Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Application of nonlinear phonological theory to intervention with six phonologically disordered children Bernhardt, Barbara May 1990

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-UBC_1990_A1 B48.pdf [ 47.18MB ]
Metadata
JSON: 831-1.0098745.json
JSON-LD: 831-1.0098745-ld.json
RDF/XML (Pretty): 831-1.0098745-rdf.xml
RDF/JSON: 831-1.0098745-rdf.json
Turtle: 831-1.0098745-turtle.txt
N-Triples: 831-1.0098745-rdf-ntriples.txt
Original Record: 831-1.0098745-source.json
Full Text
831-1.0098745-fulltext.txt
Citation
831-1.0098745.ris

Full Text

APPLICATION OF NONLINEAR PHONOLOGICAL THEORY TO INTERVENTION WITH SIX PHONOLOGICALLY DISORDERED CHILDREN By BARBARA MAY BERNHARDT B.A., The University of British Columbia, 1969 M.Sc, The University of British Columbia, 1972 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (School of Audiology and Speech Sciences) We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA September 1990 © Barbara May Bernhardt, 1990 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. The University of British Columbia Vancouver, Canada DE-6 (2/88) ABSTRACT APPLICATION OF NONLINEAR PHONOLOGICAL THEORY TO INTERVENTION WITH SIX PHONOLOGICALLY DISORDERED CHILDREN The purpose of this investigation was to examine the utility of nonlinear phonological frameworks for designing and executing an intervention program with phonologically disordered children. Six such children between the ages of 3 and 6 years participated in the study three times a week over three consecutive six-week blocks. The following general questions were addressed: 1. Will nonlinear phonological frameworks help to predict logical and attainable intervention goals for phonologically disordered children? 2. Are the separate prosodic and segmental levels of representation of nonlinear phonology psychologically real? 3. If the 'prosodic tier' has some observable clinical reality, will there be a difference in proportion and rate of syllable/word shapes acquired as a result of intervention methods that contrast the onset and rime versus those that utilize the mora a constituent? 4. If the 'segmental/melodic tier' has some observable independence, is there any advantage to be gained from targeting specified features at 'higher' versus lower' levels in the feature hierarchy in phonemic inventory intervention? An alternating block, mulitiple baseline design (counterbalanced over six single subjects) provided an opportunity to investigate the above questions. Within each six-week block, three week periods were devoted in turn to prosodic (syllable structure) training and segmental training. Prosodic subblocks were divided into two four-session sunblocks to contrast developmental change for targets presented as moraic constituents versus onset-rime constituents. Segmental ii periods were divided into two four-session subblocks to contrast developmental change for features from higher and lower levels in the feature hierarchy. Analyses during and after the study demonstrated the following with respect to the four research questions: 1. The nonlinear frameworks provided a logical model for deriving attainable intervention goals. All of the children became intelligible by the end of the project as a result of attaining the goals determined by nonlinear phonological theory. 2. Rate of attainment of syllabic and segmental goals differed, with a faster rate of change for syllabic goals overall, suggesting independence of segmental and prosodic tiers, and possible dominance of the prosodic tier. Interactions between tiers were also observed, suggesting that they are interdependent as well as autonomous. 3. Moraic and onset-rime condition quantitative results were virtually equivalent, but some qualitative differences appeared which had relevance for the each of the theories with respect to status of the onset, word-final consonants, and epenthesis. 4. Higher level features in the feature hierarchy tended to be acquired before lower level features. The nonlinear phonological frameworks stimulated a successful intervention study. Evidence gained through this study in turn contributes to the understanding of the nonlinear constructs. iii TABLE OF CONTENTS ABSTRACT ii. LIST OF TABLES xiv. LIST OF FIGURES xvii. ACKNOWLEDGEMENTS xxi. DEDICATION xxii. Chapter 1. INTRODUCTION 1 Nonlinear Phonology 2 Nonlinear Phonology in the Tradition of Generative Grammar 2 Assumptions Particular to Nonlinear Phonology 3 Phonological Process Approach to Intervention with Phonologically Disordered Children 13 Theoretical Deficiences of the Phonological Process Approach .... 14 Clinical Deficiencies of the Phonological Process Approach 15 Nonlinear Frameworks and Developmental/Clinical Phonology 16 2. METHOD 19 Subjects 19 Subject Recruitment and Participation 20 Subject Characteristics and Matching 20 Design 21 Assessment and Analysis Procedures During the Intervention Period. 24 Initial Assessment 24 Analysis during the S tudy 27 iv Intervention 27 Segmental Condition Target Selection 27 Syllabic Condition Target Selection 29 General Outline for Training Sessions 29 Treatment Methodology Specific to the Syllabic Condition 30 Analysis Following the Intervention Period 34 Feature Hierarchy Analysis 35 Syllabic Data 36 Results Reporting 37 3. SUBJECT 1: SUMMARY 38 Case History Information 38 Subject Characteristics 38 Prior Treatment 38 Phonological System at Initial Assessment 40 Phonotactics: Syllables and Words 41 Segmental Development and Feature Hierarchy 45 Intervention Plan: Block 1 49 Segmental Intervention Plan 49 Syllabic Intervention Plan 50 Results of Block 1 Intervention 51 Segmental Goals 51 Syllabic Goals 53 Intervention Plan: Block 2 54 Segmental Intervention Plan 54 Syllabic Intervention Plan 55 Results of Block 2 Intervention 56 Segmental Goals 58 Syllabic Goals 61 Block 3 and Beyond 63 v Discussion 64 Will Nonlinear Phonological Frameworks Help to Predict Logical and Attainable Goals for Phonologically Disordered Children? 64 Conversational Intelligibility Gains 64 Language Test Results 65 Are the Separate Prosodic and Segmental Levels of Representation of Nonlinear Phonology Psychologically Real? 67 Onset-Rime versus Moraic Constituency and Intervention? 67 Nonmatches and Syllable Theories at Assessment 68 Intervention Data and Syllable Theory 68 Higher versus Lower Level Features in Development? 72 Block 1 72 Block 2 72 Summary 73 4. SUBJECT 2: SUMMARY 75 Case History Information 75 Subject Characteristics 75 Prior Treatment 75 Phonological System at Initial Assessment 75 Phonotactics: Syllables and Words 75 Segmental Development 78 Intervention Plan: Block 1 82 Segmental Intervention Plan 83 Syllabic Intervention Plan 84 Results of Block 1 Intervention 85 General Results 85 Segmental Goals 86 Syllabic Goals 89 Intervention Plan: Block 2 93 vi Segmental Intervention Plan 93 Syllabic Intervention Plan 94 Results of Block 2 Intervention 95 Segmental Goals 95 Syllabic Goals 100 Block 3 and Beyond 104 Discussion 104 Will Nonlinear Phonological Frameworks Help to Predict Logical and Attainable Goals for Phonologically Disordered Children? 104 Conversational Intelligibility Gains 104 Language Test Results 108 Are the Separate Prosodic and Segmental Levels of Representation of Nonlinear Phonology Psychologically Real? 108 Interaction of the Syllabic and Skeletal Tiers 109 Onset-Rime versus Moraic Constituency and Intervention? 109 Higher versus Lower Level Features in Development? I l l Summary I l l 5. SUBJECT 3: SUMMARY 113 Case History Information 113 Subject Characteristics 113 Prior Treatment 113 Phonological System at Initial Assessment 113 Phonotactics: Syllables and Words 113 Segmental Development 121 Intervention Plan: Block 1 128 Syllabic Intervention Plan 128 Segmental Intervention Plan 130 Results of Block 1 Intervention 133 General Results 133 Syllabic Goals 133 vii Segmental Goals 142 Intervention Plan: Block 2 144 Syllabic Intervention Plan 144 Segmental Intervention Plan 146 Results of Block 2 Intervention 147 General Results 147 Syllabic Goals 148 Segmental Goals 153 Block 3 and Beyond 153 Discussion 154 Will Nonlinear Phonological Frameworks Help to Predict Logical and Attainable Goals for Phonologically Disordered Children? 154 Conversational Intelligibility Gains 154 Language Test Results 157 Are the Separate Prosodic and Segmental Levels of Representation of Nonlinear Phonology Psychologically Real? 158 Differences between Sylalbic and Segmental Conditions 158 Interaction of the Syllabic and Segmental Tiers 159 Onset-Rime versus moraic constituency and intervention? 159 Quantitative Comparisons of Moraic and Onset-Rime Conditions. 160 Qualitative Comparisons of Moraic and Onset-Rime Conditions .. 160 Higher versus Lower Level Features in Development? 162 Trained Targets: Block 1 162 Trained Targets: Block 2 163 Generalizations and Relevance of Features 163 Summary 164 6. SUBJECT 4: SUMMARY 166 Case History Information 166 Subject Characteristics 166 Prior Treatment 166 viii Phonological System at Initial Assessment 168 Phonotactics: Syllables and Words 168 Segmental Development 172 Intervention Plan: Block 1 180 Syllabic Intervention Plan 180 Segmental Intervention Plan 183 Results of Block 1 Intervention 184 Syllabic Goals 185 Segmental Goals 191 Intervention Plan: Block 2 194 Syllabic Intervention Plan 194 Segmental Intervention Plan 197 Results of Block 2 Intervention 197 Syllabic Goals 197 Segmental Goals 201 Block 3 and Beyond 203 Discussion 204 Will Nonlinear Phonological Frameworks Help to Predict Logical and Attainable Goals for Phonologically Disordered Children? 204 Conversational Intelligibility Gains 205 Language Test Results 207 Are the Separate Prosodic and Segmental Levels of Representation of Nonlinear Phonology Psychologically Real? 209 Onset-Rime versus Moraic Constituency and Intervention? 210 Higher versus Lower Levels Feature in Development? 212 Summary 213 7. SUBJECT 5: SUMMARY 215 Case History Information 215 Subject Characteristics 215 Phonological System at Initial Assessment 218 ix Phonotactics: Syllables and Words 218 Segmental Development 227 Intervention Plan: Block 1 231 Syllabic Intervention Plan 231 Segmental Intervention Plan 232 Results of Block 1 Intervention 233 Syllabic Goals 234 Segmental Goals 238 Intervention Plan: Block 2 241 Syllabic Intervention Plan 241 Segmental Intervention Plan 242 Results of Block 2 Intervention 242 Syllabic Goals 242 Segmental Goals 247 Block 3 and Beyond 249 Discussion 250 Will Nonlinear Phonological Frameworks Help to Predict Logical and Attainable Goals for Phonologically Disordered Children? 250 Conversational Intelligibility Gains 250 Language Test Results 251 Are the Separate Prosodic and Segmental Levels of Representation of Nonlinear Phonology Psychologically Real? 253 Onset-Rime versus Moraic Constituency and Intervention? 254 Higher versus Lower Level Features in Development? 255 Summary 255 8. SUBJECT 6: SUMMARY Case History Information 257 Phonological System at Initial Assessment 257 Phonotactics: Syllables and Words 257 Segmental Development 266 x Intervention Plan: Block 1 270 Segmental Intervention Plan 270 Syllabic Intervention Plan 271 Results of Block 1 Intervention 272 Syllabic Results 273 Segmental Goals 281 Intervention Plan: Block 2 284 Segmental Intervention Plan 284 Syllabic Intervention Plan 285 Results of Block 2 Intervention 286 Segmental Goals 286 Syllabic Goals 292 Block 3 and Beyond 294 Discussion 294 Will Nonlinear Phonological Frameworks Help to Predict Logical and Attainable Goals for Phonologically Disordered Children? 294 Conversational Intelligibility Gains 294 Language Test Results 298 Are the Separate Prosodic and Segmental Levels of Representation of Nonlinear Phonology Psychologically Real? 299 Onset-Rime versus Moraic Constituency and Intervention? 300 Higher versus Lower Level Features in Development? 301 Summary 302 9. DISCUSSION 305 Between-Subject Comparisons 305 Will Nonlinear Phonological Frameworks Help to Predict Logical and Attainable Goals for Phonologically Disordered Children? Are the Separate Prosodic and Segmental Levels of Representations of Nonlinear Phonology Psychologically Real? 305 Onset-Rime versus Moraic Constituency and Intervention? 314 Higher versus Lower Levels Feature in Development? 323 xi Limitations of the Study 333 Sample Size 333 Design 333 Extraneous Variables 334 Syllabic/Segmental Condition Contrast 334 Subblock Conditions 334 Future Research 335 Summary 336 Implications for Intervention 337 Intervention and Phonological Theory 338 NOTES REFERENCES 341 APPENDIX 1: BASIC WORD LIST FOR ELICITATION 345 APPENDIX 2: FEATURE SPECIFICATIONS FOR ADULT ENGLISH 347 APPENDIX 3: INTERTRANSCRIBER RELIABILITY DATA FOR Si, S4, AND SI 354 APPENDIX 4: MATCH AND NONMATCH DATA FOR ALL SUBJECTS 359 Table 1.1.--S1: Assessment and Block 1 Proportional Match Data ... 360 Table 1.2.--S1; Block 2 Proportional Match Data 361 Table 2.1.-S2: Proportional Match Data: Assessment and Block 1.. 362 Table 2.2.-S2: Block 1 Session Data for Syllabic Condition 364 Table 2.3.-S2: Block 2 Session Data for Syllabic Condition 366 Table 3.1.-S3: Proportional Match Data: Assessment and Block 1.. 370 Table 3.2.--S3: Proportional Matches in Block 1 and Block 2 372 Table 4.1.--S4: Proportional Matches: Assessment and Block 1 374 Table 4.2.--S4: Proportional Match Data: Block 2 376 Table 4.3.-S4: Diphthongs: Blocks 1 and 2 377 Table 5.1.--£5: Proportional Match Data: Assessment/Block 1 379 Table 5.2.--S5: Proportional Match Data: Block 2 381 xii Table 6.1.--S6: Proportional Match Data: Assessment and Block 1.. 383 Table 6.2.--£6_: Proportional Match Data: Block 2 385 APPENDIX 5: PHONETIC TRANSCRIPTIONS FOR ALL SUBJECTS 387 Subject 1 389 Subject 2 399 Subject 3 409 Subject 4 424 Subject 5 440 Subject 6 457 xiii LIST OF TABLES Table Page 2.1. Main characteristics of subjects during study 21 2.2. Order of treatment conditions for each subject within Blocks 1 and 2 23 2.3. Sample feature chart showing match and nonmatch data and feature differences 36 3.1. Subject information for S_j_ 39 3.2. £1: Syllable-initial clusters at Initial Assessment and at the end of Block 1.. 42 3.3. Si: Summary of singleton segmental acquisition at Initial Assessment characterized by sound class, matches and nonmatches according to word position, and feature differences 47 3.4. £1: Proportional matches for onset-rime and moraic targets across the study cast for Kolmogorov-Smirnov test 70 4.1. Subject information for S2 76 4.2. S2: Summary of singleton segmental acquisition at Initial Assessment 79 4.3. £2: Language test scores at Initial Assessment and end of Block 3 108 4.4. £2: Proportional matches for onset-rime and moraic targets across the study cast for Kolmogorov-Smirnov test 110 5.1. Subject information for S_3_ 114 5.2. S3: Proportion of spontaneous, echoic and prompted utterances per adult word shape '. 115 5.3. S3: Diphthong realization in various syllable/word types at Initial Assessment 118 5.4. £3_: Summary of singleton consonant acquisition at Initial Assessment 123 5.5. S3: Proportional matches for onset-rime and moraic targets across the study cast for Kolmogorov-Smirnov test 160 6.1. Subject information for £4 166 6.2. £4j Summary of singleton segmental acquisition at Initial Assessment 174 6.3. £4: Proportional matches for onset-rime and moraic targets across the xiv study cast for Kolmogorov-Smirnov test 211 7.1. Subject information for £5_ 216 7.2. £5_: Language test results at Initial Assessment and end of Block 3 217 7.3. S5_: WI consonant slot matches by cluster category and match type at Initial Assessment and at the ends of Blocks 1 and 2 220 7.4. £5_: WI /s/-cluster realizations at Initial Assessment and at the ends of Blocks 1 and 2 221 7.5. £5_: SI /I/-, /w/- and /j/-cluster realizations at Initial Assessment and at the ends of Blocks 1 and 2 225 7.6. £5_: SI/r/-cluster realizations at Initial Assessment and at the ends of Blocks 1 and 2 226 7.7. £5_: Summary of singleton segmental acquisition at Initial Assessment 228 7.8. £5_: Proportional matches across the study for onset-rime and moraic trained skeletal slots cast for Kolmogorov-Smirnov test 255 8.1. Subject information for £j6_ 258 8.2. £6_: Number of substitutions over total nonmatches (and percentage) for singleton consonants, by substitution type, in WI, SIWW and WF positions at Initial Assessment 259 8.3. £6_: WI singleton matches and substitutions at Initial Assessment 261 8.4. £6_: Proportion of glottal and [h]([C]) onsets in vowel-initial child forms at Initial Assessment 262 8.5. £6j WI cluster realizations at Initial Assessment 263 8.6. £6.: Substitutions for SIWW singleton consonants at Initial Assessment .... 264 8.7. £6_: Substitutions for WF singletons at Initial Assessment 265 8.8. £6_: Summary of singleton segmental acquisition at Initial Assessment 267 8.9. ££: Substitutions for SIWW singletons at end of Block 1 276 8.10. ££: Proportion of substitution types over total nonmatches for singletons within and across word positions at the end of Block 1 276 8.11. £6: Differences between Initial Assessment and Block 1 for singleton substitution types expressed as a proportion of total nonmatches 277 8.12. £6,: Substitutions for WF singletons at the end of Block 1 279 8.13. £6: WI singleton matches and substitutions at end of Block 1 282 8.14. £6: Substitution types for /s/ and /z/ singletons in WI, SIWW and WF xv positions at the end of Block 2 287 8.15. £6_: Proportional matches for onset-rime and moraic targets across the study cast for Kolmogorov-Smirnov test , 301 9.1. Difference in average proportional gains between syllabic and segmental conditions in Blocks 1 and 2, by subject 307 9.2. Feature specifications for /h/ and /V and Si's realizations of these in Blocks 1 and 2 325 9.3. Summary of intervention targets for subjects across blocks 327 9.4. Summary of intervention contrasts in terms of place and sound class and identification of faster rate of acquisition within those categories 329 xvi LIST OF FIGURES Figure Page 1.1. Alternate Representations of Word/Syllable Constituents in Onset-Rime and Moraic Theories 4 1.2. Feature Geometry for English 8 1.3. Regressive consonant assimilation in the absence of CV interaction 10 2.1. Schematization of Research Design 22 2.2. Stimuli Constituents for Onset-Rime and Moraic Subblock Condition Training 33 3.1. SI: Proportional Matches for Primary Training and Observation Targets at Initial Assessment 40 3.2. SI: Feature Geometry at Initial Assessment 48 3.3. S i : Proportional Gain in Matches for Syllabic and Segmental Targets between Initial Assessment and Block 1 Major Probes 52 3.4. S i : Proportional Matches for Syllabic Targets in Block 1 Minor Probe 54 3.5. S i : Proportional Matches for Syllabic and Segmental Targets in Block 2 Minor Probe 56 3.6. S i : Proportional Gain in Matches for Syllabic and Segmental Targets between Block 1 and Block 2 Major Probes 57 3.7. S i : Proportional Matches for Trained and Untrained Segmental Targets in Block 2 Major Probe 58 3.8. S i : Proportional Matches for Syllabic Target/st/ in Block 2 Sessions 60 3.9. S i : Developmental Progression in Proportional Matches for Trained Phonological Targets across the Intervention Period 65 3.10. S i : Feature Geometry at End of Block 2 66 3.11. S i : Proportional Differences between Moraic and Onset-Rime Match Scores for Minor Probes, Introductory Parts of Production Sessions, and Overall Session Data (Spontaneous and Echoic Responses) 69 4.1. S2: Feature Geometry at Initial Assessment 80 xvii 4.2. £2: Average Proportional Gain in Matches for Trained and Untrained Targets between Initial Assessment and Block 1 86 4.3. £2: Proportional Gain in Matches for Trained and Untrained Targets between Initial Assessment and Block 1 Major Probes 90 4.4. £2: Proportional Matches for Syllabic Targets in Block 1 Sessions 92 4.5. £2: Proportional Gain in Matches for Trained Targets between Block 1 and Block 2 Major Probes •. 96 4.6. £2: Proportional Gain in Matches between Block 1 and Block 2 Major Probes for Untrained Targets 97 4.7. £2: Proportional Matches in Block 2 Major Probe for Targets with Match Proportions Differing from Those Shown in Figures 4.5 and 4.6 98 4.8. £2: Average Proportional Gain in Matches for Trained and Untrained Targets between Block 1 and Block 2 Major Probes 99 4.9. £2: Proportional Matches for Syllabic Targets in Block 2 sessions 102 4.10. £2: Developmental Progression in Proportional Matches for Trained Segmental Targets across the Intervention Period 105 4.11. £2: Developmental Progression in Proportional Matches for Trained Syllabic Targets across the Intervention Period 106 4.12. £2: Feature Geometry at End of Block 2 107 5.1. £2: Proportional Matches for CVC and CVCV Word Shapes at Initial Assessment 116 5.2. £2: Feature Geometry at Initial Assessment 124 5.3. £3_: Proportional Gain in Matches for Syllabic and Segmental Targets between Initial Assessment and Block 1 Major Probe 134 5.4. £2: Proportional Matches for WF (CVC) Targets in Block 1 Major Probe .. 135 5.5. £2: Proportional Matches for SIWW (CVCV) Targets in Block 1 Major Probe 137 5.6. £2: Proportional Matches for Syllabic Targets in Block 1 Sessions 140 5.7. £3_: Proportional Matches for Trained and Untrained Segmental Targets in Block 1 Major Probe 143 5.8. £2: Proportional Matches for Syllabic and Segmental Targets (WI unless Otherwise Indicated) in Block 2 Major Probe 148 5.9. £2: Proportional Gain in Matches for Syllabic and Segmental Targets between Block 1 and Block 2 149 5.10. £2: Proportional Matches for Syllabic Targets in Block 2 Sessions 152 5.11. £2: Developmental Progression in Proportional Matches for Trained xviii Phonological Targets across the Intervention Period 155 5.12. £3j Feature Geometry at End of Block 2 156 6.1. £4: Feature Geometry at Initial Assessment 175 6.2. £4: Proportional WI Matches at Initial Assessment 176 6.3. £4: Proportional SIWW Matches at Initial Assessment 176 6.4. £4: Proportional WF Matches at Initial Assessment 177 6.5. S4: Proportional Gain in Matches for Syllabic and Segmental Targets between Initial Assessment and Block 1 Major Probes 184 6.6. S4:Proportional WF Matches in Block 1 Major Probe 186 6.7. £4: Proportional SIWW Matches in Block 1 Major Probe 187 6.8. S4: Proportional Matches for WI Training and Observation Targets in Block 1 major probe 191 6.9. S4: Proportional Gain in Matches for Syllabic and Segmental Targets between Block 1 and Block 2 major probes 198 6.10. £4: Proportional Matches for Syllabic Targets in Block 2 Major Probe 200 6.11. £4: Feature Geometry at End of Block 2 206 6.12. £4: Developmental Progression in Proportional Matches for Trained Phonological Targets across the Intervention Period 207 7.1. £5: Proportional Matches for Two-Element (CC) Clusters with and without Short Epenthesis, Overall and by Cluster Type, at Initial Assessment — 222 7.2. £5.: Proportional Matches for Two-Element (CC) Clusters with and without Short Epenthesis, Overall and by Cluster Type, in Block 1 Major Probe. 222 7.3. S5_: Proportional Matches for Two-Element (CC) Clusters with and without Short Epenthesis, Overall and by Cluster Type, in Block 2 Major Probe. 222 7.4. £5_: Feature Geometry at Initial Assessment 229 7.5. £5_: Average Proportional Gain in Matches for Syllabic and Segmental Targets between Initial Assessment and Block 1 Major Probes 234 7.6. £5_: Proportional Gain in Matches for Trained and Untrained Segmental Targets between Initial Assessment and Block 1 Major Probes 240 7.7. £5_: Average Proportional Gain in Matches for Trained and Untrained Syllabic and Segmental Targets between Block 1 and Block 2 Major Probes 243 7.8. £5_: Developmental Progression in Proportional Matches for Trained Phonological Targets across the Intervention Period 251 7.9. £5_: Feature Geometry at End of Block 2 252 xix 8.1. £6_: Feature Geometry at Initial Assessment 268 8.2. £6_: Proportional Gain in Matches for Syllabic and Segmental Training Targets between Initial Assessment and Block 1 Major Probes 273 8.3. £6_: Proportional Gain in Matches for Trained and Untrained SIWW Syllabic Targets between Initial Assessment and Block 1 Major Probes 275 8.4. ££: Proportional Gain in Matches for Trained and Untrained WF Syllabic Targets (excluding Liquids) between Initial Assessment and Block 1 Major Probes 278 8.5. £6: Proportional Gain in Matches for Trained and Untrained WI Segmental Targets between Initial Assessment and Block 1 Major Probes 283 8.6. Proportional Gain in Matches for Syllabic and Segmental Training Targets between Block 1 and Block 2 Major Probes 288 8.7. £6: Proportional Gain in Matches for Trained and Untrained Segmental Targets between Block 1 and Block 2 Major Probes 289 8.8. £6: Average Proportional Gain in Matches for Segmental Targets between Block 1 and Block 2 Major Probes 290 8.9. £6: Developmental Progression in Proportional Matches for Trained Syllabic Targets (except Block 2 Targets fkj and /g/) across the Intervention Period 295 8.10. £6: Developmental Progression in Proportional Matches for Trained Segmental Targets and Block 2 Syllabic targets fk/ and /g/ across the Intervention Period 296 8.11. £6: Feature Geometry at End of Block 2 ... 