The Syllable and Stress: Studies in Honor of James W. Harris 9781614515975, 9781614517368

Citation preview

Rafael A. Núñez Cedeño The Syllable and Stress

Studies in Generative Grammar

Editors Norbert Corver Harry van der Hulst Roumyana Pancheva Founding editors Jan Koster Henk van Riemsdijk

Volume 126

The Syllable and Stress

Studies in Honor of James W. Harris

Edited by Rafael A. Núñez Cedeño

ISBN 978-1-61451-736-8 e-ISBN (PDF) 978-1-61451-597-5 e-ISBN (EPUB) 978-1-5015-0049-7 ISSN 0167-4331 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available on the Internet at http://dnb.dnb.de. 6 2016 Walter de Gruyter Inc., Boston/Berlin Typesetting: RoyalStandard, Hong Kong Printing and binding: CPI books GmbH, Leck ♾ Printed on acid-free paper Printed in Germany www.degruyter.com

Contents Foreword ix Tabula gratulatoria

I

xviii

The Syllable and Glides

Laura Colantoni and José I. Hualde Constraints on front mid-vowel gliding in Spanish

1

Ellen M. Kaisse The syllabic position of glides in Spanish: insights from Pasiego vowel 29 harmony Iggy Roca Gliding ghosts or ghostly glides, and does it matter which?

51

II The Syllable – Onset, Coda, and Cross Linguistic Aspects Sonia Colina On onset clusters in Spanish: voiced onset underspecification and /f/ Fernando Martínez-Gil Syllable merger in Chicano Spanish: a constraint-based analysis

107

139

Conxita Lleó The challenge of lexically empty onsets in first language acquisition of Spanish 187 and German Rafael A. Núñez Cedeño The acquisition order of Spanish liquids in complex onsets

219

III The Syllable and the Notions of Stress and Weight John M. Lipski Spanish vocalic epenthesis: the phonetics of the mora Carlos-Eduardo Piñeros The phonological weight of Spanish syllables

245

271

W. Leo Wetzels and Ben Hermans Syllable weight, stress, and the neutralization of vowel length in Batticaloa 315 Creole Portuguese Subject index 333 Index of constraints

339

Foreword This volume is published in honor of James W. Harris’ distinguished career in general and Romance linguistics and is dedicated to him and to his enduring and enriching contribution to the theoretical research of phonology and morphology, especially of Hispanic linguistics. Retired from the Massachusetts Institute of Technology, Jim, as we fondly call him, has been and still is internationally recognized as one of the highly influential and leading figures in the area of Romance linguistics. A testament to his accomplishments and impact is found in two of his influential books, Spanish phonology (1969), and Syllable structure and stress in Spanish: a nonlinear analysis (1983), which are standard references for anyone working on theories of the properties and structure of features, syllable, and stress. Jim’s 1969 volume is the first book-long, segmentalbased treatise breaking new ground in Romance with the application of Chomsky and Halle’s (1968) generative study of The Sound Pattern of English (SPE) to the phonology of the Spanish language. In his 1983 monograph, Jim again pioneered the innovative application of the nonlinear metrical theory to Hispanic phonology and morphology, analyzing and offering solutions with rich, data-based evidence from various Hispanic dialects to longstanding problems which the standard, linear SPE model simply could not resolve. Jim’s significant contributions to the field of linguistics also contain publications of numerous papers, spanning a broad spectrum of topics. These include phonological issues of diphthongization, geometric feature representations, plural formation, presence of lexical glides, multi-layered analyses of the lexicon, syllable construction, stress placement, as well as studies of morphological phenomena of cliticization in various Romance languages, and the inter-relationship of phonology with morphology and syntax. Given the breadth of Jim’s research, it became necessary to direct one’s attention to a major aspect of his academic interests, namely, the study of the central prosodic properties of syllable and stress. That this theme is and continues to be a major venue of linguistic research of Romance phonology is reflected in the articles of the present volume, which were written by notable linguists and dedicated to Jim. In this opportune collection, couched in a variety of theoretical perspectives, leading Romance phonologists concentrate on the importance of the syllable and stress in three sections of research. Part 1 focuses on the theme of the role of glides in the syllable as it pertains to questions of their morphological and phonological conditioning, syllabic positioning, and status of lexical existence. Thus, in “Conditions on front midvowel gliding in Spanish,” Laura Colantoni and José I. Hualde give special

x

Foreword

attention to the gliding process of a front mid vowel when followed by another vowel. The purpose of the study is to understand the phonological and morphological conditions which trigger the gliding of heterosyllabic vowels in words like /te.a.tɾo/ 1 ‘theater’ with resulting output [tja.tɾo] in the popular speech of Argentinian and Mexican speakers. To this end, they constructed a paper-andpencil experiment to assess these speakers’ judgments of gliding acceptability. Two of four proposed hypotheses found support. Gliding is avoided when a minimality condition is created, and when second conjugation verbs are considered. Regarding minimality, speakers reject reducing a heterosyllabic sequence such as /le.al/ ‘loyal’ to becoming *[ljal]. Similarly, though speakers accept gliding for verbs of the first conjugation as in /dese.amos/ ‘we wish’ > [deˈsjamos], derived from desear ‘to wish’; a verb with an identical sequence of vowels, such as /se.amos/ ‘let us be’, which derives from a the second conjugation verb ser ‘to be’, is judged ill-formed, *[sja.mos] while [se.a.mos] is acceptable. In all counts, the authors conclude that speakers seem to have a very solid intuition of where gliding should occur. In “The syllabic position of glides in Spanish: insights from Pasiego vowel harmony,” Kaisse revisits the topic of the position of the glide in the syllable and its vowel-raising harmony effects in Pasiego Spanish. Her syllabic-position argument is based on the theory that the [+high] feature located in the syllabic nucleus can spread to other syllabic nuclei (van der Hulst and Weijer 1995) to yield [kumiˈɾ+ia] from underlying /komeɾ +ia/. It is this head-to-head assimilation, which allows her to explain the observations made by Hualde (1989) regarding the fact that offglides cannot spread the [+high] feature because it occupies the non-nuclear coda position, as found in [koˈxajs] from /koxajs/. An offshoot of her hypothesis is that while palatal consonants are inherently [+high], they cannot trigger vowel harmony precisely because they occupy the onset and not the nuclear position. In such a context, vowel harmony is avoided, as the underlying palatal /ʝ/ in /kɾeʝemus/ produces exactly [kɾeˈʝemus] ‘we believe’, without raised /e/. Still treating glides, in an earlier study Roca (2006) proposed the hypothesis that those segments do not exist lexically in Spanish. In “Gliding ghosts or ghostly glides, and does it matter which?,” Roca raises the same problem, and with the support of solid empirical evidence and an OT theoretical machinery, he emphatically reasserts that these segments are not only redundant but also cumbersome for the grammar. Concentrating mainly on Spanish non-verbs, his 1 For all Spanish dialects, in this volume /t/ and /d/, with their respective allophones [t] and [d, ð], represent dental articulations. Elsewhere, [β, ð, ɣ] stand for the approximant variants of /b, d, ɡ/.

Foreword

xi

arguments rest on the assumption that there is a clear relationship between stress assignment and the syllabification of heterosyllabic non-high with high vowels. Pairs such as [maˈɾi.a] ‘Maria’ ~ [maˈɾi ̯o] ‘Mario’ and [sinaˈi] ‘Sinai’ ~ [bonˈsai ̯] ‘Bonsai’, which exhibit contrasting high front vowels vs. high front vowel glides, are explained away as a result of the interaction of dominant faithfulness constraint MAX-PEAK over marked constraint ONSET, where the high vowels are at the edge of a stem. The fact that /i/ in Mario and bonsái surfaces as a glide is also due to a high constraint which prohibits the presence of tautosyllabic stressless vowels. The analysis further extends to and accounts for similar sequences found word-internally. A few unmanageable cases are covered by several high-ranked markedness constraints, such as avoiding complex nucleui, having a stressed /e/ after a heterosyllabic high vowel, and forming heterosyllabic sequences after [k]. Roca concludes by reaffirming that his analysis allows him to posit a grammar without glides. Part 2 examines the topics of subsyllabic constituents of onsets and rhymes and their relationship to their internal featural representation in Spanish, resyllabification processes in resolving both vowel hiatus in Chicano Spanish and their effects in language contacts, particularly with German, and the acquisition order of liquids in complex onsets. Regarding the properties of segments in the onset, in the OT-based article “On onset clusters in Spanish: voiced obstruent underspecification and /f/” Colina deals with the issue of grouping /f/+liquids in complex onsets along with voiceless and voiced stops by positing the hypothesis that in terms of sonority, it is underspecified for the feature [continuant]. Her work takes a step further over Martínez-Gil’s (2001) earlier proposal with respect to the phonological absence of the feature [continuant] for /f/ by claiming that this underspecification makes for a better pattern with voiced /b, d, ɡ/ as well as persevering well into the output level. With this approach, she also attempts to resolve the question of why the approximant versions of the voiced stops combine freely in onsets with liquids in clear violation of the Complex Onset Condition. Colina’s study is followed by Martínez-Gil’s in-depth approach to resolving the issue of vowels in hiatus across a word boundary by merging them into one syllable through the intervention of a resyllabification process, as proposed in his article “Syllable merger in Chicano Spanish: a constraint-based analysis.” He provides an extensive corpus of basic facts culled from a variety of existing sources from which he examines all logical combinations of abutting stressed or unstressed vowels across word boundaries. These combinations allow him to identify five patterns in the sequence /. . . V1 # V2 . . . /, where V1 undergoes significant structural changes. In pattern 1, two identical vowels are reduced to one, i.e., /se # eskapó/ > [sés.ka.pó] ‘he fled’. Pattern 2 deletes the first mid

xii

Foreword

vowel when followed by a homorganic vowel, as in /me # iɾía/ [mi.ɾí.a] ‘I would go’. Pattern 3 deletes the first vowel, as in /ábla # inglés/ [á.βli.ŋɡlés] ‘he speaks English’. For patterns 4 and 5, high and mid vowels are desyllabified, becoming a glide before another vowel, as in the respective /mi#unifóɾme/ [mju.ni.fóɾ.me] ‘my uniform’, and /me # usó/ [mju.só] ‘he used me’. Martínez-Gil first considers previous parallel OT research in which V1, the derived prevocalic glide noted before, is lexically parsed as a monomoraic member of a complex nucleus during the syllabification process, which repeats itself postlexically. Issues of sonority sequencing of non-identical high vocoids, mid-vowel raising, stress shift when syllables merge as well as strengthening of prevocalic glides in absolute syllable-initial position, allow Martínez-Gil to defend the argument that glides affiliate with the onset rather than with the nucleus. The hypothesis that the first of two vowels in a heterosyllabic sequence is incorporated as the onset of the following vowel-initial word to satisfy the undominated constraint ONSET, is one of the most innovative aspects of his paper with an OT perspective. This is evident in the input sentence /sói # iɡuál/, whose output [só.ji.ɣwál] ‘I am the same’, contains an input vocoid in a coda resyllabified now as an onset. As a second step of analysis, Martínez-Gil introduces a stratal-inspired, serial OT proposal thereby avoiding the opacity problems faced by the parallel OT approach, such as wrongly predicting that a resyllabified high vocoid becomes a consonant at the phrase level, as ley antigua ‘old law’ that turns into the wrong output *le.[ ʝ]an.ti.gua. To support his stratal OT analysis, Martínez-Gil appeals to the interaction of five markedness over eight faithfulness constraints found in the postlexical stratum and ranked in a Spanish-specific hierarchy. Besides accounting for the consonantization of resyllabified vocoids word-initially, the interplay of constraints permits the selection of optimal candidates, which result from the actual occurrences of processes of coalescence and deletion of vowels, exemplified by the five patterns of mentioned previously. Martínez-Gil identifies several potential problems that could undermine his stratal position, among them, stress shift under vowel deletion, infringement of onset binarity, merger of three-vowel sequences across the word, and wordinternal coalescence. All are reasonably explained away as a result of obeying some other kinds of constraints. He concludes noting that the stratal OT solution has the benefit of resolving issues of opacity without duplicating the empirical and theoretical shortcomings of a parallel OT account. Continuing with the thread of Martínez-Gil’s elucidation of phenomena across a word boundary, Lleó takes up the acquisition of resyllabification by Spanish and German monolingual and bilingual children in her contribution “The challenge of lexically empty onsets in first language acquisition of Spanish

Foreword

xiii

and German.” She observes that while in Spanish the coda of a preceding word normally syllabifies with a following word that begins with a vowel, as in the lexical phrase /los.#o.sos/ which turns into the output [lo.s#o.sos] ‘the bears’, in German such a strategy is usually suppressed. Instead, this language tends to insert a glottal stop in this context, changing the previous underlying phrase into [los.#ˈʔo.sos]. Her scrutiny of these two processes shows that Spanish bilingual speakers begin resyllabification at a rather late stage of development, with fewer instances of this process, and tend to insert glottal stops more often when compared to monolinguals. She attributes these results to the degree of phonological influence of one language over the other. That bilinguals are inclined to add glottal stops stems from the fact that German provides more negative transfer in terms of its prosody to the weak, heritage language, Spanish. Also viewed within a developmental research framework is Núñez Cedeno’s article “The acquisition order of liquids in Spanish complex onsets.” He examines a widely known area of phonological research in L1 acquisition literature: whether there is an acquisition order between obstruent + lateral versus obstruent + rhotics clusters, a hypothesis concomitantly advanced by Kehoe et al. (2008). Based on the statistical results of a field-work experiment conducted with 12 monolingual Spanish speaking children, he demonstrates that an obstruent + lateral cluster like the initial consonantal sequence in plátano ‘plantain’ develops gradually and is acquired before the obstruent+rhotic grouping of the princesa ‘princess’ types. He further contends that this ordering is predicated on the physical nature of the sounds involved, or to be precise, that rhotics are more difficult to produce, following Kirchner’s (1998) hypothesis that they require additional articulatory gestures for their realization. Núñez Cedeño’s findings thus provide additional support to his theoretical claim that liquids should be ordered such that the markedness constraint *rhotic-ONS outranks *lateral-ONS. Lastly, Part 3 is comprised of three analyses, one of which considers the notion of the mora in Spanish, and the remaining two are about weight and stress as they apply to Spanish and Batticaloa, a Creole Portuguese language. In the first study, “Spanish vocalic epenthesis: the phonetics of sonority and the mora,” Lipski offers an OT framework for weighing in on the role of the epenthetic vowel /e/ as a strategy in repairing impermissible consonant clusters in Spanish of the /-sC/, /-mb/, and /VCL-/ types (where L = liquids), as in the respective smoking > esmóquin ‘jacket’, mbok > emboco ‘wrestler, executioner’, and Sadr > [sa.ðeɾ] ‘a family name’. His claim hinges on the assumption that because epenthetic vowels carry no phonological weight, only a stray, unsyllabified moraic consonant in a cluster provides them with moraic content. The need to effect vocalic epenthesis is driven by the relative sonority of abutting consonants. With this unique, epenthetic approach, Lipski delivers a unified account

xiv

Foreword

to the issue of syllable-repair strategies, since the total mora count remains unchanged, whereas previous inquiries which allude to concepts like appendix are totally circular. The second and the third investigations in this section expand on the notion of weight and further confront the issue of stress in Spanish and Batticaloa respectively, but offer different solutions. In his contribution “The phonological weight of Spanish syllable,” Carlos-Eduardo Piñeros firmly argues that Spanish stress is quantity insensitive. While recognizing that past analyses have achieved simpler results in assigning stress, he contends that they are burdened by the cost of employing extraneous devices, such as extrametricality and diacritic markings, to account for exceptional stress in proparoxytone and vowel-final oxytone words of the types [piˈɾamiðe] ‘pyramid’ and [makɾaˈme] ‘macrame’, and consonant-final paroxytones like [kaˈteteɾ] ‘catheter’. Instead, he proposes an OT analysis for generating stress systematically for words by assigning syllabic trochees at the right edge of the word, such as [paɾaˈlelo] ‘parallel’, which constitutes the majority of the lexicon. The remaining forms are accounted for by idiosyncratic morphemes that interact with trochees. If the morpheme is a morphological head, it attracts stress to the vowel of that domain, as in [[foɾm + a + ˈRA]v + Ø ] ‘s/he will form’ – with a covert morpheme which provides trochee status to the last syllable; if, on the contrary, the vowel of the morpheme does not support foot head, stress is repelled, in which case it is displaced one syllable to the left, such as in [ [ˈkaɾn + IK ]a + o] ‘meat related’, where /i/ is not a foot head. With this approach, Piñeros concludes that 91% of the lexicon has predictable stress in conformity to regular footing. The exceptions carry idiosyncratic diacritics in the lexicon and are the results of the interaction between phonology and morphology, in which the latter is marked with stress-repeller morphemes. The final study “Syllable weight, stress, and the neutralization of vowel length in Batticaloa Creole Portuguese,” by Wetzels and Hermans, dissects the interaction of Batticaloa’s stress assignment with vowel quantity. This variety of Dutch-based Portuguese spoken in the east coast of Sri Lanka, has been shown to display a number of interesting prosodic properties regarding its underlying and short vowels. First, stress normally falls on the underlying long vowel of a rightmost syllable, as in /mi:di:do:r/ > [midiˈdo(:)r] ‘surveyor’. If the word contains short vowels, stress is on the left, as shown by /tosa/ > [ˈtosa] ‘to cough’. However, /mi:da/ > [ˈmi:da] ‘measure’ shows that when word-initial vowels are long, stress will be located therein. Second, long vowels not receiving stress are summarily shortened, attested in /ba:rwe:ru/ > [barˈwe:ru]. And third, noninitial stressed long vowels may have variable length, depending on whether they are in closed and word- final syllables, in which case they shorten.

Foreword

xv

To account for these facts, the authors resort to an OT approach and adopt a generalized, integrated model which calls for relating formally prominence and metrical constituency. Based mostly on the observation that Batticaloa belongs to Default-to-Opposite systems, Wetzels and Hermans rely on the notion of licensing by a word-edge mora. Their analysis of stress is conceived in terms of iambic representations and starts off with the parametrized constraint ALIGN -R, which interacts with the higher ranked initial-mora licensing constraint LIC -μi, thereby predicting main stress location near the right end of the word for most cases. The interplay of these two constraints also explains the occurrence of initial stress for words not containing long vowels, as in [ˈsindəfərə] ‘Monday’ < /sindafara/. In order to explain why word-initial underlying long stressed vowels are always long while word-internal they may be optionally shortened, the authors postulate the constraint MINIMAL SIZE , which requires the main stress constituent to have a branching head. They claim that the structure is different for both representations. For the unreduced context, the long stressed vowel has two independent prominent marks while in potentially shortened contexts one prominence is dependent on the other. In short, with the integrated model that includes two different types of heavy syllables containing long vowels, Wetzels and Hermans dispense with the argument that the opaque interaction between stress and vowel reduction in Batticaloa supports certain tenets of Sympathy Theory. All these chapters bring a host of perspectives on the issue of the syllable itself as well as its interface with stress in Spanish and Portuguese. I am thus thankful to all contributors to this volume for joining me in recognizing Jim’s dedication to these topics and for their meticulous readings and anonymous suggestions of each other’s works. I am equally grateful to the following external scholars, who, by sharing their expertise and very valuable time in reading and evaluating anonymously all contributions, made this anthology not only possible but stronger in content and argumentations. Ordered alphabetically they are: Eric Baković, Jessica Barlow, Ricardo Bermúdez-Otero, Eulàlia Bonet, Travis Bradley, Jennifer Cabrelli, Andrea Calabrese, John Grindstead, Jorge M. Guitart, Nicholas Henriksen, Eric Holt, Joan Mascaró, Alfonso Morales-Front, Lori Repetti, and Donca Steriade. The edition of any publication normally involves people other than the writer/editor. In terms of presentation, formatting, and proof-reading, I am indebted to James Compton, Bernard Issa, and E. Wyn Roberts. Also invaluable are Emily Farrell’s, Wolfgang Konwitschny’s and Lara Wysong’s punctilious editorial assistance at Mouton de Gruyter, Harry van der Hulst’s support, whose

xvi

Foreword

enthusiastic recommendations moved the editorial board to embrace the project, and W. Leo Wetzels, who proposed this work to Harry for his consideration. Last, but definitely not least, a group of notable linguists and friends were contacted to join all contributors in recognizing the profound impact Jim has made both to general linguistic theory and to Romance linguistics. In fact, some of them went even further in conveying deeply felt thoughts about Jim, which those listed on the Tabula gratulatoria below also appreciatively endorse. John Goldsmith states that “With his passion and dedication in all things related to the linguistics of the Spanish language, Jim Harris has guided the work of a generation and more of linguistic scholarship.” Bruce Hayes reminisces about Jim and wishes him “happy memories of hashing everything out in the early days of metrical stress theory.” Nicholas Henriksen goes back to his formative years in graduate school and says he spent many graduate student days reading Jim’s papers and books and, while not having met Jim personally, adds that it’s “only fair to state that many of my own positions on phonology are influenced by Harris’ work and thinking.” Michael Kenstowicz believes Jim to be “one of the leading exponents of generative grammar specializing in Hispanic languages; his research showed how many properties of these languages shed light on questions of general interest as well as how the generative approach elucidated many puzzles in the phonology and morphology of these languages. I have fond memories of the summer we spent together teaching in Girona in 1992.” John J. McCarthy, touched by Jim’s influence, further adds that “I am very grateful to Jim for teaching me, by words and example, that the data always repay close analysis.” Jorge M. Guitart relates similar sentiments when he says “Four decades later Jim Harris keeps on making me look deeper. Thank you, beloved friend and mentor.” Finally, Frank Nuessel sums up a shared feeling by all that Jim’s work on Spanish phonology “continues to be groundbreaking. His work continues to be cited in all of the work in this area of research. We thank him for his insights, precision and clarity.”

References Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Harris, James W. 1969. Spanish phonology. Cambridge, MA: The MIT Press. Harris, James W. 1983. Syllable structure and stress in Spanish: a nonlinear analysis. Cambridge, MA & London, England: The MIT Press. Hualde, José I. 1989. Autosegmental and metrical spreading in the vowel-harmony systems of northwestern Spain. Linguistics 27. 773–805. Hulst, Harry van der & Jeroen van der Weijer. 1995. Vowel harmony. In John Goldsmith (ed.), Handbook of phonological theory. 495–534. Cambridge, MA: Blackwell.

Foreword

xvii

Kehoe, Margaret, Geraldine Hilaire-Debove, Katherine Demuth, & Conxita Lleó. 2008. The structure of branching onsets and rising diphthongs: Evidence from the acquisition of French and Spanish. Language Acquisition 15. 5–57. Martínez-Gil, Fernando. 2001. Sonority as a primitive phonological feature. In Julia Herschensohn, Enrique Mallén & Karen Zagona (eds.), Features and interfaces in Romance, 203–222. Amsterdam & Philadelphia: John Benjamins. Roca, Iggy. 2006. The Spanish stress window. In Fernando Martínez-Gil & Sonia Colina (eds.), Optimality-theoretic studies in Spanish phonology. 239–277. Amsterdam & Philadelphia: John Benjamins. Kirchner, Robert Martin. 1998. An effort-based approach to consonant lenition. Los Angeles, CA: UCLA dissertation.

Tabula gratulatoria Succinctly put, the following former students, colleagues, and friends share the thoughts about Jim expressed in previous sections, and wish to distinguish him for his lasting influence on Romance linguistics and his committed collegiality. Many others would have joined with me, had they been contacted on time. To them, my most sincere apologies for missing their names. Karlos Arregi Eric Baković Paola Bentivoglio Ricardo Bermúdez-Otero Eulália Bonet Ignacio Bosque Travis G. Bradley Barbara Bullock Teresa Cabré Monné Andrea Calabrese Noam Chomsky Heles Contreras Janet DeCesaris Violeta Demonte Manuel Díaz-Campos Francesco D’Introno Timothy Face Olga Fernández Soriano Kimberly L. Geeslin Juana Gil John Goldsmith Jorge M. Guitart Morris Halle Robert M. Hammond Bruce Hayes Nicholas Henriksen Julia Herschenson Eric Holt Paula Kempchinsky Michael Kenstowicz Rochelle Lieber Gillian Lord

Marta Luján Joan Mascaró John J. McCarthy Alfonso Morales-Front Terrell Morgan Juana Muñoz-Liceras Carol Neidle Frank Neussel Francisco Ocampo Wayne O’Neil Claudia Parodi Ana T. Pérez-Leroux Carlos Piera Shana Poplack Douglas Pulleyblank David Pesestky Wayne J. Redenbarger Lori Repetti Gemma Rigau Oliver María-Luisa Rivero Mario Saltarelli Mercedes Sedano Carmen Silva-Corvalán Donca Steriade Almeyda Jacqueline Toribio Soledad Varela Dieter Wanner Mary Rose Weston Karen Zagona Mary Zampini Maria Luisa Zubizarreta

I The Syllable and Glides

Laura Colantoni and José Ignacio Hualde

Conditions on front mid-vowel gliding in Spanish Abstract: This contribution deepens our understanding of the conditions favoring mid-vowel gliding in Mexican and Argentine Spanish. Whereas this process has been widely reported, the fact that there are clear differences among words with the same phonological sequence in their tendency to undergo gliding has been largely overlooked. We report on the results of a test of syllabification intuitions administered to Argentine and Mexican speakers. Four conditions were hypothesized: (1) Minimality: gliding is dispreferred when the result is a monosyllabic word; (2) Stress: the possibility of gliding increases with distance to stress; (3) Onset: gliding is dispreferred when the first vowel in the sequence is preceded by a palatal consonant or a cluster; and (4) gliding is dispreferred in second conjugation verbs. Hypotheses (1) and (4) are generally supported by the data, while (2) and (3) receive mixed support. Overall, there are also higher rates of mid-vowel gliding reported in Mexican than in Argentine Spanish.

1 Variable syllabification of vocalic sequences The focus of this paper is a well-known variable phenomenon in Spanish phonology: the realization of mid vowels as glides in vocalic sequences, as in teatro ‘theater’, realized as [ˈtja.tɾo]. Although often considered non-standard and subject to stigmatization (to different degrees depending on the area), this process is widespread in Latin American Spanish and, to a lesser extent, it is also found in Spain. The factors that condition its occurrence, however, are far from being completely understood, in spite of having been the object of extensive research (see Garrido 2013 for an overview). Our goal is to contribute to our understanding of the realization of these sequences by exploring some phonological and morphological restrictions through a syllabification intuition survey conducted with 102 participants. According to the standard syllabification rules of Spanish, sequences containing only non-high vowels are always realized as heterosyllabic, as in (1a). Sequences where one of the vowels is high and receives the stress are also

Laura Colantoni, University of Toronto José Ignacio Hualde, University of Illinois at Urbana-Champaign

4

Laura Colantoni and José Ignacio Hualde

syllabified as a hiatus, as in (1b). The very rare instances of two identical high vowels also have a normative pronunciation in hiatus, even if stress falls elsewhere in the word, (1c). In all the other cases, the two vowels in contact constitute a diphthong. The rules summarized in (1) apply in the marking of orthographic accent, and are taken to represent careful, reading style across dialects. For instance, petróleo ‘petrol’ carries an accent mark whereas monopolio ‘monopoly’ does not, because the former is taken to be proparoxytonic, since it contains a final sequence involving two mid vowels: pe-tró-le-o. (1)

Normative syllabification of vowel-vowel (VV) sequences VV is a hiatus if a. neither V is high ( feo [ˈfe.o] ‘ugly’, teatro [te.ˈa.tɾo] ‘theater’) b. V is high and stressed (río [ˈri.o] ‘river’) c. VV are high an identical, /ii/, /uu/ (duunviro ‘duumvir’, antihispano ‘anti-Hispanic’) VV is a diphthong otherwise (dio [djo] ‘s/he gave’)

The pronunciation of many of these sequences, however, is very variable and even idiosyncratic. On the one hand, some words that should contain a diphthong according to the generalizations in (1), can be realized with a hiatus in some Peninsular varieties, e.g., cliente [kli.ˈen̪.te] ‘client’ (see Hualde and Prieto 2002; Hualde 2005, among others). On the other hand, sequences that are heterosyllabic in normative pronunciation are often realized as diphthongs instead. The fact is that words like monopolio and petróleo often rhyme, both being realized with a final diphthong [jo]. Our focus is on the variable realization of the “hiatus” sequences in (1a). In particular, mid vowels may be realized as glides when unstressed and in contact with another non-high vowel, giving rise to diphthongs – see Navarro Tomás (1970 [1918]); Harris & Kaisse (1999) for additional examples and references on this process – as illustrated in (2): (2)

Gliding of mid vowels /e.a/ > [ja] e.g. teatro /a.e/ > [aj] e.g. caerá /e.o/ > [jo] e.g. peor /o.e/ > [we] e.g. poeta /o.a/ > [wa] e.g. toalla /a.o/ > [aw] e.g. ahora

> [ˈtja.tɾo] > [kaj.ˈɾa] > [pjoɾ] > [ˈpwe.ta] > [ˈtwa.ʝa] > [ˈaw.ɾa]

‘theater’ ‘it will fall’ ‘worse’ ‘poet’ ‘towel’ ‘now’ (with stress displacement)1

1 Diphthongization with stress displacement is a marked phenomenon. Some examples with both diphthong and hiatus pronunciation accepted in the dictionary of the Real Academia Española are reúma ~ reuma [re.ˈu.ma] ~ [ˈrew.ma] ‘rheumatism’, policíaco ~ policiaco [po.li.ˈsi.a.ko] ~ [po.li.ˈsja.ko] ‘police, adj.’ Diachronically we find stress displacement in examples like REGĪNA > *reína > reina ‘queen’.

Conditions on front mid-vowel gliding in Spanish

5

The facts presented in (2) can be described as resulting from the application of the variable rule in (3): (3)

Mid vowel gliding rule unstressed /e,o/ → [j,w] / in contact with another mid or low vowel

Our goal in this paper is to advance in our understanding of the factors that condition the application of this variable rule. Of all the sequences where this rule may apply, we will focus on /ea/. We do this for several reasons. First of all, gliding in this sequence seems to be very widespread; indeed, in the varieties where gliding of non-high vowel occurs, it has been unanimously observed to affect this sequence (e.g., Garrido 2013; Hernández 2009; Sosa 2010). Secondly, the sequence /ea/ is present in a very productive group of first conjugation verbs (i.e. verbs in -ear; e.g., golpear ‘to hit’), and it is also present in some forms of second conjugation verbs (e.g., ver ‘to see’, veamos ‘let’s see’), allowing us to explore whether morphology plays a role in favoring mid-vowel gliding. The remainder of this contribution is organized as follows: In section 2, we summarize some early reports of this mid-vowel gliding phenomenon in Vulgar or Late Latin (in 2.1.) and review recent studies conducted on several Latin American Spanish varieties that have identified specific conditioning factors for gliding (in 2.2.). Then, we focus on a paper and pencil experiment that we conducted in Querétaro, México and Buenos Aires, Argentina in order to determine the extent of mid-vowel gliding in those two geographically distant varieties, and the phonological and morphological constraints that may favor the application of the rule in (3). Section 3 states our hypotheses, section 4 describes our methodology and the results of the study are presented in section 5. This is followed by a discussion (section 6) and a brief conclusion (section 7).

2 Mid-vowel gliding: diachronic and synchronic variation 2.1 Gliding in Vulgar Latin It is interesting to remark that mid-vowel gliding is by no means a new phenomenon in Spanish. In fact, the variable syllabification of unstressed eV or iV sequences was already attested in Vulgar Latin and is found throughout the history of the language. Indeed, the Appendix Probi, an anonymous text, which was probably written between 200 and 320 A.D (Díaz y Díaz 1962), reports 25

6

Laura Colantoni and José Ignacio Hualde

confusions in the pronunciation of vocalic sequences out of a total of 228 examples of confusions involving a variety of phenomena (e.g., single vowels, consonant clusters, declensions, lexical substitutions). Interestingly, two basic symmetrical patterns are reported in the document; namely, eV sequences realized as iV, as in (4), and iV sequences realized as eV 2, as in (5). (4)

eV > iV

(Appendix Probi; Díaz y Díaz 1962: 46–53) vinea non vinia cavea non cavia cochlea non cochlia

(5)

iV > eV

(Appendix Probi; Díaz y Díaz 1962: 46–53) ostium non osteum alium non aleum lilium non lileum

The first and most obvious generalization from this small dataset is that at the time there was confusion about the underlying status of vowels in sequences; mid-vowels were realized as high vowels and high vowels as mid vowels. Second, there are more instances of the former (17/25) than of the latter (8/25) in this corpus. Third, most confusions occurred in unstressed syllables at the end of the word. We will see below (2.2.) that this bi-directionality of the process is still active in present-day Spanish. Additionally, patterns of synchronic variation suggest that mid-vowel gliding is a prerequisite for high-vowel lowering. That is, high-vowel lowering in this context appears to have the status of a hypercorrected pronunciation. Although from the spelling we cannot determine syllabication with certainty, from later developments we can infer that the confusion between and indicates a tautosyllabic pronunciation of what in Classical Latin were unstressed hiatus sequences, /e.V/, /i.V/, e.g., Classical Latin vīnea /ˈwi:.ne.a/ ‘vine’ > Late Latin /ˈvi.nja/ (> /ˈvi.ɲa/). The point of this short analysis of a Vulgar Latin source, though, was to indicate that the variability that we are witnessing now is not new and that a preference for high and mid-vowel gliding in order to avoid a hiatus is found across Romance languages with greater or lesser intensity (e.g., Chitoran & Hualde 2007). Among Romance varieties, Spanish has clearly favored diphthongization throughout its history, so that a number of authors mention the existence of an ‘anti-hiatus tendency’ (Lloyd 1987: 303, Quilis 1993: 189–190). This tendency 2 There are also confusions affecting other vowels in sequences, which are not relevant for the present investigation.

Conditions on front mid-vowel gliding in Spanish

7

may even involve stress shift, diachronically (e.g., REGĪNA > re.í.na > rei.na ‘queen’) or synchronically (dialectal maestro [ma.ˈes.tɾo] > maistro [ˈmajs.tɾo] ‘master bricklayer’), as mentioned in footnote 1. The avoidance of hiatus is, in fact, a cross-linguistically common phenomenon. Avoidance strategies include, besides diphthongization, deletion of one of the two vowels and epenthesis (see, e.g., Kaisse 2014). Thus, for instance, in many western Basque dialects a palatal consonant is inserted to break the underlying sequence /i-a/, e.g., mendia [mendiʒa] ‘the mountain’, whereas in eastern dialects this sequence undergoes diphthongization, [mendja]. Regarding the sequence /e-a/, as in semea ‘the son’, in the west the hiatus is preserved, but often with rising of the mid vowel, [semia], and in eastern dialects there is diphthongization with a mid or high glide [seme̯a], [semja] (see Hualde & Gaminde 1998). In vowel sequences across word-boundaries in Spanish a variety of strategies have been attested (see Jenkins 1999, Alba 2005, Aguilar 2003, 2005, Hualde, Simonet & Torreira 2008).

2.2 Previous synchronic studies on contemporary Spanish: conditioning variables The tendency to reduce mid vowels in a sequence in one way or another has been reported in most varieties of Spanish (e.g., Martínez Gil 2000; Navarro Tomás 1977 [1918]: §68). Nevertheless, the gliding of prevocalic mid vowels, as in (3), is clearly stronger in some Spanish varieties than in others (see e.g., Aguilar 2010; Alonso 1930; Alvar 1996; Martín Butragueño 2002, Garrido 2008; 2013; but see Quilis 1993: 190–1), and is associated with different degrees of stigmatization. In Spain, the phenomenon has been particularly stigmatized, so that Navarro Tomás (1977[1918]: §68c) refers to pronunciations like real [rjal] as “vulgar”.3 There are also differences in frequency of mid-vowel gliding within Latin American Spanish. For example, in Mexican Spanish these pronunciations seem to be both widespread and more acceptable for educated speakers than in, say, Colombia (Garrido 2008, see also MacLeod 2007). Previous work has identified different variables that seem to trigger midvowel gliding (we refer the reader to Garrido 2013, Hernández 2009, and Souza 2010 for detailed literature reviews on the topic). Among the speakers’ variables, 3 Navarro Tomás (1977[1918]: §68c) adds several more examples of mid-front and back-vowel gliding in footnote 1 (e.g., pasear > pasjar ‘to go for a walk’), where he also points out that mid-vowel gliding, although considered vulgar in Spain, is found among educated speakers in Latin America.

8

Laura Colantoni and José Ignacio Hualde

level of education is repeatedly mentioned as one of the factors that accounts for different rates of mid-vowel gliding, with higher rates being found among less educated speakers (Garrido 2008; Vidal de Battini 1964). This may be at the source of the stigmatization of the phenomenon, since a tendency to diphthongize in pronunciation (e.g., /ea/ → [ja] may lead to spelling mistakes (e.g., → and → ), among less educated speakers. In the case of verbs, the phenomenon may also cause insecurity regarding verb conjugation, giving rise to non-normative (and stigmatized) forms, e.g., peliar for standard pelear ‘to fight’ and also, for instance, hyper-corrected rumean [ru.ˈme.an] ‘they ruminate’, for standard rumian [ˈru.mjan] (because infinitive forms in /-ear/ and /-iar/ are neutralized). Gliding also appears to be more frequent in casual or colloquial speech than in formal styles (Navarro Tomás 1977 [1918]; Alba 2006). Souza (2010) raises the interesting possibility that the higher rate of gliding that has been reported in more casual styles may in fact have its ultimate source in the faster speech rate correlated with colloquial styles. Beyond the inter-lectal and style-dependent variation that has been noticed, the tendency towards gliding may also be conditioned, in part, by the phonological context. A factor that has been repeatedly reported to condition the articulation of VV sequences as tautosyllabic or heterosyllabic is lexical stress. Stressed high or mid vowels in contact with another vowel do not become glides (we refer the reader to the general syllabification rules summarized in (1) above), e.g., tía [ˈti.a] ‘aunt’, fea [ˈfe.a] ‘ugly, f.’). Unstressed vowels, however, may be subject to different degrees of reduction (see Martínez Gil 2000; Navarro Tomás (1977) [1918]: §68c)4, since they tend to be shorter and more sensitive to coarticulation than stressed vowels (Chitoran & Hualde 2007; Recasens 1999), even in syllable-timed languages, such as Spanish. Among unstressed positions, it has been noticed that in Peninsular Spanish, unstressed high vowels before another vowel are more likely to preserve their syllabic character the closer to the stress they are (Hualde 1999, 2005). We may expect a similar effect of proximity to the stress in the gliding of mid vowels in sequences. Utterance-initial fortition, phonologized as word-initial, on the other hand, may favor the maintenance of hiatuses, since it has been shown (see Cho 2004) that vowels in prosodically strong positions show less coarticulation than those same vowels in prosodically weaker positions. Thus, it has been claimed that word-initiality plays a role in the preservation of exceptional hiatus in Peninsular Spanish in sequences with unstressed high vowels, e.g., diana [ˈdi.a.na] vs. 4 These authors have pointed out that reduction may range from lax vowel articulation to a glide. We refer the reader to Limanni (2014) for a detailed articulatory description of the nature of these vowels.

Conditions on front mid-vowel gliding in Spanish

9

mediana [me.ˈðja.na] (Hualde 1999, 2005:84, Hualde & Prieto 2002, Chitoran & Hualde 2007). As far as we know, this factor has not been mentioned in connection with the reduction of sequences involving non-high vowels. A third factor that has been claimed to condition vowel gliding is the place of articulation of the preceding consonant. For example, preceding coronal consonants, such as /d/ and /n/ have been reported to favor gliding of mid vowels more than any other type of consonants (see Souza 2010). Finally, morphological and lexical factors have also been taken into account. Mid-vowel gliding has been reported to be more frequent with specific verbs endings, including -ear (Garrido 2013; Hernández 2009), and with more frequent words (Souza 2010), which is consistent with the role that frequency has been claimed to play in many other reductive processes (e.g., Bybee 2001). Previous research has guided us (i) in choosing the varieties for this study (i.e. two Latin American Spanish varieties with potentially different rates of midvowel gliding, Mexican Spanish and Argentine Spanish), (ii) in selecting the target sequence (i.e., ea, which shows a high rate of gliding) and (iii) in controlling for educational factors (all our speakers have university education). In spite of the attention that mid vowel gliding has received in Spanish, it appears to us that existing work has not sufficiently explored the linguistic factors that favor or disfavor the application of the gliding rule within the word domain. In the next section, we turn to the observations that triggered our specific research questions and the hypotheses of the present study.

3 Research questions and hypotheses Our first research question concerns the linguistic (phonological and morphological) factors that condition the application of the mid-vowel gliding rule within the word domain. As mentioned in the preceding section, a number of factors have been identified in the literature on this topic, including preceding consonant, stress pattern, word class and lexical frequency. Other linguistic factors, however, have received less attention. In particular, our preliminary observations suggest that a word like peleamos ‘we fight’ is much more likely to be realized with gliding, as [pe.ˈlja.mos], than leamos ‘let us read’ is to be realized as ??[ˈlja.mos]. That is, there may be differences in the rate of gliding for sequences preceded by the same consonant, with similar stress patterns and which appear in words that belong to the same class and have similar frequencies. Since we are aware that linguistic factors may not be the same or may not have the same strength in all Spanish varieties, our second research question is

10

Laura Colantoni and José Ignacio Hualde

whether these phonological and morphological factors are active in different Spanish varieties and affect the same sequences to the same extent. In order to give a partial answer to this second question, we decided to include two varieties in our study: Argentine and Mexican Spanish. In order to address our research questions, we have formulated four hypotheses regarding the likelihood of the change /ea/ > [ja] in different environments, based both on our intuition and on previous research. Here we state the hypotheses that motivated our experiment as we originally formulated them, although, as we will see, the results of the experiment did not always offer support for the hypotheses. Our first hypothesis (H1) is that gliding is dispreferred when the result of applying the rule would be a monosyllabic word; e.g., cereal [se.ˈɾjal] ‘cereal’ will be more acceptable than leal ??[ˈljal] ‘loyal’. Although based on our observation, this hypothesis is also justified by some claims in the literature. Cabré and Prieto (2004) find that, in Catalan, diphthongization of hiatus (with a high vowel) is strongly dispreferred when the result is a monosyllabic word. Thus in Catalan there is diphthongization in cam[jo] ‘truck’, but not in m[i.ɔ]l, *m[jɔ]l ‘meow’. Unlike in Catalan, this constraint clearly does not operate in Spanish in the case of sequences with high vocoids, e.g., dio [djo] ‘s/he gave’. Nevertheless, it seems at least possible that the constraint may still affect sequences with mid vowels, disfavoring their gliding. We will refer to this constraint as the Minimality Condition. Our second hypothesis (H2) is that proximity to stress blocks gliding. In particular, in the sequence /ea/, the gliding of /e/ is expected to be more likely when the following /a/ does not bear stress. We thus predict that a pronunciation of a word like idealismo as [i.dja.ˈlis.mo] will be judged to be more likely than a pronunciation of ideal as ??[i.ˈdjal]. This hypothesis is motivated by the fact that stressed mid vowels do not undergo gliding, e.g., fea ‘ugly, fem.’ *[fja], although there are a few lexicalized cases, with stress shift in sequences of falling sonority: maestro /maéstro/ [ˈmajs.tɾo] ‘teacher; master bricklayer’, ahora /aóɾa/ [áw.ɾa] ‘now’. As mentioned above, Hualde (1999, 2005: 84) notes an effect of proximity to the stress in the syllabification of sequences with high vocoids in Peninsular Spanish; e.g., diálogo [di.ˈa.lo.ɣo] ‘dialog’ vs. dialogo [di.aˈlo.ɣo] ~ [dja.ˈlo.ɣo] ‘I converse’ vs. dialogó [dja.lo.ˈɣo] ‘s/he conversed’ (see also Simonet 2005). In this connection too, it has also been frequently observed that reduction of vowel sequences across word-boundaries to a single syllable is more likely when neither vowel is stressed (e.g., Jenkins 1999; Martínez-Gil 2000; Morris 1998). We will refer to this dispreference of gliding next to a stressed vowel as the Stress Condition hypothesis.

Conditions on front mid-vowel gliding in Spanish

11

Concerning exceptional hiatus with unstressed high vowels, Hualde (1999, 2005), Hualde & Chitoran (2007) propose an initiality condition which disfavors diphthongization in word-initial position, as was also mentioned above. Extending this condition to the /ea/ sequences we examine in this paper, the hypothesis would be that a diphthongal pronunciation [ja] would be dispreferred for word-initial sequences (e.g., peatón ‘pedestrian’ vs. apeadero ‘stopping place’). We do not test this condition in this paper because of the difficulty in finding enough examples that would allow teasing apart its effects from those of the other two conditions just mentioned. Preceding consonants have been reported to have an effect on favoring or blocking mid-vowel gliding (e.g., Souza 2010). Based on our intuition, we hypothesize that the complexity of a preceding onset acts against the gliding of mid vowels (H3). That is, gliding will be less acceptable after complex onsets (CC), as in emplear ‘to employ’ (H3a) and after post-alveolars, as in centellear ‘to sparkle’ (H3b). This constraint, which we dub the Onset Condition, would follow from the avoidance of complex patterns of articulatory timing (on CC timing, see Goldstein et al. 2007). Finally, our fourth and last hypothesis is limited to verb forms and, unlike the preceding three hypotheses, is of a morphological nature. This hypothesis states that in verbal forms, gliding will be favored with first conjugation verbs and dispreferred with second conjugation forms (H4). We will refer to this hypothesis as the Verb Class Condition. In standard Spanish pronunciation, verbs in -iar (see 6 below) may present one of two patterns, depending on whether the sequence is treated as a diphthong or a hiatus, whereas verbs in -ear only admit a hiatus syllabification: (6) -iar diphthong: -iar hiatus: -ear:

infinitive copiar ‘to copy’ enviar ‘to send’ desear ‘to desire’

present indicative copia /ˈko.pja/ ‘s/he copies’ envía /en.ˈbi.a/ ‘s/he sends’ desea /de.ˈse.a/ ‘s/he desires’

Gliding of the mid vowel in infinitives in -ear (e.g., desear > [de.ˈsjaɾ]) may cause insecurity regarding the underlying form (e.g., is it desear or desiar? copiar or copear?) and this would affect other forms of the paradigm, such as the present indicative. A quick Google search is enough to confirm that hypercorrected forms are found for verbs of both -iar types, e.g., copia ‘s/he copies’ > copea, vacía > vacea ‘s/he empties’ (stressed vowels in boldface). This is reminiscent of the vacillations attested in the Appendix Probi. The presence of hypercorrection attests to the frequency of mid vowel gliding in first conjugation

12

Laura Colantoni and José Ignacio Hualde

verbs. In contrast, in verbs whose infinitive ends in -eer, where forms with /ea/ are restricted to the subjunctive, gliding seems unlikely to us. It must be added that, in contrast to the large number of verbs in -ear, there is only a handful of verbs in -eer: creer, leer, proveer, poseer. The verbs ser and ver also have forms in -ea: sea, vea. Based on our own observations, we hypothesize that gliding of /e/ in subjunctive/imperative forms like lea, leamos, proveas, posea, vea will be uncommon.

4 Methods 4.1 Participants A total of 102 participants were recruited for this study; 54 participants were speakers of Mexican Spanish and completed the questionnaire at the Universidad Autónoma de Querétaro, México and 48 participants were from Buenos Aires, Argentina. With the exception of four of the Argentine participants, all of them were university students.

4.2 Stimuli and task This study is based on a paper-and-pencil test of phonological intuitions. A total of 41 words were used in this study (see Appendix A). Forty words contained the sequences /ea/ with unstressed /e/. In addition, we also included the word fea /ˈfe.a/ ‘ugly, fem.’, with a stressed /e/. This item was intended as a control for which no diphthong responses were expected. That is, if participants indicated a preference for a diphthong with this word, this would lead us to suspect that they were not paying enough attention to the instructions, so that their pattern of responses would warrant a second look. There were no other distractors. This list of words was presented to participants in a table format as exemplified in (6), where the instructions that were given to the participants are also included. Participants were asked to circle whether they thought that the option with a high vowel was possible in their variety. This was, thus, a simple task, whose completion required only a few minutes.

Conditions on front mid-vowel gliding in Spanish

(6)

13

Instructions for participants En español, hay palabras que tienen una pronunciación coloquial que es diferente de la ortográfica. Indique si las siguientes palabras se pueden pronunciar coloquialmente de la manera que se señala en la segunda columna, según su intuición [In Spanish, there are some words that have a colloquial pronunciation that is different from the orthographic one. Please indicate whether the following words can be pronounced in colloquial style as in the second column, according to your intuition] pelear ideal

pe-liar i-dial

SÍ SÍ

NO NO

Participants were thus asked to reflect on their intuitions regarding a possible colloquial pronunciation of 41 words with orthographic .

5 Results 5.1 General results Comparing across dialect groups first, results show that Mexican speakers accepted mid-vowel gliding in /ea/ sequences at a higher rate than Argentine speakers. The mean of ‘yes’ responses for all the test items was 40% (s.d. = 11.5) for Argentina and 49% (s.d. = 17.6) for Mexico. A t-test revealed that there was a significant difference between the two groups (p = 0.004). Although within each of the two groups there was some variation among speakers in their responses, the tendencies for the two dialects are clear. As the histogram in Figure 1 shows, 40 out of 52 Mexican participants considered a pronunciation with a glide to be possible for 16 to 30 out of the 41 items that they were asked to judge, with almost equal numbers in the three bins 14–20, 21–25 and 26–30 in which numbers of ‘yes’ responses have been divided. In comparison, Figure 2 shows that most Argentine participants accepted a pronunciation with a glide for a smaller range of items (11–20), with only 8 participants giving ‘yes’ answers beyond this range. Thus, we see a different distribution of responses with Mexican participants shifting to the higher end of the acceptance scale.

14

Laura Colantoni and José Ignacio Hualde

Figure 1: Number of Mexican participants that accepted the orthographic form with a diphthong (e.g., pe-liar) for the 41 items on our stimuli list.

Figure 2: Number of Argentine participants that accepted the orthographic form with a diphthong (e.g., pe-liar) for the 41 items on our stimuli list.

On the other hand, in spite of this overall difference in the readiness with which pronunciations with a high vowel are accepted, when individual items are considered, we find that speakers of both regions coincided to a great extent in identifying specific lexical items as being particularly prone or particularly unlikely to be realized with a high vowel. Table 1 displays the items that received less than 10% of ‘yes’ responses in each of the two dialects:

Conditions on front mid-vowel gliding in Spanish

15

Table 1: items with 80% ‘yes’ responses by dialect. Note: the words that coincided in both dialects are underlined Mexico

Argentina

Word

% yes

Word

% yes

alinear mear arrear titubeamos titubear golpear

93 91 89 87 85 83

pestañear titubear arrear alinear pelear titubeamos

92 87 83 83 83 81

In this group of most diphthong-prone words, we found again a remarkably high degree of consistency between dialects, with alinear ‘to align’, arrear ‘to drive a cattle’, titubear ‘to hesitate’, and titubeamos ‘we hesitate’ appearing in both lists. There appear to be, however, also some differences between the two geographical varieties regarding specific lexical items. In particular, the word mear ‘to pee’, which we had predicted to have a low mid-vowel gliding index based on the intuition of one of the authors (who is a native speaker of Argentine Spanish), turned out to be the one with the second highest score in the average ratings of our Mexican participants. For Argentina, on the other hand, the average rating for this word was rather low, as predicted, 15%. This word thus behaves very differently in the two geographical areas under consideration. The other word that differed substantially in its rate of accepted mid-vowel gliding across dialects was pestañear ‘to blink’. This word, which is included in Table 2 for the Argentine speakers, had a somewhat lower average rating for our Mexican group of participants, 64.8%. This may be related to the tendency to neutralize the contrast between the palatal nasal /ɲ/ and the sequence /nj/ in Argentine Spanish (see Kochetov & Colantoni 2011). We now turn to analyze the results in term of each of the four hypotheses presented in §3.

5.2 Results H1: Minimality condition Our first hypothesis stated that mid-vowel gliding would be dispreferred when the result would be a monosyllabic word. Our questionnaire contained three items to test this hypothesis, leal, real and mear. As shown in Table 3, the hypothesis was confirmed for Argentina, since two of these words, leal and real received no ‘yes’ responses and mear had only 15% of ‘yes’ responses. In general, the hypothesis was also confirmed for Mexico, where leal and real also

Conditions on front mid-vowel gliding in Spanish

17

obtained very few ‘yes’ responses. The exception is the word mear, which is, instead, among the ones with the highest percentage of ‘yes’ responses in our Mexican data, as mentioned above. For this particular word, which is somewhat vulgar, we thus find a clear difference between the dialects in degree of acceptability of the high vowel pronunciation, in spite of the substantial overall agreement reported above in §5.1. Table 3: Minimality Condition: percentage of ‘yes’ responses Word

Mexico

Argentina

leal real mear

9 15 90

0 0 15

We thus see that mid-vowel gliding with potentially monosyllabic words is dispreferred in Argentine Spanish and, less strongly also in Mexican Spanish, although with lexical exceptions, perhaps related to the vulgar or informal character of the word.

5.3 Results H2: Stress condition Leaving verbs aside, we hypothesized that, in the sequence /ea/, gliding would be less frequent when the vowel /a/ bears lexical stress. In addition to the items in Table 3, our stimuli included three additional words with stressed /eˈa/, which are listed in Table 4 together with the percentage of accepted mid-vowel gliding. Table 4: Stress Condition: percentage of ‘yes’ responses Word

Mexico

Argentina

ideal beata nuclear

31 28 54

8 21 44

Results in Table 4 do not offer strong support for the second hypothesis. Indeed, the three nouns with pre-tonic /e/ in our list, which would not have yielded a monosyllabic word if diphthongization applied, displayed variable rates of gliding within and across dialects. The only exception to this tendency was the word ideal for our Argentine participants. There are, however, other relevant data that are consistent with the hypothesis and may offer support for it. The questionnaire included three pairs of morphologically related words (i.e. beata ‘pious, fem.’ / beatitud ‘bliss’; leal ‘loyal’ / lealtad ‘loyalty’; real

18

Laura Colantoni and José Ignacio Hualde

‘real’ / realismo ‘realism’). In the morphologically simple word in these pairs, the lexical stress falls on the second vowel of the sequence, whereas in the derived word the stress falls on either the following syllable or two syllables after. If we compare the percentage of accepted mid-vowel gliding for the derived word against the simple word, as we have done in Table 5, we can see that in both dialects the percentage of accepted mid-vowel gliding was higher for the derived word, i.e., in the cases where the stress was farther away from the target mid vowel in the sequence. Nevertheless, it is important to notice that, for two of the three test items on our list, the low percentage of mid-vowel gliding in the /ea/ sequences could also be motivated by our minimality condition.5 Table 5: Stress Condition: percentage of mid-vowel gliding in morphologically related pairs. The stressed vowel is in bold face. Word

Mexico

Argentina

beata/beatitud leal/lealtad real/realismo

28/56 9/37 15/52

21/47 0/15 0/29

5.4 Results H3: Onset condition Our third hypothesis stated that gliding of /e/ would be dispreferred after a complex onset CC (H3a) and after a post-alveolar consonant (H3b). Relevant results are shown in Tables 6 and 7. As can be seen, only H3a was supported, and then only for the Argentine participants, for whom three of the four items with a complex onset received less than 10% of ‘yes’ responses. Although Mexican participants accepted the possibility of mid-vowel gliding after complex onsets at higher rates than Argentine speakers (Table 6), the opposite tendency is observed in Table 7 for sequences following a post-alveolar consonant. Even if the percentage of mid-vowel gliding for those words is rather high across dialects, Mexican speakers showed lower percentages, especially in a word like centellear ‘to sparkle’, which is realized with a true palatal [ ʝ] in the target syllable onset, as opposed to Buenos Aires Spanish, where the is realized as a post-alveolar fricative [ʒ] or [ ʃ ]. 5 Ellen Kaisse suggested to us to look into the foot structure of our stimuli in order to determine whether secondary stress (which in Spanish is postlexical) may play a role. Although this may be worth exploring, we notice that for our Mexican participants the response rates for a word like ideal, where the initial syllable could receive rhythmic secondary stress if syllabified in three syllables, [ˌi.ðe.ˈal], and beata, where it could not, [be.ˈa.ta] are very similar. Something both words have in common is that stress falls on the second vowel of the sequence: /ide.ˈal/, /be.ˈa.ta/.

Conditions on front mid-vowel gliding in Spanish

19

Table 6: Complex Onset Condition: percentage of mid-vowel gliding Word

Mexico

Argentina

emplear nuclear recrear creamos

63 54 22 18

9 44 4 4

Table 7: Post-alveolar Onset Condition: percentage of mid-vowel gliding Word

Mexico

Argentina

centellear truchear pestañear

24 56 65

42 58 92

5.5 Results H4: Verb class condition Regarding word class, our hypothesis was that gliding would be disfavored in second conjugation verbs, where the sequence /ea/ is found only in subjunctive forms. Table 8 compares the percentage of acceptable mid-vowel gliding obtained for first person plural forms in verbs belonging to the second and first conjugations. Our hypothesis was clearly confirmed in this case, with second conjugation (subjunctive) verb forms in /-eamos/ receiving extremely low rates of ‘yes’ responses. The word with the highest percentage of ‘yes’ answers in this subgroup was creamos. This word, however, is actually ambiguous. Although it is most likely to be interpreted as a subjunctive form of creer ‘to believe’, in fact it can also be an indicative form of crear ‘to create’.6 Table 8: Verb Class Condition: Percentage of mid-vowel gliding Conjugation

Word

Mexico

Argentina

2nd 2nd 2nd 2nd/(1st) 1st 1st 1st 1st

veamos leamos seamos creamos* apeamos deseamos franqueamos titubeamos

2 4 9 18 53 41 43 87

0 0 2 4 38 40 51 81

6 A fact that we did not take into account in preparing the list of test words is the possible existence of minimally contrasting words with a lexical diphthong. This is the case with leamos ‘let’s read’ vs. liamos ‘we wrap up’ and creamos ‘let’s believe/we create’ vs. criamos ‘we raise’.

20

Laura Colantoni and José Ignacio Hualde

Our questionnaire also included a long list of infinitives in -ear: afear, alinear, apear, arrear, asear, centellear, contonear, desear, emplear, franquear, golpear, hojear, mear, ojear, pelear, pestañear, recrear, titubear, truchear. The average percentage of ‘yes’ responses for this group of items was 63% for Mexico and 54% for Argentina. In general, thus, it can be concluded that midvowel gliding is moderately favored in first conjugation verbs, whereas it is strongly rejected with second conjugation verbs. We notice that there are, nevertheless, very clear differences in ratings among verbs of the first conjugation, and that some of these differences in evaluation do not follow from our other hypotheses. One example is the very high percentage of ‘yes’ responses that the word titubeamos received in both dialects. We did not predict that this word would tend to be more acceptable with gliding than the other first conjugation items in Table 8. Perhaps the fact that the two preceding syllables contain a high vowel has something to do with it. This hypothesis would need to be tested.

6 Discussion 6.1 Hypothesis evaluation Our study set forth to test four hypotheses regarding the phonological and morphological conditioning of mid-vowel gliding in /ea/ sequences for educated speakers of two Spanish varieties, Mexican and Argentine. Regarding acceptability of mid-vowel gliding in different words, two of our four hypotheses were clearly confirmed in our paper-and-pencil test of intuitions. The other hypotheses either received minor support or no support at all. The Minimality Condition Hypothesis (H1) was confirmed, as mid-vowel gliding in words that would turn out to be monosyllabic if gliding applied, was quasi-categorically rejected both in Argentine Spanish and in Mexican Spanish, with the exception of the word mear in the latter variety. The Verb Class hypothesis (H4) was also strongly supported. Whereas gliding in verbs of the first conjugation received an overall high level of acceptance, mid-vowel gliding of the same sequence in very similar phonological contexts was rejected for verbs of the second conjugation, e.g., deseamos (desear) vs. seamos (ser). To the best of our knowledge, this morphological conditioning has not been noted in previous studies. Our Stress Condition (H2) received only weak support. Some words like beata with stress on the second vowel of the sequence received higher numbers of diphthong responses than expected (for Mexican Spanish only, also ideal). A

Conditions on front mid-vowel gliding in Spanish

21

tendency was, nevertheless, observed in pairs of morphologically related words (e.g., beata/beatitud). In this case, we obtained a lower percentage of mid-vowel gliding for the simple word, where the target sequence was closer to the primary stress, than in the derived word, in which the target vocalic sequence was further away from the stress. In any event, to the extent that proximity to the stress disfavors gliding of mid vowels, this tendency has less strength than other factors, since verbs in -ear produced relatively high rates of diphthong responses in both dialects, especially Mexican Spanish. Finally, only one of the two parts of our hypothesized Onset Condition (H3) was supported, the Complex Onset Condition, and then only for the Argentine subjects. This appears to be a context where the two dialects differ. At this point we may speculate whether there could be other factors that condition frequency of mid vowel gliding that we have not tested. First of all, as mentioned in section 3, we decided not to test the Initiality Condition that has been proposed in other work (Hualde 1999, 2005, Chitoran & Hualde 2007) for sequences with high vowels. In addition, the results of this experiment suggest the possibility of other conditions on the likelihood of gliding in /ea/ sequences. We found, for instance, a greater acceptability of the diphthong pronunciation in titubeamos than in other first conjugation forms in -eamos included in our questionnaire. This was not expected. As suggested above, the presence of a high vowel in the two preceding syllables may be a contributing factor. Souza (2010) considers word frequency as a factor in this process and it may very well play a role in mid-vowel gliding. In our study, however, there were two words that were classified as frequent in Souza’s study (ideal and real) that showed very little gliding in Argentine Spanish, whereas some infrequent words, such as beata, showed higher rates of gliding in this same dialect. The large difference between dialects that we found in the acceptability of monosyllabic miar for mear, suggests that either frequency or other non-phonological variables, such as whether the word is considered colloquial or vulgar, may indeed play a role in accounting for the different rates of mid-vowel gliding of particular words. Thus, to summarize, we have identified two conditions, namely word-size (avoidance of monosyllabicity) and verb class, that seem to block gliding in both dialects, with the exception of the lexical item just mentioned for Mexican Spanish. A third condition (complex onset) may be active in Argentine but not in Mexican Spanish. Distance to stress, other things being equal, plays a weaker role.

22

Laura Colantoni and José Ignacio Hualde

Regarding similarities and differences between dialects, mid-vowel gliding in /ea/ was accepted as a possibility for all 40 items in our questionnaire, although at very different rates depending on the specific lexical item. We found substantial agreement among our subjects from both geographical varieties regarding which items are likely or unlikely to undergo gliding. At the same time, Mexicans are significantly more tolerant of diphthong pronunciations than Argentinians. That is, our study has confirmed other reports indicating that mid-vowel gliding is more advanced in Mexican than in Argentine Spanish. A high frequency of mid-vowel gliding in Mexican Spanish has been reported in a number of studies, including Perissinotto (1979), Moreno de Alba (1994) and Hernández (2009), among others. Garrido (2008), who compared Mexican against Colombian Spanish, also found that the gliding of mid vowels displayed a higher frequency in the former variety.

6.2 Categorical vs. gradient effects Our results revealed that some of the hypothesized constraints on mid-vowel gliding yielded (near) categorical responses for some lexical items. Between the two constraints that better predicted the absence of gliding, the Minimality Condition and the Verb Class Condition we want to concentrate here on the Verb Class Condition. We would like to suggest that gliding, or the lack of it, is a way of enhancing the contrast between the present indicative of verbs in -ear and the present subjunctive of verbs in -eer/-er (ver and ser), as illustrated in (7). This is the only environment where both conjugations have the same VV sequence with an unstressed mid vowel: (7)

Present Indicative (-ear) desear → de[sja]mos

Present Subjunctive (-eer/-er) leer → [le.a]mos

Given the fact that the sequence [ea], without stress on /e/, is found in many forms in the conjugation of -ear verbs (indicative: imperfect, preterit, future; subjunctive: present –nosotros form –, imperfect; infinitive, participle and gerund), paradigm effects could be responsible for extending mid-vowel gliding to all the target environments. In the case of -eer verbs, instead, there is only one verbal form where the sequence is found (nosotros form of the present subjunctive). It would be interesting to see if these paradigm effects are found in frequent and infrequent verbs alike and if they also extend to nonsense verbs. Our intuition tells us that they do extend to new verbal forms.

Conditions on front mid-vowel gliding in Spanish

23

6.3 On inter-dialectal and inter-speaker variability A final issue that it is worth exploring is the possible reasons behind the interdialectal and the inter-speaker variability that we find. As we have seen, the two dialects have different rates of mid-vowel gliding and within each dialect speakers also exhibited different degrees of acceptability of the gliding rule. A number of authors have noted lower rates of acceptance of gliding by more educated speakers (see §2.2 above for specific references). Although all participants in our study had similar levels of education, informal conversations with them suggested that not all of them attributed the same degree of stigmatization to mid-vowel gliding. In addition to educational level, thus, attitudinal factors may also play a role. Beyond the participants’ attitude towards mid-vowel gliding, there are possible articulatory motivations behind the variability observed. It has been suggested (see Chitoran & Hualde 2007) that VV sequences are an unstable mode of coordination. If this is the case, we expect to see variability among speakers and at different speech rates (see, e.g., Aguilar 1999). Depending on speech rate, vowel targets may not be achieved (see Recasens 1999). Indeed, previous studies have reported that the vowels /i/ and /e/ in a sequence are shorter than either the same vowels in hiatus or in isolation (Borzone de Manrique 1979; Hualde et al. 2008). Moreover, the duration of these vowels varies depending on the distance to the stressed syllable and on speech rate. Cross-dialectal studies (e.g., Garrido 2008) have reported that these vowel sequences are shorter in dialects that diphthongize more and shortening of the first vowel in the sequence has also been observed to correlate with a path towards diphthongization (e.g., Colantoni and Limanni 2010). Thus, given that speech rate varies across dialects and across individuals, variability in the acceptance of mid-vowel gliding should not come as a surprise.

7 Conclusion In this study we tested 102 speakers of Mexican and Argentine Spanish on their acceptance of mid-vowel gliding in words containing the sequence /ea/ in different environments. Results showed that variable syllabification of this sequence, a phenomenon that was already active in Vulgar Latin, is still synchronically active in these two Latin American varieties. We have observed differences between dialects (namely a higher rate of mid-vowel gliding in Mexican than in Argentine Spanish, which is consistent with previous reports) and

24

Laura Colantoni and José Ignacio Hualde

between individuals. Speakers’ judgments may reflect different attitudes towards a sociolinguistically marked phenomenon. To the extent that judgments reflect actual pronunciation, differences between the dialects (and among speakers) may ultimately also reflect differences in speech rate. We have also noticed that, in spite of the variability that is found, speakers seem to have rather consistent intuitions about where gliding is not supposed to occur. In particular, participants from both dialects consistently rejected midvowel gliding with second conjugation verbs and, to some extent, when the result of mid-vowel gliding would yield a monosyllabic word. At the other end of the scale, high consistency across dialects was also found in identifying certain words as being particularly prone to undergo mid-vowel gliding.

Acknowledgements We want to thank Daniel Scarpace and María Mó for their help in administering the questionnaire and E. Kaisse and an anonymous reviewer for comments on a first version of this paper.

References Aguilar, Lourdes. 1999. Hiatus and diphthong: Acoustic cues and speech situation differences. Speech Communication. 28-57-74. Aguilar, Lourdes. 2003. Effects of prosodic and segmental variables on vowel sequences pronunciation in Spanish. In Maria-Josep Solé, Daniel Recasens, & Joaquín Romero (eds.). Proceedings of the 15th International Congress of Phonetic Sciences, 2011–2014. Barcelona. Causal Productions, CD-Rom: Adelaide, Australia. Aguilar, Lourdes. 2005. Los enlaces vocálicos: ¿cuestión de dominios prosódicos? Revista Internacional de Lingüística Iberoamericana 6. 29–48. Aguilar, Lourdes. 2010. Vocales en grupo. Madrid: Arco Libros. Alba, Matthew. 2005. Hiatus resolution between words in New Mexico Spanish: A usage-based account. University of New Mexico dissertation: Albuquerque. Alba, Matthew. 2006. Accounting for variability in the production of Spanish vowel sequences. In Nuria Sagarra & Ameida J. Toribio (eds.), Selected proceedings of the 9th Hispanic Linguistics Symposium, 273–285. Somerville: Cascadilla Press. Alonso, Amado. 1930. Problemas de dialectología hispanoamericana. Buenos Aires: Facultad de Filosofía y Letras. Alvar, Manuel. 1996. Manual de dialectología hispánica. El español de América. Barcelona: Ariel. Borzone de Manrique, Ana M. 1976. Acoustic study of /i, u/ in the Spanish diphthongs. Language and Speech. 19. 121–128.

Conditions on front mid-vowel gliding in Spanish

25

Bybee, Joan. 2001. Phonology and language use. Cambridge: Cambridge University Press. Chitoran, Ioana & José I. Hualde. 2007. From hiatus to diphthongs: the evolution of vowel sequences in Romance. Phonology 24. 37–55. Cabré, Teresa & Pilar Prieto, P. 2004. Prosodic and analogical effects in lexical glide formation in Catalan. Probus 16. 113–150. Cho, Taehong. 2004. Prosodically-conditioned strenghthening and vowel-to-vowel coarticulation in English. Journal of Phonetics 32. 141–176. Colantoni, Laura & Anna Limanni. 2010. Where are hiatuses left? A comparative study of vocalic sequences in Argentine Spanish. In Karlos Arregi, Zsuzsanna Fagyal, Silvina Montrul & Annie Tremblay (eds.). Selected proceedings of the 38th Linguistic Symposium on Romance languages, 23–38. Amsterdam & Philadelphia: John Benjamins. Díaz y Díaz, Manuel. 1962. Antología del latín vulgar. Madrid: Gredos. Garrido, Marisol. 2008. Diphthongization of non-high vowel sequences in Latin American Spanish. University of Illinois at Urbana-Champaign dissertation. Garrido, Marisol. 2013. Hiatus resolution in Spanish: Motivating forces, constraining factors, and research methods. Language and Linguistic Compass 7/6. 339–350. Goldstein, Louis, Ioana Chitoran & Elisabeth Selkirk. 2007. Syllable structure as couple oscillator modes: evidence from Georgian vs. Tashlhiyt Berber. In Jürgen Trouvain & William J. Barry (eds.). Proceedings of the International Congress of Phonetic Sciences, 241–244. Saarbrücken: Universitat des Saarlandes. Harris, James and Ellen Kaisse. 1999. Palatal vowels, glides and obstruents in Argentinian Spanish. Phonology 16. 117–190. Hernández, Edith. 2009. Resolución de hiatos en verbos -ear: Un estudio sociofonético en una ciudad mexicana. The Ohio State University dissertation. Hualde, José I. 1999. Patterns in the lexicon: Hiatus with unstressed high vowels in Spanish. In Javier Gutiérrez-Rexach & Fernando Martínez-Gil (eds.). Advances in Hispanic Linguistics: Papers from the 2nd Hispanic Linguistics Symposium, 182–197. Somerville: Cascadilla Press. Hualde, José I. 2005. The sounds of Spanish. Cambridge: Cambridge University Press. Hualde, José I. & Iñaki Gaminde. 1998. Vowel interaction in Basque: A nearly exhaustive catalogue. Studies in the Linguistic Sciences 28.1. 41–77 Hualde, José I., & Mónica Prieto. 2002. On the diphthong/hiatus contrast in Spanish: some experimental results. Linguistics 40. 217–34. Hualde, José I., Miquel Simonet & Francisco Torreira. 2008. Postlexical contraction of non-high vowels in Spanish. Lingua 118. 1906–1925. Jenkins, Devin. 1999. Hiatus resolution in Spanish: Phonetic aspects and phonological implications from Northern New Mexico data. University of New Mexico dissertation. Kaisse, Helen. 2014. Dialects of Spanish and Modern Greek: Natural laboratories for the generative phonologist. LSA Presidential Address. Kochetov, Alexei & Laura Colantoni. 2011. Coronal place contrasts in Argentine and Cuban Spanish: An electropalatographic study. Journal of the International Phonetic Association 41. 313–342. Limanni, Anna. 2014. Diphthong and related processes in Spanish dialects. University of Toronto dissertation. Lloyd, Paul. 1993. Del latín al español. Madrid: Gredos. Luis, Carlos & María Amalia García Jurado. 1983. Desplazamiento fonético de vocales españolas. Letras de Hoje 17. 110.

26

Laura Colantoni and José Ignacio Hualde

MacLeod, Beth. 2007. Variable realization of vocalic sequences across Spanish dialects. Toronto: University of Toronto MA thesis. Manuel, Sharon. 1990. The role of contrast in limiting vowel-to-Vowel Coarticulation in different languages. Journal of the Acoustical Society of America 88. 1286–1298. Martín Butragueño, Pedro. 2002. Variación lingüística y teoría fonológica. México: El Colegio de México. Martínez-Gil, Fernando. 2000. La estructura prosódica y la especialización vocálica español: el problema de la sinalefa en ciertas variedades de la lengua coloquial contemporánea. In Juana Gil Fernández (ed.), Panorama de la fonología española actual, 511–560. Madrid: Arco/Libros. Moreno de Alba, José. 1994. La pronunciación del español de México. Serie de estudios de dialectología mexicana. México DF: El Colegio de México. Morris, Richard. 1998. Stylistic variation in Spanish phonology. Columbus, OH: Ohio State University dissertation. Navarro Tomás, Tomás. 1977. Manual de pronunciación española, 19th ed. Madrid: Centro Superior de Investigaciones Científicas. [1st ed., 1918]. Perissinotto, Giorgio. 1975. Fonología del español hablado en la ciudad de México: Ensayo de un método soiolingüístico. México DF: El Colegio de México. Quilis, Antonio. 1993. Tratado de fonética y fonología españolas. Madrid: Gredos. Recasens, Daniel. 1999. Lingual coarticulation. In William Hardcastle & Nigel Hewlett (eds.). Coarticulation: Theory, data and techniques, 80–104. Cambridge: Cambridge University Press. Simonet, Miquel. 2005. Prosody and Syllabification Intuitions of [CiV] Sequences in Spanish and Catalan. In Sonia Frota, Marina Vigário & Maria Jõao Freitas (eds.). Prosodies, 204– 207. Berlin: Mouton de Gruyter. Souza, Benjamin. 2010. Hiatus resolution in Spanish: An experimental study. The Pennsylvania State University dissertation. Vidal de Battini, Berta. 1964. El español de la Argentina. Buenos Aires: Consejo Nacional de Educación.

Appendix A – Stimuli list 1. 2. 3. 4. 5 6. 7. 8. 9. 10. 11.

Pelear Ideal Apear Ladear Ojear Real Mear Leal Fea Realismo Lealtad

‘to fight’ ‘ideal’ ‘to get off the horse’ ‘to lean towards one side’ ‘to look’ ‘real’ ‘to pee’ ‘loyal’ ‘ugly’ ‘realism’ ‘loyalty’

Conditions on front mid-vowel gliding in Spanish

12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41.

Beatitud Fealdad Beata Hojear Franquear Arrear Línea Desear Broncear Golpear Titubear Centellear Truchear Afear Asear Apeamos Emplear Recrear Pestañear Nuclear Deseamos Titubeamos Franqueamos Alinear Acarrear Contonear Creamos Seamos Veamos Leamos

‘beatitude’ ‘ugliness’ ‘blessed’ ‘to browse’ ‘to clear/to stamp’ ‘to drive a cattle’ ‘line’ ‘to wish’ ‘to tan’ ‘to hit’ ‘to hesitate’ ‘to shine’ ‘to copy/falsify’ ‘to make something ugly’ ‘to clean’ ‘we rustle’ ‘to use’ ‘to entertain’ ‘to blink’ ‘nuclear’ ‘we wish’ ‘we hesitate’ ‘we clear’ ‘to line up’ ‘to carry’ ‘to move’ ‘we believe, SUBJ’ ‘we are, SUBJ’ ‘we see, SUBJ’ ‘we read, SUBJ’

27

Ellen M. Kaisse

The syllabic position of glides in Spanish: insights from Pasiego vowel harmony 1 Abstract: The prevocalic glides of Pasiego trigger vowel harmony, motivating the conclusion by Hualde (1991) that they are onglides within the nucleus rather than the second member of a complex onset. Here I fill in an overlooked prediction of Hualde’s work: if Pasiego’s syllabification works like that of geographically adjacent, closely related Spanish dialects, only those prevocalic glides that are preceded by a consonant should trigger harmony. Absolute syllableinitial glides should lie in the onset and thus should not be harmony triggers. I gather evidence from the compendious descriptions of Pasiego by Penny (1969a,b). The monograph’s rich glossary of every Pasiego word encountered in Penny’s fieldwork on the dialect proves an invaluable resource. We learn that our prediction is correct – absolute syllable-initial glides do not trigger harmony. This leads us to the further conclusion that Pasiego harmony refers to the syllabic position of the trigger, not to its distinctive features.

1 Introduction This paper has three goals: first, to make a small contribution to our understanding of a vowel harmony process in Pasiego, a Northwestern Ibero-Romance dialect very closely related to Spanish; second, to use that new insight as an argument that absolute-syllable-initial glides in such dialects lie in the onset, while those preceded by consonants are in the nucleus; and finally, to consider the true nature of vowel harmony. Is harmony the spread of features from a syllabic nucleus to other syllabic nuclei (the syllable-head theory of van der Hulst and van der Weijer 1995) or is it the spread of a vocalic feature from anywhere in the syllable to a vowel, regardless of the syllabic position of trigger

1 My thanks to Patti Clark, Jim Harris, José Ignacio Hualde, Susannah Levi, Nathan Loggins, Ralph Penny, Yungdo Yun and two anonymous referees. An earlier version of this paper was delivered at the Encuentro international de lingüistica en el noroeste VI, Universidad de Sonora, Hermosillo, Mexico. Ellen M. Kaisse, University of Washington

30

Ellen M. Kaisse

or target, so long as the right vocalic features are present? (Clements and Hume 1995). Kaisse and Levi (2004) argue that Pasiego is best described as a nucleusto-nucleus spread while Turkish vowel harmony is sensitive only to the presence of a vocalic node. We will see here in some detail the reason why Pasiego harmony must rely on syllabic position – basically because one and the same underlying segment, /i/, with allophones [i], [j] and the palatal fricative [ ʝ], triggers vowel harmony only when it is demonstrably in nuclear position, while palatal consonants such as the palatal lateral [ʎ], which have all the right harmony-triggering features, but which can never occur in nuclei, do not trigger the regular harmony process. Hualde (1991) has made the elegant argument that because prevocalic glides in Pasiego trigger vowel harmony, we should conclude that they occupy a nuclear position, forming the ongliding portion of a rising diphthong. Here we fill in a previously unnoticed prediction of this argument: only those prevocalic glides that are preceded by a consonant should trigger vowel harmony. Absolute syllable-initial glides should not be expected to trigger harmony because in this environment they fill the onset, as Harris (1989) and Harris and Kaisse (1999) have established for Castilian and Argentinian Spanish.2

2 The structure of Spanish syllables with prevocalic glides 2.1 Glides in nucleus and glides in onset Consider Spanish words like [pje] ‘foot,’ [meˈɾjen̪da] ‘snack’ or [ˈbweno] ‘good.’ The first syllables of these words contain glides. But where are the glides in the 2 In this paper, as in Hualde (1989; 1991), we make the simplifying assumption that Pasiego constitutes enough of a ‘dialect’ of Spanish that arguments from the standard Castilian variety are in general transferrable to Pasiego syllable structure and vice versa. (José Ignacio Hualde has suggested to me that ‘co-dialect’ may be the better term, as it is more agnostic about the closeness of a relationship that is not quite so close as, for instance Argentinian Spanish and Castilian, the former being a recent New World development from the latter.) An anonymous referee points out that the question of the partially independent status of Pasiego with respect to standard Castilian is a difficult one to resolve, since Pasiego developed geographically near to and with continuing influence from Castilian. Not every argument given here for the syllabic structure of Spanish has been independently tested in Pasiego, to my knowledge. However, this paper’s central, non-obvious hypothesis, that syllable initial glides should not trigger vowel harmony in Pasiego, was engendered by facts from standard Castilian, suggesting the utility of understanding one variety to illuminate the other.

The syllabic position of glides in Spanish: insights from Pasiego vowel harmony

31

syllable? Do they form complex onsets with the preceding consonant, as in (1) below or do they form ongliding diphthongs with the following vowel as in (2)? We will adopt the position of Harris (1989) and Hualde (1991) that (2) is the correct structure when a consonant precedes the glide. (1)

incorrect structure for Spanish

(2)

correct structure for Spanish

Perhaps because ongliding (rising) diphthongs are relatively rare in the world’s languages, phonologists who do not specialize in Spanish have sometimes assumed that the Spanish glides are in a complex onset. Vago (1988), for instance, looking at the phenomenon of height harmony in Pasiego, adopted the structure (1) for Spanish. However, Hispanists have constructed several converging arguments that indicate that (2), the complex nucleus structure, is the correct one for Spanish. I will summarize these arguments in more detail in the next section. A convincing representative is that syllables of the form CGV are heavy for the purposes of the quantity-sensitive stress rule of Spanish (Harris 1983). Since we know that onset segments rarely figure in stress rules and that, in particular, CCV is light in Spanish when that second C is not a glide, we conclude that the Spanish postconsonantal onglide is in the nucleus. Hualde (1991) follows Harris (1983; 1989) in proposing that the glide in a Spanish CGV string is nuclear. But he offers a novel argument for this structure, based on his own 1989 article, which discussed varieties of vowel harmony found in the dialects of northwestern Spain. In Pasiego and other nearby dialects, the glide in a CGV string triggers vowel harmony; it causes the raising of preceding mid vowels just as syllabic high vowels do. In this article I investigate a prediction implicit in Hualde’s view of syllable structure which he did not pursue. Several phenomena (as presented most fully in Harris and Kaisse 1999)

32

Ellen M. Kaisse

argue that a prevocalic glide actually is in the onset if and only if there is no other segment present to fill that onset. That is, while the structure in (2), elaborated in (3) using the form [me.ˈɾjen̪.da] ‘snack’ is correct for a glide that has a consonant on its left and a vowel on its right (CGV ), the structures in (4) are correct for the consonantalized glides preceded by a vowel (VGV ) or by nothing (GV ), as in the Pasiego words [restɾaˈʝesa] ‘frolic’ and [ ʝo] ‘I.’ The terminal symbols in the tree are [j], a true glide, and [ ʝ]3, a voiced non-strident palatal fricative, the consonantalized allophone of the glide that appears in syllable-initial position. (3)

(4)

a.

b.

3 The symbol [y] is typically used by Hispanists for IPA [ ʝ]. Penny (1969b) uses [y] as a cover symbol for the more detailed transcriptions in his monograph (1969a), which distinguishes between affricate [ŷ] (which, as in Castilian, tends to occur word-initially and after a nasal) and less fully constricted fricative variants of this consonantalized glide. The precise realization of the [y] (our [ ʝ]) does not affect the argument here, and Penny’s (1969b) use of the cover symbol in an article on vowel harmony indeed suggests it should not. For a detailed description of the variants of the palatal glide in Castilian and Argentinian dialects, see Harris and Kaisse (1999) and references therein.

The syllabic position of glides in Spanish: insights from Pasiego vowel harmony

33

In other words, glides are accommodating. They fill onsets if there is no better, less sonorant candidate around to do so; otherwise, they are parsed into nuclei.4 In the rest of this paper, we refer to the environment for glides in (3) as postconsonantal, while those in (4) are called syllable-initial, or when needed, nonpostconsonantal, since they are preceded by a vowel or by nothing. Before moving on to the data from Pasiego, let’s review some of the arguments that Spanish phonologists have amassed for the differential placement of postconsonantal versus non-postconsonantal glides. All of these are explained in greater detail in Harris and Kaisse (1999).

2.2 Some arguments for the position of Spanish postconsonantal vs. syllable-initial glides 2.1.1 Stress Spanish has a 3-syllable stress window, with stress allowed to fall on the ultima, penult or antepenult. If the penult is heavy, however, stress cannot fall on the antepenult (a still-productive inheritance of the Latin stress system). In this regard, a rhymal consonant, an offglide, or, critically for us, an onglide (Harris 1983:88), are all stress-attractors and outlaw antepenultimate stress. There are words like [sa.la.ˈmaŋ.ka] ‘Salamanca’ and [be.ne.ˈswe.la] ‘Venezuela,’ but none like *[sa.ˈla.maŋ.ka] or [be.ˈne.swe.la], and work with many consultants by

4 Other languages exhibiting this kind of accommodating syllabic positioning of prevocalic glides are Korean (Yun 2004) and Slovak (Rubach 1998). Colina (this volume) also subscribes to the position that postconsonantal glides are nuclear while syllable-initial glides are in onset. Martínez-Gil (this volume) demonstrates that the syllabification generalizations discussed here only hold true lexically. Like many languages, Chicano Spanish creates additional onsets postlexically, when vowels are desyllabified in cross-word hiatus. Two of his examples will illustrate this process: (i) /ˈuno # u # ˈotɾo/ → [ˈu.no.ˈwotɾo] ‘one or the other’ (ii) /mi # uniˈfoɾme/ → [mju.ni.ˈfóɾ.me] ‘my uniform In such cases the glide typically does not harden (i) and he argues, is placed in the onset, even after a consonant (ii). In the same paper, Martínez-Gil also casts doubt on some of the traditional arguments for post-consonantal glides invariably lying in the nucleus lexically, but does not intend to dispose of all of them, particularly the ones cited here for the structural contrast of postconsonantal versus syllable-initial lexical glides. His contributions are thus largely complementary to and not incompatible with the position I espouse in this paper.

34

Ellen M. Kaisse

Harris (1983) and by Harris and Kaisse (1999) confirms that nonce words with CGV in the penult are judged ill-formed if they have antepenultimate stress. However, both nonce-word probes and existing lexical items confirm that a glide in the penult does not prevent antepenultimate stress if that glide is syllableinitial. (Such an underlying [‑consonantal] segment would have the consonantalized allophone [ ʝ] or [ɣw].) Thus there are existing words like [ˈkon.ʝu.xe] ‘spouse’ and foreign place names like ‘Iowa’ [ˈa.ʝo.ɣ wa]5; additionally nonce words like [ˈta.ʝa.ma] are readily accepted. We should also add that the actual consonantalized version of the glide is not important in these judgments. Some dialects and some speakers are less consonantalizing than others. The realization is quite variable over speakers and dialects, and a non-consonantalized [j], when it occurs, behaves just like a consonantalized one.6

2.1.2 The three segment limit on rhymes Harris (1983:15) established that rhymes in Spanish contain at most three segments. So there are forms like [ˈfweɾ.te] ‘strong,’ [ˈtɾjuɱ.fo] ‘triumph’ and [ˈklaws7.tɾo] ‘cloister’ whose first syllables contain the rhymes [weɾ], [jun], and [aws], but there are no words like *[fweɾs.te], with the rhyme *[weɾs]. What is most relevant to us is that word-initial or intervocalic glides do not count towards the 3-syllable limit on rhymes.8 There are no words like *[njuks], but the word [juks.ta.po.ˈneɾ] ‘to juxtapose’ is well-formed because the [j] has been recruited to fill an onset that would otherwise be empty, leaving the well-formed rhyme [uks]. Harris and Kaisse (1999) give a few more arguments for the differential syllabic positions of syllable-initial and postconsonantal glides, but the two above should

5 Thanks to José Ignacio Hualde for this lovely example. 6 Argentinian Spanish is shown by Harris and Kaisse (1999) to have a rare phoneme /i./, which behaves differently from normal /i/ in refusing to lose its syllabicity next to another vowel in many cases, in counting as a vowel for stress calculations, and in having a different consonantalized allophone from normal /i/ when it is forced into an onset late in the word-level phonology. Argentinian speakers do not accept antepenultimate stress when this rare segment is in onset on the surface. They therefore show a contrast between acceptable [ˈta.ʒa.ma] (from underlying /taiama/, [ʒ] being the Argentinian variant of [ ʝ]) and unacceptable [ˈta.ja.ma] from underlying /tai.ama/. (Harris and Kaisse 1999:149.) 7 The [s] must be in coda, since s+consonant is a robustly outlawed onset in Spanish. 8 Of course, glides which follow a vowel and precede a consonant, as in [ˈaw.xe] ‘apex’, must be parsed into the rhyme, since they cannot form an onset cluster with a following consonant. Spanish two-member onsets all consist of an obstruent followed by a liquid. (Harris 1983:13.)

The syllabic position of glides in Spanish: insights from Pasiego vowel harmony

35

be sufficient for our purposes. We shall now add to this list the differential behavior of postconsonantal and syllable-initial glides in Pasiego vowel harmony. The former trigger harmony, as Hualde (1989), has already discovered. But, we shall see, a close reading of Penny’s (1969a) glossary and of his article on vowel harmony (1969b) show that the latter do not.

3 Vowel harmony in Pasiego The Cantabrian mountains in Northwestern Spain are the home of several dialects which exhibit phonological processes that differ from those found in Standard Castilian Spanish. Pasiego is the name given to the dialect of Montes de Pas, meticulously documented by Penny (1969a) in a study of four villages located in the southeastern Cantabrian region of the Spanish province previously known as Santander. In addition to the primary description, Penny (1969a) (referred to as EHP (El habla pasiega) in the examples below), we have a more theoretical article on its vowel harmony processes by the same author (Penny 1969b; VH in the examples) and later articles on its vowel harmony (Hualde 1989, McCarthy 1984, Vago 1988.) Penny’s book is invaluable, as it contains an index of over 6000 words, allowing one to investigate questions not directly addressed in his text. He also meticulously records variants, cases where a word has more than one possible realization. Pasiego, like other neighboring dialects of Northwestern Spain, has at least two harmonic processes involving vowels. One, which deals with the feature [tense] or perhaps [advanced tongue root], will not concern us here. The other, a height harmony, takes the following form in Pasiego: (5)

A stressed high vowel spreads [+hi] leftward to all mid vowels.

Notice that in Pasiego, unlike other dialects of Northwestern Spain, the spread is only from stressed vowels onto unstressed vowels (suggesting, for Hualde 1989, that the domain is the metrical foot.) Consider the simple example in (6). The stem /beb-e/ ‘drink-theme vowel’ appears with its underlying mid vowels in the infinitive [beˈβ-e-ɾ] ‘to drink’ but with raised root and theme vowel before the conditional suffix [-ˈɾia], [biβ-i-ˈɾia] (6)

36

Ellen M. Kaisse

Low vowels are transparent to harmony.9 (7)

po el kaˈminU → [pu il kAˈmInU]10 along the path

‘along the path’

The domain of harmony, as we shall see, is approximately XLEX] (Selkirk 1986) – what Penny (1969b) calls a syntagma; harmony spreads [+hi] leftward not only within a lexical content word but also onto articles, clitics, conjunctions and prepositions to the left of that content word ((7), (8g–i)). Notably, harmony is not restricted to adjacent syllables. As with most harmony systems, all eligible vowels throughout the word come to agree with the rightmost stressed high vowel. (8) contains a fuller set of relevant examples. Forms showing raising via vowel height harmony are bolded. ‘to drink’

biˈβiːs

b. beˈβamus

‘we drink’ (subj.)

biˈβia

‘he was drinking’ (indic.)

c. beˈβemus

‘we drink’ (indic.)

biβiˈɾia

‘he would drink’ (condit.)

d. ˈbeβo

‘I drink’

biˈβi

‘I drank’

kumiˈɾia

‘he would eat’ (condit.)

