatic reasoning. Interpretation does not only consist in decoding and combining of what is explicitly encoded, but also in inferring additional components. Importantly, however, semantics determines the range of factors possibly taken into account. Therefore, the semantic and pragmatic side of analysis are closely related and interdependent. With respect to grammatical aspect, what has to be provided for as the semantic basis is the common denominator for all the possible interpretations that may be derived from that form. This semantics has to provide the frame for and constraints on possible interpretations.

In the next chapter, a semantic analysis of aspect is developed that allows for pragmatics to interact in a way that makes interpretation reliable even with an underspecified basis. How pragmatic principles work with the interpretation of Russian ipf aspect is elaborated in chapter 5.

4 SEMANTICS

In the present chapter, a semantic analysis of grammatical aspect will be developed that accounts for the Russian data as well as for the cross-linguistic facts. To achieve this, it is necessary to strictly separate aspect semantics from aspect interpretation1, and encoded information from inferred information. The explicitly encoded information is uniform and stable across contexts and speakers and therefore basic to all possible interpretations. This holds for the language-specific aspect systems as well as for the general cross-linguistic characterisation of grammatical aspect. The semantic basis for aspect, in whatever language it appears, consists in the establishing of a relation between a sentence base and a topic component. Despite this common basis, the language-specific differences in aspect use and interpretation cannot be denied. The English progressive, for instance, differs in certain respects from its alleged equivalents -iyordu in Turkish and the ipf aspect in Russian. It will be shown that the cross-linguistic differences are not completely idiosyncratic but can be explained in a rather straightforward way. They arise from differences in language-specific factors such as the inventory of aspect markers, the respective markedness-relations holding among them and the pragmatic inferences based on both these facts. This division of labour between semantics and pragmatics and their interaction takes a considerable load from semantics.

Terminality categories provide a conceptually grounded typology of possible instantiations of grammatical aspect. Each terminality category can be assigned one invariant semantic characterisation. Terminality categories are instantiated by language specific aspect markers2, which again can be assigned one invariant semantic basis. Thus grammatical aspect in Russian shares the basic characteristics with aspect categories of other languages. On a more specific level, aspect in Russian instantiates ‘adterminality’, the semantic basis for which is the same in all its language-specific instantiations, namely envisaging the event in the attainment of its limit(s) (Johanson 2000; cf. chapter 2, section 2.2.2). In Russian, this category is instantiated by the pf aspect as the marked member of the aspect opposition, which explicitly selects the boundaries and relates them to a bounded topic time interval. This semantics is basic to all the different interpretations the pf aspect may give rise to. The situation is more complicated with the (semantically) unmarked3 ipf aspect, which quite successfully seems to resist the assignment of one stable semantic basis.

Figure 4.1 illustrates the connection between the cross-linguistically invariant meaning of grammatical aspect, and the invariant semantics that is basic to its language-specific instantiations. The differences in formal means for expressing the respective terminality category give rise to different oppositions and to different markedness relations (indicated by ‘+’ for the marked member(s) and ‘±’ for the unmarked ones)4. The contextually relevant interpretations of the respective forms result from the application of the aspect marker onto the Aktionsart basis; aspect and Aktionsart work together in deriving the final interpretation of the finite verb in a certain context. To elaborate this interaction, the semantics of both needs to be formulated in a way that allows for their interaction. This requires figuring out the

1 ‘Aspect interpretation’ is a shortcut for ‘interpretation of an utterance containing a predicate which is

marked for a aspect`. 2 Note that ‘markers’ do not necessarily require overt expression. 3 The need to distinguish different kinds of markedness has been pointed out in chapter 2, section 2.1 (cf.

also chapter 5). 4 By the end of this chapter it will become clear how the meta-categories PERFECTIVE and IMPERFECTIVE

relate to these different levels of invariant meaning.

106 Chapter 4: Semantics

common features that make possible this operator-operandum relationship between aspect and Aktionsart (cf. the discussion on conceptual and procedural information in chapter 3, section 3.1.1).

Figure 4.1 This chapter is structured as follows: In section 4.1, the different components contributing to the semantics of grammatical aspect will be figured out. In sections 4.2-4.4, Russian, Bulgarian and Turkish will be analysed with respect to the specification and interaction of these components. Section 4.5 comprises the main results by comparing the three languages. This comparison will show that the semantic analysis developed in this chapter constitutes a cross-linguistically valid skeleton based on which the language-specific differences can be accounted for in a systematic and predictable way.

Moreover, two excursus will be provided that deal with peculiarities of the Bulgarian and the Turkish aspect systems. Section 4.4.2 proposes an analysis for the Bulgarian ipf Aorist, section 4.4.2 focuses on the so-called ‘Aorist’ in Turkish. These excursus are not intended to deliver a full-fledged account of the phenomena under consideration, but are rather meant to point out two interesting issues of the respective aspect systems.

4.1 Grammatical aspect cross-linguistically In this section, an analysis of aspect will be developed that takes aspect as one means to express the notion of finiteness (cf. Klein 2001, 2003). Aspect serves to relate a sentence base (provided by the Aktionsart basis) to a topic component and thereby marks an assertion. It will be argued that establishing this relation constitutes the general semantics of aspect in the sense of level 1 in Figure 4.1. The different cross-linguistic types of this relation are captured by the different terminality categories. It is not the whole sentence base that proves crucial in this respect, but certain parts of it which are specified by the verb and its arguments, i.e. the predicate. The topic constituent these parts are related to is in most cases a time interval – the topic time.

The analysis provided in this section constitutes the basis for the various possible interpretations of Russian aspect. Moreover, it serves as a basis for a cross-linguistically uniform and systematic account of grammatical aspect.

cross-linguistic general invariant meaning; section 4.1

1

language-specific instantiation of (1); cross-linguistic invariant meaning of the terminality categories; section 4.1.1

2

language-specific markers to express (2); invariant meaning of the respective markers; sections 4.2-4.4

3

LEVELS OF INVARIANT MEANINGGRAMMATICAL ASPECT

GENERAL SEMANTICS

INTRA AD POST

Turkish Russian Turkish

-iyordu -mekteydi -irdi

+

-di ±

pf +

ipf ±

-miş +

-di ±

107Chapter 4: Semantics

4.1.1 The common denominator As has been pointed out in chapter 2, many aspect definitions suffer from their metaphoricity and from mistaking aspect interpretations for aspect semantics. This section will outline an analysis of aspect that avoids these shortcomings. This semantic analysis is based on the assumption that semantics cannot, and in fact need not, specify all the information required for interpretation, but merely has to specify the explicitly encoded information. This encoded information has to be figured out and needs to be represented in a way that enables pragmatic principles to apply (cf. chapter 5).

The analysis that will be developed here is mainly based on the approach put forward by Klein (1994, 1995, 2001, 2003). He proposes an analysis of aspect in time-relational terms, making use of the notions situation time ‘TSit’, time of utterance ‘TU’ and topic time ‘TT’ (cf. section 4.2.1). The latter notion expresses the time interval an assertion is made about5. Klein uses these notions in order to define the semantics of tense as well. This captures the fact that both aspect and tense are related to the concept of time, albeit in different ways (cf. Comrie 1976: ‘situation external time’ vs. ‘situation internal time’). Tense is characterised by the relation between TU and TT, aspect by the relation between TT and TSit. In order to account for the language-specific characteristics of grammatical aspect, the notion of TSit needs to be refined in terms of certain features aspect is sensitive for (see below).

According to Klein’s conception of finiteness, aspect can be regarded as one of the possible means to express this notion6. He provides the following approximate definition of finiteness: “[Finiteness] is the carrier of assertion – or, more generally speaking, of validity – in an utterance, and it relates the descriptive content of a sentence to its topic component, in particular to the time span about which an assertion is made.” (Klein 2001: 1)

The decisive characteristic of finiteness thus consists in that it establishes a relation between a sentence base and a topic component and thereby marks that an assertion is made. If a situation is identified, the assertion does not cover the whole time the situation may occupy, but only a specific time span – the topic time TT7. Besides the topic time there are also other topic components that may play a role with finiteness – Klein lists topic place, topic world or a topic entity. The topic component provides the specification of the situation about which the utterance says something. To recap so far, finiteness as the carrier of assertion/validity in an utterance relates the descriptive content of a sentence to its topic component, in particular to the time span an assertion is made about, the topic time. Figure 4.2 summarises the function of finiteness according to Klein’s analysis (Klein 2001: 20): A finite sentence consists of three components, namely a topic component (basically time or entity), a sentence base (the event denoted by the predicate) and a marking of how the sentence base relates to the topic entity. This relation may be established differently in different languages – among others, by means of grammatical aspect.

Lexical verb stems are the starting point for all verb forms. When integrated into some construction, such stems are turned into inflected forms. Inflected forms may be finite or non-finite, with both marked infinitives and bare stems being non-finite. Non-finite verb forms are FIN-linkable, in other words, they can be made finite. Finiteness carries at least two distinct

5 Klein takes the notion of topic time to be an interpretation of Reichenbach’s (1947) reference time R. As

will be shown in section 4.1.2, however, TT does not fully correspond to R. 6 Klein’s analysis applies also to languages that do not have the grammatical category of aspect. The respective

notions are then expressed by lexical means or implicitly by means context. Languages like Russian, Bulgarian, Turkish and English are special in that they have grammatical means to express these notions and relations.

7 This topic time may also be an unbounded interval, in which case it is more reasonably to speak of a topic entity about which the utterance says something (cf. especially section 4.4.2 on the Turkish ‘Aorist’).

108 Chapter 4: Semantics

meaning components: a tense component and the marking of an assertion. Finiteness marks that an assertion is made and restricts this assertion to the topic situation. Languages differ with respect to the realisation of the single components contributing to the relation specified in Figure 4.2. In Indo-European, finiteness and the topic world are brought together by verbal inflection expressing the category of mood (Klein 2001, 2003). The relation between finiteness and the topic time can be established by verbal inflections as well – in languages with grammatical aspect, that relation between the sentence base and the topic time is established by exactly this category (cf. footnote 6). More specifically, aspect establishes a relation between two intervals – the time interval the assertive part of the sentence occupies and the time span an assertion is made about. The crucial elements of that assertive part of the sentence base are the features encoded on the temporal tier – phases and boundaries.

Figure 4.2 The general picture presented in Figure 4.2 can thus be regarded as illustrating the cross-linguistic semantics of aspect, the invariant meaning basic to all instantiations of this grammatical category. Several comments are at order here. The relation between the sentence base and the topic component can be established in different ways which are systematically captured by the different terminality categories (cf. Figure 4.1). Aspect does not only specify the nature of this relation, but also which part(s) of the sentence base are selected for that relation – phases or boundaries. The topic component may as well be a topic world or a topic place, but for the purposes here, time and entity will be enough. The sentence base is expressed via the verb stem and its arguments, that is, by the Aktionsart basis. This Aktionsart basis occupies a certain time interval I(e) – the run time of the event – and is FIN-linkable. Aspect relates I(e) to the topic component, in the prototypical case to the topic time interval I(TT) which identifies the part of the situation that is being talked about and in this sense individuates it8. It is this relation between I(e) – TSit in Klein’s terms – and I(TT) that is expressed by the grammatical category of aspect. Below, I(e) or TSit, respectively, will be

8 This individuation is important, since in cases where the topic time is not bounded it cannot individuate a

specific part of the situation and, consequently, no event is being introduced into the discourse (cf. Hopper & Thompson 1984). The topic constituent then is a topic entity, and what is denoted by the predicate is rather a property of that topic entity than an event (cf. section 4.4.2 on the Turkish ‘Aorist’). Breu (2005) captures this function of aspect in terms of ‘status-aspect’ as opposed to ‘focus-aspect’ that operates in the way outlined here, namely as focusing on, or selecting, specific parts of the predicate.

UTTERANCE

TOPIC COMPONENT FIN SENTENCE BASE

topic time topic entity ASPECT

verb stem and arguments

time interval I(TT)

time interval I(e)

selection and relation

109Chapter 4: Semantics

refined, since aspect relates only specific parts of what is provided by the Aktionsart basis to the topic time.

All three components – TT, relation and sentence base – may be specified and modified implicitly by means of context, or explicitly by grammatical means. In Russian, the Aktionsart can be modified by means of prefixes that function as eventuality type modifiers (section 4.1.2). Bulgarian has grammatical means to specify the boundedness feature of the topic time, whereas in Russian this can be expressed only by means of adverbials or context. Turkish has one aspect marker that explicitly relates the sentence base not to a topic time but to a topic entity. All object languages dealt with here express the assertion-part of the FIN-component by means of grammatical aspect. The language-specific aspect semantics will be topic of sections 4.2-4.4.

What happens, however, if no event is asserted, as is the case with questions, modals or counterfactuals? Predicates do not always introduce events into the discourse (Hopper & Thompson 1984), but may as well be used in a characterising function. This is the case with semantically not prototypical verbs like stative or relational verbs, and with verbs that are not used in a prototypical way (e.g., as irrealis or in generic contexts). Klein argues that in cases like these the assertion is not about a topic time, but about some topic world or topic entity. The predicate does not individuate an event, but rather ascribes some property to a participant, in most cases the grammatical subject. The assertion of a topic entity instead of a topic time will prove important with certain readings of the Russian ipf aspect, and with the analysis of a the Turkish aspect marker -irdi (the ‘Aorist’, cf. section 4.4.2).

Before applying this general semantics onto aspect in Russian, Bulgarian and Turkish, the interaction of aspect and Aktionsart and their operational interface have to be clarified. As has been pointed out in chapter 2, it is important to strictly keep apart the notions of aspect and Aktionsart. Both contribute in specific ways to the interpretation of the finite (aspectualised) verb in a certain context. This has been discussed in several studies on aspect (e.g. Comrie 1976; Bickel 1996; C. Smith 1997; Breu 1996; Filip 1999; Johanson 2000) and has been referred to, among others, by the distinctions of situation aspect vs. viewpoint aspect (C. Smith 1997), situation external time vs. situation internal time (Comrie 1976), actional content vs. aspect (Johanson 2001) or event semantics vs. propositional semantics (Filip 2005; Carlson 2000). What all these approaches share is that they conceive the relation between aspect and Aktionsart in terms of an operator-operandum relationship (Bickel 1996). Aspect (the operator) acts on the Aktionsart (the operandum) in a specific way, namely by selecting certain elements encoded by the predicate. How these elements specify Klein’s notion of ‘sentence base’ will become clear in the remainder of this section.

In order to state the interaction of aspect and Aktionsart, verbs and verb phrases need to be classified according to the features aspect is sensitive to. These features – phases and boundaries – provide the input for aspect to work on. The output of this combination of aspect and Aktionsart is the specific relation I(e)-I(TT), which constitutes the frame for the contextually relevant interpretations9. The remainder of this section deals with the input specification mainly in terms of the selection theory of aspect proposed by Bickel (1996). It will be elaborated and connected to Moens & Steedman’s (1988) proposal of an event nucleus (chapter 2, section 2.2.2).

Bickel’s (1996) selectional theory generates Aktionsarten10 out of a small set of primitive elements, namely phases and boundaries. This enables a compositional structure of

9 Putting it this way already suggests that it is not necessary to list all the possible readings but that it is

important to figure out the basic configuration. This configuration provides a frame within which different readings can be constructed.

10 Remember that the term ‘Aktionsart’ is used to refer to a classification of predicates in terms of certain features. It is not used in the Slavistic sense as derivational categories of the verb referring to modes of action.

110 Chapter 4: Semantics

Aktionsarten, since these primitive elements may be added by composition with other elements or induced by pragmatic reasoning. For the application of aspect it does not matter where the phases and boundaries come from and whether they are inherent or actual temporal boundaries – as long as they are located within VP. Elements that are located outside VP, like frequentative, durative and other temporal adverbials, are not relevant for aspect11. Nevertheless, the difference between inherent and actual boundaries matters for interpretation.

Bickel’s approach consists of two components: an Aktionsarten theory, specifying the range of possible Aktionsarten, and a theory of aspect semantics. Both can be summarised as follows (Bickel 1996: 196/197): 4.1 Aktionsarten theory

(i) Time structure is represented by a temporal tier consisting of regular alternations of phases and boundaries.

(ii) Boundaries and phases are associated with specific constituents on other tiers in semantic structure.

4.2 Aspect theory

(i) Aspect semantics is defined by operators selecting a phase and/or one or more boundaries.

(ii) Selection of a boundary entails that the associated constituent and, if any, all subsequent or preceding phases begin or cease to apply at the moment of reference tR.

(iii) Selection of a phase entails that the associated constituent applies at tR.

(iv) Selection of an unassociated phase implies that the other tiers (within associated boundaries) are relevant without applying at tR.

Bickel’s analysis offers a multi-tiered approach comparable to that proposed by Jackendoff (1987). The basic idea is as follows: “[S]emantic structure is not represented in a linear propositional format. Rather, semantic constituents are distributed over different tiers, such as a thematic tier, which contains elements like GO, BE, AD, TO, ..., and an action tier, specifying actor-undergoer relationships. Aktionsart structure is specified by a tier of its own“ (Bickel 1996: 196)

Aktionsart is specified at the temporal tier, where phases and boundaries encoding the parts of the event nucleus are located at. Associated with the temporal tier is the causal tier (cf. Michaelis 2004) that specifies the arguments of the verb. Both levels are connected with each other by associating elements of the causal tier with elements of the temporal tier: predicates are associated with phases, arguments with boundaries. Whenever phases and/or boundaries are selected, the respective associated constituents also apply. This connection allows for a mapping between the two levels and serves to explain, e.g., the definiteness-effect of the pf aspect onto direct objects (esp. incremental theme arguments) in Russian12, and the lack of this effect in Turkish, which likewise lacks overt marking of definiteness. The fact that aspect choice does not have an effect on the definiteness of object-NPs in Turkish can be explained

11 In section 4.2.1 it will be argued that such adverbials may contribute to establishing the feature of

‘boundedness’. This feature is an aspectual notion, which is, however, independent of grammatical aspect. 12 The effects resulting form the multi-tiered approach will not be dealt within this thesis. For the interaction

of aspect and definiteness compare, e.g., Filip (1997, 1999); Späth (2004).

111Chapter 4: Semantics

by the fact that aspect in Turkish is not sensitive for boundaries and thus cannot affect associated constituents (cf. section 4.4).

The temporal structure and its representation on the temporal tier captures the patterns of stasis and change that are characteristic of every situation type (Michaelis 2004), i.e. its regions and points (Jackendoff 1987) or phases and boundaries (Bickel 1996). Different situation types (Aktionsarten) occupy different patterns of phases and boundaries and can be classified accordingly (section 4.1.2). From these patterns topological inferences may be drawn that play a crucial role in deriving the different interpretations. The temporal structure indicates an idealised causative event, or in Moens & Steedman’s (1988) terms, a contingency-based tripartite event-structure that constitutes the prototypical event (Croft 1991, 1998).

The core claim of a selectional theory like Bickel’s is the following: “[A]spect marking consists of ‘selecting’ phases and/or boundaries so as to make them accessible to truth-conditional evaluation at the moment of reference.” (Bickel 1996:197)

The assumption of ‘selection’ is facilitated by the fact that the theory of Aktionsarten comprises only elements that are also relevant for the aspect theory. Selection is to be understood as highlighting or focusing certain elements of the actional content of the predicate. As pointed out above, not all elements encoded by a verb are equally relevant for aspectual issues. The decisive units of the internal phase structure are the phases (ϕ) and boundaries (τ) encoded on the temporal tier. Essential for the conception of aspect as selecting certain elements provided by the Aktionsart is the ‘Aspectual Uniformity Hypothesis’, repeated below (cf. chapter 2, section 2.2.2): “Aspect and Aktionsart representations have the same format, viz. configurations of ϕ and/or τ, and this format is the same on all levels of meaning composition (lexical semantics, morphological derivation, syntactic composition, and pragmatic enhancement).” (Bickel 1996: 204)

This identical format in terms of configurations of phases and/or boundaries, allows for the interaction of aspect and Aktionsart. Moreover, these configurations contribute the encoded information to the interpretation process.

Bickel’s assumptions outlined in ( 4.1) and ( 4.2) make interesting predictions for the functioning of pf and ipf aspect in Russian. The pf aspect is defined as selecting a boundary, which has two possible entailments: the associated constituent is affected, or the subsequent/preceding phases begin or cease to apply at the moment of reference. The first entailment accounts for the cases where perfectivisation yields quantised predicates that affect the associated constituent (the incremental theme argument) as a whole, insofar as it gets a definite reading, compare ( 4.3a). The second entailment is valid for instances where the pf aspect does not yield a quantised predicate, which is the case with delimitatives and perduratives. In these cases, it is the phase encoded by the verb that ceases to apply, and what is affected is not an argument but rather the time at which the situation is asserted to hold ( 4.3b,c). This is also the case with lexically specified consequent states (Demjjanow 1998) that begin to apply with the selection of the boundary; compare ( 4.3d): 4.3 a. On napisal pis’mo.

he write:PST:pf letter. ‘He wrote the letter.’

b. On posidel v parke. he sit:PST:pf in park:LOC ‘He sat in the park (for a while).’

c. Ona počitala knigu. she read:PST:pf book:ACC

‘She read a/the book (for a while).’

112 Chapter 4: Semantics

d. On slomal ruku. he break:PST:pf arm:ACC ‘He has broken his arm.’ The assumptions presented above can be related to the event nucleus that represents a prototypical event in terms of a preparatory phase, a culmination point and a consequent state. Verbs encode events, or more generally eventualities13, and they encode them in terms of phases and boundaries corresponding to the three parts of the event nucleus. The culmination is encoded by a boundary (τ)14, the preparation process by a dynamic phase (ϕdyn) and the consequent state by a static phase (ϕstat). Bickel’s (1996) account does not distinguish between dynamic and static phases, but in a more recent paper (Bickel 2000), he points out that it might be necessary to do so. This distinction is not decisive for all languages. For the Russian aspect system it is not necessary, but there are languages that have morphological means to access both phases distinctly. This is the case in Turkish with the markers -iyordu and -mekteydi selecting dynamic phases and -miş selecting static phases (cf. section 4.4.1). Note that a lexically static phase is different from a stative eventuality. Static phases are part of non-stative events and presuppose some prior culmination point. Stative eventualities are not part of an event nucleus (Moens & Steedman 1988: 23; cf. chapter 2, section 2.2.2).

Predicates encode parts or all of the event nucleus, and aspect operators select one of these parts. Marked members of aspectual oppositions are explicitly marked for a specific selection, while the unmarked ones are agnostic with respect to that, and, depending on the context, may deny that selection or simply not exclude it. The cross-linguistic terminality categories established by Johanson (1971; 1996; 2000; cf. chapter 2, section 2.1) relate to the event nucleus insofar as the existence of exactly these three terminality categories is predicted and justified by it:

4.4 a. INTRATERMINALITY

envisages the event within its course relates to the preparatory phase/dynamic phase

b. ADTERMINALITY envisages the event in the attainment of its crucial limit relates to the culmination point/boundary

c. POSTTERMINALITY envisages the event after the transgression of the crucial limit relates to the consequent state/static phase.

These terminality categories subsume language-specific aspect systems that are sensitive to one and the same part of the event nucleus. The language-specific aspectual categories are instantiations of the terminality categories: aspect in Russian is an instance of adterminality, the progressive aspect in English as well as the Turkish markers -iyordu and -mekteydi instantiate intraterminality; Turkish -miş is an instantiation of postterminality.

These assumptions, that will be basic to the semantic analysis of aspect developed in this chapter, are summarised in Figure 4.3. Note that this figure only shows the marked members of the respective categories. The fact that the unmarked members are not shown in Figure 4.3

13 The term ‘eventuality’ is used to refer to all encoded states of affairs. Eventualities may be instantiated by

different verb class or VP denotations that can be classified in terms of states, processes, events or in terms of the Vendlerian classes of states, activities, accomplishments and achievements. Another possibility of referring to these denotations is by means of specifying the feature combinations they are composed of (cf. section 4.1.2).

14 For the notion of ‘boundary’ cf. section 4.1.2.

113Chapter 4: Semantics

does not imply that they are semantically vacuous. Their semantics is less specific than that of the marked members, but it still can be determined in a definite way. As will be shown in chapter 5, their semantics can be derived via scalar implicatures based on Grice’s first quantity maxim (‘Make your contribution as informative as is required/possible.’) and Levinson’s (2000) corresponding Q-heuristic (‘What isn’t said, isn’t.’). Thus the markedness relations are crucial in accounting for the semantics of the language-specific aspect markers, especially for the unmarked ones.

Figure 4.3 The tripartite event-structure assumed in Figure 4.3 is taken from Moens & Steedman’s (1988) proposal, but deviates slightly with respect to the notion of ‘consequent state’ and ‘preparation phase’. Both notions are understood in a narrower sense than proposed by Moens & Steedman. They take most consequences of events to be independent of the specific time at which the event itself occurred. Thus, their notion of consequent state includes any further events “that are in the same sequence of contingently related events as the culmination” (1988: 18). Here, however, the notion of consequent state pertains only to lexically encoded consequent states (cf. the notion of lexikalisch ausgezeichneter Nachzustand ‘lexically specified consequent state’ in Demjjanow 1998), which are part of the encoded event (as is the case with inchoative and inceptive events), or may be added to it (by means of accommodation of context, cf. chapter 5). The same holds for the preparatory process. In Moens & Steedman’s framework any process that may potentially be located in the prephase of the actual event is subsumed under the notion of preparation phase. Here, the notion of preparatory process is understood in the narrower sense, namely as the lexically indicated preparation phase yielding to the culmination point15. Accordingly, the notions ‘preparatory process’ and ‘consequent state’ will be used to refer only to lexically indicated processes and states that are separated by a culmination point.

The event nucleus characterises prototypical events. A prototypical, transitive event can be understood as consisting of these three components. The tripartite structure of the nucleus defines the maximum of event-components that can be encoded by a predicate. Any event can be coerced into the direction of this prototypical event by adding the missing elements – provided context and world knowledge allow for that. The following examples illustrate the basic idea:

15 Singh & Singh (1997) refer to this narrower conception as ‘preliminary process’, as opposed th the wider

conception of Moens & Steedman’s ‘preparatory process’.

τ ϕstat ϕdyn

-ing -iyor(du) pf -miş

4. terminality category INTRA AD POST

1. event structure

2. temporal tier

3. aspect operators

consequentpreparation culmination point

114 Chapter 4: Semantics

4.5 a. He opened the window.

b. He was opening the window.

c. He has opened the window.

d. He was reaching the border. The verb open as a culminated process in Moens & Steedman’s terms is an instance of a prototypical verb: it specifies the preparatory process of opening, the change of state from the window not being open to the window being open, and the consequent state of the window being open (note that this consequent state is lexically specified). The English aspect forms of the verb highlight or select the different parts of that event. The progressive form in ( 4.5b) highlights the dynamic phase of that event, i.e. the process of opening it. The simple form in ( 4.5a) is agnostic towards that selection, it is the unmarked member of that opposition and may in certain contexts be interpreted as selecting the boundary. The present perfect in ( 4.5c) indicates that this boundary is transgressed and asserts the consequent state to hold at a certain interval of time. ( 4.5d) is an example illustrating the addition of a preparation phase by using the progressive marker16.

Turkish has an aspect marker for the selection of the dynamic phase17 as well as a marker that explicitly selects the static phase:

4.6 a. Bu kitabı okuyormuydun?

this book:ACC read:iyor:Q:PST:2Sg ‘Were you engaged in reading this book?’

b. Gözlüklerimi kaybetmişim. glasses:1SG:ACC lose:miş:1SG ‘I have lost my glasses.’ (= they are gone)

In ( 4.6a), the dynamic phase of the predicate is selected by -iyordu. This dynamic phase is actually the only part of the event nucleus that is specified by the predicate kitabı okumak (‘read the book’). ( 4.6b) gives an example of the static-phase selecting -miş. This morpheme selects the consequent state and backgrounds the rest of the event which is strongly presupposed but not asserted to hold. This explains the evidential readings -miş may give rise to. The evidentiality associated with this morpheme can be regarded as deriving from its aspectual function.

The examples in ( 4.7) illustrate the Russian case. The pf aspect selects a boundary and, if present, the lexicalised consequent state, compare ( 4.7a), and therefore gives rise to readings of ‘completedness’. The ipf aspect as the unmarked member of the Russian aspect opposition is agnostic with respect to the feature ‘boundary-selection’; compare ( 4.7b). This selection is not explicitly denied, which explains the so-called ‘general-factual’ uses of the ipf aspect and is the reason for the non-validity of characterising the ipf aspect in terms of ‘ongoing activity’ or ‘non-completedness’. 4.7 a. On napisal pis’mo.

he write:PST:pf letter ‘He wrote the letter.’

16 A detailed account of these coercion phenomena can be found in Moens (1987) and Moens & Steedman

(1988), cf. also Kamp & Reyle (1993). 17 In fact, this is a slight oversimplification, as there are some differences between the English progressive

marker and Turkish -iyordu. Compare section 4.4 for a more detailed discussion, and chapter 5, section 5.2.3 for a pragmatic explanation.

115Chapter 4: Semantics

b. On pisal pis’mo. he write:PST:ipf letter ‘He was writing the/a18 letter.’ So far, aspect has only been characterised as selecting certain elements of the semantic representation of the verb, i.e. as focussing certain parts of the event or state talked about. There is no connection to the time axis yet. This is achieved by tense and the position of the observer, or the točka otsčeta (TO) as Padučeva (1996) calls it (cf. section 4.1.2).

Note that with the analysis presented here, there are three cross-linguistic types of aspect categories that are predicted by the event nucleus and the selectional analysis of aspect. This is different from most other analyses that merely postulate two categories – perfective and imperfective. Moreover, these analyses mostly ignore the fact that the notions ‘perfetive’ and ‘imperfective’ are not applicable to all aspect languages in a uniform way. As will be evident by the end of this chapter, the terminality categories in fact may be related to two categories PERFECTIVE and IMPERFECTIVE, which are abstract meta-categories that are instantiated differently in different languages. These differences will be revealed by analysing aspect in Russian, Bulgarian and Turkish.

To sum up, this section has outlined the general semantics of aspect in terms of the relation between the assertive part of the sentence base and a topic component. This general assumption has been fleshed out by Bickel’s selection theory and has been related to the event nucleus. This allowed for a specification of the interaction between event structure, temporal tier units, aspect operators and terminality categories, which is the basis for the semantic analysis of Russian, Bulgarian and Turkish aspect developed in this chapter. The semantics of aspect sketched in this section specifies the procedural information encoded by grammatical aspect. Aspect does not encode a certain concept, but specifies the way the concept activated by the Aktionsart basis is integrated into the construction of an interpretation for the respective utterance. Aspect imposes specific requirements onto its context – the predicate it applies to – and specifies the output. The logical entry for an aspect marker has to specifiy the conditions for it to apply and the output results (see below, section 4.2).

The next section deals with the different components that contribute to the specification of aspect semantics in more detail.

4.1.2 The ingredients So far, the following factors have been specified as contributing to the interpretation of aspect: the characterisation of the Aktionsart in terms of phases and boundaries and the semantics of aspect as provided by the selection-theoretic and time-relational analysis. Before these components will be applied to aspect in Russian, Bulgarian and Turkish, they need to be specified in more detail in this section. This specification will be illustrated mainly with respect to the Russian system.

The means to represent the semantics of aspect will be topological structures, namely temporal intervals (Desclés 1993; Desclés & Guentchéva 1990a, b; Guentchéva 1989; Vazov 1998). For aspect, two kinds of intervals play a role: intervals covering the event time specified by the Aktionsart and intervals covering the topic time. Topic time intervals may impose external boundaries onto the asserted situation. These temporal boundaries are not to be confused with inherent boundaries specified by the predicate.

An interval can be defined as follows:

18 The influence of aspect on the expression of definiteness in Russian will not be dealt with here.

116 Chapter 4: Semantics

“An interval is an oriented set of contiguous instants (which are elements of a continuous line) delimited to the left and to the right by two boundaries which separate the interior (instants between the boundaries) and the exterior (instants which are not between the boundaries) of the interval”. (Desclés & Guentchéva 1990: 240b)

This can be illustrated as follows:

Figure 4.4 According to Passonneau (1987, 1988), intervals can be characterised in terms of two features: kinesis and boundedness. Kinesis pertains to the internal structure of the interval (or informally, to whether something is happening within that interval). This feature captures the difference between stative and dynamic19 intervals: change occurs within a dynamic interval with respect to the relevant predication, but not within a stative interval (Passonneau 1987: 19). The kinesis feature of the topic time is thus specified in terms of the phases and/or boundaries that are related to it by aspect.

Boundedness pertains to the way an interval is located in time with respect to other intervals. Intervals may be bounded, unbounded or unspecified for boundedness. Except for intervals comprising propositions of eternal truth that arguably hold over an unbounded period of time, all intervals are bounded in time. Languages differ, however, with respect to their capability of explicitly marking the beginnings and ends of such intervals. Decisive for the analysis of aspect are the boundedness characteristics of the topic time interval I(TT) and of the event time interval I(e). Languages that don’t have means to grammatically encode the boundedness characteristics of I(TT) (like Russian or English) specify it by lexical means, by means of context or by the discourse mode. Different boundedness characteristics of I(TT) may be plausible in different contexts. Other languages, such as Bulgarian, have grammatical means to specify the boundedness of I(TT).

Thus, a first characterisation of intervals pertains to their boundedness or non-boundedness. In the former case, the interval extends over a limited part of the referential space (Figure 4.5a), in the latter case, the interval extends from minus infinite to plus infinite (Figure 4.5b):

Figure 4.5

A second characterisation of intervals pertains to bounded intervals, which may be of three types: open, closed and semi-open (to the left or to the right). If an interval is open (Io), then both of its boundaries are open and the first and the last of the points (x) of that interval are excluded, i.e. the boundaries are not part of that interval:

19 Passonneau speaks of stative and active intervals. Here, the notion of dynamic interval is preferred.

exterior exterior interior

left bound right bound

- ∞ + ∞

b) non-bounded interval

- ∞ + ∞

a) bounded interval

117Chapter 4: Semantics

Figure 4.6 A closed interval (Ic) includes the first and the last of its points (x), i.e. the boundaries are crucially part of that interval:

Figure 4.7 Based on these definitions, the two semi-open intervals can be defined as follows: 4.8 a. semi-open to the left: (I) = {x, a < x ≤ b}

b. semi-open to the right: (I) = {x, a ≤ x < b}

The distinctions open vs. closed and bounded vs. non-bounded strictly have to be kept apart. An unbounded interval does not hold over some specific segment of time. It does not have a finite duration as it pertains from -∞ to +∞. An interval is bounded if it holds over some specific segment of time, i.e. if it is measurable within some finite duration. If such an interval is closed, the boundaries are part of the interval, if it is open, the boundaries are excluded from the interval and belong to its exterior. This distinction between open-bounded and closed-bounded intervals permits intervals with external boundaries (cf. Pianesi & Varzi 1996; for a discussion of the notion of boundary see also B. Smith 1997). Note that open, semi-open and closed intervals are all specified by being bounded.

The different interval characteristics are summarised in ( 4.9): 4.9 TYPOLOGY OF INTERVALS

a. unbounded intervals are not segmented; they do not hold over a specific segment of time

b. open intervals hold WITHIN a limited segment of time; the boundaries are not part of the intervals

c. semi-open intervals hold WITHIN a limited segment of time; they have one boundary that is part of the interval, and one boundary that is excluded

d. closed intervals hold OVER a limited segment of time; the boundaries are part of the interval

Io = {x, a < x < b} Io

] a b [

Ic= {x, a ≤ x ≤ b} Ic

[ a b ]

118 Chapter 4: Semantics

By means of intervals, situations that are referred to by linguistic means can be individuated20 and localised on the temporal axis of the speaker/hearer21. They may then enter into relations of intersection and union.

To make use of intervals for the analysis of aspect, the elements that may be part of the intervals as well as the determination of the boundaries have to be specified. The crucial elements of the theory so far are the temporal relation between a topic time and some part of the event, and the specification of that part of the event that is selected by the aspectual operator. This suggests that aspect has a double job to do: select the respective part of the sentence base encoded by the predicate and relate it to the topic time as the time the predication is asserted to be valid. The topic time is characterised by the parts of the event that are asserted – phases and boundaries – and it is this topic time that introduces external boundaries onto the situation. This interval has to be anchored on the temporal axis of the speaker/hearer, which is achieved by tense and by means of an anchoring point, namely the točka otsčeta TO, the standpoint of the observer (Padučeva 1996).

The analysis of aspect in terms of the relation between certain parts of the event and the topic time that still has to be related to the temporal axis in order to enter into tense temporal relations, captures Klein’s assumption that aspect and tense can be analysed along the same lines but are different concepts of how to deal with ‘time’. Aspect as the situation internal time (Comrie 1976) relates to temporal relations pertaining to the running time of the event, tense as situation external time (Comrie 1976) pertains to the temporal relations this asserted part of the event enters into on the time axis of the speaker/hearer.

Figure 4.8 illustrates the two different temporal relations characterising aspect and tense. Aspect is defined in terms of a relation between the run time of the event and the topic time, whereas on the temporal axis, tense-temporal relations are specified in terms of the relation between the topic time characterised in the way outlined above and the time of utterance TU:

Figure 4.8 The following examples illustrate the assumptions of Figure 4.8: 4.10 a. On medlenno dvigalsja.

he slowly move:PST:ipf ‘He was moving slowly.’

20 The situation is slightly more complicated for non-bounded intervals that do not refer to a specific segment of

time. In this case it is implausible to speak of ‘individuation’ and ‘localisation’. Localisation is achieved indirectly via the temporal localisation of the participants involved in the situation (cf. Klein 2001; cf. section 4.1.1.).

21 Or on another temporal axis, as is the case with narrations that introduce a temporal axis (or even more than one) that is different from that of the speaker/hearer.

| |

[ [ ϕdyn

I(e)

I(TT)

temporal axis of the predication:

ASPECT

TA temporal axis of the

speaker: TENSE

TO

| | [ [ TSit / TO TU / TO

119Chapter 4: Semantics

b. Da22, ja čitala ėtot roman. Yes, I read:PST:ipf this novel. ‘Yes, I have read this novel.’ Example ( 4.10a) illustrates the case where TO is simultaneous to the topic time interval, which is characterised by a dynamic phase and open boundaries. In ( 4.10b), TO is retrospective to the topic time interval and coincides with the time of utterance TU. In this case the topic time interval is closed-bounded and comprises all phases and boundaries encoded by the predicate – here a dynamic phase and a boundary. In both cases the temporal location of I(TT) and the resulting relation between I(TT) and TU is the same: I(TT) precedes TU. The specification of I(TT) in terms of boundedness and kinesis will be dealt with in more detail in section 4.2.2.

Note that TU, TSit and TO are similar to Reichenbach’s S, E and R, but differ in one important respect: they are intervals, not points23. The notion of I(TT) is not related to Reichenbach’s system, as this interval is not located on the temporal axis. This is contra Klein, who regards TT as a further development of R (cf. footnote 5). Since aspect and tense pertain to different layers, the topic time as specified by aspect has to be related to the temporal axis. This anchoring is achieved by means of tense and via the position of the observer (TO), i.e. the standpoint from which the speaker/hearer refers to the event. There are two possibilities for TO: it may be synchronous or retrospective to TSit. As indicated in Figure 4.8, TO may coincide with TU or with the situation time TSit (note, that TSit here is understood as the specific part of the situation that is element of the topic time). With present tense predicates, TU and TSit coincide and TO has to be synchronous to TSit. With past tense predications, however, TO may be both synchronous or retrospective, mainly depending on the discourse mode (režim interpretacija, Padučeva 1996): if TO coincides with TU, as is the case mainly in dialogues (rečevoj režim), it is retrospective, but it may as well be synchronous, as is the case predominantly for narration (narrativnyj režim). These different positions of TO capture the intuition behind the notion of ‘perspective from inside’ and ‘perspective from outside’, respectively, in dealing with aspect in Russian.

In Figure 4.8, the boundaries of the topic time interval are left unspecified. This is due to the fact that the exact nature of these boundaries influences the plausibility or even possibility of certain TO-positions: closed boundedness of I(TT) points out a retrospective TO; and a retrospective TO means a closed bounded I(TT), irrespective of aspect. Therefore, TO cannot be ascribed to the semantics of aspect alone, but rather depends on factors like the discourse mode and temporal adverbials (Padučeva 1996: 13-15): “Točka otsčeta založena ne tol’ko v semantiku vida, no i v semantiku obstojatel’stvo vremeni.” (Padučeva 1996: 169) [TO is built in not only in the semantics of aspect, but also in the semantics of temporal adverbials.]

The following examples illustrate the different positions of the observer (TO) for the Russian ipf aspect:

4.11 a. Kogda ja vošel, moj brat čital knigu.

when I come.in:PST:pf, my brother read:PST:ipf book:Acc ‘When I came in, my brother was reading a book.’

22 This particle indicate the discourse mode of ‘dialogue’ which is one of the indicators of a retrospective TO

(Padučeva 1996). 23 Reichenbach (1947: 290) has included a rather vague notion of ‘extension’ in his system. This does not

refer to E, S or R, but seems to illustrate some aspectual feature. As Reichenbach is not explicitly concerned with aspect, this notion remains somewhat unclear (cf. chapter 2, section 2.1.1).

120 Chapter 4: Semantics

Here, TO is synchronous to the topic time inverval. The topic time interval is included in the event time interval, as is indicated by kogda (‘when’). b. On govorit po-turecki. he speak:PRS:ipf Turkish ‘He can speak Turkish.’ In ( 4.11b), TO is again synchronous to the topic time, but I(TT) comprises all temporal tier units provided by the Aktionsart. This relation may be indicated, e.g., by discourse mode specified by the broader context in which this sentence is embedded. By this relation no specific topic time is characterised, but rather a topic entity, here: on (‘he’). In this case, aspect selects the actional content provided by the verb and relates it to the topic entity. Localisation on the time axis is achieved via this topic entity. Therefore, uttering this sentence in the past tense would implicate either that the person in question has forgotten to speak Turkish, or is not alive any more.

c. Da, ja čitala roman ‘Master i Margarita’.

yes, I read:PST:ipf novel ‘Master i Margarita’. Yes, I have read the novel ‘Master i Margarita’. The utterance in ( 4.11c) conveys the mere statement of a fact. The topic time includes all phases and boundaries provided by the Aktionsart; TO is retrospective. This is specified here by the discourse mode, which is ‘dialogue’ in this case (indicated by da, ‘yes’). Dialogue indicates that the standpoint of the speaker is located at the time of utterance. Since the location of the situation is in the past, this suggests that TO is retrospective to it. With the pf aspect, TO has to be retrospective (or prospective in case of future reference) to I(TT).

The difference between I(TT) and TO, and the difference between the synchronous and the restrospective TO, is illustrated in ( 4.12) and ( 4.13): 4.12 Konečno, ja prygala s parašjutom. of course, I jump:PST:ipf with parachute:INST ‘Of course, I have jumped with a parachute.’

4.13 On medlenno dvigalsja. he slowly move:PST:ipf ‘He was moving slowly.’

prygala s parašjutom

[ ] TT

] TO [ [ [

TU

I(e)

I(TT)

time axis

121Chapter 4: Semantics

Examples ( 4.12) and ( 4.13) illustrate that it is the topic time that is validated at TO, not the event itself. TO may be retrospective with respect to I(TT) as in ( 4.12), where it also coincides with TU, or it may be synchronous to it, as in ( 4.13). Thus, the decisive point for aspect interpretation is the topic time, characterised by its boundedness and kinesis features, and evaluated at TO. TO is the standpoint to which the speaker/hearer relates the event mediated by I(TT). The mapping of I(TT) onto the time axis in a certain relation to TU (before, simultaneous, after) is achieved by means of tense, compare, e.g., Filip (2005: 146): “A temporal operator maps the event described by a given predicate onto the time axis via its temporal location in relation to the utterance time.”

It is indeed necessary to have both notions, as they refer to different things – TO indicates the position from which the speaker refers to I(TT). Remember the dual character of Reichenbach’s reference time R that has been pointed out in chapter 2, section 2.1.1: On the one hand, it is part of the system of temporal relations, on the other hand, it specifies the aspectual perspective. In the system developed here, I(TT) corresponds to that side of R that is part of the system of temporal relations (as it is the relation between TT and TU that specifies tense), TO corresponds to that part of R that specifies the aspectual perspective (as TO determines the standpoint of the observer). Moreover, defining the asserted part of the event in terms of the relation between I(e) and I(TT) is different from locating the standpoint of the speaker on his time axis. One and the same relation of a sentence base to a topic component may be referred to from different standpoints, the only constraint being that closed-bounded topic time intervals require TO to be not synchronous to I(TT).

Grønn (2003: 47, 80), whose analysis is also based on the notion of assertion time (which he uses equivalently to TT24), does not need TO for his analysis. The assertion time is the speaker’s focus and makes visible the whole event or only part of it. In the former case TT is a ‘big assertion time’ whereas in the latter case it is ‘small’. Grønn concludes: “The effect of aspect is to widen or narrow down the assertion time relative to the event by establishing certain inclusion relations.” (Grønn 2003: 47)

According to Grønn, the decisive factor for distinguishing between different semantic configurations for the ipf aspect consists in the size of TT, which can be big or small. He distinguishes three main groups of ipf readings: the factual, the iterative-habitual and the processual one. Both the factual and iterative-habitual reading of ipf are characterised by a big TT, the processual reading by a small one. Thus, Grønn’s habitual-iterative and the factual reading cannot be distinguished by means of TT. However, both groups can be distinguished by the notion of TO. The factual group is characterised by a retrospective, the habitual group

24 Klein makes use of both notions, replacing the one by the other (assertion time TAst in Klein 1995, topic

time TT in Klein 1994, 2001).

] [ TO

I(e)

I(TT)

time axis

medlenno dvigalsja

] [ TT

[ [

TU

122 Chapter 4: Semantics

by a synchronous TO25. A synchronous TO is also characteristic of the actual-processual reading, where the difference to the habitual group consists in the explicit selection of the dynamic phase. A retrospective TO is also characteristic of the pf aspect, but pf explicitly selects the boundary, whereas for the general-factual ipf there is no such requirement; it asserts the whole event without focusing on some internal part.

The different TO’s are predicted by different inclusion-relations that may be established for a big assertion time (cf. section 4.2.1): 4.14 a. TT ⊃ ϕdyn τ ϕstat

b. TT = ϕdyn τ ϕstat If the whole event is included in the topic time, the standpoint of the observer cannot be within the event-time interval, it has to be retrospective. If both intervals are ‘simultaneous’ (Klein’s 1995 term), the standpoint of the speaker has to be synchronous to these intervals.

To recap, the semantics of aspect can be captured in time-relational terms as the relation between a sentence base with a topic component. More specifically, aspect establishes a relation between two intervals: I(e), the run time interval of the event-part selected by aspect, and the time for which the assertion is made, the topic time interval I(TT)26.

Central to the analysis of aspect is thus a mapping from an event-interval onto a time-interval. This is possible assuming that events happen in time (even states cannot escape from time) and that time proceeds as unidirectional path. Events are related to intervals of time, which results in a homomorphic mapping from events (i.e. running times of events) onto time intervals (e.g. Späth 2004: 191f): 4.15 Homomorphism between events and times

∀ei∀ej[ei ⊆ ej → time(ei) ⊆ time(ej)] As events may be built up out of different parts, this homomorphism holds for these parts as well. Consequently, an event is embedded in a time interval which then results in an event time (Späth 2004: 197f). This relation is not as immediate as the statement in ( 4.15) suggests: the event is mapped onto a topic time, and the resulting intersection of both intervals is located on the temporal tier as situation time TSit (cf. Figure 4.8). The situation time stands in a certain relation to some other relevant time span located on the temporal axis (among others, the event of the speaking/hearing process), where the relation between both may be anterior, posterior or simultaneous with that respective interval.

In addition to that mapping, the set of possible Aktionsarten needs to be clarified. Bickel’s (1996) minimal parts – phases and boundaries – and their combination will be taken as the defining criteria for different Aktionsarten. This classification incorporates two kinds of phases – static and dynamic – and a transition bound. Bickel refers to this transition bound as ‘boundary’, but the notion of boundary needs to be understood in a wider sense. It does not only refer to inherent transition bounds (i.e. culmination points), but also to external, actual temporal boundaries. Although the pf aspect in Russian is sensitive to both and does not differentiate between them, it would be wrong to take this as argument for conflating these two kinds of boundary. As was indicated in chapter 2, section 2.2.1, inherent and temporal boundaries have to be kept apart, as both can be accessed independently, even with different

25 The situation is in fact slightly more complicated (cf. section 4.2.2), but for the argument here the details

do not matter. 26 Henceforth, both intervals will also be referred to simply as e and TT.

123Chapter 4: Semantics

morphological markers (e.g. in Bulgarian). For a classification of verbs and verb phrases into Aktionsarten, the boundaries specified within VP are crucial27.

First, phase and boundaries need to be defined, and the distinction of dynamic and static phases needs to be motivated. Phases are constitutive of internally homogeneous Aktionsarten, and they may also be part of complex ones. Static and dynamic phases are both homogeneous, but differ in their exact homogeneity characteristics. Static phases are homogenous down to points, whereas dynamic phases are homogeneous down to small parts, i.e. intervals. One example to illustrate this is the verb run, that encodes a dynamic phase. It is not homogeneous down to points, because one point is not enough to define the respective event as a running-event. A bigger unit than a point is needed to characterise it appropriately. Situations consisting only of a static phase belong to the Aktionsart of states, whereas situations consisting of dynamic phases are activities. Static and dynamic phases can be defined in terms of the intervals they occupy in time: 4.16 a. static phase If P consists of a static phase, then P(x) is true in an interval i, if P(x) is true at any subinterval of zero-length (i.e. at points in time).

b. dynamic phase If P consists of a dynamic phase, and P(x) is true in an interval i, then P(x) is

true at any subinterval longer than 0. As is evident from ( 4.16), both kinds of phases exhibit subinterval-property but differ in their respective minimal parts. Furthermore, both are characterised by the part-of-relation of their constitutive parts. However, the part-of-relation with dynamic phases is a special case – it is a ‘stage-of-relation’ (Landman 1992; Rothstein 1999, 2004), meaning that there are actions that are a necessary part of the respective situation but do not constitute it by themselves. The notion of event stage is closely related to the development of the event. Landman (1992: 23f) defines event stages as follows (‘i’, ‘j’ = intervals; ‘w’ = world): 4.17 Let e be an event that goes on at i in w. Let f be an event that goes on at j in w, where i

is a subinterval of j. f is a continuation of e iff e is a stage of f. Let j be a non-final interval. f stops at j in w iff no event of which f is a stage goes on beyond i in w (i.e., at a later ending interval). This definition captures the intuition that “we cannot say that when an event stops in a world, there is no bigger event of which it is part in that world, but we can say that when it stops, there is no bigger event in the world of which it is a stage”. (Landman 1992: 23)

An event is a stage of another event, if the second one is a more developed version of the first. In order to be called a more developed version of another, a part of an event has to be big enough and share enough with the event it is a part of. This intuition will prove important in distinguishing between accomplished and achieved processes (Guentchéva 1989) in Bulgarian. Accomplished processes are terminated without being completed, i.e. they have reached a certain temporal boundary without having reached the inherent one. Such accomplished processes might be part of a bigger event, but not a stage – the explicit temporal

27 These may as well be temporal boundaries imposed on the run time of the event. Russian and Bulgarian

possess specific verbal prefixes that introduce exactly this kind of boundary, cf. section 4.2.1.

124 Chapter 4: Semantics

bounding excludes the reaching of the inherent boundary, they cannot develop further into a bigger event, cf. ( 4.18)28: 4.18 Tja pisa pismoto dva časa. she write:A:ipf letter:DEF two hours [Desclés & Guentchéva 1990b: 238] ‘She wrote the letter for two hours.’ This sentence can be paraphrased as follows29: there is an eventuality ev which is subpart of an eventuality e’ such that in the default case e’ is an eventuality of her writing a letter. In the real world there is no e’’ of which ev is a proper subpart and which is a subpart of e’. Concretely, the sentence base in ( 4.18) ‘write a letter’ would in the default case develop into the assertion that the letter is written. However, tja pisa pismoto is not part of a bigger event that would develop into that bigger eventuality denoted by ‘write a letter’. This default is explicitly cancelled by use of the Aorist; ‘write a letter’ is no continuation of pisa pismoto.

( 4.19) summarises this reasoning process: 4.19 a. pisa pismoto: ev

b. ‘write a letter’: e’

c. default case: ev → e’

d. Aorist: no ev ⊆ e’’ ⊆ e’, i.e. the event denoted by tja pisa pismoto is not part of an e’’ that would develop into e’ e. therefore: tja pisa pismoto is no stage, but possibly a part of ‘write a letter’ Above, the stage-of-relation has been illustrated with eventualities denoted by whole sentences. At the level of verb classes, this special part-of-relation pertains only to predicates encoding a dynamic phase, as those are the only predicates that have smallest parts, namely intervals, that are big enough to share something with the event. Stages of the event are more like the event itself than any arbitrary parts and share certain characteristics with their respective event. Of special importance are ‘process stages’ of an event, defined as follows (Landman 1992: 24): 4.20 Let f be an activity or accomplishment and e a stage of f. e is a process stage of f iff e has the same process characteristics as f. Accomplishments do not exclusively consist of process stages. They may as well start with a planning stage before the process starts, and they may have pause stages. This is subsumed by Landman’s (1992: 24) ‘Activity Postulate’: 4.21 The stages of accomplishment or activity e are either process, planning, or pause

stages. Planning stages precede all process stages. Pause stages are flanked left and right by process stages30.

The part-of relations that constitute static and dynamic phases can be defined as follows:

28 Section 4.3.2 will deal with this phenomenon in more detail. 29 Koenig & Muansuwan 2001 are concerned with ‘semi-perfectivity’, as they call this phenomenon, in Thai

(cf. section 4.3.2). 30 This could possibly explain progressive achievements (<τ>-predicates): the progressive selects the planning

stages and not the process stages as is the case with progressive accomplishments, i.e. <ϕdyn, τ>-predicates. For a detailed analysis of progressive achievements and the difference to progressive accomplishments see Rothstein (2004).

125Chapter 4: Semantics

4.22 a. static phase: part-of relation ∀x∀y ∈ I]x y[ [x ⊆ y ↔ (x ∪ y = y)]

b. dynamic phase: stage-of relation ∀x∀y∀z∀i ∈ I]x z[ [i]x y[ ⊆ i]x z[ ↔ (i]x y[ ∪ i]y z[ = i]x z[)]

This distinction between static and dynamic phases is important for capturing the distinction between ingressive-static and ingressive-phasal verbs. It is also important for languages that have aspect operators being sensitive to this distinction: the English progressive requires a stage-of relation as input, just like the Turkish -mekteydi. The Russian ipf aspect is neutral to this distinction, the differences surface with its interpretation in the respective input context. Turkish has an aspect marker that requires a static phase as input. This marker (-miş) introduces a TT-interval that is open on both sides and included in the static phase. The respective static phase for this marker to apply requires a preceding culmination point (inherent boundary). Static phases of this kind are always part of complex Aktionsarten, that is, they characterise consequent states. Therefore, static phases are different from lexically stative verbs that do not specify a consequent state. Since -miş requires as input a static phase characterising a consequent state, it cannot be applied to lexically stative verbs31. In this respect, -miş differs from the Russian ipf. The Turkish form is an instance of postterminality. It asserts a consequent state without taking into account the culmination leading to this state.

Taking into account the difference between dynamic and static phases, the set of possible Aktionsarten A is defined as the following set of ordered tupels (cf. chapter 2, section 2.2.2): 4.23 A = {<ϕdyn>, <ϕstat>, <τ>, <ϕdyn, τ>, <τ, ϕdyn, τ>, <ϕdyn, τ, ϕstat>32, <τ, ϕdyn>, <τ, ϕstat>} This raises the question, whether <τ>-predicates can reasonably be assumed at all, and if yes, what makes them different from Aktionsarten consisting of a boundary plus a phase. A boundary in the domain of events indicates a change-of-state, and as such requires some phase before or/and after it to be recognised as such. To put it differently, boundaries can be regarded as being unsaturated, as they need to be complemented in a certain way (B. Smith 1997: 8). In the case of Aktionsarten, this saturation is achieved when a boundary is preceded and/or followed by a phase, but not with <τ> alone. If a preceding or following phase for that boundary has to be assumed, what makes this Aktionsart different from the others? Rothstein (2004: 188) puts the difference between <τ>-Aktionsarten and those accompanied by some phase as follows: the change of state with <τ> consists of two instants (the last at which φ holds and the first at which ¬φ holds), whereas with <ϕdyn, τ>, it is an extended change that takes place at an interval held together by an incremental chain. With the latter Aktionsart, the phase leading up to the change-of-state is lexically specified, whereas in case of <τ> it is given by world-knowledge (this presumably holds for the consequent state as well).

An extended change of state may be initial or final. With an initial change of state, two cases have to be distinguished: either it is followed by a dynamic <τ, ϕdyn> or by a static phase <τ, ϕstat>. The latter possibility is rather commonly found in Uralic and Turkic languages. A final change-of-state has to be preceded by a dynamic phase (<ϕdyn, τ>), as only dynamic phases allow for developments and therefore allow for a possible culmination. And finally, a change-of-state may also happen between a dynamic and a static phase (<ϕdyn, τ, ϕstat>), which corresponds to the prototypical event representing the whole event nucleus.

The assumptions presented here are summarised in Figure 4.9.

31 Henceforth, the notion of ‘static phase’ will be used only to characterise consequent states. 32 This configuration is basic to inchoative verbs (Breu 1998).

126 Chapter 4: Semantics

Figure 4.9 The first line represents the event nucleus with all the parts that are present with a prototypical event (but that need not be present with every event). These parts are encoded by the verb in terms of phases and boundaries whose combination determines the Aktionsart of the verb accordingly. Events can be classified with respect to their boundaries. With some classes of events, a boundary is inherent – either an initial or a final one. With other classes, the boundaries are part of world knowledge according to which there is usually an initial boundary that starts the event. Since an event cannot go on forever, a final boundary can be assumed as well. These boundaries are not encoded by the verb, they have to be specified by other means or left to inference. There is also a class of states of affairs that do not specify any kind of boundaries. Such states of affairs are inalienably tied to objects/individuals they are assigned to. In the specific example illustrated in Figure 4.9, the verb/VP encodes a boundary, i.e. a culmination point.

The representation in Figure 4.9 integrates three different knowledge domains: the event level is conceptual, the verb (Aktionsart) level is lexical and the topic time and the tense relations specified on the time axis of speaker/hearer are grammatical33. What is missing in this figure is the calendar time, which is the independent time in which all the physical world is embedded34.

