The Phonetics and Phonology of Laryngeal Features in Native American Languages [1 ed.] 9789004303218, 9789004303201

Citation preview

The Phonetics and Phonology of Laryngeal Features in Native American Languages

Brill’s Studies in the Indigenous Languages of the Americas Series Editors David Beck (University of Alberta) Mily Crevels (Leiden University) Hein van der Voort (Museu Paraense Emílio Goeldi) Roberto Zavala (CIESAS-Sureste) Editorial Board Peter Bakker (Aarhus University) Nora England (University of Texas, Austin) Ana Fernández Garay (Universidad Nacional de La Pampa) Michael Fortescue (University of Copenhagen) Victor Golla (Humboldt State University) Pieter Muysken (Radboud University Nijmegen) Enrique Palancar (CNRS) Keren Rice (University of Toronto) Frank Seifart (Max Planck Institute for Evolutionary Anthropology) Leo Wetzels (CNRS/Sorbonne-Nouvelle, VU Amsterdam)

VOLUME 12

The titles published in this series are listed at brill.com/bsila

The Phonetics and Phonology of Laryngeal Features in Native American Languages Edited by

Heriberto Avelino Matt Coler W. Leo Wetzels

LEIDEN | BOSTON

Library of Congress Cataloging-in-Publication Data Names: Avelino Becerra, Heriberto, 1968– editor. | Coler, Matt, editor. | Wetzels, Leo, editor. Title: The phonetics and phonology of laryngeal features in Native American languages / Edited by Heriberto Avelino, Matt Coler, W. Leo Wetzels. Description: Leiden ; Boston : Brill, [2016] | Series: Brill’s studies in the indigenous languages of the Americas ; v. 12 | Includes bibliographical references and index. | Description based on print version record and CIP data provided by publisher; resource not viewed. Identifiers: LCCN 2015045562 (print) | LCCN 2015038981 (ebook) | ISBN 9789004303218 (E-book) | ISBN 9789004303201 (hardback : alk. paper) | ISBN 9789004303218 (e-book) Subjects: LCSH: Indians of North America—Languages—Phonetics. | Indians of North America— Languages—Phonology. | Indians of South America—Languages—Phonetics. | Indians of South America—Languages—Phonology. | Laryngeals (Phonetics) Classification: LCC PM115 (print) | LCC PM115.P56 2016 (ebook) | DDC 497—dc23 LC record available at http://lccn.loc.gov/2015045562

This publication has been typeset in the multilingual “Brill” typeface. With over 5,100 characters covering Latin, ipa, Greek, and Cyrillic, this typeface is especially suitable for use in the humanities. For more information, please see www.brill.com/brill-typeface. issn 1876-5580 isbn 978-90-04-30320-1 (hardback) isbn 978-90-04-30321-8 (e-book) Copyright 2016 by Koninklijke Brill nv, Leiden, The Netherlands. Koninklijke Brill nv incorporates the imprints Brill, Brill Hes & De Graaf, Brill Nijhoff, Brill Rodopi and Hotei Publishing. All rights reserved. No part of this publication may be reproduced, translated, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission from the publisher. Authorization to photocopy items for internal or personal use is granted by Koninklijke Brill nv provided that the appropriate fees are paid directly to The Copyright Clearance Center, 222 Rosewood Drive, Suite 910, Danvers, ma 01923, usa. Fees are subject to change. This book is printed on acid-free paper.

Contents Preface vii 1

Introduction to Laryngeal Features in Languages of the Americas 1 Heriberto Avelino, Matt Coler, and Leo Wetzels

2

Overlapping Laryngeal Classes in Athabaskan Languages: Continuity and Change 9 Keren Rice

3

Stem-Final Ejectives in Ahtna Athabascan 51 Siri G. Tuttle

4

Deg Xinag Word-Final Glottalized Consonants and Voice Quality 71 Sharon Hargus

5

Consonant-Tone Interactions: A Phonetic Study of Four Indigenous Languages of the Americas 129 Matthew Gordon

6

Phonetics in Phonology: A Cross-Linguistic Study of Laryngeal Contrast 157 Heriberto Avelino

7

The Role of Prominent Prosodic Positions in Governing Laryngealization in Vowels: A Case Study of Two Panoan Languages 180 José Elías-Ulloa

8

Pitch and Glottalization as Cues to Contrast in Yucatec Maya 203 Melissa Frazier

9

Amazonia and the Typology of Tone Systems 235 Larry M. Hyman

10

The Reconstruction of Laryngealization in Proto-Tukanoan 258 Thiago Costa Chacon

vi

contents

11

The Status of the Laryngeals ‘ʔ’ and ‘h’ in Desano 285 Wilson Silva

12

Temporal Coordination of Glottalic Gestures in Karitiana 308 Didier Demolin and Luciana Storto

Index 323

Preface This collection of studies is the result of a cooperation initiative subsidized by the Netherlands Organization for Scientific Research (grant number 236-70-005), which was started in 2007, among the Chair of Languages of the Amazon of the Vrije Universiteit Amsterdam (Leo Wetzels), the Center for Amerindian Studies at Leiden University (Willem Adelaar), the Centre d’Études des Langues Indigènes d’Amérique (Francesc Queixalos), the Programa de Pos-graduação em Antropologia Social of the Museu Amazônico at the Universidade Federal do Amazonas (Frantomé Pacheco), the Núcleo de Pesquisas em Ciências Sociais e Humanas of the Instituto Nacional de Pesquisas da Amazônia (Ana Carla Bruno) and the Núcleo de Estudos Indigenistas of the Universidade Federal de Pernambuco (Stella Telles). The main objective of this initiative was to create a practice of exchange and cooperation in the field of indigenous Brazilian languages, within which students and researchers are encouraged to take advantage of the expertise present in the cooperating institutions, and, by the creation of the AMAZÓNICAS colloquium, a platform for the exchange of ideas and for discussions of problems in the field of South-American indigenous languages and linguistics. The AMAZÓNICAS III colloquium was hosted by the Facultad de Ciencias Humanas of the Universidad Nacional de Colombia, sede Bogotá, from April 19 to 24, 2010. This volume grew out of a phonology session on the phonetics and phonology of laryngeal features, coordinated as part of this colloquium by Elsa Gomez-Imbert, Frantomé Pacheco, and Leo Wetzels. While most of the papers dealing with laryngeal features in the Amazonian languages were presented at the event, other contributions dealing with languages from outside the Amazon region were invited to be part of this volume. We wish to thank the permanent scientific committee of AMAZÓNICAS for advice and inspiration and the Facultad de Sciencias Humanas of the Universidad Nacional de Colombia, sede Bogotá for hosting it. We are also grateful to the institutions that, through their financial support, have made possible the organization of this colloquium and, indirectly, the publication of this volume: The Netherlands Organization for Scientific Research (NWO), the Vrije Universiteit Amsterdam (VU Amsterdam), the Institut de Recherche pour le Développement (IRD), the Embajada de Francia en Colombia, the Instituto Francés de Estudios Andinos (IFEA), the Programa ECOS-Nord FranciaColombia and the Facultad de Ciencias Humanas of the Universidad Nacional de Colombia.

CHAPTER 1

Introduction to Laryngeal Features in Languages of the Americas Heriberto Avelino, Matt Coler, and Leo Wetzels American languages have been fundamental to developing theories about languages. The description and analyses of New World languages by Boas, Sapir, and Lehmann were crucial to the advancement on modern linguistic science. The exceptional linguistic diversity attested in the Americas continues to challenge linguists to reconsider their science, and yield insight into the shared fabric of humanity: language. While several of the general properties of the sound patterns of indigenous American languages are well known, there are some properties that have been neglected for some time, in part due to the lack of an adequate theoretical apparatus and also the scarce and limited phonetic information available. One such property is the complex phenomenon labeled as laryngeal, which constitutes the subject of the present volume. This phenomenon involves quite different phonetic and phonological properties, such as tone, non-modal phonation, non-pulmonic production mechanisms (as in ejectives or implosives), stress and prosody, among others. While it is also known that other languages of the world have similar properties, what makes languages of the Americas special is that many of these properties co-exist in the phonologies of several individual languages in the continent. Furthermore, often the phonetic realization of one laryngeal articulation type effects and/or conditions the production of others. For instance, it is now known that tone is implemented by changes in the relative length and tension of the vocal folds and that the production of non-modal voice is controlled by complex dynamics that involve the whole laryngeal structures including width and tension of the vocal folds, but also the motion of cartilages (crycoid, tyrocricoid), and even epiglottic actions. An open challenge for phonetic research is how languages which have contrastive laryngeal features resolve the various articulatory and aero-acoustic conflicts. This passes the burden to the phonetician fieldworkers who must provide detailed and sophisticated instrumental evidence from such languages. Only when we have comparable data from multiple languages of the Americas, and the world, can we start formulating generalizations and

© koninklijke brill nv, leiden, ���6 | doi ��.��63/9789004303218_002

2

Avelino, Coler, and Wetzels

e­ xplanations about the phenomenon at a deeper level that permits addressing the fundamental questions: What is possible and why in language? The phonetic characteristics and possible interactions of the different types of non-modal phonation pose serious challenges for phonological theory. Since the naturalness condition requires that individual distinctive features relate to measurable physical characteristics and given the assumption, still supported by the majority of phonologists, that the set of distinctive features necessary for describing the sound patterns of human language is limited and universal, the phonetic study of laryngeal features is important for distinctive feature theory. As Stevens (1972) shows, in specific areas of the articulatory space, small articulator movements may cause large acoustic changes, whereas, in other regions, relatively large movements produce only minor acoustic effects. Whereas the relation between articulation and acoustics is language-independent, languages may enhance a contrast, i.e. produce ­additional acoustic cues to facilitate its perception. Until today, non-modal phonation has been impervious to establishing the relevant contrastive categories that account for the full array of vocalic and consonantal sounds they distinguish in the world’s sound patterns. Obviously, the precise knowledge of the way contrastive non-modal voice features are implemented in consonants and vowels cross-­linguistically is crucial for finding a language-independent phonetic definition for the relevant feature set, probably one that combines articulatory and acoustic properties in a way proposed for the [spread glottis] feature by Ridouane et al. (2009), or the hierarchical valve system of laryngeal articulations proposed by Edmondson and Esling (2006). Another area relevant for phonological theory concerns the question of which laryngeal features can co-exist in a given sound system and how lexical and derived segments are represented as concerns the segment-internal relations between subsets of features. Answers can be found through the study of phonological processes. These processes show how laryngeal features pattern together and, in so doing, bear witness to their functional unity while revealing their capacity for local and long-distance assimilation and under which conditions one or the other may occur. More urgently, formal phonology has yet to develop a feature model that considers the independence of particular laryngeal features (tone, phonation, etc.) but at the same time accounts for the interactions among them in such a way that, if a language has an orthogonal phonemic contrast, the production mechanisms of one may contradict those of the other. Languages of the Americas, then, are a prime source of empirical evidence to propose and test such phonological models. Laryngeal features are a recurrent topic in languages of the Americas. Nevertheless, the depth and extension of the description found in previous

Introduction

3

studies is varying. Laryngeal features are often mentioned in more general studies, however, in most cases, those studies are not oriented towards offering a specialized treatment of the phenomenon but discuss it in reference to other main subjects. Thus, although the scope of this book aims to represent a phenomenon present in many languages of the entire continent, not all the language families are included. This is hardly surprising given the history of linguistic research in the Americas. Athabascan, Mayan and Tucanoan are heavily represented as they have been the subject of previous research from which more detailed analyses can be further developed. We know that in North America, laryngeal features tend to split languages into one of two categories: (1) languages which more closely resemble ‘tonal languages’, and (2) ‘non-modal phonation languages’. The historical reconstructions and discussions about tone and laryngeal vowels in Athabascan are illustrative. Early works discussed vividly the tone inversion observed in Athabascan, but further research on the phonetics and phonology demonstrated that the tone reversals were connected directly to phonation contrasts (Hargus & Rice 2005; Kingston, 2005). Rice’s chapter offers an account of the laryngeal phonology of several Athabascan languages. Observing that the laryngeal contrast in obstruents, which involves aspiration and ejectivity for stops and voicing for continuants in Proto-Athabascan, no longer holds for several present-day daughter languages, Rice delineates phonetic and phonological motivations on how restructuring resulted in a shift in the status of [voice] as a distinctive feature for continuants. Shifting from the analysis of laryngeal contrasts to studies which deal with particular laryngeal segments, Tuttle’s and Hargus’ chapters complement Rice’s in this regard. In her discussion on Ahtna, one of the few Athabascan languages to retain word-final ejectives, Tuttle focuses on the realization of [t’] in the central dialect of this language. She shows that word-final ejectives induce laryngealization in preceding vowels, with marked decrease in amplitude. This effect is independent of intonational boundaries, which also may induce laryngealization. Both long and short vowels are laryngealized in this context. The impressionistic identification of laryngealized vowels with consonantal glottal stop is shown to relate to variable expressions of the glottal stop, including low-volume laryngealized sonorants. Tuttle’s findings are highly relevant to historical reconstructions of tonal development in the Athabascan family. Hargus’ chapter on final laryngealized consonants and voice quality in the Athabascan language Deg Xinag provides the results of a study which measures the effect of consonant laryngealization on the voice quality of a preceding vowel. In a way that is remarkably parallel to the related language Witsuwit’en,

4

Avelino, Coler, and Wetzels

the effect of laryngealization on voice quality, particularly pitch perturbation and pitch at midpoint of the preceding vowel, varies among speakers. Hargus suggests that unexpected midpoint pitch excursion may be a way of enhancing pitch perturbation before laryngealization. In Mesoamerica, recent evidence shows that laryngeal features tend to conflate rather than split in Otomanguean. All languages in this stock are tonal, and many of them have also non-modal phonation. In many of these languages the implementation of the double contrast is overt (e.g. Chinantecan, Amuzgan, Zapotecan), while in others there is a tendency to split the lexicon into tonewords and laryngeal-words. Other language families such as Mixe-Zoquean, Totonacan and Mayan, Uto-Aztecan, Tarascan, are overwhelmingly non tonal. In some of these languages, if laryngeal is present, it is limited to creaky and/or breathy vowels in addition to or derived from an underlying glottal stop and/ or glottal fricative. One remarkable exception is the Yucatecan branch of Mayan (Yucatec and Lacandon). This is the only group that has developed contrastive tone, in addition to laryngealized vowels. The Yucatec case has increased interest in the theoretical importance of the development of the feature in the language as well as for the historical implications of it. Other non-tonal languages, such as Mixe-Zoquean have non-modal phonation in vowels, however, as discussed in this volume, they seem to have been developing strategies in the phonetic implementation that make use of pitch patterns that override the underlying specification of non-modal phonation. In addition to Hargus’ aforementioned study of the effect of laryngealization on pitch in Deg Xinag, a number of contributions analyzed languages in which pitch is used to contrast words or grammatical forms. Tone makes for the subject of four cross-linguistic studies in the present volume: Gordon’s “Consonant-Tone Interactions”, Avelino’s “Phonetics in Phonology: A Cross-Linguistic Study of Laryngeal Contrast” and Elias-Ulloa’s “The Role of Prominent Prosodic Positions in Governing Laryngealization in Vowels: A Case Study of Two Panoan Languages”, and Hyman’s “Amazonia and the Typology of Tone Systems”. Gordon’s chapter explores the effects of different consonantal features on F0 through a phonetic analysis of four languages, along two dimensions: (i) the phonemic role of tone, contrasting Pirahã and Western Apache, which have phonemic tone, with Banawá and Hupa, which do not; and (ii) laryngeal contrasts, contrasting Pirahã and Banawá, which have a two-way contrast between voiced and voiceless obstruents, with Hupa and Western Apache, which have a three-way laryngeal distinction between voiceless unaspirated, voiceless aspirated, and ejective stops. Results of his analysis offer unique insights into the non-universality of F0 perturbations induces by consonants, suggesting that

Introduction

5

phonemic tone and consonants may be phonetically implemented in different ways with varied effects on F0. Based on the observation that many of the languages of the Americas show an orthogonal contrast between modal and nonmodal phonation in tone, Avelino provides a phonetic study that addresses the patterns of laryngeal contrast in three Mesoamerican languages: Yalálag Zapotec, Yucatec Maya, and Ocotepec Mixe, each of which contrasts modal and laryngealized vowels, but maintains a different status of lexical tone. Avelino performs a series of acoustic analyses to reveal that the phonetic implementation of laryngeal phonation and pitch depends in part on the tone inventory and phonological status of the laryngeal features in these languages. He shows that the phonetic implementation of laryngealization in the three languages depends on the phonemic status of tone. Avelino suggests that the phonetic implementation of laryngeal features, tone, and phonation differ across languages according to the phonemic status of those features. Elias-Ulloa’s study of prosodic positions, like Avelino’s analysis of laryngeal contrast, also involves Yalálag Zapotec. This chapter, however, provides a comparative analysis of the aforementioned language along with two Panoan languages (Capanahua and Shipibo). In this contribution, Elias-Ulloa studies the role of prominent prosodic positions in the avoidance of overlapping nonmodal phonation and tonal/pitch information within a vowel. As such, it highlights the similar behavior between the three languages with regards to the interaction of nonmodal phonation and tone/pitch. The author suggests that it is not the recoverability of phonological contrasts that is crucial to avoiding the overlap of nonmodal phonation and pitch/tone, as one may expect. That is, in the standard account, phonologies of languages that possess phonation and tonal contrasts (like Yalálag Zapotec) sequence modal and nonmodal phonation within laryngeal vowels as a strategy to facilitate the retrieval of contrastive tonal information by the listener. However, this chapter argues that languages like Capanahua and Shipibo, which show a similar pattern without the phonological contrast of phonation in tones, makes for an interesting counterexample. In these two languages, glottal stops coalesce with surrounding vowels unless the vowels are stressed. In other words, Yalálag Zapotec, Capanahua, and Shipibo can be characterized as sharing the goal of preserving phonological contrasts while avoiding weakening in prominent prosodic positions. Both the topics of laryngeal contrasts and the realization of laryngeal segments form the basis of Frazier’s chapter, “Pitch and Glottalization as Cues to Contrast in Yucatec Maya”. This contribution describes production experiments that illustrate that the phonemic categories of high tone and glottalized vowels are systematically distinguished by pitch, glottalization, and (to a lesser

6

Avelino, Coler, and Wetzels

extent) duration. Frazier outlines a discrimination task showing that listeners make use of all these cues in determining which vowel they heard, a fact that she takes to lend support to grammars that use bidirectional constraints with the same ranking values in both production and perception. When responding to manipulated stimuli, the author observes that listeners attended only to glottalization. This choice is language-specific (and therefore controlled by the perception grammar), such that when confronted with a degraded ­linguistic situation, listeners attend to the cue that is most likely to lead to success. Frazier concludes that the grammar of a language must encode the information on the reliability of certain phonetic dimensions in signaling contrast. Laryngeal segments and their effect on neighboring sounds also constitute a recurrent topic in the study of the indigenous languages of South America. Although a considerable number of South-American languages have relatively simple consonant systems, opposing a series of voiceless consonants to voiced or nasal consonants, many make a more elaborate use of contrastive laryngeal features. The contrastive use of non-modal phonation in consonants is recurrent in the languages of the Andes, such as the Chocoan languages (Colombia, Panama), Aymara (Chile, Bolivia, Peru), Callahuaya (Bolivia), Tehuelche (Patagonia), and others. In Nasa Yuwe (Colombia), vowels are contrastively aspirated or glottalized. Tone languages typically occur in the South-American lowlands. For an increasing number of tone languages the tone system is well-described. Hyman’s study addresses the question of how the tonal systems of South America fit into the overall scheme of tone system typology, taking his lead from a “property driven” approach to typology, i.e. “how different languages systematize the phonetic substance available to all languages”. From this perspective, the wide range of languages that have been termed “pitch-accent” languages, are better analyzed as restricted tone languages. As it turns out, for these languages, a tonal interpretation is always available. While the vast majority of Amazonian systems are described with two tone heights, descriptions differ with regard to the question of how the two heights are to be analyzed, as an “equipollent” two-tone opposition between /H/ and /L/ or as a “privative” one-tone system, either /H/ vs. ∅ or /L/ vs. ∅. Particularly striking is the special status of /L/ tone in a number of languages, which treat /L/ as the “marked” tone which often repels other contiguous /L/ tones. While the key issue with two-height systems is to determine which tone or tones are “phonologically active”, motivating the above distinction between privative and equipollent systems, other differences among the Amazonian tone languages relate to the way tone interacts with laryngeal features, specifically coda voicing and glottalization, properties which may restrict contrasts, and, in the case of glottalization, show an affiliation with the /L/ tone in several languages.

Introduction

7

Tone and laryngeal contrasts on vowels together occur in a number of Nambikwaran languages. Tone, contrastive glottalization, and allophonic laryngealization of vowels co-occur in languages of the Maku linguistic family. In this volume, of the South-American subcontinent, only the Amazonian languages are represented. Chacon gives a detailed reconstruction of laryngealized stops in ProtoTukanoan, according to which this polysynthetic language contrasted plain stops (*p, *t, *ts, *k) with laryngealized equivalents (instead of a voiced series). Chacon’s chapter also touches upon some domains related to laryngealization, including tones, vowels, and syllable boundaries. It also explores interesting genetic and areal implications. Noting that glottalic sounds typically pattern as an areal feature, the question of the original area inhabited by the ProtoTukanoan speakers and the events that triggered the sound changes which ultimately eliminated creaky voiced stops in the majority of Tukanoan languages today, Chacon provides some thought-provoking discussion. Laryngeal segments in Tukanoan is also the topic of Silva’s chapter which details a study of the phonological status of h and ʔ in root morphemes in the Eastern Tukanoan language, Desano. While in previous descriptions of this language, h and ʔ were treated as full consonantal segments, Silva gives an analysis of these segments according to which they are understood as prosodic realizations of suprasegmental laryngeal features that occur after the first vowel within the root. This analysis explains their unique distribution restriction as well as specific syllable patterns for which it would otherwise be difficult to account. Silva’s analysis has implications for Proto-Tukanoan (viz. Chacon’s chapter) as he shows that it is not straight-forward to posit the existence of h and ʔ as members of the Desano segmental inventory. While lacking a glottal stop in its phonological inventory, the Tupian language Karitiana presents several interesting phenomena having to do with the phonetic realization of glottal stops and the temporal realization of glottalic gestures, as shown by Demolin and Storto. A relatively frequent, but not systematic, phenomenon is the occurrence of vowels with a final burst in CVC words ending with a final unreleased voiceless stop or before nasals. That is, in this context, the vowel is articulated with an unexpected closing gesture of the glottis—contrary to the expected opening gesture necessary to produce a voiceless stop. Analyzing the outcome of an electroglottographic experiment, the authors reveal that voiceless stops and word-final nasals might involve closure of the glottis simultaneous with oral closure necessary to produce these consonants. This observation is key to explaining the unreleased nature of voiceless stops and nasals in word-final position, and the lenition and voicing of morpheme-final voiceless stops and nasals when followed by vowelinitial morphemes. The main factor involved in the account is the absence of

8

Avelino, Coler, and Wetzels

pressure in the oral cavity created by a closed or adducted glottis. It contributes to the unreleased character of word-final voiceless stops since a closed glottis prevents the elevation of intra-oral pressure necessary to produce the burst that would normally accompany the release of a stop. Lenition is facilitated because there is no pressure build up to enable the explosion of the voiceless stop or to allow for a well-defined release of the nasal. The authors leave open the possibility that the co-occurrence of both glottal stops and stress wordfinally may have a prosodic function. As we have attempted to summarize in this introduction, the reader will find in this volume a series of chapters devoted to the detailed analysis of individual languages of the Americas dealing with particular aspects of laryngeal features in a number of languages of the Americas. We believe that the readership will benefit from the diversity of languages represented as well as the theoretical orientations and methods used by the authors. The volume contains firsthand data and original analyses that will benefit both the formallyoriented linguists who might find new data to further test hypotheses, as well as the typologists, who might benefit from the new data to produce crosslinguistic comparisons. We hope that all linguists, be they students, researchers or specialists in American indigenous languages, find the contributions in this volume helpful for performing similar studies in other languages for which the patterns of laryngeal features are less known and so much needed. References Edmondson, J.A. & Esling, J.H. 2006. The valves of the throat and their functioning in tone, vocal register and stress: laryngoscopic case studies. Phonology, 23(02), 157–191. Hargus, Sharon and Keren Rice. 2005. Athabaskan Prosody. Amsterdam: J. Benjamins Pub. Kingston, John. 2005. The Phonetics of Athabaskan Tonogenesis. In: Hargus, Sharon and Keren Rice (eds.), Athabaskan Prosody. pp. 137–184. Ridouane, Rachid, G. Nick Clements, and Rajesh Khatiwada. 2010. LanguageIndependent Bases of Distinctive Features. John A. Goldsmith, Elisabeth Hume, and W. Leo Wetzels (eds.). Tones and Features, Phonetic and Phonological Perspectives. Walter de Gruyter GmbH & Co. KG, Berlin/Boston: 264–291.

CHAPTER 2

Overlapping Laryngeal Classes in Athabaskan Languages: Continuity and Change Keren Rice Languages of the Athabaskan family generally include both stops/affricates and continuants in their consonant inventories, generally at the same places of articulation. Stops/affricates are usually treated as distinguishing between voiceless unaspirated (plain), voiceless aspirated, and glottalized (ejective) phonation types, while continuants contrast in voicing. There is an overall similarity between the languages of the family, but at the same time the languages differ in interesting ways. In this paper, I seek to understand the laryngeal contrasts in the obstruent systems, examining these through the lens of what features are phonologically active, how these features enter into contrasts, and how the contrasts have changed over time. My specific goals are three-fold. First, building on Rice (1994), I focus on synchronic alternations involving stops/affricates and continuants in stems in a number of the languages of the family, presenting arguments that stops/ affricates are distinguished by features of aspiration and glottal constriction, while continuants are distinguished by voicing. In the stops/affricates, the voiceless unaspirated member is unmarked in terms of laryngeal features; with continuants, the voiceless member is unmarked, with evidence for the features and markedness relations in stems coming largely from alternations. Second, I extend the study to prefixes, again adducing evidence for the features noted above, with this evidence coming from historical change. Third, I examine a group of languages, those known as the Pacific Coast Athabaskan group, where there has been a general loss of alternations between voiceless and voiced continuants, and aspirated stops/affricates and voiceless continuants merge in some of the languages. I suggest a model of sound change that is based on the idea that the contrasts in the system have been reorganized to account for the changes that took place in this group. 1

Language Background

Languages of the Athabaskan (also called Dene) family are or were spoken in parts of Alaska and Canada (northern Athabaskan), the American southwest © koninklijke brill nv, leiden, ���6 | doi ��.��63/9789004303218_003

10

Rice

(Apachean), and parts of California and Oregon (Pacific Coast Athabaskan). I discuss examples from all these groups. See Appendix I for languages by group and sources. 1.1 The Consonant System Since this paper concerns consonant systems, it is useful to begin with an introduction to the overall shape of the consonant inventory in Athabaskan languages. Leer (2005:284) provides the reconstructed inventory in (1) for Proto-Athabaskan. (1) Proto-Athabaskan consonants (Leer 2005:284) unaspirated d dl dz dʒ aspirated t tɬ ts tʃ ejective t’ tɬ’ ts’ tʃ ’ voiceless ɬ s ʃ voiced l z ʒ nasals m n non-nasal sonorants w

dʒr tʃ r tʃ ’r (y)

gy ky ky’ xy ŋy y

G q q’ χ γ

I follow standard Athabaskan practice in using the term ‘stop’ to include both stops and affricates. The stops and continuants at a particular place of articulation are called a series—thus, there is the d series, the dz series and so on. Note that, in general, the sounds in the d row are unaspirated, those in the t row are aspirated, and those in the t’ row are ejective. I use the Athabaskanist symbols; it is important to keep in mind that the symbols d, etc. do not necessarily involve voicing, and t, etc. are aspirated. In terms of laryngeal contrasts, the terms in (1) are often used—for the stops unaspirated or plain, aspirated, ejective (glottalized); for the continuants voiceless, voiced. Note further that the lateral /l/ patterns together with the continuants in many languages.1

1  Other terminology is sometimes used. Minoura (1994) and Holton (2001) propose that for Tanana and Tanacross languages respectively, the distinction is one of fortis/lenis. Nater (1989) uses the terms lenis-voiced and fortis-voiceless for the stops/affricates in Tahltan. McDonough and Wood (2008) distinguish simplex and complex sounds, arguing that unaspirated stops are simplex and aspirated and ejective stops are complex, proposing that aspirated stops are phonemic affricates. See Hargus (2011) for discussion and critique. I step back from the details of the realization here, focusing more on natural classes as revealed through the phonology. I use the features [spread glottis], [constricted glottis], and [voice].

Overlapping laryngeal classes in Athabaskan languages

11

1.2 A Brief Introduction to Athabaskan Word Structure In addition to understanding the overall shape of the inventory, it is also important to understand something about word structure in Athabaskan languages. Three positions, stem-initial, stem-final, and prefix-initial, are generally distinguished; these are discussed in this work. See Li (1930) for early discussion and McDonough and Wood (2008) and Hargus (2010), among many others, for recent discussion; see Holton (2001) for discussion of continuants in these various positions in Tanacross, Minoura (1994) in Tanana, and Manker (2012) in Hän. Generally the full range of consonants, or close to it, is possible stem-­ initially, and most discussion of laryngeal contrasts in languages of this family focuses on this position. The range of consonants in stem-final and prefixinitial position is reduced, in terms of both places of articulation and laryngeal contrasts. Within the verb word, two types of prefixes are distinguished, usually called conjunct and disjunct. Simplifying, conjunct prefixes are largely inflectional and disjunct prefixes largely derivational. The conjunct prefixes allow a smaller range of initial consonants than do disjunct prefixes, and it is these prefixes that are of interest here. 1.3 Shifts: Looking Ahead While Athabaskan languages retain the general shape of the reconstructed Proto-Athabaskan consonant inventory, there are a number of types of shifts that have taken place. The most frequent shifts involves place of articulation: series (consonants at a particular place of articulation) tend to shift or merge in a uniform way. The details of these shifts and mergers, while interesting, are not germane to this work, and I do not address them here. Shifts in manner of articulation and laryngeal properties also occur. These shifts—synchronic phonological alternations and unconditioned neutralizations—form the core of this paper. 2

Theoretical Background

I make some key assumptions, and in this section I briefly outline the theory of contrast and markedness in which I frame this work. 2.1 Contrast I assume the Contrastive Hierarchy, the claim that features are assigned by language-particular hierarchies (Dresher 2009). Specifically, contrastive features are assigned by successively dividing the inventory until all phonemes

12

Rice

are distinguished. The notion of a contrastive hierarchy is found in Trubetzkoy (1939) and Jakobson and Halle (1956), among others, and it is developed in recent work by Dresher (2009) and Hall (2007, 2011), among others. Basically, languages can share in the features that they employ to distinguish sounds, but vary in the scope of the features. Features and classes that are present phonologically in a language are determined by phonological activity in that language. An example helps to clarify. Dresher (2009:15–17) suggests that /p b m/ can be grouped in two ways in terms of phonological patterning: /p, b/ vs. /m/, or /p/ vs. /b, m/. In one case the voiced stop contrasts with the voiceless stop in the other, it contrasts with the nasals. In the first case, Dresher argues that the feature [nasal] divides the inventory first, as in (2). (2) nasal > voice [nasal] – voiced /p/ –

+ /m/

/b/ +

[voice] [nasal]

p – –

b + –

m +

Alternatively, in the second case, [voice] is selected first, as in (3). (3) voice > nasal [voice] – /p/

+ [nasal] – /b/

+ /m/

[voice] [nasal]

p –

b + –

m + +

In (2), [nasal] is said to have scope over [voice]; in (3), the scope relation is reversed. Notice that while /p/ contrasts phonologically with /m/ in terms of nasality in (2), it does not in (3); in the latter /p/ is simply voiceless. Different phonological effects would be expected in these languages types. In the former,

Overlapping laryngeal classes in Athabaskan languages

13

only /b/ is voiced, while in the latter, both /b/ and /m/ are, with different patterns in processes involving voicing. In addition to the Contrastive Hierarchy, I assume the Contrastivist Hypothesis (Hall 2007), which states that the phonological component of a grammar operates in terms of contrastive features—only these features are phonologically active. The Contrastivist Hypothesis allows for a line to be drawn between phonology and phonetics. For instance, in (2) above, /m/ is not expected to be involved in any way in phonological processes involving the feature [voice], since it is not involved in a contrast for this feature. 2.2 Markedness I further assume that every feature has a marked and an unmarked value, and that markedness is language particular (e.g., Rice 2003, 2007). I indicate the marked value of a feature as [F] and the unmarked value as [∅]. Markedness is determined based on several criteria. One important criterion is phonological activity, with the marked value being phonologically active; see, for instance, Dresher (2009) and Rice (2007), among others. Marked values tend to be triggers in assimilation, while unmarked values tend to be targets. Neutralization is generally assumed to result in the unmarked; see, for instance, Trubetzkoy (1939/1969) and de Lacy (2006) for a recent articulation, as well as many others. Thus, marked values tend to be preserved in the absence of neutralization; unmarked values tend to emerge in processes such as neutralization and epenthesis. An additional characteristic that is often associated with markedness is that the unmarked may vary in its phonetic realization, limited by the marked, while the marked, assuming it does not neutralize, tends to be more stable in its realization; see, for instance, Lindblom (1990), Rice (2007), and Hall (2011) for discussion. Further, it is often assumed that acquisition follows a markedness path (e.g., Jakobson 1941), with unmarked values being acquired prior to marked ones. 3

Athabaskan Laryngeal Features

In Rice (1994) I evaluated three hypotheses to account for phonological laryngeal contrasts in a number of Athabaskan languages, what I called the Dual Mechanism Hypothesis (DMH), the Aspiration Hypothesis (AH), and the Voicing Hypothesis. The question that I raised was whether the phonological laryngeal distinctions for stops/affricates and continuants are identical, as proposed by Rice (1988) and Cook (2013:19), or whether these manner classes involve

14

Rice

different features, aspiration for stops/affricates and voicing for continuants.2 In particular, Cook proposes that the feature [voice] is redundant, with voiceless continuants being phonologically [spread glottis]. In Rice (1994), I argued for the Dual Mechanism Hypothesis for the languages considered—that stops/ affricates are distinguished phonologically by aspiration (and glottal constriction) while continuants are distinguished by voicing. This hypothesis provides an account for the laryngeal phonology of a number of languages of the family: it accounts for the fact that continuants participate in voicing alternations in many of the languages while stops/affricates do not [continuants are contrastive for [voice] while stops/affricates are not], and, in addition, it accounts for cases in which unaspirated stops and voiceless continuants pattern together, something expected under the Dual Mechanism Hypothesis but more difficult to account for with the other two hypotheses. I show in (4) representations of unaspirated and aspirated stops/affricates and voiceless and voiced continuants under the Dual Mechanism Hypothesis. Ejectives would be represented parallel to aspirated stops but with the feature [constricted glottis] rather than [spread glottis].3 (4) Dual Mechanism Hypothesis unaspirated stop/affricate manner [stop] laryngeal [∅] Voiceless continuant manner [∅] laryngeal [∅]

aspirated stop/affricate manner [stop] laryngeal [spread glottis] voiced continuant manner [∅] laryngeal [voice]

I proposed the representations in (5) under the Aspiration Hypothesis.

2  I frame the discussion in terms of the laryngneal features [voice] and [spread glottis], and I here compare the Dual Mechanism Hypothesis and the Aspiration Hypothesis. 3  Note that I use the feature [stop]. While stops are generally considered to be less marked than continuants cross-linguistically (e.g., Clements 2009), Shaw (1991) and Rice (1994) note that in Athabaskan languages, the reverse appears to hold. Evidence for this is drawn from languages that exhibit the so-called D-effect, combining a prefix known as D in the Athabaskan literature with a stem-initial glottal stop, yielding [t’], and with a stem-initial continuant, yielding a stop/affricate of the place of articulation of the continuant. The argument that [stop] is marked rests on the assumption that the marked feature is retained in the D-effect.

Overlapping laryngeal classes in Athabaskan languages

(5) Aspiration Hypothesis unaspirated stop/affricate manner [stop] laryngeal [∅] Voiceless continuant manner [∅] laryngeal [spread glottis]

15

aspirated stop/affricate manner [stop] laryngeal [spread glottis] voiced continuant manner [∅] laryngeal [∅]

The major difference between these hypotheses is as follows. Under the DMH, within the obstruents unaspirated stops and voiceless continuants might pattern together phonologically—both lack a laryngeal feature. In addition, phonological voicing alternations would affect continuants only since voicing is not contrastive for stops.4 Under the AH, on the other hand, unaspirated stops and voiced continuants might be expected to pattern together in terms of laryngeal features on the one hand, and aspirated stops and voiceless continuants on the other. Voicing alternations would involve [spread glottis], and might be expected to target both continuants and stops. In the following, I summarize the arguments for the DMH, briefly contrasting it with the AH. Before turning to the specific arguments, it is worthwhile to present the Contrastive Hierarchy for the obstruents under the DMH. I show this using the symbols D, T, S, and Z to stand for a voiceless unaspirated stop/affricate, voiceless aspirated stop/affricate, voiceless continuant, and voiced continuant, respectively; I do not consider ejectives. (6)

D, T, S, Z [stop] D, T [∅] D

[sg] [∅] T S

[∅] S, Z [voice] Z

4  See McDonough and Wood (2008) on the realization of what I call unaspirated. They show that the unaspirated stops are voiceless in Navajo (see McDonough 2003), while in Dëne Sųɬiné, they are voiced. This is compatible with the assumptions of the present paper, where the phonetic realization of the stops without a laryngeal feature can differ both between languages and, potentially, within a language as well. As discussed in the text, I am attending to phonological activity, with phonetic implementation building from this.

16

Rice

Phonological voicing alternations are not expected with stops, since [voice] is not contrastive for stops; likewise alternations in [spread glottis] are not expected for continuants since this feature is not contrastive for continuants. 3.1 The Unmarkedness of Unaspirated Stops I first address the unaspirated stops. Under both the DMH and the AH, the unaspirated stop is unmarked in terms of laryngeal features, a standard assumption in the literature (e.g., Lombardi 1994, but see Vaux and Samuels 2005 for discussion), and in this section I confirm this for the languages under consideration. First, in some languages, unaspirated stops vary in realization. For instance, Nater (1989) notes that the Tahltan voiced allophone of the stop is more common than the voiceless one. Holton (2001) remarks that Tanacross final plain stops vary between voiceless unaspirated, aspirated, and unreleased. See also note 4 on Navajo and Dëne Sųɬiné. Under the assumption that unmarked features can potentially be variable in their realization, this speaks to the unmarkedness of these sounds. In addition, Cook (2006) finds that in Dëne Sųɬiné the first stops to emerge in acquisition are unaspirated. Assuming that acquisition generally follows a markedness path (e.g., Jakobson 1941), the acquisition path suggests that voiceless unaspirated stops are the least marked of the stops. Stem-Initial Position: Voicing Alternations in Continuants but not in Stops In this section I examine stem-initial voicing alternations that affect continuants but not stops. In many Athabaskan languages, stem-initial continuants participate in voicing alternations, as illustrated in (7).5 Spelling is as in the source. The relevant segments are bolded. A hyphen preceding the stem indicates that the stem is preceded by a possessor. 3.2

(7) a. Navajo (Young and Morgan 1987:1) séí -zéí ‘sand’ heeɬ -yéél ‘pack, load, bundle’ 5  The continuant voicing illustrated in (7) applies in morphological contexts where the morpheme in question is the head of the construction and is preceded by phonological material. Many of the languages also have voicing alternations in verb stem-initial continuants, not illustrated here; in this case the alternations are, at least historically, phonologically conditioned. I assume that the alternations in examples such as (7) are triggered by the presence of a morpheme [voice] that occurs in this environment.

Overlapping laryngeal classes in Athabaskan languages



b. Tanacross (Holton 2001:400) se:x -zě:ɣʔ ‘saliva’ xeɬ -ɣě:leʔ ‘pack’ θέθ -ðέðʔ ‘skin’



c. Witwuwit’en (Hargus 2007:21) soq -zoq ‘foamy saliva’ ɬət -lət ‘smoke, smoky fire’ xwəs -wəs ‘thorn’



d. Koyukon (Jetté and Jones 2000) saakk -zaagge’ ‘spit, saliva’ 724 ɬoot -loode’ ‘scab, grime, dirt on body’ 422 haaɬ -ghaale’ ‘pack, backpack, burden’ 228 (gh=voiced uvular continuant)



e. Hän (Manker 2012:17) xew̥ -ɣèw ‘pack’



f. Jicarilla Apache (Phone, Olson, and Martinez 2007) saa -zaa ‘word’ 379 ɬíi̜ ̜́ -li̜’, -lí’̜ ‘horse’ 321

17

Stem-initial unaspirated stops, on the other hand, do not show systematic voicing alternations. See McDonough (2003) on Navajo, Gordon, Potter, Dawson, de Reuse, and Ladefoged (2001) on Western Apache, Nater (1989) on Tahltan, Gessner (1999) on Dëne Sųɬiné, and McDonough and Wood (2008) for general discussion. Under the DMH, different patterning of stops and continuants is expected: voicing is contrastive for continuants but not for stops. Under the AH, something extra would need to be said about this different patterning since aspiration is contrastive for both classes: [sg] would be lost in continuants in some environments, but not in stops in those same environments. 3.3 Voicelessness is Unmarked within Continuants The DMH provides an account for the distribution of continuants. Most stem-initial continuants participate in voicing alternations (see (7)), but some stems in most of the languages begin with a voiced continuant. Examples are given in (8).

18

Rice

(8) a. Navajo (Young and Morgan 1987) lóól ‘sound made by flute’ zas ‘snow’ (eastern Navajo) ghą́ą’́ ‘top, summit, peak’ (gh=voiced velar continuant) b. Tanacross (Holton 2001:398) zęy ‘black’ léʔe ‘don’t know’ c. Koyukon (Jetté and Jones 2000) zee ‘in vain, for nothing, aimlessly, without purpose’ 726 laadok ‘shawl, kerchief, scarf’ (Russian platók) 379 yo’ ‘louse’ 707 d. Tahltan (Nater 1989:33) dhath ‘snow’ (dh = /ð/) ghánje ‘goose’ (gh = /ɣ/) e. Jicarilla Apache (Phone, Olson, Martinez 2007) láí ‘law’ 314 zhaa ‘money’ 439 (zh = /ʒ/) zas ‘snow’ 437 If the voiced continuants have a marked feature associated with them and the voiceless ones do not, as under the DMH, the patterning of voiceless continuants alternating and voiced ones not follows under the common assumption that something with an unmarked feature will pattern as a target; something with a marked feature is likely to be stable (in the absence of neutralization; e.g., Avery and Rice 1989, de Lacy 2006). Under the AH, the voiceless continuants have a marked feature and so might be expected to not alternate; instead, the voiced continuants might be expected to gain a feature. See section 7 for additional discussion. To summarize, evidence from stem-initial voicing alternations suggests that stops/affricates and continuants have distinct laryngeal features, treated here as [spread glottis] in the former and [voice] in the latter. Stem-Final Position: Unaspirated Stops and Voiceless Continuants are a Class I now examine stem-final position. The obstruent inventory is smaller stem-finally than stem-initially. In the languages with the largest positional inventories (e.g., Ahtna), two types of stops contrast in this position—ejective 3.4

Overlapping laryngeal classes in Athabaskan languages

19

and unaspirated—and continuants also occur; there are not contrastive aspirated stops in this position. Thus, we find a two- rather than a three-way contrast in stops. Some languages have only a single stop series (e.g., Koyukon); in others, all final stops, except perhaps the dentals (d, t’), have spirantized (e.g., Navajo), and in others stops further neutralize, at least on the surface (e.g., Slavey). I focus here on languages with stops, examining the relationship between stops and continuants. In Rice (1994) I argued that the stem-final distribution of stops and continuants favours the DMH. As noted above, there is no contrast between unaspirated and aspirated stops in this position. Athabaskan tradition writes these with aspirated symbols. While not all authors discuss the phonetic value of these sounds, some do. For Tanacross, where there is no laryngeal contrast in stops word-finally, Holton (2000:25) notes that word-final stops may be unaspirated, aspirated, or unreleased, not unexpected given the lack of contrast. In Koyukon, Marlow (2000:lxviii) writes that final stops are released. Randoja (1990:22) says that only phonetic symbols are used syllable finally in Beaver, suggesting that the final stops are phonologically unaspirated and phonetically aspirated. Nater (1989) treats syllable-final stops as fortis-voiceless in Tahltan. Moore (2002:379) remarks that Kaska word-final stops are lightly aspirated or unreleased and unaspirated before a vowel-initial suffix. Hargus (2007:213) notes that only voiceless unaspirated stops are possible syllable-finally in many Athabaskan languages, including Witsuwit’en. Whatever the phonetic realization, there is general agreement that these final non-ejective stops are lexically (phonologically) unaspirated, as is evident from forms with vowel-initial suffixes, shown in the second column. When the suffix is added, the consonant is unaspirated, as is clearly indicated in the orthographies. (9) a. Ahtna (Kari 1990) ɬaets -laedze’ xatl -ghadle’

‘dirt, dust, gravel, ashes’ 275 ‘sled’ 212

b. Kaska (Moore 2002) gudech gudeji ‘s/he tells story/story’ 363 la:bá:t -la:bá:dé’ ‘mitts’ 373 t’ak -tagé’ ‘wing’ 379 c. Tahltan (Nater 1989:32) ba:t -bá:de ‘gloves’ se:k -zé:ge ‘saliva’ t’o:q -t’ó:ge ‘wart’

20

Rice

d. Beaver (Randoja 1990) gwòtl gwòdle ‘hit Object’ imperfective/optative 31 zət zidè ‘wake up’ 32 zəts zədzè ‘dance’ 31 In Tanacross (Holton 2000), the possessive suffix is a glottal stop. As mentioned above, variation is found in the realization of the stop when it is final; before the suffix, it is unaspirated. (10) Tanacross (Holton 2000:151) tth’á:k -tth’á:g’ ‘plate’ ɬεt -lε̌d’ ‘smoke’ Turning now to the relationship between stops and continuants, in many languages, final stops alternate with voiceless continuants under morphological conditions that are not pertinent to the discussion at hand. In the examples in (11), spirantization of stops occurs in the aspect called momentaneous imperfective, but not in the momentaneous perfective, where the stop remains.6 (11) Ahtna (Kari 1990) a. ’aets ‘step, momentaneous perfective’ 89 ’aes ‘step, momentaneous imperfective’ 89 b. ’aetɬ’ ’aeɬ

‘swim, momentaneous perfective’ 88 ‘swim, momentaneous imperfective’ 88

c. t’aats ‘cut, momentaneous perfective’ 341 t’aas ‘cut, momentaneous imperfective’ 341 To establish that this is spirantization and not stopping, it is important to note that there are also stems that end in a continuant lexically, with surface continuants in both the momentaneous perfective and imperfective form, differing only in voicing.

6  Historically, spirantization is argued to result from a suffix, reconstructed as *-s (Leer 1979) that occurs in the momentaneous imperfective. Spirantization is not simply conditioned by word-final position, as it is not found in nouns, where word-final stops are possible.

Overlapping laryngeal classes in Athabaskan languages

(12)

21

Ahtna (Kari 1990) k’aal ‘grind, file momentaneous perfective’ 247 k’aaɬ ‘grind, file, momentaneous imperfective’ 247 duus ‘crawl’ 159 kos ‘common cold, cough’ 245

Following Leer (1979) and Kari (1979, 1990), I argued in Rice (1994) that spirantization occurs when a non-dental stop is word-final in certain morphological situations, creating a voiceless continuant from an underlying unaspirated stop. (Dental stops do not have a continuant counterpart and fail to spirantize in many languages.) This is easily accounted for under the DMH, if unaspirated stops and voiceless continuants differ only by the value of the feature [stop], the simple neutralization of this feature to the unmarked value creates a voiceless continuant. 3.5 Some Challenges: Voiceless Continuants and Aspirated Stops? Arguments from voicing alternations and stem-final spirantization for the DMH strongly suggest that stops/affricates are distinguished by the feature [spread glottis] and continuants by the feature [voice]. However, in Rice (1994) I recognized that there are synchronic processes in various Athabaskan languages that are not well accounted for by the DMH, but follow naturally from the AH. Before turning to these, it is valuable to review the phonological processes that support the DMH, voicing alternations and final spirantization. Both are surface opaque. For instance, in (11) stops occur on the surface in wordfinal position in Ahtna and many other languages, the position of spirantization, under different morphological conditions. Voicing alternations are also opaque. As an example, while it often appears that the stem-initial voiced continuant occurs after a voiced segment and the voiceless continuant word-­ initially after a voiceless segment, there are counterexamples. In verbs in Slavey, for instance, there are circumstances under which a voiced continuant can follow [h], an environment in which a voiceless continuant is expected in this language. In addition, while the preponderance of word-initial continuants are voiceless, word-initial voiced continuants exist; see (8). I now address the spirantization of aspirated affricates to voiceless continuants, as found in some languages. (While I put it this in this way in Rice 1994, this is an aspect of a larger issue, non-alternating voiceless continuants. I review here the treatment in Rice 1994; see section 7 for further discussion.) For instance, in the Slavey variety Hare, voiceless aspirated affricates (the stops t, k do not spirantize) neutralize to voiceless continuants. Comparing two Slavey varieties, Déline, without spirantization, and Hare, with spirantization

22

Rice

(tsá / sá ‘beaver’; tɬe / ɬe ‘oil, lard’, cho / sho ‘down feathers’, where the first form is from Déline and the second from Hare), the result of the spirantization of an aspirated affricate is a continuant that fails to alternate in voicing. (13a) shows alternating continuants in both varieties; (13b) shows that the Hare nonalternating continuants are cognate with Déline aspirated affricates. (13) a. alternating continuants (Rice 1989) Hare sa -zá ‘month’ Déline sa -zá ‘month’ b. non-alternating continuant (Rice 1989) Hare sá -sá ‘beaver’ Déline tsá -tsá ‘beaver’ Hare is described as having lost aspirated affricates. Nevertheless, continuants from continuants and those from affricates have distinct phonological patterning. Two factors are of note. This neutralization is not morphologically conditioned, and aspirated affricates do occur in speech, with a continuant uniformly in citation forms. The spirantization of aspirated stops to voiceless continuants is unexpected given the DMH, but follows directly from the AH. In Rice (1994) I argued that this spirantization is distinguished from the processes discussed in sections 3.2 and 3.4 in being post-lexical. As above, these voiceless continuants do not alternate with voiced continuants in the way that clearly phonological continuants do. This is not a shift that led to a fundamental change in the underlying sound system. I argued in Rice (1994) that postlexically, voiceless continuants are redundantly aspirated, and thus can result from spirantization of aspirated affricates. (Alternatively, aspirated affricates could be part of the phonological inventory, spirantizing post-lexically.) This seems a reasonable account of Hare, where spirantization is variable and is not morphologically sensitive, properties that are usually associated with post-lexical processes. In the languages examined in Rice (1994), morphologically controlled processes treat unaspirated stops and voiceless continuants as a class, while an across-the-board process that shifts manner treats aspirated affricates and voiceless continuants as a class, something that is not surprising if the process is post-lexical and can reference redundant features. See section 7 for additional discussion.

Overlapping laryngeal classes in Athabaskan languages

23

3.6 Summary Representations that treat stops and continuants differently in terms of their laryngeal features receive support, with [voice] for continuants and [spread glottis] stops. 4 Prefixes In this section I address laryngeal shifts in prefixes. These are, for the most part, diachronic shifts, with, in some cases, variation within a language. I propose a pathway of change that accounts for the cross-family variation in manner and laryngeal features that is consistent with the DMH. 4.1 An Introduction to Prefixes Before turning to these shifts, some general discussion of prefixes is in order. I focus on conjunct prefixes, as this is the domain in which laryngeal shifts and manner shifts have occurred. Leer (2006) provides the following reconstructions for some of the conjunct prefixes in what he calls Pre-ProtoAthabaskan. (14) Pluralizer *qə- ([qh]) Lexical/derivational/classificatory prefix *qu-, *yə-, *tə-, *də-, *nə-, *sə Aspect-mode prefix *Gə-, *nə-, *sə-, *GυWhile Leer gives *G for some of the aspect-mode prefixes in Pre-ProtoAthabaskan, for Proto-Athabaskan he reconstructs a voiced uvular continuant; I do not address this prefix. Also of interest, is an ejective-initial prefix, reconstructed by Leer as *tʃ’rə- ‘indefinite human subject.’ 4.2 The Consonant Inventory of Prefixes As is often noted in the Athabaskan literature, the consonant inventory in prefixes is reduced from that in stems, with few contrasts in prefixes compared to those in the stem-initial inventory. The reconstructed prefix-initial inventory, from Leer (2006), is given in (15). This represents what Leer calls ‘Pre-Proto Athabaskan’.

24

Rice

(15) d t s n

q

tʃ’r y

ɣ

The reconstructed prefix inventory has unaspirated (d) and aspirated (t, q) stops, an ejective (tʃ’r) stop, a voiceless continuant (s), sounds that might be considered to be voiced continuants or sonorants ( y, ɣ), and a nasal (n). This inventory could be extended through the addition of direct object prefixes, adding a labial and some additional voiceless continuants. There is likely an additional voiced continuant as well. However, this set is sufficient to show how laryngeal features have shifted in the different languages.7 4.3 Laryngeal Shifts in Prefixes In discussion of prefixes I consider the developments that affect the stops, including the ejective, in terms of manner and laryngeal properties. There are not unconditioned laryngeal shifts in continuants. I work from the reconstructions, assuming them to be an appropriate starting point in considering changes. 4.3.1 Shifts within Stops Certain stop-initial prefixes undergo historical shifts of laryngeal properties without changing manner in some languages. The prefixes that shift begin with aspirated stops or with ejectives; the outcome of the shift is an unaspirated stop. This type of shift is not surprising: as discussed in section 3, unaspirated stops are unmarked in terms of laryngeal properties, while aspirated and ejective stops are marked, and prefixes often show less marked features than do stems (e.g., Urbanczyk 2011). In the following, I consider three prefixes that begin with aspirated stops, giving the reconstruction followed by the reflexes in several daughter languages. In many languages a morpheme maintains the reconstructed manner and laryngeal features. I follow the sources in using Athabaskan symbols. In particular, symbols like b, d, dz, g represent unaspirated stops, while t, ts, k represent aspirated stops. Some languages show variation; this is indicated in parentheses, and the language is listed in more than one place. 7  In general, conjunct prefixes support fewer contrasts than do stems. There are fewer contrasts at any consonantal place of articulation compared to the stem-initial inventory and prefixes have fewer vowel contrasts than do stems. For the tonal languages, tonal consonants are robust on stems, and not on prefixes; see, for instance, Kingston (2005) for discussion.

Overlapping laryngeal classes in Athabaskan languages

(16) no shift in laryngeal properties a. areal prefix *qʊ language Ahtna Kaska Dena’ina Lipan Apache

form ko (~hw, ku) ku (~gu) qe ki (~go)

b. third person plural *qə language Slavey Ahtna Dena’ina

form ke (~ge) q q



c. t- qualifier (inceptive, starting off, projecting out) language form Witsuwit’en, Ahtna, Tanacross, t Koyukon, Hupa, Wailaki, Tututni Dëne Sųɬiné t (~h)

In some languages the aspirated stop shifts to unaspirated. (17) shift from aspirated stop to unaspirated stop a. areal *qʊ language form Slavey go Kaska gu (~ku) Tsuut’ina gu Tsilhqút’ín gwe Lipan Apache go (~ki) b. third person plural subject *qə language Kaska Tsilhqút’ín Tsuut’ina Slavey Tliͅchoͅ Yatiì Apachean (except for Navajo)

form ge ge gi ge (~ke) ge go

25

26

Rice

c. t- inceptive language Slavey, Navajo, Sekani, Tsuut’ina, Apache, Mattole, Beaver, Kaska, Carrier

form d

Hargus (2007), in a phonetic study of Witsuwit’en, finds partial neutralization of /t/ and /d/ in prefixes in terms of duration, and suggests a pathway for this shift. The shift from an aspirated to an unaspirated stop is not surprising given the representations in both (4) and (5): the loss of aspiration from a stop creates an unaspirated consonant, as illustrated in (18). (18) voiceless aspirated stop → voiceless unaspirated stop [stop], [sg] [stop], [∅] Under the AH, where voiceless continuants are [spread glottis], the voiceless continuant might be expected to lose its feature; however, voiceless continuants do not shift in laryngeal quality except in some cases of context-­dependent voicing. There is one ejective-initial conjunct prefix. Just as the aspirated stop can neutralize to unaspirated, the ejective can as well. In (19) languages are listed in which ejectivity is retained, followed by languages in which it is lost in this prefix. (19) unspecified subject *chr’ a. languages that retain ejectivity language Koyukon, Ahtna, Slavey, Tsuut’ina, Chip, Tɬi̜cho̜ Yatiì, Sekani, Halfway Beaver, Ahtna, Koyukon Jicarilla Apache, Mescalero Apache, Lipan Apache, Plains Apache, Hupa Tututni Dena’ina

b. languages that lose ejectivity language Navajo Mattole Kaska

form ts’ ch’i tsr’ ch’e form ji dzi dze

Overlapping laryngeal classes in Athabaskan languages

27

Assuming a representation of ejectives that is parallel to that of aspirated stops, the loss of the feature [constricted glottis] is expected to yield an unaspirated stop. 4.3.2 Shifts in Manner of Articulation More interesting than cases where an aspirated or ejective stop shifts to an unaspirated stop are those in which an aspirated stop shifts to a voiceless continuant. Considering the morphemes introduced above, the following patterns are found. (20) shift of aspirated stop to voiceless continuant a. areal *qʊlanguage form Ahtna, Deg Xinag χu Witsuwit’en ho/w Navajo ha/ho Hupa xo Galice ho Tututni xwo Dëne Sųɬiné ho Tanacross xu Carrier wh Koyukon h (uvular) Holikachuk x

b. plural *qəlanguage Dëne Sųɬiné Koyukon Tanacross Carrier Witsuwit’en Tututni Galice

form he he (voiceless uvular) xε h he xə h

28 c. Inceptive *tə language Dëne Sųɬiné Gwich’in

Rice

form h (~t)8 h (~tʃ)

Under the DMH, we can imagine the following scenario. As with the shift to an unaspirated stop, [spread glottis] is lost, and, in addition, the manner of articulation is lost, yielding a voiceless continuant. The AH also provides an account: the aspirated consonant spirantizes simply through loss of its manner feature. This shift to a voiceless continuant does not distinguish the two hypotheses. In both cases, there is a loss of features, with a weaker sound resulting. As noted above, this tendency towards weakening in prefixes is not surprising. As discussed in section 3.1, the unaspirated stop is unmarked in terms of laryngeal features, and, based on Athabaskan evidence, the shift of a stop from aspirated to unaspirated results in the less marked laryngeal type. In addition, stops are considered to be more marked than continuants in Athabaskan languages (see note 3), with the manner shifts in prefixes also resulting in the unmarked. Under the DMH, we also see a shift to the unmarked in terms of laryngeal features; this can perhaps be interpreted as additional support for this hypothesis. There is one further reflex of aspirated stops, found in a few languages, a voiced continuant. (21) a. Areal *qʊ Tsek’ene gho [wə-] Halfway Beaver wə b. Third plural *qə Sekani ghə (gh=[ɣ]) Halfway Beaver ghə While these languages show weakening, as in the case when a voiceless continuant results, the resulting continuant is voiced. It appears that these prefixes 8  Two notes are in order about Dëne Sųɬiné. First, in this and some other languages, /d/ may lenite to [r] in prefixes under some conditions, and in some languages it is lost altogether under some conditions, following a typical lenition path (e.g., Gurevich 2011 for some discussion of lenition). In Dëne Sųɬiné, /θ/ can be realized as [h], as can some other consonants; see Cook (2004). [h] also patterns as epenthetic in this language. It is not clear if /t/ neutralizes to [h], or if this /t/ is lost, with [h] being epenthetic.

Overlapping laryngeal classes in Athabaskan languages

29

were reanalyzed as sonorant initial. This is appealing at least in the case of the areal, which is realized as [w] in both languages. I do not pursue this here. 4.4 Summary A variety of phonological processes as well as historical mergers suggest that stops/affricates have contrastive [spread glottis] (and [constricted glottis]), while continuants are distinguished by [voice]. This hypothesis accounts for continuants entering into voicing alternations that stops do not and voiceless unaspirated stops/affricates patterning together with voiceless continuants, as well as for the shifts in manner and phonation type that have occurred. 5

Pacific Coast Athabaskan

I now extend the study of laryngeal features in stems to the Pacific Coast Athabaskan (PCA) languages, languages that I did not address in Rice (1994). These languages differ in several ways from those discussed so far. First is in terms of the lexical phonological processes presented in section 3. The voicing alternations that are found in many other languages are overall absent in Pacific Coast Athabaskan.9 The D-effect that creates stops, found in a number of the languages, is not present in these languages, as the D-morpheme has a vowel. While spirantization occurs, it patterns somewhat differently than in the languages discussed so far. Continuant voicing alternations (section 3.3) provide evidence for voicing as a feature of continuants and not of stops; spirantization (section 3.4) provides evidence for the class of unaspirated stops and voiceless continuants, the D-effect (note 3) shows the relationship in terms of place of articulation between continuants and stops. Considering these morphologically controlled processes, the evidence from the lexical phonology that offers important support for the DMH is not strong in the PCA group. In addition, there are historical mergers of stops and continuants which have led to restructuring of the consonant inventories. The merger of aspirated stops and continuants is the most common, but unaspirated stops can also merge with these. Examples of prefix mergers for languages of this group are included in section 4; this section focuses on stems. 9  For instance, Mattole has xin, bi-xíné’ ‘song, his/her song’ (Li 1930:126), and Hupa has xe’q’, -xe:q’e’ ‘spit’ (Krauss 2005:88), without alternations (cf. Slavey shin, -yiné ‘song’, seh, -zégé’ ‘saliva’). There are some alternations between ɬ and l, at least in Hupa: ɬo:q’, -lo:q’e’ ‘fish’, ɬid, -lide’ ‘smoke’, ɬoh, -lo:de’ ~ɬo:we’ ‘scab’ (Krauss 2005, Golla 1996).

30

Rice

In the following discussion I first present the mergers and then the synchronic evidence for laryngeal classes in this group of languages. The major work on comparative Pacific Coast Athabaskan is Hoijer (1960), and I rely largely on this source; see Golla (2011) and Spence (2013) for recent discussion of the group. Hoijer focuses on the development of stem-initial consonants. The reconstructions that Hoijer uses are not always ones that would be used today. However, this is not a major issue for this work as I am interested in manner and laryngeal patterning. Hoijer’s reconstructions are acceptable in this respect; the reconstruction of place of articulation is rather where the thinking has changed. 5.1 Mergers of Consonants—Stem-Initial Position Pacific Coast Athabaskan is generally divided into two groups, Oregonian and Californian. Golla (2011:74) gives the inventory in (22) for Proto-Oregon Athabaskan, glottal stop and /h/ also occur. I use ipa except for the palatal glide, written as /y/. The languages in this group are (Upper) Umpqua, Tututni, Chasta Costa, Euchre Creek, Coquille, Galice, and Tolowa. (22) unaspirated aspirated ejective voiceless voiced

t th t’ m w

ts/tθ ts’/tθ’ s z

n

tʂ tɬ’ ɬ

tʂ’ ʂ ʐ

l

tʃ tʃh tʃ ’ ʃ ʒ

k k’ x ɣ

y

In terms of voicing, Golla (2011:75) remarks that the full set of voiceless and voiced continuants was preserved only in Upper Umpqua; in the other languages the contrasts between s and z, between ʂ and ʐ , and between ʃ and ʒ were lost. Golla’s (2011:81) Proto-California Athabaskan inventory is given in (23), again using ipa (with the exception of the palatal glide). Languages of this group are Hupa, Mattole, Kato, and Wailaki. (23) unaspirated aspirated ejective voiceless voiced

t th t’ p/m w

n

ts tsh ts’ s

tʃ tʃh tʃ’ ʃ

tɬ’ ɬ

y

l

ky kyh ky’

k k’ ɣ

ʔ h

31

Overlapping laryngeal classes in Athabaskan languages

The voiced continuants were largely lost; further mergers, discussed above, are later developments. Turning to the neutralizations, the charts below, from Hoijer (1960), illustrate the development of the continuants, voiceless aspirated stops, and voiceless unaspirated stops in stem-initial position. I omit the dentals d, t, t’, as they are uniformly maintained across this group. I also omit the ejectives, as they too are uniformly maintained. There is a merger of the aspirated stop *k and the voiceless continuant *x in all Pacific Coast Athabaskan, with /k/ resulting in some languages, and /x/ in others, while the unaspirated *g is maintained. (24)

Wailaki, Mattole, Kato, Galice Hupa, Euchre Creek, Coquille, Tolowa, Chasta Costa, Umpqua

*k

k x

*x

*g g g

*s, *ts, *dz are subject to merger in many languages, with variation between [s] and [ts] in some languages. While merger of the velars *k and *x occurs throughout Pacific Coast Athabaskan, Golla (1976:219) notes that the merger of continuants and affricates is a hallmark of Oregonian Athabaskan as opposed to California Athabaskan. Tututni reflexes are from Golla (1976); others are from Hoijer (1960). (25)

Hupa Wailaki, Mattole, Kato Euchre Creek, Coquille, Tolowa, Umpqua, Galice, Tututni Chasta Costa

*s/z s

θ

*ts  ts s/ts s

*dz dz

θ or s

For Hoijer’s reconstructed *dʒ series, more variation exists: some languages have three distinct classes, some merge two classes, and some merge all three.10 10  It is important to note that in using the term ‘merger’, I refer to the loss of laryngeal contrasts. In several cases, what are reconstructed as a continuant, voiceless aspirated stop, voiceless unaspirated stop contrast develops as a place contrast; thus a contrast is maintained, but is not realized in terms of laryngeal features. Each place of articulation is deserving of individual attention.