(8) a. beˈβeɾ

‘you (pl) drink’ (indic.)

e. koˈmeɾ

‘to eat’

f. koˈxajs

‘you take’ (2ppl subj.) kuˈxíːs

‘you take’ (2 p. pl indic.)

g. elˈpan

‘the bread’

‘the lamb’

kUn Il maˈʝIstɾU11 ‘with the teacher’

h. [see note 11]] i. po laˈkaʎe

il kUɾˈðIɾU

‘down the street

pU (I)l kAˈmInU

‘along the path’

9 There is ample evidence that [+hi] spreads, but really none that [-hi] spreads, pace McCarthy 1984. I believe McCarthy was misled by forms like the 2.p.pl subj, sint-ais, vs. 1p.pl. indic. sent-emus ‘sit’. McCarthy took these forms to indicate that the verb has an underlying /i/ in the root, since the coda glide in the affix is generally agreed not be a harmony trigger. However, the height alternation illustrated here is also present in the standard dialect and thus is not due to a synchronic vowel harmony process but rather to a much earlier, pan-Hispanic process of raising. The underlying form of the Pasiego and the standard verb is /sentir/ and the surface [e]’s are thus not the result of lowering in the neighborhood of a stressed mid vowel. 10 Capital letters indicate lax vowels (I follow Hualde’s transcription practice), generated via the tense/lax harmony system, which we are ignoring as not relevant for to the purposes of the current article. 11 Penny cites [kun] as an example of raising, suggesting its underlying and unraised form is /kon/, as in Std. Spanish, but I am unable to locate an example of it in non-raising environment to provide for comparison.

The syllabic position of glides in Spanish: insights from Pasiego vowel harmony

37

Crucial to the argument at hand is the fact, noted by Penny, Hualde, and McCarthy, that a non-syllable initial onglide in stressed syllable also causes raising (data from Penny 1969b:153–4)

(9)

a.

aɱfeˈstaɾ

‘to infect’

aɱfiˈθjon

‘infection

b.

ˈbeɾðə

‘green’

biɾðiˈɡjaɾ

‘to turn green’

c.

meɾen̪ˈdeɾ

‘to snack’

miˈɾjen̪dA

‘a snack’

d.

soɾˈβeɾ

‘to suck’

suɾβiˈtjaɾ

‘to snuffle’

e.

ˈmenus

‘less’

miˈŋɡwaɾ

‘to lessen’

f.

koˈxeɾ

‘to take’

kuˈxjo

‘he took’

g.

me lo komˈpɾo

‘he bought it for me’

mi luˈðjo

‘he gave it to me’

h.

se kaˈso

‘he got married’

si kuˈmjo

‘it’s been eaten’

i.

lo ˈmalu

‘the bad thing’

lu ˈpjoɾ

‘the worst’

j.

el ɡaˈnau

‘the cattle’

Il ˈmjEU

‘the fear’

k.

(see (14c) for unraised [ne])

ni12 ˈlunu ni ˈlutɾU

‘neither one nor the other’

We do not need to modify our harmony rule: [+hi] is still spreading from a stressed syllable: (10)

Hualde asks us to note that while onglides do trigger harmony, offglides do not. The 2ppl form of ‘grasp’ is [koˈxajs], not [kuˈxajs]. By this logic, Spanish has only rising (ongliding), not true falling (offgliding) diphthongs, since a true diphthong has both vocoids in the nucleus proper.13 Compare, for instance, [suɾ.βi.ˈt ̪jaɾ] (9d) with [ko.ˈxajs] (8f): [ˈja] in a stressed syllable causes raising of the vowels to its left, but [ˈaj] does not. Therefore, Hualde (1991) concludes, postvocalic glides are not in the nucleus (11a), but rather in the coda (11b).

12 The underling form of ‘neither’ is /ne/ in this dialect. It appears as [i] only in raising contexts (EHP:146). 13 Hualde (1989:803 note 24) cites the sole form [pUl ˈpɾAw] (underlyingly /po el pɾau/ ‘by the meadow’) as possibly showing raising of the /o/ of /po/ by a postvocalic glide. However, all other examples of postvocalic glides fail to trigger raising harmony – this case is isolated. It is also complicated by the regional differentiation of count nouns, with underlying suffix /-u/ and mass nouns with /-o/.

38 (11)

Ellen M. Kaisse

a.

b.

In such a position they cannot spread the feature [high]. This novel argument for offglides filling coda position in Spanish can be added to the previous one implicit in Harris (1989): a nuclear vowel can be followed by either a glide (plus an optional appendix [s]) or a consonant (plus an optional appendix [s]), but not by both a glide and a consonant. Thus we find ([lej] ‘law’ and [sejs] ‘six’ but not *[lejɾ]; and within words we find [peɾ.so.ˈnal] ‘personal,’ and [peɾs.pi.ˈkas] ‘perspicacious’, but not *[peɾn.so.ˈnal]). The distribution strongly suggests that postvocalic glides and postvocalic consonants fill the same slot in the syllable structure, the single coda position licensed after that vowel. (/s/ is appended in an additional appendix slot.) At the heart of Hualde’s argument is the intuition that only segments that lie in the nucleus can trigger vowel harmony. Hualde points out that only stressed vowels trigger harmony in Pasiego (and in the neighboring dialect of Tudanca (Penny 1978)). Under the assumption that stress is assigned to nuclei, a glide must lie in the nucleus to receive stress. This is a sensible assumption, but it merits some discussion. Indeed, we probably want to say that even in the many languages where unstressed vowels can trigger harmony, their nuclear position is crucial. An explicit and general statement of this idea and a hypothesis as to why it should be so is found in van der Hulst and van de Weijer’s (1995) syllable-head approach to vowel harmony. This theory attempts also to deal with the apparent non-locality of vowel harmony systems, which typically skip over intervening consonants and can affect strings of vowels over many syllables. According to van der Hulst and van de Weijer, vowel harmony is a head-to-head assimilation process. Features from the nucleus, that is the head, of a syllable percolate up to the maximal projection of N – notated as N’’ rather than as σ to parallel the X-bar notation of syntax – whence they can travel to the next N’’ and filter down again to the head of the syllable. Thus only elements in N can be the trigger or the target of vowel harmony. (12) illustrates the workings of this theory: the first syllable has a nucleus which is [+F]. That feature percolates up to the maximal projection N’’ (the syllable node) and thence rightward to the second and third syllables, as indicated by the dotted lines. [+F] then filters down to the nuclear segments of the second and third syllables.

The syllabic position of glides in Spanish: insights from Pasiego vowel harmony

(12)

39

(adapted from van der Hulst and van de Weijer 1995:509)

There are some potential problems with this theory, as pointed out perspicaciously in Levi (2000, 2004, 2008), as it cannot deal with languages such as Turkish, where certain consonants, such as the palatal lateral [ʎ], may trigger vowel harmony. We will see in section 6 that Pasiego, as described in prose in Penny (1969b) might at first glance seem to raise similar difficulties for the syllable-head theory. However, a careful survey of Penny’s index of words suggests that palatal consonants in Pasiego do not trigger true long-distance harmony, nor do they even trigger local raising of adjacent vowels in any regular fashion. Thus we adopt the syllable-head theory here, as it covers the basic data of Pasiego height harmony very well and accurately expresses Hualde’s theoretical assumptions.

4 The new observation: syllable-initial [+hi] segments do not cause raising 14 Before introducing this new data, it will be helpful to briefly discuss the underlying high vocoids of Spanish. Harris and Kaisse (1999) show at some length

14 The examples in this section were found either in Penny’s more theoretically oriented article on vowel harmony (1969b) or by searching (at random places in the alphabet) about a third of the approximately 6600-word index of Penny (1969a). Pasiego is a ‘lleista’ dialect, meaning that /ʎ/ is a phoneme and does not usually fall together with the syllable initial [ ʝ] that comes from /i/. This limits the possible examples with [ ʝ]. (Actually, a close reading of Penny’s index yields several forms with [ ʝ] that have [ʎ] in related derivatives, as well as forms that consistently show [ʎ] but which have [ ʝ] in Standard Castilian. Pasiego, then, is lleista in the sense that it has a phoneme /ʎ/, but it does not always present that phoneme in forms where we might expect it etymologically, and it allows [ ʝ] and [ʎ] to alternate as variants of the same underlying segment in certain morphemes.)

40

Ellen M. Kaisse

that [i], [j] and [ ʝ]15 are all predictable allophones of a single phoneme, /i/, which is unspecified for syllabicity but has the features [‑consonantal, +hi, ‑bk]. The same is true for the high back segments: [u], [w] and [γw]: all derive predictably from /u/. Using the front vocoids to exemplify this distribution, we can see that [i] appears when there are no adjacent segments of greater sonority to form a nucleus, as in the word [ˈpi.so] ‘floor’. [j] appears either pre- or postvocalically, and lies in the rhyme, as we argued in section 2, as in the words [pje] ‘foot’ and [lej] ‘law’. [ ʝ] appears when /i/ must fill an onset, as in [ ʝo] ‘I’ or [ˈle.ʝes] ‘laws’, the plural of [lej]. [ ʝ] is a weakly articulated, voiced, non-strident fricative, a slightly consonantalized version of the glide [j]. (For careful descriptions see Navarro Tomás 1965 and phonetic sources cited in Harris and Kaisse 1999.) Its features remain those of /i/, that is, of [i] or [j]: it is [+hi, -bk], like all palatal consonants in the feature system Harris and Kaisse adopt, but in addition, it may be [+consonantal] or not, depending on how much fortition it has undergone. In many dialects, its fortition from [j] is optional and gradient, and cross-dialectally, speakers are generally unaware that it differs from [j]. This can be observed in Pasiego in the phenomenon of mis-segmentation that Penny (EHP:94) observes: (13)

las ˈsjenas the temples

>

Pasiego

las ˈʝenas the temples

I dwell on the minimal featural and phonetic difference between [j] and [ ʝ] because, as we are about to see, only the former, being in nucleus, can trigger harmony. But I will argue that it is not the allophonic consonantality of [ ʝ] that accounts for its inability to trigger harmony. Rather, it is its position in onset, as the syllable-head theory leads us to expect. We will return to other possible but less satisfactory explanations for the inertness of [ ʝ] in section 5. We can now move on to the new data. The clearest examples of failure to raise before a [+hi] segment in a stressed syllable, that is, before [ ʝ], are found in the inter-word environments. Recall from (9g–k) that monosyllabic function words are prime targets for raising. Nonetheless, there is no raising before a Also limiting the number of possible relevant test cases is the obvious fact that words such as [ˈʝUɤU] ‘yoke’ with a consonantalized glide starting a stressed syllable that itself contains a high vowel cannot be used as evidence for or against raising by syllable initial consonantalized glides; the stressed high vowel itself is a sufficient condition for raising. 15 As mentioned earlier, I am slightly oversimplifying this allophony. Many dialects, including Pasiego, have an affricated allophone found most often phrase-initially or after a nasal. As this allophone is obviously [+consonantal] and in onset, I harmlessly collapse it here with the fricative allophone.

The syllabic position of glides in Spanish: insights from Pasiego vowel harmony

41

word whose first segment is a glide that opens a stressed syllable. Penny cites these cases as examples of failure to raise within the very section in which he is discussing cases where raising does occur, so there is no danger that he merely omitted mention of raising occurring in them: (14)

a. b. c. d.

se16 ˈno se ˈʝo lo ˈaɣu ne17 ˈʝo el ˈʝelsu

‘if not’ ‘if I do it’ ‘nor I’ ‘the plaster’

(VH p155) “ “ “ “ (EHP p152)

(rather than (rather than (rather than (rather than

*si ˈno) *si ˈʝo) *ni ˈʝo] *il ˈʝelsu]

Compare these examples with raising caused by a postconsonantal glide (9). For instance, the definite article [el] in (14d) is to be contrasted with its raised allomorph [il] in (9j) [il ˈmjEU] ‘the fear’ from /el mieu/. This failure to raise before the consonantalized glide is all the more striking since Penny (1969b) notes that there is sporadic raising before [ʎ] and [ ɲ], the underlying palatal consonants of the dialect. We come back to the contentious question of raising before these consonants at the end of the paper. Examples of lack of raising before a syllable initial palatal fricative can also be found within words. Pasiego regularly inserts a consonantalized glide intervocalically, as in the forms corresponding to the Std. Spanish verb [kɾe-ˈeɾ] ‘to believe.’ (15) contains the present tense paradigm. (EHP:121). Notice that the stem vowel is perfectly capable of undergoing raising, as shown by the 2ppl form [kɾiˈʝis], but it requires a stressed suffix vowel to cause raising. The [ ʝ] initiating a stressed syllable is not a trigger. Also notice the alternate dialect 2ppl form [kɾeˈʝejs], whose offglide does not cause raising, in accordance with Penny’s description and Hualde’s conclusion that offglides are not in the nucleus. (15) ˈkɾeʝU ˈkɾejs ˈkɾeʝə

kreˈʝer 1psg kɾeˈʝemus 2psg kɾiˈʝi:s 3psg ˈkɾeʝən

‘believe-inf.’ 1ppl 2ppl (alternate dialect form kreˈʝejs) 3ppl

Other verbs like this include [ɡoˈʝ-eɾ] ‘to hear’ and [beˈʝ-eɾ] ‘to see,’ which show no raising in the present tense paradigm save in the 2ppl. Actually, intervocalic 16 The underlying form of ‘if’ in this dialect is /se/, and it appears as [si] only in raising contexts. (EHP: 146). (Standard Spanish has /si/.) 17 As mentioned under (9k), [ne] comes from /ne/ in this dialect, unlike the /ni/ of Standard Spanish.

42

Ellen M. Kaisse

insertion is only one facet of non-etymological [ ʝ] or [j]: Penny (EHS: 69–70) states that, like many Leonese18 dialects, Pasiego has a strong propensity to add front glides. These non-etymological glides do cause raising when they are postconsonantal, as in [aθuˈljaɾ], (cf. Std. Spanish [aθoˈlaɾ]) ‘to dress timber with an adze.’ A final unproblematic example is: (16) apoleˈʝaɾ (EHP:291) ‘to support (a wall) (Std. Spanish apoˈʝaɾ) Once again, the mid vowels preceding a [ ʝ] that initiates a stressed syllable do not undergo raising.19 Several of the following examples also probably show the inability of [ ʝ] to cause raising. However, I cite them with caution, because there may be alternative explanations for their non-high vowels preceding [ ʝ]; they are therefore merely consistent with our hypothesis, no more. Penny (1969a:67; 1969b:154) points out that r can have a sporadic lowering effect on vowels. I am not terribly worried because the effect is typically for /e/ to lower to [a] after [ɾ], while my examples maintain /e/, neither lowering it after [ɾ] nor raising it before [ˈʝV.] Thus one can probably add the forms [reˈʝeɾβa] (17a) and [retɾaˈʝesə] (17b) to our cast of supporting examples, but with a small grain of salt. Similarly, Penny states that prefixes ending in [s] show a range of front vowel heights before the [s], casting some doubt on the significance of [estɾaˈʝesə], whose morphological makeup I do not know but which could conceivably have the prefix es-, a descendent of both the Latin prefixes *ex and *dis in Pasiego. Finally, Penny (1969a:67) tells us that nasals tend to lower a preceding high vowel, casting doubt on the usefulness of (17e) [empaˈʝaɾ]. Recall that the low vowel is trans18 Leonese is the larger Northwestern dialect area that includes the Cantabrian dialects to which Pasiego belongs, but also some dialects spoken in Northern Portugal and in Galicia. 19 In my close reading of Penny’s index of words, I did find a few cases where a [ ʝ] occurs in a word with a preceding high vowel that corresponds to a mid vowel in Std. Castilian Spanish. However, virtually all of these have independent potential causes for raising which have nothing to do with the [ ʝ]. Some have a stressed high vowel in addition to the [ ʝ] like [Il ˈʝUγU] ‘the yoke.’ Others are forms which have an unstressed high vowel but where many related forms in the same paradigm or closely related words have a stressed high vowel. It seems most likely that the raised vowel is present by analogy to the related forms where the stressed high vowel caused regular vowel-harmonic raising. Thus, for instance, the verb [d̪es.kus.tiˈʝaɾ] ‘to hit in the ribs’ corresponds to Std. Cast. [des.kos.tiˈʎaɾ]. It is derivative from the Pasiego noun [kus.ˈti.ʝa] ‘rib’ with stressed [ˈi], and the majority of forms of the verb have stressed [ˈi] ([kus.ˈti.ʝa] 3sg. ind. pres; [kus.ˈti.ʝen] 3pl. subj. pres.) etc.

The syllabic position of glides in Spanish: insights from Pasiego vowel harmony

43

parent to harmony, so the [a] in (17b–d) should not be an impediment to raising if raising were really triggered by a syllable-initial palatal. (17)

a. b. c. d. e.

reˈʝeɾβa (EHP:307) restɾaˈʝesə (EHP:264) retɾaˈʝesə (EHP:350) estɾaˈʝesə (EHP:264) empaˈʝaɾ (EHP:235)

‘cramp (stomach)’ ‘frolic’ ‘resemble’ ‘cough of an animal’ ‘put into a silo’

(rather (rather (rather (rather (rather

than *riˈʝeɾβa) than *ristɾaˈʝesə) than *ritɾaˈʝesə) than *istɾaˈʝesə) than *impaˈʝaɾ)

In every case that I found where an underlying mid vowel is followed by a stressed syllable beginning with the [+hi] segment [ ʝ], there is no raising. Previous arguments in the literature on Spanish (see section 2.2) already predisposed us to expect this, as underlying [‑consonantal] segments must be parsed into onsets. And according to the syllable-head theory of vowel harmony, onsets cannot trigger harmony. On the other hand, when the [+hi] segment is in the nucleus of a stressed syllable, either as the onglide [j] or the vowel [i], it regularly triggers harmony.

5 Feature-geometric vs. syllable-head accounts of vowel harmony We have argued that the height harmony system of Pasiego is best understood through the syllable-head (nucleus-to-nucleus propagation) theory of van der Hulst and van der Weijer (1995.) This is because – high vowels and postconsonantal onglides trigger harmony – onsets and offglides do not trigger harmony. Offglides are not in the nucleus in Spanish – they fill the coda position that is otherwise occupied by a consonant – the process is sensitive to the stressed status of the trigger, and stress is a property of nuclei – Palatal consonants (which perforce cannot be in nucleus, as Spanish has no syllabic consonants) do not participate in the harmony process, even though they bear the crucial harmony feature [+hi]. This last point requires some perusal of the data, and we will return to it in the next section. Other explanations for the failure of [ ʝ] to trigger harmony can be imagined but they are not satisfying and do not yield a unified account. For instance, it might be tempting to think that the added consonantality that /i/ achieves when

44

Ellen M. Kaisse

placed in an onset could prevent it from being a harmony trigger. But glides in coda do not trigger harmony, and they are certainly not consonantalized. Moreover, [ ʝ], a palatal fricative, remains [+hi], so the harmonizing feature is still present. And the consonantalization of the vocoid has such a phonetic flavor to it, being gradient, allophonic, and weak, that it is hard to credit it interfering with a grammaticized, lexical process like vowel harmony. In a stratal theory, one would place the syllabification of the underlying vocoid into onset as one of the first processes of the early lexical stratum. Consonantalization is probably postlexical, however. But let us consider why there is a robust alternative theory of vowel harmony, one which does not refer to position in the syllable but rather to the existence of a vocalic node. According to Clements and Hume (1995), any segment can trigger vowel harmony, so long as it has vocalic place features. This would include palatalized or palatal segments (18), vowels (20), and some surface glides, which are positional allophones of vowels, such as those in Spanish. Levi (2004; 2008) shows that, like Spanish, most languages have only what she calls underlying vowels – i.e. vocoids that have a vocalic place node and that show up as vowels or as glides on the surface depending on the sonority of surrounding segments. But some languages, including Turkish, have both underlying vowels and underlying glides. Underlying glides have only a consonantal place node (19). These glides do not participate in harmony cross-linguistically, and they do not alternate with vowels. They cannot appear in nucleus, so, when necessary, they trigger epenthesis of vowels rather than showing up as vowels themselves. (18)

feature geometry for [ʎ] or other palatal or palatalized consonant20

20 Levi (2004) ultimately opts for a different feature geometry than the one proposed in Clements and Hume (1995), namely that of Halle, Vaux and Wolfe (2000). The details of the geometry are not important here, but rather the fact that for harmony systems like Turkish, one must rely on the presence of features of a certain sort in trigger and target rather than on the nuclear status of the node dominating trigger or target.

The syllabic position of glides in Spanish: insights from Pasiego vowel harmony

45

(19)

feature geometry for /j/ (a true underlying glide, as in Turkish)

(20)

feature geometry for an underlying vocoid /i/ which may appear as [j] or [i]

Levi (2004; 2008) shows that the syllable-head theory of van der Hulst and van der Weijer cannot explain the behavior of vowel harmony in Turkish. Turkish backness harmony is diametrically opposed to Pasiego in two regards – Turkish glides do not trigger vowel harmony while Pasiego’s do – Turkish palatalized or palatal consonants do trigger vowel harmony while Pasiego’s do not In Turkish, the palatal lateral [ʎ] and the fronted dorsals [kj, ɡʲ] block the spread of backness features from preceding vowels and initiate harmonic domains of their own. However, the glide [j] acts like most consonants – it neither blocks nor triggers harmony. (21)

a.

jok

‘there is not’

b.

koj-u

‘cove-acc’

The back vowels in (21) co-exist happily with the front glide. Levi’s proposal is that segments that have vocalic place nodes trigger vowel harmony, wherever they may lie in the syllable. Both [ʎ] and [j] may be in an onset or coda in Turkish, but only [ʎ] is a vowel harmony trigger because, like all palatal consonants, it has a secondary palatal articulation under the vocalic node, while [j] has no vocalic node. The absence of this node also explains why Turkish [j]

46

Ellen M. Kaisse

never occupies a nuclear position, triggering epenthesis when a vowel is required, rather than serving as a vowel itself. Kaisse and Levi (2004) speculate that languages like Pasiego, where stress figures critically in the environment of harmony, may be more likely to have harmony systems that function in accord with nucleus-to-nucleus propagation. This is because nuclei are the bearers of stress. Languages like Turkish, with their stress-free harmony systems and contrasts between palatal and non-palatal laterals and dorsals, may be more prone to having harmony systems that learners will associate with triggering from a vocalic place node.

6 Do palatal consonants trigger raising in Pasiego? It is important to understand the behavior of palatal consonants with respect to raising of vowels, since this behavior impinges on whether the proper explanation for the failure of [ ʝ] to cause raising stems from its position in the onset, as I have argued. If it should turn out to be the case that the Pasiego palatal consonants cause raising, even though they clearly are syllabified as onsets, we would have to seek another explanation for the inertness of [ ʝ]. It would not be easy to describe a system where vowels, palatal nasals and liquids, and onglides were harmony triggers, but the palatal fricative [ ʝ] was not. The data from the glossary of Penny (1969a) is interestingly murky. As I will detail below, it appears that there may be some incipient harmony-like raising of preceding mid vowels initiated by the palatal sonorant consonants /ʎ/ and/ ɲ/, but true long-distance vowel harmony is never or very rarely triggered by these segments, and even local raising of an adjacent mid vowel followed by a palatal lateral or nasal has at least as many exceptions as it has exemplars. Penny (1969b:153) groups together raising before (i.e., triggered by) ‘yod’ and ‘wau’ (that is, non-consonantalized [j] and [w] in stressed syllables), and raising before the palatal consonants /ʎ/ and /ɲ/, noting however that lack of raising before the consonants does occur. However, I believe he is wrong to conflate these two kinds of cases. All of the raising examples he cites involving yod and wau are, not surprisingly, post-consonantal cases, of the type we saw in (9). This is as we expect, since [j] and [w] do not appear absolute syllable-initially in Penny’s works, which contain, unremarked upon, the examples we have cited of consonantalized, syllable-initial glides which do not trigger raising. But if the palatal consonants truly do cause the same kind of raising as nuclear vocoids, our explanation of the differential behavior of syllable-initial (putative onset)

The syllabic position of glides in Spanish: insights from Pasiego vowel harmony

47

vocoids and nucleus-internal ones falls apart. Syllable-initial [ ʝ] is just as palatal as onset [ʎ] and [ɲ] – should it not cause raising just like they do? I returned to Penny’s (1969a) invaluable glossary to see how the palatal consonants really do behave. We must keep in mind that the hallmarks of true vowel harmony in Pasiego are long distance spread, transparency of the low vowels, and the requirement that the triggering feature lie in a stressed syllable. These criteria hold true for harmony triggered by high vowels, and they also hold true for harmony triggered by postconsonantal glides. In other words, the glide in [CGV] does not trigger harmony if the vowel is unstressed. Let us determine just what the facts of palatal-triggered harmony are. The consonantal environment is really only mentioned in passing by Penny; vowel harmony is the focus of his (1969b) article. What emerges from a closer look is this: while Penny is right in noting raising triggered by these consonants in his own data, he has not realized that that consonant-induced raising, unlike vowel harmony, is very sporadic and does not have the same stressed syllable environment, or the same long-distance effect, as the well-established vowel harmony process.21 McCarthy (1984) notes many exceptions like [seˈɲor] ‘sir’ and decides raising before [ɲ] and [ʎ] is just a minor tendency, but I do not believe he did a thorough count in Penny’s glossary. After a thorough sampling, I conclude that what Penny most likely noted was a change in progress, perhaps a generalization of the environment of height harmony to include palatal consonants, but not one which was a regular part of the phonology at the time of his fieldwork. Skeptical about the likelihood of a vowel harmony process triggered by palatal sonorants but not by the derived fricative, [ ʝ], and finding only a few examples cited in Penny’s article, I read about one third of the approximately 6600 entries in the glossary to his book, sampling at various parts of the alphabet. (22) summarizes my findings. (22) a. ɲ or ʎ initiates stressed syll; raising occurs

b. ɲ or ʎ initiates stressed syll; no raising

c. ɲ or ʎ initiates unstressed syll; raising occurs

d. ɲ or ʎ initiates unstressed syll; no raising

9

12

3

18

21 I was only able to find one case where the feature [hi] spread from a palatal consonant over a long distance or through a transparent low vowel, the hallmarks of true vowel harmony in Pasiego. uɾðiˈɲaɾ cf. Std. Castilian oɾðeˈɲaɾ ‘to milk’ But even this form is unstable, occurring also as [aɾðoˈɲaɾ].

48

Ellen M. Kaisse

(22) shows that I found 42 examples of words containing a palatal sonorant preceded by an etymological mid vowel somewhere earlier in the word. Since height harmony in northwestern Spain is an innovation, I used Standard Spanish as a check on the conservative value of the vowel. In the few cases where Penny did not gloss the form with a closely related standard form or where I could not find a standard cognate myself, I did not include a form with a high vowel in a word with a palatal in any column, since I could not tell if that high vowel was the output of harmony or was etymological. I also excluded all cases with a derived high vowel that could have been raised by a stressed high vowel somewhere in the word. These decisions may have prejudiced the results slightly in favor of non-raising before palatal consonants, as I did count all instances of unstressed mid vowels coexisting with palatals as instances of non-raising. Our numbers are obviously rough, then, but the trends are evident. If consonant-induced raising resulted from the same process as vowel- and glide-induced raising, we would expect column (22a) to have an entry of 21 and column (22b) to have 0. All the cases where the consonant starts a stressed syllable (a+b) should have undergone raising. An example from column (a) is (23) suˈɲar ‘dream’ (cf. Std. soˈɲar) But we actually find a slight preponderance of cases where the etymological mid vowel remains (22b) (24)

a.

dezɣɾeˈɲau ‘uncombed’ (cf. Std. dezɣɾeˈɲaðo);

b.

desoˈʎaɾ ‘to skin’ (cf. Std desoˈʎaɾ)

These facts argue against consonant-induced harmony as a regular process or one that is due to the same rule as vowel harmony triggered by high vowels and glides. Similarly, if consonant-induced raising required that the palatal trigger belong to a stressed syllable, we would expect no entries in column (22c) and 21 entries in (22d). Here we see a family resemblance with vowel-induced raising: most of the palatals that initiate unstressed syllables fail to cause raising. This begins to look like a change in progress, spreading out from the vowel environment to encompass consonants, but without the strict grammaticalization of a fully-established phonological rule, Not surprisingly, many forms are listed both with and without raising, that is the rule is variable. (25) ˈtiʎu, ˈteʎu (EHP: 296) ‘plank’

The syllabic position of glides in Spanish: insights from Pasiego vowel harmony

49

Where does this leave us? All told, we have 42 examples in this sample where [ɲ] or [ʎ] occur in a word with a mid or high vowel preceding them. If stress is irrelevant, we should find 42 cases with high vowels and none with mid vowels. Instead we find 12 (9+3). If stress is necessary, we should find 21 cases with high vowels and 21 with mid vowels. Instead we find 9 with raising and 12 without. The safest thing seems to be to agree with McCarthy (1984) in leaving the pre- [ɲ] or [ʎ] environment out of the domain of true vowel harmony in Pasiego; formalizing such a variable process is unlikely to shed much light on the form that true phonological generalizations take when they are fully established in the grammar.

6 Conclusion A language’s vowel harmony process can be a window into its syllabic structure. The location of prevocalic glides under onset versus nucleus has been underexamined both cross-linguistically and even in Spanish, where syllable structure has received a great deal of careful attention. In this paper, we saw that when we take seriously the details of the syllabification algorithm for high vocoids, a surprising new prediction about the harmonic participation of the Pasiego prevocalic glides emerges and is confirmed in the data. And a new angle of argumentation for syllabic position emerges: the allophone of a vocoid that appears in onset position should not be a trigger for vowel harmony in languages where harmonic features spread from the head of one syllable to the head of another. We have seen that the Pasiego data is best understood when we consider vowel harmony to be a process that makes one syllable head come to share features with another syllable head. Merely looking at the distinctive features or even the feature geometry of triggers and targets is insufficient. But it must be admitted that our cross-linguistic understanding of vowel harmony is far from complete. There are also languages, such as Turkish, in which harmony triggers are definable only by their features and feature geometry. My current speculation is that Pasiego falls into the syllable-head class because its harmony process requires stress on the triggering segment, and stress is a property of syllabic nuclei. Because stress is not a factor in Turkish vowel harmony and because Turkish has phonemic contrasts between palatal and non-palatal consonants, the presence of a vocalic place node is the characteristic that allows a segment to trigger harmony in Turkish. To confirm or refute this speculative correlation, it will be fruitful to look at the harmonic behavior of palatals, glides, and stress in additional languages.

50

Ellen M. Kaisse

References Clements, G.N. & Elizabeth Hume. 1995. The internal organization of speech sounds. In John Goldsmith (ed.), The handbook of phonological theory. Cambridge, MA: Blackwell. Halle, Morris, Bert Vaux & Andrew Wolfe. 2000. On feature spreading and the representation of place of articulation. Linguistic Inquiry 31. 387–444. Harris, James W. 1983. Syllable structure and stress in Spanish: a nonlinear analysis. Cambridge, MA: The MIT Press. Harris, James W. 1989. Our present understanding of Spanish syllable structure. In Peter C. Bjarkman & Robert M. Hammond (eds.), American Spanish pronunciation, 151–96. Washington, DC: Georgetown University Press. Harris, James W. & Ellen M. Kaisse. 1999. Palatal vowels, glides, and obstruents in Argentinian Spanish. Phonology 16. 117–190. Hualde, José I. 1989. Autosegmental and metrical spreading in the vowel-harmony systems of northwestern Spain. Linguistics 27. 773–805. Hualde, José I. 1991. On Spanish syllabification. In Hector Campos & Fernando Martínez-Gil (eds.), Current studies in Spanish linguistics, 475–93. Washington, DC: Georgetown University Press. Hulst, Harry van der & Jeroen van der Weijer. 1995. Vowel harmony. In John Goldsmith (ed.), Handbook of phonological theory. 495–534. Cambridge, MA: Blackwell. Kaisse, Ellen M. & Susannah V. Levi. 2004. Vowel harmony: nucleus to nucleus or vocalic node to vocalic node? Paper presented at the 80th Annual Meeting of the Linguistic Society of America, Boston, 9 January. Levi, Susanna V. 2000. Glides, laterals, and Turkish vowel harmony. Seattle: University of Washington MA thesis. Levi, Susannah V. 2004. The representation of underlying glides: a cross-linguistic study. Seattle, WA: University of Washington dissertation. Levi, Susannah V. 2006. Phonemic vs. derived glides. Lingua. 118. 1956–1978. McCarthy, John. 1984. Theoretical consequences of Montañes vowel harmony. Linguistic Inquiry 15. 291–318. Navarro Tomás, Tomás. 1965. Manual de pronunciacíon española. Madrid: Consejo Superior de Investigaciones Cientificas. Penny, Ralph. 1969a. El habla pasiega: ensayo de dialectología montañesa. London: Tamesis Books. Penny, Ralph. 1969b. “Vowel harmony in the speech of Montes de Pas (Santander), Orbis 18:148–166. Penny, Ralph. 1978. Estudio estructural del habla de Tudanca. Beihefte zür Zeitschrift für romanishe philologie, Band 167. Tübingen: Max Niemeyer Verlag. Rubach, Jerzy. 1998. A Slovak argument for the onset-rhyme distinction. Linguistic Inquiry 29. 168–179. Selkirk, Elisabeth O. 1986. The prosodic structure of function words. In Jill N. Beckman, Laura Walsh Dickey & Suzanne Urbanczyk (eds.), Papers in Optimality Theory. UMOP 18. Amherst, MA: GLSA, 439–469. Vago, Robert. 1988. Underspecification in the dual harmony system of Pasiego (Spanish). Phonology 5. 343–362. Yun, Yungdo. 2004. Glides and high vowels in Korean syllables. Seattle, WA: University of Washington dissertation.

Iggy Roca

Gliding ghosts or ghostly glides, and does it matter which? Abstract: This article puts forward an OT integrated Spanish grammar of: 1. the parse of high vowels abutting a non-high vowel, and 2. the stress patterns of non-verbs, an empirical prerequisite. The analysis is carried out without resort to the traditional category “glide”, argued to be redundant, indeed encumbering and misleading. The keystone of the formal edifice is a low-ranking constraint calling for right alignment of word stress with the stem boundary. The unmarked tautosyllabic parse of the targeted vowel clusters in turn follows from a ranking of ONSET above ALIGN(STRESS,STEM)-RIGHT superseded where needed by MAX-PEAK to protect lexicalized syllable peaks. Some further minimal machinery accounts for some additional subsidiary departures.

0 Introduction1 Spanish, as is well known, has five vowels in its sound inventory, two of them high: i, u. It is also known that the two high vowels can occur adjacent both to each other and to a non-high vowel, hence with no onset intervening. In addition, the two abutting vowels can be tautosyllabic or belong in separate syllables. Each such eventuality is mandatory as per speaker in the word citation form in normal circumstances. The traditional approach conceives of high vowels parsed outside the syllable peak as an autonomous segmental category “glide”. The handful of exceptions to this outlook includes three works by the present author (Roca 1991, 1997a, 2006). A few other voices also declaring an exclusively vocalic nature for the segments in question nonetheless still tend to avail themselves of the term “glide”, indeed in effect of the construct, in the corresponding analyses. This paper offers an empirical demonstration that such a category does not exist, at least in Spanish: languages are many and varied, and similar in-depth investigations are needed for each of them before pronouncing against “glides” 1 I am grateful to three anonymous reviewers and the editor for comments that contributed substantially to improve the coverage and the quality of the paper. To the editor also for his friendliness and continuous help throughout the editing process. Iggy Roca, University of Essex

52

Iggy Roca

universally. The detailed examination of the Spanish situation carried out here also suggests a viable route for the enterprise elsewhere, moreover with a tangible hint that the present results may be replicable. The ensuing gain for the field would be manifest, whichever the outcome. The task is consequently well worth the effort, both material, with regard to the collection and appropriate interpretation of the relevant empirical facts, and mental, in the way of countenancing the corresponding individual gestalt shift, frequently the steepest challenge. The short answer to the paper’s title question is provided on completion of the exposition, which develops as follows. Section 1 presents an analysis of Spanish non-verb stress, on account of its close interaction with the syllabification of high vowel (H) + non-high vowel (=) clusters, in either order, a matter then examined in detail in sections 2 to 5. Section 2 gives some preliminary background, followed by an analysis, both of the systematic heterosyllabicity of =-adjacent stressed H (also in either order), and of the key role played by ONSET as a tautosyllabicity trigger. Section 3 deals with the syllabification of H= clusters, both in word-final non-desinential position and word-medially, in the latter either tautosyllabic with paroxytone word stress, or heterosyllabic with proparoxytone stress. In section 4 we examine the converse =H cluster in both closed and open syllables, pointing out the favoured correlation between stressed = and tautosyllabicity but for (C)ˈ=.H ungrammaticality. In section 5 we show word-initial CHˈ= heterosyllabicity to be favoured but for the general preferred tautosyllabicity of both the He and [k/g/x]u= clusters, bar a handful of exceptions. A summary and a succinct evaluation close the chapter in section 6. In a nutshell, thus, both H= and =H tautosyllabicity is unmarked. Marked heterosyllabicity stems from stress on the cluster’s high vowel, from specific lexical marking of this vowel as a syllable peak, or from CH.ˈ= priority wordinitially, itself overcome by general tautosyllabic preference in He and [velar C]u= clusters. Stress will in turn be seen to be contingent on both morphological structure and minimal metrical strictures.2 2 Martínez-Paricio’s (2013) recent analysis of the same topics differs significantly. We lack the space to carry out detailed comparison (some observations will however be made as we go along), but need to point out the complexity of the conceptual machinery, in contrast to the simplicity of the present proposal. Ceteris paribus, therefore, (and there is no apparent reason to believe that ceteris are not indeed paribus here) the present analysis is to be preferred. In addition, M-P’s terminological oscillation between “vowels” and “glides” (and “onglides”, “offglides”, “gliding”) running through the paper gives out an inescapable hint of confusion between the two categories. Otherwise, M-P’s core stand of accounting for “default patterns of gliding and stress [. . .] via the interaction of well-established markedness constraints” and for “exceptions to default (i.e. marked stress and the so-called exceptional hiatus) [. . . by] lexical prosodic specifications to which output must be faithful” (abstract, p. 166) belongs in the same tradition as ours here, indeed as Roca (1991, 1997, 2006).

Gliding ghosts or ghostly glides, and does it matter which?

53

1 Spanish non-verb stress The syllabification of Spanish {=, H} clusters (= = non-high vowel; H = high vowel; {=, H} = an unordered set mapping onto two ordered sets and ) is inextricably linked to stress, and consequently a succint stress grammar needs to be provided first. Our account will build on Roca (2006), itself built on Roca (2005b), and importantly at variance with much previous common wisdom. Directly relevant to our present remit, Roca’s (2006) Spanish stress model fits well with the facts of {=, H} syllabification, shedding important light on them. Roca’s (2006) stress analysis is confined to non-verbs (“N” for short), the more complex and the more thoroughly investigated of the two systems, a limitation we need to maintain here for reasons of space.

1.1 Contextual background The first issue relevant to Spanish N stress concerns the morphological status of the vowel (a, o, e, u, i) on the word’s right edge, usually a “desinence”,3 a constituent differentiated from the “stem”. The Spanish desinence is a historical relic of the Latin nominal theme vowel, and is formally (not grammatically) de facto equivalent to the case endings of other languages. The following set of forms illustrate (“]” = Spanish stem right boundary): Lat

amic-u-m

ros-a-m

mont-e-m

trib-u-m

–

Sp

amig]o

ros]a

mont]e

trib]u

bikin]i4

Also possibly with a word-final s: Lat

cosm-o-s

Mecen-a-s

diabet-e-s

vir-u-s

cut-i-s

Sp

cosm]o-s

mecen]a-s

diabet]e-s

vir]u-s

cut]i-s

3 Our “desinence” has a rich pedigree, under various denominations: cf. Saporta (1959, 1962), Hooper and Terrell (1976), Harris (1980, 1983, 1985, 1991, 1992, 1999), Klein (1989), Roca (1989, 1997, 1999, 2005a, 2005b, 2006, 2013), Morin (1999), Colina (2003), Roca and Felíu (2003), OltraMassuet and Arregi (2005), among others. 4 In the absence of a source Latin accusative theme vowel -i-, which would in any event have manifested as -e in historical oral Spanish and then dropped after a voiced coronal consonant. As to bikin]i, cf. e.g. bikin]azo, bikin]ero/a, bikin]ucho, etc. (*bikini-azo, *bikini-ero/a, *bikini-ucho).

54

Iggy Roca

The Spanish desinence exhibits a number of properties, among which stresslessness is central to our present concerns: SPANISH DESINENCE PROPERTIES a. b. c. c. d. e. f. g.

stresslessness5 -V(-s) shape word-rightmost position suppression on suffixation lexical determination lexicalized alternations autonomy from gender semantic irrelevance

plˈat]o cosm]os mecen]as vir]us carbon]o herman]o ~]a la comet]a el vástag]o

cf. cf. not hence vs. vs. and and

platˈó] (*plat]ó) cóndor] (*cónd]or) *mec]e-nas vir]al (*vir]us-al) carbón] vástag]o (*~]a) el comet]a la person]a

The following word constellations put some additional empirical flesh on the matter: UNSUFFIXED SUFFIXED

CF.

páncre]a-s

pancre-át-ic]o *pancreasático

rococó > rococo-ic]o (nonce)

cosquill]a-s

cosquill-os]o *cosquillasoso

Barrabás > barrabas-ad]a

paper]a-s

paper-it]a-s, paper-ud]o *paperasitas, *paperasudo

compás > compas-it]o, a-compas-a-r

Balcan]e-s

balcán-ic]o *balcanésico

feligrés > feligres-í]a

mecen]a-s

mecen-uch]o (nonce); mecen-azg]o mayor > mayor-azg]o *mecenasucho; *mecenasazgo

órgan]o

orgán-ic]o; organ-ic-idad *organoico; *organoicoidad

rococó > rococo-íst]a (nonce)

Jud]a-s

Jud-e]a; jud-í]o *Judasea; *judasío

Judá > juda-ísm]o

5 An anonymous reviewer raises Panamá ~ panameño as a counterexample. The matter is however straightforward: 1. in Panamá -a is not a desinence (as it is in, e.g. Málag]a ~ malagu-eño ~ malagu-ista), but part of the stem (cf. Panama-(c)íto, Panama-ísta, not *Panamito, *Panam-ísta); and 2. panam-eño exceptionally (but not uniquely: cf. also Canadá ~ canad-iense) involves stem lexical allomorphy, in contrast to regular Alcalá ~ alcala-íno, for instance. Lexical allomorphy is of course rampant in Spanish, as in many other languages.

Gliding ghosts or ghostly glides, and does it matter which?

55

The status of the Spanish desinence is exclusively morphological, and in particular it is not an expression of semantic sex as sometimes assumed. Notice, for instance, (succinctly here: fuller discussion in Roca 1989, 2000, 2005a, 2013, Roca and Felíu 2004) female Christian names like Rosario or Consuelo, with the alleged male desinence -o, correspondingly hypochoristicized as Charo, Chelo. Conversely, male hypochoristics Chema, Juanma, Josema, Rafa with the alleged female desinence -a (also Christian names like Basque Gorka ‘George’ or Joseba ‘Joseph’, or the Russian Nikita of the former Soviet leader, all familiar throughout Spain), unproblematically in all cases, indeed unnoticeably to the ordinary speaker. With regard to stress, the following parameter settings have through the years been proposed in the literature in a variety of formal frameworks (cf. e.g. Harris 1983, greatly influential):6 SPANISH NOUN STRESS TRADITIONAL METRICAL SETTINGS DOMAIN: STRESS BEARERS: IDENTITY: MARKINGS:

word syllables (≡ syllable heads) extrametricality (lexically controlled): right edge quantity sensitivity (algorithmic): heavy syllables

FOOT PROPERTIES: SIZE: SHAPE: FOOT PARSING: WORD HEAD FOOT:

binary trochee iterative right-to-left rightmost

We do not have the space to expound on the difficulties inherent in this outlook (recent observations in Oltra-Massuet and Arregi 2005 and Roca 2005b, 2006). They include, not necessarily all at once (see three paragraphs below for the symbols):

6 A considerable body of work on Spanish stress has accrued since the inception of generative phonology. Harris (1995: fn. 2) lists 18 pre-1993 “substantial” publications, to which Hooper and Terrell (1976) (couched in “heterodox” Natural Generative Phonology) needs adding. Subsequently, at least the following are worthy of mention, besides Harris (1995) itself: Lipski (1997), Roca (1997), Saltarelli (1997), Oltra-Massuet and Arregi (2005), Roca (2005b, 2006), and recently Martínez-Paricio (2013). Also Roca (1999) on Romance languages in general.

56 – – – – – –

Iggy Roca

questionable quantity sensitivity insufficiently constrained extrametricality class-internal foot asymmetries (ˈSL) ~ (ˈH) analytic challenge from the ˈV# class (NB over 100 items, and open) analytic challenge from the supermarked classes ill-motivated ancillary devices

The ensuing analyses, varying among authors, involve complex, often awkward, machineries, in various shapes and forms depending on writers and times. The intricacy of the corresponding markedness taxonomy is itself suggestive of the shortfall: (U = unmarked; M = marked; (. . .) = foot domain; S = syllable; ˈS = stressed S; L = light syllable; # = word boundary; = extrametrical; H = heavy syllable; for graphic explicitness, we henceforth supply the stressed vowel with an accent mark (e.g. ó), irrespective of the standard orthographic norm,7 wherever stress is relevant to the discussion):

STRUCTURE E.G.

U . . . (ˈSL)# car(bóno)#

M . . . (ˈSL)# car(búncu)#

STRUCTURE E.G.

. . . (ˈH)# car(bón)#

. . . X(ˈH)# car(bún)#

Roca’s (2006) rationalisation of this taxonomy to be presented next reduces the formal machinery to the minimum compatible with the facts of both stress and, innovatively, {=, H} syllabification, our specific remit here.

1.2 Analysis Roca’s (2006) Spanish stress taxonomy can be schematized as follows, with the stress bearers crucially identified with vowels, not with syllables, and metrical structure assumed only to concern the V projection, consonants in the segmental line hence orthogonal (SM = supermarked; “≺” = ‘less marked than’, in line with OT’s ‘more harmonic than’):

7 In essence, orthographic accent is not assigned to paroxytones ending in a vowel, n or s, nor to oxytones not ending in a vowel, n or s. Orthographic accent is also borne by stressed high vowels in hiatus.

Gliding ghosts or ghostly glides, and does it matter which?

57

SPANISH STRESS TAXONOMY STATUS STRUCTURE SEGMENTS METRICS SEGMENTS METRICS SEGMENTS METRICS

U (ˈV)] carbón]o a (ó) ]o carbón] a (ó)] capó] a(ó)]

≻

M (ˈVV)] órgan]o (ó a) ]o gérmen] ( é e) ] cacáo]9 a(áo)]

≻

SM8 (ˈVV)V] – régimen], etc. (é i) e ] –

This approach achieves a significant degree of simplification relative to its traditional counterparts. Specifically, the thus re-defined, independently-motivated three markedness degrees tally with the V-based stress bearer count from the stress locus to the stem right boundary (NB crucially not the word’s): U = 0, M = 1, SM = 2. The full structures responsible for these outputs follow. For graphic parsimony, I henceforth incorporate foot brackets into the segmental line, consonants metrically hence needing disregarding ( = lexically determined D): SINGULAR: STRUCTURE E.G.

U . . . (ˈV)]# car(bó)n]o# car(bón)]# ca(pó)]#

M . . . (ˈVV)] (órga)n]o# (gérmen)]#10 ca(cáo)]#

SM . . . (ˈVV)V]# – (régi)men]# –

PLURAL: STRUCTURE E.G.

U . . . (ˈV)]-s# car(bó)n]o-s# car(bón)]-e-s# ca(pó)]-s#

M . . . (ˈVV)]-s# (órga)n]o-s# (mármol)]-e-s# ca(cáo)]-s#

SM . . . (ˈVV)]D# – re(gíme)n]-e-s# –

8 In the interest of presentational convenience at this point ignoring two additional SM classes introduced on p. 59 below. 9 An atypical parse justified on, e.g. cacao > cacao-(c)íto (*cacaíto). Also cacao-tal, cacao-tero, although here the suffix may be word- rather than stem-level (cf. viñ]a#tero, yerb]a#tero), like the more familiar -avo (onc]e#avo). Further evidence for ao]# parse from Táo > tao-ísmo, tao-ísta; Máo > mao-ísmo, mao-ísta. Cf. also coca > coca-ína. The diminutive coquita ostensibly involves stem lexical allomorphy akin to the one in panameño addressed in fn. 5 above. 10 Also borrowings like (cámping)]#, plural (cámping)]s#, where the g is an orthographic carryover from English.

58

Iggy Roca

The switch in stress domain from the traditional word boundary to the stem boundary brings the markedness contrast between the three patterns down to their true bare bones, as follows: 1.

unmarked (U)

= stem-final V stress: car(bó)n]o#, car(bón)]#, ca(pó)]#

2.

marked (M)

= 1. allowing for binary trochee: (órga)n]o#), (gérmen)]#

3.

supermarked (SM)

= 2. allowing for )–# misalignment (cf. (régi)men]#, re(gíme)n]es#)

The four constraints that follow account for 1. and 2., again bearing in mind that consonants are metrically immaterial: CONSTRAINT DEFINITION/EFFECT

DATA

AL-ˈV,]

stress (ˈ) – stem (]) right alignment

car(bó)n]o#, car(bón)]#, ca(pó)]#

BINFTM

binary foot (for class ‘M’)

(órga)n]o#, (gérmen)]#, ca(cáo)]#

*IAMB

trochaic foot structure

cf. *(orgá)n]o#, *(germén)]#

AL-Ft,]

foot – stem right alignment cf. *or(gán]o)#

The root constraint AL-ˈV,] is responsible for the unmarked stress of words like car(bó)n]o#, car(bón)]#, ca(pó)]#. Were this to be the sole Spanish stress pattern (the case in French, with the desinence confined to schwa, systematically stressless and usually silent: cf. Roca 1999), these would be the only stress configurations present in the language. However, we are already aware that Spanish also countenances M stress one position left of U stress, hence pointing to a binary trochee ((órga)n]o#, (gérmen)]#), itself a joint product of BINFT M (= binary foot) and *IAMB (= trochaic, not iambic, structure: cf. *(orgá)n]o#, *(germén)]#). The tableau that follows displays the workings of the first three constraints, with AL-ˈV,] lowest ranked. The novel stylized format aims at enhancing visual accessibility, unnecessarily encumbered by the graphic profligacy of its conventional counterpart. The boxed-in winner’s line further expedites comparison with losers (“>” = evaluation winner): /M/

*IAMB, BINFT M >> AL-ˈV,]

>(gérmen)] (germén)] ger(mén)

* *! *!

Gliding ghosts or ghostly glides, and does it matter which?

59

Gérmen of course carries no desinence. In desinenced órgano, however, the three thus ranked constraints fall short of the true output (“>*” = empirically untrue winner; “♦” = empirically true loser): *IAMB, BINFT M >> AL-ˈV,]

/M/ >*or(gán]o) ♦(órga)n]o (orgá)n]o or(gá)n]o

*! *! *!

An additional also dominant constraint AL-Ft,] amends this false result: *IAMB, BINFT M, AL-Ft,] >> AL-ˈV,]

/M/ >(órga)n]o#

*

or(gán]o)# (orgá)n]o# or(gá)n]o#

*! *! *!

This grammar accounts for the stress patterns of over 99% of Spanish nonverbs. It only leaves out the marginal, albeit fully operational, SM (supermarked) class, itself comprising three subclasses. They are respectively represented in the items régimen (plural regímenes), ómicron (plural omicrónes), and (singlehandedly) carácter, simply M PO] in the singular but with an unexpected O] plural caractér]es, in contrast with the straight M cráter ~ cráter]es. Their respective origins are prescriptive, and the patterns in any event orthogonal to our “glides” remit. Consequently, I will not offer an analysis here, instead referring the reader to Roca’s (2006, 2015), fully in line with the one just advanced for the two core classes. Foot presence has so far only been assumed. Regulation of the matter falls on a constraint PARSE-σ: PARSE-σ

Syllables parse into feet

A ranking PARSE-σ >> AL-Ft,] leads to foot iteration: /M/

*IAMB, BINFT M, PARSE-σ >> AL-Ft,] >> AL-ˈV,]

>(óro)(géne)s]is oro(géne)s]is oro(géne)(s]ís) (óro)(géne)(s]ís)

* * *

**!* *!*

**

****

* ***!

* ** *****

60

Iggy Roca

The winner in a. (tableau candidates are henceforth assumed as ordered a. to z. from top to bottom) has the stem exhaustively parsed into two feet, whereas in b. and c. the candidates have one foot missing. In c. and d. the desinence is also footed. However, as Roca (1986) argues in detail and Harris (1991) corroborates,11 Spanish secondary stress applies in a domain higher than the word in the prosodic hierarchy. Moreover, it is optional, indeed often missing in ordinary discourse (Roca 1986). In consequence, AL-Ft,] needs to outrank PARSE-σ in the word domain to limit the number of feet to one, the carrier of the primary, and at this level only, stress: *IAMB, BINFT M, AL-Ft,] >> PARSE-σ, AL-ˈV,]

/M/ >oro(géne)s]is (óro)(géne)s]is oro(géne)(s]is) (oro)(gene)(s]ís)

*! *!

*!* * ***

***

*

* **

**** * *

The now established full ranking follows, with illustrations intercalated:

11 Spanish secondary stress phonology subsequently became neglected to the benefit of the phonetics, with results casting doubt on its very existence: cf. Scharf et al. (1995a, 1995b), Prieto and Van Santen (1996), Díaz-Campos (2000). A recent return to phonology (Hualde 2007, 2009, 2010, Hualde and Nadeu 2013) is focused on rhetorical or emphatic stress, deemed by Hualde more real and central in Spanish than rhythmic or alternating stress, without however discarding a possible as yet unexplored link between the two. Hyde and McCord (2012) and Buckley (2012, 2014) do address rhythmic stress in the context of the formal dilemma between gradient and categorical alignment. The suitability of this set of data for this task is however questionable on account of its well-known and widely acknowledged elusiveness. Both analyses in addition adopt Harris’s (1983) interpretation of the initial-dactyl alternative as colloquial and the foot right alignment alternative as rhetorical, despite the reverse position upheld in Roca’s (1986) subsequent analysis. The difference is far from trivial, given that colloquial speech is reasonably generally construed as natural and unmarked, and rhetorical speech as contrived and marked.

Gliding ghosts or ghostly glides, and does it matter which?

61

Two additional issues need brief attention before closing the discussion on stress. First, as indicated in the table on p. 55 above, a tradition originating in Harris (1967, 1969)12 and consolidated in Harris (1983) views Spanish N stress as sensitive to syllable weight. This claim got questionned already in the 1980s (cf. Roca 1988), then in the 1990s (Roca 1990; partially also Lipski 1997 and Roca 1997b), and again more widely in recent years (Bárkány 2002, Alvord 2003, Face 2004, Roca 2005b, 2006),13 and consequently no such assumption will be made here.14 Second, our identification of the Spanish stress bearing unit with the syllable nucleus head currently encapsulated in PARSE-σ, fully adequate for the deliberately monovocalic syllables we have been considering, henceforth needs widening to all rime-parsed vowels in order to account for the syllabification patterns of {=, H} clusters, the paper’s remit. The new formulation follows: PARSE RIME VOWELS (PARSE-V(R)): Rime vowels parse into feet We are already aware that foot shape is controlled by *IAMB, and foot size by the ranking BINFT M >> AL-ˈV,]. The corresponding updated Hasse Diagram is as follows:

12 Foreshadowed in Foley (1965: 84): “Stress the penultimate vowel if long or followed by two consonants of which the second is not a liquid, otherwise stress the antepenultimate vowel”. 13 Martínez-Paricio (2013) also declares against quantity sensitivity. The stand is, however, ambiguous, as she at the same time considers word-final “glides” and consonants moraic to get the corresponding syllables to attract stress, hence de facto construing them as heavy (see fn. 37 below). 14 The widespread traditional idea that Spanish stress is quantity sensitive misses two crucial facts: 1. Latin stress was indeed quantity sensitive, and 2. the Latin stress locus has lived through into Spanish almost unfailingy (cf. “El acento se mantiene inalterable desde el tiempo de Plauto, de Horacio, de Prudencio, hasta el de Cervantes y hasta el nuestro”; Menéndez Pidal 1962, p. 36, §5bis; cf. also Pensado 1985). Moreover, there are a handful of historical exceptions to the supposed quantity sensitivity, among which the place name Frómis.ta (a town in the Castilian heartland) is well-known. Also, and most revealingly, heavy syllables are readily skipped in modern importations: cf. Wáshington, Dússeldorf, Párkinson, etc. Indeed, the common proparoxytone distortion of the source stress in foreign names pays no regard to quantity sensitivity: e.g. Dutch Amsterdám > Spanish Ámsterdam. These facts would be bizarre if Spanish stress indeed were quantity-sensitive.

62

Iggy Roca

In the interest of presentational parsimony, we will henceforth ignore PARSEV(R) where not directly relevant to the discussion. This completes our presentation of Spanish N word stress. The remainder of the paper explores the interaction of this stress grammar with the grammar of {=, H} syllabification we will gradually build.

2 {=, H} syllabification 2.1 Preliminaries Roca’s (2006) analysis of {=, H} syllabification (H = high vowel; = = non-high vowel), on the footsteps of Roca (1991, 1997a), has no resort to the category “glide”. Here we shall draw on this analysis, expanding on it as necessary. The abstract combinatorics of stressed {=, H} clusters follows: {=, H} CLUSTER STRESS FACTORIAL TYPOLOGY V TYPES

STRESSED V

{=, H} {=, H}

H =

STRESS CONTOUR TYPES V1 STRESS V2 STRESS ˈH= =ˈH ˈ=H Hˈ=

Should Spanish {=, H} syllabification bear no relation to stress, contrary to fact, each such cluster would be liable to both a tauto- and a heterosyllabic parse. In actual fact, however, a number of stress-driven syllabification restrictions tabulated below are in place. For visual salience, illegitimate parses are crossed out, and mandatory ones boxed in. Heterosyllabicity is henceforth graphically marked with an intervening dot (“X.Y”). Tautosyllabicity is left unmarked (“XY”), except where potential unclarity advises inclusion of the conventional ligature symbol (“X͡Y”): SPANISH STRESSED {V, H} CLUSTERS SYLLABIFICATION INVENTORY Stress

Sonority Falling

Rising

heterosyllabic

tautosyllabic

heterosyllabic

tautosyllabic

+–

ˈ=.H

ˈ=H

ˈH.=

ˈH=

–+

=.ˈH

=ˈH

H.ˈ=

Hˈ=

Gliding ghosts or ghostly glides, and does it matter which?

63

The taxonomy’s gist is as follows: ˈH (in either order) ˈ=-initial cluster ˈ=-final cluster

% % %

heterosyllabicity tautosyllabicity compatible with either

Greater precision on the syllabic status of the ˈ=-final class requires introduction of an additional category “favoured” (K = velar C; L = liquid): Fixed syllabification

Favoured syllabification

Heterosyllabic

ˈH

#C(L)H.ˈ=

Tautosyllabic

ˈ=H

Kuˈ= Hˈe

In the remainder of the paper, responsibility for these patterns will be laid on the constraint rankings displayed in the Hasse diagram that follows, with the usual illustrative data intercalated (V̆ = stressless vowel; C = epenthetic consonant; Ø = null, i.e. = deleted):

This ranking connects with its stress counterpart through simple domination of ONS over AL-ˈV,]:

64

Iggy Roca

These constraints and their rankings are introduced and justified in the remainder of the paper, stepwise from where we left off for stress. For reasons of space we limit coverage to tautomorphemic clusters. In a nutshell, heteromorphemic stressed {=, H} is heterosyllabic but as regards derivational suffixes, which induce a more complex behaviour. Desinences do not behave as suffixes for syllabification purposes, but as morphological “excrescences” of the stem, that they simply “inflect”. A matter raised by reviewers needs clarification before we proceed. While the data pertaining to stress are uniform throughout the Spanish-speaking world, some aspects of {=, H} syllabification are subject to geographical, and ultimately idiolectal, variation, furthermore possibly socially or age-related determined. Clearly, thus, any report of such facts of necessity needs to be selective, criterion choice hence being a prior requirement. I will simply adhere to the Chomskyan paradigm and place the speaker at the centre of language, in particular the native speaker’s specific idiolect. Consequently, I will guide myself by my own practice and intuitions (wherever apposite labelled “RA”, for “reference accent”), robustly underpinned by three decades of sustained critical selfexamination.15 I am also familiar with the reports in the classical treatise by 15 I am thus in full agreement with Harris and Kaisse’s (1999:119) stand that “genuine understanding of particular grammars and of general phonological theory must be rooted in careful investigations of individual dialects and grows as insights from a number of such inquiries are consolidated”. Hualde (1999:188) also settles for “bas[ing] the description on my own [Hualde’s, IR] dialect faute de mieux”. My idiolect belongs in what is now often referred to as NorthernCentral Peninsular Spanish, the variety also the object of Navarro Tomás’s and many other

Gliding ghosts or ghostly glides, and does it matter which?

65

Tomás Navarro Tomás, arguably the most illustrious Spanish phonetician of all times, and am aware of a number of other pronunciations, both experientially (I can mimic some with, I think, a reasonable degree of accuracy) and scholarly.16 The task relevant to our present endeavour is not, in any event, one of complete inventory taking, doomed to fall badly short if carried out by a single individual, a collective full survey of the totality of Spanish speakers (400m?) in turn a chimeric dream. It is thus, rather, one of charting the main elements partaking in the matter and bringing out the forces acting on them, consequently my goal here. The data issue becomes particularly critical in section 5.1 below, and we will return to it there.

2.2 =-adjacent ˈH heterosyllabicity The inevitable heterosyllabicity of =-adjacent stressed H displayed in the inventory on p. 9 above is a fact of Universal Grammar.17 {ˈH, =} tautosyllabicity presupposes one of the following two configurations, respectively with rising and falling sonority (X̩ = syllable nucleus; X̯ = not syllable nucleus):

Both these structures are at odds with the familiar universal principle identifying local sonority peaks with syllable peaks. The respective legitimate parses are thus, instead, as follows:

common descriptions. As I have warned in the main text, however, any individual accent may, and most likely does, contain some subdialectal, indeed idiolectal, traits, and consequently absolute cross-coincidence cannot a priori be guaranteed. I will point out differences wherever I am aware of them. 16 Cf. Beritognolo (2008) for a variety geographically (but not necessarily factually) about as removed from mine as materially practicable. 17 An anonymous reader points to English beer [ɪə̯], law [ɔə̯] as possible counterexamples. This remark is oblivious of the fact that English “diphthongs” are generally viewed as single vocalic units with a non-steady articulation. In contrast, the Spanish ones are incontrovertibly made up of two full vowels, with no accommodation for “glides”: at its most basic, ordinary Spanish speakers can “deletrear” (spell out orally), or even slowly pronounce (should they wish to), (v)ió as disyllabic (v)i.ó (cf. ri.ó), whereas English speakers cannot pronounce (b)eer with [ɪ].[ə]. Cf. the discussion on Dioni on p. 89 below. Also fn. 48. Terminological rigour is of course of the essence in scientific inquiry.

66

Iggy Roca

This principle must be assumed to be part of GEN. In the interest of both explicitness and explanation, I shall nonetheless for the moment formalize it as an undominated constraint (SON=SYL)-PK, itself a joint corollary of both the principle of sonority sequencing, responsible for the mountain-like sonority shape of the syllable, and the peak minimal sonority parameter, in Spanish set to H (high vowel): (SON=SYLL)-PK

A Sonority Peak corresponds with a Syllable Nucleus Peak

PKMINSON(SP)

Syllable Nucleus Peak sonority ≥ H

The tableau below integrates the undominated (SYLL=SON)-PK into the stressrelated ranking already in place. For economy, the also undominated PKMINSON (SP) is simply assumed. Unless otherwise stated, all evaluations are henceforth pertinent to both i and u. Both rising and falling sonority clusters will be considered. To speed up tableau reading, candidate lines are provided with hyphens under constraints inoperative owing to absence of the required lexical trigger: (SON=SYL)-PK >> IAMB, BINFT M, AL-Ft,] >> AL-ˈV,] –

>Ma(rí.)]a (Mári)]a Ma(rí)]a̯

*!

– –

>Sina.(í)]

–

Sin(ái)] Sina̯(í)]

– –

*!

*!

*!

Both winning candidates, Marí.a and Sina.í, have the stress drawn to the stem right egde by AL-ˈV,], notwithstanding its lowest ranking. The contrastive parse of the word-final vowel, as a desinence in the top tableau but in the stem in the bottom one, is of course morphologically motivated: cf. e.g. the effect of the evaluative suffix -ucho/a on the base’s last vowel, i preserved in Sinaiúcho (i hence not a desinence), but not so a in Mariúcha (a hence a desinence). An also tautosyllabic configuration with the rime appropriately sonoritypeaked but the stress on the lower-sonority sister vowel is also ungrammatical:

Gliding ghosts or ghostly glides, and does it matter which?

67

The correct alignments are, instead, as follows:

I will dub the constraint responsible HEADSTACK,18 and formulate it as follows: HEADSTACK

A metrical head corresponds with a syllable head

The following two tableaux display the contributions of the two new constraints: HDSTACK, (SON=SYL)-PK >> *IAMB, BINFT M, AL-Ft,] >> AL-ˈV,] >Ma(rí.)]a

–

(Mári)]a Ma(rí ̯)]a Ma(rí)]a̯

– – –

*! *!

–

>Sina.(í)] Si(nái)] Si(naí ̯)] Sina̯(í)]

*!

*! *!

– – –

*!

The incompatibility of stressed ˈH with cluster tautosyllabicity hence falls out from the two principles respectively encapsulated in HDSTACK and (SON=SYL)-PK. For simplicity of presentation, indeed for formal and conceptual 18 Headship transitivity is acknowledged in Prince and Smolensky (2004:62): “the head of X’ count[s] also as the head of X’”.

68

Iggy Roca

rigour, I will henceforth assume inclusion in GEN, correspondingly omitting them from the ranking.

2.3 ONSET as tautosyllabicity inducer Our present ranking induces stem-final stress, the concomitant cluster heterosyllabicity a simple consequence of the HDSTACK and (SON=SYL)-PK GEN requirements. Both Marí.]a and Sina.í] correspondingly come out violation-free, as I have just shown and now highlight: *IAMB, BINFT M, AL-Ft,] >> AL-ˈV,] >Ma(rí.)]a >Sina.(í)]

– –

This outcome is at odds with the Spanish unmarked status of tautosyllabicity, with about a two to one advantage in lexical occurrence: Mário͡ is thus less marked than its minimal-pair partner Marí.a (the desinential vowel quality contrast is immaterial), and similarly for bonsá͡i versus Sina.í. Besides numbers, tautosyllabicity unmarkedness informs the official Spanish written accentuation conventions (cf. fn. 7 above), among the best in the world in Harris’s (1975:60) declared opinion. In line with Roca (2006), we shall achieve tautosyllabicity on a ranking of tautosyllabicity-favouring ONSET over stem-final stress favouring AL-ˈV,]. We define ONSET as follows: ONSET

A syllable has an onset

An illustrative evaluation follows. Henceforth, I split the conventional single constraint line of tableaux into two or more parallel ones, on two accounts. First, to maintain visually separate the set of constraints responsible for stress from those responsible for {=, H} syllabification, so as to enable readers to keep tabs of their respective contributions to the evaluation. And second, crucially, to allow ongoing tracking of the full mesh of rankings, as is known inevitably partly opaque on the conventional unilinear representation: cf. McCarthy (2008: 49–50). Constraints separated by commas on the same line are mutually unranked. An arrow replaces the usual ranking symbol “>>” to enhance visual salience in the context of the new format. A simple (physical or mental) 45o right turn of the constraint lines on the plane reveals the intended isomorphism with the standard Hasse diagram display.

Gliding ghosts or ghostly glides, and does it matter which?

>(Mári)]o

–

Ma(rí.)]o

–

>bon(sái)]

–

bonsa.(í)]

–

69

* *! * *!

Both these winners are at odds with the preceding ones for the respective segmental partners Marí.a (vs. Mário) and Sina.í (vs. bonsái).19. The early OT literature formalized such not unusual ranking paradoxes as local ranking reversals akin to “co-phonologies”, here hence ONSET >> AL-ˈV,] (Mário, bonsái) vs. AL-ˈV,] >> ONS (Marí.a, Sina.í ), as follows (cf. Rosenthall 1994: 147 (81)):

–

>Ma(rí.)]a (Mári)]a (Marí.)]a Ma(rí.]a)

*!

– – –

>Sina.(í)]

–

Si(nái)]

–

>(Mári)]o

–

>Mar(í).]o (Marí.)]o Ma(rí.]o)

*!

– – –

>bon(sái)]

–

(bónsai)] (bónsa)i] bonsa.(í)]

– – –

* *! * *

*!

* *!

*

*!

*! * * * **! **

*! *!

19 An anonymous reviewer objects to the (Mári)]o foot structure on grounds of syllable integrity violation. The issue of syllable integrity has recently been insightfully addressed in Hyde (2007).

70

Iggy Roca

The manifest burden on learnability incurred by such formal complexity led Roca (2006) instead to model surface heterosylabicity by means of a lexicalized syllable peak in concurrence with a faithfulness constraint MAX-PK, crucially dominant of ONS:20 MAXPK

A lexical syllable nucleus peak corresponds with a surface syllable nucleus peak

The Marí.a-Mário opposition now thus reduces to a contrast between presence and absence of /i./ in the respective lexical representations:

>(Mári)]o

–

–

Ma(rí.)]o Mari](ó)

– –

– –

>bon(sái)]

–

–

bonsa.(í)] bon(sá.i)] /i./

– –

– –

>Ma(rí.)]a

–

(Mári)]a Mari.](á) /i./

– –

>Sina.(í)]

–

Si(nái)] Si(ná.i)

–

* *! *!

*

* *! *!

*

* *! *

*!

* *

* *! *

* *!

Violations are attested in a number of languages, but the solution is straightforward: simple outranking of the constraint(s) responsible for syllable integrity. In Spanish the dominant constraint would be ONSET. 20 For antecedents see, e.g. Steriade (1984) (Rumanian), Guerssel (1986) (Ait Seghrouchen Berber), Roca (1997) (Spanish), Harris and Kaisse (1999) (Spanish). For the general theory, Levin (1985: 2) and Blevins (1995: 221). Levin’s statement is well worth quoting: “vowel/glide alternations must often be specified underlyingly as [+syllabic], but never as [-syllabic] [. . . Heads] can be marked in underlying lexical representations, while the property of being a non-head cannot”. The latter is precisely the defining property of the “glide” construct we are shunning here.

Gliding ghosts or ghostly glides, and does it matter which?

71

The scope of MAXPK needs restricting to stressed clusters on account of the characteristic tautosyllabicity of Spanish stressless clusters (cf. “syllable merger” in Roca 1986: 350–352; 1991: 605 ff.):21 cf. Sina.í vs. diminutive Sinaicíto (*Sina.icíto). The required constraint is *V̆ .V̆ , for V̆ = stressless V: *V̆ .V̆

Adjacent stressless vowels are tautosyllabic

The necessary prevalence of *V̆ .V̆ entails domination over MAXPK:

–

>Sinai(cÍ)t]o Sina.i(cÍ)t]o

*!

–

* *

The equivalent conventionally formatted Hasse diagram follows, for greater transparency:

In a nutshell, thus, cluster tautosyllabicity is driven by ONSET, itself checked by MAXPK, itself checked by *V̆ .V̆ . This simple ranking constitutes the backbone of the Spanish {H, V} syllabification grammar. In the sections that follow we review a set of complying configurations (3.1, 3.2, 4.1, 4.2), then turning to contradictory ones requiring additional machinery (4.3, 5.1, 5.2.1, 5.2.2).

21 Exceptions are possible. Hualde (1999) reports several in his own speech, some involving the cycle: cf. re.irémos < re.ír (cf. Harris’s 1969: 95–96 antecedent o.iré, not oiré, etc.) vs. reinémos < reinár (this contrast does not obtain in my speech). Others are idiosyncratic, however: those concerning the stressless di=-prefix, for instance (p. 190). Appropriate machinery to deal with such an eventuality will be introduced below: see the first tableau on p. 30 and the related discussion.

72

Iggy Roca

3 H= clusters 3.1 Word-final non-desinential All the H= clusters we have considered involve a desinential parse of =, as we know usual for the stressless word-final Spanish vowel. Desinenceless forms with a matching sonority profile also exist, however. The import of the contrast concerns the role of the stem right boundary in the determination of the segmental span falling within the range of both AL-Ft,] and AL-ˈV,], with the concomitant exclusion or not of the desinence vowel (stressable, as a rime vowel) from their respective scope. The correct stress and syllabification patterns of such desinenceless forms also follow from our present ranking. Consider first the contrast between, on the one hand, Rúbi]o# (paralleling Mári]o#) and Rubió]#, and, on the other, Marí.]a# and Sarri.á]#, as well as between Sarri.á]# and Sárri]a#, all now tabulated for ease of comparison: MORPHOLOGY DESINENCED DESINENCELESS Rúbi]o# Rubió]# Sárri]a# Marí].a# Sarri.á]# #-NONFINAL #-FINAL STRESS

TAUTO ) TAUTO SYLLABICITY HETERO

The Spanish surname Rúbi]o (cf. rúbi]o ‘blond’) matches Mári]o with regard to both stress and syllabification. A Catalan segmental homophone Rubió (unproblematically also thus pronounced by Spanish-only speakers) contrasts with Rúbio in stress. Also, not unrelatedly, in the morphological status of the final o, desinential in Rúbi]o but stem-internal in Rubió]. The contrast is significant, as the necessary stresslessness of the Spanish desinence provides a robust cue for the morphological allotment of the vowel.22 Both these stress patterns (Rúbio’s of course identical with Mário’s) fall out from the present ranking:23 22 Robust but not determining, on account of the small set of V-ending but desinence-less items with M PO stress referred to in fn 9 above. 23 In candidate e. in the Rubió tableau, ONS and AL-Ft,] are level-ranking, both violations hence fatal. To facilitate reading, however, I restrict tableau violation marking to the levelranking constraint linearly ordered first, here hence ONS.

Gliding ghosts or ghostly glides, and does it matter which?

>(Rúbi)]o

–

–

Ru(bí.)]o

–

–

>(Rubió)]

–

–

(Rúbio)] (Rúbi)o] Ru(bí.o)] Ru(bí.)o] Ru(bí.o)]

– – – – –

– – – – –

73

* *!

*!* ** * * *

*! *! *! *!

*

The tautosyllabicity of desinenceless Rubió contrasts with the heterosyllabicity of likewise desinenceless Sarri.á, another Catalan name (a Barcelona district) systematically also thus pronounced by Spanish-only speakers. Our analysis accounts for the difference on a simple assumption of an underlying nucleus head /i./ in Sarri.á, in contrast with Rubió:

>Sarri.(á)]

–

Sarri(á)] Sa(rrí.a)] (Sárri)a] (Sárria)]

– – – –

* *! *! *!

* *

*! ** **

Sarri.á and Marí.a hence share heterosyllabicity (both carry the /i./ mark), but their stress patterns differ. Responsibility for the divergence falls on the respective morphological structures, desinenceless in Sarri.á but desinenced in Marí.a, all as familiar. Sarri.á also contrasts with Sárria (likewise a toponym, in the also Spanish region of Galicia), with the Mário familiar pattern and evaluation:

74

Iggy Roca

>(Sárri)]a

–

–

Sa(rrí.)]a Sarri](á)

– –

– –

* *! *!

*

All the contrasts examined concern H=-ending words with = parsed either in the stem or in the desinence. In the former, = may be followed by an also stem-internal word-final C: – – –

Cebrián, manantiál, marciál, magiár gorrión, Gerión, gestión, halción, Albiól Asuán

Forms in this shape also tend to have final stress,24 with concomitant unmarked H= tautosyllabicity (geniál), veni.ál a rare exception to the latter in my speech:25

>geni(ál)]

–

–

(géni)al] (génial)] /i./

– –

– –

>veni.ál

–

venial

–

*!

** *!*

* *!

3.2. Word-medial H= dual patterning Hˈ= tautosyllabicity is also favoured word-medially, tautomorphemicity of course assumed:26 24 The main exception concerns a handful of m- or r-ending classicisms identifiable as SM, e.g. ar(móni)um]# or (séni)or]#. 25 Navarro Tomás (1959:159) nonetheless gives gorri.ón, avi.ón, embri.ón, all at odds with my speech. 26 The phonological overlap of the final segmental sequence with an existing suffix (e.g -ano/a in anciáno) has no grammatical or semantic support for such an analysis here.

Gliding ghosts or ghostly glides, and does it matter which?

– –

75

idióta, idióma, mediócre, periódo anciáno, Santiágo, Severiáno, Tatiána, Arriága

This pattern also falls out from the present procedure in the absence of lexical marking:

>peri(ó)d]o

–

–

peri.(ó)d]o peri(ód]o) pe(rí.o)d]o (pério)d]o (péri)od]o

– – – – –

– – – – –

*! *! *! *!

* *!* **

A small set of items, período indeed among them, unexpectedly exhibit an alternative pattern, with PPO# stress and H.= heterosyllabicity: tautosyllabic PO#

periódo, pediátra, Hesiódo, zoodiáco, . . .

heterosyllabic PPO#

perí.odo, pedí.atra, Hesí.odo, zoodí.aco, . . .

Joint vowel heterosyllabicity and foot maximality imply double lexical marking /H., M/:27

/M, H./ >pe(rí.o)d]o peri.(ód]o) peri.(ó)d]o peri(ó)d]o peri(ód]o) (pério)d]o (péri)od]o (péri.)od]o

* *! *!

*!

* * *! *! *! *!

* *!

*

*

* *

** ** **

27 The pattern is actually etymological: cf. Latin pe(rˈĭ.ŏ)du, with PPO stress and abutting short vowel heterosyllabicity, general in Latin modulo the three diphthongs ae, oe, au (Bassols 1962: 76; Grandgent 1963: 141). Its Spanish correlate is attributable to learned normativism. Roca (2006:262–263) notes compatibility of the /M/ candidate *périodo with the proposed ranking (it beats período on ONS), and rules it out on an additional constraint WSP (Weight-to-Stress Principle) with the scope restricted to vowels. The soundness of this interpretation is however uncertain. Thus, a word acríviola (included for ‘Capuchina (planta trepoleácea)’ in the Enciclopedia Universal, although not in the DRAE), with the same pattern, may be thus pronounceable

76

Iggy Roca

This pattern fails to manifest during ordinary acquisition, unsurprisingly so on account of its formal complexity. It can however be incorporated later in life by the better educated at the cost of the double marking. The process may be helped, if not directly induced, by the so-called “manía esdrujulista” (‘proparoxytone obsession’), a well-known phenomenon, both historical and contemporary, involving PO to PPO stress shift, whence *périto, *cólega, *pápiro, etc. for etymological and standard períto, coléga, papíro, etc. Some such eccentricities have actually displaced the originals in the standard: cf. e.g. modern parásito, políglota, tortícolis for prior parasíto, poliglóta, torticólis (general discussion in Alonso 1930, and practically oriented one in Gómez Torrego 1989). This bizarre behaviour can perhaps be attributed to an association of PPO stress with “poshness”, presumably on account of its not infrequent occurrence in words themselves felt as “posh”, some with learnèd PPO suffixes like -ólogo, -ófago, -íparo, -ívaro, -íatra, -ómano, -ígero, -ólatra, etc., and often related to somewhat arcane domains of human existence such as medicine or religion.28 All the word-medial H= clusters examined involve open syllables. There indeed seems to be a dearth of relevant close-syllable correlates, for no obvious reason, where “relevant” is meant to exclude the Hˈe and Kuˈ= clusters, independently shunning of heterosyllabicity and dealt with in 5.2 below. The quasiminimal PO pair hedióndo vs. Iri.óndo embodies this additional eventuality.

despite the obvious shortage of lexical support. This eventuality is actually congenial with the experimental finding reported in Shelton et al. (2012:11) that “rising and falling diphthongs do pattern differently [. . .] [F]alling diphthongs induce a significantly higher error rate [on a PPO stress experimental condition, IR] than do rising diphthongs”. As regards factual perí.odo, the external normative target was precisely the PPO Classical Latin etymon, and consequently our /M, i./ double marking was, and still is, required. The lexical allomorphy of the tautosyllabic ͡ derivative periódico (*peri.ódico) can be taken as corroborative of the contrived nature of perí.odo. Martínez-Paricio’s (2013:185) corresponding tableau (47) fails to include a false winner pe μ.ri μ.(oμh.doμ) (“μ” = mora; “h” = foot head), for no obvious reason. 28 Menéndez Pidal (1962: 40, fn 1) indeed refers to the “prestigio docto que las dignifica [i.e. a estas palabras, IR]”. An anonymous reviewer suggests that analogy may also be at play: cf, e.g. parasíto > parásito < depósito, medúla > médula < cédula. It is far from clear, however, why analogy should not have operated in precisely the opposite direction (medúla > cedúla, etc.) to achieve the general unmarked Spanish stress pattern: this may be the tendency in Aragonese local varieties, but it is not in Spanish, our sole remit here. Alonso (1930: 349 ff) also appeals to analogy, but Menéndez Pidal (1962: 40, fn 1) appositely points out that “las escasas terminaciones -ago, -igo, etc. no podían vencer las muchísimas más en -ágo, -ígo, etc. si no es por el prestigio del esdrújulo”.

Gliding ghosts or ghostly glides, and does it matter which?

77

4 ˈVH cluster 4.1 Stress – syllabicity match We are by now well aware of the compatibility of word-final falling sonority clusters with both tauto- and heterosyllabicity, respectively contingent on absence vs. presence of lexical /H./: cf. the bonsai vs. Sina.í evaluations on p. 69 above. The same situation turns up word-medially. The following forms with stressless H illustrate cluster tautosyllabicity: – – –

donáire, arcáico, zanfáina, mosáico, Lováina, poláina (also e.g. náuta with u) acéite, aféite, deléite, Bestéiro, Cabréira intróito, tróika, estóico, paleozóico, colóide, paranóide, tifóide Heterosyllabic stressed H is also attested ( signals a silent orthographic h):

– – –

para.íso, ba.ía, Isa.í.as, Bala.ídos (also sa.úco with u) ve.ículo, de.íxis Bo.ígas

Both these patterns also fall out from our machinery. The =.í patterned set crucially includes the /H./ mark responsible for unpredictable heterosyllabicity:

>mo(sái)c]o

–

–

(mósai)c]o mosa.(í)c]o (mósa)ic]o /i./

– – –

– – –

>para.(í)s]o

–

para.(ís]o) pa(rái)s]o (pára)is]o (párai)s]o

– – – –

* **! *! *!

**

* * *! *! *!

*! *

* ** **

Likewise word-initially, again respectively without and with /H./ marking:

78

Iggy Roca

>(náu)t]a

–

–

na.(ú)t]a /u./

–

–

>sa.(ú)c]o

–

*

sa.(úc]o) (sáu)c]o

– –

*

* *!

*!

*!

*

In náufrago (also gnéisico, with i), with the stress one position further back from the word right edge than is habitual (4 Vs: cf. e.g. ráfaga, with 3), an /M/ mark crucially fends off the otherwise violation-free *nau(frá)g]o:

–

>(náufra)g]o nau(frá)g]o (náu)frag]o nau(frág]o) na.(úfra)g]o

*!

– – – –

** *! *! *!

** *

The victory of a. (náufra)g]o over its c. contender (náu)frag]o on AL-Ft,] may in turn look like a mere technicality, given their stress locus identity. However, the defeat of d. *nau(frág]o) precisely on AL-Ft,] provides further evidence of the relevance of this constraint. Noteworthily, (náufra)g]o fulfills Hyde’s (2002:318) GEN’s FOOTCAP constriction imposing an upper limit of two syllables (NB not moras or vowels) on foot size (‘Feet are maximaly disyllabic’). All the data examined are thus accounted for on the set of the three mutually ranked syllable-related constraints *V̆ .V̆ >> MAXPK >> ONSET plus the stress-related subordinate AL-ˈV,]. The ˈ=H cluster to be analysed next poses two challenges, respectively concerning possibility or otherwise of tautosyllabic ˈ=HC. and of heterosyllabic ˈ=.H(C).

4.2 Parse of ˈ=HC. Our discussion of the ˈ=H cluster has up to now been confined to open syllables. Indeed, tradition claims impossibility of tautosyllabicity in stressed closed sylla-

Gliding ghosts or ghostly glides, and does it matter which?

79

bles, modulo coda s (Manáus, Cascáis), a segment privileged in more than one way in Spanish, just as in other languages. The reverse of the coin, occasional cluster heterosyllabicity before s (Ema.ús, Ana.ís), simply falls out from lexical /H./ marking, in the familiar way. For consonants other than s, Harris (1983) acknowledges the three everyday words véinte, tréinta, áunque as counterevidence to his alleged ˈ=HC. compulsory heterosyllabicity,29 but deems them irrelevant relics: “[N]ative reaction to nonce forms containing such clusters is strongly negative. (A typical comment: ‘¡Horror! ¡¿Cómo puede nadie decir tales cosas?!’)” (p. 31).30 Ungrammaticality in the face of such daily occurrence would nonetheless be surprising: think, by contrast, of the ordinary Spanish speaker’s inability to assign stress outside a word-rightmost three-syllable window (e.g. in foreign names like Finnish Kálevala or Dutch Schéveningen), or to realize the palatal consonants [ɲ], [ʎ] word-finally (Spanish speakers systematically say [l] for Catalan [ʎ] in, e.g. Maragall). In stark contrast, ˈ=HC tautosyllabicity is unproblematic in the three words in question.31 Harris’s strong claim indeed got gradually liberalized in the subsequent literature, albeit still not fully abandoned. Thus, Hualde (1991) excludes coda n and z (= [θ] in the Castilian variety) from the scope of the restriction so as to accommodate Basque names like Induráin (a celebrated cycle champion) or Zaráuz (a seaside Basque town), both with the cluster tautosyllabic throughout Spain. Also the established English loan cl[au]n (clown), supplementing Harris’s patrimonial aunque, and “in foreign names” f (Dutch Cruyff ) and [x] (Catalan Puig, with Catalan (i)g# = [tʃ ] interpreted as [x] by Spanish speakers).32 The evidence on the ground actually goes further. There is word-medial éil in the also Basque surname Aréil.za.33 Indeed, with obstruents (not necessarily fricative) in foreign words like apart.heid (NB foreign h = Spanish [x]), raid, leitmotiv, forfeit, bobsleig, or proper names in the style of Leipzig, Zweig, Reich (ch = German [ç] but Spanish [tʃ ]), Freud, Lloyd, Kuwait, Farenheit. Admittedly, not all speakers give them the pronunciations suggested by the spelling: the 29 Also auxilio, with the cluster nonetheless stressless and on assumption of an x [ks] pronunciation by no means general in ordinary Spanish: Harris’s informants are mainly Mexican (cf. e.g. 1983: 139, fn. 6), a fact possibly not irrelevant in the present connexion. 30 Reliance on informal native informant probing can be problematic precisely on account of the informality, with the concomitant poor control of the contextual and linguistic parameters, moreover often not made fully explicit. The small size and haphazardness of the samples may also render them unrepresentative. 31 At least in ordinary standard speech. In veinte, treinta, the diphthong may simplify to [e] in the compositesː ve(i)ntiuno, ve(i)ntidós, etc. Aunque appears to be more resilient to reduction in the standard. 32 The high-vowel cluster [ui] falls outside our present remit. 33 Opportunely borne by a leading politician of the Spanish Franco and immediately postFranco eras often talked about in the media.

80

Iggy Roca

coda obstruent may be left out, and the vowel cluster reduced to the peak. However, the full spelling is realisable in more careful speech, in sharp contrast with, e.g. the systematic Spanish prothesis on English word-initial vowel-less s: slogan, stop, etc. are indefectibly [e]slogan, [e]stop, etc. in Spanish-speaking lips. Bunches of everyday English words now being imported without apparent native resistence further attest to the Spanish legitimacy of tautosyllabic VˈHC: cf. e.g. mail [mˈeil], (gas/fuel)oil, hidrofoil, trail (in contrast with older ra.íl), soul (cf. Se.úl), etc. Last, but not least, a significant number of Spanish-speaking areas favour unrestricted ˈ=HC tautosyllabicity: “La preferencia del habla vulgar por el diptongo [. . .] hace que en ella abunde más la dislocación del acento en favor de la vocal más abierta: [. . .] máiz, ráiz, bául, páis [. . .]” (Menéndez Pidal 1962: 39, fn 1). Indeed, this pronunciation need not be felt as vulgar in areas outside the Castille sphere of influence: “La sinéresis de laúd, baúl, país, maíz [. . .], frecuente entre el vulgo en España, y muy extendida, aún en clases más altas [emphasis mine], en América [. . .]” (Navarro Tomás 1959:161). In sum, thus, whatever the traditional tendency for ˈ=H heterosyllabicity before a coda consonant other than s, tautosyllabicity is now found both in non-Castilian Spanish names (Arnáiz, Aréilza) and in foreign loans (m[ˈei]l, óil). Indeed, more widely in some varieties, even at the price of stress shift where necessary to overcome the sonority strictures on tautosyllabicity with regard to the head stacking and sonority peaking requirements on syllabicity of 2.2 above: ma.íz -> máiz, ba.úl -> bául, etc. This type of shift historically led precisely to the ˈ=HC. tautosyllabicity of véinte (< veínte < viínti < vi:gínti) and tréinta (treínta < tre:gínta < tri:gínta (cf. Menéndez Pidal 1962:186), in the mould of vagína > vái.na and regína > réi.na, both however with the source syllable open. The áunque issue is addressed on p. 86 below. Both patterns in hand are also accounted for on our present machinery:

>Indu(ráin)]

–

–

In(dúra)in] In(dúrain)] Indu(rá.in)] Indura.(ín)] /i./

– – – –

– – – –

>Efra.(ín)]

–

E(fráin)]

–

* *! *! *!

** **! *

* *!

*

Gliding ghosts or ghostly glides, and does it matter which?

81

4.3 ˈ=H. obligatory tautosyllabicity The second issue regarding the Spanish ˈ=H cluster concerns its obligatory tautosyllabicity in an open syllable. The relevant =H stress-cum-syllabification abstract combinatorics and their respective empirical status in Spanish follow: ( =.ˈH. ( ˈ=H.

*=ˈH. *ˈ=.H.

The impossibility of *=ˈH also falls out from the UG strictures of 2.2 above, but the ungrammaticality of *ˈ=.H needs addressing. The issue is not mentioned in Harris (1983: cf. e.g. table 1.21 on p. 14). The subsequent literature did note the impracticability of the configuration, ruling it out on a range of devices couched in various theoretical frameworks, among them the following: Roca (1991:624, (60)):

Condition: ‘V1 bears the stress peak and is more sonorous than V2’ Roca (1997a:261, (61)): ~ ˈV.H, ‘for V more sonorous than H’ Harris and Kaisse (1999: 140): Postvocalic Denuclearisation rule V [+high] | | N N

–>

V [+high] | — | N

([+high] is unstressed)

Roca (2006:258, (37)): SON FALL ’ ‘Tautopodal abutting vowel sequences of falling sonority are tautosyllabic’ The four formulae are manifestly equiparable, modulo Roca’s (2006) tautopodality condition: they de facto involve a description of the unwanted structure followed by its (static or dynamic) prohibition. Our current ranking accounts for both lexically unmarked and /H./-marked inputs, respectively leading to surface tauto- and heterosyllabicity. Consider tautosyllabicity in the absence of /H./:34 34 The resemblance with Rosenthall’s (1994) ((33), p. 17) ONS >> NODIPH ranking is manifest. NODIPH plays no part in our analysis, free of some of the clutter present in Rosenthall’s overall machinery, perhaps attributable to misconceptions about the data.

82

Iggy Roca

>(gái)t]a

–

–

ga.(í)t]a

–

–

>po(lái).n]a

–

–

pola.(í)n]a

–

–

>bon(sái)]

–

–

bonsa.(í)] bon(sá.i)] (bónsai) (bónsa)i]

– – – –

– – – –

* *! * *! * *! *!

* **! **

*!

/M/ marking would be inconsequential here, given the winners’ systematic foot binarity also without it. In turn, /H./ marking as usual enforces heterosyllabicity, with the stress on H automatically ensuing (cf. again 2.2 above):

>A.(í)d]a

–

*

(Á.i)d]a (Ái)d]a /i./

– –

*

>para.(í)s]o

–

*

pa(rá.i)s]o pa(rái)s]o /i./

– –

*

>yata.(í)]35

–

*

yat(á).i] yat(á.i)] ya(tái)]

– – –

* *

*! *

*!

