The Rise of Passives in North Slavic and Baltic

63
What makes Grammaticalization? A Look [rom its Fringes and its Components Edited by Walter Bisang Nikolaus P. Himmelmann Björn Wiemer Mouton de Gruyter Berlin . New York

Transcript of The Rise of Passives in North Slavic and Baltic

What makes Grammaticalization? A Look [rom its Fringes and its Components

Edited by

Walter Bisang Nikolaus P. Himmelmann Björn Wiemer

Mouton de Gruyter Berlin . New York

The evolution ofpassives as grammatical constructions in Northem Slavic and Baltic languages1

Björn Wiemer

Introductory remarks

The morphological structure of Slavic and Baltic languages is very similar, at least their derivation al morphology (both for verbs and for nouns), which in the following will be of special interest. 2 One of the crucial points to be made in this article is that languages sharing a large amount of morphosyntactic techniques and markers can show considerable differences in their paradig­matic organisation and grammatical status. Since most of these markers are even etymologically closely related, they also form a convenient class to show different stages of grammaticalization. For this reason I will make use of data from different Northem Slavic3 and Baltic languages, trying to show that, al­though they share a considerable amount of properties, the central · factors which decide on the (rise ofthe) grammatical status ofthese properties lies in the way they fit into particular constructions called 'passive'. Jb this we will have to add some categorial distinctions into which passives get involved. AI­though these languages are related also areally, the present study will not take up the topic of areal convergences as such. Rather I will attempt to show how the various morphological "ingredients" of passives, despite their partially common Indo-European heritage, have to be analyzed in order to disclose the language-specific contributions they make for the rise and functioning of passive constructions. In connection with this I will argue that the evolution of passives has to be judged as the result of a couple of convergent devel­opments in the application of affixes and obliquely marked noun phrases on clause level; these developments intimately interact with the lexicon. Actually, lexical changes often precede grammatical ones. Precisely for this reason lex­ical and grammatical processes have to be distinguished as clearly as possible.

The principal problem to be discussed here can also be looked at from an­other perspective. Standard, i.e. morpheme-based theories of grammatic­alization start from the observation . that very often units, which are first se­

. mantically, syntactically and morphonologically autonomous, gradually lose

272 Bjöm Wiemer

their independence in these regards. They simultaneously become dependent on other units, which they begin to determine regularly with respect to some (pre-conceived) set· of grammatical functions. From these premises the as­sumption is made (explicitly or implicitly) that grammaticalization is deter­mined (or, at least, can be described) by criteria which rest on (a) the gradual semantic and phonological erosion of linguistic material, (b) features of syn­tagmatic and paradigmatic cohesiveness and scope, and (c) syntagmatic and paradigmatic variability. Along these lines grammaticalization parameters have been successfully developed and applied by C. Lehmann ([1985] 1995); other researchers, Bybee, Heine, and Giv6n, for instance, basically abide by the same principles. Even other scholars, such as Traugott or Hopper, who take a perspective differing from Lehmann's or Bybee's, eventually base their statements on the assumption that grammaticalization concerns morphemes which, by becoming parts of what these scholars consider to belong to gram­mar, "erode" , "bleach out" and are more dependent on the context than before (though the notion of 'context' and 'scope' may actually differ).

Now, if we look from such an angle at the paths of evolution of passives in Baltic and Slavic languages, we notice that standard assumptions concern­ing grammaticalization processes turn out not to be well-suited in order to de­scribe adequately what is going on when certain morphemes begin to "collab­orate" as a passive CONSTRUCTION. Nor can this development be captured by Traugott's notion of 'subjectification'. For the "classical", morpheme-based approaches toward grammaticalization, we are thus left with a paradox: prob­ably nobody doubts that passives are a genuine and central part of the grammar of particular languages; they are language-specific and do not exist in alllan­guages, though they share some basic syntactic, semantic and pragmatic func­tions (see next section). How, then, should we call processes in and by which passives arise? It seems straightforward to call them 'grammaticalization', since such processes manifest a "movement towards grammar": But, as I will show further, some important aspects of the evolution of passives can~ot be captured by the standard criteria of grammaticalization theory, i.e. those crite­ria which center around the etymology of particular morphemes. Does it fol­low from this that grammar sometimes arises without grammaticalization, or do we have to widen, or modify, our view and criteria on grammaticalization?

I will pick up this question at the end. The article is structured as folIows: before giving the basic facts about the status quo in Baltic and Northern Slavic languages (Section 2), I provide adefinition of 'passive' which allows us to focus on its cross-linguistically pertinent representatives in Slavic and Bal­tic languages (see Table 1). Since the involved phenomena are very complex and cannot be reasonably laid down and put together even in one long article,

Passives as grammatical constructions 273

I restrict myselfto the so-called "personal passive", leaving aside passive-like constructions which lack an agreeing (nominatival) subject (otherwise. called

"impersonal" or "autonomous passive"). This notion of the passive rests on some basic assumptions made in Role&Reference Grammar (Section 1). Af­ter these two sections the main part (Section 3) will be dedicated to discussing facts of what we know about the development of passives in Baltic and North­em Slavic and theirtheoretical underpinnings (Section 3). On the basis ofthis Section 4 examines which of the standard criteria of grammaticalization are crucial for an understanding of this development and to what extent. Finally, Section 5 presents some additional criteria, which I consider necessary for a more comprehensive theory of grammaticalization.

1. Delimitation of phenomena (in the framework of Role and Reference Grammar)

By a full-ftedged 'passive' I understand a construction in which the high­est-ranking semantic argument of a transitive verb is syntactically defocussed (demoted), while a lower-ranking argument is given a syntactically privil­eged (promoted) position, namely that of the traditional subject which con­trols agreement of number and gender witb the verb and is coded in the nom­inative. The corresponding active construction is morphologically unmarked, and the privileged syntactic position is occupied by the highest-ranking ar­gument. When being demoted, in many languages it can be expressed in the guise of a prepositionless oblique case (e.g. the instrumental) or a pp (e.g. the English by-phrase). This notion of passive applies to languages with a nomina­tive-accusative alignment, and it is strict in the sense that it narrows down the notion of 'subject' and makes it applicable basically only for languages with an elaborated case system. Since, however, all the languages with which we will be dealing here are just of this type, we can leave it at this.4

Argument ranking will here be understood according to the assumptions of RRG. Very briefty, these assumptions are the following: thematic roles can be ordered and cluster along a continuum. The highest-ranking argument (doser to or identical with an agentive role) is ascribed the status of Actor, the lowest-ranking argument (closer to or identical with a patient) is ascribed the status of Undergoer. Actor and Undergoer constitute macroroles, which are defined as "generalizations across classes of specific argument positions in 10-gical structure" (Van Valin/LaPolla 1997: 142).5 Grammatical (i.e. syntactic) relations furthermore are defined by the notion of 'privileged syntactic argu­ment' (PSA). A PSAmay be either a controller or a pivot of syntactic construc-

274 Bjijm Wiemer

tions. Grammatical relations in general, then, are constructions in which there exist restricted neutralizations of argument positions inherent to the lexical entries of verbs.6 Sueh neutralizations manifest themselves in such construc­tional properties as Equi-NP-Deletion (cf. (Ia', b')), control ofreflexives, par­ticipial relativization and ellipsis of the first syntactic argument (= subject) in coordinate c1auses ifthe respective arguments are coreferentia1.7 Such proper­ties render either a controller or a controlled element (pivot). A typical repre­sentative of a PSA is the grammatical subject as defined above.

Now, if there is a passive in the language it must be possible to eliminate the predicate's highest-ranking argument (=Actor) from the syntax and to promote a lower-ranking argument (usually the Undergoer) to the position of PSA. This means that in a language with accusative alignment it is allowed to code the Undergoer (or, at least, a lower-ranking argument) of transitive verbs as subject. We may call this a derived subject. The choice of PSA in the c1ause is, thus, not wholly determined by the semantic relations of the arguments to the predicate (verb). Compare, for instance, such an example of the English passive as (1 b):

(1) a. Peter.A slapped lohn. U a'. Peterj.A slapped lohnk.U0j/*k to get into a betier mood b. lohn.U was slapped by Peter.A b'. lohnk. U was slapped by Peterj.A0*ilk to get into a better mood.

The NP lohn in (1 b) betrays the same behavioral and coding properties as Peter in the active sentence (la). That these properties do not entirely depend on the semantic relation of the syntactic argument to the predicate can be seen if one compares sentences (1 a') and (1 b'): in (1 a') it is the Actor which controls the infinitival complement, in (1 b') it is the Undergoer; but in any case it is the syntactic subject of the finite c1ause. As the examples in (1) show, passives comprise a PSA, and this is what makes them part ofthe particular language's grammar (in terms ofRRG). . ~

Th~re is another component of passive constructions, to which much less attention has been paid in the literature, namely: many (most?) passives can code the syntactically demoted (backgrounded) Actor as an oblique NP or a PP. The relation between its syntactic status and its role in the predicate's ar­gument structure is very peculiar: though in virtually every descriptive ap­proach to passives oblique agent-phrases are treated as syntactically periph­eral NPs/PPs, they code a semantic argument of the predicate, Le. represent a part of the verb' s lexical entry. This is what allows one to say that active and passive constructions are synonymous (in any case, denotationally), or that a

Passives as grammatical constructions 275

"full-fledged" passive defocusses the Actor (agent) only in terms of morpho­syntactic coding, but not in terms of changes in the lexical explication.8

Figure 1 displays the way the different linkings to the syntax in the active and the passive may be represented formally. This representation, however, takes into account only the synchronic relation between both voices by re­ferring them to an identical lexical structure. From a diachronic perspective, identity of lexical content (i.e. argument structure) is only a very late step in restructuring the morphology and syntax of the clause. This step, as I hope to demonstrate in the following, was preceded by a considerable amount of con­vergent changes mainly in the derivational morphology, but caused also by se­mantic reanalysis, analogic expansion and competition between paradigmat­ically interrelated adjunct phrases. Syntactic reanalysis played only a minor role. There is probably only one case (namely that of the genetivus auctoris in East Baltic) where it has to be ascribed a significant role in the evolution of passives (see 3.4.2). Considered from this angle, the appearance of agent phrases in passive constructions should be qualified as a I.Ond of "re-intro­duction" of a semantic argument from a syntactic adjunct into the verb's lex­ical structure; this argument then becomes (re-)established as the Actor proper (see 3.4). The guiding question which I come back to after the discussion of the data is therefore: can the various phenomena involved in the gradual rise of

[CL [C Mary [N surprised] Sally]]

~ ~ Actor

i do' (Mary ACY' 0] CAUSE [INGR surprised' (SallYINA)]

a. Active-voice linking

[CL [C Mary [N surprised] Sally]] ~ ~

Actor

i do' (MaryINA' 0] CAUSE [INGR surprised' (Sally ACY)]

b. Passive-voice linking

Figure 1. Voice linking in Role and Reference Grammar (Source: Van Valin and LaPolla 1997: 291)

276 Bjöm Wiemer

passive constructions be subsumed under grammaticalization and, if so, how should this process be described? In order to answer this I will attempt to il­lustrate why passives and the way they grammaticalize can only be reasonably captured by treating them as constructions which are composed of rather dis­parate elements, but with the common grammatical properties I have defined above following RRG.

2. Basic information on the relevant passive constructions (contemporary state)

'Northem Slavic' comprises those Slavic languages that are spoken north ofthe Balkan Peninsula and the Carpathian mouhtains (see n. 3). In these languages we can distinguish two types of passives; for short examples see Table 1. The first one consists of a verb form with a reflexive marker (RM), the second of a so-called passive participle marked by the derivational suffix -n- oder -t- (be­tween stern and ending) together with an (often optional) copula or auxiliary. The former will henceforth be called the 'RM-passive', the latter the 'analytic' or 'periphrastic passive'. The Baltic languages possess only a periphrastic pas­sive, as does modem Polish. The Baltic participles have the suffix -t- (but not

-n'- as in Slavic; see below) to mark anteriority (traditionally named "preteritel perfect passive participle") and the suffix -m- to mark simultaneity ("present passive participles"). These are etymologically closely related to their Slavic equivalents.9

In East and West Slavic, the nlt-participle, which is used for the peri­phrastic passive, shows agreement with the subject (= PSA) of the so-called personal passive in number (singular or plural), gender (masculine, femi­nine or neuter), and case (nominative). 10 The same holds for the Baltic t- and m-participles. Furthermore, all the languages which we are concemed with here use a BE-copula (Russ. byt', Pol. byc, Lith. buti, Lat. but), II but a! least' nowadays they differ with respect to whether they use a second copula of the BECOME-type Oike Germ. werden). Because the lexical meaning of such copu­lae or auxiliaries denotes a transition into a new state I refer to them as 'incho­ative copula/auxiliary'.

With respect to the RM-passive, East and West Slavic languages differ with regard to the morphological status of the RM itself: in West Slavic it is an en­clitic (Pol. sit:,., Cz. se, Slov. sa), in East Slavic an unmovable postfix -sjal

- 's, 12 i.e. an affix added to the stern and its inflectional and/or derivativational (

suffixes (therefore called "extra-inflectional affix" by Haspelmath 1990: 29). Both enclitic and postfix derive from the same etymological element *st:,.lsebe,

Passives as grammatical constructions 277

the accusative/genitive-form of the reflexive pronoun. In Common Slavic the shorter form * st:. could be used either as a ditic or a stressable pronoun, and as such it was still attested in Old Church Slavonic (OCS) and early East Slavic. Therefore, from this angle, the East Slavic languages present a more advanced stage in the morphologization of the original pronoun. This fact will be re­sumed later (see 3.1).

With respect to oblique agent phrases, it is quite weH known that the Latvian passive does not allow the use of such a phrase, at least not without some specific conditions to be specified in 3.4.2 (for this reason I have put a line in brackets into the respective rubric of Table 1). Slovak could weH be considered another such case; at least agent phrases are avoided (cf. Paulinyet al. 1968: 291) and, ifused, highly marked stylistically. In fact, such an avoid­ance is not really a grammatical restriction, but rather a matter of pragmatic (in)appropriateness, similarly, as it seems, to oblique agent NPs in Finnish or Turkish (cf. Keenan 1985: 249). All the other (contemporary) languages of this survey do more or less readily permit agent phrases to be added, either in the instrumental (standard Russian, Belarusian, Ukrainian, Czech) or in the geni­tive (Lithuanian) or in different kinds of PPs (Pol. przez + ACC. 'through', Cz. od + GEN. 'from', probably also u + GEN. 'at, by' in some NW-Russian dia­lects; see 3.4.1). These basic facts are set out in Table 1. Compare one standard example for each construction type:

(I) a. Russüm Medved' byl ubit (oxotnikom) bear.NoM.SG.M COP.PST.3.SG.NOM.M killed.NoM.sG.M hunter.INs

b. Czech Medved byl zabit (mysliveem) bear.NOM.SG.M COP.PST.3.SG.NOM.M killed.NoM.SG.M hunter.INS

c. Polish Niediwiedi zostallbyl zabity (przez bear.NoM.sG.M cOP.psT.3.sG.NOM.M killed.NoM.sG.M through mysliwegp) hunter.Acc

d. Lithuanian Lokys buvo uZmustas (medziotojo) bear.NoM.sG.M COP.PST.3 killed.NoM.SG.M hunter. INS

e. Latvian ~

Laeis tikalbija nosists bear.NOM.SG.M COP.PST.3 killed.NoM.SG.M 'The bear was killed (by the hunter).'

:.... <:u ~

• <:u

~ E :c .~

~

00 r--­('.l

Table 1. Types of passive constructions (contemporary state)

Undergoer = Subject Actor = 0 or oblique

Language

Russian (East Slavic)

Czech, Slovak (West Slavic)

Polish (West Slavic)

Lithuanian (Baltic)

Latvian (Baltic)

"Analytic"

Participle ( + agreement)

+

+

+

+

+

Copula(e)

byt' 'be' (0 in present tense), ex. (Ia)

byt 'be', ex. (Ib)

byc 'be', zostac 'become', ex. Oc)

büti 'be', ex. Od)

büt 'be', ex. Oe)

"Reflexive" Actor phrase(s) « *se etc.)

stern + postfix, ex. (Ha) bare instrumental

stern + c1itic, ex. (Hb) bare instrumental, od 'from' + GEN.

(stern + c1itic; only przez, through' + ACC.

lexicalized remnants, see 3.1.1)

bare genitive « possessive)

(-)

Passives as grammatical constructions 279

(iI) a. V magazine prodaetsja xleb in ShOp.LOC sell.PRs.3.SG.RM bread.NoM.SG

b. V obehodu se prodava eh leb in ShOp.LOC RM seIl.PRs.3.SG bread.NOM.SG 'Bread is (being) sold in the shop.'

Russian

Czech

Since participiaI constructions constitute the predominant pattern of marking the passive in the language region discussed here they will be given particular attention. I will start the examination of the "ingredients" of passive construc­tions with the so-caIled reflexive marker.

