Exigwflxfixuxh an.» a .. n t , ‘ . ..u.wfiH..:..e.no..: . _ P.3d ; , , run .3 A. . ; . . .. , Ban. ., 1 . . V , . l. , . ‘ . , . . \7 mpg . , . (.221?) . . x In». n .58 15 .u 1 2|. 2.7.. Imt y; , f I”: { v . 1 H7 .29.;QII.’ , . , Q'Ina‘..: ‘ , 9!) V 1. 0 ‘ ., :33: 2...? m y . H53”. ‘ia.’ n .l Ih. ' ‘ '4 N1 7331! 0:... . . . , \ . 34!, .8... Va!) _ x? . .3561 {3&5 a... .t W57: L. a!“ , . .. 393‘ 45?, ifiwwm L .3. . 512%" .. a... #21:... iiifiwnmm 7;»... 9;. Jails-L . r. I 2... . V: my Ana , . , . , , . if 3. Hawaii” Ian] . Mkfiw , , $21M. .“ 31.. . .4... .n...~ 3. .- This is to certify that the dissertation entitled Perceptions of African American English Dialect Density by Anglo-European American Speech-Language Pathologists presented by Gregory C. Robinson has been accepted towards fulfillment of the requirements for the PhD degree in Audiology and Speech Pathology DD, 9 DDDDD DD D] D Majdr Professor’s Signat’ure 96/6 / Date MSU is an Affirmative Action/Equal Opportunity Institution LIBRARY Michigan State University -'_______'-_-.-.-.-.-.-.-.-_-._,-D- PLACE IN RETURN BOX to remove this checkout from your record. To AVOID FINES return on or before date due. MAY BE RECALLED with earlier due date if requested. DATE DUE DATE DUE DATE DUE 2/05 p:ICIRC/DateDue.indd-p.1 PERCEPTIONS OF AFRICAN AMERICAN ENGLISH DIALECT DENSITY BY ANGLO-EUROPEAN AMERICAN SPEECH-LANGUAGE PATHOLOGISTS By Gregory C. Robinson A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Audiology and Speech Sciences 2006 ABSTRACT PERCEPTIONS OF AFRICAN AMERICAN ENGLISH DIALECT DENSITY BY ANGLO-EUROPEAN AMERICAN SPEECH-LANGUAGE PATHOLOGISTS By Gregory C. Robinson Language researchers have used the terms dialect density and dialect rate to describe the varying amounts and types of features used by speakers of a given dialect (Oetting & McDonald. 2002). Members of the general population use other terms. such as “thick dialect" or “slight accent.“ These colloquial terms often coincide with subjective statements about how easy the speaker was to understand and/0r how noticeable the use of the dialect was. Although dialects consist of features across all levels of language (phonology. morphosyntax. semantics. and pragmatics). phonological features may be the most identifiable (Wolfram & Schilling-Estes. 1998). Little is known concerning the actual role that these features play in perception of inter-speaker differences. It is important to understand the perceptibility of dialect density. because speakers of some dialects are penalized for the use oftheir dialect in some situations. For example. a speech-language pathologist (SLP) may mistake an African American English (AAE) dialect for a speech-language disorder. This may be more likely to happen with speakers who use a greater number or particular types of AAE phonological features. This study aims to determine the extent to which selected amounts and types ofAAE features contribute to subjective judgments by non-AAE-speaking speech-language pathologists in predominantly Anglo-European American school districts. The subjective judgments under investigation are perceived degree of comprehensibility (i.e.. how easily understandable the speaker would be to the general population) and perceived dialect detectability (i.e.. how noticeable the use of AAE was) as perceptual indicators of dialect density. Certified speech-language pathologists from predominantly Anglo-European American school districts in rural mid-Michigan listened to individual sentences that contained varying amounts and types of AAE phonological features. The SLPs rated each sentence on five-point ordinal scales regarding how noticeable the dialect was (1: AAE dialect was not noticeable. 5: AAE dialect was extremely noticeable) and how understandable the speaker would be to people in the general population (1: Very difficult to understand, 5: Very easy to understand). The ratings for the different sentences were compared to determine the extent to which the frequency and types of features represented in the sentences contributed to the SLPs’ judgments. The study found that both comprehensibility and dialect detectability ratings were affected by the number of features included in the sentences. The type of features included had consistent effects in the comprehensibility ratings but less consistent effects in the dialect detectability ratings. Possible explanations for these inconsistencies were discussed. Theoretical implications for prototype-based models of speech perception were addressed along with practical implications for speech-language pathologists in the areas of: (a) assessments with nonmainstream dialect speakers. and (b) accent modification. AC KNOWLEDGMENTS This dissertation would not have been possible without the invaluable input and support from several people. First. I would like to thank my advisor, Dr. Ida Stockman. She constantly supported. motivated, and inspired me during this great undertaking. She has changed my life by imparting her knowledge and determination during this process. Next. I would like to thank Dr. Dennis Preston, who helped not only to give me excellent advice through the development of this idea but also shared his unfailing joy of research. language diversity. and social activism. I also thank Dr. Peter Lapine and Dr. Brad Rakerd for their intellectual contributions and support throughout the process. The logistics of the study would not have been possible without the support of Cheryl Granzo. Director of Therapy in the Ionia Intermediate School District who helped me contact study participants. Dr. Laura Smith-Olinde shared her knowledge of statistics to assist me with data analysis. I also thank Johanna and Derrick Boult, Mary Jo Hidecker. Robin Pulford, Diane Ogiela. and Rana Alkhamra for their input and support during my entire doctoral program. Last, but certainly not least. I want to thank my family. My daughter. Madeleine. kept me on track by making me little sticky notes of encouragement reading. "Get it dun (sic)” and “work, work. work. disertashen (sic).” My wife, Monica. was always there to encourage me, push me to work harder, and force me to take breaks. She is my soul-mate and my best friend. I could not have done this without her. I also want to thank my parents, Lewis and Carolyn Robinson. who were behind me 11 %. and my parents in- law, Alan and Phyllis C lark, for their support throughout this process. TABLE OF CONTENTS LIST OF TABLES ................................................................... LIST OF FIGURES .................................................................. CHAPTER I: INTRODUCTION ................................................ A. Temrinology ............................................................ B. Two Perspective on Normal Language Variation: Folk vs. Linguistic 1. From Categorical to Gradient View of Dialect Features. 2. Attitudes Toward Non-mainstream Dialects ................ C. Dialect Detectability and C omprehensibility ................ 1. Dialect Detectability ................................... 2. Comprehensibility ..................................... a. Phonemic Difference ................................ b. Listener Familiarity .................................. D. Statement of the Problem .............................................. 1. SLPs as Participants ........................................... a. SLPs as Professionals with Power .................. b. Professional Mandates Pertaining to Linguistic Diversity ................................. c. Antiquated Paradigms Concerning Language and Culture ........................................... d. SLPs and Cross-Dialectal Assessments ............. e. SLPs and Dialect C ornprehensibility Judgments... f. SLPS and Dialect Detectability Judgments ...... 2. AAE as a Language Variety ......................................... a. Description of AAE ....................................... xi xiii ll 14 14 16 19 b. The Misdiagnosis of AAE as Disordered Language... c. AAE Phonology ............................................ 3. F requeney and Type of Features as Variables ............... a. Frequency of Features Included ............................. b. Types of Features Included ............................... E. Research Questions ....................................................... CHAPTER 11: METHOD ............................................................ A. Participants ................................................................ 1. Characteristics .................................................... 2. Recruitment ...................................................... 3. Human Subjects Approval ...................................... B. General Description of the Stimuli .......................................... C. Construction of the Stimuli ................................................. 1. Number of Features Included ..................................... IQ . Type ofDialect Features Included ............................... a. Phonological Patterns ..................................... b. Perceptual Salience Groupings ......................... DJ . Description of the Sentences ................................... 4‘s. . Preparation of the Stimulus Recordings ......................... 3. Speakers” Characteristics ................................. b. Speakers’ Conditions ...................................... c. Construction of Naturalness Rating Stimulus C Ds.. (It . Verifying Naturalness of Stimulus Sentences ................. vi 40 41 44 44 45 49 51 51 51 56 57 57 58 a. Naturalness Rater Characteristics ......................... 68 b. Naturalness Rating Task .................................. 68 c. Naturalness Rating Results ................................ 69 6. Randomization and C ounterbalancing of Stimulus Sentences. 70 a. Speaker Order ................................................ 71 b. Task Order .................................................... 72 c. Sentence Randomization ................................... 72 D. Collection ofData ............................................................ 73 1. Presentation ofStimuli ......................................... 73 2. Rating Tasks ......................................................... 74 a. Task Orientation and Familiarization ..................... 74 b. C omprehensibility ........................................... 76 c. Dialect Detectability ........................................ 79 E. Intra-rater Reliability ......................................................... 80 F. Data Processing ............................................................. 80 G. Data Analysis ................................................................ 8O 1. Compilation Comparisons ......................................... 80 2. Task-Specific Comparison of Test Variables .................. 81 CHAPTER III: RESULTS ................................................................ 83 A. Intra-rater Reliability .......................................................... 84 B. Compilation Comparisons ............................................... 85 1. Compilation Comparisons Over Both Tasks ....................... 86 2. Comparisons Specific to C omprehensibility ................... 87 vii 3. C omparisons Specific to Dialect Detectability ................... 88 4. Compilation Groupings Resulting from Compilation Differences ............................................................ 90 C. Principal Findings: Dialect Detectability Ratings ...................... 91 1. Description of Factors ............................................. 91 2. High Detectability Group Results ............................... 93 3. Frequency Effects .......................................... 93 b. Salience Effects ........................................... 95 c. Frequency - Salience Interaction Effects .............. 95 3. Low Detectability Group Results ............................... 95 a. Frequency Effects ........................................ 95 b. Salience Effects ........................................... 100 c. Frequency by Salience Interaction .................... 100 4. Post Hoc Comparison of Features for Detectability Ratings ...... 101 a. Ranking of Features ...................................... 101 b. Perceptual Salience Classification and Detectability. 103 5. Summary of Dialect Detectability Results ..................... 104 D. Principal Findings: C omprehensibility Ratings ........................ 105 1. Frequency Effects .................................................. 105 2. Salience Effects ..................................................... 107 3. Frequency by Salience Interaction ............................ 108 4. Summary of C omprehensibility Results .......................... 108 E. Written Comments from Listeners ....................................... 109 CHAPTER IV: DISCUSSION ........................................................... llO viii A. Gradient Perception of AAE Dialect by SLPs ........................... 110 1. Frequency of Dialect Features ..................................... 110 2. Perceptual Salience of Dialect Features ........................... 111 a. Comprehensibility Judgments ............................. 1 11 b. Dialect Detectability Judgments .......................... 1 14 c. Summary of Perceptual Salience Contribution ......... 117 B. Methodological Variables .................................................. 120 1. Differences Among Compilations .................................. 120 2. Differences Between Speakers ..................................... 123 C. Theoretical and Research Implications ................................... 123 1. Stevens Law and General Sensory Perception .................. 123 2. Johannson’s Factors for Non-native Accent Feature Perceptibility .................................................... 124 3. Steriade’s P-Map ................................................. 125 4. Prototype-Based Models of Speech Perception ................. 127 5. Implications for Research Methodology ............................ 128 D. Practical Implications ......................................................... 129 1. Implications for SLPs in the Public Schools ..................... 130 2. Implications for Instructors of SLPs ............................. 132 3. Implications for SLPs Engaging in Accent Modification Therapy ....................................................... 132 E. Study Limitations and Future Research .................................... I33 1. Listener Variations ................................................... 133 2. Stimulus Variations ................................................... 133 3. Task Variations ....................................................... 