<<

University of Calgary PRISM: University of Calgary's Digital Repository

Graduate Studies The Vault: Electronic Theses and Dissertations

2012-09-26 When nothing exists: The role of zero in the prosodic hierarchy

Windsor, Joseph William

Windsor, J. W. (2012). When nothing exists: The role of zero in the prosodic hierarchy (Unpublished master's thesis). University of Calgary, Calgary, AB. doi:10.11575/PRISM/28694 http://hdl.handle.net/11023/227 master thesis

University of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission. Downloaded from PRISM: https://prism.ucalgary.ca

THE UNIVERSITY OF CALGARY

When nothing exists:

The role of zero in the prosodic hierarchy

by

Joseph W. Windsor

A THESIS

SUBMITTED TO THE FACULTY OF GRADUATE STUDIES

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF

MASTER OF ARTS

DEPARTMENT OF LINGUISTICS

CALGARY, ALBERTA

SEPTEMBER, 2012

© Joseph W. Windsor, 2012

THE UNIVERSITY OF CALGARY FACULTY OF GRADUATE STUDIES DEPARTMENT OF LINGUISTICS

The undersigned certify that they have read and recommended to the Faculty of Graduate Studies for acceptance, a thesis entitled 'When nothing exists: The role of zero in the prosodic hierarchy' submitted by Joseph W. Windsor in partial fulfillment of the requirements for the degree of Master of Arts in Linguistics.

______Supervisor, Dr. Darin Flynn, Department of Linguistics

______Dr. Karsten Koch, Department of Linguistics

______Dr. Richard Zach, Department of Philosophy

______Date

Abstract

Zero (Ø) has been a phonological tool for over a century used to explain linguistic phenomena where language-specific rules either under or over-apply. This tool has often been argued to be 'dangerous' since it gives 'additional power to the system' (Stanley 1967). This long-standing controversy has lead to many authors arguing against analyses of phenomena that rely on Ø. This opposition to Ø often leads those authors to relegate a given phenomenon to the lexicon rather than maintaining a phonological explanation for a given sound change.

While the skepticism of those authors may be warranted, what is lacking from the field is a list of criteria by which to evaluate proposals of Ø. This thesis examines cases of proposed Øs in several languages to propose a list of four criteria which can be used as a general guideline for either warranting or eliminating Ø in a particular phonological analysis.

iii

Acknowledgements

For the past two years, I've been asked 'what do you study?' to which I would answer, 'nothing... literally.' It has become an inside joke to talk about the student specializes in nothing, and such an abstract topic is without doubt difficult to supervise. Because of this, I would like to extend my deepest thanks to my supervisor, Dr. Darin Flynn for his unwavering encouragement and support. No matter how crazy the idea, Darin never once told me I was wrong. He listened to each hypothesis, discussed it with me, and pointed me toward an that I should read. This often lead me to reconsider what I had previously thought, and each time, Darin made sure I was well read on a topic before committing the idea to paper. This work is a direct result of his support, direction, and enthusiasm. Additionally, I would like to thank the members of my examining committee: Dr. Karsten Koch, Dr. Richard Zach, and my neutral chair Dr. Amanda Pounder. Your questions and comments on my thesis were both insightful and challenging, and they helped me to identify avenues for future research. There are many other members of the Department of Linguistics who deserve to be thanked as well, too many to be individually named. At one point or another, almost every graduate student or faculty member helped me to formalize the ideas that are presented in this thesis. Whether it was directly tied to a class, or thinking of my thesis while reading a related article and emailing me the reference, or just sitting in front of a whiteboard and helping me to sort out my ideas. They all provided one of the most supportive and friendly environments to study in that I've ever known. Finally, no one person ever truly writes a thesis on their own. It is not just the student who stays up late at night or pulls their hair out when something just doesn't quite fit. Their family is always right there with them, feeling every set back or celebrating any achievement. In my family, this was my grandmother – who made sure I was looking after myself; my grandfather – who taught me to fix my own vehicles because "I can't help you with school, but I can make sure you get there;" my father – who listened to every crazy idea, smiled, and told me to go for it; my mother – who both worried and supported as only a mother can; my step-mother – who wanted nothing more than to see me succeed; and my step-father who made sure I arrived by driving with me 5133 of the 5212km (the tow truck was responsible for the last 79km, but he was there for that too) from Stellarton to Calgary;

iv

and, of course, Alanna – who stepped in and made sure I was living on more than just ramen noodles and coffee, and looked after both her share of the housework and mine, just so I wouldn't be distracted. I never could have done this on my own, and I hope you all know how large your role was in this process.

v

This thesis is dedicated in memory of Dr. Kenneth E. Nilsen who inspired my love of Celtic languages, literatures, and cultures. His enthusiasm and love for language is what started me down this path. Go soilsí Dia solas na bhFlaitheas dhó

vi

Table of Contents

Approval page ii Abstract iii Acknowledgements iv Dedication vi Table of Contents vii List of Numbered Examples ix List of Abbreviations and Symbols xiv Epigraph xv CHAPTER 1: Introduction 1 1.1: Phonological frameworks 4 1.2: Interface and syntactic frameworks 8 1.3: Ø in feature theory 10 1.4: Acoustic Ø 14 CHAPTER 2: Where is Ø expected in phonological theory? 18 2.1: Ø segments 19 2.2: Ø mora 34 2.3: Ø 39 2.4: Ø feet 45 2.5: Ø words 46 2.6: Ø phrases 50 2.7: Ø intonational phrases 54 2.8: Interim conclusion I 60 CHAPTER 3: Case studies of Ø 62 3.1: 'Ghost' vowels in Russian (Jer vowels) 62 3.1.1: Competing models of representation for Jer vowels 64 3.1.2: Predictions of the exceptional word analysis 69 3.1.3: 'Ghost' segments in Seri 73 3.1.4: Consequences of Seri Ø for Russian 80 3.1.5: Ø inputs and outputs for Russian Jers 83 3.1.6: Summary 86 3.2: A Ø prosodic word in Irish 88 3.2.1: Consonantal lenition to Ø in Irish mutations 88 3.2.1.2: A brief history of Irish lenitions and interface theories 89 3.2.1.3: Phonological realization of Irish lenitions 92 3.2.1.4: Lenition to palatalized Ø 98 3.2.1.5: Interim conclusion II 102 3.2.2: Phonological environment for Irish lenition 103 3.2.2.2: Past tense and a Ø prosodic word 109 3.2.2.3: Irish syntax- interface and a Ø word 110 3.2.3: Summary 122 CHAPTER 4: Conclusion 124 5: References 126 6: Appendices 136

vii

6.1: Optimality theory: The basics 136 6.2: Scans of Steele (1775: 25-8) (reproduced from Google eBook) 139 6.3: Scottish Gaelic nasal fricatives: the phonology/ interface 143 6.4: Index of constraints (alphabetical) 147 6.5: Index of languages (alphabetical) 152

viii

List of Numbered Examples

1) Syntactic Ø 2 2) Syntactic copies and Ø 2 3) Comparison of prosodic hierarchies 6 4) The prosodic hierarchy (preliminary sketch) 7 5) Irish minimal pairs (stress) 8 6) Syntactically derived Irish minimal pairs (stress) 9 7) Somali consonant contrastive hierarchy 12 8) Possible feature matrix for Somali phonemes 12 9) Definition of featural Ø (Ø-specification) 13 10) Definition of prosodic Ø (empty elements) 13 11) Comparative F2 value for Blackfoot silent 15 12) Comparative tongue height for Blackfoot silent morphology 16 13) Comparative lip aperture for Blackfoot silent morphology 16 14) The commonly posited prosodic hierarchy 18 15) H-aspiré compared to true vowels 21 16) Whole exception hypothesis (WMEH) 22 17) ANCHOR (word, σ R/L)h-aspiré 24 18) ANCHOR (word, σ R/L) 24 19) Enchaînment in h-aspiré & true vowels 24 20) CV 25 21) ONSET 25 22) DEP I-O 26 23) MAXIMALITY 27 24) MAX I-O 27 25) MPARSE 28 26) Under-application of a counter feeding rule due to Ø 28 27) The phonemic inventory of Blackfoot 30 28) ENHANCE-R φ 30 29) Vowel reduction to Ø in Blackfoot (first attempt) Mistapoota 'go away' 31 30) ENHANCE-R φ (revisited and specific to Blackfoot) 31 31) Spectrogram of Mistapoota 'go away' 33 32) Defective syllables in Friulian 34 33) /CVCαCβ/ → [CVVCα] in Friulian 35 34) *μ / –VD_STOP 35 35) Hierarchy for consonant voicing patterns in Friulian 36 36) SPREAD-R-μ 36 37) CRISPEDGEσ 36 38) Σ-BIN 37 39) An empty mora in Italian 37 40) Filling the empty mora in Italian 37 41) ANCHOR I-O 38 42) Minimal pairs in Standard Chinese 39 43) ONSET 40

ix

44) DEP I-O 40 45) Mono-syllabic words followed by a Ø beat in English verse 41 46) Poetry skewing rhyme 41 47) Weight to stress principle (WSP) 42 48) Tableau for a Ø output 43 Nuc 49) FILL 43 50) NUC 44 51) Mono-syllabic words followed by a Ø beat in English verse 44 52) A Ø foot in 18th century English 45 53) The prosodic hierarchy (revised for recursive word structure) 47 54) Minimal word size in English 48 55) WORDSIZE(μμ) 48 56) MATCHWORD 48 57) NONRECURSIVITY (NO-REC) 48 58) EXHAUSTIVITY 48 59) Constraints governing ω formation in English 49 60) *STRUC 50 61) MATCHPHRASE 50 62) A Ø phrase in 18th century English 51 63) V-to-T imperatives in Irish and vacuous syntactic structure 52 64) Shortening in Kimatuumbi 53 65) Phrasal tone insertion in Kimatuumbi (constraint) 53 66) Phrasal tone insertion in Kimatuumbi (schematization) 53 67) MATCHCLAUSE 54 68) Pitch reset in intonational phrases 55 69) CP subjects in English 56 70) Subject drop in adult English 'diary speech' 56 71) Article drop in adult English 'diary speech' 56 72) Subject + modal/ deletion 57 73) CP subject elision 57 74) Generalization of ι-boundary phenomena 58

75) CRISPEDGEι 58 76) Universal ranking of ι constraints 58 77) Contrastive focus on a Ø ι 59 78) Criteria for prosodic Ø 61 79) Vowel ~ Ø alternations in Russian 62 80) Under-application of vowel ~ Ø alternations in Russian 62 81) Russian vowel ~ Ø alternations with /o/, /ə/, and /e/ 63 82) Russian /o/, /ə/, and /e/ that resist alternation with Ø 63 83) Representation of an exceptional word under Jer deletion 64 84) *COMPLEX 65 85) The sonority sequencing profile (SSP) 65 86) The sonority scale (least to most sonorous) 65 87) Complex consonant clusters in English 65 88) SSP violations in French 66

x

89) Representation of an exceptional word under Jer ~ Ø alternation 67 90) EXHAUSTIVITY 67 91) DEP I-O 67 92) Representation of [ljón] 'linen.NOM' / [ljn-a] 'linen.GEN' without mora 68 93) DEP-μ-IO 68 94) ONSET 68 95) Tableau for [ljón] 'linen.NOM' (exceptional analysis) 68 96) Tableau for [ljn-a] 'linen.GEN' (exceptional segment analysis) 69 97) Whole morpheme exception hypothesis (WMEH) 69 98) *NUC/ə 70 99) *MID 70 100) MAX V 70 101) Tableau for [mestj] 'vengeance' (noun) (WMEH analysis) 70 102) Tableau for [mst-itj] 'avenge' (verb) (WMEH analysis) 70 103) Tableau for [mest] 'place.GEN.sg' (WMEH analysis) 70 104) Tableau for [mest-o] 'place.NOM.sg' (WMEH analysis – first attempt) 70 105) *MIDL 71 106) Tableau for [mest-o] 'place.NOM.sg' (WMEH analysis – final attempt) 71 107) Positional generalization 71 108) Jers in word-final syllables 72 109) Jers in word-initial syllables 72 110) Passive selection rules in Seri 74 111) Unexpected realizations in Seri 74 112) CV tier schematization of Seri 'be hit' and 'be felt' 74 113) Glottal stop realization in Seri passives 74 114) Paradigm of 'know' and 'give' in Seri 76 μ 115) *V$V [Low] 77 116) SPREAD-R(V-Place, Sonority) 77 117) IDENT(place) 77 118) Derivation of Seri independent future 'know' and 'give' 77 119) *[+Son, DOR] 77 120) Derivation of Seri 'know' independent future and emphatic realis 78 121) Tableau for Seri 'know' in the emphatic realis 78 122) Mora de-linking vs. mora deletion 79 123) [aa] vs. [aː] in Seri 80 124) Positional generalization 81 125) CRISPEDGE[PCat] 82 126) Representation of a Jer ~ Ø alternation: [ljna] 'linen.GEN' 83 127) Representation of [ljón] 'linen.NOM' / [ljn-a] 'linen.GEN' without mora 83 128) Tableau for [ljØna] 'linen.GEN' with the null parse (first attempt) 84 129) Prosodic parse of illicit onset clusters in Russian 85 130) Irish consonant alternations under lenition 89 131) MORPHREAL 91 132) MUTAGREE 91 133) LENITE 92

xi

134) LAZY 93 135) IDENT(continuant) 93 136) MAX I-O 93 137) Lenition triggers: an 'very' and ró 'too' 93 138) ró + /t/-adjectives 94 139) Coronal blocking 94 140) Geminate inalterability 95 141) Lenition as coronal de-linking 95 142) CORONALHOMORGANICITY (CORHOM) 95 143) IDENT(place) 95 144) CORONALSECOND (COR2) 95 145) Tableau for an gheal 'very bright' 96 146) Tableau for an tiubh 'very thick' (first attempt) 96 147) Tableau for an tiubh 'very thick' (final attempt) 97 148) Lenition triggers and /f/-initial adjectives 97 149) Alternation of /f/ R Ø under lenition 98 150) Lenition of /fj/ 99 151) [±Nasal] as a contrastive feature in Irish 100 152) [±Back] as a contrastive feature in Irish 100 153) Derivation for the lenition of /fj/ – an fhliuch 'very wet' 101 154) Constraint hierarchy governing Irish lenitions 102 155) Morphosyntactic environments for Irish lenition 104 156) Gender of possessive pronouns and lenition 105 157) Gender of nouns and lenition 105 158) Lenition environments 106 159) MATCHPHRASE 106 160) MATCHWORD 106 161) Merger inside DP 107 162) Merger outside DP 107 163) Lenition environments in Vulgar Latin vs. Old Irish 108 164) Past tense of the regular verb 109 165) V-to-T movement in Irish (first attempt) 110 166) Verbal morphology by tense in Irish 111 167) Double tense morphology in the conditional mood 112 168) V-to-T movement in Irish (effects of linearization) 113 169) The reduplicated preterite in Old Irish 114 170) Reduplication and lenition in Old Irish 114 171) Phonological forms of reduplicated preterites in Old Irish 115 172) Reduplication to fill a void in Old Irish 116 173) Modern Irish d' prefix in vowel-initial past tense verbs 120 174) Modern Irish sonorant-initial past tense verbs 121 175) Modern Irish /f/-initial past tense verbs 121 176) Sample tableau 1 137 177) Sample tableau 1b 137 178) Constraints & definitions 138

xii

179) Sample tableau 1c [ə.hɪn.in] a hiníon 'her daughter' 138 180) Nasal fricatives in Irish 143 181) Nasal fricatives in Tory Island Irish 143 182) Nasal vowels adjacent to phonemic nasal consonants 144 183) Nasal fricatives in Applecross Gaelic 144 184) Scottish Gaelic /m/~/b/ minimal pairs under lenition 145

xiii

List of Abbreviations and Symbols

List of phonological symbols List of syntactic symbols Ø Zero – a cognitive place holder Ø Phonetically null SPELLOUT  Null parse < > Orthographic form / silent copy R Radical X(P) Simultaneous head and phrase [F] Phonological feature XP A syntactic phrase X Segment Xo A syntactic head C Consonant C Complementizer V Vowel V Verb μ Mora (timing unit) v Light verb σ Syllable T Tense Σ Stress foot A Adjective ω Prosodic word N Noun φ Phonological phrase n Light noun ι Intonational phrase D Determiner υ Phonological utterance Q Quantifier / / Mental representation NOM Nominative [ ] Phonological output (or) [φ] GEN Genitive ( ) Optional (or) (ω) / (Σ) DAT Dative L or [X Left edge boundary VOC Vocative R or X] Right edge boundary MCAT Morphological category $ Morpheme boundary M Morpheme Ons Syllabic onset dim Diminutive Rh Syllabic rhyme def Definite (article) Nuc Syllabic nucleus poss Possessive Cod Syllabic coda fem Feminine GEN Generate masc Masculine  Winning candidate (output) sg Singular  Incorrect output predicted pl Plural | Not ranked relevant to neg Negative » Crucial ranking (domination) FUT Future (tense) OT Optimality Theory PST Past (tense) G&C Government and Charm VSO Verb-Subject-Object ET Exemplar Theory L1 First language MT Match Theory SSP Sonority sequencing profile WSP Weight to stress principle RS Raddoppiamento sintattico WMEH Whole morpheme exception hypothesis

xiv

Whether you see it or not is irrelevant; I can't see sub-atomic particles, but never-the-less, they're there. –Sheldon Cooper (The Big Bang Theory S04E03)

[O]ne might take the tip of the pencil and magnify it. One reaches the point where a stunning realization strikes home: The pencil tip is not solid; it is composed of atoms which whirl and revolve like a trillion demon planets. What seems solid to us is actually only a loose net held together by gravity. Viewed at their actual size, the distances between these atoms might become leagues, gulfs, aeons. The atoms themselves are composed of nuclei and revolving protons and electrons. One may step down further to subatomic particles. And then to what? Tachyons? Nothing? Of course not. Everything in the universe denies nothing; to suggest an ending is the one absurdity. –Stephen King (The Gunslinger: 288)

An idea that is not dangerous is unworthy of being called an idea at all. –Oscar Wilde

xv

Introduction | 1

1. Introduction Zero (Ø) is an amazing thing. If it is there, you might never know it and if it leaves, you might never miss it. In a typical savings account, interest is paid at a rate of 0.250%. The Ø before the decimal point is doing very little; we all understand 0.250 and .250 to be equal. Similarly, the Ø at the end adds only trivial information. Deletion of either of these integers will not significantly affect the bank or the account holder. In another position, however, Ø can be very important. Note the striking difference of interest paid out between the hypothetical savings account and a chequing account at the same institution where the percentage of interest on a chequing account would be 0.025%. In the savings account, you are making 25 cents on $100 whereas in the chequing account you make only 2.5 cents on $100. That might not seem like much, but with each additional Ø on the principal balance, the divide grows further and further apart. Clearly, Ø is of great consequence in finances, but what is it actually doing? What is the role of Ø? Ø is, by and large, a mathematical placeholder. It exists for the purpose of accuracy in the conveyance of information. It is not limited to banking and finances, but to all forms of operations. The advent of Ø in mathematical theory allowed the advancement of that theory by understanding that the position of integers is as important as their value. Prior to would both ٩∩∩|||| or ||||∩∩٩ ,a positional system, in an early Egyptian numbering system be indicative of 124. It was only following the advent of this system that mathematical theory advanced (Bourbaki 1998, Ifrah 2000, Kaplan 2000, Seife 2000, Barrow 2001). But what does all of this have to do with language? Language seems a far cry from math (see Kornai 2008), so why do we need to be concerned with Øs in natural language? Just as Øs in mathematical operations are placeholders to give a certain value to surrounding integers, Øs exist in language as what I will call 'cognitive placeholders.' The purpose of these cognitive placeholders is not to prevent the confusion between rates of interest or any such mathematical difference. The failure of these Øs to be understood is not likely to lead to bankruptcy or financial gains. However, Øs in language do exist for the clarity of a conveyed message, and may be responsible for the over or under-application of various linguistic generalizations. Just like in mathematics, Ø will be recognized not for its own properties, but its effect on neighboring segments.

Introduction | 2

A classic example of Ø as a linguistic placeholder is that of optional relative pronouns in syntactic theory. Take for example the sentences below in 1.

1) Syntactic Ø (Kim 2004: 366) a. What do you think that Mary read? b. What do you think Ø Mary read?

Given that syntactic structure is a constant, and that there must still be a complementizer in the 1b sentence, in order to get the same reading as the sentence in 1a, we must posit a null complementizer in 1b – represented by Ø. Another example of Ø in language is any instance where cliticization is expected to be possible, but ungrammatical when attempted. For example, take the following three sentences:

2) Syntactic copies and Ø (Kim 2004: 360) a. I wanna learn English (cp: I want to learn English)

b. Whoi do you want to learn English? c. *Who do you wanna learn English?

There are various explanations of the ungrammaticality of the sentence in 2c; however, a common account is the presence of a silent copy between the clitic to and host want left by Wh-movement of who to C. Because this silent copy intervenes between the potential clitic and host, cliticization of the two elements cannot take place. The question then becomes: does a silent element in the syntax necessitate the existence of a silent element in the phonology? This will be a recurring question throughout the remainder of this work, but concerning the sentences in 2, and similar examples, Kim (2004: 357) argues that what prevents the phonological cliticization of want and to in 2b and 2c is that unlike in 2a the two elements fall into separate intonational phrases, and not the intervention of a silent copy. Nespor & Vogel (2007: 56) reach a similar conclusion stating that "it is not the syntactic properties of the constructions under consideration, but rather their phonological properties that are responsible for the application or nonapplication of [various phenomena]." For Nespor & Vogel, as well as for Kim, the explanation for the pattern in 2, and phenomena like it, can be due to something other than Øs in phonology. However, Øs have always been as important to phonological theory as they have been controversial.

Introduction | 3

Work in the first part of the 20th century work such as Jakobson (1931, et seq.) or Trubetzkoy (1939) relied on the zero-specification of phonological features in a three-way opposition with positive and negative features to explain language-specific contrast as well as the under-application of certain phonological rules in Czech and French. The most notable work on this topic which is referred to as effectively an end to Ø as a featural specification possibility was Stanley (1967). Stanley concluded that rather than a Ø- specification for a given feature, redundant features are recoverable by morpheme structure rules (1967: 394) in the phonetics and not contrastively specified as Ø. Dresher (2009) notes, however, that this did not actually put an end to Ø in phonological theory, but rather, simply changed its shape. I will return to Stanley's (1967) contribution to Ø in later sections discussing Ø in feature theory (section 1.3) as well as in relation to Gouskova's (2012) whole morpheme exception hypothesis (sections 2.1 and 3.1.2). As Hartmann et al.. (2008: 2) sums up, "[e]mptiness has always played an important role in the literature in phonology most prominently in the literature on syllable structure." In fact, Ø has been proposed for more than half the phonological hierarchy (Nespor & Vogel 2007): at the level of the feature (Jakobson 1931, Trubetzkoy 1939, Archangeli & Pulleyblank 1994, Dresher 2009), segment (Marlett & Stemberger 1983, McCarthy 1994, Marlett 1997, Prince & Smolensky 2004, Hannahs 2009, Windsor 2010a), mora (Kenstowicz 1994, Yearley 1995, Nagy 1995, Absalom et al. 2002), syllable (Kenstowicz 1994, Duanmu 2000, Gussmann 2002, Lahiri & Plank 2010), the stress foot (Lahiri & Plank 2010), the prosodic word (Windsor 2010) and even the prosodic phrase (Steele 1775 qtd. in Lahiri & Plank 2010) (among many others). However, simply because Ø has been posited in all of these instances does not make it any less controversial. Especially in recent times, many authors since Stanley (1967) have argued for analyses which either did not rely on Ø or in fact banned the use of Ø all together: Côté 2008, Gouskova 2012, Gouskova & Becker to appear among many others—most notably those working in perceptual phonology: Flemming 1995, Jun 1995, Steriade 1999, Boersma 1998, Hume & Johnson 2001, Hayes et al. 2004 to name a few. The contention with Ø can be largely summed up in the title of an internet document introducing a two-level model of phonology: "with zero you can do (almost) anything" (Antworth 1991) which overtly states that despite using Ø derivationally (McCarthy 2007), "[Øs] are provided by rule application mechanism and

Introduction | 4 exist only internally; that is, [Øs] are not included in input forms nor are they printed in output forms." Though because of this exceptional power, many authors would reject Ø even internally. In this study, I will investigate Ø and its importance at each level of the prosodic hierarchy in an Optimality Theoretic framework (OT) (Prince and Smolensky 2004).1 At each level from feature to utterance, Ø must be accounted for. This is not to say that Ø is a viable structure at every given level of prosodic structure but that Ø must be accounted for at all levels. This includes levels such as the phonological utterance where Ø is not attested as well as theoretically unmotivated. Before delving into this topic, however, it is necessary to assume a particular theoretical model through which to look at Ø and its instantiations.

1.1. Phonological frameworks In this section I will argue for choosing a particular framework within which to couch the discussion of various phenomena which have relied on Ø. Before proceeding along this line, however, it is first pertinent to set up a dichotomy between two types of phonological theories – grammatical theories, and representational theories. According to Anderson (1985), linguistic explanation lies somewhere between theories of and theories of representation. The purpose of this thesis is to present a set of a-theoretical criteria which may be applied to instances of suspected Ø to evaluate the plausibility of invoking this tool in a particular environment and language. In order to establish these criteria, I must investigate various phenomena which rely on Ø. This requires a theory of grammar and of representation. Though I will ultimately argue for the use of OT phonology as the theory of grammar I adopt, the implementation of the four criteria is not dependant on use within this framework. Aside from the theory of grammar I adopt for this thesis, a theory of linguistic representation will also be required as a basic assumption for what the linguistic structure that phonology concerns itself with looks like – namely which model of the prosodic hierarchy will be assumed herein. Likewise, the evaluation criteria for Ø is not dependant

1 Because I will not go into great detail on the mechanics of OT, if the reader is unfamiliar with this theory they are directed to appendix 6.1 for a short discussion of the mechanics of the theory and on how to walk through OT tableaux, a medium of illustration that will be utilized frequently in the body of this work.

Introduction | 5 on a particular theory of prosodic representation and so the evaluation of the phenomena discussed herein will use a maximally inclusive hierarchy outlined below in 4. Certain phonological frameworks are more or less restrictive than others with regards to how much leniency they afford to Ø. For example, Government & Charm Phonology (G&C) (Kaye, Lowenstamm & Vergnaud 1985 et seq.) requires the use of Ø as a word-final empty nucleus in order to govern a final (preceding) consonant or, more generally, to break up word internal consonant clusters. As such, the role of zero in this theory is well described and theoretically necessitated. On the opposite end of the spectrum, a perception-based approach such as Exemplar Theory (ET) would advocate that something must be present in the signal in order for it to be learned, and subsequently generalized (Pierrehumbert 2001). Since Ø is not a physical manifestation within the perceived acoustic signal, ET would hold that it could not be heard and therefore could not be learned. As I began, if Ø is present, you may never know it, and if it is absent, you may never miss it. How can such a creature be recognized in the input in order to be learned? Of course, this question will need to be answered by any theory that allows Ø, but as such, it seems that ET would afford very little room for Ø to exist in phonological representations. Conversely, in Optimality Theory (OT) (Prince and Smolensky 2004) Ø is less well defined but still utilized by many authors (Duanmu 2000, van Oostendorp 2000 et seq., Absalom et al. 2002, Hannahs 2009, Windsor 2010 et seq., Wolf & McCarthy 2010, among many others). What is needed, then, is not a rehash of Øs as understood by G&C Phonology, but a working model of how to evaluate Ø as a useable phonological tool: whether the presence of Ø at a given level of the prosodic hierarchy is theoretically well-motivated, and what evidence for that Ø would look like. In OT, Ø is not a theoretical necessity as it is for G&C phonology, nor is it theoretically ruled out as it is for ET. OT will ban Ø from surfacing by way of constraints such as FILL (Tesar & Smolensky 2000: 31, Prince and Smolensky 2004: 2 29) or WORDSIZE (McCarthy and Prince 1986 [1996]). However, because all constraints are violable under the strict domination hierarchy (Prince and Smolensky 2004: 3), OT predicts that at least in some instances, the above mentioned constraints could be violated,

2 Although the constraint WORDSIZE actually pre-dates the release of the OT manuscript in 1993, I will argue in sections 2.5 and 3.2.2.3 that it is still a viable constraint. WORDSIZE essentially prevents a word that was too light from being realized as a prosodic word (see section 2.5 for a discussion of this constraint and its effect on English determiners). The FILL family of constraints, on the other hand, penalizes empty structure.

Introduction | 6 and Ø be allowed to surface. As Tesar & Smolensky (2000: 31) note: "Optimal structures may have empty structure, in violation of FILL, only when that is necessary to avoid violation of higher-ranking constraints." For these reasons, the discussion to follow will be couched in an OT framework. As noted above, in addition to selecting a particular phonological framework, a certain representational model of the prosodic hierarchy must also be chosen. Three major prosodic hierarchies stand out to be chosen from: that of Nespor & Vogel (2007), Selkirk (1986), and Beckman & Pierrehumbert (1986). Other theories of the structure of the prosodic hierarchy such as Halle & Vergnaud (1987) or Gussenhoven (1988) are very similar to one of the above mentioned theories or another. Of the three theories to be considered then, Beckman and Pierrehumbert's (1986) contribution may be overlooked for the sole fact that it is too restrictive for the present study, being concerned with only three levels of prosodic organization (Full Intonational Phrase, Intermediate Intonational Phrase and the Accentual Phrase). Comparing the other two we see that they are quite similar.

3) Comparison of prosodic hierarchies. (Shattuck-Hufnagel & Turk 1996: 206)

Nespor & Vogel Selkirk Beckman and Pierrehumbert Utterance (Utterance)

Inton. Phrase Inton. Phrase Full Inton. Phrase

Phon. Phrase Major Phrase Intermed. Inton. Phrase

-- Minor Phrase Accentual Phrase

Clitic Group --

Prosodic Word Prosodic Word

Foot Foot

Syllable Syllable

Mora

In an attempt to be maximally inclusive, but also remain maximally distinct in utilizing and naming all levels of the prosodic hierarchy, I will utilize the structure and naming conventions of the Nespor & Vogel (2007) hierarchy, but to the inclusion of levels

Introduction | 7 below the syllable. The levels to be considered below the syllable will be the mora, segment, and phonological feature. The representation of the prosodic hierarchy that I will employ is represented below.

4) The prosodic hierarchy (preliminary sketch – revisited in section 2.5)

The phonological utterance υ

The intonational phrase ɩ ɩ

The prosodic phrase φ φ φ

The clitic group C C C C

The prosodic word ω ω ω ω

The stress foot Σ Σ Σ Σ Σ

The syllable σ σ σ σ σ σ σ σ σ

The mora μ μμ μ μμ μ μμ μ μ μμ μμ μ

The segment [ðə ɹoʊd tə hɛɫ ɪz peɪvd wɪθ ædvɛɹbz] "The road to hell is paved with adverbs" -Stephen King (2000: 125)

The hierarchy represented above is meant as an example only. Parsing from one level to the next is done with considerations of stress placement from a casual recitation of the above sentence as it would be constructed in my own dialect. Below the level of the segment resides the phonological feature (not depicted). While they will not be the subject of this thesis, as the most widely discussed phonological home to Ø (Dresher 2009 and references therein), they must be at least briefly discussed. I will return to this in section 1.3. First, however, returning to the question of whether a silent element in the syntax necessitates the existence of a silent element in the phonology, certain assumptions of syntactic theory and the interface between it and phonology must also be made.

Introduction | 8

1.2. Interface and syntactic frameworks Let us briefly return to the question of what prevents cliticization in a sentence such as that in example 2 Whoi do you *wanna/ want to learn English? Is it the presence of a silent copy in the position, or is it the division of the utterance into two intonational phrases as Kim (2004) suggests? Once more, to be maximally inclusive of possible explanations, for the remainder of this work it is necessary to assume a theory of transformational grammar in order to achieve a possible solution to the above question wherein a phonologically null copy is responsible for blocking the cliticization of the two syntactic elements. Thus, I will assume a minimalist syntactic theory (Chomsky 1995 et seq.) for the remainder of this work. The next step in attempting to answer the above posed question is choosing a theory that relates syntax to phonology. Three choices stand out for the relation between syntax and phonology: Truckenbrodt's (1999) WRAP XP theory, Seidl's (2001) Minimal Indirect Reference, or Selkirk's (2009; 2011) Match Theory (MT). What separates Selkirk's MT from the others is that it is a theory of direct reference3 between the syntactic and phonological structure so that the constraints that govern phonological realization may make implicit reference to syntactic categories and domains. Dividing the interface theories into these two camps allows the motivation of MT and direct reference over the others. Owing to examples such as those in 5, Windsor (2011) argues in favor of MT over WRAP XP and therefore direct reference over indirect reference.

5) Irish minimal pairs (stress) (Green 2008: 199, Windsor 2011: 6) a) ((ˌkjan)ω (ˈkatj)ω)ω b) [(ˈkjan)ω (ˈkatj)ω]φ head cat.NOM head cat.GEN 'long-eared owl' 'head of (a) cat' In the above minimal pair from Irish the crucial difference between the two forms is the stress exhibited on the first word – secondary stress in 5a, but primary stress in 5b. Clearly there is no phonetic difference in segments between the word for 'head' in 5a and 5b. One possibility is that there is a difference in moraic structure in the representation of the forms, but as Ní Chiosáin (1991: chapter 4) outlines, the preferred syllable template in

3 I refer to MT as being a theory of direct reference because of its ability to mirror syntactic structure in phonological structure, but this theory is technically an indirect reference theory in that it makes use of violable phonological constraints in an attempt to force isomorphism between the syntax and phonology.

Introduction | 9

Irish is bi-moraic wherever possible. Therefore moraic structure cannot account for the stress difference in the above minimal pair either. The only other possibility for accounting for this minimal pair is the way in which each is separately parsed into the prosodic hierarchy. Whereas in the WRAP XP theory only phrases (phonological and syntactic) are visible at the interface, the difference between the above forms would need to be attributed to some prosodic constraint on the shape of phonological words (ω). Since there are no segmental or moraic differences between the two, it must be the difference of word structure. This could be accounted for in the WRAP XP theory by assuming lexical stress, and therefore stored recursive word structure; however, while this could explain differences such as that in 5 it does not offer a predictive theory of when such stress differences should arise. On the other hand, by assuming MT and direct reference, such forms are not only explained, but also predicted based on the constraint MATCHWORD. Undominated, this constraint parses the form in 5b, 6b as two phonological words as they each occupy separate syntactic heads. The form in 5a, 6a, being a compound noun (Green 2008: 199, Windsor 2011: 6), is a complex noun head incorporating two separate noun heads under one terminal (nominal) node (Baker 1988). The different syntactic representations can be seen below in 6.

6) Syntactically derived Irish minimal pairs (stress) a) No b) nP

No No no N(P) ((ˌkjan)ω (ˈkatj)ω)ω [(ˈkjan)ω (ˈkatj)ω]φ head cat.NOM head cat.GEN 'Long-eared owl' 'head of (a) cat'

Thus, whereas Truckenbrodt (1999) can describe the robust secondary stress phenomenon in Western Irish, the predictive power of Selkirk's MT, and therefore direct reference, is needed to achieve explanatory adequacy (Chomsky 1965). Based on the above evidence, I will assume a minimalist syntax, and MT as my theory of the interface between that syntax and OT phonology.

Introduction | 10

1.3. Ø in feature theory Feature theory in the first half of the 20th century such as that employed by Jakobson & Lots (1949) for French phonemes allowed the specification of features such as Vocality, Gravity, and Continuousness to fall into three categories. With respect to Vocality, the phoneme /g/ would be specified as a minus value, whereas /o/ would be specified as a positive value. Liquids on the other hand, /r/ and /l/, could be vocalic or not, and were listed as [±Vocalic]. However, in any given context, a particular phoneme could only be plus or minus despite being listed as able to be employed as either one or the other. With respect to a feature such as Continuousness, oral stops such as /g/ or /k/ would obviously be specified as [–Continuant] whereas fricatives such as /ʒ/ or /ʃ/ would be specified as [+Continuant]. The final category that a phoneme may be specified as with regards to a particular feature was a zero-specification. According to Jakobson & Lots, nasal stops having an oral closure, but continuous articulation were specified as [ØContinuant]. The three-way specification possibility was not uncommon practice in that time period and Ø was coming to be used to explain a plethora of phonological phenomena. However, Ø was quickly realized as perhaps a too-powerful tool for the segmental phonologist, and was being relied upon almost like a crutch. When other solutions failed, throw in a few Øs and all would be well. This dependency on Ø was the motivation for the influential paper by Stanley (1967), what some might refer to as effectively an end to Ø in phonology. Stanley argued that P(honological) rules "operate only on fully specified matrices" (1967: 398). In his argument, 'redundancy rules' were used to fill in any featural blanks in a given environment so that the P rules would not have to deal with Ø. As he states, in this argument, "redundant [… means] 'can be left blank and later predicted.'" (433). It is the "'improper use of blanks' that gives ternary power to a binary system" (Stanley 1967 qtd. in Dresher 2009: 97). The "danger" here is that "binary features will be used in a ternary manner, whereby [Ø] contrasts with both + and −" (Dresher 2009: 96). Of course, the important part of this statement is "contrast." As Saussure said, "the sound of a word […] is not in itself important, but the phonetic contrasts which allow us to distinguish that word from any other," (1916:166 qtd. in Dresher 2009: 1). In other words, what is not contrastive is, as Stanley states – redundant. Dresher goes on to build his theory of featural phonology from these two perspectives: being mindful of the exceptional power (danger) of Ø being

Introduction | 11 used in a three way contrast with + and −, as well as the need for an explanatorily adequate system of phonological contrast. Dresher outlines his system as one in which not all features need to be specified in a given phoneme, because not all languages will make use of all phonological features. This both allows for Ø specifications, but not in contrast with + and − values. One example of a featural Ø which is not in contrast with specified features comes from Somali (Flynn et al. 2012). Blevins (2006a) argued that Somali had a phonological process of voicing word- final stops contra to what is predicted by markedness theories. This position was countered by Kiparsky (2006) who argued that the distinction in Somali was one of aspiration and not of voicing. This controversy arises based on the wide variability in the phonetic realization of stops in Somali. Intervocalically, non-geminate stops could lenite to their voiced, continuant counterparts; and in word final position stops may be optionally produced as glottalized, unreleased, or ejected. While Kiparsky (2006) argues that any instantiation of voicing is phonetic enhancement of a phonological gesture (a la Keyser & Stevens 2006), Blevins (2006b: 4) contends that "designation of voicing as "phonetic implementation" is purely arbitrary. Voicing could just as well be the rule, with glottalization and devoicing attributed to phonetic implementation." In the study conducted by Flynn et al. (2012), spectral tilt, F0 onset, preceding vowel length, stop closure duration, closure voicing, and voice onset time of Somali stop consonants were measured where possible in various phonological environments. It was the conclusion of this study that the relevant phonological contrast was [±Spread Glottis] as argued by Kiparsky (2006). This evidence was used to provide the following contrastive feature hierarchy (using Dresher's (2009) system) for Somali consonants.