3. Sketch ofthe evolution ofpassives in (East) Baltic and Northern Slavic .

3.1. Reflexive markers and the RM-passive

It has become a commonplace that the I.E. reflexive marker (RM), which in quite a couple of languages of this branch has. developed into a marker of the passive, etymologically derives from a free ("anaphoric") reflexive pronoun. This is also the case with Slavic * sebe/ st:., the former ACC/GEN of the pronoun, which has become a clitic or agglutinative RM. 13 Lithuanian has the RM -si­which most probably derives from the former dative * si (common with SIavic). Latvian dialects show a variation of RM-forms (basically -si- and -sa-), and their relation to Lith. -si- has not entirely been clarified. The same holds for Old Prussian -si. 14 However, for the discussion at hand it is of hardly any rele­vance from wh ich case form of the reflexive pronoun the RM in Baltic and Slavic happened to derive etymologically. Regardless of which case it was, the respective form must have lost both its place in the case paradigm of the pronoun and its referentiality so that it became appropriate as a clitic, which marked valency changes and other, often subtle, changes in the lexical struc­ture of the verb sterns with which it occurred (see below). Furthermore, loss of paradigm and referentiality must have taken place so early in pre-documented times that it will probably remain an unsolvable problem to determine the ex­act mechanism and communicative triggers of this process. We can only rea­sonably assume that the onset of this process coincided with the reflexive mor­pheme marking a part-whole or some similar relationship between the "doer" of the denoted action and the "object" which s/he acted upon. In Slavic and Baltic this process cannot be illustrated by the help of attested data, but it may be reconstructed with confidence by the aid of comparable historical evidence

280 Bjöm Wiemer

from other languages as, for instance, from French (cf. the re~ent analysis of pertinent data conducted by Waltereit 2000). - Notwithstanding ·the difficulties in reconstructing the pre-documented stages, a broader typological basis of data proves useful in understanding what plausibly must have gone on before historieal documents began to appear. It is widely acknowledged that in many languages the RM is an extraordinar­ily polyfunctional morpheme, whose functions may be arranged on a seman­tic map showing the contiguity relations between the proper reflexive func­tion on one end and the passive (and passive-like) functions on the other end (see Figure 2). Such a map of valency-changing functions of morphemes was first proposed by Haspelmath (1987). The functional spread ofthe RM is to a large degree comparable to the polyfunctionality of ancient I.E. markers of the so-called 'medium' (-ya- in Sanskrit, -n- + ablaut in Gothic, -men- in Ancient Greek etc.); 15 by and large they were renovated by an expanding RM.

As early as the appearance ofthe first preserved Slavic documents (approx. 1,000 years ago) the RM already showed a "functional inventory" which is, leaving aside some details, identical to its distribution today (see below). Also phonetic erosion and morphosyntactic coalescence had already proceded to a considerable degree: in the earliest authentie (i.e. not translated) East Slavic texts we encounter s~ as a morpheme whose syntagmatic variability is more restricted than in those texts which were translated from OCS. In genuinely East Slavic texts s~ could no longer be shifted around and almost always oc­curred immediately after the verb (Gunnarsson 1935: 102). This meant just one step from agglutination, which was gradually completed by the four­teenth century.

Now, the crucial point is that this kind of phenomena, which are weIl known from grammaticalization studies, was not determined by the passive function of the RM as such. To realize this, we first have to consider that the passive function is only a very late stage in the expansion of the RM through the semantic space of valency changing operations. This slow "mov~~ent" can be illustrated by Figure 2 (= Fig. 17 in Haspelmath 2003). Therefore, be­fore the RM "arrived" at the passive function, it must have been applied to mark valency changes which brought about changes in the argument structure of the deriving verbs (sterns). In this respect, these changes were derivation al in the traditional sense, by which new lexical items arise. Consider, for in-

Emphatic Full fl · -7 flO re eXlve re eXIve

Grooming -7 body motion

Anti--7 .

Potential P ° -7 . -7 aSSIve

causative passIve

Figure 2. Semantically contiguous functions of reflexive pronoun and RM

Passives as grammatical constructions 281

stance, the change from causative to partitive object (Russ. cesat' 'to comb' => . cesat'sja 'to comb one's (own) hair'), from causative to autocausative (pod­njat' 'to lift, raise' => podnjat'sja 'to rise, stand up'), from causative to anti­causative (slomat' 'break.tr' => slomat'sja 'break.intr'), from causative to re­flexive-causative (stric' 'to cut hair' => stric'sja 'to have S.O.k cut ont1's hair') or to "modal-passive" (e.g. stirat' => stirat'sja in Eti noski legko stirajutsja 'These socks wash easily', with an additional alethic meaning component, i.e. a modal component indicating the objective possibility of the realization of the respective action denoted by the stern). 16 Only the RM in the passive function (both impersonal and personal) really does not change the lexical meaning of the deriving verb. It, therefore, does not enhance the number of entries in the lexicon, nor does it even influence the internal organization of single lexical entries (depending of what one's theory of the lexicon looks like), but it just indieates a different" mapping from the lexieon onto the morphosyntax. 17 In this sense it is the only really grammatical (=non-Iexical) function ofthe RM.

The whole diachronie process, whieh begins with the gradualloss of refer­entiality of the reflexive pronoun and its transition to a RM marking valency changes, and which ends up by marking just are-arrangement of mapping rules from the lexicon to the syntax (as in Fig. 1), can be represented graphic­ally as in Figure 3. The RM doing its services as a passive marker is an instance of what I would like to subsurne under 'grammatical derivation', because the morphotactie rules deriving aRM-passive from a transitive verb are just the same as with other RM-forms derived from other Russian verbs: they al­ways come between the deriving stern and inflectional affix(es). In the passive function the RM is just more productive and less restricted by the lexieal con­tent than it is, for instance, in the formation of anticausaiives, autocausatives, so-called "modal passives", etc. (see above). For a more general justification of 'grammatical derivation' see V. Lehmann (this volume). What is at issue here is not another attempt at demarcating inflection and derivation, but an at­tempt to treat grammaticalization and grammatieal categories (paradigms) as

Lexical derivation (change ofthe verb's lexical structure)

Reflexive proper 4=== ~ Passive 7 ~

grammatical derivation (preservation of argument structure, in particular of the number of argument positions)

Figure 3. From lexical valency changes to grammatical derivation (RM-passive)

282 Björn Wiemer

phenomena wh ich themselves do not depend on the question whether some meaningful distinctions in a language or processes relevant to its syntax are ex­-pressed by inftectional endings, or rather by other morphological operations. 18

3.1.1. The rise of categorial restrietions

Look now an exampie illustrating the passive function of s~ in Old East Slavic: 19

(2) loza, ne tvorjasci ploda 0

branch.NoM.SG.F NEG create.PTC.NOM.SG.F fmit.GEN .SG.M about mne, posekaetsja, a tvorjascia plod me.LOC chop off.3.SG.PRS.RM and create.PTC.SG.F.DET fmit.Acc.SG.M otrebitsja clean.3.SG.PRS.RM (Feodosij Pecerskij, 11 th c.) 'the branch which does not yi~Id me a fmit, wiU be chopped ofT, but that one which does bear a fmit, will be c1eaned '.

Many of the examples cited in the literature of the topic, however, are not un­equivocal, insofar as the passive is not the onIy possible interpretation which can be given to the verb with s~; cf., for instance (from Dankov 1981: 88):

(3) i krestisja vö Erdane otö Ioana and baptize.3.sG.AOR.RM in Jordan.Loc from John.GEN.SG (Lavrent'evskaja letopis', 12th c.) 1. 'and he was baptized in the Jordan by John' ~ passive

11. 'and he made himselfbe baptized by John in the Jordan' ~ reftexive­causative.

The passive function quite often intersected especially with the autocausative, the anticausative and the reftexive-causative. In fact, such ambiguities are en­countered until today, but in Russian some subtle, but important changes in the overall distribution of RM-verbs (or verb-forms) have occurred, wh ich have made the system of choices for marking the passive more restrictive. The most relevant one for our topic is the fact that RM-verbs with perfective (pf.) value20 can no longer be interpreted as passive, Le. the RM-passive has been . restricted to imperfective (ipf.) verbs. Cf. otrebitsja in (2) and the following °example from a Novgorod text (cited after Strel'cova 1978: 217):

Passives as grammatical constructions 283

(4) asce ... da oslepitbsja i nos da urezetb if that blind.3.sG.PRs.RM and nose.ACC.SG and cut off.3.SG.PRS emu him.DAT 'and when ... that he beblinded and one cut off his nose'.

The verb oslepiti 'to blind' is perfective. Nonetheless, this instance can unani­mously be interpreted as passive (probably nobody, leaving aside masochists, orders other people to cripple oneself, so (4) will not be interpreted as a reflex­ive-causative). In early East Slavic we can encounter examples in which the RM-passive and the periphrastic passive of verbs of either aspect are used one after another. Compare an example with an imperfective participie in a peri­phrastic construction taken from the "zitie Feodora Studita", a text more obvi­ously influenced by OCS (cited from Silina 1995: 501):

(5) i vbstavb nacatb peti psaltyrb dvdvu. i tu abie mnogyi trQsö ne slysimö byvaase. ta ce po mItve sedbsu emu se paky bescislbnyixö besovö gla(s) slysaase sja 'and as he got up, he beg an to chant the David-psalm and immediately the heavy tremor was not to be heard anymore, but as he sat down af­ter the prayer, the voice of untold demons was to be heard afresh'.

In such instances the passive function is obliterated by an alethic meaning (see above). But we can encounter comparable examples with n/t-participles and RM fonns from perfective verbs both betraying genuine passive function in one sentence even in sixteenth-century Polish:

(6) to nam przez cne kapellany 'thiS.ACC US.DA T through Upright.ACC.P~.M priest.Acc.PL.M oznajmilo Sl( I objawiono jest (1533) declared.3.pRT.N RM and proclaim.PPP.NOM.SG.N COP.PRS.3.SG lit. 'this declared itself to us by/through upright priests and is pro­claimed', i.e. 'it was proclaimed to us by upright priests'.

The retreat of the RM-passive from perfective verbs was a slow and gradual process, and it is impossible to pin down exacdy the time when it eventually died out. Today it has indeed become obsolete, although there are relics of it in modem (at least colloquial) Russian; cf. Vas tekst procitalsja 'Your text was read (through)', or more precisely: 'Your text could be read'. I douqt, how-

284 Björn Wiemer

ever, whether these relics are really to be treated as passives "pure and sim­ple": they usually bear an additional air of alethic modality (see my transla­tion). Furthermore, they do not allow an agent phrase (Vas tekst procitalsja

*nasim inzenerom 'Your text was read by our engineer'), as would be possible otherwise in the passive of modem Russian. As for diachrony, we may say that the RM -passive of perfective verbs was ousted at the latest in the course of the nineteenth century.21

All this shows that polyfunctionality of RM-forms of perfective sterns has decreased; RM-forms of imperfective sterns, however, have not. The latter ones, in turn, have been restricted to the RM-passive in East Slavic and ceased to form the analytic passive. Compare examples from early centuries such as (7). °

(7) Igumenö Stägo Andrlja Grigorii, ljubimö bo bl preze Volodimeromö, elen ze oto Vbstislava i oto vsixö ljudei (Ipat'evskaja letopis', t. 11-1, 287; 12th c.) 'because the abbot of Saint Andrej, Grigorij, was formerly revered by Volodimer, respected by Vhstislav and by all people'.

In Russian analytic passives of imperfective verbs were in use approximately until the seventeenth century, after that they began to gradually disappear from texts of differents genres (cf. Lomtev 1956: 199). Lexicalized relics have been preserved practically only in attributive use (e.g. nepisanye zakony 'unwrit­ten laws'). In the long ron, these two processes of functional decrease con­verged, so that two competitive types of passive constructions were reorgan­ized in a complementary fashion. This process is wholly understandable only in connection with the evolution of the East Slavic (and Polish) aspect system, which, however, I cannot dweIl on here.

A similar process of complementary distribution resting on the strength­ening of the aspect opposition took place in Polish, too. But it did not involve the RM-passive, which was altogether abandoned somewhere in °the second half of the nineteenth century (cf. Szlifersztejnowa 1968: 156). Sometimes examples like Wyklad si~ pisze ('The lecture is being written') are mentioned as attesting an RM -passive in modem Polish. In fact, such cases are at best remnants of an earlier pattern, they do not allow for an agent phrase (as would be usual in Polish), they are lexically highly restricted (to more or less a dozen verbs), and they cannot be used with subjects naming persons capable of act­ing on their own.22 They might therefore qualify as phraseologized units (lex­lcalizations) .

Passives as grammatical constructions 285

RM-forms of verbs with a passive function (with an agent phrase!) could still be encountered in the first half of the nineteenth century; for instance:

(8) Nakazane przez kr61a projekta order.PTc.NOM.PL through king. ACC.SG.M project.NoM.PL dopelniefL (. .. ) gotowaly sit: przez supplement.GEN .PL prepare.3 .PL.NON -VIR RM through wyznaczone do tego osoby (1807) apppoint.PTc.AcC.PL to this.GEN .SG person.Acc.PL lit. 'The supplementary projects ordered by the King prepared them­selves by the people appointed to this (task)', i.e. 'The supplementary projects ordered by the King were prepared by the people appointed to this task'.

In Polish only an analytic passive survived, but the participles are now sub­ject to ~e same complementary functional distribution over two opposed as­pects (perfective-imperfective) as are finite verb forms (in the active). That is, in modem standard Polish only nlt-participles of imperfective verbs, to the exclusion of nlt-participles of perfective verbs, are capable of indicating iter­ativity and progressivity (duration without limits), whereas, in turn, nlt-parti­ciples ofperfective verbs are the only means to refer to post-states (i.e. resulta­tives) or to singular time-Iocated events in the passive.23

In fact, contemporary Polish is the only Slavic standard language in which nlt-participles show such a consequent and clear-cut distribution and in which the RM-passive has been oustedtotally. From this point of view, Czech and Slovak have remained less straightforward: on the one hand, they allow both imperfective and perfective verbs to form aRM-passive, although the RM­passive of perfective verbs is much rarer and tends to show the same restric­tions as have just been mentioned for Polish remnants of the RM-passive. On the other hand, nlt-participles ofboth aspects can form an analytic passive, but they are not so neatly distributed over aspect functions as they are in Polish.24

Czech and Slovak show the same potential ambiguities with RM-forms which are characteristic for Russian (see above). They restrict them mainly to the third person, where it is less likely that RM-forms will be interpreted not as passives, but as auto- or anticausatives.25 Even more important, periphrastic and RM-passive do not so much complement, but partially "duplicate" each other, insofar as nlt-participles of imperfective verbs and finite forms of im­perfective verbs supplied with the RM often fulfill the same aspectual func­tions (besides marking voice). Compare the following two examples illustrat­ing the habitual present in the "high" variety of Czech (spisovnd cestina):

286 Bjöm Wiemer

(9) a. Vyznamne knihy se prekllidaji do famous.F.NOM.PL book.F.NOM.PL RM translate.PRs.3.PL into cestiny Czech.F.GEN .SG 'Famous books are (usually) translated into Czech'

b. Tyto prtice jsou v posledn{ DEM.F.NOM.PL work.F.NOM.PL COP.PRS.3.PL in last.F.Loc.SG dobl prekltidtiny do cestiny period.F.LOC.SG translate.PPP.F.NoM.SG into Czech.F.GEN.SG 'Recently these (written) works are (= have been being) translated into Czech '.

We may assume that this reflects a more ancient stage, which held also for Russian and Polish at least until the eighteenth century (see above and Section 3.2). This stage should be considered less grammaticalized than the Polish passive (for the reasons laid out above), but also less grammaticalized than the Russian passive, because Czech still lacks a clear-cut division ofRM-passive vs. periphrastic passive and imperfective vs. perfective verbs, which has estab­lished itself in Russian.

3.2. Participles: orientation of diathesis and aspectual shifts

3.2.1. Resultative and relative participles

The principal "ingredient" of analytic passives are participles. We will, there­fore, have to account for the rise of so-called passive participles and their re­lation to resultative constructions. Resultative partieiples are "verb forms that express astate implying a previous event" (Nedjalkov and Jaxontov 1988: 6).

Haspelmath 0994: 153), among others, showed that lack of inhere~t dia­thetic orientation is characteristic for many deverbal adjectives in I.E. as well as in non-I.E. languages. By this the following is meant. In many languages participles began as deverbal adjectives whose diathetic relationship to the de­riving verbs was not a stable or unified one. That is, often the argument struc­tures of deverbal adjectives are related to the argument structures of the verbs in a non-uniform way. Deverbal derivatives with one "and the same formative can in one case agree with an argument which is identical with a lower-rank­ing argument of the deriving verbs. Inasmuch as this lower-ranking argument can normally be identified as the Undergoer, such derivatives can be called 'Undergoer-oriented' (cf. Engl. to build a house ~ the built house, builtrefers

Passives as grammatical constructions 287

to, and agrees with, the patient, i.e. a typical Undergoer). In other cases, how­ever, their argument coincided with the highest-ranking argument ofthe deriv­ing verb. With two- and three-place verbs this is necessarily the Actor, so that participles taking this perspective can in general be called 'Actor-oriented'; for instance, Germ. schwören 'to swear' =:} die Geschwore-nen 'jurors' or be­dienen 'to serve, operate' =:} der Bedien-te 'employee' as lexicalized remnants from Middle High German (last example by courtesy of Petra Vogel, p.C.).26 In other cases (e.g. with Lezgian deverbal adjectives) their argument does not correspond to any argument of the deriving verb, but to an adjunct. In general, deverbal nouns that are not fixed with regard to macroroles can be called 'rela­tive participles'; cf. Haspelmath (1994: 154) for the term and examples.

Beside relative participles, some affixes may also become highly produc­tive devices of adjective-formation; cf., for instance, English adjectives with the suffix -able (manage-able, agree-able, etc.) or German adjectives with the suffix -bar (annehm-bar 'acceptable', veräußer-bar 'alienable; which can be sold' etc.). That the bulk oflinguists does not indude these forms into the para­digm of verbs, is first and foremost to be explained from the fact that these af­fixes add some semantic component to the meaning of the stern of the deriving verb. As a rule, this is, again, a modal (alethic, deontic) component. In other words, deverbal adjectives do not become true participles as long as they mod­ify the lexical meaning of the deriving verbs. They remain adjectives precisely because the respective affixes create new lexical units and do not just adapt the lexical stern to the syntactic surroundings. From this we see that productivity of affixes alone does not distinguish true participies from deverbal adjectives. Again, a crucial point with regard to grammaticalization proves to be that the lexical content of verb sterns affixed by some productive derivational mor­pheme be preserved. This is the same criterion as with RM-derivatives dis­cussed in 3.1.

