Document not found! Please try again

MORPHOSYNTACTIC FEATURE CHAINS AND ...

5 downloads 0 Views 1013KB Size Report
I knife (to) brother gave. Although it is not apparent that this language makes use of the edge condition,. Kaisse claims that the c-command condition is ...
MORPHOSYNTACTIC FEATURE CHAINS AND PHONOLOGICAL DOMAINS by Gorka Elordieta

______________

A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (Linguistics)

August 1997

Copyright 1997

Gorka Elordieta

3

Table of Contents

Acknowledgments.............................................................................................. Abstract............................................................................................................. Chapter 1: Introduction...................................................................................... 1. The main issue.................................................................................... 2. Brief overview of analyses in Prosodic Phonology............................... 3. Problems for Prosodic Phonology....................................................... 4. The proposal...................................................................................... 4.1. The theoretical framework................................................... 4.2. Mapping of feature chains onto morphosyntactic constituency........................................................................ 4.2.1. Feature chains as stem plus affixes.............................. 4.2.2. Feature checking as independent words...................... 4.2.3. Feature chains as morphosyntactic words not composed of stem plus affixes.................................... 5. Summary and organization of the dissertation.................................... Chapter 2: Feature licensing, morphological constituency and phonological domains in Basque................................................................................ 1. Introduction...................................................................................... 2. VA in Lekeitio Basque: data and generalizations............................... 2.1. Preliminary data................................................................. 2.2. Morphosyntactic distribution of VA................................... 2.2.1. Nominal contexts..................................................... 2.2.1.1. Nominal inflection................................................. 2.2.1.2. Roots.................................................................... 2.2.1.3. Derivational morphemes, compounding and word boundaries.................................................................... 55 2.2.2. Verbal contexts....................................................... 2.2.2.1. Inflected auxiliaries.............................................. 2.2.2.2. Causative verbs.................................................... 2.2.2.3. Modal particles.................................................... 2.2.2.4. Subordinating conjunctions.................................. 2.2.2.5. Across any two words......................................... 3. Clause structure, head movement and Infl licensing in Basque......... 3.1. Clause structure............................................................... 66 3.2. Head movement............................................................... 3.3. Infl (T) licensing.............................................................. 3.4. Possible alternative phrase structures............................... 3.4.1. Verb-to-Aspect...................................................... 3.4.2. Long head movement............................................. 3.5. Determiners...................................................................... 4. An analysis in terms of feature chains..............................................

v viii 1 1 3 7 15 15 30 32 40 42 44

46 46 48 48 51 51 51 55

56 56 62 62 63 64 66 69 72 87 88 98 107 112

4

4.1. Mapping between morphological domains and phonological domains................................................................................... 4.2. Empirical analysis of VA.................................................. 4.3. Supporting evidence:/n/-deletion...................................... 5. A puzzle for theories of phrasal phonology..................................... 5.1. The P1/P2 distinction among postlexical rules................... 5.2. Direct Reference Theory................................................... 5.3. Precompiled Phrasal Phonology........................................ 5.4. Prosodic Hierarchy Theory............................................... 5.4.1. The Relation-Based Approach................................ 5.4.1.1. Overview of the basic postulates and prosodic constituents...................................................................... 5.4.1.2. Predictions on the domain of application of VA.. 5.4.2. End-Based Approach............................................. 5.4.2.1. Mapping from syntactic structure........................ 5.4.2.2. Predictions on the domain of application of VA.. 5.5. Summary........................................................................ 200 Chapter 3: Additional evidence for the creation of phonological domains from morphosyntactic feature chains................................................. 202 1. Introduction.................................................................................. 202 2. French liaison............................................................................... 204 2.1. Representation of liaison consonants.............................. 205 2.2. Liaison and styles of speech........................................... 209 2.3. Distribution of liaison.................................................... 211 2.3.1. Obligatory liaison................................................ 2.3.2. Frequent liaison................................................... 2.3.3. Rare liaison......................................................... 2.3.4. Absence of liaison............................................... 2.4. Previous analyses.......................................................... 220 2.4.1. Selkirk (1972)...................................................... 2.4.2. Rotenberg (1978)................................................. 2.4.3. Direct Reference Theory...................................... 2.4.2. The End-Based Approach..................................... 2.4.4.1. Selkirk (1986).................................................... 2.4.4.2. De Jong (1990).................................................. 2.5. An analysis of obligatory liaison in terms of morphosyntactic feature chains..............................................236 2.6. Possible alternative: agreement relationships................... 2.7. Optional liaison.............................................................. 249 2.8. Prenominal adjectives.................................................... 253 2.9. Summary....................................................................... 253 3. Initial Consonant Mutation in Modern Irish.................................. 255 3.1. Introduction.................................................................. 255 3.2. Historical background on ICM in Modern Irish................. 3.3. Distribution of ICM in Modern Irish................................. 3.3.1. Verbal contexts......................................................

119 121 134 137 138 147 151 159 165 165 180 185 185 193

211 215 217 219 220 222 223 229 229 233

246

258 265 265

4

3.3.2. Nominal contexts................................................... 3.3.2.1. Determiners........................................................ 3.3.2.2. Possessive pronouns........................................... 3.3.2.3. Prepositions........................................................ 3.4. Previous proposals.......................................................... 271 3.4.1. Monphophonological analyses............................... 3.4.1.1. Triggered mutation............................................. 3.4.1.2. Inherent mutation.............................................. 3.5. ICM as syntactically determined...................................... 3.5.1. Complementizers and eclipsis................................. 3.5.2. Tense and lenition................................................. 3.6. ICM as an effect of morphosyntactic licensing................ References ................................................................................................... 289

267 267 269 270 271 272 273 276 276 277 284

Acknowledgments

There are many people I want to thank for making this dissertation possible. First and foremost, my advisor and committee chair, Jean-Roger Vergnaud, without whose advice and support I would never have even started to think about the issue investigated in this dissertation. Some of the central ideas in this work originate in discussions with him, so I will not be able to thank him enough. I admire Jean-Roger’s intelligence and his “big picture” view of linguistic theory, and his ability to relate facts of language to interfaces among different components of grammar. I deeply share this interest of his, especially regarding the interfaces of syntax, morphology and phonology. The theme of this dissertation is a good example of it. One important facet of my relationship with Jean-Roger is that he always believed in me and in the importance of the ideas expressed in this thesis, lifting me up when I hit hard times reconciling disparate arrays of data from totally unrelated languages, in what seemed a hopeless task. Those marathonian appointments in his office at Larchmont with coffee and pastries will not be easily forgotten. I must also thank the other members of my committee, Joseph Aoun, Alicja Gorecka, Mario Saltarelli and María Luisa Zubizarreta, for helpful comments, suggestions and criticism. Joseph has been a great teacher, an inspiring force really and one of the reasons why I chose the department of linguistics at USC to pursue my doctoral studies. I still wonder at his knowledge of syntactic theory, his energy and his ability to exchange in discussions from one topic or question to another with a speed that was awfully hard to catch on. It was impossible to snooze in his classes, even straight after lunch. If you dosed off, there was always a wake-up call in the form of a question or jokes about pizzas under your chair (?!). Alicja has been extremely valuable for me in acquiring the knowledge I have of phonological theory. Her careful and well prepared presentations of material are a model

of organization, clearness, and efficient teaching. She has also been very supportive of me throughout my studies, at a personal level as well, for which I will be thankful forever. Her criticisms and scepticism have stuck with me more than she thinks, pushing me to be more convincing in my arguments. María Luisa deserves my gratitude for her kind support all these years, from the moment I thought about applying to the linguistics department. I will always remember her advice not to tinker with analyses and to go deeper into the problem in a more unifying way. I hope the final version of this dissertation is worthy of her requests. Finally, I am thankful to Mario Saltarelli, the external member of my committee, for help that was much beyond the call of duty. He provided me with valuable hints and suggestions stemming from his expertise in both syntax and phonology. I simply wish I had had the opportunity to incorporate more of his suggestions into the dissertation. More professors in the Department of Linguistics at USC have given me the chance to become a well-rounded linguist. If I have failed to become one, it’s not their fault. They are Bernard Comrie, Hajime Hoji, Barry Schein, and Roger Woodard. Although I never took classes with them, Ed Finegan, Jack Hawkins and Bill Rutherford also provided a comfortable and friendly atmosphere. During these past years, I have met many graduate students in the department. All have contributed to making my stay at USC and LA as pleasant as possible. Special thanks to my good friends José Camacho, Liliana Sánchez, and above all, Magdalena Romera. I have learned a lot from you guys. I also want to thank the staff in the department of linguistics (Linda CulverWilliams, Laura Reiter, Kathy Stubaus, Don Bui and Jessica Hamilton) for being such great people and so efficient. More than anything, I want to thank them for their friendliness. That is one of the distinguishing characteristics of the department. My first experience with linguistics in the United States took place in the Department of Spanish and Portuguese at USC. There I obtained my Master’s degree in

Spanish, with a specialization in Spanish linguistics. I want to say thank you to everybody who, in one way or another, made it possible for me to embark in a further adventure into the Department of Linguistics at USC. In random order: Mario Saltarelli, Carmen Silva-Corvalán, Enric Vallduví, Ramón Araluce, Adalberto Salas, Eduardo Raposo and especially María Carreira. Finally, there are the people that are almost as happy as I am that I am finishing this six-year long (plus two more in the Spanish department) ordeal: my family back there in the Basque Country. It’s been tough on them, too. Those weekly letters from my mother telling me about family matters and life in my hometown, Lekeitio, as well as the customary clips reporting Athletic de Bilbao’s last game have been an irreplaceable link back home.

1

Abstract

Morphosyntactic feature chains and phonological domains The main objective in this dissertation is to provide a formal analysis of one aspect of the mapping between syntactic structure and morphophonological structure. It is argued that the relationships of feature specifications in the different syntactic heads driving the operation of feature checking (cf. Chomsky 1995) are primitive morphosyntactic entities in the form of feature chains: C-T, T-V, T-D, V-D, D-N, P-D. These chains are mapped to Morphological Structure and are realized morphologically in different ways, depending on the lexical specifications of the features in the chain. A feature chain may be spelled out as a single morphological word (m-word), with the shape stem+affixes, if one of the heads contains the feature [Affix]. The m-word is typically mapped as a single phonological domain, where phonological processes specified to apply in a phonological word will apply, giving rise to the effects of phonological interaction between stems and affixes. A case study showing intricate interactions between the affixal feature and categorial feature attraction is provided in chapter 1, with the phenomenon of Vowel Assimilation in Lekeitio Basque. Another possibility in the syntax-MS mapping is that neither of the members in a feature chain is an affix. In this case, there are two alternatives available in the system. First, that the heads containing the features may be realized as independent words. Each head is an m-word, and thus by default also a separate phonological word, with the consequence that no phonological rules applying within a phonological domain will affect the two heads. This is instantiated by tense particles in Vata, and by prepositions in French. Second, that the morphosyntactic cohesion established by the feature chain and instantiated by the operation of feature checking relating the features and the heads containing them is realized as a cohesive constituent at MS, as a morphosyntactic word (ms-word). This constituent is then mapped as a single phonological domain. Chapter 3

2

contains discussions of these two possible scenarios, analyzing French liaison and Irish Consonant Mutation. This system provides a unified treatment of a wide range of phenomena, explaining cases that Prosodic Phonology alone cannot capture satisfactorily.

1

Chapter 1 Introduction

1. The main issue. 1.1. The postlexical phonology of functional categories. It is a commonly observed fact among languages that functional and inflectional categories (i.e., determiners, complementizers, agreement and case, tense, aspect, etc.) very often group together with lexical categories (i.e., nouns, adjectives, verbs) forming domains for the application of phonological processes. These facts are most obvious when functional and inflectional categories are contained within a “word”, a linguistic unit that traditionally has been understood as a unit composed of a stem and all affixes attached to it. The framework of Lexical Phonology and Morphology is set to provide a formal account of these relationships (cf. Pesetsky 1979, Mohanan 1982, 1986, Kiparsky 1982, 1985, Kaisse and Shaw 1985, among others). But functional categories may also combine phonologically with adjacent words even when they are realized as separate linguistic units of a certain type. For instance, determiners which are expressed as separate words more often than not combine with the following noun or adjective to form one domain of stress assignment, and they may also enter into phonological processes which affect the segmental or featural structure of any of the combined elements. Prepositions, tense and aspect markers, complementizers, and other function words are found to display similar patterns across languages. Selkirk and Shen (1990) analyze several rules applying across words in Shanghai Chinese, and conclude that a combination of a lexical word and a function word composes a phonological word (i.e., a prosodic word or PWd, in their sense). It is lexical words that project lefthand PWd boundaries, not function words, and thus the latter fall within the PWd domain extending from the left edge of one lexical word to the left edge of the next. The rules of Obligatory Tone Deletion, Left-to-Right Tone association, and

2

Default Tone Insertion apply within a PWd. Schematically, without reproducing the actual tone realizations, (1) illustrates a minimal pair between a verb phrase in which the direct object is a pronoun and forms a phonological word with the verb and a verb phrase with a lexical direct object, which forms its own prosodic word. Additional examples are provided in (2), with the labeled structures to the left of the arrow representing syntactic structure and the representations to the right of the arrow illustrating the division of the string in phonological words: (1) a.

b.

(2) a.

b.

(V Pro Prt) taN ‘noN leq hit you has ‘has hit you’ (V) (N Prt) taN ‘mo leq hit horse has ‘has hit the horse’ V [ P [N]NP ]PP ‘z ‘laq ‘zawNhe live in Shanghai ‘live in Shanghai’



V [ Q Classif. N ]NP taw ?iq pe ‘zo pour one/a cup tea ‘pour a cup of tea’

(V P) (N)



(V Q Class.) (N)

1.2. Brief overview of analyses in the Prosodic Hierarchy Theory. Over the last fifteen years or so, several theories have been proposed dealing with the interaction between phonology and morphology on the one hand, and phonology and syntax on the other, and facts like the ones mentioned above have been part or their object of study. The theory of Lexical Phonology and Morphology (cf. Pesetsky 1979, Kiparsky 1982, 1985, Mohanan 1982, 1986) deals with the interaction between

3

morphology and phonology, and the theories of phrasal phonology such as the ones originally proposed by Kaisse (1985), Nespor and Vogel (1982, 1986) and Selkirk (1986) occupy themselves with the mapping between syntactic structure and phonological or prosodic structure. Lexical Phonology concentrates mainly on phonological processes occurring inside the word domain, across morpheme boundaries. These are the lexical rules, or rules which apply in the lexical component (i.e., the lexicon). They are claimed to have different properties from phonological rules that apply across word boundaries, called postlexical rules, or rules which apply in the postlexical component, i.e., after all syntactic operations have taken place. Postlexical rules are the focus of the theories of the syntax-phonology interface. The Prosodic Hierarchy Theory defends the view that syntactic and phonological representations are not isomorphic and that there is a distinct level of representation called Prosodic Structure which mediates between the syntactic and phonological components. This level contains a hierarchically organized set of prosodic constituents, built from syntactic structure by a finite set of parametrized algorithms of prosodic constituent formation. Phonological operations themselves do not refer to syntactic constituents, but to the already created prosodic constituents. Two main approaches have developed from this view: the End-Based approach (cf. Chen 1987, Selkirk 1986) and the Relation-Based Approach (cf. Nespor and Vogel 1982, 1986, Hayes 1989). The different models within the Prosodic Hierarchy Theory will be reviewed in some detail in chapter 2, but let us present here a brief sketch of how the End-Based Approach of Selkirk (1986) accounts for the data in (1)-(2). Basically, a theory is devised in which prosodic structure is built from syntactic structure by inserting lefthand or righthand prosodic edges or boundaries coinciding with the ledfthand or righthand edges of words or maximal projections in syntax. Thus, the following algorithms for prosodic constituent formation are proposed:

4

(3)

End parameter settings: (i) a. (ii) a.

]Word ]Xmax

b. b.

Word [ Xmax

[

(i) and (ii) present two parameters each, a matter of arbitrary selection among languages. In order to define a prosodic word, a language X may choose to select option (ia), that is, inserting boundaries to the right of words, understood as lexical syntactic heads (cf. the discussion below), whereas language Y may select option (ib), so that prosodic boundaries are inserted to the left of syntactic heads. The same situation arises for phonological phrases. The string of phonetically realized material enclosed between two word boundaries forms a prosodic word, and the material enclosed between two maximal projection boundaries forms a phonological phrase. The following diagram illustrates this mapping from syntactic to prosodic structure: (4) a.

b. c. d.

S qp NP VP 9 3 fw N PP ? NP ! ! 2 2 ! ! ! fw NP V NP N ! ! ! ! ! ! ! ! ! ! N ! N ! ! ! ! ! ! ! ! ...........]w.............]w......]w.........]w........]w ...........................]Xmax .............]Xmax ...]Xmax (_____) (_______) (__) (____) (____) (______________) (________) (____)

Pwd PPh

A crucial assumption in the the theory is that only lexical words project prosodic boundaries, not function words, as seen in the diagram in (4). Function words are deemed ”invisible” to prosodic word formation. According to Selkirk, this captures the fact that

5

tey are usually stressless or lean on adjacent categories to receive stress, and they interact phonologically with lexical words in way that two lexical words do not. They participate in phonological processes affecting that are not observed at the boundaries of two lexical words, for example. This property is stated in the Principle of Categorial Invisibility of Function Words (PCI) of Selkirk (1984), which says that “any theory must assume that function words are not identifiable as constituents of any sort in the syntax. Whether heads or not, the bracketing of function words is ‘invisible’ to the rules deriving phonological domains” (Selkirk 1986:396). As a consequence of the PCI, it is argued that function words lack the particular features of the metrical grid alignment of “real” words, and from these peculiarities of metrical grid alignment the phonological properties of function words follow automatically. Selkirk’s analysis is that function words are not assigned the silent demibeat of syntactic timing that nonfunction words receive in the syntax-phonology mapping (cf. Selkirk 1984:section 6.2.1 for general discussion), and that they are not assigned a third-level “main word stress” (cf. Selkirk 1984, sections 3.2.4 and 4.1.1). This is what is represented in the cases in (1)-(2). Shanghai Chinese would choose option (ib) for prosodic word formation. That is, a lefthand prosodic boundary is inserted at the left edge of lexical categories, and pronouns, prepositions, classifier particles, complementizers, auxiliaries and other functional categories are invisible for such insertion. They are therefore integrated in one prosodic word with adjacent lexical verbs or nouns to their left. The processes of tone sandhi mentioned above apply in this domain: (5) a. b. c. d.

V V V V

Pro N P Q

Prt Prt N Classif.

   N 

(V Pro Prt (V (N Prt (V P (N (V Q Class. (N

   

(V Pro Prt) (V) (N Prt) (V P) (N) (V Q Class.) (N)

6

As we will show in this dissertation, however, the facts of the syntax-phonology interface are somewhat more complex than what this model predicts, as there are phonological phenomena that do not obey to these principles of the mapping between syntax and phonology. In this chapter we will briefly provide two examples, which show that the Prosodic Hierarchy Theory as it is devised is not sufficient to account for such cases. We will try to provide an analysis for this type of facts, so that a more complete understanding of the mapping between syntax and phonology is achieved, more specifically regarding the behavior of functional categories.

1.3. Problems for Prosodic Phonology. Lekeitio Basque has a process of Vowel Assimilation (to be discussed in detail in chapter 2), by which the initial vowel of a determiner or inflected auxiliary assimilates in all its features to the preceding vowel, the final vowel of a nominal stem or participial verb. Basque presents a rather free word order, the neutral word order being claimed to be SOV. In Basque, determiners are bound morphemes appearing as affixes on the last word of a nominal phrase, that is, a noun or the adjective that follows it. Case markers and locative markers are spelled out as suffixed to the determiner:1 (6) a. arrebi-a sister-det.sg.

1

‘the girl (abs.sg.)’

b.

gixon-a-k man-det.sg.-erg.

‘the man (erg.sg.)’

c.

mendi txikiñ-éta-n mountain small-det.pl.-inessive

‘in the small mountains’

Only the ergative and dative case are spelled out overtly. The absolutive case is realized as a zero affix.

7

In the periphrastic conjugation, lexical verbs and inflectional categories such as agreement, tense, and mood are realized as two independent syntactic heads. In declarative and interrogative clauses that do not involve negation, the lexical verb always appears followed by the inflected auxiliary, the traditional name given to the bundle of morphemes realizing agreement (spelling out the person features of the subject, direct object and indirect object), tense, mood, and auxiliary type:2 (7) a. erosi e-b-an buy 3erg.--pres.-rt.-past

‘(s)he bought it’

b.

saldu d-o-tza-gu sell 3abs.sg.-rt-3dat.sg.-1erg.pl.

‘we have sold it to him/her’

c.

amaitxu-ko d-ó-zu finish-fut. 3abs.sg.-rt-2erg.sg.

‘you will finish it’

Observe the following examples showing Vowel Assimilation at play in both nominal and verbal contexts: (8) a. neski-a  girl-det.sg. ‘the girl’

2

neskii

b.

itxasu-a-n  sea-det.sg.-ines. ‘in the sea’

itxasuun

c.

asma e-b-an make up 3erg.--pres.-rt.-past ‘(s)he made it up’

d.

erosi e-b-é-s-en buy 3erg.--pres.-rt-erg.pl.-3abs.pl.-past ‘they bought them’



asma aban



erosi ibésen

More detailed explanations on verbal conjugation are provided in chapter 2, section 2.2.2 and section 3.3,

fn. 22. The presentation offered here is a brief sketch illustrating our point of discussion.

8

e.

apurtu é-b-en break 3erg.--pres.-rt.-erg.pl.-past ‘they broke it’



apurtu úben

As we can see in (8d,e), inflected auxiliaries may bear independent underlying accent.3 Usually, the sequence formed by a lexical verb and an inflected auxiliary constitutes one domain of stress assignment, but there are some constructions in which they may form separate stress domains. The participial verb and the auxiliary surface

3

Morphemes (i.e., roots and affixes) can be lexically accented or unaccented in Lekeitio Basque. Accented

words contain either an accented root or one or more accented affixes, with stress surfacing on the penultimate syllable of the word. In the following examples, an asterisk placed at the end of a morpheme indicates that it is accented: (i) /liburu*-a/ book-det.sg.

li.bu.rú.a‘the book’

/mendi-txik*/ men.dí.txik mountain-ablative sg. /ikasi e-b-e*-n/ learn 3erg.--pres.-rt.-erg.pl.-past

‘from the mountain’ ikasi ében

‘they learned it’

Unaccented words receive final stress when they are the last word in the constituent immediately preceding the verb. Otherwise, they surface without stress. Thus, in the following example, of the two unaccented words amen and lagunain the NP, only the latter surfaces with stress on its final syllable: (ii)

NP[amen

laguná] VP[etorri da] mother-gen. friend come aux

‘The mother’s friend has come’

See Hualde, Elordieta and Elordieta (1994) for a description of the metrical system of Lekeitio Basque. Elordieta (1997) and Jun and Elordieta (1997) contain analysesof of the intonational units that are partially formed from the distribution of accents.

9

each with their own stress when the participial verb is focalized.4 In the following examples, focus stress on the last syllable of the verb is indicated with a circunflex accent: (9) a. ikutû eban benganía Péruk touch aux floor Peru ‘Peru did touch the floor’ b.

/

ikutû uban ...

emôngo ebasan amak erregalúak gustora / give-fut. aux mother presents happily ‘The mother would give the presents happily’

c. amaitxû ében biarra ewerdirako / finish aux work midday ‘they did finish the work by midday’

emôngo obasan ...

amaitxû úben

In these sentences, the participial verb is set apart as a domain for stress assignment, and therefore it constitutes a separate phonological word from the inflected auxiliary. Still, as shown above, VA can apply normally. The participial verb also surfaces with its own stress, excluding the auxiliary, in subordinate clauses. No focalization strategy is involved here: (10) a. txakurra ikusí ebenían, aringarínga urte ében dog see aux-comp running leave aux ‘When they saw the dog, they ran away’ b. gauza asko saldú ebález, 4

pozik ewan

The participial verb receives focal stress when the event expressed by the verb and the auxiliary

constitutes the main assertion in the sentence, or when the lexical meaning of the verb is the main assertion. In the latter case, the emphatic verb ein ‘do, make’ is inserted between the participial verb and the auxiliary: (i) urtên ein dau ‘(s)he has LEFT (as opposed to going in, for instance)’ leave do aux

10

thing much sell aux-comp happy was ‘Because (s)he sold many things, (s)he was happy’

In these cases as well, VA may apply: ikusí ibenían and saldú ubález, respectively. This eliminates all possibility of referring to the phonological word as the domain where VA applies, if the phonological word is defined prosodically, that is, in terms of stress. If there is only a maximum of one (primary) stress per prosodic word, the participial verb and the auxiliary verb may constitute separate prosodic words and still form a domain for VA to apply. In principle, this is a puzzle for the Prosodic Hierarchy Theory, at least concerning the definition of prosodic words. If each prosodic word is generally associated with one main stress, the participial verb and the auxiliary in those cases should be two separate prosodic words. Since VA applies to the initial vowel of the auxiliary, the prosodic word cannot be the domain of VA.5 One might insist that leaving stress properties aside, there are other properties that make the auxiliary a weak prosodic word, and that it needs to lean on an adjacent verb to be part of a well-formed prosodic word. But what are these properties? It cannot be the functional category status per se, because there are other functional categories that do not undergo the rule, such as modal particles, subordinating conjunctions, or postpositions (cf. (11)-(13), respectively): (11) a. etorri ete díras? / *etorri ite díras? come dub. aux ‘I wonder whether they have come’ b. atrapa ei dósu / *atrapa ai dósu catch evid. aux ‘I have heard that you have caught it’

5

Stress groups are combined in intonational constituents called Accentual Phrases. As shown in Elordieta

(1997) and Jun and Elordieta (1997), in the cases at hand the verb and the auxiliary belong to two different Accentual Phrases. Thus, this intonational constituent cannot be the domain of VA, either.

11

(12) a. ekarri árren / bring despite b. kompondu ezik fix unless (13)

amaitxu árte / finish until

*ekarri írren /

*kompondu uzik

*amaitxu úrte

There have been analyses in the literature claiming that causative verbs are functional heads. If this were correct, causative verbs would add more evidence against a proposal that attributed participation in VA to functional head status, because they do not undergo the rule. Moreover, they always appear right-adjacent to the verb, moving together to the left periphery of the clause in operator-involving constructions, for example, so it is not as if their syntactic relationship is loose: (14)

paga eraiñ neutzan / pay make aux ‘I made him/her pay’

*paga araiñ

Let us conclude then for the moment that the Prosodic Hierarchy Theory is insufficient as it stands to account for the difference between inflected auxiliaries and other functional heads, or more importantly, for the similarity between auxiliaries and determiners, the two categories that undergo the rule. An alternative analysis is thus necessary, and we will try to provide one. Our analysis is based not on prosodic considerations, but on deeper grammatical properties that determiners and inflected auxiliaries share. Based on the ideas presented in the framework of the Minimalist Program, we notice that theese heads are functional heads realizing formal features that need to be checked in the linguistic derivation by features present in the lexical heads they interact phonologically with. These same grammatical properties are what distinguishes them from other elements which do not undergo the rule: modal particles, subordinating conjunctions, postpositions and causative verbs either do not have sets of

12

formal features to be checked, or do not have the same kind of formal features. These similarities and differences reveal interesting facts about the interaction between functional categories and lexical categories. We will return to this discussion in section 4.2. Another example we will consider is French liaison (chapter 3, section 2). This is a phenomenon where the final latent consonants of words are pronounced in the presence of a following vowel, in certain syntactic contexts. For the purposes of this dissertation we concentrate in the conversational style of speech in French, the so called Style I in the traditional literature on French liaison. The reason for this choice is that, as Rotenberg (1978) and many other authors point out, it is the style of speech where native judgments are most clear and where the contexts of application of liaison are less inclusive, i.e., more restrictive. More formal styles of speech induce wider contexts of liaison with less clear judgments. The relevant aspect of liaison for our purposes is that in conversational style it applies to various degrees of frequency between a functional category and what follows, but not between two lexical words. Thus, compare (15) with (16). The application of liaison is indicated with the symbol ‘∩’, and the nonapplication of liaison with a slash: (15) a. les∩arbres the trees b. ils∩ont parlé they have spoken (16) a. des femmes / intelligents women intelligent ‘intelligent women’ b. les hommes / étaient fatigués. the men were fatigued

13

What is interesting is that functional categories do not have their final consonants pronounced with the same degree of frequency in conversational speech. Determiners, demonstratives, numerals, quantifiers and clitics trigger liaison obligatorily: (17) a. des ∩ ennemis ind.det enemies ‘(some) enemies’ b. cet∩effort this effort c. mon∩oeil my eye d. vingt-trois∩oignons twenty-three onions e.

Il vous∩a donné la réponse. he you has given the answer ‘He has given you the answer’

f.

Ils∩ont entrés dans la salle they have entered in the room

They contrast with prepositions, degree modifiers, auxiliaries and copulas. With these categories, liaison does not apply obligatorily. It does occur frequently with monosyllabic forms of these elements: (18) a. dans∩une salle publique in a room public ‘in a public room’ b. fort∩intéressant. very interesting c. Je suis∩arrivée trop tôt. I have arrived too soon d. Nous sommes∩impatients de vous voir à Paris. we are impatient of you see in Paris ‘We are anxious to see you in Paris’

14

In turn, the contexts in (18) contrast with the ones in (19), exemplifying cases of polysyllabic prepositions, degree modifiers, auxiliaries and copulas. In these contexts, liaison is found rarely or is harder to make in conversational style: (19) a. pendant /? une semaine during one week b. assez /? intimes rather intimate c.

Les soviétiques auraient /? annoncé la fin des négociations. the soviets had announced the end of the negotiations

These distinctions are real, and have been checked with native intuitions of several speakers. These contrasts have also been found independently in collected sets of data from recorded conversations with native speakers of French (the Órleans corpus, as cited by De Jong 1988, 1990). As such, they should be accounted for in any theory of the syntax-phonology interface. As we demonstrate in more detail in chapter 3, a pure prosodic analysis of these facts cannot provide the answer for these differences, since determiners, demonstratives, numerals, quantifiers, prepositions, degree modifiers, auxiliaries and copulas are all functional categories. They would thus be expected to pattern similarly, unless something else is said. Within the End-Based model, De Jong (1990) tried to account for these differences among functional categories, recurring to the notion of head. The basis of the arguments is that some functional categories are not heads but modifiers or auxiliaries, whereas other functional categories are syntactic heads. Then, an additional endparameter setting is included, one that looks at heads and inserts boundaries to the left or right of syntactic heads. The stipulation is that determiners, demonstratives, numerals, quantifiers and clitics are not heads, but prepositions, auxiliaries and copulas are. Thus, fixing a choice of insertion of boundaries to the right of the latter categories would

15

separate them from following material prosodically, accounting for the lack of obligatory liaison. Since determiners, demonstratives, numerals and quantifiers are not heads, they are never separated from following material and therefore fall under the domain of obligatory liaison (cf. chapter 3, section 2, for more details). This move appears to be incompatible with current views in syntactic theory, not only in the minimalist framework but also in the earlier framework of Principles and Parameters, on which these theories base their syntactic assumptions. All the functional categories mentioned above are syntactic heads of their own projections (Det0, P0, Adv0, Aux0), and thus no syntactic differences exist among them in terms of their status as heads.6 The conclusion is that an alternative analysis is needed to account for these facts.

Like in the case of Basque, we base our analysis on the observation of the similar grammatical properties displayed by the categories triggering obligatory liaison, and at the same time, on the differences that set them apart from the categories that do not participate in liaison to the same degree.

2. The proposal. 2.1. The theoretical framework. The theoretical framework that will guide our model is the one known as the Minimalist Program, formulated and developed by Noam Chomsky (1993, 1994, 1995). I will begin by briefly sketching the basic ideas in this model, expressed in Chapter 4 of

6

In the minimalist framework, all heads that project (e.g., the head T of TP) are maximal zero-level

projections or X0max, that is, heads and maximal projections at the same time (cf. Chomsky 1995:245). However, all the functional heads mentioned above in our discussion of liaison are heads of their own projections.

16

Chomsky (1995). Then I will proceed to outline the approach I will pursue in this dissertation as a model of the mapping between syntactic and phonological structures. Building on the theory of Principles and Parameters, and especially on principles of economy and derivation and representation, the minimalist framework views Universal Grammar as including a generative device, the computational system of human language (CHL), that constructs linguistic expressions as pairs (π, λ). π is a PF representation and λ is an LF representation, each consisting of “legitimate objects” that can receive an interpretation at the articulatory-perceptual (A-P) and the conceptual-intentional (C-I) interfaces, respectively, as instructions to the performance systems. Under minimalist assumptions, PF and LF are the only levels of representation in grammar, with their own wellformedness conditions that linguistic expressions must meet. If each linguistic expression consisting of the pair of PF and LF representations (π, λ) meets the interface conditions, then the generated representation is said to satisfy the condition of Full Interpretation (FI). The grammar determines a set of derivations, and a derivation converges at an interface level if it yields a representation satisfying FI at that level, and converges if it converges at both interface levels, PF and LF; otherwise, it crashes. Additionally, a linguistic expression must be optimal, satisfying various natural economy conditions: locality of movement, shortest move, etc. Thus, the grammar generates three sets of derivations: the set D of derivations, a subset DC of convergent derivations of D, and a subset DA of admissible derivations of D. FI determines DC, and the economy conditions select DA. The set of admissible derivations DA is a subset of the set of convergent derivations DC that meet the economy conditions. The grammar starts by selecting an array of lexical items and mapping them to pairs of PF and LF representations (π, λ). The operation Merge takes a pair of syntactic objects α and β, and replaces them by a new combined syntactic object K. K has the following composition:

17

(20)

K = {γ, { α, β }}

α, β are the objects or constituents of K and γ is the label of K, which indicates which one of the constituents α and β K is a projection of. The label is actually the head of the projection. The usual structural relationships of X-bar theory are maintained in this system. If α is a minimal category and γ is its projection, then β is the complement of α. If α is a branching constituent, β is the specifier of γ. Adjunction of β to α would create a twosegment category L, with label < γ, γ >: (21)

L = {< γ, γ > α, β }

In a structure such as (21), α is not a category, but a segment of the category L. L has two segments, then: α and itself. To give a concrete example, consider the sentence in (22): (22)

Peter saw the man

The initial array for this phrase is: (23)

{man, the, saw, Peter, T}

The derivation starts from the bottom. The noun man merges with the determiner the, forming the structure in (24): (24)

{the the, man} A tree diagram representation of (24) is (25a) or (25b), using categorial labels:

(25) a.

D 2

b.

DP 2

18

the

man

the

man

Then, the DP so formed merges with a verb, saw, forming a VP: (27)

VP 3 V DP ! 2 saw the man

This VP, in turn, merges with a light verb, in a similar fashion to a Larsonian shell, deemed necessary for all transitive clauses (cf. Hale and Keyser 1993): (28)

vP 3 v VP 3 V DP ! 2 saw the man

Now, the subject merges into the specifier position of vP: (29)

vP 3 DP v’ ! 3 Peter v VP 3 V DP ! 2 saw the man

The vP is merged with T, forming the structure in (30): (30)

TP 3

19

T

vP 3 DP v’ ! 3 Peter v VP 3 V DP ! 2 saw the man

This is an illustration of how Merge works. Within each cycle in the syntactic derivation, however, after Merge has created a syntactic object, and before another step of the Merge procedure applies, there is another operation that combines syntactic objects in the derivation. It is the operation Attract/Move, a procedure that typically carries out the operation of checking formal features. Let us now move to discuss the notion of feature checking. Lexical items are composed of three types of lexical features: (a) phonological features, i.e., the phonological matrix of a lexical item, which receive an interpretation only at the A-P interface (i.e., the phonological interface); (b) semantic features, which receive an interpretation only at the C-I interface (i.e., semantic interface); and (c) formal features, which include categorial features such as [nominal], [verbal], etc., and noncategorial features such as P-features (e.g., [3 person], [femenine], [-plural]), Case features (e.g., [accusative case]), and tense features (e.g., [past]). Some of them are intrinsic, i.e., nonderivable from the derivation, and others optional, added as the numeration is formed. Phonological features are interpretable only at PF; semantic features are interpretable only at LF. Formal features are accessible in the course of the computation, by definition. Furthermore, no direct connection exists between the PF and LF interface levels, except as defined by the derivation.

20

A noun like man, for instance, contains the phonological features composing its phonological matrix, the semantic feature [human], and the formal features [nominal], [3 person]. Its optional features include the noncategorial features of Case and number. A determiner like the has phonological features (i.e., [ð@]), semantic features such as [antecedently presented], or [referential], and formal features such as the categorial feature D, which is intrinsic. It also has P-features and Case features, which are not intrinsic properties (except perhaps for [3 person]). A verb like saw contains phonological features, semantic features, and the formal features [verbal] and the Case feature [assign accusative]. Its P-features and tense are optional, that is, they are not intrinsic to the verb itself. One of the central claims in the minimalist framework is that formal features have to be checked in the course of the derivation, both of functional categories and lexical categories. T(ense) has an [assign nominative] Case feature that needs to be checked with those of a subject DP (or NP). This checking is done by raising the [nominative] Case feature and of a DP to the specifier position of T. T also has a D feature that has to be checked by raising the categorial feature of D to Spec,TP. This feature is responsible for the Extended Projection Principle effects, that is, the requirement that every sentence must have a subject. In turn, the subject gets its nominative Case assigned (checked) by raising its feature [nominative] to Spec,T:7

7

I would like to insert a short parenthesis in the discussion at this point. It has been objected against the

minimalist program that no justification is provided for the operation of feature checking. It is likely that the independent justification for it is just the mere observation, dating from the first descriptions of human language, that there are certain grammatical categories that are related: a sentence needs a subject, which is assigned nominative Case; there must be some category that licenses this Case. Subjects express P-features that are also reflected on the verb in many languages. Verbs, in turn, have tense, aspect, mood, and agreement morphemes realized on them, relating them both to inflectional categories T, Asp and to the P-

21

(31)

TP 3 [D]i T’ [nom.]i 3 : T0 vP !

[D] [ nom.]

to

! DP ! [D] ! [nom.] z-------m

v’ 2 v VP !

V0

The features in bold font indicate the moved features and the features attracting them, that is, the features entering in a checking relationship. Those in smaller font and italics represent the copies left behind of the features that have been moved. These copies are deleted by an independent operation of deletion at PF. By the operation of feature checking, the values for the feature specifications in D are compared with those of the same features in T, and if they match, the derivation is convergent and further operations may apply. Oherwise, the derivation is canceled. A valid metaphor to refer to this relationship between D and T is one of symbiosis. Each gets part of its formal features checked, as part of the necessity for convergence. T gets its D feature checked. D, in turn, gets its Case feature checked. But as a result, the case assigning features of T and the categorial [D] feature of D are also checked.

Furthermore, it is assumed that an

features of the subject. The reality of grammar is thus that categories are related to one another, by sharing features. Furthermore, only derivations in which categories share features with the same values are grammatical derivations (i.e., convergent derivations). The operation of feature checking would thus be only a formal procedure to make sure that features match, and that only derivations in which all features have been checked and match are convergent. Feature checking is then a procedure that licenses convergent derivations.

22

element X can target Y only if a feature of Y is checked by the operation (cf. Chomsky 1995:257). That is, movement is triggered only by feature checking purposes. Since features are considered to form abstract morphemes, movement is thus always triggered by morphological requirements. Under minimalist assumptions, determined ultimately by economy principles, the operation Move involves only formal features, not the whole category containing those features. This defines Move as Move F, F being a formal feature (cf. Chomsky 1995:262): (32)

F carries along just enough material for convergence

It is assumed that Move F automatically carries along the set of formal features of a category or lexical item. Thus, Move F is really Move FF[F], where FF[F] is the set of formal features of the lexical item in question. Movement may be covert or overt. If it is covert, it involves simply movement of features (i.e., of FF[F]). Overt movement is triggered by a strong feature. It involves raising the whole category, as “extra baggage”, in a kind of “generalized pied-piping” movement, in Chomsky’s own words (cf. Chomsky 1995:262). It is properties of the PF component that require such pied-piping for convergence. The reasons are left unexplained, and the differences between overt and covert movement are stated as a description of the facts, associated to a difference in nature between the PF and LF interface levels. Categorial features in the functional heads may be strong, that is, they require overt raising of the whole category, not only features. Thus, if the D feature of T is strong, it is going to attract the whole subject DP, driving to its phonetic realization in Spec,TP. We can illustrate this using the example and the structure in (30):

23

(33)

TP 3 DP T’ ! 3 Peter T vP 3 {DP} v’ 3 v VP 3 V DP ! 2 saw the man

The {DP} in Spec,vP represents a copy of the moved category, which will later be deleted at PF by the operation of copy deletion. If movement were covert, that is, if the D feature of T were not strong, only the set of formal features of D mould have moved and the lexical item Peter would have stayed in Spec,vP. Continuing with our presentation of feature specifications, T also has V-features, that attract the categorial features of a verb, i.e., [V]. Again, this movement may be overt or covert. If it is covert, the set of formal features of the verb will have moved to T, the lexical item being phonetically realized in v, which is taken to be a verbal affix requiring the overt incorporation of V. At the same time, the verb has tense features, which are not intrinsic to the verb but assigned when the verb enters in the numeration. These T features of V have to be checked as well, and this is done by movement to T. This movement is done together with the rest of formal features contained in V, as an instantiation of Move FF(F). In the following diagram we illustrate this relationship with the feature [+ past]:

24

(34)

TP 3 T’ rp T0 vP [V] [V]j tp [+past] [+ past]j DP v’ : 2 ! v VP ! 2 ! ! V v Vi ! [V]j ! [+past]j ! ! z------m

As mentioned before, this relationship between T and V is symbiotic. Each gets part of its features checked as a consequence of movement. T gets its [V] feature checked, and V gets its tense feature checked. Additionally, the verb is specified for the P-features of the subject, not as intrinsic features but as optional features, inherited once the verb enters the numeration. These features are checked once the formal features of the subject (including its P-features) are in Spec,T and the formal features of the verb raise to T, providing a case checking configuration in the projection of T:

25

(35)

TP 3 FF(Subj) T’ [P-feat.] rp T0 vP FF (V) [P-feat.]i

ti

DP v’ : 2 ! v VP ! 2 ! ! V v Vi ! FF(F) ! ( [P -feat.]j) z---------m

Although we did not mention it in our discussion of Merge, the projection TP merges in a higher constituent, CP. The head of this projection contains operator features, indicating force or type of clause (interrogative or imperative operators, for instance). A feature such as Q attracts a wh-word to Spec,CP. It must also contain some features that attract the raising of T to C, after raising of V to T, a fact observed in many languages (i.e., V2 languages, languages with V-to-I-to-C movement in operator-involving constructions). C has a categorial feature T that needs to be checked, much like T has a feature V that has to be checked. Also, we assume that T is specified for a feature in C0 (in interrogative sentences, for instance, T would be specified for the feature Q), and this feature needs to be cheked. These symbiotic feature checking operations are carried out by the raising of the feature T (with V) to C. The following diagram is an schematic illustration of the operation of feature checking in an interrogative sentence:

26

(36)

CP 3 C TP [T]i [T] 2 [Q]i [Q] T vP : [T]i ! [Q] i z---m

[Accusative Case] and P-features are among the formal features of an object of a transitive clause, and the verb has the feature [assign accusative Case] as well as Pfeatures of the object. Once the verb raises to v, checking of these features is done by raising the formal features of the object to Spec,v. Since the subject is occupying a specifier position in vP, a multiple specifier adjunction structure is posited (cf. the slightly adapted version of the diagram in Chomsky 1995:369): (37)

vP 3 FF(Obj) v’ [P-feat.]i 3 : Subj v’ ! 3 ! Vb VP ! [P-feat.] 3 ! tV FF(Obj) z--------------m

Vb stands for the complex verbal formed after raising the auxiliary to v, and tV represents the trace/copy left by the verb. The Spec is the position that the formal features of the object will occupy in order to check the accusative Case feature with Vb. The subject moves up to T, higher above. There are other configurations in which the object can check its accusative Case. If the object does not raise overtly to [Spec,Vb], the formal features of the object rsise to

27

T, where the formal features of the verb have also raised, forming the structure in (38) (cf. Chomsky 1995:371): (38)

T 3 FF(Obj) T 3 FF(Subj) T 3 Vb T

We will not discuss the details of how and why one structure is relevant over the other. What matters for our presentation is that the object and the verb are in a checking relationship, regarding the [accusative Case] and P-features. Feature checking takes place inside the DP as well. The same way T attracts the categorial feature V, the determiner attracts the categorial feature N, for convergence (cf. Chomsky 1995:337, 364). Expletives, which are Ds, trigger this movement from their associates (cf. Chomsky 1995: 364). We can assume that the determiner has P-features, present in the noun, and that movement of the set of formal features of N is required to check them as well. In turn, Longobardi (1994) suggests that referentiality features are present in the noun and the deteminer, and they have to be checked. Case is a property of D, not of N. The diagram in (39) represents this movement: (39) DP 3 D NP [N] [N]i ! : N ! [N] i z----m

28

Again, if the N-feature in D is strong, it will require pied-piping of the whole category, and the noun will be adjoined to D. Otherwise, the noun will be spelled out in its original position. Another head that assigns Case is the preposition.8 P has to check this Case assigning feature against the Case of the DP which is its complement. This Case feature is more concretely in the head D (and in its absence, in the head N). The operation Attract/Move F therefore raises the formal features of D to Spec,P, for formal checking of Case features. In order to draw a parallelism with other case assigning heads such as T and V, we could also hypothesize that P attracts a categorial feature D, which are checked in its Spec position, together with the other formal features of D: (40)

PP 3 FF(D)i P’ : 3 ! P DP ! 3 ! D NP ! FF(D)i z-----m

2.2. Mapping of feature chains onto morphosyntactic constituency.

8

Which case P assigns (accusative, oblique, or other) is language-specific. In Germanic and Romance P

assigns accusative case.

