Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
The Spanish feminine el at the syntax-phonology interface
(USC Thesis Other)
The Spanish feminine el at the syntax-phonology interface
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
THE SPANISH FEMININE EL AT THE SYNTAX-PHONOLOGY INTERFACE by Erika Elizabeth Varis A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (HISPANIC LINGUISTICS) August 2012 Copyright 2012 Erika ElizabethVaris ii Dedication For the two loves of my life: Benjamin and Remington Doggett iii Acknowledgements An author never truly writes alone. This dissertation owes to many people who have assisted, encouraged and supported me throughout the dissertation process and graduate research. I would like to take a moment here to acknowledge their support. First and foremost, my academic advisor and committee chair, Rachel Walker, has provided amazing guidance and support in all aspects of my graduate education. I would most especially like to thank her for offering inspiring encouragement whenever I felt I had run into a hiccup in my research. I left every conversation feeling inspired and refreshed with new energy to tackle any challenge that might arise. My thesis committee members, Mario Saltarelli and Louis Goldstein, have guided me in this work and many of the papers and projects leading up to the dissertation, each providing a different perspective to my research. Mario’s wide breadth of expertise in Romance linguistics, including syntax and historical linguistics, was a great assistance. In addition, he was the one who pointed out the basic data pattern to me! Louis kept my research grounded in phonetics and was an incredible support on the experimental side, as well as providing guidance on the broader theoretical picture. I have received additional assistance and support from many professors and scholars both inside and outside USC. I would like to thank Khalil Iskarous, Xiao He, Ed Holsinger, and Melissa Frazier for assistance with statistics and data manipulation in R; Maria Luisa Zubizarreta for discussions on the syntax-phonology interface; Karen Jesney for coining the phrase “correspondence on correspondences” and providing insightful feedback on the theoretical analysis; the audiences of the Linguistic Symposium of iv Romance Linguistics (LSRL) 40 and 42, particularly Jose Ignacio Hualde and Travis Bradley; Edward Flemming for discussion on the expansion of Minimum Distance; Emily Elfner for email correspondence regarding the syntax-prosody mapping interface and Match Theory; and the audience of PhonLunch and the Phonetics Lab groups at USC for support on the phonological analysis and phonetic experiments. For tolerating my countless data questions and providing many, many linguistic judgments both phonological and syntactic, I extend everlasting gratitude to Hector Velásquez, Sergio Robles-Puente, Magdalena Pire-Schmidt, Marianna Nadeu, and Dasha Henderer. Many thanks go to my fellow graduate students and friends in the Linguistics Department at USC. I greatly value the friendship and support from my fellow cohort and friends in the Linguistics Department, without whom graduate school would not have been nearly as much fun, including Katy McKinney-Bock, Mary Byram Washburn, Hector Velasquez, Xiao He, Ben Parrell, Lucy Kim, David Lee, Sarah Ouwayda, Canan Ipek, Mythili Menon, Laura Tejada, Magdalena Pire-Schmidt, and Sergio Robles-Puente. In particular, I thank my good friend Katy for her personal and professional friendship, and her patience with me in being a soundboard and guide throughout innumerable discussions on syntactic theory. Finally, I owe a loving and grateful debt to my family and friends, who have been unfailing in their support and love while I focused my energies on school and research. Above all, my husband Benjamin Doggett has been an incredible support at every stage of graduate school and the dissertation, and without him none of this would be possible. v Table of Contents Dedication ii Acknowledgements iii List of Tables viii List of Figures xi Abstract xiii Chapter One: Introduction 1 1.1 Data background: the feminine el 3 1.2 Theoretical background 6 1.2.1 Optimality Theory 6 1.2.2 X-bar structure and Minimalism 10 1.2.3 Prosodic Phonology 14 1.3 Organization of the dissertation 17 Chapter Two: The Spanish feminine el: from syntax to phonology 18 2.1 The feminine el data 19 2.2 Syntax to prosody 22 2.2.1 Syntax of the Spanish DP 23 2.2.2 Prosodic Phonology Interface 27 2.2.2.1 Clitic adjunction and prosodic recursivity 33 2.2.2.2 Cliticization of Spanish determiners 39 2.2.2.3 Analysis of word adjunction 45 2.2.2.4 Syntactic head-relation in Prosodic Phonology 59 2.2.2.4.1 Spanish conjunction and other function words 61 2.2.2.4.2 Structural alternative with no head-complement relation 68 2.2.2.4.3 Summary of comparison with structural approach 79 2.3 Prosody to phonology 80 2.3.1 Minimum Distance on adjacent vowels 81 2.3.2 Pwd max and prosodic clitic structure 88 2.3.3 MP Faithfulness 92 2.3.4 Stress and positional markedness 100 2.3.5 Conjunction 106 2.3.6 Discussion on Pwd max domain and positional markedness 113 2.4 Historical development of the feminine el 120 2.4.1 Frequency and historical change 129 2.5 Exceptions to the feminine el 134 vi 2.6 Summary of Chapter Two 139 Chapter Three: Phonetic studies: Spanish feminine el and vowel hiatus 141 3.1 Background 141 3.1.1. Spanish vowel hiatus in the phonology 141 3.1.2 The prosodic approach 143 3.1.3 Phonetic reflections of hiatus and prosody 147 3.1.4 Hypotheses 153 3.2 Experiment 1: vowel hiatus and prosodic categories 155 3.2.1 Methods 155 3.2.1.1 Materials 155 3.2.1.2 Procedure 157 3.2.1.3 Measurements 158 3.2.2 Predictions 162 3.2.2.1 Duration 162 3.2.2.2 Formants 163 3.2.3 Results 165 3.2.3.1 Duration Analyses 165 3.2.3.2 Formant Analyses 172 3.2.4 Discussion of Experiment 1 results 179 3.2.5 Summary of Experiment 1 183 3.3 Experiment 2: evidence for prosodic distinction of DNA and DAN 184 3.3.1 Methods 184 3.3.1.1 Materials 184 3.3.1.2 Procedure 185 3.3.1.3 Measurements 186 3.3.2 Predictions 186 3.3.3 Results 190 3.3.3.1 Duration Analysis 190 3.3.3.2 Formant Analysis: V1V2 vs. V2 193 3.3.3.3 Formant Analysis: DNA vs. DAN 195 3.3.4 Discussion of Experiment 2 198 3.3.5 Summary of Experiment 2 200 3.4 Comparison of Experiment 1 and 2 200 3.4.1 Duration comparison of clitic and determiner data 202 3.4.2 Formant comparison of clitic and determiner data 204 3.4.3 Discussion and summary of comparison 208 3.5 Discussion and alternatives 212 3.6 Summary of Chapter Three 214 Chapter Four: Further case studies 216 4.1 Russian pronoun allomorphy 216 4.1.1 Russian allomorphy in a structural account 222 4.2 Catalan 225 vii 4.2.1 Western Catalan 226 4.2.2 Central Catalan 234 4.3 Cibaeño Spanish 241 4.4 More on MP Faith: variations of the feminine el 248 4.4.1 Morphological gender change 249 4.4.2 Synchronic variation: masculine el 258 4.4.2.1 More on AGREE(VALUE) 267 4.5 Summary of Chapter Four 270 Chapter Five: Alternative approaches 272 5.1 Clitic group 274 5.2 Recursive PPhs 275 5.3 Frequency as a mapping factor 283 5.4 Direct reference 288 5.5 Articulatory vowel restrictions 294 5.6 Summary of Chapter Five 296 Chapter Six: Concluding remarks 298 6.1 Contributions 298 6.2 Further research 300 6.3 Summary 303 References 304 Appendices Appendix A: Control and test items for Experiment 1 325 Appendix B: Control and test items for Experiment 2 327 Appendix C: Fixed effects results formant results 328 viii List of Tables Table 1. Ranking: C1 >> C2 >> C3 8 Table 2. Ranking: C2 >> C1 >> C3 9 Table 3. MATCH-PHRASE >> NONRECURSIVITY 32 Table 4. ALIGN-R(XP,!), ALIGN-L(XP,!) >> NONRECURSIVITY 32 Table 5. Constraint violations of HEADEDNESS, PARSE-INTO-", MATCH-WORD 50 Table 6. Constraint violations of HEADEDNESS, MATCH-WORD, PARSE-", NONRECURSIVITY" 51 Table 7. Factorial Typology 52 Table 8. PARSE FNC -INTO-" HCOMP >> NONREC" >> PARSE-" 54 Table 9. LINCORR >> PARSE FNC -INTO-" HCOMP 56 Table 10. BIN-MAX(") and MATCH-" LEX 58 Table 11. Lexically indexed PARSE-" 65 Table 12. Lexically indexed PARSE FNC(D) -INTO-" HCOMP 67 Table 13. MATCH(LEXP, PWD MAX ) differentiating DNA and DAN 71 Table 14. VP short: [ TP va t [ VP t [ PP a [ DP la [ FP tienda]]]]] ‘he/she goes to the store’ 72 Table 15. VP long: [ TP compró t [ VP t[ DP el [ FP sombrero [ AP grande]]][ AP rápidamente]]] ‘he/she bought the big hat quickly’ 73 Table 16. Lexically indexed PARSE-" D,CONJ 75 Table 17. Lexically indexed MATCH(LEXP, PWD MAX ) DP 76 Table 18. Failure of MATCH(LEXP, PWD MAX ) DP 79 Table 19. Activity of MINDISTVV 97 ix Table 20. No masculine la M 98 Table 21. Activity of Pwd max domain 99 Table 22. Failure of ONSET/#! 102 Table 23. Activity of positionally marked MINDISTV'V 104 Table 24. MINDISTVV with conjunction 108 Table 25. MINDISTV'V(PWD MAX )=F1/F2:1 >> IDENT-MP >> MINDISTVV(PWD MAX )=F1/F2:1 >> IDENT-IO[HI]>> MINDISTVV=F1/F2:1 110 Table 26. Conjunction and stress 112 Table 27. MINDISTV'V evaluation 118 Table 28. Stage I: phonological feminine el with all vowel-initial words 125 Table 29. Stage II: feminine el with a-initial words (including adjectives) 127 Table 30. Stage III: feminine el with stressed a-initial words 128 Table 31. Stage IV: Modern Spanish feminine el with stressed a-initial nouns only 129 Table 32. Summary of all intervals of F2 vowel quality effects for IP and PPh: V1 173 Table 33. Summary of F1 and F2 effects for IP and PPh: V2 175 Table 34. Summary of F1 and F2 effects for Pwd and clitic V1V2 177 Table 35. Summary of F1 and F2 effects for DNA/DAN: V1V2 vs. V2 193 Table 36. Summary of F1 and F2 effects for DNA/DAN: V1V2 196 Table 37. Summary of F1 and F2 effects for clitic-verb, DN and DA 205 Table 38. PARSE FNC >> NONRECURSIVITY" 219 Table 39. PARSE FNC -INTO-" HCOMP >> NONRECURSIVITY" >> PARSE-" 220 x Table 40. Full Russian allomorphy 221 Table 41. Russian allomorphy with MATCH(LEXP, PWD MAX ) 223 Table 42. Failure of MATCH(LEXP, PWD MAX ) for Russian elided nouns 225 Table 43. W. Catalan MINDISTVV and vowel deletion 231 Table 44. W. Catalan and stress 232 Table 45. Central Catalan MINDISTV'V 237 Table 46. Cibaeño coda glide-blocking 245 Table 47. Failure of Faith Lex 246 Table 48. DEB ranking of Cibaeño Spanish 247 Table 49. Morphological gender change in arte 254 Table 50. el arte and el ave: lexical-indexation and frequency-indexation for MM 257 Table 51. Synchronic variation and AGREE(VALUE) 264 Table 52. Paradigm Uniformity and prenominal adjectives 266 Table 53. AGREE(XP HEAD ) and AGREE (VALUE) 268 Table 54. MINDISTV'V(PWD MAX )=F1/F2:1, IDENT-MP, IDENT-MM, PU(D) >> AGREE(XP HEAD ) >> AGREE (VALUE) 270 Table 55. Failure of articulatory *V x .V y 295 xi List of Figures Figure A. Wave and spectrogram for the segment la heriste ‘you hurt her’ 161 Figure B. Durations of control V1 at the clitic-Pwd boundary, IP boundary, PPh boundary and Pwd-Pwd boundary. 167 Figure C. Durations of control V2 at the IP, PPh and Pwd boundaries 169 Figure D. V2 durations for the IP and PPh boundaries, comparing test and control items for each vowel quality. 170 Figure E. Control V1 and V1V2 test sequence durations for the Pwd and clitic boundaries. 172 Figure F. Vowel charts for the 1 st , 2 nd , 3 rd and 4 th time intervals of V1 at IP and PPh boundaries 174 Figure G. Vowel charts for the 1 st , 2 nd , 3 rd and 4 th time intervals of V2 at IP and PPh boundaries. 176 Figure H. Vowel charts for Pwd and clitic vowels across the 1 st , 2 nd , 3 rd and 4 th time intervals 178 Figure I. Wave form and spectrogram for the phrase la elitista ‘the elitist [feminine]’, with the test vowel sequence ae broken up into four equal intervals. 187 Figure J. Wave form and spectrogram for the phrase el elitista ‘the elitist [masculine]’, with the control vowel e broken up into four equal intervals. 188 Figure K. Durations of V1V2 test items compared to control V2 items, for DAN and DNA sequences. 192 Figure L. Durations of DNA and DAN word sequences (both control and test). 192 Figure M. Vowel charts comparing test V1V2 sequences with control V2 vowels for the 1 st and 2 nd intervals of DAN and DNA items. 195 Figure N. Vowel charts for V1V2 test sequences of DAN and DNA items at the 1 st , 2 nd , 3 rd and 4 th intervals. 198 xii Figure O. Durations of V1V2 for clitic, DA and DN sequences. 204 Figure P. Vowel charts for clitic, DA and DN V1V2 sequences at the 1 st , 2 nd , 3 rd and 4 th intervals. 207 xiii Abstract The main theoretical goal of this dissertation is the investigation of the syntax- phonology interface through probing syntax-sensitive phonological phenomena, exemplified by a comprehensive analysis of the so-called Spanish “feminine el”, a morpho-phonological determiner alternation that is triggered by certain phonological and syntactic contexts. Specific areas of the syntax-phonology interface that I examine are syntactic triggers to prosodic clitic organization and phonetic reflections of those prosodic structures. I argue that the feminine el provides support for the inclusion of the syntactic head-complement relation in the Prosodic Phonology mapping, and produces recursive Prosodic Word (Pwd) cliticization structures. In particular, it accrues evidence for the maximal Pwd domain proposed by Itô and Mester (2006, 2007, 2008, 2010). This dissertation provides new phonetic evidence for distinguishing between maximal and minimal Pwd boundaries in different cliticization structures. Additional theoretical proposals of the dissertation include a scale of phonological contrast-driven hiatus restrictions, and the development of a morphology-phonology correspondence account of apparent mismatches in gender realization at the morphology- phonology interface, in order to address the specific feminine or masculine morphological interpretation of the Spanish article in question. The proposed scale of hiatus restrictions is a novel application of the Minimum Distance (Flemming 2004) approach to perceptual contrast, utilizing perceptual contrast in segment transitions to produce a relative scale of adjacent VV restrictions. The morphology-phonology correspondence account provided builds on previous work in the morphology-phonology xiv interface (Walker & Feng 2004, Wolf 2008) to evolve the family of IDENTITY constraints governing correspondence between morphological and phonological forms in the input and output. In examining these areas, this dissertation is the first to provide a full theoretical analysis of the Spanish feminine el phenomenon, accounting for phonological, morphological and syntactic effects. Further case studies in Spanish and Romance dialects also demonstrate the effects of the prosodic cliticization structures, hiatus restrictions, and morpho-phonological correspondence proposed. Among the broad theoretical contributions of this work, this dissertation brings new insight and evidence to the distinctions between different prosodic clitic attachments and the nature of the syntactic information on which they may rely, further developing the relationship between syntax and phonology. It additionally expands the understanding of phonological vowel hiatus by building on perceptual contrast considerations, opening a new perspective on segment sequencing. 1 Chapter One: Introduction The main goal of this dissertation is the investigation of the syntax-phonology interface. The specific aspects of this interface I examine are syntactic triggers to phonological phenomena and phonetic reflections of prosodic organization. Additional areas of investigation include contrast-driven hiatus restrictions in functionally-driven phonology and apparent gender agreement mismatches in the morphology-phonology interface. In exploring these theoretical areas, I provide a comprehensive analysis of the Spanish feminine el phenomenon, where a determiner form el—homophonous with the normative masculine determiner el—alternates with the regular feminine la. While the specific morphological interpretation is not necessarily feminine for all variations of the pattern, I use the term “feminine” el for reference to the phenomenon, in accordance with previous research (Harris 1987, Janda & Varela García 1991, Gonazález 2003, Eddington & Hualde 2008). The morpho-phonological determiner alternation is triggered by certain phonological and syntactic contexts, providing an ideal case study for the examination of phonological interfaces in linguistic theory. While the morphological and phonological aspects of the feminine el pattern are well known in the Spanish linguistics literature, this is the first theoretical analysis accounting for syntactic influence and bringing all properties of the pattern together in a complete analysis. As a phonological phenomenon that is crucially sensitive to syntactic structure, this topic provides an excellent case study for exploration of the syntax-phonology interface. The area of the interface that the feminine el bears on specifically is that of the 2 Prosodic Word (Pwd) and clitic structures. In the analysis pursued here, I argue that the feminine el provides support for recursive Pwds in prosodic cliticization structures, in particular the maximal Pwd domain proposed by Itô and Mester (2006). In addition, I propose that prosodic cliticization is dependent upon syntactic structure such that the head-complement relation in syntax invokes prosodic attachment at the maximal Pwd in prosody. This syntax-prosody mapping allows the realization of different prosodic clitic structures, depending on the syntactic input. In support of the syntax-phonology interface analysis of the Spanish feminine el, I also provide new phonetic data pertaining to the prosodic structures of Spanish determiner-lexical word and clitic-host sequences. Two acoustic experiments give evidence of varying phonetic production of vowel hiatus at established prosodic boundaries in Spanish, and also evidence for the distinct prosodic clitic structures proposed for determiner sequences with the feminine el. In the phonological analysis of the feminine el, I propose a novel application of the Minimum Distance approach (Flemming 1995, 2004) to perceptual contrast as applied to adjacent segments, accounting for phenomena restricting vowels in hiatus. I argue that a family of Minimum Distance constraints operates over adjacent vowels, requiring varying degrees of vocalic contrast. These constraints produce a gradient sensitivity to the quality of vowels in hiatus, allowing a uniform analysis of not only the feminine el pattern, but also patterns of hiatus avoidance in other areas of Spanish and Romance linguistics. 3 In addition, a correspondence approach to the morphology-phonology interface is developed in accounting for the allomorphic repair strategy chosen in the feminine el, building upon previous work in this interface area (Walker & Feng 2004, Wolf 2008). Specifically, the analysis utilizes identity correspondence between existing morpho- phonological (MP) relationships in the input, or underlying form, and the resulting MP relationships of the output or surface form. Further applications of morpho-phonological correspondence address the historical development of the language as well as synchronic variation displayed in the syntactic and morphological interpretation of determiners in Spanish. Overall, this dissertation develops several areas of linguistic interface in the analysis of the Spanish feminine el: prosodic cliticization of the syntax-phonology interface, hiatus reflections at the phonetic-prosodic interface, phonological segment adjacency as functionally driven requirements from the phonetics-phonology interface, and identity correspondence of the morpho-phonological interface. 1.1 Data background: the feminine el The Spanish feminine el is a pattern of determiner alternation, originating in a historical pattern sensitive to vowel hiatus with the definite articles as they evolved from Latin demonstratives 1 . The triggering mechanism for the alternation has gradually grown more complex over time. The present day widely recognized pattern involves the use of the normatively masculine article form el, which ends in a consonant, instead of the 1 More on the historical development of the pattern can be found in §2.4. 4 regular feminine form la, which ends in a vowel. (Spanish has obligatory gender agreement between the determiner and noun.) It is called the “feminine el” since it involves the non-normative use of el for feminine nouns. The data is presented in (1-1). The masculine determiner el is shown in (1-1a), and the feminine determiner la is shown in (1-1b). Vowel hiatus is tolerated with the feminine determiner la and all five vowels of the Spanish vowel system (a, e, i, o, u) when unstressed (1-1c-g), as well as stressed vowels other than a (1-1h-k) 2 . (Stress is marked with an accent. In the data presented below, the subscripts are as follows: D=determiner, N=noun, M=masculine, F=feminine.) (1-1) a. el DM cuchillo NM ‘the knife [m]’ b. la DF torre NF ‘the tower [f]’ c. la DF amíga NF ‘the friend [f]’ d. la DF edád NF ‘the age [f]’ e. la DF octáva NF ‘the octave [f]’ f. la DF utópia NF ‘the utopia [f]’ g. la DF idéa NF ‘the idea [f]’ h. la DF éra NF ‘the era [f]’ i. la DF hóra NF 3 ‘the hour [f]’ j. la DF úlcera NF ‘the ulcer [f]’ 2 The Spanish vowel inventory is the same in stressed and unstressed contexts. 3 In Spanish, orthographic h is silent. Thus, this sequence contains the phonological hiatus [a-o]. 5 k. la DF índole NF ‘the (emotional) character [f]’ The feminine el surfaces only when the use of the normative la would produce a sequence of a followed by stressed a (1-2a-b). The determiner in this case is still considered morphologically feminine (Harris 1987, Salvá 1988, Wolf 2008), since the DP continues to demonstrate feminine gender agreement as a whole, such as when adjectives are included (1-2c-d). (1-2) a. el água NF ‘the water [f]’ b. el ála NF ‘the wing [f]’ c. el DF água NF cristalína AF ‘the crystalline [f] water [f]’ d. el DF ála NF lárga AF ‘the long [f] wing [f]’ As shown, the phonological context for the feminine el is restricted to that where use of the feminine la would produce the specific vowel hiatus a-á. The morpho- phonological aspect of the pattern is reflected in the use of the specific allomorph el as a hiatus repair, which corresponds to an existing determiner form in the language (the masculine). A syntactic complication arises, however, in sequencing options within the Spanish determiner phrase (DP). In DPs that contain a prenominal adjective—even one that begins with stressed a—hiatus is tolerated with the regular feminine la determiner (1-3). 6 (1-3) a. la DF ámplia AF água NF ‘the ample [f] water [f]’ b. la DF ágil AF ála NF ‘the agile [f] wing [f]’ The feminine el phenomenon is restricted not only to a specific phonological context, but also to the syntactic structure represented by the determiner-noun (DN) sequence, as opposed to the determiner-adjective (DA) sequence. Explaining this syntactic trigger for a pattern of phonological hiatus avoidance is one of the primary aims of this dissertation. 1.2 Theoretical background This dissertation is framed within current theoretical models of phonology, syntax, and the interface between the two. The phonological analyses are couched in Optimality Theory (Prince & Smolensky 1993/2004), the syntactic discussion stems from perspectives of X-bar structure (Chomsky 1970, Jackendoff 1977) and Minimalism (Chomsky 1993, 1995), and work on the interface is developed within the framework of Prosodic Phonology (Selkirk 1981; Nespor & Vogel 1986; Inkelas & Zec 1990) and Match Theory (Selkirk 2009, 2011; Elfner 2011). 1.2.1 Optimality Theory Optimality Theory (OT) is a constraint-based framework of grammar in which the optimal output form is chosen under parallel evaluation of candidate outputs in relation to a hierarchy of violable constraints. Some basic background for the theory is provided 7 here, but for a fuller introduction to the theory, the reader is referred to Prince and Smolensky (1993/2004) and McCarthy (2002, 2004). The constraint set is universal cross-linguistically, and different language grammars emerge from different rankings of the constraints. The model includes mechanisms for the generation of output candidates (GEN), and the evaluation of candidates (EVAL), under a given ranking of the set of universal constraints (CON). GEN may produce an infinite number of possible output candidates for a given input, or underlying form. CON includes constraints of two categories: markedness and faithfulness. Markedness constraints evaluate the well- formedness of a candidate, while faithfulness constraints evaluate how closely the candidate corresponds to the input form. The different ranking of these constraint types produce different language patterns. EVAL is the mechanism by which the optimal candidate is chosen, based on the constraint ranking. EVAL eliminates candidates that violate higher-ranking constraints, choosing the optimal candidate, or actual output form, as the one that remains when all other candidates are ruled out. As an example, suppose we have the constraints C1, C2 and C3, ranked in that order. Given an input, GEN produces candidates (a), (b), (c), and (d) (among others). Different candidates may violate different constraints once, multiple times, or not at all. In the tableau shown in Table 1, an asterisk (*) denotes each constraint violation for the candidates. If a candidate incurs more violations of a higher-ranked constraint than another competing candidate, it is ruled out, denoted by an exclamation mark (!). The remaining competing candidates are compared at the next level of the ranking, and the one which incurs more violations than the others is ruled out, etc. The optimal candidate 8 selected by the constraint ranking is marked with an arrow ($). By convention, once a candidate is ruled out by a given constraint, its row is shaded for the subsequent lower- ranked constraints to indicate that it is no longer in competition at that ranking level. Table 1. Ranking: C1 >> C2 >> C3 /input/ C1 C2 C3 Candidate a **! Candidate b *! $ Candidate c * Candidate d * *! In Table 1, candidate (b) is ruled out because it violates the highest ranked constraint C1, while competing candidates obey it, even though (b) does not incur violations of the other two constraints C2 and C3. Candidates (a), (c) and (d) all violate C2, but candidate (a) incurs more violations than either (c) or (d), and so it is ruled out. Candidate (d) is ruled out due to its violation of low-ranking C3, which candidate (c) does not violate at all. If the constraints are re-ranked, as they may be in the grammar for another language, a different optimal candidate may emerge. 9 Table 2. Ranking: C2 >> C1 >> C3 /input/ C2 C1 C3 Candidate a *!* $Candidate b * Candidate c *! Candidate d *! * In Table 2, candidate (b) is selected as optimal, because it is the only candidate that does not violate top-ranked C2. All other candidates are ruled out by the constraint C2. The above outline provides a basic view of the mechanics of an OT grammar. In addition, there have been many expansions to the classic OT framework based on constraint families and types. One of the theoretical developments I pursue in this dissertation is the concept of Minimum Distance in Dispersion Theory (Flemming 1995, 2004). A brief overview of this concept and how it relates to the following work is provided here, with a more complete discussion in Chapter 2. Dispersion Theory was first presented in Flemming (1995) as a framework for evaluating phonological contrast, and has since been further developed in the literature with focus on vowel inventories (Flemming 2004), vowel reduction processes (Padgett 2004, Flemming 2005), vowel harmony (Ní Chiosáin & Padgett 2001), palatalization and nasalization contrasts (Padgett 1997, 2001), diphthongs or intravocalic sequences (Casali 1998, Sands 2004), and sound change (Sanders 2003). The area of Dispersion Theory I focus on is MINIMUM DISTANCE, the set of constraints designed to maximize the 10 perceptual distinctiveness of contrasts along a particular dimension. The farther apart a set of contrasting segments are along a dimensional scale, the easier they are to perceive as contrasting, while the closer they are along the dimension (the more similar they are), the more perceptually confusable they will be. In analyzing the phonological aspect of the feminine el phenomenon in §2.3.2, I apply these Minimum Distance constraints to adjacent vowel segments, yielding a family of MINDISTVV constraints sensitive to varying degrees of contrast in vowel hiatus. The OT framework is primarily used as an architecture for the phonological component of the analysis. For the syntactic component, discussions are presented using concepts from X-bar structure and Minimalism. 1.2.2 X-bar structure and Minimalism The formalism I adopt for the syntactic structures discussed in this dissertation is based primarily on X-bar structure (Chomsky 1970, Jackendoff 1977), with reference to certain ideas in the Minimalist program of syntax (Chomsky 1993, 1995). For a detailed introduction to X-bar, see Carnie (2002). Likewise, the reader is recommended to Chomsky (1995, 2000, 2001) for a complete overview of Minimalism. The basic X-bar outline of syntactic structures is presented in (1-4) below, including levels of representation at the minimal X 0 level (the terminal nodes), intermediary levels (notated with X') and the maximal XP or phrasal level. These notions of different levels of a structure re-appear in the theory of prosodic phonology pursued in this work, discussed in the next section. 11 (1-4) XP specifier X' X complement Also available to the structure are points of attachment for additional material, including complements, specifiers, and adjuncts. A complement is attached as a sister to X 0 . Functionally, it is considered crucial to the meaning of the phrase. For example, the English verb push requires a DP complement, a direct object. (1-5) *He pushed. He pushed [ DP the cart] comp . A specifier attaches to the XP structure at the XP level, as a sister to X' (see 6 above). An adjunct may attach anywhere above the X 0 level, 4 but is distinguished from complements and specifiers in that it does not affect the level of the X-bar node to which it attaches. Thus, adjunction to an XP node produces another XP, while adjunction to an X' node produces another X'. It is usually also considered non-essential material. Continuing with the example push, the addition of the adverb rapidly provides an adjunct to the VP structure, as illustrated in (1-6). 4 Note that the precise nature of adjunction and where adjuncts attach within a structure is still debated. For the purposes of my analysis, the only crucial definition of adjuncts is that it does not change the level of the X-bar node. This aspect of adjunction is not controversial. 12 (1-6) The X-bar structure is representative at different levels of the overall grammar. An underlying or base-generated D-structure may have one X-bar tree, and different transformational or movement operations result in another X-bar tree in S-structure. (1-7) D-structure Transformational rules/Movement S-structure In the more recent Minimalist Program of syntactic theory, certain proposals of X-bar are simplified, and the relationships between syntactic structure, semantic meaning, and phonological form are further developed. Rather than providing a full overview of the underlying philosophy of Minimalist syntax, I will simply outline some of the elements adopted in this dissertation. Where terminology of X-bar and Minimalism 13 are not widely disparate or contradictory—such as referring to “movement” or “Merge”—I use the X-bar terms. One aspect of the Minimalist program discussed in Chapter 4 of this dissertation involves the nature of grammatical agreement. Following Bhatt (2005) I assume a Minimalist definition of syntactic agreement, such as that between Spanish determiners (probes) and nouns (goals). (1-8) Agree: the probe % with an unvalued uninterpretable feature [f] searches for a goal & with a matching valued uninterpretable feature [F] within its c-command domain. The probe then takes the feature value of the goal. In Chapter 4, syntactic Agree is incorporated into a syntactic-phonological constraint on the manifestation of feature values, available to an OT grammar for phonological analysis of a synchronic variation of the feminine el. Another important element of the Minimalist Program is the overall architecture of the grammar, particularly the relationship between syntactic structure and the semantic and phonological components. The model of the grammar in (1-7) for the most part remains, with the addition of the phonological (PF) and semantic (LF) components branching from post-movement S-structure. 14 (1 -9) D-structure (transformational operations) S-structure Phonetic Form (PF) Logical Form (LF) One of the primary concerns of this dissertation is the mapping on the left side of the Y-grammar model in (1-9), from syntactic S-structure to the PF component, where prosodic structure, phonological constraints, and phonetic representations are assumed to be located. In the analysis of the Spanish feminine el proposed in Chapter 2, I assume that as presented in the above model, the syntactic representation at S-structure, including the key head-complement relation between lexical items, forms the input for the mapping to prosodic structure (a phonological component). 1.2.3 Prosodic phonology In mapping the syntactic S-structure to the phonological PF component, I assume a framework of prosodic phonology, reviewed further in §2.2.2. Prosodic Phonology (Selkirk 1981; Nespor & Vogel 1986; Inkelas & Zec 1990; Itô & Mester 1992, 2007; Truckenbrodt 1995, 1999; and many others) is a framework of interface between syntax and phonology, which includes the prosodic hierarchy of units in (1-10) as well as constraints on the syntactic-prosodic mapping and prosodic well-formedness. These 15 prosodic categories are visible in the phonological component, and may be referenced by phonological constraints. (1-10) The Prosodic Hierarchy ' utterance ( intonational phrase (IP) ! phonological phrase (PPh) " prosodic word (Pwd) F foot # syllable The mapping theory developed here extends recent approaches in Match Theory (Selkirk 2009, 2011; Elfner 2011) and PARSE constraint driven mapping (Itô & Mester 2008), which favor recursive prosodic structure. In the analysis of the Spanish feminine el structures, presented in §2.2.2.4, a PARSE constraint is proposed for function words to drive head-complement sensitive adjunction. This allows two different prosodic clitic structures to surface in response to differing syntactic relations between Spanish determiner-noun (DN) and determiner-adjective (DA) sequences. The structures proposed for Spanish DN and DA follow from the possibility of prosodic adjunction at different levels of a prosodic category, in much the same way that different syntactic structures are formed by attaching material at the X 0 , X' or XP levels of the X-bar schema. For the prosodic hierarchy, Itô and Mester (2006, 2007, 2008, 2010) 16 borrow Grimshaw’s (1991) extended projection terminology to dub the highest layer of a given category as “maximal” and the lowest as “minimal”. They propose that phonological constraints may reference not only the prosodic category but maximal and minimal subcategories from recursive prosodic structure. The attachment options produce the possibility of two key clitic structures—the affixal clitic and free clitic—shown in (1-13) below, and argued in §2.2.2.2 to be representative of the Spanish DN and DA sequences. In the phonological analysis, presented in §2.3, constraints parametrized to the Pwd max (" max ) domain allow differentiation between the DN sequence, included within a Pwd max , and the DA sequence, which crosses a Pwd max boundary. (1-11) a. affixal clitic b. free clitic " max ! " min " max Clitic Host Clitic Host [" max Cl [" min H]] [! Cl [" max H]] D N D A Note the difference in the prosodic boundary intervening between the clitic and its host. In the free clitic structure, the intervening boundary between the clitic and host is 17 Pwd max (1-11b), while in the affixal clitic structure, the intervening boundary between the clitic and host is lower-order, corresponding to a minimal or non-maximal level of the Pwd prosodic category (1-11a). Evidence for this distinction in the boundary intervening between DA and DN is given in the phonetic studies of Chapter 3, which finds support for a weaker boundary at DN junctures. 1.3 Organization of the dissertation The previous sections have reviewed the key theoretical concepts touched on by the current analysis of the Spanish feminine el. The remainder of the dissertation is organized as follows. Chapter 2 provides a complete analysis of the feminine el, from the syntactic structure of the Spanish DP, through the interface mapping to prosodic structure, and the subsequent phonological analysis referencing the prosodic constituents. Chapter 3 provides phonetic evidence in support of the theoretical analysis, in particular the prosodic distinction between determiner-noun and determiner-adjective sequences in Spanish. Chapter 4 discusses additional patterns elsewhere in Spanish, Romance and non- Romance languages that mirror different elements of the theoretical analysis, including syntactically-dependent cliticization, segment adjacency effects in vowel hiatus and consonant onsets, specification to the prosodic Pwd max domain, morpho-phonological correspondence and syntactic agreement. Chapter 5 addresses some alternative proposals for the syntax-prosody mapping and prosodic structures proposed for the feminine el. Chapter 6 concludes with a review of the contributions of the dissertation and avenues for future research. 18 Chapter Two: The Spanish feminine el: from syntax to phonology In this chapter, a comprehensive analysis of the Spanish feminine el phenomenon is offered, providing evidence for recursive prosodic word structure as well as phonological sensitivity to the perceptual contrast between adjacent vowels. I argue that sensitivity to word order in the phonological phenomenon is the result of different syntactic relations that affect prosodic clitic structure. In turn, the phonology accesses the prosodic distinction via constraint specification to the prosodic domain in question. In the syntactic analysis, the syntactic head-complement relation between determiners and nouns is proposed to be crucially different from the relationship between determiners and prenominal adjectives. The two different relations between the determiner and the following word are incorporated in the syntactic structure inputs for prosodic constituency formation, in which the syntactic difference is reflected via different prosodic clitic attachment for nouns versus adjectives. Determiners attaching to nouns form a recursive prosodic word structure, with minimal and maximal Pwd level labeling following Itô and Mester (2006, 2007, 2008, 2010). Determiners attaching to adjectives form a non-recursive phrasal clitic structure. The syntax-prosody mapping is addressed within the recent framework of Match Theory (Selkirk 2009, 2011; Elfner 2011), with the addition of syntactic relation reference in the parsing of function words within a maximal Pwd. The phonological analysis addressing the segment adjacency issue of vowel hiatus continues reference to Pwd max as a domain of constraint application, to explain how the morpho-phonological alternation of the Spanish determiner interacts with the proposed prosodic structures. 19 Section §2.1 reviews the Spanish data. In section §2.2, I examine the possible prosodic structures of the sequences in question in the context of current theoretical advances in the Prosodic Phonology interface, including an edge-based alignment approach (Chen 1985; Selkirk 1984, 1996; Selkirk & Tateishi 1988, 1991; Selkirk & Shen 1990; Truckenbrodt 1995, 1999; Itô & Mester 2008; and others) and Match Theory (Selkirk 2009, 2011; Elfner 2011). I show that the Spanish data is best accounted for in a structure with prosodic adjunction at the word level—inducing recursive prosodic word structure—and at the phrasal level. In §2.3, the proposed prosodic structures are applied to a phonological analysis of hiatus resolution, utilizing considerations of perceptual contrast on segment adjacency, stress, and morpho-phonological faithfulness correspondence. Section §2.4 presents the historical development of the feminine el, with discussion on the compositionality of the various phonological considerations previously illustrated. Section §2.5 addresses the lexical exceptions to the feminine el pattern, and §2.6 summarizes the main points of the prosodic analysis and how it affects phonological segment adjacency. 2.1 The feminine el data As introduced in chapter 1, the Spanish feminine el pattern is a phonologically conditioned case of allomorphy, in which phonological considerations of vowel hiatus and stress affect the choice of determiner allomorph. Word order plays a crucial role in the pattern, providing evidence for different prosodic parsing of the sequences in the syntax-phonology interface. 20 At the phonological level, Spanish tolerates most inter-syllabic and inter-word vowel hiatus, for example (2-1). (2-1) Across words Across syllables mi amíg[a á]lta ‘my tall[f] friend[f]’ tr[á.e] ‘he/she brings’ However, hiatus is limited within the DP through determiner allomorphy, the so- called “feminine el” (Harris 1987, González 2003, Wolf 2008). The alternation involves resolving hiatus of an a-á sequence specifically at a border between the definite determiner 5 and a noun. The masculine determiner is el (2-2a), and the usual feminine determiner is la (2-2b). (2-2) a. el D cuchillo N ‘the knife[m]’ b. la D torre N ‘the tower[f]’ As shown in section 1.1, hiatus is tolerated with the determiner la and all five unstressed vowels, as well as stressed vowels other than a. But, when the determiner is followed by a stressed a-initial noun, the form used is the consonant-final el, usually 5 Prescriptively, the indefinite article un/una ‘a’ is not required to participate in the alternation, but most speakers extend the determiner alternation to the indefinite as well. All other extensions of the masculine form to prenominal determiners, quantifiers and adjectives are not accepted (Diccionario Panhispánico de Dudas, Real Academia Española, online edition October 2005). For further examination of this extension, as well as non-prescriptive variation involving other prenominal determiners and adjectives, see §4.5. 21 reserved for the masculine determiner, thereby avoiding vowel hiatus (2-3). Postnominal adjectives are included to demonstrate the feminine gender of the phrase. (2-3) e[l á]gua N cristalína A , *l[a á]gua N cristalína A ‘the crystalline[f] water[f]’ e[l á]la N lárga A , *l[a á]la N lárga A ‘the long[f] wing[f]’ e[l á]da N buéna A , *l[a á]da N buéna A ‘the good[f] fairy[f]’ e[l á]guila N blánca A , *l[a á]guila N blánca A ‘the white[f] eagle[f]’ Interestingly, when the determiner-noun-adjective sequence is reordered to use a prenominal adjective that begins with stressed a, the determiner switch no longer occurs, and hiatus is tolerated (2-4). (2-4) l[a á]mplia A água N , *e[l á]mplia A água N ‘the ample[f] water[f]’ l[a á]gil A ála N , *e[l á]gil A ála N ‘the agile[f] wing[f]’ l[a á]gria A háda N , *e[l á]gria A háda N ‘the bitter[f] fairy[f]’ l[a á]lta A águila N , *e[l á]lta A águila N ‘the high[f] eagle[f]’ In summary, identical adjacent stressed vowels are restricted at the determiner noun (DN) boundary, while determiner adjective (DA) ordering blocks repair. The restrictions on adjacent vowels and stress clearly belong to the phonological sphere of 22 analysis, but the word order condition appears to be syntactic in nature. This sensitivity to syntax has been identified in previous research on the feminine el, but has not been the subject of a formal analysis (Harris 1987, Penny 2002, González 2003). The question remains: how do we account for this syntactic sensitivity in a phonological phenomenon? 2.2 Syntax to prosody I propose that the feminine el’s phonological sensitivity to word order results from underlying differences in the syntactic relations of the two word sequences, which in turn influence the formation of prosodic structure into phrasal and word level clitics. The word sequences determiner-noun (DN) and determiner-adjective (DA) encode different syntactic relations between the determiner and following lexical item. In prosodic structure, the determiner (D) is a prosodic clitic that attaches differently to the following host noun (N) or adjective (A) depending on its relationship to the host word. Specifically, N is the head of the complement to D, while A is not. The head-complement relation between D and N gives rise to a closer prosodic attachment, with D incorporating into the prosodic structure at a lower level of the prosodic hierarchy than when it attaches to A. The different prosodic structures built from the syntax are available to the phonology as triggering or blocking contexts for segment adjacency effects. In examining this flow of information from syntactic structure through prosodic structure and phonology, first we must review the syntax of the Spanish determiner phrase (DP) and the interface framework of Prosodic Phonology. 23 2.2.1 Syntax of the Spanish DP Key to the effect of word order on the feminine el is recognizing the syntactic difference between nouns and adjectives in relationship to the determiner. Since Abney 1987, a larger Determiner Phrase (DP) structure—as opposed to a Noun Phrase (NP) structure—is the most widely accepted analysis of the DN unit. The DP has as its head the determiner (D) and an NP as its complement. Adjectives may be postnominal or prenominal in Spanish, with different semantic interpretations for each position. One of the primary differences is that postnominal adjectives are restrictive, while prenominal adjectives are nonrestrictive. The examples in (2-5) demonstrate the different interpretations (from Demonte 1999). (2-5) a. Dame la rosa delicada. Give-me the rose delicate “Give me the rose (that is) delicate.” b. Dame la delicada rosa. Give-me the delicate rose “Give me the delicate rose.” In (2-5a), the adjective delicada identifies out of a group of roses, the one that is delicate. (2-5b) on the other hand, names the concept ‘rose’ as delicate. The prenominal nonrestrictive meaning is often associated with a poetic reading, as for epithets, or 24 speaker-oriented interpretations, as in the phrase un viejo amigo ‘an old friend (whom the speaker has known for a long time)’. Because of the differing semantic interpretations, these adjectives are often proposed to have different syntactic positions in relation to the determiner and noun (Bernstein 1991, 1993; Martín 1995; Demonte 1999, 2005; Zagona 2002). 6 Demonte (1999a, 2008) presents prenominal adjectives as generated in functional projections above the NP, but postnominal adjectives adjoined lower within the NP 7 . Head movement of N or NP movement to a higher functional position in the DP have been proposed to account for the surface word ordering in Romance and other languages with postnominal adjectives. Arguments for head movement of N to D are presented in Longobardi (1994, 1995) and Bernstein (1991); head movement of N to an intermediary functional projection such as Num(ber)P in Ritter (1991), Valois (1991), Picallo (1991), Bernstein (1993, 1999, 2003), and Cinque (1994); and proposals of NP movement in Cinque (2003, 2004, 2010) and Alexiadou (2003). In addition, a head-parameter approach that involves no movement is proposed in Saltarelli (2001). 6 Though see Cinque (1994, 2003, 2004, 2010) for uniform syntactic analyses in which both prenominal and postnominal adjectives are generated as specifiers to a series of functional projections between the D and NP. Bosque & Picallo (1996) follow a similar analysis, with the additional proposal that a “thematic” class of postnominal adjectives generates above the NP, while other “classificatory” postnominal adjectives are generated within NP. Dumitrescu and Saltarelli (1998) and Saltarelli (2001) propose that prenominal adjectives are actually complements to D, while postnominal adjectives are complements to N. 7 Postnominal adjectives have also been presented within a reduced relative clause (Cinque 2010), Demonte & Pérez-Jiménez (to appear), or in a small clause structure (Demonte 1999b, 2008, Gutiérrez-Rexach & Mallén 2001/2002). The precise nature of the functional projections of either prenominal or postnominal adjectives is irrelevant to the analysis pursued here. What is crucial is that the prenominal adjective is in a specifier position of the XP complement to D, while N is always in the head position of D’s lexical complement. 25 I do not enter the debate of N movement versus NP movement, but instead focus on the uncontroversial syntactic relationship between D and N. Regardless of functional projections, NP is the lexical complement to D. N is therefore the head of the lexical complement. In the previous work mentioned on this subject, APs are usually analyzed as specifiers or adjuncts in the larger DP, whether they attach to a functional XP complement or within the NP, though see Dumitrescu and Saltarelli (1998) and Saltarelli (2001) for an alternative view of adjectives as complements. In contrast to these researchers, I assume that an adjective is never the head of D’s complement, 8 as even prenominal adjectives are fundamentally optional material. Additionally, an adjective complement analysis eliminates any structural difference between A and N, forcing the distinction between them in the feminine el pattern to be reduced to lexical indexing. Indexation to a syntactic category distinction like adjectives and nouns in such a way is undesirable, since the flexibility of language often allows re-interpretation of a lexical word depending on the structure in which it is used (Borer 2005). Furthermore, while phonological effects of category distinctions have been recognized for a difference between nouns and verbs, the behavior of adjectives specifically patterns with nouns in phenomena such as stress patterning in Spanish (Smith 2011). Given this unified behavior elsewhere in the language, it would be undesirable to introduce lexical category specification to separate adjectives from nouns only for the feminine el. Rather, it is 8 In DPs with elided nouns, such as Spanish el D rojo A ‘the red one’, the noun is null, but the adjective remains in specifier or adjunct position, and does not move to the noun’s head complement position. See §2.5 for more discussion on elided noun constructions. 26 preferable that such shifts in adjective patterning are epiphenomenal, falling out of independent structural or relational restrictions. Simplified structures for DNA and DAN, focusing on only the lexical XPs, are presented in (2-6) for the phrases el D águila N blanca A ‘the white eagle’ and la D alta A águila N ‘the high eagle’. (2-6) a. DNA b. DAN In the syntactic structure, the N is always the head of the lexical complement to D, while prenominal A is not. When there is no prenominal adjective (either in a DP with no adjectives, or in one with only postnominal adjectives), D is adjacent to its head of complement N (notated HComp for clarity) (2-7a). With a prenominal adjective, D is separated from its head of complement, and is instead adjacent to the non-head A (2-7b). In the context of direct adjacency between D and the head of its lexical complement, i.e. N águila, the feminine el serves to avoid vowel hiatus. However, when D adjoins to a 27 non-head, i.e. A alta, the regular determiner la is permitted, despite the production of hiatus. (2-7) a. [ DP el [ NP águila HComp [ AP blanca]]] b. [ DP la [ NP [ AP alta] águila HComp ]] The syntactic relations of DN and DA are distinct, as is the phonological treatment of these sequences. To allow phonological sensitivity to these different relations, §2.2.2 turns to the syntax-phonology interface of Prosodic Phonology (Selkirk 1984, Nespor & Vogel 1986, Inkelas & Zec 1990, Itô & Mester 1992, 2007) as a means of reflecting the difference between DA and DN word orders. 2.2.2 Prosodic Phonology Interface As introduced in Chapter 1, Prosodic Phonology (Selkirk 1981; Nespor & Vogel 1986; Inkelas & Zec 1990; Itô & Mester 1992, 2007; Truckenbrodt 1995, 1999; and many others) is a framework of interface between syntax and phonology, utilizing a hierarchical structure of prosodic units (1-10) built from syntactic inputs and available for reference by phonological constraints. One of the important claims of the theory is that the relationship between syntax and phonology is indirect, mediated by the prosodic structure, which may be non-isomorphic with syntactic constituency. In building prosodic structure, one of the most common manifestations of Prosodic Phonology has access to syntactic XP edges, but not syntactic relations that are 28 internal to the XP (Chen 1985; Selkirk 1986; Selkirk & Tateishi 1988 1991; Selkirk & Shen 1990; and many others). Nevertheless, some researchers have proposed mapping approaches incorporating syntactic relations (Nespor & Vogel 1986, Chen 1990, Inkelas & Zec 1990, Seidl 2001). Here, I pursue a combined approach, in which structural syntactic edges provide the basis for most prosodic structures, but the internal head- complement relation is available to influence the attachment of function words. In this section, I first provide background regarding the edge-based and Match Theory structural approaches, and then address how I propose to incorporate the syntactic head- complement relation in the interface. In Optimality Theory (OT) (Prince & Smolensky 1993/2004), an edge-based approach to the analysis of the interface utilizes constraints on prosodic formation in output forms (McCarthy 1993; Selkirk 1996; Helsloot 1995; Truckenbrodt 1995, 1999; Bao 1996; Golston 1996; Peperkamp 1997; Yip 1999). An example of such a constraint governing the edge-based syntax-to-prosody mapping utilizes Alignment (McCarthy & Prince 1993) of the left and right edges of syntactic and prosodic structures. (2-8) ALIGN-L (XP, !) – Align the left edge of a syntactic XP to the left edge of a phonological phrase (!). ALIGN-R (XP, !) – Align the right edge of a syntactic XP to the right edge of a phonological phrase (!). 29 These constraints assign a violation for every syntactic XP left/right edge that does not correspond with the left/right edge of a phonological phrase. With Selkirk’s (1986, 1995) assumption that functional projections are invisible to the syntax-prosody mapping, the syntactic XPs referenced by these constraints are limited to lexical XPs. Along with forces governing the mapping from syntactic to prosodic structure, the Prosodic Phonology interface assumes certain tenets of prosodic well-formedness. Selkirk (1984) and Nespor and Vogel (1986) restrict the formation of prosodic categories with the Strict Layer Hypothesis (SLH), which assumes that prosodic constituents are strictly and exhaustively ordered within the levels of the prosodic hierarchy, forbidding recursive prosodic structures and structures with “skipped” levels. The idea of inviolable EXHAUSTIVITY and NONRECURSIVITY in prosodic structures has been challenged in the literature (Booij 1996; Peperkamp 1997; Itô & Mester 1992; Inkelas 1989; Ladd 1986, 1992; Kanerva 1989, Hayes 1995, and others), leading to a decomposition of the Strict Layer Hypothesis into separate violable constraints (Selkirk 1995). These constraints interact with the constraints on the syntax-prosody mapping to produce a wider range of prosodic structures than that predicted by a single unviolated SLH. (2-9) Strict Layer Hypothesis constraints a. LAYERDNESS: No prosodic constituent C i dominates C j , j>i b. HEADEDNESS: Any C i must dominate a C i-1 c. EXHAUSTIVITY: No C i dominates C j , j<i-1 d. NONRECURSIVITY: No C i dominates C i 30 The idea of LAYERDNESS is that a prosodic category should not dominate one that is higher in the prosodic hierarchy. For example, a Pwd should not dominate a PPh. HEADEDNESS maintains that each prosodic constituent should dominate a category that comes immediately beneath it in the Prosodic Hierarchy, so that, for example, a PPh should dominate a Pwd, and may not be made up exclusively of syllables or feet that are not parsed into Pwds. EXHAUSTIVITY states that no level of the hierarchy should be skipped, so that, for instance, there should always be an intervening Pwd between a PPh and foot. It is differentiated from HEADEDNESS in that EXHAUSTIVITY requires every foot to be dominated by a Pwd that is dominated by a PPh, etc. HEADEDNESS, on the other hand, only requires that each PPh contain some Pwd somewhere, and that each Pwd contain some foot, etc. NONRECURSIVITY requires only one layer for each category in the hierarchical structure. For example, no PPh should dominate another PPh. Selkirk (1996) suggests that LAYEREDNESS and HEADEDNESS appear to be unviolated in language patterns, but that EXHAUSTIVITY and NONRECURSIVITY may be violated in the prosodic structures of different languages. In recent developments on the syntax-phonology interface, recursive prosodic structure is argued for throughout the prosodic hierarchy (Itô & Mester 2010; Selkirk 2009, 2011; Elfner 2011), and in the Match Theory approach to the syntax-prosody mapping (Selkirk 2009, 2011; Elfner 2011), recursivity is fundamentally favored in prosodic structure. One of the more recent developments on the syntax-prosody mapping, Match Theory proposes that the forces driving the construction of prosodic structure aim for 31 complete one-to-one correspondence between the syntactic and prosodic structures. Although Match Theory is largely similar to the edge-based approach in that syntactic XP edges are matched to prosodic constituents, MATCH constraints on the syntax-prosody mapping are distinguished from edge-based Alignment by requiring complete isomorphy to the entire syntactic constituent, resulting in a preference for recursive prosodic structure. As argued in Elfner (2011), the Match Theory approach predicts the appropriate recursive prosodic structure for Connemara Irish (CI) phonological phrases, while an Alignment approach will always prefer a less recursive structure. An example of a MATCH constraint on the formation of prosodic structure is provided below (following Elfner’s 2011 definitions), along with a review of the different predictions for Match and Alignment approaches. (2-10) MATCH-PHRASE– Suppose there is a syntactic phrase (XP) in the syntactic representation that exhaustively dominates a set of one or more terminal nodes %. Assign one violation mark if there is no phonological phrase (!) in the phonological representation that exhaustively dominates the phonological exponents of the terminal nodes %. Exhaustive domination – A syntactic node % exhaustively dominates a set of terminal nodes & iff % dominates all and only the terminal nodes in &. 32 When highly ranked in a constraint grammar, MATCH-PHRASE predicts complete isomorphy with the syntax, preferring a recursive prosodic structure for the [V [S O]] transitive sentence in CI (Table 3), while Alignment constraints still allow NONRECURSIVITY to prefer a less recursive structure (Table 4) (tableaux reproduced from Elfner 2011). 9 Table 3. MATCH-PHRASE >> NONRECURSIVITY [ )* V [ TP [DP 1 ] [DP 2 ]]] 10 MATCH-PHRASE NONRECURSIVITY a. (V DP 1 DP 2 ) TP! DP 1 ! DP 2 ! b. (V (DP 1 ) (DP 2 )) TP! ** $c. (V ((DP 1 ) (DP 2 ))) *** Table 4. ALIGN-R(XP,!), ALIGN-L(XP,!) >> NONRECURSIVITY [ )* V [ TP [DP 1 ] [DP 2 ]]] ALIGN-R(XP,!) ALIGN-L(XP,!) NONRECURSIVITY a. (V DP 1 DP 2 ) DP1! TP! DP1! $ b. (V (DP 1 ) (DP 2 )) ** c. (V ((DP 1 ) (DP 2 ))) ***! 9 Also shown in Elfner (2011), the use of a WRAP-XP constraint (Truckenbrodt 1995) does not help the recursive candidate (c) win out over NONRECURSIVITY. 10 In Elfner (2011), )* is a functional projection headed by V. TP is the functional projection housing tense (T). 33 Elfner (2011) provides evidence from pitch accents in CI that support the fully recursive and isomorphic prosodic structure (V ((DP 1 ) (DP 2 ))), requiring a Match analysis of the syntax-prosody mapping. Given this and other evidence in favor of Match Theory over the edge-based Alignment approach (Selkirk 2009, 2011), I will assume a Match Theory approach to the data in Spanish. As we shall see in §2.2.2.4, the prosodic structures necessary for the Spanish feminine el data require a modification of the interface to allow reference to the syntactic head-complement relation. In particular, the head-complement relation is identified to ensure close prosodic attachment between the function word D and its host N. In the following section 2.2.2.1, I outline the background for recursive prosodic cliticization structures at the Pwd level of the prosodic hierarchy. Section 2.2.2.2 addresses the prosodic structures for Spanish determiner cliticization. 2.2.2.1 Clitic adjunction and prosodic recursivity Recursive Pwd structure has been proposed for many language patterns, expanding the Pwd category into multiple, nested levels via prosodic attachment (Booij 1996; Peperkamp 1997; Itô & Mester 1992, 2006, 2008, 2010; Selkirk 1996; Anderson 2005, Krämer 2009, Walker 2011). Itô & Mester (2006, 2008, 2010) borrow Grimshaw’s (1991) extended projection terminology to dub the highest layer of a given category as “maximal” and the lowest as “minimal”. They further propose that phonological constraints may reference not only the prosodic category but maximal and minimal subcategories from the recursive structure. Examples of such adjunction and category 34 nesting are given for the Pwd and PPh categories in (2-11), with the definition of maximal and minimal projections in (2-12). (2-11) ! PPh ( IP " maximal Pwd + maximal PPh " + X X " minimal Pwd Y Y + minimal PPh F foot " Pwd (2-12) K max(imal) =K not dominated by K K min(imal) =K not dominating K The different attachment possibilities produce different prosodic clitic structures, reviewed in Selkirk (1996) and Anderson (2005). These structures are internal clitics, affixal clitics, free clitics and Pwd clitics, presented in (2-13) with the Pwd min and Pwd max terminology. 