The development and loss of the Old Irish double system of verbal inflection

THE DEVELOPMENT AND LOSS OF THE OLD IRISH DOUBLE SYSTEM OF VERBAL INFLECTION

Glenda Elizabeth Newton

Trinity College

This dissertation is submitted for the degree of Doctor of Philosophy

Cambridge University October 2006

DECLARATION

This dissertation is a result of my own work and includes nothing which

is the outcome of work done in collaboration. It does not exceed the word

limit of 80,000 words. The research reported in this dissertation was funded by a postgraduate

award from the Arts & Humanities Research Council.


Glenda Elizabeth Newton

SUMMARY This dissertation aims to provide a new syntactic account of the Old Irish double system of verbal inflection, its development from Proto-Indo-European (PIE) and its loss between Old and Modern Irish within the framework of Chomsky’s minimalist programme.

The verb in Old Irish has a different morphological ending depending on its position in the clause. In absolute initial position it has absolute inflection and in near-initial position it has conjunct inflection. Existing theoretical accounts of the Old Irish verbal system have proposed that the choice between absolute and conjunct verbal endings is dependent on the position of the finite verb in the syntax. However, a detailed examination of the Old Irish data suggests that this cannot be the case, as the finite verb appears to occupy the same syntactic position in all clauses. A new synchronic account is proposed whereby the different verbal forms are a result of the combination of syntactic and post-syntactic operations. The development of the double system has received much attention in the philological literature, yet the origin of the system remains highly disputed. This dissertation develops a new account combining the phonological aspects of Cowgill’s Particle Theory with a generative analysis of the development of verb-initial word order between PIE and Old Irish. It is argued that the development of unmarked verb-initial word order in Old Irish can be linked both to changes in the C system and to the development of the new verbal morphology. Turning finally to the loss of the double system, it is argued that this began before the Old Irish period. Syntactic changes in pre-Old Irish meant that the double system ceased to be syntactically productive and began to be derived post-syntactically. These post-syntactic operations were not productive, applying only in limited set of environments. On the basis of Middle and Early Modern Irish data it is argued that a decrease in evidence for the different forms led to a failure to acquire the environments in which these operations applied. As a result, the double system was lost from the language as a whole. In addition to providing a new account of the Old Irish verbal system, this dissertation touches on a number of theoretical issues. First, it considers the syntax-phonology interface and the types of operation that can take place there. Secondly, it gives an example of how head movement might develop, which provides a new perspective on the link between head movement and morphology. Finally, the loss of the double system provides an example of how, despite the necessarily abrupt nature of parametric change, its effects can be gradual.

ACKNOWLEDGEMENTS

Completing this thesis has been a huge task, which would not have been possible without the help and support of many different people. First, I would like to thank my supervisor, Ian Roberts. In addition to his insightful comments and suggestions for my research, Ian has always been incredibly encouraging, supportive and enthusiastic about my work. Working with Ian has been a hugely inspiring and enjoyable experience and I am immensely grateful for all the time and energy he has put into helping me research, write and complete this dissertation. David Willis, my PhD advisor, has also played an important role in my research and my linguistic career in general. In addition to introducing me to the study of historical linguistics, he has read and commented on various papers and always been happy to discuss any aspect of my research. For this I am very grateful. Thanks are also due to Adam Ledgeway for stepping in as the assessor for my second year interview. Theresa Biberauer has had a huge influence both on me and my thesis. As a teacher, Theresa was truly an inspiration. Her enthusiasm for linguistics in general and syntax in particular rubbed off on me from my first undergraduate supervision. Throughout my time as an undergraduate, an MPhil and a PhD student Theresa has always been there to offer me encouragement and support, both academically and personally. She has read and provided invaluable comments on virtually all of my work (including the whole of this thesis) and helped me to make some of my most difficult decisions. I am incredibly grateful to her for all of this and more besides. Turning to some more general thank yous, I would like to thank the staff and students (past and present) of the Linguistics Department for making it a great place to be. On the staff side I would like in particular to thank Kasia Jaszczolt, Mark Jones, Ariel Knapman, Peter Matthews, Francis Nolan and Rachel Smith who have contributed in various ways to my PhD and my general linguistic education. My MPhil and PhD colleagues also deserve a big thank you, most notably: Fiorien Bonthuis, Faye Chalcraft, Joanne Chapter, Lila Daskalaki, Andrew Janes, Kate Ketner, Eleni Kriempardis, Theo Markopoulos, Marios Mavrogiorgos, Alyson Pitts, Matthew Reeve, Christina Sevdali, Keith Shaw, Makis Sipetzis and Edward Wilford. Special thanks are due to my “PhD mentor” Marc Richards who has read and commented on various bits of my work and always been happy to help with matters of syntax and semicolons! Now, to the ASNaCs, without whom this work would not have begun. I am very grateful to Máire Ní Mhaonaigh and Oliver Padel for letting me onto the Celtic Philology course in 2000-2001, and for all their efforts in bringing my knowledge of Irish and Welsh up to speed. This course was the beginning of my fascination with all things Celtic. Máire and also Paul Russell have always been very helpful throughout my PhD in answering questions and offering advice on Celtic matters. I would also like to thank the Trinity Celticists, Mohan Ganesalingam and Geraldine Parsons for the many useful discussions we have had over the years.

I have received a lot of helpful feedback from participants at the various conferences I have attended, including the Postgraduate Conference in Linguistics at Manchester in March 2005, Lingo at Oxford in September 2005, ConSOLE at the University of the Basque Country in December 2005 and LAGB at Newcastle in September 2006. In particular, I would like to thank Nick Zair, who I met in Oxford, for our discussions on the joys of Old Irish, for help with Latin and Greek and for offering to proofread my thesis. I feel very fortunate to have been a member of Trinity College throughout my time in Cambridge. As an undergraduate and a postgraduate it has been a great place to live and work. I am particularly grateful for their financial support, both in terms of my Research Scholarship, and the money they have provided for attending conferences. Finally I would like to thank my non-linguistics friends and my family for keeping me sane over the last three years by providing many much needed distractions from the world of linguistics. In particular in this regard I would like to attempt to express my inexpressible gratitude to Ed. You have done so much. Thank you for everything. My world is a much happier one with you in it.

LINGUISTIC ABBREVIATIONS

The following abbreviations are found in the main text and the glosses of the linguistic examples:

1 first person NAS nasalization 2 second person NEG negative 3 third person NOM nominative case ABS absolute inflection P preposition ABSL absolutive case PASS passive ACC accusative case PERF perfect Adj adjective PF Phonological Form AdjP adjective phrase PIE Proto-Indo-European Adv adverb PL plural AdvP adverb phrase PP prepositional phrase Asp aspect PRED predicate AUX auxiliary PRES present C complementizer PRET preterite CONJ conjunct inflection PRT particle COND conditional PROG progressive COP copula PT prototonic CP complementizer phrase PTC particle DAT dative case PVB preverb D determiner Q question particle DM Distributed Morphology REL relative DP determiner phrase RM Relativized Minimality DT deuterotonic S sentence F feminine SG singular FIN finiteness SOV subject-object-verb order FOC focus SUBJ subjunctive FOR force SUFF suffixed pronoun FUT future SVO subject-verb-object order GEN genitive T tense HMC head-movement constraint TOP topic IE Indo-European TP tense phrase IMPF imperfect UG Universal Grammar INF infixed pronoun v light verb INST instrumental V verb INT interrogative V2 verb-second order LEN lenition VN verbal noun LF logical form VOS verb-object-subject order LHM Long Head Movement VP verb phrase LOC locative vP light verb phrase M masculine VSO verb-subject-object order N neuter X head category N noun XP phrasal category

TEXTUAL ABBREVIATIONS The following abbreviations are used in this dissertation to refer to texts. ACC Amra Choluimb Chille. W. Stokes (ed.) (1899). The Bodleian

Amra Choluim Chille. Revue Celtique 20, 270–271.

AM Audacht Morainn. F. Kelly (ed.) (1976). Dublin: Dublin Institute for Advanced Studies.

AMC Aislinge Meic Con Glinne: K. Jackson (ed.) (1990). Dublin: Dublin Institute for Advanced Studies.

AS Agallamh na Senórach: Colloquy of the Ancients. S. O’Grady (ed.) (1892) Silva Gadelica: a collection of tales in Irish, 2 vols. Vol 1, 94–233, vol 2, 101–265.

CC In Cath Catharda: The Civil War of the Romans. An Irish Version of Lucan's Pharsalia. W. Stokes (ed.) (1909). Leipzig: Hirzel

Corm Cormac’s Glossary. W. Stokes (ed.) (1962). Three Irish glossaries. London: Williams and Norgate.

Corp Gen Corpus genealogiarum Hiberniae. M. O’Brien (ed.) (1962). Dublin: Dublin Institute for Advanced Studies.

DiAChor Di Astud Chor. N. McLeod (ed.) (1992). Early Irish contract law. Sydney: Centre for Celtic Studies.

Il Iliad. A. Murray (1999). London: Harvard University Press. LL The Book of Leinster, formerly Lebar na Núachongbála. R.

Best, M. O’Brien & A. O’Sullivan (eds.) (1954–1983). Dublin: Dublin Institute for Advanced Studies.

LU Lebor na hUidre (Book of the Dun Cow). Best R. & O. Bergin (eds.)(1929). Dublin: Dublin Institute for Advanced Studies.

Ml The Milan glosses on the psalms, Milan. Thes i, 7–483. NE Noo Ean (St John’s Gospel reprinted from the Manx Bible

1819). London: British and Foreign Bible Society PKM Pedeir Keinc y Mabinogi. I. Williams (ed.) (1930). Cardiff.

RV Die Hymnen des Rigveda. T. Aufrecht (ed.) (1877) Bonn: Marcus.

Sg Glosses on Priscian, St Gall. Thes ii, 49–224. TBC (II) Táin Bó Cúailnge from the Book of Leinster. C. O’Rahilly (ed.)

(1967). Dublin: Dublin Institute for Advanced Studies. Thes Thesaurus Palaeohibernicus: a collection of old-Irish glosses,

scholia, prose, and verse. W. Stokes & J. Strachan (eds.) (1901–3) Dublin Institute for Advanced Studies.

TT Togail Troi. LL, 1063–1117. Wb Würzburg Glosses on the Pauline epistles. Thes i, 499–712.


ABBREVIATIONS CHAPTER 1 – INTRODUCTION 1

1. Introduction: aims and objectives 1 2. Background information 4

2.1 Irish 4 2.2 Theoretical Background 6

2.2.1 An introduction to the minimalist programme 7 2.2.2 The status of head movement 12

CHAPTER 2 – A SYNCHRONIC ACCOUNT OF THE OLD IRISH VERBAL SYSTEM 18

1. Introduction 18 2. The Old Irish verbal system: the basic facts 18

2.1 Simple verbs 19 2.2 Compound verbs 21 2.3 Infixed pronouns 23

3. Syntactic analyses of Old Irish 25 3.1 Old Irish as a residual verb second language 26 3.2 Two types of raising: Carnie, Harley and Pyatt (2000) 28 3.3 Are simple verbs in C? Empirical evidence 30

3.3.1 Object pronouns 30 3.3.2 Relative clauses 32 3.3.3 Interim Summary 35

3.4 Deriving compound verbs: a problem for syntactic accounts 36 3.4.1 Initial preverbs are in C 36 3.4.2 Head movement 38 3.4.3 XP movement 43 3.4.4 Remnant movement 46

3.5 Summary 49

4. The syntax-phonology interface and the Old Irish verbal system 50 4.1 The syntax-phonology interface 50 4.2 Post-syntactic movement and the Old Irish verb 52 4.3 A new post-syntactic account 60

4.3.1 Simple verbs and do-support: a cross-linguistic parallel 60 4.3.2 Compound verbs 74

4.4 A closer look at the Old Irish data 84 4.5 Summary 89

5. Conclusion 89

CHAPTER 3 – SYNTACTIC CHANGE 92 1. Introduction 92 2. Parametric change and the role of language acquisition 92

3. Parameters in the minimalist programme 95

4. How to set parameters 98

4.1 Cue-based models of language acquisition 98 4.2 A cue-based model of syntactic change 101

5. Conclusion 106

CHAPTER 4 – THE DEVELOPMENT OF THE DOUBLE SYSTEM 108

1. Introduction 108

2. Philological accounts of the development of the double system 108 2.1 Introduction 108 2.2 Particle theory (Cowgill 1975) 112 2.3 Suffix and infix deletion theory (McCone 1979) 114 2.4 Summary 120

3. Aspects of the syntax of Proto-Indo-European 120

3.1 Introduction 120 3.2 The structure of the left-periphery 121 3.3 Word order 129 3.4 Preverbs 136 3.5 Summary 138

4. From Proto-Indo European to Pre-Old Irish: Changes in the C-system

4.1 Introduction 139 4.2 The development of V-to-C movement: a syntactic reanalysis 140 4.3 Changes in the CP from PIE to Old Irish 144 4.4 From optional to obligatory verb fronting 152 4.5 From clitic to affix: changes in the inner C position 155 4.6 Summary 160

5. The development of preverbs 160

6. The development of conjunct particles 164 6.1 Negative particles 164 6.2 Interrogative particles 166 6.3 Conjunctions 168 6.4 Summary 174

7. Conclusion 174

CHAPTER 5 – THE LOSS OF THE DOUBLE SYSTEM 177

1. Introduction 177 2. Univerbation: the loss of the double system in the syntax 179

2.1 Introduction: what is univerbation? 179 2.2 ‘Univerbation’ within the clause: generalised V-to-T movement 182

2.2.1 The loss of final syllables & the loss of V-to-C movement183 2.2.2 Syntactic evidence and the loss of V-to-C movement 186

2.3 ‘Univerbation’ within the vP: generalised v-to-V movement 194 2.4 The development of post-syntactic rules 198

2.4.1 Deuterotonic compound verbs 198 2.4.2 Absolute verbal inflection 200

2.5 Summary 201

3. The simplification of compound verbs 202 3.1 Texts 203 3.2 How the deuterotonic/prototonic alternation was lost 204 3.3 The simplification of compound verbs: a theoretical perspective 207 3.4 The causes of simplification 208 3.5 The spread of the simplification 212 3.6 Summary 216

4. The loss of the absolute and conjunct 217

4.1 How the absolute/conjunct distinction was lost 217 4.2 The loss of the absolute/conjunct distinction: a theoretical account 223 4.3 Summary 226

5. Conclusion 227

CHAPTER 6 – CONCLUSION 229 REFERENCES 233

CHAPTER 1

INTRODUCTION

1. INTRODUCTION: AIMS AND OBJECTIVES

From a linguistic perspective Old Irish is a fascinating object of study because it

differs significantly from other Indo-European (IE) languages. Nowhere is this more

apparent than in the verbal system. First, Old Irish, like Modern Irish and Welsh,

shows unmarked verb-initial word order in both main (1) and subordinate clauses (2).

1. béoigidir in spirut in corp in fect so

vivifies.PRES.3SG the spirit the body the time this

‘The spirit vivifies the body now’ (Wb 13d7)

2. as-rubart día friu-som ara celebartis a sollumnu

say.PERF.3SG God to.3PL that celebrate.PAST.SUBJ.3PL his feasts

‘God said to them that they should celebrate his feasts’ (Ml 102d3)

Verb-initial word order is relatively rare not only within the IE language family but

across all the languages of the world.1 Secondly, Old Irish has a double system of

verbal inflection, whereby the inflectional ending found on the verb differs depending

on the position of the verb in the clause. When the verb is in absolute initial position it

has independent or absolute endings (3). When the verb is not in initial position, when

it is preceded by a so-called conjunct particle, it has dependent or conjunct endings

(4).

3. léicid-som cloich asa tailm

release.PRES.3SG.ABS-emph.part.3SG.M stone out-of-his sling

‘He releases a stone out of his sling…’ (LU 6210–6211)

4. cenid leci in metur…

although.NEG allow.PRES.3SG.CONJ the metre…

‘Although the metre does not allow…’ (Ml 30a10) 1 According to The World Atlas of Language Structures (Haspelmath, Dryer, Gil & Comrie 2005: 330) of the 1228 languages under consideration only 85 (7%) are VSO.

CHAPTER 1 – INTRODUCTION

2

Old Irish is the only Indo-European language where a fully productive system of this

kind is attested.2 The idiosyncrasies of the Old Irish verbal system raise three main

research questions: (i) how does the system function synchronically in Old Irish? (ii)

How did the system develop from Proto-Indo-European (PIE)? (iii) How was the

system lost between Old and Modern Irish? This dissertation aims to address all three.

Because the Old Irish verbal system is so different from those of other languages it

offers a significant challenge to the theory of Universal Grammar. If the principles

and parameters of Chomsky’s minimalist programme (Chomsky 1993, 1995a, 2000,

2001, 2004, 2005, 2006) reflect the syntactic knowledge present in all humans at

birth, then this theoretical framework should account equally well for Old Irish as for

English or French. The first aim of this dissertation is to establish that this is the case

and to develop a coherent synchronic account of the double system of verbal

inflection within the framework of minimalist syntax.

As no other early IE language has unmarked verb-initial word order or a double

system of verbal inflection neither of these features can be reconstructed for PIE.

However, Old Irish is clearly an IE language and so these two features must have

developed from PIE origins.3 The second aim of this dissertation, then, is to establish

how and why this development took place. This topic has received much attention in

the philological literature, yet so far no definitive account has been reached. This

dissertation aims to shed new light on the problem by approaching it from a different

perspective, namely that of generative syntax.

Although Modern Irish retains verb-initial word order, it no longer has a

productive double system of verbal inflection. The question of how the double system

was lost has not been tackled in the literature. The third aim of this dissertation is to

examine this issue and provide an account of the loss of the double system consistent

with a generative theory of syntactic change.

This dissertation hopes to contribute not only to our understanding of the history of

Irish but also to our understanding of syntactic change. Building on previous

diachronic work in a generative framework by David Lightfoot and Ian Roberts, this

dissertation aims to show how syntactic change should be modelled within the most 2 There are traces of a similar system in Old Welsh, but it is not clear whether these traces reflect a fully productive system. See Rodway (1998) for discussion. 3 Although see Vennemann (2002) who suggests that some of the idiosyncratic features of Celtic may originate from contact with Semitic languages.


3

recent version of the minimalist programme (Chomsky 2005, 2006) and what the

advantages of such an approach are.

To summarise, therefore, this dissertation aims to achieve the following objectives:

• To provide a synchronic account of the Old Irish double system of verbal

inflection within the framework of Chomsky’s minimalist programme.

• To show how syntactic change can be modelled within the most recent version

of the minimalist programme and what the advantages are of such an

approach.

• To provide an account of the development of the double system of verbal

inflection and verb-initial word order from PIE consistent with this framework

for syntactic change.

• To provide an account of the loss of the double system of verbal inflection

during Middle and Early Modern Irish.

The structure of the dissertation is as follows: chapter 2 provides an introduction to

the most salient features of the Old Irish verbal system, reviews the existing

generative accounts and develops a novel synchronic account of the double system in

the Old Irish period. Chapter 3 outlines the theoretical framework for the remainder of

the dissertation, considering how syntactic change should be modelled within the

most recent formulation of minimalism. Chapter 4 reviews the existing philological

literature on the development of the double system and then goes on to propose a

novel account based within a minimalist framework. Chapter 5 looks at the loss of the

double system.

The remainder of this chapter introduces some relevant background information on

Old Irish and the minimalist framework, which is assumed throughout this

dissertation.


4

2 BACKGROUND INFORMATION

2.1 Irish

The Celtic branch of the IE language family can be divided into two main groups,

Continental and Insular Celtic.4 The Continental Celtic languages, Gaulish, Lepontic

(Cisalpine Gaulish) and Celtiberian (Hispano-Celtic), are our earliest attested

evidence of Celtic, consisting mainly of inscriptions discovered in France, Italy and

Spain dating from the sixth century BC to the fourth century AD. The evidence for

Insular Celtic is later, with the earliest inscriptions dating from the late fourth century

AD (Russell 1995: 25). The Insular Celtic languages (the Celtic languages spoken in

the British Isles, and Breton) can be divided into two groups, Brittonic and Goidelic.

The Brittonic languages originated in mainland Britain, but with the arrival of the

Angles and the Saxons in the sixth and seventh centuries AD, the British speakers

were pushed into the western peninsulas and across the English Channel to northern

France, giving rise to the three modern Brittonic languages Welsh, Cornish and

Breton. The Goidelic languages developed from the language spoken in Ireland. Scots

Gaelic and Manx were established through the migration of Irish speakers from

Ireland to Western Scotland and the Isle of Man.5 The relationship between the Celtic

languages can be seen in the diagram below.

5. Proto-Celtic

Continental Celtic Insular Celtic

Celtiberian Lepontic Gaulish Goidelic Brittonic

Irish Scots Gaelic Manx Welsh Breton Cornish

As this dissertation is concerned with Irish let us consider the Goidelic branch in more

detail. The earliest evidence of Irish is found in inscriptions written in the Ogam

script.6 The majority of the Ogam stones date from the fifth and sixth centuries AD

4 The relationship between the Continental and Insular Celtic languages is somewhat controversial. See Russell (1995: 15–18) for an overview and further references. For the remainder of this dissertation it will be assumed, following McCone (1996), that there was an Insular Celtic stage. 5 See Jackson (1951) on the historical development of the Goidelic languages. 6 See McManus (1991) for more details.


5

and are mostly memorial inscriptions of the form ‘A, son/descendent of B’. Although

the Ogam stones have provided Celticists with much useful information on the

historical phonology of Irish, their formulaic nature means that they are of little use

for the study of morphology and syntax. In particular, for our purposes, the Ogam

inscriptions make no contribution to our understanding of the Irish verbal system as

they contain no verbs.

The language of the Ogam stones, the language of the fifth and sixth centuries, is

referred to as Primitive Irish.7 Old Irish, the focus of chapter 2 of this dissertation, is

the language of the seventh, eighth and ninth centuries. The main sources of Old Irish

are glosses in Latin manuscripts, which are collated, edited and translated in Stokes &

Strachan (1901–3). The most substantial of these are the Würzburg Glosses on the

Pauline Epistles (ca 750), the Milan Glosses on the Psalms (ca 800) and the St Gall

Glosses on Priscian’s Grammar (ca 850).8 For his Grammar of Old Irish, Thurneysen

(1946) draws primarily on the language of the glosses, as these glosses are found in

(more or less) contemporary manuscripts. There are many examples of what appear to

be Old Irish texts in later manuscripts such as the twelfth-century Lebor na hUidre

‘The Book of the Dun Cow’ (Best & Bergin 1929) and The Book of Leinster (Best,

O’Brien & O’Sullivan 1954–1983). However, this evidence is less reliable as it is not

clear to what extent the texts have been modified or modernised by later scribes. The

Old Irish evidence used in this dissertation comes primarily from Thurneysen’s

Grammar and the glosses.

Middle Irish is generally agreed to begin with the biblical poem Saltair na Rann

which is dated on non-linguistic grounds to the tenth century (ca 988 – Mac Eoin

1961a). There are many contemporary Middle Irish manuscripts, and so our study of

Middle Irish is not restricted in the way that Old Irish is. However, we face a different

problem. Middle Irish texts contain to varying degrees combinations of Old Irish and

Early Modern Irish forms alongside “monstrosities which never formed part of the

living language of Ireland” (Strachan 1904: 153). This tendency of Middle Irish

scribes to archaise makes it difficult to determine what the spoken language of the

period was like, and importantly for linguistic purposes, what changes had taken place

at a given point in time. For example, a text that appears on linguistic grounds to date 7 Russell (1995) refers to the Ogam inscriptions as Archaic Irish. Thurneysen (1946: 8–9) uses the term Archaic Irish to refer to early manuscript sources, i.e. pre AD 750. We will follow Thurneysen’s usage for the remainder of this dissertation. 8 These dates are taken from Thurneysen (1946) and are generally well accepted.


6

from the early Middle Irish period may have been written later, but by a well-trained

scribe, who knew how to archaise the forms ‘correctly’. Of course this does not mean

that Middle Irish texts cannot be used for linguistic study, but they must be used with

care.

The Early Modern Irish period began in 1200, with the formation of the Bardic

schools (Russell 1995: 27). A distinction is often made between Classical Modern

Irish, the highly stylised archaic language described in the Grammatical Tracts

(Bergin 1916–55) and used for Bardic poetry and Early Modern Irish, the language

found in the prose texts of this period. As a learned literary language, Classical

Modern Irish was no-one’s native tongue. The language of the prose tales, although

still archaic, is considered closer to the language spoken during this period, and so is

of greater use for grammatical study.

The main stages of Irish are summarised in the table below:

Table 1: The main stages of Irish

400–600 Primitive Irish

600–700 Archaic Old Irish

700–900 (Classical) Old Irish

900–1200 Middle Irish

1200–1600 Early Modern Irish

1600–Present Modern Irish

2.2 Theoretical background

The analyses of the Old Irish verbal system and the changes it underwent that are

considered in this dissertation are based within the framework of generative grammar,

specifically the latest version of Chomsky’s principles and parameters framework, the

minimalist programme (Chomsky 1993, 1995a, 1995b, 2000, 2001, 2004, 2005,

2006). This section has two aims. The first is to provide an introduction to the basic

architecture and assumptions of minimalist syntax for non-specialists.9 This is the

topic of section 2.2.1. Section 2.2.2 considers the status of head-movement, an

important concept in the chapters to follow, within minimalist syntax.

9 For a more detailed introduction to minimalist syntax see Adger (2003).


7

2.2.1 An introduction to the minimalist programme

The main difference between minimalism and earlier models of generative grammar

is the focus on economy and simplicity. The minimalist programme works from the

assumption that language is an optimal system (the ‘Strong Minimalist Thesis’ see

Chomsky 2000, 2001 for detailed explanation of this view). If language is optimal it

will contain only components and operations that are “virtually conceptually

necessary”. As the primary function of the language faculty, in Chomsky’s view

(Chomsky 2002: 105–109), is to link sound and meaning, the parts of the language

faculty that are needed for this are (virtually) conceptually necessary; those parts that

are not are imperfections. This idea can be seen in the minimalist conception of the

architecture of the language faculty, shown in (6) below. Minimalism eliminates the

intermediate representational levels of earlier versions of principles and parameters

theory, maintaining that the only relevant levels of representation are the interfaces

with phonology (PF) and semantics (LF).

6.

Lexicon

syntactic derivation (‘narrow syntax’)

Spell-Out

PF LF

In addition to the interface levels, the language faculty has two further components, a

lexicon containing all the lexical items for the language in question and a

computational system (CHL), which puts these lexical items together to build syntactic

structures. Crucially, the lexicon differs from language to language, but CHL is

considered invariant cross-linguistically (Chomsky 1995a: 170). Let us consider each

of these components in turn.

The lexicon consists of two types of lexical item, substantive (or lexical) categories

such as nouns, verbs and adjectives, and functional categories. Chomsky (2000: 102)

proposes that the core functional categories are “C (expressing force/mood), T


8

(tense/event structure) and v (the “light verb” the locus of argument structure)”. 10 11

Both substantive and functional lexical items are stored as bundles of features,

encoding semantic, phonological and grammatical information. For the purposes of

narrow syntax, only the grammatical features are relevant, with semantic and

phonological features not playing a role until the derivation reaches the interfaces.12

The primary function of CHL is to build syntactic structures through the operation

Merge (Chomsky 1995a: 226). Lexical items are taken from the lexicon and placed in

a numeration or lexical array.13 CHL then takes these lexical items and merges them in

a pairwise fashion to build a sentence. So, if we take the sentence ‘John kisses Mary’,

the numeration (ignoring any functional structure) contains the lexical items [John],

[kisses] and [Mary]. The computational system takes the verb [kisses] and the object

[Mary] and merges them to form a verb phrase (VP):14

7. VP

kisses Mary

The computational system then merges the subject [John] with the VP to form the

structure below, where the verb kisses is the head of the VP, the object Mary is the

complement and the subject John is the specifier:15

10 Light v has been assigned many different functions since its introduction by Chomsky (1955/1975). Since Chomsky (1995a: 312–16) it has been assumed (although not universally) that v is present in all clauses except unaccusatives, passives and raising contexts, being the locus for Accusative Case and the external theta role. 11 These core functional categories can be seen as umbrella terms for the wider variety of categories found cross-linguistically (Chomsky 2005: 17). See Pollock (1989), Cinque (1999) on the articulated TP and Rizzi (1997) on the articulated CP. 12 It is generally assumed that phonological features play no role in narrow syntax. Under a Distributed Morphology view (Halle & Marantz 1993, 1994) this results from the fact that phonological features are added only at the point of Spell-Out. We will return to this idea below. 13 See Chomsky (2004: 107) on the difference between these two concepts. 14 The use of labels such as V’ and VP are associated with X-bar theory rather than the minimalist Bare Phrase Structure (Chomsky 1995b, Chametzky 2000). However, these labels will be retained throughout this dissertation for clarity of exposition. They should not be taken to have any theoretical significance. 15 The functional structure is omitted from this diagram for expository reasons. As discussed above, argument structure is associated with light v, so v would need to be merged before the subject is merged.


9

8. VP

John V’

kisses Mary

The operation outlined above is called External Merge as it merges lexical items from

the numeration, which is external to the derivation. However, it is also possible to

copy a lexical item that has already been merged, or a syntactic structure that has

already been formed and remerge it higher in the structure. This operation is Internal

Merge or Move. For example, in a passive clause, such as ‘Mary is kissed’, the

derivation begins in the same way as before. The verb kiss and the object Mary are

merged together from the numeration. Then the auxiliary verb (a T element) is

merged. The object Mary is copied and remerged in the subject position (Spec-TP).

9. TP

Mary T

is VP

kissed tMary

In passive clauses, then, the object moves to become the subject of the clause.

Descriptively this is reasonably uncontroversial; however, it is not explanatory. We

need an explanation as to why this movement takes place. In the case of passives, the

object is forced to move because the clause needs a subject. In English every clause

must have a subject – this is the Extended Projection Principle (EPP – Chomsky

1981). In minimalist terms, T in English has an EPP-feature and so its specifier

position must be filled. When nothing is merged there from the numeration, some

element will be moved from within the existing syntactic structure to fill this position.

Within the minimalist programme the notion of EPP is extended such that all


10

movement operations are a result of EPP-features (Chomsky 2000: 102–3).16 17

Within the lexicon each functional category is specified as being [±EPP]. Different

word orders in different languages are a result of different categories having different

EPP values; therefore, whether a given functional category is [±EPP] will vary from

language to language. We will return to this issue in some detail in chapter 3.

So whether or not movement takes place is determined by the presence of an EPP-

feature on the functional category; however, this does not explain which constituent is

moved. In order to explain this we must introduce a new operation: Agree (Chomsky

2000, 2001). When a lexical item is taken from the lexicon it has some features that

are intrinsically valued and some that are unvalued. If we consider a verb, for

example, it has no intrinsic values for tense or person and number agreement. These

values are determined by the clause in which the verb appears. This suggests that

these features are valued during the syntax. Chomsky argues further that all unvalued

features must be valued during narrow syntax due to the principle of Full

Interpretation (Chomsky 1995a: 194). Unvalued features cannot be interpreted by the

interfaces, and so if any features remain unvalued at the end of narrow syntax the

derivation will crash, i.e. be ungrammatical. Under this view, then, the Agree

operation is crucial.

To initiate an Agree relation, a category with an unvalued feature, the Probe,

searches for a category with a matching (identical) valued feature, the Goal. The Goal

must be in the domain of the Probe, i.e. it must be contained within the

sister/complement of the Probe. Moreover, the Goal must be the closest possible

target, i.e. no other categories with matching features can intervene between the Probe

and the Goal. Furthermore, the Goal must be active (Chomsky 2001: 6). A Goal is

only active if it has unvalued features of its own. This ensures that the Agree relation

is reciprocal, with the unvalued features of both the Probe and the Goal receiving

values. If these conditions are met, then Agree can take place, and the unvalued

features of both the Probe and the Goal are valued.18

16 See the following section on the distinction between head- and phrasal-movement. 17 Chomsky (2005) proposes that phase heads are assigned an Edge Feature (EF) that can trigger movement. The relationship between EF and EPP is not entirely clear, see Chomsky (2005: 22–3; 2006 for discussion). 18 Chomsky (2001: 5) argues that unvalued features are intrinsically uninterpretable and remain so after they have been valued. As a result, after a feature is valued it is deleted. This idea of deletion is problematic. Although certain unvalued features are uninterpretable at LF they may still be realised at PF. For example, subject agreement on the verb is uninterpretable at LF, but is realised phonologically in English as the suffix -s. If this feature were deleted in narrow syntax as part of the Agree operation


11

Feature valuation and the operation Agree build on earlier ideas about feature

checking (Chomsky 1993, 1995a). However, there is one crucial difference. In earlier

versions of the minimalist programme movement and feature checking were

intrinsically linked such that the only way to check and eliminate uninterpretable

features was through movement. In the latest formulation of minimalism (Chomsky

2000, 2001, 2004, 2005) this is no longer the case. Feature valuation via Agree can

take place in situ, independently of movement. An Agree relation will only result in

movement if the Probe, the target of movement, has an EPP-feature. Although Agree

and feature valuation is a necessary condition for movement, it is no longer sufficient.

There is one more fundamental difference between the recent version of

minimalism and its predecessors. Chomsky (2000, 2001) proposes that derivations

proceed cyclically, via phases. Under this system, only a restricted amount of the

structure, specifically one phase, is available to CHL at any one time. This has the

result of reducing computational complexity. The introduction of the concept of phase

has interesting implications for the operation Agree and, by association, Move. As

only a certain amount of structure is available to the computational system at one

time, the space over which a Probe can search for a Goal is limited. This has the effect

of imposing locality restrictions on Agree and Move. Let us consider how this works.

According to the Phase Impenetrability Condition (PIC) (Chomsky 2000: 108,

2001: 13–4), once a phase is complete, the domain (complement) of the phase head is

spelled-out and sent to the interfaces, leaving only the phase head and its specifier(s)

available for syntactic operations (i.e. Agree and Move). On the standard assumption

that CP and vP are phases, once the vP phase is complete, its domain/complement VP

will be sent to Spell-Out, and so will no longer be available. If any material from

within the VP, e.g. the verb or the object, is to move to a higher position in the clause

(e.g. T or C) it must move to either the head position (v) or Spec-vP before the phase

is complete to escape the phase. We will return to the concept of phases and the PIC

in more detail in chapter 2.

Having outlined the architecture and operations of the minimalist programme that

will be assumed throughout this dissertation, let us now turn our attention to a specific

theoretical issue that will play an important role in what is to follow, namely the

concept of head-movement and its status within minimalist syntax. then it could not have any effect at PF. It seems more likely that feature deletion takes place at the interfaces.


12

2.2.2 The status of head-movement within the minimalist programme

In the previous section it was argued that syntactic structures are built by the

operations Merge and Move. In this section we will consider the operation Move in

more detail. Two types of movement are empirically attested, movement of an entire

phrasal category, an XP, and movement of the head of a phrase. The existence of XP-

movement as a syntactic operation is widely accepted. Head-movement, on the other

hand, is more controversial. In recent work Chomsky (2000, 2001) has argued that

head-movement plays no part in narrow syntax and is in fact a PF phenomenon.

Chomsky invokes both empirical and conceptual arguments to support his claim. Let

us examine each in turn.

Chomsky’s main empirical argument against head-movement as a syntactic

phenomenon is that it has no semantic effects (Chomsky 2001: 37). If head-movement

took place in narrow syntax we would expect it to play a role at LF; however,

Chomsky argues, this does not seem to be the case. A verb, for example, tends to be

interpreted the same way in languages that show verb-movement to T or C and those

where the verb remains in situ. Matushansky (2006: 103) argues that this is

unsurprising as head-movement involves movement of a predicate and predicate

movement does not tend to have any semantic effects. We should only expect to find

semantic effects with head-movement that involves quantified heads, such as modal

verbs. There is ample evidence to show that movement of these heads does in fact

show semantic effects.

Roberts (2005: 148–9) argues that the possible interpretations of the English modal

may differ depending on whether or not it has undergone head-movement. In the T

position may is ambiguous between a permission and an epistemic interpretation,

(19a). If the modal is moved to the C position to form a question, it can only have the

permission reading, (19b).

19. (a) John may (well) leave permission/possibility

(b) May John (?*well) leave permission only

(Roberts 2005: 149)


13

Matushansky (2006: 104) observes along similar lines that different modal verbs can

have different scope properties. In example (20a) negation has scope over the modal

can whereas in (20b) the modal should has scope over negation.

20. (a) Yolanda can’t leave Neg > Mod

(b) Yolanda shouldn’t leave Mod > Neg

(Matushansky 2006: 104)

Both modals in (20) occupy the same structural position, however, they are

interpreted differently. Only in (20b) does the interpretation correspond to the surface

scope. In (20a), assuming that movement of the negative element is either impossible

(see Lechner 2005 for empirical arguments) or will have no semantic effect, the

interpretation corresponds to the reconstructed position of the modal. In (20a) scope is

determined on the basis of the position of the modal before head-movement, whereas

in (20b) it is determined according to the position of the modal after head-movement.

This suggests that head-movement can have semantic effects.

Lechner (2005) provides similar evidence from scope splitting constructions. In the

example in (21) below the modal can can take scope over the quantified subject,

splitting negation from its surface host.

21. Not every pearl can be of average size ¬◊ > ∀

(it is not possible that every pearl is of average size)

(Lechner 2005: 3)

Due to restrictions on where the subject can be interpreted, the only way to obtain this

reading, Lechner argues, is for the modal to be interpreted in its moved position. In

simplified terms, on the assumption that the subject originates in Spec-vP, and the

modal originates in light v, the only way that the modal will be able to have scope

over the subject is if the modal is interpreted in its moved position, T, and the subject

is interpreted in its base position, Spec-vP (see footnote 15).19

Matushansky (2006: 104–5) argues that head-movement also has syntactic effects.

If head-movement feeds syntactic processes, then it must be part of narrow syntax and 19 See also Lechner (2004) for evidence from comparatives to suggest that V2 cannot be a PF phenomenon.


14

not phonology. For example, Chomsky (2004), building on work by Marantz (1997)

proposes that verb-movement to v is necessary to assign a syntactic category to an

otherwise acategorial root. This seems to suggest that this movement should take

place in the syntax. Furthermore, Harley (2004), following among others Chomsky

(1993) and Bobaljik (1995), argues that since Scandinavian object shift, a syntactic

movement operation, depends on head-movement (cf. Holmberg’s Generalisation),

head-movement must also be syntactic.20

So, it seems that there is evidence to suggest that head-movement can have both

semantic and syntactic effects, and so cannot be simply a PF phenomenon; at least

some head-movement must take place in narrow syntax. Furthermore, Matushansky

(2006) argues that the analysis of head-movement as a phonological process creates

more problems than it solves. The main problem with such approaches, as observed

by Zwart (2001), is that we have no real understanding of how head-movement as a

phonological process would work. If head-movement is phonological it must be

determined what triggers the movement and how it is constrained. If it is a

phonological process, then we would expect it to be triggered by phonological

properties of the heads involved. Matushansky (2006) argues that this does not seem

to be the case. V-to-C movement in Germanic V2 clauses only takes place when C is

unfilled. However, V-to-T movement in French, according to Pollock (1989),

involves movement to a filled head position. It seems that for head-movement to take

place, the target can be either realized or null suggesting that the phonological

properties of a head position cannot influence head-movement.

Matushansky (2006: 99–100) argues that analysing head-movement as a

phonological phenomenon faces an even more serious problem, in that it violates the

PIC. As noted above, when a phase is complete the complement of the phase head is

sent to PF and unavailable for further syntactic operations. So, for example, when the

vP phase is complete the complement of v, i.e. VP, is spelled-out. In a language such

as French, where the verb moves to T (Pollock 1989), in order to conform to the

Head-Movement Constraint (Travis 1984 – HMC), the verb must move via light v.21

The crucial question is, when does V-to-v movement take place? If head-movement is

a phonological operation then it must take place after Spell-Out. However, at the end

20 Matushansky expresses the reservation that all of these syntactic effects of head-movement are theory-dependent and so cannot be decisive. 21 The status of the HMC in minimalist syntax is discussed below.


15

of the vP phase, only the VP is spelled-out; the light v head is still part of the syntax

and so is unavailable for movement. If head-movement takes place after the

completion of the CP phase then the phonological computation must deal with two

phases at once. This defeats the original purpose of introducing phases, namely to

decrease computational burden. Matushansky argues that what this shows is not that

head-movement must always occur before Spell-Out, but that if a head must move

across a phase boundary then it must do so in narrow syntax. Therefore, head-

movement cannot be purely a PF phenomenon.

From the arguments presented above it seems that not only is Chomsky’s main

empirical argument for abolishing syntactic head-movement invalid, but attempting to

account for head-movement as a phonological process creates more problems than it

can solve. This suggests that head-movement should be considered a syntactic

operation on a par with phrasal-movement.22 Let us turn now to Chomsky’s

conceptual arguments. Head-movement differs from phrasal-movement in a number

of ways, suggesting that it should not be part of narrow syntax.

One crucial way in which head-movement differs from phrasal-movement is that

head-movement violates the Extension Condition (Chomsky 2000: 136–8). The

Extension Condition states that the structure building operations Merge and Move

must always apply at the root, the highest point of the tree. When a head moves it

adjoins to a head position and not to the root and so does not extend the tree. This

means that unlike phrasal-movement, a moved head does not c-command its trace.

Within the minimalist literature, two main attempts have been made to account for

this distinction, namely Matushansky (2002, 2006) and Roberts (2005).

Matushansky and Roberts both point out that head-movement can be argued to

obey the extension condition if it in fact consists of two operations, one syntactic and

one morphological. Within the syntax head-movement proceeds in the same way as

phrasal-movement, targeting a specifier position. So, for example, in languages such

as French where the verb moves to T, the verb in fact targets Spec-TP, rather than the

head position.23 The syntactic movement operation is followed by a morphological

22 A further option that has been proposed to banish head-movement from narrow syntax is to suggest that all instances of head-movement should be reanalysed as remnant vP movement (Mahajan 2000, 2003; Boeckx & Stjepanovic 2001). Although in some cases there is good evidence for vP remnant fronting, it seems unlikely that it can replace head-movement across the board. This issue will be discussed in chapter 2, and so will not be dealt with in any detail here. 23 Spec-TP is the position typically associated with the subject (see section 2.2.1 above). However, according to Bare Phrase Structure (Chomsky 1995b) it is possible to have multiple specifiers. To


16

operation, m-merger (Matushansky 2002, 2006) or incorporation (Roberts 2005),

which merges two adjacent heads to form a single constituent. As the syntactic

movement operation is identical to phrasal-movement, a moved head will obey the

extension condition and c-command its trace in the syntax.24

If head-movement involves a combination of syntactic and morphological

operations it must be determined what drives these operations and when they can take

place. This is most problematic in terms of the m-merger/incorporation. The operation

m-merger converts two heads in a spec-head configuration into a single head

(Matushansky 2006: 94). Does m-merger affect all cases of this spec-head

configuration? This seems unlikely. If not, however, then there must be some way of

determining which cases will be affected and which will not.25 Roberts (2005)

proposes that his Incorporation operation will take place as a result of an Affix-feature

on the Probe. This Affix-feature is a type of EPP-feature that attracts a head rather

than a phrasal category. When the derivation reaches PF, this same Affix-feature

provides an instruction to incorporate the two heads. One single feature, then,

motivates both the syntactic movement and the morphological operation.26

We have seen in this section that there are good empirical and theoretical reasons

for allowing head-movement to be part of the syntax, and so for the remainder of this

dissertation we will assume that this is the case. Head-movement will be treated on a

par with XP-movement. It will be assumed that both processes are a result of Agree

and (re-)Merge.27 The difference between them is simply a matter of how much

material is pied-piped. This depends on the type of EPP-feature present on the Probe.

Following Roberts (2005) we will assume that head-movement is motivated by an

Affix-feature, a subtype of EPP-feature, that specifies that the head rather than an XP

must be moved. However we will leave the exact nature of this head-movement, i.e.

obtain the correct order in French, the verb must target the inner specifier and the subject the outer specifier. This being the case, an explanation is needed as to why head-(verb)-movement precedes XP-(subject)-movement. See Matushansky (2006: 82–3) on this issue. 24 This morphological operation is also argued to account for incorporation effects, which differentiate head-movement from XP-movement. 25 Marantz’s (1988) concept of morphological merger faces similar problems. 26 Matushansky’s m-merger operation does not drive the movement operation. She argues that head-movement is triggered by c(ategorial)-selection (Matushansky 2006: 75–77). This accounts for the fact that c-selection and head-movement show the same locality restrictions, but says nothing about the link between head-movement and morphology. 27 See Pesetsky & Torrego (2001) who also argue that head-movement and XP-movement are two sides of the same phenomenon.


17

whether it involves movement directly to a head position or combines movement to a

specifier and m-merger/incorporation, for further research.

One final issue that must be dealt with is the apparent locality conditions on head-

movement. Head-movement seems to be more local than XP-movement, as head-

movement cannot skip intervening heads. It is unclear, however, what role the HMC

should play in the minimalist programme. Such a specific condition does not seem in

keeping with minimalist ideas. It would be preferable if the HMC could be explained

in independent terms. If as argued above, head-movement is based on Agree in the

same way as XP-movement, then it must also be subject to the PIC. Unlike the HMC,

the PIC is independently motivated (see Chomsky 2001 et seq; Richards 2004). It will

be assumed here, then, that the PIC is the only formal locality constraint operating on

head-movement.

If the PIC is the only locality operation operating on head-movement, then how are

we to account for the locality effects that the PIC does not cover? For example, the

PIC will ensure that verb-movement always goes through light v to escape the phase,

however, it will not force the verb to move through T on its way to C. If, as argued in

section 2.2.1, movement always occurs as a result of Agree and the presence of EPP-

features, then for a head to move via an intervening head position, this position must

have the requisite features. For V-to-C movement to proceed via T, then, T must have

an Affix-feature that attracts the verb. This accounts for the effects of the HMC in

minimalist terms. However, under this view, it is also plausible that the HMC could

be flouted. If T did not contain the appropriate features, then the verb could move

directly from v to C (under the right version of the PIC – see the discussion in chapter

2) without moving via T. There is evidence to suggest that this may be the case in the

Continental Germanic languages, where although the verb moves to C in verb second

clauses, there is no evidence that it ever moves to T (Holmberg & Platzack 1995). It

will be argued in chapter 5 that this is also true for the history of Irish.

Having outlined the theoretical background for the analyses that are to follow, let us

now turn to the main subject of this dissertation: the Old Irish verbal system.

CHAPTER 2

A SYNCHRONIC ACCOUNT OF THE OLD IRISH VERBAL SYSTEM

1. INTRODUCTION

Before we can consider the diachronic aspects of the development and loss of the

double system of inflection, it is first necessary to gain an understanding of how the

system functions in Classical Old Irish, the language of the eighth and ninth centuries.

The aim of this chapter, then, is to provide a synchronic account of the double system

of verbal inflection within the framework of Chomsky’s minimalist programme.

A necessary prerequisite for the development of a generative account is an

understanding of the data under discussion. Section 2 provides an introduction to the

main features of the Old Irish verbal system. Of course this dissertation is not the first

work to attempt an analysis of the Old Irish double system within a generative

framework. Previous accounts have attempted to show that the different

morphological forms shown by the verb in Old Irish reflect different syntactic

structures. Section 3 reviews these previous syntactic accounts and argues that they

are faced with serious empirical and theoretical problems. As a result, it seems that

the different verbal forms cannot be a result of syntax alone. Section 4 investigates

how syntax and phonology can be combined to explain the different verbal forms, and

presents a new synchronic account of the double system of verbal inflection in Old

Irish.

2. THE OLD IRISH VERBAL SYSTEM

Like Modern Irish and the other modern Celtic languages,1 Old Irish has unmarked

verb-initial word order. The finite verb is in initial position in both main and

embedded clauses as can be seen in the examples below.

1. béoigidir in spirut in corp in fect so

vivifies.PRES.3SG the spirit the body the time this

‘The spirit vivifies the body now’ (Wb 13d7)

1 Except maybe Breton, which is argued to be a verb-second language (Stephens 1982; Borsley, Rivero & Stephens 1996; Borsley & Kathol 2000; Wilford 2005).

CHAPTER 2 – A SYNCHRONIC ACCOUNT

19

2. as-rubart día friu-som ara celebartis a sollumnu2

say.PERF.3SG God to.3PL that celebrate.PAST.SUBJ.3PL his feasts

‘God said to them that they should celebrate his feasts’ (Ml 102d3)

However, Old Irish differs from its modern counterpart in a number of ways. First,

Old Irish has a double system of inflection, whereby the verb has a different

morphological ending depending on its position in the clause. Secondly, in addition to

the simple verbs found in Modern Irish, Old Irish also has a number of compound

verbs consisting of a verb and one or more preverbs. Thirdly, Old Irish has enclitic

object pronouns that appear in second position. This section will examine each of

these features in turn.

2.1 Simple verbs: absolute and conjunct

Simple verbs in Old Irish consist of a verbal stem and an inflectional ending. Each

verb has five stems, which distinguish the present indicative, future, present

subjunctive, preterite active and preterite passive tenses. Further tense and aspect

distinctions can be achieved through the combination of these stems with the so-

called secondary inflectional endings. Combining secondary endings with the present

stem gives the imperfect, with the future stem gives the conditional and with the

present subjunctive gives the past subjunctive. The table below shows the weak verb

léicid ‘lets, allows’.3 Reading across the table, each row has the same (or very similar)

endings. Reading down the table, each column shows the same stem.4

2 In this example the conjunction aran ‘in order that’ functions as a generic ‘that’ complementizer. This usage is not very frequent in Old Irish. See Ó hUiginn (1997) for discussion. 3 In Old Irish the 3sg present indicative is traditionally used as a citation form instead of the infinitive. 4 See Thurneysen (1946: 326–440) and McCone (1997a) for more details on stems and endings.


20

Table 1: Conjugation of the weak verb léicid ‘lets, allows’

primary tenses Present future present subjunctive

preterite active

preterite passive

1sg 2sg 3sg 1pl 2pl 3pl

léiciu léici léicid léicmi léicthe léicit

léicfea léicfe léicfid léicfimmi léicfide léicfit

léicea léice léicid léicmi léicthe léicit

léicsiu léicsi léicis léicsimmi ------- léicsit

------- ------- léicthe ------- ------- léicthi

secondary tenses Imperfect conditional past subjunctive 1sg 2sg 3sg 1pl 2pl 3pl

no léicinn no léicthea no léiced no léicmis no léicthe no léictis

no léicfinn no léicfeda no léicfed no léicfimmis no léicfide no léicfitis

no léicinn no léicthea no léiced no léicmis no léicthe no léictis

The combination of stems and inflectional endings to mark tense, aspect and subject

agreement in Old Irish is similar in many ways to the verbal systems of other early

Indo-European (IE) languages. However, there is one significant feature of the Old

Irish verbal system that differentiates it from all the other early IE languages, namely

its double system of verbal inflection. In each tense for each person there are two

possible inflectional endings: the absolute and conjunct. These are shown in the

paradigms below for the verbs léicid ‘lets, allows’, marbaid ‘kills’ and berid ‘bears,

carries’.5

Table 2: Absolute and conjunct inflection

Absolute Conjunct Absolute Conjunct Absolute Conjunct

1sg léiciu -léiciu 1sg marbu -marbu 1sg biru -biur

2sg léici -léici 2sg marbai -marbai 2sg biri -bir

3sg léicid -léici 3sg marbaid -marba 3sg berid -beir

1pl léicmi -léicem 1pl marbmai -marbam 1pl bermai -beram

2pl léicthe -léicid 2pl marbthae -marbaid 2pl beirthe -berid

3pl léicit -léicet 3pl marbait -marbat 3pl berait -berat

The distribution of the absolute and conjunct endings is determined by the position of

the verb in the clause. When the verb is in absolute initial position it has absolute

5 The only exceptions to this are the secondary tenses, which have only conjunct inflection.


21

inflection, as in (3). When the verb is preceded by a so-called conjunct particle it has

conjunct inflection, as shown in example (4).6

3. léicid-som cloich asa tailm

release.PRES.3SG.ABS-emph.part.3SG.M stone out-of-his sling

‘He releases a stone out of his sling…’ (LU 6210–6211)

4. cenid leci in metur…


‘Although the metre does not allow…’ (Ml 30a10)

2.2 Compound verbs: deuterotonic and prototonic

In addition to simple verbs, Old Irish also has numerous compound verbs consisting

of a simple verb preceded by one or more preverbs. Preverbs are particles,

etymologically related to prepositions, that change the meaning of the verb. In some

cases this change in meaning is predictable from the combined meaning of the

preverb and the verb. For example, the verb do-beir ‘gives, brings’ consists of the

verb berid ‘carries’ and the preverb do ‘to’.7 Usually, however, the meaning is not

transparently compositional. The meaning of the verb fo-reith ‘helps’, for example, is

not a sum of its parts, the verb rethid ‘runs’ and the preverb fo ‘under’. Similarly,

many compound verbs contain a simple verb that is not independently attested. In the

case of the verb ad-cí ‘sees’, the simple verb -cí does not exist independently of the

compound.

Like simple verbs, compound verbs have different forms depending on their

position in the clause. However, unlike simple verbs, it is the stem rather than the

ending that shows the alternation. Compound verbs always have conjunct inflection

regardless of their position in the clause. When a compound verb is in absolute initial

position in the clause it is deuterotonic, the stress falls on the second syllable. When a

compound verb is preceded by a conjunct particle it is prototonic, showing the normal

Old Irish stress pattern of stress on the first syllable. The deuterotonic and prototonic

6 The verb also has conjunct endings in so-called Bergin’s Construction. We will return to this construction in section 3.1 and chapters 4 and 5. 7 Although see Dillon (1962) on the prehistory of the preverb do.


22

forms of the verbs do-beir ‘gives’, ad-cí ‘sees’ and do-gní ‘does’ are shown below for

the present indicative.

Table 3: Deuterotonic and prototonic forms

Deuterotonic Prototonic Deuterotonic Prototonic Deuterotonic Prototonic

1sg do-biur -tabur 1sg ad-cíu -aicciu 1sg do-gníu -dén(a)im

2sg do-bir -tabair 2sg ad-cí -aci 2sg do-gní -dén(a)i

3sg do-beir -tabair 3sg ad-cí -aicci 3sg do-gní -dén(a)i

1pl do-beram -taibrem 1pl ad-ciam -accam 1pl do-gníam -dénam

2pl do-berid -taibrid 2pl ad-ciid -accid 2pl do-gníith -dénid

3pl do-berat -taibret 3pl ad-ciat -accat 3pl do-gníat -dénat

As can be seen from these paradigms the deuterotonic and prototonic forms differ

greatly in terms of their phonological shape. The majority of these differences can be

explained by the difference in stress.8 Because the deuterotonic and prototonic forms

are stressed on different syllables, processes of sound change affected them

differently. For example, in the history of Irish unstressed vowels were liable to vowel

affection.9 Between two palatal consonants an unstressed vowel is raised to /i/.

Similarly, between two non-palatal consonants an unstressed vowel is lowered to /a/.

In the deuterotonic form, the second syllable is stressed and so unaffected; however,

in the case of the prototonic form the second syllable is unstressed and so susceptible

to vowel affection. This can be seen in the examples in (5) below, where the original

vowel /e/ remains in the deuterotonic forms do-bert ‘gave’ and as-beir ‘says’, but is

lowered and raised respectively in the corresponding prototonic forms.

5. (a) do-bert vs. -tubart ‘gave’

(b) as-beir vs. -epir ‘says’

(McCone 1997a: 5)

Another stress related change is that of syncope. Starting from the stressed syllable,

syncope eliminates the vowel of every second syllable. Because the stress is in

8 For more details see McCone (1997a: 4–9). 9 Stressed vowels underwent a different process of vowel affection whereby the vowel was lowered or raised depending on the quality of the following vowel, e.g. nom sg fer<*uiros ‘man’ vs. gen sg fir<*uirī (see Thurneysen 1946: 46–55).


23

different places in the deuterotonic and prototonic forms, different syllables were lost

(*).

6. (a) for-cenna vs. -foirc*nea ‘finishes’

(b) do-rósc*(c)aifet vs. -der*scaig*fet ‘they will excel’

(McCone 1997a: 6)

The other major phonological differences between the deuterotonic and prototonic

forms seem to suggest that the initial preverb of the deuterotonic compound is in

some sense separated from the remainder of the verb. So, whereas in the prototonic

form, an initial preverb ending in a vowel causes lenition of the following segment,

this lenition is blocked in the case of the deuterotonic (7).

7. (a) fo-fera vs. -foírea ‘causes’ (f > ø)

(b) do-cuaid vs. -dechuid ‘he had gone’ (‘c’ /k/ > ‘ch’ /x/)

(McCone 1997a: 4)

Although the differences between the deuterotonic and the prototonic can be

explained historically in this way, McCone (1997a: 191) argues that the forms differ

to too great an extent to be derived synchronically from a single underlying form and

were essentially a collection of irregularities. From an observational point of view,

this seems the most likely scenario; however, to prove that this is the case a full

analysis of the phonology of Old Irish as a whole and of compound verbs in particular

is necessary. As the main focus of this dissertation is on the syntactic aspects of the

double system, such an analysis will not be attempted here.10 For the remainder it will

be assumed, following McCone, that the deuterotonic and prototonic cannot be

derived synchronically from a single underlying form; however, this will be by no

means crucial to the analysis that is to follow.

2.3 Object pronouns

Aside from the double system of verbal inflection, the most striking feature

differentiating the Old Irish verbal system from its modern counterpart is the use of 10 See McCloskey (1978) and Eythórsson (2003) for analyses of the double system from a phonological perspective.


24

enclitic object pronouns. Object clitics in Old Irish behave similarly to those in other

IE languages, appearing in second position in the clause in accordance with

Wackernagel’s Law (Wackernagel 1892). At first sight this seems to suggest that the

clitic will always appear enclitic to the verb, because, as we saw above, the verb is in

initial position. However, this is not the case. The verbal complex in Old Irish can

contain of a number of different parts, such as a conjunct particle and multiple

preverbs in addition to the verb. Although it is not always attached to the verb, the

enclitic pronoun always appears second within the verbal complex. This is known as

Vendryes’ Restriction (Vendryes 1908).

When there is a conjunct particle, the pronoun is enclitic to it, and appears before

the verb, which has dependent (conjunct or prototonic form):

8. (a) Ní-s n-ágathar (cf. DT ad-ágathar)

NEG-INF.3PL fear.PRES.3SG.PT

‘He does not fear them’ (Thurneysen 1946: 255)

(b) Ní-m charat-sa

NEG-INF.1SG loves.PRES.3PL.CONJ-emph.part.1SG

‘They do not love me’ (Wb 5c6)

When there is no conjunct particle, and the verb is compound, the pronoun appears

infixed between the initial preverb and the remainder of the verb:

9. du-s n-gní

PVB-INF.3SG makes.PRES.3SG.DT

‘He makes it (f)’ (Ml 29a3)

When there is no conjunct particle and the verb is simple there are two possible

options. Either the object pronoun is suffixed to the verb in initial position, (10a), or a

dummy particle no is inserted and the pronoun is infixed between no and the verb, in

conjunct form, (10b).

10. (a) beirth-i

carry.3SG.ABS-SUFF.3SG.M

‘carries it’ (Ml 42b7)


25

(b) no-m ísligur

PVB-INF.1SG abase.PRES.1SG.CONJ

‘I abase myself’ (Wb 17d22)

In the Old Irish period, suffixed pronouns are of restricted use. They can only be used

when the pronoun and the verb are third person, or in more archaic texts, when the

verb is 1pl. Even then, this usage is not consistent, and the construction using the

dummy preverb no may be used instead with no difference in meaning (Thurneysen

1946: 271).11 The other forms of the pronoun are used regularly with the 3sg form of

the substantive verb at-tá ‘be’ to indicate possession (like Modern Irish, Old Irish has

no verb meaning ‘have’).

11. táth-ut12 ‘you (sg) have’ (lit. there is to you)

táth-(a)i ‘ he has’ (lit. there is to him)

táth-us ‘she has’ (lit. there is to her)

táth-unn ‘we have’ (lit. there is to us)

táth-uib ‘they have’ (lit. there is to them)

(Thurneysen 1946: 271)

With other verbs, there are a few examples of the first and second persons, but these

are restricted to poetic language (Thurneysen 1946: 271). These traces of first and

second person forms suggest that at one time suffixed pronouns were productive and

could be used more freely. However, in the Old Irish period, it seems to be the no

construction that is productive and suffixed forms are archaisms of restricted use.13

Having outlined the main features of the Old Irish verbal system we are now in a

position to examine the existing theories that deal with the double system of verbal

inflection.

3. SYNTACTIC ANALYSES OF OLD IRISH

As discussed in section 2, Old Irish differs from most other IE languages in two main

respects in having unmarked verb-initial word order and a double system of verbal 11 Although see Cowgill (1987), who argues that this is not the case. 12 Thurneysen (1946: 478) suggests that this -th- reflects the earlier absolute ending of the substantive verb *táith, which has been lost in other contexts. 13 See Breatnach (1977) for more detail on suffixed pronouns.


26

inflection. A successful synchronic analysis must be able to account for both of these

features. The different verbal forms are related to different positions in the clause,

with the independent (absolute and deuterotonic) forms appearing in absolute initial

position, and the dependent (conjunct and prototonic) forms appearing after a

conjunct particle. This suggests that the form of the verb could reflect the syntactic

structure in which the verb is contained. This is the basic idea behind the two existing

syntactic accounts of the double system, to which we turn in the sections below.

3.1 Old Irish as a residual V2 language: Doherty (1999, 2000)

Doherty (1999, 2000) claims that like Modern Irish (see, among others, McCloskey

1996a, Carnie 1995, Harley 1995, Duffield 1995) the verb in Old Irish is always in T.