29

Our analysis is built on work in progress by María Luisa Zubizarreta and JeanRoger Vergnaud, who claim that the relations of feature checking express primitive relationships of grammar which they call primitive chains, such as the following:9 (41)

C-T

(42)

T-V

(43)

T-D

(44)

V - D (Subj)

(45)

V - D (Obj)

(46)

D-N

(47)

P-D These are the pairs of heads that are related by operations of feature checking,

such that a feature in one head attracts and is checked by a set of formal features of the other head. C-T (cf. (41)) expresses the fact that the head C attracts the feature T, as evidenced in interrogatives in English, for example, that attracts the raising of a verb in T (or in its absence, a dummy verb do). The relation T-V in (42) is an expression of the raising of the categorial feature [V] as required by the feature V in T. T-D (cf. (43)) in turn expresses the Extended Projection Principle, and the checking of nominative Case feature of the DP subject and of T. (44) (i.e., V-D(Subj)) represents the checking relation of subject P-features of V that takes place in T, when the formal features of the subject DP move to Spec,T and the formal features of the verb raise to T. (45) (i.e., V-D(Obj)) represents the relations of checking accusative feature and P-features between a verb and its object. The chain D-N in (46) stands for the checking of the categorial feature N and

9

These ideas have been presented in Morphology classes at USC taught by Jean-Roger Vergnaud (1995-),

to whom I am indebted for many hours of discussion on the matter.

30

of P-features of the noun by raising the formal features of N to D. Finally, the feature chain P-D in (47) expresses the checking of Case features present in D and assigned by P, and of the categorial feature [D] attracted by P. Zubizarreta and Vergnaud claim that these pairs are primitive entities of grammar, which express the fact that the heads contains formal features that are related to formal features on the other head, and that the two heads must form one unit by LF. Although this relationship is expressed in minimalist terms as movement operations of feature checking, Zubizarreta and Vergnaud argue that it is not the operation of feature checking itself that expresses a primitive relation in grammar, but the chains themselves. The term we will adopt in this dissertation to refer to these pairs is feature chains, standing for the fact that the heads in each pair are related by sets of formal features. In Zubizarreta and Vergnaud’s theory, these chains are independent of phrase structure, although coexistent with it. They are present throughout the linguistic derivation, up to the moment it is sent to the PF and LF interface levels. A schematic phrase structure such as the one in (48) would have a parallel representation of feature chains, as primitive syntactic objects: (48)

CP ! 3 C TP [T] ti T vP [T] ti [D] DP(Subj.) v’ [V] 2 ti D N v VP [D] [P] [V] 3 V DP(Obj.) [V] 2 [T] D N [P] [D] [P] !

! ! ! ! ! ! ! ! ! ! ! !

C-T T-V T-D V-D(Subj.) V-D(Obj.) D-N

31

The heads in the phrase structure are related by the features that each of them possesses and that attract and are attracted in the manner laid out in section 2.1. The difference with respect to minimalism is that there are no operations of feature checking; the relationships among heads and their features are present directly in syntactic structure, by the existence of feature chains. These chains are objects at LF and PF, where they must receive an interpretation. Turning our attention to PF, the relevant interface level for our purposes, the idea in the framework of Zubizarreta and Vergnaud is that the chains presented above are primitive entities of grammar, and that they are units for morphosyntactic mapping. That is, they are morphosyntactic entities which are mapped to the morphophonological component of grammar, after the Spell-Out procedure has sent the derivation to the interpretive level of Phonetic Form (PF). We assume the existence of a morphological component in grammar, part of PF but with its own wellformedness principles and operations, where the syntactic derivation is sent before reaching the interpretive level of PF. To borrow a name already introduced in the literature, we call it Morphological Structure (MS) (cf. Halle 1990, Bonet 1991, Noyer 1992, Marantz 1992, Halle and Marantz 1993). MS is an interface level between syntax and phonology. In fact, is a syntactic representation that nevertheless serves as part of the phonology, where ‘phonology’ is understood as the interpretive component that realizes linguistic derivations phonologically. The organization of the grammar adopted is the following: (49)

(Morphological structure) (Phonetic Form)

DS (D-structure) ! SS (S-structure) 3 MS LF (Logical Form) ! PF

32

In this theory, affixes and stems are entries in the Vocabulary of a language, composed of two sets of features: phonological features, on the one hand, and morphosyntactic features, on the other. The terminal nodes in syntax are bundles of morphosyntactic features, hierarchically organized in the case of complex X0s, and it is only by an operation of Vocabulary Insertion that these bundles of morphosyntactic features receive phonological content. Vocabulary Insertion takes place at MS, and in order for Vocabulary Insertion to be successful, the morphosyntactic features of a vocabulary item must match or be non-distinct from the morphosyntactic features present in the terminal nodes of S-structure. Thus, the vocabulary entries inserted by this operation connect morphosyntactic feature bundles of syntactic structure with phonological feature complexes. This is where the actual spell-out of morphemes takes place. The phonologically realized string of morphemes and words would then pass to the actual level of PF, where the phonological processes of the language would modify those parts of the string which meet their structural description.

2.3. Mapping of feature chains onto MS objects. The feature chains we are considering are mapped to the MS level, after all feature checking has been completed, and at this level the chains may be grouped in morphological domains along several parameters of realization, which we will present in the following sections. There are two possibilities: a) that the heads in the chain form one single morphosyntactic word, or b) that they form two separate morphosyntactic words (from now on, we will refer to these units as ms-words). The main empirical basis that distinguishes these two possibilities is the degree of phonological cohesion between the heads. A close degree of phonological interaction between two heads in a feature chain indicates that the heads form one single morphosyntactic word, whereas the absence of phonological interaction between two heads in a feature chain indicates that the two heads form two independent morphosyntactic words. In other words, the morphological

33

or morphosyntactic constituents mapped from syntax to MS are interpreted as phonological objects, units where phonological processes apply. Using the example of a chain T-V, we represent in (50) and (51) respectively these two possibilities in the mapping of feature chains as MS objects: (50)

Syntactic derivation

Morphological Structure Phonetic Form

(51)

Syntactic derivation

Morphological Structure Phonetic Form

T-V Spell-Out ! [T V]ms-word [T V]p

T-V Spell-Out ! [T]ms-word [V]ms-word [T]p [V]p

Before providing practical examples of each possibility, it must be pointed out that within the first type of mapping (cf. (50)) there are also two possible modes of realization at the level of MS: the ms-word may adopt the form of what is traditionally understood as a morphological word, that is, a complex form composed of a stem and affixes, or each member in the feature chain is realized as an independent morphological word, in the sense that neither head is spelled out as an affix. In both cases, the morphosyntactic word is a close morphophonological object that is interpreted as a phonological unit.

34

Practical examples of feature chains realized as morphological words composed of stems plus affixes would be any words containing a nominal or verbal stem to which inflectional morphemes such as tense or agreement markers are added. Examples of this type abound in languages of the world. In English, for instance, tense morphemes and the remaining overt realization of P-features for third person singular in the present tense, -s are realized as affixes on the verb: (52) a. b.

We talk-ed She arrive-s

This possibility is independent of overt or covert feature checking. It is widely accepted that English does not have overt V-to-T movement, but Spanish does (cf. (53)). In both instances, the tense and P-feature features are realized as affixes on the verb: (53) a.

b.

Pedro camin-ó Pedro walk-past/3sg ‘Pedro walked’ Nosotros aprender-ía-mos we learn-cond.-1pl ‘We would learn’

The exact mechanisms by which the tense and P-feature morphemes appear morphologically realized as affixes on the verb merit some discussion. There is the possibility that it is a result of overt movement, triggered by a strong feature in T, such as the categorial [V] feature. In this case, more material than just the set of formal features of a lexical item are moved to the head above it, for reasons of PF convergence. The whole category (in this case a verb) moves to the position occupied by a T (and C subsequently, if C has a strong [T] feature). The consequence will be that the heads (i.e., the whole categories) are realized under one X0, in an adjoined structure:

35

(54) a.

T 2 V T

b.

C 2 T C 2 V T

In the discussions in the Minimalist framework, a morphological feature [affix] is mentioned as a lexical feature that forces movement. On page 269, for instance, it is stated that the head T has a feature “takes verbal affix”, and on page 321 the same is said about the light verb v. In these cases, it is the target of movement that has the affixal feature prompting movement.10 Then, on page 285, there is a brief mention of a possibility of analyzing incorporation as a result of a feature [affix] on the head that moves, so in this case it is the source of movement that has the feature [affix]. This feature is supposed to be a morphological feature, but it is never stated explictly if it is part of the set of formal features of a linguistic item, or whether it is a feature that always forces overt movement. If it does, then it is not clear how different the feature [affix] is from a strong feature, especially in cases where the affixal feature is part of the target of movement. That is, a strong feature [V] of T, for instance, would force the overt raising of the category V to the head with the strong feature, the same as a feature [take verbal affix]. An affixal feature of the source of movement (as Chomsky points out for noun incorporation) would produce the same effects: a category is adjoined overtly to the position occupied by another head, giving rise to structures of the form in (54).

10

The expression chosen by Chomsky to refer to this relationship (T or v having the feature “takes verbal

affix”) may be a little misleading, since the head that is realized as an affix (i.e., T) requires the raising of a category that is not an affix, a verb. What should be born in mind is the intended result, that is, that a verb incorporates onto T or v, forming complex heads with them.

36

For our purposes, it is sufficient that some heads may have a morphological feature [affix] that forces overt movement. This is an arbitrary feature that some heads possess and others do not. Simply put, some heads are affixes whereas others are not. It is important to stress the fact that it is a morphological feature, and not a syntactic one, because, as Bobaljik (1994, 1995) argues, affixation may be done after the syntactic derivation has been complete, that is, after the derivation has been sent to the PF component by the Spell-Out procedure, if a head which is an affix has not received the overt incorporation of another head. The affixation is done in Morphological Structure by morphological merger, under adjunction (i.e., in order for an affix and a stem to be combined, they must be adjacent). Adjacency is satisfied automatically when there has been syntactic incorporation of the verb to the inflectional head. When there has been no verb raising, however, a possible situation arises in which some structural material may intervene between the affix and the verb stem. Subjects, negation and other structural material intervening between the affix and the stem block merger. Adjacency, however, is not a purely morphophonological notion, since adverbs intervening between the affixes and stems do not block merger. This is schematized in (55): (55) a.

...Affix [YP NP[overt] [ Y...

Affix and Y (stem) not adjacent

b. c.

...Affix [YP [ Y... ...Affix [YP trace

[ Y...

Affix and Y (stem) adjacent Affix and Y (stem) adjacent

d.

...Affix [YP adverb [ Y..

Affix and Y (stem) adjacent

The lack of adjacency between affix and stem renders ungrammaticality. Bobaljik presents different phenomena from Germanic languages, Irish and Bambara as evidence for his approach. For English, Bobaljik argues that do-support is triggered because the verb cannot merge with the affixal Infl. He follows standard assumptions that the main verb does not raise to I0 in overt syntax, but argues that the features in Infl are affixal and

37

need to merge with the verb in the morphophonological component, after the derivation has been sent to PF. He does not adopt Chomsky’s (1993) proposal that the verb needs to check the features in I0 by raising to this head at LF and that the dummy verb do in inserted in the presence of an intervening head (such as negation) between I0 and V0 that blocks the raising of the verb. Chomsky’s approach is schematically represented in (56): (56)

when the phenomenon known as Object Shift in Germanic, an object pronoun may not intervene between a tense morpheme and a verb stem which has stayed in situ in syntax. If it does, the sentence is ungrammatical. No material may intervene between the inflectional head and the stem (cf. Bobaljik 1995:chapters 2, 5, 6). We are not going to present here the whole evidence provided by Bobaljik. If this author is correct, it means that the feature [Affix of x] can be weak in the minimalist sense, since it may not trigger overt syntactic raising of the stem (or more accurately, of the category that will be realized as a stem at Morphological Structure). However, as the label of the feature itself indicates understood in the traditional sense, it has an additional morphological property, that must be satisfied at MS: that it must be attached to a stem. There are conditions for this attachment, such as adjacency, as Bobaljik may be correct in assuming. For reasons of simplicity, we argue that the checking relation determined by the feature [Affix] is subsumed in each feature chain in (39)-(45). That is, on top of its set of formal, phonological and semantic features, each head could potentially possess the feature [Affix]. A distinction must be established, however, between the relationships of feature checking expressed in the chains in (39)-(45) and those demanded by the feature [Affix]. The relationships between features expressed in the feature chains in (39)-(45) are present in Universal Grammar, whereas the feature [Affix] is only an optional feature

38

of the heads in those feature chains. We will refer to the affixal feature explicitly when it is relevant, as in the discussion on Basque in chapter 2. As far as the relative order among morphemes is concerned, we assume that this ordering is not manifested as a property of syntax proper, since linearity is not a syntactic property but a morphophonological property. The number of counterexamples found to the Mirror Principle of Baker (1985, 1988) also point in this direction. We will propose then that relative ordering is part the set of phonological features of each category or lexical item (i.e., a verb, a noun, a determiner, C, T). For the morphemes in (48), for instance, we could posit the insertion rules in (51). The context in the righthand column indicates the ordering of morphemes with respect to one another: (51) a. [past/3sg] b. [cond] c. [1pl]



/-ó/

/ verb stem ___ ]

 

/-ía-/ /-mos/

/ verb stem ___ ] / verb stem-tense___ ]

So the first possibility for feature chain morphological realization is that one or more of the members in the chain is realized as an affix and at least one member is realized as a stem.11 This realization constitutes a morphological word which is typically mapped to the phonological component as a phonological word.12 This word is then a

11

In the chain C-T-V there could be two heads that are realized as affixes, C and T, the verb being the stem

they attach to. This is a pattern found in many languages. 12

Other possibilities exist, in which affixes constitute independent phonological domains from stems, as

evidenced by the so-called disharmonic prefixes in Hungarian and Italian (cf. Nespor and Vogel 1986) or disyllabic suffixes in YidiJ (cf. Nespor and Vogel op. cit.). We disregard these cases at the present time, because they will not be the object of our discussion in this dissertation. Further careful research awaits in order to incorporate these cases into our model.

39

potential domain for phonological processes to apply. These domains constitute the object of study of Lexical Phonology and Morphology (Pesetsky 1979, Kiparsky 1982, 1983, 1985, Mohanan 1982, 1986, Kaisse and Shaw 1985, among others), although it should be clear from this discussion that we do not share with this theory the view that morphological and phonological words are created in the lexicon. This is the Strict Lexicalist Hypothesis, first proposed in Jackendoff (1972) and Halle (1973), and followed by Lieber (1980), Lapointe (1980, 1988), Selkirk (1982) and Di Sciullo and Williams (1987). In the approach advocated here, morphological words are formed after syntax, at the level of Morphological Structure. Inflectional morphology is realized or spelled out after syntax, although the specifications that features have regarding their realization as affixes or independent words are a lexical property of features and thus should be contained in the lexicon, if such a level exists. What is not so clear is whether derivational morphology should be treated in the same way, that is, as taking place at Morphological Structure. Proponents of a split view of morphology locate derivational morphology in the lexicon and argue that rules of syntax distribute morphosyntactic features onto words, with actual rules of morphophonology taking place after syntax, in PF. This is the position known as the Weak Lexicalist Hypothesis, or Split Morphology Hypothesis, first suggested in Chomsky (1970), and defended most notably by Anderson (1982, 1986, 1992). A more radical approach is proposed by Fabb (1984), Sproat (1985), Baker (1988) and Lieber (1992), among others, suggesting that some derivational morphology is also done in the syntax.13 I do not have anything to say about the issue of where exactly derivational morphology operates. It is not a matter of central importance in this dissertation, and I leave it open for discussion. 13

The Weak Lexicalist Hypothesis thus rejects the idea of the existence of a morphological component per

se, as a separate module of grammar. We do not agree with this view.

40

The phonological words formed by feature chains containing affixation may also be subject to rules of postlexical phonology, in which case they will be the subject of study of Prosodic Phonology. We refer the reader to chapter 2, section 5, for a review of the major postulates of this theory. To sum, a feature chain may be realized at MS as a single morphological word formed by a stem and one or more affixes. This realization is determined by idiosyncratic properties of the heads containing the features in the chain (i.e., one or more of the heads are lexically specified as affixes). In turn, this word is interpreted as a domain in the PF interpretive component of grammar, where constituents formed at MS are mapped. This domain is then subject to any rules that are specified to apply within a phonological domain of that sort, i.e., a phonological word. We can represent this mapping between syntactic, morphological and phonological structure as follows, with the example of the feature chain T-V (the label m-word in the constituent at MS stands for ‘morphological word’, to borrow a common term denoting a combination of a stem plus affixes): (52)

Syntactic derivation

Morphological Structure

T-V

Spell-Out t [stemV - affixT]m-word

! Phonological/ ? /Phonetic Form [stem-affix]p

Studying morphological words and phonological words of this sort will not be the main objective in this dissertation, as enough attention and effort has been dedicated to it. We represent it as one of the possible mappings from syntactic structure to morphological and phonological structure. A more interesting situation is one in which a head which does not enter into feature checking relationships of the kind expressed in (39)-(45) is an

41

affix and intervenes linearly at MS between two heads which do enter in a feature chain relationship. Such a situation is exemplified in Lekeitio Basque (cf. section 1.3 of this chapter). Modal particles and causative verbs possess the feature [Affix], triggering the overt raising of the closest possible head: a participial verb or negation. The functional head T, located above modal particles and causative verbs and realized as an inflected auxiliary, is also an affix. Thus, it also triggers the overt incorporation of a participial verb or negation, the closest heads that can provide morphological support to it. Focusing for the moment on the relationships with a participial verb, the possible linear sequences at MS involving verbs, modal particles and inflected auxiliaries are: (53) a.

Verb-T

b.

Verb-modal-T

c.

Verb-causative-T

As we mentioned in section 1.3, only the sequence formed by a participial verb and the head realizing verbal inflection (i.e., T) is a context for application of VA (i.e., (53a)). This process dos not apply between a participial verb and a modal particle or between a participial verb and a causative verb (cf. (53b,c)), despite modal particles and causative verbs being heads that trigger incorporation of the verb. The solution to the puzzle is that even though lexical verbs, modal particles and tense heads may end up in one adjoined X0 in syntax, the verb and the head T are in a feature chain relationship. The tense head however contains the feature [Affix of V], apart from the strong categorial feature [V]. The strong categorial feature attracts the category Auxiliary located below it. The feature [Affix], however, is not satisfied by the raising of the auxiliary, because the auxiliary itself is a weak word, not an independent morphological word. The feature [Affix of V] is strong, and has to be checked overtly. This means that the participial verb, which is an independent word, has to raise to T, checking the affixal feature. This [Affix]

42

feature determines the realization of the feature chain T-V as a morphological word, i.e., [V-T]m-word. This m-word is mapped as a p-domain, a phonological word, and thus a potential domain where phonological processes may apply. Vowel Assimilation is one such process. The determiner in Basque is also an affix (i.e., D has the feature [Affix]), and forms an m-word with the last word in the NP that raises to Spec,D. This m-word is then mapped as a phonological word, where VA applies. This analysis explains why determiners and auxiliaries pattern the same way phonologically with respect to VA. Given the difficulties that the distribution of VA presents for theories of Prosodic Phonology, as we will see in chapter 2, this is a welcome result. It also provides empirical evidence for the reality of the feature [Affix] as a formal feature, not as a morphological feature visible only at MS. If it were purely a morphological feature, it would not trigger the overt syntactic effects it does.

and have been discussed extensively in the linguistics literature. We will not add anything thus refrain from adding anything Any word formed by a nominal stem and affixes determiner These two options is evidence for the possibility is that but the head members display a high degree of phonological cohesion. Such phonological interaction is not found between independent words which do not realize the heads forming the feature chains represented in (39)-(45). We suggest that this pattern stems from the fact that the members of a feature chain that are realized as independent words but have a close phonological interaction form a morphosyntactic constituent or domain at the level of Morphological Structure, and this domain is mapped to PF as a phonological domain (cf. section 4.2.3). Syntactic derivation

Morphological

T-V

Spell-Out t [stemV - affixT]m-word

43

Structure ! Phonological/ ? /Phonetic Form [stem-affix]p (56)

T-V

Syntactic derivation

Morphological Structure Phonological/ /Phonetic Form

[T]m-word

Spell-Out t [V]m-word

! ? [T]p [V]p

the feature chains are realized as a morphological word composed of a stem and one or more affixes. The morphological word is a potential domain for phonological rule application, either as a whole or as including separate domains within it (cf. section 4.2.1). Another possibility is that each member of the chain is realized as an independent morphological word, with no morphophonological interactions between the members in the chain (cf. section 4.2.2). The interest of this proposal for our purposes is that the morphosyntactic units containing the feature chain T-V visible at MS may be mapped as a phonological domain to PF, the component of grammar where phonological processes apply. We assume the validity of the notion of domain in phonology, and argue that there might be phonological processes that are specified to apply in such a unit, the morphosyntactic word (or ms-word, for short). This idea will provide us with a way to explain the phonological behavior of certain morphemes or bundles of morphemes realizing morphosyntactic features that cannot be captured in the theories of prosodic phonology that have been proposed (see section 5). In these theories no distinction is

44

made within the class of functional categories that are unarguably separate syntactic heads (the term used in the literature to refer to these expressions is function words). Those that behave as dependent phonologically or which interact closely with adjacent lexical heads triggering or undergoing phonological processes are simply stated to be defective categories from the prosodic point of view. They do not constitute prosodic words by themselves, and are incorporated into adjacent prosodic words, following a directionality parameter (cf. Nespor and Vogel 1982, 1986, Hayes 1989, Selkirk 1986, 1996). Some models argue that those dependent function words are prosodic words, and form clitic groups with adjacent lexical heads (cf. Nespor and Vogel 1986, Hayes 1989). Other models do not conceive dependent function words as prosodic words, and propose that they are directly integrated into adjacent prosodic words (cf. Selkirk 1984, 1986, 1996). The most radical of these approaches is probably stated in the Principle of the Categorial Invisibility of Function Words, suggested by Selkirk (1984), which claims that function words are not real words and are not identifiable as constituents of any sort in syntax or phonology. We will provide a more detailed overview of these theories in section 5, when we demonstrate the difficulties that these theories face in order to capture the domain of application of VA in Lekeitio Basque. At this point, suffice it to say that any theory which does not allow room for a distinction among functional categories is not able to explain the behavior of functional heads containing formal features of the inflectional type. As we have seen, they present the inherent property that they need to be licensed syntactically, a property that sets them apart from other function words which could fall under the ‘derivational’ class, such as adverbial particles, adpositions, or conjunctions. Our idea of morphological licensing is needed because inflectional heads may form separate prosodic domains for stress assignment (as in the case of the Basque auxiliary) but still interact phonologically with adjacent lexical heads in ways that other functional categories do not. That is, it will be shown that morphological domains may be

45

formed after syntax and that they may be a source of phonological domains, independent of the prosodic domain formation algorithms that have been shown to function successfully. Our argument is thus that the operations responsible for the creation of phonological domains are not reduceable to prosodic algorithms solely, but can stem from morphological domains created on the basis of morphosyntactic licensing. In other words, the licensing principle explains why phonological rules operating across words (syntactic terminals) often apply to elements with a high degree of morphosyntactic cohesion. (84)

[

In (84) we express this mapping from m-constituency to p-constituency:14 ]ms



[

]p

Whether a head is realized in situ or as adjoined to a head it forms a chain with is an independent matter, which The relevant idea for the approach to the interface between syntax and phonology that we will advocate in this dissertation is that

2.2.2. Feature chains as composed of independent words. A second possibility of realization of feature chains is that one or more of the elements composing the chain are realized as independent morphemes. In Vata, a Kru language of Africa, Koopman (1984) claims that there is a class of Infl elements which

14

This maping theory is in a way reminiscent of Inkelas’ (1990) ideas on the mapping between

morphological and prosodic constituency in the lexicon. The analysis I am positing here could thus perhaps be seen as an extension of her insights to the postsyntactic level.

46

are phonologically independent of the verb stem. They show no [ATR] harmony with the verb, a process which is characteristic of affixes, nor do they display tonal properties of affixes. The syntactic process of predicate clefting, for instance, copies and fronts the verb, but the tense particles are not repeated in the fronted verb (cf. Koopman 1984:150): (53) a.

lī n lī-dā zué e eat you eat-past yesterday Q ‘Did you eat yesterday?’

b. *lī-dā n lī-dā zué e eat-past you eat-PT yesterday Q ‘Did you eat yesterday?’

The interesting aspect of the tense morphemes is that according to Koopman, they trigger overt verb raising to Infl. However, these particles are not affixes, but independent words. That is, they do not possess a feature [Affix]. This points to an instance of a strong feature of T attracting the verb, which in the minimalist framework would be the categorial feature [V]. Clearer cases of feature chains with one or more of its members being realized as independent words would be those in which there is no overt raising. Complementizers in declarative subordinate clauses in German, English and Romance languages, for instance, do not possess strong features that trigger overt raising of the categorial feature T. If they did, T would be realized in C. The feature chain in these languages is thus realized with C as an independent word: (54) a. He said that you would come. b. María dijo que su madre estaba enferma. María said that her mother was sick

Determiners and prepositions in these languages also provide instances of feature chains (D-N and P-D, respectively) where the members are realized as separate

47

morphological words. D and N are realized as separate words in one case, the same as P and D in the other: (55) a.

the man

b.

el árbol the tree

c.

un enfant a child

d.

on a bridge

e.

para el marido for the husband

As regards the phonology of these chains, each morphological word in the chain is mapped as a separate phonological word. Thus, whatever rules in a language have as their domain of application the phonological word, they will apply in the phonological words realizing the feature chain. This is the possibility instantiated by Vata, where tense particles and verbs are independent domains of phonological processes such as [ATR] harmony and tone assignment processes. Schematically, we have the following representation of the mapping between syntactic and morphological structure: (56)

T-V

Syntactic derivation

Morphological Structure Phonological/ /Phonetic Form

[T]m-word

Spell-Out t [V]m-word

! ? [T]p [V]p

4.2.3. Feature chains as morphosyntactic words not composed of stem plus affixes.

48

A third possibility is that although none of the members in a chain are affixes, they still present a close relationship at the morphological and phonological level. The morphological cohesion is on surface that of two independent words (in the sense that they are not related by affixation or compounding), but the phonological mapping of the feature chain is that of a single domain. That is, the two elements interact very closely in a manner like that of phonological words mapped from a single morphological word (stem+affix). French liaison and Irish Consonant Mutation are examples of this possibility, to be studied in chapter 3. We will argue that this phonological behavior originates in the close degree of morphosyntactic cohesion between the two elements in question. Thus, we argue that in these cases the feature chain is mapped at MS as one constituent, although not in a typical manner of stem+affix or stem+stem (compounding). We will call this constituent a morphosyntactic word, to distinguish it from a morphological word understood in a traditional sense. The decision as to what constitutes a morphological word or a morphosyntactic word is a matter of lexical specification of the features in the chain or of the language in question. It is a setting of a parameter in the possible mappings from syntax to MS. This is the most interesting case and the one that will occupy most of our discussion in the studies of phonological phenomena we will present in this dissertation. In French, for instance, determiners, possessive pronouns, quantifiers and numerals are not morphologically realized as affixes of the following noun or prenominal adjective. However, they form the domain of obligatory liaison together with the following noun or adjective. Thus, the chain D-N is mapped as a morphosyntactic word, which in turn is interpreted as a phonological word. In Irish, complementizers, tense morphemes, possessive pronouns, determiners, and prepositions are not affixes, but form single phonological constituents with following

49

verbs or nouns. The feature chains D-N, P-D, C-T, and T-V are mapped at MS as morphosyntactic words, and as phonological words at PF. This option in the mapping between syntactic structure, morphological structure and phonological structure is expressed as in (57), with the chain D-N as an example: (57)

D-N

Syntactic derivation

Spell-Out t Morphological Structure Phonological/ /Phonetic Form

[D N]ms-word ! ? [D N]p

5. Summary and organization of the dissertation. The main objective in this dissertation is to provide a formal analysis of one aspect of the mapping between syntactic structure and morphophonological structure. It is argued that the relationships of feature specifications in the different syntactic heads driving the operation of feature checking (cf. Chomsky 1995) are primitive morphosyntactic entities in the form of feature chains: C-T, T-V, T-D, V-D, D-N, P-D. These chains are mapped to Morphological Structure and are realized morphologically in different ways, depending on the lexical specifications of the features in the chain. A feature chain may be spelled out as a single morphological word (m-word), with the shape stem+affixes, if one of the heads contains the feature [Affix]. The m-word is typically mapped as a single phonological domain, where phonological processes specified to apply in a phonological word will apply, giving rise to the effects of phonological interaction between stems and affixes. A case study showing intricate interactions between the affixal feature and categorial feature attraction is provided in chapter 1, with the phenomenon of Vowel Assimilation in Leketio Basque.

50

Another possibility in the syntax-MS mapping is that neither of the members in a feature chain is an affix. In this case, there are two alternatives available in the system. First, that the heads containing the features may be realized as independent words. Each head is an m-word, and thus by default also a separate phonological word, with the consequence that no phonological rules applying within a phonological domain will affect the two heads. This is instantiated by tense particles in Vata, and by prepositions in French. Second, that the morphosyntactic cohesion established by the feature chain and instantiated by the operation of feature checking relating the features and the heads containing them is realized as a cohesive constituent at MS, as a morphosyntactic word (ms-word). This constituent is then mapped as a single phonological domain. Chapter 3 contains discussions of these two possible scenarios, analyzing French liaison and Irish Consonant Mutation. Both chapter 2 and chapter 3 include critical comparisons with theories of prosodic phonology regarding predictive power, and the model advocated here will be shown to present several advantages over the alternative models in this respect.

46

Chapter 2 Feature licensing, morphological constituency and phonological domains in Basque

1. Introduction. In this chapter we analyze the phonological process of Vowel Assimilation (henceforth

VA)

in

Lekeitio

Basque.i

VA is an optional process in which a vowel assimilates in all its features to an immediately preceding vowel. The peculiarity of this process is its restricted distribution: it only applies in nominal and verbal contexts, between the final vowel of a noun or adjective and the initial vowel of a determiner or case marker, and between the final vowel of a lexical verb and the initial vowel of a following auxiliary verb. This property of VA poses serious problems for the theory of Lexical Phonology (Kiparsky 1982, Mohanan 1982, 1986) and theories of phrasal and prosodic phonology (Kaisse 1985, Nespor and Vogel 1986, Selkirk 1986), because the domains of application of the process have properties of both lexical and postlexical rules, and do not correspond to any constituent in the prosodic hierarchy. In this chapter we attempt a solution to the problem by exploiting the observation that the inflected auxiliary and the determiner are bound elements. We will argue that T and D have an [Affix] feature that makes them form a morphological word at MS with the participial verb or the last element in an NP, respectively. The affixal feature of T is strong, and requires the overt raising of the verb. The [Affix of N] feature of D is satisfied at MS, after an NP has moved to Spec,D overtly in syntax, to check a strong categorial feature N of D. VA applies within the m-words created by affixation of T to V and D to N. That is, the morphological domains so formed are mapped as phonological domains, where

47

phonological processes may be specified to apply. VA would be one such process, specified to apply between two elements contained in the same morphological word. This hypothesis discriminates an auxiliary from modal particles, causative verbs and conjunctions and connectives, which are not affixes. They are independent words, which is correlated with the fact that these elements do not undergo VA in the presence of a preceding syntactic head, i.e., a participial verb. If our hypothesis is on the right track, it will have consequences for our understanding of the mapping between syntax and phonology, since it will call for a rethinking of the algorithms for creation of phonological domains. We will have to pay more attention at how morphological relationships determine domains which are mapped as phonological constituents. The rest of the chapter is organized as follows: in section 2 we introduce the phenomenon of VA and provide the descriptive generalization that VA only occurs between nouns or adjectives and determiners, and verbs and auxiliaries. The structure of the clause and the DP in Basque is presented in section 3, showing the syntactic interactions between the elements participating in the process and the requirement that a finite auxiliary and a determiner be properly licensed. Our assumptions are compared with alternative proposals, which are shown to present several disadvantages which call for their rejection. Section 4 contains the analysis of the problem, based on the idea that T and D are morphologically deficient, and need to be morphologically licensed by receiving the incorporation of a syntactic head. The unit so formed is interpreted as a phonological domain where VA applies. Section 5 addresses the problems that the domains of application of VA pose for the different theories of the syntax-phonology interface proposed in the literature, suggesting that new facets of the interface between these modules of grammar must be paid attention to. Section 6 ends the paper with a summary and main conclusions.

48

2. VA in Lekeitio Basque: data and generalizations. 2.1. Preliminary data. VA in Lekeitio Basque is an optional rule of colloquial speech by which a syllable-initial vowel assimilates in all its features to an immediately preceding syllablefinal vowel. This rule applies word-internally in nominal contexts and across word boundaries in verbal contexts. In nominal contexts the vowel whose features are copied is always high, and the vowel copying the features is always unround. In verbal contexts, where the rule applies across words, there is more freedom in the quality of the vowels involved. This is illustrated in (1), where, for the cases in which VA applies within words, we represent syllable boundaries with a dot notation: (1) a. nes. ki.a ‘the girl’



nes.ki.i



a.ba.di.i.ri

b.

a.ba.di.a.ri ‘to the priest’

c.

i.txa.su.an  ‘in the sea/ocean’

i.txa.su.un ii

d.

be.su.é.tan ‘in the arms’



be.su.ú.tan

e.

asma eban  ‘(s)he made it up’

asma aban

f.

erosi ebésen  ‘they bought them’

erosi ibésen

g.

ostu ebasan  ‘(s)he stole them’

ostu ubasan

The restriction that in nominal contexts the vowel whose features are copied must be high is due to the application of another rule prior to VA, the rule of Vowel Raising (VR), which raises a stem-final [-high] vowel when immediately followed by a vowel-

49

initial suffix. This means that no stem-final vowels are [-high] on surface. Thus, the forms on the left in (1a-d) are themselves derived. All of them possess underlying nonhigh stem-final vowels, which get raised in the presence of the suffixes /-a/ and /-eta/ (the singular determiner and the locative plural determiner, respectively). These would be the derivations giving rise to the forms on the left in (1a-d): iii (2) a. /neska-a/ girl-det.sg.

neskia

‘the girl’

b. /abade-a-ri/  priest-det.sg.-dat.

abadiari

‘to the priest’

c. /itxaso-a-n/ sea-det.sg.-ines.



itxasuan

‘in the sea’

d. /beso-eta-n/ arm-det.pl.-ines.



besuétan

‘in the arms’



When an underlying stem-final high front vowel (i.e., /i/) and a suffix-initial vowel become adjacent, a rule inserting a voiced prepalatal fricative (i.e., /ž/) takes place. This consonant is then syllabified as an onset of the suffix-initial vowel. Underlying nonhigh front vowels which become high by VR do not trigger this rule. Thus, compare the two examples in (3), which constitute a minimal pair: (3) a. /barri-a/ new-det.sg. b.



/barre-a/  laughter-det.sg.

ba.rri.ža

‘new, the new one’

ba.rri.a ‘the laughter’

The insertion of /ž/ bleeds the application of VA, which needs two adjacent vowels to take place. An underlying high back vowel (i.e., /u/) does not trigger any kind of consonant insertion rule, and thus both underlying high back vowels and underlying mid back vowels which become high by VR intervene in VA environments.

50

As we mentioned, only VR is obligatory. VA applies optionally. It operates on those forms which have already undergone VR (cf. (2), for instance).iv

2.2. Morphosyntactic distribution of VA. The rule of VA has a very restricted domain of application. In nominal contexts, it only applies between the final vowel of a noun or adjective and the initial vowel of a following inflectional head, i.e., a determiner or case marker. It does not apply across members of compounds, or between a noun and an adjective. In verbal contexts, it applies between the final vowel of a verb and the following initial vowel of an inflected auxiliary. It does not apply between a verb and a following causative verb, modal particle, or subordinating conjunction. Capturing the domain of application of VA will be our main goal in this chapter and the theoretical model proposed will serve as the basis of our explanation for other phonological phenomena with similar distributions in other languages. Let us consider each of these contexts in turn.

2.2.1. Nominal contexts. 2.2.1.1. Nominal inflection. Nominal inflection in Basque is morphologically attached to the last word of the last constituent of the Noun Phrase, not to every constituent contained in it. Thus, when a noun is followed by an adjective, the determiner and case markers or postpositions will be added to the adjective, the noun remaining in its bare uninflected form (cf. (4d, g, i) in the examples below). The determiner in Basque has distinct singular and plural forms, with a further distinction in the plural determiner between locative and nonlocative cases: -a is the singular determiner, -ak is the plural determiner, and -eta is the plural determiner for locative cases. For each of the underlying forms in (4) we can obtain two alternative outputs, which we separate with a slash. The one on the left constitutes the obligatory surface output, and the one on the right represents the optional application of VA. We

51

mark syllable boundaries with a dot notation, to show that VA does not apply within a syllable, i.e., in diphthongs, and that it does not create tautosyllabic long vowels: (4) a. /orma-a/ wall-det.sg. ‘the wall’



or.mi.a v / or.mi.i

b.

/baso-ak*/ vi forest-det.pl. ‘the forests’



ba.sú.ak / ba.sú.uk

c.

/andra-a-k/  woman-det.sg.-erg. ‘the woman’

an.dri.ak / an.dri.ik

d.

/gixon tonto*-ak*-k/  man stupid-det.pl.-erg. ‘stupid men’

e.

/seme-a-ri/  son-det.sg.dat. ‘to the son’

se.mi.a.ri / se.mi.i.ri

f.

/ume-en*/  child-gen.pl. ‘of the children’

u.mí.en / u.mí.in

g.

/polísi gaixto-ak*-kin*/ policeman bad-det.pl.-soc. ‘with bad policemen’

h.

/etxe-a-n/  house-det.sg.-ines. ‘at/in the house’

i.

/kale estu-eta*-n/  street narrow-det.pl.-ines. ‘at/in the narrow streets’

j.

/ortu-eta*-tik*/ orchard-det.pl.-abl.

gixon ton.tú.ak / gixon ton.tú.uk



polísi gaix.tu.á.kiñ / polísi gaix.tu.ú.kiñ

e.txi.an / e.txi.in



kale es.tu.é.tan / kale es.tu.ú.tan

or.tu.e.tá.tik / or.tu.utá.tik

52

High vowels do not undergo the rule simply because they never occur in the relevant contexts, that is, there are no suffixes which begin with a high vowel. The round mid vowel /o/ does not assimilate, as shown in (5) for the proximative plural determiner and the proximative genitive plural marker: (5) a. /baso-ok*/  forest-det.pl.prox. b.

/ume-on*/  child-gen.pl.prox.

basúok / *basúuk

‘forests’

umíon / *umíin

‘of the children’

The affix expressing the meaning of superlative degree appears is attached to adjectival roots, and appears before a determiner. This affix also triggers VR on the last vowel of the root, and undergoes VA:vii (6)

/soro-en*-a/  crazy-sup-det ‘the craziest’

soruéna / soruúna

2.2.1.2. Roots. VA may also apply in underived domains, i.e., roots, although the application of the rule seems to be lexically determined: (7)

bi.ar / bi.ir si.es.ta / si.is.ta si.ar / *si.ir bi.á.je / *bi.í.je tri.an.gú.lo / *tri.in.gú.lo

‘to need’ ‘nap’ ‘through’ ‘trip’ ‘triangle’

2.2.1.3. Derivational morphemes, compounding, and word boundaries. Cases of rules where derivational and inflectional morphemes present a different behavior with respect to a phonological rule have been cited numerously in the literature

53

on morphophonology, revealing interesting implications for an understanding of the structure of the different submodules of grammar. However, for the case of Lekeitio Basque the issue of a possible distinction in distribution between derivational and inflectional morphemes with respect to VA does not arise. Derivational morphemes are consonant-initial in Lekeitio Basque, so it is not possible to test the behavior of derivational morphemes with respect to VA. Nevertheless, there is one example where a vowel-initial derivational morpheme is attached to a noun ending in a vowel, and interestingly, no VA occurs: (9)

/donosti-ar*-a/  do.nos.ti.á.rra / *do.nos.ti.í.rra Donostia-from-det.sg. ‘(a native of) Donostia/San Sebastian’

The rule of VA does not apply between two members of a compound or across words. See (10) and (11), respectively: (10) a. /buru-andi/  head-big ‘big-headed’

bu.ru.án.di / *bu.ru.ún.di

b. /etxe-ondo/  house-side ‘side of house’

e.txe.ón.do / *e.txe.éndo

c. /soro-antz/  mad-look ‘mad look, aspect’

so.ro.án.tza / *so.ro.ón.tza

(11) a. /seru asula/ sky blue ‘blue sky’



b. /etxe andiža/  house big ‘big house’

seru a.su.la / *se.ru u.su.la

etxe an.di.ža / *etxe endiža

54

2.2.2. Verbal contexts. 2.2.2.1. Inflected auxiliaries. In verbal contexts, the rule of VA applies between a lexical verb and a vowelinitial inflected auxiliary. The lexical verb is called participial verb in the traditional literature, because it is inflected for aspect, with progressive, perfective and future value, a term we will also use henceforth to refer to the lexical verb. Perfective aspect is realized by the suffixes -i, -tu, -n, and -{, to which the suffix -ko can be added to convey future tense, combined with the present tense appearing on the inflected auxiliary. Imperfective aspect is realized by the suffix -t(z)en: (12) a. eros-i dot buy-perf. aux

‘I have bought it’

b. gal-du senduan lose-perf. aux

‘you lost it’

c. eros-i-ko dot buy-perf.-fut. aux

‘I will buy it’

d. eros-ten dot buy-imperf. aux e. eros-ten neban buy-imperf. aux

‘I buy it’ ‘I used to buy it’

Inflected auxiliaries are formed by the amalgamation of agreement markers and tense and mood morphemes with the roots of the auxiliaries izan, *edun, *edin and *egin, of which only the first survives as an independent verb, the copula ‘to be’.viii The auxiliary roots are -a-, -o- and -ei-, respectively. Absolutive markers are normally prefixed to the auxiliary root, and dative and ergative markers and tense and mood morphemes are suffixed to the root, except for past tense forms of ditransitive verbs, where ergative markers are prefixed to the root (compare (13e) and (13f) below). This is a phenomenon known as ergativity displacement, which also occurs in other ergative

55

languages such as Georgian. As the examples in (13) show, inflected auxiliaries form a separate word from the participial verb:ix (13) a. apur-tu d-o-gu break-perf. 3abs.sg.-rt-1erg.pl. ‘We have broken it’ b. emo-n d-o-tzu-t give-perf. 3abs.sg.-rt-2dat.sg.-1erg.sg. ‘I have given to you’ c. etor-ten zirian come-imperf. 3abs.pl.-past ‘They used to come’ d. ikas-i l-ei-ke-t learn-perf. 3abs.sg.-rt-pot.-1erg.sg. ‘I can learn it’ e. ekarr-i d-o-zu bring-perf. 3abs.sg.-rt-2sg.erg. ‘You have brought it’ f. ekarr-i zen-du-an bring-perf. 2erg.sg.-rt-abs.sg.-past ‘You brought it’

All forms in the verbal paradigm of Lekeitio Basque present an initial consonant, representing an absolutive or ergative marker. Past tense verbal forms with a third person ergative marker are the only exception.x The third person ergative marker is not overtly realized, i.e., it is realized as {. The initial vowel in these forms is the mid front vowel e, which appears in nonpresent and irrealis forms (i.e., past tense, potential and imperative). Present tense forms show the vowel a, appearing immediately after the absolutive and ergative markers. These vowels, a and e, have been identified as the markers -a- and -e-, associated with present and nonpresent/irrealis verb forms, respectively. We thus gloss the vowel e as a nonpresent morpheme, -pres.xi

56

In this context no raising of the final vowel of the lexical verb occurs; VR is restricted to the boundaries created by nominal inflection and certain roots:xii, xiii (14) a. /žo e-ba-n/ hit 3erg.--pres.-rt.xiv-past ‘(s)he hit him/her/it’



žo eban / žo oban

b. /galdu e-ba-sxv-an/  lose 3erg.--pres.-rt-3abs.pl.-past ‘(s)he lost them’

galdu ebasan / galdu ubasan

c. /ikasi e-b-e*xvi-n/  learn 3erg.--pres.-rt.-erg.pl.-past ‘they learnt it’

ikasi ében / ikasi íben

d. /atrapa e-b-e*-s-en/  atrapa ebésen / atrapa abésen catch 3erg.--pres.-rt-erg.pl.-3abs.pl.-past ‘they caught them’

VA may apply in nonderived environments in the few verb roots that present two heterosyllabic vowels in a sequence. These are verbs borrowed from Spanish verbs ending in -ear, adapted as -ia by the rule of VR: (15)

ma.ri.a / ma.ri.i ‘make/feel dizzy’ moski.a / mos.ki.i ‘to get angry’ sor.ti.a / sor.ti.i ‘to raffle’

< Sp. marear < Sp. mosquear < Sp. sortear xvii

2.2.2.2. Causative verbs. VA does not apply between a lexical verb and a causative verb, eraiñ, which in linear sequence appears between the lexical verb and the inflected auxiliary: (16) a. paga eraiñ neutzan / pay make aux ‘I made him/her pay’

*paga araiñ

b. altza eraiñ dotzat / *altza araiñ raise make ‘I have made him/her raise’

57

2.2.2.3. Modal particles. There are two modal particles which constitute independent syntactic heads and that may intervene between the lexical verb and the inflected auxiliary. These particles do not contribute anything to the propositional content of a sentence, and their basic semantic function is to express epistemic attitudes of the speaker concerning the existence or non-existence of the state of affairs identified by other elements in the sentence. The modal particle ete appears in interrogative and exclamative sentences, and conveys a meaning of wondering, uncertainty, doubt, suspicion, on the part of the speaker about the event expressed in the sentence, and ei indicates that what is being expressed in the sentence has reported by other people and that the speaker cannot fully assure the veracity of the event denoted by the proposition. We call the particles ete and ei ‘dubitative’ and ‘evidential’, respectively. Modal verbs in other languages also have epistemological usages (e.g., may, might, in English, poder, deber in Spanish), but these particles should not be classified as modal verbs, since these particles are not verbs to begin with. They are not predicates, they do not take any arguments and they are never inflected, unlike regular verbs. Moreover, Basque does have clear modal verbs (nahi ‘to want’, behar ‘to need’), whose syntactic properties are very different from modal particles.xviii Also, note that the use of the term “modal” for these particles is not related to the grammatical concept of mood, as in indicative mood or subjunctive mood. Basque has moods independently of the modal particles. No VA occurs between a lexical verb and these particles: (17) a. etorri ete díras? / *etorri ite díras? come dub. aux ‘I wonder whether they have come’ b. atrapa ei dósu

/

*atrapa ai dósu

58

catch evid. aux ‘I have heard that you have caught it’

2.2.2.4. Subordinating conjunctions. In adverbial nonfinite clauses, the verb appears followed by a subordinating conjunction. No VA applies between these elements either: (18) a. ekarri árren / bring despite

*ekarri írren

b. kompondu ezik fix unless

/

*kompondu uzik

c. amaitxu árte / finish until

*amaitxu úrte

2.2.2.5. Across any two words. VA does not occur across any other two words, such as a direct or indirect object and a verb (cf. (19a,b), a subject and a verb (19c), or two objects (cf. (19d)). For ease of identification, we have underlined the relevant vowels: / (19) a. arraña erosi dau fish buy aux ‘(s)he has bought fish’

*arraña arosi dau

b. amari astu žako mother(dat.) forget aux ‘the mother has forgotten’

/

*amari istu žako

c. laguna etorri da / friend come aux ‘the friend has come’

*laguna atorri da

d. amumári erregalúa ein dotzagu / *amumári irregalúa dotzagu grandmother-dat. present-abs. make aux ‘We have made (i.e., bought and given) a present for grandmother’

ein

59

2.2.3. Summary of distribution of VA. To summarize our presentation so far, we have shown that in Lekeitio Basque there is an optional rule of VA by which a vowel loses all its features and assimilates to the vowel located immediately to its left. VA applies in nominal and verbal contexts. In nominal contexts, it applies between the final vowel of a noun or adjective and the initial vowel of a determiner or case marker. The final vowel of the lexical head is always high. Those vowels which were not underlyingly high become [+high] by virtue due to the application of the rule of VR, which raises the final vowel of a noun or adjective when preceding a vowel-initial inflectional suffix. We have also shown that VA can apply in certain lexically marked nonderived domains. It cannot apply, however, between two members of a compound noun or adjective, or across two lexical heads, e.g., a noun and an adjective. In verbal contexts, VA applies between the final vowel of a lexical verb and the initial vowel of a following inflected auxiliary. VR does not apply to the final vowel of the lexical verb; this rule is restricted to morphological concatenation, i.e., nominal inflection. No VA occurs between the verb and other following elements, such as modal particles, a causative verb, or subordinating conjunctions. VA cannot apply between any two other words, either; no VA is found between a subject or object and a verb or between two objects, for instance. The preliminary distributional generalization is then that VA only applies between lexical heads and following elements realizing inflectional features, such as determiners and inflected auxiliaries. The distribution of VA is important in that it unveils the serious problems that the different theories of phrasal and prosodic phonology have to face in order to account for phenomena of this type, in which a contrast exists between functional heads realizing morphosyntactic features of the inflectional sort and function

60

words which do not have these characteristics, such as adverbial or modal particles, degree adverbs, adpositions, or subordinating conjunctions. In this chapter and the ones that follow we will bring forth evidence of the empirical inadequacies of these theories to correctly capture the nature of the relevant domains. We suggest that a different framework is needed, one which takes morphosyntactic relationships among elements into account. Our alternative analysis is based on the notion of feature licensing of inflectional categories. Since the different predictions made by the theories of phrasal and prosodic phonology as well as our analysis depend crucially on the syntactic structure of the derivation being mapped to the postsyntactic phonological component, it is necessary to present first the structure of the linguistic derivation in overt syntax in Basque. This will be our task in the next section.