32 (26)

Rice

Hupa Wailaki, Kato Mattole Euchre Creek, Coquille Tolowa Tututni Umpqua Galice Chasta Costa

*ʃ, *ʒ W ʃ x ʃ ʃ ʃ ʃ

*tʃ tʃ w tʃ tʃ s, sr ʃ

sr s, ʃ s

*dʒ dʒ dʒ dʒ s ʃ, ʃr dʒ s ?

Finally, non-ejective lateral affricates merge with continuants, with, in general, the voiceless unaspirated lateral becoming l and the voiceless aspirated lateral becoming ɬ. To summarize, in the Pacific Coast languages, prefix-initial aspirated stops may remain, or may be restructured to unaspirated stops or to voiceless continuants, as illustrated in section 4. In the stem-initial inventory, aspirated stops and continuants merge at some places of articulation, with restructuring of the inventory. Mergers may occur between continuants and aspirated stops or between continuants, aspirated stops, and unaspirated stops; other potential mergers are not found. In addition to these mergers, the stem-initial alternations between voiced and voiceless continuants discussed in section 3 not generally found. 5.2 Stem-Final Spirantization in Mattole In this section I examine stem-final spirantization in Mattole, following Li (1930). Mattole exhibits what are called light-heavy stem pairs. The differences between the light and heavy stems are morphological in nature. The Mattole light-heavy pairs are given in (27). I use ‘h’ where Li uses a right facing apostrophe to indicate what he calls strong final breathing (1930:18). The question marks are from Li. The symbol ‘c’ is a voiceless alveopalatal continuant; ‘dj’ is a voiceless unaspirated alveopalatal affricate. (27) Mattole light/heavy pairs (18) light heavy -: -i -h -d -’ɬ -’d -h -ɣ (-h -w)

Overlapping laryngeal classes in Athabaskan languages

-’x -s -’s -c -’c -ɬ ? -’ -x (xw) -s, -ts -n -ŋ -ŋ’

33

-’g (-’xw -’gw) -ts (?) -’ts -dj -’dj -l -’l -’ -x (xw) -s, -ts -n -n -’n

Li (1930) proposes that the heavy form is basic. Ejective stops (written with preglottalization) spirantize, maintaining their laryngeal feature. The other stop-final forms have a voiceless continuant in the light form, except for the /d/-final form, where [h] occurs in the light form.11 The nasals shift in place of articulation. Li (1930:18) remarks on the aspiration, saying that the breathiness is morphologically significant, with the light form marked by aspiration which is realized so long as no other laryngeal property is present. Thus, stops spirantize, with unaspirated stops becoming voiceless continuants and ejectives retaining their glottalization. 5.3 Interim Summary In the Pacific Coast languages, we find the following. There is a loss of the contrast between aspirated stops and voiceless continuants at some places of articulation in all the languages; in some there is a further merger with unaspirated stops at some places of articulation. There is, in general, a loss of alternations between voiced and voiceless continuants, although both may remain in the inventories. In stem-final position, spirantization in Mattole results in a voiceless continuant. Ejective consonants retain their glottalization, and /d/, with no continuant counterpart, is realized as [h] (‘breathing’ in Li’s terms), as is /ɣ/.

11  As mentioned in note 4, Leer (1979) identifies a suffix in Proto-Athabaskan, *s, that is likely the source of the aspiration. In Mattole, more than spirantization occurs in the light forms, given the realization of d-final stems (unaspirated) with a final [h] in the light form.

34

Rice

5.4 An Account What evidence do we have regarding laryngeal features in the Pacific Coast group? Given that there are not generally alternations between voiced and voiceless continuants, the primary question that I address is the relationship between continuants and stops/affricates—are voiceless continuants best considered to be phonologically without a laryngeal feature (DMH) or do they have a laryngeal feature, namely aspiration (AH)? Evidence comes from the mergers that result in the restructuring of the inventories discussed in section 5.2, given the general absence of phonological processes in this group. I first review restructuring in prefixes (section 4). In Pacific Coast Athabaskan, depending on the language, historical aspirated stops are realized as aspirated or unaspirated stops or as voiceless continuants in prefix-initial position. This shift is compatible with the representation of unaspirated stops, aspirated stops, and voiceless continuants under the DMH; it is also compatible with the AH. Assuming the DMH, the aspirated stop could lose its aspiration, yielding an unaspirated stop, or lose both its aspiration and its manner, yielding a voiceless continuant. Under the AH, the loss of aspiration from an aspirated stop would yield an unaspirated stop; the loss of manner would result in a voiceless continuant. Turning to stem-final consonants, alternations between stops and continuants occur, at least in Mattole. There is no lexical contrast within stops in terms of aspiration: they are phonologically unaspirated or ejective in this position. The fact that the stops are unaspirated in the presence of a vowel-initial suffix points to this as their underlying status (see Li 1930). There is, further, no contrast between voiced and voiceless continuants in this position. Manner representations for stem-final obstruents are given (28), assuming the markedness of [stop]. (28) stop [stop] continuant [∅] Spirantization in stem-final position has two parts: loss of [stop] and the addition of [spread glottis]. As Li (1930) discusses for Mattole, aspiration in stemfinal position is morphologically conditioned, appearing in the light form (except with nasals, where there appears to be a shift in place of articulation), and this aspiration is probably best treated as a morpheme that is accommodated when phonological conditions allow (see note 6). Both the DMH and the AH would treat the underlying stem-final stops (heavy form) as voiceless and unaspirated; the former would treat the heavy-form voiceless continuants as voiceless and the latter would treat them as aspirated. In either case, the morphological aspiration is possible, yielding surface voiceless continuants.

Overlapping laryngeal classes in Athabaskan languages

35

Under the DMH, the stops and continuants in this position differ only by manner; under the AH they also are distinguished in laryngeal features, perhaps an economy argument for the DMH. Finally consider the manner and laryngeal shifts in stem-initial consonants. Alternations of consonants in this position are not generally found, unlike in the languages discussed in section 3. Thus, as noted earlier the evidence here is from the diachronic shifts. Recall the shifts: (29) Ejective consonants are retained. Historical merger of voiceless continuants, aspirated stops, and, in some languages at some places of articulation, unaspirated stops, to aspirated stops or, more often, to voiceless continuants, occurred. There is no evidence to suggest that these mergers are surface only—the inventory has been altered. In section 3, I suggested that in most Athabaskan languages, voiced and voiceless continuants group together, as do voiceless continuants and unaspirated stops, diagrammed in (30). (30) aspirated stops

voiceless continuants unaspirated stops voiced continuants

These relationships motivated the Dual Mechanism Hypothesis. In the PCA languages, the following relationships appear to hold instead. The voiced continuants do not interact in the system; the aspirated stops and voiceless continuants are subject to merger; they can also merge with unaspirated stops. (31) voiced continuant aspirated stop voiceless unaspirated stop continuant While the surface inventories are reasonably similar in the different language groups, the consonants nevertheless appear to be organized in different ways from the perspective of classes required for the phonology. Voiced continuants are essentially outside the system in Pacific Coast Athabaskan but integrated with voiceless continuants elsewhere, and voiceless continuants pattern with unaspirated stops in many languages but, for the purpose of mergers, with aspirated stops in PCA. What triggered the changes in Pacific Coast Athabaskan? Two factors might be relevant. One is the overall loss of alternations between voiced and voiceless continuants, essentially removing the voiced continuants from the

36

Rice

obstruent system. A second factor is the phonetic relationship of the voiceless continuants and the aspirated stops. As Vaux and Miller (2011) summarize, voiceless continuants in many languages pattern with aspirated stops, and Vaux and Miller suggest that, in at least some languages, voiceless continuants have the feature [spread glottis]. For Mattole, Li (1930:7) writes that the aspirated stops/affricates are pronounced with a velar spirantal glide in the case of t and k, and with a palatal spirantal glide in the case of tc; ts has, he notes, very weak or no aspiration. Golla (1996:367) notes strong aspiration on Hupa stops/affricates. McDonough and Wood (2008) focus on the fricative release of aspirated stops in a number of Athabaskan languages. The aspiration associated with voiceless continuants coupled with the strong continuant release at some places of articulation in these languages may have been factors that allowed the grouping of voiceless continuants with aspirated stops once the continuants were no longer held in place by voicing alternations. More, of course, must be said; for instance, there are no mergers with t and d even if their release is affricated, but I leave this aside for later work. 6

A Sketch of Language Change

I have proposed that distinct laryngeal features for stops and continuants are required to account for alternations and distribution of obstruents in many Athabaskan languages. In general, voiceless continuants group with unaspirated stops; in PCA, on the other hand, at least for the consonant shifts, voiceless continuants form a class with aspirated stops. In this section I outline a model for assigning contrastive features that allows for these two different types of systems. Here I return to the model introduced in section 2. Recall that Dresher (2009), building on insights of Trubetzkoy (1939/1969) and Jakobson and Halle (1956), among others, argues that phonological natural classes can differ from language to language, even when those languages share the same surface inventory. Dresher (2009) proposes that the features and classes that are present phonologically for a language are determined by phonological activity in that language. Features are not introduced all at once, but successively, with a continuous division of the inventory until each segment is uniquely specified. Dresher (2009) further argues that the order in which features are introduced is not fixed cross-linguistically, with variation allowed in the feature hierarchy resulting in different phonological natural classes. What might this perspective offer in the Athabaskan case? Assuming, as argued in this work, that aspiration is contrastive within the stops and voicing in the continuants, then the overall organization of the features is likely as

Overlapping laryngeal classes in Athabaskan languages

37

follows. Manner contrasts within obstruents distinguish stops from continuants; then laryngeal contrasts are made within these classes. I illustrate this with respect to the *dz series in (32), using the kind of diagram developed by Dresher (2009) and introduced in section 2. (32) manner > laryngeal ts, dz, s, z [stop] ts, dz [sg] ts

[∅] s, z

[∅] [voice] z dz

[∅] s

Stops exhibit a contrast between aspirated and unaspirated (and ejective; not shown here), while continuants show a contrast between voiced and voiceless. The following types of interactions can be predicted (see Dresher 2009 and Oxford, 2015, for details of restrictions on processes and phonological change). (33) interaction between aspirated and unaspirated stops: differ by [sg] interaction between voiced and voiceless continuants: differ by [voice] interaction between unaspirated stops and voiceless continuants: differ by [stop] These various types of interactions occur in the languages discussed in sections 3 and 4. With respect to aspirated and unaspirated stops, stem-final position does not licence both of these, but only unaspirated stops. With respect to continuants, voicing alternations occur. Finally, unaspirated stops alternate with voiceless continuants. While aspirated stops also neutralize to voiceless continuants, I argued in section 3.5 that this process has the hallmarks of a post-lexical process, and thus it need not be restricted to only contrastive features. Turning to the patterning of stem-initial consonants in the Pacific Coast languages, recall that in this group, voiced continuants do not alternate with voiceless continuants. Rather at some places of articulation mergers occurred between voiceless continuants and aspirated stops and, in some cases, with unaspirated stops as well. This suggests that there is not a phonological natural class of voiced and voiceless continuants (except, perhaps, in the laterals—see note 9; I set these aside). In terms of the Contrastive Hierarchy, this lack of interaction of voiced continuants with the other obstruents suggests that [voice] is ordered high in the hierarchy, with the voiced continuants

38

Rice

essentially patterning as sonorants, as in (34) (see, for instance, Rice 1993). I illustrate with the velar consonants. (34)

k, g, k’, x, ɣ [voice] ɣ

[∅] k g k’ x

I now consider the patterning within the segments on the right branch in (34). The voiceless continuant and the aspirated stop merge in the Pacific Coast languages; the particular outcome (k, x) depends on language/variety. I omit the ejectives as they are not implicated in any sound changes. The proposed starting point, after the shift in the position of [voice], is shown in (35). (35) manner > aspiration k, g, x [stop] [sg] k

[∅] g

[∅] x

Here the hierarchy of (32) is maintained except for the position of [voice]. If the hypothesis that the continuants were reanalyzed as aspirated in PCA is correct, the following reorganization occurred, with laryngeal features having scope over manner and, in addition, the voiceless continuant being analyzed as aspirated. (36) aspiration > manner kxg [sg] [stop] k

[∅] x

[∅] g

At this stage, the aspirated stops and continuants, a natural class, merged. Following this merger, the contrasts between the non-ejective obstruents in

Overlapping laryngeal classes in Athabaskan languages

39

the system could be represented as in (37), with a simple contrast between aspirated and non-aspirated sounds. (37) non-ejective [sg]

[∅]

The aspirated sound is realized as [k] or [x], depending on language. Such variation is not surprising, given the absence of a contrast between stop and continuant, as discussed in section 2. In at least some languages, the non-aspirated velar is usually realized as [g] but can also be a continuant [ɣ]; this variation too speaks to the lack of manner contrasts. All the languages show a merger of velars, and, as discussed above, the Oregonian languages take the mergers a step further at some places of articulation. Again, ejectives remain distinct. Beginning with the hierarchy in (37), the non-ejectives are subject to merger, with, in general, a voiceless continuant resulting: the contrast at these places of articulation is simply between ejective and non-ejective. Why might the voiceless continuant be the typical result of the merger? Voicelessness is not surprising, given that it is generally considered to be unmarked. Cross-linguistically continuancy is more surprising, as continuancy is generally thought to be marked (e.g., Clements 2003, 2009); it is perhaps due to the Athabaskan languages showing patterning that stops are more marked than fricatives (see note 3) as well as to the phonetically very strong aspiration associated with the aspirated stops. To summarize, this brief sketch treats the manner and laryngeal shifts in PCA as follows. We begin with a primary distinction between stops and continuants, with different laryngeal contrasts within these classes. The voicing contrast within the continuants was lost, with the feature [voice] relocated in the hierarchy. The strong phonetic resemblance between voiceless continuants and aspirated stops might have been the motivation for a restructuring, with these as a class; these then are subject to merger. Finally, non-ejective obstruents became subject to merger, resulting in a simple contrast between ejective and non-ejective obstruents, with the latter generally realized as continuants. This analysis is summarized below, where the basic premise that underlies the analysis is that neutralization affects natural classes of sounds. I omit ejectives from these diagrams; given the retention of ejective consonants, their laryngeal feature would be hierarchically high.

40

Rice

The first stage is shown in (38). (38)

manner > laryngeal [stop]

[∅]

[sg] [∅] [voice] [∅] With the loss of alternations between voiced and voiceless continuants, the voiced and voiceless continuants no longer formed a class, as in (39). (39) voice > manner > spread glottis [voice]

[∅] [stop] [∅] [sg] [∅]

A further change shifted the positions of [stop] and [spread glottis], creating a natural class of aspirated stops and voiceless continuants. (40)

voice > spread glottis > manner [voice]

[∅] [sg]

[∅]

[stop] [∅] With the loss of the manner contrast, a situation arose in which the aspirated and unaspirated sounds form a class. (41) voice > spread glottis [voice]

[∅] [sg]

[∅]

Overlapping laryngeal classes in Athabaskan languages

41

Finally, in some cases, the contrast in aspiration was lost, resulting in the following, with sonorants and simply an ejective/non-ejective contrast in the obstruents (ejectives are not included here). (42)

[voice]

[∅]

While I have not provided a full analysis, it appears that reordering of the features in a contrastive hierarchy provides an account of how the changes might have occurred, with neutralization of sisters, reanalysis of sounds, and reranking of features with respect to each other; see Oxford (2015) for detailed discussion of the principles of language change under the theoretical assumptions outlined here. Many questions arise, and I enumerate a few here. First, as noted earlier, why do dentals remain intact? This could be due to the absence of continuant counterparts to the dental stops. Second, why do the mergers begin with velars—the merger found throughout the area? Third, why might the laterals continue to show voicing alternations, while these are lost at other places of articulation? One might also ask what would motivate the reduction in inventory size in the PCA group in general and the Oregon group in particular, another interesting topic to pursue. 7

A Return to Voiceless Continuant/Voiceless Aspirated Stop Mergers

In this section I return to an issue addressed in section 3.3, the merger of voiceless continuants and aspirated stops. In that section, I summarized the proposal of Rice (1994), that this represents a surface merger that has not led to restructuring of the inventory. In this section I slightly shift the focus, considering languages with the following set of characteristics: (43) Alternating voiced/voiceless continuants Non-alternating voiced continuants Non-alternating voiceless continuants The Hare variety of Slavey discussed in section 3.5 is one such language. In this variety, we find the following types of patterning of continuants.

42

Rice

(44) a. alternating continuants sa -zá ‘month, sun’ b. non-alternating voiceless continuants sá -sá ‘beaver’ c. non-alternating voiced continuants le -lé ‘smoke’ zoͅ ‘only’ Southern Tutchone is similar. Data is from Krauss (2005). (45) a. alternating continuants xé -yè’ ‘grease’ 82 b. non-alternating voiceless continuants sǝ́ r sǝ̀ r’ ‘dry wood’ 91 sà’ sà’ ‘beaver’ 94 c. non-alternating voiced continuants yà’ -yà’ ‘louse’ 94 zén- ‘day’ 82 In Hare, the non-alternating voiceless continuants derive from aspirated affricates; the non-alternating voiced continuants have as one source the neutralization of ɬ and wh to l and w respectively. In Southern Tutchone, the nonalternating voiceless continuants arise from aspirated affricates and the nonalternating voiced continuants have unaspirated affricates as one source.12 Nater (1989) identifies alternating and non-alternating voiceless continuants in Tahltan. (46) a. alternating continuants (Nater 1989:32) seɬ -zéɬé ‘hook’ xən -ghǝ́ ne ‘song’ (gh = [ɣ]) b. non-alternating continuants (Nater 1989:33) tha: -tháe ‘sand’ (th = [θ]) xas -xáse ‘scar’ thon’ -thósn’e ‘star’ 12  In both languages aspirated affricates spirantize, but aspirated stops do not: Hare te̜ ‘ice’, -ké ‘foot’; Southern Tutchone tə́n ‘ice’, -kè’ ‘foot’.

Overlapping laryngeal classes in Athabaskan languages

43

Witsuwit’en (Hargus 2007) also has these three types of sounds. (47) a. alternating continuants (Hargus 2007:22) soq -zoq ‘foamy saliva’ χəy -ghih ‘root’ b. non-alternating voiceless continuants (Hargus 2007:23) χas -χas ‘fireweed’ c. non-alternating voiced continuants (Hargus 2007:23) yin ‘moss’ yəs ‘snow’ luts ‘red elderberry’ (loan) Hargus (2007:218) argues for the following featural analysis for Witsuwit’en: (48) t, ts th, tsh t’, ts’ S z s

[spread glottis] +

[constricted glottis]

[voice]

+ + –

[lowered larynx] + – – – + –

The symbol /S/ is used for the alternating continuant; /z/ and /s/ are nonalternating. [Lowered Larynx] is introduced by Hargus to account for various effects of consonants on vowels; these are beyond the scope of this work, and Hargus (2007:218) concludes that ‘Witsuwit’en is only one step removed from the basic Athabaskan plan for laryngeal features.’ Below I sketch a possible proposal for these systems, using the model proposed by Hargus (2007). Using the features [voice] and [spread glottis], the following representations are appropriate. (49)

alternating continuants non-alternating voiceless continuants non-alternating voiced continuants

s/z s z

[sg] ✓

[voice] ✓

Given these representations, only the continuant without a laryngeal feature is expected to alternate under the assumptions laid out in section 2 and

44

Rice

assuming that only a single laryngeal feature can be associated with a segment in these languages. While restructuring in this way is a possibility, it is interesting that in at least Hare and Southern Tutchone, non-alternating voiceless continuants arise primarily from aspirated affricates. In both languages, the affricate remains a possible realization. In normal speech in Hare, affricates sometimes occur, although not in citation forms, as discussed earlier. In Southern Tutchone, there is also variation in spirantization, apparently depending on variety. (50) Southern Tutchone (http://www.ynlc.ca/languages/st/st.html) Ashihik dialect Tàa’an dialect ätthä’n äthän ‘meat’ ätsia äsì ‘grandfather’ In general, Ashihik retains the contrast between an aspirated affricate and a voiceless continuant, while Tàa’an could be said to maintain the contrast phonologically, as evidenced by the presence or absence of voicing alternations, but lose it phonetically. The variation, coupled with exposure to other varieties, might suggest that the phonological inventory includes aspirated affricates, and these spirantize phonetically; if there has been restructuring, the representations in (49) seem appropriate (although they may pose challenges to the Contrastive Hierarchy). In Southern Tutchone, the unaspirated affricate (but not the dental or velar stop) may also spirantize to the non-alternating voiced continuant (e.g. dzenu ‘day’ in Ashihik, zenu in Tàa’an). This result is surprising, given the phonological representation of the unaspirated affricate in (4). However, and this is an unknown, if the phonological stops that lack a laryngeal feature are phonetically voiced, the resulting voiced continuant is not unexpected: it is likely that this process originated as post-lexical, perhaps leading later to phonological restructuring. 8

General Conclusion

I have examined the natural classes of stops and continuants with respect to laryngeal features in several Athabaskan languages. While in many of the languages, there is a major split between stops and continuants in features required, this is not true in all of the languages. In Pacific Coast Athabaskan, the major split appears to be between voiced and voiceless sounds, and within the voiceless sounds between ejective on the one hand and aspirated and unaspirated on the other, with aspirated merging, leading to restructuring of the inventory and, in general, an indeterminacy in establishing features

Overlapping laryngeal classes in Athabaskan languages

45

based on phonological patterning. In other languages where stops and continuants merge, the presence of alternating and non-alternating continuants provides a clue to the structure of the system and also points to the different patterning of affricates, which spirantize, and stops, which do not. McDonough and Wood (2008) note that stability in the phonetic realization of laryngeal contrasts is a hallmark of Athabaskan languages (see, however, Hargus 2011). Phonological stability is found across much of the family, with an invariant core contrastive hierarchy for manner and laryngeal features. However, the phonological systems of Pacific Coast Athabaskan appear to be different to a greater or lesser degree, depending on language and place of articulation. I proposed the following pathway of development of PCA. Continuant voicing alternations were lost, with in a shift of the position of [voice] in the Contrastive Hierarchy, above manner; manner and laryngeal features shifted in their ranking; at a later stage, some languages essentially contrast only aspirated/unaspirated in the non-ejective obstruents and, finally, at some places of articulation, some languages simply have a contrast within obstruents of ejective/non-ejective. Across the family, much the same surface inventory is found, but, if the arguments in this paper hold, that surface inventory can be organized in different ways phonologically. Apparently similar surface inventories can have different phonological properties; apparently different surface inventories can have similar phonological properties; and different surface inventories can show phonological different properties. To illustrate these, consider the following comparisons. Using a single surface place of articulation series of consonants to illustrate, Li (1930) gives the following inventory for Mattole, using IPA: tʃ tʃh tʃ ’ ʃ. He treats the palatal glide /j/ as a semivowel, outside the series. In Slavey, this series would be tʃ tʃh tʃ’ ʃ j, with /j/ as part of the series. The treatment of the voiced continuants as obstruents (Slavey) or sonorants (Mattole) differentiates these languages, with ʃ and j alternating in Slavey but not in Mattole. Thus, the similar surface inventories are analyzed as different phonologically. In other cases, inventories are different on the surface but are at some level the same. For instance, the closely related Hare and Déline varieties of Slavey differ in that Déline has surface aspirated affricates that Hare largely lacks, with voiceless continuants instead; Southern Tutchone exhibits a similar difference between varieties. In terms of voicing alternations, the Hare continuants that are cognate with Déline continuants and the Hare continuants that are cognate with Déline affricates pattern differently. This difference between varieties is superficial. Finally, the inventories in two languages may both look different on the surface and pattern differently phonologically. For instance, in many Pacific Coast

46

Rice

languages, more than different analysis of a similar surface system is involved. Mattole has the series (in IPA) tsh/s ts’, where Slavey has ts tsh ts’ s z. Slavey ts, tsh, s, z are realized as tsh/s in Mattole, and there is no indication that the Mattole inventory should be treated in any different way. I have only briefly sketched an analysis of change in Athabaskn languages based on the Contrastive Hierarchy. This study is worth pursuing, potentially providing insight into the phonology of the languages and also into sites where variation is likely to occur. References Avery, Peter and Keren Rice. 1989. Segment structure and coronal underspecification. Phonology, 6. 179–200. Clements, G.N. 2003. Feature economy in sound systems. Phonology 20. 287–333. ———. 2009. The role of features in phonological inventories. In Contemporary views on architecture and representation in phonology, ed. Eric Raimy and Charles E. Cairns, 19–68. Cambridge, Massachusetts: MIT Press. Cook, Eung-Do. 1984. A Sarcee grammar. Vancouver: UBC Press. ———. 2004. A grammar of Dëne Sųɬiné (Chipewyan). Winnipeg: Algonquian and Iroquoian Linguistics, Memoir 17 (Special Athabaskan Number). ———. 2006. The pattern of consonantal acquisition and change in Chipewyan (Dëne Suliné). International Journal of American Linguistics 72. 236–263. ———. 2013. A Tsilhqút’ín grammar. Vancouver: UBC Press. de Lacy, Paul. 2006. Markedness. Reduction and preservation in phonology. Cambridge: Cambridge University Press. Dresher, B. Elan. 2009. The Contrastive Hierarchy in phonology. Cambridge: Cambridge University Press. Gessner, Suzanne. 1999. Laryngeal processes in Chipewyan and other Athapaskan languages. Vancouver: University of British Columbia Masters thesis. Golla, Victor. 1976. Tututni (Oregon Athapaskan). International Journal of American Linguistics 42. 217–227. ———. 1977. A note on Hupa verb stems. International Journal of American Linguistics 43. 355–358. ———. 1996a. Hupa. Handbook of North American Indians, volume 17, ed. Ives Goddard. 364–389. Washington: Smithsonian Institution. ——— (compiler). 1996b. Hupa language dictionary, second edition/Na:tinixwe Mixine:whe’. Hoopa, California: Hoopa Valley Tribal Council. ———. 2011. California Indian languages. Berkeley: University of California Press.

Overlapping laryngeal classes in Athabaskan languages

47

Gordon, Matthew. 1995. The phonetic structures of Hupa. Fieldwork Studies of targeted languages IV (UCLA Working Papers in Phonetics, 93.), ed. Ian Maddieson, 1–24. Los Angeles: UCLA Department of Linguistics Phonetics Lab. ———. 2001. Laryngeal timing and correspondence in Hupa. In Adam Albright and Taehong Cho (editors). UCLA Working Papers in Linguistics, volume 7, Papers in Phonology 5, ed. by Adam Albright and Taehong Cho, 1–70. Gordon, Matthew, Brian Potter, John Dawson, Willem de Reuse, and Peter Ladefoged. 2001. Phonetic structures of Western Apache. International Journal of American Linguistics 67. 415–48. Gurevich, Naomi. 2011. Lenition. The Blackwell companion to phonology, ed. by Marc van Oostendorp, Colin Ewen, Elizabeth Hume, and Keren Rice. 1559–1575. Malden, Massachusetts: Wiley-Blackwell. Hall, Daniel Currie. 2007. The role and representation of contrast in phonological theory. PhD diss., University of Toronto. ———. 2011. Phonological contrast and its phonetic enhancement: Dispersedness without dispersion. Phonology 28. 1–54. Hargus, Sharon. 1988. The lexical phonology of Sekani. New York: Garland. ———. 2007. Witsuwit’en grammar: phonetics, phonology and morphology. Vancouver: UBC Press. ———. 2008. Deg Xinag lateral affricates: phonetic and historical perspectives. Poster presented at the annual meeting of Society for the Study of the Indigenous Languages of the Americas, Chicago. ———. 2010. Athabaskan phonetics and phonology. Language and Linguistics Compass 4/10. 1019–1040. ———. 2011. Stop and affricate features in Athabaskan (in general) and Deg Xinag (in particular). Paper presented at WECOL, Vancouver, British Columbia. Hoijer, Harry. 1960. Athapaskan languages of the Pacific coast. In Stanley Diamond (editor). Culture in history. Essays in honor of Paul Radin, ed. by Stanley Diamond. 960–976. New York: Columbia University Press. Holton, Gary. 2000. The phonology and morphology of the Tanacross Athabaskan language. PhD diss., University of California Santa Barbara. ———. 2001. Fortis and lenis continuants in Tanacross Athapaskan. International Journal of American Linguistics 67. 396–414. Jakobson, Roman. 1941. Child language, aphasia, and phonological universals. The Hague: Mouton. Jakobson, Roman and Morris Halle. 1956. Fundamentals of language. The Hague: Mouton. Jetté, Jules and Eliza Jones, with James Kari (editor). 2000. Koyukon Athabaskan dictionary. Fairbanks: Alaska Native Language Center.

48

Rice

Jung, Dagmar. 1999. The dynamics of polysynthetic morphology: Person and number marking in Athabaskan. PhD diss., University of New Mexico. Kari, James. 1979. Athabaskan verb theme categories: Ahtna. Fairbanks: Alaska Native Language Center, University of Alaska Fairbanks. ———. 1990. Ahtna Athabaskan dictionary. Fairbanks: Alaska Native Language Center, University of Alaska Fairbanks. ———. 2007. Dena’ina topical dictionary. Fairbanks: Alaska Native Language Center. Kingston, John. 2005. The phonetics of Athabaskan tonogenesis. Athabaskan prosody, ed. by Sharon Hargus and Keren Rice. 137–184. Amsterdam: John Benjamins. Krauss, Michael. 2005. Athabaskan prosody, ed. by Sharon Hargus and Keren Rice, 55–136. Amsterdam, Philadelphia: John Benjamins. Leer, Jeff. 1979. Proto-Athabaskan verb stem variation. Fairbanks: Alaska Native Language Center, University of Alaska Fairbanks. ———. 2005. How stress shapes the stem-suffix complex. (editors) Athabaskan prosody, ed. by Sharon Hargus and Keren Rice, 278–318. Amsterdam: John Benjamins. ———. 2006. Na-Dene Languages. The Encyclopedia of Language and Linguistics, ed. by R.E. Asher and J.M.Y. Simpson. 2665–2666. Oxford: Pergamon Press. Li, Fang-Kuei. 1930. Mattole: An Athabaskan language. Chicago: University of Chicago Press. Lindblom, Bjorn. 1990. Explaining phonetic variation: a sketch of the H&H theory. Speech production and speech modeling, ed. by William J. Hardcastle and Alain Marchal. 403–439. Dordrecht: Kluwer Academic Publishers. Lombardi, Linda. 1994. Laryngeal features and laryngeal neutralization. New York: Garland. Manker, Jonathan T. 2012. An acoustic study of stem prominence in Hän Athabascan. Fairbanks, Alaska: University of Alaska Fairbanks M.A. thesis. Marlow, Patrick. 2000. Koyukon sound system and orthography. Koyukon Athabaskan dictionary, ed. by Jules Jetté and Eliza Jones. lxvi–lxxi. Fairbanks: Alaska Native Language Center. McDonough, Joyce. 2003. The Navajo sound system. Dordrecht: Kluwer Academic Publishers. McDonough, Joyce and Valerie Wood. 2008. The stop contrasts of the Athabaskan languages. Journal of Phonetics 36. 427–49. Minoura, Nobukatsu. 1994. A comparative phonology of the Upper Tanana Athabaskan dialects. Languages of the North Pacific Rim. Hokkaido University Publications in Linguistics 7, ed. by Osahito Miyaoka. 159–196. Sapporo: Department of Linguistics, Faculty of Letters, Hokkaido University. Moore, Patrick. 2002. Point of view in Kaska historical narratives. PhD diss., Indiana University. Nater, Hank. 1989. Some comments on the phonology of Tahltan. International Journal of American Linguistics 55. 25–42.

Overlapping laryngeal classes in Athabaskan languages

49

Oxford, William. 2015. Patterns of contrast in phonological change: Evidence from Algonquian vowel systems. Language 91–2. 308–357. Phone, Wilhelmina, Maureen Olson, and Matilda Martinez. 2007. Dictionary of Jicarilla Apache/Abáachi Mizaa Iɬkee’ Siijai, ed. by Melissa Axelrod, Jule Gómez de García, and Jordan Lachler. Albuquerque: University of New Mexico Press. Poser, William J. 2009. The Carrier language. A brief introduction. College of New Caledonia Press. Randoja, Tiina. 1990. The phonology and morphology of Halfway River Beaver. PhD diss., University of Ottawa. Rice, Keren. 1988. Continuant voicing in Slave (Northern Athapaskan): the cyclic application of default rules. Theoretical morphology, ed. by Michael Hammond and Michael Noonan. 371–388. New York: Academic Press. Rice, Keren. 1989. A grammar of Slave. Berlin: Mouton de Gruyter. ———. 1993. A reexamination of the feature [sonorant]: the status of ‘sonorant obstruents’. Language 69. 308–344. ———. 1994. Laryngeal features in Athapaskan languages. Phonology 11. 107–47. ———. 2003. Featural markedness in phonology: Variation. The second Glot International state-of-the-article book: The latest in linguistics, ed. by Lisa Cheng and Rint Sybesma. 387–427. Berlin: Mouton de Gruyter. ———. 2007. Markedness in phonology. The Cambridge Handbook of Phonology, ed. by Paul de Lacy. 79–97. Cambridge: Cambridge University Press. Saxon, Leslie and Mary Siemens. 1996. Tłıįcho Yatıì Enįhtłè/A Dogrib Dictionary. Dogrib Divisional Board of Education, Rae-Edzo, Northwest Territories. Shaw, Patricia. 1991. Consonant harmony systems: the special status of coronal harmony. The special status of coronals: internal and external evidence, ed. by Carole Paradis and Jean-François Prunet. 125–157. San Diego: Academic Press. Spence, Justin David. 2013. Language change, contact, and koineization in Pacific Coast Athabaskan. PhD diss., University of California, Berkeley. Thompson, Chad. 1993. The areal prefix hu- in Koyukon Athapaskan. International Journal of American Linguistics 59. 315–333. Trubetzkoy, Nikolai. 1939. Grundzüge der Phonologie. Göttingen: van der Hoeck and Ruprecht. Translated 1969 by Christiane A.M. Baltaxe as Principles of phonology. Berkeley and Los Angeles: University of California Press. Urbanczyk, Suzanne. 2011. Root-affix asymmetries. The Blackwell companion to phonology, ed. by Marc van Oostendorp, Colin Ewen, Elizabeth Hume, and Keren Rice. 2490–2515. Malden, Massachusetts: Wiley-Blackwell. Vaux, Bert and Brett Miller. 2011. The representation of fricatives. The Blackwell companion to phonology, ed. by Marc van Oostendorp, Colin Ewen, Elizabeth Hume, and Keren Rice. 669–693. Malden, Massachusetts: Wiley-Blackwell. Young, Robert W. and William Morgan. 1987. The Navajo language: a grammar and colloquial dictionary. Albuquerque: University of New Mexico Press.

50

Rice

Appendix: Data Sources

Apachean Jicarilla Apache Mescalero Apache Plains Apache Western Apache Navajo

Jung 1999; Phone, Olson, Martinez 2007 Jung 1999 Jung 1999 Jung 1999 Young and Morgan 1987

Pacific Coast Hupa Mattole Galice Tututni Wailaki

Golla 1977, 1996a, b; Gordon 1995, 2001 Li 1930 Hoijer 1960 Golla 1976 Hoijer 1960

Northern Tsuut’ina Dëne Sųɬiné Slavey Tłįchǫ Yatiì Sekani Halfway Beaver Ahtna Dena’ina Koyukon Holikachuk Upper Tanana Witsuwit’en Carrier Tsilhqút’ín Kaska Tanacross Deg Xinag Hän

Cook 1984 Cook 2004 Rice 1989 Saxon and Siemens 1996 Hargus 1988 Randoja 1990 Kari 1990 Kari 2007 Jetté and Jones 2000 Thompson 1993 Minoura 1994 Hargus 2007 Poser 2009 Cook 2013 Moore 2002 Holton 2000, 2001 Hargus 2008 Manker 2012

CHAPTER 3

Stem-Final Ejectives in Ahtna Athabascan Siri G. Tuttle Introduction Ahtna, an Alaskan Athabascan language spoken in the south-central portion of the state, is one of few members of the language family to retain wordfinal ejectives. This study addresses the acoustic realization of /tʼ/ and /ts’/ in Ahtna, with special attention to the Central dialect. Related topics include the realization of glottal stop in this language, and the historical development of Athabascan tone. Kari (1990, 24) claims that final [tʼ] is realized in four different ways in Ahtna dialects. In Table 3.1, the words cʼezetʼ ‘liverʼ and getsʼ ʻmittensʼ are shown as he predicts their phonetics. Table 3.1

Orthography and putative ipa transliteration for Ahtna realizations of ejective-final stems cʼezetʼ ‘liverʼ and getsʼ ʻmittensʼ (Kari 1990)

Orthography ipa Orthography ipa

Western

Central

Lower

Mentasta

cʼezetʼ [cʼɛzɛtʼ] getsʼ [cɛtsʼ]

cʼezeʼt [cʼɛzɛʔt] geʼts [cɛʔts], [cɛʔs]

cʼezeʼł [cʼɛzɛʔɬ] geʼs [cɛʔs]

cʼezet [cʼɛzɛt] gets [cɛts]

The goal of this paper is to demonstrate the acoustic realization of Central Ahtna stem-final ⟨ʼt⟩ and ⟨ʼts⟩. In this dialect, is the element symbolized by the apostrophe in the orthography actually something that could be written as in the ipa translation in Table 3.1? If not, then what is it? Leer (1979, 7) mentions “preglottalization” in his discussion of stem-final ejectives in modern Athabascan languages, but is not specific as to its phonetic realization, or whether its realization seems consonantal or vocalic.

© koninklijke brill nv, leiden, ���6 | doi ��.��63/9789004303218_004

52

Tuttle

This paper will show, based on study and analysis of field and archival data, that word-final ejectives in the Central dialect clearly induce laryngealization in preceding vowels, with marked decrease in amplitude, but the full closure of canonical glottal stop is not seen as the spellings in Table 3.1 might ­suggest. Both long and short vowels are laryngealized in this context, for about the same percentage of the vowel. The identification of laryngealized vowels with consonantal glottal stop is shown to relate to variable expressions of glottal stop, including low-volume laryngealized sonorants. These findings are highly relevant to historical reconstructions of tonal development in the Athabascan family, e.g. Leer (1979), Krauss (2005), Kingston (2005). 1

Ahtna Dialects and Data

The Ahtna language is spoken in south central Alaska, mainly along the Copper River, extending along the southern side of the Alaska Range and extending to Cantwell, near Denali. The language is not being passed on, and is spoken mainly by elders, though revitalization efforts are ongoing. An estimated 30–40 people speak the language. The language analyzed in this paper is spoken by people living around the Copper Center, Tazlina and Gulkana area, on the Copper River. The consonants of Ahtna are given in Table 3.2 in ipa, with orthography in angled brackets. Where more than one form is placed in a cell, variability is indicated. Table 3.2 Ahtna consonants Labial

Stops and affricates

Nasals Fricatives Approximants

Alveolar Alveolar to Alveo-Palatal

Plain

p ⟨b⟩

Aspirated Glottalized

tʰ ⟨t⟩ tʼ ⟨tʼ⟩ m ⟨m⟩ n ⟨n⟩

Voiced Voiceless

ʍ ⟨hw⟩

t ⟨d⟩

Lateral Palatal Uvular Glottal

ts, tʃ ⟨dz⟩

tɬ ⟨tł⟩

tsʰ, tʃʰ ⟨ts⟩ tsʼ, tʃʼ ⟨tsʼ⟩

tɬʰ ⟨tl⟩ cʰ ⟨c⟩ tɬʼ ⟨tlʼ⟩ cʼ ⟨cʼ⟩

z, ʒ ⟨z⟩ s, ʃ ⟨s⟩

ɬ ⟨ł⟩ l ⟨l⟩

c ⟨g⟩

q ⟨gg⟩

ʔ ⟨ʼ⟩

qʰ ⟨k⟩ qʼ ⟨kʼ⟩ ɴ ⟨ng⟩ ʁ ⟨gh⟩ ç ⟨yh⟩ χ ⟨x⟩ h ⟨h⟩ j ⟨y⟩

Stem-final ejectives in Ahtna Athabascan

53

There are ejectives in Ahtna words in three places of articulation, with lateral articulation of coronal affricate ejectives making a fourth category. This paper is concerned with the realization of ejectives in stem-final position. Stem-final glottalized consonants are a feature of the inventories of some toneless Athabaskan languages. In all dialects of Ahtna except Mentasta, ejectives occur finally in some form—usually, however, not as canonical ejectives with a glottal release. We will focus on the small subset of non-lateral coronals in Central Ahtna because of the large number of possibilities involved, but with the intent of expanding the study to laterals, palatals and uvulars in future work. An example showing all three laryngeal settings for an Ahtna alveolar stop is shown in Figure 3.1.

Figure 3.1 Tʼaa naditaan, ʻHe brought a feather.ʼ Three laryngeal settings for alveolar stops, Central Ahtna (Tazlina)

In Figure 3.1, the phrase- and word-initial ejective tʼ is a canonical ejective, with a silent period between the oral and the glottal release and modal vowel following. In general, the “plain” /t/ tends to have a very short voice onset time, about 10 milliseconds; the aspirated /tʰ/ has a significant voice onset time, 50 milliseconds or more. The canonical syllable-initial ejective /tʼ/ in Figure 3.1 is actually an unusual token, representing a fairly formal utterance. In more relaxed speech, both syllable-initial and syllable-final ejectives vary in their realization and are seldom of the canonical form. Plain and aspirated stops, on the other hand, are more regularly realized with forms similar to that seen in this example. 2

Ejective Realization and Historical Suprasegmentalization

Ejective realization in Ahtna is interesting in the light of both synchronic and diachronic comparison within the Athabascan group and outside Athabascan within Na-Dene.

54

Tuttle

Synchronically, ejectives have been found to vary in their realization in a number of Athabascan languages, tonal and non-tonal. Specifically, the glottal feature associated with ejectives is found to be articulated not within the timing of the ejective itself, but expressed on a neighboring, usually tautosyllabic, vowel. Effects on vowels next to ejectives, along with uncanonical ejective shape, have been discovered in Witsuwitʼen (Wright, Hargus and Davis 2002), and Lower Tanana (Tuttle 1998). Effects on vowels next to other glottalic consonants have been observed in Witsuwitʼen (Hargus 2005). Gordon (2000) finds that glottal features of consonants in Hupa are realized on a preceding vowel when the consonant is in a coda, and on following vowels when the consonant is in onset position; we shall see that similar effects are present in Ahtna. Diachronically, Athabascan “suprasegmentalization” and development of tone from glottal features have been associated only with the presence of glottalic consonants in codas. Spreading of coda laryngealization to stem nuclei is claimed as the historical basis for Athabascan tonal patterns, which result in high tone in some languages and low in others, with loss of glottalization on coda consonants in tonal languages. The presence of coda glottalic consonants in non-tonal Athabascan languages (Ahtna and Hupa, for example) supports a role for coda laryngeal setting as a causative factor in tonogenesis (Leer 1979, Krauss 2005). Variation in ejective realization in the modern non-tonal languages, such as Ahtna, may provide a glimpse of the phonetics of this tonogenesis. The divided expression of the modern tonal systems has been difficult to explain, both phonologically and phonetically (Krauss 2005, Kingston 2005). Explanation has focused on phonetics (Kingston 2005, which proposes dual reflexes based on different types of historical laryngeal articulations) though phonological explanation has also been advanced (Jaker 2012) in the light of a known tone reversal in Dogrib. Proto-Athabascan “constriction” (Leerʼs term for the effects of suprasegmentalization of laryngeal features) might have a modern-day realization in non-tonal languages that have not substituted pitch differences for differences in voice quality. In this sense, Ahtna, as a non-tonal language that has coda ejectives, might qualify as diachronically “pre-tonal,” rather than non-tonal. That would mean that the mechanism of suprasegmentalization, if not the substitution of pitch for voice quality, could be directly observed, a very interesting situation. Such observation is unlikely to solve the persistent problem of dual tonal reflex within the Athabascan family. Kingston (2005:168) points out that the persistence of glottal stop in languages that have developed tone and lost final ejectives (like Lower Tanana) cannot account for the high and low reflexes of historical laryngealization through a process of tone reversal, because a new

55

Stem-final ejectives in Ahtna Athabascan

wave of tonogenesis would only affect syllables that contain the glottal trigger, and no Athabascan language has tone only on syllables ending or historically ending in glottal stop; historically ejective-final syllables are also tonal. However, the possibility of more complex, less consistent prosodic situations is not ruled out. At any given time, a language family that indulges in tonogenesis might demonstrate development, stable presence of, or loss of tonal distinctions in a particular daughter language. If suprasegmentalization of glottalic features persists as a pattern in present-day Athabascan languages, such languages might have variable, subphonological effects on voice quality; or they might have phonologized voice quality but not developed it into pitch. Either state could be classed as diachronically “pre-tonal”. On the other end of the continuum, in a language that is losing or has lost tone, will tonal vowels be modal in quality, or might they retain a voice quality feature as well? And if so, what does the loss of final ejectives and the presence of modal voice quality mean? Does it predict that tone would have developed in such a language, and been lost? The Ahtna language is relevant to these questions because the various dialects present a mini-typology of laryngeal features. Here I repeat Table 3.1 as Table 3.3, with a few additions. My hypothesis is that Kari’s spelling of the forms in the Central and Lower columns represents primarily vocalic, rather than consonantal realizations of glottalic features from coda ejectives. Table 3.3 Orthography and hypothetical ipa transliteration for Ahtna realizations of ejective-final stems cʼezetʼ ‘liverʼ and getsʼ ʻmittensʼ (Kari 1990)

Orthography ipa Orthography ipa

Western Final ejective; Modal voice? Pre-tonal 1

Central Final Ejective: Laryngealization? Pre-tonal 2

Lower Final Ejective: Laryngealization? Pre-tonal 2

Mentasta No Final Ejective: Modal Voice? Post-tonal?

cʼezetʼ [cʼɛzɛtʼ] getsʼ [cɛtsʼ]

cʼezeʼt [cʼɛzɛ̰t] geʼts [cɛ̰ts], [cɛ̰s]

cʼezeʼł [cʼɛzɛ̰ɬ] geʼs [cɛ̰s]

cʼezet [cʼɛzɛt] gets [cɛts]

In what follows, I examine data from Central Ahtna in some detail, in an attempt to determine whether this hypothesis is correct. Though the table above p ­ rovides a range of hypotheses, only the Central column will be addressed in this paper.

56

Tuttle

3 Data The goal of this paper is to discover the phonetic realization of tautosyllabic vowel-ejective sequences in Central Ahtna Athabascan. Quantitative examination of the available data is desirable for such an inquiry, but it is not without potential pitfalls. There are a number of logistical problems arising from the status of the language and the nature and frequency of the sequence in question. Coda ejectives are typologically rare, and unfortunately, even in Ahtna they are not frequent. Moreover, it is not possible to combine data sets to come up with an overall picture of “Central Ahtna” ejectives, since speakers’ production represents not only dialect, but micro-dialect or family dialect patterns. Add to these issues the fact that there are very few fluent speakers (they number in the tens), and even fewer who will tolerate the production of an elicited list. The resulting numbers do not, therefore, look like the product of a laboratory experiment conducted with speakers of a majority, or even a substantial minority language. However, I believe it is worthwhile to gather up the examples and look at them quantitatively where possible, qualitatively where that makes more sense. Because the previous discourse regarding vowel laryngealization has been focused on diachronic explanation, there may be patterns in the modern language that have escaped attention. The data we have now are not what we would like to have, but they are better than the data we will have in the future, when even fewer speakers are available to help with the inquiry. The data discussed in this paper are therefore of various types. Two sources of data come from the author’s field work: sessions recorded in the creation of a set of examples for teaching provide many tokens, both hyperarticulated and in normal speech context. In addition to these sessions, recorded in Tazlina in 2004, a Gulkana speaker was recorded in 2004 and 2005 performing elicited phrases, specifically targeting ejectives in different positions within syllables and words. In addition to these elicited examples, other material from Western and Lower Ahtna is drawn from teaching materials recorded by James Kari and published by the Mt. Sanford Tribal Consortium Native Language Program (Ainsworth et al. 2005). Older examples in recorded narrative have also been considered. Narratives examined include a Copper Center text recorded in 1978 and transcribed by Kari in 2004, and a Mentasta text recorded and transcribed by Kari in 2004. Due to the sparse distribution of final ejectives, text data are not well suited to quantitative analysis in this study, but examination of examples from the texts shows that the effects found in elicited examples are not the effect of elicitation, but are also present in spontaneous speech.

Stem-final ejectives in Ahtna Athabascan

4

57

Qualitative Phonetic Description

In this section, examples drawn from qualitative observation are presented. Figures 3.2–3.7 represent elicited utterances recorded in conversation with a female speaker from Gulkana. In the small groups of Athabascan speakers within the Copper River villages, individual, family and dialect variation are sometimes equally present, so that patterns must be established within idiolects as well as within dialects. The contrasts present in this personʼs speech are apparent in older textual examples from Copper Center as well, but are given here in context with each other, rather than mixed with the older material. 4.1 Glottal Stops Canonical glottal stop is seen in Figure 3.2, yii ełnit’aats’ (‘he cut into it’) where the word boundary between a vowel final and a vowel initial word induces an epenthetic glottal stop. Note that before the silent part of the glottal stop begins, there is a rather abupt decrease in amplitude in the vowel, and visible glottal pulses.

FIGURE 3.2 Yii ełnitʼaatsʼ, ʻHe cut into itʼ.

A lexical glottal stop is shown in Figure 3.3, balega’ hdet’aan (‘tents are pitched’) where the Russian loan balega’ contains a final glottal stop. Figure 3.3 shows this phrase.

FIGURE 3.3 Balega’ hdet’aan, ‘Tents are pitched’ (lexical glottal stop).

58

Tuttle

In Figure 3.3, there is a near silent section between the end of the last vowel in balega’ and the h of hdet’aan. Preceding this quiet section, however, there is also a section of the vowel where amplitude is decreased and irregular glottal pulses are visible. More commonly, glottal stop between vowels occurs as a sudden drop in amplitude right at the point of formant shift, with some laryngealization in the vicinity. This is shown in Figure 3.4, u’eł, ʻwith him/her’. There is no period of silence, but there is an abrupt shift in formants, and the second vowel is laryngealized.

FIGURE 3.4 u’eł, ʻwith him/her’.

Perhaps the most relevant environment to examine for present purposes would be a glottal stop preceding another stop. Such an example is given in Figure 3.5:

FIGURE 3.5 i’deznaat’, ʻsomething flashed, sparkled’.

In this case we see a noticeable closure (in this example, a full 210 ms.) which represents closure time both for the coda glottal stop and for the following

Stem-final ejectives in Ahtna Athabascan

59

plain alveolar stop. Closure time for a plain intervocalic alveolar stop may be as little as 40 ms. The vowel preceding the glottal stop in Figure 3.5 is reduced in amplitude and a little creaky preceding the consonant, but the effect is not strong in this instance. From these examples, we may see that the consonant described as glottal stop in this Ahtna speaker’s inventory is often something less than a “stop,” but makes its presence known through abruptly reduced amplitude and laryngealization whether or not a period of silence is present. 4.2 Stem-Initial Ejectives Qualitative examples of syllable-initial ejectives, from the same speaker of Central Ahtna, also show variation, but not to the level of change of manner or major class. Figure 3.6 shows k’ey lat’aadze’, ʻbirch bark.’

FIGURE 3.6 Stem-initial ejective t’, k’ey lat’aadze’, ʻbirch bark’.

In Figure 3.6, there are two ejectives, the second of which falls into the alveolar set under discussion. Notice the period of lower volume following the glottal release, as the vowel aa begins. Many, though not all, stem-initial ejectives in the data set show a period of laryngealization, along with lower volume, following the ejective. The vowel preceding this ejective also shows a brief period of lowered volume. When an ejective begins and ends a stem, the vowel between can be laryngealized at both ends, with short vowels sometimes being perturbed all the way through. More commonly there is a modal section in the middle. Figure 3.7 shows a vowel surrounded by ejectives.

60

Tuttle

FIGURE 3.7 Stem vowel with initial and final ejective, ik’ey’nest’aats’, ʻhe cut it apart by accident.’

The long vowel in the final syllable of ik’ey’nest’aats’ is perturbed briefly at the beginning, following the initial ejective, and briefly at the end, preceding the final ejective. Examples 2–7 were recorded by one person, JM of Gulkana, Alaska, and they match realizations found in longer texts by speakers from nearby Copper Center. Both villages are “central” geographically, though speakers manifest influences of various Ahtna dialects at times, making clear boundaries difficut to draw. JM’s elicited data contains the largest number of ejective examples of any of the sources consulted for this study, and for this reason, quantitative methods are applied to these data. 5 Quantitative Analysis 5.1 Background Analysis It seems important to clarify the status of the findings exemplified in section 4, since creaky voice can be found in abnormal speech, and is more frequently found in older speakers, like the Ahtna elders whose speech forms the basis for this study. Voice quality can also be used linguistically and paralinguistically in intonation and discourse. In order to distinguish the findings above from these other possible sources, statistical exploration of some of the data was performed. Small-scale quantitative analysis of the elicited Gulkana data includes measurement of vowel length, duration of visible creak, and amplitude of modal and non-modal sections of vowels, as divided based on auditory and visual examination. In addition, vowels are measured for local jitter, which records the level of irregularity of glottal pulses. Measurements were made in Praat 5.1.26. Quantitative measurements were carried out for a set of 157 syllables. The distribution of tokens to segmental environments is as given in Table 3.4:

61

Stem-final ejectives in Ahtna Athabascan TABLE 3.4 Distribution of vowel tokens in Gulkana data set, segmental environment

t’ ________ 24

ʔ ________ 5

t’ ________ t’ ________ 11 57

________ t’ 33

________ ʔ 27

The distribution of tokens to morphological environment is as given in Table 3.5: TABLE 3.5 Distribution of vowel tokens in Gulkana data set, prosodic word environment Initial

Medial

Final

Monosyllable

57

33

57

10

It might not be obvious why a monosyllable would be different prosodically from an initial or final syllable in a word. In previous studies of Athabascan prosody and morphology, monosyllables vary more widely than other word configurations, because they include a small subset of content words—a few nouns and even fewer verbs—as well as a limited set of function words. The content words have a heavy syllable minimum, and bear stress, while the function words may consist of light syllables and are variably stressed under intonation. For these reasons monosyllables are kept separate. It would be desirable to factor out intonation as a source of voice quality variation. Utterance-final position may be associated with increased creakiness in some languages, although it has not stood out as a marker of utterance finality in previous studies of Ahtna intonation (Tuttle 2008, Berez 2011). Qualitative consideration of this question is presented in section 6.2. Some speakers are creakier overall than others, and older speakers are sometimes creakier than younger ones. When the syllables in this data set that had no contiguous glottalic consonant were measured for jitter in comparison with all others, the following results were obtained. The measure used was “local jitter” in Praat 5.1.26, which is the average absolute difference between periods, divided by the average period. The entire vowel was measured for jitter in all cases, despite the fact that perturbed vowels are divisible into modal

62

Tuttle

and non-modal sections, to measure overall difference. The raw mean value for local jitter for the 57 vowels not contiguous with glottalic consonants is 1.80. This value is high enough to be considered “pathological” according to parametric standards, showing that the speaker’s voice is quite creaky even when no glottal consonants are present. This measurement provides a baseline for other observations. 5.2 Creakiness and Ejective Position Since the goal in this paper is to investigate syllables containing ejectives, we remove the tokens that contain initial or final glottal stop, and also segregate the three examples of syllables that both begin and end with an ejective. The remaining data set contains 14 vowels. Analysis of variance shows a distinct difference in duration of creakiness in vowels (presented here as percentage of duration.) TABLE 3.6 Duration of visible or auditory creak, as percentage of total duration (p < .0001)

No ejective (57) vC’ (ejective coda) (N = 24) C’v (ejective onset) (30)

Mean percentage of creak duration

Std Error

 1.47 16.4 28.4

3.8 2.5 3.4

To compare this measure, which is based on auditory and visual impressions, to an acoustic measurement, the same tokens are examined for jitter in Table 3.7. (There are four missing values in the percentage set, accounting for the different numbers in these two tables.) Table 3.7 Jitter in syllables with ejective onsets, ejective codas, and with no ejectives (p = .0024)

No ejective (56) vC’ (ejective coda) (N = 24) C’v (ejective onset) (33)

Mean jitter

Std Error

1.8 2.2 3.0

.20 .31 .26

Stem-final ejectives in Ahtna Athabascan

63

The interesting thing about this finding is not that the vowels not contiguous to ejectives are less creaky; rather, it is that the vowels following ejectives are quite a bit more creaky than those that precede ejectives. The theory of Athabascan tonogenesis has had nothing to say about onset ejectives, but these data support those of Tuttle 1998, Gordon 2000, Wright, Hargus and Davis 2002, and Hargus 2005, all of which point to coarticulation between onset ejectives and the vowels following them. While it is not the goal of this paper to make a diachronic proposal, these findings form part of a larger, growing body of data that asks a diachronic question: why does the laryngealization associated with ejective onsets have no effect on Athabascan tone? Does this mean that laryngealization in general is not the precursor of tone, and the historical answer must lie elsewhere? Or does the answer have something to do with the structure of the syllable? 5.3 Durations: Consonant or Vowel? We again return to the original question: what does Kariʼs notation for the Central Ahtna ejective-final stems represent? Is the glottal feature in the nucleus a vowel effect (which would make it a better candidate for Leer’s “suprasegmentalization”) or a consonant (which would make Kari’s spelling very literal) in ejective-final examples? That is, would we transcribe the word for “mittens” ⟨gets’⟩ as [cɛʔs] or [cɛ̰s]? We have seen that glottal stops in Ahtna have variable realization. However, they do seem to have duration. This fact might provide another test for the question: consonant or vowel? If a glottal stop is added to the vowel in an ejective-final syllables, while a glottal feature is subtracted from the final consonant, this would change the ratio of durations between nucleus and coda. If the glottal component in the nucleus is a consonant, maybe the nucleus is longer in an ejective-final stem. A quantitative approach to this question must account for any possible differences between short and long vowels. An analysis of variance on durations of stem vowels in ejective-adjacent, glottal-adjacent, and other environments was run separately for short stem vowels and long stem vowels, showing no difference in significance. The combined set, including both short and long vowels, shows interesting differences in duration between those vowels that are adjacent to ejectives, and those in other environments. Boxplot visualization is shown in Figure 3.8.

64

Tuttle

350 300 250 200 150 100

6InterEje

5CodaGlot

4CodaEje

3OnsetGlot

2OnsetEje

1NoGlot

50

Environment

FIGURE 3.8 Analysis of variance over elicited data, duration of stem vowels in ejective environments.

Onset and coda ejectives condition a difference in duration; glottal stop does not have a significant effect in this set, but there are not sufficient tokens to say much about them. Vowels in non-glottalic environments are shorter. But interestingly, in this data set there is not a significant difference between vowels that are preceded by ejectives, and vowels that are followed by ejectives. In fact, the vowels that follow ejectives are longer than those that precede them. The longest vowels are those that occur in a syllable between two ejectives. In the historical literature, lengthening preceding stem-final ejectives is associated with the process of tonogenesis. Leer (1979, 22) notes that in the tonogenetic process he calls “suprasegmentalization,” there was “a transformation from a sequence of vowel plus glottal stop into a long vowel which was suprasegmentally modified” by the glottal feature. In the Central Ahtna examples in this data set, we see lengthened and modified vowels, in very much the manner imagined by Leer for the historical stages preceding tonogenesis. However, we see them in both post-ejective and pre-ejective contexts, while the phonological explanation for tonogenesis involves only coda ejectives. Does this test help us decide whether the Kari spelling represents the phonetics appropriately? I submit that it does, because (rather than in spite of) the variable nature of glottal stop. We have seen that glottal stop can take the form of a vowel with lowered intensity and creakiness, or a robust silence. However, in both cases the realization of glottal stop took time to pronounce. The glottal component in the stems measured for this test also took time. The choice of the apostrophe, a consonant symbol, to represent the Central Ahtna variant of the ejective-closed stem, is for this speaker’s output a pragmatic decision. In the typewriter-based orthography, we have no symbol for “creak,” and it does not

Stem-final ejectives in Ahtna Athabascan

65

seem reasonable to try to invent one. Either spelling version (Cvt’ or Cv’t) signals the presence of the ejective that triggers the vowel effect, and neither reflects the speaker’s phonetics. Overall, the Ahtna writing system, as designed by Kari, is very easy to apply to each of the dialects, with only a few reading rules for each dialect. The Cv’t version could be adopted by Central Ahtna writers as a marker of identity, understood as variably expressed. Looked at in this light, the consonant symbol seems a perfect choice—even when it is read as a vowel. 5.4 Observations from Quantitative Analysis of a Central Ahtna Speaker We find through basic examination of a set of utterances collected from one speaker of Central Ahtna that voice quality is perturbed for this speaker overall. However, jitter measures and visual-auditory measurement of non-modal voice quality show a difference between vowels that are adjacent to glottalic consonants, and those that are not. In addition, measurement of durations in stem vowels in different laryngeal environments show that vowels are longer when they are preceded or followed by ejectives or glottal stop. 6

Some Qualitative Comparisons (Central Ahtna)

6.1 Vowel Quality Preceding Coda vs. Onset Ejectives Ahtna verbs, like those in other Athabascan languages, are stem-final when not suffixed. Suffixation induces resyllabification of stem-final consonants, so that a stem nucleus is no longer tautosyllabic with its final consonant. In both elicited and text data, laryngealization is variably present in stems when the final is resyllabified. A suffix vowel may bear laryngealization, as we would expect given the effect of onset ejectives on following vowels. Examples are shown in Figures 3.9 and 3.10; these examples also come from material elicited with JM.

FIGURE 3.9 Neltsatʼ, ʻhe is shiveringʼ.

66

Tuttle

In Figure 3.9, the last part of the (short) stem vowel [a] is laryngealized, there is a long closure for the [t’], and the consonant is released. Note the sharp decline in pitch in the short vowel as it becomes laryngealized. This is an intonationphrase-final verb.

FIGURE 3.10

Neltsatʼen ʻthe one who is shiveringʼ.

In Figure 3.10, the nominalized form of the verb shown in Figure 3.9 also shows laryngealization in the pre-ejective vowel [a]. In addition, however, the ejective is articulated as a glottal stop. The stem vowel in this case is not pronounced with lowered pitch, though the pitch does drop in the course of the vowel, just as in the phrase-final case. The cases shown in Figures 3.9 and 3.10 demonstrate productive suffixation: relative clause formation using a phonologically dependent affix. Historical suffixation can also be found, particularly in names of animals and plants. In the elicited and narrative data both, non-productive suffixation of ejectivefinal stems does not seem to be marked by laryngealization. An example is given in Figure 3.11.

FIGURE 3.11

Cʼutʼaatsʼi, ʻgreen-eyed gadflyʼ.

Stem-final ejectives in Ahtna Athabascan

67

The verb stem in this word is -tʼaatsʼ, ʻcut,ʼ but the word is an old formative, the suffixation unproductive. There is laryngealization following the initial ejective of the stem, but not preceding the final ejective. It is, however, produced as an ejective, with full oral and glottal releases and a silent period. From one example, we cannot jump to a conclusion about the distinction between productive and historical suffixation. However, the question is relevant to the discussion of tonogenesis. If only coda ejectives conditioned tonal development, what happens under resyllabification caused by suffixation? Further examination of voice quality effects with controlled morphological alternation could be very interesting. Vowel Quality in Utterance-Final Position: Glottalic vs. Non-Glottalic Coda Because many of the glottalic codas in these data are intonationally final, it is necessary to demonstrate that the effects of phrase-final laryngealization are of a different nature than those induced by glottalic codas in the same position. Figure 3.12 shows a non-glottalic closed syllable in phrase-final position. 6.2

ε

FIGURE 3.12

Denat, ʻit customarily shinesʼ.

In Figure 3.12 the vowel, though it is in a phrase-final syllable, is not laryngealized (jitter is 1.77, which is lower than average for this speaker.) There is a long closure to the final t, and the consonant is released. The difference between this picture and that in Figure 3.13 shows how even phrase-finally, laryngealized vowels are not neutralized with others. Figure 3.13 repeats the phrase seen in wave-form in Figure 3.5.

68

Tuttle

FIGURE 3.13

Iʼdeznaatʼ, ʻsomething flashed, sparkled.ʼ

Perturbation of the vowels can be seen in the first syllable (where the vowel is flanked by glottal stops) and in the last syllable. It will be observed in this last pair, as well as in Figure 3.7, that long vowels preceding an ejective stop are laryngealized just as short vowels are. In our observation, the co-articulatory effect of pre-ejective laryngealization seems as frequently observed in long as in short vowels. This is also true of duration as well, as seen in section 5.3. 7

Remarks on Historical/Comparative Data

Leer (1979) articulates the hypothesis, supported by much comparative data, that glottalic features in stem rhymes developed into three prosodic reflexes in modern Athabascan languages: high tone, low tone and tonelessness, the last sometimes accompanied by the survival of coda ejectives. Both Krauss (2005) and Kingston (2005) address the phonetic plausibility of this hypothesis in detail. We have not discussed pitch in this study, though pitch tracks are included in some of the figures. There has never been a suggestion that modern Ahtna is a tonal language, and work done with archival material and living speakers has not suggested that it ever was one. There is, however, clear evidence that pitch is manipulated in intonation in this language (Tuttle 2008, Berez 2011). For these reasons, no attempt has been made in this study to correlate normalized pitch with perturbed vowels. However, the findings shown in Figures 3.9 and 3.10 suggest that laryngealization is being phonologized within stems. Coda ejectives induce preceding laryngealization and onset ejectives induce following laryngealization, except in the productively nominalized case. This suggests that laryngealization is phonologized to the stem and not to the syllable. This would be a necessary condition for tonogenesis as described by Leer, so this finding supports his (1979) theory.

