Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Mundurukú : phonetics, phonology, synchrony, diachrony Picanço, Gessiane Lobato 2005

You don't seem to have a PDF reader installed, try download the pdf

Item Metadata


831-ubc_2006-13030X.pdf [ 28.51MB ]
JSON: 831-1.0092991.json
JSON-LD: 831-1.0092991-ld.json
RDF/XML (Pretty): 831-1.0092991-rdf.xml
RDF/JSON: 831-1.0092991-rdf.json
Turtle: 831-1.0092991-turtle.txt
N-Triples: 831-1.0092991-rdf-ntriples.txt
Original Record: 831-1.0092991-source.json
Full Text

Full Text

MUNDURUKU: PHONETICS, PHONOLOGY, SYNCHRONY, D1ACHRONY  by GESSIANE LOBATO PICANCO  B.A. Federal University of Para, 1999  A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY  in THE FACULTY OF GRADUATE STUDIES  (Linguistics)  THE UNIVERSITY OF BRITISH COLUMBIA  December 2005  © Gessiane Lobato Picanco, 2005  Abstract This dissertation offers an in-depth investigation of the phonology of Munduruku, a Tupi language spoken in the Amazonian basin of Brazil, approached from three interrelated perspectives: phonetic, phonological and diachronic. It examines (i) the Munduruku vowel and consonant inventories, (ii) syllable structure and syllabification, (iii) phonotactic patterns, (iv) nasal harmony, (v) consonant mutation, (vi) tone system and the tone-creaky voice interaction, (vii) reduplication, and (viii) the phonological behavior of various affixes. The phonetic investigation focuses on several acoustic properties of segments (i.e. vowels and consonants), and on phonological contrasts observed in vowels, in particular the oral-nasal and modal-creaky voice oppositions, in addition to tonal distinctions. This is done with a view to determining how and to what extent such phonetic realizations can be imposed on phonological representations. These issues constitute an important part of the study, and are particularly relevant to the discussion about the coarticulatory effects observed in the realization of stops, nasals and laryngeals. The study also offers a formal account of all major phonological processes attested in the language such as syllabification, nasal harmony, consonant mutation, tone, etc. The theoretical model adopted here is Optimality Theory (OT), which defends a representation of the structural design of grammars based upon a ranking of universal constraints. Each chapter contributes to the development of an OT-based grammar of the phonology of Munduruku by examining new aspects of the language, and by situating them in a large-scale scenario until the OT-grammar is assembled. This result is presented in the last chapter. In search of evidence for the synchronic analysis, and for a better understanding of some uncharacteristic patterns, the study turns to the historical development of the language. Using data from Kuruaya, a sister language to Munduruku, hypotheses about the stage that preceded both languages, Proto-Munduruku, are made available. In recovering this stage, and the stage that preceded the modern period, it is possible to recover many of the changes the grammar has undergone and which culminated in the synchronic patterns. Ultimately, this study argues for an approach to synchronic grammars as a composite of universal and language-specific properties, determined by diachronic changes.  Contents Abstract  ii  Contents  iii  List of tables  viii  List of figures  x  Acknowledgements  Chapter 1  x  iii  Introduction  1.1  Wuyjuyu "Our people"  1  1.2  The Tupi stock  5  1.2.1  Munduruku  5  1.2.2  Kuruaya  6  1.3  Major goal  7  1.4  Organization of the dissertation  7  1.4.1  Acoustic analysis  7  1.4.2  Phonological analysis: Optimality Theory  8  1.4.3  Diachronic analysis  10  1.4.4  Overview of the chapters  11  Chapter 2  Phonetic and phonological structures of vowels  2.1  Introduction  13  2.2  Vowel qualities  17  2.2.1  Vowel dispersion  18  2.2.2  The acoustic space of Munduruku vowels  20  2.3  2.4  Modal vowels: the oral-nasal contrast  26  2.3.1  30  Acoustic effects of nasalization on vowel height  Phonation types  33  2.4.1  The modal-creaky voice opposition  35  2.4.2  Acoustic properties of the creaky-modal contrast  37 Procedures  39  2.4.3  Results and discussion  2.5  Formant  40 frequencies  40 Overall duration  43 Fundamental frequency  43 Periodicity  50  51  Spectral tilt  Conclusion  Chapter 3  55  Phonetic and phonological structures of consonants  3.1  Introduction  56  3.2  The inventory  57  3.3  Voiceless stops  59  3.3.1  Instrumental analysis: procedures  62  3.3.2  Results  65  3.4  Voiced stops  71  3.5  Nasal stops  74  3.5.1  76  Preoralization as a coarticulatory effect  3.6  Approximants  3.7  Laryngeals: /h/and /?/  3.8  v  80 82  3.7.1  111 as a complete closure  84  3.7.2  111 as constricted voicing  87  Features for consonants  Chapter 4  90  Syllable structure and syllabification  4.1  Preliminaries  93  4.2  The basics of syllabification  95  4.2.1  Contiguity  97  4.2.2  Inter- and Intra-morphemic C O N T I G U I T Y  99  4.3  Hiatus 4.3.1  4.4  101 Vowels versus glides  Onsets 4.4.1  105 111  Contiguity versus Dependence  112  4.4.2  4.4.3 4.5  Word-initial Irl  120 Changing the ranking  123 Loss of a constraint: systematic versus accidental gaps  124  Asymmetries in V C V syllabification  126  Consonant clusters  132  4.5.1  Clusters I: voiceless stop + C  133  4.5.2  Clusters II: nasal + C  142  4.5.3  Clusters III: approximant + C  146  4.5.4  Sequences coronal + coronal  148  4.6  The imperfective {-m}  158  4.7  Conclusion  163  Chapter 5  Phonotactics: Synchrony and Diachrony  5.1  Introduction  165  5.2  The synchronic patterns  166  5.3  The development of Munduruku consonants and phonotactics  169  5.3.1  Background  169  5.3.2  Previous comparative studies on Tupi languages  171  5.4  Secondary split: the origin of *dv  171  5.4.1  The change in nasal harmony  174  5.4.2  The reanalysis of a gap  179  5.5  The origin of *ni  185  5.6  The origin of * f i , *<^i, and *rji  188  5.6.1  Between Pre-Proto-Munduruku and Proto-Mundurukii  190  5.6.2  Between Proto-Mundurukii and Pre-Munduruku  196  5.6.3  Between Pre-Munduruku and Munduruku  202  5.6.4  Summary of the changes  204  5.7  The origin of *si  206  5.7.1  Native versus borrowed vocabulary  206  5.7.2  The history of /s/ in Munduruku  207  5.8  On the history of/J"/  211  5.9  Conclusion  216  Chapter 6  Nasal Harmony  6.1  Introduction  218  6.2  Nasal harmony: general aspects  218  6.3  Theoretical assumptions  221  6.3.1  Locality  221  6.3.2  Grounding conditions  224  6.3.3  Harmony via no-disagreement (Pulleyblank 2002)  226  6.4  Nasal harmony in Munduruku  229  6.4.1  The behavior of laryngeals  231  6.4.2  Nasal vowels versus nasal consonants in harmony systems  236  6.5  Nasal harmony in Kuruaya  246  6.6  A n OT account of a historical change in nasal harmony  250  Chapter 7  Consonant Mutation  7.1  Introduction  255  7.2  Morphosyntactic aspects  258  7.2.1  Nouns  259  7.2.2  Verbs  263  7.2.3  Other cases  269  Phonological aspects  273  7.3.1  Lenition versus fortition; lenition and fortition  277  7.3.2  Discussion and analysis  282  7.3.3  The t/d alternation and nasality  289 Dialect A  294 Dialect B  295  7.3  7.4  On the history of the alternations tfAfe and p/b  300  7.5  Conclusion  308  Chapter 8  The phonetics and phonology of tone and creaky voice  8.1  Introduction  310  8.2  Tone processes and theoretical assumptions  311  8.3  High tones  317  8.4  8.5  Underlying L-tone versus surface L-tone  319  8.4.1  325  Similarity effects of the O C P  Tonal polarity  329  8.5.1  Properties of tonal polarity  330  8.5.2  Proposal: Polar tone as floating H tone  335  8.6  Unstable H tones  338  8.7  The interaction between creaky voice and tone  344  8.7.1  If [+c.g.] then L tone  345  8.7.2  Creaky voice and H tone  349  8.8  Evaluating the proposal  Chapter 9  358  Conclusion  9.1  Building the global ranking  368  9.2  Further support  371  9.2.1  The emphatic clitic {=nma}  371  9.2.2  Third person {t-}  373  9.3  9.4  Placing Reduplication in the OT-grammar  374  9.3.1  Progressive reduplication  376  9.3.2  Reduplication and t-deletion  379  9.3.4  Reduplication and consonant mutation  380  Reduplication with fixed segmentism  382  9.4.1  Reduplication and *Js  383  9.4.2  V-Reduplication and epenthetic /h/  385  9.5  Language change and its consequences  386  9.6  The OT-based grammar of the phonology of Munduruku  387  Appendix: List of constraints  390  Bibliography  395  Tables Table 2.1. F l , F2, F3 mean values and standard deviations for Munduruku vowels  21  Table 2.2. Mean values and standard deviations of formant frequencies for oral and nasal vowels  33  Table 2.3. Mean values and standard deviations of formant frequencies for modal and creaky vowels  41  Table 2.4. Duration means and standard deviations for modal-creaky vowels  43  Table 2.5. Results of significance tests: L-tone versus creaky voice  47  Table 2.6. Results of significance tests: L-tone versus H-tone  48  Table 2.7. Means and standard deviations for measures of jitter  51  Table 2.8. Results of significance tests: Duration of glottal pulses for creaky-modal pairs  51  Table 3.1. Munduruku consonant inventory  58  Table 3.2. Closure duration means for [p, t, k] in sequences V N + C V , V C V , VC+V,and V C i + Q V  65  Table 3.3. Results of closure duration for [p, t, k] in sequences V N + C V , V C V , V C + V , and V C i + Q V  66  Table 3.4. Results of a random selection of sequences V C V , V C + V and V C i + C V i  67  Table 3.5. V O T means by place of articulation  67  Table 3.6. Results of V O T for [p, t, k] in sequences V N + C V , V C V , V C + V , and V C i + C , V  68  Table 3.7. Vowel duration in word-final C V and C V C syllables  69  Table 3.8. Duration means for vowels before [p, t, k] in sequences V C V and V C + V  69  Table 3.9. Results for the palatal affricate [tf] in intervocalic position  70  Table 3.10. Duration means of V O T and burst for voiced stops word-initially  72  Table 3.11. Means for closure duration of voiced stops in sequences V C V and voiceless stops in sequences V N + C V  73  Table 3.12. Duration means of vowels preceding voiced and voiceless stops  73  Table 3.13. Duration means for the nasals /m, n/  78  Table 3.14. T-test results for nasals /m, n/ in sequences VN#, VN#, V N + V , and V N + V  79  Table 3.15. Results of closure durations for/?/ in sequences V C + ? V and V C + ? V  86  Table 4.1. Summary of duration means for voiceless stops  126  Table 4.2. Summary of duration means for nasals  127  Table 4.3. Combinations of coda x onset  133  Table 5.1. Relative frequency of Munduruku consonants in C V ( C ) syllables  167  Table 5.2. Correspondence set I: d/1 word-initially  172  Table 5.3. Correspondence set I: d/1 word-medially  172  Table 5.4. Correspondence set II: n/1 word-initially  173  Table 5.5. Correspondence set II: n/1 word-medially  173  Table 5.6. Correspondence set III: r/n intervocalically  186  Table 5.7. The word "hammock" in Tupi  187  Table 5.8. Correspondence set TV: J/d before N  191  Table 5.9. Correspondence set V : cfe/d intervocalically in oral contexts  192  Table 5.10. Correspondence set V I : [n]/n intervocalically in nasal contexts  193  Table 5.11. Correspondences in consonant mutation  197  Table 5.12. Correspondence set VII: tj/tf word-initially, intervocalically and postconsonantally  200  Table 5.13. Summary and illustrations of the evolution of ***C in Munduruku  205  Table 5.14. Correspondence set VIII: 0 i / - s i  207  Table 5.15. Correspondence set IX: sv/si  208  Table 5.16. Correspondence setX: a/i  209  Table 5.17. Correspondence set X I : psv/biv  210  Table 5.18. Correspondence set XII: J/J  212  Table 5.19. Correspondence set XIII: Ji/ki  212  Table 6.1. Cross-linguistic distribution of triggers and targets in nasal harmony  242  Table 7.1. Distribution of consonants morpheme-initially in inalienable nouns  261  Table 7.2. Distribution of consonants morpheme-initially in stative, unaccusative, transitive, and unergative verbs  265  Figures Figure 2.1. Formant plots of Munduruku vowel qualities  22  Figure 2.2. A comparison between [o] and [o~u]  24  Figure 2.3. F l vs. F2 plots of Munduruku oral-nasal vowels  32  Figure 2.4. Waveforms illustrating modal [a] and creaky [a]  38  Figure 2.5. F l vs. F2 plots of Munduruku creaky-modal vowels  42  Figure 2.6. Graphs showing FO realization for H-tone f\J, L-tone Ii/, and creaky l\l  44  Figure 2.7. Graphs showing FO realization for H-tone Id, L-tone Id, and creaky Id  45  Figure 2.8. Graphs showing FO realization for H-tone lal, L-tone lal, and creaky / § /  45  Figure 2.9. Graphs showing FO realization for H-tone /a/, L-tone lal, and creaky lal  46  Figure 2.10. Graphs showing FO realization for H-tone lol, L-tone lol, and creaky lol  46  Figure 2.11. Mean FO values for vowels with H-tone, L-tone and creaky voice  47  Figure 2.12. Waveforms showing adjacent glottal pulses of modal [a] and creaky [a]  50  Figure 2.13. Mean values of the amplitude differences between the fundamental (FO) and the second harmonic (h2) for all vowels  52  Figure 2.14. Mean values of the amplitude of the fundamental (FO) for modal-creaky pairs of vowels Figure 2.15. F F T spectra of modal [a] and creaky [a]  53 53  Figure 2.16. Means for amplitude differences between the fundamental (FO) and harmonic closest to F l for all vowels  54  Figure 3.1. A waveform illustrating duration measures  62  Figure 3.2. A waveform showing a typical burst event for voiceless stop [k]  63  Figure 3.3. Illustration of stop closure in the sequence [t-t]  64  Figure 3.4. Waveform of a word-initial Idl  72  Figure 3.5. Waveform of the Munduruku word cfyededem [djedede^m] 'to talk' Figure 3.6. Waveforms showing portions of oral and nasal vowels followed by Iml  73 76  Figure 3.7. Schematic representation of the relative timing of velic and oral gestures in preoralized nasals  80  Figure 3.8. Spectrograms of V h V sequences  83  Figure 3.9. Spectrograms of [vnv] and [vhv] sequences  84  Figure 3.10. Waveform to illustrate 111 as a complete stop  85  Figure 3.11. Expanded waveforms comparing the onsets of H-tone and L-tone vowels following 111  86  Figure 3.12. Spectrograms illustrating the realization of an intervocalic 111  88  Figure 3.13. Spectrogram of the Munduruku word wa?i?a 'gourd'  88  Figure 3.14. Spectrograms showing 111 after a glide and a nasal consonant Figure 3.15. Spectrograms illustrating the sequence [v?v]  89 90  Figure 4.1. Spectrogram showing sequences V . V versus G V  105  Figure 4.2. Spectrogram showing an intervocalic [w]  106  Figure 4.3. Spectrogram of the sequence /koa/ [k a] in the word koato 'summer'  106  Figure 4.4. Illustration of the closure phase in the sequence [p-t]  134  Figure 4.5. Waveform illustrating the sequence /p-b/ realized as a long [b]  135  Figure 4.6. Waveform illustrating the sequence /k-b/ realized as [k-b]  136  Figure 4.7. Waveform illustrating the sequence /k-dj/ realized as [g-cfe]  136  Figure 4.8. Waveform illustrating the sequence /p-d/ realized as [b-d] ( E M )  136  Figure 4.9. Waveform illustrating the sequence /p-d/ realized as [b-d] ( A K )  137  Figure 4.10. Waveform illustrating the sequence /p-d/ realized as [p-d] (JT)  137  Figure 4.11. Expanded waveform illustrating the affricate /tjV  138  Figure 4.12. Expanded waveform illustrating the stop + fricative sequence /k-jV  139  Figure 4.13. Expanded waveform illustrating the stop + fricative sequence /p-s/  139  Figure 4.14. Expanded waveform of the sequence IX-ral  140  Figure 4.15. Waveforms illustrating non-syllabic [a]  141  Figure 4.16. Waveform illustrating a full vowel [a]  142  Figure 4.17. Waveforms illustrating sequences nasal + stop  144  Figure 4.18. Waveform illustrating non-syllabic [a] in a sequence nasal + Irl ( A K )  145  Figure 4.19. Waveform illustrating non-syllabic [a] in a sequence nasal + Irl (JT)  145  Figure 4.20. Waveform illustrating non-syllabic [a] in a sequence nasal + Ixl ( A K )  146  w  Figure 6.1. Spectrograms of V h V sequences in the Munduruku words ihi 'winter' and iaham 'to bite s.t'  233  Figure 6.2. Spectrograms of V h V sequences in the Munduruku words afhi 'mother' and ihi 'winter'  234  Figure 6.3. Spectrograms of V ? V sequences in Munduruku  235  Figure 8.1. Sequences of H tones within the morpheme and in morpheme concatenation  318  Figure 8.2. Surface realization of the noun dig 'smoke' following a lexical L tone and toneless mora  321  Figure 8.3. Graph showing FO realization for default and lexical L tones  325  Figure 8.4. Graph illustrating tonal polarity of the classifier -ta  330  Figure 8.5. Graphs illustrating unstable H tones versus stable H tones  339  Figure 8.6. Waveform showing a creaky vowel in the word akobg 'banana'  347  Figure 8.7. Waveform showing loss of creaky voice in a sequence /L+L/  347  Figure 8.8. Waveform of the Munduruku noun kgbi 'sky'  351  Figure 8.9. Waveform of the noun kgbi 'sky' in the word kabi-dig 'mist'  352  Figure 8.10. Waveform of the noun kgbi 'sky' in the word kabi-kerere 'cloud'  352  Acknowledgements Many people contributed to making this dissertation possible, and here I would like to express my gratitude to all of them. First of all, this work would not exist without the support and assistance of the Munduruku community, in particular, of the various speakers with whom I worked in these 9 years of research on the language: Adalto Akay, Adonias Kaba, Antonio Tawe, Carlos Pago, Dionlsio Bor5, Edelson Munduruku, Genivaldo Kaba, Ines Kaba, Jairo Torres, Joao Maria, Jose Crixi, Lorival Boro, late Luciano Boro, Maria Zilda, Martinho Boro, Nilza Kaba, Rafael Pago, Raimundo Boro, and Valmar Kaba. As well, I am very grateful to Maria Curuaia (Apalapan-Kirie) and Paulo Curuaia (Tupao), two of the last five elder speakers of the Kuruaya language. Their energy and enthusiasm during the data-collection sessions have always amazed me. I owe a life-long gratitude to my supervisor, Douglas Pulleyblank, and to Denny Moore for their guidance and generous support in all academic, and sometimes personal, matters. Thanks for believing in me. I hope I have fulfilled your expectations. To the professors of the Linguistics Department, especially Patricia Shaw, Guy Carden, Rose-Marie Dechaine and Henry Davis, I would like to say that you have made a substantial contribution to my training in Linguistics at this university. Thank you all. I also acknowledge the important contribution of my colleagues of the Tupi Comparative Project (Museu Emilio Goeldi, Brazil): Sergio Meira, Nilson Gabas Jr., Luciana Storto, Sebastian Drude, A n a Vilacy Galucio, Didier Demolin, and Carmen Rodrigues; and the support of three friends who helped me significantly in my first linguistic steps: Ana Carla Bruno, Marilia Ferreira and Eduardo Ribeiro. I also appreciate the friendship of a number of students: Amelia Silva, Andres Salanova, Eun-Sook K i m , Jason Brown, Kayono Shiobara, Oladiipo Ajiboye, Rachel Wojdak, Sugunya Ruangjaroon, Solveiga Armoskaite, and Yunhee Chung. Thank you all for your time and support. There are no words that could possibly express my gratitude to my family. They were by my side from the beginning to the end. Particularly, Gessy Picanco, my mother, and Geiva Picanco, my sister, thank you very much for taking charge, so many times, of my mother duties to Ailime so I could pursue my dream. I am very fortunate to have all of you. xiii  To my wonderful daughter Ailime: " Y o u have always amazed me with your intelligence and understanding of the nature of my work. Y o u are only 8 years old, but you have given me all the strength and inspiration I needed to never give up." To Tyler Peterson: " Y o u have been my partner in this endeavor and in life. Thanks for your friendship and persistence, and for keeping us together through bad and wonderful times while I was writing this dissertation. I look forward to 'sharing all that is to come'."  M y academic training has benefited from the financial and technical support of the following institutions: Wenner-Gren Foundation for Anthropological Research, Conselho Nacional de Desenvolvimento Cientlfico e Tecnologico (CNPq, Brazil), Museu Paraense Emilio Goeldi, and the University of British Columbia. Fieldwork on Munduruku was financed by the Wenner-Gren Research Grant 6616, awarded to Denny Moore, and S S H R C Research Grant 4102002-0041, awarded to Douglas Pulleyblank; and fieldwork on Kuruaya was financed by the Endangered Language Fund, awarded to the author. A l l errors are my own.  CHAPTER 1  Introduction 1.1  Wifyjifyu "Our people" 1768. Bastien und Bastienne by Wolfgang Amadeus Mozart, his first published opera, is  performed in Vienna; Joseph Fourier, initiator of the Fourier series, is born in France; the Munduruku Indians, whose language is the heart of this dissertation, enter Brazilian colonial history. The Tupi were the first native American groups met by the Europeans in the early period of Brazilian colonization (in 1500). But the Munduruku, also a Tupi tribe, were first reported only in 1768, in the Roteiro da viagem da cidade do Para ate as ultimas coldnias do sertdo da provincia [Report of the trip from the city of Para to the remote colonies of the province], written by the Vicar General of Rio Negro, Jose Monteiro de Noronha. First called Matucucu or Matucaru (they call themselves Wuyjuyu "our people"; see below), the Munduruku inhabited the banks of the Maue riyer, a tributary of the Amazon. They became best known, and feared, from 1770 onwards, due to the devasting attacks they launched against white settlements and other tribes along the banks of the Tapajos river (see maps in (2) below), while expanding their territory in the region (Horton 1948, citing Manoel Baena 1885 and Almeida Serra 1779). By the beginning of the X l X t h century, the Munduruku completely dominated the region between the Madeira and Tapajos rivers, a fact that led the historian Aires de Casal (1917-1976) to name the region after them: Mundurucania. The  Munduruku  abandoned warfare  around  1796, when defeated  by a troop of  expeditionaries. Peaceful contact was then established. Memories of those moments were kept alive until more than 150 years later: "In the old days our grandfathers were still wild and fought against the white men. The whites used to come up our rivers in their canoes, and we always battled. One day a group of them came up, and there was a fight and our men were driven off. Two of our young men were wounded and were left behind. They were captured and taken away. The next time the white men came we were about to attack when the two men who were captured stood up in the canoe and told us not to do anything as these people were our friends. They then came forward and showed us clothes, knives, axes, and many other good things that the whites gave them. They said that i f we gave rubber and farinha to the whites, we too would  receive these things. The elders decided to do this and ever since we have been friends of the white men." Murphy (1960: 27) The Munduruku call themselves Wuyjuyu [wSyc^ayS] which means "our own, our people" {way- ' 1 person plural inclusive', -cfys 'with, together', and -yS 'plural'). The name Munduruku s t  (or its variants) denotes a species of ant, and was given to them by another group, the Parintintin (Stromer 1932, cited in Horton 1948). They were also known as Paikise (Father Knife), because of their head-hunting activities, or Caras-Pretas (Black Faces), because of their facial tattooing (Kruse 1934). In fact, tattooing was a noticeable feature of the Munduruku culture, an activity beginning early in childhood and continuing until the adult phase when their whole body was eventually tattooed (e.g. Leopoldi 1979). Very little is known about the Munduruku culture prior the contact. But they were wellknown for being one of the most ferocious tribes, often in the hunt for trophy heads, the greatest honor for a Munduruku warrior. War was then its best source for this activity (Spix and Martius 1823; Murphy and Murphy 1954; Acquaviva 1976; Leopoldi 1979). Socially, the group had a well-developed moiety and sib system (Kruse 1934; see also Murphy  1959), divided  into two exogamic moieties,  "White" (Iriritayu)  and "Red"  (Ipakpukayu), each of which consisting of various families. The families had eponymous animals or plants, and believed themselves to be related to them - for example, White moiety: bord 'cotton plant, ikopi 'wasp', tawe 'monkey', etc.; Red moiety: bio 'tapir', wito 'curassow', parawa 'red macaw', apak 'tree, sp.', etc. According to the Munduruku culture, this organization was determined by Karosakaibu, the mythologic hero responsible for the unification of the group. The Munduruku still make use of the moiety system and eponymous nomenclature, despite their adoption of Portuguese names. However, the former denominations, which once served to organize the group into clans, are now hardly referred to among themselves. A typical Munduruku village consisted of a men's house (eksa) and a few dwelling houses. Each village had a set of sacred instruments, in which their ancestors' spirits were believed to be embodied. These instruments were kept in the men's house (eksa) and could by no means be seen by women. Game animals (puca) were also believed to have spirit mothers (puca xi) to protect the species. A man who killed an animal without cause could have his soul taken by the animal's  spirit mother and placed in inferior animals. Hunting was acceptable only for their own subsistence, which also included fishing, gathering and horticulture. For further information about the Munduruku history, culture, and other related topics, see: Goncalves Tocantins (1877), Barbosa Rodrigues (1882), Hartt (1885), Coudreau (1897-1977), Aires de Casal (1917-1976), Kruse (1934), Nimuendaju (1938, 1949), Kempf (1945a, b), Mense (1946-1947), Horton (1948), Murphy (1954, 1956, 1958, 1959, 1960), Brown (1957), Wilson (1958), Frikel (1959), Murphy and Murphy (1974), Spix and Martius (1976), Acquaviva (1976), Ramos (1978), and Leopoldi (1979). Nowadays the Munduruku are mainly settled in the south of the Amazon, geographical centre of Brazil, as in the map in (1). Strictly speaking, the area is situated in the southwestern corner in the state of Para, shown in the map in (2) below. With a population of approximately 7,000 individuals, they are spread around more than 50 villages in the Munduruku Indigenous Reserve. (Source of maps:; originally from Queixalos and Renault-Lescure 2000.) (1)  Map of Brazil  (2)  Map insets of (i) the Munduruku territories and (ii) the Amazon basin  These more than 200 hundred years of contact had a strong effect on their aboriginal culture, and very little resembles the former cultural pattern. But, fortunately, the language survives.  1.2  The Tupi stock The Tupi stock (Nimuendaju 1948; Rodrigues 1958a-b, 1964, 1970, 1986, 1999; Loukotka  1968; Lemle 1971) comprises ten families of languages, one of which is the Munduruku family, with two languages: Munduruku (§1.2.1) and Kuruaya (§1.2.2). (3)  Tupi stock  Arikem Aweti Juruna Mawe Monde Munduruku Purubora Ramarama Tupari Tupi-Guarani  Munduruku Kuruaya  1.2.1  Munduruku  Munduruku is a Tupi language of the Munduruku family. Their linguistic situation is relatively stable, but the population living in villages close to the city are beginning more and more to make use of the national language, Portuguese. A fair amount of them have already moved to the city, and have become fluent speakers of Portuguese, especially the young people. There is also a considerable number of monolinguals, in particular, elders, women, children, and those living in more remote villages. It is unknown how many dialects the language presently has. Crofts (1967) is the only source in this area. Munduruku has been my research language since March 1996, as an undergraduate student and member of the Tupi Comparative project of the Linguistics Division of the Museu Paraense Emilio Goeldi, in Belem-Para. It was when I heard my first Munduruku word, wuykat /waykat/ 'Good afternoon'; at that time I transcribed it as [weyka]. It was not the best word to initiate the study of the language, especially because of the creaky vowels (as we w i l l see throughout this work), and which I did not notice at that time. The Munduruku data presented here are from my own field notes, unless otherwise indicated. Data collection took place anually, for a period of approximately 1-3 months every year since 1996, and mostly in Belem-Para, where the Munduruku constantly go. Fieldwork on the language also includes a visit to five Munduruku villages in 1999, and to Jacareacanga in 2003, a town near the reserve and to where many Munduruku families have already moved. During these 9 years of intermittent fieldwork, I have had the opportunity to work with 19 native speakers, who kindly collaborated with my research and to whom I express my gratitude: Adalto  Akay (26 years old), Adonias Kaba (27, Jacareacanga), Antonio Tawe (38, Missao Cururu village), Carlos Pago (28, Caboroa village), Dionisio Boro (37, Caboroa), Edelson Munduruku (26, Jacareacanga), Genivaldo Kaba (29, Porto village), Ines Kaba (66, Jacareacanga), Jairo Torres (36, Sai-Cinza village), Joao Maria (20), Jose Crixi (42, Missao Cururu), Lorival Boro (approximately 28, Missao Cururu), late Luciano Boro (approximately 45, Missao Cururu), Maria Zilda (approximately 45, Missao Cururu), Martinho Boro (62, Missao Cururu), Nilza Kaba (14, Caboroa), Rafael Pago (approximately 50, Carocal village), Raimundo Boro (27, Missao Cururu), and Valmar Kaba (approximately 28 and 23, Missao Cururu). The results of this research are presented here. A l l errors are my own. This is not the first study of Munduruku; yet, it is certainly the most detailed. The language has a relatively vast literature on different aspects of its grammar, which go from word lists and notes (for example, Friar Hugo Mense's material published by Rodrigues in 1947) to studies with pure linguistic focus (e.g. Braun and Crofts 1965; Crofts 1967, 1971, 1973, 1985; Goncalves 1987; Angotti 1998; Picanco 1997, 1999, 2001, 2002a-d, 2003a-c, 2004a-b; 2005a-b; Gomes 2000, 2002; Pica et al. 2004). However, since Braun and Crofts' (1965) preliminary proposal of the Munduruku segmental and tonal phonology, very little has been done on this part of the grammar (see also Picanco 1997, 1999, 2002a-c, 2004b). This study counters many of the previous claims about the phonology of Munduruku, and in addition, provides a detailed analysis of the majority of the phonological processes observed in the language. It also offers a systematic investigation of the phonetics of the language, which was, until Picanco's manuscript (2002d; also Peterson and Picanco 2003), completely unexplored. A s an additional benefit, the language is approached from a historical point of view, which not only helps but sometimes provides a better explanation for the various irregularities of the synchronic patterns. This diachronic investigation (see §1.4.3 below) compares Munduruku to its sister language Kuruaya.  1.2.2 Kuruaya Kuruaya is an extremely endangered language, and will be extinct soon. There are, according to Costa (1998) and Rodrigues (1999), only five elder speakers, four of them live in the Altamira city, in north-central Brazil, in the state of Para. M y research on this language began in 2002, working with two elder speakers, Maria Curuaia (82 years old) and Paulo Curuaia (89 years old).  The history of the group is unknown, and the language underdescribed. There seems to be available only a few word lists (Snethlage 1910; Nimuendaju 1930), and a preliminary study of its segmental phonology (Costa 1998; see also Picanco 2003b, 2005a-b). Although Kuruaya will not be the focal point of this dissertation, many aspects of its phonology are explored; for example, its segmental phonology and phonological processes such as nasal harmony, consonant mutation, etc. A l l the hypotheses and data for this language are based on my own field notes, unless otherwise cited. Data collection took place in Altamira in July 2002 and in January 2003. 1.3  Major goal The  investigation of Munduruku is approached  from  three viewpoints: phonetic,  phonological and diachronic. The ultimate question that the study attempts to answer is: What is the connection between how the language is spoken, how it is represented and how it was represented in the past? It is, by no means, the purpose of this work to offer a profound theoretical discussion on the matter, although theoretical assumptions have to be made. Rather, I attempt to provide a detailed description of the phonetic and phonological structures of the language, and formalize them (see §1.4.2 below). Where a purely synchronic explanation of the facts fails, the analysis turns to the historical development of the language (§1.4.3), which most of the time provides a better understanding of the patterns. 1.4  Organization of the dissertation  1.4.1 Acoustic analysis The phonetic analysis is acoustic, in which more than 1,500 tokens were examined. The data were recorded in the field using a D A T tape-recorder and a Sony lapel microphone. Three male speakers, Adonias Kaba ( A K , 27 years), Jairo Torres (JT, 36 years), and Edelson Munduruku ( E M , 26 years), all of whom speak the same variety of Munduruku and are familiar with the Munduruku orthography, were asked to read the target words in the Munduruku sentence in (4), a sentence in which the test word is pronounced without emphasis, in addition to being relatively independent syntactically so that its tones cannot be affected by neither preceding nor following words. (4)  Orthographic convention  "Ijop  gasu"  Phonemic representation  /icfeop  rjasa/  Phonetic transcription  [i<feop  JiasS] 7  "This is  now".  ( (is).. .now)  Because the orthography does not mark contrasts of tone or creaky voice, I introduced the conventional marking (y) to distinguish minimal pairs. Pairs contrasting for creaky voice, e.g. /wida/ 'clay' and /wida/ 'jaguar', which in the orthography are both written as "wida", appeared in the stimuli with the creaky vowel marked by a tilde placed under the vowel: "Ijop wida gasu" 'This is jaguar now'; pairs contrasting for tone, e.g. Id 'tobacco' which has a low tone level and Id 'path' a high tone level, both written as "e" in the orthography, received an acute accent marking the high-tone form: "Ijop e gasu" 'This is path now'. Words that were not minimally contrastive were in the stimuli as they are in the orthography. This method turned out to be efficient as the speakers could easily recognize the words after a quick training before the recordings were made. The recordings were digitized at sampling frequency of 22050 H z and analyzed using the program for speech analysis Praat (www. The acoustic analysis consists of basic measurements of duration, fundamental frequency, formant frequency, etc. The targets of the acoustic analysis include consonants and vowels, and phonological contrasts such as creaky versus modal types of phonation, oral versus nasal quality of vowels, and finally tones. Further details of each measurement are given especially in chapters 2 and 3, and as necessary.  1.4.2 Phonological analysis: Optimality Theory The phonological analysis of Munduruku is based mainly on a database with approximately 2,599 items, divided into: (i) monomorphemic items - 254 noun roots, 217 verb roots, plus 106 morphemes (e.g. affixes, postpositions, particles, etc.); and (ii) non-monomorphemic items - 489 nouns, 506 verbs, plus other 1027 items (e.g. reduplicated and inflected forms, etc.). Many of these were also elicited in larger contexts such as full sentences or small phrases (e.g. possessive constructions, nominalizations, etc.), for which the total number is not available at this point. The phonological processes examined here are formalized within Optimality Theory (Prince and Smolensky 1993; McCarthy and Prince 1993a). Optimality Theory (OT) defends a model for the architecture of grammars based upon five primary concepts. (5)  Principles of OT (from McCarthy and Prince 1994: 3) (a)  Universality - Universal Grammar (UG) provides a set C O N of constraints that are universal and universally present in all grammars.  (b)  Violability -  Constraints are violable; but violation is minimal.  (c)  Ranking -  The constraints of C O N are ranked on a language-particular basis; the notion of minimal violation is defined in terms of this ranking. A grammar is a ranking of the constraint set.  (d)  Inclusiveness - The constraint hierarchy evaluates a set of candidate analyses that are admitted by very general considerations of structural well-formedness.  (e)  Parallelism - Best-satisfaction of the constraint hierarchy is computed over the whole hierarchy and the whole candidate set. There is no serial derivation.  In OT, Universal Grammar provides the formal mechanism from which grammars are constructed; this includes the following. (6)  (a)  C O N - The set of constraints that make up grammars, all of which hierarchically organized in a ranking.  (b)  G E N - A generation function that produces the range of candidates (outputs) for every underlying representation (input).  (c)  E V A L - A n evaluation function composed of individual constraints that comparatively evaluates the outputs with respect to a given constraint in the ranking.  (7) gives an illustration of an Optimality-based approach. The constraints assess each candidate by assigning it a number of violations (indicated by an asterisk ' * ' ) ; i f a candidate violates higher ranked constraints, while others pass, this candidate is eliminated (indicated by an exclamation mark '!'). Evaluation continues until there is only one candidate left; this is the optimal output (marked by ' ° ) . , ^  (7)  ,  Constraint tableau Input  Constraint A  Constraint B  *  *• Cand 1 Cand2 Cand 3  Constraint C  *! *!  The formalization of the phonology of Munduruku within the Optimality Theory is not intended to prove or disprove its efficacy as a model. The primary purpose is to propose an account of the major generalizations that can be made about the phonology of Munduruku, and to construct a grammar based on a system of choices amongst constraints, and their ultimate hierarchical organization. From a theoretical point of view, this brings the language into the debate about the universal versus specific nature of grammars. OT defends that constraints are universal but constraint rankings are language-specific. The study of Munduruku reveals that language-specific constraints must be invoked, though minimally. In particular, it shows that what is specific to this language is by and large rooted in the various diachronic changes that took place in the development of its grammar.  1.4.3 Diach ron ic analysis The study of Munduruku goes beyond its current stage; it also examines how the language has changed over time. This was done through the reconstruction of two prehistoric stages: (i) the earlier stage of Munduruku itself (Pre-Munduruku), and (ii) the common ancestor of both Munduruku and Kuruaya (Proto-Munduruku), as schematized in (8). In some cases, the analysis requires reference to a stage previous to Proto-Munduruku, which I call Pre-Proto-Munduruku. As a convention, pre-proto-forms will be marked with three asterisks (***Pre-PMu), proto-forms with two asterisks (**PMu), and pre-forms with a single asterisk (*Pre-Mu); forms without an asterisk are synchronic. (8)  ***Pre-Proto-Munduruku  **Proto-Munduruku  (Comparative Reconstruction)  (Internal Reconstruction)  Kuruya  Munduruku  The comparative method, one of the most widely used methods in linguistic reconstruction, was the method used in the reconstruction of Proto-Munduruku. It comprises three basic steps (e.g. Fox 1995): (i) compilation of a list with regular correspondences in the lexicon; (ii) establishment of the correspondence sets; and (iii) reconstruction of hypothetical forms.  