From this point of view, resultative participles are lexically quite dose to deverbal adjectives: they shift the focus of lexical meaning from the event de­noted by the deriving verb to a correlating resultant state;27 cf.

(10) a. The Greeks destroyed Troy -7 event =:}

b. (After the war with the Greeks) Troy was destroyed -7 resultative state.

This shift affects the inherent aspectual properties of the verb stern. More so, re­sultative participles typically have other valency properties than their deriving verbs: they are predominantly formed from transitive stems:28 the verb's Ac­tor is decentered lexically, whereas the referent of the Undergoer becomes fo-

288 Björn Wiemer

cussed as the entity for which the state obtains (' object -oriented' resultatives in the tenninology of Nedjalkov 1988). This change concerns the diathesis orien­tation. Now, in order·to become true 'passive participles', the respective deriv­atives (or: the affixes associated with them) must show a stable orientation to­wards (and agreementwith) the Undergoer ofthe deriving verbs, irrespective ofthe particular semantic role ('patient', 'theme', 'stimulus', or however they may be named) and regardless of whether these verbs are telic or not. 29

The essence of what we have said so far is the following. There are two kinds of shifts which cause a lexical dissociation of the resultative participle from the verb: there is always a shift of lexically inherent aspectual properties, and typ­ically (i.e. with two- and three-place transitive verbs) there is also a lexical shift from Actor to Undergoer orientation. For this reason resultative participles are no strict part of the paradigm of those verbs from which they are derived. Now, in order for the respective participial constructions "again" to become mem­bers of the paradigm forms of verbs with identicallexical content, but with a re-arranged Actor-Undergoer-mapping into the syntax, they must, first, be ca­pable of focussing on the same event as the deriving verbs30 and, second, show a stable orientation towards the Undergoer (in opposition to the active), at least in predicative use. In other words: the participle must by itself signal a syntac­tic (not a semantic!) deranking of the lexically highest-ranking argument (= Actor), and the whole construction must not differ semantically from the mor­phologically unmarked ( = active) one. This inc1udes also its aspectual value.

Let us now see which line of development can be detected or reconstructed for the (pre)history of relevant Slavic and Baltic participies.

3.2.2. The historically attested starting-point:jixing Undergoer-orientation

Brugmann 0895: 119) already noted that object-oriented t-participles con­stituted a c1ear majority in documented stages of ancient I.E. languages. One . could perhaps say that the t-suffix was oriented simply towards the lowesi~rank­ing argument of the deriving verb, so that with intransitives automatically the single argument, coded as subject triggering agreement, was focussed. But nonetheless we can find several examples, where even with two-place verbs the t-participle focussed on the higher-, not the lower-ranking argument, i.e. where it was Actor-oriented (see 3.2.3 for an analogous situation with the Bal­tic m-participle). Compare, for example, Latin parta which could mean either '(the woman) who bore (a child)' or '(the woman) who was born', scitus either '(the one) who knows, is experienced in' or '(the one) known (to people)', or the Old Icelandic wiss meaning either 'knowledgeable' or '(the one) known'

Passives as grammatical constructions 289

with the etymologically related Gothic un-veis, which meant only '(the one) who does not know, ignorant' (Brugmann 1895: 118). This variability in inher­ent orientation can, however, be explained by the degree to which the referent of the higher-ranking argument (= Actor) is hirnself affected by the denoted action (Haspelmath 1994: 161).

With respect to grammaticalization it is an important question when and under which circumstances nlt- and m-participles became 'oriented' with re­spect to macroroles. Notice that this is not a question of their morphology, but . of their syntactic behavior based on lexical content. This process was surely a gradual one lasting over centuries, because macrorole orientation of parti­ciples could not have taken place at once with all verbs. For the time being, it is impossible to give a convincing answer to the question from which seman­tic class of verbs the macrorole orientation in these participles started. Any­way, after comparing the facts from some ancient I.E. languages, from early and contemporary Slavic and from the genetically closely related Lithuanian, we are forced to conclude that such a process must have occurred, but its on set has to be located in pre-documented times. It obviously was not finished when the first documents appeared, although it clearly was on its way to completion in the earliest documented stages of I.E. languages, among them Slavic and Baltic languages (for the diachronic Baltic data cf. Ambrazas 2001). Traits of its incompleteness can be detected in early documented stages of Slavic lan­guages; cf., ror instance, Old Polish nie-za-plac-o-n-y (<== zaplacic 'to pay' + negation, with the n-suffix) '(the one) who did not pay', and not '(the one) who was not paid'; or modem Pol. nie-wid-o-m-y (<== widziec 'to see'), meaning 'blind (= who does not see)', not '(the one) who is not seen'. From Old Czech compare za-pom-e-nu-t-y 'forgetful' (not 'forgotten') (<== zapomenuti 'to for­get') or ne-po-hnu-t-y (<== hnuti 'to bend') with the modal meaning 'which can­not be bent' (not 'which is not bent').31 See also Russ. so-pro-vozd-a-e-m-yj (bagai) (<== soprovoidat' 'to accompany') 'accompanying, attendant [not 'ac­companied'] Ouggage)', used, for instance, in contemporary Russian customs declarations.

I am unaware of any Old East Slavic examples in which a nlt-participle (or a m-participle) of a transitive verb (with the object in the accusative) that did not show Undergoer-orientation would have been used predicatively. Notice that all relevant examples given above and below are instances of attributive or substantivized use. What we, however, do find, though rarely, are predicative Actor-oriented nlt-participles from two-place verbs whose lower-ranking ar­gument is marked by a PP. Compare the following example from the Russkaja pravda, a widely known genuinely East Slavic lawbook of the eleventh cen­tury (cited after Silina 1995: 502):

290 Bjöm Wiemer

(I1) a za Zerebecbj. oze ne vösedano na nÖj. and for stallion.Acc.sG REL.M.SG NEG sit on.PTC.NOM.SG.N on hirn to grvna· kunö CONPTC grivna kuna.GEN .PL lit. 'and for a stallion j that has not been sat on itj , (it is) grivna kun', i.e. 'for a stallion that has not been sat on (by anbody) a grivna kun [= a half of a silver-taler] has to be paid'.

The pp na nb 'on hirn' represents an argument of the verb vösesti 'to mount (a horse)'; in (I 1) it is coded in the same way as it would be in the active. The par­ticiple vösedano (<= vösesti) is in the neuter gender, because there is nothing to agree with. The reason for lack of agreement is that the participie focusses on the Actor (here only implied as a generic one).

By and large, we may be confident that n/t-participles of two-place verbs whose Iower-ranking argument was coded by the prepositionless accusative, object-orientation already was the predominant pattern for n/t-participles at

. the time when the first written documents in SIavic appeared. To this extent, passive constructions already existed at that time as a result of more or Iess fixed Undergoer-orientation, if the participle was derived from a transitive verb.

3.2.3. The fixation of Undergoer-orientation with non-resultative participIes on-m-

Contrary to n/t-participles, deverbal adjectives with the suffix -m- have never been resultative. In grammars and textbooks they are called 'present passive participles', which indicates that their temporal value is not an anterior, but a simultaneous one (with respect to some reference interval). However, similar to n/t-participles, in Old Slavic their orientation towards one of the two macro­roies (Actor, Undergoer) was not yet totally stable (predictable). For Old East SIavic compare

(12) b' emyi odolevaetö bijuscumu. beat.PTc/PASS.NOM.SG.M overcome.3.PRs beat.PTc/ACT.DAT.SG.M umirajai umarjaemu(j)umu die.PTc/ ACT .NOM.SG.M.DET kill.PTc/ ACT .DA T .SG.M 'the one who is beaten overcomes the one who beats, the dying the kill­ing' (Uspenskij sbomik, 284a; cited from Krys'ko 1997: 370).

Passives as grammatical constructions 291

Umarjati 'to kill' is the causative of umirati 'to die'. Nonetheless its m-parti­ciple in (12) does not mean '(the one) who is being killed', but '(the one) who was killing'. Examples ofthis sort are quite numerous in Old East Slavic texts, not only in translations from Ancient Greek. It could be objected that the m-par­ticiple just served to substitute for the Greek medium (non-existent in Slavic). This .certainly was one of the factors which could explain the considerable fre­quency of m-participles in the Old East Slavic texts that have come down to uso However, m-participles need not be treated as mere calques from (or rather, bad translation equivalents for) the Greek, be it only for the reason that the Slavic text did by far not always follow slavishly the Greek original.3Z After all, the etymologically closely related Lithuanian m-participles often betray the same kind oflack of diathesis orientation (see below), and here we cannot suspect any influence from Greek or oes, but a productive device rooted in rural dialects. They have remained vivid until today in Lithuanian; in the Slavic languages they were on their decline already at the time of the earliest documents. 33

In Lithuanian and Latvian the situation differs from the Slavic one first of all for the reason that these languages have never possessed a RM-passive,34 despite the fact that (especially in Lithuanian) the RM is highly polyfunctional (cf. Geniusiene 1987). Furthermore, since the first texts written in Lithuanian and Latvian have come down to us from the sixteenth and seventeenth centur­ies (and the textual basis ofOld Prussian is too limited forreliable conclusions), not much can be said about the diachronic development of t- and m-participles on the basis of empirical evidence. As regards the passive not much seems to have happened since that time. Anyway, thanks to comparative methods we can deduce that the historical situation continued the situation prior to docu­mented times (cf. Ambrazas 1990: 57-63, 90). As for Lithuanian, its system of participles has remained archaic insofar as it retained a productive use of m-participles, whereas Latvian has abandoned at least their predicative use al­together(cf. Eiche 1983; Holvoet 1995: 173).

Leaving aside the elimination of predicative m-participles in Latvian, the thing we should draw attention to is the following: Lithuanian m- and t-par­ticiples of two-place transitive verbs have been unequivocally Undergoer­oriented only if they are used predicatively as part of a passive construction (see (13)-(14)), but not necessarily so in attributive use.

-(13) Suzeistieji buvo gydo-m-i

wounded.NoM.PL.M.DET COP.3.PRT heal.PTc.NOM.PL.M felceril{ medical attendant.GEN .PL 'The wounded were (being) healed by medical attendants'.

292 Bjöm Wiemer

(14) Medis buvo perpjau-t-as (su) tree.NOM.SG.M COP.3.PRTcut through.NOM.SG.M with skerspjükliu . cross-cut saW.INS.SG 'The tree was cut through with a cross-cut saw'.

If m-participles (still in some dialects also t-participles) are used attributively, their diathesis orientation is often still different and variable. Usually they are oriented towards the single argument of either a two-place verb whose Under­goer has l>een le,xically suppressed {e.g. kulia-m-o-ji masina 'threshing ma­chine' ~ kulti 'to thresh' or gydo-m-ie-ji vaistai 'herbs of healing' ~ gydyti 'to heal, cure'; compare the laUer with Undergoer-oriented gydo-mi in (13» or to the Undergoer of a one-place verb (e.g. esa-m-as-islbü-t-as-is laikas 'presentlpast tense' <= büti 'to be'). They may, however, also focus on an ad­junct of the deriving verb (e.g. miega-m-as-is kambarys 'sleeping room' <= miegoti 'to sleep') or on the Undergoer of a two-place verb (just as in the passive); e.g. kramto-m-o-ji guma 'chewing gum' (~kramtyti 'to chew').35 This variable orientation is characteristic of m-participles serving as a kind of word-fonnation device.36 They are still a productive means for coining deno­tations, for example in diverse professional branches (see the examples just given).

From these observations we are justified to draw the following conclu­sions. There must once have occurred a split in diathesis orientation due to which m-participles are Undergoer-oriented in predicative use, but continue to show variable orientation (towards either Actor or U ndergoer or none of them) in attributive (or substantivized) use. Since these participles became strictly Undergoer-oriented only in aconstruction in which they constitute the nucleus of the clause, it seems reasonable to assume that such a split was supported by this very construction centered around m-participles. The same is true of the nlt-participles. Therefore, the evolution of the analytic passive in (Northem) Slavic and (East) Baltic was to a large degree conditioned by changes in the ap­plicability of the nlt- and m-suffixes to verbs with different linkings of argu­ment structures to the syntax (including case marking). Consequently, in all these languages an analytical passive with an agreeing NP (nominative sub­ject) is possible only with nlt- or m-derivatives of transitive verbs which in the active code their lowest-ranking argumentas a typical Undergoer, i.e. in the prepositionless accusative.37 It is impossible to prove which one of these two processes - the participles being analysed as the nucleus of the clause, or the restriction to Undergoer-orientation in predicative use - preceded (and was the reason for) the other. Probably they conditioned each other mutually.

Passives as grammatical constructions 293

The "net result" of these processes is a much more dear-cut correlation be­tween syntactic distribution and diathesis orientation. m-participles have been driven out of predicative use in Latvian and subsequently ceased to serve as a passive-fonning device. In North Slavic they ceased to be productively used (ifthey had ever been) a long time aga (see n. 33). To the contrary, Lithuanian m-participles have maintained their productivity, but with transitive verbs (i.e. with verbs having accusative objects in the active) they are now Undergo­er-oriented ifused predicatively, so that they can function ~s unanimous mark­ers of the passive. In attributive use they still show distinct traits of the ancient stage of unoriented participles.

3.2.4. The aspectual shiftfrom state to event

As concems the aspectual shift from the resultative state to a correlated event, available historical data show that Undergoer-oriented participles in Baltic and Slavic in early documented times actually were used both in resultatives ("statal passive") and as passives proper focussing on the same event as the respective active verb. A temporal adjunct or so me other context conditions could quite easily change the resultative default of such participles into an event reading. For the event use compare, for instance, (15-16) on Old Czech (cited from Sticha 1985: 80) and (17) on thirteenth-century EastSlavic from the N ovgorod region (cited after Feoktistova 1961: 201):

(15) Druhd [ces!], ze vsichni veznove, second.NoM.SG.F honor.NoM.SG.F CON alI.NoM.PL captives.NoM.PL jenz byli svazanz, byli rozvtiztini , a REL COP.NOM.PL tie.PPP.NOM.PL COP.NOM.PL untie.PPP.NOM.PL and jeho ncisledovali he.Acc follow.3.pL.PRT '(lt was) another honor that all prisoners who had been tied up were untied and followed hirn'.

(16) Vedle oltare... tu byl nalezen next to altar.GEN.SG.M here COP.3.SG.PRT.M find.PPP.NOM.SG.M kriz bozi, jenz byl crOSS.NOM.SG.M GOd.ADJ.NOM.SG.M REL cOP.3.SG.PRT.M skryt od Zid6v hide.PPP.NOM.SG.M by/before (?) JeW.GEN.PL 'Next to the altar ... here a cross (ofGod) was found, which had been hidden by/before(?) the Jews'.

294 Bjöm Wiemer

(17) v leto 6715 sveriena cerkovb in year 6715 finish.ppp.NoM.SG.F church.NoM.SG.F svjatajapjatniea (dated 1207) holy friday.NoM.SG.F 'In the year 6715 the church (ofthe) Holy Friday was nit. 'is'] finished'.

Although we do not dispose of good enough data for Old Polish (compar­able to Czech), there is . no reason to assume that the Old Polish n/t-partici­pIes behaved in any other way than in early Czech (and, to this extent, in East Slavic). East Slavic and the two Baltic languages have retained the potential aspectual ambiguity (event vs. state) in this type of constructions of n/t-parti­ciples till today. 38 Th~s means that passive constructions existed at least from the time when the ambiguity with n/t-participles illustrated above arose, al­though their "passive character" depended (and still depends) to a large degree on context-conditions (see above). In the course of the history of Polish the event-state-ambiguity of the construction was restricted and eventually aban­doned; it was overcome considerably later with the establishment of a special inchoative auxiliary (see 3.3.1).

Notice further that this ambiguity could arise only with participles from highly telic and non-punctual verbs. With punctual verbs like Cz. nalezt 'to find', rendering the participie nalezen 'found' as in (16), onIy an event reading w:as possible. Participles from punctual verbs by virtue of their lexical mean­ing cannot refer to resultative states. Thus, if they were oriented towards the Undergoer (as outlined in 3.2.2), they could function only as a real passive equivalent of the active verb. This was another factor strengthening the rise of paSSIves proper.

3.2.5. Some necessary additional remarks

This remark leads us to another, more general point. One of the central consid­erations in the preceding subsections has been that resultative participles with Undergoer-orientation have become the cornerstone of periphrastic passives in Northern Slavic and, beside non-resultative m-participles, in Baltic, too. I am not sure whether in the world's languages any case of development in the opposite direction (from periphrastic passive to resultative) has been attested; as far as I know, it has not. How can this apparently unidirectional evolution be explained? The usualline of argument (cf. e.g. Haspelmath 1994: 161) goes as folIows: participial constructions with resultative meaning evolve into

Passives as grammatical constructions 295

passives inter alia by virtue of a decrease of semantic restrictions on verbs whose participles can "enter" into a construction of the type

(I8) Undergoer-subject (= PSA) BE (= copula, may be optionaI) nll-participle.

I further argued that as soon as the participle no longer derivesexclusively from a telic verb, other interpretations than a resultative one becom.e possible or even necessary. If the verb deriving the participie does not denote a change of state, a "switch" from the event to the subsequent state is no longer imagi­nable and the whole construction can be interpreted just as a kind of syntactic equivalent (or, if one prefers, "transform") ofthe same lexical unit (verb) with identical aspectual properties.

This reasoning is fine for languages like the Germanic ones, where parti­ciples of the Slavic nil-type have both considerably extended their frequency and increased their applicability to verb sterns of different aspectual default value. For instance, in German such participles can be formed from virtually any transitive verb39 so that they are able to denote also (telic and atelic) pro­ces ses and non-resultative states (cf. Haspelmath 1994: 162 for examples of attributive use). Furthennore, the BE-copula has gradually become a marked device in comparison to the BECOME-copula werden, which has supplanted it as a neutral passive auxiliary.