Stem-final ejectives in Ahtna Athabascan

69

On the other hand, the presence of laryngealization following Ahtna onset ejectives raises questions about Leer’s phonetically based theory of tonogenesis. This effect has also been measured in several other non-contiguous daughter languages: Tanana (Tuttle 1998b), Witsuwitʼen (Wright, Hargus and Davis 2002, Hargus 2007) and Han (Manker 2011). If vowel perturbation caused tone, why did vowels preceded by glottalic consonants not develop tone? As in Hargusʼ careful (2007) study of voice quality in Witsuwitʼen, it should be expected that further research in Ahtna would yield variable, and perhaps contradictory, results. Speakers of languages that did not develop tone, but retain the structures that produced tone, may produce mechanical coarticulatory effects, and may phonologize or lexicalize these effects. As shown in the Witsuwitʼen case, both pitch and voice quality effects may exist in the same community without resulting in a consistent tonal system. 7.1 Summary and Future Questions This paper set out to consider a basic descriptive question: what is the realization of the sequence spelled Cvʼt in Kariʼs (1990) Ahtna Athabaskan Dictionary. We find that the phonetic study supports Kari’s spelling, even though the spelling suggests a stop-like realization and the sequences in our data set are vocalic. Ahtna glottal stop allows for quiet, creaky vocalic duration as a realization. Using qualitative and quantitative methods, it is observed that the pre-­ glottalization notated by Kari with an apostrophe in his spelling system stands for vowel lengthening, lowered amplitude and laryngealization, in the environment of a glottalic consonant. There is qualitative evidence that the glottalic features of final ejectives are realized on preceding vowels, both long and short, in the form of low-amplitude laryngealized vowel sections. Syllable-initial ejectives also affect the vowels that follow them, but not those that precede them (compare Gordon 2000 for Hupa), except perhaps in the special case where a stem-final ejective has been resyllabified by suffixation. The effects of ejectives on contiguous vowels are also consistent with Leer’s (1979) theory of Athabascan tonogenesis; laryngealization and lengthening occur on short vowels that are contiguous to ejectives. However, questions also arise, since we see vowels following ejectives to be as subject, or more subject to laryngealization than vowels preceding ejectives, and the theory does not explain why onsets should not have had a diachronic effect. The apparent effect of morphology in the case of suffixation encourages further study of phonetic phenomena in morphological context, as well as phonological context both at the lexical and the phrasal level.

70

Tuttle

References Ainsworth, Cynthea, James Kari, Jane Nicholas, Louise Tansy Mayo and Jake Tansy. 2005. Cantwell Ahtna Language Lessons. Chistochina: Mt. Sanford Tribal Consortium. Ainsworth, Cynthea, James Kari, Etta Bell and Henry Bell. 2005. Lower Ahtna Language Lessons. Chistochina: Mt. Sanford Tribal Consortium. Berez, Andrea L. 2011. Intonation as a genre-distinguishing feature in Ahtna: A quantitative approach. Functions of Language 18:2, 210–236. Gordon, Matthew. 2001. Laryngeal Timing and Correspondence in Hupa. UCLA Working Papers in Phonology 7. Los Angeles: UCLA Department of Linguistics. Hargus, Sharon. 2007. Witsuwitʼen Grammar: Phonetics, Grammar, Morphology. Vancouver: University of British Columbia Press. Holton, Gary. 2000. The Phonology and Morphology of the Tanacross Athabaskan Language. Ph.D. dissertation, University of California, Santa Barbara. ———. 2005. Pitch, tone and intonation in Tanacross. Athabaskan Prosody, eds. Keren Rice and Sharon Hargus, pp. 249–276. Amsterdam, Philadelphia: John Benjamins Publishing Company. Jaker, Alessandro. 2012. Prosodic Reversal in Dogrib (Weledeh Dialect.) Doctoral Dissertation, Stanford University. Kari, James. 1977. Linguistic diffusion between Tanaina and Ahtna. IJAL 43:274–288. ———. 1990. Ahtna Athabaskan Dictionary. Fairbanks: Alaska Native Language Center. ———. 2007. Denaʼina topical dictionary. Fairbanks: Alaska Native Language Center. Kingston, John. 2005. The Phonetics of Athabaskan Tonogenesis. in Athabaskan Prosody, eds. Keren Rice and Sharon Hargus, pp. 137–184. Amsterdam, Philadelphia: John Benjamins Publishing Company. Kraus, Michael E. 2005. Athabaskan Tone. Athabaskan Prosody, eds. Keren Rice and Sharon Hargus, pp. 51–136. Amsterdam: John Benjamins Publishing Company. Leer, Jeff. 1979. Proto-Athabaskan verb stem variation, part one: Phonology. Alaska Native Language Center Research Papers Number 1. Fairbanks: Alaska Native Language Center. Maddieson, Ian, Caroline Smith and Nicola Bessell. 2001. Aspects of the phonetics of Tlingit. Anthropological Linguistics 43:2, 135–176. Tuttle, Siri G. 1998. Metrical and tonal structures in Tanana Athabaskan. Ph.D. dissertation, University of Washington. ———. 2008. Word definition in Ahtna Athabascan. Linguistics 46:2, pp. 439–470. Wright, Richard, Sharon Hargus and Kate Davis. 2002. On the categorization of ejectives: data from Witsuwitʼen. Journal of the International Phonetic Association 32:43–77.

CHAPTER 4

Deg Xinag Word-Final Glottalized Consonants and Voice Quality Sharon Hargus 1

Athabaskan Tonogenesis

The cause and historical results of Athabaskan tonogenesis are by now fairly well understood. Krauss 19641 first reconstructed Proto-Athabaskan as toneless, with syllable-final glottalized consonants (ejectives, glottal stop, glottalized sonorants) as the chief instigators of tonogenesis in some of the languages. Oversimplifying slightly, syllable-final glottalized consonants in ProtoAthabaskan are thought to have coarticulated with preceding vowels, causing some kind of voice quality modification or laryngealization on the vowel (e.g. *thǝts’ ‘cane’ > *thǝ̰ts’, Krauss 2005). This laryngeal modification is traditionally termed constriction in Athabaskan linguistics (Leer 1979). In addition to glottalized consonants, Proto-Athabaskan is also reconstructed with glottalized vowels, e.g. *w��ː᷑ ɬ ‘snare, net’ (Krauss and Leer 1981), similar to the system of plain vs. glottalized vowel nuclei in Eyak (Krauss 1965), sister language to all of Athabaskan. Vocalic glottalization was therefore both contrastive as in *w��ː᷑ ɬ ‘snare, net’ and predictable before glottalized consonants in some contexts,2 such as with *thǝts’ ‘cane’ > *thǝ̰ts’. None of the Athabaskan languages appears to have preserved the contrast between glottalized vs. non-glottalized vowel nuclei, as noted by Leer 1999. * I am truly grateful to Deg Xinag speakers Alta Jerue, Edna Deacon, Hannah Maillelle, James Dementi, Katherine Hamilton, Lucy Hamilton, Phillip Arrow, and Raymond Dutchman for their willingness to participate in this study. I also gratefully acknowledge the funding for this study provided by NSF (OPP-0137483). I thank RA Julia Miller for her help preparing sound files for measurement. Comments provided by Michael Krauss on an earlier draft have improved this chapter, as did those of three anonymous reviewers and editor Matthew Coler. 1  Li 1933 had also noted a connection between tone and final glottal stop in Dene Sųɬiné. See Krauss 2005 (first circulated as an unpublished typescript in 1978) for full discussion of the history of analysis of tone and tonogenesis in Athabaskan linguistics. 2  Final ejective stops and affricates gave rise to vocalic constriction only when the preceding vowel was reduced, not full (Leer 1979).

© koninklijke brill nv, leiden, ���6 | doi ��.��63/9789004303218_005

72

Hargus

Instead, vocalic constriction either evolved into tone or was lost. In some Athabaskan languages, e.g. Dene Sųɬiné (a.k.a. Chipewyan), tone resulting from constriction is high; e.g. PA *thǝts’ ‘cane’ > *thǝ̰ts’ > DS [thɛ́θ] ‘cane’ (Elford and Elford 1998). Such languages are traditionally known as high-marked languages. However, in other, low-marked languages, constriction gave rise to low tone; e.g. Tsek’ene (a.k.a. Sekani) [thǝ̀ s] ‘cane’.3 (Tonal contrasts arose in such languages when unconstricted syllable nuclei gave rise to the opposite of the tone that developed from constriction; e.g. *ʈʂhǝʈʂ ‘dry wood’ (Krauss 2005) > Dene Sųɬiné [tshɛ̀z], Tsek’ene [tshǝ́ ts].) A minority of Athabaskan languages, some of those spoken in western Alaska, western British Columbia and the Pacific Coast group, did not develop tone; e.g. *thǝts’ ‘cane’ > Babine-Witsuwit’en [thǝz], or else developed tone and lost it; e.g. certain dialects of Koyukon.4 Underlying glottalized vowels and vocalic coarticulation with (certain) glottalized consonants in Proto-Athabaskan always yielded identical tonal results in the daughter languages which developed tone; e.g. *w��ː᷑ ɬ ‘snare, net’ > Tsek’ene [mìɬ], Dene Sųɬiné [pí�ɬ̨ ]5 (cf. Babine-Witsuwit’en [piɬ]). Therefore, the vowel preceding final glottalic consonants must have coarticulated with the glottalic consonant to produce a derived glottalized vowel which was not distinct from an underlying one.6 The map in (1), from Krauss 2005, shows the geographic distribution of the tonal and toneless Athabaskan languages. On this map, Dene Sųɬiné is Chp, Tsek’ene is Sk, and Babine-Witsuwit’en is Ba.

3  Tsek’ene (a.k.a. Sekani) and Babine-Witsuwit’en data are from my field notes on those languages. Data are from the Fort Ware (Kwadacha) dialect of Tsek’ene and the Witsuwit’en dialect of Babine-Witsuwit’en. 4  In Koyukon, low tone is found only in the Lower dialect and Southern subdialect of Upper Koyukon. In the other dialects, i.e. all of the Central dialect as well as the Northern subdialect of Upper Koyukon, tone has been lost (Krauss 2000). 5  I transcribe Proto-Athabaskan vowel nasalization and glottalization with standard IPA symbols, tilde over and under vowel, respectively. But in the daughter languages I transcribe vowel nasalization with the nasal hook under the vowel, as in Dene Sųɬiné. (The daughter languages with contrastive nasality often also have contrastive tone, making two diacritic symbols necessary on some vowels. If both are superscripts, they can be hard to read.) 6  This summary glosses over many of the interesting details of Proto-Athabaskan phonology, including internal reconstruction of Proto-Athabaskan to Pre-Proto-Athabaskan (Leer 1979, 1999). See also Leer 2001 on the details of tonal development in two specific areas of the Athabaskan-Eyak-Tlingit family, Tlingit and Southern Athabaskan.

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

(1) Tone systems in Athabaskan

Key to abbreviations: Ap Apachean, At Ahtna, Ba Babine, Bl Bearlake, Bv Beaver, Ca Carrier, Chl Chilcotin, Chp Chipewyan, Dr Dogrib, Hn Han, Ho Holikachuk, Hr Hare, Ik Ingalik, Ko Koyukon, Ks Kaska, K-T Kwalhioqua-Tlataskanai, Ku Kutchin, Mt Mountain, NT Northern Tutchone, PCA Pacific Coast Athabaskan, Sk Sekani, Sl Slavey, Sr Sarcee, ST Southern Tutchone, Tc Tanacross, Tg Tagish, Ti Tanaina, Tl Tahltan, Tn Tanana, Ts Tsetsaut, UK Upper Kuskokwim, UT Upper Tanana. (Map and caption from ‘Athabaskan tone’. In Hargus, Sharon and Keren Rice, eds., Athabaskan Prosody, 2005, p. 69. Reprinted with kind permission from John Benjamins Publishing Company, Amsterdam/Philadelphia. [www.benjamins.com].)

73

74

Hargus

Leer 1999 (54 ff.) has noted that ‘the high-marked languages are for the most part geographically contiguous’ within Athabaskan, apart from Tanacross, separated from high-marked Northern Tutchone by low-marked Upper Tanana, and Chilcotin, ‘a more ambiguous exception’.7 He has further suggested: a hypothesis that assumes an early pairing of phonation and tone. Suppose that constricted vowels, either already in PA or shortly after PA, were phonationally creaky vowels accompanied by low tone . . . Such a system could develop in several ways. First, both phonation and tone could become lost; this would give rise to an atonal system. Second the distinctive phonation could be lost or at least phonologically minimalized, so that low tone becomes the salient marker; this would give rise to the system observed with (at least most) present-day low-marked languages. Third, the distinctive low tone could be lost, and the phonation type could change from “lax creaky voice” to “tense creaky voice”. This “tense creaky voice” would then readily assume high pitch as its tonal coefficient. Along similar lines, Kingston 2005 has suggested that Proto-Athabaskan may have had two kinds of glottalized consonants, ‘slack’ ones, which gave rise to lowered f0 when vowels coarticulated with them, and ‘stiff’ ones, which in other cases yielded raised f0 through coarticulation. Although the gross outlines of Athabaskan tonogenesis are clear, the details of individual languages may still hold clues to the possible state of affairs in Proto-Athabaskan prior to tonogenesis. Witsuwit’en, a dialect of the non-tonal Babine-Witsuwit’en language, has contrasts between vowel-final and glottal stop final syllables, as well as contrasts between syllables ending in modally voiced and glottalized nasals (hereafter transcribed [n’]), as shown in (2): (2) Witsuwit’en final glottalic vs. non-glottalic contrasts [je] ‘louse’ [jeʔ] ‘boy’ (VOC) [jen] ‘across’ [jen’] ‘bridge’

7  It can be seen from (1) that Chilcotin is adjacent to Carrier, which is shown as high-marked. The status of tone in Carrier is problematic (see e.g. Gessner 2003), but it seems that Carrier may have developed high tone from constriction and then generally lost it.

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

75

In Hargus 2007, I presented the results of a study of the effect of contrasts such as those shown in (2) on the voice quality of a preceding vowel, including pitch perturbation. On the basis of their means, three of ten speakers were categorized as pitch raisers, five as pitch lowerers, and one as neutral with respect to pitch perturbation. However, all speakers exhibited variation in pitch perturbation. I concluded that “like the Athabaskan family as a whole, final [glottalized] segments have both pitch raising and pitch lowering effects in Witsuwit’en . . .” (p. 132). 2

Deg Xinag

This chapter is an investigation of the phonetic features of word-final glottalized consonants in Deg Xinag (ing) (a.k.a. Ingalik, Deg Hit’an), a non-tonal Athabaskan language spoken in western Alaska. (Deg Xinag is Ik on the map in (1).) The goal is to examine the phonetic effects of glottalized consonants on preceding vowels, and to determine whether these effects are similar to or different from those previously found in Witsuwit’en, and therefore whether any further clues as to how tonogenesis took place in the family can be gleaned. The contrastive consonants of Yukon dialect of Deg Xinag8 are listed in the phonologically oriented consonant charts in (3)–(4): (3) Deg Xinag syllable-initial contrastive consonants ph

t th t’

k kh k’ q qh q’ ʔ

tθ tθh tθ’ ʦ ʦh ʦ’ tɬ tɬh tɬ’ ʈʂ ʈʂh ʈʂ’ ʧ ʧh ʧ’ v

θð

m

n

sz

ɬl

ʂʐ

ʃ

χʁ

h

ŋ j

8  All of the speakers who participated in this study are/were speakers of the Yukon dialect. The Kuskokwim dialect has a reduced set of contrasts (Krauss 1962).

76

Hargus

(4) Deg Xinag syllable-final contrastive consonants p

td

v m m’ m̥

tθ dð

ʦʣ

tɬ dl

ʈʂ ɖʐ

θð

s

ɬl

ʂʐ

kɡ

n n’ n̥

qɢ

ʔ

ŋ ŋ’ ŋ̊ j j’ j̊

Like certain other Alaskan Athabaskan languages,9 Deg Xinag has developed voicing contrasts in syllable-final position when word-final vowels were lost. If the consonant before the deleted vowel was a voiceless stop/affricate or ejective stop/affricate, it developed into a voiced consonant (5a).10 If the preceding consonant was a (voiced) sonorant, it remained voiced (5b). Original word-final voiceless consonants (not followed by a vowel) remained voiceless (if obstruents, (5c)) or became voiceless (if sonorants, (5d)): (5) Origin of final voicing contrast in Deg Xinag    11 Proto-Athabaskan a. *-kjhǝtɬ’eː ‘younger brother’ (Krauss 2005) b. *thǝŋj-eː ‘trail’ (Leer 2005) c. *ʔã̰ː-t ‘off to the side, away (punctual)’ (Leer 1989) d. *kjhaːn ‘rain’ (Leer 1987)

> > > >

Deg Xinag -ʧhǝdl thǝŋ -ʔot ʧʰon̥

Thus Deg Xinag has a two-way contrast between vowel-final and glottal stopfinal words ((6)a.), like Witsuwit’en, but a three-way contrast for nasals (and other sonorants) in this position ((6)b.):

9  Koyukon, Lower Tanana, Tanacross and (partially) upper Kuskokwim, but not Ahtna, Dena’ina, or Gwich’in. 10  And glottalization was lost from the ejective. (There are no syllable-final ejectives in Deg Xinag.) 11  The contrastive vowel inventory of Deg Xinag is less important for present purposes than the consonants, but it consists of full/long [e a o] and reduced/short [ə ŏ]. [ŏ] is transcribed in this article, but properly speaking is an allophone of /ʊ/ which occurs next to uvulars. /ə/ has an allophone [ɪ] which is common next to palatals and velars, and will seen in spectrogram transcripts. For more information on the vowel system of Deg Xinag, see Hargus 2010 and Hargus 2012.

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

77

(6) Word-final contrasts involving glottalized consonants in Deg Xinag a. oral contrasts -ʁa ‘(man’s) sister-in-law, (woman’s) brother-in-law’ -ʁaʔ ‘grease’ b. nasal contrasts ʧhon̥ eq ʧhon ʁǝɬʧhon’

‘rain’ ‘wet fog, misty rain’ ‘it rained’

Instrumental phonetic detail on the voiceless and glottalized nasals of Deg Xinag is provided next, in 2.1 and 2.2, respectively. 2.1 Acoustic Characteristics of Voiceless Nasals In Deg Xinag, the typical voiceless nasal is voiced at onset and voiceless at offset, although there is some individual variation. The pair of syllables shown in (7) for speaker AJ shows one manifestation of the [n̥ ]-[n] contrast. In the case of [n̥ ] in (7a), the nasal consonant is modally voiced ending in a brief interval of breathy voicing. In the case of [n] in (7b), the vowel before [n] becomes breathy voiced while the nasal itself is breathy voiced. Another difference between the two syllables is the overall shortness of the syllable containing [n̥ ] compared to that of [n]: [ʁon̥ ] has roughly half the duration of [ʁon]. (7) Waveforms, spectrograms12 and pitch tracks of [ʁon] vs. [ʁon̥ ] (AJ) (a) [ʁon̥ ] (from [tŏʁon̥ ] ‘it’s furry’)

12  The vertical axis of all spectrograms in this chapter displays 0–5000 Hz.

78

Hargus

(b) [ʁon] (from [tŏʁon] ‘fur’ (lit., ‘that which is furry’))

Another pair of syllables from a different speaker, ED, shows a slightly different manifestation of the contrast. In [thǝn] in (8a), the nasal is entirely modally voiced, whereas for [thǝn̥ ] in (8b), the nasal is first modally voiced and then voiceless. There does not appear to be any breathy voicing in either token. As with AJ’s [ʁon] vs. [ʁon̥ ], the overall shortness of ED’s [thǝn̥ ] compared to [thǝn] is striking.13 (8) Waveforms, spectrograms and pitch tracks of [thǝn] vs. [thǝn̥ ] (ED) (a) [thǝn] (from [nǝɬthǝn] ‘thunder’)

13  It should be noted that there are differences in the length of the words from which the two compared syllables in (8) are taken, and word length is known to effect duration (e.g.

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

79

(b) [thǝn̥ ] ‘ice’

Finally, the pair of spectrograms in (9) shows yet another version of the contrast between voiced and voiceless nasals, this time from JD. In (9a), [ŋ] is modally voiced throughout, gradually losing intensity, whereas in (9b) [ŋ̊ ] is breathy voiced, becoming voiceless before the velar nasal is released. (9) Waveforms, spectrograms and pitch tracks of [qɪŋ] vs. [qɪŋ̊ ] (JD) (a) [qɪŋ] (from /vǝloqǝŋ/ ‘his/her fingernail’)



Klatt 1973). However, monosyllabic [thǝn̥ ] should have the longer segment durations if word length is influencing the duration of the nasal consonant.

80

Hargus

(b) [qɪŋ̊ ] (from /ðǝqǝŋ̊ / ‘it dried’)

To summarize, the word-final voiceless nasals of Deg Xinag are voiced at onset and voiceless or breathy voiced at offset, the mirror image of word-initial voiceless nasals in languages like Burmese (Bhaskararao and Ladefoged 1991), which are voiceless at onset and voiced at offset. Some individuals produce breathy rather than modally voiced nasals. 2.2 Acoustic Characteristics of Glottalized Nasals Glottalized nasals are complex articulatory events, involving three distinct gestures—oral (stop) closure, velum lowering, and glottal closure—all of which can be timed in different ways (Ladefoged and Maddieson 1996: 109–111). 2.2.1 Glottalized Nasals in Other Languages In the literature on glottalized sonorants, also known as glottalized resonants (Maddieson 2005), there is a consensus that there are two main timing possibilities, pre-glottalized and post-glottalized. Events at the right edge of these complex sounds are crucial for distinguishing these two types. In pre-glottalized sonorants “the laryngeal constriction occurs early in relation to the oral closure” (Ladefoged and Maddieson 1996:53), with “phasing of the glottal constriction to the early portion of the oral constriction” (Howe and Pulleyblank 2001:49). Pre-glottalized nasals can thus be schematized as [ʔn], and any of [ʔn̰ n], [n̰ n], [n̰ n̥], [n̰ t] would be considered pre-glottalized. With post-glottalized sonorants “[the laryngeal constriction] is more delayed [in relation to the oral closure]” (Ladefoged and Maddieson 1996:53); i.e. “glottal constriction persists beyond the oral constriction” (Howe and Pulleyblank

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

81

2001:47). Post-glottalized sonorants seem to be rarer than pre-glottalized nasals. One clear example would be the intervocalic allophone of Shiwilu /ʔr/, described as “a tap accompanied by a glottal closure . . . Intervocalic /ʔr/ is post-glottalized as [r.ʔ]” (Valenzuela and Gussenhoven 2013). Post-glottalized nasals are thus schematically [nʔ]. A glottalized nasal ending in the sequence [ṼṼ�̰ ] or [Ṽn̰ ] would thus constitute a type that is neither pre- nor post-glottalized, as nasality and glottalization end at essentially the same time. I will refer to such sounds as simultaneous glottalized nasals. In Athabaskan languages, glottalized nasals mainly occur in word-final position, and are reconstructed for Proto-Athabaskan in this position. In Proto-Athabaskan, there was no contrast between *Vːn’ and *V̰ ːn (Leer 1979). Kingston 2005 noted that if the timing of the glottalization preceded the oral closure in Proto-Athabaskan, there would have been automatic coarticulation with a preceding vowel, facilitating tonogenesis and also predicting this lack of contrast between *Vːn’ and *V̰ ːn. Descriptions of glottalized nasals in Athabaskan languages are somewhat rare, in part because the consonants themselves are fairly rare, having become nasal vowels when post-vocalic in many of the daughter languages; e.g. *-t’aːn’ ‘leaf’ (Leer 1987) > Tsek’ene –[t’ǫ̀ʔ]. However, two instrumental descriptions are available. Hupa has an unusual phonological contrast between pre- and post-glottalized nasals (Golla 1970, Golla 1977), perhaps the only Athabaskan language with this kind of timing contrast. As described by Gordon 1995, Hupa pre-glottalized nasals have “creak on the end of the preceding vowel and on the beginning of the sonorant”; [eʔn], [eʔn̥ ], [eʔt] are all possible realizations. With Hupa post-glottalized nasals, “the nasal stops in post-glottalized nasals [are] typically quite short and abruptly truncated by the glottal stop” (Gordon 1995: 17–18). For Witsuwit’en, I noted (Hargus 2007:102) that “an appropriate narrow transcription for most speakers, most of the time, is [VṼṼ�̰ ʔn]: the supralaryngeal event, lowering of the velum, [straddles] the laryngeal event, glottal constriction.” The typical Witsuwit’en glottalized sonorant is thus pre-glottalized, since nasality extends beyond the glottalization. 2.2.2 Glottalized Nasals in Deg Xinag In Deg Xinag, the glottalized nasals in the measurement sample in section 3 (/n’ ŋ’/) are also preglottalized ([ʔn ʔŋ]), although there is variation between and within speakers, as summarized in (10).

82

Hargus

(10) Percentages of glottalized nasal types, by speaker

AJ ED HM JD KH LH PA RD

pre-glottalized

simultaneous

100% 100%  89% 100% 100%  90% 100% 100%

0 0 11% 0 0 10% 0 0

In the remainder of this section I provide examples of the different types of glottalized nasals found in Deg Xinag, contrasting each example of [n’] with a modally voiced word-final nasal and also with the word-internal sequence /ʔn/ in /tǝʔne/ ‘he/she says’.14 Word-internal /ʔn/ is always realized as a preglottalized nasal, and is thus a useful comparison with the word-final glottalized nasal. (11b) shows a typical realization of [n’] in Deg Xinag, where the peak of glottalization occurs at the boundary between the vowel and the nasal. (11) contrasts the syllables [thǝŋ] and [thǝŋ’] (more narrowly, [thɪŋ] and [thɪŋ’], respectively) for speaker ED. In the spectrogram of [thɪŋ] in (11a), the vowel formants decrease in amplitude as they become more nasal, ending in the low amplitude nasal consonant. (Vowels before nasal consonants become nasal in varying degrees, and vowel nasality is omitted from the transcriptions included in this chapter.) In the spectrogram of [thɪŋ’] in (11b), the pitch begins to drop, a sign of this speaker’s glottalization (3.2.1.1), as the vowel becomes nasal.

14  Some members of the imperfective paradigm of ‘say’—3SG.SBJ [tǝʔne], 1PL.SBJ [ʈʂ’ǝʔne], 3PL.SBJ [χǝʔne]—are irregular, with /ʔn/ (from *t-n; cf. Witsuwit’en /ʔǝtni/ ‘he/she says’). The remaining forms—1SG.SBJ [tǝsne], 2SG.SBJ [tene], 2PL.SBJ [tŏχne]—are regular.

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

83

(11) Waveforms, spectrograms and pitch tracks of /thǝŋ/ vs. /thǝŋ’/ (ED) (a) [thɪŋ] (from /vǝthǝŋ/ ‘his/her trail’)

(b) [thɪŋ’] (from /vǝthǝŋ’/ ‘its handle’)

Compare this speaker’s /thǝŋ’/ in (11b) with her /tǝʔne/ in (12). The wordinternal sequence is realized nearly identically to the word-final glottalized nasal: the peak of glottalization is centered between the vowel and the nasal in (12):

84

Hargus

(12) Waveform, spectrogram and pitch track of /tǝʔne/ [tɪʔne] ‘he/she says’ (ED)

(13) shows an example from speaker LH of a different kind of pre-glottalized nasal, one with an actual silent period for glottal stop. A modally voiced nasal from this speaker is provided for comparison in (13)a. In the glottalized nasal in (13b), the vowel becomes creaky approaching the glottal stop. Then there is a silent period, transcribed [ʔ] in (13b), released into a creaky voiced nasal, which then becomes modally voiced. (13) Waveforms, spectrograms and pitch tracks of /ŋan/ vs. /ŋan’/ (LH) (a) [ŋan] (from [eŋan] ‘across’)

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

85

(b) [ŋan’] ‘land’

Compare an example of word-internal /ʔn/ from /tǝʔne/ for this speaker. Here there is no silent period for glottal stop, just an interval of creaky voicing at the boundary between the preceding vowel and the nasal. Many of LH’s word-final glottalized nasals looked like (14) rather than (13b). (14) Waveform, spectrogram and pitch track of /tǝʔne/ [tɪʔne] ‘he/she says’ (LH)

86

Hargus

A slightly different version of the kind of preglottalized nasal shown in (13b) can be seen in (15b), for JD. A modally voiced nasal from the same speaker is provided in (15a) for comparison. In the glottalized nasal in (15b) the pitch rises as the vowel becomes glottalized (one of the ways in which glottalization is realized for this speaker, 3.2.1.1). The final segment is a voiceless nasal. (15) Waveforms, spectrograms and pitch tracks of /ŋan/ vs. /ŋan’/ (JD) (a) [ŋan] (from [eŋan] ‘across’)

(b) [ŋan’] ‘land’

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

87

Compare (15b) with one of JD’s productions of /tǝʔne/ ‘he/she says’. The main difference is that for JD the word-final nasal in (15b) is creaky voiced and then voiceless, whereas the prevocalic nasal in (16) is creaky voiced and then modally voiced. (16) Waveform, spectrogram and pitch track of /tǝʔne/ [tɪʔne] ‘he/she says’ (JD)

The spectrogram of [n’] in (17b) is from AJ, and shows the extreme preglottalization often produced by this speaker, extreme in that the glottalization often finished before the vowel ended. Modally voiced [n] from the same speaker is provided for comparison in (17a). (17) Waveforms, spectrograms and pitch tracks of /hon/ vs. /hon’/ (AJ) (a) [hon] (from /q’eç tŏhon/ ‘moose’)

88

Hargus

(b) [hon’] (from /kəʁehon’/ ‘(3sg) ate’)

Compare (17b) with one of AJ’s productions of /tǝʔne/ ‘he/she says’, shown in (18). There the peak of glottalization is aligned with the right edge of the vowel. The nasal is creaky voiced, then becomes modally voiced. (18) Waveform, spectrogram and pitch track of /tǝʔne/ [tɪʔne] (AJ)

The timing of glottalization with glottalized nasals seen with AJ may shed light on an echo vowel phenomenon which occurs with many [ʔ] final tokens in Deg Xinag. In this phenomenon, the vowel before [ʔ] is copied across the wordfinal glottal stop; e.g. /vǝjoʔ/ ‘its lice’ is realized as [vǝjoʔo]. AJ was the most consistent producer of echo vowels. An example of one of her productions is shown in (19):

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

89

(19) Waveform, spectrogram and pitch track of [thoʔo] (from /sǝthoʔ/ ‘my father’) (AJ)

Given the timing of the glottalization component of her glottalized nasals roughly two-thirds to three-quarters before the end of the vowel, it seems that the echo vowel is simply an artifact of the timing of [ʔ] to intersect the preceding vowel in roughly the same way. Another type of preglottalized nasal is shown in (20b) for speaker PA. In this token, glottalization begins at the vowel-nasal boundary and continues well into the nasal consonant, which is released as a voiceless nasal. Although the glottalized nasal in (b) appears to be modally voiced, at least initially, its amplitude is diminished relative to the non-glottalized nasal in (a). (20) Waveforms, spectrograms and pitch tracks for [thɪŋ] vs. [thɪŋ’] (PA) (a) [thɪŋ] (from /vǝthǝŋ/ ‘his/her trail’)

90

Hargus

(b) [thɪŋ’] (from /vǝthǝŋ’/ ‘its handle’)

The effect of glottalization on the amplitude of both vowel and nasal consonant can also be clearly seen in one of the productions of this speaker’s /tǝʔne/, shown in (21): (21) Waveform, spectrogram and pitch track of /tǝʔne/ [tɪʔne] ‘he/she says’ (PA)

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

91

Yet another kind of pre-glottalized nasal is shown in (22b) and (c) for speaker RD, contrasted with a modally voiced word-final nasal in (a). For RD, glottalization was usually marked (in part) by pitch raising. (22) Waveforms, spectrograms and pitch tracks of /hon/ vs. /hon’/ (RD) (a) [hon] (from /q’eç tŏhon/ ‘moose’)

(b) [hon’] (from /kəʁehon’/ ‘he/she ate’)

92

Hargus

(c) [ʁon’] (from /vŏʁon’/ ‘half of it’)

(22c) shows another feature of glottalization sometimes seen with RD. In the segment labeled [o̰ ], glottalization is marked by pitch lowering and creaky voice, then followed by pitch raising. The effect of pitch lowering just before pitch raising is to increase the amount of pitch perturbation, and presumably salience, of the pitch rise. This go-down-to-go-up strategy can also be seen on RD’s word-internal sequence /ʔn/ in (23): (23) Waveform, spectrogram and pitch track of /tǝʔne/ [tɪʔne] ‘he/she says’ (RD)

Finally, an example of the type of glottalized nasal classified here as simultaneous is shown next, for speaker HM. As seen in (24b), this type of glottalized nasal ends in several pulses of [n̰ ], and it is not possible to say that the

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

93

glottalization ends before the nasality, or vice versa. A modally voiced nasal from the same speaker is provided for comparsion in (24a). (24) Waveforms, spectrograms and pitch tracks of [ŋan] vs. [ŋan’] (HM) (a) [ŋan] (from [eŋan] ‘across’)

(b) [ŋan’] ‘land’

Compare the simultaneous glottalized nasal in (24b) with this speaker’s intervocalic /ʔn/ in /tǝʔne/ ‘he/she says’, shown in (25), essentially a preglottalized nasal. As is typical for Deg Xinag, the glottalization seen in (25) is centered around the vowel-nasal boundary. The principal realization is creaky voice (pitch lowering, jitter and decreased amplitude), all of which are present to a greater degree in the word-final glottalized nasal in (24b).

94

Hargus

(25) Waveform, spectrogram and pitch track of /tǝʔne/ [tɪʔne] ‘he/she says’ (HM)

Could the simultaneous glottalized nasal in (24b) be interpreted as post-glottalized? Although the final pulses appear to be creaky voice superimposed on a nasal consonant, this is not substantially different from a typical realization of word-final glottal stop, shown in (26). In this token, the vowel becomes creaky before the glottal stop. The longest glottal pulse is labeled [ʔ] in the transcription. The final pulse, release of the glottal stop, is also creaky voice superimposed on the preceding vowel, judging from the formants which are visible on the final pulse. (26) Waveform, spectrogram and pitch track of [thoʔ] (from /sǝthoʔ/ ‘my father’) (HM)

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

95

To summarize, the typical word-final glottalized nasal in Deg Xinag is preglottalized, with glottalization sometimes realized entirely on the preceding vowel, sometimes mostly on the nasal, but most often at the vowel-nasal consonant boundary. Word-final glottalized nasals are generally produced in the same way as a word-internal sequence /ʔn/ which occurs in some forms of an irregular verb in Deg Xinag. A small number of glottalized nasals were produced in which nasality and glottalization end simultaneously. As a postscript to this section, some recent phonological discussion has centered around the question of whether the timing of glottalized sonorants can be predicted from their distribution, or vice versa (Howe and Pulleyblank 2001). The glottalized nasals of both Deg Xinag and Witsuwit’en, which are restricted to word-final position, are nearly always produced as preglottalized, just as Howe and Pulleyblank 2001 would predict. In Hupa, on the other hand, word-final glottalized nasals contrast in pre- vs. post-glottalization, not as predicted. 3

A Quantitative Study of Voice Quality in Deg Xinag

The effects of glottalized consonants on the voice quality of a neighboring vowel can only be rigorously studied by instrumental means. Acoustic data was collected to investigate the questions in (27). (27) Research questions for investigation of voice quality (a) What effect does the glottalization difference among nasals have on the voice quality of a preceding vowel? (b) What effect does final glottal stop have on the voice quality of a preceding vowel, and is it similar to that of [n’]? These two questions are investigated separately since there is a 3-way contrast among nasals but only a two-way contrast involving glottal stop. My hypothesis, based on previous studies of ejectives in Deg Xinag (Hargus 2008, Hargus 2011), is that there would be inter-speaker variation with respect to pitch lowering and/or raising before [n’], and that for pitch lowerers, there would be increased jitter before [n’]. (I predicted less jitter for pitch raisers because the glottal cycle is shorter, and therefore there is less room for variability.) I also predict greater intensity drops before [n’] because of the narrow glottis restricting airflow. I also expected glottal stop to have the same pitch-lowering or raising effect as [n’] on the basis of findings for Witsuwit’en (Hargus 2007).

96

Hargus

3.1 Methods A word list was recorded with eight speakers (five female, three male) and later downsampled to 11025 samples/second. The basic word list contained eight lexical sets for nasal (28) and oral (29) tokens, as well as other minimal or nearminimal pairs for the nasal contrast. The consonant of interest was word-final in all cases. (28) Basic word list for nasal tokens15   161718 modally voiced nasal

voiceless nasal

glottalized nasal

[eŋan] ‘across’ [taŋan] ‘metal’ [eq ʧhon] ‘wet fog, misty rain’ [q’eç tŏhon] ‘moose’16

[ʈʂ’ǝχǝnekan̥ ] ‘child’ [vǝq’ŏtaŋan̥ ] ‘it’s sharp’ [ʧhon̥ ] ‘rain’ [kaɡ ŏhon̥ ] ‘he/she is eating berries’ [tŏʁon̥ ] ‘it’s furry’ [vǝɬʔaɬthǝŋ̊ ] ‘he/she is asleep’ [ntolnǝŋ̊ ] ‘it (compact object) fell down’ [ðǝqǝŋ̊ ] ‘it dried’

[ŋan’] ‘land’ [sǝ(kǝ)ŋan’] ‘my land’ [ʁǝɬʧhon’] ‘it rained’ [kǝʁehon’] ‘he/she ate’

[tŏʁon] ‘fur’17 [vǝthǝŋ] ‘his/her trail’ [vaχa kǝtalnǝŋ] ‘hammer’18 [vǝloqǝŋ] ‘his/her fingernail’

[vŏʁon’] ‘half of it’ [vǝthǝŋ’] ‘its handle’ [χǝltalnǝŋ’] ‘he/she is knocking’ [vǝqhǝŋ’] ‘her husband’

(29) Basic word list for oral tokens vowel-final

glottal stop final

[vǝʁa] ‘his sister-in-law, her brother-in-law’ [nǝŋǝqha] ‘you (sg.) go (there) and back by boat’ [kǝtθha] ‘canoe bow or stern post’ [kǝtǝle] ‘he/she is singing’ [ajʔo ðǝtɬo] ‘they (pl. objects) are there’

[vǝʁaʔ] ‘its grease’ [ŋǝqhaʔ] ‘your foot’ [ŋǝtθhaʔ] ‘your daughter’ [kǝtǝʁeleʔ] ‘he/she sang’ [vǝloʔ] ‘his/her hand’

15  It was not possible to construct a list in which word length (number of syllables) was controlled for. 16  Lit. ‘that which eats birch’. This name is attested in Osgood 1959. Speakers knew it and produced it, but the more usual term for ‘moose’ is [tǝneɡ]. 17  Lit. ‘that which is furry’ 18  Lit. ‘that with which things are nailed’

Deg Xinag Word-Final Glottalized Consonants and Voice Quality vowel-final

glottal stop final

[ǝnǝstho] ‘I’m hurrying’ [nejo] ‘he/she arrived’ [nǝkǝsqho] ‘I’m dipnetting’

[sǝthoʔ] ‘my father’ [vǝjoʔ] ‘its lice’ [ajʔo ʁeqhoʔ] ‘it (contained object) was there’

97

In comparison sets, the onset consonant of the syllable of interest was matched for place of articulation, and to the greatest extent possible, phonation type, since this can affect pitch, possibly persisting to the midpoint of the vowel. No ejectives or glottal stops were used in onset position. If a speaker did not use a word in one of the lexical sets in (28)-(29), or had a slightly different pronunciation of the sound of interest, then a suitable substitute was made if possible; e.g. [tθha] ‘cliff’ for [kǝtθha] ‘canoe bow or stern post’. If no substitute could be found, then the entire lexical set was discarded for that speaker. However, no fewer than seven lexical sets were recorded for any given speaker in either the oral or nasal dimension. Four repetitions were elicited for each lexical set. Unfortunately, despite instructions, modeling of desired pronunciation (pronouncing each word as if it were a sentence) and attempted intervention during the recording, two speakers consistently recorded the lexical repetitions with list intonation on all but the last word in the set.19 Since effects on pitch of the final consonant were investigated, any repetitions recorded with list intonation were not measured. For a particular speaker, the number of repetitions which were measured varied according to how that speaker recorded the list, and the set of measured words was balanced for preceding consonant. For example, if there were only three usable repetitions of [kǝtǝle], then only three repetitions of [kǝtǝʁeleʔ] were measured. In cases where the repetitions of a lexical set had to be pared down to match those of another lexical set, the repetitions with the best signal to noise ratio were selected for measurement. Phonetic measurements were made with Multi-Speech v. 3.2.0. Measure­ ments of pitch, jitter (relative average perturbation), and intensity were made over a 30 ms. window centered at vowel midpoint and over the peak of 19  The word list was presented to speakers in the local orthography (Krauss 1962, Kari 1974). However, most speakers could not read the orthography. Some speakers could make out some of the words by translating from the English glosses into Deg Xinag. Because of elicitation difficulties, the word list was not recorded four times through with single repetitions of each word, but rather one time through with four repetitions as each word was recognized.

98

Hargus

glottalization, which was generally at the right edge of the vowel, as discussed in 2.2.2. In the first pass through the data, tags were placed at vowel onset, vowel offset, and peak glottalization, if the latter was not located at vowel offset. In a second pass through the data, measurements were taken at vowel midpoint and over the peak of glottalization. Pitch and jitter measures were made from voiced period marks added to the waveform. One advantage of Multi-Speech, which was also used in the study of Witsuwit’en voice quality (Hargus 2007), over other acoustic analysis programs is that it is very quick and easy to correct voiced period marks when they are incorrectly placed by the program. Compare (30) with corrected voice period marks (seen in the waveform) in the Multi-Speech work environment with a spectrogram of the same token above ((13b), generated with Praat 5.1.41), with uncorrected voiced period marks and resulting deficient pitch track: (30) Multi-Speech measurement windows, [ŋan’] ‘land’ (LH)

Since the relative average perturbation measure of jitter requires three glottal cycles in order to perform a calculation, sometimes a larger window had to be used if the measured part of the vowel had low pitch and did not contain 3 glottal cycles within 30 ms. For example, in (30), the smallest window that will include the three rightmost periods in the vowel is 40 ms., shown between the two cursors in the waveform. For oral tokens, spectral tilt (h1-h2) was also measured over a 256sample (= 23.2 ms.) window at vowel midpoint and offset. Spectral tilt can

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

99

be measured in various ways (Gordon and Ladefoged 2001, Epstein and Ladefoged 2001). h1-h2 was selected for the spectral tilt measure in the current study as it presented fewer measurement issues than h1-F1 or h1-F2, and was also the spectral tilt measure used in the previous study of Witsuwit’en (Hargus 2007). A perturbation measure was calculated for each token by subtracting each midpoint measure from each endpoint measure. The perturbation measures were then subjected to inferential statistical tests, which consisted of factorial analysis of variance and the Bonferroni/Dunn post hoc test.20 Given the expectation of inter-speaker variability (which was born out), descriptive and inferential statistics are presented separately for each speaker in results sections. Vowel midpoint measures were also compared to see to what extent the effects of glottalization might be phonologized (Hyman 1976) in Deg Xinag. Phonologization is said to have occurred when a universal, intrinsic phonetic property, such as pitch depression by voiced obstruents, becomes “exaggerated beyond what can be considered universal” (Hyman 2013: 6). (31) shows a scenario for tonogenesis induced by voiced vs. voiceless obstruents outlined by Hyman 1976. At stage II the pitch lowering property of voiced obstruents has become phonologized. (31) Phonologization of post-voiced obstruent tonal perturbation (Hyman 1976) Stage I Stage II Stage III ― ― pá [ ] pá [ ] pá [―] bá [ ] bǎ [ ] pǎ [ ] Stage I > Stage II = phonologization Stage II > Stage II = phonemicization Thus for Deg Xinag, if glottalized consonants perturb pitch at word edge, can their effects also be seen at vowel midpoint, predictable but not yet contrastive? 3.2 Results for Perturbation Measures This section contains results for measures which are token-normalized, by subtracting endpoint measure from midpoint measure. 20  For a two-way comparison, as in the comparisons of voice quality before [ʔ] vs. no following consonant, the significance reported by the Bonferroni/Dunn post hoc test is the same as that of factorial ANOVA.

100

Hargus

3.2.1 Pitch Perturbation Results for pitch perturbation are presented first for nasal and then oral tokens. 3.2.1.1 Nasal Tokens Averages and standard deviations for pitch perturbation for each speaker are provided in (32): (32) Pitch perturbation, means (and standard deviations)

AJ ED HM JD KH LH PA RD

__n

__n̥

__n’

−7 (12.6)  4 (4.9) −2 (4.3) −11 (8.3) −3 (6.0) −3 (7.3) −9 (5.6) −6 (5.2)

−4 (20.4) −2 (7.9) −1 (5.5) −19 (6.6) −7 (8.8) −1 (8.2) −15 (7.2) −12 (6.4)

−37 (22.8) −46 (18.9) −56 (25.2) −2 (14.9) −67 (17.9) −22 (38.6) −36 (31.4)  85 (54.9)

There were two main effects of glottalization on pitch, as summarized in (33). (33) Significant differences in pitch perturbation (Bonferroni/Dunn) 2122232425 [n’] vs. [n]

pitch lowerers pitch raisers

AJ, ED, HM, [n’] < [n]21 KH, LH, PA JD, RD [n’] > [n]23

[n’] vs. [n̥ ]

[n] vs. [n̥ ]

[n’] < [n̥ ]22

not significantly different

[n’] > [n̥ ]24

[n̥ ] > [n] (JD),25 not significantly different (RD)

21  AD, ED, HM, KH: p < .0001; LH: p = .0079; PA: p = .0001 22  AD, ED, HM, KH: p < .0001; LH: p = .0038; PA: p = .0003 23  JD: p = .0007; RD: p = .0002 24  JD, RD: p < .0001 25  p = .0114

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

The graph in (34) of AJ is representative of the pitch lowering group. (34) Pitch lowering before [n’] (AJ)

Graphs for each of the speakers in the pitch raising group are given in (35). (35) Pitch raising before [n’] (RD, JD) (a) RD

(b) JD

101

102

Hargus

The box plot in (36) shows the extent to which each speaker is a uniform pitch lowerer or pitch raiser: (36) Box plot26 of effect of [n’] on pitch perturbation

Of the six pitch lowerers, half (ED, HM, and KH) produced no [n’] tokens with pitch raising, while the remaining three (AJ, PA and LH) produced most of their tokens with pitch lowering. Of the pitch raisers, nearly all of 75% of RD’s tokens were produced with pitch raising, whereas roughly half of JD’s tokens were produced with pitch raising. 3.2.1.2 Oral Tokens Averages and standard deviations for pitch perturbation for each speaker are provided in (37), along with significant differences: (37) Pitch perturbation means, standard deviations and significant differences

AJ ED HM JD KH

___0 (no following C)

__ʔ

significant difs

−3 (9.0) 8 (2.9) −1 (6.2) −21 (12.4) −13 (8.6)

−38 (26.3) −54 (24.0) −59 (11.7) −19 (22.3) −46 (27.4)

[ʔ] < 0 (p < .0001) [ʔ] < 0 (p < .0001) [ʔ] < 0 (p < .0001) n.s. [ʔ] < 0 (p = .0003)

26  The grey box delimits 75th-25th percentiles of data points. The bar in the middle of the grey box is the median. The whiskers above and below the grey box represent 90th and 10th percentiles, respectively.

Deg Xinag Word-Final Glottalized Consonants and Voice Quality LH PA RD

−9 (10.2) −12 (10.9) −10 (20.4)

10 (40.5) −37 (29.3) 22 (44.6)

103

n.s. [ʔ] < 0 (p = .0002) [ʔ] > 0 (p = .0012)

Speakers divided into three groups in terms of significant difference, as seen in (37). The pitch-lowering group, consisting of five speakers (AJ, ED, HM, KH, PA), had significantly greater pitch lowering before [ʔ] than in open syllables. The graph in (38) of AJ is typical of this group: (38) Pitch lowering before [ʔ] (AJ)

There was one pitch-raiser, RD, for whom there was significantly greater pitch rising before glottal stop than in an open syllable. The graph in (39) shows the pattern seen with RD: (39) Pitch raising before [ʔ] (RD)

104

Hargus

For the remaining two speakers (JD, LH), there were no significant differences in pitch perturbation before glottal stop relative to open syllables. Word-final [ʔ] thus had neutral effects on pitch for this group. Again consider a box plot of each speaker’s productions before [ʔ] to see the extent to which each speaker is a uniform pitch raiser or lowerer: (40) Box plot of effect of [ʔ] on pitch perturbation

The five pitch lowerers—AJ, ED, HM, KH and PA—do indeed produce nearly 100% of their tokens before [ʔ] with pitch lowering. As mentioned previously, the most consistent pitch raiser, RD, produces nearly 75% of his tokens with pitch raising. For the two speakers for whom glottalization did not have a significant effect on pitch perturbation, JD produced most of his tokens with pitch lowering before glottal stop, whereas LH produced most of her tokens with pitch raising. 3.2.1.3 Pitch Perturbation Summary Six of the eight speakers behaved consistently across nasal and oral contexts. Five speakers lowered pitch before both [n’] and [ʔ], and one speaker (RD) raised pitch before both types of glottal consonants. JD raised pitch before [n’], but not [ʔ], whereas LH lowered pitch before [n’], but not before [ʔ]. 3.2.2 Jitter Perturbation Results for jitter perturbation, calculated by subtracting midpoint jitter from endpoint jitter, are presented for nasal and then oral tokens. 3.2.2.1 Nasal Tokens Averages and standard deviations for jitter perturbation for each speaker are provided in (41):

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

105

(41) Jitter perturbation, means (and standard deviations)

AJ ED HM JD KH LH PA RD

n

n̥

n’

 .7 (3.73)  .5 (1.07)  .2 (.49) 0.0 (.63) 2.6 (6.78)  .1 (1.94)  .9 (3.33)  .5 (1.19)

 3.2 (5.61)   .2 (.41)   .2 (.68)  1.1 (1.63) –1.5 (4.25)   .8 (1.68)   .2 (2.43) –.1 (1.30)

 4.1 (7.11)  8.9 (9.10)  8.8 (6.86)  3.2 (4.32) 12.9 (11.80) 12.8 (12.29) 10.0 (12.89)  5.6 (8.49)

There were two patterns in terms of significant difference, shown in (42): (42) Significant differences in jitter perturbation  2728 [n’] vs. [n]

[n’] vs. [n̥ ]

[n] vs. [n̥ ]

greater jitter before [n’] ED, HM, JD, KH, [n’] > [n]27 [n’] > [n̥ ]28 n.s. LH, PA, RD no significant difs AJ n.s. n.s. n.s.

The graph in (43) of HM is typical of the majority pattern, significantly greater increase in jitter from vowel midpoint to endpoint before [n’] relative to both [n] and [n̥ ], but no significant differences in jitter perturbation before [n] vs. [n̥ ].

27  HM, JD, LH, p < .0001; PA, p = .0002; RD, p = .0021; ED, p = .0081; KH, p = .0205 28  HM, LH, PA, p < .0001; RD, p = .0005; KH, p = .0022; JD, p = .0043; ED, p = .0063

106

Hargus

(43) Jitter perturbation (HM)

3.2.2.2 Oral Tokens Averages, standard deviations and significant differences for jitter perturbation for each speaker are provided in (44): (44) Jitter perturbation means, standard deviations and significant differences

AJ ED HM JD KH LH PA RD

__ʔ

___0 (no following C)

significant difs

12.8 (21.46) 18.9 (13.90) 14.3 (7.98)  2.0 (2.59) 16.3 (12.85)  2.2 (4.37) 13.5 (9.5)  1.5 (8.13)

 1.4 (4.70)   .4 (1.15)   .4 (1.09)   .5 (1.14)  1.5 (1.69) −2.1 (7.50)  3.0 (3.14)   .1 (3.54)

[ʔ] > 0 (p = .0218) [ʔ] > 0 (p = .0043) [ʔ] > 0 (p < .0001) [ʔ] > 0 (p = .0065) [ʔ] > 0 (p = .0034) n.s. [ʔ] > 0 (p < .0001) n.s.

As seen in (44), for six speakers, there was significantly greater jitter perturbation before [ʔ]. A typical example of this pattern is shown in (45) for JD:

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

107

(45) Jitter perturbation (JD)

3.2.2.3 Jitter Perturbation Summary Five of the eight speakers (ED, HM, JD, KH, and PA) had significantly greater jitter perturbation before both [n’] and [ʔ]. Two speakers (LH and RD) had significantly greater jitter perturbation before [n’] but not [ʔ], and one speaker (AJ) had significantly greater jitter perturbation before [ʔ] but not [n’]. 3.2.3 Intensity Perturbation Results for intensity perturbation, calculated by subtracting midpoint intensity from endpoint intensity, are presented first for nasal and then oral tokens. 3.2.3.1 Nasal Tokens Average and standard deviations for intensity perturbation for each speaker are provided in (46): (46) Intensity perturbation, means (and standard deviations)

AJ ED HM JD KH LH PA RD

n’

n

n̥

−8.7 (4.75) −6.7 (3.26) −6.6 (3.89) −12.6 (4.93) −8.2 (4.13) −10.7 (5.20) −9.5 (5.22) −10.2 (5.42)

−2.1 (2.00) −1.7 (.82) −.5 (1.35) −3.9 (3.76) −.8 (1.21) −1.7 (1.64) −3.9 (2.36) −1.2 (1.16)

−4.7 (3.84) −1.5 (1.57)   .1 (1.62) −3.9 (2.85)   .5 (1.00) −1.5 (2.12) −4.6 (2.10) −2.5 (2.07)

108

Hargus

For seven of the eight speakers, there was significantly less intensity perturbation, indicating a greater drop in intensity, before [n’] than before the other nasals, as summarized in (47). (47) Significant differences in intensity perturbation  293031

energy drop before [n’]ED, HM, JH, KH, only LH, PA, RD energy drops before AJ31 [n’], [n̥ ]

[n’] vs. [n]

[n’] vs. [n̥ ]

[n] vs. [n̥ ]

[n’] < [n]29

[n’] < [n̥ ]30

n.s.

[n’] < [n]

[n’] < [n̥ ]

[n̥ ] < [n]

The graph in (48) of LH is typical of the pattern seen with most speakers, where there is a significantly greater decrease in intensity before [n’] but not [n] or [n̥ ]: (48) Intensity perturbation (LH)

A slightly different pattern was seen with AJ. For this speaker, like the others, there was a significantly greater decrease in vowel intensity before [n’] compared to [n] and [n̥ ], but there was also a significantly greater decrease in

29  HM, JD, KH, LH, PA, RD, p < .0001; ED, p = .0004 30  HM, JD, KH, LH, PA, RD, p < .0001; ED, p = .0003 31  n’ < n, p < .0001; n’ < n̥ , p < .0001; n̥ < n, p = .0060

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

109

vowel intensity before [n̥ ] compared to [n]. [n̥ ] was thus an intermediate type for this speaker. (49) Intensity perturbation (AJ)

3.2.3.2 Oral Tokens Average and standard deviations for intensity perturbation for each speaker are provided in (50), along with significant differences: (50) Intensity perturbation means, standard deviations and significant differences

AJ ED HM JD KH LH PA RD

__ʔ

___0 (no following C)

significant difs

−7.1 (4.32) −6.5 (3.34) −4.0 (1.55) −11.2 (4.46) −8.6 (3.87) −5.4 (3.17) −5.9 (3.97) −8.1 (4.61)

−5.1 (1.91) −4.9 (1.57) −3.0 (1.63) −4.7 (1.95) −2.6 (2.10) −4.1 (2.61) −7.1 (3.94) −5.9 (2.17)

n.s. n.s. [ʔ] < 0 (p = .0229) [ʔ] < 0 (p < .0001) [ʔ] < 0 (p = .0008) n.s. n.s. [ʔ] < 0 (p = .0282)

As seen in (50), there were two patterns in terms of significant difference. For half of the speakers, there were no significant differences in intensity perturbation. For the other half, there was a significantly greater decrease in intensity for vowels before glottal stop as compared to vowels in open syllables. The graph in (51) of RD is typical of this second group of speakers:

110

Hargus

(51) Intensity perturbation (RD)

3.2.3.3 Intensity Perturbation Summary For half of the speakers (HM, JD, KH, RD), there were significantly greater drops in intensity before both glottalized sounds, [n’] and [ʔ]. For three speakers (ED, LH, PA), there were drops in intensity before [n’] but not [ʔ]. For one speaker (AJ), there were drops in intensity before both [n’] and [n̥ ], but not [ʔ]. 3.2.4 Spectral Tilt Perturbation The spectral tilt perturbation measure, like the other perturbation measures, shows what is special about vowel endpoint, abstracting away from inherent spectral tilt properties of the vowel in that token. Negative values for spectral tilt perturbation show that there is an increase in glottalization from vowel midpoint to endpoint. Averages, standard deviations and significant differences for spectral tilt perturbation for each speaker are provided in (52): (52) Spectral tilt perturbation means, standard deviations, and significant differences

AJ ED HM JD KH LH PA RD

__ʔ

___0 (no following C)

significant dif

−2.2 (6.49) −4.2 (3.96) −5.1 (6.12)  1.6 (4.19)   .6 (8.46)  2.4 (9.13) −3.6 (6.95) −4.2 (7.86)

 7.7 (6.02)  7.1 (2.63)  3.5 (4.23) −2.8 (4.62) 11.0 (4.98)   .8 (4.04)  1.8 (6.87)  4.7 (4.87)

p < .0001 p < .0001 p < .0001 p = .0005 p = .0062 n.s. p = .0060 p < .0001

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

111

For six of the speakers (AJ, ED, HM, KH, PA and RD), spectral tilt perturbation before glottal stop was a smaller number than for word-final vowels, as expected. The graph in (53) of RD is typical of this pattern. (53) Negative spectral tilt perturbation before [ʔ] (RD)

For JD, spectral tilt perturbation before glottal stop was a larger number than for word-final vowels, not as expected, as shown in (54). (54) Positive spectral tilt perturbation before [ʔ] (JD)

3.2.5 Summary of Perturbation Measures Some of the significant differences seen in the previous sections are summarized in (55) for [n’] vs. [n] and in (56) for [ʔ] vs. 0. Significant differences which were not as predicted appear between parentheses in (55)–(56).

112

Hargus

(55) [n’] vs. [n], perturbation measures

AJ ED HM JD KH LH PA RD

pitch

jitter

intensity

[n’] < [n] [n’] < [n] [n’] < [n] [n’] > [n] [n’] < [n] [n’] < [n] [n’] < [n] [n’] > [n]

[n’] > [n] [n’] > [n] [n’] > [n] [n’] > [n] [n’] > [n] [n’] > [n] [n’] > [n]

[n’] < [n] [n’] < [n] [n’] < [n] [n’] < [n] [n’] < [n] [n’] < [n] [n’] < [n]

(56) [ʔ] vs. 0, perturbation measures

AJ ED HM JD KH LH PA RD

pitch

jitter

[ʔ] < 0 [ʔ] < 0 [ʔ] < 0

[ʔ] > 0 [ʔ] > 0 [ʔ] > 0 [ʔ] > 0 [ʔ] > 0

[ʔ] < 0 [ʔ] < 0 [ʔ] > 0

[ʔ] > 0

intensity

spectral tilt

[ʔ] < 0 [ʔ] < 0 [ʔ] < 0

[ʔ] < 0 [ʔ] < 0 [ʔ] < 0 ([ʔ] > 0) [ʔ] < 0

[ʔ] < 0

[ʔ] < 0 [ʔ] < 0

Glottalization was predicted to have effects on pitch, jitter and/or intensity measures. However, as can be seen from (55)–(56), speakers differ in which measures have significant effects, and whether or not [n’] and [ʔ] have uniform effects. At one extreme, for HM and KH, all seven measured properties of glottalization were significant, and [n’] and [ʔ] had similar effects. ED, RD and PA are next, with 6 significant effects. For these speakers [n’] exhibits all expected properties of glottalization but [ʔ] only 3 of 4 expected effects. AJ is next with

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

113

4 expected effects, and neither [n’] nor [ʔ] behaving uniformly. JD has 6 significant differences, but the spectral tilt perturbation effect was not as expected of glottalized consonants. [n’] has more of the expected characteristics of a glottalized consonant than [ʔ] for this speaker. For LH, [n’] exhibited all of the expected effects of glottalization, but [ʔ] none. Significant effects on jitter perturbation were found both with pitch raisers and pitch lowerers, contrary to expectations at the outset of this study. 3.3 Results for Measures at Vowel Midpoint This section contains results for measures at vowel midpoint, to demonstrate the extent to which effects of glottalization have become phonologized in Deg Xinag. 3.3.1 Pitch at Vowel Midpoint Results for pitch at midpoint are presented first for nasal and then oral tokens. 3.3.1.1 Nasal Tokens Averages and standard deviations for pitch at vowel midpoint for each speaker are provided in (57): (57) Midpoint pitch, means (and standard deviations)

AJ ED HM JD KH LH PA RD

__n

__n̥

__n’

127 (13.5) 152 (10.9) 163 (12.9) 166 (8.7) 170 (6.7) 128 (19.5) 119 (13.3) 128 (13.3)

135 (19.8) 149 (9.1) 168 (7.8) 164 (6.7) 168 (10.1) 139 (12.6) 123 (12.8) 138 (12.5)

132 (15.9) 148 (19.6) 165 (10.3) 174 (15.8) 168 (8.5) 146 (15.2) 129 (12.8) 130 (25.8)

There were significant differences in midpoint pitch only for LH and PA.

114

Hargus

(58) Significant differences in midpoint pitch  3233

pitch raisers no midpoint pitch excursions

LH32 PA33 AJ, ED, HM, JH, KH, LH, RD

[n’] vs. [n]

[n’] vs. [n̥ ]

[n] vs. [n̥ ]

[n’] > [n] [n’] > [n] n.s.

[n’] > [n̥ ] n.s. n.s.

n.s. n.s. n.s.

Results for PA are shown graphically in (59): (59) Higher midpoint pitch before [n’] (PA)

3.3.1.2 Oral Tokens Averages and standard deviations for each speaker’s midpoint pitch are provided in (60):

32  n’ > n, p = .0010; n̥ > n, p = .0373 33  n’ > n, p = .0136

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

115

(60) Midpoint pitch, means (and standard deviations)

AJ ED HM JD KH LH PA RD

__ʔ

___0 (no following C)

significant difs

125 (28.1) 134 (18.4) 166 (16.3) 204 (22.1) 186 (5.8) 152 (15.9) 135 (12.5) 153 (28.1)

118 (15.5) 142 (5.8) 165 (11.4) 155 (13.4) 172 (7.4) 119 (9.3) 111 (10.6) 122 (27.8)

n.s. n.s. n.s. [ʔ] > 0 (p < .0001) [ʔ] > 0 (p = .0004) [ʔ] > 0 (p = .0003) [ʔ] > 0 (p < .0001) [ʔ] > 0 (p = .0001)

For five of the eight speakers, vowel midpoint pitch was significantly higher before glottal stop than in open syllables. (61) is an example of this pattern, showing the significantly higher pitch before glottal stop seen with LH: (61) Higher midpoint pitch before [ʔ] (LH)

3.3.2 Jitter at Vowel Midpoint Results for midpoint jitter are presented first for nasal and then oral tokens. 3.3.2.1 Nasal tokens Average and standard deviations for jitter at midpoint for each speaker are provided in (62):

116

Hargus

(62) Jitter at midpoint, means (and standard deviations)

AJ ED HM JD KH LH PA RD

__n’

__n

__n̥

2.5 (3.78)  .6 (.24) 1.0 (.67) 1.1 (.67) 2.2 (3.09) 1.5 (1.88) 2.7 (3.76) 1.8 (1.42)

1.4 (1.94)  .7 (.42)  .6 (.43)  .9 (.42)  .8 (.47) 1.4 (2.20) 1.5 (1.87) 1.2 (.90)

1.7 (2.97)  .5 (.41)  .6 (.37) 1.1 (.78) 3.9 (4.58)  .9 (.65) 1.9 (1.90) 1.4 (1.04)

There was significantly greater jitter at vowel midpoint only for HM: vowels before [n’] had significantly more jitter than vowels before [n] (p = .0064) or [n̥ ] (p = .0228). (63) Midpoint jitter (HM)

3.3.2.2 Oral Tokens Averages, standard deviations and significant differences for jitter at vowel midpoint for each speaker are provided in (64):

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

117

(64) Means, standard deviations and significant differences for jitter at vowel midpoint

AJ ED HM JD KH LH PA RD

__ʔ

___0 (no following C)

significant difs

3.1 (4.14) 2.3 (2.77) 2.2 (3.55) 1.3 (.75) 1.2 (.77) 2.8 (2.04) 2.3 (2.43) 4.8 (6.31)

1.3 (1.24) .6 (.63) .7 (.43) .8 (.60) .5 (.36) 3.7 (7.23) 1.5 (1.45) 2.4 (3.67)

n.s. n.s. p = .0269 p = .0044 p = .0206 n.s. n.s. n.s.

