Citation
Understanding sonority

Material Information

Title:
Understanding sonority an acoustic analysis of perceptual cues in English and Russian consonant clusters
Creator:
Bray, Jodi Patrice
Place of Publication:
[Gainesville, Fla.]
Publisher:
University of Florida
Publication Date:
Language:
English

Subjects

Subjects / Keywords:
Auditory perception ( jstor )
Consonants ( jstor )
Fricative consonants ( jstor )
Listening ( jstor )
P values ( jstor )
Perception ( jstor )
Rhyme ( jstor )
Stop consonants ( jstor )
Syllables ( jstor )
Vowels ( jstor )
Dissertations, Academic -- Linguistics -- UF ( lcsh )
Linguistics thesis, Ph. D ( lcsh )
Genre:
government publication (state, provincial, terriorial, dependent) ( marcgt )
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Summary:
ABSTRACT: This research was undertaken as a step in determining if there are acoustic correlates to sonority and if perceptual cues should play a role in phonological theory, specifically in Optimality Theory. There were four main steps to the research. First, tokens of real and nonsense words with complex word-final clusters were collected from native speakers of English and Russian. Second, data was acoustically analyzed to determine how duration and rms amplitude vary in different consonants found in parallel environments. Third, the tokens were played for native speakers of English and Russian to determine which consonants were difficult for the listeners to perceive. Finally, the findings from the acoustic and perception experiments were used to demonstrate that the difficulties in perception could be predicted through salience constraints that capture how greater amplitude and length allow a segment to be better heard. It was found in this research that certain consonants are consistently louder and longer than other consonants in parallel environments. Specifically, strident fricatives and English ["upside-down r"] tended to have greater relative rms and duration than the other consonants in parallel positions. Consequently, they were also more perceptible in these positions for both English and Russian listeners. Using the findings from the acoustic and perception experiment, this dissertation proposed the need for Optimality Theory constraints based on salience to explain the allowable consonant clusters in language. Salience constraints are shown to be able to predict the perceptibility of consonants in word-final clusters. Through these findings, it is suggested that perceptibility be considered to play a role in the phenomena often explained by sonority.
Summary:
KEYWORDS: Phonology, Sonority, Consonant Cluster, Acoustics, Salience, Phonetics, Optimality Theory, Perception
Thesis:
Thesis (Ph. D.)--University of Florida, 2001.
Bibliography:
Includes bibliographical references.
System Details:
System requirements: World Wide Web browser and PDF reader.
System Details:
Mode of access: World Wide Web.
General Note:
Title from title page of source document.
General Note:
Document formatted into pages; contains xxviii, 236 p.; also contains graphics.
General Note:
Includes vita.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright Bray, Jodi Patrice. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Embargo Date:
12/15/2012
Resource Identifier:
028625137 ( ALEPH )
78098761 ( OCLC )

Downloads

This item is only available as the following downloads:


Full Text

PAGE 1

UNDERSTANDING SONORITY: AN ACOUSTIC ANALYSIS OF PERCEPTUAL CUES IN ENGLISH AND RUSSIAN CONSONANT CLUSTERS By JODI PATRICE BRAY A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2001

PAGE 2

Copyright 2001 by Jodi Patrice Bray

PAGE 3

For Kit and Jack and Marie and Clifford

PAGE 4

ACKNOWLEDGMENTS I owe great thanks to all of my committee members: Dr. Caroline Wiltshire (chair), Dr. Ratree Wayland, Dr. D. Gary Miller, Dr. William J. Sullivan, and Dr. Alice Dyson. Their comments, assistance in analyzing data, occasional home-cooked meals, and encouragement to attend conferences were all an integral part in completing this work. Thanks also to Dr. Jean Casagrande who assisted with my defense and gave me thoughtful comments on short notice. This research was partially funded by a generous gift to the College of Liberal Arts and Sciences by the Threadgill family. This dissertation could not have been completed without the Department of Communication Sciences and Disorders, especially Dr. Christine Sapienza and Dr. Harry Rothman who allowed me to use their equipment. I would like to recognize Dr. Stefan Frisch for giving me advice that helped me to begin this research. I am grateful to Kevin Bray, my dissertation coach, for his encouragement, suggestions, and formatting tips. Lisa Banks and the Library West Staff were also a significant help in doing my research. I would also like to thank Anne Taylor from the Graduate School Editorial Office for her suggestions about formatting this text. Of course, I am very thankful to those who participated in this study and those who helped find participants. My fellow phonology and phonetics students, Mohamed Al-Khairy, Rebecca Hill, Kristin Liljegren, and Russell Moon were an invaluable source of advice. Mark Skowronski’s help was vital in producing my ‘noisy sounds.’ Thanks also to my students iv

PAGE 5

for reminding me why I love linguistics. And thanks to my Program in Linguistics friends, especially Tamara Sniad, Irene Moyna, Jodi Nelms and Toru Matusuzaki, for making linguistics fun. Finally, I am extremely grateful to my mom and dad, my sister (Christine) and my brothers (Matt and Kevin), Nancy Garbolino, Bob Tatro, Joe, Joey, and Shannon Wittreich, Heather Szaszy, Rick, Elaine, and Arielle Beschen, Nicole Lightfoot, Jim Koenig, Amanda Fitch, Scott Keshanech, Korinn Braden, and my Response Analysis friends. I thank them for their emotional and financial support, free meals, oatmeal cookies, Marshmallow Peeps, Disney World respites, and for keeping a straight face every time I said, “Just one more semester.” v

PAGE 6

TABLE OF CONTENTS page ACKNOWLEDGMENTS.................................................................................................iv LIST OF TABLES...........................................................................................................xiii LIST OF FIGURES.........................................................................................................xxv ABSTRACT..................................................................................................................xxvii CHAPTERS 1 INTRODUCTION...........................................................................................................1 2 LITERATURE REVIEW................................................................................................6 2.1. Introduction..............................................................................................................6 2.2. Sonority in Phonology.............................................................................................6 2.2.1. Sonority Sequencing...................................................................................6 2.2.2. Uses for Sonority........................................................................................9 2.2.2.1. Syllable Contact Law..........................................................................9 2.2.2.2. Assimilation........................................................................................9 2.2.2.3. Metathesis.........................................................................................10 2.2.2.4. Berber................................................................................................11 2.2.2.5. Moraic theory....................................................................................12 2.2.2.6. Steriade.............................................................................................13 2.2.2.7. Stress.................................................................................................14 2.3. Phonetic Explanations of Consonant Clusters.......................................................15 2.4. Phonetic Explanations of English and Russian......................................................19 2.5. Phonetic Explanations of Russian Clusters...........................................................21 2.6. Consonant Clusters in Phonetic/Phonology Interface...........................................21 2.7. OT and Consonant Clusters...................................................................................22 3 ACOUSTICS PROCEDURAL OVERVIEW...............................................................24 3.1. Introduction............................................................................................................24 3.2. Hypotheses.............................................................................................................25 3.3. Acoustic Data Collection.......................................................................................26 3.3.1. Participants................................................................................................26 3.3.2. Stimuli.......................................................................................................27 3.3.3. Procedure..................................................................................................32 vi

PAGE 7

3.4. Acoustic Analysis..................................................................................................34 3.4.1. Acoustic Parameters to be Measured........................................................34 3.4.2. Labeling Duration.....................................................................................35 3.4.3. Words and Phrases....................................................................................36 3.4.4. Vowels, Consonants, Codas, and Rhymes................................................36 3.4.5. Vowel Labeling.........................................................................................37 3.4.6. Sonorant and [s] Labeling.........................................................................37 3.4.7. Labeling F3...............................................................................................38 3.4.8. Consonant Labeling..................................................................................39 3.4.9. Burst Labeling...........................................................................................40 3.4.10. Duration Measurements............................................................................40 3.4.11. Intensity.....................................................................................................41 3.4.12. Comparisons.............................................................................................41 4 PERCEPTION PROCEDURAL OVERVIEW.............................................................43 4.1. Introduction............................................................................................................43 4.2. Hypotheses.............................................................................................................43 4.3. Procedure...............................................................................................................44 4.3.1. Participants................................................................................................44 4.3.2. Stimuli.......................................................................................................44 4.4. Perception Data Collection....................................................................................46 4.5. Coding and Analysis..............................................................................................47 4.6. Relationships between Acoustics and Perception..................................................48 5 ENGLISH ACOUSTIC ANALYSIS OF DURATION.................................................49 5.1. Introduction............................................................................................................49 5.2. Speaking Rates.......................................................................................................51 5.3. Comparisons across Speaking Rates......................................................................51 5.3.1. Obstruents.................................................................................................52 5.3.1.1. Stops..................................................................................................52 5.3.1.2. Fricatives...........................................................................................54 5.3.2. Sonorants...................................................................................................56 5.3.2.1. Nasals................................................................................................56 5.3.2.2. Lateral approximant [l].....................................................................57 5.3.2.3. Rhotic approximant [].....................................................................59 vii

PAGE 8

5.4. Comparisons by Manner of Articulation...............................................................61 5.4.1. Stops vs. Fricatives...................................................................................61 5.4.2. Stops vs. Nasals........................................................................................63 5.4.3. Stops vs. Lateral Approximant [l].............................................................65 5.4.4. Stops vs. Rhotic Approximant []............................................................67 5.4.5. Fricatives vs. Nasals..................................................................................68 5.4.6. Fricatives vs. Lateral Approximant [l]......................................................70 5.4.7. Fricatives vs. Rhotic Approximant [].....................................................71 5.4.8. Nasals vs. Lateral Approximant [l]...........................................................71 5.4.9. Nasals vs. Rhotic Approximant []..........................................................73 5.4.10. Lateral Approximant [l] vs. Rhotic Approximant []..............................74 5.5. Ranking of Consonants in Environments..............................................................76 5.5.1. Preceding Voiceless Stops........................................................................77 5.5.2. Preceding Fricatives..................................................................................77 5.5.3. Following [l].............................................................................................78 5.5.4. Following []............................................................................................78 6 ENGLISH ACOUSTIC ANALYSIS OF RELATIVE RMS AMPLITUDE................79 6.1. Introduction............................................................................................................79 6.2. Speaking Rate........................................................................................................80 6.3. Comparisons by Manner of Articulation...............................................................81 6.3.1. Stops vs. Fricatives...................................................................................81 6.3.2. Stops vs. Nasals........................................................................................83 6.3.3. Stops vs. Lateral Approximant [l].............................................................85 6.3.4. Stops vs. Rhotic Approximant []............................................................86 6.3.5. Fricatives vs. Nasals..................................................................................87 6.3.6. Fricatives vs. Lateral Approximant [l]......................................................90 6.3.7. Fricatives vs. Rhotic Approximant [].....................................................91 6.3.8. Nasals vs. Lateral Approximant [l]...........................................................93 6.3.9. Nasals vs. Rhotic Approximant []..........................................................94 6.3.10. Lateral Approximant [l] vs. Rhotic Approximant []..............................95 6.4. Amplitude Compared to Adjacent Consonant.......................................................96 6.4.1. Word-final Consonants Compared to Adjacent []..................................97 6.4.2. Word-final Consonants Compared to Adjacent [l]...................................97 6.4.3. Word-final Consonants Compared to Adjacent Nasals............................98 6.4.4. Word-final Consonants Compared to Adjacent [s]...................................98 6.4.5. Word-final Consonants Compared to Adjacent [p]..................................99 6.5. Ranking of Consonants in Parallel Environments.................................................99 6.5.1. Preceding Voiceless Stops........................................................................99 6.5.2. Preceding Fricatives................................................................................100 6.5.3. Following Sonorants...............................................................................100 viii

PAGE 9

7 RUSSIAN DURATION ANALYSIS..........................................................................102 7.1. Introduction..........................................................................................................102 7.2. Speaking Rates.....................................................................................................103 7.3. Comparisons across Speaking Rates....................................................................104 7.3.1. Obstruents...............................................................................................104 7.3.1.1. Stops................................................................................................104 7.3.1.2. Fricatives.........................................................................................105 7.3.2. Sonorants.................................................................................................107 7.3.2.1. Nasals..............................................................................................107 7.3.2.2. Lateral approximant [l]...................................................................108 7.3.2.3. Alveolar trill [r]...............................................................................109 7.4 Comparisons by Manner of Articulation..............................................................112 7.4.1. Stops vs. Fricatives.................................................................................112 7.4.2. Stops vs. Nasals......................................................................................113 7.4.3. Stops vs. Lateral Approximant [l]...........................................................114 7.4.4. Stops vs. Alveolar Trill [r]......................................................................115 7.4.5. Fricative vs. Nasals.................................................................................116 7.4.6. Fricatives vs. Lateral Approximant [l]....................................................118 7.4.7. Fricatives vs. Alveolar Trill [r]...............................................................118 7.4.8. Nasals vs. Lateral Approximant [l].........................................................119 7.4.9. Nasals vs. Alveolar Trill [r]....................................................................119 7.4.10. Lateral Approximant [l] vs. Alveolar Trill [r]........................................120 7.5. Ranking of Consonants in Environments............................................................120 7.5.1. Preceding Voiceless Stops......................................................................121 7.5.2. Following [r]...........................................................................................121 ix

PAGE 10

8 RUSSIAN ACOUSTIC ANALYSIS OF RELATIVE RMS AMPLITUDE..............122 8.1. Introduction..........................................................................................................122 8.2. Speaking Rate......................................................................................................122 8.3. Comparisons by Manner of Articulation.............................................................123 8.3.1. Stops vs. Fricatives.................................................................................124 8.3.2. Stops vs. Nasals......................................................................................125 8.3.3. Stops vs. Lateral Approximant [l]...........................................................126 8.3.4. Stops vs. Alveolar Trill [r]......................................................................127 8.3.5. Fricative vs. Nasals.................................................................................127 8.3.6. Fricatives vs. Lateral Approximant [l]....................................................129 8.3.7. Fricatives vs. Alveolar Trill [r]...............................................................129 8.3.8. Nasals vs. Lateral Approximant [l].........................................................130 8.3.9. Nasals vs. Alveolar Trill [r]....................................................................130 8.3.10. Lateral Approximant [l] vs. Alveolar Trill [r]........................................131 8.4. Amplitude Compared to Adjacent Consonant.....................................................131 8.4.1. Word-final Consonants Compared to Adjacent [r].................................132 8.4.2. Word-final Consonants Compared to Adjacent [l].................................133 8.4.3. Word-final Consonants Compared to Adjacent Nasals..........................134 8.4.4. Word-final Consonants Compared to Adjacent [s].................................134 8.4.5. Word-final Consonants Compared to Adjacent Voiceless Stops...........134 8.5. Ranking of Consonants in Environments............................................................135 8.5.1. Preceding Voiceless Stops......................................................................135 8.5.2. Following [r]...........................................................................................136 9 ENGLISH PERCEPTION...........................................................................................137 9.1. Introduction..........................................................................................................137 9.2. English Tokens....................................................................................................138 9.2.1. Perception of Obstruents.........................................................................138 9.2.1.1. Stops in casual speech.....................................................................139 9.2.1.2. Stops in fast speech.........................................................................142 9.2.1.3. Fricatives in casual speech..............................................................145 9.2.1.4. Fricatives in fast speech..................................................................146 9.2.2. Perception of Sonorants..........................................................................148 9.2.2.1. Nasals in casual Speech..................................................................148 9.2.2.2. Nasals in fast Speech......................................................................149 9.2.2.3. Lateral approximant [l] in casual speech........................................150 9.2.2.4. Lateral approximant [l] in fast speech............................................151 9.2.2.5. Rhotic approximant [] in casual and fast speech..........................152 9.2.3. Perception Dependent upon Manner of Articulation..............................153 9.2.3.1. Preceding Stops...............................................................................153 9.2.3.2. Preceding Fricatives........................................................................154 9.2.3.3. Following [l]...................................................................................154 9.2.3.4. Following []..................................................................................155 x

PAGE 11

9.3. Russian Tokens....................................................................................................156 9.3.1. Perception of Obstruents.........................................................................156 9.3.1.1. Stops in Casual Speech...................................................................156 9.3.1.2. Stops in Fast Speech.......................................................................159 9.3.1.3. Fricatives in Casual Speech............................................................161 9.3.1.4. Fricatives in Fast Speech................................................................162 9.3.2. Perception of Sonorants..........................................................................163 9.3.2.1. Nasals in Casual Speech.................................................................163 9.3.2.2. Nasals in Fast Speech......................................................................164 9.3.2.3. Lateral Approximant [l] in Casual and Fast Speech.......................166 9.3.2.4. Alveolar Trill [r] in Casual and Fast Speech..................................167 9.3.3. Perception Dependent upon Manner of Articulation..............................169 9.3.3.1. Preceding stops...............................................................................169 9.3.3.2. Following [r]...................................................................................171 9.4. Conclusions..........................................................................................................171 10 RUSSIAN PERCEPTION..........................................................................................173 10.1. Introduction........................................................................................................173 10.2. Russian Tokens..................................................................................................174 10.2.1. Perception of Obstruents.........................................................................174 10.2.1.1. Perception of stops in casual speech...............................................174 10.2.1.2. Stops in fast speech.........................................................................176 10.2.1.3. Fricatives in casual speech..............................................................179 10.2.1.4. Fricatives in fast speech..................................................................180 10.2.2. Perception of Sonorants..........................................................................181 10.2.2.1. Nasals in casual speech...................................................................181 10.2.2.2. Nasals in fast speech.......................................................................181 10.2.2.3. Lateral approximant [l] in casual and fast Speech..........................182 10.2.2.4. Alveolar trill in casual and fast speech...........................................183 10.2.3. Perception Dependent upon Manner of Articulation..............................185 10.2.3.1. Preceding stops...............................................................................185 10.2.3.2. Following [r]...................................................................................186 10.3. English Tokens..................................................................................................186 10.3.1. Perception of Obstruents.........................................................................187 10.3.1.1. Stops in casual speech.....................................................................187 10.3.1.2. Stops in fast speech.........................................................................189 10.3.1.3. Fricatives in casual and fast speech................................................191 10.3.2. Perception of Sonorants..........................................................................191 10.3.2.1. Nasals in casual speech...................................................................191 10.3.2.2. Nasals in fast speech.......................................................................192 10.3.2.3. Lateral approximant [l] in casual and fast speech...........................193 10.3.2.4. Rhotic approximant [] in casual and fast speech..........................193 xi

PAGE 12

10.3.3. Comparison of Perception by Manner of Articulation...........................196 10.3.3.1. Preceding stops...............................................................................196 10.3.3.2. Following []..................................................................................197 10.4. Conclusions........................................................................................................198 11 PHONOLOGICAL ANALYSIS................................................................................199 11.1. Introduction........................................................................................................199 11.2. English Phonological Analysis..........................................................................200 11.2.1. Stop Deletion..........................................................................................201 11.2.2. Perception of [] clusters........................................................................204 11.2.3. Nasals Preceding Stops...........................................................................206 11.2.4. Word-final Nasals, Fricatives, and [l].....................................................211 11.2.5. Conclusions.............................................................................................212 11.3. Russian Phonological Analysis..........................................................................212 11.4. Accounting for Other Phenomena.....................................................................215 11.5. Conclusions........................................................................................................215 12 CONCLUSIONS.........................................................................................................216 APPENDICES A ENGLISH QUESTIONNAIRE..................................................................................221 B RUSSIAN QUESTIONNAIRE...................................................................................224 C NOISY SOUND SCRIPT...........................................................................................227 D ENGLISH PERCEPTION EXAMS...........................................................................228 E RUSSIAN PERCEPTION EXAMS............................................................................230 LIST OF REFERENCES.................................................................................................232 BIOGRAPHICAL SKETCH...........................................................................................236 xii

PAGE 13

LIST OF TABLES Table Page 2-1: Clements’ Sonority Hierarchy........................................................................................7 2-2: Hume and Odden’s Impedance Hierarchy.....................................................................8 3-1: English real word list......................................................................................................29 3-2: Russian real word list.....................................................................................................30 3-3: English nonsense word list.............................................................................................31 3-4: Russian nonsense words.................................................................................................32 5-1: Mean average rate of all tokens......................................................................................51 5-2: Tokens with stops...........................................................................................................52 5-3: Duration of stops............................................................................................................53 5-4: Tokens with [s]...............................................................................................................54 5-5: Mean proportional duration of [s]..................................................................................54 5-6: Mean proportional duration of stridents.........................................................................55 5-7: Mean proportional duration of [s] following []............................................................55 5-8: Mean proportional duration of [f] following []............................................................55 5-9: Tokens with nasal consonants........................................................................................56 5-10: Mean proportional duration of nasals...........................................................................57 5-11: Tokens with [l].............................................................................................................57 5-12: Mean proportional duration of [l] at different speaking rates......................................58 5-13: Tokens with []............................................................................................................59 xiii

PAGE 14

5-14: Comparisons of [] at different speaking rates............................................................59 5-15: Mean proportional duration of [p] and [s] preceding [t]..............................................61 5-16: Mean proportional duration of [t] and [s] following [l]...............................................61 5-17: Mean proportional duration of stops ([t] and [k]) and strident fricatives following []........................................................................................................................62 5-18: Mean proportional duration of [t] and [s] following []..............................................62 5-19: Mean proportional duration of [p] and [f] following [].............................................62 5-20: Mean proportional duration of voiced stops and nasals following []........................63 5-21: Mean proportional duration of [t], [d] and [n] following [].......................................64 5-22: Mean proportional duration of [p], [b] and [m] following [].....................................64 5-23: Mean proportional duration of [p], [n] and [m] preceding [t]......................................65 5-24: Mean proportional duration of [t], [d], and [l] following [].......................................66 5-25: Mean proportional duration of [p] and [l] preceding [t]...............................................66 5-26: Mean proportional duration of [p], [], and F3 of [] preceding stop.........................67 5-27: Mean proportional duration of nasals and strident fricative following []..................68 5-28: Mean proportional duration of [s] and [n] following [].............................................68 5-29: Mean proportional duration of [f] and [m] following []............................................69 5-30: Mean proportional duration of nasals and [s] preceding voiceless stops.....................69 5-31: Mean proportional duration of [s] and [l] preceding [t]...............................................70 5-32: Mean proportional duration of [s] and [l] following []..............................................70 5-33: Mean proportional duration of [s] and [] preceding voiceless stop...........................71 5-34: Mean proportional duration of nasals and [l] preceding [t]..........................................72 5-35: [n] and [l] mean average comparisons following []...................................................72 5-36: [m] and [l] mean average comparisons following []..................................................72 xiv

PAGE 15

5-37: Nasal and [] mean average comparison preceding voiceless stop.............................73 5-38: Mean proportional duration of [l], [], and low F3 of [] preceding obstruent...........74 5-39: Mean proportional duration of [l], [], and low F3 of [] preceding [t]......................75 5-40: Mean proportional duration of [l], [], and low F3 of [] preceding [d].....................75 5-41: Mean proportional duration of [l], [], and low F3 of [] preceding strident fricatives...............................................................................................................76 5-42: Ranking of consonants by proportion of rhyme and coda preceding voiceless stops..77 5-43: Ranking of consonants by proportion of rhyme and coda with low F3 of [] preceding voiceless stops.....................................................................................77 5-44: Ranking of consonants by proportion of rhyme and coda following [l]......................78 5-45: Ranking of consonants by proportion of rhyme and coda following [].....................78 6-1: Mean relative rms of word-final voiceless stops and stop bursts...................................80 6-2: Mean relative rms of [p] and [s] preceding [t]...............................................................82 6-3: Mean relative rms of stops and fricative following sonorants.......................................82 6-4: Mean relative rms of stops and fricatives following []................................................82 6-5: Mean relative rms of stops and fricatives following [l].................................................83 6-6: Mean relative rms of stop burst and fricatives following sonorants..............................83 6-7: Mean comparisons of voiceless stops, voiced stops and nasals following []..............84 6-8: Mean relative rms of [d] and [n] following []..............................................................84 6-9: Mean relative rms of stop bursts and nasals following []............................................84 6-10: Mean relative rms of [t], [d], stop bursts and [l] following []....................................85 6-11: Mean relative rms of [p] and [l] preceding [t]..............................................................86 6-12: Mean relative rms of [p] and [] preceding [t].............................................................86 6-13: Mean relative [p] and []’s lowest F3 preceding stops................................................87 xv

PAGE 16

6-14: Mean relative rms of labial and alveolar fricatives and nasals following []..............87 6-15: NES5 relative rms of word-final nasals and fricatives.................................................88 6-16: Mean relative rms of [s] and [n] following []............................................................88 6-17: Mean relative rms of [f] and [m] following []...........................................................88 6-18: Mean relative rms of nasals and strident fricatives following []................................89 6-19: Mean relative rms of strident fricatives and nasals following [] excluding NES5....89 6-20: Mean relative rms of [s] and nasals preceding voiceless stops....................................89 6-21: [s] and [l] mean average comparisons following []...................................................90 6-22: Mean relative rms [f] and [l] following []..................................................................91 6-23: Mean relative rms of [s] and [l] preceding [t]..............................................................91 6-24: Mean relative rms of [s] and [] preceding voiceless stops.........................................92 6-25: Mean relative rms of [s] and []’s low F3 preceding voiceless stops..........................92 6-26: Mean relative rms of [n] and [l] preceding voiceless stop...........................................93 6-27: Mean relative rms of [m] and [l] preceding voiceless stops.........................................93 6-28: Mean relative rms of [n] and [l] following [].............................................................94 6-29: Mean relative rms of [m] and [l] following []............................................................94 6-30: Mean relative rms of nasals and [] preceding voiceless stop.....................................95 6-31: Mean relative rms of nasals and []’s lowest F3 preceding voiceless stop.................95 6-32: Mean relative rms of [] and [l] preceding obstruents.................................................96 6-33: Mean rms of adjacent stop and [l]................................................................................97 6-34: Mean relative rms of adjacent [l] and strident fricatives..............................................98 6-35: Mean average relative rms of adjacent stops and nasals..............................................98 6-36: Mean average relative rms of adjacent stops and [s]....................................................98 xvi

PAGE 17

6-37: Mean average relative rms of adjacent [t] and [p]........................................................99 6-38: Ranking of rms of consonants preceding stops............................................................99 6-39: Ranking of consonants based on duration of rhyme preceding stops..........................100 6-40: Ranking of rms of consonants following []...............................................................101 6-41: Ranking of duration of consonants following []........................................................101 7-1: Mean average rate of all tokens......................................................................................104 7-2: Tokens with stops...........................................................................................................104 7-3: Duration of stops following sonorants...........................................................................105 7-4: Mean proportional duration of [p] and [k] preceding [t]................................................105 7-5: Tokens with fricatives....................................................................................................106 7-6: Mean proportional duration of [s] preceding stops........................................................106 7-7: Mean proportional duration of strident fricatives following [r].....................................106 7-8: Mean proportional duration of [f] following [r].............................................................106 7-9: Tokens with nasal consonants........................................................................................107 7-10: Mean proportional duration of nasals preceding voiceless stops.................................108 7-11: Tokens with [l].............................................................................................................109 7-12: Mean proportional duration of [l] at different speaking rates......................................109 7-13: Tokens with [r].............................................................................................................109 7-14: Comparisons of [r] preceding voiceless stops at different speaking rates....................110 7-15: Comparisons of [r] preceding voiced stops at different speaking rates........................110 7-16: Comparisons of [r] preceding fricatives at different speaking rates.............................111 7-17: Comparisons of [r] preceding nasals at different speaking rates..................................111 7-18: Mean proportional duration of [p], [k], and [s] preceding [t].......................................112 7-19: Mean proportional duration of voiceless stops and strident fricatives following [r]...113 7-20: Mean proportional duration of [p] and [f] following [r]...............................................113 xvii

PAGE 18

7-21: Mean proportional duration of voiced stops and nasals following [r]..........................113 7-22: Mean proportional duration of [p], [m], and [n] preceding voiceless stops.................114 7-23: Mean proportional duration of [p] and [l] preceding [t]...............................................115 7-24: Mean proportional duration of [p], [k], and [r] preceding stops..................................115 7-25: Mean proportional duration of alveolar and labial nasals [m, n] and fricatives [s, f] following [r].........................................................................................................116 7-26: Mean proportional duration of nasals and strident fricative following [r]...................116 7-27: Mean proportional duration of [s] and [n] following [r]..............................................117 7-28: Mean proportional duration of [f] and [m] following [r].............................................117 7-29: Mean proportional duration of nasals and [s] preceding voiceless stops.....................117 7-30: Mean proportional duration of [s] and [l] preceding [t]...............................................118 7-31: Mean proportional duration of [s] and [r] preceding voiceless stop............................119 7-32: Mean proportional duration of [n] and [l] preceding [t]...............................................119 7-33: Nasal and [r] mean average comparison preceding voiceless stop..............................120 7-34: Mean proportional duration of [l] and [r] preceding [t]...............................................120 7-35: Ranking of consonants by proportion of rhyme and coda duration preceding voiceless stops......................................................................................................121 7-36: Ranking of consonants by proportion of rhyme and coda duration preceding following [r].........................................................................................................121 8-1: Mean relative amplitude of [s] preceding word-final stop.............................................123 8-2: Mean relative rms of word-final voiceless stops............................................................123 8-3: Mean relative rms of word-final voiceless stops bursts.................................................123 8-4: Mean relative rms of [p, k] and [s] preceding [t]...........................................................124 8-5: Mean relative rms of stops [p, t] and fricatives [f, s] following [r]................................125 8-6: Mean relative rms of stops [t, k] and strident fricatives [s, ] following [r]..................125 8-7: Mean relative rms of [p] and non-strident [f] following [r]...........................................125 xviii

PAGE 19

8-8: Mean comparisons of stops and nasals following [r].....................................................126 8-9: Mean relative rms of stop bursts and nasals following [r].............................................126 8-10: Mean relative rms of [p] and [l] preceding [t]..............................................................126 8-11: Mean relative rms of stops and [r] preceding [t]..........................................................127 8-12: Mean relative rms of labial and alveolar nasals and fricatives following [r]...............127 8-13: Mean relative rms of [s] and [n] following [r].............................................................128 8-14: Mean relative rms of [f] and [m] following [r].............................................................128 8-15: Mean relative rms of nasals [m, n] and strident fricatives [s, ] following [r]............128 8-16: Mean relative rms of nasals and [s] preceding voiceless stops....................................129 8-17: Mean relative rms of [s] and [l] preceding [t]..............................................................129 8-18: Mean relative rms of [s] and [r] preceding voiceless stop............................................130 8-19: Mean relative rms of [n], [m], and [l] preceding voiceless stop...................................130 8-20: Mean relative rms of nasals and [r] preceding voiceless stop......................................131 8-21: Mean relative rms of [r] and [l] before [t]....................................................................131 8-22: Mean rms of adjacent voiceless stop and [r]................................................................132 8-23: Mean rms of adjacent fricative and [r].........................................................................132 8-24: Mean rms of adjacent strident fricative and [r]............................................................132 8-25: Mean rms of adjacent non-strident fricative and [r].....................................................132 8-26: Mean rms of adjacent [m] and [r].................................................................................133 8-27: Mean rms of adjacent stop and [l]................................................................................133 8-28: Mean rms of adjacent [m] and [l].................................................................................134 8-29: Mean average relative rms of adjacent stops and nasals..............................................134 8-30: Mean average relative rms of adjacent stops and [s]....................................................134 8-31: Mean average relative rms of adjacent [t] and [p]........................................................135 8-32: Mean average relative rms of adjacent [t] and [k]........................................................135 xix

PAGE 20

8-33: Ranking of rms of consonants preceding stops............................................................135 8-34:Ranking of English rms of consonants preceding stops................................................135 8-35: Ranking of rms of consonants following [r].................................................................136 9-1: Perception of word-final voiceless stops in casual speech.............................................140 9-2: Ranking of perception of word-final stops by environment in casual speech................140 9-3: Perception of [p] preceding [t] in casual speech............................................................141 9-4: Ranking of perception of word-final stops by environment in casual speech................142 9-5: Perception of voiced stops in casual speech...................................................................142 9-6: Perception of word-final voiceless stops in fast speech.................................................143 9-7: Ranking of perception of word-final stops by environment in fast speech....................144 9-8: Perception of [p] preceding [t] in casual speech in fast speech......................................144 9-9: Ranking of perception of word-final stops by environment at +6dB in fast speech......144 9-10: Perception of word-final voiced stops in fast speech...................................................145 9-11: Ranking of perception of word-final voiced stops by environment in fast speech......145 9-12: Perception of non-strident fricatives in casual speech.................................................146 9-13: Perception of [s] preceding stops in casual speech......................................................146 9-14: Perception of strident fricatives in fast speech.............................................................147 9-15: Perception of non-strident fricatives in fast speech......................................................147 9-16: Perception of [s] preceding stops in fast speech...........................................................148 9-17: Perception of nasals preceding stops in casual speech.................................................148 9-18: Perception of word-final nasals in fast speech.............................................................149 9-19: Perception of nasals preceding stops in fast speech.....................................................149 9-20: Perception of word–final [l] in casual speech..............................................................150 9-21: Perception of [l] preceding obstruents in casual speech...............................................151 9-22: Perception of [l] preceding obstruents in fast speech...................................................152 xx

PAGE 21

9-23: Tokens with []............................................................................................................152 9-24: Ranking of perceptibility of consonants preceding voiceless stops in casual speech..154 9-25: Ranking of perceptibility of consonants preceding voiceless stops in fast speech......154 9-26: Ranking of perceptibility of consonants following [] in casual speech.....................155 9-27: Ranking of perceptibility of consonants following [] in fast speech.........................155 9-28: Perception of word-final voiceless stops in casual speech...........................................157 9-29: Ranking of perception of word-final stops by environment in casual speech..............158 9-30: Perception of [p] and [k] preceding [t] in casual speech..............................................158 9-31: Perception of voiced stops............................................................................................158 9-32: Perception of word-final voiceless stops in fast speech...............................................159 9-33: Ranking of perception of word-final stops by environment in fast speech..................160 9-34: Perception of [p] preceding [t] in casual speech in fast speech....................................160 9-35: Perception of word-final voiced stops in fast speech...................................................161 9-36: Perception of strident fricatives in casual speech.........................................................161 9-37: Perception of non-strident fricatives in casual speech.................................................162 9-38: Perception of [s] preceding stops in casual speech......................................................162 9-39: Perception of strident fricatives in fast speech.............................................................162 9-40: Perception of non-strident fricatives in fast speech......................................................163 9-41: Perception of [s] preceding stops in fast speech...........................................................163 9-42: Perception of word-final nasals in casual speech.........................................................164 9-43: Perception of word-final nasals in fast speech.............................................................165 9-44: Perception of nasals preceding stops in fast speech.....................................................165 9-45: Perception of [l] preceding consonants in casual speech.............................................166 9-46: Perception of [l] preceding consonants in fast speech..................................................167 9-47: Perception of [r] in casual speech.................................................................................168 xxi

PAGE 22

9-48 :Ranking of perception of word-final stops by environment in casual speech..............168 9-49: Perception of [r] in fast speech.....................................................................................169 9-50: Ranking of perceptibility of consonants preceding voiceless stops in casual speech..170 9-51: Ranking of perceptibility of consonants preceding voiceless stops in fast speech......170 9-52: Ranking of perceptibility of consonants following [r] in casual speech......................171 9-53: Ranking of perceptibility of consonants following [r] in casual speech......................171 10-1: Perception of word-final voiceless stops in casual speech...........................................175 10-2: Ranking of perception of word-final stops by environment in casual speech..............175 10-3: Perception of [p] and [k] preceding [t].........................................................................176 10-4: Ranking of perception of word-final voiceless stops by environment in casual speech...................................................................................................................176 10-5: Perception of word-final voiceless stops in fast speech...............................................177 10-6: Ranking of perception of word-final stops by environment in fast speech..................177 10-7: Perception of [p] preceding [t] in casual speech in fast speech....................................178 10-8: Perception of word-final voiced stops in fast speech...................................................178 10-9: Perception of strident fricatives in casual speech.........................................................179 10-10: Perception of non-strident fricatives in casual speech...............................................179 10-11: Perception of strident fricatives in fast speech...........................................................180 10-12: Perception of non-strident fricatives in fast speech....................................................180 10-13: Perception of word-final nasals in casual speech.......................................................181 10-14: Perception of word-final nasals in fast speech...........................................................182 10-15: Perception of [l] preceding obstruents in casual speech.............................................183 10-16: Perception of [l] preceding obstruents in fast speech.................................................183 10-17: Perception of [r] in fast speech...................................................................................184 10-18: Ranking of perceptibility of consonants preceding voiceless stops in casual speech185 10-19: Ranking of perceptibility of consonants preceding voiceless stops in fast speech....186 xxii

PAGE 23

10-20: Perception of word-final voiceless stops in casual speech.........................................187 10-21:Ranking of environments by perception of word-final stops at in casual speech.......188 10-22: Perception of [p] preceding [t] in casual speech........................................................188 10-23: Perception of voiced stops in casual speech...............................................................188 10-24: Perception of word-final voiceless stops in fast speech.............................................189 10-25: Ranking of environments by perception of word-final stops by environment in fast speech...................................................................................................................190 10-26: Perception of word-final voiced stops in fast speech.................................................190 10-27: Perception of word-final nasals in casual speech.......................................................192 10-28: Perception of nasals preceding stops in casual speech...............................................192 10-29: Perception of word-final nasals in fast speech...........................................................192 10-30: Perception of nasals preceding stops in fast speech...................................................193 10-31: Perception of [] in casual speech..............................................................................194 10-32: Perception of [] in fast speech..................................................................................195 10-33: Ranking of perceptibility of consonants preceding voiceless stops in casual speech196 10-34: Ranking of perceptibility of consonants preceding voiceless stops in fast speech....197 10-35: Ranking of perceptibility of consonants following [] in casual speech...................197 10-36: Ranking of perceptibility of consonants following [] at in fast speech...................198 11-1: Account for stop deletion following [s]........................................................................203 11-2: Account for stop deletion following [p].......................................................................203 11-3: Account for stop deletion in two stop clusters.............................................................204 11-4: Account for stop deletion following [].......................................................................204 11-5: Duration of segments in [] + stop rhymes..................................................................205 11-6: Account for stop deletion following [].......................................................................206 11-7: Duration of nasals and [] preceding stops..................................................................207 xxiii

PAGE 24

11-8: Salience features between [a] and a stop......................................................................209 11-9: Account for nasal deletion preceding voiceless stops..................................................210 11-10: Account for preservation of [] and [s] preceding voiceless stops............................211 11-11: Account for preservation of [] clusters..................................................................211 11-12: Salience features between [a] and a word-final stop..................................................213 11-13: Account for Russian nasal + stop clusters..................................................................213 11-14: Account for Russian [s, r, l] + stop clusters...............................................................214 A-1: Multiple choice for English perception of English tokens............................................228 A-2: Multiple choice for English perception of Russian tokens............................................228 D-1: Multiple choices for Russian perception of Russian tokens..........................................230 D-1: Multiple choices for Russian perception of English tokens..........................................230 xxiv

PAGE 25

LIST OF FIGURES Figure Page 2-1: Sonority Hierarchy.........................................................................................................7 2-2: West Germanic languages, diachronic changes.............................................................9 2-3: Syllable Contact Law.....................................................................................................9 2-4: Pali progressive assimilation..........................................................................................10 2-5: Pali regressive assimilation............................................................................................10 2-6: Pali regressive assimilation of equal segments..............................................................10 2-7: Old Spanish metathesis...................................................................................................11 2-8: Berber sonority hierarchy...............................................................................................11 2-9: Berber syllabification.....................................................................................................11 2-10: Constraint on moraic prominence................................................................................12 2-11: Parsing of word ‘simply’..............................................................................................13 2-12: Incorrect parsing of ‘simply’........................................................................................13 2-13: Incorrect parsing of ‘simply’........................................................................................13 2-14: Parsing of /supta/..........................................................................................................13 2-15: Greek word internal [s].................................................................................................14 2-16: Greek word initial [s]....................................................................................................14 3-1: Antiformant marking of word ‘romp’............................................................................38 3-2: Marking of ‘harsh’.........................................................................................................39 4-1: Multiple-choice example................................................................................................47 11-1: Already existing constraints.........................................................................................202 xxv

PAGE 26

11-2: Sonority-relative Uniformity-V constraints:................................................................205 11-3: Salience Constraints.....................................................................................................209 11-4: MAX-IO word-final consonant.....................................................................................210 xxvi

PAGE 27

Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy UNDERSTANDING SONORITY: AN ACOUSTIC ANALYSIS OF PERCEPTUAL CUES IN ENGLISH AND RUSSIAN CONSONANT CLUSTERS By Jodi Patrice Bray December 2001 Chairman: Caroline R. Wiltshire Major Department: Program in Linguistics This research was undertaken as a step in determining if there are acoustic correlates to sonority and if perceptual cues should play a role in phonological theory, specifically in Optimality Theory. There were four main steps to the research. First, tokens of real and nonsense words with complex word-final clusters were collected from native speakers of English and Russian. Second, data was acoustically analyzed to determine how duration and rms amplitude vary in different consonants found in parallel environments. Third, the tokens were played for native speakers of English and Russian to determine which consonants were difficult for the listeners to perceive. Finally, the findings from the acoustic and perception experiments were used to demonstrate that the difficulties in perception could be predicted through salience constraints that capture how greater amplitude and length allow a segment to be better heard. xxvii

PAGE 28

It was found in this research that certain consonants are consistently louder and longer than other consonants in parallel environments. Specifically, strident fricatives and English [] tended to have greater relative rms and duration than the other consonants in parallel positions. Consequently, they were also more perceptible in these positions for both English and Russian listeners. Using the findings from the acoustic and perception experiment, this dissertation proposed the need for Optimality Theory constraints based on salience to explain the allowable consonant clusters in language. Salience constraints are shown to be able to predict the perceptibility of consonants in word-final clusters. Through these findings, it is suggested that perceptibility be considered to play a role in the phenomena often explained by sonority. . xxviii

PAGE 29

CHAPTER 1 INTRODUCTION Allowable consonant clusters in the world’s languages have long been accounted for in phonology by sonority and sonority sequencing. Syllables are said to increase in sonority, reaching a peak at the nucleus (usually a vowel), and then decline in sonority if the syllable has a coda. Despite the prominent role that sonority has played in phonological theory, it remains without a consistent definition. Sonority has been linked to aperture, loudness, and perceptibility. This dissertation was undertaken as a step in better understanding the acoustic correlates to sonority and understanding whether those correlates contribute to a segment’s perceptibility. This study considered two languages, English and Russian, that are known for allowing a variety of consonant clusters. Variations in duration and amplitude of consonants in word-final clusters were measured at two different speaking rates. After measurements were completed, tokens were played for native speakers of English and Russian to consider how the perception of consonants relates to their acceptability in clusters. Phonetically this dissertation is built on two suggestions about sonority made by phoneticians. First, Ohala (1990) claims that what linguists refer to as sonority is really modulation. That is, a segment’s ability to occur in a certain environment is related to its ability to modify itself in that environment and remain perceptible. Second, Ladefoged relates sonority to a segment’s “loudness relative to that of other sounds with the same length, stress and pitch” (1993, pg. 245). Related to these two ideas, this dissertation 1

PAGE 30

2 considers modifications made to a consonant’s loudness and length in a variety of word-final clusters and how these relate to the consonant’s perceptibility. For instance, it is shown in this dissertation that some consonants are louder and longer in certain word positions, and that the louder or longer consonants are more perceptible. Findings from this study were then considered in a phonological framework, specifically Optimality Theory (Prince and Smolensky 1993). Flemming (1995) suggests that in order to account for phonological phenomena, it is necessary to consider both acoustic and articulatory constraints. An OT account would allow for multiple articulatory constraints and acoustic cues to act in competition and coordination to ensure both ease of articulation and ease of perception. The focus of this research is to determine if the cross-linguistic the role that constraints on acoustics and perceptibility may play in acceptable consonant clusters. There were three primary hypotheses for this study. First, consonant higher on the sonority hierarchy would have greater proportional duration the sounds lower on the sonority hierarchy in parallel environments. Second, sounds higher on the sonority hierarchy would have greater relative root mean square (rms) amplitude than sounds lower on the sonority hierarchy in parallel environments. Third, sounds higher on the sonority would be better perceived than sounds lower on the hierarchy in parallel environments. After considering these hypotheses in English and Russian, the findings were considered in OT. Chapter 2 of this dissertation gives an overview of how sonority has been used in phonological theory. Chapter 3 outlines the data collection procedure for the acoustic data analysis. Tokens of both real and nonsense words were collected from native

PAGE 31

3 speakers of English and Russian. Then measurements were made of the duration and rms of each consonant in the word-final consonant clusters of the tokens. Chapter 4 describes the collection of the perception data. Native speakers of English and Russian listened to the nonsense word tokens. The words were presented in three levels of noise. The listeners were given a multiple-choice exam where they were asked to choose the spelling that best represented the word that they heard. The findings from the acoustic experiment are presented in chapters 5 through 8. Chapter 5 considers the acoustic analysis of the duration of consonants in parallel environments. It shows that strident fricatives and [] have longer duration than other consonants in the same environments, while stops have shorter duration than other consonants in the same environments. Chapter 6 shows the relative rms of English consonants in parallel environments. In this chapter it is demonstrated that strident fricatives and [] have greater relative rms than other consonants in the same environments and that stops have lower relative rms. Chapter 7 considers the acoustic analysis of the duration of consonants in the Russian tokens. Here it is shown that the Russian data was slightly different from the English data. In this case, the Russian trill [r] did not have longer duration than the other consonants in parallel environments while the strident fricatives did. The Russian stops, like the English stops, were found to be shorter than other consonants n parallel environments. Chapter 8 shows the acoustic analysis of the relative rms of the consonants in parallel environments. As in English, the strident fricatives had greater relative rms than the other consonants in the same environments. However, the Russian trill [r] had lower relative rms than the English [] did. The stops

PAGE 32

4 as in English had a lower relative rms than the other consonants in the same environments. In chapters 9 and 10, the perception findings are detailed. Chapter 9 gives the findings from the English listeners. It is shown that the English listeners had difficulty hearing word-final stops. They also had more difficulty perceiving nasals before stops than they did perceiving [], [l], and [s] in the same position. Chapter 10 gives the results from the native Russian speakers’ perception experiments. They also had difficulty perceiving word-final stops. They had more difficulty perceiving [r] preceding word-final stops than they did perceiving nasals, [s], and [l] in the same word positions. Finally, chapter 11 analyzes the data in a phonological framework, specifically Optimality Theory to demonstrate the role that perception can play in a phonological analysis of consonant clusters. This chapter proposes salience constraints that can account for the inability to perceive segments in certain word positions. Salience is based upon a consonant’s relative rms and duration in a particular environment. Sonority, which has been given both acoustic and articulatory definitions, could be a case where such an analysis relying upon acoustics and perception is necessary. Using a phonological framework to understand phonetic data is at root difficult since phonetics and phonology are inherently different. In phonetic analysis, there are infinite variations on speech production, while phonology seeks to find general trends. The phonological portion of this dissertation was designed to take into account the trends found in the data in order to suggest that the growing use of perception in phonological theory is supported by listeners’ reactions to certain consonant clusters. This study investigates what ranking of acoustic correlates and perceptual cues would be necessary

PAGE 33

5 to predict the perceptibility of word-final clusters in Russian and English. Consideration was given to whether the same correlates play a role in both languages and if they are weighted differently. This dissertation proposes the necessity for the concept of salience in an accounting of sonority sequencing. Through a ranking of constraints based on salience, it is shown that a segment’s duration and amplitude along with its distinction from surrounding segments motivates its perceptibility in consonant clusters.