The notations used throughout this dissertation are as follows (e.g. Hock 1991; Fox 1995). (9)  (a)  x>y  ' x ' changes/develops into ' y '  (b)  y <x  'y' develops out of ' x '  (c)  x>y/z_orx>y/_z  ' x ' changes into ' y ' following or preceding ' z  (d)  x Cy  ' x ' merges with ' y '  (e)  x > x/y  ' x ' splits into ' x ' and ' y '  The historical approach is of singular value. In addition to being the first proposed for the Munduruku  family,  it elucidates many of the  irregularities in the patterns observed  synchronically (see especially chapters 5, 6 and 7).  1.4.4 Overview of the chapters I begin the investigation of Munduruku by examining the phonetic and phonological structures of its segment inventory. Chapter 2 deals with the vowel system; the language has five vowel qualities, / i , e, a, a, o/, and additional contrasts of nasality and creaky voice. From these, four series of vowels result: oral versus nasal modal vowels, and oral versus nasal creaky vowels. The chapter begins by providing an acoustic analysis of the five vowel qualitites, and proceeds to the acoustic properties of the oral-nasal and modal-creaky oppositions. Following Gordon and Ladefoged (2001), a number of properties typically associated with phonation differences were examined; these are: (i) periodicity, (ii) formant frequency ( F l , F2, F3), (iii) fundamental frequency (FO), (iv) overall duration, and (v) spectral tilt. Chapter 3 follows up the investigation of segments by focusing on the consonant inventory, which comprises seventeen consonants: /p, t, tf, k, b, d, 63, s, f, m, n, rj, r, w, y, ?, h/. The phonetic analysis is centered around the acoustic characteristics of intervocalic stops, nasals and laryngeals, with additional investigation of stops and nasals in coda position. The acoustic analysis provides the basis for the following chapters, especially Chapter 4. Chapter 4 deals with several issues in syllable structure and syllabification, in particular: syllabification of sequences of vowels and consonant clusters, the prohibition of the sequence /t/+coronal, restriction on the occurrence of Ixl and fhJ word-initially, and fusion of the aspect suffix {-m} with voiceless stops. Phonotactic restrictions are examined in Chapter 5. Its focal point is on the history of consonants and the restrictions on their distribution. From the eleven restrictions detected 11  synchronically - *wo, *yi, *?y, *dv, *tfi, *c^i, *si, * f 9 , *mi, *ni, and *rji - the synchronic and diachronic analyses show that six of these are historical accidents, *tfi,  *si, *mi, *ni, *rji,  and five are systematic, in the sense that there is evidence from alternations that they are in fact prohibited in the language; these are: *wo, *yi, *?y, *dv, *j*9. Nasal harmony comes next, in Chapter 6. The chapter approaches the process from synchronic and diachronic perspectives, comparing the systems of both Munduruku and Kuruaya. It shows a step-by-step analysis of the changes since Proto-Munduruku, as well as their consequences for the grammar of Munduruku. The chapter concludes with an OT account of the change in nasal harmony. Another phonological process examined is consonant mutation, Chapter 7, which also receives synchronic and diachronic treatments. Munduruku has a peculiar process of voicing alternation by means of which /p, tJ7 are voiced intervocalically, whereas /d/ is devoiced postconsonantally. The historical account comes into the picture to support the hypothesis that the process is a consequence of a sound change that failed to apply regularly to some morphemes. Chapter 8 deals with the phonology of tone and phonation types. Munduruku is a language that makes use of two levels of pitch, low (L) and high (H), to lexically distinguish items. On the surface we find five tonal behaviors: stable versus unstable H tones, active versus inert L tones, and tonal polarity. The account relies on lexical distinctions to explain the five behaviors. Lexical H tones are stable, lexical L tones are active, triggering dissimilation of a following L , and toneless moras surface L but this tone is inert. Unstable H and polar tones are the manifestation of a floating H tone. The analyis then proceeds to explore the tone-creaky voice interaction. Phonologically creaky vowels do not exhibit contrasts of tones, being restricted to a L tone. Creaky voice was in the past (Braun and Crofts 1965) analyzed as a tonal feature, but Chapter 8 shows that tones are independent of phonation oppositions. The major achievement of the study is that the tonal system of the language can be reduced to two distinctive levels, rather than the four previously proposed. Chapter 9 concludes this work by assessing the pros and cons of the analysis proposed in the preceding chapters. It attempts to put together all the constraints (see list in the Appendix) out of which the grammar of the Munduruku phonology can be constructed, its OT-based grammar.  CHAPTER 2  Phonetic and phonological structures of vowels 2.1  Introduction Chapters 2 and 3 are devoted to the phonetic and phonological structures of vowels and  consonants in Munduruku. The first proposal of its segmental phonology is found in Braun and Crofts (1965). They suggest an inventory composed of sixteen consonants and six vowels, with their six nasal counterparts. (1)  1  Munduruku phonemic inventory (adapted from Braun and Crofts 1965) (a) Consonants  Stop  Labial  Alveolar  p  t  Palatal  Velar  Glottal  k  ?  b Affricate  tf  Fricative Nasal  s m  Liquid  j"  h  n  rj  r  Glide  w  y  (b) Vowels Front  Central  Back  High  i"i  ii  uu  Mid  ee  9 9  Low  1  aa  The language also contrasts modal versus creaky vowels; however, Braun and Crofts (1965; Crofts 1973,  1985) analyze nonmodal phonation as a tonal feature. See §2.2 and §2.3 for a proposal of creaky voice as a property of vowels, and especially Chapter 8 for the phonological behavior of creaky phonation and its interaction with tones.  In later work, Crofts (1973) adds two consonants to the inventory, the voiced alveolar stop hi and the palatal nasal /ji/; the vowel inventory is reduced to five oral and five nasal vowels, with lol standing for previous hi, and no contrast between HI and hi. (2)  Munduruku phonemic inventory (adapted from Crofts 1973) (a) Consonants  Stop  Labial  Alveolar  p  t  b  d  Affricate  Palatal  Velar  Glottal  k  ?  tf 43  Fricative Nasal  m  Liquid  s  J"  n  ji  h rj  r  Glide  w  y  (b) Vowels Front  Central  High  i"i  il  Mid  ee  Low  Back  o6 aa  The last modification of the inventory is found in Crofts (1985); the palatal nasal [p.] is treated as an allophone of the velar nasal /rj/ syllable-initially, as in (3). There was no other alteration of the vowel inventory.  (3)  Consonants (adapted from Crofts 1985)  Stop  Labial  Alveolar  P  t  b  d  Palatal  Velar  Glottal  k  ?  Affricate  Fricative Nasal  m  n  Liquid Glide  I  s  r w  I consider the inventory of consonants in detail in Chapter 3, maintaining the number of consonants as proposed in Crofts (1985), and based on phonetic and phonological evidence, dividing them into four classes: stops, fricatives, nasals and approximants. The laryngeals /?, h/ are grouped with approximants, along with /r, w, y/, and affricates are classified as stops. (See Chapter 3 for details.) (4)  Consonant inventory (as proposed here)  Stop  Labial  Alveolar  P  t  b  d s  Fricative Nasal  m  Approximant  w  Palatal  Velar  Glottal  k 43 J"  n ?,h  In this chapter I examine the phonetic and phonological structures of vowels, with special attention to (i) the acoustic properties of vowel qualities, (ii) the oral-nasal and (iii) modalnonmodal oppositions. To this date, there are only two preliminary studies which have approached some phonological aspects of the language from a phonetic point of view (Picanco 2002d, 2004b).  In previous proposals of the vowel system (Braun and Crofts 1965; Crofts 1973, 1985), it is maintained that vowels exhibit an oral-nasal opposition, in addition to tone. Creaky phonation, best known as laryngealization in the literature on Munduruku, was also analyzed as a tonal feature. Here I propose that creaky voice is a contrastive property of vowels, in opposition to modal vowels. (Chapter 8 presents phonological evidence for treating creakiness as a feature of vowels, like nasality.) The term "creaky voice" describes a mode of vocal fold vibration that involves harshness accompanied by lowered pitch (Ladefoged 1971). Articulatorily, creaky voicing is produced by pressing the arytenoid cartilages tightly together, allowing the vocal cords to vibrate only at the other end (Ladefoged 1971). This type of phonation is used to distinguish words such as the following. (5)  (a)  Modal voice  (b)  Creaky voice  wida  'clay'  wida  'jaguar'  ay  'rodent, sp.'  ay  'sloth' (monkey, sp.)  Braun and Crofts (1965: 26; see also Crofts 1973, 1985) propose that creaky phonation is one of the "four accents" - accent 1 is a super-high tone, accent 2 a mid level tone, accent 3 a low level tone, and accent 4 creaky voice. A n alternative analysis has already been proposed by Picanco (1999, 2002b-c; and Chapter 8), who presents phonological evidence against the tonal status of creaky voice, despite its relation to tones. In addition, it is proposed here (see also Picanco 2002b-c; Chapter 8) that there are only two contrastive tones: High (marked with an acute accent, "v") and L o w (generally unmarked, except in phonetic transcriptions where it is marked as "v"). (Tones and tone-creaky voice interaction are taken up in Chapter 8.) Another difference from the previous proposal concerns the quality of the nonlow central vowel, which, as shown in the acoustic analysis below, is a mid vowel [a] (see also Picanco 1997, 1999), not a high vowel [i] as proposed by Braun and Crofts. The chapter is organized as follows. §2.2 shows the distribution of the vowel qualities within the acoustic space, based upon values for the first two formants ( F l , F2). §2.3 takes up the oralnasal opposition, focusing on the effects of nasalization on vowel height. §2.4 turns to the phonetic aspects of another phonological contrast, the modal-nonmodal contrast. Following Gordon and Ladefoged (2001, and others), five acoustic properties were measured for Munduruku data: formant frequencies (§, overall duration (§,  fundamental  frequency (§, periodicity (§, and spectral tilt (§ The findings reveal that 16  fundamental frequency, periodicity and spectral tilt reliably signal the modal-creaky contrast in the language; the results obtained for measurements of fundamental frequency are especially important for the tone-creaky voice interaction.  2.2  Vowel qualities The phonemic vowel system of Munduruku distinguishes five vowel qualities, as illustrated  in (6). There is one vowel in the high front region, [i], three in the mid region - mid front [e], here represented phonemically by Id, mid central [a], and higher mid back [o] - and one low vowel [a]. (6) Vowel qualities (a)  lil  [i]  i-ba-pik  His/her arm is burned.  3-arm-be.burned  (b)  Id  [e]  i-ba-pek  His/her arm is yellow.  3-arm-be.yellow  (c)  lal  [a]  i-bapak  It's visible.  3-be.visible  (d)  lal  [a]  i-ba-pak  His/her arm is red.  (e)  lol  [o]  i-ma-pok  to beat someone up  30b-CAUS-be.hurt  The five-vowel system is greatly expanded by contrasts of nasality and creaky voice, resulting in an inventory with four series of vowels: oral and nasal modal vowels, and oral and nasal creaky vowels (after Picanco 1997). The series are given in (7); creaky voice is indicated by (y) placed under the vowel, and nasalization by a tilde (v) above the vowel. Vowels also contrast for tones, but only modal vowels may surface on a High or L o w tone; creaky vowels are restricted to Low tone. (On tone distinctions, see Chapter 8).  Series of vowels (after Picanco 1997)  (7)  Modal vowels  Creaky vowels  High  1  I I  Mid  ee  Low  I  oo aa  ee  9 9  0 0  aa  The Munduruku vowel system is typologically uncommon due to the gap in the back region, where it lacks the high back vowel /u/. The system is also not fully balanced in the mid region as the mid front vowel [e] is not matched in height by mid back [o]. According to Crothers (1978; see also Greenberg 1966; Maddieson 1984, and others), the most prevalent vowel systems include primarily / i , a, u/, the vowels that are located at the most extreme points in the phonetic space, and then mid vowels. Proposals have been made to explain universal patterns of vowel distribution based upon the principle of vowel dispersion (e.g. Liljencrants and Lindblom 1972; Lindblom 1986). This principle states that vowels tend to be evenly and widely dispersed in the acoustic space available for vowels (see below). Munduruku seems to contradict this principle because its system has a major gap in the high back region and an uneven distribution of mid vowels. Nonetheless, internal evidence suggests that the language is in accordance with Liljencrants and Lindblom's model (and Lindblom 1986) so that the gaps can be accounted for in a principled way.  2.2.1  Vowel dispersion  Languages tend to show a balanced and symmetrical distribution of their vowel systems. In a language with three vowels, the system is typically triangular: / i , a, u/. In one with five vowels, the system includes / i , a, u/ plus two mid vowels, usually /e, o/ (Maddieson 1984). In attempting to explain such patterns, Liljencrants and Lindblom (1972) proposed the principle of vowel dispersion, which states that vowels tend to be widely and evenly dispersed within the acoustic space for vowels. In this sense, a triangular system takes advantage of this acoustic space by distributing its vowels to the most extreme points, as in (8)a. In a five-vowel system, these three points are first filled up and the remaining vowels are then distributed to intermediate points, i.e. the mid region, (8)b.  (8)  (a)  Triangular system i  (b)  u  Five-vowel system i  u  e  o  The dispersion principle predicts optimal systems, with distributions that first include maximally distant points, and then intermediate ones, depending on the number of vowels a language has. But there are exceptions, systems that do not conform to the maximal distribution of their vowels. Munduruku, for example, has three deviations: high back [u], mid front [e] and mid back [o], which are the counterparts for [i], [o], and [e] respectively. (9)  Gaps in the Munduruku vowel system i<  > •  —  >  •  a However, let us suppose, following Disner (in Maddieson 1984), that gaps in defective systems tend to be compensated for, or complemented by, other vowels in the system. Strictly speaking, the vowels available are distributed in a way that brings the system close to one with a symmetrical distribution. Disner reports three major types of complementary vowels (p. 144): (i) a central vowel; (ii) a front rounded or back unrounded vowel; or (iii) a peripheral vowel similar to the missing vowel but lacking a counterpart of equal height and opposite rounding elsewhere in the system. Now suppose that the gaps in Munduruku are compensated for by other vowels in the system. Consider the two possibilities in (10). Compensation in (10)a corresponds to type (iii) above, in which the gap in the mid region is compensated for by the mid back [o], a vowel that is similar to missing [o] and lacks a counterpart of equal height and opposite rounding, i.e. [e]. In this sense, the system has a gap in the high back region, but is fully balanced in the mid region. A second possibility is as in (10)b, which corresponds to types (i) and (iii) together. The gap in the high back region is compensated for by the vowel [o] which is similar to missing [u] (type (iii)), and we can suppose that the mid central vowel [a], although not a perfect counterpart for  [e], balances the system in the mid region (type (i)). In this case, Munduruku is not a counterexample to the dispersion principle. (10)  Compensation of gaps (a)  i < e ^  > 9  Q  (b)  O  i < 6  <r>  a  > o 9  a  Assuming that defective systems tend to have vowels that somehow complement each other, the question is as to whether the distribution of these vowels within the acoustic space reflects their role as complementary vowels. In other words, do complementary vowels tend to be acoustically comparable, either in height or in the front-back dimension, to the missing vowel? In the following section I address this question, examining how Munduruku vowels are distributed within the acoustic space. If compensation is as in (10)a, we can suppose that [o] is closer in height to [e] than to [i]; but i f compensation is as in (10)b, then [o] is closer to [i] than to  [9]  or  [e].  To investigate i f complementary vowels are distributed as close as possible to the vowels they are compensating for, a two-sample t-test analysis was performed, with height (i.e. F l values) as a factor. Unless otherwise indicated, the significance level, here and everywhere in this dissertation, is p<0.05.  2.2.2  The acoustic space of Munduruku vowels  The first step is to determine whether the five-vowel qualities are evenly distributed in the acoustic space, and thus compatible with Liljencrants and Lindblom's model of maximization of contrast in vowel systems. The list of words selected to represent the five vowel qualities is given in (11). Three male speakers were recorded ( A K , JT and E M ) , resulting in a total of 149 tokens, with an average of 29 tokens for each vowel. (11)  Word list for vowel qualities (a)  (b)  ibaplk  'His/her arm is burned.'  ihi  'winter'  ibapek  'His/her arm is yellow.  hem  'It's really a path!'  (c)  (d)  (e)  ibapak  'It's visible.'  dkeaham  'to go up'  ibapak  'His/her arm is red.'  iaham  'to bite something'  imapok  'to beat someone up'  6ho?a  'flute'  The vowels were characterized in terms of the frequencies of the first two formants ( F l , F2); F l characterizes the height, vertical position of the tongue, and F2 the front-back, horizontal position. The formants were measured from wideband spectrograms near the middle of the vowel where they are steady, without the transition effects of consonants. Table 2.1 gives mean values and standard deviations for all vowels. Table 2.1. Fl, F2, F3 mean values and standard deviations for Munduruku vowels.  i e 3  a 0  Mean s.d. Mean s.d. Mean s.d. Mean s.d. Mean s.d.  Fl 327 17 504 40 480 34 705 56 450 39  F2 2467 294 2212 220 1387 165 1481 111 894 106  F3 3109 255 2847 210 2510 226 2436 154 2453 175  Figure 2.1 gives a representative chart of the acoustic space of Munduruku vowel qualities. In the height dimension, all three speakers distinguish three major degrees of openness: one high vowel [i]; three mid vowels - [e], [a], and [o] (which ranges from [o] to [u] and occupies a region between mid and high vowels); and one low vowel [a].  Figure 2.1. Formant plots of Munduruku vowel qualities. (3 speakers: AK, JT, and EM; about 29 tokens for each vowel.)  L  1  i  J  L 1000  The five vowel qualities proposed here diverge from previous proposals only with respect to the quality of the central vowel [a], which Braun and Crofts (1965; also Crofts 1973, 1985) posit as being a high central vowel [i], but as shown in Figure 2.1, and in (12), this vowel is much lower than what is expected for [i], and closer to a mid vowel than to [i], (12)  /e-a/:  t=2.40  s.d.=37.7  d.f.=57  p=0.020  /o-a/: t=3.07  s.d.=37.2  d.f.=55  p=0.0033  /i-a/:  s.d.=27.5  d.f.=57  p<0.0001  t=-21.4  Comparing the mid front vowel /e/ to hi above and to hi, in (13), the difference in height is significantly greater for the pair le-ol (p<0.0001) than for /e-a/ (p=0.020), making it plausible to suggest that [a] compensates as the back counterpart for [e], rather than [o]. There is also a  significant difference in height between [i] and [o], which is higher than that obtained for the pair /e-o/. (13)  /e-o/:  t=5.15  s.d =39.8  d.f.=56  p<0.0001  /i-o/:  t=-15.6  s.d.=30.1  d.f.=56  p<0.000L  This acoustic investigation thus confirms deviations in the distribution of Munduruku vowels within the acoustic space, but does not entirely exclude the possibility that compensation of the gaps by other vowels in the system may affect the acoustic realizations of these vowels. The results for Munduruku suggest that the acoustic distribution of Ii, e, 9, o, a/, with respect to the height dimension, i s i > o > 9 > e > a (">" stands for "higher than"). Taking it this way, the system distributes its vowels first to the most extreme points, and then to intermediate ones. On the phonological side, the five vowels are here distinguished in terms of the features [high, low, back], as in (14). (Below I present some evidence in support to these distinctions.) The back vowel lol is in opposition to lil so both are [+high], but lol is also [+round], in contrast to /9, al which are [+back, -round]; hi balances the system in the mid region, in opposition to hi, which is [-back]. Even though the high back vowel should be represented by hi, since it is classified as [+high], I will continue to represent it as lol in order to be consistent with the Munduruku orthography, and in which " u " stands for hi. (14)  Vowel features -back  +back -round  +high  i  -high, -low  e  -l-low  +round 0  9  a  There is some evidence suggesting that missing hi is compensated for by the mid back vowel lol, justifying stipulation of [+high] for this vowel. In normal speech, lol varies freely between [0] and [u], as seen above, and in Figure 2.2 below. There are some exceptions to this variation: okpfojt *okp[u]t 'my son (male speech)', itfojp *it[u]p 'her husband', and fojp *[u]p 'arrow'. The absence of the  alternation [o~u] in these cases might suggest a contrast, say,  between lul and lol in the back region, in that lul varies between [u] and [0], and lol is always  [o]. I believe that a contrast /u/ versus lol brings about some complications for the vowel system of the language. First, whether or not a back vowel [o] alternates with [u], there is considerable acoustic overlap in their realizations, as Figure 2.2 shows. Figure 2.2. A comparison between [o] and [o~u]. Words okpot 'my son (male speech)', itop 'her husband', imapok 'to beat s.o. up', irore 'It's loose' and aro 'parrot'. (Legend: "o" = nonalternating [o] (16 tokens), "Co~u" = [o~u] alternation in an open syllable (18 tokens), and "Co~uC" = [o~u] alternation in a closed syllable (14 tokens))  F2 1800  1200  I  '1  L  ""vv.vr  600  1  400  C  600 • o  A Co-u  « Co~uC  Second, the number of items containing a non-alternating lol is extremely small; thus far the corpus includes only the three cases cited above. A n d third, the system is symmetrical with respect to the contrasts oral-nasal and modal-nonmodal. If we make a distinction between IvJ and lol, the inventory would have six oral modal vowels, / i , e, 9, a, o, u/, but only five nasal modal vowels, ft, e, §, a, QV, five oral creaky vowels,ft,e, 9, a, u/, and five nasal creaky vowels,ft,e, §, a, u/, as there are no cases attested of a nonalternating lol in these three series. Given all the above, it is of no detriment to posit only the back vowel lol, specified for the features [+high, +round], as in (14) above. A phonological fact that supports the analysis is illustrated in (15). In combinations glide + vowel, the language disallows cooccurrence of lol with the glide /w/ in the same syllable, a prohibition that is parallel to the restriction on combinations of a high front vowel ft/ with the palatal glide lyl. (See Chapter 4 for an account of the restrictions in combinations glide + vowel.)  (15)  Combinations glide + vowel (a)  ka.wi  'clay'  (b)  ta.we  'monkey'  '2pl.CoRef  (c)  1T13.W9  'bird, sp.'  y_9.?9k  'his/her belly'  (d)  wa.?a  'my head'  y_a.?a  'his/her head'  (e)  "wo  yo.borj  'It's big'  Another observation of interest concerns the mismatch in the mid region, where the mid front vowel [e] lacks a back counterpart [o], a vowel with corresponding height and opposite backness for [e]. One possible explanation is that a diachronic change neutralized height distinctions in the back region. Kuruaya, the other language of the Munduruku family, has a system with six qualities of oral vowels: Ii, e, i , a, o, ol (Costa 1998; Picanco 2003c; and also Chapter 5), as shown in (16)a. Most importantly, Kuruaya makes a contrast between lol and hi. Comparing the two systems, a hypothesis is that the system presently observed in Munduruku may have developed out of a system fully balanced in the mid region, as it is in Kuruaya. Two major changes seem to have taken place: one merged hi and hi, neutralizing the contrast between the two back vowels and leaving hi without a counterpart in Munduruku, (16)b; and another lowered HI to hi, perhaps to balance the system in the mid region. (Chapter 5 gives more details of historical changes in the vowel inventory). (16)  (a)  Kuruaya i  i  (b)  Munduruku  o  i  o  e  •  o 9  •  On the basis of these observations, we can argue that the vowel system of Munduruku achieves Liljencrants and Lindblom's predictions of maximal dispersion, though it does so by means of compensation of gaps. Compensation does not necessarily affect the acoustic distribution of vowels, as seen above, but can be expressed phonologically, as in the case of lol which patterns with lil in Munduruku, or historically, as the change that lowered HI to hi to balance the system in the mid region. The observation made in Maddieson (1984: 154) proves to  be correct in that the majority of vowel systems that are likely to be exceptions to the dispersion theory "(...) tend toward a balanced distribution of vowels in the available space, either by complementation with a vowel of unexpected quality or by a displacement toward the gap of some or all vowels in the system." Munduruku vowel system supports this assumption. The next step in the analysis is to investigate other contrasts in the language. So far I have considered only vowels produced with a particular state of the glottis, normal voice. There is another set which requires a different adjustment of the glottis: the set of creaky vowels. Both modal and creaky vowels exhibit an opposition between oral and nasal vowels so that the inventory is quite symmetrical, although nasal creaky vowels are much less frequent than nasal modal vowels. In the following sections I examine the acoustic properties of modal-creaky and oral-nasal contrasts, beginning with modal vowels and the oral-nasal opposition. The creakymodal contrast is examined in §2.4. 2.3  Modal vowels: the oral-nasal contrast Nasalization is a contrastive properly of Munduruku vowels; there are nasal counterparts to  all five oral vowels, irrespective of phonation differences or the environment at which they occur: open and closed syllables, or adjacent to a nasal consonant, in particular, a nasal in coda position. Illustrations of the oral-nasal opposition in modal vowels are given in (17). The contrast in the environment of nasal consonants is crucially achieved at the expense of a further adjustment of velic movement. A nasal consonant is plain only i f preceded by a nasal vowel; for example, sequences /vm, vn, vrj/ result in phonetic sequences such as [vm, vn, vrj]. If preceded by an oral vowel, for example /vm, vn, vrj/, the nasal is partially oralized, resulting in phonetic sequences such as [v m, v n , v rj]. A hypothesis is that the reinforcement of the oral-nasal b  d  8  contrast in the environment of nasal consonants causes lowering of the velum to be delayed until after the oral closure is complete (see Chapter 3). In this sense, preoralization is a strategy used by speakers to preserve a phonological contrast. The oral-nasal contrast for all vowels is 2  illustrated below:  2  in (17), /e-e/ in (18), /a-3/ in (19), /a-a/ in (20), and lo-ol in (21).  These examples also illustrate leftward nasal spread to [+sonorant] segments, triggered by a nasal vowel.  Nasal harmony is examined in Chapter 6, and preoralization of nasal consonants in Chapter 3.  (17)  (a)  (b)  ihi  [ihi]  'winter'  alhl  [aihl]  'mother'  kaji  [kaji]  'sun/moon' 'pepper (tree)'  (c)  [yopti rj]  'smoke from a stick'  [ifir)?a]  'pot'  hem  [he m]  'It's really a path!' (Emphatic)  t-ae-he-m  [taefilm]  'to choose'  y-op-tirj  g  3-CL-smoke ifirj-?a pot-CL  (18)  (a)  b  3 O-choose-RED-EVIPRF  (b)  [edi rj]  'tobacco smoke'  edi  [g di ~ Idl]  'kind of net (for fishing)'  ?e  [?e]  'AUX(iliar)'  ?e  [71]  'mortar'  cfee-a-ha-m  [(feeaha m]  'to go up'  e-dirj  s  tobacco-smoke  (c)  (19)  (a)  n  CoRef.Poss-up-RED-MPRF ara  [ara]  'maracana bird'  (b)  i-way  [rway]  'to wash s.t'  [rway]  'to shoot s.t. (with an arrow)'  [6r5 n]  'I will be white.'  [onan]  'my feces'  [iaha m]  'to bite'  3 Ob-wash i-way 3 Ob-shoot  (c)  o-ra-n  d  lSu-be.white-EVIPRF 0- n§n 1- feces  (20)  (a)  i-a-ha-m  b  30b-bite-RED-IMPRF  (b)  awa  [awa]  'a baby girl'  awawa  [awawa]  'grandmother'  <^e-wa-wa-wa-m  [^ewawawam] 'to scream'  CoRef.Poss-call-RED-RED-IMPRF  (c)  afiwa-?a  [aj*iwa?a]  'fish, sp.'  [ijiwa?a]  'termite'  [6h6?a]  'flute'  [I6fi6?a]  'head of your domestic animal'  fish-CL ifiwa-?a termite-CL  (21)  (a)  6h6-?a flute-CL e-oho-?a 2-domestic.animal-CL  (b)  i-?o-m  [i?6 m]  'to eat s.t.'  fiom]  'to enter'  deko  [deko]  'coata monkey'  e-ko[?]  [ek6]  'your tongue'  b  30b-eat-IMPRF i-6m 3Su-enter  (c)  2-tongue  The hypothesis that preoralization is a strategy to maintain the oral-nasal contrast on vowels in the environment of nasal codas is supported by neutralization of the contrast in other environments. First, vowels are always nasalized between nasal consonants (e.g. nor/?a 'flea', inSn 'to sew s.t.', imorj 'to put s.t.'). In addition, vowels can be nasalized preceding heterosyllabic nasals (e.g. a/JTJma 'fish', [kdjija 'sugar-cane'), or optionally nasalized after a nasal onset (e.g., [mo]di ~ [msfjdi 'rodent, sp.', ajifmaj  ~ aji[ma] 'fish'), except when the vowel is contrastively nasal, in which case  nasalization is obligatory (e.g. noba[no] 'gun', we[nSy] 'Brazil nut'). Most importantly, oral vowels that cause preoralization also assimilate nasality in nasal harmony; compare tei-m [tel m] 'future price' and tei-Pit [tgi?it] 'It's cheap'. (Nasal harmony is b  examined in Chapter 5.) Nasal vowels occur less frequently in syllables closed by a voiceless stop (e.g. sSsSt 'monkey, sp.', naPorek 'lizard'), in contrast with nasal vowels followed by a nasal consonant (e.g. sSsSn 'to be ashamed, embarrassed', itaben 'S/he's alert'), and oral vowels followed by stops and nasals (e.g., irM 'It's white', iwek 'It's torn', imdsSn 'to clean s.t.', yakren 'She's pregnant').  2.3.1 Acoustic effects of nasalization on vowel height There is a trend across languages for nasalized vowels to exhibit height differences relative to the corresponding oral vowels (Baht 1975; Ohala 1974, 1975; Wright 1975, 1986; Beddor 1983; Maddieson 1984, and others). Beddor's (1983) study provides an excellent overview of, as  well as objections to, previous proposals about the effects of nasalization on vowel height. It reveals significant cross-linguistic tendencies concerning the direction of such changes. Her findings establish five cross linguistic patterns in the height dimension, shown in (22). In general, high vowels lower and low vowels raise due to nasalization. Beddor distinguishes between non-contextual, contrastive, nasalization and contextual-dependent nasalization, which refers to nasalization of a vowel adjacent to a nasal consonant, relevant to show systematic changes in the case of mid vowels. (22)  Cross-linguistic patterns (Beddor 1983: 99) (a)  High nasal vowels lower.  (b)  L o w nasal vowels raise.  (c)  M i d non-contextual nasal vowels lower.  (d)  M i d back contextual nasal vowels raise.  (e)  M i d front contextual nasal vowels lower, except if nasalization affects both front and back vowel height, in which case both front and back nasal vowels raise.  Acoustic data relating the effects of nasalization on Munduruku vowels were examined for 3 speakers ( A K , JT and E M ) . The focus was on vowels that exhibit phonologically contrastive nasalization, as in (23), including oral and nasal vowels contiguous to nasal consonants. A list with 20 words with contrastive nasalization was recorded. (23)  Words for comparing oral and nasal vowels in Munduruku. i/1  e/e  3 /3  a/a  ihi  winter  alhi  'mother, V o c '  kaji  'sun/moon'  hem  'path, Emphatic'  taehem  'to choose'  edlrj  'tobacco smoke'  edi  'kind of net for fishing'  cfeedhsm  'to go up'  oran  ' I ' l l be white'  onsn  'my feces'  iaham  'to bite'  awa  'baby girl'  afiwa?a  'fish, sp.'  ij1wa?a  'termite'  'pepper tree'  'maracana bird'  o/5  oho?a  'flute'  eoho?a  'head of your domestic animal'  deko  'coata monkey'  eko  'your tongue'  Cross-linguistic patterns indicate that under the effects of nasalization, high and mid vowels lower and low vowels raise. Ohala (1974, 1975) attributes the changes in vowel height to the perturbation nasalization causes in the frequency of the first formant. Wright (1986: 49) observes that because "vowel height is inversely related to first-formant frequency, it would be expected that vowel nasalization would cause high and mid vowels to lower auditorily and the low vowels either to lower or to raise depending on the speaker and vowel." There seems to be a general observation that nasal coupling manifests itself differently depending on the vowel quality, and that much of the effect nasalization has on these vowels is correlated with changes in the frequency of the first formant. The reason, as Wright (1986) and Beddor (1986) state, is because nasalization introduces a nasal formant with fixed and relatively low frequency, typically close toFl. Acoustic data of oral-nasal pairs in Munduruku indicate that nasalization has a strong effect in the quality of some vowels. This is shown in Figure 2.3.  Figure 2.3. Fl vs. F2 plots of Munduruku oral-nasal vowels. In each pair of vowels, black stands for oral vowels and white for nasal vowels. Speakers: AK, JT and EM (Average of 29 tokens for each vowel) F2  3000  2400  I  '1  1  1  ;  600  ° V \°  *• . .  0° o  — ^ __  0  / Y  / • / •  *•<> o  «  »•  yt  */ ^ ' ' 'A  90  * •  >  /  ' A  A  400  *  /  7*~\  o  • V  /«<> \  /  f* *• •  /oo  ! •  1200  •  1  '  1800  *»'  i i V  * A  * A  A  A  /  °  • fo * I 0 *•<>  •  /  /  /  • / o  ^ A _ _ - - ^  — — - 5 — • \  /  o  */ o 1  / 1  {  •° •  o  o  •  o  •  V 0 ^  o  1  o  800  •  o  1000  Overall, all vowels are affected, except for lal and lol, for which oral and nasal qualities are very similar. (24)  la-al: t=0.187  s.d.=67.6  d.f.=59  p=0.85  lb-ol: t=-1.46  s.d.=100  d.f.=59  p=0.15  There is a lowering effect of high and mid front nasal vowels; significantly. The oral-nasal pair la-al is also different: (25)  n-il:  t=-5.51  s.d.=39.4  d.f.=62  p<0.0001  /e-e/:  t=-7.77  s.d.=72  d.f.=59  p<0.0001  s.d.=56.1  d.f.=55  p=0.0077  la-al: t=-2.77  and /e-e/ differ  If we suppose that these results represent the Munduruku community as a whole, the main effect nasalization has on vowel height is lowering of the high, mid front and central vowels. Another observation is the effect of nasalization on the front-back dimension, determined by F2. Oral vowels are more likely to be concentrated in a small space, with less variation of formant frequencies, in particular F l and F2; nasal vowels, on the other hand, exhibit greater variations. The means and standard deviations of formant frequencies are given in Table 2.2. Table 2.2. Mean values (in Hz) and standard deviations of formant frequencies of oral and nasal vowels.  i  e  3  a  0  Fl  F2  F3  338  2386  3064  26  278  191  508  2223  2898  46  243  233  493  1513  2566  48  124  269  727  1502  2496  63  124  154  427  973  2448  34  140  188  1  e  5  a  0  Fl  F2  F3  392  2492  3231  50  302  214  652  2319  3059  92  169  175  534  1513  2372  62  178  315  724  1461  2455  70  188  302  464  907  2300  138  145  327  Another contrastive property of Munduruku vowels is the modal-creaky contrast. To my knowledge, Munduruku is the only Tupi language that has a contrast between nonmodal and modal voice. In the following section I shall consider a number of acoustic properties that are most likely to signal this opposition in the language, with the expectation that this analysis will contribute to the interface relation between phonetics and the phonological behavior of creaky vowels.  2.4  Phonation types The different ways the vocal cords vibrate, or do not vibrate at all, create a variety of  phonation types (Catford 1977; Ladefoged 1971). A s suggested by Ladefoged (1971; see also Catford 1964), these various glottal states may be represented in the form of a phonation continuum, "[...] defined in terms of the aperture between the arytenoid cartilages, ranging from  voiceless (furthest apart), through breathy voiced, to regular, modal voicing, and then through creaky voice to glottal closure (closest together)." (Gordon and Ladefoged 2001: 384). This continuum is schematically represented in (26). (26)  Continuum of phonation types (from Gordon and Ladefoged 2001: 384) Most open ^ Phonation type  Voiceless  ^ Most closed Breathy  Modal  Creaky  Glottal closure  During voiceless sounds the vocal cords are apart and do not vibrate, but are still close enough to allow for some turbulence in the airstream, for example, for the production of [h] (Ladefoged 1971). The voiceless-voiced opposition occurs in the majority of languages, particularly to make distinctions among consonants (e.g. voiced Ihl versus voiceless /p/). Languages may also exploit other points on the continuum to manifest linguistic oppositions. Hmong (Huffman 1987) and Gujarati (Fisher-Jorgensen 1967) make a contrast between breathy and modal voice, Jalapa Mazatec (Kirk, Ladefoged and Ladefoged 1993) contrasts creaky, breathy and modal voice, and Kedang (Samely 1990) contrasts creaky and modal voice. Munduruku contrasts creaky and modal voice (Braun and Crofts 1965; Crofts 1973, 1985; Picanco 1997). Modal is the normal way of vibration. The vocal folds are adducted along their full length and with a suitable degree of tension as to allow vibration in a rhythmic manner, i.e. opening and closing at regular intervals of time. In breathy phonation, or murmur, the vocal cords are vibrating but never completely closed; as a result, a significant amount of airflow can still escape through the glottis causing turbulence (Catford 1977; Laver 1980). In creaky phonation, often called 'laryngealization' or 'vocal fry', the arytenoid cartilages are held tightly together allowing only the front portion of the vocal cords to vibrate. The result is a sound produced as 'a rapid series of taps' (Catford 1964: 32) and at a very low frequency (Ladefoged 1971; Laver 1980). And finally, glottal closures, characterized by holding together the vocal folds so as to inhibit vibration. However, as Ladefoged and Maddieson (1996: 75) point out, glottal stops are often realized in the form of creaky voice on surrounding vowels, especially intervocalically, in the majority of languages, and not as a complete closure.  