*! *

*!

*!

*!

* *!

*

35 Yataí: “NE Arg. y Ur. Planta de la familia de las palmas [. . .]” [DRAE]. The shorter everyday everywhere ahí [a.ˈí] makes the same point.

83

Gliding ghosts or ghostly glides, and does it matter which?

So far, therefore, our basic syllable-related four-constraint ranking stands. However, the remaining double-marked /i., M/ alternative poses a challenge, precisely the one that the four aforementioned alternative constraints aim to circumvent. Our current ranking yields ungrammatical PO stress with heterosyllabicity:

*ya(tá.i)]

*

♦yata.(í)] yat(á).i] ya(tái)]

*! *!

*

* *

*

* *

*!

The b. and d. configurations, both legitimate and b. (yata.í) factual, actually follow from this ranking, respectively on /i./ single marking and no marking. On the double /i., M/ marking, however, ungrammatical a. comes out the winner: d. is ruled out on MAXPK’s dominance over ONS, and b. and c. score worse than a. on BINFT M. Roca’s (2006) tautopodality condition does ward off this result. However, the condition is tautologically inoperative on a non-tautopodal cluster: cf. yat(á)].i, for instance, with the desinential ]i# parse legitimate on Richness of the Base.36 We will overcome the difficulty with a MAXPK-dominant constraint *ˈV.H, the common denominator of the four alternative proposals listed above:37

/i., M/ >*ya(tái)] ♦yata.(í)] yat(á).i] yat(á)].i yat(á.i)]

* *! *! *!

*ǃ * *

* * * * *

*

* *

36 The OT principle of “Richness of the Base” (i.e unconstrained base = lexicon) rules legitimate all lexical representations compatible with GEN, hence confining responsibility for surface grammaticality to the constraint ranking. 37 In Martínez-Paricio (2013:183) the construct “coerced weight” causes moraicity of a wordfinal “glide” or consonant. On this basis, caμ.(raμhj μ) caray (“μ” = mora; “h” = foot head)

84

Iggy Roca

The winner yatái is fully consistent with the now minimally augmented grammar. It does not match this word’s attested yata.í pattern (it does that of the also factual bonsái), irrelevantly, however, since yata.í is derivable on /i./ single marking, as indeed was shown above (p. 82).

5 Favoured syllabifications The present ranking favours tautosyllabicity, checked by lexical /H./ where relevant. Tautosyllabicity is nonetheless infrequent in word-initial C(L)Hˈ= clusters, modulo two classes of systematic exceptions, respectively concerning the Kuˈ= sequence (K = velar consonant) and both vowel clusters iˈe, uˈe, all of which configurations are tautosyllabic, barring a true handful of idiosyncratic exceptions. It thus follows that some further machinery is needed.

5.1 #C(L)Hˈ= heterosyllabicity Our account of exceptional H= heterosyllabicity by resort to /H./ and the corresponding dominance of MAXPK over ONS is faced with the paradox that the opposite situation prevails word-initially post-consonantally, with H.ˈ= favoured over Hˈ=:38

i

á pi.áno, ti.ára, di.álogo, fi.ámbre

ó pi.ójo, mi.ópe, Di.ógenes, Ri.ója, pri.ór

u

su.áve, Du.árte, Su.árez, Lu.ánda

Su.ómi

defeats caμ.(raμh.iμ) on both ONS and a constraint *HDσHighV disfavouring high vowels in the syllable head: cf. M-P’s tableau (44). Word-medially, however, coerced weight is declared inoperative by fiat, and consequently the “glide” is not moraic, and thus not a syllable head. The strategy seems equally applicable to “ordinary” vowels (i.e. not “glides”) and could therefore be painlessly brought into our analysis. The restriction of coerced weight to the word-final segment is however ad hoc, and the complexity inherent in the incorporation of moras into Spanish seemingly not a price worth paying: cf. e.g. Dunlap (1991) for a thorough but ultimately unsuccessful attempt. 38 The fact is mentioned in Roca (1991:621) and restated in Hualde (1997), which provides a useful survey of Spanish /i/ syllabification patterns. Postconsonantally because in absolute word-initial position the tendency is for a pre-= H to parse, either as an onset, with concomitant consonantalisation (cf. yérba vs. ciérva: akin discussion in Roca 2005c: 211ff), or in the nucleus, ͡ ͡ tautosyllabically with the following vowel (ierba, iato), unless /H./ marked (cf. i.áto vs. yate, yambo, yodo). Navarro Tomás’s (1959:159) parses su.ave, cru.el, ti.ara, pri.or, pi.ano, bi.ombo, mi.asma, hi.ato, “on etymological tradition”, are all also mine.

Gliding ghosts or ghostly glides, and does it matter which?

85

This tendency is itself subject to exceptions, some systematic and some idiosyncratic, the number and distribution of the latter moreover subject to important dialectal, indeed idiolectal, variation.39 The situation thus betrays ongoing language change in the direction of tautosyllabicity, nonetheless slowed down by the seemingly universal tendency to word-initial phonic robustness (Beckman 1998, Smith 2005), here materialized as H syllabic autonomy. I am only aware of two RA exceptions to #C(L)Hˈ= heterosyllabicity, one in two allomorphs: 1. dios/diosa (arguably /dios](a)/ lexically), and 2. diócesis. On our current machinery suspension of tautosyllabicity implies /H./ marking, here hence necessary across the board but in dios(a), diócesis, idiosyncratically excluded. Consider first the ordinary syllabification pi.ano:

>pi.(á)n]o

–

pi(á)n]o

–

* *!

The ongoing /H./ loss eventually brings on tautosyllabicity:

>pi(á)n]o

–

–

pi.(á)n]o

–

–

*!

The resistance to /H./ loss expected from the habitual linguistic conservatism is here countered by the joint action of a number of forces. One may be identified with pressure from dialects/idiolects at a more advanced stage on their way towards tautosyllabicity and possibly perceived as more prestigious. In normal circumstances, however, the weight of such pressure should be 39 The experiment reported in Cabré and Prieto (2006) yielded revealing, if expected, results. Their significance is nonetheless undermined by the small size and randomness of the informant sample: 15 speakers of Peninsular Spanish mainly from Madrid and Barcelona (7 vs. 5) plus 3 from two additional cities (2 Valencia, 1 Granada), less than a drop in the ocean of the 400(?) million Spanish speakers world-wide, indeed of Spain’s 46 odd million. For pi(.)áno and di(.)álogo they report a roughly 50/50 distribution between hetero- and tautosyllabicity (p. 233).

86

Iggy Roca

moderate, given both the usual paucity of sustained interdialectal daily contact and most speakers’ (unconscious or deliberate) common loyalty to their own pronunciation. A second, more effective force promoting stressed cluster tautosyllabicity concerns the regular heterosyllabicity loss in ordinary discourse in positions away from the phrase accent (cf. Roca 1986, 1991, Simonet 2005. Chitoran and Hualde 2007), as now illustrated, with phrase-accented vowels identified by underlining: a. b. c. d.

ayer me ayer me ayer me ayer me

compré un pi.áno compré un pi(.)áno muy cáro compré un piáno de cola carísimo compré un piáno de cola carísimo de veras preciosísimo

In a. the á of piano carries the phrase accent, heterosyllabicity consequently the norm in conservative dialects. In b., however, the phrase accent is borne by the now rightmost cáro, and concomitantly i.a becomes liable to tautosyllabicity, particularly in faster speech. In c., the phrase accent rí in carí(simo) is sufficiently removed from piáno for the cluster to turn up tautosyllabic in ordinary deliveries, i.e. not artificially slowed down as in, e.g. a dictation exercise addressed to struggling school pupils. And a fortiori likewise in d. It thus follows that all Spanish speakers, those of more isolated, conservative dialects included, are liable to well-nigh daily exposure to #C(L)Hˈ= tautosyllabicity, from the speech of others and, indeed, of themselves.40 Such hiatus loss obviously takes place outside the word domain pertinent in pianíto, pianísta, pianóla, . . . The identification of the higher prosodic domain apposite for phrases lies outside our present focus, and therefore the matter cannot be gone into here. Relevantly, the historical syllable merger of the formerly heterosyllabic a.ú of aunque (< a.ún + que) still pending analysis can be readily understood from this perspective: cf. e.g. á͡unque dices que no vienes mañána . . . , from a previous a.ún que . . . Compare current fast speech no sé qué hacer a.ún vs. á͡un no sé qué hacer.41 40 Simonet’s (2005) experiment brought out a gradual positive correlation between cluster heterosyllabicity and proximity to the stressed syllable. It involved 12 Spanish native speakers from Palma (Mallorca, Spain) rating 1–5 the likelihood of H= tauto- vs. heterosyllabicity in 15 existing Spanish words and 30 invented ones. Both word groups included three sets of equal size respectively with the cluster in tonic, pretonic, and propretonic position. The findings in Chitoran and Hualde (2007) point in the same direction. 41 An anonymous reviewer objects that aunque does not bear word stress (a view expressed in Quilis 1993: 393 and perhaps implicit in Navarro Tomás: 193, §170b; IR), and consequently that the datum is irrelevant. The point intended here, however, concerns the contrast in surface

Gliding ghosts or ghostly glides, and does it matter which?

87

An account of word-initial #C(L)H.ˈ= heterosyllabicity grounded on lexical /H./ is thus practicable, and indeed is proposed in Martínez-Paricio (2013: 184– 185). It however incurs the obvious paradox inherent in the lexicalisation of the common. In particular, the lexicon is definitorily the repertory of the unpredictable, and consequently it is desirable in our context to keep lexical representations peak-free. We will thus instead attribute word-initial Hˈ= heterosyllabicity to a sufficiently high-ranking constraint *#C(L)H͡ ˈ= inhibitory of H͡ ˈ= after a #C(L) onset. We formulate it as follows, with Hˈ= underlined as a visual reminder of its focus statusː *#C(L)H͡ ˈ=

No Post-CH͡ ˈ= tautosyllabicity word-initially

*#C(L)H͡ ˈ= needs ranking above ONS and below *V̆ .V̆ . Such domination by *V̆ .V̆ allows a stress-free reformulation *#C(L)H͡ =, with overapplication (*pi.anóla ͡ for pianóla, for instance) prevented precisely by *V̆ .V̆ , all as the following evaluations demonstrate:

>pi.(á)n]o

–

–

pi(á)n]o

–

–

*!

>pia(nó)l]a

–

–

*

pi.a(nó)l]a *!

–

–

*

*

We know, however, that *#C(L)H͡ = is challenged in RA by the idiosyncratic tautosyllabicity of diós(a) and diócesis (and naturally more words are present in less conservative accents). The fact itself is unremarkable, exceptions of course being common currency in phonology. The problem, rather, concerns the formalization of this particular type of exception in these particular circumstances. Thus, OT constraint contravention is programmatically legitimized on a higher ranking of an antagonistic constraint, obviously ONS here. However, the Spanish ranking of ONS below MAXPK is by now well established, and indeed stress location between aunque (lexical stress status irrelevant) and lexically stressed aún, where aunque it originated: under no circumstances would modern aunque be stressed *a.únque, although phrasally it can be áunque, and indeed aunqué, both compatible with lexical stresslessness: cf. áunque lluéve vs. aunqué despertaré tárde. The historical stress loss or shift from aún is hence supportive of our contention.

88

Iggy Roca

follows from the logic of the model, as the function of MAXPK involves precisely ONS blockage. The converse ranking now apparently required hence gives rise to a paradox akin to the one we saw above, was faced by Rosenthall (1994) in a different context. An obvious way forward involves lexicalising H non-peakness in dios and alike as a reflection of their markedness. On the old SPE feature system this strategy would simply entail specification of the vowel as [-syllabic]. However, the feature [±syllabic] (a de facto endorsement of the specious construct “glide”, [-consonantal, -vocalic] in the original SPE formalisation and [-consonantal, -syllabic] in the subsequent alternative)42 is now all but abandoned: the syllable

42 Harris and Kaisse (1999:187) deem misguided Roca’s (1997) construal of the “vowel” vs. “glide” contrast as akin to the SPE binary feature [±syllabic], arguing that the issue is merely terminological, and that Roca’s /H./ device amounts to [+syllabic]. The matter is not disposed of so easily, however. Thus, a distinctive feature is standardly meant as a component of a segment, the way a proton and a neutron are components of an atom, whereas a syllable, and therefore a part of a syllable such as the nucleus, is a unit of grammatical structure. In particular, any type of segment, and hence of distinctive feature, can in principle parse in the syllable nucleus peak, unrestrictedly in fact in the familiar Imdlawn Tashlhiyt Berber dialect (cf. Dell and Elmedlaui 1985). More modestly and closer to home, both liquids and nasals can make up a syllable nucleus in English, but not in Spanish. Indeed, the English free variation between a nuclear and a non-nuclear nasal, in butt[n̩] vs. butt[ən], for instance, exclusively concerns structure, the phonological matter of the segment remaining constant: [+nasal], etc. in both cases. All this is of course well-known and beyond dispute. In direct contrast, the “vowel” vs. “glide” distinction effects a split of what in Spanish is manifestly one single phoneme: all Spanish ͡ speakers can at will pronounce written piano as pi.ano or piano, a device indeed made use of by poets (dieresis, syneresis), but both the word and the relevant sounds remain the same in both cases. This reality is entirely obvious to any speaker of the language (there simply cannot be any truly contrastive minimal pairs) and is hence beyond dispute. The pedigree of the “glide” construct in fact goes back to a time when the syllable was not countenanced formally and “glide” was brought in as as a de facto partial functional substitute. It is consequently no surprise that the phonetic substance of “glide” is traditionally left unstated: cf. SPE’s early [-consonantal, -vocalic], with the two negative specifications in effect relegating the segment to a phonetic limbo, or its subsequent reformulation as [-syllabic], a de facto admission of its syllabic, not phonetic, nature. More modernly, the substance of “glide” is usually left implicit, as if treading on eggshells. In a rare recent attempt to look explicitly into the matter generally, Nevins and Chitoran (2008: 1841) argue that “the glides [j,w] are not just non-nuclear versions of vowels, but also bear a subsegmental difference, which we propose is due to a feature [±vocalic]”, an implicit clear recognition of the place-holder role that the term is commnonly alloted in the literature. In the specific case of Spanish, there is simply no glide phoneme, since there is no contrast hinging on it. Phonetic research by Aguilar and associates reached the same conclusion, as would indeed be expected: cf. “the syllable components are planned before the phonetic realisation; so, the acoustic result of hiatus – two vowels in two syllables – has to be necessarily different from the result of diphthong – two vowels in one

Gliding ghosts or ghostly glides, and does it matter which?

89

rejoined the mainstream in the 1970s and has since enjoyed well-nigh universal acceptance.43 Furthermore, on the subsequent X-bar modelling now prevailing, a nucleus prehead of necessity implies a head, and consequently full nucleus lexical specification is required here: * / / /dios/, /diócesis/

( /| /| /dios/, /diócesis/

This representation runs counter the well-established practice of keeping prosodic structure out of the lexicon: cf. Levin (1985) and Blevins (1995) for theory and Levi (2004) for empirical evidence from a range of languages worldwide. Our present /H./ lexicalisation of irregular nucleus heads represents a minimal compromise imposed by fact and with precedents in the literature both for Spanish and for other languages (cf. fn. 20 above). Furthermore, compelling evidence against full nucleus lexicalisation is available from Spanish itself. Thus, lexical encoding by definition implies arbitrariness and rote learning, novel data consequently being expected to be free of it. Consider in this light the ͡ man’s first name Dionísio (Eng. Denis), with the stressless word-initial cluster io͡ ̆ ̆ tautosyllabic on *V.V. Now, a Spanish word truncation procedure (cf. Roca and ͡ Felíu 2003) selects the base’s first two syllables (Dio.ni) as the segmental body for the truncate’s designated trochaic foot, here hence with the o stressed and the final i stressless: (óni). The relevant issue concerns the new syllabic status of the cluster-initial i.: heterosyllabic in RA and similar accents, Di.óni,44 providing evidence of *#C(L)H͡ = productivity. Productivity implies default, and thus grammar, not lexicon. Consequently, we shall maintain *#C(L)H͡ = and formalize the exceptions (only 3 (= 2?) in RA, syllable, which requires a restructuring in time and frequency to adjust both vocalic segments to the syllabic frame” (Aguilar 1999: 72). The real life fact is thus, not that the segments in question are “glides” (a non-existent category), but that they (as vowels) can parse tautosyllabically, a possibility indeed realized in Spanish and appositely captured by our apparatus. Even more basically, the difference does not concern substance (purely vocalic in both), but structure (syllable nucleus peak vs. something else), and hence it cannot involve a distinctive feature. 43 The exception chiefly concerns alternative theories of syllable-internal structure, some going as far as rejecting a unified status for the construct. A reemergence of phonetic reductionism has also recently been taking place in some quarters of the general field. 44 The nickname of a Spanish law-breaker of the 1980s then well-known and often talked about, the datum hence pretty robust (223,000 Google hits on 24.12.2014). Comparable results are available from a variety of data originally novel to the speaker: unfamiliar words encountered in writing, foreign words (names in particular), specifically made up words (readily testable), acronyms, etc.

90

Iggy Roca

as we know) through lexicalization of the tautosyllabic nucleus cluster. The ͡ following comparative evaluations of diócesis and Di.ógenes illustrate, with /io/ informally standing both for the lexical mark itself and for the constraint protecting it:

͡ /M/, /io/ >di(óce)s]is di.(óce)s]is dio(cé)s]is dio(cés]is)

*! *!

–

*

– – –

* *

* *

* *!

͡ Compare Diógenes, without [io]:

/M/ >Di.(óge)n]es

–

Di(óge)n]es Di.o(gé)n]es (Dí.oge)n]es (Dí.o)gen]es

– – – –

– *!

*!

– – – –

*

*

*!

* * * *

*!

**! **

On an alternative strategy involving constraint cloning (cf. Pater 2000, 2007, 2010), ONS would be cloned onto ONSL, itself ranked above a constraint H ⇒ SYLPK (“a high vowel constitutes a syllable peak”) related to the sonority hierarchy (Prince and Smolensky 1993/2004, Baertsch 2002). A ranking ONSLEX >> H ⇒ SYLPK >> ONS indeed yields both Mári]oONSLEX and Marí]a. The need for lexical marking of empirically unmarked Mario, rather than of marked María, contradicts the direction of Spanish markedness, however, and consequently our proposed lexical solution is superior.

5.2 Two overriding principles The two anticipated classes of systematic exceptions to *#C(L)H͡ = are still pending. They inevitably compel some further additions to our original 3+1 constraint grammar (*V̆ .V̆ >> MAXPK >> ONSET >> AL-ˈV,], now incremented with *#C(L)H͡ =.

Gliding ghosts or ghostly glides, and does it matter which?

91

5.2.1 Kuˈ= tautosyllabicity The Kuˈ= sequence almost invariably turns up tautosyllabic in the three word positions: initially, medially and finally. The following sets of data illustrate for the three velar consonants [k, g, x]: #_ /k/ cuádra, cuáquero, cuóta /g/ guápo, guánte, guárdia /x/ Juán, Juána, Juánez

#. . ._. . .# escuádra, renacuájo, acuóso antiguálla, vaguáda, iguána Tijuána, enjuágue

_# pascuál, locuáz, licuór iguál, jaguár, zaguán ajuár

I am only aware of two RA exceptions: Ju.árez word-initially and capicú.a word-finally. They are both historical legacies: 1. Ju.árez, reportedly a descendent of Medieval Galician Xu.árez, with x standing for the pre-palatal fricative [ ʃ ], not the present-day Spanish velar [x]; and 2. capicúa, an encapsulation of the Catalan phrase cap i cu.a [kˈu.a] ‘head and tail’, now used as a Spanish word by Spanish-only speakers unaware of the source. Kuˈ= tautosyllabicity also has its roots in history. First, the u of qu in the Latin orthographic cluster did not stand for a full-fledged vowel [u], but for a lip rounding gesture on /k/: phonetically, thus, Spanish [ku̯V] < Latin [kwV], e.g. Lat [kwa]ndo > Sp [ku̯a]ndo ‘when’ (cf. Väänänen 1963: 53; Monteil 1970: 55–57). Spanish also inherited the [u=] cluster from other languages: Germanic (wardôn > guardia; waiþanjan > guadaña; want > guante), Arabic ((al-)Wadi al-Kabir ‘(the) big river’ > Guadalquivir, a river in Andalusia), Amerindian (Nahuatl ahuacatl > aguacate; ahuaca-molli > guacamole), and (now increasingly) from English (watchman > guachimán; Washington > [guásinton]). Cf. also French ouate > Sp guata. Syllable-initial [u=] has thus undergone the general Spanish w -> [gu] segmental reinforcement in all cases, with eventual g_ respelling at times. The change also affected the Latin monophthong-originated huerta, huevo, hueco, etc., now all [gue] colloquially. We shall construe the corresponding synchronic structure as a reflection of its historical predecessor. Simply, after a velar consonant the habitual Spanish u parse as the first element of a complex nucleus uV yields to a parse as the second element of a Ku complex onset: (NB onset/nucleus) k/ieto (quieto), l/uego (luego), but ku/ento (cuento), gu/ante (guante). As a consequence, u will lack stress bearer status (in Spanish as we know confined to rime vowels), and will hence be excluded from the stress computation. We will formalize the situation on a constraint *K/u= interposed between MAXPK and *#C(L)H͡ =, as shown in the next tableau. As usual, high-ranking ͡ is now omitted from the evaluation as irrelevant: FAITH-SEG is assumed; /io/ *K/uV

No uV complex nucleus after K

92

Iggy Roca

>(cu/án)d]o

–

–

c/u(án)d]o cu.(án)d]o

– –

– –

*!

* *!

Note that we are interpreting *#C(L)H͡ = as orthogonal to Ku/=: otherwise a. (cu/án)d]o would violate *#C(L)H͡ = and lose to c. *cu.(án)d]o on *#C(L)H͡ = >> ONSET.45 In the absence of *K/u=, the legitimate [Ku̯ =] winner would fail to come through on account of the antagonistic *#C(L)H͡=. The next tableau demonstrates, with *K/u= crossed out for visual explicitness:

>*cu.(án)d]o

–

–

♦cu(án)d]o (cú.an)d]o

– –

– –

* *! *

*!

The PPO# pattern of ventrílocuo would also fail to emerge, with or without the /M/ mark:

/M/ >*ventri(lócu)]o

–

♦ven(trílo)cu]o ♦ven(trílocu)]o ventrilo(cú.)]o ventrilo(cú.]o)

– – – –

*!

>*ventri(lócu)]o

–

–

♦ven(trílo)cu]o ♦ven(trílocu)]o ventrilo(cú.)]o ventrilo(cú.]o)

– – – –

– – – –

* *! * *!

** **!

* * *!

*! *!

** **!

*

45 A straightforward implementation of this exclusion would involve a reformulation of *#C(L)H͡ = as *#C(L)/H͡ =.

93

Gliding ghosts or ghostly glides, and does it matter which?

*Ventrilócuo is indeed the common pre-adulthood pronunciation of the word (itself possibly familiar from the world of entertainment), then persisting in less sophisticated speakers. The eventual incorporation of *K/u= in more careful speech brings on the standard realisation ventrílocuo:

–

>ven(trílo)cu/]o ventri(ló)cu/]o ventri(lócu/]o) (véntrilo)cu/]o ven(trílo)c/u]o ven(tríloc/u]o)

*!

– – – – –

* *! *ǃ *ǃ

**! ** **

*

The same ranking leads to alícuota, a technical term only accessed later in life, if indeed at all:

>a(lícu/o)t]a

–

a(líc/uo)t]a a(líc/u)ot]a (áli)cu/ot]a ali(cu/ó)t]a alic/u(ó)t]a alic/u(ót]a)

– – – – – –

*! *!

* *! *!

** ** **

* *!

* *!

*

Comparison with the familiar /M, i./ double-marked período on the updated ranking will be instructive:

>pe(rí.o)d]o (pério)d]o (péri.)od]o (péri.o)d]o peri(ó)d]o peri.(ó)d]o

*

*

*! * * *! *!

* *

*!

** ** **!

94

Iggy Roca

MAXPK is of course instrumental to the defeat of *périodo. However, it is inoperative in alícuota, metrically thus an ordinary PPO word on a par with, e.g. amígdala. The exceptional heterosyllabicity of capicú.a and Ju.árez simply involves lexical /H./, preserved on the MAXPK >> *K/u= ranking:

>capi(c/ú.)]a

–

ca(pí)cu/]a capi(cu/]á) /u./

– –

>Ju.(árez)]

–

Ju/a(réz)] Ju.a(réz)] Ju.(á)rez] (Ju/árez)] (Ju/á)rez]

– – – – –

*!

* *! *!

*

*!

*

*

*

* *

* * * *

*

*! *!

* *

*! *

The progressive tautosyllabic pronunciation Juárez follows from the familiar ͡ and the /H./ loss. The contrast between tautosyllabic acuóso, aguóso, with Ku=, K-less heterosyllabic monstru.óso, sinu.óso, suntu.óso, tortu.óso, etc. further supports the account. 5.2.2 Hé tautosyllabicity Hˈe tautosyllabicity is also robust across the board, although again not exceptionless. It is also a product of history. As is well known, the stressed Vulgar Latin mid open front vowel [ˈɛ] developed into the Spanish diphthong ié, and its back rounded correlate [ˈɔ] into ué: Latin: Spanish:

Latin: Spanish:

c[ˈɛ]ntu(m), ciénto,

d[ˈɛ]nte(m), diénte,

n[ˈɛ]ve(m), niéve,

t[ˈɛ]nda(m) tiénda

c[ˈɔ]rda(m), cuérda,

c[ˈɔ]rpu(s), cuérpo,

sch[ˈɔ]la(m), escuéla,

s[ˈɔ]mnu(m) suéño

f[ˈɛ]bre(m), fiébre,

f[ˈɔ]cu(m), fuégo,

g[ˈɛ]lu(m), hiélo,

p[ˈɔ]rta(m), puérta,

hib[ˈɛ]rnu(m), inviérno,

r[ˈɔ]ta(m), ruéda,

Gliding ghosts or ghostly glides, and does it matter which?

95

Latin occurrence of the two mid-low vowels [ɛ, ɔ] was restricted to stressed position, their mid-high counterparts [e, o] turning up instead when stressless. This eventuality is frequent in derivatives, consequently with the corresponding Spanish vowel non-diphthongal:46 [e] cf. [ˈɛ]

centéna, ciénto,

[o] cf. [ˈɔ]

cordél, cuérda,

dentáda, diénte, corpiño, cuérpo,

febríl, fiébre, fogón, fuégo,

heláda, hiélo, portázo, puérta,

invernía, inviérno, rogatíva, ruégo,

nevada, niéve, escolár, escuéla,

tendéro tiénda soñoliénto suéño

The two historical processes in question left behind a substantial trail of ié ~ e, ué ~ o alternations. They tend to be systematic in the verb conjugation, with only a small handful of exceptions, all now minimally sampled (cf. téngo):47 INFINITIVE

1st PERSON SG

1st PERSON PL

3rd PERSON PL

REGULAR apretár dormír

PRES INDIC apriéto duérmo

PRES INDIC apretámos dormímos

PRES SUBJ apriéten duérman

IRREGULAR tenér

téngo

tenémos

tiénen

In non-verbs the situation is essentially akin (cf. centéna ~ ciénto, etc. above), with the important proviso that root-bounded diphthongs are compatible with level-2 suffixes (germane discussion in Halle, Harrris and Vergnaud 1991; cf. also Bermúdez-Otero 2006): dientecíto (cf. dentál), fiebráza ( febríl), cuerpecíto (corporál), buenísimo (bondád), ciertísimo (certéza), etc. The mechanics of Spanish diphthongization has attracted attention through the years, and consequently a number of formal proposals are available in the literature. The matter is however orthogonal to our present concerns, bearing on {=, H} syllabification rather than segmental mechanics. From our perspective, a constraint *H.e in tandem with a *V̆ .V̆ >> *H.e ranking will thus suffice: *H.e

No [e] heterosyllabicity immediately after a high vowel

46 The non-diphthongal alternant may of course have a historical learnèd origin, rather than word of mouth transmission, concomitantly with written Latin as the source of the monophthongal mid vowel. 47 These forms are commonly conceived of as ‘irregular’, e.g. tèngo, vs. regular tiènes, etc.

96

Iggy Roca

The synchronic scope of Hé tautosyllabicity goes beyond alternating Hé, as the following words (some, inconsequentially, proper nouns) minimally illustrate:48 –

quiéto, riénda, biéla, Piéra, Riéra, Viéna, noviémbre, ciéncia, quiniéla, ambiénte, aliciénte, izquiérdo, audiéncia, Zubiéta, Babiéca, puérro, duénde, asuéto, berruéco, . . .

Our *H.e constraint makes no reference to alternation and consequently also takes care of Hé tautosyllabicity in this set.49 *H.e needs to outrank *#C(L)H͡ =, antagonistic in word-initial position. Thus, minimally: *H.e >> *#C(L)H͡ = >fiéra fi.éra

* *!

*H.e meets with a handful of exceptions, now sampled for RA (mutatis mutandis in other accents): i

cli.énte, Vi.élla (cf. diénte, biéla)

u

cru.él, Noru.éga, Samu.él, Marru.écos (cf. cuéllo, alcahuéta, Manuél, berruéco) *H.e therefore necessitates domination by MAXPK:

/u./

MAXPK >> *H.e >> *#C(L)H͡ =

>Samu.él Samuél /i./

* *!

>cli.énte cliénte

* *!

*

48 English notoriously splits the cluster into two syllables, hence Di.ego, Vi.ena, Ki.ev and such like, in a manner wholly alien to Spanish. Heard several times in a row on a British radio broadcast as I was writing this paper: (ˌhazi).(ˈenda) for Haçienda, the former celebrated Manchester ͡ club, from Spanish tautosyllabic hacienda [asiˈenda]. 49 Hualde (1999) appeals to analogy (“patterns in the lexicon”, i.e. majority pressure). Chitoran and Hualde (2007) propose two further factors, one supportive of tautosyllabiciy and one inhibitory: respectively a general articulatory leaning against high vowel hiatus, and a prosodyrelated vowel lengthening tendency.

Gliding ghosts or ghostly glides, and does it matter which?

97

The fragment of the ranking further developed in this section is as follows:

This completes the presentation of the full constraint ranking displayed on p. 64 above. For clarity, the following evaluations illustrate its workings on all the H= clusters just examined:

>fi(é)r]a

–

–

fi.(é)r]a /M/

–

–

>Su.(árez)] Su.(á)rez] Su.a(réz)] *! Sua(réz)] Su(árez)] /u., M/

*ǃ * *ǃ

* *!

*

–

*

– – – –

* *

* *

* *!

>Ju.(árez)]

*

*

Ju.(á)rez] Ju.a(réz)] *! J/u(árez)] J/ua(réz)] /u./

*!

>Samu.(él)]

–

Samu(él)]

–

* * *! *!

* *

* *

* *!

*

* *

* *

*

98

Iggy Roca

6 Conclusion We have seen that metrification of all rime vowels, rather than exclusively nucleus peak vowels, holds the key to both stress and syllabification. The bottom-most constraint AL-ˈV,] draws word stress to the stem end, the default Spanish pattern. Domination of AL-FT,] over AL-ˈV,] impels full foot inclusion within the stem ((órga)n]o), even if at the expense of AL-ˈV,] (*or(gán]o)). The ranking of both *IAMB and BINFT M also over AL-ˈV,] respectively induces trochaic foot shape (*(orgá)n]o) generally and lexically-triggered binary foot size (*or(gá)n]o) where lexically relevant. The prevalence of AL-FT,] generally, over PARSE-V(R) thwarts English-style word-level foot multiplication: cf. Eng (ˌorga)(ˈnici)ty vs Sp organici(dád), not *(ˌorga)(ˌnici)(ˈdad). This stress-related constraint hierarchy connects with its {H, =} syllabification complement counterpart through domination of ONS over AL-ˈV,]. {H, =} parse is ruled by a number of factors related to both stress and syllabification. The primacy of all the constraints but PARSE-V(R) over AL-ˈV,] leads to unmarked stress retraction by one position through abutting vowel tautosyllabicity where feasible (in (Mári)]o, bon(sái)], not in ma(rí)n]o, tsu(ná)m]i), nonetheless averted on ONS-dominant MAXPK where lexically relevant (Ma(rí.)]a, Sina.(í)]). MAXPK subordination to top-most *V̆ .V̆ thwarts stressless cluster heterosyllabicity ͡ (clientéla < cli.énte).50 Three further constraints rankings between MAXPK and ONS provide for additional empirical needs. *#C(L)H͡= promotes Hˈ= heterosyllabicity word-initially in such clusters (pi.(á)n]o). It is in turn overcome by the two higher-ranking mutually level-ranking constraints *H.e (di(é)t]a) and *K/u= (cu(á)tr]o), both subordinate to MAXPK to allow transgression to the benefit of /H./ faithfulness (cli.(én)t]e, Ju.(á)rez]). Last, the prevalence of FAITH-SEG over ONS blocks fulfilment of the latter at the expense of segmental substance: A.ída ≻ *A.Øída, *Aida, for instance. The interaction between markedness and faithfulness has been seen to be critical. “Markedness” (naturalness, in actuality) favours both {=, H} tautosyllabicity and stem-final stress, with faithfulness interfering where necessary to enforce, respectively, H syllabicity and foot binarity, both lexically-determined. A handful of syllabification twists fit the same canon: cf. Hé and Ku= tautosyllabicity and #C(L)H͡ ˈ= heterosyllabicity.

50 Modulo of course the familiar general proviso of dialect, indeed idiolect, diversity in this and some other areas: cf. fn. 20 above. An anonymous reviewer here reports cli.entéla in his own speech, for instance.

Gliding ghosts or ghostly glides, and does it matter which?

99

The role of history in the system needs hightlighting. First, the morphological entity of the Latin theme vowel lives on in the Spanish desinence. Second, the Latin stress locus almost invariably stays put in Spanish. Vowel drop under the appropriate conditions has in turn invalidated the Latin stress algorithm, paradoxically often seen in the literature as the crux of Spanish stress. Third, the Latin binary trochee persisted in words taken from the writing, on the whole exempt from vowel drop and now consequently /M/ bearers.51 Fourth, the (tautological) monosyllabicity of the Latin stressed mid vowels [ˈɛ, ˈɔ] lives on in the corresponding [iˈe, uˈe] Spanish diphthongs. Fifth, the also tautological orthogonality to syllabicity of the Latin [kw] rounding persists in the analogous ͡ Spanish u onset parse Ku/V. Sixth, the regular Latin hiatus of =-adjacent H (cf. Grandgent p. 147ff, section a) survives in the Spanish /H./ lexical marking, itself in the process of piecemeal loss to the benefit of naturalness. The process is however far from complete, the corresponding present-day idiolectal map as a consequence significantly variegated. Relevantly, the analysis has been carried out with no resort to “glides”, the construct thus minimally proved superfluous, and in fact revealed as a garden path inherited from an ill-grounded tradition:52 a reminder of Hyman’s (1985: 77) dictum “Perhaps the most problematic segment type for all theories of phonology is the class of glides” may not be out of place here. Analytically, the rich fabric of relations visualized in the full Hasse diagram on p. 64 above largely remains opaque on the “glide” construal. 51 Also borne by descendants of Latin PPOs with a postaccentual nonfinal syllable nucleus a, a vowel exempt from the historical post-tonic intersyllabic vowel drop (cf. Menéndez Pidal 1962: 77): rábano, huérfano, espárrago. 52 The “glide” construal is obviously deeply engrained in the field. By way of example, a helpful sympathetic anonymous reviewer opportunely writes: “Of course, Spanish does have glides at the level of surface phonetics, so that the rhetoric about absence of glides is a little misleading. The only question is whether the contrast between, say, [du.é.to] and [dwé.lo] is better captured as /dueto/ vs /dwelo/ or as /du.eto/ vs /duelo/”. The remark, reminiscent of Harris and Kaisse’s (1999) addressed in fn. 43 above, clearly merits an upfront response, as follows. Spanish does not have glides at any level, if by “glide” we mean a segment distinct from a vowel. The two words in question both contain the vowel /u/ → [u], the contrast thus not concerning segmental substance, strictly vocalic and identical in both, but rather syllabic parse, respectively in one and two syllables. Any phonetic differences instrumentally observable in the actual delivery (cf. e.g. Aguilar 1999) are artifactual on the syllabic parse (crucially not conversely), the way there may be acoustic differences between the English syllabic n in butt[n̩] and its non-syllabic alternative in butt[ən], mutatis mutandis. This does not seem a particularly hard fact to grasp, but it obviously has been for some time in this connection and apparently continues to be. Put in still plainer terms in an attempt to achieve comprehension, it is the horse that pulls the cart, not the cart that pushes the horse. The difference, of course, involves real-life fact, not mere “semantic” spin.

100

Iggy Roca

The answer to the paper’s title question is therefore no. No, it does not matter which: they both belong in the realm of the imagination, not reality. Down on Earth, “glides” simply do not exist, at least in Spanish.53

References Aguilar, Lourdes. 1999. Hiatus and diphthong: Acoustic cues and speech situation differences. Speech Communication 28(1). 57–74. Alonso, Amado. 1930. Problemas de dialectología hispanoamericana. Buenos Aires: Instituto de Filología, Facultad de Filosofía y Letras, Universidad de Buenos Aires. Alvord, Scott. 2003. The psychological unreality of quantify sensitivity in Spanish. South West Journal of Linguistics 22(2). 1–12. Baertsch, Karen. 2002. An Optimality theoretic approach to syllable structure: the split margin hierarchy. Bloomington, IN: Indiana University dissertation. Bárkány, Szuszanna. 2002. A fresh look at quantity sensitivity in Spanish. Linguistics 40. 375– 394. Bassols de Climent, Mariano. 1962. Fonética latina. Madrid: Consejo Superior de Investigaciones Científicas. Beckman, Jill. 1998. Positional faithfulness. Amherst, MA: University of Massachussets dissertation. Beritognolo, Gustavo. 2008. Hiatus in Argentinian Spanish. Proceedings of console XV. 1–23. http://www.hum2.leidenuniv.nl/pdf/lucl/sole/console15/console15-beritognolo.pdf Bermúdez-Otero, Ricardo. 2006. Morphological structure and phonological domains in Spanish denominal derivation. In Fernando Martínez-Gil & Sonia Colina (eds.), Optimality-theoretic studies in Spanish phonology, 278–311. Amsterdam & Philadelphia: John Benjamins. Blevins, Juliette. 1995. The syllable in phonological theory. In John Goldsmith (ed.), The Handbook of phonology, 206–244. Oxford: Blackwell. Buckley, Eugene. 2014. Spanish secondary stress without gradient alignment. NELS 43. 37–50. Buckley, Eugene. (forthcoming). Foot alignment in Spanish secondary stress. To appear in Jeff Heinz, Rob Goedemans & Harry van der Hulst (eds.), Dimensions of stress. Cambridge: Cambridge University Press. Cabré, Teresa & Pilar Prieto. 2006. Exceptional hiatuses in Catalan and Spanish. In Fernando Martínez-Gil & Sonia Colina (eds.), Optimality-theoretic studies in Spanish phonology, 205–237. Amsterdam & Philadelphia: John Benjamins. Chitoran, Ioana & José Ignacio Hualde. 2007. From hiatus to diphthong: the evolution of vowel sequences in Romance. Phonology 24. 37–75. Dell, François & Mohamed Elmedlaoui. 1985. Syllabic consonants and syllabification in Imdlawn Tashlhiyt Berber. Journal of African Languages and Linguistics 7. 105–130.

53 I am delighted to take this opportunity to send Jim my very special greetings, three decades plus after our first encounter. Even longer since my discovery of his pioneering ground-breaking 1969 treatise and the subsequent many hours of endeavour to decipher generative phonology. Water has relentlessly kept going under the bridge since, but the memories are still fresh. The future is a crystal ball, and I cordially convey my very best wishes.

Gliding ghosts or ghostly glides, and does it matter which?

101

Den Os, Els & René Kager. 1986. Extrametricality and stress in Spanish and Italian. Lingua 69. 23–48. Díaz-Campos, Manuel. 2000. The phonetic manifestation of secondary stress in Spanish. In Héctor Campos, Elena Herburger, Alfonso Morales-Front & Thomas Walsh (eds.), Hispanic Linguistics at the Turn of the Millenium, 49–65. Somerville, MA: Cascadilla. DRAE: Diccionario de la Real Academia Española. On-line: http://dle.rae.es/ Dunlap, Elaine. 1991. Issues in the moraic structure of Spanish. Amherst, MA.: University of Massachussets dissertation. Face, Timothy 2004. Perceiving what isn’t there: Non-acoustic cues for perceiving Spanish stress. In Timothy Face (ed.), Laboratory approaches to Spanish phonology, 117–141. Berlin: Mouton de Gruyter. Foley, James 1965. Spanish morphology. Cambridge, MA: MIT dissertation. Gómez Torrego, Leonardo 1989. Manual del español correcto. Madrid: Arco Libros. Guerssel, Mohammed 1986. Glides in Berber and syllabicity. Linguistic Inquiry 17. 1–12. Grandgent, Charles H. 1963. Introducción al latín vulgar. Madrid: Consejo Superior de Investigaciones Científicas. Spanish translation by Francisco de Borja Moll of An introduction to Vulgar Latin, Boston, MA: D.C. Heath & Co. 1907. Halle, Morris, James W. Harrris & Jean R. Vergnaud. 1991. A Reexamination of the stress erasure convention and Spanish stress. Linguistic Inquiry 22(1). 141–159. Harris, James W. 1967. Spanish phonology. Cambridge, MA: MIT dissertation. Harris, James W. 1969. Spanish phonology. Cambridge, MA: The MIT Press Harris, James W. 1975. Stress assignment rules in Spanish. In William G. Milan, John J. Staczek, & Juan C. Zamora (eds.), 1974 Colloquium in Spanish and Portuguese linguistics, 56–83. Washington, DC: Georgetown University Press. Harris, James W. 1980. Non-concatenative morphology and Spanish plurals. Journal of Linguistic Research I. 14–31. Harris, James W. 1983. Syllable structure and stress in Spanish. Cambridge, MA: The MIT Press. Harris, James W. 1985. Spanish word markers. In F.W. Nuessel (ed.), Current Issues in Spanish phonology and morphology. 34–54. Bloomington, IN: IULC. Harris, James W. 1991a. With respect to metrical constituents in Spanish. In Héctor Campos & Fernando Martínez-Gil (eds), Current studies in Spanish linguistics, 447–473. Washington, DC: Georgetown University Press. Harris, James W. 1991b. The exponence of gender in Spanish. Linguistic Inquiry 22. 27–62. Harris, James W. 1992. The form classes of Spanish substantives. In Geert Booij & Jaap van Merle (eds.), Yearbook of Morphology 1991, 65–88. Dordrecht: Kluwer. Harris, James W. 1995. Projection and edge marking in the computation of stress in Spanish. In John Goldsmith (ed.), The handbook of phonological theory, 867–887. Oxford: Blackwell. Harris, James W. 1999. Nasal depalatalization no, morphological well-formedness sí. The structure of Spanish noun classes. MIT Working Papers in Linguistics 33. 47–82. Harris, James W. & Ellen Kaisse 1999. Palatal vowels, glides and obstruents in Argentinian Spanish. Phonology 19. 117–90. Hooper, Joan & Tracy Terrell. 1976. Stress assignment in Spanish: a natural generative approach. Glossa 10. 64–110. Hualde, José Ignacio. 1991. On Spanish syllabification. In Campos, Héctor & Fernando MartínezGil (eds.), Current Studies in Spanish Linguistics, 475–493. Washington, DC: Georgetown University Press.

102

Iggy Roca

Hualde, José Ignacio. 1997. Spanish /i/ and related sounds: an exercise in phonemic analysis. Studies in the Linguistic Sciences 27: 61–79. Hualde, José Ignacio. 1999. Patterns in the lexicon: hiatus with unstressed vowels in Spanish. In Javier Gutiérrez-Rexach & Fernando Martínez-Gil (eds.), Advances in Hispanic linguistics, papers from the 2nd Hispanic Linguistics Symposium (vol. 1), 182–197. Somerville, MA: Cascadilla. Hualde, José Ignacio. 2007. Stress removal and stress addition in Spanish. Journal of Portuguese Linguistics 5(2)/6(1). 59–89. Hualde, José Ignacio. 2009. Unstressed words in Spanish. Language Sciences 31. 199–212. Hualde, José Ignacio. 2010. Secondary stress and stress clash in Spanish. In Marta OrtegaLlebaria (ed.), Selected proceedings of the 4th Conference on Laboratory Approaches to Spanish phonology, 11–19. Somerville, MA: Cascadilla Proceedings Project. Hualde, José Ignacio. & Marina Nadeu. 2014. Rhetorical stress in Spanish. In Harry van der Hulst (ed.), Word Stress. Theoretical and typological issues, 228–252. Cambridge: Cambridge University Press. http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139600408 Hyde, Brett. 2002. A restrictive theory of metrical stress. Phonology 19. 313–359. Hyde, Brett. 2007. Issues in Banawá prosody: onset sensitivity, minimal words, and syllable Integrity. Linguistic Inquiry 38.2. 239–285. Hyde, Bruce & Bethany McCord. 2012. The inadequacy of a faithfulness-based approach to Spanish secondary stress. Ms., Washington University. Rutgers Optimality Archive, ROA 1154. http://roa.rutgers.edu/content/article/files/1213_hyde_1.pdf Hyman, Larry 1985. A theory of phonological weight. Dordrecht: Foris. Klein, Philip 1989. Spanish ‘gender’ vowels and lexical representation. Hispanic Linguistics 3. 147–163. Levi, Susanna. 2004. The representation of underlying glides: a cross-linguistic study. Seattle, WA: University of Washington dissertation. Levin, Juliette. 1985. A metrical theory of syllabicity. Cambridge, MA: MIT dissertation. Lipski, John. 1997. Spanish word stress: the interaction of moras and minimality. In Fernando Martínez-Gil & Alfonso Morales-Front (eds.), Issues in the phonology and morphology of the major Iberian languages, 559–593. Washington, DC: Georgetown University Press. Martínez-Paricio, Violeta. 2013. The intricate connection between diphthongs and stress in Spanish. Nordlyd (Tromsø University Working Papers on Language and Linguistics) 40(1). 166–95. http://septentrio.uit.no/index.php/nordlyd/article/view/2505 McCarthy, John. 2008. Doing Optimality Theory. Oxford: Blackwell. Menéndez Pidal, Ramón. 1962. Gramática histórica española. Madrid: Espasa-Calpe. Monteil, Pierre. 1970. Éléments de phonétique et de morphologie du Latin. Paris: Fernand Nathan. Morin, Regina. 1999. Spanish substantives: how many classes? In Javier Gutiérrez-Rexach & Fernando Martínez-Gil (eds.), Advances in Hispanic linguistics, papers from the 2nd Hispanic Linguistics Symposium (vol. 1), 214–30. Somerville, MA: Cascadilla. Navarro Tomás, Tomás. 1959. Manual de pronunciación española. Madrid: Publicaciones de la Revista de Filología Española. Nevins, Andrew & Ioana Chitoran. 2008. Phonological representations and the variable patterning of glides. Lingua 118 (12). 1979–97. Oltra-Massuet, Isabel & Karlos Arregi. 2005. Stress-by-structure in Spanish. Linguistic Inquiry 36. 43–84.

Gliding ghosts or ghostly glides, and does it matter which?

103

Otero, Carlos. 1986. A unified metrical account of Spanish stress. In Michael Brame, Heles Contreras & Frederick J. Newmeyer (eds.), A Festchrift for Sol Saporta, 299–332. Seattle, WA: Noit Amnofer Publishing. Pater, Joe. 2000. Non-uniformity in English secondary stress: The role of ranked and lexically specific constraints. Phonology 17. 237–274. Pater, Joe. 2007. The locus of exceptionality: Morpheme-specific phonology as constraint indexation. In Leah Bateman, Michael O’Keefe, Ehren Reilly & Adam Werle (eds.). Papers in Optimality Theory III. University of Massachusetts Occasional Papers in Linguistics 32. 259–296. Amherst, MA: GLSA. Pater, Joe. 2010. Morpheme-specific phonology: constraint indexation and inconsistency resolution. In Steve Parker (ed.), Phonological argumentation: Essays on evidence and motivation. 123–154. London: Equinox. Pensado, Carmen. 1985. On the interpretation of the non-existent: non-occurring syllable types in Spanish phonology. Folia Linguisitica 19. 313–320. Prieto, Pilar & Jan van Santen. 1996. Secondary stress in Spanish: Some experimental evidence. In Claudia Parodi, Carlos Quicoli, Mario Saltarelli & María Luisa Zubizarreta (eds), Aspects of Romance linguistics, 337–356. Washington, DC: Georgetown University Press. Prince, Alen & Paul Smolensky. 1993. Optimality Theory: constraint interaction in generative grammar. ROA-537. Subsequent publication – Oxford: Wiley & Blackwell. 2004. Quilis, Antonio. 1993. Tratado de fonología y fonética españolas. Madrid: Gredos. Roca, Iggy. 1986. Secondary stress and metrical rhythm. Phonology Yearbook 3. 341–370. Roca, Iggy. 1988. Theoretical implications of Spanish stress. Linguistic Inquiry 19. 393–423. Roca, Iggy. 1989. The organisation of grammatical gender. Transactions of the Philological Society 87(1). 1–32. Roca, Iggy. 1991. Stress and syllables in Spanish. In Héctor Campos & Fernando Martínez-Gil (eds), Current studies in Spanish linguistics. 599–635. Washington, DC: Georgetown University Press. Roca, Iggy. 1997a. There are no glides, at least in Spanish. Probus 9. 233–65. Roca, Iggy. 1997b. On the role of accent in stress systems: Spanish evidence. In Fernando Martínez-Gil & Alfonso Morales-Front (eds), Issues on the morphology and phonology of the major Iberian languages, 619–664. Washington, DC: Georgetown University Press. Roca, Iggy. 1999. Stress in the Romance languages. In Harry van der Hulst (ed.), Word prosodic systems in the languages of Europe, 559–811. Berlin: Mouton de Gruyter. Roca, Iggy. 2000. On the meaning of gender. Hispanic Research Journal 1(2). 113–128. Roca, Iggy. 2005a. La gramática y la biología en el género del español. Revista Española de Lingüística 35(1). 17–44 & 35(2). 397–432. Roca, Iggy. 2005b. Saturation of parameter settings in Spanish stress. Phonology 22. 345–394. Roca, Iggy. 2005c. Strata, yes. Structure preservation, no. Evidence from Spanish. In Twan Geerts, Ivo van Ginneken & Haike Jacobs (eds), Romance language and linguistic theory 2003, 197–218. Amsterdam & Philadelphia: John Benjamins. Roca, Iggy. 2006. The Spanish stress window. In Fernando Martínez-Gil & Sonia Colina (eds), Optimality-theoretic studies in Spanish phonology, 239–277. Amsterdam & Philadelphia, John Benjamins. Roca, Iggy. 2013. El género del castellano: Entelequias y realidades. Español Actual 99. 51–95. Roca, Iggy. 2015. Spanish stress: An update. Ms. Roca, Iggy & Elena Felíu. 2003. Morphology in truncation: The role of the Spanish desinence. In Geert Booij & Jaap van Marle (eds), Yearbook of Morphology 2002, 187–243. Dordrecht: Kluwer.

104

Iggy Roca

Rosenthall, Samuel. 1994. Vowel/glide alternation in a theory of constraint interaction. Amherst, MAː University of Massachusetts dissertation. Saltarelli, Mario. 1997. Stress in Spanish and Italian: Where morphology meets prosody. In Fernando Martínez-Gil & Alfonso Morales-Front (eds.), Issues on the morphology and phonology of the major Iberian languages, 665–694. Washington, DC: Georgetown University Press. Saporta, Sol. 1959. Morpheme alternants in Spanish. In Henry Kahane & Angelina Pietrangeli (eds.), Structural studies in Spanish themes, 15–162. Salamanca: Acta Salmanticiensia. Saporta, Sol. 1962. On the expression of gender in Spanish. Romance Philology XV(3). 279– 284. Scharf, Gabriele, Ingo Heinrich, Iggy Roca & Grerzgorz Dogil. 1995a. Articulatory correlates of secondary stress in Polish and Spanish. Stuttgart, Universität Stuttgart: AIMS (Arbeitspapiere des Instituts für Maschine Sprachverarbeitung) 2(2). 243–264. Scharf, Gabriele, Ingo Heinrich, Iggy Roca & Grerzgorz Dogil. 1995b. Articulatory correlates of secondary stress in Polish and Spanish. In K. Elenius & P. Branderud (eds), Proceedings of the XIII International Congress of Phonetic Sciences’, vol. 4. 634–637. Stockholm. Shelton, Michael, Chip Gerfen & Nicolás Gutiérrez Palma. 2012. The interaction of subsyllabic encoding and stress assignment: A new examination of an old problem in Spanish. Language and Cognitive Processes 27 (10). 1459–1478. http://www.ncbi.nlm.nih.gov/ pmc/articles/PMC3524975/ Simonet, Miquel. 2005. Prosody and syllabification intuitions of [CiV] sequences in Catalan and Spanish. In Sónia Frota, Maria-João Freitas & Marina Vigário (eds.), Prosodies, 247–267. Berlin: Mouton de Gruyter. Smith, Jennifer. 2002. Phonological augmentation in prominent positions. Amherst, MA: University of Massachusetts dissertation. SPE: Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Steriade, Donca. 1984. Glides and vowels in Romanian. In Proceedings of the Tenth Annual Meeting of the Berkeley Linguistics Society. 47–64. Väänänen, Veikko. 1963. Introduction au Latin Vulgaire. Paris: Librairie C. Klincksieck.

II The Syllable – Onset, Coda, and Cross Linguistic Aspects

Sonia Colina

On onset clusters in Spanish: voiced obstruent underspecification and /f/ Abstract: Onset clusters in Spanish consist of an obstruent or /f/ plus a liquid, with the exception of /tl/ and /dl/ which are ill formed in some dialects (*tl) and or in all (*dl). While this descriptive generalization is uncontroversial, the phonological account of onset clusters has been a topic of debate amongst phonologists for decades now. Most agree that the main factor driving the onset generalization and the well-formedness of the cluster is sonority; yet, it is not entirely clear why /f/ is the only fricative grouped with the obstruents. This paper contributes to our understanding of onset clusters by: (i) presenting additional evidence regarding the sonority of /f/, in support of the proposal in Martínez-Gil (2001); (ii) introducing refinements to the onset cluster generalization/condition (OC) (Martínez-Gil 2001; Colina 2009), most importantly with regard to point of application. The current proposal rests on input and output underspecification of voiced obstruents.

0 Introduction Onset clusters in Spanish consist of an obstruent or /f/ plus a liquid, e.g., blanco ‘white’, egresar ‘to graduate’, sufrir ‘to suffer’, with the exception of /tl/ and /dl/ which are ill formed in some dialects (*tl) and/or in all (*dl). While this descriptive generalization is uncontroversial, the phonological account of onset clusters has been a topic of debate amongst Spanish phonologists for several decades now (Harris 1983, Harris 1989a, 1989b, Martínez-Gil 1997, Martínez-Gil 2001, Colina 2009a). Most analyses agree that the main factor driving the onset generalization and the well-formedness of the cluster is sonority. Yet, it is not entirely clear why /f/ is the only fricative grouped with the obstruents. Onset clusters cannot be fully understood without an account of the behavior of /f/ in this syllabic position. This paper contributes to our phonological understanding of onset clusters in Spanish by: (i) presenting additional independent evidence regarding the sonority of /f/ and of voiced stops in Spanish, supporting the proposal in Martínez-Gil (2001); (ii) highlighting the implications of this account for the rest of the phonology, specifically with regard to phonemic inventory and

Sonia Colina, The University of Arizona

108

Sonia Colina

underspecification and voiced obstruent alternations (iii) introducing refinements to the onset cluster generalization/condition (OC) (Martínez-Gil 2001; Colina 2009a), most importantly with regard to the point at which the generalization applies. The proposal put forth here – which argues that voiced stops are underspecified in the input and output of the phonology – also contributes to the phonological literature on the topic of output (perseverant) underspecification. The paper is organized as follows: the data is presented in section 1; section 2 focuses on existing attempts to account for onset cluster generalizations, including a review of the literature in Spanish and the problems to be addressed in this paper; in section 3 the account that voiced obstruents are underspecified is developed, including the data and literature on voiced obstruent alternation (3.1), the evidence, and the formal analysis (3.2). Finally, section 4 introduces the matter of the point of application of the complex cluster generalization, and how the current proposal addresses it, including some relevant Chilean data. Section 5 presents an overview of the paper and some conclusions.

1 Onset clusters in Spanish: the data Onset clusters in Spanish consist of a voiceless or voiced stop /p, t, k, b, d, ɡ/ or /f/, as the first member of the cluster, and a liquid /l, ɾ/ as the second (1). The generalization applies in word-initial position as well as word-medially. (1)

blanco emblema broma abrir dragón adrenalina glotón inglés granada egresar fluido afligir

‘white’ ‘emblem’ ‘joke’ ‘to open’ ‘dragon’ ‘adrenaline’ ‘glutton’ ‘English’ ‘granade’ ‘to graduate’ ‘fluid’ ‘to afflict’

platicar aplaudir presión apretar tracción atracar clavo aclarar cresta increpar frente sufrir

‘to talk’ ‘to applaud’ ‘pressure’ ‘to squeeze’ ‘traction’ ‘to hold up’ ‘nail’ ‘clarify’ ‘crest’ ‘to reprimand’ ‘front’ ‘to suffer’

Exceptions to the generalization include the coronals /t, d/. /dl/ is always ill formed and /tl/ is acceptable only in some varieties, mostly under the influence

On onset clusters in Spanish: voiced obstruent underspecification and /f/

109

of languages that have this sequence, like Nahuatl (2b).1 In those Spanish varieties in which /tl/ is not a well-formed cluster, the obstruent is syllabified in the coda and the liquid in the following onset, as seen in (2a). (2)

Syllabification of the cluster -tla. Heterosyllabic b. Onset cluster at.las a.tlas ‘atlas’ at.lético a.tlético ‘athletic’ At.lántico A.tlántico ‘Atlantic’

Harris (1983:33) attributes these facts to co-occurrence restrictions on two adjacent coronals, and on voice in the case of /dl/. Accordingly, he proposes the filter in (3a), which states that a consonant cluster consisting of two noncontinuant, coronal consonants that also agree in voice is not permissible. Varieties in which */tl/ is ill formed have a more restrictive filter that rules out a sequence of two [-continuant] and [+coronal] segments in the onset, irrespective of their voice specification (3b). (3)

a.

b.

Restrictions on /tl/ and /dl/ in onset clusters are not specific to Spanish and are frequent in many languages (Parker 2012: 151). More recently, phonetic evidence has shown that these clusters present issues of perceptibility, in particular it is difficult to discriminate auditorily between /tl/ /dl/, on the one hand, and /kl/ /ɡl/, on the other (Flemming 2002, 2007). The explanation is that the articulatory gestures of the liquid can obscure stop formant transitions if they overlap with the stop (formant transitions, along with stop burst, are crucial cues for stop place). Additional support for this view comes from Blevins and Grawunder (2009) who point to historical changes in Germanic and other languages involving *kl > tl and vice versa that are motivated by the difficulty involved in the perception of the contrast.

1 /tl/ is a lateral alveolar affricate in Nahuatl. Examples of Nahuatl borrowings in Mexican Spanish are Tlacuache ‘opposum’, Tlaxcala ‘place name’.

110

Sonia Colina

2 Accounting for Spanish onset clusters In order to offer an account of Spanish onset clusters and to formulate a generalization that goes beyond description (cf. section 1), the concept of sonority must be introduced. Existing analyses of Spanish onset clusters attribute the main factor driving the well-formedness of the clusters to sonority.

2.1 Onset clusters and sonority Sonority has no straightforward phonetic definition. It has been traditionally associated with perceptual salience in terms of acoustics and with degree of stricture in articulatory terms. However, recent phonetic studies indicate that intensity (i.e., loudness) is probably the most robust correlate of sonority in measurable phonetic terms (Parker 2008, 2012; Henke, Kaisse and Wright 2012), although it is unlikely to be the only one.2 In phonological terms, sonority can be considered “a unique type of relative, n-ary (non-binary) feature-like phonological element that potentially categorizes all speech sounds into a hierarchical scale (Parker 2011: 1160).” Sonority is often defined in terms of its effect on syllable phonotactics. Syllables exhibit a contour according to which the syllable rises in sonority towards the nucleus (the most sonorous point) and decreases towards the coda (Sonority Sequencing Principle, SSP, Clements 1990). In addition, the rise is maximal in the onset and minimal in the coda, according to the Sonority Cycle (Clements 1990). Since sonority is a universal but relative and gradient concept, segments are ranked according to sonority in a universal scale, such as (4). (4) Obstruents < Nasals < Liquids < Glides < Vowels The scale is universal with regard to the relative degree of sonority of the classes, so that an obstruent will always be less sonorous than a nasal and a nasal in turn will always be less sonorous that a vowel. Languages differ, however, in the number of sonority classes needed.

2 An argument in favor of the position that intensity may not be the only correlate of sonority lies in the fact that, although [f] and [θ] are similar in terms of center of gravity (distribution of intensity), [f] has higher intensity peaks than [θ]. This would mean that [f] is more sonorous than [θ], a statement that is not supported by phonotactics data. I thank an anonymous reviewer for pointing out the intensity facts.

On onset clusters in Spanish: voiced obstruent underspecification and /f/

111

While most (if not all) of the literature on Spanish onset clusters is sonoritybased, it must be acknowledged that the notion of sonority and its role in syllable phonotactics is a somewhat controversial issue in phonological theory today (cf. for instance, the topic of a recent book, The sonority controversy, Parker (2012)). Not all researchers agree that syllable phonotactics are entirely ascribable to sonority. Some argue that a perception-based account captures phonotactic generalizations, including those pertinent to onset clusters, more adequately than sonority by offering wider empirical coverage (Wright, 2004; Henke, Kaisse and Wright 2012; among others); thus they propose replacing sonority scales with scales of auditory cue robustness, according to which the cross-linguistic phonotactic patterns observed in onset clusters respond to the maximization of perceptually recoverable strings. Take, for instance, /s/ + stop clusters. Despite violating sonority requirements (because of flat or decreasing sonority, depending on the scale in use), /s/ + stop clusters are a frequent sequence in the world’s languages; they are in fact the most attested pattern in some surveys (Morelli 1999, 2003). Proponents of a perception-based account of onset phonotactics contend that a sonority-based account cannot explain this fact; in contrast, a perception-based one highlights the strong internal perceptual cues of the sibilant as the reason for the typological frequency of this cluster. It is likely that sonority alone cannot account for onset phonotactics for all languages, in particular for those that allow highly complex onset combinations. Yet, Spanish, which exhibits a rather unmarked type of onset cluster (O(bstruent)L(iquid), cf. the typology in Parker 2012), is a good example of a language in which onset clusters are mainly, if not entirely, driven by sonority, as demonstrated by the existing literature. Returning to the sonority scale, for Spanish, Harris (1989a, b) (also Hualde 2005: 72) proposes the one in (4) in which obstruents are less sonorous than nasals and nasals in turn are less sonorous than liquids. (The scale is shown again in (5), listing the Spanish phonemes in each category). Harris formulates the onset generalization as a condition that requires that the onset contain two consonants that are not adjacent in the sonority scale. (5)

Harris’s Sonority scale for Spanish (1983) Obstruents

Nasals

Liquids

Glides

Vowels

p, b, t, d, k, ɡ, f, θ, s, h, x, tʃ, dʒ

m, n, ɲ

l, r, ɾ

j, w

a, e, i, o, u

112

Sonia Colina

Thus, it is not possible to have the sequences obstruent + nasal or nasal + liquid; in sum, obstruent + liquid is in principle an acceptable cluster. Since the only non-stop obstruent allowed as the first member of the cluster is /f/, it is necessary to eliminate /θ, s, h, x, tʃ, dʒ/. Harris resorts to point of articulation restrictions to accomplish this, positing a filter that bans two adjacent [+alveolar] consonants. Although it seems fairly intuitive to group stops, fricatives and affricates in one sonority class and then try to rule out the fricatives, except for /f/, and the affricates, as Harris does, Martínez-Gil (2001) shows that this approach is fraught with difficulties. Among them he lists the use of feature [alveolar], unnecessary for Spanish and not commonly used for other languages, and the inability to rule out some impermissible clusters, such as affricate + liquid, interdental + liquid and /x/ + liquid (Martínez-Gil 2001: 213–215). Martínez-Gil (2001) argues that a more elegant and insightful analysis can be obtained by grouping only the stops and /f/ in the class of the lowest sonority. He proposes the scale in (6) (Martínez-Gil 1996, 1997, 2001; Colina 2009a). (6) Sonority scale for Spanish (Martínez-Gil 2001: 217) Obstruents

Sonorants

p, b, t, d, k, ɡ, f

θ, s, h, x, tʃ, dʒ

m, n, ɲ, ʎ, r

l, ɾ

1

2

3

4

He also argues that sonority classes are driven by manner of articulation, rather than point of articulation. According to this, the relevant onset cluster generalization is that an onset cluster in Spanish consists of two consonants that differ maximally in sonority rank (7): the first member is drawn from the group of the least sonorous consonants permissible in the onset (voiced and voiceless stops) and the second member from the most sonorous ones (liquids), i.e., from groups 1 and 4 respectively in (6). (7)

Complex Onset Condition (Martínez-Gil 2001: 219) A complex onset in Spanish is limited to two consonants that differ maximally in sonority rank.

Given the scale in (6) and the Complex Onset Condition in (7), it is clear that Spanish does not allow /s/ + stop clusters, as some languages do, because of the sonority reversal that they entail (i.e., they do not satisfy the Complex Onset Condition). Placing Spanish in a broader cross-linguistic context, one may also

On onset clusters in Spanish: voiced obstruent underspecification and /f/

113

wonder why a glide is not permissible as the second consonant in the cluster, as CG is also an unmarked cluster type. Spanish normally parses prevocalic glides in the nucleus, indicating a preference for a complex nucleus over a complex CG onset (Colina 2009a: 21; Hualde 2014: 199–200, Kaisse this volume). This analysis is further supported by the inability of glides to serve as singleton onsets: when a prevocalic glide is not preceded by a consonant that can serve as an onset, it undergoes fortition, surfacing as an obstruent, hielo /ielo/ [ ʝélo] ‘ice’, com-iendo [ko.mjén.do] ‘eating’, cre-iendo [kɾe.ʝén.do] ‘believing’; sometimes it also exhibits obstruent insertion, hueso [weso] [ɡweso] ‘bone’ (cf. however, Martínez-Gil this volume, for a proposal in which prevocalic, postconsonantal glides are in the onset). This set of facts poses a question regarding the behavior of /f/, as it is not clear why it should belong to the group of the least sonorous consonants or why it should be the only fricative possible. Generally, languages with fricatives as C1 accept more fricatives than just /f/. In accordance with the general unmarked nature of its onset clusters, Spanish appears to belong to the group of languages that do not allow fricatives as C1 in a cluster, in opposition to languages that have restrictions on some fricatives but allow others, such as English. Thus the question remains as to why is /f/ allowed in Spanish onset clusters.

2.2 /f/ and onset clusters As mentioned above, at first, it is not apparent why /f/ is the only Spanish fricative that clusters with the stops as a permissible first member of the onset, since they do not belong to the same natural class. They do not seem to belong to the same general sonority class either. Martínez-Gil (2001) argues that /f/ is grouped with the voiceless stops and voiced stops because, like them, it does not have [+continuant] in its representation. In other words, what these sounds have in common is the absence of the feature [+continuant] in their phonological representation, because they contain the opposite specification (i.e., [-continuant]) (voiceless stops) or because they have no specification at all (voiced stops and /f/). In his analysis, [+continuant] is the feature that contributes to sonority among consonants. Therefore, segments specified as [+continuant] will be more sonorous than those without this feature. This is in accordance with the view that phonologically, sonority is related to degree of stricture (i.e., [continuant]) (Clements 1990) and intensity (Henke, Kaisse and Wright 2012). It is uncontroversial that voiceless stops are [-continuant]. In Martínez-Gil’s account, /f/ is not specified for continuancy because [+continuant] is redundant

114

Sonia Colina

cross-linguistically for labiodentals, as “no language has labiodental stops. . . except as allophones” (Ladefoged 1992: 158). Voiced stops are said to be underspecified for continuancy, in agreement with previous analyses and for similar reasons (Lozano 1979; Mascaró 1984; Hualde 1989; and others). This paper presents additional evidence for Martínez-Gil’s proposal that the stops and /f/ are well-formed as the first member of an onset cluster in Spanish because they are unspecified for continuancy in the phonology; in other words, they are the least sonorous group of consonants and they belong to the same sonority class, because they do not possess the feature [+continuant] (i.e., they lack [continuant] entirely or are [-continuant]). In addition, while the voiced approximants and /f/ are [+continuant] in the phonetic output, they do not contain this feature in the phonology; this makes them less sonorous than other approximants or fricatives that are phonologically [+continuant]. The current proposal also shares Martínez-Gil’s (2001) view that sonority is a basic, indivisible unit in phonology (rather than a derived concept). Whatever features /f/, the voiceless stops and voiced obstruents share, they cannot result from a combination of features because the absence of continuant specification (i.e., [-continuant], no specification for [continuant]) is not the same as [-continuant] and cannot be obtained from the combination of binary features. Thus, although sonority classes and sonority scales rely on features and are based on feature-like elements, they do not coincide exactly with feature classes. Sonority is not a combination of features, but a primitive of phonology; that is, it is a construct that does not result from the addition of other constructs and that cannot be divided up into independent units nor reduced to a list of features. The analysis of voiced stops in Spanish proposed in section 3 offers independent evidence for the proposal that the voiced obstruents and /f/ lack a [+continuant] specification and expands on the account of onset clusters by showing its implications for the rest of the phonology (e.g., the phonemic inventory of Spanish) (in section 3) and for other dialectal varieties of Spanish, such as Chilean Spanish (in section 4). Section 4 also deals with the point of application issue, a very relevant matter for non-derivational models of phonology like OT.

3 Voiced stops and onset clusters Based on a recent account of the voiced obstruent alternation (Colina 2013), I present additional evidence for the proposal that voiceless stops, voiced obstruents

On onset clusters in Spanish: voiced obstruent underspecification and /f/

115

and /f/ lack a [+continuant] feature and that the absence of this feature makes them better onset segments than those obstruents specified as [+continuant]; in particular, I propose that the underspecification of voiced obstruents [B, D, G] persists through the output of the phonology (i.e., output underspecification, Hale & Kissock 2007, originally from Keating 1988). I show that the analysis of onset clusters in Spanish is crucially dependent on that of the voiced obstruent alternation and this account needs to be considered in any analysis of these clusters in Spanish. In addition to explaining why /f/ clusters with the stops, the analysis of voiced obstruents presented here helps us understand why the Complex Onset Condition appears to be rampantly violated by the approximant allophones of the voiced obstruents, addressing questions regarding the point of application of the complex onset generalization. It is shown that a comprehensive account of onset clusters must also explain the interactions of these segments with other aspects of the phonology of a language, such as phonemic distinctions, featural specifications and relevant phonological phenomena.

3.1 Existing analyses of voiced obstruents Spanish voiced plosives have traditionally been described as having a continuant and a non-continuant realization. In the general varieties that are the object of this study,3 the relevant generalization is that [-continuant] allophones appear after homorganic nasals and laterals, and in word-initial pre-pausal position (8), and [+continuant] allophones elsewhere (9)4. The process is affected by much stylistic and dialectal variation (cf. Soler & Romero 1999; and Kirchner 2004 for variation influenced by speech rate; for dialectal variation with experimental evidence see Carrasco, Hualde & Simonet 2012). (8)

bomba donde tango toldo

[bómba] [dón̪de] [taŋgo] [tól ̪do]

‘bomb’ ‘where’ ‘tango’ ‘awning’

3 By ‘general’ varieties I refer to those commonly presented in textbook descriptions of Spanish pronunciation. Some varieties of Spanish, which are not the focus of this study, have stops in more contexts than those presented here. 4 These facts are also supported by recent experimental evidence (Hualde, Simonet, Shosted & Nadeu 2010).

116 (9)

Sonia Colina

haba nada hago árbol algo calvo

[áβa] [náða] [áɣo] [áɾβol] [álɣo] [kálβo]

‘bean’ ‘nothing’ ‘I do’ ‘tree’ ‘something’ ‘bald’

Existing accounts propose a variety of solutions with regard to the feature that changes, the form of the underlying representation, and the directionality of the change. Harris (1984) argues that the underlying representation contains [-continuant] because cross-linguistically, this is the unmarked value for the voiced obstruents. Subsequently, in the postlexical level a Continuancy Spreading Rule spreads [+continuant] from a preceding [+continuant] segment to the voiced obstruent. In other words, the rule changes [-continuant] to [+continuant] when preceded by a [+continuant] segment. In Harris’ analysis, a condition on autosegmental representations, which requires that when two segments share association lines (e.g., the point of articulation in homorganic segments) both segments must comply with the structural description of the rule, serves to explain the behavior of /ld/, which does not become [+continuant]. /l/ does not meet the conditions of the rule because it is not a voiced obstruent and therefore /d/, the segment with which it shares PA, cannot be the target for [+continuant] spreading. In accordance with this, /lb/ and /lɡ/ are realized as [+continuant] because they do not share nodes (they are not homorganic). Harris argues that /l/ is [+continuant]. Mascaró (1984) proposes the spreading of the feature [+/-continuant] (i.e., α continuant) to a voiced obstruent that follows it and that is underspecified for continuancy. A separate rule inserts [-continuant] after a pause (Examples in 10 from Hualde 1989). (10)

In another influential proposal, Hualde (1989) also takes as his point of departure an unspecified voiced obstruent. He proposes a rule that spreads the feature [-continuant] in a homorganic group to the right onto a voiced obstruent. Subsequently, an additional rule inserts [-continuant] in a feature matrix that contains [+voice] and [-sonorant] after a pause; finally, a default feature insertion rule

On onset clusters in Spanish: voiced obstruent underspecification and /f/

117

adds [+continuant] to all voiced obstruents with no specification after the application of the other rules. Within an Optimality-Theoretic framework, Baković (1997) argues for fortition of underlying approximants. In this approach ordered rules are replaced with hierarchically ranked universal constraints. The driving force behind fortition is a constraint, STRONG ONSET, that requires a [-continuant] or oral closure in strong positions (at the beginning of a breath group, and in a homorganic cluster). In summary, in addition to the obvious differences in theoretical framework (serial or parallel), existing analyses of voiced obstruent alternation in Spanish differ with regard to directionality, the feature that spreads (for autosegmental accounts), and the form of the underlying representation: (i) directionality of the process: fortition (Hammond 1976; Baković 1997; Lozano 1979) or lenition (Harris 1969, 1984). Fortition analyses argue that a [+continuant] allophone becomes [-continuant] in strong, prominent positions, such as after a pause or after a homorganic nasal or lateral; in lenition analyses [+continuant] spreads from the preceding segment onto a [-continuant] one, thus resulting in a lenited allophone. (ii) the feature that spreads: [αcontinuant] (Mascaró 1984); [-continuant] in homorganic clusters (Hualde 1989); [+continuant] (Harris 1984). (iii) the nature of the underlying representation: [-continuant] (Harris 1969)5, [+continuant] (Hammond, 1976, Baković 1997); underspecified (Lozano 1979; Mascaró 1984; Hualde 1989; Kirchner 19986). Proponents of the underspecification account, the most numerous and most recent accounts, justify an underspecified input because of the lack of solid evidence for any of the specified forms.

3.2 Voiced stops are underspecified for [continuant] As mentioned above, I argue that Spanish voiced obstruents are underspecified for continuancy and that this underspecification persists through the output of the phonology (i.e., output underspecification, Hale and Kissock 2007, originally 5 As mentioned by an anonymous reviewer, Piñeros (2003) also proposes [-continuant] in the underlying representation. However, his proposal is for Palenquero (a creole of Spanish and African descent spoken in Colombia), rather than for the standard phonologies that are the focus of the current analysis. Furthermore, the Palenquero data are much more complex (including gemination, prenasalization, etc.), deserving separate treatment. 6 Kirchner (1998) in fact proposes an indeterminate input (rather than truly underspecified), as the choice of underlying representations turns out to be much less crucial to his analysis.

118

Sonia Colina

from Keating 1988). Due to the absence of [+continuant] in their phonological representation, voiced stops are acceptable as the first member of the cluster in onset clusters. As a result, the onset condition is operational throughout the phonology and no constraint violations are incurred for the dialects with voiced approximants in the onset. In what follows, I first present the evidence in favor of the output underspecification of voiced obstruents and then introduce a formal analysis within an Optimality-Theoretic framework. The formal analysis is preceded by some background discussion on underspecification in Optimality Theory (OT).

3.2.1 Evidence for underspecification: voiceless fricatives and variation The first piece of evidence has to do with the behavior of onset fricatives in homorganic clusters. Unlike voiced obstruents, voiceless fricatives in homorganic clusters do not become [-continuant]. Instead they retain their [+cont] specification, rather than agreeing in continuancy with the preceding nasals, as voiced obstruents do. In other words, there is no alternation in [continuant] for voiceless fricatives in the onset, as seen in (11a). (11)

Voiceless fricatives show no alternation in [continuant] a. voiceless un gitano [uŋ.xi.tá.no] ‘a gypsy-masc.’ una gitana [u.na.xi.tá.na] ‘a gypsy-fem.’ un salón [un.sa.lón] ‘a big room’ una sala [u.na.sá.la] ‘a room’ b.

voiced un gato una gata un dicho una dicha un bobo una boba

[uŋ.ɡá.to] [u.na.ɣá.ta] [un̪.dí.tʃo] [u.na.ðí.tʃa] [um.bó.βo] [u.na.βó.βa]

‘a cat-masc.’ ‘a cat-fem.’ ‘a saying’ ‘a joy’ ‘a silly one-masc.’ ‘a silly one-fem.

I argue that voiceless fricatives (with the exception of /f/) do not become stops because they are underlyingly specified as [+continuant] and this input specification must be retained. In contrast, voiced obstruents are underspecified, so there is no feature to preserve. A consequence of this is that obstruents in Spanish show a three-way contrast in continuancy: [+continuant] (voiceless

On onset clusters in Spanish: voiced obstruent underspecification and /f/

119

fricatives, except for /f/), [-continuant] (voiceless stops) and underspecified [continuant] (voiced obstruents and /f/). In an Optimality-Theoretic analysis the correct output is the result of the high ranking of a constraint that requires preservation of the input specification for [continuant] over one that requires assimilation in this feature to the preceding segment, as seen in (12). (12)

Constraints and constraint ranking for N + voiceless fricative ID -[cont]: the output matches the input with respect to [cont] (McCarthy & Prince 1995) ID -[voi]: the output matches the input with respect to [voi] (McCarthy & Prince 1995) NC[cont]: a nasal/lateral agrees in [cont] with the following homorganic obstruent. (cf. feature geometry in Padgett 1994) (Patter 1999, 2001; Martínez-Gil 2004) ID -[cont] >> NC[cont], ID -[voi]

(13) Post-nasal voiceless fricatives

As seen in (13), the voiceless fricative /s/ does not become a stop (13c–13d) because that would entail a change in its continuancy specification and therefore a violation of the top-ranked constraint, ID -[cont]. (13c) and (13d) are therefore ruled out. (13d) is worse than (13c) because it also changes the voice specification of the input, violating ID -[voi]. The two candidates that preserve [+continuant] and do not incur violations of ID -[cont], (13a) and (13b), violate NC[cont], as they are not [-continuant] like the preceding nasal. However, since they are better than (13c) and (13d), NC[cont] must be lower than ID -[cont]. (13a) is the winner because, unlike (13b), it does not violate ID -[voi]. / f/ behaves differently from the rest of the voiceless fricatives. Since, unlike the other fricatives, /f/ is underspecified for continuancy, the analysis incorrectly

120

Sonia Colina

predicts a labiodental voiceless stop in the output ([p̪]), due to the high of ranking of NC[cont], as seen in (14). (14) Post-nasal voiceless fricatives

Given that (14c) is not the correct output, a higher ranked constraint must be responsible for its elimination. Labiodental stops are extremely marked, due to the difficulties involved in forming a complete occlusion between the upper front teeth and the lower lip. Therefore, it is reasonable to propose that an undominated constraint against labiodental stops rules out candidate (14c) in favor of (14a).7 I will return to NC[cont] and the requirement that nasals agree in continuancy with a following obstruent below. The second important piece of evidence in favor of underspecification rests on phonetic variation. Phonetic studies report a great degree of variation in voiced obstruents with regard to aperture, ranging from an open approximant to a fricative (Cole, Iskarous, & Hualde 1999; Ortega Llebaria 2004; Colantoni and Marinescu 2010; Eddington 2011; Simonet at al. 2012); furthermore, some studies have also shown that in some dialects, like Madrid Spanish, the distinction between the continuant and the non-continuant allophones is gradient rather than bimodal (Carrasco, Hualde & Simonet 2012). Gradient variation can be easily explained if voiced obstruents are underspecified in the input and also (as explained in the next section) at the output of the phonology (i.e, output underspecification), as this means that they would be realized variably according to the surrounding segments, in the phonetic component. In other words, this is a targetless output that realizes its constriction variably and in a gradient manner depending on the adjacent segments. Due to its underspecification for continuancy, /f/ should also exhibit gradient variation; however, contrary to expectations, Spanish voiceless labiodental obstruents always surface as fricatives. As above, this can also be explained by phonetically motivated articu-

7 The constraint against labiodental stops would dominate the constraint against featural insertion, as the winning [+continuant] [f] must acquire output specification through insertion (a violation of the relevant DEP constraint).

On onset clusters in Spanish: voiced obstruent underspecification and /f/

121

latory constraints, as labiodental approximants are rare, in particular those without voicing. In sum, phonetic variation and the gradient nature of the phenomenon serve as evidence in favor of a specific type of underspecification, output underspecification. As will be shown with regard to the formal analysis in 3.2.2, output underspecification is an additional piece of evidence in support of input underspecification.

3.2.2 OT analysis of voiced obstruents in Spanish As mentioned above, this proposal argues that voiced obstruents in Spanish are underspecified for continuancy (i.e., input underspecification) and that underspecification persists into the output of the phonological component. In other words, the analysis proposes both input and output (i.e., phonetic) underspecification. Consequently, before introducing the analysis, a brief discussion of these concepts within OT becomes pertinent. Underspecified inputs have been objected to by some scholars (e.g., for Galician, Martínez-Gil 2004) on the grounds that underspecification places a restriction on the form of the underlying representation. According to the Richness of the Base (ROTB), a basic principle of OT, no restrictions are possible regarding the form of the input. However, an underspecified form /B, D, G/ places no more restrictions on the input than a fully specified one, e.g., /b, d, ɡ/, as it is just one of many possible inputs. In fact, since ROTB requires that no restrictions be placed on the form of the input, an underspecified input cannot be excluded from consideration. Doing so would be contrary to the ROTB. Additionally, whether the actual input is the underspecified one /B, D, G/, or another one, cannot be determined a priori, it can only be determined through the constraints and constraint ranking that select the form which is most harmonious with the output. In other words, whether the input is fully specified must be decided on the basis of the analysis and the constraints. As will be argued in this section, for Spanish /b, d, ɡ/ the constraints and the constraint ranking provide evidence for an underspecified input. Along similar lines, Inkelas shows that OT allows for what has been called principled underspecification: underlying forms, including underspecified forms, are determined by the learner through the principle of Lexicon Optimization (LO) and not by any constraints holding directly on underlying forms (Inkelas 1995: 289). She offers the alternation-sensitive definition of LO in (15):

122 (15)

Sonia Colina

Lexicon Optimization (Inkelas 1995: 289) Given a set S = {S1, S2, S3} of surface phonetic forms for a morpheme M, suppose that there is a set of inputs I = {I1, I2, I3}, each of whose members has a set of surface representations equivalent to S. There is some Ii a I such that the mapping between Ii and the members of S is the most harmonic, i.e., incurring the fewest marks in the grammar for the highest ranked constraints. The learner should choose that Ii as the underlying representation for M.

Using LO and thus placing the learner in a central position in her proposal, Inkelas demonstrates that the grammar (i.e., the constraints and the constraint ranking) sometimes selects the underspecified input, sometimes a fully specified one. The underspecified input is selected as the input among various possibilities when the mapping between that underspecified input and the output is the most harmonious in comparison with the mapping for other inputs. Or as Kager puts it (1999: 33), “whenever the learner has no evidence (from surface forms) to postulate a specific diverging lexical form, (s)he will assume that the input is identical to the surface form.” In the case of Spanish voiced obstruents, this paper argues that the learner has no conclusive evidence to select one specified input over the other ([+continuant] over [-continuant] or vice versa) and that therefore the input remains underspecified (i.e., undetermined) for continuancy. Also, it is shown that the underspecified input is preferred over fully specified ones because it is the one that incurs the least number of violations of ID -[cont] and therefore it is the most harmonious with respect to the output. In addition to Inkelas (1995), various other authors have shown the need for input underspecification in OT (Harrison and Kaun 2001; Inkelas 2006; Inkelas, Orgun and Zoll 1997; Itô, Mester and Padgett 1995). Harrison and Kaun (2001) argue that OT allows for principled underspecification and that it “. . . leaves room for the possibility that partially underspecified lexical entries will on occasion be posited (Harrison and Kaun 2001: 212).” In sum, underspecification does not go against Richness of the Base, as long as underspecified forms are determined by the learner through LO and not by any constraints holding directly on underlying forms (Inkelas 1995: 289). Colina (2013), in an analysis of Galician geada, contends that Galician obstruent alternation is similar to that of Spanish and that it requires an underspecified input, as the grammar (the set of independently established constraints and constraint ranking) selects the correct output only when it is in correspondence with an underspecified input. Other languages and processes in which an underspecified input is selected include Yoruba ATR harmony, Walpiri vowel harmony and Turkish vowel-glide alternations (Inkelas 1995).

On onset clusters in Spanish: voiced obstruent underspecification and /f/

123

Also within the OT framework, Hale and Kissock (2007) argue for a different type of underspecification that they call perseverant underspecification. This is underspecification that persists from underlying representation through phonetic representation. It results in forms that are never fully specified featurally and that are realized variably according to the surrounding segments, i.e., in the phonetic component. It is also known as phonetic or output underspecification and was originally proposed by Keating (1988), who convincingly argues that the velar fricative in Russian /ixa/ is underspecified for the feature [back] because it shows the continuous, transitional features characteristic of sounds which are dependent in their articulation upon their adjacent sounds. Output underspecification accounts for phenomena that are highly variable and gradient in nature. As mentioned above, I contend that it also explains the great degree of phonetic variation in aperture in continuant allophones, generally dependent on the adjacent sounds (cf. Bradley and Delforge 2006a, Bradley and Delforge 2006b, Bradley 2007, Colina 2009b for output underspecification in various dialects of Spanish; cf., however, Colina 2013, for a phenomenon that requires input underspecification, but does not appear to involve phonetic underspecification). Before continuing with the analysis, it must noted that the output underspecification analysis of Spanish voiced obstruents proposed relates to a controversial issue in the phonetics/phonology interface, namely how gradual phonetic phenomena should be accounted for by formal phonology. Accounts range from those that incorporate formal mechanisms such as phonetically-grounded constraints, input-output relations (cf. van Oostendorp 2008) and/or underspecification (this paper), to those that argue against formalist views of phonology (Port & Leary 2005). While the matter is still under debate, it is clear that most relevant phenomena evince the presence of a phonological component, often through underlying distinctions (cf. voicing in Catalan and Dutch, van Oostendorp 2005: 1364). In the current proposal, as shown below, a three-way contrast in continuancy [+] [–] and [ ] accounts for the alternations and phonological behavior observed in voiced obstruents (vs. voiceless stops and fricatives), thus demonstrating that, despite the important role played by the phonetics, the process is also phonological. Moving on to the OT analysis, it is now necessary to add two faithfulness constraints, DEP-[cont] and MAX-[cont], that contemplate the presence or absence (i.e., insertion or deletion) of a [continuant] feature in the output with respect to the input to the constraints and constraint ranking established in (12).

124

Sonia Colina

(16) Voiced obstruents: Constraints and constraint ranking (McCarthy & Prince 1995) ID -[cont]: the output matches the input with respect to [cont] ID -[voi]: the output matches the input with respect to [voi] DEP-[cont]: the feature [cont] present in the output must have a correspondent in the input. MAX-[cont]: the feature [cont] present in the input must have a correspondent in the output. ID -[cont] >> ID -[voi] >> DEP-[cont] >> MAX-[cont] Considering the output as the point of departure now, as is customary in OT, it becomes clear that the most harmonious input for an underspecified output must be also the underspecified input, because any other input would either violate a faithfulness constraint, as shown in (17), or it would result in the selection of the wrong output, as seen with fully specified inputs (18–19). In (17) the optimal candidate, underspecified [aBlo] (17a), does not incur any constraint violations. Candidates specified for continuancy, (17b–d), with default value insertion) violate the constraint against the insertion of this feature (be it + or –); (17d), [aplo], also incurs a violation of ID -[voi] since there is a change in the voicing specification. (17)

The underspecified input is the most harmonious input /aBlo/ [aBlo]

ID -[cont]

ID -[voi]

DEP-[cont]

MAX-[cont]

a. ☞ aBlo b. aβlo

*!

c. ablo

*!

d. aplo

*!

*

In (18) and (19) the outputs that are identical to the input, one with a voiced stop in (18c) and one with a voiced approximant in (19b), are mistakenly selected as the winners because they do not incur any constraint violations. It could be argued that a candidate like (19b) would be the winner if the output was [aβlo]; however, the output cannot be [aβlo] as it does not match the phonetic evidence, which indicates that the output has a variable degree of constriction. It should

On onset clusters in Spanish: voiced obstruent underspecification and /f/

125

be noted that, in addition to [continuant], the feature [sonorant] must also be underspecified. It is assigned in the phonetic component and it follows from the degree of constriction: the least constricted approximant realizations of the voiced obstruents are sonorants, while the more constricted fricative and stop realizations are obstruents. (18) The specified input ([-cont]) is not the most harmonious for the output /ablo/ [aBlo]

ID -[cont]

ID -[voi]

DEP-[cont]

a. aBlo b. aβlo

MAX-[ cont] *!

*!

c. ☹ ablo d. aplo

*!

(19) The specified input ([+cont]) is not the most harmonious for the output /aβlo/ [aBlo]

ID -[cont]

ID -[voi]

a. aBlo

DEP-[cont]

MAX-[ cont] *!

b. ☹ aβlo c. ablo

*!

d. aplo

*!

*

One important implication of the OT analysis in (17–19), as indicated in 3.2.1, is that it offers additional evidence for an underspecified input, by demonstrating that it is the most harmonious input for the underspecified output. Notice also that the proposed Optimality-Theoretic account with input and output underspecification is superior to derivational accounts in that it can account for phonetic variability in obstruent realizations, a descriptive fact that is not captured by existing serial accounts (Baković 1997; Harris 1969, 1984; Hualde 1989; Lozano 1979; Mascaró 1984). Because they do not incorporate output underspecification, these analyses have to assume that voiced obstruents are either approximants or fricatives, but variably both (for the same individual and variety). Moving on to the postnasal context, recall that the stop is the most common allophone, and variation is minimal (cf. however recent phonetic studies that

126

Sonia Colina

suggest that in some dialects the aperture continuum may also affect the postnasal position, e.g., Carrasco, Hualde & Simonet 2012). This seems to indicate that the [-cont] is acquired from the adjacent consonant.8 The need for obstruents to agree in continuancy with a preceding, homorganic nasal (captured by the constraint NC[cont], in [12]) can be motivated in a model of feature geometry like the one proposed by Padgett (1994) in which sharing place features means sharing the aperture nodes dominated by them. The connection noted by many researchers between nasal homorganic clusters and voiced stops is a consequence of the domination of aperture nodes by place features in feature geometry models. In (20a) a sequence of a nasal, underspecified for point of articulation, and a voiced bilabial obstruent, without a continuancy specification, becomes a bilabial nasal + bilabial stop through sharing of the place and aperture nodes (20b), a violation of LINK , a constraint that bans the double association of features (cf. [23]). Voiced obstruents, being unspecified for continuancy, share the [-continuant] specification of the nasal (20b). (20)

Feature-geometric representation for nasal + voiced obstruent clusters a.

b.