297 9.1. Representation of Clusters — Adult Targets, Children's Reduced Forms, and Clusters Articulated with Unstressed Epenthetic Vowels 317 9.2. Comparative Derivations of £Ts [fw] Substitution for Obstruent-Approximant Clusters Based on Kaye et al. (1987) and a Proposed Alternative 318 9.3. Alternate Representations of £2's Block 1 Cluster Realizations in Onset-Rime and Moraic Theories 320 xx ACKNOWLEDGMENTS Many have made invaluable contributions to this dissertation, and I would like to take this opportunity to thank: The children and their families for enthusiastic participation in the study. Dr. Carolyn Johnson, thesis advisor for wisdom, patience, and kindness in supporting me throughout the doctoral program, with particular thanks for painstaking editing of this thesis. Dr. Pat Shaw, the cornmittee phonologist for inspiring a phonological project and for spending many hours reading drafts and explaining the theories. Dr. John Gilbert, committee member for the initial encouragement to enter the doctoral program and lessons on positivism. Patience Towler and staff, speech-language pathologists, Vancouver Health Department for providing facilities for carrying out the project, as well as for referring subjects. Monica Brekelmans and staff, Vancouver School Board, and Lynne Brown and staff, Sunny Hill Hospital for Children, local speech-language pathologists for referring subjects. Many local speech-language pathologists for moral support and part-time employment. Many friends for loyal support and encouragement. And especially my husband Louis, son Travis, daughter Carmine, my parents, late grandmother, and distant family — parents-in-law, and brother and family: This document is a tribute to your encouragement, understanding, and sacrifices over the past six years. xxi To Florence May Parker (1899-1989) For your love of children xxii CHAPTER 1 INTRODUCTION Because human communication involves the interaction of many factors which are both physical and mental, the clinical field of speech-language pathology relies on developments in related disciplines, from anatomy and physiology to those concerned with the mind and its constructs. For the last twenty years, clinicians have looked to linguistic theories for models of language and methods of data collection and analysis on which to base assessment and treatment methodologies for speech and language disorders. Grunwell states that "linguistic analysis is now a well-established technique in the range of assessment procedures employed by speech and language pathologists" (1985, 1). When linguists make major changes in the description and explanation of language, speech-language pathologists have an opportunity to search for more effective assessment and intervention methodologies by deriving implications from the theoretical changes. Clinicians maximize their opportunities for successful application when they select the most robust of theories. A linguistic theory attains a robust status primarily through its capacity to describe the synchronic facts of a language more parsimoniously than previous theories. If a theory can also explain the facts of a language, it will be preferred among a group of descriptively adequate ones.1 Once 'internal evidence' (synchronic data from one adult language) has demonstrated a theory's adequacy, its robustness becomes enhanced through 'external evidence', most significantly from other unrelated languages, but also from historical change, language acquisition, and language breakdown (Kenstowicz and Kisseberth 1979,139-140). The best opportunity for successful clinical application occurs when such external evidence has begun to enhance a theory's status. As a result of the application, a clinician may be able to provide the linguist with more evidence relevant to the 1 2 theory, both factual and utilitarian. This evidence will not prove or disprove the theory, but may elucidate aspects of it that are less accessible through internal analysis. The research reported in this dissertation examines the utility of current nonlinear phonological frameworks for planning and executing an intervention program with six phonologically disordered children. In this chapter I will first outline the basic characteristics of nonlinear phonology and then relate principal concerns regarding phonological process analysis, the most prevalent phonological approach to intervention. Nonlinear Phonology Nonlinear Phonology in the Tradition of Generative Grammar Nonlinear phonology2 developed in the late 1970's, with the major impetus provided by Goldsmith's dissertation on tone (1976). Based in generative phonology, nonlinear frameworks adhere to many of the tenets of the generative grammar tradition. The following assumptions from generative grammar underlie the work for this thesis. 1. In linguistic description it is desirable to decrease the number of language-specific rules and processes to a minimum, putting the "burden" for explaining language development on structures and principles of the innate system (Chomsky 1981). 2. For all grammatical 'parameters' within a linguistic component (such as syntax or phonology), there is an 'unmarked' versus 'marked' option. Theories of markedness vary in their assumptions and definitions. However, most assume that the unmarked value for a parameter is the one available from innately endowed 'Universal Grammar' (UG). Marked values are assumed to represent the more unusual of the options and as such need 'positive evidence' from the input language. Pharyngealized consonants and complex segments such as /gu/ are exotic examples of marked options. Until a child hears them used in his language community, his speech will reflect the innate endowment — nonpharyngealized articulations of single segments. Defining which options are marked and which unmarked across languages becomes more difficult for less exotic phenomena, however, and continues to engage the attention of linguists. 3 3. Although there may be interaction among linguistic components for language processing, each is sufficiently autonomous to study as an independent system (Clements 1989b). Assumptions Particular to Nonlinear Phonology Nonlinear phonology differs (at a minimum) from original conceptualizations of generative phonology by (1) the concept of hierarchical representation of both prosodic and segmental information rather than strictly linear (sequential) representation, and (2) the concept of tier association as a cause of and constraint on phonological "rules." Elaboration of these concepts follows below. Hierarchical Representation Within the phonological component, there are a number of hierarchical levels ('tiers'). Separate tiers have been posited for larger prosodic units, such as the word, the foot, and the syllable, as well as for smaller melodic units, such as the segment and its composite features. The phonological tiers are hypothesized to be logically and empirically autonomous, while associated with each other in principled ways. Different phonologists have put forward a number of subtheories regarding those hypothetically autonomous aspects of phonological organization and behavior. Subtheories pertinent to the dissertation are discussed briefly. A. Prosodic level With respect to the 'top' of the hierarchy, phonologists disagree on (1) the status of the syllable as a constituent, and (2) the components of the syllable or its equivalent. Kaye et al. (1987) consider onset-rime strings, rather than the syllable, to be the primes at the level of prosodic structure between feet and segments (see figure 1.1). The onset (O) is a projection of the prevocalic consonant(s), and the rime (R) a projection of the nucleus (N) — itself a projection of the vowel or other most sonorant segment. Postvocalic consonants are adjoined to the rime node. According to this view, the onset and the rime are independent constituents with their own set of phonological functions, both within themselves and in relation to each other. Onset-rime representations: According to Kaye, Lowenstamm and Vergnaud (1987) O R I N I C V CV unit O R I C V CVC unit O -R O- -R N / \ V V N / \ C V V c c CVV unit CVVC unit Moraic representations: According to Hayes (1987) o I * , M o / \ M M o M M C ' V * * CV unit C ' V V or C CVV or CVC unit C V V V CVVC unit Key: O: Onset N: Nucleus V V = diphthong R- Rime O : Syllable (/ou/,/ei/,/au/,/ai/,/oi/) M : Mora V = lax vowels and tense vowels (Ji/, /u/) * In Hayes (1989), the prevocalic consonant is adjoined directly to the syllable node, but the research undertaken here utilized Hayes (1987), in which the prevocalic consonant is adjoined to the mora. The different level of adjunction does not crucially bear on the empirical consequences of the present study. * V = lax vowel or tense vowel (i.e., only I'll, /u/. See appendix 2.) V V = diphthong and diphthongized nonhigh tense vowels. This is a notable departure from Hayes in which long vowels are equivalent to diphthongs (i.e., V: = VV). In intervention sessions which focused on final consonants, vowels were all lenghtened in the oral presentation of stimuli, making a length distinction between tense and lax vowels indiscriminate. + According to stress rules, English is a 'weight by position' language, which means that the first postvocalic consonant may project a mora. In this study closed monosyllables with single vowels were considered bimoraic. Figure 1.1. Alternate representations of word/syllable constituents in onset-rime and moraic theories 4 5 Government ("a binary asymmetric relation holding... between two skeletal positions" [Kaye et al. 1987, chapter 2, 1]) proceeds from left to right within each of the constituents (onset and rime) and from right to left between constituents (from the rime to the onset). In other words, government defines domains within which segments may associate to each other. Association may result in the appearance of changes to the governed segment, depending on the feature composition of the two segments. Other phonologists (Hayes 1987; Hyman 1985; McCarthy and Prince 1986) posit: (1) that the syllable is a constituent and prime of the prosodic tier, and (2) that prosodic 'weight units' or 'morae' (M) are the critical constituents of syllables for stress assignment and other phonological rules. Weight units are realized through vowels (or other resonants). Morae may also be realized through postvocalic consonants in 'weight-sensitive' languages, where stress assignment is sensitive to certain postvocalic consonants as well as to the vowel constituents. (For this dissertation we have assumed that English is a 'weight-sensitive' language and that closed monosyllabic words are bimoraic. For further discussion see the section on treatment methodology particular to the syllabic condition in chapter 2.) Morae are grouped together to form syllables with a postulated maximum of two morae per syllable. Syllables are then grouped into 'feet' and 'prosodic words'. Drawing on McCarthy and Prince's (1986) notion of 'minimal word', a bimoraic word is taken to be the basic unit available from UG at this prosodic level. McCarthy and Prince (1986) postulate the 'core syllable' CV as a basic unit at the syllabic level. Prevocalic and postvocalic consonants that do not contribute to the marking of syllable weight are simply adjoined to the mora or syllable (the level of adjunction being currently controversial). They do not form a particular constituent in and for themselves and, as such, have no particular independent phonological functions as prosodic constituents. A demonstration of some of the differences between the onset-rime and moraic theories is shown in their alternate representations of two possible congruent phonological phenomena in a child's phonology ~ missing final consonants and lack of diphthongs (see figure 1.1). The onset-rime theorist would claim that both of these structures require branching under the rime constituent. 6 A single onset constituent is a direct projection of the initial consonant; a single rime constituent is a direct projection of the nucleus, itself a projection of the vowel. Representation of one postvocalic consonant requires a first level branching under the rime node. Diphthongs require branching under the nucleus, a lower level branching than that for the first postvocalic consonant. For representation of a CVVC word with both a diphthong and a final consonant, branching is required at both the rime and nucleus levels. A child who uses neither final consonants nor diphthongs is continuing to operate with the minimal structure required for the universally available O-R unit. The moraic theorist would claim, alternatively, that the child who uses neither final consonants nor diphthongs in one syllable is operating with a minimal prosodic weight unit for a syllable ~ viz. one mora, or perhaps, a core syllable. Setting a parametric value licensing two morae per syllable, for a weight-sensitive language such as English, would then license both final consonants and diphthongs, since both the consonant and the second part of the vowel have phonological 'weight' as prosodic units. In order for diphthongs and postvocalic consonants (or single vowels and two postvocalic consonants) to appear in one syllable, the last consonant must be adjoined at the moraic or syllabic level. (The locus of consonant adjunction is currently under debate; in this dissertation I followed Hayes [1987] in which the locus of adjunction is the mora.) The foregoing differences might result in different predictions for development. Following moraic theory, one might expect a child to acquire diphthongs and single final consonants (or diphthongs and consonants within one syllable at the same time as single vowels and two consonants in a syllable) in the same time period when learning a weight-sensitive language, because the child would learn the constitution of bimoraic syllables in the language. Following onset-rime theory, one might expect a notable time lag between acquisition of final consonants and diphthongs because of the need to 'set parameters' for multiple levels of branching, i.e., a representational complexity. If branching of the basic O-R string happens first, as a primary type of branching (the rime subdividing into nucleus and adjoined consonant), the child might acquire 7 final consonants first. Since the rime depends upon the nucleus, however, nuclear branching might take precedence over rime branching, which would result in prior acquisition of diphthongs. B. Segmental Level Proponents of nonlinear feature representations currently agree on a basic tree structure, such as the one used as the basis for this study (see figure 1.2), modified from Sagey (1986)3. Because languages have different segmental inventories, specified representations of the individual feature systems will differ for languages, but the general framework and set of marked and unmarked features is assumed to be universal. The determination of the set of universal and language-specific features and their relative markedness is an ongoing enterprise in phonology. As figure 1.2 shows, the feature hierarchy hypothesis includes the concept that features group together into loci (nodes) with relationships of dominance or sequence. The groupings reflect the observation that "certain sets of features commonly act as single functional units with respect to phonological rules" (Clements 1989b, 1). In this dissertation, the intermediate nodes, i.e., Labial, Coronal, and Dorsal Place, were considered monovalent features dominating binary or privative terminal features relevant to that place. The nonterminal nodes are: 1. The Root node: This node links the segment to the prosodic tiers "above." Major class features [sonorant] and [consonantal] define the Root node (and thus the segment) as a vowel, consonant, liquid, or glide (McCarthy 1988). The Root node dominates the terminal features [continuant] and [nasal]. 2. The Laryngeal node: The Laryngeal node defines the glottal characteristics of the segment. It is linked to the Root node. Terminal features of this lower level node include those for [voice] and [spread glottis]. 3. The Place node: The Place node defines the oral cavity characteristics of the segment. It is a dependent of the Root node above but dominates the individual Place nodes — Labial, Coronal and Dorsal. Note: Marked features are indicated on terminal nodes with plus or minus specification. Labial, Coronal, and Dorsal nodes are considered monovalent. The Root node designates the segment. It links into the prosodic tiers above. (See appendix 2 for more detail.) •Affricates are designated as 'complex segments'. 'Complex' is a mnemonic describing the branching structure of affricates with respect to [continuant]. (The term is from Lombardy [1989] but does not imply her analysis.) Figure 1.2. Feature geometry for English 8 9 Clements (1989a,b) differentiates between Consonant-Place nodes and Vowel-Place nodes. (For this study the primary concern was development of consonants. Hence Place node is undifferentiated for consonants as opposed to vowels unless otherwise specified.) 4. The Labial Place node: The Labial Place node refers to lip articulations. It is linked to the higher level Place node and dominates its terminal feature [round]. 5. The Coronal Place node: The Coronal Place node refers to articulations of the tongue blade and tip. It is linked to the Place node above and dominates its terminal features [anterior] and [distributed]. Coronal Place is considered the universally unmarked or 'default' Place (Paradis and Prunet, forthcoming). When it does not dominate any terminal features, its specification is therefore assumed at the level of the general Place node. 6. The Dorsal Place node: The Dorsal Place node refers to articulations of the tongue dorsum. It is linked to the Place node above and dominates its terminal features [high], [low] and [back]. Tier Association As stated previously, tiers are assumed to be autonomous. However, they may associate to each other (with 'association lines'). There are basic principles for association of tiers. For example, (1) more than one consitutent may be linked to a constituent on another tier, (2) association lines linking constituents must not cross, and (3) direction of association is parameterized. Association principles constrain and explain phonological operations (processes or rules). Two of the fundamental hypotheses regarding basic phonological operations are: (1) that. they may only spread, delink, insert, or delete constituents, and (2) that these operations may apply only to adjacent constituents. For assimilation to take place, for example, the affected constituent must be adjacent to the 'spreading' constituent at the relevant level of hierarchical structure. Then it can receive the specification of the designated constituent's feature(s) by spreading. Hierarchical representations reveal adjacencies not apparent in surface representations. This is illustrated in figure 1.3, an example of a nonlinear description of regressive consonant assimilation. /d C a V g/ c Root node o [-son] \ [+cons] C-Place node O i l II Root node o / [+son] \ [-cons] V-Place ** node o Dorsal Place node o [+bk] [+lo] Root node o / [-son] \ [+cons]* ace C-P node "" : | ;  Dorsal Place node ' fl [+bk] [g g] * Only salient features for each example are shown. * * According to spreading conventions, if two adjacent segments have identical specification, spreading of those features from one of them to a third is blocked by the second. (Association lines must not cross.) Following Clements (1989a, b) consonant and vowel Place nodes are independent and therefore spreading of Dorsal Consonant-Place node to an adjacent C-Place node is possible because the Vowel-Place node is on an independent (and therefore nonadjacent) tier. Other representations are possible, but this is used to illustrate consonant adjacency, spreading, and linking. Figure 1.3. Regressive consonant assimilation in the absence of CV interaction 10 11 When /dag / (dog1) is realized as [gag], but /dal/ (doll) is realized as [dal], assimilation results from consonant-consonant interaction, rather than consonant-vowel interaction. The consonants, although separated by a vowel in the surface form, have to be adjacent at some deeper level of representation in order for assimilation, i.e., spreading of features, to take place. If they were adjacent at the level of Consonant-Place (C-Place) node (Clements 1989a, b), the Dorsal Place node of /g/ could spread to the C-Place node for /d/ (an 'empty node' for reasons of 'underspecification'). If, on the other hand, vowels were tier-adjacent to consonants at all levels of representation, specified Dorsal Place node features of the vowels would block Dorsal Place consonant feature spreading (according to standard nonlinear interpretations of spreading, such as Archangeli [1988]). Dominance. Government and Underspecificaton A. Domains for government As a consequence of hierarchichal organization, some parts of the structure dominate others, circumscribing domains for phonological 'government'. In the example of [gag] for /dag/, the assimilation may reflect, at one level, a general principle of consonant government by word position — the postvocalic consonant dominates the prevocalic consonant (similar to Kaye, Lowenstamm, and Vergnaud's [1987] 'interconstituent government' from rime to onset). The dominance relationship may only appear phonetically as assimilation when segmental simplification is "required" for performance reasons (such as a young or disordered child's inability to produce two contiguous stops with different lingual places of consonant articulation within one syllable). Feature specification relationships between segments may interact with positional government to produce results such as assimilation as (see section C below). B. Underspecification A current hypothesis about feature specification is underspecification. Those who argue for underspecification claim that underlying representations of segments (or other phonological constituents) are minimal, containing information only about the idiosyncratic (distinctive) properties of the constituent. For example, a segment which is [+sonorant] may be considered underspecified for [voice]. The [+voice] feature is predictable, given the [+sonorant] feature. With respect to which features must be encoded in representations, there is convergence of opinion on some issues, but disagreement on others. Proponents of 'radical underspecification' (Kiparsky 1982; Pulleyblank 1986; Archangeli 1988) claim that the underlying representations include "only unpredictable values for features, ...[whereas] predictable values are inserted by rule during the course of the derivation" (Archangeli 1988,192). Proponents of 'contrastive underspecification' (Clements 1987; Steriade 1987) claim that underlying representations include values for features "only where that feature is being used to distinguish segments in the respective contexts; noncontrastive values are left blank" (Archangeli 1988,191). In this dissertation I have adopted a radical underspecification perspective. Appendix 2 lists the phonemes of English and the specified feature values assumed for each of them, and includes a more detailed discussion of markedness. (The reader will find reference to this appendix useful for comprehension of the data chapters.) According to the feature specifications hypothesized, the least specified consonant is A/ and the least specified vowel is the epenthetic or reduced vowel /s/. The only specified value for /t/ is postulated to be [+consonantal], since IxJ is unmarked in terms of all other consonantal features. Marked consonantal features for English, which do need to be specified, are assumed to be: [+consonantal], [+continuant], [+nasal], [+voice], [-anterior], [+distributed], and Labial or Dorsal Place. The only specified value for /s/ is hypothesized to be [+sonorant], since /a/ is a neutral vowel unmarked in terms of all other vowel features. Marked vowel features are assumed to be: [-voice], [-back], [+low] and [+round]. (See appendix 2 discussion on markedness.) In terms of development, input will both confirm default features and values as part of a language's phonology as well as present evidence for the specified values thereof. As the input evidence becomes sufficient and the child's phonological production capacity matures, one expects a child to add specified values gradually to the set of universally unspecified features. C. Underspecification and government The regressive assimilation example in figure 1.3 demonstrates a possible interaction of positional government and underspecification. In the case of /dag/, the Coronal Place node under which place features for /d/ are represented is considered the least marked or underspecified of the Place nodes for English (see appendix 2). The /g/, as a segment with a specified Place node (Dorsal), is 'more specified' than the 'empty' Place node of the IdJ, and as such is 'dominant' in specification terms. If positional government gives the postvocalic consonant dominance over the prevocalic consonant, and there is a dominance relationship in terms of feature specification, the features of the former may spread to the prevocalic position. In a developing phonological system, when representational and/or articulatory constraints block two places of consonant production within one syllable (or O-R unit), those segments which have specification can dominate the underspecified Coronal Place node, with specified features spreading to that empty node. Eventually, pressure to conform to the adult language and increasing representationaVarticulatory abilities result in accommodation of both specified and unspecified consonants within a syllable. Phonological Process Approach to Intervention with Phonologically Disordered Children For intervention with phonologically disordered children, the Natural Phonology theory with its phonological process analysis (Stampe 1969) is the most widely used application of phonological theory (e.g., Edwards and Bernhardt 1973; Ingram 1981; Hodson andPaden 1983; Grunwell 1985). Its use has resulted in improvements in assessment and intervention over structuralist theory/substitution analysis, because it focuses on generalizations about different aspects of a child's overall system. These generalizations lead to the formulation of therapeutic goals which affect more than one phoneme within the system. Effective programming notwithstanding, several aspects of the theory and its application seem insufficient, both with respect to description and explanation. 14 Theoretical Deficiencies of the Phonological Process Approach The theoretical deficiencies of the phonological process approach he in: 1. The assumption of adultlike underlying representations (UR's) derived from an intact adultlike perceptual system (Stampe, 1969). The possibility exists that children may have idiosyncratic underlying representations (Maxwell, 1984). Nettelbladt (1983) suggests, for example, that when a child's word is missing segments in the adult target, that child form may result from word shapes created by extraction of components of the adult target rather than from deletion processes within the child's own system. Not all of the adult word may be perceived and stored. That part of the adult form (phrase, word, syllable or segment) may be perceived and stored which conforms to the child's restricted syllable/word 'templates'. (The term 'template' is borrowed from nonlinear accounts of morphemic representation and word formation processes, and was introduced into accounts of child phonology by Menn [1978, 1983]). In early phases of development, one might assume that templates (and outputs) reflect the unmarked defaults provided by UG. 2. The assumed psychological reality of the processes themselves. Stampe (1969) suggested that 'natural processes' operate on the adultlike underlying representation to allow simplification for output. It may be a correct generalization about the observed data that the child's form [sip] for the input form /Jip/ demonstrates 'depalatalization', but there is no evidence that a 'process' has 'operated' on a 'representation'. 