However, he claims that this was not always the case. At some earlier period Irish was

a verb-second (V2) language, whereby the verb raised to C and one constituent was

fronted to the Spec-CP.

It is widely accepted that pre-Old Irish was not a verb-initial language (Watkins

1963, McCone 1979b). Archaic Irish texts show examples where the verb is not in

clause-initial position, but appears clause-medially or clause-finally. These examples

are known as Bergin’s Construction, as Bergin (1938) was the first to observe that in

all cases where the verb is not in clause-initial position it has conjunct inflection.

Doherty proposes that all the examples of Bergin’s Construction are in fact V2.14

Bergin’s examples are generally verb-final; however, there is usually only one

constituent preceding the verb, as in (12).

12. (a) ceso femmuin m-bolgaig m-bung

although seaweed blistered reap.PRES.1SG.CONJ

‘Although I reap blistered seaweed’ (Corm 1059/Bergin 1938: 197)

(b) fri aingel n-acallastar

to angel speak.PRET.3SG.PT

‘He spoke to an angel’ (LU 1148/Bergin 1938: 201)

According to the standard analysis (den Besten 1983), in asymmetric V2 languages

the verb is in C. If Bergin’s Construction shows V2 and all verbs in Bergin’s

14 See also Carney (1977–9); Koch (1991).


27

Construction have conjunct inflection then, under Doherty’s analysis, whenever the

verb is in C it will have conjunct inflection.

Leaving aside for the moment the validity of Doherty’s claim that Bergin’s

Construction is V2 (we will return to this in chapter 5), let us examine how this relates

to the distribution of absolute and conjunct forms in Old Irish. As discussed in

sections 2.1 and 2.2 above, the verb has dependent form (conjunct or prototonic)

when it is preceded by a conjunct particle. On the assumption that conjunct particles

are complementizers, or at least elements merged in the C position (Chung and

McCloskey 1987), Doherty draws a parallel between Bergin’s Construction and the

synchronic double system of verbal inflection, summarised below:

“Verbs are in dependent form (conjunct, prototonic) when CP is projected.”

(Doherty 2000: 19)

When there is a conjunct particle, it appears in C and so C is projected, resulting in

conjunct inflection. Similarly, in Bergin’s Construction, the verb is in the C position,

and so C is projected. In all other cases, C is not projected and so the verb has

absolute inflection.

The major drawback of Doherty’s account, in synchronic terms, is that it is

essentially the equivalent of saying ‘the conjunct form is used after a conjunct

particle’; it is descriptive not explanatory. It is not entirely clear why the projection of

a particular category in the syntax should influence the morphological form of the

verb, especially if this category is not the one that actually contains the verb in

question. Furthermore, in minimalist terms the C-projection is assumed to always be

present, regardless of whether it has any phonological realisation. In this case, there

will be no structural difference between a clause that has a conjunct particle and one

that does not.

Further problems arise if we consider the Old Irish data in more detail. In Old Irish

there are certain particles that do not cause the verb to appear in dependent form. For

example the particle ma ‘if’ is followed by independent verb forms, as shown in (13).

13. ma dud-esta ní dibar n-iris

if PVB-is.lacking.PRES.3SG.DT anything of.your faith

‘If aught is lacking in your faith…’ (Wb 25a30)


28

Although it may be the case that in Old Irish this particle does not appear in the C

position (see Newton 2005 for such an analysis), cross-linguistic evidence suggests

that such particles should be part of the C complex. If this is the case then CP will be

projected and, given Doherty’s assumptions, we would expect the verb to have

dependent form.

Similarly if we consider relative constructions, we find that in most cases the verb

has independent form. There are various ways of marking the relative, which will be

discussed in more detail in section 3.3.2. However, as can be seen in the examples in

(14), the verb in a relative construction is clearly independent (Thurneysen 1946:

312f).

14. (a) a n-ad-chiam (cf. non-rel ad-ciam)

that PVB-see.PRES.1PL.DT

‘that which we see’ (Ml 112b13)

(b) is oinfer gaibes búaid

COP one.man seize.PRES.3SG.REL victory

‘it is one man who seizes victory’ (Wb 11a4)

Cross-linguistically relative constructions show certain similarities to interrogatives,

and so are viewed as a feature of the C-projection (Chomsky 1977). In that case,

whenever the clause is relative we would expect to find a C-projection, and, under

Doherty’s analysis, a dependent verb form. This is clearly not the case.

It seems, then, that Doherty’s account of the double system of verbal inflection

falls down on both theoretical and empirical grounds, and so cannot be maintained.

3.2 Two types of raising: Carnie, Harley & Pyatt (2000)

Carnie, Harley & Pyatt (2000 – CHP) propose an alternative syntactic account of the

double system of verbal inflection whereby the morphological form of the verb is

linked directly to the verb’s syntactic position. CHP argue that Old Irish has both V-

to-T and V-to-C raising as a result of a filled C requirement. When there is a conjunct

particle, this is merged in C and so the filled C requirement is met and the verb moves

only as far as T. When C is not filled by a conjunct particle, some verbal element

must move to fill it. If the verb is simple, then the verb itself moves to C. If the verb is


29

compound, then only the first preverb moves to the C position and the remainder of

the verb is in T. This can be seen schematically in the trees below:

15. (a) C (b) C (c) C C T C T C T ní berid do T V T V T V beir tV beir

The morphological ending of the verb reflects the verb’s syntactic position. When the

verb is in C it has absolute morphology and when it is in T (or simply not in C) it has

conjunct morphology. Absolute morphology, then, can be seen as the realisation of a

C feature on the verb. When the verb is in T, this C feature will not be present and so

the default conjunct morphology will appear.

CHP account for compound verbs in a similar fashion, with the difference in

phonological form reflecting a difference in the syntax. When a conjunct particle fills

C, the two parts of the compound verb appear in T and so are spelled-out together as

the prototonic form. When there is no conjunct particle, and the initial preverb is in C,

the preverb and verb are in separate syntactic positions, and so are spelled-out

separately, resulting in the deuterotonic form.15

Linking the morphological form of the verb to the syntactic positions in which it

appears is pleasing from a theoretical perspective, as it provides an interesting account

of the different forms. A further theoretical advantage of CHP’s theory is that it draws

an interesting parallel with the Germanic V2 languages. As mentioned in section 3.1,

following den Besten (1983), asymmetric V2 languages such as German and Dutch

are generally analysed as involving verb-movement to C. In German the verb only

moves to second position in the clause in clauses with no overt complementizer.

When a complementizer is present, V-to-C movement is blocked and so the verb

remains in situ. The filled C requirement postulated by CHP for Old Irish is identical

to that found in V2 languages. When C is not filled by a conjunct particle, some

verbal element, either the verb or the first preverb, moves to fill C. The main

difference, then, between verb-initial orders in Old Irish and V2 in Germanic is that in

15 The situation with compound verbs is by no means as straightforward as this suggests. We will examine it in detail in section 3.4.


30

Germanic there is an additional topicalisation operation that attracts an XP to Spec-

CP.16

So it seems that there are certain theoretical advantages in adopting CHP’s theory.

However, the crucial test is to see how it accounts for the data. The next section

considers the evidence from simple verbs; compound verbs are the subject of section

3.4.

3.3 Are simple verbs in C? Empirical evidence

CHP invoke two types of evidence to support their claim that simple verbs in Old

Irish raise to the C position, namely object pronouns and relative constructions. We

will examine each of these in turn.

3.3.1 Object pronouns

As discussed in section 2.3, object pronouns in Old Irish are enclitic and always

appear in second position. If the verb form is simple and in initial position, the object

pronoun is suffixed to it, as in (16). If the verb follows a conjunct particle, the object

pronoun is infixed between the conjunct particle and the verb, as in (17). If there is no

conjunct particle and the verb is compound, the object pronoun is infixed between the

initial preverb and the remainder of the verb, as shown in (18).

16. Eorum is do apstalaib beirth-i

Eorum COP to apostle.DAT.PL carries-SUFF.3SG.M

‘The Eorum he applies it to the apostles’ (Ml 94b1)

17. Ní-m charat-sa

NEG-INF.1SG loves.PRES.3PL.CONJ-emph.part.1SG

‘They do not love me’ (Wb 5c6)

16 A further interesting parallel is found in East Netherlandic dialects of Dutch, where the verb shows different morphological endings in subject and non-subject V2 clauses. Zwart (1997) argues that this reflects the different position of the verb in each case. When the topicalised constituent is a non-subject the verb raises to C and shows C-morphology. When the topicalised constituent is a subject, the verb is in T and shows T-morphology. This is directly parallel to CHP’s proposal for Old Irish.


31

18. du-s n-gní

PVB-INF.3SG.F makes.3SG.PRES.DT

‘He makes it (f)’ (Ml 29a3)

According to syntactic accounts of clitic placement, object clitics have a set position

in the clause (Kayne 1991). In many languages object pronouns seem to appear

between C and T (see Kayne 1975, 1991, 1994, Cardinaletti & Roberts 2002 on

Romance; Grohmann 2000 on dialects of German; Ackema & Neeleman 2005 on

Middle Dutch). As clitics appear between these two functional projections, it is not

entirely clear to which they belong. Cardinaletti & Roberts (2002) argue that clitics

move to the left edge of the T complex. Uriagereka (1995) argues that they move to a

projection, FP, which encodes point of view and is part of the C-projection. A full

discussion of the target for and motivation behind clitic placement is beyond the

scope of this dissertation. Crucially, object pronouns in Old Irish seem to occur in a

similar position to those found in other European languages. If we consider the

example given in (17) above, we find that the pronoun appears between the conjunct

particle, a C element (Chung & McCloskey 1987) and the verb, in T. This suggests,

that as in other European languages, object clitics in Old Irish appear between C and

T. For concreteness it will be assumed that these pronouns occupy the lowest

projection of the CP, equivalent to Rizzi’s FinP. If this is the case, and the position of

the clitics remains constant, then this supports CHP’s analysis of simple verbs. CHP

argue that when the verb is simple object pronouns appear to its right and are suffixed

to it; this is what we would expect if the verb had raised to C.

The evidence from object pronouns, however, is not as clear-cut as CHP suggest.

As was seen in section 2.3, use of suffixed pronouns in Old Irish is restricted.

Generally only third singular forms of the pronoun are found attached to third singular

indicative verb forms, and even in these cases suffixed pronouns are not used

consistently. The productive pattern in Old Irish seems to be infixation using the

dummy preverb no.

The restricted use of suffixed pronouns in Old Irish is problematic for CHP in two

respects. First, the fact that suffixed pronouns are not productive and are only used in

restricted contexts means that they do not provide convincing evidence that simple

verbs move to C. The second, and perhaps more serious, problem is that CHP must be

able to account for the productive pattern. It seems to be the case that when there is an


32

object pronoun present V-to-C movement is blocked. One way to explain this could

be to say that the infixed pronoun satisfies the filled C requirement (except in certain

numbers and persons). However, if this is the case, then we would expect movement

of the initial preverb to C also to be blocked when there is a pronoun and that does not

seem to be so (see section 3.4 below). Similarly it could be argued that the dummy

preverb no is merged into the C position whenever there is an infixed pronoun, and

this satisfies the filled C requirement and prevents verb-movement to C. However,

again this does not occur with compound verbs. If no were inserted with the pronoun

we would expect this to be the case across the board, and so again movement of the

initial preverb of the compound verb would also be blocked.

It seems, then, that instead of providing evidence to support CHP’s view, object

pronouns pose a significant problem for it. In the regular, productive case simple

verbs clearly do not move to C when there is an object pronoun, as they appear with

the dummy preverb no. Under CHP’s theory there seems to be no satisfactory

explanation as to why V-to-C movement is blocked in these cases. If, on the other

hand, there is no filled-C requirement in Old Irish so the verb never moves to C and

only ever moves as far as T, then the productive system is easily accounted for and it

is only the irregular suffixed forms that require extra explanation.17 A theory that

accounts primarily for the regular construction seems clearly preferable to one that

focuses on the irregular one. The evidence from object pronouns, then, is better

accounted for if it is assumed that there is no filled-C requirement and the verb in Old

Irish only raises as far as T.

3.3.2 Relative clauses

The second piece of evidence invoked by CHP is that of relative marking. In Old Irish

there are many different ways to mark a relative clause (see Thurneysen 1946: 312–

325; McCone 1980; Ó hUiginn 1997). One way that a clause can be marked as

relative is to use a special relative form of the verb. This can be seen in example (19)

below, where the non-relative form would be gaibid ‘takes, seizes’.

17 This is reasonably straightforward. The irregular nature of suffixed pronouns can be accounted for if we assume that the combination of verb+suffix is learnt by rote as an irregularity and inserted into the appropriate syntactic context, rather than being derived through a regular syntactic process of cliticization.


33

19. is oinfer gaibes búaid diib inna chomalnad

COP one man take.PRES.3SG.REL victory of.3PL in its completing

‘It is one man of them that gets victory for completing it’

(Wb 11a4)

As noted above, relative clause marking is associated with the C position so the fact

that special relative morphology is only found on simple verbs in absolute clause-

initial position supports the idea that simple verbs move to C in Old Irish.

Like suffixed pronouns, special relative verb forms in Old Irish are restricted in

distribution to verbs in the third person and, in the earliest texts, first plural

(Thurneysen 1946: 313). In all other persons when there is no special relative form,

the dummy particle no is inserted, which either lenites or nasalizes the initial segment

of the verb and causes it to have conjunct inflection, as shown in the examples in (20).

20. (a) is hed in so no chairigur (non-rel cairigur)

COP it this PVB reprimand.PRES.1SG.CONJ

‘This is what I reprimand’ (Wb 11d1)

(b) cid no mbetha (non-rel betha)

why PVB be.PAST.SUBJ.2SG.CONJ

‘Why (is it that) you should be?’ (Wb 4c24)

It seems on closer inspection, then, that instead of supporting CHP’s theory, the

evidence from special relative forms is problematic. As is the case for suffixed

pronouns, the special relative verb forms are so restricted that they cannot provide

convincing evidence that simple verbs move to C. Moreover, CHP must be able to

account for these restrictions, explaining why movement to C is blocked in the

majority of relative contexts. This could perhaps be explained if we argue that a

relative feature in C satisfies the filled C requirement or that no is inserted into the C

position in relative contexts, preventing verb-movement to C. However, as was the

case with object pronouns, a relative feature in C does not stop the initial preverb of a

compound verb from appearing in C. As with object pronouns, it seems that in

relative contexts, the productive pattern seems to be that where the verb only moves

as far as T and the cases that seem to demonstrate V-to-C movement are exceptions,

suggesting that we should not postulate a filled-C requirement for Old Irish.


34

There is further evidence from relative clauses to suggest that the verb only moves

as far as T in Old Irish. The most common way to mark a relative clause in Old Irish

is through lenition. When the verb is compound, lenition affects the segment after the

initial preverb, as shown in (21).

21. a n- ad-chiam (non-rel ad-ciam)

that NAS PVB-see.PRES.1PL.DT

‘That which we see’ (Ml 112b13)

When the verb is simple and a special relative form is unavailable, the dummy

preverb no is inserted and the initial segment of the verb is lenited, as in (22) below.

22. is hed in so no chairigur (non-rel cairigur)



These data can be made sense of if we assume that when C is specified as relative it

causes lenition of the initial segment of the following word.18 Crucially, the particle

that appears in the C position is not lenited. If special relative verb forms move to C,

then, we would expect them not to be lenited. Although this is the case in the earliest

sources of Old Irish, by the time of the Milan Glosses (AD 800) we find that simple

relative verb forms begin to undergo lenition. This becomes widespread by the St Gall

Glosses (AD 850) (Thurneysen 1946: 315):

23. (a) indí chomallaite (non-relative comallait)

those fulfil.PRES.3PL.REL

‘Those who fulfil.’ (Ml 114b7)

(b) cisí aimser derb thechtas (non-relative techtaid)

what time definite possess.PRES.3SG.REL

‘What is the definite time that he has?’ (Sg 26a6)

18 A similar analysis has been provided for relative clauses in Modern Irish. The main difference is that in Modern Irish relative C is phonologically realised as a particle a that either lenites or nasalizes the following word (McCloskey 2001b, 2002)


35

This development suggests that although historically special relative verb forms were

in the C position and so unlenited, by the time of the Milan Glosses this is no longer

the case. The fact that such verb forms are unlenited in the Würzburg Glosses written

some fifty years earlier could imply that V-to-C movement was lost between these

750 and 800. Alternatively, and perhaps more likely, V-to-C movement may have

been lost before the time of the Würzburg Glosses, but it was not until the time of the

Milan Glosses that the full effects of this change came to be seen in the written

language.

3.3.3 Interim summary

Although CHP’s analysis of absolute and conjunct inflection in Old Irish is

theoretically appealing, the empirical evidence does not seem to support it. The

evidence that CHP provide from object pronouns and relative constructions, when

considered more closely, not only fails to support their analysis but also poses

significant problems for it. It is difficult to account for the use of no to infix pronouns

and mark relative clauses within a theory that postulates across the board V-to-C

movement. In both cases the productive pattern seems to show that the verb only

moves as far as T. The examples that seem to show V-to-C movement are marginal

and irregular in the Old Irish period. These data seem better accounted for if the verb

only moves as far as T in Old Irish. The irregular forms that appear to demonstrate V-

to-C movement are archaisms from an earlier period, when V-to-C movement was

productive.

Before we move on there is a further piece of evidence that could be taken to

support the view that simple verbs only raise as far as T in the syntax, namely stress

patterns. Generally all other elements that CHP propose can fill the C position in Old

Irish are unstressed, i.e. conjunct particles and the initial preverbs of deuterotonic

compound verbs (Thurneysen 1946: 28–30). This being the case, if a fronted verb

appears in the C position we might expect it to be unstressed. If, on the other hand,

the verb only ever moves as far as T, then we have a unitary explanation for the stress

patterns of the Old Irish verbal complex, namely stress always falls on the first

syllable of TP.

So it seems that there is good reason to believe that simple verbs do not move to C

in Classical Old Irish. However, if simple verbs only ever move as far as T in the

syntax, then the different morphological endings cannot reflect different syntactic


36

positions. This means that we need an alternative explanation for the different

morphological forms. This will be the topic of section 4 below. First, let us consider

compound verbs.

3.4 Deriving compound verbs: a problem for syntactic accounts

3.4.1 Initial preverbs are in C

CHP argue that the initial preverb of a compound verb can satisfy the filled C

requirement and appear in the C position. In order to investigate the validity of this

claim we must consider the evidence from Old Irish. Initial preverbs behave

phonologically and syntactically like conjunct particles, which suggests that both

appear in the same position, namely C.

Let us consider first the phonological features of initial preverbs. The most notable

characteristic of the initial preverb of a compound verb is that it is unstressed, hence

the irregular deuterotonic stress pattern of independent compound verb forms (see

section 2.2). Conjunct particles are also unstressed (Thurneysen 1946: 30). If we

consider the form of preverbs, they again show similarity to conjunct particles.

Preverbs, like conjunct particles, tend to be short words, consisting in most cases of

only one syllable.19

Turning to the syntactic characteristics of initial preverbs and conjunct particles we

again see similarities. The first piece of evidence that conjunct particles and initial

preverbs compete for the same syntactic position is that they are in complementary

distribution. It is only when there is no conjunct particle present in the clause that a

preverb can appear independent from its verb. When a conjunct particle is present the

initial preverb appears as a prefix on the verb. It cannot be independent. This can be

seen by comparing the independent form ad-cí with its dependent counterpart aicci.

24. (a) Ad-cí ind rígain Conaire

PVB-see. PRES.3SG.DT the queen Conaire

‘The queen sees Conaire’

19 This surface similarity is due to the fact that all preverbs and many conjunct particles developed from prepositions. See chapter 4 below.


37

(b) Ní aicci ind rígain Conaire

NEG see.PRES.3SG.PT the queen Conaire

‘The queen doesn’t see Conaire’ (Quin 1975)

The fact that independent preverbs and conjunct particles do not co-occur suggests

that they compete for the same syntactic position.

The evidence from object pronouns used by CHP also suggests that initial preverbs

and conjunct particles appear in the same syntactic position. When there is a conjunct

particle in the C position, the object pronoun appears to its right, between the conjunct

particle and the verb, (25a). When the verb is compound and in initial position the

object pronoun is infixed between the initial preverb and the remainder of the verb

(25b).

25. (a) Ní-m accai

NEG-INF.1SG see.PRES.3SG.PT

‘He does not see me’

(b) Atom- chí

PVB-INF.1SG- see.PRES.3SG.DT

‘He sees me’ (Quin 1975)

If we assume, following CHP, that the clitic has a set position in the clause, namely

between C and T, then any element which appears to the left of the enclitic must be in

the C position. As both conjunct particles and initial preverbs appear to the left of the

enclitic, they must both be in the C position.20

Evidence from relative clauses also suggests that initial preverbs and conjunct

particles occupy the same syntactic position. As noted above, the most common way

to mark a relative clause is through lenition. When the verb is compound and in initial

position the first segment following the initial preverb, i.e. the first segment of the

stressed syllable of a deuterotonic compound, is lenited (26a). Similarly when a

relative conjunct particle, for example the relative negative nad, is in clause-initial

20 The evidence from clitic pronouns should not be taken in isolation. It could be the case that the enclitics are placed post-syntactically by Prosodic Inversion (Halpern 1995) or Local Dislocation (Embick & Noyer 2001, Adger to appear). Local Dislocation will be discussed in more detail in section 4 below.


38

position, this lenites the first segment of the following verb, i.e. the first segment of

the stressed syllable of the conjunct or prototonic verb (26b).

26. (a) a n- ad-chiam (ch=/x/, cf. non-relative ad-ciam)

that.NAS PVB-see.PRES.1PL.DT


(b) sillab nad ṡluindi (ṡ=/ø/, cf. non-relative sluindi)

syllable NEG express.PRES.3SG.CONJ

‘A syllable that does not express’ (Sg 25b13)

In section 3.3.2 it was argued that in Old Irish relative C lenites the following

segment. Therefore the fact that these relative marking mutations follow the initial

preverb suggests that the preverb must in fact be in the C position. If the preverb were

lower than C then relative C would lenite or nasalize the preverb and not the

remainder of the verb.

In sum, the initial preverb of a deuterotonic compound verb resembles a conjunct

particle in terms of both phonological form and syntactic behaviour. This suggests

that initial preverbs and conjunct particles occupy the same syntactic position, namely

C. However, in spite of these similarities there is one significant difference between

conjunct particles and preverbs. If conjunct particles are complementizers then they

will be merged directly into the C position. Preverbs, on the other hand, are

semantically part of the verb (see section 2.2 above), and so we expect them to be

merged lower in the clause as part of the vP. If the preverb originates in vP, we must

explain how it gets from its base position to C.

3.4.2 Head-movement

If preverbs appear in the C position, a head position, then the most straightforward

analysis is to suggest that they originate in a head position further down in the

structure and move to C via head-movement. CHP argue that compound verbs are an

example of a Hale & Keyser (1993) type complex VP and so the preverb is the head

of light v. This is appealing because it captures the fact that the preverb and the verb

seem to form a single semantic unit (see section 2.2 above). If the verb originates in

light v it must move to C via head-movement; however, this is by no means

straightforward.


39

3.4.2.1 Traditional head-movement

Under the traditional formulation of head-movement, a moved head is subject to the

Head-Movement Constraint (Travis 1984 – HMC). When a head moves it must move

through each intervening head position. In Old Irish when the verb is compound and

C is filled by a conjunct particle the verb seems to move according to these

conditions. The verb moves from V to light v and incorporates with the preverb. This

newly formed verbal complex of verb+preverb will then move to T and be spelled-out

as the prototonic form.21

27. CP C TP ní T vP tabair v V tdo V DP tbeir

When C is not filled by a conjunct particle, the situation is more complex, as the

preverb and verb are in different syntactic positions. The preverb must move on from

T to C. One possibility, suggested by Carnie, Harley & Pyatt (1994) and Eska (1996),

is that the verb incorporates with the preverb and moves up to T, as is the case when

C is filled, but when C is empty the preverb excorporates and moves to fulfil the filled

C requirement. Even if we accept that excorporation is possible, the excorporation

needed here would be of a different class entirely from that found in other languages.

Roberts (1991) argues that it is possible for a moved head to excorporate from a head

with which it has incorporated and continue to move up the structure to a higher

position without the other head. In the Old Irish case, it is not the originally moved

21 According to Baker (1988), when a head moves through another head position and incorporates with it the higher head will appear as a suffix on the moved head. This can be seen as a result of Kayne’s (1994) Linear Correspondence Axiom, which specifies that adjunction must always be to the left. To get the desired order in Old Irish, we need to employ right adjunction, and argue that the higher head, the preverb, appears as a prefix to the verb. One way around this problem is to adopt a DM approach whereby the status of each morpheme as a suffix or a prefix is lexically encoded, and not purely a result of syntactic operations (Embick & Noyer 2001). However, this does not enable us to make the kind of predictions that Baker’s theory of incorporation does.


40

head (V) that must excorporate, but the head with which the moved head incorporated

(v). To account for the Old Irish data, then, we would have to not only allow

excorporation in our grammar, but also significantly alter the conditions under which

it can take place, making it even less restrictive. This is clearly undesirable.

If excorporation is impossible then the only other option is for the preverb and verb

simply not to incorporate in the first place; the preverb must move independently of

the verbal stem. If this is the case then the verb must move from V to T, without

incorporating light v and the preverb must move straight to C, skipping the T position,

as shown in the tree below.

28. CP C TP do T vP beir v V tdo V DP tbeir

Such movement is clearly prohibited by HMC and so this derivation is impossible

using traditional head-movement.22

3.4.2.2 Head-movement in the minimalist programme

As discussed in chapter 1, it is unclear whether the HMC plays a role in minimalism.

Of course, if we follow Chomsky (2001) and argue that head-movement is not part of

narrow syntax then it is clearly redundant. However, if we admit head-movement into

the syntax and make the assumption that it proceeds in the same way as XP-

movement then we would expect both head- and XP-movement to be subject to the

same locality conditions, namely the Phase Impenetrability Condition (PIC). Let us

assume that the PIC is the only locality constraint relevant for head-movement in

minimalism and explore whether this can help with the analysis of the Old Irish data.

22 CHP argue that independent preverb-movement can be explained using Long Head-Movement (LHM). However, they admit that the LHM necessary to account for Old Irish would have to be very different from that found elsewhere. See Adger (to appear) and Newton (2005) for further arguments against this view.


41

The PIC has two forms, which following Müller (2004) we will refer to as PIC1

and PIC2. PIC1 from Chomsky (2000: 108, 2001: 13) is given below:

Phase Impenetrability Condition 1 (PIC1)

In a phase α (HP) with head H, the domain of H is not accessible to operations

outside α (HP); only H and its edge are accessible to such operations.

The domain, or complement, of H is spelled-out as soon as the phase, HP, is

complete, i.e. as soon as the next probe is introduced. So, if following Chomsky, we

take C and light v to be phases, under PIC1 the complement of v, i.e. VP, will be sent

to Spell-Out as soon as T is merged. As a result T will not be able to probe the

contents of VP, and so nothing can move directly from V to T.

If we consider the Old Irish data in terms of the locality conditions imposed by

PIC1 we find that the result is essentially the same as it was according to the HMC.

The verb cannot move to T directly, as according to PIC1 VP becomes unavailable

when T is merged, so the verb must first move first to light v and then on to T.

If we consider now the second formulation, the locality conditions are somewhat

different. PIC2 as given by Chomsky (2001: 14) is provided below:

Phase Impenetrability Condition 2 (PIC2)

[Where Z and H are phases, in the configuration [ZP…[HPα[H YP]]] – GEN]

The domain of H is not accessible to operations at ZP; only H and its edge are

accessible to such operations.

According to PIC2 the complement of H is only inaccessible to the next phase head.

The complement of H is not spelled-out until the next phase head is merged. So, if C

and v are phases, then the complement of v, i.e. VP will not be sent to Spell-Out until

C is merged. The main difference, then, between PIC1 and PIC2 is that in the case of

PIC1 VP is sent to Spell-Out as soon as T is merged, so T cannot probe VP directly,

whereas in PIC2 VP is not sent to Spell-Out until C is merged, and so T can probe VP

directly.

Let us consider the implications of PIC2 for the Old Irish data. Under PIC2 T can

probe VP directly; this means that the verb does not have to move via light v to


42

escape the phase. V can move straight to T without moving through and incorporating

with the preverb in light v. In addition, no part of the PIC demands that movement

from v to C must go through T, and as we are assuming that the PIC is the only

locality condition on movement, then movement directly from v to C also seems

possible. As a result, the derivation given in (28) above and repeated below as (29),

which was prohibited by the HMC, is now possible.23

29. CP C TP do T vP beir v V tdo V DP tbeir

By abandoning the HMC and invoking PIC2 it is possible for the preverb and verb to

move independently of one another without incorporation. The syntactic derivation of

deuterotonic verb forms has become possible. However, we are faced with a further

problem, namely how to derive the prototonic form. According to the PIC2-based

account of head-movement proposed here we would predict that if C is filled then V

will move to T as before, but movement of the preverb will be blocked, so the preverb

will remain in light v. This would result in the order C-verb-preverb. Such orders are

not attested at any stage of Irish. To obtain the attested orders, when C is filled both

the preverb and the verb must be in T.

There are two possible ways for the preverb and the verb both to be in T. When C

is filled either V moves to T via light v and incorporates, or both V and v move

independently to T. Neither of these options is attractive. For the first scenario, it is by

no means clear why the verb would move through v when C is filled and straight to T

when C is unfilled. Even if there were a reason, it seems likely that this reason would

23 Although these movement processes are all possible under PIC2, they will only take place in this manner under the appropriate feature configurations. T must probe for a feature present only on V and not on v. If the relevant feature is present on v then T will enter into an Agree relation with v rather than V as a result of locality. Similarly, for C to probe light v, T cannot contain the relevant features.


43

involve theoretically undesirable look-ahead. In the second case, the idea of a

functional category attracting two heads independently is not theoretically

appealing.24 For this to be the case T would have to have two separate Affix-features.

Furthermore, under such an analysis T would only attract both heads when C was

filled. Again, this presumably would involve look-ahead.

To summarise this section, it seems that although from a semantic and syntactic

point of view the analysis of the initial preverb of the compound verb as a light v head

is appealing, the attested syntactic patterns involving the preverb and verb in Old Irish

cannot be derived via head-movement. Traditional head-movement adhering to the

HMC cannot account for the deuterotonic case where the verb and preverb need to

move separately. If we abandon the HMC and assume that the PIC is the only locality

condition on head-movement, then we are still no better off. If head-movement

proceeds according to PIC1 then the preverb can move directly from v to C, but the

verb must still move via v and incorporate with the preverb. If head-movement

proceeds according to PIC2 then we can obtain the derivation of the deuterotonic form

shown in (29) whereby the verb moves to T skipping v, and v moves to C skipping T;

however, if movement proceeds in this manner, then we have no account of the

prototonic form, when the verb moves via v, and both the verb and the preverb are in

C.25

It seems that if the initial preverb is a light v head, then there is no way to derive

the attested forms by head-movement.

3.4.3 XP-movement

As there seems to be no adequate solution in terms of head-movement, let us explore

an alternative scenario. If, instead of being a light v head, the preverb is in fact an XP

then it will not be subject to the HMC and so will be able to move longer distances.

Analysing the preverb as an XP has the advantage that it enables us to draw a

cross-linguistic parallel with Germanic. Like Old Irish, German has compound verbs

consisting of a preverb (referred to as a particle in the Germanic literature) and a verb

that form a single semantic unit, but are syntactically independent. In German, the 24 Although this is proposed for Romance languages where T attracts both the finite verb and a clitic (Belletti 1999). 25 PIC2 has a further disadvantage in that it is incompatible with Chomsky’s (2004) proposal that the verb must move to light v to become verbalized. If this is the case, then V-to-T movement cannot be possible, as the verb would not be verbalized. In the remainder of this dissertation we will adopt the more restrictive PIC1 over PIC2.


44

verb can move independently of the particle, leaving the particle in the base position

in the VP, as shown by the examples in (30).

30. (a) Er steht um 7 Uhr auf

He stands at 7 o’clock up

‘He gets up at 7 o’clock.’

(b) Weil er um 7 Uhr aufsteht, ist er immer müde.

Because he at 7 o’clock up-stands is he always tired

‘Because he gets up at 7 o’clock he is always tired.’

There are two main analyses of this construction in German, the small clause account

and the complex head account. In the small clause account (Kayne 1985; Hoekstra

1988; den Dikken 1995) the particle and object form a small clause that constitutes

the complement of the verb. In the complex predicate analysis, the particle and the

verb are seen as a single lexical entry, and the object is the complement of the whole

complex verb (Neeleman 1994; Ackema & Neeleman 2005; Zeller 2001). No

consensus has been reached in the Germanic literature as to which structure is

preferable.26 In both cases the necessary syntactic configurations can be derived as it

is only the verb that moves, and this is a straightforward case of head-movement. In

Old Irish, however, both the verb and the preverb move independently. Movement out

of a small clause is not well understood. For this reason we shall concentrate on the

complex head account.

Zeller (2001) argues that the particle in German is the head of a Particle Phrase

that forms the complement of the verb. In German, as in Old Irish, when the verb does

not raise and the verb and particle remain in the VP the particle precedes the verb.

This is not a problem for Zeller’s analysis, as he assumes that German is head final

and so the complement will precede the verb. If Irish compound verbs have the same

syntactic structure as particle verbs in German, then Old Irish must also be head final.

Of course, it seems plausible that in its pre-history Irish was a head-final language

(see chapter 4), however it seems unlikely that this was the case in the Old Irish

26 In fact Wurmbrand (2000) argues that both types exist, with particle verbs that are semantically compositional having small clause structure and those that are semantically opaque being complex heads. If this is the case, then the fact that compound verbs in Old Irish are in the most part semantically opaque lends further support to the claim that they should be complex heads rather than small clauses.


45

period, where the verb precedes the object in both main and embedded clauses (see

Adger to appear). Furthermore, as Old Irish shows noun-adjective, noun-genitive

orders and prepositions rather than postpositions there is reason to believe on

typological grounds that it is a head-initial language. It seems unlikely, then, that

preverbs in Old Irish are verbal complements; however, this does not mean that we

must abandon the parallel with Germanic. It is possible that preverbs in Old Irish are

phrasal, but that instead of appearing as complements to the verb, they are in fact

specifiers.27

If the preverb is a phrasal category then we do not face the same problems as we

did when we analysed it as a head. Firstly, phrasal-movement is not subject to such

strict locality constraints as head-movement. Secondly, and perhaps more

importantly, if the preverb is a phrase then movement of the preverb and movement of

the verb will not interact. They will not move through the same positions and so will

not incorporate in the syntax.

Let us assume that the preverb originates in a particle phrase in Spec-VP. For the

preverb to fill the C position it must first move as a phrase to Spec-TP, and then as a

head to satisfy the filled C condition. At first sight this appears problematic in terms

of Chain Uniformity, as traditionally chains cannot contain both heads and phrasal

elements. Under Bare Phrase Structure (BPS – Chomsky 1995b, Chametzky 2000),

however, this is no longer a problem. If, as Zeller (2001: 127–49) proposes for

German, the particle phrase contains nothing but the preverb head, then in BPS terms

the preverb will be both a head and an XP simultaneously. As a result, there is no

problem with it moving first as a phrase and then as a head (see Sportiche 1992 who

makes such a proposal for clitic movement).

Under such an account we can derive all the required orders. When there is no

conjunct particle, the preverb moves as a phrase to Spec-TP and then as a head on to

C, and the verb moves via head-movement to T. When there is a conjunct particle in

C, the preverb moves to Spec-TP as a phrase, and the verb moves via head-movement

to T.

By analysing the verbal system in this way, movement is less constrained and so

we can derive the attested orders. However, just because the movements are possible

does not mean that they take place. Movement must be motivated, and it is unclear 27 Such an analysis is also more consistent with the diachrony of preverbs. Preverbs were originally adverbial phrases that modified the verb (see Vincent 1999 and chapter 4 below for details).


46

here what this motivation would be. If we maintain CHP’s filled C-requirement, then

this could motivate the movement of the preverb from Spec-TP to C, as the preverb

will be the closest verbal element when C is merged.28 Motivating the movement of

the particle phrase to the specifier of T is less straightforward. XP-movement to a

specifier position requires that the two positions be in an Agree relationship, and that

the higher position has an EPP-feature. The movement proposed here is problematic

on both counts. First, it is unclear what kind of feature would cause an Agree relation

between T and the preverb. T must be in an Agree relation with the verb in order for

V-to-T movement to take place. For T to also Agree with the preverb, we would need

to stipulate that T has unvalued preverb features in addition to V features. The second

and perhaps more serious problem is that of the EPP-feature. The EPP on T must be

optional, as Spec-TP will only be filled when the verb is compound. However,

optional EPP-features are only allowable in the minimalist program if they have some

kind of effect on interpretation (Chomsky 2001: 34). This does not seem to be the

case here; fronting the preverb does not seem to have a pragmatic or discourse effect.

Movement of the preverb to Spec-TP would have to be obligatory whenever the verb

is compound; the presence of an EPP-feature on T would have to be lexically

conditioned. This is not possible within the theory as it stands. From a theoretical

standpoint, then, although movement of the preverb as an XP is possible, it cannot be

motivated, and so cannot be maintained as an explanation for the presence of the

preverb in C.29

3.4.4 Remnant-movement

In the last two sections it has been argued that neither head-movement nor XP-

movement can provide a satisfactory account of Old Irish compound verbs. In the

case of head-movement, the required movement processes violate locality constraints

and in the case of XP-movement the movements cannot be motivated. In this section

we will examine a final syntactic account, namely VP-remnant-movement.

28 Although, as we saw in section 3.3 the status of CHP’s filled C condition is in doubt. If simple verbs do not move to C then when the verb is simple C will not be filled. Therefore, there can be no filled C condition. 29 Ian Roberts (p.c.) points out that it may be possible for the preverb to move straight to C when C is unfilled, and to Spec-TP only when movement to C is prevented by the presence of some other element in C. This again faces the same problems in terms of motivating the movement processes, especially if as argued in section 3.3 there is no requirement for C to be filled. In addition, there are also familiar problems in terms of lookahead.


47

Remnant-movement was originally proposed by Thiersch (1985) and den Besten &

Webelhuth (1987, 1990) to account for the topicalisation of incomplete VPs in

Germanic languages. In the current syntactic climate where head-movement is

assumed by many to play no part in narrow syntax (Chomsky 2001), remnant-

movement is often employed to replace it.

Remnant-movement involves the movement of an entire phrase (usually VP or vP)

to a specifier position higher in the clause (usually the specifier of T or C). Before this

phrasal-movement takes place all the other contents of the VP apart from the verb

must be extracted from the phrase, leaving simply a VP remnant, containing the verb

and a number of traces.

There are a growing number of analyses of verb-initial languages that invoke VP-

remnant fronting (Massam 2000, 2005, Rackowski and Travis 2000, Lee 2000, 2005,

Oda 2005). It is argued that in these verb-initial languages the EPP-feature associated

with T does not target the subject as it does in subject-initial languages, but instead

targets the predicate of the clause. Verb-initial languages tend also to show predicate-

initial orders in copula clauses, as shown by the example from Niuean in (31).

31. ko’ Mele e faiaoga

COP Mele ABSL teacher

‘The teacher is Mele’ (Massam 2000: 104)

If verb-initial order results from VP fronting, then a parallel can be drawn between

verb-initial clauses and non-verbal-predicate-initial clauses, with both employing XP

fronting to satisfy T’s EPPPRED-feature.30 Old Irish shows predicate-initial orders in

copular clauses, such as that given in (32) below (Thurneysen 1946: 492–3).

32. is fota in troscud

COP long the fasting

‘The fasting is long’ (LL 13038)

So there is reason to believe that verb-initial order may involve VP-remnant-fronting.

30 Although see McCloskey (2005) who argues that this parallel is not so straightforward in Modern Irish.


48

Let us examine now what advantages a remnant fronting analysis would have for

Old Irish. If, as suggested in the previous section, the preverb is in the specifier of VP,

then when the VP is fronted to Spec-TP, both the VP and the verb in V appear clause-

initially. This movement is consistent with locality constraints and is well motivated if

we assume an EPPPRED feature on T. However, a remnant-movement approach faces a

number of empirical problems.

The first issue concerns differentiating deuterotonic and prototonic forms. If the

entire VP moves to Spec-TP, then the preverb (Spec-VP) and the verb (V) will move

together. However, for there to be a syntactic distinction between deuterotonic and

prototonic forms, i.e. for the initial preverb of deuterotonic compounds to be in C,

then the preverb must be able to move out of the moved VP to C when C is unfilled.

This movement operation is prohibited in terms of Huang’s (1982) Condition on

Extraction Domains (CED), which forbids extraction from a specifier position.

Therefore, under a VP-remnant-movement approach the difference between

deuterotonic and prototonic forms cannot be determined syntactically.

The second issue concerns object pronouns. One of the main pieces of evidence

used by Massam (2000) to argue that Niuean is a VP-fronting language is that in

certain contexts the object fronts with the VP. The position of object pronouns could

perhaps provide similar evidence for Old Irish. As Vendryes (1908) observes, the

object pronoun always appears as part of the verbal complex in Old Irish. This could

be explained if the object pronoun moves as part of the VP to Spec-TP. However, it is

not simply the case that the object pronoun appears within the verbal complex,

crucially it always appears in clause-second position. If the object pronoun moves as

part of the VP, then we would expect it to appear as the complement of the verb, so it

would follow the verb and the preverb. In cases where the clause also contains a

conjunct particle, the object pronoun would be in clause-fourth position. If the

pronoun moves as part of the VP, it must move on to clause second position. If this

movement is syntactic, it violates the CED. If it is post-syntactic, it is unclear exactly

how it would work and why it would take place. The position of the object pronoun,

then, cannot be taken as evidence for remnant-movement. In fact such an account runs

into serious difficulty in explaining the position of the pronoun.

The third and most damaging argument against a remnant-movement account is

the position of non-pronominal objects. If we are to employ remnant fronting it is

crucial that we can motivate the movement of every other element from the VP. This


49

means both direct and indirect objects. Movement of the object out of VP to the

specifier of v can be motivated on theoretical grounds, however empirical evidence is

harder to come by. Bobaljik & Carnie (1996) argue that the object optionally moves

out of the VP in non-finite clauses in Modern Irish. Even if this is the case, optional

object movement in a subset of clauses in Modern Irish can by no means be taken as

evidence that the object moves across the board in Old Irish. If evidence can be found

to show that movement of the direct object takes place, it would be very difficult to

prove that all other constituents, such as indirect objects, adverbs, prepositional

phrases, small clauses, etc. move out of the VP too. The movement of the direct and

indirect object from the VP are crucial to any remnant fronting analysis and as neither

can be shown convincingly in Old Irish, the remnant fronting approach must be

abandoned.

3.5 Summary

In this section we have examined the existing syntactic accounts of the double system

of inflection in Old Irish. In section 3.2 it was argued that Doherty’s (1999, 2000)

proposal, based on the idea that Old Irish is a residual V2 language can be ruled out

on both theoretical and empirical grounds. At first sight, Carnie, Harley & Pyatt’s

(2000) account seems to fare better both empirically and theoretically. However, a

closer look at the Old Irish data in section 3.3 suggests that this is not the case.

Evidence from the placement of object pronouns and relative clauses suggests that,

although at an earlier stage in the history of Old Irish simple verbs moved to C, in the

Classical Old Irish period they only move as far as T. In section 3.4 we turned our

attention to compound verbs. Empirically, it seems that the data from Old Irish

support CHP’s analysis in this respect. The initial preverbs of compound verbs in Old

Irish behave phonologically and syntactically like conjunct particles, suggesting that

they both appear in the same syntactic position, namely C. Although the empirical

evidence supports this conclusion, it faces seemingly insurmountable theoretical

problems. Although the initial preverb is syntactically independent, it is semantically

part of the verb. Therefore, we expect it to be merged with the verb low in the

structure and to move to C. Having examined the possibilities of head-movement,

XP-movement and remnant-movement it seems that all of these options are flawed.

The only conclusion that remains to be drawn, then, is that the orders shown by

compound verbs in Old Irish cannot be derived syntactically.


50

To reiterate, the two main conclusions of this chapter so far are that simple verbs

do not move any higher than T in Old Irish and if the initial preverbs of compound

verbs are in C, which the empirical evidence seems to suggest they are, then they

cannot move there syntactically. If the first of these conclusions is true, then the

morphological form of absolute verbal forms cannot be a result of the verb’s syntactic

position, and so an alternative non-syntactic explanation is necessary for the different

verbal endings. Similarly, if the initial preverb of compound verbs does not move to C

in the syntax then we need an alternative explanation for this too. The overall

conclusion that can be drawn at this point is that syntax alone cannot account for the

Old Irish verbal system; we need something more.

4 THE SYNTAX-PHONOLOGY INTERFACE AND THE OLD IRISH VERBAL SYSTEM

4.1 The syntax-phonology interface

If, as argued in the previous section, simple verbs only move to T and the initial

preverb of compound verbs cannot move to C in the syntax then the distribution of the

different verbal forms in Old Irish cannot be a direct result of syntactic movement

operations. For all types of verb the syntactic configuration will be identical, with the

verb (and preverb) appearing in T. To explain how the different forms come to be

inserted, then, it seems that we need more than syntax. Something must take place

post-syntactically to provide the conditions necessary for the insertion of

absolute/conjunct or deuterotonic/prototonic forms. The aim of this section is to

establish what these post-syntactic operations might be.

The first step is to consider what kinds of post-syntactic operation can take place

cross-linguistically. Post-syntactic operations can be divided into two kinds, those that

are conceptually motivated and so must occur in all languages, and those that are

empirically motivated and only occur in individual languages. Let us examine each in

turn.

According to Chomsky’s architecture of the grammar, after syntax is complete, the

derivation proceeds to the phonological component. However, the structures

manipulated by syntactic operations are somewhat different to those manipulated by

phonology. Syntactic structures are hierarchical, whereas phonological structures are

flat. Furthermore, syntactic operations refer purely to morphosyntactic features,

whereas phonological operations refer only to phonological features (Chomsky &

Halle 1968). Therefore, after syntax is complete, but before phonology begins there


51

are two post-syntactic operations that must take place in all cases: hierarchical

syntactic structures must be linearized and (assuming Distributed Morphology)

morphosyntactic feature bundles must be replaced by phonological feature bundles.

Chomsky (1995a: 340) argues that the operation responsible for the linearization of

syntactic structure might be Kayne’s (1994) Linear Correspondence Axiom (LCA).

For Chomsky the LCA is operative at the syntax-phonology interface and is

responsible for the conversion of the hierarchical output of syntax to the linear order

found in phonology.31 We will follow Chomsky on this point and little more will be

said about it below.

The idea that syntactic operations make no reference to phonological features, and

phonological operations make no reference to morphosyntactic features can be easily

explained if we assume that there are no phonological features present in the syntax,

and no morphosyntactic features present in phonology. This is one basic assumption

behind Distributed Morphology (DM – Halle & Marantz 1993, 1994). DM assumes

an operation Vocabulary Insertion, which takes place at the syntax-phonology

interface and replaces morphosyntactic feature bundles with phonological exponents

(more will be said on this operation below).

So far we have outlined two post-syntactic operations that must take place cross-

linguistically to convert syntactic structure to phonological structure, namely

Linearize and Vocabulary Insertion. However, there is a further conceptually

necessary post-syntactic operation, namely Chain Reduction.

If, following Chomsky (1995a, 2000 et seq.), we adopt a copy theory of

movement, then at the output of the syntax the derivation will contain multiple copies

of any element that has been moved. As noted in chapter 1, Chomsky (2001) argues

that Move can be seen as Internal Merge. The element to be moved is copied and

remerged in the higher position. So, for example in a passive construction, such as

(33), the object John is merged as the complement of the verb, in the VP, and then

copied and remerged in the subject position.

33. [TPJohn was [VPkissed tJohn]]

31 Kayne (1994) maintains that the LCA is operative throughout the syntax.


52

At the output of syntax, then, there will be two copies of the DP John. However, only

one of these copies can be phonologically realised as sentences of the type *John was

kissed John are ungrammatical.32 This suggests that there must be some operation that

marks one copy of a set of identical elements to be realised and deletes the subsequent

copies. This operation, which we will term Chain Reduction, marks the leftmost, or

highest copy of the element in question for realization and deletes the lower copies

(see Brody 1995, Groat & O’Neil 1996, Pesetsky 1997, Nunes 1999, 2004, Bobaljik

2002 ).33

So far it has been argued then, that the syntax-phonology interface contains three

necessary operations: Linearize, Chain Reduction and Vocabulary Insertion. These

operations are conceptually motivated, and so must take place in all languages.

Within DM, however, there are a number of further operations that are argued to take

place between syntax and phonology. These operations manipulate the output of

syntax, changing the morphosyntactic feature structure of syntactic terminals and so

influencing the way that these terminals are realised phonologically (e.g. fission,

fusion, impoverishment see Halle & Marantz 1993, Harley & Noyer 1999 for further

details). These operations operate only in particular contexts in particular languages.

Having outlined the two types of post-syntactic operation it must now be

established which type should be invoked to account for the Old Irish data. Adger (to

appear) provides an analysis using language specific post-syntactic rules. This is

outlined in section 4.2. In section 4.3 an alternative approach is developed, that relies

predominantly on conceptually motivated, obligatory post-syntactic operations.

4.2 Post-syntactic movement and the Old Irish verb: Adger (to appear)

Adger (to appear), building on work by Marantz (1988) and Embick & Noyer (2001)

within the framework of DM, proposes that the different verbal forms in Old Irish

result from post-syntactic movement processes.

Marantz (1988) argues that syntactic structures can be rebracketed after syntax

through Morphological Merger, whereby the adjacency relation between two heads X

and Y is replaced by the affixation of X to Y. Embick & Noyer (2001) argue that 32 Although see McDaniel (1989) and Nunes (1999) for examples of so-called multiple Spell-Out. 33 Nunes (1999, 2004) attempts to explain why this should be the case, whereas others simply stipulate. It could simply be a principle of the grammar that the leftmost copy of identical feature sets is realised. If the Spell-Out of syntactic terminals operates from left to right, then it could simply be the case that as a maximally economical operation, Spell-Out simply ignores subsequent copies of an element it has already spelled-out.


53

there are two types of Morphological Merger: Lowering and Local Dislocation.

Lowering adjoins a head to the head of its complement. Local Dislocation adjoins a

head to another head that is linearly adjacent to it. These two operations differ in

terms of the types of structures they act upon. Lowering specifies a hierarchical

relationship between a head and its complement, whereas Local Dislocation specifies

a linear relationship. This difference reflects the fact that these operations take place

at different points in the derivation. Lowering takes place before linearization and

Local Dislocation takes place afterwards. Adger (to appear) argues that it is the latter,

i.e. Local Dislocation, that can account for the double system of verbal inflection in

Old Irish.

Lithuanian has compound verbs that behave somewhat similarly to those found in

Old Irish. In Lithuanian the reflexive pronoun –si is a second position clitic. When the

verb is simple it appears as a suffix (34). When, however, the verb is compound it

appears between the preverb and verb (35) (Senn 1966).

34. Laikaũ ‘I consider, maintain’

Laikaũ-si ‘I get along’

35. Iš-laikaũ ‘I preserve, withstand’

Iš-si-laikaũ ‘I hold my stand’

Embick & Noyer (2001:579f) argue that the preverb is adjoined to the verb and then

this complex verb raises to T. The reflexive clitic adjoins to the left of T.

36. T

Si T V T

PV V

The first operation that takes place post-syntactically is that T undergoes Local

Dislocation with V. T is specified as a suffix, so this operation causes no change in

order, however it results in a different structure.


54

37. T Si T V

PV V V T

The clitic then undergoes Local Dislocation with the preverb. The clitic is specified as

a suffix and so is spelled-out to the right of the preverb. In the case of simple verbs,

the clitic undergoes Local Dislocation with the entire [V+T] complex, and appears as

a suffix.

38.

T T V

PV V

PV si V T

In order to obtain this result, the Left Dislocation of V and T must take place first, and

form a complex morphological word. Embick & Noyer (2001: 580) argue that the

ordering of these two operations is predicted by the principle of cyclic application –

Local Dislocation processes always operate from the more deeply embedded structure

outward.

Adger (to appear) builds on Embick & Noyer’s analysis of Lithuanian, arguing that

the phonological and morphological distinctions between the independent and

dependent verb forms in Old Irish can be accounted for in a similar way. Adger

assumes that Old Irish has an articulated C structure (cf. Rizzi 1997), and that Force is

subcategorised to be enclitic to an X0. This Force head behaves like the clitic in

Lithuanian. During syntax, the compound verb (preverb+verb) raises to T. Assuming

that conjunct particles appear as the head of Finiteness, Force is positioned

syntactically to the left of the entire verbal complex in the syntax.


55

39.

ForceP Force FinP Fin TP C V T PV V

When a conjunct particle is present, Force Locally Dislocates with the conjunct

particle.

40. Force-C-PV-V-T > C-Force-PV-V-T

This operation has no phonological or morphological effect on the verb, which will

then be spelled-out in conjunct or prototonic form. When there is no conjunct particle

and the verb is compound, Force Locally Dislocates with the preverb.

41. Force-PV-V-T > PV-Force-V-T

Force appears between the preverb and the verb and so the verb is spelled-out in

deuterotonic form. When the verb is simple, Force Locally Dislocates with the [V+T]

complex, resulting in absolute morphology.

42. Force-V-T > V+T-Force

Adger argues that this Force head has a number of effects. As we saw in section 2.2

the main phonological distinction between the deuterotonic and prototonic verb forms

is the difference in stress placement. To account for this Adger proposes the following

alignment constraint:

“Align the left edge of a phonological phrase (or perhaps intonational phrase) with the

right edge of Force”


56

This entails that the Force head will always be followed by the left boundary of a

phonological phrase. Assuming that stress is assigned to the first syllable of a

phonological phrase, stress will always follow the Force head.

The difference in phonological phrasing in compound verbs can also be used to

explain the different forms of the compound verb. When there is a conjunct particle,

the verb and preverb are in the same phonological phrase. When there is no conjunct

particle, the preverb and verb are in separate phonological phrases. This provides an

explanation as to why lenition, elision and assimilation are blocked between the initial

preverb and the remainder of the verb when the verb is deuterotonic and not when the

verb is prototonic.34 35

Let us turn now to simple verbs. When the verb is simple and there is no conjunct

particle, the Force head will Locally Dislocate with and so appear after the V+T

complex. When Force is cliticized to T absolute morphology will be inserted; in all

other cases conjunct morphology will be chosen. This provides a neat account for the

distribution of absolute and conjunct endings. Absolute endings are only found on

simple verbs in initial position. Non-initial simple verbs and both deuterotonic and

prototonic compound verbs show conjunct inflection. In the case of non-initial simple

verbs or deuterotonic and prototonic compound verbs the Force head attaches to

either a conjunct particle or a preverb. It is only when there is no conjunct particle and

no preverb, i.e. the verb is simple and in absolute initial position, that the Force head

will be attached to T.

Adger’s post-syntactic movement account provides a neat explanation of the

distribution of the different verbal forms in Old Irish. However, it also faces various

problems, both empirical and theoretical. Let us examine each in turn.

One area of the data that Adger’s analysis does not account for well is object

pronouns. As discussed in section 2.3 above, object pronouns in Old Irish appear in

clause-second position. Adger argues that in the syntax object pronouns move to the

34 Adger seems to suggest here that the deuterotonic and prototonic forms are derived from a single underlying form. As discussed in section 2.2 more research is necessary to establish whether this is the case. 35 If, as argued in section 3.4.1 above, the initial preverb of compound verbs occupies the C position, then the phonological shape of compound verbs can be accounted for if the left edge of the TP aligns with the left edge of a phonological phrase (cf. Selkirk 1986, 1995; Truckenbrodt 1995; Inkelas & Zec 1995). In this case the initial preverb of a deuterotonic compound will appear in a separate phonological phrase to the remainder of the verb. This can explain the position of the stress and also why phonological processes cannot take place between the preverb and the remainder of the verb without the need for an extra Force head.


57

Topic position of a Rizzian articulated CP structure (see Roberts 2005: 135 for a

similar proposal).

43.

ForceP

Force Top Top FinP Pronoun Fin TP C V T PV V

The object pronouns then Locally Dislocate post-syntactically in the same way as the

Force head. So, Adger contends, there are two possible derivations. In the first case

Force Locally Dislocates first with the pronoun, forming a new compound word

which then Locally Dislocates with the preverb or the [V+T] complex (depending on

whether the verb is simple or compound).

44. Force-Pronoun-PV-[V+T]

[Pronoun+Force]-PV-[V+T]

[PV+Pronoun+Force]-[V+T]

The second option is that the pronoun Locally Dislocates with the preverb or [V+T]

and then the Force head Locally Dislocates with the [[V+T]+Pronoun] or

[PV+Pronoun] compound.

45. Force-Pronoun-PV-[V+T]

Force-[PV+Pronoun]-[V+T]

[[PV+Pronoun]+Force]-[V+T]

Adger prefers the first option. In the case of simple verb, when the object pronoun is

suffixed to the verb the verb always has absolute morphology, and never special

relative morphology (see section 3.3.2 above). Adger argues that special relative


58

morphology appears on the verb when the Force head that has Locally Dislocated and

attached to T is specified as relative. If the Force head were closer to the verb we

would expect the verb to show relative inflection in relative clauses; however, this is

not the case. Therefore, this supports the view that it is the pronoun that is closer to T,

and that the pronoun conditions absolute morphology and overrides Force. Similarly,

if Force intervened between T and the pronoun, we would assume that suffixation

would be impossible as Force marks the left edge of a phonological phrase, and so the

verb and the clitic would be in separate phonological phrases.

Although Adger’s analysis can account well for simple verbs, compound verbs are

more problematic. When a pronoun is infixed between a preverb and the rest of the

verb, or between a conjunct particle and the verb, the pronoun causes the initial

segment of the verb to mutate. Each pronoun has its own mutation (e.g. first and

second singular lenite, third singular masculine nasalizes etc.). Adger (to appear: 39)

describes the situation as follows:

“If the pronoun were to dislocate first, adjoining to the preverb, and then Force were

to further adjoin, Force would be hierarchically superior, and might be expected to

exert its own mutation [on the rest of the verb – GEN] overriding that of the

pronouns.”

There are multiple problems with this argument. From an empirical point of view, the

structure Adger proposes does not make the right predictions. If the Force head

linearly follows the pronoun we would expect it to block processes such as lenition

and nasalization rather than facilitate them. The pronoun and the following verb will

be in separate phonological domains and therefore such processes will be blocked as

they are between the preverb and verb of compound verbs. In fact, from a historical

perspective it would be preferable for the Force head to precede the pronoun. One of

the puzzles that philological accounts of the double system of verbal inflection have

to deal with is the fact that the enclitic pronouns themselves are not lenited. We would

expect the pronouns to be lenited after preverbs ending in a vowel, yet they are not.

This suggests that if there was ever any kind of element that conditioned the different

sets of inflectional endings, it must have appeared between the preverb and the

pronoun and not between the pronoun and the remainder of the verb.


59

Adger’s account of object pronouns also faces a theoretical problem. For Local

Dislocation to proceed as Adger suggests, we must abandon Embick & Noyer’s

insight that Local Dislocation affects the most deeply embedded structures first. If the

pronoun Locally Dislocates with the Force head first and then this Dislocates with the

preverb and verb lower in the structure this is counter-cyclic. If post-syntactic

movement exists, it must be properly constrained. If we adopt Adger’s account then

the strict ordering of Local Dislocation processes as proposed by Embick & Noyer

cannot be maintained and so the operation becomes less constrained.

Rejecting the cyclic operation of Local Dislocation is not the only way that

Embick & Noyer’s conception of post-syntactic movement must be altered to account

for the Old Irish data. For the Force head to have a noticeable effect on the verbal

morphology Local Dislocation must occur before Vocabulary Insertion, i.e. before the

phonological forms are inserted into the derivation. However, Embick & Noyer

(2001: 562) maintain that Local Dislocation must take place after Vocabulary

Insertion, as it affects linearized strings, and in their view, linearization and

Vocabulary Insertion occur at the same time. Adopting this view, then, if Local

Dislocation precedes Vocabulary Insertion and linearization, as it would have to

under Adger’s account, then it acts on the same structures as Lowering. This again

has the effect of making post-syntactic movement less constrained and conceptually

less desirable.36

A further theoretical problem with Adger’s analysis is the postulation of an enclitic

Force head. In some languages an articulated CP structure is well supported as the

proposed syntactic positions receive overt phonological realisation. For example, in

Italian the distribution of certain complementizers alongside topics and focussed

elements provides good motivation for an articulated CP (Rizzi 1997). No such

evidence is available for Old Irish. Adger’s Force head has no phonologically overt

counterpart and it seems no real semantic, pragmatic or discourse function. Its only

role is to achieve the desired forms in the syntax. The fact that the Force head seems

to have no interface properties leads to the question of how it would be acquired.

Furthermore, even if the Force head itself could be acquired, how would children

36 This problem disappears under a different conception of Spell-Out that allows more middle ground between Linearization and Vocabulary Insertion, such as that proposed by Ackema & Neeleman (2005). In this case, the distinction between Lowering and Local Dislocation can be maintained, as both will precede Vocabulary Insertion, but Lowering will precede Linearization and Local Dislocation will follow it.


60

acquire the fact that this phonologically null element is enclitic? Although this Force

head seems to derive the required forms in most cases, neither the Force head itself

nor its consequences can be independently motivated.