In languages that do not encode aspect grammatically, the inherent constitution of verbs are taken as default and related to a default topic time (cf. chapter 2, section 2.2.3). Aspect languages, however, have the means to modify this default mapping in two respects: they can explicitly select a certain element of the temporal tier, and they can specify the explicit relation of this selected element to the topic time. For the topic time, boundary-properties different from those delivered by the verb can be specified. This does not mean cutting off the inherent boundaries but imposing external, temporal ones that may be reached before the inherent limit is reached. The aspectually relevant properties encoded by the verb remain in the background and serve as basis for further inferences that participate in the interpretation of the respective form.

Figure 4.9 illustrates one specific case of aspectual selection, namely one possible value the Russian pf aspect may acquire. This is the case in examples like the following:

33 If languages lack grammatical means to specify these features, they make use of lexical and/or contextual

means to express them. 34 Later it will turn out that in some cases a further interval is necessary – the validation time interval ‘T-Val’

which is needed, e.g., in case of iterative events.

TIME AXIS (including TU) grammatical/lexical

[ [TU

[ ] m TSit n

(τ)

preparation phase consequent stateculmination point

EVENT (its running time) conceptual

ϕdyn τ ϕstat

VERB/VP (Aktionsart) lexical

TOPIC TIME (what is asserted) grammatical/lexical

[ ]

127Chapter 4: Semantics

4.24 On rešil zadaču. he solve:PST:pf exercise:ACC ‘He solved the exercise.’ Here, the boundary of the event (the change-of-state from the process of solving the exercise to the exercise being solved) is selected and related to the topic time in such a way, that it is included in the topic time – the exercise is solved within TT. The topic time interval thus can be described as follows: 4.25 τ ∈ TT | m ≤ τ ≤ n The interior of an interval consists of points, among which the boundary constitutes one special point. In case of ( 4.25), the ‘≤’-relation indicates that this point may be the first one of that interval, or that there might be other points preceding it. The same holds for the end of the interval. This accounts for punctual verbs that are not preceded by a lexically encoded preparation process and consequent state, as well as for verbs that encode both phases. With such verbs to assert the boundary, i.e. the change of state, it is necessary to include into the interval also parts of the phases preceding and following that culmination point.

An interval specified in the way illustrated above is then projected onto the temporal axis by the tense operator and thereby enters into tense-temporal relations of anteriority and posteriority. In the above case, the boundaries of both intervals indicate that the situation time precedes the time of utterance.

This section has provided a brief illustration of the way aspect will be analysed here. The semantics of the aspect category has been defined in time-relational and selection-theoretic terms, i.e. as the relation between a topic time and the part of the Aktionsart basis selected by the respective aspectual marker. Aspect systems of different languages differ in the exact nature of that relation and in the units selected. The possibilities of this selection are constrained by the event nucleus. This event nucleus represents the prototypical event which consists of a preparation phase, a culmination point and a consequent state. These parts are immediately relevant for aspect operators and are encoded by the verb on the temporal tier in terms of phases and boundaries – symbolised by ϕdyn (dynamic phase), τ (culmination point/boundary) and ϕstat (static phase). Aspect operators select the respective matching unit and relate it to the topic time TT in a specific way. This relation between the selected element and TT constitutes the semantics of the aspectual marker and is one of the factors involved in the derivation of the respective contextual interpretation.

The basic notions that have been introduced in this section will be used in the remainder of this chapter to analyse the semantics of aspect in Russian, Bulgarian and Turkish.

4.2 Russian

The two forms instantiating the grammatical category of aspect in Russian are the pf and the ipf aspect. Both interact with the basis provided by the Aktionsart in specific ways. The present section outlines the basic semantics of aspect in Russian and combines it with the factors of boundedness and terminativity. This proves important for an account of the various readings that are possible for the aspect forms.

4.2.1 Aspect, boundedness and terminativity In chapter 2, section 2.2.1 it has been argued that it is necessary to distinguish between boundedness and terminativity, which both in turn need to be distinguished from grammatical

128 Chapter 4: Semantics

aspect. The combination of the ingredients for aspect semantics introduced in the previous section will now be applied to the aspect system of Russian. First, the basic semantics of aspect will be elaborated, then boundedness will be dealt with in more detail. Finally, the possible combinations of the three features aspect, terminativity and boundedness will be outlined. It is important to note that the following remarks pertain to the semantics of aspect, and do not model the process of interpretation. The semantics is the stable part of Russian pf and ipf aspect (level 3 of Figure 4.1), i.e. the logical entry specified by aspect and its interaction with the logical entries provided by the Aktionsart basis. This semantics directs and constrains the interpretation of the encountered verb in a certain way, but still needs to be enriched to get the contextually relevant interpretation.

According to Klein (1995), in case of the Russian ipf aspect, TT must have a common subinterval with the source state SS in complex predicates, or with the single state in one-state predicates. It may not have a common subinterval with the target state TS35. The general overlap-relation can be stated as follows: 4.26 TT overlaps T-SS and TT not overlaps T-TS The specific nature of the overlap of T-SS and TT is left unspecified. Klein distinguishes three possible relations for the ipf aspect (1995: 689): 4.27 a. TT may be included in TSit

b. TT may contain TSit

c. TT may be simultaneous with TSit Taking into account the selection-theoretic assumptions outlined above, these relations can be refined as follows: 4.28 general overlap relation:

TT ○ e & e | ϕ ⊆ e 4.29 specific overlap relations:36

a. TT ⊂ ϕdyn/ϕstat37

b. TT = ϕdyn τ ϕstat (TT = e)

c. TT ⊃ ϕdyn τ ϕstat (TT ⊃ e) Relations ( 4.29b) and ( 4.29c) can be simplified by replacing the sequence ‘ϕdyn τ ϕstat’ with ‘e’ since for these relations to hold, it is not necessary for all elements of the temporal tier to be present in the Aktionsart basis. The only requirement for the ipf aspect is stated in ( 4.28), namely that some phase has to be present (e | ϕ ⊆ e), which does not necessarily have to be dynamic38. This is important for dealing with stative verbs, which are not dynamic but still the

35 ‘T-SS’ and ‘T-TS’: time of the source state and time of the target state; remember that Klein (1994, 1995)

classifies verbs in terms of 0- ,1- and 2-state contents, the latter involving a source state and a target state. 36 Note that ‘⊂’, ‘⊃’ and ‘=’ express relations between intervals (here: the run time of the event and the topic

time), but do not necessarily make a definite statement concerning the boundedness characteristics of TT. 37 Remember that with inchoative verbs the ipf aspect may also focus upon the consequent state, i.e. the static

phase (chapter 2, section 2.1.1). 38 As has been pointed out above, aspect in Russian is not sensitive to dynamicity.

129Chapter 4: Semantics

ipf can apply – in fact, it is the only possibility in this case39. Moreover, relation ( 4.29) imposes the most specific requirements onto the input40, whereas the other two relations are possible with any kind of input satisfying the general condition in ( 4.28). Stating the possible relations41 in this disjunctive way excludes all those relations that might as well be derived from the general overlap relation: “The disjunction is probably flexible enough, as several of the relations admitted by e ○ t can in practice be excluded as irrelevant.” (Grønn 2003: 58)

Examples ( 4.30)-( 4.32) serve to illustrate the specific relations of ipf and their interpretations. Relation ‘TT ⊂ ϕdyn’ holds in ( 4.30): 4.30 a. Kogda ja vošel v komnatu, moj brat čital knigu i moja sestra govorila po

when I enter:PST:pf in room:ACC, by brother read:PST:ipf and my sister speak:PST:ipf on telelefonu. telephone:DAT

‘When I came into the room, my brother was reading a book and my sister was talking on the phone.’

b. On sdaval ėksamen, no ne sdal ego. he pass:PST:ipf exam, bot not pass:PST:pf it ‘He was trying to pass the exam, but didn’t (couldn’t) pass it.’ ( 4.30a) is an example of the actual-processual reading of the ipf aspect. This reading is commonly assumed to constitue the prototypcial, most central, default interpretation of the ipf (Padučeva 1996)42. It is possible with all non-stative verbs. The conative reading in ( 4.30b) is restricted to terminative ipf verbs. A prerequisite for this reading to arise is the presence of an inherent boundary in the semantic contribution of the Aktionsart basis, the non-selection of which (indicated by ipf) gives rise to the reading of unsuccessful attempt. The reaching of the boundary would be an implicature from the first part of the sentence, but in ( 4.30b) this implicature is cancelled by the continuation no ne sdal ego43.

The second relation possible for the ipf aspect, ‘TT = e’, states that TT is not restricted to the assertion of some specific part of the sentence base, but is unbounded. Since TT is not bounded in this case, it is more appropriate to consider the sentence base as not being related to a topic time, but to a topic entity. In this case, aspect does not select a certain part of the event description to ascribe it to the topic time, but it takes the whole description and relates it to a topic entity. This relation is the default relation for all lexically stative and relational

39 This is different with the English progressive marker -ing that requires a dynamic phase as input. Therefore

in English, lexically stative verbs may only be used in the simple form (there are, of course, exceptions where the context supports the progressive form for stative verbs, or even requires it; cf. Žegarac 1993).

40 In chapter 3 it has been pointed out that the contribution of aspect is regarded as being stable, whereas that of the context – first of all the Aktionsart it is applied to – is variable. Therefore aspect can be said to impose certain input requirements that have to be fulfilled by that context. Cases, where the input has to be changed to meet the requirements of the aspect marker will be dealt with in more detail in chapter 5.

41 Note that at this stage of analysis, the possible relations are merely stated, without specifying how they are derived. The question of how these specific relations are arrived at will be discussed in chapter 5. The general relation in ( 4.28) will then be argued to constitute the lexical entry of aspect, which is pragmatically ambiguous between the specific relations in ( 4.29) (cf. chapter 3, section 3.1.2). This makes it possible to treat the ipf aspect as one gram (cf. Grønn 2003; Steube 1997) that is underspecified with respect to its concrete realisation – with respect to both semantics and interpretation.

42 Grønn (2003) takes the general-factual reading to be central. Remember, however, the difficulties that have been pointed out to arise with the conception of core and peripheral interpretations.

43 This is the main difference between the ipf aspect with terminative verbs and the pf aspect: the ipf aspect leaves the reaching of the boundary as implicature, as it is not selected, while the pf aspect entails this boundary, as it explicitly selects it. Thus, the reaching of the boundary within TT cannot be cancelled with the pf aspect.

130 Chapter 4: Semantics

verbs (Padučeva 1996; Glovinskaja 1982, 2001). Besides these verbs, basically any verb may be used with this ipf relation. Breu (2005) refers to this function of aspect as ‘status-aspect’, delimiting it from the focusing function of aspect (this is the case with relation ( 4.29a)). In the status-use, the ipf aspect abstracts away from the internal make-up of the predicate and turns it into a secondary, or derived, state.

The following examples illustrate this relation:

4.31 a. On chorošo znal, čto ėto sejčas ne vozmožno. he well know:PST:ipf, that this now not possible ‘He knew (*was knowing) very well, that this was not possible then.’

b. V detstve ja chorošo igrala v šachmaty. in childhood I well play:PST:ipf chess ‘When I was a child I could play chess very well.’

c. Ja mnogo trenirovala i podnimala 100 kg. I much train:PST:ipf and lift:PST:ipf 100 kg ‘I trained a lot and could lift 100 kg.’ ( 4.31) is an example of a stative verb that cannot be assigned the actual-processual interpretation as there is no dynamic phase in its semantic representation; the aspectual relation is ‘TT = e’. Lexical statives are not part of the event nucleus, as they do not enter in any causal or contingency relationships (Moens & Steedman 1988: 23). Therefore, they cannot be characterised in terms of phases or boundaries44. The situation is different with examples ( 4.31b) and ( 4.31c). They would in principle allow for the relation ‘TT ⊂ ϕdyn’, but here they are used in contexts that do not give rise to this relation. In this case, we are dealing with status aspect. As will be seen below (section 4.2.2), examples like ( 4.31b,c) differ from examples like ( 4.31a) in that they are repetitive or ‘second order’ events.

The third relation, ‘TT ⊃ e’, is essential for those uses of ipf that allow for the completedness-reading. In these cases, TT is characterised as closed-bounded, including all phases and boundaries encoded by the predicate. The readings that may emerge from this relation are the general-factual use ( 4.32a) and the durative reading ( 4.32b): 4.32 a. Ja lomal ruku.

I break:PST:ipf arm:ACC ‘I (once) broke my arm.’

b. Ja guljala ot trech do pjati. I go.for.a.walk:PST:ipf from three:GEN to five:GEN ‘I took a walk from three to five.’ The breaking of the arm in ( 4.32a) happened long ago (indicated by once in the translation), and the result, i.e. the broken arm, is not visible any more. The difference between this use of the ipf aspect and the pf aspect is illustrated in ( 4.33), where the result is asserted as still being valid at the time of utterance:

4.33 Ja slomal ruku.

I break:PST:pf arm:ACC ‘I have broken my arm.’

44 Remember that lexically stative verbs are different from static phases that characterise parts of non-stative

predicates.

131Chapter 4: Semantics

This leads over to the Russian pf aspect, the semantically marked member of the Russian aspect opposition. Its meaning is more specific and more informative than that of the ipf aspect. The pf aspect imposes a specific input requirement onto the context, namely that of a boundary to be present. According to Klein (1995), the pf aspect is characterised as having a common subinterval with both SS and TS and is therefore excluded for 1- and 0-state predicates, which necessarily have to be ipf. The corresponding relation can be stated as follows: 4.34 a. TT overlaps T-SS and TT overlaps T-TS

b. TT ⊃ τ From Klein’s definition of the pf aspect in ( 4.34a) it can be concluded that an assertion is made about the change from the source state to the target state, specifically, about the transition bound between both. The pf aspect selects all boundaries encoded by the Aktionsart and relates them to the topic time. This relation is such that the boundary is properly included in TT, compare ( 4.34b). 4.35 On otkryl okno. he open:PST:pf window ‘He opened the window.’ In ( 4.35) the boundary, i.e. the transition bound from the window not being open to the state of the window being open, is asserted to hold at TT. This allows for different specifications: one where the focus is merely on that change of state, and one where in addition to the change of state the resulting consequent state is asserted as well. The former constitutes the ‘event’ reading of pf, the latter the ‘perfect’ reading. Both readings correspond to the general characterisation of the pf aspect (TT includes the boundary), the difference consisting in the exact characteristics of the boundedness-features of the TT-interval. The perfect reading illustrates the need for defining the pf aspect in terms of a boundary being included in the topic time, and not as the event being included in the topic time. With the perfect reading this is not tenable, since the consequent state still holds at the time of utterance. As the consequent state is part of the event and as it holds at TU, it cannot be properly included in the topic time. The decisive feature for pf is thus the proper inclusion of the boundary in the topic time.

To sum up so far, aspect can be characterised in selection-theoretic and time-relational terms: it selects part(s) of the descriptive content of the verb and relates this to the topic time TT. For the Russian pf aspect, the relation is ‘TT ⊂ τ’, for the Russian ipf aspect as the unmarked member of the opposition it is less specific. With the ipf aspect, TT needs to have some overlap with the encoded event, which specifies the general overlap relation ‘TT ○ e’. The only requirement is a phase to be present, i.e. TT ○ e & e | ϕ ⊆ e. These relations constitute the invariant semantic basis of pf and ipf in Russian. Both aspects constitute one gram, although the assignment of form and (procedural) meaning is rather difficult (cf. Steube 1997: 218). This is the case especially for the ipf aspect. The analysis in terms of pragmatic ambiguity (cf. chapter 5) allows for a one-to-one relation between form and meaning while at the same time capturing the fact that ipf may realise three different specific relations of TT and TSit. This suggests that in determining the specific ipf relation in a given context, pragmatics plays an important role, as these relations rather have to be inferred than decoded.

The way from the semantics of aspectualised verbs to their interpretation in a certain utterance can now be further specified as follows:

132 Chapter 4: Semantics

4.36 Aspect: from semantics to interpretation (VI) The interpretation process sketched in ( 4.36) pertains to Russian only, because the ambiguity of the ipf aspect is specific for Russian. Languages differ with respect to which aspect relations their respective markers encode, and which are left to disambiguation or other contextual resolution. ( 4.36) specifies the logical entries of the procedural information provided by the ipf aspect and that of the conceptual information provided by the Aktionsart. The interaction of aspect and Aktionsart takes place with the temporal tier information of the logical entries, where the phases and boundaries provided by the Aktionsart are selected by aspect. The temporal tier units are encoded by the predicate, the general aspectual relation is encoded by the aspect marker. Both kinds of information can be decoded by the hearer and are thus part of the semantic information delivered by the aspectualised verb. However, in order for the interpretation to be succesful, a certain amount of information needs to be inferred, even with the information provided by the logical entries. The information that has to be inferred differs with respect to the respective logical entry it arises from. If there is a mismatch between the Aktionsart basis and the input requirements imposed by aspect, the input specified by the logical entry of the Aktionsart has to be accommodated, the necessary elements – phases or boundaries – need to be induced. Considering the logical entry of the ipf aspect, the specific relation has to be inferred by disambiguating the general relation. These inferring parts are pragmatic tasks. All in all, ( 4.36) illustrates how semantics and pragmatics interact at the level of encoded information. Pragmatics is necessary in interpretation already at this subpropositional level (cf. chapter 5).

Having sketched the semantics of aspect, its independence of both terminativity and boundedness needs to be shown. Although terminativity, boundedness and aspect are all crucial in interpreting utterances containing an aspectualised verb and although they interact rather closely, they have to be kept apart. Cappelle & Declerck (2005) emphasise the need for this distinction and draw the following “threefold aspectual distinction” in terms of terminativity, boundedness and aspect (2005: 895; cf. also chapter 2, section 2.2.1): 4.37 a. (Non)boundedness is a matter of how a particular actualisation of a kind of

event is represented with respect to the question: Does the event come to an end or not?

INTER- PRETATION

inferring:adustment of required input (ϕdyn, ϕstat, τ)

decoding

decoding

inferring:the specific relation between I(e) and I(TT)

ϕdyn, τ, ϕstatWORD MEANING

conceptual information (Aktionsart)

procedural information (aspect)

semantics

pragmatics

pragmatics

133Chapter 4: Semantics

b. (A)telicity [(a)terminativity, B.S.] is a matter of whether or not we conceptualise a kind of event as having an inherent or intended endpoint (point of completion).

c. (Im)perfectivity is a matter of whether or not we view a particular event as a whole and is grammatically expressed in English by the use of simple (nonprogressive) versus progressive verb forms.45 These three notions are connected by default relations: terminative and pf predicates are per default associated with boundedness, aterminative and ipf predicates with unboundedness. The fact that this default connection can easily be overridden shows the independence of these notions (see ( 4.48) and ( 4.49) below).

The interplay of these three notions is essential for distinguishing the ipf from the pf aspect and for a systematic categorisation of the respective semantic bases of the different interpretations of aspectualised verbs. It will become clear, however, that with respect to boundedness, the mere distinction between ‘bounded’ and ‘unbounded’ is too coarse, and that this notion needs to be refined in the way indicated above: intervals may be unbounded or bounded, and bounded intervals may be open, closed or semi-open. The mere dichotomy of ‘bounded’ vs. ‘unbounded’ is also confusing insofar as open-bounded intervals seem to be classified sometimes as unbounded, sometimes as bounded. The distinction of open-boundedness as being distinct from both non-boundedness and boundedness is crucial, however. If one is aware of the fact that it is always intervals that are characterised in terms of boundedness, then this notion of open-boundedness is not contradictory any more (cf. section 4.1.2). Furthermore, since boundedness is a characteristic of intervals, it needs to be specified which interval this notion refers to in a given case.

Cappelle & Declerck (2005) speak of ‘terminative VPs’ and ‘bounded clauses’. This reflects their assumption that both notions pertain to two different linguistic entities. Whereas terminativity pertains to the level of context-independent event-semantics, ‘boundedness’ pertains to the actualisation of an event in a finite clause: “In the case of (un)boundedness event means ‘particular instance of an actualisation, as referred to by the clause, of the kind of event denoted by the V-bar’.” (Cappelle & Declerck 2005: 894)

In order for an event to be actualised, the clause has to be finite. Thus, the actualisation of an event corresponds to the assertion of a sentence base by relating it to a topic component. It is this topic component that Cappelle & Declerck’s notion of boundedness pertains to. As the topic time is the time span an assertion is made about, the boundedness characteristic (as well as its kinesis features, cf. section 4.1.2) of this interval determines the actualisation of the event in the finite clause. An open TT-interval presents the asserted part of a situation as having no explicit beginning and end (but it is still bounded), closed TT-intervals as having an explicit beginning and end. The TT-interval may as well be not bounded at all, in which case it makes an assertion over the whole past time (in case of a past-tense predicate) or a general statement that is not restricted in time at all (in case of present-tense predicates)46. The fact that boundedness characterises the topic time and the fact that the topic time plays a crucial role in the characterisation of grammatical aspect further supports Cappelle & Declerck’s claim that boundedness is an aspectual notion. They summarise the following indicators of boundedness (2005: 896-898): temporal and spatial adverbials and prepositions, the mass/count and definite/indefinite distinction of object NPs, as well as the time of utterance with the ‘now’ of the present as boundary. Languages differ with respect to the items – lexical

45 Capelle & Declerck’s characterisation of aspect pertains to English only and differs from the framework

developed here. 46 In case of an unbounded topic time, no event is actualised (cf. section 4.4.2 on the Turkish Aorist).

134 Chapter 4: Semantics

or grammatical – contributing to the determination of boundedness. In Russian, the openness, closedness or non-boundedness of the TT-interval is left implicit and has to be inferred from the aspect relation or from context. This is different in Bulgarian, which has morphological means to express the boundedness-feature of TT, at least in the past tense and with respect to the distinction of closed vs. open and non-bounded intervals (section 4.3).

Boundedness, however, is a feature of intervals in general. Since in aspect interpretation different kinds of intervals are involved, intervals other than the topic time interval can be specified with respect to boundedness. Besides I(TT), this is the event time interval I(e) and possibly one or more validation time intervals I(T-Val). These intervals each play a role in utterance interpretation. Most central to the interpretation of aspect are I(TT) and I(e). In case of Russian, I(T-Val) may also play a role for aspect (cf. section 4.2.2).

In Russian, there exist two prefixes that modify the basic eventuality with respect to its temporal boundaries: po- and pro- impose an explicit initial and a final boundary onto the run time of the event, but do not add any other meaning modifications. Babko-Malaya (1999) classifies delimiting prefixes like po- and pro- as ‘pure perfectivising prefixes’, which are opposed to resultative prefixes that add a consequent state to the lexical semantics of the basic verb. This misleadingly suggests that po- and pro- are markers of the pf aspect. The following examples illustrate that this is contrary to fact: 4.38 a. kurit’ pokurit’ pokurivat’ smoke:INF:ipf po.smoke:INF:pf po:smoke<(y)va>INF:ipf [Filip to appear] ‘to smoke’ ‘to smoke for a while’ ‘to take repeated drags’

b. Celye dva časa čuvstvoval on sebja počti sčastlivym entire two hours feel:PST:ipf he himselv almost happy:INST i popival konjačok. and po:drink<(y)va>PST:ipf brandy [Filip to appear] ‘For two whole hours he felt almost happy and repeatedly sipped at his brandy.’ ( 4.38a) shows that an aterminative predicate (kurit’, ipf), which is temporally bounded by the prefix po- (pokurit’, pf), may be secondarily imperfectivised yielding pokurivat’. Assuming that -(y)va- is an exponent of the ipf aspect, and given the fact that markers of one and the same grammatical category cannot apply recursively, po- cannot be taken as an aspect marker. This is also illustrated in ( 4.38b), with the secondarily imperfectivised predicate popival konjačok. Such derived imperfectives may, however, denote only iterativity as in ( 4.39a), but not durativity, ( 4.39b), or progressivity, ( 4.39c) (Flier 1985: 41): 4.39 a. Ja ničego ne pišu. Pokurivaju, poleživaju,

I nothing NEG write:PRS:ipf po:smoke<(y)va>PRS:ipf, po:lie<(y)va>PRS:ipf, slegka mečtaju.

slightly dream:PRS:ipf ‘I don’t write a thing. I smoke, I lie around, I day-dream.’

b. Ja dolgo *pokurival, *poležival, slegka mečtal. I long.time po:smoke<(y)va>PST:ipf, po:lie<y)va>PST:ipf, slightly dream:PRS:ipf

c. Ja v ėtot moment *pokurivaju, *poleživaju, slegka mečtaju. I at this moment po:smoke<(y)va>PRS:ipf, po:lie<y)va>PRS:ipf, slightly dream:PRS:ipf

The incompatibility of po- and pro- prefixed verbs with durativity and progressivity is explainable from the function of these prefixes as imposing boundaries on I(e). Dolgo (‘for a long time’) here cannot modify I(e) any more, since this interval is already closed bounded. It therefore could only refer to I(TT), but this again would trigger an iterative interpretation. V

135Chapter 4: Semantics

ėtot moment (‘at the moment’) cannot be used here, since the boundaries on I(e) close this interval and preclude focusing on internal parts of it – here, the dynamic phase – to yield a progressive interpretation.

Thus, po- and pro- have to be regarded as eventuality type modifiers, but they differ from other verbal prefixes in certain respects. Prefixes like po- and pro- explicitly mark the initial and final boundaries of the run time of the event, but they do not modify the semantics of the verb in conceptual terms: “Delimitatives provide a means of quantifying in time an activity [or state, B.S.] that is theoretically limitless.” (Flier 1985: 46)

Even though Babko-Malaya (1999) classifies po- and pro- as perfectivising prefixes, they are not to be mixed up with exponents of grammatical aspect, as has been shown in ( 4.38) and ( 4.39). These prefixes rather operate on the eventuality description. Since these po- and pro- modify the verb and thus are part of the VP, they are part of the input for aspect to apply. The default aspect assignment in this case is the pf aspect. To preserve these insights, that po- and pro- differ from resultative prefixes on the one hand, and from aspect proper on the other hand, it is reasonable, given the interval-characterisation presented in section 4.1.2, to classify eventuality type modifiers according to the features of the eventuality interval they modify – kinesis or boundedness. This is illustrated by extending Filip’s (2005) scheme presented in chapter 2, section 2.2.3:

4.40 [Tense [Aspect [Eventuality-Modifier* (kinesis; boundedness) [Eventuality Description]]]]

Kinesis-modifiers apply to the ϕdyn, τ, ϕstat-characteristics by conceptually modifying them, i.e. by specifying the added phases or boundaries also in terms of conceptual information. Such modifiers contribute inherent boundaries to the eventuality-description, thereby possibly changing its terminativity value from aterminative to terminative. Boundedness-modifiers specify the boundedness-characteristics of the event-time interval. Such modifiers do not add inherent boundaries, but actual boundaries and therefore do not change the terminativity value of the eventuality description. Along these lines, po- and pro- can be analysed as eventuality type modifiers that modify the boundedness characteristics – they explicitly mark the temporal boundaries of the run time of the event. It is these temporal boundaries that are then selected by the pf aspect. Such prefixes do not immediately constrain the topic time interval, but only indirectly by bounding the event time interval. By this, they give rise to a specific default TT, which may then be confirmed at the level of propositional semantics, i.e. when the sentence is asserted47. Other means to indicate temporal boundedness, like temporal adverbials48 or boundedness induced by tense, do not concern the boundaries of the run time of the verb. They thus do not modify the eventuality description and therefore these boundaries do not constitute a possible input for the pf aspect.

To sum up so far, boundedness is most appropriately considered to be a layered system49. Boundedness characterises intervals, and thus this notion has to refer to intervals – the run

47 This is different for the Bulgarian Aorist/Imperfect distinction, which directly concerns the topic time. In

addition to these two forms, Bulgarian has delimiting prefixes as well. 48 Grønn (2003: 82-84) takes durative adverbials as event modifiers and not as contributing to the assertion

time (= topic time) parameter. As has been suggested by ( 4.39b) (cf. also ( 4.43)-( 4.45) below) it is not the case that specific adverbials always modify one and the same interval, and it is a rather complex matter to figure out which adverbials (or other lexical means) pertain to which interval. This is also emphasised by Klein (2001) who argues that, depending on information structure, word order, etc., one and the same adverbial may specify either I(TSit) [here: I(e), B.S.] or I(TT). The crucial point with aspect choice in Russian is whether – in case of I(e)-bounding – the boundaries are marked within the VP and thus contribute to the input for aspect, or outside the VP.

49 Lindstedt (1985) provides an analysis of the Bulgarian aspect system in terms of layered structures relying on the notion of boundedness.

136 Chapter 4: Semantics

time interval of the event I(e), the topic time interval I(TT) or the validation time interval I(T-Val)50. The time axis of the speaker may also play a role here, as the ‘now’ of the time of utterance bounds the past tense interval to the right, and the future tense interval to the left.

The different means that serve to convey boundedness therefore may pertain to different intervals. The interval a certain lexical item contributes to is not so much defined by its lexical content, but rather by the respective temporal scope relations. A temporal adverbial like ot…do (‘from…to’) that is listed in ( 4.42) as contributing to I(TT) may as well contribute to I(T-Val). Compare the following pair of sentences: 4.41 a. Ja čitala ot pjati do šesti. I read:PST:ipf from five:GEN to six:GEN. ‘I read from five to six.’

b. Ona vyigryvala marafon ot 2001 do 2004 goda. she win:PST:ipf marathon from 2001 to 2004 ‘She won the marathon from 2001 to 2004.’ In ( 4.41a), ot…do temporally bounds the topic time interval, whereas in ( 4.41b) it bounds the validation interval. The time span in ( 4.41b) – ‘from 2001 to 2004’ – is too long for just one single event of marathon-winning and thus this utterance is interpreted as reporting on an iterative event. Each of these single events carries its own topic time, but these single events are asserted to hold within some bigger interval of time – the validation time I(T-Val) (for more details see section 4.2.2). Depending on how many such layers are involved, more than one I(T-Val) might be necessary. What matters for aspect, is the specification of the topic time interval.

In ( 4.42) some of the language-specific means that serve to specify the boundedness feature of intervals are listed, keeping in mind they may as well refer to other time intervals: 4.42 a. I(e): po-, pro-

b. I(TT): adverbials like ot…do (‘from…to’), Aorist/Imperfect in Bulgarian

c. I(T-Val): adverbials like obyčno (‘usually’)

The arrow indicates that the respective intervals are ordered according to their projection line: I(e) is projected onto I(TT), which again is projected – if needed – onto I(T-Val) (cf. Figure 4.9).

The different layers of intervals that may be characterised in terms of boundedness51 are illustrated in ( 4.43)-( 4.45) with examples from Russian and Bulgarian (a detailed analysis of aspect in Bulgarian will be delivered in section 4.3)52:

4.43 a. Ja obyčno guljala ot trech do pjati.

I usually take.a.walk:PST:ipf from three:GEN to five:GEN ‘I usually took a walk from three to five.’

b. guljat’ I(e) no explicit boundaries ot…do I(TT) boundaries on topic time obyčno I(T-Val) no boundaries on validation time past tense TU right boundary on validation time 50 As has been pointed out, this is an additional interval that is necessary for the validation of iterative,

habitual, etc. events (cf. section 4.2.2). 51 Note that here boundedness is not yet specified as open or closed. 52 The specification of the boundedness-features of the respective intervals are not necessarily explicitly

given, but may also need to be inferred.

137Chapter 4: Semantics

c. [+b TU [-b obyčno [+b ot…do [-b guljala]]]]

4.44 a. Deteto popisa malko (posle izleze i ne se vărna veče). child:DEF write:A:pf little then go.out:A:pf and not REFL come.back again ‘The child wrote a little bit (then left and did not come back again).’ b. piša I(e) no explicit boundaries po- I(e) boundaries on run time of event Aorist I(TT) boundaries on topic time malko I(TT) boundaries on topic time (confirmation) past tense TU right boundary on topic time (confirmation)

c. [+b TU [+b Aorist [+b po [-b piša]]]] 4.45 a. Štom popišech s pisalkata Mont-Blanc (vednaga mi stavaše prijatno).

whenever write:I:pf with pen:DEF Mont-Blanc always REFL feel:I:ipf good ‘As soon I wrote (for a while) with a Mont-Blanc pen (I felt good).’

b. piša I(e) no explicit boundaries po- I(e) boundaries on run time of event Imperfect I(TT) no boundaries on topic time štom I(T-Val) no boundaries on validation time past tense TU right boundary on validation time

c. [+b TU [-b štom [-b Imperfect [+b po [-b piša]]]]] Note that the time of utterance TU functions as a boundary by the ‘now’ of the present that separates the past from the future. Thus, if past time events do not have a bounded TT, TU imposes a right boundary on TT by default. This default can be overridden by explicit cancellation of the respective implicature, compare ( 4.46): 4.46 Susi played tennis this morning. In fact, she may still be playing. In addition, the number of validation time intervals I(T-Val) is not constrained to merely one. Depending on the number of relevant adverbials, there may as well be several layers of validation time intervals. For the purpose here, one layer of T-Val is enough. This is only necessary for ‘second order’ events (cf. section 4.2.2), in the case of single events, T-Val coincides with TT, and there is no need for this layer at all. As boundedness may in fact pertain to different intervals, it is important to specify the relevant interval boundedness refers to. Unless indicated otherwise, boundedness here will refer to I(TT). It is this interval that is involved in aspect semantics – aspect expresses the relation between I(e) and I(TT). I(T-Val) may be imposed by means of adverbs or by markers specifying genericity or iterativity53.

Aspect has not played a role in the above examples on boundedness. Aspect does not modify the boundedness-characteristics of intervals, but establishes a relation between two intervals. Aspect in Russian is sensitive to the presence of inherent boundaries and event time boundaries which are specified independently. The pf aspect requires boundaries to be present as input for it to apply. It does not matter whether these are inherent or temporal boundaries, as long as they are specified within VP54. The pf aspect selects the boundaries provided by the

53 Based on Czech data mainly, Filip & Carlson (1997) argue for a distinction of genericity and iterativity

from aspect. 54 The fact that boundaries are specified within the VP does not mean, that the pf aspect necessarily has to be

applied. Its application would be the default choice that may again be overriden (cf. ( 4.38)).

138 Chapter 4: Semantics

VP-layer, i.e. inherent boundaries and temporal boundaries on the run time of the event provided by po- and pro-. It is not sensitive to outer layer boundaries that are provided by adverbials or, in case of Bulgarian, by the Aorist and Imperfect. In Russian, the connection between boundedness and aspect is rather close, given the fact that the pf aspect explicitly selects boundaries and relates them to the topic time in a way that this interval includes all inherent and temporal boundaries. However, this close relationship between pf aspect and boundedness is no entailment. The notions of terminativity, boundedness and aspect are independent of each other. The specific combination of these three features constrains the range of possible interpretations for an aspectualised verb to a considerable degree.

The possible interpretations (readings) of Russian ipf and pf aspect that will be accounted for are the ones listed in chapter 2, section 2.1, repeated in ( 4.47): 4.47

As repeatedly pointed out, it is not claimed that these readings ‘exist’ as clearly distinct and unambiguously definable meanings of ipf and pf. In chapter 5, it will be argued that it is not decisive how many and which readings are postulated; what needs to be figured out is the semantic basis, that is the specific configuration of the encoded contributions of aspect and Aktionsart. This semantic configuration constitutes the basis for the interpretation process. It defines a certain domain within which all the readings that may be constructed for an aspectualised verb in a specific context are located. For the sake of argumentation, however, the readings listed in ( 4.47) will be made use of.

Russian does not have grammatical means to express boundedness (except for the two prefixes po- and pro-), but makes use of the context (encoded or inferred) to specify this feature. The ipf aspect imposes the input requirement on its context that one phase has to be present. According to the ‘Aspectual Uniformity Hypothesis’ (Bickel 1996) it does not matter whether this phase is specified by the verb or by the verb phrase. Ipf aspect may also be encountered in cases where neither verb nor verb phase seem to provide a phase as in ( 4.48c). The following examples illustrate different combinations of ipf aspect, boundedness and aterminativity:

4.48 a. Posle obeda, ja guljala v parke.

afternoon, I go.for.a.walk:PST:ipf in park:LOC ‘In the afternoon, I took a walk in the park.’ <ϕdyn>, [-b]

b. Kogda ty pozvonila, ja čitala gazetu. when you call.up:PST:pf I read:PST:ipf newspaper:ACC,

‘When you called me up, I was reading the newspaper.’ <ϕdyn, τ>, [-b]

c. Ja guljala ot trech do pjati. I go.for.a.walk:PST:ipf from three:GEN to five:GEN

‘I went for a walk from three to five.’ <ϕdyn>, [+b]

readings for the Russian ipf aspect a. actual-processual b. conative c. inactual/continuous d. durative e. general-factual f. iterative g. habitual h. potential i. permanent j. atemporal

readings for the Russian pf aspect a. event b. perfect c. pluperfect d. delimitative/perdurativee. future f. potential g. iterative h. summary i. exemplifying

139Chapter 4: Semantics

d. On vyigryval marafon ot 2000 do 2004 goda. he win:PST:ipf marathon from 2000 to 2004 year:GEN.

‘He repeatedly won the marathon from 2000 to 2004.’ <τ>, [+b] The predicate in ( 4.48a) is aterminative and the situation is temporally unbounded55, the ipf aspect may apply without problems. In ( 4.48b), the situation is temporally unbounded as well, but the predication (‘read the newspaper’) encodes an inherent boundary and is terminative. Still, the ipf aspect may apply without problems; it focuses on the preparatory phase. This does not mean that the culmination – the inherent boundary – is cut off, as aspect does not affect terminativity. The boundary is backgrounded but present, it is simply not asserted to hold at TT. Aspect may, however, influence boundedness. In this specific example, the ipf aspect presents the event of reading the newspaper as unbounded. The combination of aterminative predicate and bounded situation is shown in ( 4.48c). The external temporal boundary is introduced by the temporal adverbial, but it does not prevent the application of ipf aspect56. Somewhat more difficult is the combination in ( 4.48d), as there is no phase present, neither in the predicate, which consists only of a boundary, nor in the situation, which is presented as temporally bounded. Nonetheless, ipf is possible because here the event is interpreted as repeated, and iteration induces a phase57. Terminative predicates are not necessarily temporally bounded, e.g., when the endpoint is not reached at all (compare 4.48b), or in case the endpoint is reached many times over (vanden Wyngaerd 2001: 76). This is exactly the case in ( 4.48d), and therefore the ipf aspect is appropriate.

Table 4.1 illustrates the possible combinations of the features [±terminativity] and [±boundedness] for the Russian ipf, and the readings these combinations are basic to. At this stage, the only distinction with respect to boundedness is that between ‘bounded’ and ‘unbounded’. The result is rather undifferentiated, and illustrates the need for a more fine-grained analysis of boundedness as introduced above. This will be achieved in section 4.2.2. Note, that that for certain readings the terminativity of the verb does not matter, provided that the general input requirement for ipf – the presence of a phase – is fulfilled. The durative reading is regarded as subcase of (2), as it is possible only with aterminative predicates58.

ipf aspect in Russian terminativity/boundedness reading

1 [±t] [-b] actual-processual, conative, habitual,

iterative, habitual, potential, permanent, atemporal

2 [±t] [+b] general-factual, durative, iterative Table 4.1

Now let’s turn to the pf aspect and its combinations with boundedness and terminativity. Examples ( 4.49a-e) illustrate the possible combinations that are summarised in Table 4.2 below:

4.49 a. Ja vstal rano utrom, sdelala zarjadku, prinjal duš i I get.up:PST:pf early morning, do:PST:pf exercise:ACC, take:PST:pf shower and

55 The rather vague notion of ‘boundedness of situation’ will be specified in section 4.2.2. 56 As has been pointed out above, the crucial point is that here, the temporal boundary is not introduced within

the VP, as is the case with delimitative and perdurative verbs. Therefore, the ipf aspect is the default in this case, whereas delimitative verbs are per default pf if they are intended to denote single events. The the pf aspect with time span adverbials would be possible only to refer to repeated events, a rather marginal use in Russian.

57 The issue of iterativity is in fact more complex than indicated here. Cf. section 4.2.2, especially footnote 69. 58 Durativity with terminative predicates yields iterative events, cf. section 4.2.2.

140 Chapter 4: Semantics

pošel v universitet. go.off:PST:pf to university

‘I got up early in the morning, did some exercise, took a shower and went off to the university. <ϕdyn, τ>, [+b] (past)

b. Ja uznal ob ėtom iz gazety. I get.to.know:PST:pf about this from paper:GEN

‘I got to know this from the paper.’ <τ, ϕstat>, [+b] (past)

c. U nego takaja privyčka – kak vernetsja s progulki, srazu otkroet okno. at him such habit – whenever return:PRS:pf from walk:GEN, at.once open:PRS:pf window

[Zaliznjak & Šmelev 1997: 18] ‘He has such a habit – as soon as he comes home from a walk, he opens the

window.’ <τ, ϕdyn>, [-b] (non-past)

d. On podnimet ėtot jaščik. he lift:PRS:pf this box [Glovinskaja 2001: 215] He can lift this box. <ϕdyn, τ>, [-b] (non-past)

e. Ona vsju žizn’ prožila v derevne. she whole live PRF:live:PST:pf in village:LOC

‘She spent her entire live in the village.’ <τ, ϕdyn, τ>, [+b] (past) The central case with respect to the pf aspect is its application to the combination of a terminative predicate and a bounded situation. This may yield an event reading as in ( 4.49a), or a perfect reading as in ( 4.49b), where the lexically encoded consequent state is asserted to hold a the time of utterance. ( 4.49c) illustrates that the boundedness feature arising from the use of the aspect marker is a default only. In certain contexts, the situation may be unbounded also with the pf aspect. In ( 4.49c), the use of the present tense, as well as the inference from privyčka (‘habit’), induce the reference to a ‘second order’ event yielding the interpretations listed under (2) in Table 4.2. The unboundedness of the situation in both ( 4.49c) and ( 4.49d) is induced by the non-past tense of the verb59. The potential reading in ( 4.49d) arises from the unbounded iterativity of single events of lifting. ( 4.49e) shows, that the pf aspect may also be applied onto an aterminative verb, which then has to be temporally bounded. For pf to apply, these temporal boundaries have be marked on the verb itself. This is the case with delimitative and perdurative verbs (case 3 of Table 4.2) that turn processes and states into bounded entities. This distinguishes the pf aspect from the general-factual and durative use of the ipf aspect. In the former case, boundedness is marked explicitly on the verb, whereas in the latter case, it is conveyed implicitly by discourse pragmatic and contextual means. If boundaries are marked on the verb, they constitute an appropriate input for pf aspect to apply, even if no inherent boundaries are encoded by the verb. In case of the durative ipf aspect, boundedness is induced by temporal adverbials, in case of the general-factual use of ipf, it is induced by discourse-pragmatic means (discourse mode ‘narration’). In both cases, the boundaries are specified outside the VP, and therefore the pf aspect cannot apply.

The only combination that is not possible for the pf aspect is combination (4) in Table 4.2, where the pf aspect would be applied to VP that encodes neither an inherent nor a temporal boundary. The impossibility of this combination is predictable from the general semantic

59 This already indicates why it is not sufficient to distinguish only bounded and unbounded situations. There

is a difference between situations presented in ( 4.49c) and atemporal situations like ‘2 + 2 = 4’. It is unsatisfactory to speak of unboundedness in both cases, but still ‘boundedness’ is not appropriate for neither case. Furthermore, any situation, except for eternal truths, is bounded in time – even situations presented in the present tense. These distinctions have to be accounted for. In section 4.2.2, the notion of boundedness will be split into open- and closed-boundedness.

141Chapter 4: Semantics

characterisation of pf that requires a boundary as input from the VP, no matter where this boundary comes from – be it an inherent one or an actual, temporal one. There is no way to induce a boundary pragmatically into the VP (Bickel 1996). A boundary has to be morphologically marked on the verb.

Table 4.2 lists the various combinations of terminativity and boundedness with the pf aspect. These assumptions are still preliminary and will be specified in the next section:

pf aspect in Russian terminativity/boundedness reading

1 [+t] [+b] event, perfect, pluperfect, summary, iterative

2 [+t] [-b]60 potential, exemplifying, future 3 [-t] [+b] delimitative, perdurative 4 [-t] [-b] Ø

Table 4.2

In this section it has been shown that aspect, (a)terminativity and (un)boundedness have to be kept apart, they are independent of and orthogonal to each other. Now that the relevant factors are isolated, the task of outlining a systematic and predictive derivation of ipf readings seems to be manageable. The respective combinations of (a)terminativity, (un)boundedness and grammatical aspect help to constrain the range of possible interpretations, as will be illustrated in more detail in section 4.2.2. This claim can be illustrated more concisely usig the interval notation introduced in section 4.1.3.

4.2.2 Aspect semantics and aspect readings Table 4.1 and Table 4.2 illustrate that simply distinguishing between some notion of ‘bounded-ness’ and ‘unboundedness’ is not enough for a systematic specification of the semantic basis for possible aspect interpretations. The semantic contribution of aspect consists in establishing a topological relation between the run time of some specific part of the event and the topic time interval. Boundedness has been argued to pertain to the topic time interval (or, as will be shown later, to an outer interval of validation time). Intervals are not specified merely in terms of unbounded vs. bounded, but in terms of unbounded, open-bounded and closed-bounded.

Aspect operators map properties of events as encoded by the predicate onto properties of times (cf. also Grønn 2003: 49). The semantics of aspect is represented in terms of two intervals – I(e) and I(TT) – that are related, or rather intersected, in characteristic ways. As a result of this intersection, the topic time interval is assigned a specific kinesis characteristics (in terms of the temporal tier units that are related to it). Its boundedness features are specified independently. In Russian, this is left to inference from respective verbal prefixes or induced by means of context. The thus characterised I(TT) is related to the time axis and positioned with respect to the time of utterance by means of tense, and referred to from TO (cf. Figure 4.8).

The intersection I(e) ∩ I(TT) captures the fact that the topic time stretches only over some specific part of the run time of the event. This will be illustrated by means of diagrams in ( 4.50)-( 4.53). Before illustrating the three specific relations of ipf, consider the boundedness-features of I(e) as represented in these diagrams. I(e) is open to the right, which is due to the fact that these representations do not refer to a specific Aktionsart. They represent the whole event nucleus, and therefore I(e) contains also a consequent states. A consequent state is

60 This combination is only possible for the present tense. Therefore readings characterised by that

combination are difficult to get for the pf aspect, that does not refer to the present tense.

142 Chapter 4: Semantics

encoded by a static phase which does not contain a final boundary. Thus, I(e) is open-bounded to the right. Since events can reasonably be assumed to have some starting point, I(e) is closed to the left61. In this respect it is interesting to note that predicates only rarely explicitly specify the temporal boundaries of the event they denote (this is the case at least for the languages dealt with here). More commonly it is an inherent boundary that is incidated and this inherent boundary per default constrains the event also temporally. This will be discussed in more detail in the course of this section.

The diagrams ( 4.50)-( 4.53) illustrate the I(TT)-I(e)-mappings for the Russian pf and ipf aspect according to the relations given in ( 4.29a-c)62. Note that the relations presented in ( 4.29a-c) are general and do not yet apply to certain verb classes. Therefore, the whole event nucleus is presented as being encoded. Klein’s relational analysis in ( 4.50a)-( 4.53a) is represented in terms of intervals in ( 4.50b)-( 4.53b). 4.50 a. TT included in the dynamic phase: TT ⊂ ϕdyn

63

b. I(e)[i j[ ∩ I(TT)]m n[ → I(TT)]m n[ = {ϕdyn, m < ϕdyn < n}64

This relation is basic also to ‘true’ intraterminal forms that explicitly exclude any boundaries from their assertion scope. This pertains to both the boundaries specified by the Aktionsart basis and the topic time boundaries and is captured by the fact that the TT-interval is open-bounded. The culmination point of the denoted event is outside the scope of assertion. 4.51 a. TT simultaneous with phases and boundaries: TT = ϕdyn τ ϕstat

b. I(e)[i j[ ∩ I(TT)]m n[ → I(TT) = e

61 The probability of such a definite starting point is different with different Aktionsarten (cf. Breu 1985), but

as the diagrams below do not pertain to specific verb classes, these differences do not matter. 62 TO and tense are omitted in these diagrams. 63 As has been mentioned above, with <ϕdyn τ ϕstat>-verbs the ipf aspect may focus on the static phase as well.

The semantic configuration looks the same, but ϕdyn has to be replaced by ϕstat. 64 Read: the intersection of the semi-open I(e) with i as the initial, included boundary and j as the final,

excluded boundary, and the open-bounded I(TT) with m and n constituting the excluded initial and final boundaries, yields an topic time interval I(TT) with m and n constituting the excluded initial and final boundaries such that this interval includes ϕdyn , but excludes any temporal (initial and final) boundaries. The kinesis-feature of I(TT) is thus provided by the selected part of I(e), the boundedness feature is induced by the context that supports the relation ‘TT ⊂ ϕdyn’. The other diagrams are to be read accordingly.

[ [ i ϕdyn τ ϕstat

j AKTIONSART I(e)

] [ m

n TOPIC TIME I(TT)

ASPECT

[ [ i ϕdyn τ ϕstat j

AKTIONSART I(e)

TOPIC TIME I(TT)

ASPECT

143Chapter 4: Semantics

In this case, the TT-interval is unbounded. There is no explicit specification of possible external boundaries. This allows for two possibilities: either the topic time is not restricted to a certain time span at all (atemporal readings; note that ‘atemporal’ is somewhat exaggerating facing the fact that nothing can escape calendar time), or it is bounded by TU or via the life span of the entity it refers to (see below). 4.52 a. TT includes phases and boundaries: TT ⊃ ϕdyn τ ϕstat

b. I(e)[i j] ∩ I(TT)[m n] → I(TT)[m n] = {e, m ≤ e ≤ n}

With the relation ‘TT ⊃ e’, the TT-interval is closed-bounded. The sentence base is completely included in it.

Note that these three relations pertain to singular events. As will be shown below, the analysis of habitual and iterative events is slightly more complicated but makes use of the same notions and distinctions that characterise singular events.

It remains to be explained, why it is possible to speak of these three clearly distinguished aspectual relations that may characterise the ipf aspect in Russian. Evidence is provided by Turkish, which has morphological means to express these relations. As will turn out in chapter 5, it is the combination with a pragmatic approach that unifies these contradictory facts in Russian, namely its uniform meaning that allows it to be regarded as one gram on the one hand, and the three specific relations on the other hand.

The pf aspect as the marked member of the aspect opposition in Russian is semantically more specific than the ipf aspect. It imposes a more restricted input requirement onto its basis, namely that there has to be present a boundary which may be selected and included in the topic time interval. This is captured by the first diagram. 4.53 a. TT includes the boundary: TT ⊃ τ

b. I(e)[i j[ ∩ I(TT)[m n] → I(TT)[m n] = {τ, m ≤ τ ≤ n}

( 4.53) represents the case that the boundary is properly included in the topic time. As the boundary is one specific instance of the points included in an interval, it is possible for the boundary to be preceded and followed by other points. Punctual verbs are the only case where exclusively the boundary is asserted, with all other verbs there is always some part of the phases preceding and following the culmination point that is, and has to be, included in the

[ [ i ϕdyn τ ϕstat j

AKTIONSART I(e)

[ ] m

n TOPIC TIME I(TT)

ASPECT

[ [ i ϕdyn τ ϕstat j

AKTIONSART I(e)

[ ] m

n TOPIC TIME I(TT)

ASPECT

144 Chapter 4: Semantics

asserted time. The reading that is based on the configuration in ( 4.53) is the so-called ‘event’ reading of the pf aspect.

However, the relation ‘TT ⊃ τ’ allows for yet another configuration. This configuration differs from the former in that the consequent state is asserted to hold at the time of utterance. The ‘perfect’ reading arising from it cannot be distinguished form the event reading by means of the relation alone. The relation is the same for both possibilities, what is different, is the boundedess feature of the topic time interval:

b’. I(e)[i j[ ∩ I(TT)[m n[ → I(TT)[m n[ = {τ, m ≤ τ < n} With the perfect reading, the topic time interval is open to the right. This allows the consequent state to be valid at the moment of reference.

The relation for the pf aspect allows for a further possible characterisation of TT, namely as semi-open to the left. This constellation is basic to ‘pluperfect’ readings of the pf aspect:

b’’. I(e)[i j[ ∩ I(TT)]m n] → I(TT)]m n] = {τ, m < τ ≤ n} The relations sketched in ( 4.50)-( 4.53) constitute the semantics of aspect in Russian. Three factors are important: first, the element(s) selected from the Aktionsart basis, second, the topic time interval and third, the relation between both. Without a specification of I(TT) in terms of characterising it as unbounded, open-bounded, closed-bounded or semi-bounded, it is impossible to draw certain important distinctions. Relations like those presented above for Russian can be established for every aspect system. Languages differ in the relations they explicitly encode and in the markedness of the respective relations (for a survey of Russian, Bulgarian, Turkish and English cf. section 4.5).

These basic relations of ipf and pf aspect interact with the Aktionsart of the verb in a specific way. This interaction constrains the range of possible interpretations and allows for certain pragmatic inferences. Again, intervals will be used as means of representation. Note that what is stated are the final readings, not how they are derived. The process of interpretation will be the topic of chapter 5.

[ [ i ϕdyn τ ϕstat

j AKTIONSART I(e)

[ [ m

n TOPIC TIME I(TT)

ASPECT

[ [ i ϕdyn τ ϕstat

j AKTIONSART I(e)

] ] m

n TOPIC TIME I(TT)

ASPECT

145Chapter 4: Semantics

4.54 actual-processual reading

4.55 conative reading

Both ( 4.54) and ( 4.55) illustrate the relation ‘TT ⊂ ϕdyn’. For this relation to hold, the terminativity of the verb does not matter – what is required is a dynamic phase to be present. The difference between terminative and aterminative verbs consists in that with the former, the conative reading, ( 4.56a), is preferred (the actual-processual would be possible as well), whereas with the latter, the actual-processual reading is preferred, ( 4.56b). The boundary, if present, is outside the scope of assertion, it is not deleted but left for implicature. As such, it may be explicitly cancelled by the following context. 4.56 a. On rešal zadaču.

he solve:PST:ipf exercise:ACC ‘He was trying to solve the exercise.’

b. Kogda ja vošel, ona čitala. when I come.in:PST:pf, she read:PST:ipf ‘When I came in, she was reading.’ In case of the ‘TT ⊂ ϕdyn’, it is not possible to explicitly mention an initial and final actual boundary, both are outside the scope of assertion. This is the prototypical instance of intraterminality, where the event is envisaged without reference to any boundaries. The impossibility of including the boundaries into the scope of assertion is indicated by the relation between TT and the event-part, which is an inclusion relation: TT is included in the dynamic phase, which excludes inherent boundaries from being asserted, and TT is open-bounded, which excludes temporal boundaries from being asserted.

Since the Russian ipf aspect is not marked for intraterminality, it can be used with temporal adverbials indicating a time span like ot… do (‘from… to’) that temporally bound the situation. In this case, another relation is activated, namely ‘TT ⊃ e’. This relation is sketched in ( 4.57) and basic to the interpretations in ( 4.58): 4.57 durative reading

[ [ e

[ ]

I(e)

I(TT)

[ [ ϕdyn τ ϕstat

] [

I(e)

I(TT)

] [

[ [ ϕdyn

I(e)

I(TT)

146 Chapter 4: Semantics

4.58 a. V ėtom uščel’e c ijulja do nojabr’ geologi iskali almazy. in this gorge:LOC from Juli to Novemger geologists search:PST:ipf diamonds [Padučeva 1996: 12]

‘From Juli to November, geologists looked for diamonds in this gorge.’

b. Ja ždal tebja polčasa, potom ušel. I wait:PST:ipf you half.an.hour, then go.away:PST:pf [Padučeva 1996: 177] ‘I waited for you half an hour, and then I went away.’

With the durative reading the denoted events are not isolated from the setting (see below), but are part of a narrative chain as is evident especially with ( 4.58b).

This is different for the general-factual use of the ipf aspect, illustrated in ( 4.59) and ( 4.60): 4.59 general-factual reading

4.60 a. Ty kodga-nibud’ razbival cennuyu vazu? you ever break:PST:ipf valuable vase:ACC [Padučeva 1996: 43] ‘Have you ever broken a valuable vase?’

b. Gde knopki? – Sergej vešal kartinu, u nego sprosi. where pins? – Sergej hang:PST:ipf picture:ACC at him ask:IMP [Padučeva 1996: 49] ‘Where are the pins? – Sergej hang up the picture, ask him!’

Again, the terminativity of the verb does not matter65. All elements encoded on the temporal tier are included in the scope of assertion. This relation only allows for a closed-bounded topic time, independent of terminativity and aspect.

This notion ‘general-factual’ is in fact an oversimplification. This term is not used consistently in aspectual analyses and comprises several rather different subgroups. A thorough investigation of the Russian general-factual ipf, its subgroups and its equivalent expressions in other languages are beyond the scope of this thesis. Grønn (2003) offers a critical overview of several accounts of the general-factual reading and provides a profound analysis of the existential/experiential, ( 4.60a), and the presuppositional, ( 4.60b), subgroups66. Grønn’s existential/experiential subgroup of the factual ipf is commonly expressed by the Perfect tense in Bulgarian (and other languages that grammatically encode this tense)67. The perfect differs from the ipf aspect in that it relates two topic time intervals, and not one topic time and a situation/event time interval as is the case with aspect (cf. Pančeva 2003: 284).

Facing these difficulties, the term ‘imperfective of isolated event’ (IIE) introduced by Lindstedt (1985) seems to be more appropriate to denote the phenomena that are intended to be captured here by the notion ‘general-factual’. The IIE does not capture all the phenomena

65 Compare, however, Grønn (2003), who takes telicity as one of the defining features for factual ipf. It is not

entirely clear, though, whether he draws a distinction between inherent boundaries that affect telicity/terminativity and temporal boundaries that don’t.

66 Padučeva (1996: 32-52) also distinguishes several groups of general-factual readings. 67 Compare Sell’s (1994) discussion of the difference between the ‘statement-of-fact’ vs. the ‘general-factual’

reading in Bulgarian. The former is expressed by the Perfect (ipf or pf), the latter by the ipf A. Note again the terminological muddle.

[ [ ϕdyn τ ϕstat

[ ]

I(e)

I(TT)

147Chapter 4: Semantics

that are traditionally subsumed under the heading ‘general-factual’, but those that deal with ‘generalised events’ that are not part of some setting: “[A] generalised event is essentially an event isolated from any setting, that is to say an event not considered as a transition from a particular state of affairs into another.” (Lindstedt 1985: 217)

Lindstedt divides such IIE’s into two main groups: one where the information focus is on one of the actants while the verb is presupposed in the context, and one where the verb is in focus. The former group constitutes presuppositional IIE’s, which Lindstedt argues to be explainable on the basis of the actual-processual and conative readings of the ipf.