134 4. Speaker Variations .................................................... 135 F. Conclusions ...................................................................... 136 APPENDICES ........................................................................... 13 7 REFERENCES ............................................................................ 157 LIST OF TABLES 1a. Listing of Stimulus Sentences and Targets for Planned Alternations 54 1b. Listing of Stimulus Sentences and Targets for Planned Alternations Phonetically Transcribed ................................................. 55 2. Characteristics of 6 groups of sentences presented during comprehensibility ratings ............................................... 56 3. AAE Phonological Patterns Included in the Stimuli .................... 58 4. Ranking of Phonological Patterns for Perceptual Salience in Nonsymbolic Units ........................................................ 6O 5. The Recording Source of Each of the Stimulus Sentence Constituents 66 6. Speaker Order and Task Order for Each Stimulus Compilations. 71 7. Differences Among the Four Compilations Over Both Perceptual Rating Tasks ............................................................. 87 8. Differences Among the Four Compilations Specific to the Dialect Detectability Task ....................................................... 89 \O . Compilation Characteristics and Rank Ordering According to Average Dialect Detectability Ratings ................................. 9O 10. Sentences in the High Detectability Group With 0. 1. and 3 Features Analyzed for Dialect Detectability ..................................... 95 11. Speaker Comparison for the Mean Change in Dialect Detectability Scores Across the Different Frequency Levels in the Low Detectability Group ..................................................... 98 12. Compilation Comparison for the Mean Change in Dialect Detectability Scores Across the Different Frequency Levels in the Low Detectability Group ........................................... 100 xi 13. Ranking of Features by Mean D Scores .................................. 102 14. Comparison of Mean D Scores for Comparable Features in the High and Low Salience Groups ................................................ 104 15. Years Experience as SLP for Participants Organized by Stimulus Compilation ............................................................ I"2 ~~ xii Ix) LIST OF FIGURES . Schematic of the Folk and Linguistic Theories of Language ............ 6 . Folk Perceptual Continuum of Gradient (or Variable) Dialect Use 11 . Phonological Severity Continuum From the Hodson Assessment of Phonological Process-3 ................................................ 3 7 . Example Stimulus Table for C omprehensibility Ratings ................ 77 . Average Dialect Detectability Scores in the High Detectability Group Obtained in Relation to the Frequency and Salience of AAE Features .................................................................... 94 . Average Dialect Detectability Ratings for the Two Speakers and the Two Compilations in the Low Detectability Group Organized by Frequency and Salience .................................................. 97 . Average Comprehensibility Scores Elicited from Stimuli Obtained in Relation to the Frequency and Salience of AAE Features ........ 106 . Original Single Dimensional Model of Perceived Dialect Density ....... 118 Multidimensional Model of Perceived Dialect Density .................... 1 19 xiii CHAPTER I INTRODUCTION “Their accent is so thick you'd have to have a knife to cut. like tryin’ to cut butter." “He had only a trace of an accent, ...but as we drove further South, his accent got heavier and heavier.” “They have a definite accent. They‘re the ones that are difficult to understand.” These statements. taken from the documentary “American Tongues” (Alvarez & Kolker, 1986), are classic examples of the way the general population describes individual differences between speakers of varying regional dialects. Many times these descriptions are expressed in metaphorical terms that depict variation in density or strength (e. g., “thick dialect" or “strong accent”). Language scholars, in contrast, often use categorical terms, such as idiolects, to refer to differences among speakers of a given dialect (Brook, 1970). Recently, however, some scholars have given attention to the more gradient view of the term dialect. While the variability of speech-language patterns is the focus of much of the US sociolinguistic research, some researchers have developed methods to quantify dialect use along a continuum concerning the amount of features used by different speakers. These scholars have coined the phrases “dialect density” and “dialect rate” to holistically describe the variable use of dialect by some speakers (for a summary, see Oetting & McDonald, 2002). These studies of dialect density have been limited by their focus just on speech production. Little seems to be known about the relationship between perception of native-language dialect density and the actual speech characteristics used within native language dialects. This relationship has, however. been explored more thoroughly among second-language speakers. The inherent goal of second-language acquisition is often to achieve near-native proficiency. The normal speaker ofa stigmatized dialect may or may not desire to speak a more standard dialect. However, many listeners, who speak standard dialects. believe that the speaker ofa stigmatized dialect should have the goal to speak more in accord with the standard language variety (Niedzielski & Preston. 2003). Exploring listener perceptions ofdialect density is important because many speakers of some dialects are penalized for their speech-language pattems (Niedzielski & Preston, 2003; Preston & Robinson, 2005; Wolfram & Schilling-Estes, 1998). Speakers, who have what the general population identifies as a “thick“ dialect. may experience greater negative social consequences than do those who speak a more standard version of the same dialect. These negative attitudes may limit access to important resources, such as housing, jobs, and education for the speakers of stigmatized dialects (Labov, 1966; Preston & Robinson, 2005; Purnell, Idsardi, & Baugh, 1999; Smith, 1993). These studies suggest that several factors specific to the listener. rather than the speaker, may determine the degree to which the speaker is penalized. These factors include the listener‘s attitude toward the dialect, the experience that the listener has with the dialect, and the level of social or economic power that the listener may have over the speaker. Perceptions regarding dialect density have particular relevance for speech- language pathologists (SLPs; Preston & Robinson, 2005). The SLP is professionally certified to detect a breakdown in communication. It is the SLP’s job to differentiate between a breakdown that is attributed to a speech-language disorder and one that results from his/her own lack of experience with a client’s normal dialect patterns and/or personal attitudes toward those dialect patterns. However. this distinction is not always [‘0 easy to make when subjective ratings (such as overall intelligibility) are used in speech- language assessments. In other words, misdiagnosing a normal language difference as a disorder may result when normal speakers of a dialect are perceived to be as difficult to understand as disordered speakers of more standard dialects. Misdiagnosis may also occur when the SLP’s personal attitudes result in the judgment that the features should be classified as “incorrect” language. Such misdiagnoses may explain why African American students are over- represented on SLP caseloads when compared to the general population (US. Department of Special Education Programs, 2004). Many of them acquire a social dialect that is not spoken by most SLPs.. The dialect, commonly referred to as African American English (AAE), differs from Standard English varieties in many ways. These differences include phonological, morpho-syntatic, semantic, and pragmatic aspects (Bailey & Thomas. 1998; Pollock. et a1. 1998; Rickford & Rickford. 2000; Smithennan, 1977; Smitherman, 1994; Stockman, 1996; Terrell & Jackson, 2002. Wolfram & Schilling-Estes, 1998). The overwhelming majority (93%) of SLPs are from an Anglo-European American background (ASHA, 2005). They do not speak AAE and. therefore. may be especially prone to make a wrong diagnosis (Seymour & Bland, 1991‘). Their inexperience with AAE could cause them to judge some speakers as difficult to understand. Their attitudes toward or ignorance about the use of particular dialect features may further cause them to be reluctant to accept the features as normal language. Clinical misdiagnoses can have negative consequences. In the public school setting. for example, a misdiagnosis could influence the way a student is viewed by peers. teachers, and by him/herself. Such misdiagnosis may be primarily based upon the ability of the SLP to detect the dialect and/or the inability of the SLP to understand the dialect. AAE speakers are diverse in the AAE linguistic features that they use. Based upon the variable use of certain features. some AAE speakers may be more prone to misdiagnosis than others. This dissertation research aims to determine how the varying presence of AAE dialect features impacts Anglo-European American public school SLPs’ perceptions of how easy a speaker is to understand and the detectability of the speaker‘s dialect. T erminolo gv The meanings of words are “neither overt nor finite" (Stockman, 2000, p. 340). The terms used in reference to language variation are no exception. Even though both linguists and non-linguists use the terms “dialect” and “accent" to describe language variation, the connotations may differ widely (Malmkjaer. 1991; Wolfram & Schilling- Estes, 1998). The linguistic distinction between these two words usually involves the levels of language that are being referred to: “[A dialect is] any user-defined variety [ofa language], that is, any variety associated with speakers ofa given type, whether geographically or otherwise defined” (Malmkjaer. 1991. p. 93-94). “The term accent is generally used in the field [of linguistics] in the strict sense of a variety differing relevantly from others only in phonological respects, not in grammar or lexis” (Malmkjaer, 1991, p. 95). Another distinction is also made by language professionals. The term dialect has been defined and used by some to denote a variety present in a first language, while accent is used to denote linguistic features present as a result of a second language (Nicolosi. Harryrnan. & Kresheck, 2003). The distinction made by the general population is not always as clear. In fact, the terms are often used interchangeably (Niedzielski & Preston. 2003'). When non-linguists are quoted in this dissertation, the term accent may include features related to all levels of language associated with form, content, and use. For the purposes of this research, the investigator used the term dialect to denote a language variety that can be described in terms of particular phonological, morphosyntactic, semantic, and pragmatic features. Thus. “the phonological features of the dialect ” can be described as a particular set of the features that characterizes a dialect. The term accent will be used to refer to the pronunciation features associated with second language acquisition (i.e., “second-language accent") due to the fact that this is the term used often in the literature. Two Perspectives on Normal Language Variation: Folk vs. Linguistic Understanding perception of normal language variation within the general population also entails understanding how such variation is viewed in the scholarly literature. Linguists and the non-linguist general population may not view language differences in exactly the same way. Specifically, linguists agree that all human language systems, including dialects. are natural, rule-governed, and hence valid in their own right (Linguistic Society of America, 1997). This concept is not widely held by the general population. Preston and Robinson (2005) adapted a taxonomy developed by Niedzielski and Preston (2003) in which a Linguistic Theory of Language is contrasted with a Folk (or non-linguist) Theory of Language (see Figure l). Figure 1. Schematic of The Folk and Linguistic Theories of Language (Niedzielski & Preston. 2003) THE LANGUAGE A Folk Theory ofLanguage Good Language \ h 5 Ordinary Language Dialects "Errors" V THELANGUAGE A “Linguistic” Theory ofLanguage Dialectitl Dialect #2 Dialect #3 Etc. ldiolecttil Idiolect#2 Etc. With respect to the Linguistic T heory ofLanguage, the individual varieties (or idiolects) are the most concrete realizations of language. The conceptualizations of c‘lialects/accents and, subsequently, lcmguage are socially real but significantly more abstract than the actual productions made by speakers. Hence. a language is simply an abstract representation of idiolects and dialects determined by society to be linguistically and/or politically similar (Fasold, 1999; Linguistic Society of America. 1997). F urthermore, each dialect. and even each idiolect. is equally legitimate. Under the Folk T heory of Language (Niedzielski & Preston. 2003), the most concrete construct is The Langmge. Only the individuals who speak an idealized form of a language are viewed as truly speaking a rule-governed system ofcommunication. An individual is allowed to deviate slightly from this form and still be characterized as speaking good language. Once a person deviates too much. he/she drifts out of the realm of real language and into erred speech in which dialects and slips-of-the-tongue are of equal status. The determination of what is good language tends to correspond to the variety spoken by those persons who enjoy social prestige. namely those with wealth. education, and high social status. This distinction between the linguists" and non-linguists” views of normal language variation was infomied by interviews with 68 non-linguists residing in Southeastern Michigan (about topics related to language diversity). The transcripts of these interviews revealed that many people are very intolerant of stigmatized dialects. They found them difficult to understand at times and were frequently preoccupied with dialect features they view as “uneducated" or “incorrect" ways of speaking. The investigator of this dissertation reviewed the testimonies provided in Niedzielski and Preston’s (2003) paper and determined a tendency for the informants to discuss inter- speaker differences in terms of gradient levels of perceived dialect density or strength. These gradient temrs suggested that the informants were commenting on the degree to which the given speaker deviated from good language. From Categorical to Gradient View of Dialect Features Language researchers recognize that the speakers of any given dialect do not all use the same number and types of features (Craig. Thompson. Washington, & Potter. 