3. The lack of principled explanation of the occurrence of any of the processes. Whereas it is more than likely true that certain phonological phenomena are natural for humans, because of physical, physiological and (possibly) cognitive constraints, neither Stampe (1969) nor subsequent applications of his theory have offered any explanations why some sound changes are more natural than others. (Recently, Stevens and Keyser's [1987] enhancement theory may be bringing auditory and phonetic data to bear on the question of naturalness, however.) 4. The implications of development as a negative "progression" of process suppressions, limiting and ordering. In order to progress from realizing the UR /Jlp/ as [sip] to [Jip], a child's 15 depalatalized output has to "dedepalatalize." In order to progress from [th£lt] to [kh£lk] for cake, a child has to "back" the "fronted" IkJ. In order to realize final consonants, a child has to "undelete" a deleted consonant. This type of implied negative progression is reminiscent of "the child as Little Linguist," a syntactic developmental theory which arose from early applications of generative grammar, but which has been discounted in recent years by Pinker (1984) and others. For example, the child is not seen as first having deletion rules which remove adult syntactic markers and categories, and then "undeleting" rules which put them back. Development is seen more as a positive process. The innately available unmarked system acts as a passive constraint, allowing mappings of representations and outputs which conform to its default "specifications." At those points in time when maturation permits and evidence for alternate specifications is sufficiently overwhelming, representations are specified in concordance with markings of the input language and the range of outputs increases (see also Sadanandan 1987). Clinical Application Deficiencies of the Phonological Process Approach The lack of a constrained and coherent theory leads to extensive variability in clinical application. 1. Each observer of child phonology can generate a unique list of processes, making it as exhaustive or simple as desired. 2. An exhaustive list (such as Hodson and Paden, 1983) often has the one or more processes over the domain of a single segment, which results in a list equivalent to a substitution analysis. 3. Very often the list is conceived without notion of rank or hierarchy, making goal prediction ad hoc. (Some authors have grouped processes. For example, Nettelbladt [1983] creates a severity ranking by comparing the prevalence of syllable structure processes with less severe feature-changing processes. Hodson and Paden [1983] rank their 48 processes from Level 0 to 3 (profound to less severe), placing some syllable structure processes in their severe levels.) 16 Nonlinear Frameworks and Developmental/Clinical Phonology The introduction of groupings and rankings of processes in the disorder literature demonstrates a pretheoretical awareness of hierarchy and levels. In order to develop further our understanding of constraints and groupings or levels of organization within children's systems, nonlinear frameworks which by representation outline levels, hierarchies, and dependencies would seem beneficial. A few developmental phonologists have applied some of the concepts of nonlinear phonology to analyses of children's phonology. Menn introduced the discussion to child phonology (1978,1983) in terms of templates. More recently, Spencer (1986), Sadanandan (1987), Ingram (1988), Stemberger (1988), and Stemberger and Steel-Gammon (in press) have presented arguments about phonological development in terms of underspecification, feature hierachy, and tier structure. Application in the clinical field is more limited. It includes Spencer's (1984) analysis of one child's disordered system in terms of McCarthy's (1981) templatic morphology, Chiat's (1989) analysis of the interaction of one child's syllable structure and fricative stopping utilizing Selkirk's (1980) syllable theory, and Dinnsen's (1989) accounting for order of segmental acquisition from predictions of universal feature hierarchies. Before the research for this dissertation, there was no reported application of nonlinear phonological theory to intervention. The basic concepts presented above — autonomous and hierarchical levels, association principles, and syllable/word theories ~ lead, however, to the following general questions pertinent to phonological intervention. 1. Will nonlinear phonological frameworks help to predict logical and attainable intervention goals for phonologically disordered children? 2. Are the separate prosodic and segmental levels of representation of nonlinear phonology psychologically real? If the prosodic and segmental levels of representation are psychologically real, then it should be possible to plan phonological intervention which focuses separately on the two levels. Furthermore, if the dominance of the prosodic tier over the segmental is developmentally relevant, one might expect to find a faster rate of change for prosodic intervention than for segmental intervention. No previous studies have directly compared phonological intervention at the prosodic versus segmental levels. 3. If the 'prosodic tier' has some observable clinical reality, will there be a difference in proportion and rate of syllable I word shapes acquired as a result of interventions that contrast the onset and rime versus those that utilize the mora as a constituent? Intervention which reflects the most psychologically real representation might be more facilitative, resulting in a faster rate of acquisition of syllable/word types. 4. If the 'segmental!melodic tier' has some observable independence, is there any advantage to be gained from targeting specified features at 'higher' versus 'lower' levels in the feature hierarchy in phonemic inventory intervention? Allowing for the fact that phonologists do not agree on all details of the hierarchy, such as the total range of features, the locus of individual features, and their markedness status, one might predict that higher and, therefore, dominant features/nodes should have developmental precedence over subordinate features/nodes. Complication of this question arises from the concept of markedness, in that early inventories should reflect unmarked default features and their values according to predictions of UG (see appendix 2): 1. Root node: [-consonantal] for vowels, [-sonorant] for consonants (Since [consonantal] and [sonorant] define the segment, early specification of [+consonantal] and [+sonorant] are necessary, or there would be no segments.) 2. Laryngeal node: [-voice], [-spread glottis] for consonants; [+voice] for vowels 3. C-Place node: Coronal - [+anterior], [-distributed] 4. V-Place node: Labial [-round]; Dorsal [+back], [-low]. Individual variation in development may result in somewhat different segmental inventories at any given point of comparison between children, further complicating the issue. Different inventories may result from feature hierarchies with different sets of specified features at the point of comparison. However, the question remains, given markedness relationships and individual variation, whether a general developmental sequence is predictable from feature level, the higher (dominating) features being specified first. CHAPTER 2 METHOD To address the research questions posed at the end of chapter 1, a longitudinal intervention study was carried out with six children between the ages of 3 and 6 years. All of the children had moderately severe or severe phonological disorders. Each child participated in individual speech therapy sessions three times per week for five months. The five-month period was divided into three six-week blocks, each containing sixteen treatment sessions and three probe sessions. The first two blocks were the experimental blocks with strict condition orders. In the third block, the needs of the child at that time determined selection and order of conditions. Subjects Criteria for subject selection were: 1. Age: Between 3 and 7 years 2. Severity: A moderately severe to severe phonological disorder, characterized by Umited phonemic and syllable/word shape inventories. (In practical terms, family members had difficulty understanding the child's speech much of the time.) 3. Language environment: English as the only input language 4. Absence of any other major impairment except a language production disorder, a mild impairment of language comprehension, cognition, or motor development, or controlled otitis media 5. Parent participation: Attendance at all treatment sessions and carrying out of home activities for the duration of the project 19 Subject Recruitment and Participation E recruited subjects by letter through health units, agencies, preschools and school boards. Agency speech-language pathologists performed an initial screening after obtaining parental consent, and E did the final screening, selection and assessment. Because of limited subject availability, the selection process took four months. This resulted in different starting dates for the subjects. Two new subjects joined the project in each of three subsequent months, E assigning subject numbers by selection order. E observed SJj's phonological development for four months prior to enrolling her as a subject, because she was age 2;6 at referral (outside the 3;0 minimum age suggested). When it became apparent that her developmental rate was notably protracted, she was selected to participate in the study. No subjects withdrew, attendance was regular, and parents participated enthusiastically in the program throughout the project period. Subject Characteristics and Matching Main characteristics of the subjects are reported in table 2.1.4 All but S6 were male. As table 2.1 shows, it was possible to match two pairs of subjects for both age and gender — S4 and S3 at about 3 1/2 years — and SI and S5 at 6 years. Serendipitously, the phonological systems of the children within each of these two pairs were also similar. Although S_2 and S6_ differed in age and gender, they were matches because of similarity in types of substitutions at the segmental level. (S6's phonological system matched that of S3 and S4 in terms of hmitations at the syllabic level.) The matching facilitated counterbalancing of conditions across subjects (see below). One of the children, S_5, had a developmental delay in all areas, with the most severe impairments in language production and fine motor skills. Three of the children had general language production disorders, one had controlled otitis media, and one a notable tongue thrust during speech and swallowing. Table 2.1.-- Main characteristics of subjects during study Subject Age Severity of disorder Associated factors 1 5;10-6;2 Moderate-severe Tongue thrust; finger-sucking habit Sibling with mild phonological disorder Diverse dialectal influences 2 4;2-4;6 Moderate-severe Chronic otitis media, myringotomy Mild morphophonological delay 3 3;4-3;8 Severe Severe language production disorder 4 3;5-3;10 Severe Severe language production disorder 5 6;2 - 6;6 Moderate-severe Severe language disorder Mild cognitive delay Mild motor impairment 6 2;11 -3;3 Severe No apparent associated factors Design Both a multiple baseline and an alternating treatment design were employed. Through the application of multiple baseline and alternating treatment design methods, some inference can be made that environmentally manipulated variables have impacted on development (Hegde 1985, 30). The multiple baseline design allowed E to compare treated and untreated parts of each child's linguistic system in order to evaluate possible effects of intervention. This provided a means for addressing the first of the research questions on general efficacy of the nonlinear frameworks for intervention. Untreated phonological targets served as baselines, as did other aspects of the children's language development. Based on generalization studies (Elbert and Gierut 1986), E predicted: 1. That those phonological elements most similar in feature components (with reference to segmental intervention) or prosodic ('skeletal') slots (with reference to prosodic intervention) to actual therapy targets would show more improvement through generalization than those phonological targets with fewer commonalities. 2. That vocabulary and morphology measures might increase at an accelerated rate. Since phonological intervention often involves the presentation of words and inflectional morphemes not in the child's production system, one might anticipate accelerated acquisition rates for these measures (assuming that input frequency and intensity affect their rate of development). 3. That the area with least potential for rapid development was syntax. Although some of the therapy involved phrase and sentence practice, syntactic goals were only treated as subsidiary goals for S3, S4 and S5, the three children with language disorders. The alternating treatment design allowed E to contrast condition types relating to the last three research questions set out at the end of chapter 1: (1) intervention targeted at the segmental versus prosodic (designated frequently throughout the text as syllabic5) tier (main conditions), (2) prosodic stimuli presentation and production manipulations as moraic versus onset-rime constituents (syllabic subblock conditions), and, (3) training of higher level versus lower level features in the feature hierarchy (segmental subblock conditions). In this study, alternations took place as follows: (1) every nine sessions (i.e., approximately every three weeks) for the syllabic and segmental training subblocks, and (2) every four sessions (i.e., approximately every ten days) for the within-subblock conditions (syllabic theory contrast and feature hierarchy contrast). This design is schematized in figure 2.1. 18 sess. (6 wk.) each > B L O C K 1 B L O C K 2 Main conditions* 9 sess. each > S Y L S E G S Y L S E G Within-subblock cond. 4 sess. each > M O R OR H L F L L P MOR OR H L F LLF * S Y L = syllabic condition; SEG = segmental condition; 1 session = minor probe * * M O R = moraic; OR = onset-rime; HLF = higher level feature; L L F = lower level feature Figure 2.1. Schematization of research design These time periods were considered long enough to provide an opportunity for consolidation of each condition's targets, while short enough to provide alternation within subblocks. The design allowed for a minimum of two alternations across the time blocks, a minimum recommended for such designs (Barlow, Hayes and Nelson 1984,229). To control for order and sequence effects, conditions were counterbalanced across subjects (within matching pairs), rather than within subjects (see table 2.2). Table 2.2.—Order of treatment conditions for each subject within Blocks 1 and 2 Block No. S i S2 S3 S4 S5 S6 # of Sess. 1 8 SeLSeH3 SeL:SeH SyMrSyO SyO:SyM SyO:SyM SeH:SeL 8 SyM:SyOb s y ° : S y M SeH:SeL SeL:SeH SeH:SeL SyM:SyO 2 8 Same as Same as Same as Same as Same as Same as 8 Block 1 Block 1 Block 1 Block 1 Block 1 Block 1 aSeL = segmental condition — lower level feature in feature hierarchy (4 sessions) SeH = segmental condition - higher level feature in feature hierarchy (4 sessions) bSyM = syllabic condition — moraic theory (4 sessions) SyO = syllabic condition — onset-rime theory (4 sessions) As table 2.2 shows, a basic counterbalancing of treatment orders was achieved across the first four subjects. Main condition order assignment for the first two subjects was segmental/syllabic with opposing subblock orders — higher level feature/lower level feature versus lower level feature/higher level feature, and moraic/onset-rime versus onset-rime/moraic respectively. Main condition order assignment for S3 and S4, who commenced training in the month after S i and S2, was syllabic/segmental, with opposing subblock orders — moraic/onset-rime versus onset-rime/moraic, and higher level feature/lower level feature versus lower level feature/higher level feature respectively. S5 and S6 were assigned to opposing main treatment (syllabic/segmental) conditions and within-subblock conditions (onset-rime/moraic versus moraic/onset-rime; lower level feature/higher level feature versus the opposite). The above condition assignments ensured three subjects per condition order set (i.e., three with the order syllabic/segmental, three with the order moraic/onset-rime, three with the order higher level feature/lower level feature, and vice versa). E set out the first four condition orders ahead of time (see S i to £4 above), and then assigned subjects to condition according to the point in time at which they were accepted for the project. Main and subblock condition orders for the fifth subject were selected to contrast with those of SI, because the screening showed that S3, matched S i in terms of age, gender, and phonological system. This allowed complete counterbalancing of treatments across similar phonologies for two of the six subjects. Subblock condition counterbalancing for subjects with similar phonological systems was achieved for: (1) S 2 versus S4 (matches by age, gender, and phonological systems), and (2) S2 versus S_6_ (matches by substitution type at the segmental level). Main condition counterbalancing across phonologies similar in terms of syllabic structure was achieved for S3 and S4 versus S6. Assessment and Analysis Procedures During the Intervention Period Initial Assessment An initial assessment consisted of an audiorecorded phonological/language sample and a battery of standardized tests of language comprehension and production. (These same tests were administered at the end of Block 3.) Stimulability of phone production in isolation and nonsense syllables was probed briefly during the assessment. For the older subjects — S2, S5_, and S i -phonemic perception was tested with the Auditory Discrimination Subtest of the Test of Language Development (Primary, 1982), and, for Si, also with the Goldman-Fristoe-Woodcock Auditory Discrimination Subtest (1974). A hearing screening, an oral mechanism examination including items from Robbins and Klee (1987), and a case history filled out by the parent completed the assessment procedures. Phonological Sampling A standard set of objects and pictures provided the stimuli for a (predominantly) spontaneous single-word elicitation, which was recorded using a Nagra IV reel-to-reel tape 25 recorder and Ampex 631 tapes with an AGK D202 microphone. (See appendix 1 for the list of 164 basic words.) General conversation during the recording session provided other words (200-400 per sample) for analysis, many in connected speech contexts. All English consonants in word-initial position (WI), word-final position (WF), syllable-initial-within-word position (SIWW) and, where possible, syllable-final-within-word position (SFWW) were probed.6 In order to check for morphophonemic effects on WF consonants (becoming SIWW after derivation), derived forms were also elicited, through adjunction of [i] to nouns and [irj] to verbs, e.g., sun, sunny: rub, rubbing. (Although E attempted to elicit the derived forms after elicitation of the root word, this was not always possible in the assessment context.) In order to control for possible assimilatory effects, consonants were elicited in a variety of word shapes which included CV, VC, and QVCi words (when such words existed) as well as more complex word shapes. Vowels and diphthongs were not specifically probed, but were in fact available for analysis from the word list as well as from the conversational sample. Phonetic Transcription E transcribed the assessment and major probe tapes using a Revox taperecorder and Videoconcepts F700 dynamic earphones. The IPA (1979) symbols and diacritics supplemented by a few extra codings served for the narrow transcriptions performed. (See appendix 5 for details about transcription.) Codings relating to imitated versus spontaneous speech and single words versus connected speech were also included. E randomly selected taped, transcribed samples from half of the subject group (SI. S5. and S4} as bases for determination of intertranscriber reliability. A second individual, who was very experienced in phonetic transcription of children's speech, transcribed the taped samples independently. When transcriptions did not match, the two listened jointly to the tape in order to resolve disputed forms. Each phonetic slot with its onset or offset characteristics counted as one item. For example, E may have transcribed a child's production of /tol/ as [thol]. The second transcriber may have transcribed that same item as [to3]. In this case, the consonant difference counted as one mismatch and the diphthong difference a second, yielding no matches for the word (0/2). The Si sample contained 70 words from the Block 1 and Block 2 major probe tapes (35 each), yielding 138 consonants for comparison. (See appendix 3 for data.) For Si, the two transcribers were in agreement about 75% of the consonants prior to discussion, and 97% after discussion. Disagreements for Si's data primarily concerned degree of dentalization of the coronal obstruents and degree of voicing ~ i.e., detail not relevant to the intervention goals for S_L Because discussion led to a similar agreement level for the samples from the two tapes (96% and 97%), transcribers concluded that one tape/child was sufficient for subsequent reliability checks. For £4, transcriptions of 35 words from the Block 1 tape were compared, yielding 64 consonants and three diphthongs for comparison. When presence or absence of (nonphonemic) glottal stops was considered, transcription agreement was 69% before discussion, and 93% after discussion. Eliminating concern for glottals, there was 88% agreement before discussion, and 99% agreement following discussion. The other main initial disagreement concerned degree of voicing. Since S4's frustration level during testing resulted in frequent use of exaggerated glottal fry (aperiodic voicing) in early probe tapes, the disagreement on laryngeal features is understandable. Neither of these laryngeal characteristics was relevant for £4's intervention plans. For S5_a transcriptions of 35 words from the Block 1 tape were compared, yielding 80 consonants for comparison. There was 86% agreement before discussion, and 96% after discussion for consonant transcription. The main initial disagreements concerned transitional phenomena, which were not relevant for S5's intervention plan. In areas of disagreement, the transcribers found that they basically perceived the same phenomena, but used different descriptors before arriving at a solution through discussion. None of the critical elements relating to the intervention for each child were in disagreement, since those concerned broader phonemic categories rather than details of narrow transcription. Because the results - using stringent reliability criteria — were comparable across the first three data sets, and because mismatches were not critical to the intervention programs, no further phonetic transcription reliability checks were made. Analysis during the Study Three evaluative probes were administered per block, at sessions 9,18, and 19. The 'minor' probes at sessions 9 and 18 tested untrained words containing targets of the previous three-week condition in an elicitation task, a conversational speech sample, and a story retell (from Clinical Probes of Articulation Consistency [1981] or E-generated). They were recorded on Fuji FR-IJ cassettes using a Marantz PMD220 tape recorder and AKG D310 microphone. A week's vacation was scheduled between the 18th and 19th sessions to inhibit carryover from the condition probe to the major block probe. The major block probe consisted of the standard single-word elicitation task from the assessment plus any conversational speech elicited, recorded on the Nagra reel-to-reel equipment. During the intervention period, E recorded transcribed utterances on a subset of forms from Grunwell (1985), namely the phoneme and cluster realization charts and the phonotactics forms. These served as a basis from which to draw feature diagrams and deduce syllable/word constructs . Intervention The analyses performed led to formulation of detailed intervention plans which conformed to the experimental framework for each subject. Segmental Condition Target Selection Unestablished or very marginally established segments (less than 10% matches) were possible training targets. Segments related and unrelated to the training targets in terms of features remained as observation targets. Reduced stimulability did not necessarily eliminate segments as targets, especially if the experimental design or the child's system limited available choices. Using the feature hierarchy in figure 1.2 as a basis, E drew feature hierarchy figures for each child. (See individual chapters for details of target selection for each subject.) E then selected unestablished features at higher and lower positions in the hierarchy as intervention targets. E assumed the following with respect to classification of feature height: 1. Terminal features are lower than intermediate nodes or other nonterminal nodes, i.e., the terminal feature [round] is lower than its class node ~ Labial Place, [continuant] is lower than Root node. 2. Terminal features of a node are equivalent in height, i.e., Coronal Place feature [+distributed] is at the same level as Coronal Place feature [-anterior]. 3. Root node features are at the highest level, with [consonantal] and [sonorant] higher than [nasal] and [continuant]. (As stated in chapter 1,1 have followed McCarthy [1988] in assuming that [consonantal] and [sonorant] define Root node, as opposed to being terminal features.) 3. Laryngeal node features are lower than Root node features but higher than Place node features. Although the theory does not assume that Place node is lower than Laryngeal node, E assumed that glottal functions take precedence over shaping of the oral cavity. Production of phonemes such as /hj and /?