The pattern seen with three speakers, significantly greater jitter at vowel midpoint for syllables closed with [ʔ], is exemplified in (65) with JD: (65) Midpoint jitter (JD)

3.3.3 Intensity at Vowel Midpoint Results for midpoint intensity are presented first for nasal and then oral tokens. 3.3.3.1 Nasal Tokens Average and standard deviations for intensity at midpoint for each speaker are provided in (66):

118

Hargus

(66) Intensity at midpoint, means (and standard deviations)

AJ ED HM JD KH LH PA RD

n’

n

n̥

58.9 (3.84) 62.2 (3.67) 64.6 (3.61) 67.2 (2.46) 61.0 (2.09) 62.3 (4.79) 67.0 (3.11) 64.0 (2.69)

61.1 (2.85) 62.7 (4.21) 65.4 (2.39) 67.6 (1.81) 62.3 (1.72) 61.9 (4.00) 67.0 (2.43) 62.0 (2.49)

61.6 (2.98) 62.4 (3.67) 66.0 (1.58) 66.9 (2.70) 60.4 (1.97) 63.3 (2.79) 68.2 (2.93) 62.6 (4.39)

There were significant differences in vowel intensity at midpoint for only one speaker, AJ. For her, midpoint vowel intensity before [n’] was significantly less than that before [n] (p = .0100) and also significantly less than before [n̥ ] (p = .0015), as shown graphically in (67). (67) Differences in vowel intensity at midpoint (AJ)

3.3.3.2 Oral Tokens Average and standard deviations for intensity at midpoint for each speaker are provided in (68) along with significant differences:

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

119

(68) Midpoint intensity means, standard deviations, and significant differences

AJ ED HM JD KH LH PA RD

__ʔ

___0 (no following C)

significant difs

62.1 (4.01) 61.7 (2.22) 62.2 (2.54) 65.1 (4.48) 62.1 (2.44) 59.0 (2.58) 66.4 (2.81) 64.1 (4.19)

59.5 (3.51) 62.7 (3.18) 62.9 (2.37) 64.0 (3.77) 61.4 (1.50) 58.0 (3.63) 62.9 (3.60) 64.7 (3.11)

p = .0285 n.s. n.s. n.s. n.s. n.s. p = .0002 n.s.

There were significant differences in intensity at midpoint for two speakers, AJ and PA, with vowels before glottal stop having significantly greater intensity. The pattern seen with PA is shown in (69): (69) Differences in vowel intensity at midpoint (PA)

Intensity (dB)

80 60 40 20 0

glottal

plain

3.3.4 Spectral Tilt at Vowel Midpoint Averages, standard deviations and significant differences for spectral tilt at vowel midpoint for each speaker are provided in (70):

120

Hargus

(70) Spectral tilt at midpoint, means (and standard deviations)

AJ ED HM JD KH LH PA RD

__ʔ

___0 (no following C)

significant difs

−1.9 (2.72) .9 (3.45) .8 (1.64) 3.7 (3.39) −4.0 (3.18) 6.5 (6.36) 3.4 (4.40) 2.9 (10.48)

 .9 (3.33)  2.1 (1.99)  3.0 (2.27) 12.4 (5.65)  1.9 (4.09)  9.7 (4.45) 10.1 (5.29) 10.5 (4.62)

p = .0047 n.s. p = .0002 p < .0001 p = .0037 n.s. p < .0001 p = .0011

As seen in (70), for six speakers there was significantly less spectral tilt at vowel midpoint for vowels before [ʔ] than for vowels in open syllables, a predicted characteristic of glottalization. The graph in (71) of RD is typical of this group: (71) Spectral tilt at midpoint (RD)

3.3.5 Summary of Midpoint Measures Significant differences seen in the previous sections are summarized for each speaker for [n’] vs. [n] in (72) and [ʔ] vs. 0 in (73). Significant differences which were not as predicted appear between parentheses in (72)–(73). The number of empty cells in (72)–(73) is striking compared to the tables summarizing the perturbation measures in 3.2.5. However, all speakers except ED had at least one significant effect of glottalized consonants on voice quality at vowel midpoint.

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

121

(72) [n’] vs. [n], midpoint measures pitch

AJ ED HM JD KH LH PA RD

jitter

intensity

[n’] < [n] [n’] > [n]

[n’] > [n] [n’] > [n]

(73) [ʔ] vs. 0, midpoint measures pitch

AJ ED HM JD KH LH PA RD

[ʔ] > 0 [ʔ] > 0 [ʔ] > 0 [ʔ] > 0 [ʔ] > 0

jitter

intensity

spectral tilt

([ʔ] > 0)

[ʔ] < 0

[ʔ] > 0 [ʔ] > 0 [ʔ] > 0

[ʔ] < 0 [ʔ] > 0 [ʔ] < 0 ([ʔ] > 0)

[ʔ] < 0 [ʔ] < 0

In general, [ʔ] had a stronger effect on voice quality of a preceding vowel at midpoint than [n’]. For JD, KH and RD, all had at least one significant effect of [ʔ] on some aspect of voice quality, but [n’] did not. Perhaps this is due to what appears to be vowel shortening before [ʔ],34 although vowel duration was not measured in this study.

34  Shorter duration before syllable-final laryngeal consonants [ʔ n] is so common in the languages of the family that Leer 1979 reconstructs this process for Proto-Athabaskan.

122

Hargus

Also of note, both [ʔ] and [n’] had pitch raising effects at vowel midpoint. For two speakers, LH and PA, [n’] and [ʔ] had the same pitch raising effect. For LH, pitch was the only effect of voice quality at vowel midpoint. Note too that LH was the only speaker for whom glottalization differences among nasals had no significant effect on pitch perturbation. For this speaker, high pitch appears to be a phonologized effect of [ʔ] and [n’]. 4

Implications of Voice Quality Study for Tonogenesis

At the outset, this study was interested in answering the following questions. What effect do final glottalized consonants have on the voice quality of a preceding vowel in Deg Xinag? Do [ʔ] and glottalized nasals have uniform effects on voice quality? Are the results for Deg Xinag similar to or different from those of Witsuwit’en? What are the possible implications for understanding Athabaskan tonogenesis? In Deg Xinag, we can first note that in general [n̥ ] had no effect on the voice quality of a preceding vowel: vowels before [n̥ ] and [n] were generally not different, unlike vowels before [n’] vs. [n]. [ʔ] and [n’] had similar effects on voice quality of vowel offset for all speakers, but also some effects even at vowel midpoint, even to the point that high pitch can be said to be predictable before [n’] and [ʔ] for one Deg Xinag speaker (LH). Within Alaska, the development of high tone from historic glottalization is rare, found only in Tanacross (see (1)). In fact, the nearest tonal neighbors to Deg Xinag are the Lower and Southern Upper dialects of Koyukon, where low tone has historically arisen from final glottalization (Krauss 2000). The extent to which high pitch is predictable from final glottalization is thus likely an independent development in Deg Xinag. Comparing Witsuwit’en and Deg Xinag, results of the quantitative study of eight Deg Xinag speakers presented in this chapter are similar in many ways to results of the ten-speaker study of Witsuwit’en in Hargus 2007. Perhaps what is most striking is the variability seen in both languages, both within speakers and across speakers. In both languages, some speakers are pitch raisers, some are pitch lowerers, and some are neutral, as shown in (74).

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

123

(74) Pitch perturbation before glottalized consonants, Witsuwit’en vs. Deg Xinag    3536

pitch lowerers pitch raisers no significant difs

Witsuwit’en (10 speakers)

Deg Xinag (8 speakers)

nasal

oral

nasal

oral

6 336 1

535 4 1

6 2

5 1 2

As can be seen from (74), pitch raisers are a minority in both communities, a point which supports the hypothesis of Leer (1999) that low pitch accompanying Proto-Athabaskan constriction may have been the default. We can also compare the two languages with respect to midpoint pitch: (75) Midpoint pitch before glottalized consonants, Witsuwit’en vs. Deg Xinag

pitch lowerers pitch raisers no significant difs

Witsuwit’en (10 speakers)

Deg Xinag (8 speakers)

nasal

oral

nasal

oral

3 7

5 5

2 6

5 3

In (75) it is striking that in both languages, the pitch of vowels at midpoint was significantly higher before glottalized consonants for some speakers, more so 35  In Witsuwit’en, there was one speaker with significant pitch lowering before [n’] and significant pitch raising before [ʔ]. 36  In Hargus 2007, I divided this group into 2 pitch raisers (proper), those for whom there was significantly greater pitch perturbation before [ʔ n’] and pitch perturbation was a positive number, and a ‘flat f0’ speaker, for whom pitch perturbation before [ʔ n’] was significantly greater than in non-glottalized contexts, but pitch perturbation was not a positive number. Since I am treating JD’s pitch perturbation pattern before [n’] in this study as pitch raising, for consistency I consider the Witsuwit’en sample as containing three pitch raisers.

124

Hargus

before [ʔ] than [n’] in both languages, and that there were no speakers with significantly lower pitch at vowel midpoint. This is all the more surprising since, in terms of pitch perturbation, there were more pitch lowerers in both languages, as shown in (74). In fact, as shown in (76) with bold font, there were some speakers in both languages whose significantly different high pitch at vowel midpoint before glottalized consonants was not matched by significantly different pitch raising at vowel offset in the same context. For Witsuwit’en these are 2F, 5F, 6M; for Deg Xinag these are KH, LH and PA. (76) Witsuwit’en and Deg Xinag speakers with significantly higher midpoint pitch before glottalized consonants. Anomalous vowel midpoint and offset results are in bold. Witsuwit’en

Deg Xinag

speaker

midpoint

perturbation

speaker

midpoint

2F

high

fall

JD

high before [ʔ] only

2M 5F

high before rise [ʔ] only high fall

LH

6M

high

PA

7F

no significant difs high before rise before [ʔ], [ʔ] only fall before [n’]

KH

RD

perturbation

rise before [n’], neutral before [ʔ] high before fall before [ʔ] only [n’], [ʔ] high fall before [n’], neutral before [ʔ] high fall before [n’], [ʔ] high before rise before [ʔ] only [n’], [ʔ]

I suggest that a clue to the seemingly anomalous high pitch at vowel midpoint in Witsuwit’en and Deg Xinag may be found in (22c), the spectrogram of RD, a pitch raiser, which showed pitch raising at the right edge of the vowel before [n’] preceded by pitch lowering in the middle of the vowel. As noted above, lowered midpoint pitch would exaggerate the pitch rise at the right edge of that token. Perhaps the opposite of what is seen in (22c) takes place (occasionally or frequently?) with pitch lowerers; i.e. pitch raising in the middle of the

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

125

vowel to exaggerate the pitch fall before glottalized consonants at right edge. This kind of pitch raising would probably be indistinguishable from high pitch which might results from intonation or stress, certainly not as striking as the pitch lowering and creaky voice found at vowel midpoint in (22c). This kind of pitch raising would have nothing to do with phonologization of pitch from voice quality effects caused by glottal consonants. If this go-up-to-go-down scenario is correct, it might even explain an unusual tone shift found in one of the sub-branches of Athabaskan. Beaver and Tsek’ene are closely related languages (Leer 2006–2010, Goddard 1996), but as seen in (1), Beaver (Bv) and Tsek’ene (Sk) have developed opposite tones from Proto-Athabaskan constriction, with differences in tonal development found even at the dialect level in Beaver (Randoja 1990, Miller 2013). The development of opposite tones must have taken place after tonogenesis from syllable-final glottalization. Beaver is adjacent to the high-marked block of Athabaskan languages. Suppose that the ancestor to Beaver-Tsek’ene was high-marked, with word-final glottalized consonants raising pitch at vowel offset and midpoint, as shown in (77a). At some point there was an innovation whereby endpoint pitch raising was enhanced by lowering vowel midpoint pitch (go-down-togo-up), as shown in (77b), not just high tone before [ʔ] but high tone before all kinds of consonants. The lowered midpoint pitch lowering then came to be reanalyzed as phonemic pitch, perhaps with subsequent changes in the pitch perturbation patterns of final glottalized consonants, as shown in (77c). (77) Possible scenario for shift of high tone to low tone (a) V́ V́V́V́V́C (b) V̀ V̀V̀V́V́C (c) V̀ V̀V̀V̀V̀C In support of stage (77c), one of the interesting findings of Miller 2013 was that in high-marked Doig River Beaver, word-final glottal stops are now pitch lowering consonants. In conclusion, the seeds of both H and L marking have been found now in two non-contiguous toneless Athabaskan languages, confirming one aspect of Kingston’s tonogenesis scenario, that both high and low pitch can result from glottalization in the same language. Reconstruction of variability for the Proto-Athabaskan speech community seems plausible as well. Understanding the effects of glottalized consonants on pitch may even help explain some puzzling shifts in tone that have taken place in Tsek’ene and Beaver dialects.

126

Hargus

References Bhaskararao, P. and Peter Ladefoged. 1991. ‘Two types of voiceless nasals.’ Journal of the International Phonetic Association 21 (02):80–88. Elford, Leon W., and Marjorie Elford. 1998. Dene (Chipewyan) Dictionary. Prince Albert, Saskatchewan: Northern Canada Mission Distributors. Epstein, Melissa A. and Peter Ladefoged. 2001. ‘Phonation Types in Amerindian Languages.’ In Proceedings of the 2001 Athabaskan Languages Conference, ed. by Siri G. Tuttle and Gary Holton. Fairbanks: Alaska Native Language Center, University of Alaska Fairbanks. 154–162. Gessner, Suzanne. 2003. The Prosodic System of the Dakelh (Carrier) Language. PhD dissertation, Department of Linguistics, University of British Columbia. Goddard, Ives. 1996. Native Languages and Language Families of North America. Washington DC: Smithsonian Institution. Golla, Victor. 1970. Hupa Grammar. PhD dissertation, Department of Linguistics, University of California Berkeley. ———. 1977. ‘A note on Hupa verb stems.’ International Journal of American Linguistics 43:355–358. Gordon, Matthew. 1995. ‘The Phonetic Structures of Hupa.’ In Fieldwork Studies of Targeted Languages IV (UCLA Working Papers in Phonetics, 93.), ed. by Ian Maddieson. Los Angeles: UCLA Department of Linguistics Phonetics Lab. 1–24. Gordon, Matthew and Peter Ladefoged. 2001. ‘Phonation types: a cross-linguistic overview.’ Journal of Phonetics 29:383–406. Hargus, Sharon. 2007. Witsuwit’en Grammar: Phonetics, Phonology and Morphology. Vancouver: UBC Press. ———. 2008. Deg Xinag lateral affricates: Phonetic and historical perspectives. Paper presented at Poster presented at annual meeting of Society for the Study of the Indigenous Languages of the Americas, Chicago. ———. 2010. ‘Vowel quality and duration in Yukon Deg Xinag.’ In Working Papers in Athabaskan Languages 2009, ed. by Siri G. Tuttle and Justin Spence. Fairbanks: Alaska Native Language Center. 33–73. ———. 2011. ‘Effects of morpheme type on Deg Xinag ejectives.’ Acoustical Society of America 129:2451. ———. 2012. ‘Deg Xinag Rounding Assimilation: A case study in phonologization.’ Journal of Laboratory Phonology 3:163–193. Howe, Darin and Douglas Pulleyblank. 2001. ‘Patterns and timing of glottalisation.’ Phonology 18:45–80. Hyman, Larry M. 1976. ‘Phonologization.’ In Linguistic Studies offered to Joseph Greenberg on the Occasion of his Sixtieth Birthday, ed. by Alphonse Juilland. Saratoga CA: Anma Libri. 407–418.

Deg Xinag Word-Final Glottalized Consonants and Voice Quality

127

———. 2013. ‘Enlarging the scope of phonologization.’ In Origins of Sound Change: Approaches to Phonologization, ed. by Alan C.L. Yu. Oxford: Oxford University Press. 3–28. Kari, James. 1974. Kuskokwim Ingalik Alphabet and Key Words, Ms. Kingston, John. 2005. ‘The Phonetics of Athabaskan Tonogenesis.’ In Athabaskan Prosody, ed. by Sharon Hargus and Keren Rice. Amsterdam: John Benjamins. 137–184. Klatt, Dennis H. 1973. ‘Interaction between two factors that influence vowel duration.’ Journal of the Acoustical Society of America 54:1102–1104. Krauss, Michael. 1962. Ingalik, Ms. ———. 1964. ‘Proto-Athapaskan-Eyak and the Problem of Na-Dene I: The Phonology.’ International Journal of American Linguistics 30:118–131. ———. 1965. ‘Eyak: a preliminary report.’ Canadian Journal of Linguistics 10:167–187. ———. 2000. ‘Koyukon Dialectology and Its Relationship to Other Athabaskan Languages.’ In Koyukon Dictionary, ed. by Jules Jetté and Eliza Jones. Fairbanks: Alaska Native Language Center, University of Alaska Fairbanks. l-lxv. ———. 2005. ‘Athabaskan tone.’ In Athabaskan Prosody, ed. by Sharon Hargus and Keren Rice. Amsterdam and Philadelphia: John Benjamins. 55–136. Krauss, Michael and Jeff Leer. 1981. Athabaskan, Eyak and Tlingit Sonorants. Fairbanks: Alaska Native Language Center, University of Alaska Fairbanks. Leer, Jeff. 1979. Proto-Athabaskan Verb Stem Variation: I. Phonology. Fairbanks: Alaska Native Language Center, University of Alaska Fairbanks. ———. 1987. ‘Navajo and Comparative Athapaskan.’ In The Navajo Language: A Grammar and Colloquial Dictionary, ed. by Robert Young and William Morgan. Albuquerque: University of New Mexico Press. 264–301. ———. 1989. ‘Directional Systems in Athapaskan and Na-Dene.’ In Athapaskan Linguistics: Current Perspectives on a Language Family, ed. by Eung-Do Cook and Keren Rice. Berlin: Mouton de Gruyter. 575–622. ———. 1999. ‘Tonogenesis in Athabaskan.’ In Cross-Linguistic Studies of Tonal Phenomena: Tonogenesis, Typology, and Related Topics, ed. by Shigeki Kaji. Tokyo: Institute for the Study of Languages and Cultures of Africa and Asia, Tokyo University of Foreign Studies. 37–66. ———. 2001. ‘Shift of Tonal Markedness in Northern Tlingit and Southern Athabaskan.’ In Cross-Linguistic Studies of Tonal Phenomena, ed. by Shigeki Kaji. Tokyo: Institute for the Study of Languages and Cultures of Asia and Africa, Tokyo University of Foreign Studies. 61–89. ———. 2005. ‘How stress shapes the stem-suffix complex in Athabaskan.’ In Athabaskan Prosody, ed. by Sharon Hargus and Keren Rice. Amsterdam and Philadelphia: John Benjamins. 278–318. ———. 2006–2010. Comparative Athabaskan Lexicon, Fairbanks, Alaska, Ms.

128

Hargus

Li, Fang-Kuei. 1933. ‘Chipewyan Consonants.’ Bulletin of the Institute of History and Philology of the Academica Sinica Supplementary Volume I: Ts’ai Yuan Pe’i Anniversary Volume:429–467. Maddieson, Ian. 2005. ‘7. Glottalized Consonants.’ In The World Atlas of Language Structures, ed. by Martin Haspelmath, Matthew S. Dryer, David Gil and Bernard Comrie. Oxford: Oxford University Press. 34–37. Miller, Julia Colleen. 2013. The Phonetics of Tone in Two Dialects of Dane-z̲aa (Athabaskan). PhD dissertation, Department of Department of Linguistics, University of Washington. Osgood, Cornelius. 1959. Ingalik mental culture. New Haven: Dept. of Anthropology, Yale University. Randoja, Tiina. 1990. The phonology and morphology of Halfway River Beaver. Ph.D. dissertation, Department of Department of Linguistics, University of Ottawa. Valenzuela, Pilar M., and Carlos Gussenhoven. 2013. ‘Shiwilu (Jebero).’ Journal of the International Phonetic Association 43 (01):97–106.

CHAPTER 5

Consonant-Tone Interactions: A Phonetic Study of Four Indigenous Languages of the Americas Matthew Gordon 1 Introduction The effect of consonants on the fundamental frequency of adjacent vowels is well documented both on a diachronic level in tonogenesis as well as on a synchronic phonetic level (see Hombert et al. 1979, Bradshaw 1999, Tang 2008 for surveys). The most familiar consonant-tone interaction involves the lowering effect of voiced consonants on F0 in adjacent vowels. Less well studied and less consistent based on available data are the effects of other types of laryngeal features on F0. In an extensive cross-linguistic survey of consonant-tone interactions, Tang (2008) includes a discussion of the typology of c­ onsonant-F0 interactions not related to voicing. She observes that virtually all of the observations about the relationship between glottalization and F0 relate to glottal stop rather than ejectives and that glottalization may either raise or lower F0 depending on the language. Strikingly, even closely related languages, as in the northern branch of Athabaskan (Kingston 2005), may vary in their diachronic tonal reflexes of glottalization. Similarly, aspiration may either raise or lower F0 in adjacent vowels depending on the language and even the individual speaker. Implosive obstruents also display cross-linguistic variation in their effect on F0. Typological research indicates other interesting patterns related to consonanttone interactions. One is that the phonetic effect of consonant laryngeal features on an adjacent vowel is typically shorter in languages with an existing * Thanks to Matt Coler and multiple anonymous reviewers for their helpful suggestions on earlier drafts of this chapter. Many thanks to Dan Everett, Keren Everett, and Peter Ladefoged for graciously providing the Pirahã and Banawá data (originally collected as part of an NSF grant awarded to Peter Ladefoged and Ian Maddieson) and to the UCLA Phonetics Lab Archive (http://archive.phonetics.ucla.edu/) for making available materials from Banawá, Pirahã, and Western Apache. A debt of gratitude is owed to the speakers of all four languages for generously volunteering their time and expertise to provide the data analyzed here. Any errors and misconceptions are solely the author’s responsibility.

© koninklijke brill nv, leiden, ���6 | doi ��.��63/9789004303218_006

130

Gordon

phonemic contrast in tone. Thus, Hombert et al. (1979) found that the duration of the pitch perturbations caused by prevocalic consonants in a following vowel is limited to the 40 to 60 millisecond range in Yoruba, which has phonemic tone on vowels, whereas in English the effect exceeds 100 milliseconds. They offer a compelling explanation for this difference based on the functional role of tone in the two languages. Yoruba speakers localize the phonetic perturbations to the immediately post-consonantal phase of the vowel in order to leave the bulk of the vowel available for conveying phonemic tone. English speakers, on the other hand, need not confine pitch perturbations to the portion of the vowel directly after the consonant since there is no threat of interfering with the perceptibility of phonemic contrasts in tone. There also appears to be an asymmetry both diachronically in terms of tonogenesis patterns as well as on a synchronic phonetic level in the direction of the effect of consonantal laryngeal features on tone in adjacent vowels. Whereas vocalic tonal contrasts have commonly developed from laryngeal contrasts in prevocalic consonants, tonogenesis triggered by postvocalic consonants appears to be a considerably rarer phenomenon (Hombert et al. 1979). Hombert et al. (1979) suggest that this diachronic asymmetry is attributed to a combination of factors observed on a phonetic level in some studies: the decreased magnitude of a consonant’s effect on F0 in a preceding vowel relative to a following vowel coupled with the greater inconsistency in F0 differences in a pre-consonantal vowel as opposed to a post-consonantal vowel. The current study seeks to expand upon our typological knowledge of the effects of different consonant laryngeal features on F0 by examining phonetic data from four languages of the Americas differing along two dimensions: the nature of their laryngeal contrasts in consonants and the phonemic role of tone. Along the dimension of tone, two languages (Pirahã and Western Apache) with phonemic tone and two lacking phonemic tone (Banawá and Hupa) are examined. Along the dimension of laryngeal contrasts, two of these languages (Pirahã and Banawá) have a two-way contrast between voiced and voiceless obstruents, while the other two (Hupa and Western Apache) exploit a three-way laryngeal distinction among stops between voiceless unaspirated, voiceless aspirated, and ejective stops. Systematic phonetic study of the effect of consonants on F0 in the four targeted languages promises to enhance our understanding of three issues in the study of consonant-tone interactions. First, comparing tonal Pirahã and Western Apache with non-tonal Banawá and Hupa allows for study of the relationship between phonemic tone in vowels and subphonemic variations in F0 triggered by adjacent consonants. Second, examination of Hupa and Western Apache, both of which have aspirated and ejective stops in addition to plain

Consonant-tone Interactions

131

unaspirated stops, contributes to our typological understanding of the less well studied effects of aspiration and glottalization on F0. Finally, a comparison of the effect of the consonant on the preceding vs. the following vowel will increase our knowledge of directional asymmetries in the phonetic realization of consonant-tone interactions. 2 Methodology 2.1 Data Data from the four languages consisted of words belonging to a larger corpus designed to examine phonetic properties as part of the UCLA endangered language phonetic documentation project conducted during the 1990s. Data from three of the four languages (all except Hupa) were recorded using a DAT recorder at a sampling rate of 44.1 kHz via a high quality headmounted microphone (Shure SM10), while data from Hupa were recorded using an analog cassette recorder via a handheld microphone. The Hupa data were subsequently digitized at a sampling rate of 16Hz in preparation for analysis. The words were uttered twice in isolation by each speaker in each language. Data from four speakers were examined for Pirahã, Western Apache (two male and two female in each), and Banawá (four males), whereas data from two speakers (one male and one female) were analyzed for Hupa. A nasal sonorant, a voiceless fricative, and stops differing in their laryngeal specification were examined for their effect on F0 in each language. A voiceless and voiced stop were targeted in Banawá and Pirahã, and a voiceless unaspirated, voiceless aspirated, and ejective stop were examined in Hupa and Western Apache. Within each language, place of articulation was controlled for among the stops and nasals to the extent the data allowed. Target sounds occurred in prevocalic position and, with few exceptions, occurred after a vowel as well. The vowel following the target consonant was systematically varied between a high front vowel and a low central vowel except for Western Apache, for which the following vowel was consistently a high vowel. Tone was systematically cross-classified with the following vowel quality in the two languages with lexical tone. In both tone languages, the vowels in syllables adjacent to the one containing the measured vowel consistently carried a low tone in order to control for possible tone sandhi effects. In the three languages for which stress has been reported (all except Western Apache), the location of the target consonant relative to stress was held constant, occurring in the onset of a stressed syllable in Hupa and in the onset of an unstressed syllable in Banawá and Pirahã. In two of the languages, Hupa and Banawá, data allowed for the quality of the preceding vowel to be held

132

Gordon

constant, which allowed for comparison of the effect of the consonant on F0 in the preceding as well as the following vowel. The corpus examined for each language appears in the appendix. 2.2 Measurements A series of F0 measurements were generated using a script within Praat (Boersma and Weenink 2010). Measurements of F0 were taken at ten millisecond intervals in the vowel following the target consonant and, in the case of Hupa and Pirahã, in the vowel preceding the target consonant. In addition, three average measurements encompassing different portions of the target vowel(s) were taken, one constituting the third of the vowel closest to the target consonant (the proximal third), another the middle third (the medial third), and another composed of the third of the vowel farthest removed from the target consonant (the distal third). In the case of the vowel following the target consonant, the initiation of voicing following the release of the consonant as diagnosed from a waveform with a time-aligned spectrogram was used as the demarcation point for the onset of the vowel. The cessation of voicing was taken as the endpoint of the measurement in the following vowel. In the two languages in which the preceding vowel was also analyzed, the release of the consonant preceding this vowel was taken as the start point while the beginning of the consonant constriction was used as the demarcation for the end of the preceding vowel. Data collected using the Praat script were imported into a spreadsheet for coding of variables in preparation for statistical analysis. 2.3 Predictions It was hypothesized that the difference between languages in the phonemic role of tone correlates with differences in the effect of the consonant on the F0 in the following vowel, where this difference might be in magnitude and/ or in the temporal domain. It is thus expected that consonant-induced F0 perturbations will be durationally shorter and/or smaller in magnitude in the two languages with phonemic tone, Pirahã and Western Apache, than in the two languages without phonemic tone, Banawá and Hupa. The relationship between particular consonant laryngeal features and their effect on F0, lowering or raising, is more difficult to predict. It is hypothesized that voiced stops will lower F0 relative to voiceless ones in the two languages with voiced stops, Banawá and Pirahã. However, it is not clear how laryngeal features associated with consonants will interact with F0 in Hupa and Western Apache, both of which lack voiced stops and possess three types of voiceless stops: voiceless unaspirated, voiceless aspirated, and ejective stops. Voiceless

Consonant-tone Interactions

133

stops cross-linguistically are associated with raising of F0, but most of the data on which this observation is based come from languages contrasting voiced and voiceless stops. Carrier, an Athabaskan language related to Hupa and Western Apache that also makes an aspiration rather than a voicing contrast in stops, is reported by Tang (2008) to contradict this pattern in displaying an affinity between voiceless stops and low tone (Pike 1986). It is also unclear how ejectives will impact F0 in the present experiment since there is little cross-linguistic data bearing on this issue. Frazier (2009) shows that ejectives trigger slightly lower F0 in a following vowel compared to plain voiceless obstruents in Yucatec Maya but higher F0 relative to sonorants and glottal stop. In the vowel preceding an ejective, however, F0 is higher than before a voiceless obstruent in one dialect but not in another dialect. Glottal stop more consistently triggers lowering of F0 in a preceding vowel. The lowering of F0 triggered by glottal stop in Yucatec Maya is observed in many languages, although others display the opposite pattern of F0 raising in the vicinity of glottal stop (Hombert et al. 1979, Kingston 2005). In his discussion of the variation between closely related Northern Athabaskan languages in their tonal reflexes of glottalization, Kingston (2005) hypothesizes how different articulatory mechanisms associated with glottalization account for this variation. Cross-linguistic results for fricatives and for sonorants are likewise split between languages evincing F0 raising and those displaying F0 lowering. A final prediction relates to the directionality of the effect of consonant laryngeal features on F0 in adjacent vowels. Given the typology of tonogenesis patterns, it is hypothesized that a consonant will induce a greater effect, temporally and/or in magnitude, on the F0 of a following vowel than a preceding vowel in the two languages, Hupa and Banawá, for which data from both the preceding and following vowel were analyzed. 3 Results 3.1 Banawá Banawá is a non-tonal Arawan language of Brazil spoken by approximately 70 people (Buller et al. 1993). Target words for measurement in Banawá contained one of the consonants /m/, /s/, /d/, /t/ in disyllabic words in two intervocalic contexts: between /a/ and between /i/. In all tokens, the first vowel was stressed and the second vowel unstressed. Figure 5.1 depicts mean F0 values at three positions in the vowels adjacent to the four target consonants: a point ten milliseconds removed from the ­consonant, a window over the third of the vowel closest to the consonant,

134

Gordon Vowel Preceding_V

Following_V

130

Consonant nasal fricative

vcd stop vcl stop

Mean F0 (in Hz)

120

110

100

90 10ms

prox 1/3

med 1/3

10ms

prox 1/3

med 1/3

location Error bars: 95% Cl

FIGURE 5.1 F0 values (in Hertz) averaged over four male Banawá speakers at three points during the vowel preceding and following four consonant types. Error bars represent 95% confidence intervals.

and a window over the medial third of the vowel. To provide a sense of the time windows associated with the proximal and medial third measurements, Figure 5.2 provides average duration values for vowels both preceding and following the different consonants. As Figure 5.1 shows, there is little difference in F0 values (< 5 Hz) as a function of consonant type at any of the measurement points. The most striking result is the lowered F0 of the vowel following the consonant compared to the vowel preceding the consonant, a result that is likely attributed to a terminal pitch fall at the end of the words, all of which were uttered in isolation. This utterance-final vowel was often realized with creaky phonation, which is associated with lowered F0 that might create a baseline effect transcending

135

Consonant-tone Interactions 0.160

Consonant nasal fricative

vcd stop vcl stop

Mean Duration (in seconds)

0.140

0.120

0.100

0.080

0.060

Preceding_V

Following_V Vowel Error bars: 95% Cl

FIGURE 5.2 Duration values (in seconds) for the vowel preceding and following four different consonant types averaged over four Banawá speakers. Error bars represent 95% confidence intervals.

differences in the preceding consonant. A difference in stress between the first (stressed) syllable and the second (unstressed) syllable likely also contributed to the difference in F0 between the two vowels. A series of mixed-effects regression models were fitted to the data with F0 values serving as the dependent variable and consonant (voiced stop vs. voiceless stop vs. fricative vs. nasal) and location of the vowel relative to the target consonant (preceding vs. following vowel) serving as the predictor variables. Speaker was included as a random factor. Separate analyses were performed for the vowel preceding the target consonant and the vowel following the consonant and for different windows in time over which F0 values were calculated.

136

Gordon

These included points at ten millisecond intervals at increasing distances from the targeted consonant ranging from ten to fifty milliseconds, i.e. ten milliseconds removed from the consonant, twenty milliseconds removed, and so on up to fifty milliseconds away from the consonant. In addition, analyses were run for each of the windows comprising equal thirds of the measured vowels, i.e. the third of the vowel closest to the target consonant, the middle of the vowel, and the third farthest away from the consonant. Results of the analysis indicated that, in contrast to the location of the vowel relative to the target consonant, the consonant itself failed to act as a reliable predictor of F0 values at any of the time points in keeping with the results shown graphically in Figure 5.1. However, the random effect of speaker did act as a robust predictor of F0 values. Figure 5.3 depicts individual speaker results for mean F0 values ten milliseconds removed from the target consonant. It can be seen that speaker 3 displays the expected lowering of F0 following a voiced stop. Speaker 1 shows a slight raising of F0 following a voiceless stop. The lowering of F0 adjacent to a voiced stop for speaker 3 is also observed to a lesser degree in the preceding vowel. Note that the slight raising of F0 in the vowel after a voiced stop for speaker 4 is unlikely to reflect a reliable effect. As the F0 values averaged over the third of the vowel closest to the target consonants show in Figure 5.4, the effect of voicing on F0 for speaker 1 and voicelessness for speaker 3 is observable throughout the third of the following vowel proximal to the target consonant (although it is absent by the medial third of the vowel). In summary, the Banawá data show only a weak effect of consonant on F0 values for adjacent vowels and this effect is limited to only certain speakers. One speaker showed a lowering of F0 in adjacent vowels, more notably the following vowel, triggered by voiced stops. For another speaker, the difference between voiced and voiceless stops in their effect on F0 is better characterized as raising following a voiceless stop. In keeping with predictions discussed earlier, the F0 perturbations are more salient on a following vowel than a preceding vowel for those speakers that display any consonant-induced effects on F0. 3.2 Pirahã Pirahã is the only extant language of the Mura family and is spoken by approximately 360 people in the Amazonas region of Brazil (Lewis 2009). The Pirahã consonant inventory is similar to that of Banawá in the respects crucial for this paper except for not contrasting voiced oral and nasal stops (Everett 1986).

137

Consonant-tone Interactions 0.160140

Consonant nasal fricative

120

0.120

120

100 Speaker 2

Mean Duration (in seconds)

80 140

1

Mean F0 (in Hz) Proximal 10 milliseconds

0.140100

vcd stop vcl stop

0.10080

140 120

0.080100

3

80 140

0.060

120

100

Preceding_V

Following_V

4

80

Vowel

Preceding_V

Following_V

Error bars: 95% Cl

Vowel Error bars: 95% Cl FIGURE 5.3 F0 values (in Hertz) for four male Banawá speakers at a point in the vowel ten milliseconds before and after four different consonant types. Error bars represent 95% confidence intervals.

A crucial difference between the two languages, however, lies in the role of F0: Pirahã has a lexical contrast between high and low tone, while Banawá does not. Comparison of the two languages thus allows for testing of the hypothesis that a higher functional load associated with F0 will constrain the magnitude of F0 perturbations induced by consonants. The vowels /a/ and /i/, both high and low tone, following the consonants /s/, /p/ and /b/ were targeted for measurement in Pirahã. All of the measured vowels appeared in unstressed syllables and the preceding vowel was low-toned. The corpus thus included two

138

Gordon 140

Consonant nasal vcd stop fricative vcl stop

120 100

120 100 Speaker 2

Mean F0 (in Hz) Proximal Third

1

80 140

80 140 120 100

3

80 140 120 100

4

80 Preceding_V

Following_V Vowel Error bars: 95% Cl

FIGURE 5.4 F0 values (in Hertz) for four male Banawá speakers computed over the third of the preceding and following vowel proximal to four different consonant types. Error bars represent 95% confidence intervals.

of the four logically possible tone patterns over two syllables: low + low and low + high. Vowels adjacent to /b/ were omitted from the analysis in tokens in which /b/ was either realized as the nasal /m/ or if it was lenited to a fricative or approximant. A few tokens of /i/ following /p/ were devoiced and thus not analyzed. Some instances of /s/ were realized as a glottal fricative but were included in the measurements. The Pirahã data were separated into two groups according to the gender of the speakers. Figure 5.5 depicts mean F0 values separated for phonemic tone as a function of the preceding consonant for the male speakers at three

139

Consonant-tone Interactions Tone low

high

180 Consonant fricative vcl stop vcd stop

Mean F0 (in Hz)

160

140

120

100 10ms

prox 1/3

med 1/3

10ms

prox 1/3

med 1/3

location Error bars: 95% Cl FIGURE 5.5 F0 values (in Hertz) averaged over two male Pirahã speakers at three points during low tone and high tone vowels following three consonant types. Error bars represent 95% confidence intervals.

different time points: ten milliseconds following the consonant, the third of the vowel immediately following the consonant, and the medial third of the vowel. Results for the female speakers appear in Figure 5.6. Results in both figures are for the vowel following the target consonant. It may also be noted in Figure 5.7, which shows duration values for vowels averaged over both male and female speakers, that the time windows comprising the proximal and medial thirds of the vowel are shorter for the vowels following a voiceless stop relative to the other two consonant contexts. A series of mixed effects regression models were fitted to the data following, with certain modifications, the procedure adopted in the analysis of Banawá. As in the analysis of Banawá, separate models were fitted to data points

140

Gordon Tone low 350

high

Consonant fricative vcl stop vcd stop

325

Mean F0 (in Hz)

300

275 250 225 200 175 10ms

prox 1/3

med 1/3

10ms

prox 1/3

med 1/3

location Error bars: 95% Cl FIGURE 5.6 F0 values (in Hertz) averaged over two female Pirahã speakers at three points during low tone and high tone vowels following three consonant types. Error bars represent 95% confidence intervals.

corresponding to different time points at which F0 values were calculated. However, tone (high vs. low) replaced vowel location as the second fixed variable in addition to consonant (voiceless stop vs. voiced stop vs. fricative). An additional difference between the Piraha analysis and the Banawá one was the incorporation of separate analyses for the two genders. In the case of the male speakers, the analyses failed to show any effect of consonant type on F0 at any of the measurement points. The most salient effect for the male speakers is the higher F0 values for vowels carrying a phonemic high tone compared to those associated with a phonemic low tone. The most salient effect attributed to a consonant appears to be the lowering of F0 in low-toned vowels following a voiceless stop relative to low-toned vowels after other consonants.

141

Consonant-tone Interactions 0.150

Consonant fricative vcl stop vcd stop

Mean Duration (in seconds)

0.120

0.090

0.060

0.030

0.000 low

high Tone Error bars: 95% Cl

FIGURE 5.7 Duration values (in seconds) for low tone and high tone vowels following four different consonant types averaged over four Pirahã speakers. Error bars represent 95% confidence intervals.

Results for the two female speakers were statistically more robust. Not only tone level but also consonant preceding the vowel acted as reliable predictors of F0 values for all time points and windows throughout the vowel. In addition, there was an interaction between consonant and tone. Table 5.1 shows summary results of the regression model at the three time points corresponding to those in Figure 5.6: ten milliseconds after the vowel, the beginning third of the vowel and the middle third of the vowel. The estimate for CONSONANT fricative and CONSONANT voiced stop are in relation to CONSONANT voiceless stop and the estimate for TONE low is in relation to TONE high, both of which serve as the intercept.

142 TABLE 5.1

Gordon Log likelihood ratios, fixed effect coefficient estimates, standard error, t-values and p- values in mixed effects models for two female speakers conducted at three time points

Parameter

Estimate

Std. Error

t-value p-value

Time = 10 milliseconds Log Likelihood = 260.8 Intercept CONSONANT fricative CONSONANT voiced stop TONE low CONSONANT fricative * TONE low CONSONANT voiced stop * TONE low

281.452500 20.650000 −38.298750 −65.689167 31.540000 51.515417

7.532873 13.047318 10.653091 11.506653 19.930108 16.731345

37.363 < .001 1.583 .125 −3.595 .001 −5.709 < .001 1.583 .125 3.079 .005

Estimate

Std. Error

Time = proximal third Log Likelihood = 262.8 Intercept CONSONANT fricative CONSONANT voiced stop TONE low CONSONANT fricative * TONE low CONSONANT voiced stop * TONE low Time = medial third Log Likelihood = 280.5 Intercept CONSONANT fricative CONSONANT voiced stop TONE low CONSONANT fricative *TONE low CONSONANT voiced stop * TONE low

281.846250 18.706250 −36.675000 −66.096250 31.017083 50.099000

7.805975 13.520345 11.039315 11.923823 20.652668 17.337935

Estimate

Std. Error

284.567500 15.082500 −32.790000 −75.823214 33.869881 52.161714

9.039163 15.656289 12.783307 13.231980 23.587754 19.685581

t-value p-value

36.106 < .001 1.384 .177 −3.322 .002 −5.543 < .001 1.502 .144 2.890 .007

t-value p-value

31.482 < .001 .963 .343 −2.565 .016 −5.730 < .001 1.436 .162 2.650 .013

The most apparent effect is the raising of F0 adjacent to the voiceless fricative relative to the vowel following other consonants, a difference that is especially apparent for low-toned vowels. Pairwise comparisons indicated that F0 following a voiceless stop was significantly higher (at p < .01) than following other

Consonant-tone Interactions

143

consonants. Voiced stops are also associated with lowering of F0, most notably in the case of high-toned vowels. The pattern seen in high tone vowels for the female speakers is essentially an exaggerated mirror of the pattern seen for the male speakers. In low tone vowels, on the other hand, the higher F0 observed in fricatives relative to voiced stops for the female speakers was not found for the male speakers. It is also interesting to note that, for the female speakers, the mean F0 value for a low tone vowel following a fricative is actually higher in absolute terms than the mean F0 for a high tone vowel after a voiced stop, suggesting that listeners are able to normalize F0 judgments as a function of the consonantal context. Another salient finding holding for both males and females is that the vowel following voiced stops is affected less by the phonemic tone of the vowel than the vowel after either voiceless stops or fricatives. This result suggests that voiced stops may have an intrinsic F0 target that is less mutable than other consonants as a function of a vowel’s tone. 3.3 Hupa Hupa is a Pacific Coast Athabaskan language of Northern California (Golla 1970) spoken by eight people as of 1998 (Lewis 2009). Hupa differs from both Banawá and Pirahã in possessing a larger inventory of consonants whose effects on F0 can be evaluated. Of particular interest is the three-way laryngeal contrast in the stop series between voiceless unaspirated, voiceless aspirated and ejective stops. Both the preceding vowel (the lower high vowel /ɪ/, if present) and the following vowel (either /a/ or /ɪ/) were examined for two Hupa speakers, one male and one female. Targeted consonants were /t/, /tʰ/, /t’/, /s/, and a nasal, which was /m/ in the pre-high vowel context and /n/ preceding the low vowel. All of the vowels preceding the target consonants were unstressed, while all those following the consonant were stressed. Figure 5.8 (male speaker) and Figure 5.9 (female speaker) show average F0 values for the pre-consonantal and post-consonantal vowel at three different time points (ten milliseconds removed from the consonant, the third of the vowel adjacent to the consonant, and the medial third of the vowel) as a function of the target consonant. Nonmodal phonation in the vowel preceding the ejective for the female speaker precluded accurate F0 extraction for the pre-ejective vowel. Figure 5.10 depicts mean duration values for the pre- and post-consonantal vowels collapsed over both speakers. Although the F0 measurements were not subjected to statistical analysis due to the paucity of data points, there are several similarities for both speakers. Most conspicuous are the following effects: first, the raising of F0

144

Gordon Vowel Preceding_V

Following_V

Consonant nasal

240

fricative vcl stop

asp stop eject stop

Mean F0 (in Hz)

210

180

150

120

90 10ms

prox 1/3

med 1/3

10ms

prox 1/3

med 1/3

location Error bars: 95% Cl

FIGURE 5.8 F0 values (in Hertz) averaged over a male Hupa speaker at three points during the vowel preceding and following five consonant types. Error bars represent 95% confidence intervals.

f­ollowing both the ejective and the aspirated stop relative to the unaspirated stop, second, the lowering of F0 in vowels adjacent to the voiceless unaspirated stop and in the vowel preceding the aspirated stop, and third, the raising of F0 preceding the ejective for the one speaker for whom data was available. These effects persist throughout the medial third of both the preceding and following vowel, though they are less pronounced by the medial third in the case of the female speaker. Fricatives and nasals differ in their effect on F0 as a function of their context and the speaker. The male speaker shows a lowering of F0 in vowels adjacent to nasals and raising of F0 in vowels next to fricatives. The female speaker, on

145

Consonant-tone Interactions Vowel Following_V

Preceding_V 380

Consonant nasal fricative vcl stop

asp stop eject stop

Mean F0 (in Hz)

285

190

95

0 10ms

prox 1/3

med 1/3

10ms

prox 1/3

med 1/3

location Error bars: 95% Cl

FIGURE 5.9 F0 values (in Hertz) averaged over a female Hupa speaker at three points during the vowel preceding and following five consonant types. Error bars represent 95% confidence intervals.

the other hand, displays raising of F0 in the vowel preceding but not following a nasal, and the fricative exerts a similar effect to the nasal on the following vowel, while F0 values in the vowel preceding a fricative are similar to those associated with a following aspirated stop. 3.4 Western Apache Western Apache belongs to the Apachean branch of Athabaskan and is spoken by a population of 12,700 according to the 1990 census (Lewis 2009). It possesses the same laryngeal contrasts as Hupa but crucially differs in having a phonemic contrast between high and low tone. F0 was measured for four

146

Gordon 0.250

Consonant nasal fricative vcl stop

asp stop eject stop

Mean Duration (in seconds)

0.200

0.150

0.100

0.050

0.000

Following_V

Preceding_V Vowel

Error bars: 95% Cl FIGURE 5.10

Duration values (in seconds) for the vowel preceding and following five different consonant types averaged over two Hupa speakers. Error bars represent 95% confidence intervals.

Western Apache speakers (two male and two female) in the vowel /i/, both high tone and low tone, following voiceless unaspirated /t/, voiceless aspirated /tʰ/, ejective /t’/, fricative /s/, and nasal /n/. All the target vowels appeared in syllables belonging to the stem, which impressionistically is more prominent than affixes, although it is unclear if Western Apache has stress in addition to tone. The vowel preceding the target consonant consistently carried low tone. The corpus thus included the tonal combinations low + low and low + high. There were no instances of high tone /i/ following /n/. Figure 5.11 (male speakers) and Figure 5.12 (female speakers) depict average F0 values for the post-consonantal vowel at three different time points for

147

Consonant-tone Interactions Tone low

high

250

Consonant asp stop nasal eject stop fricative vcl stop

Mean F0 (in Hz)

200

150

100

50

0 10ms

50 ms

10ms

medial third

50 ms

medial third

location Error bars: 95% Cl FIGURE 5.11

F0 values (in Hertz) averaged over two male Western Apache speakers at three points during low tone and high tone vowels following five consonant types. Error bars represent 95% confidence intervals.

low tone and high tone vowels: ten milliseconds removed from the consonant, fifty milliseconds distanced from the consonant, and over the middle third of the vowel. Figure 5.13 shows mean duration values for the high and low tone vowels collapsed over all four speakers. Regression models fitted to data points at different distances from the target consonant speakers indicated that only tone but not consonant was a reliable predictor of F0 values for both the male and female speakers. However, for the female speakers, there was an interaction between consonant and tone that approached significance. On high tone vowels, the ejective and aspirated stops triggered raising of F0 with the effect of ejectives persisting through the middle of the vowel. Voiceless stops and fricatives were ­associated with

148

Gordon low

250

Mean F0 (in Hz)

300

Tone

high

Consonant nasal asp stop eject stop fricative vcl stop

250

200

150

10ms

50 ms

10ms

medial 1/3

50 ms

medial 1/3

location Error bars: 95% Cl FIGURE 5.12

F0 values (in Hertz) averaged over two female Western Apache speakers at three points during low tone and high tone vowels following five consonant types. Error bars represent 95% confidence intervals.

l­ owering of F0, where the effect of voiceless stops is localized to the very beginning of the vowel. The raising effect of aspirated and ejective stops and the lowering effect of voiceless stops and fricatives for high tone vowels is also observed for male speakers. The effects of the adjacent consonant on vowel F0 is negligible in the case of low tone vowels for both male and female speakers. It is also interesting to note that male speakers do not appear to use F0 to distinguish high and low tone even by the middle of the vowel (except in the case of post-ejective vowels). It is possible that the apparent neutralization of tonal contrasts for the male speakers is due to utterance-final pitch lowering, which might exert a disproportionately large effect on high tone vowels.

149

Consonant-tone Interactions

Mean Duration (in seconds)

0.300

Consonant asp stop nasal eject stop fricative vcl stop

0.200

0.100

0.000 high

low Tone Error bars: 95% Cl

FIGURE 5.13

Duration values (in seconds) for low tone and high tone vowels preceding and following five different consonant types averaged over four Western Apache speakers. Error bars represent 95% confidence intervals.

4 Discussion While some of the hypotheses introduced earlier in the paper were confirmed in the present experiment, at least to some extent, other results are somewhat surprising in light of expectations based on other studies. 4.1 Consonant Type and F0 Looking first at the relationship between consonant type and F0, the predicted lowering of F0 in the vicinity of voiced stops relative to voiceless ones was observed only in a rather limited capacity. In Pirahã, this effect was apparent only for female speakers on high tone vowels. Male speakers of Pirahã failed to

150

Gordon

show F0 lowering induced by voiced stops. Nor was there any voicing-induced lowering of F0 on low tone vowels for Pirahã female speakers. Only two of the four Banawá speakers displayed any interaction between voicing and F0 with two speakers showing higher F0 values on vowels after voiceless stops compared to voiced ones and only early in the following vowel and not in the preceding vowel. A more robust and consistent effect on F0 was associated with spread glottis consonants including fricatives and, in the two languages possessing them, aspirated stops. Pirahã showed a raising of F0 on both high tone and low tone vowels following fricatives for female speakers, although not for male speakers. The Hupa male speaker also displayed a raising effect of fricatives on F0. The female Hupa speaker had higher F0 values following a fricative than following a voiceless unaspirated stop for the third of the vowel closest to the fricative. The female speakers of Western Apache also displayed higher F0 values adjacent to a fricative compared to after a voiceless stop, although this effect was limited to low tone vowels and diminished by a point fifty milliseconds removed from the fricative. Only the Banawá data failed to show raising of F0 by fricatives. A trend was observed for voiceless aspirated stops to exert a raising effect on F0 in the two languages with these two consonant types, Hupa and Western Apache. This effect was not found, however, in low tone vowels for the female Western Apache speakers and not in the preceding vowel for the male Hupa speaker. The tendency for aspirated stops to raise F0 in adjacent vowels is consistent with results from Korean (Kim, Beddor, and Horrocks 2002, Silva 2006) and with Cantonese data presented by Zee (1980). There are at least two possible explanations for the raising of F0 triggered by aspirated stops and fricatives. Stevens (1998) suggests that increased vocal fold stiffness is necessary at the beginning of aspirated consonants to counteract the vocal fold abduction triggered by the increased intraoral pressure associated with aspirated consonants. Silva (1988) hypothesizes that increased transglottal airflow associated with aspiration is responsible for the raising effect on F0 by aspirated consonants. These accounts are compatible with the raising effect on F0 associated with fricatives in the data examined here. Like aspirated stops, fricatives are also produced with high transglottal airflow due to the abducted glottis, which could trigger antagonistic vocal tensing gestures to allow voicing of the vowel(s) adjacent to the fricative. The fact that aspirated stops are characteristically associated with greater raising of F0 in adjacent vowels than fricatives in Hupa and Western Apache suggests a higher rate of transglottal airflow for aspirated stops in these two languages. If true, this would not be surprising given that stops but not fricatives participate in an aspiration contrast in both languages. The raising effect of fricatives and aspirated stops on F0 is not consistent, however, either in the data examined here, as discussed above, or even

Consonant-tone Interactions

151

c­ ross-linguistically. Xu and Xu (2003) find that aspirated stops trigger lowering of F0 in a following vowel in Mandarin, while Cantonese data in Francis et al. (2006) indicate a similar lowering effect on F0 associated with aspirates, in contradiction to Zee’s (1980) Cantonese results. Downing and Gick (2001) cite Mathangwane’s (1999) study of Botswana Kalang’a showing a link between greater aspiration duration and tone-depressor effects. Similarly, Erickson (1975) observes interspeaker variation in the effect of aspiration on F0 in Thai: certain speakers display higher F0 values following aspirated stops relative to unaspirated ones, while others show the opposite effect. As Downing and Gick (2001) suggest, the interspeaker and the interlanguage variation in the relationship between spread glottis consonants like fricatives and aspirated stops indicates the possibility of varied articulatory and aerodynamic strategies for the production of these two consonant types. Tang (2008) suggests that lowered air pressure, presumably below the glottis, may be responsible for the lowering effect on F0 by spread glottis sounds found in some data. The interaction between aspiration and phonemic tone found in the Western Apache female speaker data such that aspirated stops raise F0 only in high tone vowels point to the complexity of the relationship between the spread glottis configuration associated with aspirated consonants and the laryngeal gestures associated with a vowel’s tonal target. Ejectives also tended to trigger raising of F0 in the two examined languages with ejectives, Hupa and Western Apache. The one exception to this generalization is the finding that low tone vowels adjacent to ejectives were realized with lowered F0 by Western Apache male speakers, although it should be borne in mind that there was no statistically robust effect on F0 by consonant type in the Western Apache data for the male speakers. Kingston (2005) offers an articulatory explanation for this variation in the relationship between F0 and ejectives, which has led to salient divergences in tonogenesis patterns even between closely related languages in the Athabaskan family. On the one hand, glottalization may be articulated with increased tension of the vocal folds, which can trigger raising of F0. On the other hand, another realization of glottalization, characteristically described as “creak”, may not be associated with this compression and may instead trigger lowering of F0. The interaction between phonemic tone on the vowel and ejectives in the Western Apache data from male speakers suggests that different articulatory and perhaps aerodynamic strategies might be employed by the same speaker in different tonal contexts in order to accommodate the realization of lexically contrastive tonal information. More generally, the observed differences across languages and across speakers within languages suggest considerable variation in the phonetic implementation strategies employed in the production of consonants.

152

Gordon

4.2 Directionality Directionality was not a particularly important factor in predicting the effect of consonant type on F0 in the examined data. There was a slight asymmetry in the Banawá data such that voicing affected the following more than the preceding vowel for the two speakers showing any effect of consonant on F0. In Hupa, the other language for which both the preceding and following vowel were measured, there were not any clear differences in the relative magnitude of the influence of consonant on the following vowel as opposed to the preceding vowel. The most interesting directional asymmetry involved aspiration in Hupa, which was associated with F0 raising in the following vowel but lowering in the preceding one. This result is perhaps not surprising given the different aerodynamic conditions present going into a consonant constriction as opposed to at the release. The effect of a spread glottis gesture on a preceding vowel is that of breathiness, which is consistently associated with lowered F0 cross-linguistically (Hombert et al. 1979). During the constriction, on the other hand, pressure builds up potentially creating the aerodynamic and articulatory conditions discussed above that may give rise to heightened F0. 4.3 Phonemic Tone vs. Subphonemic F0 Perturbations The present experiment yielded other results that were not predicted a priori. First, there was no consistent tendency for consonant-induced F0 perturbations to be smaller in the two languages with phonemic tone than in the two without despite expectations that the high functional load of F0 in conveying phonemic contrasts in tone might inhibit subphonemic effects attributed to consonants. The language displaying the least effect of consonant on F0, Banawá, lacks phonemic tone, while Pirahã, which possesses tone contrasts on vowels, evinced the most robust effect of consonant on F0. This lack of a difference between languages as a function of the role of tone is rather unexpected given Hombert et al.’s (1979) finding that the temporal duration of F0 perturbations triggered by a preceding consonant is shorter in tonal Yoruba compared to nontonal English. However, the present results mirror data presented in Frazier (2009) for two dialects of Yucatec Maya. She found that consonants actually tended to be differentiated more in terms of their effect on F0 in adjacent vowels in the western dialect, which has contrastive tone, than in the eastern dialect, which lacks tone. The fact that phonemic tone appears not to inhibit consonant-­triggered F0 perturbations suggests that speakers are adept enough at normalizing F0 as a function of consonant context to allow for a sufficiently perceptible realization of tonal contrasts. On the other hand, certain interactions between phonemic tone and consonant-induced effects on F0 observed in the present work and the Frazier (2009) study suggest some sensitivity of

Consonant-tone Interactions

153

speakers to the relationship between tone category and the effect of consonants on F0. Thus the effect of consonant type on F0 differed for some consonants depending on tone type in both of the tonal languages examined here. For examples, voiceless stops induced a stronger lowering effect on F0 in low tone vowels than in high tone vowels in Pirahã, and ejectives induced F0 lowering on low tone vowels but F0 raising on high tone vowels for male speakers of Western Apache. Further research on F0 in a broader cross-section of tonal and non-tonal languages could shed more light on the nature of the relationship between phonemic tone and consonant-induced F0 perturbations. 4.4 Gender Another finding of the present study is that female speakers generally had a stronger effect of consonant on F0 than male speakers. Thus, consonant type had an overall statistically robust effect on F0 for female speakers but not for male speakers in Pirahã and, to a lesser extent, in Western Apache. Nor did consonant reliably influence F0 for the male speakers in Banawá, although it is impossible to determine whether this result is an effect of language or gender since all the Banawá speakers in the present experiment were male. In general, it is difficult to assess the robustness of the observed divergence between male and female speakers in all the examined languages for which data from both genders were collected since the F0 trends are largely similar for the two genders even if the effect of consonant on F0 is more pronounced for the female speakers. The most salient difference between genders appears to be the lowering of F0 in low tone vowels after ejectives for male speakers in Western Apache contrasted with the raising of F0 in the same context for female speakers. This difference plausibly reflects a genuine difference between genders since creakiness, which is often associated with vowels adjacent to ejectives, has a lowering effect on F0 and also is known to be gender dependent, where creak traditionally is more canonically associated with male speech (Henton and Bladon 1988; but see Yuasa 2010 and Podesva and Lee 2010 for contrary results). The gender-dependence of the effect of creakiness on F0 is demonstrated by results of Frazier’s (this volume) study of glottalization in Yucatec Maya. She finds that glottalized vowels, which contrast with modal voiced vowels, are variably realized with different degrees of creakiness. In tokens without creakiness, F0 values are comparable between female and male speakers whereas, in tokens with a creaky realization, F0 patterns diverge between the two genders. In the case of Western Apache, low tone might be articulatorily compatible with creak, which might be more characteristic of male speech than female speech in Western Apache. This speculative hypothesis warrants further investigation for a larger number of speakers.

154

Gordon

5 Conclusion Data from four languages examined in this chapter is consistent with recent research demonstrating the non-universality of F0 perturbations induced by consonants. The most consistent effect on F0 attributed to consonants was the raising of F0 by ejectives and by spread glottis consonants such as aspirated stops and fricatives. Even this result, however, was subject to counterexamples dependent on factors such as language, gender, directionality, and tone. Female speakers generally displayed greater F0 perturbations triggered by consonants than males, although this result was also sensitive to interactions with other properties such as tone and the location of the vowel relative to the consonant. Interestingly, the hypothesis that microprosodic effects on F0 would be smaller in languages with phonemic tone was not confirmed, at least not in its simplest form. Rather, results suggest a more complex relationship between phonemic tone and subphonemic F0 perturbations. Overall the considerable variation in the F0 data both within and across languages is consistent with the view that the both phonemic tone and consonants may be phonetically implemented in different ways with correspondingly varied effects on fundamental frequency. References Boersma, P. and D. Weenink. 2010. Praat: Doing phonetics by computer (version 5.1.42) (www.praat.org). Bradshaw, M. 1999. A Crosslinguistic Study of Consonant-Tone Interaction. PhD diss., Ohio State University. Buller, B., Buller, E. and Everett, D. 1993. Stress placement, syllable structure, and minimality in Banawá. International Journal of American Linguistics 59, 280–293. Downing, L.J., and Gick, B. 2001. Voiceless tone depressors in Nambya and Botswana Kalang’a. Proceedings of Berkeley Linguistics Society 27, 65–80. Erickson, D. 1975. Phonetic implications for a historical account of tonogenesis in Thai. In Studies in Tai Linguistics in Honor of W.J. Gedney, ed. J.G. Harris and J.R. Chamberlain, 100–111. Bangkok: Central Institute of English Language Office of State Universities. Everett, D. 1986. Pirahã. In Handbook of Amazonian Languages vol. 1, ed. D. Derbyshire and G. Pullum, 200–325. New York: Mouton. Francis, A.L., Ciocca, V., Wong, V.K.M., and Chan, J.K.L. 2006. Is fundamental frequency a cue to aspiration in initial stops? The Journal of the Acoustical Society of America 120, 2884–2895. Frazier, M. 2009. Tonal dialects and consonant-pitch interaction in Yucatec Maya. In New Perspectives in Mayan Linguistics, ed. H. Avelino, J. Coon, and E. Norcliffe, 59–82. WPLMIT 59: Cambridge, MA.

Consonant-tone Interactions

155

———. this volume. Pitch and glottalization as cues to contrast in Yucatec Maya. Golla, V. 1970. Hupa Grammar. PhD diss., University of California, Berkeley. Henton, C.G. and Bladon, R.A.W. 1988. Creak as a sociophonetic marker. In Language, speech, and mind: Studies in Honor of Victoria A. Fromkin, ed. L. Hyman, V. Fromkin and C. Li, 3–29. Beckenham, Routledge. Hombert, J.-M., Ohala, J., and Ewan, W. 1979. Phonetic explanations for the development of tones. Language 55, 37–55. Kim, M.-R., Beddor, P.S., and Horrocks, J. 2002. The contribution of consonantal and vocalic information to the perception of Korean initial stops. Journal of Phonetics, 30, 77–100. Kingston, J. 2005. The phonetics of Athabaskan tonogenesis. In Athabaskan Prosody, ed. S. Hargus and K. Rice, 137–184. Amsterdam: John Benjamins. Lewis, M. Paul (ed.). 2009. Ethnologue: Languages of the World, Sixteenth edition. Dallas, TX: SIL International. Online version: http://www.ethnologue.com. Pike, E.V. 1986. Tone Contrasts in Central Carrier (Athapaskan). International Journal of American Linguistics 52, 411–418. Podesva, R. and Lee, S. 2010. Voice quality variation and gender in Washington, DC. Paper presented at NWAV 39. San Antonio, Texas. Silva, D.J. 1998. The effects of prosodic structure and consonant phonation on vowel F0 in Korean: An examination of bilabial stops. In J.R.P. King and S.R. Ramsey (eds.), Progress in Korean linguistics, pp. 1–23. Ithaca, NY: Cornell University Press. Stevens, K.N. 1998. Acoustic Phonetics. Cambridge: MA: MIT Press. Tang, K.E. 2008. The Phonology and Phonetics of Consonant-Tone Interaction. PhD diss., UCLA. Xu, C. and Xu, Y. 2003. Effects of consonant aspiration on Mandarin tones. Journal of the International Phonetic Association 33, 165–181. Yuasa, I.P. 2010. Creaky voice: a new feminine voice quality for young urban-oriented upwardly mobile American women. American Speech 85, 315–337. Zee, E. 1980. The effect of aspiration on the F0 of the following vowel in Cantonese. UCLA Working Papers in Phonetics 49, 90–97.