2003; Craig & Washington. 2000; Craig & Washington, 2002; Craig & Washington, 2004; Craig, Washington. & Thompson-Porter, 1998a; Labov, 1966; Oetting & McDonald, 2001, 2002; Stockman, 1996, 2004; Washington & Craig, 1994, 1998; Washington, Craig, & Kushmaul. 1998). For example, Craig and colleagues (2003) reported that African American children produced different amounts and types of dialect features across grades in Michigan public school districts. These individual speech patterns are often called idiolects (Brook, 1970). When members of the general population discuss speakers of various dialects in the United States (US), they may perceive these inter-speaker language differences along a continuum or gradient. rather than in a categorical manner. For example. in the documentary “American Tongues” (Alvarez & Kolker, 1986), a speaker from New England described the speech of her fiancee as they traveled South to meet his family. She began by stating that he had “only a trace of a Southern accent” before they began the trip. She stated that as they drove further South. “his accent got heavier and heavier.” As is exemplified by this comment. these gradient descriptions of language variation sometimes take the metaphorical form of increasing amounts of mass or density. For example, in the same documentary, a speaker from New York stated that the Southem dialect of Mississippi and Georgia is so thick “you‘d have to have a knife to cut it—like trying to cut butter.” Further evidence of non—linguist perception of gradient dialect use was provided in Niedzielski and Preston‘s (2003') study. An examination of the testimonies provided in their study (see above) revealed that the informants used multiple methods to convey dialect density. For example, one informant used the phrase “really, really. really. really. Southern” (p. 105) to describe the speech of one speaker. Another informant used the phrase “very strong Southern accent” (I). 107) to describe the speech of another speaker. The transcripts also revealed contrasting terms, such as “slight Southern accent” (p.110) and “no accent” (p. 97). apparently to represent the opposite extreme to speaking a “strong” or “thick” dialect. This continuum may refer to both the frequency and type of dialect features being used by a speaker. For example, one informant contrasted speakers from Tennessee with those from “smaller towns in Kentucky” (p. 99). He explained that “the Southern people from Tennessee... still had an accent but you know. they didn‘t use as many double negatives,” (pp. 99-100). Researchers who study language acquisition and disorders have documented that speakers of non-mainstream dialects produce a variety of different features and with varying levels of frequency (Craig et al., 2003; Craig & Washington, 2002; Craig & Washington, 2004; Craig et al., 1998a; Oetting & McDonald. 2001; Stockman, 2004; Washington & Craig, 1994, 1998; Washington et al., 1998). The terms dialect density or dialect rate refer to the variable rates at which specific features associated with a given dialect are present in the speech of individuals. Several methods have been used by language researchers to measure dialect rate in children. Oetting & McDonald (2002) investigated and described the different methods of dialect quantification that have been used in numerous studies (e.g. Craig & Washington, 2000; Oetting et a1, 1997; Washington & Craig. 1994. 1998). The predominate purpose of the studies cited in Oetting & McDonald’s (2002) paper was to describe the expressive language of children, who speak non-mainstream English dialects. These studies have illuminated the diversity that is present in the language ofchildren who speak non-mainstream dialects and the factors related to variable use of dialect features. The audience of these studies was primarily speech-language pathologists (SLPs) because SLPs must understand the acceptable range of variation within dialects to inform their diagnostic decisions. However, SLPs must be concerned not only with dialect production, but also with dialect perception by the SLPs, themselves. It was not within the scope of these studies to determine the perceptibility of the dialect features. Thus, they did not provide insight into the basis for the perceived differences in dialect density or variable use of features within the same dialect. Based upon the descriptions of variable dialect use provided by the informants in Niedzielski and Preston’s (2003) work. the investigator of this dissertation’s study designed a schematic continuum to illustrate a possible model of how non-linguists may view dialect use (see Figure 2). This continuum is ordinal in nature with “no accent” or “good speech” at one pole and extreme dialect use at the other. A speaker may move along the continuum depending on personal experiences. For example. in Niedzielski and Preston’s study, Speaker B stated: I developed a Southern accent in Texas... I grew up in the New York City area. went to Texas for two years. by age eight I had a Texas accent, went back to New York, I lost it pretty much in about six months... (p. 102). While it appears that the variable use ofdialect features by a speaker plays a major role in determining movement along this continuum, it is still unclear how this variability influences perceptions of dialect use by a listener. 10 Figure 2 Folk Perceptual Continuum of Gradient (or Variable) Dialect Use Very Intelligible Unintelligible Not Noticeable Very Noticeable 0 No Accent/Dialect Slight Accent/Dialect Thick/ Heavy/ Strong/ Definite Accent/ Dialect Attitudes Toward N on-nrainstream Dialects As is conveyed in Figure 2. references to variable dialect use by people in the general population tend to coincide with references to negative factors that may inhibit communication because the speech: (a) is unintelligible or difficult to comprehend (dialect comprehensibility) and/or (b) diverts the listener‘s attention away from the message communicated to a focus instead on the dialect itself( dialect detectability). It is possible that these factors may be influenced by or coincide with the negative public attitudes toward speakers of some dialects that have been documented by linguists (Wolfram & Schilling-Estes, 1998). Consider the previous example from the American Tongues documentary of the New England speaker describing the accent of her fiancee as getting “heavier and heavier” (Alvarez & Kolker, 1986). She continued by stating that her fiancée‘s speech "became filled with all these hillbilly kind ofregionalisms. You know, this real kind of ‘you-all’ stuff.” This speaker reported that she ended her engagement because she could not bear the thought of giving birth to “Southern-speaking babies.” In the Niedzielski and Preston (2003) transcripts. informants were documented often using phrases like “poorer language skills” and references to education level 11 coinciding with discussions about speakers with stigmatized dialects. For example. one informant stated that “the people from Tennessee seemed to be more up on education than those from you know the smaller towns in Kentucky” (p.99). Several studies have examined these negative attitudes toward dialects using a methodology called the “matched-guise” technique. The practical implications of having certain negative attitudes toward various dialects are probed implicitly rather than directly asking respondents to state their attitudes--attitudes that are likely to be overtly denied by respondents (Adams, 1981; Belton. 1984; de Bardelaben. 1980; Edwards, 1982; Johnson & Buttny, 1982; Labov. 1966; Labov & Ash. 1997; Lambert. Hodgson. Gardner, & F illenbaum, 1960; Marston, 1974; Mulac & Rudd, 1977; Munro, Derwing. & Flege, 1999: Preston. 1996b; Purnell et al., 1999; Smith. 1993; Stein. 1981; Voigt. 1994,). In these studies, the listeners rated speakers of different dialects on subjective categories that were apparently unrelated to the linguistic features used or not used. These categories were conceptually grouped in terms of competence and solidarity. Non- mainstream dialects are often ranked low on the dimension of competence. This category includesjudgrnents about intellectual fortitude or work ethic. Therefore, speakers of these stigmatized dialects were commonly judged as “uneducated“ and “lazy”. C onversely. non-mainstream dialects tended to be rated high on the dimension of solidarity, invoking characteristics such as “friendliness” and “trustworthiness”. For example, Preston (1999) asked Michigan respondents to rate the degree to which they felt different personality traits applied to the residents of the different states in America. The personality traits corresponded to social status (standardness) and group solidarity (friendliness). a dichotomy that is analogous to the previously mentioned dichotomy of “competence” verses “solidarity”. While the Northern and Midwestern states (i.e., the speakers perceived to have a more mainstream dialect) received high ratings for social status, the Southern states (i.e.. the more non-mainstream dialect speakers) received higher ratings for group solidarity. Similar findings were discussed by Edwards (1982), who used the same variables of social status and group solidarity. Therefore. it should not always be assumed that speakers of stigmatized dialects will be penalized by others in every social situation. However. the traits pertinent to social advancement in the broader US. society (i.e., intelligence, education, and so on) are the ones in which non- rnainstrearn dialect speakers are commonly perceived to be deficient by the wider community. According to Preston and Robinson (2005'), if professionals in positions of cultural, social, or economic power base expectations of intellectual fortitude on dialectal characteristics. then considerable social consequences can occur for speakers of non- mainstrearn dialects. For example, Purnell, Idsardi. and Baugh (1999) revealed that even relatively subtle linguistic cues can have grave social consequences for speakers. When a single person uttered the word “hello” in three different dialects to a group of listeners (Standard American English, AAE. and Chicano English), variation in vowel tenseness and pitch on the first vowel was enough to trigger identification of the dialects. When the utterance was expanded to “Hello, I‘m calling about the apartment you have advertised in the paper,” and presented to actual property owners, the Standard American English utterances elicited appointments approximately 70% of the time. The AAE and Chicano English versions (i.e., the non-mainstream dialects) elicited appointments about 30% of the time. This study suggests that if a speaker ofa stigmatized dialect is evaluated by a 13 person with social power who adheres to the Folk Theory of Language (see above). then his/her use of a stigmatized dialect may be penalized. The more a speaker uses features associated with the dialect. especially highly stigmatized features. the more likely could this be the case. The reason is that the dialect is either highly noticeable or less intelligible to the listener. as is evident in the testimonials of non-linguists (Alvarez & Kolker, 1986; Niedzielski & Preston. 2003). Thus the listener attributes this increased communicative difficulty to the speaker’s communication skills rather than to his/her own lack of experience with the dialect. Dialect Dctectahilitv and Comprehensihilit}! As previously stated, colloquial descriptions of density reflect the degree to which a dialect is detectable and/or intelligible. A term related to intelligibility is comprehensibility, or the degree of perceived difficulty in understanding a speaker (Munro & Derwing. 1995a). The general population commonly makes such perceptual judgments. For example, in the American Tongues documentary (Alvarez & Kolker. 1986), a Southern speaker commented that New York speakers “have a definite accent. They’re the ones that are difficult to understand.” The first statement attests to the degree to which the informant noticed the New York accent. and the second statement conveys the sense of reduced comprehensibility (i.e., the amount of effort it took for her to understand the speech of New Yorkers). Furthermore. the construction “They are the ones” suggests that her own variety has been accused of incomprehensibility. Both judgments contribute to perceived differences in dialect density as discussed below. Dialect Detectability Dialect detectability is the degree to which a dialect is noticed in a speaker. Detection of a dialect or second language accent could distract from a speaker’s message. even though the dialect is not difficult to understand. The investigator of this dissertation‘s study conducted an internet search (Google, 2005) using the keywords “thick accent” and “thick dialect”. The websites listed identified the extent to which gradient descriptions ofdialect use corresponded to statements of perceived dialect density. On one website, professors were reviewed by students. Two students stated that a professor spoke with a “very thick accent;” however, they said that the professor was “not very difficult to understand.” Therefore, what the students were referring to in their use of the phrase “very thick accent” was the degree to which the accent could be noticed. Within the second-language acquisition literature. the issue of detectability has been emphasized. because many non-native speakers of languages desire to achieve near native-like production. Studies examining non-native speakers of English have used primarily subjective rating scales for listeners to detect the degree of foreign accent perceived in speech samples (Brennan & Brennan, 1981a. 1981b; Burda, Scherz, Hageman, & Edwards, 2003; Derwing & Munro, 1997; F lege & Fletcher, 1992; Johansson, 1978; Munro & Derwing, 1995a, 1995b; Niedzielski, 1999; Ryan, Carranza. & Moffie, 1977). F lege and Fletcher (1992) argued that the degree of foreign accent detected by listeners was influenced by several factors related to both the listeners and the speakers. One listener effect was the number of native speakers included in the sample. When large numbers of native speakers were included in the samples. the respondents tended to rate foreign accented speakers high for accent detectability. This observation suggests that non-native