Introduction | 12

7) Somali consonant contrastive hierarchy

[+Cons] [–Cons]

[+Son] [–Son] /ʕ ħ h ʔ/

/l r j w m n/ [+Cont] [–Cont]

[COR] [PER] [COR] [PER]

[+Ant] [–Ant] [LAB][DOR] [+Ant] [–Ant] [LAB] [DOR] /ʃ / [+SG][–SG] [+SG] [–SG] [+SG] [–SG] / / / /tʃ / / k [+hi] [–hi] / / / /

"Attempting to insert [±Voice] at any level, above or instead of, any other feature will make incorrect predictions for the language" (Flynn et al. 2012: 12). Based on the above contrasts, any given phoneme in Somali may be said to have a feature matrix consisting of:

8) Possible feature matrix for Somali phonemes

/Phoneme/ ±Consonantal ±Sonorant ±Continuant CORONAL/ PERIPHERAL/ LABIAL/ DORSAL ±Anterior ±Spread Glottis ±High Ø Voice

By positing a Ø-specification of voice in Somali, we achieve Kiparsky's (2006) conclusion that any observable voicing in Somali stop-consonants is due to phonetic

Introduction | 13 implementation of phonological features. Since [Voice] is not contrastive in the language, it can be utilized by the phonetics to cue a listener to the relevant phonological feature that marks a contrast (in this case [±SG]). Phonetics 'utilize' this feature, in essence, by filling in a phonetic + or − value, filling the vacancy left by the phonology – much the same as Stanley (1967) predicted with his redundancy rules. This analysis also has the added advantage of predicting the free variation between glottalized, unreleased, and ejected final consonants as noted in Orwin (1996), Saeed (1999), Edmonston et al. (2004), Kiparsky (2006), Blevins (2006a; b), de Lacy (2006), and Conway (2008). The phonetic utilization of featural Øs by 'filling in' a phonological vacancy as discussed above is the reason this thesis will not be concerned with Ø below the level of the segment. Featural Ø and prosodic Ø (to the inclusion of the segment) function in significantly different ways. Featural Ø is a vacancy that allows phonetic enhancement to fill in the gaps so-to-speak whereas prosodic Ø is often acoustically null but exists in the form of a placeholder preventing anything from spreading into that position. Based on this observation, I will provide the functional definitions of Ø that I will assume moving forward. As was pointed out to me early on, Ø is not the typical symbol for zero, but rather it is typically used to denote an empty set. This is fortunate in that it captures exactly what is meant by zero in this thesis: Ø at any given level of the prosodic hierarchy is a bracketing structure at that level which contains no members of any of the units below. 9) Definition of featural Ø (Ø-specification) A Ø-specification is an empty slot in the featural matrix of a phoneme which may be filled in by redundancy rules.

10) Definition of prosodic Ø (empty elements) An empty element is a prosodic category which carries no pronounceable material but exists in the mental representation as a cognitive placeholder to prevent the spreading or insertion of other phonological material into that position.

In other words, featural Ø is empty but prosodic Ø is occupied by a Ø element – it has no phonetic realization, but is very much there in the representation. This division in function is also the reason that the segment is included in my discussion of prosodic Ø. Before accepting these definitions and moving forward with a discussion on where

Introduction | 14 prosodic Ø is expected one other type of Ø must also be ruled out from this study. This is acoustic Ø.

1.4. Acoustic Ø Word-final position is phonologically a weak position where we expect to see lenition occur; in some cases (if taking a large enough diachronic look) segments can lenite all the way to Ø, as is the case with final in English. What started out in Old English as [nɑme] 'name' reduced to [naɪmə], and finally the final schwa reduced all the way to Ø producing modern [naɪm]. A similar word-final weakening can be observed in Blackfoot, an Algonquian language spoken locally in Alberta, where, in some instances a word-final morpheme reduces to Ø. Unlike English, however, this morpheme is not recognizable by a frozen spelling system where the final vowel orthographically resists deletion. In Blackfoot, word final –(w)a 'proximate' and –(y)i 'obviative' are often realized as silent morphology at the end of the word. Unlike a hypothetical Ø plural morpheme that distinguishes between English deer.sg and deer.pl, the Blackfoot silent morphology is contrastive without relying on context. Teachers of this language often describe these silent as being "puffs of air" or "silent sounds" (van der Mark 2003 qtd. in Bliss & Gick 2009) and "instruct students to 'watch their mouths' in order to perceive the sounds that they are making" (ibid). In language classes here at the University of Calgary, native elder Piitaikiihtsipiimi 'Spotted Eagle' teaches these silent elements with an elaborate head nod asking students to repeat the gesture "so I know that you know it's there" (p.c. 2012). Among native Blackfoot speakers, however, the head-nod gesture is not part of the natural speech. Bliss & Gick (2009) and Gick et al. (2012) conducted a study to determine the phonological status of these silent elements. Were they the victim of phonological lenition in a weak position as Frantz (1991) claims: short vowels devoice in final position. Or are these silent elements true phonological Øs? This study looked at the words: si'káána/si'kááni 'blanket,' ki'sómma/ki'sómmi 'moon,' and miistsísa/miistsísi 'tree' in carrier sentences to elicit the production of these items with either the silent proximate –(w)a or silent obviative –(y)i. The findings of this study were that acoustically the words for

Introduction | 15

'blanket' and 'moon' did not differ in formant values between the –(w)a and –(y)i endings. The third word, 'tree,' had slight F2 and F3 differences between –(w)a and –(y)i but not significant differences as one would expect between the vowels [i] and [a]. This is summarized in the table below. 11) Comparative F2 values for Blackfoot silent morphology (Bliss & Gick 2009: 8)

This result showed that in the speech signal there were essentially no acoustic cues to the difference in morphology. Building on the native speaker instructions though, Bliss & Gick also looked at the articulatory gestures that the reduced acoustics obscured. Employing video recordings to measure lip aperture and an ultrasound machine to capture tongue position, the authors did find articulator differences in the target words that differed between –(w)a and –(y)i. For instances where the reduced morpheme was underlyingly –(w)a there was both decreased lip aperture and tongue height. In instances where the reduced morpheme was underlyingly –(y)i there was both increased lip aperture and tongue height. Both of these differences are summarized in the tables below and were largely consistent indicating statistical significance of the findings.

Introduction | 16

12) Comparative tongue height for Blackfoot silent morphology (Bliss & Gick 2009: 9)

13) Comparative lip aperture for Blackfoot silent morphology (Bliss & Gick 2009: 9)

As mentioned previously, results for the third target word, miistsísa/miistsísi 'tree,' differed slightly from the other two in terms of acoustic realization. For this word, there were slight F2 differences between the [a] and [i] vowels, but with large overlap as was seen 'blanket' and 'moon.' The major difference lies in F3 for 'tree' where F3 for [a] had a mean of approximately 3500 Hz and the mean F3 for [i] was closer to 3000 Hz. Bliss & Gick (2009: 14) conclude based on this data that "there are "soundless" vowels in Blackfoot, which are articulatorily but not acoustically distinct." I argue that this data also demonstrates that final "soundless vowels" are the victims of phonetic weakening and not instances of a phonological cognitive placeholder, or Øs. This argument stems from two

Introduction | 17 facts that the study turns out; articulatory gestures of the silent morphology are still carried out despite having little to no acoustic reality, and it cannot be categorically stated that Blackfoot silent morphology has no acoustic reality. The point I am making by distinguishing acoustic Ø from phonological Ø is that silence does not equal a cognitive placeholder. Given that phonetics is the output of phonology, another way to put it would be to say that phonetics is the acoustic realization of phonological gestures (Browman & Goldstein 1986 et seq.). Since contrastive gestures can be observed we can conclude that the silent morphology does have contrastive segments that are realized in the phonology not as Ø but as contrasting featurally-specified phonemes. Further, because Bliss & Gick observed that there were occasional differences in the acoustics, especially for one of the three target words, the acoustic reduction can be categorized as gradient (typical of phonetic processes) rather than categorical (a hallmark of phonological phenomena) (Zsiga 2000). This fact serves as evidence once again that the reduction is in the phonetic implementation, and that they are not phonological. Having established what types of Ø will be researched and the locations within phonology in which those Øs will be explored, I now turn to a discussion of where phonological Ø is theoretically motivated and what evidence of a Ø at that level would look like.

Theoretical motivations for Ø | 18

2. Where is Zero expected in phonological theory? This thesis assumes a view of the prosodic hierarchy as presented in Selkirk (2011: 437) in what she calls the "'standard theory' of prosodic structure [... wherein] the ordered set of prosodic category types" are as depicted in 14 below. In addition to Selkirk's hierarchy, I also include the mora and segment for reasons discussed in section 1.1. 14) The commonly posited prosodic hierarchy (Selkirk 2011: 437) Intonational Phrase (ɩ) Phonological Phrase (φ) Prosodic Word (ω) (Stress) Foot (Σ) Syllable (σ) Mora (μ) Segment (X)

This section looks at each of the above prosodic categories under an OT (MT) lens in order to make predictions as to where Ø is a viable structure and where we might expect to find examples of Ø in natural language. I will begin with the segment and move up through the hierarchy to the intonational phrase discussing the plausibility of a null prosodic structure at each of these levels. As opposed to the hierarchy given previously in 4, this representation of the prosodic hierarchy omits the clitic group4 and the phonological utterance.5 Within the literature currently, there is no shortage of phonological phenomena which have been analyzed as containing a prosodic Ø. Each of these analyses, either for or against Ø, relies on evidence specific to that phenomenon and how the author interprets it. In what follows, I will examine a wide scope of such phenomena and the supporting evidence that various authors give for either positing a Ø or attempting to debunk one. This survey of language data will allow generalization over several languages and phenomena to explicitly construct criteria for the evaluation of prosodic Ø which can then be applied to any phenomenon. The broad knowledge of language-specific evidence for or against Ø will in turn be utilized to evaluate each of the subsequent case studies which are used as testing grounds for the proposed criteria.

4 The omission of the clitic group will be addressed in section 2.5. 5 The omission of the phonological utterance will be addressed in section 2.8.

Theoretical motivations for Ø | 19

2.1. Ø segments In section 1.3 I discussed briefly the notion of Ø specified phonological features. When a phonological feature is specified as Ø the remainder of positively or negatively specified features in the bundle still cause the pronunciation of a phoneme. A Ø segment on the other hand has no specified features within the bundle at all at one stage in the derivation or the other. That is to say, a Ø segment could either be stored as a phoneme in the mental representation of a word or could be chosen as the winning output candidate to be handed off to the phonetic component. With these two possibilities available, we must examine constraints that both force the realization of a Ø segment and those which ban it. First, I will entertain the idea that Ø could be a stored phoneme; that is, that a Ø phoneme will enter into the phonological component alongside all other phonemes of a given lexeme. Clearly there are two families of constraints which will either ban or permit the survival of a Ø segment in the output to phonetics: MAX I-O and DEP I-O. The MAX I-O family of constraints bans the deletion of a segment and would be violated even if a cognitive placeholder, or phonological Ø, as a null segment were absolutely deleted so that even the position that the phonological Ø held did not survive to the output. Therefore, a Ø segment is a structure which still exists in the segmental tier but which contains no features to enable pronunciation. In order to delete a segment (in violation of MAX I-O) not only does the feature matrix of that segment get deleted, but so does the position (what will be referred to as the bracketing structure for higher prosodic units). To attempt to simplify this concept consider the following; there may be a set of all birds belonging to a certain genus. If those birds go to extinction, the set is reduced to a point where it contains no members, but the set still remains, it is just empty. If, however, no record of that genus ever existing remained, then scientists may never even posit the set, empty or otherwise. This would be akin to not only removing all members of a set, but in essence, deleting the set itself as well. These are the notions I am trying to establish as the difference between Ø as a cognitive placeholder as opposed to deletion.

The DEP I-O family of constraints on the other hand are dependency constraints which prohibit the epenthesis of extra prosodic information. Again, from a viewpoint of strict possibility, there are two ways in which DEP I-O could lead to the formation of a Ø. If a given bracketing structure, in this case a segment, was the stored mental representation

Theoretical motivations for Ø | 20

entering into a derivation, DEP I-O[F] could prevent the epenthesis of features into the segment and force the pronunciation of Ø (or output of a Ø to the phonetic component) rather than filling that Ø with features for acoustic saliency in pronunciation. On the other hand, although I will not argue that this is truly feasible in natural language, rather than epenthesizing a fully specified segment, languages could potentially epenthesize a Ø in order to prevent something such as vowel hiatus. The question is now that if there are four possible ways to have a segmental Ø, which of these options are well motivated theoretical options to explore in natural language and which can be dismissed. I will begin by looking at options where Ø is the stored mental representation that enters into the phonological component from the lexicon. Avoiding for the moment the question of whether or not Ø is a viable mental representation or stored linguistic unit that can be learned, if a Ø segment were in the mental representation, what would cause this not to be filled before the output to the phonetic component? Given the theory assumed by this thesis, the simple answer is a highly ranked constraint which bans that epenthesis. But what are the relevant constraints, and how would we recognize a Ø segment in the output? There are several indicators of Ø structures that I will return to frequently throughout the course of this thesis; as a launching point, I will begin with two of them: the effect of Ø on the neighboring segment(s), and contrast. The effect of Ø on neighboring segments will be recognizable only on a case by case basis due to language specific phenomena such as lenition in Irish (see section 3.2) or the lack of low vowel deletion in Seri (see section 3.1.3). at this point it is more relevant to discuss contrast. In section 1.3 I outlined briefly a theory of featural contrast (Dresher 2009) which allowed linguists to discover which phonological features were active in a given language's inventory. This is not the only concept of contrast that matters however. As stated by Saussure (1916: 166 qtd. in Dresher 2009: 1) "[t]he sound of a word [...] is not in itself important, but the phonetic contrasts which allow us to distinguish that word from any other." This is a long- standing fundamental of linguistics and it has been frequently observed that contrast (or contrast preservation) could lead to instances of phonological opacity (Łubowicz 2003 and references therein). "The main observation is that in [...] opaque mappings, a given underlying contrast is preserved on the surface but manifested as a different surface

Theoretical motivations for Ø | 21 contrast" (Łubowicz 2003: xi). Surprisingly, Łubowicz (2003: 10), in her dissertation, 6 appeals to Ø only once in her discussion of contrast and the role of GEN. She states that "[i]n addition to featural differences, the input set also contains forms that differ from the underlying form in the number of segments. Those are forms that contain fewer segments than the underlying form (via deletion), including a null set[.]" However, even though the "null set" is a brief mention in Łubowicz's dissertation, the idea of a stored Ø representation as an explanation for word-level contrasts is rampant in the literature. Perhaps the most notable (and controversial) representational Ø segment that is used to preserve contrasts is the case of French. In French, there are three phonological processes which are often analyzed as involving a Ø segment at the beginning of certain words, what Côté (2008) dubs "The French Holy Trinity": schwa, liaison, and h-aspiré. While each of these three phenomena are interesting in their own right, I will focus here only on the h-aspiré 'h-aspirated' phenomenon. The name of this phenomenon might be misleading – it is not the case that there is aspiration of initial /h/ consonants, but rather that words (often containing an in the orthographic form) somehow come to pattern with consonant-initial words even though they are pronounced as being vowel-initial. The effects of the h-aspiré can be seen with near minimal pairs in several different contexts. 15) H-aspiré compared to true vowels (Côté 2008) a) deux héros a') deux étaux 'two heroes' 'two vises' [dø _ ero] [dø z eto]

b) la hauteur b') l'odeur 'the height' 'the smell' [la otœr] [l_ odœr]

c) le héros c') l'étau 'the hero' 'the vise' [lə ero] [l_ eto]

d) beau héros d') bel étau 'beautiful hero' 'nice vise' [bo ero] [bɛl eto]

6 "The universal candidate generator [which] constructs candidate output forms […] and specifies a relation between the candidate output form and the input" (McCarthy 2002: 8).

Theoretical motivations for Ø | 22

In the above examples, 15(a-d) are instances of h-aspiré words whereas 15(a'-d') are true vowel-initial forms. The 15(a/a') pair shows the difference between the forms with regards to liaison. In French, when two words put together create a vowel hiatus context, a frequent repair strategy is to epenthesize a liaison consonant, normally [z]. Another strategy to eliminate vowel hiatus is shown in 15(b/b') where the first of the two vowels in adjacency will delete. As can be seen in 15a and 15b, when the word in question contains an h-aspiré either vowel hiatus is permitted, or is not violated for another reason. In the 15(c/c') examples, rather than preventing vowel hiatus by deletion, instead the 15c example is commonly analyzed as preventing a sequence of two consonants, [l] followed by Ø which would violate the sonority sequencing profile leading up to the nucleic vowel. Thus, when the masculine determiner [l] precedes an h-aspiré form, a schwa must be epenthesized. Finally, in the 15(d/d') examples we see instances of functional suppletion. These examples are often analyzed as showing grammatical gender disagreement (Côté 2008: 65). Even though both héros and étau are masculine nouns, étau is required to take a different form which is acoustically identical to the feminine adjective in order to prevent the vowel hiatus we see in the surface realization of beau héros. Clearly, there are differences in the realization of these forms dependant on whether they display the h-aspiré phenomenon or not which work to preserve a contrast which would otherwise be obscured by typical phonological processes. However, such a case is not without its own controversy in the literature as to whether the analyst should account for this phenomenon by positing a representational (or surfacing) Ø. Here I will present one argument against a Ø and one for and then ask the question; what would be needed to tease these two possibilities apart? Gouskova (2012) posits the whole morpheme exception hypothesis (WMEH) which will be returned to in greater detail in section 3.2.1. This hypothesis is used to analyze what have been traditionally referred to as representational Øs in Russian and states that: 16) Whole morpheme exception hypothesis (Gouskova 2012: 80) There are no lexically exceptional segments – only lexically exceptional morphemes.

Extending Gouskova's analysis to French we would analyze these forms by stating that there are no Ø exceptional segments which cause the under-application of a phonological

Theoretical motivations for Ø | 23 rule, but rather these morphemes/lexemes are stored in the mental lexicon as exceptional themselves in that they lexically prohibit certain phonological constraints from applying. One fallout of Gouskova's Russian analysis through WMEH is that it predicts the location of Ø or exceptional segments in French. As she says for Russian (2012: 86) this analysis explains why vowel-Ø alternations occur only in the final syllable of a root and never internally. I extrapolate from Gouskova's Russian analysis that what she predicts for segment-Ø alternations cross-linguistically would be that they would always be an edge effect: that alternations with Ø would either come morpheme-initial or final. In French, this pattern seems to be maintained as they are always word-initial. Côté (2008), although taking an OT approach, likewise looks at what she calls 'defective segments/structure' in French related to the three constructions which have traditionally been analyzed as Øs and claims that the alternations can be better explained through a 'perception-based alternative.' In her analysis she concludes that: "all three are more insightfully understood without use of defective segments and that syllable structure and well-formedness, which are crucial elements of the defective segment analysis, play no role" (61).7 She explains that, like the contrast preservation model, the under or over-application of a phonological process can be a perceptual trigger to aid in understanding, what I will refer to later as recoverability (Archangeli & Pulleyblank 1994). With regards to the h-aspiré words, she takes an approach very similar to what Gouskova (2012) posits for Russian: that h-aspiré forms have a special diacritic, lexically marking them as exceptional.8 From there, certain phonological processes may or may not act on this form as given by constraints which specifically target that diacritic.9 For this purpose, Côté (2008: 92) posits the constraint

ANCHOR (word, σ R/L)h-aspiré which will apply specifically at the left edge of an h-aspiré word when combined with any other word and will outrank a general, non category- specific anchor constraint.

7 See also, Pater (2004) for a comparison of lexical exceptionality with structural approaches within an OT framework. 8 This type of diacritic explanation is highly controversial, most notably with regards to Celtic initial mutations. However, despite the large number of arguments against such listings, research generally tends to return to this notion claiming that the phonological component can interact with certain morphologically encoded diacritics (Green 2007; 2008, Gorrie 2011). This will be returned to in much greater detail in section 3.2. 9 See also, McCarthy (2003) for a discussion of categorical alignment constraints.

Theoretical motivations for Ø | 24

17) ANCHOR (word, σ R/L)h-aspiré (Côté 2008: 92) At the boundary between an h-aspiré word and the preceding word, every segment at the edge of a lexical word has a correspondent at the same edge of a syllable

18) ANCHOR (word, σ R/L) Every segment at the edge of a lexical word has a correspondent at the same edge of a syllable

This idea could also be formalized in the notion of a CRISPEDGE constraint (Ito & Mester

1999) if one could argue effectively that CRISPEDGE could apply to exceptional lexemes and not just prosodic categories. The idea here being that Côté's anchoring constraints allow for "the basic intuition that h-aspiré words maintain a stronger separation between them and the preceding word" (2008: 92). While this analysis seems to be a fairly strong one which does not depend on Ø specifications, I will offer one counter argument against a particular point made by Côté with regards to this phenomenon which could point to a Ø analysis being preferable. With regards to the clear cut distinction between h-aspiré words and those beginning with a true vowel as noted in 15 above, Côté (2008: 91) points out the fact that h- aspiré words "only variably prevent enchaînement" 'sequence[ing].' 19) Enchaînement in h-aspiré & true vowels (adapted from Côté 2008: 66, 91) h-aspiré true vowel mille hibous mille îlots 'a thousand owls' 'a thousand islets' a) no effect *[mil.i.bu] *[mil.i.lo] b) enchaînement [mi.li.bu] [mi.li.lo] c) schwa insertion [mi.lə.i.bu] *[mi.lə.i.lo] d) glottal stop insertion [mil.ʔi.bu] *[ mil.ʔi.lo] e) schwa & glottal stop *[mi.lə.ʔi.bu] *[mi.lə.ʔi.lo] insertion

The above table shows that in enchaînement environments10 the phonology must enact a repair strategy whatever the relevant constraint that is violated may be. Most commonly (applying to both h-aspiré and true vowel-initial words) is elision and re-syllabification

10 Côté (2008: 91) claims this is following a consonant final word, but does not elaborate further on the prosodic or morphosyntactic conditions for elision. I would hypothesize this is likely to be an instance of a recursive prosodic word.

Theoretical motivations for Ø | 25 where the final consonant of the preceding word becomes an onset to the second word in the sequence. However, in the case of h-aspiré words, a schwa or glottal stop may be optionally epenthesized, but crucially, as the form in 19e shows, not both. In order to maintain her 'stronger left edge' hypothesis for h-aspiré forms, Côté must posit an additional constraint forcing CV sequences to be syllabified together. 20) CV (Côté 2008: 93) CV sequences are syllabified together This constraint outranks those that provide edge strengths and allow re-syllabification. On the other hand, if this is an instance of a recursive prosodic word, then there may be weaker phonological boundaries internally anyway11 and this re-syllabification could be made by a general ONSET constraint without needing to posit a new CV constraint.

21) ONSET (Kager 1999: 93)

*[σ V (Syllables must have onsets) Further to this minor oversight in her analysis, the variability in the h-aspiré words is actually predicted in a Ø onset analysis. Côte (2008: 91 and references therein) claims that the null onset analysis over-generates by expecting that the form in 19a should be possible. I do not agree with this conclusion, but even if it were the case, we see that cross linguistically a constraint banning the sequence of schwa and glottal stop is generally high ranking12 perhaps for the exact reason that Côte (2008: 73) herself identifies: Schwa serves to provide additional cues to consonants that lack perceptual salience. The likelihood of schwa omission correlates with the relative salience of the surrounding consonants (or their degree of confusability with Ø) in the absence of schwa.

As Shaw et al. (1999: 12 and references therein) note, schwa has no inherent place features, and glottal stop is specified with the least marked place feature. Therefore, either of these segments would be expected to readily alternate with Ø. However, constraints on perceptual salience would likely prevent both from surfacing in adjacency as is the case in Salish languages despite the fact that both segments are extremely frequent outside of this environment. So even if this analysis does predict both schwa and glottal stop should be

11 See Windsor (2011) for a discussion of weak prosodic word boundaries in Irish and clustering effects. 12 See Shaw et al. (1999) for discussion of this constraint in Salish or van Oostendorp (2000) for Dutch.

Theoretical motivations for Ø | 26 able to surface, then that constraint would ban that ungrammatical output. However, I contend that the aforementioned analysis actually does not predict both schwa and glottal stop should co-occur since we posit only one Ø segment in these forms. Epenthesizing one segment would not violate a general DEP I-O since one or the other would have an input correspondent, in this case Ø.

22) DEP I-O (Kager 1999: 68) Output segments must have input correspondents (No epenthesis) However, epenthesizing both would violate this constraint. By trying to insert both schwa and glottal stop, Côté is suggesting that we could epenthesize two segments where only one correspondent position in the input is available. So, does French have a Ø segment in the mental representation of its speakers? Pending further research into the exact prosodic structure of these representations (specifically word boundaries and whether or not these are recursive words) as well as research into whether the repair strategy for enchaînement is truly in free variation, or whether individual speakers have preferences as to which repair to use the answer is unclear. There is compelling evidence on each side of the debate, but at this point I don't think it is conclusive one way or the other. What French does provide at this point though, is what to look for in terms of can something be analyzed as a Ø segment representation. According to Gouskova's (2012) WMEH, exceptional morphemes should be directionally linearized with respect to particular morphemes/lexemes. We would want to find instances of h-aspiré word medially to show it was truly an exceptional segment.13 If enchaînement is part of a recursive prosodic word, while Gouskova would still argue that it was morpheme- initial, it would at least be word-medial and more could be said about the CRISPEDGE conditions in French. Further, looking at the potential [ə] or [ʔ] alternations with /Ø/, in order to posit a language has a representational Ø segment, we would also want to show

13 Tranel (1987: 35) does give examples of four French words which can be analyzed as having root-medial h- aspiré consonants which both allow what appears to be vowel hiatus on the surface as well as prevention of nasal epenthesis. The French car make, Panhard, is often pronounced as [p a ]; children reading the comic book Astérix and Obélisk often pronounce menhir as [mɛ i ] rather than [meni ] [or [mɛ ni ] (p.c. Flynn 2012)]; 'to betray' trahir is given as [trair], and 'outside' dehors' is given as [dœɔr]. Even if the number of examples is very limited, this is strong evidence that a Ø phoneme analysis is on the correct path.

Theoretical motivations for Ø | 27 that that cognitive placeholder was at least sometimes filled (in certain conditions) with some sort of segmental material. Looking at the other effects showing in 15, in order to posit a representational Ø segment, we would also want to observe clear adjacency effects in neighboring segments (at least in certain constructions). Finally, Ø should be used as a posited representation only when no other analysis can be shown to have the same predictive power that a Ø analysis shows. Looking at these four criteria, an analysis of a Ø segment in the mental representation of French speakers does seem attractive. However, at this point, based on the final criterion, it is not conclusive that Ø is the correct analysis though it may be shown to be with further research. The above brief foray into French Ø segments has provided us with four possible criteria for evaluating Ø as a mental representation. However, a segmental Ø may also arise not from being the stored representation, but derivationally. That is, Ø may result as the optimal candidate in the phonological component substituting for a fully specified representation. I will now explore the theoretical motivation for such Øs. Kager (1999) defines MAXIMALITY in very broad terms which can be further restricted to input-output correspondences, base-reduplicant correspondences, base-truncation correspondences, specific to vowels, consonants, mora, linkages, or many others.

23) MAXIMALITY (Kager 1999: 205)

Every element of S1 has a correspondent in S2 This generalized constraint is almost the same as her specific to input-output correspondence constraint, just in more simple language.

24) MAX I-O (Kager 1999: 67) Input segments must have output correspondents ('No deletion') What these two variations on the theme of anti-deletion tell us is that faithfulness attempts to keep a 1:1 correspondence between the input to and output from phonology without necessarily being concerned with featural identity constraints. Recognizing this fact, Wolf & McCarthy (2010) argue that a mapping to zero (ʘ in their convention) is more harmonic than an absolute deletion to nothing (what they characterize as ɸ). In fact, they argue that "mapping to the null output is always one option for the analyst or learner who needs to account for the failure of some known input to surface faithfully" (3). In Wolf &

Theoretical motivations for Ø | 28

McCarthy's formalization, a null parse (mapping to zero) is the result of constraints dominating MPARSE so that a phonologically optimal pronunciation cannot be achieved.

25) MPARSE (Kager 1999: 402) Morphemes are parsed into morphological constituents When this is the case, the output of the phonological component is Ø which is observable as a gap in the morphological paradigm. Here, I extend this idea to segments rather than to whole morphemes, something not ruled out by Wolf & McCarthy (2010) since they also note that this take on correspondence can be used for phonotactics as well. However, their caution is well taken: "[t]he challenge is to define the null output [Ø] in such a way that it is distinct from the candidate that has deleted all of the underlying segments" (2010: 4). Understanding this concern, the optimal evidence of a phoneme reducing to (or alternating with) Ø would be evidence in the form of the under-application of a counter feeding rule of the basic form of the derivation given in 26 below. 26) Under-application of a counter feeding rule due to Ø

Mental representation /VxCxVy/

Rule 1: Cx → Ø / _Vy VxØVy

Rule 2: Vx → Ø/ _Vy VxØVy

Phonetic realization [VxVy] In the above hypothesized construction, we outline the idea that a given language could have two basic phonological processes at work: one which deletes a certain type of consonant before a certain type of vowel, and another that deletes a vowel when it is adjacent to a second vowel. In this scenario, we predict that the deletion of the intervocalic consonant would provide a counter feeding environment where we should also see the deletion of the first vowel as well. However, due to the fact that this language's phonology didn't actually delete the segment but rather reduced it to (alternated it with) Ø to remove the offending features, there was no case of vowel hiatus and no further deletions took place. This is exactly the type of phonological phenomenon that would be predicted by, and could only be explained by a reduction to Ø but not a total deletion. One such example of this type of phenomenon may be found in Blackfoot. As discussed above, Blackfoot has silent suffixes in certain environments. I argued that this was a case of acoustic Ø and not a phonological Ø. However, this means that Blackfoot

Theoretical motivations for Ø | 29 speakers permit silent elements in their language which are contrastive. In addition to word-final silent suffixes, Blackfoot has other prosodic edge phenomena. In recording Blackfoot elder Piitaikiihtsipiimi 'Spotted Eagle' for a learning aid for our introduction to Blackfoot language class I observed that in phrase-final position a long vowel will shorten, a short vowel will delete, and a consonant will aspirate in order to mark the prosodic boundary. The question is, if a phrase-final CV string succumbs to apocope resulting in a phrase-final consonant, does that consonant aspirate? If the vowel is truly deleted, it should. If the vowel merely reduces to Ø, it should not. This is especially true since Frantz (2009: 157) states that "[r]ules apply for maximal 'feeding' and minimal 'bleeding'" except for a few instances, none of which mention phrase-final processes. In actuality, the deletion of final vowels does not cause final consonants to aspirate, so now I will examine it more closely to provide insights into reduction to Ø phenomena. Because this is not a central claim of this thesis, I will not explore in depth what constraints cause phrase-final weakening in Blackfoot. There is a possibility that this a prosodic constraint used to enhance perception by overtly marking a phrase boundary, or it could be the case that this is a lenition in a prosodically weak environment. Given that aspiration in a language which neither phonemically makes use of [±Voice] or [±Spread Glottis] (the phonemic inventory of Blackfoot is given below in 27), I will make a preliminary claim that this effect is an enhancement gesture to mark prosodic phrase boundaries rather than a lenition since it seems that the phonology is adding a feature, not weakening the sound. This will turn out not to be the correct analysis and I will later revise this to be a lenition effect, but making this initial assumption will allow for a better understanding of potential Ø-phenomena and how to confirm or debunk analyses which rely on Ø.

Theoretical motivations for Ø | 30

27) The phonemic inventory of Blackfoot14 (Elfner 2006: 12) Labial Coronal Dorsal Glottal Vowels Stops p pː t tː k kː ʔ i iː Fricatives s sː x o oː Affricates ts tːs k s k ːs a aː Nasals m mː n nː Glides w j

Given this working hypothesis, I will posit a Band-Aid perception-based constraint,

ENHANCE-R φ which will simply serve to say perceptual salience requires the overt marking of the right edge of a prosodic phrase boundary, the method by which a language does this is variable based on the language in question. For example, Italian marks the right edge of a prosodic phrase with a vowel-lengthening phenomenon (Dehé & Samek-Lodovici 2009, Langus et al. 2012), and in Blackfoot (if this working hypothesis were correct) long vowels would lose a mora, short vowels would lose their features, and consonants would gain a [+Spread Glottis] feature.15

28) ENHANCE-R φ Overtly mark the right edge of a prosodic phrase Under this analysis, where a final short vowel deletes its features but remains in the output as a Ø in order to prevent the preceding consonant from aspirating, ENHANCE-R φ is the highest ranking constraint over the anti-deletion constraint MAX I-O(vowel). MAX I-O in turn outranks the more specific anti-feature-deletion constraints MAX I-O[F]. However, MAX I-

O([Consonantal]) is clearly higher ranking otherwise consonants would also delete under this theory. Therefore, when dealing with segments that have the [+Consonantal] feature, this maximality constraint must outrank DEP I-O([Spread Glottis]). This ranking (as depicted in the tableau below) allows the vocalic features to delete but a segmental place holder to remain which takes into account Wolf & McCarthy's (2010) notion that a mapping to Ø is more harmonic than outright deletion. It also explains why consonants don't delete, but instead aspirate at the right edge of a prosodic phrase.

14 The speaker of Blackfoot I usually work with also has an [u] vowel. It is not clear to me that this vowel is in an allophonic relationship with /o/, and it may well be a separate phoneme. This question, however, is beyond the scope of this paper. 15 For a discussion of how multiple different surface phenomena can be the result of repair strategies for a single constraint see Hannahs (2011) on Welsh sonority sequencing violations and the three possible repair strategies that variably surface to avoid violation of this constraint.

Theoretical motivations for Ø | 31

29) Vowel reduction to Ø in Blackfoot (first attempt) Mistapoota 'go away'

/mistapoːta/ ENHANCE-R φ MAX MAX MAX DEP (vowel) ([Consonantal]) ([F]) ([Spread Glottis]) a) mistapoːta *! b)  mistapoːtØ * c) mistapoːth_ *! * d) mistapoːt_ *! * e) mistapoːthØ * *!

The tableau in 29 depicts the conversation immediately above: the resistance of a vowel to completely delete allows it to merely delete all of its pronounceable features which keeps the final consonant away from the right edge of a prosodic phrase and thus prevents the aspiration of the final consonant through a counter-feeding rule. From a strictly theoretical point of view, this is a perfect example of reduction to Ø without absolute deletion and this sort of under-application of a counter-feeding rule is the exact type of evidence that we would need to find in order to posit a reduction to (or alternation with) Ø that is maintained in the phonology. Unfortunately, even though the final consonant does not aspirate in this example, the phonetics will not support this analysis. The actual analysis of phrase-final enhancement in Blackfoot is actually more straight-forward.

By revising the Band-Aid perception-based constraint ENHANCE-R φ, the entire process can be simplified.