The discussion of the general-factual reading illustrates the difficulties that arise if the different possible interpretations are the subject of debate and not their respective semantic bases. Some of the subtypes proposed for the general-factual ipf depend on the lexical semantics of the predicate, others on the discourse setting. Thus, it is more reasonable to proceed as proposed in this thesis: figure out the possible semantic configurations and take these as the starting point for the discussion. The configuration in ( 4.59) is one of the possible semantic characterisations of the ipf aspect and may give rise to a variety of readings, irrespective of whether they are referred to as ‘general-factual’, ‘statement-of-fact’ or ‘IIE’. What is decisive, is the semantic basis. Here, the term ‘general-factual’ will be understood as referring to the denotative use, which is based on the semantic configuration in ( 4.59).

The Russian sentence in ( 4.61) can be interpreted either as the denotation of a fact, which presupposes the configuration in ( 4.59), or as expressing some experience with filling in forms, which presupposes the relation of two topic time intervals. Bulgarian has overt means to distinguish between both. The ipf A in ( 4.62a) denotes the fact that the filling-in happened at some time before, whereas the ipf Perfect in ( 4.62b) expresses the experience with filling in forms: 4.61 Ja uže zapolnjal anketu.

I already fill.in:PST:ipf form:ACC ‘I have already filled in the form.’

(i) It happened before. (ii) I know how to do it.

4.62 a. Az veče popălvach anketata. I already fill.in:A:ipf form:DEF ‘I have already filled in the form.’ (= It happened before.)

b. Az veče săm popălval anketata. I already AUX fill.in:P:ipf form:DEF ‘I have already filled in the form.’ (= I know how to do it.) The next relation ,‘TT = e’, is illustrated in ( 4.63). 4.63 inactual, permanent and atemporal reading

With this relation, tense plays a crucial role. Past tense, for instance, imposes a right boundary onto I(TT). This may give rise to the implicature of a lifetime effect (Musan 1997), arising due to the alternate ‘present tense’ that would have been used in case the property is still

[ [ ϕdyn τ ϕstat

I(e)

I(TT)

148 Chapter 4: Semantics

valid. Thus, the past tense in ( 4.64c) would be odd, as it would implicate that, probably due to some rather revolutionary mathematical finding, the equation is not valid any more. With ( 4.64b), the past tense would be more plausible, as a hotel might easily change its shape and the position of its windows, e.g., by means of a renovation. Still, the use of the past tense would require some explanation in the following context for why this relation is not valid any longer. 4.64 a. Ja dopisyval “Romantikov”, kogda odnaždy večerom ko nme vošel junoša... I PRF:pisat:PST:ipf “Romantikov” when some evening to me come:PST:pf boy... [Padučeva 1996: 29] ‘I was finishing “Romantikov” when some evenig a boy came to me..’

b. Okna gostinicy vychodjat na jug. windows hotel:GEN look.out:PST:ipf to South [Padučeva 1996: 10]

‘The windows of the hotel look out to the South.’

c. Dvaždy pjat’ ravnjaetsja desjati. twice five equal:PRS:ipf ten ‘Twice five is ten.’

The states of affairs denoted by the examples in ( 4.64) refer to individuated events in decreased order. ( 4.64a) refers to an event, albeit a contingent one, that is, it refers to a rather long topic time that makes it unplausible for the respective event to hold at the whole time. Rather, various pauses and phases without an event of dopisyvat’ Romantikov have to be assumed. Thus, ( 4.64a) does not refer to an ongoing process. Utterance ( 4.64b) denotes a permanent state that characterises the topic entity – in this specific case the window of the hotel. This permanent state is temporally bounded by the life span of the entity in question. The denoted state in ( 4.64c) is not permanent but atemporal, because the life span of the entity it characterises (‘two plus two’) is not bounded but atemporal. Utterance ( 4.64c) is atemporally valid. There seems to be no reasonable way to temporally constrain its validity.

What still needs to be accounted for are the iterative, habitual and potential readings. These readings will be subsumed under the heading of ‘repetitive events’ or ‘second order events’. With these readings, the additional layer of validation time comes into play. These readings differ from those mentioned so far, as they do not pertain to single events but to repeated events er that are subsumed under one temporal interval, the validation interval I(T-Val). This additional validation interval is needed, because in this cases each occurrence of the event specifies its own topic time. These multiple occurrences of the respective event are specified as sketched above, namely in terms of a topic time with specific boundedness and kinesis properties, characterised by some selected part(s) of the verb. This default topic time is specifcied by the default aspect operator that provides the minimal interpretation of an event predicate entailing event realisation (cf. section 4.1.2). This is indicated by the e in the respective diagrams. In case of iterativity and habituality, it seems plausible to assume a closed-bounded topic time for the single events, as both are generalisations over single events. The relations established for singular events are stated as ‘second order’ relations for repetitive events (er), that is, in terms of a relation between T-Val and er. Note that in these cases, TO is located with respect to T-Val.

Russian does not have the means to overtly mark that validation interval for boundedness. The difference between iterativity and habituality can be stated in terms of the boundedness-feature of their respective validation-interval which is closed-bounded in the former, but

149Chapter 4: Semantics

open-bounded in the later case68. The potential reading has the same characteristics as the habitual reading and is enforced by contextual factors, like, e.g. manner-adverbials or certain verb phrases (Glovinskaja 1982: 54-60). 4.65 iterative reading69

The boundary with the single events is closed-bounded, as the default case here is taking the whole events as such and iterate them ( 4.66a). Since with iterativity the number of events is concretely specifiable, the validation interval is closed, which in turn means that TO is retrospective. The iterative reading is also possible with the basic relation ‘TT ⊂ ϕdyn’. In this case, it is only the dynamic phase that is mapped onto the topic time intervals that are open-bounded, compare ( 4.66b). The validation time interval remains closed-bounded, however. What is interesting with such repetitive events is that Russian uses the ipf aspect. This is triggered by the plurality of events that stretches over a certain interval of time. As this interval extends in time, it is similar to phases that are part of the verb’s semantic representation. This constitutes a suitable input for aspect to apply even in cases where the predicate does not provide a phase as in ( 4.66c,d). Plurality of events, specifically iteration, takes a time span to occur and this induces the necessary phase for ipf to apply, at least in Russian (cf. footnote 69). Ipf therefore can be applied on any verb – as long as world knowledge supports that. 4.66 a. Emu dvaždy otkazyvali. him twice reject:PST:ipf [Padučeva 1996: 26] ‘They rejected him twice.’

b. On vyigryval gonku tri raza, no každyj raz čto-to he win:PST:ipf race:ACC three times, but every time something slučalos’ i on ne uspeval. happen:PST:ipf and he not succeed:PST:ipf

68 This is mainly based on evidence from Bulgarian that uses the Aorist in the former, the Imperfect in the

latter case (section 4.3). The topic time intervals projected from the single events are not decisive. 69 Iterativity in Slavic languages is a rather complex topic and cannot be treated extensively in this thesis. It is

not always the case that iterativity requires the ipf aspect. Languages like Czech, Slovene or Sorbian for instance may use both pf and ipf to convey iterativity. Aspect choice then depends on the Aktionsart: with <τ>-predicates the pf aspect is used, with <ϕdyn, τ>-predicates both aspects are possible (cf. Breu 2000). Mønnesland (1984) offers a typological overview of aspect use in Slavic for the ‘frequentative habitual’ (repeated events that are not specified for the concrete number of occurrences). He argues that in this case a conflict arises between aspect on the VP level and aspect on the sentence level because the respective adverbials (usually, often, sometimes, etc.) modify the whole sentence and not only the VP (cf. the remarks on boundedness in section 4.1.2). This conflict is solved differently in Slavic languages. In cases where the ipf takes precedence over the pf (as in Russian), an aspectual transformation takes place (Mønnesland 1984: 55) and the stativeness of the iteration takes precedence over the totality of each individual event (1984: 72). In languages where the pf is retained (e.g. Czech), no such transformation takes place.

I(e)

I(TT)

[ ] e

[ ]

[ ] e

[ ]

[ ] e

[ ]

[ ] e

[ ]

I(T-Val) [ [ ] [ ] [ ] [ ] ]

150 Chapter 4: Semantics

‘Three times he was winning the race, but everytime something happend and he failed.’

c. On neskol’ko minut stučal v okno. he some minutes knock:PST:ipf at window ‘He knocked at the window for several minutes.’

d. On vyigryvral pervenstvo ot 2001 do 2004 goda. he win:PST:ipf championship from 2001 to 2004 year ‘He won the championship from 2001 to 2004.’

The iterative reading arises if some indication of frequency is present or if the duration suggests the reference to several clearly distinct events. Iteration is based on the multiple occurrence of individuated events.

This is different for the habitual and potential readings. These are also based on repetitive events which, however, do not need to be clearly distinct or individuated. It is not a matter of single events that are individually introduced into the discourse, but a matter of a number of events that are used to characterise some topic entity. The semantic configuration is illustrated in ( 4.67): 4.67 habitual and potential reading The open-boundedness of the topic time intervals projected by the single events indicates that these interpretations are not about single, distinct events but about some characteristics arising from the sum of them. Moreover, the open-boundedness of I(T-Val) indicates that the asserted interval does not include its boundaries. These readings are illustrated in ( 4.68): ( 4.68a) gives rise to an habitual interpretation, ( 4.68b) to a potential interpretation. 4.68 a. Šapki togda nosili reže, čem sejčas.

hats then wear:PST:ipf more.seldom, than now [Glovinskaja 2001: 201] ‘Then, people wore hats less often than nowadays.’

b. Rebenok očen’ rano govoril, chodil i plaval. child very early speak:PST:ipf, walk:PST:ipf and swim:PST:ipf ‘Very early, the child could speak, walk and swim.’ To sum up, the three notions ipf aspect, [±bounded] and [±terminative] are independent of each other. The ipf aspect may be used with bounded and unbounded situations (specified by TT or T-Val) and with terminative and aterminative predicates alike. Boundedness of a situation, as it has been called in the beginning, pertains to the topic time. The distinction between unbounded, open-bounded and closed-bounded TT-intervals facilitates a more fine-grained classification of the readings analysed here (Table 4.3). In Russian, the feature [±terminative] is specified by the predicate, whereas [±bounded] is not explicitly encoded on

I(e)

I(TT)

[ ] e

] [

[ ] e

] [

[ ] e

] [

[ ] e

] [

I(T-Val) ] ] [ ] [ ] [ ] [ [

151Chapter 4: Semantics

the verb, but is induced by means of temporal adverbials, further context or discourse mode (rečevoj režim, Padučeva 1996).

Table 4.3 illustrates that one single configuration of aspect, boundedness and terminativity may give rise to several readings. The readings shown in Table 4.3 are only a subset of readings proposed for the ipf aspect, but regardless of how many and which readings might be postulated for the ipf aspect, they all derive from these configurations.

ipf aspect in Russian terminativity boundedness70 reading relation example

1 [±t] open-bounded actual-processual, conative TT ⊂ ϕdyn ( 4.56)

2 [±t] unbounded permanent, inactual71, atemporal TT = e ( 4.64) si

ngul

ar

3 [±t] closed-bounded general-factual, durative TT ⊃ e ( 4.58),

( 4.60)

1 [±t] open-bounded habitual, potential T-Val = er ( 4.68)

repe

titve

2 [±t] closed-bounded iterative T-Val ⊃ er ( 4.66)

Table 4.3 Let us now have a look at the pf aspect. The defining feature for pf is the inclusion of a boundary within TT. Consequently, I(TT) cannot be unbounded. The central combination for pf is a terminative predicate in a situation that is presented as bounded. This gives rise to the event reading:

4.69 event reading

In this case, TT includes the selected part of the event, that is, the boundary. Consequently, the boundary of the event is asserted within the time of assertion. This may be interpreted as totality of the event, or as completedness. In fact, two different boundaries are involved, the inherent, potential boundary of the terminative verb and the actual boundary induced by the TT. While the former is asserted to hold, the latter is an implicature, albeit a very strong one.

70 Remember that boundedness is a layered notion, and therefore the boundedness of the situation referred to

by an utterance is specified at the outermost level – at the level of I(TT) with single events, or in case of repetitive events at the level of I(T-Val). ‘Boundedness’ in Table 4.3 and Table 4.4 thus may refer to I(TT) or I(T-Val) depending on whether the utterance is about a single or a repetitive event.

71 Note that with the inactual and permanent readings, the topic time interval is bounded by the topic entity the predicate refers to. As atemporal states of affairs hold for entities whose life span cannot reasonably be limited, atemporal interpretations are based on unbounded topic time intervals. This is the only case where I(TT) is unbounded. The ‘=’ in the representation of the basic aspectual relation is to indicate that aspect does not focus upon one specific part of the Aktionsart, which suggests that this relation does not impose by itself boundaries on the topic time. Boundaries on TT are rather introduced via the life span of the entity that is characterised.

[ ]

[ [ ϕdyn τ ϕstat

I(e)

I(TT)

152 Chapter 4: Semantics

4.70 a. Ja vstal rano, sdelal zarjadku i prinjal duš. I get.up:PST:pf early, do:PST:pf exercise:ACC and take:PST:pf shower ‘I got up early, did some exercise and took a shower.’

b. On vošel v komnatu, otkryl okno i sel rabotat’. he come.in:PST:pf in room:ACC, open:PST:pf window and sit.down:PST pf work:INF ‘He came into the room, opened the window and sat down to work.’ The respective dynamic phase is not asserted, but constitutes the presupposition in order for the assertion to be valid. With the event reading, the consequent state is an implicature. Thus, in ( 4.70b) the state of the window being open is not asserted within TT, although it can be inferred from the pf otkryt’ (‘to open’). It can, however, be part of the topic time, as is the case with the perfect reading, which is based on the configuration sketched in ( 4.71): 4.71 perfect reading72

4.72 a. Vy otkryli okno, i poėtomu v komnate cholodno. you open:PST:pf window, and therefore in room:LOC cold ‘You have opened the window, and therefore it’s cold in here.’

b. Ja poterjal ručku i pišu teper’ karandašom. I loose:PST:pf pen:ACC and write:PRS:ipf now pencil:INST ‘I have lost my pen and now I write with a pencil.’

c. On zabolel i ne poseščaet lekcij. he PRF:get.ill:PST:pf and not attend:PRS:ipf lectures ‘He got ill and does not attend lectures.’

The perfect reading is possible with all initiotransformative verbs (cf. example 4.72c): “[U] vsech glagolov SV c «nepogašennym» inceptivnym komponentom leksičeskoe značenie vključaet takže perfektnyj komponent, kotoryj v odnych kontekstach daet perfektnoe značenie SV, a v drugich podavljaetsja.” (Padučeva 1996: 154) [With all pf verbs that specify an inceptive component, the lexical meaning includes also a perfect component, which in some contexts gives rise to the perfect reading of the pf aspect, but is suppressed in other contexts.]

The perfect reading of the pf aspect, ( 4.73a), and the existential/experiential version of the imperfect of isolated events, ( 4.73b), are rather close in meaning as both imply some resultant state that is valid at TU: 4.73 a. Ja uže zapolnil anketu. Ona nachoditsja u dežurnogo.

I already fill.out:PST:pf form:ACC it is at officer.on.duty:GEN [Rassudova 1984: 62] I have already filled out the form. It is at the officer’s on duty.

b. Ja uže zapolnjal anketu. Začem že ešče raz? I already fill.out:PST:ipf form:ACC why again [Rassudova 1984: 62] I have already filled out the form. Why again?

72 The difference between the perfect reading and the Perfect tense is that the former specifies only one topic

time and not two (cf. the remarks on the experiential subgroup of the general-factual reading above).

[ [ ϕdyn τ ϕstat

[ [

I(e)

I(TT)

153Chapter 4: Semantics

Both differ, however, in the nature of this resultant state. In case of the pf aspect it is the lexically encoded resultant state that is relevant at TU, whereas for the experiential ipf it is some inferred state of experience (Grønn 2003: 94). This state of experience is furthermore asserted at an additional topic time (Pančeva 2003: 284).

The semantics of the pf aspect can specify also another kind of topic time. Similar to the perfect reading, the topic time interval is semi-open, but to the left: 4.74 pluperfect reading 4.75 a. V vosem’ časov ona uže vyšla. at eight o’clock she alread go.away:PST:pf ‘At eight o’clock she had already left.’

b. Kogda ja vernulsja, on uže ubral komnatu. when I return:PST:pf, he already tidy.up:PST:pf room:ACC

‘When I came back, he had already tidied up the room.’

In the above examples, the time adverbial and the kogda-clause specify the final boundary of the topic time interval, which gives rise to the pluperfect reading. This reading is enforced by the particle uže (‘already’).

In the third combination of Table 4.2, the pf aspect is applied to an aterminative verb that is bounded by morphological means on the verb itself. This is the case with the delimitative and perdurative readings. Note that Klein’s (1995) definition of the pf aspect in ( 4.34) does not predict this configuration, since his definition relies on lexically encoded resultant states (‘target states’, in his terminology). In case of inherently atelic delimitative/perdurative verbs, no such resultant state is encoded. The respective configuration is presented in ( 4.76): 4.76 delimitative/perdurative reading

In this case, the VP encodes the temporal boundaries of the run time of the denoted event. These temporal boundaries do not affect the aterminativity of the predicate. But still, the input requirement of pf is fulfilled by the temporal boundaries of I(e). This is one example where the temporal boundaries of the denoted event are not left to contextual reasoning but are explicitly encoded by the verb. Therefore it is not possible to refer to the internal phase of the event, cf. ( 4.77). Note that this configuration is also possible for lexically stative verbs, which then introduce individuated events into the discourse.

I(e)

I(TT)

[ [ ϕdyn

[ ]

[ [ ϕdyn τ ϕstat

] ]

I(e)

I(TT)

154 Chapter 4: Semantics

4.77 *Ja posidela v parke, kogda načalsja dožd’. I PRF:sit:PST:ipf in park:LOC, when begin:PST:pf rain.

‘I sat in the park, when it began to rain.’ This illustrates the claim of Bickel’s (1996) ‘Aspectual Uniformity Hypothesis’ (section 4.1.1): it does not matter for pf where the boundary comes from. Inherent and actual (temporal) boundaries are treated alike.

Imposing closed boundaries on the event time interval with terminative predicates conveys terminated, but not completed processes: 4.78 On polčasa počital gazetu (i ušel). he half.an.hour PRF.read:PST:pf paper:ACC (and go.away:PST:pf) [Zaliznjak & Šmelev 1997: 94]

‘He read the paper for half an hour (and went away).’ The difference between termination and completion is expressed more systematically in Bulgarian, namely by means of the ipf Aorist (section 4.3).

A more complicated case is the combination of a terminative verb with an unbounded topic time interval. An unbounded topic time interval with the pf aspect is only possible in the (morphological) present tense, which is defined by the time of utterance. Consequently, the assertion time interval cannot be closed-bounded (this would locate the position of the speaker outside the interval), cf. ( 4.79): 4.79 potential and future reading

In ( 4.79), the topic time interval is open-bounded but the verb is terminative. With single events this is only possible with the morphological present tense, as is illustrated in ( 4.80): 4.80 a. On poimet menja. he understand:PRS:pf me ‘He will/can understand me.’

b. Podnjat’ li tebe takuju tjažest’? lift:INF:pf Q you:DAT such:ACC weight:ACC? [Glovinskaja 2001: 215] ‘Will you/are you able to lift such a weight?’

Morphologically present pf predicates also express future time reference, where the taking place of the denoted event is presented as mere possibility. Both the potential and the future tense reading are closely related in expressing modality.

The constallation of terminative predicates and open-bounded situations is encountered also in case of the exemplifying and potential readings, that both characterise repetitive events:

I(e)

I(TT)

[ [ ϕdyn τ ϕstat

] [

155Chapter 4: Semantics

4.81 exemplifying and potential reading 4.82 a. Pridet on, byvalo, k nam, sjadet i načnet rasskazyvat’

come:PRS:pf he, usually, to us, sit.down:PRS:pf and begin:PRS:pf tell:INF:ipf čto-nibud’. something [Pul’kina & Zachava-Nekrasova 1995: 340] ‘Usually, he would come to us, sit down and begin telling us something.’

b. On rešit ljubuju zadaču. he solve:PRS:pf any:ACC exercise:ACC [Glovinskaja 2001: 215] ‘He will solve/is able to solve any exercise.’

The pf aspect is possible with repetitive events in the past as well, but this requires adverbials indicating the number of iterations (summary reading): 4.83 summary, iterative reading

4.84 a. Aleksej mignul pjat’ raz.

Aleksej wink:PST:pf five times [Glovinskaja 1982: 67] ‘Aleksej winked five times.’

b. On postučal tri raza korotko i rešitel’no. he knock:PST:pf three times shortly and resolutly ‘He briefly and resolutly knocked three times.’

For the pf aspect, an open-bounded topic time interval is possible only in the (morphological) present tense. The combination of the pf aspect with an aterminative verb in a situation that is presented as unbounded or open-bounded is not possible at all, since there are no boundaries that could serve as input for the pf aspect. Contrary to phases in the case of the ipf aspect, boundaries cannot be induced pragmatically (Bickel 1996) but have to be marked on the verb (yielding a terminative verb, that in turn is a suitable input for pf to apply). Therefore, this combination is out.

It has been shown that, in spite of its semantic unmarkedness, the ipf aspect can be assigned a definite semantic characterisation. Aspect semantics, boundedness of the topic time interval

I(e)

I(TT)

[ ] τ

[ ]

[ ] τ

[ ]

[ ] τ

[ ]

[ ] τ

[ ]

I(T-Val) ] [ ] [ ] [ ] [ ] [

I(e)

I(TT)

[ ] τ

[ ]

[ ] τ

[ ]

[ ] τ

[ ]

[ ] τ

[ ]

I(T-Val) [ [ ] [ ] [ ] [ ] ]

156 Chapter 4: Semantics

and terminativity of the predicate combine in a systematic way to provide the semantic basis for its various interpretations. The application of the pf aspect is more restricted than that of the ipf aspect, since it crucially needs a boundary – inherent or temporal – to be present. Furthermore, the topic time interval can only be closed-bounded. With the potential/exemplifying reading, a validation time interval is provided by the morphological present tense, and is thus open bounded.

Table 4.4 summarises the combinations of pf aspect, terminativity and boundedness:

pf aspect in Russian terminativity boundedness73 reading relation example

1 [+t] closed-bounded event TT ⊃ τ ( 4.70)

2 [+t] semi-open perfect, pluperfect TT ⊃ τ ( 4.72), ( 4.75)

3 [-t] closed-bounded delimitative, perdurative TT ⊃ τ ( 4.77), ( 4.78)

4 [+t] open-bounded potential, future (morphological

present74) TT ⊃ τ ( 4.80)

5 [-t] open-bounded ∅ ∅ ∅

sing

ular

6 [-t] unbounded ∅ ∅ ∅

1 [+t] open-bounded potential,

exemplifying (morph. present)

T-Val = er ( 4.82)

repe

titve

2 [+t] closed-boundediterative, summary

(past) T-Val ⊃ er ( 4.84)

Table 4.4

In the analysis of the Russian ipf presented in this section, two decisive assumptions have been made, even though they are not supported by overt marking: the importance of the I(TT)-characteristics and the existence three specific relations. In the following sections, evidence from the aspect systems of Bulgarian and Turkish will be provided. Bulgarian (section 4.3) substantiates the claims that have been made with respect to the topic time characterisation (non-bounded, open-bounded, closed-bounded), as it has means to overtly specify TT. Turkish (section 4.4) supports the assumption of the three specific relations, since it has overt means to specify these relations. Moreover, the analysis of Bulgarian and Turkish will show that the basic semantic toolbox outlined in section 4.1 is not only applicable to Russian, but is in fact valid for cross-linguistic analyses. Section 4.5 will illustrate this more concisely by systematically comparing the three languages (including English) on the basis of the general semantic framework.

73 Cf. footnote 70. 74 The notion ‘non-past’ is not appropriate here, as non-past comprises also future tense. But with regard to

the future tense, boundedness behaves like with regard to the past tense (Cappelle & Declerck 2005: 898).

157Chapter 4: Semantics

4.3 Bulgarian The aim of this section on Bulgarian is to show that Slavic languages do not behave exactly the same with respect to aspect, but differ in certain details. Bulgarian has a rather complex aspecto-temporal system as compared to other Slavic languages. In Bulgarian, there are no restrictions on the use of pf/ipf. Both aspects can be used in all tenses. The past tense system is of special interest, since there aspect interacts with two forms that are rather controversially discussed with respect to their status within the Bulgarian tense-aspect system – Aorist and Imperfect. This section will illustrate that both forms are to be analysed as aspectual, which follows straightforwardly from the definition of aspect developed in the above chapters. Aorist and Imperfect will be argued to deliver overt means to specify the boundedness feature of the topic time interval.

4.3.1 Aspect semantics and aspect readings Aspect in Bulgarian cannot be analysed adequately without taking into account the distinctions elaborated above, namely terminativity, boundedness, and aspect. Bulgarian illustrates the need to distinguish between the two kinds of boundedness (inherent and actual) that have been applied in the analysis of the Russian aspect system. Lindstedt (1985: 163) makes use of the features ‘open’ and ‘bounded’ to describe the semantics of pf/ipf and A/I in Bulgarian, and he accounts for their interaction in terms of nested structures. Openness can be denoted by aterminative verbs, ipf and I, boundedness by terminative verbs, pf and A. Crucially, these features apply at different layers of aspectual nesting. However, Lindstedt does not distinguish between different kinds of boundedness.

The distinction between Aorist and Imperfect pertains only to the past tense, and it is a matter of debate whether they are best considered as temporal or as aspectual. Lindstedt seems to suggest that there is no need to draw this distinction (1985: 279)75: “The much-discussed problem of the semantic nature of the Aorist : Imperfect opposition in Bulgarian is solved: the opposition is temporal, and therefore aspectual; or aspectual, and therefore temporal.”

In fact, tense and aspect share certain characteristics and can be analysed within a common framework, namely in tense-relational terms. But it has been shown that both are different, and this difference may not be blurred. Both Aorist and Imperfect are relevant in the past tense only, but this is not sufficient to qualify them as purely temporal. They might rather be regarded as portmanteau-markers, encoding past-tense reference and aspectual features.

Here, the A/I-distinction will be conceived as aspectual, following, e.g., Maslov (1959) and Sell (1994). Maslov takes the A/I distinction as aspectual in a ‘wider sense’, but without actually specifying this any further (1959: 162). Sell analyses it as a subordinate aspectual distinction, whereas at the same time claiming that it dominates the pf/ipf distinction. Within his model of aspectual nesting, Lindstedt regards the A/I opposition as the outer, and the pf/ipf opposition as the inner layer of such nested structures. He bases this claim on his assumption that the pf/ipf distinction belongs to word formation, whereas the A/I distinction is inflectional (1985: 208). In terms of the structure proposed by Filip (2005; cf. ( 4.40)), this would amount to claiming that pf/ipf are eventuality type modifiers, whereas A/I belong to the level of propositional semantics. As a consequence, what is commonly referred to as pf and ipf aspect in Bulgarian is not ‘aspect’ as defined in this thesis, but rather a formal expression of (a)terminativity. What is aspectual then, is the A/I opposition. This is in fact the view adopted by Johanson (2000) who regards Bulgarian pf/ipf as purely actional concerning

75 Interesting in this respect are also the Bulgarian terms for Aorist and Imperfect: Minalo svăršeno vreme (‘past

completed tense’) for the Aorist, and minalo nesvăršeno vreme (‘past non-completed tense’) for the Imperfect, with svăršen (completed) and nesvăršen (non-completed) being the Bulgarian terms for pf and ipf (cf. 4.92).

158 Chapter 4: Semantics

only the [±terminativity] parameter, whereas pf/ipf in Russian is aspectual in that it signals [±AD] marking76. Johanson regards the Russian and Bulgarian pf/ipf oppositions as two functionally different types (2000: 142). [+AD] as expressed by the Russian pf aspect presupposes and subsumes the actional value [+terminative], and the aspectual value has scope over the actional ones. This is different in Bulgarian: “The Bulgarian-type use of perfectives and imperfectives is purely actional and concerns the [±t] parameter. Here, too [...], aspectotemporal parameters interact to the effect that the viewpoint values […] have scope over the actional ones.” (Johanson 2000: 142)

According to Johanson, this analysis is supported by the fact that +AD-markers such as the Russian pf aspect are incapable of expressing presentness but have future reference if used with the morphological present tense. The Bulgarian morphologically present pf does not display future reference. Future reference in Bulgarian has to be formed analytically by the particle šte and a nonpast verb. Even though Bulgarian present pf verbs cannot express an activity going on at the moment of speech, they only rarely denote future actions: “[F]ormy nastojaščego vremeni ot osnov soveršennogo vida [...] vystupajut kak nastojaščee vremja tol’ko za predelami konkretnogo nastojaščego i tol’ko pri naličii izvestnych special’nych uslovij ili že polučajut (v rjade slavjanskich jazykov široko, v bolgarskom – tol’ko v nekotorych slučajach) značenie buduščego vremeni.” (Maslov 1959: 239; emphasis B.S.) [The forms of the present tense based on the pf aspect appear as present tense only beyond the concrete present and only in the presence of the well-known special conditions, or receive a future tense interpretation (very commonly among the Slavic languages, in Bulgarian only under special conditions).]

According to Bickel & Nichols (to appear: 1-3), inflectional categories can be defined as those categories of morphology that are sensitive to the grammatical – syntactic or morphological – environment in which they are expressed. The inflectional character of aspect in Russian77 is indicated by the fact that it plays a role in a morphological rule, namely in future tense formation: future tense is formed periphrastically with ipf verbs, but synthetically with pf verbs, namely by their non-past form. As Bickel & Nichols emphasise, transparency of marking has nothing to do with inflection vs. derivation. The highly irregular mode aspect formation in Russian thus does not constitute a counterargument against its inflectional status.

In Bulgarian, the pf/ipf distinction does not play a role in future tense formation. Future tense is formed periphrastically with both pf and ipf verbs. Morphologically non-past pf verbs do not convey future reference, cf. ( 4.86b). Actually, in Bulgarian there is no restriction on tense that has to do with the pf/ipf distinction. The fact that there do not seem to be any syntactic or morphological rules referring to that distinction supports Lindstedt’s (1985) suggestion that pf/ipf is not an inflectional but a derivational category. This in turn questions the aspectual status of pf and ipf in Bulgarian and supports Johanson’s view of this distinction as terminativity marking and pertaining to event semantics. This is, of course, a rather strong claim, probably too strong and too far-reaching, but nonetheless worth considering. At this stage of research, however, this thesis will remain agnostic about this question.

The following examples introduce the forms that will be central for the purposes here: pf/ipf and A/I. Bulgarian non-past pf is used mainly in modal contexts ( 4.85a), or to denote iterative and potential actions ( 4.85b). Furthermore, it is used in idioms and proverbs ( 4.85c): 4.85 a. Iskam da se metna na njakoe konče i ... [Maslov 1959: 236] want:PRS that REFL jump:PRS:pf on some horse and... ‘I want to jump on some horse and...’

76 For a definition of these terms see above, section 4.1. 77 The status of Russian aspect as inflectional or derivational is a matter of controversy. According to the

assumption of two semantic levels (chapter 2, section 2.2.3), aspect here is regarded as an inflectional category, although not always overtly expressed.

159Chapter 4: Semantics

b. Električestvo! Nastineš samo edno butonče i gotovo! [Sell 1994: 20] electricity! push:PRS:pf only one button and ready ‘Electricity! You push just one little button and there you are!’

c. Toz, kojto padne v boj za svoboda, toj ne umira. [Maslov 1959: 244] that one, who fall:PRS:pf in war for freedom, he NEG die:PRS:pf ‘The one who falls in a war for freedeom never dies.’ The non-past pf aspect does not convey future reference in any of the above cases. Instead of the simple non-past pf like in Russian, ( 4.86a), an analytic form has to be used in Bulgarian, ( 4.86b): 4.86 a. Skoro, deti, u vas budet chleb. Otec i brat vam prinesut. soon, children, at you be:FUT bred. Father and brother you bring:PRS:pf

b. Skoro, deca, šte imate chljab. Babajko i batju šte vi donesat. [Sell 1994: 20] soon, children, part have:2Pl bred. Father and brother AUX you bring:PRS:pf

‘Soon, children, you will have bread. Father and brother will bring it to you.’ Johanson’s statement above raises the question whether viewpoint values are grammatically encoded in Bulgarian, and if yes, by which markers. Here, it will be argued that in the past tense these viewpoint distinctions are not only expressed by means of pf/ipf, but also by the A/I distinction. This is not as radical as Johanson’s view, but it still differs from traditional analyses and textbook assumptions which tend to treat A/I as tense.

Although there is a certain affinity between pf and the Aorist on the one hand, and ipf and the Imperfect on the other, ipf A and pf I are also used – the latter form, however, only in specific contexts: 4.87 a. Včera v 7 č. az otidoch / otivach na lekcii.

yesterday at 7 o’clock I go:A:pf / go:I:ipf on lectures [Sell 1994: 83] ‘Yesterday at 7 o’clock I went / was going to classes.’

b. Tja pisa pismoto dva časa. she write:A:ipf letter:DEF two hours [Desclés & Guentechéva 1990: 239] ‘She wrote the letter for two hours.’

c. Kakvoto rešeše, go osăštestvjavaše. whatever decide:I:pf, 3Sg realise:I:ipf [Lindstedt 1985: 191] ‘Whatever she decided, she (always) realised it.’

Example ( 4.87a) illustrates the combinations pf A and ipf I. The use of the ipf A in ( 4.87b) denotes an action that has reached its actual, but not its inherent boundary. In order to express the reaching of the inherent boundary, the pf A has to be used. The pf I is possible only in rather restricted contexts, like the one in ( 4.87c). By the pf aspect, the denoted action is presented as having reached its inherent boundary, by the Imperfect as not having reached actual boundaries. This results in an iterative or habitual interpretation. As is shown in ( 4.87b), Bulgarian distinguishes between processes that are completed and processes that are finished. Completed processes are also finished, but not vice versa78. The distinction between finished and completed processes is crucial for an understanding of the A/I opposition.

78 In fact, processes, i.e. verbs being characterised by a dynamic phase, can only be finished. The distinction

finished vs. completed is relevant only for finitransformative verbs.

160 Chapter 4: Semantics

To compare Bulgarian and Russian pf/ipf, Sell (1994: 26) presents the following table79:

Function Russian Bulgarian limited duration ipf pf or if unlimited duration ipf ipf limited repetition pf or ipf pf or ipf unlimited repetition ipf pf or ipf general factual ipf ipf perfect of result pf pf or ipf

Table 4.5 This table is interesting for the purposes here in that it reveals a fundamental difference between pf/ipf in Russian and in Bulgarian: Bulgarian allows for both aspects in contexts where in Russian only the ipf is possible (except for the restricted case of pf aspect in the context of limited repetition).

In case of limited duration, Bulgarian may use both the ipf aspect, as in ( 4.88a), or the pf aspect, as in ( 4.88b): 4.88 a. Tja pisa pismoto dva časa. she write:A:ipf letter:DEF two hours [Desclés & Guentchéva 1990b: 239] ‘She wrote the letter for two hours.’

b. Pregovorite ni prodălžicha dosta dălgo; naj-posle ėto kak rešichme rabotata.... negotiations our continue:A:pf rather long; at the end this how decide:A:pf work:DEF

‘Our negotiations continued for a rather long time; at the end we decided the matter like this:...’ [Sell 1994: 73] In case of unlimited duration, only the ipf aspect is possible, compare ( 4.89): 4.89 a. Ženata dălgo mălčeše. woman:DEF long.time be.silent:I:ipf ‘The woman was silent for a long time.’

b. Sam ostanal prost, Marko običaše učenieto. himself remain:P:ipf uneducated, Marko like:I:ipf education:DEF [Desclés & Guentchéva 1990: 252] ‘Though he himself remained uneducated, Marko loved education.’ The decisive difference between ( 4.88a) and ( 4.89a,b) consists in that in the former case the initial and the final points (i.e. the boundaries) of the intervals are included in what is asserted (hence limited duration), but excluded in the latter (hence unlimited duration). The ipf aspect does not select an inherent boundary, but, as will become evident below, the Aorist imposes external boundaries. The Imperfect confirms the non-boundedness of the ipf aspect by marking the asserted interval as open-bounded. If the pf aspect is used, it selects a boundary. Its combination with the Imperfect and the resulting exclusion of temporal boundaries from the asserted time render the event iterative. In case of unbounded duration, thus, only the ipf aspect is possible.

79 This table is taken from Sell (1994), but leaves out some functions that are not discussed here, e.g.

‘dissociation from the moment of speech’. Omitted are furthermore the notions of sequentiality and simultaneity, which are conveyed with pf and ipf in Russian, whereas the pf/ipf distinction is not relevant for these functions in Bulgarian.

161Chapter 4: Semantics

In cases of limited, ( 4.90), and unlimited, ( 4.91), repetition, Bulgarian again may use both pf and ipf aspect: 4.90 a. Učitel Nonin povtori mnogo păti taja fraza.

teacher Nonin repeat:A:pf many times this phrase. [Maslov 1959: 255] ‘Teacher Nonin repeated this phrase several times.’

b. Stefan digna pistoleta si i strelja dva păti văv văzducha ... Stefan pick.up:A:ipf pistol:DEF POSS and shoot:A:ipf two times into air:DEF

[Maslov 1959: 263] ‘Then Stefan picked up his pistol and shot two times into the air.’

4.91 a. Ivan, kogato ima lekcii, se chrani v stola. Ivan, whenever have:PRS:ipf lectures, eat:PRS:ipf at canteen [Radeva 2003: 111] ‘Whenever Ivan has classes, he eats at the canteen.’

b. Sedneše djado v kăta i zapeeše njakoja pesen. sit:I:pf grandpa in korner:DEF and sing:I:pf some song [Sell 1994: 79] ‘Grandpa would sit in the corner and sing some song.’ These examples show that the notions ‘limited’ and ‘unlimited’ in fact pertain to the whether the denoted interval is closed-bounded (i.e. limited) or open-bounded (e.g. unlimited).

Table 4.6 summarises these findings, including the contributions Aorist and Imperfect80:

Function Russian Bulgarian limited duration ipf A pf or if unlimited duration ipf I ipf limited repetition pf or ipf A pf or ipf unlimited repetition ipf I pf or ipf general factual ipf A ipf perfect of result pf P pf or ipf81

Table 4.6 The comparison of Russian and Bulgarian with respect to the distinction limited vs. unlimited shows that in this respect it is not the pf/ipf opposition that is decisive in Bulgarian. Limitedness and unlimitedness are expressed by Aorist and Imperfect, respectively. Since limitedness and unlimitedness have been shown to correspond to the closed-boundedness vs. open/non-boundedness of the topic time, Aorist and Impferfect encode the boundedness characteristic of I(TT). Russian cannot express this feature morphologically. It rather arises as default inference from the use of the respective aspect form (pf is bounded per default) or by means of adverbials or discourse context. Russian pf and ipf cover more functions than the Bulgarian counterpart, as Bulgarian grammatically encodes the difference between ‘limited’ and ‘unlimited’ situations by means of A/I. In both limited and unlimited situations, pf and ipf can be used depending on the terminativity feature of the predicate. Russian is not sensitive to this distinction. It treats all extended events – durative or repetitive – alike, be they limited or unlimited. The A/I distinction in Bulgarian seems to dominate the pf/ipf distinction, and the question is how both interact to convey the notion of aspect.

80 Cf. footnote 79. 81 This shows that the aspect sytem of Russian compensates quite a lot of temporal relations. Remember that

in a given context, the Russian pf aspect may not only have a perfect, but also a pluperfect interpretation. Bulgarian has overt forms for both tenses, but they will not be dealt with in this thesis.

162 Chapter 4: Semantics

In the remainder of this section it will be shown, that Aorist and Imperfect indeed specify the boundedness characteristic of the topic time interval, whereas pf and ipf are responsible for the selection part.

The semantics of Aorist and Imperfect is described in grammars of Bulgarian as follows: 4.92 a. Imperfect (minalo nesvăršeno vreme) To se svărzva s označavane na dejstvie, koeto se vărši v edin minal moment i se predstavja kato proces na izvăršvane v minaloto. (Georgiev 1999: 317)

[It is related to the expression of an action, that goes on at one moment in the past and is presented as ongoing process in the past.]

b. Aorist (minalo svăršeno vreme) Aorist [ima] značenie za protičane na dejstvieto kato cjalosten proces nezavisimo ot vida na glagola. (Georgiev 1999: 322)

[The Aorist denotes an action as completed process independently of the aspect of the verb.]

In order to analyse the semantics of the Bulgarian aspect system in the past tense, the four combinations ipf I, ipf A, pf I and pf A have to be distinguished. In ( 4.93)-( 4.95) the semantics of these combinations is shown in terms of interval-intersection analogously to the analysis provided for Russian. The aim is to figure out how these Russian relations are expressed in Bulgarian, and whether this confirms the hypotheses about the Russian system. Note that just as in Russian, one form may give rise to different configurations. 4.93 ipf Imperfect

a. TT included in the dynamic phase: TT ⊂ ϕdyn

b. I(e)[i j[ ∩ I(TT)]m n[ → I(TT)]m n[ = {ϕdyn, m < ϕdyn < n}

a’. TT simultaneous with phases and boundaries: TT = ϕdyn τ ϕstat

b’. I(e)[i j[ ∩ I(TT)]m n[ → I(TT) = e

[ [ i ϕdyn τ ϕstat

j AKTIONSART I(e)

] [ m

n TOPIC TIME I(TT)

ASPECT

[ [ i ϕdyn τ ϕstat j

AKTIONSART I(e)

TOPIC TIME I(TT)

ASPECT

163Chapter 4: Semantics

4.94 ipf Aorist

a. TT includes phases and boundaries: TT ⊃ ϕdyn τ ϕstat

b. I(e)[i j] ∩ I(TT)[m n] → I(TT)[m n] = {e, m ≤ e ≤ n}

The Bulgarian forms ( 4.93)-( 4.94) correspond to the three possible relations for the Russian ipf aspect. To express the relation basic to the Russian general-factual reading, Bulgarian uses the ipf Aorist. This suggests that it is the Aorist that is responsible for the closed-boundedness of I(TT). It furthermore illustrates the semantic unmarkedness of both ipf and Aorist (Breu 2003 makes a similar point for Molisean Slavic).

The ipf Aorist gives rise to another relation that cannot be expressed by grammatical means in Russian82:

a’. TT includes the dynamic phase: TT ⊃ ϕdyn

b’. I(e)[i j[ ∩ I(TT)[m n] → I(TT)[m n] = {ϕdyn, m ≤ ϕdyn ≤ n}

This relation imposes explicit external boundaries onto a process. If this process is terminative, as is the case here, it is asserted to be finished without being completed. This relation has also been analysed in terms of ‘neutral aspect’, a consequence arising from not properly taking into account the contribution of the Aorist. The ipf Aorist will be dealt with in more detail in section 4.3.2.

The next combination, the pf Aorist, is only briefly mentioned here as it will not play a role in the further considerations, since the focus here is on the ipf aspect: 4.95 pf Aorist

a. TT includes the boudary: TT ⊃ τ

b. I(e)[i j[ ∩ I(TT)[m n] → I(TT)[m n] = {τ, m ≤ τ ≤ n}

82 An exception are delimitative terminative VPs that express this relation. Cf. ( 4.78) in section 4.2.2.

[ [ i ϕdyn τ ϕstat

j AKTIONSART I(e)

[ ] m

n TOPIC TIME I(TT)

ASPECT

[ [ i ϕdyn τ ϕstat j

AKTIONSART I(e)

[ ] m

n TOPIC TIME I(TT)

ASPECT

164 Chapter 4: Semantics

The forth combination, the pf Imperfect, is possible only in restricted contexts mainly in case of subordination. It then expresses unlimited repetition and therefore it is not listed among the relations mentioned above, since they are all basic to simple events.

In what follows, the Bulgarian equivalents of the Russian ipf readings mentioned in section 4.2.3 will be analysed along the lines proposed for Russian.

The actual-processual and the conative reading can be represented by one configuration, compare ( 4.96). The only difference is that the latter is more probable with terminative verbs that provide the basis for the conative inference, while the aspectual characterisation is identical to that of the actual-processual reading.

4.96 actual-processual and conative reading 4.97 a. Kogato tja izlizaše ot stajata, sreštna edin poznat. when she leave:I:ipf from room:DEF, meet:A:pf one friend [Lindstedt 1985: 137] ‘When she was leaving the room, she met a friend.’

b. Vzech li gi? Ne, ne gi vze. Ama davaše gi čovekăt. take:A:pf.2Sg Q 3Pl No, no 3Pl take:A:pf:3SG but give:I:ipf 3PL man

‘Did I take it? No, you didn’t. But the man wanted to give it.’ [ibd: 220]

c. Kakvo šte praviš v CUM? Šte kupuvam fotoaparat. what AUX do:PRS:ipf in CUM? PART buy:PRS:ipf camera [ibd: 220] ‘What are you doing in the CUM? I’m going to buy a camera.’ The selected part of the Aktionsart is the dynamic phase, which is related to the topic time. The topic time interval is characterised by a dynamic phase and it is open-bounded. This open-boundedness is contributed by the Imperfect, as becomes obvious by comparing the ipf A in ( 4.98), ( 4.100) and ( 4.102), where the Aorist imposes closed-boundaries on I(TT).

Probably the most interesting case is that of ‘semi-perfectivity’83 illustrated in ( 4.98): 4.98 semi-perfectivity

83 The notion of ‘semi-perfectivity’ will become clear in section 4.3.2.

[ [ i ϕdyn τ ϕstat j

AKTIONSART I(e)

[ ] m

n TOPIC TIME I(TT)

ASPECT

[ [ ϕdyn τ ϕstat

] [

I(e)

I(TT)

[ [ ϕdyn τ ϕstat

[ ]

I(e)

I(TT)

165Chapter 4: Semantics

4.99 a. Tja pisa pismoto dva časa. she write:A:ipf letter:DEF two hours

‘She wrote the letter for two hours.’

b. Az tărsich za moite deca kunduri, no ne namerich nikakvi. I look.for:A:ipf for my children shoes, but not find:A:pf none

‘I looked for shoes for my children, but didn’t find any.’ [Lindstedt 1985: 219]

c. Toj caruva trijset godini. he reign:A:ipf thirty years [Lindstedt 1985: 175] ‘He reigned for thirty years.’ As the ipf A will be dealt with in more detail in section 4.3.2, these examples are merely intended to illustrate its interpretation. The expression of ‘semi-perfectivity’ is only possible with terminative verbs, as is illustrated in ( 4.99a,b). The ipf A marks <ϕdyn, τ>-predicates as denoting finished processes that are not completed. In case of ( 4.99a,b) the writing of the letter and the looking for shoes are imposed actual boundaries (by means of the Aorist) that do not correspond to the internal boundaries. These are marked as not being reached (by means of the ipf aspect). With aterminative verbs, as in ( 4.99c), the ipf A most probably receives a durative reading (cf. 4.101).

In Russian, the durative reading (limited duration) is expressed by the ipf aspect combined with adverbials of the type ot...do or dolgo, or by means of context. In Bulgarian, this reading is conveyed by the ipf Aorist as illustrated in ( 4.101a). The ipf I with durative adverbials is not possible, ( 4.101b), expect for the habitual reading (Lindstedt 1985: 184). This is predicted by the semantics of the Imperfect, that is characterised as indicating a non-bounded or open-bounded interval. 4.100 durative reading

4.101 a. Njakoe vreme dvamata mălčacha, posle .... some time both be.silent:A:ipf, and then [Lindstedt 1985: 178] ‘For a while, both were silent, and then ...’

b. #Tja peeše pesen tri minuti. she sing:I:ipf song three minutes [ibd: 184] *‘She sang a song for three minutes.’ ‘She used to sing a song for three minutes.’

In case of the durative reading, the topic time interval is characterised by a dynamic phase and closed boundaries. These are due to the Aorist, the phase stems from the ipf.

The ipf A may also give rise to the general-factual reading as illustrated in ( 4.102): 4.102 general-factual reading

[ [ e

[ ]

I(e)

I(TT)

[ [ ϕdyn τ ϕstat

[ ]

I(e)

I(TT)

166 Chapter 4: Semantics

4.103 a. Chodiš li na učilište? Chodich do četvărti klas. Sega rabotja vkăšti. go:PRS:ipf Q to school? go:A:ipf until fourth class. Now work:PRS:ipf at home. [Maslov 1959: 268] ‘Do you go to school? – I went to until the fourth class. Now I work at home.’

b. Nali metoch, a sega pak e mrăsno! PART sweep:A:ipf, but now again be:PRS:3Sg dirty [Radeva 2003: 117] ‘I swept, but now it is swap again and dirty!’

c. Ivan dnes dochožda. Ivan today come:A:ipf [Radeva 2003: 117] ‘Ivan came today.’ (= He was here, but now he’s gone.)

As has been pointed out above, the notion ‘general-factual’ is adopted from Russian, where it actually comprises rather different subtypes (cf. Grønn 2003; Padučeva 1996) some of which are expressed in Bulgarian by means of the Perfect. These subtypes are not regarded here, since the term ‘general-factual’ is used only to refer to the denotative use exemplified in ( 4.103).

The inactual, permanent and atemporal readings (unlimited duration) are expressed by ipf I in Bulgarian. On the surface, this combination is identical to the actual-processual and conative readings. The difference lies in the boundedness of the topic time interval: for the actual-processual and conative readings, it is open-bounded, for the inactual, atemporal and permanent reading it is unbounded (and may be bounded by means of context). A left boundary may arise via use of the past tense, which gives rise to the so-called lifetime effect (Musan 1997). 4.104 inactual, permanent and atemporal reading 4.105 a. Toj raboteše v universiteta.

he work:I:ipf at university. He worked at university. (= He was a teacher.)

b. Togava visoko go cenjacha. then high 3Sg:ACC appreciate:I:ipf [Radeva 2003: 118] ‘At that time he was highly appreciated.’

c. Afrika se namira južno ot Sredizemno more. Africa be.located:PRS:ipf south of Mediterranean Sea. [Radeva 2003: 111] ‘Africa is located to the south of the Mediterranean Sea.’

In none of the examples in ( 4.105) is an event introduced into the discourse. The predicate in ( 4.105a) would in principle supply a dynamic phase to be selected, but in this specific context no such selection occurs. The ipf is rather used as status-aspect and thus characterises the topic entity toj (‘he’) as someone who worked at the university, i.e. as somebody who was a teacher there. For the activation of the respective aspect relation, a certain amount of encyclopaedic knowledge is required. The validity claim of this assertion is bounded to the right by the time of utterance. The bounding mechanism is the same in ( 4.105b). Contrary to ( 4.105a), the relation ‘TT = e’ in ( 4.105b) does not need additional inferences to arise, as the predicate is stative and thus does not provide a unit to be selected. ( 4.105c) is an example of

[ [ ϕdyn τ ϕstat

I(e)

I(TT)

167Chapter 4: Semantics

the atemporal reading (hence non-past) where the assertion is not bounded at any level. Constraining the validity claim in this case would not yield a reasonable interpretation.

As has been mentioned above, Bulgarian marks the distinction between limited and unlimited repetition by means of the Aorist, which induces a closed-bounded topic time interval, and of the Imperfect, which induces an unbounded or open-bounded topic time interval. Unlimited repetition, ( 4.106), will be subsumed under the heading of ‘habituality’ thereby distinguishing it from iterativity, i.e. limited repetition, on the one hand and genericity84 characterised by a non-bounded TT-interval on the other hand. 4.106 unlimited repetition: habitual and potential reading Unbounded repetition is expressed in Bulgarian by means of the ipf present as in ( 4.107a) or the ipf Imperfect as in ( 4.107b). The pf I may be used to add a shade of potentiality to habituality (Sell 1994: 83) as in ( 4.107c). This also holds for uses of the pf aspect in the present tense, ( 4.107d). These latter two represent Lindstedt’s nested structure open(bounded) ‘(O(B))’. The Imperfect and the present tense contribute the non- or open-boundedness of TT (‘outer boundedness’), whereas the pf aspect emphasises the reaching of the inner boundedness in Lindstedt’s terms (‘inherent boundedness’).

The use of pf or ipf decides on how the respective events are mapped onto the TT-interval, it specifies the kinesis-feature of TT: with the pf aspect, the boundary is asserted for each event, with the ipf aspect either the processes are highlighted or the events are stated as such85. The decisive feature for the habitual/potential reading to arise is open-bounded I(T-Val) specified by the Imperfect. 4.107 a. Ivan, kogato ima lekcii, se chrani v stola. Ivan, whenever have:PRS:ipf lectures, eat:PRS:ipf in cafeteria [Radeva 2003: 111] ‘Whenever Ivan has classes, he eats in the cafeteria.’

b. Pisarjat ne beše pušač, no segiz-togiz vzemaše po cigara v usta. clerk NEG be:I:ipf smoker, but sometimes take:I:ipf cigar into mouth [Maslov 1959: 56] ‘The clerk was not a smoker, but sometimes he took a cigarette between his lips.’

c. I kato si legnechme večer da spim, toj započvaše: and whenever lay.down:I:pf evening to sleep, he begin:I:ipf:

znaeš li kolko e chubav Pirin! know:PRS:ipf Q how be:PRS:ipf beautiful Pirin [Lindstedt 1985: 189]

‘And every night when we went to bed, he would begin: “Do you know how beautiful the Pirin (mountains) are?” ’

84 I am well aware of the fact that ‘genericity’ is a rather controversial notion among linguists. Here it is used

to denote interpretations of aspecto-temporal forms that do not refer to an actual situation but rather a characterisation of some topic entity.

85 Cf. footnote 69.

I(e)

I(TT)

[ ] e/τ

] [

[ ] e/τ

] [

[ ] e/τ

] [

[ ] e/τ

] [

I(T-Val) ] ] [ ] [ ] [ ] [ [

168 Chapter 4: Semantics

b. Po cjal den ne se pribira (ipf) – sutrin izleze (pf), through whole day not come.home:PRS:ipf – morning go.out:PRS:pf, večer se vărne (pf) evening return:PRS:pf [Lindstedt 1985: 193]

‘During all the day, he does not come home – he goes out in the morning and comes back in the evening.’ The term ‘iterativity’ will be restricted to instances of what Sell calls ‘limited repetition’, which crucially needs some indication of how many times an action took place. This also includes cases such as njakolko păti (‘several times’) or mnogo păti (‘often’), but excludes često (‘often’) or ponjakoga (‘sometimes’). These latter adverbials are in fact possible, but, according to Lindstedt (1985: 202), they are usually ungrammatical with the Aorist, which is used to express bounded repetition. 4.108 iterative reading The remarks on the choice of pf or ipf aspect that have been made for the habitual reading hold here as well. The crucial difference is that with iterativity, i.e. the repetition of discrete, single events, the Aorist is used since this form imposes closed boundaries onto the validation time-interval. Aspect choice depends on the terminativity-feature of the Aktionsart basis. 4.109 a. Poveče ot desetina păti toj pravi / napravi tova. more than dozen times he do:A:ipf / do:A:pf this. [Andrejčin 1978: 191] ‘He did this more than a dozen times.’

b. Nejko izprazva / izprazni i napălva / napălni disagite njakolo păti.86 Nejko empty:A:ipf / empty:A:pf and fill:A:ipf / fill:A:pf saddlebads several times. [ibd.] ‘Nejko emptied and filled the saddlebags several times.’

Table 4.7 summarises the comparison of ipf readings in Russian and the Bulgarian equivalents. Similar to Russian, it is not always easy to decide which reading is intended in a certain context. As will be argued in chapter 5, the naming of one interpretation as inactual or permanent or atemporal is not decisive. The crucial fact is that all the readings that are possible in a certain context share the basic aspectual characterisation.

The data in Table 4.7 suggest a division of the Russian ipf aspect into two groups according to different TT-characteristics – closed-bounded vs. non-bounded/open-bounded. Bulgarian distinguishes closed-boundedness from open- and non-boundedness by means of the A/I-opposition. The data in both Russian and Bulgarian also suggest a distinction between non-bounded and open-bounded TT-intervals, but this distinction manifests itself only in context and not by overt marking. This gives rise to the question whether there is a language

86 The ipf forms here are derived imperfectives used to convey iteration (Desclés & Guentchéva 1990: 249f).

Derived imperfectives of this type cannot express an actual-processual meaning (Sell 1994; Lindstedt 1985).

I(e)

I(TT)

[ ] e/τ

[ ]

[ ] e/τ

[ ]

[ ] e/τ

[ ]

[ ] e/τ

[ ]

I(T-Val) [ [ ] [ ] [ ] [ ] ]

169Chapter 4: Semantics

that encodes all three kinds of boundary-characteristics of ipf. In the next section, Turkish will be shown to be such a language.

Reading Russian Bulgarian

1 actual-processual ipf ipf I 2 conative ipf ipf I 3 inactual/continuous ipf ipf I 4 durative ipf ipf I; ipf A 5 general-factual87 ipf ipf A 6 iterative ipf ipf I; ipf A 7 habitual ipf ipf I 8 potential ipf ipf I 9 permanent ipf ipf; ipf I 10 atemporal ipf ipf

Table 4.7 The analysis of the Bulgarian data moreover suggests, that the pf/ipf distinction Bulgarian concerns some ‘inner’ layer of aspect composition (Johanson 2000; Lindstedt 1985; Sell 1994). Sell (1994) takes this as evidence for the fact that the pf/ipf distinction in Bulgarian is not grammatical at all, but lexical. As one piece of evidence for that claim he adduces the fact that with durative and iterative events, Bulgarian may use both pf and ipf forms, depending on whether the emphasis is on the phase part of the basic verb or on its boundary. Durativity and repetition are expressed by the Aorist and the Imperfect, and this opposition dominates the pf/ipf opposition. This is different in Russian, where adverbials like dolgo (‘for a long time’) require the ipf aspect, irrespective of the properties of the basic event. Sell takes this as further support for his claim that pf and ipf in Bulgarian are relevant to the lexical level and function as eventuality-type modifiers, contrary to pf and ipf in Russian that are relevant at the propositional level.

However, the analysis developed in this section suggests the following: the pf/ipf distinction in Bulgarian concerns the selection of a temporal tier unit provided by the Aktionsart (phases or inherent/actual boundaries) and relates this selected unit to a topic time. The boundedness characteristics of the topic time interval are specified by the A/I distinction. Selection, relation and topic time are the constituents of aspect semantics. The specification of the topic time by means of the A/I distinction justifies the analysis of this distinction as aspectual.