speech grew in salience when it was in the presence of a large 15 number of native speech samples, perhaps creating an oddball effect. A second listener effect was related to the familiarity with the sentences. After listeners became familiar with the sentences, they were more likely to rate the sentences higher in accentedness than before they were familiar with the sentences. This outcome suggests a cumulative effect on the ratings. That is, dialect features become more salient the more frequently they are experienced. Additionally, the findings suggest that after one has already achieved the linguistic meaning of an utterance, processing load is somehwat relieved so that attention to accompanying features of the message can be attended to. C omprehensihil i ty Sometimes the terms intelligibility and conrprehensibility have been used interchangeably in the research. For the purposes of this dissertation, both terms relate to how the words were understood in relation to pronunciation, not in relation to semantics or syntax. The words differ in that intelligibility refers to whether the words were successfully recognized or not (i.e., a dichotomous distinction). Con-iprehensibility is defined as the gradient level of difficulty experienced in recognizing the words (Munro & Derwing, 1995a). Presumably, this is in reference to the amount of top-down processing that occurs in understanding a speaker. In other words. a sentence may be intelligible in that all of the words of the message were successfully decoded as the intended words. At the same time, the sentence had low comprehensibility in that the listener experienced a great deal of difficulty during the decoding process. Dialects of the same language need not be mutually intelligible. In fact, the distinction between the terms dialect and language is more a political distinction than a linguistic one based on mutual intelligibility (F asold, 1999). While linguists accept this 16 point (Linguistic Society of America, 1997), members of the general population in the US. may not. Many people, who speak English in the US, tend to believe that fellow speakers of English should be intelligible to them. As previously mentioned, English speakers in the US. will often comment on the language of speakers whom they find unintelligible or incomprehensible. In the Niedzielski and Preston (2003) study. one informant contrasted the common belief that dialects of English should be mutually intelligible with the observation that many of them may not be at times: And sometimes you really don’t [understand another dialect]. I told your wife earlier that you CAN understand people in various parts of the United States, because we all do speak English. but that’s not true. Because I remember listening to a man in South Carolina, and I really did not understand one word he said. (p. 110) Another informant stated his belief that newscasters are taught to speak “Midwestern” speech because it is more universally intelligible. The concept of relative inter-dialectal intelligibility was further validated by a Preston (1997) and summarized in Niedzielski and Preston (2003). Respondents rated speech patterns found in areas of the US. based upon the perceived “degree of difference” from their own speech patterns ( 1: no difference, 2: slightly different, 3: different, and 4= unintelligibly different). Michigan respondents rated Louisiana, Mississippi, and Alabama speech patterns as unintelligibly different; Indiana respondents rated Massachusetts speech patterns as unintelligibly different; and Southem respondents rated Wisconsin, New York, and all of New England speech patterns as unintelligibly different. Therefore, the concept that dialects within the US. are not always mutually intelligible appears to be readily accepted among non—linguists. During the aforementioned internet search. the terms “thick accent” and “thick 17 dialect” were paired with the phrases “hard to understand” or “difficult to understand” using the same search engine (Google, 2005). The resulting websites predominantly included discussion about literature written in regional dialects that were difficult to comprehend for the reader and criticisms of university professors who speak English as a second language. The latter case has been validated in several studies of second language acquisition. It is generally accepted that foreign-accented speech is not as intelligible to native English speakers as native-accented speech. For example, in an earlier study by Lane (1963). a list of words was read by native speakers of English. Serbo-Croatian. “Indian” (sic), and Japanese speakers and presented to native English speakers. The foreign accented speakers were an average of 36% less intelligible to the listeners than the speech of native English speakers. However. a language difference sometimes does not affect intelligibility. per se. but instead, the more sensitive judgment of con-zprehensibility. Reduced comprehensibility is often correlated with evidence of increased cognitive load. increased error rates under certain conditions. or increased response time (Munro & Derwing, 1995b). For example. Nabeleck and Donahue (1984) found that the non-native English speech in their study was just as intelligible to native English speakers as was the native English speech under normal listening conditions. When reverberation (an echo) was superimposed on the audio stimuli to compromise the speech signal. the non-native speech was rated as significantly less intelligible than the native speech. This observation suggests that there are increased cognitive processing requirements that accompany listening to non-native speech. Although the native English listener may be 18 successful in understanding non-native English speech under ideal conditions, minimal alteration of the listening conditions could make a significant difference. In a more recent study, Munro and Derwing (1995a) discovered that when a sample was rated less comprehensible by listeners, it coincided with indicators of processing costs. English sentences were read by ten native speakers of Mandarin and ten native speakers of English. English speaking listeners judged whether the sentences were true or false statements. Sentences judged to be less comprehensible by the listeners were also answered more slowly than those judged to be more comprehensible. Munro and Derwing’s study validated the subjective notion of comprehensibility with evidence that such judgments were associated with actual cognitive effort. Some researchers have used subjective judgments of comprehensibility as a dependent variable. In giving subjectivejudgments of comprehensibility, participants are asked to indicate their own personal opinion of how difficult each speaker was to understand, regardless of whether the message was actually understood or not (Burda et al., 2003; Derwing & Munro, 1997; Gass & Varonis, 1984; Munro & Derwing, 1995a, 1995b). These studies indicate that subjective judgments of comprehensibility are reliable measurements of the difficulty in understanding that may or may not be experienced by the listener. Comprehensibility and intelligibility are not static. Numerous factors, including nonlinguistic ones (i.e. preconceived notions about the speaker; Niedzielski, 1999) may affect comprehensibility. However. two specific factors will be addressed. namely: (a) phonemic difference and (b) listener familiarity, as discussed next. Phonemic ditfi'rence. Phonemic difference refers to the possibility that a dialectal l9 difference can cause the intended meaning of an isolated word to change entirely. This was illustrated in a study by Plichta and Rakerd (2002). It compared speakers of two Michigan dialects. one spoken in the Upper Peninsula and the other in the Lower Peninsula. Participants from both dialect groups perceived computer generated C VC words. The investigators discovered that when the words were formed with vowels that fell in different perceptual boundaries of the two dialects — the speakers of the two dialects perceived acoustically identical signals as different words. The perceptual effect was minimized when a short carrier phrase was included before the word that was spoken in either dialect. This phenomenon indicates that a type of cross-dialectal perceptual confusion could be somewhat neutralized with context. However, it may require additional top-down processing to resolve the confusion. if the context is ambiguous (Munro & Derwing, 1995b). Listenerfamiliaritfiv. The above cited study also illustrates another factor that influences comprehensibility and intelligibility. Familiarity with a dialect may increase a listener‘s ability to comprehend a dialect. This point was made in a study by Gass and Varonis (1984). They observed that second language teachers were more successful transcribing sentences at the end ofa sample than they were in the beginning. While this suggests that a small degree of experience can increase intelligibility, another part of this study and other studies have demonstrated that a great deal of experience with a dialect can increase its comprehensibility even further. Gass and Varonis (1984) also compared inexperienced listeners with the second language teachers using the same non-native speech samples. The authors discovered that the inexperienced listeners made significantly more errors in sentence transcription than did the second language teachers with “many years experience” (p.79). Similar findings were revealed in a study by Labov and Ash (1997). The words pronounced by speakers in Birmingham. AL were played in isolation, phrases, and complete sentences to listeners in Birmingham, Chicago, and Philadelphia. Although, listener accuracy was greater for all groups when the word was in phrases and sentences. the Birmingham listeners scored consistently higher than the other listener groups. Even when the Birmingham listeners were incorrect, their guesses were phonetically closer to the actual intended word than were the other respondents’ guesses. In another study, Gerken and Deichman (1979) observed that among untrained test administrators, African Americans made less errors than Anglo-European Americans during the transcription of responses to test items on the Weschler Intelligence Scale for Children (WISC), when the children spoke nonmainstream dialects associated with African American groups — AAE. Although they did not observe that this interaction significantly affected the children’s overall test scores, the findings indicated that a racial/dialect mismatch can cause increased errors in tasks requiring speech intelligibility. Furthermore, this increase of errors appeared to be due to inexperience with the dialect to which the person was listening. Nelson and McRoskey (1978) reported that as African American children became older, they showed a greater ability to comprehend Standard American English (SAE). The implication is that with SAE experience comprehension increases for AAE-speaking children. However, the authors did not determine how well SAE speaking children fared in their perception of AAE after some experience with the dialect. It is likely that they did not have the experience with AAE that the AAE speaking children had with standard dialect. Speakers of minority and majority dialects experience their respective home dialects during interactions with family and friends. However, minority dialect speakers have the added exposure to mainstream dialects in educational settings as well as other settings in the community (Skutnabb-Kangas. 2000). The disadvantage of being a mainstream dialect speaker is that there likely will be minimal exposure to a diversity of other (non-mainstream) dialects (Baugh, 1983). Therefore, the speaker of the minority dialect may learn to understand a wider variety of speakers than the speaker of the mainstream dialect. The findings of Gass and Varonis (1984) and Plichta and Rakerd (2002) support the position that intelligibility with unfamiliar dialects may increase slightly within the first few minutes of exposure under ideal conditions. However. the findings of Gerken and Deichmann (1979) and Labov and Ash (1997) suggest that speaking a dialect creates a comprehensibility advantage with that dialect, which surpasses the transient familiarity gains from a few minutes of exposure during an experiment. The findings of Nelson and McRoskey (1978) and Nabeleck and Donahue (1984) suggest that when the listening situation is analogous to the competing distractions inherent to daily communication. the improvements attributed to familiarity gained by a few minutes of exposure are greatly reduced and do not compare to the more robust advantage gained from the native speaking experience. However. detection of these subtle, yet important, differences in the degree to which a dialect is understood, requires a fine-grained measurement tool. Therefore, the subjective and gradient measure of perceived difficulty of comprehensibility may be more sensitive to processing costs than are the more objective measures of intelligibility that require the listeners simply to IQ Ix) indicate the word that they heard a speaker say. Statement of the Problem The research discussed thus far suggests that when listeners communicate with speakers of dialects to which they have had little conversational exposure, the listeners tend to perceive these speakers along a continuum of dialect use. The continuum appears to coincide with judgments about how detectable and comprehensible the dialect is. This research is relevant to all professionals in positions of power over speakers of stigmatized language varieties because theirjudgments may be equated with opinions about speakers’ intelligence, education level, or work ethic. However, some professionals may have a more direct impact than others. especially when those professionals are required to assess the speaker’s language specifically. Three issues provided the context for this dissertation’s research: (a) types of listeners, (b) types of speakers, and (c) types of linguistic features to investigate. SLPS as Participants The profession of speech-language pathology is highly relevant to this research. The American Speech-Language-Hearing Association (ASHA) is the certifying organization of speech-language pathologists and other communication science and disorders specialists. It stipulates the scope of practice for speech-language pathologists and audiologists. This scope of practice includes many activities related to speech, language, and swallowing assessment and intervention. Among these activities, SLPs are licensed to do two related but very different things: (a) evaluate. treat. and prevent