Catalan Journal of Linguistics

CMY CY MY CM K Y M C Centre deLingüística 9 5 Index 105 79 41 131 161 191 shortening, voweldeletionandgliding. Stress. perception study. Stressed Clitic Pronouns in Two Spanish Varieties: A Taviano (Salentino)Italian. Augmentation viaNasalizationinBrazilianPortuguese. Allomorphy. Structures in Eivissan Catalan: RecursiveProsodic Words and Shih, Shu-hao;deLacy,Paul.EvidenceforSonority-Driven Martínez-Gil, Fernando;Ohannesian,Maria.Introduction. Colantoni, Laura; Hualde, José-Ignacio; Icardo Isasa, Ane. Isasa, Icardo José-Ignacio; Hualde, Laura; Colantoni, Jacobs, Haike. Pruitt, Kathryn.RevisitingTop-DownPrimaryStress. Kenstowicz, Michael. Nevins, Andrew; Pinheiro Costa, Paula. Torres-Tamarit, Francesc; Bonet, Eulàlia. Institut Interuni ISSN 1695-6885(inpress);2014-9719(online) Catalan Journal ofLinguistics Catalan Journal Stress anditsphonologicalinteractions T eòrica delaUni https://revistes.uab.cat/catJL Cross-level interactions in Latin: Vowel v ersitari deFilologia Vol. 18,2019 The analysis of truncated vocatives in v ersitat Autònoma deBarcelona Servei dePublicacions V alenciana Prominence Verb-Clitic

Stress and its phonological interactions CCAATTALAN JJOURNAL OFL LINGUISTICS Vol. 18 Centre deLingüística OF Stress anditsphonologicalinteractions Fernando Martínez-Gil &MariaOhannesian Martínez-Gil Fernando Institut Interuni ISSN 1695-6885(inpress);2014-9719(online) C C J J L L OURNAL T INGUISTICS A https://revistes.uab.cat/catJL A eòrica delaUni V v olume ersitari deFilologia T T Vol. 18,2019 Edited by 2019 A L v ersitat 18 AN Autònoma deBarcelona V Servei dePublicacions alenciana Catalan Journal of Linguistics Editors Centre de Lingüística Teòrica de la Universitat Autònoma de Barcelona Institut Interuniversitari de Filologia Valenciana Catalan Journal of Linguistics style sheet

Informació general i subscripcions General information and subscriptions • Document format. Manuscripts should be should be listed chronologically, with the CATALAN JOURNAL OF LINGUISTICS és una revista de lingüís- CATALAN JOURNAL OF LINGUISTICS is a journal of theoretical written in English. Font 12 point Times New oldest publication at the top and the newest at tica teòrica, publicada pel Centre de Lingüística Teòrica linguistics published by the Centre de Lingüística Roman, interline at 1.5, use margins 2,5 cm the bottom. Please, provide the DOI of articles de la Universitat Autònoma de Barcelona (format per Teòrica de la Universitat Autònoma de Barcelona (1 inch). Phonetic symbols must be set in if available. lingüistes dels departaments de Filologia Catalana i de (comprising linguists from the departments of Catalan Doulos SIL or Charis SIL. a) Book: Filologia Espanyola) i per l’Institut Interuniversitari Philology and Spanish Philology) and by the Institut • Heading. Title, with title capitalisation, in Borer, Hagit. 1983. Parametric Syntax. de Filologia Valenciana (que integra lingüistes de les Interuniversitari de Filologia Valenciana (which boldface. Name of the author(s), affiliation(s) Dordrecht: Foris. universitats de València, d’Alacant i Jaume I de Castelló). comprises linguists from the universities of València, and e-mail address(es). Acknowledgements, DOI: http://dx.doi.org/10.1515/97831 CATALAN JOURNAL OF LINGUISTICS publica articles d’inves- Alacant and Jaume I at Castelló). as a footnote, called by a flying asterisk. 10808506 tigació en lingüística teòrica que desenvolupin teories CATALAN JOURNAL OF LINGUISTICS publishes research papers sobre l’estructura de les llengües particulars dins la in theoretical linguistics developing theories on the • Abstracts. A 150 word abstract is required. b) Journal article: perspectiva d’una teoria general del llenguatge humà. structure of particular languages within the perspective A Catalan version of the abstract should be Kayne, Richard S. 1993. Toward a modu- Els volums seran sovint monogràfics, editats per un of a general theory of the human language. Volumes will included. Authors are also asked to provide lar theory of auxiliary selection. Studia especialista en el tema. often be monographic, edited by a specialist in the topic. the keywords of the article. Linguistica 47: 3-31. € € DOI: http://dx.doi.org/10.1111/j.1467- El preu de subscripció anual és de 18 . The yearly subscription price is 18 . • Sections and paragraphs. Headings of 9582.1993.tb00837.x main sections: in boldface. Headings of first Comitè assessor Redacció order paragraphs: in italics. Headings of c) Book article: Josep Maria Brucart Marraco Universitat Autònoma de Barcelona second order paragraphs or others: in italics Jaeggli, Osvaldo & Safir, Ken. 1989. The (Universitat Autònoma de Barcelona) Departament de Filologia Catalana with text beginning at the same line. Null Subject Parameter and Parametric Maria Teresa Espinal Farré 08193 Bellaterra (Barcelona). Spain Theory. In Jaeggli, Osvaldo & Safir, Phone: +34 93 581 23 71 (Universitat Autònoma de Barcelona) • Text. The beginning of a paragraph should Ken (eds.). The Null Subject Parameter, C +34 93 581 23 72 Anna Gavarró Algueró be left indented, except for the first paragraph 1-44. Dordrecht: Kluwer. (Universitat Autònoma de Barcelona) Fax: +34 93 581 27 82 of a section. Words, phrases or sentences in a DOI: http://dx.doi.org/10.1007/978-94- M Maria Lluïsa Hernanz Carbó E-mail: [email protected] language other than English appearing in the 009-2540-3_1 Y (Universitat Autònoma de Barcelona) Intercanvi body of the text should be followed by a Sabine Iatridou (Massachusetts Institute of Technology) Universitat Autònoma de Barcelona translation in simple quotation marks. d) Dissertation, manuscript, or work in CM Michael J. Kenstowicz Servei de Biblioteques References should follow this format: preparation: (Massachusetts Institute of Technology) Secció d’Intercanvi de Publicacions Pesetsky, David. 1982. Paths and MY Chomsky (1981), Kayne (1981, 1982), Belletti Joan Mascaró Altimiras Plaça Cívica (edifici N) & Rizzi (forthcoming), Jackendoff (p.c.). If Categories. PhD dissertation, MIT. (Universitat Autònoma de Barcelona) 08193 Bellaterra (Barcelona). Spain Chomsky, Noam. 1994. Categories and CY the page number or an example is expressed: Gemma Rigau Oliver Phone: +34 93 581 11 93 Rizzi (1990: 20), Rizzi (1990: ex. (24)). When Transformations. Unpublished manus- CMY (Universitat Autònoma de Barcelona) Fax: +34 93 581 32 19 the reference is included between parentheses, cript, MIT. Henriette de Swart (Universiteit Utrecht) E-mail: [email protected] K • Publication rights. CATALAN JOURNAL OF Rafaella Zanuttini (Yale University) drop the ones enclosing the year: (see Rizzi Subscripció, administració, edició i impressió 1990: 20). LINGUISTICS is published under the licence Consell editor Universitat Autònoma de Barcelona • Examples. Example numbers should go system Creative Commons, according to the Teresa Cabré Monné Servei de Publicacions modality «Attribution-Noncommercial (by-nc): (Universitat Autònoma de Barcelona) 08193 Bellaterra (Barcelona). Spain between parentheses, left aligned (Arabic digits for examples in the text; lower case derivative work is allowed under the condition Ángel Gallego Bartolomé Phone: +34 93 581 10 22 of non making a commercial use. The original (Universitat Autònoma de Barcelona) E-mail: [email protected] Roman for examples in footnotes). Example work cannot be used with commercial Jesús Jiménez Martínez (Universitat de València) http://publicacions.uab.cat texts should start at the first tab. If there is Sandra Montserrat Buendía (Universitat d’Alacant) ISSN 1695-6885 (in press); ISSN 2014-9719 (online) more than one example within a number, purposes». Therefore, everyone who sends a Xavier Villalba Nicolás Dipòsit legal: B-4.112-2003 lower case alphabet letters should be used manuscript is explicitly accepting this publica- (Universitat Autònoma de Barcelona) Imprès a Espanya. Printed in Spain followed by a stop and a tab. Examples must tion and edition cession. In the same way, Imprès en paper ecològic be referred to as (1), (5a,b), (6b-d), (ii), (iii he/she is authorizing CATALAN JOURNAL OF c-e), (iv a,b), etc. LINGUISTICS to include his/her work in a journal’s issue for its distribution and sale. The Bases de dades en què CATALAN JOURNAL OF LINGUISTICS està referenciada • Glosses. Examples from languages other cession allows CATALAN JOURNAL OF LINGUISTICS — Bibliographie Linguistique — Linguistic Bibliography than English should be accompanied by a to publish the work in a maximum period of two — CARHUS Plus+ — Linguistic Working Papers Directory gloss and, if necessary, by a translation into years. With the aim of favouring the diffusion — CiteFactor — Linguistics Abstracts English. Align glosses with examples word of knowledge, CATALAN JOURNAL OF LINGUISTICS — Dialnet — LLBA (Linguistic and Language Behavior Abstracts) by word. Translations should be in simple joins the Open Access journal movement — Francis — Scopus quotation marks. — Índice Español de Ciencias Sociales (DOAJ), and delivers all its contents to different y Humanidades (ISOC) • References. References should be according repositories under this protocol; therefore, to the following pattern, with full names of sending a manuscript to the journal also entails CATALAN JOURNAL OF LINGUISTICS es publica sota el sistema de llicències Creative Commons segons la modalitat: authors included. Reference entries for the explicit acceptation by its author/s of this Reconeixement - NoComercial (by-nc): Es permet la generació d’obres derivades multiple works by the same author(s)/editor(s) distribution method. sempre que no se’n faci un ús comercial. Tampoc es pot utilitzar l’obra original amb finalitats comercials. Catalan Journal of Linguistics 18, 2019 1-4

Summary Catalan Journal of Linguistics = CatJL Volume 18, pp. 1-215, 2019 ISSN 1695-6885 (in press); ISSN 2014-9719 (online) https://revistes.uab.cat/catJL

5-8 Martínez-Gil, Fernando (The Ohio State University); Ohannesian, Maria (Universitat Autònoma de Barcelona) Introduction. Catalan Journal of Linguistics, 2019, vol. 18, pp. 5-8.

Articles 9-40 Shih, Shu-hao (UCLA); de Lacy, Paul (Rutgers University) Evidence for Sonority-Driven Stress. Catalan Journal of Linguistics, 2019, vol. 18, pp. 9-40. We argue that there is no adequate evidence for ‘sonority-driven stress’, building on Shih (2018a,b), and disagreeing with Kenstowicz (1997), de Lacy (2002a, 2004, 2006), and others. More precisely, we argue that there is no phonological mechanism that induces metrical structure to deviate from its default position for reasons that involve the direct interaction of segmental sonority and foot form. After reviewing the history of sonority-driven stress theory, we identify two broad issues with extant evidence: the lack of methodological reliability, and misattribution of cause. We argue that impressionistic descriptions of sonority- driven stress are not reliable, in the technical sense of evidentiary validity. We further argue that apparent sonority-sensitivity in foot form is a side-effect of either allophony or minor syllable behavior. Keywords: phonology; metrical structure; sonority-driven stress; sonority; stress; evidence

41-77 Pruitt, Kathryn (Arizona State University) Revisiting Top-Down Primary Stress. Catalan Journal of Linguistics, 2019, vol. 18, pp. 41-77. Metrical theory recognizes differences between primary and non-primary stresses, sometimes within the same language. In serial theories, this has often led to a parametric approach in derivation: some languages are ‘top-down’, with the primary stress assigned first, while other languages are ‘bottom-up’, where foot construction precedes primary stress placement. This paper examines two languages (Cahuilla and Yine) that have be treated as ‘top-down’ in rule-based metrical theory, and it shows that neither requires a top-down analysis in Harmonic Serialism, a derivational version of Optimality Theory. On the basis of these case 2 CatJL 18, 2019 Summary

studies it is argued that the common, intuitive notion of what makes a language ‘top-down’—a primary stress’s independence from non-primary stresses—is oversimplified. The case studies reveal the importance of theoretical framework and typological predictions in establishing the order of primary and non-primary stress assignment. The argument culminates in a concise statement of Harmonic Serialism-specific criteria for establishing that a top-down derivation is required. Keywords: primary stress; Harmonic Serialism; metrical theory; top-down; bottom-up

79-103 Jacobs, Haike (Radboud University) Cross-level interactions in Latin: Vowel shortening, vowel deletion and vowel gliding. Catalan Journal of Linguistics, 2019, vol. 18, pp. 79-103. Serial and parallel OT differ in the way they account for phonological generalizations referring to more than one level of the prosodic hierarchy. Vowel shortening in Latin is analyzed by McCarthy, Pater & Pruitt (2016) as a case in point. Vowel shortening takes place to optimize foot structure. In parallel OT, footing and shortening can be evaluated in parallel, but in serial OT footing and shortening necessarily take place in a serial derivation. In this paper, both the serial and the parallel analysis of Latin vowel shortening are critically discussed. After that, two other potential cases of cross-level interactions in Latin are addressed: vowel deletion and vowel gliding. For each of these cases, it is argued that a serial analysis has to be preferred over a parallel one. Keywords: Serial Optimality theory; Parallel Optimality theory; Latin vowel shortening, vowel deletion and vowel gliding; Cross-level interactions; Uneven trochee

105-129 Colantoni, Laura (University of Toronto); Hualde, José-Ignacio; Icardo Isasa, Ane (University of Illinois at Urbana-Champaign) Stressed Clitic Pronouns in Two Spanish Varieties: A perception study. Catalan Journal of Linguistics, 2019, vol. 18, pp. 105-129. Clitic elements are prosodically attached to another phonological word. As such, they do not carry their own stress. In general, Spanish non-prepositional object pronouns (e.g. me/te/le/lo/se) meet the definition of clitic, whether preposed (proclitic) or postposed (enclitic) to the verb. Nevertheless, in several Romance varieties, including some Spanish varieties, the stress may surface on the clitic in verb+clitic(s) groups. We refer to this phenomenon as Enclitic Stress Shift (ESS). In this paper, we report on the perception of ESS by Spanish speakers from Argentina and Spain and explore the acoustic dimensions that may affect this perception. The discussion includes comparisons with other Romance languages and implications for sound change, such as the changes in stress patterns in the development of future and conditional tenses. Keywords: stress perception; clitics; secondary stress; Spanish phonology; prominence; Argentine Spanish; Peninsular Spanish Summary CatJL 18, 2019 3

131-159 Kenstowicz, Michael (Massachusetts Institute of Technology) The analysis of truncated vocatives in Taviano (Salentino) Italian. Catalan Journal of Linguistics, 2019, vol. 18, pp. 131-159. This paper documents and discusses various descriptive generalizations and alternative analyses of the vocative truncation found in the Southern Italian dialect of Taviano that is illustrated by such formations as Filoména > Filomé. We show that the upper limit on the size of the truncate is not restricted by foot binarity nor by internally layered feet: Addoloráta > Addolorá. The lower limit of the truncate may extend to a single CV syllable in violation of foot binarity: Césare > Cé. But truncation to just the stressed vowel is rejected: Élena > *É. We situate the analysis of the vocative truncation within the overall grammar of Salentino and show that while some properties such as the loss of coda consonants can be attributed to rankings imposed by the surrounding grammar, others are peculiar to this construction. The analysis of stress in loanword adaptation figures prominently in our discussion. The final sections examine some of the morphological and syntactic properties of the Taviano vocative. Keywords: vocative; truncation; loanword adaptation; stress

161-189 Nevins, Andrew (University College London); Pinheiro Costa, Paula (Universidade Federal do Rio de Janeiro) Prominence Augmentation via Nasalization in Brazilian Portuguese. Catalan Journal of Linguistics, 2019, vol. 18, pp. 161-189. This article aims to demonstrate that dialectal and idiolectal variants of Brazilian Portuguese that exhibit rhotic metathesis (e.g. vidro > vrido ‘glass’), spontaneous nasalization of high vowels (as in diachronic hibernum > inverno ‘winter’ and non-standard ingreja ‘church’), and pretonic vowel lowering of mid-vowels are all instantiations of the same process: prominence-boosting in stressed, secondary-stressed, or word-initial positions. Keywords: spontaneous nasalization; Brazilian Portuguese; rhotic metathesis; prominence augmentation; initial syllables

191-215 Torres-Tamarit, Francesc (SFL, Université Paris 8, CNRS); Bonet, Eulàlia (Universitat Autònoma de Barcelona) Verb-Clitic Structures in Eivissan Catalan: Recursive Prosodic Words and Allomorphy. Catalan Journal of Linguistics, 2019, vol. 18, pp. 191-215. This paper deals with stress shift in verb-clitic structures in Eivissan Catalan, an understudied Romance variety. Within Balearic Catalan, this is the only subdialect in which stress shift is restricted to apply only in second conjugation pre-clitic infinitives, those that, as opposed to other conjugations, have penultimate stress when they are pronounced in isolation. Stress in second conjugation infinitives in Eivissan Catalan shifts one syllable to the right, that is, to the final syllable of the verbal stem, when one or more pronominal enclitics follow. There 4 CatJL 18, 2019 Summary

is no stress shift in pre-clitic imperatives. We claim that pronominal enclitics in Eivissan Catalan adjoin to a recursive, maximal prosodic word, and that the domain for stress assignment is the minimal, embedded prosodic word. We further analyze two cases of stress-conditioned allomorphy (i.e. allomorphy of the infinitive morph and allomorphy of the verbal root) that occur in infinitive-clitic structures. Keywords: allomorphy; clitic; Eivissan Catalan; prosodic word; recursivity; stress shift Catalan Journal of Linguistics 18, 2019 5-8

Introduction

Fernando Martínez-Gil The Ohio State University [email protected] Maria Ohannesian Universitat Autònoma de Barcelona [email protected]

The stress systems of human languages have been the subject of intensive research by contemporary phonological theory since the development of metrical phonology in the final quarter of the twentieth century. In fact, stress studies have played a prominent role in phonological theory since Chomsky and Halle’s The Sound Pattern of English (1968) (SPE), where stress is analyzed as one of the set of language’s distinctive features. The special status of stress was captured in later work such as Liberman’s (1975) metrical grid, in which “stress is defined in terms of an abstract two-dimensional array that plots metrical positions for level of prominence” (Kentowicz 1993: 553), thus providing the basic tools to capture the primary and secondary stress assignment parameters in the world’s languages, a framework enriched later on in the seminal works by Halle and Vergnaud (1987) and Idsardi (1992). It is also worth mentioning Hayes’ (1995) influential treatise on metrical theory, which according to Kager (1995: 437) put forth “a new and highly articu- lated version of metrical stress theory whose major theoretical innovation is an asymmetric foot inventory” based on the analysis of the “stress patterns of a large number of languages, many of which had not been previously analyzed in the metrical literature”. More recently, within the framework of Optimality Theory, the relation between prosodic structure constituents and stress assignment parameters such as stress assignment directionality, (un)boundedness systems, head position, footing and weight (in)sensitivity systems has been seen to be controlled by a ranked set of constraints. Much of the work on stress and stress systems carried out after SPE has been extended successfully in both comparative analyses of various languages and monographic studies of stress in languages other than English, such as Harris’s (1983, 1985, 1989a, 1989b) and Roca’s (1988, 1997) works on Spanish, or Mester’s (1994) trailblazing exploration of Latin’s quantitative trochee. In this monographic collection of studies, we have not attempted to present a comprehensive survey of how stress interacts with other aspects of linguistic structure. Rather, it brings together seven representative samples of the multiple perspectives taken in current research on this topic, each embodying a different approach to the analysis of stress and its connections with other components of the grammar.

ISSN 1695-6885 (in press); 2014-9718 (online) https://doi.org/10.5565/rev/catjl.295 6 CatJL 18, 2019 Fernando Martínez-Gil; Maria Ohannesian

The core idea of these seven contributions to this volume is the interaction between stress and other aspects of linguistic structure, with a main focus on those of a phonological nature. Stress location can be determined in a variety of ways, depending on the particular language in question: in some languages it can be dependent on purely phonological factors, as in the well-known role of syllable weight in stress locus. In others, it can be lexically encoded. And in still other languages, it can be conditioned by morphological factors, or by a combination of phonological and morphological conditions. By the same token, vowels can interact with stress in a number of ways, as seen, for example, in stress-dependent vowel harmony (Mascaró 2015), or in the intriguing issue discussed here in de Lacy and Shih’s article “Evidence for sonority-driven stress”. In fact, on the basis of an exhaustive survey of data from a variety of languages, de Lacy and Shih find no clear evidence to support the notion of sonority-driven stress, and conclude that the putative dependence of stress locus on sonority arises from impressionistic descriptions that lack methodological reliability. Pruitt’s “Revisiting top-down primary stress” explores the topic known as ‘top- down’ primary stress systems by looking at Cahuilla, Yine and other languages with a similar stress system. Framed in Harmonic Serialism theory, her analysis examines the order of primary and non-primary stress assignment in top-down and bottom-up languages, and surmises that these terms constitute theoretical or analytical tools rather than true mechanisms of stress assignment in particular languages. Five of the contributions are centered around stress in Latin and specific Romance languages. It is well known that Latin vocalism was affected by different alterations. In “Cross-level interactions in Latin: vowel shortening, vowel deletion and vowel gliding”, Jacobs explores how these processes affected more than one level of the prosodic hierarchy. He shows that though they can be accounted for by either a parallel or a serial version of Optimality Theory, it is the serial version analysis that ultimately provides a better fit with the facts. Colantoni, Hualde and Isasi’s “Stressed clitic pronouns in two Spanish varieties: a perception study” describes an experimental study which looked at how stress can shift to clitics in verb+clitic groups in Argentinian Spanish and Peninsular Spanish, despite the fact that clitics are non-tonic by definition. This phenomenon, which the authors label ‘enclitic stress shift’, is then compared with what occurs in other Romance languages such as Sardinian and Majorcan Catalan. Truncation and stress are closely related. Truncation appears in vocatives in some Romance varieties, such as the Southern Italian dialect Taviano, discussed here in “The analysis of truncated vocatives in Taviano (Salentino) Italian” by Kenstowicz. These authors explore the properties of vocative truncation and distinguish those that belong to the language’s grammar properly speaking and those that are specific to the construction in question. The article also includes an analysis of how stress is assigned in loanword adaptation. In “Prominence augmentation via nasalization in Brazilian Portuguese”, Nevins and Pinheiro investigate stress and its relation with prominence enhancement in dialectal and idiolectal variants of Brazilian Portuguese, as well as in other Romance Introduction CatJL 18, 2019 7 languages. They explore three basic strategies, namely the spontaneous nasalization of high vowels, rhotic metathesis and the lowering of mid vowels in pretonic position, whose effect is to increase prominence in primary stress, secondary stress and word-initial positions, all of them already prominent positions. This volume concludes with Torres-Tamarit and Bonet’s exploration of stress shift from a different perspective than that adopted for the similar phenomenon in Argentinian Spanish. In “Verb-clitic structures in Eivissan Catalan: recursive prosodic words and allomorphy”, the authors note that the pattern of stress shift in verb+clitic groups is much more restrictive in this understudied Catalan variety in comparison with other Balearic Catalan varieties. Framed in the Optimality Theory model, the authors’ analysis argues for the recursion of prosodic words. To sum up, the articles that make up this volume deal with a variety of issues related to primary and secondary stress assignment, including the interaction of stress with sonority, vowel changes in Latin, stress shift in verb+clitic groups in Spanish and Catalan varieties, the role of stress in truncation and loanword adaptation, and stress location in prominence augmentation strategies. We hope that readers will find this monographic volume both interesting and useful. Finally, we would like to thank the authors for their contributions, the two anonymous reviewers of each article for their guidance and insights, and Teresa Cabré for her editing of the Catalan versions of the abstracts.

References Chomsky, Noam & Halle, Morris. 1968. The Sound Pattern of English. New York: Harper & Row. Halle, Morris & Jean-Roger Vergnaud. 1987. An essay on stress. (Current Studies in Linguistics 15). Cambridge, Mass.: MIT Press. Harris, James. 1983. Syllable Structure and Stress in Spanish: A Nonlinear Analysis. Cambridge, MA: MIT Press. Harris, James. 1985. Spanish diphthongisation and stress: a paradox resolved. Phonology Yearbook 2: 31-45. Harris, James.1989a. Our present understanding of Spanish syllable structure. In Bjarkman, Peter & Hammond, Robert (eds.). American Spanish pronunciation, 151-169. Washington: Georgetown University Press. Harris, James. 1989b. The Stress Erasure Convention and Cliticization in Spanish. Linguistic Inquiry 20.3: 339-363. Hayes, Bruce. 1995. Metrical Stress Theory: Principles and Case Studies. Chicago: University of Chicago Press. Idsardi, William. 1992. The Computation of Prosody. Cambridge, MA: MIT Ph.D. dissertation. Kager, René. 1995. Review Article. Phonology 12.3: 437-464. Kentowicz, Michael. 1994. Phonology in Generative Grammar. Oxford: Blackwell. Liberman, Mark. 1975. The intonational system of English. Cambridge, MA: MIT Ph.D. dissertation. Mascaró, Joan. 2016. On the typology of metaphony/stress dependent harmony. In F. Torres-Tamarit, Francesc, Linke, Kathrin & van Oostendorp, Marc (eds.). 8 CatJL 18, 2019 Fernando Martínez-Gil; Maria Ohannesian

Approaches to metaphony in the languages of Italy, 259-276. Berlin/Boston: Walter de Gruyter. Mester, Armin. 1994. The quantitative trochee in Latin. Natural Language and Linguistic Theory 12: 1-61. Roca, Iggy. 1988. Theoretical implications of Spanish word stress. Linguistic Inquiry 19: 393-424. Roca, Iggy. 1997. On the role of accent in stress systems: Spanish evidence. In Martínez-Gil, Fernando & Morales-Front, Alfonso (eds.). Issues in the Phonology and Morphology of the Major Iberian Languages, 619-664. Washington, D.C.: Georgetown University Press. Catalan Journal of Linguistics 18, 2019 9-40

Evidence for Sonority-Driven Stress*

Shu-hao Shih UCLA [email protected] Paul de Lacy Rutgers University [email protected]

Received: December 24, 2018 Accepted: July 19, 2019

Abstract

We argue that there is no adequate evidence for ‘sonority-driven stress’, building on Shih (2018a,b), and disagreeing with Kenstowicz (1997), de Lacy (2002a, 2004, 2006), and others. More precisely, we argue that there is no phonological mechanism that induces metrical structure to deviate from its default position for reasons that involve the direct interaction of segmental sonority and foot form. After reviewing the history of sonority-driven stress theory, we identify two broad issues with extant evidence: the lack of methodological reliability, and misattribution of cause. We argue that impressionistic descriptions of sonority-driven stress are not reliable, in the technical sense of evidentiary validity. We further argue that apparent sonority-sensitivity in foot form is a side-effect of either allophony or minor syllable behavior. Keywords: phonology; metrical structure; sonority-driven stress; sonority; stress; evidence

Resum. Evidència de l’accent sota control de la sonicitat

Sostenim que no hi ha prou evidència per postular un accent controlat per la sonicitat, basant-nos en Shih (2018a,b), i estem en desacord amb Kenstowicz (1997), de Lacy (2002a, 2004, 2006), entre altres. Més concretament, defensem que no hi ha cap mecanisme fonològic capaç de fer que l’estructura mètrica es desviï de la seva posició no marcada a causa de la interacció directa entre la sonicitat dels segments i l’estructura del peu. Un cop revisats els treballs sobre l’accent sota control de la sonicitat, identifiquem dos problemes generals d’acord amb les proves que tenim a l’abast: la manca de fiabilitat metodològica i l’error en la determinació de la causa. Sostenim que les descrip- cions impressionistes de l’accent sota control de la sonicitat no són fiables, en el sentit tècnic de validesa probatòria. Finalment, argumentem que l’aparent sensibilitat a la sonicitat de l’estructura del peu és un efecte col·lateral de l’al·lofonia o del comportament de les síl·labes menors. Paraules clau: fonologia; estructura mètrica; accent sota control de la sonicitat; sonicitat; accent; evidència

* We thank the following members of the Rutgers Phonology Laboratory for their key contributions to our understanding of sonority-driven stress: Eileen Blum, Vartan Haghverdi, and Chris Oakden. We would also like to thank the following people for their comments on various precursors to this article: Adam Jardine, Akinbiyi Akinlabi, Matt Gordon, John Kingston, Catherine Kitto, and two anonymous reviewers.

ISSN 1695-6885 (in press); 2014-9718 (online) https://doi.org/10.5565/rev/catjl.256 10 CatJL 18, 2019 Shu-hao Shih; Paul de Lacy

Table of Contents 1. Introduction 4. Implications 2. Theory References 3. Evidence

1. Introduction We inquire as to whether there is adequate evidence for phonological mechanisms that induce metrical structure to deviate from its default position because of requirements that directly involve segmental sonority. The empirical effect of such mechanisms is called ‘sonority-driven stress’ (Kenstowicz 1997; de Lacy 2002a, 2004, 2006, 2007). We argue here that there is no adequate evidence for sonority-driven stress, building on Shih (2018a,b). However, we also argue that sonority can indirectly affect metrical structure: if segments of certain sonority are required to have particular moraic configurations and metrical structure is appropriately sensitive to moraic content, then segmental sonority can appear to influence metrical structure. As an example, de Lacy (2002a, 2006§5.3) argues that Gujarati exhibits sonority-driven stress. The default position for its quantity-insensitive trochaic metrical foot is at the right edge of a Prosodic Word, as illustrated in (1a). However, if an antepenultimate syllable contains the most sonorous vowel – [a] – and the penult has a less sonorous vowel, the foot retracts (1b). Additionally, if the final syllable has an [a] and both the penult and antepenult contain less sonorous vowels, the foot compresses to encompass the final syllable alone (1c).

(1) Gujarati sonority-driven stress according to de Lacy (2006: 233-4) (a) (i) [(ˈsa.ɖa)] ‘peasants’ (ii) [ap(ˈwa.na)] ‘to give’ (iii) [mu(ˈba.ɾək)] ‘congratulation’ (b) (i) [(ˈma.ni)to] ‘favorite’ (ii) [(ˈta.dʒe)təɾ] ‘recently’ (iii) [(ˈbrah.mə)ɳo] ‘priestly caste’ (c) (i) [hɛ(ˈɾan)] ‘distressed’ (ii) [ʃi(ˈkaɾ)] ‘a hunt’ (iii) [hoʃi(ˈjaɾ)] ‘clever’ (iv) [pəhɛ(ˈlã)] ‘in the past’

While de Lacy (2002a, 2006) argues that the Gujarati metrical system involves many additional complexities involving vocalic sonority, of present interest is the phonological proposal: it involves a mechanism that penalizes particular configurations of sonority and metrical structure directly. Specifically, the Optimality

Theoretic constraint *ΔFt≤{e,o} is violated whenever the nucleus of the head syllable of a foot contains a vowel that has sonority less than or equal to the ‘peripheral mid vowel’ category. When this constraint outranks other constraints on foot loca- Evidence for Sonority-Driven Stress CatJL 18, 2019 11 tion and moraic quantity, it can induce feet to deviate from their default location. This constraint is claimed to be part of a larger family of constraints that directly relate sonority levels to prosodic positions – both heads and non-heads of all categories in the prosodic hierarchy. The question we pose in this article is whether there is adequate evidence for such phonological mechanisms: Can segmental sonority directly influence the position of metrical feet? The answer presented below relies on research on sonority-driven stress under- taken by members of the Rutgers Phonology Laboratory over the past several years: Shih (2016, 2018a,b), Haghverdi (2016) and Blum (2018). This research examined cases of sonority-driven stress using experimental methodologies and acoustic analysis. Our finding is that there is no reliable evidence that metrical feet are attracted to or repelled by segments of particular sonority levels, contrary to the theories and claims of Kenstowicz (1997), de Lacy (2002a, 2004, 2006, 2007) and others. Section 2 discusses the theory of sonority-driven stress with the aim of clearly defining the issue. Section 3 reviews the evidence for sonority-driven stress. It focuses on flaws in this evidence: the unreliability of impressionistic descriptions of the phenomenon, and misattribution of cause to apparent sonority-sensitive in metrical systems. Section 4 discusses the theoretical implications of our findings.

2. Theory Research on metrical structure – whether represented as grids or feet – has argued that it is attracted to prosodic and/or morphological edges. Depending on the phonological system, metrical feet may also be affected by syllable ‘weight’ and underlying marking (‘lexical stress’) (see Kager 2007 for an overview and references). ‘Syllable weight’ almost always refers to the structure of the syllable rime – whether it contains a long vowel, diphthong, or (particular type of) coda consonant (see de Lacy 2007 for an overview). Moraic theory identifies syllable weight as moraic content – feet can place strict requirements on moraic content, which can force them to move away from edges, or change shape (Hyman 1985; Hayes 1989, 1995). A recurring question in metrical theory has been whether there are factors apart from edges and morae that can directly affect metrical feet by forcing deviation in their default location, shape, or the position of their head and non-head syllables. This issue was raised very early in metrical theory research: Prince (1983: 71ff) argued that in Passamaquoddy “a syllable containing a full vowel is prosodically heavy; a syllable containing a schwa is light.” This analysis preceded moraic theory, so at this point ‘weight’ was seen as a metrical property of some kind. Crucially, in this analysis, vowel quality can affect heaviness. Halle & Vergnaud (1987)’s theory also permits segmental properties to directly influence metrical structure. For example, Halle & Vergnaud (1987)’s analysis of Eastern Cheremis invokes a rule “Assign line 1 asterisks to full vowels” (p. 51); this rule effectively distinguishes ‘full’ vowels from ‘reduced’ (schwa-like) vowels, and allows the former to be visible to grid formation. 12 CatJL 18, 2019 Shu-hao Shih; Paul de Lacy

However, to be clear, metrical theories did not focus on non-moraic weight factors. In some overviews of metrical stress theory, there is no significant mention of them (e.g. Kager 1995, 2006), and Hayes (1995: 271) called them “a residue of cases”, of minor importance. This outlook may explain why the treatment of such factors was rather unconstrained. For example, Halle & Vergnaud (1987)’s approach does not apparently limit the factors that can affect gridmark projection: it is possible that there could be a rule such as “Assign line 1 asterisks to [round] vowels” and “Assign line 1 asterisks only to schwas”, both argued to be unattested (e.g. de Lacy 2004). Hayes (1995: ch. 7) was a major step towards restricting the theory of non- weight metrical influences. Hayes (1995: 272ff)’s theory involved rules that allowed specific non-weight factors to project marks onto a ‘prominence grid’, which can then influence stress placement. However, unlike Halle & Vergnaud (1987), Hayes (1995) placed restrictions on how non-quantity factors influenced metrical structure: “stress rules have a choice: […] to refer to a simple criterion of syllable weight (i.e. quantity, under moraic theory) … [or] to employ a rather unconstrained criterion of syllable weight (i.e. prominence…) and have access only to a more impoverished inventory of stress assignment devices.” (p. 273). In Hayes (1995: 275)’s sample analysis, prominence only affects foot structure if it respects the form of already- established feet, thus limiting the influence of prominence on foot form. At the same time as metrical theories were grappling with non-quantity influences on metrical feet, there were two other strands of relevant phonological research. One set of theories involved ‘defective schwa’: the idea that schwa – or perhaps central vowels more generally – are phonologically different from other vowels. Exactly how they are different depended on the theory – e.g. lack of features, or lack of a mora, or some other defective syllabic constituency (see van Oostendorp 2000 for an overview). Such defective representations could cause schwa to be unfootable, or cause heads to avoid syllables with schwas. Such theories are interestingly different from Halle & Vergnaud (1987) and Hayes (1995)’s approaches: there is no direct phonological mechanism that prohibits feet or heads from dominating a schwa; instead, schwas have a defective structure (e.g. lack a mora), foot-formation processes are sensitive to that lack of structure, and so feet seem to treat schwas differently from other vowels. Such sensitivity can lead to inadvertent sonority-driven stress. For example, if schwas cannot bear a mora in a system, then they cannot head a foot, and so cannot be stressed. We will return to this important distinction of direct vs. indirect sonority-sensitivity in section 3.2, and argue that there is indeed indirect sonority-sensitivity in metrical systems, but no direct sensitivity. The other strand of relevant research was ‘degenerate syllable’ theories. In such theories, syllables are permitted to lack morae in certain circumstances (see Lin 1998; Shih 2018b for overviews). Usually, schwa (or some other reduced vowel) is unable to bear its own mora. While degenerate syllables focused on the syllable and the effects of lack of moraicity on syllable-level phonotactics and processes, the connection to ‘degenerate schwa’ and metrical prominence theories like Hayes (1995)’s is clear: certain vowels – perhaps just schwa – have phonological proper- Evidence for Sonority-Driven Stress CatJL 18, 2019 13 ties that make them invisible or undesirable to feet. The effect is sonority-driven stress, albeit indirectly caused. Returning to research on prominence, a significant problem with Halle & Vergnaud (1987) and Hayes (1995)’s theories was that they did not provide clear restrictions on which non-quantity factors could influence metrical structure. Halle & Vergnaud (1987) did not provide any limits. Hayes (1995: 276) list several factors: “heavy syllable quantity, lowness in vowels, high tone, the presence of syllable-final /ʔ/, and the presence or voicing of syllable-initial consonants”, and suggests a more general principle to identify such factors: perceptual salience, or “raw prominence”, or factors that make syllables “sound louder” (Hayes 1995: 271). However, it is not clear that this list was intended to be exhaustive, and exactly how it related to phonological representation and computation is not immediately clear. Hayes (1995: 271) even entertains (though rejects) the “pessimistic conclusion” that “there are no linguistic universals of interest in this area”. Hayes (1995: 276) then suggests that any dimension that influences prominence is universally consistent in the ‘direction’ of influence: i.e. greater prominence is associated with lower vowels, so no language should show greater prominence of higher vowels. Kenstowicz (1997) presented a significant refinement of the theory of non- quantity influences on metrical stress. One of Kenstowicz (1997)’s major proposals was to separate Hayes (1995)’s non-quantity influences into different categories. Specifically, Kenstowicz (1997) isolated segmental sonority as a property that influenced metrical structure, distinct from other factors. Later on, de Lacy (2002b) added tone as another property by arguing that there were distinct, specific mechanisms that regulated tone and metrical structure. Kenstowicz (1997) further provided a restrictive theory of how metrical structure and sonority interact: there are a family of Optimality Theoretic constraints in a universally fixed ranking with the general form *PeakFoot/sonority_level and *MarginFoot/sonority_level. In other words, Kenstowicz (1997) proposed a phonological mechanism that involved direct influence of sonority values on prosodic structure – both foot heads and non-heads. De Lacy (2002a, 2004, 2006) then presented a theory that built on Kenstowicz (1997). It included a more detailed sonority hierarchy, which placed schwa on a level between peripheral high vowels (e.g. [i u]) and high central vowels (e.g. [ɨ ʉ]), given in (2). The symbol ‘ ’ means “is less sonorous than”.

(2) Sonority hierarchy⟨ from de Lacy (2002a) (a) Consonant sonority voiceless voiced voiceles voiced nasals liquids glides … stops stops fricatives fricatives (b) Vowel sonority high⟨ ⟨ mid ⟨ high ⟨ ⟨ mid ⟨ ⟨low … central central peripheral peripheral peripheral vowels vowels vowels vowels vowels ɨ ⟨ ə ⟨ i•u ⟨ e•o ⟨ a 14 CatJL 18, 2019 Shu-hao Shih; Paul de Lacy

De Lacy (2002a, 2004, 2006)’s theory also claimed that the relation between feet and sonority was a smaller part of a much larger theory of the relation between prosody and non-structural factors. The foot-sonority constraints were a part of the family of constraints *pHead/p and *pNon-head/p, where pHead is the head element of the prosodic element p, and pNon-head is any non-head element of p (more specifically, de Lacy (2002a, 2004, 2006)’s theory refers to Designated Terminal Elements – a distinction that is not significant for the present discussion). In these constraints, p ranges over any ‘prosodic’ factor – which includes sonority and tone, but crucially not subsegmental features (e.g. place of articulation, [voice]). De Lacy (2002a, 2004, 2006)’s theory provides further restrictions on exactly how the constraints can refer to p properties. For sonority, the constraints penalize lower sonority heads over higher sonority ones, and higher sonority non-heads over lower sonority ones. For example, the constraints on feet and sonority are given in (3). Each constraint mentions a prosodic category – the head syllable of d non d a foot (H Ft) or the non-head syllable of a foot ( -H Ft). Each constraint also mentions sonority categories; for example, ɨ is ‘high central vowels, and i•u is ‘high peripheral vowels’ (see (2) above for the other categories) – note that we use ‘•’ here to separate members of the same sonority category, not as an operator of any kind. The constraint *HdFt/ɨ,ə,i•u is violated when a head syllable of a foot contains a vowel that is as sonorous as the ‘high peripheral vowels’ category, or less sonorous.

(3) Foot-sonority constraints (de Lacy 2004: 3)

(a) *HdFt/ɨ (b) *Non-HdFt/a *HdFt/ɨ,ə *Non-HdFt/a,e•o *HdFt/ɨ,ə,i•u *Non-HdFt/a,e•o,i•u *HdFt/ɨ,ə,i•u,e•o *Non-HdFt/a,e•o,i•u,ə *HdFt/ɨ,ə,i•u,e•o,a *Non-HdFt/a,e•o,i•u,ə,ɨ

De Lacy (2002a, 2004, 2006) was at pains to point out that non-quantity factors could influence feet in a far less restricted way than in Hayes (1995)’s proposals. For Hayes (1995), prominence was essentially a secondary factor and foot conditions always took precedence. For de Lacy (2002a, 2004, 2006), sonority could affect feet just like morae: foot-sonority constraints could force feet to change their head position, size, and location. So, returning to the Gujarati system in (1), de Lacy (2006: 238)’s analysis can be illustrated below. Tableau (4) shows how feet retract from the default rightmost position in so the foot head can fall on a higher sonority vowel. Candidate (b) has default footing, but by doing so violates the constraint on foot heads (*HdFt/ ɨ,ə,i•u,e•o) because the head of the foot contains an [i]. In contrast, candidate (a) does not violate the head-sonority constraint, but at the expense of aligning the foot’s right edge with the right edge of the word (alignFtR). In this way, the head- sonority constraint forces feet to deviate from their default position. Evidence for Sonority-Driven Stress CatJL 18, 2019 15

(4) Gujarati i: Foot movement

/manito/ ‘favorite’ *HdFt/ɨ,ə,i•u,e•o trochee FtBin alignFtR

☞ (a) (ˈma.ni)to * (b) ma(ˈni.to) *! (c) (ˈma)ni.to *! * *

Tableau (5) shows another response to the head-sonority constraints. Candidate (b), with the default foot position, fatally violates the head-sonority constraint; so does candidate (c), where the foot has retracted to the left edge. Candidate (d) solves the problem by using an iambic foot, but fatally violates trochee, which requires left-headed feet. The solution – in (a) – is to have a degenerate (monomoraic) foot over the final syllable.

(5) Gujarati ii: Foot reduction

/hoʃijaɾ/ ‘clever’ *HdFt/ɨ,ə,i•u,e•o trochee FtBin alignFtR

☞ (a) ho.ʃi(ˈjaɾ) * * (b) ho(ˈʃi.jaɾ) *! (c) (ˈho.ʃi)jaɾ *! * (d) ho(ʃi.ˈjaɾ) *!

The Gujarati analysis above shows how the head-sonority constraints can force deviation from both the default foot position (rightmost, here), and foot size (i.e. from binary to degenerate). The idea that sonority (and tone) can directly influence prosodic structure – both its location and size – has gained wider acceptance since the early 2000s (see e.g. McGarrity 2003; Crowhurst & Michael 2005; Carpenter 2006). We summarized the history and theories of sonority-driven stress above for two reasons. One is to emphasize that non-moraic influences on stress have been the subject of research for a long time, and the theories that have developed are now well defined and extensively explored. The second reason was to allow us define the present goal. We are not directly concerned with any stress system where sonority seems to be a factor in the position of size of feet, for, as we have seen above, such systems can be produced by both phonologically direct and indirect means. Our focus is instead on phonological mechanisms that directly cause deviation from default metrical position and shape in order to form a different metrical-sonority profile; we call this direct mechanism ‘true sonority-driven stress’. We are also not directly interested in the question “Are there constraints of the form *p/p?” OT constraints are not mechanisms (or ‘processes’); a process emerges 16 CatJL 18, 2019 Shu-hao Shih; Paul de Lacy from the interaction of constraints. For example, the head-sonority constraints can also be used to force metathesis, vowel reduction, and deletion (de Lacy 2007). Our sole concern here is whether there is sonority-motivated deviation from foot position and form that is caused by a phonological mechanism that directly relates sonority and feet. Of course, exactly how that mechanism is formally expressed in a particular Generative theory is worthy of examination, and the implications of the lack of true sonority-driven stress can have wide-ranging implications for particular theories – for reasons of space, we must defer exploring those implications here (though see Shih 2018b§2.6.4 for discussion). As a final comment, we wish to mention the evidence for metrical structure, as presented in the research mentioned above (also see de Lacy 2014 for an overview). Evidence for metrical structure comes from phonological interaction and phonetic realization. For ‘phonological interaction’, phonological and morpho-phonological processes can refer to metrical structure in their conditioning environments, such as minimal word restrictions, fortition, infixation, and so on; an example is provided for Nganasan in section 3.1 (see de Lacy 2009, 2014 for extended discussion). For ‘phonetic realization’, cross-linguistic studies have shown that multiple acoustic measures may correlate with stress in vowels. Typically, stressed vowels may have a higher pitch (e.g. Lieberman 1960; Gordon 2004; Gordon & Applebaum 2010), greater intensity (e.g. Fry 1955; Lieberman 1960; Gordon 2004; Gordon & Applebaum 2010; Gordon & Nafi 2012), and longer duration (e.g. Fry 1955; Lieberman 1960; Gordon 2004; Gordon & Applebaum 2010). Differences in F1 and F2, associated with difference in vowel quality, have also been found (e.g. Gordon 2004; Garellek & White 2012). Individuals may differ as to whether they realize foot heads with all or only some of these acoustic factors. Experimental work has included measurements of all or some of these properties; the experiments we cite below examined them all. Impressionistic work is often less clear about what acoustic property is being interpreted as stress; often the location of stress is asserted without mentioning how it was perceived (e.g. Scorza 1985). So, is there solid evidence for a theory that claims there is a phonological mechanism that directly relates sonority to foot structure, thereby causing foot retraction and degeneration?

3. Evidence We cannot demonstrate that sonority-driven stress does not exist; we can only show that there is currently no adequate evidence for it. We divide the issue into two parts here. One involves ‘reliability’: a technical term that refers to whether a measure- ment device produces consistent results. Section 3.1 argues that the majority of evidence for sonority-driven stress is either produced by devices that are either unreliable, or whose reliability cannot be determined. The other part, in section 3.2, is about the causes of apparent sonority-sensitivity. We argue that there are several such causes, and so the need for direct mechanisms involving influence of feet by sonority have not been conclusively demonstrated. Evidence for Sonority-Driven Stress CatJL 18, 2019 17

3.1. Reliability We use the term ‘reliability’ here in its technical sense: a method is reliable if it can be applied repeatedly to the same data and produce the same result (e.g. Ray 2009). For sonority-driven stress, the vast majority of evidence comes from impressionistic descriptions, reported in grammars and journal articles (see Shih 2018b: ch. 5). ‘Impressionistic descriptions’ are here considered to be any description that involves a human perceiving and reporting the position of stress, unaided by mechanical analytical devices (e.g. spectrographs, pitch trackers). For example, de Lacy (2002a)’s description of Gujarati stress involved a human – de Lacy – listening to and transcribing the speech of an individual, and marking where de Lacy believed stress fell. De Lacy also recorded the speech and later listened to the recordings, and made changes to the transcriptions if he believed he had erred in his initial transcription. de Lacy did not analyze the recordings using any acoustic analysis devices. So, de Lacy (2002) is the report of de Lacy’s perceptual impressions of where stress fell – i.e. an ‘impressionistic’ description. We note that the large majority of descriptions we have consulted do not state outright that they are impressionistic; we infer that they are due to the lack of information about the methodology used to obtain data, and that the impressionistic approach is the dominant method taught and used in linguistic field research (see e.g. Cheliah & de Reuse 2011§8.3.2; c.f. Ladefoged 2003). So, for sonority-driven stress, are impressionistic descriptions of sonority- driven stress reliable? In this section, we focus on the reliability of impressionistic descriptions specifically for those cases involving peripheral vowel distinctions.

3.1.1. Au We can address the question of reliability (with caveats discussed below) by examining cases where there have been multiple descriptions of the same sonority-driven stress system. An instructive case is Au (Torricelli > Wapei- Palei > Wapei; Sandaun Province, Papua New Guinea). A series of articles and books – Scorza (1973, 1976, 1985, 1992) – provide descriptions of the stress system. Scorza (1973)’s description is as follows:

(6) Au stress (Scorza 1973) “[Stress] usually occurs on [the] third syllable of a four syllable word, the first syllable on [sic] a three syllable word, and alternates from syllable to syllable on two syllable words, that is from initial to final depending on the word. Stress is recognized by loudness and high pitch.”

Scorza (1973)’s description does not admit of a simple characterization in terms of standard metrical theory. It perhaps involves building trochees from left to right, with the second syllable being the head (producing [(ˈσσ)σ] and [(ˌσσ)(ˈσσ)]), with lexical stress in disyllables. It certainly does not involve sonority. However, Scorza (1976, 1985) presents a different description: “Stress normally occurs on the first syllable of a word” (contrary to Scorza (1973)’s claim that it falls 18 CatJL 18, 2019 Shu-hao Shih; Paul de Lacy on the third syllable in 4-syllable words, and variably on disyllables). Additionally, Scorza (1976, 1985) claims the following:

(7) Au stress (Scorza 1976, 1985) (a) The vowel /ɨ/ is never stressed unless the only vowels in the word are /ʌ/ or another /ɨ/ (e.g. [k-ɨˈsawɨn] ‘he-hides’; [hɨˈhi] ‘ironwood’; [ˈmɨtɨk] ‘man’) (b) The vowel /ʌ/ is never stressed unless the only other vowels are /ɨ/ or another /ʌ/ (e.g. [kʌˈwat] ‘he-gives’, [ˈnʌpʌɾʌ] ‘dog’)

The additional clauses from Scorza (1985) mean that Au has a sonority-sensitive system. It is possible that the non-peripheral vowels /ʌ/ and /ɨ/ are the two least sonorous vowels – the others are the peripheral /i u o a aː/ (Scorza 1985: 219). So, Scorza (1985) seems to describe a classic sonority-driven stress system: stress falls initially, unless the initial syllable contains a low sonority vowel: e.g. [kɨˈsawɨn], [kʌˈwat]. Finally, Scorza (1992) contains another description: “Stress usually occurs on the first syllable of the word, but may change under certain morphophonemic considerations.”1 Scorza (1973), Scorza (1976), Scorza (1985), and Scorza (1992) do not mention machine acoustic analysis, suggesting that all of the descriptions were impressionistic, from either direct fieldwork or recordings made in the field (Scorza 1973: 165- 166). It is likely that the stress analyses in Scorza (1973) and Scorza (1976, 1985) involved consultation of the same data as the same fieldwork dates and consultants are mentioned (Scorza 1973: 165-6 cf. Scorza 1976: 5 cf. Scorza 1985: 215). In other words, the same person analyzed the same data twice. This homogeneity is important because both the device (Scorza himself) and the data were kept constant, so allowing us to ask whether the device is reliable. It is important to emphasize that ‘reliable’ here is a technical term, and does not carry any emotional weight or judgement with it. It simply refers to whether a particular device and methodology produces the same result given the same data; the device just happens to be a human here, but we could – and should – ask the same of any device used in phonological analysis (e.g. microphones, algorithms, software). In this sense, it is clear that the device – i.e. Scorza – is not reliable. The descriptions reported in Scorza (1973) and Scorza (1976, 1985) are significantly different, even though the data was the same. Not only is there difference in whether Au is sonority-sensitive or not, but there are differences in where stress falls both on four syllable words (Scorza 1973: the third syllable; Scorza 1976, 1985: the first syllable), and where stress falls on disyllables (Scorza 1973: either syllable; Scorza 1976, 1985: the first syllable). It is unclear whether Scorza (1992)

1. Scorza’s later works do not mention the fact that earlier works differed in their stress descriptions: e.g. Scorza (1985)’s stress section does not mention Scorza (1973)’s stress description, even though Scorza (1985) includes Scorza (1973) in its bibliography. It is possible that the lack of mention is merely due to the limited space given to discussion of stress. Evidence for Sonority-Driven Stress CatJL 18, 2019 19 is a further independent description by Scorza, or an incomplete summary of previous descriptions; if it is a new description, it presents yet a different description, ignoring the sonority effects in Scorza (1976, 1985). There is an obvious problem with the reasoning presented above. The assumption was that the device – i.e. Scorza – was the same in 1973 as in 1976, 1985, and 1992. However, there may have been profound changes in Scorza’s knowledge and skill over that time, and perhaps even physical changes relevant to perception. Unfortunately, there are few clues in the publications to be sure. It is possible that the device was reliable (and accurate, and adequately precise) at one point in time, but not at others. However, this consideration raises a profound problem when relying on impressionistic descriptions: how can we ever determine reliability if we can ascribe variation in description to change in the describer, or differences between devices (e.g. people)? This problem is particularly acute because impressionistic descriptions typically do not describe the methods used. For example, none of the Au sources cited above give details about the type of elicitation methods used, the elicitation environment, speech-relevant properties of the experimental subjects, and –crucially – about the nature of the perceptual device itself – i.e. Scorza. In many cases, it was also not possible to see how the author arrived at their generalizations because either no data was presented to support the claim, or the data presented did not cover all possible word lengths and syllable shapes. For example, Scorza (1973) provides stress generalizations but no data annotated for stress in any part of the article, which is 22 pages long. Scorza (1985) provides a total of 10 words annotated for stress in 273 pages (all on p. 219). Of these 10 words, only three show stress on a non-default position: [kɨˈsawɨn] ‘he-hides’, [k-ʌˈwat] ‘he-gives’, and [k-ʌˈkintɨp] ‘he-steals’ (Scorza 1976 provides an additional relevant word: [hɨˈhi] ‘ironwood’). In other words, the evidence for the claim that Au has sonority-driven stress is three words. We do not know how representative these words are of words of this form (at worst, they may be suppletive forms). Strictly speaking, they only show that [a] and [i] attract stress away from [ɨ] and [ʌ] – no data is provided that shows the other vowels [u o aː] attracting stress away from these central vowels. There is also no data that shows the relationship between [ɨ] and [ʌ]: while [ˈmɨtʌ] ‘woman’ indicates that [ʌ] does not attract stress away from [ɨ], no word is provided that shows that [ʌ] fails to attract stress away from [ɨ] (i.e. no data shows whether Au treats [ʌ] as more sonorous or the same in sonority as [ɨ]). A final point is that the three words are all verbs, and all have the third person singular masculine prefix; it is possible that part of speech, morphological structure, or particular affixes affects stress, so one might view the examples as being inadequately morphologically diverse. Of course, this point about data broaches a much broader issue of how to identify adequate data in support of generalizations – a topic that takes us too far afield (though see de Lacy 2009, 2014 for relevant discussion). The descriptions of Au stress at least force us to the conclusion that not all impressionistic descriptions are accurate. Some might be accurate, but the variation in the Au descriptions indicate that some are not. Importantly, we have no straightforward way of telling the two apart – i.e. was Scorza (1973) right, or Scorza (1976, 20 CatJL 18, 2019 Shu-hao Shih; Paul de Lacy

1985), or Scorza (1992)? To put it another way, based on the descriptions we have available to us, does Au have sonority-driven stress or not? There is no reliable evidence either way.

3.1.2. Gujarati Au is a remarkable case because the same person described it at different times. It is far more usual for different people to describe the same case over a period of time. So, we can ask the question of reliability in such multi-device contexts, too. The most extensively described sonority-driven stress case of which we are aware is Gujarati (Indo-European>Western Intermediate Indo-Aryan; spoken in Gujarati province, India and elsewhere): thirteen descriptions of Gujarati stress are known to us, including Shih (2018a). It is possible that all of the descriptions are independent – few of them mention the other descriptions, and when they do (e.g. de Lacy 2002a) it is to point out differences rather than similarities. If humans are broadly reliable devices in regard to the impressionistic description of sonority- driven stress, we would expect a fairly strong homogeneity in the descriptions. Instead, there is remarkable divergence on almost all metrical dimensions. Focusing on just the influence of sonority on stress, Turner (1921) and Master (1925) (for disyllables) claim that Gujarati stress falls on a fixed position – there is no sensitivity to sonority. In contrast, Cardona (1965) claims that [a] attracts stress away from other vowels, and both [i] and schwa repel it onto more sonorous vowels. Mistry (1997), de Lacy (2002a) and Cardona & Suthar (2003) claim that – for disyllables – stress does not avoid schwa (cf. Cardona 1965), while de Lacy (2002a) also claims that stress does not avoid [i]. While several descriptions say that [a] attracts stress away from other vowels, Adenwala (1965) groups [a] with the mid vowels, and Campbell & King (2011) group [a] with all of the peripheral vowels. Table 1 summarizes the role of sonority in the descriptions, grouping vowels according to how the stress system purportedly favors them in the attraction of stress. Table 1 does not do justice to the diversity of the descriptions. There are also disagreements over whether syllable shape affects stress (e.g. Masica 1991 vs. Cardona & Suthar 2003). Other descriptions of syllable shape mainly focus on [ə]: for some descriptions, when [ə] occurs in a closed syllable, it does not repel stress (Cardona 1965; Doctor 2004; Schiering & van der Hulst 2010). Interestingly, outside theoretical works (de Lacy 2002, Shih 2018a,b, Bowers 2019), there is very little explicit discussion – or even acknowledgement – of the diversity and contradictions in the Gujarati stress descriptions. For example, in the most recent descriptive work, Modi (2013§4.7), no previous description is mentioned. We are not sure what to conclude from this lack of explicit discussion except to note that stress is not unique – each descriptive work we examined had very little discussion of other descriptive works for most of the topics they covered (other than mentioning previous research in their bibliographies). We do not know whether it is standard practice for descriptive works to avoid explicitly mentioning previous work as we know of some cases to the contrary (e.g. Dunn 1999’s grammar of Chukchi, which discusses previous research in detail). Evidence for Sonority-Driven Stress CatJL 18, 2019 21

Table 1. Sonority distinctions in descriptions of Gujarati stress Source Categories Penultimate stress Turner 1921 a ɛ ɔ e o u i ə Master 1925: 2s Sonority-driven stress Mistry 1997 de Lacy 2002a: 2s a ɛ ɔ e o u i ə Cardona & Suthar 2003: 2s & 3s Cardona 1965: 2s de Lacy 2002a: 3s a ɛ ɔ e o u i ə Doctor 2004: 2s Schiering & van der Hulst 2010: 2s & 3s Cardona 1965: 3s a ɛ ɔ e o u i ə Doctor 2004: 3s Adenwala 1965: 2s a ɛ ɔ e o u i ə Campbell & King 2011 a ɛ ɔ e o u i ə Modi 2013: 2s

At this point, we are faced with a similar question to that for Au: Does Gujarati have sonority-driven stress or not? Of course, there are profound problems in assuming that we can compare the various Gujarati stress descriptions. For one thing, it is not clear that any of the descriptions were describing the same peoples’ speech; it is possible that the variation in descriptions is due to different dialects (although Cardona 1965 and de Lacy 2002a both specifically mention the Ahmedabad dialect). It is also possible (and in fact almost certain) that the dialects have changed over time, from Turner (1921) to Modi (2013). As with Au, it is also not clear that every description was produced by an equally accurate device – the descriptions do not provide details as to how they arrived at their descriptions, and so it is impossible to be sure that equivalent methods were used. However, raising such questions broaches a profoundly worrying issue: if invoking dialect differences, change over time, and variation of accuracy in linguistic devices allow us to put concerns about reliability aside, then we can never ask about the reliability of impressionistic descriptions. If we cannot be sure that our devices are reliable then they are effectively worthless for scientific research. We conclude by mentioning that there are two recent studies of Gujarati stress that are not impressionistic – Shih (2018a,b) and Bowers (2019). Shih (2018a,b) reports the result of the acoustic analysis of a production experiment. Intensity, F0, F1, F2, and duration were measured to see if there were any differences in disyllables with the form [CVCa] vs. [CaCV] (where V is not [a]). According to most of the impressionistic descriptions of Gujarati, [CVCa] should have final stress while [CaCV] should have initial stress. However, Shih (2018a,b) found no evidence that 22 CatJL 18, 2019 Shu-hao Shih; Paul de Lacy stress was any different in these words: both words were consistent with having penultimate stress. We repeat the results for vowel quality (F1 and F2) from Shih (2018b) in Figure 1 below. If [a] attracts stress away from non-[a] vowels, and vowel quality reflects stress (as claimed by Cardona 1965, de Lacy 2002a), we would expect the quality of [a] to be the same in [CáCa] (●), [CáCV] (▲), and [CVCá] (⌷) because they should all contain stressed [á]s, in contrast to [CáCa] (○), which contains an unstressed [a]. However, the results group the [a]s based on their position, not their stress: [a] in initial syllables (● and ▲) has a distinct F1 from [a] in final syllables (○ and △). If stress has anything to do with F1, the results above indicate that the penult is always stressed, regardless of where [a] is. We refer the reader to Shih (2018a) for detailed statistical models that demonstrate the statistical significance of the assertions made above. In other words, comparison of impressionistic descriptions with Shih (2018a,b)’s experiment and acoustic analysis shows significant disagreement in terms of the stress-attracting powers of [a] in Gujarati. Bowers (2019) also presents an acoustic analysis of production experiments, independent from Shih (2018a)’s research. Bowers (2019) concluded that F0, F1/F2 and duration indicate that there is a difference between initial and non-initial syllables, but the results “failed to find clear acoustic correlates of sonority-driven stress as described by de Lacy (2002; 2006)” (p. 25). If acoustic analyses of sonority-driven are presumptively assumed to be more reliable than impressionistic descriptions, then it is potentially a cause for concern that so many descriptions reported that [a] attracted stress away from lower sonority vowels. We discuss the issue of how impressionistic approaches could produce errors below.

Figure 1. Gujarati vowels in post-pausal context. ●[CáCa]; ▲[CáCV]; [CVCá]; vs. ○ [CáCa]

△ Evidence for Sonority-Driven Stress CatJL 18, 2019 23

3.1.3. Nganasan Like Gujarati, there are multiple descriptions of Nganasan stress: Helimski (1998, personal communication), Castrén (1854), Prokofjev (1937), Hajdú (1964), Tereshchenko (1979), Lublinskaya et al. (2000), de Lacy (2002a), and Vaysman (2009) (Uralic > Samoyed; Siberia, Russia). De Lacy (2004) describes an elaborate sonority-driven stress system. The description was based on previous descriptions and on the author’s impressions of audio recordings: default footing involves a right-aligned quantity-sensitive trochee: e.g. [car(ˈkiː)] ‘worn out’, [hu(ˈtaruʔ)] ‘of the houses’. However, stress retracts to the antepenultimate syllable if it contains one of [a e o] (i.e. high and mid peripheral vowels) and the penult contains one of [i y u ə ɨ] (i.e. high peripheral vowels and central vowels). It is worth noting that Tereshchenko (1979)’s description does not mesh perfectly with de Lacy (2004)’s. Tereshchenko (1979: 41) states that stress on trisyllabic words is unpredictable, though de Lacy (2004) claims that most of Tereshchenko’s data fits with de Lacy (2004)’s description. Helimski (1998)’s description also does not clearly match de Lacy (2004)’s: Helimski (1998) cites [koˈruðə] ‘houses’, not *[ˈkoruðə] as expected in de Lacy (2004)’s analysis. Helimski (1998)’s description also diverges from de Lacy (2004)’s in other important ways involving optionality and syllable shape: “[Default stress] is optionally violated by the retraction of stress from a high vowel or ə to the vowel (usually an open one) in the preceding syllable: barusji ~ barusji ‘devil’) (Helimski 1998: 486). The most recent description of Nganasan (also based on the Avam dialect) is Vaysman (2009)’s. Vaysman (2009)’s and de Lacy (2004)’s descriptions agree about the default position of feet. However, there is a crucial difference: de Lacy (2004) reports that feet retract from a penultimate [i y u ə ɨ] onto a preceding low or mid peripheral vowel, while for Vaysman (2009) it only retracts from a central vowel [ə ɨ] (Vaysman 2009§2.2.1.2). So, while ‘salmon’ is [ba(ˈkunu)] for Vaysman (2009: 28), it would be [(ˈbaku)nu] for de Lacy (2004); similarly ‘clay (locative)’ is [sa(ˈðutə)nu] for Vaysman (2009), but is predicted to be [saðu(ˈtənu)] for de Lacy (2004). The apparently minor disagreement between Vaysman (2009) and de Lacy (2004) has an outsized importance. De Lacy (2004) presents Nganasan as crucial evidence for the theory of stringent sonority-foot constraints (all other sonority-driven stress systems can be accounted for using constraints with fixed ranking, as shown in Kenstowicz 1997). With Vaysman (2009)’s description, Nganasan potentially loses its power to make the theoretical distinction claimed by de Lacy (2004). Apart from the theoretical significance of Nganasan, we are left with a disagreement of impressionistic descriptions, with apparently no way to resolve them. However, Vaysman (2009) does provide an additional relevant diagnostic. As de Lacy (2008) discusses, metrical evidence can come from many sources, including metrically-sensitive processes like fortition. Relevantly, Vaysman (2009) describes a metrically-sensitive consonant gradation process. However, Vaysman (2009: 21) notes: “the foot structure that is marked by gradation does not match the stress pattern, namely the placement of the primary stress and its shifts from [ə] and [ɨ] leftwards.” (emphasis from original). Consider the 2nd person dual suffix -ti/-ði 24 CatJL 18, 2019 Shu-hao Shih; Paul de Lacy in (8). The 2nd person dual suffix is [ti] when foot-internal, and [ði] foot-initially (a specific example of a more general consonant gradation process). This analysis works only if trochaic feet are constructed from left to right, as the footing indicates. However, the default primary stress falls on the penultimate syllable. This places the primary stressed syllable at odds with the trochaic foot parse needed for consonant gradation, as shown in (8a,b vs. c,d).