With respect to markedness, opinions differ. Sell (1994) regards the pf/ipf distinction in Bulgarian as equipollent, whereas Maslov (1959) and Lindstedt (1985) take the pf aspect as the marked member in Bulgarian. The fact that it is the ipf that is used for the denotative use speaks in favour of the latter position. Both Maslov and Lindstedt regard the Imperfect as the marked member of the A/I opposition. Maslov supports this view by the fact that the combination of pf and I is the most specific one out of the four possible combinations, as it is realised only in very restricted contexts. Comrie (1976) is careful about the markedness relation between A and I. Although it seens to be generally assumed that in Romance and Slavic languages the Aorist is the unmarked member of the A/I opposition, he does not exclude the possibility that “we are dealing with two equally marked members of an equipollent opposition” (Comrie 1971: 114). The use of the ipf A for the denotation of a fact is probably the best argument in favour of its unmarked status compared to the Imperfect (cf.

87 To ensure comparability with Russian, however, the term ‘general-factual’ is maintained; the existential

variety is not dealt with here (cf. section 4.2.2).

170 Chapter 4: Semantics

also the argumentation in Breu 2003). Thus both ipf aspect and Aorist can be regarded as unmarked with respect to their paradigmatically opposed forms, pf aspect and Imperfect.

To sum up, aspect in Russian and Bulgarian differ in the explicitness of formal expression. Russian encodes this category by means of the pf/ipf opposition, leaving notions like the boundedness-characteristics of TT to contextual specification. Bulgarian is formally more explicit as it has means to specify the topic time interval with respect to boundedness – the Aorist and Imperfect. Sell (1994; a similar argument can be found in Johanson 2000) argues against the grammatical status of pf/ipf in Bulgarian and takes it to be a lexical category. The view advocated here is not as radical as Sell’s. Based on the general definition of aspect as establishing a relation between the run time of the selected part of the encoded event and the topic time, the data discussed above has shown that both pf/ipf and A/I contribute to the expression of ‘aspect’. Aorist and Imperfect are thus at least as much aspectual as they are temporal. These forms cannot be analysed as purely temporal, but are to be regarded as combining a temporal value (topic time before time of utterance) and an aspectual value (specification of the boundedness characteristics of the topic time interval)88.

4.3.2 Excursus: Semi-perfectivity This section deals in more detail with the combination of ipf aspect and Aorist. The ipf A is interesting in several respects: it shows that boundedness and terminativity, i.e. actual and inherent boundaries, are indeed distinct and that the boundedness of TT plays a crucial role in aspect semantics. It is possible for almost every Bulgarian verb to appear with the ipf A. It will be argued that this form cannot be analysed as ‘neutral aspect’, a term introduced by C. Smith (1997) and applied to the Bulgarian ipf A by Pančeva (1993)89. Pančeva’s analysis does not acknowledge the contribution of the Aorist as imposing closed boundaries on the topic time. It will be argued that an analysis in terms of ‘semi-perfectivity’, as proposed by Koenig & Muansuwan (2000) for a similar phenomenon in Thai, is to be preferred.

The distinction between accomplished and achieved processes introduced by Desclés and Guentchéva (1989; 1990a,b) serves as a starting point for the analysis of ipf A. Remember Desclés (1993) continuum from non-accompli over accompli and achevé to attributif (chapter 2, section 2.2.2), repeated in Figure 4.10:

Figure 4.10

Languages differ with respect to which point of this continuum they encode grammatically:

88 One might conjecture that Bulgarian presents an older stage in the development of the Slavic tense-aspect

system insofar as it preserves the Old Slavic Aorist/Imperfect distinction alongside with the more recent aspect system. This could be explained by the location of Bulgarian on the periphery of the Slavic language group and by the fact that the periphery tends to behave rather conservatively with respect to changes in the language system. A thorough investigation of the diachronic development of Bulgarian is beyond the scope of this thesis.

89 Neutral aspect is also claimed for Hindi (Singh 1998).

inaccompli = non accomplissement

accompli achevé attributif

preparatory process culmination

consequent state

171Chapter 4: Semantics

“L’opposition achèvement/absence de signification d’achèvement est à la base de l’opposition aspectuelle perfectif/imperfectif des langues slaves; l’opposition accompli/inaccompli se trouve à la base d’un grand nombre de langues, en particulier les langues sémitiques” (Guentchéva 1989: 28)

As can be seen from Figure 4.10, the difference between [±accompli] and [±achevé], i.e. the distinction between finished and completed processes, reveals itself only with <ϕdyn, τ>-predicates that consist of a dynamic phase and a boundary. Predicates of the type <ϕdyn> and <τ> do not make a difference between finishing and completing.

Whereas Russian has grammatical means to encode the feature [±achevé], Bulgarian may additionally express the feature [±accompli] by grammatical means. As will be shown in section 4.4, in Turkish the feature [±attributif] is grammatically encoded.

A predicate with the structure <ϕdyn, τ> may be presented as [±completed] in Russian90 ( 4.110), but as [±finished] and [±completed] in Bulgarian ( 4.111): 4.110 a. Ivan pročital knigu za dva časa. Ivan PRF.read:PST:pf book: ACC within two hours ‘Ivan read the book within two hours.’

b. *Ivan pročital knigu dva časa. *‘Ivan read the book (through) for two hours.’

c. Ivan čital knigu dva časa. Ivan read:PST:ipf book:ACC two hours. ‘Ivan was reading the book for two hours.’

In ( 4.110a), the book-reading is presented as being completed, the inherent boundary is reached. Sentence ( 4.110b) is not possible since the durative adverbial is incompatible with the terminative verb91. The [+accompli]-reading can be conveyed only with additional lexical means as demonstrated in ( 4.110c). Without the adverbial dva časa (or means of the further context), the sentence would not have the reading as finished process.

Bulgarian does not have to use time-span or durative adverbials to mark the difference between [+achevé] and [+accompli], compare ( 4.111): 4.111 a. Ivan pročete knigata (za dva časa). Ivan PRF.read:A:pf book:DEF (within two hours). [Guentchéva 1989: 28] ‘Ivan read the book (within two hours).’

b. Ivan čete knigata (dva časa). Ivan read:A:ipf book:DEF (two hours). [ibd.]

‘Ivan read the book for (two hours).’ The feature [+achevé] is expressed by the pf A, ( 4.111a), the feature combination [+accompli, -achevé] by the ipf A, ( 4.111b).

If the A/I distinction is not adequately taken into account, examples like ( 4.112), taken from Pančeva (2003: 283), seem to behave strangely, since they are not explainable in terms of pf and ipf alone:

90 Expect for the delimitative and perdurative readings, in case of which ‘+finished’ is explicitly encoded by

marking on the verb. 91 The reason for this incompatibility cannot only be the pf aspect, since there are pf verbs that are compatible

with such adverbials as has been shown in section 4.2.

172 Chapter 4: Semantics

4.112 a. Az stroich pjasăčna kula. I build:PST:Neut sand castle ‘I was engaged in building a sandcastle.’

b. Az strojach pjasăčna kula. I build:PST:ipf sand castle. ‘I was building a sandcastle.’

c. Az postroich pjasăčna kula. I build:PST:pf sand castle. ‘I built a sandcastle.’

As is evident from the glosses, Pančeva does not distinguish between I and A, but analyses these examples uniformly as past tense forms, the difference being one of aspect, which is neutral in ( 4.112a), imperfective in ( 4.112b) and perfective in ( 4.112c). Pančeva analyses the Bulgarian aspect system in terms of this allegedly three-way distinction. The term ‘neutral aspect’ is taken from C. Smith 1997, who defines neutral aspect as a “default with a positive semantics” (1997: 77) characterising sentences lacking a viewpoint morpheme. She bases the assumption of neutral aspect on empirical grounds, namely on the interpretation of aspectually vague sentences that are neither pf nor ipf and are more flexible than either viewpoint as they allow for open and closed readings92. Its assertion span includes the initial point and at least one internal stage of a situation. This distinguishes it from the pf viewpoint that includes two endpoints and from the ipf viewpoint that includes neither. Unlike the ipf aspect, neutral aspect allows closed readings by inference (1997: 81).

However, given what has been said about default aspect (chapter 2, section 2.2.3), and considering the analysis of aspect systems in terms of ‘scalar implicatures’ to be developed in chapter 5, the notion of ‘neutral aspect’ turns out to be superfluous.

Pančeva (1993: 283f) makes use of this notion and applies it to the Bulgarian aspect system. Following C. Smith, she characterises neutral aspect as being positioned between pf and ipf with respect to the following features:

criterion pf ipf neutral achievement of goal (with telic events) + - - durative adverbials (dva časa ‘for two hours’) - + + completive adverbials (za dva časa ‘within two hours) + - - durative + inclusive interpretation of time intervals (meždu 10 i 11 ‘between 10 and 11’) + - +

neutral sequence with perfective eventualities: t(e) < t(e’) + - +

Table 4.8 If the A/I distinction is properly taken into account, postulating neutral aspect for Bulgarian is not necessary at all93. Pančeva does not seem to consider this distinction, since the only difference she states for the examples in ( 4.112) is in terms of pf, ipf and neutral aspect. The difference between these examples and their behaviour with respect to the criteria adduced in Table 4.8 can be accounted for straightforwardly if Aorist and Imperfect are taken into account: Aorist conveys bounded situations, and, if used with ipf terminative predicates, it marks the event as being finished before having reached the inherent boundary. As it conveys

92 Note again the conflation of aspect and boundedness. 93 Moreover, Pančeva analyses a phenomenon very different from what the term ‘neutral aspect’ is originally

intended to refer to.

173Chapter 4: Semantics

finished processes (due to the ipf aspect), it is compatible with durative adverbials specifying the duration of that process, but it is not compatible with completive adverbials that would require the inherent boundary to be reached. The sequentiality interpretation arises from the induction of temporal boundaries by the Aorist. Consequently, sequentiality may also arise with the combination of the Aorist and ipf verbs. Examples ( 4.112) need to be analysed as ipf A in ( 4.112a), ipf I in ( 4.112b) and pf A in ( 4.112c).

The situation with the alleged neutral aspect resembles the one with delimitative and perdurative verbs, that are also claimed to occupy a dual position. This dual position arises only if the distinction between inherent/potential boundaries and actual/temporal boundaries is not considered. Both delimitative and perdurative verbs in Russian (and Bulgarian), and the ipf Aorist in Bulgarian show the need to distinguish between the two kinds of boundaries. The pf aspect with delimitatives, and the Aorist with ipf predicates highlight actual, temporal boundaries that are presented as being reached. Therefore, the predicates do not exhibit all the properties of other pf verbs and reveal some properties typical of ipf verbs. The situation is somewhat more complicated in Russian, where there is no overt distinction between inherent and actual boundaries. In contrast, it is rather clear in Bulgarian: pf/ipf relates to inherent boundaries, A/I to actual ones. There is no contradiction if an event is presented as having inherent boundaries that are not reached, as is the case with ipf A. There is no need to postulate neutral aspect.

A similar phenomenon is found in Mandarin Chinese and Thai. C. Smith (1997: 68) notes that the interaction between situation type and pf viewpoint, such that if the situation type schema has a natural endpoint, so does the perfective sentence, occurs in a number of languages, but in some it does not. This is the case, e.g., in Mandarin Chinese. The Mandarin Chinese perfectivity marker -le does not always convey completion, it does not always guarantee that the inherent boundary has been reached: 4.113 Wo zuotian xie-le gei Zhangsan de xin, keshi mei xie-wan. I yesterday write-le to Zhangsan de-letter, but not write-finish ‘I wrote a letter to Zhangsan yesterday, but I didn’t finish it.’ [C. Smith 1997: 265] According to C. Smith’s definition, -le is a marker of perfectivity, as it presents closed events (chapter 2, section 2.1). This type of closure is the same for activities, achievements and semelfactives94, but with accomplishments, i.e. <ϕdyn, τ>-verbs, there is a difference between terminating and completing. Mandarin Chinese separates the notion of completion from that of termination. The reading of completion may arise with -le, but, as example ( 4.113) shows, this is only an implicature which may be cancelled due to further information.

The aspect system of Thai exhibits a similar phenomenon. The analysis provided by Koenig & Muansuwan (2000) relies on the difference between boundedness and terminativity suggested by Depraetere (1995)95. Thai has perfective markers that do not entail completedness of the event, which is traditionally assumed to be one characterising feature of perfectivity. In case of ‘semi-perfectivity’, as Koenig & Muansuwan call this phenomenon, the boundary inherent in the predicate is not reached, but still, the predicate is bounded. The semi-pf marker khûm in ( 4.114a) indicates that the event has stopped, but that it is not completed, as is shown in ( 4.114b):

94 These notions pertain to C. Smith’s Aktionsart-classification. 95 Koenig & Muansuwan, however, do not seem to distinguish properly between boundedness/terminativity

and grammatical aspect.

174 Chapter 4: Semantics

4.114 a. Surii tέεη kloon khûm. Surii compose poem semi-pf ‘Surii composed a/the poem.’

b. Surii tέεη kloon khûm tέε jaη mâj sèd. Surii compose poem semi-pf but still not finish ‘Surii composed a/the poem, but has not finished yet.’ To account for semi-perfectivity, Koenig & Muansuwan introduce two operators: an imperfective operator Impfv that all accomplishment-stems96 carry, and the maximality operator Max encoding semi-perfectivity, a weaker notion than terminativity.

The Impfv-operator introduces a condition α (Impfv(ev, φ)): 4.115 α(Impfv(ev, φ)):

An eventuality ev and an event description φ satisfy condition α if and only if there is an e’ which (non-necessarily properly) includes ev and satisfies φ in all “inertia” worlds, i.e. in all worlds compatible with what it would mean to complete ev without being interrupted.

Koenig & Muansowan need this definition of Impfv since they assume Thai accomplishment-stems not to encode completion. Completion might occur in some inertia world.

An additional condition is introduced by the Max-operator, stating that an eventuality ev is a maximal subpart of the continuations e’ that would fit the eventuality description φ. The notion of an event’s maximality with respect to a description φ is captured by the maximality condition Max (Koenig & Muansuwan 2000): 4.116 Max(e, φ) ↔ ¬∃e’’ ∈ UE[e ⊂ e’’ ∧ φ(e’’)]

The referent of a discourse marker e satisfies the predicative condition Max(e, φ) if and only if e is the largest eventuality which satisfies φ, that is, if there is no eventuality e’’ such that e ⊂ e’’ which satisfies the eventuality description φ. Hence, an event is maximal with respect to a description φ if and only if there is no larger event e’’ which satisfies φ. Max(e, φ) states that the semi-pf marker indicates that e is bounded with respect to the eventuality description denoted by the predicate. The asserted event is in a part-of relation to the denoted event, but not in a stage-of relation, since it is not part of a bigger event that would continue into the denoted eventuality φ (cf. section 4.1.2).

Thai accomplishment-(<ϕdyn, τ>)-stems per default carry the Impfv-operator, indicating that they do not refer to complete eventualities but to – not necessarily proper – subparts of inherently bounded eventualities. Although this default relation does not hold for Bulgarian <ϕdyn, τ>-stems (default aspect marking in Bulgarian is the same as in Russian), they are characterised in the way specified by the Impfv-operator if they are marked for the ipf aspect.

The second operator, Max, which is overtly encoded in Thai by khûm, states that the event is bounded with respect to the eventuality description referred to by the predicate. Thai <ϕdyn, τ>-predicates that are marked with khûm thus encode two kinds of boundary-specifications: the <ϕdyn, τ>-stem carries the information that the inherent boundary is not reached, but may be reached if the denoted state of affairs proceeds in the usual way, and the Max-operator

96 Remember that the distinction between boundedness and terminativity (external vs. inherent boundaries)

displays itself only with accomplishment (<ϕdyn, τ>) predicates.

175Chapter 4: Semantics

imposes actual boundaries onto that state of affairs, which is then temporally bounded without having reached its inherent boundary. It is finished but not completed. The same holds for Bulgarian ipf A forms: the ipf aspect gives rise to the inference that the inherent boundary is not reached, the Aorist imposes boundaries onto the time span that is asserted – again, the denoted state of affairs is characterised as finished but not completed. It is an accomplished process (cf. Figure 4.10).

To sum up, Koenig & Muansuwan base semi-perfectivity on two factors: 4.117 a. an Impfv operator associated with accomplishment stems

b. a Max operator associated with the semi-perfective marker Applied to Bulgarian, the semi-perfectivity conveyed by ipf A <ϕdyn, τ>-verbs derives from the ipf as Impfv operator associated with the aspect marking of the predicate, and the Aorist as Max operator, that makes explicit the boundedness of the asserted event such that it is finished but not completed.

4.4 Turkish This section delivers a short overview of the Turkish aspect system. Turkish is interesting insofar, as the expression of IMPERFECTIVITY is formally and semantically quite differentiated. More precisely, Turkish has formal means to express the three specific ipf relations that have been postulated for Russian. One marker that is especially interesting for aspectual analysis – the so-called ‘Aorist’ – will be dealt with in more detail in a short excursus in section 4.4.2.

4.4.1 Aspect semantics and aspect readings The aspecto-temporal system of Turkish is rather complex (cf., e.g. Johanson 1994). This section will shortly introduce the different aspect markers and then characterise them in the same terms as the Russian and Bulgarian markers97. The focus will be on those forms that are used to express the readings of Russian ipf aspect mentioned in section 4.2.

The Turkish aspect system is based on marked intraterminality, with three intraterminal forms being opposed to one form that is agnostic in this respect, cf. ( 4.118)98: 4.118 -mekteydi/-iyordu/-irdi : -di Johanson (1971) calls this opposition between the intraterminal forms and the unmarked -di the basic aspectual opposition (Hauptaspektgegensatz) in Turkish. The system in ( 4.118) is valid only in the past tenses, as the unmarked -di is a simple past marker and cannot be used in the present tense. Each of the forms on the left side of the semicolon signals intraterminality, whereas -di is unmarked in this respect and may simply express the neutral value or, in case of appropriate context, non-intraterminality.

97 The analysis of the different markers is based on Johanson’s (1971) work on aspect in Turkish. His insights

will be embedded in the framework developed here. 98 The notations -iyordu, -mekteydi, -irdi and -di are to be taken as the past tense allomorphic representations

of the different instantiations that are influenced by vowel and consonant harmony, as well as by preceding vowels that enforce the insertion of ‘-y-’. Note that -iyordu, -mekteydi, -irdi will be used for past and non-past alike. In case a claim is valid for the non-past only, the respective forms will be represented as -iyor, -mekte and -ir. In order to avoid misleading commitments by using glosses like ‘pf’, ‘ipf’ or ’prog’, the object language forms are used in the glosses as well. The exception is the past tense marker -di which is glossed as ‘PST’.

176 Chapter 4: Semantics

Due to its semantic unmarkedness, -di may be used to convey the mere denotation of a fact. In this respect, it resembles the Russian ipf aspect in its general-factual use, cf. ( 4.119a,b). The intraterminal form -iyordu in ( 4.119c) cannot be used to ask about the mere fact of reading the book: 4.119 a. Ty čital ėtu knigu?

you read:PST:ipf this book:ACC b. Bu kitabı okudun mu?

this book:Acc read:PST:2SG Q? ‘Did you read this book?’

c. Bu kitabı okuyormudun? this book:ACC read:iyor:Q:PST:2SG ‘Were you (engaged in) reading this book? For the denotative or general-factual function, apparently, unmarked forms have to be used – unmarked with respect to the possible alternate(s) that would be more specific by focusing on one special part of the denoted event and would therefore be unsuitable to convey the mere statement of a fact (cf. the use of the simple past in the English translation of ( 4.119a,b)).

Turkish -iyordu is often considered as progressive marker, mainly due to examples like ( 4.120), which convey the actual-processual reading, and ( 4.121) conveying the conative reading: 4.120 a. Bu sabah seni Kapalı Çarşı’da gördüm; ayakkabı alıyordun.

this morning you:ACC Kapalı Çarşı see:PST:1SG; shoes:ACC buy:iyor:PST:2SG [Ersen-Rasch 1997: 226]

‘This morning, I saw you in the Kapalı Çarşı; you were buying shoes.’

b. Balon yavaş yavaş yükseliyordu. balloon slowly slowly ascend:iyor:PST.

‘The balloon was ascending slowly.’

4.121 a. İmam partiyi kaybediyordu. Imam match lose:iyor:PST [Johanson 1971: 203] The Imam was loosing the match.

b. Az kalsın düşüyordum. almost fall:iyor:PST:1SG [ibd.] ‘I almost fell.’

c. Size geliyorduk, ama misafir geldi. to.you come:iyor:PST:1PL, but guest come:PST ‘We were about to come to you, but then a guest came.’ [Ersen-Rasch 1997: 226] These examples suggest that -iyordu corresponds to the English progressive marker. However, the use of -iyordu differs from the English progressive form, as is shown by the following examples where -iyordu is applied to non-dynamic verbs: 4.122 a. Yeni gözlüklerimle seni gayet iyi görüyorum.

new glasses:1SG:with you:ACC very well see:iyor:1SG ‘I see you (* I am seeing you) very well with my new glasses.’ [Kornfilt 1997: 357]

177Chapter 4: Semantics

b. Hasan fazla çabuk konuştuğunu biliyordu. Hasan too fast talk:FNOM:3SG:ACC know:iyor:PST:3SG ‘Hasan knew (*was knowing) that he was speaking too fast.’ [ibd.] Contrary to the English progressive, -iyordu can be freely used with stative verbs without contextual pressure and without demanding reinterpretation. Kornfilt (1997: 357) thus argues that it is more appropriate to conceive of this form as continuous, and not as progressive: “If the progressive aspect is defined as the continous aspect of a nonstative (dynamic) verb [...], the term continous aspect would be the more inclusive term, including stative and nonstative verbs alike. As a matter of fact, the suffix -(I)yor can indeed be used with both types of verbs.“

This is supported by the fact that -iyordu may also exhibit modal uses and future time reference: 4.123 a. Umutsuz yaşanmıyor.

hope:without live:PASS:NEG:iyor:3SG [Johanson 1971: 121] ‘It is impossible to live without hope.’

b. Biraz sonra geliyor. a.little.bit later come:iyor ‘He will come a little bit later.’

Obviously, -iyordu comprises two relations: one where the topic time is included in the event time, cf. ( 4.124), and one, where the topic time interval is synchronous to the run time of the event, cf. ( 4.125): 4.124 -iyordu

a. TT included in the dynamic phase/event time: TT ⊂ ϕdyn/e

b. I(e)[i j[ ∩ I(TT)]m n[ → I(TT)]m n[ = {e, m < ϕdyn/e < n}

The relation TT ⊂ ϕdyn/e captures the fact that -iyordu does not necessarily require a dynamic phase as input, but may be applied to lexical statives as well. Semantically, -iyordu is characterised as selecting a phase and relating it to the topic time, the topic time interval being included in the interval of the phase. This relation is basic to the examples ( 4.120a,b) above. In ( 4.120), it is the dynamic phase of the shoe-buying and the ascending of the balloon that is asserted by -iyordu. This yields the actual-processual reading. In ( 4.122) the predicate does not provide a dynamic phase (it is static), but -iyordu is nevertheless possible.

The second possible relation for -iyordu is illustrated in ( 4.125):

4.125 -iyordu

a. TT simultaneous with phases and boundaries: TT = ϕdyn τ ϕstat

b. I(e)[i j[ ∩ I(TT)]m n[ → I(TT) = e

[ [ i ϕdyn τ ϕstat

j AKTIONSART I(e)

] [ m

n TOPIC TIME I(TT)

ASPECT

178 Chapter 4: Semantics

This relation is basic to examples ( 4.123) above, i.e. to the cases where -iyordu does not individuate an event and therefore does not introduce an event into the discourse, but is used modally or to convey future time reference.

There is another form – -mekteydi – which seems to be more comparable to the English-type progressive marker. This is indicated by the fact that it can only be used with nonstative verbs: 4.126 a. Hasan sorunu anlıyor.

Hasan question:ACC understand:iyor ‘Hasan understands (*is understanding) the problem.’ [Kornfilt 1997: 358]

b. ??Hasan sorunu anlamakta. Hasan question:ACC understand:mekte ‘Hasan understands (*is understanding) the problem.’ [ibd.]

c. *Hasan sorumun cevabını bilmekte. Hasan question:1SG:GEN answer:3SG:ACC know:mekte ‘Hasan knows (*is knowing) the answer to my question.’ [ibd.]

d. Çayım içmekteydim. tea:1SG drink:mekte:PST:1SG ‘I was (in the act of) drinking my tea.

The use of -iyor is fine with a stative verb like anlamak (‘understand’), whereas -mekte yields a marked interpretation and requires a specific context to support this use. With bilmek (‘know’), however, -mekte yields an unacceptable sentence (Kornfilt 1997: 358).

Contrary to -iyor, -mekte cannot have future time reference, as it expresses a ‘narrow present tense’: “-yor şimdiki zamanı başlamamış, uzak şimdiki zaman da bildirir: yarın geliyorum misalinde olduğu gibi. -makta, -mekte’de ise başlamış bir yakın, mutlak şimdiki zaman ifadesi vardır. Meselâ biraz sonra gelmekteyim diyemeyiz. gelmekteyim diyorsak hareket başlamıştır.” (Ergin 1998: 323)

[-yor also expresses a present tense that has not yet begun, like, e.g., yarın geliyorum (‘I come tomorrow’). -mekte, however, expresses a narrow, absolut present tense that has already begun. Biraz sonra gelmekteyim (‘I come a little bit later’) is not possible to say. Gelmekteyim (‘I am coming’) expresses an action that has already begun.]

Basic to -mekteydi is the following semantic configuration: 4.127 -mekteydi

a. TT included in the dynamic phase: TT ⊂ ϕdyn

b. I(e)[i j[ ∩ I(TT)]m n[ → I(TT)]m n[ = {ϕdyn, m < ϕdyn < n}

[ [ i ϕdyn τ ϕstat j

AKTIONSART I(e)

TOPIC TIME I(TT)

ASPECT

179Chapter 4: Semantics

Contrary to -iyordu, -mekteydi requires a dynamic phase as input. With this requirement for a stage-of relation rather than a mere part-of relation, this form resembles very much the English progressive.

A further aspect marker of Turkish is the so-called ‘Aorist’. The term ‘Aorist’ is somewhat misleading as it is used quite differently from what this term traditionally refers to, i.e. to cases like the Bulgarian Aorist. In the Turkish case, ‘Aorist’ is to be understood as defined, e.g., by Lewis (2000: 115): “This term [the Aorist, B.S.], borrowed from Greek grammar, means ‘unbounded’ and well describes what the Turks call geniş zaman ‘the broad tense’, which denotes continuing activity.”

Although Lewis and others regard the Aorist as ‘tense’, attempts to define it temporally as ‘broad tense’ (geniş zaman) are not very convincing, cf. Ergin’s (1998) slightly odd definition: “Bu eklerde üç zaman ifadesi bulunur: 1. her zaman, 2. geçmiş zamanla şimdiki zamanı içine alan bir zaman, 3. gelecek zaman: her zaman yaparım, bugün yaparım, yarın yaparım.” (Ergin 1998: 320) [With this form, three readings can be found: 1. always, 2. a time that comprises the past and the present, 3. future tense: her zaman yaparım (‘always do:ir:1Sg’), bugün yaparım (‘today do:ir:1Sg’), yarın yaparım (‘tomorrow do:ir:1Sg’).]

According to this definition, the Aorist seems to comprise almost everything. Here, it will be argued against such a temporal conception of the Aorist, and for an analysis as aspect marker.

The Aorist gives rise to a variety of rather different readings, some of which are illustrated in ( 4.128a-f): 4.128 a. Dikkat et, yoksa düşersin.

attention do:IMP, else fall:ir:2SG [Ersen-Rasch 2001: 140] ‘Look ahead, lest you fall!

b. Yemeğin yanına ne alırsın? meal:GEN side:3SG:DAT what take:ir:2SG? [Ersen-Rasch 2001: 140]

‘What side dish would you like to take?’

c. Şimdi yapmakta olduğunuz işi tekrar seçer miydiniz? now do:INF:LOC AUX:FNOM:2PL work:ACC again choose:ir Q:PST:2PL?

‘Would you once again choose the job you are doing now?’ [Tornow 1993: 102]

d. Gidip haber verecek. Telefon eder. come:CVB news give:FUT. telephone make:ir [Johanson 1971: 186] ‘He will come and bring the news. Maybe he will simply give a call.’

e. Bu su içilir / içil-mez. this water drink:PASS:if / drink:PASS:ir:NEG [Tornow 1993: 100] ‘You can (may) / cannot (must not) drink this water.’

f. Babam satranç oynar. father:1SG chess play:ir:PRS ‘My father plays chess.’ (= He can play chess.)

[ [ i ϕdyn τ ϕstat

j AKTIONSART I(e)

] [ m

n TOPIC TIME I(TT)

ASPECT

180 Chapter 4: Semantics

Several ways of analysing the Aorist have been proposed, e.g., in terms of habituality or iterativity, but none of these suggestions really captures the semantics of this form. The Aorist is not confined to one of the uses illustrated in ( 4.128), but rather comprises all of them. Furthermore, habituality and iterativity may be conveyed with the other intraterminal forms and with -di as well. These readings of the Aorist are not displayed by the form alone, but arise due to contextual influence and activation of conceptual knowledge. What needs to be stated is the stable, encoded information that is basic to all of these interpretations. But how is this general meaning to be characterised, what is the common denominator for all those uses?

G. Lewis (2000: 116) defines the semantic invariant of the Aorist as follows: “Fundamentally, yaparım means ‘I am a doer’ and according to context it may represent: ‘I habitually do’; ‘by and large I am the sort of person who does’; ‘I am ready, willing, and able to do’; ‘I shall do’.”

Contrary to -iyordu and -mekteydi, the Aorist does not report a certain behaviour, i.e. action. It does not introduce an event into the discourse, but has the effect of characterising the entity it refers to (Yavaş 1982: 46): 4.129 a. Benim kasabım iyi et satar.

my butcher:1SG good meat sell:ir:3SG [Yavaş 1982: 46] ‘My butcher sells good meat.’

b. Benim kasabım iyi et satıyor. my butcher:1SG good meat sell:iyor:3SG [ibd.] ‘My butcher is selling good meat.’ In ( 4.129a), the butcher is characterised as one that sells good meat, whereas ( 4.129b) refers to his activity of selling good meat. ( 4.129a) thus conveyes a characterisation of the butcher, ( 4.129b) conveyes his behaviour. ( 4.129b) states an event and may answer to the question “What is happening?”, whereas ( 4.129a) cannot be used as an answer to that question. Verbs marked with the Aorist thus can be regarded as being less typical instances of the syntactic category ‘verb’. Aorist-marked verbs have a lower degree of categoriality (Hopper & Thompson 1984). This is also evident with the permanent ( 4.130a) and atemporal ( 4.130b) uses of the Aorist: 4.130 a. Evin ardı ormana bakar.

house:GEN back:3SG forest:DAT look:ir ‘The back of the house looks to the forest.’

b. Demir, suda batar. iron, water:Loc sink:ir ‘Iron sinks in water.’

What all the uses of the Aorist share is the fact that in no case one specific part of the encoded event is focused upon or selected. Rather, the Aorist seems to abstract away from the internal make-up of the predicate and seems to relate it as a whole to an unbounded topic time thereby characterising it. The Aorist will be examined in more detail in section 4.4.2.

Its semantic configuration can be specified as follows:

4.131 -irdi (‘Aorist’)

a. TT simultaneous with phases and boundaries: TT = ϕdyn τ ϕstat

b. I(e)[i j[ ∩ I(TT)]m n[ → I(TT) = e

181Chapter 4: Semantics

To sum up so far, Turkish has several means to markedly express intraterminality. This poses several questions: What is the difference between -iyor(du) and -mekte(ydi)? What is the status of the Aorist with respect to the other intraterminal forms? Furthermore, how can we explain the fact that sometimes one form may be used in the proper domain of another? Again, a purely semantic account does not suffice to answer these questions. Rather, semantics has to be combined with a pragmatic framwework (see chapter 5).

Let us now turn to the semantically unmarked form -di. This form is unmarked with respect to two terminality categories: postterminality (see below) and intraterminality. In Turkish, dynamic and static phases may be selected by two different aspect markers signalling intra- and postterminality, respectively. In case of no such explicit selection, i.e. in case of -di is applied, a non-intraterminal and non-postterminal interpretation is assigned per default. This does not mean that -di expresses the negative values of both categories. It is agnostic in this respect and may receive a negative or a neutral value, depending on the context. Due to this unmarkedness, -di does not focus on specific parts of the Aktionsart basis, but it still individuates events (contrary to the Aorist). It thus can be used to convey the mere statement of a fact, that is, in the denotative function. Consequently, -di may also include an inherent boundary and thus can be pragmatically strengthened to asserting the reaching of that boundary.

The denotative use is based on the following configuration:

4.132 -di

a. TT includes phases and boundaries: TT ⊃ ϕdyn τ ϕstat

b. I(e)[i j] ∩ I(TT)[m n] → I(TT)[m n] = {e, m ≤ e ≤ n}

In this use, -di corresponds to the general-factual reading in Russian, which is illustrated by the following examples: 4.133 a. Bu mektubu ona gösterdin mi?

this letter:ACC her show:PST:2SG Q? ‘Did you show her this letter?’

b. - Bu vakte kadar ne yapıyordun? - this time until what do:iyor:PST:2SG - İş aradım. - work look.for:PST:1SG [Johanson 1971: 117]

[ [ i ϕdyn τ ϕstat j

AKTIONSART I(e)

ASPECT [ ] m

n TOPIC TIME I(TT)

[ [ i ϕdyn τ ϕstat j

AKTIONSART I(e)

m

n TOPIC TIME I(TT)

ASPECT

182 Chapter 4: Semantics

‘What have you been doing up to now?’ ‘I looked for work.’ Examples ( 4.134) show that in the context of limited duration -di is used, since intraterminal forms are excluded:

4.134 a. Çocukların resimlerine uzun uzun baktı.

children:GEN photographs:2SG:DAT for a long time look:PST ‘She looked at the children’s photos for a long time.’

b. Hasan bütün gün birşeyler mırıldandı. Hasan whole day things mumble:PST.

‘Hasan kept on mumbling things all day long.’ Here, the intraterminal forms -mekteydi and -iyordu would not be possible since they cannot include the boundaries of the asserted interval. Note that the Russian ipf aspect can be used in cases like ( 4.134), in fact, it would the only possibility here.

Being the unmarked form, -di may be pragmatically strengthened and express pf-like values in an appropriate context. The possible configurations can be illustrated as follows: 4.135 -di

a. TT includes the boundary: TT ⊃ τ

b. I(e)[i j[ ∩ I(TT)[m n] → I(TT)[m n] = {τ, m ≤ τ ≤ n} b’. I(e)[i j[ ∩ I(TT)[m n[ → I(TT)[m n[ = {τ, m ≤ τ < n}

Both configurations are characterised by including the inherent boundary in the topic time interval. In case of ( 4.135a), I(TT) is closed-bounded, which corresponds to the configuration of the event reading of the Russian pf aspect. In ( 4.135b), I(TT) is semi-open to the left and thus corresponds to the perfect reading of Russian pf.

Closed-bounded situations, i.e. situations that are presented as including their boundaries, can only be conveyed by the unmarked -di. Both intraterminality and postterminality exclude any boundaries and focus explicitly on the respective phases; they do not allow for closed-

[ [ i ϕdyn τ ϕstat j

AKTIONSART I(e)

[ ] m n

TOPIC TIME I(TT)

ASPECT

[ [ i ϕdyn τ ϕstat

j AKTIONSART I(e)

[ [ m n

TOPIC TIME I(TT)

ASPECT

183Chapter 4: Semantics

boundedness per definitionem. To convey temporally bounded or semi-bounded situations, -di has to be used. ( 4.136) is an example of a sequence of single, bounded events: 4.136 Mutfağa bir göz attım. Çaydanlığın altını yaktım;

kitchen:DAT one glance.at:PST:1SG teakettle:Acc put.on:PST:1SG, gittim yüzümü yıkadım. go:PST:1SG face:1SG:ACC wash:PST:1SG

‘I took a look at the kitchen. I put on the kettle, went away and washed my face.’ The examples in ( 4.137) illustrate cases where -di denotes a semi-open interval: 4.137 a. Acıktım get.hungry:PST:1Sg ‘I am hungry.’ (= I got hungry, and now I am in the state of being hungry.)

b. Hatırladım. remember:PST:1Sg ‘I remember.’ As has been pointed out in chapter 2, section 2.2.3, the Turkish aspect system illustrates the need to distinguish between dynamic and static phases. Turkish has one aspect marker that selects static phases: -miş selects the phase after the transgression of their relevant terminus (Johanson 1971: 286), i.e. the stative phase. Note that this use crucially presupposes the existence of a culmination point in the denoted event – otherwise, it would not be possible to speak of a static phase at all. -miş does not include any boundaries. Consequently, both inherent and actual boundaries are outside the scope of assertion, and therefore, the resultant state holds at the time of utterance99. Such resultant states are lexically specified in case of inceptively stative (<τ, ϕstat>) verbs, inceptively dynamic (<τ, ϕdyn>) or terminative accomplishment-type verbs (<ϕdyn, τ, ϕstat>). The difference between -miş as selecting stative phases, -iyordu as selecting dynamic phases and -di as being agnostic to both notions and thereby receiving a PERFECTIVE interpretation is illustrated in ( 4.138):

4.138 a. Oturdu.

sit:PST ‘He sat down.’

b. Oturuyordu. sit:iyor:PST ‘He was sitting.’

c. Oturmuş. sit:miş ‘He has sat down.’

Oturmak (‘to sit down’, ‘to sit’) is a terminative verb that encodes a initial boundary and a following phase100. With such <τ, ϕ>-predicates, the phase following the culmination point can be asserted as a dynamic phase, ( 4.138b), or as a stative phase, ( 4.138c). Given an

99 Note that -miş differs from both the Russian pf and ipf aspect. The perfect reading of the pf aspect includes

the boundary in its assertion scope since the pf aspect cannot focus merely on the consequent state. In case of inchoative and inceptive verbs, the ipf aspect may focus on the consequent state, but contrary to -miş it cannot trigger the Aktionsart basis to induce a consequent state, as it does not carry the requirement for such an input.

100 Some languages have different lexemes for the two components, e.g., Russian sadit’sja (ipf) / sest’ (pf) ‘sit down’ and sidet’ (ipf) ‘sit’.

184 Chapter 4: Semantics

appropriate context, the initial boundary may be focused upon by the unmarked -di, ( 4.138a) (cf. Johanson 1971: 214f).

Since -miş does not include the inherent boundary in its scope, this form may be used in case the resultant state of the event is under discussion, or to convey that the speaker did not witness the respective event himself. This is illustrated in ( 4.139a,b), where neither the falling of the snow, nor the coming of Erol has been witnessed by the speaker. Rather, those events are inferred from positive evidence: 4.139 a. Kar yağmış.

snow fall:miş ‘Snow has fallen.’

b. Erol gelmiş. Erol come:miş ‘Erol has come.’ (= He is here.) This gives also rise to evidential interpretations, as is shown in the following examples: 4.140 a. Misafir de baktı saatine. – Benimki durmuş dedi.

guest also look:PST his.watch:DAT – mine stop:miş say:PST [Johanson 1971: 289] ‘The guest also looked at his watch. – [Obviously] it stopped to work.’

b. Dün gece şiir yazmışım sarhoşluğumda. Kadından, yesterday night poem write:miş:1SG drunkenness:1SG:LOC wife:ABL, aşktan bahsetmişim. love:ABL tell:miş:1SG [Johanson 1971: 288] ‘Yesterday night, being drunk, I [obviously] wrote a poem. I have written about my wife, about love.’

c. Na yapmışım mürdür bey? what do:miş:1SG director mister? [ibd.] What am I supposed to have done, director?’ ( 4.140b) is interesting insofar, as it shows that -miş may also be used in order to refer to the first person, i.e. to the experiencer. Its use is appropriate if, due to some circumstance, the speaker was not witness of what he has done. The use with the first person is also commonly found in questions like ( 4.140c).

The semantic characterisation of -miş looks as follows:

4.141 -miş

a. TT included in the static phase: TT ⊂ ϕstat

b. I(e)[i j[ ∩ I(TT)]m n[ → I(TT)]m n[ = {ϕstat, m < ϕstat < n}

[ [ i ϕdyn τ ϕstat

j AKTIONSART I(e)

] [ m n TOPIC TIME I(TT)

ASPECT

185Chapter 4: Semantics

Neither Russian nor Bulgarian encode postterminality as explicitly as does Turkish by means of -miş101. With respect to Desclés continuum (chapter 2, section 2.2.2), -miş can be said to encode the feature [±attributif].

Now let us turn to repetitive events. Here, -iyordu, -di and -irdi are possible. The main difference between these forms is that repetitive readings derived by -iyordu or -di are internally inhomogeneous. Use of the Aorist derives internally homogenous readings, therefore the notion of ‘repetitive event’ does not quite fit here. This will be examined in more detail in the next section. For the time being it suffices to state the difference between internal homogeneity conveyed by the Aorist and internal inhomogeneity denoted by -iyordu and -di. However, as has been shown above by comparing -iyordu to -irdi, the situation is not as clearcut as the theory might suggest, since there is a certain overlap in use of both forms: 4.142 a. Ona iki defa ret cevabı verildi.

him two times denial answer:ACC give:PASS:PST. ‘They rejected him twice.’

b. Eskiden her cumartesi günü hamama gidiyordu. formerly every saturday bath:DAT go:ipf:PST

‘She used to go to the sauna every saturday.’ If -di is used to convey repeated events, it denotes a number of completed and clearly distinguished events (cf. the illustration in ( 4.65)), if -iyordu is used, the focus is rather on the phase of the respective events.

The most appropriate form to convey habituality understood as unbounded repetition is the Aorist, which is shown in ( 4.143): 4.143 a. Büyükbabayı iyi anımsıyorum: torunlarla gezerdi,

grandpa:ACC well remember:ipf:1SG with.grandchildren take.a.walk:ir:PST onlarla futbol oynardı, pipo içerdi, ...

with.them football play:ir:PST, pipe smoke:ir:PST ‘I remember grandpa very well: he used to go for a walk with the

grandchildren, to play football with them, to smoke a pipe, ...’

b. Eskiden her cumartesi günü hamama giderdi. a.long.time.ago every saturday bath:DAT go:ir:PST

‘A long time ago, she used to go to the sauna every saturday.’ The following pair of examples illustrates the difference between -iyordu and the Aorist with habitual events: 4.144 a. Bir öğrencim her gün geç geliyordu.

one student:1SG every day late come:iyor:PST.3SG ‘One student of mine was coming late every day.’

b. Bir öğrencim her gün geç gelirdi. one student:1SG every day late come:ir:PST:3SG ‘One student of mine had the property of coming late every day.’ ( 4.144a) makes an assertion of a heterogeneous activity consisting of multiple instances of coming late. The frequency adverbial her gün takes scope over geç geliyordu and thus asserts an event that occurred every day. This is different in ( 4.144b): the Aorist does not assert a

101 For a more detailed analysis of -miş, compare, e.g. Aksu-Koç (1988); Yavaş (1980); Johanson (2000).

186 Chapter 4: Semantics

heterogeneous activity consisting of multiple occurrences, but it denotes an inherently homogeneous state of affairs and relates it to the topic entity (the grammatical subject). The topic entity takes semantic scope over the frequency adverbial and is assigned the property of coming late every day. There is no individuation of single late-coming-events. Thus, in ( 4.144a), it is an observed behaviour that is asserted to occur every day, whereas in ( 4.144b), a topic entity is asserted that is characterised by an every-day-late-coming-property.

Table 4.9 summarises the comparison of Russian and Turkish with respect to which aspect forms are used in order to convey the different readings. This table shows that the three different aspectual relations that have been postulated for Russian (section 4.1.2) are expressed by different markers in Turkish (more on the consequences of that in chapter 5).

Reading Russian Relation Turkish Example 1 actual-processual ipf TT ⊂ ϕdyn -mekteydi; -iyordu ( 4.120) 2 conative ipf TT ⊂ ϕdyn -iyordu ( 4.121) 3 inactual ipf TT = e -iyordu; -irdi ( 4.123), ( 4.129a)4 durative ipf TT ⊃ e -di ( 4.134) 5 general-factual ipf TT ⊃ e -di ( 4.133) 6 iterative ipf T-Val ⊃ er -di; -iyordu ( 4.142) 7 habitual ipf T-Val = er -iyordu; -irdi (Aorist) ( 4.143) 8 potential ipf T-Val = er -irdi (Aorist) ( 4.128f) 9 permanent ipf TT = e -irdi (Aorist) ( 4.130a) 10 atemporal ipf TT = e -ir102 (Aorist) ( 4.130b)

Table 4.9 This table only shows the Turkish equivalents for Russian ipf aspect. To convey the readings expressed by the Russian pf aspect, Turkish uses -di. This form does not completely correspond to the pf aspect, it may as well be used to convey certain readings of the Russian ipf (cf. Table 4.11 below).

Summing up, this section has revealed that Turkish has morphological means to distinguish between the three specific relations that characterise the ipf aspect in Russian. This is summarised in Table 4.10:

IPF103 Russian Turkish 1 TT ⊂ ϕdyn -iyordu / -mekteydi 2 TT = e -di 3 TT ⊃ e -irdi (Aorist)

Table 4.10 The following triple of examples illustrates this very nicely (Ersen-Rasch 2001: 153): 4.145 a. Bir gün babam bizi en yakın kasabaya götürüyordu. Daha oraya varmadan

‘Köyümüze dönmek istiyoruz’ dedik. ‘One day, my father was taking us to the nearest county town. Before we arrived there, we said “We want to go back to our village”.’ [TT ⊂ ϕdyn]

102 Here it is the present tense, as the use of the past tense (possibly defeasibly) implies that the assertion is

not valid any longer. Atemporal statements, however, are characterised by being not restricted in their validity. 103 ‘IPF’ refers to the meta-category IMPERFECTIVITY and not to its specific instantiation.

187Chapter 4: Semantics

b. Hastalanınca babam bizi en yakın kasabaya götürdü. ‘When got ill, my father took us to the nearest county town.’ [TT ⊃ e]

c. Hastalanınca babam bizi en yakın kasabaya götürürdü. ‘Whenever we got ill, my father would take us to the nearest county town.’ [TT = e]

The unmarked -di covers the uses of the Russian pf aspect, as this is the only form in Turkish to convey – albeit indirectly – PERFECTIVITY. Due to its unmarked status, -di conveys the general-factual reading of the Russian ipf aspect, but not the readings based on the other two ipf configurations. These are explicitly marked in Turkish with -iyordu and -irdi. This is summarised in Table 4.11:

Russian Relation Turkish pf TT ⊃ τ

TT ⊃ e -di

TT ⊂ ϕ -iyordu / -mekteydi ipf TT = e -irdi (Aorist)

Table 4.11 Table 4.11 reflects the differences of markedness in the Russian and Turkish aspect systems. Russian is marked for PERFECTIVITY, Turkish for IMPERFECTIVITY. The assumption of the three specific relations for the Russian ipf aspect is justified by the fact that they are morphologically expressed in Turkish. This will become important especially with respect to the assumed ambiguity of Russian ipf between these three relations (chapter 5).

4.4.2 Excursus: The ‘Aorist’ This section will briefly outline a proposal for analysing the Turkish Aorist.

As has been shown by the examples in section 4.4.1, the use of the Aorist is neither restricted to a certain class of verbs, nor is there a class of verbs that requires the Aorist and could not be combined with some other aspect marker104. The use of the Aorist is not determined by the lexical specification of the verb, neither by some objective characteristics of the denoted situation. The use of the Aorist depends on how the speaker perceives a situation, not on the objective quantity of the event (Johanson 1971: 159; Yavaş 1982: 48). Thus, specific instantiations of permanent states have to be expressed not with the Aorist but with -iyor(du), as is illustrated in ( 4.146b):

4.146 a. Fırat ve Dicle Basra’nın kuzeyinde birleşirler.

Euphrat and Tigris Basra’GEN north:3SG:LOC join:ir:3PL [Johanson 1971: 168] ‘Euphrat and Tigris join to the north of Basra.’

b. Bunlar nereden doğuyor? Nerede birleşiyor ve hangi they where:ABL have.source:iyor?Where:LOC join:iyor and which denize dökülüyorlar? sea:DAT flow:iyor:3PL? [ibd.] ‘Where do they have their source? Where do they join and in which sea do they flow?’

104 As has been shown in section 4.4.1, the only restriction encountered in the Turkish aspect system with

respect to the combination of aspect markers and Aktionsart basis is encountered with -mekteydi, which requires a dynamic phase as input.

188 Chapter 4: Semantics

Within the semantic framework for grammatical aspect established here, the Aorist has been described as selecting the descriptive content of the predicate as a whole without focussing on certain parts of it. Since I(TT) is not bounded (cf. ( 4.131)), it functions as status aspect in the sense of Breu (2005). The descriptive content of the predicate is thus not related to a topic time but to a topic entity, which predominantly corresponds to the grammatical subject (Klein 2001). Therefore, utterances containing an Aorist do not answer to the question “What happened/is happening?” – they do not introduce an event into the discourse but characterise an entity. This is emphasised by Yavaş (1982: 48): “[…] I would like to argue that the Aorist characterizes what is typical, normal or even inherent to an entity or to a situation.”

Utterances containing the Aorist are thus not characteristic of a time period, but of an entity. Temporal restrictions on the validity of this characteristic depend on the life span of the respective entity or are imposed by temporal frames that explicitly specify the validity of that characteristic: 4.147 a. İstanbul yedi tepe üstünde durur.

Istanbul seven hill above stand:ir:3SG [Yavaş 1982: 43] ‘Istanbul stands on seven hills.’

b. ?İstanbul yedi tepe üstünde dururdu. Istanbul seven hill above stand:ir:PST:3SG ?‘Istanbul stood on seven hills.

4.148 a. İki kere iki dört eder.

two times two four make:ir:3SG ‘Two times two make four.’

b. # İki kere iki dört ederdi. two times two four make:ir:PST:3SG #‘Two times two made four.’

4.149 Babam erken yatar ama son iki senedir geç yatıyor.

my.father early lay.dow:ir:3SG but last two year late lay.down:iyor:3SG [Yavaş 1982: 45] ‘My father goes to bed early, but within these last two years he has been going

to bed late.’ The descriptive content selected by the Aorist in ( 4.147a) is ascribed to the topic entity – the city of Istanbul – thereby characterising it as being located on seven hills. Uttering the same sentence with a past tense Aorist form, as in ( 4.147b), locates this topic entity somewhere in the past, as tense establishes a relation between the topic time – here mediated via the life span of the topic entity105 – to the time of utterance. The Aorist ascribes the descriptive content to the subject. If this subject is located in the past, the utterance conveys that it does not exist any longer (again the life-time effect, Musan 1997). This interpretation is due to

105 Kratzer (1995) captures the distinction of stage-level and individual-level predicates in terms of the event

argument which the latter are lacking. It is this event argument that carries the spatiotemporal loctation (the topic time) of some state of affairs and thus individuates it as event. This spatiotemporal location is the external argument of stage-level predicates. Individual-level predicates lack an event argument and consequently lack a spatiotemporal location. In this case, the external argument is an individual, i.e. a topic entity. Kratzer conjectures that the tense predicate always relates to the external argument of a predicate (1995: 156). With predicates that individuate events, thus, tense relates to the topic time, with predicates that do not individuate events, tense relates to the topic entity. Whether there is a topic time or a topic entity does not have any impact on finiteness – an assertion is made in both cases.

189Chapter 4: Semantics

pragmatic inference, as it is defeasible by a special context (something like ‘but nowadays the hills are not visible any more’).

( 4.148b) is unacceptable, as there is hardly any context, where this (probably) general truth can be claimed to be not valid any longer – perhaps only as a consequence of new, rather revolutionary mathematical findings.

Matters are slightly different with ( 4.149), where no such a life-time effect arises and the Aorist can be used, even though the validity claim is made for an interval that is bounded by the context. This does not contradict the semantics of the Aorist. Although this form does not impose any boundaries on the topic time, boundaries may be induced by the context or by means of inference. What is asserted by the use of the Aorist is that within that period of time, the validation time, the described behaviour is or was characteristic of the topic entity. Again, it proves crucial which interval the feature of boundedness pertains to (cf. section 4.2.1).

To recap so far, the general definition of aspect as establishing a relation between a sentence base I(e) and a topic component includes the possibility that this topic component might as well be a topic entity. Aspect thus may not only relate a sentence base to a topic time, but also to a topic entity. Russian and Bulgarian do not have overt means to encode the relation of I(e) to a topic entity. This relation is merely one of the possible relations an aspect marker may be assigned in a specific context. This is different in Turkish, where the Aorist indicates that the topic component is not a topic time, but a topic entity (cf. Figure 4.1, section 4.1.1). Since the Aorist does not select a specific part of the semantic representation of the verb, and since TT is not bounded, individuation is not possible for the Aorist. Therefore, it does not introduce an event into the discourse. As has been shown in ( 4.149), boundedness may be induced by context or by time frame adverbials. Thus, predicates combined with the Aorist may have a specified temporal duration. The run time of the states of affairs106 denoted by the Aorist can be presented as having temporal boundaries. But due to their non-relatedness to some specific interval on the time axis (remember that the relation to the time-axis is established indirectly via the topic entity), such states of affairs cannot be temporally located.

In accounting for the Turkish Aorist, the notion of ‘focality’, introduced by Johanson (1971), is decisive. Focality pertains to the relatedness of the denoted eventuality to the temporal interval under discussion (1971: 15), i.e. to the topic time interval. Focality is present if there is one single, lexically or contextually determined time span that serves as topic time. Focality is closely related to temporal locatability (remember that tense establishes a temporal relation between TT and TU, but this requires an individuated topic time interval), which is a property that characterises events independently of their lexical properties (Rothstein 1999: 366). Whether stative or dynamic, terminative or aterminative – any state of affairs can be temporally located as long as it is individuated as an event. Any eventuality can be presented as temporally located or not located. Individuation vs. non-individuation of discourse entities (participants or events) is closely related to the mass-count distinction. This is the starting point for Rothstein’s (1999) analysis of the semantics of predicative adjective phrases and be, some insights of which will prove useful to analyse the semantics of the Turkish Aorist.

Contrary to, e.g., Bach (1981, 1986), Rothstein (1999) takes the mass-count distinction not as a distinction within the verbal domain, but one between predicates or entities under a particular description (1999: 377). She takes this distinction to be the difference between the denotation of verbs and the denotation of adjectives, the two main subgroups of eventualities (1999: 362). The mass-count distinction thus distinguishes verbs from adjectives107. The domain of atomic count eventualities includes terminatives as well as processes and statives. Adjectives constitute the domain of non-atomic, mass, state-like eventualities. This yields a

106 This notion is used in order to avoid the notion of ‘event’ or ’situation’ that both imply individuation. 107 I will not engage here in the discussion of parts of speech classification which is a highly controversial

issue. For a discourse-based account see Hopper & Thompson (1984).

190 Chapter 4: Semantics

distinction between stative verbs that are count predicates and statives denoted by adjectives. The latter are mass predicates or ‘M-states’ (mass-states). The crucial difference between both kinds of eventualities is temporal locatability: count eventualities can be temporally located (even lexical statives), M-states cannot. They may have a temporal duration, but they cannot be located. Rothstein illustrates this claim with the following examples (1999: 365): 4.150 a. Yesterday the witch made John clever for three hours

b. Yesterday the witch made John clever at three o’clock. For three hours in ( 4.150a) modifies clever, whereas at three o’clock in ( 4.150b) modifies made but cannot modify clever. This indicates that the M-state denoted by clever has a certain duration (three hours) but cannot be temporally located.

Rothstein analyses be as a function from the domain of M-states into the domain of count eventualities. More specifically, be denotes a particular kind of packaging operation, namely a locating function mapping from the domain of M-states to the domain of located eventualities (1999: 372). The operation complementary to packaging is that of grinding. However, Rothstein does not describe a grinding function in the domain of eventualities – presumably, because English does not have explicit means to express this function108.

Grinding is a function that maps individuals onto the stuff they are made of, i.e., a function mapping from the count domain (C) onto the mass domain (M). It maps individuals on the join of their material parts. The grinding function is defined as follows (Rothstein 1999: 371): 4.151 grinder

A function g: C → M such that for every c ∈ C: g(c) = ∪{x ∈ M: xKc}, where K is the relation ‘material part of’. In words: grinding is a function g from the count domain C to the mass domain M, such that for every element c of C the function g(c) gives the sum (∪) of all x’s that are elements of M, where x is material part of c.

This function is used in the interpretation of the following example, where ‘bicycle’ is used as a mass term (Rothstein 1999: 371): 4.152 After he had been working for an hour, there was bicycle all over the garage floor. This is an example of how a noun can be mapped from the count domain onto the mass domain. Since verbs also belong to the count domain, such a mapping should be possible for them as well. Grinding is a function that gives the stuff an element is made of. In case of verbs, it is the elements located at their different tiers (e.g. actional, causal, temporal) that encode the stuff the denoted eventualities are made off. Grinding results in just taking these elements without any further specification or restriction. Specification in the verbal domain would be, e.g., the focusing or selection of one particular unit from the temporal tier. The restriction would be the topic time to which this part is related. With grinding, no specific part of the eventuality description is selected. All the encoded phases and boundaries located at the

108 According to Jackendoff (1991: 27), grinding in the verbal domain is one of the readings of the

progressive aspect. He takes Bill is running to the store as “the process out of which the event Bill runs to the store is composed”. This differs from the conception of grinding here (and presumably from how Rothstein conceives of grinding), since Bill is running to the store denotes an individuated event and is temporally locatable. Furthermore, the aspect marker -ing explicitly focuses on one specific part of the eventuality description (the dynamic phase).

191Chapter 4: Semantics

temporal tier are asserted not by relating them to some restricted (temporally bounded) topic time interval but to some topic entity. The eventuality description is not hooked up to a specific time interval, but to the topic entity. Since the stuff the eventuality description consists of is not bounded by a topic time, no event is individuated and introduced into the discourse. The temporal bounding of that stuff is achieved indirectly via the temporal boundaries of the topic entity. Therefore, such ‘grinded events’ may have a temporal duration, but they cannot be temporally located. Compare ( 4.150a) above: clever can be modified by for three hours, but this bounding pertains to John, who is characterised by being clever for three hours. In order for clever to be temporally locatable, it should refer to an individuated discourse entity. As ( 4.150b) shows, however, clever cannot be modified by at three o’clock. Rothstein (1999) argues, that adjectives denoting M-states may be mapped onto the count domain by means of be. It seems reasonable to assume that there exists a complementary process for mapping count predicates onto the mass domain, thereby denoting M-states.