35 (2-13) a. Internal clitic b. Affixal clitic PPh PPh Pwd min,max Pwd max Clitic Host Pwd min Clitic Host c. Free clitic d. Pwd clitic PPh PPh Pwd max Pwd max Pwd max Clitic Host Clitic Host In an internal clitic structure, such as may occur with word-level affixes, the clitic and host are combined within the minimal Pwd constituent. In this case, the clitic syllable would be predicted to participate in word-level prosodic effects, such as being counted in the noun’s syllables for stress assignment. An affixal clitic structure nests the Pwd categories, with the host word receiving its own Pwd min level and the clitic attaching to form an additional Pwd max . 11 This nested structure violates NONRECURSIVITY. Free 11 In Spanish, an extended Pwd max that includes affixal clitics is not expected to be the domain in which stress is assigned. There is a distinction between word-level affixes, such as inflectional person suffixes on verbs that count in primary stress assignment, and preposed verbal clitics, that do not. This stress difference is a strong indication that some affixes attach within the minimal Pwd, while other prosodic clitics attach at higher levels, such as in an affixal or free clitic structure. 36 clitics, on the other hand, violate EXHAUSTIVITY in that the clitic syllable is included directly in the PPh without an intervening Pwd level. Pwd clitics receive their own separate Pwd category before combining with the lexical host Pwd. In this case, such clitics would be expected to demonstrate full Pwd phenomena, such as receiving their own primary stress. These different structures also predict different prosodic boundaries intervening between the clitic and its host. For example, in a free clitic structure, the intervening boundary between the clitic and host is Pwd max (2-14b), while in an affixal clitic structure, the intervening boundary between the clitic and host will be non-maximal 12 (2-14a). (2-14) a. affixal clitic b. free clitic " max ! " min " max Clitic Host Clitic Host [ Pwdmax Cl [ Pwdmin H]] [ PPh Cl [ Pwdmax H]] Itô and Mester (2008) argue that the recursive affixal clitic structure is the appropriate representation for English and German function word prosodic adjunction, in 12 In the figure given, the boundary is labeled as Pwd min , but in the case of multiple clitics, or a compound word host, it may be non-minimal. It will never be maximal, however. 37 comparison to the internal clitic structure that attaches within the minimal Pwd or the free clitic structure with adjunction at the PPh level, outside the maximal Pwd. Parsing the function word in question as a full Pwd in these languages is ruled out on independent considerations of word binarity and stress. A review of their evidence for the English case is presented below. Itô and Mester show that sequences of a function word + lexical word such as English a laguna produce two adjacent unstressed syllables at the beginning of the structure, which would be disallowed were the function word incorporated within the minimal Pwd constituent and subject to word-level stress assignment. Compare (2-15a) with a function word-lexical word sequence to the single lexical word in (2-15b) with a maximum of one unstressed syllable at the left edge, and (2-15c) in which secondary stress is assigned to the first syllable of a morphologically related form. (2-15) a. a lagúna b. allége c. àllegátion Itô and Mester (2010) also examine evidence from English r-sandhi in non-rhotic dialects to argue for recursive "-adjoined affixal clitic structure in comparison to the !- adjoined structure of a free clitic. In American English non-rhotic dialects, r-insertion 38 appears after Pwds (2-16a), but is disallowed between a reduced function word and following Pwd (2-16b). 13 (2-16) a. ide["] -r is b. t["](*-r) add t["](*-r) his troubles The locus of r-insertion is identified as preceding a maximal Pwd boundary, as compared to the minimal Pwd boundary that occurs at the left edge of the lexical host in function word adjunction. The prosodic cliticization structures shown in (2-17) demonstrate r- insertion across a maximal Pwd boundary between the words idea-r is, while r-insertion is blocked within the Pwd max between the words the(*r) idea. (2-17) " max " max " min " min th["](*-r) ide["]-r is sound Were function words adjoined at the PPh, instead of producing the recursive Pwd structure, r-insertion would be incorrectly predicted after function words. Compare the PPh r-insertion prediction in (2-18b) with the correct prediction of the recursive Pwd clitic structure in (2-18a). 13 Itô and Mester (2010) contrast the American English non-rhotic pattern with the Norwich English dialect, which does allow r-intrusion after reduced function words to, the, a, of, and other reduced vowels. 39 (2-18) a. " max b. ! " min " (max) t["](*r) add t["](-r) add While Itô and Mester (2008, 2010) argue that the free clitic PPh-adjoining structure is not appropriate for English function words in favor of the recursive affixal clitic structure, I propose that both affixal and free clitic prosodic structures are needed to account for the behavior of Spanish determiners. 14 2.2.2.2 Cliticization of Spanish determiners As in English, Spanish determiners are stressless function words and behave prosodically as proclitics, adjoining to the following lexical word in the prosodic structure 15. Also similar to English reduced function words, neither the Pwd clitic nor internal clitic structures are appropriate. In Spanish, one of the prosodic manifestations of lexical stress is pitch accent. Determiners do not receive lexical stress or pitch accent, differentiating the determiner form el from the homophonous third person masculine singular pronoun. For example, in (2-19a), the determiner el does not receive its own 14 Padgett (2012) also proposes both free clitic and affixal clitic structures for function words in Russian, although each structure is associated with a single function word type. In Spanish, the same function word—the definite article—changes its prosodic cliticization depending on the surrounding syntactic context. 15 Though unlike English, in Spanish there is no vowel reduction to schwa. 40 pitch accent, instead providing a smooth transition from starting pitch to the L*+H accent of the noun. Since it does not require its own pitch accent, it also provides more time to drop to the L* in the following word, which falls under 120Hz. In contrast, the subject pronoun él 16 in (2-19b) demonstrates a clear pitch accent, and leaves less time to drop down for the following L*, which only reaches about 140Hz. The lack of Pwd-level characteristics indicates that the determiner is not prosodified as a separate Pwd 17 . (2-19) a. el D atajo N ‘the shortcut’ el atajo está por allí L*+H L*+H L* L% elatajoesta 75 175 100 120 140 160 Time (s) 1.214 2.461 16 As has been pointed out, pronouns also occupy a D position. However, they are different from article determiners and other Ds in that they do not require an overt NP complement, instead fulfilling the NP requirement in and of itself. 17 In a focus construction, the determiner may receive contrastive stress, for example Vi EL águila (no CUALQUIERA águila) ‘I saw THE eagle (not just ANY eagle).’ In these constructions, the article is promoted to Pwd status. As is evident, the feminine el alternation still occurs, even though the prosodic triggering context has disappeared. Further discussion of the maintenance of the feminine el pattern in a prosodically opaque context such as this is addressed in !2.4. 41 (2-19) b. él S atajó V ‘he took a shortcut’ él atajó por allí cerca L* H* PPh L+H* L* L% élatajó 75 175 100 120 140 160 Time (s) 9.607 11.2 Stress patterning again provides evidence against the internal clitic structure for Spanish determiners. Were the determiner adjoined at the minimal Pwd level, it would be expected to count in default word stress patterning. In Spanish, if a word ends in a vowel, n or s, primary stress falls on the penultimate syllable, for example (2-20a) 18 . However, monosyllabic words with the same phonological conditions still receive primary stress on the lexical syllable, rather than shifting to the determiner to meet the default stress pattern (2-20b). 18 This describes the default stress pattern. Stress is lexical in Spanish, and other non-default patterns are attested, but they are all marked orthographically to indicate the non-default stress placement. In the examples in (2-20), primary stress marks have been added to show where the default stress falls, but are unmarked in Spanish orthography. 42 (2-20) a. gátos N ‘cats’ b. la D tós N , *lá D tos N ‘the cough’ Given this lack of primary stress shift with a function word + lexical word sequence, the internal clitic structure is inappropriate for Spanish determiners 19 . This leaves the affixal and free clitic structures for Spanish determiners. I argue that in the DNA and DAN sequences, the determiners are prosodified differently, using these two structures. In a DAN sequence, the determiner is a free clitic, adjoining at the PPh level, while in a DNA sequence, it is an affixal clitic, adjoining at the Pwd level to form a recursive Pwd with the noun. 19 Interestingly, stress shift does appear with post-verbal enclitics in Argentinian Spanish, where the host word’s lexical stress shifts in response to the additional enclitic syllable (Colantoni et al, 2010). For example, the verb+clitic structure pasándo+la ‘passing it’ becomes pàsandóla instead of maintaining the lexical stress placement in pasándola. I would argue that this is an example of internal cliticization, where the enclitic attaches within the minimal Pwd. 43 (2-21) a. affixal DNA b. free DAN " max ! " min " max D N D A (" max D (" min N)) (!D (" max A)… 20 ) el agua la agria the water the bitter The phonological distinction between DNA and DAN sequences in the feminine el phenomenon forms one of the primary pieces of evidence for the prosodic structures above. A detailed phonological analysis is presented in §2.3, but the basic relationship between prosodic structure and phonological alternation is as follows. Determiner alternation to the feminine el, instead of the regular feminine la, occurs when the determiner is adjacent to the noun, but not when it is adjacent to an adjective. In the DN affixal clitic structure, the D and N are both included in a maximal Pwd, with only a minimal Pwd boundary intervening. In the feminine el pattern, this is reflected in the sensitivity to Pwd-internal avoidance of vowel hiatus, requiring the shift to el in order to avoid adjacent a vowels within the maximal Pwd. Stronger phonetic cues for divisions of 20 The ellipsis here represents that there is additional structure in this phrase, corresponding to the following noun of the DAN sequence. I assume that PPh may be ternary branching for this clitic structure, instead of strictly binary. If strict binary PPhs were proposed, then the free clitic structure would involve recursive PPhs: (!(! min D (" max A)) (" max N)). While the placement of an additional non-minimal PPh is not contraindicated by Spanish phonetic properties and does not affect the analysis given here, further discussion of the implications of minimal and maximal PPh boundaries is provided in §2.2.2.3. 44 the acoustic signal have been demonstrated for higher level prosodic boundaries, in comparison to lower level prosodic boundaries (Ladd 1988). This logically results in the possibility of an increased perceptual contrast between segments across a higher, and therefore potentially stronger, prosodic boundary. I hypothesize that the minimal Pwd boundary between the vowels is not strong enough to provide a sufficient perceptual distinction between the adjacent a vowels. In contrast, with the free clitic DA structure, a maximal Pwd boundary separates the D and A, forming a higher, and potentially phonetically stronger, prosodic boundary, which I hypothesize supplies a sufficient perceptual contrast between the two vowels to determine the quality and quantity of vowels in the sequence, and allow phonological hiatus to remain. Phonetic evidence in support of these hypotheses about such boundary differences comes from an investigation of the duration of vowels in hiatus at the DN and DA junctures. As will be discussed in detail in Chapter 3, a continuous sequence of two vowels at the DA juncture is longer than vowels at DN. This phonetic effect is predicted by the higher level maximal Pwd boundary of the free clitic structure proposed for the DAN sequence forming a stronger boundary in comparison to the lower level minimal Pwd boundary between the D and A in the affixal clitic structure. To summarize, Spanish DNA and DAN sequences are best differentiated by prosodic clitic adjunction below and above the " max . The D of DN is prosodified as an affixal clitic below the " max of the lexical N (2-22a). The D of DA is prosodified as a free clitic above the " max of the lexical A (2-22b). 45 (2-22) a. (" max D (" min N)) b. (! D (" max A)) An internal clitic structure such as (" min D N) is inappropriate for either sequence, since primary stress—a word-level phenomenon—never shifts to the functional syllable. A Pwd clitic structure such as (" D) (" A) is also inappropriate, since the function word does not receive its own primary stress or pitch accent. The remaining clitic adjunction structures provide the appropriate prosodic distinction for the phonological phenomenon, and account for the greater phonetic strength of the boundary at DA in comparison to DN. 2.2.2.3 Analysis of word adjunction Returning to the prosodic structures of affixal and free clitics for the DNA and DAN sequences, the next question is how these clitic structures are generated from the syntactic structure. As mentioned in §2.2.2.1, Match Theory has been argued to produce more accurate empirical predictions than previous edge-based accounts. I therefore follow recent developments in Match Theory for the mapping of Spanish prosodic clitic structures. To produce the appropriate prosodic structures, I propose the addition of the syntactic head-complement relation in mapping constraint formulation, requiring function words to attach within the Pwd of their lexical complements. Selkirk’s (2009, 2011) MATCH constraints may only match corresponding levels of syntactic and prosodic constituency: clauses with (, XP with !, and syntactic words 46 with ". For example, a syntactic XP may not be matched with a prosodic word. Instead, the Match constraint responsible for Pwd structure is MATCH-WORD (definition from Selkirk 2011). Since Selkirk (2011) assumes that function words are invisible to the mapping from syntax to prosody, all her Match constraints are automatically constrained to lexical words. (2-23) MATCH-WORD – A word in the syntactic constituent structure must be matched by a corresponding prosodic constituent, call it ", in phonological representation. According to Itô and Mester (2010), a syntactic “word” is the phonological material associated with a terminal node in the syntactic structure. I propose a more precise definition of MATCH-WORD that operates over the morpho-syntactic terminal nodes. Since like Elfner (2011) I assume that function words are visible to the mapping, the definition also specifies lexical words over function words (2-24). The constraint favors a prosodic structure in which every terminal lexical node is prosodified into its own Pwd, and assigns a violation for every node that is not a separate Pwd in the prosodic structure output. 21 This does not forbid or favor recursive structure, but instead 21 The constraint definition provided here is more than sufficient for the purposes of the present analysis, but there are outstanding questions as to the relationship between prosodic mapping and morphological transformations, particularly given work in the framework of Distributed Morphology (Halle 1990; Halle & Marantz 1993, 1994; Noyer 1997; among others) that argues for post-syntactic movement or transformations in the PF component of the grammar (Embick & Noyer 2001). For example, should prosodic mapping occur before, after, or concurrent with morphological mergers or vocabulary insertion of phonological material? Given the sensitivity of certain morphological operations to prosodic well-formedness—like the use of English superlative –er with adjectives of one syllable—some morphological movement operations 47 simply requires that all lexical items be Pwds at some level of the prosodic representation. (2-24) MATCH-WORD – For every terminal lexical node % in the syntactic structure, assign a violation if there is no prosodic word (") in the prosodic structure that contains all and only the phonological exponent(s) of the terminal node %. The above Match constraint will disfavor an internal clitic structure, since the lexical N is not matched to its own Pwd. All other clitic structures, however, will obey the constraint because the lexical host is associated with a Pwd, whether it is a Pwd max or Pwd min (2-25). (2-25) [D [N]] MATCH-WORD (" min , max D N) * (" max D (" min N)) (" max D) (" max N) D (" max N) should be parallel with prosodic mapping and well-formedness constraints. Further research is needed for a fuller formalization of morphology-prosody interactions. 48 There must also be a constraint driving adjunction, so that the determiner clitic is not left unparsed at the Pwd level. Following McCarthy and Prince (1993) and Itô and Mester (2008, 2010), a family of PARSE-INTO-X constraints require that all elements of the terminal string be parsed at a given level of the prosodic hierarchy. (2-26) PARSE-INTO-X (PARSE-INTO-#, PARSE-INTO-F, PARSE-INTO-", PARSE-INTO-!, etc.): Every phonological element of the terminal string is parsed at the X level. The PARSE constraint relevant to the affixal and free clitic structures is PARSE- INTO-". The free clitic structure (! D (" N)) violates PARSE-INTO-", while the affixal clitic structure ("D ("N)) obeys it. I assume that PARSE-INTO-! is top-ranked in the grammar, requiring D to be included in the PPh for a free clitic structure, rather than left outside of the prosodic structure entirely 22 . PARSE-INTO-" also makes different predictions from MATCH-WORD. PARSE-INTO-" requires that all syntactic words, including function words, be in a Pwd, but not necessarily in separate Pwds. Further constraints on prosodic well-formedness necessary to produce all prosodic clitic patterns include HEADEDNESS AND NONRECURSIVITY, reviewed earlier in §2.2.2 and repeated here in (2-27). (2-27) HEADEDNESS: Any C i must dominate a C i-1 NONRECURSIVITY – No C i dominates C i 22 This constraint would be particularly relevant were function words assumed to be invisible to prosodic phrasing, as proposed by Selkirk (2004, 2011). 49 HEADEDNESS assigns a violation for any " that does not contain at least one foot. Unstressed monosyllabic determiners do not contain a foot, and therefore Pwd clitic structures do not meet the HEADEDNESS requirement for " prosodic structure. (I also assume that insertion of phonological material to create a foot is forbidden, such as by a dominating DEP-IO constraint.) NONRECURSIVITY militates against prosodic structures that contain structures of the same category, for example a Pwd dominating another Pwd. While Selkirk (2011) argues that NONRECURSIVITY is not necessary for PPh level phrasing, and suggests that it may not be necessary at all for the universal grammar, without it the affix clitic candidate harmonically bounds the other prosodic clitic forms (Table 5). Since the internal and Pwd clitics occur crosslinguistically (Selkirk 1996, Peperkamp 1997) and the free clitic structure must emerge in the Spanish data examined here as well, I assume that this constraint is present in the grammar. To leave open the possibility that NONRECURSIVITY may not apply at higher levels of the prosodic hierarchy, I will specify NONRECURSIVITY" for the remainder of the analysis. 50 Table 5. Constraint violations of HEADEDNESS, PARSE-INTO-", MATCH-WORD 23 [ DP D [ FP N HEADEDNESS PARSE-INTO-" MATCH-WORD a. (" D N) Internal clitic * b. (" D) (" N) Pwd clitic * c. ("D ("N)) Affixal clitic d. (! D (" N)) Free clitic * In Table 5, the affixal candidate of (c) obeys all constraints, and therefore will always win, regardless of constraint ranking. The Pwd clitic structure of candidate (b) is ruled out by HEADEDNESS, the internal clitic candidate (a) is ruled out by MATCH-WORD, and the free clitic candidate (d) is ruled out by PARSE-INTO-". There is no permutation of constraint rankings where the affixal structure will lose. With the inclusion of NONRECURSIVITY", the problem of harmonic bounding is eliminated. The constraints MATCH-WORD, HEADEDNESS, PARSE-INTO-", and NONRECURSIVITY" will generate all four possible clitic adjunction candidates in a 23 Recall that in the Y-model of the grammar shown in §1.2.2, the S-structure of syntax forms the input for prosodic structure formation. Therefore, I use syntactic inputs for all tableaux pertaining to building prosodic structure. 51 factorial typology. 24 The candidates are examined in Table 6, with the factorial typology given in Table 7. Table 6. Constraint violations of HEADEDNESS, MATCH-WORD, PARSE-", NONRECURSIVITY" [ DP D [ FP N HEADEDNESS MATCH-WORD PARSE-" NONRECURSIVITY" a. (" D N) internal clitic * b. (" D) (" N) Pwd clitic * c. ("D ("N)) affixal clitic * d. (! D (" N)) free clitic * 24 The typology here is simplified a little with regard to the behavior of HEADEDNESS. The Pwd clitic structure, when it occurs in a language, could be expected to still obey HEADEDNESS by inserting enough phonological material to provide a foot in the Pwd “clitic”. This would allow HEADEDNESS to be unviolated but still produce the Pwd clitic structure (following Selkirk 1996). As mentioned above, this repair is impossible in Spanish and so I do not address it here in the typology. 52 Table 7. Factorial Typology Constraint ranking Clitic structure HEADEDNESS, PARSE-", NONRECURSIVITY" >> MATCH-WORD (" D N) internal clitic MATCH-WORD, PARSE-", NONRECURSIVITY" >> HEADEDNESS (" D) (" N) Pwd clitic HEADEDNESS, PARSE-", MATCH-WORD >> NONRECURSIVITY" (" D (" N)) affixal clitic HEADEDNESS, MATCH-WORD, NONRECURSIVITY" >> PARSE-" (! D (" N)) free clitic To achieve the Spanish affixal and free clitic structures, HEADEDNESS and MATCH-WORD are ranked above PARSE-" and NONRECURSIVITY" to rule out the internal and Pwd clitic candidates. Individual ranking of PARSE-" and NONRECURSIVITY" will produce either an affixal or free clitic structure, but further modification is necessary in order to produce both clitic structures in Spanish determiners. Returning to the syntactic structure, we recall that there is a difference in the relationship between D and the following N for the DNA sequence, and between D and the following A in the DAN sequence. N is in the head position of the complement to D, while prenominal A is not. Thus, when there is no prenominal adjective, the D is directly 53 adjacent to the head of its complement N (2-28a), while it is adjacent to a non-head of its complement when there is a prenominal adjective (2-28b). (2-28) a. [ DP el [ FP águila HComp [ AP blanca]]] b. [ DP la [ FP [ AP alta] [ FP águila HComp [ AP blanca]]]] The distinction between the affixal clitic DN and free clitic DA prosodic structures relies on syntactic headedness of the host word. If the host word is the syntactic head of the lexical XP complement to the clitic, D attaches within the host’s Pwd. If the host word is a non-head of the complement, then it attaches outside of the maximal Pwd. The prosodic constraint that ensures that a function word attaches within the Pwd is Parse-into-". Expanding the structure of this constraint allows that attachment directive to specify the type of host word (2-29). (2-29) PARSE FNC -INTO-" HCOMP – For every function word f, parse f into the Pwd of the head of its lexical complement. For the Spanish DP sequences, ranking this PARSE FNC constraint over NONRECURSIVITY" and PARSE-" will result in the affixal clitic when the noun is the host (and a head) (Table 8i), but the free clitic when the adjective is the host (a non-head) (Table 8ii). 54 Table 8. PARSE FNC -INTO-" HCOMP >> NONREC" >> PARSE-" i. [D [N HComp [AP]]] DP PARSE FNC -INTO-" HCOMP NONREC" PARSE-" $ a. (! (" max D (" min N)) (" max A)) affixal clitic * b. (! D (" max N) (" max A)) free clitic *! * ii. [D [AP] [N HComp ]] DP a. (! (" max D (" min A)) (" max N)) affixal clitic * *! $ b. (! D (" max A) (" max N)) free clitic * * The DN sequence in Table 8i is correctly prosodified into an affixal clitic structure in candidate (ia), because N is a head, and D is included in N’s Pwd. The free clitic structure in candidate (ib) does not include D inside the Pwd of its complement head N, and so violates the PARSE FNC constraint and is ruled out. For the DAN sequence in (ii), D is not parsed into its lexical complement head Pwd in either case, because it is nonadajcent to N. Both candidates (iia) and (iib) incur a violation of PARSE FNC -INTO-" HCOMP and tie. NONRECURSIVITY" rules out the affixal clitic option of candidate (iia), and the free clitic structure of (iib) is selected. 55 I assume that a linearization constraint like Elfner’s (2011) LINEARCORRESPONDENCE (LINCORR) prevents movement of D to attach to N for the underlying DAN sequence. (2-30) reviews Elfner’s definition of LINCORR. (2-30) LINEARCORRESPONDENCE (LINCORR) – Assign one violation mark for every syntactic node # whose terminal nodes do not precede the set of terminal nodes dominated by a syntactic node $ which # asymmetrically c-commands. In the case of the DAN sequence of Spanish, this constraint requires D to precede A, which is in the set of terminal nodes c-commanded by D (2-31, repeated from 2-6b). (2-31) Table 9 demonstrates the ranking of LINCORR that prevents movement repair to PARSE FNC . 56 Table 9. LINCORR >> PARSE FNC -INTO-" HCOMP [ DP D [AP] [N HComp ]] LINCORR PARSE FNC - INTO-" HCOMP NONREC" PARSE-" a. (! (" max A) (" max D (" min N)) D movement *! * $ b. (! D (" max A) (" max N)) free clitic * * Candidate (a) obeys the PARSE FNC constraint by parsing D in the Pwd of its lexical complement head N, but it violates LINCORR and is ruled out. The free clitic structure of (b), with maintenance of the linear order DAN, wins as the optimal structure. 25 Other candidates that must be ruled out by additional constraints include one with ternary branching structure within the Pwd, and one that compounds the A and N together (2-32). (2-32) a. ternary branching: (" max D (" min A) (" min N)) b. compounding: (" max D (" (" min A)(" min N))) 25 I also assume that nonadjacent prosodic attachment is restricted at GEN. For example, a candidate that does not move D to be adjacent to N but which nevertheless adjoins D at N’s Pwd, excluding the intervening A, would never be posited by GEN. 57 These candidates are ruled out by constraints restricting the size of prosodic constituents. The ternary branching candidate (2-32a) is ruled out by a binarity constraint on prosodic branching. Following Elfner (2011), BIN-MAX(") requires at most two daughter branches for each Pwd (2-33). 26 (2-33) BIN-MAX(") – assign one violation mark for every " that dominates more than two daughter constituents. The compound form in (2-32b) is ruled out by a constraint MATCH-" LEX , requiring each Pwd to be matched to a single lexical syntactic word (2-34). 27 This constraint requires separate Pwds for every lexical word, but is not sensitive to functional material. Therefore, the affixal candidate (" D (" min N)) does not violate this constraint. In general, the constraint prevents rampant compounding of any two lexical words in sequence. 28 26 Mester (1994), Hewitt (1994), Selkirk (2000), and Itô and Mester (2006) have similar constraints on binarity, though with slightly different definitions. For example, Itô and Mester (2006) require at most two daughter categories, so that it is the maximal number of feet counted for each Pwd that determines the constraint violation. 27 Similar constraints regarding the mapping of lexical words, and preventing unattested word adjunction, are presented in Itô and Mester (2010) as LEX-TO-X edge alignment mapping. 28 I assume that although regular compound words also violate this constraint, compounds are protected in the language by morphological input structure marking them as a single maximal word. MATCH-WORD ranked above MATCH-" LEX will require the compound to be parsed as a whole, rather than separately. 58 (2-34) MATCH-" LEX – assign one violation mark for every " that is not matched to a single lexical word in the syntactic structure. The ranking of these BIN-MAX and MATCH-" LEX constraints over PARSE FNC will rule out the undesired candidates, as shown in Table 10. Table 10. BIN-MAX(") and MATCH-" LEX [ DP D [AP] [N HComp ]] BIN-MAX(") MATCH-" LEX PARSE FNC - INTO-" HCOMP a. (" max D (" min A) (" min N)) ternary branching *! b. (" max D (" (" min A)(" min N))) compounding *! $ c. (! D (" max A) (" max N)) free clitic * Candidate (a) violates binarity, because the outer Pwd max has three daughter nodes, pertaining to the D, the A and the N. Candidate (b) obeys binarity by compounding A and N, so that each level of the Pwd structure only has two daughters. But, it violates MATCH- " LEX because the intermediate Pwd contains two lexical words, instead of only one. The free clitic structure in candidate (c) wins out, despite violating PARSE FNC . 59 On the theoretical level, introducing an internal syntactic relation like head- complement into the mapping interface is a major addition to Prosodic Phonology. The following section outlines some of the theoretical challenges raised by this move, as well as an illustration of how a purely edge-based or structural approach failes to account for the data. 2.2.2.4 Syntactic head relation in Prosodic Phonology The issue of whether the mapping should include internal syntactic relations or only an outer skeleton of the syntax is an ongoing debate in the literature on Prosodic Phonology. Nespor and Vogel (1986), Chen (1990) and Wagner (2005) propose the use of internal syntactic relations in the mapping, including head-complement (Chen 1990). One of the primary arguments in favor of syntactic relations is empirical coverage. The additional syntactic information allows a wider range of data patterns to be predicted, including attested patterns like the Spanish feminine el. But visibility of syntactic relations internal to the XP has been strongly argued against, since it is less restrictive a theory than one that is strictly boundary based (Chen 1985; Selkirk 1986, 1996; Selkirk & Tateishi 1988, 1991; Selkirk & Shen 1990; Bao 1996). I propose several restrictions to the syntactic relations available to the interface. First, I maintain that it is the relationships of the lexical items, and not the structure of multiple possible functional projections, that are important to the prosodic mapping. Following Elfner (2011), function words are not invisible to the syntax-prosody mapping (contra Selkirk 1986, 1996), but syntactic projections that are empty of phonological 60 material do not register for the formation of prosodic constituents, which are by definition phonological. That is, one cannot have empty prosodic structure. There must be phonological material in every prosodic constituent. Because of this basic restriction on prosodic structure, it is reasonable that the syntactic information available to prosodic mapping is restricted to information pertaining only to the syntactic terminal nodes representing phonological items. The purely functional relationships between a word and its movement trace, for example, are not available to the interface. I further restrict syntactic relation visibility to only the notions of head and complement. The complement, or argument, is one of the most basic of syntactic relations. Combining two items in a head-complement relation is the first step in building a syntactic structure. It is expressed as a sisterhood relationship within the X-bar schema (Chomsky 1970) and more recently, as set Merge in the Minimalist program (Chomsky 1995, Boeckx 2008). This relation is proposed as accessible to prosody precisely because it is so important in building linguistic structure. Unlike adjuncts, which are optional, complements are crucial to syntactic structure. It is fitting for such a basic relation to be incorporated into the syntax-phonology interface. The language patterns examined in this dissertation provide evidence for the use of the head-complement syntactic relation for the parsing of function words, specifically. Whether the relation is restricted only to the mapping of function words or may apply 61 generally throughout the syntax-prosody mapping is an open question for future crosslinguistic research. 29 With these restrictions, the head-complement approach allows full advantage of increased empirical coverage. Not only does the Parse Fnc constraint account for the feminine el, but the analysis easily adapts to account for the behavior of conjunction and other function words in Spanish. On the other hand, an attempt to provide a purely structural MATCH analysis of these data—with no internal relations—faces theoretical challenges with unattested predictions and fails to account for all function word behavior. I address these issues in the following sections. 2.2.2.4.1 Spanish conjunction and other function words Like determiners, Spanish conjunction allomorphy shows phonological evidence of affixal prosodic cliticization. Unlike the feminine el, it is not sensitive to syntactic complements. In Spanish, the two coordinators y ‘and’ and o ‘or’ consist of a single vowel. When followed by a consonant or non-identical vowel, they are realized as [i] and [o], respectively (2-35a). When followed by an identical vowel, an alternate allophone surfaces to create a non-identical hiatus sequence (2-35b). 29 An example of a head-complement relation elsewhere in the syntax that may relate to the above question is the relation between a verb and its direct object. Lexical items in a VO structure may prosodify separately due to the presence of other constraints in the grammar (such as the MATCH- " LEX utilized earlier in this chapter), but if the head-complement relation applies to the mapping of lexical as well as function items, it could predict a language that forms compounds of verbs and their objects. Continuing the discussion of the VO relationship, I propose that object clitic-V sequences do not prosodify in the same way that determiner-N sequences do because the head-complement relation is not precisely the same. For a DN sequence, D is the head of the larger DP structure, while N is the head of its complement. For an ObjClitic-V sequence, V is the head of the larger structure while the object clitic is the head of its complement. 62 (2-35) a. maní [i] nuez ‘peanut and walnut’ pizarras [i] aulas ‘blackboards and classrooms’ impostores [i] engaños ‘imposters and tricks’ estos [i] otros ‘these and others’ aplicaciones [i] usos ‘applications and uses’ uno [o] dos ‘one or two’ sábanas [o] almohadas ‘sheets or pillows’ auditorios [o] estadios ‘auditoriums or stadiums’ hijas [o] hijos ‘sons or daughters’ aplicaciones [o] usos ‘applications or uses’ b. engaños [e] impostores, *[i] impostor ‘tricks and imposters’ uno [u] otro, *[o] otro ‘one or (the) other’ It is notable that in the phrases un[o o] dos and man[í i] nuez, even though there are identical o-o and i-i sequences between the conjunction and the previous words uno and maní, the allophones [u] and [e] are not triggered. I propose that this difference stems from prosodic clitic formation. The conjunction is a proclitic, attaching at the maximal prosodic word level on its right, not its left. Therefore, in the phrase uno o dos, a maximal Pwd boundary separates the o from the o-final word on its left (2-36a), while the o-o sequence resulting from the rightwards conjunction-Pwd sequence occurs within a maximal Pwd and is restricted (2-36b). 63 (2-36) a. [uno " max ] [o [dos " min ] " max ] b. [uno " max ] [u [otro " min ] " max ] Unlike the feminine el, intervening adjectives within the second conjunct XP do not block allomorphy, behaving exactly the same as nouns (2-46). Instead, I propose that Spanish conjunctions always take the affix clitic structure within the maximal Pwd (2-37). 30 (2-36) osados A guerreros N [i] audaces A marineros N ‘daring warriors and brave sailors’ crueles A invasores N [i] orgullosos A patriotas N ‘cruel invaders and proud patriots’ amplios A espacios N e increibles A paisajes N ‘wide spaces and incredible landscapes’ altas A torres N e invencibles A murallas N ‘tall towers and invincible walls’ osados A guerreros N [o] audaces A marineros N ‘daring warriors or brave sailors’ amplios A espacios N [o] increibles A paisajes N ‘wide spaces or incredible landscapes’ crueles A invasores N u orgullosos A patriotas N ‘cruel invaders or proud patriots’ audaces A marineros N u osados A guerreros N ‘brave sailors or daring warriors’ 30 Given phonetic evidence found in Chapter 3 of a prosodic distinction between the affixal clitic DN and free clitic DA, the affixal clitic analysis of conjunction predicts that in Spanish all conjunction-Pwd sequences would be produced phonetically similar to Spanish DN sequences, with no difference between conjunction-adjective and conjunctive-noun contexts. 64 (2-37) engaños N [e Conj [impostores N " min ] " max ] amplios A espacios N [e Conj [increíbles A " min ] " max ] paisajes N In terms of the prosodic mapping, the difference in behavior of determiners and conjunctions brings up the question of how other function words in Spanish are prosodified, and how to achieve these differences in the ranking of mapping constraints like the PARSE family. Conjunction is always affected by hiatus of identical vowels, determiners only when followed by a noun, and other function words not at all. For example, the preposition de ‘of/from’ and the complementizer que ‘that’ may be followed by words beginning with stressed é, and do not undergo repair for the identical stressed hiatus e-é. (2-38) d[e é]llos ‘from them’ Me dice qu[e é]che la llave. ‘He/she tells me to turn the key.’ I propose that identical vowel hiatus is repaired within a Pwd max domain but not across a Pwd max boundary, and that different prosodic cliticization structures account for the differences in vowel hiatus behavior. I take the function words mentioned above to prosodify always as free clitics, outside the maximal Pwd of the following host. The distribution of prosodic structures for the different function words of Spanish then include affixal clitics for conjunctions, affixal and free clitics for determiners, and free clitics for all other function words. To produce these different prosodic mappings, I 65 propose that the PARSE mapping constraints are lexically indexed to different function word categories. Lexical indexing of constraints in this way has been used for various other word groupings, including morphological classes and exceptional lexical items (Pater 2000, 2006, 2008, 2009). As seen in §2.2.3, the ranking of PARSE-" >> NONRECURSIVITY" produces the affixal clitic structure, while ranking NONRECURSIVITY" >> PARSE-" produces the free clitic structure. I assume that this latter ranking holds for the elsewhere case of functional words in Spanish, and that a lexically indexed PARSE-" CONJ ranked above NONRECURSIVITY" holds to produce the affixal clitic structure in conjunctions, but leaves the other function words in a free clitic structure. Table 11. Lexically indexed PARSE-" i. [ ConjP Conj [ XP Conjunct2 ]] PARSE-" CONJ NONRECURS" PARSE-" $ a. (" max Conj (" min Conjunct2)) * b. (! Conj (" max Conjunct2)) *! * ii. [ PP P [ XP X]] a. (" max P (" min X)) *! $ b. (! P (" max X)) * The candidate set (i) illustrates the application of the lexically indexed PARSE-" CONJ constraint to conjunctions, forcing an affixal clitic structure in violation of NONRECURSIVITY". The candidate set in (ii) on the other hand, does not involve a 66 conjunction, and bypasses the lexically indexed PARSE-" requirement. Instead, the nonrecursive free clitic structure is preferred with prepositions and other function words. Since the determiner is the only function word that shows sensitivity to the type of following host—head of the lexical complement—I propose that the PARSE FNC constraint is lexically indexed to determiners. Conjunctions and other function words will vacuously satisfy the lexically indexed version of the constraint, allowing PARSE-" CONJ , PARSE-" and NONRECURSIVITY" to decide the prosodic mapping of those function words. Table 12 illustrates the mapping for all function word types: determiners, conjunction, and all others (in this case, demonstrated for prepositions). 67 Table 12. Lexically indexed PARSE FNC(D) -INTO-" HCOMP i. [ DP D [ NP N]] PARSE FNC(D) -INTO- " HCOMP PARSE- " CONJ NONREC" PARSE- " $ a. (" max D (" min N)) * b. (! D (" max N)) *! * ii. [ DP D [ NP [ AP A] N] a. (" max D (" min A)) (" max N) *! $ b. (! D (" max A) (" max N) * iii. [ ConjP Conj [ XP Conjunct2 ]] $ a. (" max Conj (" min Conjunct2)) * b. (! Conj (" max Conjunct2)) *! * iv. [ PP P [ XP X]] a. (" max P (" min X)) *! $ b. (! P (" max X)) * The candidate sets in (i) and (ii) illustrate the effect of lexically indexed PARSE FNC(D) - INTO-" HCOMP for the determiner data, requiring the affixal candidate for the DN sequence, but allowing the free clitic structure for DA to fall out of the general NONRECURSIVITY" >> PARSE-" ranking. The candidate set in (iii) shows the effect of lexically indexed PARSE-" CONJ , forcing an affixal structure for all conjunctions regardless of the host. 68 Finally, the candidate set in (iv) presents the behavior of the rest of the function words of Spanish, prosodifying as free clitics. The analysis pursued here successfully accounts for the prosodic structures of not only Spanish determiners, but also the other function words in the language. The syntactic head-complement relation between clitic and host is identified in a syntax- prosody mapping constraint lexically indexed to determiners, while the behavior of conjunctions and other functional items is achieved through lexical indexing and constraint ranking of well-established prosodic mapping and well-formedness constraints. As I turn to next, an analysis of the feminine el prosodification that does not access the syntactic head-complement relation but is instead purely MATCH-driven does not fare so well in accounting for all function words in the language. 2.2.2.4.2 Structural alternative with no syntactic head-complement relation With a structural analysis of the syntax-prosody mapping that does not have access to the internal syntactic relation between lexical items, different elements are required in order to produce the necessary clitic structures of the feminine el. Ultimately, such an analysis fails to account for the behavior of other function words. The two main ways it differs from the head-complement relation approach include crucial dependence on a syntactic head-raising structure of the DP and a mapping constraint matching the syntactic XP and prosodic Pwd levels. Since the lexical head-complement relation connects the D and N regardless of additional structure, the debate between the NP-raising and N-raising syntactic accounts 69 of the DP does not affect the interface analysis. For a purely structural analysis, however, an N-raising account is necessary, because it produces a structural difference between the DN and DA word orders. With head N-raising, the noun moves out of its NP into the head of a higher functional projection complement to D. This allows it to move across adjectives to produce the postnominal adjective word order (2-39a). Prenominal adjectives are generated as the specifier to a higher FP, above the noun’s final position (2-39b). The resulting DNA and DAN structures are differentiated by the number and type of XP boundaries intervening between DN and DA. In the DN sequence, only a single FP boundary intervenes, whereas in the DA sequence, an FP and a lexical AP boundary intervene. (2-39) a. [ DP D [ FP N t [ NP [ AP A] t]]] b. [ DP D [ FP [ AP A] F, [ FP N t [ NP t]]]] An NP-raising account of the DP involves the entire NP moving to the specifier position of the higher FP, with a resulting DN structure exactly mirroring DA (2-40). (2-40) a. [ DP D [ FP [ NP N] F, [ FP [ AP A] F,]]] b. [ DP D [ FP [ AP A] F, [ NP N]]]] Taking the postmovement N-raising structures as the input for the syntax-prosody mapping, the DN and DA syntactic inputs maintain a structural difference between the 70 clitic-host sequences. The next challenge is to translate that structural difference through a mapping constraint that will produce the affixal and free clitic structures. One way to do this is to utilize a MATCH mapping constraint requiring a lexical syntactic XP to correspond to a maximal Pwd (2-41). (2-41) MATCH(LEXP, PWD MAX ) – for every lexical XP in the syntactic structure that dominates terminal node(s) %, assign a violation if there is not a maximal prosodic word (Pwd max ) in the prosodic structure that dominates all and only the phonological exponent(s) of %. Matching the syntactic XP and prosodic " categories in this way is significant departure from the category consistent matching proposed by Selkirk (2009, 2011), but is necessary in order to produce the prosodic clitic structures attested in Spanish. Without reference to higher levels of the syntactic structure, " formation can have no influence from syntactic differences. At the word level of the syntactic terminal string (X 0 ), DA and DN are the same. Adjectives and nouns are both lexical words, while D is a functional word. In the structural approach, the difference between the sequences lies in the syntactic structure above the X 0 level. Thus, " formation must be sensitive to higher- level syntactic differences. For the feminine el, MATCH(LEXP, PWD MAX ) requires the A in the DA sequence to be contained in a maximal Pwd, forcing the clitic D outside the Pwd in a free clitic structure. Since the N has head-raised outside of the lexical NP into a functional 71 projection, the constraint does not require a maximal Pwd around N in DN. Ranked above PARSE-" and NONRECURSIVITY", the constraint correctly predicts the affixal and free clitic structures of both sequences. Table 13 illustrates. Table 13. MATCH(LEXP, PWD MAX ) differentiating DNA and DAN i. [ DP D [ FP N t [ NP [ AP A] t]]] MATCH(LEXP, PWD MAX ) PARSE-" NONRECURS" $ a. (" max D (" min N)) * b. (! D (" max N)) *! ii. [ DP D [ FP [ AP A] N t [ NP t]]] a. (" max D (" min A)) *! * $ b. (! D (" max A)) * Candidate (ia) uses the recursive affixal clitic structure for DN. It vacuously satisfies MATCH(LEXP, PWD MAX ) because the functional projection preceding N is not lexical, and therefore no maximal Pwd boundary is required between D and N. Candidate (iia) on the other hand, violates this constraint in the affixal clitic structure, because the lexical AP boundary is not aligned with a Pwd max boundary. Instead, the free clitic structure in (iib) wins out for the DA word order. While the above MATCH(LEXP, PWD MAX ) analysis serves the purpose of differentiating DN and DA in the feminine el data, it faces a few challenges. First, the cross-category matching of a syntactic XP and prosodic Pwd is questionable theoretically, and appears to predict Pwd max replacement for PPh structure, the prosodic 72 category normally matched to the syntactic XP level. Specification to the Pwd max level requires nonrecursive prosodic structure, in direct opposition to the recursive tendencies noted in recent research on the prosodic mapping (Itô & Mester 2006, 2008, 2010; Selkirk 2009, 2011; Elfner 2011). If the constraint is top-ranked in a language, it predicts the possibility of overly long Pwds, widely different prosodic outputs for head-raising languages versus non head-raising languages, and different function word cliticization patterns depending on the length and recursivity of syntactic lexical XPs. For example, in a V-to-T language like Spanish or French, top-ranked MATCH(LEXP, PWD MAX ) would predict proclitic function word cliticization in short VPs, but leave the possibility for enclitic function word cliticization in longer VPs (compare the Spanish examples in Tables 14 and 15). Table 14. VP short: [ TP va t [ VP t [ PP a [ DP la [ FP tienda]]]]] ‘he/she goes to the store’ [ TP va [ VP t [ PP a [ DP la [ FP tienda]]]] MATCH(LEXP, PWD MAX ) a. (" max va a la tienda) VP! b. (" max va a la) (" max tienda) VP! $ c. (" max va) (" max a la tienda) d. (" max va) (" max a la) (" max tienda) VP! In Table 14, only one candidate obeys the MATCH(LEXP) constraint perfectly, prosodifying all words in the lower VP—a la tienda—in a single Pwd max (c). Since the function words a la are included in the lexical VP, the constraint favors proclitic 73 attachment. All other candidates have an imperfect match for the VP word sequence, either including them in a larger Pwd max with the raised verb (a), or breaking them up into separate Pwd structures (b and d). Table 15. VP long: [ TP compró t [ VP t[ DP el [ FP sombrero [ AP grande]]][ AP rápidamente]]] ‘he/she bought the big hat quickly’ [ TP compró [ VP t [ DP el [ FP sombrero [ AP grande]]] [ AP rápidamente]]] MATCH(LEXP, PWD MAX ) a. (" max compró-el-sombrero-grande-rápidamente) VP, AP grande , AP rápido ! b. (" max compró-el)(" max sombrero grande) (" max rápidamente) VP, AP grande ! c. (" max compró el sombrero grande) (" max rápidamente) VP, AP grande ! d. (" max compró) (" max el sombrero grande) (" max rápidamente) VP, AP grande ! e. (" max compró) (" max el sombrero grande rápidamente) AP grande , AP rápido ! $ f. (" max compró) (" max el sombrero) (" max grande) (" max rápidamente) VP g. (" max compró) (" max el sombrero) (" max grande rápidamente) VP, AP grande , AP rápido ! $h. (" max compró el)(" max sombrero) (" max grande) (" max rápidamente) VP 74 In the extended VP of Table 15, there is no Pwd max structure that obeys the MATCH(LEXP, PWD MAX ) constraint completely. There are several lexical XPs in a nested syntactic structure in the input that the constraint requires to be matched to their own Pwd max structures in the output. Instead, there are two options that minimally violate the constraint. Candidate (f), with proclitic attachment of the function word el, provides separate Pwd max constituents for two of the three lexical XPs: the AP grande and the AP rapidamente. Only the VP, including the words el sombrero grande rapidamente is not matched to a Pwd max . Candidate (h) also violates the constraint by failing to match the VP to a Pwd max , but does so with enclitic attachment of the function word el to the previous word compró. Both candidates tie on the MATCH(LEXP) constraint. Which one would surface in a language would be due to the interaction of lower-ranked constraints in the grammar, but both are possible outcomes for the longer VP. This stands in contrast to the shorter VP in Table 14, in which only the proclitic attachment is possible. The MATCH(LEXP, PWD MAX ) constraint thus predicts the possibility of a language changing from proclisis to enclisis in the same function word, depending on the length of the syntactic phrase. While these theoretical predictions might be attested in future language study, a more serious problem for the structural approach is the difficulty accounting for the various behaviors of all function words in Spanish, not just determiners. In order to differentiate determiners, conjunction and the remainder of the function words in Spanish, the structural approach requires lexical indexing of PARSE-" to both 75 conjunctions and determiners, and of the MATCH(LEXP, PWD MAX ) constraint to the DP. This latter direction encounters problems theoretically and empirically. First, the general divide of free clitics and affixal clitics requires lexical indexing of PARSE-". Since the constraint PARSE-" is what drives affixal cliticization for both determiners and conjunction in this analysis, it must be indexed to both in order to differentiate them from the free clitic structure of the other function words. Table 16 demonstrates. Table 16. Lexically indexed PARSE-" D,CONJ i. [ DP D [ FP N]… PARSE- " D,CONJ NONRECURSIVITY" PARSE-" $ a. (" max D (" min N)) * b. (! D (" max N)) *! * ii. [ ConjP Conj [ XP Conjunct2 ] $ a. (" max Conj (" min Conjunct2)) * b. (! Conj (" max Conjunct2)) *! * iii. [ PP P [ XP X]] a. (" max P (" min X)) *! $ b. (! P (" max X)) * Candidate sets (i) and (ii) illustrate the application of the lexically indexed PARSE-" D,CONJ constraint to determiners and conjunctions, forcing an affixal clitic structure in violation 76 of NONRECURSIVITY". The candidate set in (iii) on the other hand, is not a determiner or conjunction, and bypasses the lexically indexed PARSE-" requirement. Instead, the nonrecursive free clitic structure is preferred with prepositions and other function words. In addition, since the determiner is the only function word that shows a sensitivity to the type of following host—whether it is within a lexical or functional XP— MATCH(LEXP, PWD MAX ) must also lexically indexed, but only to the DP. The constraint interaction with lexically indexed MATCH(LEXP, PWD MAX ) DP is shown in Table 17. Table 17. Lexically indexed MATCH(LEXP, PWD MAX ) DP i. [ DP D [ FP N]… MATCH(LEXP, PWD MAX ) DP PARSE- " D,CONJ NONRECURS" PARSE-" $ a. (" max D (" min N)) * b. (! D (" max N)) *! * ii. [ DP D [ FP [ AP A]… a. (" max D (" min A)) *! * $ b. (! D (" max A)) * * iii. [ ConjP Conj [ XP Conjunct2 ] $ a. (" max Conj (" min Conjunct2)) * b. (! Conj (" max Conjunct2)) * * Candidate sets (i) and (ii) reproduce the DN and DA distinction of prosodic affixal and free clitics, with MATCH(LEXP, PWD MAX ) DP enforcing the free clitic construction in DA. 77 With lexical indexation to the determiner pattern specifically, this MATCH constraint does not affect conjunction in (iii). Thus, conjunction maintains a consistent prosodification, rather than changing depending on the following syntactic structure. The ranking in Table 17 appears to function, but DP indexation of MATCH(LEXP, PWD MAX ) is problematic. For one, it does not make sense that a constraint that only references lexical projections be indexed to a functional item like D or functional projection like the DP. 31 Indexing to AP would more closely reflect the constraint itself, but as argued in §2.2.1, such a move reduces the entire pattern to an ad hoc specification of nouns and adjectives. This raises the issue of the difference between lexical category indexing (like adjectives vs. nouns) and functional category indexing (like determiners and conjunctions). I assume that the fundamental difference between lexical items and functional items is whether they belong to an open or closed class in the lexicon. Function words must be specified for use in a small, specific set. In terms borrowed from Distributed Morphology, when the grammar is faced with a lexical category it has a broad range of phonological forms in the lexicon to choose from for insertion and of locations where it may be inserted. As argued by Borer (2005), the same word can easily be used as an adjective, noun or verb, depending on where it is inserted into the grammar. For a recent English example, the word Google is used in English as a noun and verb (Did you look in Google N ? Did you google V it?), and with the addition of minimal 31 The way the constraint must be constructed—with reference to the XP level of the syntactic structure—requires lexical indexation that references XP, but lexical specification to the XP level is also problematic. If the lexicon is assumed to primarily contain words and not phrases, even while some words’ categories may be available for indexation, as argued in the following text, XP is not a lexical category. 78 morphological material, one could imagine an adjective (That search was googlific A !). An example of a noun and adjective with the same phonological form is the word good (the greater good N , the good A man N ). On the other hand, the grammar does not have a great deal of choice in the insertion of function items, including between different functional categories. For example, the only definite article determiner choice in English is the, but no amount of morphological adjustment will allow the use of the word and as a determiner. According to this difference between lexical words like adjective and nouns, and functional words like determiners and conjunctions, I propose that since function words must be already specified as such in the lexicon—and additionally specified to the function word category—such specification is also available to phonological constraints. Since lexical words do not need to be specified in the same way in the lexicon, it is preferable to avoid such specification of them in phonological constraints. Returning to the constraint MATCH(LEXP, PWD MAX ), specification to the DP (or even AP, if such an analysis were pursued despite the previous argument) will not always prevent other function words like conjunction from displaying sensitivity to the lexical/functional XP divide represented by the MATCH constraint. For example, two DPs may be conjoined with adjectival modifications and an empty D (2-42). (2-42) [ ConjP [ DP osados A marineros N ] [e Conj [ DP inteligentes A profesores N ]]] ‘bold sailors and intelligent professors’ 79 As shown, although the coordinator is immediately adjacent to the AP inteligentes, it still uses the hiatus avoiding e allomorph, instead of y [i]. But, the DP-indexed MATCH constraint will require the AP inteligentes to be matched to a Pwd max , predicting the unattested free clitic structure (and no hiatus avoidance), as shown in Table 18. Table 18. Failure of MATCH(LEXP, PWD MAX ) DP [ ConjP [ DP osados A marineros N ] [y Conj [ DP inteligentes A profesores N ]]] MATCH(LEXP, PWD MAX ) DP PARSE-" D,CONJ NONREC" ! a. (! y (" max inteligentes) * b. (" max e (" min inteligentes) *! * Candidate (a) obeys MATCH(LEXP, PWD MAX ) DP by parsing the prenominal adjective of the second conjunct DP in a maximal Pwd. This forces a free clitic structure between the coordinator y ([i]) and the following adjective, which corresponds with unattested maintenance of vowel hiatus. The affixal structure that should win, candidate (b), is ruled out by the lexically indexed Match constraint. 2.2.2.4.3 Summary of comparison with structural approach It is clear that a purely structural approach to the syntax-prosody mapping of the feminine el encounters a number of problems, both at the theoretical level and in accounting for the data. While future research may resolve these issues, the current comparison finds the head-complement relation approach much more advantageous. The 80 lexical head-complement relation straightforwardly accounts for the behavior of determiners in the feminine el, with D prosodically attaching at the Pwd level to N as an affixal clitic, but attaching to A at the PPh level as a free clitic. These two prosodic structures—affixal and free clitic adjunction—are the starting point for phonological differentiation of the feminine el. In the affixal DNA structure, the D and N are included in one maximal Pwd and trigger use of the feminine el determiner. In the free DAN structure, D is outside of the maximal Pwd, attaching instead at the PPh level, and determiner alternation is blocked. The next section examines these phonological effects in detail. 2.3 Prosody to phonology The different clitic structures of DNA and DAN sequences affect morpho- phonological determiner alternation in the Spanish feminine el. In addition to sensitivity to the prosodic structure, the phenomenon is triggered by phonological restrictions on vowel hiatus. The phonological environment of determiner alternation is highly constrained, depending on the quality of the vowels in hiatus, and the presence or absence of stress. Recall that the forbidden vowel sequence a-á contains identical vowels that differ in stress. In the OT analysis that I will propose, Minimum Distance constraints on perceptual contrast address the influence of vowel quality, in the approach of Dispersion Theory (Flemming 1995, 2004). The prosodic Pwd max serves as a domain in constraints to produce the different phonological treatment of affixal and free clitic structures. The stress condition is accounted for through markedness augmentation of a positional 81 markedness context (Smith 2005) specific to stressed vowels. The markedness conditions on hiatus, stress and prosodic structure interact with morpho-phonological faithfulness to produce the full feminine el pattern. 2.3.1 Minimum Distance on adjacent vowels Use of the Spanish feminine el determiner prevents a sequence of two adjacent identical vowels. Non-identical vowels are tolerated, but identical a-a is not. The pattern demonstrates a particular instance of hiatus restriction that is sensitive to the quality of the vowels in hiatus. To account for vowel quality in hiatus, I turn to Dispersion Theory (Flemming 1995, 2004). As introduced in Chapter 1, Dispersion Theory was first proposed in Flemming (1995) as a framework for evaluating phonological contrast, and has since been further applied to vowel inventories (Flemming 2004), vowel reduction processes (Padgett 2004, Flemming 2005), vowel harmony (Ní Chiosáin & Padgett 2001), palatalization and nasalization contrasts (Padgett 1997, Padgett 2001), diphthongs or intravocalic sequences (Casali 1998, Sands 2004), sound change (Sanders 2003), and contrast preservation (Robles-Puente 2010). Here I extend it to intervocalic sequences, or vowel hiatus 32 . The specific set of constraints within Dispersion Theory that I expand on is MINIMUM DISTANCE, the family of constraints designed to maximize the perceptual 32 I have separated vowel hiatus from diphthongization because of the concern that diphthong vowels become semi-glides and are therefore fundamentally different from full vowels on acoustic and articulatory levels (Padgett 2008). For this reason, I leave diphthongs as a separate phonological issue. 82 distinctiveness of contrasts along a particular dimension. The farther apart a set of contrasting segments are along a dimensional scale, the easier they are to perceive as contrasting, while the closer they are along the dimension (the more similar they are), the more perceptually confusable they will be. The general constraint formula for this set of contraints is presented in (2-43), following Flemming (2004). (2-43 MINIMUMDISTANCE=Dimension:n –enforce a minimum scalar distance n along the given dimension. MINDIST constraints are in a fixed ranking, “to encode the fact that auditory distinctiveness should be maximised” (Flemming 2004, p. 239). That is, maximally distant contrastive vowel pairs will always be preferred over less distant, and less contrastive pairs. 