(8) Consonant gradation in Nganasan (Vaysman 2009: 45-46) (a) [(ˈko-ti)] ‘your (du.) ear’ (b) [(ˌbaku)(ˈnu-ti)] ‘your (du.) salmon’ (c) [(haˈhi)-(ði)] ‘your (du.) wild deer’ (d) [(ˌkərï)(gəˈʎi)-(ði)] ‘your (du.) march’

In other words, the evidence for metrical structure from phonological processes does not support either de Lacy (2004)’s or Vaysman (2009)’s impressionistic claims about primary stress location. In short, Nganasan presents phonological evidence that does not support the claims about primary-stress placement. While it is possible that Nganasan has dual disjoint metrical tiers (e.g. Parker 1998), at the very least consonant gradation provides no support for the claims about sonority-driven primary stress, and potentially contradicts it. What we see for Nganasan, then, is not only disagreement in impressionistic descriptions of stress, but conflicting results using different methodologies – impressionistic description, and metrically-sensitive allophony.

3.1.4. Armenian Haghverdi (2016) reports the results of acoustic analysis of a production experiment on Eastern Armenian stress (Indo-European > Armenian; spoken in Armenia and other locations). Unlike Gujarati and Nganasan, there is unanimous agreement among impressionistic descriptions that stress falls on the final syllable unless it is a schwa, in which case it falls on the penult (which is never a schwa when the final syllable is also a schwa) (Khachatryan 1988, Vaux 1998, Sakayan 2007, Dum- Tragut 2009). Thanks to a rich inventory of suffixes, stress can be usually observed to advance rightwards in a word, as long as the final syllable is not schwa. Compare the forms and their stress in (9), all of which have the same root.

(9) Eastern Armenian stress (Haghverdi 2018: 1) [ɑ.ˈnuʃ] ‘sweet’ [ɑ.ˈnu.ʃ-ə] ‘the sweet’ [ɑ.ˈnu.ʃ-ət] ‘your sweet’ [ɑ.nu.ˈʃ-i.t͡ sʰ-ət] ‘from your sweet’ Evidence for Sonority-Driven Stress CatJL 18, 2019 25

Armenian stress is apparently quite different from Nganasan, Au, and Gujarati because it involves several impressionistic descriptions that all agree with each other. It seems that – at least for Armenian sonority-driven stress – the detection devices (i.e. humans) are reliable. However, Haghverdi (2016)’s acoustic analysis results differ from the impressionistic descriptions’. Haghverdi (2016) found no meaningful difference between final full vowels and schwa in terms of duration, intensity, and quality. In fact, there were no significant differences between full vowels regardless of whether they were in putative stress or non-stress position. The only significant difference was in F0, with Haghverdi (2016)’s result repeated in Figure 2. The Figure shows the results of measuring, then normalizing, F0 and duration in words with the shape CVCVC (where V is not schwa) compared with CVCəC words, in both focused and non-focused contexts. In Eastern Armenian, CVCVC words should have final stress and CVCəC words should have initial/penultimate stress. Pitch is commonly claimed to peak over the stressed vowel, so we would expect CVCəC to have an early F0 peak, while CVCVC should have a late F0 peak. However, all of the forms in Haghverdi (2016)’s results have an F0 peak over the final vowel, and the F0 peak is even at the same normalized level. In other words, if stress is marked by the F0 peak, all words have final stress. What could have caused impressionistic descriptions to claim that schwa repels stress? There are two possibilities. One is that the describers were familiar with languages that signaled stress differences through vowel reduction (e.g. many dialects of English). As ‘schwa’ signaled lack of stress in the describers’ L1, it is possible that this led them to classify Armenian schwa as ‘stressless’. Another option relates to F0. If the describers were familiar with languages where F0 peaks over word- or phrase- stressed syllables (e.g. English, again), the F0 results seen in Figure 2 might provide some explanation. Notice that F0 over a penultimate vowel preceding a schwa (in red) is higher than when a full vowel precedes a full vowel (in blue). The reason for this F0 difference is incidental: schwa is relatively shorter than other vowels, and so the pitch peak in [CVCəC] words is closer to the initial V than in

Figure 2. Eastern Armenian normalized F0 contours (adapted from Haghverdi 2016, with F0 interpolation removed). 26 CatJL 18, 2019 Shu-hao Shih; Paul de Lacy

[CVCVC] words. Consequently, the interpolated F0 is higher over the first vowel in [CVCəC] words. However, this incidental difference in F0 – perhaps coupled with the association of stress with stresslessness – could have led to the misperception of stress over full vowels in [CVCəC] words. The issue that Haghverdi (2016)’s study broaches is that of ‘cross-method’ reliability: do impressionistic methods get the same results as production experiments with acoustic analysis? The answer – from both Eastern Armenian and Gujarati (and Munster Irish, below) – is no. If we assume that acoustic analysis of controlled experiments is likely to be more reliable than impressionistic methods, then we are at least left with uncertainty for Eastern Armenian. In other words, is there clear evidence that Eastern Armenian involves sonority-driven stress? Even though the impressionistic descriptions all agree, the fact that their results disagree with those of a potentially more reliable methodology cast the issue into doubt.

3.1.5. Reliability What about the reliability of descriptions of other sonority-driven stress systems? We list all the sonority-driven stress systems we know that involve peripheral vowel distinctions in Table 2 below. We judged a case as ‘sonority-driven stress’ if stress was reported to appear on a non-default syllable due to its having a particular vowel quality or sonority. For example, Ross (2009: 762) reports that the default stress position in Takia is the final syllable, as in [ta.ˈman] ‘her/his father’. However, stress occurs on the rightmost or only [a] in the word, as in [ˈna.nun] ‘her/his child’ and [ˈŋa.sol] ‘I fled’. If there is no [a], stress occurs on the last or only [e] or [o], as in [kr.ˈŋen] ‘his/her finger/toe’, [u.ˈsol] ‘you (sg) fled’, [ˈpe.in] ‘woman’. Based on the description provided by Ross, the sonority distinctions that stress is sensitive to in Takia are therefore | a > e o > i u |. The table excludes systems in which the only stress-sensitive avoidance involves schwa; we discuss such cases below. On its face, Table 2 below seems to present an impressive array of cases that support the claim that sonority-driven stress exists. However, many of the descriptions come from a single source, so there is no way to evaluate their reliability in the context of that particular language. Of course, there is an indirect way of evaluating reliability: i.e. to show that the methods used in the particular description were reliable when used elsewhere. However, this is impossible to do in practice because none of the descriptions discussed their methods in adequate detail, or at all. Even if ‘impressionistic description’ always involved exactly the same procedure, we have seen that it is unreliable from the Au, Gujarati, and Eastern Armenian case studies discussed above. The best described cases are Gujarati and Nganasan, but we have seen that the impressionistic descriptions disagree with each other, and with experimental results in the case of Gujarati. Of the 15 languages cited above, eight have multiple descriptions: Au, Gujarati, Mordwin, Kara, Nganasan, Kobon, Harar Omoro, and Umutina. Disagreements about stress assignment were found in five out of these seven: i.e. Au, Gujarati, Kara, Nganasan, and Kobon. Even more significantly, Raisin (2017) argues that the data for Kobon presented in the primary sources does Evidence for Sonority-Driven Stress CatJL 18, 2019 27

Table 2. Languages with peripheral vowel distinctions2 Language Version Sonority distinctions for stress

(a) Scorza (1973, 1992) None (see above) Au (b) Scorza (1976,1985) aː a o u i > ɨ, ʌ Cowichan Bianco (1998) a e > i > ə Gujarati See section 3.1.2; Shih (2018a, b) Uncertain – see section 3.1.2; Shih (2018a, b) Owens (1985) Harar Omoro a > ʌ, ɔ, ɨ de Lacy (2002a) (a) Schlie & Schlie (1993) a > ɛ ɔ e o ɪ ʊ i u ə Kara (b) Schlie (1996) a > ɛ ɔ e o ɪ ʊ i u ɐ (a) Davies (1981) a au ai > o e u i > ə ɨ a > e o > i u > ə > ɨ Kobon (b) Kenstowicz (1997) (based on data from Davies 1981) (c) Davies (1980) Penultimate stress Ma Manda Pennington (2013) a > e o > ə > i u > ɨ Tsygankin & Debaev (1975) Mordwin e o ä a > i u ɨ Kenstowicz (1997) Nanti Crowhurst & Michael (2005) a > e o > i (a) de Lacy (2002a, 2004) a e o > i y u ə ɨ Nganasan (b) Vaysman (2009) a e o i y u > ɨ > ə Pichis Ashéninka Payne (1990) a e o > i Takia Ross (1995, 2002, 2009) a > e o > i u a > ɛ ɔ e o > i u ɨ Telles (1995), Wetzels & Meira (2010), (based on data from Wetzels, Telles & Umutina Wetzels, Telles & Hermans (2014) Hermans (2014)) or a > ɛ ɔ > e o > i u ɨ Yessan-Mayo Foreman & Marten (1973) a > ɔ, ʌ , ɨ Yimas Foley (1991) a > i u > ɨ

2. Language (alternate name(s); iso 639-3 code): Classification; Location: Au (avt): Torricelli > Wapei-Palei > Wapei; Sandaun Province, Papua New Guinea. Cowichan (Halkomelem; hur): Salish; British Columbia, Canada. Gujarati: Indo-European>Western Intermediate Indo-Aryan; Gujarati province, India and elsewhere. Harar Oromo (Eastern Oromo; hae): Afro-Asiatic > East Cushitic; Afar, Harari, Oromia and other regions. Kara (Lemakot; leu): Oceanic >Meso Melanesian; New Ireland, Papua New Guinea. Kobon (kpw): Trans-New Guinea, Madang; Madang province, Papua New Guinea. Ma Manda (skc): Trans-New Guinea >Finisterre; Papua New Guinea. Mordwin (Erzya; myv): Uralic > Mordvin; Russia. Nanti (cox): Maipurean >Ashéninga Campa; Peru. Nganasan (Tavgi Samoyed; nio): Uralic >Samoyed; Siberia, Russia. Pichis Ashéninka (cpu): Maipurean >Ashéninga Campa; Peru. Takia (tbc): Oceanic >North New Guinea >Nuclear Bel; Papua New Guinea. Umutina (Umotina; umo): Bororoan; Mato Grosso, Paraguay. Yessan-Mayo (yss): Sepik >Tama; Papua New Guinea. Yimas (yee): Ramu-Lower Sepik >Karawari; Papua New Guinea. 28 CatJL 18, 2019 Shu-hao Shih; Paul de Lacy not support the generalizations made about sonority-driven stress by Kenstowicz (1997) and others. It is therefore difficult to evaluate what the table above actually means in terms of evidence. Potentially it means nothing. We have also seen for Eastern Armenian that mere agreement among impressionistic descriptions is not necessarily proba- tive: all impressionistic descriptions of Eastern Armenian agreed with each other, but they did not agree with experimental results. As Haghverdi (2016) and Shih (2018a,b) discuss, it is possible that the table above simply shows that humans with particular linguistic backgrounds interpret certain perceptual impressions as ‘sonority-driven stress’. One incidental phonetic property of different vocalic sonority levels is duration: higher sonority vowels have longer inherent duration than lower sonority vowels (Peterson & Lehiste 1960). Absent any of a listener’s expected perceptual cues to stress, it is possible that listeners will employ these inherent duration differences as significant, leading to the perception of sonority-driven stress. Another possibility was discussed above for Eastern Armenian: schwa is often shorter than other vowels, and so interpolated F0 might be affected by the presence of schwa in a word, leading to the misperception of stress by a perceiver who is used to attending to F0 as a stress cue. Dobrovolsky (1999: 541) discusses a similar case for Chuvash. There is no perceptual cue for stress in words that only have reduced vowels; however, Dobrovolsky suggests: “Chuvash disyllabic words … are characterized by an intonation drop across the first vowel early in the word. I suggest that this is what is interpreted as “stress,” given a lack of other robust stress measures on the word.” For relevant discussion, see Tabain et al. (2013) and Bowers (2019).

3.2. Misidentification of cause While some cases of apparent schwa avoidance may be misdescribed (e.g. Eastern Armenian in section 3.1.4, above), we argue that feet can avoid schwa. However, the cause of such avoidance is a side-effect of representational differences between schwa and other vowels, and not because there are direct restrictions on schwa and foot elements.

3.2.1. Paiwan and non-moraic schwas Shih (2018b) provides a detailed example from Piuma Paiwan (Austronesian > Paiwan; spoken in Paiwan, Taiwan). We only briefly summarize the results of Shih (2018a,b)’s experiment and acoustic analysis here as our goal is to compare them to the impressionistic descriptions’. Descriptions of Piuma Paiwan have reported a sonority-driven stress system that is sensitive to the distinction between schwa and peripheral vowels (Chen 2009a,b, Yeh 2011). The default position for stress is on the penultimate syllable (10a). However, stress falls on the final syllable if the penult contains a schwa (and the final syllable does not) (10b). If both the penult and final syllable contain schwa, stress falls on the final syllable. Evidence for Sonority-Driven Stress CatJL 18, 2019 29

(10) Piuma Paiwan’s sonority-driven stress according to Chen (2009a, b) and Yeh (2011: 116-117) (a) Default stress on the penultimate syllable [ˈkaka] ‘sibling’ [ˈgadu] ‘mountain’ [ˈvuvu] ‘grandparents’ [ˈtsaviʎ] ‘year’ [ˈɭigim] ‘needle’ [ˈtutaŋ] ‘aluminium’ [tsaˈʎiŋa] ‘ear’ [ˈpiku] ‘elbow’ [viˈtsuka] ‘stomach’ [ʎaˈvatsaq] ‘horsefly’ [ˈragəd] ‘pebble’ [ˈtidəq] ‘interval’ [maˈqipər] ‘unlucky’ (b) Stress the final syllable if the penult contains a schwa and the ultima does not [kəˈri] ‘small’ [qurəˈpus] ‘cloud’ [cəˈvus] ‘sugarcane’ [qapəˈdu] ‘gall’ [kəˈman] ‘to eat’ [kəməˈlaŋ] ‘to know’ (c) Stress the final syllable if both penultimate and final syllables contain schwa [ɭəˈʎət] ‘lip’ [ʎisəˈqəs] ‘nit’ [tsəˈməɭ] ‘grass’ [masəŋˈsəŋ] ‘to make something’

In Kenstowicz (1997) and de Lacy (2002a, 2006)’s theories, Piuma Paiwan’s stress system would be analyzed using a constraint such as *HdFoot/ə (or *DFt≤ə). Such a constraint imposes a direct requirement that foot heads not have the sonority of schwa (or lower). However, an alternative is presented by ‘defective schwa’ and ‘minor syllable’ theories – i.e. stress avoids schwa because schwa heads a prosodically defective syllable. Following Shih (2018b), we suggest that schwas in Piuma Paiwan are pressured to be non-moraic: i.e. they form a minor syllable. So, stress falls on the final syllable in [kərí] because the schwa is non-moraic (symbolized as [ə] here – e.g. [kərí]), and the head syllable of a foot must contain at least one mora. While minor syllable theory allows an alternative to the direct sonority-foot reference theories of Kenstowicz (1997) and de Lacy (2004), we believe minor syllable theory provides a number of analytical and empirical advantages, and these advantages are particularly clear in this case. Importantly, Shih (2018b)’s experiment and acoustic analysis confirm the basic claims about the position of stress by the impressionistic descriptions. F0 is a key diagnostic, as shown in Figures 3 and 4. For words with putative penultimate stress, F0 peaks over the penult (Figure 3); for words with putative final stress, F0 peaks over the final syllable (Figure 4). Thus, if F0 is a reliable indicator of foot head position, the experimental results support the impressionistic description. 30 CatJL 18, 2019 Shu-hao Shih; Paul de Lacy

Figure 3. F0 over putatively penultimate-stressed words.

Figure 4. F0 over putatively final-stressed words.

However, Shih (2018b) shows that previous impressionistic descriptions missed important generalizations about the duration of vowels in various stress environments. In words with the shape [CəˈCV] (where V is not schwa), the final vowel is long and the schwa is remarkably short: i.e. phonologically [CəˈCVːμμ] (we include moras here to be explicit). On average, [u]s in this environment were 214ms (s.d.=25ms) in duration in focused position (of peripheral vowels, only [u] was measured), compared to the much shorter stressed [u]s in [ˈCuCu] (mean=141ms; s.d.=31ms), and unstressed final [u]s (mean=156ms; s.d.=31ms). Shih (2018b)’s analysis is that Piuma Paiwan feet must contain a minimum of two moras and be right-aligned, but schwas seek to be moraless (i.e. appear in a minor syllable). Consequently, underlying /CəCV/ must undergo two changes: one is that the schwa must lose its mora, and the second is that the V must become bimoraic – i.e. long. The output is then [Cə(ˈCVːμμ)], with a penultimate non-moraic syllable and a final bimoraic syllable. The consequence of this form is that Piuma Paiwan actually involves a metrical system with a right-aligned quantity-sensitive trochee. There is no need to invoke any direct prohibition on having schwa as a foot head. Instead, the position of the foot head is a side-effect of schwa’s desire to appear in a minor syllable. For Shih (2018b) this ‘desire’ is effected by the constraint *μ/ə “Incur a violation for every schwa that bears a mora”; this constraint Evidence for Sonority-Driven Stress CatJL 18, 2019 31 is a specific instantiation of long-standing theories that regulate the sonority of syllable nuclei (e.g. Zec 2007; Prince & Smolensky 2004). Shih (2018a,b)’s theory makes a variety of additional predictions. For CVCV words like [ˈvuvu], the need for a right-aligned bimoraic foot results in the phonological form [(ˈvuμ.vuμ)], with two monomoraic vowels. (There is no motivation for a form such as *[vuμ(ˈvuːμμ)] to emerge as this would require either unfaithfulness to underlying mono-moraicity, or the appearance of long vowels, which are generally avoided). The duration results bear this prediction out, with both [u]s being significantly shorter than final stressed [u]s. For CVCə words like [ˈtsukəs], the prediction is rather striking: that the output should have the form [(ˈCVμ.Cəμ)], with a mono-moraic schwa. Without a mono-moraic schwa, the foot would either lack two moras (*[(ˈCVμ.Cə)]) or not be right-aligned (*[(ˈCVμμ)Cə]). This phonological analysis predicts that schwa in such words should have significantly longer duration than schwas in minor syllables. This is in fact Shih (2018a,b)’s finding, with schwas predicted to be in minor syllables being around 50ms in duration (with around 20ms s.d.), while the schwa in CVCə words is significantly longer with a mean of 72ms (16ms s.d.). Finally, the theory predicts that words with schwa in both penultimate and ulti- mate position necessarily involve a moraic schwa. In fact, the phonological form of words like [ɭəˈʎət] is actually [ɭə(ˈʎəːμμt)] – the first schwa has a very short duration (mean=50ms; s.d.=18ms), while the second is relatively very long (mean=134ms; s.d.=28ms) (c.f. monomoraic schwa at 72ms). Again, the phonological output [ɭə(ˈʎəːμμt)] contains a bimoraic right-aligned foot. Shih (2018a,b) shows why the competing form *[(ɭəμˈʎəμt)] loses in this competition. The value of Shih (2018a,b)’s theory is that it explains why vowels have different durations depending on their position, particularly for schwa. In contrast, theories such as Kenstowicz (1997) and de Lacy (2004)’s face difficulties in explaining cases like Piuma Paiwan because constraints against schwa in foot heads do not require schwas to be non-moraic. In fact, de Lacy (2004)’s theory does not even entertain the possibility of non-moraic schwas; after all, an underlying motivation of de Lacy (2004)’s theory is to avoid the idea that schwa can be representationally special – its specialness is computational, due to constraints that prohibit schwa in foot heads. So, two issues arise. If schwa is never representationally special (i.e. it always has at least one mora), then how does de Lacy (2004)’s theory account for the differences in duration between the three types of schwa in Piuma Paiwan? Recall that the schwas fall into three categories durationally: (a) ones with minimal (inherent) duration (means of around 40-50ms), (b) ones with duration comparable to short full vowels (means of 72-80ms), and (c) ones with duration comparable to long full vowels (means of 134-137ms). Shih (2018a,b)’s theory accounts for these different types by moraic content: the ultra-short schwas are non-moraic, the mid- dling duration ones have one mora, and the long vowels have two moras. However, if there are no non-moraic schwas – as de Lacy (2004)’s theory implies – then there is no way to explain the three durational types of schwa. So, Piuma Paiwan seems to present strong evidence for three moraic types of schwa, thereby explaining both duration and stress position. A response would 32 CatJL 18, 2019 Shu-hao Shih; Paul de Lacy be to accept de Lacy (2004)’s theory, but admit degenerate syllables, such as mono-moraic schwa. The problem with such an approach is that it then under- mines the need for constraints such as *HdFoot/ə: when stress avoids schwas, it could be due to the non-moraicity of schwa rather than because foot heads avoid low sonority vowels.

So, evidence for *HdFoot/ə would consist of a case where stress demonstrably avoided schwa, yet schwa had comparable duration to other vowels, or was otherwise demonstrably moraic. At this point, we are unaware of such evidence. Part of the problem is that determining which approach is correct requires careful analysis of vowel durations – both schwa’s and other vowels’ – in an appropriately controlled setting. Without such an analysis, reports are merely impressionistic, and as we have seen for Piuma Paiwan, impressionistic descriptions may correctly identify the location of stress (or at least, of pitch peaks) but miss duration differences entirely. There are three other relevant experimental studies involving sonority-driven stress and schwa that we are aware of. Haghverdi (2016)’s analysis of Eastern Armenian was mentioned above. In it, stress landed on a final schwa. In Shih (2018a,b)’s terms, this implies that final schwa is moraic, otherwise it could not serve as the head of a foot. Interestingly, Gordon et al. (2012) found that the mean duration of Armenian schwa was equivalent to that of high vowels ([i]=64ms (13ms s.d.); [u]=68ms (15ms s.d.); [ə]=68ms (26ms s.d.)), implying that they bear a mora, as expected under Shih (2018)’s theory. Dobrovolsky (1999) reports the results of an acoustic analysis of Chuvash, which is reported to have final stress (e.g. [sar.la.ká] ‘widely’), with retraction away from schwas (e.g. [jə.nér.tʃək] ‘saddle’), and initial stress on words with only schwas (e.g. [tə́.tə.mər] ‘we got up’). Dobrovolsky (1999: 541) reports that “R[educed] vowels that precede or follow a stressed syllable are in general extremely short and non-prominent” – just as one would expect if they were non-moraic. An interesting result is that in words with only reduced vowels there seems to be no acoustic measure that marks stress. In this case, then, schwas seem to be non-moraic in all situations, even to the point where there is no foot at all in words with only schwas. The final case we know of is Blum (2018)’s analysis of Munster Irish, covered in the next section.

3.2.2. Munster Irish and allophony-driven footing Blum (2018) discusses a case of apparent sonority-driven stress in Munster Irish. There have been several impressionistic descriptions of Munster Irish stress (Ó Cuív 1944; Ó Sé 1989, 2008; Blankenhorn 1981; Gussman 2002; Hickey 2011, 2014), and they generally converge on the following generalizations, with some slight variations (Blum 2018§2.1). Note that Munster Irish also has vowel reduction, where every unstressed short vowel becomes schwa. Evidence for Sonority-Driven Stress CatJL 18, 2019 33

(11) Munster Irish stress (data adapted from Gussman 2002) (a) Stress the leftmost long vowel or diphthong [gaˈdiː] gadaí ‘thief’, [məˈkaːntə] ‘modest’, [əməˈdaːn] amadán ‘fool’ (b) Else stress the second syllable if it contains an [a] followed by a [x] [bəˈkax] bacach ‘lame’, [kəˈdjaxtə] cuideachta ‘company’ (c) Else stress the initial syllable [ˈsoləs] solas ‘light’, [ˈokəɹəx] ocrach ‘hungry’ On its face, Munster Irish (MI) seems like at least a partial sonority-driven stress system. The sonority-driven aspect involves [ax] where the [a] is in the second syllable: e.g. [bə.ˈkax] ‘lame’, *[ˈba.kəx]. However, [ax] later in the word does not attract stress (e.g. /okərax/ → [ˈokəɹəx], *[əkə.ˈɹax]). Also, long vowels attract stress over [ax] (e.g. /ʃeːʃax/ → [ˈʃeːʃəx] ‘melodic’, *[ʃeːˈʃax]). A curious aspect of the system is that the [x] does not appear to necessarily belong to the second syllable: e.g. [bə.ˈka.xə] ‘lame (pl.)’ (Green 1996: 4). However, Blum (2018), which involved a production experiment and acoustic analysis, made an additional significant discovery: “The [a] in Cax syllables does not reduce in any unstressed position: neither in CaxˈCVː, nor in ˈCaxCax” (p. 20). In other words, [x] blocks /a/ from becoming schwa in all unstressed positions. Blum (2018)’s analytical proposal is that MI stress is driven by allophony. Unstressed short vowels seek to reduce, and the metrical system requires a trochaic foot at the left edge. In most words, these two requirements can be met: e.g. /faɹaga/ → [(ˈfa.ɹə)gə]. However, in words with the shape /CVCax/, like [bəˈkax], it is not possible to meet both requirements. One option is to stress the leftmost syllable and not reduce the /a/ (e.g. /bakax/ → [ˈba.kax] – recall that /a/ cannot reduce before [x]). The other option is to stress the second syllable and reduce the first vowel (e.g. [bəˈkax]). Blum (2018)’s analysis is that in the face of this conundrum, MI opts for the latter approach: stress moves to the peninitial syllable so that at least one short vowel can reduce. Blum (2018)’s analytical insight can be coupled with Shih (2018b)’s non- moraic schwa theory by analyzing vowel reduction as mora loss. So, MI’s stress system as requiring a left-aligned foot that is left-headed by default, but right- headed when the leftmost vowel can become non-moraic while the rightmost one cannot. This situation is formally expressed in tableau (12). Candidate (a) wins because it avoids reducing [a] before [x], and reduces unstressed syllables to non-moraic schwa (thanks to *non-head/m “Incur a violation for every vowel that (a) is in the non-head syllable of a foot and (b) bears a mora”), albeit at the expense of not having a left-headed foot. Candidates (b), (c), and (e) all fail because they fatally violate *non-head/m by having an unstressed moraic vowel in a foot. Candidate (d) avoids violating *non-head/m, but at the cost of having a reduced vowel before an [x] – blocked in MI. In other words, as Blum (2018) observes, the pressure to reduce vowels forces feet to change shape. Importantly, there is no constraint here that relates sonority directly to foot structure. There is a constraint that relates schwa to segmental context – *əx, and one that relates foot structure to 34 CatJL 18, 2019 Shu-hao Shih; Paul de Lacy moraic content (*non-head/m), but no constraint that bans a structure consisting of a prosodic element and a sonority category (e.g. *non-head/ə).

(12) Munster Irish stress movement due to allophony, adapting Blum (2018)

/bakax/ *əx *non-head/m ident-F trochee

☞ (a) (bə.ˈkamx) * * (b) (bam.ˈkamx) *! * (c) (ˈbam.kamx) *! (d) (ˈbam.kəx) *! * (e) (ˈbəm.kamx) *! *

Finally, [CaxCax] words (e.g. [ˈljax.tax] leachtach ‘liquid’) have stress on the initial syllable but reduce neither [a]; this is due to *əx outranking *non-Head/m, as seen in tableau (13).

(13) Munster Irish stress movement due to allophony #2

/ljaxtax/ *əx *non-head/m ident-F trochee ☞ (a) (ˈljamx.tamx) * (b) (ˈljamx.təx) *! * (c) (ljəx.ˈtamx) *! * * (d) (ljamx.ˈtamx) * *!

In conclusion, the apparent sonority-driven stress of Munster Irish can be ascribed to the need to block reduction of /a/ before [x], and no constraints that directly relate sonority to foot heads (or non-heads) are necessary. In general terms, restrictions on allophony can influence footing, and so appear to be sonority-driven stress.

4. Implications We have argued above that there is no robust evidence for a phonological mechanism that induces a deviation from default foot form because of a direct condition on the sonority content of a foot – i.e. ‘sonority-driven stress’. However, apparent stress avoidance of low-sonority vowels is possible when it is due to the incidental effect of other phonological conditions. When schwa is required to be moraless, foot conditions on moraic content may cause deviation from default footing, as in Piuma Paiwan. When conditions on allophony override foot restrictions, feet may also depart from their usual position, as in Munster Irish. Evidence for Sonority-Driven Stress CatJL 18, 2019 35

Our argument in section 3.1 might seem to be that impressionistic descriptions in general are unreliable. This is not necessarily the case. The reliability of any device and methodology must be examined with respect to a particular task; it is possible that humans are very reliable with some detection tasks (e.g. identifying forms of suppletive allomorphs). In this regard, we simply conclude here that in regard to sonority-driven stress, the impressionistic method is not reliable. One consequence is that future work on sonority-driven stress – and perhaps all stress – cannot be solely impressionistic or rely on impressionistic sources. For cases of putative sonority-driven stress in particular, it is necessary to detect fine distinctions in duration and examine acoustic properties (e.g. intensity, spectral tilt, F1/F2) to a degree that is beyond the capacity of humans unaided by appropriate hardware and machine analysis. However, we again emphasize that any device and methodology must be examined for reliability. This point also applies to experiments – whether production or perception, and however analyzed; it is not adequate to assume that any device – including microphones, digitizers (e.g. sound cards), palatographs, spectrographs, or analytical software – is reliable. To emphasize this point, conducting experiments is in no way an immediate panacea for methodological reliability: we are actually not advocating for acoustic experiments – or any other non-impressionistic methodology. We are instead advocating that any method used to gather evidence must be evaluated for reliability. For example, above, we determined that the method of relying on impressionistic descriptions is not reliable; all other methods require the same determination to be made. We are aware that our article has implications beyond sonority-driven stress. For metrical theory, the only potential influence on feet apart from edges is moraic content (and perhaps tone – de Lacy 2002b, 2007, though cf. Oakden 2018; also see de Lacy 2007 for other structural factors). The finding in this article means that metrical stress theories can be more restrictive, eliminating mechanisms that allowed poorly constrained reference to non-moraic properties, as in Halle & Vergnaud (1987) and Hayes (1995) (see section 2 above). Perhaps the broadest issue that we broach here is the question of “what is good evidence?” This question has rarely been addressed in theoretical phonology (though see de Lacy 2008, de Lacy & Kingston 2013 for relevant discussion). Even so, we have identified one major requirement here, widely accepted in other fields: methods must be provably reliable for them to have value in producing evidence (e.g. Ray 2009). Furthermore, sonority-driven stress is not obviously a special case in phonology. We suspect that the same discoveries about reliability of impressionistic descriptions – and even experimental work – could be made for other areas, too. It is possible that what we have uncovered for sonority-driven stress evidence indicates a broader crisis within phonology, and perhaps in other linguistics fields: i.e. that the reliability of evidentiary methods is largely unknown, and so the evidence arising from those methods is of indeterminate value. Without adequate evidence, the success of phonological theories cannot be tested. It seems to us reasonable to consider a science without testable theories to be in a state of crisis. 36 CatJL 18, 2019 Shu-hao Shih; Paul de Lacy

References Adenwala, Mridula. 1965. The structural analysis of the phonology and morphemics of Gujarati. PhD dissertation, State University of New York at Buffalo. Bianco, Violet M. 1998. Stress assignment in Halkomelem-Cowichan roots. In Dawn Bates (ed.). Papers for the 33rd International Conference on Salish and Neighboring Languages. University of Washington, Seattle. 60-76. Blankenhorn, Virginia. 1981. Pitch, quantity and stress in Munster Irish. Éigse 18: 225-250. Blum, Eileen. 2018. Allophony-driven stress in Munster Irish. Ms., Rutgers University, New Jersey. . Bowers, Dustin. 2019. Positional vs. sonority-driven stress in Gujarati: New experimental evidence. Phonological Data and Analysis 1.1: 1-28. Campbell, George L. & Gareth King. 2011. Concise compendium of the world’s languages. 2nd edn. London & New York: Routledge. Cardona, George. 1965. A Gujarati reference grammar. Philadelphia: The University of Pennsylvania Press. Cardona, George & Babu Suthar. 2003. Gujarati. In George Cardona & Dhanesh Jain (eds.). The Indo-Aryan languages. London & New York: Routledge, 722-765. Carpenter, Angela. 2006. Aquisition of a natural versus an unnatrual stress system. PhD dissertation, University of Massachusetts, Amherst. Castren, M. Alexander. 1854. Grammatik der samojedischen Sprachen. St Petersburg: Buchdruckerei der Kaiserlichen Akademie der Wissenschaften. Cheliah, Shobhana L. & Willem J. de Reuse. 2011. Handbook of descriptive linguistic fieldwork. Dordrecht: Springer. Chen, Chun-Mei. 2009a. Documenting Paiwan phonology: issues in segments and non- stress prosodic features. Concentric: Studies in Linguistics 35: 193-223. Chen, Chun-Mei. 2009b. The phonetics of Paiwan word-level prosody. Language and Linguistics 10: 593-625. Crowhurst, Megan J. & Lev Michael. 2005. Iterative footing and prominence–drive stress in Nanti (Kampa). Language 81: 47-95. Davies, John. 1980. Kobon phonology. Canberra: Australian National University. Davies, John. 1981. Kobon. Amsterdam: North-Holland. de Lacy, Paul & John Kingston. 2013. Synchronic explanation. Natural Language and Linguistic Theory 31: 287-355. de Lacy, Paul & Patrik Bye. 2008. Metrical influences on lenition and fortition. In Joaquim de Carvalho, Tobias Scheer, and Philippe Ségéral (eds.). Lenition and Fortition. Studies in Generative Grammar. Berlin: Mouton de Gruyter, 173-206. de Lacy, Paul. 2002a. The formal expression of markedness. PhD dissertation, University of Massachusetts, Amherst. de Lacy, Paul. 2002b. The interaction of tone and stress in Optimality Theory. Phonology 19: 1-32. de Lacy, Paul. 2004. Markedness conflation in Optimality Theory. Phonology 21: 145-199. de Lacy, Paul. 2006. Markedness: reduction and preservation in phonology. Cambridge: Cambridge University Press. Evidence for Sonority-Driven Stress CatJL 18, 2019 37 de Lacy, Paul. 2007. The interaction of tone, sonority, and prosodic structure. In Paul de Lacy (ed.). The Cambridge handbook of phonology. Cambridge: Cambridge University Press, 281-307. de Lacy, Paul. 2009. Phonological evidence. In Steve Parker (ed.). Phonological argumentation: Essays on evidence and motivation. Equinox Publications, 43-78. de Lacy, Paul. 2014. Evaluating evidence for stress systems. In Harry van der Hulst (ed.). Word stress: theoretical and typological issues. Cambridge: Cambridge University Press, 149-193. Dobrovolsky, Michael. 1999. The phonetics of Chuvash stress: implications for phonology. In John J. Ohala, Yoko Hasegawa, Manjari Ohala, Daniel Granville & Ashlee C. Bailey (eds.). Proceedings of the 14th International Conference of Phonetic Sciences. Berkeley, CA: University of California, Berkeley, Linguistics Department, 539-542. Doctor, Raimond. 2004. A grammar of Gujarati. Munich: Lincom. Dum-Tragut, Jasmine. 2009. Armenian: modern Eastern Armenian. Amsterdam: John Benjamins. Dunn, Michael. 1999. A grammar of Chukchi. Doctoral dissertation, Australian National University. Foley, William A. 1991. The Yimas language of New Guinea. Stanford: Stanford University Press. Foreman, Velma & Helen Marten. 1973. Yessan-Mayo phonemes. In Alan Healey (ed.). Phonologies of Three Languages of Papua New Guinea. Ukarumpa: Summer Institute of Linguistics. 79-108. Fry, D. B. 1955. Duration and intensity as physical correlates of linguistic stress. Journal of the Acoustical Society of America 27: 765-768. Garellek, M., & White, J. 2012. Stress correlates and vowel targets in Tongan. UCLA Working Papers in Phonetics 110: 65-85. Gordon, M. 2004. A phonological and phonetic study of word-level stress in Chickasaw. International Journal of American Linguistics 70: 1-32. Gordon, Matthew & Ayla Applebaum. 2010. Acoustic correlates of stress in Turkish. Journal of the International Phonetic Association 40: 35-58. Gordon, Matthew & Latifa Nafi. 2012. Acoustic correlates of stress and pitch accent in Tashlhiyt Berber. Journal of Phonetics 40: 706-724. Gussman, Edmund. 2002. Phonology: analysis and theory. Cambridge: Cambridge University Press. Haghverdi, Vartan. 2016. Armenian schwa: a phonetic and phonological analysis. MA thesis, Rutgers University, New Jersey. Hajdú, Peter. 1964. Samoiedica. Nyelvtudoma´nyi Közleme´nyek 66: 397-405. Halle, Morris & Jean-Roger Vergnaud. 1987. An essay on stress. Cambridge, Mass.: MIT Press. Hayes, Bruce. 1995. Metrical stress theory: principles and case studies. Chicago: University of Chicago Press. Hayes, Bruce. 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20: 253-306. Helimski, Eugene. 1998. Nganasan. In Daniel Abondolo (ed.). The Uralic languages. London & New York: Routledge, 480-515. Hickey, Raymond. 2011. The dialects of Irish: study of a changing landscape. Berlin: De Gruyter Mouton. 38 CatJL 18, 2019 Shu-hao Shih; Paul de Lacy

Hickey, Raymond. 2014. The sound structure of Modern Irish. Berlin & Boston: De Gruyter Mouton. Hyman, Larry. 1985. A theory of phonological weight. Dordrecht: Foris. Kager, René. 2007. Feet and metrical stress. In Paul de Lacy (ed.). The Cambridge Handbook of Phonology. Cambridge: Cambridge University Press, 195-227. Kenstowicz, Michael. 1997. Quality-sensitive stress. Rivista di Linguistica 9: 157-187. Khachatryan, Amalia. 1988. Zhamanakakits hayereni hnchuytabanutyun (Phonology of contemporary Armenian). Yerevan: Haykakan SSH GA Hratarakchutyun. Ladefoged, Peter. 2003. Phonetic data analysis: an introduction to fieldwork and instrumental techniques. Blackwell Publishing. Lieberman, Philip. 1960. Some acoustic correlates of word stress in American English. Journal of the Acoustical Society of America 32: 451-454. Lima, Stella Telles Pereira. 1995. A língua Umutina: “um sopro de vida”. MA thesis, Universidade Federal de Pernambuco (UFPA), Recife, Brazil. Lin, Yen-Hwei. 1998. On minor syllables. In Osamu Fujimura, Brian D. Joseph & Bohumil Palek (eds). Proceedings of LP’98: Item order in language and speech. Prague: The Karolinum Press, 163-183. Lublinskaya, Marina, Valentine Goussev & Tatiana Sherstinova. 2000. Nganasan mul- timedia dictionary: a collection of audio material in the Nganasan language. Demo version. Masica, Colin P. 1991. The Indo–Aryan languages. Cambridge: Cambridge University Press. Master, Alfred. 1925. Stress accent in Modern Gujarati. Journal of the Bombay Branch of the Royal Asiatic Society 1: 76-94. McGarrity, Laura. 2003. Constraints on patterns of primary and secondary stress. PhD dissertation, Indiana University. Mistry, P. J. 1997. Gujarati phonology. In A. S. Kaye (ed.). Phonologies of Asia and Africa (including the Caucasus). Vol. 2. Winona Lake, Ind.: Eisenbrauns, 653-673. Modi, Bharati. 2013. Some issues in Gujarati phonology (Languages of the World 44). Muenchen: Lincom. Ó Cuív, Brian. 1944. The Irish of West Muskerry, Co. Cork. Dublin: Dublin Institute for Advanced Studies. Ó Sé, Diarmuid. 1989. Contributions to the study of word stress in Irish. Ériu 40: 147-178. Ó Sé, Diarmuid. 2008. Word stress in Munster Irish. Éigse 36: 87-112. Oakden, Chris. 2018. Tone and stress in Lithuanian. Ms., Rutgers University, New Jersey. Owens, Jonathan. 1985. A grammar of Harar Oromo (Northeastern Ethiopia). Hamburg: Helmut Buske Verlag. Payne, Judith. 1990. Asheninca stress patterns. In Doris L. Payne (ed.). Amazonian linguistics: studies in lowland South American languages. Austin: University of Texas Press, 185-209. Pennington, Ryan. 2013. Ma Manda phonology. MA thesis, Graduate Institute of Applied Linguistics, Dallas. Peterson, Gordon E. & Ilse Lehiste. 1960. Duration of syllable nuclei in English. JASA 32: 693-703. Evidence for Sonority-Driven Stress CatJL 18, 2019 39

Prince, Alan & Paul Smolensky. 2004. Optimality Theory: constraint interaction in generative grammar. Malden, Mass. & Oxford: Blackwell. Prince, Alan. 1983. Relating to the grid. Linguistic Inquiry 14: 19-100. Prokofjev, G.N. 1937. Nganasanskij (tavgijskij) dialekt. In Iazyki i pis ‘mennost’ samoedskii i finno-ugorskii narodov. Moscow & Leningrad: Učpedgiz, 53-74. Raisin, Ezer. 2017. The stress-encapsulation universal and phonological modularity. Ms. MIT. . Ray, William J. 2009. Methods towards a science of behavior and experience. Belmont: Wadsworth. Ross, Malcolm. 1995. Takia. In Darrell T. Tryon (ed.). Comparative Austronesian Dictionary: An Introduction to Austronesian Studies. Berlin & New York: Mouton de Gruyter, 677-686. Ross, Malcolm. 2002. Takia. In John Lynch, Malcolm Ross & Terry Crowley (eds.). The Oceanic Languages. Richmond: Curzon Press, 216-248. Ross, Malcolm. 2009. Loanwords in Takia, an Oceanic language of Papua New Guinea. In Martin Haspelmath & Uri Tadmor (eds.). Loanwords in the World’s Languages: A Comparative Handbook. Berlin: Mouton de Gruyter, 747-770. Sakayan, Dora. 2007. Eastern Armenian for the English-speaking world: a contrastive approach. Yerevan: Yerevan State University Press. Schiering, René & Harry van der Hulst. 2010. Word accent systems in the languages of Asia. In Rob Goedemans, Harry van der Hulst & Ellen van Zanten (eds.). A survey of word accentual patterns in the languages of the world. Berlin: Mouton De Gruyter, 509-613. Schlie, Perry & Ginny Schlie. 1993. A Kara phonology. In John M. Clifton (ed.). Data papers on New Guinea linguistics: Phonologies of Austronesian languages 2. Papua New Guinea: SIL Academic Publications, 99-130. Schlie, Perry. 1996. Kara organised phonology data. Ukarumpa: Summer Institute of Linguistics. Available at . Scorza, David. 1973. Phonemes of the Au language. Ukarumpa: Summer Institute of Linguistics. Available at . Scorza, David. 1976. Au language: Word, phrase, clause. SIL Language and Culture Archives. Scorza, David. 1985. A sketch of Au morphology and syntax. Papers in New Guinea Linguistics 22: 215-273. Scorza, David. 1992. Au organised phonology data. Ukarumpa: Summer Institute of Linguistics. Available at . Shih, Shu-hao. 2016. Sonority-driven stress does not exist. In Gunnar Ólafur Hansson, Ashley Farris-Trimble, Kevin McMullin & Douglas Pulleyblank (eds.). Supplemental Proceedings of the 2015 Annual Meeting on Phonology. Washington, DC: Linguistic Society of America. Shih, Shu-hao. 2018a. On the existence of sonority-driven stress in Gujarati. Phonology 35: 327-364. Shih, Shu-hao. 2018b. Non-moraic schwa: Phonology and phonetics. Doctoral Dissertation, Rutgers University. 40 CatJL 18, 2019 Shu-hao Shih; Paul de Lacy

Skorik, Petr Ja. 1961. Grammatika čukotskogo Jazyka, tom 1. Leningrad: Nauka. Skorik, Petr Ja. 1977. Grammatika čukotskogo Jazyka, tom 2. Leningrad: Nauka. Tabain, Marija, Janet Fletcher & Andrew Butcher. 2014. Lexical stress in Pitjantjatjara. Journal of Phonetics 42: 52-66. Tereschenko, N.M. 1979. Nganasanskii iazyk. Leningrad: Nauka. Tsygankin, P. J. & C. Z. Debaev. 1975. Očerk sravnitel’noj grammatiki mordovskix (mokšanskogo i erz’anskogo) literaturnix jazykov. Saransk: Mordovskij gosudarst- vennyj universitet imeni N.P. Ogareva. Turner, Ralph L. 1921. Gujarati phonology. Journal of the Royal Asiatic Society of Great Britain and Ireland 3: 329-365. van Oostendorp, Marc. 2000. Phonological projection: a theory of feature content and prosodic structure. Berlin & New York: Mouton de Gruyter. Vaux, Bert. 1998. The phonology of Armenian. Oxford: Clarendon Press. Vaysman, Olga. 2009. Segmental alternations and metrical theory. PhD dissertation, MIT. Wetzels, Leo & Sergio Meira. 2010. A survey of South American stress systems. In Rob Goedemans, Harry van der Hulst & Ellen van Zanten (eds.). A Survey of Word Accentual Patterns in the Languages of the World. Berlin & New York: Mouton de Gruyter, 313-380. Wetzels, W. Leo, Stella Telles & Ben Hermans. 2014. Onsets and syllable prominence in Umutina. DELTA: Documentação de Estudos em Lingüística Teórica e Aplicada 30: 703-720. Yeh, Shih-chi Stella. 2011. Issues in Paiwan phonology. PhD dissertation, National Tsing Hua University, Hsinchu, Taiwan. Zec, Draga. 2007. The syllable. In Paul de Lacy (ed.). The Cambridge handbook of phonology. Cambridge University Press, 161-194. Catalan Journal of Linguistics 18, 2019 41-77

Revisiting Top-Down Primary Stress*

Kathryn Pruitt Arizona State University [email protected]

Received: March 6, 2019 Accepted: July 1, 2019

Abstract

Metrical theory recognizes differences between primary and non-primary stresses, sometimes within the same language. In serial theories, this has often led to a parametric approach in derivation: some languages are ‘top-down’, with the primary stress assigned first, while other languages are ‘bottom-up’, where foot construction precedes primary stress placement. This paper examines two languages (Cahuilla and Yine) that have be treated as ‘top-down’ in rule- based metrical theory, and it shows that neither requires a top-down analysis in Harmonic Serialism, a derivational version of Optimality Theory. On the basis of these case studies it is argued that the common, intuitive notion of what makes a language ‘top-down’—a primary stress’s independence from non-primary stresses—is oversimplified. The case studies reveal the importance of theoretical framework and typological predictions in establishing the order of primary and non-primary stress assignment. The argument culminates in a concise statement of Harmonic Serialism-specific criteria for establishing that a top-down derivation is required. Keywords: primary stress; Harmonic Serialism; metrical theory; top-down; bottom-up

Resum. Un altre cop sobre l’accent primari

La teoria mètrica admet diferències entre l’accent primari i el no primari, de vegades dins la mateixa llengua. En les teories serials, això ha comportat sovint un enfocament paramètric de la derivació: algunes llengües comencen per dalt (top-down), amb l’assignació de l’accent primari en primer lloc, mentre que d’altres comencen per baix (bottom-up), en les quals la construcció del peu precedeix l’assignació de l’accent primari. Aquest article examina dues llengües (cahuilla i yine) que han estat considerades top-down en la teoria mètrica basada en regles, i demostra que cap de les dues requereix aquest tipus d’anàlisi en el marc del serialisme harmònic, una versió derivacional de la teoria de l’optimitat. Prenent aquestes llengües com a base, l’article sosté que la noció general i intuïtiva del que fa que una llengua sigui top-down —l’accent primari independent del no primari— és massa simplificada. Els casos estudiats palesen la importància del marc teòric adoptat i de les prediccions tipològiques en l’establi- ment de l’ordre d’assignació de l’accent primari i no primari. L’argumentació culmina amb

* Thanks are due to John McCarthy and to two anonymous reviewers for helpful comments on this paper. I also thank editors of this issue for the invitation to submit and for their patience in shep- herding me through the process.

ISSN 1695-6885 (in press); 2014-9718 (online) https://doi.org/10.5565/rev/catjl.262 42 CatJL 18, 2019 Kathryn Pruitt una relació concisa dels criteris específics per a establir els requisits d’una derivació top-down dins el serialisme harmònic. Paraules clau: accent primari; serialisme harmònic; teoria mètrica; top-down; bottom-up

Table of Contents 1. Introduction 4. Asymmetric patterns and language 2. Defining top-down and bottom-up typology 3. The role of theory in identifying 5. Conclusion asymmetries References

1. Introduction In the rule-based literature within metrical theory, iterative stress patterns are sometimes described as fitting into one of two categories: those that require the primary stress to be assigned before non-primary stresses, and those that do not. The term “top-down” has been used for systems of the first type (e.g., Hayes 1995: 116-117), reflecting the idea that the primary stress represents a higher category of metrical structure, namely, the head of the prosodic word. From this literature we can deduce that, descriptively, languages with top-down stress are those where the primary stress is independent in some sense from the non-primary stresses. Bidirectional stress systems present one type of example, where primary stress is assigned at one edge of the word, while iterative non-primary stresses are assigned from the opposite edge.1 The data from Yine2 (Matteson 1965) in (1) illustrate such a pattern. Yine exhibits a quantity-insensitive trochaic parsing, where the primary stress foot is right-aligned and non-primary feet are parsed from left to right. Monosyllabic feet are not permitted (or, alternatively, a stress clash is prevented), so the primary stress foot appears to stand at a slight distance from the secondary stress feet in words with an odd number of syllables (i.e., a stress lapse is tolerated).

1. Other kinds of primary stress independence discussed in the literature include languages with: lexical (or otherwise non-metrical) primary stress; ‘early’ primary stress and ‘late’ secondary stress; asymmetries in quantity-sensitivity between primary and non-primary stress; different treatment of primary and non-primary stress by extrametricality/Non-Finality; and a few others. For discussions see, for example, Odden (1979), Hayes (1995: 117), Bailey (1995: Ch 1), Hurch (1996), McGarrity (2003), and Goedemans and van der Hulst (2014). 2. Maipurean, Peru (Eberhard, Simons & Fennig 2019). Formerly referred to as Piro. Revisiting Top-Down Primary Stress CatJL 18, 2019 43

(1) Yine stress (Matteson 1965: 21)3

ru.(ˈt͡çi.t͡ça) ‘He observes taboo’ (ˌʧi.ja)(ˈha.ta) ‘He cries.’ (ˌsa.lwa).je.(ˈhka.kna) ‘They visit each other.’ (ˌpe.ʧi)(ˌʧhi.ma)(ˈtlo.na) ‘They say they stalk it.’ (ˌru.slu)(ˌno.ti).ni.(ˈtka.na) ‘Their voices already changed.’ (ˌsa.ple)(ˌwhi.ma)(ˌmta.na)(ˈtna.ka) ‘They say he went along screaming again.’ (ˌka.çru:)(ˌka.khi)(ˌma.na).ta.(ˈtka.na) ‘They were joking together then, it is said.’

A top-down interpretation of this pattern is that the primary stress is first assigned at the right edge of the prosodic word, then secondary stresses are filled in by iterating disyllabic foot construction from left-to-right, stopping when the primary stress foot is reached or when only one syllable remains unparsed. An alternative “bottom-up” derivation—where the non-primary stress feet are built as the first layer of metrical structure before one is later selected as the primary stress—yields the wrong parse in words with an odd-number of syllables. For example, the seven-syllable /ru.slu.no.ti.ni.tka.na/ ‘their voices already changed’, should surface as (ˌru.slu)(ˌno.ti).ni.(ˈtka.na), with an unfooted antepenultimate syllable (underlined) and primary stress on the penult (in bold). But applying the general parsing algorithm for left-to-right trochees first yields this parse: (ˌru.slu)(ˌno.ti)(ˌni.tka)na, which has incorrectly stressed the antepenultimate syllable and made the penult the unstressed member of a foot. This analysis would then require some additional mechanism to place primary stress correctly, such as having the ability to erase the offending foot when it is time to assign primary stress or letting the secondary stress procedure ‘know’ in advance where the primary stress will be so that it does not encroach on the primary stress’s territory. The intuition, therefore, is that a top-down analysis, where the primary stress is assigned first and secondary stresses follow, is a more straightforward way to capture such patterns.4 In contrast, in most stress systems it seems that a bottom-up analysis is possible, or even preferred, because although the primary stress may have its own set of requirements, they do not appear to override the procedures for assigning secondary stresses. A language like Pintupi5 (Hansen & Hansen 1969), with data shown in (2), displays a stress pattern than can be derived with a top-down or a bottom-up derivation, while the data in (3), from MalakMalak6 (Birk 1976),

3. Throughout this paper, data are presented in IPA except where noted. Otherwise, the cited forms retain whatever conventions are followed in their respective sources; in some cases this means a relatively more phonemic transcription, and in others more allophonic. 4. This is deliberately framed here as the ‘intuitively’ preferred characterization. Section 4 revisits Yine and other bidirectional patterns from a more analytic standpoint. 5. Pama-Nyungan, Northern Territory and Western Australia (Eberhard, Simons & Fennig 2019). 6. Daly, Northern Territory (Eberhard, Simons & Fennig 2019). 44 CatJL 18, 2019 Kathryn Pruitt demonstrate a pattern where a bottom-up derivation is the most straightforward analysis. In both cases, quantity-insensitive trochees are responsible for the stress pattern, and the leftmost one is assigned primary stress. This yields a consistent first-syllable primary stress in Pintupi because all feet are left-aligned (or, assigned left-to-right), so the primary stress could be easily derived before or after secondary stress assignment. In MalakMalak, the primary stress fluctuates between initial and peninitial positions because the feet are right aligned (or, assigned right- to-left), which means primary stress placement is not easily determined until the word is fully metrified.

(2) Pintupi stress (Hansen & Hansen 1969: 163) a. (ˈpa.ɲa) ‘earth’ b. (ˈt̻u.ʈa).ja ‘many’ c. (ˈma.ɭa)(ˌwa.na) ‘through (from) behind’ d. (ˈpu.ɭiŋ)(ˌka.la).t̻u ‘we (sat) on the hill’ e. (ˈt̻a.mu)(ˌlim.pa)(ˌt̻uŋ.ku) ‘our relation’ f. (ˈʈi.ɭi)(ˌɾi.ŋu)(ˌlam.pa).t̻u ‘the fire for our benefit flared up’ g. (ˈku.ɾa)(ˌn̻ u.lu)(ˌlim.pa)(ˌt̻u.ɽa) ‘the first one (who is) our relation’

(3) MalakMalak stress (Birk 1976: 16-17) a. (ˈwu.ru) ‘arm (or rivulet)’ b. (ˈa.la)(ˌwar)7 ‘woman’ c. (ˈmu.t̻uɾ)(ˌwu.na) ‘very much’ d. t̻ɛt.(ˈwɛ.ɾa)(ˌmaŋ.kil) ‘fork-stick’ e. (ˈne̠ ŋ.ke̠ )(ˌɾe̠ .ne̠ )(ˌjuŋ.ka) ‘you-pl will lie down.’ f. wu.(ˈwun.tu)(ˌnu.nu)(ˌwak.na) ‘He would have (given) you-sg (meat).’8 g. (ˈnuŋ.ku)(ˌɾun.tu)(ˌwe̠ .ɾe̠ )(ˌwak.ka) ‘You-pl would have (given) them (meat).’

Some metrical theorists have advocated top-down-only modes of analysis (e.g., van der Hulst 1984, 1997, 2009; Bailey 1995; Hurch 1996), but in practice these proposals are usually implemented as ‘top-down-mostly’ in light of patterns like MalakMalak (or else such patterns are reinterpreted, e.g., van der Hulst

7. The three-syllable stress pattern does not conform to the pattern in longer words. I set this aside here. See Goldsmith (1990: 173ff) and Chen (1999: 115f) for discussions that involve a different footing for trisyllables, (ˈσ)(σˌσ). 8. The cited forms in (3)f and g are verbal auxiliaries (“a free form… encod[ing] person, number, tense, mood, and aspect”, Birk 1976: 4). To derive the given translations, the auxiliary is preceded by tɛ aŋ (ibid. p. 17), lit. ‘meat give’ (as glossed in Birk p. 37, for example). Revisiting Top-Down Primary Stress CatJL 18, 2019 45

1997, 2012). A more common approach in the derivational literature is to assume bottom-up parsing as the default mode of structure-building, with top-down parsing available and used by some languages, exemplified in explicit form by Hayes (1995: 116-117). In general, then, the ordering of primary and secondary stress assignment has been treated more or less parametrically, as a language-particular choice. A consequence of this parametric approach to primary stress derivation is that the distinction between ‘top-down’ and ‘bottom-up’ has been seemingly codified as an empirical classification. For example, as a footnote to an otherwise-bottom- up demonstration of metrical stress, Goldsmith (1990: 343) remarks, “[i]n a good number of languages primary or word-level stress is assigned first, and secondary stress is assigned on the basis of the position of the word-level stress that is already assigned”; no examples or other sources are cited, so the implication is that such systems must be easy to spot. In general, the understanding seems to be that languages where primary stress is independent from secondary stresses are those where a top-down analysis is warranted, as in my characterization of Yine above. However, this paper will take a closer look at several such patterns and show that the common, intuitive notion of what makes a language ‘top-down’ is oversimplified. Using Harmonic Serialism (HS; Prince & Smolensky 1993/2004; McCarthy 2000, 2016) as the theoretical backdrop, I will show that it is possible to argue within a particular analytical and typological context for an analysis that is top-down in the sense just identified, but in so doing, I will challenge the idea that it is possible to label a language or stress pattern as ‘top-down’ in a primarily empirical or atheoretical way.9,10 To some extent, this conclusion may already be deduced from the literature. For example, the stress pattern of Tübatulabal (Uto-Aztecan; Voegelin 1935) is as follows: stress falls on the final syllable and every other mora counting backward from the final; heavy syllables are stressed and restart the mora count. Representative examples are given in (4)a. This pattern has been analyzed by some with iambs, as in (4)b, and by others with moraic trochees, as in (4)c. Although Voegelin (1935: 75) hedges about which syllable is the primary stress, it has been traditionally interpreted to be on the final syllable (see Hayes 1995: 265 for discussion).

9. Two anonymous reviewers question whether this statement is not already self-evident. Indeed, practically all labels (e.g., iamb/trochee, quantity-(in)sensitive, etc.) are couched within a specific set of assumptions and theoretical framework. However, I believe making this point explicit is still valuable, particularly in this case, because doing so reveals the specific issues underlying the use of the term and also allows us to identify differences among superficially-similar models. While it is correct to recognize that terms like ‘top-down’ have no meaning outside of a particular framework, it is also valuable to identify whether or not, or to what extent, the use of the term tends to correlates with particular empirical features and whether those features signal the same kinds of analyses in different frameworks. 10. A reviewer also asks what this has to do with theories of phonological cognition. Although I make no particular claims regarding cognition in the text, my assumption is that theories of typological variation provide an indirect way to examine human linguistic potential. In this sense, character- izing a typology of stress patterns, and being clear about the various empirical and theoretical issues therein, implicitly furthers this goal. 46 CatJL 18, 2019 Kathryn Pruitt

(4) Tübatulabal stress (examples from Voegelin 1935)

a. ha.nìː.lá wɪ̀.taŋ.hà.ta.làː.ba.ʦú ‘the house (obj.)’ ‘away from the Tejon Indians’ b. Iambic analysis (Hayes 1981; Crowhurst 1991) (ha.ˌniː)(ˈla) (ˌwɪ)(taŋ.ˌha)(ta.ˌlaː)(ba.ˈʦu) c. Trochaic analysis (Kager 1989; Hayes 1995) ha(ˌniː)(ˈla) (ˌwɪ.taŋ)(ˌha.ta)(ˌlaː)ba(ˈʦu)

The iambic analysis can be derived bottom-up by building right-to-left iambic feet of the form (LˌL), (LˌH), or (ˌL), and assigning primary stress to the rightmost one. The trochaic analysis, on the other hand, introduces a difference between primary and secondary stress feet, with only the former occupying a monomoraic foot, (ˈL). To account for this, the final primary stress has to be assigned first, as initially proposed by Kager (1989: 134) and taken up by Hayes (1995: 264). Tübatulabal thus demonstrates that the same stress pattern may be ‘top-down’, or not, depending on factors external to the location of stresses themselves. Two additional cases are examined in this paper. The first of these is Cahuilla (§3), a language analyzed as ‘top-down’ by Hayes (1995) and subsequently classi- fied as such by others (e.g., Bailey 1995; Kager 1995; McGarrity 2003). However, we will see that the top-down analysis is only needed under Hayes’s (re)interpretation of the source data. In a different light, the Cahuilla stress pattern is straightforwardly accounted for with a bottom-up analysis in HS. The second case study returns to Yine (§4). We will see that despite the independence of primary stress in Yine, both bottom-up and top-down analyses are readily available in HS using common stress constraints. However, the two analyses make different typological predictions, and it is therefore only on these grounds that one analysis can be preferred over another. The two case studies provide a detailed exemplification of the central theme of this paper: that apparent primary stress independence is not sufficient grounds to justify calling a language ‘top-down’. Informed by these case studies, the paper culminates in a precise statement of the empirical criteria that would favor a top-down analysis in Harmonic Serialism. Before moving on to the case studies, the next section (§2) presents some background on HS and defines ‘top-down’ and ‘bottom-up’ in this theory, with brief comparisons to rule-based theories of stress.

2. Defining top-down and bottom-up The discussion in this paper is framed in terms of Harmonic Serialism (HS), a version of OT with a series of optimizations that iteratively apply to select an optimal output. Similar to (most) rule-based phonology, HS is a model of grammar that assumes a phonological underlying form is mapped to a surface form in a series of discrete steps, with one operation applying at a time. But like classical Optimality Theory (OT), the steps are determined not by directed rewrite rules, Revisiting Top-Down Primary Stress CatJL 18, 2019 47 but by a competition among candidates that is settled on the basis of a hierarchy of ranked constraints. HS provides a relevant framework for the argument of this paper because it is serial, meaning the relative order of primary and secondary stress assignment must be explicitly defined, and because it is constraint-based, which permits many of the insights of classical OT to be retained. These include the use of violable constraints and the connection between individual analyses and typological predictions. In fact, as we will see in this paper, it is precisely the shift from rule-based theories to constraint-based HS that reveals the shaky status of ‘top-down’ as an empirical category, rather than a theory-dependent analytical one. In this section I begin by providing more background on HS. I then define top- down and bottom-up modes of parsing in HS, with comparisons to the rule-based theory of Hayes (1995).

2.1. Harmonic Serialism An HS derivation begins with an underlying form, which is input to the grammar as in classic OT, but the candidate generation function (Gen) is restricted to producing candidates that differ from the input by the application of only one operation, contra classic, parallel OT, where Gen produces candidates that differ in any and all possible ways from an input. A winner is chosen among the HS candidates on the basis of a hierarchy of violable constraints, as in the Eval component of parallel OT. In another departure from classic OT, however, this output is then fed back into the grammar as an intermediate input, from which an additional set of one-operation candidates is produced and compared. The derivation continues in this way, looping between Gen and Eval. The derivation converges when the input to an iteration is chosen as its own output; this indicates that no additional single operations improve the harmony of the form (as determined by the constraint ranking). In general, the definition of Gen in HS is a question of ongoing research (McCarthy 2010). For stress, it has been argued that the building of one foot counts as a single operation (Kimper 2011; Pruitt 2010, 2012) and that metrical structure-building is a separate operation from deletion, epenthesis, and feature- changing (McCarthy 2008; Jesney 2011; Staubs 2013; Elfner 2016). The steps of a stress derivation in HS thus tend to look similar to those of rule-based metrical theory, though the conditions leading to the derivation are different. In HS, a whole-word metrical parse can only be optimal if each of its component feet were selected as the best possible foot at some iteration and as more harmonic than the outcome of applying other segmental and prosodic operations. I will also generally assume, following Pruitt (2010, 2012), that foot structure cannot be altered once it is built. This assumption can be derived from the harmonic improvement imperative of Harmonic Serialism in many cases (Pruitt 2012: 9), as existing foot structure has been selected as optimal at previous stages of a derivation (and may be absent from underlying forms, according to McCarthy & Pruitt 2013). 48 CatJL 18, 2019 Kathryn Pruitt

The tableau in (6) below will illustrate a derivation of Pintupi’s stress pattern, setting aside for the moment the distinction between primary and secondary stress. This illustration assumes standard stress constraints including Parse-σ, Trochee/Iamb, and FtBin (Prince & Smolensky 1993/2004) and generalized alignment (McCarthy & Prince 1993). Kager (1999) provides a summary of this theory of stress constraints in parallel OT, and Pruitt (2012) defends the use of quadratic alignment constraints in HS. The constraints are defined in (5).

(5) Stress constraints

a. Parse-σ: Assign one violation mark for every unparsed syllable.

b. Trochee: Assign one violation mark for every foot whose head is not at its left edge.

c. FtBin: Assign one violation mark for a foot that is less than two morae.

d. AllFtL/R: For each foot in a word, assign one violation mark for every syllable intervening between the left/right edge of the foot and the left/ right edge of the word.