30) ENHANCE-R φ (revisited, and specific to Blackfoot) Aspirate any segment aligned with the right edge of a prosodic phrase Aspiration works on the phonological feature [+Spread Glottis] therefore, adding [+Spread Glottis] to a consonant will result in an aspirated consonant, and adding to a vowel will cause that vowel to devoice because the vocal folds will be abducted and phonation will be impossible. Given the three possible word endings in Blackfoot (long vowels, short vowels, and consonants) we can predict what this feature will do to each of these endings. Long vowels will devoice in the second half of the vowel, short vowels will devoice but remain pronounced, and consonants will aspirate. A closer examination of the phonetics will verify this prediction. The Spectrogram given below in 31 shows that there is a segment following the final [t] consonant of the elicited word. Crucially, this consonant does not show any vocal pulses (vocal fold vibrations). This is shown by the distinct lack of pulse lines in the

Theoretical motivations for Ø | 32 spectrum above what I have labeled the voiceless [ ] segment. Further, this is not simply aspiration of the preceding [t] segment as the voiceless element still shows easily recognizable formant bands showing the periodic, or vocalic nature of the segment. Of course, the question of whether this is phonological or phonetic enhancement of a phrase boundary is still open for debate. Given the theory of contrast outlined in section 1.3, phonological phenomena can only act on contrastive phonological features (Dresher 2009) which would relegate this approach to the phonetic component as Blackfoot relies on neither [±Voice] or [±Spread Glottis] contrastively. On the other hand, because the phonetics seems to be manipulating a phonological feature across different types of segments, this does seem to be a prosodic phenomenon. The point is that final aspiration in Blackfoot shows what would be the expected result of phonological reduction to or alternation with Ø. However, it also provides an excellent caution of the dangers of assuming Ø. Although an analysis relying on Ø may offer an attractive explanation of the data, there can be simpler and better explanations. This reminds us that although Ø is a plausible phonological structure, it should be a last resort only after other analyses are shown to be inadequate to handle the facts of the phenomenon.

Theoretical motivations for Ø | 33

31) Spectrogram of Mistapoota 'go away' (Piitaikiihtsipiimi, January 14, 2012)

Theoretical motivations for Ø | 34

2.2. Ø mora The idea of an empty mora is somewhat different than that of an empty segment, though closely related. Empty mora, in one sense, bring up that old argument of the chicken or the egg; does compensatory lengthening precede or follow the deletion of a coda consonant? That is to say, does a consonant delete leaving a mora empty so it must form a new linkage with a preceding vowel, or does a vowel link to two mora allowing for the deletion of the consonant? I will not attempt to answer this question here. However, I bring it up because the idea of an empty mora, at least in one stage in a derivation, is not an uncommon one in phonology, specifically with reference to compensatory lengthening. If our terminology for this type of phenomenon is telling at all, it suggests a consonant in a moraic position deletes leaving a mora filled by Ø (note, this is not a Ø segment as defined in the last section, but rather this is an empty set at the level of the mora which does not have a link to anything at the segmental tier). Since the phonology does not delete the mora, but only the segment, that mora is re-linked to the preceding vowel and the vowel surfaces as long. This is the correct analysis according to Hayes (1989); that compensatory lengthening is the result of the phonology acting to preserve the mora count. One such example of this is the Gallo- Romance language, Friulian. Repetti (1994) in a squib in Linguistic Inquiry describes a process of mora preservation through re-syllabification and compensatory lengthening. Repetti describes several processes dependant on syllabification and mora. Most importantly, as background information, there is a constraint that forces the devoicing of syllable-final obstruents. However, Repetti notes that there are several instances where this constraint is violated and only a voiced obstruent or sonorant can occur in syllable-final position. This is when the syllable structure is either CVVC or CVCαCβ (according to Repetti's (1994) labeling convention). The analysis that she gives for these syllable-final voiced obstruents surfacing is that they head a mora in a defective syllable. 32) Defective syllables in Friulian (Repetti 1994: 187)

a) CVCαCβ b) CVVC σ σ σ σ

μ μ μ μ μ μ

C V Cα Cβ C V C

Theoretical motivations for Ø | 35

The justification for the above depicted representations is that in some instances when a

CVCαCβ syllable ends in a liquid (what Repetti seems to leave up to dialectal variation) an underlying /CVCαCβ/ will undergo compensatory lengthening and re-syllabification to

[CVVCα]. The derivation provided for this phenomenon is as given below.

33) /CVCαCβ/ → [CVVCα] in Friulian (adapted from Repetti 1994: 189)

a) Delete Cβ σ σ

μ μ μ

C V Cα Ø

b) Re-syllabify Cα σ σ

μ μ μ

C V Ø Cα c) Spread vowel σ σ

μ μ μ

C V Cα While this analysis may seem abstract, Repetti notes that McCarthy & Prince (1990) propose a nearly identical structure of extra-metrical consonants that may or may not be linked to a mora (based on dialect) and then to a syllable in Arabic. Further, there may be constraints given in the literature for why syllable final consonants devoice, but those which solely head a mora and a syllable remain voiced. DeLacy (2006: 189) gives the constraint *μ / –VD_STOP for the reported word/syllable-final voicing effects in Somali (Blevins 2006a; b). 34) *μ / –VD_STOP Assign one violation for each mora that dominates a voiceless obstruent By itself, this constraint would predict that would could not see alternations such as [gup] / [gúba] 'hunchback' (masc./fem.) which Repetti (1994: 191) argues is syllable final devoicing, not intervocalic voicing. However, if we consider the syllable nucleus to be

Theoretical motivations for Ø | 36 more important than any other moraic position to have a more sonorous element in then we can offer a hierarchy that distinguishes the difference between nucleic mora and coda mora. 35) Hierarchy for consonant voicing patterns in Friulian

*μNUC / –VD_STOP » *VD-OBS]σ » *μCoD / –VD_STOP This hierarchy prevents voiceless obstruents from being realized in a nucleic position and forces the devoicing of voiced obstruents at the right edge of a syllable so long as they are not nucleic. While the above example shows that a language may, at some point in the derivation, be thought of as having an empty mora, compensatory lengthening is usually employed for mora preservation; that is, to make sure that a Ø mora does not simply delete when the segment to be parsed into that slot is deleted. Given the OT framework this thesis assumes, we can suggest that the above instantiation is an example of a language in which

SPREAD-R-μ dominates CRISP-EDGEσ thus allowing the spread of segmental material rightward into a moraic position for mora preservation even beyond syllable boundaries.

36) SPREAD-R-μ Spread rightward any segmental material into all available moraic positions

37) CRISPEDGEσ (adapted from Ito & Mester 1999) A syllable has crisp edges and does not allow the spread of phonological features, or the sharing of prosodic structures between them

If, however, a language re-ordered these two particular constraints (and given that MAXμ was significantly high ranking), there is the potential of a language allowing a Ø mora as the output of the phonological component. In this case, like the compensatory lengthening phenomenon explored above, the analyst would also want to be presented with gapping phenomena to back up the existence of a Ø mora in the output such as the inability to blend or, elide consonants at word boundaries. A similar situation to this in Italian will be explored briefly below as a different type of Ø mora from what we were discussing above. Like the Ø segments discussed in the previous section, Ø mora do not necessarily need to be derived. In strictly theoretical terms, they could also be a stored representation. One such example is the case of raddoppiamento sintattico (RS) in Italian (Nagy 1995 and references therein, Absalom et al. 2002 and references therein, Nespor & Vogel 2007). This phenomenon is described in Nagy (1995) as the gemination of a word-initial consonant in

Theoretical motivations for Ø | 37 order to satisfy a prosodic well-formedness constraint on foot structure and stress alignment. What Nagy observes is that strive to pronounce stress on the final syllable if that syllable creates a well-formed foot to satisfy constraints on binary structure.

38) Σ-BIN (Kager 1999: 156) Feet are binary under moraic or syllabic analysis Condensing the arguments of these papers who consider an empty mora in Italian (Nagy 1995, Absalom et al. 2002); the argument is that final stressed syllables must conform to prosodic well-formedness constraints, namely that feet be binary. Final short vowels in open syllables do not satisfy this requirement, but are never-the-less occasionally stressed. In these cases, the structure is argued to be as schematized below. 39) An empty mora in Italian σ

μ μ

C V Ø The structure above denotes the possibility of an empty mora being stored in the mental representation of a speaker. Evidence for this possibility comes from RS structures which satisfy one of the requirements outlined in the previous section for positing a Ø – it should be variably filled. In instances of RS, a word-initial singleton consonant is geminated. This is only permitted in certain (lexically specific) combinations of two words where the consonant to be geminated is the initial segment of the second word, and the first word is vowel-final. Under Nagy's (1995) and Absalom et al.'s (2002) analyses, this is leftward spreading of a consonant into a vacant position left by the Ø mora in the final position of the first word. 40) Filling the empty mora in Italian (adapted from Absalom et al. 2002: 4) σ σ σ

μ μ μ μ μ μ

cantò Ø → cantò b e ne

[kanˈtɔ] [kanˈtɔbˈbɛːne]

Theoretical motivations for Ø | 38

One question that arises from this proposal is why is it the case that a consonant can spread leftward across a prosodic word boundary rather than having a vowel spread right (lengthen) to produce a long vowel not crossing any prosodic boundaries at all. Clearly, it cannot be the case that a DEP-LINK constraint is prohibiting the epenthesis of a linkage between the empty mora and a segment because that is exactly what is happening in the proposed analysis for gemination. It is possible that a hierarchy where SPREAD-L-μ dominates an ANCHOR I-O constraint and SPREAD-R-μ preventing any rightward movement but allowing leftward spreading even across prosodic boundaries.

41) ANCHOR I-O Assign one violation for each position an element occupies additional to that specified in the input (No spreading)

Given the contrast with the closely related language discussed above, Friulian, where vowels due undergo compensatory lengthening through rightward spread and coda consonants are analyzed as re-syllabifying rightward as well, this may be an unexpected hierarchy. However, it is at least a possibility and could be given supporting evidence by a moraic analysis of vowels in Italian more generally. To confirm the Ø analysis, the analyst would want to find evidence to suggest that vowels in Italian are underlyingly specified with one or two mora, and those specified with one never doubly attach to a second. The above discussion has looked at two languages that were proposed as having Ø mora. The first example was a derivationally derived Ø mora in Friulian which is filled before the output to the phonetic component, and the second was RS in Italian where a Ø mora is hypothesized to be stored in the mental representation of the speaker. While it is not immediately clear that a Ø mora may survive to the output of phonology based on these examples, it does provide clues as to the evidence that would be needed to posit a Ø mora either in the mental representation, or in the output from phonology in other cases we will investigate later. Like Ø segments, the look at the above examples have shown that evidence of a Ø mora will come from gapping phenomena and should variably be filled at least in some instances to show the existence of an empty timing slot. Immediately above the level of the mora, is the level of the syllable, and I now turn to a discussion of the theoretical plausibility of a Ø element existing at that level of the prosodic hierarchy.

Theoretical motivations for Ø | 39

2.3. Ø syllables As Hartmann et al. (2008) note, "[e]mptiness has always played an important role in the literature in phonology, most prominently in the literature on syllable structure." This quote may be somewhat misleading. It is true that Øs that are posited to exist in a given language's phonology usually make reference to syllable structure, but that is not the same as Ø syllables themselves being posited. In G&C phonology, for example, there are restrictions put on consonant clusters so that if a cluster should arise, they should be syllabified separately. One example that gives weight to this hypothesis is the Irish word for blue, gorm. According to native speakers, this word is a single syllable. However, it is regularly pronounced as two, inserting a schwa between the coda consonants. This shows the alternation /gɔrm/ becoming [gɔr.əm].16 The analysis for this alternation being that the /ɔ/ vowel can maximally govern one coda consonant, therefore the /m/ must be governed by a Ø nucleus which is, in this case, shown by the epenthesis of schwa. This is not a common pattern in Irish, but Gussmann (2002) does discuss other evidence for this type of analysis in Irish which I will not reproduce here. Perhaps a more pertinent example comes from a posited Ø onset in Standard Chinese which makes reference to syllabic constraints. Duanmu (2000: 81) shows minimal pairs in Standard Chinese, one of which relies on a Ø. 42) Minimal pairs in Standard Chinese (Duanmu 2000: 81) a) [kwaa] 'melon' a') [waa] 'frog' b) [khwai] 'fast' b') [Øai] 'love' Under Duanmu's analysis (1990 qtd. in Duanmu 2000: 80-1) Standard Chinese has only two syllable structures: CV and CVX where [Ø] represents a zero onset and where a long (doubled) vowel represents a vowel linked to two mora. I will not go into detail as to the evidence that Duanmu offers for this analysis but it largely relies on evidence from free variation, gemination, and on-glide syllabification. What is interesting is how, in terms of OT, such an output could be predicted. A Ø onset is easily predicted by two OT constraints:

ONSET, and DEP I-O.

16 See Ní Chiosáin 1991 for a discussion on how the preferred syllable in Irish is bi-moraic which is the reason I give this syllabification rather than selecting the [r] as an onset to the second syllable. However also see Gussmann (2002) for a different view of this where intervocalic consonants are syllabified as onsets.

Theoretical motivations for Ø | 40

43) ONSET (Kager 1999: 93)

*[σ V (syllables must have onsets)

44) DEP I-O (Kager 1999: 68) Output segments must have input correspondents (No epenthesis) As we understand from the degrees of harmony expressed in Wolf & McCarthy (2010), a mapping to Ø is more harmonious than outright deletion. This should also work in the opposite way; the epenthesis of a Ø consonant is more harmonious than epenthesizing all the features to make up a pronounceable consonant. Therefore, if the hierarchy exists in

Standard Chinese where ONSET is ranked above DEP I-O, we would expect to see syllables with no onset consonant in the mental representation surface with a Ø onset rather than epenthesize a full consonant into that position. Given the rest of Duanmu's discussion, this may not be the correct analysis as the Ø onset is variably filled in much the same fashion as the proposed RS Ø mora is filled in Italian (as discussed in the previous section). However, this may be the result of other constraints. The interesting point that this brings up is the possibility of a stored syllable rather than just a stored morpheme or lexeme. If this is true, is it possible to store a Ø syllable? The answer to the above question is not clear. There is evidence from many languages that syllables are stored units where onset versus coda effects are clear and intervocalic consonants of various words can pattern with either; they do so regularly and their syllabic position is not predictable based on constraints. But how would someone recognize a Ø syllable in order to store it? The answer would default to stress application in a binary system. Like the chicken or egg question addressed with compensatory lengthening, the same can be asked with respect to Ø syllables; does the application of stress in a given language allow a learner to store a Ø syllable (if such a thing exists), or would a Ø syllable be the product of a repair strategy for maintaining foot binarity? Insomuch as Øs can be learned, in principle, positive evidence like stress application with the use of universally available constraints on stress should allow these units to be learned and stored. Evidence for this hypothesis, however, is lacking and Duanmu (2000) takes the opposite stance; that Ø syllables are computed based on well-formedness constraints for foot structure. Under this analysis, if a given unit (syllable in this instance) is unable to

Theoretical motivations for Ø | 41 form a foot due to binarity constraints, a Ø syllable will be added to form a foot. Foreshadowing the next section somewhat, at this point it is necessary to discuss what foot structure is. Duanmu (2000: 126) gives an excellent description of what a foot is and what it does. He states that "[s]ince stress is part of a foot, the presence of stress implies the presence of a foot and vice versa. It is inadequate to discuss one without the other. [...] Since rhythm is the alternation between strong and weak, a typical foot contains two beats." It is because of this definition of a foot that Duanmu proposes the notion of a Ø syllable though he is far from the first to do so. One particular example of the idea that a syllable need not always be audible comes from Abercrombie (1967: 35-6 qtd. in Duanmu 2000: 128). Abercrombie notes that "the English expression thank you is often pronounced as ['kju', where the first syllable than(k) is acoustically silent. But there is still musculature action during the silent syllable which resembles the production of a stressed syllable." This notion is incredibly similar to the silent suffixes in Blackfoot discussed in section 1.4 which were analyzed as being phonetically reduced. The notion there, and it applies here as well, is that because the articulatory gestures remain, it is evidence that the phonology is still outputting all of the features, and thus the syllable is not actually filled by Ø. The next type of evidence that Duanmu uses for positing a Ø syllable comes from poetry. He proposes the notion that a stressed mono-syllabic word in verse could be followed by what he calls an empty beat. 45) Mono-syllabic words followed by a Ø beat in English verse (Duanmu 2000: 128) (Ding Ø) (dong Ø) (bell Ø) (kit-tys) (in the) (well Ø) This evidence can be taken as support for the notion that phonological elements are important for poetry and verse, and that they have some psychological reality in the mind of speakers. However, it is not clear that this evidence is particularly strong as an argument for phonological parsing. Ohala (1986: 8) cautions against the use of poetry as phonological evidence pointing to the example that again [əˈgɛn] does not rhyme with plane [pʰlejn] except in verse like in the verse quoted below. 46) Poetry skewing rhyme (Ohala 1986: 8) I'm leaving on a jet plane ([pʰlejn]) don't know when I'll be back again ([əˈgejn])

Theoretical motivations for Ø | 42

Ohala (1986: 8) goes on to caution the analyst that: Poetic conventions regarding permissible rhymes, metrics, etc., are usually quite conservative and once established tend to persist after the language has undergone sound change. They may tell us about the history of the language but not about how the language itself is represented in any speaker's brain.

Though Ohala does consider poetry to be (potentially) better evidence than things like sound change or surface structures (if used correctly given his metrics) it is still rated far below even speech errors, language games, and of course experiments. I would reflect this caution, simply because a child reproduces a rhyme like that given in 45 does not mean that Ø syllables have been phonologized in any way. However, in addition to verse, Duanmu gives six other arguments for his analysis of Standard Chinese with a Ø syllable. However, all of these pieces of evidence are based on one simple assumption. Duanmu takes a very strong stance on the Weight to Stress Principle (WSP). 47) Weight to stress principle (WSP) (Duanmu 2000: 128-9) If a syllable is heavy, then it is stressed (a syllable is heavy if and only if it is stressed)

This stance forces Duanmu to assume a Ø syllable that is always word-final in order to allow light syllables the ability to form a well-formed foot and have stress. While Duanmu does concede that there is a difference between feet that satisfy binarity through having two syllables and those which satisfy it by having two mora (both of which conform to well formed foot structure in his analysis of Standard Chinese), he does not even entertain the idea of a degenerate foot. This basic assumption is unfortunate for Duanmu, and in fact forces some of his evidence to be contradictory, though he does give arguments for maintaining his analysis in spite of this. The most egregious of which is that fact that on the same page (2000: 140), Duanmu makes the claim that there is a strong requirement for words to be disyllabic in Standard Chinese and in the immediately following paragraph states that many words in Standard Chinese have both disyllabic and monosyllabic forms with no syntactic or semantic difference. He does state the difference in the choice between syllable count forms is due to stress (and thus to weight) but this analysis is forced because of his interpretation of the WSP. Unfortunately, this is unsatisfactory evidence to posit a Ø syllable.

Theoretical motivations for Ø | 43

There is, however, a possible constraint hierarchy which could create the necessity of a Ø syllable based on some of the discussion in Duanmu (2000). Supposing a syllable is a phonological unit which could be epenthesized, the following tableau gives a possible hierarchy that could lead to a Ø. 48) Tableau for a Ø syllable output

/σσσ/ Σ-BIN PARSE-σ DEP-μ DEP-σ a)  (σσ) (σØ) *! b) (σσ) σ *! c) (σσ) (σ) *! d) (σσ) (σμμ) *!

The above tableau is a theoretical possibility. This tableau assumes that all syllables except the one specifically listed as being bi-moraic in 48d are light syllables. It suggests that 48c would be ruled out for allowing a non-binary foot. Similarly, 48b is ruled out for not parsing the final syllable. The candidate in 48d is excluded for epenthesizing a mora (either a coda, or by lengthening a vowel). This results in the optimal candidate being the one given in 48a that allows an empty syllable to be epenthesized in order to both parse the overt syllable as well as maintain binarity. Having said that, this hierarchy is suspect at best. There is a simplification of theory that suggests that a syllable minimally contain a peak, or nucleic element. Directly referencing the sonority sequencing principle (SSP) Goldsmith (2011: 177) citing Selkirk (1984 and vicariously citing Hooper (1976), and Kiparsky (1979; 1981)) states that "[i]n any syllable, there is a segment constituting a sonority peak that is preceded and/or followed by a sequence of segments with progressively decreasing sonority values." However, just as the SSP is a violable 17 Nuc constraint, so too is the constraint that forces a syllabic peak, FILL . Nuc 49) FILL (Prince & Smolensky 2004: 106) Nucleus positions must be filled with underlying segments In fact, Prince & Smolensky (2004: 140-1) give examples from Lardil which show Nuc violations of FILL for adherence to matching and alignment constraints. This results in the optimal output of one such form having an empty syllable [mar . ]. Further to this point, Goldsmith (2009: 16) discusses the "consequences of encouraging syllable

17 See Clements (1990) for several discussions on languages which violate this constraint.

Theoretical motivations for Ø | 44 representations with empty nuclei" as spawning a theory of government relations inside syllables. As discussed in the onset of this section, G&C phonology frequently allows empty nuclei. But also as discussed in the onset of this section, an empty nucleus and an empty syllable are not the same thing. Prince & Smolensky (2004: 108) take a strong stance on what a syllable is, and I adopt it here. "[T]he syllable node σ must have a daughter Nuc and may have as leftmost and rightmost daughters respectively the nodes Ons and Cod. The nodes Ons, Nuc, and Cod, in turn, may each dominate [Cs] and [Vs], or they may be empty." This gives rise to the constraint NUC which is a constraint that is "fixed in superordinate position" (Prince & Smolensky 2004: 109).

50) NUC (Prince & Smolensky 2004: 108) Syllables must have nuclei While this argument so far is strictly theory-centric, it denotes the fact that even a so-called empty syllable still has internal structure, that is, syllables (even empty ones) minimally contain a nucleus or nucleic element, even if that nucleus itself is empty. Thus, a so-called empty syllable cannot be classified as an empty set, a prosodic bracketing structure devoid of any internal components. Moving beyond the realm of theory-centric argumentation, the onus is on me to show that this fact has consequences in natural language. As previously stated, poetry and verse is anecdotal evidence at best. However, as Ohala (1986: 4) states, evidence is that which helps us to choose between competing hypotheses and unfortunately, I have none better at this point. Earlier, I gave the example of an English children's rhyme taken from Duanmu (2000) and I repeat it below as 51. 51) Mono-syllabic words followed by a Ø beat in English verse (Duanmu 2000: 128) (Ding Ø) (dong Ø) (bell Ø) (kit-tys) (in the) (well Ø) Duanmu used this as evidence from poetic meter to show that Ø syllables were psychologically real elements since they were required to keep the proper poetic structure. I instead interpret Duanmu's description of these lines as evidence that what he posits as empty syllables in fact actually carry a nucleus and therefore a mora. Duanmu (2000: 128) states: In [51] each line has three disyllabic feet. On the first line each foot consists of a monosyllable and an empty beat [...] The empty beat need not be realized as

Theoretical motivations for Ø | 45

silence, but can be filled by lengthening the preceding monosyllable. Thus, the first two feet on the first line can be read as a prolonged ding and a prolonged dong, without pauses.

If any empty element can cause a lengthening effect on a neighboring element, than that empty element is best analyzed as an empty timing slot, or mora. This analysis not only makes sense given the data that Duanmu presents, but also follows from the proposed universally high ranking constraint NUC as given in Prince & Smolensky (2004: 108-9). In sum, I conclude that while we may posit or pursue Ø segments that make specific reference to syllabic positions, a Ø syllable itself is theoretically unmotivated and therefore should not be pursued as a course of analysis. As foreshadowed in this section, the next level of the prosodic hierarchy is the stress foot on which I will now give a brief discussion of the plausibility of a Ø structure at that level.

2.4. Ø stress feet On first read, it seems that Lahiri & Plank (2010) propose a null prosodic phrase in their paper. However, on closer inspection, they use 'prosodic phrasing' as a blanket term to refer to the creation of all prosodic bracketing structures. In fact, what seemed to be a Ø phrase in their paper, I believe is actually a Ø foot. The example they give is reproduced from Steele (1775: 28) and as they note, omits much detail.18 52) A Ø foot in 18th century English (Lahiri & Plank 2010: 375) (Every) (sentence) (in our) (language), (whether) (prose or) (verse ) ( ) (has a) (rhythmus) ( ) ( pe) (culiar) (to it) (self) ( ); (that is, in the) (language of) (modern ) (sicians), ( it is) (either in) (common time) ( or) (triple time) ( vi) (delicet), (minuet time), ( or) (jigg time), ( or) (mixed ).

The authors fail to provide a gloss for their use of symbols in the above section noting only that '' is used to indicate "an empty beat" (Lahiri & Plank 2010: 373). It is my understanding from the original work (which had divided the passage into: heavy ',' light ':.,' and lightest '..') that the above symbols should mean: '' an empty stressed syllable and ' ' an empty unstressed syllable. This would mean that the brackets are used to denote foot

18 Scans of the original document (Steele 1775) have been reproduced in Appendix 6.2.

Theoretical motivations for Ø | 46 boundaries. However, given the parsing of (that is, in the) into one bracketing structure still leaves room for doubt that these brackets could in fact denote phonological phrases. For that reason I will briefly return to this discussion in section 2.6. At this point, though, I will return to the statement of what a foot is given by Duanmu (2000: 126). He states that "[s]ince stress is part of a foot, the presence of stress implies the presence of a foot and vice versa. It is inadequate to discuss one without the other. [...] Since rhythm is the alternation between strong and weak, a typical foot contains two beats." Since I have already argued that a Ø syllable, or 'empty beat,' is not theoretically well-motivated and is, in fact, not empty, the same can be true of a foot. Even if it were possible to discover a foot which contained two empty beats, the foot itself would not be empty because it would contain those beats. Further to this point, I will again point out the argument made in the previous section on Ø syllables: simply because pauses are pertinent in poetry and prose does not mean that those pauses (be they empty syllables or feet) are in any way phonologized in the mind of a speaker. Similar to the conclusion of the previous section, I argue that the possibility of a Ø foot is theoretically unmotivated and an analysis based on this notion should not be pursued. I will now move to a discussion on the possibility of a Ø prosodic word and the phonological constraints that would permit or force its realization.

2.5. Ø prosodic words Earlier, in section 1, I gave an example to show that silent syntactic copies prevented phonological cliticization: Whoi do you *wanna/ want to learn English? The unanswered question at that time was, does the existence of a silent element in the morphosyntax necessitate the existence of a silent element in the phonology? As a point of departure for this section I will answer, no – but it does influence it. In this section I will briefly explore what makes a phonological word in English and whether that lends itself to the creation of a Ø word in the instance of silent syntactic copies of the type exampled above. Ultimately, I will not argue that English has Ø prosodic words, but will use the evidence against that hypothesis to predict what a Ø prosodic word would look like if it did exist in natural language.

Theoretical motivations for Ø | 47

Earlier, again in section 1.1, I gave an example sentence to demonstrate the assumptions I make of what the prosodic hierarchy consists of. In this representation, I listed the mental representation of function words like the plus a noun as forming a clitic group. Here, I modify this assumption to reflect more recent theories of the prosodic hierarchy such as those given in Selkirk (2009; 2011), Elfner (2010; 2011; 2012), and Windsor (2011) where ill-formed prosodic words are members of recursive structures in line with the assumed matching constraints that straddle the prosody-syntax interface. The revised structure is given below in 53 with an explanation to follow. 53) The prosodic hierarchy (revised for recursive word structure)

The phonological utterance υ

The intonational phrase ɩ ɩ

The prosodic phrase φ φ φ

The clitic group ω ω ω ω

The prosodic word ω ω ω ω

The stress foot Σ Σ Σ Σ Σ

The syllable σ σ σ σ σ σ σ σ σ

The mora μ μμ μ μμ μ μμ μ μ μμ μμ μ

The segment [ðə ɹoʊd tə hɛɫ ɪz peɪvd wɪθ ædvɛɹbz] "The road to hell is paved with adverbs" -Stephen King (2000: 125)

As stated in this diagram's first instantiation on page 7; parsing from one level to the next is with considerations of stress placement from a casual recitation of the above sentence as it would be constructed in my own dialect. I will not go into detail as to the reasons for leaving function words such as 'is' or 'with' as being monomoraic, however, the reason I assert that function words ending in schwa are not well formed prosodic words is because they do not have enough moraic weight to meet minimal word requirements in English.

Theoretical motivations for Ø | 48

54) Minimal word size in English Minimal ω [ɪ] [ɛ] [ʊ] a) *(C)Vμ *[hɪ] *[hɛ] *[hʊ] b) (C)VμCμ [hɪd] [hɛd] [hʊd] c) (C)Vμμ [hi] [he] [hu]

The data above shows that only when an English word has minimally two mora (a binary 19 foot) is it well formed. This points to the relevance of the constraint WORDSIZE (McCarthy & Prince 1986: 6) as being particularly important in English, disallowing the parsing of any mono-moraic syllable into a prosodic word by itself.

55) WORDSIZE(μμ) (McCarthy & Prince 1986) A prosodic word must minimally contain two mora This markedness constraint is in opposition to the matching constraint that seeks to show isomorphism between the prosody and the morphosyntax.

56) MATCHWORD (adapted from Selkirk 2011: 439) Each syntactic head (Xo) is interpreted by the phonological component as a prosodic word (ω)

This opposition also tells us the relevant constraint ranking so as to create prosodic words in English: terminal nodes in the morphosyntax will be matched to prosodic words if and only if they bear enough phonological weight to satisfy the English word size requirement. Otherwise, they will be parsed as part of a recursive prosodic word in violation of non- recursivity minimally violating EXHAUSTIVITY.

57) NONRECURSIVITY (NO-REC) (Selkirk 1995, Truckenbrodt 1999: 240) Any two prosodic structures that are not disjoint in extension are identical in extension: there can be no Ci that dominates Cj where Ci = Cj

58) EXHAUSTIVITY (Selkirk 1995) There can be no prosodic structure Ci that immediately dominates a constituent Cj where Cj < i –1 (i.e., no ω immediately dominates a σ and so on)

19 I assume the difference between what are phonetically tense and lax vowels in English is really mono- moraic versus bi-moraic (respectively). For a complete discussion on this topic, see Lass (1976: chapter 1).

Theoretical motivations for Ø | 49

The effect of these constraints arranged appropriately both explain the inability for English to except mono-moraic syllables as prosodic words, and gives us the output of the hierarchy proposed above in 53. 59) Constraints governing ω formation in English

/D ðə N ɹoʊd / WORDSIZE(μμ) MATCHWORD EXHAUSTIVITY NO-REC a) (ðə)ω (ɹoʊd)ω *! b) [ðə (ɹoʊd)ω]φ * **! c)  (ðə (ɹoʊd)ω)ω * * *

In the above tableau, the candidate 59a is immediately ruled out despite incurring fewer violations overall than the other candidates. Despite incurring the same number of overall violations, 59c is a more optimal candidate than 59b for the fact that the determiner [ðə] does get parsed into prosodic word structure before being parsed into a phonological phrase thus incurring only one violation of EXHAUSTIVITY. Returning to the initial question that was raised, relating silent copies in the syntax to Ø words in the phonology, English clearly does not allow the existence of Ø prosodic words as WORDSIZE(μμ) is seemingly un-dominated. While silent copies in the syntax no doubt are responsible for various linguistic phenomena, they do not force the realization of a Ø word in English. Therefore, Nespor & Vogel (2007) and Kim (2004) were correct when they analyzed instances such as Whoi do you *wanna/ want to learn English? as having want and to belong to separate phonological phrases rather than having a Ø of any kind responsible for the prevention of cliticization. This analysis, however, does not preclude the possibility of a Ø prosodic word. According to the strict domination hierarchy (Prince & Smolensky 2004: 3) the specific ordering of constraints is language specific. Whereas one language may show the ranking A » B, another language may have B » A thus producing phonological differences observed between languages. Therefore, if a language had MATCHWORD as an un- dominated constraint as English does with WORDSIZE(μμ), we would expect to see a Ø prosodic word. Of course, the onus of proof would be on the analyst to show that that structure truly existed and was not deleted before the phonological output due to WORDSIZE or *STRUC constraints that seek to limit the phonological bracketing to only absolutely necessary structures.

Theoretical motivations for Ø | 50

60) *STRUC (Kager 1999: 404) Assign one violation to each instance of phonological structure (No phonological structure) Like the proposed methods for evaluating Ø segments, the analyst proposing a Ø word would want to show that the structure, at least in some instances, could variably be filled with some segmental material; highlight and explain clear adjacency effects, and demonstrate that the analysis using a Ø word has more predictive power than one not relying on abstract null structures. Unlike Ø segments, I would not expect that Ø words could be found in multiple positions (i.e., word-initially and word-finally) easily since their existence depends on the morphosyntax. While multiple sentential positions could be explained, they may not actually be expected since syntactic structure, namely headedness or the hierarchy of projections, should remain relatively constant. Unlike syllables and feet, I conclude that a Ø prosodic word is a distinct theoretical possibility. Its existence, however, would be decidedly marked as such a structure would be difficult to acquire without robust evidence for the language learner (i.e., the clear adjacency effects or variable filling as discussed above). That being said, analyses making reference to a Ø word should not be discarded, but would require multiple avenues of evidence in order for them to be excepted. From Ø prosodic words, I now move to the immediately dominating level of the prosodic hierarchy and examine the possibility of a Ø phonological phrase.

2.6. Ø phonological phrases Much like Ø prosodic words, Ø phonological phrases are predicted to be possible under the present analysis. If it were the case that an entire syntactic XP were vacant of pronounceable copies, and were spelled out to the phonological component, in much the same way that MATCHWORD could create a Ø prosodic word, MATCHPHRASE could, in theory, create a Ø phonological phrase.

61) MATCHPHRASE (adapted from Selkirk 2011: 439) Each syntactic phrase (XP) is interpreted by the phonological component as a phonological phrase (φ)

Theoretical motivations for Ø | 51

This theory of the prosody-syntax interface allows me to return to the possibility of a Ø phrase in the analysis of Lahiri & Plank (2010). As stated in section 2.4, it is not clear that the empty beats in that paper are to be interpreted as null syllables and feet instead of null phonological phrases based on the data in Lahiri & Plank's (2010: 375) example 10, given above as 52 and repeated below as 62. 62) A Ø phrase in 18th century English (Lahiri & Plank 2010: 375) (Every) (sentence) (in our) (language), (whether) (prose or) (verse ) ( ) (has a) (rhythmus) ( ) ( pe) (culiar) (to it) (self) ( ); (that is, in the) (language of) (modern mu) (sicians), ( it is) (either in) (common time) ( or) (triple time) ( vi) (delicet), (minuet time), ( or) (jigg time), ( or) (mixed ).

If this bracketing is to be interpreted as phonological phrases, Lahiri & Plank discuss the type of theory they use which would allow for that analysis. They look at two schools of thought on how prosodic constituents are created; the one embraced by this thesis which they term "syntactic determinism" citing the theories of Selkirk (1984 et seq.), Nespor & Vogel (1986 [2007]), and Truckenbrodt (1999) among others, and their choice – a strictly rhythmic theory citing Steele (1775; 1779), Sweet (1876 et seq.), Sievers (1901a; b), Paul (1916), and Luick (1923). Of these two schools of thought, the former is by far the most prevalent and "has become the standard contemporary view" (Lahiri & Plank 2010: 372). While rhythmic constraints no doubt play a role in a given language's phonology (the choice of iambic versus trochaic feet, whether they are built left to right or right to left, and constraints such as SYLL-FORM and FOOT-FORM (Prince & Smolensky 2004) are constraints based on the phonological rhythm of a language) such a theory forces the assumption of Ø constituents while direct reference theories merely permit the possibility of Ø constituents. With the utmost respect of those authors who posited strictly rhythmic theories and their contributions which allowed the advancement of linguistics in their time, more recent evidence has allowed us to advance phonological theory in a different direction. As Ohala (1986: 5) says, advancement of a theory through new evidence "enables us to discard those theories" and Ohala "leave[s] it to the reader to say whether linguistics is closer to physics or theology in the way of giving out-dated theories a decent burial." Having dismissed the outdated notion of rhythmic phrasing of phonological constituents above the foot, I will consider the possibility of a Ø phrase under the current

Theoretical motivations for Ø | 52

Match Theory (Selkirk 2009; 2011, Elfner 2010; 2011; 2012, Windsor 2011). Under this theory, let us consider a syntactic construction where an argument-less verb, let us assume an imperative, raises to the tense projection leaving a VP structure below To containing no pronounceable material. I will diagram this possibility by using an imperative verb in Irish. 63) V-to-T imperatives in Irish and vacuous syntactic structure ith! 'eat!' TP

T vP

v T V(P) [ØPST] V v V v ith ith

I will leave aside motivations for assuming this structure for the moment. I return to a discussion of V-to-T movement in Irish in section 3.2.2.3. However, what the above structure does give the possibility to is the notion there is the SPELLOUT of at least one syntactic XP, if not two, which have no pronounceable material (vP and VP). Elfner (2010; 2011; 2012) shows that in Irish (specifically the Connemara Irish dialect) aligns what she refers to as an H-L tone with the rightmost word in every phonological phrase. Unfortunately, she does not analyze imperatives, however, one could reasonably predict some phonetic variation in the above structure if it were neighbored by an Ø phrase which would also require tones associated at its edges.20 I would not expect to find this evidence as I do not think a Ø phrase is well motivated in Irish. The theory, however, does not preclude its existence, and equips the researcher with the means to test for it. In the literature on prosody-syntax interface theories, certain language phenomena are often recycled as evidence for refinements of the theory, largely concerned with length, pitch, and tone at or near phrase boundaries (Selkirk 1986; 1995; 2009; 2011, Hale & Selkirk 1987, Truckenbrodt 1999, among many others). These authors, who all argue in favor of syntactic XPs matching to phonological φs in one form or another discuss several

20 Though Truckenbrodt (1999) would argue against the matching of these empty phrases based on the Lexical Category Condition which renders syntactic phrases headed by only lexical class words invisible to the phonological component at the interface. These phrases not being headed by a lexical item would make them invisible to the phonology under Truckenbrodt's analysis.