33 (2-44) MINDIST=DIM:n >> MINDIST=DIM:n+1 >> MINDIST=DIM:n+2, etc 34 . Applied to sequences of adjacent vowels in hiatus, I propose to formulate the proposed MINDIST constraints as follows in (2-45). 33 Another option for encoding the preference for more contrastive vowels over less contrastive is to use SPACE constraints, following Padgett (2003a,b, 2004). With this approach, experimentation identifies a specific contrast as the minimal requirement for the language, and all other vowel pairs are compared to the minimal contrast SPACE requirement to evaluate violations. Future experimentation may provide the necessary measurement evidence for this SPACE approach in Spanish, but for the purposes of the current analysis I will use the more abstract MINDISTVV formula. 34 These could also be formulated in a stringent ranking, in the manner of de Lacy (2002), instead of postulating a universally fixed ranking. Either method is equal for the purposes of this analysis. 83 (2-45) MINIMUMDISTANCEVV=Dimension:n –enforce a minimum scalar distance n along the given dimension, for any two vowels in hiatus. For example, utilizing a perceptual scale of the F1 acoustic dimension shown in (2-46), the MINDISTVV constraints would follow in (2-47). The vowels [i] and [u] have a low F1 value, while the vowel [a] has the highest F1 value. The mid vowels fall at F1 values between the two extremes. Vowels may be compared to each other according to how distant they are along the F1 scale. For example, the pair [a-i] is described as having an F1 distance of 6, as the [a] is six places away from [i] on the scale, while the pair [a-e] is only separated by three. (2-46) 1 2 3 4 5 6 7 i ɪ e̝ e - æ a u ʊ o̝ o ɐ ɑ " ɔ 84 (2-47) MINDISTVV=F1:1 – For adjacent vowels V 1 and V 2 , the two vowels must differ in F1 values by at least 1. (That is, the vowels must not be at the same F1 level, such as a-a or e-e.) MINDISTVV=F1:2 – For adjacent vowels V 1 and V 2 , the two vowels must differ in F1 values by at least 2. (That is, the vowels must be at least as contrastive as a- -, for example, which excludes sequences such as a-æ and a-a.) MINDISTVV=F1:3 – For adjacent vowels V 1 and V 2 , the two vowels must differ in F1 values by at least 3. MINDISTVV=F1:4, etc. A minimum distance scale on adjacent vowels is perceptually logical. All other things being equal, it is more difficult to distinguish two adjacent vowels when they are the same (a-a, for example) than when they differ (a-i). The greater the difference between the vowels, the easier it is to distinguish one from the other. A phonetic study by Polka and Bohn (2003) provides evidence for vowels on the periphery of the vowel space as being perceptual anchors for vowel pair sequences, increasing perceptibility of the vowel sequence. Schwartz et al. (2005) re-interpret the experimental data in terms of Dispersion-Focalisation Theory (Schwartz et al. 1997), which identifies certain peripheral vowels including a, i, and u as “focal vowels”, defined by convergence of two formants (F1 and F2, F2 and F3, F3 and F4, etc.), and predicted to be maximally perceptible. In a phonetic study examining the production of Spanish phonological vowel hiatus, hiatus 85 resolution 35 was discovered more in vowel pairs between a and a mid vowel (e, o) and a and itself than in pairs between a and i (Alba 2006), suggesting the relevance of vowel distinctions along the F1 dimension. 36 The F2 dimension of vowel hiatus appears to be active in Spanish as well. Alba (2006) found a slightly greater preference for hiatus resolution between Spanish a-e than a-o, indicating that there could be increased perceptual contrast due to rounding or backness. In addition, Clements and Ridouane (2006) showed that the front/back differentiation may be magnified in perception, and listeners will pick up quite easily on the distinction in order to differentiate vowels. The determiner-host sequences of the feminine el do not provide direct evidence for the more fine-grained front/back distinction, since all determiner vowel hiatus pairs are with a, but word internal hiatus in Spanish shows tolerance of pairs that are identical along the F1 dimension, but non- identical along the F2 dimension (2-48). 37 (2-48) po.e.ma ‘poem’ le.on ‘lion’ 35 Defined in Alba (2006) as vowel production that sounded different, in some way, from the canonical underlying phonological vowels. 36 As discussed further in Chapter 3, although the feminine el (and conjunction) are the only phonological patterns restricting hiatus in Spanish, that does not mean that all vowel hiatus sequences are produced with acoustic qualities exactly matching those of singleton vowels. This could be considered a distinction between phonological hiatus restrictions and “phonetic” hiatus restrictions. 37 Though see §2.3.5 for discussion of how the MINDISTVV constraints may apply to a tendency for diphthongization of such sequences. 86 Given this data, Spanish vowel sequence patterns are most accurately described as a restriction of identical hiatus, on both the F1 and F2 dimensions. Utilizing multiple dimensions in the formulation of MINDISTVV could also account for similar effects seen in patterns of diphthong inventories in Casali (1998), who utilized Minimum Distance considerations based on the number of feature differences between the vowels of diphthongs. The scalar relationship of F1 and F2 Spanish vowel placement is shown in (2-49), utilizing the same F1 scale shown in (2-64) and following F2 vowel distinctions as presented in Flemming (2004). Both dimensions are incorporated into the appropriate MINDISTVV constraints, as defined in (2-50). (2-49) Spanish vowels: F1 i u 1 2 3 e o 4 5 6 a 7 F2 6 5 4 3 2 1 87 (2-50) MINDISTVV=F1/F2:1 – For adjacent vowels V 1 and V 2 , the two vowels must differ in F1 or F2 values by at least 1. (That is, the vowels must not be identical, such as a-a 38 . 39 ) In addition to the constraint above, an absolute restriction on all vowel hiatus may be couched in Minimum Distance terms. This constraint requires a minimum distance between adjacent vowels that is impossible to achieve given the limits of the vowel space. Functionally, it behaves the same way as the general NO HIATUS constraint introduced in McCarthy (1993). 40 (2-51) MINDISTVV=F1/F2:7 – For adjacent vowels V 1 and V 2 , the two vowels must differ in F1 or F2 values by at least 7. (That is, do not have any adjacent vowels, as they cannot be sufficiently contrastive.) 38 This constraint will also militate against identical adjacent vowels that are not a, such as i-i, o- o, u-u, etc. While the Spanish DN or DA sequences will not contain these other identical vowels due to the phonological form of the determiner (a-final, or consonant-final), the constraint shows effects against non-a identical sequences elsewhere in the language, such as with conjunction allomorphy, which is discussed later in this chapter. 39 Given the constraints of the Spanish vowel space, the constraint MINDISTVV=F1/F2:3 would also have the effect of only discriminating against a-a, since a-e and a-o obey the minimum distance of 3. If Spanish had the vowel æ, the constraint MINDISTVV=F1/F2:3 would discriminate against the hiatus sequence a-æ as well, since it would only represent a minimum perceptual distance of 1 along the F1 dimension, and 1 along the F2 dimension. 40 While I do not address them in detail in this dissertation, other Romance languages such as French and Italian show vowel hiatus sensitivity at this most restrictive level of Minimum Distance. All forms of hiatus at determiner junctures and other clitics are repaired. Like Catalan, discussed in §4.2, the repair for these languages is deletion of the feminine determiner’s a vowel. For example, French la fille F ‘the girl’, but l’enfant F ‘the baby [female], l’idee F ‘the idea’; and Italian la figlia F ‘the daughter’, but l’amica F ‘the friend [female]’, l’idea F ‘the idea’, l’unità F ‘the unity’. 88 The MINDISTVV constraints that I propose here account for sensitivity to vowel quality. In combination with considerations of the prosodic clitic structure and the augmentation of a stress boundary, they form the basis of the markedness trigger for the Spanish feminine el. 2.3.2 Pwd max and prosodic clitic structure In addition to the Minimum Distance restriction on adjacent vowels, the feminine el determiner is conditioned by the affixal clitic prosodic structure in the DN(A) word order. Sensitivity to the prosodic condition is achieved via constraint specification to the Pwd max unit. Within the maximal Pwd, a border between the clitic and host with vowel hiatus is weak and contains weaker perceptual cues than a consonant-vowel sequence. In order to highlight the vowel transition between clitic and host within this environment and make the word boundary between vowels more perceptually salient, Pwd max is signaled as a domain for applicable constraints. Constraints targeting Pwd max have similarly been used to signal an increased need for onsets at the edge of a prosodic clitic-host structure in English r-insertion of non-rhotic dialects (Itô & Mester 2008), and to signal the domain of vowel copy harmony in Servigliano (Walker 2011). On the other hand, vowels which cross a Pwd max boundary—rather than being in hiatus within the prosodic unit—show increased phonetic duration (see chapter 3). Longer duration is expected to increase perception of the distinction and word boundary 89 between vowels. Increased perception of vowels could then lead to tolerance of phonological hiatus. Similar to Smith’s (2005) Prominence Condition filter on prominence-based positional markedness constraints and with foundation in Hayes’ (1999) Inductive Grounding of constraints, I propose to account for this relationship between phonetic lengthening and phonological hiatus via a filter on the formation of perceptually enhancing sequencing constraints. The filter identifies the prosodic constituents available as domains for MINDISTVV constraints based on their phonetic properties in the language. A constraint like MINDISTVV, which serves to enhance the perception of sequences, may then be restricted to sequences internal to a domain that shows increased boundary salience, because the domain-internal sequences do not enjoy the perceptual enhancements afforded to cross-boundary sequences. (2-52) Perceptual Enhancement Domains: A prosodic domain P is available for specification to perceptual enhancement constraints if it displays phonetic properties that increase constituent boundary salience. For example, the domain P may show a number of phonetic boundary effects, such as final lengthening, pause, or salient pitch or tone. Any of these properties will highlight the prosodic constituent in the acoustic signal, allowing P to be signaled as a domain for applicable constraints on perceptual enhancement. If the domain P does not display phonetic properties that distinguish it from the surrounding structure, then it may not be a domain for constraints like MINDISTVV. This filter distinguishes between prosodic 90 constituents that are predicted to be present in a structure despite a lack of direct phonetic evidence (for example, the minimal or intermediary recursive ! in Selkirk’s 2011 analysis of ChiMwiini) and those that are phonetically marked in a language. Each language may have different phonetic properties for different levels of the prosodic hierarchy. In previous research, this has led to proposals of different prosodic categories for different languages (such as the intermediate and accentual phrase for Japanese in Beckman & Pierrehumbert 1986, in contrast to the single phonological phrase for English, Italian and other languages). In contrast, recent work by Itô and Mester (2006, 2007, 2008, 2010) argues that there is only one prosodic hierarchy of universal categories, but that each language may phonologically or phonetically mark some categories and not others. I propose that the Perceptual Enhancement Domain filter accounts for this crosslinguistic variation, by identifying which prosodic constituents are available for constraint specification. The filter predicts that if a language does not have a phonetic property increasing perceptual salience of the prosodic constituent, then the constituent cannot be targeted by phonological constraints that enhance perception. A perceptually enhancing constraint is one which when obeyed, serves to maximize perceptual salience of the speech signal. For example, a contrast-driven constraint like Minimum Distance maximizes perception of contrasting elements. Likewise, a constraint like ONSET is perceptually enhancing because when obeyed, it augments perception by breaking up the speech signal with consonants, providing a strong transition into a syllable’s vowel. 41 41 The constraints mentioned here are only a small subset of a number of constraints that could be 91 In Spanish, this Perceptual Enhancement Domain filter identifies the Pwd max as a prosodic constituent with phonetically marked boundaries. Vowels at the Pwd max boundary are longer than those that are at a Pwd min or no boundary (see chapter 3). Therefore, Pwd max is available as a domain for MINDISTVV, while Pwd min is not. While within the Pwd max constituent, vowel hiatus is restricted in Spanish, hiatus tolerance across a Pwd max boundary is extended to all vowels, including identical a-a. The Pwd max domain is combined with the MINDIST constraints presented in the previous section to produce the relevant phonological and prosodic conditioning in the Spanish feminine el. (2-53) MINDISTVV(PWD MAX )=F1/F2:1 – For adjacent vowels V 1 and V 2 within a maximal Pwd, the two vowels must differ in F1 values by at least 1. That is, the vowels must not be identical, such as a-a. Fixed ranking: MINDISTVV(PWD MAX )=F1/F2:1>> MINDISTVV(PWD MAX )=F1/F2:2 >>... MINDISTVV(PWD MAX )=F1/F2:7 interpreted as enhancing perception. Future research may further refine the range of constraints to which the filter applies, perhaps building upon concepts such as inductive grounding. I leave this area open for future research. 92 These constraints, which govern vowel quality restrictions on adjacent vowels as well as recognizing the prosodic organization of said vowels, interact with morpho- phonological faithfulness to produce the pattern of the feminine el. 2.3.3 MP Faithfulness In avoiding hiatus, the feminine el pattern violates the canonical morphological- phonological (MP) correspondence between sound forms and morphemes in the Spanish determiner set. This section addresses the interaction between MP correspondence faithfulness and the Minimum Distance markedness constraints laid out in the previous section. The resulting interim constraint ranking will be revised later in !2.3.4 as we consider the additional role of stress in the feminine el pattern. In regard to the morpho-phonological interpretation of Spanish determiners, I follow Harris (1987), Salvá (1988), and Wolf (2008) in analyzing the exceptional article el as morphologically feminine, not masculine. Spanish is a language with obligatory gender agreement, and feminine agreement is displayed throughout the DP with the feminine el, as reviewed in (2-54). (2-54) el águila F blanca F ‘the white eagle’ el hada F madrina F ‘the fairy godmother’ el hacha F afilada F ‘the sharp axe’ 93 With a feminine interpretation of the determiner form el, there is no violation of syntactic gender agreement. In contrast, another pattern of synchronic variation displays evidence of a masculine el determiner with masculine adjective agreement, supporting a feminine interpretation when there is full feminine agreement (see §4.5 for more on the synchronic variant). 42 The use of el with a feminine gender represents a switch in the phonological manifestation of the morphologically feminine determiner. To account for this interaction between morphology and phonology, I turn to developments in OT that propose correspondence relationships between these two aspects of language. Walker and Feng (2004) posit a Ternary Correspondence Model with correspondence at the Phonology-Phonology (PP) level—representing the phonological Input-Output (IO) correspondence of traditional OT—as well as at the levels of Morphology-Phonology (MP) and Morphology-Morphology (MM). Wolf (2008) presents a similar correspondence relation between morphemes and ‘morphs’, the phonological representations of morphemes. Both models suggest complex inputs composed of morphological and phonological components that are in correspondence with morphological and phonological manifestations in the output. These correspondence relations may then be manipulated via constraints. Expanding on Walker and Feng’s (2004) Ternary Correspondence Model, I use a correspondence model of MM and MP input-output relations, schematized in (2-55). 42 Another possible interpretation is that the traditionally masculine el is not actually marked for masculine gender, but represents a morphologically unmarked form: el Ø . If this were the case, then just as if it were fully masculine el M , the use of el Ø with feminine nouns would constitute a violation of syntactic agreement, since it does not match gender with the noun. The only way to obey syntactic agreement is to take the feminine gender in el F . The possibility of syntactic disagreement in this pattern is discussed further in relation to the synchronic variant in §4.5. 94 Note that MM and PP correspondence relations straightforwardly connect input and output units, whether they are morphological units or phonological units. In contrast, MP correspondence connects input and output associations between morphological and phonological units, in effect governing the correspondence of correspondence relations. (2-55) Morphological input Phonological input MM corresp. MP corresp. PP corresp. Morphological output Phonological output Constraints along the MP relation manipulate the relationship between the morpheme and its associated phonological form, for example, a faithfulness constraint requiring a morpheme associated with a specific phonological form in the input to be associated to the same phonological form in the output, as in (2-56). 95 (2-56) IDENT-MP: Let feature % at the morphological level be in correspondence with a phonological form & in the input, the morphological feature % be in correspondence with the output feature %', and the phonological form & be in correspondence with the output phonological form &"#$Assign a violation if the output feature %' is not in correspondence with the output phonological form &". (That is, don’t change the morpho-phonological association of a morpheme and its phonological form. 43 ) In the case of Spanish “feminine el”, IDENT-MP prohibits changing the gender association of the phonological form el from masculine to feminine. To illustrate this, we examine the complex morpho-phonological representation of the input. As in Wolf (2008), the input determiners are composed of the phonological form and an associated morphological feature. Additionally, I follow Tranel and Del Gobbo (2002) in listing allomorph function words, like determiners, in the input. The grammar chooses from the set of determiners, accounting for the fact that when an alternative form is required by the phonology, it is another determiner that is chosen as a replacement, and not an unrelated phonological structure 44 . 43 This constraint assigns different violations from IDENT-PP (IDENT-IO). If the phonological forms & and &' are not identical, that is a violation of IDENT-PP. IDENT-MP does not govern the identity of the input and output phonological forms to each other, but rather the association between the phonological and morphological forms in the input and output. 44 Since the grammar chooses from a set of options in the input, regular PP-IO correspondence is not violated. Were the grammar to repair hiatus by inserting phonological material, for example /la F água F / $ [lar F água F ] then PP correspondence would be violated, and MP correspondence would be obeyed. 96 { } { { } In (2-57), the set of determiners is represented with the phonological level on top and the morphological associations in brackets below. Listed on the right are condensed versions of the inputs, with a morphological index, for ease of representation. (2-57) la , el /{la F , el M }/ [feminine] [masculine] When the underlying association of the phonological form /el/ with the morphological gender [masculine] is switched to the morphological gender [feminine], as is the case with the feminine el, this incurs a violation of IDENT-MP because the morpho- phonological relationship between phonological form and morpheme is no longer the same. (2-58) IDENT-MP violation of el F el el F agua F [feminine] [masculine] Interleaved with the markedness constraint family in (2-53), IDENT-MP blocks alternation of the determiner when adjacent vowels maintain a minimal degree of contrast, and when the vowels occur across a Pwd max boundary instead of within the Pwd max unit, as in sequences with prenominal adjectives. Table 19 illustrates the working of MINDISTVV restrictions on a-a compared with other vowel qualities. 97 Table 19. Activity of MINDISTVV i. {la F , el M } água F MINDISTVV (PWD MAX )=F1/F2:1 IDENT-MP MINDISTVV (PWD MAX )=F1/F2:7 a. [la F água F ] Pwdmax *! * $ b. [el F água F ] Pwdmax * ii. {la F , el M } éra F $ a. [la F éra F ] Pwdmax * b. [el F éra F ] Pwdmax *! iii. {la F , el M } índole F $ a. [la F índole F ] Pwdmax * b. [el F índole F ] Pwdmax *! For the noun água, candidate (ia) produces the unwanted identical vowel hiatus a-a and is ruled out on Minimum Distance, leaving the switched determiner form in (ib) to win out, despite its Faithfulness violation. In era, although candidate (iia) violates Minimum Distance of seven, because IDENT-MP is ranked above MINDISTVV(PWD MAX )=F1/F2:7, the determiner switch does not occur. A similar interaction is observed with the form índole, where the disparate vowel hiatus a-i is tolerated, thanks to the ranking of MP Faith above MINDISTVV(PWD MAX )=F1/F2:7. Note that gender agreement is assumed to be top-ranked in this pattern, preventing the use of a fully masculine el M that does not violate IDENT-MP. See §4.5 for further discussion on gender interpretation and agreement. 98 There is no reason to select a masculine la for masculine nouns, a-initial or otherwise, as demonstrated in Table 20. Switching the determiner to use a morphologically masculine la would create an unwarranted IDENT-MP violation, and also an unwarranted hiatus violation in the case of a-initial nouns. Table 20. No masculine la M i. {la F , el M } ámo M MINDISTVV (PWD MAX )=F1/F2:1 IDENT-MP MINDISTVV (PWD MAX )=F1/F2:7 $ a. [el M ámo] Pwdmax b. [la M ámo] Pwdmax *! * * ii. {la F , el M } éje M $ a. [el M éje] Pwdmax * b. [la M éje] Pwdmax *! In the masculine forms, there is no need to switch to the feminine determiner form, because the masculine one already prevents any hiatus. With an a-initial noun, as in ámo, use of the feminine form in (ib) incurs a violation of Minimum Distance as well as Faithfulness. With non-a-initial nouns, as in candidate (iib), IDENT-MP is violated without any improvement on Minimum Distance markedness. The Pwd max MINDISTVV constraints in (2-53) also address the issue of prenominal adjective blocking. Table 21 demonstrates ranking of the Pwd max domain specified MINDISTVV constraint above IDENT-MP and an unspecified MINDISTVV. 99 Table 21. Activity of Pwd max domain (i) {la F , el M } águila F álta F MINDISTVV (PWD MAX )=F1/ F2:1 IDENT-MP MINDISTVV =F1/F2:1 a. [la F águila F ] Pwdmax [álta F ] Pwdmax *! * $b. [el F águila F ] Pwdmax [álta F ] Pwdmax * (ii) {la F , el M } álta F águila F $a. la F [álta F ] Pwdmax [águila F ] Pwdmax * b. el F [álta F ] Pwdmax [águila F ] Pwdmax *! In candidate (ia) there is an unwanted identical hiatus vowel sequence within a Pwd max . Candidate (ib) repairs hiatus at the expense of IDENT-MP by changing the gender exponence of the consonant-final determiner form. Candidate (iia) also includes the unwanted identical vowel sequence, but it is across a Pwd max boundary, and vacuously satisfies the Pwd max markedness constraint. Even though this candidate also violates the general MINDISTVV=F1/F2:1, this constraint must be lower ranked than MP Faith in order to account for the success of candidate (iia). Candidate (iib) unnecessarily repairs the vowel hiatus through a violation of MP Faith, and is ruled out even though it performs better than (iia) with respect to MINDISTVV=F1/F2:1. 100 2.3.4 Stress and positional markedness As mentioned in §2.1, the Spanish pattern of hiatus resolution is sensitive to word-level primary stress, as well as vowel quality and the syntactic head-complement relation. Compare (2-59a) with an unstressed initial a, (2-59b) with secondary stress on the initial syllable, and (2-59c) with primary stress on the initial syllable. (2-59) a. la amíga (*el amíga) the friend [f] b. la àmistád (*el àmistád) the friendship [f] c. el águila (*la águila) the eagle [f] I propose to account for the stress condition on the production of the feminine el via a positional markedness context for the Minimum Distance markedness constraint. Positional markedness, like positional faithfulness (Beckman 1998) refers to constraints that target a linguistically strong or prominent environment, increasing restrictions on phonological form in such positions (Smith 2000, 2005; de Lacy 2002). Following Smith (2005), positional markedness is limited to constraints that function in an augmentation capacity. That is, the position targeted by a positional markedness constraint is perceptually salient in such a way that obeying the markedness constraint in the “strong” position augments perception. For example, the onset position of a syllable is 101 perceptually strong because it breaks up the sound signal with silences, providing a reset cue for the brain when processing speech (Delgutte 1997, Wright 2004). The markedness constraint ONSET augments syllable perception, and is therefore available for positional markedness versions that target strong positions such as the stressed syllable, as ONSET/#!. Constraints which do not serve an augmentative purpose are not available for positional markedness (Smith 2000, 2005). The constraint ONSET/#! (Smith 2005) initially appears ideal to enforce the use of the marked feminine el determiner only in the event of an onsetless stressed vowel. (2-60) ONSET/#! – for all x, if x is a syllable with primary stress, assign a violation if x does not have an onset. However, ONSET/#! will not serve for the feminine el data, because MINDISTVV will always assign a violation, regardless of stress. As shown in Table 22, this creates the undesired effect of ruling out unstressed la amiga as well as forms with stressed initial non-a vowels. 102 Table 22. Failure of ONSET/#! i. /{la F , el M } águila F / ONSET/#! MINDISTVV(PWD MAX )=F1/F2:1 IDENT-MP a. [la F .á.gui.la F ] Pwdmax *! *! $b. [e.l F á.gui.la F ] Pwdmax * ii. /{la F , el M } amíga F / a. [la F .a.mí.ga F ] Pwdmax *! !b. [e.l F a.mí.ga F ] Pwdmax * iii. {la F , el M } índole F a. [la F índole F ] Pwdmax *! !b. [el F índole F ] Pwdmax * In (i) the constraint ranking functions as desired, ruling out the feminine la F in favor of the feminine el F . Either ONSET/#! or Minimum Distance will rule out the marked form la águila in favor of the feminine el form in (ib). However, in (ii) a problem emerges with the evaluation of the candidates. Although unstressed la amiga in (iia) does not violate ONSET/#!, it still violates Minimum Distance by having two adjacent identical vowels within a maximal Pwd. By the same token, the stressed form la índole in (iiia) is ruled out by ONSET/#!, even though it obeys Minimum Distance. In any ranking of Minimum Distance markedness over the faithfulness constraint IDENT-MP to trigger alternation, the unstressed vowel forms will be ruled out. In any ranking of ONSET/#! over IDENT-MP, the stressed non-a forms will be ruled out. 103 Rather than utilizing a separate ONSET constraint, I incorporate the stressed syllable positional markedness context into the MINDISTVV(PWD MAX ) constraint itself, to capture sensitivity to vowel quality, vowel stress, and the Pwd max domain in a single constraint 45 . The MINDISTVV family is at its core a restriction on perceptual transitions from one segment to another. A vowel that is more perceptually distant from the following vowel provides a better transition for that vowel. The constraint augments perception, and is thus appropriate for the positional markedness requirements laid out by Smith (2005). Specified to the linguistically strong position of a stressed vowel (syllable nucleus), the markedness constraint increases the requirement for a perceptually salient transition into a stressed vowel. Like ONSET/#!, MINDISTV'V(PWD MAX ) increases the markedness restriction of segment transition into a vowel. In the case of ONSET, the transitional segment is specifically a consonant, while with MINDISTVV, it is a vowel. The positional markedness version of MINDISTVV(PWD MAX ) is defined in (2-61) below. (2-61) MINDISTV'V(PWD MAX )=F1/F2:1– Assign a violation for any sequence of adjacent vowels V 1 and V 2 within a maximal Pwd, in which the two vowels do not differ in F1/F2 values by at least 1, and in which the second vowel receives primary stress. 45 The desired effect of requiring all conditions—vowel quality, vowel stress, and the Pwd max domain—in order to trigger use of the feminine el could also be accomplished with Local Conjunction (Smolensky 1993, Itô & Mester 1998, Lubowicz 2002, 2005, and others) of MINDISTVV(PWD MAX ) and ONSET/#!, or a Harmonic Grammar approach (Legendre, Miyata & Smolensky 1990; Potts et al. 2009; Smolensky & Legendre 2006) with those separate constraints. 104 This constraint will assign a violation only if the vowel sequence is of identical vowels, and the second one is stressed. The sequence a-á matches this context precisely. The positionally marked constraint takes the place of MINDISTVV(PWD MAX )=F1/F2:1 in the constraint ranking proposed in §2.3.3, permitting unstressed vowels and stressed non- a forms to maintain the unmarked feminine determiner form. Table 23 illustrates. Table 23. Activity of positionally marked MINDISTV'V i. {la F , el M } águila F MINDISTV'V (PWD MAX )=F1/F2 :1 IDENT-MP MINDISTVV (PWD MAX )=F1/ F2:1 a. [la F .á.gui.la F ] Pwdmax *! * $b. [e.l F á.gui.la F ] Pwdmax * ii. {la F , el M } amíga F $a. [la F .a.mí.ga F ] Pwdmax * b. [e.l F a.mí.ga F ] Pwdmax *! iii. {la F , el M } índole F $ a. [la F índole F ] Pwdmax b. [el F índole F ] Pwdmax *! Candidate (ia) violates the positionally marked MINDISTVV constraint because it contains a sequence of identical adjacent a-á at an affixal clitic-host juncture (within the 105 maximal Pwd) and where the second vowel—the end point of the vocal transition—is stressed. In this, and only this case, is the feminine el used, violating IDENT-MP (as in candidate (ib). The other candidate sets (ii) and (iii) illustrate failure of the feminine el to emerge for unstressed or non-identical vowels. The ranking of positionally marked MINDISTVV over Faith and non-positionally marked MINDISTVV is in keeping with the general ranking schema of positional markedness as presented in Smith (2005) and shown in (2-62). (2-62) Positional Markedness >> Faith >> Markedness MINDISTV'V(PWD MAX )=F1/F2:1 >> IDENT-MP >> MINDISTVV=F1/F2:1 By universal ranking all positionally marked MINDISTV'V constraints over MINDISTVV, this ranking schema predicts that stressed vowels will always be more restrictive in their hiatus tolerance. This prediction is borne out in another Romance hiatus pattern, Catalan determiner vowel deletion, which is analyzed in §4.2. The adjusted total constraint ranking for the Spanish feminine el is provided in (2-63). 106 (2-63) MINDISTV'V(PWD MAX )=F1/F2:1 >> IDENT-MP >> MINDISTVV(PWD MAX )=F1/F2:1, MINDISTVV=F1/F2:1 MINDISTV'V(PWD MAX )=F1/F2:1 IDENT-MP MINDISTVV(PWD MAX )=F1/F2:1 MINDISTVV=F1/F2:1 2.3.5 Conjunction As mentioned in §2.2.2.4, Spanish conjunction is also sensitive to vowel contrast in hiatus, lending further support to the MINDISTVV(PWD MAX ) approach. In conjunction, the identical vowel hiatus forms that are avoided are i-i and o-o. (The hiatus data are reviewed briefly in 2-64). (2-64) impostores [i] engaños ‘imposters and tricks’ engaños [e]/*[i] impostores ‘tricks and imposters’ uno [o] dos ‘one or two’ uno [u]/*[o] otro ‘one or (the) other’ As argued previously, the conjunction-Pwd sequences are analyzed as affixal clitics, attaching within the maximal Pwd of the following word. Allomorphy avoids a 107 sequence of identical vowels within the Pwd max (2-65a). This prosodification is particularly relevant in cases where identical adjacent vowel hiatus is allowed between the preceding word and the conjunction, which do not prosodify together (2-65b). (2-65) a. (" max engaños) (" max [e] (" min impostores)) b. (" max uno) (" max [o] (" min dos)) For the phonological analysis, the same MINDISTVV level sensitivity within Pwd max that we saw in the feminine el is active in conjunction, though without stress sensitivity. Conjunction allomorphy applies equally whether the following word begins with a stressed or unstressed identical vowel (2-66). (2-66) hidalgos [i] caballeros ‘noblemen and gentlemen’ caballeros [e] [i]dálgos ‘gentlemen and noblemen’ riñónes [e] [í]gados ‘kidneys and livers’ oréjas [o] naríces ‘ears or noses’ naríces [u] oréjas ‘noses or ears’ oréjas [u] ójos ‘ears or eyes’ To achieve sensitivity regardless of stress, I use the non-positionally marked constraint MINDISTVV(PWD MAX )=F1/F2:1. Rather than repair via changes to the morpho- phonological correspondence, conjunction simply changes the quality of the vowel to 108 satisfy the non-identical requirement of Minimum Distance. This repair violates the constraint IDENT-IO[HI], which militates against changing the vowel feature [high] specification from the input to output. When /o/ is realized as [u], the vowel changes from [-hi] to [+hi], and when /i/ is realized as [e], the vowel changes from [+hi] to [-hi]. 46 MINDISTVV(PWD MAX )=F1/F2:1 is ranked above IDENT-IO[HI] to ensure that the conjunction vowel only changes when it would otherwise produce an identical vowel sequence within a maximal Pwd. Table 24. MINDISTVV with conjunction i. /uno o otro/ MINDISTVV (PWD MAX )=F1/ F2:1 IDENT-IO[HI] MINDISTVV=F1/F 2:1 a. uno (o otro " max ) *! ** $ b. uno (u otro " max ) * ii. /uno o dos/ $ a. uno (o dos " max ) * b. uno (u dos " max ) *! 46 In this analysis, I assume that conjunction allomorphy repair is purely phonological, instead of a choice of two listed allomorphs in the input (as for the feminine el). Unlike the form el, which occurs as a masculine determiner elsewhere in the language, the allomorph forms [e] and [u] do not occur independently of conjunction. They do not have existing independent morphological indexing, and their phonological forms are fully predictable by constraint ranking. 109 Candidate (ia) retains the regular o form of the conjunction, but in doing so violates general MINDISTVV requirements at the sequence of uno-o and also o-otro. The violation from o-otro is the one that violates the domain restricted constraint, because it occurs within a maximal Pwd. This is the violation that rules out the candidate. Candidate (ib) wins by changing the o to u and violating IDENT-IO[HI]. In (ii), although candidate (a) still violates Minimum Distance requirements at the uno-o juncture, it does not occur within a Pwd max , and therefore passes the domain-specified constraint unrestricted. Candidate (iib) loses on an unnecessary faithfulness violation of IDENT-IO[HI]. Combined with the MINDISTVV constraint ranking obtained for the feminine el, the Spanish conjunction data rounds out the typology of MINDISTVV effects with stress, the Pwd max domain, and general hiatus. The full tableau is shown in Table 25. 110 111 In Table 25, all three distinctions are made between general MINDISTVV, MINDISTVV(PWD MAX ) and MINDISTV'V(PWD MAX ). Candidates (ia) and (ib) demonstrate the feminine el pattern, with violation of MINDISTV'V(PWD MAX ) allowing the feminine el determiner form to surface over the regular feminine la. Candidates (iia) and (iib), on the other hand, show the failure of the feminine el to appear if only MINDISTVV(PWD MAX ) is violated, instead of the positionally marked stressed vowel version. 47 The candidate sets in (iii) and (iv) illustrate the conjunction pattern, where identical hiatus within the maximal Pwd violates MINDISTVV(PWD MAX ) and allows the conjunction to appear with a shift in the vowel, away from identical hiatus (iii), but identical hiatus across a Pwd max boundary does not violate Pwd max markedness, only general MINDISTVV, and is allowed to surface with the regular form of the conjunction (iv). As noted above, conjunction allomorphy is not sensitive to stress of the following vowel, but always alternates whether or not the vowel is stressed. This effect will still fall out of the constraint ranking given, as seen in Table 26. 47 An additional candidate for the feminine el hiatus avoidance might be one which changes the quality of the determiner vowel in the same way that conjunction does: li amiga, for example. This candidate could be ruled out by a constraint IDENT[LOW], which is violated when [+low] /a/ is raised, but not violated when [-low] /i/ and /o/ lower and raise to [e] and [u]. It could also be ruled out by a constraint enforcing an MP relationship between gender and morphological vowel endings, building on the canonical –a F categorization in Spanish. The relationship between gender and morphological affixes is further discussed in §2.4 and §4.4. 112 Table 26. Conjunction and stress hijos y hijas ‘sons and daughters’ MINDISTV'V (PWD MAX )=F1/ F2:1 IDENT-MP MINDISTVV (PWD MAX )=F1/F 2:1 IDENT- IO[HI] a. hijos ([i í]jas ! max ) *! * " b. hijos ([e í]jas ! max ) * Candidate (a) has a sequence of identical V'V in hiatus within a maximal Pwd, violating the highest ranked Minimum Distance constraint as well as lower ranked non-positionally marked MINDISTVV(PWD MAX ). In the case of conjunction, either MinDist violation will trigger use of the conjunction allomorph. Spanish conjunction allomorphy pattern provides further support for a Minimum Distance perspective in comparison to an Obligatory Contour Principle (OCP)[feature] account of Spanish. For example, a constraint OCP[+low] 48 would forbid the adjacent low vowel sequence a-a in the Spanish determiners, but does not account for the similar restriction of identical forms of hiatus i-i and o-o in conjunctions. MINDISTVV, by requiring perceptual contrast along a relative scale, is not restricted to feature 48 The OCP has a long history in linguistic research, originating in Goldsmith (1976), drawing on Leben (1973), and utilized by many researchers since to account for a wide range of adjacency restrictions. OCP vowel adjacency has specifically been proposed in Antilla (2002). 113 specification. It simply requires that a sequence of adjacent vowels be sufficiently contrastive, regardless of the particular vowel in question, be it a, i or o. 49 In summary, conjunction allomorphy in Spanish supports the Minimum Distance restrictions on identical adjacent vowels. It is again confined to the maximal Pwd, a prosodic context that serves to differentiate between sequences that appear unrestricted in the grammar—such as the hiatus between the conjunction and the previous word—and those that are forbidden, despite identical phonological contexts. 2.3.6 Discussion on the Pwd max domain and positional markedness In the analysis I provide here, the prosodic unit Pwd max is a domain of application for Minimum Distance markedness, while the stressed vowel is a positional markedness position. This allows both the stress requirement and prosodic clitic structure requirement to be included in the same constraint, without recourse to constraint conjunction or weighting. However, some discussion is warranted as to how the distinction between a domain and positional markedness context is appropriate for these two requirements. A positional markedness context is by definition a prominent position, singled out by the grammar for augmentation (Smith 2005). The domain of a constraint is not necessarily prominent, but only delimits the area of application. A stressed vowel is uncontroversially prominent. But what about Pwd max ? When referring to prosodic 49 An OCP analysis combining features, such as Padgett’s (2001) [VPlace] feature class that combines height and vowel color, could serve in place of Minimum Distance. While this approach may serve for the feminine el data, it misses the connection between the Spanish pattern and hiatus restrictions in other languages like Catalan, which allows maximally contrastive hiatus but forbids others (examined further in Chapter 4). The Minimum Distance approach unifies patterns of hiatus restriction along a continuum of contrast. 114 boundaries, there is a clear hierarchy of prominence, such that higher prosodic boundaries such as IP and PPh are more salient to the listener—strong boundary tone cues, pauses, pre-boundary lengthening—than lower level prosodic boundaries such as Pwd and clitic. But when referring to the internal span of a constituent, it would be difficult to argue that the maximal Pwd is more or less prominent than another prosodic constituent. Phrases as a whole are not more prominent than words. Words are not more prominent than syllables. As an internal domain, delimiting a span of segments belonging to a constituent, the Pwd max context is thus most appropriate as a constraint domain instead of a prominent positional markedness position. We could also conceive of the onset boundary of Pwd max as a context for positional markedness. As mentioned above, prosodic boundaries fall into a hierarchy of prominence, such that higher prosodic boundaries are often more prominent, with stronger phonetic cues, than lower level prosodic boundaries. For example, IP boundaries in Spanish are associated with pauses, final lengthening, and a pre-boundary pitch contour. In comparison, PPh boundaries may or may not receive a pause, have less final lengthening (Rao 2010), and have a prominent lexical pitch accent rather than an additional pitch contour associated with the boundary. Pwd boundaries show even less phonetic cues, with no pause whatsoever, and only the presence of pitch accents on stressed syllables rather than additional prominence or word contour. Given these comparisons, it is not surprising that Pwd max and Pwd min boundaries also show prominence distinctions at the phonetic level (as will be argued in Chapter 3). In this respect, the Pwd max boundary could be considered more prominent than the Pwd min (or 115 foot, syllable, etc.) boundary. As a prominent position, the Pwd max boundary could be used as a context for positional markedness, such that markedness constraints would target the Pwd max boundary for perceptual augmentation. However, an analysis of positional markedness at the Pwd max boundary does not reflect the Spanish feminine el data. Vowel hiatus is more, not less tolerated, at a Pwd max boundary in Spanish, with the determiner-adjective juncture tolerating all hiatus, including stressed a-á. Were the Pwd max boundary in effect as a positional markedness context, we would expect to find increased sensitivity to vowel hiatus markedness at this juncture. That is, we would expect precisely the opposite pattern as what is found in Spanish determiners: more hiatus contexts (and less dispersed ones) tolerated at a Pwd min boundary, and less hiatus contexts (and more dispersed ones) required at a Pwd max boundary. The Pwd max boundary as a positional markedness context with the MINDISTVV constraints would predict the possibility of a language with this pattern. While it is clear that Spanish is not best analyzed with Pwd max boundary as a positional markedness context, future research in other languages may support this theoretical possibility. One of the benefits to utilizing constraint domain specification and positional markedness with the MINDISTVV family, rather than proposing a single “stressed Pwd max vowel” identification is the ease with which these contexts—vowel quality, syntactic clitic attachment sensitivity, and stress—can be separated. The resulting markedness constraint is essentially compositional, with individual constraints for each element available to the grammar as a whole. This facilitates an historical development of the phenomenon, in which the number of triggering restrictions gradually increased over time 116 (see §2.4 below for an historical analysis of the feminine el), and also allows the individual constraints to be utilized separately in analyses of other language phenomena (for example, see Catalan determiner vowel deletion in §4.2). A summary of the individual elements is provided in (2-67). (2-67) MINDISTVV=F1/F2:n – Do not have adjacent vowels that do not contrast by at least n (higher values of n forbid any adjacent vowels). MINDISTVV=F1/F2:1 – Do not have identical adjacent vowels MINDISTVV(PWD MAX )=F1/F2:1 – Do not have identical adjacent vowels within a maximal Pwd. MINDISTV'V(PWD MAX )=F1/F2:1 - Do not have adjacent identical vowels at an unstressed/stressed juncture within a maximal Pwd. Some discussion regarding the evaluation of word-internal contexts and other stressed vocalic elements is warranted with respect to the final MINDISTV'V constraint. The constraint definition in the Minimum Distance formula is unspecified for vowel sequencing type, and therefore the MINDIST restriction for adjacent vowels will apply to within-word vowel sequences as well as clitic-host sequences. Since Spanish has many examples of word-internal hiatus and diphthongs, such application might be considered undesirable, predicting sequencing repair within all Pwds across the language. One way to prevent this type of overapplication of MINDISTV'V repair is to call on lexical faithfulness, such that lexical Pwds like nouns and adjectives may not undergo repair 117 while functional words like determiners and other clitics can. Lexical faithfulness has been used in relation to phonological acquisition (Geirut et al. 1999), faithfulness to syntactic categories (Smith 2011), faithfulness to lexical forms in hiatus resolution (Casali 1998), and lexically indexed constraints (Pater 2000, 2006, 2008, 2009). For the specific feminine el data outlined here for Spanish, we shall see that this is mostly unnecessary, but lexical faithfulness appears in other adjacency phenomena at prosodic cliticization (see chapter 4). First, I address alternate stress sequences, and then possible counterexamples of vocalic sequencing are discussed. For the feminine el, the MINDISTV'V constraint above is applied to a sequence of adjacent vowels wherein the second vowel is stressed (Table 27a). In a word-internal sequence where the first vowel is stressed the constraint will not assign a violation, because the hiatus transition that requires increased contrast is at the beginning of a stressed vowel, not the end of one (Table 27b). 50 50 Another possible interpretation of the stressed vowel restriction is that stress elongates the vowel and takes up a longer proportion of the VV duration, which could make it more difficult to distinguish where V1 ends and V2 begins. In unstressed sequences the shift from one vowel to the other would be expected to be in the middle, whereas stress could move that transition to one direction or the other, depending on which vowel is stressed. This perspective predicts that either stressed V1 or stressed V2 would trigger restriction. Here, I continue to use positional markedness augmentation of the transition into a vowel, which limits stress sensitivity to only V2. However, future research in other languages with sensitivity to V1 or V2 stress may provide more support for considerations of duration in hiatus. 118 Table 27. MINDISTV'V evaluation MINDISTV'V(PWD MAX )=F1/F2:1 a. CV.V ! * b. CV !.V ! Diphthongs are another source of adjacent vocalic segments, and there are many words in Spanish containing diphthongs, or even beginning with stressed diphthongs. 51 If diphthongs were conceptualized as a sequence of pure vowels the Minimum Distance constraint family would apply. However, despite the similarities between high vowels [i] and [u] and the glides [j] and [w], evidence from phonological contrasts suggests that glides are indeed distinct from high vowels (Padgett 2008), and the common analysis of diphthongs is a sequence of a vowel and on- or off-glide. Moreover, even using a purely vocalic reading of diphthongs, the exact Minimum Distance constraint active in Spanish forbids identical adjacent vowels, and the constraint would not apply because the vocalic elements of diphthongs are by definition not identical. Word-internal hiatus is fairly common in Spanish, with many words containing at least underlying hiatus. For hiatus with non-identical vowels, many studies report diphthongization and other resolution strategies for these words. This move from hiatus to diphthong occurs synchronically and diachronically, depending on internal factors 51 For example, hay [áj] ‘there is/there are’ and aire [áj.re] ‘air’. Onset diphthongs in which the second element is stressed have been traditionally analyzed as a sequence of consonantal glide onset and single vowel nucleus (Navarro Tomás 1977), for example hia.to [já.to] ‘hiatus’, hia.li.no [ja.lí.no] ‘hyaline/translucent’, huér.fa.no [wér.fa.no] ‘orphan’, hui.da [wí.da] ‘flight’, hués.ped [wés.ped] ‘guest’. Acoustic differences of the high vocoids [j] and [w] in onset position also support their distinction from vowels [i] and [u] (Borzone de Manrique 1976). 119 such as stress and phrasal position, and external factors such as sociolinguistic dialect, communicative task, and speech rate (Aguilar 1999, Chitoran & Hualde 2003, Garrido 2007). Some examples of underlying hiatus that may be produced with surface diphthongization are presented in (2-68) (from Garrido 2007). (2-68) pi.a.no > [pja.no] le.ón > [ljon] pe.ón > [pjon] to.a.lla > [twa.ja] Again, the Minimum Distance constraint active in Spanish forbids identical vowel sequences, and will ignore the above cases of non-identical diphthongs and hiatus, but the presence of this family of constraints may contribute to an explanation for why there appears to be an ongoing process of diphthongization and avoidance of hiatus, whether it is word-internal or word-external. Internal identical hiatus also exists in Spanish, most notably with the vowel o, as in al.co.hól ‘alcohol’ and mó.ho ‘mold’. Of these forms, only the word alcohol contains the precise forbidden vowel sequence of a vowel followed by an identical stressed vowel: o-ó. As predicted by the MINDISTV'V constraint, this hiatus is usually not realized. Instead, the word is produced with vowel deletion as [al.kol]. The trisyllabic form may sometimes be produced in exaggerated speech, but this pronunciation registers as an 120 affectation rather than normal native speech. In the word moho, stress falls on the first of the vowels in hiatus, and the constraint does not apply, as for the form in (Table 27b). One of the advantages of the approach pursued here is its compositionality, and that each of the markedness effects may be separated and their influences identified elsewhere in language phenomena. Non-positionally marked versions of MINDISTVV(PWD MAX ), as well as a general MINDISTVV not limited to the Pwd max domain, also occur in the grammar, predicting a range of possible patterns depending on the ranking of these different separate constraints. The historical development of the feminine el shows just such a range of effects, with the gradual increase in triggering conditions of determiner alternation. 2.4 Historical development of the feminine el Historically, the Modern Spanish singular definite articles el and la developed from the Latin nominative demonstratives ille and illa (Penny 2002, Lloyd 1987). In Old Spanish, the feminine determiner had phonological allomorphs el and la resulting from the same bisyllabic form ela, which closely matched the Latin form (Harris 1987, Salvá 1988). 121 (2-69) Latin Old Spanish /illa F / " [ela F ]/[la F ] preceding consonant initial singular 52 nouns " [el F ] preceding vowel-initial singular nouns or adjectives /ille M / " [el M ] preceding masculine nouns Final vowel deletion in the feminine forms was common not only in the definite article from ela to el, but also the indefinite una manifesting as un (2-70a), and occurred before non-/a/ vowels (2-70b) as well as unstressed /a/ (2-70c) and vowel-initial adjectives (2-70d) (Penny 2002, Janda 2003). Bisyllabic ela 53 and vowel-final la occurred with consonant-initial nouns (2-70e), and occasionally co-varied with vowel-initial nouns, as well (2-70f). (2-70) a. un F escoba F ‘a broom’ Cancionero, Juan del Encina (1496) un F (h)ora F ‘an hour’ Poema del Cid b. el F escoba F ‘the broom’ Libro llamado guía de pecadores (1546) el F (h)ora F ‘the hour’ Poesías castellanas (1558), Poema del Cid c. el F alegría F ‘the happiness’ Poema del Cid el F acémila F ‘the beast of burden’ Don Quixote de la Mancha (1582) d. el F alta F sierra F ‘the high sierra’ Poesía de Luis de León (1559) 52 Because the feminine plural determiner is consonant-final (las), it has not participated in phonologically conditioned alternations, and remains unaltered. 53 Many exemplars of ela in historical documents are combinations of y ‘and’ + la ‘the’. However, there are still instances of ela that are written separately from the conjunction and clearly represent the determiner on its own. 122 el F alta F sçiençia F ‘the high science’ Cancionero de castellano de París e. ela F madre F ‘the mother’ Siete partidas (Alfonso X) la F madre F ‘the mother’ Siete partidas (Alfonso X) f. ela F agua F ‘the water’ Libro de Alexandre el F agua F ‘the water’ Libro de Alexandre In this stage of the language, the feminine determiner had multiple allomorphs ela, la and el, with the occurrence of consonant-final allomorph el being consistent with triggering by purely phonological considerations of vowel hiatus. Avoidance of an a-V sequence from Latin-like ela agua encouraged deletion of the final /a/ of the determiner, leading to forms like el F agua F . This feminine el allomorph phonetically matched the masculine determiner simply due to similarity in the Latin forms from which they developed. In the course of development to Modern Spanish, the bisyllabic option ela simplified to la, and the use of el with feminine nouns took on more and more restrictions: only /a/, only stressed /a/, and only nouns. Furthermore, loss of the bisyllabic form in the current grammar leads later generations of speakers to hypothesize only two definite article forms in the grammar: {la, el}. The former is used only for feminine forms, and of all feminine forms it is the majority allomorph. The latter is used primarily for masculine forms, with the exception of a small class of forms that meet the phonological and prosodic conditions for the feminine el F . With the wildly uneven distribution of el M and el F , two possibilities for 123 interpretation emerge for speakers: a) the el used with feminine nouns is a violation of the canonical morpho-phonological relationship, using the masculine phonological form with feminine morphological interpretation (this is the analysis developed above for the prescriptive feminine el pattern) or b) the el used with feminine nouns is in fact masculine el M , and produces violation of syntactic gender agreement. This latter option is examined further in discussion of non-standard variation of the feminine el pattern in §4.5. Returning to the gradual restrictions on the use of the feminine el in the history of the language, a historical OT account of the phenomenon involves several stages of the grammar, with the output from an earlier stage of the language forming the input for a later stage. With the general vowel deletion and loss of the bisyllabic ela, the phenomenon changes from purely phonological in character to morpho-phonological. Once this reinterpretation of the feminine el occurs, marking it as an MP violation, the phonological context in which it appears becomes more and more restricted. In the earlier stage of the language, a simple interaction between general hiatus restrictions and vowel deletion restrictions produces all three feminine definite article allomorphs ela, la and el. The markedness constraint draws on the most restrictive Minimum Distance constraint to forbid all vowel sequences: MINDISTVV=F1/F2:7. The faithfulness constraint MAX-FINALV assigns a violation to any candidate that deletes the final vowel from the input 54 . A variably ranked prosodic constraint MATCH-LEXWD 54 The drive to maintain the final vowel could also be expressed in terms of preserving contrast (Lubowicz 2003) or maintaining morphological paradigm uniformity (Steriade 1999). Final vowels in Spanish tend to contrast morphologically, such that –a is often associated with the 124 requiring all prosodic words to be lexical words 55 will drive variable loss of phonological form in the transformation of ela from a demonstrative, a full lexical word, to a determiner, a functional word. Eventually, this constraint remains higher-ranked than Faithfulness, reflecting the consistent use of the monosyllabic el F and la F forms instead of bisyllabic ela F . feminine morpheme, and –o or a zero affix is often associated with the masculine. For example, compare the indefinite articles un M vs. una F , definite articles el M vs. la F , and nominal endings –a F vs. –o M (this nominal morphology includes several exceptions, but the trend is still overwhelmingly salient. For further discussion on morphological interpretation of nominal forms in Spanish, see §4.5). A Preserve Contrast (PC) constraint could thus be utilized to prevent deletion of the final a in the feminine determiner, since this would render it non-contrastive with the masculine form. A Paradigm Uniformity (PU) constraint aimed to preserve the –a/-# determiner paradigm could also be utilized. Further discussion of Paradigm Uniformity as a force in variation of the Spanish feminine el pattern is presented in §4.5. 55 If function words are assumed to be invisible to the syntax-prosody mapping, this Match constraint is the same as MATCH-PWD, as proposed in Selkirk (2009, 2011). For clarity, it is labeled here separately as MATCH-LEXWD, and functions very similarly to Prince and Smolensky’s (1993/2004) constraint PWD$LEXWD, as well as alignment constraints on lexical word prosody mapping (Itô & Mester 2008). 125 Table 28. Stage I: phonological feminine el with all vowel-initial words i. /ela F muger F / ‘the woman’ MATCH-LEXWD MINDISTVV=F1/F2:7 MAX-FINALV a. ela F muger F *! b. el F muger F *! " c. la F muger F ii. /ela F ora F / ‘the hour’ a. ela F ora F *! *! " b. el F ora F * c. la F ora F *! iii. /ela F acémila F / ‘the beast of burden’ a. ela F acémila F *! *! " b. el F acémila F * c. la F acémila F *! iv. /ela F alta F sierra/ ‘the high sierra’ a. ela F alta F sierra F *! *! "b. el F alta F sierra F * c. la F alta F sierra F *! 126 With the consonant initial noun in (i), Minimum Distance of adjacent vowel pairs does not come into play, since there is no vowel hiatus. Instead, there is only straightforward loss of the initial vowel of the determiner in the winning candidate (ic), bringing it down to a one syllable functional word. In (ii), (iii) and (iv), the consideration of general vowel hiatus from the most restrictive MINDISTVV=F1/F2:7 knocks out the la candidates in (c), regardless of the type of vowel in the following word, allowing the final vowel deletion allomorph el in candidates (b) to win out. When the constraint MATCH-LEXWD stabilizes its ranking to consistently produce only el and la output forms, learners have no access to the bisyllabic underlying form /ela/. Instead, they posit masculine el M and feminine la F as the only determiner forms, and interpret the el appearing with feminine nouns as a marked feminine el F that violates MP correspondence. The use of this marked form is then slowly diminished, occurring with a smaller and smaller set of feminine nouns. This reflects the IDENT-MP drive to maintain normative MP relationships in the grammar. The next restriction of the pattern appears with the alternation to el occurring only with /a/-initial words. This is accounted for by using the appropriate Minimum Distance constraint: MINDISTVV=F1/F2:1, which as discussed in §2.3.1 allows all vowel hiatus except identical vowel sequences. The behavior of determiners with consonant-initial nouns do not change, but those adjacent to non-/a/ vowels now surface as la. Table 29 demonstrates. 127 Table 29. Stage II: feminine el with a-initial words (including adjectives) i. /{la F , el M } ora F / ‘the hour’ MINDISTVV=F1/F2:1 IDENT-MP MINDISTVV=F1/F2:7 a. el F ora F *! " b. la F ora F * ii. /{la F , el M } acémila F / ‘the beast of burden’ " a. el F acémila F * b. la F acémila F *! * iii. /{la F , el M } alta F sierra/ ‘the high sierra’ " a. el F alta F sierra F * b. la F alta F sierra F *! * Candidate (ib) no longer violates Minimum Distance, because the vowels ao are sufficiently distinct to pass MINDISTVV=F1/F2:1. Instead, it wins over candidate (ia) whose use of el F unnecessarily violates IDENT-MP. The winning candidates in (ii) and (iii) still occur with el F , because the identical aa sequence violates the newly active MINDISTVV constraint. A third restriction to the pattern is that of stress, where the feminine el only occurs with words that begin with a stressed a, as opposed to unstressed a. This restriction is represented by the positionally marked MINDISTV'V constraint. 