3. Clause structure, head movement and Infl licensing in Basque. 3.1. Clause structure. Before we begin this section, a cautionary note is warranted. Although in my opinion the analysis of the structure of the clause in Basque and the syntactic operations of head movement that I am going to posit have several advantages over other earlier proposals, they still constitute a tentative picture. There remain open issues that require a deeper investigation, some of which I will mention, and perhaps there exists a better solution than the one offered here. The structure of the clause in Basque, the syntactic movements responsible for the observed word order patterns, and even the position where arguments are generated are still topics of open debate among Basque linguists, and a thorough study into all these issues is far from the objectives in this dissertation. Our aim is to provide an analysis of the relations among heads, and more specifically of the attraction that the inflected auxiliary exerts over local heads. Indeed, at the end of this section I will show that with some modifications, some of the analyses that have been

61

proposed could be compatible with our main idea on the dependent nature of the auxiliary. In our opinion, this strengthens the validity of the view I want to defend. I suggest that Basque has a head-initial clause structure, with the hierarchical organization among the different projections represented in (20). I obviate any discussion as to how the phrase structure is produced through Merge. I also simplify matters somehow by assigning each projection its own Spec position and avoiding the possibility that some of the maximal projections are also minimal at the same time (cf. Chomsky 1994, 1995), because it is not an issue of importance here. vP is also omitted from the discussion, for the same reason: (20)

CP 3 C0 TP 3 T0 ModP 3 Mod0 NegP 3 Neg0 VP 3 DP V’ 3 V0 DP

This proposal runs against most of the earlier analyses of the clause structure of Basque. Generative grammarians have been assuming head-final structures for this language, following descriptive observations that heads follow their complements across almost all categories. Ortiz de Urbina (1989) provided evidence that C0 is initial, however (cf. also Ortiz de Urbina 1994, 1995 and Albizu 1991, 1992). The main argument is that in constructions involving operators, such as wh and yes/no questions, focus constructions, and negative sentences, the inflected auxiliary occurs on the left edge of the sentence. In interrogative sentences and focus constructions the participial verb

62

appears left adjacent to the inflected auxiliary, forming a verbal complex, and in negative sentences only the auxiliary appears following negation, leaving the participle stranded, in its in-situ position (cf. (21) below). No element can intervene between a wh-phrase, the target of a yes/no question, a focalized constituent or negation and the verbal element(s) that follow. This pattern suggests the verbal elements raise to C0 to enter in a Spec-head relationship with the element in Spec,CP, along the lines of Rizzi’s wh-criterion extended to all operator-involving constructions. This is unexpected under the assumption that all heads are final in Basque: (21) a. Zer esan dótzo Péruk Mirenéri? what-A say aux Peru-E Miren-D ‘What has Peru said to Miren?’ b. Peru etorri da gaur goixian? Peru come aux today morning ‘Was it Peru that came this morning?’ c. Péruk ekarri dau liburúa gaur goixian Peru-E bring aux book-A today morning ‘Peru brought the book this morning’ d. Ez dozu ekarri liburúa gaur goixian neg aux bring book today morning ‘You didn’t bring the book this morning’

Other researchers have claimed head-initial status for projections below CP. For instance, Laka (1990) proposes a head-initial ΣP, located between CP and IP, and Artiagoitia (1992) argues that I0 is initial.xix Thus, even under the assumption that Basque is head-final, not all projections can be analyzed as head-final. The structure I am positing is simply taking a step forward towards coherence in the head directionality parameter, maintaining the relative hierarchical order among heads suggested by Ortiz de Urbina (1994, 1995). This alternative proposal is compatible with Kayne’s (1994) theory of the correspondence between linear order and hierarchical structure, namely that

63

asymmetric c-command maps into linear order. The consequence of this hypothesis is that all languages are head-initial underlyingly and that they display a universal SVO word order, with a specifier-head-complement linear order. In this theory, SOV languages differ from SVO languages in that they involve movement of the object past the verb to a higher functional projection. Working within Kayne’s system, Albizu (1994) and Ormazabal, Uriagereka and Uribe-Etxebarria (1994) have recently proposed a headinitial clause structure for Basque.

3.2. Head movement. The structure I am positing predicts the correct surface linear order among the non-nominal heads in a clause. Let us present all the different linear orders observed: (22) a. Participle - inflected auxiliary:xx etorri da come aux b. Participle - modal - inflected auxiliary: etorri ei da come mod aux c. Negation - inflected auxiliary ... verb: ez da iñor etorri neg aux anybody come d. Negation - modal - inflected auxiliary ... verb: ez ei da iñor etorri neg mod aux anybody come

Assuming that only left-adjunction is possible (cf. Kayne 1994), the order in (22a) is derived by movement of the participial verb to T0 in declarative sentences. The order in (22b) is obtained in a similar way, with the participial verb raising to T0 by head-to-head

64

movement, incorporating the modal particle on its way up. These configurations are represented in (23) below: (23)

TP wp T0 (ModP) ! 3 [etorrii (eij)dal]l (Mod0) ! ! tj

VP ! V0 ! ti

In (22c) negation raises to T0 on its way to C0, which I assume is the landing position of negation, without recurring to a ΣP projection located between CP and TP (cf. Laka 1990). I follow Ortiz de Urbina (1989, 1995) in claiming that the raising of negation is prompted by the need to occupy the head position of a projection containing an operator, in this case a negative operator. This explains the left-edge position that negation and a following inflected auxiliary occupy in negative sentences. In (22d) the same process applies, with the modal head being incorporated by negation. We represent these derivations in (24):

65

(24)

CP wp C0 TP ! 3 ! T0 (ModP) [ezi (eij) dak,l]l ! 3 ! (Mod0) NegP tl ! 3 ! Neg0 VP tj ! ! ! V0 ti ! etorri

3.3. Infl (T) licensing. An interesting point to discuss is what motivates the raising of the participial verb to T. Notice first of all that in these structures movement of the participle is not obligatory in all contexts. In negative clauses, it is negation that raises to T0, and the participle remains in situ inside the VP, as evidenced by the fact that arguments and adjuncts may intervene between negation and the inflected auxiliary and the participial verb (cf. (22c,d)). Only in declarative and interrogative sentences does the participle raise to T0 (cf. (21a,b) for examples of interrogative clauses). The reason behind this distributional contrast is related to the first question. Ortiz de Urbina (1994) argues that tensed Infl is a weak or dependent element syntactically, with certain syntactic collocational restrictions.xxi He bases his argument on the observation that in operatorinvolving constructions in which a verb is attracted to C0 (verbal focalization constructions and yes/no questions, cf. Ortiz de Urbina 1989, 1994, 1995), verbal forms containing tense features cannot occur in sentence-initial position. This can be observed with inflected auxiliaries and also with synthetic verbs, which are formed by the amalgamation of a bound verbal root and agreement and tense markers, without an

66

auxiliary or aspectual markers. The majority of verbs can only be conjugated in the periphrastic form, that is, formed by the combination of a participial verb (with an aspectual marker) and an inflected auxiliary (with agreement and tense morphology), as already explained and illustrated in section 2 (cf. (12)-(14)). Synthetic verbs are thus single verbal forms realizing tensed Infl, and are subject to the same constraint against appearing in sentence-initial position as inflected auxiliaries. The following are some examples of synthetic verb forms, their morphological composition indicated by hyphenation:xxii (25) a. zure laguna gaur da-tor your friend today 3abs.sg.-come ‘Your friend is coming today’ b. etxera g-óya-z home-all. 1abs.pl.-go-pl. ‘We are going home’ c.

arraiñ asko da-kárr-e mariñerúak fish much 3abs.sg.-bring-3erg.pl. fishermen ‘The fishermen are bringing lots of fish’

The bound roots of the lexical verbs in (25) are /-tor-/, /-oya*-/, and /-kar-/ (cf. their respective corresponding forms used in periphrastic conjugations: etorri ‘to come’, žuan ‘to go’, and ekarri ‘to bring’). In (26) and (27) we illustrate the constraint against sentence-initial position for inflected auxiliaries and synthetic verbs, respectively:xxiii (26) a. Jonek liburúa erosi dau Jon book buy aux ‘Jon has bought the book’ b. *dau erosi Jonek liburúa erosi buy aux Jon book ‘Jon has bought the book’

/ *dau Jonek erosi liburúa / *dau Jonek liburúa

67

c. *dau erosi Jonek liburua? / *dau Jonek erosi liburúa? erosi? ‘Has John bought the book?’

/

*dau Jonek liburúa

(27) a. Jonek liburúa dakar Jon book brings ‘Jon is bringing the book’ b. *Dakar Jonek liburúa / brings Jon book

*dakar liburúa Jonek.

c. *Dakar Jonek liburúa? / *dakar liburúa Jonek? ‘Is Jon bringing the book?’

The ungrammatical examples in (26) become acceptable when the participial verb appears before the auxiliary, and those in (27) become grammatical when the particle ba appears to the left of the synthetic verb, to shield it from sentence-initial position: (28) a. Erosi dau Jonek liburúa buy aux Jon book ‘Jon has bought the book’ b. Erosi dau Jonek liburúa? ‘Has Jon bought the book?’ (29) a. Ba-dakar Jonek liburua ‘Jon is bringing the book’ b. Ba-dakar Jonek liburua? ‘Is Jon bringing the book?’ Negation also counts as first element in the sentence: (30) a. *dator Jon comes Jon b. Ez dator Jon neg comes Jon ‘Jon is not coming’

68

Wh-phrases shield a synthetic verb from sentence-initial position as well: (31) a. Zer dakar Jonek? what brings Jon ‘What is Jon bringing?’ b. Nor dator? who comes ‘Who is coming?’

As we mentioned above, in the standard and southern dialects, inflected auxiliaries are always preceded by participial verbs in declarative and interrogative sentences without negation. In northern dialects, however, it is possible to front the inflected auxiliary alone, without the participial verb, in constructions involving operators, i.e., in interrogatives and focus constructions: (32) a. Zer du Jonek erosi? what has Jon buy ‘What has John bought?’ b. Liburua du Jonek erosi book has Jon buy ‘Jon has bought the book’

Crucially, inflected auxiliaries pattern exactly like synthetic verbs in all dialects, that is, they cannot appear in absolute clause-initial position: (33) a. *du Jonek liburua erosi has Jon book buy b. Ba-du Jonek liburua erosi ‘Jon has bought the book’ (34) a. *du Jonek liburua erosi? b. Ba-du Jonek liburua erosi?

69

‘Has Jon bought the book?’

Synthetic verbs in the imperative mood constitute evidence that it is the feature [tense] that has this property, and not all the inflectional features realized in the auxiliary, such as agreement. Imperative forms are inflected for agreement but not for tense, and they do not show the limitations that finite synthetic forms present: (35) a. zóyaz hortik! go-2sg.imp. there-abl. ‘Go to hell!’ (lit., ‘go around there!’) b. Betor aita! come father

(From Ortiz de Urbina (1989:154))

‘Let the father come!’ Based on Rivero’s (1993) analysis for similar facts in Breton, Ortiz de Urbina (1994) proposes that this prohibition against having an element inflected for [tense] on the left edge of a clause is due to the fact that [tense] is a feature that needs to be licensed overtly in its checking domain, that is, in the projection it ends up in overt syntax. The licensing is done either by the incorporation of a lexical head onto I0, or by the presence of an element in the specifier position of the projection of I0. The participial verb and negation are heads which can license finite Infl by incorporating onto it, and wh-phrases and focalized constituents also license [tense] by virtue of filling the specifier position of the projection where I0 ends up in those constructions, i.e., Spec,CP (cf. Borsley and Rivero 1994 and Rivero 1994 for similar treatments of the clitic nature of auxiliaries in the Balkan languages, as well as the above mentioned Rivero 1993 and Borsley, Rivero and Stephens 1993 for Breton auxiliaries). As for the particle ba-, I agree with Ortiz de Urbina in assuming that it is a particle introduced by last resort to license [tense], in the absence of another head or element in the specifier position of CP. This particle is inserted in C0.xxiv This property of [tense] was already noted by Michelena (1957), who

70

mentions “the enclitic nature of finite verbal forms, at least auxiliaries” (Michelena 1957: 177, fn. 32). In a way, these constraints resemble in some sense the much discussed Wackernagel effects, if we considered one of the interpretations of these effects, namely that these effects derive from the necessity to fill the C0 position and have a maximal projection in Spec,CP. Of course, the resemblance is only distant, because on the one hand a finite verb only raises to C0 in operator-involving constructions, not in declaratives, and on the other hand, even in these constructions the finite verb in southern dialects is preceded noy just by one constituent, as in V2 constructions, but by an element in Spec, CP and the participial verb. It is this second property that will spark the core of our analysis. A schematic representation in (36) illustrates the configurations where Infl is licensed in operator-involving constructions. This configuration is the checking domain of Infl, in Chomsky’s (1993, 1995) sense. XP stands for an overt maximal projection in Spec,CP: (36) CP 3 XP C’ 3 C IP ! (Participle+) Infl Neg+Infl Ba+Infl

The parenthesis surrounding the participle indicates that it never needs to raise to C0 in periphrastic constructions in northern dialects (cf. (32)). In southern dialects such as Lekeitio Basque the participial verb has to raise to I0 and then to C0 even in the presence of an element in Spec,CP. Thus, compare (32) with (37): (37) a. Liburúa erosi dau Jonek book buy aux Jon ‘Jon has bought the book’ b. Nok erosi dau liburúa? who buy aux book ‘Who has bought the book?’

71

Ortiz de Urbina maintains that the overtly realized XP in Spec,CP is sufficient to license Infl in some northern dialects, but that in southern dialects the participial verb raises to Infl for other reasons. He follows proposals by Horvath (1986, 1995) and Tuller (1992) for focus phenomena in Chadic, Hungarian, and other languages that the functional head hosting the syntactic features [wh] and [Focus] (i.e., Infl) must be lexicalized by verb raising. He extends this idea to the feature [tense] in Basque, and argues that a lexical head must move to it, in root clauses at least. That is, the movement of a participial verb to Infl is independent of any further movements or any other syntactic operations affecting Infl. The verb raises to Infl because this head is a weak morphological head that needs to be lexicalized. The Infl head contains only inflectional morphemes realizing tense, mood, case, number, and person features, and a bound root of an auxiliary verb (for instance, from the transitive auxiliary *edun, only -u-, -o- or -esurfaces, depending on the dialect (cf. the glosses in (13) for Lekeitio Basque). Thus, the tensed I0 requires the raising of a verbal head which is morphologically strong, i.e., free or independent. This movement takes place not only in sentences involving an operator in CP (i.e., focus constructions and interrogative sentences), but in declarative sentences as well. In sentences involving operators which attract Infl to C0, the verb will necessarily appear left-adjacent to Infl in C0, because it forms a complex head with it, V-Infl. The two independent steps are represented in (38) and (39), (38) corresponding to the raising of V to Infl, and (39) to the movement from I to C. Only the latter is common to all dialects of Basque. The former is not observed in some northern dialects, that is why we indicate the verb in parenthesis in (39). Ortiz de Urbina assumes head-finality for all projections except CP, which accounts for the occurrence of negation and the inflected auxiliary in the left periphery of the clause: (38)

IP 2 VP V-I0

72

2 : t ! z--m (39)

CP 3 Op C’ (+wh, Q...) 2 C IP (V)-Infl 2 : VP t !2 ! ! t ! z------m

In negative sentences, negation raises to Infl alone, without the participle, providing lexical support for the [tense] feature in Infl. Ortiz de Urbina posits a NegP between IP and VP. Then the complex formed by negation and Infl moves to C, for what Ortiz de Urbina suggests are scopal properties of negation. Notice that under the assumption that Basque is right-headed except for C0, Ortiz de Urbina needs to assume that negation raises to C0, in order to explain the normal occurrence of negation in the left edge of the sentence, even in subordinate clauses which are complements of bridge verbs. Cf. the example from Standard Basque in (40), and its version in Lekeitio Basque (41): (40) uste dut [ez dela etorri] think aux neg aux-comp come ‘I think (s)he hasn’t come’ (41) pentzaten dot [ez dala etorri.

Another argument Ortiz de Urbina mentions in favor of the raising of Neg to C is the licensing of subject NPIs in Basque: (42) Ez zuen inork ikusi neg aux anybody see ‘Nobody saw (it)’

73

If NPI licensing is done under c-command by negation, as standardly assumed, then the subject NPI in (42) must be under the scope of negation. Thus, for a sentence such as (43) (his (47b), we have the derivation in (44) (his (49)): (43) (44)

ez du Jonek liburua irakurri neg aux Jon book read CP 3 C IP ! 3 ez-dui Jonek I’ 2 NegP I 2 ! VP ti ti 2 NP V ! ! liburua irakurri

Like inflected auxiliaries, synthetic verbs appear adjacent to negation forming one complex head with it, as suggested by sentences like the ones in (45): (45) a. Ez dator Jon etxétik. neg. come-3sg. Jon house-abl. ‘Jon is not coming from home’ b. Nóixtik ez dábil Amáya Txomiñégaz? when-abl. neg. walk-3sg. Amaya Txomin-comit. ‘Since when is not Amaya going out with Txomin?’

Ortiz de Urbina argues that the verbal bound roots in synthetic conjugations (e.g., -tor- and -bil-, in (45)) raise to I0, skipping negation, and then negation raises to the

74

complex head formed by the synthetic verb. The derivation of the sentence in (45a) could be represented as follows: (46)

CP 3 C IP ! 2 ez-datori,j Jon I’ 2 NegP I 2 ! VP tj ti,j 2 PP V ! ! etxétik ti

This long head movement derivation violates the Head Movement Constraint (cf. Travis 1994), but Ortiz de Urbina defends this analysis on the basis of similar proposals for different languages by Belletti (1990), Roberts (1992), and Rivero (1993), who adopt Relativized Minimality to allow a distinction between L-related and non L-related heads. In this approach, movement of V over Neg would not violate ECP or the HMC under the hypothesis that Neg is a non L-related head, and V and Infl are L-related (cf. also Lema and Rivero 1989, 1991, Rivero 1991, 1994 and Borsley and Rivero 1994, among others). We follow Ortiz de Urbina’s analysis of Infl licensing, which in our terms would be T licensing, and propose that this head is a weak or dependent morphological word that needs to be supported. We argue that T has a strong feature [Affix], and a participial verb and negation are independent words that can check this feature, by providing morphological support to the head T. However, when a modal particle or a causative verb are present in the structure, we could expect them to be able to support T, thus giving the order modal-auxiliary.....participial verb or causative-auxiliary....participial verb, with the participial verb in situ. Why do we then have the order participial verb-modal-

75

auxiliary and participial verb-causative-auxiliary? Why does the participial verb have to raise and adjoin to the left of modal particles and the causative verb? The only reason for movement in the minimalist program is feature checking, so the modal particle and the causative verb must have some features that attract the verb. The reason is that they are affixes as well, maybe not in the traditional sense, but they are definitely bound items. Perhaps the raising of the verb is triggered by a strong categorial feature [V] in the causative verb, and not the feature [affix]. But in the case of the modal particle it cannot a strong [V] feature, because in negative clauses the participial verb does not raise to the modal particle. Thus, at least for the modal particle, we argue that it is the feature [affix]. We already presented examples of derivations of declarative and negative sentences in (23) and (24), which we repeat here for convenience: (47)

(48)

TP wp T0 (ModP) ! 3 [etorrii (eij)dal]l (Mod0) ! ! tj

VP ! V0 ! ti

CP wp 0 C TP ! 3 ! T0 (ModP) [ezi (eij) dak,l]l ! 3 ! (Mod0) NegP tl ! 3 ! Neg0 VP tj ! ! ! V0 ti ! etorri

76

The raising of negation is independent of the morphological properties of T0; it moves attracted by features in C0. However, as a direct consequence of the incorporation of negation over T0, this head satisfies its morphological deficiency. Thus, the reason why the participial verb does not appear to the left of the inflected auxiliary in negative clauses is because only one head raises to T to satisfy this head’s licensing requirements. The head that fulfills this task will be the one closest to T, which is negation when it is present. Negation suffices to license T morphologically, as it is a free morpheme and can provide strong morphological support to it. Negation is a free word, not an affix or a clitic, as it can stand in isolation in a sentence: (49) Ez ‘No’

Since there is no reason for the participial verb to raise to Aux and Infl when negation is present, V will stay in situ. This explains why the participial verb may appear so distant from the complex head neg+Infl at the left edge of the sentence. It is important to underline that one theoretical consequence of this analysis is that it provides evidence for the feature [Affix] independently of the categorial feature [V]. T has a categorial feature [V], that cannot be strong, because otherwise there would be overt raising of the verb in all instances, not only in the absence of negation. The strong feature in T is the [Affix] feature. This feature requires the overt incorporation of a category, the closest one structurally. When negation is present, it is negation. When there is no negation, it is the participial verb. Modal particles do not provide enough lexical support, as shown in (50) and (51). The participial verb or negation has to move with the modal to license T: (50) a. *ete dau Jonek liburúa erosi dub. has Jon book buy

77

b. Erosi ete dau Jonek liburúa buy dub. has Jon book ‘I wonder whether Jon has bought the book’ (51) a. *ei dau ez dana amaitxu evid. has neg all finish b. Ez ei dau dana amaitxu ‘I have heard that (s)he has not finished everything’

The explanation is that the modal particle is an affix, as we suggested above. That is, it cannot eliminate or satisfy the feature [affix] in T because the modal particle itself is a bound form, not an independent word (i.e., it has the [affix] feature). Thus, the raising of an independent word is required to satisfy both affixal features of the modal particle and of T.

3.4. Possible alternative phrase structures. As we mentioned at the beginning of this section, it is not our goal to provide a full-fledged analysis of the structure of the clause in Basque or of all the consequences deriving from it. Rather, we focus on an analysis which, apart from deriving the correct word order among heads, leads to the conclusion that the inflected auxiliary is a morphologically dependent element that needs to be supported by another head with verbal features. Ortiz de Urbina’s (1994, 1995) analyses argue for this aspect of Basque syntax, and our analysis is thus similar in spirit to his. Independently, Albizu (1994) also pursues a similar approach, with a detailed hierarchical structure of heads which resembles the one suggested here. Indeed, for our present purposes, any alternative theory that posited the same ideas would be compatible with our analysis, however the details and mechanisms of head movement were worked out. There have been proposals that do not consider movement of negation or the verb to the inflected auxiliary, but these

78

analyses present several problems, and in our opinion the direction we pursued in this section is more advantageous. I will briefly review two such proposals in what follows.

3.4.1. Verb-to-Aspect. Laka (1990, 1993) assumes a right-headed structure for Basque, and claims that lexical verbs in this language do not climb higher than to an aspectual head located immediately higher, to their right.xxv Laka claims that lexical verbs are bound morphemes, which need to attach to another head in syntax, the closest available one being the aspectual head in Aspect Phrase, dominating VP. Notice that this is the opposite of the view defended here, that the participial verb is an independent head that supports Infl. As we mentioned in section 2.2.2.1, there are five overt aspectual markers in Basque, -tu, -i, -n, -ko, and -t(z)en. Traditional grammars assign them the following values for the features [±completed], [±future]:xxvi (52)

-tu/-i/-n: [+completed, -future] -ko: [-completed, +future] -t(z)en: [-completed, -future] Perfective aspect is realized by the suffixes -tu/-i/-n, which are lexically selected

by the verbs they attach to. When combined with an auxiliary in the present tense, a verbal form roughly equivalent to the English present perfect is obtained (cf. (53a) below, repeated from (12a)). This form can also be used to convey the meaning of ‘recent past’ (thus being almost identical to the Spanish present perfect). The simple past is derived by the combination of a verb with an aspectual suffix -tu/-i and an auxiliary conjugated in the paradigm for past tense (cf. (53b)). The suffix -ko can be added to verbs with the aspectual suffixes -tu/-i, the auxiliary appearing conjugated in the present tense (cf. (53c)). The suffix -t(z)en indicates habituality; when combined with a present auxiliary a meaning of habitual present is obtained, and when combined with a past auxiliary a

79

meaning of habitual past is derived (cf. (53d,e), respectively). The aspectual morphemes tu/-i/-n and -t(z)en are also referred to as participles in the traditional literature, thus the term participial verb to refer to a lexical verb with aspectual affixes.xxvii (53) a. eros-i dot buy-perf. aux

‘I have bought it’

b. gal-du senduan lose-perf. aux

‘you lost it’

c. eros-i-ko dot buy-perf.-fut. aux

‘I will buy it’

d. eros-ten dot buy-imperf. aux e. eros-ten néban buy-imperf. aux

‘I buy it’ ‘I used to buy it’

Laka follows an incorporation type of analysis of inflectional morphology (cf. Baker 1988), and assumes that the suffixes -tu/-i/-n, -ko and -t(z)en are located in the head of AspP. The raising of the verb to Asp0 would derive the forms erosi, erosiko, erosten and galdu in (53). This raising is sufficient to provide lexical support for the lexical verb. This situation contrasts with that of synthetic verbal forms, which as we mentioned above are amalgamations of the root of a verb and inflectional material, without an auxiliary verb. Synthetic verb forms have non-perfective and non-habitual aspect, that is, they do not appear with any of the morphemes presented in (52)-(53). They have a punctual aspect meaning, with no overt aspect marker. Recall that synthetic and periphrastic verbs show a different behavior with respect to negation. Synthetic verbs always appear right-adjacent to negation, whereas participial verbs do not; it is the inflected auxiliary that appears right-adjacent to negation. This contrast is illustrated by (54a) and (54b), respectively:

80

(54) a. Ez dator iñor batzarréra not comes anybody meeting-all. ‘Nobody is coming to the meeting’ b.

Ez da iñor etorri batzarréra not aux anybody come meeting-all. ‘Nobody has come to the meeting’

On the basis of the behavior of participial verbs, represented by (54b), Laka (1990, 1993) claims that participial verbs do not raise to Infl in overt syntax, and that participial verbs only raise to Asp0. Synthetic verbs, however, do raise to Infl, as evidenced by their morphological composition and by the fact that they appear rightadjacent to negation, where Infl surfaces in periphrastic constructions (i.e., (54b)). Laka explains this contrast by the absence of an overt aspect marker in synthetic verb forms. Since lexical verbs are bound morphemes, the absence of an aspectual head in synthetic constructions forces the verb to raise to the next head in the tree, the overt inflectional heads in I0, so as to acquire lexical support. She assumes a head-final phrase structure for Basque, and thus the occurrence of the participial verb to the left of the auxiliary is explained by having the inflected auxiliary in Infl to the right and higher than the participial verb in Asp0. Taking the minimal pair formed by the periphrastic and synthetic verb forms in (55a,b) as an example, the structures posited by Laka for these constructions can be represented as in (56) and (57), respectively: (55) a. etorri da come aux ‘(s)he has come’ b. da-tor 3abs.sg.-come ‘(s)he is coming’ (56) a.

I’ 3

b.

I’ 3

81

AspP I0 3 ! 0 VP Asp da  y ! 0 V -i ! etor(57) a.

I’ 3 AspP I0 3 ! 0 VP Asp da  y ! 0 V { ! tor-

AspP I0 3 ! 0 Asp da y ! 0 V [etorr-]i -i ! ti

VP

b.

VP V0

I’ 3 AspP I0 3 ! 0 Asp da-[tor]i y ! ti ! ti

This analysis presents a problem, however: if we do not assume that the participial verb incorporates to the auxiliary and then to Infl, we cannot explain why the participial verb appears together with the inflected auxiliary on the left edge of the clause in interrogative sentences and focus constructions, immediately following the element in Spec,CP (cf. section 3.1). We repeat here the illustrative examples introduced in (21): (58) a. Zer esan dótzo Péruk Mirenéri? what-A say aux Peru-E Miren-D ‘What has Peru said to Miren?’ b. Peru etorri da gaur goixian? Peru come aux today morning ‘Was it Peru that came this morning?’ c. Péruk ekarri dau liburúa gaur goixian Peru-E bring aux book-A today morning ‘Peru brought the book this morning’ As we mentioned before, in declarative sentences arguments and adjuncts normally appear to the left of the participial verb and the auxiliary. The structure Laka

82

would have to assume for sentences such as those in (58) would be the following (the actual example we use is (58a)): (59) CP qp Spec C’ ! qp zer IP C0 ri ! ... I’ dótzoj tp AspP I0 3 ! ... Asp’ tj 3 VP Asp0 ty ! 0 ... V esani ! ti

The dotted lines correspond to the material that would precede the participial verb and the auxiliary in declarative sentences, that is, all arguments and adjuncts (in the actual case of (58a), the subject Péruk and the indirect object Mirenéri). The actual position they occupy is irrelevant here; it suffices to assume that they must precede the verb. The participial verb would not raise to I0, as assumed by Laka, and would stay in Asp0. The inflected auxiliary raises to C0, by the principle of Tense c-command, which states that Tense has to c-command all propositional operators in the clause (cf. Laka 1990:9). Since these constructions involve an operator in CP, the head realizing the [tense] feature must raise to C0 in order to c-command this operator. Then, in order to account for the word order of the sentences in (58), in which the material that would otherwise precede the verb in declarative sentences appears to the right of the verb, we would be forced to assume that this material (i.e., all arguments and adjuncts) has extraposed to the right of the position occupied by the auxiliary (I0 or C0

83

depending on the analysis). Thus, these elements must adjoin to the right of IP or CP. Why would such an extraposition take place in operator-involving constructions? Although Laka (1990, 1993) does not address this issue, in earlier work Laka and Uriagereka (1987) and Uriagereka (1987) propose accounts based on the necessity to eliminate barriers to movement and ECP violations. They argue that a maximal projection is a barrier for movement if its specifier is lexically filled, slightly modifying Fukui and Speas’ (1986) original claim. Thus, all the arguments occupying specifier positions that arguments must adjoin to the right of IP so that the projections they are specifiers of do not become barriers for government of the traces left by other arguments moving to Spec,CP. The extraposed arguments do not leave traces in their base-generated positions but pros, with which they are coindexed. The structure for an interrogative sentence such as the one in (60a) would be the one in (60b): (60) a. Zer edaten duzu zuk? what drink aux you ‘What do you drink?’ b. (Zer1 (pro2 t1 edaten duzu ) zuk2 )

Uriagereka (1987) follows Epstein (1987) in assuming that only structurally Casemarked arguments can be lexically governed, and claims that objects in Basque are not assigned Case structurally, but inherently. Thus, the trace left by the object in (60) can only be antecedent-governed. Uriagereka argues that, against what Chomsky (1986a) assumes, IP can be a barrier in languages with rich agreement such as Basque (and Spanish). That is why the subject neds to be extraposed in constructions of object extraction, like (60), so as not to occupy the specifier of IP. According to Laka and Uriagereka (1987) and Uriagereka (1987), this can account for certain asymmetries observed in the extraction of subjects, objects and adjuncts. Sentences in which lexical material intervenes between a wh-object and the verb or verbal complex are

84

ungrammatical (cf. (61a) below), but sentences in which lexical material intervenes between a wh-subject and the verb (or verbal complex) are considered as awkward but better than the former type (always according to Laka and Uriagereka 1987 and Uriagereka 1987). This is due to the fact that there is no maximal projection with a lexical specifier intervening between the wh-subject and its trace (cf. (61b)). Similarly, it is argued that adjuncts are adjoined to IP and that therefore there is no barrier between the moved adjunct and its trace (cf. (61c)): (61) a. * CP [ Zer1 IP [ Jonek VP [ t1 edaten du]]]? what Jon drink aux ‘What does Jon drink?’ b. ?* CP [ Nork1 IP [ t1 VP [ ardoa edaten du]]]? who wine drink aux ‘Who drinks wine? c. ? CP [ Zergatik1 IP [ t1 Jonek VP [ ardoa edaten du]]]? why Jon wine drink aux ‘Why does Jon drink wine?’

This analysis suffers from several problems, both empirical and theoretical, which Uribe-Etxebarria (1989) discusses in detail, so we will not extend ourselves in this section. The first objection is that the difference in grammaticality expressed in (61) is nonexistent, according to the native speakers I have consulted, including myself. If the reason presented for the extraposition of arguments is the necessity to avoid having IP or VP as barriers for government, then sentences such as those in (61b) should be perfectly grammatical. But they are not, even from the judgments collected by Laka and Uriagereka (1987) and Uriagereka (1987). This is unexplained in their analysis. Although one could acknowledge the lesser degree of ungrammaticality that these authors ascribe to sentences with wh-adjuncts, still these sentences are not perfectly grammatical. Moreover, as Uribe-Etxebarria (1989) correctly points out, by Laka and Uriagereka’s (1987) and Uriagereka’s (1987) analysis, sentences in which an adjunct intervened

85

between a wh-object and a verb or verbal complex with a null subject are predicted to be grammatical. This is because the subject is not phonetically realized and thus IP cannot be a barrier for movement. However, such sentences are ungrammatical: (62)

* Nor zineman ikusiko duzu? who cinema see aux ‘Who are you going to see in the movie theater?’

To these, I would add sentences such as (63), in which an adjunct intervenes between a wh-subject and its trace. These sentences are ungrammatical, but under Laka and Uriagereka’s (1987) analysis they should be grammatical, since adjuncts do not occupy the specifier position of any maximal projection, according to their analysis: (63)

*Nork goizean erosi du liburua? who morning buy aux book ‘Who has bought the book in the morning?’

Compare (62) and (63) with the grammatical sentences in (64), with strict adjacency between the wh-elements and the verbal complex: (64) a. Nor ikusiko duzu zineman? ‘Who are you going to see in the movie theater?’ b. Nork erosi du liburua goizean? ‘Who has bought the book in the morning?’

Laka and Uriagereka (1987) would be forced to argue that adjuncts also have to dislocate to the right, in order to render grammaticality out of sentences such as (62) and (63). But if so, then it is unclear what triggers this movement, given that adjuncts do not occupy the specifier position of IP or any other maximal projection, and thus would never constitute barriers for government, i.e., barriers for movement. The number of additional objections to this analysis is quite significant, and it would take us too long to review them here. Due to limits of space, I will not repeat them

86

here from Uribe-Etxebarria (1989), under the belief that the criticisms presented here are sufficient to convince the reader of the disadvantages that this approach has to account for the adjacency observed between a wh-element and a verbal complex. Its problems stem from the assumption that C0 and I0 (i.e., T0) are head-final in Basque. If a movement V-to-I-to-C is assumed for Basque in interrogatives, we must conclude that C0 is more likely initial in Basque. As we mentioned before, Laka (1990, 1993) does not assume V-to-I raising in declarative sentences. Laka’s (1990, 1993) analysis thus suffers from the same problems as Laka and Uriagereka’s (1987), since it posits a head-final CP, but it still faces an additional question: if she were to maintain an analysis in which the verb moved to I0 and then to C0 in interrogative sentences, what would motivate the raising of the participial verb precisely in these sentences, and not in declaratives? The problems posed by this assumption are discussed in the next section, as the main objection against another approach to head movement in Basque that does not assume V-to-I raising in all types of sentences in Basque. Our analysis avoids these problems entirely, since the participial verb always raises to the projection occupied by the inflected auxiliary, in all declarative and interrogative sentences which do not involve negation. As we mentioned above, the only approach that posited a head-final structure for Basque which would be compatible with our proposal would be one which had an exceptional head-initial CP and which assumed participial verb raising to T0 in all matrix clauses in the absence of negation (cf. Ortiz de Urbina’s 1989, 1994, 1995). Our proposal constitutes a step ahead in consistency with the observation that at least one projection must be head-initial and that all head movement is done by left-adjunction.

3.4.2. Long head movement. Albizu (1991) follows Laka (1990) in assuming that the participial verb raises to Asp0 but does not raise to the auxiliary. Then, in order to solve the problems that Laka’s

87

analysis had to face regarding the contrast in word order in declarative and interrogative sentences, Albizu posits a head-initial CP projection. But this amounts to having the participial verb raise to C0 as well. The author argues that this movement is triggered by a wh- or focus operator in Spec,CP. Although the author is not explicit about this, we assume that he refers to the need to fulfill a generalized version of Rizzi’s (1991) whcriterion, that is, the need for a +wh-operator in Spec,CP to be in spec-head agreement at S-structure with an element carrying a wh-feature. This element is a head which must occupy the C0 position at S-structure.

Extending Rizzi’s criterion to all operator-

involving constructions would account for focus constructions. We would need to assume that a wh- or focus operator in Spec,CP requires the presence of a head specified for the same feature (+wh, +focus) in C0. Albizu claims that it is the participial verb that is inflected for these features, and therefore it raises to C0 in operator-involving constructions. But then the problem becomes accounting for the raising of the auxiliary to Comp. This is a point that Albizu leaves open, recognizing the lack of an straightforward motivation for such a movement. This author rejects a head-to-head movement analysis of the participial verb to Aux0 and I0, on the basis of the wrong order among the different heads that would result from such movement, as evidenced by sequences involving modal particles. Albizu locates the modal projection above IP, and thus the predicted order in a head-to-head movement operation originating from the participial verb would have the modal particle in a peripheral position. Assuming left adjunction among heads, to be coherent with the order V-Asp, the derived order would be *V-Asp-Aux-Mod, and not the attested V-Asp-Mod-Aux (e.g., eros-i ei dau). The alternative that Albizu proposes is one in which the participial verb moves to Comp in a long head movement fashion, jumping over the auxiliary and all intervening heads without causing a minimality violation, in a similar way to what has been proposed by Lema and Rivero (1989, 1991) for Old Spanish, and Rivero (1991, 1993, 1994) and Borsley and Rivero (1994) for Slavic and Balkan languages. Then he suggests that a

88

principle similar to the Tense c-command of Laka (1990) motivates the raising of the auxiliary to the modal particle, adjoining to its right. Then, the complex Mod-Aux raises to C0, to the position occupied by the participial verb. This operation would involve right adjunction, to derive the right word order. The representation in (65b) for the sentence in (65a) exemplifies Albizu’s proposal: (65) a. Nok žan ete dau arraiña? who eat dub. aux fish ‘(I wonder) who has eaten the fish?’ b.

CP qp Spec C’ ! qp Noki C ModP 2 ! C Mod Mod’ ! 2 tp žanj Mod Aux/I IP Mod ! ! tp ! etem dauk,l ti I’ tm eu AuxP I 3 ! ti Aux’ tl 3 AspP Aux 3 ! VP Asp tk 3 ! ti V’ tj 2 DP V ! ! arraiña th

This analysis presents two problems: first, the author does not argue convincingly in favor of a motivation for the raising of both the auxiliary and the participial verb in

89

independent movements, as mentioned above. This analysis entails that the operator in Spec,CP is triggering the raising of I0 and V0 to the same head, C0, by the Tense Ccommand Condition and the Operator Criterion, respectively. Invoking two different principles for the same triggering element seems stipulative to us, and lacking a clear explanation. Moreover, the proponents of Spec-head agreement relations between an operator and a head with the relevant features argue that it is the head Infl that carries the features +wh or +focus, not the participial (lexical) verb (cf. Rizzi 1991, Horvath to appear). Thus, such a proposal would seem unorthodox and rather odd at best. Second, in order to derive the right order among heads, the analysis suffers from the fact that it has to assume bidirectionality in the adjunction operations. On the one hand, the lexical verb adjoins to the left of the aspectual head, and on the other, the inflected auxiliary adjoins to the right of the modal, with a subsequent right-adjunction to the participial verb in C0. These problems could be avoided in our analysis. Regarding the first problem, since the participial verb always raises to T0 except in negative sentences, any subsequent movement of T0 would drag the participial verb along with it. We would only need to justify one triggering principle for one head, T0, without having to stipulate anything for V0 raising to C0 only for interrogative clauses or clauses involving other operators. We follow Ortiz de Urbina (1994, 1995) in claiming that an operator criterion motivates the movement of T to C0, contra Albizu (1991), who suggests that it is the participial verb that is attracted. As for the second problem, recall that we are assuming left-adjunction operations in our head-to-head movement analysis. To derive the right order among the different heads, we only have to assume that the modal projection is located below the tense projection and above the verb phrase. The main reason Albizu gives to locate ModP above IP is the existence of data from Standard Basque like the following, involving the modal particles ote and omen (in Lekeitio Basque, ete and ei, respectively):

90

(66) - Nork dauka dirua? who has money ‘Who has the money?’ - Mikelek omen Mikel evid. ‘I have heard that Mikel does (have the money)’ (67) - Loteria tokatu zaizu Lottery touch aux ‘You won the lottery’ - Bai ote? yes dub. ‘Maybe yes? (i.e., I wonder whether that is true)’

Albizu argues that the only way to derive sentences like those exemplified by the responses in these dialogues is through IP ellipsis, that is, by deleting the IP below the modal. However, Euskaltzaindia’s (1985) grammar cites similar examples in which they state that the modal particles can be pronounced after a pause, dislocated. In fact, these particles may sometimes appear as parentheticals, as if they were used in an adverbial manner. Thus, it is not clear that the modal particles in these constructions are in their regular position in the clause. Moreover, and perhaps more importantly, sentences such as those which serve as the response to the first utterance in (66) and (67) are ungrammatical in Lekeitio Basque.xxviii , xxix Notice that these problems arise whether we assume a head-initial or head-final analysis of Basque clause structure. Any other alternative analyses based on long distance movement would also face the same difficulties and disadvantages. An analysis that treated the movement of the participial verb as XP-movement could be proposed, such that the VP moves to the specifier of ModP, located above IP. This would assure the correct word order among heads. Let us illustrate this with the sentence in (68),

91

concentrating first on the derivation and relationships among the verbal and modal elements: (68) a. Jonek liburúa erosi éi dau Jon book buy evid. aux ‘The say that Jon has bought a book’ b.

CP 3 C ModP 3 VP Mod’ ! 3 erosij Mod0 TP ! 3 ei T0 tj ! dau

This analysis would face two problems:xxx first, there is the lack of explanation for the trigger of movement of the VP. The movement is obligatory, in all derivations except in negative sentences, and no answer is apparent for this complementary distribution. Since the movement of the VP is not optional, no focalization or scrambling operations could be responsible here. If the answer were that the VP raises so as to provide lexical support to the inflected auxiliary, we would simply have another version of the analysis we have proposed before.xxxi Negation could then be analyzed either the way we suggested here, i.e., as a head of NegP located below TP and above VP, with head-tohead movement of negation up to T0, or as a head of NegP located above IP. In either case, if the motivation for VP raising is the need to license affixal T, then the absence of VP raising in negative sentences must be taken as an indication that negation (or NegP) would be the licenser of the tensed auxiliary.