At the first iteration of stress assignment, all possible ways of building one foot are considered. The tableau in (6) shows representative candidates, with (a) no feet, (b) a left-aligned trochee, (c) a left-aligned iamb, and (d) a right- aligned trochee. The high rank of Trochee ensures that a trochaic foot wins out, and the ranking AllFtL >> AllFtR determines that it will be at the left edge, so (b) is the winner at the first step. The ranking of Parse-σ over both alignment constraints is necessary to establish iterative footing, which continues until the fourth iteration. At that point, there is one remaining unparsed syllable, but footing it runs afoul of high-ranked FtBin, so the derivation converges on candidate (i).11

11. An anonymous reviewer asks whether we might consider foot-building as decomposed into two separate operations: one that builds constituent foot structure and another than assigns headedness within that structure (analogous to the separation of footing and primary stress assignment that we will see in the bottom-up derivations). This alternative certainly deserves to be considered further, but doing so here would require an elaborated theory of metrical representations and constraints. Proposals that decouple constituency and headedness have been made in serial rule-based (Crowhurst & Hewitt 1995b) and parallel constraint-based (Hyde 2001, et seq.) frameworks, but I am not yet aware of similar extant proposals within Harmonic Serialism. Revisiting Top-Down Primary Stress CatJL 18, 2019 49

(6) Pintupi stress in HS: [ʈí.ɭi.ɾì.ŋu.làm.pa.t̻u] ‘the fire for our benefit flared up’

/ʈi.ɭi.ɾi.ŋu.lam.pa.t̻u/ FtBin Trochee Parse-σ AllFtL AllFtR 1st iteration a. ʈi.ɭi.ɾi.ŋu.lam.pa.t̻u 7 W L b. → (ˈʈi.ɭi).ɾi.ŋu.lam.pa.t̻u 5 5 c. (ʈi.ˈɭi).ɾi.ŋu.lam.pa.t̻u 1 W 5 5 d. ʈi.ɭi.ɾi.ŋu.lam.(ˈpa.t̻u) 5 5 W L 2nd iteration e. (ˈʈi.ɭi).ɾi.ŋu.lam.pa.t̻u 5 W L 5 L f. → (ˈʈi.ɭi)(ˈɾi.ŋu).lam.pa.t̻u 3 2 8 3rd iteration g. (ˈʈi.ɭi)(ˈɾi.ŋu).lam.pa.t̻u 3 W 2 L 8 L h. → (ˈʈi.ɭi)(ˈɾi.ŋu)(ˈlam.pa).t̻u 1 6 9 4th iteration (Convergence) i. → (ˈʈi.ɭi)(ˈɾi.ŋu)(ˈlam.pa).t̻u 1 6 9 j. (ˈʈi.ɭi)(ˈɾi.ŋu)(ˈlam.pa)(ˈt̻u) 1 W L 12 W 9

Admitting HS as a theory of stress has advantages over classic OT and over rule-based theories, which are explored in other work (e.g., Pruitt 2010; see McCarthy 2016 for a summary of some results). But it also brings with it further questions about the definition of Gen, in particular, how and when the distinction between primary and secondary stresses is determined. It is possible to imagine two definitions of Gen: one where the primary stress is treated in a top-down fashion (§2.2) or one where the primary stress is assigned in a bottom-up way (§2.3). These are now discussed in turn.

2.2. Top-down A logical relationship between primary and secondary stress in HS is to assume that the first foot is the primary stress foot by default, with subsequent steps building secondary stress feet on the remaining syllables. This is more or less the view assumed (implicitly or explicitly) by previous work in HS including McCarthy (2008), Elfner (2016), and McCarthy, Pater & Pruitt (2016), and, in the terms used in the introduction, could be considered a top-down model of primary stress assignment. The serial tableau in (8) below demonstrates how Pintupi’s primary stress is modeled when parsing is top-down. Because primary stress is assigned right away (and because Pintupi’s primary stress is left-aligned along with other 50 CatJL 18, 2019 Kathryn Pruitt feet), the derivation is essentially identical to the one in (6) above, which derived left-aligned trochees without reference to primary stress. The primary stress constraints I assume are defined in (7). In (8), these constraints are set apart in the tableau to show that in this case they are not necessarily active in selecting the correct derivation.

(7) Primary stress constraints

a. Headedness(PWd): Assign a violation mark for a prosodic word (PWd) that does not have a primary stress.

b. AlignHdL/R: Assign a violation mark for each syllable intervening between the primary stress syllable and the left/right word edge.

(8) Top-down derivation of Pintupi primary stress ) d σ L R - L R t t d d in F F (PW H H B ll ll d l l t a r se

/ʈi.ɭi.ɾi.ŋu.lam.pa.t̻u/ F P A A H A A 1st iteration a. ʈi.ɭi.ɾi.ŋu.lam.pa.t̻u 7 W L 1 W L b. → (ˈʈi.ɭi).ɾi.ŋu.lam.pa.t̻u 5 5 6 c. ʈi.ɭi.ɾi.ŋu.lam.(ˈpa.t̻u) 5 5 W L 5 W 1 L 2nd iteration d. (ˈʈi.ɭi).ɾi.ŋu.lam.pa.t̻u 5 W L 5 L 6 e. → (ˈʈi.ɭi)(ˌɾi.ŋu)lam.pa.t̻u 3 2 8 6 f. (ˈʈi.ɭi).ɾi.ŋu.lam.(ˌpa.t̻u) 3 5 W 5 L 6 3rd iteration g. (ˈʈi.ɭi)(ˌɾi.ŋu).lam.pa.t̻u 3 W 2 L 8 L 6 h. → (ˈʈi.ɭi)(ˌɾi.ŋu)(ˌlam.pa).t̻u 1 6 9 6 4th iteration (Convergence) i. → (ˈʈi.ɭi)(ˌɾi.ŋu)(ˌlam.pa).t̻u 1 6 9 6 j. (ˈʈi.ɭi)(ˌɾi.ŋu)(ˌlam.pa)(ˌt̻u) 1 W L 12 W 9 6

Since primary stress is automatically assigned to the candidates for the first foot, any constraints that are specific to primary stress can also have a say in where that foot is placed. In this example, primary-stress-specific constraints were set apart because the primary stress foot matches the other feet in Pintupi in all relevant properties (left-aligned, trochaic, quantity-insensitive, and binary). But when a con- Revisiting Top-Down Primary Stress CatJL 18, 2019 51 flicting primary-stress-specific constraint is higher-ranked than a general stress constraint, a top-down derivation yields a stress pattern with some asymmetry between primary and secondary stresses (as in Yine, for example, where the primary and secondary stresses are aligned to opposite edges of the word; see section 4). This implementation of ‘top-down’ differs from that of Hayes (1995) and other metrical theories that employ grids or bracketed grids for primary stress. For Hayes and others, top-down stressing involves the application of the End Rule—the rule responsible for placing a word-level grid mark (Prince 1983)—before feet are constructed (Hayes 1995: 61). The End Rule does not build foot-level constituents, but in a top-down derivation a foot head is entailed ‘under’ the word-level grid mark during foot construction by the Continuous Column Constraint (Hayes 1995: 116-117). A top-down derivation of Pintupi would thus look like that shown in (9) according to this theory; the derivation is hypothetical because when systems can be analyzed either top-down or bottom-up, the bottom-up one is usually assumed.

(9) Hypothetical top-down derivation for Pintupi in the theory of Hayes (1995)

Wd-level ( x ) ( x ) ( x ) ( x ) Ft-level ( x . ) ( x . )( x . ) ( x . )( x . ) ( x . ) ʈiɭi ɾi ŋu lam pa t ̻u → ʈiɭi ɾiŋulam pa t ̻u → ʈiɭi ɾi ŋu lam pa t ̻u → ʈiɭi ɾi ŋu lam pa t ̻u

To summarize the difference between the two approaches, for Hayes a top- down derivation means that primary stress is designated on an initial or final syllable (modulo potential extrametricality), but it is then up to the regular parsing rules to govern what kind of foot is constructed there. In contrast, as just noted, the top-down HS model just defined will begin a stress derivation by building a foot that is the optimal outcome of both the regular parsing constraints and any constraints specific to primary stress. Although it is possible to imagine a definition of Gen in HS that would bring it closer to that of Hayes’s theory, this has not yet been explored.

2.3. Bottom-up Alternatively, it is possible to define Gen in a way that delivers only bottom-up derivations. In this case, the construction of individual feet and the designation of one as the primary stress foot are distinct operations. This version of Gen can be called bottom-up, since some metrical structure must be built before the primary stress is formally identified. The derivation in (10) illustrates how this definition of Gen would work to derive the same stress pattern as above. At the first iteration, only candidates with non-primary stress are available, so the only constraints relevant for selecting among them are the general stress constraints. The derivation builds left-aligned trochees until the 4th iteration, much like the derivation in (6) above. However, at this point candidates with primary stress are included, and candidate (k) wins because it has promoted the initial syllable to primary, in simultaneous satisfaction of Head(PWd) and AlignHdL. The primary stress 52 CatJL 18, 2019 Kathryn Pruitt constraints are again set apart for expositional clarity, though there is a potential interaction between them and the general parsing constraints, which is described at the end of this section.

(10) Bottom-up derivation of Pintupi stress in HS ) d σ L R - L R t t d d in F F (PW H H B ll ll l l d t a r se

/ʈi.ɭi.ɾi.ŋu.lam.pa.t̻u/ F P A A H A A 1st iteration a. ʈi.ɭi.ɾi.ŋu.lam.pa.t̻u 7 W L 1 b. → (ˌʈi.ɭi).ɾi.ŋu.lam.pa.t̻u 5 5 1 c. ʈi.ɭi.ɾi.ŋu.lam.(ˌpa.t̻u) 5 5 W L 1 2nd iteration d. (ˌʈi.ɭi).ɾi.ŋu.lam.pa.t̻u 5 W L 5 L 1 e. → (ˌʈi.ɭi)(ˌɾi.ŋu)lam.pa.t̻u 3 2 8 1 f. (ˌʈi.ɭi).ɾi.ŋu.lam.(ˌpa.t̻u) 3 5 W 5 L 1 3rd iteration g. (ˌʈi.ɭi)(ˌɾi.ŋu).lam.pa.t̻u 3 W 2 L 8 L 1 h. → (ˌʈi.ɭi)(ˌɾi.ŋu)(ˌlam.pa).t̻u 1 6 9 1 4th iteration i. (ˌʈi.ɭi)(ˌɾi.ŋu)(ˌlam.pa).t̻u 1 6 9 1 W L j. (ˌʈi.ɭi)(ˌɾi.ŋu)(ˌlam.pa)(ˌt̻u) 1 W L 12 W 9 1 W L k. → (ˈʈi.ɭi)(ˌɾi.ŋu)(ˌlam.pa).t̻u 1 6 9 6 l. (ˌʈi.ɭi)(ˌɾi.ŋu)(ˈlam.pa).t̻u 1 6 9 4 W 2 L In Hayes’s (1995) theory, an iterative foot construction rule creates feet in a directional sweep of the word, obeying various language-specific parameters and universal principles. Then, the End Rule selects the leftmost or rightmost available syllable on which to place the word-level grid mark. The Continuous Column Constraint ensures in this case that only a syllable which already has a foot-level grid mark can be a contender for the word-level grid mark (Hayes 1995: 36).

(11) Bottom-up derivation of Pintupi [ʈí.ɭi.ɾì.ŋu.làm.pa.t̻u] (Hayes 1995: 63) Wd-level ( x ) Ft-level ( x . ) ( x . )( x . ) ( x . )( x . )( x . ) ( x . )( x . ) ( x . ) ʈi ɭi ɾi ŋu lam pa t ̻u → ʈi ɭi ɾi ŋu lam pa t ̻u → ʈi ɭi ɾi ŋu lam pa t ̻u → ʈi ɭi ɾi ŋu lam pa t ̻u Revisiting Top-Down Primary Stress CatJL 18, 2019 53

A difference between the two theories is that in HS, operations compete at each iteration, so candidates with primary stress are available as soon as there is a foot in the (local) input on which to place it. This means that, contrary to Hayes’s theory, the bottom-up Gen in HS could produce a derivation which builds a foot at step 1, assigns it primary stress at step 2, and continues selecting (non-primary-stress) feet at the next steps. Such a derivation is predicted when the constraint mandating a primary stress (here, Head(PWd)) outranks the constraint demanding additional feet (here, Parse-σ). When the opposite ranking holds, foot construction will generally precede primary stress assignment, as in the bottom-up illustration given just above in (10). Potential implications of this difference are not explored here, but are merely described as a point of reference for comparing the two theories.

2.4. Discussion A priori, both alternatives—top-down Gen or bottom-up Gen—are theoretically coherent, though each faces potential challenges with different types of stress systems. The top-down Gen defined in §2.2 has the same problem as other top-down models: a difficulty in accounting for primary stress placement in languages like MalakMalak (discussed in section 1), where primary stress is determined in reference to foot structure rather than primarily to word edges. A top-down model in HS would thus require elaboration to admit an analysis of those systems. In contrast, the bottom-up model can straightforwardly handle MalakMalak (by building right-aligned trochees and selecting the leftmost one for primary stress), but could face challenges from the traditionally ‘top-down’ languages, where primary stress seems to obey generalizations different from those of secondary stresses. These same challenges are what lead Hayes (1995), and some others in the rule-based literature, to the conclusion that languages may select either parsing method parametrically. However, there is no way to replicate this parameter in HS: allowing both top-down and bottom-up primary stress operations in Gen with the intention of letting languages ‘choose’ via ranking is an untenable solution (Pruitt 2012: 119-123, in prep).12 A fuller comparison of top-down and bottom-up definitions of HS Gen in light of the attested typology of primary stress patterns is pursued in other work. To capture both types of stress systems, Pruitt (2012, in prep) proposes a theory of primary stress that enforces top-down parsing by default13 and allows primary stress movement to account for languages like MalakMalak. The goal here is not to

12. Since languages cannot have different Gens, this would mean defining a Gen that produces candidates that assign primary stress right away and candidates that do not. But as demonstrated in Pruitt (2012), a Gen defined in this way makes the implausible typological prediction that languages may assign primary stress in some words but not others (a prediction which is termed “non-uniform culminativity” in that work). Individually, top-down and bottom-up Gens can be defined to avoid the prediction, but a Gen with both top-down and bottom-up options cannot (see Pruitt 2012: Chs 3-4). 13. In Pruitt (2012) this is accomplished with a wholesale restriction on primary stress constraint definitions, while in Pruitt (in prep) it is simply built into the definition of Gen. 54 CatJL 18, 2019 Kathryn Pruitt duplicate those efforts but instead to shine a light on the language evidence itself and on preconceived notions a ‘top-down’ language. Toward this end, the next section turns to a discussion of Cahuilla, the first case study.

3. The role of theory in identifying asymmetries This section takes the stress pattern of Cahuilla (Uto-Aztecan; Seiler 1965, 1967, 1977) as a case study in top-down stress. Hayes (1995) analyzes Cahuilla as top- down due to an asymmetry in the primary and secondary stresses in whether they are permitted to occur as degenerate feet, similar to the Tübatulabal example in section 1, and subsequent references to Cahuilla in the literature have often continued to refer to it as top-down or primary-stress-first (e.g., Bailey 1995; Kager 1995; McGarrity 2003). In the case of Cahuilla, the asymmetry hinges on a particular interpretation of the source data. If we take Seiler’s (1965, 1967, 1977) description of the stress pattern at face value, however, no asymmetry is present and a bottom- up analysis is possible. I will suggest that constraint inviolability is the fundamental issue at work in motivating the asymmetric interpretation and therefore the top- down analysis. In particular, when we shift from rule-based serial stress derivations (with typically-inviolable constraints) to HS (which inherits constraint violability from classic OT), we also see that data interpretations, and top-down arguments based on them, shift as well. Section 3.1 gives the stress pattern of Cahuilla according to Seiler’s description and provides a bottom-up analysis in HS. Section 3.2 summarizes Hayes’s arguments for his reinterpretation of the data, which introduces a primary/secondary stress asymmetry, and it shows how this motivates his top-down analysis. Section 3.3 presents counterarguments to Hayes’s interpretation of the data and discusses the role of one’s theoretical framework in establishing primary/secondary stress asymmetries.

3.1. Cahuilla stress and HS analysis Cahuilla14 is a Uto-Aztecan language of Southern California (Seiler 1977) whose stress pattern is analyzed by Hayes (1995: 132-140) with moraic trochees, where feet typically consist either of a single heavy syllable, (ˈH), or two light syllables (ˈLL). According to Seiler (1965, 1967, 1977), primary stress usually falls on the first syllable of the root regardless of its weight, and secondary stress alternates by mora rightward (and leftward, if sufficient prefixal material is present; Seiler 1977: 27). Long vowels and diphthongs count as heavy, as do syllables closed by a glottal stop.15 The forms in (12) illustrate the patterns, which are shown along

14. The historical and contemporary sources I was able to access use the language name Cahuilla, which I follow here, but an anonymous reviewer has informed me that Ivilyuat is the language name used by its speakers. The number of speakers now probably numbers fewer than ten (as documented in Eberhard, Simons & Fennig 2019). 15. Although this follows Hayes’s (1995: 132-133) text description of Cahuilla syllable weight, the discussion of weight by Seiler (1965, 1967, 1977) is more complex. See footnotes 22 and 23 for additional information. Revisiting Top-Down Primary Stress CatJL 18, 2019 55 with the quantity profiles and associated foot types they entail. Forms (a)-(c) show alternation in the absence of quantity, and (d)-(j) show that the alternation of stress is reset after a heavy syllable. In (a)-(d) primary stress occupies a (ˈLL) foot, while in (e)-(h) it is instead (ˈH). Finally, (i) and (j) are crucial forms for establishing that the primary stress remains on the first syllable of a root even when it results in a monosyllabic light syllable foot, (ˈL). These forms will be discussed in detail below.

(12) Cahuilla stress (Seiler 1965, 1967, 1977) a. kísiʎ ‘chicken hawk’ (ˈLL) (1977: 36) b. súvalwàl ‘sparrow’ (ˈLL)(ˌL) (1965: 53) c. sásmatnèkʧem ‘slim canes’ (ˈLL)(ˌLL) (1967: 140) d. táxmuʔàʔtì ‘the song, obj. case’ (ˈLL)(ˌH)(ˌL) (1977: 31/33/57) e. táːtwàl ‘blind’ (ˈH)(ˌL) (1977: 35) f. ɲáʔʧèh ‘sit down’ (ˈH)(ˌL) (1965: 52) g. qáːnkìʧem ‘palo verde, pl.’ (ˈH)(ˌLL) (1977: 27) h. háʔtìsqal ‘he is sneezing’ (ˈH)(ˌLL) (1965: 52) i. súkàʔtì ‘the deer, obj. case’ (ˈL)(ˌH)(ˌL) (1977: 28) j. ménìʔlì ‘the moon, obj. case’ (ˈL)(ˌH)(ˌL) (1965: 52)

A bottom-up analysis of this pattern is possible in HS with a ranking that parses words into trochees (Trochee >> Iamb), iterates left-to-right (Parse-σ >> AllFtL >> AllFtR), allows degenerate feet (Parse-σ >> FtBin)16, and assigns primary stress to the leftmost foot (AlignHdL >> AlignHdR). The derivational tableaux in (13) and (15) show the HS analysis with, respectively, words of the form /LLL/, parsed as (ˈLL)(ˌL), and /LHL/, parsed as (ˈL)(ˌH)ˌ(L). In (13) we see that /su.val.wal/ ‘sparrow’ is parsed into a left-aligned disyllabic trochee at the first iteration (candidate c), and a monosyllabic foot is added at the second (candidate e). At the third iteration, candidates for primary stress placement are considered, and (g) wins because of AlignHdL.17

16. Prefixal strings do not seem to tolerate degenerate feet, so this ranking is appropriate for the stem+suffix domain only. Something more will need to be said for a full analysis incorporating prefixes, which may ultimately require a cyclic derivation. As noted in the text above, prefixes are also exceptional in the reported directionality of their stress alternation. 17. Two rankings deserve additional mention: (i) Parse-σ over Hd(Pwd) determines that parsing completely precedes primary stress assignment. The opposite ranking predicts a derivation where primary stress is assigned at the second iteration of stress (i.e., as soon as an input foot is available to place primary place stress on), but both derivational paths achieve the same outcome in this case. (ii) The ranking of AllFtL >> FtBin determines that parsing will proceed from the left edge regardless of FtBin violations, which becomes relevant in (15). If this ranking is inverted, the derivation instead builds feet on H syllables first and then goes back to build (L) where needed later. But again, the outcome of the derivations is the same. 56 CatJL 18, 2019 Kathryn Pruitt

(13) Bottom-up HS derivation of [sú.val.wàl] ‘sparrow’

/su.val.wal/ AlHdL Parse-σ AllFtL Hd(PWd) FtBin 1st iteration a. su.val.wal 3 W 1 b. (ˌsu).val.wal 2 W 1 1 W c. → (ˌsu.val).wal 1 1 2nd iteration d. (ˌsu.val).wal 1 W L 1 L e. → (ˌsu.val)(ˌwal) 2 1 1 3rd iteration f. (ˌsu.val)(ˌwal) 2 1 W 1 g. → (ˈsu.val)(ˌwal) 2 1 h. (ˌsu.val)(ˈwal) 2 W 2 1

The analysis shown so far also predicts the pattern of the (ˈL)(ˌH)(ˌL) form [sú.kàʔ.tì] ‘the deer, obj. case’, with one additional assumption. Consistent with other moraic trochee stress systems, a constraint is needed to rule out (ˈLH) and (ˈHL) as possible trochaic feet (Prince 1990; Prince & Smolensky 1993/2004). I make no particular commitment to the nature of such a constraint and will assume the formulation in (14) below for the purposes of this analysis.

(14) Ft≤2μ: Assign a violation mark for any foot larger than two morae.

With the ranking Ft≤2μ >> Parse-σ >> FtBin, we correctly predict the stress pattern of /su.kaʔ.ti/, as shown in (15). At the first iteration, a left-aligned monosyllabic foot is chosen (candidate b) because it satisfies AllFtL (unlike candidate c) and does not violate Ft≤2μ (unlike candidate d); a violation of FtBin is tolerated. At the second iteration, a foot is built on the middle heavy syllable (candidate f) because doing so satisfies FtBin and does not violate Ft≤2μ (unlike candidate g). At the third iteration, the remaining syllable is parsed into a monosyllabic foot (candidate i) due to Parse-σ >> FtBin, and at the fourth, the leftmost foot is selected as the primary stress (candidate k). Revisiting Top-Down Primary Stress CatJL 18, 2019 57

(15) Bottom-up HS derivation of [sú.kàʔ.tì] ‘the deer, obj. case’

/su.kaʔ.ti/ AlHdL Ft≤2μ Parse-σ AllFtL Hd(PWd) FtBin 1st iteration a. su.kaʔ.ti 3 W 1 L b. → (ˌsu).kaʔ.ti 2 1 1 c. su.(ˌkaʔ).ti 2 1 W 1 L d. (ˌsu.kaʔ).ti 1 W 1 L 1 L 2nd iteration e. (ˌsu).kaʔ.ti 2 W L 1 1 f. → (ˌsu)(ˌkaʔ).ti 1 1 1 1 g. (ˌsu)(ˌkaʔ.ti) 1 W L 1 1 1 3rd iteration h. (ˌsu)(ˌkaʔ).ti 1 W 1 L 1 1 L i. → (ˌsu)(ˌkaʔ)(ˌti) 3 1 2 4th iteration j. (ˌsu)(ˌkaʔ)(ˌti) 3 1 W 2 k. → (ˈsu)(ˌkaʔ)(ˌti) 3 2 l. (ˌsu)(ˌkaʔ)(ˈti) 2 W 3 2

The Cahuilla data and its analysis show that degenerate feet, (ˈL), are required in two positions: (i) initially, if the primary stress occupies a light syllable immediately followed by a heavy syllable, and (ii) finally, if a light syllable would otherwise be ‘left over’ at the end of the parse. The HS analysis treats the two instances essentially the same, both arising as the optimal outcome when Parse-σ (and AllFtL) outranks FtBin. Primary stress is later assigned to the initial syllable (of the root), regardless of the type of foot built there, (ˌL), (ˌLL), or (ˌH). In other words, the HS analysis encounters no difficulty with a bottom-up analysis of the primary stress degenerate foot specifically because the ranking must allow for degenerate feet anyway (i.e., Parse-σ >> FtBin).

3.2. Hayes’s (1995) top-down analysis In contrast to the HS analysis we have just provided, degenerate feet are not typically permitted in the theory of Hayes (1995), and his analysis of Cahuilla deals with the two instances (initial and final) in disparate ways. The stress transcribed on certain final syllables in the source data is argued to be non-metrical final 58 CatJL 18, 2019 Kathryn Pruitt lengthening, so no degenerate foot need be constructed there. Hayes (1995: 137) presents three arguments this effect. First, citing Seiler (1965), he notes that that a correlate of stress in Cahuilla is a change in pitch, but this is never seen in word- final syllables. Second, citing Seiler (1957), Hayes notes that vowel allophony (for /a/ and /e/) which normally accompanies stress alternation is absent in word-final syllables, with only the stressless allophones occurring in final position. And finally, Hayes points to stress shift seen with some monosyllabic noun stems to suggest that degenerate feet are marked in Cahuilla, drawing on data from Seiler (1977). (These arguments are revisited in section 3.3 below.) The initial (primary stress) degenerate foot, however, is assumed to be genuine. This introduces an asymmetry between primary and secondary stress, which must then be derived with top-down parsing, as we now describe. In Hayes’s theory, degenerate feet are normally avoided, which is implemented in large part by the following constraint on foot-building.

(16) The Priority Clause If at any stage in foot parsing the portion of the string being scanned would yield a degenerate foot, the parse scans further along the string to construct a proper foot where possible. (Hayes 1995: 95)

Cahuilla’s stress pattern is directionally left-to-right from the first syllable of the root. Thus, in forms such as /sukaʔti/, the first two syllables scanned are the light [su] and heavy [kaʔ]. An (ˈLH) foot as in (17)a is impossible a priori in this theory, leaving L(ˈH) and (ˈL), shown in (17)b and (17)c, as the only options in principle.

(17) Possible first feet in /sukaʔti/, according to Hayes (1995) a. (su.kaʔ)ti never allowed, no (ˈLH) trochees permitted b. su(kaʔ)ti typically-preferred due to Priority Clause c. (su)kaʔti actual Cahuilla outcome, but violates Priority Clause

The Priority Clause is intended to settle the competition in favor of an unparsed light syllable, i.e., (17)b. But since Cahuilla appears to violate the Priority Clause, preferring (17)c, another factor must be present to override it. This is achieved by allowing the End Rule to occur before regular foot parsing. This entails a foot—degenerate or not—in subsequent foot building as a consequence of the Continuous Column Constraint, as discussed in section 2.2. The derivation in (18)a illustrates the top-down analysis of Cahuilla stress for the form /sukaʔti/ ‘deer, obj’, which is parsed as (sú)(kàʔ)ti according to Hayes’s interpretation of the data. The derivation in (18)b shows what would go wrong if parsing were instead bottom-up, with regular foot building preceding the application of the End Rule. In the latter case, the first syllable remains unparsed altogether because the Priority Clause is normally inviolable. Revisiting Top-Down Primary Stress CatJL 18, 2019 59

(18) Result of top-down vs. bottom-up parsing for Cahuilla18 a. Top-down derivation; Result: [(sú)(kàʔ)ti] Wd-level (x ) (x ) (x ) Ft-level (x) (x) (x) su. kaʔ .ti → su. kaʔ .ti → su. kaʔ .ti → su. kaʔ .ti b. Bottom-up derivation; Result: *[su(káʔ)ti] Wd-level ( x ) Ft-level (x) (x) su. kaʔ .ti → su. kaʔ .ti → su. kaʔ .ti To summarize, the top-down analysis of Cahuilla stress by Hayes (1995) is motivated by the fact that the primary stress may occupy a degenerate foot, which would otherwise never be constructed. The degenerate final foot in some words is reinterpreted as final lengthening, so regular parsing simply does not build degenerate feet and a bottom-up analysis consequently fails.

3.3. Final stress? The status of the final stress is crucial to motivating a primary-stress-first analysis of Cahuilla because this is what determines whether the primary and secondary stresses show an asymmetry—a necessary precondition for motivating a top-down analysis. If instead the final stress is metrical (that is, if it is really stressed), then it too requires a degenerate foot, and the regular parsing algorithm must be capa- ble of building degenerate feet. As we saw in section 3.1, this would mean that a bottom-up analysis is possible. Thus, we will now examine in some detail the arguments Hayes (1995: 137) gives for his interpretation of the final stress.

Intonation. Hayes cites Seiler (1965: 52) for the observation that stress generally correlates with a change in pitch in Cahuilla, which does not occur in word-final syllables. Specifically, Seiler notes that stress involves a rise in pitch on the stressed syllable and a fall back down to a following unstressed syllable, with magnitudes of “about a fifth” for primary stress and “about a third” for secondary, though “[w]ord-final moras always have low pitch” (1965: 52). In the corresponding description of stress realization in Seiler (1977: 26) the pitch changes of primary and secondary stresses are reported in a similar way, but the final syllable is not mentioned, though it is difficult to draw any conclusions from this omission. We may also consider the brief discussion of intonation in Seiler (1977: 25-26). Seiler

18. A bit of technicalia is suppressed here. Hayes permits languages to parametrically employ a “weak” ban on degenerate feet; this permits their construction at the end of a parse, where the Priority Clause is irrelevant. However, at the end of the derivation, an across-the-board erasure applies to any degenerate foot which has not subsequently been assigned primary stress or repaired by another rule. Thus, the predicted outcomes are the same as those represented in the derivations in (18), i.e., with the final syllable remaining unparsed. 60 CatJL 18, 2019 Kathryn Pruitt indicates that word-final syllables can receive high pitch in service of interrogative intonation, which is “marked by a stress pattern full stress – unstressed – full stress distributed over the last three syllables of the predicate (verb) of the main clause; full stress is accompanied by high pitch, unstressed by low pitch” (1977: 25). Although there is no indication that the alignment of the pitch/stress targets of interrogative intonation are metrically-constrained (i.e., affected by word-level mora alternation), this does indicate that the pitch level of the final syllable may be conditioned by higher-level intonational factors. The intonational system is not discussed in further detail in the works I was able to consult (Seiler 1965, 1967, 1977), but since there is clearly some interaction between the word- and phrase- level prosody, Seiler’s observation of low pitch for final syllables is, arguably, not clear evidence against a metrical final stress.

Vowel allophony. Hayes’s second argument concerns vowel allophony. He cites Seiler (1957) for the observation that only stressless allophones of /a/ and /e/ may occur in final position. However, Seiler 1957 and 1977 both suggest a rather more complex system of vowel quality alternation. According to Seiler (1977: 29-35), Cahuilla’s vowel system has four phonemic vowel qualities, /i, e, a, u/,19 whose allophones (front, central, or back for /a/ and closed, half-open, or open for the others) are conditioned by a combination of factors that include stress, adjacent consonants, vowels in neighboring syllables, and word position (final vs. non-final). Interestingly, for example, when a high vowel, /i/ or /u/, receives primary stress, it is either closed or half-open, usually depending on the preceding consonant; subsequent phonemically-identical vowels in the same word will then dissimilate to produce an alternation between closed and half-open (p. 30), but the association with stress could go either way: the stressed vowels may be closed and unstressed vowels half-open; or the stressed vowels may be half-open and the unstressed vowels closed. The high vowels obey somewhat different generalizations from /e/, whose behavior is described as “altogether not too clear” (p. 31), though all three, /i, u, e/, are reported to exhibit “open” allophones in word-final position (i.e., [ɪ, ʊ, æ]; p. 31), which corresponds to neither stressed nor unstressed allophones word-internally. The phoneme /a/ shows a still different generalization, where its quality varies based on whether it is the only /a/ in the word (if yes, then it is pronounced as fronted) and otherwise based on stress (back if stressed, mid if unstressed) (p. 31-32), with no mention of a particular allophone in word-final position in Seiler (1977). In other words, I would argue that it is not possible to identify a direct correlation between stress and vowel quality more generally in Cahuilla, so the behavior of word-final vowels is not clearly indicative of their metrical status.

Stress shift with monosyllabic stems. The final argument rests on the behavior of some monosyllabic noun stems, where stress is seen to shift under prefixation and which Hayes suggests is evidence that Cahuilla treats monomoraic feet as marked,

19. And a phonemic length distinction, which is discussed separately by Seiler (1977) and involves a somewhat different generalization (p. 33-34). Revisiting Top-Down Primary Stress CatJL 18, 2019 61 rather than freely allowed. The data are as follows: when monomoraic nouns like the ones in (19)a receive a personal prefix to indicate possession, the primary stress shifts to the prefix as in (19)b; but when the root is at least bimoraic, as in (20)a, the stress does not shift, (20)b. The intuition is that the monomoraic nouns will surface as stressed when no other options are available, but a prefix makes available an alternative parse that avoids a degenerate foot.

(19) Stress shift (Seiler 1977: 33, 39) a. –na20 ‘father’ –ʔaʃ ‘pet’ b. ʧém-na ‘our father’ né-ʔaʃ ‘my pet’

(20) No stress shift (ibid.) a. –túʔat ‘flour’ –júwl21 ‘younger brother’ b. ne-túʔat ‘my flour’ ne-júwl ‘my younger brother’

However, these examples are difficult to draw conclusions from for two reasons. First, the lack of stress shift in [ne-júwl] in (20)b appears to be the only potential evidence that vowel-glide sequences are bimoraic, unlike other VC rimes in the language which are monomoraic (1977: 27), and this makes the argument partly circular.22 Relatedly, stress is reported to shift only in monosyllabic roots, whereas if the shift were metrically/quantitatively motivated we would also expect to see shift in polysyllabic words with a monomoraic primary stress foot, like [(sú) (kàʔ)(tì)].23 Second, this process is rather morphologically limited, as it is reported by Seiler (1977: 39) to apply just with noun roots (only one verb root exceptionally shifts its stress in comparable environments) and to occur only with “[p]ersonal prefixes of the class P1” (ibid). The doubtful quantitative status of vowel-glide sequences and the limited character of the generalization would seem not to provide clear evidence about the status of degenerate feet in Cahuilla in general.

20. Seiler does not mark stress on the monosyllabic bare forms here. I believe this represents a choice about the phonemic representation of the stems, as stressed monomoraic words are freely permitted as surface forms in Cahuilla (e.g., [pál] ‘water’, Seiler 1977: 25). 21. Hayes transcribes the [uw] in this example as [uː] (1995: 137), but Seiler is explicit that VG sequences are phonemically (and usually phonetically) distinct from VV sequences (1977: 34-35) and consistently gives the form as yuwl (1977: 33, 39). 22. Although vowel-glide sequences are categorized with long vowels for mora count in Seiler 1977 (p. 27), in at least one other work (1965), Seiler describes the moraic status of (pre-consonantal) vowel-glide sequences as “not entirely conclusive” (p. 53). No secondary-stress-marked examples are given to confirm their status in Seiler (1977), and counterexamples can be found among those that are provided (e.g., [pàʔ.máj.ʎu.qà.li.vè] ‘where she was giving birth to’, 1977: 57, where *[pàʔ.máj.ʎù.qa.lì.ve] would be expected if maj is bimoraic). 23. In addition, a CVʔ root should not show stress shift if glottal stops are moraic in coda position, but they apparently do, e.g., [hí-jeʔ] ‘his mother’ (Seiler 1977: 347, text no. 10, line 1), not *[hi-jéʔ]. This fact, along with Seiler’s characterization of glottal stop (and glides) as potentially moraic only in pre- and/or post-consonantal position, suggests that quantity in Cahuilla might fruitfully be reexamined altogether, but this is not pursued here. 62 CatJL 18, 2019 Kathryn Pruitt

On the basis of these considerations, there does not seem to be strong phonological evidence against Seiler’s report of final stress in the relevant forms. To cement this conclusion, we will note two additional pieces of information that favor an interpretation of the transcribed final prominence as stress. The first is that Seiler only marks final stress in words that end in an odd-numbered string of light syllables. This alone strongly suggests a metrical patterning. An alternative interpretation in terms of non-metrical final lengthening would need to explain why the same emphasis is not marked on all final syllables, regardless of where they fall in the alternating count. The second reason to prefer a metrical, or stress-based, interpretation of the final prominence derives from the fact that in many final sequences of a heavy syllable followed by a light syllable, the heavy syllable is derived by a morpho-phonological rule of glottal stop insertion, which creates the final stress (Seiler 1965, 1967, 1977: 54-58) and therefore motivates the degenerate foot in final position. For example, the form /táxmuʔat-i/ ‘the song, objective case’ surfaces as [táxmuʔàʔtì], with a glottal stop inserted before the final stop of the root (Seiler 1977: 57). Without this process we expect the stress pattern *[táxmuʔàti], which is parsed (ˈLL)(ˌLL), but glottalization forces the parse (ˈLL)(ˌH)(ˌL) instead. The process is widespread in the language. Seiler (1967: 137) claims that the motivation for the process may be “such that the suffix appended last in the word must receive a secondary stress”. Seiler (1977: 57) generalizes somewhat by suggesting that glottalization ensures adjacent stresses, which serves to signal the stem/affix boundary between the heavy and the light syllable. Although we cannot confirm or refute Seiler’s explanation of the process from the available information, it suggests in any case that Seiler felt confident that final syllables are genuinely stressed in such cases.

3.4. Discussion This section has presented a case study of a language where the interpretation of the stress description affects whether the primary stress is really independent from secondary stresses. Although Hayes’s interpretation of the Cahuilla stress data involves a primary/secondary stress asymmetry that favors a top-down analysis, this section has shown that it is plausible that no such asymmetry exists, in which case a bottom-up analysis is available. Before moving on, we can address the theoretical context in which the top- down analysis of Cahuilla was proposed, since it is this context that shapes the interpretation of the data in ways that introduce primary/secondary stress asymmetries. In light of data from a large number of languages showing that patterns seeming to require degenerate feet are rare, Hayes (1995: 86-105) takes the strong position that degenerate feet are nearly always avoided by rules that build metrical structure, hence the Priority Clause (see (16) above). When constraints are generally inviolable, as in Hayes’s and many other rule-based theories, the rational way to approach observed typological rarity is to determine whether it can be pushed to the logical extreme of typological absence, and in this vein Hayes ultimately proposes that secondary stress can never occupy a degenerate foot, though primary Revisiting Top-Down Primary Stress CatJL 18, 2019 63 stress may do so (often through top-down parsing like in Cahuilla24). Thus, a few purported cases of secondary stress degenerate feet, Cahuilla among them, are given plausible reanalyses that do not require degenerate feet. Under this theory, then, primary and secondary stress are fundamentally asymmetric in whether they permit degenerate feet. In contrast, the complete avoidance of degenerate secondary stress feet has not usually been taken up in OT-based work in metrical theory.25 When well-formedness is characterized by violable constraints, it is possible to encode the markedness of a structure without assuming an outright ban. The constraint FtBin states a dispreference for degenerate feet, but it does not entail their absence since it may be overridden when higher ranked constraints conflict. However, an overall asymmetry in the markedness of foot types is derived, with binary (or non-degenerate) feet preferred, other things being equal.26 In this context, the arguments for and against the final stress in Cahuilla seem beside the point, echoing a similar observation by Crowhurst & Hewitt (1995a). Although the presence or absence of the final stress in Cahuilla is still germane to the issue of primary/secondary stress asymmetry, the fact that the final stress would require a degenerate foot is not a compelling argument one way or another, since the theory readily admits degenerate feet as a possible, if marked, option. We conclude from this case study that the independence of primary stress from non-primary stresses is not necessarily self-evident, nor is it a theory-neutral determination. Therefore, it is not generally possible to call a language or stress pattern ‘top-down’ in the absence of detailed argumentation within a particular theoretical framework.

4. Asymmetric patterns and language typology This section now turns to a top-down case study of a different nature, revisiting the stress pattern of Yine described in the introduction. In this case, the independence of the primary stress from non-primary stresses will not be challenged. Instead, we will see that this surface asymmetry may be derived in HS top-down (as alluded to in the introduction) or bottom-up, by assuming an additional constraint. This illustration achieves two ends. First, it transparently confirms the general argument here: that a language cannot be labeled ‘top-down’ pretheoretically. And second, it delves into a key piece of argumentation for one parsing mode over the other:

24. A few cases of degenerate primary stress feet do not require top-down parsing (e.g., Auca, Hayes 1995: 182ff). It is only in cases that a primary stress degenerate foot violates the Priority Clause that top-down parsing is needed to motivate its construction. 25. An exception is the work of Hyde (2001, 2002, 2016), whose theory involves reworking the foot (primary and secondary both) as necessarily binary; ambipodal syllables are permitted to ensure both foot binarity and exhaustive parsing. 26. This is true in both HS and in parallel OT, though for different reasons. In HS, Parse-σ favors larger feet at each iteration, other things being equal (Pruitt 2010), whereas in parallel OT it is alignment constraints that favor minimizing the number of feet and therefore maximizing their size to achieve comparable satisfaction of Parse-σ. (For illustrations of the structure-minimizing properties of alignment, see e.g., Elenbaas & Kager 1999; Gordon 2002.) 64 CatJL 18, 2019 Kathryn Pruitt that of typological predictions. Since the two parsing modes utilize different constraints, and since the metric for determining the adequacy of a constraint set is typological, evaluating whether the language under discussion should be analyzed top-down or bottom-up ultimately depends on the cross-linguistic attestation of stress asymmetries. This section provides a description of Yine stress in 4.1 and then provides both top-down and bottom-up analyses in HS (section 4.2). Section 4.3 highlights the different typological predictions of each analysis and compares them to the known attestation of stress systems. Finally, section 4.4 collects the results of the two case studies in this paper and uses them to give a concise summary of the empirical criteria that favor a top-down Gen in Harmonic Serialism.

4.1. Yine stress pattern Yine is a Maipurean language of Peru (Eberhard, Simons, and Fennig 2019) described by Matteson (1965). In Yine the primary stress is on the penultimate syllable, secondary stress is word-initial, and tertiary stresses appear on odd-numbered syllables counting from the left in long enough words, though the syllable immediately preceding the primary stress never receives stress. The stress pattern is thus bidirectional. (For general discussion of bidirectional stress systems see, among others, Kager 2001, Alber 2005, Hyde 2008.) We can analyze this pattern with quantity-insensitive trochees (Hayes 1995: 201, among others). The data in (21) show the stress pattern along with the trochaic feet it entails, repeating the data from (1). The primary stress foot is right-aligned (i.e., word-final), while non-primary stress feet iterate from the left edge. (Following other work, I collapse the secondary/tertiary distinction here.)

(21) Yine data (Matteson 1965: 21)27

ru.(ˈt͡çi.t͡ça) ‘He observes taboo’ (ˌʧi.ja)(ˈha.ta) ‘He cries.’ (ˌsa.lwa).je.(ˈhka.kna) ‘They visit each other.’ (ˌpe.ʧi)(ˌʧhi.ma)(ˈtlo.na) ‘They say they stalk it.’ (ˌru.slu)(ˌno.ti).ni.(ˈtka.na) ‘Their voices already changed.’ (ˌsa.ple)(ˌwhi.ma)(ˌmta.na)(ˈtna.ka) ‘They say he went along screaming again.’ (ˌka.çru:)(ˌka.khi)(ˌmana).ta.(ˈtka.na) ‘They were joking together then, it is said.’

27. The data have been converted to IPA with the following exceptions: [l] and [r] represent lateral and retroflex flaps, respectively, and [h] is a “nasal spirant”. Syllabifications follow Matteson (1965), who suggests that the syllable template in Yine places all intervocalic consonant clusters into onset position. This is based on the fact that all intervocalic consonant sequences can also occur word- initially and the observation that all words end in a vowel. Revisiting Top-Down Primary Stress CatJL 18, 2019 65

4.2. Top-down and bottom-up analyses The asymmetry in Yine is in the direction of alignment of the primary and non- primary stress feet. As described in the introduction to this paper, a top-down analysis provides a logical way to capture this difference. In HS, such an analysis would proceed as follows. At the first iteration of stress assignment, the lone right- aligned primary stress foot is built due to the ranking AlignHdR >> AllFtL. At the next iteration, a secondary stress foot is built at the left edge of the word if there are at least two unparsed syllables remaining because of the ranking Parse-σ >> AllFtL >> AllFtR. Additional feet are constructed in a left-aligned manner as long as there are additional pairs of syllables left unparsed. The derivation in (22) illustrates. At the first iteration, the AlignHd constraint will be active in selecting an optimal candidate because primary stress feet are in the candidate set. And because AlignHdR outranks AllFtL, the most right-aligned foot wins. At subsequent iterations regular iterative parsing will begin at the left edge of the word and continue rightward. Because the first iteration selected a candidate that already maximally satisfies AlignHdR (modulo Trochee and FtBin), the AlignHd constraint will not affect regular parsing, which falls instead to the other constraints.

(22) Top-down analysis of Yine (ˌru.slu)(ˌno.ti).ni.(ˈtka.na) ‘their voices already changed’ σ L - R t d in F H B l ll r oc h ee t a r se

/ru.slu.no.ti.ni.tka.na/ T F P A A 1st iteration a. ru.slu.no.ti.ni.tka.na 7 W L L b. (ˈru.slu).no.ti.ni.tka.na 5 6 W L c. → ru.slu.no.ti.ni.(ˈtka.na) 5 1 5 d. ru.slu.no.ti.ni.tka.(ˈna) 1 W 6 W L 6 W e. ru.slu.no.ti.ni.(tka.ˈna) 1 W 5 L 5 2nd iteration f. ru.slu.no.ti.ni.(ˈtka.na) 5 W 1 5 g. → (ˌru.slu).no.ti.ni.(ˈtka.na) 3 1 5 h. ru.slu.no.(ˌti.ni)(ˈtka.na) 3 1 8 W 3rd iteration i. (ˌru.slu).no.ti.ni.(ˈtka.na) 3 W 1 5 L j. → (ˌru.slu)(ˌno.ti).ni.(ˈtka.na) 1 1 7 k. (ˌru.slu).no.(ˌti.ni)(ˈtka.na) 1 1 8 W 66 CatJL 18, 2019 Kathryn Pruitt

This analysis shows that an AlignHd constraint can overcome the general foot alignment preferences when higher ranked, yielding a bidirectional pattern when parsing is top-down. In contrast, if feet must be built at a lower, non-primary stress level before being promoted to primary stress, then the AlignHd constraint cannot be active in selecting which foot to build at the first, or any, iteration. This is demonstrated in (23). Here, the satisfaction of AlignHd (and Hd(PWd) as well is limited by the fact that Gen does not produce feet with primary stress and can only promote an existing secondary stress foot to primary. Thus, AlignHdR only becomes active to select among existing feet. The result is similar in character to MalakMalak (section 1): a unidirectional stress pattern that appears to iterate left- to-right, with the rightmost foot selected as primary: *(ˌru.slu)(ˌno.ti)(ˈni.tka).na.

(23) Bottom-up Yine derivation with same constraints: wrong outcome28 ) d σ L - R t d in F (PW H B l ll d t a r se

/ru.slu.no.ti.ni.tka.na/ F P H A A 1st iteration a. ru.slu.no.ti.ni.tka.na 7 W 1 b. ru.slu.no.ti.ni.(ˌtka.na) 5 1 5 W c. → (ˌru.slu).no.ti.ni.tka.na 5 1 2nd iteration d. (ˌru.slu).no.ti.ni.tka.na 5 W 1 L e. → (ˌru.slu)(ˌno.ti).ni.tka.na 3 1 2 f. (ˌru.slu).no.ti.ni.(ˌtka.na) 3 1 5 W 3rd iteration g. (ˌru.slu)(ˌno.ti).ni.tka.na 3 W 1 2 L h. → (ˌru.slu)(ˌno.ti)(ˌni.tka).na 1 1 6 4th iteration i. (ˌru.slu)(ˌno.ti)(ˌni.tka).na 1 1 W L 6 j. (ˌru.slu)(ˌno.ti)(ˌni.tka)(ˌna) 1 W L 1 W L 12 W k. → (ˌru.slu)(ˌno.ti)(ˈni.tka).na 1 2 6 l. (ˈru.slu)(ˌno.ti)(ˌni.tka).na 1 6 W 6

28. Parse-σ outranks Hd(PWd) for this particular derivation, though the relevant losing candidates to justify this ranking are not included in (23). See footnote 17. Revisiting Top-Down Primary Stress CatJL 18, 2019 67

An additional constraint is therefore needed in order to derive the bidirectional pattern of Yine with a bottom-up Gen. This can be achieved with AlignWdR, which favors some foot aligned to the right edge of the word, irrespective of whether it is the primary stress. This constraint is usually defined as a member of the generalized alignment family (McCarthy and Prince 1993) as in (24). It was discussed by McCarthy and Prince (1993) when the generalized alignment schema was first introduced and has been used in numerous subsequent analyses by many authors.

(24) Align(PWd,R,Ft,R) (AlignWdR): Assign one violation mark for every PWd whose right edge is not aligned with the right edge of some foot.

When ranked above AllFtL, the dominant general alignment constraint in Yine, AlignWdR favors the building of a foot at the right edge before regular left- to-right iteration proceeds. When it is time to assign primary stress, the ranking of AlignHdR >> AlignHdL favors promoting the lone rightmost foot to primary. This derivation is shown in (25).

(25) Bottom-up derivation of Yine stress with AlignWd ) d σ R L - R t d d F (PW W H l l ll d a r se

/ru.slu.no.ti.ni.tka.na/ A P H A A 1st iteration a. ru.slu.no.ti.ni.tka.na 1 W 7 W 1 L b. → ru.slu.no.ti.ni.(ˌtka.na) 5 1 5 c. (ˌru.slu).no.ti.ni.tka.na 1 W 5 1 L 2nd iteration d. ru.slu.no.ti.ni.(ˌtka.na) 5 W 1 5 e. → (ˌru.slu).no.ti.ni.(ˌtka.na) 3 1 5 f. ru.slu.no.(ˌti.ni)(ˌtka.na) 3 1 8 W 3rd iteration g. (ˌru.slu).no.ti.ni.(ˌtka.na) 3 W 1 5 L h. → (ˌru.slu)(ˌno.ti).ni.(ˌtka.na) 1 1 7 4th iteration i. (ˌru.slu)(ˌno.ti).ni.(ˌtka.na) 1 1 W L 7 j. → (ˌru.slu)(ˌno.ti).ni.(ˈtka.na) 1 1 7 k. (ˈru.slu)(ˌno.ti).ni.(ˌtka.na) 1 6 W 7 68 CatJL 18, 2019 Kathryn Pruitt

The principal conceptual difference between the bottom-up analysis of Yine and the top-down analysis presented just before it is that the bottom-up analysis treats the word-final foot as a product of a high-ranked preference for some foot to be aligned with the right edge of the word, even before it is known that it will bear the primary stress. That is, the bottom-up derivation will not treat the primary stress as exceptional but will derive the alignment asymmetry among feet before the primary stress is even assigned. The AlignWd constraint thus takes the place of AlignHd in the ranking for motivating the ‘lone’ foot, though it does not replace it entirely. At a later iteration this foot will only be assigned the primary stress because the dominant AlignHd constraint matches the AlignWd constraint in its direction (i.e., both right-aligning in this case). Thus, the fact that the rightmost foot is exceptional is technically unrelated to its status as the primary stress foot.29 This will be an important factor in comparing the typological predictions of each analysis just below, because despite this difference, there is no straightforward way to use the Yine data itself to distinguish among these alternatives. Both permit the bidirectional pattern of Yine to be derived and result in identical metrical representations. It is therefore only possible to judge the top-down and bottom-up analyses on the basis of their typological predictions, to which the next section now turns.

4.3. Typological predictions

The top-down analysis of Yine does not require AlignWd, since AlignHd is available to similar effect and it refers directly, and appropriately in this case, to the primary stress foot. But AlignWdR is crucial in the bottom-up analysis because it introduces an asymmetry among stresses, which correlates later in the derivation with primary stress assignment. To compare the bottom-up and top-down analyses, then, we must ask whether AlignWd in general and AlignWdR in particular are justified constraints, which we will do by looking at their typological predictions and comparing them to the known typology. Without AlignWd, the top-down analysis makes the prediction that all bidirectional stress systems will have primary stress on their ‘lone’ foot, because without AlignWd, an AlignHd constraint that conflicts with general parsing (AllFtL/R) is the way to motivate a separate foot with opposite directionality from the rest of the stress pattern. For illustration of this point, a simplified predicted typology of unidirectional and bidirectional languages is given in (26), assuming quantity- insensitive trochaic feet and underparsing.30

29. Interestingly, Hayes (1995: 201) describes a bottom-up analysis of Yine with two foot construction rules that precede the End Rule, similar to the bottom-up analysis with AlignWd presented here. It can be inferred from a consideration of his theory that analyzing Yine top-down would confer no advantage: since the End Rule does not build feet, two separate foot construction rules (one non-iterating at the right, another iterating from the left) would be needed anyway. This is another way that a theoretical framework may dictate a choice between top-down and bottom-up analyses. 30. Underparsing refers to the absence of monosyllabic feet on ‘leftover’ syllables, which is controlled by the ranking of Parse-σ and FtBin. HS, unlike parallel OT, predicts that monosyllabic feet should be possible in both unidirectional and bidirectional stress systems (Hyde 2012; see also Pruitt 2012: Ch 5), though only certain unidirectional patterns are robustly attested with them (Hyde 2014). Revisiting Top-Down Primary Stress CatJL 18, 2019 69

(26) (Simplified) Predictions of top-down parsing (without AlignWd) Foot Alignment Primary stress alignment a. Unidirectional Left-to-Right, Leftmost primary (ˈσσ)(ˌσσ)(ˌσσ)σ AllFtLeft AlignHdLeft

AllFtRight AlignHdRight

b. Unidirectional Right-to-Left, Rightmost primary σ(ˌσσ)(ˌσσ)(ˈσσ) AllFtRight AlignHdRight

AllFtLeft AlignHdLeft

c. Bidirectional: Primary at Left and Secondaries Right-to-Left (ˈσσ)σ(ˌσσ)(ˌσσ) AlignHdLeft

AllFtRight AlignHdRight

AllFtLeft

d. Bidirectional: Primary at Right and Secondaries Left-to-Right (ˌσσ)(ˌσσ)σ(ˈσσ) AlignHdRight

AllFtLeft AlignHdLeft

AllFtRight

The attestation of language types (26)a and (26)b is uncontroversial: (26)a is seen in Pintupi (Hansen and Hansen 1969; section 1), (26)b is the pattern of Cavineɲa31 (Key 1968). Additional languages instantiating each type are described in Gordon (2002), among others. The bidirectional pattern in (26)c is found in Garrwa32 (Furby 1974), and (26)d is the pattern of Yine.33

31. Tacanan, Bolivia (Eberhard, Simons & Fennig 2019). 32. Yanyi, Northern Territory (Eberhard, Simons & Fennig 2019). 33. These attestations refer, of course, to the quantity-insensitive trochaic versions of each pattern as illustrated here. Whether the equivalent patterns with different foot types (e.g., quantity-sensitive and/or iambic) are attested is another matter, and considerable literature has amassed throughout the history of metrical theory discussing typological asymmetries between trochees and iambs, quantity-insensitive vs. quantity-sensitive systems, etc. I set aside these issues here. 70 CatJL 18, 2019 Kathryn Pruitt

In contrast, when AlignWd is included in the constraint set, as required for the bottom-up analysis of Yine, the correlation between the lone foot and the primary stress seen in (26)c and (26)d is no longer guaranteed. An analysis with AlignWd predicts the language types in (26) plus those in (27), where the lone foot and the primary stress are no longer coextensive.

(27) Additional (simplified) predictions of bottom-up parsing (with AlignWd) Foot Alignment Primary stress alignment a. Bidirectional: Foot at Left and Right-to-Left, Rightmost primary (ˌσσ)σ(ˌσσ)(ˈσσ) AlignWdRight AlignHdRight

AllFtLeft AlignHdLeft

AllFtRight

b. Bidirectional: Foot at Right and Left-to-Right, Leftmost primary (ˈσσ)(ˌσσ)σ(ˌσσ) AlignWdLeft AlignHdLeft

AllFtRight AlignHdRight

AllFtLeft

The inclusion of AlignWd in the constraint set predicts the language types in (27), essentially no matter the parsing method—with bottom-up or top-down parsing in HS and in parallel OT as well (Kager 2005). In other words, the possibility of separating the lone foot and primary stress, as in the examples in (27), is fundamentally entailed when AlignWd is included, so we should now ask whether such systems are attested. If so, this supports AlignWd as a valid constraint, meaning the top-down analysis of Yine is not required and therefore Yine is not a top-down language, despite the fact that its primary stress is independent from its non-primary stresses on the surface. On the other hand, if no languages instantiate the patterns in (27), this would constitute evidence against AlignWd and therefore against a bottom-up analysis of Yine (and other bidirectional systems). The language types in (27) have been the subject of considerable discussion in the OT metrical literature already (Kager 2001, McCarthy 2003, and other references below). In parallel OT the discussion has emerged not from a concern with AlignWd, per se, but the use of gradient alignment constraints like AllFtL/R for deriving stress vs. the *Lapse-based Rhythmic Licensing Theory of Kager (2001). But since the typological issues overlap with what we are discussing here, that literature proves quite relevant. Yet whether the languages in (27) do indeed exist is not immediately obvious from surveying this literature. Kager (2001) argues that all bidirectional stress systems have primary stress on the lone foot, claiming that Revisiting Top-Down Primary Stress CatJL 18, 2019 71 patterns like those in (27) do not exist. Portions of Kager’s arguments are echoed by several others, including McCarthy (2003)34 and Alber (2005). Nonetheless, reports of counterexamples are not difficult to find, particularly for the pattern in (27)a. Indonesian (Cohn 1989, Cohn 1993, Cohn & McCarthy 1994) and Spanish (Harris 1983, Harris 1989, Roca 1986) are the most often cited potential examples of a left-aligned non-primary-stress (which is referred to in the literature as the “initial dactyl” by Prince 1983 and others, referencing the sequence of stress-unstressed-unstressed). There are extant rejoinders to these examples, but there are also rejoinders to the rejoinders. In Indonesian the controversy surrounds the fact that the relevant forms are Dutch loans and might therefore represent stress preservation rather than a productive pattern, though Cohn (1993: 374, fn. 1) seems to dismiss this suggestion. For Spanish, Kager (2001) argues that morphological complexity may play a role in creating the pattern, but Hyde (2008; also Hyde & McCord 2012) argues that this cannot be the whole explanation. The balance of evidence seems to favor the admission of the initial dactyl pattern and therefore a constraint like AlignWdL to motivate it, no matter the parsing method (top-down, bottom-up, or in parallel). As for the pattern in (27)b, which would utilize high-ranked AlignWdRight and is therefore specifically predicted by the bottom-up analysis of Yine, things are less clear. Cases of the Indonesian- and Spanish-type, but where the secondary stress foot must be final, are generally reported not to exist (Prince 1983; Hyde 2008).35 Nonetheless, a possible example may be found in English words that exhibit (leftward) primary stress retraction, e.g., désignàte, where the primary stress seems to have drifted further to the left than strictly necessary to satisfy the requirements of NonFinality and FtBin (which would be satisfied by *de(síg)nate, for example). This is described by Pater (2000: 241f, fn. 5) as possible evidence for a constraint demanding some foot (head) at the right edge of a word. The upshot of this brief foray into the typology of bidirectional stress systems is that a constraint like AlignWd is probably required in Con to account for attested stress patterns. And if this constraint is admitted, Yine can easily be analyzed top-down or bottom-up in Harmonic Serialism, despite the surface asymmetry it exhibits between primary and non-primary stresses.

4.4. Discussion We have now examined two case studies of languages that have previously motivated top-down analyses, and I have shown in each case that there are reasons to be skeptical of such characterizations in any general sense. In short, the identifica-

34. Interestingly, McCarthy (2003) uses the absence of clear cases of languages like (27) to argue against the typical gradient formulations of AllFtL/R, presupposing that constraints like AlignWdL/R are necessarily present in Con. What I am doing here is the opposite, presupposing AllFtL/R and instead evaluating whether an AlignWd constraint is justified. See also Kager (2005), who similarly observes that these languages only arise when AlignWd is included. 35. Prince (1983: 49) discusses only the strict mirror image of the initial dactyl, the final anapest, noting that no such cases are known. 72 CatJL 18, 2019 Kathryn Pruitt tion of surface asymmetries between primary and non-primary stresses is typically not theory-neutral (as in Cahuilla), and may also depend on a particular structural analysis (as in Tübatulabal, discussed in the introduction). Further, establishing a surface asymmetry of this sort is not sufficient for calling the language ‘top-down’, since bottom-up analyses might be available with different constraints (as in Yine). However, the issues illuminated by these case studies do point to a positive conclusion: we are able to provide the following informal36 criteria for motivating a top-down analysis in Harmonic Serialism.

(28) Criteria to establish that a top-down analysis is required in HS a. Primary and secondary stress must display an asymmetry in construction.37 b. The asymmetry must not also be found among secondary stresses in some language.

The criterion in (28)a is a necessary (though not sufficient) condition for a stress pattern to potentially require (not just allow) a top-down analysis, and it technically applies in any derivational model, not just HS. Although the Cahuilla case study emphasizes that the criterion in (28)a cannot be ascertained in a theory-neutral way, it still highlights the importance of establishing such an asymmetry in order to argue for a top-down analysis. If primary and non-primary stresses obey the same general requirements for alignment and foot type, then the general stress constraints can be entrusted to put a foot in the correct place for later primary stress assignment in a bottom-up analysis, and a top-down analysis gains no traction over the bottom-up one. But if primary-stress-specific requirements appear to override those of secondary stresses, then a top-down analysis may be needed, since only with a top-down analysis may the primary stress constraints dictate footing. This is precisely the issue that arises with the different interpretations of Cahuilla’s stress pattern described in section 3. The interpretation of the data determines whether the primary and secondary stresses show an asymmetry, and therefore, whether the pattern points toward a top-down analysis. The criterion in (28)b, however, is a limiting factor on the conclusions we are able to draw from any given asymmetric pattern, and it derives from the fact that bottom-up analyses may be (and often technically are) available for asymmetric patterns using additional constraints. Specifically, as we can infer from the illustrations earlier in this section, an asymmetry in primary and secondary stresses can be

36. An anonymous reviewer requests that these criteria be stated more formally, but given the non- trivial argumentation required to establish both criteria (as Cahuilla showed for (28)a and Yine for (28)b), I believe the criteria best serve their function when viewed heuristically. 37. For the asymmetry to motivate a top-down analysis it must involve different procedures (i.e., different rules or constraints) for assigning primary and non-primary stresses, rather than being the result of a subsequent process that affects primary and secondary stresses differently. Compare languages where the primary stress syllable is lengthened (Hayes 1995: 84 cites Icelandic and Wargamay as examples). This results in an asymmetry, but does not require the procedures for primary and secondary stress foot construction to differ. Revisiting Top-Down Primary Stress CatJL 18, 2019 73 derived bottom-up when the grammar is permitted to follow two different parsing generalizations—one for the foot that will become the primary stress, and one for all other feet. In the bottom-up analysis of Yine, for example, parsing was motivated both by AlignWdR and by Parse-σ (along with AllFtL). However, in so doing, we predict the dissolution of any formal relationship between the primary stress’s exceptional behavior and the fact that it is the primary stress. Unless the rankings are connected through some as-yet-unknown mechanism, or the derivations have foresight, the establishment of an exceptional foot and the assignment of that foot as the primary stress are predicted by bottom-up analyses of asymmetries to be, typologically-speaking, a coincidence. This means that top-down arguments can be made by establishing a primary/secondary stress asymmetry in one language that only occurs between primary and secondary stresses—and never among secondary stresses—in the known typology of stress systems.38 In the case of bidirectional systems, the typological evidence seems to point to the existence of languages where secondary stresses do show asymmetries in alignment, so patterns like Yine do not offer evidence for a top-down analysis, even though its asymmetry happens to correlate with primary vs. non-primary stress. But with this criterion now made explicit, it should be possible to examine other cases of primary stress independence to determine whether any such asymmetries are indeed typological, rather than language-specific. Implicit in this discussion is the fact that finding a language that satisfies the criteria in (28) dictates not only that a top-down analysis is required, but also that a top-down Gen (with the characteristics identified in section 2.2) must be chosen over a bottom-up Gen for a complete analysis of primary stress in HS.

5. Conclusion Previous literature shows that the concept of a ‘top-down’ language has sometimes been used as an empirical category, but by examining two typical cases we have seen that the interpretation of the data and its relationship to language typology are just as important as the pattern itself in establishing that a top-down analysis is required. Therefore, the primary conclusion of this paper is that individual languages are neither top-down nor bottom-up, as these terms should be reserved to describe categories of analysis. At the same time, we made precise the characteristics of a stress pattern that would motivate a top-down analysis—and therefore a top-down definition of Gen—in Harmonic Serialism: the primary stress must be shown to be independent on some dimension not just within a particular language but also within the attested typology of stress systems. Ideally, these criteria will guide future discussions of primary and secondary stress asymmetries and their implications for derivational models of stress assignment.