2.4.1 The modal-creaky voice opposition Munduruku contrasts on vowels two modes of laryngeal vibration: modal and creaky voice. The following words illustrate oral vowels with modal and creaky types of phonation. The pair /i-j/ is illustrated in (27), /e-e/ in (28), /a-a/ in (29), /a-a/ in (30), and /o-o/ in (31). (27)  (a)  (fee-Ji-fi-m  'to spit'  CoRef.Poss-spit-RED-IMPRF (b)  ^e-Ji-fi-fi-m  'to tremble'  CoRef.Poss-tremble-RED-RED-IMPRF  (28)  (c)  wida  'kind of clay'  (d)  wida  'jaguar'  (a)  dje-de-de-m  'to play an instrument'  CoRef.Poss-play-RED-IMPRF (b)  i-de-de-m  'to grate'  30b-grate-RED-IMPRF  (c)  i-rore  'It's loose.'  3Su-be.loose (d)  i-ero-re  'It's ripe.'  3Su-be.ripe-RED.Have  (29)  (a)  i-a  'It's light'  3Su-be.light (b)  i-ya 3Su-be.salty  'It's salty.'  (c)  o-i-tfak  'It broke.'  3Su-i-be.broken (d)  ka-cfeak  'to be cold'  thing-be.cold  (30)  (31)  (a)  dat  'scorpion'  (b)  dat  'vomit (N)'  (c)  ay  'rodent, sp.'  (d)  ay  'sloth (monkey, sp.)'  (a)  i-rore  'It's loose.'  3Su-be.loose (b)  i-ero-re  'It's ripe.'  3Su-be.ripe-PvED.Have  Creaky vowels also contrast for nasalization, but these are rare in the language. The following are some examples of creaky nasal vowels. (32)  Nasal creaky vowels (a)  kara-blrj-birj  'a hole with water'  ?-?-RED  (b)  i-ma-kenerj-kenerj  'to make little cuts in s.t. (e.g. fish)'  30b-CAUS-cut-RED  (c)  i-k§y  'hole of s.t.'  3-hole  (d)  i-ma-pa 30b-CAUS-be.hurt(?)  'to beat s.t./s.o. up'  (e)  i-kara-bon-bon  'It's curly.'  3Su-?-be.curly-RED  Because creakiness is primarily characterized by lowered pitch in Munduruku (see §, Braun and Crofts (1965; Crofts 1973, 1985) proposed that creaky voice is a tonal feature. According to them, creaky voice represents the lowest pitch level, or 'accent 4'. Despite its interaction with tones, this study (see Chapter 8) differs from Braun and Crofts in treating creaky voice as a property of vowels, like nasality, not as a tonal feature. (As will see in Chapter 8, creaky and L-tone modal vowels form a single class phonologically.) No tonal contrasts are observed in creaky voice: i f a vowel is creaky, or [+constricted glottis], then it has L o w tone; modal vowels, on the other hand, may surface on a L o w or High tone. Lowered pitch is in fact one of the primary manifestations of creaky voicing in the language. A s already noticed by Braun and Crofts (1965), and confirmed later by Picanco (2002d), creakiness is usually accompanied by a pitch lower than that of modal vowels. I will not deal with the phonology of creaky voice or its interaction with tones in this chapter. First I want to investigate its phonetic aspects as a contrastive feature o f vowels to determine what acoustic cues better signal the creaky-modal contrast in Munduruku; all of which have been proven to be good indicators of phonation differences in a number of languages (e.g. Ladefoged, Maddieson and Jackson 1988; Kirk et al. 1993; Gordon and Ladefoged 2001). Because of the rare occurrence of nasal creaky vowels in the language, and absence of good pairs to be contrasted, I will limit myself to examining the creaky-modal opposition in oral vowels only.  2.4.2 Acoustic properties of the creaky-modal contrast Creaky vowels are not heavily creaky in Munduruku. The waveforms in Figure 2.4 are 3  illustrations of the contrast; they contain a modal vowel [a] (in the top waveform) in the word dat 'scorpion', and a creaky vowel [a] in dgt 'vomit'. The waveforms show the full length of the vowel, from the first complete or nearly complete cycle until the last complete or nearly complete pulse. Note that the duration is similar for both vowels: creaky [a] is only 2 ms longer  3  1 must point out that the degree of constriction varies according to the rate of speech; in a sentence, speakers  tend to produce creaky vowels with less constriction, but if the same words are pronounced in isolation, the vowels may be heavily creaky.  than modal [a]. They differ, however, with respect to the number of glottal pulses realized for each vowel: 11 pulses for the modal, and 9 for the creaky vowel. Another observation is that adjacent glottal pulses occur at more or less regular intervals of time in both cases, although individual pulses are longer in duration for [a]: approximately 8 ms for [a] versus 9 ms for [a], and getting longer towards the end as creakiness increases. Creaky voice is in this sense manifested as a gradual fall in pitch. Figure 2.4. Waveforms illustrating modal [a] and creaky [a]. Words: dat 'scorpion' and dat 'vomit', respectively.  In y  0  T i m e  9 7  (m s )  Lowered fundamental frequency, glottal pulses with longer duration, and variation between adjacent glottal pulses are some cues associated with the creaky-modal voice distinction in Munduruku. According to Ladefoged, Maddieson and Jackson (1988), Kirk et al. (1993), and Gordon and Ladefoged (2001), the various laryngeal configurations produce certain acoustic outcomes that can be used to quantify the nonmodal-modal opposition in a given language. Gordon and Ladefoged suggest a number of acoustic properties that are typically associated with phonation types, five of which were examined for creaky-modal pairs of vowels in Munduruku: (i) formant frequencies, (ii) overall duration, (iii) fundamental frequency, (iv) periodicity, and (v) spectral tilt.  4  Formant frequencies. Phonation differences may imply differences in formant values. The reason is that nonmodal phonation may involve a raising of the larynx, as Kirk et al. (1993) suggest for creaky voice in Jalapa Mazatec, or lowering, as Thongkum (1985) suggests for breathy phonation in Chong. Raising of the larynx affects frequency values by shortening the oral tract and so raising F l ; lowering would have the opposite effect. 4  Gordon and Ladefoged include another property, intensity, but this was not measured for Munduruku data  because of technical difficulties.  Duration. Nonmodal vowels may have longer duration (e.g. Kirk et al. 1993). A related observation is that nonmodal phonation is often confined to a certain portion of the vowel rather than throughout its length, sometimes lasting for less than half of the vowel (Silverman et al. 1995; Blankeship 2002). Silverman (1995) proposes that nonmodal phonation obscures the perception of other phonological contrasts, for instance the realization of tonal contrasts; with sufficient duration, and by confining nonmodal phonation to only a portion of the vowel, other simultaneous phonological contrasts are also recoverable. Fundamental frequency. In general, nonmodal phonation is usually correlated with lowered fundamental frequency, whether or not the language uses tones contrastively. This is true for Munduruku creaky vowels where creakiness can only be realized in a low tone vowel, or is lost otherwise (Picanco 2004b, and Chapter 8). Periodicity. Creaky voice is characteristically marked by aperiodic glottal pulses. Periodicity refers to measurements of the variation in the duration of consecutive glottal pulses, or jitter. Compared to modal voice, creakiness exhibits more jitter. It is possible to quantify this variation by measuring the duration of adjacent pulses in a given portion of the vowel. Spectral tilt. Creaky voice can also be characterized in terms of spectral tilt. Measures of spectral tilt compare the differences between the amplitude of the fundamental (FO) and that of higher harmonics, in particular h2 (F0-h2) and the harmonic closest to F l (F0-F1). In creaky vowels the level of energy is greater in the higher harmonics, i.e. the second harmonic has slightly greater amplitude than the fundamental, and the harmonic closest to F l has much greater amplitude. For modal vowels the second harmonic has less amplitude than the fundamental, and the harmonic closest to F l has similar amplitude. Procedures Similar measurements were conducted in a preliminary study (Picanco 2002d) where a minimal pair was examined, dat 'scorpion' and dgt 'vomit', as produced by two male speakers. The results showed that fundamental frequency is consistently used by speakers to distinguish creaky from modal vowels. Despite the differences in the degree of creakiness, both speakers produced the creaky vowel [aj with fundamental frequency lower than modal [a]. Jitter and periodicity also reliably identified differences in phonation types; formant frequencies and overall duration overlapped, and did not seem to be a strong indication of the contrast, but new measurements show that there is an effect on formant values, as we will see next.  For this study, the five pairs of modal and creaky vowels, /i-j/, /e-e/, /a-a/, /a-a/ and /o-o/, were recorded by two male speakers, A K and JT, and measured for the five properties described above. Each vowel token was divided into two halves and the measurements were obtained from the second half, given that creaky voice is mostly localized to this portion of the vowel. The formants were measured as close as possible to the middle of the vowel (between 50 and 75%), but still at a point where creakiness was relatively salient, in order to avoid consonant transition effects on both ends. Duration was measured from the first complete or near complete pulse until the last complete or near complete one. Fundamental frequency (FO) was measured at five different points of the vowel, and each point was 1/6 apart from the other; for example, i f the duration of a vowel is 150 ms, these were divided by 6 (150/6=25), so FO was measured every 25 ms of the vowel: 25, 50, 75, 100, and 125 ms. This way we can obtain the drop in pitch that is characteristic of creaky vowels. For periodicity, the last 6-10 successive pulses (excluding the very last pulse) were measured for both modal and creaky vowels; shorter vowels, e.g. / i - i / , had only 6 pulses measured, whereas /a-a/ had about 10. Spectral tilt was also measured at the same point the formant values were obtained. The following words were used for the creaky-modal contrast. (33)  The Munduruku sample words. Creaky  Modal  'to tremble'  i-i  q^efifim  'to spit'  e-e  djededem  'to play'  idedem  'to grate'  9-3  ottjak  'It broke.'  kacjjak  'to be cold'  a-a  wida  'clay'  wida.  'jaguar'  0-0  irore  'It's loose.'  ierore  'It's ripe.'  2.4.3 Results and discussion Formant frequencies The effect of creaky phonation on vowel height, determined by F l , varies depending on the vowel. Table 2.3 provides the overall means and standard deviations for the modal-creaky pairs of vowels; the results are based on the measurements of 100 tokens produced by the two speakers, A K and JT, with a total 10 tokens for each vowel.  Table 2.3. Mean values and standard deviations of the formant frequencies (in Hz) of modal and creaky vowels. Modal  Fl  F2  F3  Creaky  Fl  F2  F3  i  345  2430  2998  i  361  2387  2919  20  138  195  18  153  236  505  2289  3016  568  2232  3038  40  95  131  36  127  150  483  2015  2605  447  2079  2856  34  104  168  30  258  319  815  1629  2701  811  1672  2768  74  111  32  59  153  162  395  966  2499  435  949  2384  17  48  202  23  83  249  e  9  a  0  e  9  a  0  The vowel chart in Figure 2.5 compares the distribution of modal-creaky pairs for all tokens.  5  The mid central pair [s-g] in the words oi'tfsk 'It broke' and katfiak 'to be cold' exhibits a large coarticulatory  effect of the preceding palatal consonants [tf] and [c&J, causing them to be realized closer to front vowels, i.e. with higher F2 values, in the words examined; recall from §2.2.2 that [a] has a F2 mean value of 1387 Hz.  Figure 2.5. Fl vs. F2 plots of Munduruku creaky-modal vowels. For each group, modal vowels are in black and creaky vowels in white. Speakers: AK, JT (10 tokens for each vowel)  600  £  •L 1000  The effect of creakiness is significant only in the pairs /e-e/ and /o-o/, and there is a small difference in the pair /a-a/. (34)  /e-e/:  t=-3.52  s.d.=40.2  d.f.=18  p=0.0025  /o-o/: t=-4.12  s.d.=21.5  d.f=18  p=0.0006  /a-a/:  s.d.=33.6  d.f.=18  p=0.025  t=2.44  Creaky voice did not reach significance in the pairs /a-a/ and /i-j/. (35)  /a-a/:  t=0.130  s.d.=70.4  d.f.=18  p=0.90  /i-j/:  t=-1.78  s.d.=20  d.f=18  p=0.092 Overall duration Duration means are given in the following table. Table 2.4. Duration means and standard deviations (in ms) for Munduruku modal-creaky vowels. (2 speakers: AK and JT; 10 tokens for each vowel.) i  i  e  e  9  3  a  a  O  Q  Mean 66 69 89 95 67 79 137 153 232 198 s.d.  11 10 10 8  8  11 32  19  16  25  The differences in duration between modal and creaky are significant for the modal-creaky pairs /o-o/ and /a-a/: (36)  /o-o/: t=3.42  s.d.=22.1  d.f.=18  p=0.0030  /a-a/:  s.d.=10.2  d.f=18  p=0.023  t=-2.48  There were no significant differences in other modal-creaky pairs. (37)  /i-j/:  t=-0.634  s.d =11.1  d.f=18  p=0.53  /e-e/:  t=-1.41  s.d.=9.66  d.f=18  p=0.17  /a-a/:  t=-1.35  s.d.=27.4  d.f.=18  p=0.19 Fundamental frequency The measures of fundamental frequency showed that the creaky-modal distinction is primarily associated with differences in pitch: creaky phonation is manifested by lowered fundamental frequency (FO) relative to modal vowels. The results obtained for this property are based on the measures of 150 tokens (2 speakers): 50 tokens for L-tone vowels (=10 tokens for each vowel quality), 50 tokens for H-tone vowels, and 50 for creaky vowels. A s previously 6  explained, five points were measured for each token, chosen by dividing the overall duration of  6  The following words with High-tone vowels were used for the comparison with corresponding Low-tone and  creaky vowels: [i]: ipojlm 'It'll be heavy'; [e]: dapsem 'deer'; [5]: oksk 'Take care of me.'; [a]: wita 'grater'; [6]: aro 'parrot'.  the vowel by 6; for instance, i f the duration of the vowel was 150 ms, each point measured was 25 ms apart from the other: 25-50-75-100-125 ms, excluding thus the first and last 25 ms of the vowel. F0 contours for each token are plotted separately for A K and JT in the figures below (from Figure 2.6 to Figure 2.10). Note that the speakers realize F0 contours similarly for both types of phonation: there is a lowering effect on F0 towards the end of the vowel, with the exceptions of H-toned lol that shows a raising effect, and lal that is more or less steady. For A K , H-tones are distinctly higher in all cases; L-tone and creaky vowels, on the other hand, overlap depending on the vowel. This overlap is mostly observed in the pair / i - i / , as shown in Figure 2.6. (The results for significance tests are given below.) For other vowels, A K produces L-tone and creaky voicing with some overlap in the pairs /e-e/ and /a-a/, shown in Figure 2.7 and Figure 2.8 respectively, but the tendency is for creaky vowels to be realized with a pitch level lower than that of L-tone modal vowels. The observations for the second speaker (JT) are similar. H-tones have higher F0, although the speaker produces them only a few Hertz higher than L-tones. The results are also comparable with respect to the realization of F0 in the different types of phonation. For both speakers, the low vowel lal shows the lowest values, and the high front vowel the highest ones. Figure 2.6. Graphs showing F0 realization for H-tone, L-tone and creaky vowels. Dashed lines are for H-tone hi, solid lines for L-tone lil, and dotted lines for creaky voice Iii.  Figure 2.7. Graphs showing FO realization for H-tone, L-tone and creaky vowels. Dashed lines are for H-tone Id, solid lines for L-tone /e/, and dotted lines for creaky voice Id.  Figure 2.9. Graphs showing FO realization for H-tone, L-tone and creaky vowels. Dashed lines are for H-tone /a/, solid lines for L-tone /a/, and dotted lines for creaky voice /a/.  AK  JT  Figure 2.10. Graphs showing FO realization for H-tone, L-tone and creaky vowels. Dashed lines are for H-tone lol, solid lines for L-tone lol, and dotted lines for creaky voice lol.  AK  JT  Figure 2.11 gives the mean FO values for vowels with differences in tone and mode of phonation. The five points measured over the vowel were averaged to obtain a representative FO value for each vowel token. These values were then averaged and one single FO value representing each combination of vowel-tone and vowel-creaky voice was obtained. These are shown in the figure.  Figure 2.11. Mean FO values for Munduruku vowels with H-tone, L-tone and creaky voice. Speakers: AK, JT (10 tokens for each vowel).  The significance tests were conducted separately for both speakers, because of the differences in FO ranges of tonal realizations. H-tone, for example, is produced at about 130 Hz for A K , but below that for JT. Only the results for A K are presented here, because the small number of tokens recorded by JT (4 tokens for each vowel) were not sufficient to obtain satisfactory results. (See Chapter 8 for further details on tone distinctions.) As previously observed, there is considerable FO overlap in the pair /i-i/; the results in the table show that the two vowels exhibit no distinction with respect to pitch (p=0.091). We find a very small difference in the pair /a-§/ (p=0.048), but not as significant as the results obtained for other vowels, for which the effect of creaky voice in the realization of pitch is highly significant. Table 2.5. Results for significance tests: L-tone versus creaky voice. (Speaker: AK)  AK  i-j  e-e  9-3  a-a  o-o  t  1.87  3.85  2.26  8.59  7.70  s.d.  4.17  3.90  8.19  7.76  3.04  d.f.  10  10  10  10  10  P  0.091  0.0032  0.048  <0.0001  <0.0001  In contrast, the table below shows that L-tone and H-tone vowels are clearly distinct. In general, the probability for H and L tone levels to coincide is less than 0.0001, except for the pair /3-3/(p=0.014). Table 2.6. Results for significance tests: L-tone versus H-tone. (Speaker: AK)  AK  1-1  e-e  3-3  a-a  0-0  t  -11.8  -6.44  -2.98  -8.75  -7.39  s.d.  2.35  3.95  7.47  3.96  2.77  d.f.  10  10  10  10  10  P  <0.0001  O.0001  0.014  <0.0001  O.0001  Specifically with respect to the realization of pitch across vowels, the results indicate a correlation between F0 and vowel quality. The front high vowel IM exhibits the highest F0 values (relative to other vowels) for H - and L-tones, as well as creaky voice. There is still not sufficient data for this correlation, but some observations are worth being addressed here. Many studies have examined intrinsic F0 (IF0) in tone languages and found correlations between IF0 and vowel height (Whalen and Levitt 1995 provide an excellent overview; Connell 2002 is another good example). The assumption is that high vowels have higher F0 values than low vowels. Whalen and Levitt observed that tone languages are also subject to IF0, although the differences are mostly observed in H-tone vowels, and typically neutralized in L-tone vowels. The list used for this study does not provide a good corpus for a comparison across vowels, but in principle vowel height may influence the realization of F0 in Munduruku. The differences observed for H-tone vowels in Munduruku more or less conform to the observations made for other languages; while the high front vowel lil exhibits F0 values higher than any other vowel - a difference that is carried over to L-tones and nonmodal phonation - the low vowel lal exhibits the lowest values in similar contexts. A s discussed above, H-tone is clearly distinct from L-tone and creaky phonation for all vowels, and exhibits higher F0 values, typically between 120-130 Hz. Despite the overlaps, L-tone is relatively distinct from creaky phonation, and ranges from approximately 100 H z to 120 Hz; these values are closer to, but still higher than creaky phonation, which is realized with F0 below 100 H z for most vowels. In one extreme we find the high front vowel I'll with means of 131 H z for H-tone fil, 116 Hz for L-tone lil, and 113 H z for creaky III. A t the other extreme, we find the low vowel lal with  means of 122 H z for /a/, 104 Hz for /a/, and 77 H z for /a/. The low vowel /a/ has FO values similar to those for /e/ and lol when H-toned, but has the lowest values i f realized with a L-tone or creaky voice. In particular, creaky lal exhibits FO values that begin around 100 Hz, and fall abruptly towards the end (see Figure 2.9 above). If IFO is a function of vowel height, it seems reasonable to suppose that creaky phonation is fully compatible with a low vowel, but tends to be in conflict with a high vowel. Taken this way, the overlap in pitch between L-tone I'll and creaky lil discussed earlier can be properly understood. One of the effects of creaky phonation is FO lowering (e.g. Gordon and Ladefoged 2001), so it may be that there is a conflict between this effect and the elevated F0 characteristically found in high vowels. This is noticeable in the realization of creaky I'll in Munduruku, which has pitch higher than other vowels, and overlaps considerably with the low pitch of a modal lil. The low vowel, on the other hand, is distinctly lower in pitch during nonmodal phonation. This difference allows us to suppose that IFO is a property that is carried over, not only to tones, but also to nonmodal phonation. O f course, the correlation between F0 and vowel height as a general trend in Munduruku is not at this point well substantiated. A s I pointed out earlier, the data used for this study do not permit an accurate examination across vowels, and the observations made above may be accidental. The association of creaky voice with low pitch is not only phonetic but also phonological. The phonological behavior of creaky voice, specifically its relation to tones, is examined in detail in Chapter 8. It is demonstrated there that, phonologically, creaky voice is not a tonal property, as it has been argued since Braun and Crofts (1965). The tone associated with creaky vowels on the surface is always L , and like L-tone modal vowels, they are subject to the same processes. It is true that creaky vowels tend to have F0 values lower than modal vowels on a L tone, but I suggest that this is a phonetic effect associated with creaky voice (e.g. Gordon and Ladefoged 2001), and not a phonological property (see Chapter 8 for further discussion). While H-tone vowels act as a distinct set, creaky and L-tone modal vowels pattern together phonologically. This follows from the assumption that contrastive tones in Munduruku occupy only two areas in the pitch range: one defines High tones and the other defines L o w tones. In Chapter 8 I further suggest that, underlyingly, vowels may be modal, [-constricted glottis], or creaky, [+constricted glottis], and that this opposition is independent of tonal distinctions, as shown in (38).  (38)  Distinctions between modal and creaky vowels (LN=Laryngeal Node) (a) Modal vowels  (b) Creaky vowel  V I LN I [-eg.]  V I LN I [+c.g.]  I now turn to the results of two other properties: periodicity and spectral tilt, which are also good indicators of the modal-creaky contrast. Periodicity Measurements of the variation in the duration of adjacent pulses (ji )> have been used to tter  establish differences in phonation types (e.g. Gordon and Ladefoged 2001; Ladefoged, Maddieson and Jackson 1988). Adjacent pulses vary less during modal vowels than during creakiness, typically characterized by irregularly spaced pulses. Jitter is another property examined for modal-creaky pairs of vowels in Munduruku. The results indicate that the degree of variation during creaky phonation is higher then the variation observed in modal voice (see below). However, the difference between creaky and modal voicing may be better described as a difference in the duration of individual glottal pulses; creaky vowels are more or less realized with regular glottal pulses, but these are longer in duration. This is illustrated in Figure 2.12 which shows 50 ms displays of waveforms for modal [a], on top, and creaky [a], on bottom, in the words wida 'clay' and wida 'jaguar'. While modal [a] is realized with at least five complete cycles, creaky [a] has only four. Figure 2.12. Waveforms (50 ms) illustrating duration of adjacent glottal pulses of modal [a] and creaky [a], respectively.  Ill1%  0  0  1 2  12  1/\J 0kkw 25 Time  (m  s)  25 Time  (m  s)  3 8  5 0  38  50  The results of jitter measurements are given in Table 2.7. Creaky vowels tend to have more jitter, as the respective standard deviations imply. But note that the reference values for duration of individual pulses also differ. Overall, mean values for creaky vowels are higher, indicating that they are longer in duration. Table 2.7. Means and standard deviations for measures of jitter (in ms).  i  e  3  a  0  Modal  Creaky  Mean  7.76  8.13  s.d.  0.43  0.47  Mean  8.07  8.71  s.d.  0.49  0.58  Mean  7.97  9.54  s.d.  0.83  1.61  Mean  7.04  9.77  s.d.  0.35  1.13  Mean  8.40  9.56  s.d.  0.49  0.54  A significance test was conducted for the duration of adjacent glottal pulses for both modal and creaky voice. The results showed that the duration of glottal pulses during creaky voice is significantly higher relative to modal voice. Table 2. 8. Results for significance tests: Duration of glottal pulses for creaky-modal pairs.  i-j  9-?  a-a  0-0  t  -3.56  -5.78  -5.29  -15.9  -10.0  s.d.  0.45  0.53  1.30  0.84  0.52  d.f.  73  92  76  94  94  P  0.0007  <0.0001  <0.0001 <0.0001  O.OOOl Spectral tilt Gordon and Ladefoged (2001: 397; and Ladefoged, Maddieson and Jackson 1988) suggest that spectral tilt provides "one of the major parameters that reliably differentiates phonation  types in many languages." Spectral tilt compares the difference between the amplitude of the fundamental and the amplitude of the higher harmonics. Here I compare the amplitude of the fundamental (FO) relative to that of the second harmonic (h2) and the greatest harmonic in F l . In both modal and creaky phonation, the fundamental tends to have less energy than h2 and F l , but the differences between F0-h2 and F0-F1 are usually higher for vowels with creaky phonation. The differences between the amplitudes of FO and h2 (F0-h2) for all vowels are given in Figure 2.13. For both speakers the amplitude values for creaky vowels are higher than the values obtained for modal vowels, with the exception of the modal vowel Id which is higher (-5.4 dB) than the corresponding creaky vowel (-4.7 dB) for A K . Speakers differ in the values for each modal-creaky pair. For instance, the results for JT show a greater difference than the results for A K , but overall the amplitude difference F0-h2 serves as an indicator of phonation types in the language. Figure 2.13. Means for amplitude differences between the fundamental (FO) and the second harmonic (h2) for all vowels. F0-h2  0  -2  -4  -6  -8  -10  -12  -14  -16  dB  Interestingly, creaky vowels tend to have amplitude values much lower, often negative, than those of modal vowels, often positive. This difference is illustrated in the graph below.  Figure 2.14. Mean values of the amplitude of the fundamental for modal-creaky pairs of vowels. The vowels appear in the following order, from left to right: /i, e, a, a, o/. Relative amplitude of FO  -10-1  E3 Modal vowels M Creaky vowels  Figure 2.15 is a F F T spectra for modal [a] (in the top figure) and creaky [a] (on the bottom) to illustrate the difference in the amplitude of the fundamental, which has negative values for the creaky vowel. Figure 2.15. FFT spectra of modal [a] and creaky [a] in the Munduruku words wida 'clay' and wida 'jaguar' (AK). 40.  FO h2  Fl  Frequency (Hz)  The differences between modal and creaky phonation can also be quantified by comparing the amplitude difference between the fundamental and the harmonic closest to F l (F0-F1). The results are shown in Figure 2.16. For both speakers, creaky vowels have much greater energy in the region of F l , relative to the amplitude of the fundamental. This can also be seen by comparing FO and F l for both types of phonation in the spectra in Figure 2.15 above. The F0-F1 amplitude differences are also greater for modal phonation but not as high as the ones obtained for creaky vowels. Figure 2.16. Means for amplitude differences between the fundamental (FO) and harmonic closest to Fl for all vowels. TO-F1  -24.4  0  -2  -4  -6  -8  -10  -12  -14  -16  -18  -20  -22  -24  -26  dB  Measurements of spectral tilt, defined by amplitude differences between the fundamental and higher harmonics (F0-h2 and F0-F1), provide another good way of quantifying differences in phonation types in the language. The second harmonic and the harmonic closest to F l have greater energy relative to that of the fundamental in creaky phonation, whereas the difference is smaller in modal phonation. These results conform to general observations that spectral tilt reliably differentiates between phonation types (see Gordon and Ladefoged 2001, and references therein).  2.5  Conclusion To conclude the acoustic investigation of vowels in Munduruku, I summarize some of the  main findings of this chapter. The vowel system of Munduruku contains a gap in the high back region due to a missing lul. The gap is compensated for phonologically by lol, which I classifiy as [+high] based on phonological evidence. The imbalance in the mid region - because Id [e] does not have a back counterpart of same height hi - is the result of a historical process of neutralization between lol and lol. This hypothesis is supported by Kuruaya, a related language, which has a contrast between lei and lol. This uneven distribution is more or less balanced by /a/, which is acoustically closer to the mid front vowel lei. Taking this approach, we can suppose that the distribution of vowels in Munduruku conforms to the principle of vowel dispersion (Liljencrants and Lindblom 1972). Section §2.3 dealt with the acoustic analysis of the oral-nasal contrast. We saw that nasalization affects height differences mostly in front vowels, which are lower than their oral counterparts. The analysis then proceeded to the investigation of modal versus creaky vowels. Five acoustic properties were measured for all pairs of vowels: /i-j/, /e-e/, /a-a/, /a-a/ and /o-o/. From these, fundamental frequency, periodicity and spectralt tilt were very efficient in differentiating the contrast. Thus, speakers have available three good indicators of the creaky-modal distinction, but at this point of the investigation it is not possible to indicate one dominant perceptual cue. This study does not include perceptual data, but many studies have addressed similar questions. Klatt and Klatt (1990), for instance, conducted a perceptual experiment with American English speakers and, using synthesized speech, they observed that breathiness was more reliably identified by the amplitude of aspiration noise. However, different cues may be employed to differentiate types of phonation, meaning that a parameter used by speakers may not be the same, or not as relevant as, the one used by others in another language. Fundamental frequency may be an important candidate to the position, especially because of the restriction on the creaky-tone interaction. I will come back to this in Chapter 8.  CHAPTER 3  Phonetic and phonological structures of consonants 3.1  Introduction This chapter describes a number of acoustic properties of Munduruku consonants, with one  question in mind: To what extent must phonetic variations be considered phonologically relevant? A particularly interesting case involves the realization of voiceless stops in intervocalic position (Section 3.3). Consider three contexts for a stop - (i) internal or initial in the morpheme, preceded by a vowel ( V C V or V+CV), (ii) morpheme-finally followed by a morpheme-initial vowel (VC+V), and (iii) morpheme-finally preceding an identical stop in the following morpheme (VCi+CiV). The phonology of the language determines that V C V and V + C V sequences be syllabified as V . C V (where "." marks a syllable boundary), and V C + V and V C i + Q V as V C . C V , with gemination of C in sequences V C + V . (Syllabification is examined in detail in Chapter 4.) Phonetically, however, a voiceless stop is ambisyllabic intervocalically, independent of its position in the morpheme, and serves both as coda for the preceding syllable and onset for the following, as argued on the basis of acoustic evidence in §3.3. Therefore, sequences V C V , V + C V and V C + V should a priori be phonetically equivalent to sequences where two identical stops are brought together by morpheme concatenation, that is, V C i + Q V . But according to the phonology, only V C + V and V C i + C i V are expected to be equivalent. Measurements of closure duration for stops in these sequences contradict the phonology. The results indicate that a stop tends to have greater duration in sequences V C V / V + C V than in V C + V , and are closer to V C i + C i V sequences. In addition, the results vary according to individual stops and sequences. This is discussed in Section 3.3, which also examines other acoustic properties of voiceless stops such as Voice Onset Time, and vowel duration preceding a voiceless stop, in comparison to a vowel preceding a voiced stop and a vowel in open or closed syllables. Section 3.4 deals with the acoustic properties of voiced stops, and discusses some acoustic correlates of the voiced-voiceless distinction in the language, including Voice Onset Time (VOT), closure duration in intervocalic position, and duration of the preceding vowel. Nasal consonants are examined in Section 3.5. The contrast between oral and nasal vowels seen in Chapter 2, is a determining factor in the phonetic realizations of tautosyllabic nasals:  following an oral vowel, nasals are preoralized, and following a nasal vowel, they are plain nasals. The analysis compares preoralized versus plain nasals, and shows that preoralization can be treated as a coarticulatory effect to avoid neutralization of the oral-nasal contrast on vowels in the context of a tautosyllabic nasal consonant. Section 3.7 is devoted to the phonetic realizations of lal and 111. It shows that the realization of laryngeals is determined by context. Ihl is realized as the voiceless counterpart of adjacent vowels, and /?/ is realized as creaky voice following sonorants, and with a complete closure otherwise. Their realizations as products of the environment provide further support for their phonological behaviour, especially in nasal harmony (to be examined in Chapter 6). I conclude the chapter by discussing some important features for consonants (Section 3.8), based on phonetic and phonological observations.  3.2  The inventory The consonant inventory of Munduruku consists of seventeen consonants, illustrated below.  (The nasal velar/rj/ is realized as [n] syllable-initially (see §3.5)). Consonants /p/  apat  [apat]  'alligator'  Ibl  kaba  [kaba]  'parrot, sp.'  It/  weta  [wsta]  'my eyes'  IAI  ipada  [ipada]  'macaw, sp.'  patfa  [patfa]  'animal game'  pa(fea  [pacfea]  'sword'  ik/  bakat  [bakat]  'tree, sp.'  Isl  tasak  [tasak]  'It's (e.g. manioc) sour.'  IV  daja  [daja]  'fire, firewood'  ImJ  ajima  [ajima]  'fish'  In/  inaka  [inaka]  'despite'  pirja  [piP-a]  'fish hook'  (onset /rj/)  p3rj  [p3rj]  'one'  (coda/rj/)  Iwl  awa  [awa]  'grandmother'  ipaya  [ipaya]  'to stretch'  hi  parat  [parat]  'sieve'  111  ya?a  [ya?a]  'his/her head'  hi  iaha  [iaha]  'to bite'  The consonants are distributed at five major places of articulation: labial, alveolar, palatal, velar, and glottal, as in Table 3.1. Voicing is contrastive only in the stop series, in particular in stops articulated at more anterior places. The absence of a voiced velar Igl reflects a well known aerodynamic constraint between obstruents and voicing. A s Ohala (1983) explains, voicing, the vibration of the vocal folds, requires sufficient airflow through the glottis. During the production of a stop, air accumulates rapidly in the oral cavity - because it is produced with a complete closure - raising oral air pressure, and consequently inhibiting vibration of the vocal folds. In back articulated stops, like [g], the space between the point where the stop is articulated (velar region) and the glottis, where voicing is produced, is smaller than in front articulated stops like [b], whose point of articulation is the lips. Thus air accumulates more rapidly during the production of [g] than during the production of [b], causing voicing to be extinguished sooner for the former. Therefore, while voicing is threatened for obstruents in general, the anterior versus posterior sites of constriction also play a leading role in that voiced back-articulated stops are most threatened and, therefore, more likely to be absent in phonological systems. Table 3.1. Munduruku consonant inventory.  Labial Stops  Alveolar  P  t  b  d  Fricatives  s  Nasals  m  n  Approximants  w  r  Palatal tf  Velar  Glottal  k  S rj y  ?, h  The following sections examine the phonetic structures of consonants, and serve to support the analysis of syllabification in Chapter 4, and other phonological processes in subsequent chapters. In §3.3 I examine voiceless stops with respect to closure duration, V O T and the effect they have on the duration of the preceding vowel in sequences V C V . §3.4 takes up voiced stops, which can be characterized by the presence of a negative V O T word-initially and shorter closure intervocalically relative to voiceless stops. In §3.5 I examine nasal consonants with respect to the plain versus preoralized realizations. It is argued that preoralization results from the attempt to preserve the oral-nasal contrast in vowels in the context of a tautosylabic nasal. The glottal approximants /h, ?/ are examined in §3.7.  3.3  Voiceless stops The voiceless stops /p, t, tj", k/ are realized as plain stops [p, t, tf, k] syllable-initially, and  unreleased [p\ f, k ] syllable-finally, with the exception of /tj7 because it is restricted to onset -1  position. (2)  1  Syllable-initially: plain stops, [p, t, tf, k] (a)  parat  pa.rat  [paraf]  'sieve'  (b)  tawe  ta.we  [tawe]  'monkey'  (c)  tfo-?a  tfo.?a  [tjo?a]  'hill'  kak  [kak ]  'fox'  hill-CL  (d)  kak  n  In coda position, only voiceless stops /p, t, k/, nasals /m, n, rj/ and glides /w, y/ may occur (see Chapter 4).  (3)  Syllable-finally: unreleased variants, [p\ f, k ] 1  (a)  kip  kip  [kip]  'louse'  o-t-ap-?at  o.tap.?at  [o'tap^af]  ' A leaf fell.'  ojat  [ofaf]  'my food'  ?Ot.p9  [?6fp5]  worm  kak  kak  [kak^]  'fox'  9k-?a  9 k . ?a  [9k^?a]  'house'  3Su-3-leaf-fall  (b)  0 - fat  1- food ?ot-pa worm-CL  (c)  house-CL  Intervocalically, voiceless stops are ambisyllabic irrespective of their morphemic affiliation, i.e. either tautomorphemic or heteromorphemic. I use the term "ambisyllabicity" to refer to a phonetic effect of consonant lengthening, and "gemination" to refer to a lengthening effect that is phonologically relevant (see below). Munduruku does not have a phonological contrast between single and long consonants, but measures of closure duration and duration of a preceding vowel reveal that voiceless stops are produced as long consonants intervocalically. There are three morphological environments to consider for a medial voiceless stop: morpheme-internally ( V C V ) , morpheme-initially preceded by a vowel (V+CV), and morphemefinally followed by a morpheme-initial vowel (VC+V). In all these cases, the stop is ambisyllabic: its serves as coda for the preceding syllable and onset for the following. Some 2  examples are provided in (4). A raised " " preceding " C " indicates that closure for the stop is c  anticipated - these refer to sequences V C V and V + C V - and " C " indicates that the stop is c  orally released onto the following vowel - these refer to sequences V C + V .  