92

Second, this analysis would have to assume that in interrogative and focus sentences, in which the verbal complex is in C0, the participial head can be extracted out of the VP and the ModP, to land on C0. This movement would violate the principle against extraction out of a CED (Huang 1982), or the more general principle against extraction out of a specifier (cf. Ormazabal, Uriagereka and Uribe-Etxebarria 1994). A remaining question, both for my analysis and for any alternative analyses which posit verb movement up to T0, TP or ModP, is that this movement would place the verbal complex very high in the clause. Where would the arguments preceding the verb be located? It is claimed that the neutral word order in Basque is S-IO-DO-V. If the verb is in T0, the subject could be in Spec,TP, but where would the objects be? One obvious possibility is that there are several functional projections above VP (TP, AgrSP, AgrOP, or vP) and that the subject and objects land in the specifiers of these projections. This would mean that the verb raises only to the lowest functional projection, and that the inflected auxiliary is realized there as well. The question would then be how to derive the attachment onto the auxiliary of tense and agreement morphemes realizing features pertaining to functional heads located above the verbal complex. One possibility would be affixation under adjacency at MS, as Bobaljik suggests. Any intervening arguments would block affixation. The other possibility would be to assume that the agreement and tense features are present in the verb from the start of the numeration, and that the decision on whether to spell out these features in the position occupied by V or some higher functional projection is a matter of morphological choice of the features themselves or the language. We leave the details of this analysis open for future research. What should remain clear from this discussion is that the participial verb incorporates onto some inflectional heads composing the complex head we labeled T0.xxxii

With

all these considerations in mind, it seems that any analysis that denies the possibility of V-to-T raising in Basque has to face numerous problems and difficulties which an analysis that argues for such movement easily avoids.

93

To summarize, in this section I have discussed and presented the syntactic configuration that I assume for Basque clause structure. Building upon ideas of Ortiz de Urbina’s (1994, 1995), we have suggested that T has a strong feature [Affix] that needs to be syntactically and morphologically licensed by the overt movement of a morphologically independent syntactic head. The participial verb has to raise to T0 in order to license it. In negative clauses, it is negation that raises to T0. Modal particles do not satisfy the affixal feature of T because it is an affixal head itself. As we saw in section 2, VA also occurs between a noun or adjective and a following determiner or case marker, that is, between a nominal head and an inflectional head. In the next section we turn our attention to the relationship between nominal heads and determiners, to see whether we can draw a parallelism between verbal and nominal contexts, or more concretely, between verbal and nominal inflectional heads in Basque.

3.5. Determiners. The order between nouns, adjectives and determiners is noun-adjectivedeterminer, as already presented in section 2: (69)

NP [gixon

argal] D [-a] man thin det.sg. ‘the thin man’

In a head-final approach to Basque clause structure, there would be no way to tell whether there has been movement of the noun or adjective to the determiner. It allows an analysis in which no syntactic movement to the DP projection has taken place, the noun and adjective remaining in their base position as complements of the determiner: (70)

DP 3 NP D0

94

6 gixon argal

! -a

Adopting the now standard DP hypothesis (Szabolcsi 1987, Abney 1987), it is apparent that the determiner is the last element in its phrase, the Determiner Phrase. On the other hand, in a head-initial approach to Basque phrase structure such as the one we are proposing here, we must assume that the elements in the NP dominated by the determiner have moved upward to the determiner projection. In principle there are two possibilities for this movement to be realized: head-movement of the head noun or the adjective to the D0 position, or XP-movement to Spec,DP. Given the fact that the determiner always appears attached to the last element in the NP, instead of to the head noun, we conclude that XP-movement to the specifier position of the determiner must have occurred, and not head movement of the noun or any other head contained in the NP. If the noun had incorporated to the determiner, for instance, the following ungrammatical output would have obtained: (71)

*gixon-a argal

A stronger piece of evidence showing that XP-movement and not head movement is responsible for the order within the DP is the fact that determiners can occur attached to the last element in a relative clause: (72)

DP [ NP [ NP [ CP [gaur

goixian etorri dan] gixon ]] [-a] today morning come aux man det.sg. ‘The man that came this morning’

The possibility for a determiner to attach to an inflected auxiliary ending a relative clause constitutes clear evidence against the traditional approach in Basque linguistics that identifies the determiner as a suffix, attached to a stem in the lexicon. Affixes typically select the specific category of the stem they attach to (cf. Zwicky and

95

Pullum 1983), which in the case of a determiner is a noun or adjective, not an auxiliary verb, i.e., determiners and case-markers attach to nominal stems, not verbal stems. If a determiner is a separate syntactic D0 head which always selects an NP, however, the attachment of a determiner to nominal stems and auxiliary verbs is not a puzzle, since the determiner is a bound morpheme that needs to attach to the last word in the NP its selects. In the case of relative clauses with an empty noun, the determiner attaches to the last phonetically realized word, the inflected auxiliary. Within the assumption of uniform head-initial structure for Basque, the relative clause must have raised to the left of the noun (cf. Kayne 1994), and then the whole NP moves to the specifier position of DP, plausibly to check the NP-features of the determiner. The diagram in (73) represents the derivation giving rise to the surface order: (73)

DP qp NPi D’ qp 3 [gaur goixian etorri dan-gixon]i D0 ti : ! ! ! -a ! z---------------m

And in the simpler case of a noun or adjective to which the determiner is suffixed, such as arrebi-a ‘the sister’: (74)

DP 3 NPi D’ ! 2 arrebii D0 ti : ! ! ! -a ! z-------m

96

The raising of the noun phrase to the specifier of DP could be seen as a checking operation of the strong categorial feature [N] present in the head of the DP functional projection, i.e., D0, by spec-head agreement in the functional head’s checking domain. In the minimalist framework, the only possibility that is considered for checking the categorial feature [N] is by attracting the formal features of N to D0. I want to suggest that NP-raising to Spec,DP is also a legitimate operation that satisfies the checking of the strong [N] feature in D. After all, it is also the head N0 that moves with all its formal features including the categorial [N] feature. Whether assuming the existence of this parametric option in checking the [N] feature or not is a matter that merits further investigation. It could well be that NP-movement to Spec,DP is triggered by other factors. I leave the matter open without further discussion.xxxiii We established in the previous section that the [affix] feature of T triggers raising of a head which is an independent word, to lexicalize and license it. We propose that the determiner has a feature [affix] as well, and that it is checked by attachment to the last word in the NP in Spec,DP. The difference with respect to T is that the affixal feature is not strong, i.e., it does not trigger syntactic movement before Spell-Out, but at MS. The NP might check the features in D0, but in addition, the Basque determiner is a bound morpheme which needs to be attached to a syntactic element. It attaches as a suffix to the last element of the phrase that precedes it, in a similar fashion to phrasal clitics, of the type discussed by Nevis (1985), Poser (1985), Zwicky (1987), and Miller (1992).xxxiv The suffixation or cliticization process takes place in the PF component, after the operation of Spell-Out has sent the derivation to the interface levels of PF and LF. Perhaps the level where this process takes place could be the level of Morphological Structure, proposed by Bonet (1991), Noyer (1992), Halle and Marantz (1993), as a component of grammar where morphological operations of merger, fusion and fission affecting adjacent morphological categories take place. This level is located between the syntactic and phonological components, after the derivation is sent to PF. This phonological

97

cliticization is carried under adjacency between the elements involved, much like the possessive marker ‘s in English cliticizes to the last word of a possessor in the specifier position of DP. Marantz (1992) argues that this process takes place after vocabulary insertion, and is characteristic of clitic morphs, as discussed in Marantz (1988, 1989), Halpern (1992), and references mentioned above. Notice that this is not a process of morphological merger in the sense of distributed morphology, since the elements involved are not morphemes or syntactic heads (i.e., a determiner and the last word in the phrase located in its specifier). The diagrams in (75)-(76) illustrate this process, corresponding to the pre-SpellOut structures (73) and (74), respectively. The arrows linking the determiner to the last word in the NP in Spec,DP indicates cliticization, not syntactic movement: (75)

DP qp NP D’ qp 2 [gaur goixian etorri dan-gixon]i-aj D0 ti : ! ! tj z--m

(76)

DP 3 NP D’ ! 2 arrebii-aj D0 ti : ! ! tj z--m

We have provided a solid reliable demonstration of the structure of the clause and the determiner phrase in Basque. We are ready now to present our analysis of VA in

98

Lekeitio Basque, which crucially depends on the ideas expressed in this section on the morphological deficiency of both types of inflectional heads in Basque: T and D.

4. An analysis in terms of feature chains. What does all we have argued for so far tell us about the domain of application of VA? Note that it is not sufficient to say that the domain of application of VA is an X0, formed in syntax or phonology, because although that eliminates sequences of a noun and an adjective, a subject and a verb, a subject and a direct object, and so on, it does not account for the absence of VA between a participial verb and a modal particle, which end up in an X0 in overt syntax. Nor can it explain why there is no VA between a participial verb and a causative verb, which also end up in the same X0 in syntax. In (77) below we illustrate a schematic derivation involving movement of a participial verb onto a causative verb which selects it: xxxv (77) a. Jonek Mireneri etorri eraiñ dotzo Jon-E Miren-D come caus aux ‘Jon has made Miren come’ b.

CP qp C0 TP qp 3 [etorrii eraiñj dotzok] k T0 VPcaus ! 3 tk V0caus VP ! ! tj V0 ! ti

These are the structures that result after participial verb movement to T, a modal particle and a causative: (78) a.

T0

b.

Mod0

c.

V0caus

99

3 V0 T0

3 V0 Mod0

3 V0 V0caus

This shows that an analysis in terms of c-command relationships and/or branching configurations that the trigger and the target enter in will not work, because the ccommand relationships and branching configurations holding between the participial verb and the element to its right are identical. Then, what is the difference between (78a), which is a potential context for VA, and (78b,c), which are not? We argue that the difference lies in the nature of the feature chains involved in each case. The chain T-V is one of the primitive chains in grammar, as explained in Chapter 1. It expresses the fact that T attracts the set of formal features of V, to check the categorial feature [V]. It also expresses the fact that V also has its tense features checked in T. This is a close syntactic relationship of a symbiotic nature (to use a metaphor we introduced in Chapter 1) that is reflected morphosyntactically in the form of a domain or constituent that includes the heads realizing those features, i.e., T and V. The strong [affix] feature of T attracts the category V, making the necessary linear adjacency possible. Thus, the syntactic configuration in (78a) gets mapped at the level of MS as a morphosyntactic domain: (79)

T0 3 V0 T0



[V-T]ms-word

The grammatical relationship established by the feature chain in (78a) is different from the one in (78b,c). In the latter, there is no feature chain relationship, only a satisfaction of the feature [affix] in the modal particle and the causative verb. The causative verb may not be an affix and it attracts the verb because of a strong [V] categorial feature. In any case, the relationship with the participial verb is not one of the primitive chains of grammar, presented in Chapter 1 and repeated below:

100

(80)

C-T

(81)

T-V

(82)

T-D

(83)

V - D (Subj)

(84)

V - D (Obj)

(85)

D-N

(86)

P-D

Thus, although the syntactic configuration is the same in the three structures in (78), and in the three cases the head that the participial verb moves to has the feature [affix], there is a distinction in degree of feature checking interaction. The verb does not check tense features or φ-features in the modal particle or the causative verb. These heads do not possess features that the verb also possesses and has to check, unlike the situation with T-V (just explained), or C-T, T-D, V-D, D-N, P-D, in which both members of the chain check features that are present in the other head.xxxvi The syntactic configurations in (78b,c), then, do not get mapped as morphosyntactic domains. As for subordinating conjunctions, we could propose that they occupy the Comp node in subordinate clauses, taking nonfinite clauses as complements. Since in Basque the subordinate clause precedes the element in Comp, the syntactic theory we have been assuming here would lead us to posit a leftward movement of the subordinate clause (TP) to the specifier position of the complementizer, i.e., Spec,CP. In (80b) we provide an illustration of this type of movement, corresponding to the subordinate clause in (80a): (80) a. Diru asko euki árren, astopótro bat da Péru. money much have despite brute one is Peru ‘Despite having a lot of money, Peru is a brute’ b.

CP

101

qp TPi C’ 6 3 diru asko euki C0 ti : ! ! ! árren ! z---------------m

This configuration would be similar to the one involved in postpositions: (81) a.

b.

negua amaitxu árte winter finish until ‘until the winter ends’ PP qp VPi P’ 6 3 negua amaitxu P0 ti : ! ! ! árte ! z---------------m

If one were to propose that in (80) the feature chain C-T is involved, one possible explanation is that the complementizers ezik and árren are realized as independent words. This is a possibility already discussed in Chapter 1. The same could be suggested about the postposition árren. This hypothesis permits us to establish a difference between the sequences formed by participial verbs and tense (i.e., inflected auxiliary) and those formed by participial verbs and modal particles or causatives. On the nominal side, we have the feature chain D-N, one of the primitive chains, which expresses the raising of formal features of N to D or Spec,DP. The categorial feature [N] that checks the same feature in D, the φ-features of N check the φ-features in D, and possibly the referential features of N and D get

102

checked as well). At MS, the determiner is suffixed to the rightmost word in an NP (i.e., a noun or an adjective). The structure so formed is thus a morphological word at MS: (82)

N/A-D]



[N/A-D]m

We assume the existence of the level of Morphological Structure (MS), the morphological component of grammar located after Spell-Out, that is, after all overt syntactic operations have taken place and the linguistic derivation has been sent to PF. This is the position adopted by Bonet (1991), Noyer (1992) and Halle and Marantz (1993). For these authors, MS is an interface level between syntax and phonology. In fact, is a syntactic representation that nevertheless serves as part of the phonology, where ‘phonology’ is understood as the interpretive component that realizes linguistic derivations phonologically. The organization of the grammar adopted is the following: (83) DS (D-structure) ! SS (S-structure) 3 (Morphological structure) MS LF (Logical Form) ! (Phonetic Form) PF

In this theory, affixes and stems are entries in the Vocabulary of a language, composed of two sets of features: phonological features, on the one hand, and morphosyntactic features, on the other. The terminal nodes in syntax are bundles of morphosyntactic features, hierarchically organized in the case of complex X0s, and it is only by an operation of Vocabulary Insertion that these bundles of morphosyntactic features receive phonological content. Vocabulary Insertion takes place at MS, and in order for Vocabulary Insertion to be successful, the morphosyntactic features of a vocabulary item must match or be non-distinct from the morphosyntactic features present in the terminal nodes of S-structure. Thus, the vocabulary entries inserted by this

103

operation connect morphosyntactic feature bundles of syntactic structure with phonological feature complexes. This is where the actual spell-out of morphemes takes place. The phonologically realized string of morphemes and words would then pass to the actual level of PF, where the phonological processes of the language would modify those parts of the string which meet their structural description.

4.1. Mapping between morphological domains and phonological domains The interest of this proposal for our purposes is that the morphosyntactic units containing the feature chain T-V visible at MS may be mapped as a phonological domain to PF, the component of grammar where phonological processes apply. We assume the validity of the notion of domain in phonology, and argue that there might be phonological processes that are specified to apply in such a unit, the morphosyntactic word (or msword, for short). This idea will provide us with a way to explain the phonological behavior of certain morphemes or bundles of morphemes realizing morphosyntactic features that cannot be captured in the theories of prosodic phonology that have been proposed (see section 5). In these theories no distinction is made within the class of functional categories that are unarguably separate syntactic heads (the term used in the literature to refer to these expressions is function words). Those that behave as dependent phonologically or which interact closely with adjacent lexical heads triggering or undergoing phonological processes are simply stated to be defective categories from the prosodic point of view. They do not constitute prosodic words by themselves, and are incorporated into adjacent prosodic words, following a directionality parameter (cf. Nespor and Vogel 1982, 1986, Hayes 1989, Selkirk 1986, 1996). Some models argue that those dependent function words are prosodic words, and form clitic groups with adjacent lexical heads (cf. Nespor and Vogel 1986, Hayes 1989). Other models do not conceive dependent function words as prosodic words, and propose that they are directly integrated into adjacent prosodic words (cf. Selkirk 1984, 1986, 1996). The most radical of these

104

approaches is probably stated in the Principle of the Categorial Invisibility of Function Words, suggested by Selkirk (1984), which claims that function words are not real words and are not identifiable as constituents of any sort in syntax or phonology. We

will

provide a more detailed overview of these theories in section 5, when we demonstrate the difficulties that these theories face in order to capture the domain of application of VA in Lekeitio Basque. At this point, suffice it to say that any theory which does not allow room for a distinction among functional categories is not able to explain the behavior of functional heads containing formal features of the inflectional type. As we have seen, they present the inherent property that they need to be licensed syntactically, a property that sets them apart from other function words which could fall under the ‘derivational’ class, such as adverbial particles, adpositions, or conjunctions. Our idea of morphological licensing is needed because inflectional heads may form separate prosodic domains for stress assignment (as in the case of the Basque auxiliary) but still interact phonologically with adjacent lexical heads in ways that other functional categories do not. That is, it will be shown that morphological domains may be formed after syntax and that they may be a source of phonological domains, independent of the prosodic domain formation algorithms that have been shown to function successfully. Our argument is thus that the operations responsible for the creation of phonological domains are not reduceable to prosodic algorithms solely, but can stem from morphological domains created on the basis of morphosyntactic licensing. In other words, the licensing principle explains why phonological rules operating across words (syntactic terminals) often apply to elements with a high degree of morphosyntactic cohesion.

In (84) we express this mapping from m-constituency to p-

constituency:xxxvii (84)

[

]ms



[

]p

105

4.2. Empirical analysis of VA. In (85) and (86) we repeat the formalization of the process of morphosyntactic mapping we proposed in (79) and (82), respectively: (85)

(86)

T0 3 V0 T0



[V-T]ms-word

N/A-D]



[N/A-D]m

In (87) and (88) we provide examples of the instantiation of (84) in nominal and verbal contexts, respectively: (87) a. b.

(88) a. b.

[ umi ]NP -a] child det-sg

-->

[ umi -a ]m ‘the child’

[kale estu ]NP -étan] --> street narrow det.pl-ines.

[kale]m [estuétan]m ‘in the narrow streets’

[ ekarri ] [ eban ] bring aux

-->

[ ekarri eban ]m ‘(s)he brought it’

[ atrapa ] [ ebasan ] catch aux

-->

[ atrapa ebasan ]m ‘(s)he caught them’

If our working hypothesis is on the right track, we grasp the nature of the contexts of application of VA. It applies in the morphological constituents that realize at MS the feature chains D-N and T-V. In (89) we express this mapping from m-constituency to pconstituency, and the domain of application of VA as the phonological component would interpret it. The vowel beginning a deficient element being licensed assimilates to the left-adjacent vowel ending the licensing head:

106

(89) a. b.

[

]m



[

]p

Domain of VA:

[

]mp

(90) and (91) illustrate the optional application of VA on the examples in (87)(88), after the mapping from m-constituency to p-constituency is done: (90) a. b.

[ umi -a ]mp



umia / umii

[ estu -étan]mp



estuétan / estuútan

(91) a.

[ ekarri eban ]mp 

ekarri eban / ekarri iban

b.

[ atrapa ebasan ]mp



atrapa ebasan / atrapa abasan

We suggested that inflected auxiliaries and synthetic verbs shared the property of being morphologically deficient. Ortiz de Urbina (1994) argues that the verbal root raising to I0 in synthetic constructions is sufficient to provide lexical support to Infl. If so, then we would conclude that a synthetic verb forms an m-word after verb raising. This would explain that the initial vowel of a synthetic verb does not assimilate to the final vowel of a preceding word: (92) a. dana ekižan / *dana akižan all knew-1erg.sg.-3abs.sg. ‘(s)he knew everything’ b. bera etorren / (s)he came-3abs.sg. ‘(s)he was coming’

*bera atorren

However, synthetic verbs with an initial /d/ or /b/ assimilate in voicing to a preceding consonant; they become voiceless after a voiceless consonant ending a

107

preceding word. This process does not apply to other d-initial words, including participial verbs: (93) a.

ez daukozu / ez taukozuxxxviii neg have-2erg.sg.-3abs.sg. ‘You don’t have it’

b. lagunak badáki / lagunak padáki friend if-knows-3erg.sg.-3abs.sg. ‘If the friend knows it’ c. nok dákar žatekúa? who brings-3erg.sg.-3abs.sg. food ‘Who is bringing food?’

/

*nok tákar žatekúa?

(94) a. lagunak dirua ekarri dau / *lagunak tirua ekarri dau friend money bring aux ‘The friend has brought the money’ b. umiak bajatu dau child lower aux ‘The child has lowered it’

/

*umiak pajatu dau

The phenomenon of devoicing could be viewed as a process that suggests some sort of phonological “leaning” or dependency on the preceding element, perhaps indicating a remaining deficiency. But I do not believe this fact warrants positing the creation of an m-domain between the synthetic verb and the preceding word. In our discussion in section 3 we presented Ortiz de Urbina’s (1994) analysis that in sentences involving operators, the head Infl in C0 requires the presence of a maximal projection in Spec,CP or a head in C0. In northern and eastern dialects, this is achieved by the presence of a wh-word, a focalized constituent, or negation. In the standard and southern dialects, however, an additional requirement exists: Infl needs to be lexicalized by receiving the incorporation of a lexical head. That is why the participial verb has to raise. This lexicalization requirement holds for all constructions, not just those involving operators,

108

i.e., the verb raises irrespective of the presence or the absence of an operator in Spec,CP. Its presence in C0 is only a side effect of its incorporation onto Infl first and the subsequent attraction of Infl to C0 triggered by the operator in Spec,CP. This is represented in the diagram in (95) below, repeated from (36). The parenthesis surrounding the participle indicates that its presence is parametrized in Basque: (95)

CP 3 XP C’ 3 C IP ! (Participle+) Infl Neg+Infl Ba+Infl

For us, the raising of a participial verb, negation or the root of a verb is sufficient to lexicalize T, that is, to check the [affix] feature. Thus, I have no principled answer for consonant devoicing in synthetic verbs (except for the case in which negation incorporates onto the synthetic verb on its way to C0 or G0). Notice also that the process affects synthetic verbs even in declarative sentences, in which presumably no raising of Infl to C0 occurs and hence no condition such as (95) applies. This argues against a possible analysis of the synthetic verbs forming onemorphological word with the words preceding them: (96) lagunak liburúak daukoz / taukoz friend books has-3erg.sg-3abs.sg ‘The friend has books’

Also, if this were a productive rule applying in an m-word, like VA, we would expect consonant devoicing to be as readily available as VA. However, it is not. Only in the context of negation is the process almost obligatory. In the other cases, it is optional,

109

and according to the native judgments I have recorded, it is harder to have devoicing in those contexts than it is to make VA in the contexts in which we have seen it apply. Moreover, devoicing only affects /d/ and /b/-initial verbs, not /g/-initial verbs: (97) a. ez gara gaixtúak neg. are-1abs.pl. evil ‘We are not evil’

/ *ez kara gaixtúak

b. nórtzuk gágoz barruan? who are-1abs.pl. inside ‘Who (of us) are inside?’

/

*nórtzuk kágoz barruan?

In the absence of a better explanation, I will simply leave the matter of devoicing as a property of synthetic verbs with initial coronal and labial voiced consonants. Returning to our main point, with our analysis we can explain the fact that VA does not occur between two lexical categories: because they are not part of the same mdomain, but of separate ones, and thus they fall in different p-domains. This would include causative verbs as well: (98) a.

b. dau

[lora]mp [ederra]mp flower beautiful ‘beautiful flower’



lora ederra / *lora adarra

[ egurra]mp [ekarri dau]mp 

egurra ekarri dau / *egurra akarri

wood bring aux ‘(s)he has brought wood’ c.

[salta]mp [eraiñ]mp  jump make ‘make jump’

salta eraiñ / *salta araiñ

Notice that this analysis predicts that once an auxiliary is licensed morphologically (i.e., becomes part of an m-word), it is predicted that it will not interact with a following word phonologically. Thus, no VA occurs between a vowel-final auxiliary and a following vowel-initial word:

110

(99) a. amaitxu dogu afariža finish aux dinner ‘We have finished dinner’ b. Morphosyntactic constituency:

[amaitxu dogu]m [afariža]m

c. Phonological constituency:

[amaitxu dogu]p [afariža]p

d. amaitxu dogu afariža /

*amaitxu dogu ufariža

In negative sentences, T forms an m-word with negation, and the participial verb stays behind. Several material could separate the auxiliary from the participial verb (cf. section 3), but the inflected auxiliary and the verb may be linearly adjacent as well. Even in such cases, no VA occurs between a vowel-final auxiliary and the verb, because it is not the verb that forms an m-domain with Infl, but negation: (100) a.

ez dozu ekarri liburua gaur goixian neg aux bring book today morning ‘You didn’t bring the book today’

b.

[ez dozu]m [ekarri]m ... neg aux bring

c.

[ez dozu]p [ekarri]p



ez dozu ekarri / *ez dozu ukarri

As for the absence of VA between members of compounds, we would assume that each member is an independent word, which combines with the other to form another word: (101) a. [ [ buru ]mp - [ andi ]mp ]mp head big ‘big-headed’ b. [ [etxe]mp - [ondo]mp ]mp house side ‘house side’



buru-ándi / *buru-úndi



etxe-óndo / *etxe-éndo

111

Although the two members of a compound form an m-constituent, we assume here that the internal brackets and mp domains corresponding to each member are visible in the phonological component. This ensures the proper blocking of VA. That two members of a compound form separate phonological domains has been argued for other languages such as Sanskrit, Turkish, Hungarian, Italian, Dutch, and Malayalam (cf. Nespor and Vogel 1986:ch.3, Inkelas 1990).xxxix Modal particles, conjunctions and postpositions are separate words as well: (102)

[ekarri]mp [ete]mp bring dubit.

(103) a.

[ erosi ]mp [ árren ]mp  buy despite ‘despite buying’

erosi árren / *erosi írren

b.

[ leidu ]mp [ ezik ]mp  read unless ‘unless (someone) reads’

leidu ezik / *leidu uzik

c.

[ ikasi ]mp [ árte ]mp  learn until ‘until learning’

ikasi árte / *ikasi írte



ekarri ete

*ekarri ite

In a sequence formed by a participial verb, a modal particle and an inflected auxiliary, no VA can be observed between a modal particle and the auxiliary, because the right phonological context is absent. The dubitative modal particle ete ends in /e/, the same vowel that begins vowel-initial auxiliaries. The evidential modal particle ei does not trigger VA but it triggers /ž/-insertion on the auxiliary: (104) apurtu ei eban  break evid. aux ‘I have heard that (s)he broke it’

apurtu ei žéban

112

This process is found in nominal contexts, between a noun or adjective ending in an underlying /i/ and a determiner or case marker. The voiced prepalatal fricative is syllabified as an onset of the initial vowel of the determiner or case marker. Underlying mid front vowels which become high by the rule of Vowel Raising do not trigger this rule. Cf. the examples in (3) above, which we repeat in (105): (105) a. /barri-a/  new-det.sg. b. /barre-a/  laughter-det.sg.

ba.rri.ža

‘new, the new one’

ba.rri.a ‘the laughter’

This process is only optional between a participial verb and an auxiliary: (106) a. ekarri eban bring aux



ekarri eban / ekarri iban / ekarri žeban

b. irabazi ében win aux



irabazi ében / irabazi íben / irabazi žében

If the verb ends in a diphthong, however, a palatalization rule is obligatory, either fortition of the glide to a palatal fricative /j/ or insertion of /ž/, keeping the glide: (107) egin eban do aux



eyeban / ei žebanxl

No process of this kind occurs between a verb and a modal particle, a causative verb, a subordinating conjunction or a postposition: (108) a. egin ete dau  eiñ ete dau do dubit. aux b. egin erain dotzo do caus. aux c. egin árte do until



/ *eyete dau / *ei žete dau

 eiñ erain dotzo / *eyerain dotzo / *ei žerain dotzo eiñ árte / *eyárte / *ei žerain dotzo

113

Palatalization could be analyzed as a process which occurs in a phonological domain directly mapped from an m-domain. That is why we may find it between a participial verb and an inflected auxiliary, but not between a participial verb and a modal particle, a causative verb, a subordinating conjunction or a postposition. Since these elements constitute separate m-words, according to our theory, palatalization is correctly predicted not to occur. The participial verb raises to the modal particle and then the whole complex head incorporates onto T, an operation that checks the [affix] feature of both the modal and T. We can assume that the features of the verb as an independent head percolate and are visible to the whole complex, thus ensuring that T gets licensed. The modal is a separate m-word from the verb, and thus palatalization between the modal particle ei and an auxiliary is predicted to occur. It is a matter of linear adjacency that makes the process of palatalization apply between the modal particle and the auxiliary:xli (109) [[ekarri]m [ei]m ]m eban



[[ekarri]m [ei]m eban]m]m

We may ask ourselves what the behavior of demonstratives is with respect to VA, since they are all vowel-initial: hau ‘this’, hori ‘that’, ha ‘that one over there’; hónek ‘these’, hórrek ‘those’, hárek ‘those over there’ (as in all southern dialects, the h is not a pronounced consonant in LB). Unlike Standard Basque, demonstratives in LB precede the NP, except for hau and hori, which can appear following the last element in the NP, like determiners. In fact, like determiners, they trigger the rule of VR, by which the final nonhigh vowel of a stem becomes high, when followed by a vowel-initial suffix (cf. (66), and compare it with the examples in (1), in section 1.1.1). The prediction would thus be that demonstratives display the same behavior as determiners with respect to VA. Nevertheless, VA does not apply to demonstratives in LB: (110) a. /etxe-au/  house-dem. ‘this house’

etxiau / *etxiiu or *etxiii

114

b. /etxe-ori/  house-dem. ‘that house’

etxiori / *etxiiri

The explanation for the contrast with determiners is phonological in nature: these forms begin with a diphthong and a round vowel, respectively, which are not affected by assimilation. Determiners beginning with a round vowel do not undergo VA, for instance. Consider here the plural proximative determiner -ok, which unlike the plural determiner ak does not undergo VA: (111) a. /umí-ak/  umíak / umíik child-det.pl ‘the children’ b.

/umí-ok/  umíok / *umíik child-del.pl.prox. ‘the children (proximative)’

As mentioned in note 13, there are forms of the inflected auxiliary with a direct object and an indirect object and a third person ergative morpheme which present an initial diphthong eu-: (112) /ipiñi e-u-tz-an/ put 3erg.--pres-rt.-3dat.sg.-past ‘(s)he put it on him/her’



ipiñi eutzan

These are the only vowel-initial forms of inflected auxiliaries that resist assimilation of one or both of its members (*ipiñi iutzan, *ipiñi iitzan). This indicates that there is a phonological constraint that does not allow VA to apply to complex vocalic sequences. Only short vowels can undergo the rule of VA. In fact, there is evidence from other dialects of Basque that shows that complex syllable nuclei are resistant to vowel assimilation. In Arbizu, for instance, long vowels

115

fail to undergo the partial assimilation rule that raises a low vowel and turns it into a mid vowel (examples from Hualde 1988, chapter 2, section 3.1): (113) a. /mendi-ak/ mountain-det.pl. ‘the mountains’



b. /mendi-aa-n/  mountain-det.sg.-ines. ‘in the mountain’

mendijek

mendijaan

*mendijean or *mendijeen

Hualde (1988) analyzed this contrast as due to the inalterability of geminates (cf. Hayes 1986, Schein and Steriade 1986). Our proposal would then be that diphthongs in Basque are like geminate vowels in that they are complex nuclei, and as complex nuclei they are resistant to assimilation.

4.2. Supporting evidence: /n/-deletion. Another piece of evidence showing that there is a substantial difference between auxiliaries and modal particles is the phenomenon of /n/-deletion. This is a process in which one of the perfective aspect markers that can be added to a verb, -n, gets deleted in the presence of a following vowel-initial auxiliary. This rule of /n/-deletion provides the context for VA: (114) a. erun eban  take aux ‘(s)he took it’ b. žan ebasan  eat aux ‘(s)he ate them’

eru eban

/

eru uban

ža ebasan / ža abasan

116

Modal particles following a participial verb do not trigger such a rule. Thus, the outputs in (115), in which the final -n of the participial verb is dropped before a modal particle, are not acceptable: (115) a.

erun ete dábe?  *eru ete dábe take dub. aux ‘(I wonder whether) they have taken it’

b. žan éi dau  *ža éi dau eat evid. aux ‘I have heard that (s)he has eaten it’

In fact, no other element apart form an auxiliary triggers /n/-deletion. The final -n in a participial verb cannot delete before the causative verb eraiñ, the subordinating conjunctions árren and ezik, or the postposition árte: (116) a. emon eraiñ nazu give make aux ‘you made me give it’



*emo eraiñ nazu

b. žatekúa artun árren food take despite ‘despite taking the food’



*... artu árren

c. geu žun ezik we go unless ‘unless we go’

*... žu ezik



d. zezena sartun árte bull go in until ‘until the bull goes in’



*... sartu árte

There is an exact correlation between the contexts of application of VA and the contexts of application of /n/-deletion in verbal contexts. This correlation supports the analysis we have proposed in this section, since we could also formalize the context of application of /n/-deletion in the same terms as for VA:

117

(117) -n  {

/

[___ α

β

V ]mp

This rule deletes the perfective morpheme -n of a participial verb when followed by a vowel-initial auxiliary within the same m-constituent. The condition of being included in the same m-constituent is what distinguishes auxiliaries from modal particles, causative verbs and subordinating conjunctions and postpositions. We could not treat this process simply as a result of the incorporation of the participial verb to the auxiliary, that is, as a phonological effect of a cliticization process, since the participial verb raises in the same manner to a modal particle and a causative verb. The absence of VA and ndeletion between lexical verbs and other adjacent material apart from a finite auxiliary can be explained by the absence of a licensing relationship between the verb and these elements, that is, the degree of cohesion between participles and auxiliaries is closer than the one between participles and other elements, even within the same syntactic X0. This analysis demonstrates the significance that morphosyntactic relationships among syntactic heads has for the theory of prosodic phonology and phonological domains. The necessity of positing a source of phonological domains based on morphosyntactic considerations will become even more apparent when we review other theories of prosodic phonology and see the problems they face to capture the domain of application of VA in Lekeitio Basque. The fact that the inflected auxiliary may bear its own underlying accent, and that it may constitute a separate domain of stress assignment on its own are puzzles that call for an alternative to the idea that all phonological domains are prosodically determined. We are now ready to review critically these theories.

5. A puzzle for theories of phrasal phonology. The rule of VA presents a problem for its classification as a lexical or postlexical rule, following the assumptions of classical lexical phonology (cf. Pesetsky 1979, Kiparsky 1982, 1985, Mohanan 1982, 1986, Archangeli 1985, Kaisse and Shaw 1985,

118

Pulleyblank 1986, among others). We have seen that it applies in nonderived environments in some words, lexically specified to undergo the rule (cf. (7), (15) above), and in nominal and verbal inflection. VA cannot be simply lexical, since it applies across words, i.e., between a verb and its inflection, and it cannot be classified as postlexical rule in the classical sense either, since, contrary to what has been claimed in traditional Lexical Phonology, VA does not apply across-the-board, within and across words. It only applies in the syntactic context of a lexical element and its inflection. Moreover, as already mentioned, it may even have lexical exceptions (cf. (7)), and this is a property which is recognized for lexical rules, not postlexical rules.

5.1. The P1/P2 distinction among postlexical rules. The existence of postlexical rules which do not apply across the board but in selected syntactic contexts has been noticed over the past twenty yesrs or so. Ellen Kaisse established a distinction between P1 and P2 postlexical rules (cf. Kaisse 1985, 1990). P1 rules are the rules of connected speech or of external sandhi, which show sensitivity to morphosyntactic information, and P2 rules are the rules of fast speech for which morphosyntactic representations are not available, and are only sensitive to rate of speech. P1 rules have their context of application defined locally, on the basis of configurational relations such as c-command or conditions on edge location (cf. Rotenberg 1978, Manzini 1983, and Lobeck and Kaisse 1984, for early proposals along these lines). It is argued that these relations define parameters whose settings characterize unmarked rules of external sandhi. There are two parameters, the c-command condition and the edge-condition: (118) 1. The c-command condition: One of the words must c-command the other.xlii 2. The edge condition: The sandhi pair (i.e., the words participating in the phonological rule) must lie on the edge of the constituent that contains them.

119

The parameters may be set as follows: (119) 1a. 1b. 1c. 1d. 2a. 2b. 2c. must 2d. 2e.

Word a must c-command word b. Word b must c-command word a. Words a and b must c-command each other (government required). There is no c-command requirement. Words a and b must lie at the left edge of their containing constituent. Words a and b must lie at the right edge of their containing constituent. Words a and b must lie at both edges of their containing constituent (they exhaust that constituent). There is no edge requirement. Words a and b must lie at one edge of their containing constituent.

An illustrative example of a P1 rule according to Kaisse’s typology is lenition in Gilyak (cf. Kenstowicz and Kisseberth 1979:436-437). In this language, the initial obstruent of a word b is voiced after nasals ans spirantized after vowels. However, this lenition occurs only if word a is in certain syntactic relationships to b (in the following examples, the symbol ‘∩’ represents occurrence of lenition, and the symbol ‘/’ represents absence of lenition: (120) noun-noun compound a. q‘os ‘neck’ b. N@∩xos ‘otter neck’ c. ves / q‘os ‘crow neck’ adjective-noun a. t@f ‘house’ ∩ b. pilan d@f

‘big house’

possessive pronoun-noun a. p@x ‘paint’ ∩ b. N@N b@x ‘our paint’ direct object-verb a. v@kz-d ‘throw away’ b. ki ∩v@kz-d ‘throw away shoe’ c. Nas / p@kz-d ‘throw away belt’

120

No lenition occurs between two objects, between a subject and a verb, or between a subject and an object (the symbol ‘/’ represents absence of lenition): (121) a. ţaqo / p@kz-d knife disappeared b. ņi / ţaqo / p‘@k@n k‘im-d I knife (to) brother gave

Although it is not apparent that this language makes use of the edge condition, Kaisse claims that the c-command condition is necessary to account for the context of application of this rule. She argues that the syntactic context for lenition in Gilyak is the following: (122) Lenition occurs between a and b where b c-commands a.

This analysis is based on the assumptions that the rightmost member in noun compounds c-commands the leftmost member, a noun c-commands a preceding adjective, and a verb c-commands a preceding complement. These are the structures Kaisse posits: (123) a.

c.

N 3 N N NP d. 3 Spec N’ ! N

b.

NP 3 AP N

VP 3 NP V

In this analysis, the absence of lenition between a subject and a verb and a subject and an object is attributed to impossibility of any element in the VP to c-command

121

outside the VP, and similarly, the absence of lenition between two objects is explained by assuming that an element within an NP cannot c-command out of that NP. This approach to the syntax-phonology mapping, which posits direct reference to syntactic information for phonological rules that do not apply across the board, is called the Direct Reference Theory. We want to point out that one should be careful and not think that the DRT advocates for an isomorphism between syntactic and phonological constituents (such that every NP, VP, AP, etc. are automatically mapped as phonological phrases), or that it allows phonological rules to access all sorts of syntactic information. This is an unfair criticism to the model. The amount and type of information that the DRT allows phonological processes to access is limited, being constrained to the category-neutral, label-neutral, c-command relationships and edge conditions existing among syntactic terminal nodes, as determined by X-bar-theoretic hierarchical structure. As such, it is attractive and elegant, for its simplicity.xliii Unfortunately, the limited nature of the mechanisms that the model makes use of are insufficient to capture phonological interactions among syntactic heads which involve more than structural relations of ccommand. P2 rules, on the other hand, are fully productive, operating within and between words without regard to morphosyntactic information. They are constrained only by conditions on rate of speech and/or syllabification, and can be sensitive to pauses and intonational and phrasal boundaries. Flapping in English is an instance of a P2 rule. Using Kaisse’s examples, it can apply both to the t in sublimity and the t in write occasionally at a normal rate of speech, but not when the words are uttered in a syllableby-syllable or word-by-word pronunciation, or when pauses separate the t from the following vowel. Within a word, Flapping may apply within a stem or across any type of affixes or compound members. Syntactically, it does not seem to have any restrictions: it applies between subject and predicate, into and out of conjoined phrases, between the

122

phrasal complements of a verb, and between an NP complement and an extraposed clause: (124) a. The fruitba[D] achieved wide distribution. b. The fruitba[D] achieved and deserved a great deal of attention. c. The flying squirrel and the fruitba[D] achieved and maintained a wide distribution. d. We elected the fruitba[D] ambassador. e. We gave the fruitba[D] a shower. f. It was the fruitba[D] Annette was thinking of.

The model of grammar Kaisse suggests is represented in (125): (125) SYNTAX  S-structure

LEXICON Lexical phonology  Lexical representation Lexically interpreted surface structure  Simple cliticization 

Logical Form

POSTLEXICAL PHONOLOGY Rules of external sandhi Level P1  Pause insertion (= prosodic organization?)  Rules of fast speech Level P2 The rule of VA cannot be identified as a P1 or P2 rule, since it shows properties of both. On the one hand, VA could be classified as a P1 rule, since it has access to morphosyntactic information, as we have seen; it only applies between a lexical head and a functional head which realizes inflectional features. On the other hand, VA shows properties of P2 rules, namely sensitivity to intonational boundaries and pauses. As it is shown in (126), VA cannot apply to a vowel which is located immediately preceding a pause:

123

(126) a.

b.

neski-a etorri da # # girl-det.sg. come aux ‘The girl has come’



neskii etorri da # #

etorri da neski-a # # come aux girl-det.sg.



*etorri da neskii # #

The vowel of the determiner in neski-a in (126a) assimilates to the preceding vowel, but not in (126b), where it is utterance-final and is immediately followed by a pause. VA cannot affect vowels which are immediately followed by an Intonation Phrase boundary, irrespective or whether or not a pause immediately follows. In (127), for instance, the word alabia ‘daughter’ is the rightmost element in the topic phrase of the sentence, mariñeruen alabia ‘the fisherman’s daughter’, located before the main clause. This phrase constitutes an Intonation Phrase, which in the case of topics is signaled by phrase-final lengthening and a glottal stop at the beginning of the first Accentual Phrase in the main clause. A pause may follow the topic phrase, but is not necessary (see Jun and Elordieta 1997 for an analysis of the main Intonation Phrase properties in Lekeitio Basque, and Elordieta 1997 for a first study of the intonational structure of this language). The singular determiner -a cannot assimilate to the last vowel of the stem alabi: (127) a.

b.

mariñeruen alabi-a, bera bakárrik žun da Ondarrúra fisherman-gen.sg. daughter-det.sg. she alone/only go has Ondarroa-all. ‘The fisherman’s daughter, she has gone to Ondarroa alone’ *mariñeruen alabii, bera bakárrik žun da Ondarrúra

It is important to note that the prosodic condition is established on the segment which is the potential target of assimilation, not the word or the morpheme that contains it. Thus, VA seems acceptable with a plural determiner, even if a pause follows (cf.

124

128a). The judgments are clearer the more material is added to the determiner; (128b) is perfectly acceptable:xliv (128) a.

b.

atzo ikusi nebasan umi-ak ## / umíik ## yesterday see aux child-det.pl. ‘I saw the children yesterday’ atzo ikusi nebasan umi-ákiñ ## / umiíkiñ ## yesterday see aux child-det.pl.-com.pl. ‘I saw them with the children yesterday’

Perhaps it could be suggested that the constraint on VA is imposed by informational content, namely that VA does not apply to the last segment of a word which is part of old or background information, so as to try to preserve this information as faithfully as possible (as pointed out to me by Bernard Tranel and Moira Yip, p.c.). Indeed, in (126b) and (127) the words containing the determiner -a constitute old information, whereas in (126a) neskia is new information. It has to be said that there is no connection between informational content and possible constraints on VA. In Lekeitio Basque elements appearing in preverbal position are normally interpreted as new or focalized information, whereas elements appearing in postverbal position are interpreted as old information.xlv Thus, the words umíik and umiíkiñ in (128) constitute old information. However, VA may apply. Moreover, one could construct an example with the same word order as in (126b), in which the word neskia could be new information. In Lekeitio Basque it is possible to have a postverbal element interpreted as new information in sentences uttered as answers to a previous question, if it is separated from the verb by a pause (cf. Hualde, Elordieta and Elordieta 1994:167-170, and Osa 1990 for examples from other dialects). Still, VA is impossible: (129) - Nor etorri da? who come has ‘Who has come?’

125

- Etorri da, neski-a ## come has girl-det.sg.

/

*neskii ##

This strategy is more common with copulative verbs (cf. Hualde, Elordieta and Elordieta 1994:169): (130) - Nor da Clínton? Who is Clinton? - Clínton da Ameriketáko presidentía ## / *presidentíi ## Clinton is Amerika-of president ‘Clinton is the President of the United States’

Finally, consider sentences in which a postverbal element is interpreted as old information but is followed by more constituents. In (131), for instance, the verb is new or focalized information, and the rest of the sentence is old or bakground information. VA may apply freely, because neskia is not immediately followed by a pause: (131) etorri da neski-a goixian / come has girl-det.sg. morning ‘The girl has come in the morning’

etorri da neskii goixian

In sum, we have presented evidence that VA has mixed properties of P1 and P2 rules, the two types of postlexical rules defined by Kaisse (1985) (cf. Hualde and Elordieta 1992 for an earlier conclusion along these lines). Kaisse anticipates as a possibility that mixed rules may be found, following Kiparsky’s (1982) suggestion that phonological rules start as postlexical sound changes that evolve into acquiring sensitivity to morphosyntactic information. That is, in her model, phonological processes would start as P2 rules, sensitive only to prosodic conditions and rate of speech, and later they would become constrained by morphosyntactic conditions (P1 rules). Then it would be possible to find rules that display properties of the P1 level in the postlexical stratum,

126

still maintaining some residue of P2 properties. If this assumption is correct, perhaps VA constitutes an example of a rule in transition from P2 to P1 status. Be that as it may, in this dissertation we are not concerned with the classification of the process of VA as a P1 or P2 rule. Even if we were to classify it only as a P1 rule, we would run into serious difficulties in order to capture the correct domains of application of the rule in terms of c-command or edge conditions, as suggested by Kaisse. The theoretical model that posits direct sensitivity of phonological processes to syntactic conditions of this kind is known as the Direct Reference Theory. In what follows we will show the difficulties that this model faces in order to capture the domain of application of VA in Lekeitio Basque.