PAGE 34

CHAPTER 2 LITERATURE REVIEW 2.1. Introduction The allowable consonant clusters in the world’s languages are often explained by the sonority hierarchy and sonority sequencing. Syllables are said to increase in sonority until they reach a peak at the nucleus. Then if there is a coda, sonority decreases. Despite widespread consensus on sonority’s impact on syllable structure, much disagreement exists as to what sonority is. With phonology’s difficulties with the sonority sequencing principle came some backlash from the phonetics community questioning whether sonority even exists. This chapter outlines some past phonological analyses of consonant clusters that rely on the hierarchy, alternatives offered by phonetic research, how these analyses have been applied to English and Russian, and finally how phonetics and phonology can be interfaced to explain preferences for some consonant clusters. 2.2. Sonority in Phonology 2.2.1. Sonority Sequencing Cross-linguistically, languages prefer certain consonant clusters. For some time this was most often accounted for by the sonority hierarchy and sonority sequencing (e.g. Whitney 1865; Sievers 1893; Jespersen 1904; Saussure 1914; Grammont 1933; Hankamer and Aissen 1974; Steriade 1982; Selkirk 1984; Clements 1988, 1990). A ranking of sonority is shown in Figure 2-1 where obstruents are least sonorous and vowels and glides are most sonorous. 6

PAGE 35

7 Obstruents < Nasals < Liquids < Glides, Vowels Figure 2-1: Sonority Hierarchy There is clear evidence in phonology of a cross-linguistic preference for certain segments to occur in a certain order in syllables. Generally, sequences of segments increase in sonority towards the nucleus of the syllable where sonority reaches a peak. If there is a coda, then sonority decreases until the end of the syllable. Sonority has been associated with a wide variety of phonological phenomena. It has been linked to moraic weight (Zec 1994), contiguous segments across syllable boundaries (Murray and Vennemann 1983), the unique syllabification of Berber (Dell and Elmedlaoui 1985; Prince and Smolensky 1993), and stress placement (Kenstowicz 1997). Often, allowable consonant clusters are accounted for by the Sonority Sequencing Generalization (Selkirk 1984) and Minimal Sonority Distance (Steriade 1982). Although sonority is widely recognized by phonologists as a determiner of allowable clusters, there is great disagreement on what sonority is. Various phonological analyses have been offered to account for sonority ranking. Clements (1988, 1990) calculates sonority through major class features (Table 2-1). Table 2-1: Clements’ Sonority Hierarchy O < N < L < V + [vocoid] + + [approximant] + + + [sonorant] 0 1 2 3 Rank (relative sonority) Selkirk (1984) ranks sonority based on phonotactics and occurrence of segments in consonant clusters. The two analyses differ significantly: while Clements defined sonority using inherent features of segments, Selkirk identified the role a segment plays in syllable structure as the main defining factor. More recently, Hume and Odden (1996)

PAGE 36

8 have suggested that sonority should be replaced with 'impedance,' referring to the degree of obstruction in the vocal tract (Table 2-2). Table 2-2: Hume and Odden’s Impedance Hierarchy laryngeal V/glide liquid nasal fricative stop [-son] (8) [-cont] (4) C-place (2) V-place (1) Impedance 0 1 2 6 10 14 Despite the prominent role of sonority in phonology, sonority is plagued by four problems: Lack of a consistent definition Disagreement as to whether it should be defined as an acoustic or articulatory phenomena Exceptions to sonority sequencing The ranking of segments in the hierarchy and the details of the hierarchy. Trying to explain away these problems results in cyclical explanations of sonority. Sonority has been linked to loudness, aperture, and duration. Different analyses also vary as to how many levels exist in the hierarchy. In order to maintain the universal integrity of the hierarchy, Clements refuses to separate obstruents by voicing or manner. Conversely, Hankamer and Aissen (1974) and Steriade (1982) divide obstruents into stops and fricatives. Through evidence found in Berber syllabification, Dell and Elmedlaoui (1985) argue that obstruents should be divided still further into voiced and voiceless sounds. Zec (1994) limits the levels on the hierarchy, but allows for a variety of rankings through a set number of features that may or may not contribute to sonority in a given language.

PAGE 37

9 2.2.2. Uses for Sonority As stated above, phonologists use sonority to describe a variety of phenomena. This section outlines just a few to demonstrate how prevalent sonority and the sonority hierarchy are in phonological research. 2.2.2.1. Syllable Contact Law Murray and Vennemann (1983) rely on sonority to explain intersyllabic relationships. They found a preference for the initial consonant of an onset to be less sonorous than the final sound of the preceding coda. For instance, they showed that in the development of West Germanic languages, diachronic changes took place when a coda stop was followed by a less sonorous onset (Figure 2-2). Gothic [sat.jan] Old English [set.tan] Gothic [skap.jan.] Old English [sciep.pan] Figure 2-2: West Germanic languages, diachronic changes Through examples such as these, they developed the Syllable Contact Law (Figure 2-3). The preference for a syllable structure A$B, where A and B are marginal segments and a and b are the Consonantal Strength values of A and B respectively increases with the value of b minus a. Figure 2-3: Syllable Contact Law In the data above (Figure 2-2), the Syllable Contact Law requires a repair because the onset glide is more sonorous than the preceding coda. In this case, the glide fully assimilates to the coda. 2.2.2.2. Assimilation Similar to the accounts of Murray and Vennemann (1983), Hankamer and Aissen (1974), found that assimilation across syllable boundaries might be regressive or progressive dependent upon sonority. In Pali, consonantal assimilation across syllable boundaries is dependent upon a hierarchy (stop << fricative (s) << nasals << l << v <
PAGE 38

10 << r), where stops are least sonorous and [r] is most. They found that consonants that are more sonorous assimilate in all features to consonants of lower sonority. a. gam + ya gamma (gerundive 'go') b. lag + na lagga (p.p 'attach') c. dis + ya dissa (passive 'see') d. vak +ssa vakkha (future 'speak') Figure 2-4: Pali progressive assimilation a. kar + tum kattum (infinitive 'make') b. vas + tum vatthum (infinitive 'dwell') Figure 2-5: Pali regressive assimilation a. sup + ta sutta (p.p. 'sleep') b. sam + nisid sannisid (ind. 'be quiet') c. unmulayati ummuleti ('uproot') Figure 2-6: Pali regressive assimilation of equal segments In other words, the direction of assimilation is dependent upon whether the more sonorous segment is in the onset (Figure 2-4) or the preceding coda (Figure 2-5). If the consonants are of equal rank, assimilation is regressive (Figure 2-6). They also found that [s] assimilates to stops, suggesting higher sonority for [s], since it is always the less sonorous segment that assimilates (Figure 2-4) and (Figure 2-5). Without reference to sonority, these assimilation rules seem arbitrary, applying in one direction for some cases, but in the opposite direction for others. 2.2.2.3. Metathesis The Sonority Hierarchy has been used to explain other historical changes, such as processes of regular metathesis described by Hock (1985). Although metathesis at times seems irregular, Hock demonstrates that in some cases it is regular or systematic. In particular, cases of metathesis within syllable onsets and codas or at syllable boundaries may be due to a desire to adhere to sonority sequencing. Hock shows examples of historical changes where obstruent+sonorant sequences are metathesized from Latin to Old Spanish (Figure 2-7). Through a sonority account,

PAGE 39

11 Hock is able to give a motivation to what in the past was seen as an arbitrary metathesis of sounds. Latin titulum: *tidle OSpan. tilde 'title' capitulum: :*kabidle cabilde 'chapter' Figure 2-7: Old Spanish metathesis 2.2.2.4. Berber Dell and Elmedlaoui (1985) demonstrate a case where syllable peaks are determined not by featural descriptions of segments, but rather by scanning words for most sonorous segments. Essentially, in Berber any segment can serve as a syllabic peak. Words are scanned left to right iteratively for the most sonorous unparsed segment, following the sonority ranking in (Figure 2-8). low vowel /a/ >high vocoids /i,u/>liquids>nasals> voiced fricatives>voiceless fricatives>voiced stops>voiceless stops Figure 2-8: Berber sonority hierarchy The syllabification occurs following these steps. First, the most sonorous segment is parsed as the nucleus. Second, the segment immediately preceding the nucleus is parsed as the onset. Third, glide formation occurs if preceding segment is a vowel. Steps 1-3 are repeated until no more syllables can be formed. Berber does not allow abutting nuclei, so the process is finished once unparsed segments do not have segments to their left that could act as onsets within a syllable. Fourth, any unparsed segments are then parsed as codas (Figure 2-9). /ratlult/ Step One rA tlult Step Two r A tl U lt Step One r A tl U lt Step Two r A t l U lt Step Four r At.l Ult [rat.lult] (Capitalized segments are nuclei. Underlined segments are onsets) Figure 2-9: Berber syllabification

PAGE 40

12 Though this analysis can account for the syllabification that occurs in Berber, the iterative steps do not seem to be a plausible explanation of the data. Prince and Smolensky also attempt to account for the syllabification of Berber relying upon sonority, but do so in Optimality Theory that allows syllabification to be accounted for in one step. They do so by breaking sonority into two parts, segments that prefer to be in Margin and segments that prefer to be in Peak position, and then they rank the Margin constraints as well as faithfulness to input constraints above the Peak constraints. Besides not being iterative, the OT account demonstrates how different components of sonority can be broken down, opening the door for accounts by Kenstowicz (1997) that interleaves alignment constraints within sonority constraints to choose the most optimal output. 2.2.2.5. Moraic theory Sonority plays a role in moraic theory as well. Zec (1988, 1995) seeks to show that the mora, not the syllable, is key in capturing the sequencing of sounds. Yet, even without the syllable, Zec still depends on sonority to predict the alignment of segments and moras. A segment projecting a mora (which functions as weight within a syllable) must be equally sonorous as or more sonorous than the following segment. In a 1995 paper, Zec proposes the following constraint in Figure 2-10: i | ri rj Constraint on Moraic Prominence: Segment ri projects a mora iff it is not followed by a more sonorous segment rj. Figure 2-10: Constraint on moraic prominence The parsing of the word 'simply' in (Figure 2-11) does not violate the constraint.

PAGE 41

13 | | | s i m p l y Figure 2-11: Parsing of word ‘simply’ However those in Figure 2-12 and Figure 2-13. * | | s i m p l y Figure 2-12: Incorrect parsing of ‘simply’ * | | | | s i m p l y Figure 2-13: Incorrect parsing of ‘simply’ The first violates the constraint because [m] does not project a mora even though the following stop is less sonorous. The second fails because [p] projects a mora even though the following liquid is more sonorous. She demonstrates that this constraint plays an essential role in Pali where consonant gemination occurs when a segment that requires a moraic projection is lost due to syllable well-formedness constraints (Figure 2-14). i | | s u pi +t a [sutta] Figure 2-14: Parsing of /supta/ Regressive assimilation rather than deletion of [p] occurs to ensure that the mora that [p] projects is filled. 2.2.2.6. Steriade Steriade (1982) demonstrates the role that the sonority hierarchy plays in cluster simplification in Attic Greek. One example given is the deletion of consonants word

PAGE 42

14 medially and word-finally. For example, stop+[s] codas and [s]+stop onsets violate the Sonority Sequencing Generalization, and consequently, the [s] will be deleted (Figure 2-15). /hepstos/ hephthos /-graphsthai/ -graphthai Figure 2-15: Greek word internal [s] Steriade argues that [s]+stop sequences are allowed word initially, only because the [s] may be extrasyllabic in this position (Figure 2-16). That is, the [s] is not part of the onset. skapto skniphos Figure 2-16: Greek word initial [s] Steriade also takes the Sonority hierarchy one step farther with the Minimum Sonority Distance requirement (MSD). The MSD requires not only that segments follow sonority sequencing in syllable structure, but also that a minimum sonority distance be maintained between segments in a coda or an onset. In Attic Greek, MSD comes into play when some clusters, such as [ns], [sk], and [ng] simplify to one segment, while clusters such as [lg], with a further sonority distance, do not. The distance between segments may vary from language to language, so that one sequence is allowed in one language, but not in another. 2.2.2.7. Stress Kenstowicz (1997) uses sonority to account for stress placement. Kenstowicz assumes a sonority ranking where, for instance, [a] is more sonorous than [e]. That is lower and peripheral vowels are more sonorous and hence, not only make better syllable peaks, but also are more likely to receive stress. This topic is discussed further below.

PAGE 43

15 2.3. Phonetic Explanations of Consonant Clusters Disagreement among phonologists as well as a lack of phonetic correlates for sonority has caused skepticism among phoneticians who believe that phonological concepts should have measurable correlates (Ohala 1990; Kawasaki-Fukumori 1992). There has been some phonetic investigation into sonority, including both acoustic and articulatory studies. Traditionally, acoustic definitions seem to be more abundant with reference to noisiness, loudness, more sound, and the ability to carry farther distances. Ladefoged defines sonority as a sound's "loudness relative to that of other sounds with the same length, stress and pitch (1993, pg. 245)." However, articulatory definitions are also often found with reference to stricture and closure. Phonetic definitions of sonority are nothing new. Some early phonetic explanations of sonority include Saussure (1914) and Jakobson and Halle (1956). Saussure offered that sonority is most likely aperture. Syllables are formed by increasing and decreasing adduction and abduction. Jakobson and Halle broke sonority down into eight features. Each of the features is described in both acoustic and genetic terms. For example, the feature Vocalic was acoustically defined as " presence of a sharply defined formant structure" and genetically as "primary or only excitation at the glottis together with a free passage through the buccal tract" (p.40). Those segments that are neither optimally consonants nor vowels are sonorants. These include nasals and liquids: nasals because they have consonantal features, except for the nasal feature which superimposes "a clear-cut formant structure," and liquids because they contain features associated with both consonants and vowels. However, limitations in phonetic research made measurable analyses difficult and even seemingly impossible. For example, in 1974 Hankamer and Aissen expressed no

PAGE 44

16 doubt that phonetic correlates of sonority may exist; however, they believe that the complexity of sonority is too difficult to be answered with phonetic analysis. Of course, many advances in the ability to conduct phonetic research may have changed that strong prediction over time. More recently, efforts have been made to try to find some phonetic explanations of sonority sequencing. Experimental research on phonetic correlates for sonority includes the works of Lindblom (1983), Keating (1983), Price (1980) and Kawasaki-Fukumori(1992). While Lindblom and Keating investigated articulatory correlates of sonority, Price and Kawasaki-Fukumori considered the acoustic and perceptual correlates. Lindblom (1983) focused on how articulation affects the ability for certain segments to occur sequentially. He found that segments that are more difficult to coarticulate are found farther from each other, and segments that coarticulate with more ease are found close to each other. There is a need for ease of production by the speaker without compromising the listener's ability to distinguish segments. Lindblom looked specifically at jaw displacement of consonants in Ca:, a:C, and aC: syllables. Comparing the jaw height of consonants to the open jaw of [a], he found the following hierarchy of consonants in Swedish: s, , stops, m, j, n, v, r, l. [s] has the highest jaw placement and [l] has the lowest. This ranking closely resembles the sonority hierarchy. Segments with a higher jaw placement are less likely to coarticulate with a vowel and so are less likely to occur adjacent to a vowel. This explains why in consonant clusters [s] is often farther from a vowel than [l]. Keating (1983) built on Lindblom's proposal and considered a combination of jaw position, F1, and amplitude. From Lindblom's work, Keating hypothesized that, in

PAGE 45

17 clusters, jaw opening should steadily increase to the nucleus, and if there is a coda, then it should steadily decrease. A syllable consists of a steadily increasing F1 and amplitude that may or may not coincide with an opening of the jaw. That is because tongue position for some segments can vary independent of jaw position. While this would seem to correlate well with sonorants, it seems somewhat problematic for obstruents. Jaw height in fricatives results in higher intensity. Keating argued, though, that this inconsistency allows for the observed variation of placement of [s] in the sonority hierarchy. Raising of jaw position in [s] results in higher intensity. If intensity plays a role in sonority, [s] would make a good margin segment because of its high jaw placement and lack of coarticulation. However, it would also make a good nucleus because of its high intensity. Keating proposed that a multidimensional definition of sonority would help to explain cross-linguistic variations in segmental sequencing. Ohala (1990) perhaps offered the phonetic explanation of consonant clusters that most radically differs from the traditional phonological explanations. Ohala (1990) proposed that Sonority, as defined by linguists, should be thrown out. He suggests that weaknesses such as circularity and the failure to explain some phonotactic patterns imply that linguists are going down the wrong path. Ohala echoed a warning given by Saussure (1914) nearly a century earlier that sonority has the potential of becoming too cyclical. The sonority of a segment is decided by the role that it plays in syllable structure (as in Selkirk 1984), while the role a segment may play in syllable structure is decided by its sonority (as in Clements 1988). In addition, linguists place limitations on the patterns of sequences that the Sonority Hierarchy should account for. Past analyses have ignored or attempted to avoid fricative+stop onsets and stop+fricative codas, suggesting that sonority should not

PAGE 46

18 account for these phenomena. Instead, linguists rely on extrametricality or on labeling these examples as ‘exceptions’ to explain why this common phenomenon occurs (Morelli 1999). Ohala also argued that past accounts ignored relationships among different components of the syllable such as the frequency of some onset+nucleus combinations as opposed to others. For example, languages generally disfavor [w] before back rounded vowels or [j] before front vowels. Ohala argued that linguists who support the sonority hierarchy have drawn an arbitrary line between what sonority should and should not account for. As an alternative, Ohala suggested that combinations of segments should be accounted for by modulations. According to Ohala, nothing innate about segments requires that they occur in a certain order in syllable structure; rather, it is how the sounds interact concatenatively that limits the possible sequences. That is, certain segments occur cross-linguistically in predictable orders because of acoustic preferences for the orders. Fricatives precede stops because stops are more easily perceived before vowels than they are before fricatives. Ohala believes that "sonority" and "strength" explanations for segment sequencing should be replaced by acoustic parameters: periodicity, spectral shape, FO, amplitude. Rather than a single sonority feature, these acoustic features would result in preferences due to modulations in specific phonetic contexts. That is, the modulations of sounds because of their acoustic features would affect their perceptibility in certain contexts. Fricative+stop onsets are preferred to stop+fricative onsets because the stop is better perceived in the former. By contrast, some Sonority Hierarchies in the past have proposed that stops are less sonorous than fricatives, and thus stop+fricative clusters should be predicted by sonority sequencing in onsets. In actuality, fricative+stop clusters are more common than

PAGE 47

19 stop+fricative clusters (Greenberg 1978; Hock 1985). In addition, languages that allow stop+fricative clusters put restriction on them. For example, a language such as Sanskrit, which has many words with ksonsets, has very few psand tsclusters. And just as regular metathesis has brought about a reordering of obstruents and sonorants in syllables, so it has done with stop+fricative clusters. 2.4. Phonetic Explanations of English and Russian In this research, word-final clusters in English and Russian are compared to determine what similarities and differences exist in their production and perception. Such comparisons are interesting because both languages have a wide variety of consonant clusters. In addition, English and Russian exhibit differences in how they resolve consonant clusters into syllable. Comparisons between English and Russian phonetics include Zsiga (2000) who looked at variations in cross word boundary palatalization. Zsiga found that there is more likely to be a longer duration of overlap in English than in Russian, suggesting that a stronger overlap exists in the English gesture. Price (1980) attempted to find acoustic correlates through comparisons of minimal pairs where a sonorant is part of the onset in one word, and part of the nucleus in another. This would include such pairs as beret/bray, polite/plight, and parade/prayed. Assuming that sonority can account for the number of syllables a native-speaker detects, Price manipulated the duration, amplitude, hiss, and silence to determine how each affects the number of syllables a listener perceives. She found that (1) duration is a more effective cue to sonority than is amplitude, (2) amplitude may play a role when duration is ambiguous, (3) when duration is manipulated, voiced segments tend to be more sonorant than hiss-excited segments, which in turn appear more sonorant than silence, (4) absolute duration is more important to perceived sonority than relative sonority. (Price 1980: 342)

PAGE 48

20 Her findings, though, need further investigation since she considered only two segments (r, l) produced in only two syllable positions in one language. The most promising suggestion is that sonority may be the result of a combination of listener perception and ease of articulation (Keating 1983; Kawasaki-Fukumori 1992). Kawasaki-Fukumori (1992) does not look at the innate characteristics of segments, but rather looks at how segments interact with one another in sequences. Following Ohala (1990), Kawasaki-Fukumori considered what modulations occur between segments to determine if ease of perceptibility causes some clusters to be more likely than others. Specifically, Kawasaki-Fukumori considered spectral characteristics to determine if cross-linguistic preferences for certain clusters are acoustically driven. She wanted to determine a) if the degree of modulation affects how likely a segment is to be found, and b) if the similarities between one cluster and another limit the likelihood that a cluster would be found. That is, if it is difficult for listeners to distinguish between two clusters, such as [ps] and [ts], it is possible that a language will have one, but not likely that it would have both. Alternatively, if a language does have both, minimal pairs are not likely to be found. She does find that these auditory factors correlate with the occurrence of certain clusters. For example, voiced stop+[] clusters ([d], [b], and [g]) are spectrally distinct from one another. However, voiced stop+[l] clusters ([dl], [bl], and [gl]) are less distinct from one another. Kawasaki-Fukumori relates this to dialectal and historical variation in [gl] and [dl] clusters, that is not found in [] clusters. Flemming (1995) also referred to this in his research. Kawasaki-Fukumori concludes that ease of perception does play a role in the phonotactics found in the world's languages. However, she also concedes that some universal phonotactic patterns may be motivated by articulatory factors, so that universal

PAGE 49

21 sound patterns may be the result of "balancing the demands of both articulatory simplicity and perceptual distinctiveness (pg. 83)." 2.5. Phonetic Explanations of Russian Clusters Research on the quality of consonant clusters in Russian has focused primarily on onset clusters. Burton and Robblee (1997) focused on voicing neutralization in initial clusters. Specifically they looked at consonant clusters that were formed when words beginning with a sonorant were preceded by /s/, /z/, /d/, or /t/. They found in some cases where voicing assimilation occurred, some cues for voicing did remain, such as duration and amplitude of segments. This was particularly true if the sonorant was followed by a vowel. However, if the sonorant was followed by another obstruents the initial obstruent is likely to maintain its cues. 2.6. Consonant Clusters in Phonetic/Phonology Interface Similar to Kawasaki-Fukumori (1992), this study considers acoustic characteristics and perception. This project is distinct from past research in that it examines the relationship between acoustic measurements of consonants and their perceptibility in coda clusters and does so cross-linguistically. This research (1) looks in depth at the acoustic characteristics of segments in a variety of coda positions, (2) reviews where segments are most difficult to perceive, (3) compares the acoustics and perception to determine which acoustic cues are most salient in which word positions, and (4) applies the data to develop constraints to be used in an Optimality Theory model. The data from this study are being used to determine if sonority is phonetically supported. Similar to Ohala (1990), this work is undertaken as a reaction to the inconsistent definition given to sonority throughout phonological research. However, this study recognizes that the prevalence that sonority has played in phonology requires that it

PAGE 50

22 not be discarded completely. The overall goal of this study is to look for phonetic cues that would allow a more consistent definition for sonority. While Ohala’s explanation of consonant clusters suggests a strong separation from phonology, his analysis has sparked the interest of phonologists. Most notably, Steriade (1998) refers to Ohala’s (1990) assertion that syllable structure does not decide acceptable strings of segments, but rather follows from it. According to Steriade, the existence of syllable structure is undeniable in phonological representation, but it does not determine segmental sequences. Rather, sequences of segments are decided by ease of production and perceptions. Speaker’s intuition about syllables is decided by what sequences they feel would make an acceptable word in the language. It is possible that there is a trading relationship between articulatory, acoustic and perceptual cues to sonority across languages, and therefore a theory that would allow multiple constraints to act in competition and coordination is needed to account for this complex phenomenon. As pointed out by Clements (1990), any analysis should not compromise the cross-linguistic integrity of any definition of sonority for language specific preferences for ranking the sonority of voiced and voiceless segments or ranking fricatives as higher sonority than stops. However, understanding sonority might require a reliance on more than just articulatory features as Clements suggested. 2.7. OT and Consonant Clusters Optimality Theory (OT) (Prince and Smolensky 1993) analyses of sonority sequencing include Morelli (1999) in which she accounts for [s]+stop clusters by demonstrating that obstruent clusters need to be accounted for differently than other clusters. In her analysis, [s]+stop clusters are ill-formed, as is evidenced by their being divided into separate syllables word medially in many languages. However, this analysis

PAGE 51

23 is countered by Steriade (1999) that no clusters are ill-formed. In other words, that [s]+stop clusters are allowed in the language suggest that they are well-formed. The separation of [s]+stop clusters word-medially is the result of a word-formation preference, e.g. giving extra weight to the first syllable. Still, even in OT analyses there is a desire to call consonant clusters that do not fit the theory preferred pattern ill-formed or marked. One solution to this dilemma however may lie in the ability to inter-rank constraints on perception, articulation, and acoustics in one theory. Optimality Theory (OT) (Prince and Smolensky 1993), especially with reliance on acoustic constraints (as in Flemming 1995), which allows for various rankings of the sonority hierarchy, may prove to be an acceptable model to account for phonetic correlates of sonority engaged in a trading relation.

PAGE 52

CHAPTER 3 ACOUSTICS PROCEDURAL OVERVIEW 3.1. Introduction Past analyses have varied as to whether sonority should be defined articulatorily, acoustically, or perceptually. This study focuses on the latter two aspects of sonority. This dissertation is comprised of two studies--the first acoustic and the second perceptual. The acoustic experiment is designed to determine what modifications are made to consonants in a variety of word-final clusters in English and Russian. The perception experiment considers how these modifications affect perceptibility. Each of these studies is subdivided into an English and Russian portion. English and Russian were chosen because of their variety of word-final clusters, some of which seemingly violate the generally accepted sonority sequencing principle. English and Russian also present an interesting contrast; while there is some overlap in the clusters they allow, there also are significant differences. The acoustic study is designed to allow for the measurement of speech sounds in a variety of positions after the nucleus in a one-syllable word--in both clusters and singletons. The purpose is to determine what modifications are made to sounds dependent upon the environment that they are in and the rate at which they are spoken. An important consideration is whether sonority is an intrinsic feature of a phone or if it is a quality adopted by a phone when it falls in a certain word position. Similar to proposals made by Ohala (1990), this study considers how segments are modified in certain word positions 24

PAGE 53

25 by focusing on two acoustic characteristics: duration and amplitude. Measurements were taken so that comparisons of consonant quality could be made for individual speakers, for speakers of each language, and across the two languages. The data from the acoustic study were then applied to a perception study to determine what segments were best perceived in various word positions and in various levels of noise. The findings for both the acoustic and perceptual experiments were then applied theoretically to determine if a segment’s position on the sonority hierarchy is correlated with its perceptibility and what acoustic qualities of that segment allow it to be better perceived. 3.2. Hypotheses Because consonants higher on the sonority hierarchy are said to be louder, more sustainable, and more easily perceived, the hypotheses of the acoustic experiment are the following. Consonants lower on the sonority hierarchy would be more likely to decrease in duration as speaking rate increases. Consonants would decrease in duration when they are adjacent to consonants ranked higher on the sonority hierarchy since consonants ranked higher on the sonority hierarchy have longer duration and would be a greater proportion of the rhyme and coda. Consonants lower on the sonority hierarchy would be more likely to significantly decrease word-finally. Consonants that rank high on the sonority hierarchy would have a longer duration than those consonants ranked lower on the hierarchy in parallel environments. In terms of relative rms, it was the expected that a consonant’s relative rms would be related to its position in the sonority hierarchy. That is, segments that rank lower on the sonority hierarchy would be more likely to decrease their relative rms as speaking rate

PAGE 54

26 increased. In addition, consonants that rank higher would have a higher relative rms than consonants ranked lower on the sonority hierarchy in parallel word positions. 3.3. Acoustic Data Collection The goal of this experiment was to examine acoustic characteristics of English and Russian consonants and how these acoustic characteristics vary as a function of the position they occupy in a word and the rate at which they are spoken. From this experiment, it was determined if sonority could be influenced by constraints on the modifications of consonants, such as lengthening or increased intensity, in a variety of word-final positions. 3.3.1. Participants Five male native speakers of American English (NES) and one male native speaker of Russian (NSR) served as participants for the acoustic experiment. These participants were recruited through introductory linguistics courses taught at the University of Florida, the UF Russian program, the UF Russian Club, and personal contacts. In order to increase the pool of perspective participants, a five-dollar movie gift certificate was offered to participants who completed the project. This allowed the researcher better control of dialect, socio-economic background, and limited second-language experience. Prospective participants in both groups were administered a language background questionnaire to ensure that they met the requirements for participation. The American participants were University of Florida students who were native speakers of English all from Northern Florida. The English speakers were all in their early twenties. Four speakers had no prior exposure to Russian or any other Slavic language. One speaker

PAGE 55

27 (NES 4) reported having a high school friend whose parents were Polish, but he reported that he had little contact with them. The Russian participant was a 27-year-old native speaker of Russian from Minsk. He has been in the Untied States since 1990, but speaks English with a noticeable Russian accent. Like the English speakers, he had no history of hearing loss and did not have speech impediment or a learning disorder that might have inhibited his language development. 3.3.2. Stimuli This study focused on how individual consonants varied dependent on their environment. Therefore, the data set was designed to include consonants in at least two different word positions. All consonants occurred in at least one cluster and in isolation word-finally. Four different lists (2 for real words and 2 for nonsense words) of monosyllabic words were constructed for each language. Both real and nonsense words were collected for this study. Nonsense words were possible syllables predicted by the phonotactics of the languages. They were designed primarily for the perception experiment. They also had the added benefit of allowing some uniformity in the initial consonant in the word. Real words were also collected. The primary purpose for collecting real words was as a point of comparison with the nonsense words. This was to ensure that the speakers were pronouncing the nonsense words as if they were real words of either English or Russian. Both the real and nonsense words were used in the acoustic experiment. Only the nonsense words were used in the perception experiment. In total, four different “word” lists were elicited for each language. The first list was of real words with word-final consonant clusters. The second list was of nonsense

PAGE 56

28 words with word-final consonant clusters. The third list was of real words with word-final singleton consonants. The fourth list was of nonsense words with word-final singleton consonants. For the third and fourth word lists each of the consonants (p, t, k, b, d, , s, f, m, n, l, r) appeared word-finally. When possible, the vowels in the tokens were low mid or back vowels to reduce acoustic variations such as rounding, palatalization, and different formant transition. For Russian tokens, when a low vowel was not allowed by the phonotactics a mid back vowel [o] was used. The same procedure was followed to create the four wordlists needed for the native English speakers. Speakers of English somewhat varied the backness of the vowel dependent upon the following consonant cluster. This does suggest the need for future study in which the limits of vowels in segment sequencing is considered. The English real word tokens are shown in Table 3-1. The real word tokens used in the Russian experiment are shown in Table 3-2. The Russian word list was developed using Romanov’s Russian-English/English-Russian Dictionary (Wedel and Romanov 1964) and Bielfeldt’s Rcklufiges Wrterbuch der Russischen Sprache der Gegenwart (1965), a reverse dictionary that aided in finding the word-final clusters needed for the study. In addition, a native speaker of Russian reviewed the word list and each participant in the acoustic study was asked if any of the nonsense words were real to ensure that no real words were included in the perception experiment. Ideally, the nonsense word lists would have been identical for both languages, but two obstacles prohibited this. First, the phonotactics of each language constrained the vowels that could occur before certain clusters, and second, the lexicons of the two languages caused overlap. When an English nonsense word was a real word in Russian,

PAGE 57

29 that word was placed in the Russian real word list to allow for acoustic comparisons between these tokens. The English nonsense words are shown in Table 3-3. The Russian nonsense words are shown in Table 3-4. Table 3-1: English real word list Simple Codas1 ‘mock’ [mk] ‘far’ [fr] ‘doll’ [d] ‘gone’ [gn] ‘calm’ [km] ‘gosh’ [g] ‘rot’ [rt] ‘lob’ [lb] ‘knob’ [nb] ‘shot’ [t] ‘toss’ [ts] ‘mauve’ [mv] ‘bog’ [bg] ‘vol’ [vl] (name of sports rival) ‘wash’ [w] ‘pod’ [pd] ‘sop’ [sp] Complex Codas ‘mosque’ [msk] ‘font’ [fnt] ‘harsh’ [hr] ‘shelf’ [lf] ‘shark’ [rk] ‘romp’ [rmp] ‘farm’ [frm] ‘scarf’ [skrf] ‘false’ [fls] ‘solve’ [slv] ‘harp’ [hrp] ‘farce’ [frs] ‘part’ [prt] ‘carb’ [krb] ‘hopped’ [hpt] ‘card’ [krd] ‘walsh’ [wl] ‘bald’ [bld] ‘cost’ [kst] ‘barn’ [brn] ‘wasp’ [wsp] ‘fault’ [flt] ‘carl’ [krl] 1 The simple word lists have a variety of initial consonants because they were designed with the hope of considering both onsets and codas in this study. Otherwise the word list would have been designed with uniformity of the initial consonants. Onset consonants will be considered in future research.

PAGE 58

30 Table 3-2: Russian real word list Simple Codas ‘, [fat] ‘, [val] ‘, [sad] ‘, [lad] ‘, [dar] ‘, [tak] ‘, [mak] ‘, [gam] ‘, [pas] ‘, [bak] ‘, [ag] ‘, [rab] ‘, [xan] ‘, [gav] ‘, [na] Complex Codas ‘, [bars] ‘, [fars] ‘, [bant] ‘, [arf] ‘, [fakt] ‘, [gart] ‘, [kadr] ‘, [skarb] ‘, [vamp] ‘, [bask] ‘, [gorn] ‘, [karp] ‘, [xolm] ‘, [ark] ‘, [korm] ‘, [perl] ‘, [far] ‘, [nard] ‘, [opt] ‘, [last] ‘, [xald] ‘, [tolk] ‘, [porn]

PAGE 59

31 Table 3-3: English nonsense word list Simple Codas ‘mog’ [mg] ‘rol’ [rl] ‘dop’ [dp] ‘shon [n] ‘vock’ [vk] ‘cosh’ [k] ‘nar’ [nr] ‘fod’ [fd] ‘shoff’ [f] ‘gom’ [gm] ‘bot’ [bt] ‘nop’ [np] ‘poss’ [ps] ‘losh’ [l] ‘tob’ [tb] ‘sov’ [sv] Complex Codas ‘fomp’ [fmp] ‘gart’ [grt] ‘posk’ [psk] ‘parn’ [prn] ‘fark’ [frk] ‘tarm’ [trm] ‘farb’ [frb] ‘parl’ [prl] ‘pault’ [plt] ‘palse’ [pls] ‘fard’ [frd] ‘fosp’ [fsp] ‘polve’ [plv] ‘parsh’ [pr] ‘fald’ [fld] ‘pont’ [pnt] ‘fopt’ [fpt] ‘fost’ [fst] ‘barce’ [brs] ‘parf’ [prf] ‘palsh’ [pl] ‘farp’ [frp]

PAGE 60

32 Table 3-4: Russian nonsense words Simple Codas ‘, [vak] ‘, [an] ‘, [dap] ‘, [nar] ‘, [gan] ‘, [mas] ‘, [ka] ‘, [ga] ‘, [fad] ‘, [tab] ‘, [nab] ‘, [pal] ‘, [bat] ‘, [sag] ‘, [dam] ‘, [nap] ‘, [sav] ‘, [la] ‘, [af] Complex Codas ‘ [par] ‘ [parf] ‘ [bakt] ‘ [pald] ‘ [fark] ‘ [famp] ‘ [golk] ‘ [polm] ‘, [pask] ‘, [pant] ‘, [patr] ‘, [fard] ‘, [pln] ‘, [farb] ‘, [part] ‘, [farp] ‘, [kars] ‘, [fast] ‘, [palt] ‘, [fopt] ‘, [torm] Word lists were randomized for each speaker to ensure that the same list effects did not occur for each speaker. In addition, a buffer word was added to the beginning and end of each list that was not included in the data analyzed for this study. 3.3.3. Procedure All data were collected at the University of Florida. All participants were asked to produce the words from the wordlists using a high quality digital audio recorder (Sony, TCDD8) and a head mounted microphone (Shure, Model SM 10A). The microphone was kept at a consistent 3 cm from the speaker’s mouth and positioned at the left corner of the mouth. Recording took place in a double-walled IAC soundproof booth. Participants were read the instructions before production began. The Russian participants heard his instructions through a recorded message spoken in Russian in order to limit the effects of

PAGE 61

33 English on their production of the tokens. The instructions were read by a native Russian speaker from the Ukraine who had been in the United States for six years. Since speakers had little or no linguistic training, the tokens were written orthographically rather than phonetically. Tokens were presented to the participants in their native orthography (Roman or Cyrillic) to promote the sense that both the real and nonsense words were native to their language. Speakers were given an opportunity to review the tokens and practice saying them to promote natural pronunciation when recorded. This also ensured that the speakers understood the spelling methods used in the word lists. The English speakers were instructed that the letters ‘a,’ ‘o,’ and ‘au’ should be pronounced like the vowel in ‘hot.’ They were also instructed that word-final ‘e’ is silent. The native speakers of Russian were instructed that ‘a’ should be pronounced like the sound in the word ‘.’ The letter ‘o’ should be pronounced like the sound in the word ‘.’ This instruction was given to ensure that the participants understood how to read these new words. Each word was produced in a frame sentence "Please say _____ for me" for English and “Skazhite, pozhalujsta, _____ vslux” [skazhit poalusta ___ fslux] (“Say, please, _____ out loud”) for Russian. These relatively longer frame sentences are necessary to allow sufficient time for the desired speaking rate of the target word to be achieved. This specific English phrase was chosen since it has been shown to be useful for measuring relative amplitudes, as in Lavoie (2000). If a speaker mispronounced a word or read it with a peculiar stress pattern, that word was put into another word list for the speaker to reproduce.