Theoretical motivations for Ø | 53 instances of the above phenomena in various languages presenting clear evidence for either left or right φ boundaries. A particularly good example of this is Kimatuumbi as discussed in Truckenbrodt (1999: section 4). Kimatuumbi is analyzed as marking right φ boundaries with a shortening effect and left φ boundaries with phrasal tone insertion. Unfortunately, according to Truckenbrodt the shortening effect in Kimatuumbi is formalized as: 64) Shortening in Kimatuumbi (Truckenbrodt 1999: 237) Long stem vowels shorten, except in the prosodic word immediately preceding the right edge of a P-phrase

This essentially means that the right edge of a φ in Kimatuumbi blocks a particular phonological alternation, not displays it. However, the second φ edge effect may be more fruitful in discovering a Ø phrase. Again according to Truckenbrodt, this second phenomenon inserts an H tone between two XPs and therefore between two φs that tone surfaces on the last vowel of the syntactic phrase preceding it. In an alignment based analysis (such as that presented in Hale & Selkirk 1987) this is formalized as: 65) Phrasal tone insertion in Kimatuumbi (constraint) (Truckenbrodt 1999: 238)

ALIGN (P, L; H, R) Align the left edge of each phonological phrase with the right edge of a[n] H tone Truckenbrodt also offers a schematization of exactly what this rule looks like: 66) Phrasal tone insertion in Kimatuumbi (schematization) (Truckenbrodt 1999: 238)

P P P

(H) H H What this schematization predicts is that should a syntactic construction in Kimatuumbi be tested wherein a syntactic XP devoid of pronounceable material be linearized to the right of a syntactic XP which did contain pronounceable material, the left-most of the two phrases would show H tone insertion.21 Other tests could be used in languages that show final lengthening or other tonal phenomena. If it was found that such syntactic constructions, when pronounced, displayed a higher degree than normal of final lengthening, or if upstep (presumably from two H

21 Again, Truckenbrodt (1999) would argue against this possibility based on his Lexical Category Condition, but it is a predicted possibility under MT.

Theoretical motivations for Ø | 54 tones sequentially) were exhibited where no pronounced subsequent material was observed, these would also be evidence for a Ø phonological phrase. To the best of my knowledge, however, neither such phenomena, nor such analyses have yet been documented or presented, although such phenomena are possible under the current hypothesis. The lack of these phenomena being attested thus far would be due to constraints such as *STRUC deleting that structure even though MATCHPHRASE would seek to create it. In sum of this short section, like prosodic words, the possibility for Ø structures at the level of the phonological phrase exists. Tests for this sort of structure would, again, be clear adjacency effects and the analyst demonstrating that reliance on a Ø phrase would allow for better predictive power than any other analysis. Once more, I leave variable positions within structure out of the possibility for evidence at this level as the syntactic structure that feeds the phonology should be relatively consistent. However, one might expect to be able to create an alternate sentence structure that would allow this position to be pronounced (i.e., adding additional structure that could block raising out of this phrase in the syntax and maintain the same effect on adjacent phrases). As stated, I am not aware of any description or analysis of a language which shows this evidence, while it is possible under the present theory, I would in fact be surprised to learn of such facts. With that said, I now turn to investigating the possibility of Ø structures at the level of intonational phrases.

2.7. Ø intonational phrases The highest level of prosodic matching that MT is concerned with is the intonational phrase. Selkirk (2009; 2011) contends that just as a syntactic head should be realized as a prosodic word and a syntactic XP should be realized as a phonological phrase, the maximal syntactic projection CP also has a correspondence with prosodic constituency, namely the intonational phrase.

67) MATCHCLAUSE (adapted from Selkirk 2011: 439) Each syntactic clause (CP) is interpreted by the phonological component as an intonational phrase (ɩ)

Much like Ø phonological phrases, I am unaware of any analysis to date which makes use of a Ø intonational phrase. The existence of such a structure is, however, not prohibited

Theoretical motivations for Ø | 55 under the current theory. Before proceeding with what evidence for a Ø ɩ would look like, it is first important to understand what evidence for the existence of this structure is. Theory-centric evidence for the existence of ɩ comes from headedness. Since the concept of ɩ is that one or more φs parse into that higher structure, a given φ should be the head of ɩ. Nespor & Vogel (2007: xxiv) state that main ɩ prominence universally falls on the rightmost φ (at least in broad-focus sentences). This means that intonational stress is aligned with the rightmost φ in an ɩ and for this to be the case, the prosodic constituent ɩ must exist, otherwise languages that align phrasal stress with the left would run counter to what Nespor & Vogel (2007) claim is a universal. More concrete evidence for the existence of intonational phrases comes from a phonetic correlate with intonational boundaries known as 'reset' or 'pitch reset' (see for example Lin & Fon 2011). These sort of resets in F0 can be seen at the beginning of a new intonational phrase when the steadily downward drifting intonational (F0) contour is reset back up to a new upper level before continuing to drift back down again towards the end of an utterance. 68) Pitch reset in intonational phrases

The above pitch trace is a recording made of a sentence from Nespor & Vogel (2007: 188)

'Lions [ɩ as you know]ɩ are dangerous.' In line with the expected intonational phrase contours, pitch reset can be observed in the phonetic component as well. These intonational phrases, unsurprisingly, match up with syntactic CPs, whether they are a CP subject, or an

Theoretical motivations for Ø | 56 embedded clause (a parenthetical clause) as above. Now that I've established what an intonational phrase is and how is it recognized, is there a potential construction that could lead to this structure being filled by Ø, and what would the evidence of that Ø look like? Only one structure comes to mind as a potential place to look for a Ø intonational phrase. While most often it is the case that the subject of a clause be a DP, many languages also allow CP subjects such as English. 69) CP subjects in English CP CP C' C TP That the Ø answer is obvious surprised Hermes Building on this fact; in very limited circumstances, it is possible to omit a subject in English – in child acquisition, and in 'diary speech' (Haegeman & Ihsane 2001). 70) Subject drop in adult English 'diary speech' (Haegeman & Ihsane 2001: 330) a) Cried yesterday morning: as if it were an hour for keening: why is crying so pleasurable. b) Thought I heard something. c) Guess I should have been more careful.

In the above examples, all three sentences assume a null subject (or null topic) where the pronoun 'I' would be understood to be present, but not pronounced. In this position, a pronoun is not the only thing that can be dropped, but also articles. 71) Article drop in adult English 'diary speech' (Haegeman & Ihsane 2001: 331) a) Damn dogs are taking over the city b) *Damn dogs are taking over city (cp. *Those damn dogs are taking over city) As can be seen in the above pair, while initial articles can be dropped in 'diary speech,' medial articles cannot be dropped so readily. Looking at this data alone, one might draw the conclusion that English, among other languages, is able to drop a Do (this analysis assumes that pronouns such as 'I' are of the category D, not N). However, other data shows that this is not the whole truth. In fact, subject plus modal/copula deletion is possible under certain circumstances.

Theoretical motivations for Ø | 57

72) Subject + modal/copula deletion (adapted from Haegeman & Ihsane 2001: 331) a) (I) had to stop, (I was) wet to (the) skin b) (I'm) afraid there's not much we can do c) *I (am) afraid there's not much we can do The data in 72 highlight the fact that not only a subject may be omitted in certain English constructions, but also a modal or copular verb. In the first example, there are many syntactic elements that do not get pronounced – including subjects and one of the verbs. Examples 72b and 72c crucially show that a clitic subject plus copular verb can be omitted, but when this occurs, the subject may not survive eliding only the copular verb. This data shows that certain syntactic elements may be subject to deletion before SPELLOUT, but what about an entire CP subject? In 'diary speech' style, while context may be needed, the sentences are at least possible. 73) CP subject elision a) (That I was scared) didn't bother me none. b) (That the answer was obvious) surprised me! c) ?(That the answer was obvious) surprised Hermes. With an implied, or context provided subject, if the object is first person, it seems to be possible to drop an entire CP before pronunciation. Example 73c does not seem to be entirely ungrammatical with a third person object, but it is certainly worse than the others. In the immediately above argument, I have chosen my wording carefully. I note that the subject CP may be deleted before SPELLOUT. The reason I make this distinction is because there is no reason to assume that the deletion of these elements is due to the phonological component. To defend this statement, I argue that an L1 learner would be provided no evidence of a Ø intonational phrase that survives into the phonological component. I make this claim based on the fact that I can find no evidence of a phonological phenomenon that takes place at an ι boundary, but rather only that an ι boundary is the maximal application domain of certain phenomena. Nespor & Vogel (2007: 205-16) discuss three phenomena which apply at the level of the intonational phrase: several featurally-conditioned sandhi rules in Gorgia Toscana (Tuscan Italian), nasal spreading in Spanish, and /s/-voicing in Greek. In all three of the phenomenon they note, the rule can be summarized as follows.

Theoretical motivations for Ø | 58

74) Generalization of ι-boundary phenomena

α → β / [ι__]ι [γ] In other words, some radical /α/ becomes [β] in the environment of [γ] provided that that environment does not cross an ι boundary. This fact gives evidence for the universal high ranking of the faithfulness constraint CRISP-EDGEι.

75) CRISPEDGEι (adapted from Ito & Mester 1999: 30) The prosodic category ι has crisp edges (ι-boundary edges do not allow the spread of phonological features, or the sharing of prosodic structures across them) Since evidence of the prosodic constituent ι comes from the blocking of phenomena spreading, not from adjacency effects, no evidence would ever be given to the L1 learner of a Ø ι surviving to the output of the phonological component. I then argue for a universal ranking of constraints concerning ι-boundaries as given below in 76 which would prohibit the surfacing of a Ø ι cross-linguistically. 76) Universal ranking of ι constraints ι FILL » MATCHCLAUSE | CRISPEDGEι This ranking both allows matching constraints and crisp edge constraints to align phonological ιs to syntactic CPs and not allow phonological phenomena to cross those boundaries, but also allows the principle of stray erasure to delete any redundant, or empty, ιs so that they are never output from the phonological component. The above statement can be backed up by asking a question that recently appeared on phonetician, John Wells' blog site: "How do we accent the absence of something? A zero morpheme, if you will?" (Wells 2012). Looking specifically at contrastive focus, we can see the effects in two adjacent ιs: It wasn't blue, | it was green. (Wells 2012). If we accept that ellipsis could remove all but the direct object of the second ι (it wasn't blue, but, green) how would we accent the contrastive focus if that element were a Ø? Wells was intrigued by this question when he noted the shifted contrastive focus of a Ø morpheme in a House of Commons debate wherein he paraphrases the leader of the opposition as saying "Instead of harping on about unemployment, why doesn't he congratulate the Government on the increase in employment?" wherein he noted that the contrastive focus of 'un- employment' was 'Ø-employment.' Of particular interest is the fact that although the stress

Theoretical motivations for Ø | 59 was theoretically shifted from a Ø to the first syllable of the contrastive word, the initial vowel carrying that stress was not strengthened, but remained as what Wells called weak [ɪ]. This is a recurrent theme in this thesis: Ø cannot carry stress, nor other phonological features more generally, therefore they must be shifted to the closest available segment (see section 3.2.1.4 for a complete discussion). Given that Nespor & Vogel (2007: xxiv) argue that the head of ι is universally right-linearized, even such instances of contrastive focus would be forced to shift stress to an already stressed right ι edge thus obscuring the fact that a Ø had been there in the first place as schematized below. 77) Contrastive focus on a Ø ι

a) [ ]ι [ Ø ]ι

Universal Contrastive focus prominence on Ø element

b) [ ]ι [ Ø ]ι

Focus shift to previously stressed element

c) [ ]ι

Enhanced deletion of stress redundant structure This at least provides a measurable prediction if such an example could be found: is the focused right boundary of the ι preceding the hypothesized Ø ι somehow phonetically stressed more than would otherwise be expected? I remain skeptical that such a Ø could be learned and phonologized – however, at least this hypothesis does provide testable conditions for it.

Theoretical motivations for Ø | 60

2.8. Interim conclusion 1 In this section I have explored the theoretical possibilities for Ø structures throughout the prosodic hierarchy from the level of the segment to the intonational phrase. In some instances I have argued that there are theoretical motivations for why one would want to argue for a Ø structure, and for others I have argued that Ø is not a theoretically well motivated possibility. In the discussion that preceded, I found that both the segmental and moraic tiers could be viably analyzed as containing Øs. However, at the syllabic and foot level, while a nucleus position could be empty, a full syllable or foot would minimally have some internal structure and therefore could never actually be analyzed as a Ø structure or empty set. Above the level of the foot, due to matching constraints, I argued for the possibility of Ø prosodic words and phrases, though I noted I would be surprised to learn of an analysis that proposed a Ø phonological phrase. Building on this, at the level of the intonational phrase, despite a matching constraint which could in theory produce a Ø ι, I argued that an L1 learner would never be presented with evidence to maintain this structure in the output of the phonological component. The well-versed reader may note that some researchers (Nespor & Vogel 2007 for example) have given evidence for a phonological constituent, the utterance (υ) which is above the intonational phrase. I do not take issue with this constituent, nor deny its existence. However, having reached the maximal level to which I could match prosodic and syntactic constituents where two or more members of a given level could be side by side, I would suggest that the only Ø utterance would be a non- linguistic look to give reality to the phrase 'if looks could kill.' Given that this is extra- linguistic, this too is not a spot to look for Ø and therefore left unexamined. One important by-product of this chapter has been to identify what evidence for Ø at a given level of prosodic structure would look like, and how one would go about arguing for a Ø structure. With some degree of flexibility between the various constituents, the four pieces of evidence for a given structure were:

Theoretical motivations for Ø | 61

78) Criteria for prosodic Ø a) Positional variability (the empty element should not be restricted to morpheme/lexeme-initial or final position solely) b) Variably filled (in certain constructions, a given empty element should contain some pronounceable material) c) Adjacency effects (an empty element should clearly be recognizable though its effect on neighboring constituents) d) Predictive power (an analysis assuming an empty element must be shown to have more predictability of distribution than other competing analyses not reliant on Ø)

Though it is not a requirement that all analyses of Ø show all four of these criteria being met, the more compliant with this list, the more readily believable that analysis will be. Now that Ø as a theoretical element is better understood, I will walk through three case studies where Ø is shown to be a viable segment (in Seri), a proposed but unnecessary mora (in Russian) and both a segment and word (in Irish). I begin with a discussion on the Russian Jer vowel with a sidetrack into Seri before leading to an in depth discussion of consonantal lenition in Irish.

Case studies of Ø | 62

3. Case studies of Ø In this chapter, I build on the theoretical motivations of Ø discussed in the previous chapter. I begin by discussing Jer vowels in Russian which show an alternation with Ø in certain grammatical constructions. This discussion, as directed by Gouskova (2012) will naturally proceed to a look at alternations with Ø in Seri which will then be used to re-evaluate the analysis of a Ø mora in Russian. Following the section on Russian, I will proceed to give an in-depth analysis of Irish consonantal lenition arguing in favor of both a segmental alternation with Ø as well as a Ø prosodic word. Evidence for the proposed Øs in this chapter comes from contrast, cross-linguistic comparisons, historical developments, and phonological recoverability.

3.1. 'Ghost' vowels in Russian (Jer vowels) Like the French segmental alternations with [Ø] discussed in section 2.1, Russian (as with most Slavic languages) has been claimed to have vowel ~ Ø alternations in certain grammatical constructions. Also like French, the environment for these alternations is not strictly linked to a morphosyntactic configuration, but is rather dependent on the phonological environment which can change when different morphology is added to a root. This can be observed in examples such as those below in 79 where the addition of a genitive suffix in 79b causes the deletion (or alternation with Ø) of the vowel that surfaced in the 79a example. 79) Vowel ~ Ø alternations in Russian (Gouskova 2012: 80) a) [ljón] b) [ljØn-á] linen.NOM linen.GEG.sg 'linen' '(of) linen' This is in stark contrast to other forms such as those given in 80 where the same basic environment is attested, but no vowel ~ Ø alternation is observed. 80) Under-application of vowel ~ Ø alternations in Russian (Gouskova 2012: 80) a) [ljénj] b) [ljénj-i] laziness.NOM laziness.GEN.sg 'laziness' '(of) laziness' This process is only partially predictable and is not simply reliant on which vowel is used as can be seen by the following lists.

Case studies of Ø | 63

81) Russian vowel ~ Ø alternations with /o/, /ə/, and /e/ (Yearley 1995: 533) a) [kusok] 'place' (NOM.sg) a') [kusk-a] 'place' (GEN.sg) b) [denj] 'day' (NOM.sg) b') [dnj-a] 'day' (GEN.sg) c) [ugərj] 'eel' (NOM.sg) (Gouskova 2012: 80) c') [ugrj-u] 'eel' (DAT.sg)

d) [polon] 'full' (short form of adjective) d') [poln-ɨj] 'full' (long form of adjective) e) [beden] 'poor' (short form of adjective) e') [bedn-ɨj] 'poor' (long form of adjective) f) [mestj] 'vengeance' (noun) f') [mst-itj] 'avenge' (verb)

82) Russian /o/, /ə/, and /e/ that resist alternation with Ø (Yearley 1995: 533-4) a) [mest] 'place' (GEG.sg) a') [mest-o] 'place' (NOM.sg) b) [gravjor] 'engraver' (NOM.sg) (Gouskova 2012: 80) b') [gravjor-ə] 'engraver' (GEN.sg) c) [igərj] 'Igor, a name' (NOM.sg) (Gouskova 2012: 80) c') [igərj-u] 'Igor, a name' (DAT.sg) d) [poxoʒ] 'resembling' (short form) d') [poxoʒ-ɨj] 'resembling' (long form) e) [bel] 'white' (short form) e') [bel-ɨj] 'white' (long form) f) [gore-l] 'burned' (masc. past tense) f') [gore-la] 'burned' (fem. past tense)

Although only mid-vowels can alternate with Ø,22 as shown in 80 and 81, they do not delete in all words23 (Gouskova & Becker to appear). The examples in 81f and 82a show

22 There are examples of [a] alternating with Ø as well: [rat-ok] ~ [rt-a] 'mouth' (dim.NOM.sg / GEN.sg). Gouskova (2012: 84) argues that in these examples, the [a] vowel is underlyingly an /o/. 23 In the discussion that follows, I will adopt the terminology of Gouskova (2012) and refer to words containing a Jer vowel in the root as exceptional words. In Gouskova's terms, this label reflects the notion that these lexemes are lexically marked to show vowel deletion in their paradigm. This term can also be used to refer to words containing a Jer vowel in the analysis proposed by Yearley (1995). However, in Yearley's analysis, words containing a Jer vowel would be exceptional in that their underlying representation would contain a Ø – what we have labeled as a cognitive placeholder for the purposes of this discussion.

Case studies of Ø | 64 this remarkably well since their only root phonological difference is the palatalization of the final obstruent. The lack of predictability as to which forms will have a vowel ~ Ø alternation and which forms will maintain a root vowel even with the addition of a vowel suffix has led to a controversy in the literature as to whether there is a stored Ø vowel in the mental representation of some words, as claimed by Yearley (1995) for Russian and Marlett & Stemberger (1983) and Marlett (1997) for Seri, or if these words are lexically marked as exceptional and may vowel-drop as suggested by Gouskova (2012) and Gouskova & Becker (to appear).

3.1.1. Competing models of representation for Jer vowels The two competing analyses of the Russian Jer vowels make very different predictions of the phonological representation of the words that contain them. If Gouskova (2012) and Gouskova & Becker (to appear) are correct, the vowel ~ Ø alternations are actually the deletion of a mid-vowel in lexically marked words. The representation for such a word contains a traditional structure where segments are parsed into their respective syllable structures and then into a prosodic word24 as schematized below. 83) Representation of an exceptional word under Jer deletion ω

σ

On N

μ

[lj n - a] If the above structure were taken to be the correct representation of Russian Jer vowels, where the /o/ vowel deletes without leaving behind any trace, several markedness constraints would be violated. Some of those constraints would be *COMPLEX, MAX I-O, and the SSP.

24 I am omitting the level of the foot for this discussion as it does not seem to be pertinent to the phenomenon.

Case studies of Ø | 65

84) *COMPLEX (Kager 1999: 288) No complex syllable margins (*[σ CC (onsets are simple) *CC]σ (codas are simple)) 85) The sonority sequencing profile (SSP)25 (Gussenhoven & Jacobs 2005: 138) The sonority of a syllable increases from the beginning of the syllable onwards, and decreases from the beginning of the peak onwards

I assume, following Foley (1972 qtd. in Clements 1990: 286), that the sonority scale is that given in 86 which crucially shows (for the form schematized in 83) that nasals are less sonorous than liquids. 86) The sonority scale (least to most sonorous)26 Oral stops » Fricatives » Nasals » Liquids » Glides » Vowels The violation of these two constraints by itself is not troubling since many languages violate these markedness constraints. For example, English frequently has words that have three and four branches for a given syllable position. 87) Complex consonant clusters in English a) 'angst' σ

Rh

Nuc Cod

[ ɑ ŋ k s t ]

b) 'strengths' σ

Ons Rh

Nuc Cod

[ s t ɹ ɛ ŋ k θ s ]

25 cp. SON-SEQ Kager 1999: 267 26 Further refinements of this scale are possible so that vowels are also ranked relative to one another in that [a] » [e] [o] » [i] [u] » [ə] (de Lacy & Kingston to appear: 9) but that is beyond the scope of this paper.

Case studies of Ø | 66

Likewise, French has many words which violate the SSP by having a fall in sonority going from the nucleus to an obstruent followed by a sudden rise back up to a liquid. 88) SSP violations in French (Clements 1990: 288) a) table 'table' σ

Ons Rh

Nuc Cod

[ t a b l ]

b) autre 'other' σ

Rh

Nuc Cod

[ o t r ]

Clements (1990: 288) also includes Russian among the list of languages which show sonority reversal giving the examples: rta 'mouth.GEN,' lba 'forehead.GEN,' and mgla 'mist.' Svitlana Filonik (p.c. 2012) informs me that both rta and lba contain Jer vowels (rot and lob in the nominative case respectively). However, this is not the case with mgla. In its various cases the forms are: mgly, mgle, mglu, and mgloj. This shows that Russian does at least allow the violation of the SSP. However, it would still need to be motivated exactly why the mid-vowel would delete – what higher ranked constraints would cause the deletion of this segment in violation of *COMPLEX, MAX I-O, and the SSP? This question will be returned to in the next section. First, though, we must also explore the representation assumed by the analysis that claims that Jer vowels are a true alternation with Ø, which is not without its own problems. Yearley (1995) argues for an alternate representation of roots containing a Jer vowel. In her analysis, instances of Jer vowels are not a deletion of an underlying vowel in a certain environment, but rather the realization of a vowel when necessary. In Yearley's

Case studies of Ø | 67 analysis, the initial consonant is separated from the root by a zero allowing it to be an appendix at the prosodic word level as schematized below. 89) Representation of an exceptional word under Jer ~ Ø alternation ω

σ

Ons Nuc

μ

[lj Ø n - a] While this representation satisfies the SSP in that the lateral consonant is not part of the syllable, it does violate several other constraints such as EXHAUSTIVITY27 and DEP I-O. 28 90) EXHAUSTIVITY (Shattuck-Hufnagel & Turk 1996: 208 and references therein)

A constituent of level Cj in the Prosodic Hierarchy may not dominate a constituent of level Cj – (1 + n) (i.e., of more than one level down) (Parse exhaustively all members of a lower constituent into an immediately dominating constituent)

91) DEP I-O (Kager 1999: 68) Output segments must have input correspondents (No epenthesis) This is not the whole story according to Yearley though. An immediate question would be: how does GEN choose which mid-vowel to epenthesize into a given root. Yearley's answer to this is that it is not a vowel which is being epenthesized, but rather a mora. According to Yearley's analysis, the vowel (/o, ə, or e/) exists in the mental representation, but lacks a timing unit, or mora. By this analysis, the form above would be listed in the mental representation as schematized below.

27 Prince & Smolensky (2004: 33) note that: "GEN freely admits analyses in which segments are left out of syllable structure, and these analyses must be disposed of. Such 'underparsing' analyses violate the constraint PARSE [EXHAUSTIVITY], [... however] underparsing can lead to the satisfaction of [other constraints]. 28 See alternate definition given from Selkirk (1995) in 58.

Case studies of Ø | 68

92) Representation of [ljón] 'linen.NOM' / [ljn-a] 'linen.GEN' without mora29 a) b) μ

/ lj o n / / lj o n a /

If this representation is correct, according to syllable structure constraints (SYL-STRUC) (Prince & Smolensky 2004: 242-6) every syllable must contain one and only one nucleus and it must be the most sonorous element. The nucleus of a syllable must head a mora. In the 92a example, since there are no other options, the Jer vowel /o/ must surface to be the nucleus of the syllable. This surfacing is not in violation of DEP I-O as it was defined in 91, but rather it is in violation of DEP-μ-IO.

93) DEP-μ-IO (Kager 1999: 156) Output mora have input correspondents. Because the /o/ vowel must surface to head the syllable, a mora must be epenthesized to it to satisfy well formedness constraints (cp. Prince & Smolensky 2004: 242-6). 92b, however, does not have this same problem. In this example, the GEN suffix /-a/ contains a mora in its mental representation. When this suffix is added to the root, the /n/ is resyllabified as an onset in satisfaction of the constraint of the same name.

94) ONSET (Kager 1999: 93)

*[σ V (syllables must have onsets) Because the syllable already has a sonorous element30 that can head the syllable that also already heads its own mora, the /o/ can be left unpronounced. These two examples are schematized in tableaux below. 95) Tableau for [ljón] 'linen.NOM' (exceptional segment analysis) j /l on/ SYL-STRUC DEP-μ-IO MAX I-O ljon *! μ *  ljon μ * *! lj n

29 Yearley (1995) states that coda consonants in Russian do not add moraic weight to a syllable and I will assume this to be a correct analysis since I can find no arguments to the contrary. 30 In this particular example the /a/ vowel is even more sonorous than /o/ according to de Lacy & Kingston (to appear: 9) and makes a better syllable head.

Case studies of Ø | 69

96) Tableau for [ljn-a] 'linen.GEN' (exceptional segment analysis) μ SYL-STRUC DEP-μ-IO MAX I-O /ljon-a/ μ *! ljona μ μ *! ljo.na μ *  ljna

Now that both sides of the argument have been outlined, we will explore the potential problems and solutions for each.

3.1.2 Predictions of the exceptional word analysis The argument that Gouskova (2012) (and subsequently Gouskova & Becker to appear) wishes to make is in favor of what she calls the whole morpheme exception hypothesis (WMEH) as discussed with regards to French in section 2.1. 97) Whole morpheme exception hypothesis (Gouskova 2012: 80) There are no lexically exceptional segments – only lexically exceptional morphemes.

What this hypothesis means for the argument put forward by Yearley (1995) is that it is not the Jer vowel that is lexically listed as exceptional (stored without a mora) but rather it is the entire morphological root which is lexically stored as exceptional (in its paradigm). As Gouskova (2012: 81) states "it is well established that an account without some appeal to lexical idiosyncrasy is impossible." She makes this statement because, as previously noted, Jer vowels must be analyzed as a form of deletion, not epenthesis, otherwise the realized vowel would not be predictable: in what instances would an [o] surface instead of an [e] or [ə]? On this point, Gouskova (2012) and Yearley (1995) are in agreement. In an attempt to explain why only mid-vowels participate in the alternation,

Gouskova (2012: 95) argues for the constraints *NUC/ə and *MID which would be ranked above MAX V.

Case studies of Ø | 70

98) *NUC/ə (adapted from Gouskova 2012: 97) Assign one violation each time a vowel of low sonority ([ə]) appears in a position of prominence

99) *MID (Gouskova 2012: 96) Assign a violation mark for every peripheral vowel that is [–High] and [–Low]31

100) MAX V (adapted from Kager 1999: 178) Input vowels must have output correspondents

These constraints (including SYL-STRUC) will correctly predict the outcome of words showing alternations with Ø such as [mestj] 'vengeance' (noun) and [mst-itj] 'avenge' (verb). 101) Tableau for [mestj] 'vengeance' (noun) (WMEH analysis) j / mest / SYL-STRUC *MID MAX V  mestj * m st *! *

102) Tableau for [mst-itj] 'avenge' (verb) (WMEH analysis) j / mest-it / SYL-STRUC *MID MAX V mestitj *  mstitj *

Unfortunately, this constraint ranking as it currently exists cannot make the correct predictions for forms that do not show vowel ~ Ø alternations. 103) Tableau for [mest] 'place.GEN.sg' (WMEH analysis)

/mest/ SYL-STRUC *MID MAX V  mest * m st *! *

104) Tableau for [mest-o] 'place.NOM.sg' (WMEH analysis – first attempt)

/mest-o/ SYL-STRUC *MID MAX V  mesto **!  mest * *  msto * * m st *! **

31 For justification of this, Gouskova cites the harmony scale of Beckman (1997: 14) where it states {high, low} » mid

Case studies of Ø | 71

In un-alternating forms such as that in 103-104, this constraint hierarchy incorrectly predicts the deletion of a non-Jer mid vowel. In order to account for this, as well as to demonstrate the need for morphemes to be lexically marked, Gouskova (2012: 96) proposes the additional constraint *MIDL.

105) *MIDL (Gouskova 2012: 96)

Alternating forms are lexically linked to a *MID constraint

In order to predict the correct outputs, Gouskova ranks the lexically indexed *MIDL constrain above MAX V which are both ranked above the general *MID constraint. When this is done, revisiting the forms in 103-104, the correct output can be predicted. Further, because the constraint that we had used to determine the correct outputs in tableaux 101 and 102 is argued to be the lexically indexed *MIDL constraint, we still maintain the correct output in those tableaux as well. 106) Tableau for [mest-o] 'place.NOM.sg' (WMEH analysis – final attempt)

/mest-o/ SYL-STRUC *MIDL MAX V *MID  mesto ** mest *! * msto *! * m st *!

Because this morpheme is not lexically specified to contain a Jer vowel, the surfacing of [e] and [o] do not cause a violation of the *MIDL constraint and the correct output is thus predicted. As proof that it is morphemes that are lexically listed as exceptional and not segments, Gouskova points to the fact that Jer vowels are always found in the final syllable of a morpheme. 107) Positional generalization (Gouskova 2012: 86) Only the vowel in the last syllable of a morpheme can alternate with zero. If a vowel alternates in the first syllable, it is also the last and only syllable.

The prediction being, of course, that if Jer vowels were segmentally exceptional they should not be positionally restricted within a morpheme: if, phonemically, a vowel can be stored without a mora, it should not matter in which syllable of a morpheme that vowel is stored. If this generalization can be shown to be correct as well as predictable, then this

Case studies of Ø | 72 would be the preferred analysis since it is much simpler than postulating defective representations. This generalization seems to be backed up by Yearley's examples 9 and 10 given below as 108 and 109 respectively. 108) Jers in word-final syllables (Yearley 1995: 542) a) [kuxonj] 'kitchen' (GEN.pl) a') [kuxnj-a] 'kitchen' (NOM.sg) b) [okon] 'window' (GEN.pl) b') [okn-o] 'window' (NOM.sg) c) [spokojen] 'calm' (masc.) c') [spokojn-a] 'calm' (fem.) d) [korotok] 'short' (masc.) d') [korotk-a] 'short' (fem.) The generalization is shown to be especially valid in examples 108(b-d) where any of the vowels (being mid-vowels) have the potential to be deleted otherwise. Further, the examples of word-initial Jers that Yearley provides are mono-syllabic morphemes, also adhering to the generalization. 109) Jers in word-initial syllables (Yearley 1995: 545) a) [voʃ] 'louse' a') [vʃivɨj] 'lice-ridden' b) [ʒog] 'burned' (masc.) b') [ʒgla] 'burned' (fem.) c) [von] 'out' c') [vneʃnij] 'outward' d) [denj] 'day' d') [dnevnoj] 'daily' The same pattern as above can be seen in what Yearley calls multiple medial Jers. These vowel ~ Ø alternations only seem to take place in mono-syllabic morphemes, or the final syllable of bi-syllabic roots. In the resulting morphologically complex word, these Jers seem to be initial or medial, but conform to the generalization in that they are still in the final syllable of the individual morphemes. To further strengthen her account, Gouskova (2012: 114) states that "[i]t is even clearer in Seri than in Russian that ghost segments are nothing more than representationally encoded diacritics for morpheme exceptionality – not segment exceptionality." Given the

Case studies of Ø | 73 importance of this statement, we will now turn our attention to the so-called 'ghost segments' in Seri to investigate the predictions of a null timing slot in that language before concluding whether or not Russian Jer vowels are true alternations with Ø.

3.1.3. 'Ghost' segments in Seri To show how her theory of morpheme exceptionality over segmental exceptionality is cross-linguistically relevant, Gouskova (2012) points to 'ghost' segments in Seri (Marlett & Stemberger 1983). Although in Seri the interesting alternations do not appear in the final syllable of a morpheme, they are still positionally restricted to a morpheme edge according to Gouskova. In Seri, the relevant alternations appear in root-initial position. The generalization to be taken from the positional generalization of Gouskova (2012: 86) given previously as 107 is that morpheme exceptionality should occur at a given edge for a given language. In Russian, that edge was said to be right-linearized, and in Seri that edge would be left-linearized. According to Marlett & Stemberger (1983 qtd. in Gouskova 2012: 113) ghost consonants are the result of null timing slots32 which are not associated with any segmental features. This timing slot is rarely segmentally realized (i.e., unlike Jer vowels which are frequently segmentally realized). The purpose of, or method by which, these ghost consonants can be interpreted is their effect on allomorph selection. For example, the passive is formed by the addition of a morphological prefix which has two forms. [aːʔ-] is used for consonant-initial roots such as [aːʔ-kaʃni] 'be bitten'. [p-] is used for vowel initial forms such as [p-eʃ i] 'be defeated.' However, much like the French examples discussed in section 1.2, some vowel-initial forms behave as though they are consonant-initial, taking the [aːʔ-] prefix, such as [aːʔ-aχs] 'be hit.' For the purposes of passive morpheme selection, Marlett & Stemberger claim that the rules are as given below.33

32 Marlett & Stemberger (1983: 617) assume a segmental tier and a distinct auto-segmental tier for Cs and Vs, not a timing slot that would be associated with moraic values. 33 I will concern myself here with only one of the constructions given in Marlett & Stemberger (1983) because it is one that Gouskova (2012: 113-4) highlights specifically. Marlett & Stemberger highlight several other constructions which show the same type of phenomena though.

Case studies of Ø | 74

110) Passive selection rules in Seri (Marlett & Stemberger 1983: 628)

p- / __[ROOT V ([-p-eʃi] (/-iʃi/) 'be defeated') [+Ablaut] aːʔ- / __elsewhere ([-aːʔ-kaʃni] 'be bitten') ([-aːʔ-aːʔ-itax] 'be made to burn') By these realization rules, forms such as 'be hit' or 'be felt' which on the surface appear to be vowel initial but take the elsewhere prefix as shown in 112 can be analyzed as being separated from the prefix by an empty consonantal position word initially as schematized in 111. 111) Unexpected realizations in Seri (Marlett & Stemberger 1983: 628) a) [aːʔ-aχʃ] / *[p-aχʃ] 'be hit' b) [aːʔ-iː] / * [p-iː] 'be felt' 112) CV tier schematization of Seri 'be hit' and 'be felt' a) VVC CVCC

aː ʔ- Ø a χ ʃ b) VVC CVV

aː ʔ-Ø iː Given the realization rules expressed in 110, it is not because the roots in the above examples are consonant-initial, but rather it is by virtue of them not being vowel-initial that the elsewhere morpheme surfaces.34 A second piece of evidence for the null segments in Seri comes from glottal stop deletion in the passive when the glottal stop is in coda position of a syllable which also has a glottal stop as an onset. 113) Glottal stop realization in Seri passives (Marlett & Stemberger 1983: 628) a) /ʔa-aːʔ-sanx/ → [ʔ-aː-sanx] 'who was carried' b) /ʔi-ʔ-aːʔ-kaʃni/ → [ʔi-ʔ-aː-kaʃni] 'my being bitten' c) /ʔa-aːʔ-otʃ/ → [ʔ-aː-otʃ] 'what was sucked' d) /ʔa-aːʔ-aχʃ/ → [ʔ-aː-aχʃ] 'what was hit'

34 This type of distinction will again be shown to be crucial in section 3.2 in the discussion on /f/ ~ Ø alternations in Irish past tense verbs.