5.2. Direct Reference Theory. In this theory the application of postlexical rules between two elements a and b depends on the structural relationship of c-command existing between them, some rules also imposing branching conditions. From the assumptions on clause structure and head dependencies expressed in the previous section, this theory would not allow us to determine the context of application of VA. A participial verb adjoins to the auxiliary the same way it adjoins to a modal particle or causative verb, and thus the same c-command relationships hold between a participial verb and an auxiliary, a participial verb and a modal particle, and a participial verb and a causative verb. However, only in the first case do we find VA. As we can see in the diagrams in (132), repeated from (78), the adjunction structure that obtains when a participial verb raises to Aux0 and then to I0 is the same as the one obtained when it raises to a modal particle and a causative verb: (132) a.

T0 3 V0 T0

T0 3 V0 T0 3 0 V Mod0 b.

c.

V0caus 3 0 V V0caus

127

Note that similar results would be obtained even with a stipulation stating that the rule of VA affects the initial vowel of a word b if it c-commands a and is nonlexical, a possibility suggested by Kaisse (1985) as an alternative treatment of French liaison (Kaisse 1985:167, fn. 4). This analysis would not be able to make a distinction between auxiliaries and modal particles, since both are nonlexical, i.e., function words. As for subordinating conjunctions, we could propose that they occupy the Comp node in subordinate clauses, taking nonfinite clauses as complements. Since in Basque the subordinate clause precedes the element in Comp, the syntactic theory we have been assuming here would lead us to posit a leftward movement of the subordinate clause to the specifier position of the complementizer, i.e., Spec,CP. In (133b) we provide an illustration of this type of movement, corresponding to the subordinate clause in (133a): (133) a. Diru asko euki árren, astopótro bat da Péru. money much have despite brute one is Peru ‘Despite having a lot of money, Peru is a brute’ b.

CP qp TPi C’ 6 3 diru asko euki C0 ti : ! ! ! árren ! z---------------m

This configuration would be similar to the one involved in postpositions: (134) a.

b.

negua amaitxu árte winter finish until ‘until the winter ends’ PP qp VPi P’

128

6 3 negua amaitxu P0 ti : ! ! ! árte ! z---------------m

The c-command configurations involving subordinating conjunctions and postpositions do not help discern what is the possible c-command relationship that determines the correct domains of application of VA, since these structures are different from the one involving modal particles and causative verbs, which do not undergo the rule, either. We established the most solid proposal on the phrase structure of Basque that seemed possible, one which necessarily must have raising of the participial verb onto the inflected auxiliary. Accepting the correctness of this proposal, the reader is free to consider other alternatives by herself and check the possible structures resulting from verb-to-Infl raising. If VP raising to the specifier of IP (or ModP, when a modal particle is present) is considered, as pointed out in section 3, we would end up with identical structures: (135) a.

TP 3 VP T’ ! 2 V0 T0 t

b. VP

ModP 3 Mod’ ! 2 0 V Mod0 TP

There is no change if VP-adjunction is assumed, either: (136) a.

IP 3 VP IP ! 2 0 0 V I AuxP

b. VP

ModP 3 ModP ! 2 0 V Mod0 IP

129

This shows that trying to determine the domains of application of VA in terms of c-command relationships is not a successful approach, under every reasonable assumptions on the hierarchical relationships among heads and of verb movement.xlvi 5.3. Precompiled Phrasal Phonology. Hayes (1990) tried to eliminate from the postlexical component all rules applying across words whose structural description referred to syntactic information, suggesting that such rules belong to the lexical component, and should be considered as rules of phrasal allomorphy. This author adopts a view of the lexicon in which the lexical entries for every word include information of its syntactic and semantic properties, as well as a phonological instantiation frame, which indicates the phonological shape and context in which the word is instantiated. Words are not inserted in syntactic trees with phonological content. Only indices indicating the identity of a word and its syntactic properties are inserted under a syntactic terminal node. Phonological instantiation takes place postsyntactically. The simplest case would be exemplified by the lexical entry for a proper noun such as Bill: (137) 986 (index of Bill) (syntactic and semantic specifications) /bIl/

For a sentence such as John saw Bill, we could illustrate the two processes of lexical insertion (index insertion and phonological instantiation) as follows: (138) a.

Abstract Phrase Marker

b. Index Insertion 728 = index of John 142 = index of saw 986 = index of Bill

c. Phonological Instantiation (Postsyntactic)

130

S 2 NP VP ! 2 N V NP ! N

S 3 NP VP ! 2 N V NP ! ! ! 728 142 N jan ! 986

S 3 NP VP ! 2 N V NP ! ! ! s N ! bIl

This author views the lexicon as including a set of phrasal allomorphs for every word, generated by lexical phonological rules. Each of these allomorphs is marked to surface in certain syntactic contexts, encoded by means of phonological instantiation frames. Hayes illustrates this proposal through the rule of Hausa final vowel shortening, a process where final long vowels of verbs appear as short when the verb precedes a full NP direct object: (139) a. ná: kámà: I have-caught ‘I have caught (it)’ b. ná: kámà: ší I have-caught it ‘I have caught it’ c. ná: kámà kí:fí: I have-caught fish ‘I have caught a fish’ d. ná: ká:mà: wàxlvii Mú:sá: kí:fí: I have-caught Musa fish ‘I have caught Musa a fish’

Only in (139c) does the final long vowel of the verb ká:mà: appear as short, that is, when followed by a full NP direct object. In all other contexts, the vowel appears as long. This distribution would be captured by the following phonological instantiation frame:

131

(140) Frame 1:

/ [VP ___ NP ...], NP non-pronominal

This phonological instantiation frame states that verbs in Hausa may have an extra allomorph appearing in this syntactic context. The two allomorphs of the verb ká:mà: are ká:mà: and ká:mà. The rule of vowel shortening would then refer to this phonological instantiation frame, generating the allomorph with the short vowel: (141) V: --> V / [... ___][Frame 1] So, after syntax, at the point of phonological instantiation, one of the two allomorphs of the verb will be inserted; the shortened form will be inserted if its structural description matches the syntactic environment, and the longer form will be inserted elsewhere. Since the process of VA affects only to a specific class of elements (i.e., determiners and inflected auxiliaries), it would be desirable to consider a possible treatment of this phenomenon in terms of precompiled phrasal phonology, so as to compare it with our proposal. It turns out that the framework of precompiled phrasal phonology is not warranted for VA. A precompilation analysis of VA would force us to posit five allomorphs for vowel-initial auxiliary verbs. There would be the basic allomorph with the underlying initial /e/, an allomorph starting with the vowel /a/ to be inserted after a participial verb ending in /a/ (i.e., [Frame 1]), an allomorph starting with the vowel /i/ to be inserted after a participial verb ending in /i/ (i.e., [Frame 2]), an allomorph starting with the vowel /o/ to be inserted after a participial verb ending in /o/ (i.e., [Frame 3]), and an allomorph starting with the vowel /u/ to be inserted after a participial verb ending in /u/ (i.e., [Frame 4]). For example: (142) a. /eban/ b. /aban/ [Frame 1]

132

c. /iban/ [Frame 2] d. /oban/ [Frame 3] e. /uban/ [Frame 4] (143) a. b. c. d.

Frame 1: Frame 2: Frame 3: Frame 4:

/ [ V [...a] ___ ] / [ V [...i] ___ ] / [ V [...o] ___ ] / [ V [...u] ___ ]

Likewise, for each determiner, we would need three allomorphs: one with the underlying initial vowel (i.e., /a/ for the nonlocative singular and plural determiners, /e/ for genitive markers and locative plural determiners) and two more with the high vowels /i/ and /u/, to be inserted after the last word in an NP ending in an /i/ or /u/ ([Frame 1] and [Frame 2], respectively). Let us illustrate this for the singular determiner: (144) a. /-a/ b. /-i/ [Frame 1] c. /-u/ [Frame 2] (145) Frame 1: Frame 2:

/ [NP [...i] ___] / [NP [...u] ___]

This analysis fails to capture the fact that VA is a process first of all. Listing allomorphs for every single inflected auxiliary or determiner/acse marker of the language and specifying in the lexicon the syntactic contexts where the allomorphs surface does not tell us what really is going on: that the initial vowel of an auxiliary or determiner/case marker assimilates to the last vowel of a participial verb or of the last word in an NP, respectively. The phonological process itself, i.e., assimilation, is sufficiently productive and motivated from a phonological point of view, unlike arbitrary or lexically specified phonological changes which would seem better suited for an analysis in terms of lexical storage of allomorphs and morphosyntactic environments. But of course, acknowledging that VA is a process implies abandoning the idea of allomorphs altogether to account for this process, since the nature of the process renders

133

them superfluous. Only one basic allomorph would be necessary, i.e., the underlying form of the auxiliary or determiner. Still it could express the descriptive generalization that the rule only affects the initial vowel of inflected auxiliaries and determiners, but that is merely a description of the facts. It does not render an explanation of the nature of the domain of application of the rule:xlviii (146)

Root ! V

Root ! [V [Aux / D]

The theory of precompiled phrasal phonology as devised by Hayes is best suited to account for phenomena which affect and are triggered by specific syntactic categories or morphemes. In Hausa, only verbs undergo the rule of final vowel shortening, not other categories such as nouns or adjectives. The a/an alternation of the indefinite determiner in English is analyzed by Hayes as an instance of a precompiled rule, affecting just this particular syntactic category. A similar approach is suggested to analyze the alternation that the Spanish femenine definite determiner shows between el and la. It is el before nouns whose initial vowel is stressed, and la elsewhere: (147) la tórre ‘the tower’ la cáma ‘the bed’ la almoháda ‘the pillow’ el água ‘the water’ la álta tórre ‘the high tower’

The noun água triggers the selection of el as the allomorph for the definite determiner, by virtue of having an initial stressed /á/. The last example shows that it is only nouns that trigger allomorphy selection; the adjective álta, although starting with a stressed /á/, does not force the appearance of el. Hayes expresses the allomorphs and their environments as follows:

134

(148)

/el/ —

/ ___ [N á

/la/



VA in Lekeitio Basque, however, is not a process of this kind. It has a limited distribution, but it is not a rule that affects only a specific morpheme or syntactic category. It affects determiners and auxiliary verbs preceded by nouns, adjectives, and participial verbs. Saying that the rule applies to determiners and auxiliaries preceded by nouns or adjectives and verbs is not saying much apart from describing the problem. Such a procedure simply states in the lexicon regularities which belong to syntax, the component of grammar dealing with combinations of words. It fails to capture the generalization that only categories realizing inflectional features are capable of undergoing the process. Our analysis provides an explanation for this generalization. Moreover, it is apparent that the rule of VA cannot be a precompiled rule, even under the assumptions fostered by Hayes himself. Phonological instantiation of precompiled rules takes place after syntax but before phrasal phonology, that is, before pauses and prosodic constituency (cf. Nespor and Vogel 1986) form part of the linguistic derivation. Hayes makes the explicit claim that rules that are blocked by pause cannot be precompiled rules, but they must be rules of true phrasal phonology: “any rule blocked by pause is a rule of true phrasal phonology” (Hayes 1990:107). Since VA is sensitive to intonational boundaries and pauses, as shown in section 5.2, it cannot be treated as a precompiled rule then, . Also, VA is an optional rule of colloquial style, that is more frequent the faster the speech becomes. This is another property of phrasal rules, but not of precompiled rules: “Assuming that phonological instantiation is not sensitive to speaking rate, then rules that apply in larger domains at greater speaking rates must be true phrasal rules, and not precompiled.” (Hayes 1990:107).

135

Summarizing, the discussion of the previous two subsections suggests that VA poses problems for its classification in terms of the theory of lexical phonology. The rule could be argued to apply both lexically (within the word boundaries of a noun or adjective) and postlexically, but when it applies postlexically, it shows sensitivity to syntactic information, that is, it applies in the very restricted domain of a verb followed by an inflected auxiliary. The rule cannot be classified as a P1 postlexical rule in the sense of Kaisse (1985) either, since it is also sensitive to pauses and intonational phrasing, properties that belong to P2 rules, i.e., rules of late postlexical application and insensitive to syntactic information. We have showed that the contexts of application of VA cannot be defined under the c-command and edge conditions postulated by the Direct reference Theory, because no distinctions exist in c-command relationships in the configurations formed by the sequences V-Infl and V-Mod-Infl. We also saw that VA cannot be analyzed in terms of precompiled phrasal phonology, because it does not display properties attributed to precompiled rules, such as arbitrary or lexicalized phonological changes, and allomorph storage in the lexicon. Listing inflected auxiliaries and determiners or case markers in the lexicon as the only syntactic categories that can undergo the process means giving up on the generalization lying behind this fact, and an explanation based on it: that these elements are inflectional heads. Our analysis, presented in section 4, builds on this generalization. In the remainder of this chapter we will concentrate on the other theory of phrasal phonology, the Prosodic Hierarchy Theory. This theory, in turn, is split in two models, the Relation-Based approach and the End-Based approach. After providing an overview of the basic claims and postulates of the Prosodic Hierarchy Theory and of each of the two models representing it, we will show that they cannot predict correctly the domain of application of VA in Lekeitio Basque. This is because these models do not recognize the special properties of functional heads containing bundles of inflectional features, treating

136

them the same as other functional categories. The validity of our proposal will thus gain force.

5.4. Prosodic Hierarchy Theory. The basic postulates of the Prosodic Hierarchy Theory (PHT) are explicitly stated in Selkirk (1978, 1980ab), and Nespor and Vogel (1982, 1986). As we mentioned earlier, the main claim of the Prosodic Hierarchy Theory (PHT) is that there exists a suprasegmental, hierarchically arranged organization of the utterance, called prosodic structure. This structure is composed of a finite set of universal prosodic constituents, which are the domains of application of phonological rules and phonetic processes. From the bottom up, these constituents are the syllable, the foot, the prosodic word, the clitic group, the phonological phrase, the intonational phrase, and the utterance.xlix These constituents are mapped from morphosyntactic structure by algorithms which make reference to nonphonological notions, that is, syntactic information, but prosodic structure and the constituents that compose it are not isomorphic with syntactic structure. The motivation for the PHT comes mainly from the observation that most phonological processes applying across words seem to require access to very limited morphosyntactic information, and that syntactic constituents do not determine the domains for the application of phonological rules. The PHT acknowledges that a distinction has to be made between processes which are sensitive to pure phonological or prosodic conditioning (their object of study), and processes that are directly sensitive to morphological structure or syntactic category information in an arbitrary way. Among the latter, phonological rules that are triggered by morphemes of a derivational or inflectional type, or by compounding, are deemed as the object of Lexical Phonology (cf. Kiparsky 1982, 1985, Mohanan 1982, 1986). In some languages there have been reported cases of phonological processes that are triggered or undergone by specific morphemes in isolation, without apparent extensions to other morphemes that belong to the same

137

morphological or semantic class. The PHT does not focus on these cases, which could perhaps be treated by means of lexical specifications idiosyncratic to those morphemes, that is, by assuming that the morpheme in question is lexically specified to trigger or undergo a certain rule. On the other hand, there are some rules that apply across words that need to refer to syntactic labeled bracketings, such that their context of application is specified on the basis of categorial information (e.g., N, NP, V, VP, P, PP, etc.). The two vowel deletion rules of Greek discussed in Kaisse (1977) are illustrative examples of this type of rules. One of the vowel deletion rules applies across two words only if they are contained in an NP, and the other one imposes the condition that the two words must be part of the same VP. Another example would be the rule of Verb Final Vowel Deletion in Italian, which optionally deletes the final vowel of a word a when followed by another word b which is its complement, but only if word a is a verb. If it is a noun, the rules does not apply (cf. van Hoorn 1983, Vogel et al. 1983). These and similar examples, described in Kenstowicz and Kisseberth (1977) and Kaisse (1977), among others (see references therein), are considered to fall outside the scope of prosodic phonology and form a different subsystem of rules. Hayes (1990) claims that these rules receive a better treatment if they are considered to apply in the lexicon, as precompiled phrasal rules, given their idiosyncratic domains of application (cf. chapter 3:3.3 for discussion of precompiled phrasal phonology). In any case, the PHT is not concerned with these rules, nor with those that require reference to morphological structure. Only phonological rules applying below and above the word level that are sensitive only to phonological or prosodic factors constitute the object of study of the PHT. As far as syntax is concerned, Nespor and Vogel try to provide additional evidence against an approach that posited syntactic constituents as domains of application of phonological rules. One of the problems that would arise for such an approach would be the existence of rules which are sensitive to the length of the syntactic constituents.

138

Under a strictly syntactic approach, this would be unexpected, because the number of words is irrelevant for the definition of syntactic constituents. A constituent composed of two words should have the same phonological behavior of a constituent of the same type composed of more than two words. However, this is not the case with certain rules. For instance, Nasal Assimilation in Spanish. In this language, nasals are homorganic to a following consonant both within a word and across words: (149) a. ga[m]ba b. co[m] piedad c. la[N]gosta d. come[N] carne

‘shrimp’ ‘with pity’ ‘lobster’ ‘(they) eat meat’

Nasal Assimilation does not apply between just any two words, however. Syntactic constituency does not determine the domain of application of the rule. Rather, it depends on the length of the syntactic constituent containing the word with the nasal and the one containing the following word. Thus, compare (150) and (151), examples in which the word containing the nasal ends a subject NP and the triggering word starts a VP: (150) a.

[ NP Mi faisá[N] ] [ VP corre siempre ] . my pheasant runs always

( B / [...X___Y]Dj [Z...]Dj ...]Di

142

ii)

A --> B / [...X]Dj [Y___Z...]Dj ...]Di

c. Domain limit: i)

A --> B / [...X___Y]Di

ii)

A --> B / [X___Y...]Di

In this model of PHT, functional elements such as determiners, auxiliaries, possessive pronouns, or conjunctions are included in phonological words, clitic groups or phonological phrases with the heads with which they are associated phonologically. That is, these elements may be independent phonological words, which combine with other words to form phonological phrases, they may be clitics which attach to phonological words to create clitic groups, or they may be affixes which combine with a stem to form a phonological word. This is the typology of phonological patterns of the so called function words. For the sake of discussion, let us present the algorithms for the creation of phonological words, clitic groups and phonological phrases that Nespor and Vogel (1986) propose: (156) ω domain A. or B.

(ω = phonological word)

The domain of ω is Q.(Q = terminal element of a syntactic tree) I

The domain of ω consists of a. a stem; b. any element identified by specific phonological and/or morphological criteria; c. any element marked with the diacritic [+W]. II Any unattached elements within Q form part of the adjacent ω closest to the stem; if no such ω exists, they form a ω on their own.

Phonological words may thus be equal to or smaller than the terminal element in a syntactic tree (i.e., Q), as expressed by A and B in (156), respectively. Possibility A refers to phonological words which are composed of the stem and all affixes, or by the

143

two members of a compound together. Greek and Latin are languages which illustrate this parameter of ω-domain formation. In Greek, primary word stress must fall on one of the last three syllables of a word, the exact location of stress being predictable on the basis of stress assignment algorithms and morphological information (cf. Steriade 1988, Halle 1997 and references therein). The following examples are taken from Nespor and Vogel (1986:ch.4): (157) a. álo◊os b. patéras c. axinós

‘horse’ ‘father’ ‘sea urchin’

Compound words form a single domain for stress assignment, i.e., stress falls on one of the last three syllables: (158) a. kúkla spíti  kuklóspito doll house b. níxta pulí  nixtopúli night bird

‘doll’s house’ ‘night bird’

Possibility B Ia is exemplified by those cases in which only a stem plus affixes counts as a phonological word, that is, with each member of a compound word forming its own phonological word (e.g., Sanskrit, Turkish, Italian, Hungarian, Dutch; cf. Nespor and Vogel 1986:117-143). In Turkish, all monomorphemic and derived words have primary stress on the last syllable: (159) a. çocúk b. çocuklár c. çocuklarimíz

‘child’ ‘children’ ‘our children’

In compounding, however, stress assignment does not compute the whole complex word as a single domain of stress assignment. The first member retains primary stress on its final syllable, and secondary stress falls on the last syllable of the second

144

member. This is the pattern found in independent words which are joined together in a phonological phrase: (160) a. düğünçiçegì of yesterdayflower b.

çáy evì tea house

‘butter cup’ ‘tea house’

The rule of Vowel Harmony also provides additional evidence that compounds are separate prosodic words. This rule harmonizes all vowels in a domain for the feature [back] and high features for the feature [round]. Whereas suffixes form one single domain for harmony, members of compounds do not: (161) a. ev-in ‘of the house’ house-gen. b. vapur-un ‘of the steamer’ steamer-gen. (162) bugün this day

‘today’

Part Ib of possibility B refers to the distinctions that some languages show between prefixes and suffixes in terms of phonological word formation. In Hungarian and Italian prefixes are specified to form independent phonological words, as opposed to suffixes, which combine with the stem to form one phonological word (cf. Nespor and Vogel 1986:122-134), and affixes which form phonological words on their own by virtue of satisfying minimal prosodic size requirements such as bisyllabicity (e.g. YidiJ; cf. Nespor and Vogel 1986:134-136). In Hungarian a rule of Vowel harmony applies to spread the feature [back] (except for the vowels /i/, /i:/ and /e:/. Whereas the rule applies to a sequence stem+suffix, it does not apply to the sequence prefix+stem. Compare (163)(164) with (165) in the following examples from Hungarian:

145

(163) a. ölelés b. ölelés-nek

‘embracement’ ‘embracement (dat.sg.)’

(164) a. hajó b. hajó-nak

‘ship’ ‘ship (dat.sg.)’

(165) a. be-utazni ‘to commute in’ in commute b. fel-ugrani up jump

‘to jump up’

Part Ic refers to affixes which are idiosyncratically specified to form independent words, as in Dutch (cf. Nespor and Vogel 1986:136-140). In the phenomenon of Coordination Reduction in this language (Booij 1985), the second part of a complex word deletes under identity with the second part of the following complex word in a coordinate structure: (166) a. landbouw en tuinbouw b. land en tuinbouw ‘agri(culture) and horticulture’ (167) a. een elfjarige, twaalfjarige jongen b. een elf, twaalfjarige jongen ‘an eleven-, twelve-year-old boy’

No deletion can apply to parts of an underived word, and certain derivational morphemes cannot be deleted either: (168) a. blauwig en rodig b. *blauw en rodig ‘bluish and reddish’ (169) a. absurditeit en banaliteit b. *absurd en banaliteit ‘absurdity and banality’

146

According to Booij (1983), no pricipled distinction based on morphological or phonological grounds can be established between the suffixes that can undergo the rule and those that cannot. It is thus posited that the suffixes which do not undergo the rule are marked as prosodic words on their own. Hence the diacritic [+W]. Part II of possibility B responds to the necessity of obeying the Strict layer Hypothesis, which says that every level of prosodic constituency must exhaustively parse the entire segmental string. That is, every segment must be part of a phonological word, a phonological phrase, an intonational phrase, and an utterance. By this condition, elements that do not qualify as stems, such as conjunctions, complementizers and clitics, also form a ω, either by attaching to a ω within Q or by themselves. The existence of the clitic group is proposed on the grounds of the observation that there are phonological rules that only apply to the sequence formed by a lexical word and the clitic that attaches to them (cf. Cohn 1989, Hayes 1989liii). The clitic group is defined as follows: (170) Clitic Group Formation I. C domain The domain of C consists of a ω containing an independent (i.e. nonclitic) word plus any adjacent ωs containing a. a DCL, or b. a CL such that there is no possible host with which it shares more category memberships. II.

C construction Join into an n-ary branching C all ws included in a string delimited by the definition of the domain of C.

This definition assumes that there are elements lexically specified as clitics, with the mark [+CL], following Klavans (1982). DCL and CL stand for directional and nondirectional clitics, respectively. DCLs are idiosyncratically specified for directionality of attachment, i.e., as proclitics or enclitics. CLs would be those that only require an adjacent host.liv

147

Nespor and Vogel (1986:125-129) analyze a rule of standard northern Italian, Intervocalic s-Voicing (ISV), which applies within words (cf. (171)), but not across words (cf. (172)). The rule treats clitics as independent words (cf. (173)): (171) a. a[z]ola b. a[z]ilo c. ca[z]ina

‘button hole’ ‘nursery school’ ‘house (diminutive)’

(172) a. hanno [s]eminato have seeded b. la [s]irena the siren (173) a. telefonati-[s]i phone-reflexive b. lo [s]apevo it I-knew

*hanno [z]eminato

‘(they) have seeded’

*la [z]irena

‘the siren’

*telefonati[z]i

‘having called each other’

*lo [z]apevo

‘(I) knew it’

Additional evidence is provided by the rule of Raddoppiamento Sintattico, that lengthens the initial consonant of a word after a stressed vowel which ends the preceding word (cf. (174)). It does not apply within words, as shown in (175), and clitics behave as if they were independent words with respect to Raddoppiamento (cf. (176)): (174) a. tré [k:]ani b. piú [b:]ello

‘three dogs’ ‘more beautiful’

(175) a. sú[b]ito b. tá[v]olo

*sú[b:]ito *tá[v:]olo

‘immediately’ ‘table’

(176) a. da[m:]i give-me

(

ipiñi eutzan

We do not include such forms because diphthongs resist assimilation of one or both of its members (*ipiñi iutzan, *ipiñi iitzan). Only short vowels can undergo the rule of VA. We take this to be an independent phonological property of Lekeitio Basque, which has nothing to do with the morphosyntactic composition of the inflected auxiliary. See footnote 7 and the end of section 4.3 for additional examples and discussion. xiv

The root of the transitive auxiliary *edun is -u-, which becomes -b- by a process of intervocalic

labialization. The vowel a then would be simply an epenthetic vowel inserted between this consonant and the -n marking past tense. xv

The -s- after the ergative/absolutive plural marker is traditionally identified as an absolutive pluralizer,

used only in transitive forms to mark the presence of a third person plural absolutive, i.e., to refer to a third person plural direct object. xvi

The vowel -e*- appearing after the root of the auxiliary, b, is a pluralizer morpheme referring to a second

and third person plural absolutive participant in intransitive forms and a second and third person plural ergative participant in transitive forms. In (14c,d) it indicates the presence of the third person plural ergative morpheme. Traditional linguists do not classify the vowel -a- appearing in singular forms as a singular marker, however. It is simply taken as an epenthetic vowel (cf. fn. 10). xvii

The vowel appearing before the -r in Spanish infinitives has often been considered in Spanish

descriptive grammars as a separate morpheme, called thematic vowel. Since no notion of a morpheme or

175

grammatical category similar to a thematic vowel exists in Basque, I assume that this vowel does not constitute a morpheme on its own. Thus, the Basque adaptations in (15) can be taken as nonderived roots. xviii

Cf. Euskaltzaindia 1985, Mujika 1988 for detailed descriptive analysis on these and other modal

particles found in other dialects. For an overview of the main properties of modal particles in other languages, and discussion of previous work on the topic, see König 1991, §8.2. xix

Xabier Artiagoitia (p.c.) reminds me that Rebuschi (1984), in an article published in (1990), also posits

an initial INFL, but I have been unable to check this article. xx

As we mentioned above, the inflected auxiliary is a term used traditionally to refer to a complex form

containing morphemes that spell out person, number and case features, tense, mood, and a root of an auxiliary that does not exist as a regular verb any more in Basque (all the auxiliary verbs except one are reconstructed). Thus, the head T0 is just a cohesive label we give to this bundle of morphemes, for the sake of simplicity. How many different syntactic projections should be represented and in which hierarchical order are issues that do not concern us here. We will not consider the consequences of positing an independent Auxiliary Phrase in the structure. Suffice it to say that none of the syntactic analyses of the structure of the clause in Basque that I am aware of posit such a projection. xxi

We take Infl or I0 in Ortiz de Urbina’s analysis to be T0 in our analysis, but we will maintain Ortiz de

Urbina’s terminology throughout the presentation of his analysis. xxii

Nowadays only approximately twenty verbs can be inflected synthetically, although not all of them are

shared by all dialects. In Lekeitio Basque, for instance, only ten verbs allow synthetic inflection: ixan ‘to be’, egon ‘to be, stay’, ñu(a)n ‘to go’, etorri ‘to come’, ibilli ‘to walk’, euki ‘to have (possession)’, eru(a)n ‘to carry’, ekarri ‘to bring’, ñakiñ ‘to know’, and an accented root -iño* ‘to say’. The paradigms they form are defective, in that only present and past tense forms are possible, and some verbs only allow present tense forms. There is no semantic or syntactic criteria determining the synthetic class, in what seems a lexically idiosyncratic distinction. For a list of synthetic verbs, see the grammar of Euskaltzaindia (1985),

176

the Academy of the Basque language. For a list of synthetic verbs in Lekeitio Basque, see Hualde, Elordieta and Elordieta (1994:3.2.2.2, 3.2.2.4). xxiii

To be exact, as Ortiz de Urbina (1994) points out, the constraint on initial position refers more

specifically to clause-initial position rather than sentence-initial, as evidenced by sentences involving topic constituents preceding an inflected auxiliary or synthetic verb. These type of sentences are ungrammatical, much like (25b,c) and (26b,c): (i) a. *Jonek, dau liburúa erosi b. *Jonek, dakar liburúa? xxiv

In this respect, I disagree with Ortiz de Urbina, who assumes ba- to be inserted in the specifier of CP. I

take ba- to be a last resort spell-out of the affirmative and question morphemes in C0, introduced in the absence of an element in Spec,CP and/or a lexical head in C0. In this view, C0 would be a head that needs to be lexically filled, i.e., licensed, by elements with the relevant features. If no such elements are available in the sentence, ba- is inserted, possibly in PF. xxv

A preliminary statement on this line is exposed in Laka (1988), although the author does not yet

explicitly articulate the reasons for the absence of V-to-I movement. xxvi

Albizu (1992a) suggests -t(z)en is specified as [+completed]. This would make it indistinguishable from

-tu/-i, however. Although he does not explicitly state so, it would have to be assumed that the feature [+habitual] corresponding to -t(z)en would distinguish them. xxvii

Some verbs ending in a vowel do not add the perfective morphemes -tu/-i/-n, but still convey the

meaning associated to these morphemes when combined with an inflected auxiliary (cf. Hualde, Elordieta and Elordieta 1994:111-112 for an exhaustive list of this group of verbs). xxviii

Xabier Artiagoitia (p.c.) points out to me that the answers in (66)-(67) are not standard, and that they

most probably belong to eastern dialects. He also drwas to my attention the possibility of treating omen and ote as adverbials heading an Adverbial Phrase, which would constitute another alternative to Albizu’s

177

examples. Although worth exploring, I will not consider this alternative here. In my analysis, these particles have all the properties of heads, not of adjuncts or heads of specifiers or adjuncts, as shown by the fact that they move together with the participial verb and inflection. xxix

The question that all analyses would have to answer is how the logical scope of the modal particle over

the whole clause can be obtained. In our analysis, the answer is straightforward for interrogative and negative sentences, because the modal particle is in C0 and thus c-commands all elements in the clause. In affirmative neutral declarative sentences, we would have to assume that the wide scope of the modal particle is obtained by LF-raising of the modal particle to a projection where it c-commands the whole clause, most likely CP. xxx

One possible advantage of this analysis, however, would be that having modal particles above IP

without incorporation with the inflected auxiliary permits an account of the VP-ellipsis and parenthetical usages of modal particles in the dialects where those constructions are possible (cf. (66)-(67)). xxxi

Notice also that this alternative would have to assume the impossibility for a modal particle to license

an inflected auxiliary, much like our proposed analysis. xxxii

Laka (1988, 1993) argues that Infl in Basque forms a complex X0 in syntax, in a head-final approach of

Basque clause structure. xxxiii

Other features which could be assumed to be present in D0 would be referentiality and definiteness, at

least. Of these, definiteness has been treated as an agreement feature before (e.g., Borer 1988, 1994, Siloni 1994). The D0 position has also been analyzed as the locus of nominal inflectional features such as gender and number. As part of checking theory, we assume that the checking operation inside DP can be carried out in overt or covert syntax. In Basque, the features of the determiner are checked in overt syntax. N-to-D raising, to check referentiality features in D, is proposed by Longobardi (1994) and Chomsky (1995:364). Syntactic head-to-head movement from N0 to D0 has been argued to apply in Semitic (see Ritter 1986, 1989, Ouhalla 1988, Fassi Fehri 1989, Siloni 1989, 1990, and references cited there), Scandinavian (see

178

Delsing 1988, Taraldsen 1990, Holmberg 1992) and Romanian (cf. Grosu 1988, Dobrovie-Sorin 1987). For the particular case of Basque, on the basis of examples like (69) and (72), it seems that the most apparent analysis of the NP-D surface order is to posit NP-movement to D0 instead of head-to-head movement from N0 to D0. xxxiv

Halpern (1995) uses the term lexical clitic to refer to phrasal clitics.

xxxv

The causative verb eraiñ may well select an IP in Basque, instead of a VP, on the basis of the

diagnostics provided for causative verbs by Ritter and Rosen (1993). But even in such a case the participial verb and the causative verb would appear in the same X0 on surface, so in the end whether the causative verb selects an IP or a VP does not matter for our present purposes. As for the motivation for the movement of the participial verb to the causative verb, one possibility would be that this movement is semantic in nature, in particular for complex predicate formation. xxxvi

We leave open the possibility that the participial verb is inflected for the feature [causative]. Even if it

were, the nature of the relationship between the verb and the causative is still not as cohesive as the one holding in the primitive feature chains. xxxvii

This maping theory is in a way reminiscent of Inkelas’ (1990) ideas on the mapping between

morphological and prosodic constituency in the lexicon. The analysis I am positing here could thus perhaps be seen as an extension of her insights to the postsyntactic level. xxxviii

The /z/ of negation is the spelling remnant of what once was a voiced alveolar fricative. In Biscayan

Basque it is not voiced any longer. xxxix

The other possibility is for members of compounds to form one single phonological domain. This is

observed in Greek and Latin, according to Nespor and Vogel (1986:110-115). In our model the most logical assumption would seem to be that members of compounds are m-words. Then, in order to account for languages like Greek or Latin, we would have to say that phonological processes may select either the

179

internal m-domains, (mapped as p-domains), or the external m÷p domain. The first instantiation of the parameter would treat the members of a compound as separate phonological units, while the second instantiation would treat them as one unit. Whether or not this is correct, I leave open for future research. xl

Note that there has been deletion of the final /n/ of the participail verb in this example. This is an

obligatory process which I discuss in this section. xli

As for why ñ-insertion is optional between a participial verb and an inflected auxiliary but obligatory

between the evidential modal particle ei and an inflected auxiliary, the plausible hypothesis that we will adopt here is that it responds to a functional pressure to distinguish a modal particle from the verb ein ‘do, make’. This verb is used when the participial verb is focalized, and it may optionally drop its final nasal consonant, causing the insertion of a voiced palatal fricative /j/ or a voiced prepalatal fricative /ñ/ (cf. (107 as well, where the verb ein is used as a participial verb): (i) saldu ein eban --> saldu eyeban / saldu eiñeban sell do aux ‘(s)he SOLD it’ We can argue that making ñ-insertion an obligatory process between the evidential modal particle and an inflected auxiliary is a way to maintain a surface distinction between the emphatic particle ein and the evidential modal particle ei. No such pressure exits when a participial verb is involved, and thus the process may remain optional. xlii

The definition of c-command used is domain c-command (from Lobeck and Kaisse 1984:171): (i) Domain c-command: In the structure [Xmax ... α ...], Xmax is defined as the domain of α. α c-commands everything in its domain. α=Xj, jβ0 This definition of c-command is actually m-command, that is, a m-commands b if the first maximal projection dominating a also dominates b. To avoid any confusions, we will maintain the name ‘ccommand’, keeping in mind that it stands for m-command. This condition will be relevant in section 5, when we consider the limitations and inadequacies of the different theories of phrasal and prosodic phonology to account for the correct domains of application of VA.

180

xliii

See Odden (1987, 1990) for attempts to demonstrate the adequacy of the DRT over competing theories

of the syntax-phonology interface, such as the PHT, to be reviewed in the next section. xliv

VA seems more acceptable in (30b) than in (30a), perhaps because the former contains more material

before the pause. xlv

In Lekeitio Basque, when the whole sentence is new information (i.e., when the sentence is uttered “out-

of-the blue”, or as an answer to the question “what happened?”), indirect objects are postverbal. That is, the order among constituents in these cases is S-DO-V-IO. Incidentally, we will not consider indirect objects in our present discussion, because they always end in -ari (-ári in the dative plural or if the base root is accented), that is, they do not present a sequence of two heterosyllabic vowels followed by a pause or Intonation Phrase boundary. xlvi

If it turned out that the best analysis of modal particles is that they are particles which are adjoined to

Infl, there still would not be a distinction between verb adjunction to a modal particle or verb adjunction to Infl. xlvii

It is not clear what the meaning of the particle wà is. According to the translation and the word order in

this sentence, it might be an indirect object marker. xlviii

In fact, Hayes suggests that “rules of phonological spreading, particularly spreading of autosegments

over multiple syllables, would in many cases be impossible to treat as precompiled, and thus would have to be analyzed as true phrasal phonology.” (Hayes 1990:107). VA involves spreading of features, although in a local domain (between adjacent vowels), and therefore it has a property that would make it even less justifiable as a precompiled rule. xlix

Additional constituents have been proposed, such as the Focal Phrase, located between the Phonological

Phrase and the Intonational Phrase (cf. Kanerva 1990), the Small Phonological Phrase, intermediate between a Phonological Phrase and a Phonological Word (cf. Selkirk 1986), or the Small Word, which

181

comprises part of a Word (cf. Rice 1993). Only the ones mentioned above have received universal consideration, however, except for the clitic group, on which see below. l

Interestingly, however, Hayes and Lahiri resort to syntactic conditions such as c-command to define their

default P-phrasing and P-phrase restructuring algorithms. If their analysis is correct, it constitutes a point of contact between the two models of syntax-phonology interaction discussed in this section. I will not elaborate on this matter, however, due to space limitations and because it is not of central interest to our main proposal in this dissertation, which is the proper definition of phonological domains involving lexical heads and inflectional heads. li

Chen (1990) suggests that in order for the Relation-Based approach to be complete and accurate, it would

need to refer to the argument-adjunct distinction as well. lii

The first two principles are subsumed under Selkirk’s (1984) Strict Layer Hypothesis.

liii

Bruce Hayes was the first one to propose the clitic group as a constituent in the prosodic hierarchy, in a

1984 paper, only later published in 1989. liv

Incidentally, it must be pointed out that positing nondirectional clitics contradicts Klavans’s (1982) claim

that all clitics are specified for direction of attachment. lv

The auxiliary that does not have an underlying accent (such as the ones in (192a,b)) forms another stress

domain with whatever material follows. More exactly, it forms an Accentual Phrase with following material until the end of the utterance or until the next H*L pitch accent (cf. Elordieta 1997, Jun and Elordieta 1997). lvi

Actually, the reader should be warned that there is an error in Hale and Selkirk’s article, on page 177.

The syntactic structure that appears there for head-final languages is similar to the one posited for headinitial languages, i.e., (84). Reading the text discussing the two structures the reader notices that the structure the authors had in mind was the one we present in (86).

182

lvii

Notice, in any case, that we are dealing with complex X0s in verbal contexts, and that therefore we have

to refer to the X0s composing those complex heads. If we were not able to refer to the structure below the topmost X0s it would not be possible to make a distinction among types of heads. This is an important point which is not mentioned in the End-Based approach. Only simple X0s are taken into account. lviii

We ignore here the internal structure of the NP. Whether the right edge of theNP also coincides with the

right edge of the AP, for instance, derives the same results. lix

We are not claiming that prosodic words and phonological phrases cannot be built choosing opposite

directions of prosodic edges. This possibility is allowed by the EB theory, and no conflicts should arise if prosodic words and phonological phrases were domains where different processes were specified to apply. In the case under discussion, however, it would be claimed that the domains formed by phonological phrasing could affect the application of VA, which also applies in prosodic words. This is what creates the conflict.

202

Chapter 3 Additional evidence for the creation of phonological domains from morphosyntactic feature chains 1. Introduction. In this chapter I will provide additional examples of phonological processes which only occur between lexical heads and functional heads containing sets of formal inflectional features, in support of the explanatory line of investigation presented in the previous chapter.

Section 2 is devoted to the analysis of French liaison. This

phenomenon refers to the pronunciation of latent word-final consonants in certain syntactic configurations. Liaison applies obligatorily between determiners, possessive pronouns and nouns, and between subject pronouns and following verbs. This can be explained following the idea that feature chains are mapped as morphosyntactic constituents. An m-domain is formed at the level of Morphological Structure that comprises a functional head containing (bundles of) features and a lexical head that checks those features (D-N, V-D, D-T). This domain, in turn, is mapped as a phonological domain, where French liaison is specified to apply obligatorily. One noteworthy aspect of liaison is that unlike in Basque, no syntactic incorporation takes place between the heads which intervene in the process. That is, the morphological domain is purely reflecting a grammatical relationship. This fact provides additional evidence in favor of the existence of domains or constituents that are visible in a separate component of grammar, different from syntax and phonology. Liaison does not apply obligatorily with auxiliaries, copulas, prepositions, and degree modifiers. Auxiliaries and copulas are verbs inflected for tense and P-features, but they are the verbal heads checking those features, not the participial verb or predicate that follows them. Thus, the feature chains are self-contained in the auxiliaries and copulas, so to speak, and there is no primitive feature chain linking them to following elements. Prepositions form part of feature chains with the determiner (P-D), but they are mapped

203

at MS as independent words. As for degree modifiers, it is not apparent that they enter into checking relations with the adjectives they modify. They do not share P-features, and are not assigned Case. The idea that cohesive morphosyntactic relationships are reflected in the morphophonological component of grammar is thus evidenced once again. Although not obligatory, liaison is highly frequent with monosyllabic forms of prepositions, degree modifiers, auxiliaries and copulas. This pattern is analyzable as due to a prosodic factor, the clitic-like nature of monosyllabic function words. This analysis suggests an interesting division of morphological and prosodic factors as sources of phonological domains. The focus of section 3 is the phenomenon of Initial Consonant Mutation (ICM) in Irish. This phenomenon refers to the phonological mutations undergone by the initial consonants of lexical words immediately following functional heads. Before Old Irish, the word-final segments that triggered mutation on the following word-initial consonant were lost, and only functional heads maintained their mutating properties, in the absence of the phonological environment. Lexical categories lost their mutating properties together with the loss of their word-final segments. Although the synchronic lack of phonological transparency in the mutation environment has been taken to indicate that ICM is not really a phonological rule but an allomorphic process, triggered by certain particles, the interesting fact is that precisely those functional categories expressing morphosyntactic features maintained their mutating properties on following heads they enter into checking relations with. Of the three languages studied in this dissertation, Irish is the one that reflects all feature chains as morphosyntactic domains at MS, domains which are later mapped as phonological domains. An observation that provides support for our theory is that a there is a functional category, the negative specifier gan, that does not trigger any kind of mutation. This fact is explained from the fact that this element is not a head realizing inflectional features that need to be checked.

204

2. French liaison. In French there are two types of word-final consonants: fixed and latent. Fixed consonants are always pronounced, whereas latent consonants are pronounced only when the following word starts with a vowel, a phenomenon known as liaison. This contrast is illustrated in (1)-(2), where the consonants that are pronounced are underlined, and unpronounced consonants are enclosed in angled brackets. The adjective net ‘clear’ in (2a) has a fixed consonant, which is always pronounced, before a vowel or consonant. The adjective petit ‘small’ in (2b) has a latent consonant, only pronounced when the following word starts with a vowel (examples taken from Tranel 1995:799): (1) a. net avantage clear advantage b. net dèfaut clear defect (2) a. petit avantage small advantage b. peti dèfaut small defect

2.1. Representation of liaison consonants. Three major approaches to French liaison can be distinguished: chronologically speaking, the first approach was one which advocated for underlying consonants which got deleted when followed by another consonant or pause. Thus, Selkirk (1974) would characterize this rule as follows (cf. Schane 1968, 1974, Dell 1970 for similar alternative formulations): (3)

[-son]



{

/ __ #

{ [+cons]} { # }

A competing account is one which views liaison as consonant insertion (cf. Klausenburger 1978, Kaye and Morin 1978, Tranel 1981, Morin and Kaye 1982). The

205

analysis consists essentially of positing epenthetic consonants in certain morphologically and syntactically specified contexts. The epenthetic consonant must be lexically specified for each lenition environment. Tranel (1981) for instance captures plural verb liaison with two rules, as follows: (4) a. X ]V

- third person [+ plural ]

1 Examples:

2

3⇒

1 2 t 2 3

nous arrivons ensemble vous chantez encore

b. X ]V + third person [+ plural ] 1 Examples:

# [+syll]

# [+syll] 2

3⇒

1 2 t 2 3

ils arrivent ensemble elles chantent encore

Prenominal adjective liaison consonants are inserted by the following scheme: (5)

[+syll]]A + masc [+ sing ] 1

# 2

N

[ [+syll] 3



1 2 C 2 3

Each adjective has to be lexically specified as to the C that gets inserted. Thus, petit /p@ti/ is specified for /t/, gros /gro/ for /z/, etc. A third alternative approach is the suppletion solution, which is actually already introduced by Tranel’s insertion framework. Inserting a lexically specified consonant amounts to saying that that consonant is a suffix to the stem and present in the lexicon (Klausenburger 1984:27). Morin (1983), Klausenburger (1984) attempt to reduce liaison to the choice of suppletive forms in appropriate phonological environments, thus treating

206

liaison as allomorph selection, triggered by the neighboring phonological environment. The following table for adjectives illustrates this position:

(6) petit mauvais gros grand

pre-C, -h, -G1i, -pause

pre-V, -G2

p@ti movεz gro grã

p@tit movε groz grãt

Within the framework of autosegmental phonology, Clements and Keyser (1983) offer an account of French liaison which conceives the liaison consonants as ‘latent consonants’. These consonants are part of the underlying form, and marked as extrasyllabic, that is, they are not associated to the syllabic structure of the word which they belong to. They may be phonetically realized if the following word starts with a vowel, in which case the extrasyllabic consonant is syllabified as part of the following word. If the following word starts with a consonant, the extrasyllabic consonant cannot syllabify with the following word, and it does not surface phonetically (i.e., it is erased at the end of the phonological derivation). The workings of this analysis can be exemplified with phrases like petit ami ‘little friend’ and petit garçon ‘little boy’, which would have the following derivational representations : (7) a.