Agreement in continuancy with a preceding nasal only affects voiced obstruents: voiceless fricatives are not affected. This is because the sharing of aperture nodes only takes place when it does not require altering the aperture node of the 8 Another context exhibiting minimal (if any) variability is absolute word-initial position (postpausal). The pause favors a stop in the phonetic implementation of the underspecified output.

On onset clusters in Spanish: voiced obstruent underspecification and /f/

127

post-nasal segment. Recall that we argue that voiced obstruents are underspecified for continuancy, but voiceless stops and fricatives are specified as [-continuant] and [+continuant], respectively. In Optimality-Theoretic terms, as seen in (12), this means that ID -[cont] dominates NC[cont]. Segments underlyingly specified as [+continuant], such as voiceless fricatives, retain their [+continuant] specification due to the domination of ID -[cont] over NC[cont] (21). As explained above, /f/ does not become a stop, despite its underspecification for [continuant]. This is due to an undominated constraint that bans the highly marked labiodental stops, *p̪ (22). (21)

(22)

Feature-geometric representation for nasal + voiceless fricative clusters a.

Input

b.

Output

Feature-geometric representation for nasal + /f/ clusters a.

Input

128

Sonia Colina

b.

Output

(23) NC[cont]: Post-nasal obstruents must agree in continuancy with the preceding coda nasal, i.e. they must be [-continuant]. LINK : No double association of features (Itô, Mester & Padgett 1995; Colina 2009b) As determined above, LINK must be dominated by NC-[cont], and this in turn by ID -[cont]; however, the ranking between LINK and ID -[voi] is undetermined. (24) Voiced obstruents in NC clusters

I argue that LINK is the responsible constraint in (24) (rather than DEP-[cont]) because insertion (a DEP-[cont] violation) would incur an additional OCP violation by having two contiguous identical association lines. Note that an alternative account with an input identical to the output (i.e., with a stop) in the homorganic context would work, but it would in turn require a stop in non-nasal contexts. As mentioned above, this would not explain the phonetic variability observed in these contexts.

On onset clusters in Spanish: voiced obstruent underspecification and /f/

129

This section presented the evidence and the analysis in favor of the view that voiced obstruents are underspecified for continuancy. This view offers additional evidence for Martínez-Gil’s proposal that /f/ and the voiced obstruents belong to the same sonority class in Spanish (the least sonorous one) because they all lack a [+continuant] specification. [+continuant] is the feature that contributes to sonority. Consequently, the relevant generalization that accounts for onset clusters in Spanish is that a complex onset in Spanish is limited to two consonants that differ maximally in sonority rank (Complex Onset Condition, Martínez Gil 2001: 219) This condition raises a potentially problematic issue – point of application. Since Spanish voiced obstruents are often realized as approximants (i.e., [+continuant]), the Complex Onset Condition appears to be violated at some point. The next section discusses the point of application of the Complex Onset Condition and shows that it constitutes another piece of evidence in support of input and output underspecification of voiced obstruents.

4 The Complex Onset Condition: point of application 4.1 Point of application An important question arises with regard to the point at which the Complex Onset Condition is operative, especially since voiced obstruents are often surface approximants in Spanish, in apparent violation of such a condition. Martínez-Gil (2001), under a derivational framework, proposes that the condition is operative at the level of the underlying representation, where the voiced obstruents would be stops. This makes the sonority proposal and the Complex Onset Condition dependent on an account of voiced obstruent allophony in which the input is a stop. Although Martínez-Gil does not present the details of such an analysis, it is clear that it would be problematic for an Optimality-Theoretic account, as the Complex Onset Condition applies to the input, not to the output, in direct contradiction with the essence of OT. OT analyses (Martínez-Gil 1997, Colina 2009a) have claimed that the condition, formulated as the Maximal Sonority Distance (MSD) constraint, is violated under domination by a more highly ranked constraint, requiring the presence of an approximant realization for voiced obstruents in some contexts (i.e., nonhomorganic clusters and when not preceded by a pause).

130

Sonia Colina

While constraint violation is definitely a possible way of explaining surface approximants, the proposal argued for in this paper – in which voiced obstruents are underspecified in the input and at the output of the phonology –, presents a more economical, elegant and straightforward account. Like all constraints in OT, the Complex Onset Condition (MSD constraint) applies to output forms; in other words, the Complex Onset Condition is a phonological condition that must be satisfied at the output of the phonology. Given that in the current proposal, voiced obstruents still lack the specification [+continuant] at the output of the phonology (which will be assigned later as a matter of phonetic implementation), MSD is satisfied. This explanation is independently needed to account for why /f/ is grouped with the obstruents. In sum, the proposed analysis offers an elegant and straightforward account of the Onset Condition as it does not need to specify or restrict the point at which it applies.

4.2 Chilean dialects Data from some Chilean dialects poses some potential problems for the account that says that the onset condition applies at the output of the phonology. In these dialects, underlying oral stops are vocalized in a syllable coda, becoming glides (Lenz, Bello & Oroz 1940; Oroz 1966; Martínez-Gil 1997). (25)

/p, b/ /t, d/ /k, ɡ/

[w] [j] [j] or [w]

ap.to ét.ni.co ac.to

a[w].to é[j].nico a[j].to, a[w].to

‘apt’ ‘ethnic’ ‘act’

Underlying voiced stops also undergo vocalization when followed by a tautosyllabic liquid in word-medial position (26a–c), but not across words (26d). Voiceless stops in onset clusters are not vocalized (26e–g). (26)

a. b. c. d. e. f. g.

po[w].re ma[j].re vi.na[j].re la.[ð]ro.ga le.[p]ra le.[t]ra sa.[k]ro

*po.[β]re *ma.[ð]re *vi.na.[ɣ]re *la[j].ro.ga *le[w].ra *le[j].ra *sa[j].ro, *sa[w].ro

‘poor’ ‘mother’ ‘vinegar’ ‘the drug’ ‘leprosy’ ‘letter’ ‘holy’

Martínez-Gil (1997) convincingly shows that it is the need to improve the sonority contour of the complex onset that explains why only voiced (not voiceless) onset obstruents vocalize and are subsequently parsed in the coda. In other

131

On onset clusters in Spanish: voiced obstruent underspecification and /f/

words, voiced obstruents become [+continuant], in order to avoid an approximant as the first member of the cluster. The need for a better onset is what motivates the parsing and vocalization of the approximants in the coda. Since voiceless stops do not have approximant allophone, they do not vocalize in an onset cluster (they do, however, when they are in the coda [25]). As seen in (26d), vocalization does not take place across words in an effort to avoid misalignment of syllables and word boundaries. The above account, however, presents a possible obstacle for the analysis proposed here. If, as it is argued in this paper, voiced obstruents lack the feature [+continuant] at the output of the phonology, sonority restrictions (Complex Onset Condition) are not violated and therefore the prediction is that there would be no vocalization. The explanation to this apparent difficulty lies in the behavior of /f/ once again. /f/ patterns differently from the voiced obstruents in onset clusters, exhibiting no vocalization. This is evidence that in this dialect the feature that makes voiced obstruents more sonorous than /f/ is [+voice]. In other words, Chilean requires one additional sonority class, that divides the least sonorous class of voiceless, voiced obstruents and /f/ (those without [+continuant] in their phonological representations) into two: [-voice] and [+voice] (27): (27) Sonority scale for some varieties of Chilean Spanish Obstruents

Sonorants

p, t, k, [+voiceless, +labiodental]

b, d, ɡ

θ, s, h, x, tʃ, dʒ

m, n, ɲ, ʎ, r

l, ɾ

1

2

3

4

5

With a sonority scale like the one in (27), the Complex Onset Generalization for these Chilean dialects only allows segments in class 1 and 5 (e.g., voiceless stops and /f/, and l, ɾ). As mentioned in section 2, the sonority scale is universal with regard to the relative degree of sonority of the classes, so that an obstruent will always be less sonorous than a nasal and a nasal in turn will always be less sonorous that a vowel. Thus, at first glance, the scale in (27) may appear problematic, because /f/ is above the voiced stops. However, (27) does not place a voiceless labiodental fricative over the voiced stops, rather a voiceless obstruent that lacks a continuant specification and cannot therefore be considered a fricative at the point of application of the sonority generalization. It is reasonable to assume that language specific adjustments to the scale will depend on the sound inventory of the language (including language-specific featural specifications) and point of application. Furthermore, the apparent reversal does not involve a major sonority class (such as obstruents vs. nasals).

132

Sonia Colina

In absolute word-initial position and post-nasal environments, voiced stops do occur as the first member of the cluster (Lenz, Bello & Orozco 1940: 108). This can be attributed to the fact that the coda parsing option is not available, as the resulting nasal + stop is an ill-formed coda in Spanish *[tjemb.lo] *[tjemj.lo], and to the high ranking of the relevant constraint. In an alternative explanation that relies on the violation of MSD under domination of the constraints enforcing spirantization, Martínez-Gil (2013) explains the Chilean data through Stratal OT. Stratal OT is a modular version of OT that distinguishes two strata: the lexical stratum (word phonology) and the postlexical one (phrasal phonology) (Bermúdez-Otero 2006, forthcoming; Kiparsky 2000, 2008, 2010). Each level is a grammar with its own ranking and the output of the lexical stratum is the input to the postlexical level. For the general dialects, Martínez-Gil proposes that in the lexical stratum MSD dominates SPIR (the constraint or group of constraints responsible for spirantization)9, because spirantization is a postlexical process; this ranking is reversed in the postlexical stratum, where spirantization surfaces regardless of sonority violations in complex onsets. In the Chilean dialects referred to above, the MSD outranks faithfulness of lexical parsing. In other words, the MSD applies only lexically in most dialects, but in Chilean it is also enforced postlexically. This forces the voiced spirant resulting from S PIR into the coda of the previous syllable where it undergoes vocalization. While this approach works, Stratal OT is a controversial proposal within OT. It has been argued to introduce serialism into OT, in particular because it cannot reconstruct the principles that allowed its serial predecessor (Lexical Phonology) to set limits on the number of levels allowed, and on how levels differ from each other (cf. McCarthy 2002: 172–174). Given the current status of Stratal OT within the theory, an analysis that does not need to resort to stratal solutions is preferable. Furthermore, the stratal account is normally invoked to account for cyclity or opacity effects. The Chilean data are only a case of dialectal variation.

5 Conclusions This paper presented an Optimality-Theoretic analysis of voiced obstruents and of onset clusters in Spanish. Voiced obstruents are argued to be underspecified for [continuancy] in the input and in the output. The proposed analysis accounts 9 Spirantization (SPIR): The value of the feature continuant in voiced obstruents may not disagree with that of the preceding segment (Martínez-Gil 1997: 190).

On onset clusters in Spanish: voiced obstruent underspecification and /f/

133

for why /f/ is the only fricative accepted as the first member of an onset cluster in Spanish, providing support and independent evidence for Martínez-Gil’s proposal that what contributes to sonority is the presence of [+continuant]. Consequently, /f/ patterns with the voiced obstruents due to the absence of [+continuant] in its representation; voiced obstruents and /f/ comply with the Complex Onset Generalization, as they constitute the most sonorous class of sounds in Spanish. Output and input underspecification of voiced obstruents is consistent with recent phonetic evidence on variable and gradient realizations of voiced obstruents and with the absence of continuancy alternation of /f/ in postnasal position. It is also shown that underspecification does not go against the essence of OT and that, on the contrary, underspecified candidates must be considered in accordance with the ROTB. In the analysis presented the underspecified input is selected by the constraints and the constraint ranking as the most harmonious input for the phonetically underspecified output. In addition, the proposed analysis solves the point of application problem faced by the onset generalization in previous analyses. In line with OptimalityTheoretic tenets, restrictions on onset clusters apply to the output forms of the phonology. Finally, data from some Chilean dialects, in which vocalization is preferred to the common open (i.e., approximant) realizations of voiced obstruent, initially pose a challenge to the analysis but are also accounted for. It is shown that a comprehensive account of onset clusters must explain the interactions of these segments with other aspects of the phonology of a language, such as phonemic distinctions, featural specifications and relevant phonological phenomena.

References Baković, Eric. 1997. Strong onsets and Spanish fortition. MIT Working Papers in Linguistics 23. 21–39. Bermúdez-Otero, Ricardo. 2006. Morphological structure and phonological domains in Spanish denominal derivation. In Fernando Martínez-Gil & Sonia Colina (eds.), Optimality-Theoretic studies in Spanish phonology, 278–311. Amsterdam & Philadelphia: John Benjamins. Blevins, Juliette & Sven Grawunder. 2009. *Kl > Tl sound change in Germanic and elsewhere: descriptions, explanations, and implications, Linguistic Typology 13. 267–303. Bermúdez-Otero, Ricardo. forthcoming. Stratal Optimality Theory. Oxford: Oxford University Press. www.bermudez-otero.com/Stratal_Optimality_Theory.htm. Bradley, Travis. 2007. Prosodically-conditioned sibilant voicing in Balkan Judeo-Spanish. In Erin Bainbridge & Brian Agbayani (eds.), Proceedings of the 34th Western Conference on linguistics, 48–60. Fresno, CA: California State University.

134

Sonia Colina

Bradley, Travis & Anne Marie Delforge. 2006a. Systemic contrast and the diachrony of Spanish sibilant voicing. In Randall Gess & Deborah Arteaga (eds.), Historical Romance linguistics: retrospectives and perspectives, 19–52. Amsterdam & Philadelphia: John. Bradley, Travis & Anne Marie Delforge. 2006b. Phonological retention and innovation in the Judeo-Spanish of Istanbul. In Timothy Face & Carol Klee (eds.), Selected proceedings of the 8th Hispanic Linguistics Symposium, 73–88. Somerville, MA: Cascadilla Press. Carrasco, Patricio, José Ignacio Hualde & Miquel Simonet. 2012. Dialectal differences in Spanish voiced obstruent allophony: Costa Rican versus Iberian Spanish. Phonetica 69. 149–179. Clements, George N. 1990. The role of sonority in core syllabification. In John Kingston and Mary Beckman (eds.), Papers in laboratory phonology, 283–333. Cambridge & New York: Cambridge University Press. Colantoni, Laura & Irina Marinescu. 2010. The scope of stop weakening in Argentine Spanish. In Marta Ortega-Llebaria (ed.), Selected Proceedings of the 4th Conference on Laboratory Approaches to Spanish Phonology, 100–114. Somerville, MA: Cascadilla Proceedings Project. Cole Jennifer, Khalil Iskarous & José Ignacio Hualde. 1999. Effects of prosodic and segmental context on /g/-lenition in Spanish. In Osama Fujimura, Brian D. Joseph, & Bohumil Palek (eds.), Proceedings of the Fourth International Linguistics and Phonetics Conference, 575– 589. Prague: The Karolinum Press. Colina, Sonia. 2009a. Spanish phonology: A syllabic perspective. Washington, DC: Georgetown University Press. Colina, Sonia. 2009b. Sibilant voicing in Ecuadoran Spanish. Studies in Hispanic and Lusophone Linguistics. 2. 3–29. Colina, Sonia. 2013. Spanish voiced obstruent alternation and underspecification in OT. Paper presented at the 43rd Linguistic Symposium on Romance Languages, City University of New York, April 19–21. Eddington, David. 2011. What are the contextual phonetic variants of /b d ɡ/ in colloquial Spanish? Probus. 23. 1–19. Flemming, Edward. 2002. Auditory representations in phonology. New York: Routledge. Flemming, Edward. 2007. Stop place contrasts before liquids. In Proceedings of the 16th International Congress of Phonetic Sciences, 233–236. Hale, Mark & Madelyn Kissock. 2007. The phonetics-phonology interface and the acquisition of perseverant underspecification. In Gillian Ramchand & Charles Reiss (eds.), The Oxford handbook of linguistic interfaces, 81–101. Oxford: Oxford UP. Hammond, Robert M. 1976. Phonemic restructuring of voiced obstruents in Miami Cuban Spanish. In Frances M. Aid, Melvyn C. Resnik & Bohdan Saciuk (eds.), 1975 colloquium on Hispanic linguistics, 42–51. Washington, DC: Georgetown University Press. Harris, James W. 1969. Spanish phonology. Cambridge, MA: The MIT Press. Harris, James W. 1983. Syllable structure and stress in Spanish: A non-linear analysis. Cambridge, MA: The MIT Press. Harris, James W. 1984. La espirantización en castellano y la representación fonológica autosegmental. Estudis Gramaticals 1. 149–67. Harris, James W. 1989a. Sonority and syllabification in Spanish. In Carl Kirschner & Janet DeCesaris (eds.), Studies in Romance linguistics, 139–153. Amsterdam & Philadelphia: John Benjamins. Harris, James W. 1989b. Our present understanding of Spanish syllable structure. In Peter Bjarkman & Robert M. Hammond (eds.), American Spanish pronunciation, 151–169. Washington, DC: Georgetown University Press.

On onset clusters in Spanish: voiced obstruent underspecification and /f/

135

Harrison, David & Abigail Kaun. 2001. Patterns, pervasive patterns and feature specification. In T. Alan Hall (ed.), Distinctive feature theory, 211–236. Berlin: Mouton de Gruyter. Henke, Eric, Ellen M. Kaisse & Richard Wright. 2012. Is the sonority sequencing principle an epiphenomenon? In Steve Parker (ed.), The sonority controversy, 65–100. Mouton de Gryuter. Hualde, José Ignacio. 1989. Procesos consonánticos y estructuras geométricas en español. Lingüística 1. 7–44. Hualde, José Ignacio. 2005. The sounds of Spanish. Cambridge & New York: Cambridge University Press. Hualde, José I., Miquel Simonet, Ryan Shosted & Mariana Nadeu. 2010. Quantifying Iberian spirantization: Acoustics and articulation. Paper presented at LSRL 40. http://prosody. beckman.illinois.edu/jihualde/LSRL40–d-final.pdf Hualde, José Ignacio. 2014. La silabificación en español. In Rafael Núñez-Cedeño, Sonia Colina & Travis Bradley (eds.), Fonología generativa contemporánea de la lengua española. Washington, DC: Georgetown University Press. Inkelas, Sharon. 1995. The consequences of optimization for underspecification. In Jill N. Beckman (ed.), Proceedings of the 26th Annual Meeting of the North East Linguistic Society (NELS), 287–302. Amherst, MA: GLSA. Inkelas, Sharon. 2006. Underspecification. In Keith Brown (ed.), Encylopedia of language and linguistics, 224–226. Oxford: Elsevier. Inkelas, Sharon, Cemil Orhan Orgun & Cheryl Zoll. 1997. The implications of lexical exceptions for the nature of grammar. In Iggy Roca (ed.), Derivations and constraints in phonology, 393–418. Oxford: Clarendon Press. Itô, Junko, Armin Mester & Jaye Padgett. 1995. NC: Licensing and underspecification in Optimality Theory. Linguistic Inquiry. 26. 571–614. Kager, Renè. 1999. Optimality Theory. Cambridge, New York: Cambridge University Press. Keating, Patricia, A. 1988. Underspecification in phonetics. Phonology 5. 275–292. Kiparsky, Paul. 2000. Opacity and cyclicity. The Linguistic Review 17. 351–365. Kiparsky, Paul. 2008. Fenno-Swedish quantity: Contrast in Stratal OT. In Bert Vaux & Andrew Nevins (eds.), Rules, constraints, and phonological phenomena, 185–219. Oxford: Oxford University Press. Kiparsky, Paul. 2010. Paradigms and opacity. Chicago: The University of Chicago Press. (Center for the Study of Language and Information Series, Lecture Notes, no. 140). Kirchner, Robert. 1998. An effort-based approach to consonant lenition. UCLA dissertation. Kirchner, Robert. 2004. Consonant lenition. In Bruce Hayes, Robert Kirchner & Donca Steriade, (Eds.). Phonetically-based phonology, 313–345. Cambridge, UK: Cambridge University Press. Ladefoged, Peter. 1992. Course on phonetics. New York: Harcourt Brace. Lenz, Rodolfo, Andrés Bello & Rodolfo Oroz. 1940. El español en Chile. Buenos Aires: Imprenta de la Universidad de Buenos Aires (Biblioteca de Dialectología Hispanoamericana 6) Lozano, M. Carmen. 1979. Stop and spirant alternations: Fortition and spirantization processes in Spanish phonology. Bloomington, IN: Indiana University Linguistics Club. Martínez-Gil, Fernando. 1996. El principio de la distancia mínima de sonoridad y el problema de la vocalización consonántica en el español de Chile. Hispanic Linguistics 8. 201–246. Martínez-Gil, Fernando. 1997. Obstruent vocalization in Chilean Spanish: A serial versus a constraint-based approach. Probus 9. 165–200.

136

Sonia Colina

Martínez-Gil, Fernando. 2001. Sonority as a primitive phonological feature. In Julia Herschensohn, Enrique Mallén & Karen Zagona (eds.), Features and interfaces in Romance, 203–222. Amsterdam & Philadelphia: John Benjamins. Martínez-Gil, Fernando. 2004. Galician geada. In Julie Auger, Joseph Clancy Clements & Barbara Vance (eds.), Contemporary approaches to Romance linguistics, 299–320. Amsterdam & Philadelphia: John Benjamins. Martínez-Gil, Fernando. 2013. Spirantization and the structure of Spanish complex onsets: a Stratal OT solution. Paper presented at the Hispanic Linguistics Symposium, University of Ottawa, October 2013. Mascaró, Joan. 1984. Continuant spreading in Basque, Catalan and Spanish language sound structure. In Mark Aronoff & Richard T. Oerhle (eds.), Language sound structure: Studies in phonology, 287–298. Cambridge, MA: The MIT Press. McCarthy, John & Alan Prince. 1995. Faithfulness and Reduplicative Identity. In Jill Beckman, Laura Walsh, & Suzanne Urbanczyk (eds), University of Massachusetts Occasional Papers in Linguistics 18: Papers in Optimality Theory, 249–384. Amherst, MA: GLSA. McCarthy, John. 2002. A thematic guide to Optimality Theory. Cambridge, New York: Cambridge University Press. Morelli, Frida. 1999. The phonotactics and phonology of obstruent clusters in Optimality Theory. Baltimore, MD: University of Maryland dissertation. Morelli, Frida. 2003. The relative harmony of /s+stop/ onsets: Obstruent clusters and the Sonority Sequencing Principle. In Caroline Féry and Ruben van de Vijver (eds.), The syllable in Optimality Theory, 356–371. Cambridge: Cambridge University Press. Oroz, Rodolfo. 1966. La lengua castellana en Chile. Santiago de Chile: Imprenta de la Universidad de Chile. Ortega-Llebaria, Marta. 2004. Interplay between phonetic and inventory constraints in the degree of spirantization of voiced stops: Comparing intervocalic /b/ and intervocalic /g/ in Spanish and English. In Timothy Face (ed.), Laboratory approaches to Spanish phonology, 237–255. Berlin: Mouton de Gruyter. Padgett, Jaye. 1994. Stricture and nasal place assimilation. Natural Language and Linguistic Theory 12. 465–513. Parker, Steve. 2008. Sound level protrusions as physical correlates of sonority. Journal of Phonetics. 36. 55–90. Parker, Steve. 2011. Sonority. In Marc von Oostendorp, Colin J. Ewen, Elizabeth Hume & Keren Rice (eds), The Blackwell companion to phonology 1160–1184. Malden, MA, Oxford/Wiley Blackwell. Parker, Steve. 2012. Sonority distance vs. sonority dispersion – a typological survey. In In Steve Parker (ed.), The sonority controversy, 101–164. Munich: Mouton de Gryuter. Pater, Joe. 1999. Austronesian nasal substitution and other NC effects. In Rene Kager, Harry van der Hulst, & Wim Zonneveld, (eds.), The Prosody morphology interface. Cambridge: Cambridge University Press. 310–343. Pater, Joe. 2001. Austronesian nasal substitution revisited. In Linda Lombardi, (ed.) Segmental phonology in Optimality Theory: Constraints and Representations. Cambridge: Cambridge University Press. 159–182. Piñeros, Carlos E. 2003. Accounting for the instability of Palenquero voiced stops. Lingua. 113. 1185–1222. Port, Robert & Leary, Adam. 2005. Against formal phonology. Language 81. 927–964.

On onset clusters in Spanish: voiced obstruent underspecification and /f/

137

Simonet, Miquel, José Ignacio Hualde & Mariana Nadeu. 2012. Lenition of/d/in spontaneous Spanish and Catalan. Paper presented at INTERSPEECH. Soler, Antonia & Joaquín Romero. 1999. The role of duration in stop lenition in Spanish. In John Ohala, Yoko Hasegawa, Manjari Ohala, Daniel Granville & Ashlee C. Bailey (eds.), Proceedings of the 14th International Congress of Phonetics Sciences 99, 483–486. Berkeley, CA: Univ. of California, Berkeley. van Oostendorp, Marc. 2008. Incomplete devoicing in formal phonology. Lingua. 118. 1362– 1374. Wright, Richard. 2004. A review of perceptual cues and cues robustness. In Bruce Hayes, Robert Kirchner & Donca Steriade (eds.), Phonetically-based phonology, 34–57. Cambridge: Cambridge University Press.

Fernando Martínez-Gil

Syllable merger in Chicano Spanish: a constraint-based analysis Abstract: This paper presents an analysis of hiatus resolution in Chicano Spanish, a variety spoken in the southwest of the United States, whereby a vowel sequence across word boundaries is merged into one syllable. The theoretical interest of syllable merger lies in accounting for the five distinct patterns found in the phonetic realization of the merged syllable, which invariably involves moraic shortening: its duration being equal to that of a single vowel. The present analysis, based on Stratal OT, holds that merger involves linking a word-final vowel to the onset of a syllable; whenever dominant constraints proscribe onset incorporation, hiatus is resolved by coalescence and deletion. Challenging the assumption that merger involves the combination of the vowel sequence into a complex nucleus, Stratal OT is shown to provide an insightful account of hiatus resolution which avoids the theoretical and empirical shortcomings of previous analyses within the parallel OT framework.

1 Introduction Hiatus-avoiding strategies across languages have commanded the attention of researchers working within the framework of Optimality Theory (OT), who have argued that such phenomena are governed by constraints.1 Among them, Casali (1997, 1998, 2011) presents arguably the most comprehensive crosslinguistic surveys to date of hiatus avoiding processes, encompassing a broad range of empirical sources. This paper presents an OT account of the resolution of hiatus

1 See, e.g., McCarthy 1993, Rosenthall (1997a, 1997b), Casali (1997, 1998, 2011), Senturia (1998), Kang (1999), Kawahara and Hara (2009), and St-Amand (2012), among others. Note: Portions of this paper were presented at the 34th Linguistic Symposium on Romance Languages, held at the University of Utah in March, 2004. I am grateful to Sonia Colina and two anonymous reviewers for their comments. All errors and omissions are my own. Fernando Martínez-Gil, The Ohio State University

140

Fernando Martínez-Gil

by means of syllable merger across word boundaries in Chicano Spanish (henceforth ChSp), a variety spoken in the southwestern states of New Mexico and Texas, as described in Hutchinson (1974), Reyes (1978), and Harms (1977: 45).2 As in most Spanish varieties, when a word-final vowel is followed by a wordinitial one in ChSp, hiatus is resolved by combining the two vowels into one syllable. For convenience, in the remainder of this paper I will refer to the word-final and word-initial vowels as V1 and V2 , respectively. As in certain other varieties of Spanish, whenever V1 survives the merger process in ChSp, it surfaces as a high glide, and the merged syllable is invariably headed by V2 . Syllable merger in all varieties of colloquial Spanish, including ChSp, is known to be predominantly favored by informal speech registers and fast tempo, roughly corresponding to what Harris (1969: 7) describes as the allegretto and presto styles. In addition, prosodic factors such as stress are also known to be involved in syllable merger in Spanish. Thus, in general, the process is most favored if the two vowels targeted by merger are unstressed, but it is either highly precarious or precluded entirely when both vowels bear primary stress. As for ChSp, the various outcomes of syllable merger are determined by a number of factors, including sequential order, feature specification, applicable phonotactic constraints on syllable constituents, universal and language-specific restrictions on syllabic organization, and syntactic construction type. Other American Spanish dialects, including Mexican varieties, exhibit similar patterns (see, e.g., Marden 1938, Mattluck 1951, 1995, Bowen 1956–1957: 6, Boyd-Bowman 1960, Stockwell and Bowen 1965: 111, Cárdenas 1967), though not necessarily the whole set of hiatus-resolving strategies observed in ChSp. On the other hand, the patterns of syllable merger in ChSp differ significantly from those found in standard varieties of Iberian Spanish, where merger is determined primarily by the relative sonority of the vowel sequence: the most sonorous vowel generally heads the merged syllable, independently of word-initial vs. word-final position (cf. Navarro Tomás 1977: 147ff., Roca 1991, Hualde 1994, Hualde, Simonet, and Torreira 2008). Serial accounts of hiatus resolution in ChSp in the past typically generate the observed syllable merger patterns by positing a number of formal mechanisms

2 The studies of Jenkins (1999), and Alba (2005, 2006, 2008), deal with hiatus resolution in New Mexico Spanish varieties. They are primarily concerned with sociolinguistic and stylistic variation and/or the role of word frequency; they report additional patterns of hiatus resolution in these varieties that differ in part from those contained in the three main empirical sources consulted in this paper.

Syllable merger in Chicano Spanish: a constraint-based analysis

141

whose motivation relies to a certain extent on universal markedness constraints, but arguably fail to reach an adequate level of explanation (see, for example, the segmental treatments in Hutchinson 1974 and Reyes 1978, or the autosegmental approaches in Clements and Keyser 1983 and Martínez-Gil 2000).3 The main purpose of this paper is to present a constraint-based analysis of hiatus resolution in ChSp, following (Martínez-Gil 2000)’s proposal that a sequence of two vowels in hiatus across word boundaries /. . .V1 # V2. . ./ is merged into one syllable by a resyllabification process that incorporates V1 to the onset of the syllable headed by V2 . In the OT approach adopted here, the various patterns of hiatus resolution can be derived exclusively from the domination of a widely acknowledged set of universal markedness constraints over a number of segmental faithfulness constraints which demand identity between input and output. Hiatus avoidance is driven by an undominated syllable structure constraint, ONSET, requiring that syllables have onsets. Full integrity of V2 in syllable merger is compelled by ANCHOR- LEFT, a constraint that demands complete faithfulness to the moraic and segmental feature of a vowel located on the left edge of a morpheme/word. Quite generally, in sequences of a non-low V1 followed by a heterorganic V2 , other highly ranked constraints require that V1 be parsed as the onset of the merged syllable, thereby losing its moraic status, and becoming a glide; if V1 is mid, it is concomitantly raised to high, a result compelled by the highly-ranked phonotactic markedness constraint GLIDE -HI , demanding that nonmoraic vocoids be specified as [+high]. Elsewhere, highly-ranked constraints lead to deletion of a low V1 , or to the coalescence of V1 and V2 when both have identical feature specifications. This account differs in a number of significant details from the other two available treatments of the ChSp data, namely, Baković (2006), and Colina (2009: 59–64), most crucially in that under these authors’ set of proposed constraints and constraint interaction, rather than the onset, V1 is instead compelled to attach to the nucleus of the merged syllable, where it shares a mora with V2 , the nuclear vowel. It is shown that an analysis in which the onglide functions as the onset of the merged syllable, rather than a member of a complex nucleus, presents significant advantages over the two competing OT analyses, by providing a uniform explanation of the various alterations undergone by V1 upon the merger taking place, and also broadening the empirical coverage, as it is extended to instances of syllable merger that had been overlooked in previous accounts. While the primary goal of this work is the analysis of syllable merger

3 I will not present a critique of these accounts here; for a discussion on the shortcomings of serial approaches, see Colina (2009: 64–68).

142

Fernando Martínez-Gil

in ChSp, it is also intended as a contribution to the study of Spanish syllable structure in general. The remainder of this article is organized as follows. Section 2 presents the data on hiatus resolution in ChSp. Section 3 discusses the syllabic affiliation of glides in Spanish within the framework of moraic phonology, with particular attention to the syllabification of prevocalic glides. An OT analysis of syllable merger in ChSp is put forward in Section 4 where it is argued that hiatus is resolved by the incorporation of V1 to the onset of the merged syllable. Section 5 addresses several remaining issues, including stress shift, syllable merger and complex onsets, word-internal merger, syllable merger in three-vowel sequences, and presents a brief comparison with two other available OT treatments of ChSp syllable merger: Baković (2006) and Colina (2009). Finally, Section 6 offers some concluding remarks.

2 The data: five patterns of syllable merger in Chicano Spanish The data in (1) through (4) are illustrative of hiatus resolution by means of syllable merger in colloquial ChSp. In this process, V1 either undergoes deletion or quite generally surfaces as a glide, whereas V2 invariably preserves its syllabic status. The data illustrates three types of sequences of: a) two unstressed vowels (1); b) an unstressed vowel followed by a stressed vowel (2); and c) a stressed vowel followed by an unstressed vowel (3). The examples are representative of all logically possible combinations of two vowels across word boundaries; they represent five distinct patterns of syllable merger to which I will turn shortly.4 4 Examples in (1)–(3) are given in their standard conventional spelling followed by their (widely assumed) underlying forms, the corresponding surface representations, and the English gloss. Lexical stress is indicated with an acute accent mark. The vowel sequences that undergo merger are underlined for visual convenience. Whenever deemed necessary for the sake of clarity in (1)–(3) and in subsequent examples syllable boundaries will be indicated by dots. The data have been drawn either directly from the three primary sources (Hutchinson 1974, Reyes 1978, and Harms 1977: 45) or are fully consistent with their description of syllable merger given therein. Sequences of stressed vowels are not included because the consulted sources present a somewhat inconsistent pattern. Thus in the data provided by Hutchinson (1974) they also undergo merger. However, in the variety described by Reyes (1978: 14–16), merger of two vowels that bear lexical stress fails to take place in a number of syntactic constructions; in other contexts, merger involves temporal shortening, and stress coalesces on the merged sequence, but unlike the data in (1)–(3), no deletion or raising takes place: the segmental features of both vowels remain unaltered. I have opted to ignore this complication here, since it does not affect the main substance of my analysis.

Syllable merger in Chicano Spanish: a constraint-based analysis

143

(1) V1 and V2 are unstressed a. Identical vowels: /mi # iɡlésia/ mi iglesia /tu # unifóɾme/ tu uniforme /se # eskapó/ se escapó como Homero /komo # oméɾo/ /paɡa # adán/ paga Adán

[mi.ɣlé.sja] [tu.ni.fóɾ.me] [sés.ka.pó] [ko.mo.mé.ɾo] [pá.ɣa.ðán]

‘my church’ ‘your uniform’ ‘he fled’ ‘like Homero’ ‘Adán pays’

b. A mid vowel followed by a non-identical homorganic me iría /me # iɾía/ [mi.ɾí.a] [poɾ.ki.tá.lja] porque Italia /poɾke # itália/ como uniforme /komo # unifóɾme/ [ko.mu.ni.fóɾ.me] /no # usé/ [nu.sé] no usé

high vowel: ‘I would go’ ‘because Italy’ ‘as (a) uniform’ ‘I did not use’

c. A low vowel followed by a non-low vowel: habla inglés /ábla # inɡlés/ [á.βliŋ.ɡlés] /esta # unión/ [es.tu.njón] esta unión [u.neɾ.má.na] una hermana /una # eɾmána/ [kán̪.to.mé.ɾo] canta Homero /kánta # oméɾo/

‘he speaks English’ ‘this union’ ‘a sister’ ‘Homero sings’

d. A high vowel followed by a non-identical vowel: mi uniforme mi esposa casi olvidé casi acabó tu imagen tu esposa su opinión su amigo

/mi # unifóɾme/ /mi # espósa/ /kási # olbidé/ /kási # akabó/ /tu # imáxen/ /tu # espósa/ /su # opinión/ /su # amíɡo/

[mju.ni.fóɾ.me] [mjes.pó.sa] [ká.sjol.βi.ðé] [ká.sja.ka.βó] [twi.má.xen] [twes.pó.sa] [swo.pi.njón] [swa.mí.ɣo]

‘my uniform’ ‘my wife’ ‘I almost forgot’ ‘he almost ended’ ‘your image’ ‘your wife’ ‘his opinion’ ‘his friend’

e. A mid vowel followed by a non-identical heterorganic vowel: me usó se olvidó se acabó lo hicieron no esperó tengo amigos

/me # usó/ /se # olbidó/ /se # akabó/ /lo # isiéɾon/ /no # espeɾó/ /téngo # amíɡos/

[mju.só] [sjol.βi.ðó] [sja.ka.βó] [lwi.sjé.ɾon] [nwes.pe.ɾó] [téŋ.gwa.mí.ɣos]

‘he used me’ ‘he forgot’ ‘it ended’ ‘they did it’ ‘he didn’t wait’ ‘I have friends’

144

Fernando Martínez-Gil

(2)

V1 is unstressed and V2 stressed a.

An unstressed vowel followed by a stressed identical vowel: mi hijo tu último porque Eva lo odia Eva habla

b.

e.

‘my son’ ‘your last (one)’ ‘because Eva’ ‘he hates it’ ‘Eva speaks’

/tiéne # ípo/ /este # íxo/ /ténɡo # úno/ /lo # úsa/

[tjé.ní.po] [es.tí.xo] [téŋ.ɡú.no] [lú.sa]

‘he has hiccups’ ‘this son’ ‘I have one’ ‘he uses it’

An unstressed low vowel followed by a stressed vowel: una india Eva úsa paga Eva la otra

d.

[mí.xo] [túl ̪.ti.mo] [poɾ.ké.βa] [ló.ðja] [é.βá.βla]

A mid vowel followed by a stressed homorganic high vowel: tiene hipo este hijo tengo uno lo usa

c.

/mi # íxo/ /tu # último/ /poɾke # éba/ /lo # ódia/ /éba # ábla/

/una # índia/ /éba # úsa/ /páɡa # éba/ /la # ótɾa/

[u.nín̪.dja] [é.βú.sa] [pá.ɣé.βa] [ló.tɾa]

A high vowel followed by a stressed vowel: /mi # último/ [mjúl ̪.ti.mo] mi último /si # éba/ [sjé.βa] si Eva /mi # ótɾo/ [mjó.tɾo] mi otro /si # ábla/ [sjá.βla] si habla /su # íxo/ [swí.xo] su hijo /tu # éba/ [twé.βa] tu Eva /tu # óbɾa/ [swó.βɾa] tu obra /tu # álma/ [twál.ma] tu alma

‘an Indian-FEM .’ ‘Eva uses’ ‘Eva pays’ ‘the other-FEM .’ ‘my last (one)’ ‘if Eva’ ‘my other (one)’ ‘if he speaks’ ‘his son’ ‘your Eva’ ‘your deed’ ‘your soul’

A mid vowel followed by a stressed heterorganic vowel: te usa te odian te hablan tengo hipo como Eva no habla

/te # úsa/ /te # ódian/ /te # áblan/ /ténɡo # ípo/ /komo # éba/ /no # ábla/

[tjú.sa] [tjó.ðjan] [tjá.βlan] [téŋ.ɡwí.po] [ko.mwé.βa] [nwá.βla]

‘he uses you’ ‘they hate you’ ‘they speak to you’ ‘I have hiccups’ ‘like Eva’ ‘he does not speak’

Syllable merger in Chicano Spanish: a constraint-based analysis

145

(3) V1 is stressed and V2 unstressed a. A stressed vowel followed by an identical vowel: comí higitos Cantú usó Cheché entró habló Homero vendrá Adán

/komí # iɡítos/ /kantú # usó/ /ʧeʧé # entɾó/ /abló # oméɾo/ /bendɾá # adán/

[ko.mí.ɣí.tos] [kan̪.tú.só] [ʧe.ʧén̪.tɾó] [a.βló.mé.ɾo] [ben̪.dɾá.ðán]

‘I ate figs-DIM . ’ ‘Cantú used’ ‘Cheché went in’ ‘Homero spoke’ ‘Adán will come’

b. A stressed mid vowel followed by a non-identical homorganic high vowel: compré higuitos Cheché iría costó un chingo usó uniforme c.

/kompɾé # iɡítos/ /ʧeʧé # iɾía/ /kostó # un # ʧínɡo/ /usó # unifoɾme/

[com.pɾí.ɣí.tos] [ʧe.ʧí.ɾía] [kos.túň.ʧíŋ.ɡo] [u.sú.ni.foɾ.me]

‘I bought figs-DIM . ’ ‘Cheché would go’ ‘it cost-PAST a lot’ ‘he used a uniform’

A stressed low vowel followed by a non-low vowel: vendrá Inés está usado será Evita será Ortega

/bendɾá # inés/ /está # usádo/ /seɾá # ebíta/ /seɾá # oɾtéɡa/

[ben̪.dɾí.nés] [es.tú.sá.ðo] [se.ɾé.βí.ta] [se.ɾóɾ.té.ɣa]

‘Inés will come’ ‘it is used’ ‘it must be Evita’ ‘it must be Ortega’

d. A stressed high vowel followed by a non-identical vowel: comí uvitas comí enchiladas pedí ochenta así acabó tú imitabas Cantú está Cantú olvidó Cantú admiraba e.

/komí # ubítas/ /komí # enʧiládas/ /pedí # oʧénta/ /así # akabó/ /tú # imitábas/ /kantú # está/ /kantú # olbidó/ /kantú # admiɾaba/

[ko.mjú.βí.tas] [ko.mjéň.ʧi.lá.ðas] [pe.ðjó.ʧén̪.ta] [a.sjá.ka.βó] [twí.mi.tá.βas] [kan̪.twés.tá] [kan̪.twól.βi.ðó] [kan̪.twáð.mi.ɾá.βa]

‘I ate grapes-DIM . ’ ‘I ate enchiladas’ ‘I asked for eighty’ ‘it finished this way’ ‘you were imitating’ ‘Cantú is’ ‘Cantú forgot’ ‘Cantú was admiring’

A stressed mid vowel followed by a non-identical heterorganic vowel: /tomé # ubítas/ [to.mjú.βí.tas] ‘I took grapes-DIM . ’ tomé uvitas Cheché olvidó /ʧeʧé # olbidó/ [ʧe.ʧjól.βi.ðó] ‘Cheché forgot’ /ʧeʧé # akabó/ [ʧe.ʧjá.ka.βó] ‘Cheché finished’ Cheché acabó quedó Inés /kedó # inés/ [ke.ðwí.nés] ‘Inés remained’ /abló # ebíta/ [a.βlwé.βí.ta] ‘Evita spoke’ habló Evita habló Adán /abló # adán/ [a.βlwá.ðán] ‘Adán spoke’

Following Clements and Keyser (1983: 86), the individual surface outcomes of merger for each vowel sequence in (1)–(3) can be summarized schematically as in (4), where instances in which V1 is realized as a glide are shaded for visual convenience:

146

Fernando Martínez-Gil

(4)

/. . .i # V. . ./

/. . .u # V. . ./

/. . .e # V. . ./

/. . .o # V. . ./

/. . .a # V. . ./

i#i→i

u#u→u

e#e→e

o#o→o

a#a→a

i # u → ju

u # i → wi

e#i→i

o#u→o

a#i→i

i # e → je

u # e → we

e # u → ju

o # i → wi

a#u→u

i # o → jo

u # o → wo

e # o → jo

o # e → we

a#e→e

i # a → ja

u # a → wa

e # a → ja

o # a → wa

a#o→o

As illustrated in (4), syllable merger exhibits several systematic properties. First, the first vowel is sometimes deleted, while the second vowel is always maintained. Second, identical vowel sequences undergo coalescence. Third, when the first member of the sequence is not deleted, it invariably surfaces as a high glide, while the second member invariably preserves its syllabic status; in the process of desyllabification, mid vowels are raised to high. Fourth, merger is accompanied by temporal shortening (Reyes 1978: 6), the resulting diphthong being approximately half the timing of a lexical diphthong, as shown instrumentally by Hutchinson (1974). Finally, when a vowel that bears primary stress is desyllabified, stress shifts rightwards onto its nuclear counterpart, as illustrated by the data in (3) above. In sum, hiatus resolution by syllable merger in ChSp is accomplished by means of five distinct patterns, where the term pattern is used here to designate the set of structural changes undergone by the first vowel. I identify them in (5) as Patterns I–V: (5)

a.

Pattern I: Identical Vowel Coalescence (IDVCOAL): two identical vowels coalesce into one (e.g., se escapó /se # eskapó/ → [sés.ka.pó] ‘he fled’; cf. (1a) above).

b.

Pattern II: Mid Vowel Deletion (MIDVDEL): a mid vowel is deleted when followed by a homorganic high vowel (i.e., one with which it agrees for both backness and roundness; e.g., me iría /me # iɾía/ → [mi.ɾí.a] ‘I would go’; cf. (1b)).

c.

Pattern III: Low Vowel Deletion (LOW VDEL): a low vowel is deleted before another vowel (e.g., habla inglés /ábla # inɡlés/ → [á.βliŋ.ɡlés] ‘he speaks English’; cf. (1c)).

Syllable merger in Chicano Spanish: a constraint-based analysis

147

d.

Pattern IV: High Vowel Desyllabification (HI VDES ): a high vowel is desyllabified, and thus becomes a glide, before a non-identical vowel (e.g., mi esposa /mi # espósa/ → [mjes.pó.sa] ‘my wife’ cf. (1d)).

e.

Pattern V: Mid Vowel Raising (MIDVRAIS ): a mid vowel is raised and concomitantly desyllabified when followed by a non-identical vowel (e.g., me usó /me # usó/ → [mju.só] ‘he used me’; cf. (1e)).

Patterns I (IDVCOAL) and IV (HI VDES ), are general in colloquial Spanish. Patterns II (MIDVDEL), III (LOW VDEL), and V (MIDVRAIS ) are widely attested in other American Spanish varieties (see, e.g., Stockwell and Bowen 1965: 109ff., Dalbor 1997, ch. 8, Hammond 2001, ch. 22). A fundamental question that arises with regards to the data in (1)–(3) is whether the patterns of syllable merger summarized in (5) result from languagespecific properties, or whether they arise from general phonological principles at work in ChSp. If it can be shown that the latter is true, then the challenge presented by the data is how to capture the five distinct patterns of syllable merger in a unitary way, such that they follow with minimal stipulation and formal complexity from the operation of universal principles.

3 The syllabic affiliation of Spanish glides As we have just seen, in ChSp syllable merger V2 is preserved unaltered, whereas V1 either becomes a glide or is deleted. Thus, an unavoidable prerequisite to an adequate OT analysis of ChSp syllable merger involves establishing unequivocally the syllabic affiliation of the prenuclear glide that results from V1 desyllabification (cf. (4) above), so that the adequate constraints and constraint ranking can be identified in order to generate the observed outputs, while simultaneously ruling out all other potential competitors. An additional goal is to determine the syllabic affiliation of competing candidates containing postnuclear glides, so that they are excluded under the proposed constraints and constraint interaction. Finally, our analysis should minimally provide some insight as to why in certain instances deletion of V1 is preferred to gliding (precisely the unshaded sequences in (4) above). Consider the representation of syllable structure in moraic phonology, arguably the dominant framework in the last two decades, in which a syllable consists of at least one obligatory segmental unit associated to a mora, the unit of syllable weight. Relevant for our purposes are the three configurations in (7), where the positions X, Y, and Z stand for units of the segmental melody.

148 (7)

Fernando Martínez-Gil

a.

b.

c.

The moraic element Y in (7a) is invariably the nucleus. The segments Y and Z correspond to the traditional rhyme constituent (Steriade 1982, Harris 1983). In weight-insensitive languages, the syllabic margins are non-moraic units, and thus linked directly to the syllable node, as in (7a): the premoraic unit, X, corresponds to the traditional onset constituent; its postmoraic counterpart, Z, to the coda. Some authors assume that in these languages the weightless postmoraic unit Z is nuclear, since it shares its mora associated with the nuclear segment, as in (7b). For weight-sensitive languages, such as Spanish (Harris 1983, 1991, 1995; cf. the Weight-to-Stress constraint in Roca 2006), a postnuclear consonant is rendered moraic by position, as in (7c).5 Shown in (8) are the six logical possibilities involving the tautosyllabic combination of a nuclear vowel (represented with /a/) and a high vocoid (abbreviated as Vh). Examples (8a–c) encompass the three parsing possibilities of prevocalic glides, and (8d–f) those of postvocalic ones. (8)

a.

b.

c.

d.

e.

f.

In the remainder of this paper, I refer to the high vocoid in (8a) as an onset glide, and that in (8d) as a coda glide. Monomoraic configurations such as (8b) and (8e) will be referred to as a light diphthongs, and bimoraic ones like (8c) and 5 In spite of the seemingly compelling arguments from unmarked lexical stress to consider Spanish a language sensitive to syllable weight, this claim has been questioned in recent years, especially in the case of prevocalic glides, and it still remains controversial (see Roca 2006 and references therein, and Piñeros, this volume). It will become evident in this paper that the issue is orthogonal to the analysis of the facts on syllable merger in ChSp.

Syllable merger in Chicano Spanish: a constraint-based analysis

149

(8f) as heavy diphthongs, thus directly reflecting their corresponding syllabic weight. Each structure in (8) is followed by the particular transcription used in the remainder of this paper: diphthongs, heavy or light, are indicated with a top ligature in order to distinguish them from those containing onset or coda glides (8a) and (8d), respectively, and the heavy diphthongs (8c) and (8f) are transcribed in boldface in order to distinguish them from the light ones (8b) and (8e). It is important to point out that in terms of standard autosegmental representations the glide in light diphthongs such as (8c) and (8f) can only be construed as moraic, since it is directly dominated by the nuclear mora.6 The notion of a “moraic glide” is an apparent oxymoron; in the original formulation of Hayes (1989a) a glide is, by definition, a non-moraic vowel, in which case only the high vocoids in (8a) and (8d) qualify as glides. This is, for example, the position adopted in Hualde’s (1994) classic study of syllable merger in standard Iberian Spanish (see also Rosenthall 1997a: 4–5). And yet, there seems to be compelling evidence that in some languages glides must be analyzed as moraic, since they bear syllable weight (cf. Rosenthal 1997a, 1997b, Smith 2002, 2005). This points to an apparent caveat of moraic theory: if we accept that the distinction between a high vowel and a glide is merely one of syllabic affiliation, as widely assumed since Levin (1995) and much subsequent work, a moraic high vowel cannot possibly be distinguished structurally from a moraic high glide: namely a high vocoid will be processed as a glide only if less sonorous than its mid or low counterpart. Thus, while non-moraic high vowels, insofar as they are directly dominated by the syllable node, as in (8a) and (8d), have a direct phonological interpretation as glides, in the remaining configurations in (8) the “glideness” of a high vocoid (whether independently moraic or morasharing) can only be ascertained at the level of phonetic implementation, on the basis of relative sonority: a high vowel will be realized as a glide if the adjacent tautosyllabic vowel is of higher sonority.7 Of course, a familiar problem immediately arises if both vocoids are high, and thus of equal sonority, as in (9):

6 The conventional phonological interpretation of mora-sharing glides in OT is that they are in fact moraic (see, for example, Smith 2002a, 2005). Accordingly, Colina’s (2009, Ch. 3) assumption, that both in ChSp and elsewhere in Spanish the prevocalic glide in such structures is nonmoraic, and thus violates a constraint requiring that high vowels be associated to a mora, must undoubtedly be in error. 7 The following statement in Baković (2006: 67) is instructive in this respect: “glides are assumed to be nothing more nor less than the phonetic interpretation of a [+high] vocoid in a branching mora [i.e., the light diphthongs in (8b) and (8e) above; FMG] . . . or a branching syllable” [i.e., the heavy diphthongs in (8c) and (8f) above; FMG].

150 (9)

Fernando Martínez-Gil

a.

b.

In ChSp this situation arises when merger results in unstressed vowel sequences of non-identical high vocoids (cf. 1d–e)). Clearly, the determination of which of the two vowels is nuclear and which is a glide has to be made on a basis other than sonority, a problem to which we return below. A critical issue that arises next is that the choice of one among the six syllabic configurations countenanced by moraic theory in (8) matches the structure of merged syllables in ChSp. As we saw earlier in (4), whenever V1 survives merger the result is invariably a right-headed syllable. With this in mind, (8d–e) can be immediately discarded, since all three are left-headed (i.e., the nuclear vowel precedes the high vocoid). On the other hand, bimoraic structures such as (8c) and (8f) can also be counted out: as mentioned earlier, syllable merger in ChSp brings about a temporal (i.e., moraic) shortening of the vowel sequence, which becomes a short diphthong.8 We are left then with the task of determining whether the prevocalic high vocoid that shows up in the merged syllable is an onset glide, as in (8a), the position taken in Martínez-Gil (2000), or whether it shares the nuclear mora with the following vowel, as in (8b). The latter position is assumed in OT accounts of ChSp syllable merger, such as Baković (2006) and Colina (2009). On the basis of previous research purporting that prevocalic glides are parsed as members of a monomoraic complex nucleus in core (i.e., lexical) syllabification, these two authors assume without further discussion that such parsing simply carries over to the phrase level, where merger takes place. Leaving aside instances in which merger is accomplished by coalescence or by V1-deletion, which of the 8 Moraic shortening is most conspicuous in the merger of sequences of unstressed identical vowels, leading to the neutralization of contrasts between a sequence of two identical vowels and a single vowel, both word-internally, as in azahar ‘orange blossom’ vs. azar ‘chance’ or cohorte ‘cohort’ vs. corte ‘cut’, as well as across word boundaries, as in quiere estirar ‘he wants to stretch’ vs. quieres tirar ‘you can throw’ or estaba hablando ‘he was talking’ vs. estaba blando: ‘it was soft’ (see Navarro Tomás 1977: 66–70, 152–154, Zamora and Guitart 1982: 79, D’Introno, Guitart and Zamora 1988: 143, Quilis 1999: 375, Hammond 2001: 325–326, Monroy Casas 2004: 81ff., Hidalgo Navarro and Quilis Merín 2004: 133, Hualde 2005: 90–91, and Real Academia Española 2011: 339–340, among many others). Shortening can often be disfavored or impeded by factors such as lexical stress and final position in an intonational group (Navarro Tomás 1977: 148ff.).

Syllable merger in Chicano Spanish: a constraint-based analysis

151

two alternatives, (8a) or (8b), is the correct choice becomes, therefore, a fundamental empirical question.9 The assumption of an onset glide (8a) offers two immediate advantages over the nucleus-sharing option (8b). Consider, first, the problem posed by a merger of non-identical high vocoids, such as /. . .i # u. . ./ and /. . .u # i. . ./ (cf., e.g., mi uniforme ‘my uniform’, tu imagen ‘your image’ in (1d)), or a heterorganic mid-high vowel sequence, such as /. . .e # u. . ./ and /. . .o # i. . ./, where the mid vowel undergoes raising (cf., e.g., me usó ‘he used me’, lo hicieron ‘they did it’ in (1e) above). It is a well-known fact of colloquial Spanish that in the absence of lexical stress the nucleus of a tautosyllabic vowel sequence of equal sonority is assigned to the second member (Harris 1985: 38, Hualde 2005: 90), while the first member undergoes temporal reduction, and in many dialects, including ChSp, also (mid vowel) raising, thus reflecting the well-documented fact that Spanish favors rising diphthongs over falling ones.10 It is imperative to keep in mind that syllable merger in ChSp, as in all varieties of Spanish, is a postlexical phenomenon, since it takes place at the phrase level, although the postlexical application of syllable merger may also affect vowel sequences in hiatus wordinternally (see Navarro Tomás 1977: 66–69, 147ff, Hualde 1994), much in line with the across-the-board application of postlexical phenomena. ChSp wordinternal syllable merger will be addressed briefly in Section 5.2 below. The mapping of a prevocalic high or a mid vowel into a glide in syllable merger follows directly from the assumption that V1 is incorporated into the onset of the merged syllable, as illustrated in (10a); mid-vowel raising can be readily captured as the natural outcome of a well-known universal principle requiring that vocoids be high when associated to a syllable margin.

9 Establishing the syllabic affiliation of prevocalic glides created by syllable merger is especially challenging if we consider the well-known difficulties that arise when attempting to find reliable cues for syllabic constituency in experimental work. The task of collecting empirical evidence in support of one particular configuration over the other, thus, typically proceeds from circumstantial and/or indirect evidence. 10 Navarro Tomás (1977: 68) notes: “En los casos en que se encuentran juntas las vocales e, o, el elemento que se cierra y abrevia al producirse la sinéresis es, como en las combinaciones iu, ui, el que va en primer lugar.” [In those instances in which the vowels e, o, are combined into one sequence by syllable merger, it is the first member that undergoes closing and shortening, as is also the case for the vowel sequences iu, ui”; my translation, FM-G]. This property is amply corroborated in the experimental study by Hualde, Simonet and Torreira (2007).

152 (10)

Fernando Martínez-Gil

a.

b.

The same result, however, does not follow from a mapping involving the morasharing alternative in (10b), where it is not clear what principle, whether universal or particular to Spanish, would ensure that nuclear status is granted to V2 , in detriment to V1 , especially when V1 is of higher sonority than V2; here, reliance on phonetic interpretation alone as the criterion to select the correct outcome would evidently be of no avail. On the other hand, the assumption that merger proceeds by incorporation of V1 as a nuclear member of the merged syllable, as in (10b), encounters serious problems when attempting to account for the fact that V1 becomes a glide in heterorganic mid-high vowel sequences, such as /. . .e # u. . ./ and /. . .o # i. . ./: given its lower sonority. In such cases it is precisely a glide interpretation of V2 that would be expected in phonetic implementation.11 An added bonus of assuming an onset affiliation of prevocalic glides (8a)/(10a) is that the shift of stress from V1 onto an unstressed V2 concurrent with syllable merger follows directly from the fact that glides in Spanish (perhaps universally) cannot be primary-stress bearers, and thus when a stressed V1 is parsed into the onset of the merged syllable (cf. examples in (3d–e)), stress shifts accordingly onto V2 .12 By contrast, in cases where V1 is of equal sonority or higher sonority than V2 , it is not clear how stress-shift would follow from merger under the assumption that V1 is incorporated to the nucleus of the merged syllable, especially when merger applies to sequences of equal or falling sonority. 11 Both Baković (2006) and Colina (2009) assume that the short raising diphthongs created in ChSp syllable merger adopt the nuclear configuration (8b) above, but their characterization of Spanish falling diphthongs differs significantly. In the process of considering potential candidates to be rejected by constraint evaluation, Baković assumes that falling diphthong are always bimoraic, as in (8e), where the second member is in the coda, a seemingly controversial assumption, as we have seen earlier. In contrast, Colina maintains that they are monomoraic, in which the second member is a coda glide, as in (8d). Remarkably, candidate evaluation by both authors excludes (8e), the mirror-image of (8b), from any consideration whatsoever as a potential structural configuration for falling diphthongs in ChSp, a position that contravenes a central tenet of OT, freedom of analysis, as embodied in the principle known as the Richness of the Base. 12 An analogous process takes place when merger results in V1-deletion (cf. (3b–c)): stress is inherited by V2 , in a manner reminiscent of the property Goldsmith (1990: 27ff.) terms tonal stability in tone languages.

Syllable merger in Chicano Spanish: a constraint-based analysis

153

A second advantage of parsing of V1 into the onset of the merged syllable is that in one particular environment, namely, absolute syllable-initial position, prevocalic glides must inevitably be in the onset, as in (8a), not the nucleus, as in (8b). Harris and Kaisse (1999: 126), arguably the most comprehensive treatment of Spanish glides in the last decades, observe that in core syllabification prenuclear high vocoids are opportunistic: when no consonant is available for syllable-initial position, they attach to the onset, whereas in most dialects they undergo strengthening and become consonantal13, accounting for glide ~ consonant alternations such as le[j] ‘king’ ~ le.[ ʝ]es ‘kings’, re[j] ‘king’ ~ re.[ ʝ]es ‘kings’, U.ru.gua[j] ‘Uruguay’ ~ U.ru.gua.[ ʝ]o ‘Uruguayan’, etc., as well as the familiar glide ~ consonant contrasts in pairs like cre.c[j]en.do ‘growing’ vs. cre.[ ʝ]en.do ‘believing’ or pro.me.t[j]en.do ‘promising’ vs. pro.ve.[ ʝ]en.do ‘providing’.14 Glide consonantalization (GC) in most dialects does not apply at the phrase level, where it is counterfed by postlexical glide-to-vowel resyllabification (Harris and Kaisse 1999: 155), as illustrated in (11) with resyllabification of the second member of a falling diphthong to the initial syllable of the following word (11a), or the merger of conjunctions y ‘and’, u ‘or’ with the word-initial vowel (11b). Phrasal resyllabification results in minimal pairs such as hay una [á.jú.na] in (11a), not *[a.ʝú.na] (postlexical resyllabification, no GC), vs. ayuna [a.ʝú.na] ‘he fasts/is fasting’ (lexical syllabification; GC has applied). (11) a. hay una ley antigua

/ái # úna/ /léi # antíɡua/

[á.jú.na] [lé.jan̪.tí.ɣwa]

‘there is one-FEM .’ ‘old law’

b. padre y amigo /pádɾe # i # amíɡo/ [pá.ðɾe.ja.mí.ɣo] ‘father and friend’ /úno # u # ótɾo/ [ú.no.wótɾo] ‘one or the other’ uno u otro Because phrase level-glide-to-vowel resyllabification (11) replicates exactly the process of consonant-to-vowel resyllabification in the same environment (cf. las aves /las # ábes/ → [la.sá.βes] ‘the birds’), one is led to conclude that the syllabic readjustment observed in (11) naturally arises as a way of providing an onset to the otherwise onsetless following syllable. In short, such a distinctly consonantal behavior of word-final prevocalic glides follows naturally from a hypothesis of resyllabification to the syllable onset, not to a nuclear position. In this sense, glide-to-vowel resyllabification across word boundaries in (11a) bears a striking formal resemblance to syllable merger (11b), in that they are 13 For a comparable behavior of prevocalic glides in Pasiego, see Kaisse, this volume. 14 In order to account for GC, earlier accounts propose that prevocalic glides not preceded by a consonant are initially attached to the nucleus by core syllabification, and subsequently moved to the onset, where they undergo consonantalization (Harris 1983: 57, Hualde 1991).

154

Fernando Martínez-Gil

driven by one and the same principle: that of avoiding onsetless syllables. Clearly, a similar insight cannot be captured if prevocalic glides resulting from syllable merger are analyzed as members of a complex nucleus.15 A final piece of evidence in support of the onset affiliation of word-initial glides in Spanish phrasal phonology derives from the fact that they behave like consonants. As is well-known, coda nasals in Spanish assimilate in place of articulation to a following consonant (cf. u[n] río ‘a river’ vs. u[ň] chico ‘a boy’). In varieties that lack GC, nasals also undergo homorganic assimilation to wordinitial glides: un hielo → [uň.jé.lo] ‘an ice’, un hierro → [uň.jé.ro] ‘an iron’ by contrast, coda nasals resyllabify, but crucially fail to assimilate, to a word-initial nuclear high vowel: un imán → [u.ni.mán] ‘a magnet’ (not *[u.ňi.mán]; Hooper 1972). We have seen that prevocalic glides occupy the syllabic onset when located in absolute syllable initial position, both in lexical syllabification and at the phrase level when created by syllable merger. In addition, the assumption that prevocalic glides are attached to the onset of the merged syllable in ChSp provides a straightforward phonological account of both gliding, mid-vowel raising, and stress shift, independently of whether the glide ends up in syllable-initial position or is preceded by a tautosyllabic consonant; the same result is not accomplished under the competing hypothesis, which holds they are parsed in a complex nucleus. 15 Special emphasis is placed on this point as a response to an anonymous reviewer’s contention that glides in Spanish are never in the onset, citing Harris (1983: 61) who in fact states that a word-final glide does not resyllabify to a word-initial vowel, claiming that hay una (cf. 11a) above) is syllabified as [áj.ú.na], where the word final glide is kept in hiatus with the following word-initial vowel, as a prosodic instantiation of the word boundary. This analysis, however, is at odds with the literature on Spanish syllabification, where it is established in unequivocal terms that an intervocalic high vocoid indeed resyllabifies with the following vowel: [a.jú.na] (cf. Hooper 1972: 527, Navarro Tomás 1977: 151, D’Introno, Zamora, and Guitart 1988: 143, Hualde 1989: 825, 2005: 94–95, Harris and Kaisse 1999: 155, Quilis 1999: 182, Hidalgo Navarro and Quilis Merín 2004: 131, Monroy Casas 2004: 167, Roca 1991, 2005: 214, Real Academia Española 2011: 353, just to cite a few). One can only speculate that Harris’ reasoning for denying resyllabification in such cases is based on the premise that since syllable-initial glides are subject to GC, the failure of intervocalic glides to undergo GC postlexically must be the consequence of their failure to also undergo resyllabification. There is, however, a simpler alternative, the counterfeeding relationship between the two processes duly acknowledged in Harris and Kaisse (1999: 155, 180): intervocalic glides in Spanish, just as intervocalic consonants, are subject to resyllabification postlexically, thereby fulfilling the predominant drive to avoid onsetless syllables, but unlike what happens lexically, GC is no longer active in the phrasal domain. Oddly, the position in Harris (1983) seemingly contradicts his earlier analysis (Harris 1969: 34) in which GC is alleged to be obligatory at the phrase level, which would only be consistent with resyllabification to the onset position.

Syllable merger in Chicano Spanish: a constraint-based analysis

155

Harris and Kaisse (1999) argue that high vocoids preceded by a tautosyllabic consonant in core syllabification form a complex nucleus with the following vowel (see also Colina, Kaisse, this volume). Several arguments have been advanced in previous studies on Spanish syllable structure, related mainly to stress placement and syllable phonotactics, in favor of the nuclear affiliation of prevocalic glides, converging on the idea that they must be parsed in the rhyme, and thus are necessarily nuclear. On closer examination, however, these arguments turn out to be either open to question or entirely irrelevant for the syllabic organization at the phrasal level. Consider, first, what is arguably the most compelling (and also the most often cited) argument for assuming the nuclearity of prepeak glides. Spanish rhymes are restricted to a maximum of three segments (Harris 1983: 8–10). Accordingly, the rhyme of the initial syllable in items such as p[w]er.ta ‘door’ or pers.pi.cu.o ‘shrewd’ (underlined in these examples for clarity) is well formed, but not in hypothetical *p[w]ers.ta, because it exceeds the segmental limit. That the three-segment restriction is clearly inapplicable at the phrase level, however, is evidenced by the examples in (12), illustrating how syllable merger may combine up to six-member rhymes (see Navarro Tomás 1977: 69–72, Hualde 1994, Real Academia Española 2011: 297, 348): (12)

envidio a Eusebio mutua inspiración un elogio a Euskadi una serie a explorar

[jo̯ae̯w] [wajns] [jo̯ae̯ws] [je̯ae̯ks]

‘I envy Eusebio’ ‘mutual inspiration’ ‘a tribute to the Basque Country’ ‘a series to be explored’

A second argument in favor of the nuclearity of prevocalic glides is the alleged impossibility of antepenultimate stress when the penult contains a rising diphthong, as in hypothetical *telé.fjo.no. Harris (1983: 10–12) argues that the glide must be part of a branching rhyme, since two-segment onsets in the penult do not block antepenultimate stress (cf. te.lé.gra.fo). Translated into moraic representations, a “branching rhyme” in this context can only mean that the prevocalic glide carries syllabic weight, and that the raising diphthongs are bimoraic, as in (8c). While this analysis has been considered problematic (cf. Roca 1990), it is orthogonal to the issue of syllable merger, since, on the one hand, the merged syllable adopts a monomoraic configuration, due to temporal shortening, as we have seen earlier, and on the other, primary stress is assigned in the lexical phonology, and thus irrelevant for the phrase domain. A third argument claimed to support the nuclearity of prevocalic glides maintains that complex onsets typically exhibit co-occurrence restrictions, and yet no such restrictions can be observed between prevocalic glides and onset

156

Fernando Martínez-Gil

consonants in Spanish (Carreira 1992: 55). This observation, however, appears to be in error. Thus while sequences consisting of a palatal consonant followed by a heterorganic glide occur freely in Spanish (e.g., ch[w]é.ca ‘stump (of a tree)’, ha.bi.ch[w]e.la ‘bean’, ll[w]e.ve ‘it is raining’, ra.y[w]e.la ‘hopscotch’, po.ll[w]e.lo ‘chicken-DIM .’, cas.ta,ñ[w]e.la ‘castanet’, pa.ñ[w]e.lo ‘handkerchief’, etc.; ch = [ʧ]; ll, y = [ɟ] or [ ʝ]), there is a phonotactic restriction that disallows the combination of a palatal consonant followed by a homorganic glide. Hypothetical forms containing such a sequence are not simply inexistent in Spanish; they are systematically excluded (cf. *ch[j]e.ca, *ha.bi.ch[j]e.la, *ll[j]e.ve, *ra.y[j]e.la, *po.ll[j]e.lo, *cas.ta.ñ[j]e.la, *pa.ñ[j]e.lo, etc.). Crucially, an analogous restriction does not obtain between a palatal consonant and a following nuclear high front vowel (cf. chi.ca ‘girl’, co.chi.no ‘pig’, sal.chi.cha ‘sausage’, a.llí ‘over there’, me.lli.zo ‘twin’, si.llín ‘saddle’, en.sa.yis.ta ‘essayist’, tra.mo.yis.ta ‘trickster’, pe.que.ñi.to ‘very small’, a.ñil ‘indigo’, etc.). An adequate explanation of these facts requires invoking a phonotactic constraint that prohibits a palatal consonant and a following high vocoid from sharing place of articulation. It is evident that such a constraint is applicable exclusively to front glides parsed in the onset. A fourth argument holds that Spanish rhymes are subject to a co-occurrence restriction that prohibits any combination of a high nuclear vowel and a homorganic glide: *[ji], *[ij], *[wu], *[uw] (Harris 1983: 17). It is claimed that this restriction can only be formulated in reference to the syllable rhyme: since *[ji] sequences are malformed because the prevocalic glide is assumed to be nuclear (Colina 2011: 141–142). Thus, when the diminutive suffix -ito is added to [j]- final stems, such as ro.sa.r[j]-o ‘rosary’ or es.cri.to.r[j]-o ‘desk’, a disallowed high vocoid sequence would be created, and such conflict is avoided by the deletion of the stem-final vocoid: ro.sa.ri.to, es.cri.to.ri.to (not *ro.sa.r[jí].to, *es.cri.to.r[jí].to). Under this account, deletion follows from the assumption that the glide would be member of a (disallowed) complex nucleus (Colina 2011: 142). However, there is evidence that deletion of the prevocalic glide in such cases may be related to the presence of a preceding tautosyllabic consonant rather than to the alleged rhyme affiliation. There is a widespread misconception that there is an absolute prohibition against *[ji] sequences in Spanish. As we saw earlier, in core syllabification, a high front vocoid not preceded by a consonant is assigned to the syllable onset, where it undergoes GC. However, there are varieties of Spanish, including ChSp, that lack GC (cf. Hills 1938: 19–20, Alonso 1967: 192ff., Canfield 1976, 1981: 12, 39, 54–55, 62, 80–81, Hualde 2005: 27, 166): [j]a.mar ‘to call’, ma.[j]o ‘May’, re.[j]es ‘kings’, etc. Here, when an i-initial suffix is added to a stem that ends in a high front glide, the purportedly prohibited sequence, *[ji] surfaces in lexical syllabification without any apparent difficulty: tra.mo.[jí]s.ta (tramoy-ista) ‘trickster’, en.sa[.jí]s.ta (ensay-ista) ‘essay writer’,

Syllable merger in Chicano Spanish: a constraint-based analysis

157

a.[jí] (all-í) ‘over there’, etc. Furthermore, such [ji] sequences are also found at the phrase level in most Spanish varieties, including ChSp, by resyllabification of a word-final [j] to a word-initial /i-/: muy idiota [mu.ji.ðjó.ta] ‘very idiotic’, hay imágenes [á.ji.má.xe.nes] ‘there are images’, rey inglés [ré.jiŋ.ɡlés] ‘English king’, etc. If there is, indeed, a restriction against sequences of identical high vocoids within the rhyme, then it must be the case that the prevocalic glide in both cases is attached to the onset. In short, the correct generalization is that Spanish disallows [ji] sequences, except when the glide provides the onset to the following vowel-initial syllable in order to avoid hiatus. When preceded by a consonant, tautosyllabic sequences of identical high vocoids are prohibited by a well-known universal principle, the Obligatory Contour Principle (OCP), independent of the glide’s particular syllabic affiliation. According to a fifth argument, the nuclear affiliation of prevocalic glides is warranted by the alternations of a stressed rising diphthong and a unstressed mid vowel in many Spanish lexical items (cf. c[jé].go ‘blind’ ~ c[e].gue.ra ‘blindness’, c[wé].llo ‘neck’ ~ c[o].llar ‘necklace’, etc.). Since the diphthong is derived from a nuclear vowel, it is claimed that the first member must also be nuclear. This argument, however, is flawed; when the mid vowel is word-initial, the onglide ends up in an absolute syllable-initial position, and so it naturally undergoes GC, an unequivocal indication that it is parsed in the onset: hie.lo [ɟé.lo] ‘ice’ (cf. hel-ado ‘frozen’), hierro [ɟé.ro] ‘iron’ (cf. herr-ero ‘blacksmith’). A final argument for the nuclear status of prevocalic glides claims that they are selected for the nuclear position in hypocoristic truncation: Dan[j]él → Dáni, Anton[j]o → Toni, etc. (Colina 2012: 142). The strength of this argument, however, vanishes when we consider that formally analogous truncated names, such as Pili (< Pilar), Alvi (< Álvaro), Rosi (< Rosa), Loli (< Dolores), and many more, all contain a final -i which is absent in the corresponding base name. One is thus led to suspect that the final -i in hypocoristics like Dani, Toni, etc., is not necessarily the base’s glide; instead, as argued in Roca and Felíu (2003: 212), it can be reasonably identified as a diminutive suffix, formed by clipping the final syllable of the diminutive suffix -ito, a process presumably fostered by the popularity in the Spanish-speaking world of English nicknames, such as Danny, Tony, etc. In any event, the argument in question is clearly irrelevant for syllable merger; hypocoristic formation is a morphological operation, and thus ascribed to the lexical domain, while merger takes place at the phrase level. To sum up, prevocalic glides must be in the onset both lexically and postlexically when in absolute syllable initial position. We have established that in one additional instance, after a palatal consonant, /j/ must also be located in the onset, in order to account for the restriction on sequences of palatal consonants followed by a homorganic glide. The present proposal that syllable

158

Fernando Martínez-Gil

merger in ChSp is accomplished uniformly by attachment of V1 to the onset of the merged syllable, provides a simple and natural account of the various patterns involved: high vowel gliding and mid vowel gliding cum raising. Vowel coalescence, /a/-deletion, and mid-vowel deletion can also be shown to follow whenever such a configuration is disallowed by universal markedness principles by invoking the relevant constraints and constraint interaction; in contrast, as argued in the present section, the same results are not readily attainable under the premise that prevocalic glides become members of a complex nucleus. My primary goal in the remainder of this paper is to articulate these remarks into an OT analysis of ChSp syllable merger.

4 An OT analysis of syllable merger in Chicano Spanish In the preceding section we have discussed several facts related to syllable structure in Spanish that require a distinction between lexical vs. postlexical phonological phenomena in Spanish. It has long been known that the two levels exhibit different properties. And yet, the most challenging problems for parallel OT, i.e., one that countenances a direct input-output mapping, derive from the phenomenon of phonological opacity, commonly generated in what is known in classical generative theory as counterbleeding and counterfeeding relationships. As we have seen earlier, GC provides a typical example of an opaque interaction by counterfeeding, also known as opacity by underapplication; in most dialects it takes place obligatorily in lexical syllabification when a prevocalic glide is in syllable-initial position, but not postlexically when the structural conditions are created at the phrase level by glide-to-vowel resyllabification. In a parallel OT type of analysis, GC can be derived by subordinating the faithfulness constraint IDENT-[cons], which demands that the [-cons] input value of a high vocoid be preserved in the output, to the markedness constraint *ONSET- GLIDE , which disfavors the parsing of glides in onset position (presumably resulting from requirement that the sonority of onset segments be at least equal to that of a consonant). GC arises when the [-cons] input of a high front vocoid is mapped to consonantal [ ʝ] is sacrificed in order to satisfy the higher ranking *ONSET- GLIDE . This is essentially the approach taken in Colina (2009: 22ff.). Unfortunately, such a constraint ranking erroneously predicts GC at the phrase level. Colina resolves the problem by appealing to the subordination of both *ONSET- GLIDE and IDENT-[cons] by IDENT-PrWd[cons], an anti-allomorphy out-

Syllable merger in Chicano Spanish: a constraint-based analysis

159

put-to-output constraint which, for any given prosodic (i.e., fully derived) word, prohibits a change in the feature [cons]. Thus, in our earlier example ley antigua ‘old law’ in (11), the (wrong) output *le.[ ʝ]an.ti.gua would be ruled out since the final [j] in the prosodic word le[j] has been mapped as a consonant in the output *le[ ʝ]. A significant problem with this analysis is that faithfulness to the feature [cons] is unnecessarily duplicated; namely, the same input segment is evaluated twice by the same constraint under a different ranking: once by IDENT-[cons] in reference to the segment’s input-output mapping of the word within a phrase, and a second time by IDENT-[cons] in reference to the segment’s output realization within the word considered in isolation. The counterfeeding interaction of GC and glide-to-vowel resyllabification at the phrase level is surely resolved, but only at the cost of stipulating that satisfying input-output faithfulness to [cons] at the (prosodic) word level has priority over the same requirement at the phrase level.16 There is, however, a more pressing empirical issue: given the ranking IDENT-PrWd[cons] » IDENT-[cons], it is not clear how leyes, the plural of ley, is prevented from undergoing GC, thus yielding the wrong output *le.[j]es, which is fully-faithful to the prosodic word le[j], instead of le.[ ʝ]es, the correct form but unfaithful to the prosodic word. A slightly more complex, but seemingly analogous situation arises in a process known as liquid gliding in Cibaeño Spanish, a variety spoken in the Dominican Republic in which the liquids /ɾ, l/ are vocalized to the front glide [j] in coda position, and thus also word-finally: amo[j] (= amor) ‘love’, pa.pe[j] (= papel) ‘paper’, etc. (Guitart 1981, Harris 1983: 47–50). Dominican Spanish exhibits GC (Henríquez Ureña 1940: 138, Jiménez Sabater 1975: 108–110). Given the prosodic words amo[j], pa.pe[j], and the constraint ranking IDENT-PrWd[cons] » IDENT-[cons], vocalization takes place at the phrase level, as predicted. However, the resulting offglide ends up in syllable initial position, as a result of glide-tovowel resyllabification, and it also fails to undergo GC, hence the doubly-opaque interaction: su amor es [swa.mó.je(h)] ‘his love is’, not *[swa.mó.ʝe(h)], papel oscuro [pa.pé.jo([h]).kú.ɾo] ‘dark paper’, not *[pa.pé.ʝo([h]).kú.ɾo].17 And yet, just the 16 For a discussion of the shortcomings of output-to-output correspondence as a theory of phonological opacity, see McCarthy (2007: 44–47) and references therein. See also Harris and Kaisse (1999: 179ff.) cogent critique of the challenges one faces when attempting to account for the counterfeeding opacity that arises in the interaction of GC and glide-to-vowel resyllabification within the version of OT known as Sympathy Theory. The resyllabification of word-final glides derived by liquid vocalization to word-initial vowels in Cibaeño Spanish is corroborated by the intuitions of Prof. Junice Acosta (p.c.), herself a native speaker of this variety. In addition, I wish to thank her and Rafael Núnez-Cedeño for their input regarding the plural forms of liquid-final words in Cibaeño. 17 Coda /s/ is variably aspirated and often deleted in Dominican Spanish, hence the superscripted transcription ‘(h)’.

160

Fernando Martínez-Gil

opposite constraint ranking would be required in order to derive the plural forms amo[ɾ]e([h]), pape[l]e([h]), which are fully faithful to the input forms /amoɾ /, /papel/, instead of *amo[j]e([h]), *pape[j]e([h]), which replicate the corresponding prosodic words. This paradoxical state of affairs arises because the evaluation of input-output faithfulness to the feature [cons] at the lexical level is carried out simultaneously with the phrase level. A simpler and more satisfactory solution to the counterfeeding by underapplication type of opacity found in the interaction of GC and glide-to-vowel resyllabification is provided within the serial version of OT known as Stratal OT (e.g., Kiparsky 2000, 2008, 2010, Rubach 2003, Bermúdez-Otero 2006, 2011, forthcoming, Anttila 2006, Bermúdez-Otero and McMahon 2006, Collie 2007, among many others), a modular OT approach to the phonology-morphology interaction similar to that of Lexical Phonology (Kiparsky 1982a, 1982b, 1985, Mohanan 1986, Hargus and Kaisse 1993, and much related work). In this model, the phonological system consists of separate ordered domains (or strata), correlated with morphological categories (Stem, Word), and/or morphosyntactic ones (word vs. phrasal domain), as illustrated in (13). Each stratum is an OT grammar with its own constraint ranking, and the output of each stratum constitutes the input of the next stratum. (13)

The Stratal OT model:

In applying the model in (13) to the interaction of GC and glide-to-vowel resyllabification, we just need to assume that the dominant constraint enforcing GC in the lexical stratum is demoted to a subordinate status in the ranking that prevails in the phrase-level domain, where resyllabification takes place, and thus is no longer active (Roca 2005).

Syllable merger in Chicano Spanish: a constraint-based analysis

161

Syllable merger in ChSp and in all varieties of Spanish is a postlexical process. In Section 3 we discussed compelling empirical evidence that lexical and postlexical syllabic parsing syllabification may differ in a number of regards. In accordance with the Stratal OT model, it is assumed here that core syllabification produces the output of the lexical stratum, which thus becomes the input to the phrase-level domain, where syllable merger takes place. As we shall see in more detail later in Section 5.5 below, the potential for lexical and postlexical constraint rankings to differ, as countenanced by Stratal OT, is absolutely essential to the task of providing an adequate account of phrase-level syllable merger, while simultaneously avoiding the wrong results when the analysis is applied to core syllabification in the lexical stratum, an undesirable potential outcome in parallel OT approaches. Consider, for example, Pattern III (LOW VDEL) in ChSp. In order to account for the deletion of /a/ before another vowel, both Baković (2006) and Colina (2009), propose a ranking in which the markedness constraint NO -CODA disfavoring closed syllables dominates Max-IO, a faithfulness constraint that militates against segmental deletion. Both authors assume Spanish offglides are in the coda, and thus violate NO -CODA . Other potential candidates are ruled out by undominated constraints irrelevant for the present discussion. Given an example such as habla inglés /ábla # inɡlés/ → [á.βliŋ.ɡlés] ‘he speaks English’, /a/-deletion follows from the proposed constraint ranking NO -CODA » Max-IO, which disfavors the [a]-preserving outcome [. . .aj. . .] in favor of its /a/deleting counterpart [. . .i. . .].18 A seemingly intractable problem, however, arises immediately. Since Spanish freely allows closed syllables, the subordination of Max-IO to NO -CODA in our particular example amounts to the evidently wrong empirical claim that Spanish disallows syllables headed by /a/ and closed by an offglide, thereby predicting that, say, caimán /kaj.mán/ ‘alligator’, or cauto /káuto/ ‘cautious’ would be realized as *[ka.mán], *[ká.to], instead of the correct output forms [kaj.mán], [káw.to]. Given the constraint ranking in question, it is not clear what other (non ad hoc) constraint(s) could be brought to bear in order to prevent such an undesirable outcome. Similar considerations could be adduced, mutatis mutandis, for Pattern II, involving the deletion of a mid vowel followed by a homorganic high vowel (cf. reina [réj.na] ‘queen’, not *[rína], peine [péj.ne] ‘comb’, not *[pí.ne], and so on). 18 The ranking NO -CODA » Max-IO is explicitly stated in Colina (2009: 63). As for Baković (2006), this ranking can only be inferred indirectly: NO -CODA is briefly discussed at the outset (p. 66) in the evaluation of sequences of non-identical high vowels, but left out of the discussion in the remainder of the paper, including the final ranking of constraints (p. 70). However, NO -CODA in Baković’s analysis dominates 1-TO -1 (a constraint demanding a one-to-one linking of moras and segments); since 1-TO -1 has the same ranking as Max-V (= Max-IO), it follows by transitivity that NO -CODA must also dominate Max-IO.

162

Fernando Martínez-Gil

Let us turn now to the OT analysis of ChSp syllable merger proposed in this paper. As shown in (14), within the moraic theory of the syllable a resolution of the vowel hiatus (14a) created by syntactic concatenation within the postlexical stratum can be achieved by mapping the input into one of the nine logically possible configurations in (14b–j): (14) a.

b.

c.

d.

e.

f.

g.

h.

i.

j.

Namely, hiatus could be resolved by: 1) inserting an epenthetic consonant (= C) (14b); 2) delinking V1 from its associated mora and attaching it to the onset of the following syllable (14c); 3) associating V2 to the coda of the preceding syllable (14d); 4) merging the two syllables into a heavy diphthong (14e); 5) merging the two syllables into a short diphthong (14f); 6) creating a long vowel when V1 and V2 are identical (14g); 7) vowel coalescence when V1 and V2 are identical (14h); 8) V1 deletion (14i); and 9) V2 deletion (14j). Among the various options in (14) only the input-output mappings in (15) are required in order to derive the five hiatus-resolving syllable merger patterns found in ChSp (cf. (5) above). All resulting configurations in (15) directly reflect the moraic shortening of the input structure observed in syllable merger: (15)

a.

Pattern I (IDVCOAL) : se escapó /se # eskapó/ → [sés.ka.pó], ‘he fled’):

b.

Pattern II (MIDVDEL): me iría /me # iɾía/ → [mi.ɾí.a] ‘I would go’:

Syllable merger in Chicano Spanish: a constraint-based analysis

163

c.

Pattern III (LOW VDEL): habla inglés /ábla # inɡlés/ → [á.βliŋ. ɡlés] ‘he speaks English’

d.

Pattern IV (HI VDES ): mi esposa /mi # espósa/ → [mjes.pó.sa] ‘my wife’

e.

Pattern V (MIDVRAIS ): me usó /me # usó/ → [mju.só] ‘he used me’

Phonological derivations in OT generally involve a conflict between two types of universal constraints: markedness constraints, prohibiting some structural configuration, and faithfulness constraints, requiring that the output be a faithful rendering of the input. The set of universal constraints assumed to be at work in ChSp syllable merger are given in (16). (16) A. Markedness constraints a. ONSET: syllables must have onsets (McCarthy 1993, Prince and Smolensky 2004). b. NO DIPHTHONGS (NO -DIPH ): diphthongs, whether short (i.e., monomoraic) or long (i.e., bimoraic) are disallowed (cf. Rosenthall 1997a, 1997b).19 c. OBLIGATORY CONTOUR PRINCIPLE (OCP): sequences of identical tautosyllabic segments are prohibited (cf. McCarthy 1986, 2003, Yip 1998, McCarthy and Prince 1995a, 1999, Myers 1997, Fukazawa 1999, Kager 1999: 56, among others). d. GLIDE -HI : nonmoraic vocoids must be [+high] (Casali 1997, Rosenthall 1997a, 1997b, Roca 2005). 19 In Rosenthall (1997a: 22–24) NO -DIPH applies to heavy diphthongs only; short diphthongs are disfavored by the constraint *BRANCH -μ. For simplicity, both constraints are subsumed here under the label NO -DIPH .

164

Fernando Martínez-Gil

e. NO LONG VOWELS (NO -LV): a single vowel melody may not be linked to two moras (Rosenthall 1997a, 1997b). B. Faithfulness constraints (see McCarthy and Prince 1995a, 1999, McCarthy 2002, 2008) f. ANCHOR- L (EFT ): an input element at left edge of a morpheme/word must have a correspondent at the left edge of the morpheme/word in the output (for anchoring constraints in OT, see, among others, Cole and Kisseberth 1994, McCarthy and Prince 1995a, 1995b, 1999, Bye and de Lacy 2000, Nelson 2003). g. DEP -IO: an output segment must have an input correspondent (i.e., ‘no epenthesis’). h. MAX-IO: an input segment must have an output correspondent (i.e., ‘no deletion’). i. MAX-μ: an input mora must have a correspondent in the output. j. IDENT- [low]: the specification for the feature [low] in an input segment must have a correspondent in the output. k. IDENT- [back]: the specification for the feature [back] in an input segment must have a correspondent in the output. l. IDENT- [high]: the specification for the feature [high] in an input segment must have a correspondent in the output. m. UNIFORMITY (UNIF ): no segment in the output has multiple correspondents in the input (i.e., ‘avoid vowel coalescence’; Lamontagne and Rice 1995, McCarthy and Prince 1995a, Pater 1999, Gnanadesikan 2004, McCarthy 2008). Subset A in (16) contains the relevant structural markedness constraints. ONSET (16a) requires that syllable have onsets; a sufficiently high ranking of this constraint ensures that the input (14a) will not be chosen as the optimal candidate. NO -DIPH (16b) rules out a syllable merger that creates a complex syllable nucleus consisting of either a bimoraic or a monomoraic vocoid sequence, such as (14e) and (14f), respectively. Note that (14c–d) do not violate NO -DIPH ; since the initial or final member in each case is linked directly to the syllable node, not to a mora, they are members of onset and coda, respectively, and therefore they do not meet the narrow definition of diphthong assumed here. The OCP (16c) prohibits adjacent identical tautosyllabic units; here it is invoked to rule out structures such as (14b–f) when V1 and V2 are identical. GLIDE -HI requires (14d) that non-moraic vocoids (glides) be high, thus ruling out candidates such as (14c–d) in which the onset/coda member is a mid or a low glide. Finally, NO LONG VOWELS (16e) disallows long vowels, as in (1g).

Syllable merger in Chicano Spanish: a constraint-based analysis

165

Subset B of universal constraints (16f–m) belong to the family of faithfulness constraints; they demand faithfulness of features and segments between input and output forms. ANCHOR-LEFT (16f) requires full input-output faithfulness of any element located at the left edge of a morpheme/word. In the case of syllable merger, this constraint bars any operation that alters the moraic or segmental content of a word-initial vowel, including deletion, thus excluding candidates (14d) (where V2 has lost its input mora), and (14j) (due to V2 deletion). It can be argued that the scope of this constraint is more general than morphological entities such as morphemes or words, and that it also enforces the identity of the initial element of any of the three major prosodic categories: prosodic word, metrical foot, and syllable. In our analysis of word-internal syllable merger in ChSp in Section 5.4 below, we show how ANCHOR-LEFT is used to preserve the integrity of the vowel that heads the word’s metrical foot. DEP-IO (16g) bans the insertion of an epenthetic segment as a means to breaking hiatus, as in candidate (14b). MAX-IO (16h) penalizes the deletion of any input segment, thereby excluding (14i–j). MAX-μ (16i) requires that the input’s moraic structure be preserved in the output, and therefore it is violated by all monomoraic candidates in (14), namely, (14c–d), (14f), and (14h–j). This constraint must obviously rank low in the constraint hierarchy because it is violated pervasively in ChSp syllable merger. Finally, the feature identity constraints IDENT-[low] (16j), IDENT[back] (16k), and IDENT-[high] (16l) disfavor any alteration of the input values for each of these three major contrastive vowel features in the corresponding output. Finally, UNIF (16m) is an anti-coalescence constraint that prevents multiple input segments from having a single output correspondent. It is crucial to bear in mind that vowel coalescence does not involve deletion, and hence no violation of MAX-IO is incurred. The proposed ranking of all the constraints in (16) is shown in (17). (17)

Constraint Ranking: ONSET, DEP-IO, ANCHOR-L, NO -DIPH , GLIDE -HI , NO -LV, IDENT-[low] » OCP, IDENT- [back] » MAX-IO » IDENT- [high], MAX- μ, UNIF

Observe that with the single exception of the OCP, the remaining markedness constraints are never violated in the output of syllable merger, and thus must be undominated (the reasons for assigning a lower ranking to the OCP with regards to the other markedness constraints will be discussed later in Section 5.3). It must also be the case that faithfulness constraints DEP-IO and ANCHOR-LEFT are undominated, since hiatus is never resolved by either inserting an epenthetic consonant, nor by deletion, desyllabification (i.e., demorification), or alteration of the feature content of V2 . On the other hand, IDENT-[low] and IDENT-[back] must dominate MAX- IO since the option of deleting a low vowel is preferred over the logically possible alternative of raising it to mid or

166

Fernando Martínez-Gil

high. IDENT-[high] must be outranked by MAX- IO since the option of raising a mid vowel followed by a heterorganic vowel is preferred to deletion (cf. me usó ‘he used me’ /me # usó/ → [mju.só], not *[mu.só]). Finally, MAX-μ must also be low ranked, since desyllabification of the first vowel is pervasive in syllable merger. To conclude this section, a complete derivation for each of the five patterns of syllable merger in ChSp is given in tableaux (18)–(22) below, illustrating the evaluation of a set of likely candidates for each pattern of syllable merger.