128 Table 30. Stage III: feminine el with stressed a-initial words i. /{la F , el M } acémila F / ‘the beast of burden’ MINDISTV'V=F1/F2:1 IDENT-MP MINDISTVV=F1/F2:1 a. el F acémila F *! " b. la F acémila F * ii. /{la F , el M } álta F sierra/ ‘the high sierra’ " a. el F álta F sierra F * b. la F álta F sierra F *! * The fourth restriction to the pattern is the syntactic one, allowing prenominal adjectives to occur with the vowel-final la, but using the consonant-final el F with nouns. As presented earlier in §2.2, prosodic considerations here come into play. The DA and DN sequences represent different syntactic relationships, leading to their differing organization prosodically and incurring a stronger Pwd max prosodic boundary between the determiner and adjective than there is between the determiner and the noun. This boundary produces a slow-down effect in the production of the vowel sequences straddling the border (see chapter 3), increasing perception of the two vowels as individual elements, and alleviating the Minimum Distance restriction on perceptual contrast at the boundary, while the restriction remains within the Pwd max internal constituent. This final stage of the feminine el pattern corresponds with the markedness constraints in Modern Spanish. 129 Table 31. Stage IV: Modern Spanish feminine el with stressed a-initial nouns only i. /{la F , el M } águila F / ‘the eagle’ MINDISTV'V (PWD MAX )=F1/F2:1 IDENT-MP MINDISTV'V=F1/F2:1 " a. el F águila F * b. la F águila F *! * ii. /{la F , el M } álta F sierra/ ‘the high sierra’ a. el F álta F sierra F *! " b. la F álta F sierra F * As the feminine el develops through the history of Spanish, it becomes progressively more restricted, which is reflected in the use of a progressively complex markedness trigger, summarized in (2-71). (2-71) Stage I: MINDISTVV=F1/F2:7 (ie., NO HIATUS) Stage II: MINDISTVV=F1/F2:1 Stage III: MINDISTV'V=F1/F2:1 Stage IV: MINDISTV'V(PWD MAX )=F1/F2:1 2.4.1 Frequency and historical change In !2.2, the syntactic head-complement relation in DNA and DAN sequences was argued to affect different Pwd adjunction in the prosodic structures. The prosodic 130 analysis also accounts for phonetic slow-down results at the DA border, as will be discussed in further detail in Chapter 3. Together with sensitivity to the syntactic head- complement relation, frequency could have played a role in prosodification and slow- down at the D-A border, particularly in historical change. Word frequency has been shown to influence speech processing, with less frequent words causing delays and more frequent words facilitating speed of processing with various tasks (Forster & Chambers 1973, Landauer & Streeter 1973, Rayner 1977, Taft & Hambly 1986, Trueswell 1996, Griffin & Bock 1998, and others). In addition, frequency can affect a language change in progress, with the change spreading first among frequent words, and being delayed among less frequent words (Hooper 1976, Phillips 1984, Bybee 2002). This same frequency effect of the individual word may also function at the structural level. The prenominal adjective structure is less frequent in Spanish, particularly among those adjectives that may occur pre- or post-nominally. The postnominal adjective structure is more common. This infrequency of syntactic structure might lead to slower language processing, causing delays at the D-A border and interpretation of a larger prosodic boundary 56 . Furthermore, the infrequent structure may have had the same effect upon language change as lexical frequency, resisting use of the el F allomorph. Whether the feminine el pattern’s sensitivity to the DAN word order may be attributed entirely to a structural frequency effect rather than prosodic structure differing as a result of the difference in the syntactic head relations is still an open 56 When compared to disfluencies, this delay does not appear to be in that category. Speech is fluid, and while the extended duration at this prosodic border is statistically significant, it does not register as a long or halting disfluency. 131 question. Further research in the relationship of structural frequency on language change might shed light on this issue. Given phonetic support of prosodic boundary distinctions, both in previous literature and in the Spanish vowel hiatus data presented in Chapter 3, I will continue to assume that prosody differentiates between the DNA and DAN structures in a meaningful way. Another alternative involving frequency is the possibility that frequency determines the prosodic mapping of these sequences, instead of the PARSE D - INTO-! HCOMP constraint. This alternative is discussed in Chapter 5. At the birth of the distinction between feminine el DA and DN word order in Golden Age Spanish, I speculate that the prosodic distinction was the primary trigger for feminine el blocking. Recall that at this time of change in the language, there was still a fair amount of variability. Such variability corresponds well with predicted variability in prosody, allowing speech rate and other factors to influence whether the speaker produced a maximal Pwd boundary between the D and the following word to block the feminine el. Another competing factor on the production of the feminine el at this time is the loss of the bisyllabic ela form, requiring re-interpretation of the feminine el from a simple phonological allomorph to a marked morpho-phonological allomorph. Recall that without the surface ela form, the language learner has no data from which to posit such an underlying form for surface la and el, both used with feminine nouns. Rather, the learner hears la for almost all feminine nouns, el for all masculine nouns, and el with a few feminine nouns that continue to have feminine morphological agreement throughout the rest of the grammar. The feminine el is then re-interpreted as using the masculine 132 phonological form el, together with feminine morphological meaning. This mix of morphological and phonological forms violates the canonical morpho-phonological correspondence and creates a source of possible confusion in the grammar. Use of the marked feminine el form is slowly restricted more and more, with the further restrictions on the phonological vowel quality trigger, and the addition of sensitivity to prosodic structure. Learners still hear el with the following noun most of the time that noun is produced, and so the use of the exceptional and marked determiner form is grammaticalized to that context. The use of la with prenominal adjectives, originally induced by the most common prosodic structure, becomes grammaticalized to that syntactic context. Thus, in Modern Spanish, speakers inherit the memorized pattern that el occurs with this class of nouns, while la occurs with adjectives. As mentioned, most of the time this coincides with the prosodic structure to which the alternation was sensitive. With grammaticalization, subsequent variability of the prosodic structure has no consequence, since the pattern is fixed. Alternatively, transderivational output-output (OO) faithfulness to the normal speech rate form could be utilized to maintain the pattern at other speech rates as well, which avoids fixed grammaticalization in the theoretical analysis. This OO approach might also account for why the feminine el pattern does not shift with prosodic variability, but conjunction allomorphy does. For Spanish conjunction allomorphy, prosodic variability still plays a role. If the prosodic structure is manipulated with the introduction of intervening sentential level 133 adverbs, for example, PPh breaks are inserted between the conjunction clitic and the adverb, and the pause associated with such breaks may block allomorphy (2-72). 5758 (2-72) a. (% amplios espacios [i],) (% inesperadamente,) (% aburridas vistas) ‘wide spaces and, unexpectedly, boring views’ b. (% el uno o,) (% obviamente,) (% el otro) ‘the one or, obviously, the other’ The disparity between the effect of prosody in these two patterns may be related to the availability of structural interruptions. For conjunction, it is syntactically possible for adverbs to be inserted between the clitic and host word. On the other hand, such an interruption between the determiner clitic and its host word is not available. Or, the prenominal adjective may be viewed as precisely such an interruption, but one that triggers insertion of a different prosodic boundary (Pwd max , not PPh). Different types of 57 If these adverbs are interpreted at the NP level instead of the sentential level, then they are like prenominal adjectives in that they are included in the second conjunct XP, do not trigger parenthetical PPh break insertion, and are available for cliticization. amplios espacios e inesperadamente aburridas vistas ‘wide spaces and unexpectedly boring views’ (the fact that the views are boring is unexpected) crueles invasores u obviamente orgullosos patriotas ‘cruel invaders or obviously proud patriots’ (the patriots are proud in an obvious way) 58 There is the additional question of whether an intervening sentential adverb may interrupt hiatus between the conjunction and its syntactically adjacent XP, and continue to allow conjunction allomorphy despite the PPh break. For example: amplios espacios e, obviamente, increíbles vistas crueles invasores u, inesperadamente, orgullosos patriotas This option appears to be unacceptable, as the native speakers that I polled do not find these sentences acceptable. 134 prosodic variability are available to the two patterns: prosodic variability resulting from additional syntactic structures versus prosodic variability resulting from extra-linguistic factors such as speech rate. OO Faithfulness between normal and fast or slow speech rates may prevent pattern variability resulting from extralinguistic factors, but would not prevent structural prosodic differences from influencing allomorphy. Future research is still needed to investigate whether the fixed pattern of the feminine el is still accessible to speakers as productive, and whether it could show sensitivity to speech rate or register. A proposal for such a study is discussed in Chapter 6. If speakers productively use the feminine el with new feminine á-initial words regardless of speech rate, this may indicate a more direct flow of information from syntax to phonology. The advantages and disadvantages to such a direct-reference analysis are discussed further in Chapter 5. 2.5 Exceptions to the feminine el As mentioned in the above section on the evolution of the feminine el, the class of nouns participating in determiner alternation has grown more and more restricted through its history. The shrinking pattern is also reflected in the number of exceptions to the feminine el. As presented in Chapter 1 and here, the feminine el determiner is used with feminine nouns that begin with a stressed a. But, there are feminine á-initial words that do not participate in the alternation, and instead use the regular feminine determiner la. 135 These exceptions include nominal referents with biological gender, acronyms, letters of the alphabet, and countries and cities. Some examples are provided in (2-73). (2-73) Exceptions to the feminine el Gender referents a. la Álvarez ‘the Álvarez [woman]’ b. la árabe ‘the Arab [woman]’ Acronyms c. la AGE (Asociación de Geógrafos Españoles) ‘the AGE (Association of Spanish Geographers)’ d. la AUE (Agrupación de Unidades Especiales) ‘the AUE (Special Forces Group)’ 59 Alphabet e. la a ‘the a’ f. la hache ‘the h’ Countries and Cities g. la Ámsterdam ‘Amsterdam’ 60 h. la Austria ‘Austria’ 59 A special police force in Argentina. 60 Unlike English, countries and cities are often preceded by the definite determiner in Spanish. 136 These different groupings of exceptions have different explanations for why they deviate from the syntactic-phonological pattern of the feminine el. For referents of people in which the nominal form does not vary for men and women, the use of feminine la instead of a feminine el is clearly accounted for by considerations of disambiguation for biological gender. The phrase la Álvarez refers to a woman with the last name Álvarez, and occurs along with the phrase el Álvarez, which refers to a man with the last name Álvarez. Similarly, el árabe refers to a male Arab. Were the feminine el to be used, the distinction between the man and woman would be neutralized. In addition, these phrases may be analyzed as containing an elided noun immediately preceding an adjectival Álvarez or árabe. These phrases do in fact occur with overt nouns, as shown in (2-74). (2-74) a. la mujer Álvarez ‘the Álvarez woman’ b. la mujer árabe ‘the Arabian woman’ The elided noun construction regularly maintains the feminine la with all adjectives, even when the adjective would be analyzed as postnominal were a feminine el noun present, for example (2-75). (2-75) la alta < el (águila) alta ‘the tall one’ < ‘the tall (eagle)’ The PARSE FNC(D) analysis I provide here accounts for these apparent exceptions with noun-seeming adjectives. The adjective—whether it is a regular adjective like alta 137 or a reference to a person like Álvarez—is still not the head of the complement to D, and will therefore always prosodify with the preceding determiner in a free clitic prosodic structure, allowing the regular la determiner form. The elided noun may not produce a prosodic structure, because it has no phonological exponent. Thus, the affixal clitic structure that is required for use of the feminine el is not present in these cases. Acronyms are another source of apparent exceptions. But acronyms may also demonstrate the feminine el (2-76). (2-76) el AUE (Acta Unida Europea) ‘the Single European Act’ This example of the feminine el with an acronym in fact is due to the use of the feminine el with the first word of the acronym, acta. Acta is one of the nouns belonging to the feminine el set: it is feminine, and begins with a stressed a. Compare the acronym el AUE, which refers to a phrase beginning with a feminine el noun, to the acronym la AUE from (2-101d) which refers to a phrase beginning with a feminine noun that takes the regular feminine la: agrupación (which begins with an unstressed a). Given this evidence, I suggest that acronyms are not true exceptions to the pattern. Instead, they retain the determiner form that is used for the entire phrase when it is pronounced in full. This could be accounted for theoretically by the use of Output- Output correspondence constraints (Benua 1995) or Uniform Exponence (UE) (Steriade 1999), enforcing faithfulness of the determiner with acronyms to the determiner in their corresponding full forms. 138 The letters of the alphabet also maintain the feminine la despite the phonological context of stressed á. Again, this could be explained by UE faithfulness of determiners within the paradigm of alphabet letters, which are all feminine. The most interesting group of exceptions is the proper names of countries and cities. According to the Real Academia Española (RAE), the use of the feminine el is variable with topographic proper names. With continents like Asia and África, the RAE prescribes the feminine el, but notes that countries and cities on the other hand take the regular la. Furthermore, this variation of feminine el treatment for this group of proper names is reflected in speakers’ extending the use of la for continents in some cases. For example, Google searches for el Asia and la Asia yielded about equal number of cases (493,000 and 528,000 respectively), while the uneven distribution of el África and la África search results still show preference for the feminine el (1,080,000 and 246,000 respectively). I argue that these country and city proper names are true exceptions to the feminine el pattern, possibly linked to their proper name status. The fluctuation and extension of feminine la to all geographic proper names, along with the gradually increasing phonological restrictions and syntactic-prosodic structure freezing, may be an indication of the tenuous status of the feminine el in the language’s grammar. Further research on these exceptions as well as nonce forms may shed light on the productivity of the feminine el and its standing in the grammar. 139 2.6 Summary of Chapter Two A complete analysis of the Spanish feminine el has been proposed here, incorporating phonological sensitivity to many considerations of language. These include stress augmentation, perceptual contrast of adjacent vowels, morpho-phonological faithfulness, and importantly, the prosodic level of cliticization of the determiner to the following word. Prosodic clitic attachment within or across a Pwd max boundary accounts for an important piece of the pattern: the difference between determiner-adjective and determiner-noun word orders. A strong case is made for the presence of both recursive Pwd cliticization and phrasal cliticization in the Prosodic Phonology of Spanish. The DN word order corresponds to the recursive affixal clitic structure, where the determiner adjoins to the prosodic structure beneath the maximal Pwd, while the DA word order corresponds to a free clitic structure, where the determiner adjoins to the prosodic structures across the maximal Pwd boundary. The syntactic head-complement relation is crucial to the analysis of these distinct prosodic structures, as it forms the basis of a syntactic distinction in the syntactic inputs sent to the PF component of the grammar. The mapping from syntactic to prosodic structure is addressed via Match Theory and PARSE constraints, and the head- complement relation is introduced in the constraint PARSE FNC -INTO-! HCOMP . This mapping constraint accounts for the production of both the affixal and free clitic structures in Spanish determiner sequences, as dependent upon syntactic relations. These prosodic structures feed into the phonological analysis in the form of the constraint domain specific to the prosodic maximal Pwd. Perceptual contrast 140 considerations of Minimum Distance on adjacent vowel segments provide the sensitivity to vowel type, and sensitivity to stress is reflected in the positional markedness context of stressed vowels. The combination of these markedness constraints specific to the Pwdmax domain and positional markedness augmentation at stressed vowels produces the narrowly specified a-á trigger for feminine el determiner alternation. The feminine el repair represents a violation of morphological-phonological identity faithfulness, which contrasts with its historical development from a purely phonological phenomenon. While the current grammar is argued to have experienced grammaticalization to account for the pattern’s status as a fixed effect that no longer varies with prosodic changes, the prosodic analysis put forth captures the historical source of the pattern with respect to the syntax- phonology interface. This chapter has focused on the phonological feminine el pattern and abstract prosodic structures resulting from distinctions in the syntax. The prosodic structures presented for the DN and DA sequences here receive support not only from their different status in the feminine el, but also from phonetic data on general vowel hiatus in Spanish. The following chapter presents this experimental evidence in support of the prosodic structures proposed. 141 Chapter Three: Phonetic study: Spanish feminine el and vowel hiatus The previous chapter outlined the syntactic and phonological aspects of the Spanish feminine el pattern, proposing prosodic differences in the cliticization of the determiner to the following host noun or adjective. This chapter supports the prosodic analysis by exploring the role of prosody in general vowel hiatus in Spanish, to determine the viability of the prosodic hypothesis as a mediator between syntax and phonology in this phenomenon. The goals of the studies studies outlined here are two-fold: to investigate Spanish vowel hiatus and prosody in general, and to investigate hiatus at the specific syntactic structure relevant in the feminine el. Two experiments are presented for these investigations, with results confirming general prosodic category distinctions in manifestations of hiatus, and showing evidence of different prosodic structures for the syntactic differences to which the feminine el is sensitive. Section 3.1 provides some background to Spanish vowel hiatus and reviews the feminine el pattern, §3.2 and §3.3 outline the two experimental studies, §3.4 compares the two studies, and §3.5 discusses the theoretical implications of the phonetic data. 3.1 Background 3.1.1 Spanish vowel hiatus in the phonology As discussed in Chapter 2, Spanish has a five-vowel system of {a, e, i, o, u}, and generally tolerates vowel hiatus, with the exception of the feminine el phenomenon. The specific hiatus that is resolved by use of the masculine determiner form el is a sequence of a followed by stressed a at the determiner-noun juncture (3-2). All other combinations 142 of a followed by stressed and unstressed vowels are tolerated (3-1). For convenience, the relevant data is repeated here from Chapters 1 and 2. (3-1) Phonological vowel hiatus m[i D a]míg[a N á]lta A ‘my tall[f] friend[f]’ tr[á.e] V ‘he/she tries’ l[a D é]ra N ‘the era [f]’ l[a D ó]ra N ‘the hour [f]’ l[a D ú]lcera N ‘the ulcer [f]’ l[a D í]ndole N ‘the (emotional) character [f]’ l[a D a]míga N ‘the friend [f]’ (3-2) Feminine el with determiner-noun a-á sequences. e[l á]gua N , *l[a á]gua N ‘the water[f]’ e[l á]la N , *l[a á]la N ‘the wing[f]’ e[l á]da N , *l[a á] ‘the fairy[f]’ e[l á]guila N , *l[a á]guila N ‘the eagle[f]’ For the same set of nouns exemplified in (3-2), hiatus is tolerated when a prenominal adjective is used (3-3). 143 (3-3) Determiner-adjective blocking l[a á]gria A água N , *e[l á]gria A água N ‘the bitter[f] water[f]’ l[a á]mplia A ála N , *e[l á]mplia A ála N ‘the ample[f] wing[f]’ l[a á]gria A háda N , *e[l á]gria A háda N ‘the bitter[f] fairy[f]’ l[a á]lta A águila N , *e[l á]lta A águila N ‘the high[f] eagle[f]’ As mentioned in Chapter 1, previous research has examined the phonological and morpho-phonological aspects of this data (Gónzalez 2003; Harris 1983, 1987; Varis 2010, Varis 2011; Wolf 2008; Zwicky 1985), but there has been little work on the syntactic difference between adjectives and nouns in this phenomenon. Accounting for the noun/adjective word order distinction is one of the primary goals of this dissertation. 3.1.2 The prosodic approach The difference between determiner-noun-(adjective) (DN or DNA) and determiner-adjective-noun (DAN) sequences to which the phonological process is sensitive could be viewed as purely syntactic—the different word orders reflect different syntactic structures that affect phonological vowel hiatus 61 —or prosodic—the different word orders are phrased into different prosodic structures. In the latter case, prosody would mediate between syntax and phonology, encoding the syntactic difference via a difference in prosodic structures. 61 This direct reference option to the syntax-phonology interface is discussed further in Chapter 5. 144 Since the idea of a direct relationship between syntax and phonology has been controversial (Nespor & Vogel 1986, Inkelas & Zec 1990, Bao 1996, and others), the prosodic approach is preferable. For the full prosodic analysis, the reader is directed to Chapter 2, but a summary of the prosodic hierarchy and categories proposed in this approach is reviewed in (3-4) below, with differentiations made between prosodic words (Pwds), phonological phrases (PPhs) and intonational phrases (IPs). As argued in Varis (2009), evidence for a difference in the prosodic structures of DNA and DAN sequences, both of which feature a clitic-Pwd-Pwd sequence, is lacking. (3-4) The Prosodic Hierarchy & utterance ' intonational phrase (IP) % phonological phrase (PPh) ! prosodic word (Pwd) F foot ( syllable One of the issues in determining prosodic differences between the DNA and DAN sequences is that we do not expect PPh boundaries between any of those words. Studies on related languages, such as Italian, have found no evidence of PPh boundaries between the nouns, adjectives and determiners in these sequences (Dehé & Samek-Lodovici 2009), and work on Spanish prosody has shown the importance of binarity constraints on 145 the formation of PPhs, requiring a minimum of two lexical words per phrase (Prieto 2006). In addition, phonetic cues for PPh boundaries in Spanish indicate that the DNA and DAN sequences recorded in this study are wrapped in a single PPh, with no boundary breaks of that level (also discussed in §5.2). A second question stems from the nature of the determiner as a prosodic clitic. Selkirk (1996) and Anderson (2005) suggest that stressless determiners prosodify as clitics, and Uriagereka (1995) demonstrates similarity between Spanish determiners and the proclitic structure of accusative pronoun clitics. Following these researchers, the prosodic boundary between a determiner and the following word—noun or adjective— would be that which separates a clitic and a Pwd. In this case, if there were a prosodic difference between the DNA and DAN sequences, it would derive from different types of clitic-Pwd attachment. The boundary between the clitic and Pwd is expected to be rather weak, since clitics themselves are accentually weak, but it could have different strengths depending on the level to which the clitic attaches. A review of possible prosodic clitic attachments is included in (3-5a-d). The relevant boundary here is between the bare clitic syllable and the different prosodic levels at which it may attach to the Pwd host 62 . 62 The inter-clitic boundary examined here is distinct from any larger prosodic category of clitic group (Nespor & Vogel 1986, Hayes 1989) that subsumes the clitic and attached items. See §5.1 for a discussion of how the clitic group analysis fails to account for the Spanish data in question. 146 (3-5) a. Internal clitic b. Affixal clitic PPh PPh Pwd min,max Pwd max Clitic Host Pwd min Clitic Host c. Free clitic d. Pwd clitic PPh PPh Pwd max Pwd max Pwd max Clitic Host Clitic Host As discussed in Chapter 2, the most appropriate clitic structures for the Spanish determiner in the DNA and DAN sequences are the affixal and free clitics. In Spanish, the determiner is not a prosodic word, and it does not receive lexical stress, which rules out the prosodic word clitic structure in (3-5d). The internal clitic and affixal clitic structures have been proposed for enclisis (attachment on the right) and proclisis (attachment on the left), particularly in Romance languages (Monachesi 1996, Anderson 2005). Enclitics often show a close connection to the host, and may be subsumed within the word. Proclitics, on the other hand, show more syntactic independence, and may appear in a range of syntactic positions within the sentence. The internal and affixal structures reflect these differences in their levels of attachment. Since the Spanish 147 determiner shows similarity to the proclitic structure of accusative pronoun clitics (Uriagereka 1995) and demonstrates no classic Pwd-level phenomena that would indicate an internal clitic structure, such as primary stress assignment, an affixal or free clitic structure, rather than an internal structure, would be the most appropriate prosodic structure options to distinguish between DNA and DAN, with attachment at the Pwd or PPh level. Still, precisely because there are no other phonological Pwd phenomena available here to differentiate between the structures, we are left looking elsewhere for independent evidence to show that the sequences are prosodified into these different structures. I turn to phonetic reflections of prosodic structures in Spanish vowel hiatus in order to address this issue. 3.1.3 Phonetic reflections of hiatus and prosody As discussed above, Spanish tolerates vowel hiatus at the phonological level. This does not mean, however, that the phonetic production of two full vowels is expected. At the phonetic level, various manifestations of vowel production may occur. Previous research in Spanish has impressionistically identified coalescence and deletion at phonological vowel hiatus (Alba 2006). In other languages, phonetic measurements of vowels in phonological hiatus are shown to reflect distinctions between different prosodic environments, resulting from gradient gestural overlap that produces the perception of vowel coalescence or even deletion (for example, Zsiga 1997 for Igbo, Cabré & Prieto 2003 for Catalan, and Baltazani 2006 and Kainada 2007 for Greek). The production of 148 the intended vowels is more overlapped at the weaker prosodic boundaries, producing greater co-articulation and an effect of coalescence, and less overlapped at the stronger prosodic boundaries, producing less co-articulation. Three main patterns of phonetic vowel production are expected: sequential full vowels, vowel coalescence to varying degrees, or vowel deletion. In a dynamical gesture model of articulatory phonology (Browman & Goldstein 1992, Gafos 2006, Gafos & Benus 2006), these patterns are predicted from timing differences in the proposed organization of the articulatory vowel gestures. In this model, different tongue gestures may be planned for each vowel or consonant, and those gestures coupled together in different timing modes. The “in-phase” coupling aims to produce the gestures at the same time, resulting in gestures overlap and coarticulation. The “anti-phase” coupling aims to produce the gestures at rhythmically contrasting times, resulting in a zero gestural overlap and a sequential arrangement. If we assume a different tongue body gesture for each vowel, then differing degrees of temporal overlap producing minimal, partial and complete gestural overlap (Browman & Goldstein 1992) will predict varying degrees of coalescence from two sequential vowels. For two vowels in sequence, the gestures have minimal overlap, and would be organized in a sequential anti-phase relationship (Browman & Goldstein 1988, 2001). The gestural organization would be that shown in the gestural score (3-6). Each box represents the planned tongue body gesture for the /a/ or the /i/ of the sequence /a-i/. 149 (3-6) Gestural score: sequential vowels: /ai/ TB a i Time With the consecutive gestures above, timing overlap is minimal, and the first vowel is fully identifiable acoustically before the next vowel begins. In this case, we would expect each vowel to be produced with the same (or statistically similar) duration as when it exists as a singleton, surrounded by consonants; and in terms of vowel quality, each vowel is produced with the same gestures and formants as when it exists as a singleton. (In this figure, I further assume minimal influence of consonants on the vocalic signal, such as with a coronal consonant or glottal stop.) In short, each vowel is produced close to its canonical form, with minimal gestural overlap and transition period. Coalesced vowels would have gestures activated closer together, forming more overlap in the intended gestures. Since the articulator for both gestures is the same (tongue body), both vowels cannot be produced at the same time. Instead, the tongue produces a vocalic sound that is a blend between the two intended vowels. With complete temporal overlap, full coalescence is predicted, with the vowel sequence also shortened in duration to the length of a single vowel, shown in the gestural score in (3-7). Partial overlap is also possible, and would result in partial coalescence, shown in the gestural score in (3-8). 150 (3-7) Gestural score: complete coalescence: /ai/ TB a i Time (3-8) Gestural score: partial coalescence: /ai/ TB a i Time In these cases, the tongue body gesture planned for the intended /a/ is overlapped with the gesture for the intended /i/ for a significant portion of the duration of the sequence. Since /a/ is a low vowel and /i/ is a high vowel, gestures for the two are in direct competition, and would probably result in a mid-vowel [e]-like production for the case of complete overlap, and formant changes in partial overlap. Crucially, both vowels are expected to have formants significantly different from singleton /a/ or singleton /i/. The proposed difference between sequential vowels and coalesced vowels can be accounted for by increased rate. The speaker shortens the overall duration by layering 151 vowel gestures. This representation assumes that the duration of each individual gesture remains relatively stable regardless of speech rate. With deletion, one vowel is removed from the production altogether. One way to accomplish this is to not attempt to produce the deleted vowel, with no gestures associated with it. This would be equivalent to phonological deletion, with the vowel removed from the planning stage of a dynamical model. Alternatively, inhibitory amplitude coupling of the two vocalic gestures in the planning phase (Tilsen 2011a,b) may result in one vowel’s gesture activation receiving enough inhibition to never reach the threshold of production. Only the stronger vowel would be produced in the gestural score, which would be identical to that of an underlying singleton vowel. In (3-9), both the planning phase and the gestural score for vowel deletion are presented, where the arrows in the planning phase represent the inhibitory amplitude coupling of the gestures. The thickness of the arrow represents the amplitude of the inhibition. 152 (3-9) Vowel deletion: /ai/; where /a/ is inhibitorily coupled to /i/ Planning phase: /a/ /i/ Gestural score: TB i Time In this proposed model of phonetic vowel deletion, the vowel is still present phonologically in the planning phase, but due to the inhibitory force of being coupled to another vowel gesture with stronger activation—perhaps resulting from different categories, such as a lexical word versus function word distinction—it never reaches the production phase. The only vowel acoustically identifiable in the proposed score in (3-9) would be [i]. In addition to varying degrees of vowel hiatus production found in previous research on other languages, recent work on Spanish prosody has identified final boundary lengthening as a non-hiatus phonetic correlate to different prosodic boundaries (Beckman et al. 2002, Prieto 2006, and Rao 2010, among others). Specifically, Rao (2010) identified increased lengthening in words and stressed syllables at the end of 153 higher level prosodic phrases such as PPhs and IPs, in comparison to Pwds. PPhs and IPs were differentiated by smaller pauses but increased lengthening in the PPh cases, with the proposal that increased lengthening provided an additional, necessary cue for the inherently weaker PPh boundary. As a phonetic effect of prosodic boundaries, general lengthening would be expected to influence the behavior of vowels in hiatus at these boundaries, as well. Following these researchers, I investigate phonetic hiatus resolution patterns for Spanish inter-word sequences in order to determine whether a) phonetic vowel production reflects prosodic boundary distinctions as in the previously studied languages, and if so, b) it reflects finer-grained distinctions between determiner-adjective and determiner-noun boundaries. As a separate issue, the question of phonetic resolution type—whether complete deletion may occur or whether vowels are merely coalesced—is investigated to clarify the possible phonetic productions of phonological hiatus in Spanish. 3.1.4 Hypotheses The two main hypotheses address the issues of prosodic structures in Spanish vowel hiatus, to confirm that established prosodic levels are differentiated phonetically and to provide evidence for a prosodic difference between DNA and DAN sequences. (3-10) H A : Spanish productions of phonological vowel hiatus exhibit measurable differences as a function of different levels in the prosodic hierarchy. If the 154 vowels in hiatus span the boundary of a category of the prosodic hierarchy, the productions will differ according to the category level. H B : Spanish productions of phonological vowel hiatus exhibit measurable differences between DNA and DAN syntactic sequences. If both H A and H B are supported, there would be evidence for postulating different prosodic structures of DNA and DAN, allowing a prosodic analysis of the feminine el. A sub-hypothesis to the above two addresses the nature of vowel production in hiatus contexts, following up on previous researchers’ reports of vowel deletion as a phonetic hiatus repair strategy. As predicted by the model of gestural overlap, unless the vowels are strongly inhibitorily coupled such that one of them does not reach the production threshold, gestural overlap will result in coalescence between the two vowels rather than deletion. That is, both vowels are still present phonetically, although their manifestations may vary. (3-10ii) H C : Phonetic productions of Spanish phonological vowel hiatus result in vowel coalescence, but not vowel deletion. This hypothesis regards full deletion of a vowel, with no attempt to phonetically produce it, either in timing or vowel quality. Spanish is not generally recognized as having a consistent and predictable pattern of phonological vowel assimilation that would retain vocalic length while deleting or modifying vowel quality (for example, /a.e/ " [a:]), nor 155 assimilation that retains vowel quality while reducing a vocalic sequence to the timing of a singleton vowel (for example, /a.e/ " [æ]). Since this type of behavior does not occur phonologically, I do not expect phonetic manifestations to also be so divided. Rather, I assume that the production of Spanish vowels is directly linked to the intention to produce a full phonological vowel. The aim is to evaluate whether the underlying phonological vowel is present phonetically, and if so, how it is manifested. Investigating the way in which vowel hiatus is phonetically manifested can inform decisions about how to pursue the larger questions and what measures may reflect which prosodic distinctions. Each of the two main hypotheses is tested in a speech production experiment, and the sub-hypothesis regarding coalescence or deletion is examined along with each of the main questions. The data for both issues was collected with the same participants at the same time, but the experiment designs and results are presented individually. 3.2 Experiment 1: Spanish hiatus and prosodic boundaries The first experiment focuses on Spanish vowel hiatus at different prosodic boundaries, to determine its sensitivity to prosody. 3.2.1 Methods 3.2.1.1 Materials The four prosodic variables investigated are clitic, Pwd, PPh, and IP boundaries. The vowels in hiatus are combinations of the vowel a followed by every vowel of the Spanish five vowel system, for a total of five vowel test sequences: aa, ae, ai, ao and au. 156 To create test and control items, combinations of noun-verb words were selected for each vowel pair and put into carrier phrases at different prosodic boundaries. For example, for the ae condition, the words hotelera estaba ‘hoteliere was’ were placed into sentences in which they would be a) within a phonological phrase (Pwd condition) b) across a phonological phrase (PPh condition) or c) across an intonational phrase (IP condition). For the clitic condition, a direct object pronoun-verb pair was used. All vowel combinations were unstressed, to remove the possible effect of lexical stress upon hiatus production. With the four prosodic conditions and five vowel pairs, 20 test sentences were created. (See Appendix I for full carrier sentences of test and control items.) Control items for each word in the pair were created using one of the test words paired with a word that ended or began with a coronal consonant. Coronal consonants were preferred as they would have less influence on the surrounding vowel formants, as compared to a velar or labial consonant, which exhibit strong influences on F1 and F2 going into and out of the consonant. For example, for the ae condition, two control pairs were generated: hotelera tasó ‘hoteliere valued’ and árbol estaba ‘tree was’. In this way, the same vowel of the same word could be compared in a hiatus environment and non- hiatus environment. The only condition that did not have double controls was the clitic one, because Spanish verbal clitics are vowel-final 63 . A control was included for the vowel-final clitic with a consonant-initial verb, but the reverse was impossible given the 63 As has been pointed out, Spanish plural clitic forms end in the consonant /s/. However, pilots of this study with coda /s/ proved problematic to code due to variable s-aspiration, and these forms were deemed unsuitable for controls. 157 limitations of the language. With the four prosodic conditions and nine control word pairs, 36 control sentences were created. 3.2.1.2 Procedure Nine native speakers of Peninsular Spanish—five women and four men—residing in the Madrid area participated in the study, aged 18 to 50. They were presented with test, control and filler sentences to read out loud. Three repetitions of test sentences and two repetitions of control sentences were included. Based on pilot results, the sentences were grouped into blocks by length—long, medium and short—and randomized within length blocks. Each block was presented via PowerPoint slide presentation with appropriate timed intervals for the length of the sentences, to minimize variation in rate within each participant’s productions. Blocks of ‘long’ sentences appeared with seven seconds for each sentence, blocks of ‘medium’ sentences appeared with six seconds, and blocks of ‘short’ sentences appeared with five seconds. Each participant was presented with the blocks in a different random order. There were seven blocks total—two long, two medium, and three short—of 23-25 sentences each, and participants were invited to rest in between blocks. The total experiment duration was between 30 and 40 minutes, depending on the participant. Participants were recorded using Praat (Boersma & Weenink 2008). Vowel sequence segmentation was done manually from the wave and spectrogram forms. The beginning of a vowel sequence was identified from the appearance of F1 and F2 formants after their absence for a pause or an obstruent, or from the sharp shift in F1 from a lateral [l] to a vowel (F1 was considerably less dense for the 158 consonant than for the vowel). The end of a vowel sequence was again identified from the disappearance or sharp shift in formants. Wave forms corroborated the start and stop times with increased amplitude for vowel segments as compared to consonants. The formant measurements were obtained from the resulting timepoints using the Praat algorithm for formant extraction, and the data were analyzed R (R Core Development Team 2010). 3.2.1.3 Measurements The test and control sentences were first evaluated for prosodic structure, to confirm that they were indeed produced with the intended prosodic boundaries. This check also eliminated the concern of rate influence on the formation of prosodic phrases. While some participants regularly spoke rapidly, if the sentence was not produced with the prosodic boundary being examined, then the token was discarded, since it would not be an accurate representation of the effect of prosody on vowel hiatus manifestation. Following Prieto (2006) and Spanish To_BI conventions (McGory & Diaz- Campos 1999), PPh and IP boundaries were identified if there was an audible prominent pitch accent at the end of a phrase. Of those boundaries, those at the beginning and ends of sentences were assumed to be IPs, while those intervening between phrases sentence- medially were assumed to be PPhs. 64 In addition, there was a pause or break between the 64 This assumption has been questioned in some research. For example, Sosa (1991, 1999) argues against an intermediate PPh boundary, in favor of only IP. Given that previous research in Spanish has found a difference between sentence-medial and sentence-final boundaries of this nature (Rao 2010) as well as the results found here distinguishing between these two categories, I continue to assume a distinction between PPh and IP boundaries. 159 elements 65 . Pwd and clitic conditions were identified with no pause between VV sequences, and no clear prominent pitch dividing the words into separate phrases. Clitic words received no tonic pitch accent. I focus on two main phonetic measurements of vowel hiatus: duration of the vowel-vowel (VV) sequence, and the formants of the sequence. By comparing duration measurements of VV test items with singleton V control, deletion as a hiatus strategy can be evaluated. This is especially relevant for sequences of identical vowels, such as aa, which occur frequently in Spanish phonology, and which would lack formant distinctions between the two vowels. Average formant values are also compared among test vowels at different boundaries, to evaluate the variation of adjacent vocalic influence as dependent on the prosodic category. Acoustic measurements were taken on the duration and formants of the vowel sequences, as presented in (3-11). 65 While Spanish PPhs do not necessarily require a significant pause, when evaluating tokens for the difference between PPhs and Pwds, the clear PPhs with final pitch accent always included a short pause or breath. In addition, third party confirmation of whether certain ambiguous tokens sounded more like a PPh or Pwd boundary always ruled in favor of the non-pause tokens being Pwd. Truly ambiguous tokens were discarded from analysis. 160 (3-11) Acoustic measurements of Experiment 1 a. duration of V1 and V2 individually when a pause intervened b. duration of the total V1V2 sequence when no pause intervened c. average F1 measurements of the vowels at four equal intervals across V1 and V2 individually when a pause intervened d. average F1 measurements of the vowels at four equal intervals across the total V1V2 sequence when no pause intervened e. average F2 measurements of the vowels at four equal intervals across V1 and V2 individually when a pause intervened f. average F2 measurements of the vowels at four equal intervals across the total V1V2 sequence when no pause intervened For example, in the clitic test case for ae, the ae vowel sequence was measured for total duration, and it was broken up into four equal intervals. For each interval, an average F1 and an average F2 was calculated. An illustration of interval division is shown in Figure A. 161 Figure A. Wave and spectrogram for the segment la heriste ‘you hurt her’, showing the continuous ae vowel sequence broken up into four segments for analysis. In the IP and PPh conditions where V1 and V2 were broken by a pause, each vowel was divided into four intervals and formant averages taken for F1 and F2 at each interval. In the Pwd and clitic conditions where the V1V2 sequence was continuous, the entire sequence as a whole was divided into the four intervals and formant averages taken from each interval. This encourages separate analyses of the broken and unbroken vowels. Because of the careful prosodic monitoring of the tokens, not all variable conditions were eligible to be analyzed for every subject. For example, some subjects 162 tended to speak faster, producing Pwd boundaries where an IP or PPh was be expected, leading to missing tokens in the IP and PPh conditions for that subject. As a result, the design measurements were unbalanced. For the analysis, a mixed effects model in R, lmer (Bates & Martin 2010), was used in order to account for both subject variation and unbalanced conditions. In addition to the issue of prosodic production, formant measurements were also problematic in some cases, since some of the audio recordings produced weak formants, and some formants were not accurately identified by the formant tracker. Many /u/ vowels, for example, were produced with weak formant structures, and were not able to be accurately analyzed. When there were too many problematic tokens for /u/ conditions to produce reliable results, this vowel was left out of the analyses. Word frequency was not initially controlled for in the experimental design, due to the difficulty of accounting for the specific phonetic environment (see §3.3 on Experiment 2 for more discussion of frequency), but frequency was coded for investigation as a possible factor. 3.2.2 Predictions 3.2.2.1 Duration In examining the duration of vocalic sequences, several predictions stem from the hypothesis H A that vowel hiatus productions differentiate between levels of the prosodic hierarchy. First, it is predicted that greater prosodic boundaries would correlate with longer vowel durations of V1, in keeping with previous research on final boundary 163 lengthening in Spanish, and forming a strong phonetic differentiation between levels of the prosodic hierarchy. V2 is not expected to show the same degree of lengthening, as domain-initial lengthening effects have not been reported for Spanish, though it may occur in other languages (Fougeron 1999, 2001; Barnes 2006). 66 Inherent duration differences due to vowel quality are also expected to surface (Lehiste 1970). In unbroken V1V2 cases, the total duration is predicted to be longer for the Pwd condition than for clitics, as a continuation of expected lengthening differentiation effects between stronger and weaker prosodic boundaries. 67 Regarding the hypothesis H C of coalescence vs. deletion, I predict that unbroken V1V2 compared to a singleton control V will be longer, reflecting a lack of vowel deletion. 3.2.2.2 Formants The expected patterns of formant measurements closely follow some of the predictions for duration and the findings of previous research in vowel hiatus at prosodic boundaries. Overall, it is predicted that at greater prosodic boundaries, V1 formants would be less influenced by vowel quality of the following vowel, but that at weaker 66 Domain initial strengthening effects found by these researchers have primarily been in consonants, not vowels. Fougeron (2001) found lengthening effects for domain initial vowels in French, but these effects were only for one speaker, and were not consistent between vowels (/i/ was longer in an accentual phrase than Pwd, but /a/ was longer in Pwd than in an accentual phrase). 67 Although a lengthening distinction could also be interpreted as an effect of stress rather than prosodic boundary, the definition of free or affixal clitic coincides with a lack of stress. If a weak clitic item is stressed—such as with emphatic subject pronouns, for example—it is elevated to Pwd status in the prosody, and would no longer be considered a free or affixal clitic, but might be a Pwd clitic, with a Pwd-Pwd boundary between the clitic and its host. 164 prosodic boundaries the result of co-articulation would be greater. V2 is predicted to also show vocalic distinctions, since they are different vowels underlyingly, but co- articulation effects from the preceding a could yield less vowel distinctions, depending on the strength of the prosodic boundary. With a stronger prosodic boundary, less influence from the a would be noticed, whereas at a weaker prosodic boundary, more co- articulation would produce vowels that were closer together and less distinct. At IP and PPh boundaries, the influences of prosodic boundary and following vowel type are directly measurable on V1 and V2 separately, while at junctures between Pwds and between Pwds and clitics, the predicted coarticulation effects of V2 on V1 would emerge in the formant measurements of the first two time intervals measured, and effects of V1 on V2 would emerge in the formant measurements of the last two time intervals. In general, coarticulation effects are expected to be reflected in the degree of interaction between the factors of vowel quality and prosodic boundary, in a statistical model. At higher level prosodic boundaries (such as IP), less effect of the prosodic boundary upon the different behavior of vowels should correspond to less interaction between the factors, such that the behavior of the vowel formants is due entirely to the expected behavior of the vowels themselves. In comparison with the higher level prosodic boundaries, weaker boundaries such as at clitics would be expected to demonstrate more interaction between the factors, such that the degree of vowel dispersion in the space is in part dependent on prosodic category. 165 3.2.3 Results 3.2.3.1 Duration Analyses Final lengthening The first comparison was of durations of vowels at each prosodic boundary, to confirm the general effect of prosody on vowel lengthening. To determine the effect of prosody alone and provide a comparison across all prosodic boundaries, possible hiatus effects were removed by examining only control V1 items with the factor of prosodic boundary. These are the vowels just preceding a prosodic boundary, at the end of a phrase. Vowel type did not vary in V1, as they were all a. Both subject and word were input as random effects for the model. The mixed effects model results showed significant lengthening effects depending on the prosodic boundary. The fixed effects results in (3-12) are shown for each prosodic boundary, as compared to the clitic boundary. 68 T-values of absolute 2 or greater are considered significant, and marked with an asterisk *. 68 Another model was run including word frequency as a possible factor, but it was statistically insignificant, and a model comparison yielded no difference between the models. Therefore, the simpler one-factor analysis is reported here. 166 (3-12) Final lengthening fixed effects results (df: 7): Estimate Std. Error t-value (Intercept) 71.28 10.83 6.585 IP 55.98 10.72 5.222* PPh 69.61 10.96 6.349* Pwd 18.56 11.09 1.673 The specific comparisons between boundaries from post-hoc Tukey tests are summarized below in (3-13), and durations are illustrated for each prosodic boundary in Figure B. (3-13) PPh longer than IP p<.05 PPh longer than Pwd p<.001 PPh longer than clitic p<.001 IP longer than Pwd p<.001 IP longer than clitic p<.001 167 Figure B. Durations of control V1 at the clitic-Pwd boundary, IP boundary, PPh boundary and Pwd-Pwd boundary. Initial lengthening Although domain initial lengthening effects have not been reported for Spanish, control V2 was analyzed to investigate the possibility of vowel lengthening at the beginning of prosodic categories. As mentioned above, controls for V2 of the clitic items were not available, but the mixed effects model was run instead on the IP, PPh and Pwd 168 data. 69 The model yielded no significant domain initial lengthening effects. The t-value results are presented in (3-14), as compared to the IP condition. Figure C illustrates the lack of domain initial vowel length differences in Spanish. (3-14) Initial lengthening fixed effects: Estimate Std. Error t-value (Intercept) 78.1121 5.0887 15.350 PPh 0.9003 3.1536 0.285 Pwd 1.6954 3.3641 0.504 69 Frequency was not included as a separate factor in this analysis, as all the words in this set are of high frequency. 169 Figure C. Durations of control V2 at the IP, PPh and Pwd boundaries. Hiatus effects: (V1)V2 vs. (C)V2 Further duration results with V2 examine the IP and PPh cases separately, and include the test vowels as well as the controls. Here, the relevant factors included in the mixed effects model were prosodic boundary (whether the vowel was at the beginning of an IP or a PPh), vowel quality (a, e, i, o, u), and whether the token was in hiatus with a preceding a or a non-hiatus control token. 70 Of the three factors examined, the only one 70 While interaction was not expected between these factors, for completeness the model was also run for interaction effects. An anova model comparison yielded no statistical difference between 170 with any significant effect on V2 duration was vowel type: specifically, e was shorter than a (t = -8.58), o was slightly longer than a (t=2.151) and u was shorter than a (t=- 4.088). There was no significant difference between test and control items, and V2 duration was not different between the IP and PPh boundaries. V2 duration is illustrated in Figure D. Figure D. V2 durations for the IP and PPh boundaries, comparing test and control items for each vowel quality. the interaction model and non-interaction model. The reported t-results here are from the simpler, non-interaction model. 171 Hiatus effects: V1V2 vs. V1(C) For the unbroken V1V2 sequence at the weaker prosodic boundaries Pwd and clitic, factors of the analysis included vowel quality (aa, ae, ai, ao, au), prosodic boundary (whether the first vowel was at the end of a Pwd or clitic), and test (V1V2) vs. control (V1). 71 Control V1 was included as opposed to control V2 since there was no control V2 available for clitics, as discussed in §3.2.1.1. In the mixed effects model results (df: 10), the only significant differences occurred between test and control vowels of different underlying numbers. Test sequences of two vowels were significantly longer than controls (approximately 66 ms. longer), but vowel type and prosodic boundary were not significant. 72 The plots in Figure E show clitic and Pwd vowel comparisons for the test and control vowels. 71 Since this analysis includes test sequences of two words that vary (unlike controls in which only one word varies), single word frequency was inappropriate as a factor. Investigation of bigram frequencies as a factor for these hiatus analyses is suggested for future research (see §3.5). 72 These factors were also not significant when run on only test data, separate from the controls. 172 Figure E. Control V1 and V1V2 test sequence durations for the Pwd and clitic boundaries. 3.2.3.2 Formant analyses IP and PPh: V1 For IP and PPh hiatus V1, the mixed effects model was run on F1 and F2 measurements at each time interval, using the factors vowel quality of the following vowel (a, e, i, o, u) and prosodic boundary (IP or PPh). Neither the vowel quality of the following post-pause vowel nor the prosodic boundary itself had any significant effect on the F1 of V1 of IP and PPh items, nor was there any interaction between the two factors, across all time intervals. 173 On F2 measurements, the following vowel quality had some effect in general, but it did not show interaction with the prosodic boundary type, and boundary type did not have an overall effect. For ease of presentation, the effects of vowel quality on each time interval of F2 are summarized in Table 32. The full fixed effects results for all formant analyses are available in Appendix C. The notation n.s. indicates that the effect was not statistically significant. All vowel types are compared to the baseline sequence of aa. The vowel charts in Figure F show the dispersion of V1 a, as followed by the different vowels in hiatus, for both prosodic boundaries across all time intervals. As indicated in the mixed model and the vowel charts, only i and o showed any differentiation from the other vowels as having an effect on the previous a, but this was not consistent throughout the time intervals. Table 32. Summary of all intervals of F2 vowel quality effects for IP and PPh: V1 F2 effects Time intervals V1, a followed by: 1 2 3 4 e n.s. n.s. n.s. n.s. i Higher F2 t=4.18 Higher F2 t=2.0 n.s. n.s. o n.s. Lower F2 t=2.2 n.s. n.s. u n.s. n.s. n.s. n.s. 174 Figure F. Vowel charts for the 1 st , 2 nd , 3 rd and 4 th time intervals of V1 at IP and PPh boundaries. While V1 is underlyingly a for all vowel sequences, the following vowel i, e, a, o or u is used as the label to distinguish between a followed by i vs. a followed by o, etc. IP and PPh: V2 Possible formant effects were also examined for V2. For these items, /u/ tokens had to be removed from the analysis because there were not enough of them with accurately identified formants to provide statistical analysis. In V2, vowel type had a strong effect on F1 and F2 measurements. Prosodic boundary did not have an effect, nor was there any interaction between vowel type and prosodic boundary. The effects on F1 175 and F2 are summarized in Table 33, and the vowel charts for IP and PPh V2 vowels are shown in Figure G. Table 33. Summary of F1 and F2 effects for IP and PPh: V2 V2, in comparison to a: Time intervals F1 effects 1 2 3 4 e Lower F1 t= -6.89 Lower F1 t= -5.72 Lower F1 t= -5.98 Lower F1 t= -4.75 i Lower F1 t= -9.35 Lower F1 t= -7.02 Lower F1 t= -6.98 Lower F1 t= -5.79 o Lower F1 t= -4.78 Lower F1 t= -4.18 Lower F1 t= -3.99 Lower F1 t= -3.48 F2 effects 1 2 3 4 e Higher F2 t= 10.41 Higher F2 t= 10.67 Higher F2 t= 10.32 Higher F2 t= 10.53 i Higher F2 t= 14.74 Higher F2 t= 13.92 Higher F2 t= 12.54 Higher F2 t= 11.05 o Lower F2 t= -4.12 Lower F2 t= -4.16 Lower F2 t= -3.45 Lower F2 t= -2.46 176 Figure G. Vowel charts for the 1 st , 2 nd , 3 rd and 4 th time intervals of V2 at IP and PPh boundaries. Pwd and clitic For the Pwd and clitic vowels, the mixed effects model was run on F1 and F2 using the factors vowel type (aa, ae, ai, ao, au) and prosodic boundary (Pwd (= Pwd- Pwd) or clitic (= clitic-Pwd)). T-values for the fixed effects and p-values for post-hoc interactions across the time intervals are summarized in Table 34. The clitic and Pwd vowels are superimposed in the charts of Figure H to illustrate the different behavior of the two boundary conditions. 177 Table 34. Summary of F1 and F2 effects for Pwd and clitic V1V2 Time intervals 1 2 3 4 Vowel type n.s. n.s. n.s. n.s. F1 Interaction n.s. n.s. n.s. n.s. Vowel type ae > aa (t=4.9) ai > aa (t=8.0) ao < aa (t= -4.1) au < aa (t= -3.9) ae > aa (t=7.9) ai > aa (t=9.4) ao < aa (t= -4.1) au < aa (t= -6.6) ae > aa (t=8.2) ai > aa (t=9.5) ao < aa (t= -5.4) au < aa (t= -6.5) ae > aa (t=8.3) ai > aa (t=11.0) ao < aa (t= -2.0) au < aa (t= -4.7) F2 Interaction ae, ao and au were s. different from aa at the clitic boundary (p< .01), but not the Pwd. ae had a higher F2 at the clitic boundary than at the Pwd (p= .017) All vowels s. different from aa at both clitic and Pwd boundaries (p< .01) ae had a higher F2 at the clitic boundary than at the Pwd (p< .01) All vowels s.different from aa at both clitic and Pwd boundaries (p< .001) ae had a higher F2 at the clitic boundary than at the Pwd (p< .001). au was s. different from aa at the clitic boundary (p< .01), but not the Pwd. ae had a higher F2 at the clitic boundary than at the Pwd (p< .01) 178 Table 34. Continued Time intervals 1 2 3 4 Interaction au had a lower F2 at the clitic boundary than at the Pwd (p= .024) au had a lower F2 at the clitic boundary than at the Pwd (p= .039). au had a lower F2 at the clitic boundary than at the Pwd (p= .037) Figure H. Vowel charts for Pwd and clitic vowels across the 1 st , 2 nd , 3 rd and 4 th time intervals. These unbroken vowel sequences are all a-V2, but only V2 is labeled to show more clearly the differences between the underlying vowels. 