PAGE 62

34 Phrases were read at three self-determined speaking rates: slow, casual, and fast. Speakers were told that ‘slow’ meant a pace slower than they would normally speak and ‘rapid’ meant a pace faster than their usual speech. ‘Casual,’ they were told, meant their natural speaking pace. Speakers were told to determine for themselves what it means to speak at these rates, but at all three rates, they should be able to be understood. These different speaking rates were used to allow for comparison in acoustic cues dependent upon the rate of speech. One goal of this study is to determine what cues speakers use to ensure that a segment is perceived, and different speaking rates may allow different cues to be realized. Past research has questioned whether speaking rate consistently affects the segment quality. This question is addressed for this particular data in the acoustic analysis. 3.4. Acoustic Analysis After data collection was completed, recordings of the stimuli were digitized at 25,000 Hz using PC Kay Elemetrics, Computerized Speech Lab (CSL). Each individual repetition of each word was then stored as a separate file for further acoustic analysis using Cool Edit. Tokens were analyzed using Kay Elemetrics Multi-Speech and PRAAT software (developed by Paul Boersma and David Weenink). 3.4.1. Acoustic Parameters to be Measured The acoustic analysis was designed to determine how segments are produced differently in various word positions. Most segments were compared in at least two positions, in isolation and in at least one complex cluster dependent upon what the phonotactics allow.

PAGE 63

35 Each segment was measured paying close attention to the acoustic cues associated with each individual segment. Several acoustic features were considered to determine what plays a role in sonority in individual languages. The primary features included in the analysis were intensity and duration, since consonants that are more sonorous are often correlated with an increase in these features. Measurements of duration include not only duration of the entire phone, but also closure duration, burst duration, and duration of formant transitions. The above qualities of each consonant were compared in various positions of monosyllabic words with simple and complex codas. These factors were evaluated to determine if any quality plays a more predominant role throughout the clusters produced by individual speakers, by speakers of the same language, or by speakers in both language groups. For each token, the duration of the whole word and phrase was measured. Comparisons were made to determine if the number and quality of segments in onset and/or coda affects the duration of the word. 3.4.2. Labeling Duration Each phone and word was labeled using a PRAAT text grid, which allowed for simultaneous analysis of both the waveform and wideband spectrogram. In most cases, primary consideration was given to the wideband spectrogram that allowed for close analysis of the start and end of formants and the lowering of F3. Individual consonants and vowels were defined by their duration, which are described in the following sections. These initial duration measurements were also used later in the analysis of intensity. PRAAT textgrids can be divided into several levels where multiple durations can be labeled. For each token, as many as eight levels were used. These included phrase duration, word duration, token vowel duration, coda sonorant duration, F3 duration, coda

PAGE 64

36 obstruent duration, and stop burst duration. In addition, an eighth tier was added. This tier measured the duration of a vowel in the carrier phrase as a point of comparison with the tokens. The following sections describe which qualities were used to decide where to place labels on each tier and how they relate to defining individual phones. 3.4.3. Words and Phrases For each token, the duration of both the word and phrase was measured. This was primarily to ensure that the speakers were able to self-determine speaking rate. Speakers were told that each speaking rate should be faster than the previous speaking rate. If they succeeded, slow speech should be slower than casual and rapid speech should be faster than casual speech. For the English phrases, the duration of the phrase was measured from the beginning of the burst of [p] in ‘please’ through the end of F2 in ‘me.’ The Russian phrases were measured from the beginning of the frication in ‘Skazhite’ to the end of frication in ‘vslux.’ For the English words, the beginning of the word was measured at the end of F2 in ‘say’ until the end of the final consonant in the token as described below. The Russian words were measured from the end of F2 of the [a] in ‘pozhalujsta’ until the end of the final consonant as described below. 3.4.4. Vowels, Consonants, Codas, and Rhymes For each token, the duration of each vowel and consonant as well as the entire coda was measured at each of the three speaking rates. The duration of the coda is the sum of all of the consonants’ durations. The duration of the rhyme is the sum of the vowel and the coda. Measurements of the entire coda and rhyme were computed to allow the duration of each consonant to be considered relative to the other consonant in the coda and rhyme.

PAGE 65

37 3.4.5. Vowel Labeling Although vowels are not the focus of this study, they vary in terms of duration in the presence of different consonants. Vowel length in English is known to be longer before voiced obstruents than before other consonants (Chen 1970; House and Fairbanks 1953; Raphael 1972). In addition, length of vowel steady state is known to be longer in the presence of [l] than in the presence of [], so this distinction may be exaggerated to enable recognition of liquids in clusters. Vowel length may then serve as cue for consonants in word-final position. For vowels, onset was measured at the beginning of F2. The vowel offset was measured as the onset of the following consonant, which could be any of the consonants being studied, as discussed in the following sections. 3.4.6. Sonorant and [s] Labeling This tier included sonorants that immediately followed a vowel, e.g. [] as in ‘far’ or ‘barn’ as well as [s] when it preceded a stop as in ‘cost.’ The onset of [s] began at the beginning of the frication noise following the vowel. The onset of [l] as well as the onset of nasals as in ‘barn,’ ‘bald,’ and ‘romp (Figure 3-1),’ was measured as the beginning of the anti-formants until the end of F2, since anti-formants are known to be cues for these consonants (Kurowski and Blumstein 1984). The beginning of the anti-formants was measured at a clear decrease in amplitude using both the waveform and spectrogram. When there was discrepancy between the waveform and spectrogram, the beginning of the anti-formants was taken as the one earlier in time. For [], the onset following the vowel was measured at the beginning of the decrease of F3.

PAGE 66

38 Time (s)0.384386 2.68137 0 104 Phrase Say Word V S C B Time (s)0.384386 2.68137 (V=vowel, S=sonorant, C=word-final consonant, B= stop-burst) Figure 3-1: Antiformant marking of word ‘romp’ Offset for sonorants before an obstruent and at the end of a word was measured as the offset of F2. In some cases, there was a brief closure between a sonorant and a following fricative. In these cases, the closure was measured as part of the sonorant, not the fricative (Clements 1987). Sonorant offset before another sonorant, such as the [] in ‘barn’ was measured as the onset of the second sonorant described below. 3.4.7. Labeling F3 For English [], a subcategory was also measured. After the decrease of F3, another measurement was taken of the duration that F3 held its lowest frequency (Figure 3-2). Special consideration was given to the duration of this quality because the lowering of F3 is known to be a strong cue for [] (Lehiste, 1964; McGovern and Strange 1977).

PAGE 67

39 In addition, formant transition are known to be longer for [] than for [l], suggesting that this may be a cue used to differentiate between the two. Time (s)0.635411 2.40565 0 104 Phrase Say Word Vowel Sonorant F3 Consonant Time (s)0.635411 2.40565 Figure 3-2: Marking of ‘harsh’ 3.4.8. Consonant Labeling This category included word-final obstruents and sonorants that followed another sonorant. This would include the [m], [n], [l] in ‘farm,’ ‘barn,’ and ‘Carl,’ [t] in ‘cost,’ [s] in ‘farce,’ and [p] in ‘sop.’ For words with multiple stop consonants, an additional consonant tier was added. A strong cue for all of the sonorants that occur in word-final position is anti-formants, so these were used as the onset cue. The onset of [l] following [] was measured at the beginning of anti-formant. When nasals followed [l], the onset of the nasal was measured at an increase of the anti-formants. As described above, the

PAGE 68

40 beginning of anti-formants were measured using both the waveform and the spectrogram. The offset of these word-final sonorants was measured as the offset of F2. For fricatives, the beginning of duration was measured as the beginning of frication. In cases where there was a pause between a final fricative and the beginning of ‘for’ in the carrier phrase, the end of frication was measured at the beginning of the pause. When the two fricatives ran together, the end of frication was marked at a change in energy, either a lowering in frequency at the end of a strident fricative or a change in amplitude between two non-strident fricatives. For stop consonants, the beginning of closure was measured at the end of the vowel or sonorant consonant, i.e. the end of F2. When the stop closure ran into the following fricative in ‘for,’ the end of the stop duration was measured as the beginning of the frication noise. If there was a stop burst, the offset of the stop was measured at the end of the burst. 3.4.9. Burst Labeling Since stop burst can be a strong cue for place of articulation, stop burst duration was also measured as a subcomponent of the stop duration. Burst was measured as the beginning of the noise until the end of the noise. When there was no burst, then its duration was said to be zero. Measuring the burst duration allowed for the closure duration to be calculated as well. This is significant since stop closure duration can be used as a cue for voicing in final position and may distinguish velar stops from the other places of articulation. 3.4.10. Duration Measurements Duration measurements were extracted using the labels of above. A PRAAT script was used to extract the numbers automatically. The numbers were placed in a

PAGE 69

41 Microsoft Excel file. Using Excel, the measurements of each of the consonants and consonant qualities (e.g. stop burst and duration of lowered F3 for []) was divided by both the duration of the coda and the rhyme. Comparisons of the consonants were made using these ratios since duration may vary depending on individual speaking rates of speakers and within each speaker. 3.4.11. Intensity Intensity was measured in Root Mean Squared amplitude. RMS was chosen following Lavoie (2000) who suggested that there might be a correlation between the sonority hierarchy and relative rms in English. Using the duration marks above, rms amplitude was measured for all of the consonants marked for duration as well as their subcomponents. Since rms is not an absolute value, but rather is affected by the loudness of the utterance, rms must be considered relatively. As in Lavoie, intensity was not measured in relationship to the rest of the word, but rather to a sound in another word in the carrier phrase. This is because variation in the compositions of the word (e.g. not all tokens begin with the same phone) would not allow rms to be measured relative to each token if cross-token comparisons were to be made. For the English phrase, rms amplitude was taken from the vowel [e] in ‘say.’ For the Russian tokens, rms amplitude was taken from [a] in ‘pozhalujsta.’ The rms for the consonants in the coda was then divided by the values of the rms of the carrier phrase vowels. 3.4.12. Comparisons After all phonetic cues were measured, comparisons were made to determine if there are trade-offs between cues in different word positions and at different speaking rates. For each speaker, consideration was given to whether a change in one acoustic cue

PAGE 70

42 that may enhance perception was countered with a decrease in another acoustic cue, or if an increase in one acoustic cue was coupled by an increase in another acoustic cue. After comparisons were made within each speaker, comparisons were made across speakers of the same language to determine if the same patterns were found for all speakers. Finally, the data were compared across both languages to determine if the patterns found in Russian were the same as the patterns found in English.

PAGE 71

CHAPTER 4 PERCEPTION PROCEDURAL OVERVIEW 4.1. Introduction The goal of this experiment is to examine the perceptibility of various consonant clusters. The nonsense word tokens from the acoustic experiment were presented in various levels of noise to determine if some consonants and consonant clusters are more easily perceived than others are. In addition, tokens from both languages were used to determine which consonants are more difficult to perceive when produced by speakers of different languages. It was also considered whether clusters that more closely adhere to the Sonority Sequencing Generalization (Selkirk 1984) and Minimum Sonority Distance Requirement (Steriade 1982) are more easily perceived than those that do not. 4.2. Hypotheses Consonants higher on the sonority hierarchy are said to be more perceptible. In addition, word-final clusters have a more gradual slope allowing for fewer distinctions between adjacent consonants. Given these two observations, it was expected that (1) consonants lower on the sonority hierarchy would be more difficult to perceive, particularly when the noise outweighed that sound and (2) consonants would be more difficult to perceive adjacent to consonants that they are closer to on the sonority hierarchy. 43

PAGE 72

44 4.3. Procedure 4.3.1. Participants A group of six (3 male, 3 female) native speakers of American English (NEL) and three (1 male, 2 female) native speakers of Russians (NRL) served as listeners for the second experiment The participants in this group met the same criteria as the participants in the acoustic data collection. The Russian listeners were all from the Ukraine, but spoke Russian all of their lives at home and in school. In addition, all participants in the perception experiment were required to pass a pure tone hearing screen at octave frequency from 250 to 8,000 Hz., at 25 dB SPL using a DSP Pure Tone Audiometer by Micro Audiometrics Corporation. This hearing range is necessary to ensure that participants can hear the variety of sounds included in the experiment, particularly fricatives whose frequencies extend to a relatively high frequency region. 4.3.2. Stimuli To reduce the effect of higher levels of information (i.e. syntactic and semantic) in the perception of consonant clusters, the stimuli that were used in this experiment were the nonsense words produced by the native English and native Russian speakers in the Acoustic Experiment. These words were presented in three levels of pink noise. Pink noise was chosen to lessen the masking effects on fricatives that may occur with white noise. The idea was to create an obstacle to hearing the tokens without masking out only one type of consonant. The acoustic characteristic of pink noise with more lower frequency components is also similar to natural speech. The signal to noise ratios were 0dB S/N, +6dB S/N, and -6dB S/N. These amounts are based upon signal to noise ratios used by Guion (1997). The ratios were designed so that the noise was great enough to cause difficulty, but not so great as to cause the listener to completely guess at responses.

PAGE 73

45 These ratios were pretested to determine if they were applicable to this study. For comparison, these tokens were also presented without masking noise. A band of pink noise was added to the tokens using Matlab. A sound sample was randomly taken from 60 seconds of pink noise, so that not all samples were masked with the exact same section of noise (See Appendix C for program)1. The program measured each token’s average energy, then created three new files combining the noise with token: the first with the sound 6 dB greater than the token, the second with the noise and token of equal energy, and the third with the noise file 6 dB below the token file. For the perception experiment, only the nonsense tokens were used. The experiment was comprised primarily of tokens with complex codas, as they were the primary focus of this study. A few tokens with simple codas were also included to prevent the listeners from assuming that all of the tokens had complex codas and as a means for the researcher to check that the listeners were not always choosing the option with a complex coda. Lists were randomized using the Research Randomizer2 to create 18 lists. Six English word lists of the English tokens were created for the six English listeners, i.e. one for each listener. Three different Russian word lists were created for the three Russian listeners. In addition six word lists of the Russian tokens were created for the English listeners and three word lists of the English tokens were created for the Russian listeners. 1 Thanks to Mark Skowronski for his assistance with writing this script. 2 The Research Randomizer is available at http://www.randomizer.org/ . It uses the "Math.random" method within the JavaScript programming language to generate its random numbers.

PAGE 74

46 4.4. Perception Data Collection The perception experiment was designed using the University of Alabama at Birmingham (UAB) Software designed at the UAB Department of Speech and Hearing. This software allowed for multiple versions of the same exam with the tokens presented in random order. For the English data, the program stored their responses easing the coding. The Russian data collection had to take place on paper because UAB does not support Cyrillic fonts, so the Russian data was coded by hand. Participants were presented with one word at a time presented in the carrier phrase from the acoustic data collection ‘Please say ______ for me.’ The tokens were all normalized to 98% Peak using the Modriff software that is included in the UAB package. This was to ensure that none of the tokens was louder and thus easier to perceive. All of the casual tokens were spliced into the same casual carrier phrase. All of the fast tokens were spliced into the same fast carrier phrase. The carrier phrases that were used were chosen because they had pauses before and after the token and had less creakiness particularly during the production of ‘me.’ Each token was played only once. No repetitions of the tokens were allowed. Upon hearing the word, participants were presented with four possible choices that might be the token they heard. For each multiple choice list, there was the correct response, two responses with each of the coda consonants missing, one response with a wrong consonant, for example [l] where an [] was the actual consonant in the token, and a fifth choice of none of the above if the listener believed none of the responses was correct. Therefore, the token ‘gart,’ would have the choices given (Figure 4-1).

PAGE 75

47 a. gart b. gar c. gat d. galt e. none Figure 4-1: Multiple-choice example If no response was given in 5 seconds, the program immediately moved to the next token. The lists were broken into sets of approximately twenty-five tokens, to give the participants an opportunity to take breaks if necessary (see Appendix D and Appendix E for entire multiple choice quizzes). Multiple-choice was chosen to ensure that the listener focused on the end of the word, not the onset. The noise made the task too difficult to complete without giving the participants some idea of what they might have heard. In addition, since the nonsense words sounded similar to native words of English, it was important to prevent the listeners to rely upon the lexical knowledge of their language. The data collection took place in a quiet room using a Dell PC. The same computer was used for all data collection to eliminate any variations that may occur using different sound cards. Participants wore Senneheiser Noisegard headphones. 4.5. Coding and Analysis The responses of the perceived tokens were compared with the correct responses to determine which consonants were lost or misperceived in various levels of noise. Individual consonants were coded as perceived, misperceived, or not perceived in each word position. Responses were compared to determine if consonants of the same manner of articulation behaved similarly in terms of how they are perceived by individual listeners. Sounds were compared between speakers of the same language to determine if the same patterns existed among the same language respondents. Then, sounds were

PAGE 76

48 compared cross-linguistically to determine if the same consonants are lost in both languages in the same environments. In addition, it was considered (1) if consonants were lost or misperceived in less noise when produced by speakers of the language not native to the listener and (2) if variations in cues were related to the consonants that are misperceived or lost. 4.6. Relationships between Acoustics and Perception After both experiments were completed, the data collected were compared to determine the relationship between acoustics and perception. The acoustic measurements were compared with the consonants' perceptibility to determine which modifications provide the most salient cues. Consideration was given to whether the tokens amplitude, duration, placement in the word-final cluster or some combination of these traits was correlated to the listener’s ability to perceive individual consonants.

PAGE 77

CHAPTER 5 ENGLISH ACOUSTIC ANALYSIS OF DURATION 5.1. Introduction This chapter compares the duration of each of the consonants in the word-final clusters of all of the tokens collected. First, before analyzing the individual consonants, this chapter shows that speakers did increase their speaking rate as they were instructed to do during data collection. Second, it describes how the duration of consonants varied dependent upon the speaking rate. Third, it describes how each consonant compared to the other consonants in parallel environments. Finally, this chapter shows how the consonants rank in terms of their durations in the various environments and relates those rankings to the sonority hierarchy. It was the hypothesis of this dissertation that consonants lower on the sonority hierarchy would be more likely to decrease in duration as speaking rate increases. In addition, consonants that rank high on the sonority hierarchy would have a longer duration than those consonants ranked lower on the hierarchy in parallel environments. Finally, it was the hypothesis of this dissertation that the ranking of the consonants by their duration in parallel environments would be similar to the sonority hierarchy. Because duration is relative to the individual speaking rate of each speaker during each repetition of the tokens, the duration of the consonant is taken as a percentage of the duration of both the coda and the rhyme. Duration was not taken as a percentage of the word since the initial consonant was not the same for all tokens. The coda is the duration 49

PAGE 78

50 from the offset of the vowel to the end of the word1. The duration of the rhyme is the duration from the onset of the vowel through the end of the word. Details as to how these divisions were made can be found in Chapter 3. The data for all five English speakers were collapsed and considered together for the analysis. In addition, the real and nonsense words were considered together. Each token was produced by each of the five speakers once for each speaking rate. Comparisons were made of the proportional duration of the consonants in the real and nonsense tokens in the same environments to ensure that there were no significant differences (p .05) because of the use of nonsense words. The voiceless obstruent+[l] clusters in the nonsense words were significantly different in the fast speech productions from the real words. In the real words, the obstruents were longer, and [l] was shorter than they were in the nonsense words. For all other cases, there were no significant differences in the production of the real and nonsense words. The findings for this chapter were similar to the sonority hierarchy in the most sonorous consonant [] was longer than the other consonants in parallel environments (preceding stops and fricatives). The findings also showed that the strident fricatives [s, ] were longer than the other consonants (stops, nasal, [l], and sometimes []) in parallel environments. The nasals and [l] were shorter than [] and the stridents fricatives, but they were not always longer than the voiceless stops. 1 I recognize that this is a rather loose definition of ‘coda’ since segments at word boundaries often considered ‘extrasyllabic’ (Fudge 1969, Steriade 1982, Bielfeldt, Hans Holm. 1965. Rcklufiges Wrterbuch der russischen Sprache der Gegenwart. Berlin: Akademie-Verlag Booij, Geert 1983, Clements and Keyser 1983, De Jong 1988).

PAGE 79

51 5.2. Speaking Rates According to the instructions during the collection of the data, speakers were told to read the list of tokens in the carrier phrase, ‘Please say___ for me”. The first time they were told to read at a slow pace, the second time at a casual pace, and the third time at a fast pace. For this study only, the casual and fast speech tokens were considered. Since some consideration in this chapter is given to how the phones varied according to speaking rate, it is important to verify that speakers did successfully accomplish this task. Comparisons were made of the word duration to ensure that speakers succeeded at speaking faster from casual to fast speech. The duration of all of the tokens considered for this study were considered when determining if the speaking rate increased. The duration of all of the tokens were averaged together since the tokens were the same for all of the speakers. Considering the speaking rate of each of the tokens across speakers, the fast speaking rate was significantly faster than casual (Table 5-1). Table 5-1: Mean average rate of all tokens Casual Fast p-value 0.464 sec 0.392 sec .000 Because the speaking rates were self-determined, the casual speaking rate of NES1 does not equal the casual speaking rate of NES2. Future study might separate tokens by their actual rate rather than their self-determined rate, however such a task is out of scope with the goals of this dissertation. 5.3. Comparisons across Speaking Rates Consonants were compared at each speaking rate in the same environments. Proportional duration was compared using a t-test where the null hypothesis was no

PAGE 80

52 difference in the proportional duration of the consonant despite a change in speaking rate. Significant was considered a p-value of less than or equal to .05. When comparing speaking rates, most significant differences in quality occurred between slow and casual and slow and fast speech. This may be due to the fact that the slow speech was the speakers’ first run through each list, so in addition to speaking more slowly, speakers may have been speaking more carefully. For this study, only the casual and fast tokens were considered. 5.3.1. Obstruents 5.3.1.1. Stops Stops occurred word-finally following each of the sonorants, [s], and [p], and preceding [t] (Table 5-2). Table 5-2: Tokens with stops Environment Real Word Nonsense Word Preceding and following stop hopped fopt Following [s] cost fost Following nasal romp font fomp pont Following [l] fault pault Following [] carb card harp part shark fard farb farp gart fark Table 5-3 shows the proportion of the rhyme devoted to the stops following sonorants, [s], following [p], and preceding [t].

PAGE 81

53 Table 5-3: Duration of stops Casual Rhyme Fast Rhyme p-values Casual Coda Fast Coda p-values Following sonorants 32% 29% 0.008 46% 42% 0.018 following [s] 24% 19% 0.025 40% 34% 0.062 following [p] 29% 22% 0.06 53% 42% 0.117 preceding [t] 27% 29% 0.311 47% 58% 0.117 The duration of the word-final stops decreased from casual to fast speech in all three environments. Comparing stops preceded by sonorants, the duration of the stop significantly decreased as speaking rate increased. Considering the duration of the stops following [s], the duration of the stop on average decreased as speaking rate increased. This difference is significant as a portion of the rhyme when comparing when comparing casual and fast. Considering [t] following [p], the duration also decreased, but the difference was not significant Stops were also found in the data preceding word-final [t] (‘hopped’ and ‘fopt’). In this case the proportion of both the rhyme and coda devoted to [p] on average did increase as speaking rate increased, although not significantly. Note also that [p] increased more as a proportion of the coda than as a proportion of the rhyme. This shows that it is gaining its duration from the shortened word-final [t], not from the vowel as the sonorants do which is shown in the following section. As was expected, the duration of the obstruent stops decrease word-finally as speaking rate increase. The duration of [p] increased when it occurred preceding a word-final stop. However, this was due to the shortening of the word-final [t].

PAGE 82

54 5.3.1.2. Fricatives The fricative [s] was found both preceding stops and word-finally after sonorants as is shown in Table 5-4. Table 5-4: Tokens with [s] Environment Real Word Nonsense Word before stops wasp cost mosque fosp fost posk Following [] farce barce Following [l] false palse Table 5-5 shows the duration of [s] preceding stops and following sonorants. Table 5-5: Mean proportional duration of [s] Casual Rhyme Fast Rhyme p-values Casual Coda Fast Coda p-values preceding stops 34% 35% 0.383 60% 66% 0.062 following sonorants 47% 44% 0.052 58% 56% 0.152 When [s] appeared between a vowel and a word-final stop, its proportion of the rhyme did not significantly increase as speaking rate increased (34% and 35%). Its proportion of the coda increased more from casual to fast speech (60% and 66%), but still not significantly. Conversely, following sonorants the duration of fricatives decreased as speaking rate increased (Rhyme 47% and 44%, Coda 58% and 56%). In this case, there is significance in the difference of the proportion of the rhyme, but not the coda. Separating the sonorant+fricative clusters by the manner of the sonorants uncovers more information about the fricatives behavior after sonorant consonants.

PAGE 83

55 Adjacent to [l], the strident fricatives did not significantly change as speaking rate increased (Table 5-6). Table 5-6: Mean proportional duration of stridents Casual Rhyme Fast Rhyme p-values Casual Coda Fast Coda p-values following [l] 48% 46% 0.628 67% 68% 0.682 following [] 48% 45% 0.051 54% 51% 0.063 In fact, the duration of the fricative remained a stable part of the rhyme and the coda at all speaking rates in this environment. The duration of the strident fricatives following [] decreased as speaking rate increased (Table 5-6). Just considering the tokens where [s] follows [], the difference in duration is significant for all cases (Table 5-7). Table 5-7: Mean proportional duration of [s] following [] Casual Rhyme Fast Rhyme p-value Casual Coda Fast Coda p-value 46% 43% 0.029 53% 49% 0.013 Like the strident fricatives, the duration of [f] also decreased as speaking rate increased (Table 5-8). Table 5-8: Mean proportional duration of [f] following [] Casual Rhyme Fast Rhyme Casual Coda Fast Coda 43% 39% 50% 44% p-values 0.312 0.123 Unlike [s] though, the duration of [f] was not significantly different from casual to fast speech. This may reflect that the duration of [f] in casual speech is shorter than that of the strident fricatives giving it less of an opportunity to decrease in fast speech.

PAGE 84

56 Speaking rate then more significantly affected the duration devoted to [s] than it affected the duration devoted to [f] word-finally. In addition, [s] was most greatly affected by speaking rate when it occurred word-finally rather than when it preceded a word-final voiceless stop. The fricatives agree with the hypothesis that the duration of obstruents would decrease as speaking rate increased when they occurred word-finally. 5.3.2. Sonorants When [] occurred immediately after the vowel, it tended to increase its proportion of both the coda and the rhyme. However, this increase was not always significant. The other sonorants ([l] and the nasals) did not consistently increase or decrease with speaking rate. The remainder of this section outlines how each of the sonorants varied its proportion of the rhyme and coda according to speaking rate. 5.3.2.1. Nasals Tokens with nasal consonants, had [m] and [n] occurring word-finally after [] and occurring preceding word-final stops as is shown in Table 5-9. Table 5-9: Tokens with nasal consonants Environment Real Words Nonsense Words Preceding Stops font romp pont fomp Following [] barn farm parn tarm When nasals preceded voiceless stops, the mean average of the proportion of the rhyme and coda devoted to the nasal increased as speaking rate increased, but not significantly (Table 5-10).

PAGE 85

57 Table 5-10: Mean proportional duration of nasals Casual Rhyme Fast Rhyme p-values Casual Coda Fast Coda p-values preceding voiceless stops 20% 23% 0.057 41% 48% 0.091 following [] 29% 29% 0.738 38% 39% 0.594 For both the percentage of the rhyme and coda devoted to the nasal preceding voiceless stops at each speaking rate, the proportion devoted to the nasal is longer in fast speech than in casual speech (Rhyme casual 20% and fast 23%, Coda casual 41% and fast 48%). However, this difference was not significant. This increase in nasal duration reflects the shortening of the duration of voiceless stops word-finally described above. The same was found when nasals occur word-finally after []. In this case, the duration of the nasal devoted to the rhyme remained consistent from casual to fast speech (Rhyme casual and fast 29%, Coda casual 38% and fast 39%). Note that this is different from [s], which significantly decreased with each change in speaking rate. The duration of the nasals behaved differently from the fricatives in that they remained the same word-finally from casual to fast speech. Preceding stops though, nasals like [s] and [p] increased in proportional duration as speaking rate increased. 5.3.2.2. Lateral approximant [l] The lateral sonorant [l] was found in several positions in the data, both word-finally and preceding obstruents (Table 5-11). Table 5-11: Tokens with [l] Environment Real Words Nonsense Words [l] before voiceless stop fault pault [l] before voiced stop bald fald [l] before voiceless fricative false Walsh palse palsh [l] after [] Carl parl

PAGE 86

58 Table 5-12 shows the comparisons of the duration of [l] in each environment in both speaking rates. Grouping all of the tokens with [l] preceding obstruents together shows that the duration of [l] in this position remained essentially the same with no significant difference (Rhyme casual 24%, Rhyme fast 23%, Coda casual 34%, and Coda fast 33%). Following [], the duration of [l] slightly increased from casual to fast speech, but not significantly (Rhyme casual 37%, Rhyme fast 40%, Coda casual 44%, and Coda fast 46%). Table 5-12: Mean proportional duration of [l] at different speaking rates Casual Rhyme Fast Rhyme p-values Casual Coda Fast Coda p-values preceding obstruents 24% 23% 0.663 34% 33% 0.908 following [] 37% 40% 0.652 44% 46% 0.789 Breaking down the data such that [l] precedes only voiced obstruents, voiced stops, voiceless obstruents, voiceless stops, or voiceless fricatives gives the same results with no significant differences in the proportional duration of [l]. This was also true when [l] followed []. In this position, the proportional duration never significantly changed dependent upon speaking rate. While the other less sonorous consonants were significantly affected by speaking rate, [l] was not. Word-finally, [l] behaved more like the nasals than like [s]. The proportional duration increased from casual to fast. Preceding obstruents, [l] behaved differently than both [s] and the nasals holding steady its proportional duration from casual to fast speech.

PAGE 87

59 5.3.2.3. Rhotic approximant [] The data collected with [] in the consonant cluster included a wide variety of environments, as is shown in Table 5-13. Table 5-13: Tokens with [] Real Words Nonsense Words Preceding stops harp part shark carb card farp gart fark farb fard Preceding fricatives scarf farce harsh parf barce parsh Preceding nasals farm barn tarm parn Preceding [l] Carl parl Table 5-14 shows the differences in duration of [] in different word position dependent upon speaking rate. Table 5-14: Comparisons of [] at different speaking rates Casual Rhyme Fast Rhyme p-values Casual Coda Fast Coda p-values preceding voiceless stops 49% 50% 0.610 58% 59% 0.476 preceding voiced stops 49% 50% 0.683 58% 59% 0.661 preceding fricatives 41% 45% 0.005 47% 51% 0.013 preceding strident fricatives 40% 43% 0.084 46% 49% 0.063 non-strident fricatives 43% 49% 0.032 50% 55% 0.123 preceding nasals 49% 47% 0.454 62% 61% 0.594 preceding[l] 47% 47% 0.998 56% 54% 0.789

PAGE 88

60 Preceding the obstruents, the duration of [] always remained about the same from casual to fast speech. Preceding the stops, this difference was never significant. The same was also true when comparing the duration of [] before voiced stops. Preceding fricatives, [] significantly increased its proportion of the rhyme with from casual to fast speech. Preceding strident fricatives the relative duration also increases, however not significantly when comparing casual and fast speech. Preceding non-strident fricatives, the duration of [] also increased. In this case, as a proportion of the rhyme, the duration increased significantly from casual to fast speech. Preceding nasals, [], there is no significant difference between its duration in fast and casual speech. Preceding [l], however, there is no significant difference in proportional duration from casual to fast speech. That this is the one environment in which no significant differences are found is interesting since [l] is the most sonorous consonant that [] appears adjacent to. So [] tends to increase both its duration of the rhyme and coda as speaking rate increases. This happens most significantly when [] precedes an obstruent. The difference is least significant when followed by the sonorant consonant [l]. Note that [] is most significantly changed with speaking rate adjacent to the fricatives, especially the strident fricatives. However, [] is not significantly changed when it is adjacent to the nasals, stops, and [l].

PAGE 89

61 5.4. Comparisons by Manner of Articulation 5.4.1. Stops vs. Fricatives Fricatives and stops were found in similar environments both word-finally and preceding word-final [t]. Since fricatives are higher on some versions of the sonority hierarchy, it was believed that fricatives would have a longer proportional duration than stops in the same position. For stops, the duration included both the gap and the release. Both stops and fricatives occurred preceding [t], following nasals, following [l], and following []. Preceding [t], [s] was significantly longer in casual speech, and nearly significantly longer in fast speech than the stop [p] (Table 5-15). Table 5-15: Mean proportional duration of [p] and [s] preceding [t] Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘hopped,’ ‘fopt’ 27% 29% 47% 58% ‘cost,’ ‘fost 36% 37% 64% 75% p-values [p] vs. [s] 0.006 0.083 0.014 0.056 In a sonority hierarchy where obstruents are divided into fricatives and stops, this is to be expected as the more sonorous consonant has longer duration in the same environment. Following [l], [s] was always on average longer than [t] (Table 5-16). However, preceding [t], there was no significant difference between the duration of [s] and [t] except in the fast rhyme. Table 5-16: Mean proportional duration of [t] and [s] following [l] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘fault,’ ‘pault’ 36% 30% 55% 53% ‘false,’ ‘palse’ 44% 43% 65% 64% p-values [t] vs. [s] 0.502 0.022 0.698 0.483

PAGE 90

62 Following [], though, there were significant differences in the duration of the stops and the strident fricatives (Table 5-17). Table 5-17: Mean proportional duration of stops ([t] and [k]) and strident fricatives following [] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘part,’ ‘shark,’ ‘gart,’ ‘fark’ 33% 31% 39% 37% ‘farce,’ ‘harsh,’ ‘barce,’ ‘parsh’ 48% 45% 54% 51% p-values stops vs. strident fricatives 0.001 0.000 0.001 0.000 In this word-final position, the strident fricatives were significantly longer than the stops. Considering only the stops and the fricatives with the same place of articulation, [s] and [t], [s] was still longer than [t], although not significantly in casual speech (Table 5-18). Table 5-18: Mean proportional duration of [t] and [s] following [] Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘part,’ ‘gart’ 33% 30% 38% 36% ‘farce,’ ‘barce’ 43% 43% 51% 52% p-values [t] vs. [s] 0.148 0.002 0.098 0.002 Considering the non-strident fricative [f] and the bilabial stop [p], however, the same differences were not found (Table 5-19). Table 5-19: Mean proportional duration of [p] and [f] following [] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘harp,’ ‘farp’ 44% 41% 50% 48% ‘scarf,’ ‘parf’ 43% 39% 50% 44% p-values [p] vs. [f] 0.551 0.179 0.395 0.082 The stop [p] was longer than [f] as a proportion of the rhyme in both casual ([p] 44%, [f] 43%) and fast speech ([p] 41%, [f] 39%). As a proportion of the coda the durations of [f]

PAGE 91

63 and [p] were equal (50%) and the duration of [p] was longer in fast speech ([p] 48%, [f] 44%). However, none of the differences were significant as was the difference between [s] and [t]. Fricatives are, on average, longer than stops in parallel positions. However, when fricatives are separated by stridency, only the strident fricatives are longer than the stops in all positions. Word-finally, the only significant differences are between the strident fricatives and stops following []. 5.4.2. Stops vs. Nasals Both stops and nasals occurred word-finally following [] and preceding word-final stops. Since nasals rank higher on the sonority hierarchy, it was expected that they would have a longer proportional duration in parallel environments than the stops (Table 5-20). Table 5-20: Mean proportional duration of voiced stops and nasals following [] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘harp’, ‘part’, ‘farp’, ‘gart’ 38% 36% 43% 42% ‘carb’, ‘card’, ‘farb’, ‘fard’ 25% 24% 32% 31% ‘farm’, ‘barn’, ‘tarm’, ‘parn’ 29% 29% 38% 39% p-values Voiceless stop vs. nasal 0.022 0.021 0.221 0.446 Voiced stop vs. nasal 0.171 0.131 0.076 0.096 Comparing the nasals to both voiced and voiceless stops, the mean average duration of nasals was a longer portion of both the rhyme and coda in both casual and fast speech. However, this difference was only significant when comparing the proportion of the rhyme for the nasals and the voiceless stops. This is not surprising since the duration of stop compared to vowel duration is a voicing cue for word-final stops, so the voiced stops were not as long as the voiceless stops. However when not considering the vowel

PAGE 92

64 duration, the significant difference between the stops and nasals is lost. On average, they have a closer proportional duration. In no case was the difference in coda duration significant when comparing stops to nasals. When the nasals and stops are separated by place of articulation, as in Table 5-21, the difference is significant in fast speech when comparing [d] and [n] for both rhyme and coda. Table 5-21: Mean proportional duration of [t], [d] and [n] following [] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘part’, ‘gart’ 33% 30% 38% 36% ‘card’, ‘fard’ 23% 19% 28% 24% ‘barn’, ‘parn’ 28% 28% 36% 35% p-values [t] vs. [n] 0.573 0.573 0.945 0.797 [d] vs. [n] 0.153 0.021 0.117 0.019 Comparing the voiced labial consonants, the duration of [m] is longer than the duration of [b]. However, the difference is not significant in either casual or fast speech (Table 5-22). Table 5-22: Mean proportional duration of [p], [b] and [m] following [] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘harp,’ ‘farp’ 44% 41% 50% 48% ‘carb,’ ‘farb’ 27% 30% 36% 39% ‘farm,’ ‘tarm’ 29% 31% 39% 44% p-values [p] vs. [m] 0.001 0.000 0.000 0.099 [b] vs. [m] 0.58 0.76 0.44 0.09 Comparing the voiceless stop [p] and [m], though, [p] was a significantly longer portion of the rhyme and a significantly longer portion of the coda in casual speech. This likely represents the long stop duration to signal a voiceless stop. Both stops and nasals occur before word-final stops (Table 5-23).

PAGE 93

65 Table 5-23: Mean proportional duration of [p], [n] and [m] preceding [t] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘hopped,’ ‘fopt’ 27% 29% 47% 58% ‘font,’ ‘pont’ 22% 27% 45% 50% ‘romp,’ ‘fomp’ 21% 27% 40% 54% p-values [p] vs. [n] 0.112 0.700 0.648 0.473 [p] vs. [m] 0.328 0.737 0.546 0.879 In this case, also there was no significant difference between the durations of the stops and nasals. The durations of all three increased with speaking rate and remained almost equal. In summary, word-finally the duration of the nasals was greater than the duration the voiced stops, but not always significantly. This agrees with the hypothesis that consonants higher on the sonority hierarchy would have greater duration. However, the duration of the nasals and the voiceless stops were about equal preceding word-final voiceless stops. This did not agree with the prediction of the hypothesis. Word-finally, the voiceless stops were longer than the nasals. This did not agree with the hypothesis and was likely because of the long voiceless stops to signal a voicing distinction from the word-final voiced stops. 5.4.3. Stops vs. Lateral Approximant [l] In the data stops and [l] were found in two of the same environments, preceding stops and following []. For comparison, the tokens with stop following [] that were used were ‘part,’ ‘card,’ ‘gart,’ and ‘fard’ since [t] and [l] are both alveolar consonants. [p] was used in comparison to [l] for the tokens preceding [t] even though they are not the same place of articulation. This is because [t] does not occur in this position, yet the comparison is worthwhile since [l] is considered more sonorous than [p].

PAGE 94

66 Comparing [l] and the word-final stops, [l] is proportionally longer than voiceless stop, but not significantly except in fast speech. However, [l] is significantly longer than [d] as a portion of both the coda and rhyme in both speaking rates (Table 5-24). Table 5-24: Mean proportional duration of [t], [d], and [l] following [] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘part,’ ‘gart’ 33% 30% 38% 36% ‘card,’ ‘fard’ 23% 18% 28% 24% ‘Carl,’ ‘parl’ 37% 40% 44% 46% p-values [t] vs. [l] 0.264 0.046 0.210 0.054 [d] vs. [l] 0.006 0.003 0.008 0.006 This, however, should come as no surprise since [d] is shorter than [t] word-finally in English which is a cue for word-final obstruent voicing. What is interesting is that the duration of [t] becomes significantly shorter than [l] in rapid speech suggesting that this cue for voicing may be lost as speaking rate increase. The other place where both [l] and a stop occur is preceding a word-final [t]. In this case the stop is [p]. Although, the place of articulation is not the same, comparison were made since [l] is said to outrank [p] on the sonority hierarchy. In this position, neither consonant is consistently longer than the other is, and there is no significant difference in the duration (Table 5-25). Table 5-25: Mean proportional duration of [p] and [l] preceding [t] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘hopped,’ ‘fopt’ 27% 29% 47% 58% ‘fault,’ ‘pault’ 28% 27% 45% 47% p-values [p] vs. [l] 0.607 0.684 0.921 0.240 [l] is longer than the stops word-finally as would be expected since [l] is more sonorous than stops according the sonority hierarchy. This, though, is not found

PAGE 95

67 preceding the word-final stop [t] where there is no significant difference between [p] and [l], and at times [p] on average is actually longer than [l]. 5.4.4. Stops vs. Rhotic Approximant [] The only stop that occurs in the same environment as [] was the voiceless bilabial stop [p]. Although, these phones are not the same place of articulation, they are still of interest for this research, since [] is ranked higher on the sonority hierarchy than [p]. The tokens with [p] and [] in the same rhyme position were those with the word-final voiceless alveolar stop [t] In both speaking rates, [] was a greater proportion of both the rhyme and coda. In both speaking rates, the difference was significant with the exception of the duration of the coda in fast speech (Table 5-26). Considering only low F3, [] is still longer, but significantly only in casual speech as a portion of the rhyme. In general, this agrees with the hypothesis that [] would have a greater duration than the stop [p], but only significantly when the F3 duration is included. Table 5-26: Mean proportional duration of [p], [], and F3 of [] preceding stop Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘hopped,’ ‘fopt’ 27% 29% 47% 58% ‘part,’ ‘gart’ all of [] 53% 53% 62% 64% low F3 of [] 33% 34% 52% 54% p-values [p] vs. [] 0.000 0.000 0.009 0.314 [p] vs. F3 0.013 0.224 0.208 0.553

PAGE 96

68 5.4.5. Fricatives vs. Nasals Nasals and fricatives occurred preceding stops and word-finally following []. Word-finally, comparing only the strident fricatives and the nasals, at both speaking rates voiceless fricatives were significantly longer than nasals following [] (Table 5-27). Of course voiceless stops and fricatives are longer than their voiced counterparts word-finally, but still the comparison is significant for this study since voiceless stops are less sonorous than nasals. Table 5-27: Mean proportional duration of nasals and strident fricative following [] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘farce,’ ‘harsh,’ ‘barce,’ ‘parsh’ 48% 45% 54% 51% ‘barn,’ ‘farm,’ ‘parn,’ ‘tarm’ 29% 29% 38% 39% p-values [s, ] vs. [m, n] 0.000 0.000 0.000 0.000 This is also significant since the stridents’ duration decreased significantly with the increase in speaking rate following [], while the duration of the nasals did not. Yet, the duration of the strident fricatives remains significantly longer than the nasals. The same is true when comparing only the alveolar consonants [n] and [s] (Table 5-28). Table 5-28: Mean proportional duration of [s] and [n] following [] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘farce,’ ‘barce’ 43% 43% 51% 52% ‘barn,’ ‘parn’ 28% 28% 36% 35% p-values [s] vs. [n] 0.004 0.002 0.006 0.006 In this case, [s] is significantly longer than [n] as both a proportion of the rhyme and coda. [s] is 43% of the rhyme in both casual and fast speech, while [n] is only 28% of the rhyme.