The above way of reasoning, however, does not quite fit the data for Rus­sian (at least for standard Russian), since in the analytic passive it has been using basically no other nlt-participles than those of telic verbs (with some well-describable extensions, about which cf. Knjazev 1989: 108ff). Lexical expansion has not gone farther than to conc1usive verbs (first of all verbs de­noting speech acts and mental events, compare nalezen 'found' in (16) from Old Czech). With West Slavic, in particular with PoIish, the situation is dif­ferent. They may be compared to the Gennan case, insofar as nlt-participles are now the only means of fonning passives, and they can be formed from the absolute majority of transitive verbs and to a large extent regardless of their lexically inherent aspectual default. Actually, this means that nlt-participles can in principle refer not only to telic events and their correlated post-states, but also to processes or even to usual states (see (22)-(23) in 3.3.1). In other words: the construction type of (18) has ceased to be exclusively associated with resultatives, and the construction's aspectual value is conditioned mostly by the lexical aspectual value of the nlt-participle. In addition to this (and con­trary to German), the temporal value is also conditioned by the membership of the nlt-participle in the class of perfective vs. imperfective verbs. German

296 Bjöm Wiemer

has no grammatical aspect opposition, whereas in Polish such an opposition has evolved, and its n/t-participles have neatlybeen integrated into it (see end of3.3.1).

As for the Lithuanian analytic passive, the construction type laid down in (18) must be applied also to the non-resultative m-participles with their un­broken productivity inherited from its LE. predecessors (see 3.2.3). For them considerations about an aspectual shift are irrelevant. Roughly speaking, these participies fulfill more or less the same duties as Polish n/t-participles of im­perfective verbs and of the Russian "reflexive" passive (see Section 3.1.1). Since in the course of its history Latvian has given up the predicative use of m-participles, its passive shares features ofboth Polish and German. However, it has neither developed a grammatical aspect system (as has Polish), nor has the inchoative auxiliary tikt become a well-integrated part of the verbal para­digm (as has Pol. zostac and, even more so, Germ. werden), nor can an Actor phrase be adged (cf. 3.4.2). For these reasons its analytic passive should be considered as less grammaticalized than the passive of other languages dis­cussed here (see 3.3.2).

In general, it seems reasonable to regard the degree of lexical expansion of construction types as the one from the scheme in (18) as an important para­meter for "measuring" the degree of the grammaticalization of passives in a language. Another parameter would be the interaction of the particular con­struction with other categorial distinctions made on the verb; here we have to account for the Slavic aspect opposition, by which verbs are divided into one of the two global classes of perfective vs. imperfective verbs, each having its own set of grammatical functions (here not discussed in any detail). Notice that these criteria do not rest upon assumptions about the "morphologization" of linguistic material, but on the systematicity of functional distinctions made and onthe expansion of syntactically relevant patterns in the lexicon, under­stood as the inventory of units that serve as potential input for the construction (on this point see also the contributions by Himmelmann and V. Lehmann in

- this volume). . , Of course, we do not (and probably will never) know exactly how the

"switch" from resultative state to the correlated event came about; nor do we exactly know how "diathesis orientation in deverbal adjectives/participles took place. Detges (2000) has recently proposed a theory based on the communi­cative, speaker-based motives of such a change. As convincing as his argu­ment may be, he accounts only for the aspectual shift (from resultative state to event), for he is interested solely in the transition from resultative to per­fect. Furthermore, he elaborates only on a particular case of such a transition, namely one which justifies the view that certain resultatives are first exploited

Passives as grammatical constructions 297

as an expressive means of assuring the hearer that the speaker is bound to the action denoted by the participle because s/he (thespeaker) has made a com­mitment. But then Detges himself goes on by saying that such resultatives grammaticalize into perfects by virtue of their "communicative abuse" , which beg ins with the expansion of the rhetoric pattern on an increasing amount of verb sterns. As a consequence, the subject of the c1ause is no longer exc1u­sively interpreted as the individual who refers to an obliging speech act or to a state-of-affairs obviously caused by him-/herself, but as the agent of the ac­tion regardless of rhetoric commitment (intention); thus, a shift towards the event (e.g. a speech act) itself occurs which conventionalizes and continues spreading in the lexicon.

I basically agree with Detges's considerations, although he defers the ex­pansion of the "lexical input" into the construction (its "abuse") to a stage af­ter its grammaticalization proper (in his terms). Notice, however, that his rea­soning does not account for the diathesis orientation of the participle, nor is it helpful with regard to participles which never have been resultative (as the m-participles).40 They hold only for a resultative/perfect built on a HAVE-verb (so-called 'possessive resultative/perfect'), which then acquires auxiliary sta­tus. A further problem is that in no Baltic or Slavic language has an auxiliary of "possessive origin" been used for the purposes of an analytic PASSIVE (and hardly has there been a "HAvE-passive" inany other European language). Nor does there occur a shift from Undergoer to Actor (if it did the construction would not qualify as a passive). To the contrary, in the cases discussed by Det­ges, the Actor probably was never "ousted" from the argument structure of the verb (or the originally possessive construction). Consequen~ly, we may doubt whether his model is able to describe the cases at stake here, even if we should happen to dispose of data with a greater time-depth.

3.3. Auxiliaries (specialized copulae)

In the history of practically allianguages discussed here we can detect several verbs that, at one time or another, were on their way towards becoming pas­sive auxiliaries. In most cases these were inchoative verbs functioning as cop­ulae (as in Max became a teacher or Suddenly the valley went dark), which si­multaneoulsy functioned as existential verbs , (as in Night Jelilstarted, Then it became dark or There appearedlarose some difficulties); for an overview cf. Wiemer (1998). In a few cases, copular and existential verbs with a con­tinuative meaning also proved to be possible candidates of passive auxiliaries, such as, for instance, Lith. likti 'to remain' (= 'to continue to be'). But only in

298 Björn Wiemer

~olish and, at least to a certain degree, in Latvian, has such a copula become a standardized means of denoting just a state-changing event with the Under­goer of the participle foregrounded. 41 Such copulae can be regarded as passive auxiliaries. Although data as to their functional development are fragmentary, I will try to give a rough outline of the diachronic development of at least Pol. zostac 'to become' and to add afterwards some short amendment for Lat. tikt.

3.3.1. PoZish zostac: From continuative to inchoative meaning

Originally zostac42 was not inchoative (z 'to become'), but continuative (z 'to remain'). Please note that continuative meaning does not entail a stative or durative aspectual default. The aspectual default of zostac was right from the start of its development into an auxiliary a punctual one (see below). This must be understood in the sense that zostac at all times lexicalized the nega­tion of the inception of a new state (z 'not to change into another place/pos­ition/state'). I will return to this point below.

In the oldest documents zostac fulfilled only the function of an existen­tial-Iocational verb with PPs (as in modem Polish zostac w domu 'to remain at horne') and of a copula with nouns, adjectives and, by time, also with nlt-par­ticiples. The earliest detectable example with a participle is from the middle of the sixteenth century; cf.:

(19) Ize-s vczyniZ pnekl(ltym zostaniesz nad this-2.SG do.3.SG.PRT.M curse.PTC.INS.SG.M be(come).2.sG.PRS over wszystki zwierz~ta (Biblia Brzeska, 1563) all.Acc.PL animal.ACC.PL 'and because you did (this), you will be cursed more than all the animals'.

In this case an inchoative meaning, i.e. a meaning naming the transition into a new state, is quite obvious. Such instances, however, for centuries remained very rare. Much more frequently zostac was used with predicative adjectives; but more often than not the meaning of such predicates was either continuative ('to remain'), or it cannot be established from the available context whether zostac already gained an inchoative meaning (cf. for instance zostac glupim, bezecnym 'to ?become/?remain silly, infamous', sixteenth century). One of the problems is that early dictionaries probably did not adequately reftect real language use. For example, we may assume that zostac evolved into a passive auxiliary under the influence of Middle Low German bliven, from which Pol. zostac was calqued; at that time bliven already oscillated between a continua-

Passives as grammatical constructions 299

I BECOME:~el I state, I

t

Figure 4. Contiguity relation between continuative and inchoative meaning

tive ('to remain') and an inchoative meaning (for details cf. Weiss 1982: 205-209).43 This makes it rather unlikely that zostac did not show any inchoative function, especially with n/t-participles. We can thus conjecture that this evo­lution must have started not later than at the beginning of the sixteenth century, ,and early examples like (19) might already reftect a usage conditioned by lan­guage contact. By and large, the transition of zostac from a copula with an ex­clusively continuative meaning towards a separate inchoative meaning of an auxiliary lasted at least from the late fifteenth century till the nineteenth cen­tury, in which course zostac eventuall y became an unambiguous marker of the actional passive (::::: 'Vorgangspassiv', see n. 41), for which Middle High Ger­man gave the model; cf. Szlifersztejnowa (I968: 156, n. 131).

Let us first clarify what happened on the semantic side of the phenomenon. The conceptual relation between the negation of a change of state, which is the basic meaning of to remain and its equivalents in other languages, and the tran­sition into a new state, which is the basic meaning of to become and equivalent verbs in other languages, is based on temporal contiguity. If state2 succeeds af­ter state1, and there is an event e2 wh ich marks the transition from state1 into state2, the inchoative (= change-of-state) meaning of to become sets off the beginning of state J, whereas the continuative meaning of to remain has to be understood as the negation of e2, i.e. of a transition into state2 (see Figure 4). Event e1, in turn, marks off the point in time fromwhich state l starts. The in­choative meaning thus focusses on a telic (situation-changing) event. It may in certain cases also embrace a telic process immediately preceding this event and naturally ending by attaining it. Figure 4 is a rather informal (and coarse) way of representing the contiguity relation between inchoative and continua­tive meaning; but it suffices for our purposes.44 Now, why should a meaning shift from a negation of a change into state2 (= ..., e2) to the "left boundary" of state J (= e l ) occur? And what triggers it?

Weiss (1982: 208) suggested that a likely trigger causing a shift from con­tinuative to inchoative meaning was a contextual factor. He argues that such a switch may be explained on the basis of a semantic equivalence, which can be presented as in (20).

(20) ., BECOME + past passive participle = REMAIN + ., past passive participle.

300 Bjöm Wiemer

This equivalence basically corresponds to the middle part of Figure 4. (20) stresses the role which in the shift from continuative to inchoative meaning might have been played by negation: internal negation of the event (eI) pre­supposed by the participle (which itselffocusses on the correlated post-state; see below) must have changed to extemal negation of the same state-changing event (eI)' but already denoted by the finite verb (copula/auxiliary).

In practice, Weiss ' hypothesis implies that the respective copula underwent a switch to inchoative meaning, because it was frequently used with negated adjectives and/or n/t-participles denoting antonymic states (e.g. 'dead-alive', 'wise-silly', 'bright-dark'); cf., for instance, from modem Polish

(21) a. Zolnierz zostal niezabity soldier.NoM.sG.M become.3.SG.PRT.M NEG.kiIled.NoM.SG.M ( w bitwie )45

in battle.Loc lit. 'The soldier remained unkilled [= alive] (in the battle)' =

b. Zolnierz nie zostal zabity soldier.NoM.sG.M NEG become.3.sG.PRT.M killed.NoM.sG~M 'The soldier was not killed'.

Indeed, instances of usage corresponding to (21 a) were widespread for Mid­dIe Low German bliven. These may weIl have triggered a process by which the continuative meaning was backgrounded, whereas the inchoative mean­ing was foregrounded. UnfortunateIy, Weiss does not give any authentie ex-

. amples from Old Polish, neither could I find a single one in SIStar XI-6 (2000: 464-467), where large amounts of exampies from the oidest sources of the language are given in diverse kinds of syntactie context. Maybe, more intense in-depth studies of texts would bring some appropriate examples to light. The data hitherto available, however, suggest that examples of the sort illustrated in (21a) were by no means salient and indeed very rare. .

Notiee that the capability of zostac to denote a state-changing event is not simply a consequence of the perfective aspect, to whieh zostac beiongs. Its imperfective counterpart zostawac shows the same collocation restrictions with regard to participles, and it cannot refer to an ongoing process.46 Zostac (pf.) and zostawac (ipf.) differ only with regard to aspect (as a grammatieal category!), but they share a common, namely punctual, Iexieal aspectual de­fault, which cannot be altered. This defauit has probably not changed through the whole semailtic history of these verbs. And the shift from continuative to inchoative alone means only a lexieal evolution of the verb zostac. It does not

Passives as grammatical constructions 301

by itself clarify why zostac has become a component of a grammatical con­struction (i.e. the passive).

In order to explain this we should account for the fact that at some moment in its history the copular verb zostac began to combine with a participIe. I men­tioned earlier that zostac was much more widespread with adjectives. So its combinability with n/t-participles seems to be a case of expansion in the lex­icon (whether triggered by negated participles, as suggested by Weiss, or not must, for the time being, remain an open question). However this might have been, we can try to describe the phenomenon merely from its functional side and state the following: the participle "inherits" a considerable part, if not the entire argument structure of the verb from which it is derived; for this reason it is capable of focussing on the same event as does the verb, whereas the cop­ula loses its specific aspectual meaning. This line of argument is proposed by Schwarze with regard to Italian rimanere (2003: 25), a "semantic cousin" of Pol. zostac. This argument fits the evidence on the development of the Polish verb only insofar as zostac originally meant 'to remain, stay' (both as copula and as existential-Iocational verb), from which an inchoative meaning 'to be­come' with different parts of speech (as nominal predicates) evolved.

There are , however, some thomy details in the Polish case for which Schwarze's argument does not seem to be wholly appropriate. First of all, zostac until today is restricted mainly to n/t-participles of verbs indicating a change of state (and closely related ones). Compare: zostal zgniecony, zamel­dowany, zauwaiony 'was crumbled up, registered, noticed' are fine. Parti­ciples from punctual verbs not indicating changes of state, however, do not combine with zostac; see, for instance, *zostal uderzony, spotkany, zoba­czony 'was hit, met, caught sight of', which are no acceptable passives in con­temporary Polish47 (and most probably have not ever been). This seems to hold for It. rimanere, too. In any case, we have to take into account that pas­sive participles of verbs denoting changes of state nonnally refer not to the telic event denoted by finite fonns ofthese verbs, but to the post-state implied by such an event. Such a post-state makes them suitable for denoting resulta­tives (see 3.2.1). Nonetheless, a combination of zostac with such a participle focusses on the event (eI)' not the subsequent state (state l ). This can only mean that what determines the aspectual interpretation of the whole combination is the auxiliary, and that the resultative default of the n/t-participle can be altered by it in this particular construction. Thus, zostac becoming a passive auxiliary does not necessarily imply a loss of lexically inherent aspectual properties,48 i.e. a kind of semantic bleaching.49

This conclusion is in line with the fact that zostac does not share a couple of properties which are considered as typical for auxiliaries. The auxiliary zostac

302 Bjöm Wiemer

certainly does not lose any argument. Rather, another argument has been in­corporated, which is shared with the participie that combines with zostac to a complex predicate. Further, its paradigm is not defective, i.e. it still fully dis­tinguishes tense, person, number and gender (in the past tense). In this respect it behaves like any other perfective verb and enters into anormal aspectual pair with an imperfective counterpart, namely zostawac (see above). Given this, zostac does not conform to the role that auxiliarization would imply a de­crease of categoriality with respect to these grammatically relevant properties (cf., for instance, Hansen in this volume). Does this mean that auxiliarization is not always accompanied by a 10ss of categoriality, or does it follow from the above that zostac combining with an n/t-participle is not a proper passive? I firmly suppose the first conclusion is correct.

In connection with this consider the following. The resultative default of n/ t-participles is not altered with the BE-copula (byc), which is more widely used as the finite part ofPolish passive constructions. This, in turn, means that zostac and byc in combining with n/t-participles of perfective verbs demonstrate complementary aspectual properties, namely an opposition 'dynamic : static' (focus on el vs. focus on statel). In this respect, they form a paradigm with clear-cut complementary functions. This transparent distribution (or: "share of labour") is amplified by the fact that only byc, but not zostac can combine with n/t-participles of imperfective verbs, and that byc + n/t-participle of an imper­fective verb is reserved for the same basic functions of imperfective aspect as in the active, namely: for non-time-located states of affairs (iterativity, habitu­ality etc.; cf. (22» and ongoing processes (progressive meaning; cf. (23)):

(22) Podczas spotkafi z kolegami (zwykle) during meeting.GEN .PL with colleague.INs.pL.M usually byl oglaszany wynik cOP.3 .SG.M announce.PTClIpF.NOM.SG.M result.NoM.SG.M ostatniego konkursu last.GEN .SG.M competition.GEN .SG.M 'During each meeting with colleagues (usually) the result ofthe last competition was announced.'

(23) Gdy weszlismy na salt:., (akurat) byly when enter.l.pL.PRT into auditory.ACC.SG.F just cOP.3.PL.NON-VIR omawiane ciekawe sprawy discusS.PTC!IPF .NOM.PL. NON -VIR interesting.NOM.PL.F issue.NOM.pL.F 'As we entered the auditory, interesting issues were (just) being dis­cussed'.