In sum, Adger’s account faces various problems. Most notably, from a theoretical

perspective, the changes that need to be made to Embick & Noyer’s theory of post-

syntactic movement to account for the Old Irish data lead to a less restrictive theory.

However, even in its original form, it is not clear that post-syntactic movement is

sufficiently restrictive in terms of what motivates the operations and what constrains

their operation. In addition, a level between syntax and phonology that duplicates

syntactic movement is conceptually undesirable. Both Lowering and Local

Dislocation involve downward movement, an operation that is strictly forbidden in

the syntactic component. Allowing such operations detracts from the explanatory

force of syntax. Furthermore, Local Dislocation has essentially the same effect as

Prosodic Inversion (Halpern 1995). Embick & Noyer maintain that the two are

different, and that both need to be maintained. If this is the case, then the

Morphological level seems to duplicate not only syntactic operations, but also

phonological ones. Such duplication should not exist in an optimal system. The

conceptual problems faced by post-syntactic movement suggest that an alternative is

needed. This is the aim of the following sections.

4.3 A new post-syntactic account

4.3.1 Simple verbs and do-support: a cross-linguistic parallel

We saw in section 3.3 above that one piece of the Old Irish data that is particularly

problematic for CHP is the use of object pronouns with simple verbs. CHP assume

that suffixed pronouns are regular and productive, however, a closer look at the Old

Irish data shows that this is not the case. As noted above, suffixed pronouns are only

used when the pronoun and the verb are both third person, and even then this usage is

not consistent. The productive pattern in Old Irish is to insert the dummy preverb no

and infix the pronoun between the preverb no and the verb, which, as it is not in initial

position, appears with conjunct inflection, as shown in (46).

46. (a) no-s nguid-som

PVB-INF.3PL beseech.PRES.3SG.CONJ-emph.part.3SG.M

‘He beseeches them’ (Wb 25b9)


61

(b) ni hed no-t beir í nem

NEG it PVB-INF.2SG carry.PRES.3SG.CONJ in heaven

‘It is not this that brings you into heaven’ (Wb 6c9)

The use of the preverb no in this case bears some resemblance to the use of do-

insertion in English. It is well known that verbs do not move to T in English, and what

is often said is that verbs are united with their inflections through some kind of

downward movement from T to V, so-called Affix-Hopping (Chomsky 1957). When

Affix-Hopping is impossible, for example when a negative head intervenes, the

inflections are stranded in T and do is inserted in order to provide a host for them. A

parallel situation could be envisaged for Old Irish. In Old Irish the verb does not raise

to C; instead absolute inflections Affix-Hop from C to T. When an infixed pronoun is

present, this prevents Affix-Hopping and so the dummy preverb no must be inserted

to host the absolute inflections in C.

Traditionally Affix-Hopping has been viewed as problematic as it involves

downward movement. Downward movement is prohibited in the syntax, as a moved

element must c-command its trace.37 Under the current version of minimalism,

however, an alternative analysis is available. The appearance of tense and subject-

agreement-(φ)-features on the verb in its base position can be seen as a result of the

operation Agree. T and V both have tense and φ-features. The operation Agree values

unvalued features, therefore, after Agree has taken place the tense and φ-features on T

will be identical to those on V. At the point of Spell-Out, one set of these features

receives a phonological realisation. What appears to be downward movement, then, is

the realisation of these tense and φ-features on V rather than T. The implementation

of an Agree relation between V and T can explain how tense and φ-features can be

present on both V and T, and so can potentially be realised in either position;

however, this does not explain why these features are sometimes realised in V and

sometimes in T.

It is perhaps possible to account for the position in which tense and φ-features are

realised through the operation of Chain Reduction. In section 4.1 we saw that in

canonical cases Chain Reduction marks the leftmost copy of a moved element for

37 Embick & Noyer (2001: 584–591) and Bobaljik (2002) account for Affix-Hopping via Lowering. As argued in section 4.2, post-syntactic movement is conceptually undesirable. As a result we will discuss this approach further here.


62

realisation and deletes all subsequent copies. To account for Affix-Hopping as

proposed above, the concept of Chain Reduction needs to be extended so that it

applies not only to movement chains but also to features in an Agree relation. At first

sight this seems problematic. Once feature valuation has taken place between two

features there is no link between them; therefore, we cannot talk about chains between

valued features. However, Chomsky also makes such a claim about moved elements.

Chomsky (2001: 11) suggests that chains cannot be considered ‘real’, as the

postulation of chains or indices to mark multiple instances of a moved element would

violate the Inclusiveness Condition. If there is no concrete link between moved

elements how does the operation Chain Reduction know which of these elements

form a chain and should be subject to deletion? The two main possibilities are identity

and c-command.38 After feature valuation has taken place, a pair of features in an

Agree relation will by necessity be identical. Furthermore, in order for an Agree

relation to take place in the first place the Probe and Goal must be in a c-command

relation.39 So, it seems that the identical features resulting from Agree fulfil the same

requirements as moved elements in terms of Chain Reduction.40

Under this revised view of Chain Reduction then we would expect the tense and φ-

features to be realised in the highest or leftmost position, namely T. However, as we

saw above, this is not always the case. Let us consider the operation Chain Reduction

in more detail. So far it has been assumed that Chain Reduction is a post-syntactic

operation, and the decision as to which copy is marked for realisation is determined at

the PF-interface.41 If this is the case, then it seems plausible that this decision should

be determined by PF requirements (Landau 2006: 54):

38 Nunes (1999) argues that chains cannot be determined in terms of identity as, if this were the case, why is one occurrence of John not deleted in sentences such as Johni hit Johnj. Clearly each instance of John has a different referent, and they are, therefore, distinct. However, it is not clear how this should be accounted for within the syntax. 39 Chomsky does not explicitly mention c-command in his definition of Agree, however, he does state that the Probe searches for a Goal in its domain/complement. This will result in a c-command relation between Probe and Goal. 40 It is plausible that in some situations more than one instance of a pair of valued features will be phonologically realised. However, this also occurs in movement chains, for example in cases of multiple Spell-Out in German (McDaniel 1989). This can be explained through some independent phonological requirement demanding that the lower copy be realised (Landau 2006). 41 Landau (2006) argues that this is a result of the modular nature of the grammar – the decision as to which copy to pronounce/interpret is determined at PF/LF respectively, as there can be no interaction between the two.


63

47. (a) PF copies that are demanded by PF requirements cannot be deleted

(b) PF copies that are excluded by PF requirements must be deleted

For example, Franks (1999) argues that second-position clitics in Serbo-Croatian

move to C in the syntax and are usually spelled-out there. However, these clitics need

a host to their left. If no such host is available the top copy, i.e. that in C, cannot be

realised, and a lower copy is spelled-out instead. Similarly, Bobaljik (2002) argues

that in cases of object shift, the highest copy of the object cannot be realised if it

appears between V and T. V and T must be string adjacent in order for morphological

merger to take place. An intervening object that interrupts this adjacency requirement

cannot be realised, and so a lower copy will receive a phonological realisation instead.

It seems that a similar argument could be made for the spell-out of tense and φ-

features in English. If, as is often assumed, tense and φ-features in English are affixes,

then in order to be realised they must have a host, i.e. they must satisfy the Stranded

Affix Filter (SAF – Lasnik 1981, 1995). The SAF can be seen as a PF requirement

(Halle & Marantz 1993; Lasnik 1995; Bobaljik 2002; Landau 2006). Therefore, if at

PF, tense and φ-features have no host and are stranded in T they cannot be realised

there, and so the lower copy in V will be spelled-out instead, resulting in Affix-

Hopping. So, it seems then we have the beginnings of an explanation as to why tense

and φ-features are sometimes spelled-out in T and sometimes in V. However, there

are several aspects of the above proposal that are in need of further clarification. First,

what exactly does it mean for a feature to be an affix? Second, what constitutes ‘a

host’? In other words, how is the SAF satisfied? Let us examine each of these issues

in turn.

The first issue to be addressed is what it means for an entity to be an affix. If, as

suggested above, the SAF is a PF-requirement, operative at the syntax-phonology

interface, then [affix] cannot be a purely phonological property, as the SAF must be

satisfied before the phonological features enter the derivation. Halle & Marantz

(1993) argue that Vocabulary Insertion, i.e. the replacement of morphosyntactic

features with phonological features, is the last stage in the post-syntactic component,

marking the beginning of phonology proper (see also Ackema & Neeleman 2005:

185–6). At the point of Chain Reduction, then, phonological information is not yet

available. Therefore, the property [affix] must be encoded in the morphosyntactic

features.


64

There is empirical evidence to suggest that affixation is not purely a phonological

process. In addition to their obvious phonological deficiency, affixes also seem to

show a certain level of syntactic dependency. Unlike clitics, which can in many cases

appear freely with any type of host, affixes tend to be restricted in their distribution,

appearing only in a particular syntactic position with a particular type of host (Zwicky

& Pullum 1983: 504–5). Often, if a clitic’s phonological requirements are not met, it

seems that it can move in the phonology to find an appropriate host (so-called

Prosodic Inversion – Halpern 1995). This is not generally true of affixes.42 This

suggests that affixation, or at least part of affixation, takes place before the insertion

of the phonological material, and so the property [affix] must be specified in the

morphosyntactic features. It will be assumed in what is to follow that [affix] is a

morphological property that is associated with individual features, i.e. a

morphological subfeature relevant at the PF-interface.43

Having established what it may mean for an entity to be an affix, let us now

consider what may constitute a host and how the SAF can be satisfied at PF. It was

argued above that [affix] is, at least in part, a morphosyntactic property, therefore

affixation must be, at least in part, a morphosyntactic operation. A feature that is

specified morphosyntactically as [affix] requires a morphosyntactic host. What this

means, I propose, is that it must be able to combine in the syntax or the post-syntactic

component with another morphosyntactic feature. In the simplest case, what this

means is that a feature with the property [affix] will satisfy the SAF if it appears

under the same syntactic terminal node as another morphosyntactic feature.44 One

final question that must be raised is whether the morphosyntactic feature that provides

a host for the affixal feature requires any specific properties. I would suggest that the

only requirement is that the feature in question has a positive value. As discussed in

chapter 1, by the PF-interface, where the [affix] feature is relevant, all

morphosyntactic features will necessarily have been valued by the syntactic operation

Agree. However, these values are not necessarily positive. If a feature has its default

42 The distinction between affixes and clitics is not clear-cut. There is a tendency within the DM literature to assume that there is no distinction between them. It is perhaps possible that a distinction can be drawn between entities that are specified as morphosyntactically affixal and those that are phonologically affixal. However, we will not go any further into this issue here. 43 It should be emphasised at this point that this is entirely distinct from the syntactic Affix-feature that motivates head-movement during the syntax. 44 This does not rule out the possibility of post-syntactic operations such as Morphological Merger in the sense of Marantz (1988). This may be an alternative way in which the affixation requirement can be met. However, we will not investigate this possibility further here.


65

value, then it could be argued that its feature specification is 0. For example, present

tense, third person, singular number could all be seen as a lack of a positive value.

Features that lack a positive morphosyntactic value can have an LF interpretation and

can also be realised at PF; however, I propose, they cannot play a role in the post-

syntactic, morphological component (see Harley 1994, Harley & Ritter 2002 for a

similar proposal).

Having outlined our basic assumptions regarding the nature of affixation, let us

return to the case in hand, namely tense and φ-features in English. These features in

English are morphosyntactically specified as [affix]; therefore, in order to satisfy the

SAF and receive a phonological realisation they must occur under the same terminal

node as at least one other non-default-valued feature. When there is an aspectual or

modal auxiliary in English, there will clearly be aspectual or modal features in T.

Therefore, the [affix] property will be satisfied, and the tense and φ-features will be

realised in T and deleted in V. When there is no aspectual or modal auxiliary, for

example in the simple present or simple past tenses then there will be no positively-

valued features with which the tense and φ-features can combine, and as such they

cannot be realised in T and are spelled-out in V. It seems, then, that by implementing

the SAF Affix-Hopping in English can be accounted for by the operations Agree and

Chain Reduction, crucially without recourse to post-syntactic movement. Before we

consider the related phenomenon of do-support, let us turn to Old Irish.

It was argued in section 3.3 above that the verb in Old Irish does not raise above T.

Therefore, to account for the appearance of absolute morphology on the verb we seem

to have a case of downward movement parallel to Affix-Hopping. When C is not

filled by a conjunct particle the absolute verbal endings move from C to T. As with

Affix-Hopping above, this could also be seen as a result of Chain Reduction. If the

feature conditioning absolute inflection appears in both C and T, and like tense and φ-

features in English has the property [affix], the position in which it appears will be

determined by the morphosyntactic features contained in C and T. When C contains

positively-valued morphosyntactic features, these will provide a morphosyntactic host

for the affixal features, satisfying the SAF and allowing the realisation of these

features in the leftmost position. When C contains no positively-valued features,

however, the affixal features cannot be realised in C, as a result of the SAF, and so

will be spelled-out in T. Let us consider this proposal in more detail.


66

The first issue to consider is what the feature(s) shared by C and T in Old Irish

might be. Absolute inflections express subject agreement; therefore we would expect

them to be a realisation of φ-features. However, conjunct inflections also mark subject

agreement, and so there must be some factor that distinguishes the two. Under CHP’s

view, the difference between absolute and conjunct inflection could simply be a result

of the fact that the verb occupies different syntactic positions, with φ-features being

spelled-out as absolute inflection when they appear with a verb in C and as conjunct

when they appear with the verb in T. However, as we saw in section 3.3 above, there

is evidence to suggest that the verb does not raise above T in Old Irish. If this is the

case, then φ-features will always be realised on the verb in T, and so the distinction

cannot be explained in this way.

A further possibility is that absolute endings are the realisation of φ-features

combined with some other feature. It is difficult, however, to determine exactly what

this feature might be, as the double system of verbal inflection seems to have no clear

function in the Old Irish period.45 One possibility is that it is a force feature. As we

saw in section 3.3.2 above, in certain numbers and persons absolute verb forms can

show special relative endings. The existence of relative and non-relative absolute

endings could be seen as the presence of relative and non-relative force features in the

T position. When a relative force feature combines with T’s φ-features this results in

special relative morphology, and when a non-relative force feature combines with T’s

φ-features this results in absolute morphology. We will assume for what follows that

the relevant feature that is shared by C and T is a force feature.46 47

In the case of Affix-Hopping in English it was argued that tense and φ-features

were shared by T and V as a result of Agree. This may also be the case with the force

feature in Old Irish. However, as this feature is shared by C and T there is an

alternative possibility. Chomsky (2005, 2006) proposes that T enters the derivation

with no features of its own and instead receives its features from the phase head C.

Under this view, it is plausible that certain features could be shared by both heads.48 It

45 Although see Koch 1987; Isaac 2001. 46 There is also diachronic evidence to support the idea that absolute inflection results from a force feature. In chapter 4 it is argued that absolute endings developed due to the presence of a declarative clause-typing particle that was reanalysed as a verbal affix. 47 Invoking a force feature here draws a parallel with Adger’s account. Crucially, however, we are not postulating an extra syntactic position, simply a force feature, that is assumed to be present universally. 48 Chomsky is not entirely clear as to whether ‘spreading’ means ‘sharing’ or simply ‘passing on’. The data seems to suggest, however, that sharing is a possibility.


67

seems likely that this will be a matter of cross-linguistic variation. In English for

example, tense and φ-features appear on T and not on C and so do not seem to be

shared. In Germanic complementizer agreement constructions, however, we find φ-

features on C, suggesting perhaps that they are shared by C and T (48). Similarly, in

Modern Irish we find tensed complementizers alongside tensed verb forms,

suggesting a shared tense feature (49).

48. (a) ob-st (du) noch Minga kumm-st

whether-2SG (you) to Munich come-2SG

‘Whether you come to Munich’ (Fuß 2005: 159)

(b) wem-ma (mia) aaf Minga fon

when-1PL (we) to Munich drive

‘When we drive to Munich’ (Fuß 2005: 165)

49. (a) Níor oscail Cáit an geata

NEG.PAST open.3SG.PAST Cáit the gate

‘Cáit did not open the gate’

(b) Deir sé gur oscail Cáit an geata

say.3SG.PRES he that.PAST open.3SG.PAST Cáit the gate

‘He says that Cáit opened the gate’

Let us assume, then, that the force feature in Old Irish, like tense in Modern Irish, is

shared by both C and T. However, unlike tense in Modern Irish, force in Old Irish can

only be realised in one position, like tense and φ-features in Modern English.

Furthermore, like tense and φ-features in English, the force feature has the property

[affix] and so can only be realised in a position where it satisfies the SAF.

It was argued above that in order to satisfy the SAF a feature that is specified as

[affix] must appear under the same terminal node as an additional positively valued

morphosyntactic feature. When a conjunct particle or the initial preverb of a

compound verb (see section 4.3.2 below) appears in the C position this will provide

positively valued morphosyntactic features with which the affixal force feature can


68

combine. Therefore the SAF will be satisfied and the force feature will be realised in

C and deleted in T, as shown in the tree below.49

50. CP

C TP ⇒ [Cco-ø [Tléici…]]

Conj/Pvb until-FORCE allows.PRES.3SG.CONJ [ForceAFF] ‘Until he allows’

T VP Verb [T, φ] [ForceAFF]

When there are no other morphosyntactic features present in C, i.e. no conjunct

particle and no preverb, force will have no host, and so as a result of the SAF cannot

be realised there. In this case, C will be completely empty, and so will receive no

realisation, and force will be realised in T, in conjunction with the φ-features, giving

rise to absolute verbal inflection. This is shown in the tree below:

51. CP

C TP ⇒ [Cø [Tléicid]]

[ForceAFF] ø allows.PRES.3SG.ABS T VP ‘He allows’

Verb [T, φ] [ForceAFF]

So far in this section it has been argued that the appearance of absolute morphology in

Old Irish is conditioned in a similar way to the well-known phenomenon of Affix-

Hopping in English. Both constructions can be explained in current theoretical terms

through the use of Agree (in the English case) or feature spreading (in Old Irish) and

the post-syntactic operation Chain Reduction. In the second half of this section we

turn to the second parallel between English and Old Irish, namely in the use of do in

English and no in Old Irish.

49 When force is declarative it has no phonological realisation when it is spelled-out in C. When force is relative it can be realised as a special relative form of the preverb (see section 3.4.1 above) or as a relative complementizer, such as the negative relative nad.


69

Both do in English and no in Old Irish seem to have the status of last resort

elements, used to provide a phonological realisation for a functional projection that

for some reason needs to be realised, but has no phonological realisation of its own.

Neither do in English nor no in Old Irish are associated with a single meaning or

function, and both can be used in a variety of environments. Let us consider each in

turn.

There are 4 main constructions in which the dummy auxiliary do appears in the T

position: namely negative clauses, emphatic clauses, VP ellipsis and VP raising,

shown in the examples in (52).

52. (a) John does not know

(b) John DOES know

(c) John knows and Mary does too

(d) Bob asked Mary to leave John, and leave him she did

Biberauer & Roberts (to appear) argue that in all these contexts T contains a

[+affective]-feature, a type of polarity feature. Polarity is often argued to be

associated with T cross-linguistically (Laka 1990, Zanuttini 1997).50 The example in

(52a) is a negative clause, so it could be argued that T contains a negative polarity

feature. In (52b) the auxiliary do is emphatic. In this case then, T could be argued to

contain an emphatic-positive polarity feature. Although the auxiliary in (52c) and

(52d) is not stressed as it is in (52b), there is evidence to suggest that in these

environments, the auxiliary is still emphatic. For example, in cases of VP ellipsis and

VP raising the auxiliary cannot be phonologically reduced, so sentences such as those

in (53) are ungrammatical.

53. (a) *John has left and Mary’s too

(b) *She said she had left and left she’d

Emphatic auxiliaries cannot be phonologically reduced. This could explain why we

do not find reduced auxiliary forms in VP ellipsis and VP raising constructions.

50 Although see also Déchaine & Wiltschko (2003) who propose a parametric difference between T- and C-related PolP.


70

Clearly a [+affective]-feature on T will not always result in the insertion of do.

When there are modal or aspectual features in T then T will be realised as a modal or

aspectual auxiliary rather than do. In DM terms, this can be explained by the

elsewhere principle. The auxiliary that matches the largest number of features in T

will be inserted there. So, when there are modal or aspectual features on T, the more

highly specified modal and aspectual auxiliaries (e.g. will, can, must, is, have) match

the feature specification of T more closely and are chosen in place of do. When,

however, there are no aspectual or modal features in T, then do will be inserted.

Biberauer & Roberts’ account of do-insertion works well with the account of

Affix-Hopping provided above. When T contains a [+affective]-feature there will be a

positively valued morphosyntactic feature with which the affixal tense and φ-features

can combine. Therefore, in all the environments where we find do-insertion the tense

and φ-features can be realised in T and deleted in V.51 Let us consider now whether

we can account for no-insertion in Old Irish in a similar way.

The environments in which we find no in Old Irish are considerably more diverse

than those in which we find do in English. The dummy preverb no is used with object

pronouns, in relative clauses and when the verb has a secondary tense. We saw in

sections 2.3 and 3.3.1 above that suffixed pronouns are no longer productive in the

Old Irish period and are used only in restricted circumstances. When there is no

available suffixed form, the dummy preverb no is inserted and the pronoun is infixed

between them as in (54). In this case no seems to be associated with the presence of

an enclitic object pronoun.

54. (a) no-s nguid-som

PVB-INF.3PL beseech.PRES.3SG.CONJ-emph.part.3SG.M

‘He beseeches them’ (Wb 25b9)

(b) ni hed no-t beir í nem

NEG it PVB-INF.2SG carry.PRES.3SG.CONJ in heaven

‘It is not this that brings you into heaven’ (Wb 6c9)

51 Biberauer & Roberts (to appear) propose a different view of Affix-Hopping.


71

Turning to the relative clauses, we find that when there is no special relative form

available, a relative clause is marked through the use of the dummy preverb no

followed by either lenition, (55a), or nasalization, (55b).

55. (a) is hed in so no chairigur (non-rel cairigur)



(b) cid no mbetha (non-rel betha)

why PVB be.PAST.SUBJ.2SG.CONJ

‘Why (is it that) you should be?’ (Wb 4c24)

Due to cross-linguistically observed similarities with interrogative marking, relative

marking is associated with the C position. As a result, we can assume that whenever

the clause is relative, the C position will be specified with a feature [+wh]. So, in this

case, no seems to be associated with a wh-feature.

The third case is that of secondary tenses. When a simple verb appears in the

imperfect (56a), past subjunctive (56b) or the conditional (56c) and there is no

conjunct particle, the dummy preverb no is inserted.

56. (a) no scarinn friu

PVB part.IMPF.1SG.CONJ to.3PL

‘I should part with them’ (Wb 24a4)

(b) cia nu tiastais huaim

although PVB go.PAST.SUBJ.3PL.CONJ from.1SG

‘Although they should go from me’ (Ml 117d3)

(c) no comallaibthe

PVB fulfil.COND.PASS.SG.CONJ

‘It would be fulfilled’ (Ml 105b14)

In these cases, no seems to be an aspectual particle.

It seems, then, that unlike English do there is no one feature that can account for all

occurrences of no, not even a very general feature like Biberauer & Roberts’

[+affective]. It seems, then, that Biberauer & Roberts account of do-insertion cannot

be carried over to Old Irish. There are two possible alternatives. First, it could be


72

argued that for each different environment in which no appears no is a different

vocabulary item (VI), so in Old Irish there are three homophonous VIs no, one that

realises C [+wh], one that realises C[+objectCL] and one that realises C[+aspect]. At

first sight, this seems possible, but unappealing. If we consider the analysis in more

detail, however, there seems to be a more serious problem. If the VI for no contains

feature specifications of this kind, we might expect it to behave more like a conjunct

particle; so whenever C contains an aspect or wh-feature we might expect no to be

inserted. However, this does not seem to be the case. The particle no appears only

when there is no other element that could appear in the C position. This includes not

only conjunct particles, but also, as we shall see in the next section, the initial

preverbs of compound verbs.

An alternative is to characterise the particle no as an elsewhere morpheme. In this

case, the appearance of no is not conditioned by any one feature in particular. The VI

for no has no feature specifications, apart from the fact it can appear in C. In DM

terms, no will always compete for insertion into the C position, but will only be

inserted when there are no more highly specified VIs (i.e. conjunct particles or initial

preverbs) that match the feature content of C more closely. When C contains a

negative feature, or a conjunction feature or a preverb feature (see (57) below), the

corresponding VI will be inserted. However, if none of these features are present, the

elsewhere morpheme no will be inserted.

57. [C [+negative]] → ní

[C [+conjunction]] → con, dian, aran

[C [+preverb]] → do, fo, as, ro

[C] → no

The characterization of no as an elsewhere morpheme accounts well for the intuition

that it appears simply whenever it is needed; however, the VI for no listed above is

perhaps slightly misleading. C is not, as (57) suggests, realised as no in all clauses

where there is no conjunct particle and no initial preverb. We saw in the first half of

this section that when C contains no features other than a declarative Force feature C

receives no realisation at all. Let us consider now how these two scenarios can be

reconciled.


73

The crucial point is that C is only realised as no when C contains an extra feature

in addition to its default features. When C contains only its default features it receives

no phonological realisation. One way to account for this could be to argue that there is

a specific VI, such as that given below, that is in effect more specific than the

elsewhere case, specifying that when C has its default features it is realised as null.

58. [C [Force0 Finiteness0 v0]] → ø

However, this is problematic. Under the principles of Vocabulary Insertion, (Halle &

Marantz 1993) the VI that matches the feature content of the terminal node most

closely is inserted. When the C node is specified with an object clitic feature, it will

have the feature structure given below (assuming it is non-relative and the verb is

simple):

59. [C [Force0 Finiteness0 v0 ObjectCl]]

This feature structure is more closely matched to the VI in (58) than that for no and so

we would expect ø to be inserted rather than no.

If we return to the account of Affix-Hopping and absolute inflection presented

above a new solution emerges. It was argued above that when C contains only an

affixal Force feature this feature cannot be realised due to the SAF, as the affix does

not have a host. If we further assume, however, that the SAF prevents not simply the

Force feature, but the entire C head from receiving a phonological realisation, then we

can perhaps explain the distribution of the null C and the elsewhere particle no. When

C contains only an affixal Force feature, C is marked for deletion as it violates the

SAF. Therefore, C cannot receive a phonological realisation and so is simply not

considered during the spell-out of the syntactic terminals at Vocabulary Insertion. In

this case, then, C is necessarily null as it does not even have the option of being

spelled-out. When C contains an extra feature, the SAF is satisfied as the affixal Force

feature has a host. Therefore, C is marked for realisation. It is considered by the

operation Vocabulary Insertion and so is realised as the elsewhere particle no. Viewed

in this way the alternation between ø and no reflects whether the C head reaches the

point of Spell-Out, and whether or not it is marked for deletion during the process of

Chain Reduction. It seems then that the postulation of an affixal Force feature can


74

account not only for the distribution of absolute and conjunct endings but also for the

distribution of the particle no.

4.3.2 Compound verbs

In section 3.4.1 above it was argued that the initial preverbs of deuterotonic

compound verbs behave phonologically and syntactically like conjunct particles,

therefore it seems likely that, like a conjunct particle, this initial preverb is in the C

position. Similarly in sections 3.4.2–4 it was argued that the preverb cannot move to

C independently of the verb in the syntax, due to certain well motivated constraints on

syntactic movement operations. As deriving the necessary orders within narrow

syntax is seemingly impossible we need an alternative explanation, preferably one

that does not rely on conceptually undesirable post-syntactic movement.

The basic problem is this: the initial preverb of a compound verb behaves as if it is

in C, but how does it get there? The semantic relation between the preverb and the

verb suggests that they are merged together in the vP, therefore the initial preverb

cannot be merged in C. Similarly, it was shown in section 3.4.2 that movement of the

preverb to C, independently of verb-movement to T is impossible due to syntactic

movement constraints (either the HMC or the PIC). So, if the preverb is not in C as a

result of Merge or Move, then the only remaining possibility is Agree. It is possible to

maintain the insight that the preverb is merged in light v and the idea that it is in C, if

C has a verbal feature, that is valued by Agree with light v and then under certain

circumstances spelled-out in the C position rather than lower in the clause. As was the

case with Affix-Hopping in section 4.2, apparently illegal movement can be explained

through Agree and Spell-Out.

Before we examine this proposal in more detail, let us consider how it might work

in general terms. Conventional movement approaches run into problems in accounting

for compound verbs in Old Irish because of the behaviour of the initial preverb. When

C is filled by a conjunct particle, movement of the verb to T seems to proceed in the

expected fashion, moving through and incorporating the preverb in light v. However,

when C is not filled, the verb moves to T without incorporating light v and light v

moves independently to C. Movement approaches must explain why when C is filled

incorporation takes place, and when C is unfilled incorporation does not take place. If

the preverb never moves to C in the syntax, but is only realised there as a result of


75

Agree, then we no longer face this problem. In the syntax, the only movement that

takes place is the movement of the verb to T via light v.

Independent of the syntactic movement operation from V to T there is an Agree

relation between C and light v. C has an unvalued verbal feature that must be valued.

Although this verbal feature is always present on C and is always valued, it is only

realised phonologically when there are no other more specific features in C. When

other features, such as [neg] or [Q], or lexical items such as the conjunctions when or

so are present in C, then these are spelled-out in preference to the verbal feature. This

proposal can be seen schematically in the tree in (60) below:

60. C C T

[u v] T v

V T tv V

v V tV D From the tree above we might expect that if the preverb is spelled-out in C it will be

spelled-out twice, in both C and T, resulting in forms like *do-tabair rather than the

attested do-beir.52 However, this is clearly not the case. When some element other

than the preverb appears in C, the preverb is spelled-out with the verb in T. However,

when the preverb is spelled-out in C, the lower copy in T is deleted, resulting in the

deuterotonic verbal form. Let us look at each stage of the proposal in detail. First we

will consider the Agree relation between C and v. Secondly we will examine what

exactly a preverb is under this analysis and how it comes to be realised in the different

positions. Thirdly we will consider the deletion operation that affects the initial

preverb and how the two forms of the compound verb are derived.

4.3.2.1 Agree between C and v

The syntactic operations involved in this proposal are relatively uncontroversial. First,

the idea that the verb moves to v, incorporates with the preverb and then moves onto

T is straightforward and requires no complications to the syntax. This is head-

52 Forms such as do-tabair are found in early Middle Irish. See Strachan (1895–7).


76

movement according to the HMC or the PIC1, motivated by an Affix-feature in T.

Similarly, the Agree relation between C and v is an operation that receives cross-

linguistic support. Since den Besten (1983) it has been assumed that in the Germanic

V2 languages the verb moves to the C position. Under the theoretical assumptions of

this dissertation, this movement will be motivated by an Affix-feature on C. However,

for the verb to move to this position, there must be an Agree relation between the verb

and the C position. As a result of the PIC, C cannot be in an Agree relation directly

with the verb. To move to C, the verb must move via light v. Therefore, there must be

an Agree relation between C and light v.53 It seems a reasonable step to postulate such

an Agree relation for Old Irish, especially as it seems that at an earlier stage Old Irish

demonstrated V-to-C movement (see section 3.3 above and chapter 4 below).

Although it seems reasonable to suggest that there is an Agree relation between C

and light v, an Agree relation is dependent on the presence of features that need to be

valued. The next stage, then, is to establish what these features might be. Den Besten

(1983) proposes that verb-movement to C in V2 languages occurs because tense is

located in the C position.54 This view has received recent theoretical support from

work by Chomsky (2005, 2006). Chomsky argues that C is the true locus of tense and

agreement features. Biberauer & Roberts (2005) argue that although in the Romance

languages these features are transferred to T (resulting in V-to-T movement), in

Germanic these features remain on C and motivate V-to-C movement. This

suggestion could also apply to Old Irish. However, it does not explain why C is active

and acts as a probe. The reason for this, I suggest, is that C also has an unvalued

verbal feature. As light v is the closest head containing an appropriate valued verbal

feature, an Agree relation between C and v is established to value this feature.

4.3.2.2 A closer look at preverbs and the status of light v

Under the analysis presented here preverbs are the phonological realisation of a verbal

feature. This verbal feature can be realised either in C separate from the remainder of

the verb, resulting in the deuterotonic form, or with the remainder of the verb in T (or

lower in the clause in Bergin’s Construction – see section 4.4 below) resulting in 53 If the verb can be shown to move via the T position perhaps C might Agree with T rather than v. However, there is reason to believe that the verb does not move via T in at least some of the Germanic V2 languages. In this case, the Agree relation must be between C and light v. 54 Another prominent suggestion in the V2 literature, first proposed by Holmberg & Platzack (1995) is that the verb moves to C to check finiteness features. See Vikner (1995) for an overview of C/T approaches to V2.


77

prototonic form. Crucially for this analysis it is assumed that this feature is also

associated with light v. By making this assumption it is possible to account

straightforwardly for the different positions in which the verbal feature occurs. First,

as we saw in the previous section, there are both theoretical and empirical arguments

to support the existence of an Agree relation between C and v. Therefore, if the verbal

feature is associated with v we can account for its appearance in C. Secondly, verb

movement is generally assumed (according to the HMC, PIC1 or Chomsky’s (2004)

proposal the verbs need to be verbalised by movement to light v) to proceed via light

v resulting in incorporation,. Therefore if, as suggested here, Old Irish has V-to-T

movement, then we would expect the verb to move through and incorporate light v.

This accounts for the appearance of the verbal feature in T.

Although there are theoretical reasons for assuming that the verbal feature that is

realised as a preverb is associated with light v, if we consider the exact nature of this

feature in further detail it seems likely that it is in fact unvalued on light v and is

valued through an Agree relation with a corresponding valued feature on the verb.

In recent minimalist work, light v seems to have the status of a verbaliser.

Following Marantz (1997), Chomsky (2004) argues that verbs (and nouns and

adjectives) are stored in the lexicon as acategorial roots. It is only through association

with a light v head that these acategorial roots become verbalised. Chomsky argues

that in order for this to take place the verb root must move to light v. It is not clear

that this is necessary as the same effect could be achieved through an Agree relation.

So, it seems that light v must be in an Agree relation with V, the acategorial root, in

order to value the root’s categorial feature; however, as discussed in the previous

section, Agree relations are reciprocal; for an Agree relation to hold between v and V,

v must also have some unvalued feature that the verbal root can value. One possibility

is that in the same way as the verbal root needs functional, i.e. categorial, information

the v head needs to be specified for lexical content. Let us explore what this lexical

content could be.

It seems unlikely that the verbal root will pass on its entire semantic content to

light v. A more plausible suggestion, perhaps, is that the root simply passes on

information regarding the class to which it belongs. In many languages verbs and

nouns are divided into conjugation and declension classes. These classes are

essentially arbitrary, they have no semantic or grammatical justification; it is simply

the case that during language acquisition when a child learns a verb or noun s/he must


78

learn to which class it belongs, so that s/he can conjugate or decline it appropriately.

In French, verbs can be divided into three conjugation classes depending on the final

two letters of their infinitive form. For example, verbs with an infinitive ending -er

such as regarder ‘watch’ belong to the first conjugation, those ending in –ir, such as

finir ‘finish’ are second conjugation and those ending in –re such as vendre ‘sell’ are

third conjugation. Each verb class has a slightly different conjugation pattern, as can

be seen in the table below.

Table 4: French verb conjugation classes in the present tense

First conjugation Second conjugation Third conjugation

1sg je regarde je finis je vends

2sg tu regardes tu finis tu vends

3sg il/elle regarde il/elle finit il/elle vend

1pl nous regardons nous finissons nous vendons

2pl vous regardez vous finissez vous vendez

3pl ils regardent ils finissent ils vendent

Embick & Halle (2005) argue that each root contains a diacritic marking its

conjugation or declension class. After syntax is complete, an extra, dissociated

morpheme (see Embick 1997) is added to the v head, and the information regarding

class is transferred from the root via a concord process. The same result, it seems to

me, could be obtained, without these extra post-syntactic operations, if the verb-class

feature appears on the v head itself, as a result of an Agree relation between v and the

root (V). This is what is proposed here.55 56

There are two pieces of evidence to support the proposal that preverbs in Old Irish

are arbitrary verb-class markers. First, as we saw in section 2.2 above, in the Old Irish

period most preverbs have no meaning independent of the verb. Therefore, it seems

likely that they are functional rather than lexical elements. Secondly, there are only a 55 Old Irish also has conventional conjugation classes. See McCone (1997a) for details. 56 One possible criticism here is that within the DM literature information regarding conjugation classes is often argued to be inserted with the phonological features during vocabulary insertion as conjugation class features do not play any role in narrow syntax. If this is the case, then the verb class features that are being proposed for Old Irish cannot be exactly the same as conjugation class markers as they do seem to play a role in the syntax. This is perhaps what we would expect, because, as noted in footnote 55, Old Irish has conjugation class markers in addition to preverbs. However, the parallel still holds to the extent that the verb class features in Old Irish are arbitrary and have no semantic or grammatical justification.


79

handful of initial preverbs found in the Old Irish period and they are not used

productively. The fact that preverbs are a closed class again suggests they are

functional rather than lexical.

So far, then, it has been argued that preverbs are the realisation of a verb class

feature on either C or T. Each compound verb has a verb class feature corresponding

to the preverb with which it is associated. The value of this feature is determined by

the lexical verb, but is transmitted to light v via an Agree relation between v and V.

Having outlined our assumptions regarding preverbs let us consider how this analysis

is preferable to the other possibilities.

The only existing concrete proposal regarding the status of preverbs is that made

by CHP. CHP argue that preverbs are merged into the structure as light v heads, part

of a Hale & Keyser (1993) complex VP. CHP’s analysis resembles that presented

here in that it assumes that preverbs are associated with light v. However, there are

two major differences. First, it is assumed here that preverbs are not merged as light v

heads; they are simply the realisation of one feature that appears in light v. Secondly

we assume that the feature that appears on light v is not intrinsically associated with

light v, but instead is valued by a valued verb class feature present on the verb itself.

The idea that preverbs are the realisation of a single feature that appears in light v

is preferable to the proposal that preverbs are merged as light v heads in several

respects. First, if preverbs are merged as light v heads we might expect them to

function as light v heads do cross-linguistically, for example in terms of determining

the argument structure of a verb or expressing aspect. This does not seem to be the

case. Preverbs have no grammatical function. If preverbs are simply the realisation of

an idiosyncratic verbal feature that appears on light v then there is no reason why they

should have all the properties of a light v head.

The second argument for the analysis of preverbs as a single feature rather than a

light v head is semantic. As noted in section 2.2 above, preverbs have no meaning

independent of the compound verb of which they are a part. If preverbs are merged as

light v heads, i.e. as independent bundles of features, we might expect them to have

some independent semantic content.57 If preverbs are the realisation of a single verbal

feature, then there is no reason for them to have any semantic content. Furthermore,

if, as proposed above, the value of the verb class feature is determined by the lexical

57 Although this is not necessarily the case as compound verbs could be analysed as idioms.


80

verb itself, this provides a neat account of the fact that preverbs have no meaning

independent of the remainder of the compound verb.58

Having provided an outline and justification for the analysis of preverbs as the

realisation of verb class features we must now establish how exactly these features

come to be realised in the positions in which they do. The remainder of this section

considers how the verbal features come to receive a phonological realisation at the

point of Vocabulary Insertion. The next section examines deletion of the lower

copies.

As discussed in section 4.1 morphosyntactic features become phonologically

realised at the point of Vocabulary Insertion when phonological exponents are

inserted in place of morphosyntactic feature bundles. For each morphosyntactic

feature bundle all possible vocabulary items (VIs) that could be inserted into that

particular position compete for insertion. To be inserted in a particular context, the

morphosyntactic features associated with a VI must be non-distinct from those of the

insertion context, i.e. the VI cannot have any features that are not present in the

terminal node. However, the VI need not contain all the features of the terminal node.

The VI that wins the competition and is inserted into the derivation will be the one

with the most matching features. Let us consider how this would work in Old Irish.

In DM terms, each conjunct particle in Old Irish is an independent VI containing

morphosyntactic and phonological features. Some consist simply of C-based features

and others are independent lexical items, such as when and until.59 This is shown in

(61) below.

61. [neg] ⇔ /ní/

[Q] ⇔ /in/

[neg] [rel] ⇔ /nad/

when ⇔ /dian/

until ⇔ /con/ 58 One further option is to analyse preverbs not as v heads but as Ps. Preverbs clearly resemble and are etymologically related to prepositions, and so there is reason to support such an analysis. However, it is not entirely clear how a P head would fit into the syntactic structure and interact with C and T. It could perhaps be argued that there is a P projection as part of an articulated vP structure. However, this possibility will be left for later research. 59 The idea that lexical items such as conjunctions compete for insertion into the C position is problematic in DM. However, it is clear that such competitions take place. It is perhaps possible that these conjunctions can be broken down into morphosyntactic feature bundles. However, this will not be attempted here.


81

At the point of Vocabulary Insertion all potential candidates for insertion into the C

position will compete, and the most specific VI, i.e. the item that matches the most

features will be inserted. So, when the C position contains the features [neg] and [rel],

either /nad/ or /ní/ could be inserted as both are non-distinct from the terminal node,

neither VI contains features that are not present in the C position. However, in this

case /nad/ will be inserted in preference to /ní/ as it is more specific; it matches two

features in C rather than one. When the C position contains only the feature [neg], it

will be impossible to insert the VI /nad/ as this contains the feature [rel], which is not

present in C, and so /ní/ is the only option.

Let us turn now to preverbs. It was argued above that preverbs are the realisation

of a verb-class feature in C. However, preverbs are only realised in C when there is no

conjunct particle. In DM terms this means that the verb-class feature will only be

realised in C when no other more specific VI can be inserted. On the assumption that

the v feature in C represents a verb-class marker, we could imagine VIs such as the

following (as verb classes are arbitrary they are marked simply by numbers):

62. [v1] ⇔ /do/

[v2] ⇔ /fo/

[v3] ⇔ /as/

etc.

The only morphosyntactic features specified for these VIs then are the valued v

features. When the C position contains no other features of the type [neg] or [Q] or a

lexical item such as when, the VI encoding the valued v feature will match the C

position most closely and so be inserted there. When the C position contains both the

feature [neg] and a valued [v1] feature, however, it is the [neg] feature that wins out.

There are two possible ways this could be explained. First, it could be that all VIs that

can potentially be inserted into the C position contain all the features that are

associated with C, such as Force, Finiteness and, crucially, v, as shown in (63) below.


82

63. [Force] [Fin] [v] [neg] ⇔ /ní/

[Force] [Fin] [v] [Q] ⇔ /in/

[Force] [Fin] [v] [neg] [rel] ⇔ /nad/

[Force] [Fin] [v] when ⇔ /dian/

[Force] [Fin] [v] until ⇔ /con/

In this case, any conjunct particle will be inserted in preference to a preverb as the VI

contains an additional feature matching the C node, e.g. [neg], rather than simply the

default C features [Force] [Fin] [v].60

4.3.2.3 Preverb deletion

In the previous two sections it has been argued that in Old Irish in all clauses C has an

unvalued verbal feature that must be valued through an Agree relation with a

matching feature in v. This verbal feature encodes verb class, and it can be realised in

C, as an initial preverb, however, it will only be realised when there are no other C-

features in the C position. The next issue to be discussed is how, when the v feature is

realised in C, the v feature in T comes to be deleted.

It was argued in section 4.1 that the PF-operation Chain Reduction marks the

highest or leftmost copy of a certain feature bundle for realisation and deletes

subsequent copies. Although in most cases identical feature bundles arise by

movement, this does not have to be the case. If chains are determined in terms of

identity and/or c-command as discussed in section 4.3.1, then it is possible that two

identical feature bundles resulting from the operation Agree could also undergo Chain

Reduction. It is possible that this could be the case in Old Irish. When C is empty, i.e.

it contains no C-based features that will result in the insertion of a conjunct particle,

the only feature it contains will be a valued verbal feature. As this verbal feature was

valued by light v it will be identical to the light v heads lower in the structure, not

simply the original light v head, but also the light v head that incorporates with the

verb and moves to T. As a result, under Chain Reduction the higher copy will be

marked for realisation and the lower copies in T and light v will be deleted.

60 It is not clear that VIs contain all the information given in (55). DM assumes that VIs are underspecified. However, if this is the case, then we simply need to stipulate that inherent C features are spelled-out in preference to a valued v-feature (cf. Halle & Marantz 1993), or perhaps that C features are higher in a universal feature hierarchy (cf. Noyer 1997).


83

In the situation described above, when the C position contains nothing but a verbal

feature, this deletion operation is quite straightforward; however, this is not the only

circumstance in which the verbal feature is realised in C. C is realised as an initial

preverb when there is an object pronoun in the clause, when the verb is in a so-called

secondary tense (imperfect, past subjunctive and conditional) and when the clause is

relative. As we saw in section 4.3.1, these are all the environments that we find

insertion of the dummy particle no when the verb is simple. It was argued above, that

the insertion of the dummy particle no provided evidence that each of these

constructions involved the presence of a non-default feature in C, which provided a

host for affixal Force and prevented C from being deleted at PF. However, if these

constructions all involve extra features in the C position, then this poses a problem for

our account of compound verbs, as it means that the features in C and the features in

light v will not be identical, and so we would not expect the lower copy to be deleted.

However, if we consider each of these cases in more detail, it seems that this is not the

case, and the features in C and v can be shown to be identical.

Let us first examine the secondary tenses. It was argued above that the feature that

prevents C from being deleted in this case is an aspectual feature. Aspect marking is

typically a feature associated with light v. Therefore, for aspectual features to be

present on C, C must be in an Agree relation with light v. As a result, if C contains a

valued aspect feature, light v must contain one too. In the case of secondary tenses,

then, C will not contain a feature that is not present in v. The two positions will both

contain an aspect feature, and so will be identical. As a result, we would expect

deletion of the lower vs to take place.

A similar proposal can be made for clauses where there is an object pronoun. The

light v head is typically associated with object agreement and accusative case

assignment. Therefore, if C has object clitic agreement features, it seems likely that

these features will be valued through an Agree relation with light v. In this case, then,

as before, the extra feature present on C will also be present on v, so again, it seems

that the two heads will be identical.61

The case of relative clause marking is more problematic. Relative clause marking

is typically the result of a [wh] feature on C. These features are not typically 61 This is problematic, as we would expect object-features always to be present on C whenever there is an object. In Old Irish it seems that C is only valued for object features when the object is a clitic. This could be related to the fact that object clitics seem to raise into the C-projection in Old Irish. A better understanding of clitics is necessary before any firm conclusion can be reached.


84

associated with v. However, this does not mean that v may not be specified as [wh].

In object wh-questions in English a wh-phrase must move to Spec-CP. If this wh-

phrase is the object of the verb, it cannot move straight to Spec-CP as this would

violate the PIC. As the object is merged as the complement of V within the VP it is

sent to PF at the end of the phase and so is unavailable for movement to C. If the

object wh-phrase is to move to Spec-CP it must move first to Spec-vP to escape the

phase. In English non-wh-clauses the object does not move to Spec-vP. Light v in

English does not have an EPP-feature that attracts the object. Therefore, we need an

explanation as to why movement of the object to Spec-vP should take place in wh-

questions. One possibility would be to suggest that when C is marked as [wh] and has

an associated EPP-feature light v will be marked in the same way. This [wh]-feature

on light v cannot be a result of Agree, as movement of the object to Spec-vP must

take place before C is merged. Therefore, light v must receive this [wh] feature (and

an associated EPP-feature) in the numeration. If a similar situation holds in Old Irish,

then both C and light v will contain [wh] features and so in relative clauses, the

features in each head will be identical and again deletion of the lower v head will be

possible.62

In this section it has been argued that when the verbal feature is phonologically

realised in C it is deleted in the lower positions, within the TP and vP. This deletion

occurs before the point of Vocabulary Insertion whenever the features in C are

identical to those in light v. It seems that C and v will be identical not only when C

contains only a verbal feature, but also when C contains aspectual, object (clitic)

agreement or relative features. Once this deletion of the lower v head has taken place,

light v and the verbal root will appear in separate syntactic positions and so will be

spelled-out separately, resulting in the deuterotonic verbal form. If C contains other

C-based features, it will not be identical with the lower vs and so the v in T is not

deleted. In this case light v and the verbal root are spelled-out together in T, resulting

in the prototonic form.

4.4 A closer look at the Old Irish data

In this section a new post-syntactic account of the double system of verbal inflection

has been developed. So far we have provided an account of the main facts, i.e. the 62 This could be extended to force in general, thus explaining why the presence of force in C does not prevent deletion of the lower v head.


85

distribution of the different verbal forms and object clitic pronouns. In this section we

will consider some further aspects of the Old Irish verbal system and how the analysis

presented above can account for them.

4.4.1 Tmesis and Bergin’s Construction

As noted above, tmesis and Bergin’s Construction are archaic constructions where the

verb, instead of being in initial position, appears lower in the clause in verb-final or

verb-medial position. In Bergin’s Construction, the verb can be simple or compound.

In either case, the entirety of the verbal complex appears in non-initial position and

has dependent form, i.e. conjunct inflection if the verb is simple (64a) and prototonic

form if the verb is compound (64b).



‘Although I reap blistered seaweed’ (Corm 1059/Bergin 1938: 197)



‘He spoke to an angel’ (LU 1148/Bergin 1938: 201)

Tmesis involves only compound verbs. In this case, the initial preverb is spelled out

where we would expect, in initial position, and the remainder of the verb is lower in

the clause.

65. Ath (mór)cathu fri crícha comnámat -cuirethar

PVB (great)battalions to borders neighbour.GEN.PL send.PRES.3PL.DT

‘He dispatches (great) battalions to the borders of hostile neighbours’

(AM §15)

The advantage of the post-syntactic analysis of Old Irish presented here is that it can

provide a unitary account of both constructions. The basic feature of both tmesis and

Bergin’s Law is that the verb appears in non-initial position for stylistic effect. This

can be accounted for if we assume that in both cases the verb does not raise to T,

because T’s Affix feature has been omitted for stylistic reasons. The only difference

between the two constructions, then, is the position in which the initial preverb is


86

realised. In tmesis, the preverb is realised in C and deleted lower in the clause, in the

same way as was suggested above for all deuterotonic verb forms. In Bergin’s

Construction, where the verb is compound, the initial preverb is not realised in initial

position and is spelled out with the remainder of the verb. This is shown in the

simplified trees below:

66. (a) Bergin’s Construction (b) Tmesis:

CP CP …tabair do…beir

C TP C TP do do

T vP T vP Ø

v VP v VP do do beir beir

4.4.2 Imperatives

There is a construction in Old Irish where a verb in absolute initial position does not

have absolute or deuterotonic form. When the verb is imperative (and there is no

infixed pronoun) it has conjunct or prototonic form.

67. (a) ber brith étrunn

bear.IMPV.2SG judgement between.1PL

‘Judge between us’ (Ml 38c28)

(b) tabair digail

give.IMPV.2SG punishment

‘Inflict punishment’ (Ml 27c12)

This construction can be accounted for within the post-syntactic analysis presented

above if we postulate an imperative feature that appears in the C position. The

imperative feature combines with and provides a host for the affixal Force feature in

C, ensuring that it is realised on C and deleted in T, resulting in conjunct rather than

absolute inflection. When the verb is compound the imperative feature prevents the

realisation of the preverb in C. For this to be the case, there must be a specific

vocabulary item inserted in C when C contains the feature [imperative]. This


87

vocabulary item could be phonologically null, or it could instead have some kind of

suprasegmental phonological effect. Obviously, from the available evidence from the

texts it is difficult to determine which is the case.

4.4.3 Relative forms

We saw above that there are two main ways of marking a verb as relative in Old Irish,

either through the use of special relative verbal endings (68) or through the lenition or

nasalization of the initial segment of the verb (69), or in the case of deuterotonic

forms, the verbal stem (70).

68. is oinfer gaibes búaid diib inna chomalnad

COP one man take.PRES.3SG.REL victory of.3PL in its completing

‘It is one man of them that gets victory for completing it’

(Wb 11a4)

69. is hed in so no chairigur (non-rel cairigur)



70. a n- ad-chiam (ch=/x/, cf. non-relative ad-ciam)

that.NAS PVB-see.PRES.1PL.DT


Although we have touched on the analysis of relative clauses in the preceding

sections, let us clarify here exactly how the post-syntactic analysis proposed above

accounts for these different constructions. The basic assumption is that relative

marking in Old Irish is conditioned by the presence of a wh-feature on C. Old Irish

has no relative particles corresponding to Modern Irish aL and aN, so relative C has no

realisation of its own. However, this does not mean that the wh-feature has no

phonological effects. Whenever C contains a wh-feature it will cause the initial

segment of the verb in T to lenite or nasalize.63 In Old Irish this seems to happen

across the board, not only in the examples above where C is filled, but also as we saw

in section 3.3.2, when the verb is simple and has the special relative form.

63 The exact mechanism behind the grammatical mutations in Old Irish will be left for further research.


88

71. cisí aimser derb thechtas (non-relative techtaid)

what time definite possess.PRES.3SG.REL

‘What is the definite time that he has?’ (Sg 26a6)

Grammatical mutations are not the only phonological effect of a wh-feature on C. In

certain cases, the combination of the wh-feature with other features in C results in

special relative forms. For example, the preverbs im and ar have special disyllabic

relative forms imma/imme and ara/are. The negative particle also has a special

relative form, nad.

The special relative morphological found on simple verbs can be accounted for in a

similar way. Let us assume that the wh-feature, like the Force feature that conditions

absolute inflection, is shared by both C and T. For certain persons and tenses, the

presence of a wh-feature on T will result in the spell-out of special relative verbal

endings. This will only be the case, however, when a relevant vocabulary item exists.

In the present 3sg relative, for example, there are two possible vocabulary items that

could be inserted:

72. [present] [3sg] → /ø/

[present] [3sg] [wh] → /es/

When there is a [wh] feature on C the relative form will be inserted as it matches

more features in the terminal node. In the case of the 1sg, however, there is only one

vocabulary item available as there is no specific relative vocabulary item.64

73. [present] [1sg] → /u/

Therefore, only the non-relative form can appear.

One crucial difference between the wh-feature that conditions relative morphology

and the Force feature that conditions absolute inflection is that the wh-feature is not

affixal. The fact that simple relative verb forms are lenited in the same way as any

64 Although there is only one vocabulary item, this vocabulary item has a number of possible realisations. We shall not go into this issue here.


89

other verb forms suggests that the wh-feature is present and, in this case realised, in

both C and T.

4.5 Summary

At the end of section 3 it was concluded that syntax alone cannot account for the

different verbal forms found in Old Irish. As a result, this section has explored the

possibility of accounting for the Old Irish verbal system through a combination of

syntactic and post-syntactic operations. Section 4.1 argued that there are two kinds of

post-syntactic operation that take place at the interface between syntax and

phonology, conceptually motivated, universal, obligatory operations such as

Linearize, Chain Reduction and Vocabulary Insertion, and empirically motivated,

language specific operations such as impoverishment, fusion and fission. Section 4.2

reviewed an attempt by Adger (to appear) to account for the Old Irish verbal system

using the second type of post-syntactic operation. It was argued that this is, in fact, the

biggest problem with Adger’s analysis. Adger relies on conceptually undesirable post-

syntactic movement operations. In section 4.3 a new post-syntactic account was

developed, relying primarily on the syntactic operation Agree and the conceptually

motivated post-syntactic operations of Chain Reduction and Vocabulary Insertion.

Section 4.4. considered how this new theory of the double system can account for

various parts of the data.

5 CONCLUSION

This chapter has reviewed the existing generative literature on the Old Irish verbal

system and attempted to develop a novel, minimalist account of the data. Having

presented a description of the main features of the Old Irish verbal system in section

2, section 3 went on to examine the existing syntactic accounts for the distribution of

the different verbal forms in Old Irish. It was argued that neither Doherty (1999,

2000) nor Carnie, Harley & Pyatt (2000) can account for all aspects of the Old Irish

data. Although at first sight, CHP’s analysis seems to be well supported by data from

object pronouns and relative clauses, a closer look at these constructions in the Old

Irish period shows that this is not the case. Moreover, CHP’s analysis faces significant

theoretical problems, particularly in the area of compound verbs. Section 3.4

attempted to show that independent movement of the preverb and the verb in the

syntax is impossible. Deriving the attested orders through head-movement is


90

prohibited by constraints on syntactic movement. XP- and remnant-movement

accounts, although less constrained, face problems in terms of motivation.

Having concluded in section 3 that the different verbal forms in Old Irish cannot be

a result of syntax alone, an alternative possibility was explored, namely the

interaction of syntactic and post-syntactic rules. Adger’s (to appear) post-syntactic

account was reviewed, and it was concluded that the main problem with it was its

reliance on post-syntactic movement operations. In section 4.3 a new analysis was

developed, based primarily on the conceptually motivated post-syntactic operations of

Chain Reduction and Vocabulary Insertion. Simple verbs move only as far as T in the

syntax, and the appearance of absolute endings on an initial verb is a result of a

process parallel to English Affix-Hopping. Compound verbs also move only as far as

T in the syntax, with the verb moving via light v in the expected fashion and

incorporating with the preverb. The appearance of the initial preverb of a deuterotonic

compound verb in C is a result of an Agree relation between C and v, valuing a verbal

feature in the C position. When C contains no other non-default features, this verbal

feature is realised in C as a preverb, and the lower occurrence of v in T is deleted, via

Chain Reduction under identity.

The synchronic account presented here highlights a number of interesting

diachronic questions. First, although the double system of verbal inflection is not

derived syntactically in the Old Irish period, the evidence from relative clauses and

suffixed pronouns presented in section 3.3 suggests that this was not always the case,

and that at some point in its pre-history Irish did have V-to-C movement with

absolute inflection as a result of verb-movement to C. Generalised V-to-C movement

is not a feature of PIE and so must be a Celtic innovation. How did V-to-C movement

arise in Old Irish? This is the topic of chapter 4.

If pre-Old Irish had V-to-C movement, but this is no longer the case in the Old

Irish period, we need to explain how this movement process was lost and how the

post-syntactic operations responsible for deriving the double system arose. What is

the status of these rules? It seems likely that the situation described here for Old Irish

would be relatively unstable, with the double system being learnt as irregularities.

From this we would perhaps predict that the system was difficult to maintain and so

was lost. This clearly happened between the Old and Modern Irish periods, but how

and why? These questions are investigated in chapter 5.


91

Before we can go on to examine the diachronic issues surrounding the Old Irish

verbal system we need a better understanding of the nature and operation of syntactic

change within the framework of the minimalist programme. This is the aim of chapter

3.

CHAPTER 3

SYNTACTIC CHANGE IN A GENERATIVE FRAMEWORK

1 INTRODUCTION

In the previous chapter it was argued that the reason that the Old Irish verbal system

proves so difficult to account for synchronically within the framework of minimalist

syntax is because it is not a syntactically productive system. To gain a fuller

understanding of how the verbal system of Old Irish works then we must consider the

diachronic perspective, namely how the system developed and how it was lost. This

will be the subject of chapters 4 and 5 respectively. Before we can embark on the

diachrony of Old Irish, however, it is necessary to outline some initial assumptions

regarding the theoretical framework in which the analyses developed in the following

chapters will be based. This is the primary aim of this chapter.

This dissertation is concerned with changes affecting the Irish verbal system.

Although as will be seen in the chapters to follow the verbal system has been affected

by a wide range of changes, this dissertation focuses primarily on syntactic change.

As with the synchronic account presented in chapter 2, the diachronic account

developed in chapters 4 and 5 will be based within the framework of the minimalist

programme. This chapter outlines how syntactic change can be modelled within the

minimalist framework, and how modelling syntactic change in this way can

contribute to our understanding of the mechanisms and causes of syntactic change.

Section 2 examines generative approaches to syntactic change, focussing on how

syntactic change can be modelled within the principles and parameters framework.

Section 3 considers the concept of parametric change within the minimalist

programme. In section 4 we explore how children set their parameters during

language acquisition, and how parametric changes arise. Section 5 summarises the

major theoretical assumptions outlined in this chapter that will be used to tackle the

Irish data in the remainder of the dissertation.

2 PARAMETRIC CHANGE AND THE ROLE OF LANGUAGE ACQUISITION

The minimalist programme, as outlined in chapter 1, is a version of the principles and

parameters theory of universal grammar. According to principles and parameters

theory the language faculty is seen as consisting of a number of invariant principles

that hold cross-linguistically and a number of binary parameters that encode the

CHAPTER 3 – SYNTACTIC CHANGE

93

variation between languages. So, for example, it is often argued that the extended

projection principle (EPP) holds cross-linguistically1 – every clause must have a

subject in Spec-TP. However, there is a parameter that determines whether or not the

subject must be realised overtly if it is a pronoun, namely the null subject parameter.

In English the subject must always be expressed. In Italian, however, this is not the

case. This is shown by the examples below:2

1. (a) He speaks English

(b) * Speaks English

2. (a) Lui parla italiano

(b) Parla italiano

At birth English and Italian children have the same linguistic knowledge – they are

endowed with the same principles and parameters. However, during the course of

language acquisition they are exposed to different linguistic data and so set these

parameters differently. Linguistic change between two stages of a language can be

viewed in exactly the same way. A younger generation of language learners, although

endowed with the same principles and parameters as their parents, may, for some

reason, be faced with different linguistic data to the older generation, and so set their

parameters in a different way. If we consider again the null subject parameter, Old

French behaved similarly to Italian in that pronominal subjects did not need to be

expressed:

3. (a) Einsi partirent del port de Venise come vos avez oi

Thus left.3PL the port of Venice as you have heard

(b) Si firent grant joie la nuit

So made.3PL great joy that night (Roberts 1993: 84)

Modern French, however, patterns more like English where the presence of a subject

pronoun is obligatory. Between Old and Modern French there has been a change in

parameter settings. Over time the evidence for null subjects decreased. Phonological

1 Although see McCloskey (1996b, 2001a) who argues on the basis of evidence from Modern Irish that the EPP on T is not in fact universal. 2 Expressing the subject in null subject languages is not ungrammatical, but has an emphatic effect.