Prototypically, verbs introduce events as discourse participants, but not all verbs are semantically prototypic and refer to events. Even prototypic verbs may be used in a non-prototypic function, i.e. as not referring to events109. In many languages, this non-prototypical function is indicated by less overt verbal morphology, i.e., by formally less marked forms110. This is the case, e.g., in English, Russian and Bulgarian, where formally and semantically unmarked verb forms are used if the form is not intended to introduce an event into the discourse. However, this is different for Turkish, which has one overt form that explicitly encodes ‘non-individuation’ or ‘non-introduction’ of event – the Aorist. According to what has been said above, the Aorist can be regarded as overt marking of the grinding function in the domain of verbs – it maps from the count-domain to the mass-domain.

The examples in ( 4.153) illustrate this point again: 4.153 a. Şimdi, bu saatte, orada çorba ve ekmek dağılırdı.

now, this time:LOC, here soup and bred give.out:PASS:ir:PST [Johanson 1971: 178]

‘Now, at this time, soup and bread were given out here.’

b. O gün düşmanlar bile barışırdı. this day enemies even make.peace:ir:Pst [ibd.] ‘On this day, usually, even enemies made their peace with each other.’

The time adverbials şimdi (‘now’) and bu saatte (‘at this time’) in ( 4.153a) do not refer to one definite place in time, they do not locate the giving out of soup and bread. O gün (‘this day’) in ( 4.153b) does not locate one specific make their peace. Instead, make their peace is characteristic for enemies on that certain day. The temporally locating adverbials thus refer to a category of places in time that are all characterised by the giving out of soup and bread and of enemies making their peace (cf. Johanson 1971: 177). These utterances do not introduce events into the discourse, but characterise certain time spans by the denotations of the respective predicates. This is indicated by the use of the Aorist.

109 Cf. Hopper & Thompson (1984) for a cross-linguistic survey of this phenomenon. 110 Dahl (1995) proposes a correlation between formal and semantic unmarkedness on the one hand, and the

expression of genericity on the other hand. According to Dahl, the most general statement about generics – captured in terms of a ‘minimal marking tendency’ – is, that generic sentences tend to not contain any overt tense-aspect marking and employ the least marked tense-aspect choice in the respective language. Note, however, that this is not the case in Turkish, where a form which is marked for aspect – the Aorist – is used for generic statements. The fact that the Aorist is the least formally marked aspect form in Turkish might be taken as argument in favour of this tendency. However, since the use of the Aorist for generics is a rather new development in the Turkish language history, the minimal marking tendency does not seem to apply here.

192 Chapter 4: Semantics

To sum up, in English, Russian, and Bulgarian, the mapping from the count to the mass domain for verbs is not overtly expressed by specific markers, but results from the use of the aspectually unmarked form (which does not explicitly focus on a specific temporal tier unit) and contextual support. The atemporal, inactual and permanent readings as well as the modalised readings (e.g. potential, exemplifying) and habituality (where one is not referring to individuated, temporally localised events) can be regarded as results of a mapping from the count to the mass domain. In none of these cases the assertion is made for a topic time, but for a topic entity that is characterised in a specific way. In none of these cases, an individualised event is introduced into the discourse.

The situation is different in Turkish. The central point of this section has been that Turkish has an explicit marker for the grinding function – the Aorist. The analysis of the Aorist proposed in this section suggests that this form might indeed be regarded as expressing the mapping from count eventualities onto mass eventualities. This assumption is reflected in characterisations of the Aorist like that of G. Lewis (2000: 116) mentioned above, namely that the Aorist-marked yaparım means something like ‘I am a doer’, and not ‘I am doing’. The Aorist is characterised by the impossibility of temporal localisation. The fact that Aorist predicates cannot be located in time indicates that no event is introduced into the discourse (as these should be individuable by their temporal localisation), but rather a property that is ascribed to some participant (e.g., the grammatical subject) and temporally located via the temporal location of that participant. The temporal boundaries of the validity of this property or characterisation can be specified independently. If this is not possible, they per default coincide with the temporal boundaries pertaining the existence of that participant.

The analysis of the Turkish Aorist suggested here needs, of course, further elaboration.

4.5 Conclusion The central aim of this chapter was to develop a semantic analysis for grammatical aspect that is general enough to account for aspect cross-linguistically, but at the same time allows for the specifications needed to capture the language-specific data. In section 4.1, the cross-linguistic characterisation of grammatical aspect has been stated in terms of the relation between a topic time and some specific part of the denoted event. The selection of a temporal tier unit, the relation between I(e) and I(TT), and the boundedness-characteristics of the topic time interval constitute the elements of the basic semantic toolbox for aspect definitions cross-linguistically. The respective features and relations out of this toolbox are instantiated and specified differently (overt or by means of context) in different languages – but within predictable ranges: the relation may be one of inclusion (⊃), simultaneity (=) or being included (⊂), the selection may pertain to ϕdyn, τ or ϕstat, and the topic time interval may be unbounded, open-bounded or closed-bounded (cf. Table 4.15). Without a thorough analysis of the components involved in determining the semantics of aspect in general as well as its instantiation in specific languages, there is no way to account for the interpretation of aspectualised verbs in a systematic way.

With respect to Russian, the task was to provide a semantic basis that allows for a systematic account for the multitude of possible readings. The main focus has been on the ipf aspect, which has been assigned one invariant meaning in terms of a general overlap relation between I(TT) and I(e). This general relation has been argued to comprise three specific relations. How the postulation of these three relations is compatible with assuming the ipf aspect to be one gram, that is, to have one lexical entry and not three, will be the topic of chapter 5.

In sections 4.3 and 4.4, the Bulgarian and Turkish aspect systems have been analysed in order to figure out the different components that play a role in the semantic basis, and to test cross-linguistic applicability of the analysis proposed. Table 4.12 shows the means Bulgarian

193Chapter 4: Semantics

IPF TT ○ e

and Turkish use to express those readings of the Russian ipf aspect that have been considered here. These readings do not constitute clearly distinct meanings that could be enumerated in an a priori list. They are rather constructed in a given context, based on the respective semantic configuration.

Reading Russian Bulgarian Turkish 1 actual-processual ipf ipf I -mekteydi; -iyordu 2 conative ipf ipf I -iyordu 3 inactual/continuous ipf ipf I -iyordu; -irdi (Aorist) 4 durative ipf ipf I; ipf A -di 5 general-factual ipf ipf A -di 6 iterative ipf ipf I; ipf A -di, -iyordu 7 habitual ipf ipf I -irdi (Aorist) 8 potential ipf ipf I -irdi (Aorist) 9 permanent ipf ipf I -irdi (Aorist) 10 atemporal111 ipf ipf -ir (Aorist)

Table 4.12 The comparison of Russian, Bulgarian and Turkish with respect to the formal expression of these readings reveals an interesting three-way classification, where the respective groups are each based on one semantic configuration, corresponding to one of the three specific aspectual relations that have been proposed for Russian.

This is evident especially with the Turkish system that has morphological means to distinguish these three specific relations, compare Table 4.13:

Relation Reading of Russian ipf Turkish I. TT ⊂ ϕdyn

[TT open-bounded] actual-processual, conative -iyordu / -mekteydi

II. TT = e [TT unbounded]

habitual, continuous, potential, permanent, atemporal -irdi

III. TT ⊃ e [TT closed-bounded] general-factual, durative -di

Table 4.13 Interestingly, each of the Turkish aspect markers seems to allow for the construction of the same range of possible interpretations as the three specific relations proposed for the Russian ipf aspect (cf. chapter 5, section 5.1.2).

Moreover, the comparison of Russian, Bulgarian and Turkish has shown that analysing aspect in terms of just some vague notion of ‘imperfective’ and ‘perfective’ aspect obscures important differences. In order to maintain the notions of ‘perfectivity’ and ‘imperfectivity’ for cross-linguistic analyses, it is necessary to define them as meta-categories PERFECTIVITY and IMPERFECTIVITY. Furthermore, one has to be aware of the fact that these meta-categories may be brought about differently. It is exactly the differences within the general system of grammatical aspect that are responsible for the cross-linguistic differences in usage and interpretation. In this respect, Johanson’s (1996; 2000) terminality categories [±AD], [±INTRA] and [±POST] have proven useful, as they reveal the differences between the different kinds of ‘imperfective’ and ‘perfective’ aspect that have been taken for granted in different languages:

111 Statements that are atemporally valid may occur in the present tense only.

194 Chapter 4: Semantics

“It is thus essential to try to determine the orientation of the language-specific oppositions in the field of ‘perfectivity’ and ‘imperfectivity’ – namely intraterminality vs. nonintraterminality and adterminality vs. nonadterminality – as well as the roles the individual categories play as marked or unmarked members of them.” (Johanson 1996: 244) 112

The English progressive is different from both the Russian ipf aspect and Turkish -iyordu. Neither English nor Turkish have a form that expresses PERFECTIVITY in the same way as the Russian pf aspect does. On certain contextual conditions, the respective ‘non-imperfective’ forms can express a value similar to pf, but they do not have to. These differences between use and interpretation of the different IMPERFECTIVE- and PERFECTIVE-markers are explainable on the basis of the three terminality categories and the oppositions and markedness relations between their respective members: Russian is marked for adterminality, English and Turkish for intraterminality (cf. chapter 5, section 5.2.3 for a more detailed analysis).

LANGUAGE MARKER TERMINALITY-CATEGORY

ASPECTUAL META-CATEGORY

+INTRA -iyordu, -mekteydi, -irdi (Aorist)

intraterminality IMPERFECTIVITY

±INTRA -di pseudo-perfectivity PERFECTIVITY

+POST -miş postterminality IMPERFECTIVITY113

Turkish

±POST -di pseudo-perfectivity PERFECTIVITY

+INTRA -ing intraterminality IMPERFECTIVITY

English ±INTRA simple form pseudo-perfectivity PERFECTIVITY

+AD pf perfectivity proper PERFECTIVITY Russian

±AD ipf imperfectivity proper IMPERFECTIVITY

past A (pf/ipf)114

+AD non-past pf

adterminality PERFECTIVITY

past I (pf/ipf) Bulgarian

±AD non-past ipf

imperfectivity proper115 IMPERFECTIVITY

Table 4.14

112 In this specific paper, Johanson does not deal with postterminality. 113 Postterminality is an instance of IMPERFECTIVITY as it excludes any boundaries from the socpe of assertion. 114 The ipf A shows that in the past tense, [+AD] also pertains to the reaching of temporal boundaries, not

only to inherent ones. 115 The Bulgarian Imperfect can be regarded as an instance of intraterminality, because whenever a predicate

is used with the Imperfect, the topic time interval does not include its boundaries. This is different for the Russian ipf past, which allows for interpretations where the boundaries of I(TT) are included in the assertion. In the present tense, the boundaries of I(TT) cannot be part of that interval in neither language.

195Chapter 4: Semantics

Table 4.14 summarises the differences between the aspect systems of Turkish, English, Russian and Bulgarian in terms of the respective terminality categories, the markers instantiating them and the meta-categories the systems are based on. Note that in the Bulgarian past tense, Aorist and Imperfect play a role for the expression of aspect. Aorist forms denote events in the attainment of their limits – with regards to inherent limits in case of pf predicates or with regards to temporal limits in case of ipf predicates – and thus the Aorist pertains to the category of adterminality-markers, alongside with the pf aspect. The Aorist may confirm the default boundedness features induced by the pf aspect (closed-boundedness of TT). Otherwise, the Aorist may override the one introduced by the ipf aspect (open-boundedness or non-boundedness of TT). The fact that non-past forms behave like in Russian shows that the differences between both languages pertain mainly to the past tense.

Table 4.14 summarises the differences in bringing about PERFECTIVITY and IMPERFECTIVITY. These meta-categories may be based on different terminality-constellations. Taking into account the markedness-status of the language-specific categories, several subtypes can be distinguished (cf. also Johanson 1996), that differentiate between proper (marked) and pseudo (unmarked) instantiations of these meta-categories:

4.154 a. IMPERFECTIVITY:116 +Intra, ±Ad, +Post

b. PERFECTIVITY: ±Intra, +Ad, ±Post

These cross-linguistic types specify procedural information that is encoded – more or less specifically – in the respective language-specific instantiations: IMPERFECTIVITY delivers information like ‘topic time included in a certain phase, with topic time boundaries being excluded from the assertion’, PERFECTIVITY something like ‘all boundaries – inherent and actual – are included in the topic time interval’117. In both cases, this procedural information may be part of the semantics of the respective form (in the marked cases) or may arise as result of pragmatic strengthening and as default that may be subject to later cancellation. This means that language-specific aspect markers of these category types differ with respect to their procedural information within a predictable range. These differences, however, must not be ignored, as it is exactly them that are responsible for the differences in interpretation that are accounted for in terms of pragmatic principles (chapter 5).

The aspect systems of the languages discussed here relate to, and are based on, the event nucleus. The event nucleus represents the conceptual notion of the prototypical event and it covers the constituents an event may consist of: a preparation phase, a culmination point and a consequent state. These components may be lexically encoded by the predicate and, consequently, be accessed by aspect. This is how the three terminality categories are conceptually anchored.

The cross-linguistic characterisation of aspect can be stated as follows: Aspect relates a sentence base to a topic component. Languages differ with respect to their morphological means for encoding the different possibilities of how this general characterisation of aspect is instantiated. More precisely, languages differ with respect to which Aktionsarten are built up from the basic set of dynamic phase, static phase and boundary, with respect to which units

116 The languages analysed here have privative aspect oppositions. Therefore, there is no negative value listed

in Table 4.14. There may as well be languages with an equipollent aspect opposition in which case one member would express the positive the other the negative value.

117 Compare Bickel’s (1996: 199) cross-linguistic characterisation of IMPERFECTIVE and PERFECTIVE aspect in selection theoretic terms: “the imperfective selects exactly one ϕ” and “the perfective selects one or two subsequent τ”. Two comments are at order here: First, ‘imperfective’ and ‘perfective’ refer to the respective meta-categories; second, since ‘ϕ’ is not further specified, Bickel’s definition of IMPERFECTIVE allows to include also postterminality, just as assumed in this thesis.

196 Chapter 4: Semantics

are explicitly selected by the marked aspect operators and how they are related to the topic time, and with respect to the boundedness-features of the topic time intervals. The topic component does not have to be a topic time, but may as well be a topic entity. This is the case with unbounded topic times.

The following table shows how the basic components contributing to the semantic characterisation of grammatical aspect are modified in Russian, Bulgarian, Turkish and English. Note that this table only summarises the marked elements of the respective aspect categories:

TOPIC COMPONENT boundedness

RELATION I(TT) – I(e) includes/included /

simultaneous

SENTENCE BASEselection of ϕdyn τ ϕstat

MARKER

open-bounded I118 ⊂ ϕdyn -ing/ -iyordu /119

-mekteydi open-bounded I ⊂ ϕstat -miş closed-bounded A ⊃ τ pf120 non-bounded I = ϕdyn τ ϕstat -irdi (‘Aorist’)

Table 4.15 Of the three languages analysed here, Bulgarian is the only one that has explicit means to specify the boundedness feature of the topic time interval. The Imperfect (I) marks an unbounded or open-bounded interval, the Aorist (A) a closed-bounded interval. Turkish is the only language that has overt means to explicitly select the static phase (-miş) and to explicitly mark the topic time interval as being non-bounded (-irdi). If the topic-time interval is not bounded, the asserted situation does not consist of an event and its participants, but of a topic entity that is assigned certain properties specified by the predicate.

The analysis of aspect in Bulgarian and Turkish as compared to the Russian system has revealed interesting results. With respect to Russian, it has been shown that the topic time interval and its boundedness features are indeed important for adequately stating the semantics of aspect and the semantic basis for its interpretations. The TT-feature is not explicitly encoded in Russian. Therefore, the role it plays for the semantic characterisation of aspect might easily remain unnoticed. Moreover, it has been shown that it is justified assume the three specific relations for the Russian ipf aspect, since Turkish has means to encode them with different aspect markers. It has been shown that a small set of basic tools that may be specified within clearly delimited ranges accounts for the semantics of aspect cross-linguistically.

Figure 4.11 summarises the basic semantic configurations for the aspect markers in Russian, Bulgarian and Turkish121. Examples ( 4.155)-( 4.164) illustrate the respective relations:

4.155 a. Kogda ja vošel, on govoril po telefonu.

when I come.in:PST:pf, he speak:PST ipf at telephone:DAT ‘When I came in, he was making a telephone call.’

118 This column is valid for Bulgarian only. 119 The difference between the English progressive and -iyordu is not explainable in terms of the terminality

categories and the markedness relations alone – both are intraterminal and both are marked. Still, they differ in usage and interpretation. This can be accounted for only by taking into account the whole system of intraterminality in both languages. English has only one form to mark intraterminality, Turkish has three. This will be dealt with in more detail in chapter 5, section 5.2.3.

120 ‘pf’ pertains to Russian and Bulgarian. 121 This table does not distinguish marked from unmarked aspect operators. It merely lists the different

relations that may be conveyed the respective forms.

197Chapter 4: Semantics

b. Ty čital segodnja gazetu? you read:PST:ipf today newspaper:ACC ‘Have you read the newspaper today?’

c. Ja ne govorju po-arabski I not speak:PRS:ipf Arabic ‘I cannot speak Arabic.’

4.156 a. Ja vstretil ego na vokzale.

I meet:PST:pf him at station:LOC ‘I met him at the station.’

b. On zabolel i ne poseščaet lekcij. he get.ill:PST:pf and not attend:PRS:ipf lectures ‘He got ill and does not attend lectures.’

c. V 8 časov, ona uže vyšla. at 8 o’clock, she already go.away:Pst:pf ‘She had already left at 8 o’clock.’

4.157 a. Kogato tja izlizaše ot stajata, sreštna edin poznat. when she leave:I:ipf from room:DEF, meet:A:pf one friend ‘When she was leaving the room, she met a friend.’

b. Afrika se namira južno ot Sredizemno more. Africa be.located:PRS:ipf south of Mediterranean Sea. ‘Africa is located to the south of the Mediterranean Sea.’

4.158 a. Toj caruva trijset godini. he reign:A:ipf thirty years ‘He reigned for thirty years.’

b. Ivan dnes dochožda. Ivan today come:A:ipf ‘Ivan came today.’ (= He was here, but now he’s gone.)

4.159 a. Včera sreštnach tvojata prijatelka Ivanka. yesterday meet:A:pf:1Sg your:DEF friend Ivanka ‘Yesterday, I met your friend Ivanka.’

b. Az veče săm popălval anketata. I already AUX fill.in:P:ipf form:DEF

‘I have already filled in the form.’ (= I know how to do it.)

4.160 Çayım içmekteydim. tea:1SG drink:mekte:PST:1SG ‘I was (in the act of) drinking my tea.

4.161 a. Balon yavaş yavaş yükseliyordu. balloon slowly slowly ascend:iyor:PST. ‘The balloon was ascending slowly.’

b. Bu dünyaya bir kere geliyoruz. this world one time come:iyor:1PL ‘We come into this world only once.’

198 Chapter 4: Semantics

4.162 a. Bu kitabı okudun mu? this book:Acc read:Pst Q ‘Have you read this book?’

b. Anahtarı buldum. key:ACC find:PST:1SG ‘I found the key.’

c. Acıktım. get.hungry:PST:1Sg ‘I am hungry.’ (= I got hungry, and now I am in the state of being hungry.) 4.163 Gözlüklerimi kaybetmişim. glasses:1SG:ACC lose:miş:1SG ‘I lost my glasses.’ (= They are gone.)

4.164 Benim kasabım iyi et satar.

my butcher:POSS:1SG good meat sell:ir:3SG ‘My butcher sells good meat.’

The question that arises from the comparison of this data is, whether it is possible to account for the interpretation of the different aspect markers in a systematic way. This will be addressed in chapter 5, where focus will be on the pragmatic processes that underlie the interpretation of utterances containing Russian ipf predicates.

199Chapter 4: Semantics

Figure 4.11

example:

CULMINATIONPOINT τ

CONSEQUENT STATE ϕstat

PREPARATIONPHASE ϕdyn

event nucleus: Aktionsart:

] inaccompli [

[ accompli ]

[achevé ] ] attributif [ Desclés’ continuum:

3. Turkish 3.1 mekteydi ] [ (4.160) 3.2 iyordu a) ] [ (4.161) b) 3.3 di a) [ ] (4.162) b) [ ] c) [ [ 3.4 miş ] [ (4.163)

3.5. irdi (‘Aorist’) (4.164)

2. Bulgarian2.1 ipf I a) ] [ (4.157) b) 2.2 ipf A a) [ ] (4.158) b) [ ] 2.3 pf A [ ] (4.159) 2.4 pf P [ [

1. Russian 1.1 ipf a) ] [ (4.155) b) [ ] c) 1.2 pf a) [ ] (4.156) b) [ [

c) ] ]

5 PRAGMATICS

In chapter 3, semantics was defined in terms of information that is explicitly encoded, pragmatics in terms of information that has to be inferred. The respective inferences need to be reliable and predictable, since otherwise communication would hardly be as successful as it seems to be1. For pragmatic explanations to be reliable and predictable, it is necessary to spell out not only the respective mechanisms, but also the encoded information, that is, the semantic basis these mechanisms work on. Chapter 4 has outlined a semantic analysis of grammatical aspect. This semantic information, however, merely provides a template that needs to be supplemented by additional information supplied by pragmatic reasoning processes.

The present chapter is concerned with the pragmatic principles that apply in interpreting the Russian ipf aspect, and with the way these principles interact with the semantically provided information. The aim is to describe the regularities involved in interpreting utterances containing aspectualised verbs. It will be shown that semantics alone is not sufficient to account for the interpretation of aspect (section 5.1). This is no shortcoming of semantics, if it is embedded into a pragmatic framework (section 5.2).

Acknowledging the interaction of semantics and pragmatics2 does not only allow to state the language-specific semantic basis in rather simple terms, but also allows for a systematic cross-linguistic account of aspect (section 5.2.3). The semantic basis for aspect can be kept stable across languages, if interpretations of aspectualised verbs are not confused with semantics, but accounted for by pragmatic principles instead: “[M]uch of the confusion is due to a neglect of pragmatics in aspectology, especially when it comes to typological issues or the analysis of undescribed languages. Aspect definitions often mistake generalised conversational implicatures as part of the semantics of the form.” (Bickel 1996: 31)

The pragmatic analysis developed in this chapter focuses on the interpretation of the Russian ipf aspect. The analysis of the Bulgarian and Turkish aspect systems presented in chapter 4 was intended to substantiate the claims that have been made for the underspecified Russian aspect system, and to show the cross-linguistic applicability of the analysis. The exact specification of the interpretation process for aspect forms in Bulgarian and Turkish is beyond the scope of this thesis, but it is reasonable to assume that it proceeds roughly along the lines that will be proposed for Russian. The pragmatic principles assumed are general principles of pragmatic reasoning. Their outcomes differ depending on the respective semantic basis they interact with.

To figure out the interaction of semantics and pragmatics in the interpretation of grammatical aspect, the contributions of aspect and Aktionsart have to be separated. Both provide different kinds of information that interact on a specific level but differ in the pragmatic processes that may be needed to supplement the respective encoded information. This is shown in Figure 5.13: aspect and Aktionsart encode features that match at an

1 Yet there is hardly a one-to-one correspondence between intended speaker meaning and hearer’s meaning. 2 One account of grammatical aspect that considers pragmatics is delivered by C. Smith (1997), who argues

as follows: “Since we are concerned with the meanings conveyed by the linguistic forms, we must distinguish between semantic meaning and the implications that may arise from its presence in a sentence. We ask whether an interpretation is due to semantic meaning alone or whether it includes pragmatic inference. To answer the question we will put the sentence in contexts where the interpretation might be expected to vary. If an interpretation remains invariant, it is conveyed by the linguistic form and is part of the semantic meaning of the sentence” (C. Smith 1997: 67). However, this approach does not consider how aspect and Aktionsart interact, what the respective encoded information is and what information needs to be inferred. These specifications are important if pragmatics is intended to play a systematic role in utterance interpretation (cf. Figure 5.1).

3 Figure 5.1 is valid for the Russian ipf aspect only, because aspect markers of other languages may need different kinds of inference, depending on their semantics.

Chapter 5: Pragmatics

202

operational interface, but both differ with respect to the kind of information that needs to be inferred. The general relation of the ipf aspect has to be disambiguated to one of the three specific relations. With respect to the Aktionsart basis it might be necessary to induce missing features (phases and boundaries) in order to provide the appropriate input for aspect to apply. The former will be dealt with in terms of ambiguity and sense-generality (section 5.1.2), the latter in terms of accommodation and inferences (section 5.2.2).

Figure 5.1 The need to infer additional information is no shortcoming of semantics. Semantics does not have to account for certain processes. Some contributions to the interpretation process are better delegated to pragmatics, which keeps the semantics stable, clear and simple. Note Dascal’s (1981: 160) warning that “if pragmatics is required to do too much […] there is reason to fear that it will accomplish too little.” The same can be claimed with respect to semantics. Thus, a balanced division of labour is needed, with explicitly stated contributions of both semantics and pragmatics and an outline of their interaction.

5.1 The semantic basis

The semantic analysis developed in chapter 4 provides an analysis of aspect that captures the encoded information and provides a suitable basis for pragmatic inferences to apply. This section deals with the semantic factors and their integration into the interpretation process in more detail. Section 5.1.1 summarises the semantic framework developed so far, and section 5.1.2 focuses on the different kinds of underspecification involved in the semantic characterisation of aspect interpretation. Both sections reveal the limits of purely semantic analyses and illustrate the need for a pragmatic embedding. If pragmatics is conceded its appropriate role, semantics actually does not need to account for all the details in utterance interpretation4.

4 Remember Levinson’s claim that a proper elaboration of pragmatic principles may constrain the shape of

semantic representations.

INTERPRETATION

ENCYCLOP. ENTRY: AD HOC CONCEPTS

WORD MEANING

conceptual information (Aktionsart)

procedural information (aspect)

KIND OF INFORMATION

inferring:adjustment of required input (ϕdyn, ϕstat, τ)

decoding

decoding

inferring:the specific relation between I(e) and I(TT)

ϕdyn, ϕstat, τ

LOGICAL ENTRY: GCI

Chapter 5: Pragmatics

203

5.1.1 Encoding: Semantic information In chapter 4, the semantics of aspect has been stated in terms of the different components that contribute to the encoded information. The present section is concerned with the role the encoded information plays in the interpretation process. The focus will be on the question of how this information is figured out and combined, and on the ways in which it contributes to the interpretation process. The semantic skeleton is quite abstract. It is not reasonable to enrich it further though, since this would make it too specific and too much tied to certain contexts.

What is encoded by grammatical aspect is a relation between a specific part of the sentence base and a topic component. In case of the Russian ipf aspect, this is an overlap relation, the exact specification of which is not explicitly encoded, but needs to be inferred. The different factors that play a role in deriving the final interpretation of an ipf form in Russian have been specified in chapter 4. These factors are the internal make-up of the encoded event, the boundedness characteristics of the asserted time interval and the selection specification of grammatical aspect. The selection specification constitutes an input requirement aspect imposes onto its Aktionsart basis in order to operate on it by selecting a certain element and relating it to the topic time. Languages differ with respect to the factors that are explicitly encoded, and with respect to the factors that have to be induced pragmatically.

Based on the analysis developed in chapter 4, the following components of the semantic contribution to aspect interpretation (especially of the ipf aspect) in Russian can be specified: 5.1 a. the semantics of pf and ipf and their opposition relation

b. the Aktionsart basis

c. interaction of aspect and Aktionsart

d. one general relation for the ipf aspect

e. three specific relations for the ipf aspect Figure 5.2 illustrates how these factors contribute to the interpretation of Russian ipf aspect:

Figure 5.2

Aktionsart + aspect • default combinations • deviations from default

ipf – general relation • relation 1 • relation 2 • relation 3

ipf – specific relation • reading 1 • reading 2 • reading 3 • ...

contextually relevant interpretation

semantic markedness • pf - marked • ipf - unmarked

underspecification

underspecification

specificity

opposition

disambiguation

composition

Chapter 5: Pragmatics

204

The semantics of the aspect markers and their status within the respective opposition constitute the basis for aspect interpretation as sketched in Figure 5.2. The general semantics of aspect consists in the relation between I(TT) and I(e). The marked pf aspect is specified in terms of an inclusion relation between the boundary/boundaries (inherent or actual) provided by the Aktionsart and the topic time: ‘I(TT) ⊃ I(τ)’. The matter is slightly more complicated with the semantically unmarked5 ipf aspect. A purely structuralist markedness opposition would state its semantics in a negative way, as ‘not perfective’. However, as has been shown in chapter 4, it is possible to establish a definite word meaning for the ipf aspect, albeit a rather general and abstract one. It can be characterised in terms of a general overlap relation ‘I(TT) ○ I(e)’, with the condition ‘ϕ ⊆ e’, i.e., a phase to be encoded by the Aktionsart basis. This general relation has to be specified to one of three specific relations: 5.2 I(TT) ○ I(e) | ϕ ⊆ e

a. I(TT) ⊂ I(ϕdyn)

b. I(TT) = I(e)

c. I(TT) ⊃ I(e) The disjunctive analysis in ( 5.2) assumes the three specific ipf relations to be clearly distinct and specifiable, which in turn suggests that the general relation for the ipf aspect is ambiguous between them6. Interpreting an utterance containing an ipf verb thus requires the resolution of this ambiguity. The disambiguation again yields an underspecified representation that still has to be supplemented by additional information in order to establish the contextually relevant interpretation. However, disambiguation already restricts the range of possible ipf interpretations to a considerable degree.

The basis for the application of aspect, i.e. the assertive part of the sentence base (Klein 2001), is provided by the Aktionsart in terms of the temporal tier units ϕdyn, τ and ϕstat. The interaction of aspect and Aktionsart comprises default cases and cases of deviation from the default. The default case in Russian is the combination of pf aspect and terminative predicates on the one hand, and of ipf aspect and aterminative predicates on the other hand (Bohnemeyer & Swift 2001, 2004; cf. chapter 2, section 2.2.3). Deviations from default aspect assignment are morphologically marked on the verb, as is shown in ( 5.3): 5.3 a. On otkryval dver’. he open<yva>:PST:ipf door ‘He was opening the door.’

b. Rebenok kriknul. child cry<nu>:PST:pf ‘The child cried out.’ The default for the <ϕdyn, τ>-predicate in ( 5.3a) would be the pf aspect. The deviation from this default is indicated by the suffix -(y)va-. In ( 5.3b), the suffix -nu- indicates that the <ϕdyn>-predicate is not assigned the default ipf, but the pf aspect. For the application of the pf aspect, a boundary has to be present as input. If the required input is not provided, as is the case in ( 5.3b), it has to be induced. Deviation from default aspect assignment thus may not only consist in selecting an element that is not the default-selection, but also in mismatches

5 The notion of semantic markedness pertains to specificity or informativity. The pf aspect is semantically

marked because it is more specific with respect to the relation encoded than the ipf aspect. 6 The exact nature of this ambiguity will be dealt with in section 5.1.2.

Chapter 5: Pragmatics

205

between aspect and Aktionsart. In this latter case, the Aktionsart basis has to be accommodated by the induction of the respective element in order to provide the suitable input for the aspect marker to apply. In case of ( 5.3b), a boundary has to be induced, in case of ( 5.4), a phase is required for the ipf aspect to apply: 5.4 a. On vyigryval gonku.

he win:PST:ipf race:ACC ‘He won / was winning / used to win / … / the race.’

b. On vyigryval gonku, kogda on upal i ušibsja. he win:PST:ipf race:ACC, when he fall:PST:pf and hurt.himself:PST:pf ‘He was winning the race, when he fell and hurt himself.’

c. On vyigryval gonku tri raza. he win:PST:ipf race:ACC three times ‘He won the race three times.’ In ( 5.4a), the ipf aspect is applied to the <τ>-predicate vyigrat’ gonku (‘win the race’). Since the ipf aspect needs a phase to apply, an operator-operandum mismatch arises – the Aktionsart does not provide a phase. This mismatch, and the resulting deviation from default aspect assignment, is morphologically indicated by the suffix -yva-. The induction of the necessary phase may be achieved by different means yielding different interpretations, as indicated in ( 5.4a). Two of these possibilities are shown in ( 5.4b,c): in ( 5.4b), the phase is ‘added’, resulting in the structure <ϕdyn, τ>. The dynamic phase is then zoomed in as indicated by the kodga-(‘when’)-clause, which results in the relation ‘TT ⊂ ϕdyn’. This relation gives rise to, e.g., the actual-processual or the conative reading. The resolution of the mismatch is achieved differently in ( 5.4c), where the iterative adverbial tri raza (‘three times’) pluralizes the event, which then extends over a period of time, i.e. over a phase. This constitutes the appropriate input for the application of the ipf aspect7. These mismatch resolving operations are generally referred to in terms of ‘coercion’ or ‘recategorisation’, and are treated in more detail in section 5.2.

The markedness opposition specifies the pf aspect as semantically marked, i.e. as being more specific and more informative, and the ipf aspect as being unmarked, i.e. less specific and less informative. The pf aspect can be assigned one single, specific relation (‘TT ⊃ τ’), whereas the ipf aspect is characterised by one general relation that needs to be disambiguated to one of the three specific relations mentioned in ( 5.2a-c). The disambiguation involves specifying the topic time TT and the relevant part of the Aktionsart that is selected and related to TT8. Figuring out TT requires the specification of its boundedness feature as unbounded (in which case it more adequate to speak of a topic entity), open-bounded or closed-bounded. It is not always clear, however, which element serves to specify the topic time (cf. Klein 2001: 21f). In Russian, the specification of TT is possible only indirectly by means of temporal adverbials, context or discourse mode. There are also particles that suggest a certain TT: uže (‘already’), for example, specifies TT as closed-bounded, adverbs like medlenno (‘slowly’) or postepenno (‘gradually’) as open-bounded.

Another factor that helps to specify TT is the position of the observer, the točka otsčeta TO (cf. chapter 4, section 4.1.2). This position of the observer may be lexically given by temporal

7 This is the case especially in Russian, but not in all Slavic languages. Cf. chapter 4, section 4.2.2, for the

difficulties concerning repetitive events. 8 Grønn (2003: 169) takes the size of TT to prefer a specific interpretation of the ipf aspect. However, he does

not distinguish between the semantics of aspect in terms of the three different specific relations, and their respective interpretation. This will be decisive in section 5.1.2. The point here concerns the role of TT in the disambiguation of the general ipf relation.

Chapter 5: Pragmatics

206

adverbials, but it may also be provided by discourse mode (Padučeva 1996). A retrospective TO suggests that TT is closed-bounded, a synchronous one allows for an unbounded and an open-bounded TT. Thus, particles that unambiguously specify TO as retrospective at the same time specify the TT-interval as closed-bounded. TO may be given by temporal conjunctions like kogda (‘when’), by temporal adverbials like v vosem’ časov (‘at eight o’clock’) or by particles like uže (‘already’)9. Padučeva (1996: 12-15) lists the following factors that may determine TO: 5.5 a. discourse mode

b. adverbials indicating durativity and iterativity

c. lexical meaning of the verb A dialogue setting in most cases points out a retrospective TO. Particles like the deictic vot (‘there’) unambiguously mark such a dialogue setting and therefore specify TO as retrospective: 5.6 Vot na ėtoj stene visela kartina

there at this wall:LOC hang:PST:ipf painting [Padučeva 1996: 13] ‘There, at that wall, hung a painting.’

Durative adverbials specify an interval within which an action took place and thus indicate a retrospective TO: 5.7 Ja guljala s desjati do dvenacati. I take.a.walk:PST:ipf from ten:GEN to twelve:GEN [Padučeva 1996: 14] ‘I took a walk from ten to twelve.’ Contrary to deictic particles, such adverbials do not specify TO immediately, but rather specify TT as closed-bounded. The retrospective TO follows from this characterisation of TT.

Adverbials indicating the number of times an action took place only allow for a retrospective TO. In order to specify a definite number of executions of an action, the standpoint of the speaker/hearer has to be retrospective to the time the assertion is made about: 5.8 J segodnja dva raza guljal c sobakoj. I today two times take.a.walk:PST:ipf with dog:INST [Padučeva 1996: 14] ‘Today, I took the dog for a walk twice.’ According to Padučeva, adverbials used to answer an implicit or explicit kogda-(‘when’)-question specify TO as synchronous if they appear in a thematic position ( 5.9a), but as retrospective if they appear in a non-thematic position ( 5.9b): 5.9 a. Večerom ja pisal otčet. Pozvonil telefon. in.the.evening I write:PST:ipf report. ring:PST:pf telephone [Padučeva 1996: 14] ‘In the evening, I was writing a report. The telephone rang.’

b. Ja peredaval emu žurnal v dva časa. I hand.over:PST:ipf him magazin at two o’clock [Padučeva 1996: 14] ‘I handed him over the magazine at two o’clock.’

9 It is not always clear whether some adverbial specifies TT or TO. Both are sometimes hard to tease apart

since they are mutually dependent to a certain degree.

Chapter 5: Pragmatics

207

These adverbials, however, seem to specify TT rather than TO. It is the second sentence in ( 5.9a) that specifies TO as synchronous. Such a context is missing in ( 5.9b), and therefore TO is likely to be retrospective.

Adverbials of the type vse bol’še (‘more and more’) indicate a synchronous TO: 5.10 Koster gorel vse jarče.

camp-fire burn:PST:ipf more brighter [Padučeva 1996: 14] The camp-fire was burning brighter and brighter.

Furthermore, there are certain verbs that unambiguously specify TO. Videt’ (‘see’), for instance, allows only a retrospective TO: 5.11 Ja videl po televizoru novosti.

I see:PST:ipf on television:DAT news [Padučeva 1996: 15] ‘I saw the news on TV.’ As this discussion indicates, it is not always easy to decide whether a certain adverbial specifies TO or TT (Klein 2001: 20). TO and TT interact very closely – once TO is specified, the possibilities for TT are constrained to a certain degree and vice versa. Both TO and TT therefore help to disambiguate the general overlap relation to one of the three specific relations10.

Examples ( 5.12)-( 5.14) summarise the contribution of TO and TT to specifying the specific aspectual relation for Russian ipf: 5.12 Kogda ja vošel, ona čitala gazetu.

when I come.in:PST:pf, she read:PST:ipf paper:ACC When I came in, she was reading the paper.

Kogda (‘when’) in ( 5.12) specifies TO as synchronous with respect to the asserted interval. The specific relation is thus ‘TT ⊂ ϕdyn’. 5.13 Ja uže rasskazyval vam ėtu istoriju.

I already tell:PST:ipf you this story. I already told you this story.

Τhe particle uže (‘already’) in ( 5.13) suggests that TO is retrospective, and, consequently, the relation is ‘TT ⊃ e’. 5.14 On rabotal v universitete.

he work:PST:ipf at university. He worked at university. (= He was a teacher at university).

( 5.14) is slightly more complicated, since there is no overt clue that would give an indication of TO or TT. In this case, world knowledge makes the relation ‘TT = e’ most probable, yielding the interpretation ‘He was a teacher at university’. In principle, however, the other two relations would be possible as well. As example ( 5.14) shows, world knowledge is another important factor that helps to disambiguate the general ipf relation.

10 This interdependence of TT and TO is not surprising, given the fact that they instantiate the dual characters

of ‘reference time’ (cf. chapter 4, section 4.1.2).

Chapter 5: Pragmatics

208

The fact that in many cases the factors relevant for disambiguation have to be inferred, indicates that pragmatics plays a role at this step of aspect interpretation (cf. section 5.1.2).

The disambiguated relations do not yet supply all the information needed in order to arrive at the contextually relevant interpretation. Rather, they are still underspecified and need to be elaborated by means of context, background assumptions and conceptual knowledge. The three dots with the different readings in Figure 5.2 indicate that it is not possible to state the readings in form of a list. This list would get rather long and would not be fixed once and for all. The items on that list, i.e. the various readings, could not be enumerated in a definite way. This is supported by the ongoing discussion about how many and which readings for the ipf aspect have to be postulated.

Semantic composition is commonly assumed to be the main mechanism guiding the interpretation process. As has been pointed out in chapter 3, compositionality applies only at the level of word and sentence meaning, i.e. with the stable, invariant information explicitly encoded by the lexical items used. With respect to ipf aspect, such an invariant, explicitly encoded information that contributes to semantic composition is encountered at the level of the general and specific relations. However, the specific relations allow for different interpretations that are derived by inferring information that is stored in the encyclopaedic knowledge associated with the respective Aktionsart basis or provided by the wider context. As has been argued in chapter 3, encyclopaedic knowledge does not belong to the encoded information contributing to the interpretation process. It is variable in different respects. Compositionality thus does not apply in this domain. There is no way to compositionally derive the contextually-relevant interpretations without getting lost in the vastness of the encyclopaedia. If a semantic theory of aspect interpretation intends to achieve this as well, it would be too specific and too much tied to certain contexts and would no longer be able to ensure the stability and predictability of interpretation. As has been mentioned at the beginning – if semantics is required to do too much, it will probably accomplish too little.

The explicitly encoded information constitutes a template that has to be specified (from the general relation to the specific relations) and enriched (from the specific relation to the interpretation). This specification and enrichment is part of the pragmatic reasoning involved in utterance interpretation. As will be argued in section 5.2, both processes are in different nature and need to be explained by different principles – in terms of GCI in the former case, in terms of enrichment and loosening (ad hoc concept formation; e.g. Carston 2002, 2004b) in the latter11.

The model presented in Figure 5.2 is static insofar, as it does not refer to the processes involved in putting together the different pieces of information, and it does not incorporate the possibility of backtracking. Such a static model presents the stable, encoded parts of meaning, but it does not capture the flexibility involved in the interpretation. Figure 5.2 suggests that the derivation proceeds from one step to the next, in a strictly incremental way. This contradicts the fact that default interpretations can be cancelled by the following context, which is due to their non-monotonic, inferential character.

The limitations of semantics in accounting for the interpretation of the Russian ipf aspect is illustrated in ( 5.15):

11 GCI theory is advocated by neo-Gricean pragmatic theories, enrichment and loosening by relevance-

theoretic accounts. As the example of aspect interpretations shows, both accounts can be regarded as complementing each other instead of being in competition (cf. also Saul 2002). Instead of taking an either-or position, it should rather be acknowledged that both theories serve to account for different mechanisms that all play a role in interpretation.

Chapter 5: Pragmatics

209

5.15 On ničevo ne delal, tol’ko ležal i kuril. he nothing NEG do:PST:ipf, only lie:PST:ipf and smoke:PST:ipf [Stoll 2001: 85]

(i) He wasn’t doing anything, just lying (around) and smoking.

(ii) He didn’t do anything, only lay (around) and smoke.

( 5.15) may describe either a specific situation (i) or a general habit (ii). The interpretation depends on the broader context since the encoded information does not provide any clues considering which relation is the relevant one.

The same holds for cases where the context gives rise to a specific default interpretation, which may again be overridden by the following context: 5.16 a. Moj otec govoril po-turecki. my father speak:PST:ipf Turkish ‘My father could speak Turkish.’

b. Moj otec govoril po-turecki, kogda ja vošel. my father speak:PST:ipf Turkish when I come.in:PST:pf ‘My father was speaking Turkish, when I came in.’

5.17 a. Perevodčik govoril po-turecki.

translator speak:PST:ipf Turkish ‘The translator was speaking Turkish.’

b. Perevodčik govoril i po-turecki. translator speak:PST:ipf also Turkish ‘The translator could also speak Turkish.’ In both ( 5.16) and ( 5.17) the actual-processual, the habitual and the potential reading would be possible (Glovinskaja 1982)12. In ( 5.16a), the potential reading is the most probable one. However, it can be overriden by the following context as in ( 5.16b). In ( 5.17a), the preferred interpretation is the actual-processual one, because the potential reading would not be relevant in this case – translators are supposed to be able to speak certain languages (besides their mother tongue). Thus, this interpretation would not constitute the necessary non-trivial update of the common ground, i.e. of the presupposed information13. A slight modification of the context, though, makes the potential reading again the preferred one, as is shown by the insertion of i (‘also’) in ( 5.17b)14. The fact that the interpretations in ( 5.16a) and ( 5.17a) may be overridden by the following context demonstrates that they have to be regarded as defaults and not as semantically encoded, stable meaning. Since the respective changes in context also result in an override of the aspectual relation – ‘TT = e’ → ‘TT ⊂ ϕdyn’ in ( 5.16), and ‘TT ⊂ ϕdyn’ → ‘TT = e’ in ( 5.17) –, the specific ipf relations constitute defaults as well. In section 5.1.2, it will be discussed how this default status of the specific relations can be accommodated with the assumption that they are part of the semantics of the ipf aspect.

These examples show that the pragmatic notion of default plays a role already at the subpropositional level. Pragmatic reasoning is involved in disambiguating the general ipf relation to one of the three specific relations, but this specification may again be overriden by

12 This is the case in a sentence uttered out of the blue, where the hearer is free to construct an appropriate

context in which this sentence is interpreted. Otherwise, TO or TT would be fixed and would allow either for the actual-processual or for the potential / habitual reading.

13 This is not to deny that the speaker might intend exactly this reading, and intend to convey additional implicatures operating on the propositional content. Implicatures of this kind are not at issue here.

14 Thanks to Uwe Junghanns for pointing this out to me.

Chapter 5: Pragmatics

210

the following context. Furthermore, the knowledge needed to interpret the sentences in ( 5.16) and ( 5.17) is not encoded by the items used. The fact that translators can speak certain languages is not semantic knowledge but pertains to the encyclopaedic entry15 activated by the lexical items used. If it were semantically encoded information, it could not be overriden. In this way, pragmatics plays already a role in establishing ‘what is said’.

The following questions arise from the account delivered so far: First, how is it possible to accommodate one general ipf relation with three specific relations and the different interpretations they give rise to? Second, how can this be accommodated with the assumption of ipf constituting one gram? And third, how are the default interpretations and the deviations from these default accounted for? These questions do not need to be addressed by semantics, they have to be accounted for in pragmatic terms. Pragmatics is based on and interacts with semantic information. Thus, pragmatics is involved in specifying the general relation in terms of one of the three specific relations. Moreover, it plays a role for arriving at the contextually relevant interpretation by taking into account the encyclopaedic knowledge activated by the Aktionsart basis.

This section has outlined the contribution of the explicitly encoded information provided by aspect and Aktionsart to the interpretation of Russian ipf aspect. The encoded information constitutes the semantic component of the interpretation process. It has been argued against a too broad conception of semantics since this would make it too specific and too context-bound to be predictable and systematic. Such a broad conception would delegate too much work to semantics and thus make it rather powerless. Restricting semantics to the information encoded by the aspect marker and the Aktionsart basis and their respective interaction preserves it as a powerful means ensuring compositionality and thus reliability in interpretation. This semantic component has to be complemented by pragmatic mechanisms allowing that the meaning can outstrip that explicitly encoded information. Such mechanisms are provided by inferential heuristics giving rise to default inferences. These heuristics have to be based on principles of communication like plan recognition, goal sharing and accommodation (Thomason 1992: 352).

5.1.2 Decoding: Ambiguity and sense-generality As has been shown in the section above, the semantics of the Russian ipf aspect is defined in terms of one general overlap relation. This general relation is captured by a meta-variable ‘IPF’ that has to be specified in a given context to one of the three specific relations. In order to substitute the meta-variable for one of the specific relations, the topic time and/or the position of the speaker (TO) has to be figured out, as well as the relevant part of the sentence base that is asserted and the relation established between both. The specification of the exact relation is a rather complex matter of mutually interacting and constraining features and conditions. The main difficulty consists in the fact that neither of these features is consistently and unambiguously overtly expressed in Russian. Temporal adverbials may specify both TO or TT, but even if TO is specified as being synchronous, the exact ‘I(TT)-I(e)’ relation is not yet unambiguously given. The semantic factors contributing to the interpretation process have been illustrated in Figure 5.2. What is evident in this figure is the double underspecification involved in the derivation of aspect interpretation – underspecification of the general relation, and underspecification of the specific relations.

The disjunctive analysis of ipf aspect in terms of the three specific relations suggests that the underspecification of the general relation is an instance of ambiguity. However, this contradicts the conception of the ipf aspect as one gram with one lexical entry – at least, if

15 Remember that the lexical entry of one lexical items specifies three different (sub-)entries: a lexical, a logical

and an encyclopaedic one (chapter 3, section 3.1.1).

Chapter 5: Pragmatics

211

ambiguity is conceived in the traditional way as being caused by different lexical entries sharing the same surface realisation. This section will show that the ambiguity of ipf aspect is compatible with the assumption of one single lexical entry, if it is analysed as an instance of pragmatic ambiguity (Lascarides et al. 1996; cf. chapter 3, section 3.1.2).

The second instance of underspecification pertains to the relation of the disambiguated specific ipf relations to their respective interpretations. One central claim here is that the contextually relevant interpretations of the ipf aspect need not, and in fact cannot, be listed. The underspecification of the specific ipf relations thus cannot be of the same kind as the underspecification encountered with the general ipf relation. Rather, this type of underspecification constitutes an instance of sense-generality (Atlas 1989) characterised by underdetermination16.

Sense-generality is not resolved by selecting an interpretation from a predefined list, but by constructing an interpretation according to the information supplied by context and according to the encyclopaedic knowledge activated by the respective lexical item. At this level, there is no selection of meanings, but a construction of interpretations: “In interpreting the utterances of sense-general sentences, we are not selecting from the readings of an ambiguous sentence; we are constructing from a definite but general sense and from collateral information a specific content.” (Atlas 1989: 31)

The ambiguity of the Russian ipf aspect consists in its three specific relations that each give rise to different labelled bracketings (see below) and thus have different truth conditions17. This is plausible considering once again the nature of these relations: aspect is part of the finiteness component, as it marks an assertion by relating the sentence base to the topic time. The three relations differ in the exact nature of this relation, that is, they differ with respect to the part of the sentence base that is asserted to hold at a certain topic time interval. A difference in assertion causes a difference in truth values; thus the three relations each have different truth conditions. The ambiguity encountered with the ipf aspect concerns its logical entry and its interaction with the logical entry specified by the Aktionsart. The sense-generality encountered with the disambiguiated specific relations is based on the encyclopaedic entry of the Aktionsart (cf. Figure 5.1).

Traditionally, ambiguity is ascribed to lexical items that can be assigned multiple lexical entries. However, the discussion above suggests that it is neither desirable nor necessary to postulate multiple lexical entries for the Russian ipf aspect. But, if there is only one single lexical entry for ipf, how can it be justified to speak of ambiguity? At this point, a purely semantic analysis has to be enriched by pragmatics. To detect the ambiguity involved in the interpretation of ipf verbs, pragmatics has to be considered.

The next question that arises facing these two instances of underspecification is how they can be accommodated. How is the ipf aspect to be analysed – as pragmatically ambiguous or as sense-general? In this section it will be argued that both kinds of underspecification do not exclude each other. They are located at different levels of interpretation and play different roles in the interpretation process.

Turkish provides evidence for the postulation of the three specific relations for ipf, since it has morphological means to express each of these relations (cf. chapter 4, section 4.4.1). The relevant facts are summarised in Table 5.1:

16 Remember that ambiguity is characterised by overdetermination – the respective item gives rise to

different, clearly distinct specifications, from which the relevant one has to be chosen. 17 This does not contradict the assumption of this ambiguity to be pragmatic. As has been argued in chapter

3, section 3.2.1, pragmatics applies already at the subpropositional level and may thus be truth-conditionally relevant. The equation of truth-conditional vs. non truth-conditional with semantics vs. pragmatics is not tenable.

Chapter 5: Pragmatics

212

IPF TT ○ e

Relation Reading of Russian ipf Turkish I. TT ⊂ ϕdyn

[TT open-bounded] actual-processual, conative -iyordu /

-mekteydi

II. TT = e [TT unbounded]

habitual, continuous, potential, permanent, atemporal -irdi

III. TT ⊃ e [TT closed-bounded]

general-factual, durative -di

Table 5.1 The three possible relations for the ipf aspect and the interpretations they may give rise to18 will henceforth be referred to as group I, II and III, respectively. It is more reasonable to refer to three basic relations than to three main, or core, meanings of the ipf aspect, as is done, e.g., by Stankov (1980) and Grønn (2003). Stankov (1980) divides the semantics of ipf aspect into three main meanings19: the general-factual (obšto-faktičeskoto značenie), the actual-processual (konkretno-procesnoto značenie) and the continuous meaning (postojanno-procesnoto značenie). Grønn (2003: 22) assumes three core meanings for the Russian ipf aspect: the processual, the habitual-iterative and the factual. Approaches like these are problematic due to the fact that they take specific interpretations for core meanings, which in turn are suggested to constitute the semantics of aspect. In chapter 2, the difficulties have been pointed out that arise if semantics and interpretation of aspect are not kept apart. Core meanings are context dependent and therefore cannot constitute a reliable basis for a semantic analysis of aspect. Semantics corresponds to the explicitly encoded part of information and is as such stable across contexts. This is partly acknowledged by Stoll (2001), who replaces the semantic notion of core meanings by the pragmatic notion of core functions. She states two main functions for the ipf aspect – the durative (which seems to correspond to the actual-processual) and the general-factual. As both functions do not apply in any case and with any verb, and since they may be overridden by the following context, they are not part of the ipf semantics but constitute pragmatic implicatures. Another difficulty arising form approaches like Stankov’s or Grønn’s is that stating the meaning of aspect in terms of main interpretations refers to the interpretation of aspect-Aktionsart combinations and does not figure out the semantic specification of aspect alone. In order to state the semantics of aspect, the semantic contribution of the Aktionsart basis has to be ignored.

To sum up so far, instead of defining the ipf aspect in terms of ‘core meanings’, it is more appropriate to refer to three specific relations which constitute three possible semantic bases for its interpretation. Each relation allows for various interpretations, depending, among other things, on the respective Aktionsart basis. Table 5.1 summarises the semantic analysis of ipf interpretations in Russian: there is one invariant meaning ‘IPF’ that is basic to all ipf uses. It constitutes the semantic basis for the ipf aspect, and gives rise to three distinct specifications.

The presumption of these three relations is supported by the comparison with the Turkish intraterminal system where there are three distinct morphological expressions for each of the three relations. By the ‘substance principle’ (Zwicky & Sadock 1975: 5) this may indicate a systematic ambiguity in the case of Russian:

18 Note again that the readings listed in this table are not to be understood as clearly distinct meanings of the

respective relations. Neither does this listing claim that it is exactly these readings that are possible. Compare the remarks on the status of aspect readings in chapters 2 and 4.

19 The iterative meaning is left out, as Table 5.1 is concerned with the three first order relations (cf. chapter 4). Stankov deals with aspect in Bulgarian, but as has been illustrated in chapter 4, the basic ipf relations are the same for Bulgarian and Russian.

Chapter 5: Pragmatics

213

“If a distinction in understanding is a systematic ambiguity in some language, then that distinction is potentially realizable by a formal mark in some other language.”

However, this does not have to be the case, as Zwicky & Sadock (1975: 5) emphasise: “[F]rom the fact that a particular semantic differentia is simple and that it is formally marked in some language we can conclude nothing about the status of this distinction for any particular example in any language; both lack of specification [sense-generality, B.S.] and systematic ambiguity are consistent with these facts.”

Accordingly, the fact that Turkish has three morphologically distinct IMPERFECTIVE forms corresponding to the relations postulated as specifications of the Russian meta-variable ‘IPF’ may indicate both – systematic ambiguity or sense-generality of the relations in Russian.

Carston (2002: 273) points out that if a single form in L1 (here: Russian) has two understandings, and if these are translated into L2 (here: Turkish) by means of two forms, then the form in L1 does not necessarily have to be ambiguous. It is rather the case that no decision can be made with respect to ambiguity and sense-generality (cf. also Zwicky & Sadock 1975). Carston argues that the reasoning has to be as follows: If a single form in L1 has two understandings and this form is translated into all/most other languages by a single form, which also evinces the same understandings, then the form is not linguistically ambiguous, but sense-general with regards to the two understandings. According to Carston, this is due to the fact that the interpretive distinctions arise quite systematically due to general cognitive / pragmatic factors acting on a single lexical item.

Applied to Russian, this means that it is not the general ipf relation that is ‘translated’ from Russian into Turkish, but one of the three specific relations. As is shown in Table 5.1, the corresponding Turkish aspect markers exhibit the same range of possible interpretations as the Russian specific ipf relations. These relations are not overtly differentiated in Russian, but overtly marked in Turkish. Since Turkish has different markers for the expression of the three specific relations postulated for Russian, the partition of the general IMPERFECTIVITY-field that has been claimed to hold for Russian, appears to be quite systematic (cf. also the English progressive marker that instantiates one of these relations). Furthermore, these partitions are clearly distinct and enumerable. Both facts give supportive evidence for an ambiguity analysis of Russian ipf aspect at this level.

To shortly recap, the ipf aspect instantiates the general notion of IMPERFECTIVITY, which is pragmatically strengthened to one of the three specific relations. Although these three specific relations pertain to encoded, that is, semantic information, the processes leading to the disambiguation are of pragmatic nature. The results of the disambiguation process are defeasible. Moreover, disambiguation may have to refer to encyclopaedic knowledge about the typical length of events that helps to specify the relation of I(TT) and I(e). Since the specific relations do not yet supply all the information necessary for interpretation but are sense-general, the question whether the Russian ipf aspect is ambiguous or sense-general is no either-or-question (cf. chapter 3, section 3.1.2). One and the same expression may be both ambiguous and sense-general, depending on the level of investigation.

The question whether Russian ipf aspect is ambiguous or sense-general is also discussed in Grønn (2003). Grønn takes the ipf aspect to be vague (i.e. sense-general), since its semantics only licenses the general overlap relation ‘e ○ t’. This is the topologically least informative relation and therefore needs pragmatic strengthening to arrive at a more specific relation (2003: 168f). Since an aspectualised verb is always associated with some topic time, pragmatic strengthening always occurs in a non-ambiguous context. Therefore, according to Grønn, the general relation is not ambiguous but sense-general. He captures the range of ipf readings in terms of interpretational restrictions induced by the nature of the topic time. The role of TT is very much the same as proposed here. Grønn (2003: 170f) postulates two principles that license the factual and the progressive ipf: the general relation is strengthened to the factual ipf if it is possible to include the event in the topic time, and it is strengthened to

Chapter 5: Pragmatics

214

the processual ipf if the topic time is small compared to what would constitute the normal length of the temporal trace of the event. Except for the fact that he does not mention the third possible relation for ipf, namely ‘TT = e’, Grønn seems to take criteria comparable to those presented here in order to delimit the specific ipf relations. But he does not consider them to indicate ambiguity of ipf aspect. His argument against ambiguity derives mainly from the inappropriateness of ambiguity tests, the outcome of which he takes to be as given in advance. Therefore, such tests do not prove that ipf is genuinely ambiguous.