38. Ultimately, not unlike Hayes’s (1995) decision to draw a typological distinction between primary and secondary stress with respect to degenerate feet. 74 CatJL 18, 2019 Kathryn Pruitt

References Alber, Birgit. 2005. Clash, lapse, and directionality. Natural Language & Linguistic Theory 23: 485-542. Bailey, Todd Mark. 1995. Non-metrical constraints on stress. Doctoral dissertation, University of Minnesota. Birk, D. B. W. 1976. The MalakMalak language, Daly River (Western Arnhem Land). Pacific Linguistics B-45. The Australian National University. Chen, Matthew Y. 1999. Directionality constraints on derivation? In Ben Hermans & Marc van Oostendorp (eds.). The Derivational Residue in Phonological Optimlity Theory, 105-127. John Benjamins: Philadephia. Cohn, Abigail C. 1989. Stress in Indonesian and bracketing paradoxes. Natural Language & Linguistic Theory 7(2): 167-216. Cohn, Abigail C. 1993. The initial dactyl effect in Indonesian. Linguistic Inquiry 24(2): 372-381. Cohn, Abigail C. & John J. McCarthy. 1994. Alignment and parallelism in Indonesian phonology. Published 1998 in Working Papers of the Cornell Phonetics Laboratory. Vol. 12, 53-137. Crowhurst, Megan. 1991. Minimality and Foot Structure in Metrical Phonology and Prosodic Morphology. Doctoral dissertation, University of Arizona, Tucson. Crowhurst, Megan & Mark Hewitt. 1995a. Directional footing, degeneracy, and alignment. Proceedings of the North East Linguistic Society 25, 47-61. Crowhurst, Megan & Mark Hewitt. 1995b. Prosodic overlay and headless feet in Yidiɲ. Phonology 12: 39-84. Eberhard, David M., Gary F. Simons & Charles D. Fennig (eds.). 2019. Ethnologue: Languages of the World. Twenty-second edition. Dallas, Texas: SIL International. Online version: . Elenbaas, Nine & René Kager. 1999. Ternary rhythm and the lapse constraint. Phonology 16: 273-329. Elfner, Emily. 2016. Stress-epenthesis interactions in Harmonic Serialism. In John J. McCarthy & Joe Pater (eds.). Harmonic Grammar and Harmonic Serialism. London: Equinox. Furby, Christine. 1974. Garawa phonology. Number 37 in Pacific Linguistics, Series A. Australian National University, Canberra. Goedemans, Rob & Harry van der Hulst. 2014. The separation of accent and rhythm: evidence from StressTyp. In Harry van der Hulst (ed.). Word Stress: Theoretical and Typological Issues, 119-145. New York: Cambridge University Press. Goldsmith, John A. 1990. Autosegmental and Metrical Phonology. Cambridge, MA: Basil Blackwell. Gordon, Matthew. 2002. A factorial typology of quantity-insensitive stress. Natural Language & Linguistic Theory 20(3): 491-552. Hansen, Kenneth C. & L. E. Hansen. 1969. Pintupi phonology. Oceanic Linguistics 8: 153-170. Harris, James. 1983. Syllable structure and stress in Spanish. Linguistic Inquiry Monograph 8. Cambridge, MA: MIT Press. Harris, James. 1989. A podiatric note on secondary stress in Spanish. Ms. Massachusetts Institute of Technology. Revisiting Top-Down Primary Stress CatJL 18, 2019 75

Hayes, Bruce. 1981. A metrical theory of stress rules. Bloomington: Indiana University Linguistics Club. Hayes, Bruce. 1995. Metrical stress theory: principles and case studies. Chicago: University of Chicago Press. Hulst, Harry van der. 1984. Syllable structure and stress in Dutch. Dordrecht: Foris. Hulst, Harry van der. 1997. Primary accent is non-metrical. Revista di Linguistica 9(1): 99-127. Hulst, Harry van der. 2009. Brackets and grid marks, or theories of primary accent and rhythm. In Eric Raimy & Charles E. Cairns (eds.). Contemporary Views on Architecture and Representations in Phonological Theory, 225-245. Cambridge, MA: MIT Press. Hulst, Harry van der. 2012. Deconstructing stress. Lingua 122: 1494-1521. Hurch, Bernhard. 1996. Accentuations. In Bernhard Hurch & Richard A. Rhodes (eds.). Natural Phonology: The State of the Art, 73-96. Berlin: Mouton de Gruyter. Hyde, Brett. 2001. Metrical and Prosodic Structure in Optimality Theory. Doctoral dissertation, Rutgers University. ROA-476. Hyde, Brett. 2002. A restrictive theory of metrical stress. Phonology 19: 313-359. Hyde, Brett. 2008. Bidirectional stress systems. In Charles B. Chang & Hannah J. Haynie (eds.). Proceedings of the 26th West Coast Conference on Formal Linguistics, 270-278. Cascadilla Proceedings Project, Somerville, MA. Hyde, Brett. 2012. The odd-parity input problem in metrical stress theory. Phonology 29(3): 383-431. Hyde, Brett. 2014. Symmetries and asymmetries in secondary stress patterns. In Harry van der Hulst (ed.). Word Stress: Theoretical and Typological Issues. Cambridge: Cambridge University Press. Hyde, Brett. 2016. Layering and Directionality: Metrical Stress in Optimality Theory. London: Equinox. Hyde, Brett & Bethany McCord. 2012. The inadequacy of a faithfulness-based approach to Spanish secondary stress. Ms. Washington University. ROA-1154. Jesney, Karen. 2011. Positional faithfulness, non-locality, and the Harmonic Serialism solution. In Suzi Lima, Kevin Mullin, and Brian W. Smith (eds.). Proceedings of the North East Linguistic Society 39. GLSA, Amherst, MA. ROA-1018. Kager, René. 1989. A metrical theory of stress and destressing in English and Dutch. Dordrecht: Foris Publications. Kager, René. 1995. Review of Hayes (1995). Phonology 12(3): 437-464. Kager, René. 1999. Optimality Theory. Cambridge: Cambridge University Press. Kager, René. 2001. Rhythmic directionality by positional licensing. Paper presented at the Fifth HIL Phonology Conference, University of Potsdam. ROA-514. Kager, René. 2005. Rhythmic licensing theory: an extended typology. The Phonology- Morphology Circle of Korea. Available at . Key, Mary Ritchie. 1968. Comparative Tacanan Phonology. The Hague: Mouton. Kimper, Wendell A. 2011. Locality and globality in phonological variation. Natural Language & Linguistic Theory 29: 423-465. Matteson, Esther. 1965. The Piro (Arawakan) language. University of California Publications in Linguistics 42. Berkeley and Los Angeles: University of Califor- nia Press. 76 CatJL 18, 2019 Kathryn Pruitt

McCarthy, John J. 2000. Harmonic serialism and parallelism. In M. Hirotani (ed.). Proceedings of the North East Linguistic Society 30, 501-524. GLSA, Amherst, MA. McCarthy, John J. 2003. OT constraints are categorical. Phonology 20: 75-138. McCarthy, John J. 2008. The serial interaction of stress and syncope. Natural Language & Linguistic Theory 26: 499-546. McCarthy, John J. 2010. Studying Gen. Journal of the Phonetic Society of Japan 13(2): 3-12. McCarthy, John J. 2016. The theory and practice of Harmonic Serialism. In John J. McCarthy & Joe Pater (eds.). Harmonic Grammar and Harmonic Serialism. London: Equinox. McCarthy, John J., Joe Pater & Kathryn Pruitt. 2016. Cross-level interactions in Harmonic Serialism. In John J. McCarthy & Joe Pater (eds.). Harmonic Grammar and Harmonic Serialism. London: Equinox. McCarthy, John J. & Alan Prince. 1993. Generalized alignment. Yearbook of Morphology 12: 79-153. ROA-7. McCarthy, John J. & Kathryn Pruitt. 2013. Sources of phonological structure. In Hans Broekhuis & Ralf Vogel (eds.). Linguistic Derivations and Filtering: Minimalism and Optimality Theory. London: Equinox. McGarrity, Laura. 2003. Constraints on patterns of primary and secondary stress. Doctoral dissertation, Indiana University. Odden, David. 1979. Principles of stress assignment: a cross-linguistic view. Studies in the Linguistic Sciences 9(1): 157-175. Pater, Joe. 2000. Non-uniformity in English secondary stress: the role of ranked and lexically specific constraints. Phonology 17: 237-274. Prince, Alan. 1983. Relating to the grid. Linguistic Inquiry 14(1): 19-100. Prince, Alan. 1990. Quantitative consequences of rhythmic organization. In Proceedings of the Chicago Linguistic Society 26, Vol. 2, pp. 355-398. Prince, Alan & Paul Smolensky. 1993/2004. Optimality Theory: Constraint Interaction in Generative Grammar. Malden, MA: Blackwell. Pruitt, Kathryn. 2010. Serialism and locality in constraint-based metrical parsing. Phonology 27(3): 481-526. Pruitt, Kathryn. 2012. Stress in Harmonic Serialism. Doctoral dissertation, University of Massachusetts Amherst. Pruitt, Kathryn. in prep. Parallelism within serialism: evidence from primary stress. Ms. Arizona State University. Roca, Iggy M.. 1986. Secondary stress and metrical rhythm. Phonology Yearbook 3: 330-341. Seiler, Hansjakob. 1957. Die phonetischen Grundlagen der Vokalphoneme des Cahuilla. STUF - Language Typology and Universals 10(1-4): 204-223. Seiler, Hansjakob. 1965. Accent and morphophonemics in Cahuilla and in Uto-Aztecan. International Journal of American Linguistics 31(1): 50-59. Seiler, Hansjakob. 1967. Structure and reconstruction in some Uto-Aztecan languages. International Journal of American Linguistics 33(2): 135-147. Seiler, Hansjakob. 1977. Cahuilla Grammar. Banning, CA: Malki Museum Press. Revisiting Top-Down Primary Stress CatJL 18, 2019 77

Staubs, Robert. 2013. Pathologies of feature-driven stress. In Yelena Fainleib and Nicholas LaCara and Yangsook Park (eds.). Proceedings of the Northeast Linguistic Society 41. GLSA, Amherst, MA. Voegelin, Charles. 1935. Tübatulabal Grammar, University of California Publications in American Archaeology and Ethnology, vol. 34. Berkeley: University of California Press.

Catalan Journal of Linguistics 18, 2019 79-103

Cross-level interactions in Latin: Vowel shortening, vowel deletion and vowel gliding*

Haike Jacobs Radboud University [email protected]

Received: April 1, 2019 Accepted: July 4, 2019

Abstract

Serial and parallel OT differ in the way they account for phonological generalizations referring to more than one level of the prosodic hierarchy. Vowel shortening in Latin is analyzed by McCarthy, Pater & Pruitt (2016) as a case in point. Vowel shortening takes place to optimize foot structure. In parallel OT, footing and shortening can be evaluated in parallel, but in serial OT footing and shortening necessarily take place in a serial derivation. In this paper, both the serial and the parallel analysis of Latin vowel shortening are critically discussed. After that, two other potential cases of cross-level interactions in Latin are addressed: vowel deletion and vowel gliding. For each of these cases, it is argued that a serial analysis has to be preferred over a parallel one. Keywords: Serial Optimality theory; Parallel Optimality theory; Latin vowel shortening, vowel deletion and vowel gliding; Cross-level interactions; Uneven trochee

Resum. Interaccions entre nivells en llatí: abreujament vocàlic, elisió vocàlica i formació de diftongs

La versió paral·lela i la serial de la teoria de l’optimitat (TO) difereixen entre sí en la manera com expliquen les generalitzacions fonològiques que es refereixen a més d’un nivell de la jerarquia prosòdica. L’abreujament vocàlic en llatí ha estat analitzat per McCarthy, Pater & Pruitt (2016) com un exemple d’això. L’escurçament es produeix per tal d’optimitzar l’estructura del peu. En la primera versió la construcció dels peus i l’escurçament poden avaluar-se en paral·lel, però en la segona la construcció dels peus i l’abreujament s’han de dur a terme necessàriament en una derivació serial. En aquest article es discuteixen críticament ambdues anàlisis. Després del l’es- curçament vocàlic, s’aborden altres dos casos d’interacció entre nivells en llatí: l’elisió vocàlica i la diftongació de vocals. Per a cadascun d’aquests casos, s’argumenta que és preferible una anàlisi serial a una de paral·lela. Paraules clau: teoria de la optimitat serial; teoria de la optimitat en paral·lel; abreujament, elisió i diftongació de les vocals del llatí; interaccions entre nivells; troqueu asimètric

* I would like to thank two anonymous reviewers for their useful and helpful critical comments and remarks. All remaining errors are my own responsibility.

ISSN 1695-6885 (in press); 2014-9718 (online) https://doi.org/10.5565/rev/catjl.261 80 CatJL 18, 2019 Haike Jacobs

Table of Contents 0. Introduction 3. Vowel gliding and cross-level 1. Vowel Shortening and cross-level interactions interactions in Harmonic Serialism 4. Summary versus Parallel OT References 2. Vowel deletion and cross-level interactions

0. Introduction McCarthy, Pater & Pruitt (2016) have drawn attention to a specific difference between parallel OT (Prince & Smolensky 2004) and serial versions of OT, like Harmonic Serialism (McCarthy 2016). Within Harmonic Serialism, an input form is subject to the constraint grammar of a language and may undergo one single modification, but only if that modification makes the form more harmonic, that is, no longer violating a markedness constraint that is ranked higher than the specific faithfulness constraint that is going to be violated by that single modification. The altered input is then subject again to the same constraint grammar and the process repeats itself until no more harmonic improvement is possible. An input form thus goes step-wise though the constraint grammar to end up as the optimal surface form. In Parallel OT, there is no limitation in terms of the modifications that are allowed. This means that phonological generalizations that refer to more than one level of the prosodic hierarchy cannot be evaluated in parallel, but have to be treated differently in Harmonic Serialism. A case in point is foot-based vowel shortening. If a long vowel is shortened, because it makes a better foot, then in Parallel OT, footing and shortening can be evaluated in parallel, but not in Harmonic Serialism. This paper is structured as follows. In section 1, the Serial OT analysis of vowel shortening proposed by McCarthy, Pater & Pruitt (2016) is compared to the Parallel OT analysis of Prince & Smolensky (2004). We will focus on the different empirical predictions of the two analyses and point out a shared problem: analyzing short pre-final vowels as secondarily stressed in some word types, which as we will show in section 2 is problematic for vowel deletion. Section 2.1 motivates the use of an uneven trochee and of a foot-based vowel deletion constraint. After that, we will show in section 2.2, why a serial analysis is preferred over a parallel one. In section 2.3 we will address the cross-level interaction in Latin vowel deletion. Finally, section 3 discusses the cross-level interaction in vowel gliding and shows why a serial analysis has to be preferred over a parallel one for vowel gliding.

1. Vowel Shortening and cross-level interactions in Harmonic Serialism versus Parallel OT Cretic shortening in Pre-classical Latin is taken as an example by McCarthy, Pater & Pruitt (2016) of what they call a cross-level interaction, that is, a generalization that refers to more than one level of the prosodic hierarchy. By Cretic shortening, Cross-level interactions in Latin CatJL 18, 2019 81 the final vowel of a HLH word is shortened, as in words like /deːsinoː/ desino ‘I cease’ or /neskioː/ nescio ‘I don’t know’ that surface as (ˈdeː)(ˌsi.no) and (ˈnes) (ˌki.o) in Pre-classical Latin. Following Mester (1994) and Prince & Smolensky (2004), the rationale for vowel shortening is assumed to be the fact that it allows for a better or more complete footing: instead of analyzing the word as (ˈH)L(ˌH) it is analyzed as (ˈH)(ˌLH-), where H- stands for a heavy syllable that has been made light. A footing (ˈH)(ˌLH-) is thus equivalent to (ˈH)(ˌLL). In Prince & Smolensky (2004), following Mester (1994), shortening and footing take place in parallel, that is, the two most relevant output candidates (ˈH)L(ˌH) and (ˈH)(ˌLH-) are evaluated in parallel in the selection of the optimal output form. The relative ranking of two constraints, one asking for syllables to be parsed in a foot (Parse-σ) and the other asking that moras not be deleted (Max-μ), determines which of the two output forms is optimal: (ˈH)(ˌLH-), with one mora deleted, in Pre-classical Latin and (ˈH) L(ˌH), with no mora deleted, but with one unparsed syllable, in Classical Latin. This is illustrated in (1) and (2), in which we have indicated a shortened H syllable directly as L instead of H- in the output forms.

(1) Cretic shortening in Pre-classical Latin

/deː.si.noː/ Parse-σ Max-μ (ˈH)L(ˌH) [(ˈdeː)si(ˌnoː) *! F (ˈH)(ˌLL) [(ˈdeː)(ˌsi.no)] *

(2) Vowel length respected in Classical Latin

/deː.si.noː/ Max-μ Parse-σ F (ˈH)L(ˌH) [(ˈdeː)si(ˌnoː) * (ˈH)(ˌLL) [(ˈdeː)(ˌsi.no)] *!

In Harmonic Serialism, this analysis is not possible. Vowel shortening and footing constitute both a single modification, which implies that necessarily one has to precede the other. If the vowel is shortened for foot harmony reasons, this means that the foot, which is going to be improved by the shortening, has to be there before the actual shortening takes place. This is the way Cretic shortening is analyzed in McCarthy, Pater & Pruitt (2016). A trochaic (ˌLH) foot, although it never appears in Latin surface forms, is allowed as an intermediate foot in the harmonically serial evaluation. This is an example of what is called a ‘violation of the surface-true’. The intermediate (ˌLH) foot violates the constraint WSP (a heavy syllable may not be located in the weak part of a foot) which is surface-unviolated in Latin (cf. McCarthy, Pater & Pruitt (2016) for a more detailed discussion).1 The

1. The choice between the two possible Classical Latin output forms (ˈH)L(ˌH) or (ˈH)(ˌLH) depends on the ranking of the constraint WSP (a heavy syllable may not may not be located in the weak part 82 CatJL 18, 2019 Haike Jacobs intermediate (ˌLH) foot is then turned into a (ˌLH-) foot (or equivalently (ˌLL)), that is, with a shortened vowel, at the next round of evaluation. Let us briefly illustrate their analysis. For the derivation of surface (ˈdeː)(ˌsi.no), the following steps are assumedː /deːsinoː/ à deː.si.noː à (ˈdeː)si.noː à (ˈdeː)(ˌsi.noː) à (ˈdeː)(ˌsi.no). The relevant steps are illustrated, after syllabification at step 1, in tableaux (3) to (5) below. The NonFinality constraint is split into two constraints NonFin (Main-σ), the main stressed syllable may not be final, and NonFin (Main-Ft), the foot with main stress may not be final. The constraint RhHrm, rhythmic harmony, rules out uneven trochees, that is (ˈHL) feet and the constraint Align-R (Main-σ) requires the main stressed syllable to be the final syllable of the word (the number of syllables indicates the distance away from the word edge).

(3) Step 2: deː.si.noː going to (ˈdeː)si.noː. HLH becomes (ˈH)LH /deː.si.noː/ Non-Fin RhHrm Non-Fin Align-R Parse-σ WSP Max-μ Main-σ Main-Ft Main-σ

F (ˈH)LH σ σ * * (ˈHL)H *! σ σ * H(ˈLH) *! σ * * HL(ˈH) *! * * *

(4) Step 3: (ˈdeː)si.noː going to (ˈdeː)(ˌsi.noː). (ˈH)LH becomes (ˈH)(ˌLH) (ˈdeː)si.noː Non-Fin RhHrm Non-Fin Align-R Parse-σ WSP Max-μ Main-σ Main-Ft Main-σ

F (ˈH)(ˌLH) σ σ * (ˈH)LH σ σ *!* (ˈH)L(ˌH) σ σ *!

of a foot). If this constraint is located below Parse-σ and Max-μ, (ˈH)(ˌLH) is the predicted optimal result, if it is ranked above Max-μ, but below Parse-σ, (ˈH)L(ˌH) is predicted to be optimal. Cross-level interactions in Latin CatJL 18, 2019 83

(5) Step 4: shortening (ˈdeː)(ˌsi.noː) going to (ˈdeː)(ˌsi.no). (ˈH)(ˌLH) becomes (ˈH)(ˌLL) (ˈdeː)(ˌsi.no) Non-Fin RhHrm Non-Fin Align-R Parse-σ WSP Max-μ Main-σ Main-Ft Main-σ

(ˈH)(ˌLH) σ σ *! F (ˈH)(ˌLL) σ σ * (ˈH) L H σ σ *!* (ˈH) L(ˌH) σ σ *!

By ranking Parse-σ above WSP the analysis thus arrives at having a secondary quantity-insensitive foot (ˌLH) in the course of the derivation, at step 3 in (4), and then improving on it at step 4 in (5). McCarthy, Pater & Pruitt remark “Thus, although (ˌLH) trochees are absent from surface forms, they emerge and are later eliminated by shortening in the course of the derivation.” Iambic shortening, so-called because it takes place in words that have an iambic input shape /LH/, like /a.moː/ amo ‘I love’ surfacing as (ˈa.mo) in Pre-classical Latin, is analyzed in a similar way, as illustrated in (6) and (7), where we have added the constraint Ft-Bin, a foot must be binary (consisting of two syllables or two moras).

(6) Step 2: a.moː going to (ˈa.moː). LH becomes (ˈLH) /a.moː/ Non-Fin FtBin Non-Fin Align-R Parse-σ WSP Max-μ Main-σ Main-Ft Main-σ

F (ˈLH) * σ * L(ˈH) *! * * (ˈL)H *! σ *

(7) Step 3: (ˈa.moː) going to (ˈa.mo). (ˈLH) becomes (ˈLL) (ˈa.moː) Non-Fin FtBin Non-Fin Align-R Parse-σ WSP Max-μ Main-σ Main-Ft Main-σ

F (ˈLL) * σ * (ˈLH) * σ *! 84 CatJL 18, 2019 Haike Jacobs

A number of remarks are in order. While a (ˌLH) or a (ˈLH) trochee (with secondary stress as in (4) or with main stress as in (6)), may be allowed in the course of the derivation, a non-final main stressed foot (ˈLH) trochee obviously needs to be avoided, as this would lead to the incorrect prediction of shortening a long vowel with primary stress. This is taken care of by the Align-R (Main-σ) constraint that requires the main stressed syllable to be the final syllable of the word, as illustrated for [a.ˈmiː.kum] amicum ‘friend’ in (8).

(8) Main stress for amicum: a.miː.kum going to a(ˈmiː)kum. LHL becomes L(ˈH)L /a.miː.kum/ Non-Fin RhHrm Non-Fin Align-R Parse-σ WSP Max-μ Main-σ Main-Ft Main-σ

F L(ˈH)H σ * * (ˈLH)H σ σ! * * LH(ˈH) *! * * *

McCarthy, Pater & Pruitt (2016) mention an interesting difference between their analysis and the original parallel analysis advanced by Prince & Smolensky (2004). In that analysis, the Pre-classical Latin optimal output form for cretic / HLH/ words, that is (ˈH)(ˌLL), has an interesting competitor in the surface form (ˈLL)(ˌH), that is, a surface form that fares equally well in terms of completely footing all syllables, but in which it is not the final syllable that is shortened, but the vowel of the main stressed syllable. A potential surface output (ˈde.si)(ˌnoː) is thus compared to actual optimal (ˈdeː)(ˌsi.no). The selection of (ˈH)(ˌLL) over (ˈLL)(ˌH) is taken care of by Prince & Smolensky (2004) by invoking the Peak-Prominence constraint in (9).

(9) Peak-Prominence (Pk-Prom): Peak (x) › Peak (y) if |x| › |y|

According to the Peak Prominence constraint (x) is a better peak than (y) if its intrinsic prominence is greater than that of (y). A heavy syllable is thus a better peak than a light syllable. This is illustrated in tableau (10) where we have left out the two Non-Fin constraints. Both output forms (10a) and (10b) are equally good in terms of the relative ranking of Parse-σ and Max-μ, but differ in their compliance with the last ranked constraint Pk-Prom (cf. Prince & Smolensky 2004: 76). Cross-level interactions in Latin CatJL 18, 2019 85

(10) Parallel parsing of HLH

/deːsi.no/ WSP RhHrm A l i g n - R Parse-σ Max-μ Pk-Prom /HLH/ Main-σ a.F(ˈH)(ˌLL) σ σ * |H| b.(ˈLL)(ˌH) σ σ * |L| ! c.(ˈH)L(ˌH) σ σ *! |H| d.(ˈH)(ˌLH) *! σ σ |H| e.(ˈHL)(ˌH) *! σ σ |H|

In their further discussion of the constraint Pk-Prom, Prince & Smolensky (2004: 78-79) suggest that the constraint holds for all peaks or heads, but that the evaluation of the head of the PrWd takes priority over the evaluation of foot heads. So, in order to be precise, both output candidate (10a) and (10b) in fact entail a violation of Pk-Prom. Given that (10a) has a Pk-Prom violation of a foot that is not the main stressed foot, whereas in (10b) the main foot violates Pk-Prom, (10a) is evaluated as being better than (10b). The displayed evaluations for Pk-Prom in (10) should thus be interpreted as referring to the foot that has main stress, that is, the first foot in the output forms. The initial foot in (10b) has a L peak, but a H one in (10a) and the other output forms. McCarthy, Pater & Pruitt (2016: 12) remark that in their Harmonic Serialism analysis, the competition between (ˈH)(ˌLL) and (ˈLL)(ˌH), that is between (ˈdeː)(ˌsi. no) and (ˈde.si)(ˌnoː), is an issue that simply does not arise, given that

*(ˈde.si)(ˌnoː) is never even a candidate, much less a serious challenger to the intended winner. It is not a candidate because it would require an intermediate step with a (ˈHL) trochee, (ˈde.si). Undominated RhHrm ensures that such feet never win even at intermediate derivational steps. The HS analysis therefore has an advantage over the P-OT analysis: it can explain, and need not stipulate, which syllable is affected by cretic shortening.

McCarthy, Pater & Pruitt (2016) observe a further difference between their serial and the original parallel analysis. Two heavy syllables before the main stressed syllable are predicted to become (ˌHH) in the serial analysis, but (ˌH)(ˌH) in the parallel analysis. McCarthy, Pater & Pruitt mention as an example /au.diː.toː.te/ auditote ‘hear! (fut.pl.)’. Their serial analysis predicts (ˌau.diː)(ˈtoː)te, whereas the parallel analysis predicts (ˌau)(ˌdiː)(ˈtoː)te. This difference is due to the fact that in the serial analysis Parse-σ dominates WSP, whereas in the parallel analysis the opposite ranking holds. The parallel analysis is illustrated in (11), the serial one in is given in (12), where we pick up the derivation after main stress assignment and where we have left out lower-ranked Max-μ. 86 CatJL 18, 2019 Haike Jacobs

(11) Parallel parsing of HHHL /au.diː.toː.te/ as (ˌau)(ˌdiː)(ˈtoː)te /au.diː.toː.te/ WSP RhHrm Align-R Parse-σ /HHHL/ Main-σ a.(ˌHH)(ˈH)L *! σ * b.F(ˌH)(ˌH)(ˈH)L σ * c.(ˌH)H(ˈH)L σ * *! d.HH(ˈH)L σ * *!*

(12) Step 3: au.diː(ˈtoː)te going to (ˌau.diː)(ˈtoː)te HH(ˈH)L becoming (ˌHH)(ˈH)L /au.diː(ˈtoː)te/ Non-Fin RhHrm Non-Fin Align-R Parse-σ WSP Main-σ Main-Ft Main-σ

(ˌH)H(ˈH)L σ * *! F(ˌHH)(ˈH)L σ * * H(ˌH)(ˈH)L σ * *!

Please observe that the optimal initial foot resulting at step 3 in (12) contains a violation of WSP that cannot be improved upon further. Shortening at a next step, (ˌau.diː) going to (ˌau.di) and incurring a violation of lower-ranked Max-μ, is blocked by the higher-ranked, undominated RhHrm constraint. This means that, rather counter-intuitively, the serial analysis allows a trochaic foot (HH), while disallowing a (HL) one. This leads to yet another difference between the Parallel and the Serial OT analysis. It is not only in the stressing of two heavy syllables before the main stressed syllable, as observed by McCarthy, Pater and Pruitt, that the two analyses differ, but also in the stressing of a heavy and a light syllable before the main stressed syllable. Prince & Smolensky (2004: 79) remark: “We expect (ˌH-L)(ˈH) (ˌH) from HLHH. This avoids the Foot Form violation (HL) […] and it achieves complete foot parsing at the expense of Max-μ.” This is illustrated in (13) for a word like /kiː.wi.taː.tem/ civitatem ‘city’, where this time, given that the foot with main stress has a H peak in all three output forms, the displayed evaluations for Pk-Prom should be interpreted as referring to the initial foot with secondary stress. Cross-level interactions in Latin CatJL 18, 2019 87

(13) Parallel parsing of HLHH

/kiː.wi.taː.tem/ WSP RhHrm A l i g n - R Parse-σ Max-μ Pk-Prom /HLHH/ Main-σ a.F (ˌLL)(ˈH)(ˌH) σ * |L| b. (ˌHL)(ˈH)(ˌH) *! σ |H| c. (ˌH)L(ˈH)(ˌH) σ *! |H|

The serial analysis could never produce this result as it would require a (ˌHL) trochee at an intermediate step in the serial derivation. It is doubtful, however, whether shortening did apply to the initial vowel in such cases in Pre-classical Latin, all we know for sure is that the initial vowel was long in Classical Latin as witnessed by modern Romance reflexes such as French cité, Italian città, Spanish ciudad etc. which all show a long initial vowel, a short one would have merged as [e] with original Latin long [eː]. Furthermore, it should be observed that both analyses predict vowel shortening to take place in a pre-main stress LHL sequence. In the serial analysis, the Align-R (Main-σ) constraint, contrary to the main stress environment in (8) above, is not preventing LHL in the pre-main stress sequence from becoming optimally analysed as (ˌLH)L and then further improved by vowel shortening to surface as (ˌLH-)L. Vowel shortening in Pre-classical Latin was, however, restricted to final vowels and word-internally did affect heavy syllables (closed heavy syllables, but rarely heavy syllables with a long vowel (cf. Drexler 1969: 214; Mester 1994), only if they were immediately followed by main stress (cf. Allen 1973: 181; Lindsay 1894: 201-202). The word-internal de-weighting of closed, heavy syllables is therefore better analysed as clash avoidance and the shortening of final vowels as final vowel shortening (cf. Jacobs 2003 for a more detailed account). Such an analysis does not need to invoke intermediate trochaic feet (ˌLH), but also raises the question what it is that makes the final long vowel in cretic HLH words special. From a Classical Latin poetry perspective the obvious answer seems to be that these words, just as LLLH words, like balineum ‘bath’, cupiditas ‘lust’, maritimus ‘maritime’ or manipulum ‘a fistful’, simply do not fit in a dactylic hexameter. In Classical Latin poetry, these LLLH words simply do not occur just as words of a cretic type, HLH, were generally avoided. The only way in which LLLH words could occur in Classical poetry is in syncopated form. This is shown clearly by comparing LLLH words like manipulum to LHLH words like periculum ‘danger’, Both have main stress on the antepenultimate syllable, but differ in that [i] is long in periculum, but short in manipulum. A word like periculum frequently occurs in Vergil and Ovid both in syncopated and in unsyncopated (always as LHLL pericula) form, but manipulum only occurs in syncopated form and never in its full form. Let us briefly summarize in (14) the differences between the serial and the parallel analyses of vowel shortening in Pre-classical Latin. 88 CatJL 18, 2019 Haike Jacobs

(14) Summarizing the differences between the parallel and serial vowel shortening analyses Different predictions Parallel Serial HHHL /au.diː.toː.te/ (ˌau)(ˌdiː)(ˈtoː)te (ˌau.diː)(ˈtoː)te (ˌH)(ˌH)(ˈH)L (ˌHH)(ˈH)L HLHH /kiː.wi.taː.tem/ (ˌki.wi)(ˈtaː)(ˌtem) (ˌkiː)wi(ˈtaː)(ˌtem) (ˌLL)(ˈH)(ˌH) (ˌH)L(ˈH)(ˌH)

The most problematic aspect of both the serial and the parallel analysis of Latin vowel shortening is that a penultimate unstressed vowel is not uniformly analysed as being unstressed, but is sometimes analysed as having secondary stress, that is, as (ˈH)(ˌLL) in HLL words or in HLH words, analysed as (ˈH)(ˌLH-) in Pre-classical and as (ˈH)L(ˌH) or (ˈH)(ˌLH) (cf. footnote 1 above) in Classical Latin. This is due to the fact that the constraint Non-Fin (Main-Ft) is specified for the foot with main stress. In Jacobs (2004) it was proposed that by replacing the constraint Non-Fin (Main-Ft) by the constraint Non-Fin (Ft) (a foot may not be final) a uniform analysis of unstressed prefinal vowels as effectively unstressed obtains. The replacement was shown to be independently required for typology in Jacobs (1999) on the basis of the observation that leftward footing never leads to pre-antepenultimate stress, but that only rightward footing ever does so. McCarthy (2003) argues for a similar replacement based on arguments against foot extrametricality. A uniform analysis of unstressed prefinal vowels as unstressed is motivated by Latin vowel deletion or syncope to which we will turn in the next section where we will discuss a cross-level interaction that seems to hold for Latin vowel deletion.

2. Vowel deletion and cross-level interactions In this section we turn to another potential cross-level interaction in Latin vowel deletion. We will first, in section 2.1, motivate the use of the uneven trochee and a foot-based vowel deletion constraint. After that, we will show in section 2.2 why a serial analysis needs to be preferred over a parallel one. Section 2.3 discusses the cross-level constraint interaction involved in Latin vowel deletion.

2.1. Vowel deletion, even and uneven trochees and foot-based deletion constraints In Latin, vowel deletion affected the unstressed post-tonic vowel in all the prosodic word types illustrated in (15).

(15) Vowel deletion in Latin a. ˈHLL type lāmĭnă ‘plate’ becoming [ˈlamna] b. ˈLLH type sŏlĭdus ‘solid’ becoming [ˈsoldus] c. ˈHLH type ārĭdus ‘dry’ becoming [ˈardus] Cross-level interactions in Latin CatJL 18, 2019 89

As mentioned above, by replacing the constraint Non-Fin (Main-Ft) (a foot with main stress may not be final) by a constraint Non-Fin (Ft) (a foot may not be final), the unstressed penultimate vowels are analysed as unstressed, and, if the constraint RhHrm is ranked below Parse-σ, they all occupy the weak part of a foot in Jacobs (2004). That Parallel OT analysis is illustrated in (16).

(16) Latin stress with Non-Fin (Ft) and RhHrm dominated by Parse-σ

(16a) /HLL/ Non-Fin Ft-Bin Align-R WSP Parse-σ RhHrm (Ft) Main-σ lāmĭnă (ˈH)(ˌLL) *! σ σ (ˌH)(ˈLL) *! σ (ˈH)LL σ σ * *! F(ˈHL)L σ σ * *

(16b) /LLH/ Non-Fin Ft-Bin Align-R WSP Parse -σ RhHrm (Ft) Main-σ sŏlĭdus

(ˈLL)(ˌH) *! (ˈL)LH *! σ σ * * F(ˈLL)H σ σ * L(ˈLH) *! σ * *

(16c) /HLH/ NonFin Ft-Bin Align-R WSP Parse -σ RhHrm (Ft) Main-σ ārĭdus (ˈH)L(ˌH) *! σ σ * (ˌH)L(ˈH) *! * H(ˈL)H *! σ * * (ˈH)LH σ σ * *! (ˈH)(ˌLH) *! σ σ * F(ˈHL)H σ σ * *

The relative ranking of the constraints RhHrm and Parse-σ thus produces moraic trochees (if RhHrm dominates Parse-σ) or uneven trochees (if Parse-σ dominates RhHrm). 90 CatJL 18, 2019 Haike Jacobs

The assumption of a canonical even trochee, that is (H) or (LL) ones, and the claimed universal absence of uneven or (HL) trochees (The Iambic-Trochaic Law Hayes (1995)) was originally, among other things, based upon trochaic shortening in languages such as Fijian, where a HL sequence by shortening becomes an even trochee, as in (ˈmbu:) ‘grandmother’ with a heavy syllable turning into a light one in /ˈmbu: ŋɡu/ à (ˈmbu-ŋɡu), /HL/ à (LL), ‘my grandmother’ (cf. Hayes 1995: 145). Under a Serial OT analysis of trochaic shortening in Fijian or of tri-syllabic laxing in English, if the shortening is analysed as taking place for foot harmony reasons, an uneven trochee has to be allowed at an intermediate step of the derivation and hence cannot be excluded as a possible foot type.2 In the history of Latin, syncope never affected stressed vowels and all unstressed penultimate vowels in the word types in (15) were subject to deletion, which is the main reason why they should uniformly be analysed as unstressed. However, it is not immediately clear whether the constraint responsible for vowel deletion should really make reference to the foot. Either a constraint *Weak-V in ft (cf. Jacobs 2004), no weak vowel in a foot, or a constraint *Weak-V, no vowel in an open, unstressed non-final syllable (cf. McCarthy 2008), if ranked above a constraint militating against vowel deletion, Max-V(owel), successfully accounts for Latin vowel deletion, and will target the same vowel in the word types in (15), whether analysed with even or with uneven trochees, as illustrated in (17).

(17) Even or uneven trochees in Latin even trochees uneven trochees a. HLL type lāmĭnă (ˈlā)mĭnă (ˈlāmĭ)nă b. LLH type sŏlĭdus (ˈsŏlĭ)dus (ˈsŏlĭ)dus c. HLH type ārĭdus (ˈā)rĭdus (ˈārĭ)dus

Jacobs (2019) argues for the need of a foot-based vowel deletion constraint based on vowel deletion in words such as balineum ‘bath’ and opitumus ‘the best’. These words, that is words of four syllables with the first three syllables light, had initial stress in early Classical Latin, when all other word types did have penult or antepenult stress (cf. Lindsay 1894 and Jacobs 2003 for a formal OT account). When no reference is made to the foot for vowel deletion, output candidates (ˈbali) num or (ˈbal)num for underlying /balineum/ will be preferred over actual (ˈbal) neum. This is illustrated within parallel OT in (18) and (19). Jacobs (2019) deviates from the parallel and serial analyses discussed in section 1 by using the constraints Align W/R and Align W/L (cf. McCarthy and Prince (1993), Kager (1999) which require the word to be right- or left-aligned with a foot. In early Classical Latin Align W/L was ranked above Align W/R, whereas the opposite ranking holds for Classical Latin.

2. The uneven trochee has been motivated in, among others, Repetti (1998) and Revithiadou (2004), see also Hyde (2011), Mellander (2003) and Iversen, Patel and Ohgushi (2008) for more discussion. Cross-level interactions in Latin CatJL 18, 2019 91

(18) *Weak-V and vowel deletion in words like balineum

ba.li.ne.um ALIGN-W/L *Weak-V Max-V ALIGN-W/R Parse-σ

1.(ˈba.li)ne.um *! * σ σ * * 2.ba (ˈli.ne)um σ! * * σ * * 3.(ˈbal)ne.um *! * σ σ * * 4.(ˈbal.ne)um *! * σ * 5.(ˈba.li)num *! * σ * 6.F(ˈbal)num * * σ *

(19) *Weak-V in Ft and vowel deletion in words like balineum ba.li.ne.um ALIGN-W/L *Weak-V in Ft Max-V ALIGN-W/R Parse-σ

1.(ˈba.li)ne.um *! σ σ * * 2.ba (ˈli.ne)um σ! * σ * * 3.F(ˈbal)ne.um * σ σ * * 4.(ˈbal.ne)um *! * σ * 5.(ˈba.li)num *! * σ * 6.(bal)num * *! σ *

As illustrated in (19) a foot-based vowel deletion constraint is required in a parallel analysis. If we switch from a Parallel to a Serial perspective, a number of remarks are in order. An output candidate (ˈbal.ne)um (18/19-4) is not a viable candidate in Harmonic Serialism when the derivation goes directly from (ˈba.li)ne.um à (ˈbal.ne)um. This would imply two modifications at the same time: deletion of the vowel and footing or incorporation of the pre-final syllable into the main stress foot. If this is done stepwise, that is, vowel deletion after main stress, (ˈba.li)ne.um à (ˈbal)ne.um, and, then, foot changing, (ˈbal)ne.um à (ˈbal.ne)um, it depends on how re-footing is conceived off. If it implies deleting the foot and then constructing a new one, it is excluded as a possible derivation, given that foot deletion does not constitute a harmonic improvement. If it is seen, not as foot deletion and foot construction, but as the incorporation of an unstressed syllable into an existing foot, it is allowed, depending on the ranking of relative ranking of RhHrm and Parse-σ. If Parse-σ dominates RhHrm, a step from (ˈbal)ne.um to (ˈbal.ne)um is indeed an improvement in harmony. The more likely historical scenario seems to be that after vowel deletion at some point the word was indeed restructured or reanalysed in terms of footing in this way. In the evolution of 92 CatJL 18, 2019 Haike Jacobs balineum, there was no second round of vowel deletion given that vowel gliding (cf. Jacobs 2016) did turn the vowel in the pre-final syllable into a glide and changed the word into [balnju]. A second round of syncope did occur in a word like *deksiteros which in Early Classical Latin with initial stress was presumably parsed as (ˈdeksi)teros and after syncope at some point as (ˈdekste)ros and, after a second round of syncope as (ˈdeks)tros, ended up as dexter ‘right’ in Classical Latin (cf. Shen 2012). Similar remarks hold for output candidate (ˈbal)num (18/19-6). Although possible from a parallel perspective, the single-modification-at-the-same-time limitation of the serial analysis excludes it from being a viable output candidate. However, even if output candidate (18/19-6) and (18/19-4) were omitted from evaluation, then, still, with the *Weak-V constraint (thus without referring to the foot), output candidate (18-5), (ˈba.li)num, would wrongly be selected in a serial analysis. We therefore need, both in a parallel and in a serial analysis, an uneven trochee and a vowel deletion constraint that makes reference to the foot. Further motivation for the necessity of foot-based constraints can be found in De Lacy (2002) who motivates foot-based constraints on the basis of Dutch vowel reduction. Before turning to the cross-level interaction in section 2.3, we will first show why a serial analysis of vowel deletion has to be preferred over a parallel account.

2.2. Why a serial analysis of vowel deletion should be preferred over a parallel analysis In Jacobs (2008) it is shown that syncope in cases such as sōlĭcŭlum [soːˈli.ku.lum] > sōlĭclum [soː(ˈli)klum] ‘sun’ leads to surface prosodic opacity. In the syncopated form, main stress is on the light penultimate syllable, violating the constraint Ft-Bin, requiring that feet should be binary, that is, either consisting of two syllables or be heavy (consisting of two moras). Normally, if the penultimate syllable is light, as in locuples ‘rich’ or Patroclus ‘Proper name’ main stress is on the antepenultimate syllable. In Parallel OT, this is problematic, given that one and the same ranking could never evaluate surface forms with an identical prosodic surface shape differently, as illustrated in (20) and (21).

(20) Parallel parsing of Patroclus.

/LLH/ Non-Fin (Ft) Ft-Bin Align-R WSP Parse-σ RhHrm Main-σ

L(ˈLH) *! σ * * L(ˈL)H *! σ * * F(ˈLL)H σ σ * Cross-level interactions in Latin CatJL 18, 2019 93

(21) Problematic parallel parsing of so:ˈliklum. Stress is incorrectly predicted to shift. /HLH/ Non-Fin (Ft) Ft-Bin Align-R WSP Parse-σ RhHrm Main-σ

F(ˈHL)H σ σ * * LH(ˈL)H *! σ ** (ˈH)LH σ σ **! H(ˈLH) *! σ * *

Jacobs (2008) presents a serial analysis using OT with candidate chains (OT- CC) (cf. McCarthy 2007). Parallel transparent competitors to opaque [soːˈli.klum] such as sōlĭclum, [ˈsoːli.klum], with a stress shift, but with transparent surface stress, or [ˈsoːl.klum], with two vowels deleted, but also with transparent surface stress, constitute ill-formed candidate chains and are therefore no real competitors. This is in a nutshell the analysis. A candidate chain is fine, there is harmonic improvement, given that the high-ranked constraint HeadWord, requiring a word to have stress is no longer violated. A chain is also fine. The further modification of soː(ˈli.ku)lum into soː(ˈli)klum is again an harmonic improvement, given that *Weak-V dominates Max-V. A chain is, however, ill-formed as the change from soː(ˈli)klum to (ˈsoːli)klum contains two modifications at the same time: stress deletion and stress assignment. If the two changes are done separately, that is, soː(ˈli)klum, soːliklum, (ˈsoːli) klum, then the change from soː(ˈli)klum to soːliklum, again, is not an harmonic improvement given the high-ranked HeadWord constraint. Similar remarks hold for a transparent surface stress competitor [ˈsoːl.klum]. The serial OT analysis thus has a clear advantage over a Parallel OT analysis: metrically conditioned vowel deletion can never lead to stress shift, not even when that would lead to surface prosodic transparency (cf. McCarthy 2008 for a similar argumentation). Let us next turn to the cross-level constraint interaction in Latin vowel deletion.

2.3. Vowel deletion and cross-level interactions Let us now address the cross-level interaction in Latin vowel deletion. The crucial ranking parts required for the OT-CC analysis in Jacobs (2008) are illustrated in (22) and (23). In (22), we first give the two candidate chains for transparent vowel deletion in (ˈsoli)dum > (ˈsol)dum for which the relative ranking of *Weak-V and Max-V with respect to Ft-Bin is irrelevant. 94 CatJL 18, 2019 Haike Jacobs

(22) *Weak-V and Max-V may be ranked below Ft-Bin for transparent solidum > soldum so.li.dum Ft-Bin *Weak-V Max-V

1. *! 2.F *

However, the ranking in (22) cannot be the ranking for opaque vowel deletion in soliculum. Here, it is crucial that both *Weak-V and Max-V are ranked above Ft-Bin, as illustrated in (23).

(23) Both *Weak-V and Max-V must be ranked above Ft-Bin for opaque soˈliclum so.li.ku.lum *Weak-V Max-V Ft-Bin

1. *! 2.F * *

In (23) the two well-formed candidate chains for opaque soliclum are evaluated. Chain 23-1, is a well-formed chain. The further modification of soː(ˈli.ku)lum into soː(ˈli)klum, as in chain 23-2, constitutes also an harmonic improvement, given that *Weak-V dominates Max-V, but crucially, for that analysis to go through, the constraints *Weak-V and Max-V both have to dominate the constraint Ft-Bin, which will have no consequences for transparent vowel deletion in (22), given that chain 22-2 does not have a violation of Ft-Bin. If we translate this analysis into Harmonic Serialism, then, the full constraint ranking in which HeadWord dominates *Weak-V › Max-V, which in their turn, dominate Ft-Bin, raises the question why words like solidum and soliculum are footed as (ˈsoli)dum and soː(ˈlicu)lum to begin with, if the *Weak-V › Max-V ranking judges them to be suboptimal? Why do the wrong footing first and then improve on it by vowel deletion? In order to answer this question, it is first of all important to mention that it is not the case that (ˈLL) or (ˈHL) feet did never surface in Latin. Vowel deletion, which was an optional process that started in Pre-classical Latin and was active throughout the history of the language as witnessed by the optimus, balneum (Pre- classical) and periclum and maniplum (Classical) cases discussed above, started out by targeting mainly high vowels and by being sensitive to phonotactic constraints. Vowel deletion in words like vetulus ‘old’ only occurred in Late Latin (cf. the Appendix Probi “vetulus non veclus”). This means that, instead of the general vowel deletion constraint *Weak-Vowel in Ft, we need a more specific constraint, *Weak-high in Ft (targeting high vowels), and we also need to check vowel deletion for phonotactic compatibility. We can now assume that the suboptimal LL foot created in the derivation of soːlikulum à soː(ˈli.ku)lum à soː(ˈli)klum results directly from the footing constraints or in the case of the suboptimal HL foot in Cross-level interactions in Latin CatJL 18, 2019 95 the derivation of aːridus à (ˈaːri)dus à (ˈaːr)dus is enforced by Parse-σ, much in the same way as an uneven trochee would be required at an intermediate step in a serial analysis of trochaic shortening in Fijian or of tri-syllabic laxing in English as mentioned above. Therefore, what happens from a serial perspective is that a foot is first constructed and only after that, at a next step, inspected for possible vowel deletion, but both crucially cannot take place at the same time. This is visualized, after syllabification at step 1, in (24) for words like solidus and vetulus.

(24) Step 2: so.li.dus and ve.tu.lus going to (ˈso.li.)dus and (ˈve.tu)lus LLH becomes (ˈLL)H /LLH/ Head *Weak- Max-V Non-Fin Ft-Bin Align-R Parse-σ Word high in Ft (Ft) Main-σ

F1.(ˈLL)H σ σ * 2. LLH *! * * *

3. L(ˈL)H *! σ * * 4. L(ˈLH) *! σ *

The dotted line between the second and third constraint expresses the optionality of vowel deletion and indicates that the relative ranking between them is not fixed. The important thing to observe is that output candidate (24-3) is ruled out as it violates Ft-Bin. The foot (ˈLL)H is then inspected for possible improvement at step 3 in the derivation. In the case of a word like (ˈso.li.)dus vowel deletion constitutes a harmonic improvement as illustrated in (25), where we have left out Ft-Bin and where the two optimal output forms represent the optional status of vowel deletion.

(25) Step 3: (ˈso.li.)dus optionally going to (ˈsol)dus. (ˈLL)H may become (ˈH)H /(ˈso.li.)dus/ Head *Weak- Max-V Non-Fin Align-R Parse-σ Word high in Ft (Ft) Main-σ

F1.(ˈsol)dus * σ * 2. solidus *! * * *

F3.(ˈsoli)dus * σ σ * 4. *L(ˈLH)

Please observe that candidate (24-4), which we have repeated as (25-4), is no longer an option at step 3, given that it would require two modifications at one step: the deletion of the foot (ˈsoli) and the assignment of the foot (ˈlidus). Please observe also that optional vowel deletion does not take place in Classical Latin for words 96 CatJL 18, 2019 Haike Jacobs like (ˈve.tu)lus at step 3, due to higher-ranked phonotactic constraints which, for reasons of space, have been left out in (24) and (25). Given that in this serial analysis the vowel deletion constraint is only evaluated after a foot has been assigned, no violations have been provided for the two constraints *Weak-high in Ft and Max-V in (24). To make this more precise we repeat in (26) tableau (24), but with the violations for these two constraints, if one were able to evaluate them at step 2 and an output form (26-5) in which the vowel has been deleted.

(26) Step 2: so.li.dus going directly to (ˈsol)dus? /LLH/ Head *Weak- Max-V Non-Fin Ft- Align-R Parse-σ Word high in Ft (Ft) Bin Main-σ

F1.(ˈsoli)dus * σ σ * 2. solidus *! * * *

3. so(ˈli)dus * *! σ * * 4. so(ˈlidus) * *! σ * F5. (ˈsol)dus * σ *

If the derivation at step 2 would work like this, this would imply that at step 2 not only two modifications are required: deleting a vowel and assigning a foot, but also that footing would need to do something that never occurs in stress systems, that is, footing would need to inspect segmental features like [+high] and phonotactic well-formedness constraints in order to decide whether to assign a monosyllabic foot (as in (26-5) or not as in (ˈve.tu)lus. The derivation of opaque of soːlikulum à soː(ˈli. ku)lum à soː(ˈli)klum proceeds essentially along the same lines. The intermediate foot (ˈli.ku) is enforced by the regular footing constraints and the surface violation of Ft-Bin in soː(ˈli)klum is enforced by the higher-ranked vowel deletion constraint *Weak-high in Ft. This illustrated for step 2 and 3 in the derivation in (27) and (28).

(27) Step 2: soːli.ku.lum going to soː(ˈli.ku)lum. /soː(ˈli.ku)lum / Head *Weak- Max-V Non-Fin Ft- Align-R Parse-σ Word high in Ft (Ft) Bin Main-σ

1 (ˈsoːli)ku.lum σ σ σ! **

2(ˈsoː)li.ku.lum σ σ σ! *** 3 soː(ˈli)ku.lum *! σ σ *** 4 soː.li(ˈku)lum *! σ *** 5 soːli.ku.lum *! ****

F6 soː(ˈli.ku)lum σ σ ** Cross-level interactions in Latin CatJL 18, 2019 97

(28) Step 3: soː(ˈli.ku)lum optionally realised as soː(ˈli)klum. /soː(ˈli.ku)lum / Head *Weak- Max-V Non-Fin Ft- Align-R Parse-σ Word high in Ft (Ft) Bin Main-σ

F1 soː(ˈli)klum * * σ ** 2 soːli.ku.lum *!

F3 soː(ˈli.ku)lum * σ σ ***

Let us next turn to a second case in the phonology of Latin where a cross-level constraint interaction can be observed: vowel gliding.

3. Vowel gliding and cross-level interactions The hiatus resolution by the vowel gliding of the non-low vowels (i, e, u, o) before another vowel in Late Latin, generally dated in the first century AD, is illustrated in (29) where the French reflexes have also been indicated (cf. Pope 1934: 102; Lindsay 1984: 142; Bourciez 1974: 28; Fouché 1958: 28, among others).

(29) Vowel gliding of i, e, u, o in hiatus Classical Latin Late Latin Old French French filium [ˈfilium] [ˈfilju] fil fils ‘son’ vineam [ˈwineam] [ˈwinja] vigne vigne ‘vine’ vidua [ˈwiduam] [ˈwedwa] veve, vedve veuve ‘widow’ coacticare [koaktiˈkare] [kwaktiˈkare] cachier cacher ‘to hide’ The relation between vowel gliding in (29) and stress displays a paradox that has never been properly understood. All traditional descriptions of the facts agree that in order for vowel gliding to take place the non-low vowel in hiatus had to be an unstressed one. Bourciez (1974: 148) states: “Tout ĭ ou ě atone en hiatus avait pris la valeur de y…” [every unstressed i or e in hiatus has taken the value of j,] and Pope (1934: 102) mentions “that unstressed ĕ, ǐ, ŏ, ŭ standing in hiatus with a following vowel gradually lost syllabic value and consonantalised”. If vowel gliding was indeed restricted to unstressed vowels, it becomes immediately clear why it did not take place in disyllabic words, as illustrated by the forms in (30).

(30) No vowel gliding in disyllabic words Classical Latin Late Latin Old French French Gloss deum [ˈdeum] [deu] deo, de, dieu, dieu ‘God’ gruem [ˈgruem] [grue] grue grue ‘crane’ ego [ˈego] [eo] eo, io, jeu, je ‘I’ diem [ˈdiem] [die] di -di- -day- mea [ˈmea] [mea] meie, moie mienne ‘mine’ piam [ˈpiam] [pia] pie pieuse ‘devout’ viam [ˈwiam] [wea] veie voie ‘way’ 98 CatJL 18, 2019 Haike Jacobs

The forms in (31) further show that the absence of vowel gliding in (30) is unrelated to the vowel in hiatus being in initial position. That is, vowel gliding did take place word-initially.

(31) Vowel gliding word-initially Classical Latin Late Latin Old French French Gloss diurnum [di.ˈur.num] [dj/dʒɔrnu] jorn, jor jour ‘day’ georgium [ge.ˈɔr.gi.um] [gj/dʒɔrgj/dʒu] Georges Georges id. quietum [kwi.ˈe.tum] [kwjeta] coi tranquille ‘calm’ duodecim [du.ˈo.d.ekim] [dwodetsi] dose douze ‘twelve’ The forms in (32), finally, show that, paradoxically, stressed vowels were subject to vowel gliding. (32) Vowel gliding of stressed vowels Classical Latin Late Latin Old French French Gloss filiolum [fiˈli.o.lum] [filˈjolu] filuel filleul ‘godson’ tiliam *tiliolu [tiˈli.o.lum] [tilˈjolu] tilleul tilleul ‘lime’ gladiolum [glaˈdi.o.lum] [glaˈdjolu] glaïeul glaïeul ‘gladiolus’ mulierem [muˈli.e.rem] [muˈljere] moillier --- ‘wife’ Traditional analyses have never been able to account for the vowel gliding of stressed vowels. Jacobs (2016) shows that the paradoxical relationship between stress and vowel gliding receives a straightforward explanation when it is considered from a serial instead of a parallel OT perspective. It allows the prosody of the language to determine whether or not vowel gliding takes place, but, importantly, the vowel gliding itself can be formalized without any reference to stress at all. The constraints in (33) are needed to describe the vowel gliding. (33) Constraints for hiatus resolution No Hiatus : Avoid two adjacent hetero-syllabic vowels Max-μ : Do not delete a vowel’s mora

The hierarchy in (34a) is proposed for Classical Latin and vowel gliding that started in early Late Latin is accounted for by assuming that the hierarchy changed into the one in (34b) in Late Latin. This means that the only change in the grammar is related to vowel gliding, but that the prosodic grammar, a part of the hierarchy in (34) remained unchanged. (34a) Classical Latin hierarchy for stress and hiatus Non-Fin (Ft) >> Max-μ >> No Hiatus >> Ft-Bin >> Align-R Main- σ >> Parse-σ (34b) Late Latin hierarchy for stress and hiatus N on-Fin (Ft) >> No Hiatus >> Max-μ >>Ft-Bin >> Align-R Main-σ >> Parse-σ Cross-level interactions in Latin CatJL 18, 2019 99

In Classical Latin, the evaluation for a word like diurnum, converges stress- wise on the output form [di.(ˈur)num] in the second evaluation round, as illustrated in (35) and (36).

(35) Second step in Classical Latin: /di.ur.num/ becomes [di.(ˈur)num] / di.ur.num/ Non-Fin Max-μ No Ft-Bin Align-R Parse-σ (Ft) Hiatus Main-σ

1. (ˌdi).(ˈur) num * *! σ * 2.Fdi.(ˈur) num * σ * * 3. di.(ˈur)(ˌnum) *! * σ *

Tableau (36) shows that vowel gliding does not lead to further harmonic improvement in the next round of evaluation, but that grammar (34a) converges on [di.(ˈur)num].

(36) Third step for diurnum in Classical Latin: convergence, no vowel gliding

/ di.(ˈur)num/ Non-Fin Max-μ No Ft-Bin Align-R Parse-σ (Ft) Hiatus Main-σ

1.(ˌdi).(ˈur) num * *! σ * 2.Fdi.(ˈur)num * σ * * 3.(ˈdjur)num *! σ * 4.di.(ˈur)(ˌnum) *! * *

Output candidate (36-3) with vowel gliding is not harmonically improving, but worse when compared to winning output candidate (36-2) [di.(ˈur)num]. In Late Latin, however, given the changed ranking of the two relevant constraints, vowel gliding does lead to further harmonic improvement, as shown in tableau (37).

(37) Third step for diurnum in Late Latin (34b): vowel gliding

/ di.(ˈur)num/ Non-Fin No Max-μ Ft-Bin Align-R Parse-σ (Ft) Hiatus Main-σ

1.(ˌdi).(ˈur) num *! * σ * 2. di.(ˈur)num *! σ * * 3.F(ˈdjur)num * σ * 4.di.(ˈur)(ˌnum) *! * * 100 CatJL 18, 2019 Haike Jacobs

Tableau (37) shows that, in Late Latin, the output candidate (37-3) with vowel gliding is harmonically improving when compared to output candidate (37-2) [di.(ˈur)num] as it succeeds in avoiding a violation of the constraint No Hiatus. Let us now turn to the blocking of vowel gliding in disyllabic words (cf. (30) above), which straightforwardly results from the grammar in (34b). In (38), the fate of disyllabic viam during the second round of evaluation in Late Latin is illustrated.

(38) Third step for viam in Late Latin: convergence, no vowel gliding

/ (ˈvi)am/ Non-Fin No Max-μ Ft-Bin Align-R Parse-σ (Ft) Hiatus Main-σ

1 (ˈvi.am) *! * σ 2 (ˈvjam) *! * 3F(ˈvi)am * * σ *

The ranking of the constraints Non-Fin and Ft-Bin above the vowel gliding constraints straightforwardly accounts for the absence of vowel gliding in disyllabic words. Let us next look more closely at the effect of vowel gliding in the problematic forms in (32) above. The important thing to observe is that the shift of stress from the antepenultimate to the penultimate syllable does not imply literally a change in foot structure. That is, the foot is shortened by one syllable, but stress remains to be expressed by the same foot. In (39) we illustrate the third step in the derivation of [mu.(ˈli.e)rem] in Late Latin.

(39) Third step for mulierem in Late Latin: vowel gliding

/mu.(ˈli.e)rem/ Non-Fin No Max-μ Ft-Bin Align-R Parse-σ (Ft) Hiatus Main-σ

1mu.(ˈli.e)rem *! σ σ * * 2mu.(ˈlje.rem) *! * σ * 3Fmul.(ˈje)rem * * σ * 4*?(ˈmul.je)rem * σ σ * 5*?(ˈmul)je.rem * σ σ * *

As shown in (39-3), harmonic improvement by vowel gliding is possible in at the third step in the derivation for mulierem. A parallel account would predict, dependent on the relative ranking of Parse-σ and RhHrm, either candidate (39- 4) or (39-5) to be incorrectly selected as the optimal output. Both these output candidates are ruled out as possible output forms in Harmonic Serialism, given that Cross-level interactions in Latin CatJL 18, 2019 101 they imply two modifications at the same time, instead of the allowed single one single modification. In order to get from input /mu(ˈli.e)rem/ in (39) to unwanted [(ˈmul.je)rem] in (39-4) or to [(ˈmul)je.rem] in (39-5) two modifications are required at step 3 in the derivation: removal of stress from the antepenultimate syllable and assignment of stress to the first syllable, of which the first modification is not allowed. If stress was first undone, in one step, and then reassigned later, in a second step, the first step would not be harmonically improving, given the high- ranked Head-Word constraint, which would be violated by an intermediate form without stress or without a foot. This is a clear advantage of Serial OT over Parallel OT: once stress has been assigned, it can no longer be undone, as that would require two modifications at the same time. Let us summarize thus far. The link between glide resolution and stress is an indirect one. Vowel gliding is not directly metrically conditioned (in the sense that only unstressed vowels are subject to it), in fact, it is not metrically conditioned at all, but applies to stressed and to unstressed vowels alike. It is the prosodic system of the language that determines whether or not vowel gliding applies. More specifically, in words like deum and viam in (30), vowel gliding did not take place, because going from (ˈvi.)am to (ˈvjam) is not a harmonic improvement, contrary to the other cases where gilding did take place, as illustrated above. However, here too, we are facing a problematic cross-level interaction. The basic issue is this: the tableaux (35) to (39) start out by stressing or footing the underlying form. In order to do footing the forms need to be syllabified, given that syllable structure is required for stressing to take place, but then the following question arises: why would, for instance (38), to begin with be syllabified with a hiatus if the constraint against it is high ranked? Why first syllabify the input string as involving a hiatus and only then harmonically improve it? Contrary to stress, or feet, as discussed above, syllable structure needs to escape the requirement that once assigned it can no longer be undone. Resyllabification does occur without problems, as in (39-3) where [l] is onset before, but coda after vowel gliding, contrary to the shifting of stress to the initial syllable in (39-4). McCarthy (2007: 76) assumes that “syllabic parsing is not subject to faithfulness and therefore that syllables may be freely assigned and reassigned in chains. Metrical parsing […] is subject to faithfulness” and this is what would be needed for our analysis to go through as well. We would like to exclude (39-4) on the grounds that metrical parsing is subject to the harmonic improvement requirements in Harmonic Serialism, while at the same time maintaining a syllabified input with the possibility of freely changing it elsewhere, as in (35) to (39). So we need to assume that syllable structure can be freely assigned and reassigned at each step in the derivation. We assume that when gliding began syllable structure is present in the input forms. There is one piece of evidence that this is really needed. If we compare Classical Latin iam [jam] ‘already’ or ius [jus] ‘law’ with viam [wi.am] and piam [pi.am] it is clear that the distribution of high vowels and glides, which was predictable in Classical Latin (Steriade 1988), ceased to be so a at a later point in time when vowel gliding started. If the explanation for the non-gliding in the latter two words is related to high-ranked Non-Fin (Ft), it must be the case that the first two words, albeit with 102 CatJL 18, 2019 Haike Jacobs predictable syllabification in Classical Latin, were stored with syllable structure in Late Latin.

4. Summary In this paper, after having critically reviewed the analysis of vowel shortening proposed by McCarthy, Pater & Pruitt (2016), we have addressed two cases of cross- level interactions in Latin, one in vowel deletion and another one in vowel gliding. We have argued that a serial analysis has to be preferred over a parallel analysis, given that it straightforwardly not only accounts for prosodic opacity resulting from vowel deletion and from vowel gliding, but also for the fact that both stressed and unstressed vowels could be subject to vowel gliding.

References Allen, W. Sidney. 1973. Accent and Rhythm. Cambridge: Cambridge University Press. Bourciez, Edouard & Jean Bourciez, 1967 [1974]. Phonétique française. Paris: Klincksieck. Drexler, Hans. Die Iambenkürzung: Kürzung der zweiten Silbe eines iambischen Wortes, eines iambischen Wortanfangs. Hildesheim: Olms. Fouché, Pierre, 1958. Phonétique historique du français. Paris: Klincksieck. Hayes, Bruce. 1995. Metrical Stress Theory. Principles and Case Studies. Chicago: The University of Chicago Press. Hyde, Brett. 2011. The Iambic-Trochaic Law. In Marc van Oostendorp, Colin Ewen, Elizabeth Hume & Keren Rice (eds.). The Blackwell Companion to Phonology. Volume 2 Suprasegmental and Prosodic Phonology, 1052-1077. Iversen, John, Aniruddh Patel & Kengo Ohgushi. 2008. Perception of rhythmic grouping depends on auditory experience. The Journal of the Acoustical Society of America 124: 2263-2271. Jacobs, Haike. 1999. Constraining constraints: NonFinality and the Typology of Foot- extrametricality. In Renée van Bezooijen & René Kager (eds.) Linguistics in the Netherlands 1999, 111-129. Amsterdam: John Benjamins. Jacobs, Haike. 2003. The Emergence of Quantity-Sensitivity in Latin: Secondary Stress, Iambic Shortening and theoretical implications for ‘mixed’ stress systems. In E. Holt (eds.). Optimality Theory and Language Change, 229-247. Dordrecht: Kluwer. Jacobs, Haike. 2004. Rhythmic Vowel Deletion in OT: Syncope in Latin. Probus 16: 63-90. Jacobs, Haike. 2008. Sympathy, comparative markedness, OT-CC and Latin syncope. Probus 20: 235-255. Jacobs, Haike. 2016. L’interaction entre le système d’accentuation et la consonification des voyelles en hiatus dans la phonologie historique du français. Diachroniques 5: 79-103. De Lacy, Paul. 2002. The Formal Expression of Markedness. Ph.D. dissertation, University of Massachusetts Amherst. Lindsay, Wallace. 1894 [1963]. The Latin language. New York: Hafner. 2nd edition. McCarthy, John. 2003. OT constraints are categorical. Phonology 20: 75-138. Cross-level interactions in Latin CatJL 18, 2019 103

McCarthy, John. 2007. Hidden Generalizations. Phonological Opacity in Optimality Theory. London: Equinox. McCarthy, John. 2008. The serial interaction of stress and syncope. Natural Language and Linguistic Theory 26: 499-546. McCarthy, John. 2010. Harmonic Serialism Supplement to Doing Optimality Theory. . McCarthy, John, Joe Pater & Kathryn Pruitt. 2016. Cross-level interactions in Harmonic Serialism. . Mellander, Evan. 2003. (HL)-creating processes in a theory of foot structure. The Linguistic Review 29: 243-280. Mester, Ralph Armin. 1994. The Quantitative Trochee in Latin. Natural Language and Linguistic Theory 12, 1-61. Pope, Mildred, 1934. From Latin to Modern French. Manchester: Manchester University Press. Prince, Alan & Paul Smolensky. 1993 [2004]. Optimality Theory. Constraint Interaction in Generative Grammar. Oxford: Blackwell. Repetti, Lori. 1998. Uneven Trochees in Latin: Evidence from Romance Dialects. University of Venice Working Papers in Linguistics 8: 95-119. Revithiadou, Anthi. 2004. The Iambic/Trochaic Law revisited. In Boban Arsenijevic, Noureddine Elouazizi, Martin Salzmann & Mark de Vos (eds.). Leiden Papers in Linguistics 1.1: 37-62. Shen, Ranjan. 2012. Reconstructing phonological change: duration and syllable structure in Latin and vowel reduction. Phonology 29: 465-504. Steriade, Donca. 1984. Glides and Vowels in Romanian. Proceedings of the Tenth Annual Meeting of the Berkeley Linguistics Society, 47-64.