PAGE 97

69 Comparing the labial nasal [m] and labial fricative [f], the same pattern is found for rhyme duration (Table 5-29). Table 5-29: Mean proportional duration of [f] and [m] following [] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘scarf,’ ‘parf’ 43% 39% 50% 44% ‘farm,’ ‘tarm’ 29% 31% 39% 44% p-values [f] vs. [m] 0.007 0.013 0.014 0.834 This suggests that following [], the nasal [m] is more stable than [f] in how it interacts with []. However, the same is not found as a portion of the coda. Remember that the duration of [f] does not significantly decrease like [s] following [] likely because the duration of [f] in both speaking rates is shorter than [s]. Another environment that both [s] and nasals were found is preceding voiceless stops. The tokens used for these comparisons were those with stops with the same places of articulation, i.e. alveolar and bilabial. In these cases, the duration of [s] was significantly longer than the nasals (Table 5-30). Table 5-30: Mean proportional duration of nasals and [s] preceding voiceless stops Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘wasp,’ ‘cost,’ ‘fosp,’ ‘fost’ 35% 35% 61% 66% ‘romp,’ ‘font,’ ‘fomp,’ ‘pont’ 21% 27% 42% 52% p-values [s] vs. nasals 0.000 0.000 0.000 0.000 Considering all of the environments that nasals and [s] occurred in, nasals, which are ranked higher on the sonority hierarchy than [s], are shorter than [s] in word-final position after [] and preceding voiceless stops. Comparing [m] to [f], the duration of [f] is greater than [m] in casual speech, but in fast speech, the durations are equal.

PAGE 98

70 5.4.6. Fricatives vs. Lateral Approximant [l] Comparing the fricatives and [l], only [s] was compared with [l] since both consonants are alveolar. There are two environments where both of these consonants are found: following [] and preceding [t]. Both following [] and preceding [t], [s] on average was longer than [l]. However, the difference was greater and more significant when the consonants preceded [t] than it was when the consonants followed [] (Table 5-31). Table 5-31: Mean proportional duration of [s] and [l] preceding [t] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘cost,’ ‘fost’ 36% 37% 64% 75% ‘fault,’ ‘pault’ 28% 27% 45% 47% p-values [s] vs. [l] 0.010 0.097 0.000 0.015 Following [], on average [s] was still longer than [l] (Table 5-32). In this case, [s] is longer than [l], but not significantly. Table 5-32: Mean proportional duration of [s] and [l] following [] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘farce,’ ‘barce’ 43% 43% 51% 52% ‘Carl,’ ‘parl’ 37% 40% 44% 46% p-values [s] vs. [l] 0.435 0.316 0.378 0.121 [s] and [l] varied significantly preceding a word-final stop, but did not vary when occurring word-finally. This is different form [s] compared to the nasals where [s] was usually significantly longer than [n].

PAGE 99

71 5.4.7. Fricatives vs. Rhotic Approximant [] The only environment that both [] and [s] occurred in is preceding voiceless stops. When considering [] from the beginning of the F3 transitions, [] was a significantly longer portion of the rhyme than [s] in both speaking rates (Table 5-33)? Table 5-33: Mean proportional duration of [s] and [] preceding voiceless stop Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘wasp,’ ‘cost,’ ‘mosque,’ ‘fosp,’ ‘fost,’ ‘posk’ 34% 35% 60% 66% ‘harp,’ ‘part,’ ‘shark,’ ‘farp,’ ‘gart,’ ‘fark’ all of [] 49% 50% 58% 59% low F3 of [] 30% 30% 46% 47% p-values [s] vs. [] 0.000 0.000 0.552 0.059 [s] vs. F3 [] 0.091 0.011 0.002 0.000 If only the portion of the duration of [] devoted to the lowered F3 is considered, [s] is then significantly longer than [] (with the exception of the casual rhyme where [s] is still longer, just not significantly). 5.4.8. Nasals vs. Lateral Approximant [l] Comparisons were made of the tokens with [n] or [l] followed by a word-final [t]. Since [l] and [n] are ranked similarly on the sonority hierarchy and have similar cues, it was difficult to predict which would have a longer proportional duration under the hypothesis that consonants higher on the hierarchy would have a longer proportional duration. As is shown in Table 5-34, there was no significant difference between the duration of [n] and [l] in either speaking rate preceding [t].

PAGE 100

72 Table 5-34: Mean proportional duration of nasals and [l] preceding [t] Tokens Casual Fast Casual Rhyme Rhyme Coda Fast Coda ‘font,’ ‘pont’ 22% 27% 45% 50% ‘fault,’ ‘pault’ 28% 27% 45% 47% p-values [n] vs. [l] 0.11 0.71 0.91 0.93 [l] did, though, have a longer proportional duration in casual speech, but not significantly. Word-finally, [l] and nasals both occurred after []. Comparing just the alveolar nasal [n] and [l], [l] is a significantly longer portion of the rhyme in both speaking rates, but not a significantly longer portion of the coda (Table 5-35). Table 5-35: [n] and [l] mean average comparisons following [] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘barn,’ ‘parn’ 28% 28% 36% 35% ‘Carl,’ ‘parn’ 37% 40% 44% 46% p-values [n] vs. [l] 0.046 0.035 0.198 0.072 The same pattern can be found when comparing the bilabial nasal [m] with [l] except that in this case [l] in the casual rhyme is not significantly longer than [m] in the casual rhyme (Table 5-36). Table 5-36: [m] and [l] mean average comparisons following [] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘farm,’ ‘tarm’ 29% 31% 39% 44% ‘Carl,’ ‘parl’ 37% 40% 44% 46% p-values [m] vs. [l] 0.063 0.040 0.287 0.850 The nasals and [l] which are ranked closely on the sonority hierarchy varied as to which was greater in different word positions. [l], which is ranked higher was longer word-finally. Preceding [t], though, whether [l] or [n] was longer varied with speaking rate.

PAGE 101

73 5.4.9. Nasals vs. Rhotic Approximant [] Nasals and [] were both found preceding voiceless stops. The data in Table 5-37 show that the duration of [] and the nasals are significantly different in both speaking rates. [] is a significantly longer portion of the coda and rhyme in both speaking rates. Although, this [] includes movement of the F3 transition, this difference cannot be explained simply by attention to this early cue, since the duration of the nasal was measured at the beginning of the earliest sign of anti-formants. Table 5-37: Nasal and [] mean average comparison preceding voiceless stop Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘romp,’ ‘font,’ ‘fomp,’ ‘pont’ 21% 27% 42% 52% ‘harp,’ ‘part,’ ‘farp,’ ‘gart’ all of [] 49% 49% 57% 58% low F3 of [] 31% 30% 46% 46% p-values Nasals vs. [] 0.000 0.000 0.016 0.029 Nasals vs. [] F3 0.008 0.184 0.443 0.306 Considering, though, just the portion of [] after F3 levels off, the portion of the rhyme devoted to [] is still longer than the portion devoted to the nasal, but not significantly in fast speech. Considering the portion devoted to the coda, there is no significant difference between the nasals and []. In addition, in fast speech a longer portion of the coda is devoted to the nasal. When comparing [s] and [], [] was significantly longer when the F3 transition was included, but when it was not [s] was significantly longer than []. In

PAGE 102

74 this sense [s] was different from the nasals. As would be predicted according to the hypothesis, the duration of [] was longer then the duration of the nasals. 5.4.10. Lateral Approximant [l] vs. Rhotic Approximant [] Both [l] and [] occurred before word-final stops and fricatives in the data collected. Preceding obstruents, considering the entire duration of [] (from the lowering of F3 through maintaining a low F3), [] is significantly longer at all speaking rates than [l] (Table 5-38). Table 5-38: Mean proportional duration of [l], [], and low F3 of [] preceding obstruent Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘fault,’ ‘bald,’ ‘pault,’ ‘fald,’ ‘false,’ ‘Walsh,’ ‘palse,’ ‘palsh’ 27% 27% 45% 44% ‘part,’ ‘card,’ ‘gart,’ ‘fard,’ ‘farce,’ ‘harsh,’ ‘barce,’ ‘parsh’ all of [] 48% 50% 57% 60% low F3 of [] 28% 30% 45% 49% p-values [l] vs. [] 0.000 0.000 0.000 0.002 [l] vs F3 0.358 0.266 0.820 0.335 If only the duration of the lowered F3 is considered, though, there is no significant difference between the duration of [l] and []. This is also true when the duration of [l] and [] are compared preceding just the voiceless stop [t] (Table 5-39)

PAGE 103

75 Table 5-39: Mean proportional duration of [l], [], and low F3 of [] preceding [t] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘fault,’ ‘pault’ 28% 27% 45% 47% ‘part,’ ‘gart’ all of [] 53% 53% 62% 64% low F3 of [] 34% 34% 52% 54% p-values [l] vs. [] 0.000 0.000 0.010 0.035 [l] vs. F3 0.133 0.160 0.227 0.369 Preceding [d], the proportional duration of [] was longer than that of the proportional duration of [l] significantly in both speaking rates (Table 5-40). However, when only the duration of low F3 was considered, the duration of [l] and [] were nearly identical at both speaking rates with no significant differences. Table 5-40: Mean proportional duration of [l], [], and low F3 of [] preceding [d] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘bald,’ ‘fald’ 31% 34% 62% 63% ‘card,’ ‘fard’ all of [] 57% 60% 72% 76% low F3 of [] 33% 35% 59% 68% p-values [l] vs. [] 0.000 0.001 0.028 0.046 [l] vs. F3 0.639 0.923 0.415 0.490 Similarly, when comparing the duration of [l] and [] preceding strident fricatives, [] was significantly longer than [l] when the duration of [] including the third formant transition (Table 5-41).

PAGE 104

76 Table 5-41: Mean proportional duration of [l], [], and low F3 of [] preceding strident fricatives Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda ‘false,’ ‘Walsh,’ ‘palse,’ ‘palsh’ 24% 22% 33% 32% ‘farce,’ ‘harsh,’ ‘barce,’ ‘parsh’ all of [] 40% 43% 46% 49% low F3 of [] 23% 26% 33% 36% p-values [l] vs. [] 0.000 0.000 0.001 0.000 [l] vs. F3 0.570 0.041 0.739 0.024 However, when only the low F3 of [] was considered, the duration of [l] and [] were nearly identical. Comparing [l] and [], [] had a longer proportional duration as would be expected since [] is ranked higher on the sonority hierarchy. However, limiting the cues that are included in [] duration, [l] and [] are nearly identical in proportional duration with no significant differences between them. In this sense, [l] behaved similarly to the nasals, but differently from [s], which was significantly longer than [] when the F3 transitions were not included. 5.5. Ranking of Consonants in Environments The following section shows how the consonants are ranked in terms of their duration in parallel environments. Attention is given to how this ranking relates to the sonority hierarchy. In general, it was found that [] and the strident fricatives ranked higher than the other consonants in parallel environments.

PAGE 105

77 5.5.1. Preceding Voiceless Stops In the data examined for this study, voiceless stops were preceded by [p, s, m, n, l, ]. The ranking of these consonants in terms of their duration before voiceless stops is shown in Table 5-42. Table 5-42: Ranking of consonants by proportion of rhyme and coda preceding voiceless stops Casual and fast speech [] >> [s] >> [m, n], [l], [p] The ranking above is based upon [] including the F3 transition. If only the duration of the low F3 is included, then [s] ranks higher than [] (Table 5-43). Table 5-43: Ranking of consonants by proportion of rhyme and coda with low F3 of [] preceding voiceless stops Casual and fast speech [] = low F3 [s] >> [] >> [m, n], [l], [p] The rankings in both Table 5-42 and Table 5-43 are similar to the sonority hierarchy in that the duration of [] is greater than the durations of the stop, nasals and [l]. It is unlike the hierarchy in that there is no difference in the ranking of the nasals, [l], and [p]. Also, [s] is ranked high in both hierarchies so that if duration were correlated with duration, than in this position [s] would be most sonorous. 5.5.2. Preceding Fricatives Preceding fricatives [] was a longer proportion of both the rhyme and coda than [l]. This was true for both speaking rates. In this case, the consonant that ranks higher on the sonority hierarchy has the longer duration.

PAGE 106

78 5.5.3. Following [l] Only strident fricatives and [t] followed [l] in the data collected. The ranking of these consonants in this position is shown in Table 5-44. Table 5-44: Ranking of consonants by proportion of rhyme and coda following [l] Casual and fast speech [s] >> [t] In this case [s] was significantly longer than [t] following [l]. 5.5.4. Following [] Following [], all of the consonants considered in this study were found. Their ranking is found in Table 5-45. Table 5-45: Ranking of consonants by proportion of rhyme and coda following [] Casual and fast speech [s, ] >> [l] >> [t] >> [m, n] [f]>> [b, d] This ranking is similar to the sonority hierarchy in that the duration of [l] is greater than the nasals and the stops. The ranking of the voiceless stops and fricatives, though, breaks the hierarchy. The strident fricatives are not so problematic since some hierarchies rank [s] has the highest member of the hierarchy. That [m] is not greater than the non-strident fricative [f] and ranked lower than the voiceless stops is more problematic if sonority is correlated with consonant length. However, this is not unexpected since duration is a cue for voicing in word-final obstruents in English where voiceless obstruents are significantly longer than voiced obstruents.

PAGE 107

CHAPTER 6 ENGLISH ACOUSTIC ANALYSIS OF RELATIVE RMS AMPLITUDE 6.1. Introduction This chapter describes the findings of the comparison of each consonant’s relative root mean square (rms) dependent upon its speaking rate. Then, it compares the phones with different manners of articulation in parallel environments. Next, it compares the amplitudes of adjacent consonant. Finally, this chapter shows the rankings of the consonants dependent upon their relative rms in parallel environments and relates these rankings to the sonority hierarchy. Relative rms amplitude was taken as the phone’s rms amplitude (sound pressure expressed in Pascal) in proportion to the rms amplitude of [e] in the word ‘say’ of the carrier phrase, “Please say _____ for me” It was the expectation of this study that a consonant’s relative rms would be related to its position in the sonority hierarchy. That is, consonants that rank lower on the sonority hierarchy would be more likely to decrease their relative rms as speaking rate increased. In addition, consonants that rank higher would have a higher relative rms than consonants ranked lower on the sonority hierarchy in parallel word positions. The data for all five speakers were collapsed and considered together for the analysis. In addition, the real and nonsense words were considered together. Comparisons were made of the relative rms of the consonants in the real and nonsense tokens in the same environments to ensure that there were no significant differences (p .05) as a result of the use of nonsense words. No such differences were found. 79

PAGE 108

80 As with duration, it was found that [] and the strident fricatives had greater relative rms than the other consonants in parallel environments. For rms, [l] tended to have a greater relative rms than the stops and non-strident fricatives. The nasals, though, did not tend to have greater relative rms than the other consonants in parallel environments. 6.2. Speaking Rate As described in Chapter 5, the speakers did increase their rate of speech according to the instructions that were given to them during data collection. Consonants were compared at each speaking rate in the same environments. Relative rms was compared using a t-test where the null hypothesis was no difference in the relative rms of the consonant despite a change in speaking rate. Significant was considered a p-value of less than or equal to .05. However, stops do significantly decrease their relative rms word-finally (Table 6-1). Table 6-1: Mean relative rms of word-final voiceless stops and stop bursts Casual Fast p-values Gap and burst 0.270 0.220 0.010 Burst 0.339 0.281 0.056 This most likely reflects the weakening or deletion of the stop burst as speaking rate increased. Relative rms of the burst decreased as speaking rate increased, but not significantly. The relative rms of the sonorant sounds and the fricatives, however, was not significantly affected by increasing speaking rate from casual to fast speech. This agrees with the hypothesis of this study that obstruents would be more likely to decrease

PAGE 109

81 amplitude as speaking rate increased. What may seem surprising, though, is that the stop consonant [p] did not also significantly decrease its relative rms preceding [t]. However, this may be due to the already low relative rms of [p] in this position as will be seen later in this chapter. 6.3. Comparisons by Manner of Articulation It was expected in this research that phones which rank higher on the sonority hierarchy would have a higher relative rms than phones ranked lower on the sonority hierarchy when they are compared in parallel environments. When possible consonants were compared with consonants of the same place of articulation, i.e. [l] was compared with [t] and [d], but not [b]. Relative rms was compared using a t-test where the null hypothesis was no difference in the relative rms of the consonants appearing in the same environment. Significant was considered a p-value of less than or equal to .05. While neither the token’s speaking rate resulted in many significant differences in the relative rms of the consonants, comparing phones of different manners of articulation in the parallel positions did result in several significant differences. For instance, [] and [s] tended to have a significantly higher relative rms than other phones in the same position. The voiceless stops and stop bursts tended to have a significantly lower relative rms than other consonants in the same positions. 6.3.1. Stops vs. Fricatives Stops and fricatives had several points of comparison both word-finally and preceding word-final stops. In the latter position, [p] and [s] appeared preceding [t] in the tokens. Stops and fricatives are also found in comparable positions word-finally

PAGE 110

82 following the sonorant consonants [l], and []. The expectation was that in all of the environments, the fricatives would have a significantly higher relative rms than the stops. As is shown in Table 6-2, in both speaking rates, [s] did have significantly greater amplitude than [p]. Table 6-2: Mean relative rms of [p] and [s] preceding [t] Tokens Casual Fast ‘hopped,’ ‘fopt’ 0.140 0.143 ‘cost,’ ‘fost’ 1.109 0.839 p-values [p] vs. [s] 0.010 0.004 In addition, comparing stops and fricatives following sonorants reveals that fricatives have significantly greater amplitude in this position than do stops (Table 6-3). Table 6-3: Mean relative rms of stops and fricative following sonorants Tokens Casual Fast ‘fault,’ ‘pault,’ ‘harp,’ ‘part,’ ‘shark,’ ‘farp,’ ‘gart,’ ‘farp’ 0.250 0.202 ‘false,’ ‘palse,’ ‘scarf,’ ‘farce,’ ‘harsh,’ ‘parf,’ ‘barce,’ ‘parsh’ 1.127 0.973 p-values [p, t] vs. [f, s] 0.000 0.000 If the environments are broken down to following [l] and [] separately, the same is found (Table 6-4 and Table 6-5). Table 6-4: Mean relative rms of stops and fricatives following [] Tokens Casual Fast ‘harp,’ ‘part,’ ‘shark,’ ‘farp,’ ‘gart,’ ‘farp’ 0.264 0.215 ‘scarf,’ ‘farce,’ ‘harsh,’ ‘parf,’ ‘barce,’ ‘parsh’ 1.049 0.890 p-values [p, t, k] vs. [f, s, ] 0.000 0.000

PAGE 111

83 Table 6-5: Mean relative rms of stops and fricatives following [l] Tokens Casual Fast ‘fault,’ ‘pault’ 0.209 0.167 ‘false,’ ‘palse’ 1.504 1.349 p-values [t] vs. [s] 0.043 0.033 The values above though represent the relative rms of the fricatives compared to the relative rms of the entire duration of the stop, i.e. both the gap and the burst. Would the same findings remain true if the fricatives were compared to just the relative rms of the stop burst? Measurements were taken of just the stop bursts. Comparing the bursts with the fricatives still resulted in the fricatives having higher relative rms than the stops despite that the relative rms of just the burst is higher than the entire stop (Table 6-6). Table 6-6: Mean relative rms of stop burst and fricatives following sonorants Tokens Casual Fast ‘fault,’ ‘pault,’ ‘harp,’ ‘part,’ ‘shark,’ ‘farp,’ ‘gart,’ ‘farp’ 0.349 0.285 ‘false,’ ‘palse,’ ‘scarf,’ ‘farce,’ ‘harsh,’ ‘parf,’ ‘barce,’ ‘parsh’ 1.127 0.973 p-values [p, t, k] vs. [f, s, ] 0.000 0.000 Therefore, for the comparison of stops and fricatives, the hypothesis proved to be true that the sound that ranks higher on the sonority hierarchy would have a higher relative rms in all environments. That assumes a hierarchy in which both strident and non-strident fricatives are separated from stops. 6.3.2. Stops vs. Nasals Both stops and nasals are found word-finally following []. The stops compared to the nasals were the voiced and voiceless stops with the same places of articulation, bilabial and alveolar. Since nasals rank higher on the sonority hierarchy, it would be expected that nasals should have a higher relative rms than the stops. For the data

PAGE 112

84 collected, regardless of whether the nasals are compared to the voiced or voiceless stops, the nasals have significantly greater relative rms amplitude than the stops (Table 6-7). Table 6-7: Mean comparisons of voiceless stops, voiced stops and nasals following [] Tokens Casual Fast ‘harp,’ ‘part,’ ‘farp,’ ‘gart’ 0.263 0.178 ‘carb,’ ‘card,’ ‘farb,’ ‘fard’ 0.295 0.225 ‘farm,’ ‘barn,’ ‘tarm,’ ‘parn’ 1.084 0.798 p-values Nasal vs. voiceless stops 0.028 0.001 Nasal vs. voiced stop 0.020 0.001 Breaking the consonants down by place, the same pattern is found, but in these cases not significantly, possibly because there are less tokens to consider (Table 6-8). Table 6-8: Mean relative rms of [d] and [n] following [] Tokens Casual Fast ‘carb,’ ‘farb’ 0.254 0.195 ‘card,’ ‘fard’ 0.335 0.253 ‘farm,’ ‘tarm’ 1.261 0.753 ‘barn,’ ‘parn’ 0.926 0.844 p-values [b] vs. [m] 0.122 0.021 [d] vs. [n] 0.095 0.030 Comparing the nasals to just the relative rms of the stop burst, the mean relative rms is still significantly greater for the nasal (Table 6-9). Table 6-9: Mean relative rms of stop bursts and nasals following [] Tokens Casual Fast ‘harp,’ ‘part,’ ‘farp,’ ‘gart’ 0.352 0.268 ‘carb,’ ‘card,’ ‘farb,’ ‘fard’ 0.332 0.252 ‘farm,’ ‘barn,’ ‘tarm,’ ‘parn’ 1.084 0.798 p-values Voiceless burst vs. nasal 0.041 0.004 Voiced burst vs. nasal 0.031 0.002 In the comparison of stops and nasals, the hypothesis again proved true that the consonant ranked higher on the sonority hierarchy would have a higher relative rms. In

PAGE 113

85 chapter 5, it was shown that the voiceless stops had a longer duration than the nasals in this position, so the relationship between voiceless stops and the nasals were the opposite in duration than they were in rms amplitude. 6.3.3. Stops vs. Lateral Approximant [l] Like, the fricatives, stops and [l] occurred in several of the same positions both word-finally after [] and preceding word-final [t]. Also, like the fricatives, it was expected that [l] would have a higher relative rms than the stops. For comparison, the tokens with word-final stops used were those with [t] and [d] since they agree in place of articulation with [l]. Indeed, in all cases [l] did have a greater relative rms amplitude than [t] and [d] (Table 6-10). Table 6-10: Mean relative rms of [t], [d], stop bursts and [l] following [] Tokens Casual Fast ‘part,’ ‘gart’ [t] gap and burst 0.253 0.226 [t] burst 0.247 0.291 ‘card,’ ‘fard’ [d] gap and burst 0.335 0.253 [d] burst 0.489 0.291 ‘Carl,’ ‘parl’ 0.756 0.915 p-values [l] vs. [t] 0.012 0.003 [l] vs. [t] burst 0.005 0.008 [l] vs. [d] 0.034 0.002 [l] vs. [d] burst 0.158 0.005 This remained true when only the amplitude of the stop burst was considered. However, in casual speech [l] was not significantly greater than the stop burst of [d]. This can be explained by the fact that NES2 and NES5 had a high rms for the stop bursts in the ‘fard’ at 1.005 and 1.099 respectively. It is interesting though that the two speakers both chose to articulate this word on that way.

PAGE 114

86 Similarly, preceding [t], [l] had higher relative rms amplitude than the stop, [p] (Table 6-11). Table 6-11: Mean relative rms of [p] and [l] preceding [t] Tokens Casual Fast ‘hopped,’ ‘fopt’ 0.140 0.143 ‘fault,’ ‘pault’ 0.887 0.730 p-values 0.019 0.001 Again, the stops had a lower relative rms than the more sonorous consonant, in this case [l]. When comparing [l] with the stops, the hypothesis again proved to be true that the consonant that ranked higher on the sonority hierarchy had greater relative rms. This was true whether comparing [l] to the voiced or voiceless stops. In addition, this difference was always significant except when comparing the casual speech productions of the word-final stop burst of [d] and [l]. 6.3.4. Stops vs. Rhotic Approximant [] The only environment where both [] and a stop appeared was preceding a word-final [t]. Just as with the other sonorants, [] had a significantly higher relative rms than [p] in this position (Table 6-12). Table 6-12: Mean relative rms of [p] and [] preceding [t] Tokens Casual Fast ‘hopped,’ ‘fopt’ 0.140 0.143 ‘part,’ ‘gart’ 1.721 1.599 p-values [p] vs. [] 0.001 0.000 Comparing the stops to just the lowest point of F3 still results in [] having a significantly greater relative rms than the stops (Table 6-13).

PAGE 115

87 Table 6-13: Mean relative [p] and []’s lowest F3 preceding stops Tokens Casual Fast ‘hopped,’ ‘fopt’ 0.140 0.143 ‘part,’ ‘gart’ 1.441 1.263 p-values [p] vs. [] 0.001 .000 Again the finding in this subsection are not surprising since [] is ranked higher than [p] in the sonority hierarchy. Again, the hypothesis is proved true that consonants that rank higher on the sonority hierarchy would have greater relative rms than consonants ranked lower on the hierarchy in parallel environments. 6.3.5. Fricatives vs. Nasals Fricatives and nasals appeared in two of the same environments: word-finally following [] and preceding a word-final voiceless stop. It was expected that in all of these environments the nasals should have a higher relative rms than the fricatives. The first point of comparison was the fricatives and the nasals with the same places of articulation (labial and alveolar) following [] (Table 6-14). Table 6-14: Mean relative rms of labial and alveolar fricatives and nasals following [] Tokens Casual Fast ‘scarf,’ ‘farce,’ ‘parf,’ ‘barce’ 0.974 0.713 ‘barn,’ ‘farm,’ ‘parn,’ ‘tarm’ 1.084 0.798 p-values [f, s] vs. [m, n] 0.793 0.453 In this position, the average relative rms of the nasals is greater than that of the fricatives, but not significantly. While NES5 in all cases had a higher relative rms for the nasals than the fricatives (Table 6-15), the other speakers varied as to which consonants had a higher rms in this position. Unlike comparing the stops with nasals, there is not a clear distinction between the relative rms of fricatives and nasals with the sonorant having greater relative rms than the obstruent.

PAGE 116

88 Table 6-15: NES5 relative rms of word-final nasals and fricatives Token Casual Fast farce 2.632 1.465 scarf 0.619 0.362 barce 2.536 2.197 parf 0.765 0.637 Mean 1.638 1.165 barn 2.660 2.356 farm 5.504 1.966 parn 3.955 2.778 tarm 3.649 2.228 Mean 3.942 2.332 p-values 0.113 0.020 Considering just the alveolars [s] and [n], [s] on average had greater relative rms, but the difference in not significant (Table 6-16). Table 6-16: Mean relative rms of [s] and [n] following [] Tokens Casual Fast ‘farce,’ ‘barce’ 1.618 1.079 ‘barn,’ ‘parn’ 0.926 0.844 p-values [s] vs. [n] 0.233 0.493 When comparing the proportional duration in chapter 5, it was found that [n] was significantly longer than [s] in this position. Looking at the labial [f] and [m], the relative rms of the nasal is now longer than the fricative (Table 6-17). This difference by place is not surprising since now the strident and non-strident [s] and [f] are separated into two different data sets. Table 6-17: Mean relative rms of [f] and [m] following [] Tokens Casual Fast ‘scarf,’ ‘parf’ 0.395 0.347 ‘farm,’ ‘tarm’ 1.261 0.753 p-values [f] vs. [m] 0.202 0.099 Remember that when considering duration though, [f] had a significantly longer proportional duration than [m] in this same position. For one last comparison in this

PAGE 117

89 environment, all of the tokens with a word-final strident fricative, i.e. [s] and [] were compared with the nasals to determine if this would reveal any significant differences in rms following [] (Table 6-18). Table 6-18: Mean relative rms of nasals and strident fricatives following [] Tokens Casual Fast ‘farce,’ ‘harsh,’ ‘barce,’ ‘parsh’ 1.394 1.162 ‘barn,’ ‘farm,’ ‘parn,’ ‘tarm’ 1.084 0.798 p-values [s, ] vs. [m, n] 0.438 0.031 Now the rms of the strident fricatives is shown to be greater than that of the nasals. In addition, significant difference is shown in the fast speech tokens. Taking NES5, who as described above had high relative rms on his nasals, out of the data set shows that in all cases, the stridents had a significantly higher relative rms than the nasals (Table 6-19). Table 6-19: Mean relative rms of strident fricatives and nasals following [] excluding NES5 Tokens Casual Fast ‘farce,’ ‘harsh,’ ‘barce,’ ‘parsh’ 1.144 0.915 ‘barn,’ ‘farm,’ ‘parn,’ ‘tarm’ 0.322 0.415 p-values [s, ] vs. [m, n] 0.017 0.008 Comparisons were also made of the nasals and [s] preceding word-final stops. For this analysis those tokens with the same word-final stops were chosen (romp, font, fomp, pont, wasp, cost, fosp, fost) Preceding voiceless stops, [s] had higher relative rms in both speaking rates, although not significantly (Table 6-20). Table 6-20: Mean relative rms of [s] and nasals preceding voiceless stops Tokens Casual Fast ‘wasp,’ ‘cost,’ ‘fosp,’ ‘fost’ 1.065 0.906 ‘romp,’ ‘font,’ ‘fomp,’ ‘pont’ 0.820 0.739 p-values [s] vs. [m, n] 0.317 0.305

PAGE 118

90 Unlike the fricative and nasals in word-final position, removing NES5 from the data does not reveal any statistical difference in relative rms of these consonants in this environment. In addition breaking up the tokens by the final stop’s place of articulation does not show any significant difference in rms. Comparing the nasals and fricatives did not result in the consonant that ranks higher on the sonority hierarchy having significantly greater relative rms in parallel environments. On the contrary, the strident fricatives tended to have greater relative rms than the nasals, both preceding stops and following []. 6.3.6. Fricatives vs. Lateral Approximant [l] When comparing the fricatives with [l], the phone [s] was initially used as a point of comparison, since both are alveolar. Both [l] and [s] preceded word-final [t] and followed []. Comparing [l] and [s] following [] revealed no significant difference in their relative rms (Table 6-21), but on average [s] has a greater rms than [l] did. Table 6-21: [s] and [l] mean average comparisons following [] Tokens Casual Fast ‘farce,’ ‘barce’ 1.618 1.079 ‘Carl,’ ‘parl’ 0.756 0.915 p-values [s] vs. [l] 0.139 0.577 Similarly, when considering duration, [s] had on average a longer proportional duration than did [l], but the difference was not significant. Comparing [l] to the non-strident fricative [f] in ‘scarf’ and ‘parf’ reveals that [l] had significantly greater amplitude than [f] (Table 6-22).

PAGE 119

91 Table 6-22: Mean relative rms [f] and [l] following [] Tokens Casual Fast ‘scarf,’ ‘parf’ 0.395 0.347 ‘Carl,’ ‘parl’ 0.756 0.915 p-values [f] vs. [l] 0.109 0.010 Therefore, when comparing [l] to the non-strident fricative, [l] had a greater relative than [f], but when comparing [l] to the strident fricative, [s] had a greater relative rms. The other position where both [s] and [l] appeared was preceding [t]. While [s] on average does have a higher relative rms amplitude preceding [t] in casual and fast speech, there is no significant difference between the relative rms of [s] (Casual 1.109, Fast .839) and [l] (Casual .887, Fast .730) in this position (Table 6-23). Table 6-23: Mean relative rms of [s] and [l] preceding [t] Tokens Casual Fast ‘cost,’ ‘fost’ 1.109 0.839 ‘fault,’ ‘pault’ 0.887 0.730 p-values [s] vs. [l] 0.618 0.793 In addition, the relative rms is not consistently higher for either [s] or [l] when comparing both speaking rates. When comparing [l] to the fricatives, [l] had a greater relative rms than the non-strident fricatives, but [s] had a greater relative rms than [l]. These differences, though, were only significant when comparing the fast speech productions of word-final [l] and [f]. 6.3.7. Fricatives vs. Rhotic Approximant [] The only fricative that occurred in the same position as [] was [s]. Both appeared before word-final voiceless stops. As is shown in Table 6-24, [] had a greater relative rms preceding voiceless stops than did [s].

PAGE 120

92 Table 6-24: Mean relative rms of [s] and [] preceding voiceless stops Tokens Casual Fast ‘wasp,’ ‘cost,’ ‘mosque,’ ‘fosp,’ ‘fost,’ ‘posk’ 0.985 0.849 ‘harp,’ ‘part,’ ‘shark,’ ‘farp,’ ‘gart,’ ‘fark’ 1.630 1.650 p-values [s] vs. [] 0.027 0.000 The same results were found when comparing the duration of [] and [s] where [] had a significantly longer proportional duration of the rhyme than [s]. Comparing just the point at which [] reaches its lowest F3 still results in greater relative rms for [] than [s], however this difference is only significant in fast speech (Table 6-25). Looking at the individual speakers casual speech data reveals that the lack of significance is due to NES4 and NES5. While the other speakers tend to have higher rms of low F3 than [s] relative rms of [s] in casual speech, NES4 and NES5 do not. Table 6-25: Mean relative rms of [s] and []’s low F3 preceding voiceless stops Tokens Casual Fast ‘wasp,’ ‘cost,’ ‘mosque,’ ‘fosp,’ ‘fost,’ ‘posk’ 0.985 0.849 ‘harp,’ ‘part,’ ‘shark,’ ‘farp,’ ‘gart,’ ‘fark’ 1.262 1.284 p-values [s] vs. [] 0.394 0.006 When comparing the duration of [s] and just the low F3 of [], [s] was a significantly longer portion of both the rhyme and coda. Comparing [] and [s] preceding word-final stops, the sonorant did have greater relative rms than the obstruent as would be predicted by the hypothesis. The difference is significant except when comparing [s] to the low F3 of [] in casual speech. In this sense,

PAGE 121

93 [] is distinct from the other sonorants, which did not have significantly greater relative rms than [s]. 6.3.8. Nasals vs. Lateral Approximant [l] Both [l] and the nasals occurred word-finally following [] and preceding [t]. Since [n] and [l] are both sonorant consonants occurring next to each other on the sonority hierarchy, it was difficult to hypothesize with a strong sense of confidence which would have a higher rms in parallel environments. However, with a strict reading of Clements’ sonority hierarchy (1988, 1990), [l] would have a higher relative rms if the hypothesis of this dissertation were correct that consonants that are more sonorous should have a higher relative rms. Preceding [t], [n] and [l] were not significantly different in amplitude. [n] had greater relative rms amplitude than [l] in casual and fast speech (Table 6-26). Table 6-26: Mean relative rms of [n] and [l] preceding voiceless stop Tokens Casual Fast ‘font,’ ‘pont’ 0.955 0.932 ‘fault,’ ‘pault’ 0.887 0.730 p-values [n] vs. [l] 0.864 0.679 As well, when comparing the proportional duration of [l] and [m] in this position there was no significant difference, but on average [l] had a greater relative rms than [m] in this position (Table 6-27). Table 6-27: Mean relative rms of [m] and [l] preceding voiceless stops Tokens Casual Fast ‘romp,’ ‘fomp’ 0.669 0.546 ‘fault,’ ‘pault’ 0.887 0.730 p-values [m] vs. [l] 0.532 0.240

PAGE 122

94 Following [], there were also was no significant difference in the relative rms of [n] and [l] (Table 6-28). Table 6-28: Mean relative rms of [n] and [l] following [] Tokens Casual Fast ‘barn,’ ‘parn’ 0.926 0.844 ‘Carl,’ ‘parl’ 0.756 0.915 p-values [v] vs. [l] 0.674 0.977 When comparing the duration following [], there was a significant difference between [l] and [n]. [l] was a significantly longer portion of the rhyme than was []. Comparing [l] to the nasal [m] instead of [n] shows the same results (Table 6-29). Table 6-29: Mean relative rms of [m] and [l] following [] Tokens Casual Fast ‘farm,’ ‘tarm’ 1.261 0.753 ‘Carl,’ ‘parl’ 0.756 0.915 p-values [m] vs. [l] 0.316 0.697 However, comparing the proportional of [l] and [m], [l] was significantly longer than [m] in this position (Table 5-36). So comparing nasals and [l] neither had a significantly higher relative rms in any of their parallel environment. This is not surprising since they are both sonorant consonants with similar cues (anti-formants). Although there were no significant difference in relative rms, there were significant differences in the differences of their durations as was seen in chapter 5. 6.3.9. Nasals vs. Rhotic Approximant [] Both [] and the nasals occurred preceding voiceless stops. For comparisons, only the [] tokens with word-final [p] and [t] were used so that the environments were

PAGE 123

95 consistent. The average relative rms was greater for [] than it was for the nasal in both speaking rates (Table 6-30). Table 6-30: Mean relative rms of nasals and [] preceding voiceless stop Tokens Casual Fast ‘romp,’ ‘font,’ ‘fomp,’ ‘pont’ 0.820 0.739 ‘harp,’ ‘part,’ ‘farp,’ ‘gart’ 1.722 1.674 p-values [m, n] vs. [] 0.008 0.004 Breaking up the tokens by those with the same place of articulation for the word-final stop reveals the same findings. Comparing just the duration of the lowest portion of F3 and the nasal gives the same results with [] having greater relative rms amplitude than the nasals (Table 6-31). In casual speech, though, this difference is not significantly different. Table 6-31: Mean relative rms of nasals and []’s lowest F3 preceding voiceless stop Tokens Casual Fast ‘romp,’ ‘font,’ ‘fomp,’ ‘pont’ 0.820 0.739 ‘harp,’ ‘part,’ ‘farp,’ ‘gart’ 1.351 1.279 p-values [m, n] vs. [] 0.062 0.024 Comparing nasals and [], the consonant that ranks higher on the sonority hierarchy did have a higher relative rms. This held true whether considering the entire duration of [] or just its primary cue the duration of the low F3. 6.3.10. Lateral Approximant [l] vs. Rhotic Approximant [] Both [l] and [] occurred before word-final obstruents [t, d, s, and]. In all of the environments, [] had higher relative rms amplitude than [l] (Table 6-32). The difference was significant preceding the obstruents, as well as when just the fricatives (all of which

PAGE 124

96 were strident) were considered. The difference however was not always significant preceding just the voiced or voiceless stops. Table 6-32: Mean relative rms of [] and [l] preceding obstruents Tokens Casual Fast Preceding Obstruents ‘fault,’ ‘bald,’ ‘pault,’ ‘fald,’ ‘false,’ ‘Walsh,’ ‘palse,’ ‘palsh’ 0.950 0.903 ‘part,’ ‘card,’ ‘gart,’ ‘fard,’ ‘farce,’ ‘harsh,’ ‘barce,’ ‘parsh’ 1.674 1.647 p-values 0.000 0.000 Preceding Voiceless Obstruents ‘fault,’ ‘pault,’ ‘false,’ ‘Walsh,’ ‘palse,’ ‘palsh’ 1.008 0.830 ‘part,’ ‘gart,’ ‘farce,’ ‘harsh,’ ‘barce,’ ‘parsh’ 1.707 1.613 p-values 0.000 0.000 Preceding Voiceless Strident Fricatives ‘false,’ ‘Walsh,’ ‘palse,’ ‘palsh’ 1.083 0.893 ‘farce,’ ‘harsh,’ ‘barce,’ ‘parsh’ 1.700 1.620 p-values 0.005 0.008 Preceding Voiceless Stops ‘fault,’ ‘pault’ 0.887 0.730 ‘part,’ ‘gart’ 1.721 1.599 p-values 0.043 0.000 Preceding Voiced Stops ‘bald,’ ‘fald’ 0.914 1.099 ‘card,’ ‘fard’ 1.490 1.485 p-values 0.016 0.123 6.4. Amplitude Compared to Adjacent Consonant This section compares the relative rms of consonants that are adjacent to one another. According to the hypothesis, it is expected that of two adjacent consonants, the consonant higher on the sonority hierarchy would have a greater relative rms. Of course, this could also be accounted by the individual phone’s placement in the token. That is, word-final consonants may be less sonorous simply because they are at the end of a word.

PAGE 125

97 6.4.1. Word-final Consonants Compared to Adjacent [] In most cases, [] had significantly higher relative rms than the consonant that it was adjacent to. This is consistent with the hypothesis that in consonant cluster the more sonorous consonant would have greater relative rms. The only consonants that had greater relative rms than the [] they followed were strident fricatives in ‘farce,’ ‘harsh,’ ‘barce,’ and ‘parsh.’ In these cases, the stridents had on average a higher relative rms, although not significantly. This reflects that individual repetitions of tokens varied as to whether the relative rms of [] or the strident fricative was greater. 6.4.2. Word-final Consonants Compared to Adjacent [l] Compared to the word-final-stop [t], [l] had a significantly higher relative rms than [t] in the tokens ‘fault’ and ‘pault.’ This was true when comparing both the entire stop and just the burst (Table 6-33). Table 6-33: Mean rms of adjacent stop and [l] Casual Fast Stop 0.209 0.167 Stop burst 0.219 0.198 [l] 0.887 0.730 p-values stop vs. [l] 0.028 0.001 stop burst vs. [l] 0.034 0.001 Comparisons of the word-final fricatives and [l] were not as consistent. Speakers varied as to whether the fricative or [l] had a greater relative rms. The result is that while overall the fricatives had a higher average relative rms, there is no significant difference between the two. Breaking down the tokens by the fricative’s place of articulation also gives the same results (Table 6-34).

PAGE 126

98 Table 6-34: Mean relative rms of adjacent [l] and strident fricatives Casual Fast Strident fricative 1.251 1.068 [l] 1.083 0.893 p-values [s, ] vs. [l] 0.422 0.348 6.4.3. Word-final Consonants Compared to Adjacent Nasals Nasals that appeared adjacent to word-final stops had a significantly greater relative rms than both the stops and the stop bursts (Table 6-35). Table 6-35: Mean average relative rms of adjacent stops and nasals Casual Fast Stop 0.214 0.179 Stop burst 0.245 0.220 Nasal 0.820 0.739 p-values stop vs. [nasal] 0.001 0.001 stop burst vs. nasal 0.001 0.001 This is in accord with the hypothesis that more sonorant consonants would have a higher rms than consonants that they are adjacent to. 6.4.4. Word-final Consonants Compared to Adjacent [s] [s] occurred adjacent to word-final voiceless stops. Not surprisingly, [s] had a higher relative rms than the stops in these tokens (Table 6-36). Table 6-36: Mean average relative rms of adjacent stops and [s] Casual Fast Stop 0.247 0.189 Stop burst 0.339 0.247 [s] 0.985 0.849 p-values stop vs. [s] 0.000 0.000 stop burst vs. [s] 0.000 0.000 This would be as expected since stops in general had a lower relative rms than the other consonants.