179 3.2.4 Discussion of Experiment 1 results Duration Interestingly, Pwd final vowels were not significantly longer than clitic vowels, although there is a tendency towards greater length. Rather, the prosodic hierarchy seems to be broken up into three levels regarding final lengthening effects: PPh producing the greatest lengthening, IP producing the next greatest lengthening, and Pwd and clitic grouped together at the lowest level with the same duration. The results are consistent with Rao’s (2010) findings regarding PPh, IP and Pwd lengthening distinctions. The prosodic boundaries broken by a pause, IP and PPh, both pattern with greater final vowel duration than unbroken Pwd and clitic, and the highest level prosodic category (IP), which is also associated with a longer pause, demonstrates less lengthening than the middle level PPh category. These results indicate that Spanish does produce a vowel lengthening distinction between different levels of the prosodic hierarchy, though Pwd and clitic are grouped together. The initial lengthening results match clearly with the predictions. Previous research in Spanish has not identified the beginning of a phrase as receiving lengthening effects, and since V2 is at the beginning of a new phrase, not the end, it is unsurprising that prosody had no significant lengthening effect. An inherent duration difference between vowels a and e is to be expected, as well. For the third factor examined in IP and PP cases (a non-hiatus control vowel vs. hiatus test vowel), the number of vowels being compared is the same—one—so it is not unexpected for there to be no significant duration difference, particularly at a strong 180 prosodic boundary that has produced a significant pause or separation between vowels in hiatus. For the Pwd and clitic boundary hiatus V1V2 vowels, a tendency for longer durations at the Pwd boundary was found, but no statistically significant distinction. The lack of a distinction between the Pwd and clitic boundaries was unexpected in the initial predictions, but is in keeping with the results found from control V1 durations overall, where there was no difference in duration between Pwds and clitics. The prediction that Spanish does not delete—even phonetically—in cases of vowel hiatus is borne out in these results. The duration of two phonological vowels in hiatus significantly exceeds the length of one vowel, supporting the conclusion that they are still produced as two vowels. The percept of only one vowel cited in Alba (2006), under conditions of what is usually regarded as phonological hiatus, may be related to the level of co-articulation of the two vowels in hiatus. The gestural model discussed above predicts that vowel formants are affected by the attempt to produce another vowel immediately adjacent, and therefore the resulting vowel sequence formants may sound more like one vowel or more like the other, or they may sound like a completely different coalesced third vowel. Formants In the formant results for IP and PP V1, the close cluster of vowels in the vowel charts (in Figure F) reflects the lack of co-articulation effects from the following vowel. For the most part, the a’s are all produced very similarly to each other, despite being 181 followed by differing vowels. The tight overlap of these vowels in the charts indicates that there is not much difference in the production of a followed by different vowels across these boundaries. The main prediction regarding a differentiation between prosodic boundaries was not borne out in the formant measurements for V1, despite a difference in final lengthening between IP and PPh. This would indicate that duration and coarticulation effects are not necessarily correlated. Instead, final lengthening of V1 from the prosodic boundary appears completely independent of hiatus effects. This independence of duration and coarticulation effects is not surprising, given the pause between V1 and V2 in IP and PPh cases. The pause in speech allows ample time to reform the vowel for the next phrase, minimizing coarticulation effects, while the boundary still produces a duration distinction among the vowels at the end of a phrase. In IP and PPh V2, vowel type had a strong effect on F1 and F2 measurements, as expected considering that these are underlyingly different vowels. However, the lack of interaction between vowel type and prosodic boundary indicates that the vowels behave much the same whether they are in hiatus across an IP boundary or a PPh boundary. Formant co-articulation effects are not reflective of boundary distinctions at this level. Instead, there is very little co-articulation at these higher level prosodic boundaries, probably due to the strength of the break and pause between vowels. For the Pwd and clitic cases, recall that the specific prediction regarding the factors of vowel quality and prosodic boundary is again the presence of interaction, indicating that there is more co-articulation at the clitic boundary than at the Pwd. While 182 the details of the interaction are not consistent between vowel types across the time intervals, a broad picture emerges with respect to the effect of prosodic boundary on the formants: more influence of the following vowel type occurs at the clitic boundary than the Pwd. Although F1 did not show significant effects, in F2, the vowel type co- articulation effects were stronger at the clitic boundary than the Pwd. The following vowel had more of an effect on the formants of the underlying preceding /a/ in the early interval at the clitic boundary, while the same formant effects showed up a little bit later at the Pwd boundary. In addition, the vowel sequences ae and au showed stronger co- articulation effects at the clitic boundary than the Pwd boundary. The tendency for /a/ vowels to resemble the following vowel of the sequence early on in clitics and Pwds is illustrated in Figure H. We can see the clustering effect from underlying V1 /a/ in the early formants, which gradually disperse to their respective V2 positions, but even in the early cluster, each a is differentiated by the vowel following it. This differentiation is clearer in the clitics than the Pwds, reflecting the increased co- articulation at this boundary. Since the a is co-articulated with the following vowel, the formants during the a portion of the vowel sequence look more like the following vowel than an a unaffected by a different vowel (such as the a followed by a). While both the Pwd and clitic vowels show early co-articulation and gradual dispersion, the clitic vowels of the first time interval are already well dispersed relative to one another, while the Pwd vowels do not yet show the canonical relative spacing of the Spanish vowels to each other. 183 Overall, the prosodic boundary prediction held for the unbroken V1V2 sequences. Clitic boundaries showed more of a co-articulation effect than Pwd ones, and this co- articulation occurred early in the sequence, affecting V1 /a/ depending on the quality of the following V2. 3.2.5 Summary of Experiment 1 The results of Experiment 1 showed support for hypotheses H A and H C , which proposed hiatus productions reflecting different prosodic boundaries and vowel coalescence instead of vowel deletion, respectively. Regarding the first point, both duration and formant results supported differentiation between prosodic boundaries. Final lengthening of V1 distinguished between IP, PPh and Pwd, though the duration results did not differentiate between Pwd and clitic categories. Formant measurements, however, showed co-articulation effects in Pwd and clitic categories—with a distinction between the two. Formant measurements did not show such distinctions between IP and PPh. Taken together, the prosodic boundaries were differentiated by the phonetic manifestations of vowel hiatus, though in different ways by different cues. Regarding the question of vowel coalescence versus deletion, duration differences between control V1 and test V1V2 sequences indicate that both vowels in a V1V2 sequence are produced, rather than undergoing vowel deletion. The above findings in prosodic distinctions have promising implications for the question of the relationship between syntax and phonology at the DAN and DNA sequences in Spanish. Since prosodic structures like Pwds and clitic manifest phonetic 184 hiatus differences, then if Spanish determiner sequences are prosodified differently, they may also show phonetic hiatus differences. Experiment 2 explores this possibility. 3.3 Experiment 2: Spanish hiatus of DNA/DAN sequences The second experiment examines hiatus of DNA and DAN sequences in Spanish. While these sequences were not expected to show signs of full phonological phrase boundaries in between the DP elements—and in fact, never demonstrated such breaks in fluent speech—the phonetic reflections of hiatus were predicted to show some differentiations, particularly if there is a difference in the way these syntactic sequences prosodify. 3.3.1 Methods 3.3.1.1 Materials To create test items, combinations of feminine DNA and DAN sequences were embedded in initial position of carrier sentences to produce all five vowel combinations aa, ae, ai, ao and au at the word juncture of the definite determiner and the following noun or adjective. Control items were masculine, with a consonant-final determiner. For example, in the ae combination, the DNA test sequence was la etóloga elitista ‘the elitist ethologist’, and the DAN test sequence was la elitista etóloga ‘the elitist ethologist’. The control items for this combination were el etólogo elitista and el elitista etólogo, respectively. Since the position of the adjective is semantically related to contrastive or metaphoric readings, the carrier sentences were modified to give an appropriate context 185 reflecting the meaning change. From the five vowel combinations and two sequencing conditions, 10 test items (with la) and 10 control items (with el) were created. In the effort to restrict the phonetic environment to the appropriate unstressed vowel sequences, surrounded by coronal consonants for the least consonantal formant influences, some of the particular words used were more frequent than others. Specifically, the words alegre ‘happy’, usual ‘usual’ and italiana ‘Italian’ were of high frequency (within the top 5,000 words, cf. Davies 2006), while the others were not 73 . To account for the possible influence of lexical frequency, all items were labeled as frequent (determiner followed by one of the frequent words) or non-frequent (determiner followed by any of the other nouns or adjectives). (See Appendix B for a full list of test and control sentences.) 3.3.1.2 Procedure The same nine speakers were the participants for Experiment 2, as Experiment 2 items were presented with the items from Experiment 1. In this way, Experiment 2 items functioned as fillers for Experiment 1, and Experiment 1 items functioned as fillers for Experiment 2. Two repetitions of the test items were presented, but only one set of control items was included. Participants were recorded using Praat, and the vowels segmented manually using the same criteria outlined for Experiment 1. The formant and duration measurements were extracted from Praat, and analyzed in R. 73 Data analyses were also run with high, medium, and low frequency denominations based on Google word search results, but model comparisons of the different analyses proved the high/non- high frequency label to be most appropriate formost closely match the data. 186 3.3.1.3 Measurements The procedure for measuring was very similar to that of Experiment 1. Disfluent tokens were discarded, and the remainder was produced with the DNA or DAN sequence wrapped in a single PPh. Whether there was a PPh break after the sequence, before the rest of the sentence, was deemed immaterial to the main issue, which is the behavior of the DN or DA juncture. From the acoustic data, the following measurements were taken: (3-15) a. duration of the V1V2 sequence of DA and DN test items b. duration of control V2 in masculine adjective and noun control items c. F1 measurements at four equal intervals of both the V1V2 test sequences and singleton V2 of the control items d. F2 measurements at four equal intervals of both the V1V2 test sequences and singleton V2 of the control items 3.3.2 Predictions Since the hypothesized prosodic clitic structures for DAN and DNA involve a stronger prosodic boundary at DA (Pwd max ) than at DN (Pwd min ), vowels in DA sequences were expected to be longer than DN. In addition, a difference in duration between V1V2 test items and control V2 was expected, supporting vowel maintenance as opposed to vowel deletion. 187 Formant predictions also follow from the hypothesis of vowel maintenance. More co-articulation effects are expected in test sequences of V1V2 in comparison to control sequences of V2, reflecting the underlying presence of both vowels (in comparison to a deletion model). As in Experiment 1, the formant measurements analyzed here are averages from four different intervals of the vocalic sequence. In the test V1V2 sequence, the four intervals span two vowels total, whereas in the control V2 items, the four intervals only span a single vowel. Examples of the four interval divisions are provided below for the test sequence ae in la elitista and the control vowel e in el elitista. Figure I. Wave form and spectrogram for the phrase la elitista ‘the elitist [feminine]’, with the test vowel sequence ae broken up into four equal intervals. 188 Figure J. Wave form and spectrogram for the phrase el elitista ‘the elitist [masculine]’, with the control vowel e broken up into four equal intervals. When comparing the test and control intervals, the most likely difference between the underlying vowels would occur in the first two intervals, since those time intervals would reflect the presence or absence of V1 /a/. In the test sequence, the first two intervals correspond roughly with the underlying /a/, while in the control vowel, the first two intervals correspond with the beginning of the underlying singleton /e/. The prediction regarding comparison of formants at these time intervals is that in the test intervals, the formants will show an influence of the /a/, reflecting a lack of a-deletion, while the control intervals should only reflect formants appropriate for /e/. The clustering of vowels around /a/ values early in the sequence seen in clitics and Pwds of Experiment 189 1 should be repeated in the test items here, but not in the controls. 74 This means that in the mixed effects model, the factor vowel type will have a stronger effect in the controls than in the test items, resulting in an interaction between the two factors. In support of the prosodic distinction hypothesis between DA and DN test sequences, it is also predicted that there should be more formant co-articulation effects at DN than at DA. The weaker hypothesized prosodic boundary of DN should be reflected in an interaction between the factors vowel quality and prosodic boundary as the vowels progress in the time intervals from underlying /a/ from the determiner to the varying V2 in the noun or adjective. Vowel quality is expected to have an effect on formant structure, since different vowels carry different formant values. We expect the initial time intervals to show less effect from the following vowel at the stronger DA boundary, whereas the early intervals are expected to show formants resembling their respective following vowels sooner for the weaker DN boundary. The underlying /a/ vowels in the beginning are predicted to have more formant influence from the following V2 when there is greater coarticulation and gestural overlap. Conversely, in the latter half of the vowel sequence, greater coarticulation is expected to manifest as less distinct formants between the V2 vowels. Coarticulation from the preceding /a/ would centralize the formants, resulting in less interaction effect from the factor vowel quality. Therefore, I assume the following specific predictions regarding formant interaction: in the first two vowel time intervals, greater influence of vowel quality is expected in DN than DA (indicating more 74 Although it can take some time for vowel qualities to manifest, it is expected that control V2 will nevertheless be different from test V1V2 sequences here, because the V1 /a/ is not present in production at all. 190 coarticulation from the following V2); in the second two vowel time intervals, greater influence of vowel quality is expected in DA than DN (indicating less coarticulation from the previous V1). 3.3.3 Results 3.3.3.1 Duration Analysis Durations of vowel sequences were compared in much the same way that unbroken vowel sequences were in Experiment 1. The main difference was that there was no control case for V1, so the comparisons between V1V2 and control V1 were not available. Instead, a comparison between the test V1V2 sequence and control V2 was made. The factors examined were vowel type (aa, ae, ai, ao, au), control vowel (a, e, i, o, u) vs. test vowels, frequency (hi vs. lo) 75 , and syntactic sequencing (DNA vs. DAN). Subject was included as a random effect. 76 There were significant main effects for the factors control/test, adjective/noun, and vowel quality, as reported in (3-19). Test vowel sequences were significantly longer than singleton controls. DN vowels were shorter than 75 As mentioned, it was impossible to fully control for word frequency due to the very specific phonetic contexts surrounding the vowel sequences. Frequency was thus included as a possible factor in order to eliminate lengthening effects due only to a high or low frequency distinction, to focus more accurately on the distinction between DA and DN. The imbalance of high and low frequency also meant that it was impossible to check for interactions between frequency and vowel quality. Such interactions were not predicted for duration (for example, if a were longer in frequent words while o was longer in infrequent words), so a non-interaction model is pursued for duration analyses. But, interactions are predicted in formant analyses. Because the formant analyses are dependent upon interaction with vowel quality, frequency was left out of the formant models. Future research with a design fully balanced for frequency is necessary to comprehensively examine the possible effect of frequency on co-articulation. 76 Since only one word per vowel/syntactic condition was used, word was unnecessary as a separate random effect. (In addition, a model comparison of an analysis with word and subject and one with only subject showed no distinction.) 191 DA vowels, and vowels at low frequency words were longer than those at high frequency words. For inherent vowel duration differences, au was longer than aa. The plots in Figure K show comparisons of test V1V2 sequences and control singletons, and Figure L provides a comparison of DNA and DAN vowels. (3-16) Fixed effects of duration DNA and DAN data (df: 10) Estimate Std. Error t value (Intercept) 66.450 7.680 8.652 Test vs. control 43.110 ms 3.181 13.551* DN vs. DA -17.497 ms 3.128 -5.593* ae vs. aa -2.850 ms 5.318 -0.536 ai vs. aa 2.030 ms 4.663 0.435 ao vs. aa 9.104 ms 5.095 1.787 au vs. aa 11.911 ms 4.617 2.580* Lo vs. hi frequency 11.565 ms 3.941 2.934* 192 Figure K. Durations of V1V2 test items compared to control V2 items, for DAN and DNA sequences. Figure L. Durations of DNA and DAN word sequences (both control and test). 193 3.3.3.2 Formant analysis: V1V2 vs. V2 Initially, the mixed model was run using syntactic sequence DNA/DAN as a factor along with vowel quality and control vs. test, but the difference between DNA and DAN came out insignificant, and a model comparison yielded no significant difference between the 3-factor model and a 2-factor model using only vowel quality and control vs. test vowels. For these reasons, I report the simpler 2-factor model here. Since control items did not include V1 (a), the factor vowel quality included the V1V2 test sequence with its respective V2 control, such that aa and a constituted one level (labeled aa), ae and e constituted another level (labeled ae), etc. Both factors were significant overall, and showed interaction. The results are summarized in Table 35, and the vowel charts compared for test and control vowels in Figure M. Table 35. Summary of F1 and F2 effects for DNA/DAN: V1V2 vs. V2 Time intervals F1 effects 1 2 Vowel quality ae < aa (t=6.3) ai < aa (t=9.6) ao < aa (t=5.1) au < aa (t=6.2) ae < aa (t=6.2) ai < aa (t=8.3) ao < aa (t=3.9) au < aa (t=6.2) Control vs. test V1V2 > control V2 (t=2.0) V1V2 > control V2 (t=2.5) 194 Table 35. Continued Time intervals F1 effects 1 2 Interaction e, o and u have s. lower F1 than a in the control conditions, but ae, ao, and au were not significantly lower than aa in the test condition o has s. lower F1 than a in the control, but ao was not significantly lower than aa in the test condition Time intervals F2 effects 1 2 Vowel quality ae > aa (t=6.3) ai > aa (t=12.4) ao < aa (t= -6.2) au < aa (t= -5.5) ae > aa (t=6.0) ai > aa (t=11.3) ao < aa (t= -5.9) au < aa (t= -5.2) Control vs. test V1V2 < control V2 (t=2.0) V1V2 < control V2 (t=2.5) Interaction Test ae and ai had lower F2 than their e and i control items Test ae and ai had lower F2 than their e and i control items 195 Figure M. Vowel charts comparing test V1V2 sequences with control V2 vowels for the 1 st and 2 nd intervals of DAN and DNA items. Test sequences all consisted of a-V2, but only V2 is labeled for clearer comparison to the control vowels. 3.3.3.3 Formant analysis: DNA vs. DAN While the mixed effects model including test and control data as an additional factor did not find DNA and DAN different in their formant behavior, a closer look at these sequences is presented here with only the test V1V2 data. The mixed effects model was run on V1V2 sequences for both F1 and F2 using the factors vowel quality (aa, ae, ai, ao, au) and syntactic sequence (DNA, DAN), across all four time intervals (df: 12). The results of the statistical model showed the expected significant effect of vowel quality. A difference between DA and DN showed as a main effect only for F1 at the first interval, but interaction between vowel quality and syntactic sequence emerged 196 for F1 and F2 in the second and third intervals (Table 36). The vowel charts in Figure N illustrate the vowel dispersion for DAN and DNA structures. Table 36. Summary of F1 and F2 effects for DNA/DAN: V1V2 Time intervals F1 effects 1 2 3 4 Vowel quality ae < aa (t= -3.9) ai < aa (t= -5.3) ao < aa (t=-2.6) au n.s. ae < aa (t= -4.5) ai < aa (t= -6.7) ao < aa (t= -3.9) au < aa (t= -3.9) ae < aa (t=-3.81) ai < aa (t=-8.4) ao < aa (t=-4.7) au < aa (t=-5.5) ae < aa (t=-3.5) ai < aa (t=-7.4) ao < aa (t=-4.0) au < aa (t=-5.1) DAN vs. DNA DN < DA (t=-2.1) n.s. n.s. n.s. Interaction n.s. ae < aa at DA (p<.01) but not DN ao < aa at DA (p<.01) but not DN ao < aa at DA (p<.01) but not DN n.s. F2 effects 1 2 3 4 Vowel quality ae > aa (t= 4.7) ai > aa (t= 5.6) ao < aa (t= -5.5) au < aa (t= -6.7) ae > aa (t= 6.3) ai > aa (t= 8.9) ao < aa (t= -7.6) au < aa (t= -7.9) ae > aa (t= 5.8) ai > aa (t= 10.7) ao < aa (t= -8.6) au < aa (t= -9.0) ae > aa (t= 5.1) ai > aa (t= 10.2) ao < aa (t= -7.6) au < aa (t= -8.3) 197 Table 36. Continued Time intervals F2 effects 1 2 3 4 DAN vs. DNA n.s. n.s. n.s. n.s. Interaction n.s. n.s. ao higher in DN than ao in DA (p=.02) n.s. 198 Figure N. Vowel charts for V1V2 test sequences of DAN and DNA items at the 1 st , 2 nd , 3 rd and 4 th intervals. While all test sequences consisted of a-V2, only V2 is labeled to show more clearly the influence of V2 on the vowel formants. 3.3.4 Discussion of Experiment 2 The prediction regarding a lack of a-deletion is supported by the results. Test sequences V1V2 were significantly longer than control singleton V2. In the formant analysis regarding hiatus effects, test sequences containing the underlying /a/ had a higher F1 than control vowels in general, and the influence of the (following) vowel was 199 less in test conditions than in the controls. In F2, the direction of difference from /a/ was appropriate for the following vowel quality: e and i had higher F2 in the control cases, where there was no underlying /a/, while o and u had lower F2 in the control cases. The vowel charts in Figure K show how test vowels are clustered closer to the /a/ position of the vowel space, while the control vowels are significantly more dispersed. These results support an analysis of coalescence instead of deletion. 77 In distinguishing between the two syntactic structures DNA and DAN, the results again provide support for a distinction. Duration showed a difference between the two syntactic sequences, with DN vowels shorter than DA vowels, which is consistent with the proposed prosodic boundaries of these structures. Formant analyses were mixed regarding the DN/DA factor. When run together with control data, the factor of syntactic sequence was not statistically significant. But when the model was focused in on just the test V1V2 sequences, some differences emerged. In the first interval, DN vowels had lower F1 overall than DA vowels, which could reflect a stronger co-articulation effect of the following i, e, o or u vowel (all of which have lower F1 than a). In the second and third intervals, F1 interaction between vowel quality and syntactic sequence shows stronger differentiation from aa in the DA case than in DN. In the third interval, the F2 77 Since this formant analysis of comparing test and control determiner items to determine the level of co-articulation influence from /a/ was successful in establishing a difference, it might seem that such an analysis could also establish a difference between test and control items from the dataset of Experiment 1 with clitics and Pwds. However, the different prosodic contexts being tested were far more complex and not as conducive to controls of the same level of consistency to formulate sensible sentences. Controls for clitics and Pwds were all non-hiatus in the inclusion of a consonant, but beyond the consonant there was variability in the following or preceding vowels (see Appendix A). Since vowels may be articulatorily adjacent through an intervening consonant (Gafos 1996), the variability in the outside vowel produced controls that were inappropriate for this type of formant comparison. 200 values of ao are higher in DN than in DA, indicating more co-articulation effects of V1 a. The formant interaction effects overall display more co-articulation effects in DN. Earlier in the sequence the effects are from the V2 on the production of V1 a, and later in the sequence the effects are from V1 a on the varying V2 vowels. Because these interaction effects are relatively subtle, they did not emerge in the larger model examining control vowels, and only appeared when the test vowels (which are the only ones which could have co-articulation effects from /a/) were analyzed separately. 3.3.5 Summary of Experiment 2 The results of Experiment 2 support hypothesis H B of a phonetic hiatus difference between DNA and DAN sequences. There was a lengthening difference in the duration and co-articulation difference in the formants between the syntactic sequences. The direction of both duration and formant effects suggest a weaker prosodic boundary at DN than at DA. In addition, duration and formant analyses comparing test and control vocalic sequences provided support for hypothesis H C with evidence for coalescence, rather than deletion, as an accurate description of phonetic vowel hiatus in these cases. 3.4 Comparison of Experiment 1 and Experiment 2 One of the comparisons available for further illumination of the determiner data is to directly examine the behavior of clitics—which in Experiment 1 were direct object pronouns—with the behavior of the determiner sequences. As discussed in Chapter 2, Spanish determiners are expected to prosodify as clitics, based on previous literature, the 201 predictions of the Prosodic Phonology framework, and evidence from lexical stress of Spanish. The main purpose of comparing clitic and determiner data to each other is to determine how closely DA or DN pattern with verbal clitics. As mentioned in Chapter 2, verbal clitics do not display the same phonological hiatus restrictions as determiners or conjunctions. Therefore, we expect verbal direct object clitics to pattern as free clitics, with the boundary between a verbal direct object (DO) clitic and its following verb Pwd as strong as the Pwd max boundary between D and A. Alternatively, if DA and DN vowel sequences are both different in length or show different formant effects from clitic-Pwd vowel sequences, this might inform a decision to place definite determiners in a different category, apart from clitics. If they are not grouped with clitics in the prosodic hierarchy, then a different prosodic explanation might be more appropriate for the syntactic sensitivity in the feminine el. §3.4.1 compares duration measurements of the two datasets, and §3.4.2 compares formant measurements of the datasets. Predictions: duration Based on the evidence of §3.3, different levels of prosodic cliticization are predicted to be reflected in duration, such that junctures of free clitic-Pwd would have longer V1V2 durations than junctures of affixal clitic-Pwd. The verbal DO clitics are expected to pattern with DA (with a longer duration than DN), and DN to pattern with shorter duration than either DA or verbal clitics. 202 Predictions: formants Similar to the predictions regarding duration differences between clitics and determiners, formant effects are expected to show verbal clitics patterning with DA. Earlier in the vowel sequence, more coarticulation is expected to manifest as greater influence of vowel quality, while in the latter half, greater coarticulation is expected to manifest as less influence of vowel quality, due to centralization tendencies from the preceding underlying /a/. DN is expected to show more of these coarticulation effects than DA. Verbal clitics are expected to match their behavior with DA. 3.4.1 Duration comparison of clitic and determiner data The lmer model included test items of two vowels in an unbroken sequence, and evaluated the factors item type (clitic-Verb, D-noun, D-adjective), vowel quality (aa, ae, ai, ao, au), and word frequency of the lexical verb, noun or adjective. Subject and word were included as random factors in the analysis. In the fixed effects results, summarized in (3-17), vowel quality was not significant. Item type was significant, with sequences across clitic-Pwd boundaries longer than across a DA boundary, and DA longer than DN (by logical extension, clitic-Pwd sequences are also longer than DN). The graph in Figure O illustrates the length of all three item types. 203 (3-17) Duration fixed effects: comparing clitic-Verb, DN, DA Estimate Std. Error t value (Intercept) 104.855ms 14.539 7.212 Clitic vs. DA 25.162ms 9.990 2.519* DN vs. DA -18.205ms 8.397 -2.168* ae vs. aa 7.413ms 11.488 0.645 ai vs. aa 8.715ms 11.133 0.783 ao vs. aa 23.655ms 12.603 1.877 au vs. aa 18.140ms 13.306 1.363 Lo vs. hi frequency 6.904ms 10.111 0.683 204 Figure O. Durations of V1V2 for DA, DN, and clitic sequences. 3.4.2 Formant comparison of clitic and determiner data An analysis of formant effects was also run on the combined data with clitics and determiners. The mixed effects model was run on F1 and F2 using the factors vowel quality (aa, ae, ai, ao, au) and item type (clitic, DA, and DN). Vowel quality had a significant main effect, but item type did not. There was interaction between the factors for F1 at the first three intervals, and for F2 at all four intervals (3-24). The vowel charts in Figure P visually compare the dispersion of vowels in the vowel space for clitics, DN and DA at all four intervals. 205 Table 37. Summary of F1 and F2 effects for clitic-verb, DN and DA Time intervals F1 effects 1 2 3 4 Vowel quality ae < aa (t= -3.6) ai < aa (t= -4.5) ao < aa (t=-2.3) au n.s. ae < aa (t= -4.7) ai < aa (t= -6.8) ao < aa (t= -4.0) au < aa (t= -4.3) ae < aa (t=-3.8) ai < aa (t=-8.3) ao < aa (t=-4.7) au < aa (t=-5.7) ae < aa (t=-3.1) ai < aa (t=-6.7) ao < aa (t=-3.6) au < aa (t=-4.8) Clitic vs. DAN vs. DNA n.s. n.s. n.s. n.s. Interaction ae < aa in clitic and DA, but not in DN ao < aa in clitic, but not in DA or DN au < aa in clitic, but not in DA or DN ae < aa in clitic and DA, but not in DN ao < aa in clitic and DA, but not DN ae < aa in clitic and DA, but not in DN ao < aa in clitic and DA, but not DN n.s. 206 Table 37. Continued Time intervals F2 effects 1 2 3 4 Vowel quality ae > aa (t= 3.7) ai > aa (t= 3.7) ao < aa (t= -3.6) au < aa (t= -4.5) ae > aa (t= 4.1) ai > aa (t= 5.7) ao < aa (t= -4.8) au < aa (t= -5.1) ae > aa (t= 3.9) ai > aa (t= 7.1) ao < aa (t= -5.9) au < aa (t= -6.1) ae > aa (t= 3.5) ai > aa (t= 7.1) ao < aa (t= -5.5) au < aa (t= -5.9) Clitic vs. DAN vs. DNA n.s. n.s. n.s. n.s. Interaction ae > aa in clitic, but not DA or DN ao < aa in clitic & DA, but not DN au < aa in DA, but not clitic or DN ao < aa in clitic & DA, but not DN au < aa in clitic & DA, but not DN ao < aa in clitic & DA, but not DN ao < aa in DA, but not clitic or DN 207 i e a o u Clitic, DA & DN vowels: 1st interval F2 F1 i e a o u F2 F1 i e a o u F2 F1 2600 2200 1800 1400 1000 800 600 400 clitic DA DN i e a o u Clitic, DA & DN vowels: 2nd interval F2 F1 i e a o u F2 F1 i e a o u F2 F1 2600 2200 1800 1400 1000 800 600 400 clitic DA DN i e a o u Clitic, DA & DN vowels: 3rd interval F2 F1 i e a o u F2 F1 i e a o u F2 F1 2600 2200 1800 1400 1000 800 600 400 clitic DA DN i e a o u Clitic, DA & DN vowels: 4th interval F2 F1 i e a o u F2 F1 i e a o u F2 F1 2600 2200 1800 1400 1000 800 600 400 clitic DA DN Figure P. Vowel charts for clitic, DA and DN V1V2 sequences at the 1 st , 2 nd , 3 rd and 4 th intervals. Although these are all sequences of a-V2, only V2 is labelled for clarity. 208 3.4.4 Discussion and summary of comparison The results of direct comparison between clitics and determiners showed support for grouping determiners as low-level prosodic clitics in the prosodic hierarchy, together with verbal clitics. In the duration data sequences of determiner-lexical word (either noun or adjective) appear to pattern differently from clitics. In terms of actual measured values, verbal clitics were the longest, and then DA, and DN was the shortest. Statistically, verbal clitic vowels pattern separately from DA (they are longer), and DN vowels were shorter than DA. The differences in duration seem to indicate gradient distinctions of the prosodic category of clitic. While the status of clitic for verbal direct object pronouns is uncontroversial, determiners paired with both adjectives and nouns demonstrated behavior of an even weaker prosodic boundary than the verbal clitics. It is unclear whether this means that further clitic distinctions are necessary for the prosodic hierarchy, or whether the increased duration of verbal clitics could be due to another factor, such as the relative frequency of verbal clitics to definite articles. While further data is needed in comparing the behavior of determiners and verbal clitics, the duration patterning is consistent with the prosodic analysis of DA being separated by a Pwd max boundary, while DN is separated by a lower Pwd min boundary. The behavior of verbal clitics as longer than DA point to a prosodic clitic structure for the verbal clitics that is separated by a boundary at least as strong as Pwd max . This may also account for why verbal clitics do not show repair of identical stressed vowel hiatus sequences, such as in the phrase lo ódio ‘I hate him’. 209 The formant analyses did not display exactly the behavior predicted between the different intervals. Recall that since the first two time intervals were expected to roughly correspond with V1 (a) and the second two time intervals with V2, the effect of coarticulation was predicted to manifest differently for the different time intervals. Interaction between the prosodic and vowel quality factors for the first two time intervals was expected to correspond to increased coarticulation, such that V2 vowel quality would produces more of an effect upon the formants of V1 at a lower prosodic boundary. On the other hand, a lower prosodic boundary was expected to show less effect of vowel quality at time intervals three and four, since in greater coarticulation would mean more centralization of the vowels closer to /a/, as a result of influence from V1. However, no such distinction between the time intervals was displayed in the results. Instead, DA and clitic conditions regularly showed a greater effect of vowel quality throughout all time intervals, while DN vowel sequences were consistently less distinct from aa. It appears that the shorter DN sequences showed more fusion between V1 a and varying V2, resulting in very low distinction between the two vowels. The longer DA and clitic sequences showed a different strategy, where instead of maintaining a clear V1 a for an equal portion of the sequence, switch sooner to V2 formant qualities (for ae and ao in particular). The unequal distribution of time allotted to V1 a versus V2 is not surprising, considering that the V1 a belongs to a function word, while V2 belongs to a lexical word. In other languages, phonological hiatus repair at a functional-lexical sequence will often favor the lexical vowel (Casali 1998). The unexpected disparity between complete coalescence behavior demonstrated in DN and the sequential short a-long V2 behavior 210 demonstrated in clitics and DA may be accounted for by the difference in duration and proportions. The vowel sequences were divided into four equal intervals, regardless of how long they were. As shown in the duration results, DN was significantly shorter htan DA (and also shorter than clitics). Therefore, interval 1 for DN was shorter than interval 1 for DA and clitics. It may be that the disparity in behavior between DN and DA/clitics is due to the different time within which formants were measured. In a shorter sequence, the dynamical prosodic boundary delay in reaching the V2 target may take proportionately more time of the total sequence, resulting in an effect of more centralization throughout, whereas in a longer sequence, the vowel target(s) of both V1 and V2 would be reached in a sooner interval. Further research comparing the same absolute time may provide additional data for this comparison. The possibility of duration proportions having a strong effect on the vowel formants makes evaluation of deletion versus coalescence difficult. DA and clitic vowels may be more distinct from aa because the V1 a is not attempted in production. Given the he results of control vs. test formant comparisons in Experiment 2 data, as well as visual representation of the charts in Figure P, this is unlikely. There is a gradual dispersion of vowels within the vowel space from the first interval (with F2 values between 2000 and 1400 Hz) to the second interval (F2 values between 2100 and 1300 Hz) to the third and fourth intervals (F2 values between 2300 and 1200 Hz). Although the vowel deletion or coalescence behavior is difficult to determine with this data, there is an effect of clitic type, such that DN behaved quite differently from DA or clitics. The vowels ae or ao, for example, were not differentiated from aa in several 211 intervals of the DN sequence, while they were differentiated in clitics and DA. While such effects were not always consistent for all vowels in all intervals (for example, in interval 1 and interval 4 both clitic and DN sequences patterned together in non- significance of ao and au distinctions), groupings of the clitic and DA conditions accounted for the majority of the interactions. The formant results show DA vowels behaving more like accusative direct object verbal pronouns, while DN vowels patterned separately. In sum, the comparison of clitics, determiners and Pwds indicate that determiners are appropriately labeled as prosodic clitics. In the duration results, DA patterned with shorter duration than recognized clitics, while in formant results they patterned together with verbal clitics. Overall, the results indicate that the label “clitic” is more gradient than previously realized, with the DN context in particular being realized with weaker prosodic boundaries than verbal clitics or DA sequences. The phonetic differences between these sequences provide support for the theoretical distinctions between the possible prosodic clitic structures in (3-5) and for the prosodic analysis pursued in this dissertation of DN and DA as affixal and free clitics, respectively. The four hypothesized prosodic clitic structures predict that four different boundary effects (no boundary, clitic- Pwd min , clitic-Pwd max , Pwd max -Pwd max ) could emerge in phonetic production. The gradiency of production in the direct object pronoun, DN and DA sequences neatly supports these distinctions of prosodic clitic attachment. The specific direction of comparison—with DN patterning as a weaker prosodic boundary than DA—support the theoretical analysis of DN as an affixal clitic with only a Pwd min boundary intervening 212 between the clitic (determiner) and host (noun), while DA is a free clitic structure with a stronger Pwd max boundary intervening between the items. 3.5 Discussion and alternatives The results of the experiments confirm all three hypotheses of the study, reviewed below. (3-18) H A : Spanish productions of phonological vowel hiatus exhibit measurable differences as a function of different levels in the prosodic hierarchy. If the vowels in hiatus span the boundary of a category of the prosodic hierarchy, the productions will differ according to the category level. H B : Spanish productions of phonological vowel hiatus exhibit measurable differences between DNA and DAN syntactic sequences. H C : Phonetic productions of Spanish phonological vowel hiatus result in vowel coalescence, but not vowel deletion. The first hypothesis H A , to confirm Spanish phonetic hiatus correlates of different prosodic boundaries, was well supported in Experiment 1. Duration of category-final V1 differentiates between the larger prosodic boundaries of IP, PPh and Pwd/clitic, and co- articulatory differences in V1V2 hiatus highlight a difference between the Pwd and clitic categories. The second main hypothesis H B , regarding differences between DNA and DAN sequences, received support in Experiment 2 from duration differences and co- 213 articulation effects. The DAN sequence was considerably longer than the DNA sequence, supporting a prosodic distinction in which DA is separated by a stronger prosodic boundary. In V1V2 sequences of DN and DA, formant results showed evidence of more co-articulation in a DN sequence than DA, as well. The question of vowel deletion as a phonetic hiatus strategy, as presented in H C , was answered in the negative. Duration results from Experiment 1 support the rejection of vowel deletion as an adequate descriptor of phonetic hiatus realizations, as did results of Experiment 2. In addition, formant analyses of test and control tokens in Experiment 2 support the rejection of vowel deletion, favoring a coalescence analysis. Direct comparison analyses run on the test items in Experiment 1 and Experiment 2 in §3.4 suggest that both verbal clitics and determiners prosodify as prosodic clitic constituents. Specifically, DA sequences patterned as equal to or weaker than verbal clitics, and DN sequences displayed the effects of an even weaker prosodic boundary, particularly in the duration analysis. These distinctions support the analysis of DN at a Pwd min boundary and DA at a Pwd max boundary. As discussed, the duration results in particular showed support for the prosodic distinction between DA and DN, one of the primary goals of the study. However, an alternative explanation for the duration differences found should be recognized. Since word frequency was found to play a role (albeit a small one) in duration of vowel hiatus sequences, the longer duration of DA sequences could also be attributed purely to frequency considerations. Rather than frequency of the lexical item, however, it could be attributed to frequency of syntactic structure or bigram sequencing. A prenominal 214 adjective structure, in general, is less frequent in Spanish than a postnominal adjective structure or an NP with no modifying adjectives at all. This means that most DA sequences are less frequent than most DN sequences. Bigram frequency measures on the determiner-lexical word sequence may prove to account for the duration distinction, without reference to a weaker or stronger prosodic boundary. One problem with this interpretation is that the influence of syntactic frequency (or bigram frequency) and prosody are very difficult to tease apart. Lower frequency could very well be an influencing factor on the development of different prosodic structures. Future research on the relationship between frequency and prosodic structure may yield further insight to this issue. 3.6 Summary of Chapter Three In sum, differentiations in Spanish prosodic structures have been shown in phonetic manifestations of vowel hiatus, in keeping with previous research on other languages as well as work on Spanish prosody in non-hiatus conditions. In the established prosodic structures of IP, PPh, Pwd and (object) clitic, no evidence was found for vowel deletion as a hiatus production, contrary to previous description. Although it is a possibility that some speakers may in fact delete vowels in this context and that the participants of the study were simply not among this number, it is unlikely that deletion is a widely utilized strategy in the production of hiatus. In the variety of prosodic contexts examined here, including low level clitic-Pwd boundaries, formant and duration evidence suggest instead that vowel gestures are overlapped in fluid speech. Regarding the 215 viability of a prosodic analysis of the DN and DA distinction in the feminine el in Spanish, evidence was found in support of difference prosodic structures for each of these sequences. DA vowels were found to have longer duration than DN vowels and less co- articulation. While still patterning weaker than verbal clitics, DA and DN sequences are well within the realm of production for prosodic clitics. 216 Chapter Four: Further case studies The feminine el in Spanish is not the only phonological phenomenon that shows sensitivity to the prosodic cliticization and the Pwd max domain, adjacency restrictions, or morpho-phonological interaction. This chapter reviews prosodic cliticization phenomena of the syntax-phonology interface that show similarity to various aspects of the feminine el pattern. Russian pronoun allomorphy, addressed in §4.1, shows a distinction between the affixal and free cliticization in the prosodic structure. Catalan determiner allomorphy in §4.2 shows effects of the same Minimum Distance family of constraints on adjacent segments that was proposed in Chapter 3 for the Spanish feminine el and conjunction. Coda gliding in the Spanish dialect of Cibaeño is presented in §4.3, providing an example of further adjacency effects at the prosodic clitic-host context. Section §4.4 reviews the proposed morpho-phonological correspondence in analyzing morphological gender change and non-prescriptive gender variation of the Spanish feminine el, and §4.5 summarizes. 4.1 Russian pronoun allomorphy Like the Spanish feminine el phenomenon, Russian pronoun allomrophy shows sensitivity to the internal head-complement relation in the syntax-prosody mapping. Russian third person pronouns [ix] ‘them’, [jemu] ‘him [dat]’ [jego] ‘him [acc]’, and [jejo] ‘her [acc]’, have n-initial allomorphs when following a preposition (4-1a). However, this alternation is blocked when the pronoun is used as a possessive adjective, instead of the head of the following DP (4-1b). The data are from Billings (1996) and 217 Nevins (2011), with the syntactic structure supplied by the author. Following Abney (1987), Kayne (1993, 2000), and Adger (2003), I assume that possessives are in the specifier of a larger DP with a null head, and that pronouns are head Ds. 7879 (4-1) a. [ PP bez [ DP nix]] ‘without them’ [ PP ku [ DP nemu]] ‘to him [dat]’ [ PP v [ DP nego]] ‘in him’ [ PP v [ DP nee]] ‘in her’ b. [ PP bez [ DP ix D 0 [ NP brata]]] ‘without their brother’ [ PP k [ DP jevo D 0 [ NP domu]]] ‘toward his house’ [ PP v [ DP jego D 0 [ NP dome]]] ‘in his house’ [ PP v [ DP jejo D 0 [ NP dome]]] ‘in her house’ According to the historical analysis of this allomorphy in Billings (1996), the phonological condition for the n-initial allomorph stems from historically nasal-final prepositions whose nasal coda disappeared in most contexts, but when combined with palatal-initial pronouns became a nasal onset for the pronoun. After analogy, this resulted in the pronouns having an /n/-initial allomorph after prepositions (even a consonant-final preposition such as bez), but a /j/- (or for ix, zero-) initial allomorph in all other contexts. 78 This analysis of possessives is supported by the possibility of an overt D in possessive structures, such as English his Poss every D wish NP , or Hungarian Peter valamannyi kalap-ja lit. ‘Peter’s each hat-[def]’ (data from Adger 2002). 79 The possessive itself is part of its own DP, which is larger in the case of an NP possessive instead of a single pronoun. 218 The n-initial allomorph was first triggered when it was in a Pwd together with the preposition, suggesting the prosodic structure in (4-2a). When the n-initial allomorph does not occur, I assume a recursive free clitic structure of the preposition and possessive pronoun, as shown in (4-2b). 80 (4-2) a. (! max bez (! min nix)) affixal clitic b. (" max bez (" min ix (! min brata)) recursive free clitics In an internal head relation approach to the syntax-prosody mapping, the generation of the different prosodic clitic structures in (4-2) is straightforward. For the examples bez nix and bez ix brata, the n-initial pronoun nix stands in the head of the complement to the PP bez (4-3a) and leads to an affixal clitic structure within the Pwd max (4-3b). (4-3) a. [ PP bez [ DP nix HComp ]] b. (! max bez (! min nix)) 80 Alternative structures for the bez ix brata sequence include a free+affixal clitic structure of (" max bez (! max ix (! min brata))) and a free+Pwd clitic structure of (" bez (! max ix) (! max brata)). The current analysis current analysis assumes that Russian function words in general, including the possessive pronoun ix, prosodify in a free clitic structure, with the preposition affixal clitic structure occurring only in the case of a head-complement adjacency. If function words generally prosodify as free clitics, but the possessive prosodifies as an affixal clitic, an analysis involving lexical indexation would be appropriate (much like the analysis proposed for Spanish conjunction in Chapter 2). Ultimately, whether the possessive ix prosodifies as an affixal, free or Pwd clitic is a matter for further empirical research. 219 The vowel-initial pronoun ix is not the head of the complement to the PP (4-4a) and therefore the preposition forms a free clitic structure (4-4b) (where ix is also a free clitic attached to brata). (4-4) a. [ PP bez [ DP ix Spec D 0 [ NP brata HComp ]]] b. (" max bez (" min ix (! max brata))) The same PARSE FNC -INTO-! HCOMP constraint proposed in Chapter 2 for the Spanish feminine el accounts for Russian prosodification of the structure in (4-3). Ranked above NONRECURSIVITY!, the affixal clitic structure is generated when the preposition clitic is adjacent to its complement head, as shown in Table 38. Table 38. PARSE FNC >> NONRECURSIVITY! [ PP bez [ DP ix HComp ]] 81 PARSE FNC -INTO-! HCOMP NONRECURSIVITY! # a. (! max bez (! min nix)) affixal clitic * b. (" bez (! max nix)) free clitic *! 81 Although the phonological constraints driving use of the n-initial allomorph are not shown in this tableau, the underlying /ix/ form is used with the resulting [nix] allomorph in the candidate set, to clarify where the n-initial allomorph is attested and where it is not. 220 Candidate (a) adjoins the preposition to its lexical complement head at the Pwd max level, forming an affixal clitic structure and obeying PARSE FNC -INTO-! HCOMP . In contrast, the free clitic structure in (b) violates the syntactic relation constraint and is ruled out. When the preposition is removed from its complement head, as in when the pronoun functions as a possessive in (4-4), the ranking of NONRECURSIVITY! and PARSE-! produce a free clitic structure for all adjoined material, as shown in Table 39. Table 39. PARSE FNC -INTO-! HCOMP >> NONRECURSIVITY! >> PARSE-! [ PP bez [ DP ix Spec D 0 [ NP brata HComp ]]] PARSE FNC - INTO-! HCOMP NONRECURS! PARSE-! a. (! max bez (! min ix)) (! max brata)) affixal clitic * *! # b. (" max bez (" min ix (! max brata))) recursive free clitics * ** In Table 39, PARSE FNC assigns a violation to both candidates, since the functional preposition is not parsed in its lexical complement Pwd. 82 Candidate (a) includes affixal 82 A fourth possible candidate, which would obey PARSE FNC , involves recursive affixal cliticization, where the possessive ix and preposition bez are prosodified as affixal clitics: (! max bez (! ix (! min brata))). This candidate could be ruled out by prosodic well-formedness constraints limiting the size of a Pwd. Since the candidate obeys binary branching, it does not violate the constraint BIN-MAX(!), utilized in the analysis of Spanish. Instead, the size limitation here could take the form of a constraint like MAXLAYER-!, limiting Pwds to only two recursive layers (minimal and maximal), or a constraint like *PCat Non-min/non-max , forbidding intermediary category levels. 221 cliticization to the possessive ix, but is ruled out by NONRECURSIVITY!, leaving the free clitic candidate in (b) to win out. The ranking is consistent with the affixal clitic structure in preposition-pronoun sequences, as shown in the full tableau in Table 40. Table 40. Full Russian allomorphy i. [ PP bez [ DP ix Head ]] PARSE FNC -INTO- ! HCOMP NONRECURS! PARSE-! # a. (! max bez (! min nix)) affixal clitic * b. (" bez (! max nix)) free clitic *! * ii. [ PP bez [ DP ix Spec D 0 [ NP brata Head ]]] a. (! max bez (! min ix)) (! max brata)) affixal clitic * *! # b. (" max bez (" min ix (! max brata))) recursive free clitics * ** The above head-complement relation approach to Russian cliticization, utilizing PARSE FNC -INTO-! HCOMP , is consistent with the Spanish feminine el pattern. In both languages, a clitic adjacent to the head of its lexical complement is required to adjoin within the Pwd max level, while when it is adjacent to a non-head, the ranking of other constraints in the hierarchy determines a free clitic structure. Also like the Spanish 222 feminine el, the head-complement approach to the prosodic mapping in Russian is shown to advantage in comparison to a purely structural Match account. 4.1.1 Russian allomorphy in a structural account A structural Match approach to the Russian data encounters a few issues. In order to produce a distinction between the preposition-possessive and preposition-pronoun sequences, a cross-category MATCH(LEXP, PWD MAX ) constraint like that suggested in §2.2.2.4.2 might be used, enforcing a match between the lexical XP level and a Pwd max prosodic constituent. This constraint will provide a distinction between the two sequences. In both structures, a functional DP boundary intervenes between the preposition and the following word. However, with the presence of an NP complement in the possessive structure, a maximal Pwd is constructed around the noun, 83 forcing both the possessive and preposition into recursive PPhs. Table 41 demonstrates. 83 I assume that Russian does not have N- or NP-raising within the DP, since Russian default adjective ordering is prenominal, for example: moi D bolʃoi A krasnij A ʃarik N ‘my big red balloon’. In the only postnominal adjective construction, the noun appears to be in a focus position, preceding both the adjective and the determiner: ʃarik N moi D bolʃoi A krasnij A lit. ‘balloon my big red’. 223 Table 41. Russian allomorphy with MATCH(LEXP, PWD MAX ) i. [ PP bez [ DP ix]] MATCH(LEXP, PWD MAX ) PARSE-! NONREC! #a. (! max bez (! min nix)) affixal clitic * b. (" bez (! max nix)) free clitic *! ii. [ PP bez [ DP ix [ NP brata]]] a. (! max bez (! ix (! min brata))) affixal clitic *! ** # b. (" max bez (" min ix (! max brata))) recursive free clitics ** * Candidate (ia), the affixal clitic structure, vacuously satisfies MATCH(LEXP, PWD MAX ) because there is no lexical XP that must be matched to a maximal Pwd, and the constraint ranking of PARSE-INTO-! >> NONRECURSIVITY! rules out the free clitic structure of (ib). For candidates (iia), the noun brata is not contained in a Pwd max , and so it violates the Match constraint. Instead, candidate (iib) wins with a series of two free clitics, allowing brata to match the Pwd max exactly. While the MATCH(LEXP, PWD MAX ) produces a difference between preposition- pronoun and preposition-possessive, a problem emerges when we consider possessive constructions in which the noun has elided (4-6). Like Spanish noun elision in DPs (see 224 §2.5), Russian may elide the noun and leave the possessive alone in the DP (data from Billings 1996, cited from Ferrell 1958 and Hill 1977). (4-6) Oni ʒivut ne v jego dome, a v jee. They live not in his poss house but in hers poss ‘They live not in his house, but in hers.’ Both bolded words are possessives, not pronouns. In particular, the feminine jee is not /n/-initial, despite the fact that its complement noun is elided. If the recursive free clitic structure of (4-5b) is taken to be accurate for possessives, the elided noun forms should also follow this construction with recursive PPhs. The problem is that after elision, the noun is not present, and cannot be matched to a maximal Pwd (4-7). (4-7) [ PP v [ DP jejo D 0 [ NP dome]]] # (" max v (" min jejo (! max ?))) Following Elfner (2011), the syntax-prosody mapping should not produce empty prosodic structures where there is no phonological material. The Match constraint above will not require a Pwd max for an empty NP, and instead the lower-ranked constraints are left to determine the winning candidate. As shown in Table 42, this will produce an affixal clitic structure for possessives with elided nouns, which would trigger the /n/- initial allomorph in the phonology. 225 Table 42. Failure of MATCH(LEXP, PWD MAX ) for Russian elided nouns [ PP v [ DP jejo [ NP dome]]] ‘in hers’ MATCH(LEXP, PWD MAX ) PARSE-! NONRECURSIVITY ! !a. (! max v (! min nejo)) affixal clitic ** b. (" v (! max jejo)) free clitic *! * c. (" max v (" min jejo)) recursive free clitics, no Pwd **! * Without a noun, candidates (b) and (c) do not violate MATCH(LEXP, PWD MAX ), and the affixal clitic candidate (a) will win. As seen here, the Russian data poses considerable problems for a purely structural Match approach, but is easily accounted for in a mapping theory that has access to the internal head relation. The Russian data provides additional support for the PARSE FNC approach to cliticization, mirroring the analysis in Spanish. In both languages, adjacency to the complement head triggers cliticization within a maximal Pwd while a non-head allows cliticization across a maximal Pwd boundary. 4.2 Catalan determiners Catalan shows support for another aspect of the feminine el analysis proposed in this dissertation, namely phonological adjacency restrictions at prosodic cliticization. It is 226 specifically sensitive to the same Minimum Distance constraints on adjacent vowel segments that were proposed in Chapter 2. An analysis is pursued in §4.2.1 for the dialect of Western Catalan, and notes on Central Catalan follow in §4.2.2. 4.2.1 Western Catalan Western (W.) Catalan has a seven vowel inventory /a, e, $, ɔ, o, i, u/, which reduces to the five vowel inventory /a, e, o, i, u/ in unstressed syllables (Herrick 2004, Wheeler 2005). Like Spanish, the regular Catalan feminine determiner ends with the vowel /a/ (4-9a). In combination with vowel-initial nouns, the final /a/ of the determiner produces the possibility of hiatus. Unlike Spanish, all sequences with a stressed vowel are disallowed, and the determiner vowel is deleted to avoid hiatus (4-9b). When both vowels of the underlying determiner-noun sequence are unstressed, most vowel pairs are disallowed (4-9c). However, /i/ and /u/ vowels block deletion and allow full manifestation of the determiner’s /a/ vowel (4-9d). 84 (Data from Wheeler et al. 1999, Orbis Latinus 2002, Catalan dictionary and a native speaker informant.) 84 Like Spanish, Catalan also has exceptions to the determiner vowel hiatus pattern. Letters of the alphabet take the full form la instead of hiatus-avoiding l’. Other exceptions include words that would be homophonous with other existing words in the lexicon (for example, la anormalitat/*l’anormalitat ‘the abnormality’ contrasts with existing la normalitat ‘the normality’), and independent lexical exceptions la una ‘one [o’clock]’, la host ‘the host’, and la ira ‘the anger’. 227 (4-9) Catalan determiners a. la cama ‘the leg’ la filla ‘the daughter’ la mandra ‘the laziness’ la torre ‘the tower’ la bèstia ‘the beast’ b. l’aigua, *la aigua [láj.gwa] ‘the water’ l’era, *la era [lé.ra] ‘the era’ l’èmula, *la èmula [l$ !.mu.la] ‘the (female) rival’ l’òbra, *la òbra [lɔ́.bra] ‘the work’ l’honra, *la honra [lón.ra] ‘the honor’ l’índole, *la índole [lín.do.le] ‘the (emotional) character’ l’úlcera, *la úlcera [lúl.se.ra] ‘the ulcer’ c. l’amiga, *la amiga [la.mí.ga] ‘the (female) friend’ l’entitat, *la entitat [len.ti.tát] ‘the entity’ l’honradesa, *la honradesa [lon.ra.d$ !.za] the honesty l’obrera, *la obrera [lo.bré.ra] ‘the (female) worker’ l’emulació, *la emulació [le.mu.la.sjó] ‘the rivalry’ d. la unitat, *l’unitat [la.u.ni.tát] ‘the unity’ la idea, *l’idea [la.i.dé.a] ‘the idea’ 228 The vowel adjacency restrictions in W. Catalan are specific to prosodic cliticization of the DP, but they are not sensitive to the category of the lexical host. Prenominal adjectives behave the same way that nouns do, allowing unstressed a-i and a- u hiatus but disallowing other vowel pairs. (4-10) la incendiària mescla, *l’incendiària mescla ‘the incendiary mixture’ l’emotiva història, *la emotiva història ‘the emotional history’ l’híbrida planta, *la híbrida planta ‘the hybrid plant’ In sum, W. Catalan allows hiatus of unstressed /a-i/ and /a-u/ at the juncture of determiner cliticization, but it disallows all other sequences. Prosodic cliticization is not sensitive to syntactic structure. Instead, I propose that Catalan determiners take the affix clitic structure within the Pwd max , regardless of whether the host word is a noun or adjective. The identical prosodic structures for determiner-noun and determiner-adjective cliticization also predict identical phonetic manifestation. The increased vowel duration found in Spanish DA sequences would not be predicted for Catalan. (4-11) [ PPh [ Pwdmax la[ Pwdmin idea]] [ Pwdmax incendiària]] [ PPh [ Pwdmax la[ Pwdmin incendiària]] [ Pwdmax mescla]] The family of MINDISTVV constraints used in §2.3 that were sensitive to perceptual contrast in adjacent vowel sequences in Spanish also motivate the Catalan 229 pattern of hiatus avoidance. The only vowels allowed to be in hiatus—ai and au—are those that are furthest apart on the scale of perceptual contrast, presented in (4-12). (4-12) W. Catalan vowels along the F1/F2 dimensions: F1 i u 1 2 3 e o 4 $ ɔ 5 6 a 7 F2 6 5 4 3 2 1 Within the Minimum Distance family, the constraint relevant for the W. Catalan pattern is that which allows ai and au but forbids all less contrastive pairs. The context for this constraint is PWD MAX , which specifies sensitivity within the maximal Pwd, preventing repair across separate lexical items. In addition, I assume faithfulness to lexical items (Casali 1998; Geirut et al. 1999; Smith 2011)—represented here as FAITH LEX —is top-ranked in Catalan, allowing functional words to repair the MINDISTVV violation, but preventing full lexical words from undergoing repair. This also inhibits 230 repair to word-internal violations, such as that represented by hiatus of e and a in the word idea. (4-13) MINDISTVV(PWD MAX )=F1/F2:6 – V 1 and V 2 within a Pwd max must be at least 6 values apart on the F1 or F2 dimension. Ranked above MAX-IO, which forbids segment deletion, MINDISTVV(PWD MAX )=F1/F2:6 prevents all vowel hiatus except for [a-i] and [a-u], as shown in Table 43. As mentioned in §2.3.1, the constraint MINDISTVV(PWD MAX )=F1/F2:7 functions in a general NO HIATUS effect, since it requires a level of perceptual distance that is impossible within the constraints of the vowel space. 