94

change led to a loss of person distinctions in the verbal endings and so speakers began

to use pronominal subjects with greater frequency until the point where the use of

pronominal subjects was reanalysed as being obligatory and the parameter setting was

changed.3

Under the generative view, then, syntactic change involves a change in parameter

settings between subsequent generations. Parameters are set by a child acquiring its

language. Therefore, syntactic change must necessarily take place during language

acquisition. In this respect generativists have built on the work of Henning Andersen

(1973). Andersen observes that children do not have direct access to the grammar

they are attempting to acquire, i.e. the grammar of the older generation, but instead

they must attempt to recreate it on the basis of the linguistic evidence they hear, i.e.

the output of the earlier generation’s grammar. It is because of this discontinuity in

transmission that it is possible for changes to occur. In most cases children will

analyse the data they hear in the same way as the older generation, and language

acquisition will be “successful”. However, there is the possibility that children may

analyse the data they hear in a different way to earlier generations. This being the

case, their grammar will differ from their parents’ grammar. This difference will be

reflected in a difference in their output. This is shown schematically in the diagram

below:4

4.

The basic mechanism of syntactic change, then, under the generativist view, is

reanalysis, first proposed by Langacker (1977) and Timberlake (1977) (see also

Roberts (1993) for a discussion of the relationship between parametric change and

reanalysis). The child analyses the data that it encounters in a different way to the 3 This is a simplification. See Roberts (1993) for more details. 4 This diagram is an idealization. As pointed out by Weinreich, Labov & Herzog (1968: 145–6), children tend to acquire language primarily from their peers, i.e. other slightly older children, rather than their parents.

Grammar 1 (adult)

Grammar 2 (child)

Output 1 Output 2


95

older generation. As a result the child sets its parameters differently to the parameter

settings of the older generation. Returning to the French example, French children

reanalysed the data as suggesting that subject pronouns were obligatory, rather than

optional as specified by their parents’ grammars, and so set the null subject parameter

accordingly.

In this section it has been established that syntactic change can be modelled as a

change in parameter settings between subsequent generations. However, in order to

determine exactly what this means we must establish first what a parameter is and

what a change in parameter settings entails. This is the subject of the next section.

3 PARAMETERS IN THE MINIMALIST PROGRAM

Under the earlier government and binding style approach to the principles and

parameters model, the core of UG consisted of universal principles and a set of

parameters that were set during the process of language acquisition. Each parameter

had a specific function, for example the head parameter determined whether a

language had basic head-complement (e.g. VO) or complement-head (e.g. OV) order

or the null subject parameter determined whether languages allowed null subject

pronouns or not. In minimalism, however, cross-linguistic variation is encoded

somewhat differently.

In chapter 1 section 2.2.1 we saw that in minimalism the language faculty consists

of a computational system (CHL) and a lexicon. CHL contains invariant principles of

UG, which provide the basic mechanisms by which sentences are constructed. All

variation between languages, both lexical and syntactic, is encoded in the lexicon. In

this respect minimalism has an obvious conceptual advantage to earlier versions of

the theory. It cannot be denied that languages differ in terms of vocabulary and so to

acquire a particular language a child must acquire the lexical items that make up the

lexicon. If syntactic variation is encoded in the lexicon, then it seems likely that the

parts of syntax that need to be acquired will be acquired in the same or perhaps a

similar way to other lexical information.

Syntactic variation is encoded in a particular subset of lexical items, namely

functional categories, such as C, T and v – the so-called ‘core functional categories’

(Chomsky 2000: 102, 2001: 6). Variation is encoded in terms of grammatical features.

In different languages functional categories will contain different features or different

feature values, and this will result in different syntactic properties. Let us examine


96

how this might work. In chapter 1 we saw that there are three syntactic operations in

the minimalist programme, namely Merge, Agree and Move. Merge is the basic

structure building operation; it is not motivated by features and so is not subject to

variation.5 Agree and Move both depend on the presence of grammatical features and

so can vary both cross-linguistically and diachronically. Variation in terms of

movement is most relevant here, so let us consider this in more detail.6

We saw in chapter 1 that for movement to take place, the constituent to be moved

(the goal) and the target of movement (the probe) must be in an Agree relation.

However, although an Agree relation is necessary for movement, it is not sufficient.

For movement to take place, the probe must also contain an EPP-feature. This may be

a general EPP-feature, attracting a phrasal category, or an Affix-feature, a type of

EPP-feature that attracts a head. The presence of a particular movement operation in a

given language, then, will crucially depend on the presence of an EPP/Affix-feature.

Let us consider a synchronic example. It is well known that English and French differ

in terms of the position of the verb (Emonds 1978; Pollock 1989). In English, the verb

must appear adjacent to the direct object; no element can intervene. This is shown in

the examples in (5). In French, this is not the case, and adverbs such as souvent

‘often’ or negation, pas, can intervene between the verb and the object as shown by

the examples in (6).

5. (a) I often eat cheese

(b) *I eat often cheese

(c) I do not eat cheese

(d) *I eat not cheese

6. (a) *Je souvent mange du fromage

(b) Je mange souvent du fromage

(c) *Je ne pas mange de fromage

(d) Je ne mange pas de fromage

5 Although see Chomsky (2005, 2006) where it is argued that the operation Merge is motivated by Edge Features (EF). The recursive nature of Merge is argued to result from the fact that all lexical items have an EF that must be satisfied. Although under this view Merge is motivated by features, the fact that, according to Chomsky, all lexical items have an EF, suggests that the EF and the operation Merge will not be subject to variation. 6 For an example of variation in terms of Agree see the discussion of negative concord in Roberts & Roussou (2003: 136–154).


97

In French T has an Affix-feature, and so the verb moves to T to the left of negation

and VP-adverbs. In English T does not have an Affix-feature and so the verb remains

in situ within the VP. The cross-linguistic difference is a result of a difference in

terms of the features associated with T, specifically whether or not T has an Affix-

feature.

If we turn now to a diachronic example, we see that the same type of variation can

be found. It seems that Early Modern English was like Modern French in that it had

generalised V-to-T movement (Roberts 1985, 1993; Biberauer & Roberts to appear).

This can be seen in the examples in (7) below, where adverbs (7a) and negation (7b)

can intervene between the verb and the direct object.

7. (a) The Turkes…made anone redy a grete ordonnaunce

‘The Turks soon made ready (prepared) a great ordnance’

(Kaye, The Delectable Newsse of the Glorious Victorye of the Rhodyans

agaynest the Turkes; Gray (1985: 23), Roberts (1993: 253)

(b) If I gave not this accompt to you

‘If I did not give this account to you’

(J. Cheke, Letter to Hoby; Görlach (1991: 223). Roberts (1999: 290))

In minimalist terms this means that in Early Modern English T had an Affix-feature.

In Modern English, however, as seen in the examples in (5), T does not have an

Affix-feature. Between Early Modern English and Modern English T lost this Affix-

feature and the associated V-to-T movement. This type of parametric change, then,

can be seen as the change in value of an EPP or Affix-feature associated with a

particular functional head.

So far it has been argued that within the minimalist programme parameters are

encoded in terms of formal features on functional heads. Parametric differences in

terms of movement operations result from the presence or absence of EPP and Affix-

features on functional heads. Viewed in this way, then, during language acquisition

children must determine for each functional head which formal features are associated

with that head and whether it has an EPP- and/or an Affix-feature.7 The next question

7 In terms of EPP/Affix-features this view is quite simplistic. A given functional head will actually not have the same EPP/Affix-feature values in all contexts. For example, in declarative clauses C in English has neither an EPP- nor an Affix-feature. In root wh-interrogative clauses on the other hand, it


98

that arises is how a child determines this, how does a child set its parameters? This is

the topic of the next section.

4 HOW TO SET PARAMETERS

In section 2 above it was argued that at birth all children are endowed with the same

linguistic knowledge and it is only through exposure to different linguistic data that

children in different environments acquire different languages. The principal question

then is how do children use this evidence to set their parameters? Parameters are

generally argued to be abstract entities, far removed from the structures that make up

the linguistic data. How then do children know that a particular construction they hear

in the linguistic data corresponds to a particular abstract parameter setting? We need

something to bridge this gap. For our purposes we will adopt the cue-based model of

language acquisition to provide a link between the linguistic data and abstract

parameter settings.8

4.1 Cue-based models of language acquisition

According to cue-based models (Dresher & Kaye 1990; Dresher 1999) UG specifies

not only a set of parameters, but also a cue for each parameter, that is a piece of the

input data that provides the child with evidence for a particular parameter setting. If

this specific structure is not found in the linguistic data then the parameter will revert

to its default setting.

Dresher (1999) illustrates the cue-based theory with reference to a phonological

parameter, namely Quantity Sensitivity (QS). In English the position of word stress

depends on the distribution and location of heavy syllables within a word. Stress falls

on the penultimate syllable if that syllable is heavy, i.e. contains a long vowel (8b) or

is closed by a consonant (8c). If not, then stress falls on the antepenultimate syllable

(8a).

has both, leading to wh-movement and inversion. There are two solutions to this. Either the different types of C are different lexical items and during acquisition a child must acquire all the possible types of C. The alternative is to suggest that certain features have their own EPP/Affix specification and when C contains such a feature, this feature overrides C’s default EPP/Affix-feature values. In the present context it is not necessary to decide between the two. 8 For summaries and critiques of various other approaches to language acquisition see Dresher (1999), Lightfoot (1999: 144–149)


99

8. (a) álgebra, Cánada, América

(b) Vancóu:ver, aró:ma

(c) agénda, appéndix

In order to acquire this stress system, the child must determine that English

distinguishes between light and heavy syllables. Dresher proposes that this involves a

parameter. Children start from the assumption that all syllables are equal with regards

to stress, i.e. the default setting is for Quantity Insensitivity (QI). However, when they

encounter words of the same number of syllables with different stress patterns (i.e. the

specified cue) they change this parameter setting, and so acquire quantity sensitivity.

This is summarised in (9) below:

9. Quantity (in)sensitivity

(a) Parameter: the language does/does not distinguish between light and

heavy syllables

(b) Default: assume all syllables have the same status (QI)

(c) Cue: Words of n syllables, conflicting stress contours (QS).

(Dresher 1999: 31)

Lightfoot (1999, 2006) adopts Dresher & Kaye’s notion of cue and applies it to the

acquisition of syntax.9 Lightfoot utilises cues to explain the loss of V2 in English. The

main feature that distinguishes V2 languages from SVO languages is that in V2

languages any constituent can appear before the finite verb, whereas in SVO

languages only the subject can occupy this position. Therefore, it seems likely that the

cue for the V2 parameter (assuming for now that such a parameter exists) should be

the presence of a non-subject XP in initial position followed directly by the verb. The

child scans the input and if it finds sufficient evidence of XVS order then the V2

parameter is set positively.10 If it does not find this evidence then the V2 parameter

9 Lightfoot’s concept of a cue differs from that proposed by Dresher & Kaye as he argues that cues and parameters can in fact be seen as one and the same thing: ‘cues which are realised only in certain grammars constitute the parameters, the points of variation between grammars’ (Lightfoot 1999:149). However, it is unclear what advantage Lightfoot hopes to achieve by amalgamating these two concepts. See Roberts & Roussou (2003: 13–14) for criticism of this view of cues. 10 This notion of ‘sufficient’ or ‘robust’ evidence is employed frequently in the generative literature on syntactic change; however, it is problematic. How much evidence is sufficient? No satisfactory answer has been given to this question, and it seems unlikely that one ever will. Lightfoot (1999: 154) suggests that for the V2 parameter to be robustly cued XVS clauses must make up 30% of the trigger


100

remains in its default setting, i.e. it is set negatively.11 We will return, in detail, to the

application of the cue-based model of language acquisition to syntactic change in the

next section.

One advantage of the cue-based model of language acquisition is that it reduces the

cognitive burden on the learner compared to input matching models, such as Gibson

& Wexler (1994). If children are pre-programmed with specific information about

which structures they need to find in the trigger experience in order to fix their

parameters, then this could account for the fact that language acquisition seems to

take place so quickly and effortlessly. This is more difficult to account for if it is

assumed that children must ‘try out’ numerous different parameter settings or

competing grammars before converging on the correct one. However, the reduction in

cognitive burden is accompanied by a significant increase in the amount of

information that must be innate. If for each parameter UG specifies a cue then this

doubles the information that must be innate.12

A further advantage of the cue-based approach is that it allows us to account for

one of the problems concerning the linguistic data. Lightfoot (1991) argues that

during language acquisition children make no use of negative data. This means that

they make no use of constructions they do not hear. Lightfoot observes that this is

particularly problematic in the case of the null subject parameter. A child learning

English must learn that omitting subject pronouns is not allowed. How is this possible

without reference to information that such constructions are ungrammatical? Within a

cue-based model a child scans the linguistic data for a cue. If the child finds this cue,

then the parameter is set positively. If the child does not find the cue, then the

experience. Even if this figure is correct for the acquisition of V2 (see Westergaard to appear), it seems unlikely that the necessary level of robustness will be the same for every parameter. This dissertation will add nothing further to this issue. 11 The information the child needs to set the V2 parameter cannot come directly from the input data. In order to set this parameter, the structure needs to be at least partially parsed. For example, in order to determine whether there are examples of the structure XVS in a language the child must have some notion of categories and constituency. She must be able to determine what is a verb and what is a noun. More importantly she must have some way of determining that the subject is not a relevant initial constituent. Dresher & Kaye acknowledge this, and propose that parameter setting proceeds in an order determined by UG. This creates a learning path (see also Lightfoot 1989 and Baker’s 2001 ‘parameter hierarchy’). Initially cues are based entirely in the input, but as the child progresses along the learning path cues become increasingly abstract and grammar-internal. 12 A further question here is whether there can be a one-to-one correspondence between cues and parameters. From an acquisitional point of view this is clearly desirable. However, it is not entirely clear that this is the case. We will return to this issue in the next section.


101

parameter receives the default setting.13 The acquisition of the English setting for the

null subject parameter can be explained if the English setting is the default. The child

will scan the linguistic data for evidence of omitted subject pronouns. If the child

finds evidence of null subjects, then the parameter will be set positively, giving the

Italian setting. However, if the child finds no evidence (or insufficient evidence) for

null subjects then the parameter will have its default setting, i.e. the English setting.14

4.2 A cue-based model of syntactic change

If, following Lightfoot, we are going to invoke a cue-based model to account for

syntactic change, then we must establish what can constitute a possible cue. In this

section we will focus on the type of parameter most relevant to the topic of this

dissertation, namely head-movement. It was argued in section 3 that during language

acquisition children set their movement parameters by establishing whether a given

head is ±Affix or ±EPP. Under the cue-based approach each parameter must have a

default and a positive setting. If parameter setting is seen as assigning functional

features to functional heads, then it seems reasonable to suggest that the ‘default

parameter setting’ is for the functional head to contain no (non-default) features.15

This being the case, for a child to assign an EPP- or an Affix-feature to a functional

head, there must be positive evidence. EPP- and Affix-features must be robustly cued.

If EPP- and Affix-features are associated with cues, we must establish what the

relevant cues might be. In his discussion of the V2 parameter in the history of

English, Lightfoot (1999, 2006) argues that the relevant cue is a syntactic one. In

order to acquire V2 (in our terms, for C to be assigned +EPP and +Affix-features)

children must hear sufficient structures of the type XVS, where a single constituent

that is not the subject precedes the verb. This is clearly a syntactic cue, specifying

particular structure. If we consider an alternative parameter, namely the V-to-T

parameter, a similar structural parameter can be proposed for this. It was argued in

section 3 above that when the verb raises to T it appears to the left of any material

13 The concept of default parameter settings is somewhat controversial, as in many cases it is difficult to determine what the default is. For movement parameters in the framework developed here, however, the notion of default is relatively straightforward. The default setting involves no movement features being assigned to the functional head. For this reason, we will maintain the concept of default in what is to follow. 14 Although see Hyams (1986) who proposes that the Italian setting is the default. 15 It seems likely that each functional head will have a set of core features that are always present on that head cross-linguistically.


102

marking the edge of vP, such as vP adverbs and negation (in languages where

negation is marked in this way), or in verb-initial languages, the subject. The cue for

an Affix-feature on T then could be verb-adverb, verb-negation or verb-subject

orders. However, this syntactic evidence is not the only possibility for the acquisition

of V-to-T movement. It is possible that verbal morphology could also play a role.16

In both synchronic and diachronic work within the generative tradition a link is

often drawn between morphology and movement. The majority of this work focuses

on the correlation between rich verbal morphology and movement to T (Roberts 1985,

1993; Vikner 1997; Rohrbacher 1999). It tends to be the case that languages that have

rich verbal morphology also have V-to-T movement. So, as shown in example (10)

below, Italian has rich verbal morphology and V-to-T movement, whereas English, as

shown in (11), has impoverished verbal morphology and no V-to-T movement.

10. (a) leggo sempre il giornale

read.1SG always the newspaper

(b) leggono sempre il giornale

read.3PL always the newspaper

11. (a) I always read the newspaper

(b) They always read the newspaper

This correlation between morphology and verb-movement could be taken as evidence

that inflectional morphology is a cue for the acquisition of V-to-T movement. If the

verb is consistently found with inflections that are associated with a particular

functional head, this can be taken as evidence that the verb has moved to this

functional position. Children acquiring Italian will encounter verbs with rich

inflectional morphology, and so will set their V-to-T parameter positively. English

children, on the other hand, will not encounter such morphology and so will not set

the V-to-T parameter positively.

16 If this view is correct and V-to-T movement can be cued by verb-adverb, verb-negation or verb-subject orders or by verbal morphology then there cannot be a one-to-one correlation between cues and parameters. As noted in footnote 11 this is undesirable from an acquisitional perspective. One way around this would be to argue that the actual cue provided by UG for V-to-T movement is simply evidence that the verb is in T. Viewed in this way, verb-adverb, verb-negation and verb-subject constructions and verbal morphology simply provide evidence for the cue, they provide evidence that the verb is in T and so enable the child to set the V-to-T parameter positively.


103

However, the notion of morphology as a cue is not straightforward, as the

correlation only seems to hold one way. There are many languages that seem to show

V-to-T movement but do not show rich verbal morphology. The example in (12)

below is from the Celtic language Manx. In (12a) and (12b) the verb has the same

form although in (12a) the subject is 1sg and in (12b) the subject is 3pl.

12. (a) shen-y-fa haink mish bashtey lesh ushtey

therefore came I baptizing with water

‘Therefore I came baptizing with water’ (NE ch1v31)

(b) As haink ad gys Ean

and came they to John

‘And they came to John’ (NE ch3v26)

In languages such as Manx, then, the morphological cue for V-to-T movement clearly

plays no role. V-to-T movement must be acquired purely from syntactic evidence,

such as the fact that the verb precedes VP-adverbs, or in verb-intial languages such as

Manx perhaps, that the verb precedes the subject. From this it seems that a

morphological cue is by no means necessary for the acquisition of V-to-T movement.

The status of verbal morphology as a cue for acquisition would perhaps be on

firmer ground if evidence could be found of a case where morphology is the only cue

causing the acquisition of V-to-T movement. However, all languages with rich

inflection that could utilise a morphological cue also show the relevant syntactic

configurations such that V-to-T movement is also syntactically cued. So, for example,

in languages such as Italian that show rich verbal inflection, the verb also appears to

the left of VP-adverbs. So, it seems that in the case of V-to-T movement at least there

is no clear synchronic evidence to support the postulation of an independent

morphological cue. At best, morphological evidence can only reinforce the syntactic

evidence, making the evidence for V-to-T movement more robust. Let us now

consider the relationship between morphological and syntactic cues from a diachronic

perspective.17

17There is a third possible cue that may play a role in the setting of syntactic parameters. It is possible that the acquisition of an Affix- or EPP-feature on a given head could be cued by the feature specification of a related head. This is a kind of implicational cue, so if C is +Affix, then we might expect T and v to also be +Affix. It is often assumed in work on the history of English (Roberts 1993; Lightfoot 1999) that the presence of V-to-C movement requires V-to-T movement, due to the HMC.


104

It was argued in the previous section that children set their parameters by scanning

the linguistic data for specific cues provided by UG. If the child finds the cue the

parameter is set positively. If the cue is not found in the data then the parameter

receives its default setting. It seems that there are two possible ways in which a

parameter can change: either a cue is lost, and the parameter returns to its default

setting, or a new cue develops and the parameter changes from the default to the

positive setting. In both cases, a change in parameter settings arises as a result of a

change in the trigger experience that either obscures or creates a cue. Let us examine

how changes in the trigger experience obscured the cues and led to the loss of V-to-T

movement in English, and what this can tell us about the interaction of morphological

and syntactic cues.

It was argued above that V-to-T movement could possibly be cued by the presence

of rich tense or agreement morphology on the verb. The most likely way for

morphological cues of this type to become obscured is through phonological change.

Phonological change may eradicate the phonological differences between different

endings, causing them to fall together and the morphological distinctions to be lost.

Such a change can be seen to have affected verbs in the history of English. Due to

phonological changes such as the reduction of final vowels to schwa and the loss of

final nasals, verbal endings fell together (see Roberts 1993 for more detail). It is often

argued that this loss of verbal agreement morphology led to the loss of V-to-T

movement (Roberts 1985, 1993). When the verbal morphology became impoverished,

the morphological cue for V-to-T movement became obscured, leading to a change in

the parameter setting.

At first sight, it seems that we have a strong diachronic argument for the existence

of a morphological cue. However, on closer inspection it seems that the loss of the

morphological cue cannot have directly caused the loss of V-to-T movement.

Although the loss of the morphology was perhaps a necessary condition it was not

sufficient. V-to-T movement survived in English for at least 75 years after the loss of

verbal morphology (Roberts 1999: 292). After the morphological cue was lost, the

However, as discussed in chapter 1 it is not clear what role this plays in minimalism. Furthermore, it is not entirely clear that V-to-C movement always implies the presence of V-to-T movement. It has been shown that in the mainland Scandinavian languages although the verb moves to C in V2 constructions it does not move via T (Holmberg & Platzack 1995). Implicational cues will play no further role in this dissertation, but they are perhaps a topic worthy of further research.


105

syntactic cue still enabled V-to-T movement to be acquired. It was not until the

syntactic cues were also obscured that V-to-T was finally lost.

Roberts (1993) argues that the change in the trigger experience that caused the loss

of the syntactic cue for V-to-T movement was the rise of do-insertion. In order for V-

to-T movement to be cued syntactically, the verb must appear to the left of elements

that mark the left edge of the VP, i.e. adverbs and negation. However, an increase in

the frequency of do-insertion meant that the verb raised to T less frequently.18 As a

result the cue was obscured, and the parameter reverted to its default setting. Before

the loss of V-to-T movement in English it was possible to use the do-insertion

construction in place of verb-movement. Do-insertion was a stylistic option that could

be used in a much wider range of contexts than it can today, simply as an alternative

to verb-raising. The examples in (13) below are from Shakespeare.

13. (a) Where eyes did once inhabit

(Richard III, Barber 1976: 163, Biberauer & Roberts to appear: 21)

(b) Rough windes do shake the darling buds of Maie

(Sonnet 18, Biberauer & Roberts to appear: 22)

The increase in do-insertion that led to the loss of V-to-T movement in English

provides an example of the most common way in which a syntactic cue can be

obscured. An optional construction, which carries some kind of stylistic effect,

increases in frequency to such an extent that it has an effect on the underlying

parameter settings. Often as a result of this change in frequency the optional, stylistic

operation becomes generalised and obligatory. For example Lightfoot (1999) argues

that the loss of V2 in English resulted from an increase in subject-initial clauses at the

expense of XVS clauses. As a result of this increase in frequency, V2 was no longer

robustly cued, and so the V2 parameter returns to its default, and subject-initial SVO

clause structure becomes the normal grammatical clause structure. Of course, to truly

explain what caused these changes we need to explain what caused the stylistic shifts

in frequency to occur. Why did do-support increase? Why did subject-initial clauses

become preferred? The answers to these questions are difficult to pin down. Stylistic

18 See Kroch (1989) on the progress of this change.


106

changes could be motivated by sociolinguistic factors, or perhaps by language

contact.19

So it seems that there are two main ways in which changes in the trigger

experience could obscure a cue. Phonological change can lead to the loss of

morphological cues, and changes in the frequency of certain stylistic variants can lead

to the loss of syntactic cues. It seems plausible that cues may develop in a similar

way. A morphological cue may develop as a result of morphological change, for

example, as proposed for Germanic by Fuß (2005) a clitic may be reanalysed as an

affix (see also Alexiadou & Fanselow 2002 on the development of subject-agreement

morphology). The development of a syntactic cue may result from a change in

frequency of a stylistic variant, leading to the generalisation of a previously optional,

stylistically marked construction. It will be argued in chapter 4 that both of these

changes occurred in the history of Irish, and led to a positive setting for the V-to-C

movement parameter.

The discussion of the development of V-to-C movement in chapter 4 and the loss

of V-to-C movement in chapter 5 will also enable us to consider the interaction

between morphological and syntactic cues from a different perspective. We have seen

in this section that in the case of V-to-T movement there is no evidence for an

independent morphological cue. From both a synchronic and a diachronic perspective

there seems to be no evidence of morphological cues acting independently of the

syntactic cue. The remaining chapters of this dissertation will attempt to establish

whether the development and loss of V-to-C movement provides any evidence for

independent morphological cues. This will lead to a clearer understanding of the link

between morphological and syntactic change.

6 CONCLUSION

The aim of this chapter was to provide some theoretical background for the account of

syntactic change in the history of Irish that is to follow. We have examined how

19 If we accept that implicational cues exist as discussed in footnote 16, then there is a further way that a cue could become obscured. If the Affix-feature on T is cued by the presence of an Affix-feature on C then this cue could be lost as a result of a parametric change. For example, in the history of English, while English was a V2 language this also cued V-to-T movement. However, with the loss of V2, this cue was lost. However, the loss of V2 did not result in the immediate loss of V-to-T movement (van Kemenade 1987; Biberauer & Roberts to appear). It seems that like the loss of morphology this can only have been a contributing factor, and so it is unclear that this should be considered a cue in its own right.


107

syntactic change can be modelled within a generative framework and have developed

a theory of acquisition and language change based on the ideas of the minimalist

program.

Within the minimalist program all variation between languages is encoded in the

lexicon. Specifically, syntactic variation is encoded in the features of functional

categories. This chapter focussed on the differences in basic word order between

languages and concluded that this can be encoded in terms of the presence or absence

of EPP- and Affix-features on each functional head. These features are purely

syntactic. They are uninterpretable at the interfaces and so must be eliminated in the

syntax.

When children acquire language they must determine for each functional head

whether it is ±EPP and ±Affix. To do this children utilise the linguistic data or trigger

experience that surrounds them. In this chapter we have argued for a cue-based

approach to language acquisition. For each parameter, i.e. for each Affix- and EPP-

feature associated with each functional head there is a specific cue specified by UG. If

the children locate the required cue in the trigger experience then they postulate an

EPP or Affix-feature for that head. If the cue is not found then the parameter receives

its default setting, i.e. neither an EPP- nor an Affix-feature is postulated.

A change in parameter settings can occur in two possible ways, both of which are

dependent on the trigger experience. Changes in the trigger experience can either

cause an existing cue to become obscured or ambiguous or they can cause a new cue

to develop. The former option will result in the parameter being assigned its default

setting, i.e. no Affix- or EPP-feature will be assigned. The latter option will result in a

new positive parameter setting, i.e. a new Affix- or EPP-feature will be assigned.

Examples of each type will be provided in the following chapters.

CHAPTER 4

THE DEVELOPMENT OF THE DOUBLE SYSTEM

1. INTRODUCTION

The aim of this chapter is to provide an account of how the double system of verbal

inflection developed in Old Irish from Proto-Indo-European (PIE). As noted in

chapter 1, this dissertation is by no means the first work to tackle this issue. The

origins of the Old Irish double system have been hotly debated since the beginnings

of Celtic Philology in the nineteenth century, and as yet no firm conclusion has been

reached. However, this is not to say that this previous work has been fruitless. Many

interesting and important insights have been gained through the philological study of

the Old Irish verbal system. This chapter aims to combine these philological insights

with minimalist ideas about syntactic change to develop a new account of how the

double system of verbal inflection developed in Old Irish.

This chapter is structured as follows: section 2 reviews the two main philological

accounts of the development of the Old Irish double system of verbal inflection.

Section 3 begins the syntactic analysis, providing some remarks on the syntax of PIE

viewed within a minimalist framework. Section 4 looks at the development of V-to-C

movement in Old Irish, and how this was linked to the development of new verbal

morphology. Sections 5 and 6 consider how preverbs and conjunct particles

developed between PIE and Old Irish. Section 7 concludes the chapter.

2. PHILOLOGICAL ACCOUNTS OF THE DEVELOPMENT OF THE DOUBLE SYSTEM

2.1 Introduction

As noted in the introduction, the existing philological work on the origins of the

double system of verbal inflection in Old Irish has provided many useful insights,

which will play an important role in the account presented in the remainder of this

dissertation. This section, then, aims to provide an introduction to the two main

theories that have been proposed to account for the origin of the double system,

namely Cowgill’s (1975, 1985) particle theory and McCone’s (1979b, 1982, 1985)

suffix-/infix-deletion theory. Before we go on to evaluate these two accounts,

however, let us first remind ourselves of what it is that any theory of the origins of the

double system must explain. There are two crucial aspects of the double system that

must be explained: its phonological shape and its distribution. The distributional facts,

CHAPTER 4 – THE DEVELOPMENT OF THE DOUBLE SYSTEM

109

although typologically odd, are descriptively straightforward. Absolute endings

appear only on simple verbs in absolute initial position, as shown in (1a). When the

verb is compound or in non-initial position it has conjunct endings, as in (1b–d).

1. (a) berid cách brith for arele

bear.PRES.3SG.ABS each judgment on other

‘Each bears judgment on the other’ (Wb 29b9)

(b) do-beir inso arnab uilib cumactib dichoissin

give.PRES.3SG.DT this for.the all powers exist(?)

‘He gives this for all the powers that exist…’ (Wb 21a13)

(c) cenid leci in metur…


‘Although the metre does not allow…’ (Ml 30a10)

(d) ní tabair desimrechta híc

neg give.PRES.3SG.PT examples here

‘He does not give examples here…’ (Sg 214a5)

Furthermore, compound verbs show different stems dependent on their position in the

clause. In absolute initial position a compound verb is deuterotonic, (1b), and in non-

initial position it is prototonic, (1d).

Let us turn now to the phonological differences between the absolute and conjunct

inflection of simple verbs. These can be seen from the paradigms given in table 1

below of the verbs léicid ‘lets’, marbaid ‘kills’ and berid ‘carries’.

Table 1: Absolute and conjunct inflection

Absolute Conjunct Absolute Conjunct Absolute Conjunct

1sg léiciu -léiciu 1sg marbu -marbu 1sg biru -biur

2sg léici -léici 2sg marbai -marbai 2sg biri -bir

3sg léicid -léici 3sg marbaid -marba 3sg berid -beir

1pl léicmi -léicem 1pl marbmai -marbam 1pl bermai -beram

2pl léicthe -léicid 2pl marbthae -marbaid 2pl beirthe -berid

3pl léicit -léicet 3pl marbait -marbat 3pl berait -berat


110

As can be observed from the above paradigms, absolute verb forms tend to be longer

than conjunct forms. In the singular the absolute contains an extra syllable. In the first

and second plural the absolute has the same number of syllables as the conjunct, but

the absolute has retained a final syllable which was lost in the conjunct, instead losing

an internal syllable through syncope (e.g. 1pl absolute *beromos>ber∅mai vs. 1pl

conjunct *beromos>beram∅). Since Cowgill (1975) it has been widely accepted that

verbs in Celtic underwent a process of early *i-apocope whereby final *-i was lost

and so the IE primary and secondary endings fell together.1 2

2. *bhereti >*bheret > -beir ‘carries’ (conjunct)

As shown in (2), it seems to be the Old Irish conjunct form that developed from this

single inherited IE form. To obtain the correct absolute form, it seems that the end of

the word must have been protected from the process of *i-apocope by some kind of

phonologically overt entity. This extra phonological material would have prevented

the *i from being word-final and so would have prevented it from being apocopated.

This is shown in (3) below.

3. *bhereti+X > berid ‘carries’ (absolute)

Turning now to compound verbs, the differences between the deuterotonic and

prototonic forms are a lot more pronounced than between the absolute and conjunct.

This can be seen in the paradigms given in table 2 below for the compound verbs do-

beir ‘gives’, ad-cí ‘sees’ and do-gní ‘does’.

1 The exact phonological details of this sound change are unclear. McCone (1978) argues that *i-apocope took place across the board. However, Schrijver (1994: 159–165) argues that the only convincing evidence for *i-apocope comes from 3sg and 3pl verbal endings (-ti and -nti), the short dative endings of t- and k- stems and the prepositions fri<*writ<*writ-i ‘to’ and la<*let<*let-i ‘with’. In all these cases the final *i was preceded by a voiceless stop, suggesting that the change was phonologically conditioned. More research is necessary to determine which of these options is correct. 2 See Cowgill (1975: 41–6) for arguments against the earlier view that absolute and conjunct inflection developed from PIE primary and secondary endings.


111

Table 2: Deuterotonic and prototonic forms

Deuterotonic Prototonic Deuterotonic Prototonic Deuterotonic Prototonic

1sg do-biur -tabur 1sg ad-cíu -aicciu 1sg do-gníu -dén(a)im

2sg do-bir -tabair 2sg ad-cí -aci 2sg do-gní -dén(a)i

3sg do-beir -tabair 3sg ad-cí -aicci 3sg do-gní -dén(a)i

1pl do-beram -taibrem 1pl ad-ciam -accam 1pl do-gníam -dénam

2pl do-berid -taibrid 2pl ad-ciid -accid 2pl do-gníith -dénat

3pl do-berat -taibret 3pl ad-ciat -accat 3pl do-gníat -dénatar

The most important difference between the two forms of the compound verb is the

difference in stress. The prototonic form is stressed on the first syllable in accordance

with regular Old Irish stress patterns. The deuterotonic is stressed on the second

syllable. As we saw in chapter 2, this difference in stress placement led to the two

forms being affected differently by the various sound changes that affected the

language as a whole, such as syncope and vowel affection.

The other major phonological difference between deuterotonic and prototonic

forms is that when the verb is prototonic the initial preverb can lenite or nasalize the

following segment. In the case of the deuterotonic these mutations seem to be

blocked. In example (4a) below the [f] of the verbal stem fera is present in the

deuterotonic but not in the prototonic, where it has been lenited from /f/ to ∅. In (4b)

the final [n] of the preverb con has caused the initial consonant of the verb certar to

nasalise in the prototonic, so it is pronounced [g] in the prototonic and [k] in the

deuterotonic.

4. (a) fo-fera -foírea ‘causes’

(b) con-certar -coic[g]erta ‘corrects’

(McCone 1997a: 4)

The particle theory and the suffix-/infix-deletion theories account for the phonological

peculiarities of the absolute/conjunct and deuterotonic/prototonic forms in the same

way. Both theories postulate enclitic particles that appear in second position. These

particles are suffixed to simple verbs in absolute initial position and so protect the

final *i from apocope, and infixed between the initial preverb and the remainder of

initial compound verbs, separating the initial preverb from the remainder of the verb


112

in deuterotonic compounds. The particle theory postulates an enclitic sentence particle

with the form *es or *ed. The suffix-/infix-deletion theory derives essentially the

same effect from enclitic object pronouns. Having outlined the basic similarities

between the two theories, let us now examine each in more detail and consider how

they differ.

2.2 The particle theory

The particle theory was first introduced by Thurneysen (1907) to account for non-

lenition in compound verbs. Since then, a variety of scholars have adopted and

adapted the theory, invoking particles of varied phonological shape and function (see,

among others, Dillon 1947, Boling 1972, Kortland 1979). The most influential

version of the particle theory, however, is that of Cowgill (1975). Cowgill argues that

the morphological shape of the absolute and conjunct forms can be best accounted for

by a second position particle with the form *es. For a full account of the phonological

advantages gained by adopting the particle theory the reader is referred to Cowgill

(1975, 1985). It will be sufficient for our purposes to provide one paradigm for

illustration. The first three columns in table 3 show the development of the conjunct

forms, resulting from the loss of the final syllable between Primitive and Old Irish.

The final two columns demonstrate how the presence of the particle *(e)s prevented

apocope, giving the Old Irish absolute forms.

Table 3: The develoment of conjunct and absolute forms

Proto Celtic

Primitive Irish Old Irish Primitive Irish +*es

Old Irish

*berū > *berū > -biur * berū-s > biru *beresi > *berī > -bir * berī-s > biri *bereti > *beret > -beir * bereti-s > berid *beromos > *beromas > -beram * beromo-es > berm(a)i *beretes > *beretes > -berid * beretēs > beirthe *beronti > *berodd > -berat * beroddi-s > ber(a)it

Although it is clear that the particle theory can account for the basic distinction

between absolute and conjunct forms, namely the presence or absence of the final

syllable, many phonological objections have been raised against it (see McCone

1979b, 1982, 1985, Sims-Williams 1984, Isaac 2000). For the most part these

objections have been answered by Cowgill (1985) and Schrijver (1997). This


113

dissertation is not concerned with details of phonology and so will not enter into this

debate. For further details the reader is referred to the sources listed above.

The biggest objection to the particle theory is of a non-phonological nature. A

major problem with the particle theory is that no firm conclusion has been reached as

to the exact form and meaning of the particle. Thurneysen (1907) proposed that the

particle must have the shape *es and have developed from a form of the copula. The

problems with this etymology led Boling (1972) to propose that the particle was

originally a neuter pronoun of the form *ed. Cowgill (1975) returns to Thurneysen’s

*es, arguing that *ed has too many phonological difficulties, but again he cannot offer

an etymology. The issue was reopened by Schrijver (1994: 180f). Based on evidence

from the development of the prepositions fri ‘against’ and la ‘with’, Schrijver argues

that following the apocope of short *-i, the newly word-final *-t fell together with

word-final *-s. This being the case, Schrijver proposes that Cowgill’s particle *es

developed from *eti, a sentence connective cognate with Latin et ‘and’, Gothic iþ

‘but’, Ancient Greek éti ‘further, moreover’ and Sanskrit áti ‘beyond, very’.3 Clearly

none of these are subordinators and this characteristic could easily have been shared

by their Celtic counterpart, thus explaining why *es appears in main rather than

subordinate clauses (Schrijver 1994: 184). The cognates do not, however, offer an

explanation as to why the particle *es appeared in second position. The cognates are

all initial-position particles, but to have the required phonological effect the particle

*es in Celtic must have appeared in second position. Schrijver suggests that *es

became a second-position clitic as a Celtic innovation on the model of the other

connective clitics attested for Old Irish, namely ch<*kwe and d<*dè.

In conclusion to this section it seems that the main phonological objection to the

particle theory can be overcome. In addition, if we accept Schrijver’s etymology we

have provided an answer to the remaining criticism by giving a plausible explanation

for the function of the particle. In spite of the criticisms levelled against it, it seems

that, in general, the particle theory can be upheld. There is one major conceptual

problem with the particle theory, however, that has not been dealt with here. For the

particle to have the desired effect on the verbal forms it must have been a second

position enclitic, subject to Wackernagel’s Law, and the verb must have been in initial

position. However, the particle theory offers no explanation as to how the verb came 3 Eska (2002a) argues that Schrijver’s etymology is also supported by the Transalpine Celtic forms eti, etic<*eti-kwe and coetic<*ko-eti-kwe.


114

to be in initial position. As we will see in section 3, verb-initial order in Old Irish

cannot have been inherited from PIE, so some explanation is needed as to how it

arose. Let us leave this issue for now and consider the main alternative, namely

McCone’s suffix-/infix-deletion theory.

2.3 Suffix-/Infix-deletion theory

Having argued emphatically against the particle theory, McCone (1979b, 1982, 1985)

proposes his own account of the origins of the double system, namely the suffix-

/infix-deletion theory. McCone accounts for the morphological peculiarities of the

deuterotonic and absolute verb forms by the presence of an infixed or suffixed object

pronoun. This pronoun would have had the same phonological effect as the particle

*es, namely preventing *i-apocope in simple verbs and blocking lenition in

compound verbs. These pronouns, along with their accompanying mutations, were

then analogically deleted to give the resulting Old Irish absolute/conjunct and

deuterotonic/prototonic alternations. Initially McCone’s theory seems to have an

advantage over the particle theory as it has the same phonological effects as the

particle theory but does not invoke a historically unattested particle. However, if one

accepts Schrijver’s etymology for the particle *es as a sentence connective, then this

is no longer the case.

One area in which McCone’s theory does seem to be preferable to the particle

theory is that, building on work by Watkins (1963), it offers an account of the

development of verb-intial word order. Watkins proposes that at some earlier stage

the non-marked word order in Irish was verb-final. As unmarked verb-intial word

order is not attested in any other IE language family, it can be concluded in line with

reconstruction methodology that unmarked verb-intial order was not a feature of PIE

(see section 3 below). Therefore, it seems that the development of verb-intial order

was a Celtic innovation. Furthermore, if we consider the Continental Celtic evidence,

although fragmentary, it seems to suggest that the unmarked order in these cases was

subject-initial. This can be seen in the examples below from Gaulish (5a) and

Celtiberian (5b). 4

4 Gaulish shows some examples of VSO, such as the Voltino inscription given below: To- med- eclai obalda natina PVB- INF.1SG set.up Obalda daughter ‘(and) Obalda (their) dear daughter, set me up’. VSO seems to be a marked order. See Koch (1985); Eska (1989) for details.


115

5. (a) MARTIALIS DANNOTALI IEVRV VCVETE SOSIN

Martialis.NOM Dannotalos.GEN dedicated Ucetis.DAT this

CELICNON

monument.ACC

‘Martialis (son) of Dannotalos dedicated this edifice to Ucuetis’

(Lejeune 1988: 150)

(b) uTa osCues PousTom-ue Coruinom-ue maCasi[a]m-ue ailam-ue

and SUBJECT OBJECT OBJECT OBJECT OBJECT

amPiTiseTi

VERB5

‘And let him rebuild the cow stable… etc.’ (Russell 1995:283)

From this it seems then, that verb-intial order was not a feature of Proto-Celtic, and so

must be an Insular Celtic innovation.6 However, if we consider evidence from

Archaic Irish texts, it is possible that even Irish itself was not always a verb-intial

language, and that the shift may have taken place in the pre-history of Irish.7

There are two types of non-verb-intial orders in Archaic Irish poetic texts. These

examples are traditionally grouped into two types: Bergin’s Construction and tmesis.

Bergin’s Law (1938) states that “when the verb does not stand at the head of its clause

[i.e. in clause-initial position – GEN], particularly when it follows its subject or

object, it takes the dependent form, that is, a simple verb has the conjunct ending and

a compound verb is prototonic.” (Bergin 1938: 197) An example of Bergin’s

Construction is given in (6) below, where the verb do-ingaib ‘keeps’ has prototonic

form.

5 This gloss is adapted from Russell (1995). At present it is unclear what the exact meaning of each constituent is. 6 This view relies on the assumption that the Celtic languages can be divided in this fashion into Insular and Continental Celtic. It is not compatible with the Gallo-Brittonic hypothesis which assumes that the Brittonic languages are more closely related to Gaulish than to Goidelic. See chapter 1, footnote 4. 7 Of course, Welsh is also a verb-intial language. Although Middle Welsh seems to be V2 (Willis 1998), there is evidence that before this Old Welsh was verb-intial. This could suggest that the development of verb-intial order occurred before Irish and Welsh diverged, at a time of Insular Celtic unity. However, it is also possible that verb-intial order developed independently in both Welsh and Irish if we entertain the possibility that the necessary conditions for this development were in place in both languages at the point when they diverged.


116

6. Mortlithi márlóchet di doínib dingbatar (<do-ingaib)

plagues great-lightnings from people.DAT keep.PRES.PASS.PL.PT

‘Plagues and great lightnings are kept from the people’ (AM §12)

Tmesis is found with compound verbs. The preverb is at the beginning of a sentence,

usually infixing a pronoun, and the verb appears at the end of the clause. This can be

seen in the example in (7) below, where the two parts of the compound verb

ath-cuirethar ‘sends’ appear at opposite ends of the clause.

7. Ath (mór)cathu fri crícha comnámat -cuirethar

PVB (great)battalions to borders neighbour.GEN.PL send.PRES.3PL.DT

‘He dispatches (great) battalions to the borders of hostile neighbours’

(AM §15)

From this evidence Watkins concludes that in early Irish the unmarked word order

was verb-final, and that when the verb was in final position it had conjunct or

prototonic form.8

In addition to the unmarked verb-final orders, Watkins argues that there were

further possible unmarked orders, involving enclitic pronouns. As we saw in chapter

2, Old Irish has enclitic pronouns that are subject not only to Wackernagel’s Law

(1892) but also to Vendryes’ Restriction (1908). This means that not only must they

appear in second position in the clause, but they must also be hosted by a verbal

element, namely a simple verb, a preverb or a conjunct particle. Watkins argues that

this was also true in the prehistoric stages of Irish, so that if a clause contained an

enclitic pronoun, the verb or preverb was forced to move to clause-initial position in

order to host it. In this situation the fronted verb played no pragmatic or discourse

function, and so the construction was unmarked. The possible unmarked orders for

Irish at this stage (which are all attested in Archaic Irish texts) can be seen below:

(V=verb, P=preverb, C=conjunct particle, E=enclitic, #=clause boundary).

8 We will discuss the extent to which Bergin’s Law and tmesis constructions support the view that early Irish was a verb-final language in chapter 5. At present, the crucial point is that early Irish did not have unmarked verb-intial order.


117

8. (a) # … V#

(b) # … PV#

(c) #VE … #

(d) #PE … V#

(e) #CE … V#

Watkins then proposes that Irish underwent a process of univerbation, whereby the

separate parts of the verbal complex came together to form a single unit (we will

return to univerbation in detail in chapter 5). Such a process can be observed in many

other early IE languages; however, in these languages, the univerbated verb appears

in clause-final, not clause-initial position as it does in Irish. Watkins argues that

univerbation in Irish took place towards the beginning of the clause due to

Wackernagel’s Law and Vendryes’ Restriction; the enclitic had to remain in second

position and it had to be hosted by a verbal element. This led to the following

unmarked orders:

9. (a) # … V #

(b) #VE … #

(c) # … PV #

(d) #PEV …#

(e) #CEV … #9

It is easy to see that from this position the step to unmarked VSO order is perfectly

natural. If there are a sufficient number of enclitic pronouns in the language, there

will be more occurrences of unmarked verb-intial word order than unmarked verb-

final word order, and so the verb-final orders will be seen as exceptions, and

eventually verb-intial word order will be generalised.10

9 Both Watkins and McCone refer to this type as univerbation. This is problematic under the conventional view of univerbation as the conjunct particle and simple verb do not form a single semantic unit in the same way as the two parts of a compound verb. 10 A question arises here as to why there should have been so many object pronouns present in the language. In order to have the desired effect on both the phonological form of the verb and the shift in word order these pronouns must have been significantly more frequent than they are in the Old Irish period. Sims-Williams (1984) addresses this issue, arguing that in its prehistory Irish made use of anticipatory object pronouns, similar to those found in French examples such as je les aime les danseuses ‘I like (them) the dancers’.


118

Although Watkins’ univerbation account explains how verb-intial word order came

to be generalised it does not offer a satisfactory explanation of how the different

forms arose. The simple generalisation of verb-intial order as described above would

have resulted in prototonic and conjunct forms appearing in initial position when

there was no enclitic pronoun. This is clearly not what is found in Old Irish. McCone

(1979b) proposes a solution to this problem. McCone argues that there were two

possible ways in which this shift to verb-intial order could take place. Either the

prototonic form could shift to initial position, as discussed above, or a new initial

deuterotonic form could be created by infix deletion on the model of (9e). Learners

would have seen univerbated verbs of the form conjunct particle-enclitic-(preverb)-

verb alongside verbs of the form conjunct particle-(preverb)-verb, where there was no

change in stress pattern and the only difference between them was the presence or

absence of an enclitic pronoun. This would have then formed the model for creating

new initial compound verbs without the infix on the basis of initial preverb-enclitic-

verb forms. The enclitic and its associated mutation could simply be deleted giving

the deuterotonic form. This is shown in the proportion below:

10. #CE’(P)V…# : #C’(P)V…# = #PE’V…# : X

X = #P’V…#

McCone argues that the deletion process was “clearly preferable” as a means of

avoiding the unnecessary complication of a shift in the stress pattern depending on

whether an infix was present or not. Simple verbs then followed suit on analogy with

compound verbs.

As will be clear from the preceding paragraph, McCone’s suffix-/infix-deletion

theory is complex, and as pointed out by Cowill (1985) and Kortlandt (1994), it

requires an improbably large amount of analogy. This is a significant problem for

McCone’s theory. From an acquisitional point of view, it is expecting a lot of the

children acquiring the double system to carry out such complex analogical reasoning.

Moreover, Kortlandt (1994: 61) argues that any theory of the absolute and conjunct

based on analogy is highly implausible. He argues that if there were interaction

between the two sets of forms the most likely result would have been the replacement

of one form by the other, as happened later in Irish.


119

A further difficulty with McCone’s theory arises from the assumption that the

“univerbation” of conjunct particle and verb preceded the univerbation of preverb and

verb (9e). Firstly, as noted in footnote 9, the coming together of conjunct particle and

verb cannot strictly be seen as univerbation. Although in Old Irish the conjunct

particle is unstressed and so forms a single phonological word with the verb, it is not

part of the verb semantically or syntactically. The preverb and verb, on the other

hand, form a semantic unit and it seems likely that this semantic link could have

motivated the univerbation, causing them to become a single syntactic/phonological

unit. This is not the case for the conjunct particle and the verb. If the univerbation of

conjunct particle and verb occurred first then some further explanation is necessary.

We will return to this issue in chapter 5.

The Watkins/McCone model faces an additional problem from a theoretical point

of view. Although at first sight it appears to have an advantage over the particle

theory as it can account not only for the phonological shape of the different forms, but

also their presence in clause-initial position, when considered from within the

framework of generative syntax, the suffix-/infix-deletion theory seems to lose its

appeal.

Watkins argues that the presence of the pronominal clitics in second position

forces the verb to move to initial position, in order to comply with Wackernagel’s

Law and Vendryes’ Restriction. This proposal corresponds neither to cross-linguistic

observations regarding the behaviour of clitics nor to minimalist ideas about

movement. Clitics have received a great deal of attention in the generative literature,

with many different theories being proposed. The major difference between these

theories is whether clitic placement is determined by syntax (Kayne 1975, 1989,

1994, Uriagereka 1995), phonology (Anderson 1992, 1993, 1996) or a mixture of the

two (Halpern 1995, Bošković 2000). Although there are many different accounts, all

the proposed theories seem to agree in one respect – clitics move to attach to a

suitable prosodic host, they do not attract specific elements to provide themselves

with a host.11 In chapter 1, section 2.2 it was argued that in minimalist syntax

movement can only be driven by abstract features associated with functional heads,

namely EPP- and Affix-features. Within this theoretical framework, clitics can

undergo movement but they cannot cause movement of other elements. So, it seems 11 See also the discussion of Zwicky & Pullum’s (1983) criteria for distinguishing clitcs from affixes in section 4 below.


120

that the suffix-/infix-deletion model is incompatible with minimalist ideas about

syntax; therefore, for the purposes of this dissertation it cannot be maintained.12

2.4 Summary

In this section we have examined the two main existing accounts of the origins of the

Old Irish double system of verbal inflection. As the primary focus of this dissertation

is syntactic, the phonological debate surrounding the origins of the double system has

been put to one side and the two competing theories have been compared on

conceptual and comparative syntactic grounds. Although from a syntactic point of

view, McCone’s theory appears to be at an advantage, as it offers an account for the

distribution of the different verbal forms and the shift to unmarked verb-intial order,

this theory is not consistent with current syntactic assumptions. However, this is not

the primary argument against McCone’s theory. The excessive complexity and

implausible amounts of analogy that McCone invokes make it equally unattractive

from a pretheoretical, conceptual point of view. The particle theory has the advantage

here in that it can derive the attested forms without such analogy. After the particle

has had the necessary effect on the verbal forms it is simply lost through the process

of apocope that affected the language as a whole in the fifth century.

It was argued above that if we accept Schrijver’s etymology for the particle *es the

only remaining conceptual objection against the particle theory is that it cannot

explain the shift to unmarked verb-intial word order. The remainder of this chapter

aims to show that this objection can be overcome if the particle theory is combined

with minimalist ideas about syntactic change.

3. ASPECTS OF THE SYNTAX OF PROTO-INDO-EUROPEAN

3.1 Introduction

The aim of this chapter as a whole is to provide an account within the minimalist

framework of how the Old Irish double system of verbal inflection developed from

PIE. This section aims to provide the starting point for this account, namely an outline

of the syntax of PIE. Syntactic reconstruction is not without its problems, and we can,

12 Under this view, Vendryes’ Restriction did not cause the shift to verb-intial word order but was a consequence of it. Enclitics appear with elements of the verbal complex because the verb appears in initial position. The situation can be compared to object clitics in Romance. These appear with the verb, but this is due to independent movement of the verb and the clitic to the same functional position, namely T (Kayne 1989, 1991).


121

of course, never expect to gain a full picture of the syntax of PIE. However, following

Mark Hale (1996b), we shall take the view that it should be possible to reconstruct

certain features of PIE syntax. The reconstruction proposed in this section follows that

developed by Hale (1987, 1995), updating it into minimalist terms. The

reconstructions presented in this section are primarily based on points of similarity

between the oldest and most widely studied IE languages, i.e. Hittite, Sanskrit,

Ancient Greek and Latin, without reference to the directionality of syntactic change,

or one type of change being more natural than another. We will examine three

features of the syntax of PIE. Section 3.2 examines the structure of the left periphery

of the clause. Section 3.3 looks at word order patterns. Section 3.4 considers preverbs

and how they should be analysed within generative syntax. Section 3.5 draws together

the main conclusions from this chapter and provides a summary of the main features

being proposed for the syntax of PIE.

3.2 The structure of the left periphery

Before we can begin to consider the structure of the left periphery in PIE we must

first establish what is meant by this term. The left periphery is the left-most edge of

the clause. In generative terms this essentially means the highest functional

projection, the CP, which encodes discourse-related information. There seems to be

some degree of cross-linguistic variation as to the structure of the CP. In languages

such as English the CP seems to consist simply of a single functional position.13 The

C position may contain a complementizer, marking whether the clause is finite (11a)

or non-finite (11b). The C position may also mark illocutionary force through either

movement in a wh-question (12a) or merge of an appropriate complementizer (12b).

11. (a) John saw [CP Cthat [TP Mary [VP left]]]

(b) John asked [CP for [TP Mary [VP to leave]]]

12. (a) [CPWho Cdid [TPJohn [VP see yesterday?]]]

(b) I asked [CP Cif [TPJohn [VP saw me]]]

Crucially, in languages such as English all of these functions target the same syntactic

domain, a single CP projection. In languages such as Italian, however, this is not the

13 This is disputed. See, for example, Chomsky (1977), Haegeman (2000).


122

case. Elements that have different functions co-occur and appear in different positions

at the left edge of the clause. This led Rizzi (1997) to suggest that CP can be a cover

term for a number of independent, specialised functional projections. For example, in

Italian, it is possible to topicalise or focus a constituent by moving it to the left

periphery, as can be seen in (13).

13. Credo che a Gianni, QUESTO, domani, gli dovremmo dire

Topic Focus Topic TP

I believe that to Gianni, THIS, tomorrow we should say

(Rizzi 1997: 295)

Crucially, as shown in the example in (13) the topicalised and focussed elements can

co-occur, demonstrating that they must occupy different syntactic positions, a(t least

one) topic phrase (TopP) and a focus phrase (FocP). Italian also provides evidence for

two more functional projections. It was argued above that, in English, markers of

finiteness and force target the same generalised C position; in Italian this is not the

case. In the examples in (14) below it can be seen that the complementizers che and di

occupy different syntactic positions: che appears before the topic (14a) and di appears

after it (14b).

14. (a) Credo che, il tuo libro, loro lo apprezzerebbero molto

Force Topic IP

I believe that your book they it would appreciate a lot

(Rizzi 1997: 288)

(b) Credo, il tuo libro, di apprezzarlo molto

Topic Fin IP

I believe, your book, to appreciate-it a lot

(Rizzi 1997: 288)

Rizzi argues that che marks illocutionary force, whereas di marks finiteness, and so

two extra projections can be added to the CP, a force phrase (ForceP) and a finiteness

phrase (FinP). From this Rizzi proposes that the articulated CP with the structure

given below:


123

15. ForceP

Force TopP

Top FocP

Foc TopP

Top FinP

Fin TP

Both English-type and Italian-type languages have the same expressive power. In

minimalist terms, the same features are associated with the CP in both languages;

however, whereas in English these features are all contained within one generic C

position, in Italian they are spread among a number of individual functional

projections.

Let us turn now to the earliest IE languages. Kiparsky (1995a) argues that, as the

earliest IE languages have no complementizers, we cannot reconstruct

complementizers for PIE and so PIE cannot have had a CP. The first part of

Kiparsky’s argument seems to be valid. The earliest IE languages did not have

generic complementizers, equivalent to English that.14 These complementizers

developed later in each language, often from a 3sg neuter relative pronoun, for

example, Latin quod, Sanskrit yád, Hittite kuit and Ancient Greek hóti. However, it

cannot be concluded from this that PIE did not have a CP. As discussed above, the CP

is not just the phrase projected by a complementizer; it plays a number of different

roles. Only if none of the CP functions can be reconstructed for PIE can we conclude

that PIE had no CP.15 As will be shown in the remainder of this section, this does not

seem to be the case.

14 Although it seems to be the case that there are no that-type complementizers in the earliest IE dialects, there are various clause-typing particles that would be good candidates for appearing in the C head position (see, for example, Arad & Roussou 1997 on particles in Ancient Greek). Further research on the syntactic behaviour of these particles is necessary before a firm conclusion can be reached. 15 The fact that generic complementizers cannot be reconstructed for PIE could be taken as evidence to suggest that PIE had an articulated rather than a generic CP. However, the relationship between that-


124

Let us begin by considering the evidence from Ancient Greek. Dik (1995) argues

that Ancient Greek is a discourse configurational language where word order is

determined by pragmatics rather than syntax. In the examples in (16) below, the word

order varies in terms of grammatical relations; however, the ordering Topic-Focus-

Verb remains the same.

16. (a) Topic Focus Verb.

strouthòn dè oudeìs élaben

ostrich.ACC PTC nobody.NOM caught

‘An ostrich, NOBODY caught’

(Xenophon’s. Anabasis 1.5.3; Matić 2003: 574)

(b) Topic Focus Verb.

ho dè Kléandros oudéna epeprákei

the PTC Cleander.NOM nobody.ACC had sold

‘As for Cleander, he had not sold one of them’

(Xenophon’s. Anabasis 7.2.6; Matić 2003:575)

Matić (2003) argues that Dik’s Top-Foc-V structure, although essentially correct,

does not provide enough syntactic or pragmatic positions to account for the possible

word-order variations found in Ancient Greek. In the example given in (17) below

Kûros ‘Cyrus’ is the topic, and tônde heíneka ‘because of these’ is the focus, but what

is the status of all the elements that intervene?

complementizers and the articulated CP is not so straightforward. Italian, for example, has a generic that-complementizer, che, but no generic C position.


125

17. Topic

Kûros mèn toûton tòn lógon toîsi Íōsi kai toîsi Aioleûsi

Cyrus PTC this the story to.the Ionians and to.the Aeolians

Focus Verb

tônde heíneka élekse, hóti dè Íōnes próteron… ouk epeíthonto…

these because.of said that PTC Ionians before not obeyed

‘Cyrus told this story to the Ionians and the Aeolians for the reason that

the Ionians… had before disobeyed him…’

(Herodotus 1.141.3/Matić 2003: 600)

Matić argues that the intervening elements are all topics. This means that not only

does Ancient Greek seem to have clearly defined positions for topic and focus, it also

has the possibility of multiple topics, on either side of the focus position.16 17

Fortson (2004) provides evidence to suggest that the left periphery in Latin is

similar to that found in Ancient Greek. Latin also seems to have a number of possible

fronting positions in the left periphery. Fortson argues that in the example in (18)

below, si ‘if’ occupies the C position and a large number of constituents can precede

it.18

18. Perfidia et peculatus ex urbe et auaritia si exultant

Betrayal and embezzlement from city and greed if are-exiled

‘If betrayal and embezzlement and greed are exiled from the city’

(Plautus, Persa 555; Fortson 2004: 145)

The evidence from Ancient Greek and Latin, then suggests an articulated CP similar

to that proposed by Rizzi for Italian (cf. the tree in (15) above).

16 Matić further categorises these topic positions into more specific pragmatic roles, e.g. exclusive contrast topic (non-recursive), frame-setting topic (recursive) and continuous topic (recursive) and develops a highly detailed pragmatic structure for the clause. 17 Interestingly, Rizzi (1997) also proposes this for Italian. He argues that there is a recursive topic position either side of the focus position. 18 Of course, this presupposes that Latin has developed complementizers.


126

19.

TopP

Spec TopP

Top FocP

Spec FocP

Foc TP

If we consider the evidence from Vedic Sanskrit, a somewhat different picture

emerges. Hale (1987, 1995, 1996) proposes that there are two positions at the left

edge of the clause in Vedic.19 The inner position hosts wh-words (both relative

pronouns and interrogatives) and the outer position hosts what Hale described as

‘topicalised’ elements. Crucially, wh-elements and ‘topics’ can co-occur as shown in

(20) below, where µpnānaṃ tīrtháṃ ‘attained course’ appears to the left of the wh-

word ká.