2  A reason to assume that intervocalic stops are ambisyllabic is that a vowel preceding a voiceless stop is as  short as a vowel in a closed syllable (see §3.3.2).  (4)  V C V and V + C V (a)  o-dopa  [6.d6 .pa]  'my face'  [i .pik ]  'It's burned.'  potip  [po'.tip"']  'fish, sp.'  ka-to  [ka'.to]  'a village name'  pokaso  [p9]  'pidgin'  0 -kat  [6\kaY]  'my cultivated garden'  [i kop. af]  'One who went down.'  p  1-face i-plk  p  n  3Su-be.burned  (b)  thing-purple  (c)  k  1- garden  (5)  V C + V sequences (a)  i-kop=at  k  p  3 Su-go. down=NOM i-dip=at  [i.dip. af]  'One who is beautiful.'  p  3Su-be.beautiful=NOM  (b)  fipat=at  [jt.paVaf]  'the good one'  be.good=NOM i-kot=ap  [^.kot.'ap]  'an object for digging'  30b-dig=NOM  (c)  i-kak=ap  [i .kSk. ap"'] k  k  'an object for holding'  30b-hold=NOM d3e-6rok=at  [d3e.6rok. af ]  CoRef. Poss-hunt=NOM  k  'hunter'  Measures of closure duration were conducted for sequences V C V , V + C V and V C + V . The question addressed was as to what extent closure durations for a stop in V C V / V + C V sequences are comparable to, or diverge from, duration for stops in V C + V sequences. Here sequences V C V refer to both monomorphemic V C V and bimorphemic V+CV. I measured V + C V sequences separately from monomorphemic V C V to check i f morpheme boundary had an effect on the duration of the consonant; the statistical analysis showed that they are alike: V+pV versus V p V : p=0.15, V+tV versus V t V : p=0.35, and V+kV versus V k V : p=0.65. Given this result, sequences V C V and V + C V will be treated as a single category, namely V C V .  3.3.1 Instrumental analysis: procedures Durations were measured with reference to waveforms and spectrograms to locate the initiation of stop closure. The pairs examined compare onset ( V . C V ) versus morpheme-final (coda) stops (VC+V). Three points were tagged on each digitized token, as illustrated in the waveform in Figure 3.1. The first point (a) marks the beginning of stop closure, (b) marks the point where the stop is released, and (c) marks the initiation of voicing in the following vowel. Figure 3.1. A waveform illustrating duration measures: (a) initiation of stop closure, (b) stop release, and (c) onset of voicing for the following vowel.  (a)  (b)(c)  Two measures of duration were taken from the points marked. The first is closure duration, which was measured from the beginning of stop closure (a) to the point where the stop is released (b). The second is V O T , which measures from the stop release (b) to the beginning of voicing in the following vowel (c). For consonants with more than one burst event - the velar stop [k] in particular is characteristically marked by double release - V O T was measured from the point of the first release to the onset of voicing, as illustrated below.  Figure 3.2. A waveform showing a typical burst event for voiceless stop [k]. VOT was measured from the first burst to onset of voicing.  Measures of closure duration for intervocalic stops in V C V and V C + V sequences were also compared to the durations of identical stops in sequences V C i + C i V . These sequences are illustrated in (6). (6)  V C i + C i V sequences (a)  o-t-ap-pa  'I got feathers.'  3 Su-3 -hair-pick t-ip-pik  'The plantation is burned.'  3-bush-be.burned  (b)  0-jafcta  'my food'  1- food-CL 0-jaMsp  'my food'  1- food-CL  (c)  i-ma-kirik-kiri-n  'toties.t.'  30b-CAUS-tie-RED-IMPRF i-ma-kak-ka-ri  'to hold s.t.'  30b-CAUS-hold-RED-IMPRF  A sequence of identical consonants created by morpheme concatenation is typically realized with single closure and burst events, as illustrated in Figure 3.3 which shows the sequence [t-t] in the word ofat-ta 'my food'. A comparison with single medial stops in sequences V C V and V C + V will help answer the question as to whether single medial stops differ in duration from identical stops in sequences V Q + C i V . Figure 3.3. Illustration of stop closure in the sequence [t-t] in the Munduruku word ofat-ta 'my food'.  Finally, reference values of closure duration and V O T for each stop were also obtained from sequences V N + C V , in (7), where N is a nasal consonant and C is /p/, Ixl, or fkl. These values were used as references for comparison with the values obtained from the sequences V C V , V C + V , and V C i + C i V . Assume that closure durations of voiceless stops after a nasal consonant are the 'normal' realization, and any deviations from these will be considered shortening or lengthening. (7)  (a)  pan-pa  'one long, flexible object (e.g. snake)'  one-CL  (b)  parj-ta  'one seed-like object'  one-CL  (c)  i-men-ka  'It's real/true.'  3-true-?  A l l these sequences were chosen because closure duration is a potential acoustic property distinguishing between single and geminate consonants (e.g. Lisker 1958; Pickett and Decker 1960; Catford 1977; Lahiri and Hankamer 1988). Phonologically, V C V sequences (C=stop) are syllabified as V . C V , whereas V C + V and V C i + C i V are both syllabified as V C i . C i V , in that the 64  morpheme-final stop in sequences V C + V is lenghthened to provide an onset for the following vowel. (A complete analysis of syllable structure and syllabification is provided in Chapter 4.) Hence, a stop is expected to have shorter closure in sequences V C V than in V C + V or V C i + Q V . However, given the fact that ambisyllabicity happens irrespective of morphemic affiliation, the question is as to whether the phonetic realizations of medial stops signal their phonological behavior. In other words, are there durational differences between sequences V C V and sequences V C + V and V C i + C i V ? The results are discussed in §3.3.2. Another good indicator for single versus geminate consonants is the duration of preceding vowels (e.g. Maddieson 1985). In general, a vowel is shorter in closed syllables than in open syllables. If medial voiceless stops are ambisyllabic, it is expected that the preceding vowel be shorter than one preceding a voiced stop or word-finally in an open syllable. Durational measurements were conducted for vowels in the following contexts: word-final open syllable, word-final closed syllable, and preceding heterosyllabic voiceless and voiced stops. The results for each context are compared and discussed below. Two speakers were recorded, A K and JT, with approximately 74 tokens for each stop in all sequences - V C V : 20 tokens for each stop, V C + V : /p/=20, l\J=\9, /k/=20, V C i + C i V : 20 tokens for each stop, and V N + C V : 14 tokens for each stop. In V C V and V C + V sequences, measures of duration included both the consonant and the preceding vowel, which were compared to measures of vowels in open and closed syllables word-finally.  3.3.2  Results  Table 3.2 gives the averages and standard deviations of all tokens of each word. Table 3.2. Closure duration means (in ms) for [p, t, k] in sequences VN+CV, VCV, VC+V, and VC,+dV.  p s.d. t s.d. k s.d.  VN+CV  VCV  VC+V  VCi+CVi  156  235  188  265  21.8  28.5  25.5  28.5  152  239  193  225  31.2  31.5  29.1  23.4  130  147  169  144  20.8  41.1  41.5  40.3  Table 3.3 compares the durational differences of the various sequences. A l l sequences with [p] are significantly different from each other, but the results for [t] show similarity in V t V versus Vt+tV sequences (p=0.13), and a significant difference otherwise. Conversely, sequences with the velar stop [k] are very similar in most sequences compared, except for Vk+V versus V N + k V (p-0.0027). Table 3.3. Results of closure duration for [p, t, k] in sequences VN+CV, VCV, VC+V, and VC,+C,V.  p  t  k  V C V vs  V C V vs  V C V vs  V C + V vs  V C + V vs  VN+CV  VCi+CiV  VC+V  VN+CV  VCi+CVi  t  8.75  -3.36  5.47  3.87  -9.01  s.d.  26.0  28.5  27.0  24.1  27.0  d.f.  32  38  38  32  38  P  O.0001  =0.0018  O.0001  =0.0005  O.0001  t  7.91  1.55  4.67  3.89  -3.77  s.d.  31.4  27.8  30.4  30.0  26.3  d.f.  32  38  37  31  37  P  O.0001  =0.13  O.0001  =0.0005  =0.0006  t  1.46  0.279  -1.66  3.26  1.95  s.d.  34.3  40.7  41.3  34.3  40.9  d.f.  32  38  37  31  37  P  =0.15  =0.78  =0.10  =0.0027  =0.058  Based on these results, the observation is that [p, t, k] are distinctly longer intervocalically, but durational differences in V C V , V C + V and V Q + Q V sequences depend largely on individual stops, with no systematic relation between them. For instance, [p, t] tend to be longer in V C V sequences than in V C + V , but [t, k] pattern together i f we compare V C V versus V Q + Q V , whereas V p V versus Vp+pV are different (p=0.0018). In addition, only [p, t] are significantly different in V N + C V versus V C V and V C + V . A n additional selection of sequences V C V , V C + V and V C i + C i V for all three stops was included in the previous measurements, including data from a third speaker (EM), and the results are consistent with the results above, as shown in the following table. V C V versus V C i + C i V differ significantly for [p], but not for [t, k], whereas V C V versus V C + V , and V C + V versus V C i + C i V are significantly different for [p, t], but similar for [k].  Table 3.4. Results of a random selection of sequences VCV, VC+V and VC|+CV|.  p  t  k  V C V vs  V C V vs  V C + V vs  VCi+CiV  VC+V  VCi+CiV  t  -7.75  2.70  -11.6  s.d.  32.9  29.9  28.8  d.f.  74  78  74  P  O.0001  =0.0085  <0.0001  t  1.64  5.60  -4.86  s.d.  29.8  32.6  26.7  d.f.  73  76  73  P  =0.11  <0.0001  <0.0001  t  1.11  -0.148  1.26  s.d.  38.0  40.8  37.9  d.f.  71  77  70  P  =0.27  =0.88  =0.21  Similarly, measures of Voice Onset Time (VOT) do not favor any of the sequences. Table 3.5 gives V O T values for the voiceless stops in the same sequences. The one difference is the effect of place of articulation on burst duration, as already noticed for a number of languages (Cho and Ladefoged 1999). The velar stop [k] exhibits the greatest V O T in all sequences, with average 30 ms and up; the mean values for [p] and [t] are close but burst durations are longer for the alveolar than for the labial. Table 3.5. VOT means (in ms) by place of articulation.  p s.d. t s.d. k s.d.  VN+CV  VCV  VC+V  VCi+CiV  10.7  11.8  12.6  12.8  1.77  2.18  4.39  5.45  13.0  17.2  17.2  13.2  4.15  4.28  5.38  3.98  36.9  39.2  30.9  36.9  9.61  6.93  6.02  8.0  The V O T for [p] shows no distinction in the sequences compared. The results for [t] show a difference in V t V versus NV+tV and Vt+tV. The sequences Vt+V are similar to VN+tV and slightly different from Vt+tV (p=0.011). For [k], V O T is significantly different in V k V versus Vk+V, and slightly different in V k + V versus Vk+kV, but similar otherwise. Table 3.6. Results of VOT for [p, t, k] in sequences VN+CV, VCV, VC+V, and VQ+QV.  p  t  k  V C V vs  VCVvs  V C V vs  V C + V vs  V C + V vs  VN+CV  VCi+CiV  VC+V  VN+CV  VCi+CiV  t  1.51  -0.80  -0.77  1.54  -0.12  s.d.  2.02  4.15  3.46  3.56  4.95  d.f.  32  38  38  32  38  P  =0.14  =0.43  =0.44  =0.13  =0.90  t  2.81  3.07  -0.15  2.41  2.6  s.d.  4.23  4.13  4.85  4.91  4.72  d.f.  32  38  37  31  37  P  =0.0085  =0.0040  =0.99  =0.022  =0.011  t  0.83  0.98  4.02  -2.21  -2.66  s.d.  8.13  7.48  6.50  7.73  7.11  d.f.  32  38  37  31  37  P  =0.41  =0.33  =0.0003  =0.035  =0.012  A good indication of the ambisyllabic behavior of medial stops is a difference in the duration of a preceding vowel (Maddieson 1985). First, the duration of a vowel in Munduruku correlates with vowel height in that the lower the vowel, the greater its duration; thus lil is the shortest vowel, and lal the longest; mid vowels are in the middle. This is the pattern for both open and closed syllables, as shown in Table 3.7. Note that a vowel in a closed syllable is shorter than the same vowel in an open syllable.  Table 3.7. Vowel duration (in ms) in word-final CV and CVC syllables. (The values refer to a mean of 20 tokens for each vowel in CV# and 20 in CVC#; 2 speakers: AK and JT.)  i  cv# s.d.  cvc# s.d.  e  a  9  0  133  144  137  148  124  28.4  37.3  35.1  24.4  20.1  65  77  79  83  68  11.3  9.3  16.2  11.8  8.0  In comparison to these, the results of the phonetic length of a vowel preceding a voiceless stop in the sequences V C V and V C + V show that the vowel is also shorter, with mean values that are very similar to those obtained for vowels in closed syllables. The duration means of a vowel before [p, t, k] in the words examined are in the order of 64-88 ms for V C V sequences, and 7688 ms for V C + V . Thus, a vowel before a voiceless stop is as short as a vowel in a closed syllable, providing a better and much stronger piece of evidence for the ambisyllabicity effect of voiceless stops in the language. Table 3.8. Duration means (in ms) for vowels before [p, t, k] in sequences VCV and VC+V. (20 tokens for each sequence, 2 speakers: AK and JT).  VpV  Vp+V  VtV  Vt+V  VkV  Vk+V  Mean  87  76  88  78  64  88  s.d.  8.1  9.5  11.0  9.1  12.0  10.9  To conclude the acoustic investigation of voiceless stops, a few words can be added about the voicless affricate /tjV. This consonant is produced with a long stop component and little frication. I measured 20 tokens of the Munduruku words datfe 'hawk, sp.' and ayatfat 'woman', as produced by the two speakers, and the results, given in Table 3.9, confirm that the stop portion of the affricate is much longer than the fricative portion (means of 179 ms versus 64 ms respectively), which is in between the means obtained for [t] in sequences V N + C V (152 ms) and V C + V (193 ms). The overall mean duration of [tf], including the fricative portion, is 242 ms (s.d. 16.7). Like in the case of voiceless stops, /{$"/ seems to be ambisyllabic, hence the preceding vowel is shorter than a vowel in an open syllable (mean of 99 ms), although not as short as one 69  in a closed syllable (65-83 ms), or preceding a voiceless stop (64-88 ms); 99 ms is higher than, but still relatively close to, the duration of a vowel preceding [t] in V t V sequences (t=-2.70, s.d =13.1, d.f.=38, p=0.010). Table 3.9. Results of the palatal affricate [tf] in intervocalic position. (20 tokens, 2 speakers: AK and JT.)  Stop portion  Fricative portion  Vowel duration  Mean (ms)  179  64  99  s.d.  21.9  10.1  14.3  The conclusion we reach with the acoustic investigation of voiceless stops is that their phonetic realizations do not pattern with their phonological behaviour. The ambisyllabic character of these stops is phonologically relevant for syllabification of medial stops as geminates only in sequences V C + V ; hence V C . C V , which satisfies the requirements of the phonology by aligning a morpheme-boundary with a syllable-boundary, and at the same time provides an onset for the following syllable. (A complete analysis of syllable structure and syllabification is proposed in Chapter 4.) For the phonology, these sequences are similar to those involving a sequence of identical stops (VCi+CiV). Phonetically, however, both sequences differ with respect to closure duration; a stop tends to be shorter in a V C + V sequence than in V C V or V C i + C i V . Despite this, Chapter 4 provides phonological evidence for syllabification of V C + V as V C . C V , not V . C V . In addition, the fact that a stop has greater closure duration in sequences V C V does not appear to be phonologically significant. Phonologically, the stop behaves as a single consonant, and is syllabified as onset, i.e. V . C V . Another good reason to not assign a phonological value to durational differences is the divergence across the results of individual stops. Closure durations for /p/ are different in all sequences compared; for  the results vary between similar in some cases and different in  others; and for fk/, the results show that this stop is mostly realized with similar closure duration, with the exception of Vk+V versus Vk+kV. To integrate these results into the phonology, several arrangements would have to be made in order to capture the details of their phonetic realizations. From a phonological point of view, there is no need to make a distinction between /p/, Itl, and Ik/, because they function as a single class in all phonological processes examined in this work, as we will see in the following chapters.  3.4  Voiced stops Voiced stops lb, dl may be affected by the environment at which they occur, but their typical  realization is as [b, d] respectively. There is an optional prenasalized variant, [""b, d], that occurs n  after a nasal vowel. Relative to lb, dl, the phonetic realization of the voiced affricate /dj/ is more variable, ranging from a palatal affricate [03], to a palatal stop [}], palatalized [d ], and sometimes j  a glide [y], which occurs only intervocalically. Measurements were made for possible acoustic correlates of the voiced-voiceless distinction. One is duration of V O T in word-initial lb, dl, following Lisker and Abramson (1964; see also Keating, Linker and Huffman 1983). In Munduruku, like in many languages, voicing distinctions in word-initial stops are primarily correlated with salient prevoicing in the voiced series (negative V O T ) , whereas voiceless stops are produced with no prevoicing (positive V O T ) , as expected. Another property measured was burst duration in the voiced series, with the exception of /dy because of the variations [dj ~ j ~ d ~ y]. (The results are discussed in detail below.) J  Figure 3.4 gives the points tagged in each of the 17 tokens for the stops lb, dl word-initially: (a) marks the point where there is regular vocal fold vibration, (b) marks the stop burst and (c) the onset of the following vowel. V O T was measured as corresponding to the interval between (a) and (b), and burst duration to (b) and (c). These results will be compared to the results obtained for the voiceless series, discussed in the previous section. The words examined are: for initial Ibl, bio 'tapir' and bekitfat 'boy, child'; and for initial Idl, datfe 'hawk, sp.' and deko 'monkey, sp.'. In most tokens the two speakers were quite consistent in producing initial voiced stops with prevoicing.  Figure 3.4. Waveform of word-initial /d/ in the Munduruku word datfe 'hawk, sp.'; (a) marks the onset of prevoicing, (b) marks the onset of the burst, and (c) the offset of the burst and beginning of the vowel.  A,/W/W/\ A A A.,\.ft A 1111 'r J 1 v w vy vy Vy Vy W w v j HI Q f l 1 A  r  A  (a)  (b) (c)  A l l voiced stops have negative V O T word-initially; for the bilabial stop [b] V O T is only few ms longer than that of the voiced alveolar [d]. The voiced palatal affricate follows the pattern with mean of -107 ms. A s previously explained, burst duration could not be measured for /cfe/ because of the variation in the release. Table 3.10. Duration means (in ms) of VOT and burst for voiced stops word-initially. (17 tokens for each stop, 2 speakers: AK and JT)  b s.d. d s.d.  VOT  Burst duration  -109 ms  6 ms  22  1.9  -105 ms  11 ms  29.3  2.8  -107 ms  43 s.d.  25.6  The primary acoustic correlate of the voiced-voiceless distinction in intervocalic position is duration of the closure. A similar measure of closure duration was conducted for the voiced stops /b, d, dy in intervocalic (VCV) position, including durations of the preceding vowel. The words examined were tfebekit 'his/her child' for Ibl, djededem 'to talk' and obadip 'my relatives' for /d/,  and fodpda  'tucuma fruit' and adjodfoi 'grandmother' for Afe/. (The V C sequences  underlined correspond to the portion measured; 20 tokens were measured for each stop).  Figure 3.5 illustrates a voiced stop in intervocalic position, characteristically realized with uninterrupted vibration of the vocal cords throughout the closure, and extremely short burst release. Figure 3.5. Waveform of the Munduruku word djededem [<%ededem] 'to talk'. b  (a)  (b)  A comparison with the corresponding voiceless stops /p, t, tj7 shows that voiceless stops have closure durations that are much greater than the voiced series, as shown in the following table. The means for voiceless stops correspond to their means in V N + C V sequences because these are supposed to reflect the duration of a voiceless stop without the effect of consonant lengthening. (For /tj7 and /dj/ the means refer to stop + fricative portions.) Table 3.11. Means for closure durations of voiced stops in VCV sequences and voiceless stops in VN+CV sequences.  Closure duration  s.d.  d  t  b  p  86 ms  156 ms 62 ms  11.0  21.8  12.2  152 ms 83 ms 242 ms 31.2  24.2  16.7  With respect to duration of the preceding vowel, before a voiced stop a vowel patterns with vowels in open syllables, whereas before a voiceless stop, the vowel patterns with vowels in closed syllables. Table 3.12. Duration means (in ms) of vowels preceding voiced and voiceless stops.  VbV  VpV  VdV  VtV  VcfeV  VlfV  Duration of  Mean  109  87  141  88  140  99  preceding V  s.d.  30.7  8.1  17.6  11.0  18.6  14.3  It follows from this that unlike voiceless stops, voiced stops are not ambisyllabic, but they do have an effect on vowel duration i f we consider the difference in place of articulation to be a factor. For example, a vowel is shorter preceding Ibl than preceding Idl: 109 ms versus 141 ms. Despite the influence of place of articulation in the duration of the preceding vowel, it seems reasonable to conclude that vowels are shorter preceding voiceless stops than preceding voiced stops. To conclude, we saw that voicing distinctions in Munduruku are characterized by differences in the three parameters examined: (i) V O T in word-initial stops - voiced stops have negative voice onset time, meaning that vocal fold vibration precedes burst release; (ii) closure duration in medial stops - the voiced set is characterized by shorter closure relative to voiceless stops in intervocalic position; and (iii) duration of preceding vowels - vowels are much shorter preceding the voiceless set. I now proceed to nasal stops and the preoralized versus plain nasal variation.  3.5  Nasal stops Munduruku has three nasal stops: /m, n, rj/, each of which has at least two variants: plain  nasals [m, n, rj], and preoralized [ m, n , rj]. The velar nasal /rj/ has in addition a third variant: a b  d  e  palatal [ji] that occurs only syllable-initially, (8)c; the variant [rj] occurs syllable-finally following a nasal vowel, (9)c, and [ rj] occurs syllable-finally following an oral vowel, (10)c. 8  The bilabial /ml and alveolar /n/ are realized as plain nasals [m, n] both syllable-initially, (8)a-b, and syllable-finally following a nasal vowel, (9)a-b, and are partially oralized [ m, n] syllableb  finally following an oral vowel, (10)a-b. (8) Syllable-initially: N V and N V (a)  (b)  madi  ma.di  [madi]  'rodent, sp.'  afima  'fish'  imorj  i.morj  'to put s.t. i n . . . '  napen-pa  [naplnpa]  worm  [nobano]  'gun, rifle'  worm-CL nobano  d  (c)  (9)  rjadap  rja.dap  [pidap']  'buriti (tree, sp.)'  pirja  pi.rja  [pljia]  'fish hook'  Syllable-finally following a nasal vowel: V N (a)  (fee-pirem  d3e.p1.rem  [q^epiflm]  'to put a fire out'  [(feekon]  'to eat' (Intr.)  [ltablrj]  'S/he is alert.'  CoRef.Poss-put.out  (b)  dje-kon  c^e.kon  CoRef.Poss-eat  (c)  i-ta-berj  i.ta.berj  3-eye-be.alert  (10)  Syllable-finally following an oral vowel: V N (a)  t-irem  ti.rem  [tire m]  'S/he is w e t '  i.kon  [iko n]  'to dig something'  3Su-be.wet  (b)  i-ko-n  d  30b-dig-IMPRF  (c)  i-berj-berj  i.berj.berj  [ibe rjbe rj] 8  'S/he is full.'  8  3Su-be.full-RED  Relative to the oral closure, lowering of the velum is delayed in preoralized nasals. Compare the waveforms in Figure 3.6, which shows Iml following oral and nasal vowels. The waveform on top contains part of the preoralized bilabial nasal [ m], and the waveform at the bottom, a b  plain nasal [m]. Note that the transition from the vowel to the nasal is clearly defined in the sequence V N , whereas in the second case, vowel and nasal overlap, changing gradually from one to the other.  Figure 3.6. Waveforms showing portions of oral and nasal vowels followed by /m/ in the Munduruku words tirem 'It's wet' and cfcepirem 'to put fire out'. The top waveform contains the preoralized variant [ m] after an oral vowel; b  and in the bottom, the plain nasal [m] after a nasal vowel.  191  Time (ms)  The duration of the oral portion in a preoralized nasal is variable. The oral closure mayprecede velic lowering or closure and lowering may be more or less simultaneous, but it is never the case that lowering of the velum precedes the closure. Therefore, nasalization never interferes with the articulation of a preceding vowel in a sequence V N .  3.5.1 Preoralization as a coarticulatory effect The lowering of the velum for a nasal consonant is likely to overlap into the articulatory configuration of preceding vowels (Clumeck 1976; Manuel and Krakow 1984; Farnetani 1986; Manuel 1988; Rochet and Rochet 1991; Sole 1992). The degree of overlap varies across languages; a vowel may be nasalized throughout its duration, as in English, or partially, as in Spanish (Sole 1992). Sole attributes this difference to phonological distinctions. In languages where [nasal] is phonologically distinctive in vowels, coarticulatory effects in V N sequences must be weaker in order to maintain the oral-nasal distinction, especially when adjacent to a nasal consonant. Vowel nasalization caused by coarticulation with a nasal consonant might well be perceived by listeners as an inherent property of the vowel, and this could neutralize a phonological distinction. If, on the other hand, efforts are made to avoid such coarticulatory effects, lowering of the velum for the nasal will be delayed as much as possible so to not interfere with the articulation of the preceding vowel. From the desynchronization of velic lowering with oral closure, preoralization results. In Munduruku partial oralization is observed  only when the nasal closes a syllable, typically indicating a morpheme boundary, that is, in a position where the oral-nasal contrast may be threatened. Nasal onsets are affected neither by 3  the preceding nor by the following vowel. Instead, a nasal onset triggers nasalization on surrounding vowels, completely when a vowel precedes the nasal and partially when follows it. In the latter, nasalization lasts for less than half of the vowel. For the acoustic investigation of nasals, the targets plain and preoralized variants were analyzed for durational differences. Only the nasals /m, n/ were compared; /rj/ was excluded because of its realization as [ri] in onset position and [rj, rj] in coda. Durations were measured 8  for the following sequences: word-initial N (#NV), intervocalic N with N as onset ( V N V ) , intervocalic N following an oral vowel (VN+V), intervocalic N following a nasal vowel (VN+V), and word-final nasals following both oral and nasal vowels (VN# and VN# respectively). The wordlist is provided in (11). For preoralized nasals, duration was measured from the oral closure, therefore including the oral portion, to the onset of the following vowel. A s with the case of voiceless stops examined in §3.3 above, morpheme-final nasals followed by a morpheme-initial vowel are released. (11)  List for the acoustic analysis of nasals (a)  (b)  3  VNV  #NV npdi  'rodent, sp.'  serjemo  'lizard'  napenpa  worm  nobano  'rifle, gun'  VN+V  VN# djepirem  'to put fire out'  d3epirem=ap  's.t. used to put fire out'  cfeekon  'to eat'  d3ekon=ap  's.t. used to eat'  Stops and nasals not only close the syllable but also the morpheme; there are few exceptions with stops in  which case they seem to occur inside the morpheme (e.g. dagsem 'deer'), but I am not aware of any cases with a nasal closing the syllable but not the morpheme. Related to this is the fact that only the rightmost vowel in the morpheme is contrastively oral or nasal, so it may not be a coincidence that coda nasals are preoralized but not onsets. (For further discussion see Chapter 6.)  (c)  VN#  VN+V  tirem  'It's w e t '  tirem=at  'one who is wet'  ikon  'to dig it'  ikon=ap  's.t. used to dig'  The mean durations of nasals in the six environments are given in Table 3.13. Note that a word-final nasal (VN# or VN#) tends to have greater duration than a morpheme-final nasal followed by a vowel (VN+V and VN+V), and this has greater duration than a nasal in onset position (#NV and V . N V ) . The difference in duration between V N V and V N + V may reflect the fact that a morpheme-final nasal, like voiceless stops, are lenghthened to provide an onset for the following syllable; thus while V N V is syllabified as V . N V , V N + V is syllabified as V N . N V . Table 3.13. Duration means (in ms) for the nasals /m, nl. (2 speakers: AK and JT; average of 19 tokens for each sequence).  m s.d. n s.d.  #NV  VNV  VN+V  VN+V  VN#  VN#  84.7  102  121  132  153  147  14.8  21.4  21.5  23.6  31.3  18.0  79.6  86.7  91.3  111  132  173  25.0  12.8  15.4  27.3  27.7  35.5  Preoralized versus plain nasals seem to behave alike, as the t-test results below show. For /ml, there is a significant difference in VN# relative to V N + V , and a slight difference in VN# versus V N + V , but no major difference in the other sequences compared. Ixd, on the other hand, differs in every pair compared.  Table 3.14. T-test results for nasals /m, n/ in sequences VN#, VN#, VN+V, and VN+V.  m  n  VN# vs  VN# vs  VN#vs  V N + V vs  VN#  VN+V  VN+V  VN+V  t  0.779  3.83  2.15  -1.67  s.d.  25.5  26.8  21.0  22.6  d.f.  38  38  38  38  P  =0.44  =0.0005  =0.038  =0.10  t  -3.90  5.63  5.86  -2.74  s.d.  31.7  22.4  31.6  22.0  d.f.  35  36  34  35  P  =0.0004  O.0001  <0.0001  =0.0096  The oral portion of word-final nasals have mean durations on the order of 35 ms (s.d. 13.5) for a preoralized bilabial [ m], and 33 ms (s.d. 11.6) for [ n] (10 tokens for each sequence). This b  d  durational variation can be indicative of a coarticulatory effect by means of which velic lowering is phased differently relative to the oral closure so as to preserve the oral-nasal contrast on vowels. B y delaying lowering of the velum for the nasal, a speaker prevents anticipation of nasality in the preceding vowel, and possibly a perceptual interpretation as an intrinsic feature of the vowel. A schematic representation of the relative timing of velic and oral gestures is presented Figure 3.7, following a model of articulatory phonology as proposed by Browman and Goldstein (1989, 1992). In this model, articulatory events are gestures, defined in terms of tract variables, and each tract variable is associated with particular articulators. For example, tongue tip constrict location is a tract variable associated with tongue tip, tongue body, and jaw; velic aperture is associated with velum; glottal aperture with glottis, and so on. Since gestures characterize physical events occurring in a given space at a given time, they may interfere with each other; in other words, they overlap.  Figure 3.7. Schematic representation of the relative timing of velic and oral gestures in preoralized nasals.  vowel oral constriction velic lowering  The articulation of a nasal segment comprises two gestures: velic lowering and oral constriction (Krakow 1989). Krakow observes that word-initial nasals exhibit velic lowering more or less simultaneous with the oral constriction gesture. In word-final nasals, velic lowering gesture precedes oral constriction; this anticipation is perceived as nasalization in a preceding vowel. For the production of a preoralized nasal, the velic lowering gesture is timed relatively late so as to avoid nasality to spread to the preceding vowel. Although this may suggest a phonological process spreading the feature [-nasal] from an oral vowel to a tautosyllabic nasal, phonological evidence is provided in Chapter 6 to demonstrate that oral vowels are also subject to nasal assimilation from a nasal vowel. In other words, the phonology does not treat such vowels as inherently [-nasal]. Preoralization can thus be regarded as a mechanism to avoid a perceptual confusion between inherently nasal vowels and vowels nasalized by coarticulation with a nasal consonant, that is, it is a way of preserving a contrast that would otherwise be neutralized by a coarticulatory effect. I will return to preoralization in Chapter 6 to discuss whether preservation of the contrast in this manner should be phonologically represented.  3.6  Approximants The set of approximants comprises /r, w, y, ?, hi. Their distribution within a syllable varies,  /w, y/, may occur both as onset and coda, whereas /r, ?, h/ can only be onsets.  (12)  (a)  pa.rat  'sieve'  (b)  wi.da kaw.ta  'jaguar' 'salt'  (c)  yo.borj  'It's big' 'tortoise'  (d)  (e)  worm  ?ot.p9  i.hi  'winter'  Ixl has an additional restriction: it does not occur word-initially in native words, but may occur in borrowings (e.g. raiaPa 'can' from Portuguese lata, and rapiPip 'pencil', Portuguese lapis), in which case it is preceded by a short, nonsyllabic schwa: [ ratara] and [ rapirip"]. (Nonsyllablic [a] is examined in Chapter 4.) Approximants also have affinities with vowels, especially with respect to their behavior in nasal harmony. These are the consonants that assimilate nasality, thus all have nasalized variants [r, w, y, contexts, and are oral otherwise. 5  (13)  h] in the context of nasalization  6  (a)  ara  [ara]  'maracana (bird, sp.)'  (b)  way  [way]  'far, distant'  (c)  w-a?6  [wa?6]  'my voice, speech'  [oofio]  'my domestic animal'  1-voice  (d)  o-6ho  1-domestic.animal  4  This was first reported by Braun and Crofts (1965). Young speakers who speak Portuguese fluently  pronounce simply [rata?a] and [fapi?ipj. 5  A nasalized [y] may also be realized as [ji] syllable-initially, depending on the rate of speech: y-a-Jit=ma (3-  CL-be.small=EMPH) 'It (e.g. house) is small' is phonetically [yaJifma] or [naKt'ma]. 6  Even though the glottal 111 is transcribed as a full stop in the examples, its realization in intervocalic position  is usually as creaky voice on adjacent vowels. See section 3.7 for details.  3.7  Laryngeals: /h/and 111 Phonologically, the laryngeal approximant Ihi is restricted to medial position but may  optionally occur word-initially i f the initial syllable is underlyingly onsetless (e.g. / ' p a t h ' is [e] ~ [he], apat 'alligator' is [apaf ] ~ [hapaf], and so on). Word-medially Ihi occurs only between identical (reduplicated) vowels, except in reduplication with a fixed vowel Id in Have7  constructions, which forces Ihi to occur between different vowels. This is illustrated in (14)f. Although Ihi exhibits properties of a default consonant, its status as an independent phoneme is motivated by lexicalization of forms such as those in (14)a-c for which the non-reduplicated forms no longer exist in the language. (The phonology of Ihi is treated in Chapter 4.) (14)  (a)  ihi  'winter'  (b)  o-6ho  'my domestic animal'  F  oo  1 -domestic.animal  (c)  y-aoho=at  'one who is pale'  *yao  3Su-be.pale=NOM  (d)  (e)  (f)  (fee-a-ha-m  'to go up, climb'  o-cfee-a  'S/he climbed'  CoRef.Poss-up-RED-IMPRF  3Su-CoRef.Poss-up  i-a-ha-m  o-a  'to bite s.t'  30b-bite-RED-IMPRF  lsg-bite  t-e'i-he  t-ei  'It has a price'  3 -price-RED. Exist  's.t. bit me.'  'its price'  3-price  ' The reduplicated vowels do not always agree in tonal melody, for example, ihi 'monkey, sp.'  The phonetic aspects of Ihl are acquired from surrounding vowels. A s Ladefoged (1971) describes, Ihl is the voiceless counterpart of the following vowel. Compare the two spectrograms below, which show parts of V h V sequences in the Munduruku words (hi 'winter' (on the left) and iaham 'to bite s.t.' (on the right). Note that formant structures for vowels preceding and following the laryngeal remain steady throughout their duration. Note also that the range of frequencies for Ihl varies depending on the vowel. In {hi, phonetically [ici], the peak of energy is centered in the higher frequencies, around F2 and F3. In [aha], the region with greater energy is located in the lower frequencies, around F l and F2, with very little energy in the higher frequencies. The phonetic shape of [h] is therefore, as Ladefoged remarked, that of a neighboring vowel, without voicing. Figure 3.8. Spectrograms of VhV sequences in the Munduruku words (hi 'winter' (left), and iaham 'to go up' (right).  i  h  i  a  h  a  The affinities that the voiceless approximant Ihl has with vowels can also be extended to its compatibility with nasalization, strictly speaking, compatibility with lowered velum, the required configuration for a nasalized sound. Studies (e.g. Ohala 1974; Cohn 1990) have demonstrated that the position of the velum during the production of glottal consonants is largely determined by the context - raised i f surrounded by oral sounds, lowered i f surrounded by nasal sounds (Ohala 1974). In Cohn's (1990) study of nasalization in Sundanese, the results of airflow traces also confirm that Ihl is heavily nasalized in nasal environments.  In Munduruku Ihi assimilates nasalization from adjacent nasal vowels. Compare the two spectrograms below containing [h] in both nasal, [ifii] on the left, and oral contexts, [ihi] on the right. The nasalized sequence exhibits an uninterrupted noise throughout its duration, more evident in the region of F2/F3, which affects not only the vowels but also the laryngeal, suggesting that the velum remains lowered during its production. Figure 3.9. Spectrograms of [vhv] and [vhv] sequences in the Munduruku words ihi 'winter' and afhi 'mother, Voc.'.  1  h  1  l  h  i  Another laryngeal that shares some properties with vowels is 111. Its phonetic realization ranges from complete closure to creaky voice on adjacent vowels. The occurrence of one or the other can be predicted from adjacent segments and tone: segments determine the closurecreakiness realization, and tone determines the target for creakiness. These observations are based on the examination of 120 tokens (2 speakers, A K and JT) of words containing /?/ in the following contexts: (i) word-initially (20 tokens), (ii) preceded by a voiceless stop (40 tokens), (iii) preceded by sonorants, i.e. nasals and glides (10 tokens each), and (iv) in intervocalic position (40 tokens). The results of each context are presented in the following sections. 3.7.1  /V as a complete closure  The glottal approximant is consistently realized as a complete stop [?] when preceded by another stop, as in (15), which contain sequences Stop-? preceding both High and Low tones.  (15)  vc+?v  VC+?V  (a)  ak-?a  'house'  kak-?ao  House-CL  (b)  wiap-?ip  'rodent, sp.'  fox-?  'stick of a fan'  fan-CL  o-fat-?a  'my food (e.g. orange)'  1-food-CL  Figure 3.10 illustrates the sequence in the word kak-?ao. The release of the voiceless stop [k] is indicated in the waveform by the first point (a), and the silent period that follows it is the closure for the glottal stop. The second point (b) marks the onset of the following vowel.  8  Figure 3.10. Waveform of the Munduruku word kak?ao 'rodent, sp.' to illustrate /?/ as a complete stop.  (a)  (b)  The glottal was realized as a long and silent closure in 90% of the cases (36 tokens out of 40), but closure durations could be measured only for 25 of these. The others had no clear indication of the point where the glottal closure started as the preceding stop had no visible burst release. In 4 tokens the preceding stop was released as onset of the following vowel, and the glottal realized as creaky voice in the vowel; these are all sequences containing a L-tone vowel, 8  An interesting aspect related to the realization of [?] in C-?V sequences, not explored here, is the transition  from a silent closure period, the glottal closure, to noisy vibrations in the beginning of the following vowel (the period marked in the figure below by the vertical lines), then to constricted voice, and finally to modal voice. 