Case studies of Ø | 75

Marlett & Stemberger explain these alternations by first arguing that the final /a/ of the first morpheme and the initial /aː/ of the second morpheme elide. We can imagine that this would be due to some sort of *SUPERHEAVY constraint (cp. *3µ: Kager 1999: 268) that bans three mora in one syllable and that DEP I-O is ranked higher than MAX I-O forbidding the epenthesis of a consonant to resolve this problem. Because languages universally prefer onsets to codas,35 the preceding /ʔ/ is then syllabified as an onset. This creates the environment for /ʔ/ deletion, but only if the /ʔ/ is in coda position. As just stated, /ʔ/ (where possible) resyllabifies as an onset, as it would before a vowel. This is the reason that in 113b it is the coda /ʔ/ of /aːʔ-/ that deletes and not the glottal stop that precedes it in /ʔi-ʔ-/. Because, in 113(c-d) the glottal stop still deletes, Marlett & Stemberger (1983) take this as evidence that there is a Ø consonant preceding the vowel. Despite this being compelling evidence, Gouskova (2012: 114) offers what she thinks is a better explanation of these facts and that is that "some vowel-initial stems have stronger left edges than others." Strong edges prevent consonants from re-syllabifying from a coda into an onset for example, or from coalescing, thus preventing the [p-] passive morpheme for vowel-initial roots. However, rather than appealing to a constraint such as

CRISPEDGE (Ito & Mester 1999) Gouskova would need to argue that once again, these exceptional roots are morphologically exceptional in that they are marked as containing a strong left edge. According to Gouskova (2012: 114) this account is more explanatorily adequate than the exceptional segment analysis in that it also explains why the so-called null consonants are always word initial in Seri (or morpheme final in Russian). However, according to Marlett (1997), there is also data to support that empty consonant slots in Seri also appear in root-medial positions. The morphological realization rules, of course, do not play a part in the identification of root-medial empty consonants the way they did in identifying root-initial ones. In root-medial position, empty consonants are identified by their role in glide formation and in blocking a general rule of low vowel deletion. Consider the paradigms of the verbs 'know' and 'give' listed below.

35 But see Ní Chiosáin 1991: chapter 4 and Windsor 2011: 5 for a discussion on how Irish violates this constraint in favor of the Weight-to-Stress principle.

Case studies of Ø | 76

114) Paradigm of 'know' and 'give' in Seri (Marlett 1997: 1-2) Grammatical 'know' 'give' Additional

description morphology Underlying form of a) /-aØa/ /-æØæ/ -- the verb Subject b) [kíja] [kíjæ] k-i-__ nominalization Independent c) [isíja] [isíjæ] i-si-__ future Proximal d) [imíja] [imíjæ] i-mi-__ realis Emphatic e) [iχóa] [iχóæ] i-χo-__ realis Dependent f) [ipóa] [ipóæ] i-po-__ realis Distal g) [ijóa] [ijóæ] i-jo-__ realis Neutral h) [itáa] [it æ] i-t-__ realis Negative neutral i) [itkmáa] [itkm æ] i-t-m-__ realis Negative j) [iskmáa] [iskm æ] i-s-m-__ independent irrealis Negative emphatic k) [iχomáa] [iχom æ] i-χo-m-__ realis Negative dependent l) [ipomáa] [ipom æ] i-po-m-__ irrealis Negative distal m) [ijomáa] [ijom æ] i-jo-m-__ realis Negative subject n) [imáa] [im æ] i-i-m-__ nominalization o) Imperative [káa] [k æ] k-__ Passive subject p) [ʔapáa] [ʔp æ] ʔa-p-__ nominalization

Unlike the word initial forms in Seri, but similar to the so called exceptional segments in Russian, these root medial Øs are sometimes filled by a glide. Marlett (1997: 2) makes this analysis because the palatal glide in this position (exampled in 114(b-d)) does not appear in other verbs, nor in other members of these paradigms which do not have an /i/ vowel surfacing adjacent to an empty position. In these forms, the features of the adjacent /i/ vowel are able to spread to the empty C position in order to be pronounced as a palatal off

Case studies of Ø | 77 glide [j]. This would take place after a general language wide rule of low-vowel deletion which, as Marlett (1997: 2) describes it: "essentially deletes a (root-initial) short low vowel after another vowel." Descriptively, I will formalize this constraint as below in 115. The derivation for this process is schematized below in 118. μ 115) *V$V [Low] Root initial short low vowels are banned when following another vowel I will also propose a harmony constraint to formalize the reason for glide formation: 36 116) SPREAD-R(V-Place, Sonority) Vocalic place and sonority features are spread rightwards from a vowel to an immediately adjacent consonant.

To reflect the fact that glide formation occurs only in hypothesized empty slots, the harmony constraint must be dominated by IDENT(Place).

117) IDENT(Place) Correspondent segments have identical values for feature [F] [(Place)] (If αRβ and α is [γF], then β is [γF]) 118) Derivation of Seri independent future 'know' and 'give' (Marlett 1997: 3) Independent future (i-si-) 'know' 'give' VCVVCV VCVVCV Mental representation /i s i a Ø a/ /i s i æØæ/ VCV CV VCV CV Low vowel deletion

(*V$Vμ ) [Low] /i s i Ø a/ /i s i Øæ/ VCV CV VCV CV Feature spreading

(SPREAD-R ) (V-Place, Sonority) /i s i j a/ /i s i jæ/ Surface form [isija] [isijæ]

It is worth noting that this glide formation is not evidenced in most other examples, namely 114(e-p). Marlett (1997) explains this by stating that the labio-velar glide [w] is not in the Seri consonant inventory. I will represent this as a contrastive feature constraint (Dresher 2009) banning the combination [+Son] and [DORSAL]. 119) *[+Son, DOR] Segments containing both [+Son] and [DOR] features are banned

36 Cp. HARMONY I-O in Kager (1999: 378).

Case studies of Ø | 78

With this constraint in place, we can now account for all of the data listed in 114 above. Two selected forms 'know' are schematized below to demonstrate the difference between those that show glide formation and those which do not. 120) Derivation of Seri 'know' independent future and emphatic realis 'know' (-aØa) Independent future (i-si-) Emphatic realis (i-χo-) VCVVCV VCVVCV Mental representation /i s i a Ø a/ /i χ o a Øa/ VCV CV VCV CV Low vowel deletion μ (*V$V [Low]) i s i Ø a i χ o Øa VCV CV VCV CV Feature spreading

(SPREAD-R ) (V-Place, Sonority) i s i j a i χ o w a VCV CV *[+Son, DOR] -- i χ o Ø a Phonetic form [isija] [iχoa]

The derivation above provides a visual account of the process, but is perhaps even better explained by the use of a tableaux. 121) Tableau for Seri 'know' in the emphatic realis μ /iχoaØa/ *V$V [Low] *[+Son, DOR] MAX I-O a) iχoaØa *! b) iχo_a *! * c) iχo_ _ **! d) iχo_wa *! * e)  iχo_Øa *

Because of the distinction between Ø and deletion in the above tableau it may be difficult to parse on its own. While, as can be seen in example 120 from the independent future form of the verb, SPREAD-R(V-Place, Sonority) is an active constraint in Seri, the candidate in 121d shows that we may ignore it for this form. The candidate in 121a is ruled out because it violates the low vowel deletion rule by allowing the /o/ and /a/ vowels to surface in adjacency, as does 121b. Although one /a/ vowel deleted in 121b, when this deletion occurs, the second /a/ vowel would also need to be deleted according to this constraint, which is the observed in the 121c candidate. If there were no Ø present in this form, 121c

Case studies of Ø | 79 would be the winning candidate. However, we know that this is not the actual winning candidate because the phonetic output of this form does contain one remaining /a/ vowel. Since we know that it is not the form in 121b that wins, it must be the candidate in 121e which contains a Ø segment and prevents a violation of the low vowel deletion constraint by allowing the two vowels not to be in adjacency. Prediction of 121e as the winning candidate can be obtained by showing that it violates the anti-deletion MAX I-O only once as opposed to the two violations incurred by 121c. It is the cyclic application of a counter feeding rule that is missing from the derivation in 120 but can be easily captured by a constraint hierarchy. By themselves, the derivations in given above may seem more convenient than anything else, that I am merely selecting or inventing constraints to suite the data. However, there is one more problem to consider which should solidify this argument. Marlett (1997) in a footnote states that the rule he calls 'Short Low Vowel Deletion' may be somewhat misleading. It would be more correct to say that the features associated with the short low vowel delete, but the mora attached to it still remains. Marlett notes a difference with respect to this phenomenon based on syllable position. If the deleting vowel has a coda, /o + a/ or /o + æ] become [oː]. In pre-vocalic position (or if the following consonant is the onset to the next syllable) the vowel loss results in a short [o]. I interpret this as suggesting that syllables containing a vowel + coda consonant share a mora.37 When the vowel is lost, the mora link to the coda consonant prevents that mora from deleting and must link to the preceding vowel. If there is no coda consonant, vowel deletion also deletes the mora as it is not still linked to any segment. 122) Mora de-linking vs. mora deletion μ μ μ μ μ μ Mental representation V.VC.CV V.V.CV μ μ μ μ μ Vowel (mora) deletion V. C.CV V. .CV μ μ μ μ μ Mora re-linking (if possible):

Compensatory lengthening Vː C.CV V. .CV

37 See Ní Chiosáin (1991: chapter 4) for attested examples of mora sharing in Irish.

Case studies of Ø | 80

Of course, if the hypothesized Ø were a coda and this caused compensatory lengthening of the preceding vowel, this would perhaps be more telling evidence, however, as previously argued for Seri, onsets are preferred to codas, and therefore the empty consonant cannot be syllabified as a coda. However, according to Marlett (1997) this type of compensatory lengthening is common in Seri. This leads one to wonder why (as evidenced in 113(h-p)) [áa] or [ æ] is ever allowed to surface instead of [áː] or [ ː]. Clearly, and I believe this is the strongest evidence for a Ø in Seri, the Ø consonant prevents this elision. Marlett (1997) defends his transcription of two vowel sequences rather than a long vowel by offering comparative evidence for the phonological patterning of long versus short low vowels. While initial short low vowels delete following another vowel, long low vowels condition the deletion of the preceding vowel. Marlett gives the following example to demonstrate this distinction. 123) [aa] vs. [aː] in Seri (Marlett 1997: 3) a) /aa/ = [aː] 'will be flat' /si-aa/ → [saa] ([saː]) b) /aa/ = [aa] 'will give' /si-aØa/ → [sija] If, in the verbs 'know' and 'give,' this were not a sequence of two /a/ vowels separated by an empty consonant, the two would elide into one long vowel which would show the same phonetic output as 'will be flat' in 123a. As this is not the case, we may conclude that Ø consonants absolutely exist in Seri, and they do so in root-medial position as well as in root-initial position. This is a problem for Gouskova (2012) and Gouskova & Becker (to appear).

3.1.4. Consequences of Seri Ø for Russian At the end of section 3.1.2 I stated that if Gouskova's (2012) generalization about edge alignment and exceptionality could be shown to be correct as well as predictable, then it would be the preferred analysis over defective representations. Against Gouskova's (2012: 114 claim that: "[i]t is even clearer in Seri than in Russian that ghost segments are nothing more than representationally encoded diacritics for morpheme exceptionality – not segment exceptionality," I have shown that this is not true. If the generative prediction that if a segment can be specified as Ø in word initial position, it should be available in any onset position (not just word-initially) is the correct extrapolation from Gouskova's positional

Case studies of Ø | 81 generalization (repeated below as 124) – then Seri certainly shows the generative prediction to be borne out. 124) Positional generalization (Gouskova 2012: 86) Only the vowel in the last syllable of a morpheme can alternate with zero. If a vowel alternates in the first syllable, it is also the last and only syllable

The fact that the generative prediction of allowing Ø to serve as a feasible onset consonant in Seri (as it will also be shown to be in Irish in section 3.2) casts doubt on morpheme exceptionality hypothesis in Russian as well. This is especially true since Gouskova stated that the positional generalization is "even clearer in Seri than in Russian" (2012: 114). The second part of my original claim that morpheme exceptionality could potentially be the preferred analysis rested on whether or not that analysis could make testable predictions. A justified criticism of earlier works that posited an underlying Ø in the mental representation of Russian Jer vowels was; how does one predict whether the Jer vowel will alternate with [o], [ə], or [e]? This is impossible to do unless the underlying form is the featurally specified vowel that alternates with Ø on the surface instead of the other way around. I will make this same argument against Gouskova's exceptional morpheme analysis. In order to specify which words will contain a Jer and which will not, rather than appealing to a defective representation (a vowel that is not linked to a mora in the mental representation) Gouskova appeals to lexical specification. She suggests that morphemes containing a Jer vowel are marked with some feature [L]. This is the same as her analysis for Seri: that "some words contain stronger left edges than others" which is dictated by morphemes which are lexically marked with the feature [L]. This is significantly different than the constraints for lexical specification in Prince & Smolensky

(2004) for example. The constraint LX≈PR of Prince & Smolensky (2004: 51) does not grant lexical specification to lexically marked morphemes so that they may stand apart in the phonology of a language, it states that a lexical word (all lexical words) should be realized as a prosodic word. Further refinement of this constraint is possible so that the category-specific constraint LX≈PR(MCAT) allows lexical words of a particular morphological category to be recognized as prosodic words, but this constraint still requires a defined morphological category which is to be recognized by the phonology – it does not recognize abstract undefined features. Building on this argument, the constraint which causes some

Case studies of Ø | 82 phonological edges to be stronger than others (Gouskova's claim for Seri) is

CRISPEDGE[PCat].

125) CRISPEDGE[PCat] (Ito & Mester 1999: 30) A P(rosodic) Cat(egory) has crisp edges (a given type of prosodic category's edges do not allow the spread of phonological features, or the sharing of prosodic structures between them)

The prosodic category for this constraint may be defined as a syllable, a foot, a word, etc. but in all instances it must be a prosodically defined category, and is not the realization of an abstract morphosyntactic feature. What this means for Gouskova's analysis is that in order to have phonological constraints act on a language's words, those phonological constraints must have a phonologically defined environment to act upon – a lexical flagging in the form of an undefined feature will not do this. The fact that there is no particular morphological or prosodic category that can be linked to Jer vowels removes the predictability from this analysis. However, one possible reason that Russian does not show the positional variability that Seri does is to look at the position of the trigger for the alternation. Jer vowels, according to Gouskova, appear only morpheme finally and their alternation with Ø is triggered by the addition of a grammatical suffix. This is expected under phonological adjacency rules: why would a grammatical suffix trigger an effect multiple syllables away? Whereas I conclude that Seri stores a Ø segment phoneme, a potential Ø in the case of Russian Jers is in the phonological output after a given trigger is activated, and not a stored unit of the morpheme/lexeme. The trigger for this alternation is a suffix, and therefore, only the nearest syllable can be effected.38 In suggesting that Gouskova's analysis is less predictive than positing an underlying Ø-mora, I will also highlight the circular nature of that proposal: morphemes are lexically listed as exceptional because they surface with exceptional traits. I would also argue that since children are statistical learners (Werker & Curtin 2005) looking for patterns in their input to help them parse a speech string and learn language; using universally available phonological constraints they would be more easily able to find a pattern giving evidence for a defective representation (a mora-less vowel) than they would be able to lexically store all forms of 'exceptional' morphemes. This is further augmented by the fact that it is

38 I am thankful to one of my examiners, Dr. Koch, for offering this possible explanation as to why the positional generalization effects Russian Jers and not Seri or French to the same extent.

Case studies of Ø | 83 suggested that the exceptional characteristic of these morphemes is a feature that must be recognized by phonological interface constraints; the name and quality of the feature that causes these various phenomena is that it is specified [L]. This is to say, this feature serves no purpose but to give a mechanical solution to the phenomenon that it is intended to describe.

3.1.5. Ø inputs and outputs for Russian Jers Having explored the analysis of Russian Jers (and Seri empty consonants) that posits phonological Ø to exist in the mental representation of speakers, I will now discuss what this means for the representation this analysis predicts for forms with complex onsets which violate the SSP. This was the representation given in 89, repeated below as 126. 126) Representation of a Jer ~ Ø alternation: [ljn-a] 'linen.GEN' ω

σ

On N

μ

[lj Ø n - a] As discussed in section 3.1.1, this is the output of an underlying representation that has no mora associated with the initial vowel as schematized in 92 repeated below as 127. 127) Representation of [ljón] 'linen.NOM' / [ljn-a] 'linen.GEN' without mora a) b) μ

/ lj o n / / lj o n a / Since this schematization is the only way to predict the surfacing of a Jer vowel, I will conclude that this means Jer vowels are not linked to a Ø-mora specification in their mental representation, but rather have no mora attached to them at all – a crucial difference for cognitive placeholders. This empty moraic position gets filled in by necessity of SYLSTRUC constraints when no other option is available. The question then becomes, when Jer vowels

Case studies of Ø | 84 remain unpronounced (do not epenthesize a mora) does this necessitate a Ø segment output? As was argued earlier, a mapping to Ø is more harmonious than deletion, however in this instance; the principle of stray erasure (Prince & Smolensky 2004) must be invoked. The Ø segment receives a 'null parse' in that it gets associated to no other member of the prosodic hierarchy it is by necessity "entirely silent [which...] would render an item useless, and therefore provide the basis for an explanation for its absence from the lexicon" (2004: 58). This phenomenon certainly backs up the principle of stray erasure since when no mora is epenthesized, the null parse does dictate that the Jer vowel is entirely silent. Does this, however, entail that a Ø vowel is "useless" and therefore entirely deleted by stray erasure? This is the question we will explore with relation to the SSP. As was suggested by the representation depicted in 126 above, Ø does seem to serve a purpose, that is to separate the offending sonorant from the onset of the syllable in order to satisfy the SSP. If we take this as the case, then the output of the null parse does serve a purpose even if it is not pronounced. A constraint ranking favoring the null parse as the optimal output and satisfying the SSP is given below. 128) Tableau for [ljØna] 'linen.GEN' with the null parse (first attempt) μ SSP SYL-STRUC DEP-μ-IO MAX I-O /ljon-a/ μ *! ljona μ μ *! ljo.na μ *! * ljna μ

 lj.Øna

The above tableau shows, if this is the correct analysis, that not only is the null parse the optimal output, but that Ø exists in the phonological output as a means to separate the more sonorous lateral from the less sonorous nasal onset. This structure gives a predictable quality to what seems to be SSP violations in Russian, allowing us to hypothesize that like so many other languages, Russian does not readily violate the SSP. The prediction that this tableau makes for the rest of the language is that wherever we see an SSP violation, the offending consonant is an appendix to the prosodic word and not actually parsed into the

Case studies of Ø | 85 syllable onset. The phonology recognizes that the offending consonant should not be parsed into a syllable by the existence of a prosodic Ø being realized in the output of that word. In the proper environment (when no other syllable nucleus is available in the input) a mora will be epenthesized to that Ø and it will be realized instead as a mid vowel. Because the realized mid-vowel will not be subject to the principle of stray erasure, and is parsed into higher prosodic structure, no word-level appendix is needed, and the otherwise offending consonant can be parsed as an onset since no SSP violations will exist. Unfortunately, this prediction is not borne out. As discussed in section 3.1.1, Clements (1990: 288) gives examples of Russian words which show sonority reversal including mgla 'mist.' Unlike the other forms given so far, this SSP-violating example does not contain a Jer vowel which would separate the /m/ and /g/ consonants in the onset in the correct environment. Svitlana Filonik (p.c. 2012) informs me that in its various other cases the forms are: mgly, mgle, mglu, and mgloj. This shows that Russian either does at least allow the violation of the SSP, or if offending consonants are separated from the syllable by being parsed directly into the word, they are not reliant on a Ø segment to facilitate this parse. In her analysis of Jer vowels, Yearley (1995: 547) apparently chooses the latter option and gives the representation of mgla 'mist' as still containing a word-level appendix to satisfy the SSP. 129) Prosodic parse of illicit onset clusters in Russian (Yearley 1995: 547) ω

σ

μ

C C C V m g l a Note, the above structure differs from the one presented in 126 in that it is not a Ø consonant that forces the separation of /m/ from the /gl/ onset, but is strictly done by the SSP. Whether this is the correct analysis of these structures or the SSP is simply violated in Russian in favor of not epenthesizing additional vowels and mora is inconsequential for this analysis. What we may conclude from this form is that not all forms which appear to

Case studies of Ø | 86 violate the SSP on the surface are dependent on a prosodic Ø. By positing that it is a surfacing prosodic Ø which prevents the violation of the SSP in some instances, but not in others, we lose the predictive quality of Øs as they are concerned with Jer vowels in Russian. What this means is that, while Seri in section 3.1.3 was shown to have output phonological Ø, segmental Ø in Russian is deleted due to the principle of stray erasure: phonological Ø in Russian could alternate with a moraless but fully specified vowel in the input, but it does not serve any purpose in the phonological component and so it is more likely that these moraless vowels are simply deleted when no mora is epenthesized.

3.1.6. Summary The Jer vowel ~ Ø alternations for Russian are best analyzed as underlying mid vowels which delete when not associated to a mora. This means that the vowels associated with the Jer phenomenon are moraless in their underlying representations, but; in order to predict which of the three mid-vowels will surface when no other vowel is capable of forming a nucleus is evidence that those moraless vowels are fully specified. According to the definition of Ø that we have been working with, this means that Russian does not contain Ø mora; rather that it contains vowels that are associated with no mora in the mental representation, not a Ø mora. When no mora is epenthesized for syllable well formedness, it is possible that a Ø segment remains to allow an SSP-offending consonant to be an appendix to the word level. However, this hypothesis does not allow for greater predictive power and introduces complexity to a system where none is needed. Having said that, I contend that the best analysis of Russian Jers is still that of the exceptional segment, not an exceptional morpheme that is somehow lexically listed with a diacritic which survives to the phonological component. The exceptional segment analysis, however, does not need to rely on a prosodic Ø either at the moraic, nor segmental level in order to be complete. On the other hand, the Seri language does provide compelling evidence for a Ø representation and one which survives to the output of the phonological component. The Ø segment in Seri is optionally filled (with [j]), has positional variability (morpheme initial or medial), shows clear adjacency effects (preventing vowel hiatus effects), and allows more predictive power than the WMEH approach that argues that there are exceptional lexically

Case studies of Ø | 87 marked morphemes since diacritic-specific constraints to strengthen or weaken boundaries are not needed under this approach. I defend my analysis of Seri as a Ø consonant and not a Ø mora by pointing out that the Ø segments in that language are uniformly onsets and never nuclei or coda consonants. Since the onset is not a moraic position, Seri must display Ø segments, and not Ø mora. In this section, I examined two languages with the intention of highlighting cases of Ø segments and Ø mora. In the case of Russian Jer vowels, Ø proved to be an unnecessary assumption, or analytic tool. However, this does not mean it was a failed analysis because a strong point was highlighted: Ø and deletion are indeed different, and the analyst who wishes to investigate phonological phenomena via Ø structures must recognize not only the difference between the two, but also when Ø is redundant or cumbersome to the analysis over all. Having looked in depth at an analysis where a Ø mora was potentially the answer to a surface phenomenon, another direction would be to take what was learned from this case study and apply it to the RS phenomenon in Italian discussed in section 2.2. However, due to spatial considerations, that analysis will have to wait for future research. I now proceed to the next in-depth case study of Irish wherein two types of Ø have been previously proposed: an /f/ ~ Ø segmental alternation, and a Ø prosodic word.

Case studies of Ø | 88

3.2. A Ø prosodic word in Irish This section is organized into two major parts. Before making my proposal for the existence of a Ø prosodic word in Irish, I will first show the requirement of positing a Ø segment in Irish. This initial discussion will provide evidence of the need for the phonological component in discussing Irish initial mutations. The environment that the phonological component acts upon to produce the initial mutation lenition will be crucial in the subsequent proposal of a Ø prosodic word.

3.2.1. Consonantal lenition to Ø in Irish mutations Irish lenition is a segmental weakening phenomenon that effects word-initial consonants and has been the object of no small debate. Green (2007: 70) summarizes the problem by stating that "[t]he initial consonant mutations of the Celtic languages are of great interest to theoretical linguists because they appear to be (and are often argued to be) phonological processes that apply in morphosyntactic rather than phonological environments." Lenition is easily recognizable in Irish orthography because it is reliably written as an inserted after the initial consonant.39 The orthographic cues the reader to alternate the radical (phonemic) consonant with a regular and predictable weakened form. The table below gives the radical consonant in phonemic slant brackets on the left and the alternate mutated form in square brackets on the right. The variation in possible mutated forms is owing to whether the radical has an underlying [+Back] or [–Back] feature for velarization or palatalization respectively, but this is also predictable.

39 With the exception of /L, N, and R/ which will be discussed subsequently.

Case studies of Ø | 89

130) Irish consonant alternations under lenition40 /f/ ~ [Ø][Øj] /s/ ~ [h] [ç] /k/ ~ [x] [ç] /p/ ~ [f] [fj] /t / ~ [h] [ç] /g/ ~ [ɣ] [j] /b/ ~ [w][vj] /d / ~ [ɣ] [j] /m/ ~ [w][vj] /l / ~ [ʟ] [lj] /r/ ~ [r] [rj] /n / ~ [n ] [nʲ]41

3.2.1.2. A brief history of Irish lenition and interface theories The lenition phenomenon and its orthographic marking was captured in Vallancey's (1773 [2011]) linguistic grammar of Irish as aspirate mutation with the idea that an [h] segment was somehow combining with the underlying or radical segment to produce a mutated form. This theory was never entirely dismissed. In a dependency phonology analysis of the aspirate mutation, or lenition, Ó Dochartaigh (1979) argued that voiceless stops were characterized with the feature C and voiced stops as C. V When the correct lenition environment was encountered, the governance relationship was mutated into a viz-governance relationship whereby an additional Jacobsonian [Vocalic] feature was added to the governing [Consonantal] feature (Ní Chiosáin 1991). This additional feature was to represent the addition of aspiration which would result in the surfacing of the mutated form. The result was a featural relationship of V,C and V,C for voiceless and voiced consonants respectively.

V In this approach, lenition was considered to be a process whereby the lenited consonant was derived from the radical consonant by the addition of an extra [Vocalic]

40 The lenition of the /l / /n / and /r/ segments is not agreed on by all researchers. It was the opinion of Ó Dochartaigh (1979: 463) that they partook in lenition to a lesser degree. Phonetic evidence supporting this claim has been documented in various dialects of Irish (Finck 1899, Quiggin 1906, Ó Searcaigh 1925, Sommerfelt 1929; 1965, de Bhaldraithe 1945 [1975]; 1953 [1977], Mhac an Fhailigh 1968, Hughes 1986, Ó Siadhail 1989). Green (2007: 75) concedes that "tense sonorants" become lenited to their lax counterparts. See also, Carnie 2002 or Archangeli et al. 2011 for a full discussion of [ATR] in Irish and Scottish Gaelic respectively. 41 I use the [nj] symbol here instead of [ɲ] because it will become important to distinguish the palatal articulation as a secondary articulation which may be the result of a floating feature in subsequent sections.

Case studies of Ø | 90 element to the structure causing it to become more relatively periodic, and thus reducing the consonantal strength of the radical consonant (Ó Dochartaigh 1979: 463; Windsor 2010a: 19). Moving into more contemporary linguistic analyses, the debate has centered largely around which sub-field of linguistics is responsible for the lenition phenomenon. Is lenition a lexical choice that was purely syntactic in nature as Trask (1997) suggested? Are the initial consonants weakened due to computation by the morphological component as Massam (1983) and Green (1997; 2007; 2008) suggest? Does lenition, as a segmental weakening or alternation, belong strictly to the phonology as put forward by Lieber (1987), Ní Chiosáin (1991), Kibre (1995), and Gnanadesikan (1997) among many others? The problem for any of the above mentioned treatments of initial mutations goes back to exactly why Green (2007: 70) said the phenomenon was interesting in the first place: the alternation seems to be phonological in nature, but takes place in a morphosyntactically dictated environment. So how can it be the case that Irish lenitions belong to any one linguistic sub-field? More likely is the idea that lenitions are the surface realization of an interface phenomenon capturing the relationship between the morphosyntax and the phonology. This is exactly the notion put forward in Carnie (1991), Windsor (2010 et seq.) and Gorrie (2011), and what is expected given the direct reference between syntax and phonology argued for in section 1.2. Any number of theories have tried to provide a relationship between the morphosyntax and phonology to govern their relationship and explain Green's statement that a phonological change was taking place in a morphosyntactic environment. McCarthy & Prince (1986 [1996]:6) attempted to describe how prosodic words were understood by the phonological component by putting a constraint on the minimal prosodic size or weight of the word. By this constraint, words provided to the phonological component from the morphosyntax would seek to be isomorphic but would first have to fulfill phonological weight requirements. Prince & Smolensky (1993 [2004]: 51) proposed the constraint

LX≈PR(MCAT) that would allow a lexical item of a particular morphological category to be recognized by the phonological component as a prosodic word. Goldrick (1998 et seq.) put forward the idea of Trubidity Theory which, like the earlier constraints also attempted to force the morphosyntax and phonological components to show isomorphism to one

Case studies of Ø | 91 another. This theory was later expanded upon by van Oostendorp (2004 et seq.) wherein he argued for a theory of Mirroring – that phonology should attempt to perfectly mirror the syntax in terms of structure. Along much the same lines of those theories previously mentioned, Revithiadou (2005) argued for a conceptual model of the interface called

Colored Containment. This model, much like the LX≈PR(MCAT) constraint in Prince & Smolensky's (1993 [2004]) manuscript, listed various morphosyntactic features in the lexicon with colors which could be recognized and interpreted by the phonology in different ways. This allowed the categorization of different types of morphemes to be treated in different but predictable and systematic ways by the phonological component. Dealing specifically with the mutations of the Celtic languages, Green (2007: 84) argued in favor of a constraint called MORPHREAL which gave three conditions on the phonological realization of morphosyntactically governed mutations.

131) MORPHREAL (Green 2007: 84) A morpheme must be realized by fulfilling one of the following conditions: a. the output affixed form contains at least one segment not in the unaffixed form, and that segment(s) is coindexed with a segment(s) in the affix's input; b. the output affixed form contains a segment which is coindexed with the affix's input and that segment has a scale (or feature) value contained in the affix's but not in the unaffixed form; c. the output affixed form contains a segment which is coindexed with the affix's input and that segment has a scale value adjacent to that of the affix's input. That value does not occur in the unaffixed form.

This was later revised to MUTAGREE.

132) MUTAGREE (Green 2008: 204) Mutation triggers and targets must agree in their mutation grade.

Like the other theories discussed above, MORPHREAL/MUTAGREE provided a direct translation between the morphosyntax and the phonology.42 The notion of lenition, or a lenition , being somehow morphologically encoded and interpreted by the phonological component was once more revisited by Gorrie (2011) wherein it is argued that

42 I leave Match Theory (Selkirk 2009 et seq.) out of this discussion because it was discussed in detail in section 1.2 and is the assumed interface theory for this thesis.

Case studies of Ø | 92

43 features such as LENITE and LENITET are morphological features which are interpreted by the phonological component to produce mutations. What all of these theories ultimately argue, both those specifically dealing with Celtic mutations and those not, is that there must be communication between the morphosyntactic component and the phonological component. Ultimately, alternations taking place in the prosodic/segmental/featural domains are the result of phonological constraints acting on the input provided from the SPELLOUT of the morphosyntax. What this means here is that, putting the mutation trigger and/or environment aside for the moment, the alternation of Irish consonants under lenition is the end product of an interface phenomenon, but one that ultimately results in a phonological consonant alternation.

3.2.1.3. Phonological realization of Irish lenitions Now that it is established that there is a phonological constraint which governs Irish consonantal lenition we can explore what that constraint is; how it interacts with other phonological constraints, and how in the end this constraint will produce a cognitive placeholder, or phonological Ø. I will begin this discussion by examining what causes consonants to lenite, and what happens to them when they do so.

Windsor (2010a: 11) argues for the constraint LENITE.

133) LENITE (O'Hagan & Krämer 2004, Windsor 2010a: 11): Reduce articulatory effort – promote the surfacing of the unmarked. This constraint follows from Zipf's Law (Danesi 2004: 12) invoking the Principle of Least Effort which suggests that humans (unconsciously) strive to make use of the least possible amount of biomechanical energy to communicate as long as understanding can be maintained. It also takes into account the language-specific qualification built into the constraint LAZY (Kirchner 1998, Lombardi 2001) by being later described as the driving principle behind Celtic lenition, but not predicting what the alternations should look like other than that the mutated form be less marked than the radical, underlying, consonant.

43 This feature describes the phenomenon of S-Fortition (Ní Chosáin 1991, Green 2008) which is separate from, but undoubtedly related to lenition.

Case studies of Ø | 93

134) LAZY (Lombardi 2001: 88) minimize articulatory effort (i.e. biomechanical energy)... Language-specific lenition patterns arise from LAZY, interacting with faithfulness constraints, within an Optimality Theoretic grammar. Spirantization, for example, is analyzed in terms of rankings where LAZY dominates PRES(continuant)

For the purposes of this discussion, while faithfulness constraints to many features no doubt play a role in lenition patterns, we will concern ourselves specifically with two consonantal alternations, one which forces a stop to become continuant, and one which appears to be a deletion but that I will argue is truly an alternation with Ø. Because these alternations will be our focus, the other most pertinent constraints to consider are MAXI-O, a constraint banning deletions, and IDENT(Continuant), a feature faithfulness constraint.

135) IDENT(Continuant) (Kager 1999: 250) Correspondent segments have identical values for feature [F] [(Continuant)] (If αRβ and α is [γF], then β is [γF])

136) MAX I-O (Kager 1999: 67) Input segments must have output correspondents ('No deletion')

As discussed previously, Wolf & McCarthy (2010) have recently revised The MAX I-O constraint as an anti-deletion constraint which bans a mapping to zero. In their discussion surrounding this constraint, they make reference to the fact that while this constraint bans a mapping to zero, reduction to a cognitive placeholder (ʘ in their convention) is more harmonic than an absolute deletion to nothing (what they characterize as ɸ). To paraphrase in the terms I have outlined in this thesis; reduction to Ø is more faithful to the input than deletion as the input still has a correspondent segment in the output – a cognitive placeholder. It is this dichotomy that I will explore with specific reference to Irish mutations. The two specific alternations I will highlight are: the lenition of /t/ to [h/ç] and the lenition of /f/ to Ø. Consider the following examples: 137) Lenition triggers: an 'very' and ro 'too' a) geal 'bright' [ ʲɑl] b) an gheal 'very bright' [ən jɑl] c) ró gheal 'too bright' [roː jɑl]

As shown above, the adjective geal 'bright' undergoes lenition when merged with the quantifiers an 'very' or ró 'too.' As these two quantifiers are lenition triggers for geal, so too

Case studies of Ø | 94 are they for other adjectives beginning with lenitable consonants. First, I will show the effect of ró 'too' on adjectives beginning with our target radical /t/. 138) ró + /t/-adjectives44 a) taobhach 'lateral' [ tiwɑx] a') ró thaobhach 'too lateral' [roː hiwɑx]

b) taoisleannach 'substantial' [ tiːʃlɛnɑx] b') ró thaoisleannach 'too substantial' [roː hiːʃlɛnɑx]

c) teoranta 'limited' [ ʲoːrəntə] c') ró theoranta 'too limited' [roː çoːrəntə]

d) tiubh 'thick' [ ʲuv] d') ró thiubh 'too thick' [roː çuv]

As the examples above show, in the correct lenition environment, the radical /t/ alternates with either [h] or [ç] depending on the underlying [±Back] specification of the radical. /t/- initial adjectives merged with the quantifier an 'very,' however do not pattern like their /g/- initial counterparts given in 137. In this construction, lenition is blocked by a process known as coronal blocking (Ní Chiosáin 1991, Green 2008, Windsor 2011). This is shown below in 139. 139) Coronal blocking a) taobhach 'lateral' [ tiwɑx] a') an taobhach 'very lateral' [ən tiwɑx]

b) taoisleannach 'substantial' [ tiːʃlɛnɑx] b') an taoisleannach 'very substantial' [ən tiːʃlɛnɑx]

c) teoranta 'limited' [ ʲoːrəntə] c') an teoranta 'very limited' [ən ʲoːrəntə]

d) tiubh 'thick' [ ʲuv] d') an tiubh 'very thick' [ən ʲuv] The reason that an 'very' fails to cause lenition to /t/-initial adjectives even though, as we have already seen, it is a lenition trigger is because the contact consonants between trigger and target are both coronal. Examples such as those above are analyzed as recursive prosodic words wherein the adjacent coronal consonants come to share a single place node

44 Examples c-d are taken from Ní Chiosáin (1991: 27) and have been modified to IPA transcription. Examples a-b are based on my own production of these forms.

Case studies of Ø | 95

([CORONAL]) making the latter word immune to the rules that cause lenition due to geminate inalterability (Ní Chiosáin 1991: 34). This is depicted in 141. 140) Geminate inalterability (Ní Chiosáin 1991: 41)

Structural conditions are strictly interpreted: when a structural condition is imposed on a segment, no part of that segment may obey a contradictory structural condition

141) Lenition as coronal de-linking a) [roː hiːʃlɛnɑx] b) [ən tiːʃlɛnɑx] = COR COR c) [roː çoːrəntə] d) [ən ʲoːrəntə] = COR COR Coronal blocking shows unquestioningly that lenition is a phenomenon that interacts with other phonological constraints. Green (2008), who frequently argues against lenition being interpreted as a strictly phonological process (Green 1997, 2006, 2007, 2008), recognizes this fact and so utilizes his interface constraint MUTAGREE as being dominated by

CORONALHOMORGANICITY (CORHOM) – which must in turn be dominated by IDENT(place).