Grønn supports his claims this with the following examples (2003: 36): 5.18 Maša probovala baltijskoe pivo. I Katja tože.

Maša taste:PST:ipf Baltic beer. And Katja also. was tasting was ‘Maša Baltic beer. And so Katja.’ tasted did

5.19 Maša i Katja probovali baltijskoe pivo.

Maša and Katja taste:PST:ipf Baltic beer were tasting ‘Maša and Katja Baltic beer.’

tasted The different readings are licensed by different topic times. The elided material and the reduced conjunction in ( 5.18) and ( 5.19) have the same topic time as the overt form. Since Grønn takes the topic time to be given in advance, he concludes that this test does not prove the ambiguity of ipf; it does not rule out the possibility of one general, vague ipf meaning that varies with different topic times. Since the existence of different readings is a necessary but not a sufficient condition for ambiguity, he argues that the ipf aspect has to be treated as vague and ascribes it one general meaning that has to be compatible with its heterogeneous set of readings in context (35f). Grønn claims that treating the ipf as ambiguous requires the postulation of several ipf viewpoint operators, which would be incompatible with the assumption of one general meaning for the ipf aspect. He draws evidence in favour of his vagueness account from cross-linguistic facts, namely from similar patterns of semantic diversity of formal markers of ipf aspect. This diversity has to be reducible to some universal concept of imperfectivity, and the vagueness approach permits such a unified semantics for each member of the category of aspect.

Several problems arise with Grønn’s argumentation. He does not distinguish between the two levels of the interpretation process that are considered decisive here: semantics in terms of the general and the pragmatically strengthened relations, and the different ipf readings. Furthermore, by assuming TT to be given in advance, he does not need the disambiguation step and takes this as an argument against ambiguity. Nevertheless, he assumes TT to play a decisive role for the strengthening of the general ipf overlap relation to one of the specific relations. By arguing that TT is always given in advance, he leaves out this step of the interpretation process and take this as evidence against ambiguity – although he seems to presuppose it at the same time.

In this section, three arguments against Grønn’s rejection of ambiguity will be delivered: First, taking into account the concept of pragmatic ambiguity makes it possible to accommodate the general semantics of the ipf aspect with different, clearly distinct specifications. Second, Grønn’s argument that the respective topic times in ( 5.18) and ( 5.19) are given in advance does not invalidate the ambiguity approach. On the contrary, it is exactly the fixing of the topic time that helps resolving the ambiguity of the ipf aspect. Giving the topic time in advance is possible by means of triggering expressions that require a certain

Chapter 5: Pragmatics

215

relation between I(TT) and I(e). Third, the cross-linguistic facts in chapter 4 have shown that the meta-category IMPERFECTIVE may be instantiated differently in different languages (cf. also section 5.2.3). An account of the Russian ipf aspect in terms of pragmatic ambiguity does not contradict a cross-linguistic characterisation of IMPERFECTIVE. The semantic diversity Grønn alludes to as argument against an ambiguity approach only applies with the specific interpretations allowed by one relation. However, it is the relations, that are semantically encoded, and exactly these relations can be taken as evidence for or against ambiguity.

Ambiguity is commonly considered as arising from one expression having more than one underlying syntactic or semantic representation (e.g. Zwicky & Sadock 1975), whereas general expressions are assumed to have only one single representation. The most prominent characteristic of ambiguity is that an ambiguous expression gives rise to distinct truth values, depending on the state of affairs it refers to. A sentence containing such an expression is alternatively judged true and judged untrue with respect to a fixed state of affairs and a fixed occasion of use (Gillon 2004: 183). Thus, the alternate truth value judgement test is one possible means to detect ambiguity: “Let α be an expression. Let δ( ) be an expression frame such that δ(α) is a sentence liable to being judged with respect to a truth value. Let s be a state of affairs. If δ(α) is alternately judged true and judged not true with respect to s, then α is prima facie ambiguous.” (Gillon 2004: 166)

Gillon’s approach is interesting insofar as he considers ambiguous entities as being neither semantic representations nor formulae, but labelled bracketings. Ambiguity thus arises from expressions that are characterised by different underlying bracketings: “An expression is ambiguous if and only if it can be assigned at least two distinct labelled bracketings.” (Gillon 2004: 165)

Labelled bracketings are obvious with structural ambiguity, where the syntactic structure allows for different constituent structures: 5.20 Bill saw a man with a telescope.

a. Bill [VP saw [NP a man] [PP with a telescope]]

b. Bill [VP saw [NP a man [PP with a telescope]]] [Gillon 2004: 159]

Lexical ambiguity is a special case of distinct labelled bracketing. In this case, the labelled bracketing is the limiting one that comprises a single node and its label: 5.21 They found hospitals.

a. [S [NP they] [VP found [NP hospitals]]]

b. [S [NP they] [VP findperf [NP hospitals]]]

c. [S [NP they] [VP foundpres [NP hospitals]]] [Gillon 2004: 162f]

Found in ( 5.21a) is a shorthand for two specifications with the same phonological value and syntactic category but different morphosyntactic features of different verbs and thus different senses: third person plural past tense of to find as in ( 5.21b), or third person plural present tense of to found as in ( 5.21c). Thus the labelled bracketing in ( 5.21a) comprises in fact two bracketings that differ with respect to the label assigned to the minimal verb constituent. These minimal constituents and their labels constitute a minimal bracketing, and thus the word form found is ambiguous in the sense defined above.

Ambiguity may also arise with single lexical entries (Lascarides et al. 1996; chapter 3, section 3.1.2). In this case, ambiguity is due to different default interpertations that are characterised by different bracketings each. The effects are the same as for ambiguities based

Chapter 5: Pragmatics

216

on different lexical entries. ( 5.22a) is an instance of a semantic ambiguity, as there are multiple lexical entries for bank, whereas ( 5.22b) illustrates an instance of pragmatic ambiguity, as there exists one general lexical entry for brush:

5.22 a. John walked towards the bank this morning and so did Peter.

b. # Rembrandt used a brush and so did our janitor. Bank in ( 5.22a) does not yield a zeugma effect, since the interpretations of both lexical entries are equally possible with both conjuncts. Both John and Peter are equally likely to go to a financial institution, and both are equally likely to go to a riverside. As long as bank is interpreted equally in both conjuncts, the interpretation is fine. This is different with brush in ( 5.22b). The two conjuncts of this sentence give rise to two different specifications of brush (‘paint brush’ and ‘cleaning brush’), which are hard to accommodate and thus make the sentences odd. However, there is one general entry for brush, as indicated in ( 5.23): 5.23 a. The grandparents gave the children brushes for Christmas.

b. The grandparents gave the children brushes for Christmas: one got a paint brush, the other one got a hair brush.

In ( 5.23a), brush does not have to be specified, but may be specified. ( 5.23b) indicates that brush is not ambiguous. It is acceptable even though there are two different specific senses of brush.

Lascarides et al. attribute the zeugma effect arising from ( 5.22b) to conflicting pragmatic cues that give rise to default inferences about interpretation and discourse coherence. The default interpretations conflict and the hearer cannot solve this conflict. To explain the resulting effect, Lascarides et al. propose to link semantics to pragmatic reasoning – and this is exactly what will be pursued here for the interpretation of Russian ipf aspect. The notion of pragmatic ambiguity makes it possible for the Russian ipf aspect to be conceived as one gram with one lexical entry, which is ambiguos at the same time – pragmatically ambiguous – between the three possible specifications of its general semantics. Moreover, Russian ipf aspect is not an instance of ambiguity with respect to conceptual information as is the case with the examples presented above, but an instance of ambiguity with respect to procedural information (cf. Figure 5.1).

The conflicts in judging the acceptability of the respective utterances are due to different default interpretations that suggest different labelled bracketings, i.e. the assignment of different I(TT)-I(e) relations. Tests based on the assumption of identity of sense thus pertain to the identity of aspecutal relations. The requirement of a specific I(TT)-I(e) relation can be referred to in terms of selection restrictions. As has been shown in chapter 4 (cf. also Table 5.1), the three basic relations for Russian ipf aspect are each based on a specific TT-interval. Thus, topic-time triggers serve to fix the relevant relation.

Besides the alternate truth value judgement test mentioned above, several further tests may be used in order to detect ambiguity. Each test relies on a specific basic assumption that has to be made explicit. In what follows, some of the ambiguity tests established by Zwicky & Sadock (1975) and Gillon (2004) will be outlined and applied to Russian ipf aspect. These tests basically rely on acceptability judgments and on assumptions concerning coordination, pronouns and ellipsis, giving rise to certain effects that indicate ambiguity.

The assumption behind the first test concerns the distribution over coordinated arguments (Gillon 2004: 175):

Chapter 5: Pragmatics

217

“If one neighboring coordinated constituent satisfies a lexeme’s selection restriction, then all the neighboring coordinated constituents do as well.”

If there is one expression-frame, i.e. function, that takes two coordinated expressions, i.e. arguments, then both arguments have to satisfy the selection restrictions imposed by the function, cf. ( 5.24a). If two function-argument combinations are fine on their own as in ( 5.24b,c), but not possible in the coordinated structure as in ( 5.24d), this indicates the ambiguity of the function: 5.24 a. function(argument1 ∧ argument2)

b. function(argument3)

c. function(argument4)

d. #function(argument3 ∧ argument4) This assumption is basic to the argument coordination, or zeugma test, where the coordination of two grammatically congruent expressions as arguments of one function is not judged acceptable. This is illustrated by the ambiguity of take in ( 5.25a) and expire in ( 5.25b): 5.25 a. # He took his hat and his leave. [cf. Lascarides et al. 1996]

b. # John and his driver’s license expired on Tuesday. [Cruse 1986: 61] Although the individual parts of the conjunctions in ( 5.25) are fine on their own, their coordination yields a conflict in interpretation. This conflict cannot be solved by the hearer thus yielding a zeugma effect. In ( 5.25a), take constitues the function into which the coordinated arguments his hat and his leave are inserted. Both he took his hat and he took his leave are okay, but the coordination take(his hat and his leave) is unacceptable. The arguments his hat and his leave seem to satisfy different sense restrictions of take, and thus indicate its ambiguity. The same holds for ( 5.25b) with expire on Tuesday as the frame and John and his driver’s license as the respective coordinated expressions.

Argument coordination requires identical relations for both conjuncts, namely relations that satisfy the selection restrictions imposed by the triggering expression. In case of Russian ipf aspect, such triggers may be temporal or manner adverbials, certain kinds of clauses, or questions. Questions about whether some action has taken place at all trigger the relation ‘TT ⊃ e’. Questions of the kind čto ty delal včera v vosem’ časov (‘What were you doing yesterday at eight o’clok) or čto ty dealeš sejčas (‘what are you doing now’) trigger the relation ‘TT ⊂ ϕdyn’. Questions about characterisations of participants or about habits trigger the relation ‘TT = e’.

Examples ( 5.26)-( 5.30) illustrate the argument coordination test based on the test concerning the distribution over coordinated arguments, as illustrated in ( 5.24). The underlying assumption states that neighbouring coordinated constituents need to satisfy a lexeme’s selection restrictions. The form of the test is ‘function(argument1 ∧ argument2)’. The frame for two or more coordinated expressions can be imposed by expressions triggering a certain relation. The scope of the triggering expressions may thereby extend beyond the sentence boundary.

Kogda (‘when’) in ( 5.26) triggers the relation ‘TT ⊂ ϕdyn’. This relation serves as frame, or function20, whose sense restriction – the respective relation – is realised by the predicate of the first sentence (i.e. the argument of the function), but not by the predicate of the second sentence. Since the second sentence is the immediate continuation of the first, it should exhibit the same relation in order for the discourse to be coherent:

20 Since the aspectual relation constitutes the expression frame, or function, that takes a predicate as

argument, this also expresses the assumption of grammatical aspect dominating the VP.

Chapter 5: Pragmatics

218

5.26 Kogda ty pozvonila, ja čitala rasskaz Dama s sobačkoj. #No vse zabyla. when you call:PST:pf, I read:PST:ipf novel Dama s sobackoj. But everything forget:PST:pf

‘When you called, I was reading the novel Dama s sobačkoj. #But I have forgotten everything.’

The second sentence of ( 5.26) does not fit the relation triggered by kogda, but rather presupposes the relation ‘TT ⊃ e’. Thus the continuation sounds odd. The application of this test works as follows: 5.27 a. Ja čitala rasskaz Dama s sobačkoj, kogda ty pozvonila. → [TT⊂ ϕdyn] I read:PST:ipf novel Dama s sobackoj, when you call:PST:pf ‘I was reading the novel Dama s sobačkoj, when you called.’

b. Ja čitala rasskaz Dama s sobačkoj, no vse zabyla. → [TT ⊃ e] I read:PST:ipf novel Dama s sobackoj, but everything forget:PST:pf ‘I read the novel Dama s sobačkoj, but I have forgotten everything.’

c. # Kogda ty pozvonila, ja čitala rasskaz Dama s sobač.. # [V ___ ] no vse zabyla. when you call:PST:pf, I read:PST:ipf novel Dama s sobackoj. But everything forget:PST:pf

# ‘When you called, I was reading the novel Dama s sobačkoj. #But I have forgotten everything.’

If one neighbouring coordinated constituent satisfies a lexeme’s sense restriction, the other coordinated constituents have to do so as well. In ( 5.27c), the first coordinated constituent constituent (čitala) satisfies the sense restriction imposed by the kogda-sentence (‘TT ⊂ ϕdyn’). However, the other coordinated constituent (___, no vse zabyla) doesn’t. The verb satisfies the sense-restriction imposed by vse zabyla, namely ‘TT ⊃ e’. Basically, čitala is able to satisfy both sense restiction as shown in ( 5.27a,b). In ( 5.27c), however, it is positioned between two sense restrictions which gives rise to a conflict in interpretation thus pointing out the ambiguity of the relations that are possible for čitala.

The question in ( 5.28) indicates a dialogue setting, i.e., a discourse mode that suggests a retrospective TO. Retrospective TO’s trigger the relation ‘TT ⊃ e’. The relation in the answer, however, is ‘TT ⊂ ϕdyn’ (triggered by kogda). Therefore it is pragmatically not appropriate: 5.28 Q: Ty [TT ⊃ e čitala] rasskaz Dama s Sobačkoj?

you read:PST:ipf novel Dama s Sobačkoj? ‘Have you read the novel Dama s Sobačkoj?’

A: # Da, [TT ⊂ ϕ __ ] kogda ty pozvonila včera. yes, when you call:PST:pf yesterday ‘Yes, when you called yesterday.’

The time specification in the question of ( 5.29) indicates that the relation is ‘TT ⊂ ϕdyn’. The same relation has to be present in the answer and therefore, ja spala can only receive an actual-processual interpretation: 5.29 Q: Čto ty delal v 7 časov?

what you do:PST:ipf at 7 o’clock ‘What were you doing at 7 o’clock?’

A: Ja spala. I sleep:PST:ipf ‘I was sleeping.’

Chapter 5: Pragmatics

219

In ( 5.30), v svobodnoe vremja (‘in the free time’) indicates that the question is about a general habit: 5.30 Q: Čto ty [TT = e delaeš’] v svobodnoe vremja? what you do:PRS:ipf in free time ‘What do you do in your free time?’

A: Ja [TT = e begaju, ezžu na velosipede, čitaju, vstrečajus’ s druzjami]… I run:PRS:ipf, ride:PRS:ipf on bike, read:PRS:ipf, meet:PRS:ipf with friends ‘I jog, bike, read, meet friends…’ Due to the present tense, only two out of the three relations are possible in ( 5.30): ‘TT = e’ and ‘TT ⊂ ϕdyn’. The question is about a general habit, and accordingly it triggers the relation ‘TT = e’. Thus, the predicates in the answer need to receive an interpretation out of group II.

The next test, the predicate coordination test, is based on the following assumption (Gillon 2004: 176f): “If two constituents are coordinated, then their corresponding sense restrictions must be simultaneously satisfied.”

This assumption states that if two functions are coordinated, then their shared argument must fulfill the sense restrictions of both, as illustrated in ( 5.31a). If two functions are individually fine with the same argument, as in ( 5.31b,c), but the coordination of the functions and their application on this argument is not possible, as in ( 5.31d), this indicates that the argument is ambiguous, as it seems to satisfy different sense restrictions of the two frames: 5.31 a. function1 ∧ function2(argument)

b. function3(argument)

c. function4 (argument)

d. #function3 ∧ function4(argument) These considerations are basic to the predicate coordination test, where the coordination of two argument taking expressions applied to one and the same argument is judged unacceptabel. The ambiguity of newspaper in ( 5.32) is revealed by this test: 5.32 # The newspaper fell off the table and fired the editor.

Both the newspaper fell off the table and the newspaper fired the editor are fine, but fall and fire(newspaper), i.e. the conjunction of the frames and its application on the expressions as in ( 5.32), is not possible. This points out the ambiguity of newspaper, as it does not simultaneously satisfy the sense restrictions imposed by fall and fire.

In ( 5.33), the predicate coordination test is applied to the ipf aspect. Remember the test structure ‘function1 ∧ function2(argument)’. With respect to aspect, the framing expressions trigger certain relations which correspond to the sense restrictions that must simultaneously be satisfied by the expression inserted in the frame. In ( 5.33) the relations (functions) are triggerd by chorošo and v vosem’ časov, the inserted expression (argument) is igrat’21:

21 This example is problematic insofar, as its oddness may also be attributed to the poor compatibility of the

adverbials used. This might obscure the fact that is at issue here, namely that both adverbials trigger different relations for the ipf aspect of the predicate.

Chapter 5: Pragmatics

220

5.33 # Ivan chorošo, i Viktor v 8 časov igral na skripke. Ivan well, and Viktor at 8 o’clock play:PST:ipf on violin.

(i) Ivan chorošo [TT = e igral] na skripke. ‘Ivan could play the violin very well.’ (= He was a good violin player.)

(ii) V vosem’ časov, Vladimir [TT ⊂ ϕdyn igral] na skripke. ‘At 8 o’clock, Vladimir was playing the violin.’

# ‘Ivan [TT = e played] the violin well, and Viktor [TT ⊂ ϕdyn __ ] at 8 o’clock.’ The structures chorošo igrat’ na skripke and v vosem’ časov igrat’ na skripke are fine. The coordination of the predicates to chorošo ∧ v vosem’ časov(igrat’ na skripke) as in ( 5.33) is odd. In ( 5.33) the default interpretations (i) and (ii) of the two conjuncts are characterised by different aspectual relations triggered by the respective adverbials. Chorošo and v vosem’ časov have different sense restrictions and thus trigger different relations, which are not simultaneously satisfied by igral. The adverbials require different relations that conflict in the coordinate structure, and this conflict cannot be solved by the hearer.

Example ( 5.34) is based on the same test, and also requires that for coordinated constituents, the corresponding sense restrictions have to be satisfied. In ( 5.34) the aspectual relations are not triggered by adverbials as in ( 5.33), but by the respective predicates. Note that although the lexical content of the predicates is decisive for fixing the relations, it is not the lexical content that is tested for ambiguity – it is not conceptual ambiguity that is at issue, but ambiguity of the procedural information: 5.34 # Ja govorju po-arabski i čitaju gazetu.

I speak:PRS:ipf Arabic and read:PRS:ipf paper:ACC

(i) ## I [TT = e speak Arabic] and I [TT ⊂ ϕdyn am reading the paper].

(ii) # I [TT ⊂ ϕdyn am speaking Arabic] and [TT ⊂ ϕdyn reading the paper].

(iii) # I [TT = e speak Arabic] and [TT = e read the paper]. The oddness of ( 5.34) can be explained by the different default relations attributed to the two predicates. Basically, both verbs govorit’ (‘speak’) and čitat’ (‘read’) allow for all three ipf relations. The insertion of the verbs into the predicates of the two conjuncts in ( 5.34), however, gives rise to different default relations: ‘TT = e’ for the first conjunct and ‘TT ⊂ ϕdyn’ for the second. In interpretation (i), the two different relations are coordinated, which is the reason for the oddness. The interpreations in (ii) and (iii) each involve the same relation in both conjuncts, but assigning ‘TT ⊂ ϕdyn’ to ja govorju po-arabski and ‘TT = e’ to ja čitaju gazetu contradicts the default assignments (cf. Glovinskaja 1982: 59). Therefore, (ii) and (iii) sound odd, even though they realise the same ipf relation. In this case, coordination yields a strange effect comparable to the zeugma effect in ( 5.22b). The reason for this is that there are different default interpretations arising from the combinations speak Arabic (default: group II, Table 5.1) and read the paper (default: group I, Table 5.1). Although (ii) and (iii) are not instances of crossed-readings, they conflict by their default interpretations. This conflict can hardly be resolved and thus yields the zeugma effet. The oddness of ( 5.34) is supported by the parallel construction that suggests a parallel in the aspect relations of the predicates. This parallel conflicts with the default relations triggered by the lexical content.

In ( 5.34), the lexical contents of the respective verbs strongly suggest a specific relation. There are also verbs that exclude certain relations, a fact that can be taken as further support for ambiguity. With lexically stative verbs that do not contain a dynamic phase in their

Chapter 5: Pragmatics

221

semantic representation, it is hard to get the relation ‘TT ⊂ ϕdyn’ without considerable contextual pressure. Thus ( 5.35) sounds odd with readings based on this relation22: 5.35 #V pjat’ časov on vesil 78 kg.

at five o’clock he weigh:PST:ipf 78 kg #‘At five o’clock, he was weighing 78 kg.’

Another test for ambiguity makes use of expessions such as so + the auxiliary verb to do (cf. also Lakoff 1970). The basic assumption here is that so + to do stands in for its antecedent (Gillon 2004: 182): “The denotation of the endophoric expression [pronoun, so+to do, ellipsis; B.S.] is determined by whatever sense can be ascribed to its antecedent.”

This assumption is basic to identical sense tests (Gillon 2004: 181) such as, for instance, conjunction reduction, the possibility of crossed-readings or the possibility of reference by an endophoric expression to an antecedent.

The ambiguity of decline in ( 5.36) is revealed by this test: 5.36 # Julia declined a cocktail and Iris ___ an irregular Latin noun. [Zwicky & Sadock 1975: 20] Furthermore, conjunction reduction is possible only if the omitted part is the same in both conjuncts: 5.37 I went to a bank this morning and so did my brother. For two conjoined clauses to be eligible for conjunction reduction, they must be of the form X-A-Y and X-B-Y, where A and B are consituents of the same type (Zwicky & Sadock 1975: 18). Thus, for the conjunction reduction to be possible in ( 5.37), both instances of bank have to refer to the same concept – either to the financial instituation or to the riverside.

Another test for ambiguity is the impossibility of crossed readings with conjunction reduction. Conjunction reduction is only possible if the respective elements have the same sense. This is illustrated in ( 5.38):

5.38 Morton and Oliver tossed down their lunches. [Zwicky & Sadock 1975: 18] Logically, there are four possible interpretations: either both Oliver and Morton toss their lunches on the floor, or both eat them very hastily, or Oliver tosses his lunch on the floor and Morton eats it hastily and vice verca. However, out of these four possibilities only two are actually possible, namely those where both Oliver and Morton do the same. The fact that ( 5.38) lacks crossed-readings is indicated by ( 5.39), where additional contextual information would favor different understandings of the two predicates (Zwicky & Sadock 1975: 18) and thus makes the sentence odd: 5.39 # Morton, as always a greedy eater, and Oliver, who continued to refuse food on

principle, tossed down their lunches. If only two of the four readings are possible, this indicates the ambiguity of the respective item (here: toss).

Identical sense tests are used in ( 5.40) and ( 5.41):

22 The same holds for the English translation, as the English progressive requires a dynamic phase as input.

Chapter 5: Pragmatics

222

5.40 Ivan i Viktor igrali na skripke. Ivan and Viktor play:PST:ipf on violin:LOC

(i) ‘Ivan and Viktor [TT ⊂ ϕdyn were playing] the violin.’

(ii) ‘Ivan and Viktor [TT = e could play] the violin.’

(iii) ‘Ivan and Viktor [TT ⊃ e played] the violin.’ In order for the conjunction reduction to be possible in ( 5.40), the overt and the reduced elements must be of the same type, i.e. the relation in both conjuncts has to be the same (cf. 5.38). Thus igrali in ( 5.40) can either have the relation ‘TT ⊂ ϕdyn’ as in (i), ‘TT = e’ as in (ii) or ‘TT ⊃ e’ a in (iii). Crossed-readings in terms of different relations in both conjuncts are not possible.

Example ( 5.41) corresponds to the identical sense test in ( 5.36). Identity of sense refers to identity of relation. Conjunction reduction is only possible if the deleted part in both conjuncts belongs to the same group of readings, i.e. if both are characterised by the same aspectual relation: 5.41 Ja čitala ėtot roman i Marija tože.

I read:PST:ipf this novel and Marija also

(i) I [TT ⊃ e read the novel], and so did [TT ⊃ e __ ]Marija.

(ii) I [TT ⊂ ϕdyn was reading the novel], and so was [TT ⊂ ϕdyn __ ] Marija.

(iii) # I [TT ⊃ e read the novel], and so was [TT ⊂ ϕdyn __ ] Marija.

(iv) # I [TT ⊂ ϕdyn was reading the novel], and so did [TT ⊃ e __ ] Marija.

In ( 5.41), only two out of the four logically possible interpretations (i)-(iv) are acutally valid, namely only those with the identical relations, i.e. (i) and (ii).

Discourse mode also plays an important role as a triggering factor (Lascarides et al. 1996; Vazov & Lapalme 2000; Smith 2003), but it is not dealt with here in detail. One example for the influence of discourse mode has been given in ( 5.34) above. Another example is ( 5.45) below, where the discourse mode ‘description’ allows only for readings out of group II. Moreover, as has been pointed out repeatedly, the discoure mode ‘dialogue’ suggests the relation ‘TT ⊃ e’, i.e. a reading out of group III.

The results of the above tests concerning the different ipf relations support the presumed ambiguity of ipf aspect with respect to its general relation. The general overlap relation is an instance of pragmatic ambiguity, which is compatible with the postulation of only one lexical entry for the ipf aspect. The fact that the respective disambiguated relations can be overriden by the following context and thus have the status of defaults supports the assumption of this ambiguity being indeed pragmatic: 5.42 a. Moj otec govoril po-turecki. my father speak:PST:ipf Turkish ‘My father could speak Turkish.’

b. Moj otec govoril po-turecki, kogda ja vošel. my father speak:PST:ipf Turkish when I come.in:PST:pf ‘My father was speaking Turkish, when I came in.’ The default relation in ( 5.42a) is ‘TT = e’. In ( 5.42b) this default is overriden by the following kogda-clause that triggers the relation ‘TT ⊂ ϕdyn’. Since the specific relations constitute the

Chapter 5: Pragmatics

223

semantics of the ipf aspect, this example illustrates that defaults play a role with the specification of the semantic basis as well, and not only with respect to the final interpretation.

However, this is not the end of the story. Table 5.1 shows that the process of disambiguation alone does not yet specify the contextually relevant interpretation. The aspect relations are still sense-general and in need of further specification by means of wider context, background assumptions and conceptual knowledge. They constitute the common denominator for all the possible interpretations that may be constructed from this basis. From a linguistic point of view, it is not possible to establish a fixed list of all readings a priori. This is even acknowledged by Bondarko (1990), who otherwise proposes various discrete readings: “Častnye značenija NSV, o kotorych idet reč’ v aspektologičeskich issledovanijach, vrjad li isčerpyvajut vsju semantičeskuju zonu NSV.” (Bondarko 1990: 19) [The specific readings of the ipf aspect, which are the topic of aspectological studies, hardly exhaust the whole semantic zone of the ipf aspect.]

Interpreting an aspect form in context does not mean selecting an interpretation from a certain list, but constructing an interpretation according to encoded information, information inferred from context, background assumptions and encyclopaedic knowledge (cf. Atlas 1989)23. Each of these disambiguated relations may still give rise to several different interpretations. Examples ( 5.43)-( 5.45) illustrate the sense-generality of each relation24:

5.43 Kogda ja vošla, brat [TT ⊂ ϕdyn rešal] zadaču. Ja emu pomogala,

when I come.in:PST:pf, brother solve:PST:ipf exercise:ACC. I him help:PST:ipf, i nakonec-to my uspeli. and finally we succeed:PST:pf

‘When I came in, my brother was solving / was trying to solve an exercise. I helped him and at the end we managed (to solve) it.’

Kodga in ( 5.43) triggers the relation ‘TT ⊂ ϕdyn’, which in turn may give rise to different interpretations. Therefore, it does not matter, whether rešal zadaču is interpreted as was solving the exercise, or was trying to solve the exercise. Both interpretations are compatible with the relation triggered by kodga, and with the second sentence of the utterance.

5.44 Q: V detstve, ja [TT = e / TT ⊃ e igrala] na flejte. A ty?

in childhood:LOC, I play:PST:ipf on flute:LOC. And you? ‘When I was a child, I played the flute. And you?’

A: Da, igrala – ne chorošo, no často. Každyj raz, kodga u nas byl prazdnik, yes, play:PST:ipf – not well, but often. Every time when at us was holiday, ja pytalsja igrat’ bez ošibok, no mne nikogda ne udavalos’. I try:PST:ipf play:INF:ipf without mistake:GEN, but me never NEG succeed:PST:ipf

‘Yes, I played it – not well, but often. Every time when we had a holiday, I tried to play it without mistakes, but I never managed to do so.’

The question in ( 5.44) excludes the relation ‘TT ⊂ ϕdyn’, since for this relation to hold it has to be possible for the topic time interval to be included in the run time interval of the dynamic phase. This is excluded by the TT specified in ( 5.44) (v detstve). The other two relations, ‘TT ⊃ e’ and ‘TT = e’, are both possible. The answer in ( 5.44) realises the relation ‘TT = e’,

23 This is different from NLP applications that need to have a fixed list of possible readings in order to state

the appropriate rules of their derivation. 24 The necessity to assign definite interpretations to the respective forms is merely due to the purpose of

illustration and does not contradict the assumption of sense-generality.

Chapter 5: Pragmatics

224

allowing for different interpretations: the potential, the habitual and the iterative, which may all be realised without yielding contradiction.

In ( 5.45) it is the discourse mode ‘description’ that requires the relation ‘TT = e’:

5.45 Ja vspominaju o deduške: on guljal so vnukami I remember:PRS:ipf about grandpa: he go.for.a.walk:PST:ipf with grandchildren s nimi igral v futbol, kuril trubku, chorošo igral v šachmaty, … with them play:PST:ipf footbal, smoke:PST:ipf pipe, well play:PST:ipf

‘I remember grandpa: he used to go for a walk with the grandchildren, play football with them, he used to smoke a pipe, he played chess very well, …

In ( 5.45), all the readings have to be members of group II, but it is possible to vary the readings within this group: the potential and the habitual reading may arise because they both are characterised by the aspectual relation ‘TT = e’. The same holds for ( 5.46): 5.46 V detstve, ja igrala v tennis.

in childhood:LOC, I play:PST:ipf tennis. ‘When I was a child, I used to play / could play / usually played / … tennis.‘

The step from sense-generality to interpretation of the ipf aspect is hard to capture systematically. The sense-general specific relations provide the frame for all the possible interpretations that may be constructed in the respective context. Therefore, it is irrelevant which and how many readings are proposed. Naming and listing readings is a mere auxiliary tool; they are not to be regarded as different meanings of aspect. One generalisation with respect to the derivation of aspect readings is that frequently encountered concepts are more likely to get activated than less frequently encountered ones, which gives rise to preferred interpretations with certain aspect-Aktionsart combinations. By the principle of relevance (Sperber & Wilson 1995), all other lines of interpretation are dismissed, if these concepts yield an interpretation which is relevant in a given context. One possible way to capture this vague intuition might be the establishment of probability rankings based on native speakers’ judgements, and statistical approaches based on co-occurrences of certain lexical items. However, even such preferred readings can be overridden by the following context, and they depend on background assumptions and encyclopaedic knowledge that may very across speakers/hearers.

To sum up, the semantics of the ipf aspect can be captured by a general overlap-relation between the topic time interval I(TT) and the run time of the event I(e). This general relation is pragmatically ambiguous and may give rise to three specific relations. The ambiguity reveals itself only with reference to pragmatic reasoning, that is, pragmatics is necessary in order to determine the semantic basis of the ipf aspect in a specific context. This illustrates that pragmatics plays a role already at the subpropositional level of word and sentence meaning. The three disambiguated specific ipf relations in turn are sense-general. They provide the basis from which an interpretation is constructed in a given context. Each specific relation constrains the range of possible interpretations. Crucially, the possible interpretations for one specific relation are neither fixed nor enumerable.

For the interpretation of the ipf aspect, one of the three specific relations is selected. Based on, and restricted by this specific relation, the contextually relevant interpretation is constructed. This is summarised in Figure 5.3. The box around the three relations indicates that they have one common basis and do relate to one lexical entry: ‘ipf = TT ○ e’.

Chapter 5: Pragmatics

225

Figure 5.3

The reliability of the ambiguity tests applied above might not be uncontroversial, and the same might be true for the overall distinction of ambiguity and sense-generality, since speakers/hearers usually are not aware of this distinction in the actual interpretation process. However, having in mind cross-linguistic analyses as well as computer-guided interpretation, it is useful to distinguish the selection of meaning from the construction of interpretation. Only the former pertains to the semantics of aspect. This is important for cross-linguistic analyses, since the semantics of the respective markers can be stated and compared in a systematic way. This is also important and for computer-guided interpretation, since disambiguation is a task that can be solved by computers. The specific relations constrain the range of possible interpretations to a considerable degree and thus facilitate the task of automatic processing.

The figuring out and the combining of the mechanisms and features that play a role in disambiguating the Russian ipf aspect are guided by pragmatic reasoning. Compare example ( 5.16) above: it is encyclopaedic knowledge, knowledge about stereotypes (translators can languages other than their native language) that serves to specify the respective relation. This corresponds to Levinson’s claim that world knowledge is involved in disambiguation (2000a: 174-177). He claims these processes to be mediated by Gricean principles, more specifically, by I-inference: “Thus the kind of disambiguations often attributed to brute encyclopedic knowledge may be mediated by Gricean procedures: the speaker must presume that the addressee will use the same salient stereotype (regardless of its factual probabilities) to resolve the ambiguity.” (Levinson 2000a: 176)

Since GCI are cancellable, the recipient assumes that the speaker designs his utterance in a way that, other things being equal, the predictable GCI would go through. Thus, given the option of preserving the implicatures, an interpreter generally chooses that interpretation that will do so.

This double status of ambiguity and disambiguation in terms of both semantic encoding/decoding and pragmatic inference at the same time fits Levinson’s conception of the division of labour between semantics and pragmatics (chapter 3, section 3.2.2): reference resolution (i.e. disambiguation) is part of ‘what is said’ but may be achieved also by means of inferring and not only by means of decoding. In order to disambiguate the general ipf relation, it may not only be necessary to take into account the logical entry of the Aktionsart basis, but also the encyclopaedic entry associated with it. Since disambiguation proceeds by inferring parts of the activated encyclopaedic knowledge, it is possible to characterise it as a pragmatic

ambiguity

sense-generality

specificity

SELECTION

CONSTRUCTION

ipf: TT ○ e

TT ⊂ ϕdyn TT = e TT ⊃ e

interpret. interpret. interpret.

Chapter 5: Pragmatics

226

process. Since disambiguation is necessary for yielding the exact semantic contribution of aspect, that is, its specific word meaning, it contributes to the coded content of an utterance.

This is a crucial difference considering the goals of truth-conditional semantics and pragmatic theories, as Carston (2002: 21) points out. The former distinguishes the different meanings that are possible for one ambiguous sentence before their treatment by the theory. The proceeding of a pragmatic theory is different: “What the pragmatic theory must confront is the very different issue of how the hearer recognizes the one (or, on the occasion of a pun, two) of these n possibilities the speaker intends on a particular occasion of use.” (Carston 2002: 21)

Table 5.2 illustrates the interaction of semantics and pragmatics by summarising the interpretation of Russian ipf verbs as specified in this section (cf. Figure 5.1 and Figure 5.2):

semantic representation disambiguation proposition additional

proposition aspect

Aktionsart specific relation contextually relevant interpretation further implicatures

the coded the said the meant SEMANTICS1 PRAGMATICS1 SEMANTICS2 PRAGMATICS2

Table 5.2 The coded information pertains to the logcial entries of aspect and Aktionsart and their interaction in terms of the general overlap relation that allows for three specifications. This coded information is semantically represented and constitutes semantics1. The disambiguation of the general overlap relation is a pragmatic task. The contextually relevant interpretation is constructed from the semantic information provided by the specific relations, as well as from the encyclopaedic knowledge activated by the Aktionsart basis. This construction is again driven by pragmatic reasoning. Both pragmatic processes constitute pragmatics1 and contribute to ‘what is said’, to the proposition expressed. This proposition constitutes semantics2, the basis for further implicatures that yield additional propositions derived by pragmatics2.

5.2 Pragmatic embedding

This section outlines the pragmatic embedding of the semantic analysis of aspect. Crucially, ‘embedding’ does not imply ‘replacing’ semantics25. An elaborated semantic analysis is necessary in order to account for interpretation, since it is semantics that guarantees compositionality and thus systematicity and productivity of language. The semantic part can be elaborated more precisely if pragmatics is accounted for in a systematic way. Pragmatics may indeed contribute to shaping the semantic representation – if it is acknowledged as an equal partner in the interpretation process and as thoroughly analysed as semantics.

The pragmatic framework that will be applied to aspect interpretation is Levinson’s GCI theory (cf. chapter 3). The starting point is his suggestion to apply an implicature analysis to closed class morphemes:

25 The supplementing of semantics with pragmatic principles is also the basic idea of lexical pragmatics

advocated, e.g., by Blutner (1998a, b): “The main doctrine of Lexical Pragmatics aims at the combination of a compositional semantics with a general mechanism of conversational implicature“ (Blutner 1998b: 140). Generalised conversational implicatures explain the non-compositional character of lexical pragmatics, and they are the appropriate means to account for the non-compositional character of the interpretation of aspectualised verbs in specific utterances, and, importantly, do this in a predictable and systematic way.

Chapter 5: Pragmatics

227

“An implicature analysis of closed class morphemes may yield important insights and resolve many puzzles that arise from the apparently elusive and variable content of these expressions.” (Levinson 2000a: 98)

As will be shown in this section, the puzzles concerning the semantic characterisation of the Russian ipf aspect (cf. chapter 2) can be resolved if semantics is restricted to the encoded information that needs to be enriched and specified by means of systematic and predictable pragmatic principles. Pragmatics plays a role at basically all the different stages of the interpretation process (cf. section 5.1.2). The fact that the interpretation of ipf verbs always needs a certain amount of pragmatic reasoning is due to the semantically underspecified contribution of the ipf aspect. The impossibility to state all the potential contextually relevant aspect readings is another argument against a purely semantic analysis, as well as the fact that interpretations of aspectualised verbs seem to be defaults that can be overridden by the following context. Since semantics is assumed to provide explicitly encoded information, these default interpretations cannot be explained by semantics alone.

To emphasise it again, both semantics and pragmatics are needed and both will be shown to interact in systematic and predictable ways.

5.2.1 GCI and aspect Grammatical categories are expressed by paradigmatically opposed expressions, closed class function words or morphemes, and thus could be assumed to constitute the core of application of GCI theory (cf. chapter 3, section 3.2.2). Quite to the contrary, however, pragmatics still seems to play a rather subordinate role in this domain.

Levinson (2000a: 91) mentions the following predictions the GCI-approach allows for such a set of closed class morphemes:

5.47 a. there are to be expected sets of contrasting function words, because the

pragmatic load can then be increased through pragmatic oppositions without further increasing the set of oppositions

b. there are to be expected scalar oppositions due to informational asymmetries

c. there are likely to be contrasts pertaining to formal asymmetries Prediction ( 5.47a) is the most general one and concerns the basic assumption that a well developed pragmatic theory lightens the burden of semantics. Prediction ( 5.47b) is basic to the derivation of scalar implicatures, a sub-class of Q-inferences. These inferences rely on contrasts in informativity of paradigmatically opposed forms, here pf and ipf aspect. Prediction ( 5.47c) pertains to contrasts in formal marking giving rise to M-inferences in case the default aspect assignment as specified in chapter 2, section 2.2.3 is not valid.

Generalised conversational implicatures play a systematic role in generality narrowing, where ‘generality’ covers both the Radical Pragmatics view that the lexical content of an expression is the weakest exhibited meaning of it, and Atlas’ (1989) conception of sense-generality (cf. chapter 3, section 3.1.2). In fact, sense-generality and the Gricean maxims both complement each other, as Atlas (1989: 149) emphasises: “Grice’s pragmatics without sense-generality is blind, and sense-generality without Grice’s pragmatics is empty.”

Levinson exemplifies the role of GCI with sense-general expressions by means of possessives such as ‘Larry’s book is on negation’ (2000a: 186), where the possessive merely encodes that there exists some relation between Larry and book. Oblique cases in case-marking languages constitute another example of expressions exhibiting this kind of sense-generality. Levinson mentions the Latin ablative, which may express ‘source’, ‘cause’, ‘agency’, ‘instrument’, ‘deprivation’, or ‘direction from’. Such a general relation is specified by means of inferences

Chapter 5: Pragmatics

228

guided by Gricean principles on the one hand, and by contextual information on the other. An analysis along these lines will be applied to the case of grammatical aspect as well. The heuristics and inferences Levinson proposes operate on the stable, encoded part of grammatical aspect and the stable, encoded part of its interaction with the Aktionsart basis (cf. Figure 5.1).

What is interpreted is an aspectualised verb, that is, a combination of procedural and conceptual information. The two kinds of information differ in that the latter has, besides a logical and lexical entry, an encyclopaedic one as well. Since logical and lexical entries are stable across contexts and speakers, they constitute those components of encoded information that deliver the invariant, reliable part of communication. Both speaker and hearer know about this reliability. It is this mutual awareness that provides the basis on which Levinson’s default heuristics work to yield predictable, stable, context-invariant default inferences. Other principles of narrowing and enrichment interact with the encyclopaedic entries. It is therefore best to combine Levinson’s heuristics with accounts of ad hoc concept formation (e.g. Carston 2002), acknowledging that they operate within different domains of utterance interpretation. Criticism of GCI theory as not being generalisable to all kinds of lexical adjustment and to cases of narrowing (e.g., Carston 2004c) thus do not invalidate the central assumptions of this theory. This is illustrated in Figure 5.4:

Figure 5.4

The light grey zones indicate the utterance tokens of both speaker and hearer that overlap in the dark grey zone. This dark grey zone constitutes the stable part of conveyed information provided by the logical entries, giving rise to predictable, that is, context-independent

Aspectualised verb used in a specific utterance

Speaker UTTERANCE TOKEN encyclopaedic

variable

Hearer UTTERANCE TOKEN

encyclopaedic

variable

Speaker & HearerUTTERANCE TYPE

(lexical) / logical

stable

Aktionsart: ϕdyn τ ϕstat

aspect: relation

context-independent heuristics and

default inferences

GCI

context-dependent enrichment and

specification

concept formation

context-dependent enrichment and

specification

concept formation

Chapter 5: Pragmatics

229

inferences in the sense of Levinson’s GCI. The light grey zones constitute the variable part of the information conveyed. This kind of information stems basically from the encyclopaedic entries involved. Since the exact content of encyclopaedic entries may vary across speakers, the respective utterance tokens may vary as well. The inferences that are drawn from this information are thus context-dependent processes of concept formation like enrichment and specification that are mainly dealt with in the framework of Relevance Theory. This points out that the two ‘competing’ pragmatic theories – Relevance Theory and neo-Gricean pragmatics/Radical Pragmatics – should be regarded as complementing each other, since they apply to different phenomena.

Figure 5.4 also illustrates the basic assumptions about how communication proceeds (cf. chapter 3, section 3.1.1). The decisive fact is that both speaker and hearer are mutually aware of the encoded information and of the inferences it may give rise to. That is, in uttering a sentence, the speaker can rely on the hearer to draw certain inferences and the hearer can rely on the speaker to be aware of the fact that he will draw these inferences. The speaker has a certain utterance token in mind which he intends to convey. He needs to encode it by linguistic means, but this encoded information delivers only part of the intended utterance token. What the hearer gets as the basis for his interpretation is thus an utterance type that has to be enriched in order to arrive at an utterance token. The utterance type constitutes the reliable part of communication as it is overt to both speaker and hearer. For communication to work, there has to be a set of general principles of reasoning that are available to speaker and hearer. These principles are reliably made use of such that the speaker can rely on the hearer making use of them in interpreting the encoded information. The reasoning processes applying at this level of utterance-type meaning are guided by the heuristics proposed by Levinson. The mutual knowledge about the fact that speaker and hearer are both oriented to these heuristics provides extra, highly salient constraints to the solution of the problem of how the speaker should frame his utterance so that the proposition the recipient attributes to it has the properties that the speaker intended it to have (Levinson 2000a: 258).

In order to bridge the gap between semantic representations and propositions, a reasoning system must be involved that expands the information given. This is achieved by pragmatic processes, that is, defeasible inferences. GCI-theory offers one solution how such inferences can be constrained to the best interpretation, namely, by using heuristics to warrant presumptive meanings (Levinson 2000a: 257f). As has been pointed out in chapter 3, section 3.2.2, heuristics describe observed regularities in behaviour and thereby give rise to inferences. Fast and frugal heuristics (Gigerenzer et al. 1999) account for how real minds make decisions under constraints of limited time and knowledge. They are accurate without being too complex and without being too closely tied to a certain environment, that is, context. By paying attention to systematic informative cues and overlooking more variable and uninformative ones, such heuristics make a trade-off between generality and specificity. These features also characterise GCI, which are therefore the appropriate means to specify the regularities involved in the interpretation process. In the remainder of this section, Levinson’s GCI theory will be applied to the interpretation of Russian ipf aspect.

The first kind of GCI relies on the Q-principle which is related to Grice’s first quantity maxim. The corresponding heuristics states ‘what isn’t said, isn’t’ and gives rise to basically two kinds of inferences – clausal and scalar (Gazdar 1979). Clausal inferences are not dealt with here, but scalar inferences will be shown to play an important role not only to account for the interpretation of aspect in Russian, but also to account for some cross-linguistic prima facie puzzles concerning aspect usage and interpretation (cf. section 5.2.3). The basic background assumption with respect to scalar implicatures is a contrast set of linguistic alternates that differ in informativity and are based on a common semantic field. The concept of Q-inference is basic to Radical Pragmatics, which is occasionally exposed to criticism for

Chapter 5: Pragmatics

230

not properly defining central notions such as ‘scale’, ‘expression alternate’, ‘ordering’ and ‘context’: „In short, the supposition present in the Radical Pragmatic theory of Q1 implicatures [scalar implicatures, B.S.], which surfaces in the formulation of the Gricean argument, namely that in order to derive such an implicature the available expression alternates need to be identified, is a highly problematic one. The reason is that there is no definition of these expression alternates.“ (Scharten 1997: 44)

The problems listed by Scharten (1997) do not arise if the notion of scalar implicature is applied to closed class morphemes constituting paradigmatic oppositions. In these cases, the expression alternates of a certain semantic field are explicit and fixed, just as their ranking within the scale. Expression alternates and their ranking in terms of informativity belong to the knowledge speaker and hearer have about their language. From the restriction of analysis to grammatical morphemes it follows that we are dealing with lexical or logical scales that arise independently of contextual assumptions and are stable over contexts.

Scalar Q-inferences are characterised by two essential properties: they are meta-linguistic in that they do not rely on context but on a set of paradigmatically opposed alternates, and they are negative, since what is implicated is a presumption that the possible alternative is not the case. Scalar inferences depend on a contrast in salience and on what might have been said, but the speaker chose not to say. Crucially, the hearer has to be aware of the possible alternatives26. The heuristic that is basic to this inference relies upon the speaker’s and the hearer’s perception of the informativity-ranking (cf. ( 5.47b)) of the respective expressions with respect to the possible alternates (Hirschberg 1991: 83). In the straightforward and most commonly cited cases such as <all; some>, the scales induced by the perceived ranking can be defined in terms of entailment. However, such an entailment definition is not possible for all scales. Scales can be construed in an ad hoc way, relying on contextually established bases for the comparison of the alternatives. Elements of such scales may thus be related without having an intrinsic relationship to one another and are therefore hard to predict outside the context in which they are evoked. Hirschberg (1991) offers a detailed analysis of this kind of scalar implicatures and the difficulty of finding an adequate definition for them. What all instances of scales share, are the characteristics of partial ordering relations, so that a poset condition appears to capture all the different instances (Hirschberg 1991: 132). In fact, Hirschberg’s account relates generalised conversational implicatures with particularised ones: the former are based on the lexicon, the latter on salient contrasts in the world, where inference depends on nonce scales with a partial ordering relation defined over them (Levinson 2000a: 95ff). Thus, we are actually dealing with two quite different kinds of scales: lexical or logical scales on the one hand, and pragmatically established scales on the other.

In the domain of grammatical aspect, which is restricted to the analysis of a paradigmatic contrast between members of one grammatical category, the respective scales can be predicted outside the context. They therefore remain stable within a certain semantic field. The notion of ‘semantic field’ is in fact crucial in this respect, since this notion constitutes the basis for the alternates to be comparable (cf. the analysis of the Turkish aspect system in section 5.2.3). With pragmatically established scales, this basis is induced in an ad hoc way, but in case of lexical/logical scales established by grammatical morphemes, the scale is defined in advance, namely by the knowledge of the respective semantic field. This is part of the interlocutors’ knowledge of the language, and thus remains stable across contexts. The scale is induced by the lexical items themselves and not by context; the inferences derived from that scale are thus meta-linguistic.

26 Compare also Parikh (2000: 191f), who equally emphasises the importance of knowledge about the

possible alternates: “The content communicated depends not only on what was uttered but also, crucially, on what the speaker might have uttered but chose not to, and on [the] shared information about these choices.”

Chapter 5: Pragmatics

231

What makes the notion of generalised scalar implicature so useful is its independence of specific contexts which allows for default inferences to arise. This property of scalar implicatures has been questioned, among others, by Scharten (1997): “We may conclude that the Q1 implicatures [scalar implicatures, B.S.] are very strongly dependent on the knowledge that a speaker possesses and on what the hearer assumes the speaker to know.” (Scharten 1997: 10)

Scharten bases this criticism on examples like the following:

5.48 X is meeting a woman today.

The stronger, more informative expression would have been to use the name of that woman. And the use of the more informative expression is obligatory in order to obey the first quantity maxim ‘say as much as you can’. If the speaker does not do this, as is the case in ( 5.48), the hearer presumes this to be significant, and this presumption leads him to infer an implicit message. If X is a woman, however, no implicature of something indecent going on arises, but simply the inference that the speaker did not know the identity of that woman: 5.49 a. Whom is Ms. Johnson meeting this evening?

b. She is meeting a woman this evening. Scharten concludes that the hearer needs to have at least some knowledge of the facts in order to draw the respective inferences. This, however, contradicts the basic characteristics of generalised conversational implicatures as going through without further contextual knowledge. Accordingly, Scharten argues that it is not the content that makes a conversational implicature a generalised one, but the conditions under which it arises.

However, this objection is not valid for the analysis of grammatical morphemes that constitute oppositions based on informativity. This knowledge is part of the knowledge speaker and hearer have about their language and involves knowledge of the expression alternate that has not been used. In this case, the hearer does not need to have additional background knowledge besides the knowledge of his language in order to draw the respective inferences from the use of a certain expression. The informativity-relation remains the same across contexts, it is systematic.

With paradigmatically opposed morphemes the informativity relation can be captured in terms of a meta-scale based on the function the morphemes serve. Levinson (2000a: 97f) illustrates this with the example of the Finnish partitive and accusative case (Heinämäkki 1994: 213f) which specify the feature of boundedness in different ways. This can be reflected by the meta-scale <BOUNDED, ±BOUNDED>, instantiated by the accusative and partitive case. The accusative is specific in that the respective event is bounded ( 5.50a), the partitive is unspecific with respect to the boundedness of the event ( 5.50b), it is noncommittal about whether the event is bounded or unbounded. 5.50 a. Terttu luki kirjan. Terttu read book:ACC

‘Terttu read (all) the book.’

b. Terttu luki kirjaa. Terttu read book:PART ‘Terttu was reading a (part of the) book.’ The notion of ‘meta-scale’ indicates that the elements of the scale are not linguistic entities but semantic notions. Another possibility to capture this relationship is to state it in terms of a

Chapter 5: Pragmatics

232

specific scale based on the semantic field ‘boundedness’ instantiated by <ACC; PART>. The elements of this scale are linguistic entities. Use of the partitive case allows for the inference that the stronger claim, i.e., boundedness of the event, is not possible. The partitive case does not deny boundedness, it merely suggests that a bounded event is not being referred to.

With respect to grammatical aspect, there are basically three semantic notions a meta-scale could be based on: intraterminality, adterminality and postterminality. These are the cross-linguistic generalisations capturing the possible types of aspect categories (Johanson 1971; 1994; 2000; cf. chapter 4). The aspect markers that instantiate the respective meta-categories stand in specific markedness relations, which in turn are based on certain semantic fields. These semantic fields have to be fixed before the corresponding scale can be established. That is, a meta-scale of the form <INTRA; ±INTRA> or <AD; ±AD> is appropriate to capture cross-linguistic features of aspectual systems and compare languages as to whether, and how, they instantiate this relation, but it is too coarse to capture language-internal relations.

( 5.51)-( 5.53) show the meta-scales and the corresponding specific scales for Turkish, English and Russian:

5.51 Turkish

a. meta-scale: <INTRA; ±INTRA>

b. intraterminality-field: <-iyordu/-mekteydi/-irdi; -di> The specific intraterminality-scale in Turkish may in fact give rise to further scales, based on different semantic fields (cf. section 5.2.3). 5.52 English

a. meta-scale: <INTRA; ±INTRA>

b. progressivity-field: <progressive; simple form>

English instantiates the meta-scale <INTRA; ±INTRA> by a specific scale based on the semantic field ‘progressivity’. In a cross-linguistic perspective, this semantic field is one of the possible instantiations of the meta-scale <INTRA; ±INTRA>. 5.53 Russian

a. meta-scale: <AD; ±AD>

b. perfectivity-field: <pf; ipf> The aspect system of Russian is based on the meta-scale <AD; ±AD>. It is instantiated by the specific scale <pf; ipf> which is based on the perfectivity field. Note that there is another specific scale, which is based on the imperfectivity (non-adterminality) field: <secondary ipf; simplex ipf>27. This scale will not be dealt with here.

27 The existence of such a scale is more evident in Bulgarian, where secondary imperfectivisation is more

productive than in Russian. In case of secondary ipf verbs that are derived from a pf partner of an ipf simplex counterpart, e.g., piša (ipf) – napiša (pf) – napisvam (sec.ipf) ‘to write’, the derived secondary ipf verb (napisvam) is restricted in use. Such ‘true’ secondary imperfectives can be used to convey the actual-processual reading only in very restricted contexts. They are mainly used to convey iterativity and habituality (Lindstedt 1985: 196). Due to these restrictions on use they are more specific than the corresponding simplex ipf verb, and this allows to establish the scale <secondary ipf; simplex ipf>.

Chapter 5: Pragmatics

233

In order to capture the language-internal inferences induced by the Q-heuristics, the specific scales and the semantic fields they are based on are crucial. One and the same lexical item can be element of different scales. A scale can be established only in case the two (or more) items in question could equally be used in the same context. To give an example, the perfectivity scale <pf; ipf> in Russian can only be established in past tense contexts, since in the present tense the pf aspect does not constitute a possible alternate to the ipf. Furthermore, semantic fields and the corresponding scales can only be established if the semantics of the items has been specified in order to determine their common basis (cf. the remarks on opposition and markedness in chapter 2).

How do these scales give rise to default interpretations? By the Q-principle the hearer is entitled to assume that the speaker made the strongest statement consistent with what he knows. The Q-heuristic allows the hearer to infer that if the speaker uses the weaker element of the scale, he is not in a position to assert the stronger one – either, because it is not valid or relevant, or because the speaker does not know whether it holds or simply does not want to make the stronger commitment. This is illustrated in ( 5.54), where the ipf aspect is used, but the pf would constitute a possible alternate:

5.54 On pisal pis’mo. he write:PST:ipf letter ‘He wrote / was writing / … a letter.’ By the perfectivity-scale <pf; ipf> the hearer is entitled to draw the inference that if the speaker uses the weaker, less informative element – the ipf aspect –, the stronger one – the pf aspect – is not possible or not relevant, and that consequently no boundary is selected and asserted. Being the weaker element does not amount to explicitly denying the value of the stronger one. The use of the ipf aspect does not exclude boundary-selection. And in fact, with the general-factual reading, the boundary is included in the topic time (remember, the respective relation is ‘TT ⊃ e’). If the statement is about the general fact that an action has taken place, then the inherent boundary, if present, has to be asserted as well. Otherwise, the action could not be said to have taken place. This indicates that the scale <pf; ipf> cannot be an entailment scale, since in case of the general-factual reading, the entailment relation would the other way round. With the relation ‘TT = e’, there could not be an entailment in either direction. Thus, the scale <pf; ipf> is no entailment scale – it is not the case that the pf aspect entails the ipf. The ordering criterion for the elements of the perfectivity scale is informativity. On the basis of this scale, the general-factual reading of the ipf aspect can be accounted for in a straightforward way – it conveys the mere denotation of a fact, boundary selection is neither explicitly asserted nor explicitly excluded. Note that the perfectivity-scale <pf; ipf> in Russian is independent of aspectual pairs. The semantic specificity (markedness) of the pf aspect and the non-specificity (unmarkedness) of the ipf aspect is systematic (Stoll 2001) and thus independent of its application to verbs and independent of the existence of aspectual pairs.

To recap, Q-inferences are involved in deriving the meaning of semantically unmarked forms by giving rise to scalar implicatures. Scales are to be understood as ordered sets of semantic concepts being lexicalised or grammaticalised to the same degree, standing in a salient opposition relation (Bickel 1996: 14). The elements of such a scale differ in that the stronger one is more informative, more specific. It has one feature specification more and thus makes a stronger denotational claim. The scale for the perfectivity-field in Russian is <pf; ipf>, stating that if the speaker uses the weaker element of the scale – the ipf aspect – the hearer is entitled to infer the non-validity of the stronger element – the pf aspect. Whereas the use of the pf aspect explicitly states the reaching of a boundary within TT, the use of the ipf

Chapter 5: Pragmatics

234

aspect does not exclude the reaching of the boundary. This is captured by its semantics in terms of the general overlap relation for the ipf aspect and the three specific relations it may be strengthened to.