σ σ t! t! C V C V C ! ! ! ! ! p @ t i t

b.

σ σ t! t! C V C V C ! ! ! ! ! p @ t i t

σ σ ! t! V C V ! ! ! a m i σ σ t!y t! C V C C V ! ! ! ! ! g a r s O

We will not attempt to decide whether the phenomenon should be treated in terms of consonant deletion, consonant insertion, or suppletion. The reader is referred to Tranel (1995) for a comprehensive

review of all the proposed alternatives, as well as

207

Klausenburger (1984), which includes an excursus through the diachronic development of French liaison. Tranel (1995) argues that the most satisfactory way to represent the distinction between fixed and latent consonants is to posit fixed consonants as anchored to a skeletal C position and latent consonants as floating segments, i.e., C slots that are not linked to the melodic tier (cf. Wetzels 1987, Zoll 1994, 1996 for recent approaches on the represenattion of floating segments): (8) a.

Fixed consonants

b.

σ 8 x x x ! ! ! C V C

Latent consonants σ 1 x x ! ! C V C

Now we turn our discussion to our main topic of interest: the issue of what syntactic contexts allow for the occurrence of liaison.

2.2. Liaison and styles of speech. The contexts of occurrence of Liaison in French differ along three styles of speech: the style of conversation familière, conversation soignée and the style of lecture or discours. These styles are also known as conversation courante, conversation sérieuse et soignée and style soutenu, respectively (Fouché 1959). Following Selkirk (1972), and for ease of reference, we will call them Style I, II and III, respectively. Style I is the most colloquial or least elevated style of speech, the style that is the most restrictive in liaison contexts. The liaison contexts that appear in Style I are the “basic” contexts, which appear in the more elevated or less colloquial Styles II and III as well. These three styles are in order of increasing number of liaison contexts; the set of contexts characteristic of style n is a subset of the set of liaison contexts characteristic of style n+1. That is, if a set

208

of contexts ci is found in Style I, a set of contexts cii which includes ci is found in Style II. In turn, the set of liaison contexts that appear in Style III is ciii, which includes cii and ci. This means that Style III is the least restrictive in liaison contexts. For instance, in Style III liaison is possible between a plural noun and an adjective, between a plural adjective and a following complement, or between a verb and a following complement, but liaison is not allowed in these contexts in Style I. These contexts are exemplified in (9a-c), respectively (examples drawn from Selkirk 1972). In order to indicate the presence or absence of liaison, we will follow the practice started in Chapter 1, that is, the



symbol

between two consonants indicates that liaison takes place, and the symbol / indicates that liaison is not possible. The outputs in (9) are ungrammatical in Style I: (9) a. les marchands∩anglais the merchants English (pl.) ‘the English merchants’ b. Ces traits sont propres∩à certains individus que je connais. these features are proper of certain individuals that I know ‘These features are proper of certain individuals that I know’ c. Elle donnait∩un cours à l’université. she gave a course at the university ‘She taught a course at the university’

In our discussion, we will limit ourselves to Style I, where the basic contexts of liaison are found, because these are the only environments where the rule applies obligatorily. According to Rotenberg (1978), the set of contexts in Style I is the real linguistic phenomenon, because its common denominator is a rule which is unconsciously known by every speaker. In addition to this rule, every speaker is also aware of a certain social pressure to make more liaison in more formal social contexts. The elevated style, proper of higher social classes, responds to an effort to imitate spelling and earlier periods in the history of French in which final consonants were pronounced more often. However, there are no fixed rules of elevated liaison, and there is

209

almost no agreement about the facts of liaison in non-conversational styles (cf. the data provided by Morin and Kaye 1982, against Selkirk 1972, 1974). There is a great deal of individual variation in the frequency and consistency with which liaison is produced, and liaison in elevated styles decreases as the scale of socioeconomic classes goes down (cf. De Jong 1988, 1990). For all these reasons, Rotenberg calls the non-conversational or non-casual styles artificial and normative. I thus agree with him that the domains of liaison which should be investigated are those found in Style I, which induce obligatoriness of liaison and which are present in the speakers’ grammatical competence. From now on then, all the data and distribution of liaison will be based on the most conversational or colloquial style of speech, Style I.

2.3. Distribution of liaison. The interest of the distribution of liaison in Style I is that not all words with latent consonants make liaison obligatorily in the presence of a following vowel-initial word. Words belonging to certain categories make liaison obligatorily, while other only do so optionally, although frequently. And within the latter class, liaison is found more often with monosyllabic forms than with polysyllabic forms.ii, iii

2.3.1. Obligatory liaison. We find obligatory liaison in the following contexts: - between a determiner and a following noun or adjective: (10) a. des ∩ ennemis ind.det enemies ‘(some) enemies’ b. un∩arbre a tree c. les∩aimables marchands

210

the friendly merchants - between a demonstrative and a following noun or adjective: (11) a. cet∩effort this effort b. ces∩amis these friends - between a possessive pronoun and a following noun or adjective: (12) a. mon∩oeil my eye b. ton∩écriture your writing c. son∩opinion his opinion d. nos∩oppresseurs our oppressors e. leurs∩effets their effects - between a numeral or quantifier and a following noun or adjective:iv (13) a.

vingt-trois∩oignons twenty-three onions

b.

trente-deux∩assauts thirty-two attacks

c.

trois∩ingénieurs three engineers

d.

plusieurs∩autres enfants many other children

211

e.

aucun∩officier no officer

f.

quelques∩expériences a few experiences

- between certain adjectives and a following noun: (14) a.

mon petit∩ami my little friend

b.

de beaux∩artichauts beautiful artichokes

c.

un grand∩arbre a big tree

d.

un gros∩handicapv a big handicap

e.

un long∩hiver a long winter

f.

un bon∩avocat a good lawyer

- between object clitics and a following verb: (15) a.

Il vous∩a donné la réponse. he you has given the answer ‘He has given you the answer’

b.

Elles nous∩ont appelés they us have called ‘They have called us’

c.

Je les∩écoute I them listen to ‘I listen to them’

d.

J’en ∩ ai plusieurs I-of it have several

212

- between a verb and a following subject or oblique clitic: (16) a.

Sont∩-ils arrivés? have they (masc.) arrived ‘Have they arrived?’

b.

Est∩-elle allée à Paris? is she gone to Paris ‘Has she gone to Paris?’

c.

Ont∩-elles accepté? have they accepted ‘Have they accepted?’

d. Allons∩-y tous ensemble. go-there all together ‘Let’s go there all together’ - between a subject clitic and a following verb: (17) a.

Vous∩avez fini. you have finished

b.

Nous∩avons déjà mangé. we have already eaten

c.

Ils∩ont entrés dans la salle they have entered in the room

2.3.2. Frequent liaison. Liaison is frequent but not obligatory in the following contexts: - between a monosyllabic preposition and a following word in a complement DP: (18) a.

dans∩une salle publique in a room public ‘in a public room’

213

b.

sans∩aucune ami without any friend

c.

chez∩un ami vi home a friend ‘at a friend’s home’

- between a monosyllabic degree modifier and a following adjective:vii (19) a.

On le considère très∩ambitieux. imper. him consider very ambicious ‘He is considered to be very ambicious’

b.

C’est une personne bien∩aimable. is a person very friendly ‘He is a very friendly person’

c.

Elle est plus∩âgée que lui. she is more old than him ‘She is older than him’

d.

fort∩intéressant. very interesting

- between a monosyllabic auxiliary and a following participial verb: (20) a.

Jeanne est∩allée voir le résultat. Jean has gone to see the result

b.

Je suis∩arrivée trop tôt. I have arrived too soon

c.

Ils sont∩entrés dans la salle. they have entered in the room

d.

Les commerçants ont∩augmenté toutes les prix. the merchants have all raised the prices

- between monosyllabic forms of the copula être and a predicate:

214

(21) a.

Il est∩insupportable que tu y ailles tout seul. It is unbearable that you there go all alone

b.

Nous sommes∩impatients de vous voir à Paris. we are impatient of you see in Paris ‘We are anxious to see you in Paris’

c.

Ses amis sont∩intelligent. his friends are intelligent

2.3.3. Rare liaison. Liaison is rare with polysyllabic forms of the categories just mentioned in 3.3, i.e., prepositions, degree adverbs, auxiliaries and copulas. Thus, there is a contrast between the monosyllabic prepositions in (18) (i.e., dans, sans, chez) and polysyllabic prepositions such as pendant ‘during’, depuis ‘since’, avant ‘before’, devant ‘in front of’, après ‘after’, which consist of two syllables. Some examples are provided in (22), where the combination of a ‘/’ sign and a question mark indicate that liaison is possible, although harder to make:viii (22) a.

pendant /? une semaine during one week

b.

depuis /? un an since one year

c.

devant /? une maison in front of a house

Monosyllabic degree adverbs such as those in (19) (i.e., très, bien, plus), after which liaison is very frequent, contrast with the bisyllabic degree adverbs assez ‘rather’, or vraiment ‘truly’, after which it is harder to have liaison: (23) a. Leurs rapports avec l’administration sont assez /? intimes. their relations with the administration are rather intimate

215

b. Il est vraiment /? idiot he is truly idiotic

Similarly, after bisyllabic forms of auxiliaries and the copula liaison is not so common in Style I. Compare the examples in (20)-(21) with those in (24): (24) a.

Vous avez /? étonné tout le monde. you have surprised everybody

b.

Les soviétiques auraient /? annoncé la fin des négociations. the soviets had announced the end of the negotiations.

c.

J’avais /? avancé cette idée moi-même. I-had advanced that idea myself

d.

Les mères seront /? impatientes de voir leurs fils. the mothers will-be anxious to see their children

2.3.4. Absence of liaison. No liaison is found between any other two elements in Style I: between a noun and an adjective, a noun or adjective and a preposition in a complement Prepositional Phrase, an adverb and an adjective, a nonpronominal subject and a verb, a verb and an object, between two objects, or between an adverb and a verb. These contexts are exemplified in (25a-g), respectively:ix (25) a. des langues / étrangères languages foreign ‘foreign languages’ b. des voitures / à deux chevaux cars with two horses c.

insolents / envers leurs patrons insolent toward their employers

d.

extrêmement / amusante. extremely amusing

216

e.

les hommes / étaient fatigués. the men were fatigued

f.

Elle donnait / un cours à l’université. she gave a course at the university

g.

Donnez ces lunettes / à Paul. give these glasses to Paul

h.

les soldats ont facilement / avancé. the soldiers have easily advanced

After having presented the syntactic contexts where liaison applies, in the next section we will review the proposals that have been made to account for the domain of application of liaison. 2.4. Previous analyses. 2.4.1. Selkirk (1972). Selkirk (1972) stated that all monosyllabic non-lexical items triggered liaison. Since according to SPE conventions non-lexical words are separated from what follows by only one word boundary, #, she concluded that liaison was specified to apply when a maximum of one # separated two words. That # would be the one introduced by the lexical word following the non-lexical element. Thus, for degree modifiers, auxiliaries, copulas and clitics, she posits the structures in (26)-(29), respectively: (26)

(27)

(28)

moins plus ]Q A’‘[ # Q[ trop sont [ # [ suis V’‘ Aux ont est V’‘[

# V’[ [ sont ] est

A’[

]Aux

PP

# A[Adjective]A ]A’ # ]A’‘

V’[

[#.........] PP ]V’ # ]V’‘

A’‘ N’‘

# V[Verb]V ]V’ # ]V’‘

A’‘ N’‘

217

(29)

V’‘[

# V’[ V[ [Clitic] [Clitic] V[#Verb ]V ]V ]V’ #]V’‘

The presence of two #s intervening between two lexical words accounts for the absence of liaison between two lexical words, such as an adverb and a following verb, for example: (30)

V’‘[

# A’‘[ # A’[ A[

brusquement ]A ]A’ # ]A’‘ { toujours }

V’[# V[Verb]V

]V’ #]V’‘

Similar structures would be posited for sequences of a noun and an adjective, or a verb and its complement. Selkirk has to resort to two ad-hoc readjustment rules, in order to account for the behavior of polysyllabic non-lexical words and prenominal adjectives. Recall that polysyllabic degree adverbs, prepositions, auxiliaries and copulas do not trigger liaison as readily as monosyllabic ones. Selkirk simply stipulates a rule that adds word boundaries to polysyllabic function words, which makes them like lexical words: (31)

[C0 V C0 V C0 X ]

⇒ [ # C0 V C0 V C0 X # ]

Prenominal adjectives present the problem that they are lexical words, and thus according to SPE conventions should be separated from the following noun by two word boundaries. To solve this problem, Selkirk calls for a language-particular rule which deletes the #’s that are to the right of the adjective within the adjective phrase: (32) ⇒

[N’‘# X [A’‘# Y [A’#[A#Adjective#A]#A’]#A’‘] [N’ # [N#Noun# N] Z # [N’’# X [A’‘# Y [A’#[A#Adjective A]

2.4.2. Rotenberg (1978).

A’] A’’]

N’]# N’‘]

[N’ # [N#Noun# N] Z # N’ # N’‘]

218

Rotenberg (1978) showed the inadequacies of boundary theory as a theory of phrasal phonology, and thus criticized Selkirk’s treatment of French liaison as well. However, his model still has to acknowledge the special status of non-lexical items in phrasal phonology. Rotenberg claims that a sequence of a non-lexical item plus a lexical item forms a unit, called clitic group, in the sense that all non-lexical categories for Rotenberg are clitics, which are marked in the lexicon as proclitics or enclitics, that is, as elements which cliticize phonologically to the left or to the right of an adjacent lexical element. Then, liaison would be a c-level rule, that is, liaison occurs between two adjacent words which are included in the same clitic group, here indicated with the subscript c: (33)

W (c X w w Y ) Z

Although in our analysis to French liaison we will use the term clitic to refer to certain non-lexical elements in French, we will show that not all non-lexical items can be classified as such, and that there is a distinction among non-lexical items. The difference in phonological behavior among functional categories will be used as a diagnosis of a difference in morphological status. We will present our argument in section 3.5.

2.4.3. Direct Reference Theory. Kaisse (1985) provides an analysis of French liaison in terms of the Direct Reference Theory, the basic postulates of which we presented in chapter 2, section 5.1. She claims that the contexts of application of liaison can be delimited as follows: (34)

Liaison applies between two words a and b where b c-commands a.

The term c-command utilized by Kaisse is misleading, since it is not interpreted in the usual sense of the syntactic tradition in the Government and Binding Theory (i.e., α

219

m-commands β if the first branching node dominating α also dominates β). Rather, ccommand as utilized by Kaisse is an abbreviation for domain c-command, which is equivalent to m-command (that is, α m-commands β if the first maximal projection dominating α also dominates β). Thus, Kaisse’s claim should be interpreted to mean that liaison applies between two words a and b where the first maximal projection dominating b also dominates a. Kaisse claims that an analysis in terms of c-command can obviate the stipulation that nonlexical items always trigger liaison. Determiners, possessive pronouns, clitics, degree adverbs and prenominal adjectives are all c-commanded by the right-adjacent elements with which they make liaison, since the maximal projections containing the latter also contain the former. Kaisse follows assumptions of syntactic phrase structure of Chomsky (1981), and treats nonlexical items as specifiers of lexical maximal projections. Thus, determiners and possessive pronouns as specifiers of the maximal projection headed by a noun, i.e., N”, degree modifiers are specifiers of A”, and auxiliaries are specifiers of V”. This means that determiners, possessive pronouns, degree modifiers and auxiliaries are c-commanded by the respective head nouns, adjectives and verbs. As for prenominal adjectives, they are taken as leftward complements of nouns. Hence, in this case nouns c-command adjectives as well. Clitics are assumed to be sisters of verbs, so the c-command relation between the verb and the clitic is straighforward. Here we introduce the structural relationships Kaisse assumes for the following pairs of elements: determiner-noun, degree adverb-adjective, and prenominal adjective-noun: (35) a.

c.

N” 3 Det N’ ! N N” 3 Det N’ 3

b.

A” 3 Deg A’ ! A

220

A”

N

Clitics are assumed to be sisters of verbs, so the c-command relation between the verb and the clitic is straighforward. Several problems may be noticed in this analysis. First, the liaison context formed by a determiner and a prenominal adjective entails difficulties for a c-command treatment. A prenominal adjective cannot c-command out of its maximal projection, A”, and thus it cannot c-command the determiner: (36) a.

un∩éminent avocat an eminent lawyer

b.

un∩intelligent ami an intelligent friend

c.

N” 3 Det N’ 3 A” ! A’ ! A

Kaisse notes this problem and suggests several solutions, none of which is satisfactory, in my opinion. One possibility she suggests is that prenominal adjectives are nonphrasal, but this is purely stipulatory. Another alternative is to recur to the special status of nonlexical categories that Rotenberg referred to, namely that determiners and other nonlexical items cliticize to the following word; then this word would be able to ccommand them: (37)

N” ! N’

221

3 A” N ! A’ ! A 3 Det A

This alternative would imply going back to the distinction lexical/nonlexical that Kaisse herself wanted to avoid, so if valid it would seriously diminish the alleged force of her proposal, which is that the domains of application of rules of external sandhi are governed by structural relations of c-command among participants, without regard to syntactic class membership or category. A last suggestion of a solution to the problem is that the c-command condition could be weakened to allow liaison if b c-commands a or if a is nonlexical and c-commands b. This alternative would present the same problems as the previous one. Another problem would be posed by liaison occurring between prepositions and what follows. Prepositions are not considered specifiers, but heads of their own maximal projections, i.e., prepositional phrases. If the following material is included in a different maximal projection, as complement of the preposition, these items cannot c-command out of this maximal projection, and thus cannot c-command the preposition. However, liaison is found between monosyllabic prepositions and whatever vowel-initial material follows: (38) a. b. c.

dans∩une salle publique in a room public chez∩elle (at)home she (‘at her house’) P” 3 P N”

222

3 Det N’ 3 N

A more serious problem maybe arises when we take into account current assumptions on syntactic structure in the Principles and Parameters framework, an early model of which Kaisse is basing her analysis on. Functional categories are not considered to be specifiers of maximal projections headed by lexical categories. Rather, they are heads of their own maximal projections, taking maximal projections as complements (cf. Abney 1987, Fukui 1986, Fukui and Speas 1986, among the earliest proposals). Thus, a determiner is not a specifier of a nominal phrase, but the head of the Determiner Phrase (cf. Abney 1987, Szabolcsi 1983, 1987), which selects a Noun Phrase as a complement; a complementizer is the head of a CP (former S’), with Inflection Phrase (i.e., IP, the former S) as its complement; this IP is headed by Infl, and has an Auxiliary Phrase or a Verb Phrase as its complement; an auxiliary verb heads an Auxiliary Phrase, taking a Verb Phrase as a complementx. These are widely accepted assumptions nowadays. This being so, Kaisse’s analysis cannot account for the environments in which liaison applies, because the words following determiners and auxiliaries cannot c-command out of their maximal projections, that is, word b does not c-command word a, but liaison occurs. This is the same problem as the one posed by prepositional phrases discussed above: (39) a.

DP 3 D NP 3 N

b. Aux

AuxP 3 VP 3 V

What seems to be the only way out for Kaisse’s model would be to say that liaison applies between word a and word b when word a is a nonlexical category and it c-

223

commands word b. However, as pointed out above, this refinement involves acknowledging a distinction between lexical and nonlexical categories that the DRT does not want to make reference to. Furthermore, such a distinction is insufficient, because of the categorical judgments it imposes on the presence or absence of liaison. As we have seen in section 2.3, not all nonlexical categories behave similarly with respect to liaison. There are contexts in which liaison is obligatory, but there are other contexts in which it is

frequent,

and

others

in

which

it

is

rare.

Determiners,

demonstratives,

numerals/quantifiers and possessive adjectives make obligatory liaison, whereas prepositions, degree adverbs, auxiliaries and copulas do not. The revised analysis suggested above cannot predict these facts, since it is based entirely on the syntactic configuration in which two words appear, and those syntactic configurations are shared by determiners, prepositions, and auxiliaries/copulas alike. This major problem is shared by other accounts proposed in the literature, which we will discuss in the remainder of this section. As a result, the conclusion we will reach is that an alternative approach is necessary, one which distinguishes among the functional categories that can make liaison in Style I. The analysis proposed in the previous chapter is able to make finer grained distinctions among nonlexical categories, based on the relevance of morphosyntactic relationships, and will help explain why liaison applies obligatorily in the restricted environments in which it is set to apply. As was proved with VA in Lekeitio Basque, our analysis will show that (unfortunately perhaps) the nature of the syntax-phonology interface is not as simple as Kaisse’s DRT advocates.

2.4.4. The End-Based Approach. 2.4.4.1. Selkirk (1986). Selkirk (1986) provides an analysis of the basic contexts of liaison in terms of the End-Based Approach that she developed. Remember from our presentation in chapter 2:5.5 that Selkirk (1986) argues that there is a postsyntactic level of phonological

224

structure called Prosodic Structure. This structure is built from syntactic structure by parametrized settings that place boundaries to the right or left edges of words or maximal projections. Selkirk adduces the existence of a prosodic constituent in a variety of languages which contains a phrasal head and a preceding or following specifier material, but not any complement of the head. She calls this constituent the small phonological phrase, to distinguish it from the maximal phonological phrase, built from the setting of the parameter for Xmax. Thus, the complete set of parameters for the mapping of syntactic structure onto prosodic structure give rise to three levels of prosodic structure, respectively: the prosodic word, the small phonological phrase, and the phonological phrase. Languages choose the direction of prosodic boundary insertion for each level of prosodic structure. The settings can be represented as in (40): (40) End parameter settings: (i) a. ]Word b. Word [ Prosodic Word (ii) a. ]Xhead b. Xhead [ Small Phonological Phrase (iii) a. ]Xmax b. Xmax [ Phonological Phrase

According to Selkirk, small phonological phrases in French are created by the insertion of right-hand brackets after each phrasal head. That is, the selected parameter for prosodic constituent formation referring to the category Xhead is ]Xhead, meaning that a head ends a small phonological phrase. The string of material contained between two edges marked as ]Xhead is the domain where obligatory liaison applies. This analysis is illustrated in (44)-(46), which represent the prosodic structure of the sentences in (41)(43) after the insertion of ]Xhead boundaries. The underlined consonants indicate that liaison has occurred (cf. (35)-(37)in Selkirk 1986:396): (41)

Sais-tu quand ils inviteront un autre grand artiste? know-you when they will-invite another big artist ‘Do you know when they will invite another big artist?’

(42)

Ces très aimables enfants en ont avalé. these very nice children some-have drunk ‘These very nice children have drunk some of it’

225

(43)

On m’a souvent amené dans un énorme wagon. one me-has often taken me in an enormous train car ‘One often took me in an enormous train car’

(44) a.S[[Sais-tu]V [[quand]Comp [ils [[inviteront]V [un [autre]AP [grand]AP [artiste]N]NP]VP]]] b. ...........]Xhead ........................................]Xhead .........................................]Xhead c. (_____) (_____________________) (______________________) (45) a. S[[Ces [[très]Det [aimables]A]AP [enfants]N]NP [[en] [ont]]Aux [avalé]V ]VP]]] b. ..........................................................]Xhead ................................]Xhead c. (______________________________) (________________) (46) a. S[On VP[[m’[a]]Aux [souvent]AdvP [amené]V PP[[dans]P [un] [énorme]AP [wagon]N]NP]]]] b. .........................................................]Xhead ..............................................]Xhead c. (_____________________________) (________________________) As (44)-(46) show, Selkirk treats determiners, auxiliary verbs and prepositions as specifiers, not as heads, and thus no ]Xhead boundary can be inserted to their right. They are always included in the same domain with a following head noun or verb, to whose right a right-edge ]Xhead boundary is inserted. In Selkirk’s words,...the ]Xhead setting distinguishes between the heads of maximal projections and the heads of specifier constituents. Moreover, this setting is successful in bringing preceding function words into the small phonological phrase with the following phrasal head. This is no surprise: any theory must assume that function words are not identifiable as constituents of any sort in the syntax. Whether heads or not, the bracketing of function words is ‘invisible’ to the rules deriving phonological domains. (p. 396) Prenominal adjectives and preverbal adverbs are also considered as specifiers, which do not project a maximal projection. The author cites as evidence the fact that prenominal adjectives do not contain complements, whereas postnominal adjectives do. Although she does not provide actual examples, the following contrastive pair mentioned by Valois (1991:154) illustrates her point: (47) a. b.

L’invasion [AP improbable aux yeux des Terriens] de Jupiter. the invasion improbable to the eyes of the Earthlings of Jupiter *L’[AP improbable aux yeux des Terriens] invasion de Jupiter. the improbable to the eyes of the Earthlings invasion of Jupiter

226

In this sense, prenominal adjectives would differ from postnominal adjectives in that the former are not contained in a maximal projection AP. Selkirk concludes that they are not phrasal heads, and thus do not count as “real” heads for the insertion of prosodic boundaries. This means that no ]Xhead boundary is inserted on the right edge of prenominal adjectives, and they fall in the same domain with the following noun. This is why liaison occurs between prenominal adjectives and head nouns. Although Selkirk does not provide arguments for treating adverbs in the same manner, she simply adopts the same analysis. One objection to be raised against this analysis is that the claim that only lexical heads can project a ]Xhead boundary needs justification, especially in the light of syntactic developments in the last decade. It is currently a common tenet in generative syntactic theory (which Selkirk’s assumptions are based on) that at least some functional categories head their own syntactic projections. Thus, if we were to keep up with nowadays practice, what would be the reason to deny prosodic head status to functional heads? Is it an accident of grammar?

But perhaps more importantly, the claim for the invisibility

of function words is too general. As we saw in Chapter 2 when we discussed Vowel Assimilation in Lekeitio Basque, not all nonlexical words display the same pattern in postlexical phonology. And as we saw in section 2.3 in the distribution of liaison, a distinction must be made among function words regarding their phonological behavior. A class of function words triggers liaison whenever it is followed by another head, while another class of function words only triggers liaison optionally (very frequently) in monosyllabic forms. This is a distinction Selkirk (1972) noted, but Selkirk (1986) overlooks (cf. fn 8). Thus, this approach has to face the same problems we pointed out above for an analysis based on the Direct Reference Theory. Namely, it cannot account for the differences among function words in degree of frequency of occurrence of liaison. Treating all function words alike, as invisible to prosodic boundary insertion (cf.

227

Selkirk’s 1984 Principle of Categorial Invisibility of Function Words), leads to the impossibility of capturing the distribution of liaison stated in section 2.3. Second, although the assumption that prenominal adjectives and adverbs are nonphrasal specifiers could perhaps be maintained, what ought to be clarified is what is meant when it is said that they do not count as “real” heads. From the syntactic point of view, every linguistic expression inserted under a terminal node must be a head.xi Maybe what is meant is that they are adjoined to the following lexical material, but that is never made explicit in the analysis.

2.4.4.2. De Jong (1990). Within the framework of the End-Based approach of prosodic phonology, De Jong (1990) attempted to provide an analysis that could solve this problem. Providing empirical support from real speech data (the Orléans corpus, cf. Blanc and Biggs 1972), De Jong acknowledged the existence of different degrees of frequency in the realization of liaison, depending on the grammatical category of the word whose final consonant was involved. De Jong pointed out that liaison was less frequent after prepositions, copulas and auxiliaries, even in monosyllabic forms, than after determiners, possessive adjectives, demonstratives, quantifiers and numerals, or clitics. He suggested that the difference between obligatory and optional liaison lies in the difference between ‘real’ function words and closed-category phrasal heads. ‘Real’ function words such as determiners, possessive pronouns, demonstratives, numerals and clitics are not heads of any kind, but specifiers or modifiers, whereas prepositions, auxiliaries, copulas are function words occupying the head position of a maximal projection (Prepositional Phrase and Verb Phrase, respectively). De Jong argues that the domain of obligatory liaison is more restrictive than Selkirk’s (1986) small phonological phrase, namely the prosodic word (PWd), parsed from syntactic structure by picking out the right edge of every head. Since real function words are not heads, they will fall in the same prosodic word with an

228

adjacent head. The domain of optional liaison would be the small phonological phrase (SPP), which is built inserting prosodic boundaries at the right end of every lexical head. Then, a more inclusive domain where liaison is not found in conversationsl style would be the maximal phonological phrase (MPP), constructed picking out the right edge of lexical maximal projections. This is summarized in (48): (48)

PWd formation: SPP formation: MPP formation:

]Xhead ]Xhead, where Xhead is major-category. ]Xmax

The example in (49), slightly adapted from De Jong (1990:82), illustrates the workings of this algorithm: (49) MPP: ( ) ( ) SPP: ( ) ( ) ( ) PWd: ( ) ( ) ( ) ( ) ( ) ( ) [[ont]V [été]V [aidés]V]VP] PP[[par]P [des [enseignants]N S[Ils [[admirables]A]AP]NP]PP they have been helped by indef. teachers wonderful ‘they have been helped by wonderful teachers’

This analysis is still not satisfactory, because it assumes that determiners, possessive pronouns, numerals and quantifiers are specifiers or modifiers of some sort, not heads. However, as pointed out above, for the past decade or so it has been standardly assumed in the syntactic theory of Principles and Parameters (or even the most recent minimalist framework, cf. Chomsky 1995) that these elements are heads, much like prepositions, copulas and auxiliaries. In a simplified manner, these would be the syntactic representations for determiners and quantifiers:xii (50)

DP 3 D NP ! ! les N !

QP 3 Q NP ! ! plusieurs N !

229

amis

enfants

If this is the case, how can we maintain the distinctions among functional heads regarding their different degrees of participation in liaison? It seems clear that an alternative approach is necessary in order to solve this issue. We will present such an alternative in the next section.

2.5. An analysis of obligatory liaison in terms of morphosyntactic feature chains. We want to propose an analysis of French liaison that is readily available using the notion of feature chains and their morphosyntactic mapping discussed in Chapters 1 and 2. First of all, we will present an asymmetry between function words in French; a class of function words triggers liaison categorically, whereas another class triggers liaison depending on prosodic conditions. As we presented above, liaison is always made between determiners, demonstratives, possessive pronouns, numerals or quantifiers and the following noun or adjective (cf. (10)-(13); between certain prenominal adjectives and following head nouns (cf. (14)); and between clitics and verbs, whether the clitics follow or precede the verb (cf. (15)-(17)). It is optional, although almost obligatory, to make liaison between monosyllabic prepositions, degree adverbs, auxiliary verbs, and copulas (cf. (18)-(21)). However, liaison is harder to obtain with polysyllabic prepositions, degree adverbs, auxiliaries and copulas (cf. (22)-(24)). In contrast, no difference is perceived with polysyllabic numerals and quantifiers. According to all sources, the number of syllables does not seem to play a role in the occurrence of liaison with these elements. This contrast is illustrated below, using examples already presented above. In (51)-(52) we give examples of polysyllabic prepositions, polysyllabic degree adverbs, and polysyllabic

230

auxiliaries and copulas; in (53) we give examples of polysyllabic quantifiers and numerals: (51) a. pendant /? une semaine during one week b. depuis /? un an since one year c. devant /? une maison in front of a house (52) a. assez /? intimes. rather intimate b. vraiment idiot truly idiotic (53) a. avez /? étonné have surprised b. auraient /? annoncé had announced c. seront /? impatientes will-be impatient (54) a. vingt-trois∩oignons twenty-three onions b. trente-deux∩assauts thirty-two rounds c. plusieurs∩amis many friends

This asymmetry has to be accounted for. Our proposal is that the domains of application of French liaison can be defined in terms of the mapping from syntactic structure to morphological structure. Determiners, demonstratives, possessive pronouns,

231

quantifiers and numerals, subject and object clitics, are all functional categories which enter into feature checking relationships with the nominal heads that are their complements. They realize morphosyntactic features of agreement, i.e., the P-features person, gender and number (in the case of determiners, possessives and numerals, only number and genderxiii), which are also specified in the noun and adjectives that follow. These features are present in the head noun and the functional head above the NP, i.e., D0 (subsuming determiners, demonstratives and possessives) or the head numerals occupy (let us call it Num0). Quantifiers also have features of number, as illustrated by the agreement in number reflected on the noun and adjectives that follow. For reasons of simplicity, let us assume quantifiers are D-type or Num-type heads. This being so, a natural assumption deriving from the theory of syntax is that a relationship of feature checking is established between the heads D0/Num0 and N0, in a parallel fashion to the checking between T(ense) and the verb. The heads D/Num attract the categorial feature [N] or the set of formal features in N to check them against their own feature specifications, the same way T attracts the feature V (and the formal features in V). Indeed, T also attracts the feature D of an argument DP (the subject), triggering the raising of this feature to the specifier position of TP. If movement is overt, the whole DP raises together with D, in a pied-piping fashion (cf. Chomsky 1995). Then, the agreement relationship is manifested by raising the N to D or Num, overtly or at LF, as Chomsky (1995) argues, following Longobardi (1994).xiv Alternatively, another possibility is that this agreement relationship is manifested by raising of NP to Spec,DP, in the same way a subject in Spec,TP enters into agreement with the verb in T. Whichever possibility is the correct one, the assumption coherent with the theory of syntax is that D/Num have features that attract the raising of N/NP. For the case of French only the set of formal features of N are involved in movement, not the whole category, because in French there is no evidence of overt syntactic movement of nouns or adjectives to D0/Num0 or Spec,DP/NumP. The structures in (55) and (56) illustrate this relationship. To account for

232

the cases in which there is a determiner and a numeral, the structure in (56b) would have to be assumed, plausibly with movement of the noun to Num (checking the feature [± plural]) and then to D: (55)

DP 3 D NP [± plural] ! [± femenine] N [± referential] [± plural] : [± femenine] ! [± referential] ! ! z------m

(56) a. NumP 3 Num NP [± plural] ! : N ! [± plural] z----m

b.

DP 3 D [± plural] [± femenine]

NumP 3 Num NP

[± referential] [± plural]

:

:

!

N

z-- [± plural] [± femenine] [± referential] ! ! z-----------m

! ! !

Thus, determiners, demonstratives, quantifiers and numerals represent instances of the feature chain D-N. The fact that they participate in obligatory liaison indicates that these heads are mapped at MS as forming morphosyntactic words with following nouns: (57)

D-N



[D N]ms-word

233

As for subject and object clitics, we may follow proposals that treat them as functional heads expressing P-features and Case features, more concretely as D0s (cf. Torrego 1993, Uriagereka 1992, among others, and most recently, Everett 1996). Subject clitics agree with the following verb in person and number, and also realize nominative Case. They are heads entering into a checking relationship with the verb (for checking of P-features) and with T (checking the [assign nominative Case] feature and the categorial [D] feature of T that requires the formal features of a subject to be in Spec,T). To be more specific, they are both heads and maximal projections, in the sense of Chomsky (1994, 1995), and can move to the specifier position of TP. In French checking is overt, as evidenced by the raising of the clitic out of the VP or vP. This checking relationship is represented in (58): (58)

TP 3 D0i T’ 3 T0/V0j VP 3 ti V’ ! tj

Object clitics could fall under the same approach. They would originate as D0s from their base generated position inside the VP, and raise to the specifier of the functional projection wich checks accusative Case (AgrDOP, vP, or whatever). Alternatively, we could argue that they are object agreement markers, occupying the head of AgrDOP, as convincingly argued by Franco (1991, 1993) for Spanish, or the head of a Clitic Phrase (cf. Sportiche 1992). These clitics are thus in an agreement relationship with the verb, which is specified for the features realized by the clitic. In this hypothesis, object clitics and verbs are assumed to form a complex head, by movement of the verb to the head position occupied by the clitic. We could adopt any of the previous proposals in

234

the literature that treat object clitics and verbs as forming a complex head, either by movement of the clitic to the verb or by movement of the verb to the clitic. We will not try to decide here which approach is more satisfactory, because it would take us too far afield and because it is still a matter of debate in the literature. It suffices for us that at least in the morphophonological level object clitics form an unbreakable sequence with verbs, that cannot be disrupted by any kind intervening material (adverbs or the like). Clitics are then the D heads in the feature chains T-D and V-D(Subj), V-D(Obj). They spell out the sets of formal features in D, and as such they are part of these feature chains straightforwardly. The chains are realized at MS as morphosyntactic words: (59)

T-D



[D T]ms-word

(60)

V-D



[D V]ms-word

The generalization is now that the functional heads D0 and Num0 are the ones triggering obligatory liaison, indicating that these heads reflect their association with a nominal head (i.e., the relation of checking the features contained in them) in the morphophonological component as well. The features in D0 and Num0 associate with the lexical heads inflected for the same features, forming a morphological or morphosyntactic close unit or domain, and this domain is interpreted as a domain of application of obligatory liaison. This is represented as in (61): (61) Domain of obligatory application of liaison: [

]ms-word

 [

]p

It is important to note the difference between subject clitics and nonclitic subjects. The former form phonological domains of obligatory liaison with following verbs, whereas the latter do not: (62) a.

Ils∩ont entrés dans la salle

235

they have entered in the room b.

Les hommes / ont fatigués. the men are tired

This contrast is due to the fact that clitics realize the head D, of the feature chains T-D and V-D(Subj), whereas full DPs involve more than this head (the head D and an NP). The morphosyntactic constituent comprises only the sequence formed by the heads that directly realize the feature chains. Prepositions, degree adverbs, copulas and auxiliaries, on the other hand, do not get mapped at MS the same way. Copulas and auxiliaries, as we said, are already instantiations of the chain T-V, and this chain is mapped as a verb which constitutes an independent morphological word: (63)

T-V



[V T]ms-word

Auxiliaries and copulas are self-contained T-V chains, in which the head T has attracted the categorial feature [V] in the auxiliary. The head T and the head V may enter into feature checking relationships with other heads (T-D, V-D), but not the chain T-V itself, which is already realized morphosyntactically by the auxiliary or a copula. An auxiliary does not enter into checking relationships with following elements. If someone were to push the hypothesis that there is a [V-V] or [V-v] chain that relates an auxiliary to a past participle, we need to point out that such a chain is not mapped as a morphosyntactic word at MS in French. It suffices to point out that the auxiliary may be separated from the participle by adverbs: (64) a.

b.

Ils ont arbitrairement accusé de complot tous les étudiants. they have arbitrarily accused of complot all the students ‘They arbitrarily accused all the students of conspiracy’ Jean était probablement parti avant votre arrivée. Jean had probably left before your arrival

236

Since auxiliaries do not form morphosyntactic words with following material, they are correctly predicted not to fall in the domain of obligatory liaison. It could be argued that copulas share P-features with their predicates. However, it is not clear that this fact implies that there is a chain linking the copula to the predicate. The predicate can be a past participle (cf. (65a)), a noun (cf. (65b)), or an adjective (cf. (65c)): (65) a. Ils sont fatigués. they are tired b. Marie est artiste. Marie is artist c. Ses copains sont ambitieux. their companions are ambitious

This implies positing three different feature chains for what is the same relationship of checking of P-features: V-V, V-N, V-A. This is not a desirable move. The correct analysis is that the checking relationship holds between the copula and the Pfeatures of D (subject), which are shared by the predicate. These features in D (subject) raise to Spec,T together with the rest of the set of formal features of D (subject). That is, the checking of P-features is expressed in the feature chain V-D(subject). Since no feature chain links the copula and the following predicate, they are predicted to be realized as different morphosyntactic words, and that therefore no liaison applies obligatorily between them. This is a correct prediction. Prepositions enter in checking relations with D; the [accusative Case] feature of D and the [assign accusative Case] of P are checked, by raising the formal features of D to Spec,P. However, in light of the absence of obligatory liaison between prepositions and following determiners or nouns we must conclude that prepositions are specified not to

237

form part of one morphosyntactic word with the determiner (and the noun). This is an arbitrary property of the feature chain P-D, or of French grammar: unlike the feature chains T-D or V-D, the feature chain P-D is realized at MS as two independent words:xv (66)

P-D



[P]ms-word

[D]ms-word

This mapping is one of the possible realizations of feature chains onto morphological or morphosyntactic constituency. As stated in chapter 1, all that the grammar requires is that feature chains be realized at MS into morphosyntactic constituents, whether as single morphological words formed of stems plus affixes, as single morphosyntactic words containing the nonaffixal spell-outs of the heads in the chain, or as separate morphosyntactic words. The options are parametrized within a language and across languages for each feature chain, and the choices must be taken as arbitrary properties of the chains themselves or of each language in question. As for degree adverbs, it is not clear whether they are heads (Abney 1987) or specifiers of adjectives. What is clear is that they do not participate in feature chains of the primary sort, in which categorial features, P-features, tense features and Case features are checked. In general, then, these heads will not be mapped as morphosyntactic words together with following heads. That is why they do not make obligatory liaison. The morphosyntactic word realizing a feature chain at MS is mapped as a phonological domain, where obligatory liaison applies. The result is the same as in Basque: heads realizing bundles of inflectional features forming morphological domains that are interpreted as phonological domains. Let us illustrate our analysis with several examples of derivations from syntactic structure to phonological structure via morphological/morphosyntactic structure, leading to the formation of phonological domains of obligatory liaison:

238

(67) a. Syntactic feature chain: b. Morphosyntactic constituency: c. Phonological constituency: d. French liaison

[D-N] [les amis]ms-word [les amis]p les amisxvi

(68) a. Syntactic feature chain: b. Morphosyntactic constituency: c. Phonological constituency: d. French liaison

[D-N] [cet effort]ms-word [cet effort]p cet effort

(69) a. Syntactic feature chain: b. Morphosyntactic constituency: c. Phonological constituency: d. French liaison

[D-N] [son opinion]ms-word [son opinion]p son opinion

(70) a. Syntactic feature chain: b. Morphosyntactic constituency: c. Phonological constituency: d. French liaison

[Num-N] [trois ingénieurs]ms-word [trois ingénieurs]p trois ingénieurs

(71) a. Syntactic feature chain: b. Morphosyntactic constituency: c. Phonological constituency: d. French liaison:

[V-D] [les écoute]ms-word [les écoute]p les écoute

(72) a. Syntactic feature chain: b. Morphosyntactic constituency: c. Phonological constituency: d. French liaison

[T-D] [ils arrivent]ms-word [ils arrivent]p ils arrivent

Notice that we re not dealing with morphological words understood in the traditional sense as complex forms composed of a stem plus affixes. Some of these functional categories are affixes or bound forms that attach to an adjacent stem or word, such as object clitics and subject clitics of imperative and interrogative constructions. But determiners, demonstratives, quantifiers, numerals, are not. They form morphosyntactic words with the nouns following them, instantiating one of the possible ways to realize syntactic feature chains. The association is done independent of affixation or incorporation, as is not triggered by prosodic considerations: polysyllabic numerals express the P-feature [± plural], which needs to be checked syntactically by the [± plural]

239

feature in N. The feature chain Num-N is mapped or realized at MS as a single morphosyntactic domain, which is then mapped as a phonological domain [ ]p. It is in this domain where liaison applies obligatorily. In this respect the structures we are presenting depart from the cases of Basque, where the relevant functional heads were forming complex X0s with other heads. This indicates that checking relations are represented as primitive relations of grammar, that is, as primitive chains, which are reflected morphophonologically as well as syntactically. To prove the point that it is the realization of inflectional features what is at stake, consider the minimal pair in (73) (Bernard Tranel, p.c.; cf. also Tranel 1981:215-216). Only when it is a numeral expressing the feature number does vingt-trois ‘twenty-three’ make liaison in the presence of a following vowel-initial noun. When it stands for an ordinal numeral, which does not express number, no liaison is found: (73) a. vingt-trois ∩ étudiants twenty-three students b. vingt-trois / octobre twenty-third October ‘the twenty-third of October’

The following examples from Quebec French illustrate cases of liaison applying with right-dislocated noun phrases across an intonational boundary, which shows that the nature of the relationship between α and β is not syntactic or prosodic in nature. (74a) is taken from Tranel (1992:279), and (74b) from Morin and Kaye (1982:305). These type of sentences are not accepted by all speakers, but Tranel (1992:274) cites them as possible in Parisian French as well. On the main syntactic differences between these constructions and other dislocated structures with the preposition de, see Tranel (1992:276-8). Liaison is indicated by writing the consonant before the noun:xvii (74) a. J’en ai un, n-elephant I-clitic have one, elephant

240

b. Je vous ai vu en donner deux, z-oies I saw you clitic give two, geese

The two parts composing these dislocated constructions are separated by an intonational break, indicated by the comma, and belong to two different intonational phrases, i.e., two different prosodic constituents. The main part of the construction is able to stand alone as a grammatical sentence, that is, it constitutes a separate syntactic constituent from the dislocated part.