Catalan Journal of Linguistics 18, 2019 105-129

Stressed Clitic Pronouns in Two Spanish Varieties: A perception study*

Laura Colantoni University of Toronto [email protected] José-Ignacio Hualde Ane Icardo Isasa University of Illinois at Urbana-Champaign [email protected] [email protected]

Received: December 10, 2018 Accepted: June 6, 2019

Abstract

Clitic elements are prosodically attached to another phonological word. As such, they do not carry their own stress. In general, Spanish non-prepositional object pronouns (e.g. me/te/le/lo/ se) meet the definition of clitic, whether preposed (proclitic) or postposed (enclitic) to the verb. Nevertheless, in several Romance varieties, including some Spanish varieties, the stress may surface on the clitic in verb+clitic(s) groups. We refer to this phenomenon as Enclitic Stress Shift (ESS). In this paper, we report on the perception of ESS by Spanish speakers from Argentina and Spain and explore the acoustic dimensions that may affect this perception. The discussion includes comparisons with other Romance languages and implications for sound change, such as the changes in stress patterns in the development of future and conditional tenses. Keywords: stress perception; clitics; secondary stress; Spanish phonology; prominence; Argentine Spanish; Peninsular Spanish

Resum. Clítics pronominals accentuats en dues varietats del castellà: un estudi de percepció

Els elements clítics s’adjunten prosòdicament a una altra paraula fonològica, per tant, no porten accent propi. En general, els pronoms no preposicionals amb funció d’objecte (e.g. me/te/le/lo/ se) compleixen la definició de clític, tant si van davant (proclític) com darrera (enclític) del verb. Però en algunes varietats romàniques, incloses algunes de l’espanyol, l’accent pot aparèixer sobre el clític del grup verb+clític(s). Ens referim a aquest fenomen com a desplaçament de l’accent enclític (ESS, en anglès). En aquest article, presentem un estudi de percepció entre parlants d’espanyol de l’Argentina i d’Espanya i examinem les dimensions acústiques que poden afectar aquesta percepció. La discussió inclou comparacions amb altres llengües romàniques i implicacions en

* We are thankful to Maria-Rosa Lloret and to a second, anonymous, reviewer for their very useful comments and observations. We also want to thank Marissa Barlaz for statistical advice and Vanina Machado Araujo for her assistance in testing the Argentine Spanish speakers.

ISSN 1695-6885 (in press); 2014-9718 (online) https://doi.org/10.5565/rev/catjl.260 106 CatJL 18, 2019 Laura Colantoni; José-Ignacio Hualde; Ane Icardo Isasa relació amb el canvi fonètic històric, com el que presenten els patrons accentuals en l’evolució de les formes verbals de futur i condicional. Paraules clau: percepció de l’accent; clítics; accent secundari; fonologia de l’espanyol; promi- nència; espanyol de l’Argentina; espanyol peninsular

Table of Contents 1. Introduction References 2. Perception experiment Appendix 1 3. Discussion Appendix 2. List of stimuli

1. Introduction Although clitics are by definition unstressed elements that prosodically attach to a host word, many studies have claimed that in verb+clitic sequences, as those illustrated in (1), the stress may shift to the clitic in a process that we call Enclitic stress shift (ESS):

(1) Dáme ‘give’ – dàmeló1 ‘give it to me’ Dár ‘to give’ – dárle ‘to give her/him’ – dàrseló ‘to give it to her/him’ Dándo ‘giving’– dàndolé ‘giving to her/him’

As we will discuss below, this process has been reported in a variety of Romance languages, including some Spanish varieties. In Argentine Spanish, ESS appears to be more frequent than its alternative (stress on the verbal host) and acoustic evidence suggests that the clitic bears primary rather than secondary stress (Colantoni & Cuervo 2013; Klassen & Patience 2016). It is not clear, though, to what extent this phenomenon is observed in other Spanish-speaking varieties and whether speakers of the different varieties are sensitive to ESS. Thus, our goal here is to investigate the perception of verb+clitic sequences by two groups of participants who represent the Argentine and Peninsular varieties. The latter variety may be particularly relevant, since incipient evidence (Colantoni, Cuervo & Hualde 2010) suggests that verb+clitic sequences are sometimes doubly stressed in this dialect as well. In the remainder of this paper, we first describe the realization of verb+clitic sequences in Romance (section 1.1) including a discussion of whether the main prominence is on the verb, the clitic or both (section 1.2). In section 2, we report on a perception study, whose results are discussed in section 3, where we draw some parallels between our Spanish data and those reported in other varieties, and

1. Throughout this paper, we will use an acute accent mark to indicate (perceived) primary stress and a grave accent mark to indicate secondary stress. For clarity, we add accent marks even when they are not required orthographically. Stressed Clitic Pronouns in Two Spanish Varieties: A perception study CatJL 18, 2019 107 suggest that the stress shift observed in clitic-bearing verbal forms resembles the diachronic evolution of future and conditional tenses in Romance.

1.1. Clitics in Spanish and Romance Clitics are syntactic elements that prosodically attach to another phonological word. In Spanish direct and indirect object pronouns are prosodically attached to the verb. Clitic pronouns may appear either before the verb (as proclitic), as in (2a), or after the verb (enclitic), as in (2b). Proclisis is the rule except in affirmative imperatives, infinitives and gerunds. In periphrastic verbal forms containing a conjugated verb + infinitive or gerund (2c), the clitic can appear either before the conjugated verb or after the infinitive.2

(2a) Me lo dan ‘they are giving it to me’ (2b) Dame ‘give me’, dámelo ‘give it to me’ (but negative imperative no me lo des ‘don’t give it to me’) Darme ‘to give me’, dármelo ‘to give it to me’ Dándome ‘giving me’, dándomelo ‘giving it to me’ (2c) Me lo quieren dar ~ quieren dármelo ‘they want to give it to me’ Me lo van a dar ~ van a dármelo ‘they are going to give it to me’ Me lo están dando ~ están dándomelo ‘they are giving it to me’

Whereas, in general, the location of the primary stress in Spanish is restricted to one of the last three syllables of the word (e.g., camaleón ‘chameleon’, cucará- cha ‘cockroach’, libélula ‘dragonfly’, but not *líbelula, *cúcaracha, with stress beyond the three-syllable window), verb+enclitic forms with two enclitics may commonly show stress four syllables from the end (explicándomelo ‘explaining it to me’, explícamelo ‘explain it to me!’). In the rare cases where a verb carries three enclitic pronouns, the stressed syllable may even be located five syllables from the end, e.g. las polillas se me están comiendo los libros ‘moths are eating up my books’ à están comiéndosemelos ‘they are eating them up on me’. Enclitics are thus invisible to stress assignment on the verb and do not affect the location of the primary stress: comiéndo, except for the ESS phenomenon that we study here3 (but see below).

2. In addition to direct and indirect object clitics, the impersonal subject pronoun se, as in Se come muy bien en este restaurant ‘one eats very well in this restaurant’ can also be enclitic, e.g. Sigue comiéndose bien en ese restaurant ‘one still eats well in this restaurant’. 3. All Spanish verbs are stressed following the same pattern. Unlike in Italian, Spanish verbs do not contrast in the position of stress. The verb forms that may take enclitic pronouns are either paroxytonic (the gerund, e.g. comiéndo ‘eating’ and tú and usted(es) imperatives, e.g. cóme (tú), cóma (usted), cóman (ustedes) ‘eat!’) or oxytonic (the infinitive, e.g. comér ‘to eat’ and vos and vosotros/as imperatives, e.g. comé (vos), coméd (vosotros/as) ‘eat’). Usted ‘you-sg formal’ and ustedes ‘you-pl’ forms are used in all Spanish dialects. Vosotros/as ‘you-pl familiar’ is only used in Spain. As a second person familiar form, some dialects, including Spain, use tú and other dialects, including Argentina, use vos. See Appendix 1. 108 CatJL 18, 2019 Laura Colantoni; José-Ignacio Hualde; Ane Icardo Isasa

Clitics are words that cannot stand on their own prosodically, but must “lean” on a host. This implies lack of stress. It has thus been observed that clitic pronouns in Spanish cannot receive stress “even if contrastive focus is intended” (Hualde 2005: 258). Thus, an English sentence like I already told Mary and now I am going to tell YOU what I think could be translated in Spanish as Ya se lo he dicho a María y ahora te voy a decir a TI lo que pienso, but not as *Ya se lo he dicho a María y ahora TE voy a decir lo que pienso. Nevertheless, at least in some varieties, enclitics may in fact bear stress under certain conditions (Fernández Soriano 1999: 1256). Both stress displacement to the clitic and stress displacement to another syllable in the verbal root triggered by the presence of a clitic have in fact been reported in a wide variety of Romance dialects, including some Spanish dialects. Ordóñez & Repetti (2006: 170) summarize five different patterns in verbs bearing enclitics in Romance varieties, which range from stress stability to systematic stress shift to the last syllable. The three intermediate patterns reported by these authors include generalized penultimate stress shift, two-clitic penultimate stress shift (i.e., the stress shifts to the first clitic in a cluster) and mixed penultimate stress shift (i.e., the stress variably shifts to the penultimate syllable). Final stress shift, which is of interest here, given the patterns that have been reported in Argentine Spanish, has been observed in some varieties of Sardinian, as illustrated in (3), as well as in Gascon and Majorcan Catalan.

(3) Final stress shift in some varieties of Sardinian (Ordóñez & Repetti 2006: 170) Nárra. ‘say’ Narramí ‘tell me’ Narralá ‘say it’ Narramilá ‘say it to me’

Variation in stress patterns in verb+enclitic forms has been reported not only across languages but even among dialects of the same language. This is the situation, for instance, in Catalan (Torres Tamarit & Pons Moll 2018), where stress stability on the verb has been reported for Central Catalan, a tendency to shift the stress to the penultimate syllable has been described for Formentera Catalan and a strong tendency to shift to the final syllable is found in Minorcan and Majorcan Catalan (see also Nadeu et al. 2017). Whereas in Spanish, in general, enclitics are taken to be invisible for stress assignment purposes (e.g., using the examples above in (3): nárra, nárrame, nárra- la, nárramela), stress shift in verbs with enclitics has been reported in Argentine Spanish (Moyna 1999; Huidobro 2005; Colantoni & Cuervo 2013; Klassen & Patience 2016). In this variety, shift to the final clitic is available whenever there is at least one unstressed syllable between the stressed syllable of the verb and the clitic as in (4a).4 Sequences of two adjacent stressed syllables as in (4b) are

4. As an anonymous reviewer points out, it is interesting to note that, when ESS takes place, the stress always shifts to the final syllable, even though the unmarked pattern in Spanish is penultimate stress. Stressed Clitic Pronouns in Two Spanish Varieties: A perception study CatJL 18, 2019 109 ungrammatical. It is important to highlight, however, that even when ESS is available, it does not categorically happen. Colantoni & Cuervo (2013) report that ESS occurred in 71% of the tokens in their corpus where the phonological conditions for the process obtained. The authors identify some morphological and phonological factors as favoring ESS. For example, ESS is more frequent in imperatives than in infinitives and gerunds, and when the target word is followed by a pause than when followed by another word.

(4a) Encontràndoló. ‘finding it’ Dàmeló ‘give it to me’ Dàndomeló ‘giving it to me’ (4b) *Encontràrló ‘to find it’ *Dàmé ‘give me’

A question that remains to be explored is whether the likelihood of ESS also depends on the number of syllables counting from the stressed syllable of the verb to the end of the word; e.g. is stress shift more likely in explicándomelo ‘explaining it to me’ than in explicándolo? Further evidence of ESS in Argentine Spanish is its pervasive documentation in literary works and comics. In written texts, it is not uncommon to see imperatives with an orthographic accent on the clitic, as illustrated in these examples taken from La novela de Perón by Tomás Eloy Martínez (1991): […] le dije “Quedesé tranquila porque…” ‘I told her: stay calm, because…’ (p.33); Esfuercesé ‘make an effort’ (p. 52); imaginesé ‘imagine’(p. 44). (The standard forms are quédese, esfúercese, imagínese). Very little is known about how widespread ESS is in other Spanish varieties. For Peninsular Spanish, Menéndez Pidal (1918: 255) documents its existence in 16th century poetry, as well as in contemporary speech. He gives examples of emphatic imperatives with two stresses: “levántaté, entiéndemé” (Menédez Pidal 1918: 255), although not of infinitives or gerunds.5 Colantoni, Cuervo & Hualde (2010), in unpublished work, conducted a small pilot study and reported that ESS occurs in Peninsular Spanish, albeit less frequently than in Argentine Spanish (18% of possible instances in their Peninsular corpus, as perceived by the authors). The acoustic analysis of the data tentatively confirmed Menéndez Pidal’s intuition that these forms may receive two stresses, as in (5a). The authors also noticed that ESS appears to be restricted to phrase final position, as shown in (5b).

(5a) siéntatè ~ sièntaté ‘sit down’ (5b) vàmonós ‘let’s go’; escùchamé ‘listen to me’, but *vàmonós yá ‘let’s go now’; *escùchamé tú ‘listen to me’

5. Gili Gaya (1943) makes a similar comment regarding the double accentuation of these imperative forms. 110 CatJL 18, 2019 Laura Colantoni; José-Ignacio Hualde; Ane Icardo Isasa

Doubly stressed verb+clitic sequences have not been reported in other studies on Spanish, except for Peperkamp (1997), see below. Potentially double stressing constitutes a logical step in the change from stress on the verb to stress on the clitic (e.g. vámonos > vámonós > vamonós). The phenomenon may also be related to an emphatic intonational contour with a low tone on the verb and a prominent high accent on the last stressed syllable as one may find in utterances like vàmonos yá ‘let’s go now’ or dàme el líbro ‘give me the book’.

1.2. Prominence and cliticization The fact that doubly stressed verb+clitic sequences have been reported leads us to wonder whether ESS is better understood as a case of stress shift or as an instance of secondary stress. A variety of analyses and interpretations have in fact been proposed. Based on auditory transcriptions and comparisons with stress patterns in compounds, Moyna (1999) suggests that the case of Argentine Spanish represents an instance of secondary stress on the clitic, where the primary stress remains on the host. Colantoni & Cuervo (2013) and Klassen & Patience (2016) conducted acoustic studies of a corpus of Argentine TV soap operas and, based on the results obtained for three acoustic parameters (duration, intensity and f0), concluded that the primary stress is on the clitic in cases of ESS. Colantoni & Cuervo (2013) based their conclusions on patterns of perceived prominence; namely, they observed that in those cases in which the stress was perceived on the clitic, the vowel of the clitic was longer in duration and higher in intensity. To avoid this shortcoming and to determine whether there was evidence that clitics behaved in the same way as inflectional affixes, Klassen and Patience (2016) compared stressed clitics to unstressed clitics and to stressed words (e.g. mandameló ‘send it to me’; vs. man- dámelo vs. mandámelo vos) as well as to stressed and unstressed affixes. They largely replicated the results of Colantoni & Cuervo (2013) as concerns the realization of stressed clitics. Additionally, they found evidence in support that stressed vs. unstressed clitics pattern in the same way as stressed and unstressed affixes, in the sense that stressed elements in both cases were longer in duration, had higher intensity and higher pitch. A recent analysis of Majorcan Catalan (Nadeu et al. 2017), a variety in which final stressed clitics had been reported, also found evidence consistent with the claim that the clitic bears primary rather than secondary stress in this Catalan variety. In their study, 8 speakers of Central Catalan, a variety with no ESS, and Majorcan Catalan, which is reported to have ESS, were compared in their production of verb+clitic and verb+non-clitic pronoun sequences using a reading task. The stressed vowel of the verb and those of the enclitic and the following non-clitic pronoun were compared in their duration, mean f0 and vowel quality. The latter feature is important in Catalan, a language with systematic unstressed vowel reduction. An additional sort of evidence of the stress shifting to the clitic would be a change in the quality of the vowel of the verb, which should be reduced. Results revealed that post-verbal pronouns in Majorcan Catalan differed from Central Catalan in their Stressed Clitic Pronouns in Two Spanish Varieties: A perception study CatJL 18, 2019 111 increased duration and in the shape of the intonational contour associated with the clitic, which was convex in the former and concave in the latter. Additionally, the authors found evidence of vowel reduction associated with the verb in Majorcan but not in Central Catalan. This acoustic evidence led the authors to conclude that stress shifts to the clitic in Majorcan Catalan and that an interpretation of the facts as involving complete stress shift is more compatible with the acoustic data than an analysis involving either two prominences or a main prominence on the verb and a secondary prominence on the clitic. Alternative analyses have been proposed for ESS in Romance varieties spoken in Italy. Peperkamp (1997) suggested that the stress shift to the penultimate observed in Neapolitan when a clitic cluster is encliticized to the verb (e.g. narra- mílo) could be analyzed as an instance of double prominence (one on the verb and one on the clitic), with the prominence on the clitic being higher. Kim & Repetti (2013), instead, interpreted the case of stress shift in Campidanese Sardinian, not as a change in word level stress, but as an association of the verb+clitic structure with an HL* pitch accent. Using a word translation task, the authors elicited verb+clitic sequences from 4 speakers of the dialect, who also spoke standard Italian. They analyzed the vowel duration and the pitch accents. As concerns the former, they found lengthening of the penultimate syllable. The pitch patterns are described as being rather consistent among stimuli and participants, with an H tone associated with the lexically stressed vowel of the verb and an L tone associated with the penultimate syllable of the sequence. Lengthening and a low tone consistently associated with the rightmost prominent syllables are interpreted as an indication that the pronominal element is a weak pronoun rather than a true clitic, and thus functions as a prosodic word. This prosodic word and the verb form a phonological phrase, and a bitonal pitch accent is associated with this phrase.

1.3. Stress shift in verb+clitic sequences and the formation of the future and conditional tenses The studies that we have just reviewed have revealed that Romance varieties differ in the relative prominence patterns associated with verbs and pronouns in verb+clitic sequences. Acoustic studies have shown that when the stress is perceived to be on the clitic, duration is affected (Colantoni & Cuervo 2013; Kim & Repetti 2013; Nadeu et al. 2017) and pitch patterns may also differ from the patterns found in cases where the clitic is unstressed (Kim & Repetti 2013; Nadeu et al. 2017). This body of experimental evidence is consistent with early reports of doubly accented clitics in 16th century Spanish poetry and in literary and popular Argentine texts, where the orthographic accent is written on the clitic. Stress shifts in verbal phrases are not new in Romance varieties. In the formation of future and conditional tenses, the main stress shifted from the root to the auxiliary in a process that was variable and long and had as its end point a paradigmatic stress on the last syllable. This process closely resembled ESS, and we are not the first ones to draw such parallel. In fact, in an article on the formation of future and conditional tenses in Romance, Nocentini (2001) argues 112 CatJL 18, 2019 Laura Colantoni; José-Ignacio Hualde; Ane Icardo Isasa that the auxiliary, which had been shortened, did not have the same level of prominence as an independent word but was not deprived of an accent either. Nocentini explicitly compares the behavior of the auxiliary to that of a clitic (2001: 382). The auxiliary would, then, carry a secondary stress depending on the length of the verb+auxiliary compound. For example, it probably carried secondary stress in the first-person plural (e.g. cantáre habémus > cantárèmus, our example). In order to explain why we have a paradigmatic accent in the future tense, Nocentini invokes, among other factors, the power to attract stress that clitics have. He suggests that in sequences of root+(pronoun)+auxiliary, the increased distance between the primary and secondary stress, triggered a rhythmic pattern frequent in Latin (iambic inversion) and resulted in switching the primary stress to the auxiliary, with rhythmic retraction of the secondary stress on the infinitival root (2001: 385-386): cantáre hábeo > *cantár-àyo > *càntaráyo (> It canteró, Sp cantaré), cantáre íllud hábeo > *cantár-lu-àyo > *cantàr-lu-áyo (OSp cantàr- lo-é). Nocentini points out that the same tendency that operated in shifting the stress from the root to the auxiliary in the formation of the future is still active in many Italian varieties, which, as we have seen, show patterns of stress shift in verb+clitic sequences (see Ordoñez & Repetti 2006). Crucially, this process was not instantaneous. Although the first evidence that infinitive+HABEO sequences were used to express future tense is found already in the 3rd century, it took over 700 years for the synthetic form to fully replace the analytic one (Nocentini 2001). Moreover, Menéndez Pidal (1918: section 123) claims that, in Spanish, the future could be treated as a compound until well into the 17th century, as evidenced by the fact that pronouns were allowed between the verb and the auxiliary (e.g. dar le has ‘you will give him/her’; venir vos edes ‘you will come’– Menéndez Pidal 1918: 326). If Nocentini is right regarding the origin of stress shift in the development of Romance future and conditional tenses, we should expect that verb+clitic forms that at some point bore secondary stress on the clitic, as claimed for example by Kim & Repetti (2013), could eventually develop into one prosodic word with the main stress on the clitic (e.g. dámelò > dameló ‘give it to me’). Different varieties of a given language are expected to possibly be at different stages. Regarding the two varieties of Spanish that we are concerned with in this paper, Argentine and Peninsular Spanish, there could be independent factors that may favor the pattern with main stress on the clitic in Argentine Spanish. One such factor may be the use of different second person pronouns and corresponding verb forms, e.g. Arg. (vos) cantás vs Pen. (tú) cántas ‘you-sg sing’ (see appendix I). The fact that vos imperatives have final stress (cantá ‘sing!’, decí ‘say!’, explicá ‘explain’, introducí ‘introduce!’) may lead speakers of Argentine Spanish to interpret doubly-stressed imperative forms as having final stress, e.g.: decímelò > decimeló ‘say it to me!’. A further step may have been the spread of final stress to nonfinite verb forms with enclitics (cantármelo > cantármelò > cantarmeló, cantándomelo > cantándomelò > cantandomeló). Based on previous reports and experimental studies, we expect Argentine Spanish to be more advanced in this evolution towards final stress than Peninsular Stressed Clitic Pronouns in Two Spanish Varieties: A perception study CatJL 18, 2019 113

Spanish. Here we report on a perception study where native speakers from these two Spanish dialects were asked to judge stress prominence in stimuli from both dialects. One question that we want to address is whether previous reports regarding the relative frequency in the production of possible stress patterns in each of the two dialects is reflected in the perception of native speakers.

2. Perception experiment 2.1. Research questions Reports on ESS typically rely on the researcher’s perception of prominence. If our goal is to document the existence of the phenomenon and its distribution, this may indeed be enough, since in canonical instances the shift of prominence is perceptually salient and unambiguous. If, on the other hand, we aim at quantifying ESS based on corpus data in order to be able to determine, for instance, differences among dialects or speech styles, as in the unpublished work by Colantoni, Cuervo & Hualde (2010) mentioned in section 1, it is less obvious that one person’s perception may offer reliable data. For comparison, Spanish speakers agree on which syllable of any word carries lexical stress. In citation form of words, this syllable is acoustically enhanced. In natural speech, however, many words are produced with very little or no acoustic evidence showing which syllable is lexically stressed, so that, in the case of minimal pairs, listeners may be unable to identify the meaning that the speaker intended (Torreira, Simonet & Hualde 2014). That is, for instance, cánto ‘I sing’ and cantó ‘s-he sang’ are a minimal pair in Spanish, which native speakers clearly distinguish in citation form. However, within a phrase such as siempre que cánto/ cantó esta canción […] ‘Every time I sing/s-he sang this song […]’, the position of the stress, which serves to identify the word, may be conveyed very weakly or not at all, so that the discourse context may be needed in order to identify the word. Similarly, phrasal accent in English may be used for pragmatic purposes and phonologists agree on its expected position in different contexts (e.g. Ladd 2008). Judging phrasal prominence in actual discourse, however, becomes very subjective and unreliable. If consulted, naïve speakers do not show very high agreement in their judgments regarding which words in the phrase are prominent. With a sufficiently large number of participants, however, it is possible to determine which factors contribute to the perception of prominence for different groups of listeners, as shown in recent work by J. Cole and colleagues (Cole, Moe & Hasegawa-Johnson 2010; Cole, Mo & Baek 2010; Cole, Mahrt & Roy 2017; Roy, Cole & Mahrt 2017; for German, see also Baumann & Winter 2018). Here we use a similar methodology (asking a relatively large group of naïve, educated speakers) to study the perception of ESS in Spanish. Our research questions are, thus, to what extent naïve speakers agree on the position of the stress, whether there are differences among native speakers form Argentina and Spain in this perception, and what acoustic features listeners may use in their judgment of prominence on different syllables. 114 CatJL 18, 2019 Laura Colantoni; José-Ignacio Hualde; Ane Icardo Isasa

2.2. Methods 2.2.1. Participants A total of 36 native Spanish-speakers participated in this experiment: 15 speakers of Argentine Spanish recruited in Toronto and 21 speakers of Peninsular Spanish recruited in Irun, Gipuzkoa, Basque Country, Spain. Argentinian participants recruited in Toronto had spent no more than three years in Canada and used Spanish every day. Indeed, half of our Argentine participants lived in Argentina and were visiting relatives in Canada at the time of the experiment. Irun is a bilingual Spanish/Basque town, but where Spanish is by far the most common language of social interaction. Our Peninsular Spanish speakers all used Spanish daily and varied in their knowledge of Basque.6 Participants did not report having any hearing or speech impairments. The participants in this experiment may in general be characterized as educated speakers, but with no training in Linguistics and naïve as to the exact purpose and hypotheses of the experiment.

2.2.2. Stimuli We extracted all verb+clitic sequences that could be potential targets for ESS from 12 hours of Spanish-language TV soap operas (telenovelas): 6 hours from a series produced in Spain (“Cuéntame cómo pasó”) and 6 hours from two series made in Argentina (“Floricienta” and “Socias”).7 As mentioned in the introduction, potential targets of ESS are verb+enclitic sequences where there is at least one syllable between the lexically stressed syllable of the verb and the last syllable of the sequence (e.g., cantármelo, but not cantárlo). After eliminating tokens with background music and other types of background noise, we obtained 100 tokens to be used as stimuli for our perception experiment, 65 from the TV show produced in Spain and 35 from the two Argentine TV shows.8 Figures 1-4 show examples of four types of pitch contours observable in our stimuli. They are all from the series produced in Spain. Figure 1 shows a double-peak contour, with upstepping of the second peak, interpretable as double stress, with primary stress on the clitic (given the upstep): mìranós ‘look at us’. Figure 2 illustrates an example of main pitch prominence on the clitic: calmaté ‘calm down’. Figure 3 shows a pitch rise on the stressed syllable of the verb continuing as high plateau: sácala ‘take it out’. This contour seems more difficult to interpret in terms of prominence. There is prominence on the verb, but the clitic may or may not be perceived as prominent as well. Finally, Figure 4 shows a clear case with a single prominent pitch rise and fall on the verb: vámonos.

6. Given the great morphological differences between Basque and Spanish, we do not expect any transfer effects due to knowledge of Basque. Basque does not have clitic pronouns. Some Basque dialects have lexical pitch-accent, but that is not the case in the variety of Irun. 7. The actors, as far as we know, are all native speakers of Spanish from each of the two countries. The actors from which the Argentine tokens were obtained are all native speakers of Buenos Aires Spanish. 8. Our goal was to obtain an equal number of stimuli per variety. To do that, we selected the same number of hours of TV shows from each variety. However, in the Argentine shows, we obtained a smaller number of tokens and we had to discard many tokens, given the presence of music and overlapping conversations. Stressed Clitic Pronouns in Two Spanish Varieties: A perception study CatJL 18, 2019 115

Figure 1 (left). Illustration of double tonal prominence in mìranós ‘look at us’. Figure 2 (right). Illustration of tonal prominence on the clitic in yá calmaté ‘calm down already’.

Figure 3 (left). Illustration of an instance of pitch rise on verb followed by a sustained high tone in sácala ‘take it out’. Figure 4 (right). Illustration of a pitch rise and fall on the verb in vámonos ‘let’s go’.

Since besides pitch, other acoustic features of the speech signal, such as duration and intensity, may affect the perception of prominence on a given syllable, we acoustically analyzed our stimuli in Praat (Boersma & Weenick 2018) using ProsodyPro (Xu 2013). The three target vowels were segmented using TextGrids (see Figure 5) and labeled as follows: v1: lexically stressed vowel of the verb; v2: penultimate vowel of verb+enclitic unit; v3: final vowel of verb+enclitic unit. We extracted the mean duration, mean intensity and mean f0 (in semitones) of each of the three vowels. For duration and intensity, we then calculated the ratios v1/v3 and v2/v3. For example, a v1 to v3 ratio of “1” in intensity indicates that the lexically stressed vowel of the verb and the vowel of the last clitic have the same intensity. For pitch, we subtracted the value of the third vowel to that of v1 or v2. Differences beyond 1.5 semitones are considered to be perceptible.9

9. There is evidence that differences of 3ST play a role in communication (t’Hart et al. 1990: 29). However, Rietveld and Gussenhoven (1985) report lower thresholds. They found that differences of 1.5 STs were perceptually salient. 116 CatJL 18, 2019 Laura Colantoni; José-Ignacio Hualde; Ane Icardo Isasa

Figure 5. Example of segmentation of target vowels. Sound file: Comprándole ‘buying if for her/him’ (extracted from the Spanish soap opera Cuéntame cómo pasó).

Figure 6. v1/v3 ratio (left) and v2/v3 ratio (right) in duration of all the stimuli in our sample.

Figures 6 and 7 (produced in R (R Core Team 2017) with the package ggplot2 (Wickham 2016)) illustrate the between-dialect ratios in duration and differences in pitch, respectively. In both cases, the ratio of the Argentine data is below 1, which suggests that the final vowel tends to be longer than both the lexically stressed vowel of the verb and Stressed Clitic Pronouns in Two Spanish Varieties: A perception study CatJL 18, 2019 117

Figure 7. v1-v3 difference (left) and v2-v3 difference (right) in pitch (semitones) of all the stimuli in our sample. the penultimate vowel. For the Peninsular stimuli, on the other hand, the mean for both ratios is close to 1, indicating relatively small differences in duration among the three vowels overall. To test whether the stimuli of the two dialects differed in their duration ratio, we ran two U-tests, which returned a significant difference for both ratios (v1/v3: W = 1530.5, p-value = 0.004; v2/v3: W = 1553, p-value = 0.002). As concerns pitch, the median positive values obtained in the Argentine sample (see Figure 7) suggest that the final vowel is lower in pitch than the other two preceding vowels. The difference in pitch between v1-v3 and v2-v3 are likely to be perceptible, since these are close to the 1.5 semitone threshold (1.3 semitones). In the Peninsular data, the natural declination tendency seems not to apply and differences in semitones between the vowels compared hovered below 0 (v1-v3=-0.2st; v2-v3=-0.05). Once again, results of U-tests comparing the two samples revealed significant between-dialect differences in the v1-v3 difference (W = 660, p-value = 0.001) as well as in the v2-v3 difference (W = 801, p-value = 0.02). Intensity differences between the two samples turned out to be non-significant for both ratios (v1/v3: W = 1012, p-value = 0.3; v2/v3: W = 963, p-value = 0.2). Given the difference in the two sets of stimuli regarding pitch and durational cues, we expect our participants to perceive an accent on the clitic more frequently in the Argentine tokens.

2.2.3. Experimental procedure The perception experiment was run in Psychopy (Pierce 2007) on a laptop computer. Stimuli were presented to the participants in a randomized order generated by the program (putting together all stimuli from both dialects). Participants wore headphones and did the experiment in the presence of the experimenter, in a quiet environment (not through the internet). They were asked to identify the position of the stress by pressing ‘1’ on the keyboard, if they perceived the stress as falling on the verb (e.g. vámonos); ‘2’, if they perceived two stresses (vámonós); or ‘3’, if they perceived stress at the end of the word (vamonós). Participants had a short training session and they could listen to the stimuli only once. A screen shot illustrating the procedure in shown in Figure 8. 118 CatJL 18, 2019 Laura Colantoni; José-Ignacio Hualde; Ane Icardo Isasa

Figure 8. Screen shot of example presented in Psychopy.

2.2.4. Data analysis Both the results of the acoustic analysis of the signal performed in ProsodyPro (Xu 2013) and the participants’ responses from Psychopy (Pierce 2007) were exported to Excel, and statistical analysis was performed in R (R Core Team 2017) within the RStudio (RStudio Team 2016) environment. Binomial mixed-effects logistic regressions were fitted to the participants’ responses with the package lme4 (Bates et al. 2015), and p-values were obtained with lmerTest (Kuznetsova et al. 2017). We explore whether both groups of participants, from Argentina (ARG) and Spain (PEN), reacted in a similar way to the stimuli, and what acoustic factors determined their perception of stress on the clitic.

2.3. Results As mentioned in section 2.2.3, participants in the experiment could provide one of three responses: 1 = stress on verb, 2 = two stresses or 3 = stress on the clitic. In the first section of our results, §2.3.1, we focus on the perception of ESS. Thus, for the statistical analysis, we will group together the two answers that indicate ESS, that is, perception of prominence on the clitic (2 and 3) vs. prominence only on the verb (response 1). Then, in section 2.3.2, we examine the distribution of double stress, and, thus, we compare response 2 to the two responses that indicate that only one syllable was perceived as prominent (i.e., 1 and 3).

2.3.1. Perception of ESS In general, agreement for all 36 participants on the rating of 100 tokens was only ‘slight’ (Fleiss’ Kappa = 0.189, z= 47.3, p < 0.0001). If we consider only the 21 participants in the Peninsular group, agreement rises to ‘fair’ (Fleiss’ Kappa = 0.227, z= 32.9, p < 0.0001). Considering only the 15 Argentine participants there is no improvement in agreement. Stressed Clitic Pronouns in Two Spanish Varieties: A perception study CatJL 18, 2019 119

Table 1. Perception of prominence pattern by participant group Participant group 1= stress on verb 2 = double stress 3 = stress on clitic ARG 44.4% 29.3% 26.3% PEN 33.1% 35.2% 31.8%

Overall, the Peninsular group was more likely to perceive both stress on the clitic and double stress than the group from Argentina. As shown in Table 1, whereas ARG participants perceived a single stress on the verb 44.4% of the time, for the PEN group only 33.1% of the responses fall into this category. When we group together answers ‘2’ (= two stresses perceived) and ‘3’ (= stressed perceived on the clitic), given that in both cases there is perceived prominence on the clitic (that is, ESS), and separate the responses by the dialect of the stimulus, we obtain the density plots in Figure 9.10 Here, the left panel shows the distribution of responses of our 15 participants from Argentina and the right panel that of our 21 participants from Spain. As shown in the left panel of Figure 9, for the Argentine participants, responses to ‘arg’ (= Argentina) stimuli were grouped towards the right of ‘pen’ (= Spain) stimuli, indicating that ‘arg’ stimuli were perceived as either stressed on the clitic or having double stress more often. In these plots, the x axis indicates the number of participants who agreed on perceiving ESS. For instance, for the ‘arg’ stimuli, there were 9 tokens for which 14 participants perceived ESS and 5 tokens for which all 15 Argentine participants agreed on perceiving ESS. This is to be compared with only 2 ‘pen’ tokens for which 14 of the participants agreed on ESS and 1 ‘pen’ token where they all agreed. The gray line representing ‘pen’ stimuli reaches its peak at 6 on the x-axis because, for 11 ‘pen’ tokens, 6 Argentine participants gave a response consistent with ESS and this was the highest agreement that was reached for ‘pen’ stimuli (6/15). The participants tested in Spain (Figure 9, right) show a somewhat different distribution, with more perception of prominence on the clitic for both sets of stimuli, but much more so for ‘arg’ stimuli. That is, both groups of participants perceived more instances of ESS in the stimuli extracted from Argentine soap operas and the participants from Spain provided more frequent ‘2’ and ‘3’responses. To determine statistical significance, we ran a mixed-effect binomial regression model with perceived prominence (‘2’ and ‘3’ combined vs. ‘1’) as the dependent variable and dialect of the stimulus (Argentine, Peninsular) and participants’ dialect (Argentine, Peninsular) as fixed factors. Item and participant were included as random effects. Both fixed factors turned out to be significant (stimulus dialect, p <0.001, participants’ dialect, p < 0.01). As shown in Figure 10,11 the regression indicates that both groups of participants tended to perceive stress on the clitic more often in Argentine than in Peninsular stimuli. In addition, participants from Spain tended to perceive stress on the clitic more often in both kinds of stimuli.

10. Density plots were made with the package ggplot2 (Wickham 2016). 11. This figure was created with the package visireg (Breheni & Burchett 2017). 120 CatJL 18, 2019 Laura Colantoni; José-Ignacio Hualde; Ane Icardo Isasa

Figure 9. Density plots of responses ‘2’ (two stresses) and ‘3’ (stressed on the clitic) combined, separately for ARG (left panel) and PEN (right panel) participants. The black line shows the distribution of responses to Argentinian stimuli and the gray line shows responses to Peninsular stimuli.

Figure 10. Effect of participants’ dialect and of stimuli dialect on the perception of prominence on the clitic (double prominence included). Left panel = responses of participants from Argentina (ARG); right panel = responses of participants from Spain (PEN).

Argentinians perceived stress on the clitic in stimuli from Spain with very low frequency. Given this substantial difference in responses between the two groups of participants, in order to examine the effects of acoustic features on the perception of stress on the clitic, we decided to consider the responses of each group of participants separately. For each of the two groups, then, we fitted a binomial mixed-effects regression model with the structure indicated in Table 2. Stressed Clitic Pronouns in Two Spanish Varieties: A perception study CatJL 18, 2019 121

Table 2. Structure of mixed-effects binomial logistic regression with acoustic predictors Dependent variable Binary Response (stress on the verb vs. stress on clitic/two stresses) Fixed factors Duration ratio between V1 and V3 Intensity ratio between V1 and V3 Pitch difference between V1 and V3 Random factors Word Participant

Table 3. Output of mixed effect models with acoustic predictors only for ARG participants Estimate z-value p-value Duration -0.01 -3.23 = 0.001 Intensity -0.15 -3.11 = 0.002 Pitch 0.003 0.81 = 0.4 n.s.

Table 4. Output of mixed effect models with acoustic predictors only for PEN participants Estimate z-value p-value

Duration 0.18 0.53 = 0.6 n.s. Intensity -4.93 -1.63 =0.1 n.s. Pitch 0.008 2.35 =0.018

A summary of the output of the regression regarding the effect of the fixed factors for the Argentine participants (ARG) is shown in Table 3. For this group of participants, there is a significant effect of duration (p <0.001) and intensity (<0.01), but not of f0, on whether stress is perceived on the clitic (whether an additional stress is perceived on the verb or not). For the participants tested in Spain (PEN), the regression returns the output summarized in Table 4. Interestingly, for this group of participants neither duration nor intensity (as measured by v1/v3 ratios) appears to have a significant effect on their response. On the other hand, pitch has a marginally significant effect.

2.3.2. Perception of double stress In this final section of the results we explore the perception of double stress. The questions that we ask are the following: Is double stress perceived more often in the Peninsular than in the Argentine data? Do speakers of both geographical varieties show similar patterns of perception? The data in Table 1 above showed that ARG speakers responded ‘double stress’ only slightly less often than PEN participants (29% vs 35%, respectively). In Table 5, we show the frequency with which the Argentine and Peninsular stimuli elicited each of the three possible responses, for all 36 participants together: The stimuli taken from Argentine TV shows elicited many more ‘double-stress’ responses than those from Spanish shows (40% vs 28.8%). The difference in the responses to Argentine and Peninsular stimuli is consistent with 122 CatJL 18, 2019 Laura Colantoni; José-Ignacio Hualde; Ane Icardo Isasa

Table 5. Perception of prominence pattern by dialect of stimulus (all 36 participants) Stimulus 1= stress on verb 2 = double stress 3 = stress on clitic Argentine 328 = 26% 503= 40% 429 = 34% Peninsular 1040= 44.4% 674= 28.8% 627= 26.8%

some of the differences in the acoustic properties of the stimuli reported above in section 2.2.2. In order to explore whether the dialect of the participant or the dialect of the stimuli had a significant effect on the type of response, we created a second binary variable putting together answers ‘1’ and ‘3’, which indicate that a single stress was perceived either on the verb or on the clitic, vs. answer ‘2’ (double stress perceived). This binary dependent variable was modeled using a binomial mixed effects regression with the same structure as the one used in the previous section to test the difference between a stressed perceived on the clitic vs. double stress. The regression returned a significant effect of the dialect of the stimulus (β = 0.54, z = 2.48, p = 0.012), but no significant effect of the dialect of the participants. Both groups of participants perceived double stress significantly more often in the stimuli extracted from the Argentinian soap operas (see Figure 11). Given the non-significant effect of participants’ dialect on the perception of double stress, we did not run a mixed effects model with the acoustic variables that characterize the stimuli, as we did in the previous section.

Figure 11. Effect of participants’ dialect and of stimuli dialect on the perception of double prominence vs. single prominence (either on the verb or on the clitic). Left panel = participants from Argentina; right panel = participants from Spain. Stressed Clitic Pronouns in Two Spanish Varieties: A perception study CatJL 18, 2019 123

3. Discussion 3.1. Summary of results and answers to our research questions The results of our perception experiment revealed a more complex pattern than the one suggested by previous experimental studies. As concerns our question regarding whether participants’ dialect affected the perception of ESS in verb+clitic sequence, the answer is affirmative: we obtained different response patterns for Peninsular and for Argentine Spanish speakers. Whereas the former tended to have their responses almost equally distributed across the three categories (stress on the verb, on the clitic or both), the Argentine participants’ most frequent response was that the main prominence was on the verb. This difference in perception between participants depending on their native dialect is possibly the most solid result of this perception study. Our results tentatively suggest that each group of participants used different cues in order to judge the prominence pattern. As shown in Table 3, duration and intensity turned out to be significant cues for our Argentine participants, whereas only pitch was significant for the Peninsular Spanish participants. This is an interesting result that was not predicted and would require further testing. Notice that, in principle, the finding that our Peninsular participants provided more frequent ESS responses than our Argentine participants would appear to go against our predictions. However, in order to interpret these results, it is important to analyze the interaction between participant’s dialect and the dialect of the stimuli. One interpretation could be that, because ESS is more advanced in Argentina, speakers from this country have a higher acoustic threshold for its perception. Both possible ESS patterns, namely shift of the stress to the last syllable and double stress, were perceived more often in the examples from Argentinian TV shows than in those from Spain. This could be due to an accident of the specific stimuli that we happened to select for this study but is consistent with prior research indicating a higher frequency of ESS in Argentine than in Peninsular Spanish. Finally, our participants frequently indicated perception of double stress (in 40% of the stimuli from Argentinian soap operas and in almost 30% of those from Spanish TV shows), but there were also many cases of stress perceived only on the syllable containing the (last) clitic pronoun, especially in responses to stimuli from Argentinian soap operas (34% of all responses to stimuli in this group). These results may be interpreted as showing a stage in a diachronic path towards shifting of the (production and perception of the) stress from the verb root to the enclitic, as happened in the evolution of the future tense.

3.2. Potential task effects The methodology that we used, which has been previously applied to the perception of phrasal accent (see section 2.1), seems promising. This methodology is inherently inductive and seeks to determine whether participants can show some agreement in the way they categorize stimuli and whether the stimuli that belong to a given category share some acoustic properties. In this way, we tried to avoid some of the shortcomings of previous acoustic studies (e.g. Colantoni & Cuervo 124 CatJL 18, 2019 Laura Colantoni; José-Ignacio Hualde; Ane Icardo Isasa

2013; Klassen & Patience 2016), which relied on the categorization of the stimuli based on the researchers’ perception. Indeed, we observed in our data that individual participants had different patterns of response, but tendencies emerged at the group level. It should be noted that this was a difficult task, since participants could listen to each sound file only once, they had to rate 100 files and the sound files contained words in isolation. It is important to highlight that the stimuli differed to a great extent across many variables, such as the gender and age of the speaker, the speech rate and the loudness. Moreover, we used the examples that we obtained from 12 hours of recordings, and presented them out of context. Given the nature of the stimuli and the findings by Torreira et al. (2014) regarding the perception of lexical stress in running speech, we should expect some degree of uncertainty in participants’ responses, a degree of uncertainty that we would not expect if the stimuli had been produced in isolation with the aforementioned variables being controlled for. Our study did not include a control group, i.e. participants and stimuli from a dialect where ESS had not been reported (e.g. Mexican Spanish), as it was the case in Nadeu et al. (2017). We believe, however, that testing the perception of speakers from Argentina and Spain constitutes a valuable contribution. In future research, the responses obtained with this methodology could be also fruitfully compared with the shared judgment of experts listening to verb forms in context.

3.3. ESS and iambic inversion In our experiment the stimuli obtained from Argentine soap operas were judged more frequently to display ESS, that is, to either have the main prominence on the clitic or to have double prominence, than stimuli from Spain. This is consistent with previous reports that ESS is more frequent in Argentine than in Peninsular Spanish and with the results of the acoustic analysis of the stimuli (although, again, it also could be an accident in the selection of tokens). On the other hand, based on our own observations and the scarce reports in the literature (e.g. Menéndez Pidal 1918), we expected to find a preference for a perception of double stress (as opposed to complete shift of the stress to the clitic) in the Peninsular stimuli and, instead, a preference for final stress in the stimuli from Argentina. Although the Argentine stimuli received more ‘stress on clitic’ responses that the Peninsular stimuli, they also received even more ‘double stress’ responses. From an analytical point of view, it may be the case that, in stimuli where double stress tended to be perceived, the main stress had shifted to the clitic but a perceivable secondary stress remained on the verb stem; or it may also be the case that the primary stress remained on the stem, but the clitic had a secondary stress that was also perceived. A factor that needs to be taken into account is that we tested educated speakers who know where the stress is supposed to be, but still might have heard a prominence on the clitic, which might have led them to report two stresses rather than just one stress on the clitic. Stressed Clitic Pronouns in Two Spanish Varieties: A perception study CatJL 18, 2019 125

Independently of the specific explanation, which should be explored in future studies, our results revealed a difference in the perception of the stimuli coming from each dialect and suggest that ESS in Argentine Spanish is in the mixed penultimate stress shift category in Ordoñez & Repetti’s (2006) typology. Again, the differences in responses to the two groups of stimuli, although consistent with our expectations, may be due to the characteristics of the specific set of stimuli that we used. That is, it could be that by chance we happened to choose a greater proportion of ARG stimuli with ESS. For this reason, further experimentation is needed. The variability observed in the Argentine participants’ responses also suggests, first of all, that final stress shift is not systematic and, secondly, that speakers of Argentine Spanish have a higher threshold for perceiving ESS than speakers from Spain. They provided responses interpretable as ESS less frequently, especially to Peninsular tokens. Peninsular Spanish seems to be at a different point in the final stress shift typology, as evidenced mainly by the fact that the responses obtained from Peninsular speakers were equally distributed across the three possible categories included in the study. This suggests that at least there is some vacillation regarding the placement of the main prominence in these sequences. In section 1.3, we suggested that the differences between varieties could be linked to differences in the second person verb paradigm observed in these two dialects. Although this is a potential factor that seems to be worth considering, both the acoustic analysis of our stimuli and the acoustic cues that participants from each dialect seem to be using as a cue to main prominence suggest an additional explanation: there may be differences in the cues that convey stress in the two dialects (with a greater weight of duration and intensity in Argentine Spanish and of pitch in Peninsular Spanish). This clearly requires further investigation. What are then the parallels between ESS and the iambic inversion observed in the formation of Romance future and conditional tenses? In our stimuli, we can see that there is a tendency to perceiving some type of prominence on the clitic (either primary or secondary), as seems to have been the case in sequences of infinitive + HABEO in Late Latin. We have also observed that the tendency to perceive the prominence on the clitic is not categorical and shows cross-dialectal variation. What we do not know is whether this variability is a stable feature of the language, as it may be concluded from Menéndez Pidal’s (1918) report on the presence of doubly stressed verb+clitic sequences in 16th century poetry, or if it is a case of a change in progress.

References Bates, Douglas, Maechler, Martin, Bolker, Ben & Walker, Steve. 2015. Fitting linear mixed-effects models using lme4. Journal of Statistical Software 67(1): 1-48. Baumann, Stephan & Winter, Bodo. 2018. What makes a word prominent? Predicting untrained German listeners perceptual judgments. Journal of Phonetics 70: 20-38. Boersma, Paul & Weenink, David. 2018. Praat: doing phonetics by computer, version 6.035, [computer program]. 126 CatJL 18, 2019 Laura Colantoni; José-Ignacio Hualde; Ane Icardo Isasa

Breheny, Patrick & Burchett, Woodrow. 2017. Visualization of Regression Models Using visreg. The R Journal 9: 56-71. Colantoni, Laura & Cuervo, María Cristina. 2013. Clíticos acentuados. In Laura Colantoni & Celeste Rodríguez Louro. Perspectivas teóricas y experimentales sobre el español de la Argentina, 143-157. Madrid: Iberoamericana. Colantoni, Laura, Cuervo, María Cristina & Hualde, José Ignacio. 2010. Stress as a symptom. Paper presented at the Linguistic Symposium on Romance Languages 40, University Washington, Seattle, March 26-28. Cole, Jennifer, Mo, Yoonsook & Hasegawa-Johnson, Mark. 2010. Signal-based and expectation-based factors in the perception of prosodic prominence. Laboratory Phonology 1: 425-452. Cole, Jennifer, Mo, Yoonsook & Baek, Soondo. 2010. The role of syntactic structure in guiding prosody perception with ordinary listeners and everyday speech. Language and Cognitive Processes 25(7): 1141-1177. Cole, Jennifer, Mahrt, Timothy & Roy, Joseph. 2017. Crowd-sourcing prosodic annota- tion. Computer Speech & Language 45: 300-325. Fernández Soriano, Olga. 1999. El pronombre personal. Formas y distribuciones. In Ignacio Bosque & Violeta Demonte (eds). Gramática descriptiva de la lengua española, 1209-1273. Madrid: Espasa. Fontanella de Weinberg, María Beatriz. 2000. El español bonaerense. En María Beatriz Fontanella de Weinberg & Nélida Donni de Mirande (eds.). El español de la Argentina y sus variedades regionales, 37-61. Buenos Aires: Edicial. Gili Gaya, Samuel. 1943. Curso superior de sintaxis española. Barcelona: Bibliograf. Kuznetsova, Alexandra, Brockhoff, Per & Christensen, Rune. 2017. lmerTest Package: Tests in Linear Mixed Effects Models. Journal of Statistical Software 82(13): 1-26. Hualde, José Ignacio. 2005. The sounds of Spanish. Cambridge: CUP. Huidobro, Susana. 2005. Phonological constraints on verum focus in Argentinian Spanish. Manuscrito, SUNY. Kim, Miran & Repetti, Lori. 2013. Bitonal pitch accents and phonological alignment in Sardinian. Probus 25, 267-300. Klassen, Gabrielle & Patience, Matthew. 2016. Stressed clitics in Argentine Spanish: Which way does the clitic lean? In Alejandro Cuza, Lori Czerwionka & Daniel Olson (eds.). Inquiries in Hispanic Linguistics: From theory to empirical evidence, 149-169. Amsterdam: John Benjamins. Martínez, Tomás E. 1991. La novela de Perón. Buenos Aires. La Nación. Moyna, María Irene. 1999. Pronominal clitic stress in Río de la Plata Spanish: an optimality account. The SECOL Review 23, 15-44. Nadeu, Mariana, Simonet, Miquel & Llompart, Miquel. 2017. Stress postverbal pronominals in Catalan. Probus 29(1): 119-162 Nocentini, Alberto. 2001. La genesi del futuro e del condizionale sintetico romanzo. Zeitschrift für romanische Philologie 117(3): 367-401. Peperkamp, Shannon. 1997. Prosodic words. The Hague: HAG. Ordóñez, Francisco & Repetti, Lori. 2005. Stressed clitics? In Jean-Pierre Montreuil (ed.). New perspectives on Romance linguistics. Volume II: Phonetics, 167-181. Amsterdam: John Benjamins. Stressed Clitic Pronouns in Two Spanish Varieties: A perception study CatJL 18, 2019 127

Peirce, Jonathan. 2007. PsychoPy: Psychophysics software in Python. Journal of Neuroscience Methods 162 (1-2): 8-13. R Core Team. 2017. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Rietveld, Toni & Gussenhoven, Carlos. 1985. On the relation between pitch excursion size and prominence. Journal of Phonetics 13: 299-308. Roy, Joseph, Cole, Jennifer & Mahrt, Timothy. 2017. Individual differences and patterns of convergence in prosody perception. Laboratory Phonology 8(1): 22. RStudio Team. 2016. RStudio: Integrated Development for R. RStudio, Inc., Boston, MA URL . t’Hart, Johan, Collier, René & Cohen, Antonie. 1990. A perceptual study of intonation: An experimental phonetic approach to speech melody. Cambridge: Cambridge University Press. Torreira, Francisco, Simonet, Miquel & Hualde, José Ignacio. 2014. Quasi- neutralization of stress contrasts in Spanish. Campbell, Gibbon & Hirst (eds.). Proceedings of Speech Prosody 7, 197-201. Dublin. Torres Tamarit, Francisco & Pons-Moll, Claudia. 2018. Enclitic-induced stress shift in Catalan. Journal of Linguistics 55(2): 407-444. Wickham, Hadley. 2016. ggplot2: Elegant Graphics for Data Analysis. New York: Springer-Verlag. Xu, Yi. 2013. ProsodyPro: A tool for large-scale systematic prosodic analysis. TRASP 2013, Aix-en-Provence. .

Appendix 1 Second person Imperative forms in Argentine and Peninsular Spanish, illustrated with the verb cantar ‘to sing’. Notice that the 2nd person singular form vos, used in Argentina, and the 2nd person plural form vosotros/as, used in Spain, are etymologically related. In the Imperative, vos and vosotros/as forms have final stress. They only differ in that the vosotros/as forms have a final /d/.

Argentina Spain 2nd person singular familiar cantá (vos) cánta (tú) 2nd person singular formal cánte (usted) cánte (usted) 2nd person plural cánten (ustedes) cantád (vosotros/as) cánten (ustedes) formal 128 CatJL 18, 2019 Laura Colantoni; José-Ignacio Hualde; Ane Icardo Isasa

Appendix 2. List of stimuli Stimulus Dialect abonenme Peninsular acompañandole Peninsular acuerdate Peninsular adelantamelos Peninsular avisame Peninsular ayudandolo Argentine callate Peninsular calmate Peninsular comete Peninsular comportate Peninsular comprandole Argentine comprandole Argentine convirtiendonos Argentine creame Peninsular cuidandote Argentine cuidate Peninsular damelo Argentine decirmelo Argentine deciselo (2) Argentine dejale Peninsular dejame Peninsular dejame Peninsular dejamela Argentine dejate Peninsular dejeme Argentine digales Peninsular digame (6) Peninsular dimelo Peninsular disculpeme (2) Argentine diselo Peninsular disfrazandose Argentine embarazandome Argentine encierrame Peninsular ensename Peninsular escucheme (2) Argentine esperame Peninsular esperandole Peninsular explicamelo Peninsular fiate Peninsular imaginate (2) Peninsular imaginese Peninsular incitandonos Peninsular llamemosla Argentine Stressed Clitic Pronouns in Two Spanish Varieties: A perception study CatJL 18, 2019 129

Stimulus Dialect llevate (2) Peninsular llevele Argentine miralo Peninsular miralos Peninsular miranos Peninsular mireme Argentine olvidate Peninsular pensemoslo Argentine perdoneme Argentine permitame Peninsular pidamelo Argentine piensalo Peninsular ponetelo (2) Argentine pongase Argentine portate Peninsular preparate Peninsular prometale Argentine quedate (2) Peninsular quedese Peninsular quedese Argentine recibiendole Peninsular sacala Peninsular secate Peninsular sientate (5) Peninsular sientese Argentine sintiendome Peninsular subame Peninsular tranoslo Argentine tranquilicese Peninsular tranquilicese (2) Argentine transmitales Peninsular vamonos (4) Peninsular vayase Argentine vendiendoselo Argentine viendolo (2) Peninsular vistase Argentine

Catalan Journal of Linguistics 18, 2019 131-159

The analysis of truncated vocatives in Taviano (Salentino) Italian*

Michael Kenstowicz Massachusetts Institute of Technology [email protected]

Received: December 30, 2018 Accepted: June 5, 2019

Abstract

This paper documents and discusses various descriptive generalizations and alternative analyses of the vocative truncation found in the Southern Italian dialect of Taviano that is illustrated by such formations as Filoména > Filomé. We show that the upper limit on the size of the truncate is not restricted by foot binarity nor by internally layered feet: Addoloráta > Addolorá. The lower limit of the truncate may extend to a single CV syllable in violation of foot binarity: Césare > Cé. But truncation to just the stressed vowel is rejected: Élena > *É. We situate the analysis of the vocative truncation within the overall grammar of Salentino and show that while some properties such as the loss of coda consonants can be attributed to rankings imposed by the surrounding grammar, others are peculiar to this construction. The analysis of stress in loanword adaptation figures prominently in our discussion. The final sections examine some of the morphological and syntactic properties of the Taviano vocative. Keywords: vocative; truncation; loanword adaptation; stress

Resum. L’anàlisi dels vocatius truncats en tavià (salentí)

Aquest article documenta i discuteix diverses generalitzacions descriptives i anàlisis alternatives del truncament vocatiu en tavià, un dialecte del sud d’Itàlia, que es pot il·lustrar amb exemples com Filoména>Filomé. Es demostra que el límit superior en relació a la mida màxima del truncat no es restringeix al peu binari ni als peus mínimament recursius: Addoloráta > Addolorá. El límit inferior del truncat pot arribar a una sola síl·laba CV, amb violació del requisit de binarietat: Césare > Cé, tot i que es rebutja el truncament quan el resultat és una vocal tònica: Élena > *É. L’anàlisi del truncament vocatiu se situa en el context de la gramàtica general del salentí i demostra que, mentre algunes propietats com la pèrdua de consonants en posició de coda es poden atribuir als rànquings imposats per la gramàtica del salentí, altres són específiques d’aquesta construcció. L’anàlisi de l’accent en l’adaptació de manlleus constitueix un aspecte prominent de la proposta. La secció final examina algunes de les propietats morfològiques i sintàctiques del vocatiu del tavià. Paraules clau: vocatiu; truncament; adaptació de manlleus; accent; Salentino

* Thanks to Stanislao Zompi for sharing his language and for his many contributions to the analyses and discussions in this paper. Any remaining errors are the responsibility of the author. We also thank our two anonymous CJL reviewers for their comments. This paper is dedicated to Guglielmo Cinque in recognition of his vital role in nurturing the development of generative grammar in Italy as well as for his research in typology and emarginazione.

ISSN 1695-6885 (in press); 2014-9718 (online) https://doi.org/10.5565/rev/catjl.257 132 CatJL 18, 2019 Michael Kenstowicz

Table of Contents 1. Introduction 7. Prior-Analyses-3 2. Truncated Vocatives (D’Alessandro & van Oostendorp 2016) 3. Background 8. Summary of Phonological Characteristics of the Vocative 4. Vocatives and truncated vocatives 9. Morphological and Grammatical 5. Prior analyses-1 (Alber 2009) Properties of the Taviano Vocative 6. Prior-Analyses-2 10. Summary and conclusion (Vanrell & Cabre 2011; Cabre & Vanrell 2016; Cabre et al. 2019) References Appendix

Omaggio a G. Cinque

1. Introduction The canonical morphological operation is affixation to a base. Truncation processes seen in hypocoristic formations such as Michael > Mike are anomalous from this perspective—in two senses. First, there is no affix. Second, the output is shorter/ smaller than the base. Such truncation processes are typically divided into several subtypes. A major division is between subtractive morphology vs. prosodically targeted compressions. In the former, the change to the base is minimal and in the best-studied cases involves filling a slot in an inflectional paradigm, as in Kosati plurals (Kurisu 2001) and Icelandic deverbal nouns (Oreshnik 1972; Kiparsky 1984). Prosodically-motivated truncation comprises clippings, hypocoristics, and vocatives. The first two types usually involve mapping the base to a prosodic category such as the foot and minimal word (McCarthy & Prince 1986). Japanese formations like demonsutoréesyon ‘demonstration’ > [demo] and Masako > [masa] (Ito 1990) illustrate. The two types of nickname formation first analyzed in these terms for Spanish by Prieto (1992) and for Italian by Thornton (1996) illustrate prosodic truncations in Romance (1). We use ‘right vs. left edge’ to refer to the site of the truncation. Their inverses are often used to denote the point of alignment of the base word with the truncating template.

(1) Spanish right edge left edge base truncate base truncate Jose Jose Roberto Beto Teresa Tere Alfonso Poncho Gertrudis Gertu Ernesto Neto Francisco Francis Fernando Nando Margarita Marga Rosita Chita The analysis of truncated vocatives in Taviano (Salentino) Italian CatJL 18, 2019 133

Italian right edge left edge base truncate base truncate Francésca Fránce Salvatore Tore Alessándra Ále Filomɛ́na Mɛ́na Simóna Símo Antɔ́nio Tɔ́to Cecília Cɛ́ci Gugliɛ́lmo Mɛ́mo Robérto Rɔ́be Giusɛ́ppe Pɛ́ppe Subsequently, Feliu (2001) called attention to Spanish trisyllabic truncates such as analfabéto > [analfa] ‘illiterate’ and more recently Martinez-Paricio & Torres- Tamarit (2018) document hypocoristic formations like Estanisláo > [estanis], which they analyze with minimally recursive, internally layered trochaic feet: (es(tanis)).

2. Truncated Vocatives Recent literature has isolated a distinct type of truncation, which at least canonically is said to be associated with appellative (vocative) formations. This type has been detected in at least three distinct Mediterranean locales: Southern Italian, Sardinian, and Algherese Catalan. It has also been reported for certain Romanian dialects for kinship terms and personal names as well as certain morphologically defined imperative verb forms. Unlike hypocoristics and clippings, this formation does not have a fixed size but rather varies its length as a function of the location of stress and the number of pretonic syllables in the base. Examples from all four of these Romance varieties appear in (2). (2) a. Southern Italian (Alber & Arndt-Lappe 2012) base vocative Barbara Ba Carmela Carme Salvatore Salvato Antonella Antone b. Sardinian (Cabre & Vanrell 2016; Cabre et al. 2019) base vocative Bertulu Be Gosamu Go Simone Simo Fortunadu Fortuna Servatore Servato c. Algherese Catalan (Cabre & Vanrell 2011) base vocative Fabio Fa Francesco France Antoni Anto Nicoletto Nicole Elisabeta Elisabe 134 CatJL 18, 2019 Michael Kenstowicz

d. Maramureş Romanian (Maiden 2006) base vocative nevastă nevá nepoată nepoá tată ta Alexa Alé Părasca Părá

The descriptive generalization underlying the formation of such structures is that the truncate aligns with the left edge of the given-name base form and with its stressed vowel. Since the base is of potentially unbounded length, the truncate will vary its size stretching from one up to four or more syllables, the latter putting beyond the reach of an internally layered foot, as in Algherese Catalan Elisabéta > [elisabe]. Our goal in this paper is to document and analyze this truncation process in more detail based on data from the Italian dialect of Taviano (Salentino). But first we situate the vocative construction in the overall grammar of this dialect.

3. Background Taviano has many of the features common to Southern Italian dialects in general. It has five vowels in stressed position and lacks the open-close mid vowel distinction of Tuscan-based Standard Italian. The mid vowels /o/ and /e/ generally are barred from postonic position except that the latter may appear word-finally. In pretonic position vowel reduction occurs in the oldest layer of the vocabulary and is reflected in paradigmatic alternations found in the inflection of verbs. Here unstressed /o/ is realized as [u] while unstressed /e/ is realized as [a]—the latter in contradistinction to most other Salentino varieties, where it surfaces as [i]: [vene] 3sg, [vanimu] 1pl ‘come’ and [more] 3sg, [murimu] 1pl ‘die’. The dialect has eliminated the metaphony found in other Salentino locales (most notably Lecce, Calabrese 1984) and retains remnants in just a handful of items such as [fjerru] ‘iron’, [jentu] ‘wind’, [jernu] ‘winter’ (cf. St. Italian ferro, vento, (in)verno). Diphthongs of both rising [je] and falling [au̯ ] sonority occur as well as geminate consonants, with /b/, /dʒ/, and /ɲ/ always geminated intervocalically, and /dd/ often realized as a retroflex. The Taviano syllable template is CCVC with codas restricted to /s/, liquids, and nasals homorganic with the following onset consonant. In word-final position codas are largely absent. The latter restriction is actively enforced since loanwords ending in a consonant are regularly repaired by copy-vowel epenthesis.

(3) St. Italian Taviano film filmi ‘film’ bar barra ‘bar, café’ Úpim Úpimmi brand name Conad Konadda brand name The analysis of truncated vocatives in Taviano (Salentino) Italian CatJL 18, 2019 135

There is also a word-minimality requirement in Taviano so that many monosyllables of Standard Italian appear with a lexically determined augment composed of a coronal consonant plus /e/. Examples appear in (4).1

(4) St. Italian Taviano no none ‘no’ sì sine ‘yes’ va vaje ‘go’ 3sg chi tʃite, tʃine ‘who’ tre trete ‘three’

As far as clippings and hypocoristic formations are concerned, Taviano has the left-edge truncation structure targeting a disyllabic trochaic template that is found in many other varieties of Italian and Romance more generally (cf. 1). The nicknames display segmental and prosodic TETU effects in which a consonant is inserted or copied that are comparable to those observed in Standard Italian, Spanish (Boyd- Bowman 1955; Piñeros 2000), and English.

(5) St. Italian Taviano base nickname base nickname Alessandro Sandro Lisandru Sandru Salvatore Toto Salvatore Totu Luigi Gigi Luvitʃi, Luviddʒi Tʃitʃi, Pitʃi Antonietta Ntunjetta Étta, Tetta Domenico Meni, Menico Dumeniku Mimmu

But Taviano lacks the right-edge truncation noted above in (1) for Spanish and Standard Italian. On the other hand, there is another, less common pattern of data in which the stressed syllable is reduplicated with a CV prefix (6). As observed by an anonymous referee, such iambically stressed structures are common in French (e.g. Monique > Momo, Louis > Loulou). But in this case it is the initial syllable that is reduplicated, not the stressed syllable.