20. µpnānaṃ tīrtháṃ ká ihá prá vocad

attained.ACC.SG course.ACC.SG WH.NOM.SG here forth speak

‘Who can proclaim here the attained course?’

(RV 10.114.7c/Hale 1996a: 170)

Hale argues that, as the inner position hosts wh-elements it must be a CP, and the

outer edge, containing topics, must be a TopP. This gives the basic clause structure

for Vedic as shown in (21) below.

19 See Hock (1996) for an alternative view of the left periphery in Vedic.


127

21. TopP

Spec TopP

Top CP

Spec C

C TP

The fact that Vedic appears to have two distinct positions in the left periphery

suggests that it does not have a generic CP, i.e. it patterns more with Italian than with

English. Hale’s CP only hosts wh-phrases, so it seems more likely that it is

specialized functional projection containing simply a wh-feature.20 The function of

Hale’s TopP seems somewhat less clear. This appears to be because, in Sanskrit, the

outer position can be used for both topicalisation and focussing.21 In (22a) (repeated

from (20) above) the fronted constituent is a topic. It is not emphatic. In (22b),

however, the fronted constituent seems to be fronted for contrastive focus.

22. (a) µpnānaṃ tīrtháṃ ká ihá prá vocad

attained.ACC.SG course.ACC.SG WH.NOM.SG here forth speak

‘Who can proclaim here the attained course?’

(RV 10.114.7c/Hale 1996a: 170)

(b) Diví vái sóma µsīd, átha ihá devµḥ

In heaven indeed Soma was but here gods

‘In heaven was Soma, but here the gods’

(MacDonell 1916: 284)

20 This could be construed as a ForceP, as wh-movement typically marks interrogative force. However, if the inner position is a Force position we might expect it to be the landing site of movement that marks other types of illocutionary force, such as imperatives, or the merge position of sentential question particles. Further research is necessary to see whether this is the case. For our purposes it will be assumed, following Hale, that this position can only host wh-elements. 21 This resembles the situation in verb-second (V2) languages such as Modern German, where topicalisation and focussing both target the specifier of a general CP.


128

It seems that Hale’s label is inadequate. This outer position in Sanskrit is not a TopP

but a less specific pragmatic fronting position that can host both topics and focus

expressions.

If we examine the evidence from Anatolian, it seems that Hittite patterns with

Vedic. In Hittite a fronted wh-word can be preceded by one constituent in the syntax,

as shown in (23) below.

23. ŠEŠ-tar kuiš kuedani hatreškizzi

brotherhood who to.another writes

‘Who always writes about brotherhood to another?’

(Garrett 1990: 35)

The pragmatic function of the outer position is not entirely clear, and further research

is necessary before anything conclusive can be said on the left periphery in Hittite

(although see Garrett (1994) for some discussion of how the Hittite system developed

into Lycian). At this stage, however, it seems that the Hittite system resembles Vedic

more closely than Greek and Latin.22

Having considered the left periphery in Ancient Greek, Latin, Vedic and Hittite,

we are now faced with the question of how we should reconstruct the left periphery

for PIE. This is not straightforward. Let us assume that either the structure provided

for Ancient Greek or that provided for Vedic should be reconstructed for PIE. If the

first of these scenarios is true, and PIE had separate Topic and Focus projections, then

Vedic and Hittite must have innovated. The inherited focus position became restricted

in function, and so could only host wh-elements. Wh-movement is often seen as a

subtype of focussing (Kiparsky 1995a, Rizzi 1997, Sabel 2000), therefore it is easy to

see how a reanalysis may have taken place, which changed the feature content on the

inner C head from [+focus] to [+wh].23 After this reanalysis the inner position could

no longer be used for focussing, and so this function was assigned to the outer

position, which became a more generalised functional position, capable of hosting

both topics and focus expressions. 22 Interestingly, the Anatolian languages also show an additional possibility for pragmatic fronting, namely left dislocation. See Garrett (1994) for a discussion of the relationship between this operation and the development of verb-intial word order in Lycian. 23 This restriction of a focus position to a wh-position also seems to have happened at a somewhat later stage in Greek. Roberts & Roussou (2003: 161f) argue that between Ancient and Modern Greek focussed indefinites became reanalysed as wh-words.


129

Let us consider now the alternative possibility, i.e. that Ancient Greek and Latin

systems developed from a Vedic-style left periphery. For this to be the case, the inner

position would have become less specific in terms of feature structure, changing from

[+wh] to [+focus]. Such a change is plausible, but less natural than a change in the

opposite direction. Furthermore, Ancient Greek and Latin seem to allow multiple

topics, whereas Vedic and Hittite strictly allow only two constituents to appear in the

left periphery. It seems more likely perhaps for recursive topicalisation to become

restricted to a single position than for a single topicalisation position to become

recursive, however again, this is by no means conclusive.

In sum, it seems that all four early IE languages show evidence for multiple

positions in the left periphery that are associated with specific pragmatic functions.

This suggests that we should reconstruct an articulated CP for PIE, consisting of a

number of different functional projections, rather than a single generalised CP, as

found in English. Such a conclusion is perhaps further supported by the fact that in

their earliest stages the early IE languages did not have generalised that-type

complementizers. At present no firm conclusions can be drawn about the exact

functions of these positions, and so the remainder of this dissertation will assume a

modified version of Hale’s conception of the left periphery. It will be assumed that

PIE had an articulated CP, containing two functional projections, a Top/FocP that was

the target for pragmatic fronting processes such as topicalisation and focussing and a

CwhP which is the target for wh-movement. This is shown in the diagram below:

24. Top/FocP

Top/Foc CwhP

Cwh TP

3.3 Word order

Delbrück (1893) proposed that the unmarked word order of PIE was verb-final. This

conclusion was strengthened by the discovery of Hittite (Watkins 1976), and today it

is widely accepted among Indo-Europeanists. Hale (1987, 1995) bases his

reconstruction within a theory of syntax whereby SOV word order is accounted for by


130

the head parameter; the subject and object are merged into the derivation in the order

subject-object-verb. This can be seen in the tree in (25a). Since Kayne (1994), the

head parameter has been largely abandoned in generative work (although see

Richards 2004 who attempts to resurrect the head parameter as a PF condition on

word order; see also Saito & Fukui 1998 and Fox & Pesetsky 2004). Under a

Kaynean analysis the verb and the object are always merged in the order verb-object.

SOV languages must be derived via movement of the object or a projection

containing the object into a higher projection (usually the specifier of vP) above the

position of the verb. This is shown in the tree in (25b).

25. (a) VP (b) vP

Subject V Subject vP

Object Verb Object v

v VP

Verb tObject

Hale’s analysis can be adapted easily to bring it into line with this theoretical

development. All that is necessary is to postulate that v in PIE had an EPP-feature

which causes the object to move to Spec-vP.24

More problematic than verb-final orders are the verb-medial clauses that are

attested in the early IE dialects, where some element appears after the verb, as in the

example (26a) from Vedic, and (26b) from Homeric Greek:

24 Although this does not account for other VP-elements that can occur preverbally, such as PPs or dative objects. One way to account for this is for the entire VP to raise to Spec-vP following movement of the verb to v. This ensures that the verb follows the entire contents of VP. See Biberauer (2003) for such a proposal.


131

26. (a) tam eva tābhir āhutibhiḥ śamayitvorjaṃ lokānāṃ

him.ACC thus libations power having.appeased.ACC worlds.GEN

jayati yamaṃ devaṃ

wins Yama.ACC God.ACC

‘Having appeased him, the God Yama, with these libations, he wins the

power of the worlds’

(JB 7, 3–4; McCone 1997b: 370)

(b) Trôas d' eklinan Danaoi

Trojans.ACC PTC turned Danaans.NOM

‘And the Danaans turned the Trojans’

(Il 5, 37)

Examples such as (26a) where the object follows the verb can be easily explained

under a Kaynean theory of phrase structure if we assume that the object does not raise

to Spec-vP in these cases, but remains in situ as the complement of V. For this to be

the case the EPP-feature present on v must be optional. Within the minimalist

programme, such optionality is permitted if it has an effect on interpretation

(Chomsky 2001: 34). If these deviations from SOV order are pragmatically motivated

then it will be possible to explain them in this way. A similar scenario is invoked to

account for North Germanic object shift (Richards 2004) and the apparent optionality

of object raising during the shift from OV to VO word order in Middle English

(Biberauer & Roberts 2005).

Subject-final clauses, such as that in (26b) are more problematic. The subject is

merged in Spec-vP and so will not appear to the right of the verb at any stage of the

derivation. One possibility to account for those cases where the verb precedes the

subject (or the object if object movement is obligatory) is to suggest that the verb

moves to T. V-to-T movement is common cross-linguistically (see Pollock 1989 for

French, Roberts 1985, 1993 for Middle English, McCloskey 1996a for Modern Irish).

There is no theoretical objection to postulating V-to-T movement. However, before it

can be reconstructed for PIE, there must be sufficient evidence for V-to-T movement

in the early IE languages. Hale (1995: 166–186) argues that V-to-T movement occurs

quite frequently in Vedic. In the example in (27) below, the verb precedes the object


132

and an adverb and a preverb (see section 3.4 below), which Hale takes to mark the left

edge of the VP, suggesting that the verb has raised.

27. diváṣ caranti pári sadyó ántān

heaven.GEN.SG wander PVB in-one-day end.ACC.PL

‘They wandered around the ends of heaven in one day’

(RV 5.47.4d/ Hale 1995:186)

The major problem with postulating V-to-T movement in Vedic is that it must be

optional, as, in clauses with SOV word order, where the verb appears to the right of

the subject and the object, it cannot have raised to T. If V-to-T is optional, we would

expect its presence or absence to have an effect on interpretation. V-to-T movement,

however, is not generally associated with pragmatic effects cross-linguistically (see

Chomsky 2001); therefore the postulation of optional V-to-T movement seems

unlikely.25

If we turn to the other early dialects, it seems that there is little evidence to support

the reconstruction of V-to-T movement. Although it is often argued that the modern

Romance languages and Modern Greek have V-to-T movement (see among others

Pollock 1989 for French, Belletti 1991 for Italian and Alexiadou & Anagnostopoulou

1998 for Greek) there is no clear evidence that V-to-T was a feature of Latin or

Ancient Greek. Hittite is more rigidly verb-final than any of the other early dialects

(Luraghi 1990) and so does not provide us with any evidence for the reconstruction of

V-to-T movement. It seems that, although from a theoretical perspective V-to-T

movement is possible and perhaps desirable as a feature of PIE, this is not backed up

by the evidence from the daughter languages.

If V-to-T movement cannot be reconstructed for PIE, then there are two further

possibilities for accounting for verb-subject orders; either the subject could have been

right-dislocated or the verb could have raised to a higher functional projection within

the articulated CP. Let us consider each of these options in turn. When a constituent is

right-dislocated its grammatical role tends to be taken up earlier in the clause, either

25 The possibility cannot be entirely disregarded, however, as there is evidence to suggest that Icelandic and Faroese show variation in terms of V-to-T movement. See Hróarsdóttir, Hrafnbjargarson, Wiklund & Bentzen (to appear) on Icelandic; Thráinsson (2003) on Faroese.


133

by inflections on the verb or by a pronoun. This often seems to be the case for post-

verbal subjects and objects in the early IE languages (McCone 1997b: 370)

It has also been observed that placing an element at the right edge of the clause has

a similar pragmatic effect to placing it at the left edge. The right edge of the clause

seems to provide an additional Focus position (McCone 1997b: 372, Matić 2003:

615–9). McCone also observes that a displaced subject or object is often long or

heavy. This clearly resembles the well-documented process of heavy-NP shift (Rizzi

1990).26

Let us turn now to the second possibility, namely movement of the verb into the

left periphery. It was argued above that cross-linguistically V-to-T movement does

not have pragmatic effects. However, this does not mean that verb-movement cannot

be pragmatically motivated. In section 3.2 above it was argued that movement into the

left periphery of the clause is linked to pragmatic effects such as topicalisation and

focussing. Although in the examples given above the fronted constituent is a nominal

expression, this is not always the case. Verbs can also be fronted in the earliest IE

languages, as shown in the examples in (28) below.

28. (a) átārisur bharatµ gavyávaḥ sám

cross Bharata cow-seeking together

‘The cow-seeking Bharatas have crossed over’

(RV 3.33.12a; Hale 1995:193)

(b) êe kholon pauseien erêtuseie te thumon

or wrath check curb and spirit.ACC

‘or he should check his wrath and curb his spirit’

(Il 1,192; McCone 1979a: 251)

It is possible, then, that examples where the verb precedes the subject or object could

involve pragmatic fronting of the verb to the left periphery. This could result in verb-

intial order, as in (28) or, if some other constituent is fronted to an additional

functional projection within the CP to the left of the verb, to V2, as in (26b) or even

V3 or V4. Let us examine this process of verb fronting in more detail.

26 Right-dislocation and heavy NP shift are contentious issues in current syntactic theory, which we shall not go into here. See Beerman, LeBlanc & van Riemsdijk (1997) for discussion.


134

Watkins (1963) observes that fronting a verb is functionally equivalent to fronting

any other constituent.

“For in such other Indo-European languages, where the normal position of the

verb is sentence final [i.e. not Old Irish – GEN], any element, subject, object,

prepositional phrase, etc. may be placed in the initial position for stylistic

emphasis; the placing of the finite verb itself in this initial position is simply

another case of the same emphasis.” (Watkins 1963: 5)

If verb fronting and DP fronting have the same pragmatic function, we would expect

both types of fronting to be a result of the same process and target the same position

in the left periphery. Topicalisation and focussing tend to involve the movement of an

entire phrase to a specifier position (Rizzi 1997). As a result, we might expect

pragmatically motivated verb fronting also to involve phrasal-movement. In some

cases, there is evidence to suggest that this is the case. In the examples given in (29)

below, both the verb and its complement appear in the outer position in the left

periphery. If, as Hale (1987, 1995) suggests, Vedic only allows two constituents in the

left periphery, then the verb and object cannot have been fronted separately. They

must have moved together to target the position to the left of the position filled by the

wh-element. This suggests that the whole vP has raised.

29. (a) gávāṃ gotrám udásṛjo yád añgiraḥ

cow.GEN.PL herd.ACC.SG up-released when Angiras.VOC.SG

‘When you released the herd of cows, O Angiras’

(RV 2.23.18b; Hale 1995: 194)

(b) hárī índrasya ní cikāya káḥ svit

steed.ACC.DL indra.GEN.SG perceived who PRT

‘Who has perceived the two steeds of Indra’

(RV 10.114.9d; Hale 1987:42)

Under a Kaynean analysis, as adopted here, in order for the object to precede the verb

the object must have moved to Spec-vP, therefore, it must be the vP rather than the

VP that has fronted to the left periphery. However, this poses the problem of why the


135

subject does not appear in the clause-initial position. In (29a) the subject is expressed

through the inflections on the verb, and in (29b) the subject is the wh-element, which

will have moved out of the vP to the left periphery before the vP is moved. However,

this will not always be the case. Furthermore, there are many examples where the

verb appears in initial position independently of the object. In order to maintain the vP

fronting analysis in this situation the object and the subject must have both moved out

of the vP prior to movement of the vP to the left periphery. It seems that the position

of nominals in the early IE languages is quite free (see Dover 1960 for Ancient Greek,

Luraghi 1990 for Hittite). Therefore it could be argued that the clause structure of PIE

involves a large amount of scrambling, with the object and subject being free to move

out of the vP. However, even if it is possible to explain movement of the subject and

direct object out of the vP in this way, it is not so straightforward for adverbs and

prepositional complements, which do not tend to move.

An alternative analysis of verb fronting would be as head-movement to a

functional projection within the left periphery. There is evidence to suggest that

pragmatic effects are linked to particular projections rather than specific syntactic

positions. For example, in English an embedded clause can be marked as interrogative

either through raising a wh-phrase to Spec-CP (30a) or merging if in the C head

position (30b):

30. (a) I asked [CP which film C ø [TP you Twere [VP watching twhich film]]].

(b) I asked [CPø Cif [TP you Twere [VP watching ‘Titanic’]]].

By extension, then, it is possible that an element could be marked as a topic or a focus

simply by moving as a head to the relevant topic/focus projection. The postulation of

such pragmatically motivated head-movement raises several theoretical questions.

The first of these relates to the question of how movement is motivated in the case of

topicalisation and focussing. Until now in this dissertation it has been argued that

movement is motivated by the presence of an EPP-feature on the functional head in

question, and that a distinction can be made between EPP-features, which target

phrasal categories and Affix-features, which target heads. If topicalisation can involve

either phrasal-movement or head-movement then we are faced with a problem, as the

functional head in question must have an EPP-feature in some cases and an Affix-

feature in others. In recent work Chomsky (2005, 2006) argues that movement to


136

phase heads is motivated by an Edge Feature (EF). In principle, there is no reason

why EF may not attract either a head or phrase.27 The only necessary condition for the

EF to be checked is that some element be merged at the edge of the phase head

bearing it; this element could be either a head or a phrase. In cases of topicalisation,

then, the EF associated with the C position targets an element from within the domain

of C, and raises it so that it is re-merged at the edge of the C-projection. Introducing

the concept of EF can perhaps provide an account of movement operations that target

both phrasal categories and heads. However, it is not entirely clear what the

relationship between EF and EPP and Affix-features should be. For our purposes it

will be assumed that EF is essentially another type of movement diacritic, comparable

to an EPP- or Affix-feature. Whereas EPP- and Affix-features are specified to attract

either XPs or heads respectively, EF can attract both categories.28

In this section it has been argued, following general consensus among Indo-

Europeanists, that PIE should be reconstructed as having underlying SOV order,

derived through movement of the object to Spec-vP. However, evidence from the

daughter languages suggests that deviations from unmarked SOV order were possible.

It seems likely that these deviations had stylistic or pragmatic effects, and so could

result from a number of different processes, such as optionality with regards to object-

raising, right-dislocation, heavy-NP-shift and pragmatically motivated movement

(topicalisation or focussing) of the verb to the left periphery. This pragmatic

movement of the verb to the left periphery could result either in verb-medial orders,

or, crucially for the argument presented below, verb-intial. Furthermore, it has been

argued that it is possible that fronting of the verb to initial or near-initial position

could have involved either vP movement or movement of just the verb. Crucially,

however, in the early IE languages, and by extension PIE, this movement process was

pragmatically conditioned and therefore optional.

3.4 Preverbs

Preverbs are particles, etymologically related to prepositions, that can appear before a

verb and modify its meaning. Both preverbs and prepositions were most likely

independent adverbs in PIE, a situation that is reflected in the Anatolian languages 27 Apart from the fact that Chomsky does not consider head-movement to play a role in narrow syntax. See chapter 1, section 2.2.2 for arguments against this view. 28 The relation between EF and EPP is by no means clear in Chomsky (2005, 2006). If EF is the only feature that can motivate movement then EPP and Affix could perhaps be seen as subtypes of EF.


137

(see Boley 1985 for details of so-called Anatolian ‘place words’). Typically adverbs

are adjoined to the phrase they modify. Therefore, we would expect the adverbs that

became preverbs to have originated in Spec-VP or Spec-vP. Hale (1995: 192) argues

that this is the position they occupy in Vedic; however, Hale notes that preverbs may

also appear at the beginning of the clause, modifying the whole clause. In this case,

Hale argues, the preverb is in Spec-CP as it appears between a topicalised constituent

in TopP and wh-word in CP, as shown in the example (31) below.

31. ābhogáyam prá yád ichánta aítanµ

nourishment.ACC.SG PVB when seeking.NOM.PL went.2SG

‘When seeking nourishment you went forth’

(RV 1.110.2a; Hale 1996a: 185)

However, this does not seem to be the only possible position for clause-initial

preverbs in Vedic. McCone (1997b), following Renou (1933), argues that fronting the

preverb of a compound verb plays the same functional role as fronting the verb if the

verb is simple. In example (32) below the verb in all three clauses is imperative, and

therefore we would expect it to appear at the beginning of the clause. However, the

verb only appears in clause-initial position in the final clause. In the first clause the

preverb is fronted in place of the verb and in the second clause the negative particle

appears first.

32. ā no yajñe bhajata, mā no yajñād antárgata,

PVB us sacrifice share.IMPV.2PL NEG us sacrifice exclude. IMPV.2PL

astv evá no ‘pi yajñe bhāga iti

be.IMPV.3SG thus us PTC sacrifice share PTC

‘Give us a share in the sacrifice, do not exclude us from the sacrifice, let

there be a share in the sacrifice for us.’

(McCone 1997b: 372)


138

If fronting the preverb has the same pragmatic force as fronting the verb, then we

would expect a fronted preverb to target the same projection as a fronted verb, i.e. the

outer left-peripheral position, Hale’s TopP.29

McCone (1979a) provides evidence to suggest that this is also the case in Ancient

Greek:30

33. hôs ephat' Atreïdês, epi d' êineon alloi Akhaioi

so spoke son-of-Atreus PVB PRT approved other Acheans

‘So spoke the son of Atreus, and all the Acheans shouted assent’

(Il 3.461; McCone 1979a: 253)

Garrett (1992) observes that in Lycian a verb is only fronted when it has no associated

preverb; fronted preverbs and fronted verbs are in complementary distribution. This

also seems to be the case in Hittite, which lacks clear cases of verb fronting that

strands a preverb.

The evidence from Vedic, Ancient Greek and Anatolian suggests that the

possibility of fronting the preverb in place of a verb to the outer peripheral position,

the Top/FocP, should be reconstructed for PIE.

3.5 Summary

In this section we have explored the possibility of reconstructing some features of the

syntax of PIE that will be relevant to the analysis of Old Irish that is to follow. The

most important features of our proposed reconstruction are summarised below.

1) PIE did not have generalised that-complementizers

2) PIE had an articulated CP consisting of (at least) two specialized C heads,

which each played a specific pragmatic function. Following Hale, we will

reconstruct an outer position that hosts topicalised and focussed elements

(Top/FocP) and an inner position that hosts wh-words (CwhP).

29 Although both preverbs and verbs target the same projection there is no necessity for them to target the same position. Whereas verbs target the head position, preverbs, as adverbial phrases, are more likely to target the specifier. 30 The Latin evidence is of little use here as in Latin preverbs have for the most part been fully univerbated and appear as verbal prefixes.


139

3) PIE had unmarked verb-final order; deviations from SOV order are

pragmatically marked and are the result of a variety of processes such as

optional object raising to Spec-vP, right-dislocation, heavy NP shift and verb-

movement to the left periphery.

4) verb-intial order in PIE was optional and stylistically marked. Verb fronting

in PIE targets the Top/FocP of the left periphery, and may involve fronting the

entire vP or simply the verb itself. Either way, this movement is pragmatically

motivated and optional.

5) If a verb is compound, the initial preverb can be fronted to Top/FocP in place

of the verb to achieve the same stylistic effect.

Having established the relevant features of PIE, let us now move onto the topic in

hand – the Irish verbal system.

4 FROM PROTO-INDO-EUROPEAN TO PRE-OLD IRISH: CHANGES IN THE C-SYSTEM

4.1 Introduction

At the end of chapter 2 it was concluded that although there is good evidence to

suggest that during the Old Irish period the verb only moves as far as T, this was not

always the case. Although suffixed pronouns and special relative verbal endings are

not used productively in Classical Old Irish, it seems likely that there was a stage

before the Old Irish period when they were. At this time, then, we can assume that

Irish had V-to-C movement. The remainder of this chapter will focus on this pre-Old

Irish stage and how it developed from PIE. This section considers the development of

verb-initial word order, section 5 focuses on the development of initial preverbs and

section 6 looks at the development of conjunct particles.31

Before we go on to consider the development from PIE to pre-Old Irish, let us

make clear our assumptions regarding the verbal system of the pre-Old Irish period.

In many ways the verbal system that is being assumed for this period is similar to that

proposed by Carnie, Harley & Pyatt (2000 – CHP) for Classical Old Irish. It is

31 The term ‘pre-Old Irish’ is deliberately vague. Clearly the period to which it refers is before our first written records. It seems quite plausible that the changes described in this chapter affected Insular Celtic as a whole as the Brittonic languages also show verb-intial word order and traces of a double system of verbal inflection. However, a more thorough analysis of the Welsh evidence is necessary for this to be confirmed. For the remainder of this chapter then, we will continue to use the term ‘pre-Old Irish’ in this vague sense, to refer to a period of Irish before the beginnings of the written record and after Common Celtic.


140

assumed that pre-Old Irish has a filled C condition, i.e. C has an Affix-feature that

attracts some verbal element, i.e. the verb or an initial preverb, when it is not filled by

a conjunct particle. However, crucially, unlike CHP, it is assumed that during pre-Old

Irish there is no V-to-T movement. When C is filled, the verb and any associated

preverbs remain in their base position further down the clause. In other words, pre-

Old Irish is the pre-univerbation stage, giving the orders shown below (where

V=verb, P=preverb, C= conjunct particle):

34. (a) #V... #

(b) #P...V#

(c) #C...(P)V#

As we saw in chapter 2, the derivation of compound verbs through syntactic

movement to both C and T faces significant syntactic problems. For this reason it will

be assumed that at no stage in the history of Irish did these two movement operations

co-occur. At the stage when the verb and preverb move syntactically to C, movement

to T is prohibited. Movement to T develops only through the reanalysis of V-to-C

movement. We will return to this topic in detail in chapter 5.

This section focuses on the development of V-to-C movement. Section 4.2

explores the nature of and motivation behind the syntactic change that resulted in the

development of V-to-C movement, namely the reanalysis of stylistically marked verb-

intial order as grammatically motivated verb-intial order. Before this reanalysis could

take place, however, a number of unrelated changes must have occurred,

predominantly in the C-system. These changes are the topic of sections 4.3, 4.4 and

4.5. Section 4.6 summarises and concludes the section.

4.2 The development of V-to-C movement: a syntactic reanalysis

In section 3.3 above it was argued that the unmarked word order pattern in PIE was

verb-final. However, verb-intial word order seems to have been a possibility in the

earliest IE languages as a stylistically marked variant. verb-intial order in the early IE

languages, and by extension, PIE seems to involve either vP or verb-movement to a

topicalisation or focus position in the left periphery, part of an articulated CP.

Following Hale (1987, 1995), it was argued that PIE had two positions in the left

periphery. The outer position hosted pragmatically fronted (i.e. topicalised or


141

focussed) elements, this is a Top/FocP. The inner position hosted interrogative and

relative wh-words, a CwhP. As fronting of the verb is a subtype of topicalisation or

focussing, rather than wh-movement, we can assume that it targets the outer of these

positions. This is shown schematically in the tree in (35) below.

35. Top/FocP

Top/Foc CwhP

vP/V

Cwh TP

(wh)

(Subj…Obj)

The situation we are assuming for pre-Old Irish is somewhat different. First, there is

no evidence in Old Irish for an articulated CP (although see Roberts 2005, Adger to

appear, who argue otherwise). Old Irish does not allow fronting for emphasis or in

wh-questions (see section 6 below). The only way to emphasise a constituent is to use

a cleft sentence.32 This suggests that there are no available fronting positions in the

left periphery to host topics, foci or wh-words. Second, Old Irish has unmarked verb-

initial word order. The verb obligatorily appears in clause-initial position, unless this

position is already filled. As a result, verb fronting can no longer be a pragmatically

motivated subtype of topicalisation or focussing; it must be grammatically

conditioned. Following CHP we assume that the fronted verb in pre-Old Irish targets

the C position. This is shown in the tree in (36) below.

36. CP

C TP

V

(Subj…Obj)

32 This also seems to be the case in Scots Gaelic. See Adger & Ramchand (2005).


142

So, it seems that from PIE to pre-Old Irish, two changes took place in the left

periphery of the clause: the articulated CP found in PIE was reanalysed as an

unarticulated CP, and pragmatically motivated verb fronting was reanalysed as

grammatically motivated V-to-C movement. This is shown schematically in (37).

37. Top/FocP CP

Top/Foc CwhP C T

vP /V V

Cwh TP (Subj…Obj)

(wh)

( Subj…Obj)

Having described the changes that distinguish the left periphery in pre-Old Irish from

that proposed for PIE, we must now attempt an explanation of why these changes

took place. As discussed in chapter 3, under a generative view, syntactic change is

primarily a result of language acquisition. So, in order to explain these changes we

must explain why children stopped acquiring an articulated CP, and reanalysed

stylistically motivated verb fronting as an obligatory, grammatically motivated

process. Let us consider each of these changes in turn.

The change between an articulated and an unarticulated CP could be seen as a

structural simplification (Clark & Roberts 1993; Roberts & Roussou 2003).33 During

language acquisition children assign the simplest possible analysis to the linguistic

data they encounter. Simplicity is often determined in terms of movement operations.

A grammar with more movement operations will be more complex than a grammar

with fewer. It seems plausible that this notion of simplicity could be extended to

phrase structure. During acquisition a child will acquire the least amount of structure

consistent with the data. Following Chomsky (2000, 2001 et seq.), let us assume that

there are three core functional categories, namely C, T and v, which are present cross-

linguistically. To deviate from this basic clausal architecture, i.e. to acquire extra

33 This is not consistent with Roberts & Roussou’s (2003) definition of structural simplification. Roberts & Roussou propose that simplicity is determined in terms of feature syncretisms. Change proceeds so as to reduce the number of functional features associated with a particular functional head. Seen in this way, then, an articulated CP should be preferable to an unarticulated CP as this involves the spread of functional features over a number of separate projections, resulting in fewer syncretisms.


143

functional positions, there must be sufficient evidence. In terms of the CP, then, it

seems likely that the child will only acquire an articulated CP, i.e. extra structural

positions, if there is sufficient evidence.34 Let us consider what this evidence might

be. It seems likely that the crucial evidence for the acquisition of an articulated CP is

the co-occurrence of multiple constituents with distinct pragmatic functions in the left

periphery, i.e. to the left of TP. In Italian this is clearly the case, with topics, foci and

various complementizers co-occurring. However, if, following Hale (1987, 1995), in

PIE only two constituents, a topic or focussed constituent and a wh-element, could

appear in the left periphery, why is it not simply analysed as an unarticulated CP? The

answer to this, I propose, lies in the nature of the constituents that appear in the left

periphery. If a language has an unarticulated CP, then the only two positions available

are the head position, C, and the specifier of CP. In an unarticulated CP, then, we

would only expect the combination XP-X in the left periphery.35 If we find XP-XP,

X-XP or X-X, specifically where the two heads occupy different syntactic positions,

i.e. they are not adjoined under a single C-head, then this is evidence for an

articulated CP. Section 4.3 argues that there is reason to believe that the first two

configurations were lost in Celtic, thus reducing the evidence for an articulated CP.

However, this left the third X-X configuration. This was lost at a later stage. This is

the topic of section 4.5.

Let us turn now to the second postulated change, i.e. the reanalysis of

pragmatically motivated, optional vP/verb-movement as grammatically motivated,

obligatory V-to-C movement. In terms of the theory of change outlined in chapter 3,

what we have here is the change from an optional movement feature (EF) to an

obligatory movement feature (Affix). There are two main pieces of evidence that can

be used to distinguish an optional from an obligatory movement feature. The first of

these is frequency. Movement motivated by an obligatory feature will be more

frequent than that motivated by an optional one. Secondly, an optional feature will be

associated with pragmatic or stylistic effect. When a child acquires a movement

feature associated with a particular functional category or feature it determines

whether this feature is optional or obligatory through consideration of frequency and

34 The alternative to this, of course, is that the articulated CP is universal and present in all grammars. It seems somewhat uneconomical, however, to propose that speakers of English, for example, have numerous structural positions present in their grammars that never receive any phonological realisation. 35 This is leaving aside the possibility of multiple specifiers.


144

pragmatic effect.36 In order for a movement feature to change from optional to

obligatory then the relevant movement process must appear to become more frequent

and lose its pragmatic function.37 Therefore, in order for pragmatic verb fronting to be

reanalysed as obligatory grammatical fronting, it must become more frequent, and the

pragmatic function associated with it must become bleached. In section 4.4 a possible

explanation for such an increase in frequency is proposed, linking the increase in verb

fronting to the increase in frequency of Cowgill’s *es particle. This increase in

frequency, in turn, leads to the bleaching of the pragmatic function.

By explaining the change in frequency and associated pragmatic function we can

account for the change from optional to obligatory verb-movement; however, this

does not account for the change in the position of the fronted verb. Whereas in PIE

fronted verbs seem to have targeted the outer position of the left periphery, in pre-Old

Irish fronted verbs occupy the C position. For this to be the case, a further change

must take place, whereby the elements that fill the inner peripheral position change in

status from clitic to affix, allowing the fronted verb to be analysed as appearing in the

inner position, i.e. in the C position. This change is linked to the loss of the final piece

of evidence for the loss of the articulated CP, and is dealt with in section 4.5.

4.3 Changes in the CP from PIE to pre-Old Irish

The aim of this section is to consider how the evidence relating to the acquisition of

an articulated CP changed between PIE and pre-Old Irish. It was argued above that in

order to acquire an articulated CP, children must encounter evidence of multiple

constituents co-occurring in the left periphery, i.e. to the left of TP. Crucially, these

constituents must be in the configuration XP-XP, X-XP or X-X. It seems, then, that a

condition for the loss of the articulated CP is the reanalysis of the inner position,

CwhP, from an XP to a head. In the non-Celtic early IE languages it seems that there

is clear evidence that the inner position is filled by an XP. Hale (1987, 1995) argues

that this position hosts wh-elements, specifically relative and interrogative pronouns.

In the early IE languages both interrogative and relative pronouns are inflected for

number, gender and case (see Fortson 2004: 126–130). A relative or interrogative

36 Of course, it will also be necessary to determine whether the movement feature attracts a head, an XP or both, i.e. whether it is an Affix, EPP or EF. 37 It is unclear at this stage whether the change from an optional to an obligatory feature should be more natural than the change from an obligatory to an optional one. As under the theory of change proposed here neither option is the default case we might expect them to be equally matched.


145

pronoun that is clearly inflected for number, gender and case will be analysed as a

DP, i.e. a phrasal category. This means that as long as the pronoun that appears in the

inner position is inflected, it will be analysed as an XP, giving the configurations XP-

XP or X-XP in the left periphery. This will ensure that an articulated CP continues to

be acquired. If we turn to Celtic, however, this does not seem to be the case. Neither

Old Irish nor Middle Welsh show an inflected relative pronoun.38 39

Let us consider first the relative system in Old Irish. There are many ways of

marking a relative clause in Old Irish, suggesting that relative marking is of

composite origin (Watkins 1963). In this section we will focus on the main strategies

for relative marking relevant to the discussion here. For a full account see Thurneysen

(1946: 312–325).

As we saw in chapter 2, if the verb is simple and in initial position a relative clause

can be marked by special relative morphology on the verb, as shown in (38a), where

the non-relative form would be gaibid. Similarly the preverbs im and ar have special

disyllabic relative forms imma/imme and ara/are, as in the example in (38b).

38. (a) is óinfer gaibes buáid diib inna chomalnad

COP one man take.PRES.3SG.REL victory of.3PL in-its completing

‘It is one man of them that gets victory for completing it’ (Wb 11a4)

(b) imme-rádi

PVB.REL-think.PRES.3SG.DT

‘He who thinks’ (Thurneysen 1946: 517)

In situations where relative verb morphology is impossible, the relative clause is

marked by a grammatical mutation, either lenition or nasalization. When the verb is

compound the mutation follows the initial preverb, changing the initial segment of the

verbal stem. In (39a) the initial /k/ of the verbal stem is lenited to /x/ (written ‘ch’)

after the preverb for. In (39b) the verbal stem is nasalized after the preverb do.

38 We will return to the question of interrogative wh-words in section 6 below. There is evidence to suggest that they did not occur in the inner position of the left periphery in pre-Old Irish, but appeared in the emphatic outer position. This being the case, interrogative wh-words are of little relevance to the discussion here. 39 The Germanic languages also lost the PIE inflected relative. Gothic, Old English and Old High German all have uninflected relative markers. These languages are all V2 languages and so, like pre-Old Irish, show V-to-C movement. It is not clear, however, that these two features are related.


146

39. (a) din gním for- chomnaccuir

to.the deed PVB- happen.PRET.3SG.DT

‘To the deed which happened’ (Ml 113d3)

(b) in tan do-mberam ar menmain intiu colleir

when PVB.give.PRES.1PL.DT our mind to.3PL diligently

‘When we give our mind to them diligently’ (Ml 21a8)

If the verb is simple, but of a person or number which does not have special relative

morphology, then the verb will be preceded by the dummy conjunct particle no and its

initial consonant will be mutated, as shown in (40) below.

40. is hed in so no⋅chairigur

COP.PRES.3SG it that PVB-reprimand.PRES.1SG.CONJ

‘That is what I reprimand’ (Wb 11d1)

The leniting relative clause is obligatory when the antecedent is the subject of the

relative clause, as in (41a) and optional when it is the object, as in (41b) (Thurneysen

1946: 314).

41. (a) ind huili doíni ro- chreitset

the all men PVB believe.PERF.3PL.CONJ

‘All men who have believed’

(Ml 60b16)

(b) ni torbe do an imdibe ad-chi cách

NEG profit to.3SG.M the circumcision PVB-see.PRES.3SG.DT everyone

‘It is no profit to him the circumcision which everyone sees’

(Wb 2a2)

The nasalizing relative has a much wider range of uses than its leniting counterpart

(Thurneysen 1946: 316–319). The most frequent use of a nasalizing relative clause is

when the antecedent is an adverbial expressing either time, manner or cause, or when

the clause follows a conjunction of time, manner or cause such as in tan, in tain


147

‘when’, amal ‘as’, óre ‘because’ as in example (42a).40 A nasalizing relative clause

can be used optionally in place of a leniting relative when the antecedent is the object

of the relative clause as in (42b). A further use of the nasalizing relative is to mark the

complement of verbs of saying, thinking and possibility or ‘in all contexts where the

complement of the principal clause can be more conveniently expressed by a second

clause than by a noun’ (Thurneysen 1946: 318). This is shown in (42c).

42. (a) in tan do-mberam ar menmain intiu colleir

when PVB-give.PRES.1PL.DT our mind to.3PL diligently

‘When we give our mind to them diligently’ (Ml 21a8)

(b) it hé sidi as-m-ber sís

cop.PRES.3PL they these PVB- says.PRES.3SG.DT below

‘It is these (things) that he mentions below’ (Wb 10b13)

(c) connic do-m-berthar forcell din titul

is.able PVB-NAS-give.PRES SUBJ.PASS.SG.DT testimony to.the title

‘A testimony may be given to the title (Ml 24d14)

Both the special relative verb forms and the leniting relative clause can be accounted

for by postulating an enclitic relative particle *(i)o (Thurneysen 1946: 323, Watkins

1963: 28, McCone 1980, Schrijver 1997).41 As an enclitic this particle would have

appeared in second position, according to Wackernagel’s law. This means that with

simple verbs it would have occurred immediately after the verb in initial position,

giving us the special relative forms, 3pl bertae < *beronti-io, 1pl bermae < *beromo-

io.42 When used in conjunction with a compound verb, the particle would have

appeared between the first preverb and the remainder of the verb. This accounts for

the form of the disyllabic relative preverbs imme (imma), and are (ara) and also the

leniting relative. The vocalic ending of the particle *(i)o would have phonologically

conditioned lenition of the following segment. When the final syllables were lost,

lenition became grammatically rather than phonologically triggered, and so became a

40 It seems likely that these conjunctions followed by nasalizing clauses were in origin temporal, manner or causal adverbs. We will return to this topic in section 6 below. 41 The origin of the nasalizing relative is highly disputed. See, among others, Breatnach (1980), McCone (1980), Ahlqvist (1983), Ó hUiginn (1986). 42 The third singular form is characterised by an /s/ and is more problematic. It is most often argued that this form developed analogically, although see Schrijver (1994) for a more recent analysis.


148

relative marker.43 In this way, the enclitic particle *(i)o can account for the leniting

relative following both preverbs of compound verbs and conjunct particles.44

Further support for the particle *(i)o can be obtained by considering the

comparative IE evidence. Vedic yá-, Ancient Greek hó, and Avestan ya- suggest that

the PIE relative also had the form *io- (Fortson 2004: 130). Therefore, it seems that

the relative marker in Celtic could simply have been inherited from PIE. However,

there are two fundamental differences between the Celtic and PIE particles. First it

seems likely that the PIE relative marker, unlike that found in Celtic, was inflected.

Secondly, the PIE relative marker was not enclitic. Let us examine each of these

issues in turn.

McCone (1980) argues that a Celtic inflected relative would have had the forms

below, yielding the mutations indicated in brackets (L=lenition, N=nasalization).

Table 4: The expected forms of an inflected relative pronoun in Celtic

Masc sg Fem sg Neut sg Masc pl Fem pl Neut pl

NOM *yos (–) *yā (L) *yo(d) (L) *yoi (L) *yās (–) *yā (L)

ACC *yom (N) *yam (N) *yo(d) (L) *yōs (–) *yās (–) *yā (L)

It could be argued that with the loss of the inflected endings, lenition spread

throughout the subject antecedent relatives and nasalization spread among object

antecedents. This provides a neat explanation for both the leniting and the nasalizing

relative, tying the Old Irish evidence to PIE. However, McCone provides statistical

evidence to suggest that this cannot be the case. In the Würzburg Glosses, dated at

around AD750, nasalization is less common than lenition when the antecedent of the

relative clause is a masculine or feminine object. By the time of the Milan Glosses, 50

years later, nasalization is used more frequently than lenition in the same

circumstances. By the St Gall Glosses, the nasalizing relative is well on its way out of

use. If, as suggested by the table above, nasalization originated with masculine and

feminine object antecedents we would not expect the use of nasalization to increase in

43 For more details on the development of the grammatical mutations in Celtic see Russell (1995: ch 7) and the references therein. 44 Thurneysen (1946: 323) suggests that the Welsh relative particle a could also provide support for the Celtic relative particle *(i)o. Although this particle appears in initial position in Middle Welsh, there is evidence to suggest this was not always the case. Evans (1964: 61) argues that in early Welsh formerly compound verbs such as digawn, cognate with Old Irish do-gní ‘makes, does’ do not co-occur with the relative particle, suggesting that originally the particle would have been infixed as it is in Old Irish.


149

these situations between the time of Würzburg and Milan, and then to be lost between

Milan and St Gall. This suggests that the nasalizing relative is not a reflex of the PIE

inflected relative, but an Irish innovation.45

McCone’s second main argument against the origins of the Old Irish relative

construction being in the PIE relative pronoun is that evidence from Sanskrit and

Ancient Greek suggests that this particle appeared mostly in initial position. Although

the relative pronoun can stand in second position, this only tends to occur when some

other element has been pragmatically fronted to the left of the relative marker. The

PIE relative pronoun is by no means enclitic. However, the fact that the relative

pronoun is not reconstructed as an enclitic for PIE does not rule out the possibility

that it became an enclitic in Celtic. The relative pronoun in PIE could appear in

second position when some other element was pragmatically fronted. As noted above,

one of the conditions for the shift to unmarked verb-intial order seems to be an

increase in the frequency of pragmatically fronted verbs. An increase in pragmatic

verb fronting would have resulted in the relative pronoun appearing in second

position with increasing frequency. Schrijver (1997: 104) also points out that it would

have been perfectly possible for the relative pronoun to have remodelled itself on

other types of enclitic pronominal elements.

Having established that the Old Irish relative cannot have developed directly from

the PIE relative, McCone requires an alternative etymology. Thurneysen (1946: 323)

suggests that Celtic *(i)o may be cognate with the neuter form of the PIE relative

pronoun *iod (cf. Sanskrit yád), which has lost its final *-d, or the Hittite sentence

connective ya. McCone opts for the latter, primarily on the basis that it has been

shown by Watkins (1963) that relative markers can develop from co-ordinating

conjunctions.

One problem with McCone’s account, however, is that it provides no explanation

as to what happened to the inflected relative in Celtic. Although the Insular Celtic

languages and also Gaulish (see Schrijver 1997: 105) show no evidence of an

inflected relative, there is evidence for an inflected relative in Celtiberian. In the

Botorrita I inscription, for example, we find a masculine nominative singular ios,

masculine dative singular iomui and a feminine accusative plural ias (Eska & Ellis

Evans 1993:35). This suggests that we should reconstruct an inflected relative for 45 Further support for this conclusion comes from the fact that there are no signs of a nasalizing relative in Welsh (McCone 1980: 18).


150

Proto-Celtic. Of course, it is possible that the inflected relative was simply replaced

by a new strategy of relative marking utilising sentential connectives in the Insular

Celtic languages and Gaulish. However, this leaves us with a somewhat striking

coincidence that the relative in Old Irish seems to have developed from a connective

that had the same shape as the stem of the PIE inflected relative.

Let us explore Thurneysen’s alternative, namely that the Insular Celtic uninflected

relative marker *(i)o developed from the 3sg neuter PIE relative *iod. As noted

above, in many other early IE dialects that-complementizers seem to have developed

from the 3sg neuter relative pronoun (e.g. Latin quod, Sanskrit yád, Greek hóti, Hittite

kuit). In all these cases an inflected pronoun in Spec-CP has been reanalysed as an

uninflected complementizer filling the C head. In structural terms, the development of

the relative marker in Celtic can be viewed in the same way, with the relative pronoun

being reanalysed as a relative complementizer; however, in Celtic this change was

more far reaching, with the new relative complementizer replacing all the relative

pronouns.

For the PIE relative pronoun to have been reanalysed as an uninflected relative

marker in Celtic it must have lost or have been well on its way to losing its

inflections. There are two possible explanations for this. First, it is possible that the

inflectional endings of the relative pronoun were eroded by phonological change,

leaving a relative pronoun with minimal or no number, person or case distinctions.

This seems plausible, as, although nominal expressions maintain their inflections into

the Classical Old Irish period, pronouns do not. Pronouns have only nominative and

accusative forms (Thurneysen 1946: 257–260). Therefore, it could be the case that the

relative pronouns lost their inflections as part of a change affecting the pronominal

system as a whole. The second possibility is that new strategies were introduced for

marking relative clauses, causing the inflected relative to become restricted to use

with a certain type of antecedent. If, for example, the inherited relative was only used

with subject antecedents, then it would only appear in the nominative case. Such a

scenario is proposed by Schrijver (1997)

Schrijver (1997) suggests that subject antecedent and object antecedent relative

clauses were marked in different ways in pre-Old Irish. He argues that only subject

antecedent relatives were marked with the inherited *io relative. Object antecedent

relatives were marked by enclitic pronouns *em>*en ‘him’, used for masculine and

feminine and *ed>*e ‘it’ used with neuter antecedents. The main piece of evidence


151

that Schrijver employs to support this claim comes from the distribution of contracted

and uncontracted forms in relative clauses. In the Glosses certain compound verbs can

appear in initial position in either contracted or uncontracted form. As a result we find

forms such as tuc ‘brought’ alongside forms such as do-uc. Schrijver argues that in

relative clauses the contracted forms are used almost exclusively with neuter object

antecedents, whereas the uncontracted form is found with subject antecedents. The

reason for this, Schrijver argues, is that the neuter object pronoun *e used in relative

clauses allowed contraction, whereas the *io particle found in subject relative clauses

prevented it.

If, as Schrijver suggests, the *(i)o relative was only used with subject antecedents

it would only have surfaced in its nominative form and so would have appeared to be

uninflected for case. Of course, at this stage the relative pronoun would still have

been inflected for person and number. However, these properties are not limited to

DPs. Verbs and also complementizers can show person and number agreement (see

Fuß 2005 on inflected complementizers in Germanic). Case, on the other hand, is a

property of DPs. When the pronoun lost its case inflection it lost its distinguishing

marker as a DP; therefore, it was possible for it to be reanalysed as a C particle. After

this reanalysis, the particle underwent further phonological change, leading to the loss

of number and gender distinctions, and the change from independent word to enclitic.

The change from relative pronoun to relative complementizer can be seen as a case

of grammaticalisation, with the pronoun losing its case, number and gender features

and becoming purely a marker of relative clauses. It has often been observed that

phonological reduction occurs as a result of grammaticalization (Hopper & Traugott

1993, Roberts & Roussou 2003: 224f). As a lexical item becomes more grammatical

it can lose its independent stress and become enclitic. A similar case to that presented

here is the development of future tense marking in Romance as outlined by Roberts &

Roussou (2003: 48f). The future tense endings in the modern Romance languages

developed from the Latin verb habere. If we compare the French future tense endings

shown in (43a) with the verb avoir ‘to have’ given in (43b), we see that they are

almost identical.

43. (a) chanter-ai, chanter-as, chanter-a, chanter-ons, chanter-ez, chanter-ont

(b) ai, as, a, avons, avez, ont


152

In late Latin one option for expressing the future was to use a periphrastic

construction involving the verb habere and an infinitive. Over time the verb habere,

an independent and fully stressed word, was reanalysed as a future auxiliary (like will

in English). As a result of this reanalysis it became phonologically reduced, being

reanalysed first as a clitic, and then an affix, as seen in the modern Romance

languages.

The discussion in the preceding paragraphs has been of a somewhat tentative

nature, and more research is necessary before any firm conclusions can be drawn

about the origins of the Celtic uninflected relative marker. This section has developed

a new account, namely that the Insular Celtic relative marker could have developed

through the grammaticalisation of the PIE relative pronoun. This will be assumed in

what is to follow, although it is not crucial. The important point to note is that the PIE

inflected relative was lost, as a result of which the inner position of the left periphery

in Celtic would have contained an uninflected relative marker, crucially a head and

not an XP. This change led to the loss of the configurations XP-XP and X-XP;

however, as topicalisation of the verb could involve movement of just a head the

configuration X-X remained. The articulated CP could not be lost, then, until this

configuration disappeared. Before we go on to consider how this occurred, let us turn

our attention to the changes affecting the fronted verb.

4.4 From optional to obligatory verb fronting

Between PIE and pre-Old Irish optional, pragmatically motivated verb-movement was

reanalysed as obligatory, grammatically motivated verb-movement. In theoretical

terms this involves a change in the status of the feature motivating the movement.

Whereas in PIE the feature motivating movement of the verb was optional, in pre-Old

Irish this feature became obligatory. It was argued above that in order for this change

to take place the frequency of verb fronting had to increase and the pragmatic

function associated with a fronted verb had to be bleached. This section proposes that

not only can the increase in frequency and the bleaching be linked to one another, but

that both of these changes can be linked to Cowgill’s particle *es.

In section 2 above it was argued that although the particle theory is preferable on

conceptual grounds to McCone’s suffix-/infix-deletion theory, there are two major

problems with it. First no satisfactory etymology has been provided for the particle

*es. Secondly the theory provides no explanation for the increase in verb fronting. As


153

discussed in section 2.2 above, Schrijver (1994, 1997) attempts to solve the first

problem through his proposal that the particle *es developed from a particle of the

form *eti, a main clause connective cognate with Latin et ‘and’, Gothic iþ ‘but’ and

also Ancient Greek éti ‘further, moreover’ and Sanskrit áti ‘beyond, very’. For this

particle to have had such a significant effect on the morphology of the verb it must

have been very common. Schrijver observes that in the early IE languages there is a

tendency to use a particle to express the link between clauses overtly. Watkins (1963:

13) observes that in Hittite the connective nu is found at the beginning of almost

every clause. Watkins also gives an example from Archaic Latin that shows extensive

use of the connective -que:

44. Macel[am-que opidom] / [p]ucnando cepet. En-que Macella.ACC.SG-and town.ACC.SG fighting. ABL.SG captured.3SG In-and eodem mac[istratud bene] / [r]em nauebos same.ABL.SG magistracy.ABL.SG well situation.ACC.SG ships.ABL.PL marid consol primos c[eset copias-que] / sea.ABL.SG consul.NOM.SG first.NOM.SG carried out.3SG forces.ACC.PL-and [c]lases-que nauales primos ornauet pa[rauet-que]

fleets.ACC.PL-and naval.ACC.PL first.NOM.SG put.in.order.3SG prepared.3SG-and /cum-que eis nauebos claseis Poenicas om[nis with-and these.ABL.PL ships.ABL.PL fleet.ACC.PL Punic.ACC.PLwhole.ACC.PL item ma-] / [x]umas copias Cartaciniensis praesente[d and greatest.ACC.PL forces.ACC.PL Carthaginian.ACC.PL present.ABL.SG Hanibaled] / dicatored ol[or]om in altod marid Hanibal.ABL.SG dictator.ABL.SG them.GEN.PL in high.ABL.SG sea.ABL.SG

pucn[ad uicet] / ui-que naue[is cepe]t battle.ABL.SG conquered.3SG force.ABL.SG-andships.ACC.PLcaptured.3SG

‘He captured the town of Macella by fighting. And in the same magistracy, he

was the first consul to be successful in battle with ships on the sea. He put in

order and prepared the first naval forces and navies. And with these ships he

conquered the whole Punic fleet and also the greatest Carthaginian forces, in the

presence of Hannibal, their dictator, in battle on the high sea and he captured the

ships by force...’ 46

(Degrassi 1957: 189–90; Watkins 1963: 8–9)

46 I am grateful to Nick Zair for help with the translation of this example.


154

Schrijver (1997) also observes that in Middle Welsh the connective ac ‘and’ joins

almost every clause. If this was also the case in pre-Old Irish, or even Insular Celtic,

then we have a possible explanation for why this particle came to be used so

frequently that it could have affected the verbal inflections.47

It seems, however, that using a connective was not the only way that the

connection between clauses could be marked. Dressler (1969) argues that fronting the

verb in the early IE languages could have a cataphoric or anaphoric effect linking two

clauses together. Such an occurrence of verb fronting is seen in the example from

Homeric Greek in (28b) above, repeated as (45) below. The main verb of the second

clause, erêtuseie ‘curb’, is fronted to the beginning of the clause in order to mark the

connection between the two clauses.

45. êe kholon pauseien erêtuseie te thumon.

or wrath.ACC check curb and spirit.ACC

‘Or he should check his wrath and curb his spirit’ (Il 1, 192)

The example in (45) is particularly interesting because it shows that it was possible to

use both strategies for linking clauses together. Not only is the verb fronted in the

second clause, but there is also a connective particle te in second position. In order for

the connective particle *es to have become verbal morphology this must have also

been the case in pre-Old Irish. One way to explain this development in pre-Old Irish

is that, like the Hittite connective nu, the particle *es became used so frequently that it

was bleached of its connective meaning. As a result of this bleaching there was an

increase in pragmatic verb fronting to the left periphery in order to reassert the link

between clauses. As verb-intial clauses increased in frequency, this led to further

bleaching of verb fronting, causing the connective force to be lost and the fronted

verb to become susceptible to reanalysis.48

47 David Willis (p.c.) points out that this use of connectives could simply be a literary feature and not reflect the language spoken at the time. However, it is interesting to note that it seems to be prevalent in many different branches of IE. This perhaps suggests that it is an inherited linguistic feature rather than a shared literary one. Although see Watkins (2001) on the possibility of shared literary features. 48 See Garrett (1994) for similar ideas regarding the role of bleaching in the development of unmarked verb-intial order in the Anatolian language Lycian.


155

To sum up this section, it seems that adopting Schrijver’s etymology for the

particle *es provides an explanation not only as to why the particle was so prolific but

also why there was an increase in frequency of verb fronting. It seems likely that this

increase in verb fronting would have led to a loss in pragmatic force, and so the

conditions were met for the reanalysis of verb fronting from an optional process to an

obligatory one. Although in this section an explanation has been offered for the

change from optional to obligatory verb fronting, it has not addressed the change in

the target position of the movement, i.e. the development of V-to-C movement. This

is the topic of the next section.

4.5 From clitic to affix: changes in the inner C position

To sum up the situation so far, in section 4.2 it was argued that unlike other early IE

languages, Insular Celtic did not have an inflected relative pronoun. The inherited PIE

relative pronoun was grammaticalised and became an uninflected relative marker. As

a result, the inner peripheral position in Insular Celtic contained a head rather than a

phrase, eliminating two of the possible configurations that provided evidence for an

articulated CP. Section 4.3 argued that there was an increase in frequency of verb-

intial clauses in pre-Old Irish, due to a requirement to mark the link between clauses

overtly. This meant that the verb followed by the enclitic connective *es appeared at

the beginning of virtually every main clause, making it possible for the verb fronting

to be reanalysed as an obligatory, grammatically motivated process. However, at this

stage the reanalysis of verb fronting as V-to-C movement is not yet possible. The

fronted verb occupies the outer position of the left periphery. If the fronted verb was

in all cases a remnant XP, then we could perhaps assume at this stage that the

articulated CP had been lost. In structural terms we would have the configuration XP-

X, which is compatible with an unarticulated CP; furthermore, with the increase in

frequency outlined in section 4.4, the fronted verb will have lost its pragmatic force,

reducing the evidence for a separate Topic/Focus position in the left periphery.

However, as argued in section 3.3 above, verb fronting could simply involve the

verbal head. In this case we have the configuration X-X, and therefore evidence for an

articulated CP. It seems then that for the articulated CP to be lost and for verb fronting

to be reanalysed as V-to-C movement there must be a further change, namely the

particles that fill the inner position must be reanalysed from independent syntactic

elements to affixes. It was argued in section 4.2 that the relative marker fills the inner


156

position (Cwh). However, it seems likely that this was not the only element that could

do so. Before we consider how the change from clitic to affix took place, let us

examine another candidate for this position, namely the connective clitic *es.

Hale (1995) argues that in Vedic and PIE connective clitics are placed by prosodic

inversion (Halpern 1995). The connective clitics are adjoined to the left of the second

conjunct during the syntax. Then, if no appropriate phonological host is provided by

the syntax, the clitics move to the right of the initial prosodic word during the

phonology so that they have a phonological host to their left. The evidence Hale

provides for this is that in Vedic Sanskrit and Greek, as shown in examples (46) and

(47), connective clitics appear between topicalised phrases in the outer position

(Top/FocP) and the relative or wh-phrases that occupy the specifier of the inner

position (Spec-CwhP). This means that in the early IE languages and PIE under Hale’s

analysis the connective clitics cannot be in the inner head position, i.e. Cwh.

46. stútaṣ ca yµs tvā várdhanti mahé rµdhase nṛmṇµya

praise-songs and which you increase great giving heroism

‘And which praise-songs fortify you for great giving and heroism’

(RV 8.2.29ab; Hale 1995: 266)

47. ha dè tálain’ álokhos tíni moi ôleto moírai

the PTC wretched wife.NOM wh.DAT.SG my.DAT.SG perished fate.DAT

‘But by what fate did my wretched wife perish?’

(Euripides Phoenissae 1566; Garrett 1996: 88)

However, it was argued above that the left periphery is structured somewhat

differently in Celtic to the other early IE languages. After the change of relative

pronouns from phrases to heads and also changes in the formation of wh-questions

(on which see section 6.2 below), the specifier of the inner C position is no longer

filled; therefore it becomes possible for PIE connective clitics to be reanalysed as C-

related particles in Celtic. As noted above, Watkins (1963) suggests that the PIE

connective clitics *kwe and *de became relative markers in Old Irish. Assuming that


157

relative markers are heads appearing in the inner C position, then it seems that these

clitics must have undergone a change from sentential clitic to C-related head.49

Let us consider the evidence for a similar change with regards to the connective

*es. It was argued above that *es was bleached of connective meaning because it was

used so frequently; however, the particle continued to be used long enough to prevent

apocope of the final syllables of initial verbs. It seems unlikely that this particle would

have been completely devoid of function. One possible explanation is that when the

particle was bleached of its connective meaning it became reanalysed as a clause-

typing particle, marking declarative, non-relative clauses. Such particles are common

in Brittonic Celtic. Middle Welsh has a declarative particle of the form y and Modern

Welsh has the particles mi and fe (Evans 1964, Willis 1998). In addition, the form of

special relative verbs suggests that the relative particle *io and the connective *es

never occurred together on the same verb. The two particles seem to be in

complementary distribution (Schrijver 1994, 1997). If they are both clause-marking

particles that appeared in the same syntactic position, namely the inner C position,

marking relative and non-relative clauses, then this is to be expected.

So, for each clause the inner C position (Cwh) is filled by a clause-marking clitic. In

main, i.e. non-relative, clauses, this clitic is *es and in relative clauses it is *io. Either

the verb or the entire vP is fronted in order to mark the link between clauses. This is

shown in the trees in (48) below:50

48. (a) Main clauses: (b) Relative clauses:

Top/FocP Top/FocP

Top/Foc CwhP Top/Foc CwhP vP /V vP/V

Spec Cwh` Spec Cwh` Ø Ø

Cwh TP Cwh TP es io (Subj…Obj) (Subj…Obj)

49 For the connectives to have been reanalysed as C elements they must either have been in complementary distribution with the relative particle *io or have combined with it. The former is perhaps more likely as there seems to be no phonological evidence of a union between the two. 50 If a specifier position is empty as in the trees in (56) it is simply not projected, and so absent from the structure. I have included the specifiers in the trees in (56) to emphasise the fact that after the reanalysis of *es and *io as C heads the specifier position is empty.


158

However, if, like the relative particle *io, the particle *es appears in the inner

peripheral position, then this means that both positions in the left periphery will be

filled, and so the verb cannot be analysed as filling this position. For verb fronting to

be analysed as V-to-C movement, then, the clause marking clitics in the inner position

must be reanalysed as verbal affixes.

The change from clitic to affix is a common case of grammaticalisation (see

Hopper & Traugott 1993; Roberts & Roussou 2003). In spite of the fact that this type

of change is so common, little work has been done to show exactly how it proceeds.