disorders of speech and language, and (b) “provide services to modify or enhance communicative performance (e.g. , accent modification. transgendered voice. care and to DJ improvement of the professional voice. personal/professional communicative effectiveness)” (ASHA, 2001, p. I-29). SLPs must differentiate the former from the latter, to ensure that clients seeking accent modification are not labeled as disordered (ASHA, 1983b). This task may not always be a simple one depending upon how the SLP defines disorder and whether or not that definition is interpreted within a cultural context. Van Riper (1978) defined a speech disorder as a speech difference that 1) calls attention to itself, 2) interferes with communication, or 3) places an emotional burden on the speaker. This definition is clear. An SLP should identify if the client’s speech meets the above three criteria. If an SLP finds a client‘s speech difficult to understand or noticeable (i.e., the first two criteria of the definition), then the client should be identified as disordered by the SLP. But such a diagnosis becomes less clearly optimal when assessing speakers of dialects that stand out from the Standard English varieties and/or are difficult for the SLP to understand. Some nonmainstream dialect speakers may satisfy all of the criteria of the Van Riper definition. yet are clearly not disordered. Alternatively. Payne and Taylor (2005,) suggested that SLPs might use this traditional definition of speech disorder with the important understanding that all communication must be understood and studied within the context of culture. Within the context of culture, the three criteria of the Van Riper definition take on new meaning. The individual to be diagnosed with disordered speech is one whose speech draws attention to itself or is difficult to understand within the community from which he/she learned language namely (i.e., the home speech community). This interpretation of the Van Riper (1978) criteria may seem simple, but it is very important. It may determine whether or not a client is accurately diagnosed. If a client speaks with a dialect that is highly detectable. then an SLP, who uses the Payne and Taylor interpretation. is alerted to modify the preferred assessment procedures to be more culturally relevant—perhaps by supplementing test scores with family interviews and naturalistic observations. The SLP also may use such information to provide more informed accent modification therapy with a philosophy of empowerment and language choice, as delineated by ASHA (1983b). Alternatively. the SLP. who uses the more traditional interpretation of the criteria. may use dialect detectability as the very reason why the child should be diagnosed as disordered. Regardless of the scenario, an understanding of how particular dialect features increase or decrease the detectability of a dialect is important. SLPs are of particular interest in the study of how dialect features affect perceptual judgments for a number of other reasons: (a) They are in a position of power over speakers of stigmatized dialects; (b) they are now mandated to consider factors related to practicing with a culturally and linguistically diverse population; (0) their profession continues to contain evidence of antiquated ideologies regarding language and culture. ((11) they are likely to conduct assessments with speakers of dialects other than their own; and (e) their diagnostic judgments are likely to be affected by any highly detectable dialects and factors that limit comprehensibility. SLPs as proflssionals with power. SLPs are in a particularly relevant position of power in which their decisions are focused on the speech-language behaviors of their clients. Their decisions can influence how language variation is viewed by the client, communicative partners of the client. and society in general. For example, ifa school SLP qualifies speakers of a particular dialect for special education under the diagnosis of speech-language disordered. then the Folk Theory of Language is validated for anyone who interacts with those children (including teachers. parents. and peers). In this scenario, a person perceived to be a "language expert“ labeled the child "disordered” based upon the normal language features of the child‘s dialect. In other words. regardless of how the SLP addresses the language difference in therapy, the child has been labeled disordered as far as the teachers and parents are concerned. Furthermore the diagnosis was made by a person who should know what disordered speech is. Therefore. acceptable features of a child‘s dialect will be viewed as “errors” and corrected as such by teachers. The misdiagnosis ofdialects as disorders has broader implications than the ones described thus far. “Languages are today being killed and linguistic diversity is disappearing at a much faster pace than ever before in human history...” (Skutnabb- Kangas, 2000, p. ix). Skuttnab-Kangas asserted that the policies and procedures of public schools aromid the world are largely responsible for this “linguistic genocide” (2000, p. 569). SLPs are central to this discussion. If SLPs, as language experts, continue to label dialects as disorders. then those dialects will be increasingly perceived as illegitimate fomis of communication. They will be gradually devalued and eradicated. Since language is an intrinsic part of culture (Stockman. Boult, & Robinson, 2004'). the cultural groups associated with these stigmatized languages and dialects will become increasingly devalued by the population with social and cultural power. Cultures will be increasingly eradicated along with the language or dialect (Skutnabb-Kangas, 2000). Therefore. SLPs as professionals in positions in power are especially relevant to the study ofdialect perception. Their role in the language human rights movement, although locally focused on one student at a time. is globally relevant. The need for SLPs to be knowledgeable and accurate when diagnosing disordered speech affects not only the individual client and his/her family but also the way that other speakers of that dialect are viewed in the broader community. Understanding the way that particular pronunciation patterns cause dialects to be noticed and/or judged as difficult to understand is central to understanding the misdiagnosis of nonmainstream dialect speakers. The ability to even detect a dialect must be studied. Additionally. the dialect features that cue an SLP to worry about the client‘s comprehensibility in the general population must be examined because reduced speech comprehensibility may be used as a criterion for diagnosing speech disorders. Professional mandates pertaining to linguistic diversity. The field of communication disorders has not always attended to issues of cultural and linguistic diversity. Fortunately, in the past 30 years. the field has undergone a paradigm shift. SLPs and audiologists are now required to consider issues related to treating and evaluating an increasingly diverse population (ASHA. 1983b). This multicultural movement was fueled by a combination of social. political, legal. and practical forces in the US. These forces converged to create an increasing need for SLPs to become aware of the effects of cultural and linguistic diversity on the clinical practice of speech- language pathology. To meet this need, ASHA has mandated two requirements for the treatment of cultural/linguistic diversity. First, ASHA in 1983 recommended that SLPs and audiologists consider normal language variation when making diagnostic and treatment decisions. In their “Position Paper on Social Dialects,“ ASHA stated that SLPs and audiologists may not label clients as disordered when the speech or language difference is attributed to nomial language variation. Such variation includes speaking a native language dialect or speaking English as a second language (ASHA. 1983b). In the paper. ASHA also stated that SLPs may engage in accent modification services. which reduce the detectability or increase the comprehensibility of a client’s dialect or accent to the general population. This service must be client-initiated. and the SLP must have specific knowledge of the language variety that a client seeks to modify. A related article addressed some of the implications for the new position on social dialects (ASHA, 1983a). In this paper the author(s) stated that “dialect usage exists on a continuum. The number and type of features that have a high frequency of occurrence may vary from speaker to speaker” (ASHA. 1983a. p. 27). This statement acknowledges the very premise on which this dissertation research is based. Sixteen years after the Position Paper on Social Dialects (ASHA. 1983b) was published. the second mandate was imposed. ASHA required that information related to cultural and linguistic diversity be infused into the educational requirements of speech- language pathology and audiology students (ASHA. 1999). The mandate was intended to ensure that future SLPs and audiologists receive an education that recognizes their role as professionals in an increasingly cultural]y/linguistically diverse society. Specifically. it was meant to ensure an academic and clinical education for clinicians that did not intrinsically devalue cultures and language varieties that were not considered mainstream or standard. Education in the speech-language pathology and audiology professions should be based on the research in the field. In the context of the mandated infusion of multicultural information into the communication sciences and disorders curricula. it is surprising that so little is known about what speech patterns may contribute to an SLP's ability to detect particular dialects or influence an SLP’s perception of the speaker‘s overall comprehensibility. This kind of infonnation is needed to provide prospective SLPs with an education that allows them to serve a wide variety ofclients. Antiquated paradigms concerning language and culture. Throughout the multicultural movement in communication sciences and disorders. an increasing amount of research. especially in the profession of speech-language pathology. has been relevant to the treatment and evaluation of culturally and linguistically diverse populations. This research has focused on three areas: (a) analyses of typical and atypical patterns of speech-language development across languages and dialects (e. g. Anderson. 2002; Brice. 2002; Craig & Washington. 2002; Craig & Washington. 2004; Craig. Washington. & Thompson-Porter. 1998b; Gutierrez-Clellen. C alderon. & Weismer. 2004; Junker & Stockman. 2002; Klee, Stokes. Wong. F letcher. & Gavin. 2004; Leonard. 1989. 1999; Oetting & McDonald. 2001; Seymour & Ralabate. I985; Seymour & Seymour. 1981; Stockman. 1986. 1996; Stockman & Vaughn-Cooke. 1991; Taylor. 1999; Washington & Craig. 1994. 1998; Washington et al., 1998; Wong. Leonard. Fletcher. & Stokes. 2004,). (b) the appropriateness of clinical procedures for cross-dialectal and/or cross-cultural speech-language assessments (e. g. Anderson. 2002; C heng. 1987; Hemingway. Montague. & Bradley. 1981; Johnston & Wong. 2002; Pena. Iglesias. & Licz. 2001; Qi. Kaiser. Milan. Yzquierdo. & Hancock. 2003; Schraeder. Quinn. Stockman. & Miller, 1999; Seymour & Pearson. 2004; Seymour. Roeper. & deVilliers. 2003; Stockman. 1997. 2000; Stockman & Schraeder. 1999; Washington & Craig. 1992; Westby. 1990). and. (c) to a much lesser extent. the patterns and etiology of diseases and disorders across different racial/ethnic groups (for a discussion and summary. see Battle. 2002). As is evident from the large body of research that has emerged from the multicultural movement. much progress has been made. Yet. the work is not finished. There continues to be remnants of past ideology inherent in current research and clinical practices in the field. Research in every scientific discipline. communication sciences and disorders included. is inherently based on certain socio-political constructs (Stockman. 1997b). Such evidence may be found in the paradigms used for categorizing research in the field of communication sciences and disorders. Studies that address minority groups are often classified as multicultural. while studies that address majority cultural groups remain under the category of general speech-language research. This paradigm reflects the notion that there is a mainstream. “culture-free” (and accent/dialect- free) group of people that requires no specific identity (Stockman et al., 2004). Such a paradigm may have justification. For example. due to the void of research with minority groups prior to the multicultural movement. these groups may be understandably granted unique classification. Such classification acknowledges the timely attention that is being paid to such groups. However. regardless of the justification. the effect is the same. Ethnic/racial minority groups are marginalized as different from the generic culturally-invisible group. and the majority cultural group is not studied as a distinct. cultural entity. As a result. the majority. non—recognized cultural group has not been examined as a unique factor in the misdiagnosis of speech-language disorders in socially stigmatized groups. This is unfortunate. since members of the majority culture are frequently the gatekeepers for domains of cultural capital in the US. Their role as gatekeepers has been largely unexamined in the speech-language pathology multicultural literature. Many professionals in communication sciences and disorders consider themselves as part of the mainstream or majorin group. According to ASHA‘s 2005 census. only 7% of its members certified as SLPs identified themselves as a racial/ethnic minority. considerably less than approximately 25% of the US. population. The role of SLPs. whether members of a minority group or not. is very important in the study of multicultural issues in the field. They should not be excluded as a culture-free group tmworthy of specific mention. Their cultural beliefs and behaviors affect the extent to which stigmatized groups may access domains of social capital. As previously discussed. they are professionals with a great degree of influence on the way that language variation is perceived by the community as a whole. Therefore. they should be considered central to the study of multicultural issues in the field. particularly the study of inter-dialectal perception. Do most SLPs adhere to the Folk Theory of Language or the Linguistic Theory of Language? Do they sometimes invoke one. sometimes another? Little is known about the predominant paradigm used by SLPs when considering linguistic diversity. Preston and Robinson (2005) entertained the possibility that some SLPs could operate in accordance with the Folk Theory of Language. If such a possibility is manifested. normal language variation that includes features outside the parameters of good language could be judged disordered speech. However. it is possible that SLPs are not among the non-linguist general population group. and do not adhere to such a theory of language. In fact. many SLPs were required to pass an actual post-secondary course in linguistics during their academic 31 training. Furthermore. recent graduates were required to receive information on treating culturally and linguistically diverse populations (ASHA. I999). The required curricula of SLPs. at least those who graduated in the last six years. seems to indicate that SLPs may receive the information necessary to form a more enlightened and accurate view of linguistic diversity. Levy‘s (2004) survey examined the knowledge of undergraduate and graduate speech-language pathology students. who were supposed to receive an infusion of multicultural/multilingual issues into their coursework. The survey showed that the students demonstrated fair to poor accuracy when asked test questions about the acceptability of various linguistic forms cross-dialectally. This outcome suggests that the six or more years of training that most SLPs receive may be inadequate to overcome the years of prescriptive language education and dialectal oppression that they likely received in their earlier education (Skutnabb-Kangas. 