142) CORONALHOMORGANICITY (CORHOM) (Green 2008: 205)

In ω(...CiCj...), if Ci is coronal, then Cj is coronal

143) IDENT(Place) Correspondent segments have identical values for feature [F] [(place)] (If αRβ and α is [γF], then β is [γF])

In an attempt to ratify an under-generation problem with the CORHOM constraint as well as to reflect a more universal quality of coronals, Windsor (2010a; b; c; 2011) argues for the use of an alternate constraint, CORONALSECOND (COR2).

144) CORONALSECOND (COR2) (Windsor 2010 a; b; c; 2011: 3) Preserve the coronal place specification of the second consonant in a cluster With the necessary constraints defined, we may now try to account for the phonological realization of the alternating consonants under lenition and investigate the interactions between those constraints listed above. As a starting point, I will posit that the interface constraint(s) LENITE/MUTAGREE dominates MAX I-O and subsequently IDENT(Continuant), all

Case studies of Ø | 96

of which are ranked below the un-violated COR2. This should allow the substitution of the radical for the alternate continuants in the correct mutation environment. The effects of this ranking can be seen below. I will start with a tableau for the non-coronal an gheal 'very bright' exampled in 137 which will demonstrate the critical hierarchy between

LENITE/MUTAGREE, MAX I-O, and IDENT(Continuant). 145) Tableau for an gheal 'very bright'

j LENITE/ /ən g ɑl/ COR2 MAX I-O IDENT(Continuant) MUTAGREE ən gjɑl *! ən jɑl * ən ɑl *! ə jɑl *!

If MAX I-O and IDENT(Continuant) were ranked in the opposite order, we would incorrectly predict that deletion would take place rather than changing a stop to a continuant. As this is not the case, we may conclude that these two constraints are in the correct relationship relative to one another. Building on this, if LENITE/MUTAGREE were below IDENT(Continuant) in the ranking, lenition would likely not be observable in Irish. Finally, because COR2 is not violated in Irish, we may also conclude that this is the correct hierarchy for that constraint as well. Moving on to the target /t/-initial adjectives, I will test whether or not this ranking again predicts the correct output candidate. This is used crucially to show the importance of purely phonological constraints interacting with the interface constraints, both dominating them and being dominated by them. 146) Tableau for an tiubh 'very thick' (first attempt)

LENITE/ /ən ʲuv/ COR2 MAX I-O IDENT(Continuant) MUTAGREE ən ʲuv *! ən çuv *! * ən uv *! * ə çuv * *

Unfortunately, this ranking incorrectly predicts that in order to avoid violating the COR2 constraint but still satisfy the interface constraints, Irish would delete the first coronal in the cluster. In order to avoid this possibility, but maintain what is observed for other

Case studies of Ø | 97

consonants, we must rerank the constraints so that MAX I-O is promoted above

LENITE/MUTAGREE which preserves all other tested hierarchies in 145. When this is done, the correct output is once again predicted. 147) Tableau for an tiubh 'very thick' (final attempt)

LENITE/ /ən ʲuv/ COR2 MAX I-O IDENT(Continuant) MUTAGREE ən ʲuv * ən çuv *! * ən uv *! * ə çuv *! *

As we can see from the coronal examples above, in Irish, the constraint banning deletion of a segment crucially outranks the interface constraints that force lenition. This is an important fact for the investigation of Ø as we move on to the lenition of radical /f/. As stated at the beginning of this section, in the appropriate lenition environment, the radical consonant /f/ alternates with Ø. Crucial to our discussion, I argue that this is a case of true alternation with a cognitive place holder which, in this instance, is a Ø segment. This is not an instance of segmental deletion. At the beginning of this section, the alternations of the radical /f/ was given as [Ø] or [Øj]. This notation is quite common in the literature regardless of whether the author is arguing that lenition is an alternation between consonants45 (Green 1997; 2007, Windsor 2010 et seq., Gorrie 2011), or a phonological weakening of a phoneme (Ó Dochartaigh 1979, Ní Chiosáin 1991). Here, I will argue for the former – that consonants under lenition in Irish exhibit alternation, not reduction. To demonstrate that this is the case, we must look at the quantifier + adjective paradigm that we observed previously for /t/ and /g/ and extend this analysis to /f/. 148) lenition triggers and /f/-initial adjectives a) fuar 'cold' [ fuər] a') an/ró fhuar 'very/too cold' [ən/roː uər]

b) folamh 'empty' [ fɔləv] b') an/ró fholamh 'very/too empty' [ən/roː ɔləv] When one of the /f/-initial forms are inserted into the established constraint hierarchy for Irish lenition, we will note that the correct prediction is not made unless we make the

45 Green (1997 et seq.) argues this alternation is due to morphological constraints while Windsor (2010 et seq.) argues that the alternation is due to interface constraints between the morphosyntax and the phonology.

Case studies of Ø | 98 concession that, as Wolf & McCarthy (2010) note, mapping to zero is more harmonious than deletion. 149) Alternation of /f/ R Ø under lenition

LENITE/ /roː fɔləv / COR2 MAX I-O IDENT(continuant) MUTAGREE roː fɔləv *! roː ɔləv *!  roː Øɔləv

As the tableau in 149 demonstrates, if Ø were not a possible alternate segment with /f/ in

Irish, we would expect that the crucial ranking between MAX I-O » LENITE/MUTAGREE would force the realization of the radical /f/ in the output.46 Once again, we know that this ranking crucially exists because of the incorrect output the tableau in 146 predicts when these constraints are reversed.

3.2.1.4. Lenition to palatalized Ø Until now, I have purposefully steered the conversation away from palatalized Ø [Øj] which would come from the lenition of a radical palatalized /fj/. The reason for this is that Øj presents a potential problem for this analysis that needs to be addressed separately from the discussion motivating Ø segments more generally. Ní Chiosáin (1991: 50) discusses the lenition of labiodental fricatives as "deletion as a default rule." In her discussion, Ní Chiosáin suggests that /f/ undertakes spirantization "vacuously," since spirantization is the commonly observed outcome of lenition. Because /f/ goes to Ø (Øj) she argues that /f/ undergoes "Total Deletion [which] can be regarded as a default rule that applies whenever Lenition applies vacuously" (51). To back up this argument, she reminds the reader of /f/ under lenition in which she gives two examples of lenited palatalized /f/.

46 For an independent analysis of /f/ reducing to Ø but not deleting (i.e., a consonant with no internal features) in Scottish Gaelic, see Lodge (2009).

Case studies of Ø | 99

150) Lenition of /fj/47 (Ní Chiosáin 1991: 50-51) a) fíon 'wine' [ fjiːn] a') an fhíon 'the wine' [ənʲ iːn]

b) feoil 'meat' [ ʲoːlʲ] b') an fheoil 'the meat' [ənʲ oːlʲ] If I argue that /f/ is replaced by a cognitive placeholder or Ø segment then the question arises: how does the secondary palatal articulation move across the Ø to become realized on the preceding nasal? Clearly, in other cases of lenition there is no feature spreading from the consonant to be lenited to the previous segment as we have seen in [ən ʲuv] 'very thick.' If Ø represents an occupied segmental position then it should certainly pattern with other occupied segments in similar constructions. At first glance, it would seem more appropriate to suggest that /f/ deletes and that there is a floating [–Back] feature that attaches to the nearest consonant and causes palatalization. Then, of course, if /f/ doesn't alternate with Ø in the case of palatalized consonants, why would we argue it to be the case in non- palatalized consonants? I contend that this is not the correct explanation of the phenomenon, and that this is a case of the phonetics skewing or obscuring the true phonological nature of these segments; that Øj exists phonologically, but that the secondary articulation is phonetically realized on the preceding nasal.48 One such example of the phonetic component restructuring the phonological host to a feature is given in Carnie et al. (2010) and Warner et al. (2011) wherein they note that in Scottish Gaelic, because [+Nasal] cannot be articulated on an oral fricative, the phonetics moves this contrastive feature to a neighboring sonorant.49 In Irish, we can see that when important phonological features, necessary to disambiguate between minimal pairs (such as nasalization (151) or palatalization (152)) can no longer be phonetically realized on their host segment due to articulatory constraints, those features may spread to a preceding articulatorily viable host.

47 The lenition environment in these examples are feminine singular nominative nouns valued as definite. This is a different morphosyntactic environment than what has been shown so far, but the phonetic realities of lenition are the same in any environment. 48 For the purpose of this discussion, I assume that [±Back] is a floating feature which links to a segment but is not part of that segment's feature matrix. Otherwise, if the /fj/ was alternated with Øj we would hypothesize that it would surface as [j] instead of [Øj]. Further study is needed to confirm this hypothesis. 49 See appendix 6.3 for a full discussion of this phenomenon.

Case studies of Ø | 100

151) [±Nasal] as a contrastive feature in Irish

a) bean 'woman' [ ʲɑn] a') mean 'average' [mʲɑn] b) bún 'base' [buːn] b') mún 'urine' [muːn] c) nigh 'clean' [niː] c') di 'for her' [diː] d) bád 'boat (nom.sg / gen.pl)' [bɔːd] d') bán 'white' [bɔːn] e) fad 'duration' [fɑd] e') fan 'wait' [fɑn] Examples 151(a-c) above highlight the importance of word-initial forms that depend on [±Nasal] to distinguish meanings (even without looking at eclipsis mutation which causes voiced word-initial stops to become [+Nasal] in certain morphosyntactic environments). Examples 151(d-e) show this distinction also exists word-finally. 152) [±Back] as a contrastive feature in Irish50

a) bán 'white' [bɔːn] a') binn 'peak' [ ʲɔːn] b) pán51 'pawnshop' [pɔːn] b') peann 'pen' [pʲɔːn] c) cat 'cat' [kɑt] c') cait 'cat (pl)' [kɑ ʲ] d) bád 'boat (nom.sg / gen.pl)' [bɔːd] d') báid 'boat (nom.pl / gen.sg)' [bɔː ʲ] e) fan 'wait' [fɑn] e') fean 'fan (folding)' [ ʲɑn] Examples 152(a-b) above show lexemes that differ only with respect to [±Back] which convey a meaning contrast in word-initial position. Examples 152(c-d) show the same distinction for word-final consonants, and examples 152(e-e') show that word-initially, /f/ also makes use of this meaningful contrast. Just as Carnie et al. (2010) and Warner et al. (2011) observed a repair strategy for maintaining nasality in Scottish Gaelic was to realize the [+Nasal] feature of the derived

50 Forms 152(a-d) are taken from Ní Chiosáin & Padgett (2007: 119), examples 152(e, e') are my own. 51 Spelling is approximated, I am unfamiliar with this form as it seems to be a direct borrowing from English. The Irish word I am familiar with is siopa geallearbóra.

Case studies of Ø | 101 fricative on the preceding vowel, I argue that [–Back] cannot be phonetically realized on a Ø segment in Irish. Because [±Back] is a contrastive feature in Irish, as shown in 152 above, it is necessary to be maintained in the surface phonetic realization. In order to do this, since its phonological segment is not phonetically viable to host the palatalization realization after lenition, that secondary articulation is transferred to the preceding nasal consonant in order to maintain a meaningful contrast. In line with the previous argument that lenition was the phonological realization of an interface phenomenon which resulted in the alternation of a radical consonant for a predictable replacement; I argue that the correct representation of radical /fj/ under lenition is as given below. 153) Derivation for the lenition of /fj/ – an fhliuch 'very wet'

SPELLOUT from morphosyntax [Q very A wet]AP

Phonemic form /ən + fʲlʊx/

j LENITE/MUTAGREE replaces radical Ø

Phonological output [ən Øʲlʊx]

Phonetic realization 〚ənʲ lʊx〛

The above figure schematizes the process that results in a palatalized nasal from a palatalized labio-dental radical consonant. This process begins with the SPELLOUT of a morphosyntactic adjective phrase (AP) that takes a quantifier (Q) as the specifier. When this spells out to the phonological component the phonemes are fed into the constraint hierarchy for Irish. Because [Q A]AP is the correct morphosyntactic environment for j lenition, the interface constraints LENITE/MUTAGREE cause the substitution of the radical /f / for the alternate lenited form [Øj]. The winning candidate from the phonological component is chosen, where [Øj] as a cognitive placeholder for a consonantal slot has a psychological reality. The phonological output is then sent off to the phonetic component for articulation. At this point, because palatalization is contrastive and Ø cannot be phonetically realized, the phonetic component puts the secondary palatal articulation on the preceding consonant resulting in an alveolar nasal with secondary palatal articulation.

Case studies of Ø | 102

3.2.1.5. Interim conclusion II In the preceding discussion, I have defended the argument that consonant lenition in Irish is the phonological substitution of a phonemic radical consonant for a predictable alternate. This argument was based on the fact that the constraints governing lenition must interact with other phonological constraints and can be blocked in certain phonological environments. This view stands apart from other treatments of lenition where phonological rules apply to a phoneme to alter the internal featural structure of that segment (Lieber 1987, Ní Chiosáin 1991, Kibre 1995, and Gnanadesikan 1997 among many others). Under the feature manipulation treatments, lenition (whatever the exact environment may be) caused a series of rules to act on specific, non-natural classes of segments where they could be spirantized, debuccalized, deleted, or undergo glide formation (Ní Chiosáin 1991). Given modern theories of phonology, such a diaspora of rules is no longer desirable when a single interface constraint could better explain the phenomenon. This view is similar to, but still different from the view expressed in Green (1997 et seq.) as well. While the arguments presented here align with the notion that segments under lenition alternate with predictable replacement segments; I argue that because of the phonological interaction of other constraints such as COR2/CORHOM the observable alternations are made by the phonological component to recognize morphosyntactic features, whereas Green (1997 et seq.) argues that segments under lenition are stored in the lexicon as allomorphs which are substituted in purely by the morphological component. I argue this point largely based on the fact that it is precisely phonological constraints that govern the blocking of lenition in certain phonological environments and which govern the deletion of segments, or the lack thereof. After defending my proposal of lenition as the phonological realization of an interface phenomenon, I argued that the correct constraint hierarchy governing lenition in Irish was that given in the tableaux in 147 and 149 and is listed below in 154. 154) Constraint hierarchy governing Irish lenitions

COR2 » MAX I-O » LENITE / MUTAGREE » IDENT(Continuant) This ranking allows lenition alternations to take place insofar as they do not cause second place coronals to delink from their primary place node or any consonant to delete. This

Case studies of Ø | 103 crucial ranking suggests that under lenition conditions, /f/ did not simply delete, but rather alternated with a cognitive place holder, Ø. Because the input radical /f/ has a correspondence in the output (according to Wolf & McCarthy 2010) this output is more harmonious than what Ní Chiosáin (1991: 51) called total deletion. One potential problem from this analysis arose when it was pointed out that the secondary palatal articulation of a [–Back] radical /fj/ was transferred to a preceding nasal across the proposed Ø. To explain this spreading, a similar phenomenon of nasal feature spreading in Irish and Scottish Gaelic was cited to show that when the phonological host segment was unable to maintain a crucial featural property, the phonetic realization of that feature may be on an adjacent segment. In sum, the constraints that realize initial consonant mutation in Irish force the phonological realization of a Ø segment in the output to the phonetic component. That cognitive placeholder may be specified with a palatal secondary articulation which may be transferred to an adjacent segment in the realization of the phonetics of that feature. Having demonstrated the need for segmental Ø in Irish as the result of interface constraints, I will now move on to propose a phonological environment for lenition (as provided by the morphosyntax) and based on this evidence, propose a Ø phonological word as well.

3.2.2. Phonological environment for Irish lenition Graiméar Gaeilge na mBráithre Críostaí 'The Irish grammar of the Christian Brothers' is perhaps the most extensive descriptive grammar of the Irish language. Section four of this manuscript is devoted to examples and environments of lenition in Irish and takes up roughly ten pages. This shows the diverse range of morphosyntactic environments that give rise to lenition. These environments include, but are not limited to the following list and examples of each.

Case studies of Ø | 104

155) Morphosyntactic environments for Irish lenition a) Feminine nouns after the definite article a') bróga [ ʲrʲoːgʲə] na bhróga [nə vʲrʲoːgʲə] shoe.fem.pl def.pl shoe.fem.pl 'shoes' 'the shoes' b) The second component of a compound b') riomhaire [rivʲərʲə] poist [poːʃtʲ] riomhphoist [rivʲfoːʃtʲ] computer.NOM mail.GEN computer+mail 'computer' '(of) mail' 'email' c) Following the vocative case marker c') Seosamh [ʃoːsəv] a Sheosaimh [ə çoːsɪvʲ] Joseph.NOM VOC Joseph.VOC 'Joseph' 'O Joseph' d) After most adverbs d') geal [ ʲaʟ] an gheal [ən jaʟ] bright very bright 'bright' 'very bright' e) After certain prepositions (to, for, from, off, of, in + definite article) e') cistin [kɪʃtʲən] san chistin [san xɪʃtʲən] kitchen.NOM in.def kitchen 'kitchen' 'in the kitchen' f) After most singular possessive pronouns f') máthair [mahɪrʲ] mo mháthair [mo wahɪrʲ] Mother.NOM 1.poss mother 'mother' 'my mother' g) After (most) words denoting quantity g') cloch [klɑx] prátaí [prɑːti] cloch phrátaí [klɑx frɑːti] stone (14 pounds) potato.pl stone potato.pl.GEN 'stone' 'potatoes' '(a) stone (of) potatoes' h) Adjectives modifying a feminine noun h') abhainn [ɑwaɪnʲ] mór [moːr] abhainn mhór [ɑwaɪnʲ woːr] river.fem big river big 'river' 'big' '(a) big river' i) Genitive nouns after an indefinite subject i') buidéil [bʷɪdʲəʟ] bainne [banʲə] buidéil bhainne [bʷɪdʲəʟ wanʲə] bottle.NOM milk.GEN bottle milk.GEN 'bottle' '(of) milk' 'bottle (of) milk' j) After negation j') beidh [ ʲeː] ní bheidh [nʲiː vʲeː] be.FUT neg be.FUT 'will' 'will not' k) The past tense of regular verbs k') dún [duːn] dhún [ɣuːn] close (imperative) close.PST 'close!' 'closed'

Case studies of Ø | 105

The above list is far from exhaustive. However, the data in 155(a-j) do show a certain pattern with regards to the lexemes which show lenition: they are the second member of two closely related words. There are, of course, irregularities and exceptions to this rule. One such exception is the lenition environment outlined above in 155k. This particular example will be explored in depth in section 3.2.2. Another irregularity worth pointing out is the difference in lenition patterns between possessive pronouns and regular nouns according to gender. 156) Gender of possessive pronouns and lenition a) a máthair 3.sg.fem.poss mother 'her mother' b) a mháthair 3.sg. m.poss mother 'his mother'

157) Gender of nouns and lenition a) an bád DEF boat.m 'the boat' b) an bhean DEF woman.fem 'the woman'

As shown above in 156 and 157, it is the masculine possessive pronoun which triggers lenition, but the definite article causes lenition only to feminine nouns. I hypothesize that the difference in these lenition environments is due to the syntactic location and type of the gender feature (Ritter 1993), however, the internal syntax of the Irish determiner phrase is beyond the scope of this thesis.52 Returning to a proposed phonological environment for Irish lenitions, I argue that lenition in Irish affects the left edge of the second or subsequent word parsed into a single higher phonological unit. The higher phonological unit may be a recursive word or a phrase as schematized below.

52 See Windsor (2010a; b) for a discussion of how morphosyntactic gender features may affect the realization of lenition in Irish.

Case studies of Ø | 106

158) Lenition environments a) [ ( ω ) ( ω ) φ ]

Lenition b) [ ( ω ) ( ω ) ( ω ) φ ]

Lenition c) [ ( ω ) φ ] [ ( ω ) ( ω ) φ ] [ ( ω ) φ ]

Lenition d) ( ( ω ) ( ω ) ω )

Lenition The above prosodic structures are provided to the phonological component based on matching constraints from the morphosyntax (Selkirk 2009, 2011; Elfner 2010; 2011; 2012).

159) MATCHPHRASE (adapted from Selkirk 2011: 439) Each syntactic phrase (XP) is interpreted by the phonological component as a phonological phrase (φ)

160) MATCHWORD (adapted from Selkirk 2011: 439) Each syntactic head (Xo) is interpreted by the phonological component as a prosodic word (ω)

By making use of the above matching constraints and positing the exact phonological environment for lenition as the second or subsequent word parsed into a single higher , we may observe the differences in DP internal modification versus copular equation below.

Case studies of Ø | 107

161) Merger inside DP (adapted from Carnie 1991; Windsor 2010a) an bhean mhór DEF woman big 'the big woman'

DP

D NP an N A(P) bean mór [( )ω ( )ω ( )ω]φ / ən ban moːr / [ ən van woːr ]

162) Merger outside DP (adapted from Carnie 1991; Windsor 2010a) tá an bhean mór Be.pres DEF woman big 'the woman is big'

TP

T VP

T V DP V' [–PST] tá D N(P) V A(P) an bean mór [ ( )ω [[( )ω ( )ω]φ [( )ω]φ]φ]φ / taː ən ban moːr / [ taː ən van moːr ]

In the schematization of DP internal modification in 161, we can see that when an adjective is merged inside the DP maximal projection, it is parsed as one of the successive prosodic words inside a single phonological phrase. Therefore, it and the feminine noun both undergo lenition. However, when the adjective is merged external to the DP (162), it is parsed into its own phrase and is removed from the environment which would otherwise make lenition possible. Further examples of syntactic determinism (Carstairs 1988; Lahiri & Plank 2010) will be explored section 3.2.2.3 below. While this analysis is admittedly not complete as I do not yet have a satisfactory explanation for the role of gender features in lenition, I take the above constructions as well

Case studies of Ø | 108 as the striking pattern of second word lenition observable in Irish as evidence that this hypothesis is on the right track. Further, this hypothesis follows from historical lenition environments as attested in Old Irish c. 750CE. In Old Irish, like many other languages, lenition or consonantal weakening took place intervocalically. Matasović (2007) gives the example of Vulgar Latin as a language showing intervocalic lenition, however, not to the same extent as did Old Irish. According to Matasović (2007) in order for a phoneme to lenite in Vulgar Latin, a singleton consonant would need to occur between two vowels and must be parsed into a single prosodic word. On the other hand, Old Irish allowed singleton consonants to lenite intervocalically not only word internally but also when the initial consonant of a word was parsed into a phrase where that word was syntactically "closely associated with the preceding word," (Thurneysen 1966: 77). The difference between Vulgar Latin and Old Irish is depicted below.

163) Lenition environments in Vulgar Latin vs. Old Irish (Windsor 2010a: 11) a) Vulgar Latin b) Old Irish φ φ

ω ω ω ω

σ σ σ σ σ σ

Rh On Rh Rh On Rh On Rh

N N N N N Co

V C V V C V C V C

Lenition environments

Crucially, in Old Irish, despite the environment being able to exist between words, the phonological requirement was that the consonant to be mutated occur intervocalically. If my hypothesis is correct; as final vowels were lost due to apocope, language learners began to associate the lenition environment not with segmental environments, but with prosodic ones: lenition ceased to target intervocalic consonants between "closely associated"

Case studies of Ø | 109 syntactic words (Thurneysen 1966: 77), but rather targeted any initial consonant that was parsed as the second or subsequent word edge within a phrase or recursive word. There are three common arguments against a phonological treatment of lenition in Modern Irish. The first is that it does not affect a natural class of sounds. As proposed in section 3.2.1, I argue that lenition in fact targets the natural class of sounds [+Consonantal]. That is, all consonants partake in lenition to a greater or lesser degree (Ó Dochartaigh 1979: 463). The second argument is that there is no phonological environment for lenition to act upon. In this section, I have proposed a hypothesized phonological environment and given supporting evidence that that analysis is on the correct track. The third argument is that lenition encodes morphosyntactic information about things such as gender, case, and tense. It is this final argument I will address in the following section as I propose a Ø phonological word.

3.2.2.2. Past tense and a Ø prosodic word In the previous section, I frequently pointed out that the proposed phonological environment for lenition could work for the majority of grammatical environments for lenition save one, that given in 155k repeated below as 164. 164) Past tense of the regular verb dún [duːn] dhún [ɣuːn] close (imperative) close.PST 'close!' 'closed'

This one example is potentially counterevidence to the phonological environment proposed in the previous section, and clearly demonstrates that lenition carries with it morphosyntactic information. I propose an analysis of this form that relies on a Ø prosodic word but regularizes this form and allows generalization across all types of lenition evidenced in Irish. That is, I propose that preceding the verb in the past tense is a prosodic word filled with no pronounceable information. Beyond allowing an explanation of lenition that applies across all types of grammatical constructions that result in the mutation, evidence for this proposed Ø word will come from three sources: the syntax-phonology interface, the Old Irish reduplicative preterite (Stifter 2006), and modern Irish vowel-initial verbs with the past tense.

Case studies of Ø | 110

3.2.2.3. Irish syntax-phonology interface and a Ø word As we have seen in previous sections, Irish provides evidence for a direct 1:1 mapping between the syntax and phonology (Green 2008, Windsor 2010a; b; d; 2011, Elfner 2010;

2011; 2012). The MATCHWORD constraint that has been argued for in previous sections states that when there is a syntactic head (Xo) that head should be mirrored in the phonological component as a prosodic word (ω). I argue that in Irish, a rigidly VSOX language, the only syntactic head that precedes the verb is the tense head.53 I propose that when this head is filled with the feature [PST], at linearization, it will force a Ø prosodic word to occur before the verb in the clause and create a lenition environment. We understand Irish to syntactically be a V-to-T movement language, thus resulting in the surface VSOX word order (McCloskey 1996; 2011 among others). When this movement occurs, the verbal head (Vo) incorporates into the tense head (To). Much like the proposed prosodic structure for determiners and nouns in Windsor (2011), this creates an environment where two prosodic words are parsed into a recursive word. This is schematized below. 165) V-to-T movement in Irish (first attempt) TP

To vP

To vo DP v' [PST] subject o o o V v VP verb Vo vo DP object [( ( Ø )ω ( verb )ω )ω [ [ subject ]φ [ [ object ]φ ]φ ]φ ]φ

Lenition

This analysis is, of course, not without its problems; the most major of which are the facts that this environment is found only in the past tense, and the adjoinment direction of the incorporating head.

53 This does not include pre-verbal particles for negation or question formation. See Aquaviva (2010) or Oda (2011) for discussion of these particles and their role in initial mutations of verbs.

Case studies of Ø | 111

I argue that the formation of a Ø prosodic word is dependent on a privative feature [PST]. I will contrast this feature with three other basic tenses in Irish: the imperative, the present, and the future. With regards to the imperative, Rooryck & Postma (2007: 27) group imperative verbs with infinitives as "tenseless verbs" both semantically as well as syntactically. This is demonstrated in Irish as the imperative forms are the only forms of the verb which have no morphology (or lenition) to express a tense feature. In the case of infinitives, Irish uses the aimn briathartha 'verbal noun' and therefore the additional morphology is not related to tense. This is shown in the table below. 166) Verbal morphology by tense in Irish Imperative Infinitive Past Present Future dún 'close' dún dúnadh dhún dúnann dúnfaidh glan 'clean' glan glanadh ghlan glanann glanfaidh caith 'smoke' caith caitheadh chaith caitheann caithfidh

First comparing the past tense verbs to both the imperative and infinitive verbal noun, the difference is simple based on the hypothesized syntactic configuration given in the tree in 165. Whereas, in the past tense, the terminal To node contains an interpretable [PST] feature, the imperative and infinitive verbs are tenseless. Therefore, there are no syntactic features in the terminal To node for the phonology to recognize and thus no possibility of interpreting that head as a prosodic word. The case of the present and future tenses are slightly different. As can be seen by the table in 166, tense morphology in the Irish present and future is realized by verbal suffices whereas, if we interpret lenition as tense morphology, the past tense is marked by a prefix. Clearly there is something different going on between these three tenses. The present and future tenses display what is the expected syntactic configuration of head incorporation; that is, that the moving head be left adjoined to the tree. As can be seen in the tree in 165, in order for the proposed analysis to work, I must posit that the verbal head moves and incorporates to the left of the little-vo head it adjoins with. However, when the little vo head moves and incorporates into To, it adjoins to the right. According to Baker (1988), incorporation of a head forces the moving head to adjoin to the left of the existing head it is incorporating into. This generalization provides the verb+tense suffix since the output of the incorporation is Vo > vo > To. However, in the past

Case studies of Ø | 112 tense this analysis requires that the output be To > Vo > vo in order to get the lenition environment for the past tense verb. The consequences of these facts are that either we must consent that there are different syntactic structures for verbs in the past tense as opposed to those in the present or future, or, that the syntax alone is not responsible for the ordering of the tense morphology in relation to the verb. Windsor (2010d) proposes an OT morphology (Russell 1997, Dressler 2006, Clahsen 2006) analysis of Irish verbal morphology wherein at linearization, bound morphemes are aligned with the right edge of a lexical word 54 (ALIGN L, MBOUND R, ω) and free morphemes are aligned with the left edge of a lexical 55 word (ALIGN R, MFREE L, ω). This analysis suggests that regardless of the syntactic configuration, linearization takes place at SPELLOUT and is subject to extra-syntactic principles. Evidence that this analysis is on the correct path comes from the future conditional mood which displays both the past and the future tenses on the verbal root (Aquaviva 2010, Oda 2011). 167) Double tense morphology in the conditional mood Past Future Conditional (analytic) mol 'praise' mhol molfaidh mholfadh tuig 'understand' thuig tuigfidh thuigfeadh dún 'close' dhún dúnfaidh dhúnfadh

According to Aquaviva (2010) and Oda (2011) the lenition present at the first of the conditional verb is evidence of a [PST] feature and the bolded suffix, being almost identical to the future tense, is evidence of a [FUT] feature also effecting the same verb. I use this same argument as evidence that tense features are privative and that regardless of the 56 syntactic configuration; the location of tense is dictated by linearization at SPELLOUT. Based on the above evidence I conclude that [PST] is a free morpheme and therefore left- linearized to the verb post SPELLOUT. This is depicted derivationally below in 168.

54 Note that the verb is a root or lexical word whereas the tense information are inflectional morphemes. This difference in morphosyntactic structure allows phonology to distinguish between them and allow the proper alignment pattern to take place. 55 See also, McCarthy (2003) for a full discussion of categorical alignment constraints. 56 For a complete discussion of how alignment constraints in Irish may affect syntactic constituents at linearization, see Elfner (2011).

Case studies of Ø | 113

168) V-to-T movement in Irish (effects of linearization) a) Syntax proper TP

To vP

vo To DP v' [PST] subject o o o V v VP verb Vo vo DP object

b) SPELLOUT and Linearization

ALIGN R, MFREE L, ω

[( ( Ø )ω ( verb )ω ( [PST] )ω [ [ subject ]φ [ [ object ]φ ]φ ]φ ]φ

c) Phonological component

[( ( Ø )ω ( verb )ω )ω [ [ subject ]φ [ [ object ]φ ]φ ]φ ]φ

Lenition

This proposal allows a single unified analysis of the syntactic structure of V-to-T movement in Irish while still providing a structure that would maintain the proposed environment for phonological lenition in Irish. However, this still does not account for the idea that lenition morphologically encodes the feature [PST]. To better explain this notion and to provide further evidence for a Ø prosodic word realizing the phonetically null To in Irish, we will now turn to the reduplicative preterite in Old Irish. In Old Irish, even those authors who argue against a phonological analysis of lenition in the modern language agree that lenition was a purely phonological phenomenon which took place in the intervocalic environment that is depicted above in 163b. Lenition of the past tense verb arose from a construction now called the reduplicative preterite (Thurneysen 1966, Lewis 1984, Stifter 2006) wherein the initial syllable of a verb (with some vowel alternations) was reduplicated into a prefix. The reduplicated portion of the

Case studies of Ø | 114 root often would create an intervocalic environment to lenite the root's initial consonant. Stifter (2006: 214) gives the following formula for reduplication of a root into a preterite.57 169) The reduplicated preterite in Old Irish58 root: preterite:

C1(R)VxC2- C1Vy-C1[+ len](R)əC2

In the above formula, the reduplicated syllable creates an intervocalic environment that encapsulates the second instance of C1 between Vy and the now vowel-initial root (R). This causes the second instance of C1 to undergo lenition. This can be seen in the examples below. 170) Reduplication and lenition in Old Irish (Thurneysen 1966: 424-5) Root Reduplicated preterite Gloss a) canid cechan- 'sing' b) ma(i)did memad 'break (intrans.)' c) claidid cechlad 'dig' d) nascid nenasc 'bind' e) gleinn geglann 'learn'

The above forms taken from Thurneysen (1966) are from Old Irish texts such as the Leabhar na h-Uidre 'the book of the Dun Cow'59 and are given only in orthography which did not accurately nor reliably capture lenition. However, Stifter (2006: 214) gives reconstructed phonological forms of selected reduplicated preterites which do show the weakening of the lenited consonants. These forms are listed in 169 below.

57 Lewis (1984: 137) also notes that some future forms in Proto Celtic and Old/Middle Irish showed reduplication. In the conclusion of her paper, Lewis states that the future reduplication structures were more resistant to change than were their past tense counterparts. She proposes that one possible cause of the past tense losing the reduplicative structure more quickly than the future was that the passive form of the verb in the preterite, as opposed to the future passive, was formed on a different stem which did not have reduplication. By analogy to non-reduplicating passive preterites, the active preterite would have lost reduplication. I will not concern myself with future reduplicants (which also showed lenition in Old Irish), but will argue below that the loss of reduplication in the past tense (although possibly influenced by passive preterites through analogy) was due to a change in the phonological environment for lenition away from the segmental tier and into the prosodic level. 58 The fact that the vowel often changes in the reduplicated syllable may be due to prosodic constraints attempting to maximize distinctiveness (Urbanczyk 2005). 59 A manuscript originally scribed in 1106 with some later interpolations from circa 13th century (Thurneysen 1966: 8)

Case studies of Ø | 115

171) Phonological forms of reduplicated preterites in Old Irish (Stifter 2006) Orthographic form Phonological form Gloss a) con-bobaig /kov boβəɣʲ/ 'broke'60 b) cechang- /kʲe əŋg/ 'stepped' c) lelag- /lːʲeləɣ/61 'licked' d) dedag- /deðəɣ/ 'crushed' e) cachan- /ka əv/ 'sing'62

What the tables 170 and 171 show is that regardless of orthographic tradition at the time, lenition was frequently found in the initial consonant of the past tense verb. Lenition was the result of reduplicating a syllable to a prefix position before the verb which created an intervocalic environment that was not separated by (maximally) a prosodic phrase boundary. Interestingly though, the form in 171c and 171e show that in Old Irish consonant clusters (cl and gl) could also lenite although the most common lenition environment was described as intervocalic singleton consonants. O'Connell (1912: 6-8) gives nineteen grammatical contexts for the lenition of initial consonant clusters, many of which are instances of things such as: after proclitic prepositions and verbal particles, the second member of a compound, after inseparable prefixes, and so on. I take this as evidence that even in a time where, at least all authors that I am aware of, agree that lenition was a phonological process, the environment for lenition had already begun to shift from the segmental intervocalic domain to a second word prosodic domain. Back to the topic of the reduplicated preterite though, what do these lenition contexts actually tell us about lenition in the modern language and the encoding of grammatical tense features? I argue that the fact that the past tense of many verbs in Old Irish was formed by a reduplicated syllable prefix gives us evidence of an empty space

60 This is a different form of the verb 'to break' than that given in Thurneysen (1966) or Lewis (1984). I have not provided my own translations, but rely on the accuracy of the glosses by the authors. The difference in forms may be dialectal (see Thurneysen (1966: 12) for a discussion of dialectal differences in Old Irish based on source texts), or that the authors drew from different sources/time periods within Proto/Old/Middle Irish. 61 Stifter (2006) gives the form as /l'eλəɣ/ using the lambda symbol to symbolize lenited /l/. I have regularized the transcription to IPA. In his notes on transcription he describes lambda as being "more or less pronounced like a German or English single . The unlenited Old Irish /l/ on the other hand is pronounced somewhat stronger, probably taking a bit more time" (18). This is backed up by Ó Dochartaigh (1979) and Green (1997) who transcribe the radical sonorants as L N R and their lenited corresponding consonants as l n r respectively, the difference between which is a length contrast. Thurneysen (1966: 85) refers to the unlenited sonorants as being articulated with much greater energy and tongue tensing. He also notes that these differences are not reliably recorded in orthography with the exception that unlenited sonorants were often written with double letters. 62 This is a different form of the verb 'to sing' than is given in Lewis (1984). See footnote 60 above.