The next principle suggested by Levinson, the M-Principle (derived from Grice’s manner-maxim), is basic to the heuristic ‘what is said in an abnormal way isn’t normal’ which in turn yields inferences from marked expressions to marked situations. M-inferences are meta-linguistic in that they rely on a contrast in form of the linguistic expressions, which at the same time have essentially the same semantic characterisation. Levinson (2000a: 136) takes the relation between formal and semantic markedness to be the relation in question with M-inferences. However, with respect to Russian aspect this relation has to be conceived of as one between formal and pragmatic markedness28. The semantic unmarkedness of the ipf aspect with respect to the pf aspect remains untouched by formal and pragmatic markedness29.

With respect to aspect interpretation, M-implicatures occur in case of mismatches between aspect operator and Aktionsart basis, that is, in case the basis does not provide the necessary input for aspect to apply. The missing element (phase or boundary) may be induced by pragmatic reasoning. This supplementation of the Aktionsart basis is restricted by the conceptual knowledge about the prototypical event, i.e. by the tripartite structure of the event nucleus. Moreover, the M-principle applies to cases of morphologically marked non-default aspect assignment (cf. chapter 2, section 2.2.3). Insofar as in both cases the interpretation process requires additional assumptions and thus more effort, such aspect-Aktionsart combinations can be said to be pragmatically marked30. The process of inferring specific elements into the Aktionsart basis is usually described in terms of ‘recategorisation’ or ‘coercion’ (e.g. Moens 1987; Moens & Steedman 1988; Pulman 1997; de Swart 1998). In section 5.2.2 it will be argued that the notion of coercion has to be used carefully in dealing with aspect and is better replaced by M-inference and accommodation.

The M-heuristic and its corresponding inference is rather controversial in pragmatic literature, and is quite often subsumed under the Q-principle (e.g., Traugott 2004). However, since the principles are considered as heuristics, this conflation has to be rejected: the three different heuristics are invoked by different kinds of linguistic expressions. The M-heuristic is invoked by the form of the expression, rather than by its semantics as is the case with the Q-heuristic (Levinson 2000a: 136). Another argument against the conflation of Q- and M-heuristics is that they work on different levels of markedness: semantic markedness in the case of the Q-heuristic, pragmatic markedness in the case of the M-heuristic.

Examples ( 5.55) and ( 5.56) illustrate how M-inferences work with the interpretation of Russian aspect:

5.55 On rešil zadaču.

he solve:PST:pf exercise:Acc ‘He solved the exercise.’

In ( 5.55), the <ϕdyn, τ>-Aktionsart provides the appropriate input for the pf aspect to apply. The pf aspect selects the boundary and relates it to the topic time such that the boundary is

28 Due to the consistent aspect marking in English and Turkish, M-inferences there indeed rely on the relation

between formal and semantic marking. Pragmatic marking in English and Turkish thus pertains only to cases of aspect-Aktionsart mismatches (cf. section 5.2).

29 This is different with secondary imperfectives that are marked as compared to simple imperfectives. In this case, the markedness pertains to formal as well as to semantic and pragmatic marking. This does not affect the systematic semantic unmarkedness of the ipf aspect with respect to the pf aspect, and illustrates the need to make clear the basis on which the respective scales are established. Cf. also footnote 27.

30 Compare Jackendoff’s (1991: 17) ‘rules of construal’: “[It is] characteristic of this class of rules: the interpreter avails him/herself of them to understand otherwise ill-formed or pragmatically inappropriate utterances.”

Chapter 5: Pragmatics

235

included in TT. The straightforward interpretation is that the action has reached its inherent boundary – the exercise is solved. ( 5.55) is an instance of default aspect assignment. There is no overt aspect marking, but by default the terminative Aktionsart is assigned the pf aspect. This is different with ( 5.56), where the deviation from this default aspect assignment is morphologically marked by the suffix -a- (rešat’): 5.56 On rešal zadaču.

he solve:PST:ipf exercise:Acc ‘He was solving/tried to solve/could solve (...) the exercise.’

The use of the ipf aspect gives rise to the inference that boundary selection was not possible or not relevant – it is not excluded, however. The morphologically marked deviation from the default gives rise to the inference of a marked, non-stereotypical situation. The deviation has to be ‘explained’ by an appropriate interpretation. A plausible interpretation would be, for instance, the conative interpretation he was trying to solve the exercise. The range of possible interpretations is constrained by both Q- and M-inference, that is, ‘no explicit boundary selection’ and ‘marked situation’. For the respective interpretation to arise, encyclopaedic knowledge associated with, and activated by, the respective predicate has to be taken into account. The enrichment process is guided by the next heuristic, the I-heuristic and the corresponding inference.

The I-heuristic (‘what is expressed simply is stereotypically exemplified’) is based on the ‘Principle of Informativity’ (Levinson 1987; 2000a: 112-134; Atlas & Levinson 1981) which derives from Grice’s second quantity maxim ‘Do not make your contribution more informative than is required’. This heuristics gives rise to a variety of inferences that can be subsumed under one general instruction ‘Enrich!’. Levinson (2000a: 114) thus takes the I-heuristic to be an enrichment rule, which says that the informational content of the speaker’s utterance should be amplified to the most specific interpretation that can reasonably be taken to be the one intended by the speaker. The only meta-linguistic instruction specified by the I-heuristic is the one to enrich to a stereotype: “Inferences to the stereotype are thus not ‘generalised’ in the sense that they are independent of shared beliefs (as Q and M largely are), but they are ‘generalised’ in the sense that they follow a general principle – restrict the interpretation to what by consensus constitutes the stereotypical, central extensions.” (Levinson 2000a: 103)

The relation between the Q- and I-heuristics is described as follows: “I-implicatures are inferences from the lack of further specification to the lack of need for it, whereas Q-implicatures are inferences from the lack of informational richness to the speaker’s inability to provide it.” (Levinson 2000: 116; emphasis in original)

This principle covers a range of different phenomena, such as conditional perfection (from ‘if’ to ‘iff’), conjunction buttressing (temporal sequence, causal connectedness), bridging, inference to stereotype, negative strengthening, and many more. Facing this rather heterogeneous range of manifestations of I-inferences, the questions arises in what sense these inferences can be said to be generalised. They are generalised on an abstract level, their common denominator being the maximisation of the informational load by narrowing the interpretation to a specific sub-case of what is said. These inferences specify the sense-generality of lexical items which is typical for a big part of the vocabulary. In this way, I-inferences help to constrain the lexicon to a learnable size. This guarantees efficiency in language use, but crucially requires the complementary property of enrichment. Enrichment is made possible by means of the I-heuristic. Sense-generality is also common with grammatical meaning, or, to stick to the terminology adopted here, it is typical of lexical items specifying procedural information. Levinson’s (1995) examples of I-inferences encompass the interpretation of nominal compounds and the possessive construction in English. The relation

Chapter 5: Pragmatics

236

between the parts of the compounds and the possessor-relationship are specified only in the most general way, i.e. merely as some relation holding between the two parts, which is then specified by pragmatic inference. The same holds for aspect interpretation as well.

With respect to the interpretation of Russian aspect, such I-inferences play a double role. In a meta-linguistic sense, that is, with respect to the instruction to enrich, I-inferences are involved in the default aspect interpretation illustrated in chapter 2, section 2.3.3. The general instruction to enrich a minimal expression to a stereotype corresponds to the interpretation of an not overtly aspect-marked form in terms of event realisation (cf. Bohnemeyer & Swift 2004: 287): “It certainly seems reasonable to consider aspectual reference under event realization more stereotypical than aspectual reference under lack of realization, and thus leave the latter to overtly marked forms.”

Since this meta-linguistic kind of I-inferences pertains to the logical entries of aspect and Aktionsart and their interaction, it will be referred to as ‘Ilog’. In this sense, Ilog is complementary to M: Ilog instructs to enrich to a stereotype, whereas M indicates that the stereotype is not the case.

I-inferences are also involved in deriving the final, contextually relevant interpretation, i.e. they contribute to the process of concept formation (cf. Figure 5.4). In this case they interact with the encyclopaedic entries activated by the conceptual information provided by the Aktionsart, and can no longer be regarded as context-independent default inferences. This kind of non meta-linguistic I-inference will be referred to as ‘I’. The enrichment and specification I-inferences start off requires reference to the encyclopaedic knowledge activated by the encoded conceptual information. Since encyclopaedic knowledge is not stable and varies across speakers/hearers, interpretations may differ within the frame established by the logical entries of aspect and Aktionsart and their combination on the one hand, and by the default inferences based on them on the other. The enrichment by means of I-inference is constrained by truth-evaluability as the lower bound and relevance as the upper bound (Carston 1998a). More frequently encountered concepts are more likely to get activated than less frequent ones, and can therefore be considered as the stereotypes in question. They require less interpretational effort and are thus more likely to constitute the contextually relevant interpretation. The notion of ‘frequently encountered concept’ is not only context-dependent, but it may also differ across speakers/hearers. Therefore the question arises, whether the notion of ‘GCI’ is appropriate for I-inference at all. I-inferences rather seem to constitute particularised conversational implicatures the specific realisation of which depends on the specific context.

As has been pointed out above, the distinction of the three different heuristics and the inferences based on them is treated controversially in current approaches to pragmatics. One argument brought forward against Levinson’s heuristics is that it is not necessary to have three distinct principles. Instead, one single principle of relevance (Wilson & Sperber 1995; Carston 2002) is assumed to suffice. Moreover, the M-principle is quite commonly regarded as superfluous, and it is argued that the other two principles, i.e. Q and I, are enough to account for the facts (e.g. Traugott 2004). Levinson resists the conflation of the three principles as he regards them as heuristics that are each invoked by different kinds of linguistic expressions: “[J]ust as the use of an item from a contrast set suggests that the contrastive items would be inappropriate, or the use of a minimal expression invokes a maximal interpretation, so the use of a marked expression signals an opposing interpretation to the one that would have been induced by the use of the unmarked expression.” (Levinson 2000a: 136)

The same point can be made for the contribution of the three heuristics to the interpretation of aspectualised verbs. Q and M interact with the logical entry, and are thus meta-linguistic and context-independent. I interacts with the encyclopaedic entry and guides the interpretation to the most easily accessible interpretation. The variability in the final interpretation of an

Chapter 5: Pragmatics

237

aspectualised verb is attributable to exactly this restriction of the interpretation to what, by consensus, is the stereotypical, central extension. Stereotypes may vary across contexts and speakers/hearers. Therefore, an interpretation to a stereotype is context-dependent. The enrichment process takes place within the constraints inferred from the Q- and M-heuristics. Q and M constrain the range of possible interpretations by further specifying the procedural information delivered by aspect and the combination of aspect and Aktionsart. The range of possible interpretations is constrained, but not in the sense of certain elements being eliminated from a list. It is constrained in the sense of further specifying the explicit frame within which different interpretations can be constructed. The three heuristics thus amplify the encoded information in two respects: “[They] filter further information of two kinds: information about the structure of the world (or, rather, of stereotypical properties of the relevant domain) and meta-linguistic knowledge, that is, information about semantically related expressions” (Levinson 1995: 96f)

Meta-linguistic knowledge is basic to the Q- and M-heuristics, whereas the I-heuristic crucially relies on the structure of the word specified in the respective encyclopaedic entries.

Since M-inferences seem to be parasitic on corresponding I-inferences, the I- and M-principles contribute to the division of pragmatic labour (cf. also Horn 1984): “Whatever an unmarked expression U would I-implicate, the marked alternative (denotational synonym) M will implicate the complement of U’s denotation.” (Levinson 2000a: 137)

With respect to aspect, this pertains to default aspect assignment and morphologically indicated deviations. Default aspect assignment is achieved by means of the default aspect operator that states ‘event realisation at a minimal topic time’ (chapter 2, section 2.2.3). For aterminative predicates, the ipf aspect is the default guaranteeing event realisation, for terminative predicates it is the pf aspect. This stereotypical interpretation follows from the meta-linguistic part of the I-principle (Ilog). In case of morphologically indicated deviation from this default assignment, the M-principle states that this stereotypical interpretation is not valid. The ipf aspect with terminative verbs licenses the inference that the event is not realised within TT, the pf aspect with an aterminative verb indicates more than mere event realisation within a minimal TT – either TT is explicitly marked as closed-bounded (as is the case, albeit indirectly, with delimitatives; cf. chapter 4, section 4.2.1) or the event is signalled to have occurred once within TT (e.g. suffixation with -nu-, prefixation with s-).

The difference between the Q- and M-principle consists in the form of markedness that is derived from different kinds of alternate expressions – sets of alternates with essentially the same form and contrastive semantic information in the case of Q-inferences, contrasting forms with essentially the same inherent semantic information in case of M-inferences. Q-inferences are based on semantic markedness, M-inferences on formal markedness. Both members of the Russian aspect category can be ranked according to the informativity of their semantic content, which gives rise to Q-inferences. M-inferences are based on the formal marking indicating the deviation from default assignment sketched above.

With respect to aspect in Russian, each heuristic is invoked by a different kind of linguistic input. The Q-heuristic is invoked by the choice of the aspect marker, the M-heuristic by the application of that marker onto an Aktionsart basis. The I-heuristic is invoked by the combination of aspect marker, Aktionsart basis and the encyclopaedic knowledge activated by the latter. This is summarised in Table 5.3:

Chapter 5: Pragmatics

238

aspect (logical entry)

<pf; ipf>

aspectual marker / operator (grammatical knowledge)

SEMANTIC MARKEDNESS Q

Aktionsart (logical entry) ϕdyn τ ϕstat

semantic representation / operandum (lexical knowledge)

PRAGMATIC MARKEDNESS

M / Ilog

Aktionsart (encyclopaedic entry)

encyclopaedic information (conceptual knowledge)

I

Table 5.3 In Table 5.3, different kinds of markedness are distinguished. At the level of grammatical knowledge, it is the systematic markedness relation between pf and ipf aspect that is decisive. This markedness relation has been captured in terms of informativity and allows an ordering of pf and ipf aspect on a scale, which in turn gives rise to the Q-inference. With regard to M-inferences, a different kind of markedness is involved, namely morphological markedness. Since default aspect assignment plays a crucial role in Russian, morphological markedness corresponds to pragmatic markedness: non-correspondence to the default assignment is morphologically marked and this morphological markedness points out a marked situation that has to be explained by an appropriate interpretation. In section 5.2.2, two kinds of non-correspondence to the default will be distinguished: deviation with respect to selection/focusing of the Aktionsart basis, and deviation requiring an additional operation of accommodation, which applies before the final interpretation is assigned. Within these additional procedural constraints inferred by Q and M, the I-inference guides the interpretation process, i.e., the enrichment of the information decoded and inferred from the combination of the logical entries of aspect and Aktionsart. This process of concept formation is geared towards the most stereotypical interpretation, it follows the line of interpretation that requires the least effort.

Levinson (2000a: 155-164) proposes a ranking of the principles and their inferences, namely ‘Q > M > I’: “[I]nferences based on highly constrained sets of lexemes (Q-inferences) block those based on wider ranging contrasts in markedness (M-inferences), which in turn block those based on stereotypes about the world and the inference of maximal cohesion (I-inferences.)” (Levinson 2000a: 161)

Levinson attributes the priority of Q over M to the relative importance of informational content over expression modulation. The I-principle gives rise to an enrichment instruction, and this enrichment is constrained by the other two principles. This follows from the assumptions outlined above, namely that Q and M apply to the logical entries of aspect and Aktionsart and thus modify the procedural information that specifies the way the encyclopaedic entry of the Aktionsart is to be accessed and integrated into the interpretation process. The I-principle works on that encyclopaedic entry and starts off the enrichment process within the domain constrained by Q and M.

The assumptions on markedness and ranking of the inferences with respect to the ipf aspect in Russian are summarised in Table 5.431:

31 Levinson bases the ranking on potentially arising conflicts between the inferences. Note that in case of the

Russian aspect system, the ranking is understood in the sense of scope taking. I-inferences enrich the provided information within the constraints imposed by M and Q. M indicates the markedness of the situation expressed, but this markedness cannot invalidate the constraint imposed by Q; pragmatic markedness does not have an impact on semantic markedness.

Chapter 5: Pragmatics

239

Q > M > I semantically unmarked semantically unmarked semantically unmarked pragmatically marked

→ deviation from default pragmatically marked → deviation from default

enrichment to a specific interpretation

Table 5.4

Table 5.4 shows that the semantic unmarkedness of the ipf aspect is not influenced by morphological marking, which in turn indicates pragmatic markedness. These two markedness features remain valid with the context enrichment to the specific interpretation and serve as constraints on that interpretation.

This section has outlined the pragmatic embedding of the semantic information of aspect and Aktionsart into Levinson’s GCI-theory. GCI theory is predictive as it captures expectations about inferences that ought to arise from form and content of the particular expressions used, and about the knowledge of the possible alternates that have not been used. This theory was argued to be especially suitable for the analysis of paradigmatically opposed closed class morphemes. It has been shown how Q-, M- and I-inferences interact with different parts of the semantic information contributed by an aspectualised verb and serve to amplify this information. The semantic analysis of aspect and Aktionsart developed in chapter 4 constitutes the basis for pragmatics to interact with. Levinson’s GCI have been shown to be the appropriate principles to constrain and enrich the encoded information. Semantics and pragmatics work hand in hand, mutually constraining and enriching each other.

5.2.2 Default, deviation and mismatch This section examines cases in which aspect assignment does not proceed in the default way. Non-correspondences to the default might be of two kinds: change of the default selection as in ( 5.57a), and mismatch between aspect and Aktionsart basis as in ( 5.57b): 5.57 a. On rešal zadaču. he solve:PST:ipf exercise:ACC ‘He was solving the exercise.’

b. Zacvetali rozy. PRF:bloom:PST:ipf roses ‘The roses were starting / were about to start to blossom.’ Aspect can change the default perspective on the event. This is the case in ( 5.57a) where the focus is on the phase of a <ϕdyn, τ>-predicate, whereas the default would be the focusing of the boundary. Since aspect specifies certain input requirements that have to be fulfilled by the Aktionsart basis, its application may cause a change in the eventuality description if this requirement is not fulfilled. This is the case in ( 5.57b): the basic eventuality encoded by the <ϕdyn>-verb cvesti (‘blossom’) has been supplemented by one initial boundary, resulting in the terminative verb zacvesti (‘start to blossom’) which is pf per default. For the felicitous application of the ipf aspect in ( 5.57b), a phase needs to be induced. Solving this mismatch requires an additional operation that is captured in terms of accommodation (Lewis 1979; Thomason 1990; cf. chapter 3, section 3.1.2). Once the phase is inferred, it can be focused upon. These two deviations from default aspect assignment will be referred to as ‘mismatch’ and ‘deviation’. Both kinds of deviation give rise to M-inferences. In case of mismatch, accommodation constitutes the prerequisite for the derivation of M-inferences.

Chapter 5: Pragmatics

240

As has been pointed out in chapter 3, section 3.1.2, accommodation can be considered as repair mechanism applying if the felicity conditions or usage constraints on the use of particular expressions are not met32. Accommodation and implicature do not exclude each other; in fact, some implicatures arise from the accommodation of the context, which is triggered by operator-operandum mismatches, as illustrated in ( 5.57b). As has been shown in chapter 4, aspect operators differ in their input requirements: the Russian pf aspect requires a boundary to be present, the Russian ipf aspect is less specific in this respect and merely requires the presence of a phase. For the actual-processual reading, the input requirement is more specific, since a dynamic phase is needed. Other languages, e.g., Turkish and English, have overt markers to introduce this requirement of a dynamic phase. Thus, an input requirement may be imposed by a specific marker or introduced by a specific context.

This requirement of a certain input can be regarded as a presupposition trigger that initiates accommodation in case this presupposition is not met. Therefore, accommodation can be regarded as one of the underlying causes for implicature, that is as one of the reasons that start off pragmatic reasoning (Thomason 1990: 352). Input requirements that are not fulfilled induce a modification of the Aktionsart basis in order to provide the appropriate input for aspect to apply. This modification of the Aktionsart has to be ‘explained’ by an appropriate interpretation.

The ‘repairing’ of mismatches between aspect and Aktionsart has also been accounted for in terms of coercion (or recategorisation)33. Two advocates of this view are de Swart (1998) and Pulman (1997). de Swart (1998) takes coercion to be governed by a contextual reinterpretation process that is free as long as context supports the meaning aspects associated with the aspectual change. This process takes place before aspect applies; it does not depend on whether aspect is formally expressed or not. The details of de Swart’s (1998) approach will not be dealt with here, except for one interesting idea, namely the distinction between macro- and micro-operators. Macro-operators generalise over a number of semantic operations, captured by micro-operators34:

32 Compare also Hobbs et al. (1993) who account for this additional operation in terms of ‘abduction’. Their

approach would also be applicable to the interpretation of Russian ipf aspect, since its different interpretations are defeasible and hence derived by non-monotonic reasoning. Aspect operators do not change the semantic representation of the Aktionsart basis but require a certain input. If this input is not given, inference processes help to determine what is to be inferred in order to bridge this mismatch. In Hobbs et al., these processes are characterised as ‘backward-chaining’. The rational for abductive reasoning is the following (Hobbs et al. 1993: 81):

‘From (∀x) p(x) ⊃ q(x) and q(A), one concludes p(A)’ (∀x) p(x) ⊃ q(x) is the general principle that could explain q(A)’s occurrence. With respect to aspect, (∀x) p(x) ⊃ q(x) constitutes the default combination of aspect and Aktionsart, q(A) the observable evidence, i.e. the operator-operandum mismatch. This mismatch has to be explained in order for the expression not to be unnormal. This is achieved by means of p(A), which is the inferred, underlying cause or explanation of q(A). In most cases there are various possible p(A)’s and the contextually relevant one has to be determined by further contextual hints and/or encyclopaedic knowledge. Levinson (2000a: 60-63) criticises both the abductive approach of Hobbs et al. (1993) and Thomason’s (1990) accommodation approach, who both claim their accounts to generalise to implicatures.

33 Michaelis (2004) draws a fundamental distinction between coercion approaches and construction approaches like hers. Blutner (1998a, b) emphasises the crucial difference between the ‘coercion view’ and the ‘underspecification view’. Given this background the approach advocated here might seem problematic. However, the two rather contrary views can be unified if coercion is not understood in the sense of de Swart (1998, 2000) as changing the eventuality type, but in terms of accommodation, that is, as triggering the appropriate input for the application of aspect. The semantics of aspect is underspecified, but nonetheless aspect markers (possibly combined with further triggers provided by the context) impose their input requirements onto the context, i.e. the Aktionsart basis. If this input requirement is not met, the context has to be accommodated. It is this accommodation process that is referred to by the (coercion)-operators that will be introduced in this section.

34 de Swart’s coercion operators are the following: Ceh (event → homogeneous situation, i.e. state or process), Che (state/process → event) and Csd (state → dynamic eventuality, i.e. process or event).

Chapter 5: Pragmatics

241

“The coercion operator can be viewed as a macro operator, which generalizes over a number of semantic operators. The use of Ceh, Che and Csd, rather than specific operators INCHO, HAB, etc., illustrates the pragmatic, context-dependent nature of the reinterpretation process.” (de Swart 1998: 361)

This accounts for the fact that sentences may be underspecified in isolation. The coercion operators both enable different interpretations and ensure compositionality (de Swart 1998: 364f): “[W]e want to preserve the insight that the interpretation in terms of coercion is fully compositional, but the value of the hidden operator is dependent on linguistic context and world knowledge […]. At this point, coercion of an event into a state or process leads to a choice among the set of possible mappings {ITER, HAB, PROC, …}, and coercion of a state or process into an event leads to a choice among the set of {BOUND, INCHO, ADD-CUL, …}.”

Pulman (1997) regards aspect operators as inducing shifts in the aspectual category of the basic predicate in order for it to constitute an adequate input for the aspect marker to apply. Thomas & Pulman (1999) define coercion as follows: “When we talk of coercion, we do not mean that events actually alter – merely that our perception of them alters. We can think of a coercion as mapping between alternative views of an event.”

Contrary to de Swart, this conception of coercion does not suggest that the denoted events change. It is the perception of them that changes.

Basic to Pulman’s approach are the three aspectual categories (i.e. Aktionsarten) state, process and point and the complex categories <point, state> and <process, state> (cf. chapter 2, section 2.2.2). Aspect operators are thought of as operators that require certain input types and yield outputs of specific types. Based on Moens & Steedman (1988), Pulman (1997) and Thomas & Pulman (1999) describe a variety of coercion operators which work as functions that take the lexical basis arguments and do not change its features (cf. Jackendoff 1991: 21).

The operators proposed by Pulman (1997) are illustrated in ( 5.58):

5.58 a. iterate: point → process

b. stretch: point → process

c. pphase-of: <X, state> → X, where X is point or process

d. cstate-of: <X, state> → state

e. add-cstate: X → <X, state>, where X is point or process

f. bundle: X → point, where X is not a point The application of these operators is illustrated in ( 5.59a-d): 5.59 a. Joe is sneezing.

b. Joe is swimming.

c. Joe is winning.

d. Joe is growing old. The fact that progressives require a dynamic phase as input (cf. chapter 4, section 4.1.2) causes the coercion of a point predicate like the one in ( 5.59a) into a process, either by stretching or by iterating it35. ( 5.59b) does not require coercion, since the predicate is of the

35 Both Moens & Steedman (1988) and Pulman (1997) take the progressive to require a process as input and

yield a state as output. Given the fact that the progressive requires stage-of relations as input (chapter 4, section 4.1.2), this is not tenable. There is a difference between lexical statives, derived statives (implicitly by means of aspectual interpretation or explicitly by a respective aspect marker like the Turkish -irdi) and progressives. Compare also de Swart’s (1998: 363) slightly odd statement: “The output of the Progressive is again a state, but

Chapter 5: Pragmatics

242

appropriate type for the application of the progressive. ( 5.59c) is more complex: first, the point component of the complex predicate type <point, state> is focused upon and then this point is stretched into a process, resulting in the interpretation that Joe is in the process of winning. In ( 5.59d), only one process of coercion is required, namely focusing on the process component of the complex type <process, state>.

The advantage of this system consists in its taking into account at least some internal structure of the predicates aspect is applied to (remember that Pulman does not assume a tripartite structure). One point arising from Pulman’s coercion operators concerns an important difference between the operators he proposes: with ( 5.58c,d), the appropriate input is already provided by the Aktionsart, whereas in the other cases it has to be adjusted before aspect can apply36. Actually, the notion of coercion only applies to the latter instance, that is, to cases where the basic eventuality type needs to be modified in order to fulfill the input requirement of aspect (cf., e.g. Johanson’s (1996: 237) conception of coercion as “modification of actionality”). It is reasonable to speak of coercion, recategorisation or reinterpretation only in case the input basis has to be changed prior to the application of aspect. This is crucial, since aspect does not modify the eventuality description.

The set of operators proposed in Thomas & Pulman (1999) makes this different nature of the respective operators more transparent:

5.60 a. focus-pphase: acc/ach37 → process

b. focus-point: acc/ach → point

c. focus-cstate: acc/ach → state

d. iterate: point → process

e. stretch: point → process

f. collapse: X → point38

g. try: ach → acc

h. add-cstate: point/process → ach/acc Thomas & Pulman (1999) are more explicit about the nature of these operators, but they neglect the internal structure of the basic predicate. As can be seen in ( 5.60a-c) and ( 5.60h), they distinguish two main types of operators: ‘focus’-operators and ‘add’-operators. These

the state of an event or process being in progress is a lot more ‘dynamic’ than the underlying lexical state. The more complex internal structure accounts for the vivid colour of the description.”

36 What Pulman does not explain, however, are the conditions for the respective operator to apply (Thomas & Pulman (1999) specify some conditions). He neither states which operator is to be preferred in which context. Furthermore, each of these operators allows for different interpretations, some of which are harder to get than others. For computational purposes, a theory of aspectual coercion, or any theory of aspect interpretation, should provide some preference measure over a set of interpretations that can be assigned to a sentence. In this respect, computational approaches differ from purely linguistic ones. As has been emphasised, it is neither possible nor necessary to state the exact number and specifications of possible interpretations. What can be stated is the basic semantic configuration, but each configuration allows for various interpretations. In order to implement a theory of aspect interpretation, these interpretations have to be stated and listed in advance, such as to make it possible to state deductive rules of the form A → B (e.g. Vazov & Lapalme 2000). The abductive mode of reasoning mentioned above (cf. footnote 32) is not possible for computer applications, where the premises have to be stated explicitly in order to derive the appropriate conclusions.

37 Here, Thomas & Pulman use the terms accomplishment (acc) and achievement (ach). 38 This operator corresponds to the ‘bundle’-operator in ( 5.58). Pulman (1997) and Thomas & Pulman (1999)

take this operator to be the prerequisite for ‘iterate’ to apply: ‘bundle’ turns events into points that may then be iterated irrespective of their internal structure.

Chapter 5: Pragmatics

243

operators correspond to the distinction between the two basic types of non-correspondence to default aspect assignment mentioned at the beginning of the section, namely deviation (‘focus’) and mismatch (‘add’). The operators listed in ( 5.60d-g) are of different nature, they correspond to de Swart’s micro-operators (see below).

The system proposed here makes use of ‘focus’ and ‘add’ as two types of macro-operators. ‘Focus’ pertains to deviations from default aspect assignment, ‘add’ is triggered by operator-operandum mismatches. A third macro-operator is ‘status-change’, that applies to the derivation of second order events (habitualised, modalised, iterated, etc. events; cf. chapter 4)39.

The logical entries of aspect and Aktionsart are stable across speakers. The same is true for the application of macro-operators. In this sense, macro-operators can be said to contribute to the compositional part of interpretation (cf. de Swart 1998). The logical entry of aspect concerns the input requirement of phases and boundaries that have to be supplied by the Aktionsart basis. The macro-operators pertain to this logical entry and concern the selection/focusing of one of its elements or to their induction in case of aspect-Aktionsart mismatch. Thus, the set of elements that might be affected by coercion is restricted to ‘dynamic phase’, ‘boundary’ and ‘static phase’, and the set of possible operators to ‘focus’, ‘add’ and ‘status’. The semantic analysis of aspect as selecting units from the Aktionsart basis, and the specification of these units as phases and boundaries leads to a systematic list of such operators, which can be differentiated according to whether they add one of these units or whether they focus a unit different from the default focusing. Since the set of macro-operators is restricted by the event nucleus it can be listed exhaustively. What may be induced by these operators are the components of the event nucleus. Once the nucleus is completed, it is still possible to apply the macro-operator ‘status-change’ which abstracts away from the internal make-up of the denoted event. It is therefore necessary to clearly state the Aktionsart basis in terms of phases and boundaries, since this allows for predictions with respect to the operators that may apply, i.e. which inferences may arise.

Table 5.5 summarises the classification of operators. Not all of these operators need to be available in one single language. Whereas the list of macro-operators is exhaustive (as they pertain to a tripartite event structure of the event nucleus and thus are constrained by the conceptual possibilities), the list of micro-operators cannot be complete40 (as they pertain to the encyclopaedic entry that may vary across speakers and depending on the specific context):

MACRO

logical entry add → focus

MICRO encyclopaedic entry

pphase pphase stretch, try, process… cstate cstate result, perfect, state, … focus-aspect

boundary boundary begin, end, point, …

status-aspect status-change iterate, state, habitualise, statement-of-fact…

Table 5.5

39 The three macro-operators proposed here – add, focus and status – resemble the three forms of coercion

distinguished by van Lambalgen & Hamm (2005: 171-176), namely additive coercion, subtractive coercion and cross-coercion. Additive coercion corresponds to elaborating an Aktionsart (just like ‘add‘). Subtractive coercion is assumed to delete part of a complex Aktionsart and thus corresponds to ‘focus’. Note, however, that ‘focus’ does not delete a part of the Aktionsart, but focuses on one specific part and thereby backgrounds (but does not delete) the other part(s). van Lambalgen & Hamm apply ‘cross-coercion’ to cases where a point is turned into an activity by iteration, or where activities/accomplishments are turned into states. This operation can thus be taken to correspond to ‘status-change’.

40 Such a listing of micro-operators is useful in domains such as NLP.

Chapter 5: Pragmatics

244

Table 5.5 pertains only to deviations from default aspect assignment. The correspondences to default aspect assignment are not listed since in these cases no additional operations are required, and the interpretation can proceed straightforwardly, that is, the micro-operators guiding the specific interpretation apply immediately. Table 5.5 also illustrates that ‘add’-operations are followed by ‘focus’-operations, and the latter may occur without the former.

‘Add-phase’ and ‘focus-phase’ are possible with every IMPERFECTIVE-marker that is not marked for the selection of a consequent state: 5.61 On vyigryval gonku.

he win:PST:ipf race:ACC ‘He won / was winning / used to win / … / the race.’

5.62 Biz partiyi kaybediyorduk. we match lose:iyor:PST:1Pl ‘We were loosing the match.’ 5.63 He was reaching the summit.41 In ( 5.61)-( 5.63) the respective aspect marker requires a phase as input. As a phase is lacking with the respective predicates, it has to be induced – the context, i.e. the Aktionsart basis, needs to be accommodated for the felicitous application of the aspect marker. This is captured by the macro-operator ‘add-pphase’. Complex predicates that lexically specify a phase and a boundary (e.g. of the type <ϕdyn, τ>) do not need to be accommodated. The deviation from the default consist in the focusing of the phase instead of the boundary, the respective operator then is ‘focus-pphase’42.

Whether the macro-operator is ‘add-pphase’ or solely ‘focus-pphase’ does not have any influence on the application of micro-operators. Whether adding a pphase or focusing on it, the micro-operators that may apply remain the same and are constrained and supported by the same mechanisms.

The operators ‘add-boundary’ and ‘focus-boundary’ are available in languages with aspect markers that are marked for boundary selection and therefore require a boundary to be present in the input context. In Russian, these operators apply with in case of perfectivising prefixes/suffixes applied to aterminative verbs: 5.64 Rebenok kriknul.

child cry<nu>:PST:pf ‘The child cried out.’

It is reasonable to assume that focus-boundary does not exist in Russian, since whenever a boundary is encoded by a predicate, focusing upon it is the default case. No operator is thus needed for that. Whenever a boundary has to be induced, it is focused upon – except in case an additional operation applies that again induces a phase (secondary imperfectivisation; cf. ( 5.57b), and chapter 2, Figure 2.7). Languages that do not have aspect markers that are sensitive to boundaries cannot induce the boundary by means of aspect marking alone.

41 Compare Rothstein (2004) for a thorough analysis of progressive accomplishments (<ϕdyn, τ>) and

progressive achievements (<τ>), and the differences between the both. 42 This raises the question, whether it is possible to speak of ‘focus-pphase’ in terms of deviation from default

in English and Turkish as well, since in these languages, the notion of default aspect is not needed (cf. chapter 2, section 2.2.3).

Chapter 5: Pragmatics

245

Rather, boundary selection arises as an implicature from use of the respective non-marked markers (such as the simple form in English, or -di in Turkish).

The application of ‘add-cstate’ again depends on whether a language has a specific aspect marker that requires a consequent state, i.e. a static phase, as input. In Russian, the pf aspect cannot explicitly focus on the consequent state. Rather, this focusing may arise as an implicature mediated by discourse particles like uže (‘already’). The ipf aspect cannot unambiguously focus on the consequent state, it needs to be applied to a specific Aktionsart and needs a specific context that triggers the focusing on the consequent state:

5.65 a. (Ochotnik videl, kak) Olen’ prjatalsja .

(hunter see:Pst:ipf, how) dear hide:PST:ipf ‘(The hunter saw, how) The dear was hiding.’

b. (Nikto ego ne videl, potomu čto) On prjatalsja pod stolom. (Nobody him Neg see:Pst:ipf because) he hide:Pst:ipf under table:INS ‘(Nobody saw him, because) He hid under the table,’

The ipf aspect may either focus on the process of hiding as in ( 5.65a), or on the state of being hidden as in ( 5.65b), but it cannot trigger the induction of a consequent state. Turkish has one aspect marker that requires a static phase as input, namely the postterminality marker -miş. This marker places the focus on the consequent state of the event, the preparatory phase and the culmination point being the preconditions for it to hold, cf. ( 5.66)43:

5.66 Gözlüklerimi kaybetmişim.

Glasses:1sg:ACC lose:miş:1sg ‘I have lost my glasses.’ (= My glasses are gone.)

The third macro-operator is ‘status-change’. Basically, each event can be changed its status, as this operator abstracts away from the internal make-up of the predicate. It leaves the event representation intact and does not require any changes to be made. This operator takes the whole predicate as argument: 5.67 Ja govorju po-arabski

I speak:PRS:ipf Arabic ‘I speak Arabic.’ (= I can speak Arabic.)

Languages tend to use their respective unmarked aspect markers to induce status-change. This is not surprising given the fact that status-aspect does not focus on a specific part of the Aktionsart basis. Marked members always focus on one specific part thereby selecting it to be asserted (Russian pf – boundary; English progressive – dynamic phase, etc.) and thus cannot be used to abstract away from the lexical content of the Aktionsart. Turkish is interesting in this respect, as it has one aspect form that is marked for status-change, namely the Aorist. In chapter 4, section 4.4.2 it has been analysed in terms of a grinding function: 5.68 Babam satranç oynar.

father:1SG chess play:ir:PRS ‘My father plays chess.’ (= He can play chess.)

43 The difference between this kind of aspect marking and the perfect and pluperfect tense is that for the

former the consequent state is an input requirement specified by the aspect operator, whereas tense operators do not require a specific input.

Chapter 5: Pragmatics

246

To recap so far, there exist two kinds of deviation from default aspect assignment: mere deviation and deviation involving an operator-operandum mismatch. The former corresponds to morphologically marked deviations from default aspect assignment, the latter to the morphologically marked induction of phases and boundaries in order to meet the input requirements imposed by aspect. This induction has been captured in terms of accommodation. Accommodation causes additional interpretational efforts since this “elimination of a stumbling block” (Thomason 1990) has to be explained, or motivated, by an appropriate interpretation. The reasoning process behind this interpretation has been captured by the notion of M-inference, just as in case of mere deviation from default aspect assignment.

Table 5.6 illustrates the interaction of the different pragmatic principles outlined here with one specific example:

On vyigryval gonku. he win:PST:ipf race:ACC

‘He won / was winning / used to win / .. the race.’

Q-INFERENCE

<pf; ipf>

ipf → no boundary asserted

within/at TT

systematic semantic unmarkedness of ipf

aspect

morphological markedness (-yva-)

ACCOMMODATION

aspect: ipf

Aktionsart: <τ> → mismatch

add-pphase → <ϕdyn, τ>

M-INFERENCE focus-pphase → deviation from default

pragmatic markedness

I-INFERENCE

explanation for mismatch and

deviation → interpretation

stereotypical enrichment and specification44

Q > M > I

semantic unmarkedness > pragmatic markedness > enrichment

Table 5.6 The use of the ipf aspect (vyigryval) gives rise to the inference that the stronger alternate, the pf aspect and, consequently, the assertion of the reaching of a boundary within TT, was not possible. Therefore, a general overlap relation is inferred to hold between the event-time interval I(e) and the topic-time interval I(TT). The mismatch between aspect (ipf) and

44 At this stage, the micro-operators would apply.

Chapter 5: Pragmatics

247

Aktionsart (<τ>), which is morphologically indicated by the inserted suffix -(y)va-, requires the induction of a phase. This gives rise to the inference that the situation is marked, i.e. that it did not proceed in the default way. The default would have been the assertion of the boundary, since with terminative predicates the pf aspect is the default case. The markedness of the situation does not change the semantic unmarkedness of the ipf aspect – as concerns implicatures, Q dominates M. The markedness of the situation has to be explained and this explanation is provided by an appropriate interpretation, i.e. the specification of the encoded meaning by means of encyclopaedic knowledge activated by the basic predicate. The outcome of this process of concept formation depends on factors such as the specific situation in which the sentence is used, or specific background assumptions of the hearer. However, it is constrained by the inference that the situation is marked, i.e. does not correspond to the default – M constrains I. Taken together, this yields the ranking Q > M > I (cf. Table 5.4).

Table 5.7 incorporates the inferences needed for the interpretation of an ipf verb into the general picture of the semantics-pragmatics interaction presented in Table 5.2:

semantic representation disambigu-ation

concept formation proposition additional

proposition

aspect Aktionsart

→ Q → M/Ilog

specific relation (pragmatic strengthening)

→ I (encyclop. knowledge)

contextually relevant interpretation

further implicatures (inferences about speaker’s intentions)

the coded the said the meant SEMANTICS1 PRAGMATICS1 SEMANTICS2 PRAGMATICS2

Table 5.7 Table 5.7 accounts for the fact that inferences play a role already at the level of semantic representation and thus apply to what is coded. From the expression that has been used (here: the ipf aspect) and the knowledge about the possible alternate expression (here: the pf aspect), certain inferences arise (Q and M/Ilog). These inferences do not constitute coded information, but are based on it and are thus part of the semantic representation. This is the reason for why the coded information and the semantic representation are not presented as being congruent here (contrary to their representation in Table 5.2). The semantic representation of ipf aspect is ambiguous (cf. section 5.1.2) and pragmatic reasoning is necessary for this disambiguation process. The inferences based on the semantic representation and the disambiguated relations constrain the process of concept formation. This process takes into account the encyclopaedic information activated by the encyclopaedic entry of the basic predicate and is guided by I-inference. Concept formation differs from the other inference processes with respect to its context-dependency. It yields the contextually relevant interpretation, the proposition based on which further inferences about speaker intentions can be drawn. However, the question of concept formation and the question of how to derive further, superpropositional inferences has not been dealt with here.

Table 5.7 also shows the different components that contribute to the two kinds of semantics and pragmatics that are involved in utterance interpretation (cf. chapter 3, section 3.2.1): Semantics1 corresponds to the coded information, pragmatics1 captures all the inference processes that are needed to establish the proposition, i.e. the utterance-token meaning. This utterance-token meaning corresponds to semantics2 based on which additional inferences pertaining to pragmatics2 can be drawn.

In the preceding sections, the focus of attention has been on the Russian ipf aspect. However, the pragmatic framework presented here is applicable to cross-linguistic analyses as

Chapter 5: Pragmatics

248

well. It offers interesting insights in how aspect systems differ from each other, and it helps to derive these differences from the common basis provided by the general category of grammatical aspect. This will be illustrated in more detail in the following section.

5.2.3 Semantic invariants, pragmatic scales and cross-linguistic analysis This section illustrates the usefulness of the notion of ‘scalar implicature’ in cross-linguistic analyses in general, and in accounting for the Turkish aspect system in particular. The insufficiency of a purely semantic account for a cross-linguistic analysis of aspect has been pointed out in chapter 4 and in section 5.1 (cf. also the general remarks on cross-linguistic analysis of aspect in chapter 2). As has been shown, the language-specific IMPERFECTIVE-forms differ to a considerable degree in semantics and usage, but nevertheless exhibit certain similarities that suggest them to belong to one and the same meta-category. This can hardly be accounted for in purely semantic terms. Bickel (1996: 36) is doubtful with respect to an exclusively semantic analysis of the English instantiation of IMPERFECTIVE, i.e. the progressive form: “Given these difficulties and dilemmas, it seems doubtful whether the differences between the -ing form and a regular imperfective can all be derived from a semantic restriction on the form.”

The prima facie differences between the various IMPERFECTIVE forms discussed in chapter 4 can in fact be analysed in a consistent and straightforward way by referring to the pragmatic notion of ‘scalar implicature’ (cf. section 5.2.1).

One interesting issue for a cross-linguistic comparison of aspect systems is the at first sight puzzling fact that languages use both PERFECTIVE and IMPERFECTIVE markers to express the mere statement of a fact. As is illustrated in ( 5.69), Russian uses an IMPERFECTIVE form in this case, ( 5.69a), Turkish ( 5.69b) and English ( 5.69c) a PERFECTIVE one:

5.69 a. Ty čital ėtu knigu? you read:PST:ipf this book:ACC b. Bu kitabı okudun mu? this book:ACC read:PST:2SG Q? c. Did you read this book?

This is no longer astonishing, if the following general scales are taken in to account (cf. also Bickel 1996: 49): 5.70 a. Russian: <PERFECTIVE; ∅>

b. English: <IMPERFECTIVE; ∅>

c. Turkish: <IMPERFECTIVE; ∅> These general scales are based on the assumption that grammatical aspect can be instantiated in terms of the two meta-categories PERFECTIVITY and IMPERFECTIVITY. The scales in ( 5.70) illustrate that both English and Turkish have aspect systems whose marked members express IMPERFECTIVITY, whereas the Russian aspect system is marked for PERFECTIVITY. In order to simply state a fact, it is not necessary to explicitly select a phase or a boundary. Therefore, the respective unmarked members of the aspect systems are used – that is, a PERFECTIVE form in English and Turkish, and an IMPERFECTIVE form in Russian.

As has been outlined in chapter 4, PERFECTIVITY and IMPERFECTIVITY may be brought about differently, depending on the terminality category that is instantiated by the respective

Chapter 5: Pragmatics

249

aspect system. Accordingly, the general scales in ( 5.70) are in fact based on different terminality-categories. This is illustrated by the meta-scales in ( 5.71) that capture the language-specific types of grammatical aspect: 5.71 a. Russian: <AD; ±AD>

b. English: <INTRA; ±INTRA>

c. Turkish: <INTRA; ±INTRA> and <POST; ±POST> Moreover, it is important to take into account the language-specific markers that instantiate the meta-scales. The respective markers can be ordered along specific scales, which are based on different semantic fields. Languages differ with respect to the semantic field(s) that instantiate the meta-scales in ( 5.71), and with respect to the formal means that instantiate the specific scales. This is shown in ( 5.72): 5.72 a. Russian: perfectivity: <pf; ipf>

b. English: progressivity: <-ing; simple form>

c. Turkish: past intraterminality: <-mekteydi/-iyordu/-irdi; -di> postterminality: <-miş, -di>

non-past status-change: <-ir; -iyor/-mekte> The comparison in ( 5.72) reveals the fine-grainedness of the Turkish aspect system. The language-specific differences in fine-grainedness with respect to formal marking explain certain differences in use and interpretation of aspect markers (see below).

Figure 5.5 illustrates the relatedness of the different layers of scales to the different layers of invariant semantic meaning that have been assumed throughout this thesis:

Figure 5.5 Aspect systems can express the meta-categories PERFECTIVITY and IMPERFECTIVITY. Since these meta-categories abstract away from the notion of inherent markedness, it is possible to refer to both the English progressive form and the Russian ipf aspect as IMPERFECTIVE, and to both the English simple form and the Russian pf aspect as PERFECTIVE. These meta-categories

ASPECT

IMPERFECTIVE PERFECTIVE

INTRA AD POST

language specific aspect markers

scale

general scale; compare ( 5.70)

meta-scale; compare ( 5.71)

specific scale; compare ( 5.72)

basic invariant

cross-linguistic general invariant

language specific instantiation of (1)

1

2

language specific invariant of the markers for (2)

3

Chapter 5: Pragmatics

250

can be brought about differently, namely in terms of the three terminality categories and their markedness status. IMPERFECTIVITY in Russian is brought about by unmarked [±AD], whereas IMPERFECTIVITY in English is brought about by [+INTRA]. [+AD] is basic to PERFECTIVITY in Russian, whereas unmarked [±INTRA] is basic to PERFECTIVITY in English. This is correlated with the different meta-scales that are based on the terminality categories. Meta-scales do not rank linguistic entities, but semantic notions that are encoded by the respective language-specific aspect markers. The semantic basis of the meta-scales constitutes the common ground for the language-specific aspect markers, which in turn are ordered on specific scales. The specific scales are determined by semantic fields with respect to which the elements of the scales are ordered. The semantic fields are referred to in terms of the feature that distinguishes the informationally stronger element from the weaker one. By these scales, the hearer is entitled to draw the inference that if the weaker, semantically less specific element of the scale is used, the assertion of the stronger one is not possible or not relevant. Based on that inference, the less specific element may either negate the stronger element, or it may adopt a neutral value.

The scales in ( 5.70) explain why the Russian IMPERFECTIVE form may convey a PERFECTIVE interpretation: use of the weaker element does not exclude the specific semantics of the stronger one, it does not deny it (since this would yield an equipollent relation between the two elements). Thus, the ipf aspect does not explicitly exclude boundary-selection and, consequently, it does not exclude the event being completed and/or terminated. The same holds for the Turkish and English PERFECTIVE forms that do not necessarily convey completion and termination, but may do this in certain contexts that suggest the explicit negation of the value of the stronger element.

In order to compare aspect systems of different languages, it is necessary to specify the meta-scales as well as the specific scales. This is evident, for instance, with the English and Turkish aspect systems, that are both based on the same meta-scale <INTRA; ±INTRA>, but display considerable differences concerning the use of the respective markers. Intraterminality in English is specified by ‘progressivity’, that is, the respective form (-ing) marks progressivity (by its semantic characterisation ‘I(TT) ⊂ ϕdyn’). Intraterminality is more differentiated in Turkish, which has three markers that each instantiate intraterminality in a different way. This necessitates a more fine-grained analysis in order to reveal the relation between these markers (see below). The fact that intraterminality in Turkish is more differentiated than in English explains why Turkish uses a [+INTRA]-form, the IMPERFECTIVE Aorist, to express status-change, whereas English uses the [±INTRA]-form, i.e. a PERFECTIVE form (the simple form). The English intraterminal form is marked for the selection of dynamic phases and thus cannot abstract away from the inherent characteristics of the denoted event, whereas the intraterminal Aorist is marked for status-change. This markedness for status-change is the reason for why the Aorist cannot be used to convey the statement of a fact, contrary to the English simple form.

The scales in ( 5.71) are meta-scales. Such meta-scales, however, neither explain more specific differences in the cross-linguistic use of aspect markers, nor do they explain language-internal overlapping uses of aspect markers. The latter is evident especially in Turkish. The Turkish aspect system has been shown to cause certain difficulties for purely semantic analyses. One of the questions arising from the semantic analysis sketched in chapter 4, section 4.4, concerns the use of -iyordu with stative verbs, and its use in contexts in which typically the Aorist is used. To account for that, the Turkish meta-scale <INTRA; ±INTRA> in ( 5.71c) has to be replaced by the specific intraterminality scale, which is repeated in ( 5.73)45:

45 Johanson (1971) calls the basic opposition between the intraterminal forms and the unmarked -di the

‘opposition A’, the basic aspectual opposition in Turkish.

Chapter 5: Pragmatics

251

5.73 ‘intraterminality’: <-mekteydi/-iyordu/-irdi; -di> The scale in ( 5.73) is valid only in the past tense, since the unmarked -di is a simple past marker and cannot be used in the present tense46. Each of the forms on the left side of the scale explicitly signals intraterminality, whereas -di is unmarked in that respect and may express a neutral value or non-intraterminality. This scale shows that Turkish has several means to signal marked intraterminality. The different intraterminality-markers in turn give rise to different scales based on different semantic fields47. The different intraterminality-scales are crucial in accounting for use and interpretation of the Turkish aspect markers. Decisive for the establishment of these scales are their respective bases, that is, the semantic fields with respect to which the elements constituting the scale are opposed.

The problems that arise from a purely semantic analysis of the Turkish aspect system (cf. the data in chapter 4, section 4.4) can be accounted for in terms of the specific intraterminality scales. One of the puzzles was the fact that the alleged progressive-marker -iyordu can be applied to stative verbs ( 5.74a,b), and that it may convey future reference ( 5.75a) or modal shades ( 5.75b), i.e. it may function as status-aspect:

5.74 a. Yeni gözlüklerimle seni gayet iyi görüyorum.

new glasses:1SG:with you:ACC very well see:iyor:1SG [Kornfilt 1997: 357] ‘I see you (* I am seeing you) very well with my new glasses.’

b. Hasan fazla çabuk konuştuğunu biliyordu. Hasan too fast talk:FNom:3Sg:ACC know:iyor:PST.3Sg ‘Hasan knew (*was knowing) that he was speaking too fast.’ [ibd.]

5.75 a. Yarın geliyorum.

tomorrow come:iyor:1SG. ‘I will come tomorrow.’

b. Umutsuz yaşanmıyor. hope.without live:PASS:NEG:iyor [Johanson 1971: 121] ‘One cannot live without hope.’

In ( 5.76), these uses of -iyordu are compared to -mekteydi. This latter form is indeed a progressive marker, as it may neither be used with stative verbs ( 5.76c), nor to convey future reference ( 5.76d): 5.76 a. Hasan sorunu anlıyordu.

Hasan question:ACC understand:iyor:PST [Kornfilt 1997: 358] ‘Hasan understands (*is understanding) the problem.’

b. ?? Hasan sorunu anlamaktaydı. Hasan question:ACC understand:mekte:PST [ibd.] ‘Hasan understands the problem.’

c. *Hasan sorumun cevabını bilmekteydi. Hasan question:1SG:GEN answer:3Sg:ACC know:mekte:PST [ibd.]

‘Hasan is knowing the answer to my question.’

46 The respective scale for the present tense – <Aorist; -iyor/-mekte> – is based status-change, cf. ( 5.72c). 47 Johanson (1971) takes this opposition within the intraterminal forms (the ‘opposition B’) to be based on

the notion of ‘focality’ (compare chapter 4, section 4.4). It does not constitute an additional aspectual contrast, but rather further subdivides intraterminality.

Chapter 5: Pragmatics

252

d. *Biraz sonra gelmekteyim. a.little.bit later come:mekte:PRS:1SG. [Ergin 1998: 323] ‘I will come a little bit later.’

Apparently, the difference consists in the fact that -mekteydi is more specific than -iyordu. It requires a dynamic phase as input in order to apply, whereas -iyordu is agnostic with respect to that. Both markers can be captured by means of a scale based on progressivity: 5.77 ‘progressivity’: <-mekteydi; -iyordu> Due to its position as weaker, i.e. less specific element of the progressivity-scale, -iyordu does not necessarily express progressivity, but does not exclude it either. It does not necessarily require a dynamic phase as input. This scale accounts for the fact that -iyordu can be used non-progressively, e.g., to convey non-actual reference. In this latter use it pertains to the realm of the Aorist.

The Aorist can be regarded as status-aspect marker that abstracts away from the internal make-up of the predicate. Aorist-marked predicates do not individuate events and therefore do not introduce events into the discourse: 5.78 a. Gidip haber verecek. Telefon eder. come:CVB news give:FUT. Telephone make:ir [Johanson 1971: 186] ‘He will come and bring the news. Maybe he will simply give a call.’

b. Dikkat et, yoksa düşersin! attention do:IMP, else fall:ir:2SG [Ersen-Rasch 2001: 140] ‘Look ahead, lest you fall!’

c. Öğretmemiz iyi Almanca konuşurdu. teacher:1PL well German speak:ir:PST:3SG ‘Our teacher could speak German very well.’ The Aorist is opposed to the other intraterminal forms, as it does not report on events but is used in a characterising function. -iyordu can also be used in such a function, but its main function is to refer to events. Accordingly, -iyordu and the Aorist differ with respect to the feature parameter ‘characterising a topic entity’, or more specifically, ‘status-change’48: the Aorist is marked for this feature, whereas -iyordu does not deny it. This is the basis for the establishment of the scale ‘status-change’:

5.79 ‘status-change’: <Aorist, -iyordu> Therefore, if the speaker uses the weaker form -iyordu, the Q-heuristic allows the hearer to infer that the stronger claim was not possible or not necessary to make, which in turn gives rise to the inference of the actual reference of the situation described by the verb. This is a default only, which captures the fact that -iyordu can also be used in the domain of the Aorist.

The other intraterminal form, -mekteydi, does not constitute a possible alternate to the Aorist in that semantic field. It is, however, a possible alternate to -iyordu. As has been shown above, both forms are used in progressive contexts, but while -mekteydi is confined to these uses, -iyordu has a broader domain of application and may also be used with stative verbs

48 Johanson (1971: 138) puts it slightly differently and characterises the Aorist in a negative way, namely as

negating the actual relevance of an actional content to a certain interval of time.

Chapter 5: Pragmatics

253

conveying a continuous reading (which may shade into non-actual uses and thus pertain to the Aorist domain).

The intraterminality-scale in ( 5.73) contrasts marked intraterminality with the aspectually non-marked, neutral past tense marker -di. Use of -di defeasibly implies non-intraterminality. -di is also element of another scale, which is based on postterminality, cf. ( 5.80). This illustrates, that semantic markedness is a relative notion which can only be established with respect to a fixed basis and with respect to an alternate expression. With respect to -di, -iyordu is the marked alternate (intraterminality), with respect to -mekteydi, -iyordu is unmarked (agnostic towards progressivity).

There is another meta-scale that may instantiate IMPERFECTIVITY in Turkish, namely <POST; ±POST> (cf. 5.71c). This meta-scale is instantiated by the following specific scale: 5.80 ‘postterminality’: <-miş; -di> The marked member of this scale, -miş, expresses postterminality by explicitly selecting the lexically encoded consequent state (or inducing such a consequent state to the Aktionsart basis). The weaker element of that scale, -di, is agnostic with respect to postterminality. Thus, -di is the less informative element of two scales, of the ‘intraterminality’-scale and of the ‘postterminality’-scale. 5.81 a. Gözlüklerimi kaybetmişim. glasses:1SG:ACC lose:miş:1SG ‘I have lost my glasses.’ (= my glasses are lost)

b. Gözlüklerimi kaybettim. glasses:1SG:Acc lose:PST:1SG ‘I lost my glasses.’ The use of -miş in ( 5.81a) explicitly selects the consequent state, and the interpretation can only be that the glasses are still gone. This is different in ( 5.80b), where the use of -di does not explicitly focus on the consequent state and thus does not exclude the possibility of the glasses being found again.

In this way, the concept of pragmatic scales neatly accounts for the prima facie inconsistencies and semantic problems of the Turkish aspect system. It is possible to explain why -iyordu may be used with lexically stative verbs, even though these do not encode a dynamic phase (cf. examples ( 5.74) and ( 5.76)). Being marked for intraterminality, -iyordu requires a phase to apply. This phase cannot be stative, because Turkish has an explicit marker for the selection of a static phase, namely -miş. The remaining forms to be considered are -mekteydi and the Aorist49. The former is excluded with stative verbs as it explicitly requires a dynamic phase to be present, whereas the latter is possible with stative verbs. However, the Aorist does not report on actual events. In order locate a state of affairs denoted by a stative verb in time, thus, the only possibility is -iyordu: this form is able to introduce events into the discourse (as opposed to the Aorist, which excludes that possibility) and it is not marked for the selection of a dynamic phase (as is the case with -mekteydi).