2000). This is likely the case even with the mandated infusion of rnulticultural/multilingual information into the SLP curricula of the most recent graduates. The goal of cross-dialectal acceptance may be all the more difficult if the judgments are more attitudinally constructed than factually constructed. Perhaps there are more changes that need to be made in the clinical and academic preparation of SLPs. In spite of curricular changes. informational texts in the field may maintain past inaccuracies. For example. the most recent edition of T erminologv of Communication Disorders (Nicolosi et al., 2003'). a popular dictionary for the profession. shows that the term dialect is defined as a language variety “spoken in a given geographical area. differing sufficiently from the official standard of the larger language community" (p. 80) This definition is contrary to the Linguistic Theory of Language that states that everyone speaks a dialect. including those. who speak whatever is meant by “the official standard.” It appears to reflect the Folk Theory of Language. which claims that there is a dialect—free standard variety. The definition is also. of course. factually incorrect. There is to date no “official" language in the United States. and even those states which have made English their official language do not officially recognize one variety as “standard” (Dale & Gurevitz, 1995). In summary. it is likely that many SLPs base their practice on the Folk Theory of Language rather than the Linguistic Theory. In such case. the study of SLPs is important as a factor in the misdiagnosis of minority dialect speakers. However. in determining perceptions of dialect detectability and comprehensibility. it is not necessary to determine the conceptual paradigm used by particular SLPs under investigation. These judgments are merely the foundations on which both accurate and inaccurate diagnoses are based. In other words. regardless of whether the speaker is judged to be disordered and incorrect or litigz.ii.stic'c‘tlly different and legitimate. such judgments may be largely dependent on perceptions of the speaker’s comprehensibility and dialect detectability. SLPs and cross-dialectal assessnrents. Dialect mismatches between SLPS and their clients are likely to occur on an increasingly frequent basis. Although the US. population has been diversifying for many years, SLPs remain a comparatively homogeneous group (ASHA. 2005). ASHA has identified the racial/ethnic mismatch between the demographics of its professionals and the overall US. population as a problem to be targeted in a focus initiative (ASHA. 1995). To address the problem. the organization has recommended that efforts be made to recruit and retain professionals that are members of minority groups. However. until more minorities enter the 33 profession. the current demographics ofcertified SLPs are in contrast to those of the US. This contrast between the demographics of the profession and those of the US. is causing an increasing number of SLPs to evaluate and treat clients with accents and dialects. which may be unfamiliar. This discrepancy creates an increasing likelihood for inaccurate assessments. This problem is particularly daunting considering the inadequate amount of research that has been conducted on the ability of SLPs to make adequate diagnostic judgments about speakers of minority dialects and accents. The procedures that may be most affected by the SLP’s lack of familiarity with a dialect or accent are those involving subjective measures of the client‘s behavior. such as intelligibility judgments and other global speech-language judgments. The abundant research exposing the inadequacy of many standardized tests with culturally and linguistically diverse populations has caused disillusionment with them. (Wilson. Wilson. & C oleman. 2000). Therefore. it has been recommended that SLPs use “alternative assessment procedures" as the “best method for overcoming the inherent problems in using standardized tests” (Wyatt. 2002. p. 426). These assessment procedures include analysis of language samples collected in naturalistic contexts and the use of criterion—referenced as opposed to norm-referenced procedures. The alternatives to standardized testing may be just as inadequate. if SLPs lack adequate knowledge about typical language development in minority populations. In such a scenario. the SLP operates without normative data from standardized tests and lacks information regarding typical language patterns of diverse groups. Interpretation of these more subjective alternative assessments must then be based on the SLP’s own personal experiences. If SLPs experience primarily standard language varieties. then their judgments are likely to 34 be derived from just those experiences. In this way. subjectivejudgments may inherit the very inaccuracies associated with standardized tests based on normative data primarily from the majority language community. The same factors pertaining to the general population’s perception of non- mainstream dialects should be investigated as possible factors that could compromise the accuracy of speech-language assessments. In other words. factors pertaining to the SLP. in addition to those related to the client. should be investigated: 1) comprehensibility of non-mainstream dialects. and 2) detectability of non-rnainstrearn dialects. SLPs and dialect comprehensibilitv judgments. When assessing their clients. SLPs often depend on measures of intelligibility. As a procedure unto itself. the assessment of intelligibility has been identified as a crucial component of speech- language evaluations (Kent. Miolo. & Bloedel. 1994). However. a wide range of diagnostic procedures depend on the speech perception abilities of the SLP. not just the assessment of speech intelligibility. Reduced speech intelligibility or comprehensibility may affect any assessment procedure that requires an SLP to make perceptual judgments. inclusive of standardized test administration. phonetic transcription. language sample transcription. data gathering during treatment. etc. For example. Schmid and Yeni-Komshian (1999) examined a task that SLPs typically perform on a daily basis with their clients — the identification of mispronunciations. They reported that when Standard American English-speaking undergraduate SLP and audiology students listened to dialects and accents that differed from their own. increased cognitive load was indicated by prolonged response time and increased inaccuracy in identifying when a client had mispronounced a word. They attributed this increased cognitive load to a reduction in the SLP’s ability to comprehend non-native speech. Related to the SLP's judgments about intelligibility is the judgment of how comprehensible a client will be to the general population. In a clinical scenario. the SLP will do a phonological analysis that compares a client‘s production to an intended target. After completing such an analysis. the SLP will decide whether a client‘s speech needs treatment or not. If so. then a treatment plan is constructed. Such a plan specifies which features are to be targeted first. Several guidelines have been provided for this decision. Some guidelines require the SLP to first target those patterns that will most negatively impact speech intelligibility (Hodson. 2004). Using Hodson’s criteria. patterns involving consonant omissions are assumed to negatively impact intelligibility the most. Hodson suggests further that omission ofa post-vocalic final consonant (i.e.. final consonant deletion) will have a greater impact on intelligibility than omission ofa consonant that is in a final cluster (i.e.. final cluster reduction). She offered the following schematic of this intelligibility-process relationship (Figure 3). Figure 3 Phonological Severity Continuum From the Hodson Assessment of Phonological Process-3 (Hodson, 2004) . Acceptable Profound Severe Moderate Mlld Variations Extensive Some Some Sibilant Utterance- omissions; omissions; deviations distortions; final some extensive from Mild minimal devoicing; substitutions substitutions and from shifts of regional/ Severe articulation cultural/ categories dialectal variations Dialectal variations were included in the category of“acceptable variations" apparently at the subtle extreme of the continuum. However. some dialects may use features that are not as subtle as this schematic suggests. Dialects may contain features described under the categories of Mild. Moderate. and Severe. Little is known about how SLPs might judge speakers of dialects that include phonemic substitutions or even deletions. but two outcomes are possible. First, SLPs couldjudge even those more salient patterns as “acceptable variations.“ with the understanding that they are normal language variations. The speakers using such phonological patterns would not be categorized along the continuum of disorder and may or may not be viewed as less intelligible to the general population. Second. the SLPs could judge them with the same criteria that one used to judge a phonological disorder. That is. the SLPs will judge omission patterns as less intelligible to the general population than the speaker who does 37 not use such patterns. SLPs and dialect c‘letectability judgments. SLPs must become aware of the features that draw attention to dialects. Such awareness may help them to analyze their client’s speech more accurately. For example. if the SLP knows that final consonant deletion is a highly detectable dialect feature that contributes to decreased comprehensibility. then lie/she can re-evaluate the diagnostic evidence to ensure that the pattern did not create the illusion of a disorder. Johansson (.1978) alluded to this concept in the introduction to his study of native Swedish-speakers of English as a second language when he quoted Robinson (1973. p. 192): Research into error analysis is vital. since the teacher should know in advance the type and importance of the error a particular group of pupils would be likely to make... It may happen that the teacher becomes obsessed with insignificant errors to the exclusion ofothers that are much more important. Guidelines are required to help establish a scale of relevant priorities in error correction. This quotation is equally relevant to the SLP who engages in accent modification therapy. Albeit a slight change in wording from “error“ to “feature” is required due to the fact that dialects are systematic and legitimate in their own right (Linguistic Society of America. 1997; Wolfram & Schilling-Estes. 1998). It also is relevant to traditional clinical assessments and therapies in which SLPs engage. SLPs are expected not to fault clients for normal language variation. In this scenario. the “insignificant errors" to which Robinson (1973) referred are analogous to the normal dialectal variants. especially the highly detectable features. that could be misclassified as developmental speech errors. Misclassification is just one possible effect of highly detectable dialect features. Such features can also alert the SLP that a client speaks a nonmainstream dialect. In such a scenario. the SLP should become cautious about interpreting standardized test scores. AAE as a Language Iz'ariety As discussed thus far. much of the research on the perception of non-mainstream English dialects has pertained to English as a second language accents. However. this dissertation focused on first—language dialects. ASHA’s Position Paper on Social Dialects. stated: “A bilingual speaker may present a situation that is analogous to a speaker who uses a social dialect” (ASHA. l983a. p. 26). Therefore. much of the research on perception of second language accents may apply to the study of first language dialects. But first language dialects may be more prone to assessment inaccuracies than are second-language accents. given the Folk Theory of Language (N iedzielski & Preston. 2003). As previously discussed. The Language is the only legitimate variety of speech in the Folk Theory. Minority dialect features are considered equivalent to slips-of—the- tongue and other speech mistakes. A non-mainstream dialect is thought to result from lack of education. lack of intelligence. lack of effort. or even recalicitrance (suggesting that nonmainstream speakers actually have the ability to use mainstream varieties if they simply choose to do so). Second language accents have a very obvious justification—the speakers spoke a different language as their native language. For the SLP who operates under the Folk Theory. this means that the second language features used by his/her client are legitimized. The deviations from Good English are the result of the incomplete learning of English. The SLP is free to assume that the client probably speaks an idealized form of his/her native language. and therefore may not be uneducated or unintelligent. Therefore. the speech patterns of second language dialects may have a higher status in the minds of non-linguists and possibly SLPs than the features of 39 nonmainstream dialects. Although many dialects are spoken in the US. the AAE dialect is worthy of particular study for two reasons: (a) its speakers have been documented to vary in the number and type of features used. and (b) its speakers are frequently misdiagnosed as disordered by SLPs. Description ofAAE. AAE is the language variety spoken largely by African Americans in the US. It has vocabulary. phonology. syntax. morphology. and pragmatics that are distinct from Standard American English (Rickford & Rickford. 