Passives as grammatical constructions 303

As a result, the fonnal distribution of auxiliary (zostac vs. byc) + aspect of n/ t-participle (perfective vs. imperfective) began to double the functional dis­tribution of active verb fonns. Even more so, by virtue ofthe 'dynamic: static' distinction, backed by the morphological opposition between zostac and byc, the Polish passive nowadays displays a richer inventory of stable fonn: func­tion distinctions than the active voice (V. Lehmann 1992: 550; Weiss 1977: 103). This enhances the paradigmatic character of constructions consisting of zostac/byc + n/t-participle. I want to stress that this is the contemporary state of affairs in standard Polish, which came into existence not earlier than in the middle of the nineteenth century. The neat distribution of perfective vs. imper­fective n/t-participles, according to the aspect functions of the respective ac­tive (finite) verbs, plus the dynamic: static opposition expressed by the auxil­iarles, plus the fact that the collocation of zostac is restricted to n/t-participles show that the construction type in (18) has become paradigmatically tigther.

Another curious fact is that the combination of zostac with an adjective still renders an exdusively continuative meaning; e.g. X zostal szczodry can only mean 'X has remained generous', not ' ... has become generous'. In mod­em Polish we thus observe a complementary distribution of inchoative vs. continu~tive meaning depending on the word dass of the predicative com­plementizer; only with predicative adjectives has the old continuative mean­ing survived. It does not very much matter whether we want to represent this "split" of meaning as polysemy or in some other way, nor is it really important whether zostac is in one case an auxiliary, in the other a copula. What is crucial is the fact that the inchoative vs. continuative interpretation of the auxiliary/ copula depends on the combination with the nominal part of the predicate; the choice of word dass in the nominal part is decisive on the inchoative vs. con­tinuative interpretation of the whole construction.50

To sum up: the lexically inherent aspectual default of zostac (and of its im­perfective counterpart zostawac) has remained unaltered, it has always been a punctual one. A change has occurred only with regard to its collocation restric­tions, which have considerably been loosened in such a way that virtually any kind of predicative complement can be combined with zostac, among others, passive partici~les. The discussion above could not answer the question why collocation restrictions. have been loosened, i.e. what stimulated them. We can be quite sure that the trigger has to be found in contact with Middle Low Gennan (see Weiss 1982). But since the question of language contact is out­side the range of this paper, I want only to stress that, in any case, the gram­maticalization of the periphrastic passive, again, cannot duly be described as simply "le passage d'un mot autonome au röle d'element grammatical", as Meillet ([ 1912] 1948: 131 ) put it. Rather, it consisted in the increasingly com-

304 Bjöm Wiemer

plementary distribution of zostac vs. byc + pf. n/t-participles and of the doub­ling ofthe functions which are characteristicofthe aspect opposition (perfec­tive-imperfective) in the active. This process has ended up with a very strict paradigmaticization of zostac/byc + n/t-participles, based on cross-categorial clusterings (voice properties + grammatical aspect).

3.3.2. Latvian tikt and its "rivals"

Tuming now briefty to Latvian, we can say that the status of tikt 'to become' (originally 'to come, to reach, to get into') as a passive auxiliary is not so firmly established as zostac in contemporary standard Polish. For tikt has had some

"rivals" until the recent past, namely: tapt and kt:.üt, both meaning 'to become' (Holvoet 1994: 132), although at least in the standard language these are much less frequent than tikt (cf. Eiche 1983: 27). In Latvian dialects some other verbs may be used as a passive auxiliary, among them nakt, properly me an­ing 'to come' (Gäters 1977: 133; 1993: 293). Another important difference in comparison to Pol. zostac is the fact that tikt can combine with t-participles of verbs of virtually any aspectual class; inter alia it may be used with reference to ongoing actions (24) and states (25):

(24) Maja söbrld tiek celta house.NoM.SG.F just become.3.PRs build.PPP.NoM.SG.F (celta {=: celt 'to build, raise') 'The house is just being built'.

(25) VZ~S tiek cienits (cienits {=: cienit 'to esteem, respect') he be(come).3.PRs respect.PPP.NOM.SG.M 'He is respected'.

Therefore, its lexical aspectual default can be altered quite easily. This is rem­iniscent of Gennan werden « OHG uuerdan) and signifies that the aspectual interpretation of the Latvian analytic passive is mIed by the lexical aspectual default ofthe participle, not the auxiliary (as is the case in Gennan, too).

Finally, the paradigmatic distinction between'the passive construction with tikt and the constmction with the BE-copula büt is not as clear-cut as it is in Ger­man for werden vs. sein (for a systematic account on contemporary Latvian see Holvoet 2001b: 163-166). In this respect, the Latvian analytical passive constructions are closer to the Polish opposition of zostac- vs. byc-passive, where byc may often quite readily replace zostac, but not the other way round (whereas with the German auxiliaries the markedness relation is reverse).

Passives as grammatical constructions 305

3.4. Agent phrases

We should distinguish between agent phrases with a clearly adnominal ori­gin (as in Baltic) and those which developed from adverbal modifiers (as in Slavic), because their paths of development are cardinally different. In East Baltic the genitive of possessive origin has played the central role in the evo­lution of agent phrases. In Slavic languages, beside the bare instrumental, the following prepositions have played a major role as agent-phrases: (a) the "ab­lative" preposition ot 'from' (OCS Ot'b, Russ./Bel. ot, Pol., Cz. od); (b) the Polish "translative" preposition przez 'through'; (c) the Russian "adessive" preposition u 'at (the side of)'. In the history of each of the languages consid­ered here there have been a couple of prepositions which at one time or ' an­other showed a more or less ephemeric tendency to be used as indicators of the agent in passives and only as a consequence of language contact,51 e.g. Lith. "ablative" nuo(g), Latv. no 'from', Lith. "translative" per/par 'through' and Lith. "adessive" pas 'at (the side of)'. These were basically "implanted" by contact with Slavic languages and German (as for nuo(g) and no) and, at least in the codified standards, have been banished successfully by puristic language policy. Here I will not mention them any more.

Before analyzing the data, we have to answer a central question: what does it mean to mark an agent in the passive? Strictly speaking, this presupposes that an NP in an oblique case or a PP can be interpreted as the highest-rank­ing argument of the predicate's lexical structure, i.e. as the Actor. Therefore, it would be more appropriate to speak not of 'agent phrase', but of 'Actor phrase'. Both notions are defined differently and do not totally coincide: an 'agent' is defined with reference to certain kinds of specific case roles, irre­spective of their relation to the argument structure of the predicate, whereas the 'Actor' is not restricted to any particular inherent features of the refer­ent denoted by the respective noun, but defined strictly by its relative pos­ition in the predicate's argument structure (see Section 1). If, as is often done, "agent" only or basically refers to (usually human) instigators ("effectors" or the like) who control their actions and consciously bring about a change of state, 'agent phrases' would be excluded from passives of non-transforma­tive and/or non-agentive verbs (e.g. She was loved/supported/entertained by the whole family), because the noun in the by-phrase would not cope with this requirement. Nonetheless, probably nobody doubts that the PP by the whole family in this concrete case represents an argument of the verbs to love, to support, to entertain. For this reason, I henceforth apply the term "Ac tor phrase" for the obliquely marked, but semantically highest-ranking argument of the verb. 52

306 -Bjöm Wiemer

Thus, the following question is justified: how do such oblique NPs or PPs acquire the status of the (semantically) highest-ranking argument of the predi­cate in the clause, despite retaining the syntactie status of an adjunct? From the diachronie point of view, we may be quite sure (although ultimately this may be disputed, because in most cases historieal evidence is lacking) that such NPs and PPs became associated with verbs (verbal constructions) which already had eliminated or at least defocussed their highest-ranking argument to an extent that their lexieal structures changed. In other words: before po­tential candidates of obliquely marked Actors appeared on the floor, produc­tive techniques of morphological derivation had already led to some consid­erable changes in the organization of the lexieon in the respective language (see '3.1 and 3.2). This permits us to say that the appearance of oblique Actor phrases can, in asense, be treated as a re-introduction of the formerly "lost" or defocussed Actor (= highest-ranking semantie argument), and by this the lex­ical structure of those verbs whose derivatives (marked by the RM or as parti­ciples) constituted the ~asis of passives was re-established.

This way of reasoning corresponds weIl with the fact that Actor phrases are diachronically secondary in the sense that they derive from oblique cases (or PPs) with other functions: from oblique cases or PPs indicating the instrument, the source, the means, the transmitter (mediator), the reason (motive), or the possessor. How the "shift" to marking the Actor in the verb' s lexical struc­ture came about in every partieular case, may be a topic of some consider­able dispute, since hitherto this part of the "story of the passive" has been least thoroughly investigated. I cannot present any clear solution to the question whether the shift, e.g. from instrumental function to an agentive function (as with the instrumental case in East Slavie) or the shift from an "ablative" mean­ing (compare, e.g. German von 'from'), an "adessive" meaning (e.g. English by), or a "translative" meaning (e.g. German durch, French par, Pol. przez 'through') to an agentive one should best be described as a metonymic or ra­ther a metaphoric process. Anyway, we may at least reasonably argue that the establishment of an Actor phrase is an outcome of no less than two conver­gent processes. First, the oblique phrases whieh may develop into an oblique marker of the Actor and were named above are all more or less severely re­stricted to special referential classes: the transmitter and the possessor are, as a rule, persons; instruments and means, in contrast, typically are inanimate things or substances; a source may be a place, a thing or aperson; and a reason is an abstract entity. By eventually becoming an Actor, these restrictions dis­appear. Thus, again, the way from agent (requiring specific case roles) to Actor (which is a generalized semantic role dependent solelyon the predieate) pre­supposes an expansion to different semantic classes of verbs.

Passives as grammatical constructions 307

The question is, of course, whether the gradual loss of referential restric­tions is only the net result of achain of small steps conditioned by contigu­ity, Le. a basically metonymic change. This would be the case, e.g. in a causal event chain: the transmitter, the means or the instrument of an action are by ne­cessity in temporal and spatial contiguity with the causer of the action. Like­wise, the possessor has some kind of control over the object, as has the agent. Nonetheless, not every Actor is a causer (see above), and not every Actor can be related by contiguity to a possessor3 or to an instrument, and so on. Further­more, even in languages which can be said to have highly developed passives, inasmuch as the morphology of the predicate or the construction in (18) is con­cerned, the possibility of adding an oblique Actor not seldom proves to be re­stricted - for two diametrically opposed reasons: on the one hand, an oblique Actor may sound artificial or is impossible because of the verb' s lexical seman­tics, i.e. because the verb does not code an action with a volitional (and thus personal) agent. Compare, for instance, German, in which verbs of perception, reception, emotive attitudes and even of pos session form participial passive constructions, but the addition of a pp with von or durch often sounds inac­ceptable or is at least avoided (e.g. *Das gute Essen wurde von den Gästen ge­rochen 'The good meal was smelled by the guests'). The degree of acceptance evidently correlates in the positive with the degree of transitivity (as discussed by Hopper and Thompson 1980) that can be ascribed to the respective verb in the active. On the other hand, the predicate may be characterized as passive, but an otherwise possible oblique Agent is avoided or inacceptable, if it refers to a human (or animate) participant of the action. Probably this is why Rus­sians are often very reluctant to aceept sentenees like Dom stroitsja ?sosedom 'The house is being built by (our) neighbour', although they would aecept such a sentenee, if the Aetor is a referent lower down in the animacy hierarchy, as in, e.g. Dom stroitsja novoj stroitel'nojfirmoj 'The house is being built by a new building company'. Here the aceeptance of an oblique Actor correlates in the negative with the degree of animaey (see 3.1). Insofar as transitivity, animacy and volitionality are themselves correlated positively, the reasons for avoid­anee (or appropriateness) of an oblique Actor in the passive prove to be quite different, if not contradietory. 54 They cry out for further in-depth studies with authentic texts that would aceount for developments on the diachronic axis and treat with scrutiny competing motives of using or not -using oblique NPs or PPs in different kinds of passives. In any way, all the reasons discussed in this paragraph demonstrate that, when talking about the evolution of passives, we cannot dispense oflexical expansion, or loosening of referential restrictions.55

Second, the reverse side of the same phenomenon is the following. When . one or two distinct devices of marking an oblique Actor establish themselves,

308 Björn Wiemer

this presupposes that competing PPs or oblique cases are ousted from this function. In other words: paradigmatic tightening takes place. . Unfortunately, we da not have the means to check such hypotheses for the two most prominent Actor phrases in historical stages of Slavic languages, the instrumental and the pp Otb 'from' + GEN., because these appear as ra­ther full-fledged Actor phrases already in the earliest documents, those which were heavily influenced by oes (as concerns Otb + GEN., cf. (3) and (7» as weIl as those more genuine East Slavic ones (as concerns the instrumental). As for Slavic, I will restriet myself here to just one instance, namely: the pp u + GEN. I choose this case, because it through centuries has remained on the verge of becoming an Actor phrase in participle constructions. It therefore of­fers the possibility to demonstrate what should happen to an oblique syntagm in order to become a generalized means of marking the Actor.

3.4.1. Actor phrases in Northern Slavic

In general, PPs denoting the passive Actor are very rare in folk texts and non-sacral genres. This is also true of the ancient Novgorod dialect and the pp u 'at' + GEN. The earliest example attested by Zaliznjak (1995: 228) is from the eleventh century (26). But in this case it is not really clear whether the ref­erent named by this pp was the agent of the event, or whether the pp means only the place ofthe Sycevic-family:

(26) Ziznebude pogublene u sycevicö novbgorodbske smbrde (. .. ) 'Ziznobud, a peasant from Novgorod, was killed i. in the house ofthe Sycevics

ii. by the Sycevics'.

In any case, this ambiguity is due to a contiguous relation between the (imput­able) agents and the place where they lived. As far as I know, there are no early East Slavic examples in which u + GEN. unambiguously denotes an agent, that is: where the metonymie implicature from place, or 'sphere of interest' (see below), would really have shified to the assertive (focussed) part of the prop­osition. This situation obtains until today in colloquial Russian: although sen-tences like .

(27) U nego vse produkty kupleny at hirn aIl food-stuffs.NOM.PL bUY.PPP.NOM.PL 'All food-stuff is bought at hirn [-7 by him].'

Passives as grammatical constructions 309

(28) U nee vse osibki otmeceny at her all mistakes.NoM.PL mark.PPP.NOM.PL 'All mistakes are marked at her [~ by her].'

imply more or less strongly that the respective state was brought about by an event controlled by the referent named in the PP, this is by no means a reliable rule. Together with this, such an implicature can easily be cancelled; cf.:

(27') U nego vse produkty kupleny, no pokupal vse ego brat lit. 'All food-stuff is bought at hirn, but the one who bought it was his brother'.

(28') U nee vse osibki otmeceny, no otmetki v tetrad'vnesla ucitel'nica lit. 'All mistakes are marked at her:, but the one who made the marks in her exercise-book was the teacher'.

In fact, an agentive reading of u + GEN. has remained a matter of pragmatic in­ference (based on encyc10paedic know ledge, frames or whatsoever) and not becomeapropertyoftheconstruction [Cu + GEN.) + pastpassiveparticiple], let aloneofthePPitself(cf. also Weiss 1999: 181 withfurtherrevelantliterature).

The situation differs, however, in NW-Russian dialects round Novgorod and Pskov. Here this PP in many cases must be interpreted as the agent of the action denoted by the participle, and we can observe the moment of break­through towards conventionalization; compare two examples from Trubinskij 0995: 78, 82), who writes that u + GEN. here denotes the doer of the action, not only the person concemed by it:

(29) Vidis', zamok sporcen u kogo-to 'Look, the lock has been damaged by someone' [Iit. 'at someone'].

(30) Zveroboj u menja pucockom byl zavjazan, visel, u menja uze vykinuto 'The 8t. John's wort was tied into a bundle by melwith me, it hung, (but now) it has been thrown away by melwith me' ..

Unambiguous agentive uses, however, have remained in a c1ear minority even in those dialects. Only in approximately 6% of all text tokens does u + GEN. really denote the agent; by far the most frequent cases (c. 75%) are those in which this PP can be understood either as the agent or as the personal pos­sessor "affected" by the action, or both (Trubinskij 1984: 85-87) - just as in colloquial standard Russian (see (27-28».

310 Bjöm Wiemer

It is very likely that in Old Polish (before 1600) u + GEN. was used agen­tively by the same kind of implicature as shown above for the ancient Novgorod dialect. This assumpt10n isjustified, since SIStar IX (1982-87: 257-261) lists and illustrates the same broad inventory of functions of u which is character­istic for modem colloquial Russian. In standard Polish this inventory subse­quently somewhat reduced, but from 1420 we have an example in which u is used as the actor of adeverbal noun:

(31) Othpuszczene grzechom.DA T .PL W milego Jhesu (Hymn Salve Regina R XXIV 81, c.1420) lit. 'absolution of sins at dear Jesus', i.e. 'absolution of sins by dear Jesus'.

All this shows that the potential of developping into an Actor phrase existed for u + GEN. both in East Slavic and in Polish for a very long time. 56

3.4.2. East Baltic: Syntactic and semantic reanalysis without erosion and coalescence

Contrary to the roots of Slavic Actor phrases, which are all adverbal, the ob­lique Actor phrase in the Lithuanian passive is of ultimately adnominal ori­gin, since it derives from a former possessive modifier.57 From a synchronic point of view, the main difference between Lithuanian and Latvian consists in the fact that only Lithuanian allows a genitive NP (or a possessive pro­noun) to function as an adjunct of the predicate, i.e. adverbally, whereas the Latvian agentive genitive principally always still functions as an attribute to some nominal head. Only under very specific conditions can an Actor phrase be used independently of a nominal head (see below). From a diachronic per­spective, the Lithuanian genitival Actor phrase in the passive should be treated as the result of a continu~d development from a stage at which Latvian has stopped. Despite some obscurities in the details of syntactic reconstruction, we may assurne that in Lithuanian areanalysis took place by which an adnom­inal modifier beg an to be interpreted as the highest -ranking argument of the main predicate. .'