231 Table 43. W. Catalan MINDISTVV and vowel deletion i. la idea FAITH LEX MINDISTVV (PWD MAX )=F1/F2 :6 MAX-IO MINDISTVV (PWD MAX )=F1/F 2:7 # a. la idéa * ** b. l’idéa * *! * c. l’idé *! ii. la obrera a. la obréra *! * # b. l’obréra * iii. la emulació a. la emulació *! ** # b. l’emulació * * Candidate (ia) has vowel hiatus of a-i and e-a. The pair a-i (along with a-u) is the most distinct on the F1 scale, and passes the required minimum distance for Catalan vowel hiatus. The pair e-a violates Minimum Distance, but is saved from repair because it belongs to the lexical word idea (compare to candidate (ic). The faithfulness constraint MAX-IO rules out the determiner deletion form (ib). With the candidates for the inputs in (ii) and (iii), the deletion form wins out, since all other less contrastive vowel combinations violate MINDISTVV=F1/F2:6. 232 Also like Spanish, stress produces a markedness augmentation effect to trigger the drive for a greater perceptual boundary between the two vowels. It increases hiatus sensitivity. In the Spanish feminine el, stress increases sensitivity from allowing all hiatus (in unstressed pairs) to restricting identical vowels (in unstressed-stressed pairs). In Catalan, it increases sensitivity from allowing the most contrastive vowel hiatus (in unstressed pairs) to restricting all hiatus (in unstressed-stressed pairs). Here, the positional markedness context ('V) applies to the MINDISTVV constraint forbidding all hiatus: MINDISTV'V(PWD MAX )=F1/F2:7. This constraint enforces deletion of all hiatus at the unstressed-stressed vowel juncture. Table 44. W. Catalan and stress 85 i. la índole MINDISTV'V(PWD MAX ) =F1/F2:7 MINDISTVV (PWD MAX )=F1/F2:6 MAX-IO a. la.ín.do.le *! # b. l’ín.do.le * ii. la idea #a. la.i.dé.a * b. l’i.dé.a * *! 85 While the tableau in Table 44 does not show direct evidence for ranking positionally marked MINDISTV'V(PWD MAX )=F1:7 over MINDISTVV(PWD MAX )=F1:6, I assume that all positionally marked constraints are universally ranked higher than their non-positionally marked counterparts (see §2.3.4 and Smith 2005). 233 Candidate (ia) violates the MINDIST constraint sensitive to stress, since it has hiatus into a stressed vowel. The grammar selects the deletion form in candidate (ib) instead. The vowel hiatus candidate (iia), however, vacuously satisfies MINDISTV'V because both vowels are unstressed. It wins over (iib) by obeying faithfulness MAX-IO. The Minimum Distance effects seen here in Catalan are restricted to prosodic clitics via the Pwd max domain and lexical faithfulness. They do not occur across all lexical items, but instead only affect function words that cliticize to the following lexical item, like the determiner. Another function word that follows the MINDISTVV restrictions is the preposition de ‘of’ (which becomes d’). (4-14) de Catalonia ‘from Catalonia’ d’Italia, *de Italia ‘from Italy’ The behavior of this additional function word provides additional evidence for the use of a Minimum Distance schema, rather than an alternative analysis utilizing OCP[-hi], for example. A constraint like OCP[-hi] could account for hiatus restrictions in the determiner vowels of aa, ae, ao, a!, and aɔ. It would allow ai and au, since these vowels are not both [-hi]. However, such an analysis also predicts that the hiatus vowels ei and eu should be acceptable, since these also differ in the feature [-hi]. But, the function word de does not maintain its e vowel in hiatus even if the second vowel is a [+hi] vowel i or u (4-15). The e is always deleted in contact with other vowels. The initial 234 vowels in (4-15) are also unstressed, and would be expected to tolerate hiatus according to a [-hi] analysis of the determiner cases. (4-15) d’Italia ‘from Italy’ d’Uruguai ‘from Uruguay’ The Minimum Distance approach correctly predicts that sequences of e and a high vowel are forbidden just as sequences of a and all non-high vowels are forbidden. Rather than relying on feature distinctions, it is the perceptual contrast that is relevant. The vowel e can never combine with another vowel of the vowel space in a way that satisfies MINDISTVV=F1/F2:6. It is always too close in relative perceptual distance. 4.2.2 Central Catalan Central Catalan follows much the same pattern as Western Catalan in terms of vowel deletion in determiners. However, in Central (C.) Catalan a more extreme pattern of unstressed vowel reduction demonstrates an opacity effect in determiner vowel deletion. Unlike W. Catalan, which maintains a five-vowel inventory in unstressed syllables, C. Catalan reduces to three with full phonetic neutralization (Herrick 2004). 235 (4-16) Neutralization in C. Catalan Stressed Unstressed i i e $ a " ɔ o u u Among these unstressed vowels (which is the only context in which hiatus is allowed), the most dispersed possible forms of hiatus are !-i and !-u. The MINDISTVV constraint that allows this sequence is MINDISTVV=F1/F2:3. 86 86 MINDISTVV=F1:1 would also work here, as the constraint specifies that a distance of one or more is needed. Any combination of ! with the other unstressed vowels yields a contrast of more than one, so they would satisfy MINDISTVV=F1:1, while a combination of !-!, i-u, i-i, or u-u would violate it. 236 (4-17) C. Catalan vowels along the F1/F2 dimensions F1 i u 1 2 3 e " o 4 $ ɔ 5 6 a 7 F2 6 5 4 3 2 1 For the most part, the constraint interaction functions the same way as in W. Catalan, with MINDISTV'V preventing any hiatus with stressed vowels, and MINDISTVV=F1/F2:3 allowing !-i and !-u. 237 Table 45. Central Catalan MINDISTV'V i. la index MINDISTV'V (PWD MAX )=F1/F2:7 MINDISTVV (PWD MAX )=F1/F2:3 MAX-IO a. l".ín.d"ks *! # b. l’ín.d"ks * ii. la idea #a. l".i.dé." b. l’i.dé." *! iii. la emulació a. l".".mu.l".ció *! # b. l’".mu.l".ció * iv. la aigua a. l".áj.gw" *! # b. l’áj.gw" * The candidate sets (i) and (iv) show vowel hiatus ruled out by the stress marked MINDIST constraint. Even though the !-i, !-u, and !-a sequences are allowed by MINDISTVV(PWD MAX )=F1/F2:3, stress marked MINDISTV'V does not tolerate any hiatus, ruling out all !-a sequences (since this sequence is only possible if the a is stressed), as well as stressed !-í and !-ú. The candidate set in (ii) shows the preference for maintenance of the determiner vowel in the unstressed condition, and (iii) illustrates the 238 regular determiner vowel deletion for unstressed vowels which do not meet the MINDISTVV(PWD MAX )=F1/F2:3 qualification. There is an additional complication with derived vs. underlying vowels. Prescriptively, underlying /o/- and /ɔ/-initial nouns still trigger deletion of the determiner vowel, instead of maintaining the !-u sequence. Despite phonetic neutralization, the derived [u] does not trigger hiatus maintenance (prescriptive pattern description from native speaker). (4-18) /la ɔˈbrera/ l’obrera, [lu.ˈbre.r"]/*[l".uˈbre.r"] the (female) worker /la onraˈd$za/ l’honradesa [lun.r".ˈd$.z"]/*[l".un.r".ˈd$.z"] the honesty This pattern operates over the underlying vowels rather than the surface neutralized vowels. However, speakers may occasionally still produce nonprescriptive hiatus at the derived environments, for example la orella [l".u.rɛ!.ʎ"] ‘the ear’ instead of prescriptive l’orella [lu.rɛ́.ʎ"]. There are a few explanations that could account for this apparent opacity in vowel deletion in determiners, as well as native speaker “error” in hiatus production on derived environments. One is that vowel reduction could occur in a later post-lexical stage of the grammar, while determiner vowel deletion operates over the full, unreduced underlying vowels. In this scenario, the MINDISTVV constraints pertaining to determiner vowel deletion would be the same as those for W. Catalan (MINDISTV'V(PWD MAX )=F1/F2:7 and MINDISTVV(PWD MAX )=F1/F2:6), and unstressed vowel reduction would occur at a 239 separate stage of the grammar. Another approach is a diachronic explanation, which might better explain speakers’ tendency to produce non-prescriptive hiatus in derived environments. Determiner vowel deletion might have occurred at an earlier stage of the language’s development (4-19a), before the dialects split in their treatment of vowel reduction (4-19b and c). (4-19) a. l[a ɔ]rella # l’orella vowel deletion la unitat # la unitat b. l’orella # l’[u]rell["] vowel reduction (C. Catalan) la unitat # l[".u]nitat c. l’orella # l’[o]rell[a] vowel reduction (W. Catalan) la unitat # l[a.u.]nitat Non-deletion forms such as la idea and la unitat in C. Catalan could trigger re- interpretation of the relevant Minimum Distance constraint from MINDISTVV=F1/F2:6 to MINDISTVV=F1/F2:3, accounting for the !-i and !-u vowel pairs in the current grammar. Extension and regularization of hiatus maintenance on surface forms would then account for non-prescriptive la orella. This approach predicts that hiatus maintenance would occur on all words beginning with underlying mid-vowels /o/ and /ɔ/. 87 87 The front vowels /e/ and /$/ are not expected to show hiatus maintenance, since they reduce to schwa, and MINDISTVV=F1:3 forbids the sequence !-!. 240 Another possible explanation for hiatus in derived environments is that the “underlying” mid vowels for these forms are not actually the underlying forms posited by speakers. For some forms, such as obrera ‘worker’, a morphologically related form (òbra ‘work’) demonstrates the underlying vowel /ɔ/ in the stressed syllable, allowing speakers to recover the underlying vowel in unstressed syllables via analogy. But for other words, such as orella ‘ear’, morphologically related forms with stress on the vowel in question do not exist or are very infrequent. In these cases, speakers may posit underlying forms that match the surface reduced vowel, for example /urɛ!#"/. This would account for the production of hiatus for certain words. Both Western and Central Catalan show similar Minimum Distance constraints on vowel hiatus as those in Spanish, though at the opposite end of the spectrum. Spanish only forbids identical hiatus—the least contrastive—while Catalan forbids all but the most contrastive hiatus. Stress plays a role in both languages as well. The Minimum Distance constraint positionally-marked for stressed vowels is highly restrictive, preventing any hiatus with stressed syllables. In combination with the MINDISTVV constraint allowing contrastive hiatus, the Catalan pattern of unstressed ai/au tolerance is obtained. In Spanish, the positional markedness stressed vowel context is associated with the same MINDISTVV constraint that forbids identical hiatus, enforcing repair only in the case of identical stressed hiatus. For both Catalan dialects and Spanish, the activity of Minimum Distance on hiatus is specified to the prosodic Pwd max domain, restricting phonological hiatus repair to the affixal clitic prosodic structure. 241 4.3 Cibaeño Spanish Cibaeño Spanish is a Caribbean dialect that also shows segment adjacency restrictions at a prosodic clitic juncture. In Caribbean dialects, coda consonants show various weakening effects, including /s/ aspiration or deletion and /n/ velarization. For Cibaeño, a dialect spoken in the Dominican Republic, laterals and rhotics optionally form palatal glides in coda position (4-20a) 88 . Even when followed by a vowel-initial word, coda gliding still occurs in most contexts (4-20b). An exception remains with the determiner-noun context (4-20c). 8990 (Data from Harris 1983.) 88 Glides also occur phonemically in onset position in the language, and will be addressed later in this section. 89 No data was provided regarding prenominal adjectives in this phenomenon. There are two possibilities with regards to determiner-adjective sequences. They may pattern separately from the determiner-noun sequences, as in the prescriptive Spanish feminine el, and Cibaeño Spanish would then be predicted to have glides in determiner-adjective cases, such as e[j] alto aviso ‘the high warning’. Or, it may pattern like Catalan and adjectives would behave the same as the nouns and block gliding: e[l] alto aviso. In the first scenario, Cibaeño would maintain the prosodic cliticization distinction sensitive to syntactic structure. In the second scenario, Cibaeño would be predicted to prosodify determiner-noun and determiner-adjective sequences exactly the same, with no slow-down effect at the boundary. This second scenario also predicts that in Cibaeño the feminine el distinction between nouns and adjectives, if present, would be non-productive, a memorized pattern with a closed set of words resulting solely from the historical derivation of the language, as suggested in §2.4. Further data on the productivity of the feminine el in Cibaeño would be needed to further explore this possibility. 90 Examples of glide blocking for r are not available, since Spanish determiners do not end in r. However, the analysis given here predicts that if function words such as por ‘for’ attach as affixal clitics to the following lexical word, Cibaeño speakers would preserve the r in this word when it precedes a vowel-initial word. 242 (4-20) Coda gliding of l and r # [j] a. papel blanco > pape[j] blanco ‘white paper’ mar > ma[j] ‘sea’ él da > é[j] da ‘he gives’ el día > e[j] día ‘the day’ b. papel azúl > pape[j] azú[j] ‘blue paper’ él avisa > é[j] avisa ‘he warns’ c. el aviso > e[l] aviso ‘the warning’ As discussed in §2.2, the determiner in a phrase like el aviso is a prosodic clitic that attaches to the following host word at the Pwd level, forming a maximal Pwd. In contrast, the pronoun él stands on its own as a Pwd in and of itself, receiving lexical stress and pitch tone. As with the feminine el alternation, phonological constraints on segment sequencing operate over the Pwd max positional context to trigger or block coda gliding in Cibaeño. Like the Minimum Distance constraints for the Spanish feminine el and Catalan determiner allomorphy, perceptual contrast plays a role in segment adjacency in Cibaeño glide blocking. Instead of proscribing vowel-vowel sequences, however, here the issue is which consonant-vowel sequence is perceptually better. Many linguists have made use of a markedness scale of consonant types for syllable onsets, often couched in terms of sonority (Prince & Smolensky 1993/2004, Smith 2005, and others). In this spectrum, voiceless obstruents are the most preferred as onsets and glides or semi-vowels least 243 preferred. Voiceless obstruents have been found to provide the most salient acoustic cues for a CV transition, while glides provide the least contrastive transition (Delgutte 1997). For this analysis, I adapt Prince and Smolensky’s (1993/2004) scale of ideal onset constraints to nonsyllabic CV transitions, specifying which consonant types are dispreferred as transition cues for vowels, but not necessarily specifying the syllable position of these segments (4-21). I do not specify the onset position for consonants because the relevant segment in Cibaeño Spanish is in the coda. Furthermore, although Spanish is often cited as a language with rampant resyllabification across word boundaries, recent work on phonetic lenition of Spanish consonants demonstrates a difference in the production of VC#V and V#CV (Hualde & Prieto 2012). Given the phonetic distinction between onset and coda consonants, particularly in a lenition context like coda gliding, postlexical resyllabification is not an accurate description for the language. Yet, Cibaeño still shows sensitivity to CV sequencing reminiscent of the sonority scale of ideal onsets. I propose that the same perceptual considerations of segment transitions in the acoustic signal that are applied to onset consonants also contribute to CV sequencing sensitivity across syllable boundaries. 91 91 Another possible question regarding this language family is whether these constraints could predict a language that inserts codas instead of onsets preceding vowel-initial words. In principle, any sequencing constraint like *VV or the CV and MINDISTVV families proposed here could interact with constraints like ALIGN(WD, L, %, L) and DEP-IO to produce a pattern of coda insertion, if ALIGN(WD,L) is ranked above ONSET (for example, *VV, ALIGN(WD,L) >> DEP-IO, ONSET, NO CODA). Although such a pattern is predicted by these constraints, the presence of ONSET and NO CODA in the universal grammar should restrict its occurrence. For example, I assume that coda “strengthening” only occurs in Cibaeño because the coda is already there and tolerated in the language. These issues are open for future research. 244 (4-21) *GLIDE-V >> *RHOTIC-V >> *LATERAL-V >> *NASAL-V >>*VOICEDOBST-V >> *VCLSOBST-V The universal ranking of these constraints will always prefer lateral-vowel sequences over glide-vowel sequences. 92 Per the Perceptual Enhancement Domain filter proposed in chapter 2, the *CV family of constraints may take the maximal Pwd domain because they maximize perception of segment transitions. The Pwd max domain restricts glide-blocking to the DP. Without the Pwd max domain specification, we would see glide blocking across all words in the language. With regards to driving coda gliding, a version of Kirchner’s (1998, 2004) LAZY constraint minimizes articulatory effort of the coda consonant production. Following Padgett (2008), I assume that the featural distinction between glides and laterals/rhotics is that glides are [-vocalic] and [-consonantal] while laterals and rhotics are [+consonantal]. Thus, coda gliding violates IDENT-IO[±consonantal] faithfulness. (4-22) LAZY(CODA) – Do not expend articulatory effort in the coda. IDENT-IO[±consonantal] – Let & in the input be in corresondence with ' in the output. Assign a violation for every & [±consonantal] that does not correspond to '[±consonantal]. (That is, do not change the [±consonantal] feature value from the input to the output.) 92 Like the MINDISTVV constraints, these could also be formulated in a stringent ranking, instead of universal ranking. Either is equal for the purposes of this analysis. 245 For the purposes of this analysis, LAZY(CODA) assigns a violation mark for the more effortful consonant production of [l] in comparison to the less effortful production of [j]. The interaction of the above markedness constraints produce glide-blocking in Cibaeño Spanish, presented in Table 46. Table 46. Cibaeño coda glide-blocking i. /el D aviso N / *GLIDE-V(PWD MAX ) LAZY(CODA) IDENT[±CONS] a. (e[j] aviso ! max ) *! #b. (e[l] aviso ! max ) * ii. él S avisa V #a. (e[j] ! max ) (avisa ! max ) * b. (e[l] ! max ) (avisa ! max ) *! iii. el D día N # a. (e[j] día ! max ) * b. (e[l] día ! max ) *! Candidate (ia) glides the coda consonant of the determiner to obey articulatory effort minimization, but it does so within the maximal Pwd, triggering a violation of *GLIDE-V(PWD MAX ). Candidate (ib), with production of the /l/, obeys the segment transition restriction. In (ii), the two words are prosodified into their own maximal Pwds, and therefore satisfy the positionally marked *GLIDE-V(PWD MAX ), since the glide is external to the Pwd max . Candidate (iia) wins with the LAZY production of the coda [j]. In 246 (iii) we see the same determiner-noun context in a single Pwd max , but the consonant- initial noun allows coda gliding because the /l/ does not transition to a vowel. One prediction of the Pwd max context here is that word-internal onset glide-vowel sequences will also violate the constraint, for example ha[j]a, an auxiliary verb in the subjunctive conjugation, fa[j]o ‘mistake, failure’, [j]ato ‘hiatus’, [w]ésped ‘orphan’. The fact that these words, and many others like them, are realized in Spanish reveals the derived environment effect of Cibaeno coda glide blocking. Non-application of allomorphy from the feminine el and conjunction patterns in Spanish were addressed as a distinction between function and lexical words (§2.3). But the auxiliary verb haya is a non-lexical word, and yet maintains its glide-V sequence. An appeal to lexical faithfulness does not succeed in saving this form, as shown in Table 47. 93 Table 47. Failure of Faith Lex haya Aux [subjunctive auxiliary verb] FAITH LEX *GLIDE-V(PWD MAX ) a. [a.ja] Pwdmax *! !b. [a.la] Pwdmax Candidate (a) has an undesired glide onset within the maximal Pwd and is ruled out by *GLIDE-V markedness. Lexical faithfulness does not rule out the repaired form in candidate (b), because it is not a lexical item. 93 I assume that due to the bisyllabic phonological form of the word, this function word forms a foot, and thus receives Pwd status, rather than floating in the PPh as a free clitic. 247 Unlike the feminine el and conjunction, which are blocked from spreading throughout the langauge by lexical faithfulness, coda gliding in Cibaeño demonstrates derived environment blocking (DEB). In order to account for the fact that a glide-V sequence is blocked from occurring from coda gliding contexts, but is allowed when it occurs as part of the underlying word form, another faithfulness constraint IDENT[-consonantal] is ranked above *GLIDE-V(PWD MAX ) in the DEB schema proposed in Hall (2006). (4-22) DEB schema (from Hall 2006) Faith A >> Markedness A >> Markedness B >> Faith B Table 48. DEB ranking of Cibaeño Spanish i. haya Aux [auxiliary verb] IDENT [-CONS] *GLIDE- V(PWD MAX ) LAZY(CODA) IDENT [±CONS] # a. (a.ja ! max ) * b. (a.la ! max ) *! * ii. él S avisa V ‘he warns’ a. (el ! max )(a.vi.sa ! max ) *! # b. (ej ! max )(a.vi.sa ! max ) * iii. el D aviso N ‘the warning’ # a. (el.a.vi.so ! max ) * b. (ej.a.vi.so ! max ) *! * 248 Candidate (ia) demonstrates maintenance of the underlying glide in the auxiliary verb haya, obeying IDENT[-CONS] by maintaining the [-consonantal] feature of the glide. The glide-strengthening candidate (ib) is ruled out by this faithfulness constraint. The candidate sets in (ii) and (iii) show coda gliding and coda glide blocking, respectively. Candidate (iia) maintains the lateral in coda position, violating the markedness constraint LAZY(CODA). Candidate (iib) glides the lateral to obey markedness. It does not violate the higher-ranked IDENT[-CONS] constraint because the underlying lateral is [+cons], not [-cons]. Candidate (iiia) also maintains the underlying lateral, and violates LAZY(CODA). However, candidate (iiib) with gliding violates the markedness constraint *GLIDE- V(PWD MAX ) because determiners and nouns are prosodified inside a maximal Pwd, and therefore coda gliding produces a glide-vowel sequence inside this domain. In summary, Cibaeño segment adjacency restrictions take the form of constraints on consonant-vowel transitions, preferring less sonorous /l/ to the more sonorous and vowel-like glide [j] as a transition to the following vowel. Like the feminine el, this phonological markedness restriction again occurs at the determiner-noun prosodic cliticization context, represented by the Pwd max domain. 4.4 Variations on the feminine el Thus far, the language patterns examined have shown similarity to the Spanish feminine el in their phonological segment adjacency effects at the prosodic clitic-host environment, supporting syntactic head dependent cliticization as well as the proposed 249 MINDISTVV and other segment transition constraints. This section provides further data for another aspect of the feminine el analysis: morpho-phonological correspondence. Section 2.4 reviewed the historical derivation of the feminine el, from its roots as a purely phonological phenomenon of syllable reduction (from bisyllabic ela to allomorphs el or la) to its reinterpretation as a morpho-phonological phenomenon utilizing the masculine determiner phonological form el with the feminine morphological gender. This repair, violating proposed MP Faithfulness, assumes no change at the morphological level (feminine words remain feminine) and no syntactic disagreement between the determiner and noun. However, these are two other options for avoiding the marked a-á sequence. Section 4.3.1 addresses morphological gender change in the history of individual lexical items in Spanish, and §4.3.2 presents syntactic disagreement in non- prescriptive variation of the feminine el. 4.4.1 Morphological gender change Morphological gender change is attested in the history of Spanish nouns associated with the feminine el. The words arte ‘art’ and azúcar ‘sugar’ illustrate the possibilities of morphological gender shift in the noun as well as morphological hypercorrection. Arte began in Latin as artem, and became the feminine noun arte in Old Spanish. It participated in the same determiner allomorphy demonstrated by other vowel- 250 initial feminine nouns, taking the phonological determiner form el but producing feminine agreement within the DP. 94 (4-23) /ela F arte F / # el F arte F ‘the art’/el F arte F magica F ‘the magic art’ (General estoria I, Alfonso X) In Modern Spanish, however, this noun is no longer feminine, but masculine 95 : (4-24) /el M arte M / # el M arte M ‘the art’/el M arte M mágico M ‘the magic art M ’ The loanword azúcar ‘sugar’ entered the Spanish vocabulary from Arabic sukkar or as-sukkar (with the Arabic article) and was originally grammatically masculine, yielding el açucar in Old Spanish. However, due to its /a/-initial phonological similarity to the feminine el words (remember that in Old Spanish the determiner switch was not conditioned by stress), the morphological gender of the noun was unclear. When the phonological vowel-initial trigger became more restricted for this set of words, requiring a stressed /a/ instead of unstressed, el azúcar was probably hyper-corrected in many cases to la azúcar. The conflicting gender interpretations for this noun gave rise to two 94 As with the feminine el data, I take feminine gender agreement within the DP to be a reliable indication of the noun’s gender. It is particularly relevant to the form arte, since it ends with -e, which is otherwise ambiguous for gender. 95 The Real Academia Española gives limited examples of singular arte in the feminine (arte decorativa, arte poética, arte metálica), which are reported by native speakers as either preferable in the masculine form or, in the case of arte poética, as a fixed phrase. The plural forms presented in RAE—and confirmed by speakers—are always feminine. 251 acceptable genders today in Modern Spanish: el azúcar and la azúcar. Since this item’s gender confusion was not conditioned in the same way as the feminine el patterning with agua and arte, it will not feature largely in this analysis. But, it is a good indicator of the variability in the grammar of Old Spanish, and the awareness that speakers had of the relationship between gender morphology and phonology. Both arte and azúcar demonstrate change in the noun’s gender, in response to determiner allomorphy. When the feminine el shifted from a purely phonological phenomenon to morpho-phonological in nature, faithfulness along the MP correspondence dimension was violated (see §2.3.3 and §2.4). Nominal gender change allows the MP relations among determiners to remain faithful to the grammar’s canonical forms, but violates the Morphological-Morphological (MM) dimension of correspondence. (4-25) IDENT-MM: If & is a morpheme in the input, and ' is its corresponding morpheme in the output, & and ' must have identical morphological features. (That is, don’t change the identity of a morpheme, e.g. the identity of a noun’s morphological gender.) (4-26) IDENT-MM violation /{la F , el M } arte F / # el M arte M 252 In the development of the word arte, the morpho-phonological relationships of la F and el M are maintained, and the masculine gender for the noun is chosen to match the canonical masculine gender of el. 96 The noun’s change of gender in (4-26) is differentiated from the determiner’s morpho-phonological switch in the current feminine el pattern, repeated in (4-27). (4-27) IDENT-MP violation /{la F , el M } agua F / # el F agua F I attribute this to a difference in the underlying relationships between a noun’s phonological form, its lexeme meaning ‘art’, and its grammatical gender, which are not so straightforward as the morpho-phonological relationships in function words, or adjective endings. The morpho-phonological associations of the determiners and adjectives are presented in (4-28), linking phonological forms directly with the gender morphemes. (4-28) a. la , el /{la F , el M }/ [feminine] [masculine] 96 At first blush, this suggested derivation would seem to directly contradict the principle of faithfulness to lexical items over functional words. However, the behavior of Spanish arte as well as the synchronic pattern of masculine el variation discussed in !4.4.2 is an indication of the grammar’s strong preference for avoiding MP violations, going so far as to bend the noun’s gender to avoid MP violation. 253 (4-28) b. alt {a , o} /alt{a F , o M }/ ‘tall’, ‘high’ 97 [feminine] [masculine] In contrast to function words, which for the most part do not vary phonologically, nouns may have vastly different phonological forms and yet still carry the feminine or masculine gender morpheme. I assume that relationship between morphological and phonological correspondences in the noun are therefore more indirect. The phonological form of the noun is associated with the lexeme meaning, and morphological gender is listed in the input but without a direct association to phonological form (4-29). (4-29) arte [‘art’] [feminine] While the schemas in (4-28) and (4-29) capture the difference in having a direct morpho-phonological association versus merely listing grammatical gender in the input, the treatment of noun gender morphology is a complex issue and deserves further examination. Spanish noun endings -a,-o and -e may have canonical morphological exponences in much the same way la and el do, but there are many more exceptions in each category, intersecting with considerations of word class and biological gender (Harris 1991). I leave the possibility of analyzing nominal endings with IDENT-MM and 97 The different glosses correspond to different syntactic positions. In postnominal position, alto is best translated as ‘tall’, while in prenominal position it is best translated as ‘high’. 254 IDENT-MP constraints for the future, focusing here on the determiners and overall patterns shown in the development of Spanish. Because there is no morpho-phonological correspondence between morphological gender and a specific phonological form, arte’s morphological change from feminine to masculine does not violate IDENT-MP, but it does violate IDENT-MM in changing the gender from input to output. Table 49. Morphological gender change in arte /{el M , la F } arte F / MINDISTV'V(PWD MAX )=F1/F2:1 IDENT-MP IDENT-MM a. la F arte F *! b. el F arte F *! # c. el M arte M * Candidate (a) uses the canonical feminine la F , but violates Minimum Distance and is immediately ruled out. Candidate (b) follows the regular feminine el pattern, with use of the exceptional feminine el F , which violates IDENT-MP. Candidate (c) maintains the canonical masculine el M to obey IDENT-MP, but changes the morphological gender of the noun itself, violating IDENT-MM. Candidate (b) and (c) show the necessity of ranking of IDENT-MP over IDENT-MM in order to produce the morphological change that occurred for arte. However, IDENT-MM ranked over IDENT-MP is necessary to produce the feminine el pattern in all the other words—like agua, ave, hambre and haz —that did not change gender, an issue that I turn to next. 255 In comparing the development of arte with the development of the rest of the feminine el words, two qualities differentiate it from the other lexical items: form and frequency. Since –e is gender ambiguous, while –a is usually associated with the feminine (among adjectives, functional words, and certain nouns), the gender-ambiguous –e of arte may have allowed it to more easily switch genders in comparison to the mostly –a final nouns in the rest of the word class. Frequency surfaces when comparing arte to the other –e and consonant-final nouns of the current feminine el set: ave ‘avian’, hambre ‘hunger’ and haz ‘face’. Of these, ave and haz are low-frequency, and hambre is reported to occur with postnominal masculine adjectives, unlike its –a final colleagues (Eddington & Hualde 2008). Arte and hambre are higher in frequency, and they are the ones that demonstrate full masculine identity, either in diachrony (arte) or in synchronic variation (hambre). Frequency is known to affect the diffusion of linguistic change, such that higher frequency words are more likely to participate in phonetic sound change (Hooper 1976, Phillips 1984, Bybee 2002). I hypothesize that in the development of the feminine el, the higher frequency words first took the phonetic change of syllable reduction from bisyllabic ela to la and el, while lower frequency words would still occur with bisyllabic ela. This means that low frequency words would maintain feminine gender identity more strongly, because they would still demonstrate an a-final determiner. To account for the two differences in the development of lexical items—the morphological distinction between –e and –a final words, and the frequency distinction from low-frequency and high frequency words—I use lexically indexed constraints 256 following Pater (2000, 2006, 2008, 2009). Lexically indexed MM constraints differentiated by morphological ending will encode the preference for maintaining gender in an –a ending, which typically co-occurs with the [feminine] gender, unlike –e which occurs equally with both [feminine] and [masculine] genders. MM constraints indexed for word frequency encode the idea that low frequency words are more likely to maintain their gender and less likely to change. The constraints are outlined in (4-30), followed by an illustrating tableau in Table 50. (4-30) IDENT-MM -A – Do not switch morphological exponences for lexical items ending in –a. IDENT-MM LF – Do not switch morphological exponences for low frequency items. 257 Table 50. el arte and el ave: lexical-indexation and frequency-indexation for MM 98 i. /{la F , el M } arte F-HF / 99 IDENT-MM -A IDENT-MM- LF IDENT-MP IDENT-MM a. el F árte F (mágica F ) *! # b. el M árte M (mágico M ) * ii. /{la F ,el M } ave F-LF / # a. el F áve F (blanca F ) * b. el M áve M (blanco M ) *! * iii. /{la F ,el M } agua F-HF / # a. el F água F (fría F ) * b. el M água M (frío M ) *! * Candidate (ia) is the prescriptive feminine el pattern, with a violation of IDENT- MP. Candidate (ib) re-interprets the noun as masculine, allowing for the use of masculine el M and obeying morpho-phonological faithfulness. It violates general MM faithfulness, but as a high frequency word, it does not violate the frequency-indexed IDENT-MM- LF for low frequency items. In (ii), low-frequency ave is examined under the same constraint ranking. Here, masculine interpretation of the noun in candidate (iib) fails on high- ranking frequency-indexed MM faithfulness, enabling the feminine el pattern to emerge. 98 The ranking in (4-42) is also consistent with low frequency a-final words of this class—for example alga ‘seaweed’ or arca ‘ark, chest’—maintaining feminine gender instead of undergoing gender shift. 99 While I do not suggest that frequency is encoded as part of the lexical item, it is notated here in the same way that morphological class lexical indexation is notated, for ease of reference. 258 In (iii), the stronger MM faith for a-final agua comes into play, forbidding masculine reinterpretation of the noun in (iiib). In sum, the history of the feminine el items in Spanish demonstrates more than one pattern of repair for Minimum Distance and prosodic clitic markedness. The canonical masculine el M asserts itself in morphological reinterpretation of the noun, accounted for formally by ranking MP and MM faithfulness. 4.4.2 Synchronic variation: masculine el The previous section illustrated morphological gender change in the noun as an alternate repair to the feminine el. Another possible repair to the a-á prosodic clitic sequence that avoids use of an MP violation is to allow syntactic disagreement. The following data on synchronic variation of the feminine el show precisely that behavior. According to Janda and Varela-García (1991) and Eddington and Hualde (2008), many of the feminine el words demonstrate prenominal variation in true morphological gender, rather than merely alternating the phonological form of the definite article. Other determiners, including quantifiers and demonstratives, often take the non-prescriptive masculine form with these words, despite restriction of the Standard Spanish pattern to the article (data from Eddington & Hualde 2008). (4-31) a. este M agua F , *esta F agua ‘this water’ b. ese M hacha F , *esa F hacha ‘that axe’ c. mucho M hambre F ,*mucha F hambre ‘much hunger’ 259 d. todo M el M área F ,*toda F el área ‘all the area’ The items in (4-31) show masculine morphology in the preceding demonstratives, quantifiers and determiners. In particular (4-31d), which includes the determiner el as well as the masculine todo, provides strong evidence for a fully masculine el M instead of prescriptive el F . Were feminine gender in effect here, we would expect prescriptive toda F el F área instead of the masculine agreement shown in the quantifier todo. Preceding adjectives, too, may have a fully masculine identity. (4-32) Variant Prescriptive a. buen M alma buena F alma ‘good soul’ b. el M cristalino M agua la F cristalina F agua ‘the crystalline water’ c. el M buen M hada madrina F la F buena F hada madrina F ‘the good fairy godmother’ d. 100 el M abundante agua fría F la F abundante agua fría F ‘the abundant cold water’ Postnominal adjectives like those in (4-32c-d) still retain the feminine gender, indicating that the noun itself is yet feminine. The result is a mixed gender DP, with a 100 While the prenominal adjective abundante ends in what I assume to be a gender neutral affix -e, it is appropriate for use with either the masculine or feminine genders, and allows gender from the noun to project beyond to the determiner. Thus, the expected form for a feminine noun is la abundante N, while the expected form for a masculine noun is el abundante N. 260 masculine gender prenominally through the determiner and prenominal adjectives, but feminine agreement through the noun and postnominal adjectives. Like the pattern of morphological gender change in the noun seen previously, this synchronic variation of the feminine el shows full masculine gender in the determiner, producing a violation of syntactic gender agreement within the DP. IDENT-MP is obeyed, since there is no shift in the relationship between phonological and morphological material (no feminine el F ) and IDENT-MM is obeyed, since there is no morphological shift (the feminine noun does not become masculine). Syntactic agreement is instead the contradicting force to this pattern. A constraint on agreement is utilized here to represent the imperative for same gender exponence between the D, A and N. See Bouma (2003) for a similar syntactic AGREE constraint, and Samek-Lodovici (1996) for constraints particular to syntactic specifier-head agreement. Here, I define syntactic agreement for the constraint in terms borrowed from the Minimalist program (Chomsky 2000, 2001, 2004, 2008, Bhatt 2005). Syntactically, the Minimalist function Agree takes a Probe with unvalued feature slots—a determiner or adjective with unspecified gender, for example—and looks for a Goal with matching valued features—a noun with lexically specified grammatical gender—down the c-command domain of the Probe. An example of the DP structure before syntactic Agree is shown in (4-33). 101 101 For clarification of the c-command relationship, I use a more articulated syntactic DP structure than what is presented in §2.2. The structure presented here with intervening FP projections between D and N is still consistent with the lexical head-complement analysis proposed for the syntax-phonology interface. 261 (4-33) DP D FP [gen] AP FP alt- ‘high’ [gen] NP N águila ‘eagle’ [FEM] The D Probe will find the matching feature set [gender] in both the prenominal A and the N within its c-command domain. The A also has an unspecified [gender] feature, while N has the specified, or valued, feature [FEMININE]. Following Frampton & Gutmann (2000, 2006) and Demonte & Pérez-Jiménez (to appear), the valued [FEMININE] feature is shared among the D, A and N after the syntactic operation Agree has matched the feature sets. When the structure moves to the phonological sphere, the appropriate feminine morpho-phonological forms (la F for the D, and –a F for the A) are chosen. 262 (4-34) DP D FP [FEM] AP FP alt- ‘high’ [FEM] NP N sacerdotisa ‘priestess’ [FEM] The agreement constraint in the phonological sphere is not the exact same as the set matching syntactic Agree, since it is concerned solely with feature values. When masculine el M is used with a feminine noun like agua F , it is the feature values [MASCULINE] and [FEMININE] that do not match. The unspecified feature [gender] is still matched from the syntax. This analysis thus splits syntactic Agree into two functions, one matching the feature, and the other matching the values. This is reminiscent of Chomsky’s (2000) definition of Minimalist Agree, where Agree is a feature checking operation (matching features) subject to a separate Match condition (matching feature values). I propose that the feature-matching Agree occurs in the S-structure branch of the grammar, while value-matching Agree occurs in the PF branch of the grammar. The notion of certain syntactic operations occurring in the PF branch of grammar is well- established in previous literature (Chomsky 1995, 2001; Embick & Noyer 2001, 2005; and many others). In work on Distributed Morphology, agreement features are specifically argued to be inserted post-syntax as a morphological PF operation (Embick 263 & Noyer 2005, also citing Iatridou 1990, Marantz 1992, Chomsky 1995). I do not necessarily argue that this means that unvalued feature-checking Agree is also in PF, but the data strongly points to value-insertion and value-matching as pertaining to PF and available to constraint interaction with considerations of phonological well-formedness. This split predicts the contradicting “agreement” between the syntactic and phonological spheres seen in the Spanish synchronic variant of the feminine el, where syntactically the determiner, noun and adjective are linked by Agreement, but in the phonological sphere the values of these items do not agree. 102 Since information is assumed to travel unidirectionally from syntax to phonology (i.e., there is no circular feedback from phonology to syntax), the feature value disagreement resulting from the interaction of phonological constraints does not affect the input S-structure or LF interpretation. Grammatical gender values—unlike biological gender—do not affect the meaning of the phrase, and are not expected to be active in the LF branch of the grammar. (For more discussion on the syntactic distinction between grammatical vs. biological gender, see Atkinson 2012). I define the syntax-phonology interface constraint AGREE(VALUE) as shown in (4-35). (4-35) AGREE (VALUE): every Probe must match feature values with its Goal. 102 The PF-syntactic operation approach taken here and in Distributed Morphology has also been used to account for mismatches between syntax and phonology in the linearization and feature specification of clitics and other function words (Halle & Marantz 1993; Harley & Noyer 1999, Embick & Noyer 2001). 264 This constraint assigns a violation for every Probe (D or A, for example) that does not match the feature values of its Goal (N). Table 51 demonstrates how this constraint interacts with the markedness and faithfulness constraints seen thus far to produce the pattern of synchronic gender disagreement instead of the prescriptive feminine el. Table 51. Synchronic variation and AGREE(VALUE) /{la F ,el M } agua F / MINDISTV'V(PWD MAX ) =F1/F2:1 IDENT-MP IDENT-MM AGREE (VALUE) a. la F agua F *! b. el F agua F *! c. el M agua M *! # d. el M agua F * The D in candidate in (a) agrees syntactically with the noun, but fails on Minimum Distance in producing the marked a-á sequence. Candidate (b) represents the prescriptive pattern, using the feminine el. In the variant grammar, however, this fails on IDENT-MP, which is ranked on par with Minimum Distance. Candidate (c) uses a fully masculine el and avoids the forbidden vowel sequence, but changes the morphological gender of the noun, incurring a violation of IDENT-MM. Candidate (d) obeys markedness with use of the phonological form el, MP faith by maintaining masculine gender with the determiner, and MM faith by maintaining feminine gender with the noun. Instead, the determiner and noun do not match feature values, incurring a violation of AGREE (VALUE). 265 When prenominal adjectives enter the equation, producing the mixed gender variant requires additional analysis. With a prenominal adjective, the Pwd max condition of the Minimum Distance constraint is nullified, and there is no markedness reason to use any determiner other than the standard la F . Hence, in the prescriptive pattern, feminine el F is used directly preceding a noun, but la F is used preceding an adjective. In the mixed gender variant, however, el M is used throughout. One way to achieve this is through a Paradigm Uniformity constraint, following Steriade (1999). Instead of a noun like agua taking one determiner without an adjective and another determiner with an adjective, Paradigm Uniformity (PU) would enforce the use of the same determiner throughout the DP “paradigm”. Such a constraint is presented below, outlining the correspondence relation between the simple determiner-noun structure and the more complex determiner- adjective-noun one. (4-36) PARADIGM UNIFORMITY (D): let D be a determiner with its complement N, a noun. If a realization of D occurs with N with no additional structure, then for every correspondent D( occurring with additional structure, D and D( are the same. This constraint assigns a violation mark every time a different determiner is used for a specific noun when it has a prenominal adjective (or, for that matter, a postnominal adjective). That is, each noun should use only one determiner form, and it should be the same as the form used with the simple determiner-noun structure alone. This constraint 266 favors the mixed gender variant of the Spanish feminine el pattern over one that switches the determiner, as illustrated in Table 52. Table 52. Paradigm Uniformity and prenominal adjectives /{la F ,el M } agua F / /{la F , el M } cristalin{a F ,o M } agua F fri{a F ,o M }/ PU (D) AGREE (VALUE) # a. el M agua F el M cristalino M agua F fría F * ** b. el M agua F 103 la F cristalina F agua F fría F *! * In Table 52 above, the candidate set in (a) maintains the masculine determiner form in the adjectival form to satisfy PU(D), but incurs three violations of AGREE, since the determiner and noun do not agree in gender for the simple DN form, and both the determiner and the adjective do not match the noun Goal for the DAN(A) form. The candidate set in (b) only uses the masculine determiner in the simple form as required by Minimum Distance, but uses the standard feminine one when an adjective is added to the structure. While it incurs fewer violations of AGREE (VALUE), the paradigm is non- uniform in the use of two different determiners with the same noun, and receives a fatal violation of the Paradigm Uniformity constraint. 103 This pattern is hypothetically different from the feminine el pattern described as el F agua F /la F cristalina F agua F , but would be impossible to differentiate from it, since every time an adjective intervenes, both patterns use la F . Either of these patterns could in fact be used by prescriptive speakers. 267 In a brief summary of the constraints utilized so far in the mixed gender variant, we see two levels of ranking, illustrated in (4-36). Minimum Distance and stress LC markedness, MP Correspondence and PARADIGM UNIFORMITY are all equally ranked above AGREE (VALUE), producing a pattern that avoids the forbidden vowel sequence without using a non-standard determiner form or different determiners preceding nouns and adjectives. (4-36) MINDISTV'V(PWD MAX )=F1/F2:1, IDENT-MP, PU(D) >> AGREE (VALUE) 4.4.2.1 More on AGREE(VALUE) One outstanding problem in mixed gender variant pattern is why el M cristalino M agua F fria F is attested but el M cristalina F agua F fria F is not. Both violate agreement of the D, as the determiner and noun do not match in either form. But, since agreement is analyzed as extending from the noun, the latter form with a feminine prenominal adjective does better than the attested variation with a masculine prenominal adjective. To address this, I introduce another kind of AGREE, enforcing agreement within a domain, from the perspective of the head of the phrase. " (4-37) AGREE(XP HEAD ): for every Agreement item within an XP, its feature values must match those of the XP head. 268 This constraint assigns a violation for every item that does not match feature values with the head of the XP within which it receives Agreement. On one level, this encodes the idea that feature agreement should be consistent throughout an XP, and that the head, the flagship of the XP, carries said consistent agreement. 104 A prenominal adjective may not perform syntactic Agree within the AP, since there is no Goal with valued features inside the AP, but it does so in the DP. A postnominal adjective Agrees within the NP. Thus, for the prenominal adjective, the DP is the XP domain that the constraint governs, while for a postnominal adjective, the NP is the XP domain that the constraint governs. This explains why prenominal adjectives agree with the D of the DP, while postnominal adjectives agree with the noun. Ranking AGREE (XP HEAD ) above AGREE (VALUE) yields the appropriate synchronic variation pattern. Table 53. AGREE(XP HEAD ) and AGREE (VALUE) /{la F , el M } cristalin{a F ,o M } agua F fri{a F ,o M }/ AGREE(XP HEAD ) AGREE (VALUE) # a. el M cristalino M agua F fría F * ** b. el M cristalina F agua F fría F **! * Candidate (a) violates AGREE(VALUE) twice: once for the disagreement of the D probe with the N goal, and once for disagreement of the prenominal A probe with the N goal. It 104 In a normal syntactic-phonological derivation where syntactic Agree and AGREE(VALUE) are obeyed, this constraint will never be violated. 269 violates AGREE (XP HEAD ) once: the noun does not agree with the D, which is the head of the larger XP in which the noun participates in agreement. The postnominal A does not violate this constraint, because it participates in agreement lower down in the NP, and does exhibit agreement with the N. Candidate (b) violates AGREE(VALUE) only once, for the disagreement of the D with the N. It violates AGREE (XP HEAD ) twice: the noun does not agree with the D, and the prenominal adjective does not agree with the D. In sum, the mixed gender variation of the feminine el demonstrates syntactic disagreement repair for Minimum Distance markedness, in contrast to an MP violation (the prescriptive feminine el) or MM violation (morphological change of the noun, as for arte). Constraints requiring feature agreement between each probe and goal as well as within the XP produce the effect of splitting gender in the DP, such that prenominal adjectives match the masculine gender of the determiner, but postnominal adjectives continue to reflect feminine gender of the noun. The full constraint ranking for the mixed gender pattern is shown in Table 54. 270 Table 54. MINDISTV'V(PWD MAX )=F1/F2:1, IDENT-MP, IDENT-MM, PU(D) >> AGREE(XP HEAD ) >> AGREE (VALUE) i. /{la F ,el M } agua F-HF / MINDISTV'V (PWD MAX )=F1/F 2:1 IDENT -MP IDENT -MM PU(D) AGREE (XP HEAD ) AGREE (VAL) a. la F agua F (fría F ) *! b. el F água F (fría F ) *! c. el M água M (frío M ) *! " d. el M agua F (fría F ) * * ii. /{la F , el M } cristalin{a F ,o M } agua F / # a. el M cristalino M agua F * ** b. el M cristalina F agua F **! * c. la F cristalina F agua F *! d. el M cristalino M agua M *! e. el F cristalina F agua F *! 4.5 Summary of Chapter Four Every aspect of the feminine el analysis provided in Chapter 2 is also represented in another language pattern shown in this chapter. Prosodic sensitivity to the syntactic head-complement relation in building cliticization structure is illustrated in Russian prepositional complement allomorphy. Minimum Distance constraints on vowel hiatus, specified to the Pwd max clitic context, appear in Catalan. The Cibaeño dialect shows 271 additional segment adjacency effects at the Pwd max clitic context, in the form of constraints on consonant-vowel transitions. Delving further into the history of Spanish and non-prescriptive variations of the feminine el pattern reinforces the language’s avoidance of Minimum Distance Pwd max markedness and provides support for MP Faithfulness, by maintaining the canonical masculine el M instead allowing MM Faith and syntactic agreement violations. While each interface phenomenon examined here appears disparate, they are unified in two main areas: phonological segment adjacency and prosodic cliticization. On the phonological side, each pattern illustrates an effect of adjacency markeness, be it Minimum Distance on adjacent vowels or restrictions on CV sequences. On the structural side, each pattern shows the phonological markedness sensitivity in prosodic cliticization, with the prosodic clitic-host structure in the maximal Pwd. In addition, the main case study of the Spanish feminine el together with the behavior of other function words such as conjunctions, prepositions and complementizers, provide evidence for the co- occurrence of affixal and free prosodic clitic structures occurring in the same language. In Chapter 5, we return to the issue of the theoretical interface between syntax and phonology to examine alternate proposals in accounting for the data presented here and in Chapter 2. 272 Chapter Five: Alternative approaches The previous chapters have argued for specific syntactic-phonological interactions and perceptual contrast in vowel sequencing, producing the feminine el and other phonological patterns involving prosodic clitics, including a relation-based mapping from the syntax to prosodic affixal and free clitics. This chapter examines some alternatives for the precise prosodic structures proposed, the lexical head-complement approach to the syntax-phonology mapping, and an alternative to the MINDISTVV approach to hiatus. Much of the previous literature on the mapping from syntax to prosodic structure has been concerned with the nature of syntactic information available to prosodic structure building. Different mapping procedures have been proposed, from relation- based mapping referencing the recursive and non-recursive sides of X-bar structure (Nespor & Vogel 1986), edge-based mapping referencing only XP boundaries Chen (1985), Selkirk (1986, 1995, 2004), Selkirk & Tateishi (1988, 1991), Selkirk & Shen (1990), Truckenbrodt (1995, 1999), to Match Theory referencing whole syntactic constituents (Selkirk 2009, 2011; Elfner 2011). In Chapter 2, a head-complement augmentation to Match Theory phrasing and cliticization was examined with respect to the Spanish feminine el data, appropriately generating different prosodic clitic structures with the lexical head-complement relationship present in a PARSE mapping constraint for function words, in addition to the structural mapping constraints of previous researchers. In this chapter, several main alternatives are considered to the PARSE FNC -INTO- ! HCOMP approach to the prosodic clitics and syntax-prosody mapping. One is the clitic group category used in Nespor & Vogel (1986) and Hayes (1989), examined in §5.1. As 273 shown, this category fails to account for the prosodic difference in DNA and DAN word orders of the Spanish feminine el. In §5.2, another alternative prosodic structure is explored, with recursive PPhs, rather than clitics adjoining to the Pwd and PPh. As we will see, this mapping alternative is not appropriate for the phonetic cues of PPhs in Spanish, nor is it capable of producing a prosodic difference between the two word orders. Sections 5.3 and 5.4 discuss alternatives to the theoretical mapping between syntax and phonology, rather than different individual prosodic structures. The first of these mapping alternatives, presented in §5.3, utilizes frequency as the primary mapping factor, instead of prosodic MATCH and PARSE constraints. This approach eliminates syntactic structural or relational differences as the driving force in prosodic phrasing in favor of syntactic frequency and considerations of language processing. It makes strong predictions about crosslinguistic prosodic phrasing that may or may not be borne out in future research. The fourth alternative explored in §5.4 is that of a direct interface between syntax and phonology, bypassing prosody entirely, and relying only syntactic relations. This approach, while extreme, could arguably account for the data in question and may circumvent issues of grammaticalization of the feminine el prosody. This approach raises questions about the experimental differences found in Spanish DP sequences that suggest different prosodic structures. In addition, without a mediating prosodic structure between syntax and phonology, it also predicts unattested patterns of interaction elsewhere in language. 274 The last alternative examined here relates to the MINDISTVV constraint family, which approach vowel hiatus restrictions from a perceptual perspective. Vowel hiatus restriction could also be approached from an articulatory perspective, based on the difficulty in producing different vowel elements adjacent to each other. As shown in §5.5, this approach predicts precisely the opposite pattern as that evinced by the Spanish feminine el. 5.1 Clitic group One alternative to the recursive Pwd structure sensitive to syntactic heads is a separate prosodic constituent specific to clitics and their hosts, suggested by Nespor and Vogel (1986) and Hayes (1989) as the “clitic group”. (Reproduced from Hayes 1989: 208, where “C” refers to Clitic group.) (5-1) Clitic Group Formation a. Every content word (lexical category) belongs to a separate Clitic Group. b. Definition: the HOST of the Clitic Group is the content word it contains. c. X and Y share a category membership in C if C dominates both X and Y. d. Rule: clitic words are incorporated left or right into an adjacent Clitic Group. The group selected is the one in which the clitic shares more category memberships with the host. 275 By these definitions, the Spanish D would simply be included in the clitic group belonging to the following word in both DNA and DAN cases. In the first step, both A and N form their own clitic groups, since they are lexical words (5-2a). The clitic determiner D shares category membership with both N and A because the DP dominates all three items. Thus, D is incorporated to adjacent clitic group, regardless of whether it is an A or an N (5-2b). (2) a. D ( C A) ( C N) D ( C N) ( C A) b. ( C D A) ( C N) ( C D N) ( C A) This clitic group formation fails to differentiate between the DP sequences in question, and does not reflect the experimental evidence of a greater prosodic boundary between D and A than between D and N. In addition, the clitic group has been specifically argued against in previous literature as an unnecessary addition to the prosodic hierarchy. Instead, recursive Pwd adjunction, as discussed in Chapter 2, accounts for language patterns previously analyzed with the clitic group and maintains a smaller set of universal prosodic constituents (Peperkamp 1997, Booij 1996, Itô & Mester 1992, 2007; Selkirk 1996, 2004). 5.2 Recursive PPhs The prosodic structure proposed in Chapter 2 for Spanish determiners is that of clitic adjunction at the Pwd and PPh levels, contrasting recursive and non-recursive Pwd 276 structure in DNA and DAN sequences. An alternative approach to the data would be a recursive PPh structure, assuming a strict structural mapping with no effect of internal syntactic relations. As we will see, however, the PPh structures make the wrong predictions for independent phonetic and phonological cues at the phrasal level. Following Prieto (2006) and Rao (2010), the phonetic cues associated with the end of Spanish PPh phrases are prominent stress, (optional) rising tone, and (optional) short pause. Associated with the beginning of a new phrase is pitch reset. For example, (5-3) presents a sentence for speaker 1 of the phonetic study outlined in Chapter 3. We see a prominent pitch accent in the word hotelero, and there is a dramatic drop in F0 from the peak of the L*+H pitch accent for hotelero to re-set the pitch for the beginning of le to approximately the same level as at the beginning of the sentence. This is a good indication of a PPh phrasal break after hotelero and before le. The same pattern is evident in sentence (5-4) by the same speaker, with a DNA phrase el hotelero orondo instead of DAN el orondo hotelero. While the pitch tracks in (5-3) and (5-4) are shown for speaker 1, none of the nine speakers recorded displayed evidence of a distinct PPh break inside the DNA or DAN sequence. 