Before we can determine whether a clitic has become an affix we need a clear idea

about what the distinction between a clitic and an affix is. Zwicky & Pullum (1983:

504–5) provide criteria for making this distinction:

I) Clitics can exhibit a low degree of selection with respect to their hosts, while

affixes can exhibit a high degree of selection with respect to their stems.

II) Arbitrary gaps in the set of combinations are more characteristic of affixed

words than of clitic groups.

III) Morphophonological idiosyncrasies are more characteristic of affixed words

than of clitic groups.

IV) Semantic idiosyncrasies are more characteristic of affixed words than of clitic

groups.

V) Syntactic rules can affect affixed words, but cannot affect clitic groups.

VI) Clitics can attach to material already containing clitics, but affixes cannot.

These criteria, as set out by Zwicky & Pullum, are designed to help the linguist

determine whether a grammatical entity is a clitic or an affix. Fuß (2005) proposes

that children might use something similar during language acquisition to determine

whether a certain element should be analysed as a clitic or an affix. Let us relate this

idea to Celtic.

In section 2.3 it was argued that the main motivation from a linguist’s perspective

for the existence of the particle *es was to explain the lack of *i-apocope in verbs in

initial position. The idea that this particle prevents apocope does not provide any

evidence, however, as to whether it is a clitic or an affix. Apocope is a phonological

change and so takes place in the phonological component where a fronted verb and a


159

clitic make up a single phonological word in the same way as a verb and an affix.

However, after this apocope has taken place, verbs that appear with *es have a

different stem to verbs without *es. The verbal form that appears with *es will have a

final *-i, whereas the forms that do not occur with *es will not have this *-i, as it will

have been apocopated. This means that after apocope the particle *es appears to

condition a stem alternation. According to Zwicky & Pullum’s condition (III) above,

this is a property of affixes and not clitics. In addition, due to the frequency of

remnant vP fronting, the *es particle virtually always appears with a verb. Appearing

consistently with the same type of host, according to Zwicky & Pullum’s condition (I)

is again a property of affixes rather than clitics. On the basis of this evidence, then,

children reanalyse the *es particle as an affix rather than a clitic. On the assumption

that *io underwent the same development, then once this reanalysis has taken place,

the fronted verb can (and will) be analysed as appearing in the inner position in both

relative and non-relative clauses, and so verb-movement to the inner C position is

acquired. Furthermore, once verb fronting is analysed as targeting the inner position,

this eliminates the configuration X-X and so evidence for an articulated CP no longer

remains, and so we have V-to-C movement.

What this section has argued so far is that the development of V-to-C movement is

linked to the development of an unarticulated CP. The two changes seem to have

occurred in tandem. Interestingly, however, the analysis presented in this section also

suggests that the development of V-to-C movement is related to the development of

new verbal morphology. The reanalysis of the particle *es as an affix associated with

clause-initial verbs represents the first stage of the development of absolute

morphology. So, it seems that the development of V-to-C movement and the

development of the double system of verbal morphology are linked.

In section 2 above it was argued that one of the main conceptual advantages of

McCone’s suffix-/infix-deletion theory over Cowgill’s particle theory is that it links

the development of the double system of verbal inflection to the development of

unmarked verb-intial word order. Adopting the analysis proposed in this section, this

advantage no longer holds. By supplementing Cowgill’s particle theory with ideas

from syntactic change it seems to be possible to explain the shift to verb-intial word

order and how this is linked to the development of the double system of verbal

inflection.


160

4.6 Summary

This section has presented a new account of how verb-intial word order may have

developed in pre-Old Irish. It has been argued that the development of verb-intial

order seems to be linked not only to changes in the C-system causing the loss of the

articulated CP, but also to a morphological change, namely the development of

absolute verbal morphology. In section 4.2 it was argued that within a cue-based

theory of language change the acquisition of both an articulated CP and optional

rather than obligatory verb-movement requires the presence of certain cues in the

linguistic data. Section 4.3 argued that changes in the relative system between PIE

and Insular Celtic ensured that an XP could not appear in the inner position of the left

periphery. As a result, two of the three possible cues for an articulated CP were lost.

In section 4.4 it was argued that the tendency to mark clauses overtly led to an

increased use of clausal connectives and verb fronting. This increase in frequency of

verb fronting led its pragmatic force to be bleached, and the initial verb became

subject to reanalysis. However, at this stage verb fronting could not be analysed as V-

to-C movement as the inner position, the Cwh head, was filled. Section 4.5 proposed

that it was the reanalysis of clause typing clitics that filled the Cwh position from

clitics to affixes that resulted not only in the development of V-to-C movement but

also in the loss of the articulated CP.

Having developed an account of how simple verbs may have come to appear in

absolute initial position, let us now turn our attention to compound verbs.

5 THE DEVELOPMENT OF PREVERBS

The main (non-phonological) differences between initial preverbs in PIE and Old

Irish are equivalent to the differences exhibited by simple verbs. As discussed in

section 3.4, in PIE fronting a preverb was optional and had a pragmatic effect. In Old

Irish, on the other hand as seen in chapter 2, when C is unfilled and the verb is

compound, fronting the preverb is obligatory and stylistically unmarked. The

reanalysis undergone by preverbs, then, is essentially the same as that undergone by

simple verbs: pragmatically motivated, optional preverb fronting to a

Topicalisation/Focus projection in an articulated left periphery is reanalysed as

grammatically motivated movement of the preverb to a generalised C position. The

change in the status of preverbs, then, seems to be dependent on the same changes as

the reanalysis of simple verbs. First, fronting of the initial preverb must become


161

sufficiently frequent and lose its pragmatic force so that it is susceptible to reanalysis.

Secondly, the inner position of the CP must change so that the evidence for an

articulated CP decreases, and the preverb can be analysed as occupying the single C

head position.

Before we consider the parallels between the development of preverb-movement

and verb-movement, there is an additional issue that must be addressed. It was noted

above that preverbs were originally adverbial elements. Adverbs are typically phrases;

therefore in order to appear in the C position preverbs must have undergone a change

from XP to head.51 It seems plausible that preverbs underwent grammaticalisation,

whereby they lost their independent stress and their independent meaning. It seems

likely that the change from XP to head is connected to these changes.

As discussed in section 3.4 above, preverb fronting seems to have been

functionally equivalent to fronting of a simple verb in the early IE languages. It was

argued in section 4.4 that the main motivation for verb fronting seems to have been to

mark the link between clauses. When the verb was compound, then, the initial preverb

would have been fronted in place of the verb in order to mark this link. The increase

in frequency of fronted preverbs can be explained in the same way as the increase in

frequency of fronted verbs. After the bleaching of the connective particle *es and its

change in status to a clause typing particle, the frequency of preverb fronting

increased in order to mark the link between clauses. As a result of this, the

combination of initial preverb+*es became sufficiently frequent that it was bleached

of its pragmatic force and became subject to reanalysis.

Let us turn now to the second prerequisite for the reanalysis of the preverbs,

namely the changes to the inner position of the left periphery. It was argued in section

4 that for pragmatically fronted verbs to be reanalysed as appearing in the head of an

unarticulated CP, the evidence for the articulated CP had to decrease. Specifically it

was argued in section 4.5 that the particles that could appear in the inner C position,

namely the former connective *es and the relative marker *io had to undergo a

change from clitic to affix. This change seems quite natural in the case of verbs, as

verbs tend to show morphological endings. Preverbs on the other hand are typically

indeclinable and so do not tend to appear with affixes. However, it seems that there is

both cross-linguistic and Irish evidence to support this claim that the particles that 51 This is not necessarily the case if we adopt Sportiche’s (1992) view of clitic movement. See chapter 2, section 3.4 for details.


162

could appear in the inner C position became affixed to preverbs as well as verbs. Let

us consider the Irish evidence first.

It seems that in Old Irish preverbs are not completely indeclinable. Certain

preverbs show different forms in different contexts, suggesting that they can undergo

affixation in the same way as verbs. In chapter 2, section 3.3.2 we saw that like

simple verbs certain preverbs had special relative forms. In relative contexts the

preverbs im(m) and ar have the forms imme/imma and ara/are (Thurneysen 1946:

314). This extra vowel could be taken to reflect the presence of the relative particle

*io (Watkins 1963: 25). However, this relative marking is no longer a clitic, as its

distribution is severely restricted. Instead, it seems to appear as an affix on the

preverb. Similar traces can also be found of the non-relative particle *es. Russell

(1988) observes that certain preverbs have sigmatic forms, so for example we find the

deuterotonic sigmatic form fris-gair ‘answers, replies’ alongside the prototonic form –

frecair which must have developed from asigmatic *frith-gair. Russell further notes

that “the original pattern…seems to have consisted of sigmatic preverbs in

deuterotonic forms and asigmatic in prototonic forms” (Russell 1988: 160). These

sigmatic forms can be explained by the presence of Cowgill’s particle *es. As this

particle occurred in second position it would only have appeared after preverbs in

absolute initial position. Therefore, we would expect to find reflexes of the particle

only in deuterotonic forms and not in prototonic forms. This is exactly the pattern

Russell describes. As with the relative forms of the preverbs given above, the particle

*es seems to have become part of the preverb itself, suggesting that it is an affix

rather than a clitic.52 53

Let us consider now some cross-linguistic evidence. In certain Germanic dialects

complementizers can show number and person agreement. The examples in (50)

below come from Lower Bavarian, and those in (51) are from West Flemish.54

52 Not all preverbs have a sigmatic form when they are used in deuterotonic verbs. In most instances we would expect the final *es to have been lost through apocope as it was in the absolute verb forms. It is unclear why this was not the case with the preverb frith. 53 See Schrijver (1994) for an alternative account of the development of the sigmatic form of the preverb fris-. 54 For further details of complementizer agreement in Lower Bavarian see Bayer (1984). On West Flemish see Haegeman (1992). For recent theoretical accounts of this phenomenon see Fuß (2005) and Ackema & Neeleman (2005).


163

50. (a) ob-st (du) noch Minga kumm-st

whether-2SG (you) to Munich come-2SG

‘Whether you come to Munich’ (Fuß 2005: 159)

(b) wem-ma (mia) aaf Minga fon

when-1PL (we) to Munich drive

‘When we drive to Munich’ (Fuß 2005: 165)

51. (a) da-n ik werken

that-1SG I work.1SG

‘That I work’

(b) da gie werkt

that-2SG you work.2SG

‘That you work’ (Ackema & Neeleman 2005: 236)

Complementizers are generally considered to be indeclinable; however, it seems that

in some varieties it is possible for complementizers to appear with affixes. The

Germanic evidence, then, provides cross-linguistic support for the idea that affixes

can be associated with elements other than verbs. Fuß (2005) examines how

complementizer agreement developed in these Germanic dialects. He argues that

these inflectional affixes were in origin pronominal clitics, which were reanalysed

from clitic to affix in a similar way to that proposed for Irish above. Interestingly for

our purposes, Fuß proposes that the pronominal clitics were reanalysed as affixes first

with verbs and then the affixes subsequently spread to complementizers. The same

progression could be postulated for pre-Old Irish. The reanalysis of the C-oriented

clitics *es and *io as affixes may have taken place first in the most natural

environment, i.e. with verbs, and then secondarily in the less natural environment, i.e.

with preverbs.

It seems then that the development of preverbs from pragmatically fronted

elements to grammatically placed C elements can be explained in the same way as the

development of verbs. Having provided an account of the development of preverbs in

Old Irish, let us move on to the final class of elements that can fill C, namely conjunct

particles.


164

6 THE DEVELOPMENT OF CONJUNCT PARTICLES

In traditional grammars of Old Irish, conjunct particles are defined as particles that

cause the verb to take conjunct inflection or prototonic form (Thurneysen 1946: 28).

In chapter 2 it was argued, following Carnie, Harley & Pyatt (2000), Doherty (2000)

and Adger (to appear), that conjunct particles are complementizers, i.e. they fill the C

position. So far in this chapter, it has been argued that before the development of V-

to-C movement the only elements that could fill the ‘C position’, i.e. the head of the

inner projection of the left periphery, were enclitic clause-typing particles, such as the

relative particle *io or the non-relative particle *es, which derived from a connective.

If this is the case then we need an explanation as to how and when the Old Irish

conjunct particles developed. This is the aim of this section. The conjunct particles in

Old Irish can be divided into three main groups: negative particles, interrogative

particles and conjunctions. We will consider each of these in turn.

6.1 Negative particles

It seems that a parallel can be drawn between the development of the negative particle

and that of the preverbs in Old Irish. Fortson (2004: 149) notes that from the evidence

of the earliest IE languages it is possible to reconstruct a negative adverb of the form

*ne. In order to mark clausal negation, the default position for this particle was before

the verb. However, it seems it was also possible to move the negative particle to the

beginning of the clause for emphasis. McCone (1979a, 1997b) observes that this

process of fronting the negative to clause-initial position can have the same effect as

fronting the verb. In example (52) below (repeated from (33) above) each of the three

clauses contains an imperative, and so we would expect the verb to be in initial

position. However, in the second clause, the negative particle ma is fronted in place of

the verb.

52. ā no yajñe bhajata, mā no yajñad antárgata, PVB us sacrifice share.IMPV.2PL NEG us sacrifice exclude. IMPV.2PL

astv evá no ‘pi yajñe bhāga iti be.IMPV.3SG thus us PTC sacrifice share PTC

‘give us a share in the sacrifice, do not exclude us from the sacrifice, let there

be a share in the sacrifice for us.’ (McCone 1997b: 372)


165

If negation is fronted in the same way as verbs and preverbs to mark the link between

clauses, and marking the link between clauses is obligatory, then this provides an

explanation of how initial negation became so frequent that it was possible to

reanalyse it as obligatory.55 An interesting question at this point is whether the

negative particle co-occurred with connective particles. If the negative was fronted in

the same way as verbs and preverbs to the outer peripheral position, then we might

expect a clausal connective to be present in the inner peripheral position. There is

some evidence to suggest that this was the case.

The negative particle in Old Irish has special forms when it appears in relative

contexts and with an infixed pronoun. When a negative clause is relative, the negative

particle has the form nad. Watkins (1963: 26) argues that this /d/ is equivalent to the

/d/ that characterises Class C infixed pronouns, i.e. those used in a relative context.

This /d/, Watkins argues, derives from a PIE clausal connective cognate with Ancient

Greek dè. When the negative particle appears with an infixed pronoun it has the form

nach-. Watkins (1963) argues that this reflects the presence of the IE connective clitic

*-kwe-. So it seems that the negative particle must have co-occurred with at least some

connective clitics that occupied the C position. There is perhaps also reason to believe

that the negative particle co-occurred with the connective *es. In Old Irish the basic

form of the negative alternates between ni and ní. Only the first of these would have

developed from the PIE form *ne. Thurneysen (1946: 153) argues that the alternative

with the long vowel most likely arose from confusion with the negative form of the

copula *ne-est>*nés(t)>*nís>*ní. The same long vowel could also be achieved if the

negative particle occurred with the connective *es. Further research is necessary to

determine whether this is viable.

So, in sum, it seems that the development of negation as a conjunct particle,

occurring in the C position, can plausibly be seen as part of the same development

that led to the generalisation of verb-intial and preverb-initial word orders.

55 It is not clear why fronting the negative particle should carry the same force as fronting the verb. The negative particle cannot be considered part of the verb in the same way as a preverb. Traditionally, negation is a projection in its own right, not part of the vP. One possible explanation is that fronting emphasises polarity. Whereas fronting the verb emphasises positive polarity, fronting the negative emphasises negative polarity.


166

6.2 Interrogative particles

As discussed in chapter 2, there are two ways to form a question in Old Irish. In the

case of yes/no-questions an interrogative particle in appears in clause-initial position,

causing the verb to have dependent form. To form a wh-question, a wh-word appears

in initial position. The etymology of the interrogative particle in is unclear (although

see Pedersen (1909–1913: 391) for one suggestion). As a result this section will focus

on the history of the wh-words.

There are two classes of interrogative pronoun in Old Irish. The first class, ce,ci,

cía, are generally unstressed or weakly stressed and invariable for number and gender.

They behave like conjunct particles, being followed by a verb in conjunct or

prototonic form, as shown in the examples in (53) below.

53. (a) cía beir búar o thig Temrach

who take.PRES.3SG.CONJ cattle from house Tara.GEN.SG

‘Who takes cattle from the house of Tara?’ (LL 1566)

(b) cía acca

who see.PRET.2SG.PT

‘Who did you see?’ (LU 7135)

The second class cía ‘who’, cid, ced ‘what’ are stressed and followed by a relative

clause, as shown in (54).

54. (a) cía rannas dúib

who divide.PRES.3SG.REL for.2PL

‘Who (is it that) divides for you?’ (LL 11312)

(b) cid as dénti

what COP.PRES.3SG.REL to-be-done

‘What (is it that) is to be done?’ (Wb 12d41)

Both sets of interrogative pronouns in Old Irish formally correspond to those found in

many other IE languages, for example Vedic kas, Latin quis, Hittite kuiš ‘who’

(Fortson 2004: 130). This suggests that they descend from the reconstructed PIE

interrogative pronoun *kwo-/*kwi-. Of the two classes, Bergin (1938: 205) observes

that the unstressed variant is likely to be older. This seems plausible. The stressed


167

variant, as discussed in chapter 2, is best analysed as a cleft construction. Clefting is

the main strategy employed in Old Irish for topicalising or focussing. Newton (2006)

argues that the clefting construction developed as a result of the increase in verb

fronting. When verb fronting increased in frequency, it became impossible to front

any other constituent to initial position, and so a new strategy developed in order to

fulfil the original discourse function of fronting, namely clefting.56 It seems likely that

the development of wh-clefts occurred as part of this general development.

Let us turn our attention to the unstressed interrogative pronouns. As the older

variant it seems reasonable to suggest that these pronouns developed directly from the

PIE interrogative pronoun. As discussed in section 3.2, Hale (1987, 1995) proposes

that in PIE interrogative and relative pronouns formed a single class, appearing in the

inner position of the left periphery. If this were true for Irish, we would expect the

interrogative pronouns to have developed in the same way as the relative pronouns,

i.e. becoming second position enclitics. However, this is clearly not the case, as

interrogative pronouns appear in clause-initial position. There are two possible

explanations for this. First, it is perhaps possible that verb fronting and interrogative

clauses were simply not compatible. This would mean that if an interrogative pronoun

filled the inner position of the articulated CP, no element could move to the outer

position. As a result, the interrogative pronoun could not be analysed as a second-

position clitic, even if it became phonologically reduced, as there would be no

possible host to its left. The second option is that interrogative pronouns may have

been topicalised or focussed and so appeared in the outer position of the left

periphery.

The advantage of the first option is that under this analysis the interrogative

pronoun appears in the position we would expect, namely the inner peripheral

position that becomes the C position. Furthermore, the fact that the interrogative

pronoun is uninflected in the same way as was proposed above for the relative

pronoun, suggests that they may have undergone the same type of change, i.e. been

grammaticalised from specifiers to heads. However, under this analysis, the idea that

interrogatives cannot co-occur with a topicalised or focussed constituent is pure

stipulation.

56 See Garrett (1994) who proposes that a similar development took place in Lycian.


168

If we consider the second option, the interrogative pronoun is not in the position

we might expect, i.e. the inner peripheral position; however, this is not necessarily a

problem. The idea that the interrogative pronoun might be focussed receives some

cross-linguistic support. First, parallels are often drawn between wh-movement and

focussing (Kiparsky 1995a, Rizzi 1997, Sabel 2000). If wh-movement is a sub-type of

focussing, then this suggests that wh-elements may be emphatic, and so the possibility

of them occupying initial position seems plausible. Furthermore, there is comparative

evidence to suggest that certain types of wh-word can be fronted to Top/FocP. In

Hittite, wh-words tend to appear in second position; however, indefinite relative

pronouns appear clause-initially (Held 1957). Hale (1987) argues that the reason for

this is that, unlike other types of wh-word, indefinite relatives in Hittite are topicalised

and so appear clause-initially. It is possible that interrogative wh-words in Old Irish

could behave like indefinite relatives in Hittite. There is also evidence from within

Irish that supports the idea that interrogative pronouns in pre-Old Irish were fronted to

Top/FocP. As discussed above, it seems likely that clefting developed in Old Irish to

take over the function of the Top/FocP. With the increase in verb fronting, this

position could no longer be used to topicalise or focus any other constituent. The cleft

construction developed to take on this function. The fact that stressed interrogative

pronouns appear in a cleft construction supports the idea that interrogative pronouns

were originally in Top/FocP. When Top/FocP lost its pragmatic force, and the

interrogative pronouns that filled it were reanalysed as C heads, the wh-cleft

construction developed to replace the old relative construction.

In sum, it seems that it is possible to explain how the Old Irish interrogative

pronouns developed from PIE in two different ways. At this stage, perhaps the second

option, whereby the interrogative pronoun was fronted to Top/FocP, and was

reanalysed as a C element in the same way as fronted verbs and preverbs, is

preferable.

6.3 Conjunctions

Thurneysen (1946: 547–8) provides a list of 57 conjunctions found in the Old Irish

glosses. Of these, only three are conjunct particles, namely aran ‘in order that’, dian

‘if, when’ and con ‘so that, until’. It is only these three particles that appear in clause-

initial position and are followed by dependent forms of the verb; therefore, it seems


169

that it is only these three particles that appear in the C position. This section aims to

consider how these complementizers developed.

Thurneysen (1946:298–9) argues that aran, dian and con all developed from the

combination of a preposition and the relative particle/demonstrative an. So, for

example, dian- ‘if, when’ seems to have developed from the combination of the

preposition di and an, meaning perhaps ‘of/from that’. Similarly, aran ‘in order that’

was in origin the adverb/preverb ar ‘for’ plus an, and con ‘so that, until’ came from

the preposition co ‘to’. Thurneysen (1946: 299) suggests that originally these

prepositional phrases (PPs) were part of the main clause, later becoming reanalysed as

part of the embedded clause. This proposal receives cross-linguistic support from the

development of complementizers in other languages. It has been argued that the

complementizer that in English developed diachronically from the demonstrative that

(Kiparsky 1995a, Ferraresi 1997, van Gelderen 2005). The change in status of that

can be seen as a case of rebracketing. Initially, the demonstrative that appeared as the

complement of the verb of the main clause and referred to a clause in apposition, as

shown in (55) below. Such clauses were then subject to reanalysis, whereby that was

reanalysed as being part of the subordinate clause.

55. I know thati [the earth is round]i > I know [that the earth is round]

This type of rebracketing is quite common. The complementizer for seems to have

developed in a similar way, initially being part of the main clause, and being

reanalysed as the complementizer of the infinitival clause (see Fischer et al 2000 for

more details of this change):

56. It is good [PPfor me] [CPto go] > It is good [CPfor [me to go]]

It is easy to see how this could be applied to the Old Irish case, with the PP containing

the demonstrative being part of the main clause and referring to the second clause,

and then being reanalysed as part of the second clause.

For relative PPs to be reanalysed as complementizers in the way Thurneysen

suggests, the PP must be clause final in order for there to be ambiguity as to which

clause it belongs in. This suggests that at this stage, Irish cannot have been

predominantly verb-final. There are two possible scenarios here: either the


170

development of conjunct particles followed the shift to verb-intial word order or at

this stage, Irish was perhaps verb-medial. The first of these options is not appealing. If

the shift to verb-intial order occurred before the development of conjunct particles,

then it becomes difficult to explain why conjunct particles and initial verbs with

absolute inflection are in complementary distribution. It seems unlikely that conjunct

particles would be analysed as occupying the C position if this position were already

filled by the verb or a preverb. This being the case, at the stage when conjunct

particles developed, before the shift to verb-intial word order, Irish must have been

verb medial. We will return to this proposal in more detail in chapter 5.

In addition to being reanalysed as belonging to the embedded clause rather than the

main clause, the change from relative PPs to complementizers involves a further

change from a phrase to a head (see Roberts & Roussou (2003) and van Gelderen

(2005) where it is argued that that underwent a similar change from specifier to head

in the history of English). On theoretical grounds it seems that the rebracketing and

the change from specifer to head could have occurred in either order. However, there

is reason to believe that the conjunct particles may have gone through a pre-

complementizer stage, where they were syntactically part of the embedded clause, but

were still phrasal categories. As noted above, Old Irish has a number of conjunctions

that do not seem to appear in the C position. These particles are followed by

independent rather than dependent verb forms and cannot act as hosts for enclitic

object pronouns. This suggests that they cannot fill the C position, and must either fill

a higher position in the clause, or be clause external (cf. the German conjunction denn

‘as’). However, it seems unlikely that these non-conjunct particles should be part of a

preceding clause. In fact, in many cases non-conjunct particles occur in clause-initial

position, as can be seen in the examples in (57).

57. (a) ma chomalnit a n-grád

if fulfil.PRES.3PL.ABS their orders

‘If they fulfil their orders’ (Wb 28c7)

(b) in tan do-rolaig día in n-úaill do-rigni…

when PVB-forgive.PERF.3SG.DT god the pride PVB-do.PERF.3SG.DT

‘When God had forgiven him the pride he had shown…’ (Ml 50d15)


171

In these cases the particle clearly cannot be part of a preceding clause, suggesting that

the particle occupies Spec-CP.57 The existence of clausal connectives in Old Irish that

do not occupy the C position suggests that it is possible that conjunct particles also

went through a similar stage, whereby they occupied a specifier position in the left

periphery, before being reanalysed as C heads.

The change of the conjunct particles from specifier to head can be viewed as a case

of grammaticalisation, whereby the PP undergoes both semantic and phonological

change. Initially the conjunction meaning associated with the PP will be a secondary

meaning, derivable from its literal meaning by pragmatic inference. Over time,

however, the pragmatic link between the new conjunction meaning and the literal

meaning will become opaque, and it will become possible to analyse them as distinct

lexical items, i.e. the original PP and a conjunction.58 In addition, phonological

change will obscure the internal structure of the PP, so that it is no longer clear that it

consists of two separate elements. Once the internal structure of the PP is opaque, this

reduces the evidence that it is a phrase, and reanalysis of the PP as a head becomes

possible.

Having the phonological and semantic qualities of a complementizer, however, is

not sufficient for being reanalysed as one; it is also necessary for certain syntactic

conditions to be met. In other words, for an element to be reanalysed as a

complementizer it cannot appear with any other elements that unambiguously fill the

C position. As discussed above, the fact that conjunct particles are in complementary

distribution with initial simple verbs and the initial preverb of deuterotonic

compounds suggests that they cannot have developed after the development of V-to-C

and preverb-to-C movement. However, it was argued in sections 4 and 5 above that

before the development of V-to-C movement, the verb or preverb (or the negative

particle) moved to the outer left-peripheral position in virtually every clause and C

was filled by enclitic clause typing particles, such as the main clause particle *es and

the relative particle *io. To explain the development of conjunct particles, then, we

must explain why they did not co-occur with fronted verbs or preverbs and establish

whether they co-occurred with C particles.

57 It is possible that conjunctions such as in tan ‘when’ may have originally been used as adverbial expressions. This being the case, we would not expect them to be part of the main clause, but adjoined to the embedded clause. 58 This change is perhaps comparable to the presumed development of English because from the PP by cause.


172

The fact that conjunct particles are in complementary distribution with initial verbs

and preverbs can be easily explained. In sections 4 and 5 above it was argued that the

reason that fronted verbs and preverbs appeared so frequently in initial position was to

mark the link between clauses. As subordinating conjunctions, the primary function of

conjunct particles is to link two clauses. Therefore, whenever a conjunct particle is

present this particle marks the link between clauses and so movement of the verb or

preverb to the left periphery is not necessary.59 It could also be the case, of course,

that these conjunctions, after the rebracketing, but before they were analysed as heads,

occupied the same position as fronted verbs and preverbs. This being the case, in

addition to the functional reason for their complementary distribution, there is also a

structural explanation.

An important question that arises here is when the development of conjunct

particles took place. It was argued above that the conjunctions cannot have become C

elements after the development of V-to-C movement, otherwise they would not be in

complementary distribution with fronted verbs. This leaves us with the following

options: either the conjunctions became C elements before the development of V-to-C

movement or both changes occurred at the same time. It was argued in sections 4 and

5 that the development of verb and preverb-movement to C was linked to the change

from an articulated to an unarticulated CP. Before this change, there was no

generalised C position. It was argued above that the inner position was essentially a

restricted focus position that could only be filled by clause typing particles, marking

clauses as relative (*io) or non-relative (*es). This being the case, we would not

expect the inner position at this stage to contain conjunctions, as clause linking was

not part of its function.60 In fact, according to the analysis proposed in sections 4 and

5, clause linking is a function associated with the outer position. Ideally then, from a

theoretical perspective we would expect the reanalysis of the conjunctions as C heads

not to take place until the development of an unarticulated CP. In this way the

development of complementizers can be linked to the development of a generalised C

position.

59 It could perhaps also be the case that the development of these new conjunct particles as a new strategy for linking clauses contributed to the reanalysis of the connective particles as clause typing particles by encroaching on their function. 60 Although this inner position contains *es which was a connective, it was argued in section 4.5 above that *es did not appear as head of the inner peripheral position until it was reanalysed as a clause typing particle.


173

Empirical evidence to support the idea that the conjunctions became C elements at

the same time as the development of a generalised C position is difficult to come by.

One construction that may be taken to support this claim comes from the use of

infixed pronouns after the conjunctions aran, dian and con. These conjunctions are

used with class C pronouns. Class C pronouns are used in relative contexts and are

characterised by a /d/. Watkins (1963) argues that this /d/ comes from a connective

cognate with Ancient Greek dè, but has come to play a relative function. On the

assumption that like the connectives *es and *kwe the connective *de was reanalysed

as a relative clause marker, appearing in the inner peripheral position, then for the

conjunctions aran, dian and con to become C heads, this former connective must have

changed from a clitic to an affix associated with infixed pronouns.61 This suggests

then, that the conjunctions aran, dian and con cannot have been reanalysed as C heads

until after the change in status of the clause typing particles from clitic to affix, i.e. at

the same time as the reanalysis of initial verbs and preverbs as C elements.

One final issue to be addressed in this section is why it was that only three

expressions became C heads; what prevented the reanalysis of other nominal and

adverbial expressions as complementizers? It seems unlikely that there will be a

single answer to this question. In some cases the expression in question may not have

satisfied the semantic and phonological conditions for being reanalysed as a head.

This is evidently the case for in tan ‘when’, literally ‘the time’. However, many of the

non-conjunct particles do seem to fulfil these conditions by the Old Irish period. For

example, the etymology of the particle ma ‘if’ is unclear, and phonologically it

resembles a head rather than a phrase. However, these conditions may not have been

met at the time when the changes in the C-system took place. Thurneysen (1946: 558)

notes that in the Early Irish Laws i-nneoch ma or neoch ma is often found in place of

ma (neoch is the dative singular of the indefinite pronoun ní ‘something, anything’).

The language of the Laws is typically archaic, and so this could plausibly reflect an

earlier stage of the language. If this construction was used when the shift to verb-intial

word order took place, then i-nneoch ma would have been clearly phrasal and so

could not have been reanalysed as a C head.

A further possibility is that whether or not an expression was reanalysed as a

complementizer could have been related to whether or not that expression co-occurred 61 The reanalysis of *de was somewhat different to that of the others as it became part of the infixed pronoun system rather than the verbal morphology.


174

with a fronted verb. It was argued above that once the meaning of aran, dian and con

changed such that their primary function was to link clauses, they would no longer

appear with fronted verbs or preverbs. It could be the case that these three expressions

were the only ones that underwent this change in meaning and function before the

changes in the C-system and the shift to verb-intial word order. Therefore, although

there are other non-conjunct particles that function primarily as conjunctions in Old

Irish, this may not have been the case when the CP was restructured.

6.4 Summary

The discussion in this section has of necessity been somewhat tentative; however, it

has shown that at first sight the development of conjunct particles seems to be

consistent with our account of the development of verb-intial and preverb-initial order

in pre-Old Irish. It is plausible that the negative and interrogative particles and the

conjunctions that appear in the C position in Old Irish could have originated in the

outer position of an articulated CP, and been reanalysed as C heads as a result of the

loss of the articulated CP and the development of a generalised C position.

7. CONCLUSION

This chapter has provided a new account of the development of the Old Irish double

system of verbal inflection. Section 2 outlined the two main existing philological

accounts of the development and argued that on conceptual grounds Cowgill’s

particle theory is preferable to McCone’s suffix-/infix-deletion theory. However, a

major problem with the particle theory is that it does not explain the shift to verb-

intial word order, and the link between the syntactic position of the verb and its

morphological form. The aim of the remainder of the chapter was to solve this

problem. Having outlined the main syntactic characteristics of the verbal system in

PIE in section 3, sections 4 and 5 went on to suggest how this PIE system may have

developed into that of pre-Old Irish. Section 4 concentrated on the development of V-

to-C movement in pre-Old Irish, arguing that it this was linked to the loss of the

articulated CP and the development of a generalised C position. Section 5 provided

the beginnings of an account of the development of compound verbs, arguing that

preverb-to-C movement developed in the same way and as a result of the same

conditions as V-to-C movement. In section 6, we turned our attention to conjunct


175

particles, and attempted to show that their development can be seen to be parallel to

that of initial verbs and preverbs, in that it is linked to the changes in the C-system.

In addition to linking the development of V-to-C movement with changes in the C-

system, it was argued in section 4 that V-to-C movement is also linked to the

development of absolute verbal morphology. As discussed in chapter 3, the

relationship between verb-movement and verbal morphology has received a lot of

attention within both diachronic and synchronic generative work. In particular the loss

of verb-movement is often linked to the loss of verbal inflections (Roberts 1993;

Vikner 1995). An interesting question with regard to the Irish case, then, is to what

extent the development of the new verbal morphology led to the development of V-to-

C movement.

It seems that the development of the new verbal morphology was necessary for the

development of V-to-C movement. Without the reanalysis of the former connective

*es as a verbal affix rather than a clitic, the verb could not have been analysed as

appearing in the C position. However, although this development was necessary it

was not sufficient. As discussed in section 4, a number of other changes were

necessary for the development of V-to-C movement, most notably a significant rise in

the frequency of pragmatically motivated verb fronting. So, it seems that the link

between the development of verb-movement and the development of morphology is

by no means causal.

A further interesting question is whether V-to-C movement could have arisen

without the development of new C-based verbal morphology. It could be argued that

it was the presence of these C-based inflections on the verb that linked the verb to the

C position. If the clause-typing particles had not been present then there would have

been no link between the fronted verb and the C position. As a result, when verb

fronting became reanalysed as obligatory it would perhaps have been associated with

the T position as a result of its tense and agreement inflections. Viewed in a slightly

different way, the C-based inflections could be used as a cue during language

acquisition to show children that the verb is in the C position. However, as discussed

in chapter 3, the status of morphological cues is somewhat controversial. In the case

of V-to-T movement, languages that show a morphological cue also have a syntactic

cue. If we can show that in Irish the only cue linking the fronted verb to the C position

is the morphological one then this will provide evidence for the validity of

morphological cues.


176

With the development of conjunct particles, as discussed in section 6, there is some

syntactic evidence that the verb moves to C rather than T. Once conjunct particles are

analysed as C elements, i.e. as particles merged in the C position, then the fact that the

fronted verb is in complementary distribution with these elements provides evidence

that the fronted verb targets the C position rather than T. However, it is not entirely

clear whether this evidence can constitute a cue for acquisition. We will return to this

issue in the next chapter, when we consider the loss of V-to-C movement and the loss

of the double system of verbal morphology.

CHAPTER 5

THE LOSS OF THE DOUBLE SYSTEM OF VERBAL INFLECTION

1 INTRODUCTION

One of the main differences between Old and Modern Irish is the reduction in the

complexity of the verbal system. One significant way in which the complexity has

been reduced is through the loss of the double system of verbal inflection. In Modern

Irish all verbs in all persons and tenses have only one set of inflectional endings, used

both when the verb is in absolute initial position and when the verb is preceded by a

conjunct particle. The table below shows the forms of the verb mol ‘praise’.1

Table 1: Conjugation of the regular verb mol ‘praise’ in Modern Irish

PRESENT FUTURE PAST CONDITIONAL IMPERFECT

1sg molaim molfaidh mé mhol mé mholfainn mholainn

2sg molann tú molfaidh tú mhol tú mholfá mholtá

3sg molann sé/sí molfaidh sé/sí mhol sé/sí mholfadh sé/sí mholadh sé/sí

1pl molaimid molfaimid mholamar mholfaimis mholaimis

2pl molann sibh molfaidh sibh mhol sibh mholfadh sibh mholadh sibh

3pl molann siad molfaidh siad mhol sibh mholfaidís mholaidís

Pass moltar molfar moladh mholfaí mholtaí

(McGonagle 1988)

The loss of the double system is also apparent in the Old Irish compound verbs.

Between Old and Modern Irish all formerly compound verbs have been simplified,

and follow the same inflectional patterns as other simple verbs, as shown by the

present tense paradigms for the simple verb mol ‘praise’ and the formerly compound

verbs tabhair ‘give’ (OIr do-beir), déan ‘do, make’ (OIr do-gní) and feic ‘see’ (OIr

ad-cí).

1 These endings differ slightly between dialects, with Munster retaining more synthetic forms. See Ó Siadhail (1991) for detail. Crucially, though, none of the Modern Irish dialects retain a distinction between absolute and conjunct morphology. Scots Gaelic, on the other hand, does retain such a distinction but only in the future tense, where we find the -(a)idh ending in independent verbs and the -ø ending in dependent forms.

CHAPTER 5 – THE LOSS OF THE DOUBLE SYSTEM

178

Table 2: The present tense of mol ‘praise’, tabhair ‘give’, déan ‘do’ and feic ‘see’

1sg molaim tugaim déanaim feicim

2sg molann tú tugann tú déanann tú feiceann tú

3sg molann sé/sí tugann sé/sí déanann sé/sí feiceann sé/sí

1pl molaimid tugaimid déanaimid feicimid

2pl molann sibh tugann sibh déanann sibh feiceann sibh

3pl molann siad tugann siad déanann siad feiceann siad

Pass moltar tugtar déantar feictear

Only one trace of the Old Irish double system remains in Modern Irish. There is a

class of around ten irregular verbs that have different independent and dependent

forms. For example, the verb feic ‘see’ has the past tense form chonaic in initial

position (1c), but fhaca/bhfaca in dependent position, as shown in (1a) and (1b).2

1. (a) Ní fhaca tú mo nighean

Neg see.PAST.DEP you my daughter

‘You didn’t see my daughter’

(b) An bhfaca tú mo nighean?

Int see.PAST.DEP you my daughter

‘Did you see my daughter?’

(c) Chonaic tú í

see.PAST.IND you her

‘You saw her’ (McCloskey 2001b: 79)

There is little doubt that these different forms are suppletive in Modern Irish, with the

presence of particular C elements conditioning their insertion (McCloskey 2001b: 78).

Duffield (1995: 181) explains the appearance of the two forms in terms of f-selection.

If the T position is f-selected by a complementizer or negation, then the dependent

form is inserted. If the T position is unselected by any such C head, then the

independent form will appear. In current minimalist terms the appearance of the

different forms could be explained in terms of C-T feature spreading, whereby certain

specifications of C (e.g. negative, interrogative and certain complementizers) transfer 2 The distinction between the forms in (1a) and (1b) is a result of the different particles. The negative particle lenites, giving fhaca< faca, whereas the interrogative particle nasalizes, giving bhfaca<faca.


179

a feature to T that conditions the insertion of the dependent form. Crucially, however,

the use of independent and dependent verbal forms in Modern Irish is an irregularity,

an idiosyncratic feature of a handful of verbs. It is by no means a productive system,

and so is of little interest from a syntactic perspective.

The aim of this chapter is to explore how the double system of inflection of the Old

Irish period developed into the system we find in Modern Irish, in other words, how

the double system of verbal inflection was lost. In chapter 4 it was argued that in pre-

Old Irish the double system was syntactically productive, i.e. it depended on the

syntactic position of the verb. In chapter 2 it was argued that in Classical Old Irish the

verb only moves as far as T, and so the system cannot be syntactically productive.

Section 2 examines the change between pre- and Classical Old Irish, namely how the

double system was lost as a syntactically productive system, and how this can be

linked to what Watkins (1963) terms ‘univerbation’. In the Middle Irish period the

deuterotonic/prototonic distinction broke down as a result of the simplification of

compound verbs. This is the topic of section 3. The final stage in the loss was the loss

of the absolute/conjunct distinction during the Middle and Early Modern periods. This

is examined in section 4. Section 5 concludes the chapter.

2 UNIVERBATION: THE LOSS OF THE DOUBLE SYSTEM IN THE SYNTAX

2.1 Introduction: what is univerbation?

The process of univerbation is the coming together of two (or more) parts of a word to

form a single unit. It seems to be a common process amongst all the early IE

languages. For example, in Archaic Latin we find examples such as sub vos placo

corresponding to Latin vos supplico ‘I beseech you’ (see Vincent 1999: 1119).

Similarly in Homeric Greek we find non-univerbated examples such as those in (2a)

alongside univerbated forms as in (2b). By Classical Greek only the latter survives.

2. (a) hôs ephat' Atreïdês, epi d' êineon alloi Akhaioi So spoke Atreus PVB PTC approved all Acheans

‘So spoke Atreus and all the Acheans shouted assent

(Il 3. 461/McCone 1979a: 253)

(b) Hektori men gar epêinêsan

Hector.DAT on the other hand for praised.3PL

‘to Hector on the other hand they gave praise’ (Il 18.312)


180

The process that Watkins (1963) describes as univerbation in Irish is somewhat

different. Between pre- and Classical Old Irish there is a change in the position of the

verb, such that the verb always appears in initial or near-initial position regardless of

whether there is a conjunct particle or an initial preverb in initial position. The

development of #CV…# orders in place of #C…V# orders is not strictly univerbation

as the conjunct particle and the verb do not form a single unit (except perhaps in the

phonology as the conjunct particle is proclitic). The development of #PV…# orders in

place of #P…V# orders is perhaps closer to the strict definition as the preverb and the

verb form a single semantic unit. However, in Classical Old Irish the preverb and the

verb are still syntactically and phonologically distinct. Infixation is still possible

between the verb and the preverb and compound verbs do not show initial stress,

which we would expect of a single phonological unit. True univerbation of this kind

does not occur until the Middle Irish period when compound verbs are simplified. We

will return to this in section 3.

It seems, then, that what Watkins refers to as univerbation is not in fact

univerbation in the strictest sense; it is better characterised as a syntactic change,

resulting in the generalisation of verb-initial or near-verb-initial word order. Let us

consider this change in theoretical terms.3 It was argued in chapter 4 that pre-Old Irish

resembles the system proposed by Carnie, Harley & Pyatt (2000), having a filled C

condition that forces movement of either simple verbs or the initial preverb of

compound verbs to C in the syntax when C is otherwise unfilled. In chapter 2 it was

argued that this is no longer the case in Classical Old Irish, where simple verbs only

move as far as T and preverbs do not move in the syntax. Furthermore, whereas in

pre-Old Irish the verb remains within the vP when C is filled, in Classical Old Irish

this is not the case. The verb moves to T across the board, irrespective of whether C is

filled. The differences between the two periods can be seen in the trees in (3–5)

below, where the pre-Old Irish situation is shown on the left and Classical Old Irish

on the right. Example (3) shows the difference in the structure for simple verbs,

example (4) for compound verbs, and example (5) shows the structure for a clause

containing a conjunct particle.

3 For a summary of Watkins’ (1963) account of univerbation and the problems with it see the discussion of McCone’s suffix/infix-deletion theory in chapter 4, section 2.3.


181

3. CP CP

C TP C TP

verb T vP T vP

verb v VP v VP

V DP V DP

4. CP CP

C TP C TP

pvb pvb T vP T vP

verb v VP v VP

V DP V DP

verb 5. CP CP

C TP C TP

Conj Conj T vP T vP

(pvb)verb v VP v VP

(pvb) V DP V DP

verb Two significant changes have occurred between pre- and Classical Old Irish. First, V-

to-C movement has been lost and replaced by V-to-T movement, leading to the

‘univerbation’ of simple verbs and conjunct particles in clause-initial position.

Secondly, verb-movement to light v has been generalised. In pre-Old Irish it seems

that V-to-v movement only takes place when light v is empty, i.e. when the verb is

simple. In Classical Old Irish the verb begins to move to light v across the board,

regardless of whether there is a preverb in light v. Between pre- and Classical Old

Irish there seems to have been a change in the status of the preverbs. Whereas in pre-

Old Irish they are syntactically independent and so can block verb-movement to light

v, in Classical Old Irish this is no longer the case.


182

In terms of the approach to syntactic change outlined here, the phenomenon that

Watkins calls univerbation in Irish seems to be the result of two independent changes,

namely a parametric change, whereby V-to-C movement was reanalysed as V-to-T

movement, and a change in the syntactic status of preverbs. Sections 2.2 and 2.3

consider each of these changes in turn. Although these changes account for the

syntactic differences between pre- and Classical Old Irish, there is also a crucial

morphological difference, specifically in the way that the double system of verbal

inflection is derived. Whereas in pre-Old Irish the different forms are syntactically

derived and depend on the verb and preverb occupying different syntactic positions,

in Classical Old Irish this is no longer the case and the different forms result from

post-syntactic operations. Section 2.4 considers how and why these post-syntactic

operations may have arisen.

2.2 ‘Univerbation’ within the clause: generalised V-to-T movement

It was argued in chapter 2 that there is good evidence to suggest that in Classical Old

Irish the verb only moves as far as T. This means that if, as proposed in chapter 4,

pre-Old Irish had V-to-C movement, V-to-C movement must have been lost and

replaced by V-to-T movement at some point before the attested Old Irish period.

From a theoretical perspective the reanalysis of V-to-C movement as V-to-T

movement is fairly straightforward. Under current minimalist assumptions (Chomsky

2005, 2006), T does not have any of its own features. All of T’s tense and subject

agreement (φ) features are given to T by the C head. Under this view, C and T could

be seen as a complex category (Biberauer 2005). Instead of an Agree relation between

T and V, then, we have an Agree relation between the C-T complex and the v-V

complex. As a result, for V-to-C movement to be reanalysed as V-to-T movement, we

do not need to postulate a new Agree relation. The C-T complex still Agrees with v-V

as before; the only difference is the position of the feature triggering movement.

Whereas in pre-Old Irish the Affix-feature is associated with C, in Classical Old Irish

it becomes associated with T.

The replacement of V-to-C movement by generalised V-to-T movement is also

reasonably common cross-linguistically. Old English, Old French and Middle Welsh

are all argued to have V-to-C movement, which is lost and replaced by V-to-T

movement during the course of their history (see Fischer, van Kemenade, Koopman

& van der Wurff (2000) and the references therein for English; Roberts (1993) for


183

French; Willis (1998) for Welsh). So it seems, then, that the idea that V-to-C

movement was reanalysed as V-to-T movement in Irish receives both theoretical and

cross-linguistic support. The next task is to explain why this change might have taken

place.

As discussed in chapter 3, the abstract syntactic features that drive movement are

acquired during language acquisition on the basis of evidence in the linguistic data. In

order for V-to-C movement to be reanalysed as V-to-T movement, cues for V-to-C

movement must become obscured and cues for V-to-T movement must develop. In

the documented cases of the loss of V-to-C movement in English, French and Welsh

the loss of V-to-C movement is part of the loss of V2. As discussed in chapter 3,

Lightfoot (1999) argues that the cue for the V2 parameter is clauses of the type XVS,

where some element other than the subject precedes the verb, and the subject appears

post-verbally. A decrease in the frequency of XVS clauses will lead to the loss of the

cue for V2, and so V2 will be lost. This seems to have been the case in both English

and French where there was an increase in subject-initial clauses at the expense of the

other types before V2 was finally lost (see Roberts 1993: 142–186 for French, and

326–332 for English).4

The loss of V-to-C movement in Irish clearly differs from the other attested cases

as it is not related to the loss of V2. V-to-C movement in pre-Old Irish cannot have

been cued by XVS orders, and so we need a different cue and a different explanation

for the loss. In chapter 4 it was proposed that there are two possible cues for the

acquisition of V-to-C movement in Irish. First, there is the morphological cue, the C-

based verbal morphology linking the verb to the C position. Secondly, there is the

syntactic cue, the fact that a fronted verb does not co-occur with elements that are

merged in the C position, i.e. conjunct particles. Let us examine each of these in turn.

2.2.1 The loss of final syllables and the loss of V-to-C movement

As we saw in chapter 3, it is often argued that the loss of verbal morphology can lead

to the loss of verb-movement. The most widely cited case is that of the loss of verbal

inflection and the loss of V-to-T movement in English (Roberts 1985, 1993). Under a

4 The situation in Welsh is somewhat different. Willis (1998: 184–189) argues that although the loss of V2 was in part triggered by a decrease in object topicalisation, the most significant factor was the loss of preverbal particles. These preverbal particles appeared in the C position, and so any constituent that appeared to their left had to be analysed as appearing in Spec-CP. When these particles were lost through phonological change this evidence disappeared.


184

cue-based theory of acquisition this correlation can be seen as the loss of a

morphological cue leading to a failure to acquire the associated movement process.

When the verbal endings are lost, the morphological cue linking the verb to the T

position is lost and so evidence for V-to-T movement disappears.5

In chapter 4 it was argued that V-to-C movement developed in pre-Old Irish in

conjunction with the development of new C-based morphology. This clause-marking

suffix, Cowgill’s *es particle, linked the initial verb to the C position, and so acted as

a cue during acquisition, showing that the raised verb was in the C position rather

than the T position. Particle theorists (Cowgill 1975) argue that this particle *es was

lost in the fifth century as part of the process of apocope that affected the language as

a whole at this time (see McCone 1996 for details of this change), and this is why it is

not found in Old Irish. According to a cue-based theory of language acquisition, if

this morphological ending that links the verb to the C position is lost, then one

possible cue for V-to-C movement is obscured and so we might expect V-to-C

movement to be lost.

One problem that arises from the attempt to link the loss of V-to-C movement to

the loss of verbal morphology is that although the particle *es was apocopated, it left

behind the absolute endings.6 Even after the loss of the *es particle, the absolute

endings were clearly distinct from the conjunct. Why was this not sufficient to link

the verb to the C position? One possible answer is that for inflectional morphology to

act as a cue for acquisition of movement to a particular functional head, there must be

a clear link between the form of the ending and the function associated with the

functional head. If, as argued in chapter 4, the particle *es marked declarative main

clauses there would have been a clear correspondence between the form *es and the

function of clause typing. Clause typing is typically associated with the C head, and

so the particle *es would have clearly been the realisation of a feature associated with

5 Although, as discussed in chapter 3, the loss of verbal morphology can at best be seen as a contributing factor to the loss of V-to-T movement, as verb-movement was not lost in English until the loss of syntactic evidence for V-to-T movement. 6 David Adger (p.c.) suggests that relative endings may also have provided sufficient morphological evidence for V-to-C movement. These endings have survived in the Modern Goidelic languages and so it seems likely that they were fairly robust at this earlier stage. However, as we saw in chapter 2, the special relative endings are restricted in terms of distribution, only occurring with 3rd person and 1pl forms of simple verbs. It is perhaps possible at this stage of the language, when there were large numbers of compound verbs, that the special relative endings would not have been sufficiently frequent to maintain V-to-C movement.


185

C.7 However, once the *es particle was lost as a result of apocope, this link between

form and function was lost. The absolute endings remained but the main function of

these endings was to mark subject-verb agreement, a property of the T head rather

than the C head. The functional link between the verbal endings and the C position

disappeared; therefore the morphological cue was lost.

If we assume, then, that the morphological cue is the only evidence that the initial

verb is in the C position rather than the T position, then when this cue is lost, there is

no longer any evidence that the verb is in C. However, the verb in initial position will

satisfy the cue for V-to-T movement, i.e. it appears to the left of the subject and VP-

adverbs. The loss of the morphological cue, then, leads to a loss of evidence for V-to-

C and an increase in evidence for V-to-T resulting in a shift in parameter settings.8

A further issue to address is whether the loss of the morphological cue was the

only factor that played a role in the loss of V-to-C movement. If the acquisition of V-

to-C movement relied solely on the morphological endings of the verb, then we would

expect the loss of these endings to result in the reanalysis of verb-initial orders from

V-to-C to V-to-T movement. However, as we observed in chapter 4 there is also a

syntactic distinction between movement to C and movement to T. Verb-movement to

C only takes place when C is not otherwise filled by a conjunct particle. As a result, it

is possible that examples of the type in (6), where the verb clearly remains in situ in

the vP when C is filled, alongside examples as in (7) where the verb has moved, could

provide unambiguous evidence that the verb moves to C and not to T.

6. #C (S)(Adv)(O)V#

7. #V(S)(Adv)(O) #

The next section examines whether this is the case.

7 This is perhaps not so clear within the current theory whereby features are shared by C and T. However, even in this scenario there remains a sense in which features are inherently associated with only one position. This is especially true for traditionally C-based features such as force and finiteness, which do not tend to appear on T alone. 8 Another possible way to view this reanalysis is as a structural simplification in the sense of Roberts & Roussou (2003). If we assume that a language learner postulates the least amount of structure consistent with the evidence, then, V-to-C movement will only be postulated when there is robust, specific evidence for it. If this evidence is not present then the learner will postulate less structure, namely V-to-T movement.


186

2.2.2 Syntactic evidence and the loss of V-to-C movement

Although root/embedded asymmetries such as that shown in (6) and (7) provide clear

evidence for the linguist that a fronted verb has moved to C, this does not necessarily

mean that this evidence is also used by children. This evidence is quite unlike all the

other syntactic cues that have been proposed so far. The cues for V-to-T movement

and V2 seem to involve the identification of linear strings, e.g. verb-adverb or verb-

subject orders for V-to-T and XVS orders for V2. Once children can identify

categories and constituents, identifying these strings will be reasonably

straightforward. It seems likely that identifying that a fronted verb does not co-occur

with a complementizer would be a more complex procedure. To do so, children would

have to compare multiple clauses to see that complementizers and fronted verbs are in

complementary distribution. Perhaps more problematically, to identify that two

elements are in complementary distribution a child must utilise negative evidence,

namely that two elements cannot co-occur. Lightfoot (1991, 1999) argues that

negative evidence cannot make up part of the trigger experience. Children have no

access to information about what is ungrammatical. They must set their parameters on

the basis of structures they hear, not structures they do not, as they cannot conclude

that a construction is ungrammatical simply because they do not hear it.

From a Lightfootian perspective the fact that complementizers and fronted verbs

are in complementary distribution is also problematic as a cue for parameter setting,

as it involves information from embedded domains. Lightfoot (1991) argues that

children are degree-0 learners. This means that they make use only of evidence from

main clauses (plus ‘a bit’ at the edge of the embedded clause) during language

acquisition. As complementizers only tend to occur in embedded clauses, children

would have to utilise evidence from embedded clauses to establish that

complementizers and fronted verbs cannot co-occur.9

It seems, then, that there are good reasons for rejecting root/embedded

asymmetries as a cue for the acquisition of V-to-C movement. However, it is

interesting to note that in all the languages where V-to-C movement has been lost,

there seems to have been evidence that would undermine such a cue. In other words,

in Old English (8), Old French (9) and Middle Welsh (10) the verb (bold) did not

appear strictly in final position as it does, for example, in Modern German (11). 9 This is perhaps not such an issue for Old Irish, as Old Irish has a number of conjunct particles that appear in the C position in main clauses too.


187

8. Þæt he mot ehtan godra manna

that he might persecute good men

‘That he might persecute good men’

(Wulfstan’s Homilies 130.37–8; Biberauer & Roberts 2005: 18)

9. celui jor qui vint Nostre Sires en la cité de Jerusalem

that day that came our Lord into the city of Jerusalem

‘That day that Our Lord came into the city of Jerusalem’

(Vance 1988: 94; Roberts 1993: 98)

10. a chyt archo ed yti rodi yr eil, na dyro

and though implore.SUBJ he to.you give.VN the second NEG give.IMPV

‘And though he implore you to give him the second, do not give (it)…’

(PKM 3 19–20; Willis 1998: 53)

11. Ich weiß, daß Johann gestern ein Buch gekauft hat

I know that John yesterday a book bought has

‘I know that John bought a book yesterday’

These examples demonstrate that in Old English, Old French and Middle Welsh there

was possible evidence for verb-movement in embedded clauses, i.e. when C was

filled, thus undermining the root/embedded asymmetry. Willis (1998) argues that this

was the case in Middle Welsh, with the verb consistently moving to T in embedded

clauses. In Old English and Old French, on the other hand, the position of the verb is

variable, and so the postulation of V-to-T movement is not so clear.

Roberts (1993) suggests that many clauses that seem to show embedded V2 in Old

French are in fact examples of free inversion. Free inversion is a feature of null

subject languages, such as Italian, where the subject can appear post-verbally:

12. (a) É arrivato Gianni

is arrived Gianni

‘Gianni has arrived’


188

(b) Hanno telefonato molti studenti

have.3PL telephoned many students

‘Many students have phoned’

Old French was also a null subject language, and so it seems plausible that a number

of examples, such as that given in (9) above (repeated as (13) below), where the

subject follows the verb, could be analysed in this way.

13. celui jor qui vint Nostre Sires en la cité de Jerusalem

that day that came our Lord into the city of Jerusalem

‘That day that Our Lord came into the city of Jerusalem’

(Vance 1988: 94; Roberts 1993: 98)

Biberauer & Roberts (2005) argue that many of the English examples where the verb

is not in final position in the embedded clause can be analysed as biclausal, as in the

example in (14).

14. [Þæt hi mihton] [swa bealdlice Godes geleafan bodian]

that they could so boldly God’s faith preach

‘That they could preach God’s faith so boldly’

(The Homilies of the Anglo-Saxon Church I 232; Biberauer & Roberts 2005: 17)

A third possibility for the variation seen in embedded clauses could be right-

dislocation. It was argued in chapter 4, following McCone (1979a, 1997b) that right-

dislocation may have been a feature of many of the early IE languages, in order to

provide an extra topicalisation or focus position at the right edge of the clause, or a

site for heavy subjects or objects. Right-dislocation could also result in object- or

subject-final clauses.

So far in this section, then, it has been argued on conceptual grounds that it is

unlikely that root/embedded asymmetries provide a cue for the acquisition of V-to-C

movement. However, empirically it seems that languages that have lost V-to-C in V2

contexts tend not to show strict root/embedded asymmetries. Crucially, unlike

Modern German, Old English, Old French and Middle Welsh were not strictly verb-

final in embedded clauses. On this basis it seems likely then that for V-to-C


189

movement to be reanalysed as V-to-T movement in Old Irish, pre-Old Irish cannot

have been strictly verb-final. There is some evidence to suggest that this was the case.

As we saw in chapter 4, Bergin (1938) provides a number of examples from

archaic Irish texts where the verb is not in initial position in the clause, such as those

given in (15) below.



‘Although I reap blistered seaweed’

(Corm 1059; Bergin 1938: 197)



‘He spoke to an angel’

(LU 1148; Bergin 1938: 201)

Watkins (1963) argues that such examples provide evidence to suggest that at an

earlier stage Irish was predominantly verb-final. Although the majority of Bergin’s

original examples seem to show a clause-final verb, Bergin himself makes no

reference to the verb appearing specifically in final position. In fact there are a

significant number of examples where the verb occurs in non-final position:

16. (a) Lugaid Luath loisc trebthu trén tuath

Lugaid swift burn.PRET.3SG.CONJ dwellings strong people

‘Lugaid the Swift burned the dwellings of the strong people’

(Corp Gen 5; Carney 1979: 432)

(b) bángluinn gní glenn gaeth

bloodless.deed do.PRES.3SG.CONJ valleys wind

‘The wind of the valleys does a bloodless deed’ (Corp Gen 6; Carney 1979: 433)

From this evidence Koch (1991) proposes that pre-Old Irish was a V2 language. This

idea was further developed within a generative framework by Doherty (1999, 2000),

as we saw in chapter 2. Although many examples of Bergin’s Construction seem to


190

follow the V2 pattern, with only one constituent preceding the verb, there are also a

number of cases where more than one constituent precedes the verb, as shown in (17).

17. (a) Mortlithi márlóchet di doínib dingbatar

Great.plagues great.lightnings from the people keep.PRET.PASS.PL.CONJ

‘Great plagues and great lightnings are kept from the people’

(AM §12)

(b) ar mind n-axal n-acallad

our hero apostle speak.IMPF.3SG.PT

‘Our hero used to converse with the apostle’

(ACC §82)

Doherty (2000) argues that such examples are not inconsistent with a V2 analysis, as

many other V2 languages also show V3 orders; however, this argument is not

convincing. Although V2 languages such as Old English, Old High German and

Kashmiri do show V3 orders, these deviations can in most cases be explained in a

principled way. For example, one of the fronted elements may be a clitic, as in (18) or

a left-dislocated element, neither of which ‘count’ for the purpose of V2 (Pintzuk

1999).

18. [hiora untrymnesse] [cl he] sceal rowian on his heortan

their weakness he shall atone in his heart

‘He shall atone in his heart for their weakness’

(Cura Pastoralis 60.17; Pintzuk 1999: 136)

Similarly in Kashmiri, V3 is only found in a single construction, namely wh-questions

which also contain a topic (Bhatt 1999).10

19. rajan kemis he:v nev kita:b

Raj whom showed new book

‘As for Raj, to whom did he show his new book?’

10 Kashmiri seems to differ from the Germanic V2 languages in that it has an articulated CP (Manetta 2006). This accounts for the fact that it allows more than one projection to precede the verb.


191

The cases of V3 found in Irish do not seem to be as principled as those found in true

V2 languages and so are less easy to explain as exceptions to V2.11

Eska (2002b) observes that in most cases of Bergin’s Construction the position of

the verb can be explained by reference to poetic or stylistic factors. In example (20)

below, moving the verb out of initial position achieves the effect of alliteration.

However the verb cannot be in final position as this would be incompatible with the

end rhyme.