4.1 Pattern I: Identical Vowel Coalescence Tableau (18) exemplifies the emergence of Pattern I for 14 likely candidates. As illustrated in (18), given an input containing a sequence of identical vowels, the proposed ranking selects a single vowel (18a), the outcome of coalescence, as the optimal output. (18) Pattern I (VCOAL ; e.g., se escapó /se # eskapó/ → [sés.ka.pó] ‘he escaped’): Input: /. . . e1 # e2 . . ./

Syllable merger in Chicano Spanish: a constraint-based analysis

167

In particular, candidates (18c–n) fatally violate one or more of the higher ranking constraints, including the fully-faithful hiatus configuration (18m).20 The two remaining candidates violate MAX-μ, and (18b) fares better than its competitor (18a) on the lower ranked UNIF. The decision is thus left to the higher-ranked constraint IDENT-[high]; candidate (18b) violates this constraint, and therefore (18a) emerges as the winner. In short, the optimal output form for a sequence of two identical mid vowels, will always be a single mid vowel, because it preserves the feature identity of the corresponding input, while an alternative in which the first member is raised to high will lose out since it contravenes feature identity, as illustrated in the evaluation of the two competing candidates in (18). It is crucial to recall that the coalescence of two identical vowels by syllable merger does not entail the deletion of an underlying vowel, but rather a moraic shortening, an assumption fully consistent with our analysis of the other syllable merger patterns in ChSp. Accordingly, the winning candidate (18a) does not incur a violation of either MAX-IO or ANCHOR-L. This point is crucial, since given the proposed constraints and constraint ranking, the winner would be candidate (18b), thereby incorrectly generating Pattern V (MIDVRAIS ), which otherwise is indeed the systematic outcome of sequences consisting of a mid vowel followed by a heterorganic vowel. The solution presented in (18) is precisely an OT answer to the rule-ordering paradox first pointed out by Clements and Keyser’s (1983: 89ff) in their analysis of ChSp syllable merger regarding the interaction of mid-vowel raising and identical vowel coalescence. Instructively, such paradoxes typically arise as extraneous artifacts from the counterfeeding type of opacity created in the interaction of the two processes, namely, given the otherwise general process of mid-vowel gliding and raising in syllable merger, it seemingly becomes paradoxical for V1 to forego this process, and instead undergo coalescence when followed by an identical V2 .

4.2 Pattern II: Mid-Vowel Deletion The derivation of Pattern II, involving the deletion of a mid vowel followed by a homorganic high vowel, is illustrated in tableau (19).

20 In order to avoid unnecessary clutter, excluded from (18) and from subsequent tableaux are those constraints in (16) deemed irrelevant to the evaluation of the likely candidates considered in each tableau. Included with each candidate, for ease of reference, is its corresponding syllabic configuration in (14).

168

Fernando Martínez-Gil

(19) Pattern II (MIDVDEL ; e.g., me iría /me # iɾía/ → [mi. ɾí.a] ‘I would go’): Input: /. . .e # i. . ./

Here again, candidates (19c–k) are excluded because each of them violates one of the highest-ranked constraints. Of the two remaining candidates in competition, the domination of OCP over MAX-IO is critical in selecting (19a) as the winner over (19b), since the latter violates the OCP, while the former satisfies it, although it does so by the deletion of V1 .

4.3 Pattern III: Low-Vowel Deletion The emergence of Pattern III, involving the deletion of /a/ when followed by another (non-identical) vowel is illustrated in tableau (20).

Syllable merger in Chicano Spanish: a constraint-based analysis

169

(20) Pattern III (LOW VDEL ; e.g., habla inglés /abla # inɡlés/ → [á.βliŋ.ɡlés] ‘he speaks English’): Input: /. . .a # i. . ./

In (20), candidates (20c–g) all fail to satisfy one or more of the undominated constraints. The constraint ranking selects (20a), exhibiting /a/-deletion, as the optimal candidate over (20b), one in which the low vowel becomes a front glide and is concomitantly raised to high, creating an identical vocoid sequence, since the latter violates the high-ranking IDENT-[low]. Here, deletion of wordfinal /a/ here is preferred to subjecting this vowel to multiple feature changes.

4.4 Pattern IV: High-Vowel Desyllabification Candidate evaluation in the emergence of Pattern IV, high vowel gliding, is shown in (21).

170 (21)

Fernando Martínez-Gil

Pattern IV (HI VDES ; e.g., mi esposa /mi # espósa/ → [mjes.pó.sa] ‘my wife’): Input: /. . .i # e. . ./

As in earlier tableaux, candidates (21c–l) fall out of contention because they all fail to comply with one or more of the highest-ranked constraints. The segmentally fully-faithful candidate (21a) wins out over its remaining competitor (21b) because in the latter V1 has undergone deletion, thereby resulting in a fatal violation of MAX-IO.

4.5 Pattern V: Mid-Vowel Raising Finally, tableau (22) illustrates how the proposed constraints interact to generate Pattern V, involving the desyllabification and concomitant raising to high of a mid vowel followed by a non-identical heterorganic vowel.

Syllable merger in Chicano Spanish: a constraint-based analysis

(22)

171

Pattern V (MIDVRAIS ; e.g., me usó /me # usó/ → [mju.só] ‘he used me’): Input: /. . .e # u. . ./

In (22), candidates (22c–k) are immediately left out of contention because they fatally violate the highest-ranked constraints. Of the two remaining candidates, (22a) fares better than (22b) with regards to input-output faithfulness, and thus emerges as optimal, because it satisfies higher ranked MAX-IO, whereas candidate (22b) infringes this constraint by way of V1 deletion. To complete the present analysis, something has to be said about the interaction of No-Diph and lexically-created formed diphthongs in the postlexical stratum. It is thus critical that NO -DIPH , which actively thwarts the formation of diphthongs in the process of syllable merger, not be allowed to affect lexical diphthongs in their passage to the postlexical phonology. This outcome can be easily achieved by assuming that an undominated constraint, IDENT-DIPH , protects the identity of lexical diphthongs created by core syllabification at the lexical level. Recall that in Stratal OT the output of the lexical stratum becomes the input of the postlexical one. Under the postlexical ranking IDENT-DIPH » ONSET, NO -DIPH , lexical hiatus will still be resolved by syllable merger in the manner proposed in this section, leaving lexical diphthongs unaltered.

172

Fernando Martínez-Gil

5 Remaining issues In the rest of this paper we address several issues related to syllable merger in ChSp. They include: a) stress shift; b) the affiliation of prevocalic glides preceded by a complex onset; c) syllable merger in vowel sequences of more than two members; d) syllable merger hiatus resolution within words; and e) and competing OT analyses of ChSp syllable merger.

5.1 Stress shift Stress-shift does not present a challenge for the analysis proposed here (with the caveat mentioned earlier in fn. 4). It should be recalled that given a /. . .V1 # V2. . ./ sequence, where V1 ≠ V2 , and V1 is stressed (cf. (3) above), merger is carried out by either desyllabification of a non-low vowel or by deletion of V1 . As a result, stress is shifted onto V2 , as illustrated in the two representative input-output mappings in (23): (23) a. así acabó

b. está usado

/así # akabó/ → [a.sjá.ka.βó] ‘it finished this way’

/está # usádo/ → [es.tú.sá.ðó] ‘it is used’

Crucially, stress is invariably preserved by the nuclear vowel of the merged syllable. The transfer of stress from the first vowel to the second in (23a), as mentioned earlier, is the inevitable consequence of the fact that a glide is not a legitimate stress bearer in Spanish. In (23b), deletion of V1 does not result in a concomitant loss of stress; instead, stress is reassociated to the closest stress bearer, V2 , a distinct trademark of autosegmental stability. This points to an undominated constraint, IDENT-stress, that enforces faithfulness to lexical stress in the postlexical stratum. On the other hand, whatever constraint(s) militate

Syllable merger in Chicano Spanish: a constraint-based analysis

173

against stress-shift must obviously be ranked low enough in the constraint hierarchy of ChSp., since the resolution of hiatus by syllable merger compels the violation of such constraint(s).

5.2 Syllable merger and complex onsets There is general agreement that in core syllabification Spanish complex onsets are limited to a maximum of two members: an oral stop or /f/ plus a liquid /ɾ, l/ (Harris 1983, 1989a, 1989b, Hualde 1991). If this is indeed the case, then in one specific instance of syllable merger, namely, when V1 is preceded by a biconsonantal onset (cf. (cf. cl[j]en.te ‘client’, pr[w]e.ba ‘proof, test’, etc.), the onglide must be parsed in the nucleus, thereby avoiding an infringement of the onset binarity constraint. This property would be in apparent conflict with the proposal adopted in this paper that all prevocalic glides derived in syllable merger uniformly attach to the onset of the merged syllable: an onset glide preceded by a complex onset entail a ternary structure. However, the relaxation of lexical constraints in fast connected registers in phrase phonology is a welldocumented event. In fact, we saw earlier (cf. the examples in (12)), the lexical three-segment limit on rhymes in Spanish may be freely violated in syllable merger. As illustrated in (24) the merger of a mid vowel preceded by a complex onset in ChSp (24a) exhibits precisely the same behavior as a mid vowel preceded by a one-consonant onset (24b) (examples from (1e) above). In both instances the mid vowel is subject to gliding and raising: (24)

a.

libre unión otro idioma

[lí.βɾju.njón] [ó.tɾwi.ðjó.ma]

‘free union’ ‘another language’

b.

me usó lo hicieron

[mju.só] [lwi.sjé.ɾon]

‘he used me’ ‘they did it’

The gliding and raising of mid vowels preceded by a biconsonantal onset in (24b) follows in a straightforward manner under the constraints and constraint interaction proposed here if we assume that they attach, as a third member, to the complex onset of the merged syllable, but would seem puzzling indeed if onglides were to be parsed as member of a complex nucleus, especially in input vowel sequences of falling sonority, as in (24).

5.3 Syllable merger in three-vowel sequences I have been unable to find in the available literature on ChSp phonology any substantial references to the behavior of vowel sequences longer than two members. As far as I have been able to determine from intuitions obtained informally

174

Fernando Martínez-Gil

from several native speakers of ChSp, in examples containing three-vowel sequences, such as those shown in (25), the realization of merger is entirely consistent with the general patterns observed for two-vowel sequences, and thus can be readily accommodated into the analysis proposed in this paper: (25) a. A word-final vowel followed by a falling diphthong that begins with an identical word-initial vowel in (Pattern I): está ausente /está # ausente/ [es.táw.sén̪.te] ‘he is absent’ gente europea /xénte # euɾopéa/ [xén̪.tew.ɾo.pé.a] ‘European people’ lo oiría /lo # oiɾía/ [loj.ɾí.a] ‘I/he would hear it’ b. A word-final vowel followed by a word-initial falling diphthong: para Europa casi aumenta tu euforia este aire otro autor

/paɾa # európa/ /kasi # auménta/ /tu # eufóɾia/ /este # áiɾe/ /ótɾo # autoɾ/

[pa.ɾew.ró.pa] [ka.sjaw.mén̪.ta] [twew.fó.ɾja] [es.tjáj.ɾe] [ó.tɾwaw.toɾ]

‘for Europe’ (Pattern III) ‘it almost increases’ (Pattern IV) ‘your euphoria’ (Pattern IV) ‘this air’ (Pattern V) ‘another author’ (Pattern V)

c. A high vocoid flanked by two nuclear vowels, including a word-final falling diphthong followed by a word-initial (Pattern IV; similar to examples in (11) above): hay alguna voy arriba uno y otro uno u otro aire y agua

/ai # alɡúna/ /bói # aríba/ /úno # i # ótɾo/ /úno # u # ótɾo/ /áiɾe # i # aɡua/

[a.jal.ɣú.na] [bó.ja.rí.βa] [ú.no.jó.tɾo] [ú.no.wó.tɾo] [áj.ɾe.já.ɣwa]

‘there is some-FEM . ’ ‘I am going upstairs’ ‘both of them’ ‘one or the other’ ‘air and water’

d. A word-final postvocalic high front glide followed by a word-initial high front vowel (glide-to-vowel resyllabification; Pattern IV): soy igual hay higuitos ley antigua

/sói # iɡwal/ /ái # iɡítos/ /léi # antíɡua/

[só.ji.ɣwál] [á.ji.ɣí.tos] [lé.jan̪.tí.ɣwa]

‘I am the same’ ‘there are figs-DIM .’ ‘old law’

Importantly, examples like soy igual in (25d) clearly confirm that the OCP constraint must be dominated by the other markedness constraints in (16). They also provide unequivocal empirical evidence that OCP violations are not completely ruled out in Spanish at the phrase level. As discussed earlier, they arise when an intervocalic high front vocoid is resyllabified to a word-initial high front nuclear vowel, suggesting that the OCP constraint must be subordinated to at least ONSET and ANCHOR-L, as illustrated in tableau (26) for soy igual: /sói # iɡuál/ → [só.ji.ɣwál] ‘I am the same’. Here, an OCP violation in the winning candidate is allowed in exchange for the avoidance of an ONSET violation:

Syllable merger in Chicano Spanish: a constraint-based analysis

175

(26) Input /. . .Vj # i. . ./

5.4 Word-internal syllable merger The available sources on ChSp do not provide a detailed account on the postlexical application of syllable merger within words (known also traditionally as syneresis). In addition, descriptive statements on this topic available in the relevant literature on Southwest Spanish are sporadic and circumstantial (cf. Espinosa 1930, Hills 1938, Mattluck 1995). However, three merger patterns found across words, namely I, IV, and V, are also well attested within words.21 Pattern I (coalescence of two identical vowels) takes place within words in ChSp, given the appropriate casual speech register. Coalescence is also documented in two common verbs, creer ‘to believe’ and leer ‘to read’: cr[é]n (creen) ‘they believe’, cr[é]mos (creemos) ‘we believe’, l[é] (lee) ‘he reads’, l[é]n (leen) ‘they read’, etc.; here coalescence is much less common than in other varieties of Spanish, presumably to avoid loss of morphological exponence:, V1 is root-final and V2 the theme vowel. Pattern IV, gliding of an unstressed high vocoid followed by another vowel, is systematic in most American Spanish dialects: v[j]a.je ‘journey’, cl[j]en.te ‘client’, gr[j]e.ta ‘crevice’ pi.d[jó] ‘he asked for’, c[w]e.ro ‘leather’, cr[w]el 21 In dialectological studies of Southwest Spanish, such as those just mentioned, Pattern II, is attested in a handful of lexical items, all of them are verb forms in which stem-final /e/ is followed by the stressed theme-vowel /i/: r[í]r (re.ír) ‘to laugh’, fr[í]r ( fre.ír) ‘to fry’, engr[í]do (en.gre.í.do) ‘vain’. Pattern III, /a/-deletion, is also documented in a few words word-initially: [ó]ra (a.ho.ra) ‘now’, [o]gado (a.ho.ga.do) ‘drowned’, [o]rcar (a.hor.car) ‘to hang’. In the absence of sufficiently reliable data, one can only speculate that these two merger patterns may be lexicalized. It is worth noting that Pattern III is not attested at all in word-internal position; ca.er ‘to fall’, tra.er ‘to bring’, ba.hí.a ‘bay’, etc., do not undergo merger and become *[kéɾ], *[tréɾ], *[bía], respectively; here, hiatus is apparently the only available option (Clements and Keyser 1983: 92, n. 23). It is likely that /a/-deletion inside words is precluded by constraints demanding that the morphemic integrity of words be preserved, but I will not speculate on this matter any further.

176

Fernando Martínez-Gil

‘cruel’, c[w]ota ‘quota, share’, etc. Finally, Pattern V, desyllabification and raising of a mid vowel before a non-identical heterorganic vowel, is also found inside words in colloquial styles in ChSp, as is also the case in many non-standard American Spanish varieties: r[j]al (re.al) ‘real’, l[j]al (le.al) ‘loyal’, des[j]ar (dese.ar) ‘to desire’, pel[j]ar (pe.le.ar) ‘to fight’, p[j]ón (pe.ón) ‘laborer’, l[j]ón (león) ‘lion’, alm[w]á.da (almohada) ‘pillow’, t[w]a.lla (toalla) ‘towel’, p[w]e.ma (poema) ‘poem’, p[j]or (peor) ‘worse’, and many more. In our analysis of postlexical merger, ANCHOR-L is pivotal in precluding any alteration in the structural make-up, whether featural or prosodic, of a wordinitial vowel. The same analysis can be applied to the Pattern V type of wordinternal merger if we assume that this constraint covers not only word-initial but also the initial vowel in the metrical foot (a trochee in Spanish). It should be noted that word-internal hiatus involving a potential input to the Pattern V type of merger quite generally involves a mid vowel followed by a stressed vowel, and desyllabification invariably affects V1 , never V2: p[w]e.ma, p[j]or, etc., not *po[j].ma, *pe[w]r. A simplified derivation of postlexical word-internal merger of Pattern V is illustrated in (27), which includes some of the relevant constraints for several likely candidates (F = foot):22 (27) Input: / po.(é.ma)F /

5.5 A comparison with two competing analyses of syllable merger in ChSp At several points in this paper we have alluded to the two parallel OT approaches to syllable merger in ChSp proposed by Baković (2006) and Colina (2009). The

22 An entirely analogous result in (27) can be achieved if instead of ANCHOR-L one appeals to the positional faithfulness constraint, IDENT-σ́(F), which requires feature faithfulness to the stressed syllable (Beckman 1999).

Syllable merger in Chicano Spanish: a constraint-based analysis

177

analysis advanced in this paper differs from the proposals in those two studies in several significant ways. In this section, I will focus on three major differences, and point out the advantages of the present approach over the two competing analyses, relating to: 1) the particular assumptions on the affiliation of glides in Spanish syllable structure; 2) the nature and role of the constraints invoked to explain the asymmetrical behavior of the vowel sequence involved in merger, compelling full faithfulness to V2 to the detriment of the identity of V1; and 3) the ability to attain an adequately empirical coverage of the facts related to ChSp syllable merger. First, as stated earlier, both Baković and Colina assume that Spanish onglides are members of a complex nucleus, thus forming a structure we have earlier characterized as a short diphthong (cf. (14f) above). Both also concur on the premise that offglides are in coda position, and therefore violate a NO -CODA constraint. They differ, however, on the particular syllabic position of postvocalic glides. Following arguments suggesting that Spanish lexical diphthongs bear syllabic weight, Baković assumes that offglides are moraic coda units, as in (14e), just as coda consonants; by contrast, Colina considers them non-moraic coda elements, directly attached to the syllable node, as in (14d). As discussed earlier in Section 3, there is agreement that the syllabic affiliation of prevocalic glides to the nucleus or the onset depends on the contextual presence vs. absence of a tautosyllabic consonant. It was also suggested that most of the arguments in favor of the lexical nuclearity of onglides are either problematic or outright irrelevant for the postlexical stratum. In addition, one compelling argument was introduced suggesting the onset position of [j] when preceded by a palatal consonant. It is possible that the seemingly inflexible position adopted by both Baković and Colina on the syllabic affiliation of glides as an established fact beyond dispute is ultimately responsible for the drastically reduced number of candidates subject to evaluation, a methodological choice in evident conflict with the OT tenet of the Richness of the Base. For instance, Baković’s evaluation of the coalescence of identical mid vowels (Pattern I) is limited to 4 candidates (2006: 68), while the same pattern in Colina’s evaluation of identical high-vowel coalescence (2009: 62) only contains six tokens. As illustrated by the fourteen configurations evaluated in tableau (18), a substantial number of potential likely candidates in both studies are unfortunately left unaccounted for. Particularly significant in both Bakovic and Colina’s candidate evaluation is the absence of tokens comprising short falling diphthongs (8e), presumably on the premise that they do not exist in Spanish. I refrain from speculating here how these authors would rule out such a configuration. Colina’s account does contains a *COMPLEX N constraint that bars complex nuclei, but in order to allow short raising diphthongs to become optimal in patterns IV and V, this constraint is given the

178

Fernando Martínez-Gil

lowest rank in the hierarchy, and so it would be entirely unsuitable for purposes of ruling out candidates containing short falling diphthongs. The following example cogently illustrates the risk of precluding potential candidates. In Baković, Pattern II is analyzed as coalescence, under incomplete identity, of a mid vowel followed by a homorganic high vowel, so that an input /. . .e1 # i2. . ./ is mapped as the optimal output [. . .i1,2. . .] (2006: 70). The problem here is that under the proposed constraints and constraint ranking, the (omitted) mapping /. . .e1 # i2. . ./ → [. . .e1,2] would effectively tie with the actual winner, and it is not clear what other constraint would select one over the other: both comply with the high-ranked constraints NO -LV, ONSET, and MAX-V, and both the violate lower-ranked constraints IDENT-HI and UNIF. A second difference between the present analysis and those of Baković and Colina concerns the choice of the particular constraint responsible for protecting the complete phonological integrity of V2 from the processes of desyllabification (gliding) and deletion to which V1 is subject in ChSp syllable merger. In this paper this property is captured by the undominated constraint ANCHOR-L. In both Baković and Colina’s analysis this task is performed by a high ranking of NO -CODA : under the assumption that postvocalic glides always occupy the coda position, NO -CODA effectively rules out falling diphthongs as the optimal outcome of syllable merger. In order to derive V1-deletion in this analysis, NO -CODA must inevitably dominate MAX-V. However, as mentioned earlier, the problem is that the ranking NO -CODA » MAX-V wrongly predicts the simplification of lexical falling diphthongs, by deletion of V2 , as shown in tableau (28) for the items ley [léj] ‘law’ (28a) and boina [bój.na] ‘beret’ (28b) (the symbol ‘M’ points at the wrong candidate selected as winner, and the sad face ‘☹’ points at the actual winner): (28) a. Input: /léi/

b. Input: /bóina/

Syllable merger in Chicano Spanish: a constraint-based analysis

179

Notice that unlike Stratal OT, the option of invoking a different lexical vs. postlexical constraint ranking is not available in a parallel OT model, nor is it possible to call upon an undominated constraint IDENT-DIPH to the rescue of lexical diphthongs at the phrase level. Obviously, the correct results in (28) can be achieved by reversing the ranking so that MAX-V dominates NO -CODA . As shown in (29), under such a ranking a candidate exhibiting /a/-deletion in syllable merger would always incorrectly lose to one in which V2 is mapped as a coda: (29) Input: /. . .a # i. . ./

Finally, it is worth pointing out two obvious advantages of the present approach over the analysis in Baković (2006) and Colina (2009) in connection with the three-vowel sequences illustrated earlier in (25c–d), not considered by these authors. First, both accounts would incorrectly exclude [. . .ji. . .] outputs soy igual /sói # iɡuál/ → [só.ji.ɣwál] ‘I am alike’ (cf. 25d), since [. . .ji. . .] sequences would fatally violate the undominated constraints NO LV (Baković 2006: 66, labeled *GEM (= no geminates) in Colina 2009: 62). And second, recall that in the present account, desyllabification proceeds by attachment of V1 to the onset of the merged syllable, thereby providing an onset to the word-initial onsetless syllable, in compliance with the highly-ranked ONSET constraint. In the competing analyses V1 adjoins to the mora of V2 so that the two vowels are now gathered under a branching mora. The two competing moraic representations correspond to (14c) and (14f) above, respectively, repeated here for convenience as (30a) and (30b), and illustrated with the short diphthong [ja] generated in Pattern IV (e.g., mi amigo /mi # amíɡo/ → [mja.mí.ɣo] ‘my friend’): (30)

a.

b.

In an overwhelming majority of examples of two-vowel hiatus across word boundaries, V1 is be preceded by a consonant, which naturally becomes the onset of the merged syllable, and the constraint ONSET is thereby satisfied in

180

Fernando Martínez-Gil

both instances. It can be shown, however, that (32b) is empirically inadequate. The crucial evidence in support of this conclusion can be found in inputs comprising three-vowel sequences in which a high vocoid flanked by two non-high nuclear vowels, as in (29c–d). In glide-to-vowel resyllabification it must be the case that prevocalic glides are parsed as an onset of the merged syllable, as in (31a), as illustrated by hay alguna ‘there is some-FEM .’, which in all varieties of colloquial Spanish, including ChSp, is invariably syllabified as [a.jal.ɣú.na] (cf. my remarks and references in fn. 14), not *[aj.al.ɣú.na], which is ruled out because it fails to satisfy ONSET, and thus also falls short of resolving hiatus at all. (31)

a.

b.

In short, the essential problem for an analysis in which prevocalic glides adjoin to the nucleus of the merged syllable, as in (31b), stems from the fact the analysis proposed in Bakovic (2006) and Colina (2009) would be unable to derive as optimal the desired configuration (31a); indeed, under such an analysis glideto-vowel resyllabification is inexplicable and unmotivated, since the output that would be selected as optimal (31b) leaves hiatus unresolved.

6 Conclusion This paper presents an analysis of hiatus avoidance in ChSp Spanish in the form of five different surface patterns of syllable merger across word boundaries. Three of these patterns have been shown to also apply word-internally within the postlexical stratum. I have argued that such patterns can be derived in a straight forward manner by appealing to the interaction of a number of universal markedness and faithfulness constraints, ranked in a language-specific hierarchy at the phrase level, and framed within the serial OT model known as Stratal OT. The analysis advanced in this paper in which prevocalic glides created by syllable merger at the phrase level are parsed in the onset enjoys distinct advantages over the opposing view which contends they merge with the following vowel into a complex onset. Two competing treatments of the ChSp data within parallel OT have been shown to fall short of providing an adequate account; they also present a variety of problems, both theoretical and empirical, that do not arise

Syllable merger in Chicano Spanish: a constraint-based analysis

181

in the present analysis. It is widely acknowledged that prevocalic glides in Spanish are parsed in the syllable onset when located in absolute syllable-initial position, both in lexical syllabification and in the phrase-level, as a result of high vocoid-to-vowel resyllabification. This paper proposes a uniform onset affiliation of prevocalic glides in ChSp syllable merger, accounting for the concomitant processes of non-low vowel desyllabification and mid-vowel raising. When a number of well-documented, high-ranked markedness constraints on syllable phonotactics disallow the incorporation of the word-final vowel to the onset, alternative strategies, such as vowel coalescence and deletion, are called upon in order to resolve hiatus.

References Alba, Mathew C. 2005. Hiatus resolution between words in New Mexican Spanish: A usagebased account. Albuquerque: New Mexico: University of New Mexico dissertation. Alba, Mathew C. 2006. Accounting for variability in the production of Spanish vowel sequences. In Nuria Sagarra & Almeida Jacqueline Toribio (eds.). Selected Proceedings of the 9th Hispanic Linguistics Symposium, 273–285. Somerville, MA: Cascadilla Proceedings Project. Alba, Mathew C. 2008. Ratio frequency: insights into usage effects on phonological structure from hiatus resolution in New Mexican Spanish. Studies in Hispanic and Lusophone Linguistics 1.247–286. Alonso, Amado. 1967. La “ll” y sus alteraciones en España y América. Chapter II of Estudios lingüísticos: temas hispanoamericanos, by Amado Alonso. Madrid: Gredos. Anttila, Arto. 2006. Variation and opacity. Natural Language and Linguistic Theory 24.893–944. Baković, Eric. 2006. Hiatus resolution and incomplete identity. In Fernando Martínez-Gil & Sonia Colina (eds.), Optimality-theoretic studies in Spanish phonology, 62–73. Amsterdam & Philadelphia: John Benjamins. Beckman, Jill N. 1999. Positional faithfulness: and optimality theoretic treatment of phonological asymmetries. New York: Garland. Bermúdez-Otero, Ricardo. 2006. Morphological structure and phonological domains in Spanish denominal derivation, in Fernando Martínez-Gil & Sonia Colina (eds.) Optimality-theoretic studies in Spanish phonology, 278–311. Amsterdam & Philadelphia: John Benjamins. Bermúdez-Otero, Ricardo. 2011. Cyclicity. In Marc van Oostendorp, Colin Ewen, Elizabeth Hume & Keren Rice (eds). The Blackwell companion to phonology, vol. 4, 2019–2048. Malden, MA: Wiley-Blackwell. Bermúdez-Otero, Ricardo. (forthcoming). Stratal Optimality Theory (Oxford Studies in Theoretical Linguistics). Oxford: Oxford University Press. Bermúdez-Otero, Ricardo & April McMahon. 2006. English phonology and morphology. In Bas Aarts & April McMahon (eds.), The handbook of English linguistics, 382–410. Oxford: Blackwell. Bowen, J. Donald. 1956–1957. Sequences of vowels in Spanish. Boletín de Filología 5.5–14. Boyd-Bowman, Peter. 1960. El habla de Guanajuato. México: Imprenta Universitaria de la Universidad Nacional Autónoma de México.

182

Fernando Martínez-Gil

Bye, Patrik, and Paul de Lacy. 2000. Edge asymmetries in phonology and morphology. In Ji-Yung Kim & Masako Hirotani (eds.), Proceedings of NELS 30, 121–135. Amherst, MA: GLSA Publications. Canfield, Lincoln. 1976. Rasgos fonológicos del castellano en los Estados Unidos. Boletín de la Academia Norteamericana de la lengua española 1.17–23. Canfield, Lincoln. 1981. Spanish pronunciation in the Americas. Chicago: The University of Chicago Press. Cárdenas, Daniel. 1967. El español de Jalisco. Madrid: C.S.I.C. (RFE, anejo LXXXV). Carreira, María. 1992. The representation of Spanish diphthongs. In Christiane Laeufer & Terrell Morgan (eds.), Theoretical analyses in Romance linguistics, 19–35. Amsterdam & Philadelphia: John Benjamins. Casali, Roderic F. 1997. Vowel elision in hiatus contexts: which vowels goes? Language 73.493– 533. Casali, Roderic F. 1998. Resolving hiatus. New York: Garland Publishing Inc. Casali, Roderic F. 2011. Hiatus resolution. In Marc Oostendoorp, Colin J. Ewen, Elisabeth Hume & Keren Rice, The Blackwell companion to phonology, vol. 3. Malden, MA: Wiley-Blackwell, 1434–1460. Clements, George , and Jay Keyser. 1983. CV phonology. Cambridge, MA: MIT Press. Cole, Jennifer, & Charles Kisseberth. 1994. An optimal domains theory of harmony. Studies in the Linguistic Sciences 24.1–13. Colina, Sonia. 2009. Spanish phonology: A syllabic perspective. Washington, DC: Georgetown University Press. Colina, Sonia. 2011. Syllable structure. In José Ignacio Hualde, Antxón Olarrea & Erin O’Rourke (eds.), The handbook of Hispanic linguistics, 133–151. Oxford: Wiley-Blackwell. Collie, Sarah. 2007. English stress-preservation and stratal Optimality Theory. Edinburgh: University of Edinburgh dissertation. Dalbor, John. 1997. Spanish pronunciation (3rd ed.). Fort Worth, Texas: Holt, Rinehart & Winston. D’Introno, Francesco, Jorge M. Guitart & Juan C. Zamora. 1988. Fundamentos de lingüística hispánica. Madrid: Playor. Espinosa, Aurelio M. 1930. Estudios sobre el español de Nuevo México. Parte I: Fonética. (BDH, vol. I). Buenos Aires: Imprenta de la Universidad de Buenos Aires. Fukazawa, Huraka. 1999. Theoretical implications of OCP effects on features in Optimality Theory. Maryland: University of Maryland Dissertation. Gnanadesikan, Amalia. 2004. Markedness and faithfulness constraints in child phonology. In René Kager, Joe Pater & Wim Zonneveld (eds.), Constraints in phonological acquisition, 73–108. New York: Cambridge University Press. Goldsmith John. 1990. Autosegmental phonology and metrical phonology. Cambridge, MA: Basil Blackwell. Guitart, Jorge M. 1981. Some theoretical implications of liquid gliding in Cibaeño Dominican Spanish. In Heles Contreras & Jurgen Klausenburger (eds.), Proceedings of the tenth anniversary of the linguistic symposium on Romance languages (supplement II to Papers in Linguistics 3), 223–228. Seattle, WA: University of Washington. Hammond, Robert M. 2001. The sounds of Spanish: Analysis and application. Somerville, Massachusetts: Cascadilla Press. Hargus, Sharon & Ellen M. Kaisse. 1993. Studies in lexical phonology. San Diego: Academic Press. Harms, Robert. 1977. Phonology workbook. Austin, Texas: The University of Texas at Austin.

Syllable merger in Chicano Spanish: a constraint-based analysis

183

Harris, James W. 1969. Spanish phonology. Cambridge, MA: MIT Press. Harris, James W. 1983. Syllable structure and stress in Spanish: A nonlinear analysis. Cambridge, MA: The MIT Press. Harris, James W. 1985. Spanish diphthongization and stress: A paradox resolved. Phonology Yearbook 2.31–46. New York: Cambridge University Press. Harris, James W. 1987. Epenthesis processes in Spanish. In Carol Neidle & Rafael NúñezCedeño (eds.), Studies in Romance languages, 107–122. Dordrecht: Foris. Harris, James W. 1989a. Sonority and syllabification in Spanish. In Carl Kirschner & Janet DeCesaris (eds.), Studies in Romance linguistics, 139–153. Amsterdam & Philadelphia: John Benjamins. Harris, James W. 1989b. Our present understanding of syllable structure in Spanish. In Peter C. Bjarkman & Robert M. Hammond (eds.), American Spanish pronunciation, 151–169. Washington D.C.: Georgetown University Press. Harris, James W., and Ellen M. Kaisse. 1999. Palatal vowels, glides and obstruents in Argentinian Spanish. Phonology 16.117–190. Harris, John. 1994. English phonology. Cambridge, MA: Blackwell Publishers. Hayes, Bruce. 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20.253– 306. Hidalgo Navarro, Antonio, and Mercedes Quilis Merín. 2004. Fonética y fonología españolas. Valencia: Tirant lo Blanch. Hills, E. C. 1938. El español de Nuevo México. El español en México, los Estados Unidos y la América Central (BDH, vol. IV) (con anotaciones y estudios de Pedro Henríquez Ureña), 1– 77. Buenos Aires: Imprenta de la Universidad de Buenos Aires. Hooper, Joan B. 1972. The syllable in phonological theory. Language 48.525–540. Hualde, José I. 1989. Silabeo y estructura morfémica en español. Hispania 72.821–831. Hualde, José I. 1991. On Spanish syllabification. In Héctor Campos & Fernando Martínez-Gil (eds.), Current studies in Spanish linguistics, 475–493. Washington, DC: Georgetown University Press. Hualde, José I. 1994. La contracción silábica en español. In Violeta Demonte (ed.), Gramática del español, 629–647. México D.F.: El Colegio de México. Hualde, José I. 2005. The sounds of Spanish. New York: Cambridge University Press. Hualde, José I., Miquel Simonet & Francisco Torreira. 2007. Postlexical contraction of nonhigh vowels in Spanish. Lingua 118.1906–1925. Hutchinson, Sandra P. 1974. Spanish vowel sandhi. Papers from the parasession on natural phonology. Chicago: Chicago Linguistic Society, 184–192. Jiménez Sabater, Max A. 1975. Más datos sobre el español de la República Dominicana. Santo Domingo: Ediciones Intec. Jenkins, Devin L. 1999. Hiatus resolution in Spanish: Phonetic aspects and phonological implications from Northern New Mexico data. Albuquerque, New Mexico: University of New Mexico dissertation. Kager, Renée. 1999. Optimality Theory. New York: Cambridge University Press. Kang, Ongmi. 1999. A correspondence approach to glide formation in Korean. Korean Journal of Linguistics 24.1–15. Kawahara, Shigeto & Yurie Hara. 2009. Hiatus resolution in Hiroshima Japanese. In Muhammad Abdurrahman, Anisa Schardl & Martin Walkow (eds.), Proceedings of the north east linguistic society 38, 475–486. Amherst, MA: GLSA.

184

Fernando Martínez-Gil

Kiparsky, Paul. 1982a. Lexical phonology and morphology. In In-Seok Yang (ed.), Linguistics in the morning calm, vol. 2, 3–91. Seoul: Hanshin. Kiparsky, Paul. 198b. From cyclic phonology to lexical phonology. In Harry van der Hust & Norval Smith (eds.), The structure of phonological representations, vol 1, 131–175. Dordrecht: Foris. Kiparsky, Paul. 1985. Some consequences of lexical phonology. Phonology 2.85–138. Kiparsky, Paul. 2000. Opacity and cyclicity. The Linguistic Review 17.351–367. Kiparsky, Paul. 2008. Fenno-Swedish quantity: Contrast in stratal OT. In Bert Vaux & Andrew Nevins (eds.), Rules, constraints, and phonological phenomena, 185–219. Oxford: Oxford University Press. Kiparsky, Paul 2010. Reduplication in Stratal OT. In Linda Uyechi & Lian Hee Wee (eds.), Reality, exploration, and discovery: Pattern interaction in language and life (Festschrift for K.P. Mohanan), 125–141. Stanford, CA: CSLI Publications. Lamontagne, Greg & Kere Rice. A correspondence account of coalescence. In Jill Beckman, Laura Walsh Dickey & Suzanne Urbanczyk (eds.), Papers in Optimality Theory (University of Massachusetts Occasional Papers 18), 211–224. Amherst, MA: GLSA. Levin, Juliette. 1985. A metrical theory of syllabicity. Cambridge, MA: MIT dissertation. Marden, Charles C. 1938. La fonología del español de la ciudad de México. El español en México, los Estados Unidos y la América Central (BDH, vol. IV) (con anotaciones y estudios de P. Henríquez Ureña), 1–77, 86–187. Buenos Aires: Imprenta de la Universidad de Buenos Aires. Martínez-Gil, Fernando. 2000. La estructura prosódica y la especificación vocálica en español: El problema de la sinalefa en ciertas variedades de la lengua coloquial contemporánea. In Juana Gil (ed.), Panorama actual de la fonología español, 513–562. Madrid: Arco Libros. Mattluck, Joseph. 1951. La pronunciación del español en el Valle de México. México, D.F.: Universidad Nacional Autónoma de México. Mattluck, Joseph. 1995. Hiato, sinéresis, y sinalefa: A sociolinguistic updating. In Peggy Hashemipour, Ricardo Maldonado & Margaret Van Naerssen (eds.), Studies in language learning and Spanish linguistics in honor of Tracy D. Terrell., 280–289. New York: McGraw-Hill, Inc. McCarthy, John. 1986. OCP effects: Gemination and antigemination. Linguistic Inquiry 17.207– 263. McCarthy, John. 1993. A case of surface constraint violation. Canadian Journal of Linguistics 38.127–153. McCarthy, John. 2003. OT constraints are categorical. Phonology 20: 75–138. McCarthy, John. 2007. Hidden generalizations: Phonological opacity in Optimality Theory. London: Equinox. McCarthy, John. 2008. Doing Optimality Theory. Malden, MA: Blackwell. McCarthy John & Alan Prince. 1995a. Faithfulness and reduplicative identity. In Jill Beckman, Laura Walsh Dickey & Suzanne Urbanczyk (eds.), Papers in Optimality Theory (University of Massachusetts Occasional Papers 18), 249–384. Amherst, MA: GLSA. McCarthy, John & Alan Prince. 1995b. Prosodic morphology. In John Goldsmith (ed.), The handbook of phonological theory, 318–366. Cambridge, MA: Basil Blackwell. McCarthy, John & Alan Prince. 1999. Faithfulness and identity in prosodic morphology. In René Kager, Harry van der Hulst & Wim Zonneveld (eds.), The prosody-morphology interface, 218–309. New York: Cambridge University Press. Mohanan, K. P. 1986. The theory of lexical phonology. Dordrecht: Reidel.

Syllable merger in Chicano Spanish: a constraint-based analysis

185

Monroy-Casas, Rafael. 2004. Aspectos fonéticos de las vocales españolas (2nd ed.). Madrid: Libros en Red. Myers, Scott. 1997. OCP effects in Optimality Theory. Natural Language and Linguistic Theory 15.847–892. Navarro Tomás, Tomás. 1977. Manual de pronunciación española (19th edn.). Madrid: C.S.I.C. Nelson, Nicole E. 2003. Asymmetric anchoring. New Brunswick, NJ: University of New Brunswick dissertation. (Available at Rutgers Optimality Archive: https://rucore.libraries.rutgers.eud/ rutgers-lib/38433/pdf/1/). Pater, Joe. 1999. Austronesian nasal substitution and other NC effects. In René Kager, Joe Pater & Wim Zonneveld (eds.), Constraints in phonological acquisition, 310–343. New York: Cambridge University Press. Prince, Alan & Paul Smolensky. 2004. Optimality Theory: Constraint interaction in generative grammar. Cambridge, Massachusetts: The MIT Press. Quilis, Antonio. 1999. Tratado de fonología y fonética españolas (2nd ed.). Madrid: Gredos. Real Academia Española. 2011. Nueva gramática de la lengua española: Fonética y fonología. Madrid: Espasa Libros. Reyes, Rogelio. 1978. Studies in Chicano Spanish. Bloomington, IN: Indiana University Linguistics Club (Revised version of Doctoral Dissertation, Harvard University, 1974). Roca, Iggy. 1990. Diachrony and synchrony in word stress. Journal of Linguistics 26: 133–64. Roca, Iggy. 1991. Stress and syllables in Spanish. In Héctor Campos & Fernando Martínez-Gil (eds.), Current studies in Spanish linguistics, 599–635. Washington, DC: Georgetown University Press. Roca, Iggy. 1997. There are no “glides”, at least in Spanish: An optimality account. Probus 9.233–265. Roca, Iggy. 2005. Strata, yes; structure preservation, no. Evidence from Spanish. In Twan Geerts, Ivo van Ginneken & Haike Jacobs (eds.), Romance languages and linguistic theory. Selected Papers from ‘Going Romance’ 2003, 197–218. Amsterdam & Philadelphia: John Benjamins. Roca, Iggy. 2006. The Spanish stress window. In Fernando Martínez-Gil & Sonia Colina (eds.), Optimality-theoretic studies in Spanish phonology, 239–277. Amsterdam & Philadelphia: John Benjamins. Roca, Iggy & Elena Felíu. 2003. Morphology in truncation: The role of the Spanish desinence. In Geert Booij & Jaap van Marte (eds.), Yearbook of Morphology 2002, 187–243. Rosenthall, Samuel. 1997a. Vowel/glide alternation in a theory of constraint interaction. New York: Garland Publishing Inc. Rosenthall, Samuel. 1997b. The distribution of prevocalic vowels. Natural Language and Linguistic Theory 17.499–540. Rubach, Jerzy. 2003. Duke-of-York derivations in Polish. Linguistic Inquiry 34.601–629. Schane, Sanford. 1987. The resolution of hiatus. Papers from the 23rd annual regional meeting of the Chicago Linguistic Society (Part II), 279–290. Chicago: Chicago Linguistic Society. Senturia, Martha Baird. 1998. A Prosodic theory of hiatus resolution. San Diego, CA: University of California dissertation. Smith, Jennifer L. 2002. Onset sonority constraints and subsyllabic structure. Paper presented at the 9th International Phonology Meeting, University of Vienna, Austria. (Available at Rutgers Optimality Archive http://roa.rutgers.edu, ROA-608.) Smith, Jennifer L. 2005. Phonological augmentation in prominent positions. New York: Routledge.

186

Fernando Martínez-Gil

Steriade, Donca. 1982. Greek prosodies and the nature of syllabification. Cambridge, MA: MIT dissertation. St-Amand, Anne-Bridget. 2012. Hiatus and hiatus resolution in Québécois French. Toronto: University of Toronto dissertation. Stockwell, Robert P, and J. Donald Bowen. 1965. The sounds of Spanish. Chicago: The University of Chicago press. Yip, Moira. 1998. Identity avoidance in phonology and morphology. In Steven G. Lapointe, Diane K. Brentari & Patrick M. Farrell (eds.), Morphology and its relation to phonology and syntax, 216–246. Stanford, CA: CSLI Publications. Zamora, Juan C. & Jorge M. Guitart. 1982. Dialectología Hispanoamericana. Salamanca: Almar.

Conxita Lleó

The challenge of lexically empty onsets in first language acquisition of Spanish and German Abstract: In Spanish empty onsets are often filled by resyllabification, whereas in German resyllabification is very restricted, and a glottal stop is inserted instead. German glottal stop insertion avoids variation of form and facilitates comprehension, whereas resyllabification gives way to variation and complexity; both effects can be best accounted for by Optimality Theory. The production of lexically empty onsets by German-Spanish bilinguals from Germany is analyzed with regard to resyllabification and glottal stop insertion, and compared to the production of lexically empty onsets by Spanish and German monolinguals. Results show that in the weak (or heritage) language (i.e. Spanish) bilinguals begin to resyllabify later than monolinguals; they also apply more glottal stop insertion in Spanish than monolinguals do. It is claimed that in such language contact situations, bilingual children tend to abide by the phonology of the dominant language, if this is less marked and/or less complex than the weak language.

1 Introduction Once Generative Phonology incorporated the syllable as a crucial universal construct, several principles that had been dormant until then were discovered or confirmed. Above all, it became clear that segments were organized into syllables, in ways that allowed for insightful descriptions. Various organizations of the syllable were proposed, some still involving linearity, and others being more hierarchical. One of the hierarchical treatments was Harris’ (1983) landmark proposal for Spanish, which emphasized the crucial role of the rhyme, by showing that a sequence of five segments will constitute a well-formed syllable in Spanish, only if the rhyme has a maximum of three segments. Thus, buey [bwej] ‘ox’ is a well-formed syllable because [wej], with three segments, is a well-formed rhyme, but [bwejs] is not a well-formed syllable, because the rhyme [wejs] has four segments. His treatment of syllable structure at the lexical level fulfilled explanatory

Conxita Lleó, Universität Hamburg

188

Conxita Lleó

adequacy, as it provided an explanation for many phenomena of Spanish phonology, like /s/ aspiration, /n/ velarization or lateral and nasal depalatalization.1 However, the story of the syllable does not end at the lexicon. Beyond the lexicon, at the level of connected speech, units may be rearranged for different purposes. For instance, there is a well-known principle of Onset Maximization (see e.g. Selkirk 1982, Berg and Niemi 2000, and section 2 below), or Maximal Onset Principle (Blevins 1995: 230) which might be appealed to in Spanish, in a case where the initial syllable of a word has an empty onset, and the previous word has a word-final coda: e.g., in los hombres ‘the men’, pronounced [lo.som.bɾes]2 in Standard Peninsular Spanish. That is, the initial syllable of hombres is produced with the onset /s/, although the syllable [om] has an empty onset when it exits the lexicon, and /s/ originally enters the stage as the coda of los. This description is utterly simplified, as the Maximal Onset Principle has amply evolved, first being a Principle favoring onsets over codas, segmenting VCV into V.CV universally (Selkirk 1982, Blevins 1995). Later, a dichotomy was introduced between a Principle that attracts more material into the Onset than into the Coda (the so-called Onset Maximization proper) and a Principle that only satisfies or fills the Onset by means of associating one single C to an empty Onset position, referred to as Onset-filling in Berg and Niemi (2000), and Onset satisfaction or Obligatory Onset in Derwing (1992). Itô (1989) introduced the latter terminology, but her model of Prosodic Licensing, based on templates, is closer to Optimality Theory (see below, section 2). Independently from the label and notion preferred, the Onset Maximization Principle was originally formulated to be applied to lexical items, whereas here we are concerned with forms in connected speech. The situation in Spanish would require that we expand the Onset-filling Principle, to be applied a second time, to a sequence that includes a word boundary (#).3 Both, Harris (1983) as well as Hualde (1992) recognize the importance of this phenomenon of resyllabification, and offer solutions to be discussed in section 2.

1 Harris (1983)’s insightful analysis of the Spanish syllable inspired work in other areas of phonology. It is not accidental that the two studies on L1 phonological acquisition included in this volume focus on the syllable as the fundamental construct of prosodic structure: Núñez Cedeño deals with the acquisition of complex onsets and Lleó with empty onsets. 2 Syllable boundaries are represented by a dot, and word boundaries will be signalized, when relevant, by the pound key ‟#”. 3 As Hualde (2005: 88) explicitly states, “onsets are not maximized across word boundaries”, given that resyllabification only applies to empty onsets. That is, resyllabification in Spanish involves Onset filling, since club lindo is not resyllabified into *[klu.βlin.do], in spite of the existence of clusters like [βl] in e.g. sublime [su.βli.me].

The challenge of lexically empty onsets in first language acquisition

189

It is interesting to note that Romance languages have an equivalent of resyllabification in their Grammar, although with some differences. Let us first define the Prosodic Hierarchy and its various levels (see [1]), which will serve as reference points to the present discussion. (1)

The Prosodic Hierarchy (adapted from Nespor and Vogel 1986) Intonational Phrase Phonological Phrase Clitic Group4 Prosodic Word Foot Syllable

IPh PPh ClG PW Ft σ

In Spanish, resyllabification is obligatory within the word, e.g., in [in] + humano [umano] ‘non-human’ *[i.nu.ma.no]; as well as between a stressed (lexical item) and an unstressed word (function word), together constituting the clitic group, as for instance between a Determiner and the following Noun, as in an NP like [[los]Det [hombres]N]NP ‘the men’, obligatorily produced [lo.som.bɾes], with resyllabification, or los osos [lo.so.sos], [[los]Det [osos]N]NP ‘the bears’. Other contexts for resyllabification are two stressed words, e.g., the Phonological Phrase comprised of a Noun and a following Adjective, as in olas altas ‘high waves’, [o.la.sal.tas]. A Phonological Phrase can also consist of a Prepositional Phrase, e.g., en las alas ‘in the wings’, [en.la.sa.las], integrated by the preposition en, the article las and the noun alas. In Spanish resyllabification generally takes place between the components of an Intonational Phrase (IPh), but it is blocked between Intonational Phrases. At this level of structure, the shorter the prosodic constituents are, the higher the probability of resyllabifying is. That is, there is a higher degree of probability that the Spanish speaker resyllabifies within an IPh like Carmen ataca ‘Carmen attacks’, produced [kaɾ.me.na.ta.ka], whereas [El simpático joven] [ataca constantemente] ‘The nice youngster attacks all the time’ is generally produced with two IPhs, and resyllabification does not apply between IPhs. In Standard Peninsular Spanish, resyllabification is straightforward, in the sense that generally segments do not change their feature composition when they take over a different syllabic role. However, in the case of voiced stops, in fact /d/ is the only one appearing in the coda with a certain regularity, there is 4 Some phonologists have questioned the need for the clitic group, which according to them introduces an unnecessary prosodic level. Its function is then taken over by the Phonological Word (see e.g. Wiese 1996: 251).

190

Conxita Lleó

an important modification of the feature [cont], as in coda position it varies between [+cont], [-cont] or deletion of the segment, depending on the language variety, and as an onset it has the feature [+cont], if preceded by a vowel, because it undergoes spirantization (see Harris 1984, Lipsky 1994, among others). In Catalan, resyllabification is a general phenomenon, combined with voicing of the coda-to-onset consonant, in case it is a strident, e.g. /s/, as in the nominal constituent comprised of the plural Determiner plus Noun: els homes [əl.zɔ.məs] ‘the men’ (Hualde 1992: 488). Nonetheless, this voicing process is disappearing from certain varieties of Catalan, like the one spoken in districts of Barcelona, where the same NP is often pronounced [əl.sɔ.məs]. In Benet, Cortés and Lleó (2012) this process of devoicing is attributed to transfer from Spanish. There are other complicating factors in the realization of resyllabification in Catalan, due to different treatments of deleted codas, which may reappear or not when resyllabification applies. Thus, in Barcelona Catalan,5 pont ‘bridge’ and molt ‘much’ are generally pronounced [pɔn] and [mol], respectively. When resyllabified, they are generally produced without the final obstruent, as in pont obert ‘open bridge’ [pɔ.nu.βɛr] or molt anim ‘much courage’ [mo.la.nim], although they can also be produced [pɔn.tu.βɛrt] or [mol.ta.nim], respectively. These latter pronunciations are not the most common ones in Barcelona; they are influenced by the written language and possibly by language contact with other varieties that retain final obstruents, like the Central dialect spoken in the Balearic Islands. In French, resyllabification is instantiated in the so-called liaison, which has much in common with the Catalan phenomenon just outlined, as it also has some further difficulties related to final consonant deletion. Final consonants that are not produced in French may reappear in the liaison contexts, i.e. often when the following word begins with a vowel (Prunet 1986, Durand and Lyche 2008). For instance, word-final coda /z/ is not produced in deux [dœ] ‘two’, but it reappears in deux hommes [dœ.zɔm] ‘two men’. A word-final nasal coda turns the previous vowel into a nasal vowel, and it is not further produced, as in un [œ᷉ ] ‘a’, but un ours [œ.nur] ‘a bear’. The omission or non-production of such consonants and their reappearance in liaison contexts creates problems for word segmentation, as it has been reported for very young children, who often produce variant forms like [zur], [nur] or [ur] for ours ‘bear’, and [zɔm], [nɔm] or [ɔm] for homme ‘man’ (Chevrot, Dugua and Fayol 2009: 559). And according to Nguyen, Wauquier-Gravelines, Lancia and Tuller (2007: 3) “liaison consonants 5 There are other Catalan varieties that also delete some consonants of complex codas, but a detailed description of Catalan resyllabification is beyond the purpose of this paper.

The challenge of lexically empty onsets in first language acquisition

191

have a specific perceptual status”.6 Note that such difficulties emerge not only during L1 acquisition, but diachronically these ghost consonants also led to wrong segmentations, as the Italian word narancia (from Arabic naranj, Persan narang) lost the initial n- in the Old French orange.7 In Germanic languages, there is some resyllabification, although in general it is more restricted than in Romance languages. As Young-Scholten (1994: 204) puts it, “in German as in English, complete resyllabification does not take place, but rather syllable-final consonants may become ambisyllabic”. That is, as also proposed in Nespor and Vogel (1986: 223–225) and originally in Broselow (1984) for English, resyllabified consonants in English do not only become onsets of the following word, but have a double role, in the sense that they also keep their role as codas. Vogel (1991: 58) also notes that in English resyllabification leads to the resyllabified consonants being both, codas and onsets, as shown by a clear distinction between the groupings of two words, like those in (2): whereas in (2a) /t/ and /n/ fill onset positions, in (2b) they fill both positions, coda and onset. (2)

a.

grey towels, a name

b.

great owls, an aim

Moreover, Young-Scholten (1994: 204) argues that “in German such resyllabification is more restricted than in English, occurring only within the clitic group”, i.e., to an unstressed pronoun or clitic following the verb, as in (3), uttered with a relatively fast pronunciation, in which habe ‘have’ is reduced to hab, without the final schwa. Consequently, in German enclisis can be found, but only in fast speech or in an informal register. (3)

a.

Das habe ich [das.ha.biç] gesehen

‘I’ve seen that’

b.

Ich habe ihn [iç.ha.bi:n] schon gesehen

‘I’ve already seen him’

However, proclisis is absent in German, as a constituent comprised of an article and a noun that bears stress on the initial syllable, e.g. ein Auto ‘a car’, is

6 Traditional descriptions of French phonetics differentiate liaison from enchaînement. The latter refers to resyllabification of final consonants that are not deleted in coda position, like avec ‘with’, sept ‘seven’, lutte ‘struggle’. 7 See Harper (2001–2014). Note that the deletion of initial n- in French was produced because the consonant was equivocally interpreted as being part of the indefinite article une, given the floating status of such consonants in French (Prunet 1986). In Spanish, though, the initial n- is retained in naranja, as in Spanish the nature of these consonants is more transparent.

192

Conxita Lleó

generally produced without resyllabification, and the empty onset of the noun initial syllable is then filled with a glottal stop: [ajn.ʔaw.to], and also [ʔajn.ʔaw.to], as in German the article can also be stressed; in an informal register and/or in fast pronunciation such a constituent may also be produced without glottal stop, and without resyllabification: [ajn.aw.to] (Kohler 1977: 172–173; Kohler 1994; Wiese 1996: 248–250.) Given the variety of resyllabifications and empty onset treatments in various languages, in the present article we will examine L1 acquisition by three monolingual children, about 2- to 3-years-old, two acquiring Spanish, which is a language with generalized resyllabification throughout the entire IPh, and the third child acquiring German, with much more restricted resyllabification. We will then consider three German-Spanish bilingual children, aged between 2- and 6-years-old, who were simultaneously acquiring both languages. Finally, data from sixteen German-Spanish simultaneous bilingual children between 7- and 8-years-of-age will be considered, too. The rest of the paper is organized as follows. After closely considering the various analyses of resyllabification in adult Spanish and adult German, as well as glottal stop insertion in German, we will define the research questions. Attention will then be directed to the acquisition of word-initial lexically empty onsets by the monolingual and the bilingual children acquiring both languages simultaneously. A discussion of the acquisition facts and of their theoretical consequences for the analysis of resyllabification in Spanish will close the article, leading to the final conclusion.

2 Alternative analyses of Spanish resyllabification The first analysis of resyllabification within a model of syllabic structure is that of Harris (1983: 43–44), expressed by means of rule (4). The rule restates the surface effect of the phenomenon: a consonant in the rhyme (i.e. a coda consonant) is converted into an onset when the following word begins with a vowel. (4)

Resyllabification:

[+cons] | RHYME

→

[+cons] / ___ # V | ONSET

Ten years later Harris (1993) considered the phenomenon again, this time placing it within a model of phonology organized into four ordered strata: the first pair of strata deals with words, the first stratum being cyclic, and the second one non-cyclic; the following pair of strata deals with sequences of

The challenge of lexically empty onsets in first language acquisition

193

words, and again the third stratum is cyclic and the fourth one is non-cyclic. Although rules must respect “constituent integrity”, the rule of resyllabification is not affected by such a constraint, because “it does not incorporate unsyllabified (‘free’) segments into new constituents” (Harris 1993: 182). That is, resyllabification applies to lexicon constituency, which undergoes reorganization. In fact, the essence of the rule is the same as it used to be in 1983, except that Levin (1985)’s syllable structure is preferred over the one used earlier. Hualde (1992) questioned this formulation of the phenomenon, and reanalyzed it without appealing to any process specific to resyllabification: the CV rule (5) is applied twice, once in the lexicon and once post-lexically. In a case where a word comes out of the lexicon without an initial onset, and in connected speech it is preceded by a word ending in a consonant, this wordfinal consonant is turned into an onset, as shown in (5). (5)

CV rule: Adjoin a consonant to the left of the nucleus under the N’’ node8

Both treatments of resyllabification, Harris’ and Hualde’s can be considered similar, in the sense that both are rule-governed. Harris’ appeals to a rule that changes the status of a final consonant preceding a vowel-initial word, whereas Hualde’s assumes what he calls the universal CV rule. This rule corresponds to the Onset Maximization, or rather, the Onset-filling Principle, mentioned above, although the relationship is not made explicit. These treatments of resyllabification are unsatisfying because they only refer to the role of sub-syllabic constituents, not making explicit the fact that this phenomenon produces a mismatch between words and syllables, as one element (the coda) of a syllable is turned into an element (the onset) of another syllable that belongs to a different word. This apparently unremarkable and automatic consequence may have pervasive psycholinguistic consequences for the processing of speech, when ambiguities are involved, in L1 acquisition, and even more so in the processing of L2 (see Christophe, Dupoux, Bertoncini and Mehler 1994; Millotte, Wales and Christophe 2007). Nowadays, the theory that is best equipped to describe a phenomenon that involves such a mismatch between syllable and word is Optimality Theory.9 For instance, Downing (1998: 1) argues that “[cross-linguistically] the theory of Generalized Alignment developed within Optimality Theory provides us with a unified way of accounting for onsetless syllable exceptionality”. From a similar theoretical framework, Colina (1997) formulates Spanish resyllabification by means of the Onset constraint outranking an Align constraint. Thus, the grammar 8 The N’’ node just designates the main constituent –syllable– by means of the double bar notation, originally introduced in syntax. See Levin (1985). 9 See Féry (2004) for an overview of OT treatments of French liaison.

194

Conxita Lleó

of resyllabification in Spanish is defined by the two constraints (6) and (7), ordered hierarchically as ONSET >> ALIGN LEFT. This implies that Spanish abides by the universal preference of syllables with onsets and for that purpose it sacrifices the coincidence of the left edges of syllables and words. (See Colina 2006.) (6) ONSET: Syllables must have an Onset (7)

ALIGN LEFT: The left edge of a stem must coincide with the left edge of a syllable

Obviously, this description shares the fundamental role of Onset with the previous ones, but supersedes them by making the fundamental role of Onset explicit, and the mismatch between syllable edges and word edges visible, as shown in (8): In (8a) syllable boundaries (indicated by a dot) coincide with word boundaries (represented by the pound symbol “#”), whereas in (8b) the syllable boundary has moved, and thus it does not coincide anymore with the word boundary. This mismatch has the double positive consequence of getting rid of a coda and of filling an empty onset. The utterance of (8b) in connected speech is analyzed in (9) by means of a typical OT tableau. Although the three candidates incur one violation each, the winner is the last candidate, as it violates a constraint that is the less dominant in Spanish, namely ALIGN (PW,L,FT,L). The epenthetic [ʔ] of the first candidate violates DEP-IO, which in Spanish outranks ALIGN (PW,L,FT,L); and the second candidate violates another outranking constraint, the one against empty onsets. (8a)

Lexicon: Los.#o.sos ‘the bears’

→

(8b)

Connected speech: Lo.s#o.sos

(9) Tableau for the Spanish utterance los osos ‘the bears’ /los.#ˈo.sos/

ONSET

[los.#ˈʔo.sos] [los.#ˈo.sos] ☞ [lo.ˈs#o.sos]

DEP-IO

ALIGN (PW,L,FT,L)

* *! *

Note that (11) contains the same constraints as (9), but ordered differently: The ALIGN constraint outranks DEP, and this leads to prefer ʔ-epenthesis over resyllabification. In German, on the other hand, syllables and words generally retain their boundaries without altering their position, as in (10): both in (10a) and (10b) syllable and word boundaries coincide when words are selected from

The challenge of lexically empty onsets in first language acquisition

195

the lexicon, but also in connected speech. The only difference between (10a) and (10b) is that the latter has added an epenthetic glottal stop at the beginning of the noun, in order to satisfy the ONSET constraint. Note that words stressed on the leftmost syllable with initial empty onset are provided with an initial glottal stop, by means of Glottal Stop Insertion (Wiese 1996: 58–60). This phenomenon is automatic, which means that it is predictable and thus the result of a phonological insertion rule in classical Generative Phonology. In terms of OT, the relevant constraints are ordered in a way that they turn a stressed syllable with an empty onset into a bad candidate, as shown in Tableau 11. Note that in (11) the glottal stop is assumed to not belong to the lexical representation, although similar results could have been obtained by the opposite assumption. (10a) (11)

Lexicon: Au.to ‘car’

→

(10b)

Connected speech: das.#ʔAu.to ‘the car’

Tableau for the German utterance das Auto ‘the car’ /das.#ˈaw.to/

ONSET

ALIGN (PW,L,FT,L)

☞ [das.#ˈʔaw.to] [das.#ˈaw.to] [da.ˈs#aw.to]

DEP-IO *

*! *!

However, at the morphological level within the word, in German there is resyllabification in the case of suffixation, as e.g. in kin.d+isch ‘childish’, Ach.t+ung ‘attention’ (Wiese 1996: 65). Beyond the morphological level, there is resyllabification at the level of the clitic group, as defined by Kohler (1977), Nespor and Vogel (1986) and Young-Scholten (1994), or at the level of the Phonological Word, as defined by Wiese (1996). For instance, in connected speech, and especially in an informal register, utterances like the following, with subject postposition, can be produced: Hab ich gemacht [ha.b#iç.#ge.maxt], as shown in (3) above, with some more examples.

3 The acquisition of resyllabification and glottal stop insertion: Research questions Children acquiring Spanish as L1 do not seem to experience great difficulties with resyllabification, whereas children acquiring French tend to show various forms for one single noun, as mentioned in section 1. This type of problem

196

Conxita Lleó

hardly seems to appear in Spanish. However, to my knowledge there have not been any in-depth studies on the acquisition of Spanish resyllabification; but see Lleó, Saceda and García (2013), where the phenomenon is briefly described in relation to ten 7-year-old to 8-year-old heritage speakers of Spanish, and compared to other phonological phenomena, like spirantization. That study shows that in Spanish, resyllabification is only applied in 32% of obligatory contexts on average (from 44% in the best case to 20% in the worst case). Given the scarcity of studies on this topic, the present paper aims to report on the acquisition of empty onsets, and their repair strategies – resyllabification and glottal stop insertion – by monolingual Spanish and monolingual German children, respectively, and by German-Spanish bilingual children. The research questions are as follows: (a) Do Spanish monolingual and German-Spanish bilingual children resyllabify in Spanish, as soon as they produce two concatenated words? Do they show a developmental progress or does resyllabification appear all at once? (b) Are all cases of resyllabification by the monolingual German and by the bilinguals in German within the clitic group, as the maximum range for resyllabification in German? (c) Do monolingual German children and bilinguals in German satisfy the Onset constraint by means of a glottal stop? (b) Can a developmental progress be observed? (d) Do monolinguals and bilinguals follow the same course of development in relation to resyllabification and glottal stop insertion? The answers to these research questions will be found by means of auditory analysis, which on the one hand will show resyllabification of the preceding coda into the onset, and on the other hand will show the presence of a glottal stop. Question (a) requires that we identify the percentages of resyllabification along a certain period, as produced by the monolinguals in Spanish, and this curve must be compared with that of the bilinguals. Question (b) requires that resyllabification in German be scrutinized in relation to the range at which it is applied both by monolinguals as compared with bilinguals. Resyllabification should apply within the clitic group (enclisis), since as argued in section 2, German only admits resyllabification under enclisis. In order to answer Question (c), the percentages of glottal stop insertion must be calculated both for monolinguals and bilinguals; moreover, the resulting curves should be controlled for a growing trend. To be able to answer Question (d), percentages of resyllabification and glottal stop insertion by monolinguals and bilinguals, especially their developmental curves, should be carefully compared.

The challenge of lexically empty onsets in first language acquisition

4 Methodology of the study

197

10

The monolingual data considered for analysis stem from the database PAIDUS; the bilingual data stem from the database PhonBLA and PEDSES, as well as from ALCEBLA. For the purposes of the present study, the data from two children of the PAIDUS project, María and Miguel, were followed from the time when they started to produce utterances with two words, which for María was at age 2;2 and for Miguel at age 1;8. Their parents were from Madrid. The PhonBLA database contained data from four children exposed both to German and Spanish from birth, regularly recorded by two separate teams, a German- and a Spanish-speaking team. Three children of PhonBLA were selected, because they all had Spanish-speaking mothers from Spain (from the Madrid area); a fourth child was not considered in the present study, because his mother spoke the Chilean variety of Spanish, characterized by a high degree of consonant weakening and deletion. The fathers of the three children were speakers of the North variety of German. PEDSES involved three bilingual children; only one, whose mother was a speaker of Mexican Spanish, was incorporated into the present study. This Spanish variety hardly undergoes consonant weakening or deletion. The father again was a speaker of North German. ALCEBLA contained data from 23 7- to 8-year-old German Spanish bilingual children, out of which the data of 16 children were selected on the basis that the L1 of their mothers was Spanish and the L1 of their fathers was German. They were living in Germany (Hamburg, Bremen or Mannheim), and were about to begin the complimentary Spanish school that the German Government together with the Spanish Embassy offers to those Spanish citizens who want the language competence of their children to improve. Recordings were done in monthly sessions with the Spanish monolinguals (Spanish PAIDUS), every two or three weeks with the German monolinguals (German PAIDUS) and with the bilinguals (PEDSES and PhonBLA), until the children became 3-years-old, and once a month thereafter. The two languages of the bilinguals were recorded by different teams and, if possible, in different weeks. All data were collected in semi-spontaneous situations, playing with and talking to the child. Only in the case of the ALCEBLA database, the data were elicited in one or two sessions, on the basis of those areas of Spanish 10 I am very grateful to the DFG (German Research Foundation) and to the University of Hamburg for the support given to my research projects and to the Databases stemming from them. Special thanks go to the creation of the Research Center on Multilingualism in Hamburg (1999–2011), whose scientific and financial support was essential for the realization of the research projects. See a more detailed description of the databases in Lleó (2012).

198

Conxita Lleó

grammar that might pose some difficulty in the Spanish heritage context, as e.g. resyllabification. Utterances were transcribed and introduced into EXMARaLDA, a computer-supported transcription system created by Thomas Schmidt and Kai Wörner at the Hamburger Zentrum für Sprachcorpora (see Wörner 2012). A binary decision was made regarding resyllabification (yes/no) and glottal stop insertion (yes/no) on the basis of transcripts, complemented by auditory control. If the lexical item that in the lexicon was provided with an initial empty onset incorporated a previous coda consonant into the onset position in connected speech, this was counted as resyllabification, independently of the correctness of the consonant serving as onset; that is, cases with an onset consonant that was different from the one provided by the input were also counted as resyllabified. The data were also inspected for glottal stop insertion, based on the transcripts, which were controlled by means of auditory analysis. The team met regularly to discuss unclear cases and their transcription. In general, decisions about unclear cases were reached by consensus. Some residual cases, which could not reach consensus, were decided on the basis of the majority’s perception, which amounted to 73% agreement. Auditory analysis was quite reliable with regard to resyllabification, but glottal stop insertion provided some doubtful cases, which were discussed with the acoustic help of PRAAT. The literature has some proposals, which were not always helpful, and in those cases, we trusted our auditory intuitions. For instance, according to Dilley, Shattuck-Hufnagel and Ostendorf (1996: 428–429), following criteria signal the presence of a glottal stop or the presence of glottalization: a salient perceptual impression of a glottal gesture, irregularity in the speech waveform, and a rapid dip in F0 and/or amplitude. The perceptual impression was thus the cue that we used more regularly.

5 Results Results of the analyses are presented below, contrasting those based on Spanish monolinguals, German monolinguals, and German-Spanish bilinguals in their respective two languages. All percentages refer to percentages of target-like utterances.

5.1 Acquisition of resyllabification in Spanish by two monolingual children The development of resyllabification by the two monolingual Spanish children Miguel and María is shown in Figure 1 (beginning at the point in time at which the values of putative contexts for resyllabification are three or more). Miguel

The challenge of lexically empty onsets in first language acquisition

199

Figure 1: Developmental curves of resyllabification and glottal stop insertion for the two monolingual Spanish children, Miguel (first diagram) and María (second diagram)

begins to produce resyllabified utterances a bit earlier than María, although at 2;0 his percentage only reaches 40%, and corresponds to only two tokens. His utterances exhibit a clear increase of resyllabifications, reaching 98% at 2;4 (with 40 tokens), whereas María only reaches 42% (with 10 tokens) at the latest point in Figure 1 (age 2;6–2;7), and only Miguel shows a clearly growing profile. The other measurement included in the picture is glottal stop insertion, which is important keeping in mind the comparison with the development in German. In the case of Miguel it is interesting that both values, resyllabification and glottal

200

Conxita Lleó

stop insertion begin at 2;0 with the same quantitative values as Miguel, namely 40% and two tokens each, but after that, glottal stop insertion is reduced to a minimum, which is relevant taking into consideration that this process does not play a role in Spanish. María shows minimal values of glottal stop insertion at the three data points included in the diagram.

5.2 Resyllabification and glottal stop insertion in German by one monolingual child Figure 2 shows the proportions of glottal stop insertion and resyllabification within seven months, i.e., between ages 1;11 and 2;5, as produced by the German monolingual child, Marion. It indicates that at the initial data points Marion introduces a glottal stop more than 50% of the time (41 tokens), whenever the context presents a word ending in a consonant followed by another word beginning with a vowel, i.e., in a syllable with an empty onset. At 2;2 glottal stops are reduced (to 39 %, corresponding to 24 tokens), and at 2;5 the value goes up to 50%. Resyllabification of the coda consonant reaches 27% at 2;2, corresponding to 17 tokens, and slightly increases at 2;5 (30% with 36 tokens). A comparison of Miguel’s and María’s data on the one hand, and Marion’s on the other hand, shows almost complementary results in the two languages, the two Spanish children manifesting higher percentages of resyllabification, and the

Figure 2: ʔ-Insertion and resyllabification in Marion's data at ages 1;11–2;5

The challenge of lexically empty onsets in first language acquisition

201

German child higher percentages of glottal stops. Note that the decreasing values of Marion’s glottals are rather puzzling. They will be taken up in the Discussion. Examples of Marion’s resyllabification with a few of glottal stop insertion are shown in (12). (12) Examples of resyllabification and glottal stop insertion produced by Marion at age 2;2 (session 25)11 83.

[ˈdʌx.#pʰɑ.ˈs#iə] ‘Yes, it fits here’

Doch, passt hier

88.

[pahs.#ˈʰiɐ.#ʔaʊ̯ x.#ˈʔaɪ ̯n] ‘It fits here, too’

Passt hier auch rein

139.

[ˈhɐ̞.p#iç̞ .#kʰaɪ ̯n.#ˈtʰɐ̞.ʔn̩.tʰux] ‘I don’t have any tempo here’

Habe ich kein Taschentuch

143.

[ha.β#ɪç.#ˈkʰaɪ ̯ns] ‘I don’t have any’

Habe ich keins

146.

[ˈtʰɑ.b#ɪç.#tʰaɪ ̯ʦ] ‘I don’t have any’

Habe ich keins

172.

[dɐ̞s.#ʰiɐ̯.#ˈʔa.ləs] ‘All that here’

Das hier alles

190.

[ʔɪnt.#ˈtʰi.jɐ̞.#ˈʔɑʊ̯ hˠ] ‘And here too’

Und hier auch

207.

[hɪ.#s#ɛ:n.#ˈmã] ‘Here there is a man’

Hier ist ein Mann

215.

[pʰa.ˈs#i.jɐ̞] ‘It fits here’

Passt hier

226.

[hʲiə̯.#ˈpʰɑ:.s#ɵʃ ] ‘Here it fits’

Hier passt es

258.

[da.s#iːh.#ˈʔaʊ̯ .#hˠm̩ .#ˈvɛsː.də] ‘This is also a vest’

Das ist auch eine Weste

286.

[ˈmɑ.kʰ#ɪç.#tɛ.#nɪ.ˈç#ɑ.bm̩ ] ‘I don’t like to have this one’

Mag ich den nicht haben

11 The numbers heading each utterance refer to the numbers in the corresponding corpus. The utterances belong to session 25, when the child was at age 2;2 (years;months). Dots indicate syllable boundaries and the pound key (#) indicates word boundaries. Examples of glottal stop insertion have also been shown (e.g. numbers 88 and 172), especially in those cases that exhibit a contrast between resyllabification and lack of it.

202

Conxita Lleó

Most of these utterances are target-like, especially in an informal register, and when pronounced relatively fast. However, some of them, namely 83, 190, 207 and 215 (bold faced in the list) are not target-like: on the one hand, in (83) and (215) the child has dropped the laryngeal onset /h/ of hier ‘here’, has omitted the last coda obstruent, /t/ from passt, and has converted the previous obstruent /s/ from a coda into an onset. In (190), the coronal coda consonant of und ‘and’ is also pronounced as an onset for the word hier, and the laryngeal onset of hier has become an aspiration for the coronal /t/. On the other hand, (88) and (172) are produced target-like, without resyllabification, as the word hier has kept the initial laryngeal, and the codas of the previous words das and pass(t), respectively, have maintained their coda role (note the cluster reduction in passt). In sentence (207) the /s/ of the auxiliary is(t) has been converted into the onset of the indefinite article ein ‘a’; note that, generally, the host is an auxiliary verb like sein ‘to be’, haben ‘to have’, mögen ‘to like’, building the enclitic group with a pronoun, an article or an adverb (e.g. hier ‘here’). Thus, the child attributes a clitic role to a stressed form like hier. By doing so, the child goes beyond the clitic group as the limiting unit of resyllabification, i.e., the monolingual child is taking resyllabification too far in German.

5.3 Acquisition of resyllabification and glottal stop insertion by bilingual children The production of three German-Spanish bilingual children, Jens and Simon from the corpus PhonBLA, and Stefan, from the corpus PEDSES, was observed in relation to the development of glottal stop insertion and resyllabification, in order to compare these two measurements with each other and with those of the monolingual children in each language. Figure 3 shows the values in Spanish, and Figure 4 those in German.12

5.3.1 Acquisition of resyllabification by three bilingual children in Spanish Figure 3 shows the development of resyllabification and glottal stop insertion for the bilingual children in Spanish, Jens (between ages 2;0 and 5;7), Simon (between 2;2 and 3;8), and Stefan (between 2;3 and 2;5). Although the developmental curves are not always smooth, there are clear trends to see. The first 12 Simon’s data were first calculated in Oltmanns (2007).

The challenge of lexically empty onsets in first language acquisition

Figure 3: ʔ-Insertion and resyllabification in Spanish by three bilingual children

203

204

Conxita Lleó

diagram, corresponding to Jens, shows smoothly but steadily increasing percentages for resyllabification and decreasing percentages for glottal stop insertion up to 2;9. Both trends agree with those of the monolingual Spanish children, as shown in figures 1 and 2. But between 5;0 and 5;7 the values for glottal stops grow up to 62%. Thus, on the one hand the values for resyllabification are not different from those of the monolingual Spanish María, although much lower than those of the monolingual Miguel, and his values for glottals are much higher than those of the Spanish monolinguals. Simon’s values (second diagram) are comparable to Miguel’s, glottal stop insertion staying always at low percentages, and resyllabification steadily growing, reaching 77% at 3;1. However, resyllabification undergoes many oscillations and at time point 3;8 glottal stop insertion remarkably grows, to 29%. Such tendencies do not match those of the monolingual Spanish children, as presented in Figures 1 and 2. Note that the bilinguals are older than the monolinguals, which leads to the expectation of more target-like utterances by the bilinguals. Stefan (last diagram in Figure 3) has the least consistent values of the three bilingual children, with relatively high percentages for glottal stop insertion and low values for resyllabification, with an exceptionally high value of resyllabification (46%) at 2;4. The data analysis also stops earlier than in the case of the other two bilingual children, which might be showing an earlier stage of development.

5.3.2 Acquisition of resyllabification and glottal stop insertion by three bilingual children in German Values for resyllabification and glottal stop insertion in German are showed in Figure 4, in the same order as in Figure 3: The first diagram refers to Jens (between ages 2;0 and 6;9), the second diagram to Simon (at ages 2;3, 3;5 and 3;7), and the third one to Stefan (at ages 2;11 and 3;0). In spite of fluctuations, all three children exhibit much higher values for glottal stop insertion than for resyllabification. Nevertheless, at the last time point (6;9 for Jens, and 3;0 for Stefan) their resyllabification values reach 33% and 27% respectively, and in the case of Simon (at 3;7) 40%. Given the restrictions that German imposes on the range of resyllabification, the contexts at which resyllabification has taken place have been examined more closely. All resyllabifications produced by Stefan at 2;11 and by Simon at 3;7 are shown in (13) and (14) with examples in [12]), respectively, following the same pattern introduced in (12) for Marion (see footnote 9). Most of these productions are grammatically correct, if considered as part of an informal register,

The challenge of lexically empty onsets in first language acquisition

Figure 4: ʔ-Insertion and resyllabification in German by three bilingual children

205

206

Conxita Lleó

as they all take place between a host (accented word, generally a lexical item) and a clitic (or unstressed grammatical word, like a pronoun, article, or the auxiliary is). Stefan only resyllabifies those cases involving clitics. His resyllabifications are target-like, not occurring in non-target-like cases, as Marion’s did (compare examples in [13] and [14] with examples in [12]). (13) Resyllabifications and glottal stop insertion produced by Stefan in German at age 2;11 (session 28) 37.

[ˈkɔɡ.#mas.ˈmax.#tʰa.#ˈda] Guck, was macht er da? ‘Look, what is he doing there?’

104

[ˈç#a:p#s.#kɔɡ.#ma] ‘I have it, look’

Ich hab es, guck mal

119

[ˈe.#i.#ˈhaps] ‘I have it’

Ich habe es

121

[ˈʔi.#çaps.#ˈʔum.#da] ‘I have it up there’

Ich habe es oben da

143

[ˈja.#ɪç.#ˈmak#s] ‘Yes, I do it’

Ja, ich mache es

202

[iç.#haps] (twice) ‘I have it’

Ich hab es

205

[ja.#ɪç.#ʔe.#ˈhɛps:] ‘Yes, I have it’

Ja, ich, eh, hab es

(14)

(Non)Resyllabifications produced by Simon in German at age 3;7 (session 44)

82.

[a.bɑ.dʁ.ɪ.zaɪ ̯n.lɪ.bʁ] ‘But there is a nice one there’

Aber da ist ein lieber

123.

[dɑ.zau̯ x.hɪʁ] ‘That is here, too’

Das (ist) auch hier

161.

[nɛ:.hɑp.ɪç.nɪçtʰ.ɡə.kau̯ vt] ‘No, I havn’t bought it’

Nein, habe ich nicht gekauft

178.

[zɪztʰ.du.hɪʁ.ɪz.ʃnɛ.vɪk] ‘Do you see? Here is S’

Siehst du, hier ist Schneewitchen

The challenge of lexically empty onsets in first language acquisition

207

185/212.

[da.zɪz.rɔt.kæp.çn] ‘This is Red riding-hood’

Das ist Rotkäpchen

230.

[hɪʁz.ʃne:.vɪt.çn] ‘Here is Snow-white’

Hier ist Schneewitchen

235.

[un.kuk.mɑl.hɪʁ.ɪz.ʃne:.vɪk.çn] ‘Look, here is S’

Und guck mal, hier ist Schneewitchen

240.

[un.kɪʁ.ɪz.nɔx.mɑl.ʃne:.vɪk.çn] ‘And here is S again’

Und hier ist noch mal Schneewitchen

258.

[dɑz.ɪz.ʁɔtʰ.kæpç] ‘This is Red riding-hood’

Das ist Rotkäpchen

260.

[un.kɪʁ.ɪz.dɑ.ɡɑnts.ɡʁɔ.sə.vɔlf] ‘And here is the big wolf’

Und hier ist der ganz große Wolf

266.

dʁ.møç.tʰɪç.ɔ.kɛ] ‘This I’d like to have, ok?’

Der möchte ich, ok?

277.

[un.kuk.mɑl.hɪʁ.ɪz] ‘And look, here it is’

Und guck mal, hier ist

297.

dɑz.ʔɪz.ʔaɪ ̯ɡn.bɪltʰ] Das ist eigenes Bild ‘This is the picture of one self’

Simon produces many cases of target-like resyllabification, but tends to generalize it to further non-target-like cases, as for example in (82), (123), (185), (230) and (266). In (82) and (123), proclisis rather than enclisis occurs. In both cases, final /t/ of the auxiliary ist ‘is’ is dropped, and the preceding /z/ is cliticized to the following stressed word: in (82) to the article ein, and in (123) to the adverb auch ‘too’; both words bear stress, and thus serve as hosts. Utterances (185) and (212) undergo resyllabification of the Spanish type: although is(t) ‘is’ is a potential enclitic, the vowel has not been reduced, and the verb is stressed. In such cases, no resyllabification should occur, and Simon himself has produced it in a target-like way, without resyllabification in (297). Resyllabification with truncation of the vowel in das ist ‘that is’ has been applied in (230), which results in enclisis of the auxiliary verb. In utterance (266) the child has also chosen the resyllabified solution, which is optionally possible, as ich ‘I’ can also be treated as an enclitic. Utterances (161), (178), (235), (240), (258), (260), (277) and (297) do not involve incorrect resyllabification: They are shown with the purpose of exemplifying some subtle distinctions made by this child.

208

Conxita Lleó

5.3.3 Resyllabification and glottal stop insertion in Spanish by 7-year-old simultaneous bilinguals13 In section 5.3.1 the curve related to glottal stop insertion was shown to grow towards the end (see diagrams in Figure 3). In order to determine whether such an increase was a mere appearance not solidly supported by the evidence or a real fact in the case of the weak language of simultaneous bilinguals living in Germany, percentages of both, resyllabification and glottal stop insertion, were calculated in the Spanish utterances of 16 German-Spanish simultaneous bilingual children living in Hamburg, Bremen or Mannheim. Recall that in Spanish, to be target-like, values for resyllabification should be high, and values for glottal stop insertion should be low. Figure 5 shows the percentages of both phenomena in Spanish, glottal stop insertion is indicated in the black columns, and resyllabification in the white ones. The figure shows that on the one hand, resyllabification is applied, but in a quite restricted manner, reaching less than 40% in the majority of cases. Eleven children have values for resyllabification around 40%, but five children only around 20%. Thus, as they grow older simultaneous bilingual children living in Germany reach more non-target-like results

Figure 5: Resyllabification and glottal stop insertion in Spanish by 16 simultaneous bilinguals growing up in Germany

13 The project T4 of the Research Center on Multilingualism only dealt with the Spanish competence of German-Spanish 7- to 8-year-old bilingual children living in Germany, which means that there are no German data available. Its purpose was to describe the putative weaknesses of one of the languages of these bilingual children. Data from these children were elicited and analyzed from 2007 until 2010. My thanks go to the Research Assistant, Marta Saceda Ulloa, who did the calculations presented in Figure 5.

The challenge of lexically empty onsets in first language acquisition

209

in their weak language – not supported by the language community – than monolingual children. On the other hand, glottal stop insertion is applied by four children more often than in the Spanish of monolinguals and also more often than by the younger bilinguals presented in Figure 3, namely about 40%. Actually, in five cases the percentage of glottals is higher than the percentage of resyllabification.

6 Discussion The purpose of the present study was to find out how and when resyllabification in Spanish, on the one hand, and both resyllabification and glottal stop insertion in German are acquired by monolingual and bilingual children. For that purpose, we formulated four research questions, whose answers should be found in the results of the analyses. Research questions and results will be presented here side-by-side and discussed. The first question was (a) whether Spanish monolingual and German-Spanish bilingual children resyllabify in Spanish, as soon as they produce two connected words, and whether they show a developmental progress or else resyllabification appears all at once. According to Munz (2003), Miguel produced his first two-word utterances at 1;8, and María at 2;2. According to Figure 1, Miguel already produced 40% of resyllabification at age 2;0, and María at 2;2. Thus, in María’s results it is clearly the case that connected speech and resyllabification emerge at the same time. For Miguel, we have no data before 2;0, because of the restriction that at least three contexts of resyllabification should appear in the relevant recording session. For the bilinguals, resyllabification in Spanish begins a bit later: Jens has only 12% at 2;0–2;2, and Stefan only 5% at 2;3. Simon is the only one of the bilinguals with higher values of resyllabification, similar to the monolinguals; his data begin at 2;2 with 33%. Question (a) also asked about the presence of developmental progress in the acquisition of resyllabification in Spanish and in German. All children, monolinguals as well as bilinguals showed development, although more clearly in the case of the monolingual Miguel, who reaches more than 90% at age 2;4. The German monolingual Marion also shows growth from 6% at 1;11 to 38% at 2;2. The bilingual Jens shows some growth between 2;0 and 2;7 in Spanish, but there seems to be some stagnation thereafter: at 5;7 he has a value slightly lower than the one he had at 2;6–2;7, namely 38% at 5;7 vs. 45% at 2;6–2;7. In Spanish, Stefan goes from 5% at 2;3 to almost 50% at 2;4, but back to 10% at 2;5. And he does not show any growth in German, either: 27% at 3;0 after his initial 31% at

210

Conxita Lleó

2;11. As already mentioned, Simon is the bilingual with the highest resyllabification values in Spanish, in spite of some fluctuations. In German, Jens has very low values, but they clearly grow beginning at 2;8 (with 0%) up to 6;9, which is the last data point, with 33%. Simon has a similar profile, with low values of resyllabification in both languages, with a rise to 40% at 3;7 in German and 42% at 3;8 in Spanish. The next question (b) is whether all cases of resyllabification by the monolingual German and by the bilinguals in German are within the ClG, as the maximum range for resyllabification in German. We have seen in the preceding section that this is generally the case, but that some children tend to also apply resyllabification beyond the ClG, in cases of stressed items. This has been analyzed in relation to the examples presented in (12), (13) and (14). It should be noted that the clitic group appears to be a crucial entity in defining resyllabification in German and in describing violations leading to non-target-like resyllabification. Thus, as a consequence of the range of resyllabification in German, the ClG appears to be a necessary, or at least adequate, level of the Prosodic Hierarchy. Question (c) is whether monolingual German children and bilinguals in German satisfy the Onset constraint by means of a glottal stop. This is indeed the case. All children produce glottal stops at onset position (instead of resyllabification) to some degree. Moreover, all curves related to the glottal stop have a growing tendency, some of them preceded by a decrease of the values. The monolingual Marion has 63% at 1;11, which is reduced to 43% at 2;2. The three bilinguals show high percentages of glottal stops, with a growing tendency: Jens begins at 2;0–2;2 with 75%, and after some fluctuations he ends up with a value of 50% at 6;9. Simon starts at 2;3 with 47% and ends at 3;7 with 40%. Finally, question (d) asks whether monolinguals and bilinguals follow the same course of development in relation to resyllabification and glottal stop insertion. The findings indicate that the percentages of glottals in Spanish are extremely low for the two monolinguals (hardly over 10%), whereas the bilinguals in Spanish show relatively high values: Jens begins with 29% at 2;0, and although he shows a smooth but clear decrease to 0% at 2;9, his percentages grow again at 5;0, reaching 62% at 5;7. Simon has relatively low values of glottal stop insertion, but at 2;9 he produces 30%, and towards the end of the study, at 3;8, he reaches 29%. Finally Stefan has high values at the three data points observed: at 2;3 he reaches 56%, descends to 46% at 2;4 and increases again to 55% at 2;5. An interesting fact that emerges from these data is the relation between the two linguistic systems of the bilingual children. Whereas monolingual Spanish children have high values for resyllabification and low values for glottal stop

The challenge of lexically empty onsets in first language acquisition

211

Table 1: MLU values of three German-Spanish bilingual children in their two languages. Age

Jens Sp

Jens Ge

Simon Sp

Simon Ge

Stefan Sp

Stefan Ge

1.0 1;8 1;10 2;0 2;2 2;4 2;6 2;8 2;10 3;0 3;1

1.0 1.0 1.02 1.11 – 1.62 – 2.1 1.88 2.62 2.51

1.0 1.0 1.0 1.75 – – – 2.82 2.77 3.11 2.88

1.02 1.19 1.0 1.28 1.44 2.13 – 2.82 2.59 3.20 3.26

1.0 1.06 1.06 1.40 1.55 – – 3.01 – 3.70 4.30

1.0 1.5 1.56 1.54 2.0 1.66 1.71 – 2.24 2.68 –

1.58 1.43 1.40 1.25 1.85 1.75 – 2.21 3.2 –

insertion, the monolingual German child has high values for glottals and the values for resyllabification are relatively low and begin late. Two of the bilinguals in Spanish (Jens and Stefan) have values for glottals that are higher than expected, and lower values for resyllabification, when compared to the Spanish monolinguals. What could the reason be for such diverging results? It could be argued that glottal stops are unmarked segments, which very young children produce, also in cases that are non-target-like, which can explain Marion’s (the monolingual German child’s) developmental curve, with higher percentages of glottal stops at 1;11 than later on (Figure 2).14 On the other hand, German is the ambient language for these children living in Germany, and it is slightly dominant for them, which manifests itself in their MLU values, in general somewhat higher in German than in Spanish (see Table 1). That is, at most time points the German MLU is a bit higher than the Spanish MLU. The two languages of the bilingual child interact, with some influence going from the dominant into the heritage language, i.e., Spanish in the present case. With relation to resyllabification, being the ambient language is a necessary condition to being the dominant language of the bilingual child, as in the case of Jens and Stefan. However, it is not a sufficient condition, as Simon’s resyllabification values are not different from those of monolinguals’. If we now compare the two kinds of resyllabification, the Spanish kind vs. the German kind, we find that Spanish resyllabification begins sooner, reaches 14 Lleó, Prinz, Mogharbel and Maldonado (1996) argue that very young German as well as Spanish monolingual children produce high percentages of laryngeals (stops and fricatives), both in German and Spanish, especially in babbling. However, the children of that study were in an earlier stage than those of the present study.