1. When cliticised, these elements retain their original monosyllabic form: (i ) Kwarketùnu me disse sta kosa, ma nu+ sta+ me rikòrdu tʃi(*te) someone 1sg.clit said.3sg. this thing but not prog 1sg.clit remember.1sg who fose was.3sg ‘Someone told me this thing, but I can’t recall who it was.’ 136 CatJL 18, 2019 Michael Kenstowicz

(6) base nickname Kosimu Koko Pantaleu Lele Rosariu Sasa Salvatore Toto Euddʒénju Dʒeddʒé

3.1. Loanwords: gemination, epenthesis, and stress As indicated in the preceding section, words in the Taviano dialect must terminate in a vowel. Loanwords from Standard Italian as well as other languages that end in a consonant are adapted by epenthesizing a vowel in order to bring the word into conformity with this phonotactic restriction. Unlike in Tuscan Italian where the epenthetic vowel is [e], in Taviano the epenthetic vowel is a copy of the preceding syllable—at least in the adaptation of final consonants. (See Repetti 2011 for general discussion of epenthesis in Italian loanword adaptation). In addition, Taviano displays faithfulness to the source word in at least two respects that in combination create structures that deviate rather remarkably from native words. First, the final consonant of the loan is geminated (7). This kind of gemination is found in the adaptation of loanwords into Japanese, Arabic, and Hungarian—languages that like Italian have a geminate-singleton contrast. Two principal hypotheses have been offered as motivation for such gemination. One sees it as a strategy to satisfy a bimoraic condition on stressed syllables—in the Italian case in the face of a ban on word-final long vowels (Chierchia 1986; Repetti 1993). An alternative views the gemination as a faithfulness effect that allows the consonant in the loan to maintain its “syllabic role” as a coda in the donor language while also providing an onset to the following vowel. See Shinohara (1997) and Ito & Mester (to appear) for analyses along these lines for Japanese and Bafile (2003) for Italian.

(7) St. Italian Taviano bar barra ‘bar, café’ hotel otelle ‘hotel’ tram tramma ‘streetcar’ clan klanna ‘clan’

The Taviano data in (8) provide support for the second interpretation. These loans maintain faithfulness to stress as well as preservation of the coda status of the geminated consonant. As a result, they create words with antepenultimate stress whose penultimate syllable is closed and hence heavy—a prosodic structure that otherwise is highly anomalous and which is regularly stressed on the penult in the native vocabulary: sorélle (*sórelle) ‘sisters’. The fact that the penultimate syllable is unstressed in the loan (and in its correspondent in the donor language) challenges the weight-to-stress interpretation of the loanword gemination. Bafile (2003) makes the same observation for loans in the Florentine dialect (see below). The analysis of truncated vocatives in Taviano (Salentino) Italian CatJL 18, 2019 137

(8) St. Italian Taviano tunnel tunnelle ‘tunnel’ tennis tennissi ‘tennis’ virus virussu ‘virus’ álbum álbummu ‘album’ Fiat Fiatta brand name Úpim Úpimmi brand name Comit Komitti ‘Comit’ Seat Seatta brand name ananas ?ananassa ‘pineapple’

As shown by the relatively acceptable adaptation ?ánanassa‘pineapple,’ faithfulness to stress can even give rise to words in which the ictus falls outside the three-syllable window that otherwise governs the locus of accent in Italian and Romance languages more generally. In order to test whether or not the singleton vs. geminate contrast is maintained in the geminated loanwords with antepenultimate stress such as Úpimmi, we recorded a small sample of Taviano words and measured the duration of their segments. (See Appendix for the corpus). The words were spoken in isolation by our speaker and then in a sentential frame: [tokka+ditʃi X kjanu kjanu] ‘it is necessary to say X slowly, slowly’. First, to establish a baseline we measured the ratio of the duration of the medial consonant spanning the last two syllables with respect to the duration of the preceding stressed vowel. As in Standard Italian, Taviano stressed vowels in penultimate open syllables are lengthened relative to their duration in closed syllables. For these cases such as Michéle vs. sorélle we found that in the former the average duration ratio of the consonant relative to the preceding stressed vowel was 0.66 ms (st dev = 0.26 ms) while for the geminate it was 1.67 ms (0.40). A paired two-sample T-test found the difference to be highly significant (p < 0.001). For the case of loans with the geminated consonant but antepenultimate stress such as Úpimmi vs. native singletons such as ultimi there was still a significant difference in the ratio of the medial consonant spanning the last two syllables vis a vis the penultimate vowel: geminates had a ratio of 1.86 ms (0.51 ms) vs. singletons’ 1.10 ms (0.31 ms). The ratios are different in this case because the penultimate vowel is not stressed and hence is not lengthened. Nevertheless, there is still a difference in the duration ratios of the medial consonant relative to the preceding unstressed vowel with the geminate ratio being significantly larger by a paired T-test (p < 0.001). From these (admittedly minimal) data we conclude that the final consonants in loan adaptations such as Úpim > Úpimmi really are geminated. We also compared these ratios across the two conditions with unpaired T-tests. The ratios of the penultimate vowels in the singleton context (Michéle vs. ultimi) were significantly different (p < 0.001) while they were not in the geminate context of Úpimmi vs. sorélle. It appears that the presence of a geminate cancels the vowel lengthening effect of penultimate stress. Our Praat script also returned the F0 values for the syllables in the words com- posing our corpus. We focus on words embedded in the frame sentence context 138 CatJL 18, 2019 Michael Kenstowicz since the isolation forms seemed to be distorted by a boundary tone. For words like Úpimmi and ultimi there was a slightly falling F0 contour of mean -2.7 Hz (st dev = 5.5) between the stressed antepenultimate syllable and the following unstressed penult. But for penultimate stressed words such as otélle and Michéle there was a sharp rise in pitch between the antepenult and the penult (mean = 23 Hz, st dev = 8.7 Hz). Given that H* pitch accents are aligned with stressed syllables in Taviano, these data support the contention that the stress falls earlier in the word in loans with a geminate consonant such as Úpimmi and show the surprising strength of faithfulness to stress in the loanword adaptation process.2 The loans with the copy vowel also stand out morphologically since they sys- tematically fall into the class of indeclinables. So even though bárra ‘bar, café’ terminates in a vowel that is homophonous with the singular (typically feminine) desinence, it fails to change in the plural unlike the nativized bárra ‘iron bar’, which does inflect: bárre (pl). Another point worthy of mention is that while copy-vowel epenthesis is readily applied to the vowels [i,u,e,a] as seen in the data in (8) and (9), there is more uncertainty for the vowel [o], as seen in (10). The reason presumably is a conflict between the requirement of identity between the two vowels in the copy relation on the one hand and the otherwise general absence of [o] at the end of masculine singular nouns on the other hand. In this morphological context round vowels are regularly realized as [u]. The latter phonotactic generalization overrides the otherwise strong segmental faithfulness found in Taviano loanwords so that Légo and Ninténdoare adapted as Légu (*Légo) and Ninténdu (*Nintendo).

(10) St. Italian Taviano álcol, alcole *alkollu, ?alkullu nɛ́on *neonnu, ?neunnu ípsilon, ipsilonne *ipsilonnu caos, caosse *kaossu, ?kaussu

3.2. OT analysis The distribution of stress in native Taviano words largely aligns with Standard Italian. There is one stress per phonological word. It is located in a three-syllable window at the right edge of the word. If the penultimate syllable is closed, stress regularly falls on the penult (barring well-known, isolated exceptions such as Táranto and mándorla ‘almond’). Stress location varies lexically between antepenultimate (Césare) and penultimate (Michéle) syllables, with a statistical bias in favor of the latter. Final stress is rare outside monosyllables. For purposes of discussion, we will assume a foot-free analysis that is embodied in the following constraints.

2. The sensitivity of stress in Italian to the morphological and lexical contexts in which it is located has been demonstrated by Burani et al. (2014) in a study of the accent assigned to nonce items in a reading aloud task. The analysis of truncated vocatives in Taviano (Salentino) Italian CatJL 18, 2019 139

(11) OT stress constraints Culminativity: penalize a word with more than one stress as well as a word with no stress *Extended-Lapse-Right (*ELR): penalize a sequence of three unstressed syllables at the right edge of the word Align-Stress-Left (Al-Str-L): assign a violation mark for each unstressed syllable that intervenes between the stressed syllable and the left edge of the word *Lapse-Right (*LR): penalize a word with two unstressed syllables at the right edge Non-Finality (Non-Fin): penalize a word with stress on the final syllable Weight-to-Stress (W-to-S): penalize an unstressed heavy syllable

The rankings among these constraints that are required to account for the generalizations governing the distribution of stress noted above can be inferred from the following violation tableau for a schematic four-syllable word that contains no heavy syllable.

(12) /ssss/ *ELR Al-St-L *Lapse-Rt Non-Fin śsss * * sśss * * ssśs ** sssś *** *

Undominated *Extended-Lapse-Right prevents stress from moving to the left past the antepenult. The choice between antepenultimate and penultimate stress is controlled by lexically determined rankings between Align-Stress-Left and *Lapse-Right: Align-Stress-Left » *Lapse-Right favors stress on the third-last syllable (e.g. Césare) while *Lapse-Right » Align-Stress-Left chooses penultimate stress (e.g. Michéle). Finally, for words with a closed penultimate syllable the systematic absence of antepenultimate stress follows from ranking Weight-to- Stress over Align-Stress-Left.

(13) Weight-to-Stress » Align-Stress-Left /Roberto/ Weight-to-Stress Align-Stress-Left > Ro.ber.to * Ro.ber.to *!

With these rankings in place we can turn to the analysis of the Taviano loanwords. The native-grammar constraint against word-final consonants is enforced by truncation in vocatives but by epenthesis in loanwords. A similar discrepancy is found in other languages such as Korean (Kenstowicz 2005) where illegal conso- 140 CatJL 18, 2019 Michael Kenstowicz nant clusters are repaired by deletion in the native vocabulary but by epenthesis in loans. For Taviano, this implies the demotion of different faithfulness constraints with respect to *C#, as seen in the tableaux below in (14) for vocative Loréntsu > Loré and loanword Úpim > Úpimmi.

(14) Final-Consonant (*C#): penalize a word-final consonant Max-C: penalize deletion of a consonant Dep-V: penalize insertion of a vowel

*C#, Dep-V » Max-C /Loréntsu/ *C# Dep-V Max-C > Loré * Lorén *! Lorén(n)e *!

*C#, Max-C » Dep-V /Úpim/ *C# Max-C Dep-V > Úpimmi * Úpim *! Úpi *!

As indicated above, a common strategy to remain faithful to the syllable structure of the source word but still satisfy the cross-linguistic preference for syllables with a consonantal onset is to geminate the coda consonant—at least in languages that have the singleton-geminate contrast in their phonemic repertoire. We postu- late the constraints and rankings in (15) to account for this aspect of the Taviano loanword adaptation.

(15) Ident-Final-Coda: penalize an output in which a word-final coda consonant in the source word is not matched by a coda consonant in the loanword adaptation Onset: penalize a syllable without a consonantal onset *Geminate: penalize a geminate consonant

ident-Final-Coda, Onset » *Geminate /Ú.pim/ Ident-Final-Coda Onset *Geminate > Ú.pim.mi * * Ú.pi.mi *! * Ú.pim.i **! The analysis of truncated vocatives in Taviano (Salentino) Italian CatJL 18, 2019 141

A final factor differentiates Taviano from Tuscan Italian with respect to the adaptation of word-final consonants. Taviano resorts to copy-vowel epenthesis while Tuscan inserts the default vowel [e].

(16) St. Italian Tuscan Italian Taviano lapis lapisse lapissi gas gasse gassa

The proposed OT analysis of this aspect of the loanword grammar is seen in (17). Following Stanton & Zukoff (2018) we assume that copy vowel epenthesis entails a syntagmatic correspondence relation between the inserted vowel and its mate.

(17) *ViCoVi : penalize a correspondence relation between the vowels of successive syllables Dep-F’s: penalize the insertion of a feature root node

*C#, *ViCoVi » Dep-F’s (Tuscan) *C#, Dep-F’s » *ViCoVi (Taviano)

ep s ep s /gas/ *C# *ViCoVi D -F’ /gas/ *C# D -F’ *ViCoVi gassa *! > gassa * > gasse * gasse *! gas *! gas *!

With these rankings in place we return to the anomalous stress of the Taviano loanwords. To account for the antepenultimate stress of Úpimmi and the absence of penultimate stress, the output-output faithfulness constraint for stress must outrank Weight-to-Stress. In combination with undominated Culminativity that bars more than a single stress per word, the stress of the loan is borrowed along with its segmental structure and syllable shape.

(18) Ident-Stress-OO: penalize a loanword that fails to preserve the stressed syllable of the source word

culminativity, Ident-Stress-OO » Weight-to-Stress

/Ú.pim/ Culminativity Ident-Stress-OO Weight-to-Stress > Ú.pim.mi * U.pim.mi *! Ú.pim.mi *!

As shown by the adaptation of ánanas ‘pineapple’ as ánanassa, stress may fall outside the three-syllable window at the right edge of the word in the Taviano loanword adaptation. This will follow if Ident-Stress-OO dominates *Extended- Lapse-Rt: 142 CatJL 18, 2019 Michael Kenstowicz

(19) Ident-Stress-OO » *Extended-Lapse-Rt, Weight-to-Stress

/ananas/ Ident-Stress-OO *Extended-Lapse-Rt Weight-to-Stress > ananassa * * ananassa *!

According to Bafile (2003) Florentine loanwords can preserve the stress of the source word in contravention of the weight-to-stress requirement; but they still must shift their stress to respect the three-syllable window restriction (20). For this variety of Italian, Ident-Stress-OO falls between *Extended-Lapse-Rt and Weight-to-Stress in the constraint ranking. The Florentine examples in (20) show the spirantization of certain voiceless stops (gorgia toscana).

(20) St. Italian Florentine Taviano ananas ananasse ?ananassa filobus filobusse ??filubbussu autobus aqobusse ??autubbussu ticket tihette tikette lapis lapisse lapissi cógnac kɔ́ɲɲakke kóɲɲakka

With the analysis of the basic prosodic structure of Taviano in place, we return to the discussion of the vocative.

4. Vocatives and truncated vocatives Vocatives occur in two forms in Taviano: a full form identical to the base name and a truncated form. As near as we can tell, both types occur in the same range of contexts without major differences in meaning or pragmatic force. The main difference between the two is that the use of the truncated form normally implicates the speaker’s familiarity with the addressee and is typically used in more informal settings. The data in (21) sample a range of contexts where both freely occur: (a) calling someone out of sight, (b) calling someone within sight of the speaker to attract the addressee’s attention, (c) repeating the addressee’s name to maintain their attention, and to express pleasure (d) or displeasure (e). The Taviano forms are paralleled by the Standard Italian ones in the second half of (21).

(21) Taviano a. Mikele/Mike, addú stai? where be.2sg ‘Michael, where are you?’ b. Ah Mikele/Mike, addú sta+ bbai? where prog go.2sg ‘Oh, Michael, where are you going?’ The analysis of truncated vocatives in Taviano (Salentino) Italian CatJL 18, 2019 143

c. Senti, Mikele/Mike, sta kosa nunn e+ possibbile. hear.imp this thing neg is possible ‘Listen, Michael, that is not possible.’ d. Gwarda, Mikéle/Miké, ttʃ ’ e+bbeddu! look.imp how is beautiful ‘Look, Michael, how beautiful!’ e. Mikéle/Miké, ttʃi sta+ fatʃi? What prog do.2sg ‘Michael, what are you doing!’

St. Italian a. Michele/Miche, dove sei? b. Ah, Michele/Miche, dove vai? c. Senti, Michele/Miche, questo e impossible. d. Guarda, Michele/Miche, che bello! e. Michele/Miche, cosa fai?

4.1. Survey of data and generalizations The data in (22) illustrate the range of truncated forms for given names of different sizes and loci of the stressed syllable in Taviano. As mentioned above, the descriptive generalization noted by Alber (2009), Alber & Arndt-Lappe (2013) and Vanrell & Cabre (2011), based in part on an earlier study by Kuen (1932), is that the truncated form is identical to the base form up to the stressed vowel. Post-tonic material is deleted. Since the accented syllable is restricted to fall within a three-syllable window at the right edge of the word in Italian and the amount of pretonic material may range from zero to two or more syllables, the truncate has no fixed size. This factor makes it of particular interest in that there is no prosodic template to which the truncate is mapping—in contradistinction to hypocoristic formation, which as seen above, targets the bimoraic foot (with a single recursion as another option). Another difference from hypocoristics is that the vocative truncate is not associated with segmental markedness reductions and remains largely faithful to the base.

(22) vocatives in Taviano penultimate stress (piano) antepenultimate stress (sdrucciole) base truncate base truncate Luka Lu Tʃésare Tʃé Robbertu Robbe Kalóddʒeru Kalo Samuele Samue Kosimu Ko Filomena Filome Kristoforu Kristo Addoloráta Addolorá 144 CatJL 18, 2019 Michael Kenstowicz

4.2. Additional properties Below in (23) are more examples of vocative truncations in Taviano illustrating a number of restrictions on the output. As mentioned above, word-final consonants are largely barred in the dialect. As observed by Alber (2009) for Italian and Cabre & Vanrell (2016) for Sardinian, this factor governs the output of the vocative truncation and leads to the deletion of coda consonants from the stressed syllable of the base. These codas include /s/, homorganic nasals, liquids, as well as the first half of a geminate. Taviano thus employs two different strategies to avoid word-final consonants: copy-vowel epenthesis in the case of loanwords and truncation in the case of the vocative.

(23) more vocatives in Taviano St. Italian Taviano St. Italian Taviano base truncate base truncate Augusta Augusta Augu Roberto Robbertu Robbe Ernesto Ernestu Erne Riccárdo Rikkardu Rikka Evaristo Evaristu Evari Gilda Dʒilda Dʒi Olimpo Olimpu Oli Adelmo Adelmu Ade Alfonso Alfontsu Alfo Cataldo Kataldu Kata Franco Franku Fra Arianna Arjanna Arja Lisandro Lisandru Lisa Giuseppe Dʒuséppe Dʒusé Lorenzo Lorentsu Lore Rocco Rokku Ro Marco Marku Ma Agatella Akatedda Akate

There are, however, some additional restrictions on the Taviano truncate that cannot be traced to the background grammar. For example, while a falling sonority diphthong may terminate a word (cf. jéu váu ddái e+poi stáu kkwái ‘I go there and then stay here’), its high vocoid is clipped from the vocative truncate (24). Although there are many examples for the /au/ diphthong, it is difficult to test whether the same factor holds for /ai/ since this diphthong is rare word internally in the dialect. On the other hand, the rising sonority diphthong /je/ is faithfully preserved in the truncate. Its back vowel counterpart /wo/ is quite rare in Taviano. In Standard Italian /wo/ derives from Latin/Proto-Romance /ŏ/, which has the reflex /o/ in Salentino: cf. Latin bŏnus, Standard Italian buono, Salentino bonu. The analysis of truncated vocatives in Taviano (Salentino) Italian CatJL 18, 2019 145

(24) St. Italian Taviano base truncated vocative Paolo Paulu Pa Mauro Mauru Ma Claudio Klaudju Kla Diego Djegu Dje Daniele Danjeli Danje Gabriele Gabbrjeli Gabbrje Piero Pjeru Pje

These data suggest that the vocative aligns its right edge with the nucleus of the stressed syllable—the sonority peak. If this factor is built into the constraint that defines the right edge of the truncated vocative, the absence of coda consonants seen in (23) would not have to be attributed to the surrounding grammar but instead to a distinctive property of the vocative itself. It predicts that the same penchant for final vowels could be found in a language that otherwise licenses coda consonants.3 Some support for this interpretation comes from another restriction on the Taviano vocative: as seen in (25) truncates consisting of a single stressed vowel of the base are judged as ill-formed. Furthermore, alternatives that attempt to beef up the truncate with an augment (/Ándʒelu/ > *Áte), or an onset (/Ándʒelu/ > *Tá) or coda consonant (/Ándʒelu/ > At* ), are judged no better. On the other hand, vowel-initial bases with noninitial stress truncate easily (cf. Arjánna > Arjá) as do initial-stressed bases with a consonantal onset (cf. Márku > Má). Cabre & Vanrell (2016) report a similar restriction against the V́ … > *V́ mapping in Sardinian. Taviano speakers judge truncations to a high vowel, while still unacceptable, as not as bad as truncation to a nonhigh vowel—especially /a/ and /e/. It is conceivable that the high vowels have the tinge of a nonvocalic onset that would ameliorate the speakers’ judgments. A further phonetic study is needed to corroborate this point. The contrast between /Ándʒelu/ > *Á but Márku > Má suggests that an organic onset consonant can contribute to weight in the absence of a coda consonant as a type of ‘last resort’ and is reminiscent of some native Australian languages such as Aranda (Davis 1988) where initial onsetless syllables are skipped in the stress calculation: [kutungùla] ‘ceremonial assistant’ but [utnadawara] place name.

3. A case in point might be the Romanian dialects of Maramureş exemplified in (2d) above, although Maiden (2006) does not contain all the data needed to fully substantiate this claim. Parrott (2010) reports that penultimate stressed personal names and kinship terms in Russian can truncate their final vowel in the vocative; coda consonants are retained in some cases (Nikita > Nikit, Petrúka > Petrúk, rebjáta > rebját ‘guys’) but can be deleted in others (máma ‘mommy’> ma, pápa ‘daddy’ > pa, bába ‘granny’ > ba). Yardav (1996) describes a greater variety of such truncations for Russian; but all seem to end in a consonant or consonant cluster with no further simplification: Serjoža > Serjoš, Marinocka > Marinock, Mustafá > Mustáf, Mimi > Mim (indeclinable). 146 CatJL 18, 2019 Michael Kenstowicz

(25) St. Italian Taviano base truncate St. Italian Taviano base truncate Agata Akata *A Éva Éva *É Aldo Aldu *A Éttore Éttore *É Angelo Ándʒelu *A Íva Íva ?Í Anna Anna *A Ítalo Ítalu ?Í Ada Ada *A Ísa Ísa ?Í Élena Élena *É Ída Ída ?Í Énzo Éntsu *É Úgo Úgu ?Ú

In these cases of (25) a phonological minimality requirement against a Prosodic Word consisting of a single vowel short-circuits a morphological operation and leads to the null output as the optimal candidate. (See Cabre & Kenstowicz 1995 for a similar phenomenon in Catalan).

5. Prior analyses-1 (Alber 2009) Alber (2009) proposes a detailed analysis of the Southern Italian vocative truncation, which we review here. It is presented in the context of a comparison with the two hypocoristic truncations. The analysis is cast within the OT framework as an Output-Output relation between the base form and the truncate. The hypocoristic constructions differ from the truncated vocative by targeting a disyllabic trochaic foot—the same structure assumed for the more general stress placement in Italian. So-called ‘size-restrictor constraints’ of Foot-Binarity, Parse-syllable, and All- Ft-Left as well as the foot-form constraint Trochee impose a disyllabic, initial stress structure on the truncate. Which particular phonemes compose the truncate is controlled by anchoring constraints that specify the left and right edges of the hypocoristic with respect to the base. Alber argues that these constraints belong to the alignment family on the grounds that in Slavic languages like Russian the left edge of a truncate can pass over an onsetless syllable to target a more optimal CV structure, as in Antonjéta > Tonja. Alber’s analysis of the three truncation structures in Italian is treated as a function of the ranking of three Anchor constraints: Anchor-Left, Anchor-Right, and Anchor-Stress. For left-edge faithful nicknames such as Francésca > Fránce, Anchor-Left predominates and chooses Fránce over Césca, as seen in the tableau below (26).

(26) Anchor-Left » Anchor-Stress, Anchor-Right Francesca Anchor-Left Anchor-Stress Anchor-Right > France * sca Cesca Fran! The analysis of truncated vocatives in Taviano (Salentino) Italian CatJL 18, 2019 147

Right-edge faithful structures with penultimate stress such as Amerigo > Rigo are ambiguous between whether right-edge alignment or stressed syllable alignment is the decisive factor.

(27) Anchor-Stress, Anchor-Right » Anchor-Left Amerigo Anchor-Stress Anchor-Right Anchor-Left (Rigo) Ame (Ame) * rigo

But antepenultimately accented formations like Doménico > Méni and Cristoforo > Stofo show that anchoring to the stressed syllable must dominate anchoring to the right edge. Alber treats this factor as faithfulness between the stress of the base and the stress of the truncate.

(28) Anchor Stress » Anchor-Right, Anchor-Left Domenico Anchor-Stress Anchor-Right Anchor-Left > (Meni) co Do (Nico) *! Dome

In the majority of cases the size restrictor constraints and Contiguity predomi- nate to shift the right edge of the truncate leftward past the final syllable and thus choose Méni over Ménico and Méco.

(29) SRC, Contiguity » Align-Right Domenico SRC Contiguity Align-Right > (Meni) co (Menico) *! (Meco) ni!

Alber cites formations like Doménico > Ménico and Cristoforo > Stoforo from Thornton (1996) to argue that Anchor-Right also plays a role in defining the templates for truncation and hence must figure into the typology even though it lacks the psycholinguistic backing of Anchor-Left and Anchor-Stress, which are known to play a significant role in lexical access (Cutler 2012).

(30) Anchor-Right, Contiguity » SRC Domenico Anchor-Right Contiguity SRC > (Menico) * (Meni) co! 148 CatJL 18, 2019 Michael Kenstowicz

Contiguity is high ranked since the truncate normally forms a single block of segments excised from the base though for markedness reasons violations may occur sporadically, as in Beatrice > Bice.

(31) *Complex Onset » Contiguity /Beatrice/ *Complex-Onset Contiguity > Bice eatr Trice *!

With this background in place we turn to Alber’s (2009) analysis of the truncated vocatives. The basic intuition lying behind the analysis is that their variable and unrestricted size can be fitted within the overall typology of truncations by calling on the two major constraints Anchor-Left and Anchor-Stress that define the hypocoristics and letting them dominate the size-restrictor constraints. This move produces an ‘atemplatic’ output which is said to not be enforced by any metrical (rhythmic) considerations. The truncation of Salvatore > Salvato is shown in tableau (32) below.

(32) Anchor-Left, Anchor-Stress » SRC /Salvatore/ Anchor-Left Anchor-Stress SRC > Salvato * Salva *! Tore Salva!

Since for Alber the preservation of the stressed syllable of the base in the truncate is treated as a faithfulness effect rather than as a product of alignment, another constraint must be called on to implement the truncation of the post-tonic material. Otherwise, lower-ranked Anchor-Right will stretch the truncate to align with the base at its back end, generating the fully faithful candidate as the output (33).

(33) /Salvatore/ Anchor-Left Anchor-Stress Anchor-Right SRC Salvato re! * Salva *! Tore Salva! > Salvatore

To solve this problem Alber appeals to a member of the Coincide family of constraints from Zoll (1996): Coincide-Syll-1. This constraint requires all of the The analysis of truncated vocatives in Taviano (Salentino) Italian CatJL 18, 2019 149 syllables to align with the left edge of the prosodic word. Since they cannot pile on top of one another, deletion is the optimal repair and eliminates all but the first syllable from the output.

(34) /Salvatore/ Coincide-Syll-1 Align-Right > Salvato vato * Salvatore vatore!

Alber utilizes the Coincide-Syll-1 constraint to generate monosyllabic hypocoristics such as those in (35) below which, while not otherwise mentioned in the literature, were collected from Northern Italian students at Verona and independently confirmed for us by a speaker from Trento.

(35) Francesca Fra Cristina Cri Luisa Lu Stefania Ste

The tableau below shows this treatment of these structures in Alber’s analysis.

(36)

Base: Francesca Coincide-σ1 FtBin Anchor-L Anchor-S Anchor-R ☞ a. (Frá) * * ncesca

b. (Frán.ce) ce! * sca

c. (Cé) * Fran! sca

d. (Scá) * France! *

Without knowing more about the background grammar of such speakers, it is difficult to judge the plausibility of this analysis—in particular that Coincide- Syll-1 is the factor responsible for shaving off the right edge of the base in (34). For example, it is unclear how to account for the loss of the offglide in Páulu > Pá since the alternative Páu satisfies faithfulness for stress and Coincide-Syll-1. The substratal Northern Italian dialects are well known for unstressed vowel deletion (apocope and syncope). It is conceivable that the derivation of Frá might proceed from intermediate Frán.ce rather than being directly generated from the full name. If so, then the independent motivation for the Coincide-Syll-1 constraint is called into question and hence its role in the vocative truncation as well. 150 CatJL 18, 2019 Michael Kenstowicz

6. Prior-Analyses-2 (Vanrell & Cabre 2011; Cabre & Vanrell 2016; Cabre et al. 2019) Based on corpora collected from Sardinian, Eastern Catalan, and some Southern Italian speakers, Vanrell & Cabre (2011), Cabre & Vanrell (2016), Cabre et al. (2019) treat the vocative truncation as a byproduct of constraints aligning the truncate with nuclear pitch accents that are associated with the vocative formations. Here we focus on their discussion of Sardinian (Cabre et al. 2019), which is the most worked out analysis. Their data were collected from six native speakers who were asked to imagine two scenarios for using a vocative. In the first (‘direct call’), the speaker calls an unseen addressee assumed to be close by (e.g. in another room in a house). In the second scenario (‘insistent call’), the speaker, having received no response to the direct call, repeats the call assuming the addressee to be located more distantly but still within earshot of the speaker. Similarly to our data with Taviano (recall 21 above), the specific intonational tune can be applied to both truncated as well as non-truncated, full-form vocatives. For Sardinian it consists of a L+H* rise mapped to the first syllable and an L* assigned to the stressed syllable followed by a L% boundary tone4. The intonation contour is illustrated by the truncated form of Eleonore > Eleono with a rise on the first syllable and a low target on the stressed syllable at the right edge of the truncate (37). The two medial syllables appear to be associated with the H.

(37)

As far as we can tell, Cabre et al. (2019) propose a two-step derivation of the vocatives. In the first step, the two pitch accents associated with the vocative

4. The L+H* notation seems to imply a stress accent on the initial syllable. As near as we can tell, a %LH boundary tone would also be consistent with the Sardinian data. The analysis of truncated vocatives in Taviano (Salentino) Italian CatJL 18, 2019 151 are mapped to the full-form of the base: “Whether truncation happens or not, the intonational pattern applied to the base is the same” (p. 35). In a second step, the truncate is derived from the full form vocative by a categorical alignment constraint that requires the syllable with the L* pitch accent to coincide with the right edge of the prosodic word. Thus, in this analysis, the truncation is in effect stipulated as a product of alignment and both are compiled into a single constraint that transforms the full form vocative into the truncated form. This approach contrasts with Alber’s more granulated interpretation in which alignment to the stressed syllable is factored out into a more general stress faithfulness phenomenon that is found outside of the truncation per se and which we saw plays a significant role in Taviano loanword adaptation and more generally in output-output faithfulness in derivational morphology.

7. Prior-Analyses-3 (D’Alessandro & van Oostendorp 2016) It is precisely this point that D’Alessandro & van Oostendorp (2016) focus on in their discussion of vocative truncation. They motivate the truncation with the assumption that the nuclear tone falling on the stressed syllable of the vocative originates as a right-edge boundary tone in the construction. Truncation is interpreted as a strategy to minimize the distance between the right edge of the containing structure and the boundary tone so that they coincide. But given that the vocative construction can apparently occur with the same intonation contour in both the full form as well as the truncated form, there is no necessary connection between truncation and the vocative. At best, under this analysis, truncation is an optimizing TETU-like effect that improves the fit between the F0 contour and its segmental “text”. Future research should conduct an instrumental study of the full form and the truncated vocatives to establish in fact whether or not their pitch contours are distinct and if so whether the truncated form can reasonably be derived from a right-edge boundary tone. See Jeong (2018) for a recent study along these lines that differentiates two types of rising declaratives in English.

8. Summary of Phonological Characteristics of the Vocative While more data are surely welcome, it is possible to discern some general trends in the typology of vocatives that distinguish them from hypocoristics. First, hypocoristics are templatic while vocatives are not. Second, hypocoristics may be associated with truncation at the left or the right edge of the base. All cases of vocative truncation that we are aware of truncate material at the right edge and remain faithful to the left edge. Third, vocative truncation seems to be motivated by the desire to align the right edge of the truncate with prominence. The prominence can take a number of forms: the sonority peak of the stressed syllable (Taviano), a stressed or stressless syllable with a consonantal coda (Romanian, Russian), a pitch accent (Sardinian), or a lengthened vowel (Yupik: Miyaoka 2012: 859). Fourth, while hypocoristics are often accompanied by segmental and syllabic markedness reductions (TETU), vocatives seem more faithful to the base. In sum, the phonol- 152 CatJL 18, 2019 Michael Kenstowicz ogy of vocatives mimics the general character of phonology (and more broadly of grammar) as a whole. There are general trends with possible external motivation; but they are embedded in a system of conflicting forces that can be resolved in various ways. More research is needed to better circumscribe the overall typology.

9. Morphological and Grammatical Properties of the Taviano Vocative In addition to phonological constraints such as the minimality requirements discussed in 4.2, Salentino’s vocative truncation also obeys morphosyntactic constraints pertaining to the nature and size of possible bases. In the canonical case, the truncation base is a single word, such as a name, a nickname, a conventional or professional title, or any other form of address, including plurals (38).

(38) a. Briganti → Brigá surname b. Piɲɲatédda → Piɲɲaté nickname, lit. ‘small pot’ c. professóre → professó ‘professor’ d. vaɲɲunʧéddu/-ddi → vaɲɲunʧé ‘little boy/boys’[diminutive of vaɲɲóne ‘boy’]

A branching structure made up of two words is also relatively acceptable (39), albeit with somewhat degraded outcomes in cases of adjectival or possessive modification (40).5

(39) a. siɲɲóra Terésa → siɲɲora Teré ‘Mrs. Teresa’ b. méʃʃu Marju → meʃʃu Má ‘master craftsman [title] Mario’ c. Marku Ritʦu → Marku Rí ‘Marco Rizzo’ [name and surname]

(40) a. kummare nóʃʃa → ?*kummare no grandmother our.F.SG b. béddu vaɲɲóne → ?*beddu vaɲɲó beautiful.M.SG boy

By contrast, more complex NP structures are all decidedly unacceptable as inputs for vocative truncation, despite making for fine vocative nominals if left untruncated (41).

5. Similar examples of minimally complex vocative truncates have been noted elsewhere in the literature on vocative truncation in Italo-Romance. Rohlfs (vol. 1: 448-9) reports bella fé ‘beautiful woman’ for Abruzzese, ʦi mmo ‘uncle monk’ for Ischia’s dialect, and Jaku Fa (name + surname) for Sardinian. Comparable examples involving compound names and surnames have recently been discussed by D’Alessandro & Van Oostendorp (2016). The analysis of truncated vocatives in Taviano (Salentino) Italian CatJL 18, 2019 153

(41) a. Marku Ritʦu de Tajanu → *Marku Ritʦu de Tajá name surname from Taviano b. beɖɖu de nonna → *beɖɖu de nó darling of grandmother ‘Granny’s darling’ c. tíe ka pórti li kkjali → *tíe ka porti li kkjá 2sg that wear.2sg the glasses ‘You who are wearing glasses’

Importantly, in those limited cases where truncation can target a branching nominal phrase, truncation never applies distributively to each component word of that phrase, but always to the right edge of the phrase as a whole (42). This pattern thus differs starkly, for example, from the one displayed by vocative exponence in Latin and other Indo-European languages, where vocative “case” morphology shows up (under most circumstances, obligatorily) on every possible component word within the vocative nominal (43).

(42) a. siɲɲóra Terésa → siɲɲora Teré/*siɲɲó Teré/*siɲɲó Teresa ‘Mrs Teresa’ b. méʃʃu Marju → meʃʃu Ma/*me Ma/*me Marju ‘master craftsman [title] Mario’

(43) Latin (Plautus, 3rd-2nd century BC) uālē bene, ocul-e mī fare.imp well, eye-voc my.voc ‘farewell, my eye!’ [The nominative would have been oculus meus]

Although exempt from concord phenomena such as that in (43), Taviano’s vocative truncation does behave like canonical inflectional morphology in that it may be fed by derivational morphology; but it can never feed derivation. In general, vocative truncation appears to be a morphological “dead end,” never providing an input for further affixation, as illustrated by the impossibility of appending diminutive suffixes like -úʧʧu and -ínu to any vocative truncate. The problem with such diminutivized truncates cannot straightforwardly be reduced to the hiatus between the truncate-final vowel and the ú/í of the diminutive, given that such hiatuses are allowed elsewhere in the language (e.g. faúɲɲu ‘sultriness’, kjaútu ‘coffin’).

(44) a. Dunátu → Duná ↛ *Dunaúʧʧu b. Páulu → Pá ↛ *Paínu

The ungrammatical examples in (44) should be contrasted with the grammatical ones in (45), where diminutive morphology applies to the base rather than to the truncate. 154 CatJL 18, 2019 Michael Kenstowicz

(45) a. Dunátu → Dunatúʧʧu → Dunatú b. Páulu → Paulínu → Paulí

The fact that truncation can apply equally successfully to a non-diminutive or a diminutive base (compare the underlined forms in (44) and (45)) also shows that there is no privileged base that the process preferentially applies to. This point is further illustrated in (46) with examples involving hypochoristics and diminutives of hypochoristics.

(46) a. Dʒuséppe → dʒusé (plain base) b. Péppe → Pé (hypochoristic) c. Péppe → Peppínu → Peppí (diminutive of hypochoristic)

Finally, it is also worth noting the existence of syntacto-semantic constraints on Taviano vocativization, which apply to truncated and non-truncated vocatives alike. In particular, it turns out to be impossible for the vocative nominal to be preceded by any articles, despite the fact that they are generally obligatory on definite nominals, including proper names. For example, the relevant allomorphs of the singular definite article — (l)u and l’ for the masculine, (l)a and l’ for the feminine — cannot be omitted in argumental uses of proper names such as those exemplified in (47). But they must be omitted in the corresponding vocatives (48).

(47) a. {U/*Ø} Kósimu è+ ʃutu kkasa. def Cosimo is gone home ‘Cosimo has gone home.’ b. Adʤu kuntatu ku+ *(la) María have.1sg talked with def Maria ‘I have talked with Maria.’

(48) a. (*A) Marí(a), si+ ʃuta kkasa? def Maria be. 2sg gone home ‘Maria, are you back home?’ b. (*U) Kó(simu), tokka+ kuntamu def Cosimo is.necessary talk.1pl ‘Cosimo, we need to talk.’

As already mentioned, this restriction holds of vocatives irrespective of truncation, and instantiates a cross-linguistically frequent pattern. Similar bans on articles in vocatives have been reported for several other languages that admit or require articles before proper names. Cases in point include Seri (isolate, Mexico; Marlett 2008), Austrian German (Göksel & Pochträger 2013: 89), Modern Greek, The analysis of truncated vocatives in Taviano (Salentino) Italian CatJL 18, 2019 155 and Venetian dialects of Italo-Romance (Hill 2014: 66).6 A natural explanation for this pattern revolves around the idea that “[i]n the Vocative Phrase, definite articles do not check definiteness, since the addressee semantics is inherently definite and referential” (Hill 2014: 62). This definiteness requirement is presumably imposed — via syntactic selection or semantic definedness conditions — by the functional structure that hosts the vocative nominal. In this respect, we follow the literature (Moro 2003; Stavrou 2009, and much subsequent work) in positing a vocative- dedicated functional head Voc , which we take to host the vocative noun phrase in its specifier, and to be a part of the highest functional field at the left periphery of the clause, encoding information⁰ about discourse participants and utterance context.

(49) VocP

NP Voc’

Voc ForceP?

Other heads ⁰ in the same region of the clausal structure have been invoked in accounts of phenomena such as speaker- and addressee-oriented adverbs, allocu- tive agreement, indexical shift, and a lot more. Unlike some of these phenomena, however, the possibility of expressing a vocative seems to be cross-linguistically restricted to matrix-clause environments only, which we take to show that VocP can never be projected in the periphery of an embedded clause, no matter how structurally big.

(50) a. Siɲɲore, da-mme a salute Lord give-imp-1sg.clit the health ‘O my Lord, give me good health.’ b. Oʃi preku (*Siɲɲore) ku+ (*Siɲɲore) me daje a salute Today pray.1sg Lord comp Lord 1sg.clit gives the health ‘Today I’ll pray that He give me good health.’ c. Oʃi nne diku (*Siɲɲore) ka (*Siɲɲore) tokka+ me Today 3.dat say.1sg Lord comp Lord must 1sg.clit daje a salute gives the health ‘Today I will tell Him that He has to give me good health.’

6. Similarly but not identically, vocative nominals in Zulu do not need any “augment” (Halpert 2012: 136; Janson 2013: 230-231) — a prefix analyzed as an “intrinsic case licensor” by Halpert (2012). 156 CatJL 18, 2019 Michael Kenstowicz

10. Summary and conclusion This paper has documented and analyzed the formation of truncated vocatives in the Taviano dialect of Salentino Italian. Our findings align in general terms with those reported for Sardinian, Algherese Catalan, and certain Romanian dialects. The truncate has no fixed size in contrast to hypocoristics, which target a disyllabic trochaic foot. The vocative truncate is faithful to the left edge of the base while its right edge aligns with the base’s stressed syllable. The paper discussed whether or not additional restrictions on the truncate can be attributed to the background grammar or must be stipulated as properties of this particular construction. Our investigation of Taviano uncovered novel findings on the shape of word-final syllables, which are enforced differently between native words and loanwords. Subtle differences with regard to the locus of stress in loanword adaptation among Taviano, Florentine, and Standard Italian can be expressed as different rankings in an OT grammar. The paper concluded with a discussion of certain morphological and syntactic restrictions on the Taviano vocatives and their parallels in other languages. More cross-linguistic research is needed to better circumscribe the range of variation of this until recently largely overlooked structure.

References Alber, Birgit. 2010. An exploration of truncation in Italian. Rutgers Working Papers in Linguistics 3: 1-30. Alber, Birgit & Arndt-Lappe, Sabine. 2012. Templatic and subtractive truncation. In Trommer, Jochen (ed.). The Morphology and Phonology of Exponence, 289-325. Oxford: Oxford University Press. Bafile, Laura. 2003. Le consonanti finali nel fiorentino e nel napoletano. Rivista Italiana di Dialettologia 27: 149-178. Burani, Cristina, Paizi, Despina & Sulpizio, Simone. 2014. Stress assignment in reading Italian: friendship outweighs dominance. Memory & Cognition 42: 662-675. Boyd-Bowman, Peter. 1955. Como obra la fonetica infantile en la formacion de los hipocoristicos. Nueva Revista de Filologia Hispánica 9: 337-366. Cabre, Teresa & Kenstowicz, Michael. 1995. Prosodic trapping in Catalan. Linguistic Inquiry 26: 694-705. Cabré, Teresa & Vanrell, Maria del Mar. 2011. Accent i entonació en els vocatius de l’alguerès. In La toponomàstica de les illes del Mediterrani occidental (XXXV Col·loqui de la Societat d’Onomàstica l’Alguer, maig del 2008), València-l’Alguer: Editorial Denes, S.L., 331-341. Cabré, Teresa & Vanrell, Maria del Mar. 2016. Truncation in Sardinian: prosodic vs. intonational templates. Unpublished ms. Cabre, Teresa; Torres-Tamarit, Francesc & Vanrell, Maria del Mar. 2019. Truncation in Sardinian. Unpublished ms. Calabrese, Andrea. 1984. Metaphony in Salentino. Rivista di Grammatica Generativa 9-10: 1-140. Chierchia, Gennaro. 1986. Length, syllabification, and the phonological cycle in Italian. Journal of Italian Linguistics 8: 5-33. The analysis of truncated vocatives in Taviano (Salentino) Italian CatJL 18, 2019 157

Cutler, Anne. 2012. Native Listening: Language Experience and the Recognition of Spoken Words. Cambridge, MA: MIT Press. D’Alessandro, Roberta & van Oostendorp, Marc. 2016. When imperfections are perfect: prosody, phi-features and deixis in Central and Southern Italian vocatives. In Carrilho, Erenestina, Fieis, Alexandra, Lobo, Maria & Pereira, Sandra. Romance Languages and Linguistic Theory 10: Selected Papers from ‘Going Romance’ 28 Lisbon, 61-82. Amsterdam & Philadelphia: John Benjamins. Davis, Stuart. 1988. Syllable onsets as a factor in stress rules. Phonology 5: 1-19. Göksel, Aslı & Pochträger, Markus. 2016. The vocative and its kin: marking function through prosody. In Sonnenhauser, Barbara & Hannah, Aziz (eds.). Vocative! Addressing between System and Performance, 87-107. Berlin: De Gruyter Mouton. Feliu, Elena. 2001. Output constraints on two Spanish word-creation processes. Linguistics 39(5): 871-891. Halpert, Claire. 2012. Argument Licensing and Agreement in Zulu. Ph.D. dissertation. Cambridge (MA): Massachusetts Institute of Technology. Hill, Virginia. 2014. Vocatives: How Syntax Meets with Pragmatics. Leiden: Brill. Ito, Junko. 1990. Prosodic minimality in Japanese. Chicago Linguistic Society 26(2): 213-239. Ito, Junko, Kubozono, Haruo & Mester, Armin. To appear. A prosodic account of consonant gemination in Japanese loanwords. In Kubozono, Haruo (ed.). The Phonetics and Phonology of Geminate Consonants. Oxford: Oxford University Press. Janson, Tore. 2016. Vocative and the grammar of calls. In Sonnenhauser, Barbara & Hannah, Aziz (eds.). Vocative! Addressing between System and Performance, 219- 234. Berlin: De Gruyter Mouton. Jeong, Sunwoo. 2018. Intonation and sentence type conventions: two types of rising declaratives. To appear in Journal of Semantics. Kenstowicz, Michael. 2005. The phonetics and phonology of Korean loanword adaptation. In S-J. Rhee (ed.). Proceedings of the First European Conference on Korean Linguistics, 17-32. Seoul: Hankook Publishing Co. Kurisu, Kazutaka. 2001. The Phonology of Morpheme Realization. Ph.D. dissertation, University of Californa Santa Cruz. Kiparsky, Paul. 1984. On the lexical phonology of Icelandic. In Elert, Claes-Christian, Johansson, Irene & Strangert, Eva (eds.). Nordic Prosody III, 135-164. University of Umea. Kuen, Heinrich. 1932. El dialecto de Alguer y su posición en la historia de la lengua catalana. Anuari de l’Oficina Romànica de Linguistica i Literatura V: 121-177. Maiden, Martin. 2006. On Romanian imperatives. Philologica Jassyensia, An II, Nr. 1: 47-59. McCarthy, John & Prince, Alan. 1986. Prosodic Morphology. Ms. University of Massachusetts, Amherst and Brandeis University, Waltham, Mass. [Annotated version 1996, issued as Technical Report no. 32, Rutgers Center for Cognitive Science.] Marlett, Stephen. 2008. The form and use of names in Seri. International Journal of American Linguistics 74: 47-82. Martinez-Paricio, Violeta & Torres-Tamarit, Francesc. To appear. Trisyllabic hypocoristics in Spanish and layered feet. Natural Language & Linguistic Theory. 158 CatJL 18, 2019 Michael Kenstowicz

Miyaoka, Osahito. 2012. A Grammar of Central Alaskan Yupik. Berlin: De Gruyter Mouton. Moro, Andrea. 2003. Notes on vocative case: a case study in clause structure. In Quer, Josep & Schroten, Jan (eds.). Romance Languages and Linguistic Theory 2001, 247-261. Amsterdam: John Benjamins. Oresnik, Janez. 1972. On the epenthesis rule in Modern Icelandic. Arkiv for Nordisk Filologi 87: 1-32. Parrott, Lillian. 2010. Vocatives and other direct address forms: a contrastive study. In Grønn, A. & Marijanovic, I. (eds.). Russian in Contrast, Oslo Studies in Language 2(1): 211-229. Passino, Diana. 2008. Aspects of Consonantal Lengthening in Italian: A Case Study of Gemination of Word-final Consonants in the Adaptation of Loanwords. HALCCSD. Revised version of University of Padua Ph.D. dissertation, 2005. Piñeros, Carlos Eduardo. 2000. Prosodic and segmental unmarkedness in Spanish truncation. Linguistics 38: 63-98. Prieto, Pilar. 1992. Truncation processes in Spanish. Studies in the Linguistic Sciences 22(1): 143-158. Repetti, Lori. 1993. The integration of foreign loans in the phonology of Italian. Italica 70(2): 182-196. Repetti, Lori. 2012. Consonant-final loanwords and epenthetic vowels in Italian. Catalan Journal of Linguistics 11: 167-188. Rohlfs, Gerhard (1966-1969). Grammatica Storica della Lingua Italiana e dei Suoi Dialetti. 3 vols. Turin: Einaudi. Shinohara, Shigeko. 1997. Analyse phonologique de l’adaptation japonaise de mots étrangers. Ph.D. dissertation. University of Sorbonne-Nouvelle Paris III. Stanton, Juliet & Sam Zukoff. 2018. Prosodic identity in copy epenthesis: evidence for a correspondence-based approach. Natural Language & Linguistic Theory 36(2): 637-684. Stavrou, Melita. 2009. Vocative! Ms. Aristotle University of Thessaloniki. Thornton, Anna. 1996. On some phenomena of prosodic morphology in Italian: accor- ciamenti, hypocoristics and prosodic delimitation. Probus 8: 81-112. Vanrell, Maria del Mar & Cabré, Teresa. 2011. Troncamento e intonazione dei vocativi in Italia centromeridionale. In Gili-Fivela, B., Stella, A., Garappa, L., & Grimaldi, M. (eds.). Contesto comunicativo e variabilità nella produzione e percezione della lingua: Atti del 7º convegno AISV. Roma: Bulzoni. Yardoff, Michael. 1996. Modern Russian vocatives: a case of subtractive morphology. Journal of Slavic Linguistics 4(1): 133-153. The analysis of truncated vocatives in Taviano (Salentino) Italian CatJL 18, 2019 159

Appendix

barra ‘iron bar’ barra ‘bar, cafe’ Brindisi placename karémme Puppets symbolizing Lent kastími ‘curse’ 2 sg. Fiatta brandname sorelle ‘sisters’ mamma ‘mama’ Mikéle personal name otelle ‘hotel’

piata ‘take.imp.2sg 3fem.sg for yourself’

piatele ‘take.imp.2sg those.fem ridimmi ‘tell me again’

ridissi ‘say again’ (priterite 1.sg)

rifatta ‘made again’ (participle fem) ste#ʃʃéme ‘these idiots’ tandemme ‘tandem’ (two-seat bicycle) tennissi ‘tennis’ tramma ‘tram’ ultimi ‘final’ pl. upimmi brand name

venime ‘come imp.2sg-1sg.dat’

Catalan Journal of Linguistics 18, 2019 161-189

Prominence Augmentation via Nasalization in Brazilian Portuguese*

Andrew Nevins University College London [email protected] Paula Pinheiro Costa Universidade Federal do Rio de Janeiro [email protected]

Received: June 3, 2019 Accepted: August 13, 2019

Abstract

This article aims to demonstrate that dialectal and idiolectal variants of Brazilian Portuguese that exhibit rhotic metathesis (e.g. vidro > vrido ‘glass’), spontaneous nasalization of high vowels (as in diachronic hibernum > inverno ‘winter’ and non-standard ingreja ‘church’), and pretonic vowel lowering of mid-vowels are all instantiations of the same process: prominence-boosting in stressed, secondary-stressed, or word-initial positions. Keywords: spontaneous nasalization; Brazilian Portuguese; rhotic metathesis; prominence augmentation; initial syllables

Resum. Augment de la prominència per nasalització en el portuguès del Brasil

Aquest article vol demostrar que els següents tres fenòmens que exhibeixen les varietats dia- lectals i idiolectals del portuguès del Brasil, i.e. la metàtesi ròtica (e.g. vidro > vrido ‘vidre’), la nasalització espontània de les vocals altes (com el canvi diacrònic hibernum > inverno ‘hivern’ o la forma no estàndard ingreja ‘església’) i l’abaixament de les vocals medials pretòniques, són manifestacions del mateix procés: reforçament de la prominència de les posicions tòniques, d’accent secundari o de síl·labes inicials. Paraules clau: nasalització espontània; portuguès del Brasil; metàtesi ròtica; augment de la pro- minència; síl·labes inicials

* We thank Livia Camargo Souza, Gean Damulakis, Guilherme Garcia, Clàudia Pons-Moll, Arthur Santana, Filomena Sandalo, Leo Wetzels, and the editors and reviewers for valuable comments and suggestions on the analysis.

ISSN 1695-6885 (in press); 2014-9718 (online) https://doi.org/10.5565/rev/catjl.291 162 CatJL 18, 2019 Andrew Nevins; Paula Pinheiro Costa

Table of Contents 1. Overview 4. Nasality as a Contributor of Weight 2. Prominence-Demanding Positions 5. Conclusions and future directions in Brazilian Portuguese References 3. Spontaneous Nasalization Appendix: List of Non-Standard forms as Prominence-Augmentation with Spontaneous Nasalization

1. Overview Nasality on vowels is a well-known feature of Portuguese, a diachronic development from nasal coda consonants, with the existence of forms such as Spanish lana, Italian lana, Catalan llana, Romanian lână ‘wool’ alongside Portuguese lã [lɐ̃] clearly demonstrating the latter language lost the coda consonant altogether, in favor of nasalization of the preceding vowel. There are, however, what are called ‘non-etymological’ cases of nasalization found in Portuguese (alongside the well-known case of muito [mũj.tʊ]̃ ‘many’, standardly nasalized and simply not orthographically indicated), and these are the non-standard (and sometimes less prestigious) pronunciations such as idioma [ˌĩ.d͡ ʒi.’õ.mɐ] ‘language’, igreja [ĩ.ˈgɾe.ʒɐ] ‘church’, identidade [ĩ.ˌdẽ.t͡ ʃiˈda.d͡ ʒɪ] ‘identity’ and idiota [ĩ.d͡ ʒiˈɔ.tɐ] ‘idiot’ with a initial nasal vowel, the last of which is even rendered, tongue-in- cheek, in the official promotional poster for the movie Bob Esponja: O Incrível Resgate).1 All of these preceding forms – some of the most representative and well- known cases – have in common the fact that they involve a high front vowel in the initial syllable, and through a quantitative analysis of over two hundred examples presented within this paper, we conclude that there is a pattern informing why spontaneous nasalization arises in examples such as these, as opposed to being randomly articulatorily distributed throughout the language. In short, we claim that spontaneous nasalization is a strategy to perceptually enhance prominence, and tends towards particular vowels in particular syllabic positions, arising indeed potentially ‘spontaneously’, as opposed to being a case of variable dialectal lexicalization. Thus, during a visit to a virtually unknown fishing village in Brazil named Picinguaba [̩pi.s̩ ĩ.ˈgwa.bɐ], a highly literate Portuguese speaker called it [̩p̩ ĩ.si.ˈgwa. bɐ], apparently transposing the nasal vowel to the first syllable. This ‘real-time error’ was not the result of dialectal divergence (as the toponym is so low frequency that one cannot claim there are two lexicalized variants), but rather, we contend, revealing of a latent and recurring tendency that occurs in the language: to make stressed syllables even more prominent than they otherwise would be. Specifically, the foot structure of this four-syllable word is composed of two trochees, whereby the initial syllable has secondary stress (see Collischonn 1994). The strong syllable within the first foot should ideally have more prominence than the weaker syllable in this foot. Transposing the nasality from the prosodically weaker syllable onto the stronger one

1. See . Prominence Augmentation via Nasalization in Brazilian Portuguese CatJL 18, 2019 163 accomplishes just that, in a manner similar to what Smith (2005) called Prominence Augmentation. The intuition is that certain prominent positions, in particular, stressed and/ or initial syllables, demand the presence of prominence-increasing material (a clearly violable tendency, but one present, the force of which when exerted can be observed). Highlighting this particular case with a version of Zoll’s (1998) Coincide constraints, which demand marked segments such as nasal vowels in initial syllables, would look as follows:2

(1) Tableau for Nasal Metathesis towards Initial Syllable

/ ̩pi.sĩ.ˈgwa.bɐ/ Coincide(σ1,NasalV) Dep [+nasal] Linearity F [̩pĩ.si.ˈgwa.bɐ] * [ ̩pi.sĩ.ˈgwa.bɐ] *! [ ̩pĩ.sĩ.ˈgwa.bɐ] *!

With this small ‘error’ in encoding and reproducing the town’s toponym, we have an entire phonological tendency writ small: what properties of this particular word are responsible? This requires spelling out the following three questions for Brazilian Portuguese: (A) when does a syllable require prominence (i.e. a theory of secondary stress and prominent positions), as in the case of the initial syllable of this four-syllable paroxytone word? (B) what kinds of syllables are needier than others (e.g. those with higher vowels, or those with one or fewer consonants in the onset)? (C) What kinds of processes (and segments) can be employed to enact prominence-boosting, and how can these be formally represented? The Stress-to-Weight Principle (SWP; e.g. Prince 1990) is employed to explain cases in which a stressed syllable (whether lexically or predictably stressed) undergoes additional phonological processes to gain weight, such as addition of a coda consonant (in Italian raddoppiamento sintattico; Borrelli 2002) or post-tonic syncope in Tonkawa (Gouskova 2007), and this principle affords insight into a wide range of strategies. Indeed, glide insertion before [s] in final-stressed syllables in the Carioca dialect (Rio de Janeiro) of Brazilian Portuguese (BP) (e.g. rapa[j]s ‘guy’) has been analyzed in Nevins (2015) as an instance of the SWP, under the assumption that coda [s] is non-moraic and hence such stressed syllables require additional weight. Nonetheless, there are certain processes affecting primary (and secondary) stressed syllables that cannot be easily modeled in terms of weight- addition via moraic structure per se, but nonetheless do involve a broader notion of prominence (see e.g. Ryan 2014 for application of this notion for weight-to-stress cases), as shown in (2a)-(2d):

2. We include secondary stress as the input to the tableau for expositional purposes, remaining agnostic as to whether this is the result either of underlying assignment, or the presence of a constraint not shown in this tableau, or the output of an earlier stratum of phonological computation (as secondary-stress has a post-lexical character in BP; Collischonn 1994). 164 CatJL 18, 2019 Andrew Nevins; Paula Pinheiro Costa

(2) Prominence Scales employed in Augmentation a. diphthongs > simplex nuclei b. complex onsets > onsetful > onsetless c. nasal vowels > oral vowels d. low vowels > mid vowels > high vowels

The scales in (2) can be derived from a combination of increased duration for all of these, as is clearly true for cases of more segmental material (2a-b), and has been phonetically demonstrated for (2c) by Moraes & Wetzels (2003) and for the vowel scale in (2d) by Crosswhite (1999). We contend that the phonologized effects of weight are stated in terms of weight-sensitive comparisons in phonological terms, although they phonetically derive from increased duration.

2. Prominence-Demanding Positions in Brazilian Portuguese As is generally known throughout the Romance languages, main stress on all lexical items besides verbs (which are outside the scope of this study) falls within a three-syllable window from the right edge, enabling minimal triplets (e.g. sábia ‘sage’, sabía ‘knew-imperf.’, sabiá ‘thrush’), with penultimate stress being the default. Final stress occurs predictably when the final syllable is heavy, ending in a glide or rhotic (e.g. ane[w] ‘ring’, abajur ‘lampshade’, hero[j] ‘hero’). This quantity-sensitivity, discussed by Bisol (1992) and Wetzels (2007), does not always hold for final sibilants, which is what compels glide epenthesis. The productivity of these principles, as discussed by Hermans & Wetzels (2012) can be found in new coinages such as acronyms and pharmaceuticals, which follow these principles blindly. None of these bear orthographic accents; the accents shown below indicate where stress is assigned phonologically:

(3) Novel formations showing adherence to Main-Stress Assignment a. Final WSP: Detran, Procon, Funai, Susau, Benes, Proer, Varsul, Anpol, Efexor, Dorless, Beserol b. Default penult otherwise: Óvni, Bradésco, Texáco, Úfba, Clopsína c. Final epenthesis leads to penult: Fapéspi, Telprági, Valméti, Nisulídi

More spectacular cases come from blend first-names, which are a goldmine for study given the onomastic creativity of anthroponyms in Brazil: Gustavo+Maria = Gusmar, and the famous case of the six soccer players from the 1970 World Cup Tostão+Pele+Rivelino+Carlos Alberto+Gérson+Jairzinho yielding the first name Tospericargerja (born that year), whose unusual name bears no need for an orthographic accent because, as inventive as his parents were, his name still has default penultimate stress. Given the Weight-to-Stress-Principle holding over final syllables, followed by penultimate stress as a default, exceptional lexically-marked stress is therefore any Prominence Augmentation via Nasalization in Brazilian Portuguese CatJL 18, 2019 165 case of antepenultimate stress or any case of penultimate stress when the final is heavy. Indeed, antepenultimate stress is not only unproductive, but has a tendency to be levelled out to penultimate stress, as found in colloquial pronunciations such as (4), in which the medial vowel is deleted (see França 2009, among others):

(4) Examples of antepenultimate stressed words undergoing syncope to become default penutimate-stressed: a. árvore [ˈah.vo.ɾɪ] > arvre [ˈah.vɾɪ] ‘tree’ b. abóbora [a.ˈbɔ.bo.ɾɐ] > abobra [a.ˈbɔ.bɾɐ] ‘pumpkin’ c. xícara [ˈʃi.ka. ɾɐ] > xicra [ˈʃi.kɾɐ] ‘teacup’ d. córrego [ˈkɔ.ɦe.gʊ] > corgo [ˈkɔɦ.gʊ] ‘creek’ e. estômago [iʃ.ˈtõ.ma.gu] > estomgo [iʃ.ˈtõ.gu] ‘stomach’ f. ônibus [ˈõ.nɪ.bʊs] > onbus [ˈõj.bʊs]̃ ‘bus’ g. cônjuge [ˈkõ.ʒu.ʒɪ] > conge [ˈkõ.ʒɪ] ‘spouse’

As Herman & Wetzels (2012) point out, there are two sets of regular antepenultimate stress: those that involve a high vowel in a hiatus context causing throwback in (5), and those (Leo Wetzels, pers. comm, and Asaf Bachrach, pers. comm) with prestressing suffixes (or possibly pseudo-suffixes synchronically) in (6):

(5) Antepenultimate stress with high vowel + vowel sequences:3 a. Antônio [ɐ̃.ˈtõ.njo] ‘proper name’ b. pátria [ˈpa.tɾjɐ] ‘homeland’ c. tábua [ˈta.bwɐ] ‘board’ d. Tânia [ˈtɐ̃.njɐ] ‘proper name’ (6) Antepenultimate stress with prestressing sufixes -ico, -ido, -ula:4 a. elétrico [e.ˈlɛ.tɾɪ.kʊ] ‘electric’ b. plastico [ˈplas.tɪ.kʊ] ‘plastic’ c. timido [ˈʧi.mɪ.dʊ] ‘timid’ d. rapido [ˈha.pɪ.dʊ] ‘rapid’ e. flâmula [ˈflɐ̃.mʊ.lɐ] ‘banner’ f. edícula [e.ˈd͡ ʒi.kʊ.lɐ] ‘cottage’

3. In Portuguese orthography, antepenultimate stress is marked by an acute accent generally, but by a circumflex accent when a nasal vowel (i.e. before a nasal consonant). 4. Pseudo-suffixes of this sort may also be at play with anthroponyms such as Jéfferson, Éverton, with antepenultimate stress. 166 CatJL 18, 2019 Andrew Nevins; Paula Pinheiro Costa

Cases like (5) thus form an ‘island of reliability’ among the exceptions (although even the last two forms undergo sporadic glide metathesis, forming ta[w] ba and Ta[j]na as an application of the SWP). The cases in (6) form a set of recur- rent exceptions where high vowels repel stress recurrently with these formatives and may require only listing the (pseudo-suffixes) in question (-imo, ido, -ico, -ula,…). Thus, while antepenultimate stress is lexical and exceptional, cases like (5-6) do seem to reduce the overall unpredictability given smaller pattern-governed cases within the exceptionality. Now, what about secondary stress in BP? In general, secondary stress seems to be assigned separately (and perhaps after) the marking of primary stress. Thus, while there is variation among speakers in terms of whether a given lexical item has penultimate or non-penultimate stress for a handful of forms (e.g. rúbrica [ˈhu.bɾɪ.kɐ] vs. rubrica [hu.ˈbɾi.kɐ] ‘rubric’; projétil [pɾɔ.ˈʒɛ.t͡ ʃiw] vs. projetil [pɾɔ. ʒɛ.ˈt͡ ʃiw] ‘projectile’), and for segmental properties (e.g. assovio [a.soˈvi.ʊ] and assobio [a.soˈbi.ʊ] ‘whistle’ are in free variation), there is no free variation between antepenultimate and final stress known for any form (say ássovio and assovió). Given the lack of much interspeaker variation in main stress, it is notable that there is indeed persistent variation in secondary stress, as pioneeringly discussed by Collischonn (1994); see also Buckley (2016) and Hualde & Nadeu (2011) for related phenomena in Spanish. In English and Dutch, secondary stress can also be diagnosed by a syllable that is not a schwa (e.g. not a reduced vowel). Thus, English differentiates rái.derwith [ə] vs rá.dar, both with the same main stress but different secondary stress. In BP, secondary stress is diagnosable by relative prominence, though the acoustic correlates are complex (see Moraes 2003). Importantly, as shown by Collischonn (1994), there are two strategies for assigning secondary stress, neither of which is apparently weight-sensitive. There is an initial dactyl strategy, found in words with an odd number of syllables before the main stress, and a rhythmic strategy.