/ is independent of any oral cavity shaping, whereas production of any phoneme with oral cavity shaping also requires laryngeal function (marked or unmarked). 4. Place node is higher than Labial Place and Dorsal Place nodes, which are at equivalent levels. 5. As the default Place node, Coronal Place node is higher than Labial Place and Dorsal Place nodes, since it is assumed at the level of Place node. In order to select segments as training targets and observation targets, E then consulted feature specifications for English, relating features to the segments for which those features are distinctive (appendix 2 in this thesis). Since establishment of features was the goal, often more than one segment became a training target during a feature subblock. For example, in the case of SI. the feature [-anterior] was a lower level terminal feature target. The segments /J7 and AV were both selected for establishment, with /if and/tf/ remaining the baseline observation targets. This provided a maximal contrast of a 'complex' versus simple (nonbranching) segment, as well as a voiced versus unvoiced [-anterior] segment. The training of a voiceless fricative and a voiced affricate in the same time frame provided a means of evaluating the effects of training on development of voicing and 'complex' structure, although the focus in therapy sessions was on 29 [-anterior] tongue placement, and not on the complex/noncomplex distinction or the voicing distinction. The syllabic data dictated choice of word position for target training. Only well-established word positions served as slots into which to insert the new segments, in order to maintain as much independence as possible between the segmental and syllabic conditions. Syllabic Condition Target Selection Syllabic targets included two types: (1) unestablished or marginally established syllable/word shapes, e.g., CVC, CCVC, CVCVC, and (2) extension of segment or sound class word position realization. In the latter case, E assumed that dominance of the prosodic tier over the segmental tier implies that syllable/word position determines licensing of segments, rather than the reverse. Marginally established syllable/word positions remain underspecified in comparison to other syllable/word positions for the features of a language. For example, at one point in development, coronal fricatives may be licensed for realization in WF position, but not in WI position or clusters. (Autonomy of syllabic constituents is evident in the fact that some may be specified for a given set of features whereas others are not.) Development did not entail learning the feature composition of coronal fricatives, but the strengthening of the system's syllable positions by the realization of more segments within those positions. The segmental data dictated choice of segments to insert into the new syllable slots. Only segments well-established in some word position became targets, in order to maintain as much independence as possible between the segmental and syllabic conditions. As was the case for the segmental condition, reduced stimulability did not necessarily eliminate possible targets, and some syllable/word shapes and segments remained as untrained observation targets. General Outline for Training Sessions Training sessions were audiotaped on the Marantz PMD220 using the AKG D310 microphone and Fuji FR-II tapes. All training words differed from probe words and the former did not contain any unestablished segments or word shapes except the target for that subblock condition. Sessions in all conditions followed this general outline: 1. Awareness training for the training target: mildly amplified stories, songs, poems, etc., which focused on the target (no vocal response required) 2. Auditory perceptual contrast training of the target and its error pattern with minimal or near minimal pairs or trios (only at the beginning of a subblock) 3. Imitation training with auditory, visual and physical cueing from the target in isolation, through permissible syllable/word shapes to sentence patterns, then spontaneous sentences and finally conversation Minimal pair contrasts were used frequently throughout production training, although, for syllabic targets, E ensured that only those constituents particular to the onset-rime or moraic condition were made salient in contrast training (see section below on intervention specific to the syllabic condition). 4. Practice in monitoring of others and self at the sentence level and above 5. Tangible and/or social reinforcement for correct productions at all levels, with special prizes being earned for use of set numbers of targets in conversation 6. Parent participation within the sessions, followed by a brief counselling session concerning required home practice. Parents were asked to note any changes in spontaneous speech outside the therapy situation, and any information pertinent to subject history during the intervention period which could threaten internal validity of the project (Ventry and Schiavetti 1980,104). Treatment Methodology Specific to the Syllabic Condition Counting Conventions for Moraic Constituency A. Monosyllabic words E assumed that, in English, a mora may be realized by either a vowel or the first tautosyllabic consonant following the vowel. Following this assumption, the following were considered bimoraic: 1. CVC monosyllabic words In English monosyllabic words, lax vowels are followed obligatorily by consonants, suggesting that the consonant has prosodic relevance, i.e., syllabic weight. It is not 31 obligatory, however, that a consonant follow a tense vowel (/i/ and /u/ only in this study). This suggests that consonants may not have syllabic weight after tense vowels, as the vowels are bimoraic. While the difference between tense and lax vowels is acknowledged in priniciple, it was not considered an independent factor in a certain type of intervention session focusing on CVC development. In that type of session, the method of presentation involved addition of C 2 to a C[V:::] unit (£3_ and S6J, in which both lax and tense vowel stimuli were presented with exaggerated length, i.e., with an indeterminate number of morae. The targeted final consonant was treated as an adjoined unit in those intervention sessions. With increasingly rapid articulation, syllable timing was reduced to that of the adult model - at which point the moraic status would presumably be the same as that of the adult model (bimoraic with both lax and tense vowels, the WF consonant being adjoined in cases of syllables with tense vowels). 2. Syllables with diphthongs, i.e., two identifiable (usually lax) vowel units (in this study /au/, /al/, hi/, /ou/, /8I/) 3. CVCC or CVVC monosyllabic words (W=diphthong). B. Disyllabic words 1. In general the possibly bimoraic status of the unstressed N suffix in English was ignored when relating moraic status of CVCV and CVC words for S3, S4, and S6 (see Block 1 syllabic intervention plans for each subject), because (1) the focus for moraic status was on the first (stressed) syllable, and because (2) the children had no difficulty pronouncing [i(:)] in open syllables. 2. For disyllabic words, other complexities arose. For example, according to the counting methodology, the first syllable of the underlying form of a derived diminutive word such as doggie /dag+i/ would appear to have two morae (/dag/ — V=l, C2=l), as would the first syllable of the underlying form of tractor (/track/ -- V=l, C3=l). The bimoraic status of the /dag-/ portion of doggie changes, however, when it occurs in the derived diminutive ([da'gi1] — Vi=l), whereas it does not change for tractor ([trask'ta-] ~ Vi=l, C3=l). Onset creation (premoraic consonant adjunction) for the second syllable of doggie {l-M becomes [-.gi]), leaves only one constituent in the first syllable - the vowel - as a candidate for moraic assignment. The first syllable of tractor, on the other hand, remains closed at the surface level ([trask'ta*-]) and, thus, the first syllable has bimoraic status. Stimuli Presentation At the outset of imitation training during syllabic subblocks, E presented segments, syllables and/or words through speech (that is, auditory cues) with simultaneous visual and physical manipulation cues. These cues were provided in such a way as to emphasize phonological constituents according to the theory in question (see figure 2.2). In the onset-rime condition, the maximal onset and maximal rime were presented as discrete stimuli. For example, a CVC word was presented as C and VC ([th]-PAUSE-[ap]); a CCV word was presented as CC and V ([st]-PAUSE-[£i]) In the moraic condition, the immediately prevocalic consonant was considered "attached" to the following vowel making the basic constituent (core syllable) CV. Consonants were added to this (inseparable) CV unit. For example, to establish the CVC word shape when final consonants were missing in the child's system, separate CV and C stimuli were presented. (See note above on monosyllabic words for one type of methodology. See the Block 1 intervention section for S4 for another.) To establish WI clusters, a CCV word was presented as a C and a CV, i.e., [s]-PAUSE-[tei]. While it could be argued that either presentation method could be construed as the other method — the O-R '(C)C - V presentation could be construed as moraic as well because the moraic unit was presented separately from the prevocalic adjoined consonant(s) ~ it was assumed that constituent status was the important variable. Thus, 'CC, if presented separately, was then a unit, or constituent (a domain with internal structure and its own set of operations for phonological rules). Because onset clusters are not constituents in moraic theory, E did not present them as such in the moraic condition. One could construe the CV moraic constituent as the basic O-R unit. However, presentation of a CV unit does not focus on the onset and rime as separate and independent constituents, which was desirable in the onset-rime condition. Onset-rime stimuli 0 \ R O j R 1 : * j f \ / \ = $ C : V C C C : V C + VC stimuli CC + V stimuli for CVC word shape training for WI cluster training Moraic stimuli A o: : o ' E \ ! I : M : , M = I \ / A - C G C V CV + C stimuli C + CV stimuli for CVC word shape training for WI cluster training Key: O: Onset N: Nucleus R: Rime O : Syllable M: Mora *Dotted lines delineate divisions for stimuli presentation. Constituents on one side of the dotted line were presented separately from constituents on the other side of it. Figure 2.2. Stimuli constituents for onset-rime and moraic subblock condition training 33 It could be argued that phonetic considerations favored the moraic condition from the outset, since the phonetically motivated glottal onset of the VC must be suppressed in the onset-rime condition; that is, C + [?]VC impedes pronunciation of CVC. When E preceived this to be a problem for a child, a rapidly alternating series of ([?])VC, C, VC, C, VC was presented for elicitation. Rapid speech elicitation facilitated suppression of the [?] onset. A possible advantage for onset-rime condition training, on the other hand, may have been contrast training for onsets in auditory stimulation and production training, e.g. [st] versus [t] versus [s] (heightening awareness of the difference between the child's production and the adult target) as well as rhyming activities heightening awareness of the rime. Other than these biases, there was an additional problem for /s/ clusters. In both theories, the /s/ may be considered extrametrical when preceding a stop word initially (Kiparsky 1979). This puts the Is/ outside of the "true" onset for /sp, st, sk /. It had to be treated as an onset constituent in this study to maintain consistency with the basic decision to contrast rime and onset. An uncontrollable biasing factor could have been the choice of lexical items and probe items for each condition. The "magic" of changing damp into stamp (and thus remembering that type of word) may be greater for a particular child than changing ick into stick, or vice versa. This kind of experimental bias could neither be predicted nor controlled, but it was assumed to be equivalent for conditions. Once a child demonstrated phonetic control of the syllable/word structure target, procedures for establishing phrases and sentences were the same as for segmental targets. Analysis Following the Intervention Period In order to analyze the results of the various probes more precisely, E performed computer assisted analyses following the study, using spreadsheets Excel and WINGZ on the Macintosh Plus computer. The majority of the analyses reported in the results chapters are based on the initial assessment and the major probes at the ends of Blocks 1 and 2, the two experimental blocks. For some of the children, analyses based on the minor probes and session data complement the major probe results. 35 Feature Hierarchy Analysis E recorded matches and discrepancies between the child form and the adult model for each phoneme per word position (consonants for all children, and consonants and diphthongs for S4's Block 1 and 2 probes). For each child, proportional matches and gains were calculated for each phoneme in the major probe, with descriptions made of feature differences and changes based on the feature hierarchy. This approach is relational, in that features or syllable shapes are noted as marked or unmarked with respect to the adult model presented in appendix 2. 'Omissions' of cluster and diphthong elements were entered as missing Root nodes in the feature hierarchy data, as were 'omissions' of singletons. The example in table 2.3 demonstrates the feature descriptions for the word face > [haet]. (Because there are a large number of spreadsheets for data analysis they are not included in the dissertation.) An anomaly arises in the description of phonological development when a child's system appears to have a specified value for the wrong segments. For example, [+distributed] is labelled a 'specified' feature in English for /r/ and interdentals in this report, but for many children, it appears to be a feature of coronal obstruents when produced with an interdental tongue position ("frontal lisp"). In this case the 'distributed' aspect of the coronals as produced incorrectly has the quality of an underspecified 'default', taking the place of the [(+)anterior] coronal default for English. An articulatory basis for the "incorrect" default can be observed and serve as an explanation for the phonological anomaly - a consistent interdental tongue position. Sometimes the unexpected default appears to have no observable articulatory or perceptual basis. In such cases, the anomaly can only be described as such. Table 2.3.—Sample feature chart showing match and nonmatch data and feature differences Adult Target % Match Match Tokens Non-match Tokens Phonetic Realization or Omission Feature Differences: Types Feature Differences: Tokens f 0% 0 1 h Place node unspecified3 1 ei 0% 0 1 as SYLb: VV> V SEGb: Dors: [-hi, -lo] unspecified 1 s 0% 0 1 t Root node: [+cont] unspecified 1 aThe Root node features [+consonantal] and [+continuant] were assumed to be specified. However, no articulator node was specified, hence the 'Place node unspecified' designation. The [h] was a default [+continuant] consonant. Although /h/ has no underlying [continuant] specification (McCarthy 1988, 92), the Laryngeal node [+spread glottis] specification of /h/ was assumed to be equivalent in function to the [continuant] specification of the Root node. (See section on features in chapter 9, however.) bBoth syllabic and segmental differences between /ei/ and [ae] are addressed within the feature chart. Syllabic Data General Syllabic Data For each child, E calculated proportional matches and gains for general syllable/word shapes for each major probe, and for individual targets relating to that syllable/word target. The latter were available in the feature charts (see above). The former were derived from confusion matrices which distinguished between adult target syllable/word shapes and child forms. For S3 and S4, these were similar to Grunwell's (1985) Phonotactic Possibilities: Contrastive Analysis. Adult target syllable/word shapes were listed on the vertical axis and child syllable/word shapes on the horizontal axis. Whereas Grunwell used individual matrices for mono-, di- and polysyllables, E included all in the same matrix in the current analysis. This was done because very often an adult polysyllabic target (CVCVC) was realized as a monosyllable by the child (CV), and, less often, an adult target monosyllable (CV) as a child reduplicated or diminutized disyllable, e.g., doggie for dog (CVCV). For SI and S_5, only the WI clusters were under observation, and 37 hence the confusion matrices only included words with initial clusters. For S2 and £6_, E was able to derive syllabic data directly from the feature charts. Moraic versus Onset-Rime Condition Results Testing for differences between the moraic and onset-rime conditions depended on whether the same or different segments were involved in the opposing conditions. For example, in the case of Si, /kw/ was a target in the moraic condition, and /tw/ a target in the onset-rime condition. Major probe data was one source for comparison, but because of the limited number of tokens in the major probe, E also used minor probe data. When the comparison between onset-rime and moraic conditions involved the same cluster or segments, as in Block 2 for Si (i.e., /st/), E analyzed therapy session data. Interobserver reliability was checked for one session of Si's sessions. Using an audiotape tape of the session, E and a second observer independently scored Si's spontaneous and echoic responses as correct or incorrect. Prior to discussion, agreement between E and the second observer was 87% for scoring.of correct and incorrect imitative responses and 88% for scoring of spontaneous responses. After discussion, there was 100% agreement for imitative responses, and 99% agreement for spontaneous responses. Initial disagreements were due primarily to differential assignment of responses to echoic versus spontaneous categories. Because agreement was high, no further reliability checks were made of session data. Results Reporting The next six chapters contain individual case studies for the six children. Each chapter includes: 1. A summary of pertinent biographical data related to the disorder 2. A summary of language test results 3. A detailed analysis of the child's phonological system at Initial Assessment 4. The intervention plans for Block 1 and Block 2, with rationales for target selection 5. The results of Block 1 and Block 2 intervention CHAPTER 3 SUBJECT 1: SUMMARY [ kan a i ? av m a i vwe i : : k neu] Case History Information Subject Characteristics During the reported period of the study, this boy was 5; 10 - 6;4. The phonological disorder was associated with the following (see table 3.1): 1. A tongue thrust during speech and swallowing 2. Finger-sucking 3. Mild difficulty with phonemic discrimination 4. Dialectal variation in input by changing residency ~ Vancouver (3 years), Australia (2 years), Vancouver (1 year) ~ and parent's dialect — Mother English, father Australian 5. Middle child of three boys, with the younger sibling having a mild articulation disorder Prior Treatment SI was initially assessed at the age of 2;9 for unintelligible speech. At that time his phonetic inventory consisted of voiced and glottal stops, glides /j/ and/w/, and fricatives /f/ (as a substitution for final voiceless sibilants) and /0/ (as a substitution for final Iff). Syllable/word shapes were reduced, with many missing final and medial consonants. He received intermittent speech therapy in Vancouver and Australia with focus on several final consonants, as well as /f/ and l\J in all positions. At the age of 5;2 (the most recent assessment prior to the project), the following were missing: liquids, alveopalatals, affricates, /hi, WI voiceless stops and some strident fricatives. Although there were no word reductions, cluster reduction was frequent. From 5;6 to 5;9 he received infrequent articulation therapy focusing on high frequency words, /5/, and an introduction to HI in WI position. 38 Table 3.1.—Subject information for SI AGE 5; 10 - 6;2 for reported period GENDER Male BIRTH ORDER Second of 3 boys; third has mild phonological disorder DIALECT Mother -- English; Father — Australian; Residence — Vancouver, Canada (previously Vancouver and Australia) HEARING Screened within normal limits at beginning of project PHONEMIC Test of Language Development (Primary) (1982) Auditory DIS CRIMINATION Discrimination Subtest - borderline average (S.S. of 7) Goldman-Fristoe-Woodcock Audit. Discrim.Test (1974): -1.5 SD. ORAL MECHANISM Tongue thrust and Class 1 malocclusion Finger-sucking habit Some lack of independence for jaw and tongue movements, such that elevation of the tip and body of the tongue could not be performed with an open jaw before training LANGUAGE Age-appropriate SKILLS PREVIOUS Between 2;9 and 5;0, intermittent therapy in Vancouver and Australia for THERAPY final consonants, HI and IxJ with other speech-language pathologists A few sessions in the 6 months prior to the project with a different speech-language pathologist who introduced 16/ and IV No changes noted in the year prior to the study 39 Phonological System at Initial Assessment In figure 3.1 baseline data for syllabic and segmental targets are displayed. Absence of training target data for the segmental condition indicates a lack of matches for those training targets at time of asssessment. ed JCS U E 80.00% 70.00% 20.00% 0.00% HHHHHHH 111 H H H •1 HHHHH H i IH^H H • I • • • H i H H H pi ii j- -"" iB" Hi HHHI •1 m • •1 H i HHI1 •1 l u 1 1 i i i i i i H i 111 • I III 1 E l i • I 11 , 1 1 i n 1 1 i i i • Training syllabic targets | | ' c c ~ | | Observation seg. targets" M SYL: WI CC: 1 syl: trained H SYL: WI CC: 2 syl: trained • SYL: TOT WI CC: EPEN x • SYL: TOT WI CC: EPEN V SYL: SIWW CC - observ. SEG: ASPIRATION SEG: COMPLEX V SEG: OBST - Not complex* Note: There were no matches for syllabic training target WI #CC in 4-syllable words and segmental training targets /l/ and alveopalatals. There were no matches for syllabic observation targets #CCC and segmental observation targets /h/, /r/, and the interdental fricatives. Clusters were judged with and without EPEN, i.e., short epenthesis (reduced vowel). *OBST - Not complex = coronal obstruents (stops and fricatives) with appropriate lack of branching (complex) structure Figure 3.1. SI: Proportional matches for primary training and observation targets at Initial Assessment Phonotactics: Syllables and Words Word structure and 'skeletal slot' realizations matches with adult targets (i.e. not necessarily segmental matches) were found for: 1. Number of syllables in a word 2. Number of final consonants in a word 3. Diphthongs (some nonsignificant dialectal differences) On the other hand, syllable-initial consonant clusters were reduced or simplified in comparison with the adult target (see table 3.2). As the length of the word increased, the frequency of reduction increased. In WI position, the following was observed in terms of cluster matches for 'skeletal slots' (not necessarily segments): 1. Monosyllabic words: 11/26 (42.31%) matches for diconsonantal clusters, with nonmatches being singletons. Neither diphthongs nor postvocalic consonants appeared to influence realization of initial cluster elements, indicating that limitations of output were not a limitation based on allowable skeletal slots per se. 2. Disyllabic words: 8/26 (30.77%) matches for diconsonantal clusters, with nonmatches being singletons. 3. 4-syllable words: No matches (two attempts) for diconsonantal clusters (singleton realizations). 4. Triconsonantal clusters: No matches (four attempts). For the three mono- and disyllabic items, two slots were filled and for the trisyllabic item one slot was filled. Syllable-initial diconsonantal clusters within words matched only once out of seven times: eyebrows - [?al.vwau5]. Not all compounds had clusters — the clusters in airplane and icecubes were realized as singletons. The segmental data further demonstrates the delay in cluster development. With the exception of /gj/ in one of two items (argue), no consonant clusters matched adult targets for both segments. The first segment of the cluster matched the adult target 37.31% of the time and the second element 24.14% of the time. There were further differences in realization of the two consonantal-type elements depending on their segmental composition. Table 3.2.—£1: Syllable-initial clusters at Initial Assessment and at the end of Block 1 Cluster Initial Assessment End of Block l a (/sn/)b SYLP: (CC) (1/3) SEG: /s/> [+distributed\d SYL:/sn/>CC+ SEG: /s/ > [^distributed^ /sp, sm, sk, (sn)/ SYL: C SEG: /s/RN >0 SYL: CC+ SEG: /s/ > [+distributed\ /St/ SYL: C SEG: /s/> [+distributed]; /t/RN> 0 NO CHANGE /CI/: SYL: C SEG:/l/RN>0e SYL: CC+; epen. (1/2) SEG: /s/RN > 0 (1/2); /l/>(j] or[lj] /Cj/ SYL: C SEG: /j/RN>0 (except SIWW [£)]) SYL: CC* except/mju./; SEG: Match except /mju./ > /my/ /C/[w] SYL: CC or (C)f SEG: Ci matches: When Lab or after epen. vowel : Usually Ci Lab[+cont], i.e., If] SYL: CC*,+ SEG: Ci: Matches except for most coronals : C 2 remains [w] /sCC/ SYL:CorCC SEG: Lab[f]or[fw] SYL: C(C)+(Epen.) SEG: Variable aSummary of major and minor probe data — highlights only. Changes show stars(*) for trained targets and + for untrained targets. ^Marginal acquisitions in brackets CSYL = syllabic information; SEG = segmental information ^Italics = inappropriate marking according to appendix 2 Feature Specifications eOne [j]for/l/ f19/32 with CC, otherwise C 42 /sAclusters Variability in skeletal slot realization was noted for /s/-clusters depending on the second element of the cluster. This fact at the outset indicates a bidirectional influence between the segmental and syllabic tiers. Syllabic constraints resulted in missing segments but segmental constraints also appeared to influence which clusters were realized with two elements. 