212

Conxita Lleó

much higher values and seems to work in a straightforward manner. German resyllabification, on the other hand, requires more grammatical and prosodic knowledge. Since it only happens in clitic contexts, as the word that can have an onset from a previous coda must be in enclitic position (its host is on the left-hand side), the child needs a clear-cut prosodic classification of what is stressed and what is not stressed. The child also needs relatively good knowledge of verbal morphology, since resyllabification in German generally takes place between the verb and the postposed weak pronoun, a subject or an object. In Spanish, resyllabification matters are more transparent and have a larger range, up to the IPh. Still, German seems to be influencing degree of resyllabification in Spanish, especially in the case of Jens and Stefan. It is interesting to note that the analyses of the German utterances of bilinguals have shown that there is no influence in the other direction, from Spanish to German. This may be due to Spanish being the weak language of some bilinguals, whereas German is slightly dominant, for some of the bilingual children living in Germany. However, it could also be due to purely internal factors in the following sense. On the one hand, resyllabification in the Spanish way results in optimal syllables, in optimally segmented syllables. On the other hand, glottal stop insertion results in optimal words, in optimally segmented words. In Lleó and Vogel (2004) it was argued that German is a demarcative language, in the sense proposed by Trubetzkoy (1939), whereas Spanish is a grouping language, because phonological phenomena have higher constituents (generally the Intonation Phrase) as their relevant range (see Chen 1990 for the notion of “grouping language”). That proposal can now be complemented by the idea that Spanish is grouping, but it also is a syllabic language in the sense that it has a preference for optimally segmented syllables. The relevant constraint creates a misalignment between syllables and words, as the syllabic constituents do not match the lexical constituents anymore. Moreover, in Spanish resyllabification involves variation in word production, because a word that begins with a vowel can be alternatively produced with an initial consonant – e.g., o.so, (u.)no.so ‘a bear’, (e.)lo.so ‘the bear’, (u.no.)so.sos ‘some bears’, etc. This same tendency to avoid variation, for the sake of uniformity, has also been observed in these bilingual children with regard to processes like spirantization and assimilation of Place of Articulation (PA) of nasals (Lleó submitted). In sections (1) and (2) various theories of resyllabification have been presented: Harris (1983) and (1993), Hualde (1992) and Colina (1997). A further question should be posed about the role of data from L1 acquisition in deciding what theory is more adequate to account for resyllabification. In other words, what information should appear in the child’s grammar regarding resyllabification?

The challenge of lexically empty onsets in first language acquisition

213

The Spanish resyllabification phenomenon is a relatively simple one, which could be formulated in an additional rule of the kind presented by Harris (1983), or by reapplying an already existing rule, namely the CV rule, as in Hualde (1992). However, such rules lack descriptive adequacy for several reasons. On the one hand, they are structure-changing rules, which several phonologists have tried to get rid of (Steriade 1988, Rice 1989, among others). For instance, Rice (1989) rejects them, but has to pay the price of enlarging the notion of extraprosodicity. Transposed to the child acquiring the L1, this would mean that isolated words cannot be syllabified, and the child has to wait until connected speech emerges, in order to syllabify. By this trick, resyllabification is avoided, and replaced by syllabification. On the other hand, phonological rules were very powerful and introduced much arbitrariness into the child grammar, e.g., deletion rules, which deleted certain segments for the child, and their effects had to be neutralized in order to acquire the adult language. Resyllabification creates a complex relationship between syllables and words, because of the emergence of the misalignment between their edges. However, such rules are mute precisely about this important result: They do not make explicit that the resulting utterances involve a mismatch between syllable- and word-edges. An alternative way of representing resyllabification consists in choosing the target-like candidate for production, on the basis of the hierarchy of constraints (6) >> (7), i.e. ONSET >> ALIGN LEFT. This hierarchy of constraints penalizes syllables with empty onsets in contexts where material to fill the onset is available. In German, the most straightforward mechanism to avoid such syllables with empty onsets is glottal stop insertion – see Tableau (11) – reformulated in OT as a violation of DEP-IO. Note that German has two mechanisms to fill empty onsets: glottal stop insertion if the syllable is stressed, and resyllabification (with ambisyllabicity) if the syllable is unstressed, i.e., enclitic. Later on, child German also contains such verbal constructions with postposed subject, like ha.bich ‘I’ve’ or kan.nich. ‘I can’. These potentially resyllabified expressions are very limited in number, when compared with Spanish reyllabification. Thus, it is plausible that young children may produce these expressions as mimicries, imitating them as if they were lexical items, which would mean that no resyllabification is yet involved. However, the overextensions of resyllabification shown in (11), (12) and (13), as produced by young children in German, indicate that resyllabification is involved. In the case of the bilinguals Stefan and Simon, this is true relatively late: at 2;11 and 3;7, respectively. However, in the case of the monolingual Marion, she seems to already apply resyllabification at an early age, namely at 2;2. The question of whether the first productions of such “formulaic” German expressions do already involve resyllabification or not cannot be conclusively answered at this point of the investigation.

214

Conxita Lleó

Trying to match the language of the parents with percentages of resyllabification does not provide any clear correlation. For instance, most of the children in Figure 5 show the same constellation of languages at home as the younger group of simultaneous bilinguals (Figures 3 and 4), i.e., the mothers’ L1 is Spanish and they speak to their children in Spanish; the fathers’ L1 is German and they talk to their children in German. However, the child labeled TIK has the highest resyllabification rate of the whole group, with rather low percentage of glottals, and the child labeled TR has the lowest percentage of resyllabification of the group, and one of the highest percentages of glottals (almost 40%). A few cases do show a match between the languages at home and the percentages of resyllabification, e.g., JP’s and JC’s fathers are Spanish speakers, which means that both parents have Spanish as L1, and these children reach relatively good results in comparison to the rest of the group. However, these seem the exception rather than the law. In fact, the lack of correlation between the language of the parents and the children’s language has been found quite often in this type of study (see e.g. Kerswill 1996, Lleó and Cortés 2013). From a different perspective, these results show that as the bilingual children grow older, resyllabification does not increase, on the contrary, it seems that it is Transfer that shows an increase.

7 Conclusion Spanish, like other Romance languages, has a resyllabification process, which provides an onset to onsetless word-initial syllables. A language like German uses another mechanism to fill an onset: If a word initial syllable has an empty onset, a glottal stop is generally epenthesized in that position. Thus, it can be argued that both processes, resyllabification and glottal stop insertion, preserve the unmarkedness of syllable structure by providing an onset, albeit by means of different mechanisms: Spanish compromises word identity by manipulating syllable structure at the left edge of words, whereas German avoids modifying syllable structure. Still, German also has resyllabification, but in a more restricted form: under enclisis, and in a fast and informal register. The study reported here presents longitudinal child data on the acquisition of resyllabification, produced by two Spanish monolinguals from Madrid (Spain), by one German monolingual from Hamburg (Germany), and by three GermanSpanish simultaneous bilinguals from Hamburg (Germany), Spanish being the weak or heritage language (HL), not supported by the ambient language. Crosssectional data from a group of sixteen 7–year-old German-Spanish simultaneous

The challenge of lexically empty onsets in first language acquisition

215

bilinguals from Hamburg, Bremen or Mannheim (Germany) have also been presented. Whereas Spanish monolinguals begin to resyllabify very soon – as soon as they produce two-word utterances – with very high percentages of resyllabification, two of the bilinguals have lower percentages of resyllabification, and higher percentages of glottal stops in word-initial onset position. They thus do not resyllabify as much, often (but not always) inserting a glottal stop instead. An explanation for the lack of resyllabification of these bilinguals is looked for in the other language, German, which provides negative transfer, the reason being that HL (Heritage Language) speakers receive less input in the HL, and tend to substitute fragments of its hierarchy of constraints by the hierarchy of the other, dominant language. Ultimately, these bilinguals avoid word variation, and form variation, in general.

References Benet, Ariadna, Susana Cortés & Conxita Lleó. 2012. Devoicing of sibilants as a segmental cue to the influence of Spanish onto current Catalan phonology. In Kurt Braunmüller & Christoph Gabriel (eds.), Multilingual individuals and multilingual societies. Hamburger studies on multilingualism (HSM 13). Amsterdam & Philadelphia: John Benjamins. Berg, Thomas & Jussi Niemi. 2000. Syllabification in Finnish and German: Onset filling vs. onset maximization. Journal of Phonetics 28. 187–216. Blevins, Juliette. 1995. The syllable in phonological theory. In John A. Goldsmith (ed.), The handbook of phonological theory, 206–244. Cambridge, MA: Blackwell. Boersma, Paul & David Weenink. 2011. PRAAT: Doing phonetics by computer [computer program]. Version 5.3.02, retrieved Nov. 2011 from http://www.praat.org/ Broselow, Ellen. 1984. An investigation of transfer in second language phonology. International Review of Applied Linguistics in Language Teaching 22 (4). Chen, Matthew Y. 1990. What must phonology know about syntax? In Sharon Inkelas & Draga Zec (eds.), The phonology-syntax connection. 19–46. Chicago & London: The University of Chicago Press. Chevrot, Jean-Pierre, Celine Dugua & Michel Fayol. 2009. Liaison acquisition, word segmentation and construction in French: a usage-based account. Journal of Child Language 36. 557–596. Christophe, Anne, Emmanuel Dupoux, J. Bertoncini & Jacques Mehler. 1994. Do infants perceive word boundaries? An empirical study of the bootstrapping of lexical acquisition. Journal of the Acoustical Society of America 95. 1570–1580. Colina, Sonia. 1997. Identity constraints and Spanish resyllabification. Lingua 103 (1). 1–23. Colina, Sonia. 2006. Optimality-theoretic advances in our understanding of Spanish syllable structure. In Fernando Martinez-Gil & Sonia Colina (eds.), Optimality-theoretic studies in Spanish phonology, 172–204. Amsterdam & Philadelphia: John Benjamins. Derwing, Bruce L. 1992. A “pause-break” task for eliciting syllable boundary judgments from literate and illiterate speakers: preliminary results for five diverse languages. Language and Speech 35, 219–235.

216

Conxita Lleó

Dilley, Laura, Stephanie Shattuck-Hufnagel & Mari Ostendorf. 1996. Glottalization of wordinitial vowels as a function of prosodic structure. Journal of Phonetics 24, 423–444. Downing, Laura J. 1998. On the prosodic misalignment of onsetless syllables. Natural Language and Linguistic Theory 16, 1–52. Durand, Jacques & Chantal Lyche. 2008. French liaison in the light of corpus data. Journal of French language studies 18, 33–66. Féry, Caroline. 2004. Liaison and syllable structure in French. Manuscript. Harper, Douglas. 2001–2014. Online etymology dictionary. http://www.etymonline.com. Harris, James W. 1983. Syllable structure and stress in Spanish. A nonlinear analysis. Cambridge, MA: The MIT Press. Harris, James W. 1984. La espirantiación en castellano y la representación fonológica autosegmental. Estudis gramaticals 1, 149–167. Bellaterra: Universitat Autònoma de Barcelona. Harris, James W. 1993. Integrity of prosodic constituents and the domain of syllabification rules in Spanish and Catalan. In Kenneth Hale and Samuel Jay Keyser (eds.), View from Building 20, 177–193. Cambridge, MA/London, England: The MIT Press. Hualde, José-Ignacio. 1992. On Spanish syllabification. In Manuel Campos & Fernando MartínezGil (eds.), Current studies in Spanish linguistics, 475–493. Washington, DC: Georgetown University Press. Hualde, José-Ignacio. 2005. The sounds of Spanish. Cambridge University Press. Itô, Junko. 1989. A prosodic theory of epenthesis. Natural Language and Linguistic Theory 7, 217–259. Kerswill, Paul. 1996. Children, adolescents and language change. Language Variation and Change 8, 177–202. Kohler, Klaus J. 1977. Einführung in die phonetik des Deutschen. Berlin: Erich Schmidt. Kohler, Klaus J. 1994. Glottal stops and glottalization in German. Phonetica 51, 38–51. Levin, Juliette. 1985. A metrical theory of syllabicity. Cambridge, MA: MIT dissertation. Lipski, John M. 1994. Spanish stops, spirants and glides: From consonantal to vocalic. In M.L. Mazzola (ed.), Issues and theory in Romance linguistics, 67–86. Washington, DC: Georgetown University Press. Lleó, Conxita. 2012. Monolingual and bilingual phonoprosodic corpora of child German and child Spanish. In Thomas Schmidt & Kai Wörner (eds.), Multilingual corpora and multilingual corpus analysis. Hamburger studies on multilingualism 14. 107–122. Amsterdam & Philadelphia: John Benjamins. Lleó, Conxita. Submitted. Acquiring bilingual phonologies (2L1 and L2): Interface or Variation? In Susann Fischer & Christoph Gabriel (eds.). Forthcoming. Grammatical interfaces in Romance linguistics. Berlin & New York: Mouton De Gruyter. Lleó, Conxita & Susana Cortés. 2013. Modelling the outcome of language contact in the speech of Spanish-German and Spanish-Catalan bilingual children. In Johannes Kabatek & Lucia Loureido (eds.), Special Issue on language competition and linguistic diffusion: Interdisciplinary models and case studies. International Journal of the Sociology of Language 221, 101–125. Lleó, Conxita, Michael Prinz, Christliebe El Mogharbel & Antonio Maldonado. 1996. Early phonological acquisition of German and Spanish: A reinterpretation of the continuity issue within the Principles and Parameters Model. In Carolyn E. Johnson & John H.V. Gilbert (eds.), Children’s Language Volume 9, 11–31. Mahwah, New Jersey: Lawrence Erlbaum Associates.

The challenge of lexically empty onsets in first language acquisition

217

Lleó, Conxita, Marta Saceda Ulloa & Izarbe García Sánchez. 2013. Promoting the weak language of German-Spanish bilingual children living in Germany. In Kristin Bührig & Bernd Meyer (eds.), Transferring linguistic know-how into institutional practice. Hamburger studies on multilingualism (HSM 15). 99–120. Amsterdam & Philadelphia: John Benjamins. Lleó, Conxita & Irene Vogel. 2004. Learning new segments and reducing domains in German L2 Phonology: The role of the Prosodic Hierarchy. International Journal of Bilingualism 8 (1). 79–104. Millotte, Séverine, Roger Wales & Anne Christophe. 2007. Phrasal prosody disambiguates syntax. Language and Cognitive Processes 22. 898–909. Munz, Nina. 2003. Resilbifizierung in L1–Erwerb des Spanischen. Hamburg, Germany: University of Hamburg MA thesis. Nespor, Marina & Irene Vogel. 1986. Prosodic phonology. Dordrecht: Foris. Nguyen, Noël, Sophie Wauquier-Gravelines, Leonardo Lancia & Betty Tuller. 2007. Detection of liaison consonants in speech processing in French. Experimental data and theoretical implications. In Pilar Prieto, Joan Mascaró & Maria-Josep Solé (eds.), Segmental and prosodic issues in Romance phonology, 3–23. Amsterdam & Philadelphia: John Benjamins. Oltmanns, Silke. 2007. Resilbifizierung im L1-Erwerb eines deutsch/Spanischen bilingualen Kindes. Hamburg, Germany: University of Hamburg MA thesis. Prunet, Jean-François. 1986. Liaison and nasalization in French. In Carol Neidle & Rafael Nuñez Cedeño (eds.), Studies in Romance languages, 225–235. Dordrecht: Foris. Rice, Keren D. 1989. On eliminating resyllabification into onsets. In Proceedings of the Eighth West Coastal Conference on Formal Linguistics 8, 331–346. Selkirk, Elisabeth. 1982. Syllables. In Harry van der Hulst & Norval Smith (eds.). The structure of phonological representations, vol. 2, 337–383. Dordrecht: Foris. Steriade, Donca. 1988. Greek accent: A case for preserving structure. Linguistic Inquiry 19, 271– 314. Trubetzkoy, Nikolai S. 1939. Grundzüge der phonologie. Prag: Travaux du Cercle Linguistique de Prague. Vogel, Irene. 1991. Prosodic phonology: second language acquisition data as evidence in theoretical phonology. In Thom Huebner & Charles A. Ferguson (eds.), Crosscurrents in second language acquisition and linguistic theories, 47–66. Amsterdam & Philadelphia: John Benjamins. Wiese, Richard. 1996. The phonology of German. Oxford: Clarendon Press. Wörner, Kai. 2012. Finding the balance betwen strict defaults and total openness: Collecting and managing metadata for spoken language corpora with the EXMARaLDA Corpus Manager. In Thomas Schmidt & Kai Wörner (eds.), Multilingual corpora and multilingual corpus analysis, 383–400. Amsterdam & Philadelphia: John Benjamins. Young-Scholten, Martha. 1994. On positive evidence and ultimate attainment in L2 phonology. Second Language Research 10. 193–214.

Rafael A. Núñez Cedeño

The acquisition order of liquids in Spanish complex onsets Abstract: Most research on children’s development of tautosyllabic obstruent + liquid clusters has shown that in this sequence the least sonorous consonants are first to emerge. Few, however, have raised the question of whether there is a gradual or categorical developmental order in their liquids. Based on cross-sectional data coming from 12 monolingual Spanish-speaking children, the present study shows there is a gradual developmental order whereby obstruent+lateral emerge before obstruent+rhotic clusters. As a corollary of this proposal, I further suggest that this order obeys the physical properties inherent in the production of a rhotic over a lateral in child speech: Young children have articulatory difficulty in producing rhotics. This study thus supports a hierarchy where the markedness constraint *rhotic-ONS outranks *lateral-ONS and further confirms previous works by Kehoe et al. (2008) on the existence of an acquisition order for the two types of clusters in question.

1 Introduction In his 2008 study, Núñez Cedeño investigated the emergence of complex onsets in the speech of Seihla, a monolingual Dominican Spanish-speaking girl. He showed that she was inclined to reduce obstruent+liquid onset clusters, cateNote: I am most grateful to Jennifer Taveras and María Eugenia Ricart, directors of the preschools where the study was carried out, for allowing me to use their schools for the research reported herein. I am especially indebted to all parents and their children who willingly agreed to participate in the project. I am also thankful to Mandy Faretta-Stutenberg for sharing her statistical expertise and to Junice Acosta for her assistance in transcribing all tokens phonetically. Very special recognitions go to Laura Bartlett, Travis Bradley, Sonia Colina, Conxita Lleó, John Grindstead, and Fernando Martínez-Gil for helping me considerably with their probing comments and suggestions, thus making sure the paper’s arguments were stronger and free from infelicities. The usual exemptions apply. Also, this paper benefited from comments and questions made by audiences at the 64th Annual Kentucky Foreign Language Conference and LSRL 41. Rafael A. Núñez Cedeño, University of Illinois at Chicago

220

Rafael A. Núñez Cedeño

gorically favoring the survival of the initial obstruent1, the least sonorous consonant, as has been attested in the development of onsets of other languages (Fikkert 1994, Lleó and Prinz 1996, Ohala 1999, Rose 2000, Grijzenhout and Joppen 2002, Freitas 2003, Barlow 2003, Pater and Barlow 2003, Gnanadesikan 2004, Barlow 2007). He concluded, following similar arguments adopted by other investigators, that consonantal onset reductions follow a “sonority-based onset selection” (Pater and Barlow 2003: 490), which is guided by a universal sonority scale where the least marked consonantal segments are obstruent stops. Observing a pattern by which Seihla was making a distinction between the acquisition of obstruent+rhotic2 vs. obstruent+lateral clusters, Núñez Cedeño further advanced the idea that the latter developmentally precedes the articulation of the former, a hypothesis that Kehoe et al. (2008) also suggest. In the present study, based on cross-sectional data from 12 monolingual Spanish- speaking children, I expand Núñez Cedeño’s (2008) findings and the seminal idea on the acquisition order of liquids raised in the study by Kehoe et al. (2008) and propose that the obstruent+lateral sequence emerges before obstruent+rhotic clusters and further suggest that these complex onsets develop gradually over time. As a corollary of this proposal, and still in support of studies by Núñez Cedeño and Kehoe et al., I further maintain that this order, rather than just being based on the cross-linguistic patterns of acoustic correlates observed by Parker (2002), also corresponds to the physical properties inherent in the production of a rhotic over a lateral in children’s speech: they have articulatory difficulty in producing rhotics. Thus it is assumed, following the ideas developed in Kirchner (1998), that rhotics require more complex articulatory gestures to produce. This study provides additional support for a hierarchy where the markedness constraint *rhotic-ONS outranks *lateral-ONS. To frame the discussion, I have organized the paper as follows. Section 2 presents an overview of Spanish syllable structure as produced by adults, which is assumed to underlie children’s own structure. Section 3 gives an overview of the development of complex onsets and introduces the Sonority Dispersion Principle, which not only enhances the effects of the universal sonority hierarchy but is also at the heart of better explaining and capturing the two-consonant cluster reduction to the least sonorous one. Section 4 discusses the methodology

1 From here on, the term obstruent in complex onsets should be understood to represent either the plosives /p,t,k,b,d,ɡ/ only or the fricative /f/, as discussed in section 2. 2 Here and elsewhere in the paper, the term ‘rhotic’ refers to the tap /ɾ/, as only the tap is involved in this type of cluster. This decision is independent from the controversial issue of whether Spanish has one or two distinctive rhotics, which will not be addressed in this paper (see Núñez Cedeño 1994, Bradley 2006).

The acquisition order of liquids in Spanish complex onsets

221

employed in carrying out the research and coding the data. Section 5 presents the complex onset data produced by all children. Section 6 provides a general discussion of the findings and delves into the phonetic reasons why liquids (lateral and rhotic) must be ordered. Section 7 looks into the order of acquisition of /l/ over /ɾ/ and proposes a refined constraint ranking for liquids. Section 8 concludes the paper and suggests avenues for future research.

2 Spanish syllable structure Considering first the phonological inventory for consonants given in (1), I will assume the syllable structure in (2) for standard Latin American Spanish, which essentially replicates the structures proposed by several researchers (Harris 1983, Núñez Cedeño 1985, and Hualde 1991): (1)

/ p t k b d ɡ f s m n ɲ l ɾ ʧ Ɉ x w j /3

(2)

Syllable structure a.

b.

As shown in (2a,b), the syllable may be headed by an optional Ons(et) and a Rhyme, which contains one of five primary vowels as its obligatory nucleus. Since I am focusing on onsets only, I will disregard the structure of the rhyme and instead will center my attention on onsets. The C represents a skeletal slot to which a C(onsonant) melody is attached. In (2a) any one of the consonants in (1) may occupy that C, thus being a viable simple onset. This consonant assignment is essentially guided by the universal tendency to privilege the basic, 3 In this inventory I am assuming that /Ɉ/ underlies the palatal plosive [Ɉ] and the fricative palatal [ ʝ], whose realizations are contextually driven: the former occurs before a pause and after a nasal, whereas the latter appears elsewhere, as a process of spirantization. It should be noted, however, that the level of constriction of either one varies according to the dialect, from a glide or fricative realization to a stop, which have led to discussions on the phonemic status underlying these allophones (Hualde 2005: 164–72). I further take the arguable position that /j,w/ exist phonologically in Spanish.

222

Rafael A. Núñez Cedeño

non-branching CV syllabic prototype. The natural assumption here is that when confronted with the unstructured segmental sequence VCV, languages are likely to construct their syllable by providing it with an onset first, thus syllabifying this string as V.CV, rather than VC.V. Similarly, upon encountering an unstructured syllabic string like CVC, Spanish first syllabifies it as CV.C, and then incorporates the final C as a coda in a second pass, yielding CVC (Hualde 1991, Núñez Cedeño 1985). Alternatively, to account for the canonical CV typology from an OT view, a grammar may have two markedness constraints, interplaying with each other: ONSET, *[σ V, requiring a syllable to begin with a consonant, and *CODA, *C]σ, banning syllables that end in a consonant. If the type CVC exists, as it does in Spanish, it is because *CODA is out-ranked by a few crucial faithfulness constraints (Colina 2009: 11–18). This CVC syllabic type can be seen in the first and final syllables of the word ful.mi.nar ‘fulminate’. The tendency to accept the CV canonical shape in Spanish is not always respected, as shown by the first syllables of a.plicar, ‘to apply’, a.presar ‘to arrest’ om.bligo ‘navel’, a.tlas4 ‘atlas’, and a.fronta ‘s/he faces’, given the optionality of Onset, as in representation (2a). The structure in (2b) shows that onsets can be more complex, consisting of the maximal sequence C1C2, where C1 are either the plosives /p,t,k,b,d,ɡ/ or the fricative /f/, and C2 represents the simple rhotic /ɾ/ or the lateral /l/, except for */dl/, which phonologically does not exist though phonetically it may surface (Guitart 1999: 250). Borrowing from Núñez Cedeño (2008), (3) and (4) provide instances of word initial and internal onset realizations, shown in boldface and marked by the syllable boundary symbol [.]5: (3) Non-branching onsets petaca [pe.ˈta.ka] dato [ˈda.to] fulminar [ful.mi.ˈnaɾ] loro [ˈlo.ɾo] rosado [ro.ˈsa. ðo] gentío [hen̪.ˈti.o]

‘tobacco pouch’ ‘datum’ ‘fulminate’ ‘parrot’ ‘pink’ ‘crowd’

bodega galeno señal mucho carreta cabulla

[bo.ˈðe.ɣa] [ɡa.ˈle.no] [ˈse.ɲal] [ˈmu.ʧo] [ka.ˈre.ta] [ka’βu.ʝa]

‘grocery store’ ‘physician’ ‘signal’ ‘a lot’ ‘cart’ ‘hemp’

4 As is well-known, in Mexican and other varieties of Latin American Spanish tautosyllabic /tl/ is the norm, although in Peninsular Spanish it may be parsed as heterosyllabic (Hualde 2005: 74). 5 The acute mark “ ʹ ” on vowels reflects Spanish spelling.

The acquisition order of liquids in Spanish complex onsets

(4) Possible complex onsets playa [ˈpla.ʝa] ‘beach’ premio [ˈpɾe.mjo] ‘award’ blanco [ˈblaŋ.ko] ‘white’ bruja [ˈbɾu.ha] ‘witch’ Nestlé [nes.ˈtle] ‘Nestle’ drama [ˈdɾa.ma] ‘drama’ glotón [ɡlo.ˈton] ‘greedy’ crema [ˈkɾe.ma] ‘cream’ gremio [ˈɡɾe.mjo] ‘syndicate’ flota [ˈflo.ta] ‘fleet’ fracaso [fɾaˈka.so] ‘failure’

aplicar apresar ombligo cumbre atlas tundra mangle suscribe mugre conflicto afrontar

[a.pli,’kaɾ] [a.pɾe.ˈsaɾ] [om.ˈbli.ɣo] ̞ [ˈkum.bɾe] [ˈa.tlas] [ˈtun̪.dɾa] [ˈmaŋ.ɡle] [sus.ˈkɾi.βe] [ˈmu.ɣɾe] [koɱ.ˈflik.to] [a.fɾon̪.ˈtaɾ]

223

‘to apply’ ‘to arrest’ ‘navel’ ‘summit’ ‘atlas’ ‘tundra’ ‘mangrove tree’ ‘subscribe’ ‘dirt’ ‘conflict’ ‘to face’

Except for the admissible combinations /h+l, ɾ/ in Dominican Spanish and syllable-initial/ /tl/ for both Dominican and Mexican Spanish, no other kinds of C1C2 onsets exist in general Spanish. Still, there are some researchers who have argued for the existence of [C+w/j] onsets for adults’ speech (D’Introno 1995: 401–402, Hammond 2001: 122, 134, Martínez-Gil, this volume, among others), but generally the most accepted position is that glides are constituents of the rhyme.

3 Background on complex onset development in Spanish The complex onsets listed in (4) belong to adult grammar and have been used as the point of departure for the analysis of their acquisition in Spanish and for other languages. Basically, all data collected thus far generally reveal that when children attempt to reproduce complex onsets of the types obstruent+ liquid, their outputs show that the obstruent consonant is retained while the liquid is normally effaced, as has been documented in the works cited in the introduction, of which (5) gives representative examples. (5)

Onset reduction in various languages Portuguese creme /kɾεmi/ [kε] prédio /pɾεdju/ [pεdu] French

cle fleur

/kle/ /flœR/

[kε] [βœ]

‘cream’ ‘building’

(Freitas 2003)

‘key’ ‘flower’

(Rose 2000)

224

Rafael A. Núñez Cedeño

Dutch

trein brand

/tɾein/ /bɾand/

[tεi] [bat]

‘train’ ‘fire’

(Fikkert 1994)

German

brot frau

/bʁo:t/ /fʁau/

[bo:] [bau]

‘bread’ ‘woman’

(Grijzenhout and Joppen 2002)

English

friend blue

/fɾεnd/ /blu/

[fε] [bu]

Spanish

fresa

/fɾesa/

[fesa]

(Gnanadesikan 2004) ‘strawberry’

(Barlow 2003, Núñez Cedeño 2008)

The main force driving this retention results from the interplay of the effects between the Universal Sonority Hierarchy in (6) and the Sonority Dispersion Principle given in (7). The scale in (6) is adapted from Parker (2002: 240), which is a more fine-grained sonority distinction among segments when compared to that of other researchers who group fricatives and plosives under the obstruent class, and do not distinguish between voiced and voiceless consonants (Foley 1977, Anderson and Ewens 1987, Clements 1990, Blevins 1995, among others). (6) Universal Sonority Hierarchy vowels > glides > liquids > nasals > voiced fricatives > voiceless fricatives > voiced plosives > voiceless plosives Assuming that sonority is defined in terms of a correlation that may exist between major class features and physical properties of sounds (Clements 1990: 284), (6) says that sounds are scaled in terms of their sonority, with the most sonorous on the left and the least sonorous on the right. Put differently, (6) captures the idea that when reducing consonantal clusters, children tend to allow the least sonorous segments to surface. But in order to work effectively with the data in (5), (6) must work in tandem with Clements’ (1990) Sonority Dispersion Principle, which I partially paraphrase from Kenstowicz (1994: 283) in (7). (7)

Sonority Dispersion Principle (SDP) A syllable prefers to maximize the sonority slope in demisyllables formed by Onset+vowel while it tends to minimize that slope in the rhyme demisyllables.

The SDP exemplifies a ranking for unmarked syllables of the CV-type in which the best onsets show a “sharp and steady rise in sonority”; in other words, they are pre-nuclear single obstruents (Clements 1990: 304). If the onset

The acquisition order of liquids in Spanish complex onsets

225

is maximized, then the ones that show the best profile consist of pre-nuclear obstruent+liquid combinations. The best rhyme, in turn, shows the best profile when it has a post-nuclear “gradual drop in sonority” or, said differently, when it ends in a single sonorant6. If the SDP, together with the hierarchy (6), capture the manner in which consonants order themselves with respect to a syllable peak in adults’ speech productions, it is reasonable to assume that children will tend to maximize the contrast between the syllable nucleus and its prenuclear marginal segment, thereby privileging the CV core syllable, while minimizing the contrast between the syllable nucleus and its postnuclear marginal segment, when there is one. Since liquids are simply too vowel-like to meet the SDP requirement when consonant simplification occurs in complex onsets, the normal tendency is for liquids to disappear while obstruents are usually preserved. It is not always the case that obstruent liquid clusters are reduced to C1, in a C1C2 sequence. Lleó and Prinz (1996) provided partial evidence showing that two of their four Spanish speaking children presented a higher percentage of cluster reductions to the most sonorous C2 segments rather than to the least sonorous C1. They document, for instance, that the target cluster in /floɾ/ ‘flower’ reduced to [l], [lojs] or [ˈlole:], or by the same token, /ɡlobo/ ‘balloon’ was produced as [ˈlobo]. A similar but sporadic output was produced for /ɡlobo/ in the present study, as Table 1 shows. Most studies on onset acquisition reviewed above discuss essentially the reduction of one of the two consonants. Kehoe et al. (2008) and Núñez Cedeño (2008), however, were first in addressing the accepted assumption that the appearance of stop+lateral and stop+rhotic happens simultaneously or categorically. Although the study by Kehoe et al. centered mainly on the phonological representation of CGV sequences in monolingual French and Spanish-speaking and German-Spanish bilingual children, they also raised the question of the acquisition order of obstruent+liquid clusters vs. the acquisition order of the CGV sequence, and further indicated that the obstruent+lateral cluster precedes the acquisition of the obstruent+rhotic sequence (2008: 26, 28–32). The authors, however, did not delve into the reasons why obstruent+lateral precedes obstruent+ rhotic in the development of children’s speech. This was in essence the focus of Núñez Cedeño’s (2008) research.

6 Contrarily, the least optimal profile for a single onset in a core syllable is one whose nucleus is headed by glide. For a complex onset, the least preferred combination begins with a liquid followed by glide. For a simple rhyme, the less optimal syllable has a nucleus, which ends in an obstruent. For complex rhymes the post-nuclear liquid+obstruent cluster is the least optimal (Clements 1990: 305).

226

Rafael A. Núñez Cedeño

Basing himself on longitudinal data from the acquisition of liquids in the speech of Seihla, he documented that the lateral /l/ is first to appear, followed by the simple rhotic /ɾ/, which manifests itself allophonically as a voiced dental stop or the approximant [ð], before being rendered as an actual rhotic (Núñez Cedeño 2008: 19). In the present study, we will assume the order of liquid acquisition reported therein. In that study, Núñez Cedeño also provided evidence which shows that the obstruent+lateral cluster is first to emerge. In light of these findings, he went on to suggest that the reason for this apparent order rests not solely on the intensity correlates of the segments in question, as Parker (2002) argues, nor in their distributional patterns, but on the rather simple axiom that /l/ is articulatorily easier to produce than the rhotic /ɾ/. Thus, in the present study, I take the stance that in order to produce speech sounds, the motion of neuromuscular fibers are activated by nerve impulses through an electrochemical reaction “which causes a brief twitch of the fibers. . . , and which consumes a quantity of the basic “fuel” of muscle tissue. . .” (Kirchner 1998: 37). The integrated interplay of electrochemical reactions with muscular fibers results in articulatory effort. To the extent that a given speaker minimizes the gestural effort required to articulate a given sound, a consonant in our case, then it is understood here that such a sound is easier to articulate than one that maximizes additional articulatory gestures. This assumption raises some important implications. On the one hand, it impacts the Universal Sonority Hierarchy (6), especially in its recast markedness constraint formulated by Pater and Barlow (2003), which makes no scalar distinction among liquids. On the other hand, it assumes ( Kirchner (1998:38) that the development of gestural efforts is inherently intertwined with the knowledge the speaker has about his or her phonological grammar. As stated in the Introduction, this study aims to expand on and support Núñez Cedeño’s (2008) original idea by asking the question: do children use laterals in onset clusters before they use rhotics in onset clusters? The fact that they do will be shown with a plethora of data coming from actual outputs produced by monolingual Dominican Spanish-speaking children.

4 Methodology Before presenting the analysis of the data, let us first discuss the methodology used to obtain them. The research included 12 Dominican Spanish-speaking monolingual children, 9 females and 3 males, whose age ranged between 2;0 to 3;2 years, with a mean age of 2;10.5. The fact that children were attending expensive preschools in exclusive neighborhoods, allowed me to categorize them

The acquisition order of liquids in Spanish complex onsets

227

broadly as belonging either to the middle or upper class. They all lived in the city of Santo Domingo and were enrolled in the preschools where all interviews took place. At first, the children were divided into four groups of three. Each child from each group was interviewed individually, one group per day, every two weeks, for two months. This division had a two-pronged purpose. First, it allowed the collection of consistent speech samples from each child at a given age. Second, a child was assigned to a specific group to facilitate communication and planning with staff on follow-up interviews. The children were also examined four times over a period of two months. One week before initiating the work session, I established rapport with the children by reading stories and playing games with them. This activity was carried out for two hours daily and was subsequently followed by the regular work sessions. A work session consisted of recording and eliciting children’s speech using images or objects presented on a computer; these represented 60 target words containing singleton and grouped consonants. Samples of both images and words are reproduced in the Appendix. These words were divided into two sets of 30 each. Each set was presented twice either in the first or in the second month. A typical interaction with a child consisted of, for example, showing him/her a picture of a tiger (the expected response is “tigre” in Spanish), and they were asked ¿conoces ese animal? ‘do you know that animal?’, ¿qué es? ‘what is it?’, ¿dónde lo has visto? ‘where have you seen it it?’ A unidirectional small Olympus DS-2 digital voice recorder was placed discreetly behind the computer, so as not to distract the child’s attention. Four children failed to respond to the images presented. In that case, through dialogs I elicited spontaneous productions to obtain the missing words. If the words contained grouped consonants, they were also included in the data because oftentimes some children skipped the stimuli which were presented to them.

4.1 Stimuli The two sets of recordings were examined for consistency of production. Only the last set with consonantal clusters, which exhibited the characteristics shown in (8), was included for the final analysis of the results. Josué’s output was an exception in that it was only obtained once because his parents left the city before completion of the research. The following illustrates representative examples of words with complex onsets, shown in boldface type, used in this study.

228

Rafael A. Núñez Cedeño

(8) Samples of complex onsets7 /pl/

/pr/

/bl/

/br/

/tr/

/dr/

ˈplátano

prinˈcesa

ˈbloque

ˈbruja

esˈtrella

draˈgón(ˈcito)

cumpleˈaños

presiˈdente

ˈblanco(gato)

somˈbrero

ˈtrompa

maˈdrina

pluˈmita

profeˈsora

biblioˈteca

liˈbreta

viˈtrina

paˈdrino

ˈplanta

exprimiˈdor

ˈmueble

alˈfombra

maesˈtra

laˈdrón

plaˈtillo

ˈprima

taˈblero

‘plantain’

‘princess’

‘block’

‘witch’

‘star’

‘(little)dragon’

‘birthday’

‘president’

‘white(cat)’

‘hat’

‘trunk’

‘godmother’

‘little feather’

‘professor’

‘library’

‘notebook’

‘glass case’

‘godfather’

‘plant’

‘juicer’

‘furniture’

‘rug’

‘teacher’

‘thief’

‘saucer’

‘cousin’

‘checkerboard’

/kr/

/gl/

/kl/

tracˈtor

‘tractor’ /gr/

/fl/

/fr/

ˈclavo

cruciˈfijo

ˈglobo

ˈgrande

ˈflecha

ˈfresa

ˈchicle

ˈlocrio

ˈregla

ˈtigre

ˈflorero

refrigeraˈdor

biciˈcleta

craˈyola

iˈglesia

ˈnegro

flamboˈyán

reˈfresco

chanˈcleta

escriˈtorio

ˈsangre

griˈtando

fluˈcito

fraˈnela

clip

escriˈbir

canˈgrejo

ˈflaco

ˈfruta friˈjol fregaˈdero

‘nail’

‘crucifix’

‘balloon’

‘big’

‘arrow’

‘strawberry’

‘chewing gum’

‘rice with meat’

‘ruler’

‘tiger’

‘vase’

‘refrigerator’

‘bicycle’

‘crayola’

‘church’

‘black’

‘a kind of tree’

‘soda pop’

‘slipper’

‘desk’

‘blood’

‘crying’

‘small suit’

‘undershirt’

‘clip’

‘to write’

‘crab’

‘skinny’

‘fruit’ ‘bean’ ‘sink’

7 I am making the assumption that voiced obstruents are stops at the lexical level (see also Martínez-Gil 2001). As is fairly well-known in Spanish phonology, the stops /b, d, ɡ/ become [b, d, ɡ] after a pause and after a nasal, and for /d/ also after a lateral. Elsewhere they are realized as the approximants [β, ð, ɣ]. All children adhered consistently to these patterns.

The acquisition order of liquids in Spanish complex onsets

229

As (8) shows, there were monosyllabic as well as polysyllabic words, which contained the grouped consonants both in word-initial and internal positions, followed or preceded by stressed or unstressed vowels of different quality.

4.2 Coding Children’s recorded outputs were transferred from the recorder to a computer program. All speech tokens extracted from the computer were transcribed phonetically by the researcher with the assistance of one graduate student, who specializes in phonology and phonetics and is a speaker of the dialect. The transcriptions were done independently of each other. The total number of tokens included in the research was 696. Reliability agreement on transcriptions for both transcribers was close to 96% , based on a total of 367 tokens for the second set, which is what I am using for final calculations, as noted above. Only instances of lexically unexpected outputs were discarded, as e.g., if trying to describe a sad face the child did not say triste ‘sad’ but instead produced llorando ‘crying’; this word did not have an onset cluster and thus it was discarded. However, if the child said e.g., planeta ‘planet’ instead of the expected globo, ‘balloon’, this word was included since it has an obstruent+lateral cluster, albeit not the expected one. Inaudible sounds or hard-to-understand productions were also discarded. Productions counted as correct were those containing full complex onsets, deletions of C1 or C2 in C1C2 as in [ˈpatano] for [ˈplatano] or [ˈlobo] for [ˈɡloβo] because of their obvious approximations to intended targets, epenthesis as in [niɣlesja] for [iˈɣlesja], substitutions of the glide [j] for liquids, as in [ˈpjuma] for [ˈpluma], sound substitutions, as in [fan̪ˈtuflas], [ˈtlen], [ˈneðɾo] for the respective [pan̪ˈtuflas], [ˈtɾen], [ˈneɣɾo], and deletion of non-targeted syllables, as in [ˈɡleta] or [ˈkleta] for [bisiˈkleta]. Cases of sound substitutions were considered for the analysis, although were not necessarily counted as correct, because children are attempting to produce those target-like segments. Other studies have explained why these substitutions occur, a topic that will not be pursued here (Morales-Front 1999, Barlow 2003, Pater and Barlow 2003).

5 Results In this section I am addressing the order of acquisition of consonantal clusters in onset position. Specifically, I am studying the obstruent+rhotic and obstruent+ lateral onsets, which first will be examined in terms of their general behavior, to

230

Rafael A. Núñez Cedeño

be followed by an analysis on how they pattern regarding their manner of articulation and voicing. In response to the question of whether children use laterals in onset clusters before they do for rhotics in the same environment, the main prediction is that an implicit order of development exists in which the former are acquired before the latter. A corollary of this hypothesis is that the development occurs gradually. Regarding the complex onset types /obs+rhot/ or /obs+lat/8, children generally produced the four different output types presented in Tables 1 and 2. In both Tables, the types [obs+Ø] and [obs+G] refer to a sequence of CC onsets where the second C, representing either the lateral or the rhotic, is deleted or changed into the glide [j]. In Table 1, the /obs+rhot/ onset is also realized as the respective [obs+rhot] and [obs+lat] cluster types, the latter suggesting that children transformed the rhotic input into a lateral output. Conversely, in Table 2, the input /obst+lat/ could be manifested phonetically as either an [obs+lat] or as an [obst+rhot] type, where the latter represents a change from a lateral input into a rhotic output. Because of their very sporadic realizations, the types in which the first C is deleted in CC groups were not included. Mauro, for example, provided one single instance in which [ɾ] is preserved, out of a total of 17 outputs. In a similar fashion, Josué, with 14 token productions, deleted only one initial C of such a group. As Table 1 shows, he transformed all rhotics into the lateral [l]. In Table 2 there were two separate instances of deletion of obstruents while the lateral consonant remains unaffected. These non-canonical deletions, by virtue of their rare occurrences, will not be counted in the calculations (cf. Kehoe et al. 2008). While Table 2 shows Camille and Susa evincing 100% productions of the [obs+rhot] type, the most remarkable pattern for most children is to delete or transform the input /ɾ/ into [l]. In contrast, Table 2 clearly demonstrates that, except for Kate, Jamie, and Maria, who averaged 28% deletions, most children tended to produce high percentages of [obs+lat] clusters, with four of them even reaching the 100% threshold. This table is in stark contrast to the outputs recorded in Table 1. Based on comparisons of the respective [obs+rhot] and [obs+lat] outputs for both input groups displayed in Tables 1 and 2, Table 3 shows the mean and standard deviation in percentages. Overall, this table establishes that children performed much better with [obs+lat] as opposed to [obs+rhot] types, with mean scores of 71% vs. 49%, 8 For the sake of expository convenience, the segment classes appearing within slashes and brackets are used here as mnemonic, abbreviatory devices to refer to phonological and phonetic representations, respectively.

The acquisition order of liquids in Spanish complex onsets

231

Table 1: Input complex onsets of /obs+rhot/ and percentages of output types for all children Input: /obs+rhot/ Subject

Age

# of tokens

[obs+Ø]

[obs+rhot]

[obs+lat]

[obs+g]

Camille Kate Jamie Diego Mauro Hanna Jade Maya Sofia Susa Maria Josue

2;6 3;0 2;11 3;0 2;0 2;1 3;2 2;1 3;2 3;2 2;0 2;7

20 18 17 20 16 8 19 19 14 19 19 13

0% (0) 17% (3) 24% (4) 45% (9) 25% (4) 100% (8) 16% (3) 26% (5) 14% (2) 0 26% (5) 38% (5)

100% (20) 83% (15) 0% (0) 20% (4) 50% (8) 0 42% (8) 42% (8) 79% (11) 100% (19) 68% (13) 0

(0) 0 29% (5) 35% (7) 25% (4) 0 42% (8) 32% (6) 7% (1) 0 5% (1) 62% (8)

0 0 47% (8) 0 0 0 0 0 0 0 0 0

Table 2: Input complex onsets of /obs+lat/ and percentages of output types for all children Input: /obs+lat/ Subject

Age

#of tokens

[obs+Ø]

[obs+lat]

[obs+rhot]

[obs+g]

Camille Kate Jamie Diego Mauro Hanna Jade Maya Sofia Susa Maria Josue

2;6 3;0 2;11 3;0 2;0 2;1 3;2 2;1 3;2 3;2 2;0 2;7

15 15 14 15 15 8 16 14 14 17 12 6

0 33% (5) 36 (5) 0 0 100% (8) 12% (2) 0 7% (1) 0 25% (3) 100% (6)

100% (15) 67% (10) 36% (5) 100% (15) 100% (15) 0 88% (14) 100% (14) 93% (13) 100% (17) 75% (9) 0

0 0 0 0 0 0 0 0 0 0 0 0

0 0 28% (4) 0 0 0 0 0 0 0 0 0

respectively. When a paired sample t-test is applied to both clusters in Tables 1 and 2, an important difference emerges between the outputs [obs+rhot] and [obs+lat], namely, that the latter is significantly more prevalent among the given subject sample, t (11) = –2.630, p < .05. Clearly, all children appear to have less difficulty in producing [obst+lat] than [obst+rhot] complex onsets, results that compare favorably with those reported by Kehoe et al. (2008: 26).

232

Rafael A. Núñez Cedeño

Table 3: Mean percentages and standard deviation of [obs+rhot] and [obs+lat] of unmodified output liquid Input/Output

Mean

N

Std. Deviation

[obs+rhot] [obs+lat]

49% 71%

12 12

.37 .38

Table 4: Samples Statistics of outputs for Pair 1: /obs+rhot/ and Pair 2: /obs+lat/

Pair 1 Pair 2

[obs+rhot] [obs+lat] [obs+rhot] [obs+lat]

Mean

N

Std. Deviation

49% 20% 0% 71%

12 12 12 12

.37 .20 .00 .38

Still, by taking a closer look at Tables 1–2, we can quantitatively determine the effects of changes that the inputs in question have undergone. Observe in Table 1 that the /obs+rhot/ input shows more noticeable variations among children than its counterpart /obs+lat/ in Table 2. It is apparent in the former that if the clustered /ɾ/ is to transform itself into a different consonantal segment, it will do so preferably toward [l]. Jamie and Josué may variably substitute the glide [j] for both liquid inputs, but the other children suppressed this process9. Note further that the grouped /l/ in Table 2 was consistently realized as [l]. For this cluster, there was no change gravitating toward [ɾ]. This becomes more apparent when we examine and compare the mean percentages for cluster types within and across group in both tables, as Table 4 shows. For Pair 1, the mean findings are clear: There was a higher percentage of retention of a rhotic, 49%, than conversion to a lateral, with 20%. Recall, however, that in Table 1 rhotic-deletion or conversion to [l] confirmed that these processes are highly favored by children. As for Pair 2, the [obs+lat] output is an obvious winner with 71% mean score in relation to 0% for lateral-to-rhotic conversion. Again, these results compare robustly with the outcomes observed in Table 2, which clearly show that for [obs+lat] inputs children favor lateral retention.

9 Since this is a cross-sectional study, it is difficult to characterize this suppression in terms of development, as correctly pointed out by Conxita Lleó (p.c.). It may be the case that these children had gone through a stage in which they already applied [Cj] or have not simply reached it yet.

The acquisition order of liquids in Spanish complex onsets

233

Thus far we have been able to assess the general behavior of liquids in complex onsets. Now let us turn to the analysis of these clusters according to two additional measures. That is, we want to find out whether the manner and voicing of initial obstruents may have an effect on the kind of changes the liquid might undergo. In other words, we seek to determine whether the initial obstruent plays a role in deleting the liquid or changing it into something else. We have already seen that /ɾ/ is typically converted into [l] or is deleted, and both /ɾ/ and /l/ may switch to [j]. Let us consider Table 5 and see how the segments are distributed in percentages. This table contains two sets. The upper rows for both sets of complex onsets refer to the changes they have undergone, which could be liquid deletion, /ɾ/ changing into [l], or substitution of a glide for a liquid. The far left column contains a breakdown of possible inputs, whether they are headed by voiced/voiceless stops or the fricative /f/, followed by a liquid. Table 5: Outcomes of [obs+liquid] cluster depending on the manner and voicing of the input obstruent

Vced+rhot Vless+rhot Vced+lat Vless+lat

Stop+ø

Changed to stop+lat/rhot

Changed to stop+glide

23% (21/90) 33% (21/63) 15% (8/55) 22% (14/64)

24% (22/90) 24% (15/63) 0% (0/55) 0% (0/64)

4% (4/90) 3% (2/63) 2% (1/55) 5% (3/64)

/f/+ø Vless+rhot Vless+lat

14% (4/28) 4% (1/28)

Outcomes of [f+liquid] clusters Changed to Changed to /f/+lat/rhot /f/+glide 21% (6/28) 0% (0/28)

4% (1/28) 7% (2/28)

For the first percentage set, children appear to perform differently with obstruent+rhotic clusters: When the onset is voiceless the rhotic is deleted more often. In relation to changing the rhotic into a lateral or a glide, both clusters behave equally well. A similar performance occurs when stops are followed by laterals: When the former are voiceless, the lateral is deleted more often than when they are voiced. In both cases, no lateral is ever substituted by [ɾ], though a change to a glide is marginally possible. The observed deletion patterns suggest that it is likely for a liquid to be deleted from a cluster when a voiceless stop precedes it. This is an unexpected result because in the sonority contour of two abutting consonants, a voiceless stop+liquid is much better than a voiced stop+liquid.

234

Rafael A. Núñez Cedeño

A more important comparison, however, is the set of stop-headed onsets in Table 5 compared to the fricative-headed set. The percentages of responses indicate a larger rate of liquid deletion for stop-headed onsets. This conforms to the sonority hierarchy requirements, where a syllable starting with a stop is preferred over one beginning with a fricative. Such preference dovetails neatly with Martínez-Gil’s (2001) generalization about the phonotactics of complex onsets, which states that two consonants must differ maximally in their sonority rank10. It seems, on the other hand, that there is no significant difference when converting the rhotic into a lateral in stop-headed as opposed to fricativeheaded onsets, the difference being a mere 3% favoring the former. In summary, the major distinction between the sets observed in Table 5 is that once again the percentages indicate that when the complex onset is retained, the rhotic input is quite prone to be produced as a lateral output.

6 Discussion Overall, the results above appear to be consistent with the prediction that /obs+lat/ onsets are easier to produce by children. The reason why this cluster is easier to articulate than /obs+rho/ appears to stem from co-articulation factors. For a child who is acquiring his/her phonology, it will be easier to articulate a complex onset containing two segments that involve complementary gestures of production, such as the combination stop+lateral. While stops may have ballistic movements during their articulation, the contact for a lateral is prolonged, rather than momentary and ballistic-like, as expected in the articulation of a tap (Catford 1977:130). Such co-articulatory effects of hit-and-run motions for each member of the [obs+rhot] sequence may be a very complex, dynamic attribute for a child to handle in the early stages of language acquisition. The natural default for the child is to fall back to a lateral, which lacks the rhotic’s inherent ballistic motion; in other words, the production of the lateral requires minimization of gestural activities. Therefore, and regarding the contrasting percentages of 71% vs. 49% noted above, the inference could be made that /l/ is acquired before /ɾ/. This assumption is confirmed in Núñez-Cedeño’s (2008) study in which of the two liquids the singleton /l/ was shown to be acquired first. 10 In this study, Martínez-Gil argues that obstruent stops and the fricative /f/ lack the [+continuant] specification. Thus they may freely combine with liquids in onsets, except that dental stop+lateral clusters are disallowed by a filter. It appears that this filter must be relaxed, because unlike Castilian Spanish where it is enforced, Dominican Spanish, as noted in section 2, allows only initial /tl/ onsets while disallowing /dl/.

The acquisition order of liquids in Spanish complex onsets

235

6.1 Theoretical consequences on the ordering of liquids In light of the existence of the alternations [obs+lat] and [obs+rhot] from the single input /obs+rho/, which points toward asymmetrical results in relation to the /obs+lat/ sequence, two reasonable assumptions have been made. First, there is an order in acquiring complex onsets, where [obs+lat] precedes [obs+ rhot], which the data and results seem to corroborate. Second, from this ordering it can be inferred that [obs+lat] and [obs+rhot] cluster types develop gradually. Now, a study that includes cross-sectional data may not be the most suitable approach for providing a definitive and reliable answer for confirming a gradual development and order of acquisition hypothesis over a categorical one. A firmer basis for such a conclusion has been shown to come from observations made with data over a long period of time11. In effect, and as noted above, in analyzing the longitudinal data of Seihla’s phonological development, Núñez Cedeño (2008) argued and provided evidence for an order of development of liquids in which the lateral /l/ was shown to precede the rhotic /ɾ/. To explain this hypothesis, and following other arguments which come from the behavior of both liquids in cross-linguistic12 and acquisition studies, he adopted the rather simple assumption that the lateral is articulatorily easier to produce than a tap, for the reasons given therein (Núñez Cedeño 2008: 20–21), and at present summarized in section 5. Having established this order, he went on to demonstrate that the stop+lateral combination is acquired before stop+rhotic clusters. Given the order of acquisition of /l/ over /ɾ/, Núñez Cedeño (2008) also argued that while some researchers claimed that the Universal Sonority Hierarchy (6) was based on sonority markedness, (Pater and Barlow 2003), it needed to be refined to capture markedness distinctions among liquids, as their scale makes no distinctions in terms of articulatory efforts. As a consequence, (6) encapsulates neither the presence of obstruent+lateral emerging before obstruent+rhotic clusters in these children’s speech nor the development of similar sequences in other Spanish dialects and Romance languages. Arguing from an OT perspective, Núñez Cedeño (2008) then proposed that the early development of liquids 11 The longitudinal study by Lleó (this volume) on the resyllabification in Spanish and German is yet another fine empirical example for testing acquisition hypotheses. 12 Additionally, various cross-linguistic investigations have shown that there is a preferentially ordered sequence for acquiring liquids. In his review of phonological acquisition studies, in particular on Templin’s (1957) study, Ingram (1989: 364) concludes that the /l/ of English is acquired earlier than the /ɹ/. Smith’s (1973: 13–22) analysis of the longitudinal data of his son’s grammar provides evidence showing that /ɹ, l/ neutralize to /l/ and not the other way around. Adult Maya Quiche grammar contains both types of liquids; however, when children acquire the language their early output exhibits only laterals (Pye, Ingram, and List 1987).

236

Rafael A. Núñez Cedeño

is possibly banned by the markedness constraints *rhotic-ONS and *lateral-ONS , which disallow a rhotic or a lateral in onset positions, respectively. The proposed sonority hierarchy in (9), which subsumes the articulatory difficulties alluded to above, incorporates the sonority-based hierarchy (6) for consonants only13. (9) *glide-ONS >> *rhotic-ONS >> *lateral-ONS >> *nasal-ONS >> *fricative-ONS >> *stop-ONS Hierarchy (9) says that there is a preference in markedness for less sonorous sounds in onsets. Put differently, (9) predicts that an onset beginning with a rhotic is worse off than one starting with a lateral.14 Such ranking would explain why children generally do not have early rhotic productions, simply because their presence would violate the high-ranked *rhotic-ONS constraint15. On the other hand, when liquid acquisition kicks in, the first to appear is /l/ in singleton onsets 13 In support of this ranking, Parker (2002: 240) offers a final, phonological version of a Universal Sonority Hierarchy in which laterals are more sonorous than taps and trills (cf. Selkirk 1984), who provided a provisional, opposite ranking for these segments). 14 Foley (1977: 38) observes that during the evolution of Spanish, initial /ɾ/ strengthened to /r/ but not initial /l/. Similarly, strengthening occurs for /l/, which changes to /r/ after /ɾ/ in Dutch, Norwegian, and Greek, all of which indicate that phonologically /ɾ/ is stronger or more sonorous than /l/. In light of these phonological tendencies, Foley proposes that /ɾ/ is stronger than /l/. Though Foley dismisses phonetic measurements as a basis for ranking liquids, Parker’s phonetic correlates as well as phonological evidence provide a firm basis for sustaining that laterals outrank consonantal rhotics in sonority. We have also argued that the acquisition literature points in this direction. A good example is Barlow’s study (2003, footnotes 11, 15) in which she also documents the late emergence of rhotics as opposed to /l/. This led her to posit a high-ranked *VIBRANT constraint in children’s speech, though she stops short of articulating reasons for this order. 15 For Bonet and Mascaró (1997), this constraint would need further refinements but in opposite directions. These researchers analyzed the representation and distribution of /r/ and /ɾ/ in the onset for Spanish and Catalan and argued that their phonetic realizations may be driven by sonority effects. They say that since the trill occurs in syllable initial position, like obstruents, while the tap is the only one that appears as the second segment of a complex onset, like laterals and glides, it is reasonable to suggest that the former is less sonorous than the latter, provided that the trill is assumed to be the unmarked case. Following this line of reasoning, Bonet and Mascaró propose a sonority scale in which the trill has a low sonority value while the tap’s is high. This would mean that (9) would have to be modified so that the tap patterns with glides and the trill with stops. The implication for such a scale for the acquisition of rhotics in single and complex onsets is that the trill would most likely surface early in children’s speech because they pattern with voiceless stops, which, as is fairly well-known in language development studies, tend to occur early. The evidence provided in the present study and other acquisition research show this not to be the case (see footnotes 8 and 9).

The acquisition order of liquids in Spanish complex onsets

237

and in [obs+lateral] before [obs+rhotic] clusters develop precisely because *lateral-ONS is low-ranked, and thus subject to minimal violation.

7 Conclusion Before Núñez-Cedeño (2008) and Kehoe et al. (2008), the majority of research on onset acquisition has assumed that the appearance of stop+lateral and stop+ rhotic clusters occurs gradually or categorically. In the current study, the evidence adduced has indirectly shown that this is not the case. It has been argued that children tend to show considerable variation in mastering the sequences [obs+lat] and [obs+rhot]. The data have shown that it is more likely for a child to realize the /obs+rhot/ sequence as [obs+lat] or [obs+Ø] than to convert the /obs+lat/ sequence into a [obs+rhot]. To change a rhotic for a lateral is not unique to the children for the current study. For children learning other Spanish dialects and Romance languages, the acquisition research literature also supports the argument that taps and trills in non-branching onsets are generally substituted with [l] (see Barlow 2003, 2005 and the references cited therein). Furthermore, I attempted to demonstrate whether the changes that the rhotic experimented in /obs+liquid/ sequences were driven by the initial obstruent. The findings revealed that neither the manner nor the voicing of the obstruent seems to play a role in the altered rhotic. I was able to determine, however, that a stop-headed cluster was more likely to have a changed rhotic than a fricative-headed one. The reason for this preference, though minimal, lies in the fact that in complex onsets there is greater distance in the slope of sonority between a stop and a liquid than between a fricative and a liquid, a fact that is captured both in (6) and in the markedness constraint hierarchy in (9). The prediction that children tend to effect fewer changes to liquids headed by stops as opposed to those headed by fricatives holds marginally at best with the set of data analyzed. Absent from the findings is the role that stress may play in triggering the occurrence of liquids in these complex onsets. A general survey of the data suggests that it plays no role: Change is equally likely in stressed and unstressed syllables. This appears to run counter to positional faithfulness constraints which could conceivably target stressed syllables and favor cluster retention more so than in unstressed syllables (Beckman 1997: 2). Still, there is a need to determine what their behavior would be in relation to the distance from the stressed syllable, that is, whether a cluster which is more distant from a stressed syllable is more likely to effect a change on the liquid in one direction or the

238

Rafael A. Núñez Cedeño

other. At first blush, one would suspect that since the null hypothesis rests on articulatory efforts, distance from a stressed syllable would not affect the outcome. An analysis that contains complex onsets in words with three or more syllables would test such hypotheses. Finally, the clusters for /f/+liquid in (8) are mostly represented with wordinitial tokens. Future research would have to control for non-initial clusters of this type. I venture to speculate, however, that such findings would unlikely invalidate the results because the other clusters analyzed here, in spite of being more equally distributed, appear not to have negative effects.

References Anderson, John M. & Colin J. Ewen. 1987. Principles of dependency phonology. Cambridge: Cambridge University Press. Barlow, Jessica A. 2003. Asymmetries in the acquisition of consonant clusters in Spanish. Canadian Journal of Linguistics/Revue Canadienne de Linguistique 48: 179–210. Barlow, Jessica A. 2005. A phonological change and the representation of consonant clusters in Spanish: A case study. Clinical Linguistics and Phonetics 19:659–679. Barlow, Jessica A. 2007. Constraint conflict in the acquisition of clusters in Spanish. In Fernando Martínez-Gil & Sonia Colina (eds.), Optimality theoretic studies in Spanish phonology, 525–548. Amsterdam & Philadelphia: John Benjamins. Beckman, Jill. 1997. Positional faithfulness, positional neutralization, and Shona vowel harmony. Phonology 14:1–46. Bradley, Travis. 2006. Spanish rhotics and Dominican hypercorrected /s/. Probus 1–53. Blevins, Juliette. 1995. The syllable in phonological theory. In John A. Goldsmith (ed.), The handbook of phonological theory, 206–244. Cambridge, MA: Blackwell. Bonet, Eulalia & Joan Mascaró. 1997. On the representation of constrasting rhotics. In Fernando Martínez-Gil & Alfonso Morales-Front (eds.), Issues in the phonology and morphology of the major Iberian languages, 127–150. Washington, D.C.: Georgetown University Press. Catford, J.C. 1977. Fundamental problems in phonetics. Bloomington & London: Indiana University Press. Clements, N. G. 1990. The role of the sonority cycle in core syllabification. In John Kingston & Mary E. Beckman (eds.), Papers in laboratory phonology 1. Between the grammar and physics of speech, 283–333. Cambridge: Cambridge University Press. Colina, Sonia. 2009. Spanish phonology: a syllabic perspective. Washington, D.C.: Georgetown University Press. D’Introno, Francesco, Enrique del Teso, & Rosemary Weston. 1995. Fonética y fonología actual del español. Madrid: Cátedra. Fikkert, Paula. 1994. On the acquisition of prosodic structure. Leiden: University of Leiden dissertation. Foley, James. 1970. Phonological distinctive features. Folia Linguistica 4:98–92. Foley, James. 1977. Foundations of theoretical phonology. Cambridge: Cambridge University Press.

The acquisition order of liquids in Spanish complex onsets

239

Freitas, M. Joao. 2003. The acquisition of onset clusters in European Portuguese. Probus 15:27– 47. Gnanadesikan, Amalia. 2004. Markedness and faithfulness constraints in child phonology. In René Kager, Joe Pater & Wim Zonneveld (eds.), Constraints in phonological acquisition, 73–107. Cambridge: Cambridge University Press. Grijzenhout, Janet & Sandra Joppen. 2002. The lack of onsets in German child phonology. In Ingeborg Lasser (ed.), The Process of language acquisition, 319–39. Frankfurt & Berlin Peter Lang Verlag. Guitart, Jorge M. 1999. Sentido y sonido. Washington, D.C.: Georgetown University Press. Hammond, Robert M. 2001. The sounds of Spanish: analysis and application. Somerville, MA: Cascadilla Press. Harris, James W. 1983. Syllable structure and stress in Spanish: a nonlinear approach. Cambridge, MA: The MIT Press. Hualde, José Ignacio. 1991. On Spanish syllabification. In Héctor Campos & Fernando MartínezGil (eds.), Current studies in Spanish linguistics, 475–493. Washington, D.C.: Georgetown University Press. Hualde, José Ignacio. 2005. The sounds of Spanish. Cambridge & New York: Cambridge University Press. Ingram, David. 1989. First language acquisition. Cambridge: Cambridge University Press. Kenstowicz, Michael. 1994. Phonology in generative grammar. Cambridge, MA: Blackwell. Kehoe, Margaret, Geraldine Hilaire-Debove, Katherine Demuth & Conxita Lleó. 2008. The structure of branching onsets and rising diphthongs: evidence from the acquisition of French and Spanish. Language Acquisition 15.5–57. Kirchner, Robert Martin. 1998. An effort-based approach to consonant lenition. UCLA dissertation. Lleó, Conxita & Michael Prinz. 1996. Consonant clusters in child phonology and the directionality of syllable structure assignment. Journal of Child Language 23:31–56. Martínez-Gil, Fernando. 2001. Sonority as a primitive phonological feature. In Julia Herschensohn, Enrique Mallén & Karen Zagona (eds.), Features and interfaces in Romance, 203– 222. Amsterdam & Philadelphia: John Benjamins. Morales-Front, Alfonso. 1999. The roles of templates in the acquisition of phonology. In Javier Gutiérrez-Rexach & Fernando Martínez-Gil (eds.), Advances in Hispanic linguistics, 72–98. Somerville, MA: Cascadilla Press. Núñez Cedeño, Rafael A. 1985. On the three-tiered syllabic theory and its implications for Spanish. In Larry D. King & Catherine A. Maley (eds.), Selected papers from the XIIIth Linguistic Symposium on Romance languages, 261–286. Amsterdam & Philadelphia: John Benjamins. Núñez Cedeño, Rafael A. 1994. The alterability of Spanish geminates and its effects on the Uniform Applicability Condition. Probus 5:3–19. Núñez Cedeño, Rafael A. 2008. On the acquisition of Spanish onsets. Southwest Journal of Linguistics 27.1–30. Ohala, Diane K. 1999. The influence of sonority on children’s cluster reductions. Journal of Communication Disorders 32:397–422. Parker, Stephen. 2002. Quantifying the sonority hierarchy. Amherst, MA: University of Massachusetts dissertation. Pater, Joe & Jessica A. Barlow. 2003. Constraint conflict in cluster reduction. Journal of Child Language 30:487–526.

240

Rafael A. Núñez Cedeño

Pye, C., David Ingram & H. List. 1987. A comparison of initial consonant acquisition in English and Quiche. In K. E. Nelson and A. van Kleeck (eds.), Children’s Language, vol. 6, 175–90. Hillsdale, New Jersey: Earlbaum. Rose, Yvan. 2000. Headedness and prosodic licensing in the L1 acquisition of phonology. Montreal: McGill University dissertation. Selkirk, Elizabeth. 1984. On the major class features and syllabic theory. In Mark Aronoff & Richard T. Oehrle (eds.), Language sound structure, 107–136. Cambridge, MA: The MIT Press. Templin, M. 1957. Certain language skills in children. University of Minnesota Institute of Child Welfare (Monograph Series 26). Minneapolis: University of Minnesota Press.

Appendix Picture Samples and Spanish Consonantal Clusters Each subject was presented with drawings or pictures of the objects listed under each consonant cluster below. Then they were asked: ¿Qué es esto? (‘What is this?); ¿En dónde lo ves o se encuentra? (‘Where do you normally see it or where is it normally found?’). Questions were followed by a brief dialog with the subject regarding the object being described. Picture samples

241

The acquisition order of liquids in Spanish complex onsets

PL

PR

BL

BR

TR

DR

plátano

princesa

bloque

bruja

estrella

dragón(cito)

cumpleaños

presidente

blanco(gato)

sombrero

trompa

madrina

plumita

profesora

biblioteca

libreta

vitrina

padrino

planta

exprimidor

mueble

alfombra

maestra

ladrón

platillo (volador)

prima

tablero

alfombra (mágica)

tractor

‘plantain’

‘princess’

‘block’

‘witch’

‘star’

‘(little)dragon’

‘birthday’

‘president’

‘white(cat)’

‘hat’

‘trunk’

‘godmother’

‘little feather’

‘professor’

‘library’

‘notebook’

‘glass case’

‘godfather’

‘plant’

‘juicer’

‘furniture’

‘(magic) carpet’

‘teacher’

‘thief’

‘(flying) saucer’

‘cousin’

‘checkerboard’

CL

CR

‘tractor’

GL

GR

FL

FR

clavo

crucifijo

globo

(casa)grande

flecha

fresa

chicle

locrio

regla

tigre

florero

refrigerador

bicicleta

crayola

iglesia

negro(perro)

flucito

refresco

chancleta

escritorio

sangre

flaco

franela

clip

escribir

gritando

flamboyán

fruta frijol fregadero

cangrejo ‘nail’

‘crucifix’

‘globe’

‘big house’

‘arrow’

‘strawberry’

‘chiclet’

‘rice with meat’

‘ruler’

‘tiger’

‘vase’

‘refrigerator’

‘bicycle’

‘crayola’

‘church’

‘black’

‘a small suit’

‘soda pop’

‘slipper’

‘desk’

‘blood’

‘skinny’

‘undershirt’

‘clip’

‘to write’

‘crying’

‘a kind of tree’

‘fruit’ ‘bean’

‘crab’

‘sink’

III The Syllable and the Notions of Stress and Weight

John M. Lipski

Spanish vocalic epenthesis: the phonetics of sonority and the mora Abstract: This study presents a model of Spanish vocalic epenthesis based on a combination of observed and elicited data. It is proposed that the presence or absence of a mora dictates the position and phonetic realization of “repair” vowels. When faced with impermissible onset and coda clusters, e.g. in attempting to pronounce foreign words, Spanish exhibits a number of syllabic repair strategies involving epenthesis of /e/ or insertion of a ghost (svarabhakti) vowel. It is proposed that the unifying predictor of the position and phonetic nature of epenthetic vowels is the presence/absence of a mora. Since epenthetic vowels do not add phonological weight, i.e. carry no mora, a full vowel added during the repair of non-permissible clusters must acquire its mora through licensing. The only available licenser is an unsyllabified moraic consonant. Svarabkahti vowels cannot acquire a mora, hence their fleeting nature and the fact that they are typically not lexicalized.

1 Introduction The Spanish syllable has figured prominently in linguistic analyses, including structuralist, generative, and Optimality Theory approaches (e.g. taking as landmarks Granda 1966, Harris 1983, Colina 2006). The fundamental consonant patterns can be described in straightforward fashion. Any Spanish consonant can occur as a simple onset, while Spanish onset clusters have a maximum of two elements, the first of which must be a stop or the fricative /f/, and the second of which must be a liquid (/l/ or /ɾ/);1 the clusters *tl- and *dl- are disallowed. Word-final coda consonants are limited to the set {/d/, /s/, /l/, /ɾ/, /n/, and marginally /x/}, but word-internally (as well as word-finally in foreign borrowings), any Spanish consonant except for /x/ and the palatals /j/, /ɲ/, and /ʎ/ can appear in the coda. Coda clusters in patrimonial Spanish words only appear word-internally, and take the form /-Cs/, where the first consonant comes from the set {/b/, /d/, /l/, /ɾ/, /n/}. Whereas most of the research has been directed at 1 Under the assumption that semivocalic elements in rising diphthongs such as bueno ‘good’ and diez ‘ten’ are located in the nucleus (e.g. Harris 1983: 25). John M. Lipski, The Pennsylvania State University

246

John M. Lipski

characterizing actually occurring syllable types, there has been a complementary effort to account for non-occurring combinations, typically invoking sonority sequencing as the primary well-formedness criterion. Less mention is made of repair strategies involving non-permissible clusters, except for the omnipresent addition of prothetic /e/ to word-initial /sC-/ clusters in foreign borrowings or Spanish-influenced interlanguage: ski > esquí. In the contemporary Spanishspeaking world, there is an ongoing process of phonotactic adaptation of foreign words containing phonotactically non-permissable clusters, sometimes only temporarily as the result of transitory current events or attempts to speak another language, and sometimes more permanently; syllabic repair strategies are commonly applied in such cases. The present essay offers a broader analysis of Spanish consonant-cluster repair strategies involving vocalic epenthesis, addressing both the placement of the epenthetic vowel and its phonetic and phonological content. The following facts are drawn together: (1) Word-initial /sC-/ clusters are repaired by adding /e/ before the /s/ (stop > estop) rather than between /s/ and the following consonant (*setop). (2) The same occurs with word-initial clusters of clearly decreasing sonority (/mba-/ > emba not *meba). (3) Word-initial clusters of equal sonority or adjacent points on the sonority scale (e.g. ps-, kn-) do not add an epenthetic /e/ either before the first consonant or between the consonants but rather are realized with a fleeting schwa-like svarabhakti vowel between the consonants. (4) Word-final consonant clusters of increasing sonority are repaired by inserting /e/ between the consonants rather than following the consonants, even when a permissible word-initial cluster would occur (Sadr > Sader, not *Sadre). It will be proposed that relative sonority in consonant clusters is intimately related to the mora, which has most often been evoked to account for stress assignment and the minimal prosodic word. Rather than defining the mora epiphenominally (e.g. as indicated by the presence of a long vowel, a geminate consonant, or a stress-attracting syllable), a sonority-based approach will be presented, together with an empirical phonetic correlate of the presence or absence of a mora: the placement and realization of full epenthetic or svarabhakti vowels. The remainder of the study is organized as follows. The treatment of word-initial /sC-/ clusters in terms of sonority is combined with a discussion of putative violations of preferred syllable configurations. This is followed by a discussion of other word-initial and word-final clusters that violate sonority-sequencing requirements for Spanish, and which are repaired by vocalic epenthesis when borrowed or when Spanish speakers attempt spontaneous pronunciation. The

Spanish vocalic epenthesis

247

mora is then defined in terms of relative sonority, and the ensuing model is given an Optimality Theory demonstration. An appendix contains a brief description of the phonetics of vocalic epenthesis as a repair strategy in Spanish.

2 Epenthesis in Spanish: /s/ + C clusters and sonority sequencing The role of sonority in Spanish syllable formation is widely accepted, and forms the core of most proposals for syllabification and syllable well-formedness (e.g. Colina 2006, 2009; Harris 1983, 1989; Hualde 1991, 1999; Morelli 2003, Shepherd 2003). In Spanish syllables, sonority rises strictly from onset to nucleus, and generally falls strictly from nucleus to coda (with the special case of complex codas, all of which occur word-internally in patrimonial Spanish words and have /s/ as the second element). Assuming the general sonority hierarchy OBSTRUENT Carlitos. CONTIG by itself does not appear to provide sufficient justification for the choice of prothetic esC- rather than epenthetic *seC- as a Spanish repair strategy. A revised OT model will be proposed in a following section. There is considerable cross-linguistic evidence, including data from first language acquisition and language disorders, that in word-initial /s/ + C clusters the /s/ is frequently not part of a complex onset, but may be part of an appendix (references summarized in Goad & Rose 2004; also Davis 1990), or in Government Phonology linked to a preceding empty nucleus (Kaye 1992). This is true even in clusters with indisputably rising sonority, such as /s/ + SONORANT. To explain Spanish prothetic /e/ simply in terms of the appendix status of clusterinitial /s/ is circular, since the putative appendix status of /s/ is based precisely on the existence of phenomena such as epenthesis, syllabification, and placeidentity constraints. In the search for a more enlightening solution, it is useful to examine other configurations that trigger epenthesis in Spanish. In this fashion it may be possible to situate the “special” status of /s/ + C clusters within a more general model.

4 Epenthesis in Spanish word-final complex codas Unlike, e.g., Portuguese, Spanish freely tolerates foreign words with non-Spanish single codas (e.g. club, grog, input, internet, Facebook); although these consonants may be reduced or deleted in casual speech (presumably due to the markedness of these coda consonants), epenthetic vowels are not added. Similarly, Spanish tolerates complex codas not conforming to the Spanish /-Cs/ configuration, as long as they exhibit decreasing sonority, as in York, comfort, Grant, sometimes with deletion of the final consonant in colloquial speech but never with epenthesis (e.g. Bonet 2006: 317). In contrast to onset position, there is no strong Minimal Sonority Condition for complex codas, merely the requirement that sonority decrease monotonically (respecting, e.g. Clements’ 1990 Dispersion Principle: maximize sonority difference in the onset and minimize in the coda). The 4 Strictly speaking this statement is not quite accurate, since deletion of some word-internal coda consonants commonly occurs in most varieties of Spanish.

250

John M. Lipski

many Catalan surnames such as Front, Brucart, Guitart, Gelabert found throughout the Spanish-speaking world do not produce the phonotactic unease provoked by initial sC- clusters. In coda position /s/ does not appear to be more sonorous than other obstruents. To the extent that Spanish speakers actually pronounce both segments in foreign words with complex codas, there is no observable tendency to insert epenthetic vowels either in /-sC/ codas or in /-Cs/ codas: neither English box [baks] nor ask [æsk] triggers epenthesis in the L2 English of Spanish speakers. Even complex codas containing two non-sibilant obstruents do not normally provoke spontaneous epenthesis: soft [sɔft] and act [ækt] can be pronounced by most Spanish speakers without phonotactic distress. The variability of reduction in coda clusters of non-increasing sonority stands in contrast to the exceptionless nature of word-initial epenthesis. The same does not hold for word-final complex codas that exhibit rising sonority. In such cases, an epenthetic vowel (usually the Spanish default vowel [e]) is inserted between the two coda consonants, even when the coda cluster could occur as a Spanish complex onset. Thus, for example, the Arabic name al Sadr (which became prominent during the second Gulf War) is typically pronounced [al.sa.ðeɾ] rather than *[al.sa.ðɾe], despite the fact that /dɾ-/ is a perfectly acceptable Spanish onset cluster. Other examples (from Bonet 2006: 318) include single > [siŋ.ɡel] ‘a single-play record,’ Lidl > [li.ðel] ‘name of a supermarket chain,’ and numerous spontaneous pronunciations of English words ending in OBSTRUENT + (SYLLABIC ) LIQUID, such as table > [te.βel], finger [fiŋ.ɡeɾ], apple [a.pel], and so forth.5 The position of epenthetic vowels in wordfinal clusters of rising sonority appears at first glance to be a counterexample to Spanish syllabification algorithms that maximize onsets rather than codas (Harris 1983, 1989; Hualde 1991, 1999) as well as a counterexample to contiguity. Epenthesis also occurs – both in momentary approximations to foreign words and in assimilated loans – when accommodating words ending in complex codas with rising sonority that do not represent possible Spanish onset clusters, such as Arabic Ibn [i.βen] and the Spanish-influenced realization of English words ending phonetically in syllabic consonants, e.g. the pronunciation of happen as [xa.pen]. In such cases, the epenthetic vowel could in theory be inserted after the two consonants (*[iβ.ne], *[xap.ne]) but this does not occur. Epenthesis in word-final coda clusters does not respect contiguity, which casts additional doubt on a contiguity-based explanation for esC- rather than *seC in repair of /sC-/ onset clusters. An avoidance of complex onsets could in principle be invoked in those instances where the coda cluster is a possible Spanish onset 5 In all of these instances except for the cluster /dl/, the OBSTRUENT + LIQUID groups form possible onsets in Spanish.

Spanish vocalic epenthesis

251

(Bonet 2006: 336 considers but rejects this possibility), but Spanish shows no other signs of avoiding complex onsets.6 Bonet (2006) analyzes cases such as Sadr > [sa.ðeɾ] in terms of alignment constraints, i.e. maintaining the alignment of the right edge of the syllable with the right word boundary. Since there are no productive epenthesis processes in Spanish that target coda clusters with falling sonority, the alignment approach produces the correct answers with no counterexamples. At the same time, this proposal yields the incongruous result that syllable-word boundary alignment is highly ranked at the end of the word, but is easily over-ridden word-initially, as evidenced by the treatment of /sC-/ onset clusters.

5 More on Spanish word-initial epenthesis: decreasing sonority onset clusters A look at the phonotactic accommodation of other non-permissible onset clusters in Spanish suggests alternatives to both alignment and contiguity as the ultimate motivating force behind the placement of epenthetic vowels in Spanish. In addition to the well-documented treatment of words beginning with sC- clusters, there are other instances of epenthetic vowels in Spanish used to repair nonpermissible complex onsets, and which argue against the “special” status of /s/ in onset clusters. Consider first onset clusters that exhibit monotonically decreasing sonority, similar to sC- onsets. A typical case occurs with prenasalized obstruents, which have come into contact with Spanish at various points in its history. The most extensive contacts involved African languages, and there is indirect evidence that Africans who acquired Spanish as a second language under conditions of forced servitude (first in Spain, later in Spanish America) prenasalized word-initial voiced obstruents (Lipski 1992) as well as retaining African lexical items containing prenasalized consonants. The Afro-Iberian creole language Palenquero spoken in San Basilio de Palenque, Colombia, bears witness to this process, with Spanish-derived words such as ndo < dos ‘two,’ ngande < grande ‘large,’ ndulo < duro ‘hard,’ mbulo < burro ‘donkey,’ etc. (Friedemann &

6 Colina (2006) does analyze the lack of resyllabication in combinations such as club latino ‘Latin club’ as *clu-bla-ti-no not in terms of *COMPLEX ONSET but rather as a manifestation of alignment constraints. For onset clusters beginning with a voiced obstruent, Spanish has a history of occasionally resyllabifying the obstruent into the coda of the preceding syllable, thence into the nucleus as a glide, e.g. pa-dre > pai-re ‘father’ (Lipski, 1994; Martínez-Gil 1996, 1997; Piñeros 2001).

252

John M. Lipski

Patiño Rosselli 1983: 99–100). Literary texts from early modern Spain and Latin America contain imitations of Africans’ halting attempts at speaking Spanish, and what were probably prenasalized consonants were perceived as containing a prothetic /e/, or in a few rare cases, /a/: (1)

No ensa (< sa) discreto ‘[we] are not discreet’ (Lope de Vega 1894: 363; ‘El santo negro Rosambuco’) [Spain, 17th century] gente embrancas (< branca) ‘white people’; Estornudar gente enblancas (< blancas) ‘white people are sneezing’ (Andrés de Claramonte, El valiente negro en Flandes, Claramonte 1951) [Spain, 17th century] si el cuerpo tenemo enpreto (< prieto/preto) ‘if our bodies are black’ (Vélez de Guevara, El negro del serafín, Sánchez 1979) [Spain, 17th century] Bailar como un andimoños (< demonio) ‘to dance like a demon’ (Lope de Vega “La madre de la mejor”; 1893: 368) [Spain, 17th century] mi pecho está girviendo como agua que pela engallina (< gallina) ‘my chest is boiling like water that scalds chickens’ (Benítez del Cristo, Los novios catedráticos (1930); [Cuba, 19th century] si cabeza m’enduele (< me duele) bamo la casa Mundo ‘if my head hurts, let’s go to Mundo’s house’ (Cabrera 1971: 517) [Cuba, early 20th century] aprende a mandá primero que a engoberná (< gobernar) ‘learn to rule before governing’ (Cabrera 1970) [Cuba, early 20th century)

African words, mostly from Bantu languages, borrowed into Afro-Cuban cults, often added a prothetic /e/: mbala > embala ‘boniato,’ ndoki > endoki ‘witch doctor,’ nkento > enkento ‘wife,’ nganga > enganga ‘witchcraft,’ mbok > emboco ‘wrestler, executioner’ (García González & Valdés Acosta 1978: 21; Núñez Cedeño, Alúm, & Nodal 1985: 270; also Núñez Cedeño 1988: 150). As in the previous examples, when contemporary Spanish speakers are faced with foreign words beginning with prenasalized consonants the phonotactic resolution almost always involves a prothetic [e], e.g. Nkrumah [eŋ.kɾu.ma] (surname of the first president of Ghana), Nguema [eŋ.gé.ma] (surname of the current president of Equatorial Guinea), etc. When Spanish speakers with no knowledge of Guaraní read Paraguayan literature in which Guaraní lexical items are commonly inserted, a prothetic vowel is also pronounced in words like mboy [em.boj] ‘how much.’ The Guaraní variety spoken in the indigenous communities of Misiones province in Northeastern Argentina is Mbyá [mbɯ.á], which is pronounced by monolingual Spanish speakers as [em.bi.á]. In phonotactic repairs of prenasalized consonants (implicitly analyzed by Spanish speakers

Spanish vocalic epenthesis

253

as NASAL + OBSTRUENT clusters), continguity is respected, but alignment is not: prenasalized consonants are never incorporated through the addition of an epenthetic vowel between the two “halves”: *meb-, *ned-, etc. A related case involves word-initial (preconsonantal) syllabic consonants. In the traditional Spanish dialect of northern New Mexico, syllabic sonorants occasionally arise (Espinosa 1925, Lipski 1993, Piñeros 2005). One of the most frequent combinations, still to be heard in this archaic dialect, involves the possessive mi ‘my’ + noun beginning with a labial consonant, e.g. mi papá > [m̥ .pa.pá] ‘my father,’ mi paquete [m̥ .pa.ké.te] ‘my package.’ Although the syllabic nasal has survived in vernacular speech, for many younger speakers it has accreted a prothetic vowel: empapá, a variant that appears in a regionalist literary text (Arellano 1992: 91):7 (2)

Les dio muncho miedo y corrieron derecho a la casa de enpapá como era la más cerca ‘They got really scared and ran straight to my dad’s house, since it was the closest.’ Salió enpapá cuando oyó el bullicio y ya entraron las tres más muertas que vivas. ‘my dad came out when he heard the racket and the three women entered, more dead than alive.’

As with the case of word-initial prenasalized consonants, the prothetic vowel extruded from word-initial syllabic nasals does not respect alignment conditions, although contiguity is maintained.

6 The treatment of complex onsets with non-decreasing sonority There are also occasions when Spanish speakers are faced with onset clusters containing elements either of equal sonority or exhibiting rising sonority but contiguous on the sonority scale, thereby violating sonority sequencing. Examples 7 The treatment of the syllabic nasal mirrors that of prenasalized obstruents, and also parallels the evolution of preverbal clitics in Catalan, which in proclitic position evolved from the configuration C + /e/ to /e/ + C e.g. me > *m̥ > em (Moll 1952: 192; Badia i Margarit 1981: 294–296). Espinosa observes that the same process has been proposed for proto-Indo European evolution into Latin, for example *km̥ tom > centum ‘hundred,’ *dekm̥ > decem ‘ten’ (Brugmann 1905: 133; Sommers 1914: 41–46). Similar developments occurred in Oscan and Umbrian (Lindsay 1894: 273–275).

254

John M. Lipski

include psi ‘the Greek letter,’ Phnom Penh (the capital of Cambodia), Tbilisi (capital of the Republic of Georgia), Knack, Knoll, Knecht (surnames of Germanic origin, common in parts of South America), B’nai Brith, Mladić (former Serbian general), etc. Although assimilated borrowings typically delete the first consonant (e.g. (p)sicología ‘psychology’; Hualde 2005: 77), when attempting to pronounce such words for the first time Spanish speakers frequently strive to retain both consonants by means of repair strategies. Unlike what occurs in the case of initial clusters of falling sonority and final clusters with rising sonority, Spanish speakers do not usually make use of the epenthetic [e] that is readily available as a repair strategy in other environments. Inserted instead is a svarabhakti vowel (Hualde 2005: 113) or truncated timing slot between the two consonants. Unlike the case of complex onsets with falling sonority and complex codas with rising sonority, the epenthetic vowel in these complex onset repairs is typically shorter than a full Spanish [e], may be devoiced, and exhibits a less well-defined formant structure than other Spanish vowels, more often resembling the excrescent segment characteristically found within OBSTRUENT + LIQUID onset clusters (Bradley 2002, 2006, Bradley & Schmeiser 2003; Ramírez 2006), although being schwa-like in having a centralized articulation rather than being a fleeting duplicate of the following vowel. This element may not represent a target of phonological epenthesis but rather the “pulling apart” of individual consonantal gestures, which momentarily opens the vocal track and extrudes a brief vocalic sound (e.g. Davidson & Stone 2003, Gafos 2002). Alignment of left word and syllable boundaries is maintained but contiguity is apparently violated, even when an acceptable Spanish coda would result if a prothetic rather than an inter-consonantal vowel were inserted.

7 Moras and the licensing of epenthetic vowels The survey of synchronically productive epenthetic processes in Spanish highlights the potential shortcomings of postulating alignment, contiguity, or coda constraints as the primary motivation for an apparently cohesive phenomenon. Needed is a unifying predictor of the position and phonetic nature of the epenthetic vowel. Such an element is the MORA , the unit of phonological weight. The concept of the mora has a long history in phonological theory, and refers in general to a unit of syllabic length (“long” vs. “short” syllables) and/or weight (potentially attracting stress accent). Seen in these terms moras are defined epiphenomenally, as indicated by McCawley (1978): “The only reasonable definition of ‘mora’ that I am aware of is ‘something of which a long syllable consists

Spanish vocalic epenthesis

255

of two and a short syllable consists of one.’” Based on extensive cross-linguistic research Hyman (1985) offers a comprehensive proposal for the mora as a phonological prime (cf. also Broselow 1995, Hayes 1989, inter alia), and the mora has made its way into models of Spanish syllable and word structure (e.g. Colina 2006, Harris 1992, Hualde 1999, Lipski 1997, among others). With respect to consonants, onsets do not carry moras; all moraic consonants are ultimately found in coda position. The aforementioned studies embody an epiphenomenal definition of the mora, e.g. found in long vowels, geminate consonants, or “heavy” stress-attracting syllables. Syllables, however, are typically defined in terms of sonority, and Spanish syllabification algorithms make essential use of relative sonority (e.g. Harris 1989; Hualde 1991, 1999), and it is therefore appropriate to directly define the mora based on sonority. In order to account for vocalic epenthesis in Spanish the following sonority-based definition of the mora is consistent with observed phenomena: (3)

Define as MORAIC (in Spanish) any segment immediately followed by a segment OF LOWER SONORITY (or by no segment at all).

This is not the first time that moras have been defined in terms of the relative sonority of contiguous segments. Zec (1995: 91–92) offers a similar proposal as a necessary condition for a segment to project a mora,8 with the difference that the sonority of the moraic segment be equal to or higher than the following segment. In Spanish this works for word-internal sequences of consonants (Spanish has no phonological geminates or long vowels), but as shown below, does not account for the treatment of foreign words beginning with two consonants of equal sonority.9 For Spanish, the condition in (3) is assumed to be not only necessary but also sufficient, i.e. every segment so defined is moraic. The sufficiency condition is implicit in Zec’s model, and is explicitly required for Spanish. Under

8 For Zec the mora is the phonological prime, and the sonority conditions merely allow moras to be projected. It is not clear whether this difference has any empirical consequences. 9 For learned and non-patrimonial words with single coda obstruents, e.g. doctor, Magdalena, the coda consonant is almost always realized as a weak approximant or is vocalized into the nucleus (e.g. Lipski 1994, Martínez-Gil 1990, 1996, 1997; Piñeros 2001) while the following onset consonant may retain an occlusive articulation, even extending to the voiced obstruents /b/, /d/, and /ɡ/, which are more frequently realized as approximants in other prevocalic environments.