2000). Historically. AAE may be related to creole languages that evolved from a pidgin system that combined Western European languages with West African languages (Mufwene. 1998). It has also. however. acquired features as a result of the early second- language learning experience. contact with a variety of English dialect speakers. and. of course. changes and innovations in the variety initiated long after slavery (Wolfram and Schilling-Estes 1998). The linguistic features associated with the dialect have been identified in multiple sources (Bailey & Thomas. 1998; Craig et al., 2003; Craig & Washington. 2004; F asold. 1981; Luelsdorff, 1975; Oetting & McDonald. 2001; Pollock et al., 1998; Smitherman. 1977. 1994; Stockman. 1986. 1996). Speakers of the dialect differ in the number and types of features used (ASHA. l983a; Stockman. 2004; Terrell & J ackson. 2002). Factors such as education level and socio-economic status have been discussed in several papers (Craig et al., 2003; Craig & W ashington, 2002; Craig & Washington. 2004; Oetting & McDonald. 2001. 2002; Stockman. 2004; Washington & Craig, 1994, 1998). The misdiagnosis ofAAE as disordered language. African American children are 40 over-represented on speech-language pathology caseloads across the country (US. Department of Special Education Programs. 2004). The main reason for this disparity is the use of assessment procedures that are highly biased against AAE (Seymour & Bland. 1991). Most SLPs who are Anglo-European American speakers of a mainstream English variety are unfamiliar with AAE. Consequently. they may judge the speakers unfairly based on their own beliefs about how understandable the dialect is. Speakers whose variety of AAE is either very noticeable or not so comprehensible to the SLP may be misdiagnosed more often than other speakers of AAE. Given this possibility. it is important to determine how the varying use of AAE features influence the perceptual judgments of SLPs. AAE phonology. AAE has a systematic phonological system that differs significantly from other fomis of American English (Bailey & Thomas. 1998; Luelsdorff. 1975; Pollock et al.. 1998; Stockman. 1996. 2004). Common AAE phonological patterns have been described in several sources (Bailey & Thomas. 1998; Pollock et al.. 1998; Stockman. 1996); however. speakers of AAE are diverse in the frequency and types of phonological features they might use (Bailey & Thomas. 1998; Pollock et al.. 1998; Stockman. 2004; Wolfram. 1986). For example. one speaker of AAE might use multiple phonological features commonly associated with the dialect in every sentence spoken. while another one may use relatively few features. As stated previously. this quantification of AAE features produced by a speaker is termed dialect density or dialect rate (Craig & Washington. 1998). Different speakers also may use different types of features. For instance. one speaker might substitute /v/ or /f/ for final voiced/voiceless -th sounds. while another one substitutes /d/ or /t/. Yet another speaker may not substitute 41 any of these sounds for voiced/voiceless —th but still use other phonological patterns associated with AAE. F urthermore. a single speaker may use any. all. or even none of these possibilities across a variety ofdifferent contexts and still be considered a speaker of AAE because of other morphosyntactic, semantic. pragmatic. or intonational patterns. Although AAE's phonology has been studied less than its morpho-syntax (Bailey & Thomas. 1998). the phonology may have the greater impact on the dialect’s perception. It has been long known that phonological features of nonnative accents carry more perceptual weight than morpho-syntactic features and could trigger greater disturbances in intelligibility (J ohansson. 1978). One possible reason for the perceptual impact of phonological features is that they occur more frequently than morpho-syntactic ones. That is. there are more phonemes in any unit of speech than there are morphernes. Stockman, Guillory. Newkirk. and Seibert (in submission) discovered that AAE phonological features accounted for a larger proportion of the dialectal variance in the speech of African American children than did morpho-syntactic ones. The effect of phonological features may occur so frequently that they contribute to a gestalt perception of dialect rather than specific awareness of features. Whereas. morpho-syntactic features may be more categorically salient. During development. AAE phonology tends to be retained in the classroom past the age when standard grammatical patterns may be used (Craig et al.. 2003; Craig & Washington. 2004; Stockman. 1996). A possible reason for the longevity of AAE phonological features in children could be that the individual phonological features of their dialect are apt to operate below the level of consciousness. while morpho-syntactic features may be more obvious to them (Wolfram & Schilling-Estes.l 998). Thus. to code switch toward a standard dialect. the specific morpho-syntactic features will be targeted by the speaker first. The phonology may still contribute to the overall perception of dialect. yet the speaker may remain unaware of what or how to sound more “standard.” Therefore. when learning to code-switch into a standard dialect. the children may retain AAE phonological patterns. even when doing their best approximation of the standard dialect. Although they may not be readily noticed and changed by speakers. phonological features associated with any stigmatized dialect still can be just as socially stigrnatizing as morpho-syntactic features. Wolfram and Schilling-Estes (1998) suggest that consonantal features are especially prone to social stigma compared to features affecting vowel sounds. While AAE phonological features can occur in any word position. this dissertation’s research focused on just those sounds in the final position of words. This is the position where the majority of AAE phonological features occur (Bailey & Thomas. 1998; Luelsdorff, 1975; Pollock et al.. 1998; Stockman. 1996. 2004). Phonological features of AAE may cluster at the ends of words because word and syllable final consonants (i.e.. codas) are more marked than are consonants in word and syllable initial positions (i.e.. onsets). Languages and dialects are prone to modify final consonants to create a less marked form (Steriade. 2004). This expectation was validated in AAE when Craig and colleagues (2003) examined the occurrence of nine different types of AAE phonological features among children in grades 2 through 5. Three of the four patterns used most often involved final consonants: (a) Substitutions of th (“with” /w10/ 9 [wlt] or [wlt]) (b) consonant cluster reduction (e.g.. “fast” /faest/ -) [faes]). and (c) postvocalic consonant reduction (“feet” /fit/ 9 [fi]). 43 F requencv and Type ofFeatures as Variables The research reviewed thus far suggests that several variables should be considered in studies that aim to determine the impact of dialect features on cross- dialectal perception. The study by Johansson (1978) and the testimonials from non- linguists (Alvarez & Kolker. I986; Niedzielski & Preston. 2003) suggest that the two judgments of dialect detectability and dialect comprehensibility are direct indicators of dialect density perception. Listener judgments may be affected by numerous factors. two of which are related to the dialect patterns used in a given unit of perceived speech: (a) number of dialect features present. and (b) type of dialect features present. Frequency offl'atures included. If a speaker uses more dialect features in a given unit of spoken language. then it is logical to assume that the presence of a particular dialect would be detected more quickly and easily. For example. Ryan. Carranza. and Moffie (1977) revealed that the degree of perceived foreign accent was positively correlated with the number of segmental substitutions in a brief reading passage. This outcome offers perceptual justification for the use various methods for calculating dialect density within the child language acquisition and disorders literature. as summarized by Oetting & McDonald (2002'). These meters of variable use of recognized dialect patterns commonly reflect an average of the number of features used per some linguistic unit. The term dialect density appears to be analogous to the concept of physical density (mass per unit of volume). In physical density. each gram representing a unit of mass is equal to every other gram. Thus. increases in density are invariably linear on a ratio scale (with an absolute zero). A . . . . t substance measurrng Sg/cm3 IS five times more dense than a substance measuring lg/cm”. 44 This may be where the similarities between dialect density measurement and physical density measurement end. Dialect features may not affect perceptual judgments equally. That is, a phrase containing five phonological dialect features per unit of language may or may not be five times more perceptually salient than one containing one dialect feature for the same unit of language. Perceptualsalience may depend on what type of dialect features are included in the sentence. Also. each additional feature may affect perception in differing ways. It is possible that there is a threshold and/or a plateau for the perception of dialect features. A threshold means that a certain number of features per unit of language would be needed to even detect that the dialect is being spoken. A plateau means that the inclusion of a certain number of features may cause the perceptual judgments to level off. creating little change when additional features are added. Types of features included. Research in the field of perceptual dialectology suggests that the degree to which a dialect is noticed camrot be explained fully by the sheer number of dialect features present in a unit of language. In some instances. a single linguistic cue carries so much perceptual weight that it can override the perception of other dialect features in the same context. For example. Graff. Labov. and Harris ( 1983) reported that when the onset of the diphthong /aU/ was acoustically altered to become [an] in the speech of an African American. Philadelphia judges identified the speaker as Anglo-European American —— despite the presence of AAE phonological features included throughout the rest of the sentence. J ohansson ( 1978) hypothesized that all linguistic features were not perceived equally. He studied the perceptibility of the various speech errors made by native. Swedish-speakers. who were learning to speak English. A fomrula was developed for the 45 perceptually-based evaluation of non-native error patterns. The formula involved: (a) determining whether an error caused increased difficulty for the native listener to understand the message of the non-native speaker. and (b) determining the degree to which the error was “irritating” to the listener. These two components are strikingly similar to the two components of intelligibility/comprehensibility and detectability used by the non-linguists in the testimonials previously described in this chapter (Alvarez & Kolker. 1986; Niedzielski & Preston. 2003). An error that is comprehensible but is not irritating would constitute the lowest grade oferror. In the context ofthis dissertation. this interpretation can be rephrased as follows: a dialect feature that does not reduce understanding and is not very detectabile is less likely to impact the perceptual judgments of SLPs than other features. Johannson concluded that: (a) phonological errors carried more weight than morpho-syntactic errors. (b) phonemic errors were judged as more serious than subphonemic errors. and (c) subphonernic errors that differed greatly from the target were judged more severely than were those that did not differ as much from the target. F lege and Hammond (1982) also focused on the relative salience of various phonological patterns in non-native speech. Using a delayed mimicry paradigm. they asked native English speakers enrolled in first-year Spanish classes to read sentences in a “typical Spanish accent.” The type of sound substitutions used by the students was analyzed. The students used several sound substitutions that are typically heard in Spanish accented English. but some were used more than others. The relative ranking of the sound substitutions used suggested that some substitutions were more salient than others. However. the delayed mimicry paradigm is based on several assumptions. These 46 assumptions pertained to the folk (i.e.. non-linguist) awareness oflinguistic variation. as discussed by Preston (1996a). He asserted that the awareness of dialect features by non- linguists fell along four dimensions: (a) availability. (b) accuracy. (c) detail. and (d) control. Availability concerns the degree to which a given person might attend to linguistic details. He/she may range from complete oblivion to a preoccupation with a linguistic event. Accuracy refers to the level at which the person can correctly convey the existence of dialect features. This dimension ranges from absolute accuracy to complete inaccuracy. Detail regards the degree to which the person is able to refer to specific elements of a language variety. This scale ranges from a general awareness of the linguistic phenomenon to the ability to cite specific details about it. C ontrol concerns the person’s ability to imitate the dialect. Listeners may have absolute. partial. or no imitative control over production ofa given dialect. An ability to imitate. even accurately. for example. might not indicate a higher setting on the other scales of awareness. To summarize. evidence supports the notion that some linguistic patterns may be more salient to listeners than are others. While the number of dialect features in a unit of language may affect the degree to which a dialect is perceived. it does not entirely explain variable perceptual judgments. Instead. the specific types of features presented must be considered. Furthermore. a linear. ratio-scaled relationship between the number of features present and perceptual judgments should not be assumed. The determination of which features are likely to be detected more than others may be based on perceptual salience (i.e.. the degree to which listeners can detect a feature in non-meaningful speech). Recent research on optimality theory of phonology 47 suggests that. cross-linguistically, speakers tend to use phonological alternations that are the least perceptually distant from an intended phonemic prototype. This is done while trying to obey constraints within the phonological system (Steriade. 