19 out of 40 tokens examined showed this pattern. The mean duration for these noisy vibrations is 29 ms (s.d. 11).  i.e. sequences V C - 7 V . Tone of the following vowel is another factor that influences the duration of/?/. Following a stop, it tends to be realized as a complete closure, but is consistently realized as such only when preceding a H-tone vowel. Before a L-tone vowel it may alternate with nonmodal phonation. The results are summarized in Table 3.15. Table 3.15. Results of closure durations for 111 in VC+?V and VC+?V sequences.  VC-?V  VC-?V  111 realized as a  Individual  complete closure  means  20 tokens  87 ms  (100%)  s.d. 44.8  16 tokens  61 ms  (80%)  29.7  Total  20 tokens  20 tokens  A comparison of the means for each sequence shows the influence o f tones on the duration of 111. The glottal approximant has greater duration preceding a H-tone vowel than preceding a L-tone vowel (87 ms versus 61 ms respectively). This is because 111 is strongly coarticulated with an adjacent, preceding or following, L-tone vowel. This finding is important because it supports a phonological restriction on glottal constriction and tone in the language: only L-tone vowels may surface with constricted voicing. Figure 3.11 gives a typical example of 111 preceding both H-tone, waveform on top, and L tone vowels, waveform in the bottom. Note that 111 does not exhibit abrupt releases in either case; the transition is from a silent closure period to a period of noisy vibration (not always present as seen in fn 8), followed by constricted voicing, and changing gradually to normal. Figure 3.11. Expanded waveforms comparing the onsets of H-tone vowels, on top, and L-tone vowels, on bottom, following 111 in the words ak?a' house' and ofat?a 'my food'.  INK,  K  IN  The effect of constricted voicing is greater, i.e. it lasts longer, in a L-tone vowel (35 ms versus 22 ms respectively). In addition, L-tone vowels are more likely to be affected than H-tone vowels; overall 90% of the 20 tokens for L-tone vowels exhibit constricted voicing, whereas only 65% of the samples with H-tone vowels showed similar effects. The next section provides further details on the ?-tone interaction and coarticulatory effects with neighbouring vowels.  3.7.2  /%'as constricted voicing  In intervocalic position and following nasals and glides, 111 tends to be realized as a heavy type of creaky voice. Strictly speaking, its variation as a phonation type occurs between [+sonorant] segments. Illustrations of an intervocalic 111 preceding both H and L tones are given in the words below. (16)  V?V (a)  wa?e  V?V 'bowl'  o-?a  'axe'  axe-CL  (b)  o-de?o  'my scent'  1-scent  ipada-?a  'macaw's head'  macaw-CL  Here tones play a leading role. The spectrograms in Figure 3.12 illustrate the pronunciation of 111 in intervocalic position, preceding and following H and L tones. The spectrogram on the left represents the sequence V ? V , and the one on the right the sequence V ? V . Ill is predominantly realized as heavily constricted voicing in both cases; but coarticulation with either vowel, preceding or following the 111, depends crucially on tone. In the first sequence, V ? V (L?H), the vowel on the left is most affected by creakiness, whereas in the second case, V ? V (H?L), creakiness goes to the right.  Figure 3.12. Spectrograms illustrating the realization of an intervocalic 111. The spectrogram on the left is a sample of the word wa?e 'bowl', and on the right a sample of the word o?a 'axe'.  Another example is given in Figure 3.13. This time a H-tone vowel is intercalated between two glottal approximants; the word is wa?i?a 'gourd'. Similarly, both /?/s surface entirely as creaky phonation, and the vowels affected are the two L-tone vowels. The vowel that is intercalated between them, a H-tone A/, remains modal. Figure 3.13. Spectrogram of the Munduruku word wa?i?a 'gourd'.  Note, however, that in all spectrograms above, the region where creakiness is most salient is not in the vowel itself but at the point where the formants change, i.e. during the transition from one vowel into the other. It is this boundary, a syllable boundary in phonology, that is marked by intense glottal activity, suggesting that /?/ is still in syllable-onset position and more or less realized independently of the vowel. Therefore the crucial difference between creaky voice as a 88  contrastive feature of vowels and creakiness triggered by a glottal approximant is timing of constricted voicing. Intensification of glottal activity in the former is synchronized with the vowel itself, not truncated to the transition into the following segment, as discussed in Chapter 2. This distinction is crucial here because, i f glottal activity for 111 during the vowel to vowel transition is salient enough and overrides the perception of creakiness in the vowel, then 111 can have a phonological status on its own and be perceived by listeners as independent of the vowel. By timing both types of constricted voicing in this manner, the contrast between creaky phonation on vowels and the glottal consonant is maintained. But vowels are not the only sonorants that determine the realization of 111 as creaky phonation. The following spectrograms illustrate 111 after a glide lyl and a nasal Irjl in the words dyardyTa 'orange' and nogja 'flea'. Here too the 111 is realized with greater glottal activity near the transition to the following vowel. In the first spectrogram, glottal activity overlaps with the articulation of the sequence glide-?-vowel, going from modal to creaky then back to modal [yyaa] - and the creaky portion concides with the transition from the glide to the vowel, the point where the syllable boundary is phonologically: dyi.ray.7a. Interestingly, in ndgPa only the beginning of the vowel is affected. Figure 3.14. Spectrograms showing 111 after a glide and a nasal consonant. The spectrogram in the left shows only the underlined part of the word chirdv?a 'orange', and the second spectrogram shows the entire word n6rj?a 'flea'.  [ r a y  y  a  a]  [n  6  rj  a  a]  Given these observations, the participation of /?/ in nasal harmony is not problematic for the phonology. Like vowels, nasals and glides, the glottal approximant belongs to the group of [+sonorant], the segments that assimilate nasalization. Two examples of/?/ in nasal contexts are given in the spectrograms in Figure 3.15. Its phonetic realization as a phonation type in these contexts is perfectly compatible with velic lowering, allowing the laryngeal and adjacent vowels to be nasalized altogether. Figure 3.15. Spectrograms illustrating the sequence [v?v] in the Munduruku words wa?d 'my voice, language', and pira?e'dry fish'.  [w  a  a  u  u]  [(p)"i  r a t  |  g]  Timing of creaky voicing is the core difference between a creaky vowel and /?/ in intervocalic position. In the former, creakiness is timed to the vowel itself, whereas in the latter it is timed to a syllable boundary. Creakiness in a sequence V ? V is heavier near the vowel-tovowel transition, overriding the creaky effect on adjacent vowels. This leads to an analysis of /?/ as a true consonant.  3.8  Features for consonants Many of the considerations dealing with the acoustic properties of Munduruku consonants  discussed above, along with their phonological behavior (to be discussed in different chapters), lead me to propose six major features to distinguish consonants, and to which I will be referring in the following chapters.  (17)  Maj or features for consonants P  b  t  d  k  s  J"  sonorant  n  +  + + + +  + +  + + + +  +  rj  r  consonantal  + + + + + +  + + + +  continuant  - -  - + + - - - +  - -  +  nasal anterior voice  + + - + •  -  •  -  + -  1  ni  (±)  h  + +  y  V  V  y  y  + + + + +  + + + + + +  + + +  w  +  +  -  -  [sonorant]. The feature [sonorant] plays a major role in the phonology of Munduruku. It distinguishes the class of segments that may undergo nasality, [+sonorant], from the class that blocks the process, [-sonorant]. The classification of laryngeals II, h/ in this class has already been proposed by Chomsky and Halle (1968), but the proposal presented here is primarily supported by the acoustic investigation, which showed that the phonetic realization of laryngeals is largely dependent on the context they occur: Ihi is realized as a voiceless counterpart of adjacent vowels, and 111 as creaky phonation between [+sonorant] segments, in addition to their compatibility with processes referring to this class (e.g. nasal harmony).  [consonantal]. This feature distinguishes consonantal from vocalic segments (i.e. /w, yl and vowels). A fundamental difference between a creaky vowel and 111 is also the  feature  [consonantal]. Even though 111 may be realized as creaky phonation on adjacent vowels, timing of creakiness is phased so to coincide with a syllable boundary, more specifically, with an onset position, a position at which [+consonantal] segments occur.  [continuant]. The feature  [continuant]  distinguishes  affricates,  [-continuant],  from  fricatives, [+continuanf]. The stop portion in affricates seems to be more salient than the fricative portion, as shown in §§3.4-3.5. In addition, affricates pattern with other stops phonologically, for example, in consonant mutation (to be examined in Chapter 7). The glottal 111 is ambiguous with respect to this feature: it is [-continuant] following [-sonorant], and [+continuant] following a [+sonorant] segment. This consonant can perhaps be analyzed as unspecified for continuancy, acquiring the feature from adjacent segments.  [nasal]. Nasality is contrastive in the stop series, distinguishing voiced oral stops from nasal ones; and most importantly, it is also contrastive on vowels. Nasal vowels trigger nasal harmony, whereas oral vowels, and other sonorants, undergo the process. Nasal harmony is examined in Chapter 6.  [anterior]. Especially important for coronals, the feature [anterior] distinguishes alveolars, [+anterior] - It, d, s, n, r / - from palato-alveolars, [-anterior] -/tf, cfe, J, y/.  [voice]. The feature [voice] is important in the stop series to differentiate /p, t, tJ7 from lb, d, dy, the only voiced-voiceless set in the language. Voicing contrasts are discussed in greater detail in Chapter 7, which examines a process of voicing alternation and neutralization.  The following chapters present phonological motivations for the classification of consonants according to the features proposed above. Other features may be introduced later as needed.  CHAPTER 4  Syllable structure and syllabification 4.1  Preliminaries This chapter focuses on the syllable structure and syllabification patterns in Munduruku. The  representation adopted here is as depicted in (1). I follow Kenstowicz (1993: 253; see also Prince and Smolensky 1993) in assuming that "the syllable is a projection of the single primitive category 'nucleus'." I also assume the proposal of Shaw (1992, 1996, 2002) that the nucleus is optimally headed by a mora, i.e. a vowel. Onset and coda are optional, and are occupied by consonants. (1)  A syllable in Munduruku  The moraic versus non-moraic value distinction of segments makes a difference in the parsing of a segmental string into syllables. It is this requirement that establishes, for example, the difference between a sequence glide + vowel, (2)a, and a sequence vowel + vowel, (2)b, which are heterosyllabic (V.V). (2) (a) UR  sp. /we/ 'rodent, sp.' SR  (b)/o-e/ UR  a  'my path' SR a a I I  N N  I I u n  we  w e [we]  oe  o e  [u.e]  The mora is also the tone-bearing unit (see Chapter 8), so only vowels bear a tone on the surface. A l l else being equal, we expect that there will be as many syllables and tones in a word as there are moras. The example below illustrates this pattern. 93  (3)  /i-ea/  'It's swollen'  UR  SR  i e 3  I a  a  a  i  e  9  [i.e.9]  Further, the hypothesis that a syllable contains a single mora is consistent with another phonological process that has the syllable as a domain: reduplication. The most pervasive pattern of reduplication in Munduruku targets the final syllable of a base. Because vowels are parsed into separate syllables, the reduplicative morpheme copies only the second vowel in a V V sequence (call this V-reduplication), and Ihi is epenthesized to provide an onset for the reduplicative morpheme (details in §4.3). (Syllable boundaries are marked by a period.) (4)  (a)  i-a  i.a  ->  30b-bite  (b)  i-a-ha-m  'bitting s.t'  30b-bite-RED-IMPRF  t-ae  ->  30b-choose  ta.e  ->  t-ae-he-m  'choosing s.t.'  30b-choose-RED-IMPRF  (5) V-reduplication (RED = a ) (a)UR  (b) SR  (c) Base-RED  I N I J*  I  a  l ia  i  aha  Finally, Munduruku has a phonotactic restriction on the cooccurrence of Idl with a nasalized vowel, so that a sequence dv is never found. (Phonotactic patterns are examined in Chapter 5.) Because of this prohibition, an unusual case of blocking of nasality has arisen in the language.  As we will see in Chapter 6, nasality spreads leftwards from a nasal vowel, affecting all [+sonorant] segments. If/d/ is on the way, as the examples in (6), nasality stops before reaching the syllable containing the stop, and that syllable surfaces oral. Note that (6)b contains a sequence of vowels in which case only the second vowel is nasalized. This pattern provides the last piece of evidence that two or more vowels in sequence are heterosyllabic. (6)  Nasalization and /d/ (a)  da?6-rek  ->  da.'ro.rek  [nas]  'lizard, sp.'  [nas]  lizard-?  (b)  o-dae  o.da.e  H  I [nas]  'S.o. picked me'  [nas]  1 Ob-choose The following  sections  examine  various aspects related  to  syllable structure and  syllabification in the language. I begin by introducing the inventory of syllables and the range of constraints that determine the parsing of segment strings into syllables (Section 4.2). Section 4.3 presents an account of W  sequences and discusses a mismatch between phonetic and  phonological syllabification. Section 4.4 takes up some issues on onsets, in particular, issues related to (i) the restriction on word-initial Irl, (ii) ambisyllabic stops, and (iii) syllabification of V C V sequences. Then the analysis proceedes to an investigation of intervocalic clusters, focusing on the coarticulatory effects of some clusters (Section 4.5), and the restriction on a particular sequence of coronal segments (Section 4.5.4). The last section (Section 4.6) examines place assimilation regarding the imperfective suffix {-m}. 4.2  The basics of syllabification The inventory of syllables in Munduruku is limited to four types: V , C V , V C , and C V C , all  of which may be initial, medial or final in the word, as illustrated in (7).  (7) Syllable types  V  (a)  (b)  cv  (a)  (b)  cvc  (a)  (b)  vc  (a)  (b)  Initial  Medial  Final  a.pat  do.a  'alligator'  'crab'  'spider'  i.e.9  so.e.dgp  bi.o  'It's sore.'  'fish, sp.'  'tapir'  kaji  'sun, moon'  'pidgin'  'bird, sp.'  mg.di  ko.ra.ra  'rodent, sp.'  'fence'  'bird, sp.'  ?Ot.p9  na.pen.p9  pa.rat  'worm'  'centipede'  'sieve'  kak  ka.sop_.ta  po.tip  'fox'  'star'  'fish, sp.'  ok. pot  co.ot  'my son (masc.)'  'hawk'  'bird, sp.'  ay  e.jt  'sloth (moneky, sp.)'  'to raise (a child)'  'honey'  I begin the account of syllable structure by looking at the basics of syllabification. The language permits syllables with and without onset or coda, but it is required that these positions are occupied at most by a single consonant. A related prohibition is observed for nuclei; the language disallows more than one vowel in the nuclear position. This requirement can be achieved by ""COMPLEX (McCarthy and Prince 1993), for which I adopt a generalized version. (8)  *COMPLEX( ef) D  Given a string of segments (Cs and Vs), and three positions within a syllable (onset, nucleus and coda), parsing of more than one segment to the same position is prohibited.  The requirement that syllables must have nuclei (Prince and Smolensky 1993) and weight (Shaw 1996) is enforced by the two constraints below. Both constraints are undominated in Munduruku. (9)  (a)  a N u c - Syllables must have nuclei. (Prince and Smolensky 1993)  (b)  aMORA - Syllables must have weight (as encoded by the mora). (Shaw 1996)  It is evident from the examples in (7) that the universal preference for C V syllables, encoded by the constraints in (10), is violated in Munduruku syllables. The presence of V and types of syllable shows that both (10)  (a)  ONSET  (b)  NOCODA  ONSET  VC/CVC  and N O C O D A are violable in the language.  - A syllable must have onset. - A syllable must not have coda.  A crucial point for the analysis is the input-output dependence (DEP) relation, the 'antiepenthesis' constraint. In this study I make use of a related notion of dependence, namely (McCarthy and Prince 1994). The use of  CONTIGUITY  CONTIGUITY  instead of  DEP  will be  motivated in §4.4, which deals with a special case of epenthesis. First, let me introduce how CONTIGUITY  will be employed here.  4.2.1 Contiguity McCarthy and Prince (1994: 123) define CONTIGUITY as in (11). (11)  Contiguity a.  I-CONTIG  ("No Skipping")  The portion of Si standing in correspondence forms a contiguous string. Domain 0R) is a single contiguous string in S i . b.  O-CONTIG ("NO  Intrusion")  The portion of S2 standing in correspondence forms a contiguous string. Range (5R) is a single contiguous string in S2.  I-CONTIG  holds of input strings to guarantee that none of their internal elements is deleted in  the output.  O-CONTIG  holds of output strings to avoid internal epenthesis.  Crucially,  CONTIGUITY  applies only to internal elements; neither deletion nor epenthesis at the periphery of  a string is subject to the constraint. This effect is essential for the analysis of Munduruku, as we will see later in §4.4. McCarthy and Prince explain, "[...] the map xyz -> xz violates I-CONTIG, because the Range of x,z is not a contiguous string in the input. But the map xyz  is {x, z}, and  xy does not violate  I-CONTIG, because xy is a contiguous string in the input. The constraint 0-CONTIG rules out internal epenthesis: the map xz  xyz violates 0-CONTIG, but xy -> xyz  does not." (p. 123)  McCarthy and Prince do not consider reordering of elements in a string as a possible violation of CONTIGUITY; this is attributed to LINEARITY, the 'anti-metathesis'  constraint  (McCarthy and Prince 1994). However, reordering of elements also creates a different substring, if internal to the morpheme. For example, the map xyz -> yxz violates CONTIGUITY twice, because y,x and x,z are not contiguous strings in the basic form. The definition of CONTIGUITY assumed here determines that deletion and epenthesis, as well as reordering, must be marked as violations of the constraint. CONTIG-IO(Def)  (12)  The portion of Si standing in correspondence forms a contiguous string, as does the correspondent portion of Si. (From Kager 1999: 250) Range (91) is a single contiguous string in S i .  It is important to establish the types of structures to which CONTIGUITY applies, and the types of elements that make up a string. Here I distinguish two types of structures. One refers to a string of segments (Cs and Vs) that form a Morpheme, and the other refers to a string of morphemes that form a Word. I refer to the first type as "/«/ra-morphemic contiguity," and 1  represent it by M-CONTIGUITY  (M=Morpheme); the  second type  is "/Mter-morphemic  contiguity," or W-CONTIGUITY (W=Word).  1  I use the term Word to refer to any morphosyntactic form of a lexical item (Trask 1996); for example,  inflected forms of a verb (e.g. go, goes, went, etc.) are, in this sense, different words.  4.2.2 Inter- and Intra-morphemic CONTIGUITY W-CONTIGUITY and M-CONTIGUITY are defined (13) and (14) respectively. (M-CONTIGUITY  is a non-segregated version of McCarthy and Prince's (1994) constraints CoNTiGUlTY(Root) and CONTIGUITY (Af).) W-CONTIGUITY  (13)  The portions (i.e. morphemes) of W(ord) standing in correspondence form a contiguous string, as do the correspondent portions of W .  2  M-CONTIGUITY  (14)  The portions (i.e. segments) of M(orpheme) standing in correspondence form a contiguous string, as do the correspondent portions of M .  The  two constraints make different predictions. M-CONTIGUITY prevents  metathesis,  epenthesis and deletion of segments inside the morpheme; W-CONTIGUITY, on the other hand, prevents reordering of morphemes and epenthesis, but not deletion, of segments at morpheme boundaries. W-CONTIGUITY  effects. Suppose that {a-P-y} is an input string of morphemes that form a  word. The map a-P-y -> a-x-P-y, where x is an epenthetic segment (C or V ) not associated with the representation o f either a or p, violates W-CONTIGUITY, because epenthesis o f x creates substrings, {a-x} and {x-p}, that are not contiguous in the input. Similarly, the map a-P-y -> a-y-P violates W-CONTIGUITY, because reordering of p and y creates substrings, {a-y} and {y-P}, that are not contiguous strings in the input. Suppose now that Ixyl are segments that realize a morpheme a. The map (xv) -P a  (x) -P a  does not violate W-CONTIGUITY, because deletion o f y does not segregate the string of morphemes; the output string is still {a-P}.  2  It is important to clarify that the term "portions of the Word", as I make use of it here, refers specifically to  "individual morphemes" that are put together to form a word; therefore  W-CONTIGUITY  morphemes. In this sense, a word composed of a single morpheme would be exempted.  evaluates strings of  The application of W-CONTIGUITY is illustrated below and further discussed in §4.4 which examines V-reduplication. First of all, since W-CONTIGUITY does not prevent deletion inter-morphemically, the faithfulness constraint M A X must be posited to avoid deletion in general. Other M A X constraints and their ranking relative to each other will be introduced as I examine particular cases. (15)  MAX-IO  Input segments ( C and V ) must have output correspondents.  The ranking for constraints discussed so far, and others, will be presented in the following sections as I present evidence for their ranking. A case is suggested in Tableau 4.1. *COMPLEX dominates all other constraints to avoid tautosyllabic C C or V V sequences. Below it, we find W CONTIGUITY  (Shorthand: W-CONTIG), which prevents  epenthesis  of segments  between  morphemes, and then ONSET; M A X - I O and N O C O D A are lower in the ranking. (The position of M-CONTIG in the ranking is discussed in §4.4.) Tableau 4.1. Syllabification of kasopta 'star'.  ka-sop-ta  •COMPLEX  W-CONTIG  ONSET  MAX-IO  NOCODA  thing-flame-CL a.  b.  ka.sop.ta  c.  d.  *! * *! *!  The function of W-CONTIG is to avoid epenthesis at morpheme boundaries. Given three morphemes {a=ka, P=sop, y=ta}, we obtain the following contiguous strings in the input: {kasop} and {sop-ta}. Epenthesis of I'll, illustrated by candidate (c), violates inter-morphemic contiguity (W-CONTIG) because the vowel, which is not associated with either sop or ta, breaks the string {sop-ta} into two substrings, {sop-i} and {i-ta}, and these are not contiguous strings in the input. Conversely, deletion of /p/ in sop, illustrated by candidate (d), does not violate W CONTIG,  because the substring {so-ta} still corresponds to the string {0-y}.  A candidate such as would win according to the ranking proposed in the tableau, because the realization of {y=ta} as {at} does not violate W-CONTIG, although a violation of M 100  CONTIG  can be claimed here. The input string /ta/ corresponds to the substring at, in which case  the segments have been reordered: t-a  a-t. But given that the segments are still "contiguous"  in the output, let us assume M - C O N T I G is not violated. Despite this, would still be ruled out. We w i l l see later in this chapter that the language has a constraint, ALIGN-R, which determines that the right edge of a morpheme boundary coincide with the right edge of a syllable boundary. The constraint is introduced in §4.4.3 where some issues concerning syllabification of sequences V C V are examined. According to the ranking proposed there (see especially Tableau 4.22), is banned because /p/ in /sop/ must be aligned with the right edge of a syllable, thus ka.sop.ta wins. Let us now turn to the analysis of sequences vowel + vowel, and the difference between phonetic and phonological syllabification. 4.3  Hiatus The language has five vowel qualities / i , e, 9, a, of, most of which can be combined to form a  sequence W , syllabified as V . V . (Several arguments for syllabification of V V sequences as V . V were provided in the beginning of this chapter.) (16) shows that the high front vowel iii may be combined with any heterosyllabic vowel. (16)  / i / + V sequences (a)  i-i  i-in§m  ->  i.i.nSm  'to sew s.t'  ->  i.e.9  'It's sore.'  i.9  'It's light' (weight)  ->  'day'  ->  b'i.o  'tapir'  3 Ob-sew  (b)  i-e  i-eg  3Su-be.sore  (c)  i-9  i-9  3Su-be.light  (d)  i-a  kabi-a  sky-light/clear  i-o  bio  101  Sequences Id + vowel are also frequent, but led Combinations of the 2  n d  and /ea/ in particular are unclear.  person prefix {e-} with a verb or noun beginning with Id or Id have  different outcomes. In the sequence /e- + a.../, the vowel of the 2  n d  person prefix is always  deleted, as shown in (17)g; but i f Id belongs to other prefix, for instance {oce-} 'lpl.excl.', (17)f, there is an alternation between forms with and without Id. A s for led, it seems that this sequence is permitted in the language, as shown by examples in (17)c-d. The exception is again the 2  n d  person prefix {e-}, in which case the vowel of the root is deleted, (17)e. This suggests  that sequences led and led are generally allowed in the language, although there is a prohibition referring specifically to combinations with the person prefix {e-} '2sg', and is, therefore, morphologically conditioned. (17)  lei + V sequences (a)  e-i  te.i  'its price'  o.Je.e  'my skin'  i.e.9  'It's swollen.'  ->  (fee.9  'to go up'  ->  erj-?a *e9rj?a  'your knee'  ->  o.tja.cfeem ~ ofeadjem  'We (excl.) arrived.'  t-ei 3-price  (b)  e-e  o-j*ee 1-skin  (c)  e-9  i-e9  3Su-be.swollen  dje-9  (d)  CoRef.Poss-up  e-9rj-?a  (e)  2-knee-CL  (f)  ?e-a  otfe-adjem lpl.excl-arrive  e-ad3em  (g)  ->  a.c^em *ead3em  'Arrive!'  'bat'  2Su-arrive  (h)  e-o  mareo  A s for sequences fal + vowel, these are commonly found; only in the case of/aa/ which may optionally alternate between [aa] or simply [a], as in (18)d. (18)  /a/+ V sequences (a)  a-i  sa-i  ->  sa.i  'his/her feet'  ->  'a tree (sp.) fiber'  ->^a.a.dpm  'to send s.o.'  3-foot  (b)  a-e  koba-e tree(sp.)-fiber  (c)  a-a  i-ma-c^a-a-dja-m  30b-CAUS-go-RED-MPRF  (d)  a-a  o-ba-a-jlri ~  'my thumb'  'to put s.t. inside'  l-finger-?-AUG  (e)  a-o  i-ma-6m 30b-CAUS-enter  In sequences /a/ + vowel, illustrated in (19), and lol + vowel, in (20), no cases of the combinations /aa/ and /oa/ were found. It is not clear at this point whether the language prohibits these sequences or this is simply an accidental gap in the corpus. (19)  lal + V sequences (a)  a-i  ka-dai thing-plant  ->  ka.da.i  'plant'  (b)  a-e  t-a-erg  ->  ta.e.rg  'It's soft, tender'  dag  ->  da. 9  'fast, quick'  t-ao  ->  ta.o  'his/her leg, bone'  0.1.0  'It healed.'  'ember'  do.a  'spider'  o.o.ho  'my domestic animal'  3-CL-be.soft  (c)  a-9  (d)  a-a  (e)  a-o  3-leg/bone  (20)  lol + V sequences (a)  o-i  o-'i-o 3Su-?-be.healed  (b)  o-e  t-aboe 3-ember  (c)  0-9  (d)  o-a  doa  (e)  o-o  o-6ho  ->  1-domestic.animal  Another point to consider concerns the distinction between sequences vowel + vowel and sequences glide + vowel, which are examined next.  4.3.1 Vowels versus glides Tones are the best phonological indication for syllable breaks, and make a difference in cases such as those in (21). The pairs contrast two syllables versus one, or two moras versus 3  one. Therefore, the opposition is two tones versus one. The first column contains sequences of vowels, each of which belongs to a syllable (V.V) and is thus assigned a tone in phonology. The second column contains sequences glide + vowel ( G V / V G ) , which form a single syllable, thus get a single tone. (21)  V . V versus G V (a)  o-e  [ue]  'my path'  we  [we] 'peccary'  [ua]  's.t. bit me'  wa  [wa] 'to cry'  say  [say] 'bird, sp.'  1-path  (b)  o-a  1 Ob-bite  (c)  sa-i  [sal] 'his/her feet'  3-foot  Phonetically, the duration of a vowel is greater than the duration of a glide, especially in word-initial position. This is illustrated in Figure 4.1, which compares oa [ua] versus wa [wa]. Figure 4.1. Spectrogram showing sequences V.V versus GV in the words oa 's.t. bit me.' and wa 'to cry'.  /oa/  's.t. bit me'  /wa/ 'to cry'  For the sake of exposition, the vowel lol in the examples oe 'my path', oa 's.t. bit me', etc., is transcribed as the high back vowel [u], but recall from Chapter 2 that lol varies freely between [o] and [u].  But durational differences may be obscured by other factors. For example, glides have longer duration in intervocalic position, due to the coarticulation with both preceding and following vowels. The spectrogram below shows an intervocalic [w] in the Munduruku word awa 'grandmother', which is realized with duration similar to that of [u] in oa 's.t. bit me'. Figure 4.2. Spectrogram showing an intervocalic [w] in the Munduruku word awa 'grandmother'.  Figure 4.3 gives a slightly different example. The word kogto 'summer' has three tones and therefore three syllables: Phonetically, however, the sequence /ko/ is realized as a labialized consonant [k ], therefore phonetically two syllables: [k]. w  w  Figure 4.3. Spectrogram of the sequence /koa/ [k a] in the word /koato/ 'summer'. w  A n argument in favor of three as opposed to two syllables, comes from the speakers' intuition about syllabification. Most speakers can whistle the tones of a word: one whistle per tone per mora. In cases like kogto, they consistently whistle three tones even though it is phonetically [k ato]. This suggests that speakers whistle what is determined by the phonology, w  not how the word is pronounced (see related discussion in Mohanan 1986). It is evident that phonetics and phonology disagree on the status of vowels and glides, but the findings do not provide evidence to reliably differentiate both. The asymmetry is only 106  resolved in the phonology through the assignment of tones on vowels. In other words, vowels, unlike glides, are tone-bearing units, and since moras are the units that can bear a tone (Pulleyblank 1986; and Chapter 8), the difference between vowels and glides is that vowels are moraic. (22) (a) /we/ UR  'rodent, sp.' SR  /o-e7  'my path'  UR  SR  u.u  0 a 1 I H u  a u  we  ->  L  we  [we]  (b) /say/ [say]'bird, sp.' UR  SR  ->  oe  o e  /sa-i/ [sa.i]  'his/her feet'  UR  SR a a  a  say  [u.e]  s a y  sa-i  s  a i  By assuming moras in the underlying representation of vowels, and a constraint demanding faithfulness to an underlying mora, MAX-U., we can prevent the vowel from changing into a glide and therefore losing its moraic status. (23)  MAX-U. - Input moras must have output correspondents.  Syllabification of W  sequences is illustrated in the following tableau. First, 4  W-CONTIG  prevents epenthesis at morpheme boundaries, ruling out candidate (d). Second, glide formation is  4  A candidate that would be better than ieo in the tableau is hies, which violates ONSET only twice. I did not  include it in the range of candidates because both forms iea and hies cooccur in the language as free variants. It is desirable here that the ranking predicts the possibility of having an epenthetic element word-initially, because wordinitial epenthesis is possible, though optionally. I will return to this point in §4.4.  a fatal violation of the constraint demanding faithfulness to moras (MAX-U,), illustrated by candidate (a). Finally, two vowels parsed to a nucleus position is ruled out by *COMPLEX. This gives us the winner, candidate (b), which violates ONSET three times but retains all of its moras. Tableau 4.2. Syllabification of W  1J 1 l-e a a.  sequences, (i-ea 'It's swollen.')  •COMPLEX  MAX-U.  W-CONTIG  ONSET  NOCODA  a a  i|||§|||||  *!  ye 3 b.OCTCT  11 1  ***  UU.U  i e3 C.  OCT  1 N  A  ifillillli  *!  U.U.U  i e9 d.  a a  a  MA  *!  hi he ha  Under certain conditions, the vowels lol and lil of the 1 and 3 person possessive prefixes, st  rd  {o-} 'my N ' and {i-} 'his/her N ' , have variants that are glides, {w-} and {y-} respectively. These variants occur in inalienable possession before nouns that begin with vowels. The examples in (24) illustrate the variants {o-, i-}, and (25) illustrates the variants {w-, y-}. (24)  1 and 3 person markers as plain vowels st  (a)  rd  o-doy  o.doy  'my blood'  'his/her arm'  1-blood  (b)  i-ba 3-arm  (c)  o-adjem  casern  'I arrived.  i.o.kok  'It's dirty.'  lSu-arrive  (d)  i-okok  ->  3Su-be.dirty  (25)  1 and 3 person markers as glides st  (a)  rd  w-a?a  ->  wa.?a  'my head'  'my knife'  ya.?a  'his/her head'  y9.?9k  'his/her belly'  1-head  (b)  w-e-kise 1-Poss-knife  (c)  y-a?a 3-head  (d)  y-9?ak  ->  3-belly  The analysis I propose for the alternations o-/w- and i-/y- in possessive constructions is that these possessive prefixes have two allomorphs: one is a full vowel, {o-, i-}, the other is a glide, {w-, y-}, respectively, each of which is selected to a particular context. The tableau illustrates selection of a glide preceding a vowel. MAX-U, is not violated by either candiate because both forms, with and without a mora, are specified in the input; but the selection of {w-} is better because it satisfies ONSET.  Tableau 4.3. Selection of the variant {w-} of the 1 person possessive prefix. st  {o-, w-}a?a •COMPLEX  1 -head a.  o.a.?a  b.^  wa.?a  MAX-Ji  W-CONTIG  ONSET  NOCODA  But not all vowel-initial nouns take the variants {w-, y-}. There are cases that take {o-, i-} instead; for example, oho 'domestic animal' takes {o-}: o-oho *w-ohd 'my domestic animal'. The selection of {o-, i-} or {w-, y-} also depends on the quality of the initial vowel. This is because the language bans sequences *yi/*iy and *wo/*ow (see Chapter 5 on phonotactic restrictions). A s shown in (26), the vowels /o, i / share many features with /w, y/ respectively, differing only with respect to the feature [vocalic] in which case /w, y/ are [-vocalic]. 0  w  high  +  +  +  +  back  +  +  -  -  round  +  +  -  -  vocalic  +  -  +  _  5  y  The similarity between o/w and i/y counts as a fatal violation of the Obligatory Contour Principle (Leben 1973, Goldsmith 1976, McCarthy 1986; Odden 1986), which prohibits a sequence of identical elements. (27)  Obligatory Contour Principle (OCP) At the melodic level, two adjacent identical elements are prohibited.  We can recast the prohibitions *wo/*ow and *yi/*iy in Munduruku in terms of the OCPconstraints in (28). Note that a prohibition on sequences of segments that share more than one feature, here [+high, +round] and [+high, -back], is preferable because it does not exclude other combinations of segments that share only one feature; for example, wi or yo which share [+high].  5  Features for vowels were discussed in Chapter 2.  (28)  Sequential prohibitions (a)  *HIBK-HIBK  A sequence [+high, -back]-[+high, -back] is prohibited within a syllable.  (b)  *HIRD-HIRD  A sequence [+high, +round]-[+high, +round] is prohibited within a syllable.  A n illustration of the restriction *wo is provided in the following tableau. The candidate woho is ruled out by *HIRD-HIRD despite the fact that it provides an onset for the root-initial vowel. Tableau 4.4. OCP and selection of {o-, w-} {o-, w-}6ho * H l B K 1 -animal  4.4  a.  wo.ho  b.«"  o.o.ho  HIBK  *HIRDHIRD  •COMPLEX M A X - U W-CONTIG  ONSET  *! **  Onsets A l l Munduruku consonants can be onsets. (29) illustrates the full range of consonants in this  position, both word-initially and word-medially. Restrictions exist, however, for Irl and Ihl which do not occur word-initially, except under specific conditions, as discussed below. Word-initially  Word-medially  Ipl  poy  'tortoise'  i.pi  'It hurts.'  Ibl  bi.o  'tapir'  'canoe'  It/  ta.we  'monkey, sp.'  'village name  Idl  do.a  'spider'  ka.dirj  'dust'  Id  tjo.kon  'toucan'  da.rje  'hawk, sp.'  1)1  cfea.ray.Pa  'orange'  'cocoa'  Ikl  kip  iouse'  'banana'  Isl  say  'bird, sp.'  'knife'  IV  Jlk  'mosquito, sp.'  da.ja  'fire, firewood  Iml  ma.di  'rodent, sp.'  'fish'  Ixxl  norj.?a  'flea'  'gun'  rja.s§  'now, today'  pi.rja  'fish hook'  pa.rat  'sieve'  M  —  M  wi.da  'jaguar'  a.wi  'needle'  lyl  ya.?a  'his/her head'  a.ya.tjat  'woman'  111  ?  'worm'  wa.?e  'bowl'  i.hi  'winter'  IhJ  4.4.1 Contiguity versus Dependence The occurrence of Ihi word-initially is not entirely banned (see below), but it seems to be an innovation that has its roots in the function of the consonant in the language. A s discussed in Chapter 3, IhJ occurs in between identical (reduplicated) vowels; its status as an independent phoneme is based on reduplicated forms that have been lexicalized as such, and for which the corresponding non-reduplicated forms are no longer used in the language. These are provided in (30). The occurrence of fhl in non-lexicalized cases is illustrated in (31) below. (30)  (a)  o-6ho  'my domestic animal'  *6  1-domestic.animal (b)  y-aoho=at  'one who is pale'  3Su-CL=N0M (c)  ihi  'winter'  (d)  ihi  'monkey, sp.'  (e)  6h6-?a  'flute'  flute-CL  'ao  In general, the distribution of Ihl is constrained by the phonology to a specific role in reduplication: a default consonant. Its function is to provide an onset for a vowel in V reduplication, a pattern of reduplication in which the base is a syllable formed by a single vowel, as shown in (31). (See §4.3.3 for another context where [h] surfaces as a default onset.) (31)  Sequences of identical vowels via reduplication (a)  (b)  (c)  (d)  (e)  i-a  'to bite s.t.'  i-a-ha-m  'bitting s.t.  30b-bite  30b-bite-RED-IMPPvF  cfee-a 'to go up'  cfee-a-ha-m  CoRef.Poss-up  CoRef.Poss-up-RED-IMPRF  t-ae  t-ae-he-m  'to choose s.t.'  'going up, climbing'  'choosing s.t'  30b-choose  30b-choose-RED-IMPRP  kabi-a'day'  o^e-kabi-a-ha-m  sky-light  CoRef.Poss-sky-light-RED-IMPRP  o-i-o 'S/he got well.  i-may-o-ho-m  3Su-?-be.well  30b-CAUS-be.well-RED-IMPRF  'dawning'  'make s.o. get well'  The examples in (32) show that Ihl is prohibited i f the sequence of vowels does not result from reduplication. (32)  Non-reduplicated sequences of vowels (a)  Msa  'It's clean, new'  *ihisa  'my skin'  "ojehe  'bird, sp.'  *tfohot  3Su-be.clean  (b)  o-fee 1-skin  (c)  tfoot  113  The disparity of the distribution of IhJ is then related to its function in language: to separate two adjacent heterosyllabic vowels in V-reduplication. This is the context where it was phonologized and also the context where it mostly occurs synchronically, with some innovations (to be discussed below). The occurrence of Ihi in reduplication follows from the principles that drive syllabification. The requirement that a vowel-initial syllable in V-reduplication must have an onset is, contradictory as it seems, also a requirement of the ranking that preserves onsetless syllables elsewhere in the language. Here the role of W-CONTIGUITY is crucial, because it treats the reduplicative morpheme as any other morpheme, irrespective of its phonetic realization. This is illustrated in Tableau 4.5. Tableau 4.5. V-reduplication with epenthetic /hi. •COMPLEX  i-a-RED -m CT  MAX-U  W-CONTIG  ONSET  NOCODA  30b-bite-RED-IMPRF a.  b.  ya.ham  c.  d.  i.a.ham  *** i  *! * |*  ** **  Epenthesis of Ihi does not violate W-CONTIGUITY because the reduplicant (RED) is simply a syllable, with no specific information about the segments that compose it. Thus contiguity is required for the complex BASE-RED, but does not impose restrictions on the segments that realize the reduplicant. For example, candidate (d) obeys W-CONTIGUITY because Ihi belongs to the syllable that realizes the reduplicative morpheme. In this sense, the sequence hVis for R E D as i is for the 3 person object marker, and m for the aspect marker. Therefore, Ihi does not separate RD  the input string {a-RED}. The constraint is violated only i f there is reordering of input morphemes; for example, RED is separated from the base by an intervening segment that is independent of it, such as the suffix {-m}  'Imperfective',  illustrated by candidate  (c). The output  string is {BASE-m-RED},  corresponding to an input {BASE-RED-m}, in which case {BASE, RED} is one contiguous string, and {RED, -m} is another. Reordering breaks the string {BASE, RED} in favor of {BASE, -m}, and cancels {RED, -m}. The presence of Ihi in the reduplicant does not affect any of these.  Therefore, forms with and without Ihl are equally good according to W-CONTIGUITY, but reduplication with the laryngeal is better because of the prohibition on onsetless syllables, enforced by ONSET.  Epenthesis of Ihl is a compelling argument for the use of CONTIGUITY. To demonstrate the importance of this constraint for the analysis of Munduruku, let us first consider an account using DEP instead. One possibility is to treat V-reduplication as a case of The Emergence of the Unmarked (TETU: McCarthy and Prince 1994). Onsetless syllables are permitted generally, but not in the reduplicant. In an analysis that invokes T E T U , two faithfulness constraints must be posited: DEP-IO, to prevent epenthesis elsewhere, and D E P - B R , to prevent epenthesis in the reduplicant. (33)  (a)  DEP-IO  Any segment in the output must have a correspondent in the input.  (b)  DEP-BR  Any segment in the Reduplicant must have a correspondent in the Base.  Because we want onsetless syllables to surface, D E P - I O must dominate ONSET, and because we want to force /h/-epenthesis in reduplication, ONSET must dominate D E P - B R . Thus the ranking must be D E P - I O »  ONSET »  DEP-BR.  The first problem with this ranking is shown in  Tableau 4.6, which contains the same candidates examined in Tableau 4.5 above, but this time with D E P instead of CONTIGUITY. Note that the winner candidate is not the one with an epenthetic Ihl. To rule out candidate (c), LINEARITY, the anti-metathesis constraint, could be invoked, and i f ranked above D E P - B R , candidate (d) would be chosen. But the point I attempt to make here is that the ranking DEP-IO » not be an option.  ONSET »  DEP-BR  still predicts that epenthesis should  Tableau 4.6. V-reduplication as TETU. •COMPLEX  i-a-RED -m 0  MAX-u  DEP-IO  ONSET  DEP-BR  3 Ob-bite-RED-IMPRF *** I  a.  b.  ya.ham  c.  **  d. ©  i.a.hatn  **  *!  *!  The analysis I propose accounts for both marked and unmarked structures, and is therefore more economical. This is also in part because I believe that i f a restriction does, or does not, hold for the language in general, then it does, or does not hold for particular aspects of the same language. For example, i f onsetless syllables are generally permitted in Munduruku, then they are permitted in reduplication. Another argument in favor of CONTIGUITY is illustrated in (34). Ihl begins to extend its function as a default consonant to other environments in the language, and with the same function: to serve as an onset in onsetless syllables. This pattern is mostly found word-initially, where forms with and without Ihl are in free variation. (34) (a)  a.pat ~  ha.pat  'alligator'  (b)  e  he  'path'  (c)  i.l.S9 ~  'It hurts.'  (d)  on  fion  T  (e)  i.e.9  hi.e.s  'It's swollen.'  ~  The account defended here fails to generate the variation, but the ranking can predict that Ihl in word-initial position is possible. Tableau 4.7 gives an example. Epenthesis of Ihl appears in four contexts: between the person prefix and the verb root, candidate (a); morpheme-internally, candidate (b); word-initially, candidate (c); and both word- and morpheme-initially, candidate (d). Candidates (a) and (d) fail W-CONTIG. Epenthesis of Ihl morpheme-internally violates M CONTIG, because it segregates the morpheme {ea}. But i f Ihl is epenthesized word-initially, both M-CONTIG  and W-CONTIG are respected, and the candidate is optimal.  Tableau 4.7. Ihi word-initially but not morpheme-internally  i-ea  M-CONTIG  a.  i.he.9  b.  i.e.ha  c. ^  hj.e.a  d.  hi.he.a  MAX-u,  W-CONTIG  ONSET  *!  iiiiiiii  NOCODA  iiiiitii  *!  ** *!  *  Note that candidate (d) would be optimal i f ONSET dominated W - C O N T I G since it incurs less violations of ONSET relative to (a) and (c). Now, suppose that a change in the ranking takes place, and ONSET is promoted: ONSET »  W-CONTIG.  Reranking predicts (d), but not (b),  because epenthesis internal to the morpheme is ruled out by M-CONTIG. Therefore, i f the phonology of the language decides to avoid onsetless syllables, it is predicted that the change will be at the periphery of morphemes, not morpheme-internally. This prediction is consistent with cross-linguistic observations that epenthesis at the edges of the morpheme or word are preferred over those internal to them (McCarthy and Prince 1994; see also Peterson 2004 for a good discussion of epenthesis and contiguity in Kabardian). A l l else being equal, the predictions in terms of language change are as in (35). (35)  Historical change involving epenthesis M-CONTIG »  W-CONTIG »  ONSET  Effect: Epenthesis at word boundaries possible.  M-CONTIG »  ONSET »  W-CONTIG  Effect: Epenthesis at word and morpheme boundaries possible.  ONSET »  M-CONTIG »  W-CONTIG  Effect: N o onsetless syllables.  The alternations already observed in Munduruku suggest that the language may be taking a step further in the change above. This is consistent with the idea that epenthesis first takes place at the edges of morphemes and then internally, and not vice-versa. The analysis also succeeds in accounting for reduplication with fixed segmental material.  There is a pattern of reduplication in Munduruku that has a fixed vowel Id, which derives verbal predicates from nominal constructions; I refer to them as Exist(ential)-predicates, although there are two related interpretations: (i) as predicates of possession, when the noun (N) has an overt possessor - e.g. 'I have a house', or literally ' A house exists in my possession'; and (ii), as existential predicates when N does not have an overt possessor - e.g. 'There is a canoe', or ' A canoe exists'. In this kind of reduplication, the final syllable of the noun is copied but the vowel is always Id. This is illustrated in the following examples. 6  (36)  Reduplication with fixed vowel Id (a)  (b)  (c)  (d)  6  w-e-kobe  w-e-kobe-be  (  1-Poss-canoe  1-Poss-canoe-RED.exist  'my canoe'  'I have a canoe.'  o-dak-?a  0- dak-?a-?e  (o.dak.?a)  l-house-CL:round  1- house-CL:round-RED. exist  'my house'  'I have a house.'  ako-ba  (  ako-ba-be  banana-CLxylindrical.hard  banana-CL: cylindrical .hard-RED .exist  'banana'  'There are bananas.'  noban6-nom(  nobano-nom-nem  rifle.gun-CL:powder  rifle.gun-CL:powder-RED.exist  'gun powder'  'There is gun powder.'  Reduplication with Id copies features such as nasality and creaky voice, but not tone. Thefixedvowel is on a  L-tone irrespective of the tone of the base. This can be achieved by the ranking CONTIG »  ONSET »  in the reduplicant.  MAX-L  »  MAX-BR,  M-CONTIG  »  MAX-U »  W-  which preserves L-tone of Id while forcing other features to be realized  (e)  0- 6ho  (o.o.ho)  o-oho-he  1- domestic.animal  1 -domestic. animal-RED. exist  'my domestic animal'  'I have a domestic animal.'  A s before, i f the syllable to be copied is composed of a single vowel, Ihi is inserted as an onset for the reduplicant vowel Id, as shown in the examples below. (37)  V-reduplication in reduplication with fixed vowel Id (a)  t-e'i  (te.i)  t-ei-he  3-price  3-price-RED.exist  'its price'  (b)  i-jee  'It has price.'  (i-je-e)  i-Jee-he  3-skin  3-skin-RED.exist  'its skin'  'It has skin.'  Following a proposal by K i m and Picanco (2003; see also Chapter 9), I assume that the vowel Id is underlyingly specified in the representation of the reduplicant, as shown in Tableau 4.8. A s a vowel, Id is moraic, therefore, it is required by MAX-U. to be realized in the output. And again, epenthesis of Ihi does not violate  W-CONTIG  7  because the vowel Id does not define  R E D on its own, only part of it. If we assume that R E D is Id by itself, then the input morpheme should be {-e}, not R E D ; in this sense, R E D would be either unnecessary or independent of {-e}, and this is not the case here.  7  By realizing both Id and Ihi in the reduplicant, faithfulness to the base is violated (MAX-BR), but this is not  crucial to the point being discussed here;  MAX-BR  is discussed in Chapter 9.  Tableau 4.8. Epenthesis of Ihl in reduplication withfixedvowel Id.  t-ei-RED  CT  a.  1 e te.i.e  b. ®"  te.i.he  c.  te.hi.he  d.  te.i.hi  M-CONT1G  MAX-|J,  W-CONTIG  ONSET  * *! *!  *  As seen in the tableau above, the role of M-CONTIGUITY is to avoid epenthesis morphemeinternally, and this is the reason why candidate (c) is banned. If an epenthetic element is required, epenthesis will first target the edges of morphemes. In this language it is more important to preserve W sequences morpheme-internally than at morpheme-boundaries.  4.4.2 Word-initialM Another restriction on word-initial onsets is the occurrence of Irl in this position. This consonant is the second most frequent consonant in onset position in the language (see Chapter 5) even though most consonants have a distribution that include word-initial position. However, the restriction on a word-initial Irl affects native words, but not borrowings. The words below are borrowings from Portuguese; they are already consolidated in the lexicon of the language, as it is evident from both phonological and grammatical affinities with native words, in particular, their association with classifiers and adaptation to the native phonology. For example, Portuguese IV is replaced by Irl since IV is not a phoneme of the native inventory; also, final Isl in lapis is deleted since Isl is not a possible coda in Munduruku; and finally, the Portuguese stressed vowel is reinterpreted as a H-tone vowel. (38)  (a)  rapi-?ip  'pencil'  (Portuguese lapis)  'can'  (Portuguese lata)  pencil-CL  (b)  rata-?a can-CL  (c)  papera-dap  'paper'  (Portuguese papet)  paper-CL: flat, flexible  The innovation these borrowings brought to the language is the introduction of a new environment for Irl, previously restricted to medial position. The issue here is that borrowings, but not native words, allow Irl word-initially. Suppose the language has a constraint *#r that bans Irl in this position. The question then is: 8  What forces the phonology to allow borrowings to violate this restriction? First of all, Irl may not be actually word-initial. A s first noticed by Braun and Crofts (1965), a word-initial Irl is phonetically preceded by a non-syllabic schwa (see §4.5 for more details on this non-syllabic vowel): [ r]api?ip 'pencil' and [ r]ata?a 'can'. It may also be pronounced as a plain [r], 3  3  9  especially by those more familiar with Brazilian Portuguese. Here I consider the forms with and without a non-syllabic vowel to be phonologically deviant with respect to the prohibition on the occurrence of Irl word-initally, despite the fact that the variant [ r] occurs in the pronunciation of 3  these words. Given this, there seems to be other major requirements that force the phonology to tolerate the introduction of a new pattern in the language, and although these are in conflict with the constraint *#r, they still have priority. I believe that this is related to the fact that, when the borrowings were introduced, Irl was the best replacement, the 'output' most similar and more faithful to the source, i.e. to III, than any other consonant available in the language's inventory. The phonemic inventory of Munduruku comprises seventeen consonants (as already seen in Chapter 3). From these only Idl and Irl are relatively close to III. While the alveolar stop is associated with III in terms of place of articulation and (phonetic) voicing, Irl shares place of  8  Of course, the constraint *#r only describes a prohibition so is not properly formulated, but motivation for it  can be based on a similar prohibition in other languages. For example, certain dialects of Brazilian Portuguese, [r] also does not occur word-initially. 9  Braun and Crofts report that native words are also pronounced this way word-initially, but in all examples  used to illustrate the variant, the initial consonant is not underlyingly Irl as they suppose, but Id/. In later work, Crofts (1985) reports these forms with Idl, and comments that its pronunciation resembles that of [r]. I have witnessed /d/-flapping only intervocalically, and for some of the speakers I have worked with, perhaps suggesting a dialectal variation. Given this, there seems to be a good reason to believe that a word-initial Irl is in fact an innovation in the language.  articulation, voicing, and most importantly, it is linked to III by the feature  [+sonorant].  Therefore Irl is closer to III than Idl. Such considerations lead to the hypothesis that it is more important to respect sonorancy than a restriction. The choice for [+sonorant] can be achieved by positing a high-ranking constraint that preserves this feature. (39) MAX[+sonoranf] Input specifications of the feature [+sonorant] must have output correspondents.  MAX[+son] must dominate the phonotactic restriction on a word-initial Irl, *#r. Tableau 4.9. From Portuguese lata to Munduruku rata. MAX[+son]  lata a. ^  rata  b.  data  *#r  * *!  The phonology must permit word-initial /r/'s in Munduruku, otherwise words like rata Pa and rapiPip would not be possible. What complicates matters is that native words do not reflect the proposed ranking. On the contrary, they suggest an opposite relation between faithfulness and the markedness constraint, one in which *#r dominates MAX[+son]. This is the ranking that bans Irl word-initially and retains it medially, as shown in the following tableau. (In the tableau 'x' represents a consonant that is different from Irl.) Tableau 4.10. Ranking for an underlying hi in native words.  ra  *#r_  a.  ra  b. ^  xa  ara  *!  * *#r_  c. ^  ara  d.  axa  MAX[+son]  MAX[+son]  *!  This divergence, I believe, is not because the phonology treats native and borrowed items differently. A n alternative hypothesis which I would like to defend in the next section is that the 122  language once had a prohibition *#r, but it was lost over time; synchronically, Irl is 'free' to occur word-initially. The gap in native words reflects the period at which *#r was still active, whereas the borrowings reflect the new pattern. Changing the ranking Suppose that at an early stage of Munduruku, shown in (40)a, *#r dominated MAX[+son], and so determined the distribution of Irl in general. This of course banned all /r/'s from wordinitial position and, hypothetically speaking, changed it into x.  10  A t this stage, forms such as rata  and rapi were not possible. A t a later stage, (40)b, *#r lost its position for MAX[+son]. A t this point, the input form of a previous Irl is no longer Irl, but has been coined (reinterpreted) as Ixl, the output of the former ranking. This is due to one of the basic principles in OT, Lexicon Optimization (Prince and Smolensky 1993), which asserts that input forms echo outputs. I assume that Lexicon Optimization plays a leading role in language change by coining the output of an early stage as the input of the following. Here is when rata and rapi came into the language; the change in the ranking favored a new environment for Ixl, but left a gap in the distribution of the consonant. Synchronically, only novel forms have it in word-initial position. (40) Consequences of changing the ranking (a)  Input:  Ir&l I oral  *#r  »  MAX[+son]  Output: [jca][ara] (Lexicon Optimization)  (b)  Input:  Ixallaral  MAX[+son] »  *#r  Output: [xa] [ara]  10  A word-initial Ixl underlyingly is hypothetical, in respect to one of basics of OT, the Richness of the Base,  which says that no constraints hold at the level of underlying representations. There is, however, no languageinternal or comparative evidence to support the hypothesis. Amongst Tupi languages, at least Mekens (Galiicio, p.c.) and Mawe (Sergio Meira, p.c.) show patterns similar to Munduruku; Karo (Gabas Jr. 1988, 1999), Gaviao (Denny Moore, p.c), and Aweti (Sebastian Drude, p.c.) all exhibit /t/-flapping in which a morpheme-final l\l is realized as [r] before a morpheme-initial vowel. Thus, the consonant seems to be preferred in intervocalic position in several Tupi languages, and since long ago.  A strong argument in favor of the change in (40) comes again from borrowings. Compare the word djardy 'orange (tree)' to rata 'can', which correspond to Portuguese laranja and lata respectively. The question here is: Why does Portuguese IV correspond to Munduruku /03/ in djardy but Irl in ratal Suppose that djardy came into the language when *#r was still active, i.e. at the stage in (40)a. This prevented IV from being introduced into the language as Irl, because *#r banned it. Another option would then be Idl, the most similar consonant after Irl. This implies that dytray entered the language first as dardy, changing later to djardy: laranja —> daray > c^aray (the notation ">" means 'developed into'). This seems to be correct: not only is the change d > dj a real diachronic change in Munduruku (Chapter 5 discusses it in great detail), but also the correspondences lAfe versus 1/r in laranja/q^aray versus lata/rata can be explained in a principled way. The word rata entered the language later; strictly speaking, at the stage in (40)b. Given that *#r could be violated at that period, Portuguese IV could then be implemented as Irl. The discrepancies between borrowings and native words can then be considered as the result of a change the constraint *#r underwent. Does this mean that *#r is still active under certain conditions, or that it lost completely its function in the language? Loss of a constraint: systematic versus accidental gaps A plausible answer, one that follows from OT's basic concepts, is that reranking of constraints took place - grammars differ only in constraint ranking (Prince and Smolensky 1993). If we take different stages of a language to represent a different grammar,  11  then a  constraint operating at an early stage may be ranked higher or lower at a following stage. Suppose that two languages (X and Y ) exhibit a restriction on a segment a, and Universal Grammar (UG) has a markedness constraint *a prohibiting a to occur in an environment z. Does the restriction on a mean that both languages have the constraint *a operating in their grammars? The idea defended here, and corroborated by the diachronic analysis of Munduruku 11  By "different grammar" I do not mean that two stages of a language are completely unrelated, but that  changes that take place at one stage have an effect on the following and, consequently, in the former grammar.  in Chapter 5, is that it is not mandatory that *a be part of the synchronic grammar of a language. Let us begin by assuming that *a is a constraint in grammar X . In this grammar *a outranks Faith(ulness) to a , so the absence of a in a given context is a systematic gap, accounted for by the ranking *a » (41)  Faith-a banning a in a principled way.  Systematic gap a...  *a  Faith-a  * b.  a...  *!  In grammar Y , on the other hand, the gap on the distribution of a is not systematic, meaning that *a does not consistently ban a from context z; an example of this can be the difference between borrowed and native items discussed above. This cannot be because of the ranking *a »  Faith-a, since it bans a altogether; for this grammar, *a must be ranked sufficiently lowly  such that its effects are not evident. In other words, learners of this language have no indication whatsoever that *a is active in the grammar. I believe that a child acquiring grammar Y will acquire former borrowings as part of the native vocabulary; for them, a can and does occur in z. Therefore, the gap in the distribution of a in this language can be considered as accidental, the consequence of a sound change. To illustrate the point, assume that at stage I of Munduruku the ranking was *#r  »  MAX[+son]. This banned all /r/s word-initially: hence laranja 'orange' entered the language first as dardy, not *raray. When MAX[+son] gained a higher position in the ranking, it opened a window for a word-initial Irl; hence lata 'can' became rata. B y Lexicon Optimization, stage II did not have Irl word-initially in the native vocabulary, but could have it in novel forms. Consequently, children acquiring the language at stage II find Irl in that position, though not frequently; for them, rata and rapi are not borrowings, but lexical items that are part of the native vocabulary. Therefore, *#r has no synchronic motivation, and therefore needs not be acquired. This issue is important because the assumption that a constraint can be completely 12  12  So far I have not found any cases in which a morpheme-initial Irl may occur both word-initially and word-  medially to test whether there is alternation of some sort. If such cases do not exist in the language, then children  demoted from its function makes a difference. Under this view, a diachronic change in one area causes deterioration on another. In the case at stake, promotion of MAX[+son] caused the deterioration of *#r. This hypothesis not only predicts that borrowings may have a word-initial Irl, but also allows for an internal change producing a similar result. For example, a vowel is deleted word-initially leaving Irl in that position ( V r V > rV). But i f we assume that *#r and MAX[+son] were simply reranked relative to one another, the prediction is that Irl will occur word-initially only i f it replaces a [+sonorant] segment. To sum up: the hypothesis is that a gap on the distribution of a segment a may be accidental, meaning that there is no synchronic motivation for the activity of a constraint such as *a. Here I introduced the idea using *#r as an example; other cases will examined in Chapter 5.  4.4.3 Asymmetries in VCV syllabification In Chapter 3,1 compared closure durations for voiceless stops in a number of sequences: (i) after a nasal consonant - V N + C V ; (ii) in onset position - V C V ; (iii) in morpheme-final position followed by a morpheme-initial vowel - V C + V ; and (iv) in combination with an identical stop V C i + C i V . The mean durations, repeated in Table 4.1, showed that voiceless stops are in fact longer (ambisyllabic) in intervocalic position, but the length of individual stops varies according to the stop and the sequence (see below).  13  Table 4.1. Summary of the duration means (in ms) for voiceless stops.  VN+CV  VCV  VC+V  VCi+CiV  p  156  235  188  265  t  152  239  193  225  k  130  147  169  144  The issue here is the mistmatch between phonetics and phonology. First, morphological operations  combining two  identical stops ( V Q + Q V )  create  geminate-like  consonants  phonologically, and as such are expected to have greater duration than single stops in sequences  would in fact have no evidence for *#r. 13  Ambisyllabicity is also signaled by duration of the preceding vowel. A vowel preceding a voiceless stop is as  short as one in a closed syllable.  V C V . The difference was significant for Ipl (p<0.0001) but not for Ixl or Ikl (p=0.11 and p=0.27 respectively). Similarly, a morpheme-final stop followed by a vowel (VC+V) is lengthened in normal speech (see below on careful speech) to provide an onset for the following vowel. This can be characterized phonologically as gemination; thus sequences V C + V should in principle pattern with sequences V C i + C i V ; but they do not. A comparison of the results show a significant difference for /p, t/ (p<0.0001 for both stops) but no difference for Ikl (p=0.21). A different result was obtained for nasal stops. A s expected, a nasal is longer in V N + V or V N + V sequences, where they are lengthened, than in V N V . Table 4.2. Summary of the duration means (in ms) for nasals.  VNV  VN+V  VN+V  m  102  121  132  n  86.7  91.3  111  Finally, it is well known that the universal parsing of V C V strings is V . C V . However, the two sequences, V C V and V C + V , differ phonologically in Munduruku. A n intervocalic stop or nasal is associated with an onset position only i f V C V strings are either monomorphemic (42), or the morphological boundary is V + C V , (43). (42) V C V  (43)  (a)  a.pat  'alligator'  (b)  de.ko  'monkey, sp.'  (c)  a.jlma  'fish'  (d)  'rifle/gun'  V + CV (a)  i-pak  i.pak  'It's red.'  e.kop  ' Y o u went down.'  (b)  e-kop  ->  2Su-go.down  (c)  o-jo-morj  ->  'I put i t '  ->  o.n§y  'my teeth'  lSu-30b-put  (d)  o-n6y 1-tooth  Syllabification as  V.CV  in these examples is determined by  ONSET.  Tableau 4.11. VCV -> V.CV. Word: deko''monkey, sp.'  deko  W-CONTIG  a. "  de.ko  b.  dek.6  tS  ONSET  MAX-IO  NOCODA  *  *!  The constraint also predicts that a morpheme-initial consonant is associated with an onset position. Tableau 4.12. V+CV -> V.CV. Word: ipak 'It's red.'  i-pak  W-CONTIG  ONSET  MAX-IO  NOCODA a. f  i.pak  b.  ip.ak  *  *  But the situation is different with nasals and stops in  VC+V  sequences because the  phonology of the language requires their realization as codas. In this position only voiceless stops /p, t, kl, nasals /m, n, rj/ and glides /w, y/ occur. (44)  14  Word-medially  Word-finally  /p/  ka.sop.ta  'star'  wi.ap  'fan'  Ixl  o.jat.?a  'my food (e.g. round fruit)'  fipat  'to be good'  14  The occurrence of 111 in coda position is discussed in Chapter 8.  IkJ  ok.pot  'my son (male speech)'  kak  'fox'  Iml  'object for squeezing manioc'  dap. s em  'deer'  Inl  i.en.morj  'I put the meat'  tjo.kon  'toucan'  ¥  norj.?a  'flea'  yo.borj  'It's big.'  /w/  kaw.ta  'salt'  da.few  'fish, sp.'  lyl  'snake'  ay  'rodent, sp.  If the morpheme boundary is V C + V , the universal preference for V . C V is overridden by another that requires syllabic and morphological boundaries to be aligned. That is, oral and nasal stops are obligatorily associated with a coda position, not only in normal but also in careful speech. In normal speech, these consonants close the syllable it is associated with and is released on the following vowel, as illustrated in (45). (45)  Normal speech (a)  i-dip=at  ->  i.dip. at  'The beautiful one'  p  3Su-be.beautiful=NOM  (b)  w-e-wiap-ep  we.wi.ap. ep  'I have a fan.'  ta.en. en  '(a fruit) with flesh'  p  1-Poss-fan.RED.exist  (c)  t-a-en-en  ->  n  3-CL:seed-flesh.meat-RED.exist  (d)  6m-6m (ayatf.a=y5)->  6m. 6m  '(The women) are going in.'  m  enter-RED woman=PL  This distribution of closure and burst events between two different syllables suggests that the consonant is a geminate in the phonology, serving both as coda for the preceding syllable and onset for the following. Gemination is not the resyllabification of a coda consonant: it can be seen as a way of satisfying the alignment of a morpheme boundary with a syllable boundary. Evidence for this requirement  comes especially from observations 129  in careful speech: a  morpheme-final consonant remains in coda position, and the following onset is either realized by an epenthetic [h] or syllabification is V C V . (Thus far I have observed this alternation only in reduplication, between base and reduplicant, and at word boundaries.) (46)  Careful speech (a) ~ i.dip.[h]at  'The beautiful one'  (b)  we.wi.ap.ep ~ we.wi.ap.[h]ep  'I have a fan.'  (c)  ta.en.en ~ ta.en.[fi]en  '(a fruit) with flesh'  The alignment of the right edge of a morpheme with the right edge of a syllable is represented here by the constraint below. (47)  ALIGN-R  - Align (Morpheme, Right, Syllable, Right)  The right edge of a morpheme must be aligned with the right edge of a syllable.  Here I assume non-crisp edge alignment, following Ito and Mester (1994: 38). (48) Alignment: Dfn Align(Catl, Edgel, Cat2, Edge2)  Let Edgel, Edge2 be either L or R. Let S be any string. Then, for any substring A of S that is-the-content-of-a C a t l , there is [a] substring B of S that is-the-content-of-a Cat2, such that there is a decomposition D(A) and a decomposition D(B) of B , both sub-decompositions of a decomposition D(S) of S, such that Edgel(D(A))=Edge2(D(B)).  ALIGN-R  prevents a morpheme-final consonant to be syllabified exclusively as an onset  preceding a morpheme-initial vowel, and i f dominated by ONSET, gemination is forced. In the tableau, brackets mark morpheme boundary and periods mark syllabic division.  15  15  Gemination violates another constraint, CRlSPEDGE[a] (Ito and Mester 1994); this constraint is not discussed  because its application does not seem to be crucially required for the facts examined here.  Tableau 4.13. Gemination of coda consonants in normal speech. Input: idip at 'the beautiful one' 7  7  i-dip=at 3Su-be.beautiful=NOM  W-CONTIG  ONSET  ALIGN-R  NOCODA  a. CT CT CT llliiliilllillll  UN  i.]dip.]at] b. CT CT CT  1  foA  **  *  i.]d'ip.] at] p  C. CT CT CT  MA  *  *!  llllllplilllll  i.]di.p]at]  The variation in careful speech cannot be accounted for by the ranking proposed here. Epenthesis of [h] in reduplication (e.g. wiap-ep  wi.ap.hep 'I have a fan') satisfies ONSET but  violates W-CONTIG. Likewise, an output without [h] (e.g. wi.ap.ep), satisfies W - C O N T I G but violates ONSET. This goes back to the discussion on the possibility of a change promoting ONSET.  If W-CONTIG loses its status relative to ONSET, wi.ap.hep is predicted; that is, epenthesis  at morpheme boundaries may become a regular repair strategy to avoid onsetless syllables in the language. The presence of [h] in careful speech may be an indication that this is on its way. The analysis of sequences V C i + C i V is similar to the one proposed for V C + V . That is, sequences of identical consonants are treated as geminates, as shown in the representations in (49). This analysis is supported by acoustic investigations (see Chapter 3) which show that stops in V C i + C i V sequences are realized with single closure and burst events, like a geminate. (49)  (a)  (b)  i-yoy-yoy -> i.yoy.yoy 30b-bake-RED 'to bake s.t' CT  o J" a  t  a  a  rj  It is worth noting, however, that sequences V C i + C i V and V C + V may differ in closure duration depending on the stop. A s previously discussed, this phonetic difference is observed for /p/ and l\l only; for Ikl, sequences Vk+kV and V k + V are not distinct. Given the discrepancy of the results, I will reserve this issue for future research. For now, let us assume they are phonologically similar. As a geminate, two identical consonants in a sequence V C i + C i V satisfy the right alignment constraint and ONSET. M A X - I O is violated by candidate (a) because it deletes the final consonant of the base, despite the fact that deletion satisfies N O C O D A . Tableau 4.14. Syllabification of a sequence of identical consonants. Word: -fat-ta 'food-CL'. jat-ta  W-CONTIG  a.  Ja.]ta  b. « •  fat.]ta  ONSET  ALIGN-R  MAX-IO  NOCODA  *!  The next issue to be examined in syllabification in Munduruku concerns consonant clusters. I begin by looking at some phonetic and phonological aspects of the clusters that are allowed, and then proceed to those that are not, one in particular: sequences coronal + coronal (§4.5.4). 4.5  Consonant clusters The language allows a large set of medial clusters, all of which are heterosyllabic. We have  already seen that this is because of *COMPLEX, which forbids more than one consonant in coda or onset positions. N o w we are going to see what clusters are predicted, what are attested, and their phonetic details. Considering all possible combinations of coda and onset (8 consonants in coda position, /p, t, k, m, n, rj, w, y/ and 17 in onset /p, t, tf, k, b, d, 03, s, J", m, n, rj, r, w, y, ?, hi), there are 136 possibilities for C C clusters. From these, 8 are excluded because they involve sequences C-/h/ and Ihi does not occur post-consonantally in normal speech, as seen earlier; another 8 clusters 16  t-tf, t-d, t-dj, t-s, t-J", t-n, t-r, t-y - do not occur because of a prohibition on sequences l\l + coronal C (see below in §4.5.4). The others are given in Table 4.3; attested clusters are indicated by  16  It does occur in careful speech though. This was discussed in §4.4.3.  the ones not possible are indicated by ' * ' , and those for which I have found no examples, but also no evidence that they cannot occur, are indicated by ' ? ' . Examples are provided in the following sections. Table 4.3. Combinations of coda x onset.  p  t  k  b  d  s  I  m  p  y  y  y  y  y  y  y  y  y  y  t  y  y  mm  y  y  *  Iff!  *  k  V  y  y  y  y  y  y  y  y  m  y  y  y  y  y  ?  y  y  n  y  y  y  y  y  t  t  rj  y  y  y  y  y  y  7  w  y  y  y  7  y  y  y  y  y  y  y  y  y  %  <te  ill: y  n >  r  w  y  ?  y  y  y  y  imi  y  y  y  lil! y  mm  '•>  y  9  ?  y  y  y  y  y  y  y  •»  y  y  y  y  y  y  y  y  y  y  y  y  y  y  y  y  y  mm y  y  o  ?  7  y  >  y  y  y  y  y  y  y  y  y  y  • * :  y  y  111  y  4.5.1 Clusters I: voiceless stop + C Consonant clusters in which the first consonant is a voiceless stop are by and large the most suitable clusters for coarticulatory effects in Munduruku. In almost all combinations, a voiceless stop is phonetically influenced by the  following  onset. This is interesting  because,  phonologically, it is an onset consonant that is influenced by a coda consonant in Munduruku (see especially Chapter 7 on voicing alternation). The distinction between phonetic and phonological processes is important here. Phonetic effects do not involve the manipulation of features and are gradient; phonological processes, on the other hand, involve manipulation of features and are categorical (e.g. Keating 1996). In this section I focus on the coarticulatory effects, relating them to phonological ones. The first C C cluster to be examined consists of a sequence of voiceless stops. (Here and in the rest of the section, clusters in a box are those consisting of a sequence N + coronal, which always result in deletion of Ixl. These are examined later in §4.5.4.  (50)  Voiceless stop + voiceless stop  p  k  P  t  ka.sop.ta  o.tap.tfo.tfo  i.m9.?it.kap.kam  'I got the  'star'  'I saw a leaf.'  'to help a baby be born'  feather.' t  k  ?ot.po  o.fatta  'worm'  'my food'  ok.pot  'my son'  'my husband'  i-ma-tfoat-tfoat-m  | i.kot.kon  j  'make s.o. look'  j  'digging s.t.'  da.jek.tfo  'rodent, sp.'  'to tie s.t. up'  Heterorganic clusters ( V C i + C V ) tend to have greater duration (average of 278 ms, s.d. 2  34.4), but this result is no different from the result obtained for a sequence /p-p/ (p=0.24). They still differ from homorganic clusters in that C i is often, though not always, released. Figure 4.4 gives an illustration of a sequence Vp+tV; the arrow marks the burst release for [p]. Figure 4.4. Illustration of the closure phase in the sequence [p-t] in the Munduruku word kasop-ta 'star'.  It is also possible to have a cluster formed by voiceless stop + voiced stop. Here phonetics and phonology diverge. Phonologically, the language has a process of consonant mutation (see details in Chapter 7) in which voiced stops in onset position agree in [-voice] with a preceding voiceless stop. Voicing alternations are restricted to particular classes of morphemes and subject to certain morphosyntactic conditions; elsewhere, sequences of stops may disagree in voicing, as the examples in (51). In the cluster /p-b/, the verb -bog 'be big' alternates with -pog after a voiceless stop, but this is because consonant mutation begins to be generalized in the language. 134  (51)  Voiceless stop + voiced stop  p  t  k  b  d  tap.borj ~ tap.porj  ?ip_d39k.p9  'the leaf is big'  'armadillo, sp.'  'wood louse'  ya.ko.bat.ban  i. do .dot  'to hug s.o'  'S/he's tattooed.'  i.bik.bik  cfte.dok.don  'It's narrow.'  'to swim'  43  ' a.c^o.djot I | 'grandmother' i adjok.d^orj 'bathing'  Phonetically, however, these sequences may be optionally pronounced as a sequence voiceless + voiced or voiced + voiced, in which the coda stop assimilates voicing from the following onset. The waveforms in the figures below illustrate three sequences voiceless stop + voiced stop: /p-b/, /k-b/, /k-dj/, as produced by two speakers (JT and E M ) . The sequences in the figures 4.5 and 4.6 below show voiced and voiceless variants of voiceless stops preceding Pol in -bog 'be big', as produced by the same speaker (JT). He pronounces the sequence /k-b/ as [k-b], maintaining the difference in voicing; this is perhaps facilitated by the difference in place of articulation since the sequence /p-b/ in tsp-bog is produced as a long [b]. (In the waveforms vertical lines mark stop closures.) Figure 4.5. Waveform illustrating the sequence /p-b/ realized as a long [b]. Word: tm-bon "The leaf is big.' (JT)  Figure 4.6. Waveform illustrating the sequence /k-b/ realized as [k-b]. Word: wsk-hon 'My belly is big' (JT)  If  |lf|  ill  For the second speaker (EM), the velar stop Ikl is produced as [g] before /dj/. Figure 4.7. Waveform illustrating the sequence /k-oy realized as [g-cfe]. Word: ackok-ckon 'bathing' (EM)  4  1  Voicing assimilation varies across speakers. The waveforms below illustrate the voicedvoiceless variation in the same word, davdoddpddp 'armadillo, sp.', as produced by three speakers. The target is the sequence /p-u7. For both E M and A K , illustrated in the figures in 4.8 and 4.9 respectively, -dap-dap is pronounced as [dab-dap]. Figure 4.8. Waveform illustrating the sequence /p-d/ realized as [b-d]. Word: daydodapdap 'armadillo, sp.' (EM)  Figure 4.9. Waveform illustrating the sequence /p-d/ realized as [b-d]. Word: davdodapdap 'armadillo, sp.' (AK)  For JT, voicing distinctions were maintained and the sequence surfaces as [p-d]. Figure 4.10. Waveform illustrating the sequence /p-d/ realized as [p-d]. Word: davdodapdap 'armadillo, sp.' (JT)  The variation in voicing assimilation suggests an interpretation in terms of coarticulatory influence of voiced stops, rather than the phonological manipulation of the feature [+voice]. If we assume that [+voice] spreads to a preceding stop, two problems arise for the phonology: (i) the process is optional, and (ii) the language has already a process of voicing alternation, which is obligatory and affects stops in onset position. The differences between the two are outlined in (52). The cooccurrence of both in the same language is unlikely because they have conflicting properties. While progressive assimilation demands that a following voiced stop be realized as voiceless, the other would require the opposite: that a coda stop be realized as [+voice]. In addition, progressive assimilation targets only the stops that contrast phonologically for the feature [voice], thus excluding Ikl. Contrary to progressive assimilation, regressive assimilation is not structure-preserving, a property that is typical of coarticulatory effects. (52)  Differences:  Regressive assimilation  Progressive assimilation  is optional  is obligatory  is not structure-preserving  is structure-preserving  targets all voiceless stops  does not include Ikl  To continue with clusters, a voiceless stop may also be combined with a fricative. There are three reasons to believe that these do not form affricates. First, all stops can be combined with either Isl or /J/, except for Ixl which deletes because of the prohibition on sequences Ixl + coronal (§4.5.4). Second, the voiceless affricate /t$7 contrasts with the sequence t+f in that Ixl is always deleted in the latter. Finally, the same morpheme-initial fricative can be combined with different stops; for example, wak-sa 'my liver' versus vop-sa 'his/her liver'. (53)  Voiceless stop + fricative  p  t  k  s  5  dap.sem  Jepjep  'deer'  'two'  S9.S9t  ]  'monkey, sp.'  