Acoustic Correlates of Lenis and Fortis Stops in Manitoba Saulteaux by Adam J.R. Tallman A Thesis submitted to the Faculty of Graduate Studies of The University of Manitoba in partial fulfilment of the requirements of the degree of MASTER OF ARTS Department of Linguistics University of Manitoba Winnipeg Copyright © 2011 by Adam J.R. Tallman i Abstract This study investigated some of the acoustic correlates of lenis-fortis contrast in Saulteaux Ojibwe based on speech from six speakers in Manitoba. Four acoustic correlates of lenis and fortis stops in intervocalic position were measured; consonant duration, postaspiration, preaspiration and sonority level. It was assumed that although the speakers displayed much variation in terms of what correlates marked the contrast overall the contrast would be maintained with a similar degree of robustness across the speakers. It was hypothesized that despite the ubiquitous variation, the speakers would trade off acoustic correlates in order to maintain the contrast. Bias-reduced multiple logistic regression models were used in order to assess the (non)importance of each correlate by speaker. Multi-model inference was used in order to choose the best model for each speaker based on their speech. Some trading relations between the correlates were discovered across the speakers, however, the precise quantitative weightings between them were difficult to assess for a number of reasons. The relevance of the current study for Ojibwe dialectology is discussed. ii Acknowledgements This thesis was written with the enthusiastic help of influence of many individuals. I would like to start off by thanking my supervisor Kevin Russell. He is responsible for shaping my views of phonology and phonetics and helping guide me through some profound and exciting intellectual revolutions. The other committee members deserve thanks as well. I have profited from many stimulating and inspiring discussions with Rob Hagiwara on phonology and phonetics. H.C. Wolfart continues to be a role model for me in terms of his unconditional love for knowledge. He also taught me how to write when I was an undergraduate. My conversations with Jim Hare have been few but have introduced me to the importance of interdisciplinary and broader ethological perspectives on my own field. Although David H. Pentland was not on my thesis committee, I wish to thank him for originally keeping my interest in the field as an undergraduate. He also introduced me to Algonquian linguistics generally. Much thanks also go out to Roger Roulette my original Saulteaux teacher here at the University of Manitoba for sharing his language with me with such enthusiasm and jest. I would also like to thank the people at ALM for letting me use recordings from their library for this project. Much thanks go out to the anthropologist Maureen Matthews for giving a broader context to my work and helping procure additional recordings. Bill Tredway has been such a good friend to me throughout the writing of this thesis it is difficult for me to express. His deep compassion and commitment to the Ojibwe language has been a huge source of inspiration. He is also a fountain of polymathematical knowledge that still amazes even after hours of conversations over tea. iii This thesis is dedicated to my linguistic consultant, Ojibwe teacher and friend Boo Bumble (Ron Hobson) from Long Plain reservation. I have truly enjoyed the company and conversation of such a biting and clever person who has taught me things that will always be between him and myself and that cannot be mentioned here for many reasons. Gichi-miigwech for your time and incredible patience with me. Thanks go out to my friends who have supported me throughout and got me through this process emotionally; Jesse Stewart for standing in solidarity with me in being perpetually confused about statistics; Nedja Lucena for being there for me even though far away; Ana Paula Brandão for being a close friend to me during the isolating period of thesis writing, listening to my complaints, and even reading drafts of my thesis from continents away; thanks go out to my Winnipeg friends Conrad Sweatman, Lasha Mowchun and Kelci Stephenson for much emotional support. This thesis would also not have been possible without the loving support of my parents, Ross and Barbara Tallman who have gone through so much on my behalf. Acoustic Correlates of Lenis and Fortis Stops in Manitoba Saulteaux by Adam J.R. Tallman Contents 1. Introduction 2. Phonological Grammar 2.1. Evolution 2.2. Morphophononological Correspondences 2.3. Discussion 3. Theoretical Framework: Probabilistic Phonology 4. Phonetic Encoding I: Methodology 4.1. Speakers and Data 4.2. Measurement 4.2.1. Consonant Duration 4.2.2. Postaspiration 4.2.3. Preaspiration 4.2.4. Voicing 4.3. Probability Models 4.3.1. Phonetic Contrast as a Logit Function 4.3.2. Separation 5. Phonetic Encoding II: Places of Articulation 5.1. Bilabials 5.1.1. Consonant Duration 5.1.2. Postaspiration 5.1.3. Preaspiration 5.1.4. Voicing 5.1.5. Loglinear Models I: Factors 5.1.6. Loglinear Models II: Multiple Regression 5.2. Dentals/Alveolars 5.2.1. Consonant Duration 5.2.2. Postaspiration 5.2.3. Preaspiration 5.2.4. Voicing 5.2.5. Loglinear Models I: Factors 5.2.6. Loglinear Models II: Multiple Regression 5.3. Velars 5.3.1. Consonant Duration 5.3.2. Postaspiration 5.3.3. Preaspiration 5.3.4. Voicing 5.3.5. Loglinear Models I: Factors 5.3.6. Loglinear Models II: Multiple Regression 6. Discussion and Summary Figures 1a. Density distributions for /d/-/t/ on the consonant duration parameter (Boo). 