1. Cluster with both elements realized: /sn/ in one item, i.e., snake > [0 n: eik]. In appendix 2,/s/is designated as the maximally underspecified [+continuant] consonant and/n/as the maximally underspecified [+nasal] consonant. Both share the default Place node for English — Coronal (Paradis and Prunet, in press). This sharing of default coronality perhaps facilitated realization of the two elements for reasons of phonological and phonetic simplicity. However, sharing of coronality was not the sole factor enabling [sn] production, since £1 did not produce /st/ with both elements. According to Kiparsky (1979), the /s/ of /sn/ is not extrametrical. Thus both elements could be incorporated into the syllable structure without the need for specification of an extrametrical constituent. (If extrametrical segments are not licensed prosodically — licensing of them being the marked case — they cannot be realized.) A further factor enabling [sn] production was possibly word-specific and pragmatic: because snakes hiss, the [s] is a "natural" onset to the word. 2. Clusters with missing Root nodes: /sp/, /sm/, /sk/, /st/, and two of three possible /sn/ tokens. The second elements in all of these clusters are noncontinuant consonants. At the prosodic level, the universal minimal word has only one prevocalic consonant, and it is therefore predictable that a child will produce clusters with one element in early phases of phonological development. Given a choice of two segments, the more likely one to appear is (1) the better-established segment, and (2) the one which is not extrametrical. The omission of 'extrametricar /s/ (Kiparsky 1979) can account for the realizations of /sp/ as [p] and /sk/ as [k]. However, /st/ did not pattern with the other /s/-stop clusters but was produced as the child's default [+distributed\ coronal continuant [0]. This seems to be the result of simultaneous interaction of segmental and syllabic constraints. At the outset of encoding, only one skeletal slot apparently was licensed for realization from the prosodic tier. However, at the level of segmental encoding, two minimally specified consonantal coronals appeared to be 'competing' for realization. As a singleton, WI A/ was realized 6/16 times with continuancy (once as [0] and the rest of the time as [tG]). Thus there was not a clear contrast between the two coronals. The segment with more specified features - /s/ by virtue of its consistent [+continuant] Root node specification — was realized. The realization of of /sm/ and /sn/ (two out of three tokens) was similar to that of /sp/ and /sk/ clusters, in that the /s/ was missing. The extrametricality hypothesis would not predict this outcome, because the /s/, as the less sonorous of the two syllabifiable elements, would then be realized. However, nasals were better-established as a sound class. Segmental factors probably played a facilitating role in this case. 3. /s/-clusters with [w] as a second element (from /w/ or/1/) were realized with two elements as [fw]. In this way, they were equivalent to stop clusters with [w] (see the section on stop-initial clusters below). This divergent /s/-cluster fact once again reaffirms that syllable structure realization depends partially on segmental composition. It also supports the extrametricality hypothesis that /s/, when followed by a (sonorous) approximant, can have a surface realization, because specification of an extrametrical constituent is not required 4. Triconsonantal clusters with /s/ (/skw/, /str/, /spr/, /skr/): The [fw] realization of elements in monosyllables is ambiguous for such clusters. The [f] may have been a realization of /s/, the stop being omitted. However, it is more likely that the (extrametrical) /s/ has not been licensed for realization. In the second major probe the stops do in fact surface but the /s/ does not, suggesting that the latter interpretation is more justifiable. The singleton [f] substitution in screwdriver > [fuvwalvo9] may be a syllable-level assimilation from other labiodentals in the word. Stop-Initial Clusters Stop-glide clusters (phonemic or phonetic glides) behaved similarly depending on the glide involved. 1. Clusters with [w] as the second element: These were realized most of the time with both consonant slots filled. When both elements were present, the first matched the target some of the time if the target was labial (e.g., bread > [bwAd9], a connected speech utterance). However, in single word utterances, the use of vowel epenthesis also facilitated the appearance of the first element as itself (e.g., grey > [gaweiJ3]). When there was no epenthesis (and always in connected speech), spreading of specified (Labial Place) and unspecified ([+continuant]) features of [w] resulted in [fw] substitutions for clusters with nonlabial and even one labial initial consonant (e.g., truck > [fwAk]; brush > [VWA6]). In a few cases, only the first element was produced (as a segmental match), e.g., for WI /br/, /0r/ and /kw/ in disyllabic words and one of two monosyllabic words starting with /fr/ (i.e., front-). This was perhaps a developmentally earlier type of cluster realization and is consistent with predictions of the sonority hierarchy hypothesis that the least sonorous element will be realized. However, in the case of the disyllablic items twenty and quiet, when [fw] did not surface, only the [w] appeared. This appears to violate the extrametricality hypothesis, but is consistent with /s/-nasal cluster realization. In these cases, the /w/ is phonemic, and may have been a better-established segment than the stops (see note on A/ above under the section on /st/-clusters), and thus a more likely candidate for realization. (This also shows that the [f] must have arisen late in the derivation, following the encoding of [w].) 2. By contrast with the [w] clusters, neither /j/ 7 nor [j] from A7 appeared for /CjAclusters, except once in SIWW position after/g/ (argue >[augju]). One anomaly was noted for /si/, which patterned with the [w] clusters, since /l/ > [w] in this context (e.g., sleep > [fwip]). As there were two tokens of [w] for /V in the singleton data, the [fw] is within the general sound patterns of the system. Segmental Development and Feature Hierarchy Matches with adults targets were observed for the majority of vowels and diphthongs, AV and vocalic l&l being exceptions8. A few minor differences from the Vancouver area dialect appeared, probably related to the parents' regional dialects. (See table 3.3, figure 3.2, and appendix 5, table 1.1.) For all consonants and cluster segments, 39.96%9 of the child forms matched adult targets. Match data revealed the following: 1. Nasals, glides, final stops, and labial fricatives were established. 2. Syllable-final (SF) position was advanced over syllable-initial (SI) position for segmental realizations (42.36%), a tendency also noted above for clusters. 3. WI position had the fewest matches (30.33%). Nonmatches included both missing sound classes and within class feature differences as follows: 1. Missing sound classes were liquids and nondistributed coronal fricatives/affricates. 2. The lack of aspiration for initial voiceless stops plus missing /h/ denoted lack of specification for the feature [+spread glottis] 3. Some other voicing mismatches were noted but represented only 1.36% of total feature differences. Summary of Feature Differences The following represented the most salient of the feature differences in descending order (refer to figure 3.2): A. Unspecified child forms where adult targets are specified 1. Root node: [+consonantal] was unspecified for liquids (11.62% of total differences [percentage expressed in terms of total feature differences throughout this section]), although it was specified once for HI (1/35 times [2.86%]). 2. Root node: There was no specification (i.e., missing element) for some consonants — 7.52% of total differences. 3. Laryngeal node: [+spread glottis] was unspecified phonemically (6.15% of total differences), resulting in [?] for /h/, and was also not available as a redundant feature for many unaspirated voiceless WI stops (5/37 matches [13.51%]).10 4. 'Complex' structure of SI affricates was not in place (1.83% of syllable-initial total differences, but in six out of eight cases in absolute terms). Table 3.3.—Si: Summary of singleton segmental acquisition at Initial Assessment characterized by sound class, matches and nonmatches according to word position, and feature differences Sound class Word position matches Word position nonmatches Main feature differences Vowels Most [+rd] u or *sl Rhotacization u Glides ~/w, j/ All Nasals All Stops [-voice] WF, SFWW, (SIWW)2 WI, SIWW (/p/,/t/) WI: [-(-spread glottis] u Cor: [+dstr] *s : 'complex' *s Stops [+voice] SFWW, (WI), (SIWW), (WF) Wl(all); SIWW.WF (/d/) Minor infrequent errors Cor: [+dstr]*s Fricatives --Labial WI, (SIWW), (WF) SIWW, WF (/v/ only) Minor infrequent errors Fric/Affricates — Coronal None All [+dstr] *s or u (/67, /a/) 'complex' *s or u Liquids None All Both: [+cons] u Ixl: Lab Place for Cor* Ihl None All LN: [+sp gl] u Note: Similar tables appear for all Ss-*u= unspecified; italics, *s = 'specified inappropriately with reference to appendix 2 Feature Specifications for English 2Parentheses around word position initials indicate partial establishment: 40 - 80% matches under the Matches column, 10 - 40% matches under the Nonmatches column. These are arbitrary cutoffs used as guidelines for selecting targets. 47 Root node ([+cons]) >_0 [+son] _ - l + n a s ] ([cont]*) (Laryngeal _-~ node) 0~ ( [voice]) (([+spread glottis])) Place node Labial node ([+round])** Note: Parentheses indicate marginal features in system development. Double parentheses indicate very marginal features. Italics indicate inappropriate feature. Dotted lines represent marginally established nodes or features. *'Complex' branching structure inconsistently applied * * [u] not [+rd] &[+dstr] = default interdental tongue place Figure 3.2. SI : Feature geometry at Initial Assessment 48 B. 'Specified11 child forms where adult targets are unspecified or otherwise specified 1. Place node - Coronal Place: The[+distributed] interdental or dental place was a feature of sibilants and some coronal stops (19.83% ). This describes the interdental tongue position. The'default' coronal segment for SI was not [+anterior] but [+distributed]. Coronal place node was underspecified in the extreme, as the lack of clearly distinguishable stops, fricatives, and affricates demonstrates. 2. Place node: Labial for Coronal Place node appeared for realizations of /r/ (see section on Block 1 segmental intervention plan in chapter 5 for further discussion of glide substitutions for liquids), a few tokens of /0/ (13.16%), and stops in clusters with stops and [w]. 3. Root node: [+continuant] appeared as a feature of the first element of clusters with [w]. This feature difference and the one above were often concurrent and facilitated the appearance of two-element clusters. 4. Coronal fricatives and stops had 'complex' branching structure in WI position 9/41 times (2.45% of total syllable-initial [SI] differences). Intervention Plan: Block 1 As shown in table 2.2, Si's major treatment conditions proceeded in the order segmental/ syllabic. Subcondition orders were lower level feature/higher level feature, and moraic/onset-rime. Segmental Intervention Plan Lower Level Feature Targets: /J7 and/qV — Coronal node [-anterior] (all word positions) Higher Level Feature Target: IV - conjunction of [+consonantal] and [+sonorant] (WI and SIWW positions) Rationale 1. By establishing alveopalatals, the coronal place node would be specified with one of its marked features — [-anterior] — and the child's 'default' category [+distributed] would be circumscribed. Furthermore, by using one voiceless fricative and one voiced affricate in training, distinctions between complex and noncomplex segments, and voiced and unvoiced exemplars of 50 the [-anterior] class, could be made more salient (although the segments were not specifically trained to contrast these features). Generalization could be observed by testing /tf/ (and iy, if it could be elicited). At the onset of treatment, Si was stimulable for [J] phonetically some of the time in isolation and syllables, but could not produce a [<&] easily. A rounded fricative or affricate, produced with a low flat tongue position, was the usual nonmatching imitative production. 2. Because the the Root node feature [-(-consonantal] needed to be specified for/1/ (in conjunction with [+sonorant]), /l/ was considered a higher level target than the alveopalatals. (Even if a feature [+lateral] had been invoked for /V, N would have been higher, because the locus for the [+lateral] feature would have been the Place node, above Coronal.) The syllable-initial positions were targeted, because of a reasonable degree of stimulability and because of the parents' dialect, in which the postvocalic /V is barely audible. Imitative errors usually involved palatalization as [D] (merging with usual substitution), or protrusion as [L] {[^-distributed]). The III was chosen over Irl for liquid establishment because III was more phonetically stimulable than Irl. Because Si's mother did not use Ihl consistently in her dialect, Ihl was not considered an appropriate intervention target, although the [+spread glottis] feature was under observation throughout the study both for/h/ and for aspirated stops. Syllabic Intervention Plan Moraic Subblock Targets: Syllable-initial clusters /kw/, Ik]/, /pj/ Onset-Rime Subblock Targets: Syllable-initial clusters/tw/, /fj/, /vj/, /mj/ Rationale Because all other clusters contained unestablished segments, only /w/- and /j/-clusters were available as intervention targets. One dorsal and one labial cluster were chosen for the /jAclusters to provide a maximal contrast. No clusters were immediately stimulable in syllables. The onset-rime condition had more /j/ targets than the moraic condition, because of limitation of lexical items. This may have given an advantage to the onset-rime condition. However, the inherent difficulty of onset-rime target /tw/ (with the underspecified and somewhat problematic WI /t/) was considered an inequality favoring the moraic condition with its /kw/ target. No clusters were immediately stimulable in syllables. Results of Block 1 Intervention Figure 3.3 shows gains in proportional matches for trained and untrained segmental and syllabic targets. The greatest proportional gain was for aspiration on untrained WI stops - a probable byproduct of training in the moraic condition, during which aspirated stops were individual stimuli for stop-glide clusters. For trained targets, the average proportional gain was greater for the syllabic condition (16.65%; SD 5.36%) than for the segmental condition (2.14%, SD, 5.36%). Segmental Goals Lower Level Feature SI used no alveopalatals ([-anterior]). He marked the contrast between complex and noncomplex segments more accurately, completely eliminating the stop onset to initial coronal fricatives. However, cross-category substitutions still remained, as 7/14/t/ targets were realized as affricates, and 17/19 affricate targets were realized as fricatives in WI position. It appeared that the specified feature [-(-continuant] was overriding the unspecified feature [-continuant] for the coronal stops/affricates, most of the time for affricates, and some of the time for the stop. Higher Level Feature SI produced 3/20 (15%) SI N tokens ([+consonantal] /[+sonorant]) and one /bl/-cluster. A change other than the trained specification of [+consonantal] was the use of a vocalic onset to /!/ (three nonmatching tokens). Thus the segment with the higher level features seemed to be developing at a faster rate, even though it was not yet mastered. CQ 2 3 <u S on i < E 1 B '3 SYL:trained SEG: Untrained • SYL: WI CC - 1 syl. Dj] SEG: WI /l/-train. 0 XSEG-trained M SYL: WI CC - 2 syl. g SEG: fbV 0 SD SEG • SYL: TOT WI CC EP. x * • SEG: ASPIRATION @ XSYL-trained g SYL: TOT WI CC EP. V FJ SEG: COMPLEX V E l 5£> S Y L • SYL: SIWW CC-unt** S SEG: WI hi > COMP. x + Note: There were no matches for segmental training target alveopalatals nor for observation targets: #CCC, /h/, M and interdentals. *EP = short epenthesis (reduced vowel) **unt. = untrained + C O M P . = complex structure (affricate) Figure 3.3. SI: Proportional gain in matches for syllabic and segmental targets between Initial Assessment and Block 1 major probes Other Feature Changes Unrelated to Segmental Targets Increases in matches are noted with + and * for trained and untrained segments respectively. Progressive feature changes (without matches) on trained segments are denoted with ++, and on untrained segments with **. Regressive changes are denoted with &. A. Unspecifed child forms where adult forms are specified Laryngeal node: Range of distribution for redundant feature [+spread glottis] was expanded in the stop category (22/52 matches [42.31%]). This feature change possibly resulted from 53 syllabic condition training, since aspirated isolated phones [kh] and [th] were stimuli in training of the /k/- and/t/-clusters. ** B. 'Specified child forms where adult targets are unspecified or otherwise specified 1. Root node: [^continuant] for the first element of [w]-clusters was reduced in range of application to IV- and /d/-clusters only. + 2. Place node: Labial for Coronal Place was reduced in range of application to It I- and /d/-clusters, /r/, and/9/. + Syllabic Goals Moraic Subblock Because of consistent epenthesis use, no exact matches were noted for/kw/in the major probe, although the /k/ and /w/ elements matched as segments within the clusters in all four instances. All moraic target clusters with /j/ matched. A minor probe in the previous week had slightly different results (see figure 3.4). For both SI positions, /kw/ was completely correct 33.33% of the time. If short epenthesis was considering a minor phonetic timing deviation and thus ignored as a mismatch, 50% of the productions matched the target. Clusters with /j/ were completely correct 76.19% of the time, and 80.95% of the time when short epenthesis was not considered an error. Onset-Rime Subblock No matches were noted for /tw/ in the major probe, and IV was correct only twice in five attempts, ([f] surfaced for/t/ twice and [d] once.) There was some progression in realization of the /mj/ clusters, in that one production had a rounded front vowel for /ju/ (/mju/ > [my] in musicwriting') but there were no matches. Again, results in the minor probe were slightly different. For/tw/, there were 10% matches without epenthesis and 35% with short epenthesis. Clusters with /j/ matched 80% of the time (no epenthesis). 54 100.00% 80.00% S3 20.00% 0.00% H MOR:/kw/:Epcn \ • MOR:/Cj/:Epen x M OR: /Cj/ M MOR:/kw/:Epen x H OR:/tw/: Epen V H MOR:/Cj/ :Epen V I I OR:/tw/: Epen x Figure 3.4. SJ.: Proportional matches for syllabic targets in Block 1 minor probe Other Syllabic Changes Overall there was a dramatic change in initial cluster production with elimination of the [f] assimilation for all but coronals, and an increase in the use of epenthesis. If items with short duration epenthesis were counted as matches, 64% of the child forms matched the cluster targets in terms of number of elements. Furthermore, most /s/-clusters were realized with both elements. The percentage gain for WI clusters was 13% for single-syllable words, and 22.19% across all word shapes (counting all items with epenthesis as mismatches). For SIWW clusters, there was an overall improvement of 17.5%. Intervention Plan: Block 2 Segmental Intervention Plan Lower Level Feature Targets: /J/ and /qV ~ [-anterior] (all word positions as in Block 1) 55 Higher Level Feature Target: WI and SIWW /V, WI AV-clusters -- [+consonantal] /[+sonorant] Rationale SI had not mastered the Block 1 segmental targets, and hence they were repeated. Because WI clusters were developing rapidly in terms of skeletal slot establishment, development of /Cl/-clusters was considered a segmental goal at this point. Expanding the range of targets for /V was a logical next step, given the incipient generalization observed for / l / in the major probe. Syllabic Intervention Plan Moraic Subblock Targets: 1. /kw/ (one half-session for generalization to conversation) 2. /st/in WI position Onset-Rime Subblock Targets: 1. /tw/ (one half-session for generalization to conversation) 2. /st/in WI position Rationale Of all of the WI two-element clusters, the only one without two elements in the probe was /st/. In WF position, /st/ was established at the syllabic level and in one case /ts/ was produced correctly as [+anterior] rather than [+distributed\. These facts motivated the choice of /st/ as a syllabic goal. Choosing a syllabic target which would require some phonetic (and therefore) segmental training was a departure from the general design. Because the only other option was triple consonant WI cluster training (which would also involve /s/), and because of the general need to address the WI /s/ and IxJ neutralization, E decided to proceed with /st/ as a syllabic target. To maintain the syllabic focus, the data was then analyzed separately in terms of WI skeletal slots and in terms of features. Utilizing the same target for both conditions provided an opportunity to evaluate the effects of introducing stimuli as onset-rime versus moraic constituents without the influence of phonemic factors. In order to minimize the intercondition interference, an attempt was made to match lexical item to condition so as to maximize independence and meaningfulness of stimuli, for example, teta* for moraic, in which adding an initial [s] (o an unaspirated stop provided two words for contrast, versns ^ a a n s for onset-rime, in which the addition of [s,J provided two words for contrast. Results nf Rlock 2 Tntervf nfi'™ Figures 3.5 and 3.6 show results from the minor and major probes in Block 2 respectively. M MOR:/kw/:Ep V | MOR:/kw/:Ep x U OR:/tw/: Ep V • OR:/tw/: Ep x M / st/: TOT- CC H SEG: N | SEG: Alveop. Ill /•V AJ7 Figure 3.5. SI: Proportional matches for syllabic and segmental targets in Block 2 minor probe The minor probes results demonstrate a subblock advantage for the syllabic targets, but the major probe results show an advantage for the segmental targets (average proportional gain of 80.85%, SD, 17.65% compared with 48.92%, SD 7.25% for syllabic targets). This reflects the mastery gained within the week's vacation between the 18th and 19th sessions for Block 2 segmental targets - a consolidation period. As can be observed, best results were achieved for syllabic targets /kw/ and/tw/ (when short epenthesis was ignored as an error). The lowest number of matches were observed for segmental targets Afe/ and N. However, segmental target /J/ had a greater match proportion than syllabic targets /tw/ (when epenthesis was counted as a mismatch), and had an only slightly lower match proportion than syllabic target /st/. M 3 ca 2 3 CQ S <§ •S 03 # 100.00% -q 90.00% 80.00% -= 70.00% -= 60.00% -= 50.00% 40.00% -= 30.00% -= 20.00% - | 10.00% - | 0.00% -= -10.00% 11 111 11 III 1111 11 III Sy: Trained Seg: Trained | /kw/ - EPEN V * • SIWW A/ B SFA/ f§] ASPIRATION Hj /tw/ - EPEN V * [LTD /Cl/: EPEN V B a t / @ XSYL-trained H /st/: CC slot* g/oV 0 COMPLEX V 0 XSEG-trained H wi/i/ 0/1/ 0/h/ En] XSEG-untrained Figure 3.6. SI: Proportional gain in matches for syllabic and segmental targets between Block 1 and Block 2 major probes As can be seen, best results were achieved for SIWW M, /Cl/ (short epenthesis ignored as an error), /dy, and /J/. The apparent regression from the m;inor probe to the major probe for the syllabic target /kw/ reflects particular difficulty with one item - quarter, produced without a [w]. Segmental Goals Figure 3.7 shows Block 2 major probe proportional matches for trained and untrained segmental targets. 100.00% E 20.00% 0.00% HI Alveopal. comp +sp gl Untrained • WI/1/ I I HI TOT: EPEN V HJ] C O M P L E X \ II srww HI • /*/ ^ FRIC-not COMP * • /C1/:EPEN V H /x/ H/h/ • SFWW/1/ H/tj/ |gj ASPIRATION H WF/V ^ ALVEOP. TOT Note: There were no matches for /zl, ft/, AV, or /8/. Figure 3.7. SI: Proportional matches for trained and untrained segmental targets in Block 2 major probe Lower Level Feature Target: Alveopalatals ~ [-anterior] In the major probe, all alveopalatals except III were used in conversation. Over all phonemes and word positions, 40/61 (65.57%) child forms matched adult targets. Voicing was appropriate across all items. There were unequal results across word positions and phonemes. The best results were for WF /dy (100% matches) and /J/ in SIWW and WI position (86% and 83.33% respectively). Over all word positions the rank order of matches was Afc/ (80%) > /J7 (73.91%) > /If/ (55%) > /y (no matches on three attempts). In terms of features, there were also unequal results across word positions. In WI position, the 'default' feature [^-distributed^ was completely eliminated and the feature[-anterior] specified; in WF position, however, the feature [+distributed] was eliminated for the four/by words and for six of the ten /J7 words but not at all for the three /tf/ words (65% matches). Furthermore, specification or lack thereof for 'complex' structure was not yet completely established syllable initially (although now only in the alveopalatal class). Affricates were realized as fricatives 7/25 times, and fricatives as affricates 2/16 times. Higher Level Feature Target: / l / - [+consonantal] /[+sonorant] In the major probe, generalization of III was apparent across all word positions, even though only SI singletons and clusters had been training targets. Overall, there were 27/45 (60%) matches for /l/ as a singleton. Results varied across word position. For SI position, there were matches for 82.14% of the words (12/17 [70.59%] for WI position ,11/11 for SIWW position). Among the clusters, only /pi/ (42.86%) and /kl/ (33.33%) had matches, ignoring short epenthesis as an error. In the untrained SF positions, two out of three SFWW items matched and two out of fourteen (14.29%) of the WF items. Feature changes included establishment of [+consonantal] specification at the Root node across many items, elimination of vocalic onset, lengthening of the initial singleton, and one case of appropriate use of [+distributed\ (interdental place). One case of inappropriate specification of palatalization ([-anterior]) was noted. 