277 (5-3) (" El orondo hotelero) (" le dió la bienvenida). The fat hotelier to him/her gave the welcome ‘The fat hotelier welcomed him/her.’ el orondo hotelero le dió la bienvenida L*+H L*+H PPh L+H* L* L*+H L% 1 - el orondo hotelero 75 125 100 Time (s) 0 2.214 (5-4) (" El hotelero orondo) (" comía con gusto). The hotelier fat ate with gusto ‘The fat hotelier ate with gusto.’ el hotelero orondo comió con gusto L*+H L*+H PPh L* L* L% 1 - hotelero orondo 75 125 100 Time (s) 0 1.868 278 In the Match Theory approach of the syntax-prosody mapping, if the MATCH- PHRASE constraint were top-ranked, it would predict multiple layers of PPhs within the DAN and DNA sequences of sentences (5-3) and (5-4), not only the one we see evidence for. The definition of MATCH-PHRASE is reviewed in (5-5), and its predicted isomorphic prosodic structures are shown in (5-6). 105 (5-5) MATCH-PHRASE– Suppose there is a syntactic phrase (XP) in the syntactic representation that exhaustively dominates a set of one or more terminal nodes &. Assign one violation mark if there is no phonological phrase (") in the phonological representation that exhaustively dominates the phonological exponents of the terminal nodes &. Exhaustive domination – A syntactic node & exhaustively dominates a set of terminal nodes ' iff & dominates all and only the terminal nodes in '. (5-6) a. [ DP D [ FP N i [ NP [ AP A] i]]] (" D (" N (" A))) b. [ DP D [ FP [ AP A] N i [ NP i]]] (" D (" (" A) N)) 105 I do not assume that functional projections are invisible to the syntax-prosody mapping, and therefore MATCH-PHRASE would predict PPhs for any functional XP that has a terminal node word in it. Therefore, if a head-raising account of the DP is assumed, as presented in (5-6), the FP containing head-raised N would still match to a PPh. The same PPh matching applies to an NP- raising approach. 279 As mentioned above, Spanish has effects of final lengthening, prominent pitch and optional raising and pauses at the end of PPhs. In the data collected, these effects do not occur inside the DP sequences, but only at the end of the second lexical word. If a fully recursive structure is assumed as in (5-6), then Spanish PPh effects must be assigned to maximal PPh structure. In this case, the outside PPh of the DP should correspond to a maximal PPh in Spanish. Elfner (2011) suggests a Function Word Adjunction Principle, which states that function words are visible to the syntax-prosody mapping, but that they do not contribute to the minimal/maximal distinction, instead taking their subcategory level from the lexical phrase to which they attach. 106 This principle would yield the following structure in (5-7). (5-7) (" maximal D (" maximal N (" min A))) (" maximal D (" maximal (" min A) N)) While this phrasing might account for (maximal) phrase-final effects only occurring at the end of the second lexical word, it does not account for phrase-initial pitch 106 The Function Word Adjunction Principle is examined here in relation to the PPh level of phrasing, but if as Elfner (2011) proposes it is a universal restriction on GEN, it also has implications for the possible clitic structures available to languages. Instead of the DN affixal clitic adjoining at Pwd min and forming a Pwd max , it would be (! min D(! min N)). Instead of the DA free clitic adjoining at Pwd max and forming a PPh, it would be (! max D (! max A)). With these new prosodifications, the current analysis would not be substantially changed, as there would still be a difference between DA and DN sequences (though, it would require the domain Pwd min for MINDISTVV, not Pwd max , to apply to DN and not DA). However, it does make undesirable predictions regarding other prosodic phenomena that would usually be assigned to the Pwd min level, such as stress. If stress were assigned within the Pwd min domain, the prosodification (! min D(! min N)) would predict stress shift with nouns preceded by determiners. Both the “affixal” clitic and internal clitic of this new prosodification pattern are within a Pwd min . 280 reset. The left PPh boundary following the D in both word orders predicts PPh-initial effects at both the D and the following N or A. We do not have pitch reset between the D and the following word; therefore there cannot be a maximal PPh boundary between them. The hypothetical "-phrasing above is completely isomorphic with the syntactic structure input and differentiates between DNA and DAN sequences, but does not account for the phonetic phrasal cues. The next avenue for possible "-phrasing involves an incomplete match between the syntactic and prosodic structures and accounts for phonetic cues, but does not produce a prosodic difference between DNA and DAN word orders. Non-isomorphic prosodic phrasing inside the DP is not too surprising, since phrasing around individual words is unattested elsewhere in Spanish (Prieto 2006), as well as for DPs in Italian (Dehé & Samek-Lodovici 2009) and Connemara Irish (Elfner 2011). Indeed, Elfner (2011) proposes that a minimal binarity constraint on the formation of prosodic structure is responsible for non-isomorphy in the Connemara Irish DP, which prevents the A from appearing with its own phrase. Applied to the Spanish data, the prosodic structure changes from one with three levels as in (5-7) to only two, as in (5-8) below. This also changes the subcategory level of the PPh that contains the N and A. (5-8) a. (" ? D (" min N A)) b. (" ? D (" min A N)) 281 The subcategory level of the outer PPh is debatable. Under Elfner’s (2011) Function Word Adjunction Principle it would be another minimal PPh, since the function word PPh would take its subcategory level from the lexical PPh to which it adjoins. But, this raises questions again about the minimal or maximal domain of phonetic PPh effects in Spanish. If it is another minimal phrase (5-9a), then PPh-initial effects must be assumed at the PPh min level, predicting pitch reset between the D and following word. If Spanish function words do not follow the Function Word Adjunction Principle, as argued in §2.2, 107 and participate fully in the subcategorial distinctions of the syntax-prosody mapping, the outer PPh might be maximal (5-9b), or at least non-minimal. With this structure, we might posit initial effects at the maximal PPh, but final effects at the minimal PPh. 108 (5-9) a. (" min D (" min N A)) (" min D (" min A N)) b. (" max D (" min N A)) (" max D (" min A N)) The phrasing in (5-9b) may account for the phonetic cues of Spanish PPhs, but requires treating Spanish function words exactly the same as lexical items, contra Elfner’s (2011) proposal as well as Selkirk (1996, 2004), who argues that function words are invisible to 107 The behavior of Spanish function words in comparison to the Connemara Irish data presented in Elfner (2011) could be accounted for by formulating the Function Word Adjunction Principle as a violable constraint instead of a restriction on GEN. 108 Since recursive PPh phrasing is assumed throughout the sentence, the “maximal” PPhs shown in these examples might not always be truly maximal, because they could be subsumed by a larger phrase, for example in a fairly long sentence spoken rapidly. 282 the syntax-prosody mapping and instead take their (often non-isomorphic) prosodic structure due to interaction of constraints on the mapping of lexical words. The treatment of function words as invisible to the mapping does not predict an outer " phrase wrapping the entire DP sequence, but rather prosodifies only the N and A (5-10), again assuming binarity effects. (5-10) D (" N A) D (" A N) Another mechanism—such as the proposed clitic adjunction in the previous section— would be required to either include the D in the single ", yielding the single phrase analysis of (5-11), or build an additional outer recursive ", yielding structures like those in (5-9b). Like the recursive " structures in (5-9b), the single " structure is consistent with the phonetic effects of Spanish PPhs. (5-11) (" D N A) (" D A N) Both phrasal analyses in (5-9b) and (5-11) face a problem in regard to the Spanish data. Neither one of them predicts a difference in the prosodification of DN and DA sequences. Either there is a single PPh boundary between the function word D and the following lexical word N or A (5-9b), or there is a single PPh wrapping the entire 283 sequence (5-11). If we assume an indirect reference between syntactic structure and phonological constraints, such that the phonology may reference prosodic structure and not the syntax directly, then the prosodic structure must differentiate between the DNA and DAN sequences for the feminine el. In light of these problems with an entirely PPh phrasal analysis of prosodic distinction, I return to the prosodic adjunction of clitics at the Pwd and phrasal levels to differentiate between DNA and DAN word orders in the feminine el. I argue it is not different phrasal adjunction, but a contrast between word adjunction and phrasal adjunction that distinguishes these sequences in the phonology. Determiner allomorphy should be considered a phonological reflection of clitic adjunction depending on the presence or absence of ! max boundaries, not " boundaries. As regards the parsing of PPhs, either phrasing (5-9b) or (5-11) is compatible with the phrasal effects noted in Spanish, and any distinction between the two is outside of the scope of this work. 5.3 Frequency as a mapping factor Returning to the affixal and free clitic structures proposed for the feminine el, another alternative is available regarding the formation of these structures. Frequency has been discussed in §2.4, §3.5 and §4.3 as having an effect on the historical development of the feminine el and possibly influencing the production of DN and DA sequences. Here, it is evaluated as an alternative factor for mapping prosodic structures. This approach maintains the affixal and free prosodic structures presented in this dissertation, but instead of relying on differences in the syntactic relations of input DA and DN sequences, 284 the frequency of such structures directly affects language processing and production, which is in turn interpreted as differences in prosodic organization. As mentioned in §3.5, the prenominal adjective structure in Spanish is less frequent than a postnominal adjective structure or an NP without an adjective. Low frequency words are associated with a delay in language processing and production, while higher frequency words are processed more rapidly. 109 If such frequency effects for lexical words were extended to syntactic structures, we would predict that less frequent syntactic structures would be processed more slowly, causing delays and duration differences in production. This duration lengthening in the production of less frequent structures could be interpreted by the listener as a different prosodic organization, with higher level prosodic boundaries corresponding to increased duration. Higher level prosodic boundaries would be directly associated with low syntactic frequency. This frequency factor could account for the differences in prosodic organization and phonetic production of DNA and DAN sequences, as schematized in (5-12). (5-12) DN HF # increased speed # lower level prosodic boundary (Pwd min ) DA LF # lower speed # higher level prosodic boundary (Pwd max ) 109 See chapter 3 on Spanish vowel hiatus production; Landauer and Streeter (1973) on lexical frequency and word recognition; Trueswell (1996) on lexical frequency and syntactic ambiguity resolution; Forster and Chambers (1973) on lexical frequency and visual word naming and lexical decisions; Taft and Hambly (1986) on lexical frequency and auditory lexical decisions; Rayner (1977) on lexical frequency and reading; Griffin and Bock (1998) on lexical frequency and spoken word production. 285 Despite the possibility of structural frequency as a factor on prosodic mapping, frequency cannot be the sole factor determining prosodification. If it were, we would predict that all IP level phrase boundaries would represent the onset of less frequent syntactic structures than those associated with PPh and Pwd boundaries. By extension, syntactic clauses would be predicted to be less frequent and more difficult to process than syntactic phrases. On the one hand, the element of syntactic recursivity may make smaller constituents like phrases more frequent than larger constituents in the sense that each larger constituent is composed of smaller constituents, which would therefore occur more often. On the other hand, larger constituents are also in many cases required for full grammatical sentences, and therefore highly frequent with respect to the proportion of utterances. Given these considerations in the nature of syntax, it is questionable whether frequent or infrequent is an appropriate descriptor for these constituents. While frequency may not be the only factor governing the formation of prosodic structures, it could be an influencing factor in addition to the syntactic structure inputs at the lower levels of the prosodic hierarchy. This approach follows Itô and Mester (2007) in dividing the mapping of prosodic structure above and below the Pwd, such that the grammar treats the mapping and well-formedness of higher-level prosodic units differently from the mapping and well-formedness of lower-level prosodic units. 110 At 110 Itô and Mester (2007) argue that the Pwd category is grouped together with PPh and IP with respect to Strict Layer Hypothesis violations, while the foot and syllable units are grouped together below the Pwd level. They specifically argue that at the Pwd level (as well as PPh and IP), there are no violations of LAYERING (i.e., no skipping of levels). I depart from this view in arguing that there are LAYERING violations in the free clitic construction, with a syllable directly dominated by a PPh instead of an intervening Pwd. Instead of grouping Pwd with the higher level categories in this respect, I assume that it is grouped with the lower level prosodic units in regard to LAYERING, as well as in regard to the possible effect of frequency. 286 these lower levels of the prosodic hierarchy, frequency, not syntactic structure, could be the main factor influencing prosodic construction. At the higher levels of the prosodic structure, syntactic inputs would remain as the primary source of prosodic structure building. While this frequency approach to prosodic structure building may fit the particular Spanish determiner cliticization data, it also makes strong predictions about Pwd structures crosslinguistically. If syntactic frequency determines whether a string of words is prosodified as containing Pwd min or Pwd max boundaries intervening, we would predict a language with a sequence of Pwd min level constituents (without Pwd max boundaries at all), corresponding simply to the more frequent syntactic structure. This would be directly contrary to the recursive notion of Pwd min and Pwd max subcategories. Another issue for an approach depending on syntactic frequency is how to define it. How would syntactic structures be compared to evaluate their frequency? And how would lexical frequency play a role? If there are two infrequent lexical words in a standard, frequent structure, would the overall structure be considered more or less frequent? These questions lead to rethinking “structural” frequency, in favor of the notion of bigram frequency, which is more compositional. Bigram frequency is the frequency with which two words occur together in the string (whether or not they are in a particular syntactic structure). Bigram frequency may thus account for certain “structural” frequencies, like the higher occurrence of determiner-noun strings in comparison to determiner-adjective strings in Spanish, but it also takes into account the individual 287 lexical frequencies of the two words. If one word is very infrequent, then its bigram with a frequent word will still be relatively infrequent. Again, bigram frequency makes certain predictions about prosodic cliticization if it is taken to be an influencing factor in prosodification. Since some words are more frequent than others—and some words are more frequent in certain positions than others—bigram frequency predicts that prosodification will be dependent on the specific lexical item. That is, it predicts that a language such as Spanish would prosodify determiners with some prenominal adjectives as free clitics while they would be affixal clitics with other, more frequent prenominal adjectives. An example of such a distinction between frequent and infrequent prenominal adjectives is shown in (5-13). (5-13) a. Frequent prenominal adjective (" (! max la D (! min buena A )) (! max madre N )) ‘the good mother’ b. Infrequent prenominal adjective (" la (! max cariñosa) (! max madre N ) ) ‘the affectionate mother’ It remains to be seen whether this strong prediction is attested in Spanish or other languages. 111 111 There are a handful of adjectives in Spanish that often occur prenominally, including buen/buena ‘good’, gran/grande ‘great’, mal/mala ‘bad’, viejo/vieja ‘old’, and nuevo/nueva ‘new’. Some adjectives commonly appear prenominal to certain nouns, such as the phrase alto funcionario ‘high functionary’, but do not commonly occur prenominally with other nouns (though it is possible). The majority of these common prenominal adjectives are consonant-initial, and therefore are not appropriate for the vowel hiatus tests utilized in Chapter 3. However, future research may yield other phonetic cues for differentiation between the hypothesized Pwd min and 288 5.4 Direct reference Another option in accounting for the data in question is to use a direct reference interface. Direct reference theories of the syntax-phonology interface have been recently proposed by Wagner (2005, 2010) and Pak (2008), in which the prosodic domains referenced by the phonology are not subject to separate well-formedness conditions pertaining to the prosodic grammar, but instead follow directly and solely from syntactic information. In a similar direct reference proposal of the Spanish feminine el (Varis 2009), I suggested that phonology could reference the syntactic head-complement relation directly rather than a prosodic constituent. The syntactic head-complement relation was proposed as a positional markedness environment for phonological constraints, encompassing the two syntactic constituents and the border between them. In direct reference, the difference visible to phonology between Spanish DNA and DAN word orders is not the prosodic cliticization structure, but the syntactic relations themselves. In the DN sequence, the relationship between the D and N is that of head of the phrase (D) and the head of its complement (N). In the DA sequence, the relationship between the D and A is that of head of the phrase (D) and non-head of the complement (A). A positional markedness environment would target the head-of-complement (HComp) relationship. With the Minimum Distance markedness constraint for the feminine el, such an environment would augment the perceptual transition between the two items. Pwd max boundaries of affixal and free clitics, and according to the frequency approach these words would be expected to demonstrate cues of the Pwd min boundary, as with affixal clitics. 289 (5-14) MINDISTV'V(HCOMP):F1/F2=1 - For adjacent vowels V 1 and stressed V 2 across a head of complement boundary, the two vowels must differ in F1 or F2 values by at least 1. With this syntactically-sensitive constraint, the transition between an item and the head of its complement is targeted for increased perception, highlighting the syntactic relationship between the items. With the Spanish data, the DN sequence matches the HComp environment, while the DA sequence does not, since A is in the specifier position of the FP complement. Highlighting the head-complement relationship in this way is consistent with the notion of positional markedness, and augmenting well-formedness for an already prominent position. Here, the head-complement relation is the prominent “position”. As argued in §2.2, the head-complement relationship is one of the most, if not the most, important of syntactic relations. Phonological strengthening of the complement relation merely augments and highlights an already vital relationship in another way. The direct approach closely follows the cliticization analysis laid out in Chapter 2, with the replacement of the syntactic HComp environment instead of the prosodic Pwd max . (5-15) MINDISTV'V(HCOMP)=F1/F2:1 >> IDENT-MP >> MINDISTVV(HCOMP)=F1/F2:1, MINDISTVV=F1/F2:1 290 A few challenges for the direct approach enter when analyzing the additional data in Spanish and other languages. For Spanish conjunction and Catalan vowel deletion, for example, the HComp sensitivity must be replaced with a general syntactic complement target. Both of these language patterns are not sensitive to headedness vs. non-headedness within the complement, as reviewed in (5-16) with intervening adjectives or adverbs. (5-16) Spanish conjunction culto *y/e inteligente Head ‘cultured and intelligent’ culto *y/e inevitablemente educado Head ‘cultured and inevitably educated’ Catalan vowel-deletion la/*l’ idea Head ‘the idea’ la/*l’ incendiària mescla Head ‘the incendiary mixture’ In a direct interface approach, this would need to be couched as a head_complement (Comp) context, instead of head_head of complement (HComp), such that the border between the head D or Conjunction and the edge of its XP complement is highlighted. The difference between the Comp and HComp contexts are schematized in (5-17) for the DP structure. 291 (5-17) DP Comp D FP AP FP A NP HComp D N For both Catalan vowel deletion and Spanish conjunction, the relevant positional markedness context would be (Comp), allowing alternation to occur at the border between the head and the first item contained in its complement. The Comp relation is identified for the Catalan DP (5-18a) and Spanish conjunction (5-18b) below. (5-18) a. [ DP la Dhead [ FP incendiària APspec [mescla Nhead …]]] Comp b. [ ConjP culto APSpec [e head [ AP [inevitablemente AdvP ] educado APhead ] Comp This is compared with the Spanish feminine el case, which uses the HComp context to differentiate between the presence or absence of a prenominal adjective. Without an adjective intervening, the adjacent clitic D and following host word are in an 292 HComp relationship (5-18a). When an adjective intervenes, the clitic’s HComp partner is farther removed, and the adjacent host is instead merely HC (5-19b). (5-19) a. [ DP el Dhead [ FP [agua Nhead …]]] HComp b. [ DP la Dhead [ FP agria APspec [agua Nhead …]]] Comp HComp An advantage to the direct HComp approach to the syntax-phonology interface is that it avoids the issue of grammaticalization in the phonological pattern. 112 Recall that the current pattern of the feminine el always uses el F when the determiner is directly adjacent to the noun, even if individual Pwds are promoted to PPh status, or determiners are promoted to Pwds, in stylized speech. Instead of the current phonological pattern being a memorized alternation or in OO Correspondence with specific speech registers, direct access to syntax allows the phonology to maintain the feminine el as productive in the synchronic grammar, but still independent of prosodic variability. Further research on the productivity of the phenomenon may help to distinguish between the predictions of the direct interface approach and the indirect prosodic approach pursued in this dissertation. One issue with the direct approach is that it does not recognize the phonetic differences between DNA and DAN sequences as related to prosodic structure. The 112 Though see §2.4.1 for how transderivational correspondence could account for non-variability resulting from extra-linguistic influences on prosody. 293 previous direct proposal in Varis (2009) assumed no phonetic distinction between these two sequences and therefore, no evidence for distinct prosodic structures. Given the evidence presented in Chapter 3 that there are phonetic differences between DA and DN, the direct interface theory must then assume that these differences are not due to different prosodic structures. Instead, the direct interface explanation for the phonetic distinction might be syntactic structure frequency, as discussed in the previous section. This frequency-related delay would have to be unrelated to the slow-down effects of different prosodic boundaries. Further work on the phonetic reflections of prosody and frequency is needed to tease apart the predictions of these different hypotheses. Another issue with the direct use of syntactic relations in the phonology, in comparison to utilizing the relation merely to influence aspects of prosodic structure formation, is that it requires much more syntactic information in order to produce the attested outcome, and therefore suffers from overgeneration. As discussed above, it is not only the head of complement (HComp) relation that must be visible to phonology, but also the relationship between a head and its complement XP (HC) that must be available. In contrast, the head-complement prosodic approach proposed in Chapter 2 only views the HComp relation between the terminal node lexical items, and allows the ranking of other prosodic mapping and well-formedness constraints to govern conjunction and other function word behavior. The direct approach’s inclusion of head and XP level structure in the phonology raises questions about how much syntactic structure is available. If phonology may access any or all syntax directly, we would predict languages whose 294 phonological alternations are wholly dependent upon syntactic relations such as c- command or long-distance anaphora, for example. 5.5 Articulatory vowel restrictions Another alternative to the feminine el analysis pursued in this work provides a different perspective to the MINDISTVV constraint family proposed for hiatus sensitivity. The MINDISTVV constraints address vowel hiatus from a perceptual approach, favoring more contrastive adjacent vowel pairs over less contrastive ones to augment the perception and identification of the hiatus vowels. One could also imagine restrictions on vowel hiatus stemming from articulatory or prosodic considerations, an approach taken by Kawu (2000) and Orie and Pulleyblank (2002). An articulatory approach to vowel hiatus resolution emphasizes the drive to diminish the difficulty in producing one vowel sound immediately followed by another vowel sound. The idea is that heterosyllabic vowels in sequence requiring different articulatory configurations are more difficult to produce than a single articulatory configuration. An example of such a hiatus constraint is presented in (5-20). 113 (5-20) *V x .V y – Assign a violation for a sequence of non-identical heterosyllabic vowels. One way to repair the articulatory difficulty of forming the appropriate contours for two different vowels is to eliminate one vowel altogether, i.e. deletion. Another repair 113 In Orie and Pulleyblank (2002), this constraint is presented as a restriction on heterosyllabic vowels that are linked to different root nodes. 295 that would satisfy the constraint in (5-20) is to assimilate one vowel to the other, forming a sequence of two vowels that are identical. This would satisfy the articulatory requirement by allowing the tongue to remain in position for the same vowel sound. This last assimilation repair is in direct conflict with the perceptual contrast approach accounting for the Spanish feminine el and other language patterns examined in this dissertation. Were such an articulatory perspective on hiatus assumed for Spanish, the identical adjacent vowels aa would be favored over more contrastive vowels, when in fact it is an identical sequence specifically that is forbidden. Table 55. Failure of articulatory *V x .V y /{el M ,la F } águila F / *V x .V y IDENT-MP ! a. la F águila F b. el F águila F *! As shown in Table 55, the constraint *V x .V y does not restrict identical hiatus. Candidate (a) satisfies the articulatory restriction in that the heterosyllabic vowels are of the same quality, and do not require any change of the vocal tract. The feminine el candidate (b) violates IDENT-MP Faith and would be ruled out in this ranking, wrongly predicting la águila as the optimal form. While this articulatory approach may account for other language patterns that demonstrate a preference for assimilation, an assimilatory repair is the precise output that is forbidden in Spanish. Perceptual contrast, on the other hand, accounts for patterns that 296 show dissimilation of adjacent vowels, and is clearly the driving factor governing hiatus patterns for Spanish determiners, conjunctions and other function words in Romance. 5.6 Summary of Chapter Five As discussed in this chapter, the various contending alternatives to the PARSE FNC approach to prosodic cliticization are either incapable of producing the data patterns attested, or are theoretically deficient. The Clitic Group constituent does not distinguish between DN and DA word orders in Spanish, instead grouping all words of the DP into a single prosodic constituent. PPh phrasing, without reference to word adjunction, suffers from the same problem. Frequency as a mapping factor provides an interesting and possibly productive avenue of investigation, but makes strong predictions regarding the prosodic structures of frequent and infrequent prenominal adjectives in Spanish that may or may not be borne out. A direct reference account may produce the necessary syntactic sensitivity in phonological segment adjacency, and has the advantage of circumventing any issue of grammaticalization and opacity of prosodic structure in the Spanish feminine el. But, it is theoretically less restricted than an approach to the syntax-prosody mapping which limits the influence of syntactic relations in prosodic structure formation, producing the potential for overgeneration. In addition, it ignores phonetic evidence of distinct prosodic structures for the Spanish DN and DA sequences, or requires the assumption that there is no relation between phonetics and prosody for these structures. On the purely phonological side of the analysis, an articulatory approach to vowel hiatus 297 may be appropriate for other language patterns, but predict exactly the opposite pattern of hiatus tolerance from that shown in Spanish and Romance clitic contexts. The next chapter concludes with a review of the main contributions of this dissertation and areas for future research. 298 Chapter Six: Concluding remarks The work presented here examined the Spanish feminine el in all aspects, providing a number of contributions both to the study of the Spanish language and to the theories of phonological interfaces. The main contributions and implications of the feminine el analysis are reviewed in this chapter, together with discussion of directions for future research in this and other interface phenomena. 6.1 Contributions This dissertation was the first comprehensive analysis of the feminine el, a complex language phenomenon tying in multiple areas of language, including phonology, phonetics, morphology, syntax, and language change. The overall analysis thus provides several contributions to linguistic theory. Prosodic cliticization distinctions were proposed in accounting for sensitivity to syntactic sequencing differences in the Spanish DP, supporting the presence of recursive and non-recursive Pwd cliticization structures in the same language, depending on subtle differences in the syntactic input. In this respect, the pattern also supports the maximal and minimal prosodic level distinctions pursued in recent developments of the Prosodic Phonology interface (Itô & Mester 2006, 2007, 2008, 2010; Elfner 2011). Another contribution made in the syntax-phonology interface is the development of the syntax-prosody mapping at the Pwd level of the prosodic hierarchy. Extending the PARSE mapping constraints proposed by Itô and Mester (2008), this dissertation proposes a constraint directing function words to be parsed in the Pwd of their lexical complement 299 head: PARSE FNC -INTO-! HCOMP . This constraint introduces the syntactic head-complement relation into the prosodic mapping in a limited context, and makes strong empirical predictions regarding prosodic patterns in language, such as the possibility of PF movement of a function word to be adjacent to its lexical complement head, as I elaborate on in the next section. In the phonological analysis, a novel application of Minimum Distance constraints on perceptual contrast was proposed for adjacent vowels. These new MINDISTVV constraints utilize the functionally-driven notion of contrast for segment transitions to provide a unified account of gradient sensitivity to vowel hiatus in the Spanish feminine el, Spanish conjunction, and Catalan function words. The historical development of the feminine el and synchronic patterns of variation also provided support for an analysis of morpho-phonological (MP) correspondence. The IDENT-MP constraint proposed in this analysis of the current feminine el morpho- phonological repair accounts well for the historically increasing restrictions on the set of words participating in the alternation, as well as patterns of variation using alternate repair forms. The concept of “correspondence on correspondences”, or MP relationships, further develops previous theoretical work in this interface area (Walker & Feng 2004, Wolf 2008). In exploring the feminine el data, new phonetic evidence was provided for distinctions of the lower levels of the prosodic hierarchy, particularly different clitic-Pwd junctures. Evidence from duration and formant co-articulation effects distinguished between determiner-noun and determiner-adjective sequences in Spanish. While these 300 sequences prosodify identically at the phonological phrasal level, the acoustic data provides support for a distinction in their organization as different prosodic clitics. 6.2 Further research Many aspects of the theoretical analysis and phonetic data presented in this dissertation raise interesting questions for future research. The relationship between syntax and prosodic cliticization is one that deserves continuing research, in several respects. First, the PARSE FNC constraint proposed here predicts the possibility of function word movement to attach adjacent to its lexical complement head, as well as other possible interaction patterns, such as ternary prosodic branching or compounding. While these candidates are ruled out in Spanish by higher-ranking constraints such as LINCORR and BIN-MAX(!), the possibility remains for other language patterns. Function word movement in PF has been proposed in previous work on Distributed Morphology (Embick & Noyer 2005), and further research into the prosody of other languages may provide evidence in favor of (or contra) the syntactic head-complement relation as introduced in this constraint. In this dissertation, the feminine el and other language patterns like Russian prepositional allomorphy have been analyzed in a mapping framework sensitive to both structural and relational syntactic information, but inclusion or exclusion of internal syntactic relations (such as headedness) in the syntax-phonology interface is an area of strong debate. Future investigation into additional data patterns that have been proposed with phonological sensitivity to syntactic relations (e.g. Chen 1990, 301 Billings 1996, Pensalfini 2002), may provide additional support for one perspective over the other. The vowel adjacent MINDISTVV constraints proposed in this dissertation bring up questions about the nature of adjacency restrictions. In the feminine el and additional data patterns examined in Chapter 4, two different sources of segment adjacency markedness surfaced: MINDISTVV for vowels (Spanish and Catalan), and a scale of C-V transition requirements (Cibaeño Spanish). Intuitively, these are very similar restrictions on segment adjacency, in that certain transitions of one sound to another are preferred over others for perceptual reasons. Future research is needed to examine the relationship between these different restrictions, and whether or not they may be unified to capture the similarities between them. The phonetic data collected here has promising implications for future research in the prosodic interface between syntax and phonology. The differences found in DA and DN sequences, as well as their similarities and differences from verbal direct object clitics, indicates that phonetic reflections of prosodic clitic structures are far more gradient than previously realized. This opens up an area of future study in prosody, to further examine the subtle phonetic differences in productions of clitic and Pwd structures. Furthermore, the issue of frequency may be investigated further in relation to these phonetic reflections of prosodic clitics. Future studies overtly controlled for structural frequency should yield more data on its effect on prosodic boundaries. In addition, I suggest that frequency be further examined in historical studies, to explore the 302 possibility of its influence upon the development of prosodic structures, and other phonological phenomena that appear sensitive to prosodic structures. The historical aspect of the feminine el data, and synchronic variation in morphological interpretation, return focus to the Spanish phenomenon to answer the question: where does the language go from here? The number of lexical exceptions to the feminine el pattern’s decreasing set of feminine el nouns, together with alternate repairs evinced in synchronic masculine interpretation of the determiner and feminine regularization in child language, suggest that the feminine el is unstable. Future research is needed into the productivity of the pattern, to see what speakers will do with novel forms that provide the phonological and syntactic triggers. A nonce word experiment, for example, would be one of the next steps to examine productivity and determine the current direction for the language’s grammar. A study designed for this purpose would present novel words in the same phonological context as the restricted class, to reveal whether speakers will continue to produce the feminine el, or whether the feminine el is restricted to a closed class of words. A sample stimuli paradigm for the nonce word ‘aca’ is suggested in (6-1). (6-1) a. aca roja AFem ‘red aca’ # la/el aca roja to test the acceptability of el F or la F b. aca roja AFem ‘red aca’ # la alta aca roja/el alta aca roja/el alto aca roja to test the acceptability of el F , el M , or laF with prenominal adjective alto M or alta F 303 Additionally, this research direction could yield information on the precise nature of the current fixed pattern by determining whether variable prosodic factors would have an effect, or whether speakers have internalized the pattern as purely syntactic, which would indicate a direct flow of information from syntax to phonology. 6.3 Summary In sum, this dissertation provides an important contribution to linguistic analysis of the Spanish language in its thorough treatment of the feminine el. In doing so, it adds significant developments to the theoretical syntax-phonology interface, as well as areas of the morphology-phonology interface and phonetic-prosodic research. It is hoped that the analyses and data presented will inspire further theoretical and empirical research in these interface areas. 304 References Abney, Steven. 1987. The English noun phrase in its sentential aspect. Ph.D. dissertation: MIT. Adger, David. 2003. Core Syntax. Oxford University Press. Aguilar, Lourdes. 1999. Hiatus and diphthong: acoustic cues and speech situation differences. Speech Communication 28:57-74. Alba, Matthew C. 2006. Accounting for Variability in the Production of Spanish Vowel Sequences. In N. Sagarra & A. J. Toribio (Eds.), Selected Proceedings of the 9th Hispanic Linguistics Symposium (pp. 273-285). Somerville, MA: Cascadilla Proceedings Project. Alexiadou, A. 2003. Adjective Syntax and (the absence of) noun raising in the DP. Proceedings of the Workshop on Head-movement. UCLA. 1-39. Anderson, Stephen R. 2005. Aspects of the Theory of Clitics. Oxford Studies in Theoretical Linguistics. Oxford University Press. Antilla, Arto. 2002. Morphologically Conditioned Phonological Alternations. NLLT 20, 1: 1-42. Atkinson, Emily. 2012. Gender features on n and the root: An account of gender in French. Presented at the Linguistic Symposium on Romance Languages (LSRL) 42, Southern Utah University, April 20-22, 2012. Baayen, R. Harald. 2008. Analyzing Linguistic Data: A Practical Introduction to Statistics Using R. Cambridge: Cambridge University Press. Bacovi$, Eric. 2000. Harmony, dominance and control. PhD dissertation: Rutgers University. Bacovic!, Eric. 2006. Hiatus resolution and incomplete identity, in Fernando Martinez- Gil, Sonia Colina (eds.) Optimality-theoretic studies in Spanish phonology, pgs. 62-73, John Benjamins Publishing Company. Baltazani, Mary. 2006. Focusing, prosodic phrasing and hiatus resolution in Greek. Laboratory Phonology 8: 473-494. Bao, Zhiming. 1996. “Local tree geometry and the phonology-syntax interface”, in U. Kleinhenz (ed.), Interfaces in Phonology, 27-45. [Studia Grammatica 41.] Berlin: Akademieverlag. 305 Barnes, Jonathan. 2006. Strength and Weakness at the Interface: Positional Neutralization in Phonetics and Phonology. Berlin/New York: Mouton de Gruyter. Bates, D. and M. Martin. 2010. lme4: Linear mixed-effects models using S4 classes. R package version 0.999375-37. Beckman, Jill, 1998. Positional Faithfulness. University of Massachusetts, Amherst: dissertation. ROA-234. Beckman, Mary and Janet Pierrehumert. 1986. Intonational Structure in Japanese and English, Phonology Yearbook III:15-70. Beckman, Mary, Manuel Diaz-Campos, Julia Tevis McGory, and Terrell A. Morgan. 2002. Intonation across Spanish, in the Tones and Break Indices framework. Probus 14.9-36. Bello, Andrés. 1988. Gramática de la Lengua Castellana. Con notas de Rufino José Cuervo. Madrid: Arcos/Libros. Benua, Laura. 1995. “Identity effects in morphological truncation”, in J. Beckman, S. Urbanczyk & L. Walsh (eds.), University of Massachusetts Occasional Papers in Linguistics 18: Papers in Optimality Theory, Amherst, MA: Graduate Linguistic Student Association. Bermúdez-Otero, R. 1999. Constraint interaction in language change [Opacity and globality in phonological change.] PhD dissertation, University of Manchester /Universidad de Santiago de Compostela. Bernstein, Judy B. 1991. DP in French and Walloon: evidence for parametric variation in nominal head movement. Probus 3:2; 101-126. Bernstein, Judy B. 1993. Topics in the syntax of nominal structure across Romance. Ph.D. dissertation: CUNY. Bernstein, Judy B. 1999. Inversion in Romance: Insights from the Nominal Domain, presented at the Linguistic Society of America annual meeting, Los Angeles, January 8, 1999. Bernstein, Judy B. 2003. Word Order and Agreement in Possessive Constructions, presented at the Lisbon Workshop on Agreement, Universidade Nova de Lisboa, July 10, 2003. 306 Bernstein, Judy B. and Christina Tortora. 2005. Two types of possessive forms in English. Lingua 115:1221-1242. Bhatt, Rajesh. 2005. Long distance agreement in Hindi-Urdu. Natural Language & Linguistic Theory 23:757–807. Billings, Loren A. 1996. “Sandhi phenomena and language change”, in Ursula Kleinhenz (ed.), Interfaces in phonology. Berlin: Akademie Verlag, 60-82. (Studia grammatica 41.) Boeckx, Cedric. 2008. Bare Syntax. New York: Oxford University Press. Boersma, Paul and Bruce Hayes. 2001. Empirical tests of the Gradual Learning Algorithm. Linguistic Inquiry 32: 45-86. Boersma, Paul and David Weenink. 2008. Praat: doing phonetics by computer (Version 5.0.05) [Computer program]. Retrieved from http://www.praat.org/ Booij, Geert. 1996. Cliticization as prosodic integration. The Linguistic Review 13: 219-242. Borer, Hagit (2005), In Name Only: Structuring Sense, Volume I. Oxford: Oxford University Press. Borzone de Manrique, Ana. 1976. Acoustic study of /i,u/ in the Spanish diphthong. Language and Speech, vol 19, issue 2, pp 121-128. Bosque, Ignacio and Carme Picallo. 1996. Postnominal adjectives in Spanish DPs. Journal of Linguistics, vol. 32, no. 2, September 1996. Accessed online January 29, 2012. URL: http://www.jstor.org.libproxy.usc.edu/stable/4176372. Bouma, Gerlof. 2003. Doing Dutch pronouns automatically in Optimality Theory. Proceedings of the EACL 2003 Workshop on The Computational Treatment of Anaphora, Budapest. Browman, Catherine and Louis Goldstein. 1988. Some notes on syllable structure in articulatory phonology. Phonetica 45: 140-155. Browman, Catherine and Louis Goldstein. 1992. Articulatory Phonology: an Overview. Haskins Laboratories Status Report on Speech Research, SR 111/112: 23-42. Browman, Catherine and Louis Goldstein. 2001. Competing constraints on intergestural coordination and self-organization of phonological structures. Le Cahier de l’ICP, Bulletin de la Communication Parlée 5: 25-34. 307 Buckley, Eugene. 1997. Alignment in Manam stress. ROA: 183. Bybee, Joan. 2002. Word frequency and context of use in the lexical diffusion of phonetically conditioned sound change. Language Variation and Change 14:261- 290. Cabré, Teresa and Pilar Prieto. 2003. Vowel contact resolutions across word boundaries in Catalan. Proceedings of the 15th International Congress of Phonetic Sciences, Maria Josep Solé, Daniel Recasens, Joaquín Romero (eds.), pp. 1687-1690. Barcelona: Causal Productions Pty Ltd. Camacho, Jose. 1997. The Syntax of NP Coordination. PhD dissertation: USC. Camacho, Jose. 2003. The structure of coordination: conjunction and agreement phenomena in Spanish and other languages. Kluwer Academic Publishers. Carnie, Andrew. 2002. Syntax: a generative introduction. Malden, MA: Blackwell. Casali, Roderic F. 1998. Resolving Hiatus. Outstanding dissertations in Linguistics. Garland Publishing, Inc., New York. Catalan Dictionary. 2000. Routledge Reference. Chen, Mathew. 1985. Tianjin tone sandhi: Erratic rule application. Ms, University of California, San Diego. Chen, Matthew. 1990. “What must phonology know about syntax?”, in Inkelas & Zec (eds.) The Phonology-Syntax Connection. University of Chicago. Chitoran, Iona and José Ignacio Hualde. 2003. From hiatus to diphthong: the evolution of vowel sequences in Romance. Phonology 24: 37-75. Chomsky, Noam. 1970. Remarks on nominalization. In R. Jacobs and P. Rosenbaum (eds.), Readings in English Transformational Grammar. Waltham, MA: Blaisdell Publishing, 184-221. Chomsky, Noam. 1993. “A Minimalist Program for Linguistic Theory”, in K. Hale and S.J. Keyser (eds.) The View from Building 20. Cambridge: MIT Press. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press. 308 Chomsky, Noam. 2000. “Minimalist Inquiries: the framework”, in Roger Martin, David Michaels and Juan Uriagereka (eds.) Step by step: Essays on minimalist syntax in honor of Howard Lasnik, 89–155. Cambridge, MA: MIT Press. Chomsky, Noam. 2001. “Derivation by phase”, in Michael Kenstowicz (ed.) Ken Hale: A life in language, 1–52.Cambridge, MA: MIT Press. Cinque, Guglielmo. 1994. “On the evidence for partial N-movement in the Romance DP”, in Guglielmo Cinque et al. (eds.) Paths Towards Universal Grammar: Essays in honor of Richard S. Kayne, Washington DC: Georgetown University Press. Cinque, Guglielmo. 2003. The dual source of adjectives and XP- vs. N-raising in the Romance DP, presented at NELS 34, SUNY, Stonybrook NY, November 7-9, 2003. Cinque, Guglielmo. 2004. Deriving Greenberg’s universal 20 and its exceptions. Ms., University of Venice. Cinque, Guglielmo. 2010. The syntax of adjectives: a comparative study. Linguistic Inquiry Monographs 57, MIT Press. Clements, George & Ridouane, Rachid. 2006. Quantal Phonetics and Distinctive Features: a Review. Proceedings of ISCA Tutorial and Research Workshop on Experimental Linguistics, Athens, Greece. Colantoni, Laura, María Cristina Cuervo & Jose Ignacio Hualde. 2010. Stress as a symptom. Presented at the Linguistic Symposium on Romance Languages (LSRL) 40, March 26-28, 2010. Seattle, Washington. Colina, Sonia. 1997. Identity constraints and Spanish resyllabification. Lingua 103 (1997): 1-23. Crosswhite, Katherine. 2000. Sonority driven reduction. Berkeley Linguistics Society Proceedings. Crosswhite, Katherine. 2004. Vowel reduction. In Bruce Hayes, Robert Kirchner and Donca Steriade (eds.) Phonetically Based Phonology, pp.191-231. Cambridge University Press. Davies, Mark. 2006. A frequency dictionary of Spanish: core vocabulary for learners. Routledge. 309 Davis, Julia L. 2000. French liaison: a case study of the syntax-phonology interface. PhD dissertation: Indiana University. De Lacy, Paul. 2002. The formal expression of markedness. Ph.D. dissertation: University of Massachusetts, Amherst. GLSA Publications. ROA-542. Dehé, Nicole & Vieri Samek-Lodovici. 2009. On the prosody and syntax of DPs: evidence from Italian noun-adjective sequences. Natural Language Linguistic Theory 27: 45-75. Delgutte, Bertrand. 1997. Auditory Neural Processing of Speech, in W.J. Hardcastle and J. Laver (eds.) The Handbook of Phonetic Sciences. Oxford: Blackwell, pp. 507-538. Demonte, Violeta and Isabel Pérez-Jiménez. To appear. Closest Conjunct Agreement in Spanish DPs. Syntax and Beyond. Folia Linguistica. Demonte, Violeta. 1999a. “A Minimal Account of Spanish Adjective Position and Interpretation”, in Grammatical Analyses in Basque and Romance Linguistics, 45. Demonte, Violeta. 1999b. Semántica composicional y gramática. Los adjectivos en la interficie léxico-sintaxis. RSEL 29: 2. Demonte, Violeta. 2008. “Meaning-form correlations and the order of adjectives in Spanish”, in Ch. Keney and L. McNally (eds.) The Semantics of Adjectives and Adverbs, Oxford: Oxford University Press. Dumitrescu, Domnita and Mario Saltarelli. 1998. Two types of predicate modification: evidence from the articulated adjectives of Romanian. In José Lema and Esthela Treviño (eds.) Theoretical Analyses on Romance Languages. Amsterdam: John Benjamins. Eddington, David and José I. Hualde. 2008. El abundante agua fría: hermaphroditic Spanish nouns. Studies in Hispanic and Lusophone Linguistics 1.1: 5-31. Elfner, Emily. 2011. Syntax-prosody interactions in Irish. Ph.D. dissertation: University of Massachusetts, Amherst. Embick, David and Rolf Noyer. 2001. “Movement Operations after Syntax,” Linguistic Inquiry 32:4, 555–595. 310 Embick, David and Rolf Noyer. 2005. “Distributed Morphology and the syntax- phonology interface”, in G. Ramchand and C. Reiss (eds.) The Oxford Handbook of Linguistic Interfaces. Oxford University Press. Ferrell, James. 1958. “The systems of the third person pronoun with particular reference to the category of the preposition in Slavic languages”, in American Contributions to the Fourth International Congress of Slavicists (=Slavistic printings and reprintings 21). The Hague: Mouton de Gruyter, 77-100. Flemming, Edward. 1995. Auditory representations in phonology. Dissertation: University of California, Los Angeles. Flemming, Edward. 2004. Contrast and perceptual distinctiveness, in Hayes, Kirchner and Steriade (eds.) Phonetically Based Phonology, pp. 232-276. Cambridge University Press. Flemming, Edward. 2005. A phonetically-based model of phonological vowel reduction. Retrieved from http://web.mit.edu/flemming/www/. Forster, Kenneth I. and Susan M. Chambers. 1973. Lexical access and naming time. Journal of Verbal Learning & Verbal Behavior 12: 627-35. Fougeron, Cécile. 1999. Prosodically conditioned articulatory variations: a review, Working Papers in Phonetics 97:1-73. Fougeron, Cécile. 2001. Articulatory properties of initial segments in several prosodic constituents in French. Journal of Phonetics 29:109-135. Frampton, John and Sam Gutmann. 2000. Agreement is feature sharing. Ms. Northeastern University. Frampton, John and Sam Gutmann. 2006. “How sentences grow in the mind: efficient computation in Minimalist Syntax”, in Cedric Boeckx (ed.), Agreement systems, 121-157. Amsterdam: John Benjamins. Gafos, Adamantios and Stefan Benus. 2006. Dynamics of phonological cognition. Cognitive Science [2005 Rumelhart Prize Special Issue Honoring Paul Smolensky: Optimization and Grammar in the Cognitive Science of Language] 30 (5), pp. 905-943. Gafos, Adamantios. 1996. The articulatory basis of locality in phonology. Dissertation: Johns Hopkins University. 311 Gafos, Adamantios. 2006. Dynamics in grammar: comment on Ladd and Ernestus & Baayen. Laboratory Phonology 8: Varieties of Phonological Competence, eds. Goldstein, M. L., Whalen, D. H. and Best, C. Mouton de Gruyter: Berlin/New York, pp. 51-79. Garrido, Marisol. 2007. Diphthongization of mid/low vowel sequences in Colombian Spanish. In Jonathan Holmquist et al (eds.) Selected Proceedings of the Third Workshop on Spanish Sociolinguistics, pp. 30-37. Somerville, MA: Cascadilla Proceedings Project. Geirut, Judith A., Michelle L. Morrison and Annette Hust Champion. 1999. Lexical constraints in phonological acquisition. Journal of Child Language 26:261-294. Gnanadesikan, Amalia E. 1995. Markedness and faithfulness constraints in child phonology. University of Massachusetts, Amherst: Ms. Gnanadesikan, Amalia E. 1997. Phonology with ternary scales. University of Massachusetts, Amherst: Ph.D. dissertation. Goldsmith, John A. 1976. Autosegmental Phonology. PhD dissertation, MIT. Golston, Chris. 1996. Direct Optimality Theory: representation as pure markedness. Language 72: 713-748. González, Carolina. 2003. Interacción fonológico–morfológica de los artículos definidos en español e italiano. In Fernando Sánchez Miret (ed.), Actas del XXIII Congreso Internacional de Lingüística y Filología Románica, Salamanca, 24-30 septiembre 2001, vol. I. Tübingen: Max Niemeyer Verlag, 343–355. Griffin, Zenzi M. and Kathryn Bock. 1998. Constraint, word frequency and the relationship between lexical processing levels in spoken word production. Journal of Memory and Language 38:313-338. Grimshaw, J. 1991. Extended projections. Ms, Brandeis University. Gutiérrez-Rexach, Javier and Enrique Mallen. 2001/2002. "NP Movement and Adjective Position in the DP Phases." In Julia Herschensohn, Enrique Mallen, and Karen Zagona (eds.), Features and Interfaces in Romance. Amsterdam and Philadelphia: John Benjamins. Hall, T.A. 2006. Derived environment blocking effects, Natural Language and Linguistic Theory 24:803-846. 312 Halle, Morris and Alec Marantz. 1993. “Distributed Morphology and the pieces of inflection”, in K. Hale and S. Keyser (eds.) The View from Building: 20 Essays in Linguistics in Honor of Sylvain Bromberger, MIT Press, Cambridge,MA, 111–176. Halle, Morris and Alec Marantz. 1994. Some key features of Distributed Morphology. In A. Carnie, H. Harley & T. Bures (eds.) MIT Working Papers in Linguistics 21: Papers on Phonology and Morphology. Cambridge, MA: MITWPL, 275- 288. Halle, Morris. 1990. An Approach to Morphology. NELS 20:1. GLSA, University of Massachusetts, Amherst, 150-84. Harley, Heidi and Rolf Noyer. 1999. Distributed Morphology, Glot International 4:4, 3-9. Harris, James W. 1983. Syllable Structure and Stress in Spanish: A Nonlinear Analysis. Linguistic Inquiry Monographs, No. 8, pp. 1-158. Cambridge, MA. Harris, James W. 1987. Disagreement rules, referral rules and the Spanish feminine article el. Journal of Linguistics 23: 177-183 Harris, James W. 1991. The exponence of gender in Spanish. Linguistic Inquiry 22 no. 1: 27-62 Hayes, Bruce. 1989. “The prosodic hierarchy in meter”, in P. Kiparsky and G. Youmans, eds., Rhythm and Meter, pp. 201-260. New York: Academic Press. Hayes, Bruce. 1995. Metrical Stress Theory: Principles and Case Studies. University of Chicago Press. Hayes, Bruce. 1999. “Phonetically driven phonology: the role of Optimality Theory and inductive grounding”, in Michael Darnell, Edith Moravscik, Michael Noonan, Frederick Newmeyer, and Kathleen Wheatly (eds.), Functionalism and Formalism in Linguistics, Volume I: General Papers, John Benjamins, Amsterdam, pp. 243-285. Helsloot, Catharine Josephine. 1995. Metrical prosody: a template-and-constraint approach to phonological phrasing in Italian. Holland Academic Graphics. Herrick, Dylan. 2004. Neutralization is complete in Catalan vowel reduction. In the proceedings of the 2004 Linguistic Society of Korea International Conference, Yonsei University, Seoul, Republic of Korea, July 28-30, pp. 91-100. 313 Hewitt, Mark. 1994. Deconstructing foot binarity in Koniag Alutiig. Ms. University of British Columbia [ROA-12-0000 http://ruccs.rutgers.edu/roa.html]. Hill, Steven P. 1977. The n-factor and Russian prepositions: their development in 11 th - 20 th century texts (=Slavistic printings and reprintings 118). The Hague-Paris- New York: Mouton de Gruyter, 127-360. Holt, D. Eric 2003. Optimality Theory and Language Change. Dordrecht: Kluwer Academic Publishers. Hooper, Joan Bybee. 1976. “Word frequency in lexical diffusion and the source of morphophonological change”, in W. Christie (ed.), Current progress in historical linguistics, Amsterdam: North Holland, pp. 96-105. Hualde, José Ignacio and Mónica Prieto. 2002. On the diphthong/hiatus contrast in Spanish: some experimental results. Linguistics 40, 2: 217-234. Hualde, José Ignacio and Mónica Prieto. 2012. Lenition of sibilants in Catalan and Spanish. Presented at the Linguistic Symposium on Romance Languages (LSRL) 42, Southern Utah University, April 20-22, 2012. Hualde, Jose Ignacio. 2010. Secondary Stress and Stress Clash in Spanish. Selected Proceedings of the 4th Conference on Laboratory Approaches to Spanish Phonology, Cascadilla Proceedings Project: Somerville, MA. Iatridou, Sabine. 1990. About Agr(P), Linguistic Inquiry 21.4:551-577. Inkelas, Sharon and Draga Zec. 1990. The Phonology Syntax Connection. University of Chicago Press. Inkelas, Sharon, C. Orhan Orgun, and Cheryl Zoll. 1997. The implications of lexical exceptions for the nature of grammar. In Iggy Roca (ed.) Derivations and Constraints in Phonology, pp. 393-418. Oxford University Press. Inkelas, Sharon. 1989. Prosodic constituency in the lexicon. Ph.D. dissertation: Stanford University. Iscrulescu, Cristian. 2003. Realize-Morpheme: one Constraint, or a Family? WECOL Proceedings. 314 Iscrulescu, Cristian. 2005. Morpheme Realization and Morphological Coalescence: Evidence from Romanian. In Randall S. Guess and Edward J. Rubin (eds.) Theoretical and Experimental Approaches to Romance Linguistics: selected papers from the 34th Linguistic Symposium on Romance Languages (LSRL), Salt Lake City, March 2004, pgs. 145-159. John Benjamins Publishing Company. Itô, Junko and Armin Mester. 1992. Weak layering and word binarity. Ms.: UC Santa Cruz. Published in Takeru Honma, Masao Okazaki, Toshiyuki Tabata and Shin- ichi Tanaka (eds.) 2003. A New Century of Phonology and Phonological Theory. A Festschrift for Professor Shosuke Haraguchi on the Occasion of His Sixtieth Birthday. Tokyo: Kaitakusha. 26–65. Itô, Junko and Armin Mester. 1998. Markedness and word structure: OCP effects in Japanese. ROA 255. Itô, Junko and Armin Mester. 2006. Prosodic adjunction in Japanese compounds. Proceedings of FAJL 4, Osaka (MIT Working Papers in Linguistics). Itô, Junko and Armin Mester. 2007. Categories and projections in prosodic structure. Handout presented at OCP 4 (Old World Conference in Phonology), January 2007, Rhodes, Greece. Itô, Junko and Armin Mester. 2008. The extended prosodic word. Ms: UC Santa Cruz. Itô, Junko and Armin Mester. 2010. “The onset of the prosodic word,” in (Steve Parker, ed.) Phonological Argumentation: Essays on Evidence and Motivation. London: Equinox. Jackendoff, Ray. 1977. X-bar syntax: a study of phrase structure. Cambridge, MA: MIT Press. Janda, Richard D. 2003. “Phonologization” as the start of dephoneticization—or, on sound change and its aftermath: of extension, generalization, lexicalization and morphologization. In B.D. Joseph & R.D. Janda (eds.), The Handbook of Historical Linguistics, pp. 401-422. Oxford: Blackwell. Janda, Richard D. and Fabriola Varela-García. 1991. On lateral hermaphroditism and other variation in Spanish ‘feminine’ el. In Papers from the 27 th regional meeting of the Chicago Linguistic Society, 27: 276-290. Kager, René & Wim Zonneveld (1999b), "Phrasal phonology: an introduction", in R. Kager & W. Zonneveld (eds.), Phrasal Phonology, 1-34. Nijmegen: Nijmegen University Press. 315 Kainada, Evia. 2007. Prosodic boundary effects on duration and vowel hiatus in Modern Greek. In Proceedings of the XVIth ICPhS, pp. 1225-1228, Saarbruecken, Universitaet des Saarlandes. Kanerva, Jonni M. 1989. Focus and phrasing in Chichewa phonology. Ph.D. dissertation: Stanford. Kawu, Ahmadu Ndanusa. 2000. Hiatus resolution in Nupe, Journal of West African Languages 28.2:27-44. Kayne, Richard S. 1993. Toward a modular theory of auxiliary selection. Studia Linguistica 47.1: 3-31. Kayne, Richard S. 2000. On the Left Edge in UG. A Reply to McCloskey, Syntax 3:44- 51 (reprinted in Movement and Silence). Kiparsky, Paul. 2000. Opacity and cyclicity. The Linguistic Review 17: 351-365. Kirchner, Robert Martin. 1998. An effort-based approach to consonant lenition. Ph.D. dissertation: UCLA. Kirchner, Robert Martin. 2004. “Consonant Lenition” in Hayes, Kirchner & Steriade (eds.) Phonetically Based Phonology. Koopman, Hilda. 2006. Agreement configurations: in defense of “Spec head”. In C. Boeckx (ed.) Agreement Systems. John Benjamins. Krämer, Martin. 2009. The Phonology of Italian. Oxford University Press. Kurisu, Kazutaka. 2001. The Phonology of Morpheme Realization. University of California Santa Cruz: dissertation. ROA-490. Labov, William. 1963. The social motivation of a sound change. Word 19: 273-309. Ladd, D. Robert. 1986. Intonational phrasing: the case of recursive prosodic structure. Phonology Yearbook 3:311-340. Ladd, D. Robert. 1988. Declination 'reset' and the hierarchical organization of utterances. Journal of the Acoustical Society of America 84:530-544. Ladd, D. Robert. 1992. “An introduction to intonational phonology”, in Docherty and Ladd (eds.) Papers in Laboratory Phonology II: Gesture, segment, prosody. Cambridge University Press. 316 Landauer, T.K. and L.A. Streeter. 