'I slept'  !  'my liver'  'my back'  Phonetically, the primary difference between affricates and a sequence stop + fricative is that the stop portion in an affricate is much longer than the fricative portion, as seen in Chapter 3, and illustrated in Figure 4.11. Figure 4.11. Expanded waveform illustrating the affricate /tjV. Word: bekitfat 'child, boy' (JT)  Conversely, in stop + fricative sequences, the fricative has a more prolonged period of frication, as shown in the two waveforms below.  Figure 4.12. Expanded waveform illustrating the stop + fricative sequence /k-j7. Word: wskfqbi 'my back' (JT)  Figure 4.13. Expanded waveform illustrating the stop + fricative sequence /p-s/. Word: dapsem 'deer' (JT)  The next cluster to look at is formed by a stop + nasal. For these, I have found no examples of a stop before lul or /rj/. A s for the sequence /t-n/, this is predicted to not be possible because of the prohibition on N + coronal. (54) Voiceless stop + nasal m p  n  5  o.tap.morj 'I put the leaf...'  t  o.tit.morj 'I put the flower...'  k  o.ygk.morj 'I put the hollow object...'  In a sequence stop + nasal, nasalization and stop do no overlap. The waveform below illustrates the sequence /t-m/; after the stop is released (indicated in the waveform by an arrow), voicelessness continues for a short period before nasalization starts.  Figure 4.14. Expanded waveform of the sequence /t-m/. Word: ototmon T put the bunch.' (JT)  To conclude the investigation of sequences voiceless stop + C, let us examine those formed by stop + approximant. In this group, the most interesting cluster is the sequence stop + Irl; but before examining it, I want to briefly comment on other sequences. We have already seen the details of the sequence stop + /?/ in Chapter 3, so I will not examine it here. A s for the sequences stop + Iwl and stop + lyl, the effect of the approximant is similar to that of the voiced stops examined above; that is, a glide may also cause a preceding stop to surface voiced. For example, IkJ in the sequence lk-wl in iwekwek 'It's torn' may be produced as [gw]. (55)  voiceless stop - approximant  p  t  w  r  y  ?  rip.wero  tap.rat  o.dop.yS  op.?ak  'woodpecker'  'The leaf is white.'  'my arrows'  'spear'  i.bay.wat.wan  ka.?o.ri.rit  a.ya.tfa.ya  o .jat.?a  'to help s.o'  'sand, beach'  'women'  'my food'  k  i.wek.wek  yak.rerj  jlk.ya  ak.?a  'It's torn.'  'She's pregnant'  'mosquitoes'  'house'  Sequences stop + Irl are more interesting for the following reasons. First, syllabification of the cluster /p-r/ in tapr3t 'the leaf is white' as /p.r/ (tzprst -> tap.rat) offends preference laws for syllable contact (Vennemann 1988), and constitutes a tautosyllabic cluster in many languages. The preferred syllabification would thus be *ta.pr9t. However, this is not possible in Munduruku because ""COMPLEX, shown in Tableau 4.15, determines that all C C sequences must be  heterosyllabic. A n alternative, illustrated by candidate (c), would be to insert a vowel and create a new syllable,, but this is also bad because epenthesis violates W-CONTIG. The optimal candidate is thus tsp.rst, despite its conflict with the syllable contact law. Tableau 4.15. Syllabification of the cluster /p-r/.  t-ap-rat a.  t]a.p]rat]  b.  t]ap.]rat]  c.  t]a.p]a.rat]  •COMPLEX  W-CONTIG  ONSET  ALIGN-R  *!  *  **  *!  But the phonetic details of clusters such as /pr/ are not so simple. The transition from /p/ to Irl is not simply [p.r] as the phonology says. The cluster is in fact realized with an epenthetic, non-syllabic vowel: [tap^rat], as shown in the waveforms below. Phonologically this vowel is invisible: speakers recognize only two syllables, tap.rst, and therefore two tones, L - L . Another difference between non-syllabic [ ] and a full [a] is length. In open syllables, vowels tend to be 3  longer than those in closed (see Chapter 3). Compare the two sequences [ ra] in Figure 4.15 to 3  the sequence [ara] in Figure 4.16. Figure 4.15 shows waveforms of the word tsprat [tap rat]. 3  Note that [ ] is noticeably shorter (average of 69 ms, s.d. 9.7) than a full [a] in a closed syllable 3  (mean 79 ms, s.d. 16.2, as seen in Chapter 3). Figure 4.15. Waveforms illustrating non-syllabic [a]. Word: taprat [tap'rat] "The leaf is white'. (AK/JT)  t a p  [] 3  r a  t  t a p  fl r  a  t  Now compare these to a full vowel [a] in an open syllable and preceding Irl, shown in Figure 4.16. A s a regular vowel, [a] is clearly longer (mean 180 ms, s.d. 8.2) that a transitional [ ] in a 3  similar context.  Figure 4.16. Waveform illustrating a full vowel [a]. Word: imarsn [imars n] 'to make s.t. white'. (JT) d  The question here is: 'Is this a phonetic or a phonological process?' According to the analysis developed so far, i f [ ] is treated as a full vowel in the phonology, two problems arise. 3  First, W-CONTIGUITY does not allow epenthesis at morpheme boundaries. A n d second, as a full vowel, [ ] would have to be moraic, and as such be assigned a tone on surface, but speakers 3  whistle only two tones, meaning that ['] in the word tzprdt does not bear a tone for them. Thus, the answer is: a transitional schwa is not phonological. If it is non-syllabic and not a tone-bearing unit, therefore it is not an epenthetic vowel; it serves exclusively to release the stop in order to avoid an ill-formed contact between syllables. This explains why a transitional [ ] is shorter than 3  a full schwa in a similar sequence. But voiceless stop + Irl is not the only sequence to exhibit a transitional schwa; this effect is also observed in nasal + Irl sequences, examined next.  4.5.2 Clusters II: nasal + C In clusters formed by a nasal and another consonant, the sequence nasal + Irl also exhibits a non-syllabic schwa, as in the case of voiceless stop + Irl. (56) and (57) illustrate combinations nasal + stop, voiceless or voiced; these are produced without influence of either consonant. (There are no examples available for the followings sequences: m-d, n-d, n-03, rj-rfe.)  (56)  Nasal + voiceless stop  m  k  P  t  o.kam.ta  o.tom_jj6.tfo  akom.kom  'object for  'my nipple'  'I saw the mix'  'falling into the water'  squeezing manioc' n  na.pen.po  fin.tom  ton.tfay  'centopede'  'porridge mix'  'his/her hips'  'It's true, real'  ka.?6rj.tot  tforjjbrj ap.?ip  i.karj.karj  'one long,  'broom'  'a stick for  'It's tight'  flexible object' (57)  pounding s.t'  Nasal + voiced stop b m  n  d  A i.e.rem.borj 'It's wide.'  43  a.c^em.d^em 'to arrive'  d3e.b9n.b9n  'to stumble' t)  d3e.ban.barj  d3e.d6rj.dorj  'to crack'  'to swim'  The waveforms below illustrate sequences nasal + stop. The waveform on the left contains the sequence rj-p in the Munduruku word pSrj-pa 'one finger', and the waveform on the right shows a portion of the the word bdij-bi(wat) 'bracelet', illustrating the sequence sequence rj-b.  The examples in (58) and (59) illustrate the sequences nasal + fricative and nasal + nasal respectively. There are no particular observations to be made about these sequences. Nasal + fricative  m  n  rj  (59)  s  J"  ta.jlm.j~im  'It smells a bit.'  'dog (slang)'  tfo.konjlrj  'back of the knees'  'a big toucan'  i-cfron-sarj  ^e.firufir)  'S/he's fat.'  'to walk around'  Nasal + nasal  m  n  m  n  g  l.i.nam.nam  i.morj.morj  'will go (for sure)'  'sewing s.t'  'putting s.t.'  i.m§n.m§n  ti.n3n.nen  'to tie s.t'  'water mixed with feces'  d3e.m3n.m3rj  o.tirj.rjon  'to lean'  'porridge mixed with s.t.'  'I inhaled smoke.'  But for sequences nasal + approximant, (60), one combination in particular deserves attention: the sequence nasal + Ixl, examined in detail below. (60)  Nasal + approximant r  w m  n  rj  ?  y  i.rem.ram  o.k§m.?a  'It's not so blue.'  'my breast'  ka.wen.wen  tan.rat  kan.?i  'talking'  'Its feces are white'  'I baked the meat.'  'swallow'  darj.wi  tirj.rat  norj.?a  'from'  'The smoke is white.'  'flea'  Like sequences stop + Irl, the sequence nasal + Irl is also characterized by the presence o f a transitional [ ], as shown in the waveforms in Figure 4.18 to Figure 4.20. 3  Figure 4.18. Waveform illustrating non-syllabic [a] in nasal + hi. Word: tigrSt 'The smoke is white.' (AK)  t i  8  rj  H  r  5  t  Figure 4.19. Waveform illustrating non-syllabic [a] in nasal + Ixl. Word: tirjrgt 'The smoke is white.' (JT)  t i l )  H  r  a  t  Figure 4.20. Waveform illustrating non-syllabic [a] in nasal + Irl. Word: tomrst 'The flour is white.' (AK)  t o  m  [a]  r  a  t  The sequence rj-r in particular shows that a [ ] is in fact non-syllabic. Recall the nasal velar 3  /rj/ is realized as a palatal [n] syllable-initially, thus resyllabification would leave /rj/ in onset position, preceding ["]. However, the nasal is still realized as [rj] in tirjrSt: [ti rj'r3t], not 8  [rl.ji3.rat]. Therefore [ ] does not form a syllable with the preceding stop, oral or nasal. A n 3  illustration of syllabification of sequences nasal + Ixl is given in the following tableau. Tableau 4.16. Syllabification of sequences nasal + Irl.  tirj-rat  •COMPLEX  a.  ti.rj]rat]  b.  tirj.]rat]  c.  ti.rj]a.rat]  W-CONTIG  ONSET  •!  ALIGN-R  |||||;|;|||||  *  4.5.3 Clusters III: approximant + C There seems to be no coarticulatory effect in sequences approximant + C. The only observation concerns the cluster approximant + /?/ in that the glottal is realized as a heavy type of creaky voicing, as already examined in Chapter 3. The examples below illustrate all combinations. (61) illustrates the sequence approximant + voiceless stop; (62) the sequence approximant + voiced stop; (63) the sequence approximant + fricative; (64) the sequence approximant + nasal; and finally (65) illustrates the sequence approximant + approximant. Examples for the following sequences were not found in the corpus: w-k, w-dj, w-m, w-n, w-rj, and w-w.  (61)  Approximant + voiceless stop t  tf  da.pew.pew.da  kaw.ta  'humming-bird'  'salt'  'to bite s.t'  i.rj9y_t[§rj  cfre.koy.koy  'penis'  'snake, sp.'  'S/he is quiet/sad.'  'to row'  p >  w  y  (62)  *  Approximant + voiced stops  w  b  d  ya.daw.daw  'cocoa  'S/he's scared.'  43  fruit' y  (63)  p9V.b9  poy.da  d3e.k9y._d30.036m  'snake'  'cassava'  'to hear'  Approximant + fricative  w  y  (64)  s  J" ?a.?a  'hawk, sp.'  'fish, sp.'  cke.sav.sev  o.tayji  'She has a dress'  'my wife'  Approximant + nasal m  n  5  i.mgy.mSy  a.rj9y.rj9y  'to nail s.t.'  'fish, sp.'  'to think/meditate'  w y  k  (65)  Approximant + approximant w w  y  r  y  ?  daw.rgk.?a  to.Jaw.y9  p9w.?i  'fish, sp.'  'chiefs'  'to blow'  i.yoy.yoy  d3a.ray.Pa  'It's baked.'  'orange'  (fee. way. way va.rov.r9v 'to laugh'  'It's roundish.'  I now turn to the one cluster that is phonologically prohibited: sequences Ixl + coronal.  4.5.4 Sequences coronal + coronal While there seems to be no restrictions on clusters with distinctions in manner of articulation, in the place dimension not all clusters are tolerated; in particular, the sequence Ixl + coronal is prohibited. To explain the facts, assume the representations for coronals in (67) below, following the geometry proposed by McCarthy (1988), which treats [sonorant] and [consonantal] as articulator-free features, located at the root node and necessarily present in the representation of every segment (see Halle 1995). (66)  [sonorant consonantal  [nasal]  Laryngeal  [continuant]  Place  Labial  Coronal [distr]  Dorsal  [ant]  There are nine coronal consonants in the consonant inventory of the language: /t, d, tj", cfe, s, J*, n, r, y/; these differ in terms of the following features: (i) [sonorant] distinguishes obstruents, [-son], from nasals and approximants [+son]; (ii) [continuant] distinguishes stops, [-cont], from fricatives, [+cont]; (iii) [nasal] distinguishes nasal from oral stops; (iv) [voice] distinguishes voiceless from voiced stops; and (v) [anterior] distinguishes alveolar, [+ant], from palatal segments, [-ant].  (67)  Representations for coronals (b)  (a)  d  (c)  ti-  [-son]  t-son]  [+cons]  [+cons]  [-cont] Place  [vce]  Cor [+ant]  [-cont]  [-cont]  Lar Place  Place  Cor I [+ant]  Cor I  [-ant]  (f)  (d)  [-son] [+cons]  [vce]  (g)  [+cont]  cont] [nas]  [-cont]  Lar Place  Place  Place  Cor I [-ant]  Cor I [+ant]  Cor  (h)  I  I  [-ant]  r  (i)  y  [-son]  [+son]  [+son]  [+cons]  [+cons]  [-cons]  [+cont]  [+cont]  [+cont] Place  Place  Place  Cor  Cor I [+ant]  Cor  [-ant]  I  [-ant]  Combining coronals that may occur in coda position, /t, n, y/, with coronals that occur in onset position, there are 27 possibilities for sequences coronal + coronal. From these, only sequences Ixl + coronal are illicit; the others have no restriction, but there are two for which I have not found examples: n-d and n-cfe.  17  (68)  Possibilities for sequences coronal-coronal l i i l l *t-d n-t  (?)n-d  y-t  y-d  *t-tf  *t-d3  *t-s  *t-S  (?)n-d5 n-s y-tf  y-<te  y-s  y-J  *t-n  *t-r  *t-y  n-n  n-r  n-y  y-n  y-r  y-y  Well-formed coronal + coronal clusters are illustrated in (69) and (70). Despite the two gaps in the corpus (n-d, n-cfe), I assume that clusters Inl and lyl + coronal are all licit in Munduruku. (69)  Sequences Inl + coronal (a)  n-t  fin-tom  ->  Jin.torn  ->  tan.fay  pancake-flour 'flour for pancake'  (b)  n-d  ?  (c)  n-c  t-an-tfey 3-feces-hips 'his/her hips'  (d)  17  n-d3  ?  Sequences of identical consonants were examined in §4.4.3 above, where I argued that they form geminates.  In this case, the sequence t-t is licit as it functions as a geminate phonologically.  (e)  n-s  i-en-sak  i.en.sak  3-flesh-be.sour 'The meat is sour.'  (f)  n-f  tfokonjlji  ->  tf6.kon.Ji.Ji  toucan-AUG 'a big toucan'  (g)  n-r  t-on-rat  ->  tan.rat  3-feces-be.white 'Its feces are white.'  (h)  n-y  o-sa-en-vov  lSu-3-flesh-bake 'I baked the meat'  (b)  y-t  awav-to-to-da  ->  ->  poy.da  ->  i.rjay.prj  cassava-purple-PvED-CL 'sweet potato, sp.'  (70)  Sequences lyl + coronal (a)  y-d  poyda 'sweet manioc, sp.'  (b)  y-c  i-nay__prj 3Su-thought?-be quiet 'S/he is quiet, sad.'  (c)  y-j  i-ma-kay-cfeo-m  3 Ob-CAUS-?-see-IMPRF 'to warn someone'  (d)  y-s  tfe-sav-sev  tfe.say.sey  3.Poss-dress-RED.exist 'She has a dress'  (e)  y-f  i-tayji  3-wife 'his wife'  (f)  y-n  wen5-v-nom  we.nay.nom  nut-CL-flour 'nut flour'  (g)  y-r  y-a-royjiey  ya.roy.rey  3Su-CL-RED.exist 'It's (e.g. ball) is round.'  Two coronals in sequence are illicit i f the first is Ixl. This sequence is repaired with deletion of the alveolar stop. (71)  Ill-formed coronal + coronal sequences (a)  t-d  way-dot-dot  -> dot  -»  lpl.incl-be.painted-RED 'We're painted.'  (b)  t-tf  i-ma-foatfoat-m 3-CAUS-look-RED-IMPRF 'to make s.o. look'  (c)  t-03  t-i-dodjot-djot-m  ti.d6.d50.030n  3-CL-CAUS-carry-RED-IMPRF 'bringing water'  (d)  t-s  s5t_s§t  slsat  monkey-RED 'monkey, sp.'  (e)  t-f  o-fatjet  ->  o.jajet  ->  i.r___rot=at  1-food-RED.exist 'I have food.'  (f)  t-r  i-r9t_T9t=at 3-be.white-RED=NOM 'the white moiety'  (g)  t-y  ayatfat=y§  woman=PL 'women'  (h)  t-n  ?  Coronals have been long argued to have certain properties that other consonants do not have (Paradis and Prunet 1991). Within the markedness theory as proposed by Kean (1975), Ixl is universally unmarked and coronal is the least marked articulation. Many subsequent works support this observation. Kiparsky (1985), for example, emphasized the tendency of coronals to undergo processes of assimilation, to which consonants with other places resist (see also Avery and Rice 1988a-b). Further evidence comes from consonantal harmony which mostly involves coronals (Shaw 1991). Coronals are also special because they exhibit more contrasts of place and manner of articulation than do other consonants (Keating 1991), and are frequently free to occur in 153  positions in the syllable where other consonants are disallowed (Yip 1991). Because of this, some linguists have assumed that coronals lack Place features (e.g. Avery and Rice 1988a-b; Shaw 1991; Y i p 1989, 1991; Rice 1992), or even that they have their own sonority value in the sonority hierarchy (Steriade 1982; Selkirk 1984). The restriction on coronal clusters in Munduruku poses a problem for the view that coronal segments are unspecified for place features. Part of the problem is that the phonology restricts cooccurrence of coronals by disallowing the universally unmarked consonant Ixl to be in sequence with another coronal. On the other hand, the process corroborates general observations about the special behavior of coronals, since there is a clear distinction between coronals and non-coronals in the language. The proposal formulated here accounts for the special behavior of coronals, arguing that deletion results from the interaction of two major principles in phonological theory, the Obligatory Contour Principle (OCP) and the Sonority Sequencing Principle (SSP). The idea that speech sounds are organized according to their inherent sonority is not new (see Clements 1990 for a good overview). Sonority has played a fundamental role in determining well-formed syllable divisions. A n optimal syllable is typically characterized by a string of sounds whose degree of sonority increases towards its peak (e.g. Vennemann 1988). Vowels are the best candidates for these peaks since they possess the highest degree of sonority; consonants vary: liquids are more sonorous than nasals, and these more sonorous than obstruents. In the sonority scale proposed in (72), I follow Clements (1990) in assuming that there is no further division so that coronals and other obstruents are treated alike in terms of relative sonority. (72)  Sonority hierarchy: Vowel > Glide > Liquid > Nasal > Obstruent  One of the ideas defended here is that the phonology of Munduruku organizes segments based primarily on the feature [sonorant]. Many phonological processes affecting segments refer to this feature. One of them is examined here; another case was presented in §4.4.2 with the discussion of word-initial Ixl; and Chapter 6 provides a further piece of evidence from nasal harmony. The hierarchy of Munduruku segments proposed here groups together vowels, glides, liquids, and nasals under the group of [+sonorant] segments; and stops and fricatives under the group of [-sonorant]. A n d still in conformity with the behavior of segments in the language, it is important to distinguish vowels and approximants, [+continuant], from obstruents, [-continuant]. 154  Faithfulness to [+sonorant] dominates faithfulness to [-sonorant], because in many phonological processes to be examined here, [+sonorant] wins over [-sonorant]; and as w e w i l l below, these are dominated b y faithfulness to [+continuant]. Faithfulness constraints are defined i n (73). (a)  (73)  MAX[+cont] Input [+continuant] segments have output correspondents.  (b)  MAX[+son] Input [+sonorant] segments have output correspondents  (c)  MAX[-son] Input [-sonorant] segments have output correspondents.  (d)  Ranking:  MAX[+cont] »  MAX[+son] »  MAX[-son]  Because the clusters w e w i l l be l o o k i n g at are repaired b y deleting one o f consonants, w e want to avoid another strategy such as changes i n Place specifications.  18  The constraint that  guaratees this, and w h i c h is also important f o r a process examined i n §4.6, is M A X P A T H P L A C E . M A X P A T H - P L A C E (Shorthand: M A X P - P L )  (74)  A n y input path between Place specifications and an anchor must have a correspondent path i n the output.  Furthermore, it is assumed that the special behavior o f coronals can be accounted f o r without i n v o k i n g the underspecification o f place features, f o l l o w i n g Smolensky (1993). Restrictions on the coronal place are independently motivated b y another constraint i n the grammar. I suggest an OCP constraint prohibiting any sequences o f coronal plus coronal (see Shaw 1976-1980). *COR-COR - A sequence coronal-coronal is prohibited.  (75)  *COR-COR and MAXPATHPLace j o i n the ranking to determine what coronal + coronal sequences are licit or illicit. The ranking is as i n (76). 18  Epenthesis is also banned because it violates either M - or W-CONTIGUITY constraints, which are highly  ranked in the language. Thus I will not refer to candidates with epenthetic segments in the tableaux.  (76)  *COMPLEX »  MAX[+cont] »  *COR-COR »  MAX-IO »  ONSET »  ALIGN-R »  MAX[+SOII] »  MAXP-PL  »  MAX[-SOII]  A n example of the ranking in sequences coronal + coronal is given in Tableau 4.17. The sequential prohibition on coronals does not hold of clusters such as nasal + coronal, indicating that MAX[+son] dominates * C O R - C O R to avoid deletion of a nasal coda. * C O R - C O R is dominated by ALIGN-R because, as in the case of candidate (d), alignment is crucial in deciding which coronal can be onset, and this cannot be a morpheme-final consonant. Dissimilation of Ixl to [k] passes * C O R C O R but fatally violates M A X P - P L . Tableau 4.17. Well-formed clusters: [+sonorant]-[-sonorant]. Word: Jintom 'flour for pancake'  Jm-tom  MAX  ONSET  ALIGN-R  [+cont] a. ^  Jin.Jtom]  b.  fi-]t6m]  c.  jm.]6m]  d.  jln]6m]  e.  j*in.]k6m]  MAX  MAXP-  •CORCOR  MAX-IO  [+son] PL * *! *!  *!  The next tableau illustrates two [+sonorant] coronals. A n y attempt to delete one of them is ruled out by the faithfulness constraint preserving this feature. Tableau 4.18. Well-formed clusters: [+sonorant]-[+sonoranfJ. Word: tUnrgt 'Its feces are white.'  tan-rat  MAX  t+cont] a. f  tan.Jrat]  b.  ta.Jrat]  c.  ta.n]at]  ONSET  ALIGN -R  MAX  MAXP-  *CORCOR  MAX-IO  [+son] PL * *!  *!  *  This ranking predicts that whichever [+sonorant] segment forms a cluster with another coronal, it should not be affected by *COR-COR. The following tableau shows a case in which a coronal + coronal sequence emerges in a prefix-root combination.  Tableau 4.19. [+sonorant]-[+sonorant]. Word: way-rat 'We (incl.) are white.'  way-rat  MAX  ONSET  ALIGN-R  [+cont]  MAX  MAXP-  *COR-COR  MAX-IO  [+son] PL *  a. ^  way.]rat]  b.  wa.]rat]  *!  c.  wa.y]at]  *!  *  lllllllli  IIIIIIIIIII  This ranking holds absolutely of Ixl + coronal sequences, as show in Tableau 4.20. Because neither Ixl nor Isl are sonorants, MAX[+son] is irrelevant for this cluster; yet it does not pass the O C P constraint, shown by the failure of candidate (b), therefore the morpheme-final stop is deleted. This decision is made by A L I G N - R which protects a morpheme-final consonant, as in candidate (c). Tableau 4.20. Ill-formed clusters: *t-s. Word: sasat 'monkey, sp.' sat-RED  MAX  ONSET  ALIGN-R  [+cont] a.  sa.]sat]  b.  sat.]sat]  c.  s!.t]at]  d.  sak.]sat  MAX  MAXP-  *COR-COR  MAX-IO  [+son] PL * *! *! *!  Another ill-formed cluster is illustrated in Tableau 4.21, this time with a [+sonorant]. Tableau 4.21. Ill-formed clusters: *t-r.Word: -ra-rat 'be white' rat-RED  MAX  ONSET  ALIGN-R  ra.Jrat]  b.  rat.]rat]  c.  ra.t]at]  MAXP-  *CORCOR  MAX-IO  [+son] PL  [+cont] a. ^  MAX  * *! *!  To conclude this section, one additional observation is important here. If we compare the relevance of the low-ranked constraints for the grammar, M A X - I O in particular, we see that it is  still decisive in picking the optimal candidate. To illustrate the point, consider again the word kasopta 'star' which was introduced in the beginning of this chapter (§4.2). Candidate (b) is as optimal as the winner (a), until it is evaluated by M A X - I O , which preserves input segments. Tableau 4.22. Syllabification of the word kasopta 'star'. *COMP  ka-sop-ta  W-CONT  ONS  ALIGN-R  MAX  MAXP-  [+son] PL  *COR  MAX  -COR  -10  ka.]sop.]ta] ka.]so.]ta] ka.]so.p]i.ta] ka.]so.p]ta] *!  ka.]so.p]a.ti]  e.  In the following section I present further evidence that faithfulness to input segments conforms to the sonority scale. Vowels and glides are higher than nasals, and these higher than obstruents. 4.6  The imperfective -m The imperfective aspect marker is the labial nasal consonant {-m}, which surfaces as Iral 19  after morpheme-final vowels, (77), and is deleted after glides, (78), and nasals, (79). (77)  {-m} following a morpheme-final V (a)  ?6 + m  (b)  tfa + m  (c)  (d)  19  ?6m  'to eat' (Trans)  ->  tfim  'to go'  aoka + m  ->  aokam  'to k i l l '  e-?§ + m  -»  e?5m  'to die'  There is also the instrumental marker {-m}, as in kise-m 'with a knife', that shows the same pattern.  (78)  {-m} following glides (a)  way + m  (b)  (79)  way  'to laugh'  w§y + m  ->  w§y  'to shoot with an arrow'  tabidaw + m  ->  tabidaw  'to get lost'  paw + m  ->  paw  'to blow'  6m + m  ->  6m  'to enter'  acfeem + m  ->  acfcem  'to arrive'  kon + m  ->  kon  'to eat' (Intr)  napon + m  ->  napon  'to flee, run away'  m5rj + m  ->  m5rj  'to put something down'  ?6rj + m  ->  ?6rj  'to sweep'  {-m} following nasals (a)  (b)  (c)  In contrast, after a morpheme-final stop the imperfective fuses with the stop, surfacing with the place of articulation of the stop and nasality of the nasal. (80)  Segment fusion (a)  (b)  kom  'to go down'  kap + m  kam  'to go through, pass'  dakat + m  dakan  'to cut'  kon  'to dig'  kaplrj  'to work'  ad3orj  'to bathe'  kop + m  kot + m  (c)  ->  ->  kapik + m acfeok + m  ->  There is a caveat here: the imperfective {-m} must be realized on the surface, but succeeds only i f there is a coda position available for it, or the root-final consonant is less sonorous than ImJ. In competing with a segment of equal or higher sonority, the affix loses and is deleted. The process shows a root-affix asymmetry, where the realization of the suffix is determined by the degree of sonority of a root-final segment. (81)  Root-final segment  {-m}  Vowel  realized as coda  Glide  not realized  Nasal  not realized  Stop  realized via fusion with stop  Fusion is schematically represented in (82). This process associates the Place node of a segment with another segment, and deletes the stricture features of the first, [+cons], [-son], and [-cont]. (82)  Fusion in Munduruku [+oonoonantal]  [+consonantal]  [ sonorant]  [+sonorant]  nasal] [-cont] Place  Place  Clearly, fusion violates MAX[-son] and other faithfulness constraints such as M A X P A T H PLACE, discussed earlier. It is also evident that fused forms deviate from their inputs, but can this be characterized as M A X - I O violation? I follow Kager (1999) in assuming that fusion does not violate M A X - I O , based on the fact that it does not involve the deletion of segments. Kager explains that fusion is "a 'split' correspondence between a pair of input segments and a single output segment." (p.62) In deletion, the input lacks an output correspondent.  2 0  20  Fusion violates LINEARITY, but this constraint is not discussed here. We can assume that it is ranked low so  as to not interfere with highly ranked constraints.  (83)  Correspondences diagrams (following Kager 1999) (a)  Fusion  (b)  Input:  xi  Output: MAX-IO:  Deletion  y2  zi,  Input:  xi  yi  Output:  2  Not violated  y2 Violated  To account for the facts, I propose the following representations for glides, nasals and stops. (84)  Representations for glides, nasals and stops, (a)  Glides  (b)  (c)  Nasals  Stops  +sonorant  +sonorant  "-sonorant  -consonantal  +consonantal  +consonantal  [+cont]  [+nas] [-cont]  Place  Place  Place  Given the sonority scale Glide »  Liquid »  [-cont]  Nasal »  Obstruent, the generalization is that  the imperfective is deleted in favor a root-final glide, [+continuant], or nasal, [+sonorant], and combines with a root-final stop, [-sonorant]. This is predicted by the ranking MAX[+cont] MAX[+son] »  MAXPATH-PLACE  »  »  MAX[-son] as proposed earlier. It is also necessary to  prevent the realization of the imperfective as nasalization in a vowel. This effect can be obtained via MAXPATH[nasal], following Pulleyblank (1996). (The interaction of DEPPATH[nasal] with MAXPATH[nasal] is discussed in the analysis of nasal harmony in Chapter 6.) (85)  M A X P ATH [nasal]  Any input path between [+nasal] and an anchor must have a correspondent path in the output.  Tableau 4.23. The imperfective {-m} after vowels. Word: ?o-m 'to eat' (Trans.)  ?6-m  •COMPLEX  MAXPATH  ALIGN-R  ?6]m]  b.  ?6]  MAXP-  MAX-IO  [+son] PLACE  [nas] a.  MAX  lllllllit  •!  Now compare the six candidates in Tableau 4.24. Candidate (e) is fully faithful to the input but is excluded by *COMPLEX. Candidates (c) and (d) fail MAX[+cont]; (d) fails because fusion of /ml with lyl yields in loss of the feature [+cont] oily I. Candidate (b) retains the feature [+nas] of the imperfective but is banned because MAxPATH[nas] does not allow [+nas] to be associated with an anchor other than the one determined underlyingly. Tableau 4.24. Combination glide + m; word: way 'laugh'.  5  •COMPLEX  way-m  W-  MAX  MAXPATH  CONT1G  [+cont]  [nas]  ALIGN-R  MAX  MAXP-  [+son]  PLACE  a. ®° way]  b.  way]  c.  wa]m]  *!  d.  wan]  •!  e.  way]m]  f.  wa.y]am]  •! | | | | ; | | | | ; ; | | ; % .„ :*::||: : ; : :  ;  ;;;  •! •!  A n illustration of the combination nasal + -m is given next. Again, *COMPLEX rules out both nasals as a tautosyllabic cluster in the output, candidate (b), and A L I G N - R favors napon over napdm.  Tableau 4.25. Combination nasal + -m; word: napon 'to flee, run away'. 7  7  napon-m  •COMPLEX  W-CONTIG  ALIGN-R  MAX  MAX  MAXP-  [+son] PLACE  [+cont]  *  a. f  na.pon]  b.  na.ponjm]  c.  na.po]m]  *!  d.  na.p6n.]m]a  *!  lllliiillllll  *!  The fusion of {-m} with a root-final stop is straightforward. A s an example, consider the input takgt-m -> takgn 'to cut' in Tableau 4.26. This ranking shows that it is better to have a split (fused) output than to have only one of the consonants. Tableau 4.26. Combination of stop+m. Word takat-m 'to cut'.  takat-m  •COMPLEX  W-CONTIG  MAXPATH  ALIGN-R  4.7  ta.kan]  b.  ta.ka]m]  c.  ta.kat]  d.  takat]  e.  ta.kat]m]  f.  ta.ka.t]am]  MAXP-  [+son] PLACE  [nas] a.  MAX  * *! *! *  *! P ::l *: :::|ll  *!  :  *!  :  ;  lllllliill!  Conclusion A l l the facts examined in this chapter can be generated by the general ranking schematically  represented in (86). The lines indicate crucial dominance; constraints in the same row are coranked, meaning that there is no evidence so far to ranking them relative to each other. A t the bottom of the ranking we find marginal constraints such as MAX[-son] and N O C O D A . Their presence or absence in the ranking makes no difference for the selection of optimal outputs.  (86)  Ranking CTMORA  aNuc M-CONTIG  •COMPLEX *HIBK-HIBK *HIRD-HIRD  W-CONTIG, M A X [ + c o n t ] , MAX-U,  M A X P ATH [nas]  ONSET  ALIGN-R  MAX[+son]  MAXPATH-PLACE  *COR-COR  MAX-IO  MAX[-son], NOCODA  M a n y o f the constraints introduced i n this chapter are important for other phonological processes, and w i l l be shown as needed in the f o l l o w i n g chapters.  CHAPTER 5  Phonotactics: Synchrony and Diachrony 5.1  Introduction Restrictions on the distribution of consonants and vowels are often assigned to synchronic  grammars. In this chapter I address the importance of historical information in explaining the nature of phonotactic restrictions. In particular, I pursue the idea that these restrictions accompany the evolution of languages, emerging and changing as the language evolves. This hypothesis finds support in the development of Munduruku consonants and their phonotactics. The majority of phonotactic patterns observed in the language is accidental, rooted in the various sound changes take took place in the language. These changes left gaps that need not be synchronically motivated, either phonetically or phonologically. Following Kiparsky (1982), sound change involves (i) innovation - new patterns that come into being, either through borrowings or through changes on one part of the system and which have an effect on others; and (ii) restructuring - the difference between a grammar being acquired by a child and the grammar of those whose grammar is already constructed. It is "the resulting revision in the phonological representations." (p. 3) I hope to show that restructuring takes place in different ways. For example, §5.4 deals with a case of reanalysis of a gap. The language acquired a phonotactic restriction on a sequence /d/ + nasal vowel (*dv) as a result of a secondary split caused by a change in nasal harmony. This change affected not only the phonemic inventory but also grammatical structures because of the conflict with consonant mutation (Chapter 7). This conflict resulted in the reanalysis of *dv as a language-specific constraint, with a major role in the phonology of the language. There are also cases of gaps that are repaired later in the grammar. One of these is a gap on the distribution of /dy (§5.6). This consonant was born in the language with two significant gaps: it did not occur word-initially or before a high front vowel lil. The first gap has already been repaired by novel forms borrowed from Portuguese, but the second remains. Conversely, there are gaps that emerge in the history of a segment with a previous regular distribution. This is illustrated by the history of /#/ (§5.6.) and Isl (§5.7), which synchronically do not occur before a high vowel lil, a combination that was once possible in the language.  The focal point of the historical analysis is to show that most restrictions on the distribution of segments are not systematic, in the sense that the language does not really prohibit these combinations. Rather, they are accidental gaps caused by sound changes. The distinction between systematic and accidental gaps is established here based, though not strictly, on the presence versus absence of alternations. Forms for which there is no evident variation, but are somehow absent or restricted, are likely to be accidental, and may be rooted in diachronic changes.  5.2  The synchronic patterns This section is devoted to the set of restrictions on the distribution of consonants and vowels,  as manifested synchronically. I hope to show that a purely synchronic analysis of these patterns is not satisfactory, and would require a set of language specific statements that obscure and neglect their diachronic nature. Then I proceed to examining these patterns from a diachronic perspective to show that they result from a series of sound changes, and are better understood this way. To identify the restrictions on possible combinations of consonants and vowels, a total of 1252 occurrences of consonants were counted in C V ( C ) syllables, with one instance per variant of each morpheme. For example, morphemes that participate in consonant mutation (see §5.6 and Chapter 7) have voiceless (e.g. tfd 'to go', -pa 'arm') and voiced (e.g. q^9, -bg) variants; for these, two entries were counted for each consonant: one for /tf, p/, and another for /03, b/. Those that do not vary were counted only once. For example, the two words okPa 'house' and witaPa 'rock' contain the same morpheme -Pa 'round object'; in this case, only one occurrence o f / ? / preceding /a/ was counted. Also, reduplicative morphemes were not included, with the exception of Ihl because it mostly occurs in reduplication, as we saw in chapters 3 and 4. Thus the reduplicative morpheme was considered for the distribution of Ihl in  {irom djeS'Xo go  up'), but excluded for the distribution of/w/ in dse-way-way 'to laugh'; this form was counted as having only one occurrence of /w/ before /a/. Another case of reduplication included was reduplication with a fixed vowel, as the fixed vowel Id examined in Chapter 4. For example, the reduplicative morpheme -Pe in witaPaPe 'There are rocks' was considered for the distribution of /?/ before Id.  Table 5.1 gives the relative frequency of consonants in onset position in all C V ( C ) syllables counted. The last two columns show occurrences before nasal (v) and creaky (y) vowels, which are already included in the total of 1252. The numbers in brackets are explained below. Table 5.1. Relative frequency of Munduruku consonants in CV(C) syllables.  Total  _v  _y  23  128  11  7  16  13  93  5  7  11  46  30  109  10  21  12  6  50  21  114  III!  13  12  6  20  45  64  147  12  21  a>  11  10  8  15  45  9  2  (i)  10  8  6  22  47  5  2  s  d>  11  35  12  13  72  18  9  s  37  17  n  8  2  64  7  4  m  u>  2  27  6  5  41  7  2  n  0  7  8  8  7  30  12  4  ill  15  10  7  9  41  5  3  w  16  21  14  46  16  7  y  0  3  5  4  11  23  4  5  r  14  40  22  26  31  133  22  9  h  3  3  1  2  3  12  4  1  ?  14  8  8  11  15  56  16  0  Total  197  199  238  334  284  1252  163  117  _i  _e  9  _a  0  p  27  12  33  33  b  30  14  20  t  15  7  d  25  k  rjLP]  1111 97  Based on these distributions, the following phonotactic restrictions can be stated. (1) Phonotactic restrictions (a)  /w/ does not cooccur with lol: *wo/*ow  (b)  lyl does not cooccur with III: *yi/*iy  (c)  111 does not cooccur with a creaky vowel: *?y 167  (d)  /d/ does not occur before nasalized vowels: *dv  (e)  /J7 does not occur before /a/: * f a  (f)  It}, dj/ do not occur occur before / i / : *tfi, *d5i. A n exception to these is the  alternating morpheme fitfdlq^dja'many, (g)  much, very'.  Isl does not occur before lil: *si. A n exception to *si is pasia 'to go for a walk',  from Portuguese passear. There is also basiaPa 'metal basin', also from Portuguese bacia. This word seems to be mostly used by those living in the city, and may not be familiar to most speakers, unlike pasia 'to go for a walk' which also appears in Crofts (1986). (h)  /m, n, rj/ do not occur before lil: *mi, *ni, *rji [pi], /mi/ occurs in kamifa 'shirt',  from Portuguese camisa.  