60 Changes over Time for Trained Segmental Targets The changes noted for alveopalatals and /l/ occurred during the syllabic block. The single word data from the segmental subblock minor probe did not show this degree of generalization. (This parallels the situation in Block 1, where phonetic stimulability for alveopalatals increased while syllabic targets were being trained.) The token economy, reinforcing spontaneous use of any targets in conversation, possibly promoted this generalization. At the end of the segmental subblock, the minor probe showed more matches for alveopalatals than for M, a reverse of Block 1. Of the 38 alveopalatals, 15 child forms matched the adult targets (39.47%), in comparison with none at the end of Block 1. Only 5/25 (20%) of the AV words were matches, compared with 3/22 (13.64%) matches at the end of Block 1. However, the matches occurred for alveopalatals only in WF (9/15 matches [60%]) and SIWW positions (6/15 [40%]), with no matching WI targets. Recalling that for Si, WI position was the weakest in terms of matches in general, the apparent advantage for the alveopalatals over AV may be less than it seems. The three WI matches for M demonstrate an advance of M over the alveopalatals for the most troublesome word position. Other Feature Changes Unrelated to Segmental Targets Unspecified child form where adult form is specified (resulting in matches) Laryngeal node: Redundant feature [+spread glottis] was now specified more frequently for for/h/ (6/14 matches [42.86%]) and was present for more unvoiced WI stops (41/46 matches [89.13%]) Areas of No Change in Feature Specification 1. Coronal node: [+distributed] interdental or dental place use continued for /s/ and /z/ realization. 2. Root node: [+consonantal]/[+distibuted] remained unspecified for A/. 3. Root node: [+continuant] remained unspecified for /57. Syllabic Goals Major probe data were insufficient to compare moraic and onset-rime conditions. (Surface realization of two skeletal slots for/st/was incipient, as 5/9 [55.56%] matching tokens appeared in the major probe, and 11/17 [64.70%] appeared in the minor probe.) The minor probe data and the intervention session data from the syllabic subblock were examined, the latter only for/st/ (see figure 3.8 for session match data). i H M: A L L SESSIONS: ECHOIC H M: A L L SESSIONS: SPONT H I M: A L L SESS: TOTAL • M: INTRO TO SESSIONS: ECHOIC 11 M: INTRO TO SESSIONS: SPONT • M: INTRO TO SESSIONS: TOTAL • OR: A L L SESSIONS: ECHOIC H OR: A L L SESSIONS: SPONT H OR: A L L SESSIONS: TOTAL ggj OR: INTRO TO SESSIONS: ECHOIC • OR: INTRO TO SESSIONS: SPONT 0 OR: INTRO TO SESSIONS: TOTAL Note: Si ' s echoic and spontaneous responses in moraic and onset-rime conditions are shown across all sessions and across introductory parts of the subblocks, when stimuli were presented as discrete moraic or onset-rime constituents. Figure 3.8. Si: Proportional matches for syllabic target /st/ in Block 2 sessions Moraic Subblock A. Minor probe data If epenthesis was ignored, 23/23 matches were noted for /kw/. If epenthesis was acknowledged as an error, 69.57% of the items matched the targets. B. Session data for /st/ (skeletal slot results) Overall, for moraic condition echoic responses, there were 100/121 (82.64%) diconsonantal imitations. For spontaneous (including self-echoic or self-corrected) responses there were 252/280 (90%) diconsonantal productions, yielding a total of 352/401 (87.78%) skeletal slot matches in therapy sessions. In the introductory parts of the production training sessions, E made word constituency salient with verbal and visual cues. For these introductory portions, 34/36 (94.44%) moraic condition echoic responses were correct, whereas 106/121 (87.6%) spontaneous responses were correct. It appears that imitation was somewhat better when word constituents were presented separately than when the word was presented as a unit (66/85 - 77.64%). However, once the child could pronounce a word without a verbal model, direct access to constituency (by visual cueing) was perhaps not necessary (correct responses = 146/159 - 91.82%). Onset-Rime Subblock A. Minor probe data If epenthesis was ignored, 16/21 (76.19%) matches were noted for/tw/. If epenthesis was acknowledged as an error, 10/21 (47.62%) of the items matched the targets. B. Session data for /st/ (skeletal slots) Overall, for onset-rime condition echoic responses, there were 71/78 (91.13%) matches. For spontaneous responses there were 67/86 (77.91%) responses, yielding a total of 138/164 (84.15%) total matches. If only the introductory parts of the production sessions are examined, 54/59 (91.52%) onset-rime condition echoic responses were correct, whereas 34/37 (91.89%) spontaneous responses were correct. Word constituency presentation did not apparently influence correct imitation of words, but there was a notable gap between the introductory parts of sessions and total correct spontaneous productions of words assigned to this condition. Other Syllabic Notes From the feature data, it was apparent that the segments of clusters matched the adult target less than 50% of the time. This was because of the substitutions for /s/ ([0]) and/r/ ([w]). As noted, epenthesis continued to be used in stop and approximant clusters. Block 3 and Beyond Since the syllabic goals had been met in terms of number of consonant slots in diconsonantal clusters, a short program introducing triconsonantal clusters followed, complementing a review of /st/. The major foci of Block 3 were necessarily segmental: sorting out [+distributed] from [+anterior] (/s/ and /z/ versus /0/ and /57) and establishing [+consonantal] /[-fsonorant] with [^distributed] (/r/). In Block 3, there was some success in sorting out [+distibuted\ from [+anterior], although many of the alveolar sibilants remained phonetically dental. SI gained facility in imitation of [r]. Syllabic goals were met. In a follow-up speech elicitation, four months later, no change was observed for /r/ and alveolar sibilants. A weekly therapy program over a five-month period led to generalization of /r/ in syllable-initial position (including clusters). Because his parents' dialect does not include the postvocalic allophone, this was not a priority for therapy. As secondary dentition has been slow to erupt, and as finger-sucking and tongue thrust continue, the /s/ and /z/ are not yet amenable (at age 8) to consistently correct phonetic placement. Epenthesis in clusters gradually disappeared, particularly following success with /r/-clusters. Discussion In terms of the research questions the following interpretations are offered. 64 1. Will nonlinear phonological frameworks help to predict logical and attainable intervention goals for phonologically disordered children? The nonlinear frameworks did lead to reasonable targets for intervention. The intervention project may have been successful for reasons other than the reasonable choosing of targets, such as the intensive stimulation and the mutually enhancing effect of a variety of approaches. However, the generalizations noted below are a phonological indication that the chosen targets were relevant for inducing change in the system. Conversational Intelligibility Gains There was a remarkable improvement in intelligibility12 due to the attainment of conversational proficiency for predicted targets /V, /J/, /dy and diconsonantal cluster production, epenthesis notwithstanding (see figures 3.9 and 3.10). Untrained targets which generalized to conversation from the trained ones were, predictably [-anterior] /tf/ and (dentalized) /s/, clusters with [w] or [j] as substitutions, and two skeletal slot positions for all /s/-clusters except /st/. Other untrained targets which improved were voiceless initial stops (possibly since [-fspread glottis] stops were presented in both moraic and onset-rime conditions as models) and/h/. The latter was never presented in therapy, but may have improved as an extension of [+spread glottis] specification for WI stops. Facility with imitative production of [r] (unstimulable at the onset of intervention) may be attributed to the acquisition of /l / and the alveopalatals, since /r/ requires the specification of [+consonantal]/[+sonorant] as well as a retracted ([-anterior]) tongue position. In the year prior to the study and in the four-month break following, minimal changes occurred, lending support to the perception of the project's effectiveness. 100.00% Assess Block 1 Block 2 WI, SIWW/l/ ' * « /J/ SYL: #st as #CC m i /ay — SYL: #CC, EPEN \ Note: At Block 2, / l / data includes clusters (100%, short epenthesis ignored as an error). Clusters with epenthesis that were designated as matches had very short epenthetic vowels. When the epenthetic vowel was as long as other unstressed vowels (according to E's transcription), the child form was counted as a mismatch. Figure 3.9. SI: Developmental progression in proportional matches for trianed phonological targets across the intervention period Language Test Results Language test scores (within or above normal limits to begin with) improved in ways which support the conviction that phonological intervention, and not general language development, are responsible for the gains incurred. On the Peabody Picture Vocabulary Test (Form L), Si's score improved from a SS of 88 to a SS of 121 (average = 100) with nonoverlapping confidence intervals. On the Test of Language Development (Primary), there was improvement for vocabulary, word discrimination, and word articulation subtest scores and equivalence or slight regression of scores for syntactic/morphological subtests. These results were expected, since the enterprise of phonological intervention involves presentation of many (new) words, phonemic discrimination training, and phonetic training, with minimal emphasis (in the general case, and particularly for SI) on syntax. Note: Parentheses indicate partially established features in system developnr Italics indicate inappropriate feature. Dotted lines represent partially-establi constituents. *'Complex' branching structure more consistently applied to affricates; **[+conf] unspecified for the voiced interdental; [+cons] unspecified for /r/ onl &Less frequent interdental tongue place Figure 3.10. SI: Feature geometry at end of Block 2 66 2. Are the separate prosodic and segmental levels of representation of nonlinear phonology psychologically real? At the end of the two blocks, there were quantitative differences between the syllabic and segmental results. At the end of Block 1, there was a clear gain in percentage and types of syllable/word shape matches in comparison with segmental targets, the latter of which had not generalized to conversation. The segmental targets introduced in Block 1 were not mastered until the end of Block 2 (which accounts for the noticeable difference between Block 1 and Block 2 for segmental targets). By contrast, rate of change decreased for syllabic targets in Block 2. Generalization of the specific syllabic target /st/ had not been achieved at the end of Block 2, although widespread changes had occurred across the cluster targets in Block 1. Because /st/ also required segmental learning at the level of the coronal node, it was an expectedly difficult target. With respect to phonological development and the autonomy of tiers, it appears that there was some independent development that could be exploited in intervention. For this particular subject, syllabic changes were more dramatic at the outset and led to improved intelligibility within six weeks of the onset of therapy. Initial cluster development had begun prior to intervention. Taking advantage of this beginning development led to immediate results. Segmental changes occurred at a slower rate (note /r/ and /s/ particularly), both for phonological and phonetic (articulatory) reasons. Targeting the two tiers independendy and alternately was synchronous with this slower-changing aspect of the system. Complete autonomy of tiers was not supported by this subject's data, however, as the cluster data (particularly the /s/-clusters) showed. Where two tiers interact, especially negatively as for /st/, we might expect a slower rate of change, as was the case for SL 3. If the 'syllabiclprosodic tier' has some observable clinical reality, will there be a difference in proportion and rate of syllable shapes acquired as a result of interventions that contrast the onset and rime versus those that utilize the mora as a constituent? 68 Nonmatches and Syllable Theories at Initial Assessment In onset-rime theory, the onset is considered a constituent made up of up to two elements (not including extrametrical /s/). Kaye et al. (1987) suggest that intraconstituent government proceeds from left to right, with deletion and spreading being possible results of this government. In those cases where the first onset consonant was retained and the second was missing, there is support for the hypothesized direction of government (for deletion processes). However, the apparent Labial spreading from [w] to the preceding consonant, resulting in [fw] clusters, would seem to contradict the intraconstituent direction of government hypothesis (for diagram showing spreading for [fw], see figure 9.2). It does, on the other hand, seem to support the concept of onset as a constituent with its own set of operations. (No cases of leftward spreading link labiality from a labial vowel onto the preceding consonant; in other words, the labial spreading is a consonantal onset interaction with no assistance from the rime.) It is difficult to derive a moraic interpretation of the assessment cluster data, since onsets do not have any moraic status in that theory except where rules involve spreading from onset to coda position for creation or deletion of morae. The adjunction or lack thereof of consonants to a mora may be considered more of a segmental issue than a syllabic one. That only certain segments surfaced, depending on degree of feature-sharing (a segmental tier phenomenon) may support this aspect of the moraic theory. Intervention Data and Syllable Theory Quantitative Analysis Scores were overall slightly higher for moraic condition targets (according to the mean difference shown in figure 3.11) across the study. Greater differences were noted in the minor probes than across session data, however. ai O S o JS I i u c -10.00% 11 M O R - OR: B L 1 M I N O R • M O R - OR: A L L SESSIONS (E) • M O R - OR: B L 2 M I N O R • M O R - OR: A L L SESS. (S) • M O R - OR: INTRO SESSIONS (E) • M O R - OR: A L L SESS. (TOT) • M O R - OR: INTRO SESS. (S) U M O R - OR: X D I F F E R E N C E B M O R - OR: INTRO SESS. (TOT) M O R - O R : SD Figure 3.11. Sl_: Proportional differences between moraic and onset-rime match scores for minor probes, introductory parts of production sessions, and overall session data (spontaneous and echoic responses) In order to evaluate the aforementioned differences statistically, the Kolmogorov-Smirnov nonparametric measure (comparing cumulative frequency distributions of proportional matches for the two conditions) was chosen. Individually, there were eight different comparison measures available across the study for O-R targets, including four miniprobe comparisons, two major probe comparisons, and two session subblock comparisons. There were seven comparisons for moraic targets (only one major probe measure, due to insufficient data). The following null (HQ) and alternative hypotheses (Hi) were tested using the Kolmogorov-Smirnov measure (see table 3.4). HQ: There will be no difference between the proportional match scores for all of the moraic (m) and onset-rime (n) measures across the study. 70 Hi: Moraic targets had proportionately higher proportional match scores across the study. Table 3.4.—£1: Proportional matches for onset-rime and moraic targets across the study cast for Kolmogorov-Smirnov test <10% 31-35% 36-40% 46-50% 76-80% 81-85% 86-90% 91-95% 100% s m(x) 0/7 0/7 0/7 2/7 2/7 4/7 5/7 5/7 7/7 sn(x) 1/8 2/8 3/8 3/8 6/8 7/8 7/8 8/8 8/8 p>0A The largest discrepancy between number of proportional match data sets falls in the 76-80% column. This yields a test statistic of (8)(7)(.4625)=25.9, which fails to reach the level of significance at the p > 0.1 level (one-tailed). Qualitative Analysis Although the slight advantage for the moraic conditions was not statistically significant for this child, the actual differences will be discussed below in order to ascertain whether there may be some clinical or linguistic relevance to the observed difference. Specifically, the moraic condition was more successful for/kw/ and for spontaneous productions of /st/ over all sessions. The difficulty with the onset-rime target /tw/ may, however, have reflected difficulty with WI AV at the segmental level. A minimally specified phoneme, such as IX/ in English, may be more inaccessible for change than a more highly specified phoneme, by virtue of having nothing remarkable to identify itself. For Si, WI III was often affected by the interdental default tongue position ([+distributed\) for coronals, making change difficult for articulatory reasons. In the onset-rime subblock /t/ itself was not a focus when targeting the /tw/-and /st/-clusters. In the moraic subblock the edge member of the cluster (i.e., Ikf) was modelled and imitated separately from the consonant directly preceding the vowel. In other words, the type of stimuli in the onset-rime condition may have contributed to slower rate of change for 71 spontaneous productions of /tw/ and /st/, because of general negative interactions of the segmental and syllabic tiers for coronals and clusters. Results were more contradictory for /st/ session data. Over all therapy sessions for /st/, the onset-rime condition had a higher percentage of correct echoic responses, whereas the moraic condition showed a higher percentage of correct spontaneous productions of /st/ syllables and words. However, during the introductory parts of the production sessions (step #3 in training), E presented direct verbal and visual models of word constituency. For these parts of the sessions echoic moraic condition trials had a 3% advantage over the onset-rime trials. Since both conditions had over a 90% rate of correct responses, the result may not be clinically interesting. Spontaneous productions are the primary responses of interest in an intervention program, but those particular results for /st/ session data are also somewhat ambiguous. During the introductory parts of production sessions, onset-rime words had a greater success rate, but over all sessions, there were a greater number of matches in the moraic condition. This may have reflected an extraneous variable ~ the greater number of trials for the moraic condition, giving more practice. A learning effect from the moraic to the onset-rime condition may have resulted in success in the introductory portion of the onset-rime program, but the onset-rime condition itself and/or the lesser number of trials failed to promote the same degree of generalization. A factor not apparent in the training target data per se may indicate some general learning about clusters based on stimuli in the onset-rime condition. From the assessment to the end of Block 1, the use of epenthesis increased, enabling consistent realization of WI stops in clusters. Onset-rime subblock training methods may have promoted use of epenthesis, since the [w] or [j] of the [tw]-, [mj]- or [fj]-cluster was presented as an [°.w] or [i.j] to make a pronounceable constituent separate from the rime, e.g., twine > [r^ -wuln]. 4. If the 'segmental tier' has some observable independence, is there any advantage to be gained from targeting specified features at 'higher' versus 'lower' levels in the feature hierarchy in phonemic inventory intervention? Block 1 At the end of Block 1, AV was correct in WI position 3/22 times, whereas there were no matches for the alveopalatals. It appeared that the higher Root node feature [+consonantal] conjoined with [+sonorant] was being more quickly acquired than the lower Coronal node feature [-anterior]. This was a prediction of the feature hierarchy approach. Block 2 At the end of Block 2, both sets of targets ~ A/ and the alveopalatals - had generalized to conversation, demonstrating an apparent "catch-up" effect for the [-anterior] feature in comparison with the Block 1 major probe, and a "catch-up" effect for III from the Block 2 minor probe. There was a slight advantage for the alveopalatals over III overall and in WI position at the time of the major probe, although there had been no matching WI alveopalatals a month before, at the time of the minor probe. One of the notable therapy observations at the beginning of Block 2 was the greater ease of articulation for alveopalatals than in Block 1, even though intervention had focused solely on III and the syllabic targets in the interim. The rapid increase in generalization of alveopalatals in Block 2 may indicate that phonological learning was being held back primarily by articulatory factors for this set of targets. The double "catch-up" effects may indicate that the feature hierarchy level did not influence rate of development in Block 2, after the phonemes had begun to emerge. However, the earlier success with III at the end of Block 1, which continued marginally for the trained word positions in Block 2, seems to lend partial support to the relevance of height in the feature hierarchy for development and intervention. Furthermore, by targeting two features simultaneously, it is possible that the higher level feature may positively influence establishment of a lower level feature, particularly if, as in this case, this other feature is a marked feature (i.e., [-anterior]) of a default place node for the first (Coronal). In support of feature relevance are the generalizations observed for related untrained targets. For example, /tf/ became evident in the child's speech at the same time as its [-anterior] counterparts - /J"/ and /dy - although not at a lower match proportion level, the predicted outcome for an untrained target. The general learning regarding the specification or lack thereof for 'complex' structure among stops, fricatives and affricates appeared to have been facilitated through the training of /]7 and /qV, two segments differing in terms of the branching/nonbranching structure. The facility gained with imitation of Irl (unstimulable at the outset) may suggest a general development in articulatory skill, but also may suggest the extension of the conjunction of [+consonantal] with [+sonorant] gained from the learning of IV. From the syllabic condition training, a byproduct appeared to be the consolidation of aspiration on WI voiceless stops (the redundant feature [+spread glottis] possibly learned from practise with [kh] as a separate constituent). Once the voiceless stops were consistently produced, /h/ appeared in the child's speech - suggesting generalization of the redundant [+spread glottis] feature to the phonemic level. Beyond feature hierarchy predictions, word position was shown to be relevant for the process of segmental acquisition (as has been noted often by other observers, e.g., Branigan [1976]; Chiat [1989]). The generally more difficult word position for SI was WI position, once again observed by the emergence of WF alveopalatals first (in spite of training in all word positions). For /V there may have been easier access to WI than to WF position because of the allophonic/dialectal variation between the word positions. However, the fact that it was possible to establish one of the targets — / l / — in the more generally difficult word position may have facilitated the emergence of the other WI targets - the alveopalatals. Whatever the feature hierarchy model provides in terms of a reference for a child's phonemic system needs, the interaction of the segmental and syllabic variables must also be identified. Summary SI achieved all of his phonological goals for the project, gaining conversational proficiency for trained segmental targets HI, /J/ and l&J, and trained syllabic cluster targets /Cw/ and /Cj/ and skeletal slot specification for /st/-clusters. Gains were greatest for syllabic targets in 74 Block 1, and segmental targets in Block 2. Interaction of segmental and syllabic constraints made the syllabic target /st/ more difficult than the syllabic targets for Block 1. Generalization to untrained targets took place in both conditions in predictable directions ~ in the segmental condition from /J/ and /qV to /tf/ (another [-anterior] phoneme), and in the syllabic condition from /Cw/ and /Cj/ to skeletal slot marking and specific phoneme realization for all clusters. Specification of 'complex' structure for affricates as opposed to fricatives and stops was a byproduct of the alveopalatal training, although it was not a direct objective of the training. A probable byproduct of the cluster training was the appearance of aspiration (the redundant feature of [+spread glottis]) for WI stops, a feature which may have transferred to the phonemic level, resulting in emergence of /h/, acquired spontaneously during the intervention period. Using the Kolmogorov-Smirnov test, there was no statistically significant difference between proportional matches for onset-rime and moraic conditions. Moraic condition training appeared slightly more successful than onset-rime training for spontaneous productions of /st/ and overall for /kw/. However, the imitative parts of the /st/ training showed a slight advantage for the onset-rime constituents. Si's general difficulty with A/ may have impeded development of /tw/, especially since the A/ was not presented as a separate stimulus, but as a unit with the glide. The use of epenthesis generally fostered development of cluster element realization. This may have been a result of onset-rime subblock training, in which the [w] or [j] was often lengthened to an [u] or [i], to make a pronounceable constituent. The marginally faster rate of acquisition for AV, with its higher level feature conjunction [+consonantal]/[+sonorant], compared with the alveopalatals, with their lower level feature [-anterior], may imply relevance of the feature hierarchy for Si's development. Lack of change for untrained targets hi and the interdentals ([+distributed]) both during and following the project supports the conviction that intervention was responsible for change. CHAPTER 4 SUBJECT 2: SUMMARY [ b a b w s benha? , p n o u W A ? ] Case History Information Subject Characteristics During the reported period of the study, this boy was 4;2 - 4;6. The phonological disorder was associated with: 1. A history of chronic otitis media from age 1 year, with a myringotomy three months prior to study for "glue ear." (Hearing screening at bimonthly intervals throughout the study showed hearing within normal limits.) 