2.6. Possible alternative: agreement relationships. There is another possible alternative way of accounting for the distinction between the domains of obligatory and optional liaison which is compatible in spirit with the analysis we have proposed so far. This alternative, explored in Elordieta (1997b), would be to treat obligatory liaison as a reflection of agreement relationships. Determiners, demonstratives, possessive pronouns, and numerals agree with the following nouns or adjectives in P-features. Prepositions, degree adverbs, auxiliaries and copulas do not agree with the following syntactic heads (adjectives, past participles and adjectives or nouns). The contrast which best reflects the correctness of this proposal is that between polysyllabic quantifiers and numerals on the one hand and polysyllabic prepositions, degree adverbs, auxiliaries and copulas on the other. Recall that liaison is found obligatorily (or is hard not to make liaison) after the former, whereas liaison is only optional and less frequent after the latter. In the illustrative examples in (75), for example, repeated from (54), vingt-trois ‘twenty-three’, trente-deux ‘thirty-two’ and plusieurs ‘several’ agree with their respective head nouns in the feature [+plural], as illustrated by the appearance of the plural morpheme realized as -s on the head noun: (75) a. vingt-trois[+pl]∩oignons[+pl] twenty-three onions

241

b. trente-deux[+pl]∩assauts[+pl] thirty-two rounds c. plusieurs[+pl]∩amis[+pl] many friends

One way to capture this correlation between agreement relationships and obligatory liaison is to claim that two words α and β enter in a liaison environment when word α is a functional head that agrees in one or more phi-features with word β. This rule would be defined as in (76): (76)

Domain of obligatory application of liaison: [α [+φ] β [+φ]] Object clitics would be accounted for assuming the object agreement hypothesis

of Franco (1993), or alternatively, by assuming that complex X0s are domains of liaison. Prepositions, degree adverbs, auxiliaries and copulas do not share feature specifications with other following heads. A copula and its predicate usually agree in number and gender, but there are examples where no agreement is displayed (cf. (77)). In contrast, it is impossible to find examples with lack of agreement between determiners, possessives, quantifiers and numerals and nouns or adjectives (Tranel, p.c.): (77) a. Jean et Paul sont venue et parti, respectivement (*venues, *parties) John and Paul have come(sing.) and left(sing.), respectively b. Jean et Paul sont bon et beau, respectivement (*bons, *beaux) John and Paul are good(sing.) and handsome(sing.), respectively

Nevertheless, if agreement were the only factor involved in liaison, there would be no way to explain why liaison does not apply in conversational style to the last consonants of plural nouns before an adjective that agrees in number and gender (see (78)

242

below, repeated from (25a)). Why is liaison obligatory between a determiner and a following noun or adjective, but rare or absent between a noun and an adjective? (78) a. b.

langues / étrangères languages foreign faits / historiques facts historical

In order to explain this restriction on nouns, reference to a lexical/functional distinction is crucial. That is, only functional heads realizing agreement features trigger obligatory liaison in Style I. Reference to this distinction, however, is only a description of the facts, and does not explain the grammatical factor(s) responsible for it. The analysis presented in section 2.5 provides an analysis of this asymmetry based on the notion of morphophonological reflex of feature checking relationships. Namely, determiners, demonstratives, quantifiers/numerals are the syntactic heads containing inflectional features, which need to be checked by a nominal head. This relationships is reflected morphologically as well by having these features (or more exactly, the morphemes realizing these features) form morphological domains with nominal heads. Since nouns are not functional heads that need to be checked/licensed, they do not enter into morphophonological domains with following lexical heads, including adjectives. To conclude our analysis of the domains of obligatory liaison, we point out that liaison applies obligatorily within a word, between a consonant-final prefix and a vowelinitial stem, for instance. Indeed, orthography reflects the occurrence of liaison, as final consonants of prefixes are written before vowel-initial stem, but ommitted otherwise. Observe the following examples (taken from Selkirk 1972:304) (79) a.

désassembler ‘to disassemble’ désinfecter ‘to disinfect’ cf. débloquer

‘to unblock’

243

b.

c.

défavoriser mésintelligence mésestimer cf. méfait mécompte

‘to disfavor’ ‘misintelligence’ ‘to belittle, reject’

sousestimer

‘to underestimate’

sous-alimenté cf. souligner

‘underfed’ ‘to underline’

2.7. Optional liaison. Now the question arises: how do we account for the high frequency of liaison with monosyllabic forms of prepositions, degree adverbs, auxiliaries and copulas? Moreover, what could be the reason why polysyllabic forms of these categories trigger no liaison or rare liaison? This behavior can be explained if we assume that these function words display prosodic properties of clitics when they are monosyllabic. It is a well attested fact across languages that clitics are function words, and quite regularly they are monosyllabic. If this assumption is correct, then we propose that cliticization forms a domain of optional application of liaison in French. In other words, these facts would lead us to posit the existence of the clitic group as a prosodic domain in French, where liaison applies optionally. This domain could be represented as in (80): (80)

Domain of optional application of liaison: p[α p[β

]]

α is a clitic (i.e., a monosyllabic function word), and β is a p-domain originating from an m-domain. That is, the clitic has a p-domain as its host, rather than a particular lexical head. The resulting structure is one of adjunction, rather than strict sisterhood

244

under one p-node, which would make prepositions, degree modifiers, auxiliaries and copulas indistinguishable from the functional heads discussed above, which do form one simple p[ ] domain, without boundaries, with lexical heads. Observe the following illustrative examples of (81): (81) a.

doute]] doubt



sans∩aucune doute

[très p[incommode]] very uncomfortable



très∩incommode

p [est p[allée]]



est∩allée

p [sont p[intelligents]]



sont∩inteligents

p [sans

p[aucune

without any

b. c.

p

has gone

d.

are

intelligent

Polysyllabic forms of prepositions, degree adverbs, auxiliaries and copulas are not clitics, and so liaison is not found as frequently or as naturally as with monosyllabic forms. 2.8. Prenominal adjectives. Different sources in the literature on French liaison state that the final latent consonant of prenominal adjectives is always pronounced. The judgments I obtained from my native informants, however, indicate that this generalization is not accurate. It looks more like only the most frequently used prenominal adjectives trigger liaison. First of all, not all adjectives may appear in prenominal position in French, and of those which can, only a small set sounds natural in prenominal position in conversational style. These adjectives are: petit ‘small, little’, gros ‘big’, grand ‘tall’, bon ‘good’, vieil ‘old’, long ‘long’, and a few other adjectives with suppletive forms (i.e. with fixed final consonants), such as bel ‘beautiful’, nouvel ‘new’, fol ‘crazy’, mol ‘half-hearted’.

245

Other adjectives are not normally used in prenominal position in colloquial French. Adjectives such as eminent, intelligent, or ambitieux ‘ambitious’, for instance, do not trigger obligatory liaison in prenominal position. Indeed, liaison is quite hard to get in these cases: (82) a. l’eminent/?avocat the eminent lawyer b. cet intelligent /? enfant this intelligent child c. un ambitieux /? ingénieur an ambitious engineer

We must conclude then that liaison is lexicalized for some adjectives. Observe also that except for petit, all of the adjectives listed above as making obligatory liaison are monosyllabic, a property that perhaps adds a clitic-like behavior to them.xviii

bad

officer

2.9. Summary. In this section we have offered an analysis of French liaison in conversational style of speech or Style I which provides further empirical support for the validity of the theoretical proposal expressed in chapters 1 and 2. This proposal argues that syntactic relationships of feature checking among heads are primitive relationships in grammar, which we named feature chains. These chains are entities that have grammatical relevance not only in syntax, which is the level where the operations of feature checking take place (overt syntax or LF), but also in the morphophonological component. The idea is that feature chains have to be realized as morphological or morphosyntactic entities in the level of Morphological Structure (MS). There are basically two possibilities of feature chain realization: a) that the heads in the chain form one single morphosyntactic word, irrespective of whether one of the heads is an affix or not; and b) that the heads in a

246

feature chain are realized as two independent morphosyntactic words. The main empirical evidence for a difference between these two possibilities is evidenced by the differences in phonological interaction between the heads. The morphosyntactic constituent formed by a feature chain is mapped as a phonological constituent at PF, and thus may be a domain for phonological rule application. When the heads forming the feature chain are mapped as separate words, however, they may not undergo any kind of phonological interaction, or at least in a lesser degree than when they form one morphosyntactic word. Whether a feature chain is mapped as a single morphosyntactic word or as two separate words are options subject to parametric variation among languages, depending on particular choices made by grammars, encoded as lexical specifications for each head or for each feature chain. In French, the feature chains D-N, Num-N, T-D, V-DSubj, V-DObj are specified to be mapped as morphosyntactic words, and these domains are interpreted as phonological constituents at PF, where liaison is specified to apply obligatorily. These chains are realized by the sequences formed by a determiner, a quantifier, or a numeral and a noun/adjective, a subject clitic and a following verb, and an object clitic and a verb. The heads in these domains are not affixes, except perhaps for clitics (see the discussion in section 2.5). This analysis can predict why other function words such as prepositions, degree modifiers, auxiliaries and copulas do not form domains of obligatory liaison with following words. Prepositions are specified not to form morphosyntactic words with the head it forms a feature chain with, i.e., the determiner. Degree modifiers do not intervene in feature chains with other heads (i.e., it is not apparent what kind of features they may check), and auxiliaries and copulas are already morphological realizations of feature chains in themselves (i.e., T-V), not with following heads. Finally, since two lexical words are not in a feature chain relationship, they are predicted not to form liaison environments.

247

Our proposal constitutes a satisfactory account of the differences among function words in their behavior with respect to liaison. Alternative analyses presented in the Prosodic Hierarchy Theory framework reviewed in section 2.4 cannot explain these differences, either because they treat all function words in a similar way, or because they attribute the contrasts to distinctions in syntactic structure which are nonexistent. The frequent occurrence of liaison with monosyllabic prepositions, degree modifiers, auxiliaries and copulas is attributed to the clitic-like nature of monosyllabic forms of these categories. This assumption finds support in the fact that monosyllabic function words are very often clitics in languages of the world. These categories form clitic groups by adjoining to adjacent heads or p-domains originating from m-domains. The clitic groups so formed are the domains of optional (frequent) application of liaison. Polysyllabic forms of these categories will not form clitic groups, and this is evidenced by the rare occurrence of liaison with these forms. One important contrast among function words that stands out in French is the one between polysyllabic forms of prepositions, degree adverbs, auxiliaries and copulas

and polysyllabic forms of

quantifiers and numerals. While the former rarely have their final consonants pronounced in conversational style, the latter still make obligatory liaison. This supports our analysis of liaison in terms of feature chains. Finally, it was shown that the occurrence of liaison with prenominal adjectives should be better understood as a case of lexicalization. This is because only a few adjectives may appear in prenominal position, and of those which may appear in that position, only a reduced group of monosyllabic adjectives triggers liaison in colloquial style. In our last section, we will discuss the phenomenon of Irish Initial Consonant Mutation.

3. Initial Consonant Mutation in Modern Irish.

248

3.1.

Introduction. Initial Consonant Mutation (henceforth ICM) is a common term used to refer to

the phonological alternations undergone by the initial consonants of words in certain morphologically and syntactically specified contexts. The range of phenomena classifiable as ICM is interesting for our purposes because these are phonological processes which are directly conditioned by morphological and syntactic factors, apart from the purely phonological. In some cases, the mutation environment is not phonologically transparent, in the sense that the changes introduced in the initial consonant do not seem to derive from the phonological features of the trigger and the target. Bantu languages and Celtic languages display perhaps the most studied cases of ICM. In this chapter we will concentrate on Modern Irish, where the underlying initial consonants of verbs, nouns and in some cases adjectives undergo two forms of mutation, lenition and eclipsis, depending on the syntactic and morphological environment in which these verbs, nouns and adjectives occur. Previous analyses of Irish ICM treated the phenomenon as determined by an arbitrary list of particles, lexically specified to trigger lenition or eclipsis on the initial consonant of the following word. This position resulted from the inability to establish natural generalizations in terms of morphological or syntactic properties of the triggering contexts, and from the fact that the phonological changes were not synchronically transparent. The mutations in the initial consonants did not seem to be directly effected by the phonological feature matrices of the triggering particles, since the word-final segments that triggered mutation at earlier stages of the language were lost. However, closer inspection of the syntactic structure of Irish unfolds underlying regularities in mutation contexts. Verbs in the past tense undergo lenition in their initial consonants, and eclipsis when preceded by an element occupying the complementizer position. Nouns mutate their initial consonants when preceded by elements occupying the

249

determiner position (i.e., D0) in the syntactic tree, such as determiners themselves, possessive proclitics and the head nouns in construct state nominals. We capitalize on the generalizations found by Duffield (1991, 1995, 1996, 1997) to propose an analysis along the lines suggested in this dissertation. We claim that although the phonological changes induced are idiosyncratic to the particles triggering them, still the contexts are systematically constrained in such a way such that they are reduced to functional heads realizing inflectional features of tense, agreement and Case and the lexical heads morphosyntactically associated to them. It will be argued that the relationship between these elements is one of feature checking, in which the lexical heads license the functional heads (by the operation of feature checking) in the sense explained in Chapters 1 and 2 and in the discussion in section 2 of this chapter, and that by virtue of this relationship these heads form one morphological constituent. This unit is then mapped as a phonological constituent at PF, where mutation is specified to apply between a particle α and a lexical head contained in the same m-constituent. The notion of feature chain realization will serve us to explain why mutation was reduced (from a possibly wider range of contexts) to those involving a functional head and a lexical head. It will also help us to provide an account of the distinction between functional heads that trigger mutation and those nonlexical elements which do not, such as negative specifiers. These are not heads realizing morphosyntactic features, and thus they do not enter into relations of morphosyntactic licensing (i.e., feature checking) with other elements. Following our hypothesis, these elements will not be part of the same domain in the morphophonological component, where these relations may be visible. The rest of the discussion on ICM is organized as follows: in section 3.2 we provide a historical background to the origin of ICM in Irish. We also mention the analysis proposed by Grijzenhout (1995) in terms of deletions of aperture positions and associations of sonorant voicing nodes. In section 3.3 we present a distribution of the basic environments where ICM is found in Modern Irish. In section 3.4 we criticize

250

previous proposals of the problem. Section 3.5 introduces the generalization discovered by Duffield (1991, 1995) that serves as the basis of our analysis, which we finally present in section 3.6.

3.2. Historical background on ICM in Modern Irish. In Modern Irish there are two major mutations, known as lenition (also called aspiration) and eclipsis (also called nasalization). Lenition is a process by which stops spirantize and fricatives either turn to [h] or else delete. By eclipsis, on the other hand, unvoiced consonants voice and voiced ones nasalize. In the table below we represent the effects of the rule, using IPA symbols. It is important to mention that Irish consonants come in contrastive palatalized and velarized varieties. We have marked the palatalized varieties with the apostrophe (’), following modern Irish phoneticians’ device. In the leftmost column we have written the unmutated consonants; the column in the middle corresponds to the lenited consonants, and the third column contains the eclipsed consonants. Immediately before each column with the phonetic representation of the consonants, we have included the orthographic representation of these consonants in each case (from Rotenberg 1978, who cites Dillon and Ó Cróinín 1961, Oftedal 1962, and Mhac an Fhailigh 1968, among others, as sources of general information on Irish ICM; cf. also Ó Cuív 1986):

251

(83)

unmutated Orthogr. p t c b d g f s m

lenited

eclipsed

Orthogr. p p’ t t’ k k’ b b’ d d’ g g’ f f’ s š m m’

ph th ch bh dh gh fh sh mh

Orthogr. f f’ h h’/ç x ç wv γ j γ j φxix h h’ wv

bp

b b’ dt gc mb nd ng bhf __xx __

d d’ g g’ m m’ n n’ N N’ w v _ _ _ _

These changes were phonologically motivated at earlier stages of the Irish language. In Archaic Irish (300-600 AD), intervocalic lenition became an active phonological process which transformed plosives into fricatives when flanked by vowels, or by a vowel and a sonorant consonant.xxi This change affected only short plosives; long or geminate plosives underwent loss of length at a later stage, becoming short plosives, when lenition was no longer active as a phonological rule (cf. Kelly 1978). The following chart depicts the presumed consonant alternations in Archaic Irish standardly assumed (from Grijzenhout 1995, who cites Kelly 1978; cf. also Thurneysen 1946): (84)

Common Celtic Irish Intervocalic Lenition Irish Loss of Length

p t k pp tt kk fθx p t k

b d g bb dd gg vð◊ b d g

m w

The fricatives /f/, /θ/ and /x/ did not exist as underlying consonants in Archaic Irish. Grijzenhout (1995) disputes the assumption that a voicing contrast existed among plosives in Archaic Irish. She suggests instead that the contrast was among aspirated and unaspirated plosives. She thus provides the following chart (p. 63): (85)

Common Celtic Irish Intervocalic Lenition

ph th kh f θ x

pph tth kkh

p t k pp tt kk f θx

m w

252

ph th kh

Irish Loss of Length

p t k

That is, unaspirated and aspirated short plosives changed into fricatives, and the nasal consonant m changed into the labial fricative w. Archaic Irish intervocalic lenition took place word-internally, affecting stem-final consonants before inflectional affixes, but also across word-boundaries, affecting word-initial consonants after inflectional suffixes. Although several analyses of ICM have been proposed in the literature, in linear and nonlinear frameworks, none of them has solved satisfactorily the puzzle that ICM poses. Grijzenhout (1995) criticizes these past proposals, and analyzes these changes adopting Steriade’s (1993) aperture position theory and Rice’s (1993) sonorant voicing theory to explain lenition and nasalization, respectively. She suggests that plosives change into fricatives by losing their Aperture poition indicating the closure phase (with or without a specification for aspiration). The outcome is an Aperture position for release without a contrast for aspiration or voicing:xxii (86) a. aspirated plosive  fricative A0 Arel ! [aspirated] b.



Arel

unaspirated plosive  fricative A0 Arel 

Arel

In late Archaic Irish, a rule of Apocope deleted final vowels and final -VC sequences, although the mutating properties remained. The minimal pair in (87) exemplifies the applications of lenition, loss of length and Apocope word-internally: (87)

(h)

a. *kat us b. *kattos

Lenition

 

loss of length

*kaθus vacuous

 

Apocope

vacuous *katos

 

[kaθ] cath ‘battle’ [kat] cat ‘cat’

253

The example in (88) illustrates the application of lenition across words. The initial consonant of the nominative masculine singular *kattos gets lenited because it is preceded and followed by vowels: (88)

Common Celtic *esjo kattos



Early Archaic Irish

......

Modern Irish

*esjo xattos

...... 

a chat

[a xat]

‘his cat’

The interesting aspect of Irish mutation is that when final vowels or -VC sequences were lost by Apocope, the mutations in initial consonants were maintained (cf. Oftedal 1962, Ternes 1986). An example is presented in (89), for the genitive singular noun maq(q)i/mac(c)i ‘of the son’, which evolved into meic in Old Irish. After the loss of the -i of the genitive, the mutation triggered by this vowel was maintained into Old Irish: (89)

Archaic Irish Old Irish

maq(q)i/mac(c)i son-gen.sg.

>

Old Irish

meic son-gen.sg.

meic Choncobar son-gen. Concobar ‘Concobar’s son’

In fact, the environments in which initial mutations were found or absent were instrumental in reconstructing the original existence of final vowels or -VC sequences in Archaic Irish, for which no documented records existed. The following paradigm illustrates the relationship between the case endings of the masculine noun *wiros ‘man’ of Archaic Irish and the mutation environments maintained in Old Irish, even after the loss of case endings by Apocope: (90) Nominative Genitive Dative

Archaic Irish *wiros  *wirí  *wirú 

Old Irish [f’er] [f’ir’] + initial lenition [f’iurw]

254

Evidence for initial nasalization (eclipsis) in Archaic Irish is less available. Its effects are found solely in Old Irish data, where some word-final nasals escaped Apocope. But in any case, it seems clear that eclipsis must have started in Archaic Irish, around the same period as intervocalic lenition. Like lenition, initial nasalization was maintained in Old Irish even after word-final nasals were lost. In this period, unaspirated plosives were prenasalized. Grijzenhout (1995) also offers an analysis for Old Irish nasalization, based on Rice’s (1993) proposal of the sonorant voicing (SV) autosegmental node. She argues that nasalization involves insertion of a SV node in the Aperture position for closure, i.e., A0:xxiii (91)

unaspirated plosives  A0 Arel



prenasal plosives A0 Arel ! SV

Recall that in Modern Irish these are the changes induced on initial consonants: (92) a.

Lenition plosive fricative

 

fricative deletion, or laryngeal sound (s -> h)

b. Nasalization voiceless plosive voiced plosive

 

voiced plosive sonorant

For lenition, Grijzenhout’s analysis is again that the A0 position is deleted; this leaves only the Arel position, which explains why a fricative is the output: (93) a. plosive  fricative A0 Arel b.



Arel

fricative  zero

255

Arel



{

For nasalization in Modern Irish, this author argues that a SV node is inserted and associated to the Arel position: (94) a.

voiceless plosive  A0 Arel

b.



voiced plosive A0 Arel ! SV

voiced plosive



sonorant

A0 Arel ! SV



A0 Arel SV

This concludes our presentation of the phonological analysis of ICM. We have chosen to merely sketch it, since it is not our main concern in this dissertation. For more details, the reader is referred to Grijzenhout (1995), and references therein. What is relevant for our purposes is defining the contexts of application of ICM. They are well described for Modern Irish, but unfortunately not for Archaic Irish. This is a period from which only a few stone inscriptions can be used as testimony of the language, and to my knowledge there is no description of the contexts of application of ICM in Archaic Irish. For Old Irish we find almost the same problem, although Jackson (1953) and other Celtic scholars mention the “close speech-group” as the context where ICM occurred, always triggered by some proclitic ending in a vowel or a nasal. Not all word-initial consonants following word-final vowels or nasals mutated. Determiners, numerals, possessive pronouns, prepositions and past tense morphemes are cited as triggers of ICM. For a list of environments where lenition and nasalization can be found, see Thurneysen (1946:142-150). What all these authors mention is that from the loss of

256

word-final sequences, the conditioning of these changes was not phonological anymore, but grammatical. That is, certain particles and grammatical formatives trigger these mutations on the initial consonant of the following word (a noun, an adjective, or a verb). We provide examples of mutation phenomena in these contexts followingly.

3.3. Distribution of ICM in Modern Irish. 3.3.1. Verbal contexts. We can provide the following examples illustrating mutation in verbs, for which we have taken the verb cuir ‘put’ as a model (taken from Duffield 1991, based on the Northern dialect of Gweedore, Co. Donegal). Those in (95) are cases of eclipsis, and those in (96) are cases of lenition. For both sets of examples, we have underlined the mutated consonants: (95) a. An gcuireann sí grá isteach ina gcroí? Q put she love inside in-their heart ‘Does she instill love in their hearts?’ b. Nach gcuireann sí grá isteach ina gcroí? neg-Q ‘Doesn’t she instill love in their hearts?’ c. Creidim go gcuireann sí grá isteach ina gcroí. comp ‘I believe that she instills love in their hearts’ d. D’fhiafraigh sé an gcuireann sí grá isteach ina gcroí. comp ‘He asked whether she instills love in their hearts’ e. Creidim nach gcuireann sí grá isteach ina gcroí. neg-comp ‘I believe that she doesn’t instil love in their hearts’ f.

D’fhiafraigh sé nach gcuireann sí grá isteach ina gcroí. ‘He asked whether she doesn’t instil love in their hearts’

(96) a. Do chuir síad leis an ól

é.

257

past put they with the drink him ‘They drove him to drink’ b. Nío-r chuir síad leis an ól é. neg-past ‘They didn’t drive him to drink’ c. A-r chuir síad leis an ól é. Q-past ‘Did they drive him to drink?’ d.

Ná-r chuir síad leis an ól é. neg-Q-past ‘Didn’t they drive him to drink?’

e.

Creidim gu-r chuir síad leis an ól é. I-believe comp-past ‘I believe that they drove him to drink’

f.

Creidim ná-r chuir síad leis an ól é. neg-comp-past ‘I believe that they didn’t drive him to drink’

g. Níl fhios agam a-r chuir síad leis an ól é. Q-past ‘I don’t know whether they drove him to drink’ h.

Níl fhios agam ná-r chuir síad leis an ól é. ‘I don’t know whether she didn’t drive him to drink’

i.

Cé a chuireann grá isteach ina gcroi? who aL instils love in their hearts ‘Who instils love in their hearts?’

As we can see from these data, there are certain particles that trigger lenition, and others that trigger eclipsis. Among the first are do, níor, ar, nár, gur, a, and among the second are an, nach, go. The synchronic phonological shape of these morphemes indicates that nowadays the triggering context of mutation in verbal contexts cannot be identified phonologically. Of the five leniting particles, only two end in a vowel (the past tense marker do and the particle aL, on which see below), and of the eclipsing particles, only one ends in a nasal (the interrogative complementizer an). The ancestors of these

258

particles did end in the appropriate contexts triggering mutation in Archaic Irish, but these final sequences were lost later, and the descendants of these particles maintained the mutating properties, as a morphological feature. From Old Irish on, the mutating properties of these and other particles became phonologically arbitrary features.

3.3.2. Nominal contexts. 3.3.2.1. Determiners. In nominal contexts, nouns and adjectives also get mutated in the presence of a definite determinerxxiv, a possessive proclitic, and certain prepositions. When preceded by a determiner, what kind of mutation is produced depends on the P-features of the entire Determiner Phrase. Like the particles introducing clauses presented in the previous section, the ability to induce mutation was inherited by these determiners from their ancestors, which ended in the appropriate word-final sequences triggering phonologically motivated mutation. After these final sequences were lost, the mutation-inducing properties were maintained as idiosyncratic features of these elements. Masculine singular nouns in the genitive case get their initial consonant lenited, whereas feminine singular nouns get their initial consonant lenited in the nominative and accusative case (cf. (97) below). This is because the determiners inflected for these case, gender and number specifications ended in a vowel. In Modern Irish, however, these determiners end in a nasal (i.e., the homophonous an). As an example of the irrelevance of the phonological environment for mutation, suffice it to state that if the phonological environment were active in Modern Irish, this determiner should trigger eclipsis.

259

Initial onsonants of both masculine and feminine nouns in the nominative case get eclipsed when they are preceded by a plural determiner (cf. (98)). Again, the plural determiner is na, which does not end in a nasal. Its ancestor did. Like with the singular dteterminer, the paradoxical state that has been reached in Modern Irish is that if the plural determiner were to trigger mutation based on its final segment, it would cause lenition on the following noun, instead of the observed eclipsis. The following table illustrates mutation on a masculine and a feminine noun: (97)

cat ‘cat’ (masculine)

clann ‘family’ (feminine)

singular nom. acc. gen.

an cat an cat an chait

singular nom. acc. gen.

plural nom. acc. gen.

na cait na cait na gcat

plural nom. acc. gen.

an chlann an chlann na clainne

na clanna na clanna na gclann

3.3.2.2. Possessive pronouns. The type of mutation that will surface on a noun when preceded by a possessive proclitic also depends on the number and gender specifications of the possessive proclitic; singular proclitics trigger lenition and plural proclitics trigger eclipsis: (98)

Singular

Plural

1

mo theach ‘my house’

ár dteach ‘our house’

2

do theach ‘your house’

bhur dteach ‘your house’

3 masc

a theach ‘his house’

a dteach ‘their house’

260

3 fem

3.3.2.3.

a h-ínionxxv ‘her daughter’

Prepositions. Most common prepositions trigger lenition when followed by indefinite nominal

complements, that is, nouns without an overt determiner. They trigger mutation on whatever appears to their right, independent of its syntactic category (examples taken from Duffield 1996): (99) a.

Thit sé de chrann fell he from tree ‘he fell from a tree’

b.

trí shioc through frost ‘through frost’

c.

do Mháire for Mary ‘for Mary’

d. Dá dtráchtainn [ar [ghníomh a dhéanamh]] if I-should-speak on deed ptc do-VN ‘If I should speak of performing some deed’ e. ag argóint [faoi [chéard a ba cheart domhsa a dhéanamh]] prog argue about what ptc be right to-me ptc do-VN ‘arguing about what I should do’

Three prepositions (chomh, le, go) trigger H-prefixation, a different kind of mutation, which we will discuss in section 3.5. 3.3.2.4. 3.4. Previous proposals.

261

We are not going to present here a review of the different analyses proposed to account for the phonological changes characteristic of ICM. Several attempts have been made, most of them faulty, and we refer the reader to Grijzenhout (1995) for a comprehensive criticism of past proposals and an alternative analysis. The important point to bear in mind for our purposes is that very few attempts have been made to come up with generalizations concerning the mutation contexts. The apparently arbitrary character of mutation, with phonological changes that do not follow from the neighboring segments, have led researchers to believe that such a task is impossible, that ICM is chaotic and unsystematic. Only a description of the main mutation contexts is foud in the traditional literature (cf. Ó Siadhail 1989, and especially Christian Brothers 1990 for lists of environments and examples). In the next paragraphs we are going to review briefly the few studies that have focused on this topic from a generative point of view.

3.4.1. Morphophonological analyses. Many authors have distinguished two kinds of ICM in the Celtic languages: triggered and inherent, or, in Oftedal’s (1962) terms, projected mutations and incorporated mutations. The first type refers to mutations triggered by specific morphological or syntactic forms or constructions, always preceding the word which undergoes mutation. The second type refers to cases in which mutation is observed with no triggering element being present. These cases have been considered to be part of the number, gender, case, definiteness, and tense-marking systems. The same way as ablaut and change in quality of final consonant can be allomorphs of the [+past] morpheme, the grammatical marking of the noun or adjective (for instance, the bundle of features [+gen, +masc, +sing, +def] may be morphologically signaled by a lenited initial.

3.4.1.1.

Triggered mutation.

262

According to the criteria introduced above, all the examples in (93)-(94) are cases of mutation triggered by an idiosyncratic property of the particles which immediately precede the verb or a noun. These particles are merely listed in the lexicon as having this property of triggering phonological changes on the initial consonant of elements following them (Hamp 1951, Oftedal 1962, Rotenberg 1978, Thomas-Flinders 1981, Massam 1983, among others). Rotenberg (1978) suggests rules like the following, where the elements triggering mutation include an abstract morphological feature [+L] or [+N], standing for lenition or nasalization (eclipsis), respectively (cf. Massam 1983 for a similar approach):xxvi (95) a.

b.

+cons -voice α cor +cont +obst α voice ->

-> α nas +voice

+cont -sonorant - α cons < α unit>

/ [+L] ___

/ [+N] ___

3.4.1.2.

Inherent mutation. Rotenberg (1978) and Massam (1983) treat the cases represented in (90)-(92) as inherent mutation, that is, they claim that mutations in these cases appear as reflexes of a morphological characteristic of the lexical item. Rotenberg and Massam analyse the mutation in the verbal contexts in (91) as a word formation rule which prefixes the past tense morpheme to a verb. They adopt this analysis in view of the fact that the realization of the past tense morpheme do preceding a verb is only optional: (98)

(Do) chui sí. past put she ‘she put’

Rotenberg suggests that lenition on the verb is part of the morphological marking of the past tense, and thus it is a lexical word formation process. Massam accounts for this mutation case positing that the past tense morpheme (at least in root declarative clauses) may be realized as a bundle of phonological features which form a floating autosegment, and this autosegment links to the initial consonant of the verb, causing

263

lenition. She only considers cases without the morpheme do, and thus she also takes this morphophonological process to take place in the lexicon. As for the cases in (92), Rotenberg argues that they have nothing to do with the preceding word (i.e., the determiner), but rather they are part of the morphological realization of the Case, number and gender features of nouns and adjectives. He bases his position on the observation that nouns get their initial consonants lenited when genitive, masculine, singular and definite, even in the absence of a definite determiner: (99) a.

b.

capall Shéamais horse Séamas-gen-masc-sg-def ‘Seamas’ horse’ mac Sheáin son Seán-gen-masc-sg-def ‘Sean’s son’

The same would hold of adjectives qualifying a noun which is feminine, nominative and singular, i.e., when marked that way themselves (adjectives agree in case, gender, and number with the noun they accompany): (100) an bhean mhór shaibhir the woman-nom-fem-sg big-nom-fem-sg rich-nom-fem-sg ‘the rich big woman’ Massam (1983) adopts a different approach to these cases, considering them instances of triggered and inherent mutation at the same time, for the lenition phenomena. She assumes that it is the determiner that has the property of inducing lenition on the following noun, and this noun on a following adjective. Thus, the same way as she posited a floating autosegment as a suffix of trigger words, she posits a floating autosegment as a kind of suffix in the underlying representation of the determiner and the noun, and these autosegments link to the feature matrix of the initial consonant of the following word. That is, the autosegment from the determiner is linked to the initial consonant of the noun, and the autosegment of the noun links onto the initial consonant of the following adjective, like in a chain operation. The motivation for proposing such an operation comes from the existence of what she calls the Case Realization Generalization, stated in (101): (101) Case Realization Generalization item

If Case is realized by lenition, the leniting segment must appear on every lexical in the phrase which has this Case.

264

That is, Massam interprets the mutation processes in DPs as a phonological reflect of agreement in Case among the elements inside DPs. Summarizing our discussion in this section, we have seen that ICM in Irish cannot receive a satisfactory explanation in pure phonological terms; the changes are phonologically unmotivated, and the syntactic environment preceding the element undergoing mutation has to be considered. Two types of ICM phenomena have been distinguished in the literature: triggered and inherent. The first type corresponds to those cases in which a particle or grammatical formative has the property of triggering idiosyncratic mutation (either eclipsis or lenition) on the initial consonant of the following verb or noun. The second type corresponds to those cases in which mutation is not dependent on the presence of any particle or grammatical formative, and is rather a phonological reflex of certain bundles of morphological features carried by the verb or noun. In the following section we will present a different proposal, which argues for a syntactic determination of mutation contexts, and which will prove empirically and explanatorily superior to these previous analyses.

3.5.

ICM as syntactically determined. Duffield (1991, 1995, 1996) departs from the view that distinguishes two kinds of ICM in Irish, and claims that all mutation is determined syntactically, not lexically. He argues that a mere listing of particles triggering mutation misses an important generalization: that a lexically filled Tense node triggers lenition on the following verb, and that complementizers and interrogative particles trigger eclipsis. Duffield captures this in what he calls the Mutation Trigger Constraint: (102) Mutation Trigger Constraint a. A lexicalized C0 node creates an Eclipsis domain. b. A lexicalized T0 node creates a Lenition domain.

265

3.5.1. Complementizers and eclipsis. The elements lexicalizing the Complementizer head node in Modern Irish are the proclitic particles in (90), an, nach, go, and they all trigger eclipsis. Duffield provides several convincing arguments to demonstrate that these elements are located in C0 at sstructure (i.e., before Spell-Out), but we will not go into the details of his analysis here. He addresses the question of whether these particles are all base-generated in C0, or whether they originate somewhere else and are subsequently moved to C0 in the course of the derivation. He assumes that interrogative particles are either generated in C0 or they are generated under Infl or some other functional projection below CP and subsequently raise to C0. Then, instead of maintaining that it is an arbitrary lexical property of go, an, and nach that they trigger eclipsis, Duffield concludes that it is much more revealing and economical to say that a lexicalized C0 node creates an eclipsis domain. That it is the syntactic context, rather than the lexical item, which is responsible for the obligatory mutation effect.

3.5.2. Tense and lenition. The elements that lexicalize the T0 node in the examples in (91) are the preterite allomorphs do and the clitic -r (leaving (91i) aside for the moment). The first one appears when there is no more material preceding the verb, that is, in root declarative clauses (cf. (91a)). In nearly all dialects the particle do gets deleted with consonant-initial verb stems (except /f/-initial)xxvii, and thus the only way to distinguish the preterite stem from a bare stem or the imperative is through lenition on the initial consonant of the preterite stem, since none of these forms have tense suffixes attached to them, unlike the other tenses. Before vowel-initial verb stems, however, it has to surface as a prefix on the stem, dropping the vowel o:

266

(103) a.

(do) phóg si

é.

(past) kiss she him ‘She kissed him’

b.

*(d’) fhag siad an mhonarcha ar a sé a chlog. past leave they the factory

at

six o’clock

‘They left the factory at six o’clock’

c.

*(d’) ith sé an bia. past eat he the food ‘He ate the food’

The second past allomorph, -r, appears as a clitic of a complementizer or negation (cf. (91b-h). Duffield argues this cliticization to be phonological, not syntactic. The distribution of this clitic depends on the existence of an element in Comp, not on the verb stem in Irish.xxviii

Duffield assumes that in Irish the verb does not raise to Tense,

which he locates above the AgrP projection, and that the subject stays in-situ, in the specifier of VPmax. This derives the VSOX word order. We illustrate the syntactic structure of Irish clauses proposed by Duffield (1995), slightly simplified: (104)

CP 3 Spec C’ 3 C0 TP 3 T0 NegP

267

3 Neg0 AgrSP 3 AgrS0 VPmax 3 DP V’ 3 AspP 3 Asp0 VP 6

A context of lenition is exemplified below: (105) a. b.

Ar chuir síad leis an ól é. (= (91c)) CP 3 Spec C’ 3 C0 TP ! 3 a T0 AgrSP ! 3 -r AgrS0 VPmax ! to chuiri DP V’ ! 6 síad

ti leis an ól é

Another element which can lexicalize the T0 node and thus create a domain for lenition is negation. This case is particularly relevant for Duffield’s analysis, because the initial consonant of the verb following negation gets lenited when the verb is inflected for any tense, not just the past tense: (106) Ní chuireann síad leis an ól é. neg put they with the drink him ‘They don’t drive him to drink’

268

Duffield assumes that sentential negation constitutes a separate projection of its own, between TP and AgrSP, and that its head must raise to take scope over any and all other functional nodes. When CP is projected, Duffield argues that negation morphemes are base-generated in C0. To the mutation environments determined by lexically realized T0 and C0, Duffield adds another context of syntactically determined mutation: the one triggered by a lexically realized Aspect node, Asp0: (107) A lexicalized Asp0 node creates a lenition environment

This context is different from the ones above, because it is clause-medial, rather than clause-initial. It occurs in non-finite constructions in which the direct object is preposed, to a position which is still open to debate among researchers working on Irish, but which Duffield chooses to identify as Spec,AspP. Observe the following examples: (108) a. [ í an teach a cheanach] a ba mhaith liom. her the house ptc buy-VN ptc were good with-me ‘I would like her to buy the house’ b.

An fear atá sí a phósadh ... the man ptc-be she ptc marry-VN ‘The man (that) she is marrying...’

In these sentences, the particle a lexicalizes (i.e., overtly realizes) Asp0, and as a consequence, the non-finite verb (glossed as VN, that is, ‘verbal noun’) has its initial consonant mutated. This is the schematic representation of these sentences:

269

(109)

AgrP 3 Agr0 VP1 3 Spec V’ 3 1 V AspP 3 Spec Asp’ 6 3 Asp0 VP2 3 2 V NP 6 tj í an teachj a cheanach ... atá sí tj a phósadh tj

The most important aspect of Duffield’s work is that it eliminates all cases of inherent mutation, arguing that lenition is not a lexical property of verbs in the past tense, as Rotenberg and Massam suggested. Rather, as stated in clause (174b) of his Mutation Trigger Constraint, he claims that it is a phonological property of a lexically filled Tense node. Copulative constructions in which the copula is in the past tense and the following predicate gets its initial consonant mutated is a good piece of evidence in favor of this proposal. Although no agreement has been reached as to the exact position that the copula occupies in the clause, Duffield assumes that it sits in T0. Interestingly, nouns and adjectives that follow the copula get their initial consonants mutated:xxix (110) a. Ba mhuintéoir é be-past teacher him ‘He was a teacher’ b. Nío-r mhuintéoir é neg-past teacher him ‘He wasn’t a teacher’ c. Creidim gu-r mhuintéoir é believe-1sg comp-past teacher him ‘I believe he was a teacher' d. a-rbh iascaire é?

270

Q-past fisherman him ‘Was he a fisherman?’

This seriously weakens all analyses of inherent mutation (Rotenberg 1978, Massam 1983 and almost all traditional accounts), at least in what respects mutation triggered by a past morpheme. If lenition were a property of elements inflected for the past tense, we cannot explain the lenition process undergone by the initial consonants of lexical elements such as nouns and adjectives, which cannot be inflected for the past morpheme or any tense morpheme. The copula or its substitutes lexicalize the T0 node and this triggers lenition on the following element. I will add another counterexample to the inherent mutation hypothesis, to show the necessity to refer to a lexically realized trigger. In Construct State Nominals, the head noun appears initially, followed by possessor NPs. The head noun appears without a determiner. Feminine nouns in nominative/accusative case are crucial, because by the inherent mutation hypothesis they would be predicted to have lenited initial consonants. However, they do not, and appear in their base form: (111) máthair an fhir mother-fem det-masc-gen man-masc-gen ‘The man’s mother’ However, a following adjective does get lenited: Sheáin. (112) clann mhaith family-gen good-gen Sean-gen ‘Sean’s good family’

This connects us to nominal contexts. In cohesion with his analysis of verbal mutation, Duffield states the following condition on nominal mutation: (113) Lexicalized D0 triggers lenition, unless otherwise specified.

271

This means that lenition is the elsewhere case, and determiners and possessive proclitics which are lexically specified to trigger eclipsis have priority, so to speak, and override the occurrence lenition. This is the working of the well known Elsewhere Principle. Now, all this discussion brings us to our main concern, which is how to analyze ICM in Modern Irish in our framework. We build on the excellent job done by Duffield, who has broken with centuries of tradition in Celtic studies, and has shown that ICM is not chaotic or unworthy of an effort to find linguistically meaningful generalizations. His main objective in dealing with ICM in Irish was to use the identification of mutation triggers as a tool for identifying the syntactic position of those triggers and the form of syntactic representations. We step on his work now, and provide a formalization to the generalizations discovered by Duffield in terms of our hypothesis of feature chains being mapped as morphosyntactic constituents. Otherwise, his generalizations will remain just that, interesting and intriguing generalizations. The question to be tackled is why those functional categories trigger mutation on the following heads. In other words, why have the mutation contexts been reduced to these cases in Modern Irish? We have the opportunity to go one step further and show that the contexts of application of ICM are closer to recurrent contexts of application of other processes in the languages in the world than hitherto assumed. We maintain that ICM is not best treated as a precompiled rule of phrasal phonology, against Hayes’s (1990) suggestion. The phonological changes are synchronically arbitrary, and the question of which particles trigger which type of mutation seems idiosyncratic to the particles in many instances. But the generalizations unveiled by Duffield, of which for limits of space we only offered a sketch, point to a significant role of syntax in determining the contexts where mutation applies.

3.6. ICM as an effect of morphosyntactic mapping of feature chains.

272

In Modern Irish, ICM is a reflection of the grammatical process of the morphosyntactic mapping of feature chains being at play. Mutation applies to initial consonants of nouns following determiners, possessive pronouns and prepositions, and to initial consonants of verbs following complementizers and tense and aspect markers. Although synchronically these morphemes do not end in the segments that induced the phonological changes in Archaic Irish, they have maintained the phonological property of causing mutation on the following consonant. The crucial observation is that these particles are all functional categories which are followed by their lexical complements. As we have had the opportunity to see in this dissertation, this is a recurrent trend among languages. We will argue that mutation is a phonological process which is triggered by all functional categories which are syntactic heads in Modern Irish, and that mutation is specified to apply in an m-constituent, or more precisely, in the phonological domain formed by a functional category and the lexical head it merges with at Morphological Structure. This merging is an instantiation of one of the possible mappings to MS of feature chains. Irish is the most radical case, since it maps all feature chains as morphosyntactic domains, including the chain P-D, which in French was not mapped as one constituent. We represent this process as in (114): (114) a. D-N b. P-N c. T-V d. C-T

   

[D N]ms-word [P N]ms-word [T V]ms-word [C T]ms-word

These constituents are then interpreted in the phonological component as phonological domains, where the operation of mutation takes place. Let us illustrate this analysis with one example from each type of mutation, starting with eclipsis in verbal contexts:xxx (115) Base form:

go gcuireann that put

273

a.

Syntactic constituency

CP [ C[go TP [ AgrP[ Agr[cuireann

b.

Morphological constituency

c.

Phonological constituency

p[go

cuireann]

d.

Mutation (eclipsis)

p[go

gcuireann]

[go cuireann]ms-word

For tense morphemes: (116) Base form:

ar cuir Q-past put

a. Syntactic constituency

TP [T[ar [AgrP[ Agr[cuir...

b. Morphological constituency

[ar cuir]ms-word

c. Phonological constituency

p[ar

cuir]

d. Mutation (lenition)

p[ar

cuir]

For the heads realizing the Asp0 node: (117) Base form:

a ceanach ptc buy-VN

a.

Syntactic constituency

AspP[ Asp[a VP[ V[ceanach

b.

Morphological constituency

[a ceanach]ms-word

c.

Phonological constituency

p[a

ceanach]

274

d.

Mutation (lenition)

p[a

cheanach]

For nominal contexts, the following representation of the sequence determinernoun illustrates the derivation hypothesized by our analysis: (118) Base form:

an cait the cat-masc.gen.sg

a. Syntactic

DP [ D[an] NP[ N[cait]]]

constituency b. Morphological constituency

[ an cait ]ms-word

c. Phonological constituency

p[

an cait]

d. Mutation (lenition)

p[an

chait]

Duffield (1993, 1995) argues, convincingly in my opinion, that possessive proclitics raise to the D0 position, and thus the structure for these elements would be the same as for determiners. This provides a representative sample of how the analysis developed in previous chapters applies also in a straightforward way to Irish ICM. We hope to have provided enough evidence in favor of the systematic properties of this phenomenon, which, leaving aside lexically specified mutation types for each syntactic position or combination of phifeatures, behaves like other phenomena studied in this dissertation. The criticisms raised against the Prosodic Hierarchy Theory in Chapters 1-2 are applicable here as well. In both approaches within this theory (e.g., the Relation-Based approach and the End-Based approach) it would be possible to describe the domain of application of ICM. In the RB approach, it would be assumed that the domain of application of ICM is the phonological word, analyzing complementizers, negative

275

particles, determiners and possessive pronouns as contained in one phonological word with the following verb or noun. If these elements were analyzed as clitics, the clitic group would be analyzed as the domain of application. However, there is one particle, the negative specifier gan, which does not trigger any mutation. This is a pure function word, similar to pas in French. Thus, this particle should fall under the same prosodic word with the following lexical item and be a trigger of mutation. However, it does not. Then, the lack of mutation is an accident. In our approach, this is expected. Gan is not a head, and thus it is not part of a feature chain. In the EB approach, the right edges of lexical heads could be set as the parameter responsible for the creation of prosodic boundaries and prosodic constituents. Since functional heads such as the ones triggering mutation do not count for prosodic boundary insertion, they would fall within one prosodic constituent with the following lexical head. However, the alternative model of syntax-phonology interface proposed here is preferable, since it represents in a more principled way the nature of the relationships between functional categories and lexical categories after Spell-Out. Our model represents what it means to say that functional categories and lexical categories are included in the same prosodic word. Moreover, it would have to explain why gan is an exception to being included in a prosodic word with adjacent material. i

G1 stands for underlying glides, which require the pre-C allomorph (cf. les whiskeys, with no liaison), and

G2 stands for derived glides (cf. les oiseaux, with liaison). ii

Some of the examples used in this section are from Selkirk (1972, ch. 3), and others have been collected

by me from several native speakers. iii

It must be pointed out that consonant feature constraints also play a role in constraining the frequency or

naturalness ofliaison. In liaison environments, it is much easier to pronounce word-final /t/ and /n/ than to pronounce /s/, a fact which is well documented in the traditional literature on French liaison.