In addition to the restrictions in (1), there are also restrictions on word-initial and coda consonants, as seen in Chapter 4. Recall from Chapter 4 that IT, hi occur word-initially only under specific conditions; and in coda position only the voiceless stops /p, t, k/, nasals Ira, n, rj, and glides /w, yl occur. The restrictions *wo, * y i and *?v can be attributed to the Obligatory Contour Principle (OCP). (*wo and * y i were already examined in Chapter 4.) The prohibition *ja can also be synchronically motivated as a language-specific requirement (see §5.8 for details). But excluding these that are systematic, and for which we can find synchronic motivation from alternations, the rest is better understood from a historical point of view. First of all, what is striking about the phonotactic restrictions in Munduruku is the number of consonants that cannot form a syllable with the high front vowel I'll. Similarly, a prohibition such as *dv does not appear to have any phonetic or phonological motivation. The language has two voiced stops, Ibl and Idl, and the affricate /dj/, but only the alveolar stop is incompatible with a nasalized vowel. However, as we examine the history of these restrictions, we come to the conclusion that they result from a series of sound changes. In fact, sequences such as dv, tfi, dp, si, mi, ni and rfi [pi] would have been synchronically possible i f the language did not change at earlier periods. This chapter is intended to provide a principled account of these, and show that motivations for them can only be found historically. A t the same time, it is argued that while a restriction such as 168  *dv must be regarded as a real constraint in the language (§5.4), *tfi, *c^i, *rji (§5.6), *ni (§5.5), and *si (*§5.7) are merely accidental gaps.  5.3  The development of Munduruku consonants and phonotactics  5.3.1  Background  The reconstruction of Proto-Munduruku consonants proposed in this chapter is by means of the Comparative Method, the method that has been most extensively used in historical studies of languages. The primary goal is to demonstrate that the synchronic patterns of Munduruku directly reflects its history. A s introduced in Chapter 1, this study deals with two levels of reconstruction, schematized in (2) - one at the level of family (Proto-Munduruku), which compares Munduruku and its sister language  Kuruaya (Comparative  Reconstruction),  Munduruku (Pre-Mu), which refers  and another language-internally,  to an earlier stage of Munduruku only  Pre-  (Internal  Reconstruction). In some cases a hypothesis must refer to the stage before Proto-Munduruku, Pre-Proto-Mundurukii (Pre-PMu). ^2)  ***Pre-Proto-Munduruku  **Proto-Munduruku  (Comparative Reconstruction)  (Internal Reconstruction)  Kuruya  Munduruku  Munduruku and Kuruaya are the only languages of the Munduruku family (Rodrigues 1964, 1970, 1986). Unlike Munduruku, Kuruaya is nearly extinct; there are only five aged speakers (Costa 1998; Rodrigues 1999), all of whom in their 80's and up. Previous works on Kuruaya were limited to word lists (Snethlage 1910; Nimuendaju 1930) and a preliminary description of its phonology (Costa 1998), with which I disagree on a number of points (see Picanco 2003c). Unless otherwise cited, Kuruaya data is from my own fieldwork, collected in two field  trips  (2002 and 2004) to Altamira, Para, Brazil, where I worked with two speakers: Mrs. Maria  Curuaia (82 years old) and Mr. Paulo Curuaia (89 years old). The data on Kuruaya used in this study consists of a word list with approximately 400 lexical items. The Kuruaya phonemic inventory consists of seventeen consonants (Picanco 2003c; cf. Costa 1998), shown in (3)a, and six oral vowels, with four nasal counterparts, (3)b. (3)  Kuruaya phonemic inventory (Picanco 2003) (a) Consonants  Stop  Nasal  Lab  Alv  P  t  b  d  m  n  Fricative  s  Lateral  1  Approximant  w  Pal  Velar  Glottal  k  r  (b) Vowels Nasal  Oral i  i  o  e  a  o  A s noted by Picanco (2003c), Kuruaya IV varies phonetically between an interdental voiced fricative [6] and a plain lateral [1]. The voiced stop Idl is produced further back, and is mostly realized as a retroflex [dj, except before IV where it is a plain palatal affricate [03]. The contrast Ibl-lml is observed only before oral vowels; there is neutralization to [m] in nasalised contexts. Likewise, Idl and Inl are neutralized to [n] in nasalised contexts; they are classified here as independent phonemes, but at this point of the investigation I cannot present strong evidence that they contrast in oral contexts. Like Idl, Inl also has a palatal allophone [ji] before HI. In nasal contexts, lb, d, 1, w, y, r, ?, hi have nasal allophones: [m, n,1/8, w, y,T, ?, fi].  1  1  The nasal allophone of IV is mostly realized as [1], but its realization as [5] or even [n] is also attested.  Other details about the phonology of Kuruaya will be discussed as necessary.  5.3.2 Previous comparative studies on Tupi languages The reconstruction of Proto-Munduruku and Pre-Munduruku developed in this dissertation is the first proposed for the Munduruku family. A considerable number of historico-comparative studies have been proposed for the Tupi-Guarani family (e.g. Rodrigues 1958a-b, 1985; Lemle 1971; Jensen 1989, 1990, 1998; Dietrich 1990; Mello 1992, 2000; Schleicher 1998; Rodrigues and Cabral 2002; Gildea 2002); and there are a few available for other Tupi families: Gabas Jr. (1991) compares the person-marking systems of Ramarama and Monde; Moore and Galucio (1993) propose a reconstruction of Proto-Tupari consonants and vowels; Moore (1994) deals with some syntactic aspects common to several Tupi languages; Storto and Baldi (1994) deal with vowel shift in' Karitiana; and Picanco (2003b, 2005b) compares nasal harmony in Munduruku and Kuruaya. (See also Rodrigues and Fargetti 2005 on the Juruna family, and Moore, Macedo and Lacerda 2005 on the Monde family.) Other studies have compared languages from different families to show their genetic relationship (e.g. Rodrigues and Dietrich 1997; Rodrigues 1980; Drude and Meira 2005). A s for Proto-Tupi, Rodrigues (1995) proposes a preliminary reconstruction of its phonemic inventory. The historical analysis offered here for the Munduruku family is innovative in two respects. It is the first reconstruction proposed for a Tupi family that includes a detailed step-by-step analysis of the changes, and it is the first to reconstruct phonological processes such as nasal harmony (Picanco 2005b; also §5.4 and especially Chapter 6), and voicing alternation (§5.6 and especially Chapter 7). This study is not intended to provide a reconstruction of the full phonemic inventory of Proto-Munduruku; the focus will be on the facts that are crucial for explaining particular synchronic patterns.  5.4  Secondary split: the origin of *dv The sound change which accounts for the restriction *dv in Munduruku is exemplified by  the correspondence sets d/1 word-initially, in Table 5.2, and word-medially, in Table 5.3; and n/1 which appears below in the tables 5.4, word-initially, and 5.5 word-medially. In Kuruaya [1] and [Tj are allophones of the same phoneme IV, which correspond systematically to the Munduruku phonemes Idl and Ixd respectively. The oral-nasal difference of the context is crucial.  First consider the correspondence in oral environments, word-initially and word-medially. Table 5.2. Correspondence set I: d/1 word-initially.  Munduruku  Kuruaya  Gloss  doa  loa  spider  darak  laik  bow  dacfee  lade(?)  rodent, sp.  daydo  layb(?)  armadillo  deko  leko  coata monkey  dokoa  lokoa(?)  ant, sp.  daja  lafa  fire, firewood  dat  lat  scorpion  da?I  la?i  to dance  5.3. Correspondence set I: d/1 word-medially.  Munduruku  Kuruaya  Gloss  ipada  sipala  red macaw  akadap  akalap  cocoa, sp.  madi  mali  rodent, sp.  daydo  layb(?)  armadillo  odop  obp  my arrow  edoti  eloti  your vestments  odao  olao  my leg, bone  odjodit  odolit  my uncle  kawidada  wilala  clay (Mu), sand (Ku)  tado?a  talo?a  uxi (fruit, sp.)  Now consider the correspondence set n/1, which is found in nasal contexts, both wordinitially and word-medially. Where Munduruku has the phoneme Inl, or a dialectal variation 172  between /n/ and Id/ (see §5.4.2 on this dialectal variation), Kuruaya has the nasalized allophone of IV, [1]. Unlike the correspondence set d/1, the set n/1 is not completely regular. A gap exists in the context of a high front vowel [i] so that it does not include cases of the sequences ni/Ti. The reason is that not only Munduruku lacks ni, but the corresponding sequence in this set is also missing in Kuruaya. This gap explains another restriction in Munduruku, *ni, caused by an independent development that took place in the context / I and preceded the changes examined here. This will be examined later in §5.5. Table 5.4. Correspondence set II: n/1 word-initially.  Kuruaya  Munduruku  Gloss  nobano  [nobano]  lobalo  [lomalo]  rifle, gun  norj?a  [norj?a]  Ion  [lorj]  flea  na?o-rek  [na^orek]  la?o-renan [la?ofenan]  lizard  da?6-rek  5.5. Correspondence set II: n/1 word-medially.  Kuruaya  Munduruku  Gloss  olarem  [olarem]  I stink.  napon [napon]  lapan  [lapan]  to run away/flee  wen§-y  [wenay]  welai  [welai]  Brazil nut  o-ba-nS  [ob|n§]  i-i-la(?)  [tlla]  his/her finger/toe nail  o-n§y  [on§y]  olay  [olay]  my teeth  o-nan  [on§n]  olan  [olan]  my feces  onarem  [onafem]  odarem  The regularity of the correspondence sets d/1 and n/1, where two distinct phonemes in Munduruku correspond to an allophonic variation in Kuruaya, suggests a historical change that not only changed the pronunciation of an original segment, but also introduced a new phonological contrast in the language: Idl versus Inl. This type of differentiation, termed  'secondary split' (Hoenigswald 1960), refers to a process by means of which a change elsewhere in the language causes the allophonic variants of a phoneme to become independent without "...canceling the phonetic difference between the allophones in question" (p. 94). This is the proposal I offer for Munduruku, and which sheds light on important aspects of its phonology, in particular on the restriction *dv. The hypothesis is that *dv originated from a secondary split, caused by a change in nasal harmony. Except for the oral-nasal distinction of the environments, which produced oral Idl and nasal Inl outcomes, the change is unconditioned and affected every lexical item containing the proto-phoneme. The details of the change in nasal harmony will be examined in Chapter 6. Here I only summarize the facts that are important for the discussion about the origin of *dv.  5.4.1 The change in nasal harmony To account for the emergence of *dv in Munduruku, we must first understand its relation to nasalization. Nasal harmony is a pervasive phenomenon in both Munduruku and Kuruaya; the process spreads the feature [+nasal] from a vowel to other segments on the left. In Munduruku nasality is assimilated by [+sonorant] segments and blocked by [-sonorant]. This is illustrated in (4); (4)a shows assimilation and (4)b blocking. (4) Nasal harmony in Munduruku ar§  ->  9f§  'maracana bird'  nobano  ->  nobano  'gun, rifle'  lw§y  'to shoot s.t. (with an arrow)'  en§n  'your feces'  tjokon  tjokon  'toucan'  ikoero  ikoefo  'fly'  kad3araw?a  'a big pot'  i-w§y 30b-shoot e-n§n 2-feces  kad_araw-?a pot-CL  ->  1-Poss-cotton-CL  In Kuruaya nasality is not blocked. Both [+sonorant] segments and [+voice] stops assimilate nasality, surfacing with nasal allophones; [-voice] segments are transparent. (5)a shows assimilation of [+sonoranf], (5)b assimilation of [+voice], and (5)c transparency. (5) Nasal harmony in Kuruaya (a)  pawa  -»  pawa  'banana'  ant  ant  'anum (bird, sp.  wela-1  welai  'Brazil nut'  lobalo  lomalo  'gun, rifle'  de-ya  neya  'they'  omiloa  'my chin'  pafawato  'macaw, sp.'  poraka  poraka  'two'  w-e-aikon  weaikon  'my bench'  Brazil.nut-CL  (b)  3-PL o-bi-loa 1-mouth-?  (c)  parawa-to macaw-purple  1-Poss-bench  In comparing the two systems we can easily identify their similarities and differences. For similarities, both languages exhibit leftward nasal harmony and in both the trigger is the rightmost nasal vowel in the morpheme. They differ with respect to the number of targets, which in Kuruaya also includes [+voiced] stops, and in the fact that Kuruaya has a system with transparency whereas Munduruku has one with opacity.  Where both systems are similar, we can reliably suppose that they reflect the system of the proto-language. Thus Proto-Mundurukii had leftward nasal harmony and the trigger was the rightmost vowel in the morpheme. Where they differ, we must then ask which language mostly reflects the proto-system. For reasons that will become clear below, I assume that nasal harmony in Kuruaya mostly reflects nasal harmony in Proto-Munduruku, where targets were [+sonorant] and [+voice] segments, and [-voice] segments were transparent. Therefore, the development is from a system with assimilation and transparency to a system with assimilation and opacity, and from a system where [+sonorant] and [+voice] were targets to a system where only [+sonorant] is the target. A n account of nasal harmony is proposed in Chapter 6; of interest here is the relation between the historical change in the system and *dv in Munduruku. Assuming that Kuruaya is closer to Proto-Mundurukii than Munduruku, the hypothesis is that the correspondence sets 611 and n/1 discussed above are the reflexes of an allophonic variation determined by nasal harmony. In other words, the Munduruku phonemes 161 and /n/ developed out of a single consonant, a consonant with oral and nasal allophones. Let us represent this consonant by * * L which stands for 'lateral', probably a voiced lateral fricative [I3], given the variation [1] ~ [6] observed in Kuruaya, as described in §5.3.1. Like Kuruaya IV, the protoconsonant * * L had two allophones: one oral, [L], occurring in oral contexts, and the other nasal, [L], occurring in nasal contexts. (6)  **L  [L]/_v  Proto-Munduruku  [I]  IJ  Suppose now the sound change in (7). A t an intermediate stage of the language, PreMundurukii, * * L developed into *d by a process of consonant fortition, and which affected word by word containing * * L throughout the entire lexicon. If nasal harmony in Pre-Munduruku was still characterized by assimilation and transparency, the pre-phoneme *d became of course a target for nasalization. (Recall that [+voice] consonants were amenable to nasalization, surfacing with nasal variants.) Thus *d exhibited an alternation similar to that of the proto-phoneme * * L : [d] preceding an oral vowel, and [n] preceding a nasal vowel.  (7)  Fortition (a)  Pre-Mundurukii  Proto-Mundurukii **L  *d  [d]/_v  [L]/_v  [L] / .  [n]/_v  Fortition can be said to have been a change in manner, typically attested in changes by lexical diffusion (Labov 1981). If * * L was a voiced lateral fricative [I3], the change * * L > *d then basically involved a change in the stricture feature [continuant], as schematized in (8). (8)  **L  >  *d  [sonorant]  -  -  [continuant]  +  [consonantal]  +  +  [coronal]  +  +  [voice]  +  +  >  -  * * L > *d is illustrated by the examples below which contain nouns with phonemic oral and nasal vowels, and vowels nasalized via nasal spread. By fortition, all instances of the protophoneme * * L changed into *d in Pre-Mundurukii, and by nasal spread *d was realized as [n] before a nasal vowel, and [d] otherwise. (For expository reasons, tones are not marked in reconstructed forms.) (9)  Proto-Munduruku  Pre-Mundurukii  **Loa  [Loa]  >  *doa  [doa]  'spider'  **L6rj  [Lorj]  >  *d5rj  [norj]  'flea'  **LobaL6  [LomaLo]  >  *dobado  [nomano]  'gun, rifle'  **Lapan  [Lapan]  >  *dapan  [napan]  'to run/flee'  Let there now be a change in nasal harmony: the pre-system with assimilation plus transparency developed into another with assimilation plus opacity, thereby causing, in many cases, loss of the environment that conditioned the allophonic variation [d]/[n]. A n example of  this situation is provided by the Munduruku words nobano 'gun, rifle' and napon 'to run away/flee' which correspond to Kuruaya lobaldand lapan respectively. Here Munduruku has lx\l where Kuruaya has the nasal variant [I]. (10)  Kuruaya  Munduruku  (a)  2  1 6 m al o  n o b a n o  (b)  l a p a n  n a p o n  The conditioning environment for the variant [n] was lost in cases where the vowel was not contrastively nasal but nasalized via assimilation, as in Pre-Mu *dobadd 'rifle/gun' and *dapdn 'to run/flee'. This created a new environment for [n], which may now occur before oral vowels, and is, therefore, in opposition to [d]. A t this point the grammar turned the allophones [d] and [n] into distinct phonemes, without affecting the phonetic differences of both. But the process of phonologization had an effect on the distribution of the phoneme /d/. While lx\J was phonologized in both oral (e.g. in cases with an intervening voiceless stop) and nasal contexts (where [n] immediately preceded a nasal vowel), the oral stop Id/ was phonologized only in oral contexts, as illustrated in (11). The change *d > d/n did not involve a change in a particular feature. Both [d] and [n] coexisted in the language at the stage nasal spread was altered. The secondary split happened because, since the conditioning environment for the nasal variant was lost in cases such as [nobano], speakers had to determine the underlying forms for the former allophones. These were, by Lexicon Optimization, the ones closest to the outputs, namely /d/ and Inf.  2  The correspondence set b/m, as in nomano/nobano, is examined in chapters 6 and 7. To anticipate, it is argued  that Pol and /ml were already independent phonemes in Proto-Munduruku. By the time the change in nasal harmony took place, the underlying form Ihl was retained, or the outcome would have been similar to the alternation d/n, where the nasal variant was preserved.  (11)  From Pre-Munduruku to Munduruku (NH=nasal harmony) Pre-Munduruku  *doa  *dobado  *dorj  N H (not blocked)  [doa]  [nomano]  [norj]  N H (blocked) + Secondary split  doa  nobano  norj  doa  nobano  norj  [doa]  [nobano]  [norj]  Munduruku  This brings us to the synchronic restriction *dv, which is nothing but a reflex of the distribution of the oral allophone of the pre-phoneme */d/. In other words, *dv emerged from a series of diachronic changes; it is therefore a historical accident. But how do these changes affect the synchronic grammar of Munduruku? Is *dv a real constraint in the language or is it simply an accidental gap as proposed in Chapter 4 for the absence of word-initial Ixl in the native vocabulary? I argue next that *dv is the reanalysis of a gap, and therefore, a language-specific constraint.  5.4.2 The reanalysis of a gap The diachronic change * * L > *d > d/n must be regarded as phonemic because a new contrast between Idl and /n/ was created. B y Lexicon Optimization, the synchronic grammar of the language must assign Id/ and Ixil to the underlying representations of the former allophones of Pre-Munduruku *d. This left a gap on the distribution of Idl, which is now a restriction on its cooccurrence with a nasalized vowel (*dv). But the change had also an effect on grammatical structure. It created a third variant in an already existing morphophonological alternation: consonant mutation. The process alters voicing in morpheme-initial consonants which may surface voiceless or voiced depending on the preceding segment (see details in Chapter 7). The alternation is between p/b, t/d, and tfVcfe. The voiceless variants surface following consonants, and voiced ones following vowels and glides. Consonant mutation is illustrated in (12).  (12)  Consonant mutation (a)  (b)  (c)  ayarjat  patet  tojaw batet  woman name  chief  'woman's name'  'chiefs name'  tawe  dapsem  toy  monkey blood  deer  blood  'monkey's blood'  'deer's blood'  o-t-ap-tfo-tfo  o-t-a-cfeo-ifeo  lSu-3-CL-see-RED  lSu-3-CL-see-RED  'I saw it (e.g. a leaf)'  'I saw it (e.g. corn seeds)'  doy  name  O f interest here is the alternation t/d. This is an old phenomenon in the family since a 3  parallel alternation is found in Kuruaya. A s expected, the voiced variant /d/ in Munduruku corresponds to IV in Kuruaya. Munduruku  Kuruaya  Gloss  (a)  tap / dap  tip / lip  leaf  (b)  ta/da  ta/la  seed  (c)  tap / dap  tap / lap  hair  (d)  ti/di  ti/li  liquid  (13)  Before underlying nasal vowels, the alternation is t/n in Munduruku, corresponding to t/fl] in Kuruaya.  3  In Chapter 7 I present all the details of these alternations and give a historical explanation of the synchronic  patterns (see also §5.6).  (14)  (a)  (b)  tawe  n§y  dapsem t§y  monkey tooth  deer  'monkey's theeth'  'deer's teeth'  wen§-y-nom  Jin-tom  nut-CL:nut-CL:powder  pancake-CL:powder  'Brazil nut mix'  'mix for pancake'  (Kuruaya: tay / [l]ay)  tooth  (Kuruaya: torn /  [Tjgm)  But where the vowel was nasalized by spreading, in cases similar to nobano, both oral and nasal forms are attested in Munduruku. I believe this is a dialectal variation, as speakers from different villages tend to prefer one form over the other. In one dialect, dialect B , the nasal variant prevailed, including in forms that were involved in consonant mutation. In the other dialect, the oral variant was retained i f the vowel was not underlyingly nasal, probably by analogy to the alternation t/d illustrated in (13) above, that is far more frequent. (15) (a)  Dialect A  Dialect B  waje daka  waje nak3  'cocoa tree branch'  daja naboe  'burning coal'  cocoa branch  (b)  da]a  daboe  firewood ember  The major effect that the change *d > d/n had on consonant mutation was the conflict in the choice between /d/ and /n/ in representing the voiced variant in contexts where nasality was lost. In the context of an underlyingly nasal vowel, the voiced variant is unambiguously /n/. This provides a valuable argument for the assumption that the gap left on the distribution of l<3J has been reanalyzed as real constraint, *dv, defined in (16). (16)  *dv - The sequence dv is prohibited.  Consider the following examples from dialect A . Nasal harmony spreads to the left, nasalizing all [+sonorant] segments it encounters. In (17), nasalization should reach /d/ but  nasalization is blocked right before a syllable containing /dV, and that syllable surfaces oral. (17)  (a)  y-abi-dawSn  ->  'to sharpen a tip of s.t'  o.da.fem  'I stink'  ->  da.?6.Tek  'lizard'  ->  o.da.e  'S.o. picked me.'  3-tip-sharpen  (b)  o-darem lSu-be.fetid  (c)  da?o-rek lizard-?  (d)  o-dae 1 Ob-choose  In dialect B, the corresponding forms are fully nasalized. (18)  (a)  yablnawSn  (b)  onafem  (c)  na?6rek  (d)  onae.  We can summarize the patterns as in (19). The dialects differ only with respect to the environment where a vowel was a target in nasal spread; before an oral vowel the variant is invariably /d/, and before an underlyingly nasal vowel the variant is always /n/. (19) Summary of the patterns _v Dialect A  d  n  Dialect B  d  n  V...V  These patterns are complex because they involve the interaction of three processes: (i) the change *d > d/n and change in nasal harmony, (ii) nasal harmony as it is synchronically, and (iii)  consonant mutation. This chapter deals with (i), and attempts to show that *dv must be regarded as a language-specific constraint, not an accidental gap. The accounts of nasal harmony and consonant mutation are proposed in chapters 6 and 7 respectively. This means that the alternations in (19) will be explained step-by-step as all processes involved are examined. The constraint *dv is crucial for the analysis, and the reason why it must be posited for Munduruku is as follows. Consider the proto- and pre-forms of alternating morphemes in (20); (20)a illustrates a morpheme with an oral vowel (_v); (20)b illustrates * * L followed by an underlying nasal vowel (_v); and (20)c and (20)d show cases of a nasal vowel somewhere in the morpheme but not immediately adjacent to * * L (_v...v). In the examples, only the voiced variants are provided, but recall that * * L and *d had a voiceless variant [t] postconsonantally. Also, recall that nasalization in Proto-Mundurukii was not blocked, thus * * L was realized as [L] in the morphemes in (20)b-d. In Pre-Munduruku, every instance of * * L changed into *d by fortition, as discussed earlier. Since nasalization did not change until after Pre-Munduruku, *d also had a variation [d]/[n], similar to the alternation [L]/[L]. (20)  Proto-Mundurukii  Pre-Munduruku  (a)  **Lao  [Lao]  >  *dao  [dao]  'leg/bone'  (b)  **Lay  [Lay]  >  *day  [nay]  'tooth'  (c)  **Lak9  [Lak§]  >  *daka [naka]  'branch'  (d)  **Lae  [Lae]  >  *dae  'to choose'  [nae]  Between Pre-Munduruku and the modern period, nasal harmony underwent changes. A t this point, obstruents became  opaque, blocking nasalization in cases such as (20)c. As a  consequence, the oral and nasal allophones of *d became independent. However, a complication arose in consonant mutation. This initial alveolar in alternating morphemes, * * L in P M u and *d in Pre-Mu, had three variants, [t, d, n], regularly distributed according to adjacent segments: [t] after consonants, [d] before oral vowels and [n] before nasal vowels. Differentiation of [d] and [n] brought about a conflict in consonant mutation, especially in cases where there was an intervening obstruent (e.g. Pre-Mu *daka 'branch'), since this was the context that established the contrast between Id/ and in/. But did phonologization of [d] and [n] also apply to the morphemes involved in consonant mutation? A t first glance, this seems to  be the case in dialect B where the form is realized on the surface as nak§, contrary to dialect A where we find dakS. I will save the discussion of this controversy for Chapter 7, after we examine nasal harmony (Chapter 6), and the details of consonant mutation. Aside from these dialectal differences, there is a good reason to argue for the constraint *dv in the synchronic grammar of Munduruku. First of all, by positing *dv, the chances that the oral variant in words such as dae'to choose' will surface as *dde are excluded.  4  Tableau 5.1. The effect of *dv  dae  Faith  *dv  a.  dae  b.  dae  *!  Secondly, no other choice exists in cases where the vowel is underlying nasal (e.g. day versus nay), so that the output will always be nSy. This is the pattern that figures in both dialects, and despite the differences in the selection of oral and nasal variants, neither A nor B have outputs such as *dde and *ddy. Tableau 5.2. dv again  *dv  day a.  nay  b.  day  Faith (*)  *!  Finally, the constraint *dv also captures the fact that neither dialect has yet acquired sequences dv, either via borrowing or nasal spread. Prior to the change in nasal harmony, *dv was not required because the realization of Pre-Mundurukii *d as [n] was entirely predictable from nasalization. After the change, the choice between [d] and [n] came to be in conflict with  4  An output in which nasalization spreads all the way to /d/ (i.e. nde) also satisfies *dv, and this is the situation  in dialect B. For further details, see Chapter 7.  nasalization, especially because [n] is no longer an allophone of Idl, it has a phonemic status on its own. The patterns in consonant mutation are a good example of this conflict. When speakers produce dae 'to choose' or darem 'to be fetid' as [da.e] and [da.fem], they violate the requirements demanding nasalization to spread to as many eligible targets as possible, but they do so in order to satisfy a more important requirement in the language, namely *dv. The gap left on the distribution of 161 after phonologization of 161 and Inl is now a real constraint, and a crucial one. In addition to the change * * L > *d > d/n, which explains the restriction on the cooccurrence of the voiced alveolar stop and a nasalized vowel (*dv), several sound changes took place in the language, giving rise to other patterns observed synchronically. I now turn to the restriction *ni and the historical change that gave rise to it. 5.5  The origin of *ni In the previous section I called the reader's attention to a gap in the correspondence set n/[Tj:  there were no cases of the correspondence before hi. This can be attributed to the fact that the change * * L > *d > d/n did not include the sequence **[LYJ, otherwise ni would have been developed in the language, as did the sequence di: * * L i > *di > di. Synchronically, *«/', or *rii, is another phonotactic restriction in Munduruku. In this section I tentatively examine the origin of *ni in the language. Very few vestiges exist that can reliably explain it, but a few words may shed light on a possible historical explanation. I believe that * * L i disappeared from the family before Proto-Munduruku. The primary reason is because Kuruaya lacks the sequence 7f, although it has li, as in -//* 'liquid', and ni, in which case Inl is phonetically realized as nasal palatal [p], as in ini [ijii] 'hammock'. Three related words allows us to establish the correspondence set given in Table 5.6. However, this set is found to occur only intervocalically, a restriction for which I offer no explanation here.  Munduruku  Kuruaya  Gloss  waretay  wariitay  warem-ap-?a  wirfi-pa? [wirfipa]  [wajfitay]  najapalm genipa  weriipa? ar5  mi  rjrfl]  hammock  The hypothesis is that the proto-sequence **L1 underwent a sound change independent of other Proto-Munduruku **L's, and before the change * * L > *d > n took place; this is the reason why it did not include sequences **L1. Let us suppose that * * L was phonetically realized as [n] before the nasal high vowel hi. (See Clements and Sonaiya (1990) on a similar realization of sequences /lv/ in Yoruba.) Abstracting away from other changes - e.g. changes in vowel quality and the absence of nasalization in the Munduruku word waretay, I hypothesize that the proto-sequence **Lf, realized as [rii], developed into Pre-Munduruku Irl, which in turn merged with other /r/'s in the language. The change is shown below. (21)  **L1 [rii] > *rv:  Proto-Munduruku  **iLi  [irii]  Pre-Munduruku  *ifi  [ifi]  Munduruku  ara  [ffa]  It may be that «f was not simply the phonetic realization of **L1, but it had already been phonologized in Proto-Munduruku, suggesting that the change **L1 > **rii took place in PreProto-Munduruku, the  stage before Proto-Munduruku. There  are two reasons for this  assumption. One is the fact that Kuruaya lacks the sequence /fiV, although it has /rii/. Another is that, i f * * L > *d took place between Proto- and Pre-Munduruku and did not include the sequence **L1, then these forms had already been phonologized as **rii, and therefore were no longer associated with * * L . Chronologically, Pre-Proto-Munduruku * * * L occurred before all vowels, including hi. Between Pre-Proto-Mundurukii and Proto-Mundurukii, ***L1 changed into **rii.  Between Proto-Mundurukii and Pre-Mundurukii, **rii developed into *ri, whereas * * L developed into *d. (22)  Relative chronology  Pre-Proto-Mundurukii  Proto-Munduruku **vrii  Pre-Mundurukii  ***  ***L  **L  *vn  Munduruku  vrv  *d  d  n  ***Loa  ***L6rj  **irii  **Loa  **Lorj  *ifi  *doa  *dorj  gr§  doa  norj  'hammock'  'spider'  'flea'  i L 1  But a problem arises when we look at the word "hammock" in other Tupi languages, shown in Table 5.7. Rodrigues and Dietrich (1980) reconstruct the form *ini 'hammock' for ProtoTupi-Guarani (PTG), and Moore and Galiicio (1993) reconstruct *e/irii for Proto-Tupari (PTu). Karitiana is the only language to have hi, like Munduruku, whereas most Tupian languages have Ini. (The word in Mekens was provided by A . Galiicio, in Gaviao by D. Moore, in Karitiana by L . Storto, in Mawe by S. Meira, and in Aweti by S. Drude.) Table 5.7. The word "hammock" in Tupi.  PTG  PTu  Mekens  Gaviao  Karitiana  Mawe  Aweti  Mund  Kuruaya  *iril  *erii/irii  eni  ini  ere-mi  ini  ?im  ara  ini  Given this, there is little comparative evidence for the change * * * L > **n before I'll, but the correspondences in Table 5.7 do show that **rii > *fi > rv is a plausible route. Furthermore, we cannot ignore the fact that Kuruaya does not have the sequence IT, and that there are vestiges of the correspondence rv/rii in the Munduruku family. The question as to whether it developed from the pre-proto-sequence ***L1 can only be answered i f the majority of Tupian languages are considered, and a comparison of this sort is beyond the scope of this study.  5.6  The origin of *t$i, *dy, and *ni The change * * L > *d > d/n was not the only change in the phonemic inventory of the  language. There was another, much more complex series of changes by means of which a single consonant developed into four others. These changes are diagrammatically represented in (23). A step-by-step reconstruction will be provided below, but the reader should keep in mind that we are going to look at changes that took place at three different stages: (i) between Pre-Proto- and Proto-Munduruku (§5.6.1), (ii) between Proto- and Pre-Munduruku (§5.6.2), and (iii) between Pre-Mundurukii and the present (§5.6.3). These changes explain the following phonotactic restrictions: *tfi, *cfei, and *rji [pi]. Four Munduruku sounds - three are phonemes, /j", tf, ty, and the other, [n], is an allophone of the nasal velar /rj/ syllable-initially - are all historically linked to a single consonant: PreProto-Mundurukii * * * C , which stands for a postalveolar stop, probably the (IPA) palatal stop Id. ***C had two surface realizations: [tTJ preceding the high front vowel, and [C] elsewhere. 5  Between Pre-Proto-Munduruku and Proto-Munduruku the allophone [C] became voiced intervocalically, hence [D], and an affricate [ffj in other contexts (i.e. word-initially and following a consonant); these in turn merged with other ***tf's (§5.6.2). Because of nasal harmony, [D] probably had a nasal realization [N] in nasal contexts. This is corroborated by the fact that in Kuruya the contrast between the voiced stops fb, df and nasals /m, n/ is neutralized to [m, n] in nasal contexts. Between Proto-Mundurukii and Pre-Mundurukii (§5.6.2), proto-**D changed into a voiced affricate  carrying along the allophony determined by nasal assimilation: [dj] preceding oral  vowels, and [p] preceding nasal vowels, except in consonant mutation (§5.6.3). A t this stage the change in nasal harmony took place (as explained in section 5.4.1 above). This also had an  5  1 must clarify that even though /f, tf, 05/ and [n] are historically linked to ***C, they all have other sources as  well. For example, there are at least three sources for /j7: ***Ci and Proto-Mundurukii **k V and **J" (see §5.8). J  Likewise, not all /qV's come from ***C; some of them were introduced in borrowings, e.g. dpromo 'pumpkin', from Portuguese jerimum.  impact here, propelling the split of *&, into two, and the merger of its nasal allophone [p] with the nasal velar /rj/ (i.e. a primary split). (23)From Pre-Proto-Munduruku ***C to Munduruku /JV, /tj7, Ity, [p]. (Changes that took place between stages may not be chronologically parallel.) ***C  Pre-Proto-Munduruku **C^tj/_i Voicing o f * * * C  C / elsewhere  [nasal] assimilation **tfi  " f / elsewhere [D]/v_v  Proto-Munduruku  **D [N]/v_v  **D -> *d3  Pre-Munduruku Primary split  <5  Munduruku  It is precisely the allophonic alternation [tfi]/[C] in the pre-proto-stage which has given rise to the synchronic prohibitions *t$i, *c^i, and *rji [pi]. A t each stage in the history of * * * C , the resulting forms underwent changes independently. In Proto-Munduruku ***C had already been split into **tf and * * D ; later in Pre-Mundurukii, it was split into *tf and  and finally into /J, f,  dy and [p] in the modern stage. A s it can be noted in the diagram, the phoneme l&J was born in the language with two gaps in its distribution: (i) it never occurred before lil, and (ii) it was restricted to intervocalic position. The latter restriction has already changed because of some borrowings that came into the language (§5.6.1).  As for /rj/, the restriction *rji is more recent, and was caused by the sound change *rj > j" / _ i . Clearly, then, only i f we consider the development of * * * C , can we provide a satisfactory explanation for the patterns.  5.6.1 Between Pre-Proto-Munduruku and Proto-Munduruku The Munduruku words below have an ancestor in common, the voiceless palatal stop * * * C , in the pre-proto-language. (24)  (a)  jlk  (b)  jjl  (c)  otayji  (d)  tfa  (e)  dacfce  (f)  rage  In (24)a-c two changes took place: first, ***C becomes an affricate [rj] before a high front vowel in Pre-Proto-Munduruku; and second, the stop component of [tfl is lost, and [rj] becomes simply a fricative [J] in Munduruku. Only the evidence of Kuruaya allow us to reconstruct this route. If we compare the sequence fi in Munduruku with related forms from Kuruaya, we obtain the correspondence set of Table 5.8. Before the high front vowel / i / , where Munduruku has [J] Kuruaya has the palatal affricate [cfe], which is an allophone of/d/ in this context.  Munduruku  Kuruaya  i-fi  i-di(?)  [Icfei?]  his/her mother  dik  [**]  mosquito  wejik  wedik  [wed3ik]  potato  bojijim  biydim  [biycfcim]  fear of s.t.  daydo-jiji  laylo-di?  [laylod3i?]  armadillo, sp.  ijl-ba  Idl-bi?  [i<fel-bi?]  tropical creeper  Gloss  This provides evidence for the reflex of the affricate [tfj before the high vowel, but does not provide evidence for * * * C itself. For that we must establish another correspondence set qyd, given in Table 5.9: Munduruku /dj/ corresponds to Kuruaya Idl, phonetically a retroflex [dj. This set has three important restrictions: (i) it is found to occur only between vowels; (ii) it is never found in the context of nasalization; and (iii) it is in complementary distribution with the set jVd, i.e. it does not occur before a high front vowel lil.  Kuruaya  Gloss  kidap  shelter  acfcore  adore?  to be old  adjo / adjo  ado  what  dadje  lade?  peccary  atfeok  adok  to bathe  kayac&e  kiyade  yesterday  o-tje-ifeacfea  o-tje-dada?  I defecated.  o-de-po  S/he is lying.  i-dotfe  It's there  o-do-la  I cooked it  6-dolit  my uncle  e-we-di  with you  Munduruku  1 Su-CoRef.Poss-defecate o-cjje-po 3Su-CoRef.Poss-lye i-cfeotfe 3-be.there o-c&o-da lSu-30b-cook o-c&odit 1-uncle e-we-cfca 2-Refl-together  In the context of nasalization we find the third correspondence set in this group, which is in complementary distribution to both jVd and dyd. The palatal allophone of Munduruku /rj/ [p.], corresponds to Kuruaya Inl. Like the correspondence c^/d, the set [p]/n is also restricted to intervocalic position, and does not occur before the high vowel / i / . (Recall from §5.5 that the Kuruaya sequence rii [jfi] corresponds to the sequence rv in Munduruku; see Table 5.6 above.)  Munduruku  Kuruaya  Gloss  tarje  ta[p]e  tane  mouse  pirja  Pi[p]a  pina  fish-hook  akorje  ako[ji]e  okplne?  M u : necklace; K u : belt  orjebi  o[ji]ebi  onebi?  my armpit  yarjoba  ya[ji]6b9  yanobi?  his neck  orjebit  o[ri]ebit  onebit  my grandson/daugther  arjoka  a[ji]6ka  anoka  place  owarjo  owa[n]6  owano  my brother/sister  This gives three sounds in correspondence; (25)  Kuruya  [<fe]ik  i[<fe]i?  la[dje?  ta[n]e  Munduruku  [J]'ik  j[T]i  da[d3]e  ta[p]e  'mosquito'  3-mother  'peccary'  'mouse'  and three sets in complementary distribution: (26)  Summary of the correspondence sets _i J/d  v_v  v_v ji/n  These sounds have in common the property of being realized in the postalveolar region, except for [n] which is, impressionistically, alveolar. This suggests that the original sound out of which they developed was also postalveolar; hence ***C is [+coronal, -anterior]. Before a high front vowel, both Munduruku and Kuruaya exhibit a plain palatal - [dj] in Kuruaya and [f] in Munduruku - but these diverge in manner and voicing. Before examining a 193  solution to this problem, let us first suppose that ***C was realized as a voiceless affricate [rj] before a high front vowel, and [C] elsewhere. (27)  ***C  rj / _ i  Pre-Proto-Munduruku  C / elsewhere  In Munduruku the voiced reflexes [05] and [p] are only found intervocalically, and do not include the high front vowel  Synchronically, both /dj/ and /rj/ are rare word-initially. For  example, there are only four nouns in the corpus that begin with IfoJ: djardy 'orange (tree)', dpromo  1  pumpkin', kaq}jaraw?a ' a big pot', and dprawi 'a table for domestic use', all of which  are borrowings from Portuguese: laranja, jerimum, caldeirdo and jirau, respectively. Kuruaya has IdJ in the phonemic inventory, which occurs freely word-initially: da 'one', doridoy 'rabbit', etc. However, I have not found any cases of the correpondence dj/d in any other position than intervocalically, suggesting that not all /d/'s of Kuruaya originate from * * * C . In Munduruku, on the other hand, the primary source for /cfe/ is the allophone of ***C that became [+voice] intervocalically. Thus the distribution of /d$/ reflects the distribution of this allophone; that is, it is restricted to intervocalic position in the native vocabulary; its occurrence in word-initial is due to borrowings. The restriction on the correspondence <fe/d and j