2. A mild delay in language production, reflected in inconsistent use of copula and auxiliary BE, pronominal case errors and sibilant morphophonemes 3. A mild attention deficit possibly resulting from the middle ear history (see table 4.1). Prior Treatment S2 had participated in a group articulation therapy program for three months prior to the study. Focusing on fricatives had led to production of some sibilants in WF and SIWW positions. Phonological System at Initial Assessment Phonotactics: Syllables and Words S2's phonological system was established at the syllabic level with exceptions in cluster development and positional distribution of coronal obstruents and /!/ (see appendix 4, table 4.1). 75 Table 4.1.—Subject information for S2 AGE 4;2 - 4;6 for reported period GENDER Male BIRTH ORDER Second of 2 boys FAMILY HISTORY HEARING OTHER DEVELOP. HISTORY ORAL MECHANISM English-speaking (Canadian-raised parents) Father firefighter Mother school staff assistant (not employed during study) Chronic otitis media from age 18 months Myringotomy three months prior to study Hearing screened within normal limits at onset and at periodic intervals throughout study Mild attention deficit probably secondary to hearing loss history No unusual pregnancy, birth or medical history Age-appropriate motor skills No apparent deficits in cognitive and social skills Normal for speech production except for minimal overjet Stimulable for all sounds of English in isolation (some difficulty with suppression of tongue tip for velars) LANGUAGE Comprehension* PREVIOUS THERAPY 1. Peabody Picture Vocabulary Test (Form L) — average range 2. Test for Auditory Comprehension of Language (Revised) — average range Phonemic discrimination* Auditory Discrimination Subtest of the Test of Language Development (Revised) - average range Production 1. Structured Photographic Expressive Language Test - Preschool — borderline low average range. Morphological errors reflecting phonological errors (/s/ and M morphemes), copula /auxiliary omissions and pronominal case errors 2. Language sample consistent with SPELT-P Sentences of 4-6 words, with some longer complex sentences Three months group articulation therapy for stimulation of fricatives (some change for sibilants) with another speech-language pathologist *See table 4.3 for scores. 76 Syllable-Initial Cluster Development The more advanced categories were /CI/ clusters and /Cj/ combinations. A. /l/-clusters 1. Most SI /l/-clusters were produced with two consonants and full epenthetic vowels: /CI/ > [Cal], although /pi/ once had a reduced epenthetic vowel ([C9l]short epenthesis). 2. Both WI /pi/ and /fl/ were realized with one segment only, /pi/ two of four times: /pi/ > [p], /fl/ > [1] (one matching exemplar). B. Stop-glide clusters 1. The WI /Cj/ combination was realized with /Cj/ two out of three times (once with short epenthesis) and /C/ once, although the SIWW /Cj/ combination was realized only with /C/. 2. The sole token of a /kw/ cluster was realized with the coronal [t] only. C. /r/ and /s/-clusters Syllable-initial /r/ and /s/-clusters consisted of one segment: /Cr/, /sC(C)/ > C. Syllable-Final Clusters Syllable-final clusters consisted of two elements with the following exceptions: A. Nasal clusters The coronal /nt/- and /nd/- clusters were realized with one element some of the time: /nt/ > [t] (five out of nine cases) and /nd/ > [n] (one out of two cases). B. Stop-sibilant clusters As with the nasal clusters, clusters with two coronals were more often reduced than other combinations. 1. The /ts/-cluster was realized with [t] only once in three attempts. 2. The solitary /st/-cluster exemplar had both elements missing. 3. The /ks/ cluster was realized with [?] three out of six times. (The WF /k/ was realized as [?] — also a common realization of WF/t/). C. Liquid clusters The /l/-clusters were slightly more advanced than the /r/-clusters in terms of skeletal slot realization, just as the HI was better-established as a segment. 1. Although HI was specified for [+consonantal] syllable initially, there was a vocalic substitution for/]/ in /IC/ clusters. For/lk/, the IkJ appeared as [?], its usual WF substitution. 2. Clusters with /r/ were realized three out of seven times with two elements, the first a vocalic substitution for /r/, three times with the second consonant only, and in the case of /rf/ not at all. Positional Syllabic Constraints Coronal segments had different realizations in different word positions: 1. In the case of WF position, IxJ had no Place feature word finally --WF IxJ > [?] 7/19. 2. As noted above, SF HI was realized as a vowel, i.e., without [+consonantal] specification. 3. Coronal sibilants, although usually realized with an interdental tongue position word finally (and occasionally in SIWW position), had no Place feature word initially. All fricatives in fact were realized as [h] in WI position. Only the interdentals deviated from this pattern some of the time. There were three alternate substitutions for /HJ, which matched 4/12 times, but otherwise appeared as [w] (four out of eight times), [h] (three times), or [d] (once). The unvoiced cognate was realized as [h] in thumb but as [?] in thing (both Laryngeal node substitutions). Segmental Development In general, the segmental system showed development within all sound classes, with word position (syllabic) constraints evident for stops, fricatives and liquids as noted above. The only completely established classes were vowels and glides (see table 4.2, figure 4.1, and appendix 4, table 2.1). The order of match data across word positions (excluding clusters) showed more advanced development for syllable-initial position over syllable-final position: WI (54.5%) > SIWW (48.7%) > WF (31.78%) > SFWW (18.75%). Across all consonants and cluster segments, 43.55% of the child forms matched adult targets. Table 4.2.—£2: Summary of singleton segmental acquisition at Initial Assessment Sound class Word position matches Word position nonmatches Main feature differences Vowels Adequate for intelligibuity Occasional centralization of/e/ [-bk] u1 Glides — /w, j, hf All Nasals /m/,/n/ /rj/: Dorsal Place u Stops [-voice] /p/: WF, SIWW /t/-SIWW, (WI)1, (WF) /k/-all (/p/-WI) WI: [+sprgl] u61% /k/: Dorsal Place u WI/t/: [+dstr]m*2\9% WF/t/: Place node u 37% Stops [+voice] /b/-All /d/-All /g/ - None /g/: Dorsal Place u Fricatives -Labial Neither Both WI: Place node u SIWW,WF/v/: RN: [+cont] u, default Place node SIWW/f/: Place u (3) or RN: [+cont] u (2) + default Place node (1) WF/f/:RN: [+cont] u (3), no Place or RN: [+cont] m (2) + default Place (1) Fric/Affricates — Coronal None OW) All others WI and (SIWW): [h] = RN:[+cont] m; LN: [+spr gl] m, no Place WF, ((SIWW)): [+dstr] *m or u (for /5/) or (R: 0 or [+cont] u) Most 'complex' u Liquids HJ - Syllable-initial HI - Syllable-final A-/-A11 Both: [+cons] u; *Labial Place for Coronal (Syllable-final: R:0) Single parentheses indicate partial establishment: 40 - 80% under the Matches column, and 10 -40% under the Nonmatches column. Double parentheses indicate marginal development — less than 10%. 2u= unspecified; *m, italics = specified inappropriately with reference to appendix 2 79 (Labial O node)& [+round] (Coronal 6 node) : Dorsal node)&& O (l+distributedl) ([bk]) [hi] [lo] Note: Single parentheses and dotted lines indicate partially established nodes or features. Italics indicate nonmatching feature. *'Complex' branching structure for affricates marginally established or used for nonbranching consonants **[+spread glottis] established except for aspiration on WI stops &Established except for fricatives &&Established for vowels only Figure 4.1. S2: Feature geometry at Initial Assessment 80 Match data revealed the following: 1. Overall, vowels and glides matched adult targets sufficiently so as not to interfere with intelhgibility. 2. The order of sound class establishment was vowels/glides > nasals > voiced stops > voiceless stops > liquids/fricatives. Nasals were well established except for Dorsal Place /rj/. Stops were also established for place except for Dorsal. Overall, the WI voiceless stops that were produced (/p/ and IxJ) matched 40.54% of the time, because of lack of [+spread glottis] specification (aspiration) and inconsistent dentalized [+distributed] production of the coronals. Some tokens of SIWW and WF IxJ were not produced, or realized as [?]. This suggested either a complete lack of Root node specification or specification of Root node [-(-consonantal] and Laryngeal node (the default being [?]). As noted above, fricative production was incipient. In WF and SIWW positions, there were some /s/, /z/, and /]7 sibilant productions, most of which were inappropriately [^distributed]. When the [+distributed\ sibilant did not appear, a [t], [d] or [h] was the substitution. Occasionally, the voiced fricatives /z/ and /v/ were missing altogether. In the case of the stop substitutions, Place node was specified, but Root node feature [-(-continuant] was unspecified. In the case of the common WI [h] substitution, one could say that the Root node [+consonant/+continuant] and Laryngeal node [-(-spread glottis] features were specified, but that Place Node was not specified. The [h] substitution for fricatives paralleled the [?] substitution for stops IxJ and /kj, but for the opposite syllable position. One liquid - JIJ - was established, but only syllable initially (including in clusters with epenthesis as noted above). In summary, the Place node as an intermediate node was not as well established as Root node and Laryngeal node for consonants, with no specification at all for Dorsal Place. Where established, Place node markings were dependent on sound class and syllable position. For stops, Labial Place was better-established than Coronal across positions. For fricatives, there were some coronals, but no labials. The liquid JrJ as well as SF JIJ were realized as glides in the Labial Place category, rather than as coronals. Feature Differences for Consonants Nonmatch feature differences were as follows (in an approximate descending order determined pardy by word or consonant frequency and partly by feature difference salience): A. Unspecified child forms where adult targets are specified 1. Place node: WI fricatives (and some fricatives in other word positions) were unspecified for intermediate or specific Place node, as was WF /t/ (some of the time) and SIWW and WF /k/ (most of the time). 2. Place node: Dorsal Place was not specified for consonants. 3. Root node: There were missing segments (unspecified) in clusters, and in WF position --WF/t/, /k/,/v/,/z/,and/r/. 4. Root node/Coronal node: [-(-consonantal] and[+distributed] were unspecified for /r/. 5. Branching structure ('complex') was lacking for affricates except occasionally in WF position. 6. Root node: [-(-continuant] was unspecified for WI /b7, WF /8/, SIWW and WF /v/ and (occasionally) sibilants. 7. Laryngeal node: Voiceless initial stops were unaspirated 38/62 times (61.29%), showing lack of development for redundant feature [+spread glottis] 8. Laryngeal node: [+voice] was unspecified for WF coronal fricatives/affricate substitutions for /z/, /5/ and /qV 18/31 times (58.06%). B. 'Specified' child forms where adult targets are unspecified or otherwise specified 1. Place Node: Coronal [+distributed] dental or interdental place was noted some of the time for sibilants and coronal stops. 2. Labial for Coronal Place was noted for/r/ and optionally for /57 (one word — that). Intervention Plan: Block 1 In Blocks 1 and 2, conditions for £2 followed the pattern: segmental/syllabic. The segmental subblock order was lower level feature/higher level feature. The syllabic subblock order was onset-rime/moraic. In this way, S_2 contrasted with S6 for subblock orders. 83 Segmental Intervention Plan Lower Level Feature Subblock Target: WI/k/ — Dorsal Place node Higher Level Feature Subblock Target: WF and SIWW /v/ - Root node [-(-continuant]; Labial Place node Rationale A. Basic rationale The most salient lacks in the system, irrespective of word position, were Dorsal Place for consonants and Labial Place for fricatives. B. Dorsal Place Because /g/ had both Root node and Place node specification across word positions, it could have been a target in any word position. The small English number of words containing /g/ compared with larger numbers containing the other dorsal consonants made /g/ a less desirable target choice. The nasal /rj/ was not considered salient enough perceptually to be a target, particularly considering the acoustic similarity of [inj and [in.]. This left /kj as the best choice. Unfortunately, substitutions for /k/ were not consistent across word positions. Furthermore, the only word position where consistent Root and Place node markings resulted in definable consonantal substitutions for/k/ was WI position (i.e., not a [?] substitution location). In order to adhere to the general experimental design, which permitted targeting of new segments in well-established word positions only, WI position was the word position choice for establishment of/k/. B. Labial fricatives Interaction of segmental and syllabic tiers was evident from positional differences in substitutions for fricatives. The WI [h] substitution was considered more a syllabic than a segmental concern, both because of its limitation to syllable-initial position and its application across fricative groups. In WF position, the limited evidence of sibilant production demonstrated application of the Root node [+continuant] feature through the unmarked Coronal Place node. However, this feature was not applied to all fricatives in this word position. Among the similarly low frequency fricatives /v/, /67 and AV, the /v/ was selected because of (1) its generalization potential to A7 across word positions and (2) the articulatory tendency for crossover between Coronal Place alveolar and interdental sibilants, rendering the coronals phonetic as well as phonological targets. To establish /v/ in WF and SIWW positions, it was necessary to specify the Root node [+continuant] feature and Labial Place node in order to avoid the default [d] substitution. The establishment of two specifications for /v/ was not considered more complex than the establishment of Dorsal Place node for AV because (1) labial stops were well established, and (2) [-(-continuant] was specified for sibilants in WF position. Syllabic Intervention Plan As described above, word and syllable shapes were reasonably well developed for this subject. However, there were strong interactions of word position and segment development. Syllabic goals for S2, then, in part reflected word and syllable shape development and, in part, the increase of range of distribution of segments across word positions. In general, the goals were: 1. To establish two consonant slots for a greater range of WI clusters 2. To establish one WI fricative with Place node marking - an interactive syllabic/segmental goal. Onset-Rime Subblock Targets: 1. /si, st/ 2. WI /z/ (a portion of one session) Moraic Subblock Targets: 1. /sp, sw/ 2. WI M (a portion of one session) Rationale 1. /s/-clusters: Syllable slot positions for diconsonantal clusters were evident with the A7-clusters and stop-glide clusters, albeit with epenthesis. However, £2 produced no AY- or /s/-clusters. Because he produced a few sibilants accurately without [+distributed], the /s/ was considered a better candidate than /r/ for syllabic development within the experimental design. Two /s/-clusters were targeted in each subcondition, one with a sonorant and one with a stop. Place of sonorant and stop was held constant within the condition. The /s/-nasal and /sk/-clusters were generalization targets. The former were expected to generalize by place from the appropriate condition, and the latter from the /st/ condition. 2. WI /z/: The /z/ was chosen rather than the /s/ (1) in order to provide a voiced segment contrast with /v/ (a segmental target), and (2) to compare it with the /s/ of the cluster training. By involving it minimally in both conditions, cross-condition comparisons could be made. Results of Block 1 Intervention General Results Changes occurred within both segmental and syllabic conditions. £2 now used some Labial fricatives, although there was only one match for the other segmental target ~ dorsals (an [rj], with low transcription confidence). In terms of the syllabic goals, there was an increase in two-element clusters within the /s/- and /r/-cluster categories, as well as one token of a sibilant fricative prevocalically. (See appendix 4, table 2.1, and figures 4.2 - 4.4). Figure 4.2 shows the average proportional gains for segmental and syllabic targets and the differences between trained and untrained targets in both conditions. Untrained segmental targets include only those sharing critical distinctive features with the target, i.e., all other dorsal and labiodental consonants. Untrained syllabic targets are other /s/-clusters and WI coronal fricatives. If exact matches are considered, the segmental condition was more successful than the syllabic for trained targets in this block, since there were no actual matches for trained syllabic targets. However, if development for the syllabic tier is considered in terms of consonant slot matches (with consonantal substitutions or compensatory lengthening), the syllabic condition appeared to have greater overall influence on the phonological system at this time. (As mentioned above, S2 produced more /r/-clusters with two elements as well, another generalization at the syllabic level.) 86 M 8 PQ 2 c D E E & C •a 60 bo > < SEG trained SEG untrained Dors+Labiodent.** SEG: A l l Dors+Labiodent. SYL CC slots+WI z trained + • SYL #/sC/+ cor fric untrained + + LTJ SYL CC slots+cor fric untrained B SYL: A l l trained +untrained targets XA SYL:A11 tr.+unt. CC slots+WI cor fric •There were no exact matches in terms of segments for syllabic trained targets - - WI /sp/, /st/, /si/, /sw/, /z/ ••Includes WI, SIWW, and WF /f/; WI /fl/, WI /v/; SIWW and WF /k/; WI, SIWW, WF /g/ and WI /gl/ as well as /rj/ + C C slots include actual consonants or compensatory lengthening of the prevocalic consonant + + Includes /sm/, /sn/, /sk/, WI /s/, /;/, /6/, /6/ Figure 4.2. S2: Average proportional gain in matches for trained and untrained targets between Initial Assessment and Block 1 major probes Segmental Goals Lower Level Feature Subblock As stated above, there was only one (low confidence) match in the major probe for dorsals, although S2 had produced tokens of /k/ in single words and short phrases during the treatment sessions. Higher Level Feature Subblock On the other hand, S2 generalized labial fricatives, both across phonemes (from /v/ to Iff) and across word positions. Although only SIWW and WF /v/ had been targeted, S2 produced one 87 token each of WI /v/ and /f/ correctly, as well as three matching tokens of SIWW /v/ (out of eight attempts) and one WF /v/ (out of six). The word scarf also had a WF /f/. Some mismatches showed progressive change in feature specification as well as a split in establishment of the two features - the higher level [+continuant] and the lower level Labial Place. The WI /v/ mismatches were [b] and [bv], indicating that Labial Place was specified, and [+continuant] was partially established. One SIWW/v/ token (of five) appeared as [5], indicating that [+continuant] was specified, although Labial Place was not. Similarly, S_2 produced both tokens of WF/f/ as [0], again indicating specification of [+continuant] without specification of Labial Place. Comparison of the Lower and Higher Level Features According to the match data, S_2 learned the labial fricatives in advance of dorsal stops. The establishment of the Higher Level Root node feature [+continuant] in conjunction with the Labial Place node appeared to be easier for S2 than establishing Dorsal Place for consonants. As was predicted, the double specification for /v/ did not appear to be more complex than the single specification needed for IkJ, since Labial Place and [-fcontinuant] were already established for other phonemes. Other Feature Changes for the Overall Segmental Data Other changes are listed here in terms of the feature hierarchy. Where training may have influenced the change indirectly, comments to that effect are made. The more prevalent changes are noted first, but the order is only relative, because of varying number of tokens across phonemes. Increases in matches are noted with + and * for trained and untrained segments respectively. Progressive feature changes (without matches) on trained segments are denoted with ++, and on untrained segments with **. Regressive changes are denoted with &. A. Unspecified child forms where adult targets are specified 1. Root node: More segments were present, with specification for the liquid in /r/-clusters and all WF /t/ (but see #5 below). The change for /r/ was a probable generalization from syllabic condition training of /s/-clusters. There was one SIWW /r/-cluster without epenthesis (/br/ > [bw] in Barbara-) ** 2. • Laryngeal node: More voiceless initial stops were aspirated (redundant feature [+spread glottis]), with a 17.73% gain in matches for /p/ and a 5.85% gain for /t/. Training in both conditions probably influenced the increase in [+spread glottis] specification, since the segmental training of fk/ involved highlighting of WI voiceless stops Ixj and IkJ and the syllabic condition cluster training involved contrasting of the allophonically different pairs ~ /st/ versus Ixj and /sp/ versus /p/. *, ** 3. Branching structure ('complex') was more common for WF /tf/ (3/4 versus 2/6 tokens in assessment probe). S2 produced one token of WI /dy as [dsw] (in the letter name G_), possibly indicating awareness of the two components of the affricate. ** 4. Place Node: There was beginning generalization, i.e., specification, for WI /z/ (two matches out of four attempts), WI /s/ (one match out of thirteen attempts), and WI /0/ (one match out of three attempts). These advances probably reflect training in both conditions, since [+continuant] obstruents were targets in both conditions. However, training in the syllabic condition involved sibilants more directly. +, *,++ ,** 5. Place node specification for WF Ixj: There was a decrease in proportion of matches and an increase in [?] use. This may reflect an overgeneralization of suppression of coronal default substitutions for /k/ and /v/. & 6. Labial Place for WF Pol unspecified: The three tokens of WF Pol were realized as as [d], [p] and a labiodental stop, regression possibly induced by training on /v/. & 7. Root node: [+continuant] was specified for WI /0/ and WF /6/ ((one match each for three attempts each). These probably reflect training in both conditions for fricatives. * 8. Root node: [+continuant] was unspecified for WI 151, an unexpected regression, considering training of [+continuanfJ. & 9. Laryngeal node: [+voice] was unspecified for the sole exemplar of WF /5/ and one of three tokens of WF IdrJ. ** 89 B. 'Specified child forms where adult targets are unspecified or otherwise specified The one imitative match for WI /r/ (roar'), indicated marginal change — Root node [+consonantal] and Coronal Place [+distributed] specification rather than Labial for one token of A/. * Areas of No Change in Feature Specification 1. Root node: Some tokens of WF fk/, /v/, /z/, and /r/ missing (no Root node). 2. Laryngeal node: [+voice] was unspecified for WF substitutions for /z/. 3. Place node: Coronal [+distributed] interdental place was still observed some of the time for sibilants and coronal stops. Syllabic Goals Syllabic changes were assessed with both major probes and session data. Clusters A. Skeletal slot changes noted in major probe Prior to training S_2 produced no two-element /r/- or /s/-clusters. After Block 1, he produced 10/22 two-element untrained /r/-clusters and 4/24 two-element/s/-clusters, including [hi] for a trained cluster - /si/ — and [s:n] for the untrained cluster — /sn/. (See appendix 4, table 2.1, and figures 4.2 and 4.3.) Although fewer of the trained /s/-clusters had two segments, the prevocalic consonant was lengthened in 14/24 of the /sC/ examples, seven by prolongation of the sonorant or stop and seven by aspiration of unvoiced stops. The seven cases of compensatory lengthening were: smooth > [omud30] = nasal emission preceding nasal; snake (2) > [n:{ei/M}?]; snowing > [n«o"wl~nJ; slipper >[l'ip^u9]; spider > [p-hald£3ol: spy. > [b«al]. Combining matches plus tokens with any kind of lengthening, syllabic changes were noted for 75% of the /s/-clusters. Discounting aspirated stops, which do not represent two skeletal positions, there were identifiable syllabic changes for 50% of the /s/-clusters. The prevocalic lengthening seems to suggest constitutent status for the onset. B. Matches for /s/-clusters in the major probe There were three matches for /sn/ (an untrained cluster) in the words snake and snaksy. (S and SE). As observed across the study, the salience of the "hiss" of the snake (WI [s]) plus the coronality shared by /s/ and /n/ seem to facilitate this production. In all of the elicitations £2 prolonged the [s:] - an expected outcome because of the training methods and the general tendency to lengthen clusters (see assessment data for/l/-clusters). M 8 CQ o c to E 1 E 1 oo <U - C 1 .S 53 t o # 50.00% ~zj 40.00% 30.00% -20.00% 10.00% -0.00% -= -10.00% -_ -20.00% -= -30.00% -40.00% --50.00% -60.00% --70.00% -80.00% -90.00% -100.00% Seg. targ. Syl targ Observation targets • SIWW/ v/ H /sC/ > C, C L * * B WI/v/ 0 wi/t/ • WF/v/ • /sCC/ > CC E2 wi/p/ • WF/t/ H /scy > cc* • /Cr/ > CC * E3 WF/b/ rjwF/5/ Note: There was one matching token each for untargeted WI /f/, /s/, and /r/, and SF /rj/ and /z/. There were no other matches, including for segmental condition target /k/ *sC = WI /sAcluster; /Cr/ = WI /r/-cluster * * C L = 14 cases of lengthening of C i in the child's form (including 7 aspirated stops) Figure 4.3. S2: Proportional gain in matches for trained and untrained targets between Initial Assessment and Block 1 major probes C. Nonmatch changes other than compensatory lengthening for clusters 1. Triconsonantal /s/-clusters: Prior to Block 1, £2 produced these clusters with only one consonant, but the Block 1 sample revealed three of seven tokens with two elements, one each with /skr/, /spr/ and /str/. 2. £2 produced the /sl/-cluster once as [hi], following the pattern for /s/ production. 3. The two-element /r/-clusters were produced with full epenthetic vowels. WI Sibilants Although there were no exact matches for WI /z/, two of the four tokens were close substitutions - [05] and [z>] - with Place node marking. S2 generalized Place node specification for WI fricatives, producing one token each of /s/, AV, A7, and /v/. Segmental training probably influenced the emergence of the latter two as well. Onset-Rime Subblock Since there was no change in matches at the major probe level, the session data was utilized in order to evaluate change within target groups for subblock conditions. (See appendix 4, table 2.3, and figure 4.4.) 