PAGE 127

99 6.4.5. Word-final Consonants Compared to Adjacent [p] The word-final [t], which following [p], had on average a higher relative rms than the preceding [p] in the tokens ‘hopped’ and ‘fopt’ (Table 6-37). Table 6-37: Mean average relative rms of adjacent [t] and [p] Casual Fast [t] 0.404 0.263 [p] 0.140 0.143 p-values [t] vs. [p] 0.004 0.077 Although not significant at all speaking rates, it is worth noting that this pattern was true for all of the tokens collected from all of the speakers. 6.5. Ranking of Consonants in Parallel Environments The following subsections show the ranking of the consonants based upon their relative rms in a variety of environments. In general, the strident fricatives and [] tended to have greater relative rms than the other consonants in parallel environments. The stops tended to have the lowest relative rms, except after [] where the relative rms of the nasals was less than the stops. 6.5.1. Preceding Voiceless Stops Preceding voiceless stops, a variety of consonants occurred [p, s, m, n, l, ]. Ranking of these consonants based on their relative rms preceding voiceless stops is shown in Table 6-38. Table 6-38: Ranking of rms of consonants preceding stops Casual and fast speech [] >> [s] >> [n] >> [l] >> [m] >> [p] The ranking was the same for both casual and fast speech. The ranking is similar to the sonority hierarchy in that ranking of [] is high and the ranking of the stop [p] is the

PAGE 128

100 lowest. However, the ranking of the nasals, [s] and [l] do not adhere to the sonority hierarchy. The ranking is similar to the duration ranking as a proportion of the rhyme shown again in Table 6-39. Table 6-39: Ranking of consonants based on duration of rhyme preceding stops Casual and fast speech [] >> [s] >> [m, n], [l], [p] Here also [] outranks [s]. As a proportion of the duration though it was not possible to distinguish between the stop [p] and [l] and the nasals. 6.5.2. Preceding Fricatives In the data collected only [l] and [] were found preceding fricatives. The ranking of their relative rms adhered to the sonority hierarchy with [] outranking [l] in all speaking rates. The same was found when comparing the duration of [] and [l] where [] had the longer duration. 6.5.3. Following Sonorants All three sonorants occurred following the vowel and before a word-final consonant. The nasals though were only found preceding voiceless stops, so there are no consonants to compare there. Following [l] only stops and fricatives occurred. The ranking of their relative rms in this environment did adhere to the sonority hierarchy with fricatives outranking the stops. The same was true when comparing the duration. Following [], the phone with the highest relative rms was [s]. The nasals and [l] did not have greater relative rms than each other. They did though have a greater relative

PAGE 129

101 rms than [f] which had a greater relative rms than the stops. This would give the ranking in Table 6-40. Table 6-40: Ranking of rms of consonants following [] Casual and fast speech [s, ] >> [n], [m], [l] >> [f] >> [p, t, k] >> [b, d] This ranking is similar to the sonority hierarchy in that the ranking of the sonorant consonants is greater than the stops and the fricative [f]. However, it was not possible to rank the [l] and the nasals. Also, the strident fricatives outranked the sonorants. This ranking also suggests a division between stops and fricatives. A division that allows the strident fricatives to be ranked even higher than the nasals and [l]. The ranking in Table 6-41 has some similarities to the ranking of the duration of the consonants following []. Table 6-41: Ranking of duration of consonants following [] Casual and fast speech [s, ] >> [l] >> [p, t, k] >> [m, n] [f]>> [b, d] In both rankings, the strident fricatives are ranked highest and the voiced stops are ranked lowest. In the duration rankings, though, the voiceless stops are ranked higher than the nasals and non-strident fricatives because length is a cue for voicing of word-final obstruents in English, but this contradicts the sonority hierarchy. The ranking of the relative rms is more similar to the sonority hierarchy with nasals and [l] ranked above the obstruents except for the strident fricatives.

PAGE 130

CHAPTER 7 RUSSIAN DURATION ANALYSIS 7.1. Introduction In addition to the analysis done with English data, preliminary consideration was given to data collected from a native speaker of Russian. Like English, Russian allows a variety of consonant clusters some of which are similar to English clusters but with more limitations on the clusters allowed in word-final position. This chapter outlines the similarities and differences in the Russian speakers production of the clusters that were also found in the English data. Because of different phonotactic constraints than English, not all of the same clusters could be considered in Russian. Before analyzing the individual consonants, this chapter shows that the speaker did increase his speaking rate as they were instructed during data collection. Second, it describes how the duration of consonants varies dependent upon the speaking rate. Third, this chapter considers how each consonant varies dependent upon its position in the token. Fourth, it describes how each consonant compares to the other consonants in parallel environments. Finally, this chapter summarizes the findings describing how they relate to the sonority hierarchy. Because duration is relative to the rate each repetition of the tokens, the duration of the consonant was taken as a percentage of the duration of both the coda and the rhyme. Duration was not taken as a percentage of the word since the initial consonant was not the same for all tokens. The coda is the duration from the offset of the vowel to 102

PAGE 131

103 1 I recognize that this is a rather loose definition of ‘coda’ since word-final segments are in some cases often considered ‘extra-syllabic’. the end of the word1. The duration of the rhyme is the duration from the onset of the vowel through the end of the word. Details as to how these divisions were made can be found in Chapter 3. The analysis of the Russian data found several differences from the English data. While the English [] tended to be longer than the other consonants in parallel environments, the Russian trill [r] tended to be shorter than the other consonants in parallel environments. In addition, there was more variation of the rankings of the duration of the consonants in different environments. While sometimes the strident fricatives were longest, other times the stops were longer. 7.2. Speaking Rates According to the instructions during the data collection, the speaker was told to increase the speaking rate from slow to casual to fast as he read the lists of tokens in the carrier phrase “Skazhite, pozhalujsta, _____ vslux” (“Say, please, _____ out loud”). Since some consideration in this chapter is given to how the phones varied according to speaking rate, it is important to verify that he did successfully accomplish this task. Comparisons were made of the word duration to ensure that he succeeded at speaking faster for each repetition. The duration of all of the tokens considered for this study were considered when determining if the speaking rate increased. Considering the speaking rate of each of the tokens, the fast speaking rate was significantly faster than casual (Table 7-1).

PAGE 132

104 Table 7-1: Mean average rate of all tokens Casual Fast p-value 37% 31% 0.000 7.3. Comparisons across Speaking Rates Consonants were compared at each speaking rate in parallel environments.en comparing speaking rates, most differences in quality occurred between slow and casual and slow and fast speech. This may be because the slow speech was the speaker’s first run through each list, so in addition to speaking more slowly, the speaker may have been speaking more carefully. This study, though, compares only the casual and fast speech. 7.3.1. Obstruents Wh 7.3.1.1. Stops Stops occurred word-finally following each of the sonorants, [s], and [p], and preceding [t] (Table 7-2). Table 7-2: Tokens with stops Environment Real Words Nonsense Words Following and preceding stops [opt] [fakt] [fopt] [bakt] Following [s] [last] [plast] [bask] [fast] [pask] Following nasal [bant] [frant] [shtamp] [vamp] [famp] [pant] Following [l] [palt] Following [r] [karp] [part] [shark] [farp] [gart] [fark] Unlike the English examples, the duration of the stops actually increased slightly from casual to fast speech (Casual Rhyme 335, Fast Rhyme 34%, Casual Coda 59%, Fast Coda 61%) (Table 7-3).

PAGE 133

105 Table 7-3: Duration of stops following sonorants Casual Rhyme Fast Rhyme Casual Coda Fast Coda Following sonorants 33% 34% 59% 61% Following [s] 35% 19% 61% 36% Following [p] and [k] 34% 29% 57% 48% In the other word-final positions, the duration of the stops decreased. Considering the duration of the stops following [s], the duration of the stops on average decreased as speaking rate increased (Casual Rhyme 35%, Casual Coda 61%, Fast Rhyme 19%, Fast Coda 36%). Following [p] and [k], the duration of [t] also decreased (Casual Rhyme 34%, Casual Coda 57%, Fast Rhyme 29%, Fast Coda 48%). Considering the proportion of the rhyme and coda devoted to the voiceless stops that precede [t], the proportion of the both the coda devoted to both [p] and [k] increased from casual to fast speech (Table 7-4). Table 7-4: Mean proportional duration of [p] and [k] preceding [t] Casual Rhyme Fast Rhyme Casual Coda Fast Coda [p] 29% 33% 47% 57% [k] 23% 29% 40% 49% Unlike the English stops, stops in Russian did not always decrease with speaking rate. They tended to actually hold on to their proportion of the rhyme and coda despite increases in speaking rate. 7.3.1.2. Fricatives The fricatives [s], [], and [f] were found following [r]. The fricative [s] was found preceding word-final voiceless stops (Table 7-5).

PAGE 134

106 39% 41% 71% 68% Table 7-5: Tokens with fricatives Environment Real Word Nonsense Word Before stops [last] [plast] [bask] [fast] [pask] Following [r] [fars] [far] [mar] [kars] [par] When [s] appeared between a vowel and a word-final stop, its proportion of the rhyme and coda increased from casual to fast speech (Table 7-6). Table 7-6: Mean proportional duration of [s] preceding stops Casual Rhyme Fast Rhyme Casual Coda Fast Coda 23% 33% 39% 64% The same pattern was found in English were the duration of [s] increased with speaking rate. Following [r] the proportional duration of strident fricatives decreased from casual to fast as both a proportion of the rhyme and coda (Table 7-7). Table 7-7: Mean proportional duration of strident fricatives following [r] Casual Rhyme Fast Rhyme Casual Coda Fast Coda 45% 36% 76% 69% In English, the duration of the strident fricatives also decreased with speaking rate following []. The proportion of the rhyme devoted to [f] decreased from casual to fast speech, but the proportion of the coda devoted to [f] actually increased (Table 7-8). Table 7-8: Mean proportional duration of [f] following [r] Casual Rhyme Fast Rhyme Casual Coda Fast Coda

PAGE 135

107 In English following [], the duration of [f] decreased from casual to fast speech. [f] in English was a greater portion of the rhyme in casual speech (43%) giving it more of duration to decrease in English (39% in fast speech). Similar to English, the speaking rate seemed to affect the duration of the word-final strident fricatives more so than the non-strident fricative [f]. As with English, this might be because the duration of the strident fricatives is longer in casual speech than the duration of the non-strident fricatives giving them more duration to lose with the increase in speaking rate. 7.3.2. Sonorants 7.3.2.1. Nasals Tokens with nasal consonants, had [m] and [n] word-finally after [r] and [l] and occurring preceding word-final stops as is shown in Table 7-9. Table 7-9: Tokens with nasal consonants Environment Real Words Nonsense Words Preceding voiceless stops [bant] [frant] [shtamp] [vamp] [famp] [pant] Following [l] [xolm] [polm] Following [r] [korm] [gorn] [porn] [torm] When nasals preceded voiceless stops, the mean average of the proportion of the rhyme and coda devoted to the nasal increased as speaking rate increased (Table 7-10).

PAGE 136

108 Table 7-10: Mean proportional duration of nasals preceding voiceless stops Casual Rhyme Fast Rhyme Casual Coda Fast Coda Preceding voiceless stops 24% 27% 43% 48% Following [l] 50% 47% 75% 75% Following [r] 40% 41% 72% 67% The percentage devoted to the rhyme remained increased from casual to fast speech. This is similar to what was found in English where the proportion devoted to the rhyme and coda increase with speaking rate. Word-finally after [l], from casual to fast speech the proportion devoted to the rhyme decreased while the proportion to the coda stayed the same. This may be due to the fact that in this position the [l] nearly deletes such that the nasal in both casual and fast already has the greatest portion of the coda. The decrease in rhyme duration is more due to an increase in the vowel duration, not related to an increase in [l]. As when the nasal followed [l], from casual to fast speech the duration difference was not as great. Note that this was not what occurred in English where the duration of the nasals decreased with speaking rate after []. This was due most likely to the interaction of the English []’s interaction with the vowel that allowed it to become a greater portion of the rhyme and coda with increases in speaking rate. Since the Russian trill did not interact in this way, the duration of the nasal increased with speaking rate and seemed to level out from casual to fast speech. 7.3.2.2. Lateral approximant [l] The lateral sonorant [l] was found in the data preceding [m] and [t] (Table 7-11). [l] does not (as I understand it) occur in the same variety of positions, and like [l] in English tends to delete preceding word-final consonants, i.e. English ‘calm’ and ‘palm.’

PAGE 137

109 Nonetheless, [l] was collected in the nonsense word [palt] and preceding the word-final nasal [m]. Table 7-11: Tokens with [l] Environment Real Words Nonsense Words Preceding voiceless stop [palt] Preceding [m] [xolm] [polm] The proportional durations of [l] in both environments is shown in Table 7-12. Preceding [t] the duration of [l] decreased from casual to fast speech. Table 7-12: Mean proportional duration of [l] at different speaking rates Casual Rhyme Fast Rhyme Casual Coda Fast Coda Preceding [t] 30% 14% 48% 30% Preceding [m] 20% 23% 29% 36% The duration of [l] preceding [m], on the other hand, increased from casual to fast speech. 7.3.2.3. Alveolar trill [r] For the Russian trill [r], comparisons were made to the English [] since they behave similarly in terms of where they are allowed to occur in consonant clusters. As with the English data, [r] was found preceding stops, fricatives and nasals (Table 7-13) Table 7-13: Tokens with [r] Real Words Nonsense Words Preceding stops [karp] [part] [shark] [skarb] [nard] [farp] [gart] [fark] [farb] Preceding fricatives [arf] [fars] [far] [mar] [parf] [kars] [par] Preceding nasals [gorm] [gorn] [porn] [torm]

PAGE 138

110 However, in the Russian data, tokens with word-final nasals had [o] as the vowel in the nucleus. Comparisons were made between all of the tokens, but it is recognized that differences found might be attributed to the different nucleus. In the English data [] increased significantly with each increase of speaking rate. The same was not found for the Russian [r] preceding stops (Table 7-14). Table 7-14: Comparisons of [r] preceding voiceless stops at different speaking rates Casual Rhyme Fast Rhyme Casual Coda Fast Coda 14% 19% 26% 34% However, this increase does not compare to the fast speech [] in English where the duration of [] preceding voiceless stops is 49-50% of the rhyme and 58-59% of the coda. Preceding voiced stops the duration of the coda devoted [r] decreased or remained about the same with increases in speaking rate (Table 7-15). Table 7-15: Comparisons of [r] preceding voiced stops at different speaking rates Casual Rhyme Fast Rhyme Casual Coda Fast Coda 17% 17% 42% 31% The duration of the rhyme devoted to [r] remained the same from casual to fast speech at 17%, and decreased as a proportion of the rhyme form 42% to 31%. In English, the duration of [] increased in from casual to fast speech. Preceding fricatives, [r] increased its proportion of the rhyme with each speaking rate increase (Table 7-16).

PAGE 139

111 Table 7-16: Comparisons of [r] preceding fricatives at different speaking rates Casual Rhyme Fast Rhyme Casual Coda Fast Coda Preceding fricatives 15% 17% 26% 31% Preceding strident fricatives 15% 16% 24% 31% Preceding non-strident fricatives 16% 18% 29% 32% This remained true when breaking up the fricatives by stridency. The difference though is not as great compared to the increased proportion in English. Preceding all fricatives, the proportion of the rhyme increased from 41% to 45% from casual to fast speech and 47% to 51% of the coda. Preceding nasals, [r] was longer in fast speech than in casual speech (Table 7-17). Table 7-17: Comparisons of [r] preceding nasals at different speaking rates Casual Rhyme Fast Rhyme Casual Coda Fast Coda 16% 20% 28% 33% In English, also the duration of [] decreased from casual to fast speech. However, the actually durations for English (Rhyme 49%-47%, Coda 62%-61%) were longer than they were in Russian The Russian [r] behaved differently than the English []. Overall, the English [] was longer than the Russian [r]. In addition, the English [] increased with speaking rate preceding all obstruents. However, the Russian [r] decreased as a proportion of the coda preceding voiced stops. Both the English [] and Russian [r] decreased from casual to fast speech preceding nasals.

PAGE 140

112 7.4 Comparisons by Manner of Articulation 7.4.1. Stops vs. Fricatives Fricatives and stops were found in similar environments both word-finally and preceding word-final [t]. Since fricatives are higher on the sonority scale, it was believed that fricatives would have a longer proportional duration than stops in the same position. For stops, this included both the gap and release duration. Both stops and fricatives occurred preceding [t] and following [r]. Preceding [t], [s] was longer in both fast speech, than the stops [p] and [k] (Table 7-18). However, [s] was shorter than [p] and [k] as a proportion of the coda in casual speech Table 7-18: Mean proportional duration of [p], [k], and [s] preceding [t] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda [opt], [fopt] 29% 33% 47% 57% [fakt], [bakt] 23% 29% 40% 49% [last], [plast], [fast] 23% 37% 38% 75% Conversely, in English [s] was significantly longer than [p] in this environment in both speaking rates. In English the duration of [s] was about 10% greater than the rhyme duration of the stops and about 20% greater than the coda duration of the stops (see the stops and fricatives were much closer, and in casual speech, the duration of [p] was greater than the duration of [s]. Following [r], though, there were differences in the duration of the stops and thestrident fricatives (Table 7-19). Table 5-31 in Chapter 5). In Russian, the duration of

PAGE 141

113 Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda Table 7-19: Mean proportional duration of voiceless stops and strident fricatives following [r] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda [karp], [part], [shark], [farp], [gart], [fark], [farb] 33% 37% 64% 65% [fars], [far], [mar], [kars], [par] 45% 36% 76% 69% As in English, in this word-final position the fricatives were a longer proportion of the rhyme and coda than the stops. Considering the non-strident fricative [f] and the bilabial stops [p] and [n], the duration of [f] is longer than the duration of the voiced stop [b] except in casual speech. [p] was a longer proportion of the rhyme and coda in both speaking rates (Table 7-20). Table 7-20: Mean proportional duration of [p] and [f] following [r] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda [karp], [farp] 40% 46% 76% 77% [skarb], [farb] 45% 34% 76% 66% [arf], [parf] 39% 41% 71% 68% For the Russian speaker, strident fricatives are a longer portion of both the rhyme than stops in the same word position. The non-strident fricative [f] though was not a longer proportion of the rhyme and coda than the stop [p]. 7.4.2. Stops vs. Nasals Both stops and nasals occurred word-finally following [r] and preceding [t]. Since nasals rank higher on the sonority hierarchy, it was expected that they would have a longer duration following [r] than the stops, but in the data collected that did not prove to be true (Table 7-21). Table 7-21: Mean proportional duration of voiced stops and nasals following [r]

PAGE 142

114 [karp], [part], [farp], [gart] 33% 37% 64% 65% [skarb], [nard], [farb] 32% 39% 58% 69% [korm], [gorn], [porn], [torm] 40% 41% 72% 67% In casual speech, the duration of the nasals was greater then the duration of both the voiced and voiceless stops. As speaking rate increased, though, the duration of the nasal and the stops became almost equal as both a proportion of the rhyme and coda. Comparing [p] and [n] before [t], the durations of each are about the same in both speaking rates (Table 7-22). Table 7-22: Mean proportional duration of [p], [m], and [n] preceding voiceless stops Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda [opt], [fopt] 29% 33% 47% 57% [bant], [frant], [pant] 29% 31% 56% 55% [tamp], [vamp], [famp] 19% 23% 30% 41% [p] and [n] in this position behave similarly in that the duration of both increases with speaking rate. The same was true in English where the duration of the nasals and stops both increased with speaking rate and were about equal. The duration of [m] is also greater from casual to fast speech. The duration of [m] preceding a voiceless stop is also less than both the duration [n] and [p] preceding a voiceless stop. Stops and nasals did not have much of a difference in duration in parallel environments with the exception of [m] preceding [p]. Word-finally, nasals and voiced stops behave similarly increasing their duration with speaking rate. 7.4.3. Stops vs. Lateral Approximant [l] In the data collected, both [l] and [p] were found preceding a word-final [t]. In fast speech [p] was a about twice the duration that [l] was as part of the rhyme and coda (Table 7-23).

PAGE 143

115 Table 7-23: Mean proportional duration of [p] and [l] preceding [t] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda [opt], [fopt] 29% 33% 47% 57% [palt] 30% 14% 48% 30% In casual speech, they were both about the same duration. This is different from what was found in English where [l] and [p] were about the same duration in this environment in both speaking rates (27-29%) of the rhyme in both speaking rates and (45-58%) of the coda (See [p] behaved similarly to both the English [p] and [l] in terms of duration. However, the Russian [l] was unique in terms of its great loss of duration from casual to fast speech. 7.4.4. Stops vs. Alveolar Trill [r] Table 5-25 in Chapter 5). The Russian The stops [p] and [k] and the liquid [r] all occurred preceding a word-final [t]. In both speaking rates, both [p] and [k] were a longer proportion of both the rhyme and coda than [r] (Table 7-24). Table 7-24: Mean proportional duration of [p], [k], and [r] preceding stops Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda [opt], [fopt] 29% 33% 47% 57% [fakt], [bakt] 23% 29% 40% 49% [part], [gart] 15% 20% 25% 36% This shows how differently the Russian [r] behaved from the English []. When considering [] from the beginning of the F3 transition, in English [] was a longer portion of both the rhyme and coda than [p] in this position. When the transition was not included, [] was still a greater proportion than [p] in all speaking rates with the exception the fast coda.

PAGE 144

116 In Russian, the consonant that ranks higher on the sonority hierarchy, [r], has a shorter duration than the ones that ranks lower [p, k]. In this respect, Russian differs from English were the consonant that ranked higher had a greater duration in most cases. 7.4.5. Fricative vs. Nasals Nasals and fricatives occurred preceding stops and word-finally following [r]. When only considering consonants with the same place of articulation, word-finally fricatives were longer than the nasals in casual speech as both a proportion of the rhyme and coda (Table 7-25). Table 7-25: Mean proportional duration of alveolar and labial nasals [m, n] and fricatives [s, f] following [r] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda [arf], [fars], [parf], [kars] 45% 36% 76% 69% [korm], [gorn], [porn], [torm] 40% 41% 72% 67% However, as speaking rate increased, the proportion of the rhyme devoted to fricative decreased and the proportion devoted to the nasal increased. However, the duration of each decrease with speaking rate. The same holds true when comparing the nasals to just the strident fricatives [s] and [] (Table 7-26). Table 7-26: Mean proportional duration of nasals and strident fricative following [r] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda [fars], [far], [mar], [kars], [par] 45% 35% 76% 66% [gorn], [korm], [porn], [torm] 40% 41% 72% 67% Comparisons were also made pairing the nasals and fricatives by place of articulation. Comparing [s] and [n], the same pattern was found as when comparing the nasals with

PAGE 145

117 the strident fricatives with the exception that the duration of [n] devoted to the coda is now greater than the proportion devoted to [s] (Table 7-27). Table 7-27: Mean proportional duration of [s] and [n] following [r] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda [fars], [kars] 46% 36% 76% 67% [gorn], [porn] 44% 44% 79% 72% Comparing the labial nasal [m] and labial fricative [f], the duration devoted to [f] remained greater than the proportion devoted to [m] in both speaking rates (Table 7-28). Table 7-28: Mean proportional duration of [f] and [m] following [r] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda [arf], [parf] 39% 41% 71% 68% [korm], [torm] 37% 37% 66% 62% This is similar to English where the duration devoted to [f] was greater than [m] in both speaking rates. Another environment that both [s] and nasals were found is preceding voiceless stops. In these cases, the duration of [s] was longer than the nasals in fast speech, but not in casual speech (Table 7-29). Table 7-29: Mean proportional duration of nasals and [s] preceding voiceless stops Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda [bask], [last], [plast], [fast], [pask] 23% 33% 39% 64% [tamp], [vamp], [bant], [frant], [famp], [pant] 25% 25% 48% 49% This is different from English where [s] had a longer proportional duration than the nasals in this position in both speaking rates.

PAGE 146

118 In Russian, nasals and fricatives behaved slightly differently than how they did in English. In English, the fricatives were greater than the nasals in all cases with the one example of [f] and [m] in fast speech where they were equal portions of the coda. In Russian, however, the duration of [s] tended to decrease from casual to fast speech, while the duration of the nasals tended to increase or remain the same. This resulted in the nasals at time being longer than the duration of [s] in fast speech. 7.4.6. Fricatives vs. Lateral Approximant [l] The only environment where both fricatives and [l] occurred in the data was preceding word-final [t]. Preceding [t] the duration of [s] was greater than [l] (Table 7-30). Table 7-30: Mean proportional duration of [s] and [l] preceding [t] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda [last], [plast], [fast] 23% 37% 38% 75% [palt] 30% 14% 48% 30% In this environment, the English [s] was a longer proportion of both the rhyme and coda in both speaking rates (significantly in all cases except as a portion of the fast rhyme). So in this case Russian and English were similar with the exception of the Russian casual speech (See Table 5-31 in Chapter 5). 7.4.7. Fricatives vs. Alveolar Trill [r] The only environment that both [r] and [s] occurred is preceding voiceless stops. In both speaking rates [s] is a longer proportion of both the rhyme and coda than [r] (Table 7-31).

PAGE 147

119 Table 7-31: Mean proportional duration of [s] and [r] preceding voiceless stop Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda [last], [plast], [bask], [fast], [pask] 23% 33% 39% 64% [part], [ark], [gart], [fark] 14% 19% 26% 34% This is different from English where [] was greater than [s] when taking in to consideration the F3 transitions. When the F3 transitions were not included in t English [s] was a greater duration than []. In the case of the Russian trill, such transitions of the vowel do not exist. In Russian, then [s] had a longer proportional duration than [r]. 7.4.8. Nasals vs. Lateral Approximant [l] Comparisons were made of the tokens with [n] and [l] followed by a word-final [t]. As is shown in Table 7-32, the duration of [n] was greater than the duration of [l] with the exception of as a proportion of the casual rhyme. Table 7-32: Mean proportional duration of [n] and [l] preceding [t] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda [fant], [pant] 29% 31% 56% 55% [palt] 30% 14% 48% 30% In English in this environment the duration of [n] and [l] were much closer in duration preceding [t] both about 23-27% of the rhyme and 36-50% of the coda (See Chapter 5). 7.4.9. Nasals vs. Alveolar Trill [r] Table 5-34 in Nasals and [r] were both found preceding voiceless stops. The data in Table 7-33 show that the duration of [r] is shorter than the duration of the nasals in both speaking rates as a portion of both the rhyme and coda.

PAGE 148

120 Table 7-33: Nasal and [r] mean average comparison preceding voiceless stop Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda [vamp], [bant], [famp], [pant] 25% 25% 48% 49% [karp], [part], [farp], [gart] 14% 17% 25% 30% In English, though, the duration of [] was longer the duration of the nasal at all speaking rates preceding a word-final stop. This held true whether or not the F3 transition was included as a portion of []. In English, the consonant ranking higher on the sonority hierarchy had a longer duration than the one with lower sonority, i.e. [] >> [n, m]. However, this was not true in Russian. 7.4.10. Lateral Approximant [l] vs. Alveolar Trill [r] Both [l] and [r] occurred before word-final [t] in the data collected. Comparing [r] and [l] in these tokens, [r] is a greater proportion of both the rhyme and coda (Table 7-34). Table 7-34: Mean proportional duration of [l] and [r] preceding [t] Tokens Casual Rhyme Fast Rhyme Casual Coda Fast Coda [palt] 30% 14% 48% 30% [part], [gart] 15% 20% 25% 36% The consonant that ranks higher on the sonority hierarchy, [r], has a longer duration preceding [t]. This was the same finding in the English data. 7.5. Ranking of Consonants in Environments The following section shows how the consonants are ranked in terms of their duration in parallel environments. Attention is given to how this ranking relates to the sonority hierarchy and how the ranking in Russian differs from the ranking in English.

PAGE 149

121 7.5.1. Preceding Voiceless Stops In the data examined for this study, voiceless stops were preceded by [p, k, s, m, n, l, r]. The ranking of these consonants in terms of their duration before the stops is shown in Table 7-35. Table 7-35: Ranking of consonants by proportion of rhyme and coda duration preceding voiceless stops Casual speech [l] >> [m, n] >> [p, k] >> [s] >> [r] Fast speech [s] >> [p, k] >> [m, n] >> [r] >>[l] In fast speech, the durations of [l] and the nasals rank above the other consonants. The fast speech ranking is like a mirror image of the sonority hierarchy in that the obstruents are ranked at one side of the hierarchy while the sonorants are ranked at the low end. However, the ranking of the individual sonorants and the obstruents do not also mirror the ranking. Instead the fricatives are ranked above the stops and [r] is sandwiched between [l] and the nasals. 7.5.2. Following [r] Following [r], the consonants [t, s, , f, m, n, l] were found. The rankings of these consonants following [r] can be found in Table 7-36. Table 7-36: Ranking of consonants by proportion of rhyme and coda duration preceding following [r] Casual speech [s, ] >> [p, t, k] >> [b, d] >> [f] >> [n, m] Fast speech [p, t, k] >> [b, d] >> [m, n] >> [s, ] >> [f] The casual speech rankings are similar to the rankings in English in that [s] and [] are ranked highest. It is also similar in that the nasals are ranked below voiceless stops. However, in fast speech the ranking is different from English with the strident fricatives ranked low on the hierarchy.

PAGE 150

CHAPTER 8 RUSSIAN ACOUSTIC ANALYSIS OF RELATIVE RMS AMPLITUDE 8.1. Introduction This chapter considers the relative root mean square amplitude (rms) of the consonants in the tokens collected from the Russian speakers. As with the English tokens, amplitude was measured relative to a vowel in the carrier phrase. In this case, the Relative rms amplitude was taken as the phone’s rms amplitude (sound pressure expressed in Pascal) in proportion to the rms amplitude of the final [a] in the word ‘pozhalujsta’ of the carrier phrase, “Skazhite, pozhalujsta, _____ vslux”. Comparisons were made of the rms dependent upon speaking rate and the position in the token. Comparisons were also made to the English data to determine if the Russian speaker patterned similarly to the native English speakers. Because there was only one Russian speaker, comparisons were made using the raw data comparing averages. Consideration was given to whether the production of the consonants was different from the production of the consonants by the native English speakers. When possible t-tests were run, however, it was difficult to determine significance with only a few tokens. 8.2. Speaking Rate As described in Chapter 7, the speaker did increase his speaking rate according to the instructions that were given to him during data collection. Consonants were compared 122 at each speaking rate in the same environments.

PAGE 151

123 In English, the relative amplitude of [s] preceding a word-final stop varied significantly with the change in speaking rate. Similarly, the rms of [s] decreased preceding word-final stops in Russian although not significantly in the data collected (Table 8-1). Table 8-1: Mean relative amplitude of [s] preceding word-final stop Casual Fast p-value 0.594 0.497 0.587 In the English data, word-final voiceless stops significantly decreased relative rms as speaking rate increased. For the native Russian speaker, the relative rms of the word-final stop increased from casual to fast speech (Table 8-2). Table 8-2: Mean relative rms of word-final voiceless stops Casual Fast p-value 0.176 0.218 0.107 Similarly, the word-final stop burst increased from casual to fast speech (Table 8-3). Table 8-3: Mean relative rms of word-final voiceless stops bursts Casual Fast p-value 0.246 0.303 0.045 This may reflect that the word-final stops rarely occur in Russian since words are usually declined with word-final affixes. 8.3. Comparisons by Manner of Articulation It was expected in this research that phones which rank more highly on the sonority hierarchy would have a higher relative rms than phones ranked lower on the sonority hierarchy when they are compared in the same environment. Consonants were compared in parallel environments. When possible consonants were compared with

PAGE 152

124 consonants of the same place of articulation, i.e. [r] was compared with [t] and [d], but not [b]. While neither the tokens speaking rate nor the phone’s placement in the token resulted in many consistent differences, comparing phones of different manners of articulation in the parallel positions did result in differences that were consistent across speaking rates. Unlike English, [s] and [r] did not always tend to have a higher relative rms than consonants in the same position. 8.3.1. Stops vs. Fricatives Both stops and fricatives occurred word-finally and preceding word-final stops. In the latter position, [p], [k], and [s] appear preceding [t]. Stops and fricatives were both found in word-finally following [r]. The expectation was that in all of the environments, the fricatives would have a higher relative rms than the stops. As in English when comparing stops and fricatives, the hypothesis proved to be true that the sound that ranks higher on the sonority hierarchy would have a higher relative rms in all environments. As is shown in Table 8-4, fricatives had a greater relative rms than did the stops preceding [t]. Table 8-4: Mean relative rms of [p, k] and [s] preceding [t] Tokens Casual Fast [opt], [fakt], [fopt], [bakt] 0.129 0.133 [last], [plast], [fast] 0.653 0.545 Following [r], the same is found (Table 8-5). The relative rms of the fricatives was greater whether comparing the fricatives to the entire stop (both the gap and burst) or just comparing the fricatives to the burst.

PAGE 153

125 Table 8-5: Mean relative rms of stops [p, t] and fricatives [f, s] following [r] Tokens Casual Fast [karp], [part], [shark], [farp], [gart] Stop gap and burst 0.212 0.221 Stop burst 0.329 0.376 [arf], [fars], [parf], [kars] 0.447 0.429 Comparing the stops to the strident fricatives, the fricatives still have a greater relative rms (Table 8-6). Table 8-6: Mean relative rms of stops [t, k] and strident fricatives [s, ] following [r] Tokens Casual Fast [part], [shark], [gart], [fark] Stop gap and burst 0.154 0.125 Stop burst 0.168 0.160 [fars], [far], [mar], [kars], [par] 0.978 1.020 Comparing the non-strident fricative [f] to the stop [p], though, the difference between the stop and fricative is not as great. In addition, they vary as to which has the greater relative rms in the different speaking rates (Table 8-7). Table 8-7: Mean relative rms of [p] and non-strident [f] following [r] Casual Fast [p] 0.164 0.235 [f] 0.155 0.294 8.3.2. Stops vs. Nasals Both stops and nasals are found word-finally following [r]. Nasals were compared to the voiced stops and voiceless stops with the same places of articulation, bilabial and alveolar. Since nasals rank higher on the sonority hierarchy, it was expected that nasals would have a higher relative rms than the stops. For the data collected, regardless of whether the nasals are compared to the voiced or voiceless stops, the nasals did have relative rms amplitude than the stops (Table 8-8).

PAGE 154

126 Table 8-8: Mean comparisons of stops and nasals following [r] Tokens Casual Fast [karp], [part], [farp], [gart 0.212 0.280 [skarb], [nard], [farb] 0.108 0.185 [korm], [gorn], [porn], [torm] 0.470 0.439 p-values nasal vs. voiceless stops 0.033 0.200 nasal vs. voiced stop 0.000 0.086 This same pattern was found in English where the nasals had significantly greater amplitude than the stops. Comparing just the stop burst relative rms and the nasals, the mean relative rms is still greater (Table 8-9). Table 8-9: Mean relative rms of stop bursts and nasals following [r] Tokens Casual Fast [karp], [part], [farp], [gart] 0.197 0.279 [skarb], [nard], [farb] 0.099 0.313 [korm], [gorn], [porn], [torm] 0.470 0.439 So, in the comparison of stops and nasals the hypothesis held true that nasals would have a greater relative rms than the less sonorous stops in the same environment. 8.3.3. Stops vs. Lateral Approximant [l] Both stops and [l] occurred in the data preceding a word-final [t]. Considering the [l] and [p] before a word-final [t] [l] had the greater relative rms amplitude for [l] (Table 8-10). Table 8-10: Mean relative rms of [p] and [l] preceding [t] Tokens Casual Fast [opt], [fopt] 0.178 0.224 [palt] 0.891 0.972 Again, the stops had a lower relative rms than the more sonorous consonant, in this case [l].

PAGE 155

127 8.3.4. Stops vs. Alveolar Trill [r] The only environment where both [r] and a stop appeared was preceding a word-final [t]. Just as with the other sonorants, [r] had a higher relative rms than [p] in this position (Table 8-11). Table 8-11: Mean relative rms of stops and [r] preceding [t] Tokens Casual Fast [opt], [fopt] 0.178 0.224 [fakt], [bakt] 0.079 0.087 [part], [gart] 0.979 0.445 Again the finding in this subsection are not surprising since [r] is ranked higher than [p] in the sonority hierarchy. 8.3.5. Fricative vs. Nasals Both fricatives and nasals appeared word-finally following [r] and preceding a word-final voiceless stop. It was expected that in all of these environments the nasals should have a higher relative rms than the fricatives. Consideration was first give to fricatives and nasals with the same places of articulation (labial and alveolar) following [r] (Table 8-12). Table 8-12: Mean relative rms of labial and alveolar nasals and fricatives following [r] Tokens Casual Fast [arf], [fars], [parf], [kars] 0.743 0.812 [gorn], [korm], [porn], [torm] 0.502 0.421 In this position, the fricatives had a greater relative rms than the nasals in both speaking rates. This is contrary to the expectation that the fricatives would have a greater relative rms in this position. This is contrary to what was found for the native English speakers who on average had a higher relative rms for the nasals in this position although the difference was not significant suggesting that this was not true for all of English speakers.

PAGE 156

128 Comparing only the alveolar [s] and [n], the same was found with [s] having a greater relative rms in both speaking rates (Table 8-13). Table 8-13: Mean relative rms of [s] and [n] following [r] Tokens Casual Fast [fars], [kars] 0.740 0.565 [gorn], [porn] 0.581 0.452 However, comparing the non-strident [f] to the nasal [m], the Russian speaker behaved similarly to the English speakers with a higher relative rms for the nasal than for the fricative (Table 8-14). Table 8-14: Mean relative rms of [f] and [m] following [r] Casual Fast [arf], [parf] 0.155 0.294 [korm], [torm] 0.424 0.391 As with English the relative rms of [s] is greater than that of [n], while the relative rms of [m] is greater than that of [f]. This is not surprising since now the strident and non-strident [s] and [f] are separated into two different data sets. When all of the tokens with a word-final strident fricative, i.e. [s] and [], were compared to the nasals, not surprisingly the rms of the fricatives was greater than that of the nasals (Table 8-15). Table 8-15: Mean relative rms of nasals [m, n] and strident fricatives [s, ] following [r] Tokens Casual Fast [fars], [far], [mar], [kars], [par] 0.978 1.020 [gorn], [korm], [porn], torm] 0.502 0.421 Finally, comparisons were also made of the nasals and [s] preceding word-final stops (Table 8-16). In this case, in both speaking rates, the nasals had a greater relative rms than [s]. In English, the opposite was found. [s] had a greater relative rms than the nasals in all speaking rates.

PAGE 157

129 Table 8-16: Mean relative rms of nasals and [s] preceding voiceless stops Tokens Casual Fast [bask], [last], [plast], [fast], [pask] 0.653 0.545 [tamp], [vamp], [bant], [frant], [famp], [pant] 0.783 0.635 Word-finally then, the strident fricatives had a greater relative rms than the nasals. The non-strident fricative [f] though had a lower relative rms than the nasals word-finally. This is similar to what was found in English. The Russian speaker differed from the English speakers in that the relative rms of the nasals preceding word-final stops was greater than the relative rms of [s] in this position. 8.3.6. Fricatives vs. Lateral Approximant [l] Comparing [l] and [s] is difficult since only one token of [l] before [t] was collected. Considering these tokens [l] has a greater relative rms preceding [t] in both speaking rates (Table 8-17). Table 8-17: Mean relative rms of [s] and [l] preceding [t] Casual Fast [last], [plast], [fast] 0.680 0.439 [palt] 0.891 0.972 In fast speech, the relative rms of [l] (.972) was more than twice the relative rms of [s] (.439). Although, this may not be significant as it relates to Russian as a whole, it is worth noting for the perception experiment, that in the data collected then [l] has a greater relative rms in the environment of preceding [t]. This is not what was found in the English data where [s] on average had a greater relative rms than [l]. 8.3.7. Fricatives vs. Alveolar Trill [r] As in the English data, the only fricative that occurred in the same positions as [r] was [s]. Both appeared before word-final voiceless stops. As is shown in Table 8-18, [s]

PAGE 158

130 had a greater relative rms than [r] in both casual and fast speech. In addition, the relative rms of [r] greatly decreased from casual to fast speech. Table 8-18: Mean relative rms of [s] and [r] preceding voiceless stop Casual Fast [last], [plast], [bask], [fast], [pask] 0.991 0.735 [part], [ark], [gart], [fark] 0.758 0.482 This is contrary to what was found in the English data where [] held on to its relative rms in both speaking rates, but decreased for the fricative. 8.3.8. Nasals vs. Lateral Approximant [l] Both [l] and the nasals occurred preceding [t]. In this position, [l] had a greater relative rms than the nasals in both speaking rates (Table 8-19). Table 8-19: Mean relative rms of [n], [m], and [l] preceding voiceless stop Tokens Casual Rhyme Fast Rhyme [bant], [pant] 0.825 0.442 [vamp], [famp] 0.741 0.829 [palt] 0.891 0.972 Similarly, in English in parallel environments [l] and the nasals varied as which had a relative rms at different speaking rates. 8.3.9. Nasals vs. Alveolar Trill [r] Both [r] and the nasals occurred preceding voiceless stops. For comparisons, only the [r] tokens with word-final [p] and [t] were used so that the environments were consistent. For this Russian speaker the findings were the opposite than it was for the native English speakers. For the Russian speaker the relative rms was less for [r] than it was for the nasals in both speaking rates (Table 8-20).

PAGE 159

131 Table 8-20: Mean relative rms of nasals and [r] preceding voiceless stop Tokens Casual Fast [bant], [vamp], [famp], [pant] 0.677 0.737 [karp], [part], [farp], [gart] 0.575 0.616 For nasals, the consonants that rank lower on the sonority hierarchy (i.e. nasals) had a greater relative rms than [r] in a parallel environment. 8.3.10. Lateral Approximant [l] vs. Alveolar Trill [r] Both [r] and [l] occurred preceding [t]. In English, [] always had a greater relative rms than [l] in all environments. In Russian, though, this did not prove to be true (Table 8-21). Table 8-21: Mean relative rms of [r] and [l] before [t] Tokens Casual Fast [palt] 0.891 0.972 [part], [gart] 0.979 0.445 In fast speech the relative rms of [l] was greater than the relative rms of [r], but the opposite was true in casual speech. Therefore, unlike in English, neither consistently had greater relative rms. 8.4. Amplitude Compared to Adjacent Consonant This section compares the relative rms of consonants that are adjacent to one another. According to the hypothesis, it is expected that of two adjacent consonants, the consonant higher on the sonority hierarchy would have a greater relative rms. Of course, this could also be accounted for by the quality of the individual phone or by the individual phone’s placement in the token. That is, word-final consonants may be less sonorous simply because they are at the end of a word.