This reanalysis consists of at least two steps, the first of which was also un­dertaken by Latvian and which is a precondition for the second step, essentially taken only by Lithuanian. The first step can be characterized as the internal re­organization of dependency relations between two initially co-subordinated

Passives as grammatical constructions 311

modifiers of a noun, the first modifier being a possessive pronoun or a genitival NP, the second a t- or m-participle (32a). Subsequently, the possessive modi­fier became adependent ofthe participle (32b):

~~ 7 (32) a. mana lmätes padziedäta dziesma ---7

mY.F.NoM.SG/mother.F.GEN.SG sing.PPP.F.NOM.SG song.F.NoM.SG lit. 'my/mother's (and the) sung song' =:: 'the song which is minel

mother's and wpich is sung'---7

~~ b. manalmätes padziedäta dziesma

'the song which is sung by me/mother'.

Example (32b) represents the stage of contemporary Latvian. Potential Actor phrases in the genitive can occur practically only as attributes of t-participles, i.e. they cannot leave the NP headed by the participIe. It is important to stress that this reanalysis takes place not on the level of the c1ause, but only inside the complex NP: two originally juxtaposed modifiers of the same noun are restruc­tured in such a way that one of them, namely the participle, becomes the head of the other, i.e. the genitive/pronoun. The internal dependency relation be­tween participle and genitive/pronoun is retained if the participle ( + copula) is used as the predicate of the c1ause,58 as in Dziesma ir manalmätes padziedäta 'The song is sung by me/by mother'. In Latvian the possessive noun (or pro­noun) cannot be understood as a syntactically independent denotation of the Actor, although it might be used under marked focus; e.g.

(33) [[m biedriba] NP/PSA [ir [strädnieku DEM.NOM.SG.F association.NoM.SG.F cOP.3.PRS worker.GEN.PL

dibinäta] ] NUCLEUS]CORE

found.PPP.NOM.SG.F 'It was the workers who founded the association'.

Cf. Holvoet (1994: 133), who also remarks that the agentive genitive strad­nieku cannot be used sentence-finally, i.e. after the participle. So this is only the threshhold ofthe second step by which this noun (or pronoun) moves out of its NP-internal dependency. This second step can be modelIed by a com­bined dependency-constituency representation with a Lithuanian example, which corresponds to (33):

312· Björn Wiemer

~ (34) a. [[ [Darbas] NP/PSA [(yra) mano atliktas]NUCLEUS]CORE]CLAUSE ~~

b. [[ [Darbas ] NP/PSA [(yra) atliktas ] NUCLEUS]CORE mano ]CLAUSE ~ .

'The work was performed by me'.

We see that this type of reanalysis restructures the whole c1ause, not only NPs. Furthermore, the position of the possessive NP/pronoun is no longer fixed with regard to the participle, i.e. it can appear virtually at any place in the c1ause. This is the main difference with regard to Latvian. Another, actually minor dif­ference is the fact that all Lithuanian possessive pronouns used to indicate the Actor are indec1inable (mano.POss.l.SG, tavo.POss.2.SG, savo.POSS.REF, etc., see (34)), whereas Latvian uses forms dec1ined for gender and number (manis. POSS.M.1.SG, mana.poss.F.l.SG, etc., see (32)). Although this renders a hith­erto unsolved problem for the relative chronology of the possessor phrase and the dec1ension type of possessive pronouns, we can be sure that the reinterpre­tation of adnominal genitival nouns and possessive pronouns as adjuncts on the c1ause level, which took place in Lithuanian, but not in Latvian, is a rather recent phenomenon (Holvoet 1995). Their common roots for both languages are to be sought only in the former possessive nature of the agentive, which was ancient enough to be a common source for both East Baltic languages.

Two further remarks are appropriate here. First, beside the syntactic changes just described, there is an obvious semantic relationship between pos­session and control of action. As said above, both are cognitively contiguous to each other, allowing, thus, a shift by implicature from possession to control of action. Such a shift is based on the same principle as metonymy.59 However, there is a problem, rightly noticed by Hol voet (1995: 177): how should a pos­sessor of a participle (more precisely: of the state denoted by the particple) be interpreted (e.g. in (32b), (33) or (34a))? Since such a semantic relation seems rather weird, we must conc1ude that the possessive force of the genitival NPI possessive pronoun must have been lost at the moment when the syntactic re­organization of the dependency relations on NP level took place. This is the only conc1usion implied by Holvoet. I should add to his observation that to­gether with its loss of possessive force the genitival NP/pronoun was already reinterpreted as the controller of the action denoted by the participle.6o This means that semantic reanalysis, consisting in a contiguity-based shift, pre­ceded syntactic reanalysis.

Passives as grammatical constructions 313

Second, I mentioned above that the more advanced stage in Lithuanian shows that the NP/pronoun denoting the Actor can be freely placed at any pos­ition in the c1ause. This actually runs counter to one of Lehmann' s parameters

. of grammaticalization, namely: syntagmatic cohesiveness (bondedness). In­stead of becoming tighter the whole construction called 'passive' consists of parts which can "wander around" in the c1ause more freely than they could be­fore independent Actor phrases established ihemselves.

As for grammatiealization, the problem now is that we have two concom­itant processes, which are responsible for the evolution of three-partite' pas­sives: a semantic shift based on cognitive contiguity, and syntactic reanalysis of dependency and constituency structure; but no such processes as phonetic erosion or syntagmatic coalescence are involved. If we followed the line of argument displayed in Haspelmath (1998), no grammaticalization at all took place. Haspelmath argues not only that reanalysis need not lead to grammatic­alization, but even that it is no necessary step towards grammaticalization. In the given case this would bring us to the very strange conc1usion that the pro­ces ses described above, which led to a grammatical construction called 'pas­sive', cannot be named 'grammaticalization'.

3.4.3. Resumptive remark conceming the agent

Analogical problems would arise with all other kinds of Actor phrases, too. The "re-introduction" of the Actor into the lexical structure of the predi­cate (verb) is attained mostly by semantie reanalysis, analogieal expansion (whether based on contiguity or not) and paradigmatic tightening achieved at by the reduction of competing potential Actor phrases. Syntactic reanaly­sis played a role only with the oblique agent in Lithuanian and its preceding stages, which are still reftected in Latvian. We are confronted with very simi­lar problems if we look at the verbal parts of passive constructions.

4. Discussion of standard parameters - why they are not sufficient

After having surveyed the basic set of data on passive constructions in North­ern Slavic and Baltie languages from a diachronic and synchronie point of view I will turn now to an examination of the diachronically retrievable pro­cesses of passive formation in Northern Slavic and Baltie from the viewpoint of common criteria of grammaticalization which have been elaborated by c. Lehmann 0985; 1995).

314 Bjöm Wiemer

To begin with, the criteria of semantic and phonetic erosion cannot at all be applied to the participles which came to serve as the main part of analytic pas­sives. Semantic erosion ('bleaching') is something that, by and large, distin­guishes the RM from the reflexive (anaphoric) pronoun to which it is etymo­logically related; cf. Common Slavic orthotonic *st:/sebe > clitic Pol. sit:., Cz. se, Slov. sa,61 wordform-final affix (postfix) -sja/'s in East Slavic. But this is only a net result, which ensued much earlier than the RM must have begun to be used in passive function, i.e. as a marker of a grammatical construction by which the argument structure of the predicate does not change. Both bleaching and attrition started already at the moment when the reflexive pronoun began to indicate various kinds of changes in semantic and/or syntactic valency, i.e. when it began to serve as a marker oflexical derivation (see 3.0. For the same reason phonetic erosion is no useful criterion either. It begins much earlier be­fore the RM begins to fulfill the services of a passive marker. Extemally, it does not differ from the RM as a marker of various kinds of valency changes.

Similarly, the development of Pol. zostac from a copula and existential verb with exclusively continuative function to an inchoative auxiliary has not been accompanied by anY phonetic erosion. Even semantic bleaching has not accompanied this shift: as an auxiliary, zostac when combining with a n/t-par­ticiple shares the second (= lower-ranking) argument with that participle. But it does not lose its lexical aspectual value (a punctual one) nor its telic nature (see 3.3.0. As for Latvian tikt, the erosion of the word-final vowel « tikti) is nothing peculiar for this particular verb, bilt part of an overall process in Latvian, which has lost word-final vowels as a consequence of the general change from a variable word-stress to a fixed initial word-stress system.

Neither can phonetic erosion be observed with the participles which are constitutive for analytic passives. German and other languages have lost case and agreement suffixes on participles used in passives, but this is part of a gen­eral tendency conceming adjectives and participles in predicative use. So we need not expect the languages examined here to show the same tendency. All of them have retained quite a rich case system together with mIes of agree­ment within NPs; these roles pertain also to the link between the nominative subject (= PSA) and predicative participles. It appears to be an overall typlog­ical characteristics of these languages. Thus, the maintenance of agreement mIes in passives shows that phonetic erosion is not a reliable feature, but only an accidental fact in the evolution of passive constructions of just some lan­guages.

Finally, with the establishment of Actor phrases phonetic erosion is no sen­sible criterion either. It is, however, a tricky question whether the evolution . from an agent phrase proper to an Actor phrase can be regarded as semantic

Passives as grammatical constructions 315

bleaching. In some sense, it can. Namely insofar as the Actor embraces very different specific semantic roles and, together with this, is not restricted to per­sonal, or animate, referents (cf. 3.4). But if this process should be considered semantic, or functional, bleaching, it will be just another name for the succes­sive loosening of selectional restrictions on the types of referents which can occupy the verb' s argument positions. In turn, this loosening would coincide with the expansion of the particular passive construction in the lexicon, i.e. more and more verbs can be "entered" into a constmction by which only the Actor-Undergoer-mapping into the syntax is changed (and nothing more); see Figure I. Again, such a process has not normally been described in gramm at­icalization studies, although it leads to a tightening of paradigmatic relations in the grammatical system (see below).

Let us now look at scope criteria (syntagmatic weight). Adecrease of scope (condensation), again, has affected only Pol. zostac and Lat. tikt, insofar as de­crease of scope is a typicaI by-product of auxiliarisation. NPs in oblique cases or PPs do not diminish their scope by being interpreted as Actor phrases. The Baltic genetivus auctoris shows even wider scope than the possessor phrase from which it dec1ines; the reason is simply that the possessor is an adnominal modifier, i.e. has scope over a NP, whereas the agent or Actor phrase is syntac­tically an adjunct modifying the whole clause (on its arguable syntactic status see n. 8). Furthermore, neither participles nor RM-forms of verbs show any­thing similar to decrease of syntagmatic scope. The cline: stressable * s~/ sehe > c1itic s~ > agglutinated -sja/'s in East Slavic, being accompanied by a logs of referentiality of the reflexive when it began to serve as a marker of valency reduction, represents a process which, in fact, led to a decrease of syntagmatic scope from the c1ausal to a lower level of constituency. But, again, these losses concern lexical derivation as weIl; in no way are they directly related to the passive function of the RM as such.

Syntagmatic cohesion and variability do not play any decisive role either. The reasons are the same as with erosion and cond~nsation (decrease of scope): although c1iticization and agglutination of * s~/ sehe to Pol. si~, Cz. se, Slov. sa, East Slavic -sjal' s (as weIl as Lith. -si-) represent processes of a typical gram­maticalization c1ine, they appeared independently and long before the rise of RM-passives.62 These morphological processes did not occur at all with the participies and copulae which form. part of analytic passives. In Slavic and Baltic languages, whose word order mIes are pragmatic rather than syntac­tic, participies and copulae abide by just the same communicative (pragmatic) principles of word order as do. any other syntactic units of these languages; there is nothing specific to passivization that would restrict these basic com­municative principies. Thus, syntagmatic cohesion and variability are useless

316 Bjöm Wiemer

as parameters "measuring" grammaticalization of passive constructions, at least in these languages.

Let us come then to paradigmatic cohesion and variability. These two para­meters appear to be the only ones which really do touch upon crucial aspects of the evolution of passives. To begin with cohesion, we should notice that the number and variation of potential Actor phrases in passives has diminished, mostly to a degree that only one type of oblique NP or PP is left: the genitive (vs. nuo + GEN.) in Lithuanian, the instrumental (vs. otlvid + GEN.) in Rus­sian, Belarusian and Ukrainian, the PP przez + ACC. (vs. od + GEN. and the bare instrumental) in Polish. Czech is less clear in this respect, but since the actual distribution of od + GEN. (vs. bare instrumental) functioning as Actor phrase has, as far as I know, not been investigated thoroughly, I cannot con­tribute any definite statements concerning these languages.

A similar picture obtains for copular verbs which for some time were can­didates of becoming passive auxiliaries. At least in standard Polish zostac 'to become' has established itself firmly as such an auxiliary; possible "rivals" (stac sit:., zrobic sit:. 'to become; to appear') have remainedjust copulae or ex­istential verbs with an inchoative or ingressive value (cf. Wiemer 1998: 181-184, 194-196). Similarly, Lat. tikt has "overcome" other verbs which have been showing a tendency of developing into a passive auxiliary (see 3.3.2). In any case, paradigmatic variability decreased both with the choice of the aux­iliary and with the choice of oblique case or PP to denote the Actor: only one element out of a set of possible candidates survived in each of these functions . .

A good case to illustrate a decrease in paradigrnatic variability would be the inchoative auxiliary in Polish. Pol. zostac and byc developed a clear-cut opposition of 'dynamic' vs. 'static' (= resultative) passive; zostac itself re­mained restricted to nlt-participles derived from perfective verbs (3.3.1). By this, the construction ofthe type in (18) has become tighter. More generally, a factor which has led to a considerable paradigmatic tightening of passive con­structions in Northem Slavic, in particular in Russian and Polish, is the inti­mate entanglement of the type of the passive construction with the opposition of perfective vs. imperfective aspect: passive constructions (RM- vs. analytic passive in Russian, choice of auxiliary and aspect of nlt-participle in Polish) no longer vary, but practically show a complernentary distribution of func­tions with forms; cf. 3.1 and 3.3.1. The choice of the constructions practi­cally depends on the aspect functions, which are in this way "copied" from the functional distribution of perfective and imperfective verbs in the active.

The evolution of the specifically inchoative auxiliary in Polish and the gen­eral clear-cut clustering between aspect rnernbership of the verb stern (per­fective vs. imperfective) and type of passive as weIl as the introduction of a

Passives as grammatical constructions 317

single type of Actor phrase can all be regarded as steps in a continued pro­cess of grammaticalization, after the participles already had established a sta­hle orientation towards the Undergoer ofthe deriving verb and RM-fonns of verbs had "made their way" up to the passive function.

It might be objected that categorial correlations (or clusterings) as these ones are by themselves no argument in favor (nor in disfavor) of grammatical­ization, even if they cause a tightening of paradigmatic relations. At least, ar­guments of this sort have not normally been discussed as possible criteria of grammaticalization. However, an increase in correlation between categories which are expressed by the same morphemes or constructions often leads to an increase in abstractness of content, i.e. something we might ,call 'bleach­ing'. More so, this kind of development appears to be unidirectional. It hap­pens under the condition that the interacting categories themselves are al­ready abstract enough, i.e. sufficiently independent of the lexical content of the sterns which enter into these constructions (or are affixed by the respec­tive morphemes). Ifwe continue this line ofthought we arive at the conclusion that increase of grammaticality is directly correlated, first, with a decrease of restrictions on the sterns to which the constructions or morphemes can be ap­plied. Second, they are correlated with a decrease of functional variation be­tween the involved constructiöns, morphemes and categorial clusterings.

5. What hitherto has been ign~red or neglected

The above discussion allows for different, at least in part diametrically op­posed conclusions. We may infer either that structural criteria like those for­mulated by C. Lehmann are insufficient to characterize at least certain types of grammaticalization phenomena. Or we may argue that what has been going on in Northem Slavic and Baltic and has eventually led to what we call 'passive' is no grammaticalization at all. Morpheme-based approaches should indeed be forced to take the latter position. For them, grammaticalization would stop where it should actually start for construction-based approaches. For instance, Haspelmath (1990: 51) would not treat as cases of grammaticalization the ex­pansion of a productive derivational suffix on lexical sterns, so that such phe­nomena like the application of an originally intransitive marker in a passive function do not count as grammaticalization. This is somewhat striking given the fact that Haspelmath does admit the reflexive marker Oike Slavic * st:l sehe > Russ. -sja/'s) a grammaticalized marker of the passive. This marker is like­wise a derivational device, and above 1 showed that it gained its grammatical function of marking the passive not by. virtue of semantic and phonetic ero-

318 Bjöm Wiemer

sion, but at the end of a long-Iasting process oflexical expansion. There is thus no principal difference between the story of the "reflexive passive" and the story of other intransitivising markers.

I think that the following option should be pursued: we are in need of sup­plementary criteria, and it may turn out that criteria of the sort as have been applied since Lehmann's ground-breaking work on grammaticalization are too narrowly aimed at describing certain kinds of phenomena. We therefore have to ask the question whether we want to embrace a wider range of phe­nomena to be subsumed under' grammaticalization' and formulate additional criteria in order to cope with them. In addition, it might appear reasonable to ascribe different weight to various criteria that have been applied so far and which could be suggested on the basis of analyzing the evolution of pas­sive constructions. The establishment of passives cannot adequately be ac­counted for from an exclusively or primarily morpheme-based grammatical­ization theory. How can the passive be considered a grammatical category if so few of the tradition al criteria of grammaticalization theory are suitable to describ~ what has been going on with the relevant morphological parts of pas­sive constructions to set up this category? Passives are constructions, and they have to be described in terms that less heavily (or not at all) rely on criteria of "morphologization". The grammaticality of passives does not depend on the degree of morphologization (on a grammaticalization cline) of the differ­ent components (e.g. RM: clitic or postfix). Passives are often syncretic with other categories (e.g. resultatives) that are expressed by the same (combina­tions of) forms. None of their components (RM forms of verbs, participles, copulae, agent NPs/PPs) differs morphologically from forms with other para­digmatic status, among others from forms that are traditionally regarded as DERIV ATION AL formations.