256

John M. Lipski

this definition, all vowels (as well as postnuclear, but not prenuclear glides) carry a mora, as do all coda consonants.10 Given this definition of moras, the existence and placement of epenthetic vowels in Spanish can be characterized more systematically. Since epenthetic vowels do not add phonological weight (Hyman 1984: 71f.; Piggott 1995; Piggott & Singh 1985), i.e. carry no mora, a full epenthetic vowel added during the repair of non-permissible onset and coda clusters must acquire its mora through licensing. The only available licenser of an epenthetic vowel is a “stray” moraic consonant (which, by virtue of being part of a non-permissible Spanish cluster, is originally unsyllabified). The repair strategies described in sections 2–6 can be re-cast in terms of the presence of unsyllabified moras. Beginning with word-final clusters of increasing sonority, in a word like Sadr, the /a/ and the /ɾ/ are both moraic while the /d/ is not. Since the /d/ is not moraic, it cannot be syllabified as a coda and must therefore be part of an onset. The /ɾ/, being moraic, cannot become part of a (complex) onset, thus ruling out an eventual resolution of the form *[sa.ðɾ. . .]. All moras must be associated with syllabic rhymes and since in Spanish all syllables must contain a nuclear vowel, an epenthetic vowel is licensed by the moraic /ɾ/, which becomes the coda of the syllable [ðeɾ]. A similar series of deductions pinpoints the location of the epenthetic vowel in the Spanish-influenced pronunciation of English happen as [xa.pen]/*[xap.ne]. The /n/ is moraic and must therefore be syllabified as a coda, while the /p/ is non-moraic and can only be an onset. In this analysis, alignment considerations are coincidental, as is the violation of contiguity. Relative sonority configurations, together with the need for an epenthetic vowel to be licensed by a moraic consonant (in coda position) converge on a single location for the epenthetic vowel. An identical procedure predicts the position of epenthetic vowels in wordinitial clusters of declining sonority including nasal + C and /s/ + C. In mboy, the /m/ is moraic and the /b/ is not. An epenthetic vowel between the /m/ and the /b/ would encounter no moraic consonant as a licenser, thus ruling out 10 If it is assumed that /s/ is actually more sonorous than stops, then the first element of complex codas such as those in obstáculo ‘obstacle,’ bíceps, etc. should not be moraic. When the coda obstruent is actually pronounced in such words, however, it never emerges as a stop, but rather as some sort of weak approximant, arguably with sonority levels approaching that of glides. The same reasoning could be applied to single-obstruent codas, as in doctor, acto ‘act,’ fiebre aftosa ‘hoof and mouth disease.’ However word-internally it will be necessary to add a condition to the definition, namely that an element is moraic if followed by an element of equal sonority, just in case it is also preceded by an element of higher sonority. This will account for words like himno ‘hymn, burla ‘mockery,’ alrededor ‘around,’ while not assigning a mora to the /t/ of Tbilisi.

Spanish vocalic epenthesis

257

*meboy as a possible repair. The epenthetic vowel must be placed before the /m/, which being moraic licenses the inserted vowel and serves as a coda. A word like stress receives a similar treatment. Neither the /t/ nor the /ɾ/ is moraic, thereby ruling out *setrés, which would contain an unlicensed epenthetic vowel. The /s/, being moraic, emerges as a coda consonant licensing the prothetic vowel, producing the actually occurring estrés.11

8 The repair of initial clusters with nondecreasing sonority Onset clusters that do not decrease in sonority, such as pn-, kn-, tb-, ml-, etc. contain no moraic consonants; therefore neither consonant can be syllabified as a coda. Although ultimately deletion of the first consonant may emerge as a permanent solution (possibly due to the markedness of a non-moraic consonant that cannot be syllabified as an onset), in “first encounters” with foreign words speakers often strive to accommodate every segment in some fashion. In onset clusters with non-decreasing sonority the only potential place for an epenthetic vowel is between the two consonants. In this situation a full epenthetic vowel is unable to acquire a mora from either of the flanking consonants and this is reflected in the effervescent nature of the inserted vocalic element, which almost never emerges as a full [e] or a truncated copy of the following vowel but rather more like a fleeting svarabhakti vowel with schwa-like articulatory and acoustic traits.12 An epenthetic vowel has never become lexicalized in such situations, as has occurred with epenthetic vowels licensed by moraic consonants (e.g. estrés, escáner), and the historical trend of eliminating the first consonant highlights 11 Singh (1985: 273) discuss the asymmetries of prothesis and anaptyxis, and argues that sonority sequencing is responsible: “An unacceptable word-initial cluster is broken up by the insertion of a vowel before the most sonorous segment.” This accounts for the difference between /s/+C with prothesis and stop + liquid with anaptyxis. Davidson et al. (2004) tested English speakers’ attempts to pronounce non-permissible onset clusters, containing consonants of equal sonority or rising sonority but adjacent on the sonority hierarchy. Epenthesis of a weak schwa (between the two consonants) was by far the most common repair strategy. Davidson and Stone (2003) used ultrasound for English speakers attempting to pronounce initial zgand determined that the target was actually a schwa and not an excrescent vowel formed by gestural mistiming. 12 According to Fleischhacker (2001: 85) epenthetic vowels are placed “exactly where they are least auditorily obtrusive – between the members of an underlying obstruent + sonorant cluster, but before a sibilant + stop cluster – unless a conflicting constraint forces otherwise (e.g. acrossthe-board anaptyxis, to satisfy a phonotactic banning consonants in non-prevocalic position).”

258

John M. Lipski

the difference between first approximations – which may include unconventional and uncomfortable strategies and articulations – and phonologically stable accommodation, which only draws upon fully integrated segments articulatory gestures. The epenthetic segments that arise when Spanish speakers attempt to pronounce non-permissible onset clusters with non-decreasing sonority fit the definition of semisyllables (e.g. Cho & King 2003: 187): (a) no mora; (b) no nucleus; (c) no coda; (d) no stress/accent/tone; (e) prosodically invisible; (f) well-formed onset; (g) restricted to morpheme peripheral position. By allowing semisyllables into the prosodic hierarchy, it is possible for all elements to be associated with a syllable (full or semi), without necessarily being associated with a mora or a foot.13 In the accommodation of words such as Tbilisi, Knack, etc. no full epenthetic vowel will produce a moraic syllable since neither of the two consonants is moraic, and therefore cannot be syllabified as a coda. Synchronically, the prohibition against deleting consonants in onset clusters is very strong, and the least phonological disruption is produced by the insertion of a semisyllable, in effect a brief intrusive element that overcomes the articulatory unfamiliarity of the cluster without adding a clearly distinguishable epenthetic vowel.

9 A basic OT model The presence and phonological placement of epenthetic vowels in non-permissible clusters can be illustrated within the framework of Optimality Theory by means of a series of independently-motivated constraints, including the following: MAX: no deletion DEP: no epenthesis SSC: sonority sequencing condition (syllable onsets must increase in sonority and syllable codas must decrease in sonority) MSD: maximal sonority distance for onset clusters SYL-μ: all syllables must contain a mora μ-SYL: all moras must be syllabified with a full vowel nucleus

13 Kiparsky (2003: 156) offers an alternative definition of semisyllable, which differs from the above traits in being essentially an unsyllabified mora, lacking onsets (or non-branching onsets). In Spanish, semisyllables are not moraic, in contrast to Kiparsky’s proposal.

Spanish vocalic epenthesis

259

The constraint MAX is un-dominated; synchronically Spanish does not routinely resolve non-permissible consonant clusters through deletion. The sonority-based constraints SSC and MSD are also un-dominated, as are the requirements that every syllable contain a mora and that every mora be associated with a full syllable. In the following tableaux the symbol ζ will be used to refer to the non-moraic svarabhakti vowel used to break up onset clusters with nondecreasing sonority. The standard assumption that the Spanish default vowel [e] is non-moraic when added through epenthesis is also in effect. It is also the case in Spanish that the underlying mora count is not altered in epenthetic repair. The following tableaux exemplify the interaction of moras, sonority, epenthetic vowels, and svarabhakti vowels. The Spanish epenthetic vowel [e] can only be added when a previously unsyllabified moraic consonant is available to license the vowel from the coda position. The svarabhakti vowel [ζ] remains non-moraic and non-syllabic. In each instance the total mora count remains unchanged, irrespective of the presence of epenthetic or excrescent vowels; this is a fundamental facet of Spanish phonological repair. The derivation ski > [es.ki] with prothetic [e] is given in Tableau 1. The /s/ is moraic and therefore requires a full epenthetic vowel (the Spanish default vowel /e/) in order to be realized. The syllabified moraic consonant always appears in coda position. Tableau 1: ski > [es.ki] /ski/ | | μμ

SSC

[ski]

!*

[si]/[ki] [se.ki]

MAX

SYL-μ

μ-SYL *

!* !*

*

☞ [es.ki] [sζ.ki]

DEP

* *

!*

*

(*)

[sζki]

!*

(*)

[ζs.ki]

!*

(*)

The derivation for psi > [pζsi] containing an onset cluster with non-decreasing sonority is shown in Tableau 2. Neither consonant is moraic, which precludes epenthesis of the default vowel /e/. Since /ps-/ is a non-permissible onset cluster in Spanish the svarabhakti vowel appears between the two consonants; it cannot

260

John M. Lipski

appear before the (non-moraic) /p/ because this would result in a mora-less syllable. Tableau 2: psi- > [pζsi] /psi/ | μ

MSD

[psi]

!*

[si]/[pi]

MAX

SYL-μ

μ-SYL

DEP

!*

[pe.si]

!*

*

[ep.si]

!*

*

[pζ.si]

!*

(*)

☞ [pζsi]

(*)

[ζp.si]

!*

(*)

The derivation of Sadr > [sa.ðeɾ] containing a coda cluster with rising sonority is shown in Tableau 3. The /ɾ/ is moraic and requires syllabification by the full epenthetic vowel /e/. Tableau 3: Sadr > [sa.ðeɾ] /sadr/ | | μμ [saðɾ] [sað]/[saɾ]

SSC

MAX

SYL-μ

μ-SYL

DEP

!* !*

sað-ɾ

!*

[sa.ðɾ]

!*

[sa.ðζɾ]

!*

sa.ðɾζ

!*

*

(*)

[sa.ðɾe]

!*

*

*

☞ [sa.ðeɾ]

*

In Tableau 4 the derivation of the word-initial syllable mba > [em.ba] with decreasing sonority in the onset cluster is shown. The /m/ is moraic and is incorporated into coda position together with the full epenthetic vowel /e/.

Spanish vocalic epenthesis

261

Tableau 4: mba- > [em.ba] /mba/ | | μ μ

SSC

[mba]

*!

[ba]/ma]

MAX

SYL- μ

μ- SYL

DEP

* !*

*

[m.ba]

!*

[me.ba]

!*

*

*

[mζ.ba]

!*

*

(*)

!*

(*)

[ζm.ba] ☞ [em.ba]

*

10 Summary and conclusions The preceding sections have proposed a model to account for the nature and placement of epenthetic vowels as syllable repair strategies in Spanish, based on relative sonority relationships among contiguous segments. Taking as a point of departure the postulate that epenthetic vowels lack phonological weight, it has been proposed that a full epenthetic vowel (in this case the Spanish default vowel /e/) can only be licensed by an unsyllabified moraic consonant, i.e. the second element in word-final clusters of increasing sonority or the first element of word-initial clusters of decreasing sonority. The moraic consonant then becomes the coda of the epenthetic syllable, whose nucleus is the default vowel /e/. In phonotactically non-permissible onset clusters of non-decreasing sonority neither consonant is moraic and a full epenthetic vowel cannot be licensed; ifSpanish speakers attempt to pronounce both consonants a non-moraic svarabhaktivowel semi-syllable is usually inserted between the two consonants.14 The common thread uniting all Spanish consonant cluster repairs is that the mora count remains unchanged.15 The placement and phonetic nature of the epenthetic 14 Broselow (1987: 293; 1992) demonstrates that many of the same repair strategies are used by English speakers when faced with onset clusters with decreasing or equal sonority. 15 Although moras have frequently been invoked to account for stress placement in quantitysensitive languages, it is not axiomatic that moras are phonologically relevant only in quantitysensitive languages. The present study is orthogonal to the ongoing debate as to whether contemporary Spanish is a quantity-sensitive language (see Piñeros’ article in this volume).

262

John M. Lipski

vowel in effect represent empirical phonetic correlates of moraicity; the appendix contains a visual illustration. This preliminary exploration embodies the potential for adding another dimension to the understanding of Spanish syllable structure.

References Archangeli, Diana. 1988. Aspects of underspecification theory. Phonology 5.183–207. Arellano, Juan Estevan. 1992. Inocencio: Ni pica ni escarda, pero siempre se come el mejor elote. México: Grijalbo. Badia i Margarit, Antoni. 1981. Gramàtica històrica catalana. Valencia: Eliseu Clement. Benítez del Cristo, Ignacio. 1930. Los novios catedráticos. Archivos del Folklore Cubano 5(2). 119–46. Bonet, Eulalia. 2006. Gender allomorphy and epenthesis in Spanish. In Fernando Martínez-Gil & Sonia Colina (eds.), Optimality-theoretic studies in Spanish phonology, 312–338. Amsterdam & Philadelphia: John Benjamins. Bradley, Travis. 2002. Gestural timing and the resolution of /Cɾ/ clusters in Romance. ROA 529– 0702. Bradley, Travis. 2006. Spanish complex onsets and the phonetics–phonology interface. In Fernando Martínez-Gil & Sonia Colina (eds.), Optimality-theoretic studies in Spanish phonology, 15–38. Amsterdam/Philadelphia: John Benjamins. Bradley, Travis & Ben Schmeiser. 2003. On the phonetic reality of Spanish /ɾ/ in complex onsets. In Paula Kempchinsky & Carlos-Eduardo Piñeros (eds.), Selected Proceedings of The 6th Hispanic Linguistics Symposium, 1–20. Somerville MA: Cascadilla. Broselow, Ellen. 1987. Non-obvious transfer: on predicting epenthesis errors. In Georgette Ioup & Steven Weinberger (eds.), Interlanguage phonology: the acquisition of a second language sound system, 292–304. Cambridge, MA: Newbury House. Broselow, Ellen. 1992. Transfer and universals in second language epenthesis. In Susan Gass & Larry Selinker (eds.), Language transfer in language learning, revised edition, 71–86. Amsterdam & Philadelphia: John Benjamins. Broselow, Ellen. 1995. Skeletal positions and moras. In John Goldsmith (ed.), The handbook of phonological theory, 175–205. Cambridge, MA: Blackwell. Brugmann, Karl. 1905. Abrégé de grammaire comparé des langues indo-européennes. Paris: Klincksieck. Cabrera, Lydia. 1970. Refranes de negros viejos. Miami: Ediciones CR. Cabrera, Lydia. 1971. El monte. Miami: Ediciones C. R. Carlisle, Robert. 1991. The influence of environment on vowel epenthesis in Spanish/English interphonology. Applied Linguistics 12.76–95. Cho, Young-mee Yu & Tracy Holloway King. 2003. Semisyllables and universal syllabification. In René Kager, Joe Pater & Wim Zonneveld (eds.), Constraints on phonological acquisition, 183–212. Cambridge: Cambridge University Press. Claramonte, Andrés de. 1951. La gran comedia del valiente negro en Flandes. In Ramón de Mesonero Romanos (ed.), Dramáticos contemporáneos a Lope de Vega, 491–509. Madrid: Biblioteca de Autores Españoles.

Spanish vocalic epenthesis

263

Clements, G. N. 1990. The role of the sonority cycle in core syllabification. In John Kingston & Mary Beckman (eds.), Between the grammar and physics of speech: Papers in Laboratory Phonology 1, 283–333. Colina, Sonia. 2006. Optimality-theoretic advances in our understanding of Spanish syllable structure. In Fernando Martínez-Gil & Sonia Colina (eds.), Optimality-theoretic studies in Spanish phonology, 172–204. Amsterdam & Philadelphia: John Benjamins. Colina, Sonia. 2009. Spanish phonology: A syllabic perspective. Washington, DC: Georgetown University Press. Davidson, Lisa, Peter Musczyk & Paul Smolensky. 2004. The initial and final states: Theoretical implications and experimental explorations of Richness of the Base. In René Kager, Joe Pater, & Wim Zonneveld (eds.), Constraints on phonological acquisition, 321–368. Cambridge: Cambridge University Press. Davidson, Lisa & Maureen Stone. 2003. Epenthesis versus gestural mistiming in consonant cluster production: an ultrasound study. In Mimu Tsujimura & Gina Garding (eds.), WCCFL 22 Proceedings, 165–178. Somerville, MA: Cascadilla Press. Davis, Stuart. 1990. Italian onset structure and the distribution of il and lo. Linguistics 28. 43– 55. Eddington, David. 2001. Spanish epenthesis: formal and performance perspectives. Studies in the Linguistic Sciences 31. 33–53. Fleischhacker, Heidi. 2001. Cluster-dependent epenthesis asymmetries. UCLA Working Papers in Linguistics 7. 71–116. Friedemann, Nina S. de & Carlos Patiño Rosselli. 1983. Lengua y sociedad en el Palenque de San Basilio. Bogotá: Instituto Caro y Cuervo. Gafos, Adamantios. 2002. A grammar of gestural coordination. Natural Language and Linguistic Theory 20.269–337. García González, José & Gema Valdés Acosta. 1978. Restos de lenguas bantúes en la región central de Cuba. Islas 59. 3–49. Goad, Heather & Yvan Rose. 2004. Input elaboration, head faithfulness, and evidence for representation in the acquisition of left-edge clusters in West Germanic. In René Kager, Joe Pater & Wim Zonneveld (eds.), Constraints on phonological acquisition, 109–157. Cambridge: Cambridge University Press. Granda, Germán de. 1966. La estructura silábica y su influencia en la evolución fonética del dominio ibero-románico. Madrid: Consejo Superior de Investigación Científica. Gutiérrez-Rexach, Javier. 2006. Sonority scales and syllable structure: toward a formal account of phonological change. In Fernando Martínez-Gil & Sonia Colina (eds.), Optimality-theoretic studies in Spanish phonology, 447–469. Amsterdam & Philadelphia: John Benjamins. Harris, James. 1983. Syllable structure and stress in Spanish: A non-linear analysis. Cambridge, MA: The MIT Press. Harris, James. 1989. Sonority and syllabification in Spanish. In Carl Kirschner & Janet Decesaris (eds.), Studies in Romance linguistics, 139–153. Amsterdam & Philadelphia: John Benjamins. Harris, James. 1992. Spanish stress: The extrametricality issue. Bloomington, IN: University Linguistics Club. Hayes, Bruce. 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20.253– 260. Hualde, José Ignacio. 1991. On Spanish syllabification. In Héctor Campos & Fernando MartínezGil (eds.), Current studies in Spanish linguistics, 475–493. Washington, DC: Georgetown University Press.

264

John M. Lipski

Hualde, José Ignacio. 1999. La silabificación en español. In Rafael A. Núñez-Cedeño & Alfonso Morales-Front (eds.), Fonología generativa contemporánea de la lengua española, 170– 188. Washington, DC: Georgetown University Press. Hualde, José Ignacio. 2005. The sounds of Spanish. Cambridge: Cambridge University Press. Hyman, Larry. 1985. A theory of phonological weight. Dordrecht: Foris. Kaye, Jonathan. 1992. Do you believe in magic? the story of S + C sequences. SOAS Working Papers in Linguistics 2. 293–313. Kiparsky, Paul. 2003. Syllables and moras in Arabic. In René Kager, Joe Pater & Wim Zonneveld (eds.), Constraints on phonological acquisition, 147–182. Cambridge: Cambridge University Press. Lindsay, W. M. 1894. The Latin language. Oxford: Clarendon Press. Lipski, John. 1992. Spontaneous nasalization in Afro-Hispanic language. Journal of Pidgin and Creole Languages 7. 261–305. Lipski, John. 1993. Syllabic consonants and New Mexico Spanish: The geometry of syllabification. Southwest Journal of Linguistics 12. 109–127 [published 1998]. Lipski, John. 1994. Spanish stops, spirants and glides: From consonantal to [vocalic]. In Michael Mazzola (ed.), Issues and theory in Romance Linguistics (selected papers from the Linguistic Symposium on Romance Languages XXIII), 67–86. Washington, DC: Georgetown University Press. Lipski, John. 1997. Spanish word stress: The interaction of moras and minimality. In Fernando Martínez-Gil & Alfonso Morales-Front (eds.), Issues in the phonology and morphology of the major Iberian languages, 559–593. Washington, DC: Georgetown University Press. Lipski, John. 2002. Epenthesis vs. elision in Afro-Iberian language: A constraint-based approach to creole phonology. In Teresa Satterfield, Christina Tortora & Diana Cresti (eds.), Current issues in Romance languages, 173–188. Amsterdam & Philadelphia: John Benjamins. Lombardi, Linda. 2003. Markedness and the typology of epenthetic vowels. Rutgers Optimality Archive ROA-578. Martínez-Gil, Fernando. 1990. Topics in Spanish historical phonology: A non-linear approach. Los Angeles, CA: University of Southern California dissertation. Martínez-Gil, Fernando. 1996. El principio de la distancia mínima de sonoridad y el problema de la vocalizaci6n consonántica en el español dialectal de Chile. Hispanic Linguistics 8. 201–246. Martínez-Gil, Fernando. 1997. Obstruent vocalization in Chilean Spanish: A serial versus a constraint based approach. Probus 9. 167–202. Martínez-Gil, Fernando. 2001. Sonority as a primitive phonological features: Evidence from Spanish complex onset phonotactics. In Julia Herschensohn, Enrique Mallen & Karen Zagona (eds.), Features and interfaces in Romance: essays in honor of Heles Contreras, 203–222. Amsterdam & Philadelphia: John Benjamins. McCawley, James. 1978. Notes on the historyu of accent in Japanese. Recent developments in history phonology, 287–307. The Hague: Mouton. Moll, Francisco de B. 1952. Gramática histórica catalana. Madrid: Gredos. Morelli, Frida. 2003. The relative harmony of /s+Stop/ onsets: Obstruent clusters and the sonority sequencing principle. In Caroline Féry & Ruben van de Vijver (eds.), The syllable in optimality theory, 356–371. Cambridge: Cambridge University Press. Núñez Cedeño, Rafael A. 1988. The Abakuá secret society in Cuba: Language and culture. Hispania 71.148–154.

Spanish vocalic epenthesis

265

Núñez Cedeño, Rafael A., Rolando Alúm, & Roberto Nodal. 1985. The Afro-Hispanic Abakuá: A study of linguistic pidginization. Orbis 31 (1–2). 263–284. Piggott, G. L. 1995. Epenthesis and syllable weight. Natural Language and Linguistic Theory 13. 283–326. Piggott, G. L. & Rajendra Singh. 1985. The phonology of epenthetic segments. Canadian Journal of Linguistics 30. 415–453. Piñeros, Carlos-Eduardo. 2001. Segment-to-syllable alignment and vocalization in Chilean Spanish. Lingua 111. 163–188. Piñeros, Carlos-Eduardo. 2005. Syllabic consonant formation in traditional New Mexico Spanish. Probus 17. 253–301. Ramírez, Carlos Julio. 2006. Acoustic and perceptual characterization of the epenthetic vowel between the clusters formed by consonant + liquid in Spanish. In Manuel Díaz-Campos (ed.), Selected Proceedings of the 2nd Conference on Laboratory Approaches to Spanish Phonetics and Phonology, 48–61. Somerville, MA: Cascadilla Proceedings Project. Rose, Yvan & Katherine Demuth. 2006. Vowel epenthesis in loanword adaptation: Representational and phonetic considerations. Lingua 116. 1112–1139. Sánchez, Roberto. 1979. Luis Vélez de Guevara, El negro del seraphín: A critical and annotated edition with introduction. Tallahassee, FL: Florida State University dissertation. Shepherd, Michael. 2003. Constraint interactions in Spanish phonotactics: An optimality-theory analysis of syllable-level phenomena in the Spanish language. Northridge, CA: California State University MA thesis. ROA 639-Shepherd. Singh, Rajendra. 1985. Prosodic adaptation in interphonology. Lingua 67. 269–282. Sommer, Ferdinand. 1914. Handbuch der lateinischen Laut- und Formenlehre. Heidelberg: Carl Winters. Vega Carpio, Lope de. 1893. Obras de Lope de Vega. Madrid: Real Academia Española. Zec, Draga. 1995. Sonority constraints on syllable structure. Phonology 12. 85–129.

266

John M. Lipski

Appendix: Visualizing epenthesis An example of the prothetic /e/ in the traditional New Mexico Spanish realization of mi papá > empapá is given in Figure 1.

Figure 1: Realization of mi papá by a speaker of traditional New Mexican Spanish

As an informal demonstration of Spanish vocalic epenthesis five Spanish speakers (from Argentina, Colombia, Ecuador, Spain, and the United States) were presented with a number of foreign words beginning and ending with consonant clusters that are phonotactically impermissible in Spanish, and were asked to pronounce them “as though they were Spanish words.” Onset clusters with prenasalized consonants (e.g. Nguema) almost always resulted in a prothetic [e], as did final clusters with non-decreasing sonority. Initial clusters with non-decreasing sonority added a svarabhakti vowel between the two consonants; the excrescent vowels were very short, often partially devoiced, and acoustically more centralized than Spanish /e/. Figure 2 shows the epenthetic [e] in the pronunciation of Nguema and Al Sadr by one of the Spanish speakers. Figure 3 shows svarabhakti vowels in the same speaker’s pronunciation of the word Ptaki (a surname), and Tblisi. In Figure 4 the svarabhakti vowels and the two instances of epenthetic [e] are situated within the vowel space for representative tonic vowel tokens from the

Spanish vocalic epenthesis

267

same speaker is shown. The epenthetic vowels licensed by moraic consonants (in Nguema and Al Sadr) cluster very close to tonic /e/, while the svarabhakti vowels not licensed by a moraic consonant (in Ptaki and Tblisi) are centralized and schwa-like, as well as being much shorter than epenthetic /e/.

Figure 2: Epenthetic vowels in a Spanish speaker’s pronunciation of Nguema and Al Sadr

268

John M. Lipski

Figure 3: Svarabhakti vowels in a Spanish speaker’s pronunciation of Ptaki and Tblisi

Spanish vocalic epenthesis

Figure 4: Full, epenthetic, and svarabhakti vowels of a Spanish speaker

269

Carlos-Eduardo Piñeros

The phonological weight of Spanish syllables Abstract: This article conclusively demonstrates that the Spanish stress system is quantity insensitive. It proves that quantity sensitivity suffers from circularity because the savings in stress marks that it allegedly yields are counterbalanced by an exorbitant number of extrametricality marks and, even if one accepts the stipulation that finality within certain domains renders prosodic units invisible, proper accentuation of many words is still not possible. Another factor that makes quantity sensitivity unviable is that heavy syllables reduce mobility within the accentual window and, although this seems to have the positive effect of making the distribution of stress stricter, it is actually a disadvantage because it adds to the pile of data that cannot be derived. The proposed alternative maintains that, in two thirds of the lexicon, stress is assigned by universal principles requiring the projection of a word-final syllabic trochee, while in the remaining third this footing is minimally altered by two types of morphologicallycontrolled irregularity.

1 Introduction Five decades after Foley (1965) proposed the first generative analysis of Spanish stress, there are still fundamental issues about the accentuation of words in this language that remain poorly understood. The following three are but a sample: a) is the stress domain the prosodic word or the stem? b) Are Spanish prosodic feet trochaic or iambic? And c) is the system quantity sensitive or insensitive? Views on these basic matters are discordant, even in the most recent works. With respect to the stress domain, compare, for instance, Martínez-Paricio (2013) and Hualde (2012); in relation to the foot type, contrast Gibson (2011) and Roca (2006a, this volume), and with regard to quantity sensitivity, see Shelton (2007) and Ohannesian (2004). Despite resorting to numerous theoretical and experimental approaches, advances in this area have been hampered by the fact that what appears to be a simple prosodic system is complicated by various morphological factors. The one thing that fifty years of research have clearly established Carlos-Eduardo Piñeros, University of Auckland

272

Carlos-Eduardo Piñeros

is that Spanish stress is highly conditioned by the morphology, which obscures the generalizations that one expects to see in a well-behaved prosodic system. The present work focuses on question (c) above, but it inevitably touches on questions (a) and (b) because these issues are interrelated. My goal is to conclusively demonstrate that heavy syllables do not exist in any guise in Spanish, so that the view that this language is quantity sensitive can be abandoned for good. The main achievements that will be made are the following. First, I will develop a solid analysis of the regular pattern (Section 2). Second, I will prove that extrametricality – a formal device needed to salvage quantity sensitivity – is untenable, both theoretically and empirically (Sections 3, 4, and 6). Third, I will argue that Spanish has two types of prosodically-irregular morphemes, which, when functioning as the head of the word, have the ability to push stress minimally away from the syllable that phonology alone would choose as the optimal stress bearer (Section 5). The latter is the thesis I submit to account for the existence of two robust patterns which deviate from the norm: retraction and advancement. Although this thesis will be reasonably well substantiated, its formalization must be left for a subsequent study because developing the theoretical machinery needed for that would require delving into aspects of the phonology-morphology interface which extend far beyond the scope of this article.1

2 Transferring stress from the lexicon to the grammar One of the few points of agreement in the debate about Spanish stress is that it may be distinctive. In effect, this suprasegmental may be the only sound

1 The ideas presented here could not have been conceived without the valuable contributions that James W. Harris has made to our understanding of Spanish phonology and morphology. I feel profound admiration for Jim’s work and wish to thank him for the many things I have learned from his publications – they have always been a model for me to follow and a way to discover what doing linguistics is about. I also wish to state clearly that my ardent argumentation against Spanish quantity sensitivity is not an attempt to tarnish the image of excellence that his research has rightfully earned. On the contrary, I mean to honor him because it is by emulating the rigor that characterizes his work that I have come to view the Spanish prosodic system as depicted below. My gratitude also goes to Rafael A. Núñez Cedeño for inviting me to participate in this festschrift and to two anonymous reviewers for the feedback they offered me to improve the presentation and argumentation of this article. It is solely me who bears the responsibility for any errors that might remain.

The phonological weight of Spanish syllables

273

property distinguishing the members of pairs or triplets of words such as those in (1). (Given that it is a property of the syllable as opposed to the segment, stress is signaled in these and subsequent examples by using boldface on all letters representing the relevant unit.) (1)

a.

pa.pa / pa.pá ‘Pope’ / ‘dad’

b.

du.re / du.ré ‘may it last’ / ‘I lasted’

vá.li.do / va.li.do / va.li.dó ‘valid’ / ‘I validate’ / ‘he validated’ lí.mi.te / li.mi.te / li.mi.té ‘limit’ / ‘may he limit’ / ‘I limited’

These data suggest that stress is a phonemic property; thus, one that needs to be stored in the lexicon. The underlying representation of papa and papá would be, for instance, /ˈpapa/ and /paˈpa/, respectively. This was indeed the view adopted by structuralist studies such as Llorach (1950) and Stockwell and Bowen (1965), for which stress was utterly unpredictable. In sharp contrast with this, generative approaches contend that the locus of stress is largely predictable, which implies that it is an emergent property; it ensues from the workings of the grammar (e.g. Foley 1965, Harris 1969, 1975). This alternative interpretation finds strong support in the observation that, given a word of any number of syllables, stress is more likely to appear in certain positions than in others. One phenomenon that attests to this is that stress is systematically drawn towards the right periphery of the word. Consider in this regard the examples in (2), which show that derivational suffixation, the language’s most productive type of word formation, tends to induce a rightward stress shift.2 (2)

ro.jo ro.jez ro.ji.zo en.ro.je.ce en.ro.je.cer en.ro.je.ci.do en.ro.je.ce.dor en.ro.je.ci.mien.to

‘red’ ‘redness’ ‘reddish’ ‘he reddens’ ‘to redden’ ‘reddened’ ‘causing reddening’ ‘reddening’

2 The derivational suffixes appearing in these words are –ez ‘quality’, –iz ‘propensity’, –ece ‘inchoativity’ –r ‘infinitive/nominalization’, –d ‘participle/adjectivization’, –dor ‘agentivity’, and –mient ‘action/effect’.

274

Carlos-Eduardo Piñeros

These data clearly show that the Spanish stress system is right-bounded; that is to say that the stress locus must remain close to the right edge of the word. This distributional condition is corroborated by the fact that there are a limited number of stress contrasts that the language allows. As the examples in (3a) show, three is the maximum. Attempts to add further contrasts fail because that would require freedom to move stress an unlimited distance away from the right edge of the word (3b). (3)

a.

es.pe.ci.fi.có es.pe.ci.fi.co es.pe.cí.fi.co

b.

*es.pe.ci.fi.co *es.pe.ci.fi.co

‘he specified’ ‘I specify’ ‘specific’

A common conceptualization of the phenomena in (2) and (3) involves the postulation of an accentual window encompassing the last three syllables of the word: [ . . . σ σ σ ]). This is a useful descriptive device for it subsumes the three main stress patterns that exist: oxytone (4a), paroxytone (4b), and proparoxytone (4c). (4)

a.

[ ... σ σ σ ]

b.

[ ... σ σ σ ]

c.

[ ... σ σ σ ]

ma.ra.ve.dí ‘old Spanish coin’

pa.ra.le.lo ‘parallel’

pi.rá.mi.de ‘pyramid’

de.sa.ca.tó ‘he disobeyed’

me.mo.ri.za ‘he memorizes’

lle.gá.se.mos ‘might we arrive’

Another factor that works to the benefit of the generative approach is that there are significant differences in the frequency with which the window positions are used. It turns out that paroxytones occur in large numbers, oxytones in moderate numbers, and proparoxytones in relatively small numbers. The results of a word count conducted by Morales-Front (1999: 221, 2014: 244) confirm that this order of preference is true. As seen in (5), paroxytones account for almost two thirds of the corpus, oxytones make slightly less than one third, and proparoxytones represent a small residue. (5)

a.

Paroxytones:

58,423

64.20%

b.

Oxytones:

25,215

27.71%

c.

Proparoxytones: Total:

7,362

8.09%

91,000

100.00%

The phonological weight of Spanish syllables

275

The type of word-final segment is also an important factor. Subtler patterns arise depending on whether the word ends in a vowel or in a consonant. The breakdown in (6a) shows that it is specifically when the word ends in a vowel that penultimate stress is favored; when the word ends in a consonant, that pattern is highly disfavored. Oxytones behave in the opposite way (6b), whereas proparoxytones mimic the tendencies exhibited by paroxytones (6c). (6)

a.

b.

c.

Paroxytones V-final: C-final: Oxytones V-final: C-final: Proparoxytones: V-final: C-final:

57,911 512

63.64% 0.56%

573 24,642

0.63% 27.08%

7,327 35

8.05% 0.04%

The strong lead that V-final paroxytones have over all other patterns provides an empirical basis to assume that the penult is the normal stress locus: [ . . . σ σ σ ]. Most nouns, adjectives, verbs, and adverbs pattern with the representative examples in (7), which indicates that this is a cross-categorial generalization. (7)

a.

b.

Nouns pi.no ca.bi.na pa.na.de.ro

‘pine’ ‘booth’ ‘baker’

c.

Adjectives ma.lo do.ra.da pe.re.zo.so

‘bad’ ‘golden’ ‘lazy’

d.

Verbs to.co co.ci.ne vi.gi.la.ba

‘I touch’ ‘may he cook’ ‘he watched’

Adverbs pron.to ma.ña.na tem.pra.ni.to

‘soon’ ‘tomorrow’ ‘early, dim.’

The generality of this pattern suggests that there is an advantage to placing stress on the penult, which I argue is that it makes it possible to prosodify the word in a very simple way. The structure of the foot, the prosodic unit that lies between the syllable and the word, is at the heart of this thesis. Metrical Theory has revealed that stressed syllables are those that acquire prominence by virtue of acting as foot heads and that the syllable that gains primary stress is the one that functions as the head of the main foot, the unit that functions as the head

276

Carlos-Eduardo Piñeros

of the prosodic word (Liberman and Prince 1977, Selkirk 1980). The fact that the penult is the syllable that normally bears primary stress means, therefore, that Spanish prefers to prosodify its words by projecting a main foot, which uses that particular syllable as its head. In Optimality Theory (Prince and Smolensky 1993/2002), such a system can be modeled by positing a small set of prosodic wellformedness constraints, which are universal, yet violable (8–10). (8)

MAIN -RIGHT:

MAIN FOOT RIGHT (Tesar and Smolensky 2000) Align the main foot with the word, right edge.

(9)

FOOT BIN :

FOOT BINARITY (Prince and Smolensky 1993) Feet are binary at the mora or syllable level.

(10)

TROCH :

TROCHAIC RHYTHMIC TYPE (Prince and Smolensky 1993) The head of the foot precedes the tail.

It should be underscored that each one of these requirements has an empirical basis (Piñeros, 2016). MAIN -RIGHT is grounded on the demarcative function of stress; that is to say that, when the main foot is final, its head can effectively signal to the listener that the end of the prosodic word is approaching. FOOT BIN stems from the relational nature of stress: for an element to be perceived as prominent, it is essential that it be in a relationship with another element which is less salient. Lastly, TROCH is one of the two logical ways in which the prominent element (the head) and the non-prominent element (the tail) can be ordered with respect to one another. As illustrated in tableau (11), strict compliance with these constraints provides a plausible explanation for the abundance of paroxytones in the language. Candidates (11a) and (11b) place stress on the preantepenultimate syllable, thereby causing the main foot to either stand away from the right edge of the word or to encompass more than two syllables. The next two candidates opt to place stress on the antepenult, but they do not fare any better because that position also forces the main foot to either admit more than two syllables (11c) or to stand apart from the end of the word (11d). The important revelation that these hypothetical forms make is that MAIN -RIGHT and FOOT BIN are the principles responsible for the right-bounded nature of the system. In other words, to the extent that stress is retracted further back than the penult, its ability to signal the termination of the prosodic word and to actualize a prominent/non-prominent relationship is compromised. The next candidate on the list shows that the objections against (11a–d) are resolved when stress is shifted forward one more syllable; namely, to the penult (11e). This move is truly rewarding for it allows satisfaction of both MAIN -RIGHT

The phonological weight of Spanish syllables

277

and FOOT BIN without causing an infraction of TROCH . It is in this respect that candidate (11f), which chooses to shift stress all the way to the ultima, disappoints. Note that, while it manages to satisfy MAIN -RIGHT and FOOT BIN , it contravenes TROCH because the head of the foot comes after the tail. That problem does not arise for (11g), which also shifts stress to the ultima, but takes the precaution of discarding the foot tail. Unfortunately, this is condemned by FOOT BIN . (11)

The penult is the normal stress bearer Input: /paɾalelo/3

MAIN -RIGHT

FOOT BIN

a.

[(ˈpa.ɾa).le.lo]

b.

[(ˈpa.ɾa.le.lo)]

*!

c.

[pa.(ˈɾa.le.lo)]

*!

d.

[pa.(ˈɾa.le).lo]

☞ e.

[pa.ɾa.(ˈle.lo)]

f.

[pa.ɾa.(le.ˈlo)]

g.

[pa.ɾa.le.(ˈlo)]

TROCH

*!

*!

*! *!

The conclusion ensuing from (11) is that the placement of stress on the penult is desirable because it conduces to a prosodic word of utmost simplicity; it is erected on a foot which is final, binary, and left-headed: [ . . . (σ σ)]. Since this treatment accounts for about 64% of Spanish prosodic words, significant savings in lexical storage are indeed possible. Such savings are genuine because the effect of the proposed analysis is not to shift the memory burden from the lexicon to the grammar but, rather, to lift that burden entirely. This is thanks to the fact that the principles in (8–10) are inferable from the forms learners are exposed to. While it is feasible to derive the leading stress pattern from universal principles, total removal of stress-related information from the Spanish lexicon remains a desideratum because there is still roughly 36% of the vocabulary that has not yet been accounted for. Since this is a substantial amount, any analyst aiming to demonstrate that stress is largely predictable is faced with the challenge of finding a way to further reduce the number of words requiring stress marking. In the next section I show how quantity sensitivity has been put to that use. 3 True to form, this generative analysis assumes that stress is not present in the input form.

278

Carlos-Eduardo Piñeros

3 The intervention of quantity sensitivity It is well known that, in some languages, the placement of stress is conditioned by the makeup of the syllable, while in others that factor is inconsequential. Languages of the first type are said to be quantity sensitive (QS), whereas those of the second type are said to be quantity insensitive (QI). A distinction between light and heavy syllables is warranted in QS systems because, when the stress domain includes both a light and a heavy syllable, stress is systematically placed on the latter (Zec 2011, Goedemans and van der Hulst 2011). As for most other languages, it is generally assumed that the weight of Spanish syllables depends entirely on the complexity of the rhyme, the stretch formed by the nucleus and the coda. Syllables with a complex rhyme are said to be heavy, while those with a simple rhyme are considered light. Since this assumption renders the presence and complexity of the onset inconsequential, that component of the syllable will be disregarded hereafter. The standard QS analysis of Spanish stress recognizes three main types of heavy syllable, which, abstracting away from onsets, can be portrayed as [VC]σ, [V̯ V]σ, and [VV̯ ]σ. In the first type, a coda consonant is the segment that provides extra weight to the syllable (e.g. pez ‘fish’), while in the other two types it is the nonsyllabic vowel of a rising or falling diphthong that has that effect (e.g. pie ‘foot’ and ley ‘law’). The heaviness of such syllables is usually represented by assigning a unit of weight – a mora (µ) – to each rhyme segment. This has the important effect of making all heavy syllables bimoraic; hence, prosodically uniform: [VµCµ]σ , [V̯ µVµ]σ , and [VµV̯ µ]σ . A complementary step that is needed to obtain quantity sensitivity is to ensure that light syllables have lesser weight. This is accomplished by limiting light syllables to those whose rhyme consists solely of one vowel and attributing a single mora to that segment: [Vµ]σ . From these assumptions, a well-defined dichotomy emerges: a set of heavy (H) syllables, including [VµCµ]σ , [V̯ µVµ]σ , and [VµV̯ µ]σ , and a set of light (L) syllables, consisting solely of [Vµ]σ . Given that the demands of the constraint FOOT BIN can be met at either the mora or syllable level (9), the availability of heavy syllables is bound to have repercussions for footing. In particular, a heavy syllable will suffice to create a binary foot by virtue of being bimoraic, whereas a light syllable will have to combine with another syllable if it is to be part of a wellformed foot. Let us explore the consequences that this has for a language where the principles in (8–10) are dominant, as happens in Spanish. When a word ends in an LL sequence, quantity sensitivity is, of course, imperceptible; footing proceeds as usual: [ . . . (LL)]. The word paralelo, for

The phonological weight of Spanish syllables

279

instance, remains exactly as in (11e), except that, now, it can be made explicit that all of its syllables are monomoraic: [paµ.ɾaµ.(ˈleµ.loµ)]. The addition of moras to the transcription helps to appreciate the fact that FOOT BIN is satisfied both moraically and syllabically in this case. For words ending in an HL sequence, quantity sensitivity does not affect the stress locus either; however, it should be noted that a different type of trochee emerges. Take, for instance, vi.si.tan.te ‘visitor’, where it is clear that the footed syllables contribute a total of three moras: [biµ.siµ.(ˈtaµn̪µ.teµ)]. Although this suggests an infringement of FOOT BIN , in actuality, that is not the case because binarity does not have to be met both moraically and syllabically. Satisfaction at one of these levels suffices, which means that the structure [ . . . (HL)] is also in strict compliance with the constraints in (8–10). It is specifically in the case of words ending in an LH or in an HH sequence that the adoption of quantity sensitivity has a significant impact on footing. The weight of the ultima is the critical factor. When bimoraic, that syllable alone manages to provide both the head and tail of the foot, which makes it unnecessary to draw on preceding syllables to satisfy FOOT BIN . Thus, rather than disyllabic, the foot can now be monosyllabic: [ . . . (H)]. This is how QS analyses account for oxytones such as those in (12). The prosodic structure of ti.ra.dor and de.ser.tar, for instance, would be [tiµ.ɾaµ.(ˈðoµɾµ)] and [deµ.seµɾµ.(ˈtaµɾµ)]. (12)

a.

ti.ra.dor di.gi.tal co.ra.zón sa.ta.nás mo.rro.coy

‘shooter’ ‘digital’ ‘heart’ ‘evil person’ ‘turtle’

b.

de.ser.tar re.don.del me.nor.quín ho.lan.dés ma.lan.gay

‘to desert’ ‘circle’ ‘Minorcan’ ‘Dutch’ ‘plant’

A boost in storage savings is the main appeal of quantity sensitivity. It turns out that about 27% of Spanish prosodic words pattern with those in (12), thereby offering a great opportunity for profit. To be more precise, holding quantity sensitivity responsible for final stress makes it possible to argue that 91% of the lexicon is free from the burden of stress memorization. Harris (1983), Dunlap (1991), Rosenthall (1994), Lipski (1997), and Martínez-Paricio (2013) are some of the many works that have taken this stance. For a generative analysis, the ultimate goal is, of course, to achieve 100% predictability, but this has proven unattainable because the portion of the lexicon that is left – an estimated 9% – includes proparoxytones such as those in (13a) and oxytones such as those in (13b), both of which fly in the face of phonological principles.

280 (13)

Carlos-Eduardo Piñeros

a.

cá.ma.ra ró.tu.lo má.xi.me tíl.bu.ri ím.pe.tu

‘camera’ ‘label’ ‘especially’ ‘tilbury’ ‘impetus’

b.

Pa.na.má ro.co.có ma.cra.mé bis.tu.rí am.bi.gú

‘Panama’ ‘rococo’ ‘macramé’ ‘scalpel’ ‘buffet’

The first of these patterns presents the difficulty that the foot would have to be either ternary or internal (i.e. [ . . . (σ σ σ)] or [ . . . (σ σ) σ]), while the second one is not any easier because the foot would have to be either unary or iambic (i.e. [ . . . (σµ)] or [ . . . (σ σ)]). The challenge lies in the fact that, for any of these alternatives to emerge, the undominated status of the constraints in (8–10) would have to be suspended for certain words, a move that would be no different from claiming that a single language can have multiple grammars.4 In a theory conceding such an option, coexisting grammars in which proparoxytones and oxytones are optimal can be easily set up by demoting MAIN -RIGHT, FOOT BIN , or TROCH below some antagonistic constraint; however, the merits of this approach are counterbalanced by a lack of explanatory power given that there would be no limit on the kinds of prosodic structures that could be generated through the same strategy. One could claim, for instance, that there is a parcel of the Spanish lexicon where all of the principles in (8–10) have been subdued by antagonistic ones, in which case a severely ill-formed structure such as *[σ (σ σ σ) σ σ σ ] would be sanctioned.5 While it is clear that quantity sensitivity is of no use in accounting for the patterns in (13), its supporters have found another application for it. They claim that heavy syllables cause the three-syllable window to narrow down to two or even one syllable (Harris 1983, 1992a, 1995, Roca 1991, 1997a, 2006a, Shelton 2007, among others). The restrictions inducing such effects are synthesized in (14–15).

4 There are plenty of analyses (both within rule-based and constraint-based models) that have taken this path. Roca (1988, 1990, 1997a, 2005, 2006), Rosenthall (1994), Harris (1995), Hammond (1995), Lipski (1997), Oltra-Massuet and Arregi (2005), Gibson (2011), and MartínezParicio (2013), and Meinschaefer (2015). Some of these proposals openly declare the adoption of several stress algorithms or constraint rankings (i.e. co-phonologies), while others endorse multiple grammars in a more subtle way: through parochial or lexically-indexed constraints. 5 Note that the large number of syllables in this hypothetical form is not the problem because heptasyllabic words such as parapsicología do exist; yet, they exhibit the expected footing (e.g. [pa.ɾa.si.ko.lo.(ˈxi.a)]).

The phonological weight of Spanish syllables

281

(14) Two-syllable window: a. No antepenultimate stress with a closed penult Example: a.par.te ‘apart’ vs. *a.par.te

(15)

b.

No antepenultimate stress with a rising diphthong in the penult Example: a.cia.go ‘ill-fated’ vs. *a.cia.go

c.

No antepenultimate stress with a falling diphthong in the penult Example: a.cei.te ‘oil’ vs. *a.cei.te

d.

No antepenultimate stress with a closed ultima Example: vo.lá.til ‘volatile’ vs. *vo.la.til

e.

No antepenultimate stress with a rising diphthong in the ultima Example: fa.mi.lia ‘ill-fated’ vs. *fa.mi.lia

One-syllable window: a. No antepenultimate stress with a falling diphthong in the ultima Example: pe.je.rrey ‘fish’ vs. *pe.je.rrey b.

No penultimate stress with a falling diphthong in the ultima Example: sa.mu.rái ‘samurai’ vs. *sa.mu.rai

According to (14a–c), antepenultimate stress is illicit whenever the penult is heavy. This generalization makes perfect sense because when a word ends in a HL sequence there is always enough material within these two syllables to construct a wellformed foot (e.g. [ . . . (HL)]); however, the other restrictions on the list are not as reasonable. The behavior of ultimas is the matter of concern. In congruence with standard quantity-sensitivity assumptions, one would expect that when a word ends in a heavy ultima both penultimate and antepenultimate stress would be ruled out because that syllable contains enough moras to support a wellformed foot by itself (e.g. [ . . . (H)]); yet the restrictions in (15a–b) stipulate that stress must remain on the ultima solely if this syllable contains a falling diphthong. Oddly enough, if the ultima is closed or if it contains a rising diphthong, all it can rule out is preantepenultimate stress (14d–e). This raises the following question: if all heavy syllables are bimoraic, why do they differ in their ability to condition stress mobility within the window? QS analyses tend to look away from this matter to emphasize that, if quantity sensitivity were rejected, the restrictions in (14–15) would be a mystery; however, an unbiased opinion will recognize that, even with the adoption of quantity sensitivity, some of these restrictions do not make sense.

282

Carlos-Eduardo Piñeros

A supplementary argument for quantity sensitivity comes from /s/-hypercorrection (Núñez Cedeño 1986, 1988). In varieties of Dominican Spanish where syllable-final /s/ is generally deleted, hypercorrected forms featuring an intrusive [s] in that position may arise. Compare, for instance, standard yo ‘I’ with hypercorrected yos. An interesting claim that has been made about this phenomenon is that [s] cannot be inserted in the coda of a penult belonging to a proparoxytone (Núñez Cedeño 1986, 1988, Harris 1992a). It is said, for instance, that hi.po.pó.ta.mo ‘hippopotamus’ can be hypercorrected as his.po.pó.ta.mo, hi.pos.pó.ta.mo, or hi.po.pós.ta.mo but never as *hi.po.pó.tas.mo. This judgment has been construed as evidence for quantity sensitivity on the reasoning that the absence of the latter form is what one expects to see if the inserted coda consonant adds weight to the penult. Sensible as this may seem, I show next that the argument does not withstand close scrutiny because it is based on an inaccurate characterization of the process.6 Morgan (1998) analyzes the speech of a tourist guide who, in a sample of 3,100 words, produced an impressive number of hypercorrections: 115. One of the revelations made by this study is that there is a striking parallelism between the deletion and insertion of /s/. This conclusion is grounded on two discoveries: neither of these processes is bound to a specific morphological or syntactic category and both go through an intermediate stage in which the outcome is a glottal sound: [h]. It had been previously said that the insertion of this variant was precluded by that of a fully articulated [s] (Núñez Cedeño 1988: 331); however, Morgan (1998: 90) demonstrates that the former is not only possible but, in fact, quite common (45/115 = 39%). More relevant to the ongoing discussion is the discovery that there are several contextual factors which promote insertion; notably, word-final position and the presence of a following voiceless stop. In Morgan’s corpus, 93 of the attested hypercorrections occurred at the end of the word (80.87%), while 68 happened before a consonant, which, regardless of whether it belonged to the same or the following word, was preferentially a voiceless stop (56/68 = 82.35%). The strong influence exerted by these factors has been corroborated by two recent studies. Bullock and Toribio (2010: 20) report that 78 of the 103 hypercorrections present in their corpus happened word finally (75.73%), while 72 occurred in the presence of a following consonant, which, regardless of whether it belonged to the same or the following word, was a voiceless stop in the overwhelming majority of cases (69/72 = 95.83%). Similarly, 6 It is worth mentioning that /s/-hypercorrection has also been said to be informative for other matters, including the maximal size of the Spanish syllable rhyme, the structure-preserving nature of hypercorrection rules, and the inalterability of geminates (Núñez Cedeño 1986, 1988). Not being relevant for stress assignment, such ramifications are not explored here.

The phonological weight of Spanish syllables

283

the statistical analysis presented in Bullock et al. (2014: 31) reveals that 70 of the 82 attested hypercorrections were word final (87.37%), while 57 took place before a consonant, which, regardless of whether it belonged to the same or the following word, was once again a voiceless stop in most cases (48/57 = 84.21%). The consistency of these results enables Bullock et al. (2014: 24) to reject the claims that /s/-hypercorrection belongs to the lexical level and that an intervening word boundary prevents it from ‘seeing the following segment’ (Núñez Cedeño 1988: 326). They demonstrate that this view is amiss because, just like the insertions occurring word internally, those occurring word finally are highly sensitive to the presence of a following voiceless stop. It follows from this that the process in question belongs to the phrase level and, since it is indisputable that primary stress is assigned at the word level, the conclusion emerges that hypercorrection patterns cannot tell us anything about the feasibility or nonfeasibility of stress patterns. In light of the above, let us reconsider the question of why *hi.po.pó.tas.mo has not yet appeared in any /s/-hypercorrection corpus. It is easy to see that all syllables preceding or following the penult of hi.po.pó.ta.mo make better insertion targets because they are either in contact with a following voiceless stop or in final position. That is to say that, since ta is an internal syllable and since the consonant that follows it is a voiced nasal, the probability that it be chosen as the locus of insertion is lower than that of all of its contenders. It must be recognized, however, that this reasoning does not discard *hi.po.pó.tas.mo altogether because the aforementioned fieldwork studies also found that, even in the absence of the favoring segmental and positional conditions, insertion can still take place (e.g. pres.ña.da < pre.ña.da ‘pregnant’). The idea then comes to mind that *hi.po.pó.tas.mo might be impossible for prosodic reasons; more specifically, because the intrusive consonant changes the light penult into heavy, thereby enabling it to capture stress and preclude the proparoxytone pattern. This hypothesis is not defensible, however, for two reasons. First, because it has already been established that hypercorrection belongs to the phrase level (i.e. it applies at a stage when words have already been stressed) and, second, because evidence showing that this process is perfectly compatible with proparoxytones has also been unearthed (Bullock et al. 2014: 21). Consider the sentence Mi nombre e’ Eri[s]ka ‘My name is Erika’, which illustrates both deletion and insertion, one happening in the verb es ‘is’, the other in the noun E.ri.ka.7 This example is especially noteworthy because it proves that the penult of proparoxytones is a feasible insertion target, thereby invalidating the claim that 7 The latter word is a proparoxytone, which, being of foreign origin is allowed to flout orthographic conventions (i.e. it may be written without an accent mark).

284

Carlos-Eduardo Piñeros

hypercorrections such as *hi.po.pó.tas.mo are impossible due to their prosodic configuration. Other examples of this sort were reported in the past, albeit not within the frame of a systematic fieldwork study. Lipski (1997: 588) mentions sá.bas.do from sá.ba.do ‘Saturday’ and ór.gas.no from órgano ‘organ’, which he incidentally observed in Caribbean and New Mexico Spanish. When the rectifications made by the studies cited above are incorporated, the characterization of /s/-hypercorrection becomes irreconcilable with the claim that this process supports quantity sensitivity. One of the obstacles to maintaining such a view is the paucity of relevant examples given that hypercorrection is rare overall, but especially so in word-internal position, the environment where the allegedly impossible pattern needs to be tested. In point of fact, the context that supplies most examples (i.e. word-final position) transparently shows that this is a phrase-level process; hence, inconsequential for lexical stress. The lack of grounding for the allegations is further exposed by the fact that, among the few relevant examples, there are bona fide counterexamples (e.g. E.ri[s].ka from E.ri.ka). I therefore conclude that /s/-hypercorrection cannot be included in the defense for quantity sensitivity. To recapitulate, the case for Spanish quantity sensitivity rests on two main arguments: economy, concerning the pattern in (12), and restrictiveness, regarding the width of the accentual window (14–15). We have seen that, with the adoption of quantity sensitivity, about 27% of the lexicon can be added to the 64% whose stress can be generated by the constraint system in (11); thus, approximately 91% of Spanish words would not require stress marking. Furthermore, window-narrowing effects are expected to arise because, if the system is both right-bounded and quantity-sensitive, the weight of the rightmost syllable will determine whether or not the main foot will need to incorporate previous syllables in order to achieve binarity. It should be noted that, while the argument based on economy is founded on positive evidence (12), that based on restrictiveness depends entirely on negative evidence. In effect, at the heart of the restrictions in (14) is the presumption that proparoxytones with a heavy penult do not exist (e.g. *a.par.te, *a.cia.go, *a.cei.te) and, likewise, the restrictions in (15) presuppose that proparoxytones and paroxytones with a falling diphthong in the ultima are nonexistent (e.g. *pe.je.rrey, *sa.mu.rai). Since this is an empirical issue – similar to the one we encountered with /s/-hypercorrection above – a simple way to refute the window-narrowing restrictions is to show that the structures they prohibit do in fact exist. This is the tactic that Larramendi (1729), Roca (1988), Ohannesian (2004) and a series of experimental studies including Bárkányi (2002a, b), Alvord (2003), Face and Alvord (2005) and Face (2006) have used to

The phonological weight of Spanish syllables

285

counter quantity-sensitivity claims. The data assembled by these works speak strongly in favor of quantity insensitivity; however, being confined to loans and nonce words, such evidence tends to be downplayed on the grounds that it is representative of peripheral systems rather than the core grammar. To make the case for quantity insensitivity compelling, it is therefore essential that both arguments for quantity sensitivity be set straight. One must prove not only that the restrictions in (14–15) do not hold, but also that (12) is not the product of quantity sensitivity. I shall work on the latter task first.

4 Quantity sensitivity requires extrametricality One of the disadvantages of analyzing the Spanish stress system as QS is that tampering with the weight of heavy syllables becomes a necessity. The pattern illustrated in (16) is one of several cases in which the weight of the final syllable needs to be altered to obtain the right results. (16)

ca.té.ter di.fí.cil i.ma.gen ba.ga.ces Ji.mé.nez fa.mi.lia

‘catheter’ ‘difficult’ ‘image’ ‘type of cheese’ ‘last name’ ‘family’

Words like these pose a challenge for quantity sensitivity because they end in a syllable that qualifies as heavy, yet fails to attract stress (cf. 12). This requires that a provision be made to reduce the weight of heavy syllables. A common strategy to deal with (16) is to declare the last mora of the word extrametrical; that is to say that, while present, that unit is invisible (Harris 1983, den Os and Kager 1986, Dunlap 1991, Rosenthall 1994, among many others). By way of illustration, ca.té.ter and di.fí.cil would be [kaµ.(ˈteµ.teµɾ)] and [diµ.(ˈfiµ.siµl)], where angle brackets identify the extrametrical mora. It is apparent from these representations that the ultima could not be stressed under these conditions because the foot would have to be unary (e.g. *[kaµ.teµ.(ˈteµɾ)] and *[diµ.fiµ.(ˈsiµl)]). The thesis advanced by this analysis is then that the pattern in (16) is due to an extrametricality mark, which the lexical entry of any word that behaves in this unexpected way must carry. In light of the fact that the frequency of this pattern is extremely low (0.56%), lexical marking seems to be an appropriate way to

286

Carlos-Eduardo Piñeros

deal with it; but it should be pointed out that, by adopting this solution, generative analyses openly concede that Spanish stress is not fully predictable. An immediate issue that arises with the introduction of extrametricality is that, if any number of prosodic units could be extrametrical, there would be no limit on how far stress could retract; yet, unlimited stress retraction cannot be allowed because it would make it impossible to account for the existence of right-bounded systems. In anticipation of this problem, Hayes (1980) proposes to restrict extrametricality to a single prosodic constituent located at the right periphery of the stress domain, a condition that QS analyses of Spanish must adopt in order to ensure that, in words such as those in (16), stress does not retract to the antepenult. On this strict interpretation, prosodic structures such as *[(ˈkaµ.teµ.teɾ)] and *[(ˈdiµ.fiµ.sil)] are illicit because they include a non-peripheral extrametrical unit. Restricting extrametricality in this way does not prevent it, however, from inducing more than one pattern of stress retraction. We have already seen that, when applied to a heavy ultima, extrametricality renders that syllable light, in which case stress is pushed back to the penult (16). A second form of retraction arises when extrametricality is applied to a light ultima, in which case that syllable becomes weightless and stress has no choice but to land on the antepenult. To illustrate this effect, let us return to the pattern in (13a). In the case of pí.ca.ro, for instance, strict extrametricality yields [(ˈpiµ.kaµ.ɾo)], whose foot is binary and properly aligned, despite including three syllables. Considering that the number of words that behave in this way is relatively small (8.05%), economy does not suffer terribly if they too are granted extrametricality marks; however, this group is sizable enough to prove that the parcel of the lexicon with unpredictable stress is not minute. If extrametricality were limited to the above exceptional cases, it would serve the objective of the generative program quite well; however, the number of words requiring extrametricality marking is so high that it becomes a liability. An unfortunate consequence of adopting quantity sensitivity is that even the words that follow the most common patterns require tampering with the weight of the final syllable. This is made evident by singular/plural pairs such as those in (17), which show that the heavy ultima resulting from the addition of the plural marker (i.e. –s) fails to attract stress. Given that pluralization is the main source of word-final consonants in nouns and adjectives, one can infer that this behavior is widespread in those lexical categories. Verbal forms are not different in this respect. The representative examples in (18) evince that the word-final consonant contributed by the markers of several person/number morphemes does not add weight to the ultima either and, since all tenses accept these inflectional categories, it is obvious that the phenomenon is commonplace in verbs too.

The phonological weight of Spanish syllables

287

Verbal and non-verbal inflections thus attest that the weight of heavy ultimas is systematically ignored.8 (17)

(18)

a.

mo.rro.coy / mo.rro.co.yes ti.ra.dor / ti.ra.do.res co.ra.zón / co.ra.zo.nes

‘turtle’ ‘shooter’ ‘heart’

b.

ve.ci.no / ve.ci.nos la.gu.na / la.gu.nas co.pe.te / co.pe.tes

‘neighbor’ ‘lagoon’ ‘tuft’

c.

pí.ca.ro / pí.ca.ros pá.li.da / pá.li.das cé.li.be / cé.li.bes

‘sly’ ‘pale’ ‘celibate’

a.

de.fi.no / de.fi.nes

‘I define’ / ‘you define’

b.

co.lo.ca / co.lo.can

‘he sets’ / ‘they set’

pa.ga.ba / pa.gá.ba.mos

‘I paid’ / ‘we paid’

Given such a plethora of counterexamples, insisting that Spanish stress is quantity sensitive requires that an exorbitant number of extrametricality marks be posited, but this is inadvisable because the savings in lexical storage that have been achieved would be compromised. Note that, if for the purpose of reducing stress marking by 27% one conceded that most words require extrametricality marking to obtain their inflected forms, no actual savings would ensue. To keep the cost of extrametricality low, QS analyses turn to morphology; they capitalize on the distribution of inflectional morphemes to generate the majority of extrametricality marks. Harris (1983: 115) proposes, for example, a rule of predictable extrametricality, whose effect is to render all word-final inflectional consonants extrametrical. While this move significantly reduces the need for storage, it is not a panacea; its benefits are counterbalanced by the fact that some of the data cannot be derived. Singular/plural pairs such as those in (17c) are one of the problems. We have already seen that proparoxytones can be generated by declaring the only mora of a light ultima extrametrical (e.g. [(ˈpiµ.kaµ.ɾo)]); however, their plural forms cannot be derived in the same fashion because a single extrametrical unit is not enough to guarantee that stress will remain on the antepenult when the ultima is heavy: [(ˈpiµ.kaµ.ɾos)]. As seen in this example, two extrametrical units 8 In (18a), (18b), and (18c), the endings –s, –n, and –mos are the markers of ‘second person singular’, ‘third person plural’, and ‘first person plural’, respectively.

288

Carlos-Eduardo Piñeros

are needed for that purpose. The issue is then that stress would have to fall on the penult in such plurals (e.g. *[piµ.(ˈkaµ.ɾoµs)]), which is contrary to fact. Incorrect predictions are also made for the plural forms of words following the pattern in (16), which, as the examples in (19) show, pluralize in two different ways; those that are vowel-final take –s (19a), while those that are consonant final take –es (19b). As concerns the plural forms in (19a), the prediction quantity sensitivity makes is that stress will fall on the ultima because the presence of three segments in the rhyme of that syllable renders it trimoraic; hence, capable of supporting a binary foot, even if extrametricality conceals one mora (e.g. *[paµ.laµ.(ˈsiµ̯ oµs)]). Regarding the plural forms in (19b), the expectation is that stress will be penultimate because pluralization introduces in this case not one, but two segments, both of which are incorporated into the rhyme of a new heavy ultima. Since strict extrametricality cannot render heavy syllables entirely weightless, the farthest left that stress can fall in such forms is the penult (e.g. *[diµ.fiµ.(ˈsiµ.leµs)]). It is easy to see that lifting the restrictions on extrametricality is the only way in which this approach could generate the correct plurals, but doing so would compromise the three-syllable window (e.g. *[(ˈdiµ.fiµ.si.les)]. In a nutshell, the problem with extrametricality is that it undergenerates when restricted, and it overgenerates when relaxed. (19)

a.

fa.mi.lia / fa.mi.lias pa.la.cio / pa.la.cios de.sa.güe / de.sa.gües

‘family’ ‘palace’ ‘drainage’

b.

ca.té.ter / ca.té.te.res di.fí.cil / di.fí.ci.les i.ma.gen / i.má.ge.nes

‘catheter’ ‘difficult’ ‘image’

The deficiencies exposed above make it clear that the marriage between quantity sensitivity and extrametricality is highly undesirable. From a descriptive point of view, this union is harmful because it makes it impossible to segregate the patterns that the system rejects from those that it allows and, from an explanatory point of view, it is vacuous, for it is circular to posit heavy syllables and subsequently render one of their moras invisible. These problems are flagrant in the treatment of the ultima. In a prosodic system that is right-bounded and projects a single foot per word, the ultima is the position where the distinction between light and heavy syllables is expected to play a decisive role; yet, the mission that extrametricality is brought in to accomplish is to deprive most heavy ultimas of the property that makes them so. That being the ultimate result, there is no point in positing heavy syllables in the first place.

The phonological weight of Spanish syllables

289

5 The role of morphology Generativists have long admitted that a purely phonological account of Spanish stress is impractical because certain stress patterns cannot be derived from sound structure alone. Recall, for instance, that there are words which, despite having identical segmental strings, are accented differently (1). In such cases, morphological constituency must also be taken into account. This was at first thought to be necessary only for verbs (Harris 1969), but it was eventually realized that morphology impinges on stress assignment across all lexical categories (Harris 1975, Hooper and Terrell 1976, Whitley 1976, Fink 1978, Elman 1979, Herslund 1982, Núñez Cedeño 1985, Otero 1986, and most other subsequent work). To disentangle the various ways in which morphology bears on accentuation, I propose that, according to their prosodic behavior, Spanish morphemes fall into three classes: a) stress abiders, b) stress repellers, and c) stress attractors. Developing ideas sketched in Piñeros (2016), I argue next that ordinary morphemes do not interfere with the placement of stress, but there are exceptional ones which can displace it from the expected landing site. Some push it ], while others push it rightwards (e.g. [ . . . ]; yet, leftwards (e.g. [ . . . in both cases displacement is minimal – no greater than one syllable.

5.1 Stress abiders Prosodically normal morphemes are stress abiders. What I mean by this is that they do not interfere with the assignment of stress by the phonology; more specifically, by the constraints that generate a word-final syllabic trochee (8–10). Consider in this regard the examples in (20), where the middle column presents the morphological segmentation of the word. The following conventions are adopted. The inner brackets demarcate the ‘stem’, the constituent hosting the root and any accompanying derivational suffixes. The outer brackets demarcate the morphological word, which, in the patrimonial lexicon, consists of the stem plus a ‘desinence’. The latter is the stretch of morphs actualizing various grammatical meanings, including phi features (e.g. gender, number, and person) and morphological class.9 The root morpheme is underlined to facilitate its identification. 9 As a detailed discussion of Spanish inflection would take us too far afield, the reader is referred to some of the most useful sources on this topic: Harris (1980, 1985, 1991, 1992b, 1996) and Roca (1989, 2000, 2005b, 2006b).

290 (20)

Carlos-Eduardo Piñeros

em.bus.tes

[ [embust]n + e + s ]

‘clever lie, plural’

em.bus.te.ros

[ [embust + er]n + o + s ]

‘clever lier, plural’

em.bus.te.rí.as

[ [embust + er + í]n + a + s ]

‘clever lying, plural’

The fact that all words in (20) are paroxytone suggests that their morphemes hold no prosodic irregularity. As a matter of fact, every member of the stem and the desinence here can be shown to be a stress abider. To do so, let us begin by examining the contents of the stem. It will be helpful to clarify from the start that, although stress abiders often contribute the segmental material for the syllable that ends up bearing stress, this does not warrant their qualification as stress attractors. The unmistakable trait of stress abiders is not that they accept stress, but rather that they do nothing to deflect it from its normal locus. It follows from this that the prosodic status of such morphemes depends entirely on their distribution. When a stress abider happens to be in the position where the phonological principles in (8–10) dictate stress should fall, it will be stressed; otherwise, it will be unstressed. Applying this criterion to the examples in (20) leads to the conclusion that the stems [embust]n, [embust+er]n, and [embust+er+í]n consist exclusively of stress abiders because each morpheme therein receives stress only when it is propitiously located. A clarification also needs to be made concerning the desinence. It is apparent from the examples in (20) that this part of the word is consistently unstressed; nonetheless, this does not warrant the conclusion that its members are stress repellers. In actuality, desinential morphemes are prosodically regular for they do nothing to deflect stress from the penult. This point is closely related to a misconception too often found in the literature, which is that the stem is the only part of the word that matters for stress assignment; in other words, that the stress domain excludes the desinence (Hooper and Terrell 1976, Otero 1986, Roca 1988, 1990a, b, 1997b, 2005a, 2006b, this volume, Harris 1992a, 1995, Hualde, 2012).10 10 A qualification that needs to be made is that not all of the works cited immediately above admit to be supporters of the hypothesis that the stem is the relevant domain. Some of them endorse that view in an implicit manner because, while expressing discontent with the dismissal of the word, they draw on formal devices which bind them to the opposite stance. In Harris (1992a: 28), for instance, the statement is clearly made that stress is to be computed on the word, not just the stem; yet, the following provisions, which target –V(s) desinences, are part of the algorithm proposed therein (p. 30). a) An inflectional consonant is extrametrical b) A word-final vowel is extrametrical. Since the adoption of these stipulations entails that all words ending in a –V(s) desinence will have their final syllable ignored, such a proposal is in effect identical to the declaration others openly make that the stem, rather than the word, is the stress domain.

The phonological weight of Spanish syllables

291

Contrary to that view, I contend that the role of the desinence is no less important than that of the stem because without either one of them a word-final syllabic trochee could not be constructed. Recall from the evaluation in (11) that the footing [. . . (σ σ)] – the structure that explains why paroxytones are so abundant – would not be possible without the participation of the ultima, the syllable where the segments of desinential morphemes typically end up. It follows from this that excluding the desinence from the stress domain would be a costly blunder because it would preclude the possibility of employing universal principles to derive the regular pattern (8–10). There is no denying that resorting to language-specific constraints to accent two thirds of the lexicon would be a major setback in the agenda of the generative program. Another unfortunate consequence of expelling the desinence from the stress domain is that footing would necessarily have to start at the penult and proceed leftwards (i.e. [ . . . (σ σ) σ]), as a result of which a fundamental generalization would be missed: Spanish prosodic feet are trochaic (σ σ); they embody a relationship between a prominent and a non-prominent element sequenced in that specific order.11 The key to understanding the prosodic status of the desinence lies precisely within the canonical foot. It is easy to verify in (20) that the distribution of desinential morphemes is such that their segmental material invariably ends up in the last syllable of the word, exactly where the tail of a final syllabic trochee would be. It follows naturally from this that, if the desinence does not bear stress, it is neither because it lies outside the stress domain nor because it is extrametrical, but simply because it is in charge of supporting the foot tail (e.g. [em.(ˈbus.t ̪es)], [em.bus.(ˈt ̪e.ɾos)], [em.bus.t ̪e.(ˈɾi.as)]). This behavior is perfectly normal and, thus, congruent with the claim that the words in (20) consist exclusively of stress abiders.

5.2 Stress repellers With the above understanding of what prosodically normal morphemes are, let us consider the irregularity introduced by stress repellers. These are morphemes with an exceptional ability: they can exempt one of their vowels from the duty 11 Truncation provides unambiguous evidence that Spanish prosodifies its words by building syllabic trochees (see Prieto 1992, Lipski 1995, Colina 1996, Piñeros 2000a, b, Felíu 2001, Roca and Felíu 2003). Creations such as [(ˈpo.li)] from [po.li.(ˈθi.a)] ‘police’, [(ˈʧa.ʧo)] from [mu.(ˈʧa.ʧo)] ‘boy’, [(ˈka.ɾo)] from [ka.ɾo.(ˈli.na)] ‘Caroline’, and [(ˈti.na)] from [kɾis.(ˈti.na)] ‘Christine’ conspicuously show that the prosodic word is erected on two syllables and that greater prominence is granted to the first one.

292

Carlos-Eduardo Piñeros

of supporting the foot head. In effect, when belonging to certain morphemes, any of the five Spanish vowels – /i, e, a, o, u/ – can elude stress, even if that segment belongs to the syllable that the principles in (8–10) designate as the optimal stress bearer. When this happens, stress is forced to land on the next available syllable, thereby causing the foot to engulf the antepenult: [ . . . (σ σ σ)]. The examples in (21) and (22) illustrate this phenomenon with a common stress repeller, the suffix –IC ‘relation’. Stress repellers are written in small capitals so that they can be readily identified. 12 (21)

(22)

a.

to.no

[ [ton]n + o ]

‘tone’

b.

to.na.da

[ [ton + ad]n + a ]

‘tune’

c.

tó.ni.co

[ [tón + IC ]a + o ]

‘tonic’

a.

car.ne

[ [carn]n + e ]

‘meat’

b.

car.na.da

[ [carn + ad]n + a ]

‘bait’

c.

cár.ni.co

[ [cárn + IC ]a + o ]

‘meat related’

In (21a) and (22a), the stem consists solely of the root, which happens to be a stress abider; while in the other two examples of each set, the stem includes the root and a suffix. Observe that, in (21b) and (22b), the suffix –ad ‘collectivity’ proves to be a stress abider because it is the source of the vowel that heads the penult and this syllable is willing to head the foot. This state of affairs contrasts sharply with that in (21c) and (22c), where the assignment of stress to the penult is frustrated by the suffix –IC . The most important observation is that, despite being properly located to receive stress, the vowel supplied by that morpheme refuses to support the foot head, thereby forcing the vowel of the antepenult to shoulder the responsibility: [(ˈto.ni.ko)] and [(ˈkaɾ.ni.ko)]. That the prosodic anomaly emanates from –IC is confirmed by the fact that all adjectives derived with this suffix have antepenultimate stress. Some additional examples are ba.s+e ‘base’ > bá.s+ I .C + o ‘basic’, co.n+o ‘cone’ > có.n+I . C + o ‘conical’, es.fe.r+a ‘sphere’ > es.fé.r+I . C + o ‘spherical’. Numerous other suffixes behave in this peculiar way. To name but a few, the list of Spanish stress repellers includes –UL ‘diminutive’ (e.g. glo.b+o ‘globe’ > 12 It is pertinent to mention that Spanish also has the suffix –ic ‘diminutive’, which, despite being segmentally identical to – IC ‘relation’, is a different morphological entity. An example where the former is present is the noun a.ba.ni.co (i.e. [[aban+ic]n+o] ‘fan’), derived from the primitive word [aban]n+o] ‘fan’. Here we see that –ic ‘diminutive’ is a stress abider; hence, does not require special comment. Our attention at this time is on – IC ‘relation’ because its prosodic behavior is unusual.

The phonological weight of Spanish syllables

293

gló.b+U. L + o ‘globule’, cir.c+o ‘circus’ > cír.c+U. L + o ‘circle’), –ID ‘quality’ (e.g. mor.b+o ‘morbidity’ > mór.b+I . D + o ‘morbid’, ray+a ‘ray, fish’ > rá.y+I . D + o ‘shaped like a ray’,), –FON ‘sound’ (e.g. an.gl+o ‘English’ > an.gl+ó+.FO. N + o ‘English speaking’, fran.c+o ‘French’ > fran.c+ó+.FO. N + o ‘French speaking’). Furthermore, there are also quite a few root morphemes which exempt their last vowel from the obligation of acting as foot head. Some examples of this are BÚ. FA . L +o ‘buffalo’, SÍ . LA . B +a ‘syllable’, CHÉ . VE . R +e ‘nice’, and ím.PE . T +u ‘impetus’. It is important to highlight that stress repellers are restricted in a principled manner. To be able to exempt one of its vowels from serving as foot head, the morpheme in question must be the head of the morphological word, a role that Spanish systematically assigns to the last member of the stem. A brief digression to introduce this concept is necessary. The last member of the stem earns the designation of morphological head by virtue of being the morpheme that determines key morpho-syntactic properties of the word, such as syntactic category, morphological class, gender (for nouns), and tense and mood (for verbs). A comparison between a primitive word and its derivatives helps to substantiate this claim (23). (23)

a.

[ [lan]n + a + s ]

‘wool, plural’

b.

[ [lan + os]a + o/a + s ]

‘wooly, plural’

c.

[ [lan + os + idad]n + Ø/e + s ]

‘fuzz, plural’

In the primitive word (23a), the root encounters no rival in securing the role of morphological head because it is the only member of the stem. This privilege empowers that morpheme to rule that the word will be a noun, with feminine gender, and a member of morphological class A.13 In the first derivative (23b), by contrast, the suffix –os takes on the role of morphological head and determines, in that capacity, that the word will be an adjective and follow default morphological-class assignment (i.e. class O or class A, depending on whether masculine or feminine gender is acquired via agreement). The second derivative, (23c), shows that the suffix –idad replaces –os in the role of morphological head, thereby gaining the power to rule that the word will be a noun, with feminine gender, and a member of morphological class Ø/E.14 The point of 13 After Roca (2005b, 2006b), I use mnemonic labels (e.g. class A for –a, class O for –o, class E for –e, and so on) to designate Spanish morphological classes. 14 This is a peculiar morphological class because it alternates; it surfaces as Ø in the singular, but as –e in the plural.

294

Carlos-Eduardo Piñeros

greatest significance is that the morpheme that governs the morpho-syntactic properties of the word is consistently the last member of the stem. Returning to the behavior of stress repellers, the condition that no such morpheme can enforce its irregularity unless it is the head of the morphological word provides a principled explanation for the fact that stress retraction is never iterative. Even if a word includes several repellers, stress cannot retract by more than one syllable because, since there is a single morphological head per word, only one repeller can enjoy privileged status; all others have to behave as regular morphemes. This phenomenon is illustrated in (24–25), where the stem of each word consists exclusively of stress repellers. In (24a) and (25a), the position reserved for the morphological head is occupied by the root, by virtue of which this morpheme manages to relieve its last vowel from the duty of serving as foot head. In (24b) and (25b), by contrast, the role of morphological head is taken over by a suffix, which deprives the root of its authority. The upshot of this shift of power is that the prosodically irregular vowel of the suffix now gets to be excused, while that of the root is forced to fulfill its prosodic duty. (24)

(25)

a.

có.le.ra

[ [CÓLER ]n + a ]

‘anger’

b.

co.lé.ri.co

[ [COLÉR + IC ]a + o ]

‘angry’

a.

fós.fo.ro

[ [FÓSFOR ]n + o ]

‘phosphorous’

b.

fos.fó.ri.co

[ [FOSFÓR + IC ]a + o ]

‘phosphoric’

The proposed condition also explains the regularization of proparoxytones via suffixation. When this word-formation process introduces a stress abider in the governing position, the prosodic irregularity of any other morpheme within the stem is obliterated. In (26b–c), we witness, for instance, how a root, which induces stress retraction when it is final in the stem (26a), can do nothing to that effect when a stress abider pushes it out of that position. Stress-repelling suffixes can be disempowered in this way too, as evinced by the examples in (27b–c). (26)

(27)

a.

es.pá.rra.go

[ [ESPÁRRAG ]n + o ]

‘asparagus’

b.

es.pa.rra.gue.ra

[ [ESPARRAGU + er]n + a ]

‘asparagus plant’

c.

es.pa.rra.ga.do

[ [ESPARRAGU + ad]n + o ]

‘asparagus stew’

a.

gló.bu.lo

[ [glób + UL]n + o ]

‘globule’

b.

glo.bu.lo.so

[ [glob + UL + os]a + o ]

‘globulous’

c.

glo.bu.li.na

[ [glob + UL + in]n + a ]

‘globulin’

The phonological weight of Spanish syllables

295

The existence of stress repellers is relevant to the quantity-sensitivity debate because it confirms that extrametricality is unsuitable to formalize stress retraction. The data in (21–27) leave no doubt that the abnormal prosodic behavior is induced by the last morpheme of the stem, a discovery that leads Harris (1983: 115) to forsake strict extrametricality and hold instead that finality in either the word or the stem is a sufficient condition to warrant prosodic invisibility (e.g. oasis [oµ.(ˈaµ.siµs)] ‘oasis’, nómada [(ˈnoµ.ma.ðaµ)]). By making this amendment, Harris aims to reconcile the theory with the facts, but this is to no avail because the theory becomes too permissive. Consider, for example, his analysis of the noun tó.rax ‘thorax’, segmented as [[torak]+s]. Departure from strict extrametricality is felicitous for this type of word because, as pointed out before, two moras must be written off if stress is to fall on the penult when the ultima is trimoraic (e.g. [(ˈtoµ.ɾaµks)]); however, having that option is counterproductive for other words. For li.bé.lu.la ‘dragon fly’, for instance, things go terribly wrong because the freedom to posit both a word-final and a stem-final extrametrical unit enables stress to fall beyond the antepenult (e.g. *[(ˈliµ.βeµ.lu.la)]). Realizing that the admission of stem-final extrametrical units is a serious problem, den Os and Kager (1986: 37) introduce a rule of extrametricality transfer, whereby the diacritic [+E], which is carried by any stem containing stress repellers, is discharged on the desinence. This treatment guarantees that the extrametricality bearer will be word-final even when the trigger is not, but such a solution is unacceptable because it misrepresents the facts. In cólera, for instance, the derivational stem would discharge the prosodic irregularity of the root on the desinence (28), thereby giving rise to the parsing [(ˈkoµ.leµ).ɾa], where the ultima is made weightless and left unfooted. While consistent with strict extrametricality, this analysis loses touch with reality because it portrays the penult as regular and the ultima as irregular when the data evinces that things are the other way around (21–27).15 (28)

→

Given that neither a strict nor a lax interpretation of extrametricality provides insight into stress retraction, the ineluctable conclusion is that there is no justification for such a mechanism. This result further undermines quantity 15 To obtain the corresponding plural form, den Os and Kager are forced to exclude the plural morpheme from the stress domain, which they accomplish by delaying pluralization to a late stage, the so-called compound level. This analysis thus employs level ordering as an additional way to tamper with the weight of heavy syllables, another ad hoc and unenlightening device.

296

Carlos-Eduardo Piñeros

sensitivity because Section 4 has already demonstrated that, without quantity sensitivity, heavy syllables are intractable.16

5.3 Stress attractors The footing of Spanish prosodic words is also subject to the influence of stress attractors. These are morphemes capable of drawing stress towards one of their vowels, even when that segment is not the one that is best positioned to support the foot head. The distortion caused by stress attractors is palpable in minimal pairs such as those in (29–31), where we witness a curious phenomenon: the existence of morphemes whose signifiers are segmentally identical, but prosodically different. As the middle column reveals, the nearly-homophonous morphemes are tense/mood suffixes sitting at the right periphery of the stem: –e ‘subjunctive’, –É ‘preterit’, –ra ‘past subjunctive’, –RÁ ‘future’, –re ‘future subjunctive’, and –RÉ ‘future’. To facilitate their identification, stress attractors are written in capital letters. (29)

(30)

(31)

a.

for.me

[ [form + e]v + Ø ]

‘may I/he form’

b.

for.mé

[ [form + É]v + Ø ]

‘I formed’

a.

for.ma.ra

[ [form + a + ra]v + Ø ]17

‘might I/he form’

b.

for.ma.rá

[ [form + a + RÁ]v + Ø ]

‘he will form’

a.

for.ma.re

[ [form + a + re]v + Ø ]

‘should I/he form’

b.

for.ma.ré

[ [form + a + RÉ]v + Ø ]

‘I will form’

Another aspect in which these finite verb forms are peculiar is the lack of overt morphs in the desinence. There are two factors behind the zeroes appearing in the corresponding morphological structures (middle column). One is a general morphological process of restructuring, which deletes the first person marker – o in all tenses except for the present indicative (cf. [[form]+o] ‘I form’, [[beb]+o] 16 Analyses which give up quantity sensitivity while clinging on to extrametricality remain unable to explain stress retraction. They may be able to generate proparoxytones through ad hoc devices (e.g. lexically-indexed anti-alignment constraints), but fail to provide a sensible reason for the existence of this phenomenon. A recent example of this approach is Meinschaefer (2015). 17 The suffix appearing immediately after the root in (30) and (31) is the theme vowel, –a, which has several functions including that of signaling conjugational class. For discussion on the derivational status of this morpheme, see Harris (1987, 1989, 1992a).

The phonological weight of Spanish syllables

297

‘I drink’, [[viv]+o] ‘I live’); the other one is that third person and singular number are default values which Spanish never marks. Like repellers, stress attractors are systematically restricted. Their ability to displace stress depends on their position within the stem. The key observation in this regard is that the tense/mood suffixes inducing final stress in verbal forms always occupy the position reserved for the morphological head. This privilege empowers them not only to impose their morpho-syntactic features but also to compel the observance of their prosodic irregularity. I am thus claiming that the prosodic contrasts in (29–31) are due to a difference in morphological composition. While the morphological head of the first member of each pair is a regular morpheme, the morphological head of the second member is a stress attractor, which uses its influential status to circumvent the phonological principles in (8–10). That the suffixes –É ‘preterit’ and –RÁ/–RÉ ‘future’ are the source of the anomaly is corroborated by the fact that they invariably capture stress (e.g. can.t+É ‘I sang’, bai.l+É ‘I danced’, man.d+É ‘I commanded’, can.t+a.+RÉ ‘I will sing’, bai.l+a.+RÉ ‘I will dance’, man.d+a.+RÉ ‘I will command’, can.t+a.+RÁ ‘he will sing’, bai.l+a.RÁ ‘he will dance’, man.d+a.+RÁ ‘he will command’). It goes without saying that discovering a commonality between the two forms of displacement – retraction and advancement with respect to the penult – would be a welcome result because it would offer opportunities for the development of a unified formal analysis in the future. The outlook for this enterprise is indeed positive for it can be demonstrated that the two deviations from the norm are subject to the same structural condition. It turns out that neither type of prosodically-irregular morpheme can impinge on stress assignment unless it is acting as the head of the morphological word. The prediction this makes is that, like repellers, attractors will lose the ability to displace stress when ousted from the governing position, in which case the stress pattern of the word will be decided by the new morphological head. The examples in (32–33) attest that this is exactly how Spanish stress assignment works. (32)

(33)

a.

Is.ra.el

[ [ ISRAEL ]n + Ø ]

‘Israel’

b.

is.ra.e.li.ta

[ [ ISRAEL + it ]a + a ]

‘from Israel’

c.

is.ra.e.lí

[ [ ISRAEL + Í ]a + Ø ]

‘from Israel’

a.

Se.fa.rad

[ [ SEFARAD ]n + Ø ]

‘Sepharad’

b.

se.far.di.ta

[ [ SEFARD + it ]a + a ]

‘from Sepharad’

c.

se.far.dí

[ [ SEFARD + Í ]a + Ø ]

‘from Sepharad’

298

Carlos-Eduardo Piñeros

In each of these sets, there is a primitive word, a toponym, and two derivates, demonyms. The stem of the toponym consists solely of the root (32a and 33a), which, being a stress attractor, exercises its influence as morphological head to draw stress towards its final vowel. This contrasts sharply with the first demonym (32b and 33b), where stress assignment can proceed in congruence with (11) because, since the suffix –it ‘gentilic’ is a stress abider, its rise to the role of morphological head disempowers the root without imposing any other prosodic irregularity. The alternative demonym is relevant because it illustrates the perpetuation of the anomaly. This happens whenever a stress attractor relays another one in the role of morphological head. In (32c) and (33c), the suffix –Í ‘gentilic’ becomes the new morphological head, thereby enabling its only vowel to snatch stress away from the root. Additional examples such as I.RAQ > I.RA.QU+Í, O.MÁN > O.MA.N+Í, and NA.ZAR > NA.ZA.R+Í ratify that this suffix is prosodically irregular.18 The proposal has now been substantiated that the displacement of stress in either direction is governed by a key morpheme: the morphological head. This is proof that morphology exerts a powerful influence over phonology. I submit, as a principal conclusion of this study, that the type of morpheme in charge is the factor that ultimately determines the stress pattern of Spanish prosodic words. Normally, it is an abider that controls the morphological word (because abiders are the most numerous) and, in that case, penultimate stress – as per (11) – is possible. By contrast, when either a repeller or an attractor takes over, minimal displacement to the left or to the right ensues. The interplay between these conflicting forces is nicely illustrated by the adjective is.ra.e.lí.ti.co ‘Israelitic’, whose stem hosts one morpheme of each type: [ISRAEL+ít+IC ]+o].19 Because the repeller is the head of this structure, the stress pattern can be no other than proparoxytone; however, in (32a), where the attractor is in power, stress

18 Following a proposal by Kiparsky (1991), several analyses have argued that catalexis – the absence of segmental content for certain prosodic units – is the source of Spanish oxytones. Nouveau (1994) tries this tactic in combination with quantity sensitivity, while Meinschaefer (2015) does so in combination with quantity insensitivity. The QS analysis draws on a catalectic mora to obtain a wellformed moraic trochee (e.g. [paµ.naµ.(maµ[µ])] for Pa.na.má ‘Panama’), while the QI analysis draws on a catalectic syllable to obtain a wellformed syllabic trochee (e.g. [paµ.naµ.(maµ[σ])], instead). (By convention, catalectic units are represented between brackets.) It is obvious that these analyses miss a crucial point: stress advancement is not induced by an ethereal entity, but by the morphological head. 19 Recall that capital letters are used for attractors, lower case for abiders, and small capitals for repellers.

The phonological weight of Spanish syllables

299

must be oxytone, and in (32b), where the abider is in command, stress must be paroxytone.20 The phenomena uncovered above shed new light on the issue of quantity sensitivity. Stress attractors reveal that the interpretation of the pattern in (12) as evidence for heavy syllables is amiss. The fact that morphemes may end in either a vowel or a consonant is at the heart of the matter. This triviality becomes relevant because, when the desinence is non-overt, the final segment of a stress attractor serving as morphological head ends up in the rhyme of the final syllable (see 29–31). With such a parsing, the anomaly of the stress attractor inevitably reflects on the ultima, thereby creating the illusion that the unit that throws stress forward is phonological, rather than morphological. To discern what is truly happening, let us begin by considering the scenario in which the final member of the stem is a stress attractor ending in a vowel. This is the origin of V-final oxytones, a pattern that is damaging to quantity sensitivity because it evinces that CV, the syllable type that is generally recognized

20 An anonymous reviewer has made an interesting suggestion concerning the formal representation of stress repelling and stress attracting morphemes. Since a reluctance to serve as foot heads is the defining property of stress repellers, the recommendation is made that they be represented as lexically specified foot tails. This is illustrated for the suffix – IC ‘relation’ in (a) below. By the same token, since an undue willingness to serve as foot heads is the defining property of stress attractors, it is suggested that they be represented as lexically specified foot heads. This is illustrated for the suffix –Í ‘gentilic’ in (b) below. (a)

(b)

The reviewer’s recommendation further instructs that a family of FAITHFULNESS constraints be introduced and ranked above the principles responsible for regular stress (8-10), so that the preservation of the lexically-specified prosodic structures can be ensured. A complementary step that needs to be taken with that is to fine-tune the power of FAITHFULNESS because, if this constraint family were absolutely dominant, it would cause overgeneration; that is to say that stress would potentially fall anywhere in the word since all prosodically irregular morphemes would have to be respected unconditionally. To prevent such a disastrous result, additional constraints would need to be deployed, but this is highly undesirable because, on the one hand, they would be of an ad hoc nature and, on the other hand, they would duplicate the work of the principles in (8–10), which, as demonstrated in (11), are the reason why stress gravitates towards the end of the word. The reviewer goes on to acknowledge that, when FAITHFULNESS constraints are used to derive the irregular patterns, the question of whether the system is quantity sensitive or insensitive becomes trivial because the structural complexity of the syllable and foot no longer matter. To me, such difficulties are a red flag, a sign that the formal devices being used are excessively powerful.