PAGE 160

132 8.4.1. Word-final Consonants Compared to Adjacent [r] Comparing [r] to its adjacent word-final consonant gave the same findings in Russian that it did in English. [r] had a higher relative rms than adjacent word-final stops whether considering just the entire stop or just the burst (Table 8-22). Table 8-22: Mean rms of adjacent voiceless stop and [r] Casual Fast Stop 0.158 0.161 Stop burst 0.197 0.250 [r] 0.758 0.526 Including all of the word-final fricatives [f, s, ], the fricatives had a higher relative rms than [r] in both speaking rates (Table 8-23). Table 8-23: Mean rms of adjacent fricative and [r] Casual Fast Mean fricative 0.743 0.812 Mean [r] 0.468 0.450 If the tokens are broken up by whether or not the fricatives are strident, it is shown that the strident fricatives have a greater relative rms than [r] (Table 8-24). Table 8-24: Mean rms of adjacent strident fricative and [r] Casual Fast Strident fricative 0.978 1.020 Mean [r] 0.531 0.401 However, when [r] is compared to the adjacent non-strident [f], now [r] has a greater relative rms in [arf] and [parf] (Table 8-25). Table 8-25: Mean rms of adjacent non-strident fricative and [r] Casual Fast Nonstrident fricative 0.155 0.294 Mean [r] 0.309 0.575

PAGE 161

133 This same pattern was true in English where English [] had a greater relative rms than adjacent consonants except when compared to the strident fricatives. The last consonants that [r] occurred adjacent to were nasals in [gorn], [korm], [porn] and [torm] (Table 8-26). In these cases, [r] had a greater relative rms than the nasals. The same was found in English. Russian and English both had [r] and [] with a relative rms greater than adjacent consonants except strident fricatives. Given that [r] and [] rank high on the sonority hierarchy this is what was expected. Table 8-26: Mean rms of adjacent [m] and [r] Casual Fast Mean nasal 0.470 0.439 Mean [r] 0.508 0.700 8.4.2. Word-final Consonants Compared to Adjacent [l] Compared to the word-final-stop [t], [l] had a higher relative rms than [t] in the tokens and [palt]. This was true when comparing both the entire stop and just the burst (Table 8-27). Table 8-27: Mean rms of adjacent stop and [l] Casual Fast Stop 0.156 0.088 Stop burst 0.199 0.167 [l] 0.891 0.972 Comparing [l] to the adjacent [m] in [xolm] and [kolm] reveals that [l] has a greater relative rms in this environment (Table 8-28).

PAGE 162

134 Table 8-28: Mean rms of adjacent [m] and [l] Casual Fast [m] 0.405 0.475 [l] 0.964 1.099 So in all of the tokens with [l], [l] had a greater relative rms than the adjacent consonant. 8.4.3. Word-final Consonants Compared to Adjacent Nasals Nasals that appeared adjacent to word-final stops had a greater relative rms than both the stops and the stop bursts (Table 8-29). Table 8-29: Mean average relative rms of adjacent stops and nasals Casual Fast Stop 0.209 0.222 Stop burst 0.280 0.314 Nasal 0.852 0.685 As with English, this agrees with the hypothesis that more sonorant consonants would have a higher rms than consonants that they are adjacent to. 8.4.4. Word-final Consonants Compared to Adjacent [s] [s] occurred adjacent to word-final stops. Not surprisingly [s] had a higher relative rms in these tokens (Table 8-30). Table 8-30: Mean average relative rms of adjacent stops and [s] Casual Fast Stop 0.161 0.315 Stop burst 0.224 0.346 [s] 0.594 0.497 This would be as expected since stops in general had a lower relative rms than the other consonants. 8.4.5. Word-final Consonants Compared to Adjacent Voiceless Stops The word-final [t] which followed [p] had and average higher relative rms than the preceding [p] in the tokens [opt] and [fopt] (Table 8-31).

PAGE 163

135 Table 8-31: Mean average relative rms of adjacent [t] and [p] Casual Fast [t] 0.280 0.438 [p] 0.178 0.224 [t] also had the greater relative rms when comparing it with [k] in the tokens [fakt] and [bakt] (Table 8-32). Table 8-32: Mean average relative rms of adjacent [t] and [k] Casual Fast [t] 0.280 0.438 [k] 0.079 0.087 8.5. Ranking of Consonants in Environments The following subsections show the ranking of the consonants’ relative rms in the environments of preceding stops and following [r]. It is shown that Russian was not the same as English. This is especially true preceding word-final stops where [s] and [r] were did not have the highest relative rms. 8.5.1. Preceding Voiceless Stops Preceding voiceless stops, a variety of consonants occurred [p, s, m, n, l, r]. Ranking of these consonants based on their relative rms preceding voiceless stops is shown in Table 8-33. Table 8-33: Ranking of rms of consonants preceding stops Casual and fast speech [l] >> [m, n] >> [s] >> [r] >> [p] The ranking was the same for both casual and fast speech. The ranking is quite different from English where [] and [s] outranked the other consonants (Table 8-34). Table 8-34:Ranking of English rms of consonants preceding stops Casual and fast speech [] >> [s] >> [n] >> [l] >> [m] >> [p]

PAGE 164

136 8.5.2. Following [r] Following [r], nasals, stops, strident and non-strident fricatives occurred. In both speaking rates, the ranking of the environments according to the relative rms of [r] is found in Table 8-35. Table 8-35: Ranking of rms of consonants following [r] Casual and fast speech [s, ] >> [n], [m] >> [f] >> [p, t, k] The ranking of the stops and the fricative [f] actually varied with speaking rate. This ranking is similar to the ranking of the sonority hierarchy in that the ranking of the sonorant nasals is greater than the stops and the fricative [f]. The strident fricatives outranked the sonorants as was found in English.

PAGE 165

CHAPTER 9 ENGLISH PERCEPTION 9.1. Introduction After the analysis of the tokens was completed, a perception experiment was undertaken to determine if the relative rms and duration of a consonant correlated to its perceptibility. Six native speakers of English (three men and three women) listened to both the English and Russian tokens embedded in the phrase ‘Please say ______ for me.’ The phrases were embedded in three levels of pink noise (+6db, 0dB, -6db). Listeners took a multiple-choice test in which they were told to choose the word that most sounded like the word that they heard. For details on how the perception experiment was designed, see chapter 4. When data collection was completed, responses were coded as all consonants in cluster perceived, one or more consonants not perceived, or one or more consonants misperceived. See Appendix D for the multiple choices associated with each token. The remainder of this chapter details the finding as for each consonant. All of the English tokens in the perception experiment were produced by the same speaker (Speaker 2). He was chosen primarily because he often broke between the token and the next word in the carrier phrase ‘for.’ This allowed ease in cutting the tokens without worrying about the word-final stop burst or fricative blending with the [f] in ‘for.’ After analyzing the data, it was expected that (1) consonants would be more difficult to perceive adjacent to consonants that they are closer to on the sonority 137

PAGE 166

138 hierarchy and (2) consonants lower on the sonority hierarchy would be more difficult to perceive, particularly when the noise outweighed that sound. Patterns were found in terms of which consonants were unable to be perceived in which environments. Listeners had difficulty perceiving word-final stops, especially following [s] and [p]. Even when they did perceive the presence of the stops, they did not always perceive the correct place of articulation. They also did not always perceive nasals when they perceived word-final stops. In addition, listeners did always perceive [l] when it followed []. The listeners had the least difficulty perceiving the strident fricatives [s, ] and []. However, they did at times misperceive the place of articulation of [s]. The patterns were essentially the same when the listeners heard the English and Russian tokens with exception that Russian [r] was more difficult to detect than English []. 9.2. English Tokens 9.2.1. Perception of Obstruents With the exception of the strident fricatives, the obstruents tended to have shorter duration and lower rms than the other consonants being considered. For this reason, it was expected that these would be the consonants more likely to be either misunderstood or not heard at all. For the word-final stops, this proved to be true. They were often not heard at all and when there was more than one option as to the place of articulation of the stops, the listeners often chose the wrong option. The obstruents [s] and [p] also occurred preceding stops. Since [s] was longer and had a greater relative rms than the stops that preceded, it was expected that would not be

PAGE 167

139 misheard or missed in this position. As for [p], it was not certain whether or not it would be misheard or misperceived preceding [t], since they are of equal sonority. 9.2.1.1. Stops in casual speech As stated above it was expected that stops would be misheard or misunderstood, particularly when they occurred word-finally. Table 9-1 shows the results of the perception experiment. The +6dB column represents those repetitions where the token was greater than the noise by 6dB. The -6dB column represents those repetitions where the token was less then the noise by 6dB. At 0dB, the noise and the token were equal. ‘Not Heard’ means that the listener chose the option in which the stop was not represented in the spelling. ‘Misheard’ means that the listener chose the option where the spelling represented a different consonant. The count of the ‘not heard’ and ‘misheard’ is out of the number of listeners since each listener had the opportunity to hear each token only once in each noise level. When ‘N/A’ appears in a table this means that the listeners were not given the option in the multiple choice of misperceiving the consonant’s place or manner of articulation. This description of the tables hold true for the perception of all of the consonants in this chapter. As can be seen in the table, at different levels of noise, the listeners varied as to whether or not they could perceive the word-final stops. Table 9-2 shows the rankings of the ability of the listeners where the greater the environment fall on the hierarchy the more likely it is that the stop would not be perceived. In both speaking rates, the only environment that had a significantly greater chance of not being perceived was following [l] which was significantly greater than the other environments at an 80% level of confidence.

PAGE 168

140 Table 9-1: Perception of word-final voiceless stops in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard Stop+[t] fopt 0 N/A 0 N/A 1 N/A Sum 0 N/A 0 N/A 1 N/A Percentage Missed 0% 0% 17% [s]+stop fosp 1 1 1 1 1 1 fost 0 N/A 0 N/A 1 N/A posk 3 0 0 1 1 1 Sum 4 1 1 2 3 2 Percentage Missed 22% 8% 6% 17% 17% 17% nasal+stop fomp 1 0 0 0 0 0 pont 0 N/A 1 N/A 1 N/A Sum 1 0 1 0 1 0 Percentage Missed 8% 0% 8% 0% 8% 0% [l]+stop pault 3 0 2 1 3 0 Sum 3 0 2 1 3 0 Percentage Missed 50% 0% 33% 17% 50% 0% []+stop farp 0 0 0 0 0 0 gart 0 1 0 2 0 4 fark 0 1 1 3 1 2 Sum 0 2 1 5 1 6 Percentage Missed 0% 11% 6% 28% 6% 33% Table 9-2: Ranking of perception of word-final stops by environment in casual speech +6dB [l__] >> [s__] >> [m, n __] >> [__], [p__] 0dB [l__] >> [m, n __] >> [s__], [__], [p__] -6dB [l__] >> [s__], [p__] >> [m, n __] >>[__]

PAGE 169

141 Most notable about all of the rankings is that stops were least often perceived following [l] in all three noise levels and were most often perceived following [] in all three levels of noise. It should not go unnoticed that in the tokens collected [l] agrees in place of articulation with the word-final [t]. However, this alone does not explain why [t] is not perceived as easily following [l], since comparing ‘pault’ to ‘fost,’ ‘pont,’ and ‘gart’ gives the same results. Also interesting is that the perception of [t] following [p] became more difficult as speaking rate increased. It should also not go on unnoticed that when the listeners were given options as to what that stop might be, they did often report mishearing the place of articulation. So far then looking at word-final stops in relationship to the hypothesis that the perception of consonants would be more difficult adjacent to consonants that they are closer to on the sonority hierarchy particularly in noisier contexts bears out to an extent. The stops are more likely perceived adjacent to [], which is the consonant on the further end of the sonority hierarchy. In addition, the perception of the stops became more difficult adjacent to other stops as the noise level increased. Looking now at [p] when it precedes [t], in this case, it also became more difficult for the listener to perceive the [p] (Table 9-3). Table 9-3: Perception of [p] preceding [t] in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [p]+stop fopt 1 0 1 0 2 0 Sum 1 0 1 0 2 0 Percentage Missed 17% 0% 17% 0% 33% 0%

PAGE 170

142 Putting this environment in the rankings below, shows that the environment ranks high on the hierarchy (Table 9-4). Table 9-4: Ranking of perception of word-final stops by environment in casual speech +6db [l__] >> [s__] >>[__t] >> [m, n __] >> [__], [p__] 0dB [l__] >> [__t] >> [m, n __] >> [s__], [__], [p__] -6dB [l__] >> [__t] >> [s__], [p__] >> [m, n __] >>[__] Voiced stops were found following [l] and []. In this position also, the listeners had more difficulty hearing stops when they followed [l] except in the 0dB signal to noise ratio (Table 9-5). Table 9-5: Perception of voiced stops in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [l]+vd. stop fald 2 N/A 0 N/A 2 N/A Sum 2 N/A 0 N/A 2 N/A Percentage Missed 33% N/A 0% N/A 33% N/A []+vd stop farb 0 0 0 3 0 3 fard 1 0 1 0 1 1 Sum 1 0 1 3 1 4 Percentage Missed 8% 0% 8% 25% 8% 33% Considering all of the stops in casual speech, they were most often most difficult to perceive following [l] and least difficult to perceive following []. The other environments varied as to their affect on the perception of word-final stops. 9.2.1.2. Stops in fast speech It was expected that there would be more difficulty hearing the tokens perceived at the fast speaking rate than at the casual speaking rate (Table 9-6). The rankings of the

PAGE 171

143 environments dependent upon the perception of the word-final stops are shown in Table 9-7. Table 9-6: Perception of word-final voiceless stops in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard stop+[t] fopt 0 N/A 1 N/A 2 N/A Sum 0 N/A 1 N/A 2 N/A Percentage Missed 0% N/A 17% N/A 33% N/A [s]+stop fosp 0 1 3 0 2 1 fost 1 N/A 4 N/A 1 N/A posk 0 0 1 1 1 0 Sum 1 1 8 1 4 1 Percentage Missed 6% 8% 44% 8% 22% 8% nasal+stop fomp 0 0 0 0 0 0 pont 0 N/A 0 N/A 0 N/A Sum 0 0 0 0 0 0 Percentage Missed 0% 0% 0% 0% 0% 0% [l]+vl stop pault 0 0 1 0 0 0 Sum 0 0 1 0 0 0 Percentage Missed 0% 0% 17% 0% 0% 0% []+vl stop farp 0 0 0 0 0 0 gart 2 0 2 0 3 0 fark 0 3 0 4 0 3 Sum 2 3 2 4 3 3 Percentage Missed 11% 17% 11% 22% 17% 17% As in the casual speech, when the adjacent consonants were adjacent to or equal to the word-final stops on the sonority hierarchy, it became more difficult to detect the word-final stops as the noise level increased. The fast speech differed from the casual

PAGE 172

144 speech in that the listeners always perceived the word-final stop adjacent to [l] in dB and +6dB signal to noise ratios. In addition, they were more likely to not perceive the [t] adjacent to [] in fast speech token ‘gart’ and adjacent to [s] in fost,’ but did not have this trouble when the final stop was a [p] or a [k]. Now having adjacent consonants with the same place of articulation poses a problem with perception as was true in casual speech for [lt] and [ld] clusters. Table 9-7: Ranking of perception of word-final stops by environment in fast speech +6dB [l__] >> [s__] >> [m, n __] >> [__], [p__] 0dB [s__] >> [l __] >> [p__] >> [__], [m, n__] -6dB [p__] >> [s__] >> [__] >> [m, n __], [l__] Considering two adjacent stops, the listeners several times did not detect [p] in all three noise levels (Table 9-8). When this environment is considered with the other environments, it gives the rankings in Table 9-9. Table 9-8: Perception of [p] preceding [t] in casual speech in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [p]+stop fopt 3 0 2 0 2 0 Sum 3 0 2 0 2 0 Percentage Missed 50% 0% 33% 0% 33% 0% Table 9-9: Ranking of perception of word-final stops by environment at +6dB in fast speech +6dB [__t] >> [l__] >> [s__] >> [m, n __] >> [__], [p__] 0dB [s__] >> [__t] >> [l __] >> [p__] >> [__], [m, n__] -6dB [__t], [p__] >> [s__] >> [__] >> [m, n __], [l__] As with the casual speech, the listeners did not always perceive the word-final voiced stops (Table 9-10). The fast speech differed from the casual speech in that the in

PAGE 173

145 the dB the listeners always perceived [l], but did not always perceive [] giving a different ranking in this environment. The rankings of the environments based on the perception of the voiceless stops are founding Table 9-11. Table 9-10: Perception of word-final voiced stops in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [l]+vd. stop fald 1 N/A 2 N/A 0 N/A Sum 1 N/A 2 N/A 0 N/A Percentage Missed 17% N/A 33% N/A 0% N/A []+vd stop farb 0 1 0 0 0 1 fard 0 0 0 0 1 0 Sum 0 1 0 0 1 1 Percentage Missed 0% 8% 0% 0% 8% 8% Table 9-11: Ranking of perception of word-final voiced stops by environment in fast speech +6dB, 0dB [l __] >> [__] -6dB [__] >> [l__] 9.2.1.3. Fricatives in casual speech Strident fricatives occurred following [l] and []. The non-strident fricative [f] occurred following []. In the casual speech production of ‘palse,’ ‘palsh,’ ‘barce,’ and ‘parsh,’ the listeners always perceived the presence of the word-final strident fricatives, regardless of the signal to nose ratio. In addition, they never misperceived the place of articulation of the fricatives. One speaker though did misperceive the word-final [s] in ‘barce’ as [] in the greater noise level. Looking at the non-strident fricative [f], the

PAGE 174

146 listeners did report not hearing the [f] in the equal noise to token ratio and in the greater level of noise (Table 9-12). Table 9-12: Perception of non-strident fricatives in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [f] parf 0 0 2 0 1 0 Sum 0 0 2 0 1 0 Percentage Missed 0% 0% 33% 0% 17% 0% The difference between the perception of the strident and non-strident fricatives at 0dB was significant at a 90% confidence level. Preceding word-final stops, one listener did report not hearing [s] at dB (Table 9-13). Table 9-13: Perception of [s] preceding stops in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [s]+stop fosp 0 N/A 0 N/A 0 N/A fost 0 1 0 1 1 1 posk 0 N/A 0 N/A 0 N/A Sum 0 1 0 1 1 1 Percentage Missed 0% 17% 0% 17% 6% 17% It may be worth noting that in this instance, [s] was followed by a stop with the same place of articulation in the token ‘fost.’ Listeners also misperceived [s] in this same token as the alveopalatal []. 9.2.1.4. Fricatives in fast speech In fast speech, also the listeners always detected the presence of the word-final strident fricatives (Table 9-14).

PAGE 175

147 Table 9-14: Perception of strident fricatives in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [l]+fricative palse 0 0 0 0 0 0 palsh 0 0 0 0 0 0 Sum 0 0 0 0 0 0 Percentage Missed 0% 0% 0% 0% 0% 0% []+fricative barce 0 4 0 2 0 2 parsh 0 0 0 0 0 0 Sum 0 4 0 2 0 2 Percentage Missed 0% 33% 0% 17% 0% 17% Several speakers though did misperceive the [s] as [] in the token ‘barce.’ Considering the non-strident fricative, listeners again did report not detecting the word-final [f] in ‘parf’ (Table 9-15). Table 9-15: Perception of non-strident fricatives in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard parf 1 0 2 0 3 0 Sum 1 0 2 0 3 0 Percentage Missed 17% 0% 33% 0% 50% 0% Preceding a word-final stop in fast speech, [s] was always detected in all three levels of noise (Table 9-16). For the most part strident fricatives, unlike stops were always detected although listeners could not always discern the place of articulation. However, non-strident fricative posed some problems for perception, particularly in greater noise levels.

PAGE 176

148 Table 9-16: Perception of [s] preceding stops in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [s]+stop fosp 0 N/A 0 N/A 0 N/A fost 0 1 0 1 0 1 posk 0 N/A 0 N/A 0 N/A Sum 0 1 0 1 0 1 Percentage Missed 0% 17% 0% 17% 0% 17% 9.2.2. Perception of Sonorants Word-finally both nasals and [l] were found in the data. Since sonorants were longer and had a greater relative rms than the stops and [f] in this position, it was expected that they would be less likely than the obstruents to be missed or misperceived in this position. 9.2.2.1. Nasals in casual Speech Nasals occurred word-finally after [] and preceding word-final stops. Word-finally after [], nasals were always detected in the tokens ‘parn’ and ‘tarm’. In addition, they always correctly perceived the place of articulation of the nasal. Preceding a word-final voiceless stop was a different story (Table 9-17). Table 9-17: Perception of nasals preceding stops in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard nasal + stop fomp 1 0 3 0 2 0 pont 2 0 1 0 1 0 Sum 3 0 4 0 3 0 Percentage Missed 25% 0% 33% 0% 25% 0% In this position, speakers reported not hearing the nasal in all three levels of noise. The difference between perceiving a nasal preceding voiceless stops compared to word-finally

PAGE 177

149 Missed could be predicted at a 95% percentile in all three noise levels. This is consistent with the relative rms for nasals, which was greater following [] than it was preceding stops. Nasals were also a significantly greater proportion of the rhyme following [] than they were preceding stops. Interestingly though they were a significantly shorter proportion of the coda after []. However, this does not appear to have affected the nasals’ perceptibility. 9.2.2.2. Nasals in fast Speech In the fast speech also, the listeners also for the most part perceived the presence of the word-final nasals (Table 9-18). Table 9-18: Perception of word-final nasals in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [] + nasal tarm 1 0 1 0 0 0 parn 0 0 0 0 0 0 Sum 1 0 1 0 0 0 Percentage Missed 8% 0% 8% 0% 0% 0% One listener though did report not hearing the word-final [m] in ‘tarm’ in the +6dB and 0dB sound ratios. Again, though, listeners did have difficulty perceiving the nasals when they preceded word-final voiceless stops (Table 9-19). Table 9-19: Perception of nasals preceding stops in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard nasal + stop fomp 4 0 4 0 5 0 pont 1 0 2 0 1 0 Sum 5 0 6 0 6 0 Percentage 42% 0% 50% 0% 50% 0%

PAGE 178

150 In both casual and fast speech listeners had more difficulty hearing nasals word-medially, than they did word-finally. 9.2.2.3. Lateral approximant [l] in casual speech Following [], [l] was perceived by speakers when the amplitude of the token was greater than the amplitude of the noise (Table 9-20). Table 9-20: Perception of word–final [l] in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [l] parl 0 N/A 1 N/A 3 N/A Sum 0 N/A 1 N/A 3 N/A Percentage Missed 0% N/A 17% N/A 50% N/A However, when the noise levels increased the listeners had a difficult time perceiving [l] missing [l] half of the time. This agrees with the hypothesis that listeners would have a difficult time perceiving consonants when they are adjacent to a consonant that is close to them on the sonority hierarchy. When [l] occurred adjacent to a word-final obstruent though, the listeners did not have difficultly perceiving the presence of a consonant (Table 9-21). When listeners were given an option as to what [l] might be, two listeners misidentified [l] as [] in the token ‘palsh..’

PAGE 179

151 Table 9-21: Perception of [l] preceding obstruents in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [lt] pault 0 N/A 0 N/A 0 N/A Sum 0 N/A 0 N/A 0 N/A Percentage Missed 0% N/A 0% N/A 0% N/A [ld] fald 0 0 0 0 0 0 Sum 0 0 0 0 0 0 Percentage Missed 0% 0% 0% 0% 0% 0% [l] + fricative palse 0 0 0 0 0 0 palsh 0 0 0 2 0 0 Sum 0 0 0 2 0 0 Percentage Missed 0% 0% 0% 17% 0% 0% 9.2.2.4. Lateral approximant [l] in fast speech In fast speech, the listeners always perceived the presence of [l] word-finally after [] in the token ‘parl.’ In all of the presentations of this token, the listeners were able to detect the word-final [l]. As well, they always detected the presence of the consonant when it occurred before an obstruent. In this case, a listener again misperceived [l] as [] in ‘palsh’ (Table 9-22).

PAGE 180

152 Table 9-22: Perception of [l] preceding obstruents in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [lt] pault 0 N/A 0 N/A 0 N/A Sum 0 N/A 0 N/A 0 N/A Percentage Missed 0% N/A 0% N/A 0% N/A [ld] fald 0 0 0 0 0 0 Sum 0 0 0 0 0 0 Percentage Missed 0% 0% 0% 0% 0% 0% [l] + fricative palse 0 0 0 0 0 0 palsh 0 0 0 1 0 1 Sum 0 0 0 1 0 1 Percentage Missed 0% 0% 0% 8% 0% 8% 9.2.2.5. Rhotic approximant [] in casual and fast speech The English [] tended to be longer and have greater relative rms, so it was expected that of the sonorants it would be least likely to be not heard or misperceived. [] occurred before each of the other consonants (Table 9-23). Table 9-23: Tokens with [] Environment Tokens Environment Tokens Environment Tokens [] + vl stop fark, farp, gart []+strident fricative barce, parsh [] + nasal parn, tarm [] + vd stop farb, fard [f] parf [l] parl In addition [] occurs highest of the consonants on the sonority hierarchy, so given the hypothesis that consonants higher on the hierarchy should be more likely to be perceived, it was not a surprise to find that listeners always reported hearing [] in all of the levels

PAGE 181

153 of noise. In addition, when listeners were given an option of reporting a misperception of [] as [l], they never did. The same was true in fast speech. All speakers reported hearing [] in all contexts at all levels of noise. One listener misinterpreted [] in ‘tarm’ in the -6dB signal to noise ratio as and [l]. As in casual speech, this is not surprising given that []’s duration and relative rms in all of these positions. 9.2.3. Perception Dependent upon Manner of Articulation The second hypothesis of the perception experiment was that consonant higher on the sonority hierarchy would be more likely perceived than consonants lower on the sonority hierarchy in parallel environments. Overall word-finally stops, [l], and [f] posed the most problems for the listeners particularly in greater noise. Conversely, strident fricatives were always perceived word-finally, although at times they were identified as the wrong place of articulation. Preceding a word-final consonant, the consonants that presented the most problems for perception were stops and nasals. The following subsections detail the perception of the consonants in the different environments. 9.2.3.1. Preceding Stops The consonants that were found preceding voiceless stops were [s, p, m, n, l, ]. In retrospect, the list should have included [f] + stop tokens, such as ‘loft,’ to determine where the non-strident fricative would have fallen on the hierarchy. Given the difficulty detecting [f] word-finally, it is possible that it would have been difficult to detect in this environment, but that question will have to be left to future study.

PAGE 182

154 Preceding voiceless stops in the casual speech tokens, [] (three separate tokens0 and [l] (one token) were always perceived in all levels of noise. Of the three tokens with [s] before stops, [s] was perceived in nearly all cases except for once where it was not perceived at dB signal to noise ratio in the token ‘fost.’ Both nasals (two tokens) were missed in all three levels of noise several times. This gives the ranking in Table 9-24 where a high ranking means that a consonant was more likely to be not perceived preceding a stop. Table 9-24: Ranking of perceptibility of consonants preceding voiceless stops in casual speech +6dB, 0dB, -6dB [m, n] >> [s] >> [l], [] It is also noteworthy that [s] and [l] were both at times misperceived as another consonant while [] never was. In fast speech, the findings were essentially the same with the exception that the listeners always perceived [s] given the ranking in Table 9-25. Table 9-25: Ranking of perceptibility of consonants preceding voiceless stops in fast speech +6dB, 0dB, -6dB [m, n] >> [s], [l], [] 9.2.3.2. Preceding Fricatives Preceding fricatives both [l] and [] occurred. The presence of both was always predicted in this environment in all levels of noise. However, [l] was misperceived by speakers as an [] in the token ‘palsh.’ 9.2.3.3. Following [l] Following [l], voiced and voiceless stops occurred as well as strident fricatives. In this position the strident fricatives were always perceived in both casual and fast speech.

PAGE 183

155 In both the casual and fast speech production of ‘pault’ and fald,’ the listeners at times missed the word-final stops. In the casual speech they more often missed the voiceless stops. In fast speech, they more often missed the voiced stop. 9.2.3.4. Following [] All of the consonants considered in this study were found following [] (p, t, k, b, d, m, n, l) and they varied as to how well they were perceived in this position. A higher ranking represents a significance difference at an at least 80% confidence level with the exception of the strident fricatives and nasals. They were never missed, but the statistics did not show it to be significant (Table 9-26). Table 9-26: Ranking of perceptibility of consonants following [] in casual speech +6dB [d, b] >> [p, t, k], [s, ], [f], [m, n], [l] 0dB [f] >> [p, t, k], [d, b], [l] >> [m, n], [s, ] -6 dB [l] >> [p, t, k], [d, b], [f] >> [m, n], [s, ] In the fast speech tokens [l] is ranked low, in this case always being perceived after [] (Table 9-27). Table 9-27: Ranking of perceptibility of consonants following [] in fast speech +6dB [p, t, k] >> [f], [m, n] >> [d, b], [s, ], [l] 0dB [f] >> [p, t, k] >> [m, n] >> [d, b], [l] [s, ] -6 dB [f] >> [p, t, k] >> [d, b] >> [m, n], [s, ], [f] As in the casual speech productions though, voiced and voiceless stops as well as [f] were the consonants were not perceived following [] more often than the strident fricatives and the nasals.

PAGE 184

156 9.3. Russian Tokens After the English-speaking participants completed the previous perception experiment, they were asked to complete a second experiment where they listened to the tokens produced by the native Russian speaker. Having the English speakers listen to Russian tokens is of course problematic. Most obviously, in Russian there is no []. Instead, there is the trill. During debriefing with the English subjects, at least two listeners remarked that the speaker rolled his [r]’s. When asked if they thought this made him harder to understand they remarked that it was ‘funny’ and they wanted to know where he was from. This relieved some anxiety that listeners would not know how to hear this ‘exotic’ sound. Despite what problems though having participants listen to a language they did not know, they were presented with these tokens to determine if there were any patterns in the participants misunderstanding of tokens. 9.3.1. Perception of Obstruents 9.3.1.1. Stops in Casual Speech One way that the Russian speaker differed from the English speakers was that his stop consonants were less likely to be weakened word-finally, so it was expected that his stops would be less likely to be misunderstood or unheard. However, this did not prove to be true (Table 9-28).

PAGE 185

157 Table 9-28: Perception of word-final voiceless stops in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard Stop + [t] [fopt] 0 N/A 0 N/A 0 N/A [bakt] 0 N/A 0 N/A 1 N/A Sum 0 N/A 0 N/A 1 N/A Percentage Missed 0% N/A 0% N/A 8% N/A [s] +stop [fast] 0 0 0 0 1 0 [pask] 1 2 0 1 0 2 Sum 1 2 0 1 1 2 Percentage Missed 8% 17% 0% 8% 8% 17% nasal +stop [famp] 2 0 0 1 0 0 [pant] 3 N/A 5 N/A 2 N/A Sum 5 0 5 1 2 0 Percentage Missed 42% 0% 42% 17% 17% 0% [l] + vl stop [palt] 2 0 2 1 5 0 Sum 2 0 2 1 5 0 Percentage Missed 33% 0% 33% 17% 83% 0% [r] + vl stop farp 0 0 0 0 2 1 gart 1 0 0 0 0 0 fark 1 2 1 3 0 4 Sum 2 2 1 3 2 5 Percentage Missed 0% 44% 22% 11% 0% 22% As with the English tokens, the listeners had a difficult time perceiving the word-final stops and often when they did detect their presence, they misidentified their place of articulation. The rankings of the perception of the environments based upon whether or not the listeners could perceive the stops in them are shown in Table 9-29.

PAGE 186

158 Table 9-29: Ranking of perception of word-final stops by environment in casual speech +6dB [m, n __] >> [l__] >> [s__] >> [p__] >> [r__] 0dB [m, n __] >> [l__] >> [r__] >> [s__], [p__] -6dB [l__] >> [m, n __] >> [s__], [p__] >>[r__] At dB signal to noise ratio, the rankings is similar to that of the perception of the English tokens. The listeners had the most difficulty when [t] followed [l] and the least difficulty when the stops followed [r]. The middle rankings were not the same with the listeners having more difficulty perceiving stops after nasals. Similarly, at +6dB and 0dB, the listeners had more difficulty hearing the stops following nasals in the Russian tokens than they did in the English tokens (Table 9-30). Table 9-30: Perception of [p] and [k] preceding [t] in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [p] + stop [fopt] 2 0 1 3 2 2 [bakt] 1 0 2 3 3 2 Sum 3 0 3 6 5 4 Percentage Missed 25% 0% 25% 50% 42% 33% As with the English tokens, the listeners did not always perceive the word-final voiced stops (Table 9-31). Table 9-31: Perception of voiced stops +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [r] + stop [farb] 0 2 0 3 0 3 [nard] 2 N/A 2 N/A 1 N/A Sum 2 2 2 3 1 3 Percentage Missed 17% 33% 17% 50% 8% 50% The perceptions of the word-final English and Russian stops were very similar. The listeners had difficulty with both voiced and voiceless stops. In addition, for both

PAGE 187

159 languages, in two stop clusters, the listeners tended to choose the spelling representing the second stop rather than the first when they detected only one stop. 9.3.1.2. Stops in Fast Speech It was expected that there would be more difficulty hearing the tokens perceived at the fast speaking rate than at the casual speaking rate (Table 9-32). Table 9-32: Perception of word-final voiceless stops in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard stop + [t] [fopt] 0 N/A 0 N/A 0 N/A [bakt] 4 N/A 1 N/A 1 N/A Sum 4 N/A 1 N/A 1 N/A Percentage Missed 33% N/A 8% N/A 8% N/A [s] +stop [fast] 1 0 3 0 1 0 [pask] 0 1 1 1 0 1 Sum 1 1 4 1 1 1 Percentage Missed 8% 8% 33% 8% 8% 8% nasal +stop [famp] 0 0 0 0 0 0 [pant] 0 N/A 2 N/A 1 N/A Sum 0 0 2 0 1 0 Percentage Missed 0% 0% 17% 0% 8% 0% [l] + stop pault 1 0 3 1 2 0 Sum 1 0 3 1 2 0 Percentage Missed 17% 0% 50% 17% 33% 0% [r] + stop farp 0 0 0 0 0 0 fark 3 2 2 2 0 3 gart 2 1 2 2 0 1 Sum 5 3 4 4 0 4 Percentage Missed 28% 17% 22% 22% 0% 22%

PAGE 188

160 The rankings of the environments based upon the difficulty of perceiving the word-final stops are found in Table 9-33. Table 9-33: Ranking of perception of word-final stops by environment in fast speech +6dB [p __] >> [r__] >> [l__] >> [s__] >> [m, n__] 0dB [l __] >> [s__] >> [r__] >> [m, n__], [p__] -6dB [p, k] >> [s] >> [r] >> [m, n], [l] The ranking for the casual and fast speech were identical in dB signal to noise ratio. In both cases, it became difficult for the listeners to perceive [t] when it followed [l] and it was easiest to perceive the stops when following [r]. In the fast speech tokens, the listeners did not detect the stops that preceded stops (Table 9-34). Table 9-34: Perception of [p] preceding [t] in casual speech in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [p] + stop [fopt] 0 0 0 0 0 0 [bakt] 0 1 0 3 2 1 Sum 0 1 0 3 2 1 Percentage Missed 0% 8% 0% 25% 17% 8% In the token [bakt], they also tended to misperceive [k] as [r]. The casual speech perceptions of these tokens were different in that the listeners tended to choose the spelling with the [t] rather than the preceding consonant when they detected only one stop. Reviewing the tokens with word-final voiced stops, again the listeners had problems with these tokens (Table 9-35).

PAGE 189

161 Table 9-35: Perception of word-final voiced stops in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [r] + vd. stop [farb] 0 1 0 3 0 2 [nard] 1 0 0 1 0 2 Sum 1 1 0 4 0 4 Percentage Missed 8% 17% 0% 67% 0% 67% In this case, the listeners either did not perceive the word-final stops or misperceived the place. 9.3.1.3. Fricatives in Casual Speech The findings were the same for the strident fricatives in Russian as they were for the strident fricatives in English. As Table 9-36 shows, the listeners always detected the presence of the fricative. Table 9-36: Perception of strident fricatives in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [r] + fricative [kars] 0 0 0 0 0 0 [par] 0 1 0 1 0 1 Sum 0 1 0 1 0 1 Percentage Missed 0% 8% 0% 8% 0% 8% Also similar to the English, the strident fricative was also misperceived as the wrong place of articulation. In this case, [] was misidentified as [s]. Also, as with the English tokens, the listeners had a difficult time perceiving the presence of [f] (Table 9-37).

PAGE 190

162 Table 9-37: Perception of non-strident fricatives in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [rf] parf 2 1 1 1 1 1 Sum 2 1 1 1 1 1 Percentage Missed 33% 17% 17% 17% 17% 17% What is different from the perception of the English fricatives is the difficulty that the listeners had with [s] preceding a voiceless stop. At all three signal to noise ratios, listeners reported not hearing [s] (Table 9-38). Table 9-38: Perception of [s] preceding stops in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [s] + stop [fast] 1 0 2 0 3 0 [pask] 0 N/A 0 N/A 0 N/A Sum 1 0 2 0 3 0 Percentage Missed 8% 0% 17% 0% 25% 0% 9.3.1.4. Fricatives in Fast Speech In fast speech as in casual speech, the listeners did not have difficulty perceiving the presence of word-final strident fricatives (Table 9-39). Table 9-39: Perception of strident fricatives in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [r] + fricative [kars] 0 0 0 0 0 1 [par] 0 0 0 1 0 1 Sum 0 0 0 1 0 2 Percentage Missed 0% 0% 0% 8% 0% 17%

PAGE 191

163 Again, listeners did misidentify the place of articulation of [] as [s], particularly when the noise level was greater. Also, as in the casual speech tokens, the listeners were more likely to miss the presence of word-final [f] (Table 9-40). Table 9-40: Perception of non-strident fricatives in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [rf] [parf] 2 1 0 1 2 0 Sum 2 1 0 1 2 0 Percentage Missed 33% 17% 0% 17% 33% 0% As in the casual speech, a listener did report not detecting the presents of [f] in [fast] in the dB signal to noise ratio (Table 9-41). Table 9-41: Perception of [s] preceding stops in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [s] + stop [pask] 0 0 0 0 0 0 [fast] 0 1 0 0 1 0 Sum 0 1 0 0 1 0 Percentage Missed 0% 17% 0% 0% 8% 0% As with the English tokens, the listeners again had little difficulty detecting the strident fricatives regardless of their environment. However, when given the opportunity, they did misidentify the place of articulation of the fricative. 9.3.2. Perception of Sonorants 9.3.2.1. Nasals in Casual Speech Nasals occurred word-finally after [] and preceding word-final stops (Table 9-42). In the casual speech tokens, the listeners had some difficulty perceiving the wordfinal nasals, and this difficulty seemed to increase with increased noise.

PAGE 192

164 Table 9-42: Perception of word-final nasals in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [l] + nasal [polm] 0 1 1 2 2 1 Sum 0 1 1 2 2 1 Percentage Missed 0% 17% 17% 33% 33% 17% [r] + nasal [torm] 0 N/A 1 N/A 1 N/A 0 1 0 0 1 1 Sum 0 1 1 0 2 1 Percentage Missed 0% 17% 8% 0% 17% 17% The listeners though did not have the same problem with the nasals preceding stops in the tokens [famp] and [pant]. In these tokens the listeners always perceived the presence of the nasal, and only once misidentified the place of a nasal in the 0dB presentation of the token [famp]. This is the opposite of what was found with the English tokens. In that case, it was the nasals preceding the stops that were the problem. In the Russian data, the nasals were a smaller proportion of the coda preceding stops than they were word-finally after [l] and [r] (43%, 75%, and 72% respectively). The relative rms was greater though preceding stops than it was after [r] or [l] (.685 before stops, .475 after [l], and .421 following [r]). Perhaps it was this greater relative rms that helped the listeners hear the nasals preceding stops. 9.3.2.2. Nasals in Fast Speech In fast speech, the listeners again had even more difficulty perceiving the nasal following [r] (Table 9-43).

PAGE 193

165 Table 9-43: Perception of word-final nasals in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [l] + nasal [polm] 0 4 0 3 0 3 Sum 0 4 0 3 0 3 Percentage Missed 0% 67% 0% 50% 0% 50% [r] + nasal [torm] 1 0 1 0 2 0 0 0 2 0 1 0 Sum 1 0 3 0 3 0 Percentage Missed 8% 0% 25% 0% 25% 0% This is consistent with the acoustic findings since the relative rms and the proportion of the coda of the nasals decreased from casual to fast speech. Also that the listeners were more likely to perceive [m] following [l] in fast speech than they were in casual speech is consistent with the acoustic findings as the relative rms of [m] increase following [l] from casual to fast speech. Preceding stops, the listeners were less likely to perceive the nasals preceding stops in fast speech than they were in casual speech. In addition, this difficulty increased with increased noise (Table 9-44). Table 9-44: Perception of nasals preceding stops in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard nasal + stop [famp] 1 0 2 0 2 0 [pant] 0 0 0 0 0 0 Sum 1 0 2 0 2 0 Percentage Missed 8% 0% 17% 0% 17% 0% This is consistent with the acoustic analysis of the nasals preceding stops since the relative rms of the nasals decreased from casual to fast speech. Although the listeners are

PAGE 194

166 able to detect the nasals the majority of the time, they do seem to cause more problems for perception than the strident fricatives. 9.3.2.3. Lateral Approximant [l] in Casual and Fast Speech In casual speech, the listeners always detected the [l] in the token [palt] (Table 9-45). However, they did not always detect the [l] in [polm]. Table 9-45: Perception of [l] preceding consonants in casual speech +6dB 0dB -6dB [lt] [palt] 0 N/A 0 N/A 0 N/A Sum 0 N/A 0 N/A 0 N/A Percentage Missed 0% N/A 0% N/A 0% N/A [lm] [polm] 1 N/A 1 N/A 2 N/A Sum 1 N/A 1 N/A 2 N/A Percentage Missed 17% N/A 17% N/A 33% N/A In the fast speech tokens, two of the listeners did not always detect the presence of [l] preceding [t] (Table 9-46). NEL6 missed the [l] in +6dB and 0dB. NEL3 missed the [l] in dB. That participants would miss the fast speech [l], but not the casual speech [l], might reflect that [l]’s rhyme duration decreased by one half from casual to fast speech (30% to 14 %). Its relative rms increased (.891 to .972) and the relative rms of [t] decreases (.156 to .088), but still [l] was not detected by these listeners. This suggests that the shorter duration may have played a key role in the listeners’ inability to detect [l]. The [l] in [polm] again causes some problems for the listeners, although interestingly they report detecting [l] in dB signal to noise ratio. That the listeners have fewer problems with the fast speech [l] than with the casual speech [l] again may relate to the acoustic findings. In this case the proportion of the rhyme devoted to [l] increased from casual to fast speech (20% to 23%).