1b. Density distributions for /d/-/t/ on the postaspiration parameter (Boo). 2a. Density distributions for /d/ on the consonant duration and postaspiration parameter (Boo). 2b. Density distributions for /t/ on the consonant duration and postaspiraiton parameter (Boo). 3a. Hypothetical density distributions with means 40 and 80 and standard deviations 10 each. 3b. Hypothetical density distributions with means 40 an 80 and standard deviations 20 each. 4. Ojibwe Dialects. 5. Saulteaux speakers from Manitoba used in this study. 6. Spectrogram and waveform of apabiyin 'seat'. 7. Spectrogram and waveform of dabaziiwag 'they flee'. 8. Spectrogram and waveform of aagaskoonsag 'little prairie chickens'. 9. Spectrogram and waveform of bekaa 'wait'. 10. Spectrogram and waveform of odikoon 'branch'. 11. Spectrogram and waveform of akawe 'first'. 12. Spectrogram and waveform of obakitewaan 'He hits him'. 13. Spectrogram and waveform of odaabaan 'car'. 14. Spectrogram and waveform of waabooz 'rabbit'. 15. Spectrogram and waveform of anishinaabe 'Indian'. 16. Spectrogram and waveform of gi-gii-diba`igem 'You guys payed it.' 17. Spectrogram and waveform of dagwaagig 'it is autumn'. 18. Density distributions for /b/-/p/ on the consonant duration for MR. 19. Loglinear curve for Pr(/p/) on consonant duration for MR. 20a. Hypothetical example of quasi-separate data structure. 20b. Hypothetical example of completely-separate data structure. 21. Consonant duration of /b/-/p/ by speaker. 22. Postaspiration duration of /b/-/p/ by speaker. 23. Quantile-quantile plots for postaspiration in /p/ by speaker. 24. Quantile-quantile plots for preaspiration in /p/ by speaker. 25. Histograms for sonority values in /b/s by speaker. 26a. Categories for /b/ and /p/ on consonant duration for JB. 26b. Categories for /b/ and /p/ on consonant duration for WM. 27a. Loglinear curve of Pr(/p/) on postaspiration for JB. 27b. Loglinear curve of Pr(/p/) on postaspiration for SF. 28a. Histograms for /b/s according to sonority values for Boo (left) and JB (right). 28b. Loglinear curves of Pr(/p/) on sonority for Boo (dotted) and JB (solid). 29a. Loglinear curve for Pr(/p/) on sonority for JB's best model. 29b. Loglinear curve for Pr(/p/) on consonant duration for MR's best model. 30a. Loglinear curves for Pr(/p/) on consonant duration with sonority at 1 (solid) and 0 (dotted) for Boo's best model. 30b. Loglinear curves for Pr(/p/) on sonority with consonant duration at 59 (dotted) and 102 (solid) for Boo's best model. 31. Consonant duration of /d/-/t/ by speaker. 32. Postaspiration of /d/-/t/ by speaker. 33. Quantile-quantile plots for preaspiration in /t/ by speaker. 34. Duration of preaspiration in ms before long vs. short vowels by speaker. 35. Spectrogram and waveform of mitigoog 'trees'. 36. Histograms for the density of /d/s falling in each sonority value by speaker. 37. Loglinear curves for Pr(/t/) on consonant duration with postaspiration at 0, preaspiration at 0 and sonority at 0 for Boo (solid) and SF (dashed). 38. Loglinear curves for Pr(/t/) on postaspiration with consonant duration at 80 ms, preaspiration at 0 ms and sonority at 0. 39a. Loglinear curves for Pr(/t/) on consonant duration with postaspiration and preaspiration at 0 for WM (solid) and MR (dashed). 39b. Loglinear curves for Pr(/t/) on preaspiration with consonant duration at 90 ms and 0 postaspiration for WM (solid) and MR (dashed). 40. Quantile-quantile plots for consonant duration in /g/ by speaker. 41. Spectrogram and waveform of mashkode-bizhikiwan 'black cow' produced by SF. 42. Spectrogram and waveform of ako EMPH produced by MR. 43. Consonant duration of /g/-/k/ by speaker. 44. Postaspiration of /g/-/k/ by speaker. 45. Quantile-quantile plots for postaspiration in /g/ by speaker. 46. Quantile-quantile plots for preaspiration in /k/ by speaker. 47. Preaspiration before short and long vowels for /k/ by speaker. 48. Histograms for sonority values in /g/s by speaker. 49a. Density distributions for /g/ and /k/ on consonant duration for JB. 49b. Density distributions for /g/ and /k/ on consonant duration for NT. 50. Density distributions for /k/ on postaspiration for MR (solid), SF (dotted) and JB (dashed).