Banawa Cons t d m

Appendix: Corpora for the Four Languages (Measured Vowels in Bold) Word bata kiti bada bidi bama kimi

Gloss ‘rotten’ ‘small’ ‘name’ ‘small’ ‘catfish’ ‘corn’

156 s

Pirahã Cons p

b

s

Hupa Cons t tʰ t’ s n/m

Gordon basa kisi

‘to put a stick up high’ ‘to descend’

Word páˈʔai ʔáːpaˈhai kohoˈaipí hoˈaipi bágiˈái abaˈgi ˈsabí kaiˈtiabi ʔíˈsiːsí taːˈhoasi

Gloss ‘fish’ ‘bird arrow’ ‘eat’ ‘name’ ‘thief’ ‘toucan’ ‘angry, mean’ ‘buzzard’ ‘fat, body oil’ ‘sand’

Word naˈtɪl ʍɪˈtaʔ mɪˈtʰɪs ʍɪˈtʰaʔ t’e ʍɪˈt’ah ʍɪˈsɪts’ nɪˈsat tʃɪˈmel ʍɪˈnaːʔ

Gloss ‘they go about’ ‘my mouth’ ‘through it’ ‘my father’ ‘blanket’ ‘my pocket’ ‘my skin’ ‘it is deep’ ‘lizard’ ‘eye’

Western Apache Cons Word t pɪʔaːtɪ́ pɪtɪɬ tʰ nagotʰíːh pɪhastʰiːn t’ hat’íːko pɪt’iːs s/x jɪzɪsxíː hɪsiː n pɪnɪʔ

Gloss ‘it’s female’ ‘his blood’ ‘I hope it will rain’ ‘her old man’ ‘he wants’ ‘cottonwood tree’ ‘He killed it’ ‘I missed it’ ‘his land’

CHAPTER 6

Phonetics in Phonology: A Cross-Linguistic Study of Laryngeal Contrast Heriberto Avelino 1 Introduction During the last decade there has been an increase in phonetic studies of phonation in languages of the world (Gordon and Ladefoged 2001, Michaud and Mazaudon 2006, Keating et al. 2011). Languages of the Americas have had a preponderant role in providing data to show the possibilities of human languages in controlling laryngeal structures and dynamics for linguistic functions (Avelino 2010, Blankenship 2002, Ladefoged 1988, Silverman et al. 1995). Descriptive grammars of diverse language families in the Americas include labels such as ‘rearticulated’, ‘glottalized’, and ‘aspirated’, among others, to describe phenomena that can be interpreted as the contrastive use of laryngeal settings in producing vowels. These accounts, based on the ample experience of fine fieldwork linguists, while not being intended to be phonetically elaborated, strongly suggest that in some languages there is, at least, a triple-way contrast in phonation, modal-laryngealized-aspirated, as in some Otomanguean and some Mixe-Zoquean languages, while others have only a modal-­laryngealized contrast, as in Mayan or Nadahup, for instance. However, phonation is not the only prominent laryngeal feature in languages of the Americas; early studies such as Sapir (1922) and Pike (1948) described complex tone patterns in the languages of the new continent. Tone can be found throughout the Americas, from North American to Mesoamerican to Amazonian languages, among * I would like to thank the speakers of the languages who participated in this study Ana Daysi Alonso, Atanasio Dzib, Dianela Marin, Engracia Perez, Estela Canseco, Federica Diaz, Francisco Limeta, Jose Bollo, Margarita Cortez, the late Mario Molina, Pascual Vera, Rodrigo Martinez, as well as those who preferred to remain anonymous. Thanks to Jaime Morales, Joel Aquino, Juana Vazquez, Victor Canto, Xilonen Luna for facilitating the fieldwork; I am grateful for the comments, suggestions and discussion to Christfried Naumann, Christian Dicanio, Didier Demolin, Donca Steriade, Frank Seifart, Harald Hammarström, Henrik Bergqvist, Ian Maddieson, John Ohala, Jonathan Amith, Larry Hyman, Lyle Campbell, Martin Haspelmath, Melissa Freizer, Pam Munro, Paul Haggerty, Peter Ladefoged (✝), José Elías-Ulloa, Ryan Shosted, Søren Wichmann, Susanne Michaelis, the members of the UCLA American Indian Seminar and the Seminar at the MPI-EVA, Leipzig. I am the sole responsible for the contents of this article. © koninklijke brill nv, leiden, ���6 | doi ��.��63/9789004303218_007

158

Avelino

others, in systems that show a great divergence on a number factors such as the number of tones, the grammatical use of tone, and morphophonological sandhi-like processes. Although, in general, it is frequent to find voice qualities associated with different tones (e.g. creaky voice is often associated with low tone (Huffman 1987, Cao and Maddieson 1992)), what makes languages of the Americas special is that to some extent tones and phonation are independent features, so that in some languages all tone-phonation combinations are attested.1 Indeed, this pattern is predicted from a purely phonetic perspective. Ladefoged (1973), for instance, pointed out that the ‘features of the larynx’ can be controlled independently and therefore can coexist in a single phonological system (see also Hirose 1997). Later work (Silverman et al. 1995, Silverman 1997, Blankenship 2002, inter alia) further showed that the coexistence of underlying features does not entail simultaneity, but the features can be implemented phonetically as ‘phases’ across the duration of the vowel, so that the course of non-modal phonation does not persist through the entire vowel. In spite of the relative independence of the features of the larynx, and considering the recent findings on the timing in the implementation of non-modal phonation, I would like to add another piece to the discussion by arguing in this chapter that the phonetic implementation of tone and phonation, in particular the loci or anchoring of modal and non-modal phonation within the span of the vowel, depends also in part on the phonemic status of these features in the particular language(s). More specifically, I claim that although purely phonetic considerations (e.g. bio-mechanical or acoustic) give solid grounds to the explanation of tendencies in sound patterns, they cannot alone explain nor predict the patterns of allophonic variation observed across languages. Thus, the goal of this chapter is twofold: First, to provide an analysis of the phonetic properties of non-modal phonation in three languages, Yalálag Zapotec (Otomanguean), Yucatec Maya (Mayan) and Santa María Ocotepec Mixe (Mixe-Zoquean); and second, to test the phonological hypothesis that the patterns of phasing of tone and laryngealization throughout the vowel depend on the number and inventory of tone categories, so that in languages with crowded inventories, as in Yalálag Zapotec, in which both laryngeal dimensions are contrastive, the underlying features do not overlap. The null hypothesis predicts that the phonetic implementation of tone and phonation will be maintained, while 1  Thanks to Leo Wetzels for pointing out that other languages, such as Bor, a dialect of Dinka, spoken in South Sudan contrasts three tones and four types of phonation. Indeed, other languages of the world may have orthogonal contrasts, however, in languages of the Americas this property is spread all over the continent including different unrelated families and linguistic areas. The status of this feature, whether it is a contact phenomenon or can be traced back to an old genealogical trait deserves further investigation.

159

Phonetics In Phonology

­ eutralization will be disfavoured. The alternative hypothesis claims thus, that n in languages with a simpler tone system and contrastive laryngealization, like Yucatec Maya, there is the possibility of implementing laryngeal as pitch modulations rather than with creakiness; finally, in a language with no tonal contrasts, like Santa María Ocotepec Mixe, variations in pitch will be exploited to enhance the underlying contrasts MODAL vs. LARYNGEAL. The rest of the chapter is organized as follows. In section 2, I give a brief background of the languages studied, in section 3, I describe the methods and language materials, in section 4, I present the results, and in section 5, I finish with a discussion about the theoretical implications derived from the empirical findings of the three languages investigated. 2 Background A number of languages of the Americas show an orthogonal contrast between two laryngeal features: modal vs. non-modal phonation and tone. In this section I present a phonetic study addressing the patterns of laryngeal contrast observed in three Mesoamerican languages: Yalálag Zapotec (YZ), Yucatec Maya (YM) and Santa María Ocotepec Mixe or simply Ocotepec Mixe (OM). The three languages examined here YZ, YM and OM contrast modal and laryngealized vowels; however, they crucially differ in the status of lexical tone. In YZ there is a contrast between high, low and falling tones; YM has phonemic high and low tones only on long modal vowels; and OM does not have phonemic tone, but has a modal-laryngealized phonemic contrast. Table 6.1 illustrates the contrasts of laryngeal features in the three languages. TABLE 6.1

Tone and phonation contrasts in Yalálag Zapotec, Yucatec Maya and Ocotepec Mixe Tone

Phonation Modal

Laryngealized

Yalálag Zapotec

High Low Falling

zé bà bê

‘each’ ‘tomb’ ‘echo’

zḛ́ bà̰ bḛ̂

‘wall’ ‘animal’ ‘morning’

Yucatec Maya

High Low

t�ʃáːk t�ʃàːk

‘rain’ ‘parboil’

t�ʃa̰ːk

‘amaranth’

taːk

‘mother’

ta̰ːk

‘cornfield’

Ocotepec Mixe

160

Avelino

2.1 Yalálag Zapotec Yalálag Zapotec or Di’ll Wra’ll is an Otomanguean language spoken in Villa Hidalgo, in the Municipality of Villa Alta, in the highlands of Oaxaca, Mexico by 2115 people (INEGI 2002). Previous descriptions of Zapotecan languages mention a ‘rearticulated’ vowel type that has been traditionally represented as a sequence of two vowels with an intervening glottal stop (Long 1999, López Cruz 1997, Munro 1999, Nellis and Nellis 1983). Nevertheless, there are a number of phonological arguments to support a laryngealized non-modal phonation type as the underlying feature of ‘rearticulated’ vowels. First, there is a minimality constraint on stressed syllables to be bi-moraic (Avelino Becerra 2004). Although there is no length contrast, laryngealized vowels constitute a single heavy syllable, not two syllables as the practical orthography vʔv may suggest.2 Second, the two vowels of the ‘rearticulated’ type always have the same quality, a phenomenon identified as translaryngeal harmony (Steriade 1987). This is in sharp contrast with underlying vʔv sequences which might have different vowel qualities.3 This is typical, for instance, of the series of person clitics attached to the verb stem: vʔ]stem = vʔ]person (-áʔ 1SG, -óʔ 2SG, -èʔ 3SG.FORMAL) where each vowel belongs to a different morpheme and forms two separate syllables, as in examples (1) and (2).4 (1) Ll-achayra’-o’ HAB-suffer-2SG ‘You are suffering.’ [ʒàt�ʃâjràʔóʔ] (2) B-a-yilj-o’-e’ perf-progress-look.for-2SG-3SG.FORMAL ‘You used to look for him.’ [βàjìlʒòʔèʔ]

2  I carried out a series of uncontrolled experiments that might be considered as further evidence showing that rearticulated vowels are a single complex unit, rather than a sequence of independent vowels. Some speakers were instructed to tap, whistle or hum words containing rearticulated vowels. Speakers produced rearticulated words with a single whistle, hum or tap, just as they did with monosyllabic words with modal vowels. 3  See (Borroff 2007) for similar facts in Yatzachi Zapotec. 4  Glottal stop is contrastive en Zapotecan languages. The fact that these forms fail to show translaryngeal harmony gives evidence of its phonemic status as a consonantal segment rather than a laryngeal vocalic feature.

161

Phonetics In Phonology TABLE 6.2 Tone and phonation contrasts in Yalálag Zapotec Tone

Phonation

 

Modal

Laryngealized

High  

zé jín

‘each’ ‘a chest of clothes’

zḛ́ jín̰

‘wall’ ‘chili’

Low

bà

‘tomb’

bà̰

‘animal’

 

gà

‘nine’

gà̰

‘basket’

Falling

bê

‘echo’

bḛ̂

‘morning’

Third, rearticulated vowels only have one tone-bearing unit; if there were two vowels in the rearticulated nucleus, one would expect more combinations of tones that in fact are not attested. These arguments, together with external evidence from similar descriptions of other Zapotecan languages (López Cruz 1997, Munro 1999, Nellis et al. 1983) make the laryngealized phonation a more parsimonious analysis than the sequences of two vowels interrupted by a glottal stop.5 The contrast between modal and laryngealized phonation of YZ is exemplified in Table 6.2.6 2.2 Yucatec Maya Yucatec Maya or Màaya t’àan is a Mayan language spoken by about 706,000 people in the Yucatan Peninsula, Mexico and Belize (Lewis 2009). Yucatec Maya together with Lacandon are the only languages of the family that have developed contrastive tone.7 5  Pierrehumbert and Talkin have discussed that glottal stops are rarely realized with a full oral constriction but often produced as lenited creakiness (Pierrehumbert and Talkin 1992). The phonetic realization of underlying sequences /vʔv/ in Yatzachi Zapotec discussed by Borroff is very similar to the production of underlying laryngealized vowels in YZ. 6  For the time being I keep in this table the traditional ‘rearticulated’ label and the corresponding transcription with a glottal stop. 7  In Southern Lacandon tone is preserved including voice alternations, but in Northern Lacandon the contrasts have been eroded (Berqvist, p.c.). There are early reports of other mayan languages as having tone: Mocho, Tuzantec and Tzotzil. Most recent descriptions of Mocho (Palosaari 2002) conclude that the language has also innovated incipient tone. Indeed, the evidence shows that tone is not productive in the lexical contrast in these languages.

162

Avelino

TABLE 6.3 High-Low contrast in Yucatec Maya High

t�ʃáːk mí:s ʔéːk’ ʔáːk

‘rain’ ‘to sweep’ ‘black’ ‘turtle’

Low

t�ʃàːk mìːs ʔèːk’ ʔàːk

‘parboil’ ‘cat’ ‘star’ ‘herb’

TABLE 6.4 Modal-Rearticulated contrast in Yucatec Maya Modal

paʃ pèːt�ʃ pòːt

‘music’ ‘tick’ ‘to pierce’

Rearticulated

paʔaʃ peʔet�ʃ poʔot

‘break’ ‘rough’ ‘spider’

Yucatec Maya vowel nucleus includes modal long vowels, short vowels and the so-called ‘rearticulated’ or ‘broken’ nuclei (Pike 1946, Blair and VermontSalas 1965, Fox 1970, McQuown 1970, Fisher 1976), inter alia. The High-Low tone contrast occurs only on long modal vowels and it is neutralized in ‘rearticulated’ and short vowels. Examples of these contrasts are shown in Table 6.3 and Table 6.4 above. Yucatec Maya exploits both the tone contrast High vs. Low and the phonatory contrast of modal vs. rearticulated in its lexicon as well as in the inflectional verb system. The phonetic implementation of this three-way prosodic opposition is similar, in verb inflection and in the lexicon. The received wisdom in grammatical studies is that tone and rearticulated vowel type contrasts are used in the expression of voice inflection paradigms in transitive verbs (Bricker et al. 1998, Orie and Bricker 2000). Accordingly, the active voice carries the underlying form of the verb root, the antipassive has low tone, middle Fieldwork by Avelino on San Bartolo Tzotzil in 2010 did not reveal reliable tone ­contrasts. In so far as we know Yucatec and Southern Lacandon are the only Mayan languages with full documented productive contrastive tone in the lexicon and in the morphology.

163

Phonetics In Phonology TABLE 6.5 Verb inflection illustrating the vowel nuclei alternations of tone and laryngealization in Yucatec Maya Active

Antipassive

Middle

Passive

Gloss

tin tàab cháach tin taʔak

tàab nahen chàach nahen taʔak nahen

táab cháach taʔak

taʔabi chaʔach taʔakbi

‘to graft’ ‘to grasp’ ‘to hide’

voice is marked by high tone and the passive voice changes the nucleus of the root to a rearticulated vowel. Table 6.5 illustrates the three tone classes of verbs with Low, High and ‘rearticulated’ nucleus. As the table indicates, rearticulated vowels undergo no additional prosodic changes across the paradigm. Although this description has been repeated for a long time in several works, the study of Avelino et al. (2011),8 is the first in-depth investigation of the phonetic properties of phonation and pitch involved in the grammatical alternations of Yucatec Maya. Its results depart from previous accounts showing that the underlying tone distinction is maintained throughout different grammatical voices except for low tone in the passive, in which case the vowel becomes laryngealized. However, one of the the most important findings from Avelino et al. (2011) for the present study is that—contra previous descriptions maintaining that passive voice would be pronounced with a ‘rearticulated’ vowel nuclei—these forms are produced instead with falling tone but not necessarily with the prototypical rearticulated vowel [vʔv].9 2.3 Santa María Ocotepec Mixe Santa María Ocotepec Mixe or Ayuuk, is a Mixe-Zoquean language spoken in the district of Totontepec Villa de Morelos, Northeastern Oaxaca, Mexico.10 Like other Mixe languages (cf. Elson (1992), Wichmann (1995)), it has a complex vowel system. The variety of Mixe to which OM belongs (North Highlands) is the most complex of the Mixe-Zoquean family. It includes nine vowel qualities and a series of long, ‘aspirated’ and ‘rearticulated’ vowels in addition to the 8  Results often cited in the literature from an earlier report as Avelino et al. (2007). 9  See also Frazier (2011) and this volume for additional data corroborating and expanding these findings. 10  The number of speakers for the Totontepec Mixe variety is 5,200 (Lewis 2009). Santa María Ocotepec Mixe seems to be the same language or very close to Totontepec Mixe.

164

Avelino

vowels with modal phonation and normal length /i  i̵ e ǝ a æ u o ɔ/.11 In this paper, I will concentrate only on the contrast between modal and rearticulated vowels exemplified in Table 6.6 below. A phonological analysis showing the phonemic status of rearticulated vowels in OM is out of the scope of this paper, so here, I will consider the working hypothesis that these vowels correspond to laryngealized non-modal phonation (see Dieterman 2002 for a similar claim for Isthmus Mixe). For the present study it would suffice to provide two arguments in favor of the phonemic status of laryngealized vowels as opposed to the representation of a sequence of three segments. First, just like in YZ and YM, the rearticulated vowels are tautosyllabic, and the two vocalic portions of the reaticulation have the same vowel quality. Second, most of the roots are monosyllabic in Mixe, so if rearticulated vowels were to be considered as two syllables it would be a striking exception to the overall shape of the Mixe lexicon. This brief summary of the phonemic status of laryngealization and tone in the three languages studied introduces the basic information to develop the phonetic analysis properly in the following section. 3

Phonetic Analysis

3.1 Methods and Materials Six speakers of YZ (3 females, 3 males); eight speakers of YM (3 females, 5 males); 3 speakers of OM (2 females, 1 male) were recorded. The consultants were born and raised in their hometowns and use their native language on a daily basis. All the participants are bilingual in Spanish, but their first and native language is dominant. The same protocol was followed for the recording and analysis of the three languages: Consultants were instructed to read and repeat a list of words that reflected the contrasts of interest. Between 3 and 6 repetitions of each word were recorded. Utterances were recorded digitally. To obtain as much of the acoustic signal as possible from the glottal source, a microphone 11  In fact, Mixe vowel systems, including OM, are even more complex: grammatical studies report additional series of vowel nuclei so that vowels in Mixe languages altogether have: (1) short modal vowels, (2) long modal vowels, (3) short modal vowels aspirated towards the offset, (4) long modal vowels aspirated towards the offset, (5) short modal vowels checked by a glottal stop at the offset, (6) long modal vowels checked by a glottal stop at the offset, (7) rearticulated vowels, (8) rearticulated vowels with an additional aspiration towards the offset. See Crawford (1963), Reyes Gómez (2009), Schoenhals and Schoenhals (1965), Suslak (2005) for Totontepec Mixe.

165

Phonetics In Phonology TABLE 6.6 Modal-Rearticulated contrast in Ocotepec Mixe Modal

tak hæm pet tsǝk tsiv hi̵ts ʔok tæk muk

‘bald’ ‘it is over there’ ‘climb up’ ‘love’ ‘pitcher’ ‘dough’ ‘grandmother’ ‘old’ ‘together’

Rearticulated

taʔak pæʔæm peʔet tsǝʔǝk tsiʔiv i̵ʔi̵ts koʔok tæʔæk muʔuk

‘milpa’ ‘nest’ ‘sweep’ ‘cure’ ‘squash’ ‘vomit’ ‘fruit’ ‘clavillo plant’ ‘brave’

was set at a distance of 3 cm in front of the speaker. Words were pronounced in carrier sentences of the type “let us say five times”. The sentences did not include non-modal phonation that may affect the target word. Fundamental frequency and three spectral measures, H1*-H2*, H1*-A1*, H1*-A3* (Hanson 1995; 2001), were taken from a fast fourier transform calculated over a 26 ms window at three equidistant points within a vowel. Fundamental frequency measurements were taken at the same three points in the vowel. The measures were obtained with the Praat software (Boersma and Weeninck 2013). 4 Results The first and most conspicuous result was the wide variation of the phonetic patterns of laryngeal vowels, both across-subjects as well as within subjects in each language. Laryngealized vowels in the three languages were produced in a continuum ranging from a vowel interrupted by a full glottal stop, to a vowel with creaky phonation restricted to the middle portion, and vowels with attenuated creakiness in the second half. Figure 6.1 illustrates the patterns found in the three languages. It shows three columns, one per language; the rows are canonical examples of the variation found for laryngeal vowels. The majority of laryngealized vowels in YZ were produced with three phases of phonation starting with a modal portion followed by a section of creakyness and a final section with modal phonation. I will call this the ‘trough pattern’ of nonmodal phonation. The top figure (a) in the column of YZ shows the canonical

166

Avelino

‘­rearticulated’ type, with a clear glottal stop and two vocalic gestures of modal phonation at the beginning and towards the end of the vowel. The second figure below, (b) in this column, shows again the three phases already described, with the difference that the middle portion, indicated by the arrow, has a prominent creakiness as it can be seen from the wide spaced cycles (about six cycles in this example). The third panel (c) in the column shows also a similar three-phase pattern, albeit creakiness was attenuated. The second column of the figure illustrates the patterns found in YM. The first panel (d) corresponds to the typical ‘rearticulated’ vowel. The other two patterns found in YM are characterized by an initial vowel section with modal phonation and ending with creakiness, as indicated by the arrow in (e). I will call this the ‘laryngealfinal’ pattern of non-modal phonation. Notice that, just as in YZ, in YM there are instances in which the laryngealization is reduced significantly, as shown in the bottom figure of the YM column in panel (f). The last column of the figure shows representative examples of the patterns found in OM. The top spectrogram in (g) corresponds to the cases in wich laryngealization was very prominent. In these cases, laryngealization was present from the middle of the vowel and extended throughout the end. i.e. a ‘laryngeal-final’ pattern. Other pronunciations showed the trough-pattern illustrated by the figure in the second panel (h). Finally, a great number of instances were pronounced with weak laryngealization as in the bottom figure of the column in panel (i). 4.1 Fundamental Frequency The F0 contours of laryngealized and modal vowels differed in the three languages. I will report the results of males and females to account for the corresponding obvious sex differences in pitch. Figure 6.2 summarizes the findings of pitch by sex and vowel type. Yalálag Zapotec. YZ females showed a clear separation of high and low tones across both phonation types unlike the data for males which seem to have closer F0 values in both phonation types, with exception of the beginning of the vowel on which the differences are maximized. The F0 of laryngealized vowels in both, males and females alike, showed a troughpattern with the lowest pitch value occurring in the middle portion of the vowel. In contrast, the pitch frequencies of modal vowels were either flat or slightly rising towards the end.12

12  To facilitate the comparison with YM I will only include here the discussion of high and low tones. The reader can refer to Avelino Becerra (2004) and Avelino (ms) for an account including the falling tone.

167

Phonetics In Phonology

FIGURE 6.1 Variation of laryngealized vowels in YZ, YM and OM. laryngealized modal high laryngealized females modal low laryngealized modal high laryngealized males modal low laryngealized females modal

laryngealized males modal laryngealized females modal

laryngealized males modal Hz

initial

mid

final

FIGURE 6.2 F0 contours of YZ, YM and OM by phonation types.

168

Avelino

Yucatec Maya. In YM high and low tones are well differentiated in both sexes. The most prominent difference is that females showed a falling contour whereas males had a more steady level pitch. Overall, the F0 differed substantially according to phonation type and in the case of modal vowels whether it was high or low tone. The figure showed a slight rise from vowel onset to offset in high tone, and a slight fall in low tone for females. In contrast, the F0 of laryngealized vowels started at the same level as high tone and reached lower frequencies than those of low tone vowels. For males, the F0 of laryngealized vowels started at frequencies higher than high tone vowels and dropped towards the end of the vowel (see section 5 for further discussion). Taking these obervations together, laryngealized vowels can be characterized by a dramatic falling contour in both females and males. Ocotepec Mixe. A consistent trend was found a across speakers in which laryngealized vowels had a falling contour, whereas modal vowels had a rising contour. 4.2 Spectral Measurements Overall, the most conspicuous finding is that laryngealized vowels showed a pattern in which the last portion of the vowel had increased properties of creakiness in contrast with the initial section, which showed properties similar to those of modal vowels at equivalent points in the vowel. A trough-pattern of phonation revealed by spectral measurements was consistent in YZ and observed also in YM and OM. Nevertheless, one of the findings of this study concerns a methodological/technical issue: Although it is possible to observe general tendencies of spectral differences that could be correlated with phonation type, the data showed that not every measure is a suitable predictor of phonation type in every language, i.e. for a given language there are measurements that are better indicators of a difference between laryngealized and modal phonation.13 The measurement that reflected better the overall pattern in the three languages was H1*-A1*. In general, the results of H1*-H2* and H1*A3* were less reliable in differentiating modal and laryngealized vowels across the three languages. The H1*-A3* measurement was reliable in YZ but was nonsignificant in YM and only differentiated phonation in OM males. The results of the H1*-A3* showed a tendency to a trough-pattern. It is worth mentioning that H1*-H2*, a measurement which has been used as an indicator of differences in phonation (Hanson 2001, Stevens and Hanson 1995) did not show a homogeneous pattern across the three languages either; however, it revealed

13  Similar remarks have been made for other languages Esposito (2010), Keating et al. (2011), Garellek (2011). Spectral measurements might not be use indiscriminately.

169

Phonetics In Phonology ‒2.5 ‒5 ‒7.5 ‒10 ‒12.5 ‒15 ‒17.5 ‒20 ‒22.5 ‒25 ‒27.5 0 ‒2 ‒4 ‒6 ‒8 ‒10 ‒12 ‒14 5

6 4 2 0 ‒2 ‒4 ‒6 5 4 3 2 1 0 ‒1 ‒2 ‒3 ‒4 1 0 ‒1 ‒2 ‒3 ‒4 ‒5 ‒6 ‒7

0 ‒5 .10 .15 .20

initial

mid H1-H2

final

.25

initial

17.5 15 12.5 10 7.5 5 2.5 0 ‒2.5 ‒5 16 14 12 10 8 6 4 2 25 20 15 10 5 0 ‒5 ‒10

mid final initial mid H1-A1 H1-A3 modal laryngealized modal laryngealized females males

Yalalag Zapotec

Yucatec Maya

Ocotepec Mixe

final

FIGURE 6.3 Spectral correlates of modal and laryngealized vowels in YZ, YM and OM.

differences between modal and larygealized vowels in the following cases: YZ showed a tendency to a trough-pattern but in YM and OM the tendency was towards a laryngeal-final pattern, i.e. an increased creakiness towards the end of the vowel. Figure 6.3 summarizes the results of spectral measures. 5 Discussion Previous research has shown the relevance of the timing and phasing of vocalic non-modal phonation with respect to other features, notably tone, in languages where they are contrastive (Blankenship 2002, Di-Canio 2008, Garellek and Keating 2011, Silverman 1997).14 However, there is no prediction about what the anchor of non-modal phonation (e.g. creakiness, breathiness) throughout the overall timing of a LARYNGEAL vowel would be, nor what is the relation between non-modal phonation and other features such as tone. The findings of the present study allow me to propose a typology of the primary ways in which languages differ from one another in the implementation of non-modal phonation. It is maintained here that languages of the World will align to the 14  See also Herrera Z. (2000), Maddieson and Hess (1986) for further data.

170

Avelino

types summarized in Table 6.7. The table schematizes of the PHONATION BEAR UNIT (PBU) taking scope over the syllable nucleus.15 The PBU is divided in three sections or phases, each of which could function as anchors to the implementation of non-modal phonation. Thus, a first type of languages are those in which the non-modal phonation feature is anchored to the first portion of the PBU. The second type includes languages in which non-modal phonation is restricted to the last portion of the PBU. A third type, the ‘trough’ type describes languages in which the non-modal phonation is implemented in the middle portion of the PBU, whereas the extremes remain modal.16 A fourth possible type is represented by the ‘throughout’ type. In this case, nonmodal phonation extends throughout the entire PBU. A fifth possible type is the one in which the non-modal phonation in discontinuous across the PBU (i.e. expressed in the extremes of the PBU). This type is unattested, and moreover it is predicted that there would not be such a language. Unlike the ‘trough’ type, there is no functional explanation for a type in which the marked voice quality has to be implemented in the periphery of the PBU. The results obtained in this study suggest that the phonetic implementation of laryngeal features, tone and phonation, differ across languages according to the phonemic status of such features. In YZ the contrast between modal and non-modal phonation is consistently preserved across tone patterns. This indicates that there are no neutralizing processes commonly reported in other languages, but there is a fine control of the actions of the larynx to produce the implementation of the two features within the same vowel. The trough-pattern found in both fundamental frequency and spectral measures analyses is the crucial observation showing that in YZ the two features do not overlap in the phonetic implementation, but are distributed throughout the vowel: LARYNGEAL is produced in the middle of the vowel, and pitch differences are implemented at the beginning and end portions of the vowel. I have hypothesized elsewhere (Avelino Becerra 2004, Avelino ms) that this pattern could be the phonologization of the optimal phonetic arrangement, which, everything else being equal, would produce extremely demanding and contradictory laryngeal settings entailed by some of the features (i.e. high-creaky), and likely to be recovered in perception (Silverman 1997). The data analysis 15  It is assumed that the scope of the PBU is language specific. It could range from the syllable nucleus, or rhyme, but could also be other prosodic unit such as the mora, the entire syllable or even the word and beyond. 16  One could predict logical possibilities in which even that another voice quality type could also be adjacent to the targeted non-modal phonation, for instance a sequence of modalcreaky-breathy. Such types of languages are still to be documented.

171

Phonetics In Phonology TABLE 6.7 Typology of phasing and anchoring of non-modal phonation

Initial

← Phonation Bear Unit → Mid

Final

non-modal initial non-modal final trough throughout non-modal extremes

showed that in YM and OM modal and laryngealized vowels are not only distinguished by their spectral properties alone but the modulation of pitch patterns are also used as cues to signal the contrast.17 It was peculiar in YM that laryngealized vowels showed higher F0 values than the corresponding modal ones. This result was unexpected as laryngealized vowels are usually associated with creakiness, which often entails a lower frequency. One possible interpretation of the data is that YM speakers implement the laryngealized category not with a loose tension of the vocal folds, but with an increased tension of the whole laryngeal structures, including of course the vocal folds, which, in turn, produces the global effect of an increased fundamental frequency. In fact, the auditory impressions at the moment of making the recordings in fieldwork was close to this description, this is to say, the cover term ‘rearticulated’ in YM does not correspond to creakiness, but may be better categorized phonetically as ‘pressed’ or ‘tense’ voice (Laver 1980).18 If this were the case, the greater 17  A preliminary perceptual study manipulating the stimuli in such a way that the laryngeal constriction ranged from glottal stop to heavy laryngealization to modal voice values across three pitch categories High, Low and Falling has shown that in the three languages stimuli of a greater constriction is identified primarily as the canonical LARYNGEAL category (i.e. the rearticulated form [vʔv]) across the three tone categories. However, stimuli with attenuated creakiness anchored to the middle portion of the vowel [vv̰v] or plain modal vowels with a falling pitch were also identified as LARYNGEAL. These results indicate that in perception the target of laryngeal vowels is the typical rearticulated vowel, but a vowel with falling pitch and no laringealization cues could also have good identification scores. 18  The literature is rather misleading about labels and terms for phonation types (see Gerratt and Kreiman 2001). My impression from talking with fellow fieldworkers is that there is also an important confusion in the use of terms to refer to non-modal phonation.

172

Avelino

pitch found in ‘rearticulated’ vowels would be explained straighforwardly as F0 increases with subglottal pressure. However, one of the most interesting findings of the present study was that in YM and OM the difference between modal and laryngeal vowels was often more reliable in the F0 patterns than in the spectral differences. As the results showed, in these two languages there was a tendency to produce underlying ‘rearticulated’ nucleus as a laryngealfinal pattern, i.e. as a sequenced gesture of a modal vowel followed by nonmodal phonation. There is a phonetic explanation to this pattern: The relative stiffness of the vocal folds determines in part the rate of vocal fold vibration. Decreasing the vocal folds stiffness entails also the shortening and thickening of the vocal folds area/length in creaky phonation; this, in turn, will produce a decreasing in pitch. In this way, a falling contour will be produced as the result of a transition from modal to non-modal phonation. The interesting fact, however, is that a falling tone contour which is non-contrastive in both languages, is used as the actual exponent of the underlying contrast in phonation. I would like to offer an explanation to this problem. Stevens and Keyser (2010) introduced the notion of feature enhancement in which some acoustic/articulatory non-contrastive features are superimposed to underlying specifications to enhance their perceptibility. The data suggests that YM and OM speakers are reinterpreting the falling pitch contour as an inherent property of rearticulated vowels in addition to their spectral properties. If the acoustic target of vʔv is ‘undershot’ (Lindblom 1990; 2004) a coarticulatory effect would yield acoustic reduction, which, in turn, will increase its confusability with modal vowels. In that case, other sub-phonemic cues, crucially pitch, may help to recover the contrastiveness of the phonemic category. The data presented here seem to show evidence along these lines, and further suggest a change in progress in YM and OM: A vʔv target is being replaced by a falling pitch target, which originally was just a sub-phonemic feature that increased the perceptual salience of rearticulated vowels. If the change in pitch entailed by ‘rearticulated’ vowels has been misperceived by YM and OM listeners, they might be shifting the phonemic contrast from phonation to tone. In the speech of contemporary YM In p ­ articular, impressionistic language descriptions wrongly collapse two different phonation types under the category of ‘laryngealization’ or even ‘creakiness’. Informally, I distinguish two types of laryngealization: on the one hand, there is the ‘bored’ laryngealization in which the vocal folds are abducted, and there is an overall lax tension of the larynx; on the other hand, there is the ‘angry’ laryngealization (like when people are grunting to express displeasure), in which there is a tight constriction of the vocal folds and larynx in general, and an increased subglottal pressure. The latter corresponds to pressed voice, whereas the former should be used for creakiness properly.

Phonetics In Phonology

173

and OM speakers/listeners, a sudden turning point in the spectral slope and pitch anchored to the middle of the vowel signals the underlying prototype of a laringealized vowel. This strong cue in the production would be enough for the listeners to recover the underlying laryngeal feature. Thus, these results indicate that the patterns of non-modal phonation are constrained by the presence of other laryngeal features, crucially, pitch. The typology of tone-phonation interaction is completed by languages in which none of the features is contrastive but have alternations that resemble the phenomena just described here. Elías-Ulloa (this volume) presents Capanahua and Shipibo as instances of such a type. According to his account, in these languages there is a ‘complete coalescence’ of glottal stops flanked by vowels which surface as laryngealyzed [v̰], only in unstressed positions, while the same string will surface as a ‘partially coalesced’ (represented by Elías-Ulloa as vʔ, but implemented phonetically as [vv̰]) with high pitch on the first vocoid portion, only in stressed syllables. Elías-Ulloa reasons that because in these languages there are no demands on the preservation of phonemic features, the Panoan pattern arises as a mechanism to avoid the weakening of prominent prosodic positions, a hypothesis that, he conjectures, could be extended to Yalálag Zapotec. A similar hypothesis has been put forward earlier by González (2003, 2005) for the Panoan family. In her account, segmental processes, such a glottal deletion (concomitant to coalescence) are primarily a rhythm-governed phenomenon that ensures the well-formedness of syllables within the feet in a strong-weak pattern.19 Contra this view, I shall offer an alternative account showing that rhythmic processes that produce consonantal alternations do not contradict but complement and are consistent with the explanation presented in this paper: that the alignment of non-modal phonation and pitch in Capanahua and Shipibo is a function of the implementation of underlying features, and their exponents in the phonetic implementation. Let us consider the arguments. First, Elías-Ulloa offers a explanation for Panoan languages that is consistent and strengthens the analysis offered for Mesoamerican languages in this paper. He says “Capanahua and Shipibo avoid the overlap of high pitch and laryngealized voice in prominent positions [i.e. vowels in stressed syllables]”. Because it is precisely stress what indicates the prominent position, which in turn is realized by high pitch and concurrent syllable weight, the anti-coalescence of Capanahua and Shipibo follows the very same heuristics of YZ, YM and OM: to align the realization of the exponents of the underlying 19  There are minor differences between González and Elías-Ulloa: González deletion of coda /ʔ/ corresponds to Elías-Ulloa’s coalesced forms; González concludes that rhythmic processes have the function of increasing the prominence of the head footed syllable.

174

Avelino

features to avoid mistmaches in the phonetic implementation. The generalization found in Mesoamerican languages holds: pitch and laryngealized voice have to be phased in such a way that they do not overlap each other; which syllable within a word is the locus of contrast is an independent but certainly important consideration. However, the fact that the contrast in the languages discussed is a root-controlled phenomenon does not predict the alignment nor the phonetic implementation and anchoring of the contrastive features, an issue unsolved by Elías-Ulloa’s proposal.20 Second, the data presented by Elías-Ulloa seems to run afoul his own proposal. According to the hypothesis in Shipibo a glottal stop (i.e. the laryngealization on the vowel) does not occur at all on stressed vowels; his Fig. 7.6 (this volume) intends to illustrate this behavior with the word /ʧɨː—ʔati/ → [ʧɨː.a̰ti]. However, the actual figure and the segmentation provided show clearly that the last portion of the vowel of the root is heavily laryngealized, rendering [ʧɨɨ̰:a̰ti]. Likewise, Figure 7.7 in ElíasUlloa (this volume) intends to show that the vowel with secondary stress in / ka∫kaka∫ka—ʔati/ → [ˈka∫kaˌka∫ka̰ˌʔati] is modal. However, the graphic shows that the first portion of the vowel surfaces, indeed, with a different quality, allegedly with creakiness [ˈka∫kaˌka∫ka̰ˌa̰ati]. A similar pattern is observed in his Figure 7.8, which could be represented more precisely in a narrow transcription as [ȡʑa .ˈbβɨɨ̰.ˈa̰ani]. The data indicates that there is coarticulation in positions that are not predicted by the hypothesis of anti-coalescence in prominent positions. Third, identifying that the contrasts occurs in prominent positions does not predict the actual patterns observed in the Mesoamerican languages investigated here, nor the types found across languages. The analysis maintained in this paper advocates for a dynamic perspective of phonology, in which phonology and phonetics is considered in a unified framework (Flemming 2001); in this context, the Capanahua and Shipibo data further support the thesis accounting for Mesoamerican languages, namely that the 20  Stress and tone in Panoan is a difficult and still controversial topic. Some authors, such as Loos (1969:186 and passim) would consider tone to be underlying. There are a number of exceptions observed by Loos if stress is considered to be underlying and high tone an associated sub-phonemic cue. In nouns high pitch occurs in stressed syllables and on “non-final post-stressed syllables” (:187) as in ˈsóntáko ‘young girl’; however, “in verbs the rule does not apply, for final unstressed syllables do not always have low tone” Compare ́ ï�́ ‘put it in the canoe’ (Loos 1969:188). Instead, “nánï�ḱ in ‘he put it in the canoe’ vs. ˈnánï�w Loos suggests that if tone is considered to be underlying “it is possible to conclude that stress is predictable on the basis of tone: the first high tone of a word is stressed” (:ibid see Loos for further derivational analysis of stress assignment). Other Panoan languages, such as Kakataibo, Amawaka, Yaminawa and Marinawa can be analyzed as H vs. ∅, and others, like Chakobo, with Low marked suffixes (Zariquiey (2012) and personal communication).

Phonetics In Phonology

175

arrangement of the patterns emerging in each language and the typology of feature anchoring are driven by the fundamental notion of phonological contrasts and their particular phonetic expression. 6

Concluding Remarks

The evidence discussed in this paper has shown that the phonetic implementation of LARYNGEAL in the three languages investigated, Yalálag Zapotec (Otomanguean), Yucatec Maya (Mayan) and Santa María Ocotepec Mixe (MixeZoquean) depends on the phonemic status of TONE. This is a generalization that is also observed in Nadahup, Chibchan, Arawakan and Tupian and might be extended to other languages of the Americas (see the overview by Storto and Demolin (2012)). In YZ a consistent trough-pattern was found, anchoring the creakiness to the middle portion of the vowel while leaving modal voice in the extremes of the vowel. It was argued that this pattern is an optimal configuration to the implementation of high, low and falling tones, and larygealization. In contrast, I showed that YM and OM with a simpler system or lacking tone contrasts at all, respectively, also has the trough-pattern as the canonical type, however, the phonetic implementation of LARYNGEAL can be dispensed with and substituted by concurrent sub-phonemic pitch cues, namely, the falling pitch contour which entails a shift anchored to the middle portion of the vowel and observed in the production of the laryngeal-final pattern. In sum, the implementation of tone and phonation in the languages investigated creates patterns that avoid the overlapping of their respective gestures, and at the same time, enhance their production and presumably their perception. References Avelino, H. 2010. Acoustic and electroglottographic analyses of nonpathological, nonmodal phonation. Journal of Voice 24 (3), 270–280. ———. Unpublished manuscript. The phonetics and phonology of non-modal phonation variability: Evidence from Yalálag Zapotec. Avelino, H., E. Shin, and S. Tilsen. 2011. The phonetics of laryngealization in Yucatec Maya. In H. Avelino (Ed.), New Perspectives in Mayan linguistics, pp. 1–20. Cambridge: Cambridge Scholars Publishing. Avelino, H., S. Tilsen, E. Kim, and J. Pynes. 2007. The phonetics of laryngealization in Yucatec Maya. In Paper presented at the 81st Annual Meeting of the Linguistic Society of America, Anaheim, California, pp. 1. LSA.

176

Avelino

Avelino Becerra, H. 2004. Topics in Yalálag Zapotec, with Particular Reference to Its Phonetic Structures. Ph. D. thesis, University of California, Los Angeles. Blair, R.W. and R. Vermont-Salas. 1965. Spoken (Yucatec) Maya. Book 1: Lessons 1–12. Chicago: University of Chicago, Dept. of Anthropology. Blankenship, B. 2002. The timing of nonmodal phonation in vowels. Journal of Phonetics 30, 163–191. Boersma, P. and D. Weeninck. 2003. Praat. doing phonetics by computer. Borroff, M.L. 2007. A Landmark Underspecification Account of the Patterning of Glottal Stop. Ph. D. thesis, Stony Brook University. Bricker, V.R., E. Po’ot Yah, and O. Dzul de Po’ot. 1998. A dictionary of the Maya language: as spoken in Hocabá, Yucatán; with a botanical index. Salt Lake City: Univ. of Utah Press. Cao, J. and I. Maddieson. 1992. An exploration of phonation types in wu dialects of chinese. Journal of Phonetics 20, 77–92. Crawford, J.C. 1963. Totontepec Mixe phonotagmemics. Summer Institute of Linguistics of the University of Oklahoma. DiCanio, C.T. 2008. The phonetics and phonology of San Martn Itunyoso Trique. Ph. D. thesis, University of California Berkeley. Dieterman, J. 2002. Secondary palatalization in Isthmus Mixe: A phonetic and phonological account. Ph. D. thesis, The University of Texas at Arlington. Elson, B.F. 1992. Language in context: Essays for Robert Longacre. In S.J.J. Hwang and W.  Merrifield (Eds.), Reconstructing Mixe-Zoque, pp. 577–592. Dallas: Summer Institute of Linguistics and The University of Texas at Arlington. Esposito, C.M. 2010. The effects of linguistic experience on the perception of phonation. Journal of Phonetics 38 (2), 303–316. Fisher, W.M. 1976. On tonal features in the yucatecan dialects. In M. McClaran (Ed.), Mayan linguistics, pp. 29–43. Los Angeles: University of California, Los Angeles. American Indian Studies Center. Flemming, E. 2001. Scalar and categorical phenomena in a unified model of phonetics and phonology. Phonology 18, 7–44. Fox, J.A. 1970. Proto-Mayan Accent, Morpheme Structure Conditions, and Velar Innovations. Ph. D. thesis, University of Chicago, Chicago. Frazier, M. 2011. Tonal dialects and consonant-pitch interaction in Yucatec Maya. In H.  Avelino (Ed.), New Perspectives in Mayan linguistics, pp. 21–20. Cambridge: Cambridge Scholars Publishing. Garellek, M. 2011. The timing and sequencing of coarticulated non-modal phonation in english and white hmong. Journal of Phonetics 40 (1), 152–161. Garellek, M. and P. Keating. 2011. The acoustic consequences of phonation and tone interactions in jalapa mazatec. Journal of the International Phonetic Association 41, 185–205.

Phonetics In Phonology

177

Gerratt, B.R. and J. Kreiman. 2001. Toward a taxonomy of nonmodal phonation. Journal of Phonetics 29 (4), 365–381. Gordon, M. and P. Ladefoged. 2001. Phonation types: a cross-linguistic overview. Journal of Phonetics 29 (4), 383–406. Hanson, H.M. 1995. Glottal characteristics of female speakers. Ph. D. thesis, Harvard University. ———. 2001. Towards models of phonation. Journal of Phonetics 29, 451–480. Herrera Z.E. 2000. Amuzgo and zapotec: two more cases of laryngeally complex languages. Anthropological linguistics 42, 545–563. Hirose, H. 1997. Investigating the Physiology of Laryngeal Structures. In W.J. Hardcastle and J. Laver (Eds.), The Handbook of phonetic sciences, Number 4 in Blackwell Handbooks in Linguistics, Chapter 4, pp. 116–136. Oxford: Blackwell Publishers Ltd. Huffman, M.K. 1987. Measures of phonation in Hmong. Journal of the Acoustical Society of America 81 (2), 495–504. INEGI. 2002. XII censo general de poblacio´n y vivienda, 2000. Perfil sociodemo­ gráfico. México. Aguascalientes, Ags.: Instituto Nacional de Estadística, Geografía e Informática. Keating, P., C. Esposito, M. Garellek, S. ud Dowla Khan, and J. Kuan. 2011. Phonation contrasts across languages. Proceedings of ICPhS 2011, 1046–1049. Ladefoged, P. 1973. The features of the larynx. Journal of Phonetics 1, 73–83. ———. 1988. Discussion of phonetics: a note on some terms for phonation types. 1988), Vocal physiology: Voice Production, Mechanisms and Functions, Raven Press, Ltd., New York, 373375. Laver, J. 1980. Phonetic Description of Voice Quality. Cambridge: University of Cambridge Press. Lewis, M.P. (Ed.). 2009. Ethnologue: Languages of the World (Sixteenth ed.). Dallas, TX, USA: SIL International. Lindblom, B. 1990. Explaining Phonetic Variation: a sketch of the H&H theory. In W.J.  Hardcastle and A. Marchal (Eds.), Speech Production and Speech Modelling, pp. 403–439. Dordrecht: Kluwer. ———. 2004. The organization of speech movements: Specification of units and modes of control. In J. Slifka, S. Manuel, and M. Matthies (Eds.), From Sound to Sense: 50+ Years of Discoveries in Speech Communication, Boston: MA. Long, R.C. 1999. Diccionario Zapoteco de San Bartolomé Zoogocho Oaxaca. Mexico City: Instituto Lingüístico de Verano. López Cruz, A. 1997. Morfología verbal del Zapoteco de San pablo Güilá. Bachelors Thesis. Escuela Nacional de Antropología e Historia. INAH SEP. México. Maddieson, I. and S. Hess. 1986. Tense and lax revisited: more on phonation type and pitch in minority languages in china. UCLA, Working Papers in Phonetics 63, 103–109.

178

Avelino

McQuown, N.A. 1970. El acento del Maya Yucateco. In Congreso Internacional de Americanistas: Lingüística e indigenismo moderno de América, Lima, Perú. Instituto de Estudios Peruanos. Michaud, A. and M. Mazaudon. 2006. Pitch and voice quality characteristics of the lexical word-tones of Tamang, as compared with level tones (Naxi data) and pitch-plusvoice-quality tones (Vietnamese data). In Proceedings Speech Prosody 2006, Dresden, pp. 823–826. Mouton. Munro, P. 1999. Di’csyonaary X:te’n. Di’zh Sah Sann Lu’uc. San Lucas Quiavin Zapotec Dictionary. Diccionario Zapoteco de San Lucas Quiavin. Los Angeles: Chicano Studies Research Center, UCLA. Nellis, N. and J.G. Nellis. 1983. Diccionario zapoteco de Juárez: zapoteco-español, españolzapoteco, Volume 27. México: Instituto Lingüístico de Verano. Nellis, N., J.G. Nellis, and D.A. Bartholomew. 1983. Gramática zapoteca. In Diccionario zapoteco de Juárez. Vocabularios y Diccionarios Indígenas, Volume 27 of “Mariano Silva y Aceves”. México: Instituto Lingístico de Verano. Orie, O.O. and V.R. Bricker. 2000. Placeless and historical laryngeals in Yucatec Maya. International Journal of American Linguistics 66 (3), pp. 283–317. Palosaari, N.E. 2002. Topics in Mocho’ phonology and morphology. Ph.D. thesis, The University of Utah. Pierrehumbert, J.B. and D. Talkin. 1992. Lenition of /h/ and glottal stop. In G. Doherty and D.R. Ladd (Eds.), Papers in Laboratory Phonology II, pp. 90–117. Cambridge University Press. Pike, K.L. 1946. Phonemic pitch in Maya. International Journal of American Linguistics 12, 82–88. ———. 1948. Tone languages: a technique for determining the number and type of pitch contrasts in a language, with studies in tonemic substitution and fusion. University of Michigan Press. Reyes Gómez, J.C. 2009. Fonología de la lengua Ayuuk de Alotepec, Oaxaca. Bachelors Thesis. Escuela Nacional de Antropología e Historia. INAH, SEP. México. Sapir, E. 1922. Athabaskan tone. American Anthropologist 24 (3), 390–391. Schoenhals, A. and L.C. Schoenhals. 1965. Vocabulario mixe de Totontepec. Instituto Lingüístico de Verano. Silverman, D. 1997. Laryngeal complexity in otomanguean vowels. Phonology 14, 235–262. Silverman, D., B. Blankenship, P. Kirk, and P. Ladefoged. 1995. Phonetic structures in Jalapa Mazatec. Anthropological linguistics, 7088. Steriade, D. 1987. Locality and feature geometry. Proceedings of the North East Linguistic Society 17, 595–619. Stevens, K.N. and H.M. Hanson. 1995. Classification of glottal vibration from acoustic measurements. In O. Fujimura and M. Hirano (Eds.), Vocal fold physiology. Voice quality control, pp. 146–170. San Diego, CA: Singular Publishing Group Inc.

Phonetics In Phonology

179

Storto, L. and D. Demolin. 2012. The phonetics and phonology of south american languages. In L. Campbell and V. Grondona (Eds.), The Indigenous Languages of South America: A Comprehensive Guide, Volume 2 of The World of Linguistics, pp. 331–390. Berlin: Mouton. Suslak, D.F. 2005. The future of Totontepecano Mixe: Youth and language in the Mixe highlands. Ph. D. thesis, University of Chicago. Wichmann, S. 1995. The relationship among the Mixe-Zoquean languages of Mexico. University of Utah Press. Zariquiey, R. 2012. The interaction of tone and stress in kashibo-kakataibo (panoan). Workshop “Tone: Theory and Practice”. Heriberto Avelino and Ian Maddieson, organizers. Max Planck Institute-Leipzig.

CHAPTER 7

The Role of Prominent Prosodic Positions in Governing Laryngealization in Vowels: A Case Study of Two Panoan Languages José Elías-Ulloa 1 Introduction There exist languages in which both tones and types of phonation contrast. In these languages, when vowels in non-modal phonation bear a contrastive tone, portions appear in modal voice. An illustrative case is Yalálag Zapotec, spoken in the state of Oaxaca, Mexico. According to Avelino-Becerra (2004), Yalálag Zapotec has three contrasting tones: high, low, and falling (e.g. /já/ ‘sweathouse’, /jà/ ‘bell’, /jâ/ ‘cane’). With regard to phonation types, Yalálag Zapotec contrasts modal vowels versus laryngealized vowels (e.g. [ze] ‘each’, [zḛ] ‘wall’).1 In minimal pairs of Yalálag Zapotec words that have laryngealized vowels but differ in terms of high, low or falling tone, those laryngealized vowels present portions of their duration in modal voice. Figure 7.1 shows the case of the minimal pair: /bà̰/ ‘animal’ versus /bá̰/ ‘smooth’. Both words have the same laryngealized vowel /a̰/. In the former case, there is a low tone while in the latter, a high one. The laryngealization in both vowels can be observed by looking at the vertical striations in the spectrogram. They indicate that the glottal pulses occur at a much slower and non-periodic rate. Interestingly, both vowels show sections of modal phonation, particularly from the beginning toward the middle of each vowel. * I would like to thank Carolina Gonzalez, Seunghun Lee, Heriberto Avelino-Becerra as well as the audience of Amazonicas III (UNAL, Bogota, Colombia) for their comments and feedback. I would also like to thank an anonymous reviewer for the comments and suggestions. This work has been supported by the National Science Foundation under Grant BCS-0966257. 1  Avelino-Becerra (2004:77) transcribes Yalálag Zapotec larygealized vowels with a superscripted glottal stop that interrupts the vowel. Thus, for instance, the word for ‘wall’ is transcribed as [zeʔe]. In this paper, I will transcribe them as vowels with the subscripted diacritic [˷] to emphasize the idea that these vowels behave as single phonological units.

© koninklijke brill nv, leiden, ���6 | doi ��.��63/9789004303218_008

Prominent Prosodic Positions In Governing Laryngealization

181

Figure 7.1 Yalálag Zapotec: /bà̰ / ‘animal’ versus /bá̰/ ‘smooth’ (Avelino-Becerra 2004:162).

In his article on laryngeal complexity in Otomanguean vowels, Silverman (1997:257) explains that the presence of modal voice in vowels with a phonological specification for non-modal phonation is a way of facilitating the maintenance and recoverability of other phonological contrasts. This occurs because non-modal phonation weakens the acoustic cues that help the listener recover phonological contrasts (see also similar remarks in Gordon 1998; Gordon and Ladefoged 2001; Blankenship 2002; Avelino-Becerra 2004, among others). Thus, in Yalálag Zapotec, the presence of modal voice in non-modal phonation vowels provides the listener with the opportunity to recover the cues to distinguish between high, low and falling tones. Although the presence of modal voice facilitates the listener’s recovery of acoustic cues—a task that is certainly more challenging during non-modal phonation—I disagree that the presence of modal phonation in non-modal vowels is necessarily related to the maintenance or recoverability of phonological contrasts. In this article, I will examine the case of two Panoan languages, spoken in the Peruvian Amazon: Capanahua and Shipibo. Neither has tonal or phonation contrasts. However, both behave like Yalálag Zapotec in that nonmodal phonation, although non-contrastive, is avoided in places where it can weaken prominent prosodic positions associated with high pitch. In Section 2, I examine the common phonetic realization of glottal stops as creaky/laryngealized phonation and present evidence for considering the glottal stop a consonantal segment in Capanahua and Shipibo. In Section 3, I provide a detailed characterization of the behavior of glottal stops in both Panoan languages. This section includes a discussion of their stress patterns since the phenomenon of glottal coalescence interacts with the metrical system in both cases. Finally, in Section 4, I discuss two approaches to account for the data presented. I argue in favor of considering prominent prosodic positions as the crucial and unifying force behind the behavior of vowel laryngealization

182

Elías-Ulloa

in languages that lack phonation or tonal contrasts as well as in those, like Otomanguean languages, that do have those contrasts. 2

On the Phonetic Realization of Glottal Stops

Before examining the cases of Shipibo and Capanahua, I will discuss the phonetic realization of glottal stops. Traditionally, a glottal stop is described as a regular stop consonant, that is, a segment during the articulation of which the airstream is stopped from coming out of the mouth by the closure of the vocal folds. However, the phonetic realization of glottal stops is not like the realization of other stop consonants. During the realization of a non-laryngeal stop, the articulators create a closure somewhere in the oral cavity, which completely stops the airflow. In a spectrogram, this closure is seen through the lack of acoustic energy, a white space as the one shown in Figure 7.2 for the realization of the voiceless bilabial stop [p]. In contrast, in most languages, the glottal stop tends to be realized as creaky voice on surrounding vowels, rarely as a complete glottal closure (Ladefoged and Maddieson 1996). The typical phonetic realization of a glottal stop looks more like a glottal glide as depicted in the spectrogram in Figure 7.3 through the conspicuously visible vertical striations that appear between the vowels (see more on the realization of glottal stops in Boroff 2007). By comparing the realization of both stop consonants in Figure 7.2 and Figure 7.3, one is left wondering in what sense the “glottal stop” is a stop or even a consonant.2 The answer to this question relies on the phonology of the language studied. In some languages, the evidence indicates that the glottal segment patterns as a consonant; in others, it behaves as a laryngeal feature associated to vowels or consonants. As we will see shortly, in Capanahua and Shipibo, glottal stops behave as consonants at the phonological level but tend to be realized as creaky voice on vowels or creaky glottal glides at the phonetic level. Thus, based on their phonological 2   For some languages like Blackfoot (Bortolin and McLennan 1995, Peterson 2004) or Munduruku (Lobato Picanço 2005), it has been reported that glottal stops tend to be realized with a full glottal closure. However, even in these languages the realization of glottal stops as laryngealization is possible. For example, Peterson 2004, based on Bortolin and McLennan 1995, mentions that there is an undergoing change in the phonetic realization of glottal stops in Blackfoot. While older speaker realize it with a full glottal closure, younger speakers realize it as creaky voice or they completely elide it.

Prominent Prosodic Positions In Governing Laryngealization

Figure 7.2 A typical phonetic realization of a bilabial stop [p] in an intervocalic environment.

Figure 7.3 A typical phonetic realization of a glottal stop [ʔ] in an intervocalic environment.

183

184

Elías-Ulloa

behavior, I consider glottal stops in both Panoan languages discussed in this study to be consonantal segments. The discrepancy between the phonological behavior and the phonetic realization observed in glottal stops is not unexpected. The prosodic context in which a segment occurs has a great impact on its phonetic realization (see Beckman 1998; Gonzalez 2003 for more examples of the interaction between prosody and segments). Thus, I depart from the idea that a phonological unit in the segmental inventory of a language has to be categorized once and for all as belonging to a single natural class, say, vowel, consonant or glide at all levels of representation. Different phonological requirements/constraints can force segments to change their nature. In Shipibo, for example, the voiced retroflex affricate, /ɖʐ /, can be realized, according to the prosodic environment, as an affricate, [ɖʐ ], in stressed and word-initial syllables but as a fricative [ʐ ] or even as a retroflex approximant with little or non-friction at all in unstressed syllables or word non-initial syllables (Elias-Ulloa 2011). Since I will be examining the cases of Capanahua and Shipibo, two very close Panoan languages, I will make explicit why I consider the “glottal stop” to be a consonantal segment, typically realized at the phonetic level as a glottal glide, in both languages and not, say, laryngealization associated with vowels, as is the case, for example, in Yalálag Zapotec. In both Capanahua and Shipibo, the evidence for the consonantal nature of the glottal stop comes from the interaction of syllabification, stress and processes of nasalization. Both languages have suffixes that begin with a glottal stop. When they follow a stem that ends in a consonant, the glottal stop of the suffix behaves as a syllable onset. Let us examine the data in (1) from Shipibo.3 In (1a), the root /bɨkʊn/ ‘myopic’ is followed by the verbalizer suffix /-ʔati/, which has an initial glottal stop. The result of combining these two morphemes is the form [bɨ.ˈkʊ̃ n.ʔa.ti] ‘to steam up’.4 The glottal stop of the suffix surfaces as the onset of the third syllable. We know this because the main stress falls on the second syllable. In Shipibo, the main stress occurs on the second syllable if it is closed. Otherwise, it falls on the initial syllable. In contrast in (1b), the suffix that follows the root does not have a glottal stop. In this case, the nasal consonant of the root becomes the onset of the third syllable. As expected since the second syllable is no longer considered closed by the phonology of the language, the main stress jumps back to the initial syllable. 3  A similar pattern can be found for Capanahua (see Loos and Loos 1998:193, Elias-Ulloa 2009:186–7). 4  Nasal consonants in coda position coalesce with the vowel in their rhyme. This yields the observed form in (1a): [bɨ.ˈkʊ̃ .ʔa.ti].

Prominent Prosodic Positions In Governing Laryngealization

185

The second piece of evidence that the glottal stop in (1a) behaves as a consonant comes from nasalization. Shipibo, as other Panoan languages, strongly nasalizes vowels that precede nasal codas. Moreover, when a nasal coda is followed by a glide (i.e. /j, w/) or a glottal stop, it does not appear as an independent segment. In that context, nasal codas completely coalesce with the preceding vowel (Elias-Ulloa 2011). In spite of this phenomenon of coalescence, the phonology of Shipibo still considers the resulting nasalized vowel as a sequence of a vowel followed by a nasal coda, as supported by the position of the stress on the second syllable. In contrast, in (1b), since the nasal consonant does occur as the onset of the third syllable, it does not trigger nasalization in the preceding vowel. If the glottal stop in (1a) were not a consonantal segment but a laryngeal specification of the following vowel, then we would wrongly expect the form *[ˈbɨ.kʊ.na̰.ti] to occur. This unattested form would have the stress on the initial syllable since the second syllable is open, and the nasal consonant would emerge as a full-blown onset nasal segment, as occurs in (1b). (1) a. /bɨkʊn -ʔati/ → [bɨ.ˈkʊ̃ n.ʔa.ti] →[bɨ.ˈkʊ̃ .ʔa.ti] (Shipibo) myopic –VZ1 ‘to steam up’ (Elias-Ulloa 2011)

b. /bɨkʊn -a/ → [ˈbɨ.kʊ.na] (Shipibo) myopic -PP2 ‘for one’s vision to become clouded over’ (PP2) (Elias-Ulloa 2011)

The phonotactics also provides evidence that the glottal stop behaves like a consonant. If the glottal stops of Shipibo were a laryngeal specification on vowels and, thus, not counted as consonantal segments able to occur in the syllable margins, then we would expect Shipibo to have closed syllables with shapes like *[CV̰ C], where the vowel appears with laryngealization and followed by a single coda. As long as the coda is [n], [s], [ ʃ ], or [ʂ], this syllable would meet all the phonotactic requirements Shipibo imposes on its syllables. However, that type of syllable is unattested in the language. The reason is that the glottal stop counts as a consonant, not as a laryngeal specification on vowels and therefore from the perspective of Shipibo phonology such a syllable shape would be considered to have an illegal complex coda: *[CVʔC].5 5  Although Shipibo does allow the occurrence of complex codas (Elias-Ulloa 2011), these are subject to strict restrictions: (i) the maximum number of consonants in a complex coda is two and (ii) the first consonant must be the nasal [n] and this should be followed by a fricative (e.g. [ˈĩns.pi] ‘head side’, [ˈpʊ̃ nʂ.tɨ.ti] ‘to cut the arm’).

186 3

Elías-Ulloa

Glottal Stops in Capanahua and Shipibo

In Capanahua and Shipibo, stress is predictable. Main stress falls on the second syllable if closed (see the data in 2b and 2c). Otherwise, by default, it occurs on the initial syllable (see 2a and 2d). Syllables with main stress always appear associated with a high pitch and longer duration. The data in (2) show that stress is predictable from the syllables that form the word. In (2a), the second syllable is open and thus the stress falls on the initial syllable. In contrast, in (2b), the same second syllable has received the ergative suffix, /-n/, which turns it into a closed syllable. This makes the main stress jump onto that syllable. In (2c), the second syllable is closed and, as expected, the stress occurs on it. However, when the suffix /-n/ is added, as in (2d), the second syllable of the word becomes open, and the main stress jumps back onto the initial syllable. (2) a. [ˈba.kɨ] ‘Child’

b. /bakɨ -n/ → [ba.ˈkɨn] ‘Child’ (ERG)



c. /tʃaɖʐaʂ/ → [tʃa.ˈɖʐaʂ] ‘Catalan’ (sp. of bird)



d. /tʃaɖʐaʂ -n/ → [ˈtʃa.ɖʐa.ˌʂan] ‘Catalan’ (sp. of bird—ERG)

In both languages, words longer than two syllables can present secondary stresses (see 2d, 3b, 4a–b). As occurs with main stress, closed syllables with secondary stress tend to be marked by a high pitch (Elias-Ulloa 2011). Thus, words can present more than one high pitch. However, the culminativity of main stress is reflected through two cues. First, high pitch associated with secondary stresses always appears downstepped compared to the high pitch associated with the syllable bearing the word-main stress. Second, the vowel of main-stressed syllables shows a greater duration than vowels of syllables with secondary stress and, in turn, vowels of the latter tend to show a greater duration than vowels of unstressed syllables. Secondary stress can also fall on open syllables but they would rarely show a high pitch. Instead, their vowels tend to have a greater duration than vowels in unstressed syllables.

Prominent Prosodic Positions In Governing Laryngealization

187

A few suffixes have lexical stress. For example, when the imperative suffix /-ˈwɨ/, which comes marked with a lexical stress, combines with the verb root /pi/ ‘to eat’, the main stress occurs on the second syllable instead of the expected initial one: [pi.ˈwɨ] ‘eat!’ However, when the verb /pi/ combines with the tense suffix /-kɨ/, which does not have a lexical stress, the resulting form shows the main stress on the initial syllable: [ˈpi.kɨ] ‘ate’. If the suffix with lexical stress occurs after the second syllable of the word, and it is not adjacent to the syllable with main stress, then the suffix with lexical stress appears bearing a secondary stress. This can be observed when the verb /pi/ ‘eat’ combines with the causative suffix /-ma/ and the imperative yielding the form: [ˈpi.ma.ˌwɨ] ‘make (him) eat!’ If the suffix with lexical stress occurs adjacent to the syllable carrying the main stress, it surfaces unstressed. We can see this behavior if the verb /pi/ ‘eat’ combines with the suffix for third person plural /-kan/ and the imperative: [pi.ˈkan.wɨ] ‘eat!’ The high pitch associated with an initial syllable bearing main stress can spread over the adjacent post-tonic syllable. Elias-Ulloa (2011) found that this phenomenon mainly occurs in the speech of Shipibo female speakers. Thus, for example, the word /ʃʊntakʊ/ ‘young woman’, uttered by Shipibo-female speakers, can optionally surface as: [ˈʃʊ́ n.tá.kʊ] or [ˈʃʊ́ n.ta.kʊ]. For male speakers, the high pitch associated with main stress does not spread: [ˈʃʊ́ n.ta.kʊ]. Similar patterns have also been reported for Capanahua (Loos 1969). Even in the case where the high pitch spreads, the vowel of the main-stressed syllable is readily identified by its greater duration compared to the vowel of the posttonic syllable. To sum up, neither of the two Panoan languages discussed in this study has tonal/pitch contrasts. The distribution of stressed syllables, with the exception of some morphemes that bear lexical stress, is generally predictable from syllable structure. All lexical words must obligatorily have a high pitch associated with the main-stressed syllable. Closed syllables bearing secondary stress attract additional high pitches. In turn, this means that the distribution of high pitches is predictable from the distribution of stressed syllables. The syllable bearing main stress is the most prominent syllable in the word. It shows the vowel with the greatest duration, and its high pitch is conspicuously higher than other high pitches that might be present in the word. Shipibo and Capanahua have latent consonants, namely, consonants that cannot surface as codas due to phonotactic restrictions. Both languages only allow as codas sibilant and nasal consonants (in addition, Capanahua allows the glottal stop). Latent consonants appear once they manage to occupy the onset of a syllable. Although latent consonants cannot surface as codas, the

188

Elías-Ulloa

phonology considers their syllables as closed. In this sense, they behave as any other coda consonant. See data from Capanahua in (3) and compare it to the stress pattern in (2c) and (2d). (3) a. /kapɨp/ → [ka.ˈpɨ] (Capanahua) ‘Caiman’

b. /kapɨp -n/ → [ˈka.pɨ.ˌpan] (Capanahua) ‘Caiman’ (ERG)

Both languages have in their segmental inventory a glottal stop consonant /ʔ/ (Elias-Ulloa 2009, 2011). In non-prominent prosodic positions (i.e. unstressed syllables), glottal stops completely coalesce with the vowels of those syllables, resulting in laryngealization running throughout the entire duration of the prosodically weak vowel. In contrast, when the vowel belongs to a stressed syllable, glottal stops only coalesce partially with them. In those cases, we observe laryngealization running only through the second half of those prosodically strong vowels. The first half appears in modal voice or with little non-modal phonation. First, I present the case of Capanahua (Loos 1969, Safir 1979, Gonzalez 2003, Elias-Ulloa 2009). In Capanahua, coda glottal stops show a rhythmic pattern reflecting the prosodic structure of the language. They completely coalesce with the preceding vowel in unstressed syllables but only partially in stressed syllables. The data in (4) show this pattern (see Loos 1969:182–3, Elias-Ulloa 2009:174–9).6 I represent complete glottal coalescence by the subscripted diacritic [   ̰ ] underneath the affected vowel. Partial coalescence is represented by the sequence of a vowel and a glottal stop [Vʔ]. (4) a. /tʊʔkʊ -taʔ -ki/ → [ˈtʊʔ.kʊ.ˌtaʔ.ki] (Capanahua) Frog -EV -DECL ‘It is a frog’

b. /tʊʔkʊ -ma -taʔ -ki/ → [ˈtʊʔ.kʊ.ˌma.ta̰ .ki] Frog -NEG -EV -DECL ‘It is not a frog’



c. /baʔkiʃ/ → [ba̰.ˈkiʃ ] (Capanahua) ‘Yesterday, tomorrow’

(Capanahua)

6  Loos (1969) reports the phenomenon as a glottal stop deletion. Elias-Ulloa (2009) presents acoustic evidence for reexamining the phenomenon as glottal stop coalescence.