PAGE 195

167 Table 9-46: Perception of [l] preceding consonants in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [lt] [palt] 1 N/A 1 N/A 1 N/A Sum 1 N/A 1 N/A 1 N/A Percentage Missed 17% N/A 17% N/A 17% N/A [lm] [polm] 2 N/A 1 N/A 0 N/A Sum 2 N/A 1 N/A 0 N/A Percentage Missed 33% N/A 17% N/A 0% N/A 9.3.2.4. Alveolar Trill [r] in Casual and Fast Speech Unlike the English [], the native English speakers had difficulty perceiving the Russian [r] when it preceded word-final stops (Table 9-47). In the other environments, listeners did not have difficulty detecting the presence of the consonant, but some listeners misidentified [r] as [l] when they were given the option in the multiple choices. This misidentification of [r] though is not surprising since [r] was not native to their dialect. The ranking of the environment based on whether or no the listeners detected the presence of [r] is found in Table 9-48. At signal to noise ration of 0dB, the listeners always detected the presence of the consonant.

PAGE 196

168 Table 9-47: Perception of [r] in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [r] + vl stop [farp] 1 0 0 1 0 2 [gart] 0 N/A 0 N/A 0 N/A [fark] 1 N/A 0 N/A 1 N/A Sum 2 0 0 1 1 2 Percentage Missed 11% 0% 0% 17% 6% 33% [r] + vd stop [farb] 0 0 0 0 0 0 [nard] 0 0 0 1 0 0 Sum 0 0 0 1 0 0 Percentage Missed 0% 0% 0% 8% 0% 0% [r] + strident fricative [kars] 0 1 0 1 0 0 [par] 0 N/A 0 N/A 0 N/A Sum 0 1 0 1 0 0 Percentage Missed 0% 17% 0% 17% 0% 0% [r] + non-strident fricative [parf] 0 N/A 0 N/A 0 N/A Sum 0 N/A 0 N/A 0 N/A Percentage Missed 0% N/A 0% N/A 0% N/A [r] + nasal [torm] 0 0 0 1 0 1 [gorn] 0 N/A 0 N/A 0 N/A Sum 0 0 0 1 0 1 Percentage Missed 0% 0% 0% 17% 0% 17% Table 9-48 :Ranking of perception of word-final stops by environment in casual speech 6dB, dB [__p, t, k] >> [__d, b], [__s, ], [__f], [__m, n] In fast speech, the listeners had a more difficulty perceiving [r] (Table 9-49). In addition, the inability to perceive [r] increased with speaking rate.

PAGE 197

169 Table 9-49: Perception of [r] in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [r] + vl stop [farp] 0 0 0 0 1 0 [gart] 1 N/A 1 N/A 2 N/A [fark] 0 N/A 0 N/A 0 N/A Sum 1 0 1 0 3 0 Percentage Missed 6% 0% 6% 0% 17% 0% [r] + vd stop [farb] 0 0 0 0 0 0 [nard] 1 0 1 0 2 1 Sum 1 0 1 0 2 1 Percentage Missed 8% 0% 8% 0% 17% 8% [r] + strident fricative [kars] 0 0 1 2 0 1 [par] 0 N/A 0 N/A 0 N/A Sum 0 0 1 2 0 1 Percentage Missed 0% 0% 8% 33% 0% 17% [r] + non-strident fricative [parf] 0 N/A 0 N/A 0 N/A Sum 0 N/A 0 N/A 0 N/A Percentage Missed 0% N/A 0% N/A 0% N/A [r] + nasal [torm] 0 0 1 0 1 0 [gorn] 0 N/A 0 N/A 1 N/A Sum 0 0 1 0 2 0 Percentage Missed 0% 0% 8% 0% 17% 0% 9.3.3. Perception Dependent upon Manner of Articulation 9.3.3.1. Preceding stops In the Russian, as in the English, the consonants that were found preceding voiceless stops were [s, p, m, n, l, ].

PAGE 198

170 Preceding voiceless stops in the casual speech tokens, [r] (three separate tokens) and [l] (one token) were always perceived in all levels of noise. There were two tokens with [s] preceding voiceless stops and two with nasals preceding stops. The perception of these consonants before the stops was different from the perception of these consonants in the English tokens. The rankings of the English listeners misses of the consonants before voiceless stops are found in Table 9-50. Table 9-50: Ranking of perceptibility of consonants preceding voiceless stops in casual speech +6dB [p, k] >> [r] >> [s] >> [m, n], [l] 0dB [p, k] >> [s] >> [m, n], [l], [r] -6dB [p, k] >> [s] >> [r] >> [m, n], [l] In the Russian tokens, the nasals and [l] were always perceives in all levels of noise. Some listeners missed these sounds when listening to the English tokens. Also, the Russian trill was ranked higher than the English []. This was not unexpected since the trill is not native to these participants’ dialect of English. The fast speech findings were more similar to the perception of the English tokens (Table 9-51). Table 9-51: Ranking of perceptibility of consonants preceding voiceless stops in fast speech +6dB [l] >> [r] >> [p, k], [s], [m, n] 0dB [m, n], [l] >> [r] >> [p, k], [s] -6dB [p, k], [m, n] [l], [r] >> [s] The decreased ability to perceive the nasal in fast speech may be the result of the decrease in relative rms from casual to fast speech (.852 to .685) while the adjacent stop’s relative rms increased (.280 to .314). Conversely, the proportion of the coda devoted to the nasal though actually increased from casual to fast (43% to 48%), so the added duration in relation to the stop did not seem to aid perception.

PAGE 199

171 9.3.3.2. Following [r] In the Russian data nearly all of the consonants being considered were found following [r] (p, t, k, b, d, m, n). These consonants varied as to how well they were perceived in this position (Table 9-52). Table 9-52: Ranking of perceptibility of consonants following [r] in casual speech +6dB [f] >> [b, d] >> [p, t, k] >> [s, ], [m, n] 0dB [f] >> [m, n] >> [b, d] >> [p, t, k] >> [s, ] -6 dB [f], [m, n] >> [p, t, k] >> [b, d] >> [s, ] Most notable about these rankings is that the strident fricatives are always ranked low. The listeners always detected their presence in the tokens. Also, in the decreased noise, levels the listeners had more difficulty detecting the presence of the nasals following [r] than they did the stops. In fast speech, there were some similarities with and differences from the casual speech Table 9-53. Table 9-53: Ranking of perceptibility of consonants following [r] in casual speech +6dB [f] >> [p, t, k] >> [m, n], [b, d] >> [s, ] 0dB [m, n] >> [p, t, k] >> [s, ], [f], [b, d] -6dB [f], [m, n] >> [p, t, k], [b, d], [s, ] As with the casual speech, the listeners had difficulty with [f] and the nasals particularly in the increased noise. The listeners had less difficulty with the stops in fast speech than they did in casual speech. This is consistent with the fact the relative rms and the duration of the stops increased from casual to fast speech (0.157 to 0.161). 9.4. Conclusions The English listeners had similar difficulties perceiving the English consonants as they did perceiving the Russian tokens. The consonants lower on the sonority hierarchy, stops, caused the most amount of trouble for perception. These stops were more difficult

PAGE 200

172 adjacent to consonants that they are closer to on the sonority hierarchy, i.e. [s] and other stops. Most interestingly in nasal+stop clusters, listeners often heard the stop rather than the nasals. While the perception of the English and Russian clusters were similar, they did differ in terms of the perception of the [] and [r]. This is not surprising though given the greater duration and relative rms of [] over [r]. In general, the listeners tended to have the least difficulty perceiving the consonants that were longer and had a greater relative rms.

PAGE 201

CHAPTER 10 RUSSIAN PERCEPTION 10.1. Introduction A perception experiment was undertaken with Russian listeners to determine if the relative rms and duration of a consonant correlated to its perceptibility. Three native speakers of Russian (two women and one man) listened to both the English and Russian tokens in the phrase ‘Skazhite, pozhalujsta, _____ vslux’ (‘Say, please, _____ out loud’). The phrases were then embedded in three levels of pink noise (+6db, 0dB, -6db). Listeners took a multiple choice test where they were told to choose the word that best sounded like the word that they heard. For details on how the perception experiment was designed, see chapter 4. When data collection was completed, responses were coded as all consonants in cluster perceived, one or more consonants not perceived, or one or more consonants misperceived. The coding and the formatting of the tables in this chapter are the same as those for the previous chapter. The remainder of this chapter details the finding as they result to each consonant. The findings from the Russian speakers were similar to the findings from the English speakers. The Russian speakers had difficulty perceiving word-final stops. They also had little trouble perceiving the strident fricatives. They differed from the English listeners in their perception of [r] and the nasals. Unexpectedly, they had difficulty perceiving clusters with the trill [r]. The English listeners did not have as much difficulty with these same clusters. The Russian listeners though did not have trouble perceiving 173

PAGE 202

174 nasals preceding stops. The rest of this chapter outlines the Russian listeners perception of the tokens produced by the Russian and English speakers. 10.2. Russian Tokens The productions of the Russian tokens were different from the English tokens. As described in Chapter 7 and Chapter 8, the duration and relative rms of the consonants in were not the same in the two languages. This may relate to the differences in the perception of the Russia and English tokens. 10.2.1. Perception of Obstruents 10.2.1.1. Perception of stops in casual speech Obstruents in Russian differed from the obstruents in English. In particular, the stops in Russian were more likely to hold on to their duration as speaking rate increased. This did not, though, ensure that the listeners perceived the word-final stops (Table 10-1). Interestingly, the word-final stops were less often perceived were less often perceived in the +6dB signal to noise ratio than were the stops in +6dB. The ranking of the environments based on the perception of the word-final stops is found in Table 10-2. The rankings are similar to the English perception of the English tokens in that the perception of [t] after [l] became highest ranked with increase in noise. It is different though in that the increase in increased the perception of the word-final stops. Also the like ranking of following [r] in the hierarchy may be misleading one speaker responded ‘none’ to the token [part] at 0dB in casual speech and another coded [part] as ‘none’ at –6dB in casual speech.

PAGE 203

175 Table 10-1: Perception of word-final voiceless stops in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard stop + [t] [fopt] 1 N/A 2 N/A 0 N/A [bakt] 2 N/A 1 N/A 0 N/A Sum 3 N/A 3 N/A 0 N/A Percentage Missed 50% N/A 50% N/A 0% N/A [s] +stop [fast] 0 0 1 2 0 1 [pask] 0 1 1 1 0 2 Sum 0 1 2 3 0 3 Percentage Missed 0% 17% 33% 50% 0% 50% nasal +stop [pant] 2 N/A 1 N/A 0 N/A [famp] 0 0 0 0 0 0 Sum 2 0 1 0 0 0 Percentage Missed 33% 0% 17% 0% 0% 0% [l] + vl stop [palt] 1 N/A 2 N/A 1 N/A Sum 1 N/A 2 N/A 1 N/A Percentage Missed 33% N/A 67% N/A 33% N/A [r] + vl stop [part] 0 1 0 0 0 0 [fark] 0 3 1 1 0 2 [farp] 0 0 1 0 0 0 Sum 0 4 2 1 0 2 Percentage Missed 0% 44% 22% 11% 0% 22% Table 10-2: Ranking of perception of word-final stops by environment in casual speech +6dB [p, k__] >> [m, n __], [l__] >> [s__], [r__] 0dB [l__] >> [p, k__] >> [s__] >> [r__] >> [m, n__] -6dB [l__] >> [p, k__], [s__], [m, n __], [r__] The stops preceding a word-final stop were less perceived increased noise (Table 10-3). Adding the perception of [p] into the db ranking above give the ranking in Table 10-4.

PAGE 204

176 Table 10-3: Perception of [p] and [k] preceding [t] +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [p] + stop [bakt] 0 0 0 0 0 3 [fopt] 0 0 0 0 2 0 Sum 0 0 0 0 2 3 Percentage Missed 0% 0% 0% 0% 33% 50% Table 10-4: Ranking of perception of word-final voiceless stops by environment in casual speech -6dB [__t] >> [l__] >> [p, k__], [s__], [m, n __], [r__] Considering the voiced stops, in all cases the native Russian speakers were able to perceive the stops in the token [fald]. The native English speakers were not always able to perceive the stops in this environment. 10.2.1.2. Stops in fast speech In fast speech, the listeners were more likely to perceive the word-final voiceless stops (Table 10-5). This varied from the perception of the English tokens by the English speakers who did not always perceive the word-final stops in fast speech. However, this is consistent with the findings in the acoustic analysis. While English stops bursts and durations decreased with an increase in speaking rate, the Russian stops became stronger with increases in speaking rate. Even though these listeners detected the presence of the stops, they did often misidentify the place of articulation of the stops, particularly after [r]. The ranking of the environments dependent upon the ability to perceive the word-final stops are found in Table 10-6.

PAGE 205

177 Table 10-5: Perception of word-final voiceless stops in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard stop + [t] [bakt] 1 N/A 0 N/A 0 N/A [fopt] 0 N/A 0 N/A 0 N/A Sum 1 N/A 0 N/A 0 N/A Percentage Missed 17% N/A 0% N/A 0% N/A [s] +stop [fast] 0 1 1 0 0 0 [pask] 0 0 0 3 0 1 Sum 0 1 1 3 0 1 Percentage Missed 0% 17% 17% 50% 0% 17% nasal +stop pont 1 N/A 2 N/A 0 N/A fomp 0 0 0 0 0 0 Sum 1 0 1 1 0 0 Percentage Missed 17% 0% 33% 33% 0% 0% [l] + stop pault 0 N/A 0 N/A 0 N/A Sum 0 N/A 0 N/A 0 N/A Percentage Missed 0% N/A 0% N/A 0% N/A [r] + stop part 0 2 0 0 0 0 fark 0 2 0 3 0 3 farp 0 0 0 0 0 1 Sum 0 4 0 3 0 4 Percentage Missed 0% 44% 0% 33% 0% 44% Table 10-6: Ranking of perception of word-final stops by environment in fast speech +6dB [p, k__], [m, n __] >> [s__], [l__], [r__] 0dB [n__] >> [s__] >> [p, k__], [m, n__], [l__], [r__] -6dB [p, k__], [s__], [m, n __], [l__], [r__] In fast speech, a stop preceding [t] was only not detected once in the 0dB signal to noise ratio (Table 10-7).

PAGE 206

178 Table 10-7: Perception of [p] preceding [t] in casual speech in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [p] + stop [bakt] 0 0 1 0 0 0 [fopt] 0 0 0 0 0 0 Sum 0 0 1 0 0 0 Percentage Missed 0% 0% 17% 0% 0% 0% This is different from how the English speakers responded to these same tokens. In casual speech, they often did not perceive the [p] preceding [t] instead opting for the spelling with the ‘t’ rather than the ‘p’ or [k]. Considering the voiced stops word-finally, one speaker responded that they did not detect the word-final [b] at dB in fast speech (Table 10-8). Table 10-8: Perception of word-final voiced stops in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [r] + vd. stop [farb] 0 N/A 0 N/A 1 N/A Sum 0 N/A 0 N/A 1 N/A Percentage Missed 0% N/A 0% N/A 33% N/A In addition, another listener responded ‘none’ to the same token at dB signal to noise ratio suggesting that something about this cluster may have become more difficult to perceive with the increase in speaking rate. In general, then, the perception of the voiceless stops in fast speech did not become more difficult with an increase in noise. On the contrary, the listeners were more likely to detect the stop in the increased noise. However, the voiced stop cluster [rb] became more difficult with increased noise.

PAGE 207

179 10.2.1.3. Fricatives in casual speech The presence of the word-final strident fricatives was always detected by the Russian speakers (Table 10-9). Table 10-9: Perception of strident fricatives in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [r] + fricative [kars] 0 0 0 2 0 1 [par] N/A 0 N/A 0 0 N/A Sum 0 0 0 2 0 1 Percentage Missed 0% 0% 0% 67% 0% 33% When given the opportunity, though, the Russian speakers misidentified [s] as []. The same also held true for the non-strident fricative [f] as is shown in the Table 10-10. Table 10-10: Perception of non-strident fricatives in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [rf] parf 0 0 0 1 0 0 Sum 0 0 0 1 0 0 Percentage Missed 0% 0% 0% 33% 0% 0% Here the listeners always sensed the presence of [s], but one listener did misidentify the [f] as an [s] in one instance. Note that this is different from the native English speakers perception of [f] which became more difficult with the increase in noise. Considering [s] preceding the stops in the tokens [fast] and [pask], the listeners always perceived, the listeners always perceived its presence in all noise levels. The perception of the Russian strident fricatives was similar to the perception of the English strident fricatives by the English speakers. In most cases, the listeners detected their presence in both languages, but occasionally misidentified their place of articulation. The

PAGE 208

180 non-strident fricative [f] though was different in Russian than it was in English. While the native English speakers did not always detect it, the native Russian speakers did, although they did at times misidentify the place of articulation. 10.2.1.4. Fricatives in fast speech The perception of the strident fricative in fast speech was similar to the perception in casual speech, with one speaker not perceiving the [s] in dB in the token [kars] (Table 10-11). Table 10-11: Perception of strident fricatives in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [r] + fricative [kars] 0 0 0 0 1 0 [par] 0 0 0 0 0 0 Sum 0 0 0 0 1 0 Percentage Missed 0% 0% 0% 0% 17% 0% As for the non-strident fricative [f], one listener did report not detecting the non-strident fricative [f] in fast speech at 0dB. In addition listeners misperceived it as an [s] in all three noise levels (Table 10-12). Table 10-12: Perception of non-strident fricatives in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [rf] [parf] 0 1 1 1 0 1 Sum 0 1 1 1 0 1 Percentage Missed 0% 33% 33% 33% 0% 33% The listeners always perceived [s] preceding the voiceless stops in [fast] and [pask]. The strident fricatives seemed to produce fewer problems for perception by the native speakers of Russian. This is consistent with how the English listeners responded to

PAGE 209

181 these same tokens. It is also what one would predict given the findings from the acoustic analysis, since the strident fricatives had longer duration and greater rms than [f]. 10.2.2. Perception of Sonorants 10.2.2.1. Nasals in casual speech Nasals occurred word-finally after [r] and [l] and preceding word-final stops. Word-finally listeners always perceived the presence of the nasal. Although when given the opportunity, [m] was at times misperceived as [n] in [torm] (Table 10-13). Table 10-13: Perception of word-final nasals in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [l] + nasal [polm] 0 N/A 0 N/A 0 N/A Sum 0 N/A 0 N/A 0 N/A Percentage Missed 0% N/A 0% N/A 0% N/A [r] + nasal [torm] 0 2 0 1 0 2 Sum 0 2 0 1 0 2 Percentage Missed 0% 67% 0% 33% 0% 67% Preceding word-final stops listeners always perceived the presence of the nasals and never misperceived the place of articulation in the tokens [famp] and [pant]. In this position, in order for speakers to misidentify the place of articulation of the nasal, they would also have had to misperceive the place of articulation of the stop. 10.2.2.2. Nasals in fast speech As in the casual speech, the listeners had little difficulty detecting the presence of the word-final nasals. One speaker though responded that there was no word-final nasal in the token [torm] in the dB signal to noise ratio (Table 10-14).

PAGE 210

182 Table 10-14: Perception of word-final nasals in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [l] + nasal [polm] 0 N/A 0 N/A 0 N/A Sum 0 N/A 0 N/A 0 N/A Percentage Missed 0% N/A 0% N/A 0% N/A [r] + nasal [torm] 0 2 0 0 1 1 Sum 0 2 0 0 1 1 Percentage Missed 0% 67% 0% 0% 33% 33% Preceding word-final stops, listeners never missed the nasals when they preceded word-final stops in the tokens [pant] and [famp]. However, two listeners did answer ‘none’ to the +6db presentation of [pant]. The native English speakers did not respond in the same way to those Russian tokens. Instead, they had trouble with token [famp] where listeners reporter there being no [m]. However, they had no difficulty identifying [pant]. 10.2.2.3. Lateral approximant [l] in casual and fast Speech The native Russian speakers responded to the Russian [l] much in the same way that the native English speakers. In the casual speech tokens they listeners replied that there was no [l] in the casual speech production of [palt] (Table 10-15). Then in the Fast speech production, the presence of [l] was noticed, but it was misidentified as [r]. Similarly, [l] was detected in the casual speech production of [polm], but not necessarily in the fast speech productions (Table 10-16). Again, the Russian speakers had the same difficulties with the Russian [l] as the native English speakers did. Remember from Chapter 9 that [l] was a greater proportion

PAGE 211

183 of the rhyme in fast speech [palt] than in casual speech [palt] and [l] was smaller proportion of the rhyme in fast speech [polm] than in casual speech [polm]. Table 10-15: Perception of [l] preceding obstruents in casual speech +6dB 0dB -6dB [lt] Not Heard Misheard Not Heard Misheard Not Heard Misheard [palt] 1 0 0 0 1 0 Sum 1 0 0 0 1 0 Percentage Missed 33% 0% 0% 0% 33% 0% [lm] [polm] 0 1 0 0 0 0 Sum 0 1 0 0 0 0 Percentage Missed 0% 33% 0% 0% 0% 0% Table 10-16: Perception of [l] preceding obstruents in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [lt] [palt] 0 1 0 0 1 0 Sum 0 1 0 0 1 0 Percentage Missed 0% 33% 0% 0% 33% 0% [lm] [polm] 1 0 0 1 1 1 Sum 1 0 0 1 1 1 Percentage Missed 33% 0% 0% 33% 33% 33% 10.2.2.4. Alveolar trill in casual and fast speech The native speakers of English had some difficulty perceiving the presence of the Russian [r] in casual speech when it preceded a voiceless stop. The Russian speakers also had some problems with the [r] tokens. The speakers never responded that they did not perceive [r] or that they misperceived [r], but two of the listeners often replied ‘none’ to these tokens. What is not certain is why they did this, but these tokens did pose some problems. The tokens that they had the most difficulty with were ‘parf,’ ’part,’ ‘farb’ and

PAGE 212

184 ’parsh’. The native speakers of English did not have the same problems with these tokens, despite that the Russian trill was not native to their language. As in the casual speech, for the most part listeners did perceive the presence of [r] (Table 10-17). Table 10-17: Perception of [r] in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [r]+vl stop [Farp] 0 N/A 0 N/A 1 N/A [Part] 0 N/A 0 N/A 0 N/A [Fark] 0 N/A 0 N/A 0 N/A Sum 0 N/A 0 N/A 1 N/A Percentage Missed 0% N/A 0% N/A 11% N/A [r]+vd stop [farb] 0 0 0 0 1 0 Sum 0 0 0 0 1 0 Percentage Missed 0% 0% 0% 0% 33% 0% [r]+strident fricative [kars] 0 0 0 0 0 0 [par] 0 0 0 0 0 1 Sum 0 0 0 0 0 1 Percentage Missed 0% 0% 0% 0% 0% 33% [r]+non-strident fricative [parf] 0 N/A 0 N/A 0 N/A Sum 0 N/A 0 N/A 0 N/A Percentage Missed 0% N/A 0% N/A 0% N/A [r]+nasal [torm] 0 N/A 0 N/A 0 N/A Sum 0 N/A 0 N/A 0 N/A Percentage Missed 0% N/A 0% N/A 0% N/A

PAGE 213

185 There were two instances in which the listeners reported not hearing an [r] in the token at signal to noise ratio of dB. Although there were not enough misses to be significant, it is worth mentioning that these instances of not hearing [r] occurred with different listeners. With more data, it would not be a surprise to find that the Russian listeners have more difficulty perceiving the fast speech [r] than the English speakers have hearing []. In the acoustic analysis, it was found that the [r] duration and relative rms decreased with speaking rate while the [] duration and rms increased. 10.2.3. Perception Dependent upon Manner of Articulation 10.2.3.1. Preceding stops The consonants that were found preceding voiceless stops were [s, p, k, m, n, l, ]. The ranking of the perceptibility of the consonants preceding the stops at +6dB and –6dB is found in Table 10-18. Table 10-18: Ranking of perceptibility of consonants preceding voiceless stops in casual speech +6dB, dB [l] >> [p, k], [s], [m, n], [r] [r] is ranked low, but it should be noted that in the dB presentation of [part], NRL2 responded ‘none’. In the 0dB signal to noise ratio none of the responses included the listeners choosing spellings with the sonorants not represented. In fast speech, NRL2 responded ‘none’ to several tokens with word-final stops. She responded none to [pant] in all three noise levels, [part] at 0dB, and [fast] at dB. Putting aside her ‘none’ responses gives the ranking in Table 10-19. At +6dB, none of the listeners responded ‘none.’

PAGE 214

186 Table 10-19: Ranking of perceptibility of consonants preceding voiceless stops in fast speech 0dB [p, k] >> [l], [s], [m, n], [r] -6dB [l] >> [p, k], [s], [m, n], [r] 10.2.3.2. Following [r] In the Russian data nearly all of the consonants being considered were found following [r] (p, t, k, b, d, , m, n). In this environment, the listeners (especially NRL2) often chose ‘none,’ so it is difficult to determine how well they perceived the consonants that followed [r]. NRL1 responded ‘none’ to [part] in casual speech at dB and [farb] in fast speech at dB. NRL2 responded ‘none’ to [parf] at dB in casual speech and +6dB in fast speech. In fast speech, she responded ‘none’ to [part] at 0dB and dB in fast speech, [torm] at 0B, [par] at 0 dB. As a result, it is difficult to know how to rank the consonants in this environment. This problem that the first two Russian listeners had with [r] is interesting though since the English listeners did not have the same problem. The third Russian listeners did not respond ‘none’ to the [r] tokens as the other two did. He did, though, report that he did not detect the [f] in the fast speech token of [parf] at 0dB. 10.3. English Tokens When listening to the English tokens, the Russian participants had the advantage of having lived in the United States for a year or more. In addition, the listeners had started taking English classes in grade school. Therefore, it was expected that they would do better than the native speakers of English did listening to the Russian tokens. Still, it was expected that there would be some problems. For example, Russian does not have the English [], so it was uncertain how they would react to it in the nonsense words. In

PAGE 215

187 Missed addition, Russian has word-final devoicing of obstruents. It was unclear what if any problems these factors would cause when the listeners heard the English tokens. 10.3.1. Perception of Obstruents 10.3.1.1. Stops in casual speech Like the native speakers of English, the native Russian speakers had difficulty perceiving the word-final voiceless stops (Table 10-20). Table 10-20: Perception of word-final voiceless stops in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard stop+[t] fopt 1 N/A 1 N/A 2 N/A Sum 1 N/A 1 N/A 2 N/A Percentage Missed 33% N/A 33% N/A 67% N/A [s]+stop fost 0 1 1 1 0 2 posk 0 3 0 2 0 2 Sum 0 4 1 3 0 4 Percentage Missed 0% 67% 17% 50% 0% 67% nasal+stop pont 2 N/A 2 N/A 2 N/A fomp 0 0 0 0 0 0 Sum 2 0 2 0 2 0 Percentage Missed 33% 0% 33% 0% 33% 0% [l]+vl stop pault 1 N/A 2 N/A 1 N/A Sum 1 N/A 2 N/A 1 N/A Percentage Missed 33% N/A 67% N/A 33% N/A [] + vl stop part 1 1 0 0 1 1 fark 0 3 0 0 1 0 garp 0 0 0 0 1 0 Sum 1 4 0 0 3 1 Percentage 11% 44% 0% 0% 33% 11%

PAGE 216

188 Missed This was particularly true when they were presented in greater noise. Following [s] they were always able to perceive the stops in all three noise levels, but at times they misperceived the place. The listeners did have problems perceiving the word-final stops in the other environments, but varied as to which environments had the most instance of no detection. This gives the ranking in Table 10-21. Table 10-21:Ranking of environments by perception of word-final stops at in casual speech +6dB [p__], [l__], [m, n __] >> [__] >> [s__] 0dB [l__] >> [p__], [m, n __] >> [s__], [__] dB [p__] >> [m, n __], [l__], [__] >> [s__] Considering [p] when it preceded [t], only one listener reported not perceiving it in the 0dB presentation of casual speech. It seems that these listeners were less likely to perceive the final [t] than they were to perceive the preceding [p] (Table 10-22). Table 10-22: Perception of [p] preceding [t] in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [p] + stop fopt 0 0 1 0 0 0 Sum 0 0 1 0 0 0 Percentage Missed 0% 0% 33% 0% 0% 0% As with the English perception of the English the tokens, the Russian listeners did not always perceive a word-final voiced stop (Table 10-23). Table 10-23: Perception of voiced stops in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [l] + stop fald 2 N/A 2 N/A 1 N/A Sum 2 N/A 2 N/A 1 N/A Percentage 67% N/A 67% N/A 33% N/A

PAGE 217

189 10.3.1.2. Stops in fast speech It was expected that there would be more difficulty hearing the fast tokens than casual tokens. This would seem to be most likely for the native Russian speakers listening to the word-final English stops, which were shorter and had lower relative rms word-finally (Table 10-24). Table 10-24: Perception of word-final voiceless stops in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard Stop + [t] fopt 1 N/A 1 N/A 0 N/A Sum 1 N/A 1 N/A 0 N/A Percentage Missed 33% N/A 33% N/A 0% N/A [s] +stop fost 1 0 0 3 0 2 posk 0 0 0 1 0 1 Sum 1 0 0 4 0 3 Percentage Missed 17% 0% 0% 67% 0% 50% nasal +stop pont 0 N/A 0 N/A 0 N/A fomp 0 0 0 0 0 1 Sum 0 0 0 0 0 1 Percentage Missed 0% 0% 0% 0% 0% 33% [l] + vl stop pault 2 N/A 1 N/A 0 N/A Sum 2 N/A 1 N/A 0 N/A Percentage Missed 67% N/A 33% N/A 0% N/A [] + vl stop part 1 0 0 3 0 1 fark 1 1 0 0 0 3 farp 0 0 0 0 1 0 Sum 2 1 0 3 1 4 Percentage Missed 22% 11% 0% 33% 11% 44%

PAGE 218

190 Again, which environment the listeners were less likely to perceive the word-final stops varied in the different levels of noise, but interestingly in the highest level of noise, the listeners only missed the stops after []. The rankings of the environments are found in Table 10-25. Table 10-25: Ranking of environments by perception of word-final stops by environment in fast speech +6dB [l__] >> [p__] >> [__] >>[s__] >> [m, n __] 0dB [l__], [p__] >> [s__], [m, n __], [__] -6dB [__] >> [p__], [s__], [m, n __], [l__] In fast speech, the listeners always predicted the presence of [p] preceding [t] in all levels of noise. In addition, they always correctly identified the part of speech. Instead, if they missed a consonant in the two stop cluster, they were less likely to perceive the [t] as was shown above in Table 10-24. As in casual speech, the Russian listeners were not always able to perceive the word-final [d] in ‘fald’ (Table 10-26). Table 10-26: Perception of word-final voiced stops in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [l] + stop fald 1 N/A 2 N/A 1 N/A Sum 1 N/A 2 N/A 1 N/A Percentage Missed 33% N/A 67% N/A 33% N/A The English listeners also had the same difficulty with the word-final voiceless stops. This is likely because [d] was shortened in this position by the English speakers since shorter stop duration is a cue for voicing in English.

PAGE 219

191 In both the casual and fast speech examples, the listeners had trouble detecting the presence of the word-final stops. This pattern was also true for the native English speakers listening to these same tokens. 10.3.1.3. Fricatives in casual and fast speech Like the native the English speakers, the native Russian speakers always detected the presence of word-final [] in both casual and fast speech for the token ‘parsh’. The native Russian speakers were not presented with the ‘barce’ as it is a real-word in Russian. The listeners also did not have a problem detecting [s] preceding a stop. In both casual and fast speech, the listeners never chose a spelling without the fricative represented in the tokens ‘fost’ and ‘posk.’ For the word-final non-strident [f], all of the Russian speakers correctly detected its presence, but one listener misidentified it as [s] in the 0dB noise ratio. This is different from the native English speakers who responded that there was no word-final [f] in ‘parf’ in dB signal to noise ratio. In the fast speech, the listeners always detected and never misperceived the word-final [f]. 10.3.2. Perception of Sonorants 10.3.2.1. Nasals in casual speech Nasals occurred word-finally after [] and preceding word-final stops. Word-finally, the listeners always perceived the word-final stops, but at times misidentified their place of articulation (Table 10-27).

PAGE 220

192 parn 0 1 0 1 1 2 Table 10-27: Perception of word-final nasals in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [] + nasal tarm 0 0 0 0 0 0 parn 0 1 0 2 0 0 Sum 0 1 0 2 0 0 Percentage Missed 0% 17% 0% 33% 0% 0% The English listeners also were always able to detect the word-final nasals. As for the nasals preceding the word-final stops, the Russian listeners only missed the nasals at +6dB (Table 10-28). Table 10-28: Perception of nasals preceding stops in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard nasal + stop fomp 1 0 0 0 0 0 pont 1 0 0 0 0 0 Sum 2 0 0 0 0 0 Percentage Missed 33% 0% 0% 0% 0% 0% Some of the English listeners reported not detecting the nasals in these same tokens in all three noise levels. 10.3.2.2. Nasals in fast speech In the fast speech tokens, word-finally the listeners perceived the nasals for the most part, but again they misidentified the place of articulation of the stop in several instances (Table 10-29). Table 10-29: Perception of word-final nasals in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard [] + nasal tarm 0 0 0 1 0 1

PAGE 221

193 Sum 0 1 0 2 1 3 Percentage Missed 0% 17% 0% 33% 17% 50% Preceding the word-final stops though was another story. Here, the listeners did not always predict the presence of the nasals (Table 10-30). Table 10-30: Perception of nasals preceding stops in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard nasal + stop fomp 1 0 1 0 1 1 pont 1 0 0 0 1 0 Sum 2 0 1 0 2 1 Percentage Missed 33% 0% 17% 0% 33% 17% In addition to the no hear and misperceptions, one listener responded ‘none’ to the fast production of ‘pont’ in the dB signal to noise ratio. 10.3.2.3. Lateral approximant [l] in casual and fast speech In both casual and fast speech, listeners always perceived the presence of [l] preceding an obstruent in the tokens ‘pault’ and ‘fald.’ The native Russian speakers also correctly identified the [l] as [] as the native English speakers did. This though is most likely because Russian natively has a trill not []. 10.3.2.4. Rhotic approximant [] in casual and fast speech Although the English [] tended to be longer and have greater relative rms, the cues for [] are not the same for Russian [r]. As a result, it was uncertain whether the native Russian speakers would be able to detect [] despite its long duration and greater relative rms. While the English speakers always predicted the presence of [] in the collected tokens, the Russian speakers did not (Table 10-31).

PAGE 222

194 Table 10-31: Perception of [] in casual speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard []+vl stop part 0 N/A 0 N/A 0 N/A fark 0 N/A 0 N/A 0 N/A farp 0 N/A 0 N/A 0 N/A Sum 0 N/A 0 N/A 0 N/A Percentage Missed 0% N/A 0% N/A 0% N/A [] + vd stop farb 0 0 0 0 0 0 fard 0 0 0 0 0 0 Sum 0 0 0 0 0 0 Percentage Missed 0% 0% 0% 0% 0% 0% []+strid fric parsh 0 0 0 0 0 2 Sum 0 0 0 0 0 2 Percentage Missed 0% 0% 0% 0% 0% 67% [f] parf 0 N/A 0 N/A 0 N/A Sum 0 N/A 0 N/A 0 N/A Percentage Missed 0% N/A 0% N/A 0% N/A []+nasal parn 0 N/A 0 N/A 1 N/A tarm 0 N/A 0 N/A 0 N/A Sum 0 N/A 0 N/A 1 N/A Percentage Missed 0% N/A 0% N/A 17% N/A What is interesting about this is that unlike the English speakers, the Russian speakers had taken classes in English. In addition, the English [] was longer and had greater relative rms than [r]. Although the listeners were not told that they were listening to a native speaker of English, all three reported either during the debriefing or during data collection that they knew they were listening to someone with an American accent. One

PAGE 223

195 Missed of the Russian listeners missed [] in the casual speech production of [tam]. Two listeners misperceived [] as [l] in the -6dB production of ‘parsh.’ In the fast speech tokens, two listeners had difficulty with the perception of [] in the tokens in the dB signal to noise ratio (Table 10-32). Table 10-32: Perception of [] in fast speech +6dB 0dB -6dB Not Heard Misheard Not Heard Misheard Not Heard Misheard []+vl stop part 0 N/A 0 N/A 0 N/A fark 0 N/A 0 N/A 0 N/A farp 0 N/A 0 N/A 1 N/A Sum 0 N/A 0 N/A 1 N/A Percentage Missed 0% N/A 0% N/A 11% N/A []+vd stop farb 0 0 0 0 0 0 fard 0 0 0 0 0 0 Sum 0 0 0 0 0 0 Percentage Missed 0% 0% 0% 0% 0% 0% []+strident fricative parsh 0 0 0 0 1 0 Sum 0 0 0 0 1 0 Percentage Missed 0% 0% 0% 0% 33% 0% [f] parf 0 N/A 0 N/A 0 N/A Sum 0 N/A 0 N/A 0 N/A Percentage Missed 0% N/A 0% N/A 0% N/A []+nasal parn 0 N/A 0 N/A 0 N/A tarm 0 N/A 0 N/A 1 N/A Sum 0 N/A 0 N/A 1 N/A Percentage 0% N/A 0% N/A 17% N/A

PAGE 224

196 Comparing the Russian responses to [] as compared to [r] it is also interesting to note that the Russian listeners did not respond ‘none’ to the [] tokens in the way that they did to the [r] tokens from Russian. In debriefing with the listeners, two of the listeners reported that they felt the English [] would make it difficult for Russians to complete this task. However, it seems that the Russian trill may have caused more trouble for them than the English []. It is uncertain whether this was due to the imperceptibility of [r] or perhaps the listeners were hearing a dialect of Russian that they were not comfortable. 10.3.3. Comparison of Perception by Manner of Articulation As in the English perception, the second hypothesis of the perception experiment was that consonant higher on the sonority hierarchy would be more likely perceived than consonants lower on the sonority hierarchy in parallel environments. Overall word-finally stops, [l], and [f] posed the most problems for the listeners particularly in greater noise. Conversely, strident fricatives were always perceived word-finally. 10.3.3.1. Preceding stops The consonants that were found preceding voiceless stops were [s, p, m, n, l, ]. The rankings of the perception of the consonants preceding the voiceless stops are in Table 10-33. Table 10-33: Ranking of perceptibility of consonants preceding voiceless stops in casual speech +6db [m, n] >>[l], [p], [s], [] 0db [p] >> [m, n], [s], []

PAGE 225

197 At +6dB in casual speech, speaker 2 responded ‘none’ to the token ‘pault’. At dB in casual speech, none of the Russian listeners reported not detecting the consonants that preceded the stops. As for the consonants preceding stops in fast speech, the listeners varied as to which consonant they did not hear dependent upon the level noise (Table 10-34). Table 10-34: Ranking of perceptibility of consonants preceding voiceless stops in fast speech +6dB [l] >> [p] >> [] >> [s] >>[m, n] 0dB [l], [p] >> [s], [m, n], [] -6dB [] >> [p], [s], [m, n], [l] The inability to perceive [] before stops increased in the noise levels, while the inability to detect [p] and [l] decreased. 10.3.3.2. Following [] The consonants that followed [] in the English tokens that were presented to the Russian listeners were [p, t, k, b, d, s, m, n]. The Russian listeners tended to have the most difficulty perceiving the stops. The rankings of the consonants that were difficult to perceive after [] are found in Table 10-35. At 0dB, the respondents never chose the spelling without the consonant after []. Table 10-35: Ranking of perceptibility of consonants following [] in casual speech +6dB [p, t, k] >> [b, d], [s, ], [f], [l] -6dB [p, t, k] [b, d] >> [s, ], [f], [m, n], [l] Recall that when the English speakers listened to the English tokens they had difficulty perceiving the word-final [f] after []. The Russian listeners did not have this same problem.

PAGE 226

198 The fast speech perception was similar to the casual speech perception of consonants after [] in that the listeners seemed to have the most trouble when [] was followed by a stop (Table 10-36). Table 10-36: Ranking of perceptibility of consonants following [] at in fast speech +6dB [p, t, k] >> [b, d] >> [s, ], [f] , [m, n], [l] -6dB [b, d] >> [p, t, k] >> [s, ], [f], [m, n], [l] Again, in the fast speech productions the respondents never chose the spelling with no consonant after []. The same was also true with the casual speech tokens. One of the listeners responded ‘none’ to the token ‘tarm’ in the dB presentation of the token. 10.4. Conclusions Comparing the Russian listeners’ perception of the Russian data to the English listeners’ perception of the English data, the listeners had different problems. While the English listeners tended to have difficulty perceiving nasals preceding stops, the Russians did not. However, the Russian listeners did have difficulty perceiving tokens with the trill [r] n it, while the English listeners did not have the same difficult perceiving tokens with []. Both the Russian and English listens did have difficult perceiving word-final stops. Neither groups had difficulty perceiving strident fricatives. Comparing the listeners’ ability to perceive each other’s languages, it is interesting that the English listeners had less difficulty with the [r] tokens than the Russians did. Perhaps this is because [r] is not found in English, so it stood out to the listeners. Similarly, the Russian listeners did not have much difficulty with the [] tokens.