(7) Variable Secondary Stress with Odd-Number of Pretonic Syllables: a. maràcujá [ma.ˌɾa.ku.ˈʒa] ~ màracujá [ˌma.ɾa.ku.ˈʒa] ‘passionfruit’ b. abàcaxí [a.ˌba.ka.ˈʃi] ~ àbacaxí [ˌa.ba.ka.ˈʃi] ‘pineapple’ c. Aràcajú [a.ˌɾa.ka.ˈʒu] ~ Àracajú [ˌa.ɾa.ka.ˈʒu] (toponym)

According to Collischonn, these are optional, automatic, and post-lexical, and even affected by late fast speech rules such as glide formation. Thus tetrasyllabic ìn.vi.á.vel [ˌĩ.vi.ˈa.vɛw] ‘unviable’ has initial secondary stress, whereas trisyllabic in.vjá.vel [ĩ.ˈvja.vɛw] does not, and pentasyllabicro.(dò.vi)( á.rja) [hoˌdo.vi.ˈa.ɾjɐ] ‘bus station’ has secondary stress on do, whereas tetrasyllabic (rò.do)(vjá.rja) [ˌho. do.ˈvja.ɾjɐ] has it on ro. Parallel work on Spanish, e.g. by Buckley (2016), has suggested that an initial dactyl is a possible footing strategy to achieve initial prominence, but in fact, considering even longer words opens the possibility that a ternary dactyl is simply a permitted foot-type in the inventory of secondary-stress feet. Thus the heptasyl- Prominence Augmentation via Nasalization in Brazilian Portuguese CatJL 18, 2019 167 labic city (originally multimorphemic in its Tupi-Guaraní source, but not analyzed as such in BP) in São Paulo is (Ì.ta).(qùa.que.ce).(tú.ba), with secondary stress on the first and third syllables, and main stress on the sixth. Interestingly, with even-numbered pretonic syllables, there is no room for such variation, and thus the derived demonym is an Ì.ta.qùa.que.cè.tu.bén.se, with secondary stress on the first, third, and fifth syllables, and main stress on the seventh. Derived demonyms themselves, however, can exhibit medial dactyls, as in A.ràcaju.én.se, in which the second syllable alone has secondary stress, and the main stress falls on the fifth syllable.5 A final factor to consider in secondary stress is its apparent reassignment in cases of stress clash. Like English thirtéén vs thírteen mén or Tennesséé vs Ténnessee Wílliams, BP disallows stress clash caused by adjacent words (or com- pounding-suffixes): café ‘coffee’ càfezínho ‘coffee-dim.’ (and not cafézínho); see Sandalo & Truckenbrodt (2002) for phrasal examples. But is this retraction, or simply erasure of the stress altogether, and reassignment based on Initial Dactyls? There is evidence that it is erasure + reassignment:6

(8) Variable secondary stress following stress-clash reassignment: a. picolé [ˌpi.ko.ˈlɛ] ‘popsicle’ b. picòlezínho [piˌkɔ.lɛˈzĩ.ɲʊ] ‘popsicle-dim.’, rhythmic trochee pattern c. pìcolezínho [ˌpi.kɔ.lɛˈzĩ.ɲʊ] ‘popsicle-dim.’, initial dactyl pattern

With this overview of secondary stress assignment, we can now delimit the set of prominence-demanding positions in BP as those with main stress, those with secondary stress, where secondary stress is assigned by either rhythmic counting leftward from the main stress (always found in cases of an even-number of pretonic syllables), or in case of an odd-number of pretonic syllables, variation between the rhythmic pattern and use of dactyls (either medial or initial). There is, however, one additional position we must add to the inventory of prominence-demanding syllables, going above and beyond those resulting from foot structure, and that is the word-initial syllable itself. The word-initial syllable, we contend, is a crucial site for prominence boosting, even if not directly footed as a locus of secondary stress. Many articulatory studies have found prominence boosting of segments at the edges of initial domains, as distinct from that of stressed syllables; see Fougeron (2001) and Cho & Keating (2009). We appeal to the findings of Smith (2005) that the initial position is one that requires prominence-boosting for the psycholinguistic purposes of lexical access and word-recognition, in addition to potentially for reasons of speech planning by the speaker. For phonologically-based studies of the importance of the initial syllable, see Beckman (1998) and Becker et al. (2018), among many others.

5. Medial dactyls can also be observed as the result of phrasal footing, e.g. Peço vocês encarecida- mente [(ˈpe.so.vo)(ˈses.en.ca)(ˈɾe.ci.da)(ˈmen.te)] ‘Ask.1sg. you.pl warmly’. 6. We thank Leo Wetzels for suggesting this generalization. 168 CatJL 18, 2019 Andrew Nevins; Paula Pinheiro Costa

Summarizing this section overall, we expect prominence-boosting processes to affect (and not affect) the following syllables:

(9a) Syllables potentially demanding prominence-boosting: i. Main stress7 ii. Secondary Stress iii. Initial Syllable

(9b) Syllables never demanding prominence-boosting: i. Non-initial syllables with no secondary stress

Thus, consider the frequently described phenomenon of pretonic mid-vowel lowering in Northeastern dialects of Brazilian Portuguese (as described in oversimplified terms in Nevins (2012), for example). As the careful study of Santana (2019: 204) shows, this mid-vowel lowering does not in fact affect all syllables, but only secondary-stressed or initial ones. Thus, for a word like televisão ‘televi- sion’ with two pretonic mid-vowels, we expect the following possibilities, given the variability in secondary-stress and footing outlined above:

(10) Possible and Impossible Patterns of Prominence Boosting in NE BP: a. Prominence-Boosting on head of initial dactyl and initial syllable: (tɛ̀.le.vi)são b. Prominence-Boosting on head of R>L trochee and initial syllable: tɛ(lɛ̀.vi)são c. Impossible pattern: (9a) with lowering of second, non-head: (tɛ̀.lɛ.vi)são, or (àpɛ).(li.do) ‘nickname’

This concludes the overall outline of the patterns to be described in this paper: we have (a) a set of prominence-boosting positions, (b) a set of prominence-needy syllable types, and (c) a set of processes that seek to boost syllables that fall in both (a) and (b). Thus, in (10), the prominence-boosting positions are those of initial position and/or secondary stress, the prominence-needy syllables are those with lower sonority (e.g. closed mid-vowels), and the process to achieve prominence-boosting is mid-vowel lowering, a process specific to Northeastern BP and outranked by constraints on faithfulness in other dialects of Portuguese.

7. Interestingly, Brazilian Portuguese final syllables headed by underlying /i/ or /u/ typically attract stress, e.g. bambú ‘bamboo’, sushí ‘sushi’. This typologically surprising attraction of stress to low-sonority nuclei may be related to differentiating them from the underlying /e,o/ that post-toni- cally reduce to [ɪ,ʊ] and are never stressed. In fact, words with final /i/ without final stress, e.g. táxi ‘taxi’ may even be analyzed by speakers as with an underlying /e/, as found in the possible vocative form [tak.se]. Whatever the analysis of stress-attracting final /i,u/ may be, it is noteworthy that no examples of spontaneous nasalization in our corpus are found with such vowels. Presumably, in final open syllables, they have enough prominence on their own, without nasalization (though notably, in Carioca dialects they may develop schwa-offglides, e.g. [bɐ̃ˈbuə, suˈʃiə] ‘bamboo, sushi’). Prominence Augmentation via Nasalization in Brazilian Portuguese CatJL 18, 2019 169

A similar pattern to (10) can be found in rhotic metathesis in the Romance languages; thus, there is alternation between crocodillo and cocodrillo ‘crocodile’ showing variation between stressed syllable and initial syllable, whereby medial unstressed syllable is never touched (see Damulakis 2005: 66). This kind of rhotic metathesis is found in non-standard forms in BP (and for coda cases such as 11d, seems to hold independently of whether the rhotic realization is retroflex, velar fricative, or glottal fricative):

(11) Rhotic metathesis to the primary/secondary stressed-syllable a. [ˈvi.dɾʊ] > [ˈvɾi.dʊ] ‘glass’ b. [iʃ.ˈtu.pɾʊ] > [iʃ.ˈtɾu.pʊ] ‘rape’ c. [fits.ˈpɛ.tɾɪ.kɪ] > [fits.ˈpɾɛ.tɪ.kɪ] (‘Fitzpatrick’ proper name) d. [ˌla.gahˈʧi.ʃɐ] > [ˌlah.gaˈʧi.ʃɐ ] ‘lizard’

There is ample documentation of the diachronic phenomenon of rhotacism in Portuguese (e.g. branco ‘white’, compare Spanish blanco and French blanc) which is often colloquially extended to newer words (e.g. Leblon > Lebron ‘toponym’). The reason for such substitution, often not explicitly commented upon, is arguably to improve the sonority profile of the onset, as /r/ is more sonorous than /l/ (and indeed, this feeds into the restriction that [tl, dl] are banned as word-initial onsets while [tr, dr] are allowed; see Moreton 2002 for pertinent discussion from English, and Pons-Moll 2008 for the proposal that taps are more sonorous than laterals in Romance). Rhotacism is so common it often overshadows the arguably distinct, stress-related phenomenon of rhotic metathesis: the migration of the /r/ from a complex onset in an unstressed syllable into that of a stressed syllable. The cases reviewed herein are entirely non-standard and have not been treated in diachronic literature on Portuguese, as they are sporadically found, and unattested in standard variants. Consider moreover the oft-noted example problema > pobrema ‘problem’. While this may have the appearance of ‘rhotacism’, this would not explain the disappearance of the /r/ in the first syllable. Similarly, wholesale liquid metathesis (of the kind found in miraklo > milagro ‘miracle’) is not at stake as the form is not plobrema. Instead, what is happening is that the unstressed syllable is robbed of its rhotic, as an unstressed syllable is not allowed more prominence than a stressed one. Thus, we treat problema > poˈbrema as a case of metathesis to the stressed syllable. Similarly, Lipski (1992a) who collects examples from Spanish such as fábrica > frábica ‘factory’ and dentro > drento ‘inside’, as well as initial-syllable trans- positions like petróleo > pretóleo ‘petroleum’, mentions (p. 100) “This apparent leftward skewing of consonant clusters may have to do with relative prominence, articulatory energy, or the position of the stress peak within the word.” Clàudia Pons-Moll (pers. comm) cites examples from Majorcan Catalan such as padrí > pradí, which depart from the stressed syllable, but crucially move to the initial syllable. While examples of leftward metathesis by only one syllable are found (as metathesis is often sporadic), we contend that, parallel to (10), rhotic metathesis can 170 CatJL 18, 2019 Andrew Nevins; Paula Pinheiro Costa be to a primary/secondarily stressed syllable, or to the initial syllable, but is less likely to migrate to a non-initial, unstressed syllable. Rhotic metathesis is widely discussed for the Iberian languages, but here we limit our attention to BP, where it manifests itself as a prominence-augmentation phenomenon. In a tellingly parallel fashion, Blevins & Garrett (1998) cite the example of Bagnères-de-Luchon, for which “Posttonic r has moved one syllable to the left, into the historically stressed syllable (which is often also the initial syllable).” The analysis they posit is exactly within the spirit adopted here: “In addition, long-distance movement processes evidently move segments or features into syllables which are in some sense more prominent: stressed syllables, as in Colville; or initial syllables, as with Bagneres-de-Luchon French and Ancient Greek. We cannot fully explain these patterns, but they are consistent with the view that the segments and features in Table 1 occupy long durational windows which allow for their reinterpretation in nonhistorical positions. This perceptual reinterpretation involves segments and features which are drawn to positions of prosodic prominence.” (p. 527). Rhotic metathesis is thus already discussed in the literature as a case of prominence-augmentation, and we contend that the nasal vowel metathesis found in (1) above is exactly parallel to such cases. In what follows, we turn to spontaneous nasalization – the introduction of nasal vowels in non-etymological positions, with the same motivation, but in cases for which no existing nasal vowel can be said to have been transposed. Such cases, to our knowledge, are not found with rhotics (e.g. spontaneous insertion of rhotics as the second member of an onset cluster).

3. Spontaneous Nasalization as Prominence-Augmentation We repeat below the prominence scales from (2):

(2) a. diphthongs > simplex nuclei b. complex onsets > onsetful > onsetless c. nasal vowels > oral vowels d. low vowels > lower-mid vowels > upper-mid-vowels > high vowels

Many authors (especially Smith 2005) have proposed that certain phonological processes can be understood in terms of prominence augmentation, especially when initial or secondarily-stressed syllables do not have sufficiently prominent segmental material. Many phenomena, both within standard and non-standard varieties of BP, can be understood within this logic. For example, Wetzels’ (1997) neutralization of mid-vowels to the lower, [-ATR] version in proparoxytones (dactylic lowering) and in penultimate-stressed words that violate the Final-WSP (spondaic lowering) is once such case, motivating the distribution of this vowel in m[ɔ]vel ‘furniture’, ab[ɔ]bora ‘pumpkin’, d[ɔ]lar ‘dollar’, [ɔ]rfã ‘orphan’, F[ɛ]lix ‘proper name’, Pers[ɛ]fone ‘proper name’ and so forth. Once we view vowel height as a kind of sonority that increases the prominence of a syllable, these marked foot-types receive greater sonority in the head position, again using the logic of Coincide as defined in Zoll (1998), as shown in the tableau in (13) for dactylic lowering. Prominence Augmentation via Nasalization in Brazilian Portuguese CatJL 18, 2019 171

(12) Definition of Coincide family of constraints

Coincide (strong position, prominent property) (i) x (x is a strong position y (y=prominent property v Coincide (x,y)) • Coincide (x,y) will be true if y=x, y dominates x, or x dominates y ∀ ∃ (ii) Assess one mark for each value of x for which (i) is false.

We can now apply this to the position Head-Dactyl, as a prominence-augment- ing position. In terms of why the head of a dactyl should be a prominent position, consider the representations in Martínez-Paricio & Kager (2015) and in Hermans & Wetzels (2012), according to which the head of a dactyl, in a recursive binary representation, is the head of a head of a foot, and hence doubly strong.

(13) Tableau for Dactylic Lowering:8

/ per(ˈse.fo.nɪ) / Coincide(Head-Dactyl, Dep [+low] Dep [-ATR] [-ATR]) F [per(ˈsɛ.fo.nɪ)] * [per(ˈse.fo.nɪ)] *! [per(ˈsa.fo.nɪ)] *!

The collection of spontaneous nasalization data in BP we analyze here is based on Costa (2019), and composed of 219 examples in which nasalization occurs in positions that deviate from the etymological and/or standard forms of such words in BP. The data are based on written forms collected from social media and blogs. While one might protest that written forms are not revealing, the fact is that one does not always have a recording device handy while spontaneous nasalization is uttered. Nonetheless, the existence of non-etymological nasalization in BP forms such as indioma ‘language’, ingreja ‘church’, indiota ‘idiot’, indentidade ‘identity’, ingnorante ‘ignorant’ are widely known to all speakers of the language, even if dismissed as mere errors. Souza e Melo (2015) collected written data from ninth- year students in the city of Recife, and reports that she “has observed as a teacher that this kind of nasalization is very widespread among students” (p. 79, translation ours); see also passing mentions in Marroquim (2008: 47).9 Alves (2004), working

8. As mentioned earlier, we include secondary stress (and footing) as the input to the tableau for expositional purposes, remaining agnostic as to whether this is the result of a constraint not shown in this tableau, or the output of an earlier stratum of phonological computation (as secondary-stress has a post-lexical character in BP; Collischonn 1994). See also footnote 12 for suggestive evidence that spontaneous nasalization is a process with a late character, following the determination of rhotic allophony. 9. Such forms are also frequently commented upon in blogs, e.g. (accessed 1 June, 2019). At the moment, we have only conducted a thorough study of this phenomenon in Brazilian varieties of Portuguese. According to Ezra Champal Nhampoca (pers. comm), Mozambican varieties with extensive contact 172 CatJL 18, 2019 Andrew Nevins; Paula Pinheiro Costa on the quilombo Portuguese of Gurutubana, Minas Gerais, notes the recurrence of “spurious nasalization” in word-initial position and with the vowel /i/ (p. 21). Cases of nasal epenthesis with initial onsetless [i] in fact have some diachroni- cally enshrined cases, the most famous being Latin hibernum, which became inverno in Spanish and Portuguese.10 While most of the well-known examples cited above start with a word-initial onsetless high front vowel, this is by far not the only set of such cases (and indeed, it is difficult to claim that these all have negative connotations from a pseudo-prefix in- as this would be unlikely for the cases of ‘church’ and ‘identity’). All written sources were taken from ones with more than 50 non-repeated hits, and the top hits were verified as ones with a clear author who was a speaker of Portuguese. In addition, to make sure that these patterns were not purely due to orthotactic trends (e.g. word-initial sequences), Costa (2019) collected audio samples in individual recordings, in which specific words were targeted by means of elicitation and interviews, largely with interviewees without university-level education, which we contend inhibits this kind of spontaneous nasalization, an otherwise natural and understandable trend given the phonology of the language, but one that is stigmatized when a ‘known’ form exists – unlike the very case with which we began this article. The following is a sample of spontaneous nasalization as found in Costa (2019):11 the full set occurs in the appendix.

(14) Examples of Spontaneous Nasalization via Epenthesis: a. idiota [ˌĩ.d͡ ʒi.ˈɔ.tɐ] ‘idiot’ b. igreja [ĩ.ˈgɾe.ʒɐ] ‘church’ c. ironia [ˌĩ.ɾo.ˈni.ɐ]12 ‘irony’

with Bantu languages such as Changana may exhibit non-etymological nasality, e.g. enconomia ‘economy’. European Portuguese is likely to maintain a different system of nasalization, as apocope can generate the existence of emerging minimal pairs such as com [kõ] ‘with’ vs kone [kõn] ‘(sushi) cone’, potentially yielding a system like contemporary French. 10. Interestingly, hibernar was re-borrowed from Latin with the meaning ‘to hibernate’, and this item has re-undergone spontaneous nasalization in our sample: [ˌĩ.beɦ.ˈnah]. 11. The oft-commented form morta(n)dela [ˌmoh.ta.ˈdɛ.lɐ] ‘mortadela’ has an etymologically nasal source, as it appears as a variant in Italian itself (see , accessed 1 June 2019). 12. Mattoso Câmara Jr. (1970) assumes that nasal vowels pattern like closed syllables with a coda consonant, thereby explaining why the distribution of rhotics in BP – in particular, the ‘strong’, or fricative rhotic (as opposed to the tap) does not occur after a closed syllable, as in Israel ‘Israel’, enredo ‘storyline’, Henrique ‘proper name’, honra ‘honor’, genro ‘son-in-law’. However, our collection of spontaneous nasalization includes [ĩ.ˈɾa.dʊ] ‘irate’ and [ˌũ.ɾu.ˈgwaj] ‘Uruguay’, in which nasalized vowels precede the tap. These could be taken to suggest either that Mattoso Camara’s generalization that nasalized vowels are equivalent to closed syllables is not absolute (perhaps as already challenged by the existence of the proper name of a soccer player Dãr.ley), or that the spontaneous nasalization under scrutiny in this paper is a ‘late’ process, imposed after the determination of rhotic allophony. Of further relevance to the question of timing of rhotic allophony are the facts noted by Leo Wetzels (pers. comm), that while the fricative rhotic is largely not found after diphthongs, the process of l-vocalization can create such sequences for words such as melro ‘blackbird’, guelra ‘gill’, or bilro ‘bobbin’, which yield a diphthong before the fricative rhotic. Prominence Augmentation via Nasalization in Brazilian Portuguese CatJL 18, 2019 173

d. usufruir [ũ.ˌzu.fɾu.ˈih] ‘to make use of’ e. ebuliçao [ĩ.ˌbu.li.ˈsãw̃] ‘boiling’ f. ocorrer [õ.ko.ˈɦeh] ‘occur’ g. abacaxi [ã.ˌba.ka.ˈʃi] ‘pineapple’ h. afligir [ˌa.flĩ.ˈʒih] ‘to ail’ i. aipim [ˌa.ĩ.ˈpĩj̃] ‘yucca’ j. pichaçao [ˌpĩ.ʃa.ˈsãw̃] ‘grafitti’ k. pudim [pũ.ˈdĩj̃] ‘pudding’ l. cobaia [kõ.ˈbaj.jɐ] ‘guinea pig (i.e., test case)’

Ideally, these data points can all be evaluated in a perceptual study – in a same/ different or ABX task, or one in which participants rate the degree of deviation amongst two pronunciations – to see how acceptable and/or perceptually salient these are (and potentially with pseudo-words as well). Of 219 total examples of spontaneous nasalization, 182 of them (83%) occurred in the initial syllable. A breakdown by vowel quality is found below:13

(14) Spontaneous Nasalization targets, by individual vowel Vowel Occurrences % i 175 80% u 18 8% e 10 5% o 11 5% a 5 2% Total 219

As can be verified, there is an extremely high rate of occurrence for [i], followed by [u], though nearly ten times less. While one might generalize specifically to [i] as opposed to high vowels specifically, it is important to recall that in BP, the overall distribution of [i] is much higher than that of [u], and even more pronounced in initial position (where words like umbigo ‘navel’ are often non-standardly pronounced as imbigo). Summing up [i] and [u] with the caveat that [u] is rarer to begin with, high vowels make up 88% of the total targets of spontaneous nasalization. (See Cedergren & Sankoff 1975 for a similar rate of vowel nasalization of high vowels vs non-high vowels in Panamanian Spanish). In fact, Hajek & Maeda (2000) report that “at low levels of velic opening, nasalization is likely to be perceptually more salient in high vowels”. They reanalyse the oft-cited diachronic height hierarchy of nasalization, bringing many counterexamples to the fore, and concluding

13. Onsetless initial /e/ (or at least, etymological/orthographic ) is very frequently reduced to [i], particularly when nasalized or preceding sibilants. 174 CatJL 18, 2019 Andrew Nevins; Paula Pinheiro Costa that the oft-noted effect of low vowels is often brought about by increased duration, and that when duration is controlled, studies such as House & Stevens 1956 find that “synthetic low vowels require almost three times as much velic opening as high vowels before they are identified as nasal by American listeners”. As for the front-back asymmetry found in BP, this is in fact consistent with an assertion made by Ohala (1975), who suggests that given two vowels with similar F1 values, those with higher F2 values are easier to nasalize.14 The epenthesis of nasalization on an initial, onsetless high vowel is shown below, using an adaptation of Coincide specific to this position:15

(15) Tableau for Epenthetic Nasality in Initial, Secondary-Stressed Position

/i.di.ɔ.ta/ Coincide (NasalV) Dep [+nasal] *NasalV F [ˌĩ.d͡ ʒi.ˈɔ.tɐ] * * [ˌi.d͡ʒi.ˈɔ.tɐ] *!

The same process of prominence-boosting occurs below, in which spontaneous nasalization is found in the initial syllable – even if not secondarily-stressed:

(16) Tableau for Epenthetic Nasality in Initial Position

/i.ˈgɾe.ʒɐ/ Coincide (σ1,NasalV) Dep [+nasal] *NasalV F [ĩ.ˈgɾe.ʒɐ] * * [i.ˈgɾe.ʒɐ] *!

As for the high vowel effects, these could be captured by a family of Coincide constraints specific to each vowel height, with Dep(Nasal) interleaved in terms of tolerance, as shown for example in (17):

(17) Height-Specific Rankings for Prominence-Boosting Nasalization:

Coincide (σ1,NasalVhigh) >> Dep [+nasal] >> Coincide (σ1,NasalVmid) >> Coincide(σ1,NasalVlow)

As can be observed in the examples in (14), the majority of these examples involve prominence-boosting on an initial syllable, a secondary-stressed syllable, or a stressed-syllable (and moreover, usually a high vowel, and in fact an onsetless

14. In terms of the following consonant, 41% of the 219 BP cases of spontaneous nasalization occurred before fricatives. As discussed by Ohala & Busà (1995), the airflow required by fricatives causes articulatory opening and subsequent ambiguous perceptual effects that can either lead to nasal deletion before fricatives (e.g. gans > goose in English) or nasal insertion (e.g. bonaça > bonanza in Spanish). 15. In this paper, we adopt the position that nasal vowels involve addition of the feature [+nasal], though this is perfectly compatible with this being the result of addition of a coda nasal that triggers regressive nasalization. Prominence Augmentation via Nasalization in Brazilian Portuguese CatJL 18, 2019 175 high vowel). Lipski (1992b) contains a number of parallel examples from Afro- Hispanic bozal Spanish, as listed below, which he contends represents “the addition of a [+nasal] autosegment to the beginning of certain words” (p. 264).

(19) Spontaneous Nasalization in Bozal Spanish: a. suplica > sumprica ‘begs’ b. ripito > rimpito ‘rhythm’ c. despachar > dimpacha ‘to send’ d. repica > rimpicá ‘to chime’ e. iglesia > ingresia ‘church’ f. ofrece > unfrece ‘offer’

Lipski (1992b: 285) concludes that these “apparent word-internal nasal consonants, added syllable-finally, in reality represented vowel nasalization, possibly combined with low-level epenthesis of an occlusive nasal element before the following consonant”. Let’s now consider the breakdown by stress in BP:

(20) Spontaneous nasalization within sample, breakdown by stress Stress level Occurrences % Primary stress 24 11% Secondary stress 83 38% No stress 112 51% Total 219

As can be seen, primary stressed cases are the least type-frequent (although certain token-frequent cases, such as mendi(n)go [mĩ.ˈd͡ ʒĩ.gʊ] ‘mendicant’ exist.16 Some of these may be chalked up to a tendency for harmony, such as ni(n)cho ‘niche’, which is so commonly used in this form that it may be changing in the language, and related phonologically-triggered instances may be due to aggressive reduplication (in the sense of Zuraw 2002 for non-etymological cases of insertion such as sherbert) with an existing syllable with nasality already present. In order to model the fact that speakers resist spontaneous nasalization on a primary stressed syllable, Ident(±nasal)-σ́ would outrank all of the relevant Coincide constraints. Secondary-stressed ones make up a significant percentage, cases such as ve.i(n) cuˈlar [ve.ˌĩ.ku.ˈlah] ‘to convey’. However, for some of these, given the remarks above that secondary stress is variable, we cannot be fully certain that the version

16. A potential etymology offered for the word gringo ‘foreigner’ for Spanish involves an epenthetic nasal in the word griego ‘Greek’, accompanied by reduction of the diphthong. See Sayers (2009) for extended discussion. 176 CatJL 18, 2019 Andrew Nevins; Paula Pinheiro Costa with spontaneous nasalization indeed was rendered with secondary-stress on the nasalized syllable. Finally, let us consider the completely non-accented ones. Of these, how many are in the initial syllable? 88%, as can be seen below, greatly confirming the tendency already reported and modelled by Coincide (σ1,NasalV).

(21) Spontaneous nasalization in the 112 unstressed syllables, breakdown by syllable position Position Occurrences %

σ1 98 88%

σ2 13 12%

σ3 1 1% Total 112

As this table shows, the overwhelming majority of cases of spontaneous nasalization in the initial syllable occur when it is unstressed. Of the thirteen items with an unstressed non-initial syllable, one is aim.pim ‘yucca’, most likely a case of aggressive reduplication in Zuraw’s sense (e.g. sporadic occurrences of sancrosanct in English, or Ur.ber for ‘uber’ in BP)17, and the other is mortandela ‘mortadela’; see footnote 11 for discussion of this case. There are potentially thus even fewer.

Summing up, therefore, constraints such as Coincide (σ1,NasalV) are intended to guarantee that spontaneous nasalization will be attracted to the initial syllable, and its height-specific family of subconstraints to those with high vowels. The guiding intuition is that nasalization increases prominence, and that the initial syllable, and specifically when it is a shorter-duration high vowel, needs this increased prominence. For durational data on pretonic high vowels vs non-high vowels, see Faveri (1991) and Moraes (1999), both of whom found pretonic high vowels to be up to 40% shorter than the non-high vowels. Indeed, as mentioned in footnote 7, variants of the Carioca dialect of BP specifically enhance stressed high vowels (e.g. aˈm[iə]go ‘friend’, açaˈ[iə] ‘açai’, Banˈg[uə] ‘toponym’). However, why should nasal epenthesis (or metathesis) occur more with onsetless initial syllables? We turn to this last issue in the final section.

4. Nasality as a Contributor of Weight Before proceeding, we wish to include here some novel data from a study of Brazilian Portuguese loanwords as adapted into Maxakalí, an indigenous Macro- Jê language spoken in Minas Gerais, Brazil, as reported in Silva et al. (2019).

17. A related case in an initial syllable is sobrancelha ‘eyebrow’, which has both the variants sombracelha, with metathesis (perhaps affected by the existing lexeme sombra ‘shadow’), and sombrancelha, with epenthesis; the latter case perhaps may be analyzable as aggressive reduplication. Prominence Augmentation via Nasalization in Brazilian Portuguese CatJL 18, 2019 177

(22) Spontaneous Nasalization in Initial Onsetless High Vowels in Maxakalí Loanwords a. açúcar [aˈsukəh] > [aˈtɕoɰ] ‘sugar’ b. espelho [isˈpeʎʊ] > [ɪj̃ ˈpæj]̃ ‘mirror’ c espora [isˈpɔɾə] > [ɪj̃ ˈpʊə̃ ̯ ] ‘spur’ d. Oliveira [oliˈveɾə] > [ũnĩˈbɛə̯ ] (proper name)

While the authors treat this phenomenon as the result of a constraint specific to the native language, *#OralV, the question immediately arises as to why Maxakalí, too, disprefers onsetless oral initial syllables (particularly as the examples above are not found in Pinheiro’s corpus of BP forms with spontaneous nasalization). It is already known that pretonic high vowels are shorter than pretonic non- high vowels (see above; Faveri 1991 and Moraes 1999). It is also known that nasal vowels are longer than their non-nasalized counterparts (Moraes & Wetzels 1992). Finally, onsetful syllables are naturally longer than onsetless syllables, by virtue of having more segmental material. Thus, following the work of Topintzi (2010) and Ryan (2014), let us consider onsets as contributing to syllable weight, and following that of Moraes & Wetzels, let us consider nasal vowels as bearing an additional unit of weight as compared to their oral counterparts. It would be straightforward to have a version of the SWP as follows for a word such as i(n)çar ‘to hoist’ which undergoes spontaneous nasalization. Coincide would require two moras, which could be satisfied either by [+nasal] epenthesis or by onset epenthesis, with the later considered costlier.

(23) Initial Tableau for Nasal Epenthesis as Weight Satisfaction

/i.ˈsah/ Coincide(σ1,2μ) Dep (Cons) Dep(+nasal) F [̩ĩ.ˈsah] * [i.ˈsah] *! [ti.ˈsah] *!

The problem with this style of analysis is that it would mean syllables with an onset consonant and a nasal vowel would have three moras. At this point, we arrive at a theory with potentially more weight distinctions than are required for stress assignment and quantity-sensitivity per se. This immediately brings us to the gradient model of Brazilian Portuguese stress in Garcia (2017), similar to the gradient model of weight presented in Ryan (2014). In such a model, specific gradient values are assigned to onsetful vs onsetless syllables and to those with coda consonants / nasalization. In the absence of specific numerical values at present, we can abstractly formalize a scale as in (24), leading to the revised tableau in (25): 178 CatJL 18, 2019 Andrew Nevins; Paula Pinheiro Costa

(24) Gradient values of weight contribution a. Onsets contribute an amount Wo of syllable weight; Nasalization

contributes an amount WN of syllable weight, where potentially WN ≠WO

b. Vowels bear an inherent amount of weight Wv, a function of their height

c. The total weight of a syllable is thus Wv + WN + WO

c. Let Wv + WN = WT, a threshold for sufficient prominence for initial syllables

(25) Revised Tableau for Nasal Epenthesis as Weight Satisfaction

/i.ˈsah/ Coincide(σ1,WT) Dep (Cons) Dep(+nasal) F [̩ĩ.ˈsah] * [i.ˈsah] *! [ti.ˈsah] *!

More refined models of gradient stress, such as those in Garcia (2017), who adopts a logistic regression, though without specific constraint formulations, would be needed to confirm these with specific durational values, ideally in a manner akin to that developed in Ryan (2014) for gradient onset weight. In closing, we note that although onsets themselves are never inserted wholesale, rhotic metathesis does occur, by hypothesis, to increase the prominence according to the second member of a branching onset. Dep(Cons), therefore, might be more ideally formulated in terms of insertion of a prosodic onset node, and when one already is present, the metathesis of a rhotic to its second position is less intrusive. In a sense, these results support the split margin hypothesis of Baertsch & Davis (2003) that the second position of an onset and the coda have something in common – in BP, both can be mustered to increase syllable weight. Let us consider an amendment of (24) with the split margin hypothesis:

(26) Gradient values of weight contribution with the Split margin hypothesis a. Initial Onsets contribute an amount Wo of syllable weight b. A coda consonant (and nasalization) contributes the same amount of syllable weight as the second member of an onset, both being margins

contributing an amount WM of syllable weight, where potentially WM > WO

b. Vowels bear an inherent amount of weight Wv, a function of their height

c. The total weight of a syllable is thus Wv + WM + WO

c. Let Wv + WM = WT, a threshold for sufficient prominence for initial syllables Prominence Augmentation via Nasalization in Brazilian Portuguese CatJL 18, 2019 179

With the addition of (26b), we may consider one last interesting datum, observed by the authors, by a child acquiring BP who substituted nasalization for a complex onset. The word in question, driˈblar, with a complex onset, was simplified (with rhotacism or rhotic metathesis to the stressed syllable). Removal of the rhotic placed the coronal stop and the high front vowel in contact, feeding affrication (see Cristófaro Silva et al. 2017 for phenomena of this sort in child BP) to [͡ʤi]. Most interesting, however, is the epenthetic nasalization that occurs [ʤĩˈbɾaɦ], by hypothesis as compensation for the loss of the second member of the onset, in order to preserve WT (the constraints compelling rhotic metathesis are not included in this tableau):

(27) Tableau for nasalization as compensatory weight addition in onset simplification

/ dɾiˈblaɦ/ Coincide(σ1,WT) *ComplexOns Dep(+nasal) F [d͡ ʒĩˈbɾah] * * [d͡ ʒĩˈbɾah] *! * [dɾiˈbɾah] **!

While compensatory lengthening (i.e. rime augmentation) under loss of margin /r/ has been explored for Samothraki Greek by Topintzi (2010) and by Cristófaro Silva et al. (2017) for child BP, the case in (27) would be the first to our knowledge of the inverse, namely addition of margin coda (indicated via [+nasal]) to compensate for loss of onset weight. It also raises the question of other sources of coda epenthesis in BP for the sake of achieving Coincide(σ1,WT), and it may be that cases such as ur.ber ‘Uber’ occur instead of epenthetic nasalization specifically because nasal [ũ] is extremely marked in BP, as mentioned above. Thus, summarizing with broad brushes the patterns of prominence augmentation, for the mid- vowels, spondaic lowering to [-ATR] (as in (13)) is available; for /i/, nasalization is preferred (as lowering any further would require more feature changes), and for /u/, where it occurs, rhotic epenthesis may be a strategy.

5. Conclusions and future directions We have suggested that primary – and secondary – stressed syllables (and initial syllables) are like magnets, in drawing more segmental material toward them in order to enrich the asymmetry between them and unstressed syllables. The need for this additional segmental material arises most sharply with the durationally shortest syllables: high vowels, in onsetless syllables. Repair strategies include epenthesis or metathesis of nasality, mid-vowel lowering, and rhotic metathesis, all of which increase syllable prominence (duration, most acutely). Strategies such as glide insertion or raddoppiamento sintattico in Italian may also be seen as prominence- enhancement (and indeed, raise the question of whether non-etymological gemination in Italian forms such as repubblica occur quantitatively more often when the 180 CatJL 18, 2019 Andrew Nevins; Paula Pinheiro Costa stressed nucleus that benefits durationally is a high vowel). Of course, there are cases of spontaneous nasalization outside of high vowels in onsetless initial syllables, but the quantitative predominance of the pattern we have identified is what we aim to capture through the analysis presented above. A full theory of gradient weight must take into account not only the relative contribution – both phonologically and phonetically – of nasalization versus branching onsets (which we have argued may be in a compensatory relation), but also that of vowel height itself, and crucially how these factors all interact. Thus, future studies may benefit from greater exploration of the relatedness of spontaneous nasalization of word-initial vowels with the phenomenon of vowel lowering of word-initial onsetless vowels discussed in Jiménez & Lloret (2013), according to whom Valencian Catalan lowers unstressed but word-initial espina, escala, entendre to aspina ‘thorn’, ascala ‘stairs’, antendre ‘to understand’, respectively. From this perspective, spontaneous nasalization in BP forms part of a much larger pattern of positional markedness within Iberian languages. One of many final questions with which we wish to end is the relationship between vowel lowering (of the mid-vowels, as discussed above for Northeastern dialects) and nasalization. To our knowledge, these processes do not apply jointly. One might contend that either adding nasalization or enacting vowel lowering is enough, and that too many deviations from faithfulness are incurred by epenthesizing both marked features. On the other hand, this particular combination, namely [-ATR] mid vowels that are [+nasal], is generally ruled out in the phonology of Brazilian Portuguese.18 Nonetheless, a complete theory of gradient weight may be able to derive why nasalization or vowel lowering provide just the right amount of additional weight for prominence-boosting, without the need for applying both.

References Alves, Diocles Igor Castro Pires. 2004. O processo de nasalização no dialeto quilombo- la Gurutubano. Masters’ thesis, Pontificia Universidade Catolica de Minas Gerais. Baertsch, Karen & Stuart Davis. 2003. The split margin approach to syllable structure. ZAS Papers in Linguistics 32: 1-14. Becker, Michael, Andrew Nevins, Filomena Sandalo & Érick Rizzato. 2018. The acquisition path of [w]-final plurals in Brazilian Portuguese. Journal of Portuguese Linguistics 17.4: 1-17. Beckman, Jill N. 1998. Positional faithfulness. Doctoral dissertation, University of Massachusetts Amherst.

18. In dialects found within the states of São Paulo and Paraná, however (Maximiliano Guimarães, pers.comm; Filomena Sandalo, pers. comm) there is however an ‘island of reliability’ in exceptionally allowing nasal [-ATR] mid vowels only with the back vowel [ɔ̃] in stressed syllables and only before [m] in words such as fome, nome [fɔ̃mɪ, nɔ̃mɪ] ‘hunger’, ‘name’, and even forming a minimal pair among verbs some [sõmɪ] ‘sum up 3.sg’ vs some [sɔ̃mɪ] ‘disappear (imper.)!’, not distinguished in standard BP dialects. Spontaneous nasalization is not expected here, as these are allophonically nasal in any event; it is only their [-ATR] character that is unexpected. Prominence Augmentation via Nasalization in Brazilian Portuguese CatJL 18, 2019 181

Bisol, Leda. 1992. O acento e o pe metrico binario. Cadernos de Estudos Linguisticos 22: 69-80. Blevins, Juliette & Andrew Garrett. 1998. The origins of consonant-vowel metathesis. Language 74.3: 508-556. Borrelli, Doris. 2002. Raddoppiamento Sintattico in Italian. A Synchronic and Diachronic Cross-Dialectal Study. New York, London: Routledge. Buckley, Eugene. 2016. Foot Alignment in Spanish Secondary Stress. In Jeff Heinz, Rob Goedemans & Harry van der Hulst (eds.). Dimensions of Stress. Cambridge: Cambridge University Press, 79-100. Câmara Jr., J. Mattoso. 1970. Estrutura da língua portuguesa. Petrópolis: Vozes. Cedergren, H. & D. Sankoff. 1975. Nasals: A Sociolinguistic Study of Change in Progress. In Charles A. Ferguson, Larry M. Hyman & John J. Ohala (eds.). NasálFest. Stanford. Cho, T. & Keating, P. 2009. Effects of initial position versus prominence in English. Journal of Phonetics 37.4: 466-485. Collischonn, Gisela. 1994. Acento secundário em português. Letras de Hoje 29.4: 43-53. Costa, Paula Pinheiro. 2019. A nasalização espontânea no português do brasil: uma análise multifatorial. MA thesis, Universidade Federal do Rio de Janeiro. Cristófaro Silva, Thaïs, Maria Cantoni, Nívia Oliveira & Izabel Miranda. 2017. Segmental loss and phonological representation. In Geoff Lindsey & Andrew Nevins (eds.). Sonic Signatures. Amsterdam: John Benjamins, 216-230. Crosswhite, Katherine. 1999. Vowel reduction in Optimality Theory. Doctoral dissertation, UCLA. Damulakis, Gean. 2005. Fenômenos Variáveis sob uma Óptica Formal. MA thesis, Universidade Federal do Rio de Janeiro. Faveri, Claudia Borges de. 1991. Analise da duração das vogais orais do portugues de Florianopolis-Santa Catarina. MA thesis, Universidade Federal de Santa Catarina. Fougeron, C. (2001). Articulatory properties of initial segments in several prosodic constituents in French. Journal of Phonetics 29: 109-135. França, Sebastião Adauto. 2009. O apagamento da vogal postônica não-final por fal- antes de Jaru, Estado de Rondônia. Maringá 31.2: 169-182. Garcia, Guilherme D. 2017. Weight gradience and stress in Portuguese. Phonology 34.1: 41-79. Gouskova, Maria. 2007. The reduplicative template in Tonkawa. Phonology 24: 367-396. Hermans, Ben & Leo Wetzels. 2012. Productive and Unproductive Stress Patterns in Brazilian Portuguese. Letras & Letras 28.1: 77-114. House, A. S. and K. N. Stevens. 1956. Analog Studies of the Nasalization of Vowels. Journal of Speech and Hearing Disorders 21: 218-232. Hualde, J. I. & Nadeu, M. 2011. Rhetorical stress in Spanish. In Word Stress: Theoretical and Typological Issues. Cambridge: Cambridge University Press, 228-252. Jiménez, Jesus & Maria-Rosa Lloret. 2013. Vocalic adjustments under Positional Markedness in Catalan and other Romance languages. In Camacho-Taboada, V., Jiménez-Fernández, Á. L., Martín-González, J. & Reyes-Tejedor, M. (eds.). Information Structure and Agreement. Amsterdam/Philadelphia: John Benjamins, 319-336. 182 CatJL 18, 2019 Andrew Nevins; Paula Pinheiro Costa

Lipski, John M. 1992a. Metathesis as template matching: a Spanish example. Folia Linguistic Historica 11(1-2): 89-104. Lipski, John M. 1992b. Spontaneous nasalization in the development of Afro-Hispanic language. Journal of Pidgin and Creole Languages 7.2: 261-305. Marroquim, Mário. 2008. A língua do nordeste: Alagoas e Pernambuco. EDUFAL. Martinez-Paricio, Violeta & Rene Kager. 2015. The binary-to-ternary rhythmic continuum in stress typology: Layered feet and non-intervention constraints. Phonology 32.3: 459-504. Moraes, Jõao. 1999. Um algoritmo para a correçao/simulaçao da duraçao dos segmentos vocalicos em portugues. In Ester M. Scarpa (ed.). Estudos de Prosodia. Editoria UNICAMP, 69-84. Moraes, Jõao. 2003. Secondary Stress in Brazilian Portuguese: perceptual and acoustical evidence. In Proceedings of the 15th ICPhS, Barcelona. Moraes, Joao & Wetzels, W. Leo. 1992. Sobre a duraço dos segmentos vocalicos nasais e nasalizados em Portugues. Um exercicio de Fonologia Experimental. Cadernos de Estudos Lingüsticos 23: 153-166. Moreton, Elliott. 2002. Structural constraints in the perception of English stop-sonorant clusters. Cognition 84: 55-71. Nevins, Andrew. 2012. Vowel Lenition and Fortition in Brazilian Portuguese. Letras de Hoje 47.3: 228-233. Nevins, Andrew. 2015. Review of Natalie Operstein (2010). Consonant structure and prevocalization. Phonology 32: 307-317. Ohala, John. 1975. Phonetic explanations for nasal sound patterns. In Charles A. Ferguson, Larry M. Hyman & John J. Ohala (eds.). NasálFest. Stanford. Ohala, John & Maria Grazia Busà. 1995. Nasal loss before voiceless fricatives: a perceptually-based sound change. Rivista di Linguistica 7: 125-144. Pons-Moll, Claudia. 2008. Regarding the sonority of liquids. Some evidence from Romance. Paper presented at the 38th Linguistic Symposium on Romance Languages, University of Illinois at Urbana-Champaign. Prince, Alan. 1990. Quantitative consequences of rhythmic organization. Chicago Linguistic Society 26(2): 355-398. Ryan, Kevin M. 2014. Onsets contribute to syllable weight: Statistical evidence from stress and meter. Language 90: 309-341. Sandalo, Filomena & Hubert Truckenbrodt. 2003. Some notes on phonological phrasing in Brazilian Portuguese. Delta 19.1: 1-30. Santana, Arthur. 2019. Neutralização das vogais átonas no português brasileiro. PhD thesis, Universidade de São Paulo. Sayers, William. 2009. An Unnoticed Early Attestation of gringo ‘Foreigner’: Implications for Its Origin. Bulletin of Spanish Studies 86(3): 323-330. Silva, Mário Coelho, Andrew Nevins & James White. 2019. Domains and Prominence in Nasal Harmonization of Maxakalí Loanwords. To appear in International Journal of American Linguistics. Smith, Jennifer L. 2005. Phonological augmentation in prominent positions. Routledge. Souza e Melo, Sandra Maria Mendes. 2015. Processos fonológicos presentes na escrita: um estudo de caso com alunos do 9o ano de uma escola da rede estadual de Recife. Universidade Federal de Pernambuco. Prominence Augmentation via Nasalization in Brazilian Portuguese CatJL 18, 2019 183

Wetzels, W. L. M. 1997. The lexical representation of nasality in Brazilian Portuguese. Probus 9.2: 203-232. Wetzels, W. L. (2007). Primary word stress in Brazilian Portuguese and the weight parameter. Journal of Portuguese Linguistics 5: 9-58. Zoll, Cheryl. 1998. Positional Asymmetries and Licensing. ROA-282. Zuraw, Kie. 2002. Aggressive reduplication. Phonology 19: 395-439.

Appendix: List of Non-Standard forms with Spontaneous Nasalization

Num Standard Form Nasalized form (written) Phonetic Transcription 1 acolchoado aconchoado [a.ˌkõ.ʃu.ˈa.dʊ] padded 2 afligir aflingir [ˌa.flĩ.ˈʒih] to ail 3 aipim aimpim [ˌa.ĩ.ˈpĩ] yucca 4 abacaxi ambacaxi [ã .ˌba.ka.ˈʃɪ] pineapple 5 abundância ambundância [ˌa.bũ.ˈda.sjɐ] abundance 6 aviamento avinhamento [a.ˌvĩ.ɲa.ˈmẽ.tʊ] goodwill 7 bandido bandindo [ba.ˈd͡ ʒĩ.dʊ] bandit 8 bagunça bangunça [ba.ˈgũ.sɐ] mess 9 banqueta banquenta [ba.ˈkẽ.tɐ] footstool 10 banquete banquente [ba.ˈkẽ.t͡ ʃɪ] banquet 11 biblioteca binblioteca [bĩ.ˌbli.o.ˈtɛ.kɐ] library 12 cidadão cindadão [ˌsĩ.da.ˈdaw̃] citizen 13 ciúme cinhume [sĩ.ˈɲu.mɪ] jealousy 14 cobaia combaia [kõ.ˈbaj.jɐ] guinea pig 15 comigo comingo [ku.ˈmĩ.gʊ] with-me 16 colônia conlônia [kõ.ˈlõ.njɐ] colony 17 corromper conrromper [ˌkõ.ɦõ.ˈpeh] to corrupt 18 conselho consenlho [kõ.ˈsẽ.ʎʊ] advice 19 convencido convencindo [ˌkõ.vẽ.ˈsi.dʊ] convinced 20 cozinha conzinha [kõ.ˈzĩ.ɲa] kitchen 21 detrimento dentrimento [ˌdẽ.tɾi.ˈmẽ.tʊ] detriment 22 desviar desvinhar [ˌd͡ ʒiʃ.vĩ.ˈɲah] to divert 23 distinguir dinstinguir [ˌd͡ ʒĩʃ.t͡ ʃĩ.ˈgwih] to distinguish 24 distintivo distintivo [ˌd͡ ʒĩʃ.t͡ ʃĩ.ˈt͡ ʃi.vʊ] distinctive 25 empoderado emponderado [ˌĩ.põ.de.ˈɾa.dɐ] empowered 26 empoderamento emponderamento [ˌĩ.põ.ˌde.ɾa.ˈmẽ.tʊ] empowerment 27 eclipse enclipse [ĩ.ˈkli.pɪ.sɪ] / [ẽ.ˈkli.pɪ.sɪ] eclipse 28 educação enducação [ĩ.ˌdu.ka.ˈsãw̃] education 184 CatJL 18, 2019 Andrew Nevins; Paula Pinheiro Costa

Num Standard Form Nasalized form (written) Phonetic Transcription 29 educador enducador [ĩ.ˌdu.ka.ˈdoh] educator 30 engajado enganjado [ˌĩ.ga.ˈʒa.dʊ] engaged 31 eleger enleger [ˌĩ.le.ˈʒeh] to elect 32 elucidar enlucidar [ĩ.ˌlu.si.ˈdah] to elucidate 33 essencial enssencial [ĩ.ˌsẽ.si.ˈaw] / [ẽ.ˌsẽ.si.ˈaw] essential 34 etiqueta entiqueta [ˌĩ.t͡ ʃi.ˈke.tɐ] label 35 entupido entunpido [ˌĩ.tũ.ˈpi.dʊ] stuffed up 36 feiura feinhura [fẽj.ˈɲu.ɾɐ]̃ ugliness 37 fruta frunta [ˈfɾũ.tɐ] fruit 38 governo governo [gõ.ˈveɦ.nʊ] government 39 hesitei hensitei [ˌĩ.zi.ˈtej] hesitated-1sg 40 hesitou hensitou [ˌĩ.zi.ˈtow] hesitated-3sg 41 hibernar himbernar [ˌĩ.beɦ.ˈnah] to hibernate 42 hidratado hindratado [ˌĩ.dɾa.ˈta.dʊ] hydrated 43 hidratar hindratar [ˌĩ.dɾa.ˈtah] to hydrate 44 higiênico hingienico [ˌĩ.ʒi.ˈẽ.nɪ.kʊ] hygienic 45 hiato hinhato [ĩ.ˈɲa.tʊ] hiatus 46 hipnotizado hinpnotizando [ˌĩ.pi.ˌno.t͡ ʃi.ˈzã .dʊ] hypnotized 47 hipnotizar hinpnotizar [ˌĩ.pi.ˌno.͡ ʃi.ˈzah]t to hypnotize 48 hipócrita hinpócrita [ĩ.ˈpɔ.kɾɪ.tɐ] hypocrite 49 hipoteca hinpoteca [ˌĩ.po.ˈtɛ.kɐ] mortgage 50 história hinstória [ĩʃ.ˈto.ɾjɐ] history 51 histórico hinstórico [ĩʃ.ˈto.ɾɪ.kʊ] record 52 horrível honrrível [õ.ˈhi.vew] / [ũ.ˈhi.vew] horrible 53 humilhação huminlhação [u.ˌmĩ.ʎa.ˈsaw̃] humiliation 54 ideal indeal [ˌĩ.de.ˈaw] ideal 55 ibérico imbérico [ĩ.ˈbɛ.ɾɪ.kʊ] Iberian 56 impressão imprensão [ˌĩ.pɾẽ.ˈsaw̃] impression 57 impressionado imprenssionado [ĩ.ˌpɾẽ.sjo.’na.dʊ] impressed 58 impressionar imprenssionar [ĩ.ˌpɾẽ.sjo.’nah] to impress 59 ibope inbope [ĩ.ˈbɔ.pɪ] Brazilian Institute of Public Opinion and Statistics 60 ebulição inbulição [ĩ.ˌbu.li.ˈsãw̃] / [ẽ.ˌbu.li.ˈsãw̃] boiling 61 içamento inçamento [ˌĩ.sa.ˈmẽ.tʊ] hoisting 62 içar inçar [ĩ.ˈsah] to hoist 63 ícone íncone [ˈĩ.kõ.nɪ] icon Prominence Augmentation via Nasalization in Brazilian Portuguese CatJL 18, 2019 185

Num Standard Form Nasalized form (written) Phonetic Transcription 64 iconografia inconografia [ˌĩ.ko.ˌno.gɾa.ˈfi.ɐ] iconography 65 economia inconomia [ĩ.ˌko.no.ˈmi.ɐ] economy 66 econômico inconomico [ˌĩ.ko.ˈno.mɪ.kʊ] economic 67 icterícia incterícia [ˌĩ.ki.te.ˈɾi.sjɐ] jaundice 68 idade indade [ĩ.ˈda.d͡ ʒɪ] age 69 idealismo indealismo [ĩ.ˌde.a.ˈliʃ.mʊ] idealism 70 ideia indeia [ĩ.ˈdɛj.jɐ] idea 71 idêntico indêntico [ĩ.ˈdẽ.t͡ ʃɪ.kʊ] identical 72 identidade indentidade [ĩ.ˌdẽ.t͡ ʃiˈda.d͡ ʒɪ] identity 73 indigente indingente [ˌĩ.d͡ ʒĩ.ˈʒẽ.t͡ ʃɪ] indigente 74 indignação indinguinação [ˌĩ.d͡ ʒĩ.ˌgi.na.ˈsaw̃] indignation 75 indignado indinguinado [ĩ.ˌd͡ ʒĩ.gi.ˈna.dʊ] indignant 76 idioma indioma [ˌĩ.d͡ ʒi.ˈõ.mɐ] language 77 idiossincrasia indiossincrasia [ˌĩ.djo.ˌsĩ.kɾa.ˈzi.ɐ] idiosyncrasy 78 idiota indiota [ˌĩ.d͡ ʒi.ˈɔ.tɐ] idiot 79 idiotismo indiotismo [ĩ.ˌd͡ ʒi.o.ˈt͡ ʃiʃ.mʊ] idiocy 80 editar inditar [ˌĩ.d͡ ʒi.ˈtah] to edit 81 idolatra indolatra [ˌĩ.doˈla.tɾɐ] idolize-3sg 82 idólatra indólatra [ĩ.ˈdɔ.lɐ.tɾɐ] idolatrous 83 idolatrar indolatrar [ĩ.ˌdo.la.ˈtɾah] to idolize 84 idolatria indolatria [ĩ.ˌdo.la.ˈtɾi.ɐ] idolatry 85 idôneo indoneo [ĩ.ˈdõ.njʊ] legitimate 86 idoso indoso [ĩ.ˈdo.zʊ] elderly 87 efetuar infetuar [ĩ.ˌfe.tu.ˈah] to conduct 88 ignição ingnição [ĩ.ˌgi.ni.ˈsaw̃] ignition 89 ignóbil ingnobil [ˌĩ.gi.ˈnɔ.bɪw] ignoble 90 ignorado ingnorado [ĩ.ˌgi.no.ˈɾa.dʊ] ignored 91 ignorância ingnorancia [ĩ.ˌgi.no.ˈɾa.sjɐ] ignorance 92 ignorante ingnorante [ĩ.ˌgi.no.ˈɾa.t͡ ʃɪ] ignorant 93 ignorar ingnorar [ĩ.ˌgi.no.ˈɾah] to ignore 94 egoísta ingoísta [ˌĩ.go.ˈiʃ.tɐ] selfish 95 igreja ingreja [ĩ.ˈgɾe.ʒɐ] church 96 igual ingual [ĩ.ˈgwaw] equal 97 igualar ingualar [ˌĩ.gwa.ˈlah] to equalize 98 igualdade ingualdade [ˌĩ.gwaw.ˈda.d͡ ʒɪ] equality 99 iguaria inguaria [ˌĩ.gwa.ˈɾi.ɐ] delicacy 186 CatJL 18, 2019 Andrew Nevins; Paula Pinheiro Costa

Num Standard Form Nasalized form (written) Phonetic Transcription 100 iate inhate [ĩ.ˈɲa.t͡ ʃɪ] yacht 101 ilegal inlegal [ˌĩ.le.ˈgaw] illegal 102 ilegítimo inlegítimo [ˌĩ.le.ˈʒi.t͡ ʃɪ.mʊ] illegitimate 103 ilegível inlegivel [ˌĩ.le.ˈʒi.vew] illegibile 104 ileso inleso [ĩ.ˈle.zʊ] unscathed 105 ilícito inlicito [ĩ.ˈli.si.tʊ] illicit 106 ilimitado inlimitado [ĩ.ˌli.mi.ˈta.dʊ] unlimited 107 ilógico inlógico [ĩ.ˈlo.gɪ.kʊ] illogical 108 iludir inludir [ˌĩ.lu.ˈdih] to delude 109 Iluminismo inluminismo [ĩ.ˌlu.mi.ˈniʃ.mʊ] Enlightenment 110 ilusão inlusão [ˌĩ.lu.ˈzaw̃] illusion 111 ilustrar inlustrar [ˌĩ.luʃ.ˈtɾah] to illustrate 112 ilustre inlustre [ĩ.ˈluʃ.tɾɪ] illustrious 113 equivalente inquivalente [ĩ.ˌki.va.ˈlẽ.t͡ ʃɪ] equivalent 114 equívoco inquivoco [ĩ.ˈki.vo.ku] misconception 115 irracional inracional [ĩ.ˌha.sjo.ˈnaw] irrational 116 irradiação inradiação [ĩ.ˌɦa.d͡ ʒja.ˈsãw̃] radiation 117 irado inrado [ĩ.ˈɾa.dʊ] irate 118 irrealizável inrealizável [ˌĩ.ɦe.ˌa.li.ˈza.vew] unattainable 119 irreconciliável inreconciliável [ĩ.ˌɦe.kõ.ˌsi.li.ˈa.vew] unreconcilable 120 irreconhecível inreconhecível [ˌĩ.ɦe.ˌkõ.ɲe.ˈsi.vew] unrecognizable 121 irrecuperável inrecuperável [ˌĩ.ɦe.ˌku.pe.ˈɾa.vew] unrecoverable 122 irrecusável inrecusável [ĩ.ˌɦe.ku.ˈza.vew] undeniable 123 ironia inronia [ˌĩ.ɾo.ˈni.ɐ] irony 124 ironizar inronizar [ĩ.ˈɾo.ni.ˈzah] to mock 125 irreal inrreal [ˌĩ. ɦe.ˈaw] unreal 126 irredutível inrredutível [ĩ.ˌɦe.du.ˈt͡ ʃi.vew] relentless 127 irrefletido inrrefletido [ĩ.ˌɦe.fle.ˈt͡ ʃi.dʊ] thoughtless 128 irrefutável inrrefutável [ĩ.ˌɦe.fu.ˈta.vew] irrefutable 129 irregularidade inrregularidade [ĩ.ˌhe.gu.ˌla.ɾi.ˈda.d͡ ʒɪ] irregularity 130 irrelevante inrrelevante [ĩ.ˌɦe.le.ˈvã .t͡ ʃɪ] irrelevant 131 irreligioso inrreligioso [ˌĩ.ɦe.ˌli.ʒi.ˈo.zʊ] unreligious 132 irremediável inrremediável [ˌĩ.ɦe.ˌme.d͡ ʒi.ˈa.vew] irretrievable 133 irremovível inrremovível [ĩ.ˌɦe.mo.ˈvi.vew] unremovable 134 irreparável inrreparável [ĩ.ˌɦe.pa.ˈɾa.vew] unrepairable 135 irresponsável inrreponsável [ĩ.ˌɦejʃ.põ.ˈsa.vew] irresponsible Prominence Augmentation via Nasalization in Brazilian Portuguese CatJL 18, 2019 187

Num Standard Form Nasalized form (written) Phonetic Transcription 136 irrepreensível inrrepreensível [ˌĩ.ɦe.ˌpɾe.ẽ.ˈsi.vew] irreproachable 137 irresistível inrresistível [ĩ.ˌɦe.ziʃ.ˈt͡ ʃi.vew] irresistible 138 irrestrito inrrestrito [ˌĩ.ɦejʃ.ˈtɾi.tʊ] unrestricted 139 irreverente inrreverente [ĩ.ˌɦe.ve.ˈɾẽ.t͡ ʃɪ] irreverent 140 irrigar inrrigar [ˌĩ.ɦi.ˈgah] to irrigate 141 irritação inrritação [ĩ.ˌɦi.ta.ˈsãw̃] irritation 142 irritar inrritar [ˌĩ.ɦi.ˈtɐh] to irritate 143 isenção insenção [ˌĩ.zẽ.ˈsaw̃] exemption 144 isento insento [ĩ.ˈzẽ.tʊ] exempt 145 esgrima insgrima [ĩʃ.ˈgɾi.mɐ] fencing 146 isolado insolado [ˌĩ.zo.ˈla.dʊ] isolated 147 isolar insolar [ˌĩ.zo.ˈlah] to isolate 148 isopor insopor [ˌĩ.zo.ˈpoh] styrofoam 149 específico inspecífico [ˌĩʃ.pe.ˈsi.fɪ.kʊ] specific 150 expectativa inspectativa [ˌĩʃ.peˌki.ta.ˈti.vɐ] expectation 151 especulou inspeculou [ĩʃ.ˌpe.ku.ˈlow] speculated-3sg 152 esplêndido insplendido [ĩʃ.ˈplẽ.d͡ ʒɪ.dʊ] splendid 153 esplendor insplendor [ˌĩʃ.plẽ.ˈdoh] splendor 154 espontâneo inspontâneo [ˌĩʃ.põ.ˈta.nɪw] spontaneous 155 isqueiro insqueiro [ĩʃ.ˈke.rʊ] lighter 156 esquema insquema [ĩʃ.ˈke.mɐ] scheme 157 estagnado instagnado [ĩʃ.ˌta.gi.ˈna.dʊ] stagnant 158 histérica instérica [ĩʃ.ˈtɛ.ɾɪ.kɐ] hysterical 159 estima instima [ĩʃ.ˈt͡ ʃi.mɐ] esteem 160 estimulo instimulo [ˌĩʃ.t͡ ʃi.ˈmu.lʊ] stimulate-1sg 161 estímulo instímulo [ĩʃ.ˈt͡ ʃi.mʊ.lʊ] stimulus 162 extraordinário instraordinário [ˌĩʃ.tɾa.ˌoh.d͡ ʒi.ˈna.rɪw] extraordinary 163 estúpido instúpido [ĩʃ.ˈtu.pi.dʊ] stupid 164 ítem intem [ˈĩ.tẽj]̃ item 165 interromper intenrromper [ĩ.ˌtẽ.hõ.ˈpeh] to interrupt 166 itinerante intinerante [ĩ.ˌt͡ ʃi.ne.ˈɾa.t͡ ʃɪ] itinerant 167 itinerário intinerário [ĩ.ˌt͡ ʃi.ne.ˈɾa.ɾɪw] itinerary 168 exame inzame [ĩ.ˈza.mɪ] exam 169 eximir inzimir [ˌĩ.zi.ˈmih] to shirk 170 irregular inrregular [ĩ.ˌhe.gu.ˈlah] irregularity 171 jejum jenjum [ʒẽˈʒũ] fasting 188 CatJL 18, 2019 Andrew Nevins; Paula Pinheiro Costa

Num Standard Form Nasalized form (written) Phonetic Transcription 172 joelho jonhelho [jũ.ˈɲe.ʎu] knee 173 liberdade linberdade [ˌlĩ.beɦ.ˈda.d͡ ʒɪ] liberty 174 livro linvro [ˈlĩ.vɾʊ] book 175 mandioca mandinhoca [ˌma.d͡ ʒĩ.ˈɲɔ.kɐ] manioc 176 maracujá maracunjá [ma.ˌɾa.kũ.ˈʒa] passionfruit 177 mendigo mendingo [mĩ.ˈd͡ ʒĩ.gʊ] mendicant 178 melancolia menlancolia [mẽ.ˌla.ko.ˈli.ɐ] melancholy 179 misturar minsturar [mĩʃ.ˈtu.ɾah] to mix 180 mortadela mortandela [ˌmoh.ta.ˈdɛ.lɐ] mortadela 181 necessidade nencessidade [nẽ.ˌse.si.ˈda.d͡ ʒɪ] necessity 182 nicho nincho [ˈnĩ.ʃʊ] niche 183 nítido níntido [ˈnĩ.t͡ ʃɪ.dʊ] clear 184 obrigado obringado [ˌo.bɾĩ.ˈga.dʊ] thank you 185 ocorreu oncorreu [ˌõ.ko.ˈɦew] occurred-3sg 186 pentecostal pentencostal [pẽ.ˌtẽ.koʃ.ˈtaw] pentecostal 187 pentecostes pentencostes [ˌpẽ.tẽ.ˈkoʃ.t͡ ʃɪʃ] Pentecost 188 picadeiro pincadeiro [ˌpĩ.ka.ˈde.ɾʊ] arena 189 pichação pinchação [ˌpĩ.ʃa.ˈsãw̃] grafitti 190 picolé pincolé [ˌpĩ.ko.ˈlɛ] popsicle 191 piolho pinholho [pĩ.ˈɲo.ʎʊ] louse 192 possibilidade possinbilidade [ˌpo.sĩ.ˌbi.li.ˈda.d͡ ʒɪ] possibility 193 pudim pundim [pũˈd͡ ʒĩ] pudding 194 reiterar reinterar [he.ˌĩ.te.ˈɾah] to reiterate 195 reivindicar reinvindicar [hẽj.ˌṽ ĩ.d͡ ʒi.ˈkah] to claim 196 restrição restrinção (ões) [ˌhejʃ.tɾĩ.ˈsaw̃] restriction 197 sanduíche sanduinche [ˌsa.du.ˈĩ.ʃɪ] sandwich 198 seguro segunro [si.ˈgũ.ɾʊ] secure 199 sibilante simbilante [ˌsĩ.bi.ˈla.t͡ ʃɪ] sibilante 200 sigilo singilo [sĩ.ˈʒi.lʊ] confidentiality 201 sigiloso singiloso [ˌsĩ.ʒi.ˈlo.zʊ] confidential 202 sobrancelha sombracelha [ˌsõ.bɾa.ˈse.ʎɐ] eyebrow 203 sobrancelha sombrancelha [ˌsõ.bɾã.ˈse.ʎɐ] eyebrow 204 sujeito sunjeito [sũ.ˈʒej.tʊ] subject 205 tranquilo tranquinlo [tɾa.ˈqwĩ.lʊ] tranquil 206 trufado trunfado [tɾũ.ˈfa.dʊ] truffled 207 uruguai unruguai [ˌũ.ɾu.ˈgwaj] Uruguay Prominence Augmentation via Nasalization in Brazilian Portuguese CatJL 18, 2019 189

Num Standard Form Nasalized form (written) Phonetic Transcription 208 usufruir unsufruir [ũ.ˌzu.fɾu.ˈih] to make use of 209 útero untero [ˈũ.te.ɾʊ] uterus 210 utilizado untilizado(a) [ũ.ˌt͡ ʃi.li.ˈza.dʊ] used 211 utilizar untilizar [ũ.ˌt͡ ʃi.li.ˈzah] to use 212 utopia untopia [ˌũ.to.ˈpi.ɐ] utopia 213 veicular veincular [ve.ˌĩ.ku.ˈlah] to convey 214 ventilador ventinlador [vẽ.ˌt͡ ʃĩ.la.ˈdo] fan 215 Vicente vincente [vĩ.ˈsẽ.t͡ ʃɪ] proper name 216 vídeo vindeo [ˈvĩ.d͡ ʒjʊ] video 217 vigência vingência [vĩ.ˈʒẽ.sjɐ] validity 218 viado vinhado [vĩ.ˈɲa.dʊ] deer 219 viagem vinhagem [vĩ.ˈɲa.ʒẽj]̃ voyage

Catalan Journal of Linguistics 18, 2019 191-215

Verb-Clitic Structures in Eivissan Catalan: Recursive Prosodic Words and Allomorphy*

Francesc Torres-Tamarit SFL, Université Paris 8, CNRS [email protected] Eulàlia Bonet Universitat Autònoma de Barcelona [email protected]

Received: December 10, 2018 Accepted: May 17, 2019

Abstract

This paper deals with stress shift in verb-clitic structures in Eivissan Catalan, an understudied Romance variety. Within Balearic Catalan, this is the only subdialect in which stress shift is restricted to apply only in second conjugation pre-clitic infinitives, those that, as opposed to other conjugations, have penultimate stress when they are pronounced in isolation. Stress in second conjugation infinitives in Eivissan Catalan shifts one syllable to the right, that is, to the final syllable of the verbal stem, when one or more pronominal enclitics follow. There is no stress shift in pre-clitic imperatives. We claim that pronominal enclitics in Eivissan Catalan adjoin to a recursive, maximal prosodic word, and that the domain for stress assignment is the minimal, embedded prosodic word. We further analyze two cases of stress-conditioned allomorphy (i.e. allomorphy of the infinitive morph and allomorphy of the verbal root) that occur in infinitive-clitic structures. Keywords: allomorphy; clitic; Eivissan Catalan; prosodic word; recursivity; stress shift

Resum. Les estructures verb-clític en eivissenc: mots prosòdics recursius i al·lomorfia

Aquest treball investiga el desplaçament accentual en estructures verb-clític en eivissenc, una varietat romànica poc estudiada. Dintre del balear, aquest és l’únic subdialecte en el qual el des- plaçament accentual només afecta infinitius de la segona conjugació seguits de clític, és a dir, els que, en contrast amb altres conjugacions, tenen accent a la penúltima síl·laba quan no van seguits de clític. En eivissenc, quan el verb va seguit d’un o més enclítics pronominals, l’accent dels infinitius de la segona conjugació es desplaça una síl·laba cap a la dreta, que és l’última síl·laba del radical verbal; no hi ha desplaçament quan els clítics van precedits d’un imperatiu. Proposem

* This work received support from grant FFI2016-76245-C3-1-P (Agencia Estatal de Investigación, AEI, and Fondo Europeo de Desarrollo Regional, FEDER), and, for Bonet, from Generalitat de Catalunya, grant 2017 SGR 634. We want to thank Donca Steriade and Joan Mascaró for discussion of the data and possible analyses, as well as two anonymous reviewers for their comments on the first version of this paper.