In fact, the question about grammaticalization of passives becomes even more complicated by the fact that passives, defined as in Seetion 1, on a whole are not a category which should (or should not) be applied obligatorily under well-definable context conditions. That is, no language user is really forced to use the passive instead of the active in the same way as, for instance, s/he has to choose a certain case, an agreement pattern or to abide by a specific word or­der mle (e.g. in order to distinguish assertion froni question). All the more this is true with regard to Actor phrases, which are more often than not optional even within otherwise fully developed passive constructions. Active and pas­sive (as weH as other voice constructions) cannot by its very nature be com­pared to members of a paradigm in the usual sense of this term, because they are not determined by syntactic mIes on clause or NP level, but themselves set up a frame for clausal syntax. If a passive turns out to be an obligatory syntac-

Passives as grammatical constructions 319

tic construction, it will have turned into something different from the passive, namely an ergative construction (cf. Estival and Myhill1988 for such cases). Markedness relations then become inversed: the ergative (former passive) is the neutral way of coding Actor and Undergoer, the Actor being in a marked case. This, however, has not happened in Baltic or Slavic.

Briefty, neither 'intraparadigmatic' nor 'transparadigmatic variability' (C. Lehmann 1995: 138) seem to apply as viable criteria for characterizing the degree of grammaticality of the passive as such. If a language has a passive we can in every case determine whether the speaker uses a passive or a cor­responding active construction. But speakers are in no obligatory way driven . into using the passive if they do not want to. The grammaticality of thc passive rests solelyon the fact that it does not alter the lexical meaning of the predi-

/ cate in comparison to the active, as the unmarked member of the morphosyn­tactic opposition.

What does count as an agreeable parameter of the degree of grammaticality is (a) the extent to which the construction type having a strictly passive func­tion has expanded in the lexicon, (b) how clear-cut the morphosyntactic char­acteristics are, and (c) in which way they intersect with other verbal categor­ies (e.g. aspect) rendering functionally complementary expression classes, i.e. complex paradigms. A further criterion is (d) whether there is a clear choice for an oblique marking of the Actor (highest-ranking argument of the predi­cate). Once we have recognized these criteria - in addition or even instead of the standard criteria of morpheme-based grammaticalization theory - we can begin to inquire which of them should be given more weight and which of them considered as minor ones. But this would be the task for another paper.

Notes

1. I would like to thank Miriam Butt, Christoph Schwarze and Petra Vogel for their patience in commenting and/or proof-reading an earlier version of (parts of) this paper. I am also grateful to Josef Bayer and Uli Lutz for their useful discussion conceming generative conceptions of the passive, in particular of the agent -phrase. Of course, the usual disclaimers apply.

2. Whether and to what extent these shared features are a result of common I.E. heri­tage or rather of convergent development due to extensive and long-Iasting lan-guage contact will not be at stake here. .

3. This term refers to the territory of contemporary East and West Slavic standard lan­guages and their respective dialects: the first group consists of Russian" Belaru­sian and Ukrainian, the second of Polish, Czech, Slovak and the two Sorbian stand­ard languages. I will, however, totally exclude Sorbian here and focus mainly on

320 Bjöm Wiemer

Russian and Polish. Baltic, represented by the only extant (East Baltic) languages Lithuanian and Latvian, is a different, though genetically closely related branch within the I.E. family.

4. The more so, as so-called dative or accusative subjects, which also occur in these languages (cf. Russ. Mne.DAT ne rabotaetsja::::: 'Itdoes not workquiterightforme' for a "dative subject", Pol. Trzt:.sie jq.ACC lit. '(lt) shivers her', i.e. 'She is shivering' for an "accusative subject"), can be created only from a quite limited range of verbs and often only under the scope of an additional clausal negation or a manner adver­bial (e.g. Russ. Emu rabotaetsjaxorosols trudom::::: 'For hirn it works well/badly'). Perhaps even more crucial is the fact that such "subjects" (at least in the languages investigated here) do not show some behavioral properties ascribed to typical sub­jects since Keenan (1976) or in the framework of Role and Reference Grammar (RRG; cf. Van Valin and LaPolla 1997).

5. Although there may be more than two arguments, only two per predicate Oogical structure) can be assigned macrorole status.

6. Cf. Van Valin/LaPolla (1997: 274): "(. .. ) grammatical relations (syntactic rela­tions) exist in a language only where the b~havioral patterns of a language give evi­dence of a syntactic relation independent of semantic and pragmatic relations; that is, only where the behavior patterns are not reducible to semantic or pragmatic rela­tions can we say there is evidence of syntactic relations" (see also ibd.: 251-255).

7. The latter property is not a genuine syntactic, but a pragmatic (discourse-syntactic) one. Here there is no need (nor space) to neatly distinguish subjects and topics.

8. Since I would like to avoid misunderstandings and unnecessary controversies, but do not pursue here a general discussion of theories dealing with the syntax-seman­tics interface, the following remarks should suffice. It is crucial to distinguish be­tween semantic arguments, which are here understood as necessary components of the verb's lexical structure (its lexical entry), and their syntactic realisation, how­ever both may be represented formally. If we accept this distinction it will be pos­sible to say that although obliquely marked Actor NPs/PPs do not belong to the core, but to the periphery, they are nonetheless arguments 9f the respective predi­cate. 'Core' and 'periphery' are syntactic notions, and for this reason oblique Ac­tor NPs/PPs are usually treated as adjuncts (which is another syntactic notion). Ap­proaches differ, however, as far as the linking mechanism from semantics to syntax and vice versa is concemed. Some scholars from the generative camp have as­sumed intermediate concepts such as 'argument-adjuncts', which were invented in order to capture the discrepancy between argument structure and its representation in the syntax (see e.g. Stroik 2000: 138 for some comprehensive discussion). Such concepts have been linked up with the theory-internal notions of 'external argu­ment' and 'theta roles'. The latter differ from arguments in that, roughly speaking, they can lexically bind an argument and prevent it from being expressed in the syn­tax (thence the notion of 'existential binding'); cf., for instance, Wunderlich (1993: 731 et passim). Anyway, the necessity of theta roles seems now to be questioned by an increasing number of generativists (U. Lutz, p.c.), and it does not play any role in RRG. Beside RRG, also LFG (in its newer approaches) makes a clear-cut dis-

Passives as grammatical constructions 321

tinction between semantic and syntactic level and operates exclusively with linking algorithms between these two levels based, inter alia, on argument hierarchies (cf. Dalrymple 2001: 195-213). As for HPSG, passivization is described as a lexical role. This rule, though, actually affects only the verb's syntactic valency (belonging to its subcategorization list), whereas the assignment of semantic roles is preserved (cf. Pollard and Sag 1994: 305).

9. Participles with the rn-suffix are still used in standard Russlan, but, probably be­ing calqued from Church Slavonic, they already ceased to be used productively in the earliest attested times and are considered bookish (see further 3.2.2-3, n. 33).

10. In the impersonal passive, i.e. if a nominative NP as PSA is lacking, the singular neuter form is chosen (as a kind of default for non-agreement).

11. In Russian it must be omitted in the present tense. In Ukrainian, Belarusian and also basically in colloquial and dialecta1 Polish, as weIl as in the Baltic languages the present tense BE-copula can be omitted.

12. These are the two allomorphs in standard Russian, with -'s occurring after vowels, -sja after consonants. Belarusian and Ukrainian have only -sja.

13. The reconstructed Slavic forms *sebe and *s~ originally distinguished accusative and genitive, and * s~ could be stressed, too. But by the time of the first written documents the case distinction had already been virtually lost and been replaced by adistinction ofclitic (*s~) vs. non-clitic (*sebe) form, so that both forms could be used either as genitive or as accusative (cf. Gadolina 1963: 57). It is unclear how this paradigmatic levelling refers chronologically to the beginning aggluti­nation of *s~ (see below). It is, however, plausible to assume that agglutination started only after * st:, began to be used without stress.

14. See the detailed account and discussion in Endzelin (1922: 705-708). 15. Haspelmath's findings and hypotheses have been substantialized by Geniusiene's

(1987) implicational relations between different types ofRM-verbs with valency changes as weIl as, for instance, by Kemmer's (1993) investigation (to name but two prominent studies).

16. Cf. Geniusiene (1987) for a detailed and systematic account of valency changes caused by reflexive markers and typologically relevant implications of their func­tional evolution.

17. See the theoretical comment in n. 8. 18. For arguments in favour of such a view on grammaticalization see also Wiemer

(2001). 19. Example cited from Dankov 0981: 89). 'Old East Slavic' here and henceforth re­

fers to the East Slavic dialect continuum until the fourteenth century. Only from thence does it make sense to speak of Russian, Belarusian and Ukrainian, respect­iveIy.

20. That is, those which correspond to today's verbs of perfective aspect. 21. Cf. an example from a travel account of the eighteenth century: knjaz' Vaxtang,

kotoryj pri saxe podarestom sidel, otpustilsja 'knight Vaxtang, who was taken intoarrestatthe Shah's court, has been released' (citedfromKrys'ko 1997: 376). Today we would have the analytic passive byl.M.NoM.SG otpuscen.M.NoM.SG.

322 Bjöm Wiemer

22. A sentence like Zdenek sili my je can 'be interpreted only as 'Zdenek is washing hirnself', not as 'Zdenek is being washed' (if Zdenek, for example, is an infant). Contrast this with.,Russian Rebenok moetsja allowing for two readings: (i) 'The child is was hing itself' or (ii) 'The child is being washed'.

23. Reference to time-Iocated events is mainly possible only in cornbination with the BEcoME-copula zostac, but, in turn, this copula cornbines only with n/t-participles ofperfective verbs (see 3.3.1).

24. For the basic facts cf., among others, Grepl and Karlik (1998: 132), Rytel-Kuc (1990: 34f., 92), Sokolova (1993: 36), Sticha (1979: 65).

25. Anyway, such an ambiguity is not excluded even in the third person, as can be seen from such examples like Cz. Pokhidti. se za znalce, which rnust be disambiguated by the context. For its meaning rnay be not only passive ('He is considered to be an expert'), but also reflexive (' He considers hirnself to be an expert').

26. This division, as resting on the notion of macroroles in RRG, is a sernantic one. It basically coincides with the syntactic classification 'subject-oriented' vs. 'ob­ject-oriented', which was introduced in 1983 by Nedjalkov and Jaxontov (see Ned­jalkov and Jaxontov 1988: 9 for the English version) with respect to resultative con­structions. Since in NOM-Acc-Ianguages in the active the Actor tends to be coded as subject (in the nominative) and the Undergoer as object (accusative without preposition), for the present purpose both classifications render identical results.

27. This refers to what Haspelrnath (1990: 59) calls 'inactivization'. 28. Of course, resultative participles of intransitive (unaccusative) verbs are weIl at­

tested, as weIl. But an irnplicational hierarchy of the following sort seems to hold: if a language has subject-oriented resultatives, it also has object-oriented resulta­tives (cf. Nedjalkov andJaxontov 1988: 17,23).

29. On this aspect cf. also Haspelrnath (1994: 161), who casually speaks of 'syntac­ticization'. This part of,the "passive story" is mirrored, as it were, by the way in which the Actor phrase is generalized over specific semantic roles (agent, experi­encer, etc.); see 3.4.

30. To this aim it is not necessary that the Actor be explicitly named (as an oblique NP or PP). See section 3.3.2.

31. Cf. Gebauer (1929: 619), also for further examples on Old Czech. Havninek (1937: passim) gives a lot ofpertinent examples from different Slavic and the Bal­tic languages.

32. For instance, the m-participle in (12), which in this context can only be under­, stood as Actor-oriented, corresponds to the active form 'tO'Ü avaLQo'ÜV'to~ 'the killing (one)' in the Greek original.

33. The m-suffix as a productive means of deriving verbal adjectives has survived only in standard Russian (in colloquial Russian and Russian dialects such adjectives are virtually absent), but these forms, which are derived only frorn sterns of perfective verbs, have additional alethic or, more rarely, deontic rneaning (cornparable to Eng­lish adjectives on -able). Cf., for instance, ustanovi-m-yj 'which can b.e established' (<= ustanovit'PF 'to establish'), opredeli-m-yj 'definable' (<= opredelit'PF 'to de­fine'),prenebrezi-m-yj 'disregardabIe, negiectabIe' (<= prenebrec'pF 'to disregard,

Passives as grammatical constructions 323

despise'). Undergoer-oriented participles based on -m- are fonned only from im­perfective sterns and have likewise remained "bookish"; cf. opredeljaem(yj) 'be­ing defined' (<= opredeljat' ipf. 'to define'), prenebregaem(yj) 'being despised, neglected' (prenebregat' ipf. 'to disregard, despise') etc. Their text frequency is considerably lower than the frequency of nlt-participles, and they are not only vir­tually unknown in West Slavic, but also in Ukrainian and Belarusian they became obsolete a long time ago. Here the function of marking a passive with imperfective verbs has been taken over by RM -fonns (see 3.1).

34. Passively used RM-forms occur only extremely rarely in certain dialects which have been in elose contact with Slavic varieties.

35. Notabene, the participle in Engl. chewing gum is Undergoer-oriented, contrary to the regular function of -ing in the continuous fonns of two-place verbs. This is a case of lexicalization with the same diachronic background which I have been elaborating on here for the Baltic and Slavic languages.

36. Analogous examples can be found in archaic Latvian dialects; cf. Endzelin (1922: 779).

37. There are some quite systematic exelusions to this case-conditioned rule, but these concem only two-place verbs goveming a genitive or instrumental object.

38. For Russian cf. Knjazev (1989: 97), for Lithuanian Geniusiene and Nedjalkov (I988: 376f., 379).

39. There are, in fact, some participles from intransitive sterns with these affixes, which are elearly resultative (e.g. verrottet 'decayed', abgefallen 'fallen off'). Such participles, that do not change the inherent macrorole orientation of the de­riving stern, are elear archaic remnants, i.e. cases of lexicalization.

40. Nor does Detges really bother about the relation to the lexicon Oexical units), as he restricts hirnself to a morpheme-based approach towards grammaticalization, for which principles of functional expansion may be deferred to processes after grammaticalization proper. The same is true with regard to Haspelmath (1990).

41. This elumsy way of circumscribing could in short be replaced by the Gennan tenn 'Vorgangspassiv'. The problem is, however, that the Gennan werden-passive is not restricted to denoting changes of state and punctual events so that it-can readily re­fer to ongoing processes and even states themselves (e.g. Das Auto wird gerade gewaschen 'The car is being washed', Unser Geld wird von uns selbst verwaltet 'Ourmoney is administered by ourselves'). This is impossible with the Polish aux­iliary zostac pf.lzostawac ipf.; Latvian tikt, however, quite readily combines with a t-participle of an activity (atelic) verb, and it can even combine with t-participles of state verbs (cf. Nau 1998: 38 and p.c.); see 3.3.2.

42. In.Old Polish we also eI!counter ostac sit:. with the same meaning. Both ostac sit:. and zostac are etymologically related to Russ. ostat'sja 'to remain', which how­ever did not shift from the continuative to the inchoative meaning and, therefore, did not show a tendency to develop into a passive auxiliary.

43. Bliven was also borrowed by Scandinavian, at a time when it already had both continuative ('to remain') and inchoative ('to become') function, but later on established itself in an exelusively inchoative value, in which, just as in Polish, it

324 Bjöm Wiemer

began to be used as passive auxiliary (bliva) in Swedish; cf. Weiss (1982: ibd.). 44. An elaborated formal description has most recently been given by Schwarze

(2003). 45. Example (21a) sounds strange for a contemporary ear, at least in isolation, be­

cause it raises the effect of a counterexpected, non-altered state. Some additional context would be required to make such an example sound more natural.

46. Combinations of imperfective zostawac with n/t-participles of imperfective verbs would be ungrammatical (e.g. *Zdj~cia zostajq (wlasnie) pokazywane ipf

.,

intended meaning: 'The photos are (just) being shown'). A sentence like Teraz zostajq pokazanePf. zdj~cia 'Now photos are shown' can be uttered only in the context of a narrative present. This is probably the main difference between Pol. zostac!zostawac and German werden, as the latter demonstrates hardly any re­strictions on passives denoting ongoing processes and can be used to denote states, because it by itself is aspectually indifferent.

47. Instead, the BE-copula (auxiliary) byc, an impersonal or an active construction would have to be used.

48. Schwarze (2003: 25) seems to imply just this, since with respect to It. rimanere he writes that reference to the event is made only by the passive participle, not by nmanere.

49. From this perspective, PoL zostac as an auxiliary behaves the same way as an in­choative copula with nouns: here it also denotes an inchoative event (e,), not the negation of a transition into a new state (e2). X zostal nauczycielem, wdowcem, etc. 'X has become (a) teacher, widower' etc. unanimously denotes an inchoative event and cannot be translated as 'remained a teacher' etc. (contrary to adjectives, see below). Therefore, in no case has zostac lost the potential to refer BY ITSELF to the change of state (e I).

50. It is remarkable that the combination of zostac with a participle "clusters" with nouns around the inchoative meaning (see preceding note), leaving apart the adjectives (rendering a continuative meaning), though the semantics of typical adjectives is certainly closer to the semantics of verbs than is the semantics of typ­ical nouns.

51. otb + GEN. most probably was an Gld Church Slavonic "import" into Gld East Slavic. od + GEN. in early West Slavic (Polish and Czech) may weIl have been in­fluenced by Latin and/or German, as may have been Pol. przez by German durch (both meaning literally 'through') .. But their agentive function was long-Iasting and had quite a stable place in the languages. In contrast to this, an agentive func­ti on of Bel. praz and East Slavic crezlcerez 'through' might have been calqued from Polish przez, but in the 10ng ron these prepositions did not show any stable tendencies of becoming a marker of the passive agent.