Prominent Prosodic Positions In Governing Laryngealization

189

Figure 7.4 /tʊʔkʊ/ → [ˈtʊʔ.kʊ] ‘frog’ (Capanahua: partial glottal coalescence).

Figure 7.4 and Figure 7.5 show two examples of the behavior of the glottal stop in Capanahua. In the word [ˈtʊʔ.kʊ] ‘frog’, Figure 7.4, the glottal stop occurs as the coda of a stressed syllable. In this context, it coalesces only with the final edge of the stressed vowel (see arrows). The vowel, given its prosodic status, is associated with a high pitch. In contrast, in Figure 7.5, the glottal stop occurs as the coda of an unstressed syllable. The main stress is on the second syllable. In this environment, the coda glottal stop fully fuses with the unstressed vowel, which surfaces completely in creaky voice. Unlike the previous case, this time, a high pitch appears associated with the second syllable. The important generalization in Capanahua is its lack of tonal or phonation contrasts. That is, the language does not oppose high versus low tones or modal versus creaky voice. High pitch is regularly assigned to stressed syllables and creaky phonation results from the coalescence of its glottal stop consonant with a surrounding vowel. However, in spite of the lack of these contrasts, Capanahua presents a similar pattern as languages like Yalálag Zapotec that do have tonal and phonation contrasts. In Capanahua, syllables closed by a glottal stop, associated with a high pitch, present vowels with an initial phase of modal phonation followed by another phase in laryngealized phonation. Here we cannot invoke the idea that the sequencing of modal and non-modal phonation within a vowel is to facilitate the listener’s recoverability

190

Elías-Ulloa

Figure 7.5 /baʔkiʃ/ → [ba̰ .ˈkiʃ ] ‘yesterday, tomorrow’ (Capanahua: full glottal coalescence).

of phonological contrasts, like tonal oppositions, since Capanahua lacks those contrasts. The glottal stop of Shipibo has a more restricted distribution. In Shipibo, glottal stops can only be found as the initial segment of certain suffixes like /-ʔati/ (VZ1), /-ʔiti/ (VZ2), /-ʔiɖʐa/(INTENS), etc. (Elias-Ulloa 2011).7 In all these suffixes, the glottal stop is followed by a vowel. As a consequence of this restriction, glottal stops in Shipibo only occur as syllable onsets. See data in (5). (5) a. /bɨkʊn -ʔati/ → [bɨ.ˈkʊ̃ n.ʔa.ti] →[bɨ.ˈkʊ̃ .ʔa.ti] (Shipibo) Myopic -VZ1 ‘To steam up’ (Elias-Ulloa 2011)

b. /ikʊn -ʔiɖʐa/ → [i.ˈkʊ̃ n.ʔi.ɖʐa] →[i.ˈkʊ̃ .ʔi.ɖʐa] (Shipibo) true -INTENS ‘Very true’ (Loriot, Lauriault, Day 1993:429)

7  There is, however, some variability in the realization of glottal stops in Shipibo. Some speakers consistently realize them while others speakers do not (see Elias-Ulloa 2011). Those that do not realize the glottal stop treat it as a latent segment. Thus; for those speakers, a word like /bɨkʊn -ʔati/ ‘to steam up’ surfaces as [bɨ.ˈkʊ̃ .a.ti]. This realization does not have creaky voice or any type of laryngealization but we can still argue the glottal stop is latent since the final nasal of the root behaves as a coda and not as the onset of the following vowel.

Prominent Prosodic Positions In Governing Laryngealization

191

FPO

Figure 7.6 /tʃɨː -ʔati/ → [ˈtʃɨː.a̰ .ti] ‘to make (somebody) anemic’ (Shipibo).

The Shipibo glottal stop shows a similar behavior to its Capanahua counterpart. It fully coalesces with an adjacent unstressed vowel. This does not happen if the vowel is stressed and as a consequence carries a high pitch. Figure 7.6 shows an instance of this behavior. The verbalizer suffix /-ʔati/ has been added to the root /tʃɨː/ ‘anemic’. The glottal stop then appears in an intervocalic position with a stressed vowel at its left side and an unstressed one at its right side. The resulting form, [ˈtʃɨː.a̰.ti] ‘to make (somebody) anemic’, has laryngealization running throughout the unstressed vowel [a̰ ], while the stressed vowel [ɨː] and its high pitch are hardly affected by this non-modal phonation. In Figure 7.7, the same verbalizer suffix, /-ʔati/, has been added to the root /kaʃkakaʃka/, onomatopoeia for the sounds one makes when gargling. However, there is an important difference with regard to the case examined above. The glottal stop occurs flanked on its left by an unstressed vowel and on its right by a stressed vowel. This is the opposite of what was illustrated in Figure 7.6. In the resulting form, [ˈkaʃ.ka.ˌkaʃ.ka̰.ˌa.ti], shown in Figure 7.7, the first vowel of the suffix this time emerges in modal voice because it is stressed and bears a high pitch. The preceding unstressed vowel appears completely in creaky voice. Shipibo has another type of glottal stop that it uses as a prosodic marker: an optional epenthetic glottal stop. Shipibo tends to insert it between vowels of different words. In the cases where both vowels are stressed, the glottal stop is realized as creaky voice at the transition between both vowels. Neither of the vowels appears laryngealized since they are stressed and carry a high pitch. This can be observed in Figure 7.8. The words are /ɖʐabβɨk/ ‘two’ and /ani/ ‘big’. They correspond to the two first words of the sentence: /ɖʐabβɨk ani bawaɖʐa

192

Elías-Ulloa

Figure 7.7 /kaʃkakaʃka -ʔati/ → [ˈkaʃ.ka.ˌkaʃ.ka̰ .ˌa.ti] ‘to gargle’ (Shipibo).

Figure 7.8 /ɖʐabβɨk ani/ → [ɖʐa.ˈbβɨ.ʔˈa.ni] ‘two big . . .’ (Shipibo).

jʊbβɨkan bikɨ/ ‘the sorcerer brought two big parrots’ (lit. ‘two big parrots, the sorcerer brought’). The final consonant of /ɖʐabβɨk/ ‘two’ is a latent coda (it does not surface but the final syllable behaves as closed and attracts the stress: [ɖʐa.ˈbβɨ]). Thus, like Capanahua, Shipibo does not have any tonal or phonation contrast. The distribution of high pitch is predictable. It associates with stressed syllables. In turn, as described at the outset of this section, the stress assignment in Shipibo is predictable and sensitive to syllable structure (closed versus open

Prominent Prosodic Positions In Governing Laryngealization

193

syllables) as well as the position of those syllables within the prosodic word. At the phonetic level, Shipibo shows laryngealized vowels but this is the result of a phonological phenomenon of glottal stop coalescence with surrounding vowels, not a lexical specification on vowels. In spite of the lack of tonal or phonation contrasts, Shipibo avoids laryngealized vowels when they carry a high pitch. Shipibo goes a step further. It has a stricter restriction than Capanahua. While Capanahua allows the sequencing of modal and non-modal voice in a stressed vowel associated with a high pitch, Shipibo does not. Shipibo only laryngealizes unstressed vowels. This pattern can be clearly seen when a glottal stop appears trapped between two stressed vowels (see Figure 7.8). In these cases, it materializes as creaky voice only at the transition. As in Capanahua, it cannot be claimed that Shipibo avoids laryngealization of vowels carrying a high pitch in order to facilitate the recoverability of tonal contrasts by the listener since the language lacks those contrasts. 4

Discussion and Conclusions

The main idea proposed by Silverman (1997) in order to understand why ­non-modal phonation avoids overlapping with tonal information within a vowel is that if overlapping occurred, the listener could not easily recover the contrastive tonal information. Thus, for instance, in Yalálag Zapotec, if laryngealization extended over the entire duration of a vowel, it would make it difficult for the listener to recover whether the vowel carries a low, high or falling tone. This difficulty is rooted in the weakening effect non-modal phonation has on different acoustic cues vowels carry and in particular, in the ­fundamental frequency, the physical correlate of tone/pitch. Silverman (1997) thus proposes that the sequence of modal and non-modal phonation is expected to occur in languages that contrast tones and modal voice versus non-modal voice. Shipibo and Capanahua have a high pitch that predictably associates with stressed syllables and vowels with creaky voice that result from the coalescence of a glottal stop segment with a surrounding vowel. Unlike languages like Yalálag Zapotec that possess contrastive non- modal phonation and tones (see Avelino-Becerra in this volume), Shipibo and Capanahua lack those contrasts. One of the goals of this article is to highlight the similar behavior these three languages show with regard to the interaction of non-modal phonation and tone/pitch. The three languages present strategies to avoid the overlap of those properties. Capanahua and Yalálag Zapotec resort to sequencing modal and non-modal phonation so there is at least a section of the affected vowel

194

Elías-Ulloa

that appears in modal voice and from which the listener can recover the cues of the pitch/tone associated with that vowel. Shipibo is stricter. It avoids the occurrence of laryngealization in vowels carrying a high pitch. In this article, I argue that what is key to avoiding the overlap of non-modal phonation and pitch/tone is not the recoverability of phonological contrasts but the avoidance of the weakening of prominent prosodic positions. Following Beckman (1998), prominent prosodic positions are those that enjoy some perceptual advantage via either psycholinguistic or phonetic prominence. They are psycholinguistically prominent in the sense that they are crucial in processes of lexical storage, access and retrieval. They are phonetically prominent in that they provide the listener with strong physical cues that assist the parsing of phonological information, which in turn helps with processes of lexical access and retrieval (see also Kingston 1985, 1990; Steriade 1995, Kirchner 1996, de Lacy 2002, 2006; Zoll 2004, Benua 2003, among others). Prominent prosodic positions include root-initial syllables, word-initial syllables, stressed syllables, syllable onsets, long vowels, and roots. Their counterparts, non-prominent positions, include unstressed syllables, non-initial syllables, affixes, clitics, and function words. Prominent prosodic positions are characterized by maintaining phonological contrasts (that can be neutralized elsewhere), triggering processes that can enhance their prominence and inhibiting those that can weaken it. The interaction between non-modal phonation and pitch/tone in the three languages examined in this article receives a common account once the role of prominent versus non-prominent position is acknowledged. Non-modal phonation is known to debilitate phonetic cues in vowels since it lowers their intensity, weakens their formants, and makes it more difficult to retrieve the pitch/tone associated with them since the fundamental frequency is severely affected (see Gordon 1998; Gordon and Ladefoged 2001). Capanahua and Shipibo avoid the overlap of high pitch and laryngealized voice in prominent prosodic positions (in this case, vowels in stressed syllables). Both languages accept the overlap in unstressed vowels, which are non-prominent positions. Let us examine each case in turn. Capanahua has a general process of glottal coalescence where coda glottal stops fuse with the preceding vowel. In Capanahua, the [constricted glottis] feature that belongs to glottal stops must be realized. It cannot be deleted (see Elias-Ulloa 2009). This causes a conflict when a coda glottal stop occurs adjacent to a stressed vowel since the [constricted glottis] feature must surface but without debilitating the prominent prosodic position. Capanahua solves the conflict resorting to a phonetic sequencing of modal and non-modal phonation within the stressed vowel. Thus, the stressed vowel can keep its lexical specifications recoverable

Prominent Prosodic Positions In Governing Laryngealization

195

within the modal phase but the [constricted glottis] feature of the glottal stop does not fail to be realized. This is a phonetic strategy since phonologically, as I have shown in section 2, the language considers the laryngealization in the second half of the vowel an independent segment, a sequence of a vowel and a glottal stop. In contrast, in unstressed vowels, the conflict does not arise. The language does not have any urgent pressure to inhibit the process of glottal coalescence when it affects a non-prominent position. Shipibo chooses a different strategy compared with Capanahua. Shipibo is not willing to compromise the prominence of stressed syllables, not even by using a phonetic sequencing of modal and non-modal phonation as Capanahua does. For Shipibo, laryngealization can only occur in unstressed vowels. Therefore, the [constricted glottis] feature of glottal stops looks for an adjacent unstressed vowel. If there is no adjacent unstressed vowel as in the case illustrated in Figure 7.8, the glottal stop materializes as a short creaky voice transition at the very edge of the vowels surrounding it and without upsetting much their modal voice. What about Yalálag Zapotec? Following Beckman (1998), one of the functions of prominent prosodic positions is to host phonological contrasts, like the tonal contrasts Yalálag Zapotec possesses. The phonology is licensed to trigger phenomena that enhance prominent positions (cases of fortition) and also to inhibit phenomena or phonological specifications whose occurrence can weaken them. The later situation arises, in Yalálag Zapotec, when a vowel is lexically associated with the [constricted glottis] feature but belongs to a prominent syllable.8 If this feature appears during the entire duration of a vowel, it will weaken other phonological specifications the vowel has—not only its tone, but also information about formant structure. As in the case of Capanahua, Yalálag Zapotec chooses to resolve the conflict by resorting to the sequencing of modal and non-modal phonation. This phonetic strategy allows the phonetic cues of the prominent prosodic position to be enhanced so the listener can retrieve them. 8  At a first glance it might seem strange to talk about prominent prosodic positions for a tonal language like Yalálag Zapotec, but it is not. A prominent prosodic position, like a stressed syllable, is a site within the prosodic structure that can show an asymmetric behavior with regard to non-prominent positions. In languages that lack lexical tones, prominent prosodic positions are usually informally referred as ‘stressed syllables’ (although, as I mentioned before, there are other prominent prosodic positions that are not related to the head of metrical feet). Tonal languages, as any other human language, also have prosodic structure and their prominent positions are able to interact with tones (see Blumenfeld 2004, and also de Lacy 1999, 2002 for an excellent discussion of the interaction of tones and prosodic structure in Ayutla, a Mixtec language).

196

Elías-Ulloa

Before exploring other possible but unsuccessful alternatives to the proposal put forward in this article, I would like to comment on Avelino-Becerra’s analysis of Yalálag Zapotec, Yucatec Maya and Santa María Ocotepec Mixe, also published in this volume. These three languages present a phonological contrast between vowels in modal phonation and laryngealized vowels. With regards to tones, Yalálag Zapotec has a more crowded tonal space than Yucatec Maya. Yalálag Zapotec opposes three tonal units while Yucatec Maya only two. Interestingly, Santa María Ocotepec Mixe is not a tonal language. In spite of only two of the languages examined being tonal, none allows their laryngealized vowels to surface completely creaky. Their underlying laryngealized vowels can surface with an initial phase in modal phonation followed by a phase in creaky voice towards the end of the vowel. Alternatively, they show an initial phase in modal phonation followed by a creakiness period (which can vary in its degree of laryngealization) and then a final period of time towards the end of the vowel that can vary between modal phonation and attenuated creakiness. The laryngealized vowels of Santa María Ocotepec Mixe present a significant difficulty for Avelino-Becerra’s analysis.9 The analysis he proposes depends on the idea that tonal languages that have contrastive phonation want their non-modal phonation vowels to have a phase of modal phonation so they can preserve their underlying tones, and help the listener retrieve those tones. However, Santa María Ocotepec Mixe does not have tones and yet its laryngealized vowels still behave in a way Avelino-Becerra’s analysis predicts should occur in a language with tonal contrasts. In an attempt to solve this problem, Avelino-Becerra’s analysis suggests that somehow Santa María Ocotepec Mixe is (becoming) a tonal language, although phonologically there seems little or no evidence to support such a claim. In constrast, my position is that Santa 9  Avelino-Becerra (p.c.) does not see the case of Santa María Ocotepec Mixe as problematic for his proposal since its pitch could be the phonetic manifestation of the underlying feature: [constricted glottis]. That is, sometimes Santa María Ocotepec Mixe forces its laryngealized phonation to be phonetically implemented as pitch modulations so the underlying feature can be retrieved. I agree with Avelino-Becerra’s position that languages show different strategies for dealing with featural specifications in conflict with some other aspect of phonology/ phonetics: sometimes the specification is suppressed; other times it is changed or relocated. However, the behavior observed in Santa María Ocotepec Mixe does not follow from the analysis proposed in his article, which is built on the idea of recoverability of underlying tonal information under the pressure of also preserving underlying non-modal phonation. Since Santa María Ocotepec Mixe lacks underlying tones, the reason why its [constricted glottis] feature is restricted only to a portion of the vowel hosting it or is realized as pitch modulations remains a mystery.

Prominent Prosodic Positions In Governing Laryngealization

197

María Ocotepec Mixe is not very different from Shipibo and Capanahua. None of the three languages has tonal contrasts to preserve but all strive to avoid having the feature [constricted glottis] spread along the entire vowel when the vowel is in a prominent prosodic position. One of the more interesting results in Avelino-Becerra’s article (this volume) is that in the three languages examined, independently of whether they are tonal or not, there is a tendency for having in modal phonation a significant part of the initial phase of a laryngealized vowel. Sometimes, to different degrees, this occurs in the final section of the vowel, too. The initial and final phases correspond to vowel transition zones—areas that are crucial in order to retrieve information not only about the place of articulation of an adjacent consonant but also formant information about the height and backness of the vowel itself and also its fundamental frequency. From my point of view, Avelino-Becerra’s findings actually point to a tendency at the phonetic level for languages to ensure that the listener retrieves most phonetic cues at the very beginning of a vowel. In particular, there is a pressure for this to happen in the case of vowels that occupy a prominent prosodic position. Some of those cues might correlate with underlying features. Others do not. In the case of laryngealization, Avelino-Becerra’s findings seem to show a tendency for languages to avoid having laryngealization on the transition zones between consonants and vowels, areas that occur within vowels and that are crucial for retrieving the different phonetic cues of adjacent segments and the vowel itself. Thus, in languages that choose this strategy, laryngealization is preferred towards the middle of the vowel or, as we will see next, towards the second half of the vowel. There also appears to be a strong tendency to have the transition between an onset and its vowel within the syllable free of non-modal phonation instead of between a vowel and a coda or the onset of another syllable. This would explain why, in Avelino-Becerra’s study, laryngealization mostly concentrates on the central portion of the vowel or in general, towards the second half of the vowel but not towards the initial part. Here I would like to be clear about the two different issues involved when comparing Avelino-Becerra’s analysis and the phenomena of laryngealization I discuss in this article. The account I develop here is to explain a phonological pattern; that is, why laryngealization is avoided in certain syllables but allowed in others in languages like Shipibo and Capanahua. The solution I propose is that those syllables, where laryngealization is inhibited, are syllables that occupy prominent prosodic positions. My proposal, by itself, has nothing to say about the different degrees of phonetic implementation of laryngealization and the position of laryngealization within the vowels that host that

198

Elías-Ulloa

f­ eature. Avelino-Becerra’s article tackles those phonetic issues. However, I differ from Avelino-Becerra’s proposal in one important matter: the idea that the sequencing of modal and non-modal phonation within a vowel is somehow related to the presence of contrastive tones. Shipibo, Capanahua and Santa María Ocotepec Mixe show that cannot be the case. They do not have con­ trastive tones. Instead, I suggest, that in languages like Yalálag Zapotec, in which the laryngeal and tonal features must be preserved, syllables that bear tones (i.e. prominent prosodic positions) inhibit the occurrence of complete laryngealization within the vowel. The particular location of non-modal phonation towards the middle or towards the second half of the vowel is an issue of phonetic implementation related to the retrieving of phonetic cues of onset consonants and of the vowel. This phenomenon becomes more powerful when the vowel occurs in a prominent prosodic position. Thus, in all the languages Avelino-Becerra and I examine, phonology can demand the preservation of underlying features and enforce the asymmetry between prominent and non-prominent prosodic positions (e.g. by inhibiting processes that could weaken them, hosting tones or attracting high pitch to certain syllables) while at the sub-phonemic level, phonetics organizes in a biased way the sequencing of information (e.g. phonetics cues should be localized during the first half of a vowel, the [constricted glottis] feature is realized later so that phonetic cues can be easily retrieved by the listener). In the following paragraphs, I discuss why this proposal cannot be only about making phonetic cues retrievable but it should refer to the asymmetry between prominent and non-prominent positions, too. Let us examine two alternative analyses to this account based on prominent positions and see why they are flawed. First, one could argue phonological contrasts, and not prominent positions, can be viewed as the determining factor behind why languages like Capanahua, Shipibo and Yalálag Zapotec avoid overlapping non-modal phonation and pitch/tone. In this view, Yalálag Zapotec resorts to modal/non-modal phonation sequencing to facilitate the maintenance and recoverability of tonal contrasts by the listener (as opposed to avoiding the weakening of prominent prosodic positions). Under this alternative, in Capanahua, the strategy of sequencing modal and non-modal voice is not to help the listener retrieve tonal contrasts since the language lacks them. The sequencing would occur to help the listener retrieve other phonological contrasts (e.g. vowel quality). The problem with this analysis is that it will have to invoke the concept of prominent prosodic positions in some way. The idea that ‘what motivates the phonation sequencing in Capanahua is the retrieval of vowel-quality contrasts’ only works for stressed vowels, not for their unstressed counterparts.

Prominent Prosodic Positions In Governing Laryngealization

199

Unstressed vowels in Capanahua also contrast in vowel quality (i.e. /i/ vs. /a/ vs. /ɨ/ vs. /ʊ/). Why, then, does Capanahua not sequence modal and laryngealized voice in unstressed vowels? One would have to argue that it is because somehow the maintenance of vowel quality in stressed vowels is more important. But this addition already introduces the role that prominent prosodic positions play and that alone can explain the pattern observed in the three languages. A second alternative, also based on phonological contrasts, might argue that in languages like Shipibo and Capanahua, the surface contrast between stressed and unstressed syllables is made recoverable by allowing a better perception of the phonetic cue that signals stress, which is high pitch (see also the discussion in Wetzels and Meira 2010).10 Let us evaluate this possibility by examining the case of Shipibo, for instance, and see how far this hypothesis stretches. As already mentioned, syllables with main stress are cued by the obligatory presence of a high pitch and a longer duration in vowels. However, syllables with secondary stress present a slightly different behavior with regard to high pitch. Closed syllables with secondary stress have the ability to attract a high pitch but this is not obligatory. Open syllables with secondary stress do not usually carry a high pitch. During elicitation or in careful speech, syllables with secondary stress do tend to occur with a high pitch (Elias-Ulloa 2011). The point here is that even in cases where a syllable with secondary stress does not receive a high pitch, we still observe the behavior described above for non-modal phonation: partial coalescence of the glottal segment with a stressed vowel but full coalescence when the vowel is metrically unstressed. This discards the possibility of partial glottal coalescence being a strategy triggered to make recoverable a high pitch so the contrast between stressed and unstressed syllables is maintained. In contrast, an analysis based on prominent prosodic positions, proposed in this study, can straightforwardly account for why modal and non-modal voice sequencing occurs in stressed syllables but not in unstressed ones in Capanahua, for why laryngealized voice is restricted to unstressed vowels in Shipibo and also for why phonologically laryngealized vowels hosting tones in Yalálag Zapotec show a modal phase.11 Prominent prosodic positions avoid 10  Thanks to an anonymous reviewer for pointing out this alternative. 11  An anonymous reviewer rightly asks me to bear in mind that the two contrastive arguments, the prominent-position account vs the phonological-contrast account, presented in this study are not mutually exclusive. The reviewer says “It could be that prominent prosodic positions indeed play an important role in the avoidance of overlapping nonmodal phonation and pitch information, but that does not rule out that languages which

200

Elías-Ulloa

being weakened. They inhibit processes that can debilitate them (Capanahua and Shipibo do not allow glottal stops to coalesce with stressed vowels). They can trigger processes that enhance their prominence (Yalálag Zapotec sequences the realization of the [constricted glottis] feature associated with some of its vowels so other phonological information contained in the vowel can be better retrieved by the listener). Prominent prosodic positions create an asymmetry attested in many languages between them and their counterparts. Under this view, non-prominent positions are expected to undergo a number of weakening processes. An approach that relies on the retrieval of phonological contrasts by the listener, as the key factor for understanding the sequencing of modal and non-modal phonation, cannot account for why unstressed vowels undergo full laryngealization when they coalesce with a glottal stop in languages like Shipibo or Capanahua. After all, laryngealization will make the listener’s task of retrieving the vowel-quality contrasts of unstressed vowels more difficult. From the perspective of that approach, it is also surprising to find languages that lack phonation and tonal contrasts, like Capanahua, but that sequence modal and laryngealized phonation within their vowels. References Avelino-Becerra, H. 2004. Topics in Yalalag Zapotec, with particular reference to its phonetic structures. Department of Linguistics. Los Angeles, University of California. PhD. dissertation. Beckman, J. 1998. Positional faithfulness, University of Massachusetts, Amherst. PhD. dissertation. Benua, L. 2003. Transderivational Identity: Phonological relations between words (excerpt of Benua 1997/2000). Optimality Theory in phonology: a reader. J. McCarthy. Malden, MA and Oxford, UK, Blackwell: 419–437. have both contrastive [. . .] maximize this strategy to facilitate retrieval of contrastive tonal information by the listener”. This is certainly a logical possibility. However, since an account based on prominent prosodic positions can alone explain the distribution of partial versus total laringealization observed in the vowels of Panoan languages like Shipibo and Capanahua as well as in the vowels of Otomanguean languages like Yalálag Zapotec; meanwhile, an account based on phonological contrasts can only account for cases where tones and non-modal phonation are phonologically underlying, I take the position that an account based on phonological contrasts needs to show that it is independently needed, in a phenomenon where prominent prosodic positions cannot account for the data observed.

Prominent Prosodic Positions In Governing Laryngealization

201

Blankenship, B. 2002. “The timing of nonmodal phonetion in vowels.” Journal of Phonetics 30:163–91. Blumenfeld, Lev. 2004. “Tone-to-Stress and Stress-to-Tone: Ancient Greek Accent Revisited”. In Proceedings of the Annual Meeting of the Berkeley Linguistics Society, Volume 30(1). Borroff, M. 2007. A Landmark Underspecification Account of the Patterning of Glottal Stop. Stony Brook University. PhD. dissertation. Bortolin, L. and S. McLennan. 1995. Blackfoot. Retrieved from: http://www.umt.edu/ ling/BLG/ BlackfootDictionary/materials.html. de Lacy, P. 1999. Tone and prominence. Ms, University of Massachusetts, Amherst. Available as ROA-333 from the Rutgers Optimality Archive. ———. 2002. “The Interaction of tone and stress in Optimality Theory”. Phonology 19:1–32. ———. 2002. The Formal Expression of Markedness. University of Massachusetts, Amherst. PhD. dissertation. ———. 2006. Markedness: reduction and preservation in phonology. Cambridge; New York, Cambridge University Press. Elias-Ulloa, J. 2006. Theoretical Aspects of Panoan Metrical Phonology: Disyllabic Footing and Contextual Syllable Weight. New Brunswick, Rutgers University. PhD. dissertation. ———. 2009. “The Distribution of Laryngeal Segments in Capanahua.” International Journal of American Linguistics vol. 75:159–206. ———. 2011. Una documentación acústica de la lengua shipibo-conibo. (Con un bosquejo fonológico) [An Acoustic Documentation of the Shipibo-Conibo language (with a phonological outline)]. Editorial de la Pontificia Universidad Católica del Perú. Gonzalez, C. 2003. The Effect of Stress and Foot Structure on Consonantal Processes, University of Southern California. PhD. dissertation. Gordon, M. 1998. The phonetics and phonology of non-modal vowels: a cross-linguistic perspective. Berkeley Linguistics Society. Gordon, M. and P. Ladefoged. 2001. “Phonation Types: A Cross-Linguistic Overview.” Journal of Phonetics 29(4): 383–406. Kingston, J. 1985. The Phonetics and Phonology of the Timing of Oral and Glottal Events, University of California, Berkeley. ———. 1990. Articulatory binding. Papers in Laboratory Phonology I.J. Kingston and M.E. Beckman. Cambridge, UK, Cambridge University Press: 406–434. Kirchner, R. 1996. Cues or contexts in feature licensing constraints. Manuscript. ROA 162. Ladefoged, P. and I. Maddieson. 1996. The sounds of the world’s languages. Oxford, OX, UK; Cambridge, Mass., USA, Blackwell Publishers.

202

Elías-Ulloa

Loos, E.E. 1969. The phonology of Capanahua and its grammatical basis. Oklahoma, SIL and University of Oklahoma. Loos, E.E. and B. Loos. 1998. Diccionario Capanahua-Castellano. Yarinacocha, Pucallpa, Perú, Instituto Lingüístico de Verano. Loriot, J., E. Lauriault, and D. Day. 1993. Diccionario shipibo-castellano. Yarinacocha, Perú, Ministerio de Educación: Instituto Lingüístico de Verano. Peterson, T. 2004. Theoretical issues in the representation of the glottal stop in Blackfoot. Seventh Workshop on American Indigenous Languages, University of California, Santa Barbara. Picanço, G. Lobato. 2005. Mundurukú: Phonetics, Phonology, Synchrony, Diachrony, University of British Columbia. PhD. dissertation. Safir, K.J. 1979. Metrical Structure in Capanahua. Papers on Syllable Structure, Metrical Structure, and Harmony Processes. K. Safir. Cambridge, Mass., MIT Dept. of Linguistics: 95–114. Silverman, D. 1997. “Laryngeal complexity in Otomanguean vowels.” Phonology 14(2): 235–262. Steriade, D. 1995. Positional neutralization. North East Linguistic Society 24, University of Massachusetts, Amherst. Wetzels, L. and S. Meira. 2010. ‘A Survey of South American Stress Systems.’ In Harry van der Hulst, Rob Goedemans and Ellen van Zanten (Eds.) A Survey of Word Accentual Patterns in the Languages of the World (XLW1) Berlin, Walter de Gruyter: 313–381. Zoll, C. 2004. Positional Markedness, Positional Faithfulness and Licensing. Optimality Theory in Phonology: A Reader. J. McCarthy. Oxford, Blackwell Publishers: 365–378.

CHAPTER 8

Pitch and Glottalization as Cues to Contrast in Yucatec Maya Melissa Frazier 1 Introduction Yucatec Maya (a Mayan language of Mexico, henceforth abbreviated as YM) is one of the few Mayan languages to use a tonal contrast. Long vowels are produced with either low tone (e.g. /mìis/ cat) or high tone (e.g. /míis/ sweep). There is also a third type of long vowel used in YM; GLOTTALIZED vowels are produced with initial high pitch and with creaky voice. The results of production experiments show that HIGH TONE and GLOTTALIZED vowels differ in several phonetic dimensions: GLOTTALIZED vowels tend to have higher initial pitch, to have a larger pitch span, to be produced with more creaky voice, and to be longer than HIGH TONE vowels. The fact that multiple phonetic dimensions are at play in this phonological contrast has implications for what strategies listeners use in differentiating the contrast as well as for how to model the phonological grammar and its relation to phonetics. These implications are explored in this chapter via two perception experiments conducted in Yucatan, Mexico. It has been demonstrated for various phonemic contrasts that the same phonetic properties that distinguish different phonemes in production are used by the listener as cues to perceiving the contrast. For example, both F1 and duration systematically differ in the productions of tense and lax vowels in English (Peterson and Barney 1952; Peterson and Lehiste 1960), though the reliability of such cues varies by dialect. Escudero and Boersma (2004) show that speakers of different dialect groups attend to those cues that are * I would like to thank the participants of the experiments reported on here; Grant McGuire, David Mora-Marin, Elliott Moreton, and Jennifer L. Smith for providing discussion and comments on various aspects of this chapter; and Christopher Wiesen for providing statistical consulting. This data previously appeared in my doctoral dissertation (Frazier 2009). This work was supported by the Luis Quirós Varela Graduate Student Travel Fund (Institute for the Study of the Americas at the University of North Carolina, Chapel Hill) and the Jacobs Research Fund (Whatcom Museum, Bellingham, WA).

© koninklijke brill nv, leiden, ���6 | doi ��.��63/9789004303218_009

204

Frazier

the most c­ orrelated with the contrast in that dialect: Scottish English speakers distinguish /i/ from /ɪ/ primarily by F1 and only minimally by duration, whereas duration is predominantly used by Southern British English speakers. As another example, the average VOT (for both voiced and voiceless stops) is greater in English than in Spanish, and when performing a discrimination task, the crossover point (the point on the VOT continuum that begins to elicit more ‘voiceless’ responses than ‘voiced’ responses) has a higher VOT value for English listeners than for Spanish listeners (Liberman et al. 1958; Lisker and Abramson 1964, 1970; Cho and Ladefoged 1999). Given the above facts, one would expect YM listeners to make use of all available phonetic dimensions—pitch, glottalization, and vowel duration— when distinguishing between HIGH TONE and GLOTTALIZED vowels. This expected result is confirmed when listeners respond to natural stimuli produced by native speakers. This result is also in line with experimentation that shows that perception strategies are dependent on the specific phonetic productions of a language and thus must be accounted for with language-specific perception grammars, and particularly with bidirectional grammars, in which the production and perception grammars mirror each other. Listeners reacted differently to manipulated stimuli by attending to glottalization alone and ignoring pitch. I argue that the manipulated stimuli forced participants to focus on only one phonetic dimension, and so listeners made use of the phonetic cue that is the most reliable at distinguishing the contrast between HIGH TONE and GLOTTALIZED vowels. This means that the grammar of a language must not only account for how listeners use multiple cues in ‘ideal’ language situations but also how they focus on the most useful cue in ‘degraded’ language situations. This chapter proceeds as follows. In §2, I present the phonological properties of the vowel system of Yucatec Maya and the results of production experiments that show how different phonetic properties indicate a contrast between HIGH TONE and GLOTTALIZED vowels. Two perception experiments are discussed in §§3–4.1 The first perception experiment used natural stimuli to test how accurate listeners are at discriminating between HIGH TONE and GLOTTALIZED vowels and to determine which phonetic dimensions listeners use to make their choice. The second perception experiment used m ­ anipulated 1  The methodology and results of all production and perception experiments are presented in my dissertation (Frazier 2009); the results of production experiment 1 are also reported in Frazier (2011). In this chapter, I focus on the data most relevant to the discussion at hand, and, as such, I present data pooled from different speakers and different tokens than those reported in previous work.

Pitch And Glottalization As Cues To Contrast In Yucatec Maya

205

stimuli to test how listeners are influenced by the interaction of pitch and glottalization. Conclusions are presented in §5. 2

Production of Vowel Contrasts in Yucatec Maya

Yucatec Maya is spoken by some 700,000 people in Yucatan, Campeche, and Quintana Roo, Mexico and Belize (Gordon 2005). In addition to five contrasting vowel qualities ([i e a o u]), there is a four-way contrast of suprasegmental features, henceforth referred to as VOWEL SHAPE. The four vowel shapes, described in (1) along with a minimal quadruplet, can appear with any of the vowel qualities. Each vowel in YM must occur with one of these shapes. Thus, SHORT vowels are the only vowels not marked for tone; all long vowels bear tone. Throughout this chapter, vowel shapes are denoted by small capital letters so that they are not confused with generic phonological/phonetic properties (i.e. ‘GLOTTALIZED’ is a phonemic category of YM, whereas ‘glottalization’ is a phonetic property).2 (1) vowel shapes of YM (Bricker et al. 1998)3 SHORT /v/ short duration, mid pitch, modal voice LOW TONE /v̀v/ long duration, low pitch, modal voice HIGH TONE /v́v/ long duration, initial high pitch, modal voice GLOTTALIZED /v́ v̰ / long duration, initial high pitch, creaky voice (during the medial or final portion of the vowel)

chak ‘red’ chaak ‘boil’ cháak ‘rain’ cha’ak ‘starch’

2  Throughout this chapter, ‘glottalization’ is used to denote phonation type (whether modal or creaky) in order to remain consistent with the literature on YM. 3  Bricker et al. (1998), along with other major phonological descriptions of YM (e.g. Blair and Vermont Salas 1965), represent the GLOTTALIZED vowel as /vʔv/ or /v́ʔv/. These vowels are hence often called ‘rearticulated vowels’ in the literature. The representation /v́v̰/ is supported by phonetic research which shows that this vowel shape is canonically produced with creaky voice (and not a full glottal stop) and with initial high pitch (Frazier 2009, 2011; Avelino et al. 2007; see §2.2.2). Because these vowels are more often creaky than rearticulated, I follow their other naming convention and refer to them as GLOTTALIZED vowels here.

206

Frazier

Two production experiments were designed to document the realization of the suprasegmental properties of the vowel system of YM. We will see in this section that the phonetic dimensions of vowel length, pitch, and glottalization all contribute to this contrast, and so the perception experiments of §§3–4 are designed to measure the effect of these properties on listeners’ responses in discrimination tasks. 2.1 Methodology For both production experiments, speakers were recorded from multiple towns in Yucatan, Mexico. The results indicate a dialectal split in the production of pitch and vowel length. Because these dialect differences are not the focus of this chapter, I only present data from speakers from Mérida (the capital of Yucatan) and Santa Elena (approximately 65 km south of Mérida).4 These towns are both located in the western half of Yucatan, and these speakers produce the vowel shapes in ways that are most congruent with previous literature on YM vowels. The two production experiments mainly differ in the context of the target word. In production experiment 1 the target words were produced in isolation; in production experiment 2 the target words were produced in frame sentences (and data reported on here comes from target words in phrase-final position). Nineteen speakers (1 female and 6 males from Mérida; 7 females and 5 males from Santa Elena) who participated in production experiment 1 are reported on here. These participants read 100 words (25 with each vowel shape) in isolation. Not all the data collected in production experiment 1 is reported on here. Only words with HIGH TONE or GLOTTALIZED vowels are of interest. The wordlist contained both nonce forms and existing forms; the results from nonce forms are not included below. Furthermore, the wordlist for speakers from Santa Elena contained some polysyllabic words, where measurements were taken from a vowel in a non-final syllable. Because there is variation in the production of suprasegmental features that is conditioned by the position of the syllable within the word, these polysyllabic words are also excluded here. Thus, the results presented below represent the production of 20 words with a HIGH TONE vowel and 19 words with a GLOTTALIZED vowel by each speaker from Mérida and 15 words with a HIGH TONE vowel and 16 words with a GLOTTALIZED vowel by each speaker from Santa Elena. All target words are of the form CVC. 4  The other towns represented in the production experiments were Sisbicchén, Xocén, and Yax Che, which are all clustered around Valladolid, the largest city in the eastern part of Yucatan. The reader is referred to Frazier (2009) for more information about these dialect differences.

Pitch And Glottalization As Cues To Contrast In Yucatec Maya

207

Seventeen participants are reported on here for production experiment 2 (3 males from Mérida; 10 females and 4 males from Santa Elena).5 These participants read a total of 144 sentences (36 target words, 9 with each vowel shape, of the form CVC embedded in four different frame sentences each). The target word was positioned phase-initially, phrase-medially, phrase-finally (after a HIGH TONE vowel), and phrase-finally (not after a HIGH TONE vowel). Only the last of these contexts is reported on here, and the relevant frame sentence is Tu ya’alaj __. ‘S/he said __.’6 Again, only HIGH TONE and GLOTTALIZED vowels are relevant, and so the following data for production experiment 2 comes from 9 HIGH TONE and 9 GLOTTALIZED vowels as spoken by each participant. Measurements were taken from target words only (as spoken in isolation in production experiment 1 and in phrase-final position in production experiment 2).7 All measurements were extracted using PRAAT (Boersma and Weenink 2006). For each target word, the boundaries of the vowel were demarcated (as determined by the onset and offset of F2) so that vowel length could be calculated, pitch values in Hz were extracted at 10 ms intervals for the duration of each vowel, and each vowel was coded for glottalization type. 2.1.1 Coding Glottalization Type Gordon and Ladefoged (2001) summarize a body of literature that documents the acoustic and aerodynamic differences among modal, creaky, and breathy voice and conclude that, while the reliability of these properties varies from language to language, seven main characteristics can be used to differentiate these phonation types: periodicity, intensity, spectral tilt, fundamental frequency, formant frequencies, duration, and airflow. Though all of these properties are quantifiable, there is no way to determine a cut off point that

5  An additional speaker from Santa Elena was recorded, but his data was rejected because he did not produce the requested sentences. 6  A more detailed gloss for this frame sentence is as follows:  T-uy a’al- aj __.  COMP.ASP-3ERG say- COMP.ASP/TRANS.STATUS __. (The ⟨y⟩ is considered part of the morpheme ⟨u⟩, and is used before vowel-initial verbs. In standard orthography, the letter ⟨y⟩ is placed with the verb, as in ‘Tu ya’alaj ’). 7  Both the production and perception experiments reported on in this chapter compare target words in two different contexts: isolation and phrase-final. These two contexts are analogous and thus comparable in that the literature on prosody treats words spoken in isolation as being phrase-final. Additionally, the full results of these experiments as presented in Frazier 2009 show that acoustic properties of YM vowel shape are similar in words spoken in isolation and in phrase-final position.

208

Frazier

divides the values of a given property into those that equate with each of the phonation types. For example, spectral tilt (e.g. the amplitude of the second harmonic minus the amplitude of f0) tends to be negative for creaky voice, but this does not mean that one can simply measure spectral tilt and categorize those tokens with negative spectral tilt as being produced with creaky voice and those with positive spectral tilt as being produced with modal voice.8 Similar problems arise for any of the aforementioned phonetic properties. To determine the glottalization type of each token obtained in the production experiments, I observed the waveform and spectrogram for periodicity, intensity, and f0 (similar to the methods of Redi and Shattuck-Hufnagel 2001). I found that, in YM, a departure from modal voice was most consistently indicated by a weakening of intensity, as seen in Fig. 8.1. In this token, it is clear that the vowel is not produced with sustained modal voice, but the only visual indicator of a change in phonation type is lowered intensity and a slight lowering of f0. Such tokens were classified as being produced with creaky voice. Irregularly spaced glottal pulses (i.e. aperiodicity) and irregularity in the intensity of consecutive glottal pulses (possibly due to diplophonia) were less consistently present in productions of creaky voice in YM. See Fig. 8.2 for a token with creaky voice that exhibits all of these properties. Thus, the visual cues of lowered intensity and aperiodicity were used to indicate a token that was produced with creaky voice during some portion of vowel production.9 While lowered f0 was commonly present, it was never the sole indicator of a departure from modal voice. A small percentage of GLOTTALIZED vowels were produced with a full glottal stop interrupting vowel production. Any vowel that was interrupted by a gap between glottal pulses of 20 ms or more was coded as having a glottal stop. Additionally, I discovered that many tokens in YM did not display all of the canonical properties of creaky voice but rather showed only a momentary but extreme drop in intensity that indicated a departure from modal voice. This pattern occurred often and such tokens gave an auditory impression of creaky voice. Because these tokens were clearly not produced with only modal voice .

8  It has also been documented that there can be an interaction between tone and spectral tilt (Blankenship 1997), which could be especially problematic for YM, where tone and creaky voice are used phonemically. 9  Creaky voice most commonly occurred during the medial portion of the vowel, such that the vowel began and ended with modal voice. A smaller set of vowels were produced with creaky voice that began in the middle of the vowel and continued to the end. Only a few tokens showed creaky voice either initially or throughout vowel production.

Pitch And Glottalization As Cues To Contrast In Yucatec Maya

ʃ 0

ḭ Time (s)

i

m 0.4698

Frequency (Hz)

5000

i

209

0

0

Time (s)

0.4698

FIGURE 8.1 Creaky voice indicated by low intensity. This token of xi’im /ʃ íḭm/ ‘corn’ is produced by a male from Mérida.

but were also clearly not produced with a lengthy stretch of creaky voice, they are given the label of ‘weak glottalization’. Four glottalization types are thus referred to in this chapter: modal voice (no glottalization at any point in vowel production), weak glottalization (indicated by a brief dip in intensity), creaky voice (at some point during vowel production), and a glottal stop (interrupting vowel production). These glottalization types are henceforth abbreviated as mod., w.g., cr., and g.s. (respectively), and examples are presented in Fig. 8.2. 2.1.2 Measuring Pitch As explained above, pitch measurements (in Hz) were extracted at 10 ms intervals for the duration of the vowel using PRAAT. Seven pitch values are referred to in this chapter: the maximum pitch value produced during vowel production, the minimum pitch value produced during vowel production, and five pitch values produced at normalized time points (beginning of vowel, 25%, 50%, and 75% of vowel production, and end of vowel). These five pitch values are used to define pitch contours for each vowel, and the maximum pitch value minus the minimum pitch value defines the pitch span of the vowel. Maximum

210

Frazier

m

ts’

a

0

u ṵ

0

0.3588

Time (s)

u

k 0.5586

Time (s)

Frequency (Hz)

5000

Frequency (Hz)

5000

p’

0

0

Time (s)

t 0

ḭ Time (s)

i

0

0

Ɂ

Ɂ 0.323

0 5000

0.5586

Time (s)

e

Ɂ

e Time (s)

s 0.511

Frequency (Hz)

i

Frequency (Hz)

5000

0.3588

0

0

Time (s)

0.323

0

0

Time (s)

0.511

FIGURE 8.2 Examples of four types of glottalization. Top row: modal voice, ma’ats’ /máa̰ ts’/ ‘hull corn’; weak glottalization, p’u’uk /p’úṵk/ ‘cheek’ Bottom row: creaky voice, ti’i’ /tíḭʔ/ ‘there’; full glottal stop, e’es /ʔéḛs/ ‘show’

Pitch And Glottalization As Cues To Contrast In Yucatec Maya

211

and minimum pitch values were obtainable for every token, but pitch values at the five normalized time points were not always available (due to aperiodicity generally caused by creak or full glottal closure).10 These missing values were recorded as such, but did not result in throwing out the obtainable pitch values for a given vowel. In order to make meaningful cross-speaker comparisons of pitch values, pitch measurements in Hertz are transformed into SEMITONES OVER THE BASELINE (s/b). This transform is based on the work of Nolan (2003), who found that pitch spans (of intonational contours) showed the least interspeaker variation when measured in semitones, and the work of Pierrehumbert (1980), who showed that a speaker-specific baseline could be used to scale pitch measurements and, again, minimize inter-speaker variation. As shown in (2), s/b is calculated by using the standard equation for deriving semitones from Hz—12*log2(Hz/reference Hz)—but instead of using a constant value for the ‘reference Hz’, a speaker- and context-specific baseline is used. The baseline is (somewhat arbitrarily) defined as the average pitch value produced at the mid point of LOW TONE vowels in a given context (i.e. isolation or phrase-final) by a given speaker. Thus, for example, a pitch measurement of 2 s/b indicates a value that is 2 semitones above that particular speaker’s baseline (in that particular context). As shown in Frazier (2009), this transform is successful at minimizing inter-speaker variation in YM. (2) equation used to transform Hz into semitones over the baseline s/b = 12*log2(Hz/baseline Hz) where baseline Hz is the average pitch value produced at the mid point of LOW TONE vowels for a given speaker in a given context 2.2 Results 2.2.1 Vowel Length GLOTTALIZED vowels are slightly longer than HIGH TONE vowels, though this difference is only statistically significant11 for production experiment 1, as shown in Table 8.1.

10  Vocal fold vibration during the production of creaky voice was not always aperiodic, and so there are many pitch measurements that come from portions of vowels where creaky voice occurs. See §2.2.4 for further discussion of pitch and creaky voice. 11  All t statistics in this chapter are calculated using a mixed linear regression model to account for multiple observations within subjects.

212 TABLE 8.1

Frazier Mean vowel length (ms) for high tone and glottalized vowels

exp. 1 (isolation) exp. 2 (phrase-final)

high tone

glottalized

t

p

df

200 186

208 188

2.25 0.31

.02 .76

625 288

2.2.2 Glottalization For GLOTTALIZED vowels, modal voice and creaky voice are the two most common types of glottalization, while HIGH TONE vowels are (unsurprisingly) almost always produced with modal voice (see Table 8.2). While it is clear that these two vowel shapes differ in terms of glottalization, the fact that so many GLOTTALIZED vowels are produced with modal voice suggests that glottalization alone does not signal a contrast. TABLE 8.2 Distribution of glottalization types

glottalized high tone

mod.

exp. 1 (isolation) w.g. cr. g.s.

exp. 2 (phrase-final) mod. w.g. cr. g.s.

35% 94%

13% 44% 8%  2%  4% 0%

59% 96%

9% 1%

29%  3%

3% 0%

2.2.3 Pitch The average pitch contours of HIGH TONE and GLOTTALIZED vowels as spoken in isolation are shown in Fig. 8.3. GLOTTALIZED vowels start with higher pitch, which then drops more rapidly, than HIGH TONE vowels. Both vowel shapes end with relatively low pitch. GLOTTALIZED vowels have a larger pitch span than HIGH TONE vowels (x̄G LOT = 6.1 semitones; x̄H I = 4.2 semitones; t(624) = 7.82, p < .01). There are some gender-based differences in the production of pitch, as shown in Fig. 8.4. Namely, GLOTTALIZED vowels are produced with higher pitch (relative to the speaker’s baseline) by males than by females. This means

Pitch And Glottalization As Cues To Contrast In Yucatec Maya

213

FIGURE 8.3 Average pitch contours (production experiment 1; isolation). g = glottalized; h = high tone; asterisks indicate time points with statistically significant differences in pitch values for each vowel shape (** (p < .01); * (p < .05))

FIGURE 8.4 Average pitch contours by gender (production experiment 1; isolation). g = glottalized; h = high tone

that the initial pitch of GLOTTALIZED vowels is significantly higher than the initial pitch of HIGH TONE vowels for males (t(355) = 5.40, p < .01) but not for females (t = 0.71(288), p = .48). This result is discussed in more detail below. Even though productions by the two genders differ in terms of initial pitch, GLOTTALIZED vowels have a larger pitch span for both genders (x̄G LOT,male = 5.9 semitones; x̄H I,male = 4.8 semitones; x̄GLOT,female = 6.5 semitones; x̄ HI,female = 3.4 semitones). The pitch contours produced in production study 2, as shown in Fig. 8.5, show that both vowel shapes begin and end with the same pitch values in this context, though the pitch of GLOTTALIZED vowels drops earlier in vowel production. Statistical analysis shows that differences between the two contours are mostly nonsignificant.

214

Frazier

FIGURE 8.5 Average pitch contours (production experiment 2; phrase-final). g = glottalized; h = high tone; asterisks indicate time points with statistically significant differences in pitch values for each vowel shape (** (p < .01); * (p < .05))

Again, the pitch contours of GLOTTALIZED vowels are produced differently by the two genders (see Fig. 8.6). GLOTTALIZED vowels do have higher initial pitch than HIGH TONE vowels in phrase-final context when spoken by males (t(118) = 2.61, p = .01), whereas HIGH TONE vowels have higher pitch throughout vowel production (until the end of the vowel) for females (for initial pitch, t(162) = –2.36, p = .02).

FIGURE 8.6 Average pitch contours by gender (production experiment 2; phrase-final). g = glottalized; h = high tone

Pitch And Glottalization As Cues To Contrast In Yucatec Maya

215

Pitch spans are larger for GLOTTALIZED vowels, whether the data is averaged across all participants (x̄ GLOT = 6.1 semitones; x̄H I = 4.3 semitones; t(616) = 3.18, p < .01) or across each gender separately (x̄ GLOT,male = 6.5 semitones; x̄H I,male = 4.7 semitones; x̄G LOT,female = 5.9 semitones; x̄ HI,female = 4.1 semitones). 2.2.4 The Interaction of Pitch, Glottalization, and Gender The results presented in the previous section indicate that GLOTTALIZED vowels do not have the same pitch contours for both males and females, even when using the semitones over the baseline transform, which is designed to minimize cross-speaker variation. This result can be better understood by looking at the pitch contours that are produced with different glottalization types. In Fig. 8.7 we see the average pitch contours for GLOTTALIZED vowels (as produced in production experiment 1) by gender and glottalization type (modal voice, weak glottalization, and creaky voice).12 When GLOTTALIZED vowels are produced without glottalization (modal voice only), they have roughly the same contour for males and females and are produced with higher pitch than that of HIGH TONE vowels. However, the pitch contours of those vowels produced with either weak glottalization or creaky voice are strikingly different between the two genders. For males, creaky voice and weak glottalization are correlated with even higher initial pitch, while, for females, glottalization is correlated with lower initial pitch and dramatic decreases in pitch during the middle portion of the vowel (where glottalization is canonically produced). The cause of this distinction is the fact that females and males are producing creaky voice with similar f0 values (in Hz).13 For example, the average pitch value produced at the middle time point of GLOTTALIZED vowels produced with either weak glottalization or creaky voice is 147 Hz for males and 167 Hz for females, as compared to the middle time point of GLOTTALIZED vowels produced with modal voice, which is 156 Hz for males and 212 Hz for females. This indicates that pitch produced during creaky voice is not a function of a speaker’s natural pitch range in the same way that pitch produced during modal voice is. Pitch produced during creaky voice is far below the baseline for females but near, or even slightly above, the baseline for males. Thus, when 12  Those vowels produced with a full glottal stop are excluded here because of how few tokens had this glottalization type and because no pitch values can be extracted from the portion of the vowel with a glottal stop. 13  A similar result was found for speakers of English. Blomgren et al. (1998) measured f0 during productions of ‘modal register’ and ‘vocal fry’ and found that the average f0 for females was much higher than males during modal register (211.0 Hz for females as compared to 117.5 Hz for males) but not during vocal fry (48.1 Hz for females and 49.1 Hz for males).

216

Frazier

FIGURE 8.7 Average pitch contours of glottalized vowels by gender and glottalization type (production experiment 1; isolation). m = modal voice; w = weak glottalization, c = creaky voice The thin dashed line shows the average pitch contour of high tone vowels.

these pitch values are transformed into s/b, a measurement that is sensitive to a speaker’s natural pitch range, we get the gender-based differences documented in Fig. 8.7. Because males more consistently produce GLOTTALIZED vowels with higher pitch than HIGH TONE vowels, a male’s voice is used for the stimuli of perception experiment 2 (see §4). 2.3 Summary The above description of HIGH TONE and GLOTTALIZED vowels has shown that length, pitch, and glottalization all contribute to this contrast (in increasing order of importance). GLOTTALIZED vowels are more likely to be produced with glottalization, with a larger pitch span, with higher initial pitch (especially by males), and with a slightly longer duration (though this last characteristic is not very robust). The perception experiments presented in §§3–4 are designed to determine which of these cues are attended to by listeners when determining whether they heard a GLOTTALIZED or a HIGH TONE vowel. 3

Perception Experiment 1: Natural Stimuli

As we saw in §2, the phonetic dimensions of length, pitch, and glottalization systematically differ in productions of HIGH TONE and GLOTTALIZED vowels

Pitch And Glottalization As Cues To Contrast In Yucatec Maya

217

in YM. However, there is also a high degree of overlap between the permissible values of a given parameter for both vowel shapes. In other words, many productions of HIGH TONE vowels are legal productions of GLOTTALIZED vowels, and vice versa (with the one notable exception of a full glottal stop, which is only produced with GLOTTALIZED vowels). Accordingly, there are two goals of this perception experiment. The first is to determine if listeners can correctly identify tokens with HIGH TONE and GLOTTALIZED vowels when given no other contextual information. The second is to determine which of the various phonetic dimensions influence the listeners’ decisions. 3.1 Methods 3.1.1 Participants Sixteen speakers who participated in the two perception experiments are reported on here (2 males from Mérida and 9 females and 5 males from Santa Elena). All had also participated in production experiment 2, and some had participated in production experiment 1. All participants are fluent in Spanish in addition to YM; all speak YM in the home and in daily life. Two participants from Santa Elena are also proficient in English. The two participants from Mérida are originally from smaller towns on the western side of the peninsula. 3.1.2 Procedure and Stimuli Participants heard a stimulus and were asked to choose whether they heard a word with a HIGH TONE vowel or a word with GLOTTALIZED vowel. Experiments were run with PRAAT. The participant listened to each stimulus through headphones and then used a mouse to select the appropriate word on the computer screen. For each choice on the computer screen, a word was displayed in both YM and Spanish. The Spanish translation was given to ensure that there were no orthography misunderstandings. Additionally, there was a repeat button on the screen that participants could use up to three times if they wished to hear the stimulus again. Before the experiment began, participants went through a practice perception task. This was done to ensure both that the participant understood the instructions and that they were comfortable using a mouse to interact with the computer screen. If necessary, the participant was shown how to use a computer mouse. This usually required the practice session to be done more than once. There was one participant who did not like using the mouse and did not want to learn. She instead used her finger to point to a box on the screen, and I clicked on the box for her. Participants were encouraged to ask questions if necessary and were compensated for their time.

218

Frazier

The stimuli used for this experiment were natural, unmanipulated recordings of the words k’a’an /k’áa̰n/ ‘strong’ and k’áan /k’áan/ ‘hammock’. The Spanish translations given for these words were ‘fuerte’ and ‘hamaca’, respectively. Each word was spoken by 24 different speakers of YM (from Mérida, Santa Elena, and Sisbicchén); the words were recorded during production experiment 1.14 Participants heard each of the 48 stimuli once and were asked to choose between k’a’an and k’áan as the word they thought they heard, as detailed above. 3.2 Results The first goal of this experiment is to determine to what extent listeners can correctly identify HIGH TONE and GLOTTALIZED vowels without any contextual clues. As shown in Table 8.3, participants performed better than chance (Rao-Scott χ2(1) = 19.2, p < .01),15 though it is doubtful that one should consider accuracy rates of 63–64% as highly successful. This data suggests that, at the very least, participants are not merely guessing and, to a limited extent, can identify HIGH TONE and GLOTTALIZED vowels on the basis of phonetic production alone without any semantic or other discourse cues. TABLE 8.3 Confusion matrix (perception experiment 1) stimulus response k’a’an (glot.) k’áan (h.t.)

k’a’an (glot.)

k’áan (h.t.)

240 (62.5%) 144 (37.5%)

138 (35.9%) 246 (64.1%)

The second goal of the experiment is to determine which phonetic cues influence the listeners’ decisions in discriminating between HIGH TONE and GLOTTALIZED vowels. The results (see Fig. 8.8) show a statistically significant 14  As discussed in §2, speakers from Sisbicchén produce pitch and length differently than speakers from Santa Elena and Mérida. Participants were not told that they were listening to speakers from different locations, and there is no evidence in the data to suggest that participants responded differently to stimuli from different dialect groups. 15  The Rao-Scott χ2 is analogous to Pearson’s χ2 and is adjusted for multiple observations within subjects.

Pitch And Glottalization As Cues To Contrast In Yucatec Maya

219

FIGURE 8.8 Percentage of times response = glottalized vowel on the basis of glottalization type, vowel length, initial pitch, and pitch span of the stimulus.

effect of all measured phonetic dimensions: glottalization type (Rao-Scott χ2(3) = 55.2, p < .01 ), vowel length (z = 3.97, p < .01), initial pitch (z = 4.31, p h, which occurred in a diverse set of languages, both WT and ET, suggesting some kind of areal motivation or coincidence (cf. Chacon 2014). (ii) Aspiration of *t and *k in KOR.

261

Reconstruction of Laryngealization in Proto-Tukanoan

(iii) Aspiration of all word initial *C in WAN (see also Waltz 2002). (iv) *t > d in DES, SIR, YUP and KUB in word-medial position. *t also changed to [r̃]and [n] in KUE and to /r/ (~[r̃] and [n]) LET, TAN and YAH (where given that KUE lacks other important changes found in LET, TAN and YAH, it is possible that there is some sort of areal influence; cf. Chacon 2014). The reflexes of *C’ are somewhat more complex. In (3) a summary of these changes is presented. A sample of the actual cognate sets that support the reconstruction of *C’ is given in (4):3 (3) Reflexes of Proto-Tukanoan Laryngealized Stops Proto- Context sound

Kor Sek Sio Mai

Kue Let Bas Des Kub Tuk Bar Tat Pir Tuy Tan Mak Sir Kar Wan Yur Yah Yup Pis

*p’

#_ V_V]STEM ~V_V]STEM #_V*p

p h h h

p’ h h h

p’ h h h

ʔb h h h

p/b h h h

b vʔp b p

b b b b

b b b b

b b b p

b vʔp vʔb b

b p b b

b p b b

b vʔp vʔb b

b p b b

*t’

#_ ɽ V_V]STEM th ~V_V]STEM th

d t t

ʈ’ t t

ʔd t t

r/l l/t n

r r t d vʔd d

d d d

d d d

d d vʔt t vʔd d

r t d

d vʔt vʔd

d t d

*k’

#_ k V_V]STEM kh

k k

k’ k

g/∅4 k/g k ?

g k

g g

k ∅

∅ ∅ vʔk k

∅ k

k vʔk

k k

g g

3  The orthography used for each language in (5) is almost the same as the orthography of the sources. It is more attentive though to the phonetics and phonological representation, though. Nasalization is marked in two different ways according to the pattern of nasality in each language and: (1) in stems where vowels are all nasalized, a tilde “~” appears before the first letter of the word. In these cases, any segment representing a voiced consonant is phonetically nasalized, e.g. KUB /~ba/ [mã] ‘macaw’ (cf. Gomez-Imbert 2004). (2) in words that have one nasal and one oral syllable, nasalization is marked in the nuclear vowel, e.g. SIO /hẽka/ [hẽka] ‘firewood’. In this case as well, a voiced consonant in the same syllable of a nasalized vowel is phonetically nasalized. The representation of Proto-Tukanoan will also have nasalization being marked All the words above were adapted from Huber and Reed (1992) and Chacon (2014), who researched on several dictionaries and word-lists about Tukanoan languages. 4  Dialectal variation (Lev Michael p.c 2014).

262

Chacon

(4) a. *p’ in stem-initial and medial positions Kor Sek Sio Mai Kue Tan Bas Des Kub Tuk Kar Wan P-T pia p’ia p’ia ʔbia bia jeha jeha jiha jiha –

Gloss

bia bia bia bia bia bia bia *p’ia CHILI – jiba jeba jeba je’pa jepa ja’pa *jip’a LAND/GROUND

b. *t’ in stem-initial and medial positions Kor Sek Sio Mai

Kue Tan

Bas Des Kub Tuk Kar

Pir

P-T

Gloss

rɨkhɨ dɨkɨ ʈ’ɨkɨ ʔdɨkɨ – ~ri’kia rɨkɨ ~dɨkɨ dɨkɨ ~dɨkɨ ~nɨkɨ ~dɨkɨ *t’ɨkɨ HEAVY gɨda gɨda kɨda ɨ’ta ɨta kɨ’ta *k’ɨt’a EXCREMENT kɨta ĩta k’ɨta (g)ĩta gɨta ita

c. *k’ in stem-initial and medial positions Kor Sek

Sio

Mai

Kue Tan Bas

Des

Kub

Tuk

Kar

Wan

P-T

Gloss

kãho k’ãho k’ãho gãho kiau ~abu ~gabo ~gabi ~kabu ~o’be ~abo ~ka’bo *k’ãp’o EAR ɨ̃khe õkwe ũkue ũkue – – ~igi ~i’gi ~ue ~e’ke ~ke ~ke ɨ̃kw’e NOSE

The most significant changes to observe are the following: (i) in general, all ET languages underwent a change of voicing *C’ in stem-initial position. (ii) On the other hand, WT languages merged *C’ with *C in stem-medial position. These changes are the most representative of the split between ET and WT languages. Thus one can only reconstruct *C’ as a class of obstruents in the stem-initial and stem-middle position due to independent and related evidence from both branches of the family. Reflexes of *C’ in stem-medial position in ET have a more diverse pattern, involving voicing of the stop and laryngealization of the preceding vowel, which suggests a retention of *C’ stem-internally for a longer period in these languages. It is also noticeable that nasalized syllables seem to have favored voiced reflexes of *C’. On the other hand, the diverse reflexes of *C’ in stem-initial position in WT languages suggest that this was a late change in this branch.