300

Carlos-Eduardo Piñeros

as the epitome of light weight, behaves as though it were heavy (e.g. re.cor.t+É ‘I clipped’, re.cor.t+a.+RÉ ‘I will clip’, re.cor.t+a.+RÁ ‘he will clip’, ku.wai.t+Í ‘Kuwaiti’). All phonologists will agree that the fact that stress falls on the final syllable of such words has nothing to do with phonological weight. Because no interpretation of quantity sensitivity would ever hold that syllables with simple rhymes can attract stress, this pattern indisputably shows that the true stress attractor is a morpheme. The structure of the syllable where the vowel of that morpheme ends up is circumstantial. The reason why QS analyses have not been deterred by such pellucid counterevidence is that the number of V-final stress attractors is very small. As a result, their overall impact is little (0.63%). To the handful of suffixes cited above, I can only add a couple more: -Í ‘first person preterit for the second and third conjugations’ (e.g. per.d+Í ‘I lost’, dor.m+Í ‘I slept’) and –Ó ‘third person preterit for all conjugations’ (e.g. sol.d+Ó ‘he soldered’, per.d+i+Ó ‘he lost’, par.t+i+Ó ‘he left’). This scarcity extends to stress-attracting roots, which are clearly exotic (e.g. BO.GO.TÁ ‘Colombian capital’, BER.NA.BÉ ‘Barnaby’, YO.LOM.BÓ ‘Colombian municipality’, CO.LI.BRÍ ‘humming bird’, BEL.CE.BÚ ‘Beelzebub’. Stress attractors ending in a consonant are the source of C-final oxytones (12), which I now proceed to show have nothing to do with quantity sensitivity. The true facts are manifest in (34) and (35). A comparison between these two sets of data reveals that the lack of an overt desinence is directly linked to the emergence of oxytones. Observe that, in (34), the desinence, which includes abstract features for gender (masculine) and for morphological class (alternating), is covert, while in (35) an overt morph standing for feminine gender and morphological class A is present. The covertness of the desinence is pivotal

Another reason why the type of formalization sketched above is not pursued here is that the suggested representations are inadequate on multiple counts. First, they are off-target because, as shown above, the prosodic irregularities of repellers and attractors are not spread over the entire foot but rather localized on a specific segment, which is always a vowel. Second, they are wasteful because they posit fully-fledged prosodic constituents (i.e. syllables and feet), when there are independently needed principles in the grammar that can generate such structures. And third, they are dubious because they require that the structure of morphemes be defined in relation to their neighbors when, in fact, each morpheme merits an independent entry in the lexicon because it has its own meaning, form, and morpho-syntactic properties. In conclusion, I am convinced that stipulating foot structure in underlying representations and forcing its preservation via FAITHFULNESS constraints is not the appropriate way to deal with prosodically irregular morphemes. There are several alternatives that are worth exploring, but space limitations force me to leave the formal aspects of my proposal for a subsequent study.

The phonological weight of Spanish syllables

301

because it is only under that condition that stress falls further forward than phonology dictates (34).21 (34)

(35)

a.

ir.lan.dés

[ [ ir + land + ÉS ]a + Ø ]

‘Irish, masculine’

b.

ho.lan.dés

[ [ ho + land + ÉS ]a + Ø ]

‘Dutch, masculine’

c.

ze.lan.dés

[ [ ze + land + ÉS ]a + Ø ]

‘Zealander, masculine’

a.

ir.lan.de.sa

[ [ ir + land + ES ]a + a ]

‘Irish, feminine’

b.

ho.lan.de.sa

[ [ ho + land + ES ]a + a ]

‘Dutch, feminine’

c.

ze.lan.de.sa

[ [ ze + land + ES ]a + a ]

‘Zealander, feminine’

Given that stress consistently looks for the vowel of the suffix –ÉS (e.g. ir.lan.dés vs. ir.lan.de.sa), the sensible conclusion is that rhyme complexity is irrelevant.22 The actual reason why stress is final in (34) is that –ÉS ‘gentilic’ is a stress attractor governing the morphological word. In that capacity, this suffix is able to draw stress towards its vowel despite the fact that there is another vowel in the domain which is better positioned to act as foot head. Such disregard for phonological considerations warrants the claim that heavy syllables are a morphological mirage. Spanish has stress-attracting morphemes, not stressattracting syllables. In case the reader remained unconvinced, the data in (36) and (37) corroborate that the presence of a coda in the rhyme of the ultima is trivial. It is conspicuous in (36) that the adjective japonés derives from the noun Japón and that both surface with final stress because a stress attractor is the governor of each: JAPÓN in one and –ÉS in the other. Parallel outcomes are not possible for both forms in (37) because, although a stress attractor is also the morphological head, the desinence undergoes a major change; it goes from covert in the singular (37a) to overt in the plural (37b). Because pluralization interpolates a vowel in the desinence and that vowel engenders a new syllable, the need to displace stress vanishes in (37b). Put another way, the demands of the morphology now agree with those of the phonology with the fortunate result that both can be honored. Of special significance is the fact that the segmental identity that exists between the ultimas of (36b) and (37b) is disregarded. What matters is that the ending

21 The data in (34) and (35) evince another morphological quirk, which is that a change in gender can lead to a change in morphological class. In this case, the word goes from class Ø/e in the masculine (e.g. irlandés/irlandeses) to class A in the feminine (e.g. irlandesa/irlandesas). 22 This can be verified by comparing the ultimas of a pair of oxytones such as ir.lan.dés and is.ra.e.lí, where we see that stress is drawn to –ÉS and –Í regardless of the makeup of the syllable.

302

Carlos-Eduardo Piñeros

–es of the word japonés is associated with a stress attractor, whereas that of the word jabones is not. This behaviour certifies that final stress is driven by morphology, not phonology. (36)

(37)

a.

Ja.pón

[ [ JAPÓN ]n + Ø ]

‘Japan’

b.

ja.po.nés

[ [ JAPON + ÉS]a + Ø ]

‘Japanese’

a.

ja.bón

[ [ JABÓN ]n + Ø ]

‘soap’

b.

ja.bo.nes

[ [ JABON ]n + e + s ]

‘soap, plural’

Unlike their V-final counterparts, C-final stress attractors are well represented in the lexicon (27.08%). Suffixes such as –AL ‘relation’ (e.g. ide+AL ‘ideal’, mort+AL ‘deadly’), –CIÓN ‘action/effect’ (e.g. don+a+CIÓN ‘donation’, med+i+CIÓN ‘measurement’), –DOR ‘agentivity’ (e.g. don+a+DOR ‘donor’, med+i+DOR ‘measurer’), –DAD ‘quality’ (e.g. bon+DAD ‘goodness’, mal+DAD ‘wickedness’), and –ÓN ‘intensification’ (e.g. llor+ÓN ‘crybaby’, mir+ÓN ‘nosy’) are very prolific and there are numerous others like them. Roots also contribute to the abundance of oxytones as there are plenty that, being C-final, attract stress towards their last vowel (e.g. MAR– ‘sea’, AZUL– ‘blue’, PARED– ‘wall’, MAÍZ– ‘corn’, BETÚN– ‘shoe polish’, ARROZ– ‘rice’, CICATRIZ– ‘scar’). It is important to add that the main reason for the imbalance in the number of V-final and C-final stress attractors is that, for the sake of smooth concatenation, Spanish designs its morphemes in such a way that most roots end in a consonant, while most derivational suffixes begin with a vowel and end in a consonant (e.g. [[form+AL+iz+a +CIÓN]+Ø] ‘formalization’, [[por+os+IDAD]+Ø] ‘porosity’). This evinces a general preference for C+V morpheme junctures, which fosters C-final stress attractors, but discourages their V-final counterparts. The above analysis demonstrates that the positive evidence that has been adduced in favor of quantity sensitivity is spurious. It should be clear by now that the pattern in (12) is unrelated to syllable weight. Final stress, in both C-final and V-final words, is due to the empowerment of stress-attracting morphemes. In light of these findings, the generativist claim that Spanish stress is mostly predictable must be revised because it cannot stand on strictly phonological grounds. Sound structure suffices to accentuate nearly two thirds of the lexical stock – the 64% represented by V-final paroxytones – but there is a sizeable portion for which morphological information is needed to arrive at the correct accentuation. I agree that there is a further 27% for which stress can also be predicted, but the basis for that prediction is morphological. Given that the

The phonological weight of Spanish syllables

303

desinence is typically instantiated by a vowel, the presence of a consonant at the end of the word signals a gap in the inflectional paradigm and this serves as a telltale that a stress attractor is acting as the morphological head because all such morphemes allow the desinence to be covert when they are in command (e.g. [[ide+AL]+Ø] ‘ideal’, [[don+a+CIÓN]+Ø] ‘donation’, [[don+a+DOR] +Ø] ‘donor’, [[bon+DAD]+Ø] ‘goodness’, [[llor+ÓN]+Ø] ‘crybaby’). It is thanks to this morphological information that final stress is largely predictable and it is solely on these grounds that the predictability rate can be boosted from 64% to 91%. (Recall that the phonological alternative – quantity sensitivity – yields inconsistent results and is, therefore, an unreliable predictor.) Further increases to the predictability rate are not possible because, since there is no contextual property – phonological, morphological, syntactic, or semantic – that correlates with the presence of stress repellers, there is no means whereby additional predictions could be made. This leaves a residue of 9%, consisting mostly of proparoxytones, for which stress is unpredictable (e.g. sá.ba.na ‘sheet’ vs. sa.ba.na ‘savannah’). Another important inference that follows from the above is that a purely morphological analysis, as proposed by Hooper and Terrell (1976), is disadvantageous because it misses the phonological generalization that overarches the largest portion of the lexicon: a final syllabic trochee (11). The fact that this configuration is present in two thirds of prosodic words is so significant that any analysis that misses it is doomed.23 While much research is still needed to arrive at a full understanding of Spanish stress, sufficient progress has now been made to state with confidence that, despite morphological interference, Spanish stress is essentially phonological.

6 Further problems for quantity sensitivity The contention has been made in the past that one of the benefits of positing heavy syllables is that the distribution of stress becomes stricter; hence, more precise (Harris 1983, 1992a, 1995). From this standpoint, any complex rhyme in the penult prevents stress from retracting to the antepenult and one type of complex rhyme in the ultima keeps stress from reaching the penult (14–15). A comprehensive look at the lexicon reveals, nevertheless, that the facts are

23 That is indeed the case of all accounts that skip the ultima: those that explicitly uphold the stem as the stress domain (e.g. Roca 1988), as well as those that argue for the word instead, but treat –Vs inflectional endings as extrametrical (e.g. Harris 1992a).

304

Carlos-Eduardo Piñeros

not exactly so (38–39). Over the years, evidence has been mounting that all of the restrictions attributed to the complexity of syllable rhymes are violable (Ohannesian 2004). (38)

(39)

Counterevidence to the two-syllable window: a. Antepenultimate stress with a closed penult Example: Fró.mis.ta ‘locality in Spain’ b.

Antepenultimate stress with a rising diphthong in the penult Example: Tó.cua.ro ‘locality in Mexico’

c.

Antepenultimate stress with a falling diphthong in the penult Example: vó.lei.bol ‘volleyball’

d.

Antepenultimate stress with a closed ultima Example: ó.mi.cron ‘Greek letter

e.

Antepenultimate stress with a rising diphthong in the ultima Example: mag.ní.lo.cuo ‘grandiloquent’

Counterevidence to the one-syllable window: a. Antepenultimate stress with a falling diphthong in the ultima Example: Tem.per.ley ‘locality in Argentina’ b.

Penultimate stress with a falling diphthong in the ultima Example: yó.quey ‘jockey’

While it is true that the exceptions are limited to toponyms, loans, and learned words, it would be unwise to exclude them from this investigation because their use by speakers who have no knowledge of the donor languages indicates that they have become part of the Spanish lexicon. This leads me to disagree with the claim that this sort of data is irrelevant (Harris 1992a) and subscribe instead to the view that all forms used by native speakers must be explained (Roca 1988, 1990, 2005a, 2006b, this volume, Ohannesian 2004). If the items in question were not sanctioned by the grammar, their stay in the lexicon would be transient because grammatical processing would force them to conform or perish; yet, some of them have existed for centuries and show no signs of remitting (e.g. Frómista and Tócuaro). Another justification for the inclusive approach is that excluding any area of the lexicon from the pool of data would be methodologically unsound for the conclusions would be necessarily partial. Scientific rigor requires that all possible patterns be granted a place in the system, even if their frequency implies that they are marginal. The inclusion of the less common patterns is also demanded by the fact that their numbers are growing, a trend

The phonological weight of Spanish syllables

305

that seems irreversible since borrowing keeps introducing new items (e.g. búmeran ‘boomerang’, pénalti ‘penalty’, ráquetbol ‘racquetball’, póquemon ‘Pokémon’, póney ‘pony’, cówboy ‘cowboy’, and so on). When examined from this broader perspective, the restrictions in (14–15) cannot be accepted as signs of ungrammaticality, they are merely dispreferences.24 Outside the generative literature, there has been strong opposition to the claim that certain syllables reduce mobility within the accentual window by virtue of attracting stress. Numerous experimental studies have resorted to nonce words to test the validity of the window-narrowing restrictions and they too have concluded that the placement of stress to the left of syllables with complex rhymes cannot be ungrammatical because it is not systematically ruled out. In one experiment, speakers were presented with trisyllabic nonce words embedded in full sentences and were asked to judge whether the item written in bold could be a Spanish word, seemed strange, or was utterly impossible (Bárkányi 2002a). Results show that, in the worse cases, the rates of rejection for words containing quantity-sensitivity violations was in the vicinity of 50%; that is to say that the skipping of the alleged heavy syllables was never deemed decidedly impossible (40). This line of research has considerably increased the visibility of quantity insensitivity in recent years. Bárkányi (2002b), Alvord (2003), Face (2004, 2006), and Face and Alvord (2005) also argue, on the basis of nonce words, that Spanish speakers do not distinguish between light and heavy syllables.

24 For reasons that remain largely undetermined – but among which frequency is likely to be a major player – some loanwords progress to more advanced levels of nativization than others. For instance, there are dialects in which pé.nal.ti and vó.lei.bol have modified their prosodic structure to become the more native-like pe.nal.ti and vo.lei.bol. While one can surmise that such stress shifts are caused by the enforcement of quantity sensitivity, this line of thought has the unfortunate consequence of leaving those loanwords that have not yet reached the same level of nativization out of the grammar, as if they were unknown to the monolingual Spanish speakers who use them. Quantity insensitivity, by contrast, does not require the extradition of any words from the Spanish grammar to an unknown jurisdiction. In the analysis I propose, the cause of such stress shifts is a change in the sound structure of prosodically irregular morphemes. To be more precise, the mutation of a stress repeller into an abider will induce a stress shift from the antepenult to the penult (pé.nal.ti > pe.nal.ti). Similarly, the mutation of a stress repeller into an attractor will cause a stress shift from the antepenult to the ultima (e.g. vó.lei.bol > vo.lei.bol). Even proparoxytonism – the rare but well attested phenomenon whereby a paroxytone turns into proparoxytone – can be explained in this fashion; it is the mutation of a stress abider into a retractor (e.g. vér.ti.go < ver.ti.go ‘dizziness’). My point is that all such changes happen because of morphological restructuring, not because of quantity sensitivity.

306

Carlos-Eduardo Piñeros

(40) Acceptability of nonce words (Bárkányi 2002a) [CV.CVC.CV] [CV.CV̯ V.CV] [CV.CVV̯ .CV] [CV.CV.CVC] [CV.CV.CV̯ V] [CV.CV.CVV̯ ] [CV.CV.CVV̯ ]

Possible 31% 16% 19% 39% 18% 16% 19%

Strange 42% 45% 28% 42% 41% 36% 32%

Impossible 27% 39% 53% 19% 42% 47% 49%

I cannot help but notice the irony of the fact that the bulk of evidence for quantity insensitivity has come from loans and nonce words when, in actuality, the patrimonial lexicon offers ample proof of it. Minimal prosodic words, most of which are disyllabic but may also be monosyllabic, are one of the contexts where it is crystal clear that a correlation between stress and syllable structure does not exist. One finds, for instance, accented monosyllables such as yo ‘I’, tú ‘you’, ve ‘he sees’, va ‘he goes’, sí ‘yes’, no ‘no’, té ‘tea’, fe ‘faith’, vi ‘I saw’, di ‘I gave’, da ‘he gives’, and so on, coexisting with others of greater complexity, such as par ‘pair’, cal ‘lime’, pan ‘bread’, red ‘net’, luz ‘light’, pie ‘foot’, ley ‘law’, dio ‘he gave’, doy ‘I give’, etc. Since syllables of the CV type are treated identically to those of the CVC, CV̯ V, and CVV̯ types, this is prime evidence that Spanish stress assignment is blind to syllable structure. I submit that, in a system like this, it is unfounded to label some syllables heavy and others light because their behavior reveals that they are prosodically on a par. Larger prosodic words ratify that differences in rhyme complexity do not veer stress in the expected direction. Items in which a syllable with a simple rhyme manages to secure stress despite the availability of syllables of greater complexity occur throughout the lexicon and involve all positions of the accentual window (e.g. mó.vil ‘mobile’, bam.bú ‘bamboo’, hun.dí ‘I sank’, por.tá.til ‘portable’, al.cá.zar ‘fortress’, Mar.tí.nez ‘last name’, com.po.nes ‘you compose’, far.ma.cia ‘pharmacy’, es.tu.dio ‘I study’, san.ti.gua ‘he makes the cross sign’, chim.pan.cé ‘chimpanzee’, pas.par.tú ‘passe-partout’, con.ver.só ‘he conversed’ ré.ming.ton ‘rifle’, Sá.muel.son ‘Samuelson’). The simple fact that this is possible proves that syllable structure and stress are not systematically correlated in Spanish. The indisputable truth is that syllables of any type can capture stress and, when various syllable types are available, no particular one is favored. The only case in which rhyme complexity appears to create a bias has already been shown to be a morphological phenomenon (34–35).

The phonological weight of Spanish syllables

307

When the search for evidence is extended beyond the stress system, additional signs of quantity insensitivity can be found. A promising place to look for them is the sound inventory. As is well known, the possibility of packing several moras in one syllable offers QS languages the option of using length distinctions. Long vowels, which bear two moras (Vµµ) can be distinguished from short ones (Vµ) and, similarly, long consonants, which are associated with one mora (Cµ), can be opposed to short ones (C). In such systems, syllables containing a long vowel are always heavy, while syllables containing a short vowel may be heavy or light depending on whether they also contain a coda consonant associated with a mora. While languages vary as to whether vowels only, consonants only, or both major classes exploit length distinctions (Zec 2011), a correlation between the presence of long segments and the existence of heavy syllables is inevitable because the most economical way to add weight to a syllable is by lengthening one of its segments.25 In QI languages, on the other hand, neither type of length distinction is viable because, if syllables are to have equal weight, no such unit can be granted multiple moras and, consequently, no segment can be long. The expectation thus arises that the sound inventories of QI systems will consist exclusively of short segments, precisely what we find in Spanish. Phonological processes are another source of independent evidence. Changes to the number of segments in the rhyme of the ultima are relevant because they are expected to alter the weight of that syllable and thereby impinge on the structure of the foot. Word-final consonant deletion, a variable process that applies most frequently to obstruents, fits the bill. Deletion of word-final /d/ is especially common (41). (41)

e.dad

[e.ˈðað]

~

[e.ˈða]

‘age’

sa.lud

[sa.ˈluð]

~

[sa.ˈlu]

‘health’

pa.red

[pa.ˈɾeð]

~

[pa.ˈɾe]

‘wall’

QS analyses claim that the reason why words such as those in (41) are oxytone is that their final consonant provides the ultima with a second mora, thereby affording the construction of a binary foot on a single syllable (e.g. [eµ.(ˈðaµðµ)]). Given that the well-formedness of the foot hinges on the presence of two segments in the rhyme of the ultima, this view predicts that word-final consonant

25 It is important to understand that the expected correlation is not specifically between long vowels and heavy syllables, but rather between long segments (whether they are vowels or consonants) and heavy syllables.

308

Carlos-Eduardo Piñeros

deletion will trigger some form of adjustment. One way to maintain prosodic stability would be to rescue the floating mora by linking it to the adjacent vowel, but that is clearly not happening here for there is no evidence of vowel lengthening (e.g. *[eµ.(ˈðaµµ)]). The loss of the consonant along with its mora would then require that the foot be restructured, but that is not happening here either for there is no change in stress locus (e.g. *[(ˈeµ.ðaµ)]). As seen in (41), the actual reaction is inertness. That nothing is done to repair the destabilized prosodic structure is quite informative; it confirms that the prosodic structure of oxytones is not dependent on the number of segments parsed by the ultima.

7 Conclusion While generativists acknowledge that morphology introduces an element of unpredictability in Spanish stress, they are also aware that achieving the highest possible predictability rate is a desirable goal and quantity sensitivity offers them a seemingly reasonable way to do so on the basis of phonological properties alone. Unfortunately, such an enterprise cannot succeed for the simple reason that syllable structure and stress do not correlate in this language. In this article, I have demonstrated that, from a theoretical point of view, quantity sensitivity is untenable. It leads to circularity because the savings in stress marks that it allegedly yields are counterbalanced by an exorbitant number of extrametricality marks and, even if one accepts the stipulation that finality within certain domain/s renders prosodic units invisible, proper accentuation of many words is still not possible. Another factor that makes quantity sensitivity unviable is that heavy syllables reduce mobility within the accentual window and, although this seems to have the positive effect of making the distribution of stress stricter, it is actually a disadvantage because it adds to the pile of data that cannot be derived. In a nutshell, there is no empirical grounding for the restrictiveness imposed by heavy syllables. There is incontrovertible evidence that the penult is the normal stress locus and that, rather than a peculiarity of nouns, this is a cross categorical generalization; we see it in the noun cua.der.no ‘notebook’, in the adjective com.pli.ca.do ‘complicated’, in the verb in.si.nú.a ‘he insinuates’, in the adverb bas.tan.te ‘enough’, and in most of their congeners. The first significant contribution of this study has been to provide a natural explanation for this fundamental fact. I have argued that paroxytones abound in the language because of their simplicity; they feature a syllabic trochee strategically located at the end of the word, a structure that comes at no cost to the Spanish grammar because it

The phonological weight of Spanish syllables

309

follows from universal principles (8–10), which learners can infer from the forms they are exposed to. This finding has made it possible to conclude that phonology alone accounts for the accentuation of a large portion of the lexicon, an estimated 64%. I have also demonstrated that, in the remaining portion of the lexicon, stress assignment is conditioned by the morphology; more specifically, by the rise of prosodically-irregular morphemes – attractors or repellers – to the role of morphological head. When that happens, the normal stress locus cannot be maintained; stress is minimally displaced to the right or to the left as the size of the foot decreases or increases by one syllable. This is the origin of the second and third most common basic patterns: oxytones and proparoxytones, respectively. The proposed analysis recognizes that the desinence is part of the stress domain (i.e. the full word) and that stress assignment relies on that constituent, even when it is covert. An important observation in this regard is that the rise of a stress attractor to the role of morphological head may coincide with a covert or an overt desinence (e.g. [[JAPON+ÉS]+Ø] ‘Japanese, masculine, singular’ vs. [[JAPON+ES]+e+s] ‘Japanese, masculine, plural’, [[JAPON+ES]+a] ‘Japanese, feminine, singular’, and [[JAPON+ES]+a+s] ‘Japanese, feminine, plural’). When the desinence is covert, meeting the demands of the stress attractor requires departing from the syllabic trochee; that is to say that the unit that the morphology submits as stress bearer differs from that recommended by the phonology. While this conflict is invariably resolved in favor of the morphology (e.g. ja.po.nés has final stress despite there being another vowel which is better positioned to act as foot head), a harmonic solution is possible when the desinence is overt. In such a case, both morphology and phonology can be satisfied because they concur that stress should fall on the penult. The upshot of this agreement is that the irregularity of the stress attractor becomes hidden; it can be satisfied without distorting the foot (e.g. ja.po.ne.ses, ja.po.ne.sa, and ja.po.ne.sas appear normal despite the fact that a stress attractor is in command). The demise of quantity sensitivity is precipitated by the discovery that the clue that context provides to guide the accentuation of oxytones is actually morphological: a gap in the inflectional paradigm. Given that a vowel is the typical exponent of the desinence, the presence of a consonant at the end of the word points to an empty paradigmatic position, which constitutes a reliable clue that a stress attractor is in command because all such morphemes allow the desinence to be covert when they govern the word (e.g. [[PERDIZ]+Ø] ‘partridge’, [[cont+a+DOR]+Ø] ‘accountant’). Thanks to this discovery, it is legitimate to say that 91% of the lexicon has predictable stress because, to the 64% that follows regular footing, one can add the 27% that is affected by the contextuallyinferable irregularity.

310

Carlos-Eduardo Piñeros

Reaching full predictability is impossible, however, because the system also includes stress repellers, the presence of which does not correlate with any contextual property that could serve as the basis for further predictions. This means that storing stress-related information in the lexical entry of some words is ineluctable, but it should be underscored that a single record suffices. The proper place for that record is, of course, the entry of the prosodically-irregular morpheme (e.g. –IC ‘relation’), not every word where that morpheme is present. The key finding regarding the recalcitrant pattern has been that the rise of a stress repeller to the role of morphological head is what induces antepenultimate stress and that further displacement to the left is unwarranted because, there being a single morphological head per word, the demands of stress repellers co-occurring with the one sitting in the governing position are not binding. From this analytical perspective, the factor that ultimately decides the stress pattern of the word is the type of morphological head it has. Abiders in that role allow the phonology to yield a word-final syllabic trochee (paroxytones), whereas attractors and repellers exercise their power to shorten or lengthen that foot minimally (oxytones and proparoxytones, respectively). The next step in this research program is to develop a formal analysis that integrates the phonological principles in (8–10) with the veto powers of the morphological head, a formidable task that I shall tackle elsewhere. Considering how intricate the interactions between phonology and morphology are, the day when all questions about Spanish stress can be given a satisfactory answer may still be far away, but for the time being at least the following fundamental points have been solidly established: (42)

a.

Spanish stress is not entirely phonological or morphological; it is essentially phonological, although strongly conditioned by the morphology.

b.

Like in all other languages, the domain of primary stress is the word, not the stem.

c.

Spanish feet are trochaic, not iambic.

d.

Spanish is quantity insensitive, not quantity sensitive.

e.

From a prosodic point of view, Spanish morphemes fall into three categories: stress abiders, stress repellers, and stress attractors.

f.

A final syllabic trochee is behind the abundance of paroxytones, which follow form universal prosodic principles.

g.

Prosodically-irregular morphemes can only impinge on stress assignment when acting as the head of the morphological word; they are disempowered when removed from that role.

The phonological weight of Spanish syllables

311

h.

Oxytones are predictable from a gap in the inflectional paradigm, which betrays the presence of a stress attractor in the governing position.

i.

All proparoxytones are unpredictable because the presence of a stress repeller in the governing position does not correlate with any contextual property.

References Alvord, Scott M. 2003. The psychological unreality of quantity sensitivity in Spanish: Experimental evidence. The Southwest Journal of Linguistics 22(2). 1–12. Bárkányi, Zsuzsanna. 2002a. Primary stress in Spanish. In Teresa Satterfield, Christina Tortora, & Diana Cresti (eds.), Current issues in Romance languages: Selected papers from the 29th Linguistic Symposium on Romance Languages (LSRL), Ann Arbor, 8–11 April 1999, 17–31. Amsterdam & Philadephia: John Benjamins. Bárkányi, Zsuzsanna. 2002b. A fresh look at quantity sensitivity in Spanish. Linguistics 40. 375–394. Bullock, Barbara E. & Almeida Jacqueline Toribio. 2010. Correcting the record on Dominican [s]hypercorrection. In Sonia Colina, Antxon Olarrea, & Ana Maria Carvalho (eds.), Romance Linguistics 2009: Selected papers from the 39th Linguistic Symposium on Romance Languages (LSRL), 15–24. Amsterdam & Philadelphia: John Benjamins. Bullock, Barbara E., Almeida Jacqueline Toribio, & Mark Amengual. 2014. The status of s in Dominican Spanish. Lingua 143, 20–35. Colina, Sonia. 1996. Spanish noun truncation: The emergence of the unmarked. Linguistics 34(6). 1199–1218. den Os, Els & René Kager. 1986. Extrametricality and stress in Spanish and Italian. Lingua 69. 23–48. Dunlap, Elaine. 1991. Issues in the moraic structure of Spanish. Amherst, MA: University of Massachusetts dissertation. Elman, Jeffrey L. 1979. Spanish noun and adjective stress assignment: A nonphonological account. In James P. Lantolf, Francine Frank, & Jorge M. Guitart (eds.), Colloquium on Spanish and Luso-Brazilian Linguistics, 1–18. Washington, DC: Georgetown University Press. Face, Timothy L. 2004. Perceiving what isn’t there: Non-acoustic cues for perceiving Spanish stress. In Timothy L. Face (ed.), Laboratory approaches to Spanish phonology, 117–141. Berlin & New York: Mouton de Gruyter. Face, Timothy L. 2006. Cognitive factors in the perception of Spanish stress placement: implications for a model of speech perception. Linguistics 44. 1237–1267. Face, Timothy L. & Scott M. Alvord. 2005. Descriptive adequacy vs. psychological reality: The case of two restrictions on Spanish stress placement. Studies in the Linguistics Sciences 32(2). 1–16. Felíu, Elena. 2001. Output constraints on two Spanish word-creation processes. Linguistics 39. 871–891.

312

Carlos-Eduardo Piñeros

Fink, Robert. 1978. Spanish stress assignment: a mixed phonological and morphological approach. Glossa 12. 105–124. Foley, James, A. 1965. Spanish morphology. Cambridge, MA: MIT dissertation. Gibson, Mark. 2011. A typology of stress in Spanish non-verbs. Ianua. Revista Philologica Romanica 11. 1–30. Goedemans, Rob & Harry van der Hulst. 2011. Chapter 15: Weight-sensitive stress. The world atlas of language structures online. http://wals.info/chapter/15 (accessed 8 April 2013). Hammond, Mike. 1995. There is no lexicon! Rutgers Optimality Archive 43. http://roa.rutgers. edu/files/43–0195/43–0195–HAMMOND-0–0.PDF (accessed 10 April 2014). Harris, James W. 1969. Spanish phonology. Cambridge, MA: The MIT Press. Harris, James W. 1975. Stress assignment rules in Spanish. In William G. Milan, John J. Staczek, & Juan C. Zamora (eds.), 1974 Colloquium on Spanish and Portuguese Linguistics, 56–83. Washington, DC: Georgetown University Press. Harris, James W. 1980. Nonconcatenative morphology and Spanish plurals. Journal of Linguistic Research, 15–31. Harris, James W. 1983. Syllable structure and stress in Spanish: A nonlinear analysis. Cambridge, MA: The MIT Press. Harris, James W. 1985. Spanish word markers. In F. H. Nuessel Jr. (ed.), Current issues in Hispanic phonology and morphology. Blomington, IN: Indiana University Linguistics Club, 34–54. Harris, James W. 1987. The accentual patterns of verb paradigms in Spanish. Natural Language & Linguistic Theory 5, 61–90. Harris, James W. 1989. How different is verb stress in Spanish? Probus 1, 241–258. Harris , James W. 1991. The exponence of gender in Spanish. Linguistic Inquiry 22 (1), 27–62. Harris, James W. 1992a. Spanish stress: The extrametricality issue. Bloomington, IN: Indiana University Linguistics Club. Harris, James W. 1992b. The form classes of Spanish substantives. In G. Booij & J. van Marle (eds.), Yearbook of Morphology 1991. Netherlands: Kluwer Academic Publishers, 65–88. Harris, James W. 1995. Projection and edge marking in the computation of stress in Spanish. In John Goldsmith (ed.), The handbook of phonological theory, 867–887. Oxford: Blackwell. Harris, James W. 1996. The syntax and morphology of class marker suppression in Spanish. In K. Zagona (ed.), Grammatical theory and Romance languages: Selected papers from the 25th Linguistics Symposium on Romance Languages (LSRL XXV), Seattle, WA, 2–4 March, 1995. Amsterdam: John Benjamins, 99–122. Hayes, Bruce Philip. 1980. A metrical theory of stress rules. Bloomington, IN: Indiana University Linguistics Club. Herslund, Michael. 1982. The stress pattern of Spanish. Acta Linguistica Hafniensia 17. 105– 129. Hooper, Joan B. & Tracy Terrell. 1976. Stress assignment in Spanish: A natural generative analysis. Glossa 10. 64–110. Hualde, José Ignacio. 2012. Stress and rhythm. In José Ignacio Hualde, Antxon Olarrea & Erin O’Rourke (eds.), The handbook of Hispanic linguistics, 153–171. Chichester, West Sussex, England: Wiley-Blackwell. Kiparsky, Paul. 1991. Catalexis. Manuscript, Stanford University, Stanford California. Liberman, Mark and Prince, Alan. 1977. On stress and linguistic rhythm. Linguistic Inquiry 8. 249–336.

The phonological weight of Spanish syllables

313

Larramendi, Miguel de. 1729. El imposible vencido. Arte de la lengua bascongada. Salamanca: Antonio Joseph Villargordo Alcaraz. Lipski, John M. 1995. Spanish hypochoristics: Towards a unified prosodic analysis. Hispanic Linguistics 6(7). 387–434. Lipski, John M. 1997. Spanish word stress: the interaction of moras and minimality. In Fernando Martínez-Gil & Alfonso Morales-Font (eds.), Issues in the phonology and morphology of the major Iberian languages, 559–593. Washington, DC: Georgetown University Press. Llorach, A. 1950. Fonología española. Madrid: Editorial Gredos. Martínez-Paricio, Violeta. 2013. The intricate connection between diphthongs and stress in Spanish. Nordlyd 40. 166–195. Meinschaefer, Judith. 2015. Right-alignment and catalexis in Spanish word stress. Manuscript, Freie Universität Berlin, Berlin. Morales-Front, Alfonso. 1999. El acento. In Rafael A. Nuñez Cedeño & Alfonso Morales-Front (eds.), Fonología generativa contemporánea de la lengua española (1st ed.), 203–230. Washington, DC: Georgetown University Press. Morales-Front, Alfonso. 2014. El acento. In Rafael A. Nuñez Cedeño, Sonia Colina, & Travis G. Bradley (eds.), Fonología generativa contemporánea de la lengua española (2nd ed.), 235–265. Washington, DC: Georgetown University Press. Morgan, Terrell. 1998. The linguistic parameters of /s/-insertion in Dominican Spanish: A case study in quantitative hypercorrection. In Javier Gutiérrez-Rexach & Luis Silva-Villar, (eds.), Perspectives on Spanish linguistics 3, 79–96. Columbus, OH: The Ohio State University. Nouveau, Domique. 1994. Catalexis in word stress: evidence from Spanish and Dutch. In Erin Duncan, Donka Farkas, and Phillip Spaelti (eds.), Proceedings of the Twelfth West Coast Conference on Formal Linguistics, 51–66. Center for the Study of Language and Information, Stanford, California. Nuñez Cedeño, Rafael A. 1985. Metrical analysis of Spanish verbal stress. Revista Argentina de Linguistica 1. 107–132. Nuñez Cedeño, Rafael A. 1986. La /s/ ultracorrectiva en dominicano y la estructura silábica. In José D. Moreno de Alba (ed.), Actas del II congreso internacional sobre el español de América, 337–343. Mexico: Universidad Nacional Autónoma de México. Nuñez Cedeño, Rafael A. 1988. Structure-preserving properties of an epenthetic rule in Spanish. In David Birdsong & Jean Pierre Montreuil (eds.), Advances in Romance linguistics, 319– 335. Foris: Dordrecht. Ohannesian, María. 2004. La asignación del acento en castellano. Barcelona: Universitat Autònoma de Barcelona dissertation. Otero, Carlos. 1986. A unified metrical account of Spanish stress. In Michael Brame, Heles Contreras, & Frederick Newmeyer (eds.), A festschrift for Sol Saporta, 299–332. Seattle: Noit Amrofer. Oltra-Massuet, Isabel & Karlos Arregi. 2005. Stress-by-structure in Spanish. Linguistic Inquiry 36. 43–84. Piñeros, Carlos-Eduardo. 2000a. Prosodic and segmental unmarkedness in Spanish truncation. Linguistics 38. 63–98. Piñeros, Carlos-Eduardo. 2000b. Foot-sensitive word minimization in Spanish. Probus 12. 291– 324. Piñeros, Carlos-Eduardo. 2016. El acento. Enciclopedia de lingüística hispánica. London/New York: Routledge Publishing Company.

314

Carlos-Eduardo Piñeros

Prieto, Pilar. 1992. Truncation processes in Spanish. Studies in the Linguistic Sciences 22. 143– 158. Prince, Alan & Paul Smolensky. 1993/2002. Optimality Theory: Constraint interaction in generative grammar. Rutgers Optimality Archive 537. http://roa.rutgers.edu/files/537-0802/5370802-PRINCE-0-0.PDF (accessed 10 April 2014). Roca, Iggy. 1988. Theoretical implications of Spanish word stress. Linguistic Inquiry 19. 393– 423. Roca, Iggy. 1989. The organisation of grammatical gender. Transactions of the Philological Society 87 (1), 1–32. Roca, Iggy. 1990. Diachrony and synchrony in word stress. Journal of Linguistics 26: 133–164. Roca, Iggy. 1991. Stress and syllables in Spanish. In Héctor Campos & Fernando Martínez-Gil (eds.), Current studies in Spanish linguistics, 559–635. Washington, DC: Georgetown University Press. Roca, Iggy. 1997a. On the role of accent in stress systems: Spanish evidence. In Fernando Martínez-Gil & Alfonso Morales-Front (eds.), Issues in the phonology and morphology of the major Iberian languages, 619–664. Washington, DC.: Georgetown University Press. Roca, Iggy. 1997b. There are no “glides”, at least in Spanish: An optimality account. Probus 9. 233–265. Roca, Iggy. 2000. On the meaning of gender. Hispanic Research Journal 1 (2), 113–128. Roca, Iggy. 2005a. Saturation of parameter settings in Spanish stress. Phonology 22. 345–394. Roca, Iggy. 2005b. La gramática y la biología en el género del español (1.ª parte). Revista Española de Lingüística 35 (1), 17–44. Roca, Iggy. 2006a. The Spanish stress window. In Fernando Martínez-Gil & Sonia Colina (eds.), Optimality-theoretic studies in Spanish phonology, 239–277. Amsterdam & Philadelphia: John Benjamins. Roca, Iggy. 2006b. La gramática y la biología en el género del español (2.ª parte). Revista Española de Lingüística 35 (2), 397–432. Roca, Iggy & Elena Felíu. 2003. Morphology in truncation: The role of the Spanish desinence. In Geert Booij & Jaap van Marle (eds.), Yearbook of Morphology 2002, 187–243. Dordrecht: Kluwer. Rosenthall, Samuel. 1994. Vowel-glide alternations in a theory of constraint interaction. Amherst, MA: University of Massachusetts dissertation. Selkirk, Elizabeth O. 1980. On prosodic structure and its relation to syntactic structure. Bloomington, IN: Indiana University Linguistics Club. Shelton, Michael. 2007. An experimental approach to syllable weight and stress in Spanish. College Park, PA: Pennsylvania State University dissertation. Stockwell, Robert & Jean Donald Bowen. 1965. The sounds of English and Spanish. Chicago: University of Chicago Press. Tesar, Bruce & Paul Smolensky. 2000. Learnability in Optimality Theory. Cambridge, MA: The MIT Press. Waltermire, Mark. 2004. The effect of syllable weight on the determination of spoken stress in Spanish. In Timothy L. Face (ed.), Laboratory approaches to Spanish phonology, 171–191. Berlin: Mouton de Gruyter. Whitley, Stanley. 1976. Stress in Spanish: two approaches. Lingua 39. 301–332. Zec, Draga. 2011. Quantity-sensitivity. In Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, & Keren Rice (eds.), The Blackwell companion to phonology, 1335–1362. Oxford: Blackwell Publishing.

W. Leo Wetzels and Ben Hermans

10 Syllable weight, stress, and the neutralization of vowel length in Batticaloa Creole Portuguese Abstract: In this study we discuss the stress patterns of Batticaloa Creole Portuguese (BCP), which we analyze as an iambic system, using data from Smith (1977). BCP has a contrast between long and short vowels in stressed and unstressed syllables. Underlying long vowels that are not stressed surface as short vowels, while stressed long vowels vary in length, except word-initially, where lexical length is systematically maintained. Crucial in our analysis is the formal representation of the relation between prominence and metrical constituency for which we propose a generalized integrated model. From Zoll (1997) we borrow the idea of licensing to account for Default-to-Opposite systems, of which BCP is an example. We argue that the generalized integrated model provides an explanation of why in BCP long vowels in a monosyllabic iamb behave differently from the ones that occur in a disyllabic iamb.

1 Introduction Batticaloa Creole Portuguese (henceforth BCP) is the dialect of Indo-Portuguese spoken by descendants of Portuguese and Dutch who live in the provincial town of Batticaloa on the east coast of Sri Lanka. The phonology of this language is described in detail in Smith (1977). An interesting aspect of BCP is the partial neutralization of the phonological contrast between short and long vowels in both stressed and unstressed syllables. At the lexical level, long vowels can appear in any position in the word. At the output of the phonology, however, all unstressed long vowels are short. Furthermore, a stressed long vowel is realized with variable length in word internal position and tends to be fully short in closed or word-final syllables, especially in rapid speech (Smith 1977: 66). Only in a word-initial stressed syllable is the length of an underlying long vowel always fully realized.

W. Leo Wetzels, Vrije Universiteit Amsterdam, Universidade de São Paulo (USP) Ben Hermans, Meertens Instituut, Vrije Universiteit Amsterdam

316

W. Leo Wetzels and Ben Hermans

The neutralization facts of BCP have been discussed by Koontz-Garboden (2000) from the perspective of Sympathy Theory (McCarthy 1999, 1999b). The author’s analysis is based on the assumption that both stressed and unstressed long vowels are shortened already at the phonological level, unless they are located in the stressed initial syllable. Since some shortened vowels behave as long for the quantity-sensitive stress rule, vowel shortening and stress assignment interact opaquely. Koontz-Garboden argues that the opaque interaction between stress and vowel reduction in BCP provides an argument in favor of the explanatory potential of Sympathy Theory. In this paper we present an alternative analysis of stress and vocalic length in BCP. Our analysis closely follows Smith’s original proposal according to which the length contrast is preserved under stress in the phonology. Since there is no phonological rule of stressed vowel reduction, there can be no opacity involved in the distribution of word stress at this level. We also agree with Smith when he judges that the variable vowel length in stressed syllables is best accounted for in the phonetic component1. One question that must be answered by any account of the BCP neutralization facts is why the initial syllable of BCP is special, in the sense that a stressed long vowel maintains its full length only and always in this position. KoontzGarboden identifies this fact as an instance of Positional Faithfulness (Beckman 1997, 1998). We propose that the full realization of length in the initial (stressed) syllable follows from the fact that BCP has an iamb which must minimally be bisyllabic. If a word-initial long vowel is parsed as an iamb, it must be parsed with two syllables in order to realize the minimal size requirement. This entails that there is an important structural difference between the second half of a word-initial stressed long vowel and the second half of a word-internal stressed long vowel. We will argue that it is this structural difference which explains the different phonetic realization of word-initial stressed long vowels as compared to wordinternal ones. Another interesting aspect of BCP is that it represents a clear case of a Default-to-Opposite system (Zoll 1997). These systems come in two types. In some languages, the first heavy syllable of the word receives stress, and, if there is no heavy syllable, stress goes on the last syllable; in other languages, the last heavy syllable is stressed, but if the word contains no heavy syllable, the first syllable receives the stress. BCP is an instance of the latter type. Zoll (1997)

1 We moreover take it as a general truth that rules like stressed vowel shortening in BCP, which are gradient and optional, do not create opacity.

Syllable weight, stress, and the neutralization of vowel length

317

accounts for these systems in terms of licensing. In the course of our analysis of BCP it will become clear that we fully agree with her way of analyzing these systems. Before we present the data and the details of our analysis, we must make explicit some aspects of the stress theory as we see it. For the relation between prominence and metrical constituency, we follow the predominant view that prominence and metrical structure are integrated, or conflated, along the lines of Hammond (1984), Halle and Vergnaud (1987), Hayes (1994), and others (but see Hyde 2001, 2002 for a different perspective). However, we propose to take this model to its natural limit by generalizing it to the lowest levels of prosodic constituency: the mora and the syllable levels. As we will show in our discussion of the BCP phonology, the generalized integrated model allows us to explain why in BCP long vowels in a monosyllabic iamb behave differently from the ones that occur in a disyllabic iamb. In section 2 we make explicit our assumptions regarding the representation of stress. In section 3, we present the data illustrating the BCP stress system. In section 4 we propose our analysis of this system.

2 The theoretical framework Most phonologists assume that the representation of stress consists of two modes: prominence and metrical constituency. Differences of opinion concern the question how the relation between these modes is to be modeled. On the one hand, there are those who propose that prominence is located in a dimension that runs parallel to the dimension where constituency is represented, as in Hyde (2001; 2002). In such theories one finds representations like that below. As usual, headedness is represented by straight lines, dependency by slanted lines. (1)

318

W. Leo Wetzels and Ben Hermans

The figure in (1) shows the prosodic organization and the relative prominence of the prosodic constituents of a word consisting of six light (monomoraic) syllables. The syllables are parsed as left-headed binary feet, i.e., trochees. The trochaic feet are dominated by the word constituent, of which the final foot carries the main stress. The prominence mode, which runs parallel to the prosodic mode, is created by a set of projection rules. In figure (1) every mora projects a grid mark, symbolized by an asterisk representing a prominence unit. Similarly, the head syllables and the head foot each project a grid mark. The relative stress assigned to the different prosodic units occurs in the phonetic interpretation component and is derived from the relative height of the grid mark columns present in the output of the phonological grammar. In a different modeling of the relation between relative prominence and prosodic constituency, these modes are conflated, as was first proposed by Hammond (1984). (2)

One advantage of the integrated way of representing prosodic structure and prominence is its lesser complexity. The grid marks at level 2 each represent a foot themselves while the one at line 3 represents the word tree. As compared to the representation in (1), the only remnants of the prosodic mode in figure (2) are moras and syllables. Due to the influential studies by Halle and Vergnaud (1987) and Hayes (1994), which adopt Hammond’s original proposal, the integrated model of stress representation is the one that is currently most popular. Following Hermans and Wetzels (2012), we propose to extend this approach to the syllable level by eliminating the σ-node and confer the units of the basic line the status of a syllable, as illustrated in (3): (3)

Syllable weight, stress, and the neutralization of vowel length

319

In figure (3), line 1-constituents are syllables, line 2-constituents are feet, and the line 3-constituent is the main stress-constituent. Furthermore, following Prince (1985), we assume that unbounded constituents are disallowed: all constituents are maximally binary. Consequently, the main stress constituent in (3) is illicit for being ternary. The idea to generalize Hammond’s original proposal to the syllable level allows us to characterize the difference between various foot types in a simple way. Consider, for instance, the difference between the ‘even’ moraic trochee and the ‘uneven’ moraic trochee2. Consider a sequence of two syllables, the first of which contains a long vowel and the second a short vowel. (4)

Cross-linguistically, there are two ways to parse the string in (4) into trochees (Mellander 2004). In some languages the syllable with the long vowel absorbs the entire trochee, without the neighboring short syllable. In other languages, the trochee comprises both the long and the short syllable. This difference can be formalized straightforwardly on the assumption that in the first type every mora projects a grid mark on line 1, whereas in the second type a dependent mora is allowed. The difference between these trochaic feet is made explicit in (5): (5)

a.

b.

The structure in (5a) represents the ‘even’ moraic trochee. Both moras of the long vowel project a line 1-constituent. Since a foot is maximally binary, these two constituents fully absorb the trochee, so that the next line 1-constituent is excluded from the foot. In the case of an ‘uneven’ trochee, (5b), the two moras of the long vowel project a single branching line 1-consituent. Consequently, there is still room for parsing the next line 1-consituent as part of the same foot. In sum, an even moraic trochee can be derived by a prohibition on dependent moras on line 1. In the representational model proposed here, every mora 2 Evidence for the existence of the uneven moraic trochee can be found in Mellander 2004.

320

W. Leo Wetzels and Ben Hermans

creates its own ‘phonological syllable’, except that, in the case of an uneven trochee, a line 1 syllable is allowed to dominate a dependent mora, so that a short vowel can be parsed in the same trochaic foot as the preceding long vowel.3 To create the word-prosodic structures of the type proposed, we follow the practice of metrical phonology, defining the relevant constraints as Parse-X constraints. (6)

a.

PARSE -μ A mora is parsed by a line 1-constituent

b.

PARSE -Line 1 A line 1-constituent is parsed by a line 2-constituent

c.

PARSE -Line 2 A line 2-constituent is parsed by a line 3-constituent

In a language where feet are trochaic (left headed) and where the Main Stress Constituent is Right headed, a sequence of 6 light syllables is parsed in the following way: (7)

In (7) every mora is parsed as a line 1 constituent. These, in turn, are parsed as left-headed line 2 constituents, of which the final and prefinal feet are parsed as 3 One might wonder what happens if, in languages with even moraic trochees, foot construction takes place from right to left. In such a setting of the directionality parameter, violations against syllable integrity can be created, such that the two moras of the long vowel in (5a) are not parsed in the same foot. However, it has been shown by Kager and Martínez-Paricio (undated) that this type of syllable integrity violation indeed exists. There are indeed languages in which even moraic trochees are built from right to left in such a way that the two moras of a long vowel are parsed as parts of different feet. The question then arises why this does not always happen in languages where even moraic trochees are built from right to left. A possible explanation can be given in terms of structural relations. In line with earlier proposals (cf. Rubach 1993) it can be argued that, within the domain of a long vowel, the second mora is the dependent one (as we have indicated with the slanted line between V and μ in (5)). Preservation of syllable integrity would then follow from a constraint disallowing a mora that is a dependent at the segmental level to function as a head at the foot level.

Syllable weight, stress, and the neutralization of vowel length

321

a right-headed line 3 constituent. Phonetically, the structure in (7) is interpreted as a sequence of syllables with alternating trochaic rhythm and with main stress on the prefinal syllable. In the following section we present the representative data illustrating the stress system of BCP.

3 The data In BCP, a long vowel can appear in any syllable of the word at the underlying level. The rightmost syllable containing a long vowel is stressed. Underlying long vowels that are not stressed are shortened. The following examples4 were taken from Smith (1977). Vowel length is expressed with a colon, which occurs between parentheses when vowel reduction is gradient and optional at the phonetic level. (8)

underlying level

surface level

gloss

weight structure

/mi:də/ /mi:di:/ /mi:di:do:r/

[ˈmi:də] [miˈdi(:)] [midiˈdo(:)r]

measure to measure surveyor

ˈHL HˈH HHˈH

The examples in (8) show that stressed long vowels realize their full length systematically only at the beginning of the word. Otherwise their length is variable, depending on the structure of the syllable (open vs. closed), their position within the word (word-internal vs. word-final), and speech rate. In (9) we provide some additional examples showing that stress goes to the rightmost syllable containing a long vowel. The second example in (9c) illustrates that closed syllables do not attract the main stress. (9)

a.

/iskurda:m/ /ba:rwija:/

[iskurˈda(:)m] [bərwiˈja(:)]5

darkness to shave

LLˈH HLˈH

b.

/muski:tu/ /ba:rwe:ru/

[musˈki(:)tu] [bərˈwe(:)ru]

fly barber

LˈHL HˈHL

c.

/ka:ndija/ /ka:zantu/

[ˈka:ndijə] [ˈka:zəntu]

candy at home

ˈHLL ˈHLL

4 And all other examples we will use in this study. 5 In BCP stressed as well as unstressed short /a/ is phonetically realized as [ə], whether underlying or derived by unstressed vowel shortening.

322

W. Leo Wetzels and Ben Hermans

In words without long vowels, the initial syllable is stressed, as is illustrated with the examples in (10). (10)

/tosa/ /wahupu/ /kanaj/ /gɔrgal/ /sindafara/

[ˈtosə] [ˈwəhupu] [ˈkənəj] [ˈgɔrgəl] [ˈsindəfərə]

to cough subject fellow throat Monday

ˈLL ˈLLL ˈLLL ˈLL ˈLLLL

The first two words in (10) contain only CV syllables, which are unquestionably light. Again, the third and fourth examples show that CVC syllables behave as CV syllables by not attracting main stress. Being heavy, syllables with long vowels are bimoraic, whereas closed syllables are light or monomoraic. This is illustrated in (11). (11)

a.

long vowels are heavy

b.

closed syllables are light

The stress system of BCP is clearly a default-to-opposite system. We have seen earlier that default-to-opposite systems come in two types. In one subsystem, the first heavy syllable receives the word stress; if the word has no heavy syllable, the last (light) syllable is stressed. In the second subsystem, the last heavy syllable of the word receives word stress; if there is no heavy syllable in the word, the first (light) syllable is stressed. Clearly, BCP is an example of the second type. The most influential approach towards default-to-opposite systems is developed in Zoll (1997). The fundamental hypothesis of her approach is that a stressed light syllable must be ‘licensed’ by the peripheral mora. This idea will play an important role in the analysis we elaborate for BCP, especially in order to explain why the initial long stressed vowel is realized with full length in BCP, whereas a long stressed vowel in any other position is realized with variable length.

4 The analysis We analyze BCP main stress as representing an iambic system. This means that the foot structure of a word like /muski:tu/ [musˈki(:)tu] ‘fly’ is as follows:

Syllable weight, stress, and the neutralization of vowel length

323

(12)

In the representation (12) the head of the foot is a heavy (branching) syllable (line 1-constituent). The foot is right-headed, so it has a dependent line 1-constituent to the left of the head. If a word has more than one heavy syllable the last syllable containing a long vowel receives the word stress, as we have seen in (9). This is explained with an alignment constraint, requiring stress to be located near the edge of the word. For BCP, we need a constraint requiring the (iambic) foot to be located near the right edge of the word. We formulate this constraint in (13). (13) ALIGN -R The right edge of a line 2-constituent is aligned with the right edge of the word. The second important constraint requires that a stressed light syllable be licensed by the left edge of the word. In our system the general structure of a stressed light syllable is as in (14). (14)

Phonologically, a stressed light syllable has the status of a foot with a nonbranching head (at line 1). It is this structure, then, which must be licensed by the left word-edge in BCP. We formulate the licensing constraint in the following way: (15)

LIC -μi If a non-branching line 1-constituent is the head of a line 2-constituent, it must be licensed by the initial mora.

324

W. Leo Wetzels and Ben Hermans

Let us now look at a representative case, like /ba:rwe:ru/ [bərwe(:)ru] ‘barber’ (9b). This form has two underlying long vowels. Stress appears on the syllable containing the last long vowel of the word, which is the penult syllable. We can explain its stress pattern by the ranking LIC -μi >> ALIGN -R. We show this with the tableau in (16). (16) LIC -μi » ALIGN -R /ba:rwe:ru/ a.

LIC -μi

(ba:r) we: ru

**!

b. ☞ ba:r (we:) ru c.

ba: we: (ru)

ALIGN -R

* *!

The parentheses in the tableau represent syllables (line 1-constituents) that are foot heads. In (16a), the first heavy syllable is the foot head, which is located two syllables from the right word edge. For this reason it violates ALIGN -R twice. In the candidate (16b), the prefinal (heavy) syllable is the foot-head, a pattern which violates ALIGN -R only once. If the foot head were on the last syllable, as in the candidate (16c), ALIGN -R would be satisfied, but this would create a violation of LIC -μi, since the last syllable is light. Since LIC -μi dominates ALIGN -R, the last candidate cannot be optimal. The optimal candidate is the second candidate, because it does not violate LIC -μi and because it minimally violates ALIGN -R. In the tableau in (16), we have abstracted away from an important aspect of the BCP stress system. In Batticaloa, unstressed long vowels are always completely reduced. We propose that this is the consequence of three interacting constraints. One constraint is W-to-S (Weight-to-Stress), the constraint that requires that a heavy syllable be the head of a foot. (17)

W-to-S A branching line 1-constituent must be the head of a line 2- constituent

The second constraint is ALIGN -R, the constraint requiring that a foot, in fact any foot, be aligned with the right edge of a word. The third important constraint is a faithfulness constraint. It is formulated in (18). (18) IDENT /\-IO If a vowel is long at the underlying level, it is long at the output level.

Syllable weight, stress, and the neutralization of vowel length

325

We show next how these three constraints interact in such a way that a long unstressed vowel surfaces as a short vowel. Consider the following tableau. (19) W-to-S, ALIGN -R » IDENT /\-IO /ba:rwe:ru/ a.

W-to-S

(ba:r) (we:) ru

ba:r (we:)ru

IDENT /\-IO

**!*

b. ☞ (bərwe:) ru c.

ALIGN -R

* *!

*

*

In the candidate (19a), both heavy syllables are the head of a line 2-constituent. Although this satisfies W-to-S, it grossly violates ALIGN -R; the last foot violates this constraint once and the first foot adds two additional violations. In candidate (19c), no foot is constructed over the first heavy syllable. Although this reduces the number of ALIGN -R violations, this candidate violates W-to-S. In the candidate (19b), the first long (unstressed) vowel is shortened. Now W-to-S is no longer violated. However, it violates the faithfulness constraint IDENT /\-IO. Since this constraint is dominated by W-to-S the second candidate is optimal. The tableau in (19) shows that, if both W-to-S and ALIGN -R are ranked above IDENT /\-IO, only the last long vowel projects a foot, while all other long vowels to the left of the last one are categorically shortened.6 Let us next consider the words that do not contain long vowels. Why do they receive initial stress? A representative case is /sindafara/ [ˈsindəfərə] ‘Monday’. It has the following structure at the mora and syllable levels. (20)

We have already established in the tableau in (16) that LIC -μi dominates ALIGN -R in BCP. This ranking accounts for the fact that a light syllable cannot be stressed, unless it is located in word-initial position. This readily accounts for

6 BCP may also show secondary stress. Its distribution, however, is not determined by vowel duration. We will not discuss secondary stress in this paper.

326

W. Leo Wetzels and Ben Hermans

the fact that in a form without long vowels the initial syllable is stressed, even though it maximally violates ALIGN -R. We show this in the following tableau. (21)

LIC -μi » ALIGN -R /sindafara/ a.

sindəfə (rə)

LIC -μi

ALIGN -R

*!

b. ☞ (sin)dəfərə

***

c.

sin(də)fərə

*!

**

d.

sində (fər)ə

*!

*

In this tableau, the parentheses indicate the line 1 head of a line 2-constituent. In all candidates in (21), this line 1-constituent is a non-branching head of a line 2 constituent. With the exception of the candidate (21b), all other candidates in (21) violate LIC -μi, because they all contain a non-branching foot-head, in a noninitial position of the word. Since LIC -μi dominates ALIGN -R, violation of LIC -μi is fatal and the number of violations of ALIGN -R does not matter, at least not when these two constraints are in conflict. We now turn to the question of why word-initial stressed long vowels cannot be shortened, whereas stressed long vowels in non-initial position are shortened in a gradient manner in the phonetic component. In a model in which syllables are represented as line 1 grid marks, this differential treatment of underlying long vowels can be explained straightforwardly. It follows from two assumptions. It must be the case, as we have assumed here, that feet (line 2-constituents) are iambic (right headed) in BCP. We must furthermore postulate that the Main Stress Constituent requires a branching head. This constraint is formulated in (22).7 (22)

MINIMAL SIZE The head of a MSC (line 3-constituent) must be a branching line 2-constituent.

This constraint requires that the head foot must minimally have two positions at line 1 (it must be bisyllabic). Consider now a word like /ba:rwija:/ [bərwiˈja(:)] ‘to shave’, which is stressed on the final syllable. Its structure is as follows: 7 See Hermans and Wetzels 2014 for another example of a language with a Branchingness requirement imposed on the Main Stress Constituent.

Syllable weight, stress, and the neutralization of vowel length

327

(23)

The representation (23) contains a branching iambic foot, thereby satisfying MIN SIZE . Next, compare the structure in (23) with that of a word like /ka:zantu/ [ˈka:zəntu] ‘at home’, to which we provisionally assign the structure in (24): (24)

In principle, there is nothing wrong with the representation above, because monosyllabic iambs are allowed, provided they contain two moras. However, the proposed representation violates MIN SIZE because the head of the line 3constituent is a non-branching line 2 constituent. Satisfaction of this constraint can be achieved by changing the dependent second mora of the long vowel into a head mora. The resulting structure is given in (25), in which the constraint MIN SIZE is satisfied: (25)

328

W. Leo Wetzels and Ben Hermans

Now, consider the difference between the representations in (23) and (25), especially regarding the structure of the stressed long vowels. In (23), the second mora of the stressed (word-final) long vowel occupies a dependent position in a line 1-constituent. On the other hand, in (25), the second mora of the stressed (word-initial) long vowel occupies a head position in a line 1-constituent. It appears that this structural difference has phonetic consequences in BCP, as it explains why word-initial stressed long vowels, being the only ones that realize two prominence marks on line 1, are immune to gradient shortening, whereas non-initial stressed long vowels are partially or totally reduced. It all follows from the fact that foot structure is basically iambic, combined with the requirement that the foot which is the head of the main stress constituent must branch. Apart from MIN SIZE , which exists in other languages as well, nothing more needs to be said to explain the difference between initial stressed long vowels and word internal stressed long vowels. One last question remains to be answered, which is why, in representations of the type given in (25), LIC -μi is not violated? Notice that the line 1-head of the foot (the line 2-constituent) does not branch and that it is not the initial syllable. And yet, as we claim, it is licensed by the initial mora. Phonologists have used the concept of licensing in two ways: firstly, licensing can be done by domination relations. For instance, if it is required that a non-branching line 1-constituent, which is the head of a line 2-constituent, be licensed by the initial mora, then licensing is satisfied, if the line 1 constituent dominates the initial mora. There is also another way to satisfy licensing; it is often called ‘indirect licensing’ (cf. Kager 1999, for an overview). This form of licensing is met if the structure to be licensed is (autosegmentally) linked to the licensor. Clearly, in representations of the type given in (25), where the non-branching line 1 constituent does not occupy the initial position, indirect licensing applies. Although this non-branching constituent does not occupy the word-initial position to the extent that it does not dominate the initial mora, it is still licensed by the initial mora, because it is linked to it. After all, it is part of a long vowel. In our view, this shows that Zoll’s approach to default-to-opposite systems in terms of licensing is correct, as it turns out that, next to direct licensing, indirect licensing is relevant in a subset of these systems, of which BCP is an example. The analysis of BCP proposed in this section was made possible by our proposal to integrate the syllable structure with the prominence mode. To see this, consider the representation of a syllable whose nucleus is a long vowel in the mora theory as proposed by Hyman (1985):

Syllable weight, stress, and the neutralization of vowel length

329

(26)

The first mora in figure (26) is the syllable head, while the second mora is its dependent, because generally a syllable is left-headed. In models where the syllable structure runs parallel to the prominence plane, a syllable with a longvowel nucleus will get assigned one of the two representations illustrated in (27): (27)

a.

Long vowel projects one grid mark

b.

Long vowel projects two grid marks

It is easy to see that there is no direct connection between the prominence mode (grid marks) and the prosodic structure mode in the figures (27a, b). More particularly, there is no difference in prosodic structure between a prominenceprojecting mora and a mora that does not project prominence. Let us next compare the representations in (27) with the ones in (28), used in this study: (28)

a.

Long vowel projects one grid mark

b.

Long vowel projects two grid marks

The usual representation of a long vowel is the one in (28a), while the representation in (28b) emerges under specific circumstances, as in BCP, in which language MIN SIZE creates this structure. The representations in (28a) and (28b) are very different, particularly as concerns the relation between the mora and syllable levels. In (28a), the first mora is defined as the head of a syllable on line 1, with the second mora being in a

330

W. Leo Wetzels and Ben Hermans

dependent position. On the other hand, in (28b), each of the two vocalic moras takes up a head position on line 1. This means that the second mora in (28b) is prosodically stronger than the second mora in (28a). As argued in this section, it is this structural difference that is responsible for the different phonetic manifestation of non-initial stressed long vowels as compared to the word-initial stressed long vowels in BCP. It is not obvious how this difference could be naturally expressed in the parallel representation model. Therefore, our decision to extend the integrated model to include the syllable structure was not only a logical extension. The fully integrated model also offers the possibility of projecting two different types of heavy syllables containing long vowels, for which empirical evidence was found in BCP.

Conclusion We have analyzed the shortening of underlying long vowels in BCP as occurring on two different levels, separating categorical shortening, which is phonological, from gradient shortening, which is phonetic, following the original proposal by Smith (1977). If this way of handling the facts is correct, the interaction between shortening and stress placement is not opaque and cannot be considered an argument in favor of Sympathy Theory, as proposed by Koontz-Garboden (2000). We have also been able to explain the different behavior of word-initial stressed long vowels as compared to all other stressed long vowels without specifically referring to initial prominence. In our analysis, this difference follows from the hypothesis that BCP has iambic foot structure, together with the constraint MIN SIZE , which requires that the (main) foot be bisyllabic. Crucially, this explanation is dependent upon our contention that syllables are integrated in the prosody/prominence plane, thus generalizing earlier proposals in which only higher level constituents are integrated into this plane. We also have shown that Zoll’s licensing approach to default-to-oppose systems is correct, because BCP strongly suggests that ‘indirect licensing’ can play a role in these systems.

References Beckman, Jill. 1997. Positional faithfulness, positional neutralization and Shona vowel harmony. Phonology 14. 1–46. Beckman, Jill. 1998. Positional faithfulness. Amherst, MA: University of Massachusetts dissertation. Halle, Morris & Jean-Roger Vergnaud. 1987. An essay on stress. Cambridge, MA: The MIT Press.

Syllable weight, stress, and the neutralization of vowel length

331

Hammond, Michael. 1984. Constraining metrical theory: A modular theory of rhythm and destressing. Los Angeles, CA: UCLA dissertation. Hayes, Bruce. 1994. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press. Hermans, Ben & W. Leo Wetzels. 2012. Productive and unproductive stress patterns in Brazilian Portuguese. Revista Letras & Letras 28. 77–115. Hermans, Ben & W. Leo Wetzels. 2014. Branchingness and stress in Korlai Portuguese. In René Kager, Janet Grijzenhout & Koen Sebregts (eds.), Where the principles fail. A festschrift for Wim Zonneveld on the occasion of his 64th Birthday, 77–88. Published by UIL OTS, Utrecht. Hyde, Brett. 2001. Metrical and prosodic structure in Optimality Theory. New Brunswick, NJ: Rutgers University dissertation. Available in Rutgers Optimality Archive as ROA-476. Hyde, Brett. 2002. A restrictive theory of metrical stress. Phonology 19. 313–359. Hyman, Larry. 1985. A theory of phonological weight. Dordrecht: Foris. Kager, René. 1999. Optimality Theory. Cambridge: Cambridge University Press. Kager, René & Violete Martínez-Paricio. Undated manuscript. Antepenultimate mora effects – typology and representation. Koontz-Garboden, Andrew. 2000. Opacity in Batticaloa Creole Portuguese stress assignment: motivation for candidate-to-candidate faithfulness. MS, available in Rutgers Optimality as ROA-395. McCarthy, John. 1999a. Sympathy and phonological opacity. Phonology 16. 331–399. McCarthy, John. 1999b. Sympathy, cumulativity, and the Duke-of-York gambit. In Karen Baertsch & Daniel Dinnsen (eds.), Optimal green ideas in phonology, volume 1: 57–91. Bloomington, IN: Indiana University working papers in Linguistics. Mellander, Evan. 2004. (HL)-creating processes in a theory of foot structure. Linguistic Review 20. 243–280. Prince, Alan. 1985. Improving tree theory. Berkeley Linguistics Society 11. 471–490. Rubach, Jerzy. 1993. The lexical phonology of Slovak. Oxford: Clarendon Press. Smith, Ian. 1977. Sri Lanka Creole Portuguese phonology. Ithaca, NY: Cornell University dissertation. Zec, Draga. 2007. The Syllable. In Paul de Lacy (ed.), The Cambridge Handbook of Phonology, 161–194. New York: Cambridge University Press. Zoll, Cheryl. 1997. Conflicting directionality. Phonology 14. 263–286.

Subject index African languages 248, 251 Africans 251–252 Afro-Cuban 252 Afro-Iberian creole 251 ALCEBLA 197 Align Constraint 193–194 ambisyllabicity 213 American Spanish 140, 147, 175–176 anaptyxis 257fn Arabic 91, 191, 250 Argentine Spanish 3, 9, 16–17, 20–23, Argentina 5, 12–13, 15–20, 252, 266, 304 articulatory effort 226, 235, 238 Bantu 252 Basque 7, 55, 79, 155 Batticaloa Creole Portuguese 14, 315 bilingual 187, 192, 196–198, 202–205, 208– 215, 225 binarity 82, 98, 173, 279, 284 Branchingness 326 Buenos Aires Spanish 18 Cambodia 254 Catalan 10, 72–73, 79, 91, 123, 190fn, 236, 250, 253fn Castilian Spanish 35, 42fn, 80, 234fn Chicano Spanish 33, 139–142, 158 Chilean 108, 132, 197 – dialects 130–133 – Spanish 114, 131 core syllabification 153, 155–156, 161, 171, 173 clitic group 189, 191, 195–196, 202, 210 coarticulation 8 coda 34fn, 36fn, 37–38, 43–45, 79–80, 109–110, 128, 132, 148–149, 152fn, 154, 159fn, 161–162, 177–179, 188–194, 196, 198, 200, 202, 245, 247–251fn, 254, 256–258, 261, 278, 282, 301 – clusters 245, 247, 250–251, 256 – consonant 80, 177, 192, 198, 200, 202, 245, 247fn, 248–250, 256–257, 278, 282, 307

– constraints 254 – position 38, 43, 159, 177–178, 190, 191fn, 250, 255–256, 259 Colombia 22, 117fn, 251, 266, 300 complex – coda (see also coda) 190fn, 247, 249–250, 254, 256fn – nucleus 31, 91, 113, 139, 150, 154–156, 158, 173, 177 – onset 18–19, 21, 29, 31, 91, 111–112, 115, 129–130, 132, 142, 155, 172–173, 180, 188fn, 219–223, 225fn, 227–231, 233– 235, 236fn, 237–238, 247, 249–251, 253–254, 256 – Onset Condition 19, 21, 112, 115, 129– 131 CV rule 193, 213 decreasing sonority 111, 246, 249, 251, 257– 258, 260–261, 266 default-to-opposite system 315–316, 322, 328 demarcative language 212 demorification 165 derivational suffixes 64, 273fn, 289, 302 desinence 53–55, 58–60, 64, 66, 72, 74, 99, 289–291, 295–296, 299–301, 303, 309 disyllabification 146–147, 165–166, 169– 170, 172, 176, 178, 181 devoicing 190 diphthong 4, 11–12, 14–16, 19fn, 20–22, 30– 31, 37, 65fn, 75fn, 76fn, 79fn, 88fn, 94– 95, 99, 146, 148–151, 152fn, 153, 155, 157, 162–164, 171, 174, 177–179, 245fn, 278, 281, 284, 304 – (see also lexical, rising diphthong) diphthongization 4fn, 6–7, 10–11, 17, 23 Dispersion Principle 220, 224, 249 Dominican Spanish 159, 219, 223, 226, 234fn, 282 DRAE 75fn, 82fn Dutch 61, 79, 123, 224, 236fn, 279, 315 Dutch-based Portuguese 14

334

Subject index

electrochemical reactions 226 empty onset 187–188, 192, 194–196, 198, 200, 213–214 – (see also onset) Enciclopedia Universal 75fn enclisis 191, 196, 207, 214 epenthetic vowel 245–246, 249–251, 253– 254, 256–261, 267 Equatorial Guinea 252 even moraic trochee 319, 320fn excrescent vowel 257fn, 259, 266 EXMARaLDA 198 extrametricality 55–56, 271–272, 285–288, 295–296fn, 308 faithfulness 98, 132, 141, 159–160, 165, 171– 172, 176fn, 177 – constraints 70, 123, 141, 163–165, 176fn, 180, 222, 237, 299fn, 300fn – violation 248 feature geometric representation 126–127 feature geometry 44–45, 49, 119, 126 finality 271, 295, 308 Finnish 79 foot – domain 56 – head 76fn, 83fn, 275, 292–293–294, 296, 299fn, 301, 309, 324, 326 – structure 18fn, 58, 69fn, 300fn, 322, 328, 330 – tail 277, 291, 299fn fortition 8, 40, 113, 117 French 58, 91, 190–191, 193fn, 195, 223, 225, 293 front glide 42, 45, 156, 159, 169, 174 Galician 91, 121–122 GEN 66, 68, 83fn Generalized Alignment 193 German 79, 187, 191–192, 194–197, 199– 202, 204–206, 208fn, 209–214, 224, 235fn – monolinguals 187, 197–198 – Spanish bilinguals 187, 192, 196, 198, 202, 208–209, 211, 214, 225 Germanic 91, 109, 191, 254 Germany 187, 197, 208, 211–212, 214–215

Ghana 252 glide 3–4, 7–8, 13, 29–38, 40–42, 44–49, 51, 52fn, 59, 61fn, 62, 65fn, 70fn, 83fn, 84fn, 88, 89fn, 99–100, 110–111, 113, 122, 130, 140–142, 145–159, 164, 172, 174, 177–178, 180–181, 221fn, 223–224, 225fn, 229–230, 232–233, 236fn, 247, 251fn, 256 consonantalization 153 – to-vowel resyllabification 153, 158–160, 174, 180 gliding 3–13, 15–24, 51, 52fn, 147, 154, 158– 159, 167, 169, 173, 175, 178 glottal stop 187, 192, 195–196, 198, 200– 201, 210–211, 214–215 – insertion 187, 192, 195–196, 198–202, 204, 206, 208–210, 212–214 Government Phonology 249 gradient variation 120 gradient shortening 328, 330 grammatical word 206 grouping language 212 Guaraní 252 head syllable 318 headship transitivity 67 heavy syllable 55–56, 61fn, 271–272, 278, 280–281, 285, 288, 295fn, 296, 299, 301, 303, 305, 307–308, 316, 322–325, 330 height harmony 31, 35–36, 39, 43, 47–48 heterosyllabicity 3–4, 8, 52, 62, 63–65, 68, 73, 75–87, 89, 94–95, 98, 109, 222 heritage language 187, 211, 214–215 hiatus 4, 6–8, 10–11, 23, 33fn, 52fn, 56fn, 86, 88, 96fn, 99, 139–142, 151, 154fn, 157, 162, 165, 167, 171, 173, 175fn, 176, 179–181 – resolution 139–142, 146, 172 high vowels 3–4 , 6, 8–9, 11, 21, 31, 42–43, 47–49, 51, 56fn, 84fn, 149fn, 161fn homorganic – nasals 115, 117 – laterals 115, 117 hypochoristics 55

Subject index

Iambic 58, 271, 280, 310, 315, 323, 326, 328 – foot 323, 327, 330 – system 315 ,322 Iberian Spanish 140, 149 identical vowel coalescence 146, 166–167 indirect licensing 328, 330 Indo-Portuguese 315 inflectional morpheme 287 Intonation Phrase 212 – (see also IPh) IPh 189, 192, 212 Kikongo 248 Kimbundu 248 laryngeal onset 202, 211 – (see also onset) lateral 109fn, 115, 117, 119, 188, 219–221, 226, 228, 230, 232–234, 235fn, 236fn Latin 5–6, 23, 33, 42, 53, 61fn, 75fn, 76fn, 91, 94–95, 99, 253fn Latin American Spanish 3, 5, 7, 9, 221, 222fn lexical – allomorphy 54fn, 57fn, 76fn – diphthong 19fn, 146, 171, 177, 179 – (see also diphthong) – marking 52, 75, 90, 99, 285 level 160, 171, 187, 228fn, 283, 315 – phonology 132, 155, 160 lexicon 83fn, 87, 89, 96fn, 188, 193–195, 198, 271–273, 277, 279, 280, 284, 286, 289, 291, 300fn, 302–304, 306, 309 Lexicon Optimization 121–122 liaison 190, 191fn, 193fn licensing 245, 254, 256–257, 315, 317, 323, 328 liquid 34fn, 46, 61fn, 63, 88fn, 107–112, 130, 159fn, 173, 219–221, 223–226, 229, 232–238, 245, 247–248, 250 – gliding 159 LO 121–122 – (see also Lexicon Optimization) loanwords 305 main stress constituent 319–320, 326fn, 328 markedness 56–58, 68, 88, 90, 98, 141, 158, 236, 249, 257

335

– constraints 52fn, 141, 163–165, 174, 181, 222, 236 Maximal Onset Principle 187 Maximal Sonority Distance 129 metrical constituency 315, 317 Mexican Spanish 7, 9–10, 12, 15, 17, 20–22, 109fn, 197, 223, 266 mid vowel 3–4, 6–11, 18, 22, 31, 35, 36fn, 42–43, 46, 48–49, 95fn, 99, 143–147, 151, 157–158, 161, 166–167, 170, 173, 176–178 – gliding 3, 5–9, 11, 13, 15–24, 167 – raising 151, 154, 167, 170, 181 Minimal Sonority Condition 249 Minimal Sonority Distance 247, 258 Misiones province 252 MLU values 211 monomoraic 148, 150, 152fn, 155, 163, 165, 279, 318, 322 – syllable 318 – vocoid 164 Montes de Pas 35 mora 76fn, 78, 83fn, 84, 141, 147–150, 162, 164–165, 179, 245–247, 254–261, 276, 278, 285, 287–288, 298fn, 307–308, 317–320, 322–323, 325, 327–330 moraic 61fn, 84fn, 141, 148–150, 165, 255– 261 – coda 177 – consonant 245, 256–257, 259, 261, 267 – phonology 142, 147 – representation 155, 179 – segment 255 – shortening 139, 150, 162, 67 – status 141 – structure 165 – syllabl 258 – theory 149–150, 162 – trochee 298fn, 319, 320fn moraicity 83fn, 262 morphological – class 289, 301fn – excrescences 64 – exponence 175 – head 293–294, 298fn, 299, 301, 303, 309–310 – restructuring 305

336

Subject index

– word 289, 293–294, 297–298, 301, 310 MSD 129–130, 132, 258–260 – (see also Maximal Sonority Distance) Nahuatl 91, 109 NC clusters 128 negative transfer 215 neuromuscular fibers 226 New Mexico 140, 253, 266, 284 non-branching onset 222, 237, 258fn non-decreasing sonority 253, 257–259, 261, 266 nonce words 34, 285, 305–306 Obligatory Contour Principle 157, 163 OCP 128, 157, 163–165, 168, 174 – (see also Obligatory Contour Principle) offglides 37–38, 41, 43, 52fn, 161, 177 offgliding 37 onglides 29, 37, 43, 46, 52fn, 173, 177 ongliding 30–31, 37 onset 3, 11, 18, 29–34, 40, 43–47, 49, 51, 70fn, 84fn, 87, 91, 99, 109–115, 118, 131, 139, 141–142, 148–158, 162, 164, 173, 177, 179–181, 188fn, 190–196, 198, 213–215, 220–224 225fn, 230, 233– 234, 236, 245, 247–250, 255–258, 278 – (see complex/ empty/ laryngeal/nonbranching onset) – acquisition 225, 237 – condition 11, 18, 21, 118, 130 – clusters 107, 109–111, 113–115, 118, 129– 133, 219, 226, 230, 245, 247–248, 250– 251, 253–254, 257–259, 261, 266 – filling principle 188 – generalization 107, 111, 133 – maximization 188, 193 – position 49, 154fn, 158, 177, 188, 191, 198, 210, 215, 229, 236, 249 – phonotactics 111 opacity 158, 159fn, 160, 167, 316 – effect 132 open syllable 52, 76, 78–81, 248 Optimality Theory 118, 139, 187–188, 193, 245, 247–248, 258, 276 OT 51, 69, 83fn, 87, 114, 118, 121–125, 129– 130, 132–133, 139, 141–142, 147, 149fn,

150, 152fn, 158, 159fn, 160–164, 167, 171–172, 176–177, 179–180, 193fn, 194– 195, 213, 222, 235, 249, 258 – (see also Optimality Theory) output underspecification 107, 115, 117–118, 120–121, 123, 125, 129 oxytone 56fn, 274, 299, 307 PAIDUS 197 palatal 48–49, 245 – consonants 30, 39–41, 43, 45–48, 79, 157 – fricative 30, 32, 41, 44, 46, 91 – glide 32fn – lateral 30, 39, 45–46 – nasa 16, 46 – segment 44 – sonorant consonants 46–47 Palenquero 117fn, 251 parallel OT 158, 161, 176, 179–180 paroxytones 56fn, 272, 274–276, 284, 291, 302, 308, 310 Pasiego 29–35, 36fn, 38–43, 45–49, 13fn – vowel harmony 29–30 PEDSES 197, 202 Peninsular dialects 247 Peninsular Spanish 8, 10, 15fn, 8fn5, 188– 189, 222fn perseverant underspecification 108, 123 phonetic underspecification 121, 123 – variation 120–121, 123 phonological – grammar 226, 318 – syllable 320 – weight 245, 254, 256, 261, 271, 300 – word 189fn, 195 phonotactic constraints 40 PO 59, 72fn, 75–76, 83 – (see also paroxytones) positional faithfulness 316 – constraint 176fn, 237 postlexical phonology 171 PPO 75–76, 92, 94, prenuclear glide 141, 256 prenasalized obstruents 251, 253fn primary stress 21, 140, 146, 152, 155, 275– 276, 283, 310 principled underspecification 121–122

Subject index

processing of speech 193 proclisis 191, 207 proparoxytones 52, 61fn, 76, 274–275, 279– 280, 282–284, 287, 294, 296fn, 298, 303, 305fn, 309–311 PP0, 75–76, 92, 94 – (see also proparoxytones) prosodic – constituency 317–318 – domain 86 – feet 271, 291 – hierarchy 60, 189, 210, 258 – licensing 188 prothetic vowel 252–253, 257 prosthesis 80, 247fn, 257fn quantity insensitivity 285, 298fn, 305–307 quantity sensitivity 55–56, 61fn, 271–272, 277–282, 284–286, 288, 295–296, 298fn, 299–300, 302–303, 305, 308– 309 Real Academia Española 4fn, 150fn, 154fn, 155fn Richness of the Base 83, 121–122, 152, 177 ROTB 121,133 – (see also Richness of the Base) Romance languages 6, 55fn, 139fn, 189, 191, 214, 235, 237, 248 rhotics 219–220, 226, 230, 236fn rhyme 4, 34, 40, 148, 155–157, 173, 187, 192, 221, 223–225, 256, 278, 282fn, 288, 299–301, 307 rising diphthong 30, 76fn, 151, 155, 157, 245fn, 281, 304 rising sonority 247, 249, 250, 253–254, 257fn, 260 – (see also diphthong) Russian 55, 123 San Basilio de Palenque 251 schwa 58, 191, 246, 254, 257, 267 sibilants 248 secondary stress 18fn, 60, 325fn semantic sex 55 semisyllables 258 serialism 132

337

simultaneous bilingual 192, 208, 214 sonority , 40, 44, 62 – dispersion principle 220, 224 – distance 247 – hierarchy 90, 234, 236, 247, 257fn – scale 111–112, 114, 236fn, 246, 248, 253 – (see also Universal Sonority Hierarchy) – sequencing 246–247, 253, 257fn – condition 255fn – sequencing principle 66, 110 Spanish 3, 5–11, 13, 18fn, 29, 30fn, 31, 33– 35, 36fn, 37–39, 41–44, 49, 51–53, 54fn, 58, 60fn, 66, 70fn, 75fn, 76fn, 79, 81, 84fn, 88fn, 89, 91, 96fn, 99–100 – desinence 53–55, 72, 99 – dialects 29, 235 – monolinguals 196–198, 204, 209, 211, 214–215 – phonology 3, 188, 228fn, 272fn – prosthesis 80 – secondary stress 60 – speakers 65, 79, 85fn, 86, 88fn, 214, 246–247, 250, 252–254, 258, 261, 266, 305 – stress 53, 55fn, 56–58, 61, 76fn, 99, 271– 272, 274, 278, 285–287, 289, 292, 297, 302–303, 306, 308, 310 – syllables 30, 247, 271, 278 – syllable structure 142, 155, 177, 220–221, 262 – varieties 7, 9–10, 20 – (see also dialects) spirantization 132, 190, 196, 212, 221fn Sri Lanka 315 standard Spanish 11, 41fn, 48 stem 35, 53, 54fn, 60, 64, 66, 74, 98, 156, 160, 194, 197, 234, 271, 289–298, 303fn, 310 – boundary 51, 58 – level 57fn – vowel 41 Stratal OT 132, 139, 160–161, 171, 179, 180 stray moraic consonant 256 stress, – abiders 289–291, 310 – attractors 33, 289–290, 296–297, 299– 300, 302, 310

338

Subject index

– advancement 298fn – bearer 55–57, 91, 152, 172, 272, 277, 292, 309 – condition hypothesis 10 – domain 58, 271, 278, 286, 290–291, 295fn, 303fn, 309 – repellers 289–295, 299fn, 303, 310 – retraction 98, 286, 294–295, 296fn – shift 7, 10, 76, 80, 142, 146, 152, 154, 172– 173, 273, 305fn – window 33 stressed syllable 23, 37, 40–43, 46–48, 86fn, 176fn, 195, 237–238, 275, 315– 316 suffixation 54, 195, 273, 294 supermarked class 56, 59 svarabhakti vowels 246, 259, 266–269 syllabic – consonant 43, 250, 253 – language 212 – rhymes 256 – sonorants 253 – trochee 271, 289, 291, 298fn, 303, 308– 310 syllabification 3–5, 8, 10–11, 23, 29, 33fn, 44, 49, 52–53, 56, 61–64, 68, 71–72, 81, 84–85, 95, 98, 109, 141–142, 146, 150, 153–156, 158, 161, 171, 173, 181, 213, 247–250, 255, 260 syllable – head theory 29, 39–40, 43, 45 – merger 71, 86, 139–142, 146–147, 148fn, 149–155, 157–158, 161–167, 171–173, 175–177, 179–181 – nucleus 61, 65, 88fn, 89fn, 99fn, 164, 225 – nucleus peak 66, 70, 88fn, 89fn – peak 51–52, 70, 90, 225 – rhyme 156, 282fn – structure 30–31, 38, 49, 141–142, 147, 155, 158, 187, 193, 214, 220–221, 262, 306, 308, 328–330 – weight 61, 147, 148fn, 149, 302, 315 Sympathy Theory 159fn, 316, 330 synchronic variation 5–6

tap 220fn, 234–235, 236fn tautosyllabicity 52, 62–63, 65, 67–68, 71, 73–74, 77–81, 84–87, 91, 94 ,96, 98 trill 236fn trochaic foot structure 58 trochee 55, 176, 319 – (see binary/moraic/syllabic trochee) Turkish 39, 44–46, 49, 122 – vowel harmony 30, 49 underspecification 107–108, 115, 117–118, 120–123, 125, 127, 129, 133 uneven moraic trochee 319fn universal markedness 141, 158, 180 Universal Sonority Hierarchy 220, 224, 226, 235, 236fn unsyllabified – moras 256 – moraic consonant 245, 259, 261 velarization 188 Verb Class Condition 11, 19, 22 vocalic sequences 3, 6 vocalization 130–131, 133, 159 voicing 121, 123–124, 190, 230, 233, 237 vowel – coalescence 158, 162, 164–165, 181 – (see identical vowel coalescence) – deletion 158 – harmony 29–31, 32fn, 35, 36fn, 38–39, 43–49 – (see also Pasiego and Turkish vowel harmony) Vulgar Latin 5–6, 23, 94 Walpiri 122 word – final nasal coda 190 – segmentation 190 Yoruba 122

Index of constraints *#C (L )H ͡ˈV 87, 89–92, 96, 98 *CODA 222, 248 *COMPLEX N 177 *COMPLEX ONSET 251fn *fricative-ONS 236 *GEM 179 *glide-ONS 236 *HD σH ighV 84fn *H.é 95–96, 98 *IAMB 58–61, 67–68, 98 *Ku.ˈV 63–64 *K/u= 99 *lateral-ONS 219–220, 237 *nasal-ONS 236 *ONSET- GLIDE 158 *rhotic-ONS 219–220, 236 *stop-ONS ,X 236 *V ̆.V ̆ 63–64, 71, 78, 87, 89–90, 95, 97–98 *'V.H 83, 85, 87, 90, 92–94, 97 AL- FT,]

58–60, 66–68, 72, 78, 98 58–61, 63, 66–69, 72, 78, 90, 98 ALIGN LEFT 194, 213 ALIGN (PW,L,FT,L) 194–195 ALIGN - R 323–326 ANCHOR - L 161, 164–165, 167, 174, 176, 178 ANCHOR - LEFT 141, 165 – (See also, ANCHOR - L ) AL-ˈV,]

BINFT M 58–61, 66–68, 83, 98 BRANCH -μ 163fn

ID -[cont] 119, 122, 124–125, 127–128 ID -[voi] 119, 124–125, 128 IDENT-[back] 164–165 IDENT-[cons] 158–159 IDENT-[ DIPH ] 171, 179 IDENT-[HI] 178 IDENT-[high] 164–167 IDENT /\-IO 324–325 IDENT-[low] 164–165, 169 IDENT-PrWd[cons] 158–159 IDENT-stress 172 IDENT-σ́(F) 176 LIC-μi

323–326 LINK 126, 128

MAIN -RIGHT 276–277, 280 MAX 51, 258–261 MAX -[cont] 124–125 MAX -μ 164–167 MAX - PK 70, 88, 91, 98 MAX - V 161fn MINIMALSIZE 326 MSD 129–130, 132, 258–260 NC [cont]

119–120, 126–128 161, 177–179 – (see also *CODA ) NODIPH 81fn NO - LV 164–165, 178 NO - CODA

91, 98 276–280 FOOTCAP 78

107–108 128, 163–165, 168, 174 – 1-TO -1 161fn ONS 63, 69–70, 72fn, 75fn, 81fn, 83–84, 87–88, 90, 98, 219–220 ONSET 51–52, 68–69, 70fn, 71, 78, 90, 92, 117, 141, 163–165, 171, 174, 178–180, 194–195, 213, 222 – (see also ONS) ONS LEX 90

GLIDE - HI

PARSE - V (R )

OC CONTIG - IO

248–249

120fn, 194, 248, 258–261 DEP -[cont] 123–125, 128 DEP - IO 164–165, 194–195, 213 DEP

FAITH -SEG FOOTBIN

141, 163–165

OCP

61, 98 1 320 PARSE -Line 2 320 PARSE -Line

HEADSTACK

67

340

Index of constraints

320 59–61 PKMINSON (SP ) 66

TROCH 276–277

(SON = SYL )-PK 66–68 SON FALL ’ 81 SPIR 132 SSC 258–261 SSP 110 SYL-μ 258–261

WSP

PARSE -μ

PARSE -σ

UNIFORMIT Y ( UNIF )

75 324–325

W -to-S

μ-SYL 258–261

164–165, 167, 178