20. Bángluinn gní glenn gáeth

bloodless.deed do.PRES.3SG.CJ valleys wind

Góeth di muir muiredach maéth

Wind from sea Muiredach gentle

‘The wind of the valleys does a bloodless deed;

Gentle Muiredach (is) a wind from the sea’ (Corp Gen 6.47–48)

In example (21) a final verb allows for the repetition of the initial syllable in two

adjacent words.

21. i cunn chomlán comeillgiter

in sense perfect put.together.PRES.PASS.PL.CJ

‘Which are put together in perfect good sense’ (DIAChor §23)

Eska concludes that, as the position of the verb in examples of Bergin’s Construction

seems to be determined by poetic devices, there is little we can tell about the clause

structure of pre-Old Irish. It is true that due to the poetic and highly stylised nature of

the evidence it is difficult to draw any firm conclusions from Bergin’s Construction.

However, if pre-Old Irish was strictly verb-final in the same way as Modern German

we would perhaps not expect to find these deviations, at least not to such an extent.

11 There is some evidence for V2 orders in other Celtic languages, namely Breton and Middle Welsh (Willis 1998). However, although these languages do display V2 word order patterns this type of V2 does not seem to be the same as that found in Germanic. For example, both Middle Welsh and Breton show VSO order in embedded clauses and have a number of sentence particles in main clauses that seem to appear in the C position. This suggests that the verb may not be in C in Celtic V2 languages as it is in Germanic. Although pre-Old Irish may have shown some V2 sentence patterns it is difficult to give this any theoretical significance as V2 does not seem to be a uniform phenomenon cross-linguistically.


192

This suggests that pre-Old Irish was not strictly verb-final and that there was some

variation in terms of which orders were allowed. Let us turn now to some comparative

IE evidence.

Watkins (1963) argues on the basis of evidence from other early IE languages that

pre-Old Irish was a verb-final language. We saw in chapter 4 that there is evidence

from the Hittite, Vedic, Ancient Greek and Latin to suggest that verb-final order

should be reconstructed as the unmarked word order for PIE. However, this does not

entail that strict verb-final order was maintained in Irish up until the development of

verb-initial order. In fact, as we saw in chapter 4, even within the earliest IE

languages deviations from verb-final order were possible, as shown in (22) for Vedic

and (23) for Homeric Greek.

22. tam eva tābhir āhutibhiḥ śamayitvorjaṃ lokānāṃ

him.ACC thus libations power having.appeased.ACC worlds.GEN

jayati yamaṃ devaṃ

wins Yama.ACC God.ACC

‘Having appeased him, the God Yama, with these libations, he wins the

power of the worlds’

(JB 7, 3–4; McCone 1997b: 370)

23. Trôas d' eklinan Danaoi

Trojans.ACC PTC turned Danaans.NOM

‘And the Danaans turned the Trojans’

(Il 5, 37)

It was argued in chapter 4 that these deviations from verb-final order seem to have

been stylistically marked, and so were most likely derived either through verb-

movement to the left periphery or movement of another constituent to the right edge

for emphasis. As there is evidence for heavy-NP-shift and other right-dislocation

processes in Modern Irish (McCloskey 1999), it does not seem unreasonable to

suggest that these existed earlier in the language. Furthermore, it is possible that with

the increase in frequency of verb fronting and the resulting changes in the C-system in

Irish, as outlined in chapter 4, a new strategy was necessary to take over from the

pragmatic fronting process that had been lost. The increase in verb fronting meant that


193

other elements could no longer be fronted for emphasis. As a result new strategies

were needed. It is possible, then, that the loss of pragmatic fronting could have also

led to an increase in the frequency of right-dislocation. If this was the case, then we

might expect an increase in frequency of object, subject or even adverb-final clauses

in pre-Old Irish.

So, it seems that we should expect at least some deviation from verb-final order in

pre-Old Irish and perhaps even more than in the other early IE languages as a result of

the loss of the option for pragmatic fronting.12 Furthermore, as a null subject

language, we might expect pre-Old Irish to have allowed free inversion, as seen in

Old French above. This would also have led to an increase in non-verb-final orders,

and more specifically to an increase in verb-subject orders. As we saw in chapter 3,

verb-subject orders could provide a syntactic cue for V-to-T movement. Therefore, an

increase in verb-subject (and also plausibly verb-adverb) orders would have increased

the evidence for V-to-T movement and contributed to its development.

In sum it seems that the change we are suggesting for Old Irish, namely the

reanalysis of V-to-C movement as V-to-T movement is well supported both

theoretically and cross-linguistically. It has been argued that the main motivation for

the loss of V-to-C movement was the loss of the clause typing particle *es, which was

lost during the process of apocope in the fifth century. After this particle was lost, the

fronted verb was no longer linked to the C position. A cue for the acquisition of V-to-

C movement was lost, and furthermore, a cue for V-to-T movement developed as

these fronted verbs now provided positive syntactic evidence for V-to-T movement,

as the verb appeared to the left of the subject and VP-adverbs. In this section we have

also examined the possibility of a syntactic cue for V-to-C movement. It has been

argued that the fact that a fronted verb is in complementary distribution with conjunct

particles (C elements) is unlikely to be a cue during language acquisition as it makes

reference to negative evidence and also information found in embedded clauses.

However, there does seem to be a correlation between the loss of V-to-C movement

and word order variation in embedded clauses. Unlike Modern German, Old English,

Old French and Middle Welsh do not show strictly verb-final orders when C is filled.

These three languages all allow some degree of variation in word order, which could

12 These pragmatic processes may have also led to a change from underlying OV to VO. We saw in chapter 2 that Old Irish is a VO language, whereas PIE is OV. See McCone (1979c) for details of such a change in Luwian, and Taylor (1990) for a similar change in Ancient Greek.


194

be analysed as verb-movement. It is possible that this is prerequisite for the loss of V-

to-C movement. There is evidence to suggest that pre-Old Irish also showed variation

in non-verb-initial clauses, thereby fulfilling this requirement.

An interesting question at this point is why this correlation between the loss of V-

to-C movement and variation in embedded orders should exist (if it is not purely

coincidental). One possible answer is that variation in embedded contexts, especially

cases of right-dislocation or free inversion where the subject appears to the right of

the verb, will provide positive evidence for the acquisition of V-to-T movement. In

Irish it seems that this evidence, in combination with that from the main clause after

the loss of the morphological cue for V-to-C movement, increases the evidence for V-

to-T movement, leading to a shift in the position of the Affix-feature.

2.3 ‘Univerbation’ within the vP: generalised V-to-v movement

The reanalysis of V-to-C movement as V-to-T movement, as outlined above, accounts

for part of the change between Pre- and Classical Old Irish. The development of V-to-

T movement in place of V-to-C movement ensures that the verb moves to initial

position regardless of whether the C position is filled. This means that whenever the

verb is simple, whether or not the C position is filled, the verb will move out of the vP

to T, resulting in the following orders:

24. (a) #V…#

(b) #CV…#

Although the development of V-to-T movement provides us with the attested orders

for simple verbs, the situation for compound verbs is less straightforward. The shift of

an Affix-feature from the C position to the T position will simply mean that whatever

is in v is attracted to T rather than C. This means that we would still expect tmesis

orders, and furthermore, we would also expect tmesis orders when C is filled, as the

initial preverb should be attracted to T:

25. (a) #P…V#

(b) *#CP…V#


195

However, the clause type shown in (25b) is unattested at any stage of Irish. This

suggests, therefore, that there must have been another change that occurred either

before or simultaneously with the development of V-to-T movement that prevented

this order from occurring. Crucially in Classical Old Irish both the verb and the

preverb appear in clause-initial or near-clause-initial position. This suggests that the

verb always moves through light v. Whereas previously the presence of a preverb

blocked V-to-v movement it seems that this is no longer the case, and so verb-

movement occurs regardless of whether a preverb is present or not.

The development of generalised V-to-v movement does not seem to be a

parametric change. For the verb to move to light v, light v must have an Affix-feature.

However, it was argued in chapter 4 that this was already the case in pre-Old Irish.

For simple verbs to move to the C position they must move via light v to escape the

phase, in accordance with the Phase Impenetrability Condition (PIC – Chomsky 2000,

2001). In pre-Old Irish, however, this movement only took place when light v was not

filled with a preverb. The crucial difference between pre- and Classical Old Irish,

then, is that in pre-Old Irish the preverb can fulfil light v’s Affix-feature and block V-

to-v movement, whereas in Classical Old Irish it cannot.

The change in the preverb’s ability to satisfy an Affix-feature reflects a change in

its syntactic status. One way to understand this change is as an example of

grammaticalisation. As we saw in chapter 4 preverbs started off as independent

adverbial phrases that had independent meaning and could be merged in a number of

different positions in the clause, modifying a noun or a verb or the clause as a whole.

With the changes in the C-system, preverbs were reanalysed from phrases to heads.

Instead of being merged as specifiers to vP, they became merged as v heads. This is

the first stage of the grammaticalisation. By the Classical Old Irish period preverbs

have no lexical meaning independent of the verb with which they are associated.

Instead, it was argued in chapter 2, they are verb-class markers, purely functional

elements. In Classical Old Irish preverbs are simply the realisation of a verb-class

feature, they are no longer an independent bundle of features merged in the v position.

So, it seems that preverbs have undergone a change from lexical to functional item,

and from XP to head to (valued verb-class) feature. As a head the preverb could

satisfy the Affix-feature, as simply a realisation of the light v feature it cannot. What

needs to be explained then is how and why this change took place.


196

The grammaticalisation of the preverb can be seen as a loss of features. As a head,

a preverb constitutes of a bundle of features. Grammaticalisation involves the loss of

these features, until only v’s basic functional features remain.13 The loss of lexical

features reflects the semantic bleaching that is typically involved in

grammaticalisation processes. The preverb is originally an independent lexical item

that can combine with a verb to produce different meanings. To start with these

meanings are pragmatically determined; however, over time they become codified,

the formerly pragmatic meaning becomes the literal meaning and the earlier literal

meaning is lost. From the non-compositional nature of compound verbs in Old Irish

(see chapter 2) it is clear that this process has taken place, and the preverb has lost its

semantic features.

It seems, then, that there is reason to believe that the semantic changes necessary

for grammaticalisation have occurred. However, the fact that these semantic changes

have taken place does not necessarily mean that a syntactic change will occur. We

saw above that particle verbs in German are semantically univerbated, but they still

consist of two syntactically independent components. Although preverbs have lost

their lexical features they will still be analysed as heads in the syntax if there is

positive evidence available to the acquirer to suggest that this is the case.

The clearest evidence that preverbs are independent syntactic heads is that they

move independently in the syntax. In pre-Old Irish preverbs move from v to C in the

syntax. Moreover, preverbs are never incorporated with a moved verb in pre-Old

Irish, as a verb can only move when v is empty, i.e. when there is no preverb.

Between pre- and Classical Old Irish there is reason to believe that both of these

pieces of evidence disappear. First, it was argued in section 2.2 that between pre- and

Classical Old Irish V-to-C movement was reanalysed as V-to-T movement. As the

Affix-feature associated with C was lost, this means that no movement to C could

take place. Therefore, although initial preverbs clearly occupy the C position, they

cannot move there in the syntax. As a result, this piece of evidence for the syntactic

independence of preverbs is lost.14

13 The last stage in the grammaticalisation process is from valued feature to zero. This is what happens in the simplification of compound verbs in Middle Irish, as we shall see in section 3. 14 Of course, we would expect with the shift in the Affix-feature that preverbs would simply start raising to T instead of C. However, this does not seem to be the case. When children are acquiring Irish there is good evidence that the initial preverb is in C, as we shall see below.


197

In section 2.2.2 above it was argued that one of the possible causes of the

development of V-to-T movement was an increase in the level of variability in the

position of the verb when C was filled. We saw in the previous section that it is

possible that there may have been an increase in processes such as free inversion or

right-dislocation in pre-Old Irish, which contributed to the development of V-to-T

movement. When the verb is compound and free inversion or right-dislocation takes

place, the subject (or other moved element) will appear to the right of both the verb

and the preverb, as shown in (26) below. This will provide evidence, therefore, that

both the verb and the preverb have moved to T.

26. (a) Ní-m [in fer] tabair ⇒ Ní-m tabair [in fer]

(b) Do-m [in fer] beir ⇒ Do-m beir [in fer]

From this discussion, then, it seems that univerbation in the vP is heavily dependent

on univerbation at the clause level. Univerbation in the vP is essentially caused by the

reanalysis of V-to-C movement as V-to-T movement, as this change takes away the

evidence that preverbs are syntactically independent. However, if this is the case, then

univerbation within the vP must follow univerbation in the clause, and so we would

expect orders such as (25b) above, repeated as (27) below, which are simply not

found.

27. *#CP…V#

This problem can be solved, however, if we assume that both types of univerbation

occurred simultaneously. When the new generation change their parameter settings

such that V-to-C movement is lost and V-to-T movement develops, this means that

they have to reanalyse the syntactic status of preverbs so that they are consistent with

the new grammar. For this to be the case the evidence from preverbs can play no role

in parameter setting. If preverbs were taken into account, they would have provided

clear evidence for movement to C; however, they seem to be discounted. Under a cue-

based theory of language acquisition and language change, this is easily explained.

For a change in parameter settings to take place not all evidence for the earlier

parameter setting needs to be lost; the evidence simply needs to drop below a


198

particular threshold. Furthermore, parameters are set on the basis of specific cues. It

seems likely that the cue for an Affix-feature on C, i.e. for V-to-C movement, will be

based around the position of the verb, not preverbs. Therefore, we might expect the

evidence from preverbs to be ignored. So, within one generation, on the basis of the

evidence from the position of the verb there is a change in parameter settings, from V-

to-C movement to V-to-T movement, which causes preverbs to lose their syntactic

independence. However, children are still faced with evidence that the initial preverb

is in C. They must find a new way to analyse this data. This is the topic of the next

section.

2.4 The development of post-syntactic rules

In the two preceding sections it has been argued that between pre-Old Irish and the

attested Classical Old Irish period there were two major changes affecting the syntax

of the language. First, V-to-C movement was reanalysed as V-to-T movement, i.e. the

Affix-feature associated with C was lost, and an Affix-feature was instead assigned to

T. Secondly, the initial preverbs of compound verbs underwent a change in status

from heads to (valued v) features, losing their syntactic independence such that they

could no longer satisfy v’s Affix-feature. This led to the development of generalised

V-to-v movement. As a result of these changes neither the deuterotonic/prototonic

alternation nor the absolute/conjunct distinction could be syntactically derived.

Children must then find a new way to derive these morphological forms post-

syntactically. This section will examine each construction in turn.

2.4.1 Deuterotonic compound verbs

After the syntactic changes outlined above, deuterotonic compound verbs could no

longer be produced in the syntax. After the development of generalised V-to-T

movement and V-to-v movement, the verb moves to T in all clauses, incorporating the

light v head on the way. This new grammar, then, can only produce prototonic forms,

where the verb and the preverb appear together in T. It is no longer possible for the

preverb to move independently to C. Although it is no longer possible to produce

deuterotonic compounds syntactically, speakers who have acquired the new grammar

still need to produce deuterotonic compounds to ensure that the output of their new

grammar is sufficiently similar to the output of the old grammar. The way they do


199

this, I propose, is by developing a system of post-syntactic realisation rules, as

described in chapter 2.

In chapter 2 it was argued that deuterotonic compound verbs are a result of post-

syntactic processes. The C head in Old Irish has an unvalued verbal feature. This

verbal feature is valued during the syntax through an Agree relation between C and

light v. When there are no other C-based features present in C, this verbal feature is

phonologically realised in the C position. When this occurs, C is identical with the

lower light v heads in both TP and vP, and so the lower v heads receive no

phonological realisation. As a result, the initial preverb and the remainder of the verb

are spelled-out in separate positions, the preverb in C and the verb in T, resulting in

the deuterotonic stem. If Old Irish speakers really did use such a system to produce

deuterotonic stems, then this system must be acquirable. The remainder of this section

will consider whether this is the case.

The crucial point regarding the development of this new way of forming

deuterotonic compounds is that it does not involve any new operations in the syntax.

The basic syntactic operation involved is Agree between C and v. As we saw above,

for the verb to move to T, the C-T complex must be in an Agree relation with v-V.

The valuation of C’s verbal feature can be seen as a result of this Agree operation.

The difference between pre-Old Irish and Classical Old Irish involves the realisation

of this verbal feature. Classical Old Irish speakers with the new grammar reanalyse

the initial preverbs that appear in the C position as the phonological realisation of C’s

verbal feature. In DM terms what this means is that a new set of vocabulary items are

created that are in competition for insertion into the C position. Children acquiring the

new grammar acquire a set of vocabulary items such as those below:

28. [v1] ⇔ do

[v2] ⇔fo

[v3] ⇔at

etc.

As discussed in chapter 2, preverbs in Old Irish are a closed class. There are around

ten different preverbs, and so a child must acquire ten new vocabulary items. This is

not an unreasonable suggestion. Once the child has acquired these new vocabulary


200

items, the conditions under which they are inserted are determined by the process of

Vocabulary Insertion. Therefore, the child does not need to acquire these conditions,

the element of competition involved in Vocabulary Insertion will ensure that they are

inserted only when there is no other suitable vocabulary item that matches more

features contained in the terminal node.

The conditions under which the lower copy of light v is deleted are also

determined by independent postsyntactic principles. Old Irish speakers will not need

to acquire the fact that the lower copy of the preverb is deleted; this will take place

automatically as a result of the Chain Reduction operation, which deletes subsequent

copies of an item that has been marked to receive a phonological realisation. So far,

then, it seems that the only part of the post-syntactic approach to deuterotonic

compounds that must be acquired is the new vocabulary items for initial preverbs, the

new candidates for insertion into the C position. The rest of the process simply falls

out from principles of the post-syntactic component.

In sum, it seems that in terms of the post-syntactic processes involved in the

formation of deuterotonic compound verbs, the only part that has to be acquired is the

value of the verbal feature associated with the verb in question and how this verbal

feature is realised in C. The Agree relation, the environments in which this realisation

appears and the deletion of the subsequent copy all result from basic, conceptually

necessary post-syntactic operations, and so do not need to be acquired. It seems then

that acquisition of the deuterotonic verbs in Classical Old Irish is relatively

straightforward. Let us turn now to the other post-syntactic operation described in

chapter 2, namely that necessary for producing absolute inflection.

2.4.2 Absolute verbal inflection

In chapter 2 it was argued that, like deuterotonic compound verbs, absolute inflection

in Old Irish was the result of a post-syntactic process. After the loss of V-to-C

movement absolute inflection can no longer be the result of the syntactic position of

the verb. However, after the loss of V-to-C movement, children acquiring the new

grammar will still encounter absolute inflections and will need to find a new strategy

for producing them. Crucially, as was the case for deuterotonic compounds, this

strategy must be acquirable.

In chapter 2 it was argued that the distribution of absolute inflection resembles

Affix-Hopping in English. Absolute inflections are a realisation of the combination of


201

a verb’s φ-features with an affixal force feature. When C contains a host, this force

feature is realised on C rather than T, and so does not combine with the verb’s φ-

features, which are then realised as conjunct endings. When, however, C does not

contain a host the force feature is realised on T, combining with the φ-features to give

absolute inflection. Let us consider, then, what parts of this system must be acquired.

The existence of a Force feature does not need to be acquired; this comes for free

as part of UG. However, as we saw above, this feature is usually associated with C.

During language acquisition children must determine which features are associated

with which functional head. In the case of C and T the main decision to be made is

whether the feature in question is shared by both categories or present on only one.

Children acquiring this Old Irish system, then, will need to learn that Force is shared

by both T and C, and that it is affixal. Once these facts have been acquired the

question of whether this feature is realised on C or T should come down to basic PF

principles. When C is empty the Force feature cannot be realised there, according to

the Stranded Affix Filter and so is realised in the lower position. Similarly, when it is

realised on C, it cannot be realised on T, as a result of Chain Reduction. So far, then,

children need to acquire that there is a Force feature that can appear on both C and T,

and that this feature is affixal. The only other information that needs to be acquired is

the different realisations. Children must learn how the different combinations of

Force+φ-features are realised phonologically, i.e. they must learn the phonological

forms of the absolute endings.

Although it seems that the acquisition of absolute and conjunct endings is more

complex than the acquisition of deuterotonic compounds, this is not really the case. In

the case of compound verbs, children must learn for each individual verb which class

it belongs to, i.e. how a light v feature in C is realised and how this affects the

realisation of the remainder of the verb. The acquisition of the absolute-conjunct

system involves only functional features and so although there is perhaps more to

acquire, it only needs to be acquired once, and from there it will apply productively to

all verbs.

2.5 Summary

This section has provided an account of how the verbal system changed from that of

pre-Old Irish, as described in chapter 4, to that of Classical Old Irish, as seen in

chapter 2. It has been argued that the ‘univerbation’ that occurred between these two


202

periods was in fact the combination of two separate changes, the reanalysis of V-to-C

movement as V-to-T movement, and the change in status of preverbs resulting in the

development of generalised v-to-V movement. These changes seem to be closely

linked to one another, occurring within the same generation. The changes in

parameter settings are summarised in the table below.

Table 3: A summary of the changes between pre- and Classical Old Irish

Pre-Old Irish Classical Old Irish C + Affix - Affix T - Affix + Affix v + Affix + Affix Preverb X0 (valued v) Feature

As a result of these two syntactic changes neither deuterotonic verbal forms nor

absolute inflections could result from movement of the preverb or the verb to C.

Therefore, new post-syntactic operations developed that enabled speakers who

possessed the new grammar to match the output of their grammars to that of the older

generations who had the old grammar. At this point, we have reached the Classical

Old Irish period. As a result of certain syntactic changes neither the deuterotonic

forms nor absolute inflection are syntactically productive. The double system has

been lost as a syntactically productive system. In the following sections we consider

the next stage, namely how the double system was lost completely from the language

as a whole. Section 3 examines the simplification of compound verbs in the Middle

Irish period and section 4 explores the loss of the absolute/conjunct distinction in

Early Modern Irish.

3 THE SIMPLIFICATION OF COMPOUND VERBS

In the previous section it was argued that after univerbation the

deuterotonic/prototonic alternation was no longer syntactically productive. In

Classical Old Irish deuterotonic compounds were derived post-syntactically, through

the realisation of a verbal feature in C. This section looks at how this system broke

down and how compound verbs were simplified.

As the simplification of compound verbs took place for the most part during

Middle Irish there is considerable evidence to be found in texts from this period.

Section 3.1 provides a brief account of the texts from which the evidence used in this


203

section is drawn. Section 3.2 explores how compound verbs were simplified in

descriptive terms. Section 3.3 examines how these descriptive facts can be accounted

for theoretically. In section 3.4 we consider how the simplification progressed through

the language, and what the reasons for this might have been. Section 3.5 provides a

summary.

3.1 Texts

To investigate the simplification of compound verbs it is necessary to have a variety

of texts whose dates of composition span the entire Middle Irish period, i.e. from the

tenth to the twelfth centuries. I do not intend to provide a detailed absolute

chronology for the loss of the double system, so an understanding of how the texts

relate to one another is of more importance than their exact date. The texts in the

corpus can be broken down into three groups: early Middle Irish, middle Middle Irish

texts and late Middle Irish.

The early Middle Irish text is a long religious text Saltair na Rann (SR), believed

to date from 988 (Mac Eoin 1961a). Turning now to middle Middle Irish, the corpus

contains four predominantly prose texts, the second recension of Táin Bo Cúailnge

(TBC II) from the Book of Leinster, Togáil Troí (TT), Cath Catharda (CC) and

Aislinge meic Conglinne (AMC). TT seems to be the oldest, with Mac Eoin (1961b)

suggesting a date of composition in the eleventh century. Jackson (1990: xx) suggests

that maybe the second half of that century is more accurate. Through careful analysis

and comparison of the language of the two texts, Jackson (1990) argues that AMC is

later than TT, suggesting that it was composed in the last quarter of the eleventh

century. CC and TBC II are both later. Dating TBC II is difficult, due to the fact it has

an earlier source, found in the tenth-century Lebor na hUidre (TBC I), and the

tendency of the scribe to archaise (Jackson 1990: xxi). Dillon (1927) dates it at around

1100. Jackson suggests the more tentative ‘early twelfth century’, around the same

date as the manuscript in which it is found. CC is dated later than this. Sommerfelt

(1915) suggests 1150. O’Daly (1943), in her comparison of CC with TBC II, dates it

earlier, in the second quarter of the twelfth century. Importantly, she still considers it

later than TBC II. The late Middle Irish text15 is Acallam na Senórach (AS), a long

15 Jackson (1990) classes TBC II and CC as late Middle Irish. He does not include the texts I have classed as late Middle Irish.


204

text of around 8000 lines. Nuner (1958–9) argues, based on both linguistic and non-

linguistic evidence, that AS was composed between 1200 and 1225.

For all the texts discussed above lists of the verbal forms have been published. The

majority of the data in this section is drawn from these lists. For SR – Strachan

(1895–7), TT – Mac Eoin (1961b), CC – Sommerfelt (1915–8), TBC II – O’Daly

(1943), AS – Nuner (1958–9).

Having established the source of the data that is to follow, let us now turn to the

change in question, the simplification of compound verbs.

3.2 How the deuterotonic/prototonic alternation was lost

As we saw in section 1 above, in Modern Irish there are no compound verbs. With the

exception of the irregular verbs, all verbs in Modern Irish have only one stem used in

both independent and dependent position. During the Middle Irish period all

compound verbs were simplified and the deuterotonic/prototonic alternation was lost.

There were two main ways that a verb could become simplified in the Middle Irish

period. First, the compound verb in question could simply be replaced by a

synonymous simple verb. For example, we find that do-cuirethar and fo-ceird ‘puts’

are both replaced by the simple verb cuirid. Similarly Old Irish ad-fét ‘tells, relates’ is

replaced by the synonymous simple verb indisid.16 This is clearly a matter of use,

whereby speakers are choosing a simple verb instead of a compound, leading to the

loss of the compound form. However, this strategy only affects the small class of

compound verbs that have synonymous simple verbs and so is not widespread. For

the majority of compound verbs the simplification involves the innovation of a new

verbal stem, modelled on either the prototonic or the verbal noun. Some examples are

given in tables (5) and (6) below.17

16 McCone (1997a: 193) gives indisid as an example of a simple verb that has developed from the verbal noun of in-fét, which has the form indisiu. 17 There are also many examples where it is not entirely clear whether it is the prototonic or the verbal noun that has provided the new stem. See McCone (1997a: 194).


205

Table 5: New simple verbs that developed from the prototonic stem

OIr DT OIr PT OIr VN New simple verb English

do-fócaib -tócaib tócbail tócbaid ‘lifts’

do-róscai -derscaigi derscugud derscaigid ‘excels’

imm-soí -impai impód impoid ‘turns’

as-luí -élai élúd élaid ‘escapes’

con-tuili -cotlai cotlud cotlaid ‘sleeps’

(McCone 1997a: 192)

Table 6: New simple verbs that developed from the verbal noun

OIr DT OIr PT OIr VN New simple verb English

con-oí -cumai coimét coimétaid ‘protects’

do-ruimnethar -dermainethar dermat dermataid ‘forgets’

con-utaing -cumtaing cumtach cumtaigid ‘builds’

do-fich -dích dígal díglaid ‘requites’

(McCone 1997a: 193)

After simplification, then, compound verbs, like simple verbs, have a single stem.

However, this is not the only change that they undergo. The formerly compound verbs

also exhibit a change in their morphosyntactic behaviour.

After simplification formerly compound verbs begin to show absolute verbal

endings. As discussed in chapter 2, only simple verbs show absolute endings;

therefore, for a compound verb to have absolute endings it must have been simplified.

This is shown in (29a) and (29b) below. Special relative endings are also only found

with simple verbs. After a compound verb has been simplified it can show special

relative endings, as shown in (29c) and (29d) below.

29. (a) fácbaid (AMC 662) for OIr fo-ácaib ‘leaves’

(b) frecraidh (AMC 1455) for OIr fris-gair ‘answers

(c) dirges (SR 1008) for OIr do-rig ‘strips, lays bare’

(d) roiches (AS 7090) for OIr ro-saig ‘reaches’

The second morphosyntactic change associated with simplification is the use of the

dummy preverb no. As we saw in chapter 2, in Classical Old Irish no is used with


206

simple verbs when the verb is relative or in a secondary tense or in order to infix an

object pronoun. Crucially, compound verbs never appear with the preverb no.

Therefore, once a compound verb begins to appear with the dummy preverb no this

suggests that it has become simplified. In Middle Irish, relative clauses begin to be

marked simply by lenition of the initial consonant of the verb, so we do not find

examples of no being used in this environment. Similarly during the Middle Irish

period enclitic pronouns become replaced by independent pronouns. As a result, there

are very few examples of no being used here. However, the use of no with secondary

tenses is retained throughout the Middle Irish period.18 Some examples are given in

(30) below.

30. (a) no theilgidís ‘they used to throw’ (imperfect) (TT 1865) (OIr do-léctis)

(b) no derrscaiged ‘he might excel’ (past subj) (TT 381) (OIr durosced (Ml

129b13))

A third characteristic of a simplified compound verb is that it begins to appear with

prefixed ro in the past tense. In Old Irish the perfective particle ro appears in initial

position only when the verb is simple. When the verb is compound, in most cases the

particle ro is infixed between the initial preverb and the remainder of the compound

verb (Thurneysen 1946: 339–341).19 After simplification, a formerly compound verb

will appear with prefixed ro. This is shown in the examples in (31) below.

31. (a) ro fastad pass sg (TT 1143) vs. adrosoid 3sg (Ml 97d16)

ad-suidi ‘stops, detains, hinders’

(b) ro thocaib (TT 368) vs. dorogbad (Ml 17a13)

do-gaib ‘removes’

(c) do thomaill (AMC 1285) vs. dorumalt (SR 2909)

do-meil ‘consumes’

It is not entirely clear whether the change in the position of ro is a cause or an effect

of simplification. During the Middle Irish period the ro particle changed in function

18 This continues into Early Modern Irish, although the dummy preverb no is replaced by do. 19 There are exceptions to this. See McCone (1997a: chapters 9–11) for a full account of the use of ro and alternatives in Old Irish.


207

from a perfective particle to a past tense marker (McCone 1997a: 185–186). This

change from aspect to tense marker seems to coincide with the shift in its syntactic

position, from within the vP to the C position. Viewed in this way, it seems we have a

case of grammaticalisation, to at least some degree independent of the simplification.

The details of this change will be left for further research.

Having outlined the main features of the simplification of compound verbs, let us

consider how these facts can be reconciled with our theoretical account.

3.3 The simplification of compound verbs: a theoretical perspective

In chapter 2, and section 2 above, it was argued that in Classical Old Irish the

deuterotonic/prototonic alternation is not syntactically productive. The prototonic

form of the verb is derived regularly in the syntax through movement of the verb

through light v to T. To derive the deuterotonic form, however, further post-syntactic

operations are necessary, whereby the light v feature of the verb in question is

phonologically realised in the C position, and deleted lower in the structure. The

crucial difference between compound and simple verbs, then, is that compound verbs

have a positively valued verb-class feature that can be realised in the C position.

Simple verbs, on the other hand, have a verb-class feature with the default value, i.e.

0, which has no phonological realisation. Under this view, the simplification of

compound verbs can be seen as the failure to acquire a positively valued verb-class

feature, in other words, a verb is assigned to the wrong verb class.20

This theoretical view of the simplification can account for all the features of

simplification described in section 3.2. Let us consider each in turn.

It was observed above that when a compound verb is simplified it is always the

deuterotonic form that is lost, with a new stem developing from either the prototonic

or the verbal noun. When the verb-class feature loses its positive value it can no

longer be realised in C. Therefore the conditioning environment for the deuterotonic

form will be lost and so only the prototonic form can appear.21

Turning now to the morphosyntactic changes, it seems that these can also be easily

accounted for. Once a verb is analysed as having a default value for its verb-class

20 We might also expect a similar change to take place where a compound verb is assigned to another compound verb class. This seems to be the case for the Old Irish verb ro-cluinethar ‘hears’, which becomes associated with the preverb at- in Middle Irish by association with the verb ad-cí ‘sees’. 21 Of course, more needs to be said regarding how exactly the different phonological forms arise, but this will be left for further research.


208

feature, it will behave syntactically like a simple verb. The C head will still Agree

with light v as it did before, but when the verb-class feature has the default value, it is

not realised in C. This means that in the secondary tenses and with infixed pronouns

the dummy preverb no will be needed to realise the C position in relative clauses,

secondary tenses and when there is an infixed pronoun. Moreover, a default verb-

class value has no morphosyntactic realisation and so cannot combine with force

features in C (see chapter 2 above). This means that when there are no other C-based

features in C to combine with force, force will be realised on T, resulting in absolute

inflection.

So far we have seen that the theoretical account of compound verbs developed in

chapter 2 provides a pleasing account of the way in which compound verbs were

simplified. Up to this point our discussion has been predominantly descriptive,

accounting for how compound verbs were simplified. The next question is why this

was the case. This is the topic of the next section.

3.4 The causes of simplification

McCone (1997a: 191) argues that compound verbs were simplified because they were

“a collection of irregularities…difficult to preserve on the scale obtaining in Old

Irish”. This idea, that compound verbs were lost because they were irregular fits in

well with our theoretical account. In our view, compound verbs are irregular, as they

have an extra feature, a valued v feature, that is not present on simple verbs. This

extra feature makes the acquisition of compound verbs more difficult than that of

simple verbs, and so makes compound verbs more susceptible to loss.

Viewed in this way the simplification of compound verbs is essentially a process

of analogy, whereby the irregular compound verbs are subsumed under the productive

pattern of simple verbs. Kiparsky (1982, 1995b, 2000) views analogy as ‘grammar

optimization’. One way in which this can take place is through the removal of

idiosyncratic feature specifications from lexical entries (Kiparsky 1995b: 644). This is

exactly what we are proposing for Irish. The idiosyncratic feature specification, the

verb-class value, is lost and replaced by the default, namely 0.

Although irregular forms are more likely to be lost, this does not entail that they

will be. The languages of the world manage to retain many irregularities. If children

hear sufficient evidence for an irregular form then the irregular form will be acquired.

Crucially for an irregular form to be lost there must be a decrease in evidence for that


209

form. To explain the loss of the deuterotonic/prototonic alternation we must establish

what the relevant evidence is for the acquisition of the system and how this

disappeared in the Middle Irish period.

We saw above that to acquire the deuterotonic form of a compound verb it is

necessary to learn that the verb in question has a positively valued verb-class feature

that can be realised in the C position. There are three main pieces of evidence the

child could utilise to acquire this fact. From a syntactic perspective, children could

deduce from the fact that part of the verb precedes infixed object pronouns that the

initial preverb is in the C position. From a morphophonological perspective, if as

argued in chapter 2, stress and initial mutations both fall on the initial element of the

TP, then the fact that the initial preverb is unstressed and unmutated will provide

evidence that it is in the C position. There is evidence to suggest that all of these cues

that the verb is in C decrease in frequency in the Middle Irish period. Let us examine

each in turn.

It is well known that enclitic object pronouns are replaced by independent object

pronouns in the Middle Irish period (see McCone 1997a: 169–178; Jackson 1990: 93–

96 for details of this change). This development can be seen in the texts from my

corpus in the table below.

Table 6: Proportion of infixed to independent object pronouns in Middle Irish

SR TT22 AMC23 CC TBC24 AS25

Infix 396 100% 56 90% 91 97% 90 68% 311 89% 21 6%

Indep 026 0% 6 10% 3 3% 43 32% 40 11% 320 94%

Total 396 62 94 133 351 341

Jackson (1990: 96) argues on the basis of evidence from the Annals that the loss of

enclitic object pronouns was virtually complete by 1200. However, if we consider our

latest text, namely AS, which dates from around this time, there are very few

22 Data from Jackson (1990: 96) 23 Data from Jackson (1990: 96) 24 Data from Mac Gearailt (1997–1998). 25 Data from Nuner (1958–1959: 340). 26 There are one or two pronouns that could be interpreted as independent object pronouns in SR. However, as the number of infixed pronouns in SR is so high the inclusion of these cases makes little difference to the proportions given above.


210

examples of infixed pronouns, and in those that we do find the pronoun can only

appear in a limited set of environments, notably after the particle ro (examples (32a)

and (32b)) and the preverb do (32c).27

32. (a) tuc urchar don mhuic co ros- marbh

give.PRET.3SG shot to.the pig so PVB-INF.3SG kill.PRET.3SG

‘He cast a spear at the pig so that he killed it’

(AS 1557)

(b) Cid ro-t- imluaid & ro-t- aistrig ille?

what PVB-INF.2SG stir.PRET.3SG & PVB-INF.2SG bring.PRET.3SG hither

‘What has impelled you and & brought you hither?’

(AS 3898–3899)

(c) is ann do-m- riacht in da mnai

COP then PVB-INF.1SG reach.PRET.3SG.DT the two women

‘It is then that the two women reached me’

(AS 4090)

If AS contains only these remnants of the system of infixed pronouns then it seems

likely that it fell out of the spoken language earlier than 1200. CC and TBC also show

a highly simplified system of infixed pronouns, with a limited number of forms

(1sg -m, 2sg -t, 3sg/pl -s, 1pl -r) occurring in limited environments, again mainly with

ro and do. Mac Gearailt (1997–1998: 495) observes that this system is strikingly

similar to that found in the Bardic poetry of the Early Modern Period, suggesting that

even by the eleventh century infixed pronouns were probably a purely literary

phenomenon.

Let us turn now to the morphophonological evidence. In chapter 2 we saw that in

Old Irish stress seems to be associated with the first element of TP. Therefore, any

element that appears before the first stressed syllable must be part of the CP (or an

earlier clause). As long as a deuterotonic verb is stressed on the second syllable, this

will provide evidence that the first preverb is in C. There is evidence to suggest,

however, that this stress pattern is not retained throughout the Middle Irish period.

One of the first groups that undergo simplification, according to Jackson (1990), are

27 I am grateful to Geraldine Parsons for help with the translation of these examples.


211

verbs where the first preverb ends in a vowel and the next syllable starts with a vowel,

such as do-icc ‘comes’, do-adbat ‘shows’ and fo-acaib ‘gets’. Even in the Old Irish

glosses these verbs undergo a process of vowel elision so that the deuterotonic shows

the same stress pattern (and the same form) as the prototonic, as shown for the

preterite of do-icc (do-ánic/tánic) in (33a) and the present tense of do-adbat/tadbat in

(33b).

33. (a) tánic aimser mo idbartese

come.PRET.3SG time my offering.VN

‘the time of offering me has come’

(Wb 30d11)

(b) tadbat- som tra da rect…

show.PRES.3SG- emph.part.3SG.M then two laws

‘He shows forth, then, two laws…’

(Wb 4d10)

As a result, there is a loss of evidence that these verbs have a valued light v feature

realised in the C position, so they are subject to reanalysis as simple verbs. Jackson

(1990) argues that after these verbs have become simplified, other compound verbs

follow suit, showing first a change in stress, resulting in the falling together of the

deuterotonic and prototonic forms and then full simplification. Jackson (1990: 116)

provides the following examples from AMC: imrásium ‘we rowed’, imthét ‘goes

around’, which show prototonic stem in the deuterotonic position, but retain conjunct

endings. Although it is easy to see in the case of verbs that have undergone vowel

elision, this is not the case for the other verb type. As stress is not marked in the

manuscripts it is difficult to tell for sure whether such a stress shift did in fact take

place. However, it seems plausible that there would have been pressure to

amalgamate the irregular compound verbs to the strict regular initial stress pattern.

In chapter 2 it was argued that TP is also the domain of grammatical mutations.

When the clause is relative it is the initial consonant of the TP that is lenited. While

lenition falls on the second syllable of a compound verb this provides clear evidence

that the first preverb is in C. In Middle Irish all preverbs begin to lenite in non-relative


212

as well as relative clauses.28 As a result of this lenition becomes part of the phonology

of the compound verb rather than a grammatical process dependent on a particular

syntactic structure, and so it seems less likely that it would act as a cue for the

position of the preverb. A further point to bear in mind is that main clause lenition

would in many cases have made the deuterotonic form more like the prototonic. For

example, one of the main differences between do-beir and tabair is that the /b/ is

lenited to /v/ in the prototonic but not in the deuterotonic. After main clause lenition

this phonological difference would have disappeared. In certain other cases, notably

where the verbal stem began with an /f/, main clause lenition could have led to further

cases of vowel elision. For example, the lenition of /f/ in do-fócaib ‘lifts up’ could

have contributed to its reanalysis as the simple verb tócbaid (AMC 278).29 It seems

then that main clause lenition decreased the evidence for deuterotonic compound

forms in a number of different ways.

So far then, we have seen that there seems to have been a decrease in evidence that

the initial preverb of a compound verb is in the C position. These changes could then

have led to the loss of the compound verb. However, compound verbs were not all

lost at once. The process was gradual, spread out over several centuries. In the next

section we will examine the path that this change followed and explore some possible

reasons for this.

3.5 The spread of the simplification

If we are to trace the spread of the simplification throughout the verbal system we

must establish some criteria to determine whether or not a particular verb has been

simplified. In section 3.2 we saw that there are four main ways to determine whether

or not a compound verb has been simplified. First, it will only have a single stem, and

no longer show an alternation between deuterotonic/prototonic forms. Secondly, if a

verb is simple it will show absolute inflections. Thirdly, a simplified verb will appear

with the dummy preverb no in the secondary tenses. Finally, once a verb is simplified

28 The motivation behind this change is argued to be the loss of the neuter gender and the associated neuter infixed pronouns that were followed by lenition. After the loss of the neuter gender, the reason for this lenition was no longer clear. As a result it became reanalysed as general, main clause lenition caused by a preverb or conjunct particle. See McCone (1997a: 173) and Jackson (1990: 77) for more details. 29 One piece of evidence that supports the idea that verbs of this kind underwent vowel elision is hypercorrect forms. Forms such as do-farraid ‘arrived’ (SR 5715) or do-fuc ‘took’ (SR 1285) for Old Irish do-arraid and do-uc appear in early Middle Irish. The inclusion of an intervocalic /f/ in these environments suggests that it has been elided elsewhere.


213

it will appear with prefixed ro in the past tense. From these four diagnostics it should

be clear at what point a verb has become simplified. However, this is not always the

case. Within the same text a single verb may both fulfil and flout the same criterion.

For example in AMC we find the 3sg perfect form of do-meil ‘consumes’ with infixed

ro, do-romel (53) and also a 3pl perfect form with prefixed ro, ro thomailset (343).

Even more frequently we find that a verb fulfils one of the criteria and flouts another.

For example again in AMC we find the verb at-reig ‘rises’ in simplified form with a

3sg present special relative ending: érgius (299) alongside 3sg perfect forms

containing an infixed pronoun and infixed ro: atom-raracht (848).30

There are three possible explanations for the co-occurrence of simplified and

unsimplified forms. First, it could be that both forms co-existed in the spoken

language of the Middle Irish period. Alternatively, both forms could have existed in

the written form of the language, and been used to convey different stylistic effects.31

A third possibility is that the co-occurrence of the different forms reflects conflict

between the spoken language and the more archaic literary standard. Non-simplified

verbal forms are likely to reflect the literary standard, whereas simplified forms

reflect the spoken language. A careless or badly trained scribe may allow forms from

the vernacular to creep in from time to time. Bearing in mind that scribes are more

likely to archaise than to innovate, it seems plausible that if a verb shows one

simplified form it can be considered simplified.32

Let us turn our attention now to the order in which verbs are simplified. The most

influential factor that determines at which point a verb simplifies seems to be its

frequency. It seems unlikely to be a coincidence that the verbs that retain two stems

into the Modern Irish period are the most frequent. It can be seen from table 7 below

that the verbs that correspond to the Modern Irish irregular verbs, such as tabhair (do-

beir), déan (do-gní), feic (ad-cí), abair (as-beir), faigh (fo-gaib) clois/cluin (ro-

cluinethar), téidh (téit), beir (berid) are consistently the most frequent.

30 Interestingly, infixed ro seems to be preserved quite frequently when there is also an infixed object pronoun. It is not entirely clear why this is. 31 This is clearly the case later in the language. The Bardic Grammatical Tracts (Bergin 1916–55) give lists of all the possible alternative verb forms that are allowed in verse. It is perhaps possible that something similar is at play here. 32 Although see the discussion of do-icc below.


214

Table 7: The most frequent verbs, excluding the copula and the substantive

SR TT AMC CC TBC AS 1 do-beir (218) do-icc (123) do-beir (55) do-icc (313) do-icc (317+) do-beir (104)

2 téit (144) do-beir (113) do-gní (47) do-beir (249) téit (234) do-gní (83)

3 do-gní (134) téit (107) as-beir (37) gaibid (233) gaibid (204) gaibid (49)

4 ráidid (87) do-gní (88) gaibid (29) do-gní (231) berid (160) do-icc (44)

5 do-icc (60) do-roich (57) do-icc (28) ad-cí (145) do-gní (153) ad-cí (42)

6 do-tét (57) as-beir (55) berid (24) téit (91) do-beir (143) fo-gaib (42)

7 gaibid (57) gaibid (52) benaid (21) as-reig (90) as-beir (102) as-beir (39)

8 ro-cluinethar

(46)

ad-cí (46) at-aig (20) as-beir (79) ad-cí (96) téit (38)

9 berid (45) ro-cluinethar

(38)

ad-cí (17) ro-icc (72) do-tuit (72) cuirid (29)

10 fo-gaib (35) marbaid (35) as-reig (17) fo-gaib (67) as-reig (64) berid (26)

This is easy to explain under the account of the loss of compound verbs presented

here. It was argued above that, in order for a verb to be acquired as compound, a child

needs to encounter sufficient positive evidence that the verb in question has a

positively valued verb-class feature realised in C. If the required threshold is not

reached, then the verb-class feature will receive its default value and the verb will be

simplified. We saw in the previous section that during the Middle Irish period the

evidence that the initial preverb was in C decreased, making it more difficult for a

child to encounter sufficient evidence to acquire the valued verb-class feature.

However, the more frequent a verb is, the greater the chance that the child will

encounter sufficient evidence. As the evidence as a whole decreases only the most

frequent verbs will provide enough evidence to acquire the verb-class feature.

Although frequency is conceivably the most important factor, it is perhaps not the

only one. Certain classes of verbs are more likely to be simplified than others. For

example we saw above that according to Jackson (1990) the first class of verbs to be

simplified were those where the initial preverb ends in a vowel and the syllable

following the preverb begins with a vowel, such as fo-acaib ‘leaves’. This is also to

be expected, as these verbs will be the first to lose the stress-based cue for their

acquisition as compounds. In our earliest Middle Irish texts it is predominantly these

so-called class 1 verbs that show signs of simplification. In the middle Middle Irish

texts we find that virtually all class 1 verbs are simplified and some class 2 verbs, i.e.


215

those verbs where either the preverb ends, or the remainder of the verb begins, with a

consonant, are beginning to be simplified. By late Middle Irish only the irregular

verbs (i.e. ad-cí, as-beir, do-beir, do-gní, fo-gaib, ro-cluinethar and also do-roich)

remain unsimplified.33

Although for the most part the evidence seems to support Jackson’s chronology for

simplification, it is clear that the phonological shape of the verb is not the only factor

relevant in determining at what stage a verb is simplified. For example, the verb do-

icc ‘comes’ appears in its contracted form tic in the Glosses, and although there are

sporadic examples of simplification, such as no-s-ticc (AMC 1007) where no is used

to infix a pronoun and do ticedh (AS 2377) where do appears as a dummy preverb in

the imperfect, these uses are by no means consistent and do-icc shows no sign of

absolute endings even in AS. In this case, the crucial factor seems to be frequency. As

can be seen in table 5 the verb do-icc is one of the ten most frequent verbs in all six

texts. As it is so frequent, sufficient evidence will remain to acquire it as an irregular

form, whereas other class 1 verbs appear much less frequently and are subject to

simplification.

Sommerfelt (1915) observes that verbs that contain infixes are also likely to

survive longer. These infixes need not be productive. Sommerfelt argues that verbs

with petrified infixes also seem to retain their deuterotonic forms longer. Clearly if a

verb is used frequently with an infixed pronoun, then this will provide clear structural

evidence that the initial preverb is in C. However, this is not so clear in the case of

petrified infixes as they have become part of the verbal stem, and can no longer mark

the position of the preverb. However, petrified pronouns provide morphophonological

evidence. Typically stress falls after an infix. Therefore, if there is a petrified infix

this could allow the verb to maintain stress in second position and prevent the stress

from shifting to the initial syllable.34

A further pattern that is observed by Sommerfelt is that whether a verb is

simplified or not seems to depend on the form of its initial preverb. Preverbs seem to

33 Even the irregular verbs that survive into Modern Irish are beginning to show absolute endings, for example ní tabraid (AS 7685) for Old Irish ní tabair ‘he does not give’ and atcluinmíd (AS 3289) for Old Irish ro-chluinemmar (Ml 112b13) ‘we hear’. It seems likely that this reflects the breakdown of the absolute/conjunct distinction rather than the simplification of compound verbs. 34 Apart from one instance of at-bélat ‘they will die’ (889), the examples Sommerfelt gives under this category are in the most part from irregular verbs that avoid simplification as a result of their frequency, e.g. at-cím ‘I see’ (ad-cí), at-bérim-si ‘I say’ (as-beir), at-rághat ‘they rise’ (as-reig) and at-clúinim ‘I hear’ (ro-cluinethar). Therefore, the role of this factor is perhaps doubtful.


216

be lost in a sequence. This can be seen in the table below, which shows which

preverbs appear in each text.

Table 8: Preverbs in Middle Irish texts

FR SR TT AMC TBC CC AS

at/ad/as at/ad/as at/ad at at at at

do do do do do do do

fo fo fo fo fo fo fo

ro ro ro ro ro ro

im im im im im im

con con con con con

for for for for

fris fris fris

ar ar

in

This pattern is particularly interesting in view of the account of compound verbs

given above. The loss of a preverb reflects the loss of a particular verb class. Verb

classes, i.e. particular values for v, seem to be lost one by one. It seems likely that this

again is related to frequency. The preverbs that survive until the end of the Middle

Irish period are those associated with irregular verbs, i.e. ad-cí, do-beir, do-gní, fo-

gaib. Of these preverbs, it is only really do that survives as a verb class, as the other

preverbs only occur with one or two verbs. This is unsurprising as do is by far the

most frequent initial preverb in the Old Irish period (Ganesalingam 2003).35

3.6 Summary

This section has examined how the distinction between deuterotonic and prototonic

forms was lost in the Middle Irish period. Following McCone it was argued that

compound verbs became simplified because they were irregular. In order to acquire a

verb as compound a child must learn that the verb in question has an extra feature,

35 This process continues into Early Modern Irish, with do replacing ad in ad-cí and ad-chluin (OIr ro-cluinethar) (McManus 1994: 394).


217

namely a positively valued verb-class feature that can be realised in the C position. To

acquire this, children need to hear specific evidence. During the Middle Irish period

this evidence is gradually lost, and as a result children cease to assign positively

valued verb-class features and compound verbs are lost. This change does not affect

all compound verbs at the same time; simplification seems to spread throughout the

lexicon, affecting one verb at a time, with one verb at a time losing its valued verb-

class feature. The main factor determining this is frequency; however, certain other

factors also seem to play a role, for example, the phonological shape of the verb,

whether it occurs with a petrified infix and the verb class to which it belongs.

4 THE LOSS OF ABSOLUTE AND CONJUNCT INFLECTION

In section 2 above it was argued that as a result of the loss of V-to-C movement

between pre- and Classical Old Irish absolute and conjunct inflection was lost as a

syntactically productive system and came to be derived post-syntactically.

Throughout the Middle and Early Modern periods the distinction between absolute

and conjunct inflection was gradually eroded so that by Modern Irish each verb in

each tense only has one set of personal endings. In this section we will examine the

way in which the double system of inflection was lost, particularly the order in which

this change affected different tenses and moods, and consider what this can tell us

about how and why the system was lost. Like the simplification of compound verbs,

the loss of absolute and conjunct morphology took place at a time for which we have

a great deal of written evidence. To consider this change in full detail would involve a

close inspection of this textual evidence, from all the different dialects of Irish, and

also Scots Gaelic and Manx. Unfortunately such a sizeable task is beyond the scope of

this dissertation. This section, then, will provide an overview of the changes involved

in the loss of the absolute/conjunct distinction, with the more detailed study being left

for further research. Section 4.1 examines how the loss of the absolute/conjunct

distinction progressed throughout the language and section 4.2 considers how these

data support the theoretical approach outlined in this dissertation.

4.1 How the absolute/conjunct distinction was lost

In Old Irish the absolute/conjunct distinction was found only in the primary tenses,

namely the present indicative, present subjunctive, future and preterite. By Early

Modern Irish this distinction remained only in the present and the future. As we saw


218

in section 1, by Modern Irish this alternation has been lost completely. This section

examines the order in which the absolute/conjunct distinction was lost, beginning

with the loss in the preterite indicative and the present subjunctive during Middle

Irish, and then the loss in the present and future in the Early Modern period.

We saw in section 3 that the past tense underwent some significant changes during

the Middle Irish period. In Old Irish a distinction could be drawn in terms of both

form and function between the simple preterite and the perfect, formed with the

particle ro. In Middle Irish this distinction broke down. The particle ro changed in

function from a perfective to a generic past tense marker. As a result, the augmented

ro-forms were in competition with the older simple preterite formations, and began to

encroach on their territory. By late Middle Irish this change is well on its way, and in

AS we find only 74 past tense forms without ro compared to 470 with ro. This

increased use of ro meant that the absolute endings hardly surfaced. As a result, the

evidence for them decreased and they ceased to be acquired, with the only past tense

absolute ending that survives into Early Modern Irish being the 3sg s-preterite ending

-(a)is, as can be seen in the table below.36 The use of this form continues to decline

throughout the period (McManus 1994: 408).

Table 9: The past tense in Old and Early Modern Irish

OLD IRISH PRETERITE OLD IRISH PERFECT EARLY MIDDLE IRISH PAST

marbsu -marbus ro marbus do mholas

marbsai -marbais ro marbais do mholais

marbais -marb ro marb molais, do mhol

marbsaimmi(?) -marbsam ro marbsam do mhol(s)ama(i)r

do mholsam

___________ -marbsaid ro marbsaid do mhol(s)abha(i)r

do mholsaidh

marbsait -marbsat ro marbsat do mhol(s)ada(i)r

do mholsad

36 This table shows two different verbs, marbaid ‘kills’ and molaid ‘praises’. The reason for this is that although regular by the Early Modern period, in Old Irish molaid patterns as a deponent verb. The use of marbaid provides a clearer comparison.


219

The loss of the absolute/conjunct distinction in the present subjunctive was also well

underway in the Middle Irish period. In part this can be explained through

phonological and morphological changes. For example, a new 1sg ending -(e)ar

developed that replaced both the absolute and conjunct form.37 Similarly, the 2sg

endings had already fallen together in the Old Irish period, both having the form -ae.

This is not the case for the other persons and numbers. It is interesting to note that as

with the past tense, in the present subjunctive, with the exception of the 2pl form,

which we will return to below, it is the conjunct ending that survives into the Early

Modern period.

Table 10: The present subjunctive in Old and Early Modern Irish

OLD IRISH EARLY MODERN IRISH

marba -marb molar

marbae -marbae mola

marbaid -marba mola

marbmai -marbam molam

marbthae -marbaid molta(oi)

marbait -marbat

mola(i)d

One possible reason for this could be that the present subjunctive is most frequently

used after a conjunct particle. It can be seen from the table below, that in the Middle

Irish texts from our corpus (except SR) there are very few examples of verbs in the

present subjunctive without a conjunct particle.

37 On the origin of this ending see McCone (1997a: 217); McManus (1994: 407).


220

Table 11: The proportion of verbs in the present subjunctive used with or without a

conjunct particle

SR38 TT AMC CC TBC AS

Ind 57 41% 339 14% 3 9% 4 7% 16 11% 2 4%

Dep 82 59% 18 86% 32 91% 51 93% 126 89% 51 96%

Total 139 21 35 55 145 53

If the present subjunctive is used consistently with a conjunct particle, then the

absolute forms will not surface sufficiently frequently to be acquired and so are likely

to be lost.

Let us turn now to the tenses where the absolute and conjunct distinction survived

the longest, namely the present and the future tenses. Although these tenses maintain

an absolute/conjunct distinction at the end of the Middle Irish period, this distinction

is not found in all tenses. In the present tense a double set of endings is found in the

3sg, 1pl and 3pl; the future tense also shows a distinction in the 1sg in addition to

those found in the present. This can be seen in the paradigm below.

Table 12: The present and future tense in Early Modern Irish

PRESENT TENSE FUTURE TENSE

molaim -molaim molfad -molabh

mola(e) -mola(e) molfa -molfa

molaid -mol(ann) molfaidh -molfa

molma(o)id -molma(o)id

-molam

molfama(o)id -molfam

moltaoi

molaid

-moltaoi

-molaid

molfaidhe

molfaithe

-molfaidhe

-molfaithe

molaid -mola(i)d molfaid -molfad

The 1sg and 2sg present and 2sg future endings do not distinguish absolute and

conjunct even in the Old Irish period (Thurneysen 1946: 456–457). The 2pl, on the 38 Many of the examples of subjunctive forms without a conjunct particle are jussive subjunctives. It is possible that this was more frequent in the written than the spoken language. It is interesting to note that it is found most frequently in the religious, poetic text SR. 39 Two of these verbs are compound verbs, and so show conjunct rather than absolute endings.


221

other hand, survives in Old Irish, and is lost during the course of the Middle Irish

period. During Middle Irish there is confusion between the 2pl absolute and the 2pl

conjunct in the present, future and the a-subjunctive, with the absolute ending -(a)ithe

gradually replacing the conjunct -(a)id. In SR the absolute and conjunct endings

seemed to be used correctly. However, by TT and AMC only examples of the

absolute ending are found. This continues in CC and AS. It is interesting that this

change took place across all three tenses. This suggests that, at least to a certain

extent, the personal endings were not linked to a specific tense. Children learnt two

sets of endings, one for the primary tenses and one for the secondary tenses, and then

used these across the board.

The question of how and why this change took place are not entirely clear. The fact

that the absolute/conjunct distinction should be lost in 2pl forms before any others

could perhaps again be related to frequency. In our Middle Irish texts we find that the

2pl is the least frequent personal ending.

Table 13: The frequency of each personal ending in Middle Irish texts SR TT AMC CC TBC AS

1sg 101 4% 40 3% 110 15% 74 2% 290 9% 125 9%

2sg 179 6% 25 2% 70 10% 76 2% 294 9% 141 10%

3sg 1767 62% 890 55% 310 43% 2013 48% 1807 53% 543 37%

1pl 61 2% 7 0.4% 18 3% 58 1% 142 4% 136 9%

2pl 5740 2% 6 0.4% 2 0.3% 8841 2% 28 0.01% 5342 4%

3pl 389 14% 432 27% 54 8% 1111 26% 588 17% 234 16%

Pass 292 10% 213 13% 153 21% 786 19% 247 7% 220 15%

Total 2845 1613 717 4206 3396 1452

This suggests that there may have been less evidence available to acquire the

distinction between the two forms and so this distinction was more likely to be lost.

This does not, however, account for the fact that it was the absolute form that replaced

the conjunct. O’Daly (1943: 38) suggests that the reason for this was because the 2pl

conjunct ending -(a)id was identical in form to the 3sg absolute ending. The

40 49 of the 57 2pl forms in SR are imperatives. 41 53 of the 88 2pl forms in CC are imperatives. 42 18 of the 53 2pl forms in AS are imperatives.


222

generalisation of the 2pl absolute ending meant that the 2pl and 3sg could be clearly

distinguished from one another. Crucially, the generalisation of the absolute ending in

the 2pl seems to have occurred in the middle of the Middle Irish period, before the

loss of the absolute endings in the present subjunctive, thus explaining why the 2pl is

the only ending to maintain the absolute rather than the conjunct form.

Let us turn now to the other endings that survive into Early Modern Irish. The first

of these endings to be lost seems to be the 3pl. McManus (1994: 397) observes that in

the present tense the former absolute and conjunct endings are used interchangeably

in the Early Modern period, with the absolute form becoming more frequent. In the

Early Modern version of the Old Irish tale Scéla Mucce Meic Dathó (C. Breatnach

1996), for example, we find only the -aid ending.43 Similarly, the 1pl endings become

confused in Early Modern Irish. The direction of this change is not entirely clear.

McManus (1994: 401) observes that it is more common for the conjunct to replace the

absolute in prose texts, at least in the future. However, if we look further down the

line at how this ending developed into the Modern Irish dialects it seems that it is the

absolute ending that is most prevalent. In Munster, although there are traces of the

conjunct ending -am in the future, the absolute ending is more frequent, surviving in

the present as well as the future, e.g. molaimíd, cuirimíd ‘we praise, we put’ and

molfaimid ‘we will praise’. In Ulster, only the absolute forms survive, and only in the

present as molamaid and cuireamaid (O’Rahilly 1972: 170–1). It seems likely, then,

that for a time both the absolute and conjunct endings survived side by side, being

used interchangeably (see O’Rahilly 1972: 170, fn 2), with the absolute eventually

winning out. Again, it seems that it is the most phonologically marked ending that

tends to be favoured, with -muid being used in preference to -am.44

Let us turn finally to the 3sg forms. In Middle Irish a new 3sg conjunct ending

-(e)ann develops in the present tense.45 Throughout the Middle and Early Modern

periods this ending is interchangeable with the original ending, -ø, so we find forms

such as -mol alongside -molann (McManus 1994: 396). In the fifteenth century we

43 It was argued above that the 2pl conjunct ending -aid was lost because it was identical with the 3sg absolute. This was not the case here. Although the 3pl ending is spelled -aid there would have been a difference in the pronunciation, with the 3sg ending with lenited /ð/ and the 3pl with unlenited /d/ (</t/). 44 In the Modern Irish dialects this -muid ending has been reanalysed as a pronoun. For this process of degrammaticalization to have taken place, the ending must have been used to a certain degree of frequency. 45 The origin of this ending is disputed. See Breatnach (1994: 293); McCone (1997a: 205–7) for further details.