Reconstruction of Laryngealization in Proto-Tukanoan

263

It should also be noticed that *k’ has more diverse reflexes than other *C’. Chacon (2014) proposes that *t’ was the first consonant of this class to change from Proto-Tukanoan, followed by *p’, whereas *k’ had a later change in all Tukanoan languages in the ET and WT branches. That *k’ has undergone a later change in the family follows from the typological preferences of languages for “glotallic” sounds in the velar region and voiced sounds in more anterior positions (Maddieson 1984:40, see also section 3). Also, the many instances of ∅ ‘zero’ as reflexes of *k’ suggest a similar sequence of changes as *k’ > g > ∅ independently shared across the major subgroups in the family. These changes in TAN and a dialect of MAI does not seem to be directly related, while there seem to be a genetic or areal relation between TUK, TAT, BAR and KAR. 2.1 Laryngealized Consonants versus Voiced Stops Reconstruction of *C’ instead of voiced stops is based on a variety of factors. First, several languages present distinct reflexes that in different ways indicate some sort of laryngeal trace in Proto-Tukanoan stops distinct from sounds with plain voiceless or modal voice phonation. SIO (cf. Wheeler 1987) has several laryngealized stops (/p’/, /s’/ and /k’/ plus a laryngealized retroflex /ʈ’/); SEK also presents retention of laryngealized stops, though evidence for that is variable in the existing sources (see Johnson and Levinson 1990; Piaguaje et. al, 1992; Vallejos 2013); MAI has been reported to have ‘pre-glottalized’ voiced stops (/ʔb/ and /ʔd/, cf. Velie 1975 and Velie et al. 1976, though see section 3.2.1); TUK (Ramirez 1997), WAN (Stenzel 2007), PIR (Waltz 2002) and TAN (Huber and Reed 1992) have laryngealized vowels preceding a reflex of *C’ in stem medial position. In all these cases, a reconstruction of a voiced series *b, *d and *g instead of a laryngealized series would imply in unmotivated reflexes, since there should be no reason for why modal voice would create such a diverse pool of different realizations of laryngeal features. Second, in cases of stem-initial voiced stops and stem-medial voiceless stops as reflexes of *C’ in ET languages, it would be extremely rare that the alleged proto-sounds *b, *d and *g would get devoiced only between vowels (stem-medial position) but retained fully voiced in stem-initial position.5 Along the same lines, before a nasalized vowel, the languages where *C’ changed into a voiced stop underwent an additional sound change where 5  In fact, from a voiced stop series one could expect devoicing in word initial position or an unconditioned sound change where all proto voiced stops would get devoiced. However, the scenario of the reflexes in (4) clearly discourages the postulation of a voiced series of proto-stops.

264

Chacon

voiced stops became nasalized, e.g. *p’ > b > m and *t’ > d > n.6 In cases where *C’ was retained (e.g. SIO) or became voiceless (e.g. KOR) there is no nasalization of reflexes of *C’ in a nasalized context. Nasalization is a feature associated with lenition processes and is likely to cause voicing of an adjacent stop, rather than devoicing. So it would seem very unlikely that *b, *d and *g would get devoiced (and denasalized) before a nasal vowel in stem-initial or medial position in languages like KOR and SIO. In section 3, a more solid phonological analysis, discussing the interplay of laryngeal features will be provided. 2.2 Laryngealized Consonants versus Laryngealized Vowels The laryngeal feature that have been analyzed as a feature of consonants in this paper cannot be interpreted as a feature of vowels, as one might suggest. The stronger arguments against this proposal are outlined below. First, if laryngealization were a feature of vowels, one would expect that it would have been retained both in the first and in the second vowel of stems, not only in the first vowels of stems in a few ET languages (unless some independent reason related to the prosody of stem-initial vs. stem-final syllables can be discovered in future investigations). Second, assuming laryngealized vowels would incur in accepting a quite strange metathesis of laryngealization in Proto-Tukanoan words: for instance, words that I reconstruct as *C’VCV would have to be reconstructed as *CV’CV, and one would have to assume that the laryngeal trace of the vowel was assimilated to the stop and deleted from the vowel; the worse case, however, is in words that I reconstruct as *CVC’V, which would have to be reconstructed as *CVCV’; here one would have to assume that the laryngeal trace of the second vowel moved to the first vowel in order to account for the ET reflexes CV’CV. Third, one can wonder why one never finds a laryngeal trace in an environment surrounding a proto-nasal stop *m and *n. If there were laryngealized vowels in Proto-Tukanoan, than what could be the motivation for restricting the distribution of laryngealized vowels and nasal stops in the same syllable?7 Finally, even if one assumes that laryngealized vowels existed in the protolanguage, there are many phonological changes related to laryngeal stops in the

6  In synchronic terms, most linguists working with ET languages treat [m], [n], [ɲ] and [ŋ] as allophones of /b/, /d/, /j/ and /g/, respectively (cf. for instance Gomez-Imbert 2004, Ramirez 1997; Stenzel 2013; Velie 1976). 7  Synchronically nasalized vowels can be combined with laryngealization, as the TUK word ~de’e ‘mauritia flexuosa (palm species)’ illustrates.

265

Reconstruction of Laryngealization in Proto-Tukanoan

evolution of Proto-Tukanoan that it would be necessary to assume that laryngealized vowels were part of the phonological system of Pre-Proto-Tukanoan, making such a proposal even less concrete. 3

On the Phonetics and Phonology of Laryngeal Contrast in Proto-Tukanoan

In this section I analyze Proto-Tukanoan laryngeal contrasts by a set of distinctive features, combined with a more detailed phonetic characterization *C’ and *C. These will prove important for explaining phonological and morphophonological process in Tukanoan languages, the sound changes involving *C’ and *C and the consonant inventory in Proto-Tukanoan. I am restricting the discussion in this section in terms of [+ consonantal] sounds, not dealing directly with [ʔ] and [h] (for [ʔ] see section 4). The theoretical framework I use for the phonological analysis of Proto-Tukanoan consonants and related sound changes are the ones put forward by Rice (this volume) and Kiparsky (1995). The chart below presents the Proto-Tukanoan consonants classified by the major class features [± consonantal] and [± sonorant], the privative laryngeal feature [constricted glottis] and two manner features [± continuant] and the privative feature [nasal]: (5) A subset of Proto-Tukanoan consonants and feature specification p’

p

Consonant + + Sonorant Constricted glottis ✓ Continuant Nasal

w

m

t’

t

tj’

tj

n

c

+

+ +

+

+ +

+

+ +

+ +

+

✓

✓

✓

✓

s

j

+

+ +

k’

k

+

+

✓

ʔ

✓

h

+

The feature [constricted glottis] (henceforth [cg]) captures all *C’ as a natural class. It allows for a possible range of phonetic realization of this feature in Proto-Tukanoan and intermediate proto-languages, ranging from creaky-voice to ejectives and implosives, as discussed in section 3.1 below. Segments that are unmarked for [cg] have different laryngeal features implemented by default

266

Chacon

and redundant rules in Proto-Tukanoan. A [+ consonantal][– sonorant] segment could be realized as [spread glottis] (henceforth [sg]) if the context (e.g. stressed syllable, special register, etc.) would implement it as a default feature. If no context specifications were available, this class of obstruents would surface as a plain voiceless sound, i.e. unmarked for a laryngeal feature. The feature [voice] is redundantly assigned to [+ sonorant] segments in the phonetic output, a relevant fact for nasal harmony and for the voiced reflexes of *C’ (section 3.3). The feature [± continuant] is relevant not only for the contrast of segments, but also to processes that involve lenition of [+ consonantal][cg] sounds (see section 3.2.1). The feature [nasal] is also important not just for phonemic contrast but also to nasal harmony (see 3.3), being a feature also active among vowels. On the Phonetics of the Laryngeal Contrast among Obstruents in Proto-Tukanoan In preparation for a phonological account of laryngeal contrast, it is important to consider the phonetic realization of *C’ and *C in Proto-Tukanoan and intermediate proto-languages. The reflexes of *C’ and *C can be summarized in the chart below 3.1

(6) Summary of *C’ and *C reflexes types Environment

*C’ [CG] REFLEXES

*C [ ] REFLEXES

Stem-Initial Position

– Voiceless Stops – Voiced Stops – Laryngealized stops

– Aspirated stops – Voiceless consonants

Stem-Medial Position

– Voiceless Stops (merged with reflexes of *C) – Voiceless Stop with laryngealization of the preceding vowel – Voiced Stops

– Voiceless Stops – Voiceless Stop with aspiration of the preceding vowel – Voiced stops (merged with reflexes of *C’)

It is likely that [cg] in Proto-Tukanoan was realized as creaky voice, which can offer a unified explanation for the range of reflexes summarized in the chart above. On the other hand, unmarked obtruents were realized as plain or aspirated stops, depending on the phonological context.

Reconstruction of Laryngealization in Proto-Tukanoan

267

According to Ladefoged and Maddieson (1996), creaky voice is similar to modal voice in that it presents vibration of the vocal folds. On the other hand, it also presents some amount of constriction of the glottis, somewhat in a midpoint between full aperture of the glottis as in voiceless stops and full closure of the glottis as in ejectives, which are by definition voiceless sounds, since full closure of the glottis does not allow the vocal folds to vibrate. Such characteristics in creaky voice may explain the set of reflexes ranging from voiceless, to voiced and laryngealized sounds. Ladefoged and Maddieson (1996) also show that languages can vary as to the timing between the oral gesture of the articulation of a stop and the production of laryngealization. Some languages present pre-laryngealization, when glottal constriction precedes or occurs early on in the articulation of a stop. Others have post-laryngealization, when glottal constriction is delayed with respect to the articulation of a stop; or simultaneously when laryngealization can be active throughout the production of the stop. This sort of timing variation is important to explain the different reflexes of *C’ in stem-initial versus stem-medial position. It is also possible that this type of variation is analogous to what happens with aspiration in voiceless stops in several languages (e.g. Gaelic, cf. Ladefoged and Maddieson 1996), where in word initial position, stops are produced with post-aspiration, while in word medial or final position they are produced with pre-aspiration. This fact is important to consider the evolution of [sg] in Tukanoan languages, as well as vowel laryngealization and aspiration. In addition, it should be noted that the feature [cg] could have developed different phonetic realization in intermediate proto-languages according to different points of articulation. As Maddieson (1984) states, it is common for phonologists to recognize a class of ‘glottalic’ sounds in languages with a series of ejectives in the velar, palatal and alveolar place of articulation, but with an implosive in the labial place of articulation (or else with a labial ejective and an alveolar implosive instead). It is likely that such a variation also existed at some point in the history of creaky voiced stops in Tukanoan languages, what could explain variation in the reflexes between *p’, *t’ and *k’, for instance. On the Phonological Change of [Constricted Glottis] in the Tukanoan Family The discussion in this section will privilege the phonological process and sound changes that affected [cg] and other relevant laryngeal and manner features. It basically deals with cases of retention of [cg], assimilation of [cg] bearing sounds to other laryngeal or manner features and neutralization of [cg]. 3.2

268

Chacon

The feature [cg] has been retained as a contrastive feature of [+ consonantal] sounds only in the WT languages SIO, SEK and MAI.8 In all other languages, [cg] obstruents have been reinterpreted in favor of [voice], [sg] or became simply unmarked for a laryngeal feature (i.e. plain voiceless segments). In the following sections the phonological basis of [cg] evolution will be explored in each branch of the family. 3.2.1 The Evolution of [cg] in WT Languages In the conservative language SIO, [cg] is still a contrastive feature of obstruents. However the constrast of Proto-Tukanoan *C’ and *C can only be found in stem-initial position (Wheeler 1987, Bruil 2014), given the merger of *C’ and *C stops in stem-medial position in WT language. A (simplified) phonemic chart of SIO consonants is given below (adapted from Bruil 2014). (7)

Siona consonants

Consonant Sonorant Constricted glottis Continuant Nasal

p’

p

+

+

✓

w

m

t’

t

s’

s

n

t∫

+

+ +

+

+

+

+

+ +

+

+

✓

✓

✓

+

+

✓

j

+ +

k’

k

+

+

✓

ʔ

✓

h

+

In SIO, [cg] obstruents are realized as creaky voiced or as ejectives, which often causes the following vowel to be perceived as creaky voiced as well (Wheeler 1987:85, Bruil 2014:92). Unmarked obstruents are realized as plain or aspirated in word initial position (Wheeler 1987:84).9 In stem-medial position, where *C’ and *C merged, obstruents can produce post-aspiration of the preceding vowel if the syllable where the vowel is located is not closed by a glottal stop, as in [tuhtu] ‘wind’ (Bruil 2013:90). This suggests that unmarked obstruents are by default assigned the feature [sg] in this environment, which in turn assimilates to the preceding vowel. 8  M AI is a more difficult case, though. See section 3.2.1. 9  Similar facts have been described to Sekoya del Putumayo (Vallejos 2013). We will return to SEK laryngeal features in section 3.3.2 when discussing nasal harmony.

Reconstruction of Laryngealization in Proto-Tukanoan

269

The motivation for a default implementation of [sg] in unmarked obstruents might be related to strengthening the degree of phonetic differentiation in the language between the two classes of obstruents. A parallel phenomenon occurs with [cg] obstruents. Bruil (2014:93) reports that [cg] obstruents are realized as approximants in intervocalic position, with the exception of /k’/.10 Thus /p’/ > [β] and /t’/ > [ɾ]. This allophonic process seems to occur categorically stem-internally,11 and with variation across stem boundaries in compounds and in fast speech. In a featural analysis, [cg] obstruents assimilate to the feature [+ continuant], triggered by the surrounding vowels, concomitantly to [cg] delinking: [cg] > [+ continuant]. Given that [cg] is assumed to be more lenis than [sg] or plain voiceless sounds (see Fallon 2002 and Lavoie 2001), making a [cg] obstruent [+ continuant] and an unmarked obstruent [sg] can be interpreted as a process that maximize the phonetic differences between [cg] and unmarked consonants, even if ultimately that amounts to the delinking of [cg]. Thus, in the phonetic form, [cg] is non-contrastive in stem-medial position in SIO, although it is actually its presence in the underlying level that mark the target obstruents for the assimilation to [+ continuant]. In fact, [cg] in intervocalic position does not exist as a feature of obstruent in any Tukanoan language, given the merger of *C’ and *C in WT and the reinterpretation of [cg] in various ways in ET languages (see section 3.2.2). In MAI, the first phonological descriptions attest for the presence of [cg] stops. Velie (1975) and Velie et al. (1976) analyze /ʔb/ and /ʔd/ as pre-laryngealized voiced stops ([cg][voice]), which occurs in several words Chacon (2014) basis the reconstruction of Proto-Tukanoan *p’ and *t’.12 On the other hand, /g/ as reflex of *k’ is not pre-glottalized, another instance of asymmetry in the reflex of *C’ concerning the velar point of articulation (see section 2). Because *w changed to /b/ in this language, MAI has a unique consonantal inventory in the Tukanoan family. See the chart below (adapted from Velie et al. 1976):

10  /k’/ is realized as [g], another idyosincratic case of *k’ reflexes (see section 2). 11  Which cannot be reconstructed to Proto-Tukanoan, given the merger of *C’ and *C in stem-medial position. 12  More recently, Lev Michael (p.c 2011), who is doing a detailed phonological investigation of the language, reported that the sort of laryngeal contrast analyzed by Velie (1975) and Velie et al. (1976) does not exist in the language anymore.

270

Chacon

(8) Maihɨki consonants

Consonant Sonorant Constricted glottis Voice Continuant Nasal

p

ʔb

b

t

ʔd

n

s

+

+

+

+

+

+ +

+

✓ ✓

✓ ✓

✓

+

✓

j

+

k

g

+

+

ʔ

✓

✓

+

h

+

All [voice] segments have nasal allophones, as discussed in section 3.3.2. The complex articulation of [cg] and [voice] in /ʔb/ and /ʔd/ are reflexes of a change where *p’ and *t’ assimilated to [voice] from the adjacent vowel, becoming pre-glottalized and voiced. Similarly in the ET branch, two different changes show a parallel development to what happened in MAI: one change was from a [cg] obstruent to a [voice] obstruent, while in another change [cg] obstruents neutralized with unmarked ones in stem-medial position, while also causing the preceding vowel to become laryngealized (see section 3.2.2). KOR represents a more extreme outcome of the evolution of [cg]. Obstruents contrast as plain versus aspirated, as illustrated in the chart below (adapted from Cook and Gralow [1993:2]): (9) Koreguahe consonants ph p w wh mh m th t ɽ s

Consonant + + + + Sonorant + + Constricted glottis Spread glottis ✓ ✓ ✓ Continuant + + Nasal ✓

n

ñh ñ

+ + + + + + + + + + + ✓ ✓

kh k ʔ

+ + + +

✓

+ +

j

✓

✓ ✓ ✓

+

h

+ ✓

✓

+

It is remarkable the reinterpretation of the feature [cg], where unmarked obstruents in Proto-Tukanoan became aspirated ([ ] > [sg]) and [cg] obstruents lost

Reconstruction of Laryngealization in Proto-Tukanoan

271

their laryngeal feature ([cg] > [ ]). Thus, the unmarked feature [sg] became the marked feature. Also, KOR is highly innovative in the Tukanoan family because of the [sg] contrast of nasal stops. To the date, tokens of aspirated nasal sounds have no correspondence with other words in ET languages, which suggests that they are cases of borrowing or independent innovation. It is likely that a combination of language internal reasons and language contact could explain the fate of [cg] and the rise of [sg] in KOR. The internal reason for this change was likely the merger of *C’ and *C in WT languages, which can be interpreted as a neutralization process, where the marked feature [cg] was delinked and the segment became unmarked. A default rule in the stem-medial position yield those obstruents as [sg]. Then, unmarked consonants in stem-initial position also became by default [sg], which ultimately made [cg] phonetically less important for maintaining phonemic contrast, causing the loss of [cg] and simplification of the phonology. In SIO an interesting allomorphic process triggered by stress provides further information about the evolution of [cg] in Proto-Western-Tukanoan. Wheeler (1987:90) describes a stress system in SIO based on iambic feet with exhaustive parsing of phonological words.13 Primary stress is assigned to the left-most stressed syllable. When a suffix bearing an initial stop that is the reflex of *C’ is in an unstressed syllable, the stop has a “lenis” realization: the same previously mentioned [+ continuant] allophones of /p’/ > [β] and /t’/ > [ɾ] in stem-medial position, and a [voice] allophone for /k’/ > [g]. When that suffix is in a stressed syllable, however, it has a “fortis” realization, which is a [– continuant] obstruent, unmarked for a laryngeal feature. Thus in SIO (Wheeler 1987) morphemes like -k’o ‘feminine’ and -p’ɨ ‘classifier round’ have alternate forms in stressed syllables as [-ʔkho] and [-ʔphɨ] (where [ʔ] is realized in the coda of a preceding syllable) and in unstressed syllables as [-go] and [-βɨ] respectively. In KOR, the same type of alternation occurs, where -ɽe ‘referential’ alternates between [-ɾe] and [-ʔthe] (Cook and Criswell 1993). In KOR these alternations can only be understood from a diachronic perspective, given that synchronically [cg] is not contrastive among [+ consonantal]. Also, stress related rules are not as productive as in SIO (Cook and Criswell 1993:11). Because of the exceptional nature of these alternations in KOR, forms such as [-ʔthe] are analyzed as archaisms, traces of the old class of [cg] obstruents in Proto-Tukanoan. On the other hand, the SIO data calls for a synchronic explanation. These alternations seem to suggest that stress related rules neutralize [cg] obstruents and unmarked obstruents in a peculiar way. In the one hand, [cg] is still 13  The direction of parsing, left to right or the contrary, is unclear from Wheeler (1987) data.

272

Chacon

maintained in the underlying representation of these suffixes (realized as the coda of the preceding syllable), suggesting a structure-building nature of this process. On the other hand, the obstruent surfaces as [sg], which mirrors the neutralization process of *C’ and *C in WT languages. This calls for a simplified explanation in terms of rule ordering: 1. 2. 3. 4.

Stress causes the delinking of [cg] from its original segment, as a fortition process; The feature [cg] is kept in the underlying representation of the morpheme as an unassociated feature; Syllabification associates the floating [cg] with the coda of the preceding syllables; A default surface rule assigns [sg] to the obstruent;

These alternations are a morphophonological manifestation of the process that neutralized *C’ and *C in stem-medial position in Proto-Western-Tukanoan. The difference is not only between synchrony and diachrony but also a lexical phonological one, given that the environment for the stress alternation is across morpheme boundaries, and for the neutralization of *C’ and *C was at the root level. It seems that the neutralization was categorial in the root level, but is yet allophonic at the stem level. In addition, given that roots in Tukanoan languages are in general CVCV, if the iambic feet in SIO were also the feet pattern in Proto-Western-Tukanoan, *C’ and *C neutralization could be explained as the result of stress placement in the second syllable of stems. 3.2.2 The Evolution of [cg] in Eastern Tukanoan Languages All Eastern Tukanoan languages have voiced reflexes of *p’ and *t’ in steminitial position. Reflexes of *k’ are more diverse in stem-initial position, suggesting a later change in the more recent intermediate proto-languages (see section 2). All ET languages have [voice] as the most inclusive feature of its sound inventory, especially because [sonorant] segments are by default [voice]. Lacking [nasal] sounds is another major characteristics of ET languages, a fact directly related to the development of [voice] in these languages (see section 3.3). Voicing of *C’ in stem-initial position was an assimilation process triggered by the adjacent vowel (which by default is [voice]). Hence, [voice] became the contrastive feature between *p’ > b and *t’ > d in stem-initial position in Proto-Eastern-Tukanoan, while [cg] was only marginally contrastive for the velar stops *k and *k’ and glottal sounds *h and *ʔ. Thus, [cg] as a contrastive

273

Reconstruction of Laryngealization in Proto-Tukanoan

feature of obstruents was restricted mostly to stem-medial position. As a consequence, different kinds of changes affecting *C’ in stem-medial occurred, which represents the diversification of major subgroups within the ET branch. DES has a typical consonantal inventory for Western-ET languages (see section 1), as illustrated below (adapted from Silva 2012): (10) Desano consonants

consonantal sonorant voice constricted glottis continuant nasal

p

b

+

+

✓

✓

w

+

+

t

d

s

+

+

+

✓

✓

+

j

+

+

k

g

+

+

h

✓

+

ʔ

✓

In all Western-ET languages (see section 1), *C’ in stem-medial and in steminitial position became voiced, [cg] > [voice], and unmarked obstruents remained voiceless. In Southern-ET languages, *C’ in stem-medial position merged with *C, though leaving a trace of laryngealization in the vowel preceding reflexes of *p’ and *t’, as for example the word /~ja’bu/ ‘yam’ in TAN (from Proto-Tukanoan *jãp’o). This also occurred with Eastern-ET languages (see below). Given that this sort of reflex of *C’ occurred in languages from different ET subgroups, it is likely that Proto-Eastern-Tukanoan had *C’ as a pre-laryngealized stop in stemmedial position, causing the preceding vowel to be laryngealized, such as *sĩk’e [sĩʔk̰e] ‘to rub’ (~cige in BAS and sã’ke in TUK).14 In the Eastern-ET languages that present this pattern, such as TUK, WAN, PIR, all reflexes of *C’ stops in stem-medial position (*p’, *t’ and *k’) laryngealized a preceding vowel; on the other hand, *C made the preceding vowel aspirated. As a result, [cg] and unmarked obstruents have been neutralized, as the laryngeal contrast was now realized in the preceding vowel. Thus, before a voiceless stop stems are only of two types in these languages: (i) those whose 14  Western-ET is the only branch where no language has laryngealized vowels as reflexes of *C’, which is because in this branch not only all [cg] obstruents became [voice], but also because *ʔ merged with zero (with the exception of DES and SIR, see section 4).

274

Chacon

first vowel is laryngealized, as in ~da’ka [nãʔkã́] ‘do/be together’ in WAN; (ii) and those whose first vowel is post-aspirated, as in ~daka [nãhkã́] ‘miriti fruit’ also in WAN (see Stenzel 2007:332).15 Although the diachronic facts can explain in a simplified way the source for the distributional patterns of vowel aspiration and laryngealization, the synchronic analysis is more complex. Besides proposals that analyze aspiration and laryngealization as segments in coda position, Stenzel (2013) proposes an interesting analysis of laryngealized vowels as a glottal suprasegmental that is underlyingly assigned to individual morphemes. This is an interesting lexicalization path, where a previous contrastive feature of obstruents became a suprasegmental of certain morphemes. On the other hand, Stenzel (2013) follows Gomez-Imbert (2011) proposal that vowel post-aspiration is the result of an early geminate voiceless stop *CC whose first member had debuccalized (i.e *C.C > h.C). Chacon (2014) shows that some cognate sets suggest the reconstruction of *tt and *kk in Proto-Tukanoan (but not *pp), though in order to support Stenzel (2013) and Gomez-Imbert (2011) analysis, one would have to postulate that the aspiration as result of the debucalizaton of *CC was extended by analogy to all reflexes of *C in stem-medial position. In any case, vowel post-aspiration in these ET languages as well as in the WT language SIO is the result of an acoustic enhancement of an old [cg] contrast, where the unmarked obstruents (by default [sg]) would be the source for the vowel assimilation to the feature [sg]. Finally, two allomorphic processes in ET language bear evidence on the early contrast of [cg] in Proto-Eastern-Tukanoan. Ramirez (1997:278) provides an analysis of several nominal morphemes in TUK with two alternating forms: a “lenis” (default) form and a “fortis” form, which occurs in a context where a deverbal suffix has been diachronically deleted. Thus, the classifier morphemes [-wa] ‘arch-like’, [-rɨ] ‘pot-like’ and [-ga] ‘rounded’ have their “lenis” forms when combined directly to a noun, and a “fortis” form [-pha] ‘arch-like’, [-thɨ] ‘potlike’ and [-kha] ‘rounded’ when combined to a verb, in which case the standard

15  WAN is an interesting language in this regard because it is the only ET language that contrast plain voiceless stops with post-aspirated voiceless stops. This contrast is restricted to word initial position only (cf. Waltz 2002; Stenzel 2013). Aspiration is the result of retention of Proto-Tukanoan voiceless stops in stem-initial position. Contact with Arawakan languages has been suggested (Stenzel and Gomez-Imbert 2009) to be responsible for developing such aspirated stops. We now have evidence that in the Tukanoan family [spread glottis] was a default feature for unmarked obstruents in certain phonological contexts.

Reconstruction of Laryngealization in Proto-Tukanoan

275

deverbal suffix -di has been deleted (e.g. *ãyu-di-ga [be.good-DEVERBALIZERROUNDED] > [ãyukha]).16 TUK has exactly the same consonantal inventory as DES (see [10] above). Thus, aspiration is not contrastive and can only be motivated in the “fortis” forms above by the diachronic analysis of a fortition process in Proto-EasternTukanoan where [cg] alternated with unmarked obstruents, which by default surfaced as [sg] in word-internal position, similarly to the stress alternations in SIO (see section 3.2.1). Thus in TUK: (*p’ > b >) w alternates with (*p >) ph; (*t’ > d >) r alternates with (*t >) th; and (*k’ >) g alternates with (*k >) kh. The morphemes that trigger the alternating forms are reconstructed to Pre-Tukano as *-sa ‘futurate’ and *-di ‘nominalizer’ by Ramirez (1997). In fact, the reconstruction of these morphemes to Proto-Eastern-Tukanoan yield the forms *-sa ‘go’ (cf. Barnes 1992:16) and *-ti ‘nominalizer’.17 So, one may interpret that the alternate forms of the bound morphemes in TUK are the result of the diachronic process described below: 1. 2.

Deletion of a bound-morpheme, which created a floating [sg] feature. Linking [sg] to the initial obstruent of the following morpheme, causing the delinking of the previous [voice] (< *[cg]) feature.

As a result, the alternating morphemes have been lexicalized with two allomorphs: the “basic” forms with reflexes of the change [cg] > [voice] and the “alternate” forms with a residual [sg], which bear evidence to the old [cg] versus (default) [sg] contrast in Proto-Tukanoan. Gomez-Imbert (2004:59) describes a similar process with bound-morphemes in BAS and TAT. In BAS there is alternation between [b] and [h], [d]/[r] and [t], [g] and [k]. In TAT the alternation is between [w] and [p], [r] and [t] and “ø” (zero) and [k]. In both languages, the basic forms are the reflexes of *p’, *t’ and *k’, and the alternate forms are reflexes of *p, *t and *k respectively. In BAS *p > h; and in TUK *k’ > g > ø and *p’ > b > w in bound morpheme initial position, similarly to TUK. Gomez-Imbert (2004) analyzes the reflexes of the forms with *C’ reflexes as the basic since they are unconditioned. On the other hand, the alternate forms that have reflexes of *C are conditioned by a “latent” consonant. For instance, the morpheme be(t) ‘negation’ has /t/ as latent segment, which—when not surfacing—spreads its manner and laryngeal features, including [-voice], to 16  Glosses are mine. 17  Reconstruction of this morpheme is still tentative. It is based on the fact that while several ET languages have -di others have -ti.

276

Chacon

the following consonant. On the other hand -~koa ‘emphatic’ does not have a latent segment and, thus, does not trigger the alternating forms. Thus, Gomez-Imbert (2004) analysis of BAS and TAT data provides synchronic evidence to the diachronic hypothesis of allomorphic alternations in TUK. Moreover, in the comparison of alternations in TUK versus those in BAS and TAT it is intriguing why in the latter there is no sign of [sg]. This could be related to the fact that BAS and TAT processes are synchronic, so [sg] (an archaic feature from Proto-Tukanoan) is never retained in the synchronic phonology of these languages, while in TUK the feature is clearly an archaism of only a dozen of allomorphs. 3.3 Nasalization and Laryngeal Features Perhaps all Tukanoan languages have nasal harmony, i.e. when [nasal] assimilates from a source (segment, syllable or morpheme) and target all segments of a particular natural class within a given domain (e.g. phonological word). The discussion of nasal harmony in this section is restricted to segments that function as targets and blockers of nasal harmony. It supports a strong correlation of [+ sonorant], [voice] and [nasal] assimilation in these languages and explains the exceptions of nasal harmony due to the different ways Tukanoan languages have developed the feature [voice] and created a natural class encompassing voiced obstruents and sonorants. 3.3.1 Nasal Harmony in ET languages In ET languages, all [voice] segments have nasal allophones (e.g. /b/ > [m], /g/ > [ŋ] and /a/ > [ã]). Thus Proto-Tukanoan nasal obstruents *m and *n have actually merged with [voice] obstruents /b/ and /d/, since there are no contrastive environments such as syllables as [bĩ] or [mi] in any ET language. Voiceless sounds are transparent to nasal harmony in root internal position, where one does find contrast of different syllabic patterns, such as [pi] and [pĩ]. The situation is different across morpheme boundaries, where a voiced sound will be a natural target for [nasal] assimilation, but a voiceless one will block it. See the examples below from TUK (Ramirez 1997:60–1): (11a) ~kadi-ya [kãɽ̃ĩj̃ã ] sleep-IMPERATIVE ‘sleep’ (11b) ~bede-pɨ [mẽnẽpɨ] inga.tree-LOCATIVE ‘At the ingá tree’

Reconstruction of Laryngealization in Proto-Tukanoan

277

Thus, in ET languages [voice] segments are universal targets for [nasal] assimilation, whereas voiceless segments block [nasal] assimilation across morpheme boundaries, but are transparent to this process morpheme internally. This suggests that the feature [voice] has a default [-voice] value implemented to unmarked obstruents in a cycle posterior from the root level, where the absence of a laryngeal feature would make unmarked obstruents transparent to nasal harmony. 3.3.2 Nasal Harmony in WT Languages In WT languages nasal harmony has a more diverse pattern across the different languages, as well as important differences from ET. In SIO the target for [nasal] assimilation are all segments classified as [+ sonorant]. This is true both in root-internal position (as in /sɨjo/ [sɨ̃ɲ̃õ] ‘yellow’) and across morpheme boundaries (as in /t’ɨ̃o-je/ [t’ɨ̃õɲæ̃ ] ‘he brings’) (cf. Wheeler 2000:184). On the other hand, [– sonorant] segments are blockers for [nasal] assimilation, both root-internally (as in /p’ãp’i/ [p̰ãβi] ‘to touch’) and across morpheme boundaries (as in /p’ãa-p’i/ [p̰ããβi] ‘he did not do’) (Bruil 2014:124). Segments /h/ and /ʔ/ that are unmarked as [± sonorant] seem to be transparent to [nasal] assimilation (cf. Bruil 2014). A similar pattern seems to hold for KOR (cf. Cook and Criswell 1993; Dupont 1988), though with an interesting exception. KOR /ɽ/ (a reflex of *t’) is unlike other “sonorant” sounds in the language, since it blocks [nasal] assimilation, as in /kãhoɽo/ [kãhõɾo] ‘ear’ (Cook and Criswell 1993:8). In addition, the [+ sonorant] segment /j/ is transparent to nasal harmony in root-internal position (presenting no nasalized allophone, but allowing spreading of [nasal]). However, it is a normal target for [nasal] assimilation across morpheme boundaries, as exemplified by the word /jẽmẽ-jo/ [ɟẽmẽɲo] ‘tongue’. The exceptionality of /ɽ/ can be explained diachronically. Since this phoneme is the reflex of *t’, a sound that originally was [– sonorant] in Proto-Tukanoan, /ɽ/ is the only [+ sonorant] non-nasal sound in KOR that was not [+ sonorant] in the proto- language. Thus, it seems that KOR nasal harmony system has not yet “revised” /ɽ/, given it the same status it had as Proto-Tukanoan *t’. KOR /j/ is actually the reflex of the merger between different coronal sounds, including *j and the stop *tj’, as Chacon (2014) demonstrates. Thus in root internal position it seems that the neutralization of *j and *tj’ favored the unmarked (default [– sonorant]) feature, which by principle is and was opaque to nasal harmony. The same is not true for bound morphemes in KOR where /j/ is the initial consonant: all of them, it seems, have derived from Proto-Tukanoan *j. In SEK, targets of [nasal] assimilation in root internal position include all [+ sonorant] sounds (/w/ > [w̃ ], /j/ > [ɲ] and vowels) and the obstruent

278

Chacon

/d/ > [n], (Johnson and Levinson 1990). Vallejos (2013) also shows that in the Putumayo variety of SEK /ʔ/ has [ŋ] as a nasal allophone, a clear innovation from the Aguarico variety documented by Johnson and Levinson (1990). Indeed, [n] as an allophone of /d/ (< *t’) is also a clear innovation from other WT languages (see for instance KOR /ɽ/). Given that /d/ and [+ sonorant] segments are both [voice], SEK nasal harmony is more similar to the ET pattern. In fact, as the innovation concerning the nasal allophone of /ʔ/ suggest, SEK seems to be neutralizing [cg] and [voice] sounds as equal targets for [nasal] assimilation. In MAI, nasal harmony only occurs tautomorphemically, according to Michael et al. (2012). All [voice] obstruents are natural targets for [nasal] assimilation including [+ sonorant] segments: /ʔb/ > [ʔm], /b/ > [m], /ʔd/ > [ʔn], /j/ > [ɲ]. Velie et al. (1976) also include /g/ > [ŋ], though Michael et al. (2012) treat /g/ as a blocker of nasal harmony. Both analysis, however, agree that obstruents unmarked for a laryngeal feature (/p/, /t/, /s/ and /k/) do not assimilate to [nasal]. Michael et al. (2012) also provide evidence that these obstruents block nasal harmony in stem-medial position, since there are no morphemes where the first syllable is nasal and the second syllable bearing an unmarked obstruent is also nasal. Thus, given the changes where [cg] > [cg] [voice] in MAI, and given that [+ sonorant] segments are also [voice], the feature [voice] creates a natural class for [nasal] assimilation.18 3.3.3 Concluding Remarks about Nasal Harmony and Laryngeal Features The discussion in the previous sections support the following generalizations: – All reflexes of Proto-Tukanoan [+ sonorant] segments are targets for nasal harmony. – All reflexes of Proto-Tukanoan [– sonorant] segments unmarked for [cg] block nasal harmony. – The great majority of targets of nasal harmony are ultimately [voice] in Tukanoan languages, whether [voice] is an underlying feature of [– sonorant] obstruents (reflexes of [voice] assimilation to [cg] obstruents), or a redundant feature of [+ sonorant] segments. – Some [voice] segments are blockers for nasal harmony. These include segments that are underlying [cg] (such as SIO [β] and [ɾ] allophones of /p’/ and /t’/), or changes involving loss of [cg], such as KOR /ɽ/ (< *t’), and potentially MAI /g/ (< *k’). 18  This is true despite the inconsistencies among Michael et al. (2012) and Velie et al. (1976), given that reflexes of *k’ tend to pattern differently from other *C’ reflexes in several Tukanoan languages.

279

Reconstruction of Laryngealization in Proto-Tukanoan

– The sole [cg] segment that is a target for nasal harmony is Putumayo SEK /ʔ/, a clear innovation from the Aguarico SEK variety, where /ʔ/ is a transparent segment. Based on the comparison of the Tukanoan languages, one can conclude that Proto-Tukanoan had [+ sonorant] segments as the natural targets for [nasal] assimilation, [– sonorant] segments as blockers and the unmarked consonants for the feature [sonorant] as transparent segments. However, given that in Proto-Tukanoan the only [voice] segments were also [+ sonorant], it is complicate to tell which feature was the relevant one for creating the condition for [nasal] assimilation. It is possible to conclude, though, that Tukanoan languages show a strong correlation between nasality and voicing. In section 2 it was also highlighted that nasalization was an important conditioning environment for the change [cg] > [voice]. Inconsistencies in the nasalization of [voice] segments have to do with the different ways each language has undergone the change [cg] > [voice], and how they have integrated [voice] in their phonology. This is an additional piece of evidence that [voice] is an innovation in Tukanoan languages. 4

Glottal-Stop, Laryngealization and Supra-Segmentals

This section presents an additional source of laryngealization in Tukanoan languages, namely the glottal stop *ʔ. In addition, it discusses other aspects of laryngealization in Proto-Tukanoan and in its daughter languages, such as tone and syllable boundary. It is possible to reconstruct a glottal stop *ʔ for Proto-Tukanoan independently from *C’: (12) Reflexes of *ʔ Kor

Sek

Sio

Mai Kue Tan

Bas

Des

Kub Tuk

Kar

Wan P-T

Gloss

~de’e wa’i jɨ’ɨ ma’a – we’e

~de’e wa’i jɨ’ɨ ma’a – –

~de’e wa’i jɨ’ɨ ma’a k’õ’a –

– bai ji ma – –

~dee wai jɨ ~baa ~goa wei

~de’e wa’i jɨ’ɨ ~ba’a ~goa –

~dei – jɨ ~ba ~kua wei

~dee wai jɨɨ ~baa ~owa –

~da’a wa’i jɨ’ɨ ~ba’a ~ko’a –

Miriti Fish/Meat I path bone black-ink

– uai jehɨ ma koa –

– wa’i ji’i ~ba’a ~u’a –

~de’e wa’i jɨ’ɨ ~ba’a ~o’a –

*ne’e *wa’i *jɨ’ɨ ma’a k’õ’a we’e

280

Chacon

As one can see, many languages merged *ʔ with ∅ ‘zero’ (actually this has implications in supra-segmental phonology, which will be discussed below). Also, one can only reconstruct *ʔ in stem-medial position between two vowels. This fact support Gomez-Imbert’s (2011) idea that /ʔ/ is a default stop in Tukanoan languages, whose main function is to create a CV syllable. It is interesting that some languages that have lost *ʔ also developed different ways to keep two vowels apart, such as KUE using /h/ in one of the words from the cognate set above, or TAT and KAR that epenthesized consonants with a homorganic place of articulation, as one in the following words in comparison with TUK: TAT õwa : TUK õʔa ‘bone’, TAT pɨga : TUK pɨʔa ‘two’.19 Besides reflexes of *ʔ, which always occur between vowels, Tukanoan languages have also been analyzed as having /ʔ/ in syllable coda position: (C)VʔCV. Reflexes of *C’ are one important source for this latter kind of laryngealization. In languages that laryngealization as reflexes from *C’ and *ʔ, there seems to be no reason to treat them as separate sounds, as they have the same laryngealization effect on the preceding vowel. For instance, in TUK (cf. Ramirez 1997) and WAN (cf. Stenzel 2013), laryngealization is treated as a supra-segment assigned to certain morphemes and always associated with the first vowel (or mora) in a word (see section 3.2.2).20 Nevertheless, there are many instances of /ʔ/ in CVʔ syllables that do not reconstruct to Proto-Tukanoan straightforwardly, as for instance in SIO /haʔsu/ : TUK /asi/ ‘hot’, SIO /tɨʔte/ : TUK /dɨte/ ‘to cut’. This calls for further comparative studies for determining the different sources of laryngealization in Tukanoan languages. In the ET languages that have some form of laryngealization, /ʔ/ is associated with a surface low tone and creaky voicing in the preceding vowel. In WT, on the other hand, languages like KOR, SEK and SIO have the glottal stop associated with high-pitch (as in KOR) or stress (as in SIO and SEK). In MAI, a WT language that did not retain the glottal stop, the reflex of *ʔ is a low tone (Wheeler 1992). On the other hand, in the ET languages that also did not retain /ʔ/, such as BAS and KUB, reflexes of *ʔ and *C’ are associated with 19  There are many cases of diphthongs in Tukanoan languages, including in those languages that retained *ʔ. It is still premature to say how productive were diphthongs in ProtoTukanoan as well as to what extent was *ʔ productive as a default consonant. 20  A crucial fact for treating laryngealization as a supra-segment in TUK and WAN is the exceptional sequence of a vowel + laryngealization + stop (e.g. CVʔCV), which would be an exceptional case of coda if laryngealization were treated as a segment (cf. Ramirez 1997; Stenzel 2013, 2007). The lack of laryngealization in other environments also prevents from treating laryngealization as a feature of vowels.

Reconstruction of Laryngealization in Proto-Tukanoan

281

surface low tone in the first vowel of words, while at the same time a floating High tone can get associated with morphemes to the right, such as suffixes or a different stem in compounds. See the examples below from KUB: (13a) ~dei bohɨ miriti.palm leaf ‘miriti palm leaf’

[nẽ̀i ̃́ bóhɨ́] (from *~de’e ‘miriti’)

(13b) kɨbo foot ‘a foot’

[kɨ̀bóbá]

=ba CL.TIED_UP

(from *k’ɨp’o ‘foot’)

Thus, WT and ET languages have opposite patterns regarding the way /ʔ/ and reflexes of *ʔ and *C’ interact with the prosody of stress and tone. In addition, the fact that languages that have no instance of /ʔ/, such as KUB and MAI, have reflexes such high or low tones is a very important sign that the directionality of sound changes is actually from segmental laryngealization to supra-segmental proprieties like tone. Thus, it seems less probable that tones in Proto-Tukanoan gave rise to segmental laryngealization. This analysis is consistent with the general trend in tonogenesis, where segmental proprieties related to particular laryngeal settings can foster the emergence of tone systems. 5 Conclusion This paper discussed the reconstruction of laryngealized consonants in ProtoTukanoan and analyzed the phonological basis of related sound changes. It has been presented and refuted alternative hypothesis, such as whether laryngealization is a feature of vowels rather than obstruents or whether ProtoTukanoan had [voice] instead of [constricted glottis] as the contrastive feature of obstruents. Another alternative would be to think that laryngealization is derived from supra-segmental phonology, such as tone. The data analyzed in this paper show the reverse, that segmental laryngealization was responsible for the emergence of tone patterns. The analysis of sound changes from Proto-Tukanoan discussed the diachronic evolution of the feature [constricted glottis], its relation to other laryngeal features such as [voice] and [spread glottis], and the assimilation, neutralization and reanalysis processes that affected the laryngealized obstruents.

282

Chacon

Reconstruction of laryngealization has some interesting genetic, areal and pre- historic implications. First, the recognition of creaky voiced stops in ProtoTukanoan implies the revision of the classification of Tukanoan family made by Waltz and Wheeler (1972) to what is presented in (2) above and extensively argued for in Chacon (2014). Second, given that glottalic sounds usually seem to bundle as an areal feature world-wide, it makes one wonder what was the linguistic composition of the area occupied by Proto-Tukanoan speakers and what might have triggered the sound changes that eliminated laryngealized stops in the majority of Tukanoan languages. In the present, only Nadahup (e.g. Epps 2005) as neighbouring languages to ET languages presents laryngealized stops. In addition, glottalic sounds are more common as an areal feature in the Andes region, rather than in the Amazon, though there are some scattered cases of laryngealized/glottalic stops in present Amazonian languages, and this picture could be more different in the past as well. Finally, it should be mentioned that this is the first systematic approach to laryngealization in Proto-Tukanoan. Future studies will likely refine this analysis, especially by adding more details to the phonetics of Proto-Tukanoan laryngealized stops and understanding in a deeper way the relationship between laryngealization, syllable structure and tones. References Bruil, Martine. 2014. Clause-typing and evidentiality in Ecuatorian Siona. LOT: Utrecht, the Netherlands. ISBN: 978-94-6093-133-8. Chacon, Thiago. 2014. A revised proposal of Proto-Tukanoan Consonants and Tukanoan Family Classification. International Journal of American Linguists. Vol. 80, n. 3. pp. 275–322. Cook, Dorothy and Linda L. Criswell. 1993. El idioma koreguaje (Tucano occidental). Bogotá: ILV. Dupont, Carlos M. 1988. Armonía nasal en la lengua koreguaje. Cuadernos de lingüística hispánica. Tunja: Universidad Pedagógica y Tecnológica de Colombia, 105–125. Epps, Patience. 2005. A grammar of Hup. PhD dissertation. University of Virginia. Fallon, Paul D. 2002. The synchronic and diachronic phonology of ejectives. Outstanding Dissertations in Linguistics, series ed. by Laurence Horn. New York: Routledge. Gomez-Imbert, Elsa. 2011. Famille Tukano. Dictionnaire des langues du monde. Paris: Presses Universitaires de France. ———. 2004. Fonología de dos idiomas Tukano orientales. Amerindia, no 29. Paris: CELIA.

Reconstruction of Laryngealization in Proto-Tukanoan

283

Huber, Randal Q. and Robert B. Reed (compilers). 1992. Vocabulario comparativo: Palabras selectas de lenguas indígenas de Colombia. Bogota: Instituto Lingüístico de Verano. Johnson, Orville and Stephen Levinson. 1990. Gramática secoya. Cuadernos Etnolinguísticos 11. Quito: Instituto Lingusitico de Verano. Kiparsky, Paul. 1995. The Phonological Basis of Sound Change. Handbook of phonological theory, ed. by John Goldsmitth. Malden: Blackwell. Ladefoged, Peter and Ian Maddieson. 1996. The sounds of the world’s languages. Cambridge: Blackwell. Lavoie, Lisa M. 2001. Consonant Strength; phonological patterns and phonetic manifestations. Outstanding dissertations in linguistics. New York: Garland Publishing. Maddieson, Ian. 1984. Patterns of Sounds. Cambridge Studies in Speech Science and Communication. Cambridge: Cambridge University Press. Malone, T. 1986. Proto-Tucanoan and Tucanoan genetic relationship. Colombia: Instituto Linguistico de Verano. Ms. Mason, J.A. 1950. The Languages of South American Indians, Bureau of American Ethnology. Bulletim 143, v.6. 157–317 and map. Michael, Lev, Stephanie Farmer, John Sylak. 2012. Syllabic Nasality and nasal consonant harmony in Maihɨki. Handout presented at the conference Amazonicas 4: Lima, Peru. Piaguaje, Ramón, Elías Piaguaje, Orville E. Johnson, and Mary Johnson, compilers. 1992. Vocabulario Secoya. Quito: Instituto Lingüístico de Verano. Ramirez, Henri. 1997. A fala Tukano dos Ye’pâ-Masa: Tomo I: Gramática; Tomo II: Dicionário. Manaus: Inspetoria Salesiana Missionaria da Amazônia, CEDEM. Rodrigues, Aryon D. 2007. Proto-Tupí. Línguas e Culturas Tupí, ed. by Cabral, Ana Suelly Arruda Câmara and Aryon Dall’Igna Rodrigues. Volume I. Campinas: Editora Curt Nimuendajú. Silva, Wilson. 2012. A Reference gramar of Desano. PhD dissertation. University of Utah. Stenzel, Kristine. 2013. A Reference Grammar of Kotiria (Wanano). Lincoln: University of Nebraska Press. ———. 2007. Glottalization and Other Suprasegmental Features in Wanano. International Journal of American Linguists. 73:331–366. University of Chicago Press. Stenzel, Kristine and Gomez-Imbert, Elsa. 2009. Contato Lingüístico e Mudança Lingüística no Noroeste Amazônico: o caso do Kotiria (Wanano). Revista da ABRALIN 8:71–100. Vallejos, Rosa. 2013. El Secoya del Putumaryo: aported fonológicos para la reconstrucción del Proto-Tucano Occidental. Liames 13. pp. 67–100.

284

Chacon

Velie, Daniel. 1975. Bosquejo de la fonología y gramática del idioma orejón (coto). Datos Etno-Lingüísticos, 10. Lima: Instituto Lingüístico de Verano. Velie, Daniel, Ruth Brend, Esther Powlison. 1976. Fonología del Orejón. Datos Etnolinguísticos: Colección de los Archivos del ILV 2008. Lima, Peru: Instituto Linguistico de Verano http://www.sil.org/americas/peru/show_work.asp?id=52278. Waltz, Nathan E. 2002. Innovations in Wanano (Eastern Tucanoan) When Compared to Piratapuyo. International Journal of American Linguistics, Vol. 68, No. 2. 157–215. Waltz, Nathan and Alva Wheeler. 1972. Proto-Tucanoan. Comparative Studies in Amerindian Languages, ed. by Esther Matteson. 19–49. The Hague: Mouton. Wheeler, Alva. 1992. Comparaciones Lingüisticas en el Grupo Tucano Occidental. Estudios Comparativos Proto Tucano, ed. by Stephen H. Levinsohn. 17–53. Santafé de Bogotá: Alberto Lleras Camargo. ———. 1987. Gantëya Bain. El pueblo siona del río Putumayo, Colombia. Vol. I: Etnología, Gramática y Textos; vol. 2: Diccionario. Bogotá: Instituto Linguistico de Verano.

CHAPTER 11

The Status of the Laryngeals ‘ʔ’ and ‘h’ in Desano Wilson Silva 1 Introduction This study investigates the laryngeal segments h and ʔ that occur in Desano, an Eastern Tukanoan language (ET hereafter). The Tukanoan language family has 20 languages, which are divided into two main branches: Western and Eastern.1 These languages are spoken in Northwest Amazonia, in the Brazilian, Colombian, Peruvian, and Ecuadorian territories. The estimated number of speakers of all ET languages combined is only 28,000. Tukano, the largest, has c.10,000 speakers (Stenzel 2004:20), and Pisamira is the smallest, with fewer than 50. Desano is spoken in communities along the Vaupés River and its tributaries in Brazil and Colombia, and also along the Tiquié River in Brazil. Desano is spoken by approximately 200 people. Typologically, Desano is polysynthetic and agglutinating. It is morphologically complex, consisting of a root/stem and suffixes. Desano is almost exclusively a suffixing language. This chapter is organized into six main sections: Section 2 introduces the question under investigation, the data, the procedures used for the analysis and some basic information about Desano phonology. Section 3 presents a description of the occurrence of the glottal segments, and an outline of the * I would like to thank Joyce McDonough and Lyle Campbell for their comments on earlier stages of this study. I also thank the three anonymous reviewers for their helpful questions, comments and criticisms on earlier drafts of this chapter. Of course, I am responsible for any mistake and shortcomings it might contain. A version of this study was presented at the international conference Amazonicas III, held at Universidad Nacional de Colombia, in Bogota, in April 2010. I thank the audience for their feedback. I gratefully acknowledge support from the Endangered Language Fund, Hans Rausing Endangered Language Documentation Project (FTG-155), UNESCO/FUNAI-Museu do Índio of Rio de Janeiro (Projeto de Documentação de Línguas Indígenas), and NSF/NEH-DEL (PD-50023-12) for funding my research on Desano. 1  The Eastern Tukanoan languages are: Bará, Barasana, Taiwano (or Edúuria), Desano, Karapana, Kubeo, Makuna, Piratapuyo (or Waikhana), Pisamira (or Papiwá), Siriano, Tanimuka, Retuarã, Tatuyo, Tukano, Tuyuca, Wanano (or Kotiria) and Yurití. The Western Tukanoan languages areː Korenguaje, Orejón, Secoya and Siona (Gomez-Imbert 2007:402, fn.2). Barnes (2006:131) divides the Tukanoan Family in three groups: Eastern, Middle and Western.

© koninklijke brill nv, leiden, ���6 | doi ��.��63/9789004303218_012

286

Silva

previous analyses of the glottal stops in Desano. In Section 4, I discuss the status of the glottal segments h and ʔ. A brief overview of how these glottal segments have been analyzed diachronically is presented in Section 5, followed by the conclusions in Section 6. 2

Background: The Question, the Data and Previous Analyses

The aim of this chapter is to investigate the phonetic and phonological status of the glottal segments h and ʔ, which occur in root and bound morphemes in Desano. Roots can be monosyllabic ((C)V) or disyllabic ((C)V. (C)V), as illustrated in (1) for plain roots and (2) for laryngealized roots. Syllable structure in Desano is (C)V, except for the laryngeals ʔ and h which may appear in coda in the first syllable of roots. Vowel sequences are bisyllabic (V.V). Thus laryngeals may appear in the coda of the first syllable or the onset of a vowel initial second syllable of roots. (1) Plain roots V i ‘do’ CV ye ‘jaguar’, nẽ ‘buriti fruit’2 CV.CV kãɾĩ ‘to sleep’, gobe ‘hole’ CV.V wea ‘clay’, bua ‘to go down’ V.CV ãɾĩ ‘to be’, abe ‘sun’ V.V ãĩ ‘to take’, oa ‘skunk’ (2) Laryngealized roots CV.ʔV duʔu ‘to release’, duʔa ‘to yank’ V.ʔV oʔo ‘to give’, aʔʉ ‘dad’ CVʔ.CV gʉʔɾa ‘stool’, nõʔmẽ ‘maiden’ Vʔ.CV ãʔɾĩ ‘to say’, õʔmã ‘to run’ V.hV oho ‘banana’, eha ‘to arrive’ CV.hV duhu ‘to leave’, duha ‘to stay’ CVh.CV mʉ̃ htã ‘to break’, dehko ‘water’ Vh.CV ahsi ‘be hot’, ohte ‘to plant’ In previous descriptions of the phonology of the language, Kaye (1965, 1970) and Miller (1976, 1999) treated the laryngeals h and ʔ as full glottal consonant segments. In the present analysis, I consider the distribution patterns and 2  The tilde is used to indicate nasalization.

The status of the laryngeals ʔ and h in Desano

287

their phonetic realization of these segments and argue they are better treated as a suprasegmental laryngeal feature that is assigned to root morphemes in Desano. In the ET literature, the glottal stop ʔ is referred to as ‘glottalization’, and the glottal fricative h, as ‘aspiration’. I will retain this terminology to refer to these laryngeal features. As I discuss in detail in the next sections, the laryngeal sounds ʔ and h show distributive patterns that are distinct from other consonants. First, they are the only segments that occur in coda position, and the only consonants that appear in consonantal clusters, as illustrated in (1) and (2). It is important to note that these segments always appear after the first vowel of the root morpheme. Second, both segments may create a short echo vowel, so that words like duʔa ‘to yank’ and duha ‘to stay’ can appear phonetically as [duʔua] and [duhua], respectively. Third, these segments (usually h) may occur in suffixes: they appear as allophones of the phonemes [k] and [ɾ], so -ka ‘dubitative’ and -ɾã ‘plural animate’ appears phonetically as [-ha] and [-hã] respectively. Lastly, the occurrence of h and ʔ causes the shortening of the proceeding vowel. According to this analysis, the occurrence of h or ʔ in roots is due to the realization of the laryngeal features ‘glottalization’ [+ constricted glottis] and ‘aspiration’ [+ spread glottis] which are realized either as ʔ or h respectively (Kenstowicz, 1994:493). I use phonetic evidence to support the claim that the phonetic realization of h and ʔ is better treated as one single prosodic phenomenon that I call suprasegmental laryngealization. In the next section I present an overview of the Desano segment inventory based on the analysis of my field data. 2.1 Desano Consonant and Vowel Inventories Desano has a total of 17 phonological segments, 11 consonants and 6 vowels presented in Table 11.1 and Table 11.2, respectively. These segments are characterized by features following Kenstowicz (1994). All consonants, except ʔ and h, occur word-initially. The language does not allow coda consonants except for ʔ and h. The segments [d] and [ɾ] (flap) are in complementary distribution: [d] occurs in word initial position and [ɾ] word internally.3 Voiceless obstruents are unaspirated. In initial position these segments may exhibit allophonic aspiration; however, aspiration is not contrastive in Desano. The glottal ­segments 3  The alveolar flap /ɾ/ is found in the Dihputiro dialect of the Papurí River and the retroflex flap /ɽ / (which used to be represented by l in the orthography by speakers) is found in the Boreká dialect of the Tiquié River and Vaupés. Currently, after workshops on orthography development, the group has decided to represent these two sounds with the alveolar flap symbol ‘ɾ’ in the orthography.

288

Silva

ʔ and h appear in parenthesis in Table 11.1 because they are the focus of this investigation. TABLE 11.1 Desano consonant phonemes

PLOSIVE

[– voiced] [+ voiced] FRICATIVE APPROXIMANT

LABIAL

CORONAL

VELAR

GLOTTAL

p b

t d s y

k g

(ʔ)

w

(h)

TABLE 11.2 Desano vowel phonemes

[back] [front] [high] [round]

/i/

/e/

/ʉ/

/u/

/o/

/a/

– + + –

– – – –

– – + –

+ – + +

+ – – +

+ – – –

Desano has six underlying contrastive vowel segments. Table 11.2 illustrates the vowel distinctions according to the features [back], [front], [high] and [low].4 The features presented in Table 11.2 allow an interpretation of the vowel /ʉ/ as a high central unrounded vowel. Kaye (1965:6; 1970:12) characterizes the vowel /ʉ/ as a high back unrounded vowel. Miller (1976:108; 1999:9) describes the vowel /ʉ/ as a high central unrounded vowel. A plot of two formants (F1 vs. F2) provides a description of the vowel qualities in Desano. Figure 11.1 shows that the vowel ʉ in Desano should be characterized as a high central (unrounded) vowel. This account shows a symmetrical vowel system (Lindblom 1988).

4  These vowels are representative of the proto-vowel system for the Tukanoan languages (Barnes 1999:210).

The status of the laryngeals ʔ and h in Desano

FIGURE 11.1

289

Formant plots of the six vowels of Desano.5

2.2 Syllable Structure In Desano, the structure of the syllable is predominantly CV and V. The syllables in the language are generally open, except that there are also occurrences of the syllable shapes (C)Vʰ and (C)Vˀ, as shown in (3e, f, h, i).6 The glottal stop and the glottal fricative are the only segments that can occur in coda position. A vowelinitial syllable can only occur word-initially or after another vowel, but there are no VC.V syllables. This phenomenon follows mostly from the fact that there are no (C)VC syllables, except where the final C is glottal stop or a glottal fricative. (3) a. bui b. ~ʉbʉ c. ~yega d. do’e e. ~ne’ka f. ~a’di

[bu.i] CV.V [ʉ̃ .mʉ̃ ] V.CV [ɲẽ.ŋã] CV.CV [do.ˀe] CV.CV [ɲẽˀ.kã] CVʔ.CV [ãˀ.ɾi] Vʔ.CV

‘agouti’ ‘boy’ ‘knee’ ‘traira fish’ ‘to be tired’ ‘to say’

5  Formants means for F1, F2: [i] 280, 2023; [e] 501, 1663; [ʉ] 328, 1824; [a] 612, 1303; [u] 328, 1046; [o] 473, 999. 6  These syllable shapes are similar to the ones presented by Miller (1999:15). Kaye (1965:20) postulates the structure of Desano syllable as (C)V([ʔ] or [h]), where C is an “optional consonant followed by the syllabic nucleus plus an optional choice of either glottalization or aspiration”.

290

Silva

g. eha h. yʉkʉ i. uti

[e.ʰa] V.hV [yʉʰkʉ] CVh.CV [uʰti] Vh.CV

‘to arrive’ ‘tree’ ‘wasp’

Any consonant (both oral ones and their nasal allophones) can occupy the onset position of word-initial syllables, with the exception of the glottal stop ʔ and glottal fricative h. Word-internally both the glottal stop ʔ and the glottal fricative h can occupy the onset slot, like the other consonants, as in (3d, 3g). All syllables must have a vowel nucleus; sequences (C)V1V2 (where V1 and V2 are different vowels), as in (3a), are analyzed as two separate syllables (CV1.V2). Thus, these vowel sequences in Desano are generally considered to belonging to separate syllables and are not diphthongs. The reason for this is because each vowel in a (C)V1V2 sequence has a separate energy burst; that is, there are two separate intensity pulses, as shown in the spectrograms for the words [ŋãĩ] ‘parakeet’ and [nõã] ‘bone’ in Figure 11.2. a. [ ŋ

FIGURE 11.2

ã

ĩ ] ‘parakeet’ b.

[ n

õ

ã ] ‘bone’

Spectrograms for (C)V1V2 sequences.

Furthermore, when asked to say words with (C)V1V2 sequences slowly, speakers tend articulate these two vowels separately. Kaye (1965) also reports that, “when speaking slowly, a speaker of Desano will cut sequences of the type CVVCV into sequences of the type CV V CV (where a space represents a pause, i.e., a syllable boundary)” (pp. 36–37). While the vowels in (C)V1V2 sequences can be considered as belonging to different syllables, the sequences of identical vowels, i.e. (C)V1V1 (where V1

291

The status of the laryngeals ʔ and h in Desano

and V1 are the same vowels), are considered as belonging to the same syllable. The examples in Figure 11.3 show the spectrograms for the words ga [gaa] ‘hawk’ and ye [yee] ‘jaguar’. There is only one energy burst with one intensity pulse. a. [ g

FIGURE 11.3

a

a] ‘hawk’

b.

[y

e e] ‘jaguar’

Spectrograms for (C)V1V1 sequences.

2.2.1 Syllable Quantity and Weight In the discussion above, we saw that, in general, syllables in Desano tend to be ‘codaless’. The exception being the (C)V.CV laryngealized roots, in which ʔ and h can fulfill the coda position in the first syllable. Each vowel (rhyme) is assigned to one single unit of quantity (mora). Onsets are extra- moraic, and, thus, are directly linked to the syllable (cf. Hayes 1989). The laryngeal sounds ʔ and h have no effect on weight, as these exceptional closed syllables do not bear stress. Therefore, I analyze them as sharing the same mora with the preceding vowel. In (4) the various syllable shapes found in Desano are each associated with a mora. (4) Syllable shapes and moraic association σ σ σ σ σ µ

µ

V

C V

µ

µ

Vʔ

Vh

σ

µ

µ

C Vʔ

C Vh

Accordingly, root morphemes (plain roots and laryngealized roots as illustrated in Table 2.8) are analyzed as having a bimoraic structure.7 As mentioned above, plain root morphemes of the type (C)V1V2 are considered disyllabic, and thus have a bimoraic structure, as illustrated in (5). 7  Bimoraic structure has been analyzed for other Eastern Tukanoan languages, such as Tukano (Ramirez 1997:53–56), Barasana (Gomez-Imbert and Kenstowicz 2000:421), Tatuyo (GomezImbert 2005), and Wanano (Stenzel 2004).

292

Silva

(5) a. bui [bui] ‘agouti’ σ σ

b. oa [oa] ‘skunk’ σ σ

µ µ

µ µ

C V. V b u i

V. V o a

Many monosyllabic plain roots of (C)V shape are realized with a long vowel; however, in Desano, vowel duration is not phonemically distinctive. These monosyllabic plain roots are realized phonetically with long vowels and also interpreted as having bimoraic structure, as illustrated in (6).8 (6) a. ga [gaa] ‘hawk’ σ

b. ye [yee] ‘jaguar’ σ

µ µ

µ µ

C V V g a a

C V V y e e

Although the great majority of roots in Desano are bimoraic, there are a small number of monomoraic roots in the language. A small sample is shown in (7).9 (7) a. ~bo b. ~de c. ~bu

[mõ] [nẽ] [mũ]

‘piraiba fish’ ‘buriti fruit’ ‘umari fruit’

2.3 Suprasegmentals: Nasalization and Tone Nasalization and tone have been established as two suprasegmental features in ET languages. In this section, I briefly outline the main characteristics of these suprasegmental in Desano. The description of nasality and tone outlined here does not intend to be exhaustive; instead, I point out some basic

8  The bimoraic structure accounts for the assignment of stress and tone. Generally stress and tone falls on the second mora of a bimoraic morpheme. 9  These roots tend to be lengthened when they occur with no additional morphology, for example, [ga] may be realized as [gaa].

293

The status of the laryngeals ʔ and h in Desano

characteristics of these suprasegmentals in order to postulate, later, a third supra­segmental: laryngealization. 2.3.1 Nasalization It is generally accepted that, in the Eastern Tukanoan languages, nasalization is a property of the morpheme rather than a contrastive feature per se (Barnes, 1999).10 Thus all vowels and voiced consonants have nasal counterparts when they occur in nasalized morphemes. No voiceless consonants, with the exception of the fricative /s/, nasalize (Silva 2008). As mentioned, nasality is a suprasegmental feature of the morpheme. Morphemes are inherently marked as oral [– nasal], nasal [+ nasal] or are unmarked for nasality [∅ nasal]. The unmarked morpheme receives the [± nasal] feature from the morpheme to its left (progressive assimilation) or to its right (regressive assimilation) through spreading. The examples in (8) illustrate nasal harmony in Desano with cases of both progressive (8a) and regressive (8b) assimilation (data from Kaye 1971:38–39).11 (8) a. [+ N] σ

σ

b. [– N] σ

s e d a + d u pineapple-CLASSIFIER:round [sẽnãnũ] ‘pineapple’

σ

[+ N] σ

σ

σ

b a ʔ a + d i + d a ‘because (we) ate’ [baʔanĩnã]

2.3.2 Stress and Tone Stress usually falls on the second syllable of a disyllabic root morpheme, as shown in (9). Desano exhibits a two-level tone system: high and low. Verbal and nominal roots have one of the three tone patterns HH, HL, LL and a number of verbal affixes exhibit lexical high tone. High tone usually falls on the stressed syllable. As stress and tone spreading is being currently investigated, I do not mark tone in the transcriptions of this study. 10  Analyses of nasalization as a feature of morphemes can be found in Kaye (1970, 1971) for Desano; Bivin (1986) for Desano, Siriano, Tukano, Tuyuca, Kotiria (Guanano/Wanano); Barnes (1996) for Tuyuca; Gomez-Imbert (1997) for Barasano; Ramirez (1997) for Tukano; Miller (1999) for Desano; and Stenzel (2004, 2007) for Kotiria (Wanano). 11  For more details on nasal harmony in Desano and how the rules of spreading apply, see Kaye (1970, 1971) and Silva (2012).