Case studies of Ø | 116 available in the phonology that preceded the lexical verb. In order to mark grammatical tense, the phonology was required to pronounce that left-linearized vacant element which it did through phonological reduplication. This is schematized below. 172) Reduplication to fill a void in Old Irish

/([PST])ω (verb √)ω/ MATCHWORD WORDSIZE FILL DEP-IO a) ( Ø )ω (C1V1C2V2C3)ω *(!) *(!) b) (C1V1C2V2C3)ω *! c) ( Ø (C1V1C2V2C3)ω)ω *! d) (C4V3)ω (C1V1C2V2C3)ω *! e) (C1V1)ω (C1V1C2V2C3)ω

In the above tableau, I am proposing the same constraint that I argue is highly ranked in the modern Irish, MATCHWORD, is equally highly ranked in Old Irish. This constraint forces the realization of a prosodic word to encompass the phonetically null To containing the [PST] feature. It is unclear to me whether the constraint prohibiting the null phonological word is

WORDSIZE (McCarthy and Prince 1986 [1996]) or a generic FILL constraint (Prince & Smolensky 2004: 106) that requires that the syllabic elements be filled by segmental material. On the one hand, the null element that I am proposing for this structure is a Ø prosodic word which would be ruled out by WORDSIZE. On the other hand, the reduplicated material that we see fill this void is maximally a light syllable, so it makes sense that a syllabic constraint be at work here. Since the anti-epenthesis constraint DEP I-O also seems to be reasonably highly ranked, Old Irish was forced to reduplicate a contiguous syllable into the preceding prosodic word so that all of the above constraints could be satisfied. This is the structure we see in 172e. Based on the same syntactic structure and SPELLOUT that I o o o propose in 168 for modern Irish ([TP T [V T ]]), the Old Irish structure would be two prosodic words contained within a recursive prosodic word. This prosodic structure puts the second (non-reduplicated) instance of C1 inside of the phonological environment for lenition as it was in the Old Irish period: a singleton consonant that appears intervocalically where the two vowels are not separated by a prosodic phrase boundary. So how do we get from the segmental to the prosodic tier as an environment for lenition as I am proposing? As I suggested on the preceding page, lenition of initial consonant clusters in Old Irish (O'Connell 1912: 6-8) gives evidence that the intervocalic singleton consonant was already not the only environment for lenition in this period, but

Case studies of Ø | 117 also where the word to be lenited was syntactically closely associated with the preceding word (Thurneysen 1966: 77). I propose that two things happened in parallel to produce an empty prosodic word that is the realization of a phonetically null To: children acquiring the language, probably in the late Old to Middle Irish period, shifted the phonological environment for lenition away from the segmental tier to the prosodic tier and, due to Zipf's Law (Danesi 2004, Windsor 2010a), the reduplication was able to be dropped since lenition still marked the existence of a preceding word. As is widely discussed in the literature, one of the first, and perhaps most difficult tasks for a first language learner is to segment the speech signal into words (Stager & Werker 1997, Houston & Jusczyk 2000, Singh et al. 2004, Werker & Curtin 2005, Singh 2007, Curtin 2009, Schmale et al. 2010, Fikkert 2010 among many others). What makes this task so difficult for infants is that, even before knowing the meaning of the words that they are trying to segment from the signal, infants need to figure out the boundaries of these phonological and/or morphosyntactic units (Singh 2007: 3). One such way an infant is able to parse the signal prior to learning word-meaning associations is in learning language- specific phonological contrasts and cues; what Singh (2007: 5) refers to as "the statistical frequency with which particular segments occur in the input." Irish is rich with helpful cues to the language learner to help them parse words; Irish is a rigidly stress-initial language,63 and allows only one prosodic foot in a prosodic word which is anchored on the left edge64 (Green 1997; 2008, Windsor 2011). Stress is a highly salient feature of language and a major word-learning aid for infants (Curtin 2009). Additionally, as infants are learning word pairings (after the first year when they begin to associate meanings to words (Schmale et al. 2010: 1)), Irish marks initial word boundaries not only with stress, but with initial mutations such as lenition. As infants attune to segmental differences in the stressed syllables of words, lenition alternations become a source of word boundaries to help segment the speech signal. This is at the time when infants have honed in on meaningful sound alternations in their native language (10-12 months) (Stager & Werker 1997: 381)

63 The Southern Irish, or Munster, and East Mayo dialects do not have such a straight-forward stress system. See Green (1997: 117-130) for a full discussion of stress shift in these dialects. 64 Recursive prosodic words in Irish such as those found in compounds may exhibit a secondary-primary stress pattern with a foot aligned with the left edge of each of the two internal prosodic words, but this is the exception that proves the rule (Green 2008, Windsor 2011).

Case studies of Ø | 118 and just when their "overly specified word representations are thought to give way to more abstract ones" (Schmale et al. 2010: 1). Putting all of this together in addition to the fact that "learning words involves learning phonological representations" (Fikkert 2010: 227) we can suggest that when faced with a multitude of grammatical constructions which show lenition before they have a word-meaning association; it is not the grammar that an infant is learning. Instead, since infants are statistical learners (Singh 2007: 5), and they are trying to find phonetic as well as prosodic cues to segment words from the speech signal (Houston & Jusczyk 2000: 1570) it is more likely that they will try to regularize the input to give a common environment for a segmental alternation – one that specifically helps them mark prosodic word boundaries and aides them in the segmentation of the speech signal. As Curtin (2009: 1-2) summarizes:65 Prior to learning the mappings between words and their meanings, infants must parse the continuous speech stream into units that correspond to words in the ambient language. Infants' experience with their native language's sound structure provides a foundation for segmenting the speech stream. In particular, knowledge of legal sound combinations of their native language (Mattys & Jusczyk 2001), context-dependent alternations of sound patterns ([Jusczyk et al. 1999]), and language-specific rhythmic patterns ([Curtin et al. 2005, Jusczyk et al. 1999, Polka et al. 2002, Thiessen & Saffran 2007]) guide segmentation.

I propose that this is exactly what children are doing when learning Irish – looking for a common phonological environment for a segmental alternation. The only common environment is as proposed above: anywhere that two prosodic words are parsed into a single higher prosodic unit, lenition will occur at the left edge of the second or subsequent word (Windsor 2010 et seq.). Certainly in Old Irish, first language learners had a reduplicative syllable filling the tense position to cue them to this environment; but how does this transition to the modern language where I propose a Ø prosodic word which can only be recognizable by its effect on the neighboring verb? Earlier, I alluded to Zipf's law, also known as the Principle of Least Effort (Danesi 2004, Windsor 2010a). This law/principle suggests that as talkers, we are basically lazy: "humans (unconsciously) strive to make use of the least possible amount of biomechanical

65 See also Langus et al. (2012) for a discussion of how prosody allows the parsing of hierarchical (syntactic) structures in unfamiliar languages.

Case studies of Ø | 119 energy to communicate without losing the meaning behind their utterance" (Windsor 2010a: 10). What Zipf's law means for Old to Modern Irish is: why reduplicate a syllable when the meaning of that reduplication is recoverable from the phonological effect on the neighboring segment? As the phonological environment for lenition changed, the need for reduplication was lost and thus the additional phonological process was dropped. This notion is echoed in Urbanczyk (2005: 211) when she states that "the notion of enhancement can be extended beyond a segmental level." Here, Urbanczyk is specifically referring to the phonologically conditioned allomorphy of reduplication which may be complete or partial reduplication in several languages based on the recoverability of meaning from the output. The arguments put forward in Urbanczyk (1999; 2005) lend themselves to the present analysis in two ways. First, in instances of reduplication such as those evidenced in Old Irish, phonological enhancement is a common cross-linguistic strategy for facilitating recoverability of grammatical information. If a reduplicated syllable were to be dropped, prosodic enhancement of the remaining units is therefore expected. Second, the fact that it is not just featural or segmental enhancement that occurs to facilitate comprehension (as was the focus of Keyser & Stevens 2006), but also prosodic units above the segment may also influence enhancement gestures. This supports the proposed analysis in suggesting that if an environment for a phonological phenomenon shifted from the segmental level to the prosodic word level, enhancement gestures or segmental alternations would still be an expected outcome where they could enhance the recoverability of the underlying morphosyntactic structure and/or grammatical information. At the end of section 3.2.2, I stated that there were three major arguments against treating lenition in Modern Irish as a phonological phenomenon. At this point, I have argued contra to the first argument that lenition does indeed effect a natural class of sounds: [+Consonantal] sounds. I have provided arguments for the existence of a phonological environment for the sound change contra to the second point of contention. Now I turn to the third problem with treating modern Irish lenition as a phonological phenomenon; that lenition of the verb in the past tense is a morphological marker that encodes tense information. Building on the previous argument, why reduplicate a syllable when the meaning behind reduplication is recoverable from the effect on the neighboring segment, this raises the question of what is actually encoding grammatical tense information. If my

Case studies of Ø | 120 analysis is correct, what is encoding grammatical features such as tense is not the consonant showing lenition. Rather, lenition is a phonological phenomenon which allows recovery of a grammatical construction by cueing the listener to the syntactic structure which has given rise to a particular phonological form. This analysis allows the Ø prosodic word (and thus the phonetically null To in the syntax) to encode the grammatical tense information showing greater isomorphism between the morphosyntax and phonology and thus a preferred structure (McCarthy & Prince 1986, Goldrick 1998 et seq., Prince & Smolensky 2004, van Oostendorp 2004 et seq., Selkirk 2009; 2011, Gorrie 2011, among others). This proposal bears with it the prediction that if the morphosyntactic tense information were not recoverable from a given structure; that is, from lenition, that Modern Irish would use another strategy in those instances. This hypothesis is in fact borne out. As argued previously, lenition effects the natural class of sounds [+Consonantal]. This means that only consonant-initial verbs in the past tense can use lenition as a means of recoverability for a null prosodic word. That in turn means that in order to express morphosyntactic tense features in vowel-initial verbs, Irish would require another strategy. Like the Old Irish reduplicative preterite, Modern Irish fills this tense head position with segmental information. Unlike Old Irish, this is not done with a reduplicated phonological syllable, but simply by epenthesizing a single consonant, [d]. Examples of this can be seen in the table below. 173) Modern Irish d' prefix in vowel-initial past tense verbs Verb Imperative Past Present Future a) ól 'drink' ól d'ól ólann ólfaidh b) imigh 'depart' imigh d'imigh imíonn imeoidh c) éirigh 'rise' éirigh d'éirigh éiríonn éireoidh d) inis 'tell' inis d'inis insíonn inseoidh e) athraigh 'alter' athraigh d'athraigh athraíonn athróidh

Further to the above evidence, the fact that a morpheme is spelled out only when lenition does not allow for the recoverability66 of a morphosyntactic [PST] feature gives additional support to the proposed analysis. This fact also supports the idea that all [+Consonantal] segments are able to be lenited since even though [l, n, r] are not orthographically marked for lenition, neither are they marked with a [d] prefix as vowels are.

66 For a discussion on phonological recoverability, see Archangeli & Pulleyblank (1994: 103).

Case studies of Ø | 121

174) Modern Irish sonorant-initial past tense verbs Verb Imperative Past Present Future a) leag 'lay low' leag leag leagann leagfaidh b) léim 'jump' léim léim léimeann léimfidh c) lig 'let' lig lig ligeann ligfidh d) nigh 'clean' nigh nigh ním nífidh e) rith 'run' rith rith ritheann rithfidh

One potential confounding factor is /f/-initial verbs in the past tense. Recall from section 3.2.1 that /f/ lenites to Ø, which I argued was a true phonological alternation, not a deletion. However, /f/-initial verbs in the past tense pattern with both vowel-initial and consonant-initial ones. In addition to /f/ leniting to Ø in the past tense, [d] is also prefixed to the form. 175) Modern Irish /f/-initial past tense verbs Verb Imperative Past Present Future a) fan 'wait' fan d'fhan fanann fanfaidh b) féach 'appear' féach d'fhéach féacheann féachfidh c) fiafraigh 'ask' fiafraigh d'fhiafraigh fiafraíonn fiafróidh d) fill 'return' fill d'fhill filleann fillfidh e) foghlaim'learn' foghlaim d'fhoghlaim foghlaimíonn foghlaimeoidh

This does present a potential problem for the analysis of /f/ alternating with a phonological Ø segment in that once the radical /f/ lenites, if it is replaced by the Ø we would expect that it would still pattern with consonant-initial forms. Alternately, I propose that even though Ø is the phonological output of /f/ in a lenition context, it is not perceptually salient and therefore does not allow the recoverability of the preceding [PST] feature. Since the morphosyntactic [PST] feature cannot be recovered from the phonology, as it can't be with vowel-initial forms, the epenthesis of the additional [d] is required. This is similar to the case of passive selection rules in Seri discussed in section 3.1.3 in that the elsewhere case was chosen not because Ø caused certain forms to behave as though they were consonant initial, but rather prevented them from being recognized as vowel initial. Here, Ø is not making a form vowel initial, but because Ø is devoid of features, it is simply not processed as consonant initial, forcing the realization of the [d] prefix to allow recoverability.

Case studies of Ø | 122

3.2.3. Summary In the preceding case study, I investigated two instances of proposed Øs in Irish – first at the segmental level, and then at the level of the prosodic word. This study centralized around the topic of the Irish initial mutation, lenition. One of the goals of this study was to motivate that Irish lenition did indeed have a phonological component to it, which has been controversial in the literature for several decades (Ó Murchú 1972, Ó Dochartaigh 1979, Massam 1983, Lieber 1987, Carnie 1991, Ní Chiosáin 1991, Kibre 1995, Gnanadesikan 1997, Trask 1997, Green 1997; 2007; 2008, Windsor 2010 et seq., Gorrie 2011, among many others). Three major objections towards including lenition in the phonological component were noted: the natural class of sounds effected, the phonological environment, and the idea that lenition is a stored allomorph that encodes grammatical information. Evidence for my proposal that lenition was the realization of a phonological phenomenon that is influenced by the SPELLOUT of the morphosyntax came from many areas. First, I provided arguments by several phoneticians documenting Irish as well as phonologists working with Irish over the years that all [+Consonantal] segments took part in lenition to a greater or lesser degree (Ó Dochartaigh 1979 and references therein). This was further backed up later by the morphological evidence presented in the analysis of the past tense where the sonorant consonants not typically thought of as participating in lenition patterned with other consonant initial forms even though they were not orthographically marked. The next argument was that the constraints that governed lenition interacted with other phonological constraints, such as those which prevent the delinking of second position coronals from their primary place node (Ní Chiosáin 1991, Green 2008, Windsor 2011). By showing the constraint ranking relevant to segmental lenition, I was also able to defend the notion that radical /f/ in Irish under lenition alternates with a true phonological Ø. After presenting this notion, several possible counter arguments were presented such as the secondary palatal articulation of the radical /f/ in some cases being realized on a preceding nasal, or that /f/-initial verbs in the past tense in some respects patterned with vowel-initial forms. Each of these potential problems were explained using phonetic enhancement67 (Keyser & Stevens 2006) or through recoverability (Archangeli & Pulleyblank 1994).

67 For a discussion of other similar Gaelic phonetic enhancements, see Appendix 6.3.

Case studies of Ø | 123

The subsequent discussion went from the segmental level to the prosodic to discuss an environment for lenition. I claimed that wherever two prosodic words are parsed in to a single higher prosodic unit (word or phrase) lenition occurs at the left edge of the second or subsequent word. Although the fine structure of the Irish DP was left for future research, this analysis captured the fact that lenition was a second position (word) phenomenon. The one seeming exception to this rule was the verb in the past tense which on the surface appeared to be in first position. This lenition was motivated by interface constraints which matched the syntax and the phonology to allow the tense head to form a Ø prosodic word. Evidence for this analysis was taken not only from current theories of syntax and of the interface, but also from historical and L1 acquisition evidence. From this evidence I was able to defend the idea that lenition itself did not encode grammatical information, but rather it is the surface effect of a phonological phenomenon which allows for the recoverability (Archangeli & Pulleyblank 1994) of the underlying morphosyntactic structure and therefore grammatical information. Based on the above arguments I conclude that the Irish language does indeed allow prosodic Øs to surface both at the prosodic word and segmental levels. One observation that can be made is that it is possible to hypothesize that a language which allows Ø at one level of the prosodic hierarchy will be more tolerant of Øs existing at lower levels of the hierarchy. With only Irish data to evidence this trend thus far, though, this is merely a possibility.

C o n c l u s i o n | 124

4. Conclusion In this thesis, I have explored the theoretical notions of a Ø structure at various levels of the prosodic hierarchy. This research has taken on a two-part approach to the controversial issue of whether Ø is a possible phonological structure or not. I have looked at the theoretical side – does an OT framework, or phonology more generally, allow for Ø structures at certain levels of the prosodic hierarchy, in inputs to or outputs from the phonological component? In these sections, I have attempted to look at languages which could potentially show evidence for or against a Ø structure at a particular level of the prosodic hierarchy. In many of these smaller case studies, such as French h-aspiré, no clear answer was given although I offered speculation as to what further research was needed to determine one way or the other. In other cases, such as Blackfoot, clear phonetic evidence showed that a Ø structure was not well motivated. In either case, much was learned about what phonological Øs are, and how they could be recognized. In order to investigate prosodic Ø, it was first necessary to understand what that structure really is, and the difference between Ø and absolute deletion. In order to resolve these issues, I offered a definition that stated Ø is a cognitive placeholder with psychological reality in the mind of the speaker. Ø, as such, is a phonological empty set, or bracketing structure at a particular prosodic tier containing no internal structure. For this reason, those prosodic structures which necessarily minimally contained internal structure (i.e., a syllable minimally contains a nucleus even if that nucleus is headed by a Ø segment) were argued to not be phonological viable Ø structures. Coming out of this discussion and the brief investigations into languages which may or may not give evidence for Ø structures, four criteria were established for evaluating potential instances of prosodic Ø with some degree of differences based on the level of the hierarchy that the researcher was working on. a) The element should be variably filled. At some stage in the derivation, or in certain constructions, the analyst should be able to highlight an instance where some pronounceable material fills the otherwise Ø element. This was shown with [z], [ʔ], and [ə] in French; [j] in Seri; and [f] in Irish for segments, initial geminates in Italian for mora, and a [d] prefix (reduplicative prefix in Old Irish) in Irish Ø words among various other examples for various tiers.

C o n c l u s i o n | 125

b) The element should show clear adjacency effects. This is highly dependent on the language in question, but typically resembles the under-application of a counter feeding rule to produce the expected results. These adjacency effects are what are required to allow the recoverability, and thus the learnability, of the Ø structure. This type of evidence came from vowel hiatus in French and Seri for segments, compensatory lengthening in Friulian for mora, and lenition in Irish for words. c) The element should show positional variability: i.e., not be consistently morpheme/lexeme initial or final. This evidence was relaxed for prosodic levels of the word or higher since their distribution is limited by the morphosyntax. However, these examples were shown primarily from the existence of French h-aspiré and Seri 'ghost' consonants occurring not just in root-initial position, but in rare circumstances, also morpheme internally. d) The analysis overall must be demonstrated to have more predictive power than alternative analyses. This is a difficult criterion to summarize as it is not only language, but also analysis specific. The goal here is to show explanatory adequacy. Thus, if a WMEH or exemplar explanation can better not only explain but also predict a Ø alternation, it would be the preferred analysis. However, if that analysis fails to show a distinction between words/morphemes/segments which show alternation and those which do not, but a Ø analysis does, then the Ø analysis is the preferred one. Evidence of this sort came from restricting the types of 'diacritics' the phonological component could act upon to well- defined categories. Rather than simply labeling a morpheme/lexeme as 'h-aspiré' or 'exceptional' in some way, the phonological component acts on morphologically defined categories such as free or bound morphemes, or on all phonologically featural specifications or categories in a given language. This was done by remembering that the phonological component is that which deals with the categorical, not the gradient (see Zsiga 2000 as discussed in section 1.4 for further elaboration). Further support for this type of evidence came from cross-linguistic comparisons, historical evidence, and language learning possibilities. These converging evidence sources in addition to the simplification over other theories allowed for the strong arguments in favor of Ø segments and words most notably in Irish, and should be extended to another analyses of Ø for other phenomena and other languages.

R e f e r e n c e s | 126

5. References

Abercrombie, David. 1967. Elements of general phonetics. Chicago: Aldine. Absalom, Matthew; Stevens, Mary & Hajek, John. 2002. A typology of spreading, insertion and deletion: or, What you weren't told about Raddoppiamento Sintattico in Italian. Proceedings of the 2002 Conference of the Australian Linguistic Society. 1-11. Anderson, Stephen. 1985. Inflectional morphology. In Shopen, Timothy (ed.) Language typology and syntactic description:Grammatical categories and the lexicon Vol 3. Cambridge: Cambridge University Press. 150-201. Antworth, Evan L. 1991. Introduction to two-level phonology. Available at: http://www.sil.org/pckimmo/two-level_phon.html#05 Last accessed May 18, 2012. Aquaviva, Paolo. 2010. Dependent verbs and tense in Modern Irish morphosyntax. Presented at the 6th Celtic Linguistics Conference: Dublin, Ireland. September 12. Archangeli, Diana & Pulleyblank, Douglas. 1994. Grounded phonology. Cambridge, MA: MIT Press. Archangeli, Diana; Berry, Jeff; Ji, Sunjing; Josephs, Keisha; Hunt, Nicole; Fisher, Muriel & Carnie, Andrew. 2011. ATR in Scottish Gaelic tense sonorants. In Carnie, Andrew (ed.) Formal Approaches to Celtic Linguistics. Newcastle upon Tyne, UK: Cambridge Scholars Publishing. 283-306. Baker, Mark C. 1988. Incorporation: A theory of grammatical function changing. Chicago: University of Chicago Press. Barrow, John D. 2001. The book of nothing: Vacuums, voids, and the latest ideas about the origins of the universe. New York: Vintage Books. Beckman, Jill. 1997. Positional faithfulness, positional neutralization, and Shona vowel harmony. Phonology 14(1). 1-46. Beckman, M. E. & Pierrehumbert, Jannet. 1986. Intonational structure in Japanese and English. In Ewen, C. & Anderson, J. (eds.) Phonology Yearbook 3. 255-309. Blevins, Juliette. 2006a. An overview of evolutionary phonology. Theoretical Linguistics 32(2). 117-66 Blevins, Juliette. 2006b. Reply to commentaries. Theoretical Linguistics 32(2). 245-56. Bliss, Heather & Gick, Bryan. 2009. Articulation without acoustics: "Soundless" vowels in Blackfoot. Proceedings of the 2009 annual conference of the Canadian Linguistic Association. Boersma, Paul. 1998. Functional phonology. PhD Dissertation, University of Amsterdam, Amsterdam. Bourbaki, Nicolas. 1998. Elements of the history of mathematics. New York: Springer- Verlag. Brenner, Dan; Davis, Andréa; Warner, Natasha; Carnie, Andrew; Fisher, Muriel; Schertz, Jassamyn; Hammond, Michael & Archangeli, Diana. 2011. Can you say [ ] or [x ]? Aerodynamics of nasalized fricatives in Scots Gaelic. Presented at the 162nd meeting of the Acoustical Society of America: San Diego. November 3. Carnie, Andrew. 1991. A domain based phonology: The evidence from Modern Irish. Proceedings of the Leiden Conference for Junior Linguists 3, Leiden University, the Netherlands. 59-76.

R e f e r e n c e s | 127

Carnie, Andrew. 2002. A note on diphthongization before tense sonorants in Irish: An articulatory explanation. Journal of Celtic Linguistics 9. 13-31. Carnie, Andrew; Archangeli, Diana; Fisher, Muriel; Warner, Natasha; Brenner, Dan; Davis, Andrea; Gorrie, Colin; Mathiew, Lionel; Schertz, Jassamyn; Clymer, Micaya; Sung, Jae-Hyn & Ward. 2010. Scottish Gaelic experimental phonology. Presented at the 6th Celtic Linguistics Conference: Dublin, Ireland. September 12. Carstairs, Andrew. 1988. Some implications of phonologically conditioned suppletion. In Booij, Geert & van Marle, Jaap (eds.) Yearbook of Morphology 1988. Dordecht: Foris. 67-94. Chomsky, Noam. 1965. Aspects of the theory of syntax. Cambridge, MA: The MIT Press. Chomsky, Noam. 1995. The minimalist program. Cambridge, MA: MIT Press. Clahsen, Harald. 2006. Dual-mechanism morphology. In Brown, K. (ed.) Encyclopedia of language and linguistics 2nd ed. New York: Elsevier. 11-5. Clements, G. Nick. 1990. The role of the sonority cycle in core syllabification. In Kingston, John & Beckman, M. (eds.) Papers in Laboratory Phonology 1. Cambridge, MA: Cambridge University Press. 283-333. Conway, Elizabeth A. 2008. An analysis of Somali pronunciation errors. MA Thesis: Hamline University, Saint Paul, MN. Côté, Marie-Hélène. 2008. Empty elements in schwa, liaison and H-aspiré: The French holy trinity revisited. In Hartmann, Jutta M. Hegedűs, Veronika & Van Riemsdijk, Henk (eds.) Sounds of silence: Empty elements in syntax and phonology. New York: Elsevier. 61-104. Curtin, Suzanne. 2009. Twelve-month-olds learn word-object associations differing only in stress patterns. Journal of Child Language 36. 1157-65. Curtin, Suzanne; Mintz, T. H. & Christiansen, M. H. 2005. Stress changes the representational landscape: Evidence from word segmentation. Cognition 96. 233- 62. Danesi, Marcel. 2004. A basic course in anthropological linguistics. Toronto: Canadian Scholars' Press Inc. de Bhaldraith Tomás. 1953 [1977]. Gaeilge Chois Fhairrge: An deilbhíocht. Dublin: Dublin Institute for Advanced Studies. de Bhaldraith, Tomás. 1945 [1975]. The Irish of Cois Fhairrge, Co. Galway. Dublin: Dublin Institute for Advanced Studies. de Lacy, Paul & Kingston, John. To appear. Synchronic explanation. Natural Language and Linguistic Theory. de Lacy, Paul. 2006. Transmissibility and the role of the phonological component. Theoretical Linguistics 32(2). 185-96. Dehé, Nicole & Samek-Lodovici, Vieri. 2009. On the prosody and syntax of DPs: Evidence from Italian noun adjective sequences. Natural Language and Linguistic Theory 27. 45-75. Dresher, B. Elan. 2009. The contrastive hierarchy in phonology. New York: Cambridge University Press. Duanmu, San. 1990. A formal study of syllable, tone, stress, and domain in Chinese language. PhD Dissertation. MIT, Massachusetts. Duanmu, San. 2000. The phonology of Standard Chinese. New York: Oxford University Press.

R e f e r e n c e s | 128

Edmonston, Jerold A.; Esling, John H. & Harris, Jimmy G. 2004. Supraglottal cavity shape, linguistic register, and other phonetic features of Somali: http://ling.uta.edu/~Jerry/somali.pdf Last accessed: May 20, 2012. Elfner, Emily. 2006. The mora in Blackfoot. Unpublished MA thesis. Calgary: University of Calgary. Elfner, Emily. 2010. Tonal evidence for prosodic phrasing in Connemara Irish. Presented at the 6th Celtic Linguistics Conference: Dublin, Ireland. September 10. Elfner, Emily. 2011. The interaction of linearization and prosody: Evidence from pronoun postposing in Irish. In Carnie, Andrew (ed.) Formal Approaches to Celtic Linguistics. Newcastle upon Tyne: Cambridge Scholars Publishing. 17-40. Elfner, Emily. 2012. Syntax-prosody interactions in Irish. PhD Dissertation. University of Massachusetts, Amherst. Fikkert, Paula. 2010. Developing representations and the emergence of phonology: Evidence from perception and production. In Fougeron, Cécile; Kuehnert, Barbara; Imperio, Mariapaola & Vallee, Nathalie (eds.) Laboratory Phonology 10. Berlin: De Gruyter Mouton. Finck, F. N. 1899. Die araner Mundart: Ein Beitrag zur Erforschung des Westirischen. Marburg: Elwert. Flemming, Edward. 1995. Auditory representations in phonology. PhD Dissertation, UCLA, California. Flynn, Darin; Windsor, Joseph W.; Carveth, Vincent; Kennedy, Lisa & Hashi, Awil. 2012. Unpacking the phonology of Somali consonant alternations. Unpublished MS: Calgary, University of Calgary. Foley, James. 1972. Rule precursors and phonological change by meta-rule. In Stockwell, R. P. & MaCaulay, R. K. S. (eds.) Linguistic change and generative theory. Bloomington: Indiana University Press. 96-100. Frantz, Donald G. 1991 [2009]. Blackfoot grammar. Toronto: University of Toronto Press. Gick, Bryan; Bliss, Heather; Michelson, Karin & Radanov, Bosko. 2011. Articulation without Acoustics: "Soundless" Vowels in Oneida and Blackfoot. Journal of Phonetics 40: 46-53. Gnanadesikan, Amalia E. 1997. Phonology with ternary scales. PhD Dissertation, University of Massachusetts, Amherst. Goldrick, Matthew. 1998. Optimal opacity: Covert structure in phonology. Unpublished MS, John Hopkins University: New Jersey. Goldsmith, John. 2011. The syllable. In Goldsmith, John, Riggle; Jason & Yu, Alan C. L. (eds.) The handbook of phonological theory 2nd ed. Malden: Wiley-Blackwell. 164-96. Gorrie, Collin. 2011. Adjective agreement in Gaelic: A case for morphophonological features. In Carnie, Andrew (ed.) Formal approaches to Celtic linguistics. Newcastle upon Tyne: Cambridge Scholars Publishing. 321-36. Gouskova, Maria & Becker, Michael. To appear. Nonce words show that Russian yer alternations are governed by the grammar. Natural Language and Linguistic Theory. Gouskova, Maria. 2012. Unexceptional segments. Natural Language and Linguistic Theory 30. 79-133. Green, Antony D. 1997. The prosodic structure of Irish, Scots Gaelic, and Manx. PhD Dissertation: Cornell University, New York.

R e f e r e n c e s | 129

Green, Antony D. 2006. The independence of phonology and morphology: The Celtic mutations. Lingua 116. 1946-85. Green, Antony D. 2007. Phonology limited. Potsdam: Universität für Linguistik. Green, Antony D. 2008. Coronals and compounding in Irish. Linguistics 46(2). 193-213. Gussenhoven, Carlos & Jacobs, Haike. 2005. Understanding phonology 2nd edition. New York: Oxford University Press. Gussenhoven, Carlos. 1988. Intonational phrasing and the prosodic hierarchy. Phonlogica 1988. 89-99. Gussmann, Edmund. 2002. Phonology: Analysis and theory. New York: Cambridge University Press. Haegeman, Liliane & Ihsane, Tabea. 2001. Adult null subjects in the non--drop languages: Two diary dialects. Language Acquisition 9(4). 329-46. Hale, Mark & Selkirk, Elisabeth O. 1987. Government and tonal phrasing in Papago. Phonology yearbook 4. 151-83. Halle, Morris & Vergnaud, J. R. 1987. An essay on stress. Cambridge, MA: The MIT Press. Hannahs, S. J. 2009. The interaction of sonority sequencing and foot structure in Welsh. Presented at the Formal Approaches to Celtic Linguistics Conference: Arizona. March 28. Hannahs, S. J. 2011. Unity in diversity in Welsh: The avoidance of sonority sequencing violations. In Carnie, Andrew (ed.) Formal approaches to Celtic linguistics. Newcastle upon Tyne: Cambridge Scholars Publishing. 359-74. Hartmann, Jutta M. Hegedűs, Veronika & Van Riemsdijk, Henk (eds.). 2008. Sounds of silence: Empty elements in syntax and phonology. Oxford: Elsevier. Hayes, Bruce. 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20(2). 253-306. Hayes, Bruce; Kirchner, Robert & Steriade, Donca (eds.). 2004. Phonetically based phonology. Cambridge: Cambridge University Press. Hooper, Joan B. 1976. An introduction to natural generative phonology. New York: Academic Press. Houston, Derek M. & Jusczyk, Peter W. 2000. The role of talker-specific information in word segmentation by infants. Journal of Experimental Psychology 26(5). 1570- 82. Hughes, A. J. 1986. The Gaelic of Tangaveane and Commeen, County Donegal. PhD Dissertation, Queen's University of Belfast, Belfast. Hume, Elizabeth & Johnson, Keith (eds.). 2001. The role of speech perception in phonology. San Diego: Academic Press. Ifrah, Georges. 2000. The universal history of numbers: From prehistory to the invention of the computer. New York: John Wiley & Sons, Inc. Ito, Junko & Mester, Armin. 1999. The phonological lexicon. In Tsuimura, Natsuko (ed.) Handbook of Japanese linguistics. Oxford: Blackwell. 62-100. Jakobson, Roman & Lots, John. 1949. Notes on the French phonemic pattern. Word 5. 151-8. Jakobson, Roman. 1931. Phonemic notes on Standard Slovak. (Published in Czech in Slovenská miscellanea (Studies presented to Albert Pražak)). Bratislava.

R e f e r e n c e s | 130

Jun, Jongho. 1995. Perceptual and articulatory factors in place assimilation: An optimality theoretic approach. PhD Dissertation, UCLA, California. Jusczyk, Peter W.; Hohne, Elizabeth & Bauman, Angela. 1999. Infants' sensitivity to allophonic cues for word segmentation. Perception and Psychophysics 61. 1465- 73. Kager, René. 1999. Optimality theory. New York: Cambridge University Press. Kaplan, Robert. 2000. The nothing that is: A natural history of zero. Oxford: Oxford University Press. Kaye, J. D., Lowenstamm, J. and Vergnaud, J. R. 1985. The internal structure of phonological Elements: A theory of Charm and Government. Phonology Yearbook 2: 305-328. Kenstowicz, Michael. 1994. Phonology in generative grammar. Cambridge, MA: Blackwell Publishers. Keyser, Samuel J. & Stevens, Kenneth N. 2006. Enhancement and overlap in the speech chain. Language 82(1). 81-109. Kibre, Nicholas J. 1995. A model of mutation in Welsh. MA thesis, University of California at Santa Barbara. Kim, Young-Sun. 2004. English null complementizer as a clitic and intonational phrase boundary. Studies in Generative Grammar 14(3). 349-73. King, Stephen. 2000. On writing: A memoir of the craft. New York: Scribner Classics. Kiparsky, Paul. 1979. Metrical structure assignment is cyclic. Linguistic Inquiry 10(3). 421-41. Kiparsky, Paul. 1981. Remarks on the metrical structure of the syllable. In Dressler, Wolfgang U.; Pfeiffer, Oskar E. & Rennison, John E. (eds.) Phonologica. Innsbruck. Kiparsky, Paul. 2006. The amphichronic program vs. evolutionary phonology. Theoretical Linguistics 32(2). 217-36. Kirchner, Robert. 1998. An effort-based approach to consonant lenition. PhD Dissertation: UCLA, California. Kornai, András. 2008. Mathematical linguistics. Cambridge, MA: Springer. Lahiri, Aditi & Plank, Frans. 2010. Phonological phrasing in Germanic: The judgment of history, confirmed through experiment. Transactions of the Philological Society108(3). 370-98. Langus, Alan; Marchetto, Erika; Bion, Ricardo Augusto Hoffmann & Nespor, Marina. 2012. Can prosody be used to discover hierarchical structure in continuous speech? Journal of Memory and Language 66. 285-306. Lass, Roger. 1976. English phonology and phonological theory: Synchronic and diachronic studies. New York: Cambridge University Press. Lewis, Wendy. 1984. The loss of reduplication in Middle Irish. Proceedings of the Harvard Celtic Colloquium 4. 121-40. Lieber, Rochelle. 1987. An integrated theory of autosegmental processes. Albany: State University of New York Press. Lin, Hsin-Yi & Fon, Janice. 2011. The role of pitch reset in perception at discourse boundaries. ICPhS XVII. 1242-5. Lodge, Ken. 2009. Fundamental concepts in phonology: Sameness and difference. Edinburgh: Edinburgh University Press.

R e f e r e n c e s | 131

Lombardi, Linda. 2001. Segmental phonology in optimality theory. New York: Cambridge University Press. Łubowicz, Anna. 2003. Contrast preservation in phonological mappings. PhD Dissertation, University of Massachusetts, Amherst. Luick, Karl. 1923. Deutsche Lautlehre: Mit besonderer Berücksichtigung der Sprechweise Wiens und der österreichischen Alpenländer. Leipzig: Deuticke. MacAulay, Donald. 1992. The Scottish Gaelic language. In MacAulay, Donald (ed.) The Celtic Languages. Cambridge: Cambridge University Press. 137-248. Marlett, Stephen & Stemberger, Joseph P. 1983. Empty consonants in Seri. Linguistic Inquiry 5. 617-39. Marlett, Stephen A. 1997. Empty consonants in root-medial position. Working Papers of the Summer Institute of Linguistics, University of North Dakota Session 41. 1-4. Massam, Diane. 1983. The morphology of Irish Mutation. MIT Working papers in Linguistics 5: 10-29. Matasović, Ranko. 2007. Insular Celtic as a language area. The Celtic languages in contact: Papers from the workshop within the framework of the XIII international congress of Celtic studies. 93-112. Mattys, Sven L. & Jusczyk, Peter W. 2001. Word segmentation by 8-month-olds: When speech cues count more than statistics. Journal of Memory and Language 44(4). 548-67. McCarthy, John and Prince, Alan. 1986[1996]. Prosodic morphology. Report no. RuCCS- TR-32. New Brunswick, NJ: Rutgers University Center for Cognitive Science. Reprinted in Selected Works of John J. McCarthy. Available at: http://works.bepress.com/john_j_mccarthy/54. Last accessed May 30, 2012. McCarthy, John J. & Prince, Alan. 1990. Prosodic morphology and templatic morphology. In Eid, Mushira & McCarthy, John (eds.) Perspectives on Arabic Linguistics II (Papers from the Second Annual Symposium on Arabic Linguistics). Amsterdam: John Benjamins. 1-54. McCarthy, John J. 1994. The phonetics and phonology of Semitic pharyngeals. In Keating, P (ed), Papers in laboratory phonology III: Phonological structure and phonetic form. London: Cambridge University Press. 191-233. McCarthy, John J. 2002. A thematic guide to optimality theory. Cambridge: Cambridge University Press. McCarthy, John J. 2003. OT constraints are categorical. Phonology 20. 75-138. McCarthy, John J. 2007. Derivations and levels of representation. In de Lacy, Paul (ed.) The Cambridge handbook of phonology. Cambridge, Cambridge University Press. 99-118. McCloskey, Jim. 1996. On the scope of verb movement in Irish. Natural Language and Linguistic Theory 14. 47-104. McCloskey, Jim. 2011. The shape of Irish clauses. In, Carnie, Andrew (ed.) Formal Approaches to Celtic Linguistics. Newcastle upon Tyne, UK: Cambridge Scholars Publishing. 143-78. Mhac an Fhailigh, E. 1968. The Irish of Erris, Co. Mayo. Dublin: Dublin Institute for Advanced Studies. na mBráithre Críostaí, GG. 1999. Irish grammar by the Christian Brothers. Baile Átha Cliath: An Gum.