1973. Structural differences between common and rare words: failure of equivalence assumptions for theories of word recognition. Journal of Verbal Learning and Verbal Behavior 12.2: 119-131. Leben, William. 1973. Suprasegmental Phonology. PhD dissertation, MIT. Legendre, Géraldine, Yoshiro Miyata, and Paul Smolensky. 1990. Can connectionism contribute to syntax?: Harmonic grammar, with an application. Report CU-CS- 485-90. Computer Science Department, University of Colorado at Boulder. (www.cs.colorado.edu/department/publications/reports/docs/CU-CS-485-90.pdf.) Lehiste, Ilse. 1970. Suprasegmentals. MIT Press: Cambridge, MA. Lloyd, Paul M. 1987. From Latin to Spanish. Philadelphia: American Philosophical Society (Memoirs, Vol. 173). Lubowicz, Anna. 2002. Derived Environment Effects in Optimality Theory. Lingua 112, 243-280. Lubowicz, Anna. 2003. Contrast Preservation in Phonological Mappings. Ph.D. dissertation, University of Massachusetts, Amherst. GLSA Publishing. ROA-554. Lubowicz, Anna. 2005. Infixation as morpheme absorption. ROA-773. Marantz, Alec. 1992. Case and Licensing, ESCOL 91:234-53. Martín, Juan. 1995. On the syntactic structure of Spanish noun phrases. Ph.D. dissertation: USC. McCarthy, John J. (ed.) 2004. Optimality Theory in Phonology. Oxford: Blackwell Publishing. McCarthy, John J. 1993. A Case of Surface Constraint Violation. Constraint-Based Theories in Multilinear Phonology, Carole Paradis and Darlene LaCharité (eds.), special issue of Canadian Journal of Linguistics 38, 169–195. McCarthy, John J. 2002. A Thematic Guide to Optimality Theory. Cambridge University Press. McCarthy, John J. and Alan S. Prince. 1993. Prosodic Morphology I; Constraint interaction and satisfaction. MS, University of Amherst, Massachusetts and Rutgers University. 317 McCarthy, John J. and Alan S. Prince. 1999. Faithfulness and Identity in Prosodic Morphology, in René Kager, Harry van der Hulst, and Wim Zonneveld (eds.) The Prosody Morphology Interface. Cambridge: Cambridge University Press. Pp. 218-309. ROA- 216. McGory, Julia Tevis and Manual Diaz-Campos. 1999. Sp-ToBI (Spanish Tones and Breaks Indices) webpage. http://www.ling.ohio-state.edu/~tobi/sp- tobi/spanish.html Mendeloff, Henry. 1969. A Manual of Comparative Romance Linguistics, Phonology and Morphology. Catholic University of America Press, Inc., Washington, DC. Mester, Armin. 1994. The quantitative trochee in Latin, Natural Language and Linguistic Theory 12:1-61. Monachesi, Paola. 1996. The Syntax of Italian Clitics. Ph.D. dissertation: Tilburg. Munn, Alan Boag. 1993. Topics in the Syntax and Semantics of Coordinate Structures. PhD dissertation: University of Maryland. Navarro Tomás, Tomás. 1977. Manual de pronunciación española, 19th ed. Madrid: Consejo Superior de Investigaciones Científicas (Publicaciones de la Revista de Filología Española) [1st ed., 1918]. Nespor, Marina and Irene Vogel. 1982. Prosodic domains of external sandhi rules, in H. van der Hulst and N. Smith, eds., The Structure of Phonological Representation, Part I. Foris: Dortrecht. Nespor, Marina and Irene Vogel. 1986. Prosodic phonology. Dordrecht: Foris Publications. Nevins, Andrew. 2011. “Phonologically Conditioned Allomorph Selection,” in van Oostendorp, Marc, Colin J. Ewen, Elizabeth Hume and Keren Rice (eds), The Blackwell Companion to Phonology. Blackwell Publishing, 2011. Blackwell Reference Online. 21 September 2011 http://www.blackwellreference.com/subscriber/tocnode?id=g9781405184236_ch unk_g9781405184236101 Ní Chiosáin, Máire and Jaye Padgett. 2001. Markedness, segment realization and locality in spreading. In Linda Lombardi (ed.) Segmental Phonology in Optimality Theory, Cambridge University Press, pp. 118-156. Noyer, Rolf. 1997. Features, Positions and Affixes in Autonomous Morphological Structure. Garland Publishing, New York. 318 Ohala, John. 1993. “The phonetics of sound change”, in Carrie S. Masek, Roberta A. Hendrick, and Mary Frances Miller (eds.) Papers from the Parasession on Language and Behavior, Chicago Linguistic Society, pp. 178-203. Chicago: Chicago Linguistic Society. Orbis Latinus. 2002. Descriptive Catalan Grammar. http://www.orbilat.com/Languages/Catalan/Grammar/index.html. Webpage material provided by Alberto José Miyara, Jordi Floriach I Ytxart, and Michele Brunelli. Accessed May 16, 2012. Orgun, Cemil Orhan. 1996. Sign-based morphology and phonology: with special attention to Optimality Theory. PhD dissertation: UC Berkeley. Orie, Olanike Ola & Douglas Pulleyblank. 2002. Yoruba vowel elision: minimality effects, Natural Language & Linguistic Theory 20.1:101-156 Padgett, Jaye. 1997. Perceptual distance of contrast: vowel height and nasality, Phonology at Santa Cruz 5: 63-78. Padgett, Jaye. 2001. “Contrast dispersion and Russian palatalization”, in Elizabeth Hume and Keith Johnson (eds.) The role of speech perception in phonology, Academic Press, pp. 187-218. Padgett, Jaye. 2003a. Contrast and post-velar fronting in Russian, Natural Language and Linguistic Theory 21.1:39–87. Padgett, Jaye. 2003b. The emergence of contrastive palatalization in Russian. In Eric Hold (ed.) Optimality Theory and language change, Dordrecht: Kluwer Academic Press, 307-335. Padgett, Jaye. 2004. Russian vowel reduction and Dispersion Theory. Phonological studies 7, pp. 81-96. Kaitakusha, Tokyo. Padgett, Jaye. 2008. Glides, vowels and features. Lingua 118:1937-1955. Elsevier. Padgett, Jaye. 2012. “The role of prosody of Russian voicing”, in Borowsky, Kawahara, Shigya and Sugahara (eds.) Prosody Matters: Essays in Honour of Lisa Selkirk (Advances in Optimality Theory), Equinox Publishing, 2012. Pak, Majorie. 2008. The postsyntactic derivation and its phonological reflexes. Ph.D. dissertation: University of Pennsylvania. 319 Pater, Joe. 2000. Nonuniformity in English stress: the role of ranked and lexically specific constraints. Phonology 17:2. 237-274. Pater, Joe. 2006. The locus of exceptionality: morpheme-specific phonology as constraint indexation. In Leah Bateman & Adam Werle (eds.) Univeristy of Massachusetts Occasional Papers: Papers in Optimality Theory III, pp. 1-36. Amherst, MA: Graduate Linguistic Student Association. Pater, Joe. 2008. Gradual learning and convergence. Linguistic Inquiry 39/2:334-345. Pater, Joe. 2009. Morpheme-specific phonology: constraint indexation and inconsistency resolution. In Steve Parker (ed.) Phonological Argumentation: Essays on Evidence and Motivation, pp. 123-154. London: Equinox. Penny, Ralph. 2002. A history of the Spanish language, 2 nd edition. Cambridge: Cambridge University Press. Pensalfini, Rob. 2002. Vowel harmony in Jingulu, Lingua 112:561-586. Peperkamp, Sharon. 1997. Prosodic Words. HIL dissertations 34. The Hague: Holland Academic Graphics. Phillips, Betty S. 1984. Word frequency and the actuation of sound change. Language 60:320-342. Picallo, Carme. 1991. Nominals and nominalizations in Catalan. Probus 3.3: 279-316. Plank, Frans. 1984. Romance disagreements: phonology interfering with syntax. Journal of Linguistics, 20: 329-349. Polka, L. and O.-S.Bohn. 2003. Asymmetries in vowel perception, Speech Communication 41, 221-231. Potts, Christopher, Joe Pater, Karen Jesney, Rajesh Bhatt and Michael Becker. 2009. Harmonic Grammar with linear programming: from linear systems to linguistic typology. Ms., University of Massachusetts Amherst. [ROA-984]. Prieto, Pilar. 2006. Prosodic phonology in Spanish. In Sonia Colina and Fernando Martínez-Gil (eds.) Optimality Theoretic Advances in Spanish Phonology, pp 39-60. Amsterdam: John Benjamins. Prince, Alan and Paul Smolensky. 1993/2004. Optimality Theory: Constraint Interaction in Generative Grammar. Rutgers University Center for Cognitive Science Technical Report 2. ROA- 537-0802. 320 R development core team (2010). R: A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing, http://www.R- project.org. Rao, Rajiv. 2010. Final Lengthening and Pause Duration in Three Dialects of Spanish. Selected Proceedings of the 4th Conference on Laboratory Approaches to Spanish Phonology (edited by Marta Ortega-Llebaria). Cascadilla Proceedings Project. Somerville, MA. Rayner, Keith. 1977. Visual attention in reading: eye movements reflect cognitive processes. Memory and Cognition 5.4:443-448. Ritter, Elizabeth. 1991. Two functional categories in noun phrases: evidence from Modern Hebrew. Syntax and Semantics 25: 37-62. Robles-Puente, Sergio. 2010. Dissimilation and deletion of vowels in Basque dialects: an OT analysis. Ms., University of Southern California. Russell, Kevin. 1995. Morphemes and candidates in Optimality Theory. ROA- 44. Saltarelli, Mario. 2001. An integrated theory of adjectives: the alternating adjectives of Romance. In P. Bouillon and E. Viegas (eds.) Description des Adjectifs pour les Traitements Informatiques. Proccesdings of Traitement Automatique des Langues Naturelles, Cargèse, Corsica. July 12-17, 1999. P. 46-56 Salvá, Vicente. 1988. Gramática de la Lengua Castellana, según ahora se habla. Estudio y edición de Margarita Lliteras. Vol. I. Editorial Arco Libros, S.A., Madrid. Samek-Lodovici, Vieri. 1992. A unified analysis of cross-linguistic morphological gemination, Proceedings of CONSOLE-1 Utrecht, The Nederlands. Samek-Lodovici, Vieri. 1996. Constraints on subjects: an Optimality Theoretic analysis. Ph.D. dissertation: Rutgers University. Sanders, Robert N. 2003. Opacity and Sound Change in the Polish Lexicon. Ph.D. dissertation: University of California, Santa Cruz. Sands, Kathy. 2004. Patternings in the vocalic sequences of the world’s languages. Dissertation: UCSB. 321 Schwartz, Jean-Luc, Christian Abry, Louis-Jean Boë, Lucie Ménard, and Nathalie Vallée. 2005. Assymmetries of vowel perception, in the context of the Dispersion-Focalisation Theory. Speech Communication 45:425-434. Schwartz, Jean-Luc, Louis-Jean Boë, Nathalie Vallée, and Christian Abry. 1997. The Dispersion-Focalisation Theory of Vowel Systems, Journal of Phonetics 25, 255-286. Seidl, Amanda. 2001. Minimal Indirect Reference: a Theory of the Syntax-Phonology Interface. Outstanding Dissertations in Linguistics, Routledge: New York. Selkirk, Elisabeth and Koichi Tateishi. 1988. “Constraints on minor phrase formation in Japanese”, in Papers from the 24th Annual Regional Meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society. Selkirk, Elisabeth and Koichi Tateishi. 1991. “Syntax and downstep in Japanese”, in Interdisciplinary Approaches to Language: Essays in Honor of S.-Y. Kuroda. Dordrecht: Kluwer. Selkirk, Elisabeth and Tong Shen. 1990. “Prosodic domains in Shanghai Chinese”, in Sharon Inkelas & Draga Zec (eds.) The Phonology-Syntax Connection. Chicago: University of Chicago Press, 313-337. Selkirk, Elisabeth. 1978. On prosodic structure and its relation to syntactic structure. In T. Fretheim (ed.) Nordic Prosody. Trondheim: TAPIR, 111-140. Selkirk, Elisabeth. 1981. On the nature of phonological representation, in J. Anderson, J. Laver and T. Meyers (eds.) The Cognitive Representation of Speech, Amsterdam: North Holland. Selkirk, Elisabeth. 1984. Phonology and Syntax: the Relation between Sound and Structure. MIT Press: Cambridge, MA. Selkirk, Elisabeth. 1996. The prosodic structure of function words. In James L. Morgan & Katharine Demuth (eds.) Signal to Syntax: bootstrapping from speech to grammar in early acquisition. Lawrence Earlbaum Associates, Inc. Selkirk, Elisabeth. 2009. On clause and intonational phrase in Japanese: the syntactic grounding of prosodic constituent structure. Gengo Kenkyu 136: 35-73. Selkirk, Elisabeth. 2011. “The syntax-phonology interface”, in John Goldsmith, Jason Riggle, and Alan Yu (eds.) The handbook of phonological theory. Blackwell Publishing, Oxford. 322 Seo, Misun. 2003. A segment contact account of the patterning of sonorants in consonant clusters. Ph.D. dissertation, Ohio State University. Smith, Jennifer L. 2000. Positional faithfulness and learnability in Optimality Theory. In Rebecca Daly and Anastasia Riehl (eds.) Proceedings of ESCOL 99:203-214. Ithaca: CLC Publications. Smith, Jennifer L. 2005. Phonological Augmentation in Prominent Positions. Outstanding Dissertations in Linguistics. Routledge, New York. Smith, Jennifer L. 2011. Category-specific effects, in Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume and Keren Rice (eds.) The Blackwell Companion to Phonology, Vol IV: Phonological Interfaces. Blackwell Publishing. Smolensky, Paul and Géraldine Legendre. 2006. The Harmonic Mind: From Neural Computation to Optimality-Theoretic Grammar. Cambridge, MA: MIT Press. Smolensky, Paul. 1993. Harmony, markedness, and phonological activity. Handout of keynote address, Rutgers Optimality Workshop—1, October 23. ROA-87. Sosa, Juan Manuel. 1991. Fonética y fonología del español hispanoamericano. Ph.D. dissertation: University of Massachusetts, Amherst. Sosa, Juan Manuel. 1999. La entonación del español: Su estructura fónica, variabilidad y dialectología. Madrid: Cátedra. Steriade, Donca. 1999. “Lexical Conservatism in French Adjectival Liaison”, in B. Bullock, M. Authier and L. Reed (eds.) Formal Perspectives in Romance Linguistics, John Benjamins, pp. 243-270. Taft, Marcus & Gail Hambly. 1986. Exploring the Cohort Model of spoken word recognition. Cognition 22:259-282. Temkin-Martínez, Michal. 2010. Non-conformity in phonology: exceptionality and variation in Modern Hebrew spirantization. PhD dissertation: USC. Ticio, Maria Emma. 2003. “On the structure of DPs”, Dissertations Collection for University of Connecticut. Paper AAI3095847. Tilsen, Sam. 2011a. Effects of syllable stress on articulatory planning: evidence from a stop-signal experiment. Journal of Phonetics. Tilsen, Sam. 2011b. Metrical regularity facilitates speech planning and production. Laboratory Phonology. 323 Tranel, Bernard and Francesca Del Gobbo. 2002. Local Conjunction in Italian and French phonology, in C. Wiltshire and J. Camps (eds.) Romance Phonology and Variation, John Benjamins, 191-218. Truckenbrodt, Hubert. 1995. Phonological phrases: Their relation to syntax, focus, and prominence. PhD dissertation: MIT. Truckenbrodt, Hubert. 1999. On the relation between syntactic phrases and phonological phrases. Linguistic Inquiry 30:219–255. Trueswell, John C. 1996. The role of lexical frequency in syntactic ambiguity resolution. Journal of Memory and Language 35:566-585. Uriagereka, Juan. 1995. Aspects of the syntax of clitic placement in Western Romance. Linguistic Inquiry, 26: 79-123. MIT Press. Valois, Daniel. 1991. The internal syntax of DP and adjective placement in French and English. Proceedings of NELS 21, GSLA: 367-382. Varis, Erika. 2009. “Complimentary” Heads in Spanish: Syntactic Phonology. In proceedings for Israel Association of Theoretic Linguistics 25: October 12-13, 2009. Varis, Erika. 2010. The (masculine) art of Spanish: a diachronic shift along the morphology-phonology interface. Linguistic Symposium on Romance Languages (LSRL) 40, March 26-28, 2010. Seattle, Washington. Varis, Erika. 2011. Vowel hiatus and Dispersion Theory. Presented at the Linguistic Society of America (LSA) 85th Annual Meeting, January 6-9, 2011. Pittsburgh, Pennsylvania. Vennemann, Theo. 1988. Preference Laws for Syllable Structure. Berlin: Mouton de Gruyter. Wagner, Michael. 2005. Prosody and recursion. Ph.D. dissertation: MIT. Wagner, Michael. 2010. Prosody and recursion in coordinate structures and beyond. Natural Language & Linguistic Theory 28:183-237. Walker, Rachel & Feng, Bella, 2004. A Ternary Model of Morphology-Phonology Correspondence. WCCFL 23 Proceedings, B. Schmeiser, V. Chand, A. Kelleher and A. Rodriguez (eds.), pp. 773-786. Somerville, MA: Cascadilla Press. 324 Walker, Rachel. 2011. Vowel Patterns in Language. Cambridge University Press. Wheeler, Max, Alan Yates, and Nicolau Dols. 1999. Catalan: A Comprehensive Grammar. Routledge. Wheeler, Max. 2005. The Phonology of Catalan. Oxford University Press. Wolf, Matthew Adam. 2008. Optimal Interleaving: serial morphology-phonology interaction in a constraint-based model. PhD dissertation: University of Massachusetts, Amherst. Wright, Richard. 2004. A review of perceptual cues and cue robustness. In Bruce Hayes, Robert Kirchner and Donca Steriade (eds.) Phonetically Based Phonology, pp. 34-57. Cambridge University Press. Yip, Moira. 1999. “Feet, tonal reduction and speech rate at the word and phrase level in Chinese”, in René Kager and Wim Zonneveld (eds.) Phrasal Phonology, 171- 194, Nijmejen: Nijmejen University Press. [ROA-159-1196] Zagona, Karen. 2002. The Syntax of Spanish. Cambridge University Press. Zsiga, Elizabeth C. 1997. Features, gestures and Igbo vowels: an approach to the phonology-phonetics interface. Language 73, 2: 227-274. Zwicky, Arnold M. 1985. Rules of allomorphy and phonology-syntax interaction. Journal of Lnguistics, 21: 431-436. 325 Appendix A: Control and test items for Experiment 1 Test items VV Prosody Item aa IP Al pez no le gustó la atunera. Ató la red y él no pudo escapar. PP Cuando vino la atunera, ató la red. Pwd La atunera ató la red de pesca. Clitic La ató para no dejar escapar los peces. ae IP Les gustaba la antigua hotelera. Estaban contentos con ella. PP Cuando los dirigía la primera hotelera, estaban contentos. Pwd La hotelera estaba irascible con las empleadas. Clitic La heriste con ese escándalo. ai IP Ya no es la heredera. Hirió a la señora con su comportamiento irresponsable. PP Cuando supo quién era la heredera, hirió al viejo en venganza. Pwd Dicen que la heredera hirió al viejo para adelantar la muerte. Clitic La hirió cuando descubrieron su aventura amorosa. ao IP No quise acercarme a la atleta. Olía mal. PP Cuando ganó la atleta, olía a sudor y triunfo. Pwd La atleta olía el sudor de sus rivales. Clitic La atleta estaba muy sudorosa, la olí desde un kilómetro. au IP Me gusta la mezcla especial de la heladera. Usó chocolate, nueces y vainilla. PP Después de llegar la heladera, usó las palas para empezar el helado. Pwd La heladera usó tres sabores en su helado especial. Clitic La leche estaba muy fresca, y la usó en hacer un helado riquísimo. Control items, for the respective test VV sequences VV Prosody Item aa IP Al pez no le gustó la atunera. Temblaba cuando ella se le acercó. IP La atunera no le gustó a él. Ató la red y él no pudo escapar. PP Cuando vino la atunera, tiró la red. PP Cuando llegó el pescador, ató la red al bote. Pwd El pescador ató la red al bote. Pwd La atunera tomó la red del agua. Clitic La tasó para venderla según la norma. ae IP Le gustaba el antiguo árbol. Estaba rodeado de memorias. IP Les gustaba la antigua hotelera. Tasó las reservaciones con rapidez. PP Cuando florecía el árbol, estaba bonito. PP Cuando vino la hotelera, tasó las reservaciones. Pwd El árbol estaba en la bonita playa del hotel. Pwd La hotelera tasó las reservaciones según los clientes. Clitic Esa manta la tejió en sólo dos días. ai IP Ya no es la heredera. Tiró el testamento a la basura. 326 VV Prosody Item IP Ya no está el chofer. Hirió al viejo y lo despidieron. PP Según lo que dice él, hirió al viejo en venganza. PP Cuando supo quién era la heredera, tiró el vaso a la pared. Pwd Dijeron que él hirió al viejo y le echó la culpa a ella. Pwd Dicen que la atleta tiró la jabalina justo a tiempo. Clitic La tiró cuando llegó al fin de la carrera. ao IP Me gustaban las flores del árbol. Olían dulces y frescas. IP Corrió duro la atleta. Tosió fuertemente después de la carrera. PP Cuando terminó la atleta, tosió fuertemente. PP Cuando florecía el árbol, olía el aroma desde un kilómetro. Pwd El árbol olía a flores y verano. Pwd La atleta tosió mucho después de la carrera. Clitic Aunque la atleta estaba muy sudorosa, la toleraba porque había ganado. au IP Me gustó la mezcla especial de la heladera. Tomé dos bolas a la vez. IP El experto de bonsái crió ese árbol. Usó herramientas especiales para controlar el crecimiento. PP Cuando vino la heladera, tomó nota de los pedidos. PP Cuando empezaba a crecer el árbol, usó las herramientas para controlarlo. Pwd El árbol usó la luz para crecer. Pwd La heladera tomó un poco de cada sabor. Clitic La leche estaba muy fresca, y la tomó con gusto. 327 Appendix B: Control and test items for Experiment 2 Test items VV Syntax Item aa DNA La atunera alegre pescaba con ganas. DAN La alegre atunera pescaba por la mañana. ae DNA La etóloga elitista hablaba con arrogancia. DAN La elitista etóloga asistió a la universidad exclusiva. ai DNA La italiana irascible se enojaba fácilmente. DAN La irascible italiana se quejaba mucho. ao DNA La hotelera oronda comía con gusto. DAN La oronda hotelera le dio la bienvenida. au DNA La usurera judía salió de la ciudad. DAN La usual amante se ahogó por el amor. Control items V Syntax Item a DAN El alegre atunero pescaba por la mañana. DNA El atunero alegre pescaba con ganas. e DAN El elitista etólogo asistió a la universidad exclusiva. DNA El etólogo elitista hablaba con arrogancia. i DAN El irascible italiano se quejaba mucho. DNA El italiano irascible se enojaba fácilmente. o DAN El orondo hotelero le dio la bienvenida. DNA El hotelero orondo comía con gusto. u DAN El usual amante se ahogó por amor. DNA El usurero judío salió de la ciudad. 328 Estimate Std. Error t-value (Intercept) 1695.018 68.321 24.810 a followed by e 20.910 28.065 0.745 a followed by i 94.440 22.576 4.183* a followed by o -101.219 53.680 -1.886 a followed by u 46.719 24.877 1.878 PP vs. IP 21.659 22.438 0.965 Fixed effects second interval F2: V1 Estimate Std. Error t-value (Intercept) 1616.272 59.482 27.173 a followed by e 14.568 35.108 0.415 a followed by i 69.986 34.823 2.010* a followed by o -88.760 40.957 -2.167* a followed by u 64.480 37.037 1.741 PP vs. IP 24.173 23.388 1.034 Fixed effects third interval F2: V1 Estimate Std. Error t-value (Intercept) 1562.706 61.030 25.606 a followed by e 8.532 44.612 0.191 a followed by i 59.619 44.621 1.336 a followed by o -68.594 51.496 -1.332 a followed by u 81.898 48.155 1.701 PP vs. IP 12.016 30.714 0.391 Fixed effects fourth interval F2: V1 Estimate Std. Error t-value (Intercept) 1539.51 56.72 27.140 a followed by e 18.95 51.58 0.367 a followed by i 73.26 52.41 1.398 a followed by o -87.61 62.81 -1.395 a followed by u 52.95 54.33 0.975 PP vs. IP -27.63 33.93 -0.814 IP vs. PP: V2 Fixed effects first interval F1: V2 Estimate Std. Error t-value (Intercept) 674.54 29.09 23.187 e -220.31 31.96 -6.893* i -285.90 30.59 -9.346* o -138.50 28.95 -4.784* PP vs. IP 85.22 18.79 4.534* 329 (AC-6) Fixed effects second interval F1: V2 Estimate Std. Error t-value (Intercept) 676.900 37.489 18.056 e -232.231 40.612 -5.718* i -298.320 42.507 -7.018* o -152.155 36.389 -4.181* PP vs. IP 46.889 16.824 2.787* (AC-7) Fixed effects third interval F1: V2 Estimate Std. Error t-value (Intercept) 652.795 32.673 19.980 e -212.358 35.536 -5.976* i -277.536 39.773 -6.978* o -143.470 35.926 -3.993* PP vs. IP 14.914 19.502 0.765 (AC-8) Fixed effects fourth interval F1: V2 Estimate Std. Error t-value (Intercept) 619.18 37.40 16.556 e -182.74 38.45 -4.753* i -246.49 42.56 -5.792* o -141.63 40.75 -3.475* PP vs. IP 21.24 22.14 0.959 (AC-9) Fixed effects first interval F2: V2 Estimate Std. Error t-value (Intercept) 1468.539 58.031 25.306 e 538.335 51.725 10.408* i 1058.948 71.844 14.739* o -290.673 70.756 -4.108* PP vs. IP -8.931 42.300 -0.211 (AC-10) Fixed effects second interval F2: V2 Estimate Std. Error t-value (Intercept) 1495.122 56.763 26.340 e 522.477 48.969 10.670* i 1028.434 73.896 13.917* o -265.335 64.162 -4.135* PP vs. IP -19.157 42.417 -0.452 330 (AC-11) Fixed effects third interval F2: V2 Estimate Std. Error t-value (Intercept) 1541.762 58.325 26.434 e 465.418 45.120 10.315* i 963.759 76.875 12.537* o -227.601 65.930 -3.452* PP vs. IP -38.953 40.853 -0.953 (AC-12) Fixed effects fourth interval F2:V2 Estimate Std. Error t-value (Intercept) 1564.04 57.58 27.162 e 424.07 40.26 10.533 i 901.53 81.56 11.053 o -161.82 65.86 -2.457 PP vs. IP -49.37 38.17 -1.294 Pwd vs. clitics: V1V2 Estimate Std. Error t-value (Intercept) 680.51 40.34 16.870 ae -47.44 48.57 -0.977 ai -67.67 51.87 -1.305 ao -44.58 53.29 -0.837 au -77.71 52.93 -1.468 Pwds vs. clitics -43.80 48.80 -0.898 ae:Pwd 28.10 65.49 0.429 ai:Pwd -28.99 69.71 -0.416 ao:Pwd 100.24 72.01 1.392 au:Pwd 50.62 70.78 0.715 Estimate Std. Error t-value (Intercept) 699.995 64.540 10.846 ae -83.892 83.918 -1.000 ai -148.880 84.756 -1.757 ao -89.773 87.499 -1.026 au -154.145 85.039 -1.813 Pwds vs. clitics -27.505 83.417 -0.330 ae:Pwd 47.385 111.074 0.427 ai:Pwd 5.182 118.252 0.044 ao:Pwd 118.410 120.135 0.986 au:Pwd 66.548 119.025 0.559 Estimate Std. Error t value (Intercept) 686.50 93.93 7.309 ae -116.90 128.30 -0.911 ai -214.34 129.81 -1.651 331 ao -119.39 129.16 -0.924 au -209.69 128.72 -1.629 Pwds vs. clitics -4.64 127.28 -0.036 ae: Pwd 80.85 168.85 0.479 ai: Pwd 12.45 180.23 0.069 ao: Pwd 62.21 181.00 0.344 au: Pwd 26.39 180.55 0.146 Estimate Std. Error t value (Intercept) 639.8376 157.2890 4.068 ae -127.5346 220.3362 -0.579 ai -173.5295 220.6265 -0.787 ao -88.5560 221.6968 -0.399 au -169.0504 221.4674 -0.763 Pwd -6.2755 219.6034 -0.029 ae: Pwd 155.4324 291.0003 0.534 ai: Pwd -7.6492 310.7148 -0.025 ao: Pwd 8.2691 311.4788 0.027 au: Pwd -0.7699 311.0916 -0.002 Estimate Std. Error t value (Intercept) 1679.20 75.98 22.102 vowelae 237.16 48.22 4.918* vowelai 289.02 36.20 7.984* vowelao -228.15 55.42 -4.117* vowelau -201.43 51.22 -3.933* actualPwd 42.77 32.03 1.335 ae: Pwd -146.40 43.97 -3.329* ai: Pwd -37.21 47.77 -0.779 ao: Pwd 44.10 58.97 0.748 au: Pwd 101.36 53.39 1.898 332 Multiple comparisons of means: Simultaneous Tests for General Linear Hypotheses Fit: lmer(formula = F2_1 ~ vowel_actual + (1 + vowel | subj) + (1 | word), data = testunbrokenokcond) Linear Hypotheses: Estimate Std. Error z value Pr(>|z|) aa_Pwd - aa_clitic == 0 42.770 32.033 1.335 0.9220 ae_clitic - aa_clitic == 0 237.157 48.217 4.919 <0.01 *** ai_clitic - aa_clitic == 0 289.022 36.196 7.985 <0.01 *** ao_clitic - aa_clitic == 0 -228.154 55.415 -4.117 <0.01 ** au_clitic - aa_clitic == 0 -201.430 51.220 -3.933 <0.01 ** ae_Pwd - aa_Pwd == 0 90.758 48.809 1.859 0.6286 ai_Pwd - aa_Pwd == 0 251.810 38.732 6.501 <0.01 *** ao_Pwd - aa_Pwd == 0 -184.071 62.306 -2.954 0.0703 au_Pwd - aa_Pwd == 0 -100.074 52.592 -1.903 0.5973 ae_Pwd - ae_clitic == 0 -103.629 30.425 -3.406 0.0174 * ai_Pwd - ai_clitic == 0 5.558 36.597 0.152 1.0000 ao_Pwd - ao_clitic == 0 86.853 49.491 1.755 0.7025 au_Pwd - au_clitic == 0 144.126 43.662 3.301 0.0244 * Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) (AC-17) Fixed effects for second interval F2 Estimate Std. Error t value (Intercept) 1598.60 69.07 23.146 vowelae 367.31 46.55 7.891 vowelai 521.25 55.36 9.416 vowelao -239.68 58.82 -4.075 vowelau -371.42 56.09 -6.622 actualPwd 26.72 33.45 0.799 ae: Pwd -184.23 45.58 -4.042 ai: Pwd -90.27 51.75 -1.744 ao: Pwd 17.08 63.95 0.267 au: Pwd 110.52 57.69 1.916 333 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means: Fit: lmer(formula = F2_2 ~ vowel_actual + (1 + vowel | subj) + (1 | word), data = testunbrokenokcond) Linear Hypotheses: Estimate Std. Error z value Pr(>|z|) aa_Pwd - aa_clitic == 0 26.720 33.449 0.799 0.9979 ae_clitic - aa_clitic == 0 367.313 46.549 7.891 <0.01 *** ai_clitic - aa_clitic == 0 521.252 55.358 9.416 <0.01 *** ao_clitic - aa_clitic == 0 -239.685 58.816 -4.075 <0.01 ** au_clitic - aa_clitic == 0 -371.425 56.085 -6.622 <0.01 *** ae_Pwd - aa_Pwd == 0 183.079 47.133 3.884 <0.01 ** ai_Pwd - aa_Pwd == 0 430.982 57.667 7.474 <0.01 *** ao_Pwd - aa_Pwd == 0 -222.608 67.231 -3.311 0.0261 * au_Pwd - aa_Pwd == 0 -260.908 58.960 -4.425 <0.01 *** ae_Pwd - ae_clitic == 0 -157.514 31.447 -5.009 <0.01 *** ai_Pwd - ai_clitic == 0 -63.549 39.707 -1.600 0.8097 ao_Pwd - ao_clitic == 0 43.797 54.470 0.804 0.9978 au_Pwd - au_clitic == 0 137.237 47.203 2.907 0.0841 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) (AC-18) Fixed effects for third interval F2 Estimate Std. Error t value (Intercept) 1604.69 65.16 24.625 ae 437.42 53.14 8.231 ai 684.41 71.96 9.512 ao -286.73 52.74 -5.437 au -473.12 73.25 -6.459 Pwds vs. clitics -15.71 34.85 -0.451 ae: Pwd -195.13 47.51 -4.107 ai: Pwd -99.03 54.57 -1.815 ao: Pwd 23.70 65.44 0.362 au: Pwd 170.80 60.49 2.824 334 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means: Fit: lmer(formula = F2_3 ~ vowel_actual + (1 + vowel | subj) + (1 | word), data = testunbrokenokcond) Linear Hypotheses: Estimate Std. Error z value Pr(>|z|) aa_Pwd - aa_clitic == 0 -15.705 34.846 -0.451 0.99998 ae_clitic - aa_clitic == 0 437.423 53.143 8.231 < 0.001 *** ai_clitic - aa_clitic == 0 684.411 71.953 9.512 < 0.001 *** ao_clitic - aa_clitic == 0 -286.742 52.738 -5.437 < 0.001 *** au_clitic - aa_clitic == 0 -473.113 73.250 -6.459 < 0.001 *** ae_Pwd - aa_Pwd == 0 242.290 53.457 4.532 < 0.001 *** ai_Pwd - aa_Pwd == 0 585.378 74.041 7.906 < 0.001 *** ao_Pwd - aa_Pwd == 0 -263.042 61.693 -4.264 < 0.001 *** au_Pwd - aa_Pwd == 0 -302.312 76.275 -3.963 0.00236 ** ae_Pwd - ae_clitic == 0 -210.838 32.915 -6.406 < 0.001 *** ai_Pwd - ai_clitic == 0 -114.738 41.496 -2.765 0.12138 ao_Pwd - ao_clitic == 0 7.995 55.336 0.144 1.00000 au_Pwd - au_clitic == 0 155.097 48.873 3.173 0.03936 * Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) Estimate Std. Error t value (Intercept) 1615.69 64.02 25.239 ae 460.50 55.20 8.342 ai 726.49 65.76 11.047 ao -121.36 61.42 -1.976 au -368.37 77.81 -4.734 Pwds vs. clitics -19.48 34.22 -0.569 ae: Pwd -199.90 46.43 -4.305 ai: Pwd -73.01 52.06 -1.403 ao: Pwd -22.66 63.54 -0.357 au: Pwd 172.44 60.00 2.874 335 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means: Fit: lmer(formula = F2_4 ~ vowel_actual + (1 + vowel | subj) + (1 | word), data = testunbrokenokcond) Linear Hypotheses: Estimate Std. Error z value Pr(>|z|) aa_Pwd - aa_clitic == 0 -19.48 34.22 -0.569 0.9998 ae_clitic - aa_clitic == 0 460.50 55.20 8.342 <0.01 *** ai_clitic - aa_clitic == 0 726.49 65.76 11.048 <0.01 *** ao_clitic - aa_clitic == 0 -121.37 61.41 -1.976 0.5484 au_clitic - aa_clitic == 0 -368.37 77.81 -4.734 <0.01 *** ae_Pwd - aa_Pwd == 0 260.60 55.71 4.678 <0.01 *** ai_Pwd - aa_Pwd == 0 653.48 67.90 9.624 <0.01 *** ao_Pwd - aa_Pwd == 0 -144.03 69.04 -2.086 0.4693 au_Pwd - aa_Pwd == 0 -195.93 81.16 -2.414 0.2649 ae_Pwd - ae_clitic == 0 -219.38 31.84 -6.889 <0.01 *** ai_Pwd - ai_clitic == 0 -92.49 39.23 -2.358 0.2957 ao_Pwd - ao_clitic == 0 -42.14 53.50 -0.788 0.9978 au_Pwd - au_clitic == 0 152.97 47.91 3.193 0.0370 * Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) Experiment 2 DNA vs. DAN (AC-20) Fixed effects of first interval of F1 Estimate Std. Error t value (Intercept) 682.680 23.653 28.863 DA vs. DN -31.588 15.290 -2.066 ae -69.274 17.765 -3.900 ai -83.525 15.693 -5.322 ao -45.299 17.221 -2.630 au -30.317 21.356 -1.420 ae: DN 47.221 23.700 1.992 ai: DN 4.984 22.284 0.224 ao: DN 36.692 24.144 1.520 au: DN -35.741 29.448 -1.214 336 (AC-21) Fixed effects of second interval of F1 Estimate Std. Error t value (Intercept) 703.445 23.827 29.523 DA vs. DN -40.347 20.204 -1.997 ae -105.908 23.470 -4.513 ai -138.169 20.726 -6.666 ao -90.505 22.747 -3.979 au -112.073 28.165 -3.979 ae: DN 60.242 31.304 1.924 ai: DN 17.919 29.439 0.609 ao: DN 68.370 31.877 2.145 au: DN 5.231 38.862 0.135 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means Fit: lmer(formula = F1_2 ~ adjnoun_codeV + (1 | subj), data = testcond) Linear Hypotheses: Estimate Std. Error z value Pr(>|z|) adjective_ae - adjective_aa == 0 -105.908 23.470 -4.513 <0.01 *** adjective_ai - adjective_aa == 0 -138.169 20.726 -6.666 <0.01 *** adjective_ao - adjective_aa == 0 -90.505 22.747 -3.979 <0.01 ** adjective_au - adjective_aa == 0 -112.073 28.165 -3.979 <0.01 ** noun_aa - adjective_aa == 0 -40.347 20.204 -1.997 0.5927 noun_ae - adjective_ae == 0 19.894 23.797 0.836 0.9979 noun_ai - adjective_ai == 0 -22.428 21.469 -1.045 0.9890 noun_ao - adjective_ao == 0 28.023 24.522 1.143 0.9793 noun_au - adjective_au == 0 -35.116 33.365 -1.053 0.9883 noun_ae - noun_aa == 0 -45.666 20.580 -2.219 0.4356 noun_ai - noun_aa == 0 -120.250 20.860 -5.765 <0.01 *** noun_ao - noun_aa == 0 -22.135 22.176 -0.998 0.9921 noun_au - noun_aa == 0 -106.842 26.837 -3.981 <0.01 ** Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) 337 (AC-22) Fixed effects of third interval of F1 Estimate Std. Error t value (Intercept) 686.43 20.67 33.21 DN vs. DA -36.07 22.18 -1.63 ae -98.33 25.76 -3.82 ai -190.97 22.74 -8.40 ao -118.68 24.95 -4.76 au -170.62 30.81 -5.54 ae: DN 24.98 34.34 0.73 ai: DN 32.02 32.31 0.99 ao: DN 83.44 34.94 2.39 au: DN 24.53 42.56 0.58 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means Fit: lmer(formula = F1_3 ~ adjnoun_codeV + (1 | subj), data = testcond) Linear Hypotheses: Estimate Std. Error z value Pr(>|z|) adjective_ae - adjective_aa == 0 -98.332 25.759 -3.817 <0.01 ** adjective_ai - adjective_aa == 0 -190.969 22.738 -8.399 <0.01 *** adjective_ao - adjective_aa == 0 -118.677 24.953 -4.756 <0.01 *** adjective_au - adjective_aa == 0 -170.618 30.813 -5.537 <0.01 *** noun_aa - adjective_aa == 0 -36.068 22.185 -1.626 0.8308 noun_ae - adjective_ae == 0 -11.089 26.104 -0.425 1.0000 noun_ai - adjective_ai == 0 -4.046 23.549 -0.172 1.0000 noun_ao - adjective_ao == 0 47.369 26.857 1.764 0.7517 noun_au - adjective_au == 0 -11.538 36.488 -0.316 1.0000 noun_ae - noun_aa == 0 -73.353 22.583 -3.248 0.0372 * noun_ai - noun_aa == 0 -158.947 22.910 -6.938 <0.01 *** noun_ao - noun_aa == 0 -35.240 24.304 -1.450 0.9081 noun_au - noun_aa == 0 -146.088 29.422 -4.965 <0.01 *** Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) 338 (AC-23) Fixed effects of fourth interval F1 Estimate Std. Error t value (Intercept) 653.47 22.08 29.591 DN vs. DA -36.25 25.69 -1.411 ae -103.24 29.82 -3.462 ai -191.96 26.32 -7.294 ao -116.24 28.88 -4.025 au -183.28 35.60 -5.148 ae: DN 2.91 39.75 0.073 ai: DN 47.96 37.41 1.282 ao: DN 79.80 40.41 1.974 au: DN 41.72 49.21 0.848 (AC-24) Fixed effects of first interval F2 Estimate Std. Error t value (Intercept) 1701.81 75.40 22.572 DN vs. DA -21.89 36.85 -0.594 ae 200.70 42.82 4.687 ai 212.78 37.83 5.624 ao -226.60 41.51 -5.459 au -346.01 51.52 -6.716 ae: DN -27.68 57.13 -0.485 ai: DN 73.47 53.71 1.368 ao: DN 95.00 58.21 1.632 au: DN 179.62 71.02 2.529 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means: Tukey Contrasts Fit: lmer(formula = F2_1 ~ adjnoun_codeV + (1 | subj), data = testcond) Linear Hypotheses: Estimate Std. Error z value Pr(>|z|) adjective_ae - adjective_aa == 0 200.70 42.82 4.687 < 0.001 *** adjective_ai - adjective_aa == 0 212.78 37.83 5.624 < 0.001 *** adjective_ao - adjective_aa == 0 -226.60 41.51 -5.459 < 0.001 *** adjective_au - adjective_aa == 0 -346.01 51.52 -6.716 < 0.001 *** noun_aa - adjective_aa == 0 -21.89 36.85 -0.594 0.99987 noun_ae - adjective_ae == 0 -49.57 43.44 -1.141 0.97953 noun_ai - adjective_ai == 0 51.58 39.19 1.316 0.94822 noun_ao - adjective_ao == 0 73.11 44.81 1.632 0.82742 noun_au - adjective_au == 0 157.73 61.03 2.584 0.21937 noun_ae - noun_aa == 0 173.02 37.56 4.607 < 0.001 *** noun_ai - noun_aa == 0 286.24 38.04 7.525 < 0.001 *** noun_ao - noun_aa == 0 -131.60 40.51 -3.249 0.03689 * noun_au - noun_aa == 0 -166.39 49.00 -3.396 0.02277 * 339 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) (AC-25) Fixed effects of second interval F2 Estimate Std. Error t value (Intercept) 1653.284 70.966 23.297 DN vs. DA 9.097 39.028 0.233 ae 285.796 45.348 6.302 ai 356.532 40.063 8.899 ao -332.516 43.964 -7.563 au -434.055 54.546 -7.958 ae: DN -42.552 60.502 -0.703 ai: DN 17.089 56.884 0.300 ao: DN 132.294 61.645 2.146 au: DN 181.420 75.200 2.413 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means Fit: lmer(formula = F2_2 ~ adjnoun_codeV + (1 | subj), data = testcond) Linear Hypotheses: Estimate Std. Error z value Pr(>|z|) adjective_ae - adjective_aa == 0 285.796 45.348 6.302 <0.01 *** adjective_ai - adjective_aa == 0 356.532 40.063 8.899 <0.01 *** adjective_ao - adjective_aa == 0 -332.516 43.964 -7.563 <0.01 *** adjective_au - adjective_aa == 0 -434.055 54.546 -7.958 <0.01 *** noun_aa - adjective_aa == 0 9.097 39.028 0.233 1.0000 noun_ae - adjective_ae == 0 -33.455 46.000 -0.727 0.9993 noun_ai - adjective_ai == 0 26.187 41.503 0.631 0.9998 noun_ao - adjective_ao == 0 141.391 47.445 2.980 0.0817 noun_au - adjective_au == 0 190.517 64.620 2.948 0.0892 noun_ae - noun_aa == 0 243.244 39.773 6.116 <0.01 *** noun_ai - noun_aa == 0 373.622 40.289 9.274 <0.01 *** noun_ao - noun_aa == 0 -200.222 42.892 -4.668 <0.01 *** noun_au - noun_aa == 0 -252.635 51.888 -4.869 <0.01 *** Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) 340 (AC-26) Fixed effects of third interval F2 Estimate Std. Error t value (Intercept) 1659.90 67.90 24.447 DN vs. DA -12.67 43.19 -0.293 ae 290.75 50.18 5.795 ai 474.32 44.32 10.701 ao -417.07 48.64 -8.574 au -544.69 60.32 -9.030 ae: DN 13.72 66.94 0.205 ai: DN -21.81 62.94 -0.346 ao: DN 195.71 68.20 2.870 au: DN 166.66 83.18 2.004 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means Fit: lmer(formula = F2_3 ~ adjnoun_codeV + (1 | subj), data = testcond) Linear Hypotheses: Estimate Std. Error z value Pr(>|z|) adjective_ae - adjective_aa == 0 290.750 50.176 5.795 < 0.001 *** adjective_ai - adjective_aa == 0 474.322 44.324 10.701 < 0.001 *** adjective_ao - adjective_aa == 0 -417.070 48.641 -8.574 < 0.001 *** adjective_au - adjective_aa == 0 -544.685 60.322 -9.030 < 0.001 *** noun_aa - adjective_aa == 0 -12.674 43.185 -0.293 1.00000 noun_ae - adjective_ae == 0 1.048 50.892 0.021 1.00000 noun_ai - adjective_ai == 0 -34.479 45.916 -0.751 0.99911 noun_ao - adjective_ao == 0 183.038 52.480 3.488 0.01679 * noun_au - adjective_au == 0 153.987 71.463 2.155 0.47990 noun_ae - noun_aa == 0 304.471 44.005 6.919 < 0.001 *** noun_ai - noun_aa == 0 452.516 44.581 10.150 < 0.001 *** noun_ao - noun_aa == 0 -221.358 47.448 -4.665 < 0.001 *** noun_au - noun_aa == 0 -378.025 57.404 -6.585 < 0.001 *** Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) 341 (AC-27) Fixed effects of fourth interval F2 Estimate Std. Error t value (Intercept) 1689.084 64.255 26.287 DN vs. DA -57.093 46.200 -1.236 ae 274.588 53.675 5.116 ai 484.282 47.410 10.215 ao -395.756 52.030 -7.606 au -533.723 64.491 -8.276 ae: DN 55.842 71.602 0.780 ai: DN -1.232 67.327 -0.018 ao: DN 214.745 72.936 2.944 au: DN 179.686 88.946 2.020 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means Fit: lmer(formula = F2_4 ~ adjnoun_codeV + (1 | subj), data = testcond) Linear Hypotheses: Estimate Std. Error z value Pr(>|z|) adjective_ae - adjective_aa == 0 274.588 53.675 5.116 <0.01 *** adjective_ai - adjective_aa == 0 484.282 47.410 10.215 <0.01 *** adjective_ao - adjective_aa == 0 -395.756 52.030 -7.606 <0.01 *** adjective_au - adjective_aa == 0 -533.723 64.491 -8.276 <0.01 *** noun_aa - adjective_aa == 0 -57.093 46.200 -1.236 0.9653 noun_ae - adjective_ae == 0 -1.251 54.436 -0.023 1.0000 noun_ai - adjective_ai == 0 -58.325 49.112 -1.188 0.9733 noun_ao - adjective_ao == 0 157.652 56.121 2.809 0.1287 noun_au - adjective_au == 0 122.593 76.400 1.605 0.8415 noun_ae - noun_aa == 0 330.430 47.071 7.020 <0.01 *** noun_ai - noun_aa == 0 483.050 47.696 10.128 <0.01 *** noun_ao - noun_aa == 0 -181.011 50.744 -3.567 0.0129 * noun_au - noun_aa == 0 -354.038 61.397 -5.766 <0.01 *** Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) 342 Comparison of Experiment 1 and Experiment 2 (AC-28) Fixed effects of first interval of F1: clitic, DA, DN Estimate Std. Error t value (Intercept) 682.8028 24.3544 28.036 Clitic vs. DA 0.9053 15.1476 0.060 DN vs. DA -31.2350 17.3687 -1.798 ae -71.5358 20.1178 -3.556 ai -80.7468 17.7624 -4.546 ao -45.5688 19.4403 -2.344 au -33.0852 23.8437 -1.388 Clitic:ae 20.5856 24.0035 0.858 Noun:ae 51.5274 26.7986 1.923 Clitic:ai -6.9256 22.2365 -0.311 Noun:ai 1.9336 25.2723 0.077 Clitic:ao -21.0786 23.9104 -0.882 Noun:ao 42.4279 27.1561 1.562 Clitic:au -64.3896 27.5512 -2.337 Noun:au -33.6876 33.0111 -1.020 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means: Tukey Contrasts Fit: lmer(formula = F1_1 ~ vowel_adjnoun + (1 | subj), data = testcond) Linear Hypotheses: Estimate Std. Error z value Pr(>|z|) aa_clitic - aa_adjective == 0 0.9053 15.1476 0.060 1.0000 aa_noun - aa_adjective == 0 -31.2350 17.3687 -1.798 0.8921 ae_adjective - aa_adjective -71.5358 20.1178 -3.556 0.0270 * ai_adjective - aa_adjective -80.7468 17.7624 -4.546 <0.01 *** ao_adjective - aa_adjective -45.5688 19.4403 -2.344 0.5403 au_adjective - aa_adjective -33.0852 23.8437 -1.388 0.9871 aa_noun - aa_clitic -32.1404 15.8226 -2.031 0.7673 ae_clitic - aa_clitic == 0 -50.9503 13.0917 -3.892 <0.01 ** ai_clitic - aa_clitic == 0 -87.6724 13.3753 -6.555 <0.01 *** ao_clitic - aa_clitic == 0 -66.6474 13.9206 -4.788 <0.01 *** au_clitic - aa_clitic == 0 -97.4748 13.9407 -6.992 <0.01 *** ae_noun - aa_noun == 0 -20.0084 17.6534 -1.133 0.9983 ai_noun - aa_noun == 0 -78.8132 17.9555 -4.389 <0.01 *** ao_noun - aa_noun == 0 -3.1409 18.9000 -0.166 1.0000 au_noun - aa_noun == 0 -66.7727 22.8580 -2.921 0.1759 ae_clitic - ae_adjective == 0 21.4909 18.6666 1.151 0.9981 ae_noun - ae_adjective == 0 20.2924 20.3654 0.996 0.9996 ae_noun - ae_clitic == 0 -1.1985 15.2797 -0.078 1.0000 ai_clitic - ai_adjective == 0 -6.0203 16.3182 -0.369 1.0000 343 ai_noun - ai_adjective == 0 -29.3014 18.3783 -1.594 0.9563 ai_noun - ai_clitic == 0 -23.2811 15.9785 -1.457 0.9801 ao_clitic - ao_adjective == 0 -20.1733 18.5965 -1.085 0.9990 ao_noun - ao_adjective == 0 11.1928 20.8210 0.538 1.0000 ao_noun - ao_clitic == 0 31.3661 17.2851 1.815 0.8849 au_clitic - au_adjective == 0 -63.4843 23.0488 -2.754 0.2594 au_noun - au_adjective == 0 -64.9226 28.1432 -2.307 0.5683 au_noun - au_clitic == 0 -1.4383 21.5546 -0.067 1.0000 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) (AC-29) Fixed effects of second interval of F1: clitic, DA, DN Estimate Std. Error t value (Intercept) 702.914 23.474 29.944 Clitic vs. DA 2.151 17.112 0.126 DN vs. DA -38.904 19.622 -1.983 ae -106.713 22.728 -4.695 ai -135.839 20.067 -6.769 ao -89.253 21.961 -4.064 au -114.857 26.935 -4.264 Clitic:ae 17.081 27.118 0.630 Noun:ae 62.268 30.275 2.057 Clitic:ai -24.522 25.122 -0.976 Noun:ai 15.086 28.551 0.528 Clitic:ao -27.392 27.011 -1.014 Noun:ao 71.665 30.677 2.336 Clitic:au -59.231 31.124 -1.903 Noun:au 9.462 37.291 0.254 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means: Tukey Contrasts Fit: lmer(formula = F1_2 ~ vowel_adjnoun + (1 | subj), data = testcond) Linear Hypotheses: 344 Estimate Std. Error z value Pr(>|z|) aa_clitic - aa_adjective == 0 2.1506 17.1120 0.126 1.0000 aa_noun - aa_adjective == 0 -38.9044 19.6223 -1.983 0.7970 ae_adjective - aa_adjective -106.7134 22.7276 -4.695 <0.01 *** ai_adjective - aa_adjective -135.8391 20.0669 -6.769 <0.01 *** ao_adjective - aa_adjective -89.2526 21.9614 -4.064 <0.01 ** au_adjective - aa_adjective -114.8566 26.9348 -4.264 <0.01 ** aa_noun - aa_clitic -41.0549 17.8749 -2.297 0.5765 ae_clitic - aa_clitic == 0 -89.6324 14.7905 -6.060 <0.01 *** ai_clitic - aa_clitic == 0 -160.3614 15.1108 -10.612 <0.01 *** ao_clitic - aa_clitic == 0 -116.6445 15.7263 -7.417 <0.01 *** au_clitic - aa_clitic == 0 -174.0878 15.7488 -11.054 <0.01 *** ae_noun - aa_noun == 0 -44.4450 19.9439 -2.229 0.6282 ai_noun - aa_noun == 0 -120.7528 20.2856 -5.953 <0.01 *** ao_noun - aa_noun == 0 -17.5874 21.3505 -0.824 1.0000 au_noun - aa_noun == 0 -105.3946 25.8209 -4.082 <0.01 ** ae_clitic - ae_adjective == 0 19.2316 21.0879 0.912 0.9999 ae_noun - ae_adjective == 0 23.3640 23.0071 1.016 0.9995 ae_noun - ae_clitic == 0 4.1325 17.2626 0.239 1.0000 ai_clitic - ai_adjective == 0 -22.3717 18.4356 -1.214 0.9967 ai_noun - ai_adjective == 0 -23.8180 20.7625 -1.147 0.9981 ai_noun - ai_clitic == 0 -1.4464 18.0506 -0.080 1.0000 ao_clitic - ao_adjective == 0 -25.2413 21.0066 -1.202 0.9970 ao_noun - ao_adjective == 0 32.7608 23.5193 1.393 0.9869 ao_noun - ao_clitic == 0 58.0021 19.5267 2.970 0.1567 au_clitic - au_adjective == 0 -57.0807 26.0376 -2.192 0.6554 au_noun - au_adjective == 0 -29.4424 31.7892 -0.926 0.9998 au_noun - au_clitic == 0 27.6383 24.3483 1.135 0.9984 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) 345 (AC-30) Fixed effects of third interval of F1: clitic, DA, DN Estimate Std. Error t value (Intercept) 685.155 22.320 30.697 Clitic vs. DA 2.118 19.371 0.109 DN vs. DA -35.104 22.215 -1.580 ae -97.864 25.729 -3.804 ai -188.299 22.718 -8.289 ao -117.066 24.860 -4.709 au -172.250 30.488 -5.650 Clitic:ae -20.162 30.700 -0.657 Noun:ae 25.441 34.273 0.742 Clitic:ai -28.355 28.441 -0.997 Noun:ai 28.852 32.324 0.893 Clitic:ao -21.543 30.577 -0.705 Noun:ao 89.480 34.724 2.577 Clitic:au -20.640 35.233 -0.586 Noun:au 33.345 42.209 0.790 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means Fit: lmer(formula = F1_3 ~ vowel_adjnoun + (1 | subj), data = testcond) Linear Hypotheses: Estimate Std. Error z value Pr(>|z|) aa_clitic - aa_adjective == 0 2.118 19.371 0.109 1.0000 aa_noun - aa_adjective == 0 -35.104 22.215 -1.580 0.9599 ae_adjective - aa_adjective -97.864 25.729 -3.804 0.0112 * ai_adjective - aa_adjective -188.299 22.718 -8.289 <0.01 *** ao_adjective - aa_adjective -117.066 24.860 -4.709 <0.01 *** au_adjective - aa_adjective -172.250 30.488 -5.650 <0.01 *** aa_noun - aa_clitic -37.222 20.235 -1.839 0.8736 ae_clitic - aa_clitic == 0 -118.025 16.745 -7.048 <0.01 *** ai_clitic - aa_clitic == 0 -216.654 17.108 -12.664 <0.01 *** ao_clitic - aa_clitic == 0 -138.609 17.803 -7.786 <0.01 *** au_clitic - aa_clitic == 0 -192.890 17.828 -10.820 <0.01 *** ae_noun - aa_noun == 0 -72.423 22.579 -3.208 0.0820 ai_noun - aa_noun == 0 -159.447 22.967 -6.942 <0.01 *** ao_noun - aa_noun == 0 -27.586 24.168 -1.141 0.9982 au_noun - aa_noun == 0 -138.906 29.226 -4.753 <0.01 *** ae_clitic - ae_adjective == 0 -18.043 23.873 -0.756 1.0000 ae_noun - ae_adjective == 0 -9.663 26.045 -0.371 1.0000 ae_noun - ae_clitic == 0 8.381 19.544 0.429 1.0000 ai_clitic - ai_adjective == 0 -26.236 20.872 -1.257 0.9951 ai_noun - ai_adjective == 0 -6.252 23.505 -0.266 1.0000 ai_noun - ai_clitic == 0 19.984 20.433 0.978 0.9997 346 ao_clitic - ao_adjective == 0 -19.424 23.776 -0.817 1.0000 ao_noun - ao_adjective == 0 54.377 26.620 2.043 0.7593 ao_noun - ao_clitic == 0 73.801 22.104 3.339 0.0559 au_clitic - au_adjective == 0 -18.521 29.474 -0.628 1.0000 au_noun - au_adjective == 0 -1.759 35.977 -0.049 1.0000 au_noun - au_clitic == 0 16.762 27.558 0.608 1.0000 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) (AC-31) Fixed effects of fourth interval of F1: clitic, DA, DN Estimate Std. Error t value (Intercept) 650.7993 26.3116 24.734 Clitic vs. DA -11.4976 24.2762 -0.474 DN vs. DA -34.2705 27.8433 -1.231 ae -99.3665 32.2467 -3.081 ai -189.5438 28.4729 -6.657 ao -111.0614 31.1555 -3.565 au -184.0739 38.2065 -4.818 Clitic:ae -26.7660 38.4766 -0.696 Noun:ae -0.6201 42.9543 -0.014 Clitic:ai 8.4243 35.6458 0.236 Noun:ai 44.8544 40.5126 1.107 Clitic:ao -9.0579 38.3212 -0.236 Noun:ao 85.8626 43.5153 1.973 Clitic:au 36.5774 44.1554 0.828 Noun:au 58.3116 52.8955 1.102 347 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means: Tukey Contrasts Fit: lmer(formula = F1_4 ~ vowel_adjnoun + (1 | subj), data = testcond) Linear Hypotheses: Estimate Std. Error z value Pr(>|z|) aa_clitic - aa_adjective == 0 -11.498 24.276 -0.474 1.0000 aa_noun - aa_adjective == 0 -34.270 27.843 -1.231 0.9961 ae_adjective - aa_adjective -99.366 32.247 -3.081 0.1160 ai_adjective - aa_adjective -189.544 28.473 -6.657 <0.01 *** ao_adjective - aa_adjective -111.061 31.156 -3.565 0.0258 * au_adjective - aa_adjective -184.074 38.206 -4.818 <0.01 *** aa_noun - aa_clitic -22.773 25.360 -0.898 0.9999 ae_clitic - aa_clitic == 0 -126.133 20.987 -6.010 <0.01 *** ai_clitic - aa_clitic == 0 -181.119 21.442 -8.447 <0.01 *** ao_clitic - aa_clitic == 0 -120.119 22.312 -5.384 <0.01 *** au_clitic - aa_clitic == 0 -147.496 22.343 -6.601 <0.01 *** ae_noun - aa_noun == 0 -99.987 28.299 -3.533 0.0296 * ai_noun - aa_noun == 0 -144.689 28.786 -5.026 <0.01 *** ao_noun - aa_noun == 0 -25.199 30.287 -0.832 1.0000 au_noun - aa_noun == 0 -125.762 36.624 -3.434 0.0400 * ae_clitic - ae_adjective == 0 -38.264 29.919 -1.279 0.9942 ae_noun - ae_adjective == 0 -34.891 32.642 -1.069 0.9991 ae_noun - ae_clitic == 0 3.373 24.496 0.138 1.0000 ai_clitic - ai_adjective == 0 -3.073 26.160 -0.117 1.0000 ai_noun - ai_adjective == 0 10.584 29.459 0.359 1.0000 ai_noun - ai_clitic == 0 13.657 25.607 0.533 1.0000 ao_clitic - ao_adjective == 0 -20.556 29.794 -0.690 1.0000 ao_noun - ao_adjective == 0 51.592 33.358 1.547 0.9663 ao_noun - ao_clitic == 0 72.148 27.702 2.604 0.3511 au_clitic - au_adjective == 0 25.080 36.937 0.679 1.0000 au_noun - au_adjective == 0 24.041 45.081 0.533 1.0000 au_noun - au_clitic == 0 -1.039 34.534 -0.030 1.0000 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) 348 (AC-32) Fixed effects of first interval of F2: clitic, DA, DN Estimate Std. Error t value (Intercept) 1699.57 76.05 22.347 Clitic vs. DA -46.60 47.71 -0.977 DN vs. DA -22.54 54.70 -0.412 ae 206.93 63.36 3.266 ai 209.09 55.94 3.738 ao -220.18 61.23 -3.596 au -334.83 75.09 -4.459 Clitic:ae 41.38 75.60 0.547 Noun:ae -38.60 84.40 -0.457 Clitic:ai 122.21 70.03 1.745 Noun:ai 75.28 79.59 0.946 Clitic:ao 20.94 75.30 0.278 Noun:ao 81.58 85.53 0.954 Clitic:au 257.98 86.77 2.973 Noun:au 176.54 103.97 1.698 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means: Tukey Contrasts Fit: lmer(formula = F2_1 ~ vowel_adjnoun + (1 | subj), data = testcond) Linear Hypotheses: Estimate Std. Error z value Pr(>|z|) aa_clitic - aa_adjective == 0 -46.601 47.706 -0.977 0.9997 aa_noun - aa_adjective == 0 -22.536 54.701 -0.412 1.0000 ae_adjective - aa_adjective 206.933 63.359 3.266 0.0694 ai_adjective - aa_adjective 209.088 55.941 3.738 0.0151 * ao_adjective - aa_adjective -220.181 61.226 -3.596 0.0235 * au_adjective - aa_adjective -334.835 75.093 -4.459 <0.01 *** aa_noun - aa_clitic 24.066 49.832 0.483 1.0000 ae_clitic - aa_clitic == 0 248.313 41.231 6.022 <0.01 *** ai_clitic - aa_clitic == 0 331.302 42.124 7.865 <0.01 *** ao_clitic - aa_clitic == 0 -199.237 43.842 -4.544 <0.01 *** au_clitic - aa_clitic == 0 -76.857 43.905 -1.751 0.9105 ae_noun - aa_noun == 0 168.335 55.598 3.028 0.1348 ai_noun - aa_noun == 0 284.364 56.549 5.029 <0.01 *** ao_noun - aa_noun == 0 -138.599 59.524 -2.328 0.5516 au_noun - aa_noun == 0 -158.290 71.989 -2.199 0.6506 ae_clitic - ae_adjective == 0 -5.221 58.789 -0.089 1.0000 ae_noun - ae_adjective == 0 -61.134 64.139 -0.953 0.9998 ae_noun - ae_clitic == 0 -55.913 48.122 -1.162 0.9979 ai_clitic - ai_adjective == 0 75.613 51.393 1.471 0.9784 ai_noun - ai_adjective == 0 52.741 57.881 0.911 0.9999 ai_noun - ai_clitic == 0 -22.872 50.323 -0.455 1.0000 349 ao_clitic - ao_adjective == 0 -25.657 58.568 -0.438 1.0000 ao_noun - ao_adjective == 0 59.047 65.573 0.900 0.9999 ao_noun - ao_clitic == 0 84.703 54.438 1.556 0.9647 au_clitic - au_adjective == 0 211.377 72.590 2.912 0.1801 au_noun - au_adjective == 0 154.009 88.634 1.738 0.9154 au_noun - au_clitic == 0 -57.368 67.884 -0.845 0.9999 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) (AC-34) Fixed effects of second interval of F2: clitic, DA, DN Estimate Std. Error t value (Intercept) 1646.653 78.022 21.105 Clitic vs. DA -78.656 53.494 -1.470 DN vs. DA 7.832 61.340 0.128 ae 292.817 71.048 4.121 ai 358.107 62.730 5.709 ao -332.253 68.654 -4.840 au -429.045 84.202 -5.095 Clitic:ae 91.615 84.771 1.081 Noun:ae -52.532 94.641 -0.555 Clitic:ai 208.597 78.531 2.656 Noun:ai 12.413 89.252 0.139 Clitic:ao 115.025 84.440 1.362 Noun:ao 139.965 95.900 1.459 Clitic:au 230.756 97.298 2.372 Noun:au 200.181 116.576 1.717 350 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means: Tukey Contrasts Fit: lmer(formula = F2_2 ~ vowel_adjnoun + (1 | subj), data = testcond) Linear Hypotheses: Estimate Std. Error z value Pr(>|z|) aa_clitic - aa_adjective == 0 -78.656 53.494 -1.470 0.978 aa_noun - aa_adjective == 0 7.832 61.340 0.128 1.000 ae_adjective - aa_adjective 292.817 71.048 4.121 <0.01 ** ai_adjective - aa_adjective 358.107 62.730 5.709 <0.01 *** ao_adjective - aa_adjective -332.253 68.654 -4.840 <0.01 *** au_adjective - aa_adjective -429.045 84.202 -5.095 <0.01 *** aa_noun - aa_clitic 86.488 55.878 1.548 0.966 ae_clitic - aa_clitic == 0 384.432 46.235 8.315 <0.01 *** ai_clitic - aa_clitic == 0 566.704 47.237 11.997 <0.01 *** ao_clitic - aa_clitic == 0 -217.228 49.162 -4.419 <0.01 *** au_clitic - aa_clitic == 0 -198.289 49.232 -4.028 <0.01 ** ae_noun - aa_noun == 0 240.285 62.345 3.854 <0.01 ** ai_noun - aa_noun == 0 370.520 63.413 5.843 <0.01 *** ao_noun - aa_noun == 0 -192.288 66.745 -2.881 0.194 au_noun - aa_noun == 0 -228.863 80.721 -2.835 0.216 ae_clitic - ae_adjective == 0 12.959 65.922 0.197 1.000 ae_noun - ae_adjective == 0 -44.700 71.922 -0.622 1.000 ae_noun - ae_clitic == 0 -57.659 53.963 -1.068 0.999 ai_clitic - ai_adjective == 0 129.941 57.630 2.255 0.608 ai_noun - ai_adjective == 0 20.244 64.905 0.312 1.000 ai_noun - ai_clitic == 0 -109.696 56.428 -1.944 0.820 ao_clitic - ao_adjective == 0 36.369 65.671 0.554 1.000 ao_noun - ao_adjective == 0 147.796 73.526 2.010 0.781 ao_noun - ao_clitic == 0 111.427 61.042 1.825 0.880 au_clitic - au_adjective == 0 152.099 81.397 1.869 0.861 au_noun - au_adjective == 0 208.013 99.382 2.093 0.726 au_noun - au_clitic == 0 55.914 76.118 0.735 1.000 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) 351 (AC-35) Fixed effects of third interval F2: clitic, DA, DN Fixed effects: Estimate Std. Error t value (Intercept) 1655.280 78.067 21.203 Clitic vs. DA -86.222 56.708 -1.520 DN vs. DA -17.682 65.027 -0.272 ae 291.290 75.317 3.868 ai 476.181 66.500 7.161 ao -426.029 72.778 -5.854 au -543.272 89.260 -6.086 Clitic:ae 168.903 89.865 1.880 Noun:ae 9.611 100.328 0.096 Clitic:ai 222.960 83.251 2.678 Noun:ai -23.719 94.617 -0.251 Clitic:ao 199.913 89.514 2.233 Noun:ao 217.513 101.660 2.140 Clitic:au 255.949 103.143 2.481 Noun:au 198.276 123.578 1.604 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means: Tukey Contrasts Fit: lmer(formula = F2_3 ~ vowel_adjnoun + (1 | subj), data = testcond) Linear Hypotheses: Estimate Std. Error z value Pr(>|z|) aa_clitic - aa_adjective == 0 -86.222 56.708 -1.520 0.9708 aa_noun - aa_adjective == 0 -17.682 65.027 -0.272 1.0000 ae_adjective - aa_adjective 291.290 75.317 3.868 <0.01 ** ai_adjective - aa_adjective 476.181 66.500 7.161 <0.01 *** ao_adjective - aa_adjective -426.029 72.778 -5.854 <0.01 *** au_adjective - aa_adjective -543.272 89.260 -6.086 <0.01 *** aa_noun - aa_clitic 68.540 59.236 1.157 0.9980 ae_clitic - aa_clitic == 0 460.193 49.014 9.389 <0.01 *** ai_clitic - aa_clitic == 0 699.141 50.076 13.962 <0.01 *** ao_clitic - aa_clitic == 0 -226.116 52.116 -4.339 <0.01 ** au_clitic - aa_clitic == 0 -287.323 52.190 -5.505 <0.01 *** ae_noun - aa_noun == 0 300.901 66.092 4.553 <0.01 *** ai_noun - aa_noun == 0 452.463 67.225 6.731 <0.01 *** ao_noun - aa_noun == 0 -208.516 70.754 -2.947 0.1648 au_noun - aa_noun == 0 -344.996 85.568 -4.032 <0.01 ** ae_clitic - ae_adjective == 0 82.681 69.884 1.183 0.9974 ae_noun - ae_adjective == 0 -8.071 76.243 -0.106 1.0000 ae_noun - ae_clitic == 0 -90.752 57.207 -1.586 0.9586 ai_clitic - ai_adjective == 0 136.738 61.094 2.238 0.6211 ai_noun - ai_adjective == 0 -41.400 68.805 -0.602 1.0000 352 ai_noun - ai_clitic == 0 -178.138 59.818 -2.978 0.1517 ao_clitic - ao_adjective == 0 113.691 69.614 1.633 0.9473 ao_noun - ao_adjective == 0 199.832 77.941 2.564 0.3783 ao_noun - ao_clitic == 0 86.141 64.710 1.331 0.9914 au_clitic - au_adjective == 0 169.727 86.286 1.967 0.8076 au_noun - au_adjective == 0 180.595 105.347 1.714 0.9232 au_noun - au_clitic == 0 10.868 80.688 0.135 1.0000 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method) (AC-36) Fixed effects of fourth interval F2: clitic, DA, DN Estimate Std. Error t value (Intercept) 1689.455 76.674 22.034 Clitic vs. DA -114.662 58.336 -1.966 DN vs. DA -66.258 66.895 -0.990 ae 268.701 77.480 3.468 ai 483.178 68.410 7.063 ao -414.995 74.867 -5.543 au -539.351 91.820 -5.874 Clitic:ae 219.896 92.446 2.379 Noun:ae 57.698 103.210 0.559 Clitic:ai 269.131 85.643 3.142 Noun:ai 2.917 97.335 0.030 Clitic:ao 330.247 92.083 3.586 Noun:ao 244.156 104.577 2.335 Clitic:au 329.612 106.104 3.106 Noun:au 214.868 127.123 1.690 Simultaneous Tests for General Linear Hypotheses Multiple Comparisons of Means: Tukey Contrasts Fit: lmer(formula = F2_4 ~ vowel_adjnoun + (1 | subj), data = testcond) Linear Hypotheses: 353 Estimate Std. Error z value Pr(>|z|) aa_clitic - aa_adjective == 0 -114.662 58.336 -1.966 0.8063 aa_noun - aa_adjective == 0 -66.258 66.895 -0.990 0.9996 ae_adjective - aa_adjective 268.701 77.480 3.468 0.0366 * ai_adjective - aa_adjective 483.178 68.410 7.063 <0.01 *** ao_adjective - aa_adjective -414.995 74.867 -5.543 <0.01 *** au_adjective - aa_adjective -539.351 91.820 -5.874 <0.01 *** aa_noun - aa_clitic 48.404 60.937 0.794 1.0000 ae_clitic - aa_clitic == 0 488.597 50.423 9.690 <0.01 *** ai_clitic - aa_clitic == 0 752.309 51.515 14.604 <0.01 *** ao_clitic - aa_clitic == 0 -84.748 53.612 -1.581 0.9598 au_clitic - aa_clitic == 0 -209.739 53.688 -3.907 <0.01 ** ae_noun - aa_noun == 0 326.398 67.991 4.801 <0.01 *** ai_noun - aa_noun == 0 486.096 69.157 7.029 <0.01 *** ao_noun - aa_noun == 0 -170.839 72.784 -2.347 0.5379 au_noun - aa_noun == 0 -324.483 88.022 -3.686 0.0175 * ae_clitic - ae_adjective == 0 105.234 71.890 1.464 0.9791 ae_noun - ae_adjective == 0 -8.561 78.433 -0.109 1.0000 ae_noun - ae_clitic == 0 -113.794 58.851 -1.934 0.8247 ai_clitic - ai_adjective == 0 154.469 62.849 2.458 0.4539 ai_noun - ai_adjective == 0 -63.341 70.781 -0.895 0.9999 ai_noun - ai_clitic == 0 -217.809 61.535 -3.540 0.0286 * ao_clitic - ao_adjective == 0 215.585 71.610 3.011 0.1411 ao_noun - ao_adjective == 0 177.898 80.176 2.219 0.6358 ao_noun - ao_clitic == 0 -37.687 66.567 -0.566 1.0000 au_clitic - au_adjective == 0 214.950 88.763 2.422 0.4808 au_noun - au_adjective == 0 148.610 108.366 1.371 0.9886 au_noun - au_clitic == 0 -66.340 83.002 -0.799 1.0000 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Adjusted p values reported -- single-step method)
Abstract (if available)
Abstract
The main theoretical goal of this dissertation is the investigation of the syntax-phonology interface through probing syntax-sensitive phonological phenomena, exemplified by a comprehensive analysis of the so-called Spanish “feminine el”, a morpho-phonological determiner alternation that is triggered by certain phonological and syntactic contexts. Specific areas of the syntax-phonology interface that I examine are syntactic triggers to prosodic clitic organization and phonetic reflections of those prosodic structures. I argue that the feminine el provides support for the inclusion of the syntactic head-complement relation in the Prosodic Phonology mapping, and produces recursive Prosodic Word (Pwd) cliticization structures. In particular, it accrues evidence for the maximal Pwd domain proposed by Itô and Mester (2006, 2007, 2008, 2010). This dissertation provides new phonetic evidence for distinguishing between maximal and minimal Pwd boundaries in different cliticization structures. ❧ Additional theoretical proposals of the dissertation include a scale of phonological contrast-driven hiatus restrictions, and the development of a morphology-phonology correspondence account of apparent mismatches in gender realization at the morphology-phonology interface, in order to address the specific feminine or masculine morphological interpretation of the Spanish article in question. The proposed scale of hiatus restrictions is a novel application of the Minimum Distance (Flemming 2004) approach to perceptual contrast, utilizing perceptual contrast in segment transitions to produce a relative scale of adjacent VV restrictions. The morphology-phonology correspondence account provided builds on previous work in the morphology-phonology interface (Walker & Feng 2004, Wolf 2008) to evolve the family of IDENTITY constraints governing correspondence between morphological and phonological forms in the input and output. ❧ In examining these areas, this dissertation is the first to provide a full theoretical analysis of the Spanish feminine el phenomenon, accounting for phonological, morphological and syntactic effects. Further case studies in Spanish and Romance dialects also demonstrate the effects of the prosodic cliticization structures, hiatus restrictions, and morpho-phonological correspondence proposed. Among the broad theoretical contributions of this work, this dissertation brings new insight and evidence to the distinctions between different prosodic clitic attachments and the nature of the syntactic information on which they may rely, further developing the relationship between syntax and phonology. It additionally expands the understanding of phonological vowel hiatus by building on perceptual contrast considerations, opening a new perspective on segment sequencing.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Minimal contrast and the phonology-phonetics interaction
PDF
Syntax-prosody interactions in the clausal domain: head movement and coalescence
PDF
Prosody in contact: Spanish in Los Angeles
PDF
The phonological dimension of grammatical markedness
PDF
Prosodic recursion and syntactic cyclicity inside the word
PDF
Tone gestures and constraint interaction in Sierra Juarez Zapotec
PDF
Generalized surface correspondence in reduplicative opacity
PDF
Sources of non-conformity in phonology: variation and exceptionality in Modern Hebrew spirantization
PDF
A reduplicative analysis of sentence modal adverbs in Spanish
PDF
The locative alternation: unaccusative constructions and subject position
PDF
Agreement on the left edge: the syntax of left dislocation in Spanish
PDF
Individual differences in phonetic variability and phonological representation
PDF
Sound sequence adaptation in loanword phonology
PDF
Harmony in gestural phonology
PDF
Adding and subtracting alternation: resumption and prepositional phrase chopping in Spanish relative clauses
PDF
Soft biases in phonology: learnability meets grammar
PDF
Beatboxing phonology
PDF
Copy theory of movement and PF conditions on spell-out
PDF
Articulatory dynamics and stability in multi-gesture complexes
PDF
A unified syntactic account of Mandarin subject nominals
Asset Metadata
Creator
Varis, Erika Elizabeth
(author)
Core Title
The Spanish feminine el at the syntax-phonology interface
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Linguistics (Hispanic Linguistics)
Publication Date
08/01/2012
Defense Date
08/01/2012
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
clitics,feminine el,interface,OAI-PMH Harvest,phonology,prosodic phonology,Spanish,syntax
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Walker, Rachel (
committee chair
), Goldstein, Louis (
committee member
), Saltarelli, Mario (
committee member
)
Creator Email
erikavaris@gmail.com
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-83338
Unique identifier
UC11290110
Identifier
usctheses-c3-83338 (legacy record id)
Legacy Identifier
etd-VarisErika-1104.pdf
Dmrecord
83338
Document Type
Dissertation
Rights
Varis, Erika Elizabeth
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
clitics
feminine el
interface
phonology
prosodic phonology
syntax