Furthermore, it is possible to explain another cross-linguistic issue raised in chapter 4, namely the difference between the English progressive marker and the Turkish -iyordu, concerning the fact that the progressive is applicable to lexically statives only with contextual support, whereas -iyordu is fine by itself and does not give rise to additional inferences as is the case in English (e.g. agentivity or ‘acting-as-if’; cf. Žegarac 1993). This difference in use might

49 The unmarked -di is not at issue here.

Chapter 5: Pragmatics

254

be astonishing in consideration of the fact that both -ing and -iyordu are marked for intraterminality and therefore should be expected to behave in a similar way. But again, taking into account the language-specific scales, this is no longer surprising. The English aspect system is based on progressivity, that is, the intraterminal form requires a dynamic phase as input. Stative verbs do not provide such a dynamic phase and therefore, the use of the progressive marker is excluded50. This is different in Turkish, which does not only have an explicit progressive form (-mekteydi), but also an intraterminal marker that does not necessarily require a dynamic phase as input (-iyordu) and may therefore readily be applied to stative verbs.

Moreover, the scales proposed for Turkish also capture the somewhat instable situation in the Turkish aspect system. The more recent form -iyordu has pushed the Aorist out of its original domain of the actual present and past tense into non-actual, modal uses. -iyordu nowadays tends to intrude more and more into the domain of the Aorist, whereas -mekteydi is increasingly being used in spoken language in order to express progressivity (G. Lewis 2000: 111).

The contextually relevant interpretation of the respective Turkish aspectual forms then proceeds along the same lines as in Russian, i.e. it is guided by I-inference to stereotypical relations by resorting to background assumptions and conceptual knowledge.

5.3 Conclusion

This chapter has integrated the semantic analysis of aspect and Aktionsart into the pragmatic framework of Levinson’s GCI-theory. Figure 5.6 illustrates the interpretation of the Russian ipf aspect taking into account the pragmatic principles outlined in the above sections. The basic semantic skeleton is not changed, but embedded in the pragmatic framework.

Figure 5.6

50 This is slightly oversimplified, since there are specific cases, where the combination of stative verb and

progressive is not only possible but also required. However, a detailed analysis of the English progressive is beyond the scope of this thesis.

verb + Aktionsart • default combinations • deviations and mismatches

IPF as meta-variable • relation 1 • relation 2 • relation 3

ipf – specific relation • reading 1 • reading 2 • reading 3 • ...

contextually relevant interpretation

scalar implicature

disambiguation

enrichment

scale: <pf; ipf>

ambiguity

generality

specificity

Q-/M-inference

disambiguation

I-inference

Chapter 5: Pragmatics

255

There are several differences as compared to Figure 5.2 (section 5.1.1). First of all, the notion of privative opposition is replaced by the more flexible notion of lexical/logical scale established on the basis of a certain semantic field. The elements of a scale are ordered according to informativity. The use of the stronger element is more informative, since the encoded information is more specific and, consequently, the range of possible interpretations is more restricted than in case of the less specific element. The Russian aspect system is based on the semantic field ‘perfectivity’, the respective scale being <pf; ipf>. The pf aspect is marked for boundary-selection, which is not excluded by the ipf aspect.

Furthermore, the two instances of underspecification illustrated in Figure 5.2 are specified in terms of ambiguity and sense-generality. The ipf aspect is ambiguous between three specific relations, and the specific relations are sense-general with respect to the final interpretation. Both kinds of underspecification are solved by recourse to pragmatic reasoning. In case of ambiguity, pragmatics interleaves with semantic representation (cf. Table 5.7). Disambiguation is possible by means such as explicit specification of TO and/or TT, discourse mode, triggering expressions or verb semantics. Figuring out these factors and combining them may require not only semantic decoding, but also pragmatic reasoning. It is only after this disambiguation, that the exact semantic contribution of the ipf aspect can be specified.

The sense-generality of the three specific relations is resolved by pragmatic principles guiding the interpretation process. The pragmatic processes involved in this enrichment process constitute the main subject of relevance theoretic approaches. They have not been dealt with here.

The decisive improvement in Figure 5.6 is the inclusion of pragmatic principles, indicated by the dotted arrows and boxes that serves to illustrate their defeasible character. Figure 5.6 thus integrates the non-compositional component involved in the interpretation of aspectualised verbs. The respective reasoning processes are nonmonotonic, they are defaults and can be overridden by the following context. Therefore, this model allows for the possibility of backtracking as well.

The account of aspect interpretation delivered in this chapter incorporates semantic and pragmatic factors to an equal degree. Semantics and pragmatics work hand in hand in the interpretation of grammatical aspect. Acknowledging the contribution of pragmatics to the interpretation process enables a semantic analysis of grammatical aspect that is cross-linguistically applicable and at the same time captures the language-specific differences in use and interpretation of the respective aspect markers in a systematic way.

6 CONCLUDING REMARKS

6.1 Summary

The main issues of this thesis have been outlined in chapter 1 as follows: (a) What is the common denominator that justifies the subsumption of certain linguistic

phenomena under the label of ‘grammatical aspect’? (b) How do aspect and verbal basis (‘Aktionsart’) interact? (c) Facing the variety of possible interpretations for the Russian ipf aspect – is it an instance

of multiple ambiguity or is there one invariant meaning basic to all of its uses? (d) What is the status of the various aspect readings? (e) Can the cross-linguistic differences in aspect use and interpretation be accounted for in

a systematic, predictable way? These specific questions concerning semantics and interpretation of grammatical aspect in general and of Russian ipf aspect in particular have been dealt with against the background of a more general question concerning the interaction of semantics and pragmatics: (f) How do semantics and pragmatics interact in the interpretation of underspecified

linguistic expressions?

As point of departure, chapter 2 has provided an introduction to the issue of analysing grammatical aspect. Two of the main difficulties in dealing with aspect in Russian concern the lack of consistent overt marking and the multitude of possible interpretations. The major problem with many existing theories on Russian aspect consists in the confusion of semantics and interpretation. These two issues relate to the cross-linguistic perspective as well: aspect categories of different languages vary to a considerable degree in usage and interpretation, but at the same time exhibit enough similarities to suggest that they belong to one general category. This requires an analysis of aspect that captures the common basis of this general category and specifies the features in which the language specific instantiations deviate from it and differ from each other. Several basic distinctions have been introduced in chapter 2: first, the distinction of both terminativity and boundedness from aspect and second, the distinction of aspect and Aktionsart. The latter distinction has been justified by the conceptual notion of event nucleus. The aspectually relevant features of those encoded by predicates have been argued to be phases (dynamic ϕdyn, and static ϕstat) and boundaries (τ). Their various combinations allow for a classification of verbs and VPs into Aktionsarten. Aspect systems differ with respect to which of these features they are sensitive to, and at which level of Aktionsart composition this sensitivity manifests itself.

Chapter 3 was concerned with the interaction of semantics and pragmatics in utterance interpretation in general. In order to state this interaction adequately, it needs to be explicated what exactly is understood by ‘semantics’ and ‘pragmatics’. These notions have been defined in terms of the kind of information they contribute to the interpretation process – semantics as encoded information and pragmatics as inferred information. With respect to aspect interpretation, the semantic contribution – the explicitly encoded information – is delivered by the aspect marker and the Aktionsart basis it applies to. This information is stored in the

Chapter 6: Concluding remarks

258

respective lexical entries. The place of interaction for aspect and Aktionsart consists in their logical entries where the Aktionsart provides temporal tier units, i.e. phases and boundaries, and where aspect specifies certain input requirements (in terms of these temporal tier units) and a certain output in case the requirements are fulfilled. Semantic information does not deliver all the information necessary for arriving at a truth-conditional proposition. To bridge the gap between encoded information and the proposition expressed, pragmatic principles guiding the inference of additional information play a crucial role. The pragmatic principles that apply in case of the interpretation of closed-class morphemes like grammatical aspect, have been argued to consist in Levinson’s generalised conversational implicatures (GCI). These implicatures apply to the information specified by the logical entries of aspect and Aktionsart. The logical entries and the inferences arising from them have been the main concern of this thesis, whereas encyclopaedic entries and the inferences arising with them have not been dealt with. The pragmatic principles that apply there are best captured by another framework, namely that of Relevance Theory.

The aim of chapter 4 was to figure out the semantics of aspect and its interaction with the Aktionsart basis. Based on the cross-linguistic characterisation of aspect in terms of the relation between the topic time interval I(TT) and the run time interval of the event I(e), the aspect systems of Russian, Bulgarian and Turkish have been analysed. The main focus was on the Russian ipf aspect. Bulgarian and Turkish have supported the claims made for Russian and have illustrated the cross-linguistic applicability of the general semantics outlined in this chapter. Another goal of chapter 4 was to show that it is neither possible nor necessary to figure out and list all the interpretations that are possible for the ipf aspect. What needs to be stated are the different semantic configurations that are possible for one aspect-Aktionsart combination. These configurations provide the basis for, and impose constraints on, the various interpretations may be constructed in a given context. The semantics provided in chapter 4 has been stated in a way that allows for its interaction with pragmatics.

The pragmatic principles applying to that semantic basis have been outlined in chapter 5, as well as the interaction of semantics and pragmatics in aspect interpretation. The pragmatic framework applied was that of Levinson’s (2000a) generalised conversational implicatures (GCI) that have been argued to be especially appropriate for paradigmatically opposed morphemes. GCI arise from heuristics based on the form of the expressions used and on the knowledge about their possible alternative expressions. Since they are based on linguistic knowledge, they are metalinguistic and context-independent. GCI work together in constraining the process of concept formation that guides the context-specific interpretation of an aspectualised verb in an actual utterance. GCI and concept formation constitute two different mechanisms of pragmatic reasoning that do not exclude, but rather complement each other. The main concern was to illustrate how GCI apply to the interpretation of the Russian ipf aspect. Chapter 5 has also illustrated the advantages of such an analysis in terms of semantics and pragmatics in the cross-linguistic analysis of grammatical aspect.

Summing up, the questions of Chapter 1 have been answered as follows: The common denominator, the invariant semantic contribution of grammatical aspect, consists in the establishment of a relation between two intervals, I(TT) and I(e) and in the selection of a specific part of the denoted event encoded by the respective predicate. I(TT) comprises the time the assertion is made about, I(e) stretches over the part of the run time of the event that has been selected by aspect. There are thus three basic components that contribute to the semantics of aspect: I(TT), I(e) and ‘selection’. Theses components can be modified in predictable ways: I(TT) may be unbounded, open-bounded or closed-bounded, the selection may pertain to one of the temporal tier units ϕdyn, τ or ϕstat, and the relation between the two intervals may be ‘⊂’, ‘=’ or ‘⊃’. Different combinations of these components allow to state the semantic basis for all the aspect operators considered in this thesis. This conception of

Chapter 6: Concluding remarks

259

aspect semantics also specifies its interaction with the Aktionsart basis: the matching features are provided at the temporal tier in the logical entry of the Aktionsart and specified as input condition in the logical entry of the aspect marker.

The ipf aspect in Russian has been ascribed one invariant semantic basis that is present in all its occurrences – the overlap relation between I(TT) and I(e). This general relation is ambiguous between three specific relations that are each sense-general and allow for a range of possible interpretations to be constructed from them. Thus the question of ambiguity vs. sense-generality is no either-or question but depends on the level one is concerned with. Since the different readings are constructed from the sense-general specific relations, they cannot be enumerated in an a priori list as would be the case with disambiguation possibilities. Discussions about these readings, that is, which and how many should be proposed, in fact deal with the wrong issue – what is crucial instead are the basic semantic configurations. The interpretations to be constructed from them are not part of a semantic analysis but need to be accounted for in pragmatic terms, more precisely, in terms of concept formation along the lines proposed by, e.g., Carston (2002).

Acknowledging the equal contribution of semantics and pragmatics allows for a systematic cross-linguistic analysis of aspect. The semantics is stable within the domain specified above, differences can be attributed to different outcomes of the pragmatic principles applied to these slightly different semantic bases. The notion of scalar implicature has turned out useful in this respect.

6.2 Outlook

This thesis has covered only a small part of the issues that need to be addressed in dealing with aspect. One future direction of research in that field could be directed towards natural language processing (NLP) (Sonnenhauser 2003, 2004a). The analysis presented here might serve as a starting point for the identification the semantic and pragmatic factors in the aspecto-temporal system of Russian that are necessary for such NLP applications to work.

Although Thomason (1990) is probably too radical in claiming that pragmatic theories are only of use if they can be implemented in NLP applications, he is right in pointing out that this should be one goal of such analyses. The formalisation of an approach like the one provided here should therefore be carried out in a way that facilitates its implementation. This goal is still far away from the current state of analysis, however, the first steps towards this direction have been made. Without a principled account of pragmatic, i.e. inferential, principles, applications in natural language processing will necessarily fail, since the semantically encoded information provides only part of the input that is necessary for interpretation. The lexically given information has to be combined with information from former parse-states – since the interpretation is modelled as processing incrementally from left to right – as well as with general heuristics of pragmatic reasoning and with access to conceptual knowledge. Since the pragmatic principles applied here are based mainly on form and expression alternates, they should be formalisable in a suitable way.

For NLP, the decisive input factors have to be stated as well as the principles of their interaction with each other and with the context. Contrary to human interpretation of underspecified forms, where interpretations are constructed predominantly by abductive reasoning, computer driven interpretation has to select meanings from a definite list of possibilities. For the ipf aspect this means that a list of readings has to be compiled, the factors involved their derivation have to be fixed and rules of interaction have to be stated. These rules must be expressible in terms of the deductive form A → B (cf. Vazov & Lapalme 2000), where A is the premise and states the conditions for the rules to apply, and B is the conclusion, i.e. the appropriate reading in a given context.

Chapter 6: Concluding remarks

260

IPF TT ○ e

The semantic analysis of Russian aspect developed here has revealed a classification into three groups that are each based on one specific relation between I(TT) and I(e) and that each allow for a certain range of interpretations to be constructed. This is repeated in Table 6.11:

Group TO Relation Reading of Russian ipf I. synchronous TT ⊂ ϕdyn

[TT open-bounded] actual-processual, conative

II. synchronous TT = e [TT unbounded]

habitual, continuous, potential, permanent, atemporal

III. retrospective TT ⊃ e [TT closed-bounded]

general-factual, durative

Table 6.1 With respect to NLP, the disjunctive analysis of the ipf aspect proves useful. Even though this disjunction may not be consciously present in human processing (cf. Poesio 1996a), it provides a suitable means for computers to select the appropriate interpretation from. Stating the ipf aspect merely in terms of being ‘not perfective’ would be much too general to yield reasonable results. Since this grouping relies on basic semantic configurations, it should be possible for every encountered aspectualised verb to be assigned to one of the three groups mentioned in Table 6.1.

A first premise-conclusion pair consists in the following form: 6.1 IF such-and-such semantic configuration

THEN such-and-such reading The semantic configuration can be figured out by means of TO or TT, which may both be explicitly given. In chapter 5, section 5.1, the connection between TO and TT has been pointed out and some of the means to specify the one or the other (or the one by the other) have been shown. Because of the close relation between TO and TT, only one of the two parameters may be enough to specify the relevant aspectual relation.

Accordingly, ( 6.1) can be specified as follows: 6.2 a. IF TO syn/retro

THEN relation 1/2/3

b. IF TT open-bounded/unbounded/closed-bounded THEN relation 1/2/3

Once the relation is determined, the range of interpretations is constrained to those that have been assigned to the respective group in advance. Note again, that this a priori list differs fundamentally from human interpretation.

The sequel to ( 6.2) is thus the following: 6.3 IF relation 1/2/3

THEN reading out of group I/II/III The readings are each connected with a specific micro-operator (chapter 5, section 5.2.2),. The main difficulty in this respect is to decide which micro-operator applies in which context,

1 Note again that this table does not list repetitive events.

Chapter 6: Concluding remarks

261

i.e. which interpretation should be chosen from those specified within one group. This difficulty has to do with what has been called ‘concept formation’ in human processing in the preceding chapters.

In the remainder, it will be illustrated how a rule-guided interpretation, that incorporates also pragmatic principles, might work. This requires stating the input factors, determining default combinations and providing solution strategies in case of deviations from that default.

Among the input factors required for aspect interpretation are the following (cf. chapter 5, section 5.1.1): verbs indexed for the phases and boundaries they encode, lexical items (e.g., NPs, adverbials, preposition phrases, etc.) indexed for whether they add phases or boundaries, and aspect markers indexed for what element they select and for their status within the language specific semantic markedness relation. Based on these latter specifications of aspect markers – selection and markedness status – Q-inferences are drawn. In case of the Russian ipf aspect this is the inference ‘no boundary selection’, which excludes one specific aspect relation, namely ‘TT ⊃ τ’. In order to figure out which specification of the general ipf relation ‘TT ○ e’ holds in a given context, the determination of TO and/or TT is crucial. The determination of TO and TT is important insofar, as both constrain the possible interpretations of the unmarked aspectual partner by disambiguating ‘IPF’. In order to capture M-inferences, the default combinations of aspect and Aktionsart have to be stated, as well as rules for resolving possibly occurring mismatches. M-inferences then can be pinned down by ‘add’-, ‘focus’- and ‘status’-operators (chapter 5, section 5.2.2; cf. also the proposals in Pulman 1997 and Thomas & Pulman 1999). The meta-linguistic Q- and M-inferences can be handled purely on the basis of the lexically given input and on the basis of the possible alternatives, which have to be stated as well. The most difficult problem is the question of how to specify verbs for the conceptual knowledge they provide access to, which is indispensable for I-inference to be drawn. One means to cope with this problem is corpus analysis in order to detect regularities and coocurrences of lexical items that might hint to a conceptual connection. Since the factor ‘probability’ cannot be completely eliminated, a condition has to be implemented preferring the shortest line of reasoning (Thomas & Pulman 1999). Finally, a mechanism must be implemented that parallels the stop mechanism in human interpretation, i.e. a mechanism that stops the search for information to be built in the interpretation process. The reasoning stops as soon a truth-conditional interpretation is achieved.

The default case is a match of basis and aspect marker2, that is, the verbal basis provides the necessary input for the marker to apply. For the Russian ipf aspect, the conditions have to be stated under which the three possibilities (Table 6.1) are activated. In this respect, TO and TT – primarily specified by temporal or manner adverbials (e.g. vse bol’še ‘more and more’, chorošo ‘well’) – are decisive.

Certain adverbials and particles fix TO as retrospective and the reading as being one out of group III. The rule for this line of interpretation can be stated as follows (adopted from Vazov & Lapalme 2000):

6.4 IF ipf is applied to a verb providing a phase

AND if there is an adverbial/particle fixing TO as retrospective THEN the reading is out of group III.

6.5 a. GENERAL-FACTUAL READING

Ja uže rasskazyval vam ėtu istoriju. I already tell:PST:ipf you:DAT this story:ACC ‘I already told you this story.’ 2 Remember that ‘marker’ is the general term that subsumes also cases of default aspect, where there is no

overt morphological marking.

Chapter 6: Concluding remarks

262

b. DURATIVE READING Ja guljala ot trech to pjati. I take.a.walk:PST:ipf from three:Gen to five:Gen ‘I took a walk from three to five.’

The decisive items here are uže (‘already’) in ( 6.5a), and ot…do (‘from … to’) in ( 6.5b). Both specify TO as retrospective and thus the interpretation to be one out of group III.

Both interpretations can be overridden if I(TT) is changed, or if an additional I(T-Val) is induced. This may be the case with adverbials of the type vsegda (‘always’) or obyčno (‘usually’): 6.6 a. HABITUAL READING

Ja obyčno guljala ot trech to pjati. I usually take.a.walk:PST:ipf from three to five ‘I usually took a walk from three to five.’

b. [TOsyn; I(T-Val); -b obyčno [TOretro; I(TT); +b ot...do[TOsyn; I(e); -b guljal]]]

The layered representation in ( 6.6b) reflects the incremental way of interpretation, which leaves the inner parts intact. The interpretation of aspect forms processes incrementally, i.e., information once provided and processed cannot be undone. The boundedness characteristics of the outermost interval are relevant for discourse advancement or non-advancement, respectively.

A synchronous TO can be connected with both an open-bounded or an unbounded TT (group I and II, Table 6.1). The respective rule is stated in ( 6.7).

6.7 IF ipf is applied to a verb providing a phase AND if there is an adverbial fixing TO as synchr., and TT as open-bounded/unbounded THEN the reading is out of group I/II ( 6.8a) is an instance of a synchronous TO combined with an open-bounded TT, ( 6.8b) illustrates the combination of a synchronous TO with an unbounded TT: 6.8 a. ACTUAL-PROCESSUAL READING

V vosem’ časov, ja čitala. at eight o’clock, I read:PST:ipf ‘At eight o’clock, I was reading.’

b. INACTUAL READING Ran’še, on rabotal v universitete. before he work:PST:ipf at university:LOC ‘He used to work at university.’ (= ‘He was working as a teacher.’)

The above examples have illustrated default cases in which aspect and Aktionsart match. However, deviations from that default may occur as well. As has been outlined in chapter 5, section 5.2.2, non-correspondence to default aspect assignment can be of two kinds: deviation and mismatch. Both require an additional operation: either, an operation that makes explicit the non-default selection, or one that adds the feature that is necessary for the application of the respective aspect marker. After these operations have been carried out, specific operators yielding the contextually relevant interpretation apply. The application of the micro-operators proceeds just as in the default cases, and is confronted with the same difficulty of figuring out the correct micro-operator in the respective context.

Chapter 6: Concluding remarks

263

With respect to non-correspondences to the default case, the make-up of the basic Aktionsarten is important in order to predict possible inferences. Depending on the semantic representation of the verb, implicatures or presuppositions may arise. Ipf applied to <ϕ, τ>-predicates leaves the reaching of the final boundary as an implicature, ipf applied to <τ, ϕ>-predicates leaves the initial boundary as presupposition. Furthermore, the semantic representation of the verb indicates the ‘add’- and ‘focus’-operators that may apply3. This illustrates, that the semantic representations of verbs provide background and frame for pragmatic reasoning.

In ( 6.9a), the pf aspect is applied to a <τ>-predicate. This is the default case and, consequently, no additional operation is necessary. 6.9 a. Ivan vyigral gonku. Ivan win:PST:pf race:ACC

‘Ivan won the race.’

In case of ( 6.9b), however, the Aktionsart basis does not provide the necessary feature for the application of the ipf aspect:

b. Ivan vyigryval gonku (četyre raza). Ivan win: PST:ipf race:ACC (four times).

‘Ivan won the race four times / was winning the race.’ The application of the ipf aspect in ( 6.9b) requires a phase to be present, which the verb vyigrat’ (‘win’) does not provide. Thus, the ‘add-phase’-operator has to be applied. This operator causes the introduction of micro-operators such as ‘stretch’ or ‘try’4. In most cases, context provides the necessary cues for disambiguation (e.g. a when-clause), if not, one has to rely on some ‘probability-condition’ (see above). The rules for the possible interpretation of ( 6.9b) can be stated as follows: 6.10 a. IF ipf is applied to a verb providing no phase, AND a lexical item indicating iteration is present

THEN add the phase and apply ‘iterate’

b. IF ipf is applied to a verb providing no phase AND an adverbial/clause indicating incidence is present

THEN add the phase and apply ‘stretch’ The application of the ipf aspect onto a <τ>-basis is morphologically and, consequently, pragmatically5 marked, but this does not have an impact on the semantic unmarkedness of the ipf aspect. Although interpretation in terms of coercion is compositional, the specific reading depends on linguistic context and world-knowledge (de Swart 1998), as is shown in ( 6.11)6: 6.11 On rešal zadaču.

he solve:PST:ipf exercise:ACC

3 Remember that the application of operators is constrained by the event nucleus. 4 The application of ‘status-change’ would give rise to micro-operators like ‘iterate’ or ‘habitualise’. 5 Remember that in Russian the implication is ‘morphological (formal) markedness → pragmatic

markedness’, and not ‘morphological (formal) markedness → semantic markedness’. 6 The readings listed here involve different degrees of context-dependency.

Chapter 6: Concluding remarks

264

(i) ACTUAL-PROCESSUAL READING ‘He was solving the exercise.’ (ii) CONATIVE READING ‘He tried to solve the exercise.’ (iii) GENERAL-FACTUAL READING ‘He (*solved)7 the exercise.’ Whereas ( 6.11iii) can be disambiguated by fixing TO as retrospective, ( 6.11i) and ( 6.11ii) cannot be distinguished by TO alone, since both require it to be synchronous. The distinction between the possible readings is left to context and world-knowledge. Gaining probability values and assigning them to interpretations by a statistical approach that takes into account judgements of native speakers seems to be a possible way (cf. Glovinskaja 1982 for the habitual, potential and actual-processual reading). Pulman & Thomas (1999) take the interpretation requiring the shortest line of coercion as the most probable one (i.e. the interpretation that requires the least amount of ‘add’- and ‘focus’-operators). Their approach resembles the weighted abduction approach proposed in Hobbs et al. (1993), that assigns assumability costs to the different interpretations and in case of multiple possible interpretations chooses less expensive (and thus most probable) one.

The discussion above has outlined a first starting point, a brief and rather sketchy outline of how a rule-guided derivation of aspectual readings might look like. However, a number of problems remain to be solved. Corpus analyses and the appropriate annotation of verbs, aspect markers and adverbials are the prerequisite for formulating rules that enable the systematic derivation and computation of the readings. Furthermore, the contribution of factors located outside the sentence context (i.e. on the paragraph or discourse level) has to be taken into account.

7 Note that this general-factual reading arises only in special contexts. With this specific example it is

difficult to be preserved in the English translation.

REFERENCES ACL 2004. Workshop on Text Meaning and Interpretation. Barcelona Akman, V. & M. Surav 1996. Steps towards formalizing context. AI Magazine 17/3, 55-72 Aksu-Koç, A. 1988. The Acquisition of Aspect and Modality. The Case of Past Reference in

Turkish. Cambridge: Cambridge University Press Andrejčin, L. 1978. Osnovna Bălgarska Gamatika. Sofia: Nauka i Izkustvo Atlas, J. & S. Levinson. 1981. It-clefts, informativeness, and logical form: radical pragmatics

(revised standard version). Cole, P. (ed). Radical Pragmatics. New York: Academic Press, 1-61

Atlas, J. 1989. Philosophy without Ambiguity. Oxford: Clarendon Press Atlas, J. 1997. On the modularity of sentence processing: semantical generality and the language

of thought. Nuyts, J. & E. Pederson (eds). Language and Conceptualization. Cambridge: Cambridge University Press, 213-228

Babko-Malaya, O. 1999. Zero Morphology: A Study of Aspect, Argument Structure and Case. Ph.D. Dissertation. Rutgers

Bach, E. 1981. On time, tense, and aspect: an essay in English metaphysics. Cole, P. (ed). Radical Pragmatics. New York: Academic Press, 63-81

Bach, E. 1986. The algebra of events. Linguistics and Philosophy 9, 5-16 Bach, K. 1994. Conversational impliciture. Mind and Language 9, 124-162 Bach, K. 1999. The semantics-pragmatics distinction: what it is and why it matters. Turner, K.

(ed). The Semantics/Pragmatics Interface from Different Points of View. Oxford: Elsevier, 65-84

Beaver, D. 1997. Presupposition. Van Benthem, J. & A. ter Meulen (eds). The Handbook of Logic and Linguistics. Oxford: Elsevier, 939-1008

Beaver, D. 1999. Presupposition accommodation: a plea for common sense. Moss, L; J. Ginzburg & M. de Rijke (eds). Logic, Language and Computation, Vol. 2. Stanford: CSLI Publications, 21-44

Bennett, W. et al. 1989. Towards a computational model of aspect and verb semantics. Machine Translation 4, 247-280

Bickel, B. & J. Nichols. to appear. Inflectional morphology. Shopen, T. (ed). Language Typology and Syntactic Description. Cambridge: Cambridge University Press (Draft version 2/18/01)

Bickel, B. 1996. Aspect, Mood and Time in Belhare. Zürich: ASAS Bickel, B. 1997. Aspectual scope and the difference between logical and semantic representation.

Lingua 102, 115-131 Bickel, B. 2000. Unlogischer Aspekt. Breu, W. (ed). Probleme der Interaktion von Lexik und

Aspekt. Tübingen: Niemeyer, 1-20 Blakemore, D. 2002. Relevance and Linguistic Meaning. The Semantics and Pragmatics of

Discourse Markers. Cambridge: Cambridge University Press Blutner, R. 1998a. Lexical underspecification and pragmatics. Ludewig, P. & B. Geurts (eds).

Lexikalische Semantik aus kognitiver Sicht. Tübingen: Narr, 141-171 Blutner, R. 1998b. Lexical pragmatics. Journal of Semantics 15, 115-162 Blutner, R. 2000. Some aspects of optimality in natural language interpretation. Journal of

Semantics 17, 189-216 Blutner, R.; A. Leßmöllmann & R. van der Sandt. 1996. Conversational implicature and lexical

pragmatics. Proceedings of the AAAI Spring Symposium on Conversational Implicature. Stanford, 1-9

Bohnemeyer, J. & M. Swift. 2001. Default aspect: the semantic interaction of aspectual viewpoint and telicity. Talk delivered at the conference Perspectives on Aspect. Utrecht

Bohnemeyer, J. & M. Swift. 2004. Event realization and default aspect. Linguistics and Philosophy 27, 263-296

References

266

Bondarko, A. V. (ed). 1987. Teorija Funkcional’noj Grammatiki. Vvedenie, Aspektual’nost’, Vremennaja Lokalizovannost’, Taksis. Leningrad: Nauka

Bondarko, A. V. 1990. O značenijach vidov russkogo glagola. Voprosy Jazykoznanija 39/4, 5-24 Borik, O. & T. Reinhart. 2004. Telicity and perfectivity: two independent systems. Keynote

lecture delivered at the Eight Symposium on Logic and Language, Debrecen Borik, O. 2002. Aspect and Reference Time. Ph.D. Dissertation, University of Utrecht Breu, W. 1985. Handlungsgrenzen als Grundlagen der Verbalklassifikation. Lehfeldt, W. (ed).

Slavistische Linguistik 1984. München: Sagner, 9-34 Breu, W. 1988. Resultativität, Perfekt und die Gliederung der Aspektdimension. Slavistische

Linguistik 1987. München: Sagner, 42-74 Breu, W. 1996. Komponentenmodell der Interaktion von Lexik und Aspekt. Slavistische

Linguistik 1995. München: Sagner, 37-74 Breu, W. 1998. Aspektuelle Verbklassen, insbesondere Inchoativa und Inzeptiva. Berger, T. & J.

Raecke (eds). Slavistische Linguistik 1997. München: Sagner, 55-80 Breu, W. 2000. Zur Position des Slavischen in einer Typologie des Verbalaspekts. Breu, W. (ed).

Probleme der Interaktion von Lexik und Aspekt. Tübingen: Niemeyer, 21-54 Breu, W. 2003. Flexivischer und derivativer Verbalaspekt im Moliseslavischen. Berger, T. & K.

Gutschmidt (eds). Funktionale Beschreibung slavischer Sprachen. Beiträge zum XIII. Internationalen Slavistenkongreß in Ljubljana. München: Sagner, 63-81

Breu, W. 2005. Fokusovyj i statusovyj vid v verchnelužickom razgovornom jazyke. Lehmann, V. (ed.). Semantika i Struktura Slavjanskogo Vida IV. München: Sagner (in print)

Bunt, H. & B. Black. 2000. The ABC of computational pragmatics. Bunt, H. & B. Black (eds). Abduction, Belief and Context in Dialogue. Amsterdam, Philadelphia: John Benjamins, 1-46

Cann, R. et al. 2004. The Dynamics of Language. Final draft (december 2004) Cappelle, B. & R. Declerck. 2005. Spatial and temporal boundedness in English motion events.

Journal of Pragmatics 37, 889-917 Carlson, G. 2000. Weak Indefinites. Ms. University of Rochester Carston, R. 1998a. Implicature, explicature, and truth-theoretic semantics. Kasher, A. (ed),

Pragmatics. Critical Concepts, Vol. IV. London, New York: Routledge, 436-479 Carston, R. 1998b. Informativeness, relevance and scalar implicature. Carston, R. & S. Uchida

(eds). Relevance Theory: Applications and Implications. Amsterdam: John Benjamins, 179-236

Carston, R. 1999. The semantics/pragmatics distinction: a view from relevance theory. Turner, K. (ed). The Semantics/Pragmatics Interface from Different Points of View. Oxford: Elsevier, 85-125

Carston, R. 2002. Thoughts and Utterances. Oxford: Blackwell Carston, R. 2004a. Relevance theory and the saying/implicating distinction. Horn, L. & G. Ward

(eds). Handbook of Pragmatics. Oxford: Blackwell, 633-656 Carston, R. 2004b. Explicature and semantics. Davies, S. & B Gillon (eds). Semantics. A Reader.

Oxford: Oxford University Press, 817-845 Carston, R. 2004c. Word Meaning and Lexical Pragmatics. Talk delivered at the Graduate School

“Universality and Diversity”, Leipzig Ceyda Arslan, Z. 2001. “Perfect” in Turkish: the real puzzle. ms. Boğaziçi University Channell, J. 1994. Vague Language. Oxford: Oxford University Press Chierchia, G. & S. McConnell-Ginet. 2001. Meaning and Grammar. An Introduction to

Semantics. Cambridge, London: MIT Press Cole, P. (ed). 1981. Radical Pragmatics. New York: Academic Press Comrie. B. 1976. Aspect. An Introduction to the Study of Verbal Aspect and Related Problems.

Cambridge: Cambridge University Press Croft, W. 1990. Possible verbs and the structure of events. Tsohatzidis, S. L. (ed.). Meanings and

Prototypes. Studies in Linguistic Categorization. London, New York: Routledge, 48-73 Croft, W. 1991. Syntactic Categories and Grammatical Relations: The Cognitive Organization of

Information. Chicago: University of Chicago Press

References

267

Croft, W. 1998. The structure of events and the structure of language. Tomasello, M. (ed). The New Psychology of Language. Mahwah, London: Erlbaum, 67-92

Cruse, D. 1986. Lexical Semantics. Cambridge: Cambridge University Press Csató, E. 1994. Tense and actionality in Hungarian. Thieroff, R. (ed). Tense Systems in European

Languages. Tübingen: Niemeyer, 231-246 Dahl, Ö. 1995. The marking of the episodic/generic distinction in tense/aspect systems. Carlson,

G. & J. Pelletier (eds). The Generic Book. Chicago: University of Chicago Press, 412-425 Dascal, M. 1981. Contextualism. Parret, H.; M. Sbisà & J. Verschueren (eds). Possibilities and

Limitations of Pragmatics. Amsterdam, Philadelphia: John Benjamins, 153-177 de Swart, H. 1998. Aspect shift and type coercion. Natural Language and Linguistic Theory 16/2,

347-385 de Swart, H. 2000. Tense, aspect and coercion in a cross-linguistic perspective. Butt, M. & T.

King (eds). Proceedings of the Berkeley Formal Grammar Conference. Stanford: CSLI Publications

Dekker, P. & R. van Rooy 2000. Bi-directional optimality theory: an application of game theory. Journal of Semantics 17, 217-242

Demjjanow, A. 1998. Eine semantische Analyse der Perfektivierungspräfigierung im Russischen. Fallstudie pere-. München: Sagner

Depraetere, I. 1995. On the necessity of distinguishing between (un)boundedness and (a)telicity. Linguistics and Philosophy 18, 1-19

Desclés, J.-P. 1993. Un modèle cognitif d’analyse des temps du français: méthode, réalisation informatique et perspectives didactiques. Catégories Grammaticales: Temps et Aspects. Seoul, 1-18

Desclés, J.-P. & Z. Guentchéva. 1990a. Is the notion of process necessary? Colloque international sur l’aspect. Cortona, University of Pisa, 1-17

Desclés, J.-P. & Z. Guentchéva 1990b. Discourse analysis of aorist and imperfect in Bulgarian and French. Thelin, N. (ed). Verbal Aspect in Discourse. Amsterdam: John Benjamins, 237-261

Dölling, J. to appear. Semantische Form und pragmatische Anreicherung: Situationsausdrücke in der Äußerungsinterpretation. Zeitschrift für Sprachwissenschaft

Dowty, D. 1986. The effects of aspectual class on the temporal structure of discourse: Semantics or pragmatics? Linguistics and Philosophy 9/1, 37-61

Dowty, D. 1991. Thematic proto-roles and argument selection. Language 67, 547-619 Durst-Andersen, P. 1992. Mental Grammar: Russian Aspect and Related Issues. Columbus, Ohio:

Slavica Egg, M. 1994. Aktionsart und Kompositionalität. Berlin: Akademie Verlag Ergin, M. 1998. Üniversitetler için Türk Dili. İstanbul: Bayrak Ersen-Rasch, M. 1997. Türkisch für Sie. Grammatik. Ismaning: Hueber Ersen-Rasch, M. 2001. Türkische Grammatik. Ismaning: Hueber Filip, H. & G. Carlson. 1997. Sui generis genericity. Penn Working Papers in Linguistics, Vol. 4.

University of Pennsylvania, Philadelphia, 91-110 Filip, H. 1996. Quantification, aspect and lexicon. Kruijff, G.-J.; G. Morrill & D. Oehrle (eds).

Proceedings of the ESSLLI ’96 Conference on Formal Grammar. Prague: Charles University, 43-56

Filip, H. 1997. Integrating telicity, aspect und NP semantics: the role of thematic structure. Toman, J. (ed). Formal Approaches to Slavic Linguistics 3. The College Park Meeting 1994. Ann Arbor: Slavic Publishing, 61-96

Filip, H. 1999. Aspect, Eventuality Types and Noun Phrase Semantics. New York: Garland Filip, H. 2000. The quantization puzzle. Pustejovsky, J. & C. Tenny (eds), Events as Grammatical

Objects. Stanford: CSLI Publications, 3-60 Filip, H. 2001. On accumulating and having it all. Talk delivered at the conference Perspectives

on Aspect. Utrecht Filip, H. 2003. Prefixes and the delimitation of events. Journal of Slavic Linguistics 12/1-2, 55-101

References

268

Filip, H. 2005. On accumulating and having it all. van Hout, A.; H. de Swart & H. Verkuyl (eds). Perspectives on Aspect. Dordrecht: Kluwer, 125-148

Filip, H. to appear. Measures and indefinites. Carlson, G. & F. Pelletier (eds). Reference and Quantification: The Partee Effect. Stanford: CSLI Publications

Flier, M. & A. Timberlake (eds). 1985. The Scope of Russian Aspect. Columbus: Slavica Flier, M. 1985. The scope of prefixal delimitation in Russian. Flier, M. & A. Timberlake (eds).

The Scope of Russian Aspect. Columbus: Slavica, 41-58 Fodor, J. A. 2001. Language, thought and compositionality. Mind and Language 16, 1-15 Forsyth, J. 1970. A Grammar of Aspect. Cambridge: Cambridge University Press Frederking, R. 1996. Grice’s maxims: “Do the right thing.” Talk delivered at the AAAI

Symposium on Computational Implicature. Stanford, CA Friedman, V. 1985. Aspectual usage in Russian, Macedonian, and Bulgarian. Flier, M. & A.

Timberlake (eds). The Scope of Russian Aspect. Columbus: Slavica, 234-246 Gazdar, G. 1979. Pragmatics. Presupposition and Implicature. New York: Academic Press Georgiev, S. 1999. Morfologija na Bălgarskija Knižoven Ezik. Veliko Tărnovo: Abagar Gigerenzer, G. et al. 1999. Simple Heuristics That Make Us Smart. Oxford: Oxford University

Press Gillon, B. 2004. Ambiguity, indeterminacy, deixis, and vagueness. Davis, S. & B. Gillon (eds).

Semantics. A Reader. Oxford: Oxford University Press, 157-187 Glovinskaja, M. 2001. Mnogoznačnost’ i Sinonimija v Vido-Vremenno Sisteme Russkogo

Glagola. Moskva: Azbukovnik Glovinskaja, M. J. 1982. Semantičeskie Tipy Vidovych Protivopostavlenija Russkogo Glagola.

Moskva: Nauka Grice, P. 1989. Studies in the Way of Words. Cambridge: Harvard University Press Grønn, A. 2003. The Semantics and Pragmatics of the Russian Factual Imperfective. Ph.D.

Dissertation. University of Oslo Guentchéva, Z. 1989. Implications aspecto-temporelles en français et en bulgare. Sǎpostvitelno

Ezikoznanie XIV/5, 26-37 Harnish, R. 1976. Logical form and implicature. Bever, T; J. Katz & T. Langendoen (eds). An

Integrated Theory of Linguistic Ability. New York: Crowell, 313-392. Reprinted in: Kasher, A. (ed). Pragmatics. Critical Concepts, Vol. IV. London, New York: Routledge, 230-341

Heinamäkki, 1994. Aspect as boundedness in Finnish. Bache, C.; H. Basbøll & C. Lindberg (eds). Tense, Aspect and Action: Empirical and Theoretical Contributions to Language Typology. Berlin: Mouton de Gruyter, 207-233

Hirschberg, J. 1991. A Theory of Scalar Implicature. New York: Garland Hobbs, J. et al. 1993. Interpretation as abduction. Artificial Intelligence 63, 69-142 Hopper, P. & S. Thompson. 1984. The discourse basis for lexical categories in universal

grammar. Language 60/4, 703-752 Horn, L. R. 1984. Toward a new taxonomy for pragmatic inference: Q-based and R-based

implicature. Schiffrin, D. (ed). Meaning, Form, and Use in Context. Washington DC: Georgetown University Press, 11-42

Jackendoff, R. 1987. The status of thematic relations in linguistic theory. Linguistic Inquiry 18/3, 369-411

Jackendoff, R. 1991. Parts and boundaries. Cognition 41, 9-45 Jakobson, R. 1971. Zur Struktur des Russischen Verbums. Selected Writings, Vol. II. The Hague,

Paris: Mouton, 3-15 Janssen, T. 1997. Compositionality. van Benthem, J. & A. ter Meulen (eds). Handbook of Logic

and Language. Oxford: Elsevier, 417-473 Jaszczolt, K. 1999. Default semantics, pragmatics, and intentions. Turner, K. (ed). The

Semantics/Pragmatics Interface from Different Points of View. Oxford: Elsevier, 199-232 Jaszczolt, K. 2002a. Against ambiguity and underspecification: evidence from presupposition and

anaphora. Journal of Pragmatics 34, 829-849

References

269

Jaszczolt, K. 2002b. Semantics and Pragmatics. Meaning in Language and Discourse. London: Longman

Jespersen, O. 1924. The Philosophy of Grammar. Chicago: University of Chicago Press (Reprinted 1992 with an introduction and index by J. D. Cawley)

Johanson, L. 1971. Aspekt im Türkischen. Uppsala: Berlingska Boktryckeriet Johanson, L. 1994. Türkeitürkische Aspektotempora. Thieroff, R. (ed). Tense Systems in European

Languages. Tübingen: Niemeyer, 247-266 Johanson, L. 1996. Terminality operators and their hierarchical status. Devriendt, B. et al. (eds).

Complex Structures: A Functional Perspective. Berlin, New York: de Gruyter, 229-258 Johanson, L. 2000. Viewpoint operators in European languages. Dahl, Ö. (ed). Tense and Aspect

in the Languages of Europe. Berlin, New York: de Gruyter, 27-179 Kamp, H. & U. Reyle. 1993. From Discourse to Logic. Dordrecht: Kluwer Kempson, R. et al. 2001. Dynamic Syntax. Oxford: Blackwell Kissling, H. 1960. Osmanisch-Türkische Grammatik. Wiesbaden: Harrassowitz Klein, W. 1994. Time in Language. London, New York: Routledge Klein, W. 1995. A time-relational analysis of Russian aspect. Language 71/4, 669-695 Klein, W. 2001. On finiteness. Ms. MPI Nijmegen Klein, W. 2003. Finiteness, universal grammar and the language faculty. Ms. MPI Nijmegen Koenig, J.-P. & N. Muansuwan. 2000. How to end without ever finishing: Thai semi-perfectivity.

Journal of Semantics 17/2, 147-184 Kornfilt, J. 1997. Turkish. London: Routledge Koschmieder, E. 1953. Das türkische Verbum und der slavische Verbalaspekt. Münchner

Beiträge zur Slavenkunde. Festschrift für Paul Diels. München, 3-15 Kratzer, A. 1995. Stage-Level and individual-level predicates. Carlson, G. & J. F. Pelletier (eds.),

The Generic Book. Chicago: University of Chicago Press, 125-175 Krifka, M. 1998. The origins of telicity. Rothstein, S. (ed). Events and Grammar. Dordrecht:

Kluwer, 197-235 Lakoff, G. 1970. A note on vagueness and ambiguity. Linguistic Inquiry 1/3, 357-359 Landman, F. 1992. The progressive. Natural Language Semantics 1, 1-32 Lascarides, A. & N. Asher. 1993. Temporal interpretation, discourse relations and commonsense

entailment. Linguistics and Philosophy 16, 437-493 Lascarides, A. et al. 1996. Ambiguity and coherence. Journal of Semantics 13/1, 41-65 Lehmann, V. 1993. Die russischen Aspekte als gestufte Kategorien. Die Welt der Slawen 38/2,

265-297 Lehmann, V. 1997. Der Aspekt – wie lexikalische Kategorien grammatische Funktionen

motivieren. Slavistische Linguistik 1996, München: Sagner, 137-154 Lehmann, V. 1999. Aspekt. Jachnow, H. (ed), Handbuch der sprachwissenschaftlichen Russistik

und ihrer Grenzdisziplinen. Wiesbaden: Harrassowitz, 214-242 Leinonen, M. 1982. Russian Aspect, ‘Temporal’naja Lokalizacija’, and Definiteness/

Indefiniteness. Helsinki Leiss, E. 2000. Artikel und Aspekt. Die grammatischen Muster von Definitheit. Berlin, New York:

de Gruyter Levinson, S. 1983. Pragmatics. Cambridge: Cambridge University Press Levinson, S. 1987. Minimization and conversational inference. Verschueren, J. & M. Bertuccelli-

Papi (eds). The Pragmatic Perspective. Amsterdam, Philadelphia: John Benjamins, 61-129 Levinson, S. 1995. Three levels of meaning. Palmer, R. (ed). Grammar and Meaning. Cambridge:

Cambridge University Press, 90-115 Levinson, S. 1997. From outer to inner space: linguistic categories and non-linguistic thinking.

Nuyts, J. & E. Pederson (eds). Language and Conceptualization. Cambridge: Cambridge University Press, 13-45

Levinson, S. 2000a. Presumptive Meanings. Cambridge, London: MIT Press Levinson, S. 2000b. H.P. Grice on location on Rossel Island. Chang, S. S.; L. Liaw & J.

Ruppenhofer (eds). Berkeley Linguistics Society 25, 210-224.

References

270

Lewis, D. 1979. Scorekeeping in a language game. Journal of Philosophical Logic 8, 339-359; Reprinted in: Portner, P. & B. Partee (eds). 2002. Formal Semantics. The Essential Readings. Oxford: Blackwell, 162-177

Lewis, G. 2000. Turkish Grammar. Second Edition. Oxford: Oxford University Press Lindstedt, J. 1985. On the Semantics of Tense and Aspect in Bulgarian. Helsinki: University Press Lyons, J. 1977. Semantics. Vol. I and II. Cambridge: Cambridge University Press Marten, L. 2002. At the Syntax-Pragmatics Interface. Verbal Underspecification and Concept

Formation in Dynamic Syntax. Oxford: Oxford University Press Maslov, J. S. 1959. Glagol’nyj vid v sovremennom bolgarskom literaturnom jazyke (značenie i

upotreblenie). Voprosy Grammatiki Bolgarskogo Literaturnogo Jazyka. Moskva: Akademija Nauk SSSR, 157-312

Maslov, J. S. 1974. Zur Semantik der Perfektivitätsopposition. Wiener Slavistisches Jahrbuch 20, 107-122

Maslov, J. S. 1985. An outline of contrastive aspectology. Maslov, J. S. (ed). Contrastive Studies in Verbal Aspect. Heidelberg: Groos, 1-44

Mehlig, H. R. 1981. Satzsmantik und Aspektsemantik im Russischen. Zur Verbklassifikation von Zeno Vendler. Slavistische Linguistik 1980. München, 95-151

Michaelis, L. 2004. Type shifting in Construction Grammar: an integrated approach to aspectual coercion. Cognitive Linguistics 15/1, 1-67

Moens, M. 1987. Tense, Aspect and Temporal Reference. Ph. D. Dissertation, Edinburgh Moens, M. & M. Steedman, M. 1988. Temporal ontology and temporal reference. Computational

Linguistics 14/2, 15-28 Mønnesland, S. 1984. The Slavonic frequentative habitual. de Groot, C. & H. Tommola (eds).

Aspect Bound. Dordrecht: Kluwer, 53-76 Musan, R. 1997. Tense, predicates, and lifetime effects. Natural Language Semantics 5, 271-301 Nakhimovsky, A. 1988. Aspect, aspectual class, and the temporal structure of narrative.

Computational Linguistics 14/2, 29-43 Padučeva, E. V. 1996. Semantičeskie Issledovanija. Moskva: Jazyki Russkoj Kul’tury Pančeva, R. 2003. The aspectual makeup of perfect participles and the interpretation of the

perfect. Alexiadou, A. et al. (eds). Perfect Explorations. The Hague: Mouton de Gruyter, 277-308

Parikh, P. 2000. Communication, meaning, and interpretation. Linguistics and Philosophy 23, 185-212

Parikh, P. 2001. The Use of Language. Stanford: CLSI Publications Partee, B. 1973. Some structural analogies between pronouns and tenses in English. Journal of

Philosophy 70, 601-609 Passonneau, R. 1987. Situations and intervals. Proceedings of the 25th Annual Meeting of the

Association for Computational Linguistics. Stanford University, 16-24 Passonneau, R. 1988. A computational model of the semantics of tense and aspect. Computational

Linguistics 14/2, 44-60 Perrett, D. 2000. The Dynamics of Tense Construal in Haddiyya. Ph.D. Dissertation, University of

London Pianesi, F. & A Varzi. 1996. Events, topology, and temporal relation. The Monist 79/1, 89-116 Pinkal, M. 1985. Logik und Lexikon. Die Semantik des Unbestimmten. Berlin: de Gruyter Poesio, M. 1996a. Underspecification and the semantics/pragmatics interface. Proceedings of the

5th International Conference on the Cognitive Science of Natural Language Processing. Dublin, 177-185

Poesio, M. 1996b. Semantic ambiguity and perceived ambiguity. van Deemter, K. & S. Peters (eds). Ambiguity and Underspecification. Stanford: CSLI Publications

Pul’kina, I. M. & E. B. Sachava-Nekrasova. 1995. Russkij Jazyk. Praktičeskaja Grammatika c Upražnenijami. Moskva: Russij Jazyk

Pulman, S. 1997. Aspectual shift as type coercion. Transactions of the Philological Society 95/2, 279-317

References

271

Radeva, V. (ed). 2003. Bulgarische Grammatik. Hamburg: Buske Rassudova, O. P. 1984. Upotreblenie Vidov Glagola v Russkom Jazyke. Moskva: Izdatel’stvo

Moskovskogo Universiteta Reichenbach, H. 1947. Elements of Symbolic Logic. New York: Macmillan Robinson, E. A. 1997. The cognitive foundations of pragmatic principles: implications for

theories of linguistic and cognitive representation. Nuyts, J. & Pederson, E. (eds). Language and Conceptualization. Cambridge: Cambridge University Press, 253-271

Rothstein, S. 1999. Fine-grained structure in the eventuality domain: the semantics of predicative adjective phrases and Be. Natural Language Semantics 7, 347-420

Rothstein, S. 2004. Structuring Events. A Study in the Semantics of Lexical Aspect. Oxford: Blackwell

Ruhl, C. 1989. On Monosemy. A Study in Linguistic Semantics. Stony Brook: State University of New York Press

Sag, I. 1981. Formal semantics and extralinguistic context. Cole, P. (ed). Radical Pragmatics. New York: Academic Press, 273-294

Saul, J. 2002. What is said and psychological reality: Grice’s project and relevance theorists’ criticisms. Linguistics and Philosophy 25, 347-372

Savaşır, İ. 1986. Habits and abilities in Turkish. Slobin, D. & K. Zimmer (eds.). Studies in Turkish Linguistics. Amsterdam, Philadelphia: John Benjamins, 137-146

Scharten, R. 1997. Exhaustive Interpretation: A Discourse-Semantic Account. Ph.D. Dissertation, Catholic University Nijmegen

Schoorlemmer, M. 1995. Participal Passive and Aspect in Russian. Utrecht: Led Sell, G. 1994. A Comparison of Verbal Aspect in Russian and Bulgarian. Ann Arbor: UMI Singh, M. & M. Singh. 1997. The role of semantic context in the temporal interpretation of

events. Proceedings of the IJCAI-95 Workshop on Context in Natural Language Processing. Montreal

Singh, M. 1998. On the semantics of the perfective aspect. Natural Language Semantics 16, 171-199

Smith, B. 1997. Boundaries. An essay in mereotopology. Hahn, L. (ed). The Philosophy of Roderick Chisholm. LaSalle: Open Court, 534-561

Smith, C. 1997. The Parameter of Aspect. Dordrecht: Kluwer Smith, C. 2003. Modes of Discourse. Cambridge: Cambridge University Press Smith, C. to appear. Aspectual entities and tense in discourse. Kempchimsky, P. & S. Roumyana

(eds). The Syntax, Semantics and Acquisition of Aspect. Dordrecht: Kluiwer Sonnenhauser, B. 2003. Aspect and the Semantics-Pragmatics Interface. Proceedings of RANLP-

03. Borovets Sonnenhauser, B. 2004a. Underspecification of ‘Meaning’: The Case of Russian Imperfective

Aspect. Proceedings of the ACL-04 Workshop on ‘Text Meaning and Interpretation’. Barcelona, 89-96

Sonnenhauser, B. 2004b. Aspect in Russian and Turkish. Semantics and pragmatics of a grammatical category. Turkic Languages 8/2, 245-270

Späth, A. 2004. Determination im Satzkontext. Habilitationsschrift, University of Leipzig Spencer, A. & M. Zaretskaya. 1998. Verb prefixation in Russian as lexical subordination.

Linguistics 36, 1-39 Sperber, D. & D. Wilson. 1998. The mapping between the mental and the public lexicon.

Carruthers, P. & J. Boucher (eds). Language and Thought. Cambridge: Cambridge University Press, 184-200

Sperber, D. & D. Wilson. 1995. Relevance. Communication and Cognition. Oxford: Blackwell Stalnaker, R. 1972. Pragmatics. Davidson, D. & G. Harman (eds), Semantics of Natural Language.

Dordrecht: Reidel, 380-397 Stankov, V. 1980. Glagolnijat vid v Bǎlgarskija Knižoven Ezik. Sofia: Nauka i Izkustvo Steube, A. 1997. Der russische Aspekt und die Ereignisrolle des Verbs. Junghanns, U. & G.

Zybatow (eds). Formale Slavistik. Frankfurt: Peter Lang, 213-227

References

272

Stojanov, S. 1999. Gramatika na Bălgarskija Knižoven Ezik. Veliko Tărnovo: Abagar Stoll, S. 2001. The Acquisition of Russian Aspect. Ph.D. Dissertation, University of California Thelin, N. 1978. Towards a Theory of Aspect, Tense and Actionality in Slavic. Uppsala Thelin, N. 1990. Verbal aspect in discourse: on the state of the art. Thelin, N. (ed). Verbal Aspect

in Discourse. Amsterdam: John Benjamins, 3-88 Thomas, J. & S. Pulman. 1999. Bidirectional interpretation of tense and aspect. Bunt, H. et al.

(eds). 1999. Proceedings of the 3rd International Workshop on Computational Semantics. Tilburg, 247-263

Thomason, R. 1990. Accommodation, meaning, and implicature: interdisciplinary foundations for pragmatics. Cohen, P.; J. Morgan & M. Pollack (eds). Intentions in Communication. Cambridge, London: MIT Press, 325-363

Timberlake, A. 1982. Invariance and the syntax of Russian aspect. Hopper, P. (ed). Tense – Aspect. Between Semantics and Pragmatics. Amsterdam, Philadelphia: John Benjamins, 305-331

Timberlake, A. 1985. Reichenbach and Russian aspect. Flier, M. (ed). The Scope of Slavic Aspect. Columbus: Slavica Publishers, 153-168

Tornow, U. 1994. Das Verb im Türkischen. Wiesbaden: Harrassowitz Traugott, E. 2004. A critique of Levinson’s view of Q- and M-inferences in historical pragmatics.

Journal of Historical Pragmatics 5/1, 1-25 Turner, K. 1997. On ‘Pragmatic considerations in semantic analyses’. Pragmatics and Cognition

5/1, 163-176 van der Sandt, R. 1992. Presupposition projection as anaphora resolution. Journal of Semantics 9,

333-377 van Lambalgen, M. & F. Hamm. 2005. The Proper Treatment of Events. Oxford: Blackwell vanden Wyngaerd, G. 2001. Measuring events. Language 77/1, 61-90 Vazov, N. & Lapalme, G. 2000. Are the temporal structures of texts algorithms? Proceedings of

the 7th International Conference on Principles of Knowledge Representation and Reasoning. Breckendridge, 76-86

Vazov, N. 1998. Identification des Valeurs Aspecto-Temporelles des Situations. Le Case de Passé Composé. Ph.D. Dissertation, ISNA-CAMS, Paris-Sorbonne

Vendler, Z. 1967. Verbs and Times. Vendler, Z. (ed). Linguistics in Philosophy. Ithaca, New York: Cornell University Press, 97-121

Wiemer, B. 2000. Presuppozicii i implikatury v tolkovanijach predel’nych sobytij i sootnosimych s nimi processov. Naučno-Techničeskaja Informacija, Serija 2: Informacionnye Processy i Sistemy №1, 31-43

Wilson, D. & D. Sperber. 1993. Linguistic Form and Relevance. Lingua 90, 1-25 Wilson, D. & D. Sperber. 1998. Pragmatics and time. Carston, R. & S. Uschida (eds). Relevance

Theory: Applications and Implications. Amsterdam: John Benjamins, 1-22 Wilson, D. 2005. New directions for research on pragmatics and modularity. Lingua 115/8, 1129-

1146 Yavaş, F. 1980. On the Meaning of the Tense and Aspect Markers in Turkish. Ph.D. Dissertation,

University of Kansas Yavaş, F. 1982. The Turkish aorist. Glossa 16/1, 40-53 Zaliznjak, A. A. & A. D. Šmelev. 1997. Lekcii po Russkoj Aspektologii. München: Sagner Žegarac, V. 1993. Some observations on the pragmatics of the progressive. Lingua 90, 201-220 Zhang, Q. 1998. Fuzziness, vagueness, generality, ambiguity. Journal of Pragmatics 29, 13-31 Zwicky, A. & Sadock, J. 1975. Ambiguity tests and how to fail them. Kimball, J. (ed.). Syntax

and Semantics IV. New York: Academic Press, 1-36