294 (9) a. b. c. d.

Silva

[ba.á] [o.á] [go.bé] [gah.kí]

‘food’ ‘skunk’ ‘hole’ ‘monkey’

In (9) stress falls on the syllable that bears the H tone, which is usually the second syllable of the root morpheme. 3

The Laryngeal Segments in Desano

The distribution patterns of the laryngeal segments ʔ and h are discussed in this section, I also provide an overview of previous treatments of these segments. 3.1 The Glottal Fricative h This segment presents an interesting issue in Desano phonology. Kaye (1965) does not include h in the consonant inventory. He notes that the distribution of h contradicts the generalization of the (C)V syllable structure. According to Kaye, the occurrence of h can be explained by a generative rule that applies to morphemes. Thus, this “rule provides for each sequence VV́ to be rewritten VhV́ (where VV is a sequence of any two identical vowels and ´ indicates [+ accent])” (p. 47). Kaye (1970) includes h in the phonemic inventory. In both works this author treats the occurrence of h between two vowels, and h before voiceless consonants as two different phenomena. The former is a result of an accent placement rule and the later a result of pre-aspiration rule. According to Miller (1999), the glottal fricative h is a full consonant, which only occurs intervocalically and is followed by “the echo vowel that precedes [it]” (p. 12). I start by introducing the occurrence of the glottal fricative h that occurs before voiceless segments within the root. In the ET literature this has been called ‘pre-aspiration’. In root morphemes in Desano (as well as other ET languages in the Vaupés), there is a (pre-) aspiration that occurs systematically before voiceless segments /p, t, k, s/ in the onset of the second syllable (C)VCV roots, as illustrated in (10a) and (10b).12 This pre-aspiration does not occur before voiced consonants; thus aspiration is disallowed in a structure like (10c).

12  Although this pre-aspiration is a characteristic of ET languages, it has been reported to occur in at least one WT language: Siona (see Wheeler 1987:89; and Wheeler and Wheeler 1962:101).

295

The status of the laryngeals ʔ and h in Desano

(10) a. VʰC[– voice]V b. CVʰC[– voice]V c. *CVʰC[+ voice]V

ahsi-ɾi ‘be.hot-NOMINATIVE’ wehko ‘parrot’ *gohbe but gobe ‘hole’

Figure 11.4 shows the sound wave and spectrograms for the words [aʰsiɾi] ‘hot’ and [weʰko] ‘parrot’. The vowel and the aspiration are highlighted together as my claim is that, in Desano, the laryngeal feature ‘aspiration’ is realized after the leftmost (first) vowel of the root morpheme. This is not to say that aspiration is a property of the vowel. a. [ a

h

s

FIGURE 11.4

i

ɾ

i]

b. [ w

e h

k

o]

Sound wave and spectrograms for [aʰsiɾi] ‘hot’ and [weʰko] ‘parrot’.

The vowels in the initial Vʰ and CVʰ syllables of VʰCV and CVʰCV roots (Figure 11.4) are as long as (i.e., have about the same timing) the vowels in initial V and CV syllables of VCV and CVCV roots, illustrated in Figure 11.5 for the words [gobɛ] ‘hole’ and [yeba] ‘ground’. a. [ g

FIGURE 11.5

o

b

ɛ]

b.

[ y  e    b

Sound waves and spectrograms for [gobɛ] ‘hole’ and [ yeba] ‘ground’.

a]

296

Silva

Note that the timing of the leftmost vowels of CVCV roots (Figure 11.5) is about the same as (i.e. as long as) the leftmost vowels of VʰCV and CVʰV roots (Figure 11.4). It should also be noted that, for some speakers, this aspiration that occurs before voiceless segments is often realized as a voiceless vowel. Thus, examples in (10a) and (10b) can also be realized as [aḁsiɾi] and [wee̥kó] respectively, showing that [h] is virtually indistinguishable from a voiceless vowel in that environment. Furthermore, this aspiration does not occur across morpheme boundaries. Thus, in that context, aspiration has a predictable realization: it occurs before voiceless consonants within the root morpheme. Moving on to the cases where h is realized between two vowels: Besides its predictable occurrence before voiceless consonants within the root morpheme, the glottal fricative h also occurs between vowels, as illustrated by the forms VhV and CVhV listed in (2) above, and repeated below in (11). (11) a. VhV roots oho ‘banana’ uhu ‘pacu fish’ eha ‘to arrive’ b. CVhV roots weha ‘paddle’ buha ‘to cross’ behe ‘to classify’ The forms in (11b), with the CVhV shape also have an alternative realization, in which the glottal fricative is not realized; instead, a voiceless counterpart of the proceeding vowel is realized, as illustrated in (12). (12) a. [ɲãhãɾĩke] ~ [ɲãḁ̃ɾike] b. [yʉhʉgʉ] ~ [yʉʉ̥ gʉ] c. [mẽhẽõ] ~ [mẽẽ̥õ]

‘enterǃ’ ‘one man’ ‘mommy’

The glottal fricative h, in (12), occurs between two identical vowels. I argue that the second vowel is an echo of the first one (i.e. CV1hV1). This inter­ pretation helps to explain the instances in which the glottal fricative occurs between two different vowels (i.e. CV1hV2), as in (13) below. In these cases, the glottal fricative may be realized as a voiceless vowel that is identical to the vowel it precedes (as in (12)); or alternatively be followed by the first vowel (i.e. it echoes the first vowel).

The status of the laryngeals ʔ and h in Desano

(13) a. b. c. d.

[yuhi] ~ [yuu̥ i] ~ [yuhui] [mãhı̃]́ ~ [mãḁ̃ı]̃́ ~ [mãhãĩ] [mʉ̃ hĩ] ~ [mʉ̃ ʉ̥̃ i] ~ [mʉ̃ hʉ̃ ĩ] [gahi] ~ [gaḁi] ~ [gahai]

297

‘support’ ‘to turn around’ ‘caranã leaf’ ‘another’

As I have observed, the glottal fricative only occurs within the root. There are only a few instances in which [h] occurs in suffix-initial position, and in these cases, [h] is an allophone of another segment, as illustrated in (14). Thus, although the occurrence of [h] in suffix-initial position can be explained as allophonic variation of other phonemes, the [h] in root-medial position is not an allophone. (14) a. [-ka] ~ [-ha] b. [-ɾã] ~ [-hã]

‘DUBIDATIVE’ ‘PLURAL.ANIMATE’

There are at least three interesting things raised by the data above illustrating the occurrence of the glottal fricative h. First, the glottal fricative never occurs in the initial position of a word—though there are a few cases with VCV roots, where C is /h/, the first vowel is frequently deleted, and then /h/ is realized in the beginning of the word. For example, /óhó/ → [hó] ‘banana’; /úhú/ → [hú] ‘pacú fish’; /eho/ → [eheó] ~ [heó] ‘to feed’).13 Second, CVCV roots, where the second C is /h/, also have an alternative realization: the glottal fricative is not realized; instead a voiceless counterpart of the preceding vowel is realized, as in (12) and (13). The fact that this (pre-) aspiration occurs only before voiceless consonants in Desano seems to follow a pattern observed across languages. Typologically, pre-aspiration has been characterized as a process that occurs before a subset of consonant segments, specifically voiceless stops (cf. Helgason 2002). I move now to the description of the occurrence of the glottal stop ʔ. 3.2 The Glottal Stop ʔ In previous descriptions, the glottal stop that occurs in Desano has been described both as a phonetic ‘glottalized vowel’ and as a full consonant segment. Below, I provide an overview of these previous, followed by the alternative analysis, suggested by this current study. 13  We noticed during the orthography development workshops that some Desano speakers tend to write these words without the initial vowel, even though they pronounce the initial vowel in careful speech. Another note on orthography is that even though preaspiration is not contrastive, speakers decide to mark it with the letter h in the written form of the language.

298

Silva

3.2.1 The Glottal as ‘Glottalized Vowel’ Kaye (1965) describes the occurrence of the glottal stop within the root as a property of the preceding vowel. According to Kaye, the glottal stop only occur between two similar vowels, i.e., in (C)VCV roots, where the second vowel is an echo of the first. Kaye illustrated this with the examples reproduced here in (15). echo vowel (15) a. dʉ’ʉ [dʉ́ ʔʉ́ ] ‘chipmunk’

echo vowel b. ~kú’dí [kũʔũnĩ] ‘to bite’

Thus, each plain vowel would have a glottalized counterpart (iˀ, eˀ, aˀ, oˀ, uˀ) in the inventory. Kaye (1970) extends his previous analysis and characterizes glottalization as a property of the syllable; i.e., the vowel of the first syllable is glottalized. He refers to ‘glottalized syllables’, which ‘may not be phonemic’, because of its restrictive distribution—only the first syllable of a word (or root morpheme) may be glottalized. He treats glottalization as a contrastive feature of the vowel in sequences of two identical vowels V1_V1 or a vowel before a consonant V_C. However, in both studies, Kaye does not account for root morphemes that have two different vowels V1_V2 as illustrated in (16); in fact, he does not present any data of this type in either of his works. (16) a. b. c. d.

[biʔa] [waʔi] [gʉʔa] [guʔi]

‘to close’ ‘fish’ ‘we’ (exclusive) ‘turtle’

3.2.2 The Glottal Stop as a Consonant Segment The glottal stop has been treated as a consonant segment in most of the ET languages (cf. González de Pérez 2000). Miller (1999), treats these two segments as full consonant phonemes on the basis of comparing minimal pairs of contrasting the glottal stop /ʔ/ with the glottal fricative /h/, as in (17). (17) a. [oʔo] ‘to want’ b. [oho] ‘banana’ Miller (1999) also mentions that the glottal stop can only occur in intervocalic position. Thus, when the glottal stop is immediately followed by another consonant, an echo vowel occurs after the glottal stop, as in (18). (18) a. /pʉʔɾʉ/ [pʉʔᶶɾʉ] ‘after’ b. /oʔake/ [oʔoake] ‘sweep’

299

The status of the laryngeals ʔ and h in Desano

Phonetically, the glottal stop and the glottal fricative are unusual when compared to the other consonants in the language, because they are the only segments that cause the shortening of the preceding vowel. Figure (11.6) shows the sound waves and spectrograms of the root morphemes [koɾe] ‘before’ and [koʔɾe] ‘to wait’, respectively. The timing of the first vowel in the CV sequence in Figure (11.6a) is as long as the sequence CVʔ in Figure (11.6b). a. [ k

o

FIGURE 11.6

ɾ

e ]

b. [ k o ʔ

ɾ e ]

Sound waves and spectrograms for [koɾe] ‘before’ and [koʔɾe] ‘to wait’.

This phonetic fact triggered only by the glottal stop (and the glottal fricative as shown in Figure 11.4) might be an indication that glottalization is a prosodic feature occurring in the first vowel of the root morpheme, rather than a full consonant segment. 3.3 Similarities between the Occurrence of h and ʔ There are some striking similarities regarding the occurrence of the glottals ʔ and h. The distribution of these segments is very similar. Both segments only occur after the first vowel within the (C)VV roots, as in (19). These glottal segments do not occur in affixes. (19) a. (C)VV [oho] [duha]

‘banana’ ‘to stay’

b. (C)VV [oʔo] [duʔa]

‘to give’ ‘to yank’

300

Silva

Although these seem to be compelling examples of h and ʔ as contrastive segments, they are not considered full consonants segments, but the realization of the laryngeal feature in the root morpheme (discussed in Section 4 below). These glottal sounds also occur before consonants in (C)VCV roots, as in (20). However, in this case, the occurrence of the glottal fricative h is predictable: it always occurs before a voiceless segment. (20) a. (C)VhCV [mʉ̃ htã] [dehko] b. (C)VʔCV [mʉ̃ ʔtã] [deʔko]

‘to break’ ‘water’ ‘to go ahead’ ‘half’

While the glottal fricative can only appear before voiceless consonants, the glottal stop can appear either before voiceless consonants, as in (20b), or before voiced ones, as in (21). (21) a. [mõʔmẽ] b. [õʔmã]

‘to work’ ‘to take away’

Based on the data presented above illustrating the occurrence of the glottal stop (and the glottal fricative), I note the following characteristics regarding its distribution: (i) the glottal stop only occurs in medial position within the root morpheme; (ii) although Desano syllables are ‘codaless’, the glottal stop and the glottal fricative are the only segments that occur in coda position. In such contexts, it is always the first syllable in the root morpheme that might have a coda. These observations lead to an alternative analysis, treating ‘glottalization’ as a prosodic phenomena in Desano. 4

The Status of the Glottal Segments h and ʔ in Desano: An Alternative Analysis

In the previous sections we saw that the glottal segments h and ʔ have similarities regarding their occurrence within the root morpheme. They also have restricted distribution when compared to the other consonant segments of the language. In this section, I present an alternative analysis treating the glottal segments as a prosodic feature of the root.

The status of the laryngeals ʔ and h in Desano

301

There have been a few analyses proposing glottalization as a suprasegmental feature in ET languages. For example, Malone (1987) claims that glottalization in ET languages developed from the suprasegmental in the proto-language. Ramirez (1997) analyzes the glottal stops in Tukano as the realization of a ‘laryngealized’ tone. The most recent analysis is Stenzel (2007) for Wanano. Her analysis is similar to Ramirez’s insofar as she postulates a third suprasegmental feature (besides nasalization and tone), called ‘glottalization’. Glottalization as a prosodic feature is also found in Mixtec. Macaulay and Salmons (1995) state that laryngealization in Mixtec is a feature of roots. An alternative analysis to account for the laryngealization in Desano is based on Macaulay and Salmons’ analysis for Mixtec, considering language specific differences. For example, while laryngealization in Mixtec is marked by the feature [+ constricted glottis], which is realized as ʔ; in Desano roots, laryngealization is marked by the features [+ constricted glottis] and [+ spread glottis] which are realized either as [ʔ] or [h], respectively. Following this approach, the laryngeal roots in Desano presented in (2) are optionally marked in the lexicon for the floating glottal features (either [+ constricted glottis] or [+ spread glottis]), which then attaches to the leftmost vowel, as illustrated in (22a) and (22b). (22) a. /dua/ → [duha] ‘to stay’ [+ spread glottis] b. /dua/ → [duʔa] ‘to yank’ [+ constricted glottis] Plainly stated, according to this analysis, Desano roots are optionally marked in the lexicon for one of the laryngeal features [+ spread glottis] or [+ constricted glottis], which occurs after the first vowel of the root. In this approach, the target for the laryngeal features is the mora. Thus, [constricted glottis] or [spread glottis] associates to the right edge of the first mora. This way, the feature will always be displayed in intermoraic position, as a glottal stop or a glottal fricative after the first vowel of the root morpheme; and can be associated with the preceding mora. Then, in CV.V roots the feature will surface as the onset segment of the second syllable, as shown in (23a) for duha ‘to stay’ and (23b) for duʔa ‘to yank’.

302

Silva

(23) a. σ µ

σ

σ

µ →

C VhV d u a b. σ µ

σ

µ

σ

µ →

µ

C VhV d u a

σ

σ

µ →

C VʔV d u a

µ

C V h V d u a

σ

µ

σ

σ

µ →

σ

µ

C VʔV d u a

[duha] ‘to stay’

µ

C V ʔ V d u a

[duʔa] ‘to yank’

In (C)V.CV roots (i.e., in roots which the syllable following the glottal feature already has an onset consonant), the feature is not linked to the syllable node, although it remains linked to the mora of the preceding vowel. The adjunction to the preceding mora maintains the canonical (C)V syllable structure (and its ‘codaless’ characteristic), as shown in (24). (24) a. σ µ

σ µ

σ →

C VhC V g a k i b. σ µ

C VʔC V p o g a

µ

C VhC V g a k i

σ µ

µ

σ

σ →

µ

[gahki] ‘monkey’

σ µ

C VʔC V p o g a

[poʔga] ‘manioc flour’

This analysis accounts for the timing of Vh and Vʔ sequences, as well as for the echo vowels in roots with (C)V1(C)V1, where the second vowel is the echo of the first; i.e., the echo- vowel is realized after the glottal segments is attached to the underlying vowel. This analysis can also explain the pattern illustrated in (25) below, when glottalized roots are reduced. In Desano, disyllabic glottalized

The status of the laryngeals ʔ and h in Desano

303

(C)V.V roots like ~bʉ’ʉ [mʉ̃ ˀʉ̃ ] ‘you’ become monosyllabic when suffixed, i.e., the second syllable of the root is deleted; however, the remaining syllable retains the glottal feature, [mʉ̃ ʔʉ̃ ] → [mʉ̃ ʔ-]. (24) Root reduction and association of the glottal gesture σ σ σ σ σ µ = µ

µ

µ

µ

m ʉ̃ ʔ ʉ̃ + ɾ e → m ʉ̃ ʔ ɾ� ẽ you (sg.) + OBJ ‘to you’

In (25), when the suffix -ɾe is attached to the root, the root is reduced by deleting its second syllable; however, the [+ constricted glottis] feature that is manifested in the root morpheme remains because it is associated with the mora of the first vowel in the first syllable of the root. 5

Diachronic Analysis of h and ʔ in ET languages

In this section, I present a brief overview of how these glottal segments have been analyzed diachronically. There are different hypotheses regarding the glottal segments in ET languages. In general, a proto segment *h has been postulated in the proto-language, which has been deleted in the beginning of a word but retained within the word (cf. Waltz and Wheleer 1972, Barnes 1980, Wheeler 1992, Ardila 1998, Chacon 2014).14 The glottal stop has also been reconstructed as segment in the proto-­ language (cf. Waltz and Wheeler 1972, Wheeler 1992); however, Barnes (1999), following Malone (1987), does not include the glottal stop in the inventory of proto-consonants. According to Ardila (1998), the glottal stop in Desano only occurs between identical vowels, and for this reason it does not have a phonological status. However, we have seen in the examples listed in (7) and (8) that the glottal stop also occurs between different vowels and between the first vowel, and before the consonant of the second syllable in (C)VCV roots. 14  Another hypothesis, pointed out by an anonymous reviewer, is that “the exceptional behavior of glottal h (echo, metathesis) could be an areal, not a genealogical feature, shared by languages of two families,” Tukanoan (Tukano) and Arawakan (Tariana and Baniwa).

304

Silva

Malone (1987) offers a diachronic analysis to explain the glottal stop occurring in intervocalic position of the first syllable within root-morphemes in ET languages. His analysis postulates a proto *ʔ that surfaces in morphemes of the CVV root type, this way, the preferred CV structure is maintained in the second syllable. Chacon (forthcoming) also reconstructs a proto *ʔ. He observes that the ‘true glottal stop’ of the proto language occurs in (C)VCV root-shape, where the second C is the glottal stop (note that it is similar to the root-shape proposed by Malone [1987]). According to Gomez-Imbert (2011), the glottal stop that occurs in intervocalic position is a ‘default’ consonant that is empenthesized in (C)VV roots to maintain the canonical CV syllable structure of ET languages. Thus, a (C)VV root becomes (C)VCV, where the second C is a glottal stop. There are some asymmetries if we treat the glottal stop as a default consonant in order to keep a CV structure. First, only root morphemes are targets (the glottal stop does not occur in suffixes, even in the ones that have a CVV structure). Second, there are many root-morphemes that have a CVV root shape, but do not have a glottal stop in order to break the vowel sequence. Furthermore, affix morphemes composed of a single V occur frequently in Desano (and in ET languages in general) and are homophonous in different paradigms. For example, in Desano there are two -a morphemes. In the verbal paradigm -a marks non-present, and in the nominal paradigm, -a marks animate plural. These single -V morphemes (and many others) are not preceded by glottal stops as would be expected in order to reconstruct and maintain a CV structure. 6 Conclusions In this study, I present a description of the occurrence of the glottal segments h and ʔ in Desano, and describe how they differ from the other consonant segments. I propose an alternative analysis, which threats h and ʔ as a prosodic realization of a suprasegmental laryngeal feature that occurs after the first vowel within the root. This study shows that there can only be a single instance of aspiration or glottalization per root. Furthermore, this analysis differs significantly from previous accounts that treated h and ʔ as a full consonant segment (Miller 1999), or a property of the (first) syllable (Kaye 1970). The alternative approach proposed seems more appropriate than analyses that treat the glottal stop as a full contrastive consonant segment for four reasons: (i) it explains the distribution restrictions of the glottal stop and glottal fricative when compared to the other consonants in the language: they

The status of the laryngeals ʔ and h in Desano

305

only occur after the first vowel within the root morphemes; (ii) it explains the participation of h and ʔ in syllable structure: they can only be an onset rootinternally and it is the only segment allowed in coda position; (iii) it provides an underlying (C)V syllable structure, as well as a basic (C)V(C)V root template which is valid for Desano and other languages within the family; and (iv) it provides an explanation for the root-morphemes contrast. The contrast is analyzed as a suprasegment affecting specific roots (laryngeal roots). Finally, diachronically, these segments cannot be easily reconstructed as a full consonant segment in the protolanguage. The analysis presented here is surely not the only possible explanation (see for example, Kaye 1965, 1970 and Miller 1999 for alternative accounts). It is, however, my hope that this analysis provides a foundation on which future phonetic and phonological investigations may be based. References Ardila, Olga. 1998. Apectos Fonológicos de las Lenguas Tucano-Orientales: Una Visión Comparativa. In Forma y Función, v. 11, Santafé de Bogotá: Universidad Nacional de Colombia, pp. 41–54. Barnes, Janet. 1980. La Reconstrucción de Algunas Formas del Proto Tukano-BarasanoTuyuca. In Artículos en Lingüística y Campos Afines, 8, pp. 37–66. ———. 1996. Autosegments with three-way lexical contrasts in Tuyuca. In International Journal of American Linguistics 62.31–58. ———. 2006. The Tucanoan Languages. In Encyclopedia of Language & Linguistics, Second Edition, volume 13, ed. by Keith Brown, (ed.) pp. 130–142. Oxford: Elsevier. ———. 1999. Tucano. In The Amazonian Languages, ed. R.M.W. Dixon and A.Y. Aikhenvald. Cambridge: Cambridge University Press, pp. 207–226. Bivin, William E. 1986. The Nasal Harmonies of Twelve South American Languages. M.A. thesis, The University of Texas at Arlington. Blevins, Juliette. 1995. The Syllable in Phonological Theory. In The Handbook of Phonological Theory, ed. John A. Goldsmith. Blackwell, pp. 206–244. Chacón, Thiago. 2014. A revised proposal of Proto-Tukanoan consonants and Tukanoan family classification. In International Journal of American Linguistics, forthcoming. Gomez-Imbert, Elsa. 1997. Morphologie et Phonologie Barasana: Approche Non Linéaire. Ph.D. dissertation, Université Paris 8. Saint-Denis. ———. 2005. Fonología de dos idiomas tucano del Pirá-Parana: Barasana y Tatuyo. In Amerindia 29/30.43–80.

306

Silva

———. 2007. Tukanoan Nominal Classification: The Tatuyo System. In Indigenous Language of Latin America 5: Language Endangerment and Endangered Languages: Linguistic and Anthropological Studies with Special Emphasis on the Languages and Cultures of the Andean-Amazonian Border Area, ed. Leo Wetzels. Leiden: CNWS Publications, pp. 401–428. ———. 2011. Famille Tukano. In Dictionnaire des langues du monde. Paris: Presses Universitaires de France. Gomez-Imbert, Elsa, and Michael Kenstowicz. 2000. Barasana Tone and Accent. In International Journal of American Linguistics 66.419–63. González de Pérez, M.S. and M.L. Rodríguez de Montes (Eds.). 2000. Lenguas indígenas de Colombia: una visión descriptiva. Santafé de Bogotá: Instituto Caro y Cuervo. Hayes, Bruce. 2009. Introducing Phonology. Wiley-Blackwell. Helgason, Pétur. 2002. Preaspiration in the Nordic Languages: synchronic and diachronic aspects. Ph.D. dissertation, Stockholm University. Kaye, J. 1965. Two conceptions of Desano phonology. M.A. thesis, Columbia University, New York. ———. 1970. The Desano Verb: problems in semantics, syntax and phonology. Ph.D. dissertation, Columbia University, New York. ———. 1971. Nasal harmony in Desano. In Linguistic Inquiry 2.37–56. Kenstowicz, M. 1994. Phonology in Generative grammar. Cambridge, MA: Blackwell. Lindblom, Björn. 1988. Phonetic invariance and the adaptive nature of speech. In, Working Models of Human Perception, Elsendoorn, Bouma. Academic Press, London pp. 139–173. Macaulay, Monica and Joe Salmons. 1995. The Phonology of Glottalization in Mixtec. In International Journal of American Linguistics, Vol. 61, no. 1, pp. 38–61. Malone, T. 1987. Proto-Tucanoan and Tucanoan genetic relationship. Ms. Instituto Linguistico de Verano, Colombia. Miller, Marion. 1976. Fonología del desano. Sistemas fonológicos de idiomas colombianos, Tomo III. Instituto Lingüístico de Verano. Bogota. pp. 105–11. ———. 1999. Desano Grammar—Studies in the Languages of Colombia 2. The Summer Institute of Linguistics, University of Texas at Austin. Ramirez, H. 1997. A Fala Tukano dos Ye’pâ-Masa, Tomo I: Gramática. Manaus: CEDEM. Silva, Wilson. 2008. Acoustic analysis of voiceless obstruents and nasal harmony in Desano. In Amerindia, special edition, La structure des langues amazoniennes, pp. 301–19. ———. 2012. A Descriptive Grammar of Desano. Ph.D. dissertation, University of Utah. Stenzel, Kristine. 2004. A Reference Grammar of Wanano. Ph.D. dissertation, University of Colorado. ———. 2007. Glottalization and other suprasegmental features in Wanano. In International Journal of American Linguistics, vol. 73, no. 3, pp. 331–66.

The status of the laryngeals ʔ and h in Desano

307

Waltz, Nathan E. and Alva Wheeler. 1972. Proto Tucanoan. In Comparative Studies in Amerindian Languages, edited by Esther Matteson et. al. Paris: Mouton, pp. 119–149. Wheeler, Alva. 1987. Gantëya Baín: El Pueblo Siona del Rio Putumayo, Colombia. Tomo I: etnología, gramática, textos; Tomo II: diccionario. Santafé de Bogotá: Colombia, Instituto Linguistico de Veraño. ———. 1992. Comparaciones Lingüísticas en el Grupo Tucano Occidental. In Estudios Comparativos Proto Tucano, ed. J. Barnes, A. Wheeler and M. Wheeler. Santafé de Bogotá, Colombia: Editorial Alberto Lleras Camargo, pp. 17–53. Wheeler, Alva and Margaret Wheeler. 1962. Siona Phonemics (Western Tucanoan). In Ecuadorian Indian Languages: I, ed. Benjamin Elson. Linguistics Series 7, Summer Institute of Linguistics, University of Oklahoma, pp. 96–111.

CHAPTER 12

Temporal Coordination of Glottalic Gestures in Karitiana Didier Demolin and Luciana Storto 1 Introduction Karitiana, an endangered language from the Arikém branch (Tupi family) spoken in the state of Rondonia in Brazil, shows interesting phenomena concerning glottalic consonants. Indeed, as several other languages of this linguistic family, Karitiana has no clear glottal stop in its phonological inventory, even though glottal stops do exist phonetically. Glottal stops in Karitiana are predictable in the onset of stressed syllables (Storto 1999). In addition, the language presents a number of interesting phenomena related to the phonetic realization of glottal stops and to the temporal realization of glottalic gestures, as do other Tupi languages. A frequent, but not systematic, phenomenon is the occurrence of vowels with a final burst in CVC words ending with a final unreleased voiceless stop (Figures 12.1a & 12.1b). The phenomenon also appears word internally (Figure 12.5). In such instances, it is often the case that vowels which are ending with bursts precede most of the allophones of the Karitiana nasals, all of which are voiced. In order to understand the phenomenon, there are three main questions to be addressed: (1) Why is there a burst at the end of vowels before stops and nasals? (2) How can we describe this phenomenon precisely? (3) Does it reflect anything particular about the phonological system of the language? To help answer these questions, three speakers participated in experiments in which acoustic and EGG data were recorded simultaneously. This was motivated by some observations made in the field (Storto & Demolin 2002) that led us to hypothesize that voiceless stops might involve a closure of the glottis simultaneous with the oral closure necessary to produce these consonants. One important thing to note from the start is that data of this paper come * We wish to thank Luiz Karitiana, Nelson Karitiana, Inácio Karitiana, Marcelo Karitiana and João Karitiana for their participation in the experiments. We thank, as well, Angelique Amelot, John Kingston, Adrian Fourcin, Evelyn Abberton and James Kirby for helpful suggestions. Data were collected during several fieldwork trips in Brazil.

© koninklijke brill nv, leiden, ���6 | doi ��.��63/9789004303218_013

309

Temporal coordination of glottalic gestures in Karitiana

EGG

Frequency (kHz) Audio

from a small set and are limited in amount. However, the occurrence of bursts at the end of vowels observed for every subject who participated in the study makes the description of this phenomenon potentially important for future investigation. The paper has therefore to be considered as a purely descriptive contribution to a phenomenon that should receive a more detailed and quantitative treatment in the future. This is particularly true because recent observations made by the authors in Dâw and Pirahã suggest that the phenomenon presented here might not be specific to Karitiana. This could influence the phonological pattern of these languages in various ways as we will discuss in the final part of the paper.

0.5 0 ‒0.5 8 6 4 2 0 0,5 0 ‒0,5 ‒1

7.2

7.25

7.3

7.35

7.4

7.45

7.5

7.55

7.6

7.65

EGG

Frequency (kHz) Audio

FIGURE 12.1A Audio waveform, wide band spectrogram and EGG signal of the word [mbap̚ ] ‘lame’. Arrows shows the burst at the end of the vowel on the audio waveform and the interruption of the EGG signal.

0.6 0.4 0.2 0 ‒0.2 ‒0.4 8 6 4 2 0 0,5 0 ‒0.5 ‒1

5.9

6

6.1

6.2

6.3

6.4

6.5

FIGURE 12.1B Audio waveform, wide band spectrogram and EGG signal of the word [hobmã] ‘drowned’. The arrow shows the burst at the end of the vowel on the audio waveform.

310 2

Demolin and Storto

Material and Method

2.1 Method Data were obtained in several recording sessions. First, data were collected with three male speakers who were recorded with acoustic and electroglottographic EGG signals simultaneously. The second session involved three speakers recorded with acoustic and EGG signals, oral airflow (Oaf) and intraoral pressure (Ps) for bilabial consonants. Finally, data were obtained with six speakers with the acoustic signal during a session which was designed to obtain recordings from vowels in different contexts. The EGG system was a Fourcin Laryngograph. Acoustic data were obtained with a Shure head microphone. This allows recording the sound signal with a constant distance from the lips of the speakers. The aerodynamic data were obtained with the MacQuirer system designed by Scicon.

FIGURE 12.2 Electrode position on one side of the thyroid cartillage (Laryngograph system). Electrodes are posited on each side of the thyroid cartillage at the vocal folds level. A weak induced electric flow emitted through the electrodes measures the glottal resistance.

2.2 Material The relevant words were recorded in isolation and in short carrier sentences. Each word or sentence was repeated three times. Table 12.1 gives a sample of the words containing post-vocalic bursts and phonetic occurrences of glottal stops which have been observed in the recordings of the first experiment. TABLE 12.1 Data recorded showing some of the words where post-vocalic bursts and glottal stops were observed. Numbers indicate the occurrence of bursts at the end of vowels in the data

[gep ]̚  (2) [nam̚] (3) [apĩmbik ̚] (2) [kabm̚] (2) [opĩ] (2) [mbap ]̚  (3) [ãmãŋ] (1) [põm̚] (1) [opo] (3)

‘flea’ ‘rotten’ ‘to stumble’ ‘now’ ‘to cut’ ‘lame’ ‘to plant’ ‘bird species’ ‘penis’

[hobmã] (2) [opĩm̚] (1) [õmbi] (1) [ʔõm̚] (1) [ʔop ]̚  (1) [ʔotʔ̚ o:p ]̚  [ʔotʔ̚ ot] [ʔobmã] [ʔõʔĩ sokõʔĩ]

‘drowned’ ‘spinal cord’ ‘basket’ ‘shadow’ ‘hole’ ‘bamboo’ ‘species of egret’ ‘pierced’ ‘not to tie up’

Temporal coordination of glottalic gestures in Karitiana

311

2.3 Electroglottography In order to clarify the interpretation of the data, a short word about electroglottography seems necessary. Most data analyzed for the paper, that is, those data elicited with three speakers, as mentioned above, show synchronized recordings of the audio waveform and of the EGG signal. The essential difference between both is that the EGG signal measures the impedance at the glottis, i.e. the contact between the vocal folds, while the acoustic recording (audio waveform) shows short time pressure variations produced by the opening and closure of the glottis (Figure 12.3). The important thing to bear in mind is that the EGG technique essentially transduces the vocal fold contact area. Accordingly, we will not refer to ‘closing’ or ‘opening’ but to ‘contacting’ and ‘decontacting’ phases of the vibratory cycle. Therefore the EGG is sensitive to the covert changes of the contact area during glottal closure, and it also reflects the growth and loss of contact area along the length of the vocal folds as glottal closure and opening succeed each other (Baken & Orlikoff 2000). During the closed phase of the vocal fold oscillation the electrical pathway is optimized by the contact of the vocal folds. There are, however, degrees of contact. Over the course of the ‘closed’ phase of each 2.450

2.465

2.470

2.475

2.480

2.485

2.490

2.495

2.500

Audio waveform

Maximum vocal fold contact area

Minimum vocal fold contact area FIGURE 12.3

One glottal cycle (T)

EGG

Audio waveform and electroglottographic (EGG) displays of the recording of a vowel, illustrating the difference between both signals. EGG shows maximum vocal fold contact area when the signal reaches a peak and the minimum when it is low as indicated by the arrows. The comparison of both signals is given for one glottal cycle (T) as shown by the double arrows on both signals.

312

Demolin and Storto

glottal cycle, the contact of the vocal folds varies from minimal (valley represented in the signal) to maximal (represented by a peak in the signal) as contact involves more of the vertical dimension of the vocal folds (see Hayward 2001 for more details).

EGG

Frequency (kHz)

Audio

2.4 EGG and Glottal Stops in Karitiana The first set of analyzed data concerns the realization of glottal stops in various positions in words. This will allow us to differentiate the acoustic and EGG signals of glottal stops from those of other phenomena described later in the paper. Figure 12.4 shows the spectrogram and audio waveform of the words [ʔõʔĩ sokõʔĩ] ‘not to tie up’. Figure 12.4 displays the EGG signal and the audio waveform of the same words. In order to make the discussion about the interpretation of the EGG recordings clear it is worth examining these data in detail. 0.6 0.4 0.2 0 ‒0.2 ‒0.4 8 6 4 2 0.5 1 0,5 0 ‒0.5 ‒1 ‒1.5

2.6

FIGURE 12.4

2.8

3

3.2

3.4

3.6

3.8

4

4.2

4.4

4.6

Audio waveform, wide band spectrogram and EGG signal of the clause [ʔõʔĩ sokõʔĩ] ‘not to tie up’. The EGG shows that glottal stops start by a sharp increase in the signal. Then there is a gradual decrease of the signal until a fall occurs before the following vowel. The second and third [ʔ] show a couple of abrupt opening and closure after the preceding vowel. The lowering of the EGG signal before [s] and during [k] indicate the loss of contact between the vocal folds.

Glottal stops are generally realized by a beginning that is either abrupt or marked by a few glottal pulses. They end by an abrupt start on the first vibratory cycle of the following vowel. This can be seen on the spectrogram and audio waveform of Figure 12.4 for each vowel coming after a glottal stop. The EGG

Temporal coordination of glottalic gestures in Karitiana

313

show that glottal stops start by an abrupt increase in the signal. This is due to the vocal folds contact giving little resistance to the electric flow. Then, there is a gradual decrease of the signal because there is little change in the resistance during the glottal closure. The second and third glottal stops show a couple of changes in contact and loss of contact and closure at the end of the preceding vowel. The other important thing to note on the EGG signal is the transition between a glottal stop and the following vowel. Since there is no more contact between the vocal folds, there is a drop in the signal at this moment. This is different from the transition between the voiceless consonants [s] and [k] where it can be seen that there is an increase in the EGG signal due to the adduction of the vocal folds for the following vowel. When the glottis opens after a vowel before a voiceless consonant there is a gradual decrease in the signal because there is no more contact between the vocal folds. 3

Vowels with a Closing Burst

The phenomenon has been observed in three different contexts: (1) within words in VC sequences where C is a voiceless stop; (2) word finally before a final unreleased voiceless stop and (3) before nasal consonants. The contexts are presented separately in the following sections. 3.1 A Glottal Burst at the End of a Vowel before a Voiceless Stop Two examples of vowels ending with a burst are given in Figure 12.1a and 12.1b (representing the single vowel of the word mbap̚ and hobmã) and in Figure 12.5 (the first vowel of the word ʔapĩmbik̚ ). Both show that the last vowel’s cycle is realized by a clear burst on the audio waveform. Figure 12.1a shows the waveform and spectrogram of a word ending with an unreleased voiceless bilabial stop while Figures 12.5 gives the audio waveform, the spectrogram and the EGG signal of an intervocalic voiceless bilabial stop. The EGG signal of Figure 12.5 suggests that the vowel with a burst and the following voiceless stop are produced with an adduced glottis. The initial glottal stop, observed phonetically in this instance of the word [ʔapĩmbik ̚] ‘to stumble’ shows an abrupt increase on the EGG followed by a gradual decrease of the signal. This is due to fact that there is a resistance at the glottis, as the vocal folds are in contact. The gradual decrease in the signal is the consequence of a change in the vocal fold contact area, as there is less contact in the lower part of the vocal folds. The final part of the glottal stop displays a fall in the EGG due to the opening of the glottis.

314

Demolin and Storto

EGG Frequency (kHz) Audio

The vowel [a] shows a burst at the end in the waveform. The EGG signal shows an increase towards the end of the vowel, followed by a reduction of the signal’s amplitude and leveling of the signal. This reduction in amplitude is due to the fact that the bilabial closure for [p] is accomplished while the focal folds are still vibrating. The EGG signal suggests that the intervocalic [p] is realized with an adduced glottis, i.e. with contact between the vocal folds. The EGG signal during the intervocalic stop [p] starts by a small diminution. This is followed by an increase which then stays almost flat before a final diminution. This suggests that there is little change in the impedance at the glottis. The strong burst on the audio waveform at the end of [p] is likely due to the fact that the larynx moves from a low position for [a] to a higher position for [i] which increases the air pressure in the sealed volume between the glottis and the lips. This results in a strong burst similar to ejective consonants. The final consonant, the unreleased final voiceless velar stop [k ]̚  shows a loss of contact at the beginning when the EGG signal goes down. This is immediately followed by an increase and a flattening of the EGG signal. This suggests that there is a gradual increase in contact between the vocal folds. The contact between the vocal folds in the final part of the unreleased stop [k ]̚  prevents air pressure to rise behind the velar closure. This likely explains why there is no burst at end of this consonant.

0.5 0 ‒0.5 6 4 2 0 0,5 0 ‒0,5 ‒1

1.7

FIGURE 12.5

1.8

1.9

2

2.1

2.2

2.3

2.4

2.5

2.6

2.7

Audio waveform, wide band spectrogram and EGG signal of the word [ʔapĩmbik̚] to ‘stumble’. Arrows on the audio waveform indicate the vowel [a] ending with a burst and the initial burst of the following [p]. The final unreleased voiceless velar stop is identified by the F2/F3 approximation as indicated by the double arrow. Arrows on the EGG show the signal during: the initial glottal stop [ʔ], the intervocalic voiceless bilabial stop [p] following the vowel with the final burst and the final voiceless velar stop [k̚].

315

Temporal coordination of glottalic gestures in Karitiana

A Glottal Burst at the End of a Vowel before a Final Unreleased Voiceless Stop Figure 12.1a shows that the word, [mbap ]̚  ‘lame’ ends with an unreleased voiceless bilabial stop [p ]̚  . The audio waveform and the EGG signal suggest that there is a contact between the vocal folds at this moment. The EGG oscillations end at the moment where there is the burst on the audio waveform, and then rises gradually. Therefore, it seems that the closing gestures of the lips and of the glottis are produced simultaneously. This results in the rise of acoustic pressure triggering the burst at the end of the vowel. 3.2

A Glottal Burst at the End of a Vowel before a Voiced Nasal Consonant A number of vowels ending with closing burst have been observed before the allophones of voiced nasal consonants that are described by Storto (1999) as pre-oralized [bm, dn, gŋ] or final unreleased nasals [m̚] (see Figure 12.1b), and for a reason we choose to illustrate one of these cases. In these data, again, the final cycle of the vowel is realized with a burst on the audio waveform as can be observed with the word [kabm̚] presented in Figure 12.6. Another example can be observed in Figure 12.1b. There is an increase in the amplitude of the EGG signal at the end of the vowel suggesting a greater contact and maybe tension of the vocal folds. This is followed by a ‘fall and rise’ movement of the EGG signal during the voiced segment which is due to a rising falling movement of the larynx. 3.3

EGG Frequency (kHz) Audio

0.5 0 ‒0.5 8 6 4 2 0,5 0 ‒0,5 ‒1 3.4

FIGURE 12.6

3.5

3.6

3.7

3.8

3.9

4

Audio waveform wide band spectrogram and EGG signal of the word [kabm̚ ] ‘now’. The arrow on the audio waveform indicates the burst at the end of the vowel preceding the pre-oralized stop [bm]. Arrows on the EGG show the increase and decrease of the signal at the end of the vowel [a] and during the final consonant [bm].

316

Demolin and Storto

3.4 Sequences of Oral and Glottal Gestures The importance of the timing between glottalic and oral gestures is illustrated in Figure 12.7, in which we find sequences of oral and glottal gestures in consonants. An unreleased alveolar stop is followed by a glottal stop in the word [ʔotʔ̚ o:p ]̚  ‘bamboo’. The interpretation of the EGG signal for the initial glottal stop is similar to what has been discussed for Figure 12.4. The consonant cluster [tʔ̚ ] constitutes a sequence of open to close glottalic gestures. The EGG signal of the initial vowel [o] of the sequence shows a loss of contact between the vocal folds. There is then an increase in the EGG signal between the two [o] vowels. This shows an increase in contact between the vocal folds at the end of the voiceless alveolar stop. The acoustic signal does not display this event. The initial consonant [t]̚ is identified by its rising formant transition accounting for the locus of an alveolar consonant (Figure 12.7). This contrasts with the flat transition at the beginning of the following vowel which accounts for the transition between a glottal stop and a vowel. An observation the of the final consonant shows that the unreleased voiceless alveolar stop is produced with a superimposition of the oral and glottal closures. The audio waveform does not show a release burst. The EGG shows that there is a vocal folds contact accompanying the oral closure as the signal

EGG Frequency (kHz) Audio

0.6 0.4 0.2 0 ‒0.2 ‒0.4 8 6 4 2 0.5 1 0.5 0 ‒0.5 ‒1 ‒1.5

1

FIGURE 12.7

1.2

1.4

1.6 1.8 2 2.2 2.4 2.6 2.8 Audio waveform wide band spectrogram and EGG signal of the word [ʔotʔ̚ o:p] ‘bamboo’. The vertical arrow on the EGG indicates the glottal closure at the beginning of the second glottal stop [ʔ] of the word. The two horizontal arrows show the sequence of the open and closed glottis accounting for the voiceless alveolar stop [t]̚ and the glottal stop [ʔ]. The line between dots show the approximate timing of the glottal closure made during the final unreleased voiceless stop [p ̚ ]. The vertical arrow between the audio waveform and the EGG shows that no burst accounts for the alveolar stop, suggesting that the alveolar closure is lasting during the glottal closure. The horizontal arrow shows the signal during the sequence.

Temporal coordination of glottalic gestures in Karitiana

317

gradually goes down and is maintained steady for a while. This example is particularly important to understand the difference between what happens in a sequence of open and closed glottis versus the superimposition of oral and glottal gestures in word-final position when a stop is unreleased. 4

Why is there a Burst at the End of Some Vowels?

The main factor involved in the vowels showing a final burst is a last final cycle with greater intensity. There is therefore a pulse with higher pressure on the acoustic waveform. The explanation that we are proposing is that the burst at the end of the vowels (whether they precede voiceless stops or voiced segments) is the consequence of a greater or stiffer constriction of the glottis and vocal folds at the end of these vowels. This is produced almost simultaneously with an oral closure. Final voiceless stops are often produced with the oral closure gesture of the stop made with an adduction of the vocal folds. In this case, there is a partial or complete superimposition of the two gestures. This is contrary to the expected open glottis gesture of a voiceless stop. These settings likely favour the higher acoustic pressure accounting for the burst at the end of vowels. The burst at the end of Figure 12.5 shows that at the moment of the oral closure (where voicing stops on the audio waveform) there is a sharp difference in amplitude between two consecutive oscillations of the vocal folds. They are followed by two weak amplitude oscillations. This is quite different from sequences where a vowel is followed by a voiceless stop where the EGG waveform gradually diminishes in amplitude. Again this suggests that the oral closure is accompanied by a closing gesture of the vocal folds. Before voiced segments, such as pre-oralized stops (e.g. [bm]), there is an increase of the glottis closure towards the end of the preceding vowel, suggesting a greater glottal resistance. Although the explanation we are proposing should be confirmed on a better physical basis we believe that this phenomenon is important. The ‘abruptness’ of the phenomenon, i.e. the fact that it is observed on a single pulse, makes it difficult to explain. We initially thought that it was due to some contact of the speakers’ lips with the microphone or by a clipped signal. However this was gradually ruled out by the systematic presence of the phenomenon at the end of the vowels and not anywhere else in the signal of the recoded data.

318

Demolin and Storto

Recently we observed similar phenomena in Dâw and Pirahã suggesting that the phenomenon is not specific to Karitiana and might be a phonetic feature to be explained for these languages. There is another important feature to note that deserves further investigation. Indeed Karitiana data observed here suggest that there is very little coarticulation between vowels and consonants in the contexts that we discussed. As it was mentioned above, VC sequences whether word internally or when the last C is a final unreleased stop show that the timing of the oral and glottal closures makes the transition between the vowel (that can show a burst at the end) and the consonant well marked. There is little coarticulation between the vowel and the consonant. The same can be observed with nasal consonants that show no coarticulation, i.e. no nasalization, with oral vowels since there is always either some pre- or post-oralization when an oral V precedes or follows a nasal consonant as can be seen on Figures 12.1b, 12.5 and 12.6. Even when a nasal vowel is in contact with a nasal consonant there is a well marked transition between the two segments. This feature might be specific to Karitiana but further investigations would be necessary to see if this feature is found in other Tupi languages. 5

Phonological Consequences

The glottal closure or vocal folds adduction observed above at the end of voiceless stops explains why they are not exploded as shown in Figure 12.1b, 12.5 and 12.7 (Storto & Demolin 2002). Indeed there is no pressure (Po) build up in the oral cavity if the glottis is closed from the start of the consonant. That is, with no airflow coming from the lungs into the oral cavity because of the glottal closure, the latter does not suffer a significant increase in pressure. A second possibility, illustrated in Figure 12.1a ([mbap ]̚  ‘lame’), shows that after an initial rise of Po which is the consequence of the lip closure in the final consonant, Po gradually diminishes without any apparent leaking at the lips or nostrils. This is due the expansion of the oral cavity between the lips and the glottis with a lowering movement of the larynx. In the case of the nasals, there is no Po because of the lowered velum, that lets the air escape through the nasal cavity. The unreleased stops found in Karitiana are not an isolated case in Tupi languages (Storto & Demolin 2012). This occurs in similar cases in English, i.e. with final voiceless stops which can alternate with glottal stops (Laver 1994). It would be interesting to check whether other Tupi languages also display an adduced or closed glottis simultaneous with an unreleased stop.

Temporal coordination of glottalic gestures in Karitiana

FIGURE 12.8

319

Audio waveform Po, and EGG of the word [mbap̚ ] ‘lame’. Double arrows show ( from left to right) the beginning of the bilabial closure, the maximum of Po corresponding to slight a increase of the EGG signal. It can be observed that Po gradually falls after the maximum that is reached sightly after the beginning of the bilabial closure.1

In Karitiana, there is a phonological process of lenition, in which a morpheme final voiceless consonant becomes a voiced approximant when a suffix starting with a vowel is added to the word. The same phenomenon occurs with nasal consonants. This phonological process has been described in the representation in Table 12.2, adapted from Storto (1999):2 1  Data from this Figure were recorded using an EVA2 portable working station that allows us to record, simultaneously, the acoustic signal, oral airflow, intraoral pressure and the EGG signal (see Demolin 2011 for details on the method). 2  Storto (1999:38) uses the IPA symbol of voiced bilabial and velar fricatives respectively to describe some of the sounds that result from lenition, whereas we use the symbol for a bilabial and velar approximant instead. The velar approximant symbol is used by Storto to indicate the sound resulting from lenition of bilabial stops and nasals in the environment of round vowels.

320

Demolin and Storto

TABLE 12.2 Voicing and lenition of stops and nasals

p t k ɟ

→ → → →

ʋ ɾ ɰ j

m n ŋ ɲ

→ → → →

ʋ̃ ɾ̃ ɰ̃ j̃

The following data illustrate the process: (1) a+taktak+a → ataktaɰa 2s-to swim-imperative ‘swim!’ (2) a+tat+a → ataɾa 2s-to go-imperative ‘go!’ (3) a+hɨɾɨp+a → ahɨɾɨʋa 2s-to cry-imperative ‘cry’ The phenomenon of lenition is easily explained taking into account the articulatory setting of the final voiceless stops or nasals. Since the oral closure is made with none or almost no pressure (Po) in the oral cavity due to the glottal closure or adduction, the stop is (acoustically) weakened as there is no burst. The adducted vocal folds found with these consonants favors voicing when in contact with a following voiced segment. Therefore, the glottis setting (closed or adducted) (i) prevents Po build up in the oral cavity which favors lenition and (ii) also weakens the oral closure which becomes ‘looser’ and more like an approximant at the same place of articulation. Voicing assimilation by the stop is favored by the glottis setting (adducted) when a voiced segment follows, because glottal adduction is a prerequisite for voicing. 6 Conclusion The apparent random occurrence of vowels ending with bursts is related to a specific glottal setting following the vowel. The voicing and lenition processes found in Karitiana are naturally explained by the superimposition of two gestures (one glottal and one oral) having a variable timing and amplitude.

Temporal coordination of glottalic gestures in Karitiana

321

The co-occurrence of glottal stops and stress at the end of words indicates, as well, that the closing gesture of the glottis may have a prosodic function in the language, and should be subject to further investigation. The fact that other Amazonian languages display similar phenomena suggests that it might be an important phonetic feature of these languages which on the other end have quite complex laryngeal features as laryngalization or glottalization of vowels and consonants (Storto & Demolin 2012). References Baken R.J., and Orlikoff, R.F. 2000. Clinical measurements of speech and voice. San Diego: Singular. Demolin, D. 2011. Aerodynamic techniques for phonetic fieldwork. Proceedings of the International Congress of Phonetic Sciences, Hong Kong. 82–85. Hayward, K. 2001. Experimental Phonetics. London: Longman. Laver, J. 1994. Principles of Phonetics. Cambridge: Cambridge University Press. Storto, L. 1999. Aspects of a Karitiana grammar. Ph.D. dissertation. MIT. Storto, L. and Demolin, D. 2002. The phonetics and phonology of unreleased stops in Karitiana. Proceedings of the Twenty-Eighth Annual Meeting of the Berkeley Linguistics Society. 487–497. ———. 2012. The phonetics and phonology of South American languages. In L. Campbell & V. Grondona (Eds.) The Indigenous Languages of South America: A comprehensive Guide. Berlin/Boston: De Gruyter Mouton. 331–390.

Index accent, accentual 237, 239, 240, 241, 243, 244, 245, 246, 247, 254, 294 acoustic 1, 2, 4, 51, 62, 77, 80, 98, 103, 158, 164, 172, 175, 181, 182, 193, 207, 274, 308, 310, 311, 312, 315 aerodynamic 151, 152, 207, 310 affricates 9, 10, 13, 14, 18, 21, 22, 29, 31, 32, 34, 36, 42, 44, 45, 47, 52, 71 airflow 95, 150, 182, 207, 310, 318 allophon 6, 16, 81, 158, 238, 269, 270, 271, 272, 276, 277, 278, 287, 290, 297, 308, 315 aspirated, unaspirated, post-aspirated, aspiration 3, 6, 9, 10, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 44, 45, 52, 129, 130, 131, 132, 133, 143, 144, 145, 146, 147, 148, 150, 151, 152, 157, 163, 260, 261, 266, 267, 268, 270, 271, 273, 274, 275, 287, 289, 294, 295, 296, 297, 304, 306 assimilation 2, 13, 264, 267, 268, 269, 270, 272, 274, 276, 277, 278, 279, 281, 293, 320 bimoraic, see moraic breathy voice, breathiness 4, 32, 33, 77, 78, 79, 80, 152, 169, 207 burst 7, 290, 291, 308, 309, 310, 313, 314, 315, 316, 317, 318, 320 coda 6, 54, 55, 58, 62, 63, 64, 65, 67, 68, 185, 187, 188, 189, 190, 192, 194, 197, 223, 237, 251, 252, 256, 271, 272, 274, 280, 286, 287, 289, 291, 300, 302, 305 constricted 14, 27, 29, 43, 72, 74, 194, 195, 196, 197, 200, 265, 267, 268, 270, 273, 281, 287, 301, 303 continuant 3, 9, 11, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 47, 49, 265, 266, 268, 269, 270, 271, 273 contrastive, contrastive hierarchy, contrastiveness 1, 2, 4, 5, 6, 11, 12, 13 14, 15, 16, 17, 19, 29, 36, 37, 41, 44, 45, 46, 71, 72, 75, 76, 99, 151, 152, 157, 158, 159, 161, 169,

172, 173, 174, 180, 181, 193, 196, 198, 235, 238, 239, 240, 244, 251, 268, 269, 271, 272, 274, 275, 276, 281, 287, 288, 293, 298, 300, 304 creak, creaky, creakiness 4, 7, 59, 60, 61, 62, 63, 64, 69, 74, 81, 84, 85, 87, 88, 92, 93, 94, 125, 134, 151, 153, 155, 158, 159, 165, 166, 168, 169, 170, 171, 172, 174, 175, 181, 182, 184, 189, 190, 191, 193, 195, 196, 203, 205, 207, 208, 209, 210, 2011, 212, 215, 216, 220, 223, 225, 226, 227, 230, 232, 265, 266, 267, 268, 280, 282 culminativity 186, 239, 241, 242 diachronic 23, 35, 53, 54, 55, 56, 63, 69, 129, 130, 254, 271, 274, 275, 276, 277, 281, 286, 303, 304, 305 disyllabic 133, 241, 286, 291, 293, 302 electroglottography 7, 308, 309, 310, 311, 312, 313, 314, 315, 316, 317, 319 echo 88, 89, 287, 296, 298, 302 ejective 1, 3, 9, 10, 14, 15, 18, 19, 23, 24, 26, 27, 30, 31, 32, 33, 34, 35, 37, 38, 39, 41, 44, 45, 51, 52, 53, 54, 55, 56, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69 fundamental frequency 129, 154, 165, 166, 170, 171, 193, 194, 197, 207 glottal, glottalic, glottalized 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 71, 72, 74, 75, 76, 77, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 103, 104, 109, 110, 111, 112, 113, 115, 119, 120, 122, 123, 124, 125, 129, 131, 133, 138, 150, 151, 153, 157, 160, 161, 165, 166, 173, 174, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 191, 193, 194, 195, 199, 200, 203, 204, 205, 206, 207, 208, 209, 210, 211, 212, 213, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 224, 225, 226, 227, 228, 229, 230, 231, 232, 237, 251, 252, 253, 254, 257, 258, 263, 267, 268, 269, 270, 272, 274,

324 glottal, glottalic, glottalized (cont.) 279, 280, 282, 285, 286, 287, 288, 289, 290, 294, 296, 297, 298, 299, 300, 301, 302, 303, 304, 308, 310, 311, 312, 313, 314, 315, 316, 317, 318, 320, 321 glottal stop 3, 4, 5, 7, 20, 30, 51, 52, 54, 55, 57, 58, 59, 62, 63, 64, 65, 68, 69, 71, 74, 76, 81, 84, 85, 88, 94, 95, 96, 97, 103, 104, 109, 11, 115, 119, 125, 129, 133, 160, 161, 165, 166, 173, 174, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 191, 193, 194, 195, 200, 208, 209, 210, 215, 217, 220, 221, 225, 226, 227, 230, 232, 252, 268, 279 pre-glottalization 51, 69, 80, 81, 82, 84, 91, 95, 263, 269, 270 intraoral 7, 150, 310 jitter 60, 61, 62, 65, 67, 93, 95, 97, 98, 104, 10, 106, 107, 112, 113, 115, 116, 117, 121 laryngeal, laryngealization, laryngeal vowel  1, 2, 3, 4, 5, 6, 7, 8, 9, 13, 15, 16, 18, 23, 24, 28, 29, 33, 34, 35, 36, 38, 39, 43, 44, 45, 52, 54, 55, 56, 58, 59, 63, 65, 66, 67, 68, 69, 129, 130, 132, 133, 157, 159, 170, 173, 182, 237, 251, 257, 260, 263, 264, 265, 266, 268, 271, 275, 276,, 277, 278, 281, 287, 295, 300, 301, 304, 321 lenition 7, 264, 266, 319, 320 markedness 9, 11, 13, 16 unmarked 9, 13, 16, 17, 18, 21, 24, 28, 39, 260, 265, 266, 268, 269, 270, 271, 273, 274, 275, 277, 278, 279, 293 merger 11, 29–39, 41, 268, 269, 271, 277 Mesoamerica 4, 157, 159, 173, 174 metrical structure 237, 239 moraic 160, 242, 244, 245, 252, 259, 291, 292, 301 bimoraic 160, 244, 245, 247, 252, 259, 291, 292 nasal 6, 7, 10, 12, 24, 33, 34, 52, 74, 76, 77, 78–97, 100, 104, 107, 108, 113, 115, 117, 122, 123, 131, 134, 135, 136, 138, 143–149, 184, 185, 187, 238, 250, 259, 263–266, 268, 270–272, 276, 277–279, 290, 292, 293, 301, 308, 313, 315, 318, 319, 320

Index neutralization 11, 13, 18, 21, 22, 26, 31, 39, 41, 42, 148, 159, 267, 271, 272, 277, 281 oral closure 7, 80, 81, 308, 316, 317, 320 perception, perceptibility, perceptual, categorical perception 2, 5, 52, 130, 170, 172, 175, 194, 199, 203, 204, 206, 216, 217, 218, 220, 221, 222, 224, 227, 228, 229, 231, 232 phonation 1–5, 9, 29, 74, 97, 134, 143, 157–161, 163–173, 175, 180–182, 188, 189, 191–200, 207, 208, 258, 263 phonological contrast 5, 44, 81, 171, 181, 190, 194, 195, 196, 198, 199, 200, 203 phonologization 55, 68, 99, 125, 170 pitch 4–6, 54, 55, 57, 66, 68, 74, 75, 77, 78, 79, 82–104, 112–115, 121–125, 130, 134, 148, 159, 163, 166, 168, 170–175, 181, 186, 187, 189, 191, 192, 193, 194, 196, 198, 199, 203–210, 211–232, 235, 239, 240, 241, 244, 249, 253, 280 pitch-accent 6, 240, 241, 243, 244 plain roots 286, 291, 292 post-glottalization 80, 81, 94 pre-aspiration 267, 294, 294n, 297 prominent prosodic positions 5, 173, 181, 182, 188, 194, 195, 195n, 197, 198, 199, 199n, 200, 200n prosodic feature 299, 300, 301 proto-language 265, 266, 267, 272, 277, 301, 303 proto-vowel 259, 288n spectral, spectrum 165, 166, 168, 169, 170, 171, 172, 173, 229 spectral tilt 98, 99, 110, 111, 113, 119, 120, 121, 207, 208, 208n spectrogram 76n, 79, 82, 86, 87, 98, 124, 132, 166, 180, 182, 208, 226, 290, 291, 299, 312, 313 spirantization 20, 20n, 21, 22, 29, 32, 33, 33n, 34, 44 spread glottis 2, 10n, 14, 14n, 15, 16, 18, 43, 150, 151, 152, 154, 266, 274n, 281, 287, 301 stem 9, 11, 14n, 16, 16n, 17, 18, 19, 21, 23, 24, 24n, 29, 30, 31, 32, 33, 33n, 34, 35, 37, 51, 53, 54, 59, 63, 64, 65, 66, 67, 68, 69, 146,

Index 160, 184, 241, 246, 252, 261n, 262, 263, 264, 265, 266, 267, 268, 269, 270, 271, 272, 273, 274, 276, 277, 279, 280, 281, 285 stops 3, 4, 5 6, 7, 9, 10, 10n, 12, 13, 14, 14n, 15, 15n, 16, 17, 19, 20, 20n, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 31n, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 44, 45, 51, 52, 53, 54, 55, 57, 58, 59, 62, 63, 64, 65, 66, 68, 69, 71, 71n, 74, 76, 80, 84, 85, 94, 95, 97, 103, 104, 109, 115, 119, 125, 129, 130, 131, 132, 133, 135, 136, 139, 140, 142, 143, 144, 145, 147, 148, 149, 150, 151, 153, 154, 160, 160n, 161, 161n, 164n, 165, 166, 171n, 173, 174, 180n, 181, 182, 182n, 184, 185, 186, 187, 188, 188n, 189, 190, 190n, 191, 193, 194, 195, 200, 204, 205n, 208, 209, 215n, 217, 220, 221, 225, 226, 230, 252, 258, 260, 261, 263, 264, 266, 267, 268, 269, 271, 272, 273, 274, 277, 279, 280, 282, 308, 313, 314, 315, 316, 317, 318, 319n, 320, 321 stress 1, 5, 7, 61, 125, 131, 133, 135, 143, 146, 160, 173, 174, 174n, 181, 184, 185, 186, 187, 188, 189, 191, 192, 193, 194, 195, 195n, 198, 199, 200, 235, 237, 239, 240, 242, 243, 244, 245, 246, 247, 250, 252, 254, 266, 271, 272, 275, 280, 281, 291, 292n, 293, 294, 308, 321 suprasegmentals 7, 53, 54, 55, 63, 64, 205, 206, 274, 287, 292, 293, 301, 304 syllable 7, 55, 60, 62, 63, 67, 68, 69, 72, 74, 77, 78, 82, 103, 115, 131, 146, 160, 170, 170n, 173, 184, 185, 186, 187, 188, 189, 190, 195, 197, 198, 199, 206, 235, 237, 239–240, 241, 242, 243, 245–247, 250, 251, 252, 253, 254, 255, 259, 261–2, 264, 266, 268, 270, 273, 274, 278, 280, 281, 282, 284, 286, 289, 291, 294, 302, 304, 305 tone 1, 3, 4, 5, 6, 7, 8, 54, 55, 69, 71n, 72, 74, 125, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147, 148, 149, 150, 151, 152, 153, 154, 158, 159, 161, 162, 163, 164, 166, 168, 169, 170, 172, 173, 175, 203, 205, 228, 235, 236, 237, 238, 239, 240, 241, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252, 253, 254, 255, 256, 257, 258, 279, 280,

325 281, 282, 292, 292n, 293, 294, 301, 319, 324, 325 contrastive 72n, 152, 161, 162n, ​180, 193, 196, 198 tone heights 6, 235, 237, 246, 247, 238, 249, 250, 251, 254 tonogenesis 54, 55. 63, 64, 67, 68, 69, 71, 74, 75, 81, 99, 122, 125, 129, 130, 254, 281 typology 6, 169, 241 of consonant-F0 interactions 129 of feature anchoring 175 of phrasing and anchoring of non-modal phonation 171 of tone-phonation interaction 173 of tonogenesis patterns 133 phonological 237–238 voicing alternations 14, 15, 16, 17, 18, 21, 29, 36, 37, 41, 44, 45 vowels 3, 4, 6, 19, 29, 43, 52, 53, 54, 55, 57, 58, 59, 60, 62, 63, 64, 65, 66, 67, 68, 69, 71, 72, 74, 75, 76, 81, 89, 95, 97, 98, 99, 109, 113, 115, 116, 117, 118, 119, 120, 121, 122, 124, 125, 129, 130, 131, 132, 133, 134, 135, 136, 140, 141, 142, 143, 148, 150, 151, 152, 153, 157, 160, 162, 163, 164, 166, 168, 169, 172, 173, 174, 175, 180, 181, 182, 184, 186, 187, 188, 191, 193, 194, 195, 196, 197, 198, 199, 200, 203, 204, 205, 206, 207, 208, 209, 215, 220, 223, 225, 226, 228, 252, 263, 269, 270, 274, 280, 281, 286, 287, 288, 289, 290, 291, 292, 293, 295, 296, 298, 299, 301, 302, 303, 304, 305, 308, 309, 310, 311, 312, 313, 314, 315, 316, 317, 318, 319, 320, 321 echo vowels 88, 89, 287, 294, 296, 298, 302 laryngealized vowels see laryngealized, vowels nasal vowels see nasal, vowels nucleus 63, 65, 161, 162, 163, 290, 292 rhyme 291 waveform 98, 132, 208, 226, 309–317, 319 word 4, 11, 53, 57, 61, 180, 186, 187, 239–46, 249–52, 313, 321