52. In a similar vein Keenan (1985: 261) argued that the case role of an 'agent phrase' may not at all be agentive.

53. For instance, the Latvian possessive genitive (or pronoun) begins to develop into an adnominal agentive modifier at the moment when the possessor no longer co­incides with the doer (cf. Holvoet 2001a: 374).

Passives as grammatical constructions 325

54. Both faetors were aeeounted for by Shibatani (1998), but he did not notiee that a eonfliet eould arise between them.

55. Beside this, whatever the answer to the question whether these processes are metaphorie or metonymie will be, it would refer, first of all, to the lexical evolu­tion of eertain semantie classes of nouns. This would not suffiee as an explanation why these ehanges are so erucial for grammar and, from this angle, ean be treated as steps in a grammaticalization process.

56. Notiee, by the way, that u + GEN. eould render a potentially agentive meaning (by implieature) with the RM-passive, too. Compare forinstanee U Olega vse bumagi terjajutsja/uniCtozajutsja/ szigajutsja, lit. 'At Oleg all doeuments get lost/are des­troyed/bumt'. Nonetheless, a potential agentive meaning of u + GEN. has been diseussed mainly in connection with partieipial eonstruetions that are closely eonneeted to Undergoer-oriented resultatives (see above). Trubinskij (I995: 82) explieitly denies that u + GEN. with aRM-form eould be interpreted agentively in NW-Russian dialeets.

57. ParalleIs ean be found in other aneient I.E. languages (cf. Fraenkel1928: 95), al­though in many eases (e.g. in Vedic) the genitival possessor oeeurred only as the . first member of eompound nouns (cf. Hettrieh 1990: 60).

58. Holvoet (1995: 178) treats this as aseparate step towards a passive eonstruetion. Sinee, however, the change from attributive to predicative use of participles is by no means a feature restricted to the (analytic) passive, but may occur, for instance, also with adjeetives, I will not consider it as aseparate step.

59. Cf. Hopper and Traugott (1993: 80) and a lot of other researchers involved in grammaticalization studies.

60. The same semantic shift has to be assumed for genitival modifiers of deverbal nouns, as in Holvoet's own example The Greeks' siege ofTroy.

61. We can neglect the fact that the clitics in West Slavic languages can also be used for marking true reflexivity and reciprocity, in which cases they do refer to dis­tinct roles of identical participants in the respective situations . .

62. This pertains also to so-called impersonal passives - an issue not taken up here.

References

Ambrazas, Vytautas 1990 Sravnitel'nyj sintaksis pricastij baltijskixjazykov [Comparative syntax

of participles in the Baltic languagesJ. Vilnius: Mokslas. 2001 Lietuviq kalbos pasyvo raidos bruozai [Outlines of the evolution of the

passive in Lithuanianl Acta Linguistica Lithuanica 45: 11-38. Brugmann, Karl -

1895 . Die mit dem Suffix -to- gebildeten Partizipia im Verbalsystem des Latei-nischen und des Umbrisch-Oskischen. Indogermanische Forschungen 5: 89-152.

326 Bjöm Wiemer

Dalrymple, Mary 2001 Lexical Functional Grammar. (Syntax and Semantics 34.) San Diego:

Academic.Press. Dankov, Vasilij N.

1981 Istoriceskaja grammatika russkogo jazyka (vyrazenie zalogovyx otnosenij u glagola) [Historical grammar of Russian (expression of voice in the verb)]. Moscow: Vyssaja skola.

Detges, Ulrich 2000 Time and truth: The grammaticalization of resultatives and perfects

within a theory of subjectification. Studies in Language 24 (2): 345-377. Eiche, Aleksandra

1983 Latvian Declinable and Indeclinable Participles. Their Syntactic Func­tion, Frequency and Modality (A Synchronie Study based on Latvian Fiction ofthe I960s and I970s). Stockholm: Almqvist & Wiksell Inter­national.

Endzelin, Jan 1922 Lettische Grammatik. Riga: Gulbis.

Estival, Dominique and Myhill, John 1988 Formal and functional aspects of the development from passive to

ergative systems. In: Passive and Voice, M. Shibatani (ed.), 441-491. Amsterdam/Philadelphia: John Benjamins.

Feoktistova, A. S. 1961 K istorii sostavnogo skazuemogo s prisvjazocnoj cast'ju, vyraZennoj

pricastiem stradatel' nogo zaloga prosedsego vremeni (na materiale Novgorodskix pamjatnikov pis'mennosti XII-XVI vv.) [On the his­tory of the complex predicate with a copular component expressed by a past passive participle (on the material of literary Novgorodian texts of the 12-16th centuries)]. In Issledovanija po leksikologii i gramma­tike russkogo jazyka, V. I. Borkovskij and S. I. Katkov (eds.): Moscow: . Izd-voAN SSSR, 194-208.

Fraenkel, Ernst 1928 Syntax der litauischen Kasus. Kaunas: Valstybes spaustuve.

Gadolina, Margarita A. 1963 . Istorija form licnyx i vozvratnogo mestoimenij v slavjanskix jazykax

[History of the personal and reflexive pronoun forms in the Slavic lan­guagesl. Moscow: Izd-vo AN SSSR.

Gäters, Alfreds 1977 Die lettische Sprache und ihre Dialekte. (Trends in Linguistics.

State-of-the-Art Reports 9.) The Hague: Mouton. 1993 Lettische Syntax (Die Da inas) , ed. by H. Radtke. Frankfurt/Main:

Gebauer, J an 1929

Lang.

Historiclai mluvnice jazyka ceskiho. Da IV: Skladba [Historical gram­marofCzech. Part 4: Syntax]. Prague: CeskaAkademie Ved a Umenf.

Passives as grammatical constructions 327

Geniusiene, Emma 1987 The Typology 01 Reflexives. (Empirical Approaches to Language Typ­

ology 2.) Berlin/New York: Mouton de Gruyter. Geniusiene, Emma and V. P. Nedjalkov

1988 Resultative, passive, and perfect in Lithuanian. In Typology olResultative Constructions, V. P. Nedjalkov (ed.), 3-69-386. (Typological Studies in Language 12.). AmsterdamlPhiladelphia: John Benjamins,

Grepl, Miroslav and Petr Karlfk 1998 Skladba cestiny [Czech syntax]. Olomouc: Votobia.

Gunnarsson, G. 1935 Studien über die Stellung des Reflexivs im Russischen. Uppsala: A.-D.

Lundequistska Bokhandeln. Haspelmath, Martin

1987 Transitivity Altemations 01 the Anticausative Type. (Institut für Sprach­wissenschaft. Arbeitspapier Nr. 5, Neue Folge.) Cologne: Universität zu Köln.

1990 The grammaticization of passive morphology. Studies in Language 14 (I): 25-72.

1994 Passive participles across languages. In Voice: Form and Function, B. Fox and P. J. Hopper (eds.), 151-177. Amsterdam/Phi1adelphia: John Benjamins.

1998 Does grammaticalization need reanalysis? Studies in Language 22 (2): 315-351.

2003 The geometry of grammatical meaning: Semantic maps and cross-lin­guistic comparison. In The new psychology ollanguage, M. Tomasello (ed.), 211-242. (Vol. 2.) Mahwah, NJ: Lawrence Erlbaum.

Havninek, Bohuslav 1937 Genera verbi v slovans/cYch jazycich II [Genera verbi in Slavic lan­

guagesl. Prague. Hettrich, Heinrich

1990 Der Agens in passivischen Sätzen altindogermanischer Sprachen.

Holvoet, Axel

Göttingen: Vandenhoeck & Ruprecht. (Nachrichten der Akademie der Wissenschaften in Göttingen I. Philologisch-historische Klasse, J g. 1990, Nr. 2.)

1994 Notes on the Latvian passive. Linguistica Baltica 3: 131-140. 1995 On the evolution of the passive in Lithuanian and Latvian. In: Analeeta

Indoeuropaea Cracoviensia, W. Smoczynski (ed.), 173-182. (I. Safa­rewicz memoriae dedicata.) Krakow: Universitas.

Holvoet, Axel 2001a Impersonals and passives in Baltic and Finnic. In Circum-Baltic Lan­

guages, vol. 2: Grammar and Typology, Ö. Dahl and M. Koptjevska­ja-Tamm (eds.), 363-389. (SLCS 55.) AmsterdamlPhiladelphia: John Benjamins.

328 Bjöm Wiemer

HoIvoet, Axel 2001b Studies in theLatvian verb. (Baltica Varsoviensia4.) Cracow: Wyd-wo

VJ. Hopper, Paul J., Thompson, Sandra A.

1980 Transitivity in grammar and discourse. Language 56 (1): 251-299. Hopper, Paul J., Traugott, Elisabeth Closs

1993 Grammaticalization. Cambridge: Cambridge Vniversity Press. Keenaq, Edward L.

1976 Towards a universal definition of 'subject'. In Subject and Topic, C. Li (ed.), 303-333. New York: Academic Press.

1985 Passive in the world's Ianguages. In Language Typology and Syntactic Description. Vol. 1: Clause Structure, T. Shopen (ed.), 243-281. Cam-bridge: Cambridge Vniversity Press. .

Kemmer, Suzanne 1993 The Middle Voice. (Typological Studies in Language 23.) Amsterdaml

Philadelphia: John Benjamins. Knjazev, Jurij Pavlovic

1989 Akcional'nost' i statal'nost' (Ix sootnosenie v russkich konstrukcijax s pricastijami na -n, -t) [Dynamic and static interpretation (Their mu­tual relation in Russian constructions with nlt-participles)]. (Specimina Philologiae Slavicae 81.) Munich: Sagner.

Krys 'ko, Vadim B. 1997 Istoriceskij sintaksis russkogo jazyka (Ob "ekt i perexodnost') [Histor­

ical Russian syntax (Objects and transitivity)]. Moscow: Indrik. Lehmann, Christian

1985 Grammaticalization: Synchronie variation and diachronie change. Lin­gua e Stile 20 (3): 303-318.

1995 Thoughts on Grammaticalization. (LINCOM Studies in Theoretical Linguistics 1.) Munich/Newcastle: Lincom Europa.

Lehmann, Volkmar 1992 Le preterit deictique et Ie preterit narratif en polonais modeme. In Lin­

guistique et slavistique. Melanges offerts Cl Paul Garde. M. Guiraud-­Weber and Ch. Zaremba (eds.):545~557. Aix-en-Provence.

Lomtev, Timofej P. 1956 Ocerki po istoriceskomu sintaksisu russkogo jazyka [Essays on Rus­

sian historical syntax]. Moscow: Izd-vo MGV. Meillet, Antoine

1948 L' evolution des formes grammaticales. In Linguistique historique et linguistique generale. A. Meillet (ed.), 130-148. Paris [Reprint from:

Nau, Nicole 1998

Scientia (Rivista di scienza) 12, 'XXVI, 6, 1912.]

Latvian. (Languages ofthe world/Materials 217.) Munich/Newcastle: LINCOM Europa.

Passives as grammatical constructions 329

Nedjalkov, Vladirnir Petrovic (ed.) .1988 Typology 0/ Resultative Constructions. (Typological Studies in Lan­

guage 12.) Amsterdam/Philadelphia: John Benjamins. Nedjalkov, Vladirnir Petrovic and Sergej Evgen'evic Jaxontov

1988 The typology of resultative constructions. In Typology 0/ Resultative Constructions, Nedjalkov, V. P. (ed.), 3-62. (Typological Studies in Language 12.) Amsterdam/Philadelphia: John Benjamins.

Pauliny, Eugen, Ruzicka, Jozef, Stole, Jozef 1968 Slovenskti gramatika [Slovak grammar). Bratislava: Slovenske peda­

gogicke nakladatelstvo. Pollard, Carl, Sag, IvanA.

1994 Head-Driven Phrase Structure Grammar. Chicago: University of Chi­cagoPress.

Rytel-Kuc, Danuta 1990 Niemieckie passivum iman-Sätze a ich przeklad w jt:.zyku czeskim i

polskim [The Gennan passive and man-clauses and their translation in Czech and Polish.1 (Slavica 84.) Wroclaw: Ossolineum.

Schwarze, Christoph 2003 'Bleiben' und 'werden'. Zur Polysemie von it. rimanere. In Kognitive

romanische Onomasiologie und Semasiologie, A. Blank and P. Koch (eds.), 19-32. Tübingen: Niemeyer.

Shibatani, Masayoshi 1998 Voice parameters. In Typology 0/ Verbal Categories (Papers pre­

sented to Vladimir Nedjalkov on the occasion 0/ this 70th birthday), L. Kulikov and H. Vater (eds.). 117-138. (Linguistische Arbeiten 382.) Tübingen: Niemeyer.

Silina, V.B. 1995 Glagol [The verb]. In Drevnerusskaja grammatika XII-XIII vv, V. V.

Ivanov (ed.), 374-506. Moscow: Nauka. SIStar: Slownik staropolski IX [Dictionary of Old Polish]

1982-7 XI-6 (2000). Cracow: Wyd-wo PAN. Sokolova, Miloslava

1993 Semantika slovesa a slovesny rod [The semantics of the verb and voice]. Bratislava: Veda.

Strel' cova, M.I. 1978 0 svjazi vido-vremennyx i zalogovyx znacenij v pricastijach na -m-,

-n-, -t- v drevnerusskomjazyke (po pamjatnikam XI-XIV vv.) [On the relation of aspecto-temporal and voice functions of participles with the suffixes -m-, -n-, -t- in Old Russian (on the basis of literary texts from the 11th-14th centuries)J. In Problemy teorii grammaticeskogo zaloga,V. S. Xrakovskij, V. S., 213-219. Leningrad: Nauka. .

Stroik, Thomas S. 2000 Syntactic Controversies. (LINCOM Studies in Theoretical Linguistics

24.) Munich/Newcastle: LINCOM Europa.

330 Bjöm Wiemer

Szlifersztejnowa, Salomeja 1968 Bieme czasowniki zaimkowe (reflexiva) w jt:.zyku polskim [Pronominal

(reflexivet verbs with passive meaning in Polish]. Wroclaw: Ossoli­neum, Wyd-woPAN.

Sticha, Frantisek . 1979 0 uzfvanf a vyznamu dvojf fonny trpneho rodu v soucasne spisovne

cestine [On the use and meaning of two types of expressions for the passive in contemporary standard Czech]. Nase fee 62( 1) 57-71.

1985 Pozmimky 0 staroceskem pasfvu [Remarks on the Old Czech passiv]. Listyjilologicke 108: 77-82.

Trubinskij, Valentin Ivanovic 1984 Oeerki Russkogo Dialektnogo Sintaksisa [Essays on Russian dialecta1

syntax]. Leningrad: Izd-vo LGU. 1995 0 novom russkom trexclennom passive [About a new three-partite pas­

sive in Russian]. In Sevemo-russkie govory 6, Gerd, A. S. (ed.), 78-84. Sankt-Peterburg: Izd-vo SPbGU.

Van Valin, Robert D. Jr., LaPolla, Randy J. 1997 Syntax: Structure, meaning and function. Cambridge: Cambridge Uni­

versity Press. Waltereit, Richard

2000 What it means to deceive yourself: The semantic relation of French re­flexive verbs and their corresponding transitive verbs. In Reflexives:

Weiss, Daniel

Forms and Functions, Z. Frajzyngier and T. S. Curl (eds.), 257-278. (Typological Studies in Language 40.) AmsterdamlPhiladelphia: lohn Benjamins.

1977 Syntax und Semantik polnischer Partizipialkonstruktionen (Im Rahmen einer generativ-transformationelien Sprachbeschreibung). (Slavica Helvetica 10.) Beme: Lang.

1982 Deutsch-polnische Lehnbeziehungen im Bereich der Passivbildung. In Literatur- und Sprachentwicklung in Osteuropa im 20. Jahrhundert, E. Reißner (ed.), 197-218, Berlin.

1999 Ob odnom predloge, sdelavsem blestjascuju kar' eru (Vopros 0

vozmoznom agentivnom znacenii modeli "u + imj~") [About a prep­osition with a splendid career behind it (On a possible agentive meaning of the model "u + noungen")]. In Tipologija i teorija jazyka: ot opisan­ija k ob ''jasneniju (K 60-letiju Aleksandra Evgen'eviea Kibrika), E. V. Raxilina and Ja. G. Testelec (eds.), 173-186. ,Moscow: Jazyki russkoj kul'tury.

Wiemer, Bjöm 1998 Puti grammatikalizacii inxoativnyx svjazok (na primere russkogo,

pol'skogo i litovskogo jazykov) [Ways of the grammaticalization of in­choativecopulae (the example ofRussian, Polish, and Lithuanian)]. In Lexikologie und Sprachveränderung in der Slavia, M. Giger, T. Menzel,

Passives as grammatical constructions 331

and B. Wiemer (eds.), 165-212. (Studia Slavica Oldenburgensia 2.) Oldenburg: BIS

2001 Aspektual' nye paradigmy i leksiceskoe znacenie russkix i litovskix glagolov (Opyt sopostavlenija s tocki zrenija leksikalizacii i gramma­tikalizacii) [Paradigms of aspect and the lexical meaning of Russian and Lithuanian verbs (A comparative attempt with regard to lexicaliza­tion and grammaticalization)]. Voprosy jazykoznanija 2001-2,26-58.

Wunderlich, Dieter 1993 Diathesen. In Syntax: Ein internationales Handbuch zeitgenössischer

Forschung (An International Handbook 0/ Contemporary Research), J. Jacobs, A. von Stechow, W. Sternefeld, and T. Vennemann (eds.), 730-747. Vol. 1. Berlin/New York: Mouton de Gruyter.

Zaliznjak, Andrej Andreevic 1995 Drevnenovgorodskij dialekt [The dialect of ancient NovgorodJ. Mos­

cow: Jazyki russkoj kul'tury.