276

iv

The numbers cinq ‘five’, six ‘six’, sept ‘seven’, huit ‘eight’, dix ‘ten’, and perhaps also the quantifiers

plus ‘more’ and tous ‘all’ display mixed properties; their final consonants are pronounced before vowels and not before consonants, but they are also pronounced before a pause. Tranel (1996) proposes a double underlying representation for these words, one with a latent final consonant and one with a fixed final consonant. v

The orthographic h is not pronounced in French, and thus all words written with an initial h are vowel-

initial phonologically. vi

The preposition chez makes obligatory liaison when followed by the pronouns eux ‘their (masc.)’ and

elle(s) ‘her/their (fem.)’, and optional when followed by a full noun phrase. All other prepositions make optional liaison no matter what follows. Tranel (1981:246) suggests that the word combinations chez eux and chez elle(s) have been lexicalized with liaison, much like accent aigu ‘acute accent’ and Etats-Unis ‘United States’. Also, it appears that liaison is more frequent with the preposition en than with any other preposition, especially when en is followed by a bare noun (e.g. en avion ‘by plane’, en argent ‘made of silver’). Tranel (1981:247) provides an explanation based on the phonological nature of the linking consonant ([n]), and he also suggests (personal communication) that the segmental lightness of en compared to the other prepositions could be another factor responsible for this difference, the assumption being that lighter words tend to cliticize. vii

It appears that liaison feels less natural with trop ‘too’. This fact is attributed to the nature of the final

consonant, /p/ (cf. footnote 3). viii

Selkirk (1972) actually claims that there is no liaison after polysyllabic non-lexical items (i.e., degree

adverbs, prepositions, auxiliaries/copulas). Jean-Roger Vergnaud however informs me that liaison is still possible, although the distinction with the corresponding monosyllabic forms exists. Namely, that it is harder not to make liaison after monosyllabic degree adverbs, prepositions, auxiliaries/copulas than it is to make liaison, whereas the contrary occurs with polysyllabic forms of these elements.

277

ix

Some of these contexts are possible in Style II, others are found only in Style III, and still others are

impossible in either style. For some speakers it is not as bad to make liaison between a verb and its direct object as it is to make liaison between a noun and an adjective or between a subject and a verb. Still, the same speakers judge liaison in the context V-Obj rare or awkward in natural conversational style, which is the style of speech we are concentrating on. x

Some researchers do not posit a separate projection for auxiliary verbs, and locate them in I0.

xi

Chomsky (1994) argues that some terminal nodes might be heads and maximal projections at the same

time, but all terminal nodes are heads at least. xii

For the purposes of the discussion, we treat possessive pronouns and numerals as determiners and

quantifiers, respectively, although there have been proposals that analyze them as heads of their own maximal projections. Since in any of these analyses possessives and numerals are still taken to be heads, this issue does not affect the validity of our point. xiii

This is irrelevant of whether these features are overtly realized or not. Determiners and demonstratives

realize gender, although numerals do not. Possessives are specified for the gender of the possessor, not of the possessee, i.e., the following noun/adjective. The ‘singular’ specification for the feature number is not overtly realized, but it is a feature present in those categories (e.g., as [-plural]). xiv

Longobardi (1994:559-562) claims that D also contains the feature [± referential], which has to be

checked by a noun with the right features. [+R] is checked if the D is in a chain with an object-referring expression, and [-R] is checked if the D is in a chain not containing any object-referring expression. xv

xvi

To be more precise, the determiner head forms a morphosyntactic word with the following noun.

The liaison consonant is underlined. Its actual phonological realization is an the onset of the following

syllable, i.e., the initial syllable of the following word. This is the phenomenon known in the traditional literature on French liaison as liaison avec enchaînement ‘liaison with linking’.

278

xvii

I am thankful to Bernard Tranel for informing me of these facts.

xviii

Liaison is perhaps not only lexicalized for some adjectives, but for certain frequently occurring pairs

adjective-noun. Even though the adjective mauvais ‘bad’ is not as frequent in prenominal position as the adjectives mentioned above, according to judgments I collected from several native speakers, it seems easier to make liaison in (i) than in (ii): (i) mauvais{enfant bad child (ii) xix

mauvais /? officier

Ó Dochartaich (1978) maintains, however, that not all the features of this consonant are deleted. The

underlying neutral or palatalized nature of the consonant is left to attach to, for instance, the preceding definite determiner an, giving a neutral or palatalized [n] depending on the original quality of the /f/ segment. xx

Rotenberg (1978) does not provide any specification for the column of eclipsis for the last two

consonants. Crill (1984), however, gives the phonetic representations [s] and [š] for an eclipsed s, and [m] and [m’] for an eclipsed m. xxi

Grijzenhout (1995:Ch.3) dates Irish intervocalic lenition in the first half or the middle of the sixth

century. Lenition in the British Celtic languages is dated in the late sixth or early seventh century. xxii

It must be pointed out that although this analysis avoids many of the problems that previous accounts

had, it still lacks an explanation as to why a vowel triggers deletion of the A0 position in the following consonant. xxiii

The same criticism we raised for Grijzenhout’s analysis of lenition in Archaic Irish musr be raised for

nasalization as well: the author does not explain how the SV node is inserted under the Arel node. We must assume that it is due to soem process of spreading of the SV node of the preceding nasal consonant.

279

xxiv

There is no overt indefinite determiner in Irish. To convey indefiniteness, nominals are expressed in

their bare form. ‘A man’, for instance, is expressed as fear, whereas ‘the man’ would be expressed as an fear. xxv

The third person feminine possessive pronoun does not trigger lenition on consonant-initial nouns (i.e.,

a teach ‘her house’), but it causes /h/ prefixation on vowel-initial nouns. This is why in the table above we have included the noun ínion ‘daughter’. xxvi

Thomas-Flinders (1981) adopts a similar approach to the mutation phenomena of the Leurbost dialect

of Scots Gaelic, positing actual segments (i.e., /N/, /g/ and /h/) in the underlying representations of formatives that trigger non-lenition processes. These segments then are linked onto the initial consonant of the following word. xxvii

xxviii

This proclitic is often maintained before consonant-initial stems in formal and poetic registers.

Duffield suggests a very interesting explanation for the allomorphy do/-r. When it is not deleted, do

appears in unsupported contexts, that is, in root clauses; -r, however, appears whenever there is an element in C0, and its phonological shape could be explained as a two-step process: first, the d becomes the coda of the word in C0 (the vowel o gets deleted somehow), and second, the d suffers a rhotacization process in coda position which changes it into r. xxix

It has to be mentioned that the copula only surfaces in full form (i.e., spelt out as ba) in root affirmative clauses (cf. (110a)), getting deleted or substituted by -r when preceded by an element in Comp, such as negation (cf. (110b)) or a complementizer (cf. (110c)). After vowel-initial verbs, it is possible to see some remnants of the copula, in the form of -rb or -rbh (cf. (110d)). xxx

McCloskey (1992) in fact offers independent evidence suggesting that complementizers in Irish lower at

PF to the position occupied by the verb, over elements in the specifier of the projection(s) intervening between the complementizer and the verb.

289

References Abney, Steven. 1987. The English noun phrase in its sentential aspect. Doctoral dissertation, MIT, Cambridge, Mass. Albizu, Pablo. 1991. Sobre la existencia del Movimiento Largo de Núcleos en euskera. Ms., Instituto Universitario Ortega y Gasset, Madrid. Albizu, Pablo. 1992a. Tiempo y aspecto verbales en euskera: notas en torno a la teoría de los tiempos como Estructura Argumental Temporal. Ms., Instituto Universitario Ortega y Gasset, Madrid. Albizu, Pablo. 1992b. Where is COMP in Basque?. Ms., University of Southern California, Los Angeles, Calif. Albizu, Pablo. 1994. Word order in Basque and its implications for the theory of focus. Ms., University of Southern California, Los Angeles, Calif. Anderson, Stephen. 1982. Where’s morphology? Linguistic Inquiry 13:571-612. Anderson, Stephen. 1984. Kwakwala syntax and the Government-Binding theory. In Syntax and Semantics 16: The syntax of Native American languages, ed. Eung-Do Cook and Donna Gerdts, 21-76. New York: Academic Press. Anderson, Stephen.1986. Disjunctive ordering in inflectional morphology. Natural Language and Linguistic Theory 4:1-31. Anderson, Stephen. 1992. A-morphous morphology.Cambridge: Cambridge University Press. Archangeli, 1985. An overview of the theory of lexical phonology and morphology. MIT working papers in linguistics 7, 1-16. MITWPL, Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Artiagoitia, Xabier. 1992. Verbal projections in Basque and minimal structure. Doctoral dissertation, University of Washington, Seattle, Wash. Published in 1995 as a supplement of the International Journal of Basque Linguistics and Philology, Donostia-San Sebastián. Baker, Mark. 1985. The Mirror Principle and morphosyntactic explanation. Linguistic Inquiry 16: 373-416. Baker, Mark. 1988. Incorporation: A theory of grammatical function changing. Chicago: University of Chicago Press.

290

Belletti, Adriana. 1990. On the morphosyntactic nature of the sequence ‘Aux+Past Participle’ in Italian. In Grammar in progress, ed. Joan Mascaró and Marina Nespor, 25-32. Dordrecht: Foris. Benmamoun, Elabbas. 1992. Functional and inflectional morphology: Problems of projection, representation and derivation. Doctoral dissertation, University of Southern California, Los Angeles, Calif. Benmamoun, Elabbas. 1996. Agreement asymmetries and the PF interface. Paper presented at the Fifteenth West Coast Conference on Formal Linguistics. University of California, Irvine, Calif. Bernstein, Judy. 1993. Topics in the syntax of nominal structure across Romance. Doctoral dissertation, City University of New York, New York. Bickmore, Lee. 1990. Branching nodes and prosodic categories. In The phonology-syntax connection, ed. Sharon Inkelas and Draga Zec, 1-17. Stanford, Calif. and Chicago, Ill.: CSLI Publications and University of Chicago Press. Blanc, Michel, and Patricia Biggs. 1972. L’enquête socio-linguistique sur le français parlé à Orléans. Le français dans le monde 85: 16-25. Bobaljik, Jonathan David. 1994. What does adjacency do? In MIT working papers in linguistics 22: The morphology-syntax connection, 1-32. MITWPL, Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Bobaljik, Jonathan David. 1995. Morphosyntax: The syntax of verbal inflection. Doctoral dissertation, MIT, Cambridge, Mass. Bonet, Eulàlia. 1991. Morphology after syntax: Pronominal clitics in Romance languages. Doctoral dissertation, MIT, Cambridge, Mass. Booij, Geert. 1985. Coordination reduction in complex words: a case for prosodic phonology. In Advances in nonlinear phonology, ed. Harry van der Hulst and Norval Smith, 143- 160. Dordrecht: Foris. Borer, Hagit. 1988. On the morphological parallelism between compounds and constructs. Yearbook of Morphology 1:45-66. Borer, Hagit. 1994. Deconstructing the construct. Ms., University of Massachusetts, Amherst. Borsley, Robert, and María Luisa Rivero. 1994. Clitic auxiliaries and incorporation in Natural Language & Linguistic Theory 12:373-422. Polish.

291

Borsley, Robert, María Luisa Rivero, and Janig Stephens. 1993. Long head movement in Breton. Ms., University of Wales, Bangor, University of Ottawa, and Cardiff Institute of Higher Education. Cardinaletti, Anna, and Michal Starke. 1994. The typology of structural deficiency: On the three grammatical classes. Ms., University of Venice and University of Geneva/ MaxPlanck Berlin. Chen, Matthew. 1987. The syntax of Xiamen tone sandhi. Phonology Yearbook 4:109149. Chen, Matthew. 1990. What must phonology know about syntax? In The phonologysyntax connection, ed. Sharon Inkelas and Draga Zec, 19-46. Stanford, Calif. and Chicago, Ill.: CSLI Publications and University of Chicago Press. Cho, Young-mee Yu. 1990. Syntax and phrasing in Korean. In The phonology-syntax connection, ed. Sharon Inkelas and Draga Zec, 47-62. Stanford, Calif. and Chicago, Ill.: CSLI Publications and University of Chicago Press. Chomsky, Noam. 1970. Remarks on nominalization. In Readings in transformational grammar, ed. Roderick A. Jacobs and Peter S. Rosenbaum, 184-221. Waltham, Mass.: Ginn and Co. Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: Foris. Chomsky, Noam. 1986a. Barriers. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1986b. Knowledge of language: Its nature, origin, and use. New York: Praeger. Chomsky, Noam. 1993. A minimalist program for linguistic theory. In The view from Building 20: Essays in linguistics in honor of Sylvain Bromberger, ed. Kenneth Hale and Samuel Jay Keyser, 1-52. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1994. Bare phrase structure. MIT occasional papers in linguistics 5. MITWPL, Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Reprinted in Government and Binding Theory and the Minimalist Program, ed. Gert Webelhuth, 383-439. Oxford: Blackwell. 1995 Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Chomsky, Noam, and Morris Halle. 1968. The sound pattern of English. New York: Harper and Row. Christian Brothers. 1990. New Irish grammar. Dublin: C.J. Fallon.

292

Cinque, Guglielmo. 1993. On the evidence for partial N movement in the Romance DP. Ms., Università di Venezia. Clark, Mary M. 1990. The tonal system of Ìgbo. Dordrecht: Foris. Clements, George N. 1978. Tone and syntax in Ewe. In Elements of tone, stress, and intonation, ed. Donna Jo Napoli, 21-99. Washington, D.C.: Georgetown University Press. Clements, George N., and Samuel Jay Keyser. 1983. CV phonology. A generative theory of the syllable. Cambridge, Mass.: MIT Press. Cohn, Abigail. 1989. Stress in Indonesian and bracketing paradoxes. Natural Language & Linguistic Theory 7:167-216. Condoravdi, Cleo. 1990. Sandhi rules of Greek and prosodic theory. In The phonologysyntax connection, ed. Sharon Inkelas and Draga Zec, 63-84. Stanford, Calif. and Chicago, Ill.: CSLI Publications and University of Chicago Press. Crill, Hildred. 1984. Celtic mutations and post-SPE phonology. Ms., University of North Carolina, Chapel Hill. Crisma, Paola. 1990. Functional categories inside the noun phrase: A study of the distribution of nominal modifiers. Tesi di laurea, Università di Venezia. Déchaine, Rose-Marie. 1993. Predicates across categories. Doctoral dissertation, University of Massachusetts, Amherst. Déchaine, Rose-Marie, and Victor Manfredi. 1995. Cohabitations of syntax and phonology. Paper presented at HILP 2, Free University Amsterdam, January 26, 1995. De Jong, Daan. 1988. Sociolinguistic aspects of French liaison. Doctoral dissertation, Free University of Amsterdam. De Jong, Daan. 1990. The syntax-phonology interface and French liaison. Linguistics 28:57- 88. Dell, François. 1970. Les règles phonologiques tardives et la morphologie dérivationnelle du français. Doctoral dissertation, MIT, Cambridge, Mass. Delsing, Lars Olof. 1988. The Scandinavian noun phrase. In Working papers in Scandinavian syntax 42, 57-80. Dillon, Myles, and D. Ó Cróinín. 1961. Teach yourself Irish. English Universities Press Ltd.

293

Di Sciullo, Anna-Maria, and Edwin Williams. 1987. On the definition of word. Cambridge, Mass.: MIT Press. Dobrovie-Sorin, Carmen. 1987. A propos de la structure du groupe nominal en roumain. Rivista di grammatica generativa 12:123-152. Duffield, Nigel. 1991. Particles and projections. Doctoral dissertation, University of Southern California, Los Angeles, Calif. Duffield, Nigel. 1993. Irish construct state nominals and the radical pro-drop phenomenon. Proceedings of NELS 23, 113-127. GLSA, University of Massachusetts at Amherst. Duffield, Nigel. 1995. Particles and projections in Irish syntax. Dordrecht: Kluwer. Duffield, Nigel. 1996. Distributed mutation. Paper presented at NELS 27, McGill University, Montréal. Elordieta, Gorka. 1994a. Vowel Assimilation in Lekeitio Basque and its implications for the minimalist theory. Proceedings of the twenty-third Western Conference on Linguistics, ed. Sharon Hargus, Gerald R. McMenamin and Vida Samiian, 81-93. Fresno, Calif.: California State University, Fresno. Elordieta, Gorka. 1994b. On prosodic government. MIT working papers in linguistics 23: Student Conference in Linguistics, 115-133. MITWPL, Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Elordieta, Gorka. 1997. Accent, tone, and intonation in Lekeitio Basque. In Issues in the phonology and morphology of the major Iberian languages, ed. Fernando MartínezGil and Alfonso Morales. Washington, D.C.: Georgetown University Press. Emenanjo, E. ‘Nolue. 1978. Elements of Modern Ìgbo grammar. Ibadan: Oxford University Press. Emonds, Joseph. 1985. A unified theory of syntactic categories. Dordrecht: Foris. Euskaltzaindia (Royal Academy of the Basque Language). 1985. Euskal Gramatika: Lehen Urratsak-II. Iruña: Institución Príncipe de Viana. Everett, Daniel. 1989. Clitic doubling, reflexives, and word order alternations in Yagua. Language 65:339-372. Everett, Daniel. 1996. Why there are no clitics. An alternative perspective on pronominal allomorphy. Dallas, Tex.: Summer Institute of Linguistics and The University of Texas at Arlington.

294

Fabb, Nigel. 1984. Syntactic affixation. Doctoral dissertation, MIT, Cambridge, Mass. Fassi Fehri, Abdelkader. 1989. Generalised IP structure, Case and VS order. In Working papers in linguistics 10: Functional heads and clause structure, 75-113. Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Fouché, Pierre. 1959. Traité de prononciation française,2nd edition. Paris: C. Klincksieck. Franco, Jon. 1991. Spanish object clitics as verbal agreement morphemes. MIT working papers in linguistics 14: Student Conference in Linguistics, 99-114. MITWPL, Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Franco, Jon. 1993. On object agreement in Spanish. Doctoral dissertation, University of Southern California, Los Angeles. Fukui, Naoki. 1986. A theory of category projection and its applications. Doctoral dissertation, MIT, Cambridge, Mass. Revised version published as Theory of projection in syntax. Stanford, Calif.: CSLI Publications (1995). Distributed by University of Chicago Press. Fukui, Naoki, and Margaret Speas. 1986. Specifiers and projection. In MIT working papers in linguistics 8: Papers in theoretical linguistics, 128-172. MITWPL, Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Gómez, Ricardo. 1994. Acerca del origen de las formas finitas del verbo vasco. Ms., University of the Basque Country. Grijzenhout, Janet. 1995. Irish consonant mutation and phonological theory. Doctoral dissertation, Utrecht University. Published by the Research Institute for Language and Speech, Utrecht University, Utrecht. Grimshaw, Jane. 1991. Extended projection. Ms., Brandeis University. Grosu, Alexander. 1988. On the distribution of genitive phrases in Rumanian. Linguistics 26: 931-949. Hale, Kenneth, and Samuel Jay Keyser. 1993. On argument structure and the lexical expression of syntactic relations. In The view from Building 20: Essays in linguistics in honor of Sylvain Bromberger, eds. Kenneth Hale and Samuel Jay Keyser, 53-109. Cambridge, Mass.: MIT Press. Hale, Kenneth, and Elisabeth Selkirk. 1987. Government and tonal phrasing in Papago. Phonology Yearbook 4:151-183. Halle, Morris.1973. Prolegomena to a theory of word formation. Linguistic Inquiry 4:316.

295

Halle, Morris. 1990. An approach to morphology. Proceedings of NELS 23, 150-184. GLSA, University of Massachusetts at Amherst. Halle, Morris, and Alec Marantz. 1993. Distributed morphology and the pieces of inflection. In The view from Building 20: Essays in linguistics in honor of Sylvain Bromberger, eds. Kenneth Hale and Samuel Jay Keyser, 111-176. Cambridge, Mass.: MIT Press. Halpern, Aaron. 1995. On the placement and morphology of clitics. Stanford, calif.: CSLI Publications. Hamp, E. 1951. Morphophonemes of the Keltic mutations. Language 27:230-247. Hayes, Bruce. 1986. Inalterability in CV phonology. Language 62:321-351. Hayes, Bruce. 1989. The prosodic hierarchy in meter. Rhythm and Meter, ed. Paul Kiparsky and Gilbert Youmans, 201-260. San Diego: Academic Press. Hayes, Bruce. 1990. Precompiled phrasal phonology. In The phonology-syntax connection, ed. Sharon Inkelas and Draga Zec, 85-108. Stanford, Calif. and Chicago, Ill.: CSLI Publications and University of Chicago Press. Holmberg, Anders, ed. 1992. Papers from the Workshop on the Scandinavian Noun Phrase. Report 32, Department of General Linguistics, University of Umeå. Hoorn, H. van. 1983. Cancellazione della vocale finale prima di consonante. Ms., University of Amsterdam. Horvath, Julia. 1995. Structural focus, structural case and the notion of featureassignment. In Discourse configurational languages, ed. Katalin E. Kiss. New York: Oxford University Press. Hualde, José Ignacio. 1988. A lexical phonology of Basque. Doctoral dissertation, University of Southern California, Los Angeles, Calif. Hualde, José Ignacio. 1991. Basque phonology. London and New York: Routledge. Hualde, José Ignacio. 1996. Basque accentuation. In Word-prosodic systems of the languages of Europe, ed. Harry van der Hulst. Berlin: Mouton. Hualde, José Ignacio, and Gorka Elordieta. 1992. On the lexical/postlexical distinction: Vowel assimilation in Lekeitio Basque. Studies in the Linguistic Sciences 22:159164. Hualde, José Ignacio, Gorka Elordieta and Arantzazu Elordieta. 1994. The Basque dialect of Lekeitio. Supplements of the International Journal of Basque Linguistics and

296

Philology, vol. 34. Bilbo and Donostia: Universidad del País Vasco/Euskal Herriko Unibertsitatea and Diputación Foral de Gipuzkoa. Huang, James C.-T. 1982. Logical relations in Chinese and the theory of grammar. Doctoral dissertation, MIT, Cambridge, Mass. Hyman, Larry. 1985. Word domains and downstep in Bamileke-Dschang. Phonology Yearbook 2:47-83. Inkelas, Sharon. 1990. Prosodic constituency in the lexicon. New York: Garland Publishing. Inkelas, Sharon. 1993. Nimboran position class morphology. Natural Language and Linguistic Theory 11:559-624. Inkelas, Sharon, and Draga Zec. 1988. Serbo-Croatian pitch accent: the interaction of tone, stress and intonation. Language 64:227-248. Itô, Junko. 1986. Syllable theory in prosodic phonology. Doctoral dissertation, MIT, Cambridge, Mass. Jackendoff, Ray. 1972. Semantic interpretation in generative grammar. Cambridge, Mass.: MIT Press. Jackson, Kenneth. 1953. Language and history in Early Britain. Edinburgh: Edinburgh University Press. Jun, Sun-Ah, and Gorka Elordieta. 1997. Intonational structure of Lekeitio Basque. To appear in the Proceedings of the ESCA workshop on intonation, University of Athens. Kaisse, Ellen. 1977. Hiatus in Modern Greek. Doctoral dissertation, Harvard University, Cambridge, Mass. Kaisse, Ellen. 1983. The syntax of Auxiliary Reduction in English. Language 59:93-122. Kaisse, Ellen. 1985. Connected speech: The interaction of syntax and phonology. Orlando: Academic Press. Kaisse, Ellen. 1990. Toward a typology of postlexical rules. In The phonology-syntax connection, ed. Sharon Inkelas and Draga Zec, 127-143. Stanford, Calif. and Chicago, Ill.: CSLI Publications and University of Chicago Press. Kaisse, Ellen, and Patricia Shaw. 1985. On the theory of lexical phonology. Phonology Yearbook 2:1-30.

297

Kaye, Jonathan, and Yves-Charles Morin. 1978. Il n’y a pas de règles de troncation, voyons! In Proceedings of the Twelfth International Congress of Linguists, ed. Wolfgang Dressler and W. Meid, 788-792. Innsbruck: IBS. Kayne, Richard. 1994. The antisymmetry of syntax. Linguistic Inquiry Monograph 25. Cambridge, Mass.: MIT Press. Kelly, Deirdre Mary. 1978. Morphologization in Irish and Southern Paiute. Doctoral dissertation, University of Texas at Austin. Kenstowicz, Michael, and Charles Kisseberth. 1977. Topics in phonological theory. New York: Academic Press. Kidima, Lukowa. 1990. Tone and syntax in Kiyaka. In The phonology-syntax connection, ed. Sharon Inkelas and Draga Zec, 195-216. Stanford, Calif. and Chicago, Ill.: University of Chicago Press. King, Harold. 1970. On blocking the rules for contraction in English. Linguistic Inquiry 1:134-136. Kiparsky, Paul. 1982. Lexical morphology and phonology. In Linguistics in the morning calm, ed. I.-S. Yang, 3-91. Seoul: Hanshin Publishing Co. Kiparsky, Paul. 1985. Some consequences of the theory of lexical phonology. Phonology Yearbook 2:85-138. Klausenburger, Jürgen. 1978. Liaison 1977: the case for epenthesis. Studies in French Linguistics 1:1-20. Klausenburger, Jürgen. 1984. French liaison and linguistic theory. Stuttgart: Steiner. Klavans, Judith. 1982. Some problems in a theory of clitics. Doctoral dissertation, University College London, 1980. Published by Indiana University Linguistics Club. König, Ekkehard. 1991. The meaning of focus particles. London: Routledge. Koopman, Hilda. 1984. The syntax of verbs: From verb-movement rules in the Kru languages to Universal Grammar. Dordrecht: Foris. Laka, Itziar. 1988. Configurational heads in inflectional morphology: The structure of the inflected forms in Basque. International Journal of Basque Linguistics and Philology 32: 343-365. Laka, Itziar. 1990. Negation in syntax: On the nature of functional categories and projections. Doctoral dissertation, MIT, Cambridge, Mass.

298

Laka, Itziar. 1993. The structure of inflection: A case study in X0 syntax. In Generative studies in Basque linguistics, ed. José Ignacio Hualde and Jon Ortiz de Urbina, 21-70. Amsterdam: John Benjamins. Laka, Itziar, and Juan Uriagereka. 1987. Barriers for Basque and viceversa. Proceedings of NELS 17, 394-408. GLSA, University of Massachusetts at Amherst. Lakoff, George. 1970. Global rules. Language 46:627-639. Lapointe, Steven. 1980. A theory of grammatical agreement. Doctoral dissertation, University of Massachusetts, Amherst. Lapointe, Steven. 1988. Toward a unified theory of agreement. In Agreement in natural language: Approaches, theories, descriptions, ed. Michael Barlow and Charles A. Ferguson, 67-87. Stanford, Calif.: Center for the Study of Language and Information. Lasnik, Howard. 1981. Restricting the theory of transformations: A case study. In Explanation in linguistics, ed. Norbert Hornstein and David Lightfoot, 125-145. London: Longman. Lasnik, Howard. 1994. Verbal morphology: Syntactic structures meets the minimalist program. Ms., University of Connecticut. Lema, José, and María Luisa Rivero. 1989. Long Head Movement: ECP vs. HMC. In NELS 20:333-347. GLSA, University of Massachusetts, Amherst. Lema, José, and María Luisa Rivero.1991. Types of verbal movement in Old Spanish: Modals, futures and perfects. Probus 3:237-278. Lieber, Rochelle. 1980. On the organization of the lexicon. Doctoral dissertation, MIT, Cambridge, Mass. Lieber, Rochelle. 1983. New developments in Autosegmental Morphology: Consonant mutation. In Proceedings of the Second West Conference on Formal Linguistics, 165- 175. Stanford University Linguistics Association, Stanford University, Stanford, Calif. Lieber, Rochelle. 1987. An integrated theory of autosegmental processes. Albany: State University of New York Press. Lieber, Rochelle. 1992. Deconstructing morphology: Word formation in syntactic theory. Chicago: University of Chicago Press. Lobeck, Anne, and Ellen Kaisse. 1984. On the domain of locality conditions. In Proceedings of the Third West Coast Conference on Formal Linguistics, 170-178. Stanford Linguistics Association, Stanford University, Stanford, Calif.

299

Longobardi, Giuseppe. 1994. Reference and proper names: A theory of N-movement in syntax and Logical Form. Linguistic Inquiry 25:609-665. Manzini, Maria Rita. 1983. Syntactic conditions on phonological rules. In MIT working papers in linguistics 5, 1-9. MIWPL, Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Marantz, Alec. 1988. Clitics, morphological merger, and the mapping to phonological structure. In Theoretical morphology, ed. Michael Hammond and Michael Noonan, 253-270. San Diego: Academic Press. Marantz, Alec. 1989. Clitics and phrase structure. In Alternative conceptions of phrase structure, ed. Mark Baltin and Anthony Kroch, 99-116. Chicago: University of Chicago Press. Marantz, Alec. 1992. How morphemes are realized phonologically. Paper presented at DIMACS Workshop on Human Language. Martín, Juan. 1995. On the syntactic structure of Spanish noun phrases. Doctoral dissertation, University of Southern California, Los Angeles. Massam, Diane. 1983. The morphology of Irish mutation. MIT working papers in linguistics 5, 10-19. Department of Linguistics and Philosophy, MIT, Cambridge, Mass. McCloskey, James D. 1992. On the scope of verb movement in Irish. Ms., University of California at Santa Cruz. To appear in Natural Language and Linguistic Theory. McHugh, Brian. 1990. The phrasal cycle in Kivunjo Chaga tonology. In The phonologysyntax connection, ed. Sharon Inkelas and Draga Zec, 217-242. Stanford, Calif. and Chicago, Ill.: University of Chicago Press. Mhac an Fhailigh, Éamonn. 1968. The Irish of Erris, Co. Mayo. Dublin: Dublin Institute for Advanced Studies. Michelena, Luis. 1957. Las antiguas consonantes vascas. In Miscelánea homenaje a André Martinet, ed. D. Catalán, 113-157, La Laguna. Reprinted in Sobre historia de la lengua vasca, 116-189. Supplements of the International Journal of Basque Linguistics and Philology 10 [1988]. Donostia-San Sebastián: Diputación Foral de Gipuzkoa. Miller, Philip. 1992. Clitics and constituents in Phrase Structure Grammar. Doctoral dissertation, Utrecht University. Milner, Jean-Claude. 1967. French truncation rule. Quarterly Progress Report 86, 273283. Research Laboratory of Electronics, MIT, Cambridge, Mass.

300

Mohanan, Karuvannur P. 1982. Lexical Phonology. Doctoral dissertation, MIT, Cambridge, Mass. Distributed by Indiana University Linguistics Club, Bloomington, In. Mohanan, Karuvannur P. 1986. The theory of Lexical Phonology. Dordrecht: Reidel. Morin, Yves-Charles. 1983. On the morphologization of word-final consonant deletion in French. Paper read at the Phonetics Congress, Utrecht 1983. Morin, Yves-Charles, and Jonathan Kaye. 1982. The syntactic bases for French liaison. Journal of Linguistics 18:291-330. Mujika, José A. 1988. Partículas modales de la flexión verbal. International Journal of Basque Linguistics and Philology 22:463-478. Murasugi, Kumiko. 1992. Crossing and nested paths: NP-movement in accusative and ergative languages. Doctoral dissertation, MIT, Cambridge, Mass. Muysken, Peter. 1977. Syntactic developments in the verb phrase of Ecuadorian Quechua. Dordrecht: Foris. Napoli, Donna Jo, and Marina Nespor. 1979. The syntax of word-initial consonant gemination in Italian. Language 55:812-841. Nespor, Marina. 1990. On the separation of prosodic and rhythmic phonology. In The phonology-syntax connection, ed. Sharon Inkelas and Draga Zec, 243-258. Stanford, Calif. and Chicago, Ill.: University of Chicago Press. Nespor, Marina. 1997. The phonology of clitic groups. Ms., HIL/University of Amsterdam. Nespor, Marina, and Irene Vogel. 1982. Prosodic domains of external sandhi rules. In The structure of phonological representations (PartI), ed. Harry van der Hulst and Norval Smith, 225-255. Dordrecht: Foris. Nespor, Marina, and Irene Vogel. 1986. Prosodic phonology. Dordrecht: Foris. Nevis, Joel. 1985. Finnish particle clitics and general clitic theory. Doctoral dissertation, Ohio State University. Published by Garland in 1988. Noyer, Rolf. 1992. Features, positions, and affixes in autonomous morphological structure. Doctoral dissertation, MIT, Cambridge, Mass. Ó Cuív, Brian. 1986. Sandhi phenomena in Irish. In Sandhi phenomena in the languages of Europe, ed. Henning Andersen, 395-414. Berlin: Mouton de Gruyter. Ó Dochartaigh, Cathair. 1978. Lenition and Dependency Phonology. Éigse 17:457-494.

301

Oftedal, Magne. 1962. A morphemic evaluation of the Celtic initial mutations. Lochlann 2:93- 102. Oka, Toshifusa. 1993. Shallowness. In Papers on case and agreement II: MIT Working Papers in Linguistics 19, ed. Colin Phillips, 255-320. Ormazabal, Javier, Juan Uriagereka, and Myriam Uribe-Etxebarria. 1994. Word order and wh-movement: towards a parametric account. Paper presented at GLOW 17, University of Vienna, April 1994. Ortiz de Urbina, Jon. 1989. Some parameters in the grammar of Basque. Dordrecht: Foris. Ortiz de Urbina, Jon. 1994. Verb-initial patterns in Basque and Breton. Lingua 94:125153. Ortiz de Urbina, Jon. To appear. Residual verb-second and verb-first in Basque. In Discourse configurational languages, ed. Katalin E. Kiss. New York: Oxford University Press. Osa, Eusebio. 1990. Euskararen hitzordena: komunikazio zereginaren arauera. Leioa: Servicio Editorial de la Universidad del País Vasco. Ó Siadhail, Mícheál. 1989. Modern Irish: grammatical structure and dialectal variation. Cambridge: Cambridge University Press. Ouhalla, Jamal. 1988. Movement in noun phrases. Paper presented at the LAGB Meeting, Durham, March 1988. Payne, Doris, and Thomas Payne. 1989. Yagua. In Handbook of Amazonian languages, vol. 2, ed. Desmond Derbyshire and Geoffrey Pullum, 249-474. Berlin: Mouton de Gruyter. Pesetsky, David. 1979. Russian morphology and lexical theory. Ms., MIT, Cambridge, Mass. Pollock, Jean-Yves. 1989. Verb movement, Universal Grammar, and the structure of IP. Linguistic Inquiry 20:365-424. Poser, William. 1985. Cliticization to NP and Lexical Phonology. In Proceedings of the Fourth West Coast Conference on Formal Linguistics, 262-272. Stanford Linguistics Association, Stanford University, Stanford, Calif. Pulleyblank, Douglas. 1986. Tone in lexical phonology. Dordrecht: Foris.

302

Rebuschi, Georges. 1990. On the non-configurationality of Basque and some related phenomena. International Journal of Basque Linguistics and Philology 24:325342. Rice, Keren. 1991. Prosodic constituency in Hare. (Athapaskan): Evidence for the foot. Lingua 82:201-245. Rice, Keren. 1993a. A reexamination of the feature [sonorant]: the status of ‘sonorant obstruents’. Language 69:308-344. Rice, Keren. 1993b. The structure of the Slave (Northern Athabaskan) verb. In Phonetics and Phonology 4. Studies in lexical phonology, ed. Sharon Hargus and Ellen Kaisse, 145-171. San Diego, Calif: Academic Press. Riemsdijk, Henk van. 1990. Functional prepositions. In Unity in diversity: Papers presented to Simon C. Dik on his 50th birthday. Dordrecht: Foris. Riemsdijk, Henk van. 1996. The extension of projections. Ms., Tilburg University. To appear in the Tsuru University Working Papers in Linguistics. Ritter, Elizabeth. 1986. NSO noun phrase in a VSO language. Ms., MIT, Cambridge, Mass. Ritter, Elizabeth. 1989. A head-movement approach to construct-state noun phrases. Linguistics 26:909-929. Ritter, Elizabeth, and Sarah Thomas Rosen. 1993. Deriving causation. Natural Language & Linguistic Theory 11:519-555. Rivero, María Luisa. 1991. Long head movement and negation: Serbo-Croatian vs. Slovak and Czech. The Linguistic Review 8:319-351. Rivero, María Luisa. 1993. Finiteness and second position in Long Head Movement languages: Breton and Slavic. Paper read at the meeting of the Eurotyp Project Group 3, Lund, April 1993. Rivero, María Luisa. 1994. Clause structure and V-movement in the languages of the Balkans. Natural Language & Linguistic Theory 12:63-120. Rizzi, Luigi. 1979. La teoria della traccia e processi fonosintattici. Rivista di Grammatica Generativa 4:165-181. Rizzi, Luigi. 1991. Residual verb movement and the wh-criterion. Ms., University of Geneva. Rizzi, Luigi, and Leonardo Savoia. 1993. Conditions on /u/ propagation in southern Italian dialects: a locality parameter for phonosyntactic processes. In Syntactic theory and the dialects of Italy, ed. Adriana Belletti, 252-318. Torino: Rosenberg & Sellier.

303

Roberts, Ian. 1992. Two types of head movement in Romance. Ms., University of Wales, Bangor. Rotenberg, Joel.1978. The syntax of phonology. Doctoral dissertation, MIT, Cambridge, Mass. Sánchez, Liliana. 1996. Syntactic structure in nominals: A comparative analysis of Spanish and Southern Quechua. Doctoral dissertation, University of Southern California, Los Angeles. Schachter, Paul, and Fe T. Otanes. 1972. Tagalog reference grammar. Berkeley: University of California Press. Schane, Sanford. 1968. French phonology and morphology. Cambridge, Mass.: MIT Press. Schane, Sanford. 1974. There is no French truncation rule. In Linguistic studies in Romance languages, ed. R.J. Campbell et al., 89-99. Washington, D.C.: Georgetown University Press. Schein, Barry, and Donca Steriade. 1986. On geminates. Linguistic Inquiry 17:691-744. Schuchardt, Hugo. 1923 [1947]. Primitiae Linguae Vasconum. Salamanca: Colegio Trilingüe-CSIC. Selkirk, Elisabeth. 1972. The phrasal phonology of English and French. Doctoral dissertation, MIT, Cambridge, Mass. Selkirk, Elisabeth. 1974. French liaison and the X notation. Linguistic Inquiry 5:573-590. Selkirk, Elisabeth. 1978. On prosodic structure and its relation to syntactic structure. Paper presented at the Conference on Mental Representation in Phonology. IULC, 1980. Published in Nordic Prosody II, ed. Thorstein Fretheim (1980), 111-140. Trondheim: Tapir. Selkirk, Elisabeth. 1980a. Prosodic domains in phonology: Sanskrit revisited. In Language Sound Structure, ed. Mark Aronoff and Richard T. Oehrle, 107-136. Cambridge, Mass.: MIT Press. Selkirk, Elisabeth. 1980b. The role of prosodic categories in English word stress. Linguistic Inquiry 11:563-605. Selkirk, Elisabeth. 1982. The syntax of words. Cambridge, Mass.: MIT Press. Selkirk, Elisabeth. 1984. Phonology and syntax: The relation between sound and structure. Cambridge, Mass.: MIT Press.

304

Selkirk, Elisabeth. 1986. On derived domains in sentence phonology. Phonology Yearbook 3:371-405. Selkirk, Elisabeth. 1995. The prosodic structure of function words. In University of Massachusetts Occasional Papers 18: Papers in Optimality Theory. GLSA, University of Massachusetts, Amherst. Selkirk, Elisabeth, and Tong Shen. 1990. Prosodic domains in Shanghai Chinese. In The phonology-syntax connection, ed. Sharon Inkelas and Draga Zec, 313-337. Stanford, Calif. and Chicago, Ill.: University of Chicago Press. Siloni, Tali. 1989. Le syntagme nominal en hébreu. Ms., Université de Genève. Siloni, Tali. 1990. Hebrew noun phrases: Generalized noun raising. Ms., Université de Genève. Siloni, Tali. 1994. Hebrew noun phrases. Doctoral dissertation, Université de Genève. Sportiche, Dominique. 1992. Clitic constructions. Ms., University of California, Los Angeles. Sproat, Richard. 1985. On deriving the lexicon. Doctoral dissertation, MIT, Cambridge, Mass. Steriade, Donca. 1993. Closure, release and nasal contours. In Nasals, nasalization and the velum (Phonetics and Phonology 5), ed. Marie K. Huffman and Rena A. Krkow, 401- 470. San Diego, Calif: Academic Press. Stowell, Timothy. 1981. Origins of phrase structure. Doctoral dissertation, MIT, Cambridge, Mass. Szabolcsi, Anna. 1983. The possessor that ran way from home. The Linguistic Review 3:89- 102. Szabolcsi, Anna. 1987. Functional categories in the noun phrase. In Approaches to Hungarian, vol. 2, ed. István Kenesei, University of Budapest. Taraldsen, Tarald. 1990. D-projections and N-projections in Norwegian. In Grammar in progress, ed. Marina Nespor and Joan Mascaró, 419-431. Dordrecht: Foris. Ternes, Elmar. 1986. A grammatical hierarchy of joining. In Sandhi phenomena in the languages of Europe, ed. Henning Andersen, 11-21. Berlin: Mouton de Gruyter. Thomas-Flinders, Tracy. 1981. Initial lenition in Celtic: evidence for inflection as phonological process. In UCLA Occasional Papers 4. Inflectional morphology: introduction to the Extended Word-and-Paradigm Theory, ed. Tracy ThomasFlinders, 72-83. Department of Linguistics, UCLA, Los Angeles, Calif.

305

Thurneysen, Rudolf. 1946. A grammar of Old Irish. Translated from the German by D. A. Binchy and Osborn Bergin. Dublin: The Dublin Institute for Advanced Studies. Torrego, Esther. 1993. Pronouns and determiners. A DP analysis of Spanish nominals. Ms., University of Massachusetts. Tranel, Bernard. 1981. Concreteness in generative phonology. Berkeley: University of California Press. Tranel, Bernard. 1987. The sounds of French. Cambridge: Cambridge University Press. Tranel, Bernard. 1992. On suppletion and French liaison. In Romance languages and modern linguistic theory, ed. Paul Hirschbühler and Konrad Koerner, 269-308. Amsterdam and Philadelphia: John Benjamins. Tranel, Bernard. 1995. Current issues in French phonology: Liaison and position theories. In The handbook of phonological theory, ed. John Goldsmith, 798-816. Oxford: Blackwell. Tranel, Bernard. 1996. Exceptionality in Optimality Theory and final consonants in French. In Grammatical theory and Romance languages, ed. Karen Zagona, 275291. Amsterdam and Philadelphia: John Benjamins. Trask, Larry. 1977. Historical syntax and Basque verbal morphology. In Anglo-American contributions to Basque Studies. Essays in honor of Jon Bilbao, ed. William Douglass, R. Etulain and William Jacobsen, 203-217. Reno, Nev.: University of Nevada. Travis, Lisa deMena. Parameters and effects of word order variation. Doctoral dissertation, MIT, Cambridge, Mass. Tuller, Laurice. 1992. The syntax of postverbal constructions in Chadic. Natural Language & Linguistic Theory 10:303-334. Uriagereka, Juan. 1987. Government in Basque. University of Connecticut Working Papers in Linguistics 1. Uriagereka, Juan. 1992. Extraction parameters. A case study on underspecification. Ms., University of Maryland. Uribe-Etxebarria, Myriam. 1989. Basque wh-movement revisited. Ms., University of Connecticut. Valois, Daniel. 1991. The internal syntax of DPs. Doctoral dissertation, University of California, Los Angeles.

306

Vanelli, L. 1979. Una forma supplettiva dell’articolo e la sua fonosintassi. Rivista de Grammatica Generativa 4:183-206. Vogel, Irene, M. Drigo, A. Moser, and I. Zannier. 1983. La cancellazione di vocale in italiano. Studi di Grammatica Italiana, 191-230. Vogel, Irene, and István Kenesei. 1990. Syntax and semantics in phonology. In The phonology-syntax connection, ed. Sharon Inkelas and Draga Zec, 339-363. Stanford, Calif. and Chicago, Ill.: University of Chicago Press. Wetzels, Leo. 1987. The timing of latent consonants in Modern French. In Studies in Romance languages, ed. Carol Neidle and Rafael Núez-Cedeño, 283-317. Dordrecht: Foris. Zec, Draga. 1987. Interactions of prosodic and syntactic constituency. Ms., Stanford University, Stanford, Calif. Zec, Draga. 1988. Sonority contraints on prosodic structure. Doctoral dissertation, Stanford University, Stanford, Calif. Zec, Draga. 1993. Rule domains and phonological change. In Phonetics and Phonology 4. Studies in lexical phonology, ed. Sharon Hargus and Ellen Kaisse, 365-405. San Diego: Academic Press. Zec, Draga, and Sharon Inkelas. 1987. Phonological phrasing and the reduction of function words. Paper presented at the annual meeting of the Linguistic Society of America, San Francisco, Calif. Zec, Draga, and Sharon Inkelas. 1990. Prosodically-constrained syntax. In The phonology- syntax connection, ed. Sharon Inkelas and Draga Zec, 365-378. Stanford, Calif. and Chicago, Ill.: University of Chicago Press. Zoll, Cheryl. 1994. Ghosts segments and optimality. Proceedings of the West Coast Conference on Formal Linguistics 12, 183-199. Stanford, Calif.: CSLI Publications. Zoll, Cheryl. 1996. Doctoral dissertation, University of California, Berkeley. Zsiga, Elizabeth. 1992. A mismatch between morphological and prosodic domains: Evidence from two Igbo rules. Phonology 9:101-135. Zubizarreta, María Luisa, and Jean-Roger Vergnaud. 1995. Primitive chains. Work in progress, University of Southern California, Los Angeles, Calif. Zwicky, Arnold. 1969. Phonological constraints in syntactic descriptions. Papers in Linguistics 1:411-463.

307

Zwicky, Arnold. 1970. Auxiliary reduction in English. Linguistic Inquiry 1:323-336. Zwicky, Arnold. 1977. On clitics. Blomington, Ind.: Indiana University Linguistics Club. Zwicky, Arnold. 1985. Clitics and particles. Language 61:283-305. Zwicky, Arnold. 1987. Suppressing the Z’s. Journal of Linguistics 23:133-148. Zwicky, Arnold, and Geoffrey Pullum. 1983. Cliticization vs. inflection: English n’t. Language 59:502-513. Zwicky, Arnold, and Geoffrey Pullum. 1986. The principle of phonology-free syntax. Ohio State University Working Papers in Linguistics 32:92-99.

Suggest Documents