1. /st/ versus /sV: The /st/-cluster results were slightly higher than those for /si/, both in imitated and spontaneous productions during training sessions — 75% matches versus 65.82% and 76.92% versus 55.56%. The interest generated by games involving "stop" may have bolstered/st/ matches, although the "slot" games also had entertainment value. Feature-sharing for /s/ and A/ is greater than for /s/ and AV, which may aslso have contributed to the slight advantage for /st/. 2. WI/z/: No session productions ofWI/z/were recorded. Auditory awareness of the/z/ with the favored song Zip-a-dee-doo-dah and some home practice were the "treatments" provided. Moraic Subblock 1. /sw/ and/sp/: £2 had greater success with both of these targets in imitated productions -- 94.12% and 86.36% respectively — than in spontaneous productions, where the success ranking was reversed ~ 54.17% and 64.29%. 100.00% 80.00% 60.00% — 40.00% — 20.00% - | 0.00% — I I I I I I Onset-Rime targets 'A Moraic targets • TOT: OR: /st/* H TOT: MOR: /z/ • TOT: MOR: /sC/ M TOT: OR: /si/ • TOT: MOR: /sp/ • TOT: MORAIC JH TOT: OR ED TOT: MOR: /sw/ *A11 targets word-initial Figure 4.4. S2: Proportional matches for syllabic targets in Block 1 sessions Moraic Subblock (continued) 2. WI IzJ: During the last session of the subblock, a portion of time was devoted to WI /z/ production of the form CV+C. S2 found both imitation and spontaneous whole-word production easy, as the 90.91% and 100% proportions indicate. He enjoyed saying zoom, which may have enhanced that ease. Comparison of the Onset-Rime and Moraic Conditions Overall, the session training results were equivalent across conditions (73.73% for moraic targets and 71.35% for the onset-rime targets). It was not necessary to present isolated syllabic constituents for imitation most of the time, since S2 imitated whole words easily. Without consitutent stimuli distinctions, a portion of the differences between conditions was lost for cluster training. However, the treatments did differ in that the onset-rime sessions focused on activities which contrasted the onsets, whereas the moraic sessions focused on the /s/ as external to the CV(C) unit(s). Intervention Plan: Block 2 Segmental Intervention Plan Higher Level Feature Subblock Target: /kl -Dorsal Place (extended contexts - WI, SIWW, and /kl/ WI clusters) Lower Level Feature Subblock Target: WF /J/ - Coronal Place node [-anterior] (including minimal auditory and phonetic stimulation for the affricates) Rationale Because generalization of dorsal consonants had not occurred, they were retargeted. The training contexts were expanded to include SIAVW position and WI /kV-clusters in order to strengthen the generalization potential. In the Block 1 major sample, S2 produced both SIWW /k/ and /kl/ as [t] (for all /kl/-clusters, and some of the time for SIWW IkJ). Consequendy, only Dorsal Place node had to be specified. Since dorsals had been the first targets in Block 1, they were trained first again in Block 2, with the expectation that /v/ or If I training would follow. However, at the end of the dorsal consonant subblock, it was evident that the labiodentals were sufficiently established except in WI position. In this way, the labiodentals were then similar to the coronal fricatives, and as such became syllabic rather than segmental targets. Segmental goals could have included /r/ or the alveopalatals at this point. The training of /r/ would have permitted training of a higher level set of features ([+consonantal]/[+sonorant]) in conjunction with a lower feature ([+distributed]), similar to the training of /v/ in the first block. However, the /r/ was less stimulable phonetically than the /J/. For £2, who had a mild attention deficit, it was necessary to have attainable session goals in order to maintain his interest. Thus the alveopalatals were chosen in spite of the fact that the [-anterior] feature of the Coronal Place node could be considered "lower" than the Dorsal Place node. (Working with human subjects requires that practical and ethical considerations override experimental design from time to time.) The WF /J/ was targeted because WF position was established for sibilants. Syllabic Intervention Plan Onset-Rime Subblock Targets: 1. WI/f/, /fl/ - 2 sessions 2. WI /s/ before lax vowels (sVC) -1 1/4 sessions 3. /st/, /si/ - 3/4 session with homework after first WI /s/ session for /st/ Moraic Subblock Targets: 1. WI /z/, /v/ -1/2 session each 2. WI /s/ before long vowels or diphthongs (sV V[CJ) -1 1/4 sessions 3. /sp/, /sw/ -1/2 session each Rationale A. Goal expansion As noted above, S2 began to generalize labiodentals at the time of the first major probe, continuing to do so during the first part of the Block 2 segmental training period. Word initially, they continued to be realized like the other fricatives (with an [h] substitution), and thus were transferred to the syllabic condition. The /f/ was selected in addition to /v/ because of its greater frequency in English. In order to maintain equivalence across subconditions, WI /f/ and the /fl/ cluster were trained in the onset-rime condition, and WI /z/ and /v/ in the moraic. Although the /z/ may have had some advantage from being a target under both conditions previously, less than half a session per condition had been the time allotment in Block 1, insufficient time to result in learning. In addition, WI /s/ was also targeted in this block. Because it was a target in both subconditions, different vowel contexts were chosen to provide a distinction in word sets (lax for O-R, in which VC rime units were obligatory, versus tense vowels and diphthongs for moraic stimuli, in which CVV units needed to stand alone). Increasing the number of training targets for establishment of WI Place node for fricatives maximized generalization potential. B. /s/-clusters Because /s/-clusters had not generalized at the end of Block 1, they were retargeted in their respective subconditions. Results of Block 2 Intervention Syllabic goals were met during this period. There was some restricted progress with segmental goals of dorsal consonant and /J/ production (see appendix 4, table 2.2, and figures 4.5 - 4.8). The generalization of training session proficiency to conversation occurred within a two-week period at the end of the block, following the three "events" listed below: 1. E informed £2 that he made E happy when he used his new way of talking and when he did not act "silly". 2. £2's grandfather had a major heart attack, which brought a serious mood to the home environment. 3. £2 took swimming lessons, during which time he had many opportunities to use the /sw/ cluster and the /sp/ cluster (Speedo. the brand name on his swimsuit). Segmental Goals Lower Level Feature Subblock As noted above, there was advancement in conversational use of dorsal consonants, primarily for the trained segment IkJ in WI and SIWW positions (12/31 [38.71%] and 3/11 [27.27%]). S2 produced one matching token each for WI and SIWW Igl and one WF [go] (all echoic), as well as two tokens of [rj]. Higher Level Feature Subblock The major probe had one matching token of an alveopalatal: WI [f] in sheep (E), where the III was in an untargeted word position. The other four WI /J/ productions were [s] (three) and [sp] (sheep - S). In other words, the syllabic goal to establish Place node for sibilants was attained, but the segmental goal to establish [-anterior] was still to be attained. The overgeneralization of [sp] for the /J/ in sheep in an assimilatory context was an unexpected and interesting mismatch. It indicated a phonological awareness of the need for both Place node marking and a sibilant prevocalically, but uncertainty with respect to the actual specified values for that sibilant. In nonlinear terms, the lack of definite place specification prevocalically led to Labial Place spreading from the postvocalic consonant to the prevocalic position. Rather than producing [f], however, which could have been the result of the specified [-(-continuant] value coalescing with Labial Place, S2 produced a cluster with the feature [-(-continuant] and Labial Place realized on separate segments. The need to produce a coronal sibilant and a specified place was stronger than the need to produce a syllabic match. This was in tune with the epenthesis noted for liquid clusters, where the syllabic timing was sacrificed for segmental realization. ca 2 s 1 Vi a 2 03 C •a t o • WI/k/ gOR:/ f l / (noep) @ MOR: WI M g] MOR: /sp/ H SIWW/k/ • WI/s/* • OR: /st/** 0 MOR: /sw/ • OR: WI /{/ • MOR: WI M EE OR: /si/ E /sC/ > CC Note. There were no matches for trained targets /kl/ or WF ///. *Four /s/ words were realized with dentalized [S] and counted as matches. **A11 /s/-clusters were word-initial. Figure 4.5. S2: Proportional gain in matches for trained targets between Block 1 and Block 2 major probes Another aspect of the [sp] substitution concerns its unity as a constituent. On the one hand, one could surmise that the sharing of features across two segments indicates that the two are a constituent with an onset boundary. The moraic condition cluster [sp], although never introduced as a constituent in therapy, was possibly identified as such. However, the separate segment production could also be a result of the moraic condition methods, which deliberately encouraged the "addition" of [s] to a /p/ or /w/ word. Without specification of the prevocalic consonant beyond the recognition that a sibilant was required, spreading to the prevocalic position led to a [p]. To satisfy the sibilant requirement, an [s] may have been adjoined at the output stage. s 2 m B o ca i c • — c • — ca bo # 100.00% 80.00% 1 Trained segments in syllabic condition: Other word positions Trained segment in Seg. Condition W F position srww/f/ WF/f/ SIWW M H WF M • WF/z/ HD/sn/ • WI III • ICxI > CC O WI/p/ EJ wi/t/ • WF/t/ R wi /o/ Note: In addition, there was one match each for WI and SIWW /g/ and WF lb/, and two for /q/. Figure 4.6. S2: Proportional gain in matches for untrained targets between Block 1 and Block 2 major probes Comparison of the Lower and Higher Level Features The dorsals developed in advance of the alveopalatals, which reflects the longer training period for the former. However, the fact that even one token of III appeared after one subblock of training demonstrates the relative difficulty of establishing Dorsal Place for this child, in comparison with the establishment of the lower level feature [-anterior]. 90.00% -n JS E • WI/s/(/S/)* H M : WI M • /sn/ @ WI/p/ E3 WF/ty H /sC/ > CC M SIWW M EH /Cr/ > CC • WI/t/ *For /s/, 4/14 matches were dentalized variants. Figure 4.7. S2: Proportional matches in Block 2 major probe for targets with match proportions differing from those shown in figures 4.5 or 4.6 Other Feature Changes for the Overall Segmental Data Changes other than those noted above are listed here in terms of the feature hierarchy. Where training may have influenced the change indirectly, comments to that effect are made. The more prevalent changes are noted first, but the order is only relative, because of varying number of tokens across phonemes. Increases in matches are noted with + and * for trained and untrained segments respectively. Progressive feature changes (without matches) on trained segments are denoted with ++, and on untrained segments with **. Regressive changes are denoted with &. CN s 2 CQ S .5 CQ 00 # bb < 100.00% 80.00% 60.00% 40.00% 20.00% 0.00% ISD-25% J , § _ -LW-19.89% "I ^ I |SD-6.04% I ^ [SZM6.13% "IT Segmental cond. avg. Syllabic condition averages • SEG: trained EI SYL: /sC/ untrained H SY: A l l WI cor. fric. SEG: untrain.:Dors, [-ant] HD A l l WI /sC/ • SYL: A l l trained 0 SEG: A l l Dors & [-ant] • SYL: WI /f,fl,z,v,s/ ^ SYL: A l l untrained ^ SYL: /sC/ trained 0 SY: WI cor. fric. untrain. ES SYL: A l l Figure 4.8. S2: Average proportional gain in matches for trained and untrained targets between Block 1 and Block 2 major probes Feature changes (continued) A. Unspecified child forms where adult forms are specified 1. Place node: As a result of syllabic condition training, Place node was specified for WI M, /v/, and HI, and an increased number of /s/ tokens (55.56%) +,++ 2. Root node: Except for a few tokens of /r/ (WF singletons and WF and WI clusters) and 4/5 Is/ tokens in triconsonantal WI /s/-clusters, all consonants were specified with Root nodes. ** 3. Laryngeal node: Fewer voiceless initial stops were aspirated (redundant feature of [+spread glottis] less in use), with a 16.43% regression in matches for /p/ and a 4.44% regression for III. This regression may be spurious, as the increase may have been for Block 1, or may reflect an overgeneralization from the proficiency gained in /s/-cluster production, where unaspirated stops are required. & 4. Root node: [+continuant] was specified for all fricatives except two tokens of /G/ 100 (one WI and one SFWW, missing Root node), one WI /5/, and one WF /v/ (of > [Ad] ) . One SIWW HI and /v/ each had labiodental place but a [-continuant] (stop) value. ** 5. Place node specification for WF /{/: Match proportions increased and number of [?] substitutions decreased (9/44 total tokens compared with 17/49 in Block 1). This change is similar to the general decline in missing Root nodes. *,** 6. Labial Place for /b/ specified: The temporary regression noted for /b/ disappeared. * 7. Place node: A marginal change for WF /kf was Place node specification for one token (as [t], a change from only [?] substitutions). ** 8. Branching structure ('complex') was in place for SIWW /tf/ once. B. 'Specified child forms where adult targets are unspecified or otherwise specified 1. Place node: Coronal [+distributed] interdental place was observed less of the time for sibilants. This directly reflects the training with sibilants, although it was not an explicit goal. + 2. Labial Place for Coronal Place was observed for the majority of WF /9/ tokens, once each for WI and SFWW /6/. & 3. Coronal Place [+distributed] rather than Labial was specified for one token of /r/ in SIWW /br/ (similar to Block 1 results but for a different word position). ** Areas of No Change in Feature Specification 1. Root node: [+consonantal]/[+sonorant] remained unspecified for/r/. 2. Laryngeal node: [+voice] remained unspecified for WF sibilants/affricates. Syllabic Goals Clusters: Skeletal Slot Changes Observed in Major Probe After Block 2 training, £2 used all /s/-clusters in conversation except /sk/ and /sm/, for which the single token each showed a missing /s/. The most successful clusters were trained targets /sp/, /sw/, and /si/ (all 100%) with untrained target /sn/ at 80% (similar to the score at the end of Block 1) and trained target /st/ at 50%. This represented two-element /s/-cluster production for 85% of the total, and a proportional gain of 68.33%. 101 The eight triconsonantal clusters were realized with two consonants, and there was one with three. S2 also increased his production of /r/-clusters with two elements to a total of 68.18%, representing a 22.73% gain. In other words, there was an observable difference between trained and untrained targets, but an overall increase in syllable structure realization (see appendix 4, table 2.2, and figures 4.5 - 4.8). WI Fricatives All attempts at WI /z/, /v/ and /f/ were matches, and 55.56% of the WI /s/. Both the previous training with WI /z/ and the /v/ in Block 1, plus the very few word types to remember in English for /z/ and /v/ probably.contributed to the higher success rate for the other fricatives over /s/. It is also possible that the /s/, as the least specified fricative, is less accessible than the others because of lack of specification. The other interdental fricatives showed less change word-initially, particularly the /0/, for which three of the four nonmatches continued to be [h] substitutions. Onset-Rime Subblock Block 1 session data was utilized to evaluate change within target groups more exactly (see appendix 4, table 2.4, and figure 4.9). 1. /st/ versus /si/: Contrary to the outcome of Block 1, there was a marginal facility noted for /si/ over /st/, although only in imitated and not in spontaneous productions (both 68.18%) during training sessions — 77.78% matches versus 69.73% with a total of 72.5% versus 68.67%. The overall scores represented a decline in session proficiency from Block 1 for /st/ (down from 75.77%) and a gain for/si/ from 62.5%. On the major probe £2 pronounced the two /si/ words with [si], whereas two of the four /st/ words lacked /s/. Whether the difference in tokens is significant, or whether the sharing of features of /s/ and IxJ turned out to be a disadvantage for faster generalization (as it was for £1, £3_ and £5) cannot be determined. 2. WI HI and /fl/ versus WI /s-VC/: Session data demonstrated an equivalence for total productions of HI and /s/ (73.38% versus 72%). However, spontaneous productions of /s/ were at a lower level of proficiency - 56.25% versus 75.86% for/f/. The session data for spontaneous /s/ production was equivalent to the major probe match data of 55.56%, but the /f/ data in the major probe showed complete mastery at 100%. The /fl/ data was better overall than /f/ and /s/ data with an 83.33% total. Contributing to this was the high level of imitative matches at 95.65%. On the major probe £2's productions matched two of the three /fl/ targets, without epenthesis. 100.00% 80.00% 60.00% 40.00% 20.00% 0.00% I I I Onset-Rime Targets Moraic Targets • OR: III • OR: /si/ • M: /z/ E3 M: /sp/+/sw/ M OR: /ny IED OR: /st/+/sl/ • M: M E3 TOT: MOR M OR: /sV(C)/ • TOT: OR E2 M: /sp/ • OR: /st/ • M: /sVV(C)/ El M: /sw/ Figure 4.9. S2: Proportional matches for syllabic targets in Block 2 sessions Moraic Subblock 1. /sw/ and /sp/: As in Block 1, there was greater success both of these targets in imitated productions — 92% and 90.48% respectively - than in spontaneous productions, where the success was greater than in Block 1 and consistently better for /sw/ this time at 61.11 % and 71.43%. 2. WI Izl, M and /sV V(C): S2 was 100% successful with the previously introduced WI Izf in both the sessions, and the major probe. Success with the WI /v/, a segmental target in WF 103 and SIWW positions in Block 1, was slightly lower, but still higher than clusters, showing an 85.24% total match level, with 95.45% matches in imitation and 75% spontaneously. WI /s/ before long vowels or diphthongs matched less consistently than the voiced fricatives, but at a slightly higher level than the /s-VC/ from the O-R condition (possibly showing a sequence effect), with a total of 74.47% matches, averaging from 86% matches in imitation and 60% in spontaneous productions. Comparison of the Onset-Rime and Moraic Conditions Overall, the session training results showed a slight advantage for moraic targets in this block (82.61% for moraic targets versus 73.98% for the onset-rime targets). (See appendix 4, table 2.4, and figure 4.7). There was little difference between the overall block results for O-R targets (71.35% in Block 1), although there was a gain for the moraic targets (73.98% in Block 1). This probably reflected the ease noted for voiced WI fricatives rather than the methodology. In the major probe, there was no difference between WI /s/ targets before lax versus tense or long vowels and diphthongs, the criterion of difference across conditions. Each group showed five matches and four nonmatches. As noted above, the session data seemed to favor the tokens in the moraic condition, but this may have been a sequence effect. In any event, it did not carry over to the major probe sample. In terms of clusters, S2 mastered the /sp/ and /sw/ moraic targets by the time of the major probe, whereas the O-R /st/ target was still in flux. This may reflect (1) training differences, or (2) difficulty with two relatively underspecified adjacent coronal consonants, or (3) the fact that motivation to pronounce swim and Speedo was high at the time of the major probe sample, the use of these promoting more rapid generalization for /sp/ and /sw/. During this syllabic subblock, S2 began to engage in sound play which consisted of replacing the prevocalic consonant (the therapy target) with labial consonants, both during onset-rime and moraic sessions. This suggested he was segmenting words according to onset and rime constituents. For example, in session 31 (an O-R session) a substitution for £ljj2 was [blp], for funny was [bAni] and for stupid was [bupld]. In session 32, he reconstituted piano as [vijaenou] and [wijsenou], and sandwich as [baenwltS]. In rehearsing an O-R series of tan-fan-stand-sand 104 during session 35, he added his own set of [baen] and [spaen]. The latter substitution recalls [spip] for sheep discussed earlier, and might suggest that the [sp] was a unitary onset consituent, rather than an assimilated [p] with extrametrical [s] adjunction. Block 3 and Beyond A final cycle was necessary to stabilize dorsal stop production in conversation. Alveopalatals were also retargeted due to some inconsistency in conversation. In order to reduce S2's frontal lisp, a contrast program with /z/ and /5/ was conducted. One year post-therapy, £2 used alveopalatals inconsistently (substitutions were [+anterior]), and had not yet developed /r/ (at least WF /r/), according to mother's report. Discussion In terms of the research questions the following interpretations are offered. 1. Will nonlinear phonological frameworks help to predict logical and attainable intervention goals for phonologically disordered children? As with the other subjects, the nonlinear frameworks did lead to reasonable targets for intervention. The intervention for £2 may have been successful for reasons other than the choice of targets, such as the intensive stimulation and the mutually enhancing effect of a variety of approaches. However, the nonlinear frameworks did lead to attainable goals. Conversational Intelligibility Gains There was improvement in intelligibility due to the attainment of conversational proficiency for dorsal consonants, fricatives and clusters. (See figures 4.10 - 4.12.) 105 80.00% -3 Assess Block 1 Block 2 — WI/k/ «»srww/k/ «** SIWW /v/ WF/v/ 1 1 1 WI /J/ Note: /v/: Block 1 only. /J"/: Block 2 but in WF position. There were no matches for /kl/. Figure 4.10. S2: Developmental progression in proportional matches for trained segmental targets across the intervention period That the system reflected phonological change as a result of training can be inferred from (1) the changes in trained targets detailed above, (2) the types of generalizations which occurred, and (3) lacks in development for observation targets. £2 generalized within voicing cognate pairs, such as from WF /v/ to WI HI or WI /z/ to WI /s/ in Block 1. He also generalized between word positions for fricatives, e.g., from trained SIWW and WF /v/ to WI /v/ and HI, from WI HI to WF HI, from WI IzJ to WF /z/, from WF /J7 to WI /J/. Within sound class generalization occurred, from Place node specification for /z/ to the same for /9/ and from Dorsal Place stop Ikl to Dorsal Place nasal /rj/. Between sound class generalization occurred for the assignment of cluster skeletal slots, such as from /sC/-clusters to /r/-clusters. Lack of development for untrained targets such as HI, affricates and liquid cluster epenthesis also demonstrates that the training had an impact on the changes. Those phonemes not completely mastered at the end of the phonological intervention 106 project were still not in use one year following the project, suggesting that the therapy was responsible for change. Assess Block 1 Block 2 « " WI/z/ WI /si (JSI) <»* /si/ •• /sw/ »**/fl/ »'» WI hi " | /st/ v /sp/ — WI/f/ Note. Only the clusters and WI /z/ were Block 1 syllabic targets. See /v/ in figure 4.8. Figure 4.11. S2: Developmental progression in proportional matches for trained syllabic targets across the intervention period Root node ([+cons]J • • • -O [+son] [+nas] [+cont]* Laryngeal node O ' [voice] ([+spread glottis]**) Labial node [+round] (Coronal O . (Dorsal ~ 0 node)& : ~-_ node)&& _-~ E ([-ant]) ([+distributed]) ([bk]) [hi] [lo] Note: Single parentheses and dotted lines indicate partially established nodes or features. Italics indicate nonmatching feature. *'Complex' branching structure in place for SIWW and WF affricates **[+spread glottis] partially in use for WI stops &Coronal node partially established for [-anterior] and [+distributed] &&Partially established for consonants Figure 4.12. S2: Feature geometry at end of Block 2 107 Language Test Results All comprehension language test results at the end of the project showed small positive changes in performance - scores were still within normal limits but at a slightiy higher level. Table 4.3.—S2: Language test scores at Initial Assessment and end of Block 3 TEST PRETEST SCORE POSTTEST SCORE Peabody Picture Vocabulary Test (Form L) 5.5.: 100 5.5.: 102 Test of Audit. Compreh. of Lang. (Rev.) 5.5.: z = -.33 5.5.: z =+.20 Test of Lang. Dev.-(Primary): Audit. Discrim.Subtest Raw score: 18/20 (normal limits) Raw score: 20/20 In terms of language production, observation of language sample data and the Structured Photographic Expressive Language Test - Preschool (SPELT-P) revealed gains in sibilant morphophonology, with continuing pronominal case errors and inconsistent omission of BE. Language test results were as expected — small gains in vocabulary, phonemic discrimination and sibilant morphophonology alongside minimal gains in syntax — and as such lend support to the claim that the changes in phonology were a direct result of intervention. 2. Are the separate prosodic and segmental levels of representation of nonlinear phonology psychologically real ? At the end of Block 1, there were changes in both segmental inventory and syllable structure. Changes in syllable structure were proportionately greater and more pervasive across the total system. The segmental inventory changes were limited to the labiodental fricative category, most for SIWW /v/ (3/8 = 37.5%) and including one token each for WF and WI /v/ and WI /f/. Dorsal consonants remained unchanged, except for one low confidence transcription of /rj/. Syllable structure change showed 75% of the /sC/-clusters with either two elements (16.67%) or compensatory onset lengthening (58.33% , o