223

start to find examples of -(e)ann being used in place of absolute endings in prose

texts, reaching equal numbers in the seventeenth century. O’Rahilly (1972: 132)

observes that the original absolute -(a)id ending disappeared from literary usage in the

present tense in the mid-eighteenth century, but survived until the early twentieth

century in the spoken language in Ulster.46 Again, then, it seems that a period of

confusion, with the two endings being used in free variation, is followed by the

eventual loss of the less marked form. The -(e)ann ending spread further throughout

the verbal system, becoming the generic present tense ending. In the future, on the

other hand, it seems to be the absolute form that has been generalised, with -(a)idh

being most common not only in the 3sg future forms, but also across the paradigm.

Again, it seems likely that there would have been a period of fluctuation, with the -

(a)idh form winning out because it is more marked than the conjunct -a ending.

Having provided a brief survey of how the absolute and conjunct endings were lost

in Middle and Early Modern Irish, let us now consider how these facts fit with our

theoretical account of the double system of inflection.

4.2 The loss of the absolute/conjunct distinction: a theoretical account

In section 2 above it was argued that in order to acquire the absolute and conjunct

distinction children have to learn that there is an affixal Force feature that is present

on both C and T. When C contains positively valued features, the Force feature

combines with them and so is realised on C and deleted on T. In this case, the φ-

features cannot combine with Force and so are phonologically realised as conjunct.

When C is empty, on the other hand, the Force feature cannot be realised there, as

there are no features with which it can combine and so it is realised on T. It combines

with φ-features, and is realised phonologically as absolute endings. It was argued in

section 2 above that in order to acquire this system there are two main pieces of

information that must be learned. First, a child must learn that the Force feature is

affixal and shared by both C and T. Secondly, a child must acquire the different

phonological realisations that correspond to the different feature configurations, i.e.

that φ-features are spelled-out as conjunct endings and φ-features+Force are spelled-

out as absolute endings. This being the case, it seems then that the loss of absolute

and conjunct inflection could occur in one of two ways. Either there could be a 46 Interestingly this -(e)ann ending which is so prevalent in Modern Irish is not found in Scots Gaelic or Manx. O’Rahilly (1972: 132) suggests that this ending was lost in these dialects, as there is evidence for it in early written Scots Gaelic.


224

change in narrow syntax, namely a change in the position of the Force feature, or a

change in the vocabulary items (VIs – the endings), either in the morphosyntactic

features they encode or in the way they are realised phonologically. It seems that it is

the second option that is at play here.

We saw above that the first stage in the loss of the absolute/conjunct distinction

was its loss in the past indicative and present subjunctive. It was argued above that in

this case, the conjunct endings were generalised, as a result of a significant decrease

in the frequency of the absolute endings. Let us consider this in theoretical terms. In

Old Irish, for each person children would have had two separate VIs, one for absolute

and one for conjunct endings. This is shown below for the 3sg of the s-preterite:

34. (a) [Tpast] [φ3sg] ⇔ /ø/

(b) [Tpast] [φ3sg] [Force] ⇔ /(a)is/

In order to acquire each of these endings a child needs to hear them sufficiently

frequently. If the frequency of one of these endings decreases, say for example

through an increased use of conjunct particles, then there will be insufficient evidence

for it to be acquired, and this ending will be lost. As a result, the new grammar will

contain only one VI, corresponding to the conjunct ending, and this will be inserted

both when the verb is in absolute initial position and when it follows a conjunct

particle.

35. [Tpast] [φ3sg] ⇔ /ø/

At this stage there is no change in the syntax. The force feature may still appear on

either C or T, however, the VI no longer makes any reference to this feature, and can

appear both when it is present and when it is absent.47

We saw above that by the beginning of the Early Modern period the

absolute/conjunct distinction survives only in the present and future tenses, and in

47 This change could also be seen as a merger of the two VIs. In this view the two VIs remain but they both have the same realisation, in this case /ø/. Although it is possible for more than one VI to have the same realisation (for example, the genitive, plural and 3sg present endings are all realised as /s/ in English), it seems unlikely in this case, where the two VIs would be so similar, and the observed distribution is possible from simply one set of morphosyntactic features, that a child would postulate more than one VI.


225

these tenses only in the first and third persons. As a result of phonological and

morphological changes the distinction had simply been lost in the other tenses and

persons. As the distinction between absolute and conjunct survived in only a few

environments, it seems likely that it would have been more difficult to acquire as a

productive system. In other words, it would have become increasingly difficult to link

absolute endings to the presence of a Force feature on T. There are two possibilities as

to what may have occurred at this stage. First there may have been a change in the

syntax, whereby the Force feature ceased to be transferred from C to T. The

alternative is that there was a change in the VIs for the endings in the present and

future tenses such that they no longer made reference to the presence of a force

feature. Let us consider each of these options in turn

The idea that the last stage in the loss was the loss of the conditioning environment

for the absolute/conjunct alternation is appealing from an acquisitional perspective.

As discussed above, force is primarily associated with the C head; it is a C-based

feature. Under this view, for the Force feature to appear on T is the non-default case.

In order to acquire that the force feature is transferred to T, then, a child will need

sufficient evidence that this is the case. Once the absolute/conjunct system starts to

break down, the relationship between absolute endings and declarative force will

become obscured, leading to a reduction in evidence for the presence of a force

feature on T. However, absolute endings are not the only evidence that force is

present on T. In chapter 2 it was argued that one of the main arguments for

postulating a force feature on T is the presence of special relative marking on simple

verbs. In relative contexts certain forms of the verb show special relative endings,

most notably the 3sg ending -es, -as. This continues into Early Modern and Modern

Irish (except in the Munster dialect). As long as these relative verbal endings continue

there will be evidence for a Force feature on T. Therefore, it seems unlikely that the

loss of the absolute/conjunct distinction was a result of a change in the syntax.

Let us turn now to the second option. The loss of the link between absolute

morphology and declarative main clause marking could also be captured through a

change in the vocabulary items. Owing to the loss of the absolute/conjunct distinction

in other tenses and persons, it seems likely that the acquisition of the link between

absolute morphology and declarative main clauses will become more difficult. When

this link becomes obscured it will no longer be possible for children to acquire a VI

such as that below, that associates a particular phonological form with a force feature:


226

36. [Tpresent] [φ3sg] [Force] ⇔ /(a)id/

However, unlike in the past indicative and present subjunctive, when acquiring the

present and future tenses children will be faced with two different endings for the first

and third persons. Once the force feature is taken out of the equation, these two

endings will surface in exactly the same morphosyntactic contexts. They will both

have the features shown below:

37. (a) [Tpresent] [φ3sg] ⇔ /(a)id/

(b) [Tpresent] [φ3sg] ⇔ /(e)ann/

Once the link with force is lost, then, we would expect the absolute and conjunct

endings to be used interchangeably, resulting in free variation. As we saw above, this

seems to be the case in the Early Modern and Modern periods.48 Although there are

many examples of morphological doublets of this kind cross-linguistically, their

existence is generally considered to be unstable, with one variant replacing the other

(Kroch 1994). Again, this seems to be what happened in Irish, although again, further

research is necessary to determine exactly how long the doublets survived.

4.3 Summary

In this section we have explored, in general terms, how the distinction between

absolute and conjunct inflections was lost between Old and Modern Irish. The loss

seems to have involved a number of different changes, e.g. phonological changes that

caused two endings to fall together, the development of new morphological endings

that replaced both the absolute and the conjunct and syntactic changes that led to a

decrease in frequency of the absolute forms. As a result of these changes, by the Early

Modern period the distinction remained only in a few limited environments. This 48 Although it is possible that the two endings were in complete free variation, it is perhaps more likely that the choice of one over the other was determined by some other factor. In contact situations, for example, where doublets similar to this occur there is often a sociolinguistic distinction between the two variants (see Kroch 1994 and the references therein). We saw in the previous section that the ending that seems to win out is the most phonologically marked. This suggests that the choice between them is a matter of language use, rather than a principle of the grammar. Further research into the distribution of the different forms in the Early Modern and Modern periods could shed some light on this issue. For our purposes, however, the crucial point is that there is no distinction in terms of morphosyntactic features.


227

being the case, it became more difficult to acquire it as a productive system. The link

between absolute endings and the presence of a force feature was lost, and so the two

endings could appear in the same morphosyntactic environments. This led to free

variation, an unstable state of affairs, resulting in the eventual replacement of one

form by the other.

5 CONCLUSION

This chapter has provided an account of how the double system of verbal inflection

was lost between Old and Modern Irish. In section 2 it was argued that the first stage

in the loss of the double system was the syntactic change involved in univerbation.

The reanalysis of V-to-C movement as V-to-T movement, and the development of

generalised V-to-v movement resulting from a change in status of initial preverbs

from heads to simply valued v features, meant that neither the deuterotonic/prototonic

alternation nor the distinction between absolute and conjunct inflection could be

derived syntactically. At this stage, the double system was lost as a syntactically

productive system and so it began to be derived post-syntactically, as described in

chapter 2.

After univerbation and the loss of V-to-C movement, extra rules were needed to

derive the deuterotonic forms. Children had to learn for each compound verb that it

had a positively valued verb-class feature, and that this feature could be realised in C.

In order to do this they needed sufficient evidence. During the course of the Middle

Irish period this evidence was lost through the replacement of infixed object pronouns

by independent pronouns and phonological processes such as vowel elision, a

possible stress shift and the development of main clause lenition. As a result, children

could not acquire the correct verb-class value and compound verbs became

simplified. This evidence did not disappear for all verbs all at once. The change

spread through the lexicon affecting one verb at a time. The main factor determining

this spread was frequency.

The loss of the absolute/conjunct distinction also in part resulted from the loss of

V-to-C movement. After this children had to learn that certain feature configurations

could be realised in different ways, specifically that φ-features alone were realised as

conjunct endings, and φ-features combined with a declarative force feature were

realised as absolute endings. During the Middle and Early Modern periods the

evidence for these different endings decreased. The distinction was lost first in the


228

past and present subjunctive where the verb appeared frequently with a conjunct

particle. The loss in the present and future tenses was more gradual. The

absolute/conjunct distinction remained in the 3sg, 1pl and 3pl forms. However, as the

absolute/conjunct distinction had become restricted to these contexts, it became more

difficult to acquire it as a fully working system. The evidence for the link between

absolute morphology and declarative force was significantly reduced. Therefore, the

different endings were acquired simply as free variants, identical in terms of

morphosyntactic features. Over time one of the variants became preferred, usually the

variant with the most phonologically marked form, and the other disappeared. This

was essentially a matter of language use.

From the discussion in this chapter it is clear that the loss of the double system was

a gradual process, beginning with univerbation in around the sixth century, and not

reaching completion until the early twentieth century. Although the change that

initiated the loss, i.e. the parametric change involving the reanalysis of V-to-C

movement as V-to-T movement, was abrupt, the consequences of this change were

gradually implemented, first through the loss of deuterotonic forms and then through

the loss of the absolute/conjunct distinction. In this case, the gradual implementation

of the change can be seen as a case of lexical diffusion, affecting lexical items in the

case of compound verbs, and functional items in the case of the absolute/conjunct

distinction. This provides an example of how, although parametric change is abrupt,

its complete implementation can be gradual.

CHAPTER 6

CONCLUSION

The main aim of this dissertation has been to provide a new account of the double

system of verbal inflection in Old Irish, specifically how it developed from PIE and

how it was lost between Old and Modern Irish.

Chapter 1 sets out the relevant background, familiarising the reader with Old Irish

and its position within the Celtic language family and also the main features of

Chomsky’s minimalist programme. In this chapter we also introduce an important

theoretical issue, namely the status of head-movement within the minimalist

programme, outlining the main assumptions relevant for the remainder of the

dissertation.

Chapter 2 develops a new synchronic analysis of the double system of verbal

inflection in the Classical Old Irish period (eighth and ninth centuries). Having

introduced the main features of the Old Irish verbal system, the existing generative

literature on the subject is reviewed. Through close consideration of the Old Irish data

it emerges that the existing syntactic accounts of the double system are fundamentally

flawed. With regards to simple verbs it is argued that evidence from infixed pronouns

and relative clauses suggests that, although prior to the Old Irish period the verb

moved to the C position, in Classical Old Irish the verb moves only as far as T in the

syntax. It is observed that adopting this view also provides a satisfying account of

stress placement and grammatical mutation within the verbal complex in Old Irish,

with both operations applying within the TP domain.

Turning to compound verbs, it is argued that the initial preverbs of deuterotonic

compound verbs are best analysed as C elements due to their phonological and

syntactic parallels with conjunct particles. However, from a theoretical perspective it

is seemingly impossible for them to move to the C position in the syntax. Various

possible analyses are investigated, including head-movement, XP-movement and

remnant-fronting, all of which face significant theoretical and empirical difficulties.

As simple verbs always appear in the T position, absolute and conjunct

morphology cannot be a reflection of the verb’s syntactic position. Furthermore, as

the initial preverb of compound verbs cannot move to C in the syntax, the distinction

between deuterotonic and prototonic forms similarly cannot be syntactically derived.

CHAPTER 6 – CONCLUSION

230

Therefore, it is concluded that in order to account for the double system of verbal

inflection synchronically it is necessary to invoke post-syntactic operations.

A new post-syntactic account of the double system is developed, which, unlike that

proposed by Adger (to appear), does not invoke conceptually undesirable post-

syntactic movement. A parallel is drawn between the distribution of absolute

inflection in Old Irish and Affix-Hopping in English. Absolute inflection is argued to

result from the presence of an affixal force feature combining with the verb’s φ-

features on T. This force feature is shared by both C and T. When C contains valued

features, affixal force combines with them and is deleted on T, resulting in conjunct

inflection. When C contains no valued features other than force, affixal force has no

host and so cannot be realised there, in accordance with the Stranded Affix Filter. In

this case, force is realised on T in combination with the φ-features, giving rise to

absolute inflection. The distribution of the dummy preverb no was also examined, and

argued to parallel do-insertion in English.

With regard to compound verbs it is argued that the initial preverb of the

deuterotonic form is best analysed as the phonological realisation of a valued verb-

class feature in the C position. This verb-class feature on C is valued by an Agree

relation with light v, but is only realised when no other inherent C features are present

in the C position. When the verb-class feature is realised in C the lower v head is

deleted under identity, giving rise to the deuterotonic form. In all other cases v is

spelled-out alongside the verb resulting in the prototonic form.

Chapter 3 marks the beginning of the diachronic part of the dissertation,

introducing the theoretical framework to be adopted in chapters 4 and 5. Following an

outline of the main principles of parametric change within a minimalist framework, a

cue-based theory of syntactic change is developed, building on work by Dresher &

Kaye (1990), Dresher (1998) and Lightfoot (1999). In particular, a question is raised

as to the role played by morphology in acquisition and syntactic change.

Chapter 4 presents a novel account of the development of the double system of

verbal inflection from PIE. The main philological accounts of the development are

reviewed. Cowgill’s (1975) particle theory is adopted in favour of McCone’s (1979b)

suffix-/infix-deletion theory on conceptual grounds. However, the Particle Theory as

it stands cannot account for the link between the development of the double system

and the development of verb-initial word order. The remainder of the chapter attempts


231

to fill this void, supplementing Cowgill’s particle theory with an account of the

development of verb-initial word order.

An overview of the main syntactic features of PIE is presented in order to provide

a starting point for the diachronic account. Crucially, evidence is presented to suggest

that PIE had an articulated CP, to which it was possible to move the verb for stylistic

or pragmatic effects. The development of unmarked verb-initial word order in Old

Irish is argued to result from the reanalysis of this optional, pragmatically motivated

verb-movement as obligatory, grammatically motivated verb-movement to the C

position. This development seems to be linked to both the loss of the articulated CP

and also the development of new C-based verbal morphology marking clause type.

The development of V-to-C movement was not the only consequence of the

changes in the CP. The initial preverb of compound verbs also began to appear in the

C position. Furthermore, it seems that a new class of conjunct particles developed,

specifically negative and interrogative particles and certain conjunctions.

Chapter 5 considers the loss of the double system. It is argued that the first stage

involves the loss of the double system as a syntactically productive process. This

occurs as a result of what Watkins (1963) terms ‘univerbation’. Between pre- and

Classical Old Irish V-to-C movement is reanalysed as V-to-T movement and preverbs

undergo a change in status from heads to valued v features, resulting in generalised V-

to-v movement. As a consequence of these syntactic changes, neither the

absolute/conjunct distinction nor the deuterotonic/prototonic alternation can be

derived syntactically. However, children acquiring the language are still faced with

these different forms, and so have to find a new way to derive them. It is at this point

that the post-syntactic operations outlined in chapter 2 develop.

After the Old Irish period the double system is lost from the language as a whole.

The first stage of this is the simplification of compound verbs. This change is argued

to be essentially analogical, with the irregular compound forms adopting the regular

pattern of simple verbs. Acquiring a verb as compound involves acquiring the fact

that the verb in question has a positively valued v feature. During the Middle Irish

period a reduction in the evidence for these v features leads to a failure to acquire

them, and so the valued v feature is lost, and compound verbs are reanalysed as

simple. This change is gradual, affecting one verb at a time. The point at which a verb

is simplified is determined primarily by frequency.


232

The final stage in the loss of the double system is the loss of the distinction

between absolute and conjunct endings. This change is argued to have occurred in

two stages. The distinction disappeared first in the past indicative and present

subjunctive, seemingly as a result of an increase in the use of conjunct particles in

these tenses. This led to the decrease in frequency and subsequent loss of the absolute

forms. After the loss of the double system in these tenses, distinct endings remained

only in the first and third persons of the present and future tenses. Due to the limited

number of environments in which the absolute/conjunct distinction survived it

became more difficult to acquire the link between absolute morphology and the force

feature. Therefore, the system was lost. The two endings survived for a while as free

variants, before one ending, usually the most phonologically marked, became

ubiquitous.

In spite of the seemingly narrow empirical focus of the dissertation, the preceding

chapters have touched on a number of broader theoretical issues. The synchronic

analysis of Old Irish in chapter 2 led to an investigation of the interface between

syntax and phonology and the operations that can take place there. From a diachronic

perspective, we have considered the development and loss of head-movement and the

thorny issue of the link between head-movement and verbal morphology. By

considering the development of head-movement alongside the loss, within the C-

domain rather than the T domain, evidence has been provided to support the existence

of a link between morphology and head-movement. Finally, the investigation of the

loss of the double system has provided an example of how, despite the necessarily

abrupt nature of parametric change, its effects can be gradual, spanning many

centuries, through the process of lexical diffusion.

REFERENCES

233

REFERENCES

Ackema, P. & A. Neeleman (2005). Beyond morphology: interface conditions on

word formation. Oxford: Oxford University Press.

Adger, D. (2003). Core syntax: a minimalist approach. Oxford: Oxford University

Press.

Adger, D. (to appear). Postsyntactic movement and the Old Irish verb. To appear in

Natural Language and Linguistic Theory.

Adger, D. & G. Ramchand (2005). Merge and Move: Wh-dependencies revisted.

Linguistic Inquiry 36, 161–193.

Ahlqvist, A. (1983). Two notes on relative marking in Old Irish. Celtica 15, 10–12.

Alexiadou, A. & E. Anagnostopoulou (1998). Parametrizing AGR: word order, V-

movement and EPP-checking. Natural Language and Linguistic Theory 16,

491–539.

Alexiadou, A. & G. Fanselow (2002). On the correlation between morphology and

syntax. J-W. Zwart & W. Abraham (eds.) Studies in Comparative Germanic

Syntax. Amsterdam: Benjamins, 219–242.

Andersen, H. (1973). Abductive and deductive change. Language 49, 765–93.

Anderson, S. (1992). A-morphous morphology. Cambridge: Cambridge University

Press.

Anderson, S. (1993). Wackernagel’s revenge: clitics, morphology, and the syntax of

second position. Language 69, 68–98.

Anderson, S. (1996). How to put your clitics in their place. The Linguistic Review 13,

165–191.

Arad, M. & A. Roussou (1997). Particles and C-positions in Classical Greek.

Unpublished ms: University College London and University of Wales, Bangor.

Baker, M. (1988). Incorporation: a theory of grammatical function changing.

Chicago: University of Chicago Press.

Baker, M. (2001). The atoms of language: the mind’s hidden rules of grammar. New

York: Basic Books.

Bayer, J. (1984). COMP in Bavarian syntax. The Linguistic Review 3. 209–274.

Bhatt, R. (1999). Verb movement and the syntax of Kashmiri. Dordrecht: Kluwer.

Beerman, D., D. LeBlanc & H. van Riemsdijk (1997). Rightward movement.

Amsterdam: John Benjamins.

REFERENCES

234

Belletti, A. (1991). Generalized verb movement: aspects of verb syntax. Torino:

Rosenberg & Sellier.

Belletti, A. (1999). Italian/Romance clitics: Structure and derivation. H. van

Riemsdijk (ed), Clitics in the Languages of Europe. Berlin: De Gruyter, 543-

579.

Bergin, O. (1916–1955). Irish grammatical tracts I–V. Supplements to Ériu 8–10, 14,

17.

Bergin, O. (1938). On the syntax of the verb in Old Irish. Ériu 12, 197–214.

Best, R. & O. Bergin (1929). Lebor na hUidre (Book of the Dun Cow). Dublin:

Dublin Institute for Advanced Studies.

Best, R., M. O’Brien & A. O’Sullivan (1954–1983). The Book of Leinster, formerly

Lebar na Núachongbála. Dublin: Dublin Institute for Advanced Studies.

den Besten H. (1977/1983). On the interaction of root transformations and lexical

deletive rules. W. Abraham (ed.), On the Formal Syntax of the Westgermania.

Amsterdam: John Benjamins, 47–131.

den Besten, H. & G. Webelhuth (1987). ‘Remnant topicalisation and the constituent

structure of the VP in the Germanic SOV languages’ GLOW Newsletter 18, 15–

16.

den Besten, H. & G. Webelhuth (1990). Stranding. G. Grewendorf, & W. Sternefeld

(eds.) Scrambling and Barriers. Amsterdam: John Benjamins, 77–92.

Biberauer, T. (2003). Verb Second (V2) in Afrikaans: A Minimalist Investigation of

Word Order Variation. PhD dissertation, University of Cambridge.

Biberauer, T. (2005). Splitting not spreading: a new perspective on the C/T

connection. Paper presented at GLOW V in Asia (Delhi), October 2005.

Biberauer, T. & I. Roberts (2005). Changing EPP-parameters in the history of

English: accounting for variation and change. English language and linguistics

9, 5–46.

Biberauer, T. & I. Roberts (to appear) Subjects, tense and verb-movement in

Germanic and Romance. To appear in Proceedings of GLOW V in Asia.

Bobaljik, J. (1995). Morphosyntax: the syntax of verbal inflection. PhD dissertation,

MIT.

Bobaljik, J. (2002). A-chains at the PF interface: Copies and ‘covert’ movement.

Natural Language and Linguistic Theory 20, 197–267.

REFERENCES

235

Bobalkik, J. & A. Carnie (1996) A minimalist approach to some problems of Irish

word order. R. Borsley & I. Roberts (eds.) The Syntax of the Celtic Languages

Cambridge: Cambridge University Press, 223–240.

Boeckx, C. & S. Stjepanovic (2001). Head-ing for PF. Linguistic Inquiry 32, 345–

355.

Boley, J. (1985). Hittite and Indo-European place word syntax. Die Sprache 31, 229–

241.

Boling, B. (1972). Some problems of the phonology and morphology of the Old Irish

verb. Ériu 23, 73–101.

Borsley, R. & A. Kathol. (2000). Breton as a V2 language. Linguistics 38,

665–710.

Borsley, R., M. Rivero, & J. Stephens (1996). Long head movement in Breton. R.

Borsley & I. Roberts (eds.) The Syntax of the Celtic Languages. Cambridge:

Cambridge University Press, 53–74.

Borsley, R. & I. Roberts (1996). The Syntax of the Celtic Languages. Cambridge:

Cambridge University Press.

Bošković, Ž. (2000). Second position clitics: syntax and/or phonology? F. Beukema

& M. den Dikken (eds.) Clitic phenomena in European Languages. Amsterdam:

John Benjamins, 71-119.

Breatnach, C. (1996). Patronage, Politics and Prose: Ceasacht Inghine Guile, Sgéala

Muice Meic Dha Thó, Oidheadh Chuinn Chéadchathaigh. Maynooth: An

Sagart.

Breatnach, L. (1977). The suffixed pronoun in Early Irish. Celtica 12, 75–107.

Breatnach, L. (1980). Some remarks on the relative in Old Irish. Ériu 3, 1–9.

Breatnach, L. (1994). An Mheán-Ghaeilge. K. McCone, D. McManus, C. Ó Háinle &

L. Breatnach (eds.), Stair na Gaeilge in ómós do Phádraig Ó Fiannachta.

Maynooth: Department of Old Irish, St Patrick’s College, 221-333.

Brody, M. (1995). Lexico-logical form: a radically minimalist theory. Cambridge,

MA: MIT Press.

Cardinaletti, A. & I. Roberts (2002). The Extended Projection Principle as a condition

on the tense dependency. P. Svenonius (ed.) Subjects, expletives, and the EPP.

Oxford: Oxford University Press, 125–156.

Carney, J. (1979). Aspects of Archaic Irish. Éigse 17, 417–435.

REFERENCES

236

Carnie, A. (1995). Non-verbal predication and head movement. PhD dissertation,

MIT

Carnie, A., H. Harley & E. Pyatt (1994). The resurrection: Raising to Comp, evidence

from Old Irish. Studies in the Linguistic Sciences 24, 85–100.

Carnie, A., H. Harley & E. Pyatt (2000). VSO order as raising out of IP? Some

evidence from Old Irish. A. Carnie & E. Guilfoyle (eds.), The syntax of verb-

initial languages. Oxford University Press, New York, 39–59.

Chametzky, R. (2000). Phrase structure: from GB to Minimalism. Oxford: Blackwell.

Chomsky, N. (1955/1975). The logical structure of linguistic theory. New York:

Plenum.

Chomsky, N. (1957). Syntactic structures. The Hague: Mouton.

Chomsky, N. (1977). On wh-movement. P. Culicover, T. Wasow & A. Akmajian

(eds.), Formal Syntax, Academic Press, New York, 71–132.

Chomsky, N. (1981). Lectures on government and binding. Dordrecht: Foris.

Chomsky, N. (1993). A minimalist program for linguistic theory. K. Hale & S. J.

Keyser (eds.) The view from building 20. Essays in linguistics in honor of

Sylvain Bromberger. Cambridge MA: MIT Press.

Chomsky, N. (1995a). The minimalist program. Cambridge MA: MIT Press.

Chomsky, N. (1995b). Bare phrase structure. G. Webelhuth (ed.) Government and

binding theory and the minimalist program. London: Blackwell, 383–439.

Chomsky, N. (2000). Minimalist inquiries: the framework. R. Martin, D. Michaels &

J. Uriagereka (eds.), Step by step: essays in minimalist syntax in honor of

Howard Lasnik. Cambridge MA: MIT Press, 89–156.

Chomsky, N. (2001). Derivation by phase. M. Kenstowicz (ed.), Ken Hale: a life in

language. MIT Press: Cambridge MA, 1–52.

Chomsky, N. (2002). On nature and language. Cambridge: Cambridge University

Press.

Chomsky, N. (2004). Beyond explanatory adequacy. A. Belletti (ed.) Structures and

beyond: the cartography of syntactic structures, volume 3. Oxford: Oxford

University Press. 104–131.

Chomsky, N. (2005). On phases. Ms MIT.

Chomsky, N. (2006). Approaching UG from below. Ms MIT.

Chomsky, N. & M. Halle (1968). The sound pattern of English. New York: Harper &

Row.

REFERENCES

237

Chung, S. & J. McCloskey (1987) Government, barriers and small clauses in Modern

Irish. Linguistic Inquiry 18, 173–237

Cinque, G. (1999). Adverbs and functional projections. New York: Oxford University

Press.

Clark, R. & I. Roberts (1993). A computational approach to language learnability and

language change. Linguistic Inquiry 24, 299–345.

Cowgill, W. (1975). The origins of the Insular Celtic conjunct and absolute verbal

endings. H. Rix (ed.), Flexion und Wortbildung: Akten der V. Fachtagung der

Indogermanischen Gesellschaft, Regensburg, 9.–14. September 1973. Reichert,

Wiesbaden, 40–70.

Cowgill, W. (1985). On the origin of the absolute and conjunct verbal inflections of

Old Irish. B. Schlerath (ed.), Grammatische Kategorien: Funktion und

Geschichte. Reichert, Wiesbaden, 109–18.

Cowgill, W. (1987). The distribution of infixed and suffixed pronouns in Old Irish.

Cambridge Medieval Celtic Studies 14, 1–5.

Déchaine, R. & M. Wiltschko (2003). Negation at the left periphery. Evidence from

Algonquian and Salish. L. Carmichael, C-H. Huang & V. Samiian (eds.),

Proceedings of WECOL 2001. Fresno, Ca: CSU Fresno, 104-117.

Degrassi, A. (1957). Inscriptiones Latinae Liberae Rei Publicae. Fasculus prior.

Firenze: La nuova italia.

Delbrück, B. (1893). Vergleichende Syntax der indogermanischen Sprachen.

Strassburg: K. J. Trübner.

den Dikken, M. (1995). Particles: On the syntax of verb-particle, triadic, and

causative constructions. Oxford: Oxford University Press.

Dillon, M. (1927). Nominal predicates in Irish I. Zeitschrift für Celtische Philologie

16, 313–356.

Dillon, M. (1947). Celtic and the other Indo-European languages. Transactions of the

Philological Society, 15–25.

Dillon, M. (1962). History of the preverb to. Éigse 10, 120–126.

Dik, H. (1995). Ancient Greek word order. Amsterdam: Gieben.

Doherty, C. (1999). Verb second and tmesis in Early Irish. S. Chang (ed.)

Proceedings of Twenty-Fifth Annual Meeting of the Berkeley Linguistics

Society. Berkeley CA: Berkeley Linguistics Society, 98–108.

REFERENCES

238

Doherty, C. (2000). Residual verb second in Early Irish: On the nature of Bergin’s

Construction. Diachronica 17 5–38.

Dover, K. (1960). Greek word order. Cambridge: Cambridge University Press.

Dresher, E. (1999). Charting the learning path: cues to parameter setting. Linguistic

Inquiry 30, 27–68.

Dresher, E. & J. Kaye (1990). A computational learning model for metrical

phonology. Cognition 34, 137–195.

Dressler, W. (1969). Eine textsyntaktische Regel der idg. Wortstellung. Zeitschrift für

vergleichende Sprachforschung auf dem Gebiet der Indogermanischen

Sprachen 83, 1–25.

Duffield, N. (1995). Particles and projections in Irish syntax. Dordrecht: Kluwer

Academic Publishers.

Embick, D. (1997). Voice and the interfaces of syntax. PhD dissertation, University

of Pennsylvania.

Embick, D. & M. Halle (2005). On the status of stems in morphological theory. T.

Geerts & H. Jacobs (eds.) Romance languages and linguistic theory.

Amsterdam: John Benjamins, 37–62.

Embick, D. & R. Noyer (2001). Movement operations after syntax. Linguistic Inquiry

32, 555–596.

Emonds, J. (1978). The verbal complex V-V in French. Linguistic Inquiry 9, 151–

175.

Eska, J. (1989). Interpreting the Gaulish inscription of Voltino. Bulletin of the Board

of Celtic Studies 36, 106–7.

Eska, J. (1994). Rethinking the evolution of Celtic constituent configuration.

Münchener Studien zur Sprachwissenschaft 55: 7–39.

Eska, J. (1996). On syntax and phonology within the early Irish verbal complex.

Diachronica 13: 223–257.

Eska, J. (2002a). Absolute and conjunct and Cowgill. Paper presented at the East

Coast Indo-European Conference, University of Pennsylvania, June 2002.

Eska, J. (2002b). Re-assessing Bergin’s Rule. Paper presented at Harvard Celtic

Colloquium, October 2002.

Eska, J. & D. Ellis Evans (1993). Continental Celtic. M. Ball & J. Fife (eds.) The

Celtic Languages. London: Routledge, 26–63.

REFERENCES

239

Eythórsson, T. (2003). The interaction of clitics and stress assignment in Old Irish.

Paper presented at the Fourth Celtic Linguistics Conference, Cambridge,

September 2003.

Evans, D. S. (1964). A grammar of Middle Welsh. Dublin: Dublin Institute for

Advanced Studies.

Ferraresi, G. (1997). Word Order and Phrase Structure in Gothic. PhD Dissertation,

University of Stuttgart.

Fischer, O., A. van Kemenade, W. Koopman & W. van der Wurff (eds.) (2000). The

syntax of early English. Cambridge: Cambridge University Press.

Fortson, B. (2004). Indo-European language and culture. Blackwell, Oxford.

Fox, D. & D. Pesetsky (2004). Cyclic linearization of syntactic structures. Theoretical

Linguistics 31, 1–46.

Franks, S. (1999). Optimality Theory and clitics at PF. K. Dziwirek, H. Coats & C.

M. Vakareliyska (eds.) Proceedings of Formal Approaches to Slavic Linguistics

7. Ann Arbor: Michigan Slavic Publications, 101–116.

Fuß, E. (2005). The rise of agreement. A formal approach to the syntax and

grammaticalization of verbal inflection. Amsterdam: John Benjamins.

Ganesalingam, M. (2003). Multiple preverbs in Old Irish. MPhil dissertation,

University of Cambridge.

Garrett, A. (1990). The syntax of Anatolian pronominal clitics. PhD dissertation,

Havard University.

Garrett, A. (1992). Topics in Lycian syntax. Historische Sprachforschung 105,

200–212.

Garrett, A. (1994). Relative clause syntax in Lycian and Hittite. Die Sprache 36, 29–

69.

Garrett, A. (1996). Wackernagel's Law and unaccusativity in Hittite. A. Halpern & A.

Zwicky (eds.), Approaching second: Second position clitics and related

phenomena. Stanford, CA.: CSLI Publications, 85–133.

Gelderen, E. van (2005). Grammaticalization as Economy. Amsterdam, John

Benjamins.

Görlach, M. (1991). Introduction to Early Modern English. Cambridge: Cambridge

University Press.

Gray, D. (1985). The Oxford book of late medieval prose and verse. Oxford: Oxford

University Press.

REFERENCES

240

Groat E. & J. O’Neill (1996). Spell-out at the interface: achieving a unified syntactic

computational system in the minimalist framework. W. Abraham, S. Epstein, H.

Thráinsson & J-W Zwart (eds.). Minimal ideas: syntactic studies in the

minimalist framework. Amsterdam: John Benjamins, 113–139.

Grohmann, K. (2000). Towards a syntactic understanding of prosodically reduced

pronouns. Theoretical Linguistics 26, 175–210

Haegeman, L. (1992). Theory and description in generative syntax: a case study in

West Flemish. Cambridge: Cambridge University Press.

Haegeman, L. (2000). Inversion, non-adjacent inversion and adjuncts in CP.

Transactions of the Philological Society 98, 121-160.

Hale, K. & S. Keyser (1993). On Argument Structure and the Lexical Expression of

Syntactic Relations. K. Hale & S. Keyser (eds.) The View from Building 20.

Cambridge MA: MIT Press, 53–110.

Hale, M. (1987). Notes on Wackernagel’s Law in the language of the Rigveda.

Calvert Watkins (ed.) Studies in memory of Warren Cowgill (1929–1985).

Berlin/New York: De Gruyter, 38–50.

Hale, M. (1995). Wackernagel’s Law in the language of the Rigveda. Unpublished

Ms, University of Concordia.

Hale, M. (1996a). Deriving Wackernagel’s Law: prosodic and syntactic factors

determining clitic placement in the language of the Rigveda. A. Halpern & A.

Zwicky (eds.), Approaching second: second position clitics and related

phenomena. Stanford CA: CSLI Publications 165–198.

Hale, M. (1996b). Theory and method in historical linguistics. Unpublished MS,

University of Concordia.

Halle, M. & A. Marantz (1993). Distributed morphology and the pieces of inflection.

K. Hale & S. J. Keyser (eds.) The view from building 20. MIT Press, Cambridge

MA, 111–176.

Halle, M. & A. Marantz (1994). Some key features of Distributed Morphology. A.

Carnie & H, Harley (eds.) Papers on phonology and morphology, MIT Working

Papers in Linguistics 21, 257–288.

Halpern, A. (1995). On the placement and morphology of clitics. Stanford CA: CSLI

Publications.

REFERENCES

241

Harley, H. (1994). Deriving the Morphosyntactic Feature Hierarchy. A. Carnie & H.

Harley (eds.) MIT Working Papers in Linguistics 21, Papers on Phonology and

Morphology, 289–320.

Harley, H. (1995). Subjects, events and licensing. PhD dissertation, MIT.

Harley, H. (2004). Merge conflation and head movement: the first sister principle

revisited. K. Moulton & M. Wolf (eds.) NELS 34. Amherst: University of

Massachusetts,

Harley, H. & R. Noyer (1999). Distributed Morphology. GLOT International 4:

3–9.

Harley, H. & E. Ritter (2002). Person and number in pronouns: A feature-geometric

analysis. Language 78, 482–526.

Haspelmath, M., M. Dryer, D. Gil & B. Comrie (2005). The world atlas of language

structures. New York: Oxford University Press.

Held, W. (1957). The Hittite relative sentence. Language Dissertation 55. Baltimore:

Waverly.

Hock, H. (1996). Who’s on first? Towards a prosodic account of P2 clitics. A.

Halpern & A. Zwicky (eds.), Approaching second: second position clitics and

related phenomena. Stanford CA: CSLI Publications 199–270.

Hoekstra, T. (1988). Small clause results. Lingua 74, 101–139.

Holmberg, A. & C. Platzack (1995). The role of inflection in Scandinavian syntax.

New York: Oxford University Press.

Hopper, & E. Traugott (1993). Grammaticalization. Cambridge: Cambridge

University Press.

Hróarsdóttir, T., G. Hrafnbjargarson, A. Wiklund & K. Bentzen (to appear). The

Tromsø guide to Scandinavian verb movement. To appear in Working Papers in

Scandinavian Syntax.

Huang, J. (1982). Logical relations in Chinese and the theory of grammar. PhD

dissertation, MIT.

Hyams, N. (1986). Language acquisition and the theory of parameters. Dordrecht: D.

Reidel.

Inkelas S. & D. Zec (1995). Syntax-phonology interface. J. Goldsmith (ed.) The

handbook of phonological theory. Oxford: Blackwell, 535–549.

Isaac, G. (2000). The most recent model of the development of absolute and conjunct

inflection. Ériu 51, 63–68.

REFERENCES

242

Isaac, G. (2001). The function and typology of absolute and conjunct flexion in early

Celtic: Some hints from Ancient Egyptian. Transactions of the Philological

Society, 145–168.

Jackson, K. (1951). Common Gaelic: The Evolution of the Gaelic Languages. Sir

John Rhys Memorial Lecture, British Academy, London.

Jackson, K. (ed.) (1990). Aislinge meic con glinne. Dublin: Dublin Institute for

Advanced Studies.

Kayne, R. (1975). French syntax: the transformational cycle. Cambridge, MA: MIT

Press.

Kayne, R. (1985). Principles of particle constructions. J. Guéron, H.-G. Obenauer &

J.-Y. Pollock (eds.) Grammatical Representations. Dordrecht: Foris,

101–140.

Kayne, R. (1989). Null subjects and clitic climbing. O. Jaeggli & K. Safir (eds.) The

null subject parameter. Dordrecht: Kluwer, 239–261.

Kayne, R. (1991). Romance clitics, verb movement and PRO. Linguistic Inquiry 33:

409–442.

Kayne, R. (1994). The antisymmetry of syntax. Cambridge, MA: MIT Press.

Kemenade, A. van (1987). Syntactic case and morphological case in the history of

English. Dordrecht: Foris.

Kiparsky, P. (1982). Explanation in Phonology. Dordrecht: Foris.

Kiparsky, P. (1995a). Indo-European Origins of Germanic Syntax. A. Battye & I.

Roberts (eds.) Clause structure and language change. New York: Oxford

University Press, 140–170.

Kiparsky, P. (1995b). The Phonological Basis of Sound Change. J. Goldsmith (ed.)

Handbook of Phonological Theory. Oxford: Blackwell, 640–670.

Kiparsky, P. (2000). Analogy as Optimization. A. Lahiri (ed.) Analogy, Levelling,

Markedness. Principles of Change in Phonology and Morphology. Berlin:

Mouton de Gruyter, 15–46.

Koch, J. (1985). Movement and emphasis in the Gaulish sentence. Bulletin of the

Board of Celtic Studies 32, 1–37.

Koch, J. (1987). Prosody and the Old Celtic verbal complex. Ériu 38, 143–176.

Koch, J. (1991). On the prehistory of Brittonic syntax. J. Fife & E. Poppe (eds.)

Studies in Brythonic word order. Amsterdam: John Benjamins, 1–43.

REFERENCES

243

Kortland, F. (1979). The Old Irish absolute and conjunct endings and questions of

relative chronology. Ériu 30, 35–53.

Kortland, F. (1994). Absolute and conjunct again. Münchener Studien zur

Sprachwissenschaft 55, 61–68.

Kroch, A. (1989). Reflexes of grammar in patterns of language change. Language

variation and change 1, 199–244.

Kroch, A. (1994). Morphosyntactic variation. Papers from the 30th meeting of the

Chicago Linguistic Society 2, 180–201.

Laka, I. (1990). Negation in syntax. On the nature of functional categories and

projections. PhD dissertation, MIT.

Landau, I. (2006). Chain resolution in Hebrew V(P) fronting. Syntax 9, 32–66.

Langacker, R. (1977). Syntactic reanalysis. C. Li (ed.) Mechanisms of syntactic

change. Austin: University of Texas Press, 57–139.

Lasnik, H. (1981). Restricting the theory of transformations: A case study. N.

Hornstein & D. Lightfoot (eds.) Explanations in linguistics. London: Longman.

Lasnik, Howard (1995) Verbal morphology: syntactic structures meets the minimalist

program. H. Campos & P. Kempchinsky (eds.) Evolution and revolution in

linguistic theory: Essays in honor of Carlos Otero. Washington DC:

Georgetown University Press, 251–275. [Reprinted in Lasnik, H. (1999).

Minimalist analysis. Oxford: Blackwell.]

Lechner, W. (2004). Ellipsis in comparatives. Berlin: Mouton de Gruyter.

Lechner, W. (2005). Interpretive effects of head movement. Ms., University of

Tübingen.

Lee, F. (2000). VP remnant movement and VSO in Quiavini Zapotec. A. Carnie & E.

Guilfoyle (eds.), The syntax of verb-initial languages. New York: Oxford


Lee, F. (2005). Force first: clause-fronting and clause typing in San Lucas Quiaviní

Zapotec. A. Carnie, H, Harley & S. Dooley (eds.) Verb first: on the syntax of

verb-initial languages. Amsterdam: John Benjamins, 91–106.

Lejeune, M. (1988). Recuil des inscriptions gauloises ii/1, textes gallo-étrusques,

textes gallo-latins sur pierre. Paris: Centre National de la Recherche

Scientifique.

Lightfoot, D. (1989). The child’s trigger experience: degree-0 learnability.

Behavioural and Brain Sciences 12, 321–334.

REFERENCES

244

Lightfoot, D. (1991). How to set parameters: Arguments from language change.

Cambridge, MA: MIT Press.

Lightfoot, D. (1999). The development of language: Acquisition, change and

evolution. Oxford: Blackwell.

Lightfoot, D. (2006). How new languages emerge. Cambridge: Cambridge University

Press.

Lightfoot, D. & N. Hornstein (1994). Verb Movement. Cambridge: Cambridge

University Press.

Luraghi, S. (1990). Old Hittite sentence structure. London: Routledge.

McCloskey, J. (1978). Stress and syntax in Old Irish compound verbs. S. Schmerling

& C. Smith (eds.) Texas Linguistic Forum 10. Austin, Texas: Department of

Linguistics, University of Texas at Austin, 58–71.

McCloskey, J. (1991). Clause Structure, Ellipsis and Proper Government in Irish.

Lingua 85, 259–302.

McCloskey, J. (1996a). On the Scope of Verb Movement in Irish. Natural Language

and Linguistic Theory 14, 47–104.

McCloskey, J. (1996b). Subjects and subject positions in Irish. R. Borsley & I.

Roberts (eds.) The Syntax of the Celtic Languages. Cambridge: Cambridge


McCloskey, J. (1999). On the right edge in Irish. Syntax 2, 189–209.

McCloskey, J. (2001a). On the distribution of subject properties in Irish. W. D.

Davies & S. Dubinsky (eds.) Objects and other subjects. Dordrecht/Boston:

Kluwer Academic Press, 157–192.

McCloskey, J. (2001b). The morphosyntax of wh-extraction in Irish. Journal of

Linguistics 37: 67–100.

McCloskey, J. (2002). Resumption, successive cyclicity, and the locality of

operations. S. Epstein & D. Seely (eds.) Derivation and explanation in the

minimalist program. Malden, MA: Blackwell, 184–226.

McCloskey, J. (2005). A note on predicates and heads in Irish clausal syntax. A.

Carnie, H, Harley & S. Dooley (eds.) Verb first: on the syntax of verb-initial

languages. Amsterdam: John Benjamins, 155–174.

McCone, K. (1978). The dative singular of Old Irish consonant stems. Ériu 29, 26–38. McCone, K. (1979a). Aspects of Indo-European sentence patterns and their role in the

constitution of the Old Irish verbal system. DPhil dissertation, University of

REFERENCES

245

Oxford. McCone, K. (1979b). Pretonic preverbs and the absolute verbal endings in Old Irish.

Ériu, 30, 1–34.

McCone, K. (1979c). The diachronic possibilities of the Indo-European ‘amplified’

sentence: A case history from Anatolian. Brogyani, B. (ed.), Studies in

Diachronic, Synchronic, and Typological Linguistics: Festschrift for Oswald

Szemérenyi on the Occasion of his 65th Birthday. Amsterdam: John Benjamins,

467–488.

McCone, K. (1980). The nasalizing relative clause with object antecedent in the

glosses. Ériu 31, 10–27.

McCone, K. (1982). Further to absolute and conjunct. Ériu 33, 1–29.

McCone, K. (1985). The absolute and conjunct verbal inflection in Old Irish. B.

Schlerath (ed.), Grammatische Kategorien: Funktion und Geschichte.

Wiesbaden: Reichert, 261–270.

McCone, K. (1996). Towards a chronology of ancient and medieval Celtic sound

changes. Maynooth Studies in Celtic Linguistics I.

McCone, K. (1997a). The Early Irish Verb. Maynooth: An Sagart.

McCone, K. (1997b). Delbrück’s model of PIE word order and the Celtic evidence. E.

Crespo & J. Ramon (eds.) Berthold Delbrück y la sintaxis indoeuropea hoy:

actas del coloquio de la Indogermanische Gesellschaft, Madrid 21–24 de

septimbre de 1994. Madrid: Wiesbaden, 363–396.

McDaniel, D. (1989). Partial and multiple wh-movement. Natural Language and

Linguistic Theory 7, 565–604.

MacDonell, A. (1916). Vedic grammar for students. Delhi: Oxford University Press.

Mac Eoin, G. (1961a). On the date and authorship of Saltair na Rann. Zeitschrift für

Celtische Philologie 28, 51–67.

Mac Eoin, G. (1961b). Das Verbalsystem von Togail Troi. Zeitschrift für celtische

Philologie 28, 73–136, 149–233.

Mac Gearailt, U. (1997–1998). Infixed and independent pronouns in the LL text of

Táin Bo Cuailnge. Zeitschrift für Celtische Philologie 49–50, 494–515.

McGonagle, N. (1988) Irish grammar: a basic handbook. Conamara: Cló Iar-

Chonnachta.

McManus, D. (1991). A guide to Ogam. Maynooth: An Sagart.

REFERENCES

246

McManus, D. (1994). An Nua-Ghaeilge Chlasaiceach. K. McCone, D. McManus, C.

Ó Háinle & L. Breatnach (eds.), Stair na Gaeilge in ómós do Phádraig Ó

Fiannachta. Maynooth: Department of Old Irish, St Patrick’s College, 335–446.

Mahajan, A. (2000). Eliminating head movement. GLOW Newsletter 44: 44–5.

Mahajan, A. (2003). Word order and (remnant) VP movement. S. Karimi (ed.) Word

order and scrambling. Oxford: Blackwell, 217–237.

Manetta, E. (2006). Peripheries in Kashmiri and Hindi-Urdu. PhD dissertation,

University of California, Santa Cruz.

Marantz, A. (1988). Clitics, morphological merger and the mapping to phonological

structure. M. Hammond & M. Noonan (eds.) Theoretical Morphology. San

Diego: Academic Press, 253–270.

Marantz, A. (1997). No escape from syntax. Don’t try a morphological analysis in the

privacy of your own lexicon. Proceedings of the 21st annual Penn Linguistics

Colloquium. Penn Working Papers in Linguistics 4, 201–225.

Massam, D. (2000). VSO and VOS: aspects of Niuean word order. A. Carnie & E.

Guilfoyle (eds.), The syntax of verb-initial languages. New York: Oxford


Massam, D. (2005). Lexical categories, lack of inflection and predicate fronting in

Niuean. A. Carnie, H, Harley & S. Dooley (eds.) Verb first: on the syntax of

verb-initial languages. Amsterdam: John Benjamins, 227–242.

Matić, D. (2003). Topic, focus and discourse structure: Ancient Greek word order.

Studies in Language 27, 573–633.

Matushansky, O. (2002). Moving a-head. K. Hiraiwa & J. Sabbagh (eds.), Minimalist

Approaches to Clause Structure, MIT Working Papers in Linguistics 50.

Matushansky, O. (2006). Head movement in linguistic theory. Linguistic Inquiry 37:

69–109.

Müller, G. (2004). Phase impenetrability and wh-intervention. A. Stepanov, G.

Fanselow, & R. Vogel (eds.) Minimality Effects in Syntax. Berlin: Mouton de

Gruyter, 289–325.

Neeleman, A. (1994). Complex predicates. PhD dissertation, University of Utrecht.

Newton, G. (2005). The Old Irish double system of verbal inflection: towards a

synchronic analysis. F. Chalcraft & E. Sipetzis (eds.), Cambridge Occasional

Papers in Linguistics 2, 149–172.

REFERENCES

247

Newton, G. (2006). The contribution of the Old Irish verbal system to the

reconstruction of Proto-Indo-European syntax. N. Yanez-Bouzita (ed.) Papers

in Linguistics of the University of Manchester (PLUM) 2005, 95–112.

Noyer, R. (1997). Features, Positions and Affixes in Autonomous Morphological

Structure. New York: Garland Publishing.

Nuner, R. (1958–9). The verbal system of the Agallamh na Senórach. Zeitschrift für

celtische Philologie 27, 230–309.

Nunes, J. (1999). Linearization of chains and phonetic realization of chain links. S.

Epstein & N. Hornstein (eds.) Working minimalism. Cambridge MA: MIT

Press.

Nunes, J. (2004) Linearization of chains and sideward movement. Cambridge MA:

MIT Press.

Oda, K. (2005). V1 and wh-questions: a typology. A. Carnie, H, Harley & S. Dooley

(eds.) Verb first: on the syntax of verb-initial languages. Amsterdam: John

Benjamins, 107–134.

O’Daly, M. (1943). The verbal system of the LL Táin. Ériu 14, 31–139.

O’Rahilly, T. (1932). Irish dialects past and present. Dublin: Dublin Institute for

Advanced Studies.

Ó Siadhail, M. (1991). Modern Irish: grammatical structure and dialectal variation.

Cambridge: Cambridge University Press.

Ó hUiginn, R. (1986). The Old Irish nasalizing relative clause. Ériu 37, 33–87.

Ó hUiginn, R. (1997). Aspects of clause subordination in Celtic. F. Josephson (ed.)

Celts and Vikings: proceedings of the Fourth Symposium of Societas

Celtologica Nordica. Göteborg : Meijerbergs institut för svensk etymologisk

forskning, 69–87.

Pedersen, H. (1909–1913). Vergleichende Grammatik der keltischen Sprachen.

Göttingen: Vandenhoeck und Ruprecht.

Pesetsky, D. (1997). Optimality theory and syntax: movement and pronunciation. D.

Archangeli & T. Langendoen (eds.) Optimality theory: an overview. Malden,

MA: Blackwell, 134–170.

Pesetsky, D. & E. Torrego (2001). T-to-C movement: causes and consequences. M.

Kenstowicz (ed.), Ken Hale: a life in language. Cambridge MA: MIT Press,

355–426.

Pintzuk, S. (1999). Phrase Structures in Competition: Variation and Change in Old

REFERENCES

248

English Word Order. New York: Garland.

Pollock, J-Y (1989). Verb movement, universal grammar and the structure of IP.

Linguistic Inquiry 20, 365–424.

Quin, E. (1975). Old Irish workbook. Dublin: Royal Irish Academy.

Rackowski, A. & L. Travis (2000). V-initial languages: X or XP movement and

adverbial placement. A. Carnie & E. Guilfoyle (eds.), The syntax of verb-initial

languages. New York: Oxford University Press, 117–141.

Renou, L. (1933). La separation du préverbe et du verbe en védique. BSL 34, 49–96.

Richards, M. (2004). Object shift and scrambling in North and West Germanic: a

case study in symmetrical syntax. PhD dissertation, University of Cambridge.

Rizzi, L. (1990) Relativized Minimality Cambridge MA: MIT Press

Rizzi, L. (1997). The fine structure of the left periphery. L. Haegeman (ed.), Elements

of grammar: handbook of generative grammar. Dordrecht: Kluwer 281–338.

Roberts, I. (1985). Agreement parameters and the development of English modal

auxiliaries. Natural Language and Linguistic Theory 3, 21–58.

Roberts, I. (1991). Excorporation and minimality. Linguistic Inquiry 22, 209–18.

Roberts, I. (1993). Verbs and diachronic syntax: a comparative history of English and

French. Dordrecht: Kluwer Academic.

Roberts, I. (1999). Verb movement and markedness. M. deGraff (ed.) Language

creation and language change. Cambridge, MA: MIT Press, 287–328.

Roberts, I. (2005). Principles and parameters in a VSO language: A case study in

Welsh. Oxford: Oxford University Press.

Roberts, I. & A. Roussou (2003). Syntactic change: a minimalist approach to

grammaticalization. Cambridge: Cambridge University Press.

Rodway, S. (1998). Absolute forms in the poetry of the Gogynfeirdd: functionally

obsolete archaisms or working system? Journal of Celtic Linguistics 7, 63–84.

Rohrbacher, B. (1999). Morphology-driven syntax: a theory of V-to-I raising and pro-

drop. Amsterdam: John Benjamins.

Russell, P. (1988). Preverbs, Prepositions and Adverbs: Sigmatic and Asigmatic.

Transactions of the Philological Society, 144–172.

Russell, P. (1995). An introduction to the Celtic languages. London: Longman.

Sabel, J. (2000). Partial wh-movement and the typology of wh-questions. U. Lutz, G.

Müller & A. von Stechow (eds.), Wh-scope marking. Amsterdam: John

Benjamins, 409–446.

REFERENCES

249

Saito, M. & N. Fukui (1998). Order in phrase structure and movement. Linguistic

Inquiry 29, 439–474.

Schrijver, P. (1994). The Celtic adverbs for ‘against’ and ‘with’ and the early apocope

of *-i. Ériu 45, 151–189.

Schrijver, P. (1997). Studies in the history of Celtic particles and pronouns.

Maynooth Studies in Celtic Linguistics II.

Selkirk, E. (1986). On derived domains in sentence phonology. Phonology 3, 371–

405.

Selkirk, E. (1995). The prosodic structure of function words. J. Beckman, L. Dickey

& S. Urbanczyk (eds.) Papers in optimality theory. Amherst: GLSA, 439–69.

Senn, A. (1966). Handbuch der litauischen Sprache. Heidelberg: Carl Winters

Universitätsverlag.

Sims-Williams, P. (1984). The double system of verbal inflection in Old Irish.

Transactions of the Philological Society, 138–201.

Sommerfelt, A. (1915–21). Le système verbal dans In Cath Catharda. Revue Celtique

36, 24–62, 295–334 & 37, 230–246 & 38, 25–47.

Sportiche, D. (1992). Clitic Constructions. L. Zaring & J. Rooryck (eds.) Phrase

Structure and the Lexicon. Dordrecht: Kluwer Academic Publishers.

Stephens, J. (1982). Word order in Breton. PhD dissertation, University of London.

Stokes, W. & J. Strachan (1901–1903). Thesaurus palaeohibernicus: a collection of

old-Irish glosses, scholia, prose, and verse. Dublin: Dublin Institute for

Advanced Studies.

Strachan, J. (1895–7). The verbal system of Saltair na Rann. Transactions of the

Philological Society, 1–75.

Strachan, J. (1904). The infixed pronoun in Middle Irish. Ériu 1, 153–179.

Taylor, A. (1990). Clitics and configurationality in Ancient Greek. PhD dissertation,

University of Pennsylvania.

Thiersch, C. (1985). VP Scrambling and the German Mittelfeld. Unpublished ms:

University of Tilburg

Thráinsson, H. (2003). Syntactic Variation, Historical Development, and Minimalism.

Hendrick, R. (ed.). Minimalist Syntax. Oxford: Blackwell, 152-191.

Thurneysen, R. (1907). On certain initial changes in the Irish verb after preverbal

particles. Ériu 3, 18–19.

REFERENCES

250

Thurneysen, R. (1946). A Grammar of Old Irish: An Enlarged Edition with

Supplement translated by Daniel A Binchy & Osborn Bergin. Dublin: Dublin

Institute for Advanced Studies

Timberlake, A. (1977). Reanalysis and actualization in syntactic change. C. Li (ed.)

Mechanisms of syntactic change. Austin: University of Texas Press, 141–77.

Travis, L. (1984) Parameters and Effects of Word Order Variation. PhD Dissertation,

MIT.

Truckenbrodt, H. (1995). Phonological phrases: Their relation to syntax, focus and

prominence. PhD Dissertation, MIT.

Uriagereka, Juan (1995) Aspects of the syntax of clitic placement in Western

Romance. Linguistic Inquiry 26, 79–123.

Vance, B. (1988). Null subjects and syntactic change in Medieval French. PhD

dissertation, Cornell University.

Vendryes, J. (1908) Grammaire du vieil-irlandais. Paris.

Vennemann, T. (2002). Semitic > Celtic > English: The transitivity of language

contact. M. Filppula, J. Klemola & H. Pitkänen (eds.), The Celtic roots of

English (Studies in Languages 37). Joensuu: University of Joensuu, Faculty of

Humanities, 295–330.

Vikner, S. (1995). Verb Movement and Expletive Subjects in the Germanic

Languages. New York: Oxford University Press.

Vikner, S. (1997). V-to-I movement and inflection for person in all tenses. L.

Haegeman (ed.) The new comparative syntax. London: Longman, 189–213.

Vincent, N. (1999). The evolution of c-structure: prepositions and PPs from Indo-

European to Romance. Linguistics 37, 1111–1153.

Wackernagel, J. (1892). Über ein Gesetz der Indogermanischen Wortstellung.

Indogermanische Forschungen 1, 333–436.

Watkins, C. (1963). Preliminaries to a historical and comparative analysis of the

syntax of the Old Irish verb. Celtica 6, 1–49.

Watkins, C. (1976). Towards Proto-Indo-European syntax: problems and pseudo-

problems. S. Steever, C. Walker and S. Mufwene (eds.), Papers from the

parasession on diachronic syntax. Chicago: Chicago Linguistic Society, 305–

326.

Watkins, C. (2001). How to kill a dragon: aspects of Indo-European poetics. New

York: Oxford University Press.

REFERENCES

251

Weinreich, U., W. Labov & W. Herzog (1968). Empirical foundations for a theory of

language change. W. Lehmann & Y. Malkiel (eds.) Directions for historical

linguistics. Austin: University of Texas Press. 95–195.

Westergaard, M. (to appear). Triggering V2: The Amount of Input Needed for

Parameter Setting in a Split-CP Model of Word Order. A. Belletti, E. Bennati,

C. Chesi, E. DiDomenico & I. Ferrari (eds.) Language Acquisiton and

Development: Proceedings of GALA 2005. Cambridge: Cambridge Scholars

Press. 658-671.

Wilford, E. (2005). An analysis of the Breton auxiliary system. MPhil dissertation,

University of Cambridge.

Willis, D. (1998). Syntactic change in Welsh: a study of the loss of verb-second

Oxford: Clarendon Press.

Wurmbrand, S. (2000). The structure(s) of particle verbs. Unpublished ms, McGill

University.

Zanuttini, R. (1997). Negation and clausal structure: A comparative study of

Romance languages. New York: Oxford University Press.

Zeller, Jochen (2001) Particle Verbs and Local Domains Amsterdam: John

Benjamins.

Zwart, J-W. (1997). Morphosyntax of Verb Movement. A Minimalist Approach to the

Syntax of Dutch. Dordrecht: Kluwer Academic Publishers.

Zwart, J-W. (2001). Syntactic and phonological verb movement. Syntax 4, 34–62.

Zwicky, A. & G. Pullum (1983). Cliticization vs. inflection: English n’t. Language

59, 502–513.

The development and loss of the Old Irish double system of verbal inflection

Documents

Transcript of The development and loss of the Old Irish double system of verbal inflection