PAGE 227

CHAPTER 11 PHONOLOGICAL ANALYSIS 11.1. Introduction The experiments in the previous chapters were undertaken to determine if a consonant’s ability to appear in certain positions in a syllable is correlated to its proportional duration, its relative rms, and its ability to be perceived in that position. This chapter considers the interplay between contiguous segments and the ability to perceive individual segments in a coda. Codas are significant for understanding the relationship between perception and articulation because the gradual slope suggests easy production, but the gradience also implies less of a distinction between features. This dissertation considers the acoustic qualities of the segments. Further study, then, would also consider the role of articulation in the preference for certain clusters. Research by Flemming (1995) and Ct (2000) suggest the need for constraints on perception to be included in an Optimality Theory framework (Prince and Smolensky 1993). That is, deletion and corrections such as epenthesis are motivated by the need for listeners to perceive the segments in a word. One question that this dissertation addresses is whether these perception constraints are motivated by the listeners’ reactions to the consonant clusters. Given tokens where consonant clusters are produced, are some clusters truly more difficult to perceive, as the perception constraints would suggest? One problem with understanding data from an acoustic experiment in a phonological framework is that phonetics and phonology are inherently different. As 199

PAGE 228

200 Hayes (1999) points out phonetics, is gradient, but phonology is not, so phonological analyses are not wholly dependent upon phonetics. Consonants are lengthened or shortened, made louder or softer differently from person to person and experience to experience. Phonological analysis should take into consideration the major trends from phonetics analyses, but does not and should not take into account every single variation in production and perception. What this chapter does is try to account for the ranking of the imperceptibility of certain consonant clusters in chapters 9 and 10 using OT constraints. While keeping in mind the data presented in the acoustic analysis. However, it does not attempt to account for every slight variation found. That is, if one listener could not perceive a single consonant in one presentation of a token, a phonological model should not account for that. 11.2. English Phonological Analysis In English, there are limitations to the possible word-final consonant cluster, particularly when the vowel is low mid or back vowel. For instance, in a nasal+stop cluster the nasal and stop should agree in place of articulation. There is a preference for [l] to occur preceding a voiceless stop of the same place of articulation. In addition, an [l] between a low back vowel and a nasal is often deleted. These limitations can be accounted for by ease in articulation as well as ease in perception. The acoustic cues of [l] (anti-formants and lowering of F2) are lost between [a] and [m]. Notice [l] can occur between [] and [m] as in ‘film’ where F2 is high on the vowel allowing for a clear transition. In all of the English tokens, [] was not a problem for perception; in all tokens with [], the listeners perceived an []. From the acoustic analysis recall that [] was

PAGE 229

201 longer and louder than the other sonorants in the same positions. In addition, [] lengthened with speaking rate, but not at the expense of a following consonant. Instead, it gained length by interacting with the preceding vowel. This agrees with the notion of [] coloring of vowels preceding []. The strident fricatives [s, ] were also little problem for perception with the exception that the place of articulation was not always perceived correctly. Also similar to [], speaking rate had little effect on strident fricatives. Word-final nasals were also not a problem. The most expected problem for perception was the stop + stop clusters. Stops were also difficult word-finally after the other consonants. Given the weakening in terms of relative rms and duration with increases in speaking rate, this was not surprising. Less anticipated problems that will need to be accounted for were word-final [l], [f], and nasals preceding stops. The tableaus in this chapter do not represent how the tokens were produced, but rather as most listeners perceived them. For all of the tokens, some listeners were able to perceive the tokens as the speakers intended. A high ranking of MAX-IO (Figure 11-1.a) would account for the perception of all of the segments in the token. Instead, what this dissertation hopes to capture is why the individual segments were not clearly enough pronounced to be perceived by all of the listeners. 11.2.1. Stop Deletion The simplest perception to account for is stop deletion following fricatives. Acoustically, an inability to perceive stops after fricatives is motivated by the similarity of the frication noise of the fricative and the release burst of the stops. In the production

PAGE 230

202 of the English tokens, these fricatives and stops blended together with little stop-gap between the two. The deletion of these stops can easily be accounted for by relying on two constraints suggested by Ct (2000). The first is a relevant markedness constraint, CV, that requires consonants be adjacent to vowel (Figure 11-1.b). The second is a MAX-IO constraint. MAX-IO constraints require that any segment that is in the input be represented in the output. In this case, the constraint specifically requires that any segment adjacent to a vowel in the input, be present in the output. MAX-C/V__ prohibits deletion of a consonant that is adjacent to a vowel (Figure 11-1.c). IDENT-F (11-1.d) captures that listeners were given the choice in some cases of a spelling that misrepresented the place or manner of articulation of the segment being produced. These variations in the place and manner features could also have been accounted for by MAX and DEP constraints. IDENT was chosen for simplicity in the tableaus. a. MAX-IO Every segment of the input has a correspondent in the output. (McCarthy and Prince 1995) b. C V A consonant is adjacent to a vowel. (Ct 2000) c. MAX-C/V__ Do not delete a postvocalic consonant (Ct 2000) d. IDENT (F) Correspondent segments are identical in feature F (McCarthy and Prince 1995) Figure 11-1: Already existing constraints Ranking constraints C V and MAX-C/V__ above MAX-IO correctly predicts the forms where the fricative is preserved and the word-final stops are lost, as the listeners usually perceive the fricatives, but often do not perceive the stops (Table 11-1).

PAGE 231

203 Table 11-1: Account for stop deletion following [s] Max-C/V__ C V MAX-IO /fasp/ [fasp] *! [fap] *! * [fas] * [fast] /fast/ [fast] *! [fat] *! * [fas] * [fasht] *! /pask/ [pask] *! [pak] *! * [pas] * [pasp] *! Now, the ranking also predicts the most common failure in perception by the listeners. In the two stop clusters, listeners were not likely to hear the word-final stops. This ranking also accounts for the loss of the word-final [t] following [p] (Table 11-2). Table 11-2: Account for stop deletion following [p] MAX-C/V__ CV IDENT-F MAX-IO /fapt/ [fapt] *! [fap] * [fat] *! * [falt] *! Some listeners did though, at times choose the form [fat] rather than [fap]. I propose that this is not because the listeners were more likely to perceive the word-final stop rather than the first stop, but because the listeners were unable to detect the place of articulation of the stop. They merely detected the presence of a stop and chose a multiple-choice answer that represented that. This explanation agrees with the general phenomena throughout data collection of the listeners choosing the wrong form whenever a word-

PAGE 232

204 final stop was present, e.g. choosing [pasp] for [pask] or [fast] for [fasp]. I suggest that this is accounted for with a low ranking of an IDENT-F constraint (Table 11-3). Table 11-3: Account for stop deletion in two stop clusters MAX-C/ V__ C V IDENT-manner IDENT-place MAX-IO /fapt/ [fapt] *! [fap] * [fat] (*) * * [falt] *! The addition of the IDENT-F constraint shows a relationship between input and output, such that a feature in the input should be the same at output. Ranking IDENT-place low allows for the output to not agree in place of articulation of the stop with the input. 11.2.2. Perception of [] clusters The account of the stop in Table 11-13 deletion can also account for the deletion of stops after [] (Table 11-4). Table 11-4: Account for stop deletion following [] MAX -C/ V__ C V IDENT-F MAX-IO /fap/ [fap] *! [fap] *! * [fa] * [fak] *! * /gat/ [gat] *! * [gat] * * [ga] * [gap] *!

PAGE 233

205 However, this account does not show that stops are less likely to be missed in this position than they are following fricatives or other stops. In the data produced by the English speakers, the proportion of the rhyme devoted to the vowel preceding [] was small (Table 11-5). Table 11-5: Duration of segments in [] + stop rhymes vowel [] stop Casual 15% 49% 36% Fast 15% 50% 35% Acoustically, the English [] has the benefit of having the primary feature of a lowered F3. This allows for coalescence with the vowel. Since F3 frequency is not a cue for vowels, [] can easily blend with the vowel. Ct (2000) introduces uniformity constraints that capture the fact that segments that are more sonorous are more likely to coalesce with vowels (Figure 11-2). Although [] was perceived as a separate segment by the listeners, they production of [] was such that it was taking over the vowel’s place in the rhyme. a. UNIFORMITY-V [-sonorant] No vowel in the output corresponds to itself and a [-sonorant] segment in the input. b. UNIFORMITY-V [-approximant] No vowel in the output corresponds to itself and a [-approximant] segment in the input. c. UNIFORMITY-V [-vocoid] No vowel in the output corresponds to itself and a [-vocoid] segment in the input. Figure 11-2: Sonority-relative Uniformity-V constraints Relying upon the uniformity constraints, now the output that is predicted is one in which the coalesced [a] is perceived correctly as two segments, but is actually one (Table 116).

PAGE 234

206 Table 11-6: Account for stop deletion following [] MAX-C/V__ CV MAX-C IDENT-F UNIFORMITY-V [-vocoid] /fap/ [fap] *! >[f(a)p] * [fap] *! [fa] *! [fak] *! /gat/ [gat] *! >[g(ar)t] * [gat] *! [ga] *! [gap] *! The other uniformity constraints would be ranked high to prevent the vowel from coalescing with an obstruent or approximant. 11.2.3. Nasals Preceding Stops While the listeners nearly always perceived the other consonants, they did not always perceive the nasals in this environment. It is important to note that the nasals were not lost all of the time or even the majority of the time. However, they were missed by listeners more often than the other consonants preceding stops. Several factors may have contributed to the imperceptibility of the nasals. The nasals had a greater relative rms than the adjacent stops, but were about 40% of the coda. In coda clusters, stops and nasals are the same place of articulation and nearly the same manner. While coalescence of [] and a vowel allows for a successful perception of [], this does not occur for nasals. The nasals are lost between the vowel and the voiceless stops for some listeners. In the acoustic analysis breaking down the rhyme reveals that the vowel is a greater

PAGE 235

207 proportion of the rhyme for these speakers preceding nasals than it is preceding [] (Table 11-7). Table 11-7: Duration of nasals and [] preceding stops vowel or nasal stop Nasal Casual 50% 20% 30% [r] Casual 15% 49% 36% Nasal Fast 49% 23% 26% [] Fast 15% 50% 35% Remember that [] was measured from the beginning of the F3 transition and the nasal was measured from the earliest instance of anti-formants. While the nasals are not deleted here completely acoustically, the nasal is short enough to be lost for some of the listeners. This unwillingness for the nasal to coalesce can be explained by a high-ranking UNIFORMITY-V [-approximant]. However, relying upon this analysis would wrongly predict the deletion of [s] in the same environment. The next question is why in this case would the listener be more likely to perceive the stop than the nasal. This is what Ct’s analysis fails to capture. That is, it does not account for why some sonorants are more difficult to detect than others in the same environment are. For instance, it is true as Ct and others have pointed out that [l] deletes in [lm] clusters in English and other languages (think of ‘calm’ and ‘palm’). This is accounted for by the repetition of the primary cue of antiformants. However [l] does not delete in English following [], as in ‘film’ and ‘kiln.’ An analysis of sonorants’ and other consonants’ behaviors between a vowel and consonant should account for how the consonants’ cues are less pronounced surrounded by the vowel and consonant. An inability to detect a nasal between a vowel and a stop may relate to a nasal’s inability to make clear its cues in this environment. In particular, the duration of the nasals as

PAGE 236

208 Recall from the acoustic analysis that [l] in the in the nonsense words was longer than in the real words, so [l] is coded as +duration. It remains to be seen if had [l] not had this duration, would it have been perceived? produced by these native English speakers was shorter than the other sonorants1. A nasal having the same place of articulation as its following consonant would also be unable to rely upon formant transitions as a cue. Flemming (1995) uses an OT framework to show that perceptibility is based upon a segment’s ability to distinguish itself in a given environment. Similarly, this analysis relies upon different consonant’s ability to have distinct formant frequencies, duration, and loudness in a given environment in order to ensure perceptibility. I propose the need for constraints based on salience that capture a consonant’s distinctness in a particular environment. These constraints would capture the unique formant transitions, the loudness, and the duration of different consonants in parallel environments. What makes [] perceptible is its ability to optimize its cues in a variety of environments. The primary cue of [] is its lowered F3. Since F3 is not a cue in vowels, this allows [] to stand out next to vowels and to encroach on the vowel’s space in the rhyme. This is in turn allows [] to be long. In the data collected from these English speakers, [] was also louder than the other sonorants. Nasals and [l] rely on antiformants as a primary cue. The sonorants with antiformants are not able to have the great rms that [] does. Table 11-8 shows how the cues related to [] and [s] relate to the cues related to the nasals and [l]. [s] is coded as both + and –Unique Transitions because it can occur preceding stops with the same and different place of articulation. Salience features are dependent upon the segment’s place in a cluster. This is because a segment’s 1

PAGE 237

209 perceptibility is not on reliant upon the segment itself, but how the segment is able to remain distinct in the context of other segments. For instance, [l] between [] and a nasal would have the feature of +Unique Transition, since in this position the F2 transition of [l] is distinct. In this table [l] is marked as +Duration because the nonsense word productions of [l] were significantly longer than the real words. Table 11-8: Salience features between [a] and a stop [] [s] [l] [n, m] [p] Duration + + -(+) Unique Transitions + + + Loudness + + Anti-Formant + + These features are called ‘salience features’ based on their ability to alert the listener to the presence of the various consonants. Each of these salience features would be ranked as MAX-IO constraints as in Figure 11-3. a. MAX-C/ [a]__ (duration) A consonant that is able to maintain duration following [a] is present in the output b. MAX-C/ [a]__ (loudness) A consonant that is able to maintain loudness following [a] is present in the output c. MAX-C/ [a]__ (unique transitions) A consonant that is able to maintain unique formant transition [a] is present in the output d. MAX-C/ [a]__ (antiformant) A consonant that is able to maintain antiformants following [a] is present in the output Figure 11-3: Salience Constraints Salience constraints would be universally ranked such that length and loudness are ranked higher than unique transition and anti-formant, since the latter two do not guarantee that a segment can hold onto its duration in a consonant cluster.

PAGE 238

210 The salience constraints are interranked with a MAX-IO constraint that requires that a word-final consonant present in the output is present in the input (Figure 11-4). MAX-C/ __# A word-final consonant present in the input is present in the output. Figure 11-4: MAX-IO word-final consonant The MAX-C/ __# constraint is ranked higher than MAX-C/ [a]__ (antiformant), but below the other Salience constraints to ensure that the other consonants are realized in the consonants clusters while the nasals are not (Table 11-9). Table 11-9: Account for nasal deletion preceding voiceless stops MAX-C/ V__ (Duration, Loudness, Trans) CV IDENT-F MAX C/__# MAX-C/ V__ (antiformant) /famp/ [famp] *! * [fap] * [fam] *! [fank] *! /pant/ [pant] *! [pat] * [pan] *! [palt] *! This ranking would also account for why [], [s], and [l] would be less likely to delete in this position than the nasals (Table 11-10). By breaking up MAX-C/V__ according to the existence of salience features, then it is possible to predict the perception of the different consonant clusters by the listeners.

PAGE 239

211 [pas] *! * Table 11-10: Account for preservation of [] and [s] preceding voiceless stops MAX-C/ V__ (Trans, Duration, Loudness) CV IDENT-F MAX-C/__# MAX-C/ V__ (Antiform) /fap/ [fap] *! [fap] *! [fa] * [fak] *! *! /fast/ [fast] *! [fat] *! [fas] * [fasht] *! * 11.2.4. Word-final Nasals, Fricatives, and [l] While stops often were often missed word-finally following [] and [l], strident fricatives did not. Ct (2000) would account for this by ranking a constraint MAX-C(-stop), which requires that any segment other than a stop present in the input be present in the output. This constraint would be ranked higher than CV, so that any consonant that is not a stop would be preserved regardless of its place in the word. Including this constraint, though, would wrongly predict that nasals preceding stops should be preserved. This constraint would have to be altered to MAX-C(-stop)/__# to predict that consonants other than stops should be preserved word-finally (Table 11-11). Table 11-11: Account for preservation of [] clusters MAX -C/ V__ (Trans, Duration, Loudness) MAX-C (-stop) /__# IDENT-F CV MAX-C (+stop)/__# MAX-C/V__ (Antiform) /pa/ [pa] * [pa] *! [pa] *!

PAGE 240

212 In the English data though, listeners had difficulty perceiving [l] and [f] word-finally after []. It is questionable whether these two misperceptions should be accounted for. [l] in the NES2’s casual speech production of ‘parl’ (this was the production that listeners had difficulty with) had a shorter duration (21%) than the fast speech production (56%) and the average of all of the speakers’ production (35%). As for [f] it had a lower relative rms and duration than the strident fricative [s], but it was also occurring before an [f] in frame sentence, so this may have influenced the production. However, the lack of perception of these segments could be accounted for by breaking down MAX-C(-stop)/__# with salience features. 11.2.5. Conclusions The problems that listeners had with the consonant clusters were consistent. These consistencies could be accounted for by the ranking of salience features that act as cues to the listeners of the presence of a given segment. Given these features, it remains to be seen if they can also account for the perception of the Russian tokens. 11.3. Russian Phonological Analysis Before analyzing the Russian data, it is necessary to understand how the perception of the Russian data differed from the perception of the English data. [l] was several times not perceived before stops. The Russian trill [r] was more of a problem than was the English [] for both the Russian and English listeners. The nasals preceding stops were not a problem for perception by the Russian listeners and were less of a problem for the English listeners than was the perception of the English nasals preceding stops.

PAGE 241

213 [palt] *! In Russian, the salience features would be different from the salience features in English because as was shown in the acoustic analysis, they were produced differently (Table 11-12). Unlike English [], Russian [r] would be -Duration and -Loudness given the shorter duration and lower relative rms. Table 11-12: Salience features between [a] and a word-final stop [r] [s] [l] [n, m] [p, k] Duration + + Unique Transitions + + + Loudness + + Anti-Formant + + [l] in this Russian data is also [-duration] given its shorter duration. The Russian nasals were approximately the same relative rms and duration as the English nasals. However in Russian with the shorter duration of [l] and [r], nasals were louder and longer than the other sonorants. Now ranking the Duration and Loudness high gives predicts the right outcome as in Table 11-13. Table 11-13: Account for Russian nasal + stop clusters MAX-C/ V__ (Duration, Loudness) CV IDENT-F MAX C/__# MAX -C/ V__ (Trans, Antiform) /famp/ [famp] *! * [fap] *! * [fam] * [fank] *! /pant/ [pant] *! * [pat] *! * [pan] *

PAGE 242

214 The ranking in Table 11-13 can also account for the form with [s, r, l] preceding stops as is shown in Table 11-14. Table 11-14: Account for Russian [s, r, l] + stop clusters MAX-C/ V__ (Duration, Loudness) CV IDENT-F MAX-C/__# MAX-C/ V__ (Trans, Antiform) /fast/ [fast] *! [fat] *! [fas] * [fasp] *! /palt/ [palt] *! [pat] * [pal] *! [part] *! /fark/ [fark] *! [far] *! [fak] * [farp] *! The Russian listeners also had difficulty perceiving [l] between [o] and [m]. Again, this can be accounted for with the concept of salience In this case [l] sandwiched between a back rounded vowel and [m] is unable to rely upon antiformants or unique formant transitions since a back rounded vowel already has a lowered F2. The same constraints that played role in English played a role in the Russian perception with different results. Having a trill that is unable to interact with the vowel in the same way the English [] did inhibit the Russian listeners ability to detect the presence of [r]. It is worth noting that the Russian listeners did not have the same difficulties with the English [] that they had with the Russian [r].

PAGE 243

215 11.4. Accounting for Other Phenomena Some other tendencies that can be accounted for from both the English and Russian data include the increased likelihood of dropping a word-final voiced stop over a word-final voiceless stop. Also, listeners were more likely to drop a word-final stop if it agreed in place of articulation with the preceding sonorant. This can be accounted for as Ct (2000) accounts for these phenomena. MAX constraints that require that any place of articulation or voicing distinction present in the input would allow for a preference for preserving an [lt] cluster over a [ld] cluster. These constraints would also allow for a preference for preserving a [sp] cluster over a [st] cluster. 11.5. Conclusions In the acoustic data, Russian and English varied as to the rms and the duration of the consonants in the same environments. These variations affected the listeners’ ability to perceive different consonants in different word positions. In particular, the ability to perceive consonants been vowels, and stops varied dependent upon the duration and the relative rms of a consonant affected the listeners’ ability to distinguish pre-final consonants. Sonority sequencing in coda clusters requires a gradual slope. However, the gradual slope can cause consonants to become lost between the vowel and a word-final stop in both English and Russian. Salience features that are dependent upon the acoustic characteristics of a consonant in a particular environment are able to predict the perceptibility of a consonant in an environment. These features do so by capturing both what is unique about a particular segment and how well the segment’s qualities can be optimized in a given environment.

PAGE 244

CHAPTER 12 CONCLUSIONS This dissertation has considered how consonants vary in terms of relative rms and proportion of rhyme and coda dependent upon place in word-final clusters and how those variations affect the perceptibility of the consonants. It was found that consonants do vary dependent upon word-position and that not all consonants vary in the same way. In turn, it was found that these variations affect the perceptibility of individual segments and that this perceptibility is similar from listener to listener. Sonority has been used to account for why certain segments are allowed to fall in certain positions in consonant clusters. Consonants that are more sonorous have been said to be louder, longer, and more easily perceptible. As was seen in this study, the consonant that ranks highest on the sonority hierarchy, [], was the most easily perceived segment and it was louder and longer than the other consonants and was able to maintain this loudness and length despite changes in speaking rate. Using Optimality Theory constraints, this ability to preserve length and loudness was related to salience that suggest that length and loudness is related to a consonant’s perceptibility in consonant clusters. Salience is not unlike Ohala’s (1990) idea of modulations, that a segment’s ability to appear in certain consonant clusters is not inherent to the consonant, but is related to its ability to optimize its cues in certain word positions. Salience, as it relates to the English consonants, is also similar to the sonority hierarchy where [] outranks the other 216

PAGE 245

217 sonorants in terms of its ability to optimize its length, rms and F3 transition and remains perceptible at various speaking rates. The English [] has advantages both articulatorily and acoustically that allow it to remain perceptible and be articulated in a variety of word positions. F3 transitions allow it to blend well with the vowel while preserving the vowel, since only F1 and F2 are needed to distinguish English vowels. In addition, the lowering of F3 can be achieved by extending the lips and/or retroflection allowing for the production of [] adjacent to a variety of places of articulation. Of the sonorant consonants, the listeners were less likely to report perceiving nasals than [] and [l] preceding word-final stops. In English, the nasals usually have the same place of articulation as following stops. This does not allow the nasals to have formant transitions that are distinct from the stop. In addition, nasals have antiformants as a primary cue which are inherently lower in amplitude than producing an []. These factors may cause nasals to be less distinguishable than the other sonorants. Considering the obstruents in English, the strident fricatives were better able to sustain duration and relative rms in various speaking rates and were better perceived in a variety of speaking rates. If these factors do contribute to what is sonorous, this suggests that the obstruents are not all equal on the sonority hierarchy as Clements (1988, 1990) suggests; instead, they would be separated as has been suggested by Hankamer and Aissen (1974), Steriade (1982), Dell and Elmedlaoui (1985). This may suggest that the strident fricatives are ranked higher than some, if not all of the sonorants similar to Lindblom (1983) and Keating (1983). The Russian consonants were different from English in terms of their perceptibility. While the nasals were the sonorants least likely to be perceived in English,

PAGE 246

218 they were the sonorants most likely to be perceived in the Russian data. This suggests that the Russian trill would be ranked lower than the nasals on the sonority hierarchy in Russian. Unlike the English [], the Russian trill does not have unique F3 transitions. In addition, the Russian speaker did not extend its duration or increase its relative rms as the English speaker did with []. Similarly, the Russian [l] was also not well perceived and not as long or loud as the English [l]. This suggest that either nasals outrank both [l] and [r] on the Russian sonority hierarchy or that while the salience features are universally ranked, the salience features of the consonants vary dependent upon the phonetics of the language and the consonant’s position in a word. This study has suggested that perceptibility as well as articulation should be considered in Optimality Theory. Listeners were able to perceive consonants based upon their loudness, duration, and ability to optimize formant frequencies in word-final clusters. In addition, OT constraints related to the listeners’ ability to be alerted to the presence of a consonant were able, in both English and Russian, to predict how listeners would misperceive consonant clusters. This dissertation does not suggest that salience alone plays a role in acceptable consonant clusters. Of course, articulation would also be important in determining the acceptability of consonant clusters. What this dissertation has strived to show is that acoustics and perceptibility plays some role in consonant clusters and that it should also be considered in an account of sonority sequencing. These salience features may be proved to play a role in some of the other processes that sonority has been used to explain. For instance, Kenstowicz (1997) uses sonority to explain stress placement. There stress is placed on the most sonorant segment.

PAGE 247

219 If stress were related to duration and loudness, this could be related to salience features in that stress is placed on the segment where duration and loudness could be optimized. The syllable contact law (Murray and Vennemann 1983) requires that in abutting syllable the onset of the second syllable be less sonorous than the coda of the previous syllable. After understanding the role that salience plays in onsets, this might be explained by the better perceptibility of less sonorous segments in onset positions. For instance, stops are better perceived before vowels where their bursts are contrasted against the vowel. Hankamer and Aissen (1974) show examples where assimilation is either regressive or progressive in direction of less sonorous segment. This may be to strengthen the weaker segment that might otherwise be imperceptible. Future consideration might also be given to consonant clusters word-initially to determine if the same consonants are produced and perceived as well in reverse clusters and if the consonants should have the same salience features in onsets. Salience feature could be used to account for why [s] precedes stops in both onsets and codas. Other studies related to this research might consider the perception of consonants in different types of noise, such as white noise or babble. Also of interest might be consonant clusters following different vowels. For instance, the data collected as part of this study also included words such as ‘film’ and ‘kiln.’ In these words, the change in F2 transitions from the vowel to [l] could be emphasized following the front vowel. This allowed for a longer [l] duration. In these tokens, listeners were always able to detect [l], despite being followed by a nasal. In this case, salience features are affected not only by a following consonant, but also by the preceding vowel.

PAGE 248

220 Using an acoustic analysis, this dissertation supports the need for acoustics and perceptibility in an account of sonority and the sonority hierarchy, particularly as it relates to acceptable word-final consonant clusters. Salience constraints capture the unique qualities of segments, as well as the segments’ abilities to optimize those qualities in particular environments. Universally ranked salience constraints could capture the necessity for certain cues to perception in order for consonants to perceived in consonant clusters. In addition ranking the constraints, but not the individual segments could allowable sequences of consonants vary from language to language and within languages different word positions.

PAGE 249

APPENDIX A ENGLISH QUESTIONNAIRE Date: ____________ Participant #: _______ Participant Questionnaire Please fill out the questionnaire to the best of your knowledge. All information will be kept confidential. Name (First/ Last): _______________________________________________________ Phone Number: _________________________ e-mail: ___________________________ Sex: ______________ Date of Birth: _________________ Place of Birth (city, state, country): ___________________________________________ Occupation: _____________________________________________________________ If a student, parents’ occupations: ____________________________________________ Native Language(s): ______________________________________________________ Mother’s Native Language: _________________________________________________ Father’s Native Language: __________________________________________________ What language(s) do you speak at home? ______________________________________ What language(s) do you speak in school? ____________________________________ What languages have you studied? ____________________________ (language) from _______ to ________ 221 ____________________________ (language) from _______ to ________

PAGE 250

222 ____________________________ (language) from _______ to ________ What other languages do you speak fluently:________________________________ _________________________________________ Have you ever studied Russian or spent extensive time with native speakers of Russian?____ Have you ever studied Japanese or spent extensive time with native speakers of Japanese? ______ Have you ever studied Polish or spent extensive time with native speakers of Polish? ______ Have you ever studied Ukrainian or spent extensive time with native speakers of Ukrainian? ____ In what cities have you lived in for more than six months? Begin with the most recent. Include Gainesville if you have been here 6 months. ___________________________(country, city, state) from __________ to ________ ___________________________(country, city, state) from __________ to ________ ___________________________(country, city, state) from __________ to ________ ___________________________(country, city, state) from __________ to ________ ___________________________(country, city, state) from __________ to ________ Have you had a cold or ear infection within the past month? ___________________ Have you ever been diagnosed with a hearing impairment? (if yes, explain)__________________________________ _________________________________________________________________ a speech impediment? (if yes, explain)___________________________________

PAGE 251

223 _________________________________________________________________ a learning disorder that affects your ability to read or spell? (if yes, explain) __________________________________________________________________ Thank you!

PAGE 252

APPENDIX B RUSSIAN QUESTIONNAIRE Date: _______________ Participant #: ________ Participant Questionnaire Please fill out the questionnaire to the best of your knowledge. All information will be kept confidential. Name (First/ Last): _______________________________________________________ Phone Number: _________________________ e-mail: ___________________________ Sex: ______________ Date of Birth: _________________ Place of Birth (city, country): ___________________________________________ Occupation: _____________________________________________________________ If a student, parents’ occupations: ____________________________________________ Native Language(s): ______________________________________________________ Mother’s Native Language: _________________________________________________ Father’s Native Language: __________________________________________________ What language(s) do you speak at home? ______________________________________ What language(s) do you speak in school? _____________________________________ What languages have you studied? 224 __________________________________ (language) from _______ to ________

PAGE 253

225 __________________________________ (language) from _______ to ________ __________________________________ (language) from _______ to ________ Other languages you speak fluently: __________________________________________ In what cities have you lived in for more than six months? Begin with the most recent. Include Gainesville if you have been here 6 months. ___________________________(city, state, country) from __________ to ________ ___________________________(city, state, country) from __________ to ________ ___________________________(city, state, country) from __________ to ________ ___________________________(city, state, country) from __________ to ________ Have you had a cold or ear infection within the past month? ___________________ Have you ever been diagnosed with a hearing impairment? (if yes, explain)__________________________________ _________________________________________________________________ a speech impediment? (if yes, explain)__________________________________ _________________________________________________________________ a learning disorder that affects your ability to read or spell? (if yes, explain) _________________________________________________________________ When did you start studying English? ______________________________________ How long have you been in the United States? _______________________________

PAGE 254

226 How much overall would you estimate that you have used Russian and English within the past six months? At home: English: _____% Russian: _____% Other _____% At school: English: _____% Russian: _____% Other _____% With friends: English: _____% Russian: _____% Other _____% Thank you!

PAGE 255

227 wavwrite(z0,fs,16,'./noisySounds/part_fastr0.wav'); wavwrite(z6,fs,16,'./noisySounds/part_fastr6.wav'); APPENDIX C NOISY SOUND SCRIPT % jody1.m: This file does what Jodi wants it to. % July 11, 2001 clear all; format compact; % Open .wav files: [x,fs]=wavread('./sounds/part_fastr.wav'); [noise,fs]=wavread('./sounds/noise.wav'); % Extract noise segment from total noise: lenX = length(x); beginN=floor(rand(1,1)*(length(noise)-lenX-1))+1; endN=beginN+lenX-1; y=noise(beginN:endN); % Calculate energy in each segment: energyX = sum(x.*x); energyY = sum(y.*y); % Find scale factor for -6, 0, and 6 dB: k_6 = sqrt(energyX/energyY*10^(-(-6)/10)); k0 = sqrt(energyX/energyY*10^(-(0)/10)); k6 = sqrt(energyX/energyY*10^(-(6)/10)); % Combine noise and signal: z_6 = x + k_6*y; z0 = x + k0*y; z6 = x + k6*y; % Scale output to remove clipping: z_6 = z_6/max(abs(z_6))*.99; z0 = z0/max(abs(z0))*.99; z6 = z6/max(abs(z6))*.99; % Save results to new wav file: wavwrite(z_6,fs,16,'./noisySounds/part_fastr_6.wav');

PAGE 256

228 1. pard 2. pald 3. pod 4.pall 1. farb 2. fard 3. fab 4.far APPENDIX A ENGLISH PERCEPTION EXAMS Table A-1: Multiple choice for English perception of English tokens 1. barsh 2. bar 3. barce 4. bas 1. bot 2. bok 3. ba 4.bart 1. cas 2. cosh 3. carsh 4. car 1. dop 2. darp 3.dar 4. da 1. fard 2. fald 3. fod 4.fall 1. farb 2. fard 3. fab 4.far 1. fark 2. fokt 3. farct 4.fart 1. fald 2. fard 3. fot 4. far 1. fark 2. fock 3. far 4. farp 1. fop 2. farp 3. far 4. falp 1. fard 2. far 3. fod 4. fot 1. fom 2. fomp 3. fop 4.fonk 1. falt 2. fopt 3. fot 4.fop 1. foss 2. fot 3. fosht 4. fost 1. gar 2. gat 3. gart 4. garp 1. palse 2. poss 3. pall 4. parce 1. poff 2. parf 3. par 4. parp 1. parl 2. par 3. pall 4. pa 1. pon 2. par 3. parn 4. parl 1. parce 2. par 3. parsh 4.posh 1. pault 2. pall 3. pott 4. pod 1. pock 2. posk 3. poss 4. posp 1. sha 2. sholf 3. shoff 4. shov 1. talm 2. tam 3. tarm 4. tar 1. ta 2. tob 3. tarb 4. talb 1. fosp 2. fost 3. foss 4. fop 1. dop 2. darf 3. doff 4. dar 1. pont 2. ponn 3. pot 4. pault Table A-2: Multiple choice for English perception of Russian tokens 1. carsh 2. car 3. carce 4. cos 1. bot 2. bok 3. ba 4.bart 1. cas 2. cosh 3. carsh 4. car 1. dop 2. darp 3.dar 4. da

PAGE 257

229 1. nald 2. nard 3. not 4. nar 1. fark 2. fock 3. far 4. farp 1. fop 2. farp 3. far 4. falp 1. fard 2. far 3. fod 4. fot 1. fom 2. fomp 3. fop 4.fonk 1. folt 2. foapt 3. foat 4.foap 1. foss 2. fot 3. fosht 4. fost 1. gar 2. gat 3. gart 4. garp 1. gom 2. ga 3. gob 4. gon 1. larsh 2. losh 3.larce 4.lalsh 1. poff 2. parf 3. par 4. parp 1. parce 2. par 3. parsh 4.posh 1. pault 2. pall 3. pott 4. pod 1. pock 2. posk 3. poss 4. posp 1. sha 2. sholf 3. shoff 4. shov 1. talm 2. tam 3. tarm 4. tar 1. dop 2. darf 3. doff 4. dar 1. pont 2. ponn 3. pot 4. pault 1. sharn 2. shom 3. sharm 4. shon 1. sharf 2. shoff 3. shar 4. sharp 1. bocked 2. bock 3. bot 4. bart 1. goan 2. gore 3. gorn 4. gorl 1. pelm 2. peln 3. penn 4. pell 1. perl 2. per 3. pell 4. peln 1. poln 2. polm 3. poem 4. poll 1. tolk 2. toke 3. toll 4. tork 1. tolm 2. tome 3. torm 4. tore

PAGE 258

230 1. 2. 3. 4. 5.None APPENDIX D RUSSIAN PERCEPTION EXAMS Table D-1: Multiple choices for Russian perception of Russian tokens 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None Table D-1: Multiple choices for Russian perception of English tokens 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None

PAGE 259

231 1. 2. 3. 4. 5.None 1. 2. 3. 4. 5.None

PAGE 260

232 130. LIST OF REFERENCES Bielfeldt, Hans Holm. 1965. Rcklufiges Wrterbuch der russischen Sprache der Gegenwart. Berlin: Akademie-Verlag Booij, Geert. 1983. French C/ alternations, extrasyllabicity and lexical phonology. The Linguistic Review, 3, 181-207. Burton, Martha W. & Karen E Robblee. 1997. A Phonetic Analysis of Voicing Assimilation in Russian. Journal of Phonetics, 25, 97-114. Chen, M. 1970. Vowel length variation as a function of the voicing of the consonant environment. Phonetica, 22, 129-159. Clements, George N. 1987. Phonological feature representation and the description of intrusive stops. In A . Bosch, B. Need, and E. Schiller (eds.), CLS23: Parasession on Autosegmental and Metrical Phonology (pp. 29-50). Chicago: Chicago Linguistics Society. Clements, George N. 1988. The sonority cycle and syllable organization. In W. Dressler,; H. Luschutzky, O. Pfeiffer, and J. Rennison (eds.), Phonologica 1988. Cambridge: Cambridge U. Press. Clements, George N. 1990. The role of the sonority cycle in core syllabification. In J. Kingston and M. Beckman (eds.), Papers in Laboratory Phonology I: Between the Grammar and Physics of Speech (pp. 282-333). Cambridge: Cambridge University Press. Clements, G.N. and S. Jay Keyser. 1983. CV Phonology: A Generative Theory of the Syllable. Cambridge, MA: MIT Press. Ct, Marie-Hlne. 2000. Consonant Cluster Phonotactics: A Perceptual Approach. PhD dissertation. Massachusetts Institute of Technology. De Jong, Daan. 1988. Sociolinguistic Aspects of French Liaison. PhD dissertation. Vrije Universiteit Amsterdam. Dell, Franois and Mohamed Elmedlaoui. 1985. Syllabic consonants and syllabification in Imdlawn Tashlhiyt Berber. Journal of African Languages and Linguistics, 7, 105-

PAGE 261

233 Flemming, Edward Stanton. 1995. Auditory Representations in Phonology. PhD dissertation. University of California, Los Angeles. Fudge, Erik. 1969. Syllables. Journal of Linguistics, 23, 253-287. Grammont, Maurice. 1933. Trait de phontique. Paris: Delagrave. (Repr., 1965). Greenberg, J. H. 1978. Some generalizations concerning initial and final consonant clusters. In J. H. Greenberg (ed.) 243-79. Guion, Susan. 1997. Velar Palatalization: Coarticulation, Perception, and Sound Change. PhD dissertation. University of Texas at Austin. Hankamer, Jorge and Judith Aissen. 1974. The sonority hierarchy. In M. W. Lagalay, R. A. Fox, and A. Bruck (eds.), Chicago Linguistic Society10: Parasession on Natural Phonology (pp. 131-145). Chicago: CLS. Hayes, Bruce. 1999. Phonetically driven phonology: the role of Optimality Theory and Inductive Grounding. In M. Darnell, E. Moravscik, M. Noonan, F. Newmeyer, and K. Wheatly, (eds.), Functionalism and Formalism in Linguistics, Volume I: General Papers (pp. 243-285). Amsterdam: John Benjamins. Hock, Hans Henrich. 1985. Regular Metathesis. Linguistics, 23, 529-546. House, A.S. and G. Fairbanks. 1953. The influence of consonant environment upon the secondary acoustical characteristics of vowels. Journal of the Acoustical Society of America, 25, 105-113. Hume, Elizabeth and David Odden. 1996. Reconsidering [consonantal]. Phonology, 13, 345-377. Jakobson, Roman and Morris Halle. 1956. Fundamentals of Language. New York: Mouton Publishers. Jespersen, Otto. 1904. Lehrbuch der Phonetik, Leipzig and Berlin. Kawasaki-Fukumori, Haruko. 1992. An Acoustical Basis for Universal Phonotactic Constraints. Language and Speech, 35, 73-86. Keating, Patricia A. 1983. Comments on the jaw and syllable structure. Journal of Phonetics, 11, 401-406. Kenstowicz, Michael. 1994. Phonology in Generative Grammar. Oxford: Blackwell. Kenstowicz, Michael. 1997. Quality-sensitive stress. Rivista di Linguistica, 9, 157-188,

PAGE 262

234 Kurowski, Kathleen and Sheila E. Blumstein. 1984. Perceptual integration of the murmurand formant transitions for place of articulation in nasal consonants. Journal of the Acoustical Society of America 76, 383-390. Ladefoged, Peter. 1993. A Course in Phonetics, 3rd Edition. Fort Worth: Harcourt Brace College Publishers. Lavoie, Lisa. 2000. Phonological Patterns and Phonetic Manifestations of Consonant Weakening. PhD dissertation. Cornell University. Lehiste, I. 1964. Acoustical characteristics of English selected consonants. International Journal of American Linguistics, 30 (No. 3, Part 4). Lindblom, B. 1983. Economy of speech gestures. In P. F. Macneilage (ed.) The Production of Speech (pp. 217-245). New York: Springer-Verlag. McCarthy, John and Alan Prince. 1995. Faithfulness and reduplicative identity. In J. Beckman, L. Walsh Dickey, and S. Urbanczyk (eds.) University of Massachusetts Occasional Papers in Linguistics 18, 249-384. McGovern, K and W. Strange. 1977. The perception of /r/ and /l/ in syllable-initial and syllable-final position. Perception and Psychophysics, 21, 162-170. Morelli, Frida. 1999. "Are s+STOP clusters really special?" in O.Fujimura B.Joseph and B.Palek (eds.) Proceedings of the 1998 Linguistics and Phonetics Conference. The Karolinum Press, Prague: 137-161. Murray, Robert W. and Theo Vennemann. 1983. Sound change and syllable structure in Germanic phonology. Language, 59, 514-528. Ohala, John J. 1990. Alternatives to the Sonority Hierarchy for Explaining Segmental Sequential Constraints. Papers from the Regional Meetings, 2, 319-338. Price, Patti Jo. 1980. Sonority and Syllabicity: Acoustic Correlates of Perception. Phonetica, 37, 327-343. Prince, Alan, and Paul Smolensky 1993. Optimality Theory. Tech. Rep. #2 of the Rutgers Center for Cognitive Science, Rutgers, NJ. Raphael, L.1972. Preceding vowel duration as a cue to the perception of the voicing characteristic of word-final consonants in English. Journal of the Acoustical Society of America, 51, 1296-1303. Rubach, Jerzy and Geert Booij. 1990. Syllable structure assignment in Polish. Phonology, 7, 118-158.

PAGE 263

235 Saussure, Ferdinand de. 1914. Cours de Linguistique Gnrale. Lausanne and Paris: Payot. Selkirk, Elisabeth. 1984. On the major class features and syllable theory. In Aronoff and Oehrle (eds.), Language Sound Structure. Cambridge, MA: MIT Press. Sievers, E. 1893. Grundzuge der Phonetik. Leipzig: Breitkopf and Hartel. Steriade, Donca 1982. Greek Prosodies and the Nature of Syllabification. Ph.D. dissertation, MIT. Steriade, Donca. 1999 "Alternatives to the syllabic interpretation of consonantal phonotactics," in O.Fujimura B.Joseph and B.Palek (eds.) Proceedings of the 1998 Linguistics and Phonetics Conference. The Karolinum Press, Prague: 205-242. Wedel, E. and A.S. Romanov. 1964. Romanov’s Russian-English/English-Russian Dictionary. New York: Washington Square Press. Weijer, Jeroen van de. 1995. Continuancy in liquids and in obstruents. Lingua, 96, 45-61. Whitney, W. D. 1865. The relation of vowel and consonant. Journal of the American Oriental Society, 8. Reprinted in W.D. Whitney, Oriental and Linguistic Studies, Second Series, Charles Scribner's Sons, New York: 1874. Zec, Draga. 1994. Sonority Constraints on Prosodic Structure. New York: Garland Publishing. Zec, Draga. 1995. The role of moraic structure in the distribution of segments within syllables. In J. Durand and F. Katamba (eds.), Frontiers of Phonology: Atoms, Structures, Derivations. New York: Longman. Zsiga, Elizabeth C. 2000. Phonetic alignment constraints: Consonant overlap and palatalization in English and Russian. Journal of Phonetics, 28, 69-102.

PAGE 264

BIOGRAPHICAL SKETCH Jodi Bray was born in Queens, New York in 1971. She spent most of her younger years in Bethlehem, PA, graduating from Liberty High School in 1989. For her college education, she decided to attend the University of Florida in Gainesville, Florida where she earned a degree in English Literature in 1993. She then worked in Princeton, New Jersey for two years as a research assistant at Response Analysis Corporation in the Social and Policy department. When she chose to continue her education, she returned to the University of Florida earning an MA in 1996 and a PhD in 2001 from the Program in Linguistics specializing in phonology and phonetics. While earning her degree, she spent several semesters teaching introductory linguistics and English as a Second Language courses. She also worked for one semester with the aid of a College of Liberal Arts Threadgill Dissertation Fellowship. She was involved in university activities serving as president of the linguistics club and as a representative to the Graduate Student Council. The research in this dissertation represents her work investigating sonority and the acoustic cues of segments in word-final consonant clusters. This research was completed with the supervision of Caroline R. Wiltshire. 236