R e f e r e n c e s | 132

Nagy, Naomi. 1995. Double or nothing: Romance alignment strategies. Penn Working Papers in Linguistics 2(2). 93-102. Nespor, Marina & Vogel, Irene. 1986 [2007]. Prosodic phonology. Riverton, NJ: Foris Publications USA. Ní Chiosáin, Máire & Padgett, Jaye. 2007. Contrast, comparison sets, and the perceptual space. In Parker, Steve (ed.) Phonological argumentation: Essays on evidence and motivation. London: Equinox. 108-26 Ní Chiosáin, Máire. 1991. Topics in the phonology of Irish. PhD Dissertation, University of Massachusetts, Amherst. Ó Cuív, Brian. 1944. The Irish of West Muskerry, Co. Cork. Dublin: The Dublin Institute for Advanced Studies. Ó Dochartaigh, Cathair. 1979. Lenition and dependency phonology. Éigse 17: 457-94. Ó Murchú, Séamus. 1972. Réamhanálú guta mar chlaochlú tús focail 'Reanalysis of vowels as/in connection to initial mutations.' Éigse 14. Ó Searcaigh, S. 1925. Foghraidheacht Ghaedhilge, an tuaiscirt. Belfast: Browne & Nolan. Ó Siadhail, Micheal. 1989. Modern Irish: Grammatical structure and dialectal variation. Cambridge: Cambridge University Press. O'Connell, Rev. F. W. 1912. A grammar of Old Irish. Belfast: Mayne, Boyd & Son, Ltd. Oda, Kenji. 2011. Preverbal particles and irregular verbs in Irish. In Carnie, Andrew (ed.) Formal Approaches to Celtic Linguistics. Newcastle upon Tyne, UK: Cambridge Scholars Publishing. 229-54. O'Hagan, Kingsley & Krämer, Martin. 2004. The acquisition of Irish initial consonant mutations and the universal hierarchy of place features. Presented at the 12th Linguistics Associaion of Great Britain Annual Meeting: University of Surrey, Roehampton. August 31. Ohala, J. J. & Ohala, M. 1995. Speech perception and lexical representation: The role of vowel nasalization in Hindi and English. In Connell, B. & Arvaniti, A. (eds.) Phonology and phonetic evidence. Papers in laboratory phonology IV. Cambridge: Cambridge University Press. 41-60. Ohala, John J. 1986. Consumer's guide to evidence in phonology. Phonology Yearbook 3. 3-26. Orwin, Martin. 1996. A moraic model of the prosodic phonology of Somali. African Languages and Cultures. Supplement 3, Voice and Power: The Culture of Language in North-East Africa. Essays in Honour of B. W. Andrezejewski. 51-71. Pater, Joe. 2004. Exceptions in optimality theory: Typology and learnability. Presented at the Workshop on novel data in phonological theory: New York University. April 9. Paul, Hermann. 1916. Deutsche Grammatik, 4 vols. Halle: Niemeyer. Pierrehumbert, Janet. 2001. Exemplar dynamics: word frequency, lenition and contrast. In Bybee, Joan & Hopper, P. (eds.) Frequency effects and the emergence of linguistic structure. Amsterdam: Benjamins. 137-157. Polka, L.; Sundara, M. & Blue, S. 2002. The role of language experience in word segmentation: A comparison of English, French, and bilingual infants. Presented at the 143rd meeting of the Acoustical Society of America: Special session in memory of Peter Jusczyk: Pittsburgh. June 3.

R e f e r e n c e s | 133

Prince, Alan & Smolensky, Paul. 2004. Optimality theory: Constraint interaction in generative grammar. Cambridge, MA: MIT Press. Quiggin, E. 1906. A dialect of Donegal. Cambridge: Cambridge University Press. Quin, E. G. 1975. Old-Irish workbook. Dublin: Royal Irish Academy. Repetti, Lori. 1994. Degenerate syllables in Friulian. Linguistic Inquiry25. 186-93. Revithiadou, Anthi. 2005. Colored containment and the typology of interface constraints: A case study from lexical accent systems. Workshop on Freedom of Analysis: CASTL. Ritter, Elizabeth. 1993. Where's gender? Linguistic Inquiry 24(4). 795-803. Rooryck, Johan & Postma, Gertjan. 2007. On participle imperatives. In van der Wurff, Wim (ed.) Imperative clauses in generative grammar. Amsterdam: John Benjamins. 273-96. Russell, Kevin. 1997. Optimality theory and morphology. In Archangeli, Diana & Langendoen, D. Terence (eds.) Optimality theory: An overview. Malden: Blackwell Publishing. 102-33. Saeed, John. 1999. Somali. Philadelphia: John Benjamins Publishing Co. Saussure, Ferdinand de. 1916. Cours de linguistique générale. Publié par Charles Bally et al.bert Sechehaye; avec la collaboration de Albert Riedlinger. Éd. Critique préparée par Tullio de Mauro. Paris: Payot. Schmale, Rachel; Cristià, Alejandrina; Seidl, Amanda & Johnson, Elizabeth K. 2010. Developmental changes in infants' ability to cope with dialect variation in word recognition. Infancy. 1-13. Seidl, Amanda. 2001. Minimal indirect reference: A theory of the syntax-phonology interface. New York: Routledge. Seife, Charles. 2000. Zero: The biography of a dangerous idea. New York: Penguin Books. Selkirk, Elisabeth O. 1984. Phonology and syntax: The relation between sound and structure. Cambridge, MA: MIT Press. Selkirk, Elisabeth O. 1986. On derived domains in sentence phonology. Phonology Yearbook 3. 371-405. Selkirk, Elisabeth O. 2009. On clause and intonation phrase in Japanese: The syntactic grounding of prosodic constituent structure. Gengo Kenkyu 136: 35-73. Selkirk, Elisabeth O. 2011. The syntax-phonology interface. In, Goldsmith, John; Riggle, Jason & Yu, Alan C. L. (eds.) The handbook of phonological theory 2nd ed. Malden: Wiley-Blackwell. 435-84. Shattuck-Hufnagel, Stefanie & Turk, Alice E. 1996. A prosody tutorial for investigators of auditory sentence processing. Journal of Psycholinguistic Research 25(2). 193- 247. Shaw, Patricia A., Blake, Susan J., Campbell, Jill & Shepherd, Cody. 1999. Stress in hən'q'əmin'əm' (Musqueam) Salish. In Proceedings of the Workshop on Structure and Constituency in Languages of the Americas IV. University of British Columbia. Sievers, Eduard. 1901a. Grundzüge der Phonetik: Zur Einführung in das Studium der Lautlehre der indogermanischen Sprachen. Leipzig: Breitkopf & Haertel. Sievers, Eduard. 1901b. Sprachgeschichte: Phonetik. In Paul, Hermann (ed.) Grundriss der germanischen Philologie, 2nd ed. Vol 1. Strassburg: Trübner. 283-319.

R e f e r e n c e s | 134

Singh, Leher; Morgan, James L. & White, Katherine S. 2004. Preference and processing: The role of speech affect in early spoken word recognition. Journal of Memory and Language 51. 173-89. Sommerfelt, A. 1929. South Armagh Irish. Norsk Tidsskrift for Sprogviden-skap 2. 107- 91. Sommerfelt, A. 1965. The phonemic structure of the dialect of Torr, Do. Donegal. Lochlann 3. 237-54. Stager C. L. & Werker, Janet. 1997. Infants listen for more phonetic detail in speech perception than in word learning tasks. Nature 388. 381-2. Stanley, Richard. 1967. Redundancy rules in phonology. Language 43. 393-436. Steele, Joshua. 1775 [1779]. An essay towards establishing the melody and measure of speech. London: J. Almon. Available free on Google Play (Google eBook). Steriade, Donca. 1999. Alternatives to syllable-based accounts of consonantal phonotactics. In Fujimura, O; Joseph, B. D. & Palek, B. (eds.) Proceedings of LP '98 1. 205-45. Stifter, David. 2006. Sengoidelc: Old Irish for beginners. Syracuse: Syracuse University Press. Sweet, Henry. 1876. Words, logic, and grammar. London: Philological Society. Ternes, Elmar. 1973. The phonemic analysis of Scottish Gaelic. Dublin: Dublin Institute for Advanced Studies. Tesar, Bruce & Smolensky, Paul. 2000. Learnability in optimality theory. Cambridge, MA: MIT Press. Thiessen, E. D. & Saffran, J. R. 2007. Learning to learn: Infants' acquisition of stress- based strategies for word segmentation. Language Learning & Development 3. 73- 100. Thurneysen, Rudolf. 1946 [1966]. A grammar of Old Irish. Dublin: The Dublin Institute for Advanced Studies. Tranel, Bernard. 1987. The sounds of French: An introduction. Cambridge, UK: Cambridge University Press. Trask, Robert. L. 1997. The history of Basque. London: Routledge. Trubetskoy, N. S. 1939. Grundsüge der Phonologie. Göttingern: Vandenhoeck & Ruprecht. Truckenbrodt, Hubert. 1999. On the relations between syntactic phrases and phonological phrases. Linguistic Inquiry 30(2). 219-55. Urbanczyk, Suzanne. 1999. Double reduplications in parallel. In van der Hulst, Harry; Kager, René & Zonneveld, Wim (eds.) The prosody - morphology interface. Cambridge: Cambridge University Press. 390-428. Urbanczyk, Suzanne. 2005. Enhancing contrast in reduplication. In Hurch, Burnard & Mattes, Veronika (eds.) Studies on reduplication. Berlin: Mouton de Gruyter. 211- 37. Vallancey, Major Charles. 1773 [2011]. A grammar of the Iberno-Celtic, or Irish language. Dublin: R. Marchbnak. van der Mark, Sheena. 2003. The phonetics of Blackfoot pitch accent. MA Thesis, University of Calgary. van Oostendorp, Marc. 2000. Phonological projection: A theory of feature content and prosodic structure. New York: Mouton de Gruyter.

R e f e r e n c e s | 135 van Oostendorp, Marc. 2004. The theory of faithfulness. Unpublished MS: Amsterdam: Maeertens Instituut/KNAW. Wagner, Heinrich. 1958 [1981]. Linguistic atlas and survey of Irish dialects 4 vols. Dublin: The Dublin Institute for Advanced Studies. Warner, Natasha; Carnie, Andrew; Brenner, Dan; Clymer, Micaya; Mathieu, Lionel; Sung, Jae-Hyun; Schertz, Jessamyn; Archangeli, Diana; Hammond, Michael; Fisher, Muriel & Gorrie, Colin. 2011. Nasalization and frication in Scottish Gaelic. Presented at the 14th International Congress of Celtic Studies: Maynooth, Ireland. August 5. Werker, Janet F. & Curtin, Suzanne. 2005. PRIMIR: A developmental framework of infant speech processing. Language Learning and Development1(2). 197-234. Windsor, Joseph W. 2010a. Phono-morphosyntax and all points in between: Irish lenitions and the prosodic phrase. Unpublished honours thesis: Calgary, University of Calgary. Windsor, Joseph W. 2010b. Phono-morphosyntax and all points in between: Irish lenitions and the prosodic phrase. Presented at the 6th Celtic Linguistics Conference: Dublin, Ireland. September 10-12. Windsor, Joseph W. 2010c. CORONALSECOND: A universal constraint. Presented at the Alberta Conference on Linguistics: Calgary. October 30. Windsor, Joseph W. 2010d. Irregular verbal morphology in Irish: Contrasting OT morphology with word and paradigm. Unpublished MS: Calgary, University of Calgary. Windsor, Joseph W. 2011. On the boundaries of Irish prosodic words. Calgary Working Papers in Linguistics 27. 1-14. Wolf, Matthew & McCarthy, John J. 2010. Less than zero: Correspondence and the null output. In Rice, Curt (ed.) Modeling ungrammaticality in optimality theory. London: Equinox Publishing. 17-66. Yearley, Jennifer. 1995. Jer vowels in Russian. Papers in Optimality Theory II. (University of Massachusetts occasional papers in linguistics) Jill Beckman, Laura Walsh Dickey, & Suzanne Urbanczyk (eds.). 533-71. Zsiga, Elizabeth C. 2000. Phonetic alignment constraints: Consonant overlap and palatalization in English and Russian. Journal of Phonetics 28. 69-102.

A p p e n d i c e s | 136

6. Appendices 6.1 Optimality theory: The basics Optimality theory (OT) (Prince & Smolensky 1993 [2004]) is a theory of (primarily phonological)68 grammar wherein hierarchically organized universal constraints filter potential output candidates allowing the optimal one to surface. The primary interaction between constraints in this theory is between two families: markedness and faithfulness. Markedness constraints seek to change the mental representation before the output to phonetics in order to make it easier to pronounce (i.e., reducing consonant clusters, assimilating place of articulation of neighboring sounds to reduce effort, or epenthesizing consonants to avoid vowel hiatus to name a few). Faithfulness constraints seek to maintain the mental representation exactly as it is (i.e., prohibit the deletion of a segment, prevent change to a particular feature, or oppose to epenthesis of additional segments to name a few). These constraints are arranged hierarchically for every particular language so that in language α we might find that constraint A is ranked above constraint B whereas in language β, constraint B is more important that constraint A. In text, this ranking would be denoted as two angled brackets so that language α would be listed as having A » B and language β would have B » A. Another option available in this theory is 'not ranked relevant to.' This status is often used when two like constraints do not have an effect on each other (i.e., if they were both trying to preserve a phonological feature, but were either both ranked above or below the markedness constraint that would try and alter that feature), or when there is insufficient data to demonstrate which of the two constraints are more important in a given language. When one of these cases occurs, the constraints are written with a standing line between them. So if A and B are not ranked relative to each other, but both are organized above C, it would be written as A | B » C. The hierarchy of these constraints allows them to evaluate potential candidates for output that are provided by the function GEN (generate), which are typically listed along the Y axis on the left side of a tableau with the constraints listed along the X axis at the top. In a tableau, constraints that are in fixed superordinate position to another are

68 See Kager (1999) for a discussion of OT syntax, Russel (1997) for a discussion of OT morphology, or Zsiga (2000) for a discussion of OT phonetics.

A p p e n d i c e s | 137 separated by a solid line. Constraints that are not ranked relative to each other are separated by a jagged line. 176) Sample tableau 1 /mental Constraint A Constraint B Constraint C Constraint D representation/ Candidate α Candidate β Candidate γ

The above tableau shows three candidates that will be evaluated by four constraints where the organization is A » B | C » D. When a candidate incurs a violation of a constraint, it is marked with an asterix: *. If the violation is critical (i.e., violates a high ranking constraint thus removing it from the list of possibilities) an exclamation mark is put after the asterix: *!. If it is unclear where a candidate incurred the critical violation (if it violated each of two constraints not ranked relative to one another) a bracketed exclamation mark is put after each of the violations: *(!). In this system the number of violations is not as important as not violating the most important (leftmost) constraints. Thus, a candidate that violated constraints B, C, and D but not A would be preferable to a constraint that only violated constraint A. The winning candidate is demarcated by a rightward pointing finger: .69 177) Sample tableau 1b /mental Constraint A Constraint B Constraint C Constraint D representation/ Candidate α *! Candidate β *(!) *(!)  Candidate γ *

In the above tableau candidate α violated constraint A which was a critical violation because it was the highest ranking constraint where another candidate was still eligible to continue being evaluated. Candidate β violated both constraints B and C, but because they are not ranked relative to one another, it is impossible to know if the critical violation occurred in B or C so both are marked. At this point, both candidates α and β are ineligible to be the winning candidate, but candidate γ is still shown as incurring a violation of

69 A leftward pointing finger () is also occasionally used in tableaux. This denotes an incorrect output and is typicaly used to show a flaw in an argument or to demonstrate that a particular constraint ranking must be incorrect.

A p p e n d i c e s | 138 constraint D. The grey shading indicates that any further violations are irrelevant since the fate of the candidate has already been determined. In order to see exactly how this works we need to define the four constraints and put in linguistic forms for candidates. To do so, we will look at [h] epenthesis in Irish (Windsor 2010a: 27). In the following tableau the constraint list will be replaced with the following: 178) Constraints and definitions

a) IDENT(Cons) Correspondent segments have identical values for a feature [F] [(consonantal)] (If αRβ and α is [γF], then β is [γF])

b) MAX I-O All input segments must have output correspondents (no deletion)

c) *V HIATUS Vowels may not surface in adjacency

d) DEP I-O All output segments must have input correspondents (no epenthesis)

When these constraints are plugged into our sample tableau, we can evaluate the GEN candidates for [h] epenthesis in Irish. 179) Sample tableau 1c [ə.hɪn.in] a hiníon 'her daughter'

/ (əɪnin)ω / IDENT(Cons) MAX I-O *V HIATUS DEP I-O a) (əɪnin)ω *! b) (əjnin)ω *! c)  (əhɪnin)ω * d) (ə_nin)ω *!

In the above tableau, we can see that the candidate in 179a is ruled out because it faithfully allows two vowels (bolded) to surface in adjacency. The candidate in 179b is eliminated because it unfaithfully changed the status of the consonantal feature of the high front vowel to make it a consonant in order not to have two vowels side by side. To avoid the same problem, the candidate in 179d deletes the high front vowel in violation of the MAX I-O constraint making it also ruled out. Finally, the winning candidate in 179c does incur a violation of the DEP I-O constraint, but it is still the optimal candidate in this sequence and so it is the one that is allowed to surface. For more thorough, but still easy to parse discussions of the mechanics of OT, the interested reader is directed to Kager (1999) or Tesar & Smolensky (2000).

A p p e n d i c e s | 139

6.2. Scans of Steele (1775: 25-8) (reproduced from Google eBook)

A p p e n d i c e s | 140

A p p e n d i c e s | 141

A p p e n d i c e s | 142

A p p e n d i c e s | 143

6.3. Scottish Gaelic nasal fricatives: the phonology / phonetics interface As further evidence for the phonetic skewing of a phonological feature, this appendix is used to demonstrate that the phonetic component obscures the location of another phonological feature by realizing it on an adjacent segment in an almost entirely parallel construction. In instances of word-medial lenition70 of the bilabial nasal (now often word- final due to apocope) some scholars have documented instances of surfacing nasal fricatives in several different dialects (Wagner 1981, MacAulay 1992, Green 1997 and references therein). The idea behind these nasal fricatives would be that a lenited /b/ would result in a [–Nasal] alternate and a lenited /m/ would result in a [+Nasal] alternate.71 180) Nasal fricatives in Irish (de Bhaldraithe72 qtd. in Green 1997: 66)

a) lamh 'hand' [ʟaː ] b) samhradh 'summer' [sa raɣ] Beyond these (admittedly limited) examples, Wagner (1958 [1981]) documents a speaker73 who also demonstrated nasalized fricatives, both the labiodental and velar voiced spirants. 181) Nasal fricatives in Tory Island Irish (Wagner 1958[1981])

a) ceanglochaidh mé 'I shall tie' [kʲ ɣlah ə mʲɛ] b) an cheangal tú 'did you tie?' [ɛ xʲ ɣə l tu] c) bhí sé coimhéad orm 'he was watching me' [vʲi ʃɛ ko ɑ d ɔro m] c') var: ...mo choimhéad... [...mə xo ɑ d] Investigation of Wagner's (1958[1981]) Irish dialect atlases did not confirm the nasal fricatives in dialects surrounding those reported in the dialects studied in Ó Cuiv (1944) or de Bhaldraithe (1945) but instead showed nasalized vowels that, according to Green (1997: 66) and MacAulay (1992: 88) is the survival of nasalization spreading even when the

70 A remnant from Old Irish where lenition affected all intervocalic singleton consonants (Windsor 2010a and references therein). 71 Recall that lenition of the radicals /b/ and /m/ result in either a [w] or [v] depending on the underlying [±Back] feature. 72 I can find no mention of nasal fricatives in the original source, but rather, de Bhaldraithe (1945: 46) and Ó Cuiv (1944: 54) claim that vowels were nasalized (weakly) adjacent to derived [w] or [vj] where the underlying form is a lenited /m/. This would allow the distinction between lámha [lɑ ː] 'hand' and lá [lɑː] 'day' (Ó Cuiv 1944: 54). In a previous section, Green (1997: 57) cites Thurneysen (1946) and Quin (1975) as claiming Old Irish had the consonants [ ] and [ ʲ]. Stifter (2006: 18) transcribes this sound as [μ] stating that this sound was pronounced like [β] but with a nasal quality and is the output of a lenited /m/. He notes, however, that in other publications, this segment was frequently transcribed as [ ] or [ ʲ]. I would argue that, because the non-palatalized lenited /m/ in modern Irish is often realized as [w] that the Old Irish consonant was in fact [β] being more of a glide than a fricative and thus able to carry nasalization as discussed below. 73 Seamus O'Mhínan (Jimmy (James) Meenan), a small farmer over age 50 who had almost no English, living on Tory Island at a time when still Irish was the only spoken language.

A p p e n d i c e s | 144 adjacent consonant has been denasalized. Some examples of nasal vowels adjacent to a denasalized fricative are given in 182. Importantly, to the best of my knowledge, Irish has never been analyzed as showing a distinction between oral and nasal vowels nor has it been argued that Irish has phonemic nasal vowels. 182) Nasal vowels adjacent to phonemic nasal consonants (Wagner 1958[1981])

a) lamh 'hand' [ʟ ːvʲe] b) samhradh 'summer' [tɑ rə] c) ramhar 'fat' [ra wər] A similar phenomenon of nasal spreading from nasal fricatives is also found in Scottish Gaelic according to Ternes (1973 qtd. in Carnie et al. 2010). In the Applecross dialect of Scottish Gaelic, nasalization reportedly spread throughout words containing a nasal, both those with a surfacing nasal consonant (as in 183a) and those with an underlying nasal which has been spirantized (as in 183b). 183) Nasal fricatives in Applecross Gaelic (Carnie et al. 2010)

a) seanmhar 'grandmother' [ʃɛ n ɛ r] b) friamh 'root' [fr ] I raise this point here because these instances of nasalization surfacing on fricatives (in both Irish as well as Scottish Gaelic) has repeatedly be reported even though, according to Ohala & Ohala (1995) among others, a nasalized fricative should be an articulatorily impossible gesture: a speaker should not physically be able to generate enough air pressure to both resonate in the nasal cavity and create the turbulent air flow in the oral cavity for frication. This dichotomy between reported phonetics and phonetic principles of articulation caused Carnie et al. (2010); Warner et al. (2011); and Brenner et al. (2011) to do a phonetic study of Scottish Gaelic nasal fricatives. I will summarize their findings here and argue that these results can be extended to the very closely genetically related Irish language. As with Irish, (one of) the empirical issue(s) that Carnie et al. wanted to put to the test was whether or not such a thing as phonetically realized nasal fricatives was possible. The phonological phenomenon behind this would be, as it is claimed for Irish, that lenited

A p p e n d i c e s | 145

/b/ would result in [v] whereas lenited /m/ would result in [ ]. Those authors offer the following short list of minimal pairs.74 184) Scottish Gaelic /m/~/b/ minimal pairs under lenition (Carnie et al. 2010)

a) bhean / mhean 'wife / small' b) bheag / mhèag 'small / whey' c) bhalla / mhala 'wall / eyebrow' In order to test this question, fourteen native Scottish Gaelic speakers participated in both production and perception experiments, eight women and ten men aged between 24-80, all of whom had been monolingual Gaelic speakers until at least age five and still use Gaelic daily at least to some extent. In the production experiment, speakers were fitted with separate nasal and oral air flow masks and were recorded reading a list of 120 Gaelic words. Both spectral and aerodynamic measurements were made to determine the amount of frication and nasal airflow present in the target consonants. The results of this first experiment were dichotomous. Of the fricatives analyzed ([ f x ɣ h ]) only [h ] could reliably be nasalized as this place of articulation does not present the same physical challenges as the nasalization of buccal fricatives outlined above would. Of the remaining buccal fricatives, very few could be truly identified as nasal fricatives (27 out of 1666 tokens). That is, those tokens exhibited relatively high nasal airflow extending over 10% of the consonant. Many of the remaining tokens were shown to have strong F2 and F3 values indicating that they were more accurately analyzed as approximates or vowels or they exhibited nasal airflow only in the transition to or from a vowel. The few tokens chosen as "true nasalized fricatives" demonstrated either weak nasalization, and/or weak frication (Warner et al. 2011) as would be predicted by Ohala & Ohala (1994). Where nasalization was predicted to occur on a fricative due to an underlying phonological nasal segment (lenition of /m/), speakers exhibited several repair strategies: some speakers neutralized the distinction showing no nasalization at all (what Carnie et al. (2010) describes as the most common outcome), other speakers instead realized the nasalization on the preceding vowel. For some speakers, biphasic consonants were

74 Carnie et al. (2010) offer only an orthographic representation of these minimal pairs and I do not feel confident in my own Scottish Gaelic pronunciations to give approximate transcriptions.

A p p e n d i c e s | 146 observed so that one segment would show the fricative and the next would exhibit the nasal feature. For yet other reported nasal buccal fricatives speakers did display some nasal airflow but in these instances the target buccal fricative was realized instead as a glide with either weak frication and/or weak nasalization – resolving the articulatory problem of insufficient airflow. Warner et al. (2011) summarize these findings in three points noting: most often, the nasalization distinction is neutralized except on back fricatives ([x ɣ h ]) and underlying (lenited), non-initial /m/. Rarely is nasalization maintained on the preceding vowel, especially for [x] and /m/ (where more nasal airflow is observed on the fricative/approximate). [h], as expressed above, can be nasal where there is no physical/aerodynamic conflict. For the perceptual experiment (Carnie et al. 2010), a native Gaelic speaker was recorded producing a list of thirty three nonce words provided with orthographic or consonants to produce the distinction between nasalization realizations in the consonants or preceding vowels. The nonce words were played for the participants and they were asked to determine if the word was orthographically represented as or . A trend of correct selections was observed, but it was marginal and Carnie et al. note that there were too few tokens used in this study to make a statistical generalization and that the observation was based on researcher perception not statistically relevant data. Warner et al. (2011) reports this finding as listeners being able to perceive the nasal distinction, but just barely, stating that the preceding vowel would provide as much information to the listener as any other segment given that the nasalization was demonstrated on the previous vowel when present at all in concert with the preliminary airflow results (Carnie et al. 2010). Given this conclusion, Warner et al. (2011) asked the critical question: is this nasal spreading of derived nasals phonetic spreading of the nasal quality or phonological spreading of a feature? Since interceding vowels only demonstrate nasal airflow through roughly half of their duration and nasalization can still be observed across these vowels, Warner et al. (2011) conclude that this is an instance of true phonological feature spreading and is not due to purely phonetic transition. Relating this to the discussion of the spreading of a phonological [–Back] feature across a lenited /fj/ onto a preceding nasal (section

A p p e n d i c e s | 147

3.2.1.4), this shows how the phonetic component may obscure the phonological output by realizing a feature on an adjacent segment than the one the phonology suggest it should be associated with. Both of these instances of phonetic manipulation of the phonological outputs in Gaelic (both Irish and Scottish) highlight the fact that phonology is not restricted by articulatory principles and may attempt to place impractical features on a particular host segment i.e., [–Back] on a Ø segment or [+Nasal] on a spirant. Since the phonetic component cannot readily make these features realized on the specified phonological segment, it shifts the relevant (and contrastive) feature to the next articulatorily viable segment.

6.4. Index of constraints (alphabetical) *[+Son, DOR] Segments containing both [+Son] and [DOR] 77-78 features are banned *COMPLEX No complex syllable margins (*[σCC (onsets 64-67 are simple) *CC]σ (codas are simple)) *MID Assign a violation mark for every peripheral 69-71 vowel that is [–High] and [–Low] *MIDL Alternating forms are lexically linked to a *MID 71 constraint *NUC/ə Assign one violation each time a vowel of low 69-70 sonority ([ə]) appears in a position of prominence *STRUC Assign one violation to each instance of 49-50, 54 phonological structure (no phonological structure) *SUPERHEAVY (*3μ) No trimoraic syllables 75 *V HIATUS Vowels may not surface in adjacency 138 μ *V$V [Low] Root initial short low vowels are banned when 77-78 following another vowel *VD-OBS]σ Voiced obstruents are banned in syllable-final 36 position *μ / –VD_STOP Assign one violation for each mora that 35 dominates a voiceless obstruent *μCod / –VD_STOP Assign one violation for each mora that 36 dominates a voiceless obstruent if that segment is not nucleic *μNuc / –VD_STOP Assign one violation for each mora that 36 dominates a voiceless obstruent if that segment is also nucleic ALIGN (P, L; H, R) Align the left edge of each phonological phrase 53 with the right edge of an H tone

A p p e n d i c e s | 148

ALIGN L, MBOUND R, ω Align the left edge of a bound morpheme with 112 the right edge of a prosodic word ALIGN R, MFREE L, ω Align the right edge of a free morpheme with 112 the left edge of a prosodic word ANCHOR (word, σ R/L) Every segment at the edge of a lexical word has 24 a correspondent at the same edge of a syllable ANCHOR (word, σ R/L)h-aspiré At the boundary between an h-aspiré word and 23-24 the preceding word, every segment at the edge of a lexical word has a correspondent at the same edge of a syllable ANCHOR I-O Assign one violation for each position an 38 element occupies additional to that specified in the input (no spreading) CORONALHOMORGANICITY In ω(...CiCj...), if Ci is coronal, then Cj is 95, 102 (CORHOM) coronal CORONALSECOND (COR2) Preserve the coronal place specification of the 95-98, second consonant in a cluster 102 CRISPEDGE A given type of prosodic category's edges do 24, 26, not allow the spread of phonological features, 75 or the sharing of prosodic features between them CRISPEDGE[PCat] A P(rosodic) Cat(egory) has crisp edges (a 82 given type of prosodic category's edges do not allow the spread of phonological features, or the sharing of prosodic features between them) CRISPEDGEι The prosodic category ι has crisp edges (ι- 58 boundary edges do not allow the spread of phonological features, or the sharing of prosodic structures across them) CRISPEDGEσ A syllable has crisp edges and does not allow 36 the spread of phonological features, or the sharing of prosodic structures between them CV CV sequences are syllabified together 25 DEP I-O Output segments must have input 19, 26, correspondents (no epenthesis) 39-40, 67-68, 75, 116, 138 DEP I-O([Spread Glottis]) Output Spread Glottis features must have input 30 correspondents DEP I-O[F] Output features must have input correspondents 20 (no feature epenthesis) DEP-LINK Output links must have input correspondents 38 DEP-μ-IO Output mora have input correspondents 68-69 ENHANCE-R φ Overtly mark the right edge of a prosodic 30-31 phrase

A p p e n d i c e s | 149

ENHANCE-R φ (Blackfoot) Aspirate any segment aligned with the right 31 edge of a prosodic phrase i EXHAUSTIVITY There can be no prosodic structure C that 48-49, 67 immediately dominates a constituent Cj where Cj < i –1. (i.e., no ω immediately dominates a σ and so on.) Alternate: A constituent of level Cj in the Prosodic Hierarchy may not dominate a constituent of level Cj – (1 + n) (i.e., of more than one level down) (Parse exhaustively all members of a lower constituent into an immediately dominating constituent) FILL Assign one violation to every instance of empty 5-6, 116 prosodic structure Nuc FILL Nucleus positions must be filled with 43 underlying segments FOOT-FORM A blanket constraint that enforces the 51 composition of feet, binarity, rhythmic pattern, stress contours and weight IDENT(Cons) Correspondent segments have identical values 138 for a feature [F] [(consonantal)] (If αRβ and α is [γF], then β is [γF]) IDENT(Continuant) Correspondent segments have identical values 93-98, for feature [F] [(Continuant)]. 102 (If αRβ and α is [γF], then β is [γF]) IDENT(Place) Correspondent segments have identical values 77, 95 for feature [F] [(place)]. (If αRβ and α is [γF], then β is [γF]) LAZY Minimize articulatory effort (i.e. biomechanical 92-93 energy) LENITE Reduce articulatory effort – promote the 92, 95- surfacing of the unmarked 98, 101- 102 LENITET Stop coronal fricatives in a lenition 92 environment Lexical Category Condition Constraints relating syntactic and prosodic 52-53 categories apply to lexical syntactic elements and their projections, but not to functional elements and their projections, or to empty syntactic elements and their projections LX≈PR Let each lexical word be realized as a prosodic 81 word

LX≈PR(MCAT) Let each lexical word of a given morphological 81, 90-91 category be realized as a prosodic word

A p p e n d i c e s | 150

MATCHCLAUSE Each syntactic clause (CP) is interpreted by the 54, 58 phonological component as an intonational phrase (ι)

MATCHPHRASE Each syntactic phrase (XP) is interpreted by the 50, 54, phonological component as a phonological 106 phrase (φ) o MATCHWORD Each syntactic head (X ) is interpreted by the 9, 48-50, phonological component as a prosodic word 106, 110, (ω) 116 MAX I-O Input segments must have output 19, 64, correspondents (no deletion) 66-68, 75-79, 84, 93- 98, 102, 138 MAX I-O(Consonantal) Input Consonantal features must have output 30-31 correspondents MAX I-O(vowel) (MAX V) Input vowels must have output correspondents 30-31, 69-71 MAX I-O[F] Input features must have output correspondents 31 (no feature deletion) MAXIMALITY Every element of S1 has a correspondent in S2 27 MAXμ Input mora must have output correspondents 36 (no mora deletion) MORPHREAL A morpheme must be realized by fulfilling one 91 of the following conditions: a. the output affixed form contains at least one segment not in the unaffixed form, and that segment(s) is coindexed with a segment(s) in the affix's input; b. the output affixed form contains a segment which is coindexed with the affix's input and that segment has a scale (or feature) value contained in the affix's but not in the unaffixed form; c. the output affixed form contains a segment which is coindexed with the affix's input and that segment has a scale value adjacent to that of the affix's input. That value does not occur in the unaffixed form MPARSE Morphemes are parsed into morphological 28 constituents

A p p e n d i c e s | 151

MUTAGREE Mutation triggers and targets must agree in 91, 95- their mutation grade 98, 101- 102 NONRECURSIVITY (NO-REC) Any two prosodic structures that are not 48-49 disjoint in extension are identical in extension: there can be no Ci that dominates Cj where Ci = Cj NUC Syllables must have nuclei 44-45 ONSET *[σ v (syllables must have onsets) 25, 39- 40, 68 i–1 PARSE Phonological constituents of a level C must 67 be parsed directly into a dominating constituent of a level Ci PARSEσ Syllables must be parsed directly into stress 43 feet SPREAD-L-μ Spread leftward any segmental material into all 38 available moraic positions SPREAD-R(V-Place, Sonority) Vocalic place and sonority features are spread 77-78 (HARMONY I-O) rightwards from a vowel to an immediately adjacent consonant SPREAD-R-μ Spread rightward any segmental material into 36, 38 all available moraic positions SSP (SON-SEQ) The sonority of a syllable increases from the 43, 64- beginning of the syllable onwards, and 67, 83-87 decreases from the beginning of the peak onwards SYLL-FORM (SYLL-STRUC) A blanket constraint that enforces the language 51, 68- specific constraints governing syllabic 71, 83-85 templates; requirements for onsets/codas, whether complex clusters are permissible, and demanding that nuclei be filled (among others) WORDSIZE The minimal phonological weight of a prosodic 5, 48-49, word is X where X is equal to a particular 116 phonological value (i.e., a foot, a moraic weight, etc.) WORDSIZE(μμ) A prosodic word must minimally contain two 48-49 mora WRAP XP Each XP is contained in a phonological phrase 8-9 WSP If a syllable is heavy, then it is stressed (a 42 syllables is heavy if and only if it is stressed) Σ-BIN Feet are binary under moraic or syllabic 37, 43 analysis

A p p e n d i c e s | 152

6.5. Index of languages (alphabetical) Arabic 34 Blackfoot 14-17, 28-33, 124 English 2, 7, 8, 14, 41, 44, 45, 46-49, 51, 55-59, 65, 139-142 French 10, 21-27, 58, 62, 66, 124-125 Friulian 34-36, 38 Greek 57 Irish 8, 39, 52, 74, 75, 79, 81, 88-123, 124-125, 138, 143-147 Italian 36-38, 40, 57, 87, 124 Kimatuumbi 53-54 Lardil 43 Middle Irish 114-115 Old Irish 108-109, 113-120, 124, 143 Proto Celtic 114-115 Russian 22, 62-73, 75, 80-87 Scottish Gaelic 98, 99, 100, 143-147 Seri 62, 64, 72-83, 86-87, 121, 124-125 Somali 11-13, 35 Spanish 57 Standard Chinese 39-45 Vulgar Latin 108 Welsh 30