ISSN 1695-6885 (in press); 2014-9718 (online) https://doi.org/10.5565/rev/catjl.253 192 CatJL 18, 2019 Francesc Torres-Tamarit; Eulàlia Bonet que els enclítics pronominals de l’eivissenc s’adjunten a un mot prosòdic recursiu màxim i que el domini d’assignació d’accent és el mot prosòdic mínim, subordinat. També analitzem dos casos d’al·lomorfia motivada per l’accent, que es troba en les estructures infinitiu-clític, al·lomorfia que afecta el morf d’infinitiu i al·lomorfia que afecta l’arrel verbal. Paraules clau: al·lomorfia; clític; eivissenc; mot prosòdic; recursivitat; desplaçament accentual

Table of Contents 1. Introduction 5. Conclusions 2. Data References 3. Analysis 4. Allomorphy in infinitive-clitic structures in Eivissan Catalan

1. Introduction The domain for stress assignment generally corresponds to lexical words, that is, roots plus their affixes, which in prosodic phonology are said to map onto prosodic words. Morphosyntactic constituents larger than one lexical word do not usually correlate with prosodic domains for stress assignment (but see Kaisse 2017 for cases in which stress assignment can span over two lexical words when both occur as frequent collocations in Macedonian). One exception to this generalization is the behavior of clitics, a phonological term referring to function words that depend prosodically on lexical words, their prosodic hosts. In such cases, the domain for stress assignment can be larger than the strict lexical word as it includes clitics. Selkirk (1995) and Peperkamp (1997) (and later Itô & Mester 2009) have shown that clitics prosodify with their hosts in a variety of ways that generate a continuum that ranges from more cohesive to less cohesive host-clitic structures. The ways in which clitics prosodify with their hosts has an impact on stress assignment, and studying the stress patterns of host-clitic structures therefore contributes to a better understanding of the general principles governing stress assignment, particularly of the prosodic domains in which it applies. Pronominal clitics in Romance offer a rich empirical domain for studying stress assignment in structures containing enclitics (see Repetti 2016 for an overview and for previous works cited therein). The goal of this paper is to investigate the prosodic and morphological properties of verb-clitic structures in Eivissan Catalan, which has received no attention in the literature so far. In Catalan, pronominal clitics are encliticized to imperatives and infinitives (and also gerunds). Eivissan Catalan is of interest here because it exhibits a highly restrictive case of enclitic-induced stress shift that has not been attested before. In Eivissan Catalan, stress shift is restricted to occur only in penultimately-stressed second conjugation infinitives when followed by enclitics (e.g. [kunəˈʃəɾ=lu] ‘to know him’ cf. [kuˈnəʃə] ‘to know’), in which stress shifts one syllable to the right. Verb-Clitic Structures in Eivissan Catalan CatJL 18, 2019 193

When second conjugation infinitives are pronounced in isolation the locus of stress is the penultimate syllable of the verb, while with enclisis clitics remain unstressed but the stress shifts to the last syllable of the verb, even in the presence of enclitic clusters in which there would be enough phonological material to build a metrical foot on top of the clitics (e.g. [məɾəˈʃəɾ=tə=ləs] ‘to deserve them yourself’ cf. [məˈɾəʃə] ‘to deserve’). The pattern of stress shift in Eivissan Catalan deviates from the patterns of stress shift found in all the other subdialects of Balearic Catalan (Torres-Tamarit & Pons-Moll 2019), and is therefore valuable for establishing a more solid typology of pronominal encliticization in Romance. The main claim of this paper is that prosodic categories such as the prosodic word can be hierarchically self-embedded (see Bennett 2018 and works cited therein for arguments in favor of recursive prosodic categories), and that enclitics in Eivissan Catalan adjoin to a recursive prosodic word that lies outside the domain of stress assignment, which is restricted to minimal prosodic words. A process of word-final /ɾ/ deletion that affects infinitives will also be discussed that gives support to an analysis of stress shift based on recursive prosodic words and that goes against positing additional prosodic categories such as the clitic group (Vogel 2009) or the prosodic word group (Vigário 2010). Second conjugation infinitives with enclitics in Eivissan Catalan further exhibit an interesting case of allomorph selection of one of the two available infinitive morphs, as well as a case of verbal root allomorphy. Allomorph selection in infinitives followed by enclitics has not been reported before and constitutes an interesting case of phonologically-conditioned allomorphy (see Bonet & Torres-Tamarit 2010, 2011 for cases of allomorphy in imperatives followed by enclitics in Balearic Catalan). The remainder of this paper is organized as follows. In section 2 we describe the stress patterns found in verb-clitic structures in Central Catalan and Balearic Catalan: Majorcan, Minorcan, Formenteran, and Eivissan. In section 3 we briefly review Itô & Mester’s (2009) previous approach to the prosodification of clitics, and provide an OT analysis of verb-clitic structures in Eivissan Catalan. Section 4 describes the allomorphic alternations found in infinitive-clitic structures with respect to isolated infinitives in Eivissan Catalan and proposes an OT analysis of the data based on the lexical ordering of second conjugation infinitive allomorphs (Mascaró 2007). Section 5 concludes the paper.

2. Data 2.1. Stress in verb-clitic structures in Catalan Balearic Catalan is a dialect of Catalan that is further subdivided into Formenteran, Eivissan, Majorcan and Minorcan Catalan (each island of the Balearic archipela- go having its own subdialect). Whereas Majorcan, Minorcan and, more recently, Formenteran Catalan have received some attention in the literature (see Torres- Tamarit & Pons-Moll 2019 and previous references cited therein), the phonology of pronominal enclitics in Eivissan Catalan is understudied. New data on Eivissan Catalan are presented in this paper (see also Marí Tur 2017 and Ribas Tur 2018). From a theoretical point of view, these data give new evidence in favor of pho- 194 CatJL 18, 2019 Francesc Torres-Tamarit; Eulàlia Bonet nologically optimizing stress shift in verb-clitic structures, as well as recursive prosodic words. In Majorcan and Minorcan Catalan, stress always shifts to the last syllable of the whole verb-clitic structure. Therefore, the location of stress in a verb-clitic structure always differs from the location of stress when the verb is pronounced in isolation.1 Stress shifts irrespective of the conjugational class of the verb, the morphosyntactic features of the clitic(s), whether the verb is an imperative or an infinitive, and whether there is one or more clitics. Some examples of stress shift in Majorcan Catalan appear in (1) with one enclitic in (1a,b) and two enclitics in (1c). Vowel reduction to schwa in the verb further demonstrates that stress shifts in forms with enclisis (see also Nadeu et al. 2016). The same stress patterns are found in Minorcan Catalan, not exemplified in (1).

(1) Stress shift in Majorcan Catalan a. imp.2sg imp.2sg=3sg.f.acc [ˈkantə] [kəntə=ˈlə] ‘sing!’ ‘sing it!’

b. imp.2pl imp.2pl=3sg.dat [kənˈtaw] [kəntəw=ˈli] ‘sing!’ ‘sing to him/her!’

c. inf. inf=3sg.f.acc=1sg [kənˈta] [kəntəl=lə=ˈmə] ‘to sing’ ‘to sing it to me’

In Formenteran Catalan, by contrast, stress is not shifted to one and the same position. Instead, stress shifts either to the last syllable of the whole verb-clitic structure, or to the penultimate syllable. When the whole sequence ends in a consonant (except for the plural morph [-s]), stress shifts to the final syllable of the verb-clitic sequence. If it ends in a vowel, stress shifts to the penultimate syllable. These data support an analysis of enclitic-induced stress shift as a metrical optimizing strategy. In Formenteran Catalan, stress shifts in order to obtain a right-aligned moraic trochee (i.e. a left-headed, binary foot at the level of the mora). In second person plural imperatives, (2b), there is no stress shift if one enclitic is added because the last syllable of the verbal stem is already a heavy syllable and can therefore be parsed as a moraic trochee at the expense of right-alignment. Stress shift in Formenteran Catalan is illustrated in (2) with some examples.

1. There are some exceptions in Minorcan and certain subvarieties of Majorcan, in which stress shifts to the penultimate syllable of the whole sequence in some contexts due to output-output faithfulness (see Torres-Tamarit & Pons-Moll 2019). Verb-Clitic Structures in Eivissan Catalan CatJL 18, 2019 195

(2) Stress shift in Formenteran Catalan a. imp.2sg imp.2sg=3sg.f.acc [ˈkantə] [kənˈtə=lə] ‘sing!’ ‘sing it.fem!’

b. imp.2pl imp.2pl=3sg.dat [kənˈtaw] [kənˈtaw=li] ‘sing!’ ‘sing to him/her!’

c. inf. inf=1sg =3sg.f.acc [kənˈta] [kəntəɾ=ˈmə=lə] ‘to sing’ ‘to sing it to me’

Central Catalan differs from these two dialects in not having stress shift in verbs with enclitics, as illustrated in (3).

(3) Absence of stress shift in Central Catalan a. imp.2sg imp.2sg=3sg.f.acc [ˈkantə] [ˈkantə=lə] ‘sing!’ ‘sing it.fem!’

b. imp.2pl imp.2pl=3sg.dat [kənˈtɛw] [kənˈtɛw=li] ‘sing!’ ‘sing to him/her!’

c. inf. inf=1sg =3sg.f.acc [kənˈta] [kənˈtar=mə=lə] ‘to sing’ ‘to sing it to me’

2.2. Stress in verb-clitic structures in Eivissan Catalan Eivissan Catalan (henceforth EC) stands apart from the rest of Balearic subdialects and Central Catalan. First of all, there is no stress shift in imperatives with enclitics. Second, stress shift is only found in penultimately-stressed second conjugation infinitives when followed by one or more pronominal enclitics in the subvariety of EC of interest here. This is not a coincidence: first and third conjugation infinitives in Catalan have final stress (e.g. [kənˈt-a] ‘to sing’, underlyingly /kant-a-ɾ/; [kuˈz-i] ‘to sew’, underlyingly /kuz-i-ɾ/), with stress falling on the thematic vowel; but most second conjugation infinitives have penultimate stress (e.g. [ˈtem-ə] ‘to fear’, [ˈromp-ɾə] ‘to break’). Very few second conjugation infinitives have final stress in Catalan (e.g. haver ‘to have (aux.)’, saber ‘to know’, poder ‘to be able to’, soler ‘to usually do’, valer ‘to be worth’, voler ‘to want’). These second conjugation infinitives with final stress are not affected by stress shift. Third, stress is only shifted one syllable to the right, that is, to the last syllable of the verbal stem, and therefore never falls on any enclitic. Before presenting the analysis of the stress pattern in infinitives with enclisis in EC, we need an excursus into the morphology of second conjugation infini- 196 CatJL 18, 2019 Francesc Torres-Tamarit; Eulàlia Bonet tives. Most second conjugation infinitives in Catalan have two surface endings: [-ə(ɾ)] and [-ɾə], with the final [ɾ] in [-ə(ɾ)] being deleted word-finally if no enclitic follows due to a process of deletion of rhotics in word-final position in Catalan (e.g. [kuˈnəʃ-ə], [kunəˈʃ-əɾ=u]). According to Mascaró (1985), second conjugation infinitives have no thematic vowel. This is why stress falls on the verbal root in these second conjugation verbs. First and third conjugation verbs, as well as second conjugation verbs like saber ‘to know’, do have a thematic vowel, which is stressed in infinitives. For EC, we assume that the infinitive morpheme has two allomorphs: /-ɾ/ and /-əɾ/. If the last consonant of the verbal root is a sonorant (e.g. /m-, ɲ-, r-/) or a sibilant (e.g. /s-, ʃ-/), the allomorph /-əɾ/ is selected, which is pronounced [-ə] due to the process of word-final deletion of rhotics. If the last consonant of the verbal root is a stop (e.g. /p-, b-, t-, d-/) or the labiovelar glide (/w-/), the infinitive takes /ɾ/, which is pronounced [-ɾə] due to the insertion of a final epenthetic schwa, required for syllabification, as assumed in Mascaró (1985) and subsequent work on Catalan morphophonology. Selecting /-ɾ/ is possible because stops (spirantized when voiced) can create a well-formed complex onset with /-ɾ/ (e.g. [.pɾ], [.βɾ], [.tɾ], [.ðɾ]), and the heterosyllabic contact [w.ɾ] is not rising. A short list of second conjugation verbs in EC is given in (4).

(4) Second conjugation infinitives in Eivissan Catalan (adapted from Mascaró 1985: 112) inf. in [-ə] after sonorants or sibilants inf. in [-ɾə] after stops and /w-/ [ˈtem-ə] ‘to fear’ [ˈromp-ɾə] ‘to break’ [ˈpɾem-ə] ‘to press’ [ˈbat-ɾə] ‘to shake’ [ˈplaɲ-ə] ‘to feel sorry’ [ˈpɛɾð-ɾə] ‘to lose’ [əmˈpəɲ-ə] ‘to push’ [ˈbəw-ɾə] ‘to drink’ [ˈvəns-ə] ‘to defeat’ [ˈkaw-ɾə] ‘to fall’ [ˈtɔɾs-ə] ‘to twist’ [ˈrəβ-ɾə] ‘to receive’ [ˈkɾeʃ-ə] ‘to grow’ [kuˈnəʃ-ə] ‘to know’ [məˈɾəʃ-ə] ‘to deserve’ [ˈkor-ə] ‘to run’

There is still another type of verb within the second conjugation class whose verbal root ends in either /-n/ or /-l/. These verbs would be expected to select the infinitive allomorph /-əɾ/, but take instead the allomorph /-ɾ/, and avoid the heterosyllabic rising consonant cluster that would be created by the concatenation of the alveolar nasal or lateral with the rhotic through epenthesis of [-d-]. Some examples appear in (5). Verb-Clitic Structures in Eivissan Catalan CatJL 18, 2019 197

(5) Second conjugation infinitives in [-d-ɾə] in Eivissan Catalan (adapted from Mascaró 1985: 112) [ənˈsen-d-ɾə] ‘to turn on’ [ˈpɾən-d-ɾə] ‘to take’ [ˈvən-d-ɾə] ‘to sell’ [diˈsɔl-d-ɾə] ‘to dissolve’ [ˈmɔl-d-ɾə] ‘to grind’ [rəˈzɔl-d-ɾə] ‘to solve’

We will come back to these verbs in section 4. As said before, stress shifts in EC when penultimately-stressed second conjugation infinitives are followed by pronominal enclitics. For now, we exemplify stress shift with those infinitives that end in [-ə(ɾ)]. No matter the number of clitics that follow the verb, stress in EC shifts only one syllable to the right, that is, to the last syllable of the verbal stem, leaving all enclitics unstressed. This is exemplified in (6). Note that the final rhotic in /-əɾ/ surfaces when an enclitic follows as it is not final anymore.

(6) Stress shift in penultimately-stressed second conjugation verbs in Eivissan Catalan

a. inf inf=2sg [kuˈnəʃ-ə] [kunəˈʃ-əɾ=tə] ‘to know’ ‘to know you’

b. inf inf=1sg=3sg.f.acc [məˈɾəʃ-ə] [məɾəˈʃ-əɾ=mə=lə] ‘to deserve’ ‘to deserve it myself’

c. inf inf=loc [ˈkor-ə] [kuˈr-əɾ=i] ‘to run’ ‘to run there’

In the next section we briefly introduce the representations and the constraint set used in Itô & Mester (2009) to account for the prosodification of clitics. Building on their proposal, we will present an OT analysis of the EC data in section 3.3.

3. Analysis 3.1. The prosodification of clitics The basic tenets of Prosodic Phonology (Selkirk 1981; Nespor & Vogel 1986) are that phonological constituent structure is not necessarily isomorphic with syntactic constituent structure, and that phonological constituent structure is organized along a hierarchy of a finite set of universal prosodic categories referred to as the prosodic hierarchy (i.e. the syllable < the foot < the phonological word < the phonological phrase < the intonational phrase < the utterance). A considerable 198 CatJL 18, 2019 Francesc Torres-Tamarit; Eulàlia Bonet body of work within OT has been devoted to the prosodification of function words since the pioneering work of Selkirk (1995). Building up on this work, Peperkamp (1997) established a typology of the prosodification of pronominal enclitics in Italo- Romance (see also Anderson 2011 for an overview). Later on, Itô & Mester (2009), following Selkirk (1995), elaborated on the notion of recursive prosodic words in their analysis of function-lexical structures in German and English. In this paper we assume the representational principles and the constraint set of Itô & Mester (2009). In Itô & Mester (2009), four possible sites are identified for function words in a prosodic tree structure without any intervening prosodic categories between the prosodic word and the phonological phrase (as in Selkirk 1995 and Peperkamp 1997). These four sites are illustrated in (7). In (7a), the function word projects its own prosodic word and is therefore a prosodically independent form with respect to the lexical word. Both prosodic words are further parsed into a single phonological phrase. In (7b), the function word amalgamates with the prosodic word that contains the lexical host (i.e. an internal clitic in Selkirk’s terminology). In (7c), the function word is adjoined to a recursive prosodic word that dominates the inner prosodic word containing the host (i.e. an affixal clitic in Selkirk’s terminology). Finally, in (7d) the function word is directly attached to the phonological phrase (i.e. a free clitic in Selkirk’s terminology). In what follows, w stands for prosodic word and φ for phonological phrase.2

(7) Four possible sites for function words in a prosodic tree structure (based on Itô & Mester 2009: 150) a. full-w fnc

((lex)w (fnc)w)φ b. amalgamated fnc

((lex fnc)w)φ c. w-adjoined fnc

(((lex)wfnc)w)φ d. φ-attached fnc

((lex)wfnc)φ

3.2. Optimal parsing of func-lex structures The specific parsing of func-lex structures depends on the language-particular ranking of a set of constraints on prosodic representation of different types: parsing constraints, interface constraints, size constraints, and tree form constraints. In Itô & Mester (2009), prosodic parsing is triggered by the satisfaction of a set of Parse- into-X constraints, where X is a variable that stands for all prosodic categories. For

2. In the early days of Prosodic Phonology, the Strict Layer Hypothesis (Selkirk 1984) advocated both full prosodic parsing and no recursivity. The theory was relaxed later on with the advent of Optimality Theory. Verb-Clitic Structures in Eivissan Catalan CatJL 18, 2019 199 our purposes it is enough to make use of just one parsing constraint: Parse-into-w, defined in (8). Parse-into-w is violated by φ-attached function words (7d), in which the function word is not parsed into any prosodic word.

(8) Parse-into-w (Itô & Mester 2009: 139) Every element of the terminal string is parsed at the w-level.

Interface constraints are responsible for matching one edge (i.e. left or right) of certain syntactic constituents with one edge of a specific prosodic category.3 For the cases at hand, in which function words are pronominal enclitics, we just need one interface constraint: Lex-to-w(R), formulated in (9).4 This constraint penalizes the structure in (7b); with proclisis this constraint would not be violated.

(9) Lex-to-w(R) (Itô & Mester 2009: 142) Every lexical word is right aligned with a prosodic word.

Lex-to-w(R) is violated by function words that amalgamate with their hosts at the right edge because the right edge of the lexical word does not match the right edge of any prosodic word. Another crucial constraint is FootBinarity, a size constraint defined in (10) that excludes full-w function words, that is, prosodic words that contain subminimal, degenerate feet, on the proviso that function words are short enough.5

(10) FootBinarity (based on McCarthy & Prince 1993: 58) Every foot must be bimoraic or disyllabic.

Finally, a fourth constraint is needed to account for the full typology depicted in (7), the tree form constraint No-Recursion in (11).

(11) No-Recursion (Itô & Mester 2009: 145) An element is parsed only once into a given category. Assign one violation mark for each additional parse of an element into the same category.

We claim that pronominal enclitics in EC are w-adjoined function words (as in 7c) and therefore violate No-Recursion. The optimal parsing of pronominal enclitics as w-adjoined function words depends on the constraint ranking illustrated in the tableau in (12). Candidate (12d), with a full-w enclitic, violates FootBinarity because the clitic projects its own prosodic word and is therefore parsed by a

3. In constraint-based theories of the phonology-syntax interface it is consensual that only lexical syntactic categories can be referred to by interface constraints. 4. A more recent theory of the phonology-syntax interface is Match Theory (proposed in Selkirk 2009, 2011; see also Elfner 2011 and Itô & Mester 2013). Match Theory requires units of syntactic structure to have corresponding units in the prosodic structure. Lex-to-w(R) could be reformulated as Match(Lex, PWd), in which the latter does not make reference to edges. 5. This is only true for monomoraic function words. 200 CatJL 18, 2019 Francesc Torres-Tamarit; Eulàlia Bonet subminimal foot. Candidate (12c) exemplifies a case of an amalgamated enclitic, which is ruled out because it fatally violates the interface constraint Lex-to-w(R). Candidate (12b), with a φ-attached enclitic, violates the parsing constraint Parse- into-w because the enclitic is not parsed by any prosodic word and it simply skips this level to directly attach to the phonological phrase. Finally, the winning candidate is candidate (12a), with a w-adjoined enclitic, which violates the constraint No-Recursion. This constraint is dominated by all the other constraints mentioned so far.

(12) Constraint ranking for w-adjoined enclitics in Eivissan Catalan

lex fnc FtBin Lex-to-w(R) Parse-into-w No-Rec

☞ a. (((lex)w fnc)w)φ *

b. ((lex)w fnc)φ *W L

c. ((lex fnc)w)φ *W L

d. ((lex)w (fnc)w)φ *W L

In the next section we give evidence for recursive prosodic words in verb-clitic structures in EC and discuss their prosodification.

3.3. Recursive prosodic words in Eivissan Catalan: prosodic domains Leaving aside Formenteran, Majorcan and Minorcan Catalan, whose pronominal enclitics are necessarily amalgamated function words due to the observed systematic stress shift, we argue that, in EC, pronominal enclitics are w-adjoined function words, derived by the constraint ranking in (12). In such a representation, the lexical category, the verb, coincides with the boundaries of a minimal prosodic word wmin, which is further dominated by a maximal prosodic word wmax that integrates the enclitic(s), as illustrated in (13), parallel to (7c).6

(13) Minimal and maximal prosodic words in verb-clitic structures in Eivissan Catalan

(((V)wmin cl)wmax)φ

The structure in (12b), in which the clitic directly attaches to the phonological phrase, must be excluded because postverbal clitics affect the location of stress; clitics cannot lie outside the domain of stress assignment, which is the prosodic

6. It is enough for our purposes to simply distinguish between minimal and maximal words. See Itô & Mester (2007, 2009, 2012, 2013), van der Hulst (2010), Selkirk (2011), Elfner (2015), Martínez- Paricio (2012), Martínez-Paricio & Kager (2015), Bennett (2018), among others, for recursivity in metrical and prosodic phonology and the necessity of labels based on dominance relations in recursive prosodic structures. Verb-Clitic Structures in Eivissan Catalan CatJL 18, 2019 201 word. The structure in (12c), in which the clitic amalgamates prosodically with the verbal host must also be discarded because postverbal clitics never receive stress in EC. Finally, the structure in (12d), in which the clitic is parsed into its own prosodic word, is also inadequate for EC for the same reason: clitics never receive stress.7 We should now ask what kind of evidence is available that gives support to the representation in (13) with recursion as adjunction. Independent evidence comes from segmental phonology. In Catalan (except for Valencian Catalan), rhotics delete in word-final position (see Bonet & Lloret 1998).8 Some morphophonological alternations are illustrated in (14) for EC.

(14) Word-final deletion of rhotics in Eivissan Catalan [fusˈte] ‘carpenter’ cf. [fustəˈɾiə] ‘carpentry’ [ˈma] ‘sea’ cf. [məɾiˈne] ‘sailor’ [rəˈmo] ‘noise’ cf. [rəmuˈɾətə] ‘small noise’

In addition, consonantal clusters formed by a nasal or a lateral plus a homorganic stop undergo deletion of the stop in word-final position (except for Valencian and Majorcan Catalan). Some morphophonological alternations are illustrated in (15) for EC.

(15) Word-final cluster simplification in Eivissan Catalan [ˈsan] ‘saint’ cf. [səntəˈðat] ‘holiness’ [ˈven] ‘wind’ cf. [vənˈtaðə] ‘strong wind’ [ˈfɔn] ‘wellspring’ cf. [funˈtətə] ‘small wellspring’

Both deletion of rhotics and cluster simplification apply in word-final position, also when they are followed by a vowel-initial word, as seen in (16).

7. Peperkamp (1997) also proposes a recursive prosodic word as an explanation for enclitic-induced stress shift in Neapolitan. In Neapolitan, imperatives followed by one enclitic have no stress shift, but when two enclitics follow, stress falls on the first enclitic and the verb surfaces with secondary stress (e.g. pòrta=tíl=lə). This is not what we find in EC, in which only one stress surfaces in infinitives followed by one or more enclitics. Two different analyses can be assumed for infinitives followed by two enclitics in EC: (i) every enclitic is attached independently to a recursive prosodic word, or (ii) all enclitics are adjoined to the same maximal prosodic word. We will assume that the domain for stress assignment is the minimal prosodic word. If this is assumed, the constraint responsible for determining the minimal prosodic word as the domain for stress assignment would be enough to exclude building a foot on top of the enclitics. Then, all clitics can be simply adjoined to the same maximal prosodic word, minimizing the number of violations of No-Recursion. If the domain for stress assignment were any prosodic word, then another constraint should exclude the possibility of parsing enclitics into a foot. This constraint could be a Head-to-Lex constraint “requiring heads of prosodic categories to be contained in lexical (not functional) material (Itô & Mester 2009: 187)”. 8. In nouns and adjectives, word-final rhotic deletion applies only after a stressed vowel. In Eivissan and Formenteran Catalan rhotics also delete after unstressed vowels in verbs. 202 CatJL 18, 2019 Francesc Torres-Tamarit; Eulàlia Bonet

(16) Application of both deletion processes before vowel-initial words in Eivissan Catalan [ˈma uˈβɛɾtə] ‘open see’ (*[ˈmaɾ uˈβɛɾtə]) [ˈven uɾəkəˈnat] ‘hurricane wind’ (*[ˈvent uɾəkəˈnat])

In verb-clitic structures, however, both deletion processes are blocked, as shown in (17).

(17) Blocking of deletion processes in verb-clitic structures in Eivissan Catalan [kənˈtaɾ=mə] ‘to sing to me’ (*[kənˈta=mə]) [kumˈpɾant=i] ‘buying there’ (*[kumˈpɾan=i])

From the data in (16) it is clear that the prosodic domain of rhotic deletion and cluster simplification is not the phonological phrase, but the maximal word (e.g.

((ˈma)wmax (uˈβɛɾtə)wmax)φ). If enclitics attached directly to the phonological phrase, one would expect the two deletion processes to apply in verb-clitic structures, contrary to facts (e.g *[kənˈta=mə], with a structure ((kənˈta)w=mə)φ, instead of [kənˈtaɾ=mə]). If a recursive structure is assumed for verb-clitic structures, the rhotic and the consonantal cluster are not final at the level of the maximal word, only at the level of the minimal prosodic word, and hence deletion is not expected.9 If the prosodic domain for stress assignment in EC is the minimal prosodic word, and the domain for the deletion processes is the maximal prosodic word, all the observed facts follow, including the locality effect of stress shift in verb-clitic structures, which is circumscribed to move one syllable to the right, that is, to the last syllable of the verbal stem, but not further. In verb-clitic structures, the verb is prosodified on its own minimal prosodic word. In those second conjugation verbs ending in /-əɾ/, the process of deletion of rhotics is blocked because the rhotic is not final within the maximal prosodic word, the one that contains the enclitic. The fact that deletion does not apply makes the last syllable of the verbal stem a heavy syllable, and therefore stress shifts one syllable to the right to conform with the unmarked stress pattern of Catalan: a right-aligned moraic trochee, as illustrated in (18). Parentheses mark foot boundaries and square brackets prosodic word boundaries. In isolation, second conjugation verbs have penultimate stress because, as opposed to first and third conjugation verbs, they have no thematic vowel, which is always stressed.

9. One could assume that the prosodic domain of both processes of deletion is the clitic group, a distinct prosodic category. However, such an analysis would need to posit such a category for every

non-cliticized word, in which the deletion processes also apply (e.g. (((ˈma)w)ClGr ((uˈβɛɾtə)w)ClGr)φ). These forms face the problem of “too much structure”. See Itô & Mester (2009) for more arguments against the clitic group. Verb-Clitic Structures in Eivissan Catalan CatJL 18, 2019 203

(18) Stress shift in Eivissan Catalan V V=cl

[ku(ˈnəʃ-ə)]wmin, max [[kunə(ˈʃ-əɾ)]wmin=tə]wmax

The constraints responsible for stress assignment in EC are the size constraint FootBinarity-μ, which disfavors feet that are not binary under a moraic analysis; the foot form constraint Trochee, which disfavors feet whose head is not left-aligned, and which dominates Iamb; and a specific version of the alignment constraint All-Feet-Right, that is, All-Feet-Right(wmin), which needs to refer to the right edge of minimal prosodic words. In the tableau in (19) we illustrate stress shift in second conjugation verbs followed by one enclitic. The last syllable of the verbal stem being heavy, a foot can be built at the right edge of the minimal prosodic word without incurring a violation of All-Feet-Right(wmin). For the selection of the infinitive allomorph /-əɾ/ over /-ɾ/, see section 4.

(19) Stress shift in a second conjugation verb followed by one enclitic in Eivissan Catalan

All-Feet- All-Feet- Trochee FootBin-μ Iamb kunəʃ-{ɾ > əɾ}=tə R(wmin) L(wmin)

☞ a. [[kunə(ˈʃəɾ)]wmin tə]wmax ** *

b. [[ku(nəˈʃəɾ)]wmin tə]wmax *W *W *L L

c. [[ku(ˈnə)ʃəɾ]wmin tə]wmax *W *W *L L

In EC, there is simply no stress shift in first and third conjugation verbs with enclisis because these infinitives always have final stress, and therefore metrical optimization is satisfied vacuously, as illustrated in (20). In this example, the isolated infinitive ends up surfacing with a degenerate foot due to the process of word-final deletion of rhotics (a case of phonological opacity between segmental deletion and stress assignment). In verb-clitic structures, as already seen, the rhotic does not delete and the foot is binary under a moraic analysis. There is no stress shift because stress is already final in the isolated infinitive.

(20) No stress shift in first conjugation verbs in Eivissan Catalan V V=cl

[kən(ˈt-a)]wmin, max [[kən(ˈt-aɾ)]wmin=mə=lə]wmax

In Formenteran Catalan, however, clitics are amalgamated within the same prosodic word that contains the host, and therefore there is only one prosodic word in structures with enclisis. This prosodic word is both minimal and maximal, and therefore stress can shift all the way to the right as shown in (21) to optimize the 204 CatJL 18, 2019 Francesc Torres-Tamarit; Eulàlia Bonet metrical structure and, in the presence of two enclitics with the shape CVCV, a metrical foot is built on top of them.10

(21) Stress shift in first conjugation verbs in Formenteran Catalan V V=cl

[kən(ˈt-a)]wmin, max [kənt-əɾ=(ˈmə=lə)]wmin, max

In the next section we describe and analyze the two cases of allomorphy found in infinitive-clitic structures in EC.

4. Allomorphy in infinitive-clitic structures in Eivissan Catalan In this section, we focus on the phonological alternations found between infinitives in isolation and infinitives followed by pronominal enclitics in both Eivissan (EC) and Formenteran Catalan (FC). As will be seen, in EC the infinitive can surface with the ending [-əɾ] when followed by enclitics, even though the form in isolation has [-ɾə]. In FC, those verbs ending in [-ɾə] in isolation surface sys- tematically with [-ɾəɾ], that is, with a double realization of the infinitive morph (an option also found in EC for verbs ending in [-ɾə]: [[rum(ˈp-əɾ)]wmin=mə]wmax 11 and [[rum(ˈp-ɾəɾ)]wmin=mə]wmax; therefore, there is variation in EC). Before discussing the data in more detail, let us first present the analysis of allomorph selection of the infinitive morph in isolated forms. As mentioned in section 2, we assume that EC, as well as FC, has two lexically ordered infinitive morphs: /-ɾ/ and /-əɾ/ ({ɾ > əɾ}). We claim that it is the phonology that governs the selection of the allomorph since it depends on the last consonant of the verbal root. The constraint set that regulates allomorph selection appears in (22).

10. For EC an analysis in terms of amalgamated function words is also possible, as long as an asymmetric output-output constraint is added that requires sequences with two enclitics to be stressed like sequences with one enclitic, which acts as the base. The form with two

clitics [məɾə(ˈʃ-əɾ)=mə=lə]w has stress on the last syllable of the verb because it is based on [məɾə(ˈʃ-əɾ)=mə], with the less severe violation of All-Feet-R; forms with one vocalic clitic, like (6c), [ku(ˈr-əɾ=i)] ‘to run there’, would have a perfectly right-aligned moraic trochee in this analysis. In our recursivity-based analysis, asymmetric output-output constraints are unnecessary. Forms with one vocalic clitic, however, do pose a challenge to the recursivity- based analysis. Resyllabification of the infinitive morph as the onset of the vocalic clitic triggers ex to t in a violation of both L - -w(R) and F B : [[ku(ˈr-ə)]wmin ɾ=i]wmax. Maintaining such a form as the optimal candidate requires another type of output-output constraint, in the line of the model Optimal Paradigms (McCarthy 2005). In this case the paradigm would consist of the infinitive plus all the possible clitics and an OP-Ident(stress) constraint would favor paradigms with uniform stress. Since most of the clitics start with a consonant they would drag infinitives with a vowel-initial clitic into their stress pattern. We thank an anonymous reviewer for drawing our attention to the issues posed by vowel-initial clitics. 11. Our impression is that variation between [-əɾ] and [-ɾəɾ] in EC is a case of free variation, but further investigation is needed to elucidate whether this variation is dependent on extralinguistic factors. Verb-Clitic Structures in Eivissan Catalan CatJL 18, 2019 205

(22) Constraint set regulating the allomorph selection of the infinitive morpheme

a. Coda-Condition (based on McCarthy 2008: 277) Assign one violation mark for every coda that licenses a consonantal place feature (it prohibits all coda consonants except for homorganic ones).

b. Syllable-Contact (McCarthy 2008: 229) Assign one violation mark for every heterosyllabic consonant cluster with rising sonority.

c. Priority (Mascaró 2007: 726) Respect lexical priority (ordering) of allomorphs.

d. Dep-C Assign one violation mark for every consonant in the output that does not have a correspondent in the input.

The effect of each constraint is shown in the following tableaux. First consider an infinitive in which the last consonant of the root is a sibilant (23). In such a case, syllabifying the sibilant in coda position implies a fatal violation of Coda- Condition (candidates 23b and 23c) and of Syllable-Contact if no epenthetic consonant intervenes (candidate 23b). The winner, which has undergone final rhotic deletion through constraints not included in the tableau, surfaces with the less preferred allomorph, violating Priority, and therefore satisfies the two undominated constraints, Coda-Condition and Syllable-Contact.

(23) Tableau for /kunəʃ-{ɾ > əɾ}/

kunəʃ-{ɾ > əɾ} Coda-Cond Syll-Cont Priority Dep-C ☞ a. ku.ˈnə.ʃə * b. ku.ˈnəʃ.ɾə *W *W L c. ku.ˈnəʃ.ðɾə *W L *W

If the last consonant of the verbal root can be syllabified as a complex onset together with the rhotic, the preferred allomorph /ɾ/ is selected, as shown in (24). Candidate (24a) harmonically bounds all other candidates: there is no reason to select the less preferred allomorph /əɾ/, and consonant epenthesis introduces a gra- tuitous violation of Dep-C. We omit the constraint Dep-V, which is also violated by the winning candidate with a final epenthetic schwa. 206 CatJL 18, 2019 Francesc Torres-Tamarit; Eulàlia Bonet

(24) Tableau for /romp-{ɾ > əɾ}/

romp-{ɾ > əɾ} Coda-Cond Syll-Cont Priority Dep-C ☞ a. ˈrom.pɾə b. ˈrom.pə *W c. ˈron.dɾə *W

A similar situation is observed for verbs whose last consonant is /-w/. In such cases, the labiovelar glide can be syllabified in coda position without incurring a violation of Syllable-Contact.12

(25) Tableau for /tɾɛw-{ɾ > əɾ}/

tɾɛw-{ɾ > əɾ} Coda-Cond Syll-Cont Priority Dep-C ☞ a. ˈtɾɛw.ɾə b. ˈtɾɛ.wə *W c. ˈtɾɛw.ðɾə *W

Finally, verbs ending in /-n/ or /-l/ can take the preferred allomorph /ɾ/ because by epenthesizing [-d-] forms like (26a) do not violate any markedness constraints. It is crucial that Priority outranks Dep-C.13

(26) Tableau for /ənsen-{ɾ > əɾ}/

ənsen-{ɾ > əɾ} Coda-Cond Syll-Cont Priority Dep-C ☞ a. ən.ˈsen.dɾə * b. ən.ˈse.nə *W L c. ən.ˈsen.ɾə *W L

12. We assume a version of Coda-Condition that affects segments in the coda with the feature [+consonant]. For this reason the winner in (25a), with a glide in coda position, does not violate the constraint. 13. We exclude from the tableau in (26) a candidate like [ən.ˈsen.rə], in which the flap maps onto a trill (an output found in some varieties of Catalan). If we assume that trills are as sonorous as stops (Bonet & Mascaró 1997), this candidate would not violate Syllable-Contact, but it would be ruled out by a higher-ranked identity constraint against changing the feature value that distinguishes taps from trills. Pons-Moll (2011) assumes that the trill is more sonorous than nasals but less sonorous than flaps. A candidate like [ən.ˈsen.rə], with a trill, would still violate Syllable-Contact, but to a lesser degree than [ən.ˈsen.ɾə], with a flap. Verb-Clitic Structures in Eivissan Catalan CatJL 18, 2019 207

Turning now to infinitives with enclitics, second conjugation verbs that end in [-ə(ɾ)] remain unaltered with respect to the infinitive allomorph selected both in EC and in FC. A few examples are illustrated in (27). The only difference has to do with stress placement due to the presence of a recursive prosodic word in EC, but not in FC.

(27) Stress shift in verbs ending in [-ə(ɾ)] a. EC

inf. [mə(ˈɾəʃ-ə)]wmin, max

inf.+1Cl [[məɾə(ˈʃ-əɾ)]wmin=mə]wmax

inf.+2Cl [[məɾə(ˈʃ-əɾ)]wmin=mə=lə]wmax b. FC

inf. [mə(ˈɾəʃ-ə)]wmin, max

inf.+1Cl [məɾə(ˈʃ-əɾ)=mə]wmin, max

inf.+2Cl [məɾəʃ-əɾ=(ˈmə=lə)]wmin, max

Interestingly, second conjugation verbs ending in [-rə] and [-d-ɾə] do not surface as such under enclisis. In EC, the ending is instead [-əɾ], following the pattern of forms like [[məɾə(ˈʃ-əɾ)]wmin=mə]wmax, and in FC, the infinitive morph is expressed twice (i.e. [-ɾəɾ]).14 Some examples appear in (28).

(28) Allomorphy in verbs ending in [-ɾə] a. EC

inf. [(ˈrom)p-ɾə)]wmin, max

inf.+1Cl [[rum(ˈp-əɾ)]wmin=mə]wmax (opt. [rum(ˈp-ɾəɾ)]wmin=mə]wmax)

inf.+2Cl [[rum(ˈp-əɾ)]wmin=mə=lə]wmax (opt. [[rum(ˈp-ɾəɾ)]wmin= mə=lə]wmax) b. FC

inf. [(ˈrom)p-ɾə)]wmin, max

inf.+1Cl [rum(ˈp-ɾəɾ)=mə]wmin, max

inf.+2Cl [rump-ɾəɾ=(ˈmə=lə)]wmin, max

14. In Majorcan Catalan, the epenthetic [-d-] in verbs whose last consonant ends in /-n, -l/ is maintained as in FC, and the infinitive morph is repeated, surfacing with total assimilation to the following consonant (e.g. [ənsən-d-ɾəl=(ˈlo)] ‘to turn it.m on’). However, we have observed that it is also possible to take the thematic vowel of third conjugation verbs in the presence of enclitics (e. g. [əpɾən-i-l=(ˈlo)] ‘to learn it.m’). This possibility is also attested in imperatives with enclitics (e.g. [əpɾə(ˈn-i=l)] ‘learn it.m!’) (see Bonet & Torres-Tamarit 2011 for more details on imperatives). Future research on this topic in Majorcan and Minorcan Catalan is still necessary to fully understand allomorphy in infinitives followed by enclitics. 208 CatJL 18, 2019 Francesc Torres-Tamarit; Eulàlia Bonet

(29) Allomorphy in verbs ending in [-d-ɾə] a. EC

inf. [ən(ˈsen)-d-ɾə]wmin, max

inf.+1Cl [[ənsə(ˈn-əɾ)]wmin=mə]wmax

inf.+2Cl [[ənsə(ˈn-əɾ)]wmin=mə=lə]wmax b. FC

inf. [ən(ˈsen)-d-ɾə]wmin, max

inf.+1Cl [ənsən-(ˈd-ɾəɾ)=mə]wmin, max

inf.+2Cl [ənsən-d-ɾəɾ=(ˈmə=lə)]wmin, max

Furthermore, in EC, second conjugation verbs that end in /-w/ can present another peculiarity, root allomorphy. In these verbs, the ending [-ɾə] of the infinitive is not only replaced by [-əɾ], but the verbal root, instead of ending in the expected /-w/, ends in /-ɡ/; this is the root allomorph used in the subjunctive mood, and also in gerunds and participles (i.e. mou [ˈmɔw] ‘(s)he moves’ cf. mogui [ˈmɔɣ-i] ‘(s)he moves.subj’, moguent [muˈɣen] ‘moving’, mogut [muˈɣut] ‘moved’).15 In FC, as expected, the infinitive ending is just [-ɾəɾ].16

(30) Root allomorphy in verbs whose root ends in /-w/ a. EC

inf. [(ˈmɔw)-ɾə]wmin, max

inf.+1Cl [[mu(ˈɣ-əɾ)]wmin=mə]wmax

inf.+2Cl [[mu(ˈɣ-əɾ)]wmin=mə=lə]wmax b. FC

inf. [(ˈmɔw)-ɾə]wmin, max

inf.+1Cl [muw(ˈ-ɾəɾ)=mə]wmax

inf.+2Cl [muw-ɾəɾ=(ˈmə=lə)]wmin, max

To sum up, in both EC and FC the variation found in isolated infinitives (e.g. [məˈɾəʃ-ə] vs. [ˈromp-ɾə]) is neutralized under enclisis: all infinitives end in [-əɾ] (optionally in [-ɾəɾ]) in EC, and in [-ɾəɾ] in FC. On top of this, EC also exhibits a case of root allomorphy in verbs whose root ends in /-w/. As shown below, we analyze the allomorphy of the infinitive morph under enclisis as triggered by the same constraints responsible for stress assignment in Catalan, with no need to make use of new constraints. Furthermore, we propose

15. Some examples have been obtained from spontaneous speech in colloquial environments where native speakers of EC were not aware of the fact that the linguist was scrutinizing their speech: [ʒəˈɣ-əɾ=lə] (cf. [ˈʒɛw-ɾə] ‘to lay down’), [kuˈɣ-əɾ=lə] (cf. [ˈkɔw-ɾə] ‘to cook’), [bəˈɣ-əɾ=lu] (cf. [ˈbəw-ɾə] ‘to drink’), [tɾəˈɣ-əɾ=lus] (cf. [ˈtɾɛw-ɾə] ‘to remove’). 16. Root allomorphy is also attested in younger speakers of FC. Verb-Clitic Structures in Eivissan Catalan CatJL 18, 2019 209 that root allomorphy is due to a markedness constraint that prohibits /w/ in the onset position of stressed syllables (this prohibition is in fact observed in all verbal forms in Catalan, as will be shown). We start by recalling the analysis of stress shift in second conjugation verbs ending in [-əɾ], which was shown in (19), repeated below in (31) for convenience.

(31) Stress shift in a second conjugation verb followed by one enclitic in Eivissan Catalan

All-Feet- All-Feet- Trochee FtBin-μ Iamb kunəʃ-{ɾ > əɾ}=tə R(wmin) L(wmin)

☞ a. [[kunə(ˈʃəɾ)]wmintə]wmax ** *

b. [[ku(nəˈʃəɾ)]wmintə]wmax *W *W *L L

c. [[ku(ˈnə)ʃəɾ]wmintə]wmax *W *W *L L

The constraint Priority, defined in (22c), becomes relevant in infinitives ending in [-ɾə] in isolation, as well as the constraint *AffixRepetition, defined in (32), which disfavors multiple output correspondents of a single morpheme.17

(32) *AffixRepetition Assign one violation mark for every affix that has more than one corresponding morph. (See Ortmann 1999)

When stress shifts to the last syllable of the verbal stem in order to satisfy All- Feet-R(wmin), there are only two possibilities to also satisfy FootBinarity-m, as illustrated in (33). One possibility is to express the infinitive morpheme twice with the same allomorph and medial schwa epenthesis, (candidate 33d).18 This is the solution in FC. Another possibility is to select the less preferred infinitive allomorph /əɾ/, as in (33a); the last segment of the verbal form of this candidate can be syllabified as a coda and therefore makes the last syllable of the verbal stem heavy. In this analysis, the allomorphy of the infinitive morpheme in verb-clitic structures is triggered by the constraints responsible for stress shift, All-Feet-R(wmin) and FootBinarity-m, violated by candidates (33b) and (33c), respectively. In FC, Priority must be ranked above *AffixRepetition to model the optimality of candidate (33d). Each allomorph of the infinitive morpheme is identified with a subindex for clarity.

17. We could alternatively interpret the doubling of the infinitive morph as a case of consonant copy epenthesis (Kawahara 2007). 18. A homophonous candidate would be one in which, instead of expressing the first allomorph twice, the two allomorphs of the same morpheme are both expressed. This candidate would violate Priority in addition to *AffixRepetition and therefore would be harmonically bounded by the candidate in which the same allomorph, the least marked, has been repeated. 210 CatJL 18, 2019 Francesc Torres-Tamarit; Eulàlia Bonet

(33) Tableau for /romp-{ɾ > əɾ}=lə/ in Eivissan Catalan

All-Feet- *AffRep FtBin-m Prior romp-{ɾ1 > əɾ2}=lə R(wmin) * ☞ a. [[rum(ˈp-əɾ2)]wmin=lə]wmax *W L b. [[rum(ˈp-ɾ1ə)]wmin=lə]wmax *W L c. [[(ˈrom)p-ɾ1ə]wmin=lə]wmax *W L d. [[rum(ˈp-ɾ1əɾ1)]wmin=lə]wmax

The same analysis applies for infinitives ending in [-d-ɾə] in isolation, in which the epenthetic [-d-] does not surface in forms with enclitics and the less preferred allomorph /əɾ/ is also selected. As mentioned above, second conjugation verbs whose root ends in the labiovelar glide /-w/ do not only show allomorphy of the infinitive morph as illustrated in (33), but also root allomorphy. In these verbs, the infinitive takes the verbal root found in the subjunctive and the other impersonal forms, gerund and participle, which end in /-ɡ/ (moure [ˈmɔwɾə] ‘(s)he moves’ but moguer-la [muˈɣəɾ=lə] ‘to move it.f’; cf. mogui [ˈmɔɣ-i] ‘(s)he moves.subj’, moguent [muˈɣen] ‘moving’, mogut [muˈɣut] ‘moved’). We hypothesize that this case of root allomorphy is also the consequence of stress shift together with a specific markedness constraint against /w/ in the onset position of a stressed syllable, *ˈσ[w, defined in (34).

(34) *ˈσ[w Assign one violation mark for every labiovelar glide [w] syllabified in the onset position of a stressed syllable.

This markedness constraint might seem language-specific, but it is well- known that onsets of higher sonority are dispreferred cross-linguistically, and that markedness constraints can refer to specific prosodic contexts.19 This markedness constraint is never violated in Catalan verbal forms with available allomorphs, and is responsible for alternations such as those illustrated in (35). In the first and second plural forms of the present indicative stress is final. In such cases, in order to prevent forms with [w] in the onset position of a stressed syllable, the last consonant of the verbal root is [β] in Central Catalan. All forms of the present subjunctive, as well as the gerund and the participle (leaving aside some highly

19. In nominal forms, the constraint *ˈσ[w is also active. It is always satisfied whenever a root ending in /-w/ has an allomorph ending in /-b/ (in Central Catalan) or /-v/ (in Balearic Catalan) (e.g. bla[w] ‘blue.m’, but bla[v/β]ós ‘bluish’ cf. *bla[w]ós). It is also optionally satisfied by means of epenthesis of a voiced velar stop [ɡ] when output contiguity would not be altered, that is, in word-initial position (e.g. [ɡ]whisky cf. *[w]hisky, but cali[w] ‘embers’ and acali[w]at ‘reduced to embers’ cf.

*acali[ɣw]at). In denominal verbs like creuar [kɾəˈwa] ‘to cross’, the constraint *ˈσ[w is violated in stressed syllables because the verb does not have an alternative available allomorph. Verb-Clitic Structures in Eivissan Catalan CatJL 18, 2019 211 irregular verbs), take the root allomorph ending in /-ɡ/, which surfaces as [ɣ] due to post-sonorant spirantization. For other verbs, roots ending in [w] alternate with roots ending in [j] (35b) (another case of root allomorphy). In Balearic Catalan, the first and second plural forms of these verbs are instead monosyllabic: [ˈsəjm] and [ˈsəjs]. What is crucial is the fact that forms like *[bəˈwəm] or *[səˈwəw] are banned in all dialects of Catalan, and that the markedness constraint *ˈσ[w is surface-true in Catalan verbs.

(35) Alternations triggered by the satisfaction of *ˈσ[w in Central and Balearic Catalan verbal forms a. PI (‘to drink’) b. PI (‘to sit down’) bec [ˈbɛk] / [ˈbək] sec [ˈsɛk] beus [ˈbɛws] / [ˈbəws] seus [ˈsɛws] beu [ˈbɛw] / [ˈbəw] seu [ˈsɛw] bevem [bəˈβɛm] / [ˈbəjm] seiem [səˈjɛm] / [ˈsəjm] beveu [bəˈβɛw] / [ˈbəjs] seieu [səˈjɛw] / [ˈsəjs] beuen [ˈbɛwən] / [ˈbəwən] seuen [ˈsɛwən]

The stress shift motivated by enclisis would cause a violation of the markedness constraint *ˈσ[w in those verbs whose root ends in the labiovelar /-w/: *[muˈw-əɾ=lə]. The question now is how to account for the actual repaired form [muˈɣ-əɾ=lə]. One possibility is to interpret the change as the result of a phonological process of strengthening, as suggested by an anonymous reviewer. One problem with this approach is that it is difficult to justify why the outcome is a velar consonant [ɣ] instead of a labial [v]. An alternative approach is to relate the consonant change to root allomorphy. These verbs do have an allomorph with /ɡ/ (/mɔɡ/) but also an allomorph with /v/ (/mɔv/). The allomorph with /v/ is found in the Imperfective Indicative (e.g. movia) and in first and second person plural of the present indicative (e.g. movem, moveu), while the allomorph with /ɡ/ is found in the subjunctive (e.g. mogui), the impersonal forms (e.g. moguent ‘moving’, mogut ‘moved’) and the first person present indicative (e.g. moc, with final devoicing), as well as some forms of the now archaic synthetic form of the past indicative (e.g. mogueres [muˈɣeɾəs] ‘you moved’). The choice of the allomorph with /ɡ/ could be attributed to two different explanations, one of them being a frequency effect, given that the allomorph with /ɡ/ is found in more forms of the conjugation than the allomorph with /v/. The other explanation would be related to morphosyntactic factors. The allomorph with /ɡ/ is found in the other non-personal forms, gerund and participle, while the allomorph with /v/ is found only in personal forms in EC and shares no morphosyntactic features with the infinitive. We leave for further research the exact formalization of this line of approach. 212 CatJL 18, 2019 Francesc Torres-Tamarit; Eulàlia Bonet

5. Conclusions In most Balearic varieties there is stress shift with pronominal enclisis (in infinitives and imperatives). In Eivissan Catalan, however, stress shift is found only in penultimately-stressed second conjugation infinitives, not in infinitives of other conjugations or imperatives. In isolation, most second conjugation infinitives surface with paroxytonic stress (e.g. [məˈɾəʃə] ‘to deserve’; [ˈrompɾə] ‘to break’; [ənˈsendɾə] ‘to turn on, to light’), contrary to all other infinitives, which are oxytonic. With enclisis, stress shifts one syllable to the right irrespective of the number of enclitics following the second conjugation infinitive (e.g. [məɾəˈʃəɾ=tə] ‘to deserve you’; [məɾəˈʃəɾ=tə=lə] ‘to deserve it for you’). In first and third conjugation infinitives followed by enclitics, which surface with final stress without clitics, there is no stress shift (e.g. [kənˈtaɾ=tə=lə] ‘to sing it to you’; [kuˈziɾ=tə=lə] ‘to sew it for you’). In this system, stress shift is circumscribed to affect only one conjugation class of infinitives; stress shift is strictly local, that is, stress only shifts one syllable to the right, and stress never falls on pronominal enclitics. This pattern contrasts with that of Formenteran Catalan, a closely related variety in which stress shifts to the penultimate syllable of the verb-clitic structure if it ends in a vowel and to the final syllable if it ends in a consonant in both infinitives and imperatives (e.g. [məɾəˈʃəɾ=tə] ‘to deserve you’; [məɾəʃəɾ=ˈtə=lə] ‘to deserve it for you’; [kəntəɾ=ˈtə=lə] ‘to sing it to you’; [kuziɾ=ˈtə=lə] ‘to sew it for you’). The pattern described for Formenteran Catalan is parallel to the one described for Lucanian, Italo-Romance (Peperkamp 1997), but the pattern of enclitic-triggered stress shift in Eivissan Catalan has never been described before. In this paper we have proposed an OT analysis of the prosodification of verb-clitic structures in Eivissan Catalan based on recursive adjunction, following Itô & Mester (2009) (similar to the analysis of Neapolitan in Peperkamp 1997).

The result is the following prosodic structure: ((inf)ω cl)ω)φ, in which there is a maximal prosodic word that contains the clitic(s) and a minimal prosodic word that contains the verbal host. At the same time, the constraint ranking for unmarked stress assignment in Catalan, which builds a right-aligned moraic trochee within the domain of the minimal prosodic word, is responsible for stress shift in second conjugation verbs. Finally, this paper has also accounted for the allomorphy found in second conjugation infinitives when followed by enclitics. Three types of second conjugation infinitives exist in Eivissan Catalan and most other dialects: those that end in [-ə(ɾ)], those that end in [-ɾə], and those that end in [-d-ɾə], with an epenthetic [d]. As seen, [ə(ɾ)]-ending infinitives have stress on the last syllable of the verbal stem with enclitics. Interestingly, all other second conjugation infinitives under encliticization surface as if they were [ə(ɾ)]-ending verbs (e.g. [rumˈp-əɾ=tə] ‘to break you’, but [ˈromp-ɾə] ‘to break’; [ənsəˈn-əɾ=tə] ‘to turn on/to light to you’, but [ənˈsen-dɾə] ‘to turn on/to light’). A distinct case of allomorphy in second conjugation infinitives whose root ends in /w/ is also accounted for. A constraint against [w] in the onset position of a stressed syllable in verbs induces, in the context of enclisis, the selection of a root allomorph that ends in /-ɡ/, which is also found in the other non-personal forms of the verb. Verb-Clitic Structures in Eivissan Catalan CatJL 18, 2019 213

References Anderson, Stephen. 2011. Clitics. In van Oostendorp, Marc, Ewen, Colin, Hume, Elizabeth & Rice, Keren (eds.). The Blackwell Companion to Phonology, 539- 559. Malden, MA: Blackwell. Bennett, Ryan. 2018. Recursive prosodic words in Kaqchikel (Mayan). Glossa: A journal of general linguistics 3(1): 67. Bonet, Eulàlia & Lloret, Maria-Rosa. 1998. Fonologia catalana. Barcelona: Ariel. Bonet, Eulàlia & Mascaró, Joan. 1997. On the representation of contrasting rhotics. In Martínez-Gil, Fernando & Morales-Front, Alfonso (eds.). Issues in the phonology and morphology of the major Iberian languages, 103-126. Washington D.C.: Georgetown University Press. Bonet, Eulàlia & Torres-Tamarit, Francesc. 2010. Allomorphy in pre-clitic imperatives in Formenteran Catalan: An output-based analysis. In Colina, Sonia, Olarrea, Antxon & Carvalho, Ana Maria (eds.). Romance Linguistics 2009. Selected papers from the 39th Linguistic Symposium on Romance Languages, 337-352. Amsterdam: John Benjamins. Bonet, Eulàlia & Torres-Tamarit, Francesc. 2011. Les formes d’imperatiu seguides de clític: un cas de conservadorisme lèxic. In Lloret, Maria-Rosa & Pons-Moll, Clàudia (eds.). Noves aproximacions a la fonologia i la morfologia del català, 37-61. Alacant: Institut Interuniversitari de Filologia Valenciana. Elfner, Emily. 2011. Syntax-prosody interactions in Irish. Doctoral dissertation, University of Massachusetts, Amherst. Elfner, Emily. 2015. Recursion in prosodic phrasing: Evidence from Connemara Irish. Natural Language & Linguistic Theory 33 (4): 1169-1208. Itô, Junko & Mester, Armin. 2007. Prosodic adjunction in Japanese compounds. In Miyamoto, Yoichi & Ochi, Masao (eds.). Formal Approaches to Japanese Linguistics (FAJL) 4, 97-111. Cambridge, MA: MITWPL. Itô, Junko & Mester, Armin. 2009. The extended prosodic word. In Grijzenhout, Janet & Kabak, Baris (eds.). Phonological domains: Universals and deviations, 135-194. Berlin & New York: Mouton de Gruyter. Itô, Junko & Mester, Armin. 2012. Recursive prosodic phrasing in Japanese. In Borowsky, Toni, Kawahara, Shigeto, Shinya, Takahito & Sugahara, Mariko (eds.). Prosody matters: Essays in honor of Elisabeth Selkirk, 280-303. London: Equinox. Itô, Junko & Mester, Armin. 2013. Prosodic subcategories in Japanese. Lingua 124: 20-40. Kager, René & Martínez-Paricio, Violeta. 2015. The binary-to-ternary rhythmic continuum in stress typology: Layered feet and non-intervention constraints. Phonology 32 (3): 459-504. Kaisse, Ellen. 2017. The domain of stress assignment: Word-boundedness and frequent collocation. In Bowern, Claire, Horn, Laurence & Zanuttini, Raffaella (eds.). On looking into words (and beyond), 17-41. Berlin: Language Science Press. Kawahara, Shigeto. 2007. Copying and spreading in phonological theory: Evidence from echo epenthesis. In Bateman, Leah, O’Keefe, Michael, Reilly, Ehren & 214 CatJL 18, 2019 Francesc Torres-Tamarit; Eulàlia Bonet

Werle, Adam (eds.). Papers in Optimality Theory III (University of Massachusetts Occasional Papers in Linguistics 32). GSLA Publications. Marí Tur, Roser. 2017. El segment velar en la flexió verbal de l’eivissenc: Anàlisi de la variació distribucional, geogràfica i generacional. Doctoral dissertation, Universitat de Barcelona. Martínez-Paricio, Violeta. 2012. An Exploration of minimal and maximal metrical feet. Doctoral dissertation, University of Tromsø-Arctic University of Norway. Mascaró, Joan. 1985. Morfologia. Barcelona: Enciclopèdia Catalana. McCarthy, John J. 2005. Optimal paradigms. In Downing, Laura, Hall, T. Alan & Raffelsiefen, Renate (eds.). Paradigms in phonological theory, 170-210. New York: Oxford University Press. McCarthy, John J. & Prince, Alan. 1993. Generalized alignment. Yearbook of Morphology 12. Retrieved from . Mascaró, Joan. 2007. External allomorphy and lexical representation. Linguistic Inquiry 38(4): 715-735. McCarthy, John. 2008. Doing Optimality Theory: Applying theory to data. Malden, MA: Blackwell. Nadeu, Marianna, Simonet, Miquel & Llompart, Miquel. 2016. Stressed postverbal pronominals in Catalan. Probus 29(1): 119-162. Nespor, Marina & Vogel, Irene. 1986. Prosodic Phonology. Dordrecht: Foris. Ortmann, Albert. 1999. Affix repetition and non-redundancy in inflectional morphology. Zeitschrift für Sprachwissenschaft 18(1): 76-120. Peperkamp, Sharon. 1997. Prosodic words (HIL dissertations 34). The Hague: Holland Academic Graphics. Pons-Moll, Clàudia. 2011. It is all downhill from here: A typological survey of the role of Syllable Contact in Romance languages. Probus 23(1): 105-173. Repetti, Lori. 2016. The phonology of postverbal pronouns in Romance languages. In Tortora, Christina, den Dikken, Marcel, Montoya, Ignacio & O’Neill, Teresa (eds.). Romance Linguistics 2013. Selected papers from the 43rd Linguistic Symposium on Romance Languages, 361-378. Amsterdam: John Benjamins. Ribas Tur, Maria. 2018. La realització de les formes verbals seguides d’enclítics en eivissenc. Unpublished undergraduate thesis, Universitat Autònoma de Barcelona. Selkirk, Elisabeth. 1981. On the nature of phonological representation. In Anderson, John, Laver, John & Myers, Terry (eds.). The cognitive representation of speech, 379-388. Amsterdam: North Holland. Selkirk, Elisabeth. 1984. Phonology and syntax: The relation between sound and structure. Cambridge, MA: MIT Press. Selkirk, Elisabeth. 1995. The prosodic structure of function words. In Beckman, Jill, Walsh Dickey, Laura & Urbanczyk, Suzanne (eds.). University of Massachusetts occasional papers 18: papers in Optimality Theory, 439-469. Amherst, MA: GLSA. Selkirk, Elisabeth. 2009. On clause and intonational phrase in Japanese: The syntactic grounding of prosodic constituent structure. Gengo Kenkyu (Journal of the Linguistic Society of Japan) 136: 35-73. Verb-Clitic Structures in Eivissan Catalan CatJL 18, 2019 215

Selkirk, Elisabeth. 2011. The syntax-phonology interface. In Goldsmith, John, Riggle, Jason & Yu, Alan. The Handbook of Phonological Theory, 435-484. Oxford: Blackwell. Steriade, Donca. 1999. Lexical conservatism in French adjectival liaison. In Bullock, Barbara, Authier, Jean-Marc & Reed, Lisa (eds.). Formal Perspectives in Romance Linguistics, 243-270. Amsterdam: John Benjamins. Steriade, Donca. 2008. A pseudo-cyclic effect in Romanian morphophonology. In Bachrach, Asaf & Nevins, Andrew (eds.). Inflectional identity, 313-360. Amsterdam: John Benjamins. Torres-Tamarit, Francesc & Pons-Moll, Clàudia. 2019. Enclitic-induced stress shift in Catalan. Journal of Linguistics 55: 407-444. van der Hulst, Harry. 2010. A note on recursion in phonology. In van der Hulst, Harry (ed.). Recursion and human language, 301-342. Berlin & New York: Mouton de Gruyter. Vigário, Marina. 2010. Prosodic structure between the prosodic word and the phonological phrase: Recursive nodes or an independent domain? The Linguistic Review 27: 485-530. Vogel, Irene. 2009. The status of the clitic group. In Grijzenhout, Janet & Kabak, Baris (eds.). Phonological domains: Universals and deviations, 15-46. Berlin & New York: Mouton de Gruyter.

Catalan Journal of Linguistics Editors Centre de Lingüística Teòrica de la Universitat Autònoma de Barcelona Institut Interuniversitari de Filologia Valenciana Catalan Journal of Linguistics style sheet