2004). In optimality theory. constraints are ranked according to importance as determined from the language community (Prince and Smolenski.l993). They involve competing desires for faithfulness to a prototype and reluctance to use marked features. For instance. if a speaker has a constraint against final voiced obstruents (a relatively marked feature) and produces the word head. then lie/she will most likely satisfy this constraint by producing /bit/. because the /t/ has the least perceptual distance from /d/ (the intended prototype). when compared to other options (e.g.. consonant deletion). Steriade‘s (2004) framework for evaluating perceptual distance between phonological alternations is called the “P-map.” The P-map is an abstract representation of the perceptual distance between phonetic forms. Steriade discusses how the P-map includes phonological repairs to satisfy the constraint against final voiced obstruents common in human languages. Several types of repairs are: epenthesis (inserting a vowel after the final voiced obstruent; /bid/—> /bidi/). deleting the final consonant altogether (/bid/—-> /bi/) . or changing the obstruent into a continuant (lbidl—r /biz/). Yet. a fourth option. devoicing the final consonant (lbid/-+ /bit/)_. was the option used by most of the .world's languages to satisfy this constraint. Furthermore. Steriade showed that this outcome was perceptually the least noticeable when compared to other processes to avoid a final voiced obstruent. The P-map is a theoretical model. and a complete representation of the relative perceptual salience of all possible phonological changes in all phonetic environments 48 does not yet exist. Steriade. however. did provide some guidance in determining relative perceptual salience between phonological patterns. She suggested judgments of similarity or degree ofconfusability could be used. Steriade asserted that the phonological patterns used to satisfy phonological constraints by most of the world‘s languages are the ones that are the least perceptually salient. This assumption has important implications for the present study. Dialect features with the greatest perceptual salience are not likely to be used widely by the speakers of AAE. Therefore. they could either increase the detectability ofthe dialect by virtue of their perceptual salience or be ignored as random anomalies by virtue of their rarity in AAE speakers. Conversely. if a feature has low perceptual salience. it may not be readily detected by listeners. or it may be identified as characteristically iconic of AAE due to its ubiquitous nature among speakers of the dialect. It is unknown how these possibilities will reveal themselves in the perception of dialect detectability. Irvine (2001) alluded to these possibilities in relation to the power of specific dialect features to trigger attitudes about speakers. He suggested that some features can immediately register the perception of the dialect. Such features may be used more frequently by speakers of the dialect and become semiotically linked in the minds of the listeners. In terms of the perception of dialect detectability. a feature with low perceptual salience may contribute greatly to ratings ofdialect detectability just because it is readily identifiable as a frequently used feature of the dialect. Research Questions Three factors were considered in the study of perceived dialect density: (a) the listeners. (b) the speakers. and (c) the dialect patterns used. In this investigation. the 49 listeners were Anglo-European American SLPs from predominantly Anglo-European American school districts. The speakers were native AAE speakers. The sentences spoken included phonological patterns affecting word-final consonants that vary in both number and type. The research was guided by the following general question: Are Anglo-European American SLPs‘ perceptions of dialect detectability and comprehensibility influenced by the number and type of AAE phonological features present within the sentence? The perception ofdialect detectability and the perception ofcomprehensibility are two separate issues. that were studied separately. Therefore. the general question involves the following two specific questions: (a) Are Anglo-European American SLPs’ perceptions of dialect detectability influenced by the number and type of AAE phonological features included in the sentence? (b) Are Anglo-European American SLPs’ perceptions of comprehensihility influenced by the number and type of AAE phonological features included in the sentence? While the perception of dialect detectability and comprehensibility will be analyzed separately. they have similar hypotheses. It was hypothesized that the sentences with the greater number of features would be judged as having more detectable use of AAE dialect and reduced perceived comprehensibility. It is also hypothesized that the inclusion of perceptually salient features will increase dialect detectability and decrease comprehensibility. CHAPTER 11 METHOD The experiment was designed to determine the impact of selected AAE phonological features on perceptions of comprehensibility and dialect detectability by SLPs in predominantly Anglo—European American public school districts. It examined the effects that would be attributed to the number of features present. type of features present. or a combination of both aspects. Two categories of features. less salient and more salient. were explored. with varied frequencies in sentences. The two categories of perceptual salience were based on results of a preliminary study to be described briefly in this section (see Appendix A for a detailed description). The overall aim of the dissertation‘s main study was to answer the following question: Is the perceived degree of comprehensibility and dialect detectability influenced by: (a) the number of AAE features contained in a sentence and (b) their relative perceptual salience grouping? Participants Characteristics The participants in this study consisted of 16 public school SLPs. They completed a brief questionnaire that probed their conversational experience with AAE. their experience with treating AAE-speaking clients. their own personal use of AAE. and their number of years experience as an SLP. Total years ofexperience ranged from 3 to 33 years (M: 11.88. SD = 8.77). To qualify for participation in the study. the SLPs had to indicate that (a) they had experienced conversational exposure to AAE less than 5% of the time. (b) they had had fewer than three clients within the last 3 years that spoke AAE. (0) they. themselves. did 51 not speak AAE. and (d) they had worked as an SLP for at least 2 years. Each participant also passed an air conduction hearing screening at 25db HL for 1000. 2000. and 4000Hz. The investigator administered the screening using a calibrated Beltone portable audiometer (model # 119) with bilateral earphones in a quiet school office setting. Recruitment The participants were recruited from school districts in Ionia and Eaton counties. two adjacent rural mid-Michigan counties. The school districts had student populations that were 90% Anglo-European American and less than 5% African American. as indicated by the county’s census from the preceding school year 2003 (Amrie E. Casey Foundation). The investigator attended a regularly scheduled. Ionia county-wide SLP meeting to recruit participants. During the meeting. the study’s procedures were described. but the variables and hypotheses remained confidential. A sign-up form was distributed to collect contact information from those SLPs that were present. Following the meeting. some SLPs provided contact information for fellow SLPs that were not present as well as contact information for potentially interested SLPs in adjacent counties. The SLPs were contacted via telephone. email. and/or in person for scheduling. They were compensated $15 for their time. In total. 14 SLP participants represented seven different school districts in Ionia county. and 2 SLPs represented two different school districts in Eaton county. Human Subjects Approval The study was approved by the Michigan State University Internal Review Board as acceptable with regard to the protection of participant rights and privacy prior to engaging participants in the experiment (see Appendix C). General Description of the Stimuli The independent variables in this study were (a) frequency of AAE features included. and (b) relative salience of AAE features included. The stimulus set consisted of 18 different sentence wordings selected to feature various combinations of these two variables (see Table 1a). Word combinations varied only in the subject-noun and object of preposition positions. The words of the sentence recordings were produced with different pronunciation patterns. as follows: Six different sentences had no AAE phonological features. six had one AAE feature. and six had three AAE features (see Table la/b). A more detailed description of the sentences and phonological patterns appears on pages 59-61. In total. the participants made 42 ratings: 36 ratings for the dialect detectability task and 6 ratings for the comprehensibility task. The 18 items in Table 1b were presented twice. During one presentation. they were spoken by one AAE speaker. and during the other presentation they were spoken by another AAE speaker (both in randomized orderings). creating a total of 36 stimulus sentences. The participants rated each of the 36 sentences individually during the dialect detectability task and rated the 36 sentences presented in 6 groups of 6 items each (see Table 2) during the comprehensibility task. 53 Table 1a Listing of Stimulus Sentences and Targets for Planned Alternations Sentences and distribution of phonological processes. Highlighted cells represent planned phonological features. Phonological Features Present: F Dk=F inal Deletion of /k/. FDt=Final Deletion of /t/. FRk=Final Cluster Reduction of sk. FRt=Final Cluster Reduction of st. t/th=Substitution of [t] for /0/. and f/th=Substitution of [f] for /0/. Percep- tual Phonolog Salience i-cal No. of of Features Sent. Adverb Features Features Present No. Det Noun Verb Prep. Det Obj -ial may Mon- 0 More None 1 A sock be with my teeth day may Mon- 0 More None 2 A cat be with my sock day may Mon- 0 More None 3 A tooth be with my boat day may Mon- 0 Less None 4 A desk be with my teeth day may Mon- 0 Less None 5 A guest be with my desk day . may Mon- 0 Less None 6 A tooth be with my list day may Mon- 1 More FDk 1 A sock be with my teeth daL may Mon- 1 More FDt 2 A cat be with my sock day may Mon- 1 More t/th 3 A tooth be with my boat day may Mon- 1 Less FRk 4 A desk be with my teeth day may Mon- 1 Less FRt 5 A guest be with my desk day may Mon- 1 Less f/th 6 A tooth be with m list da FDk, t/tn, may Mon- 3 More t/th 1 A sock be with my teeth day FDt, t/th, may Mon- 3 More FDk 2 A cat be with my sock day t/th, t/th, may Mon- 3 More FDt 3 A tooth be with my boat day FRk, f/th, may Mon- 3 Less f/th 4 A desk be with m teeth day FRt. f/th, may Mon- 3 Less FRk 5 A guest be with my desk day f/th, f/th, may Mon- 3 Less FRt 6 A tooth be with my list day 54 Table lb Listing of Stimulus Sentences and Targets for Planned Alternations Phonetically Transcribed Phonological Features Present: F Dk=F inal Deletion of fk/. F Dt=F inal Deletion of /t/. FRk=Final Cluster Reduction of sk. FRt=Final Cluster Reduction of st. t/th=Substitution of [t] for /S/. and f/th=Substitution of [I] for /S/. Percep- tual Phono- No. of Salience logical Ad- Fea- of Features Sent. verbi tures Features Present No. Det Noun Verb Prep Det Obj al me mAn- 0 More None 1 3 80k bl W19 maI ”£19 de a me WIS maI mm- 0 More None 2 kaet bi SCIk de 3 me WIS maI mAn- 0 More None 3 ”£119 bi bot de a me WIS maI mAn- 0 Less None 4 dSSk bi tiS de a me WIS maI mAn- 0 Less None 5 gSSt bl desk de 3 me WIS maI lIst mAn- 0 Less None 6 we b' de a me WIS maI mAn- 1 More F Bk 1 SG bi tiS de 8 me WIS maI mAn- 1 More FDt 2 Re bi sak de a me WIS maI mAn- 1 More t/th 3 tut bi bot de 3 me WIS maI mm- 1 Less FRk 4 dSS bi tiS de a me WIS maI mm- 1 Less F Rt 5 988 bi desk de 3 me WIS maI lIst mAn- 1 Less f/th 6 tur I" de FDk. tlth, a me WIS maI mAn- I 3 More t/th 1 sq bi tit de FDt, tlth, a me WIS maI mAn- 3 More FDk 2 1% bi SCI de tlth, tlth, 9 me WIS maI mAn- 3 More FDt 3 tut bi b0 de FRk, f/th, 3 me WIS maI mm- 3 Less f/th 4 des bi t i 1” de F Rt, f/th, 8 me WIS maI mAn- 3 Less F Rk 5 988 bi des de flth, flth, a me WIS maI mAn- 3 Less F Rt 6 tuf bi lIs de Table 2 Characteristics of Six Groups of Sentences Presented During Comprehensibility Ratings Stimulus Group Salience Frequency 1 Less 0 2 Less 1 3 Less 3 4 More 0 5 More 1 6 More 3 Note: The 6 groups contained the sentences listed in Tables 1a and 1b organized by perceptual salience and frequency (number of features). Construction ofthe Stimuli Number ofFea/ures Included The number of dialect features (i.e.. frequency) was represented by three groups of stimulus sentences: (a) six sentences with no AAE features. (b) six sentences with one AAE feature. and (c) six sentences with three AAE features. When one feature was present. it always occurred in a word located in the subject-noun position. When three features were present. they occurred in words located in the subject-noun, the preposition. and the prepositional object-noun positions. Words containing AAE features are represented in the shaded cells in Tables 1a and 1b. Specific reasons for the three frequencies tested were. as follows: (a) the zero- feature level was used as a control condition in which all final consonants were 56 pronounced in accordance with a standard dialect. (by) the one-feature level was used to examine the individual effect of each phonological feature on the dialect detectability ratings. and (c) the three-feature level was used to provide an extreme condition in which a sentence contained AAE patterns in all positions of opportunity (i.e.. a condition in which every opportunity for AAE consonant phonological features was manifested). Ttpe of Dialect Features Included Phonological patterns. The types ofdialect features included in the stimulus sentences consisted of the phonological patterns presented in Table 3. These six specific phonological patterns represented the three most frequently occurring consonantal patterns used by African American children in a study by Craig and colleagues (2003): (a) “th” substitutions. (b) cluster reduction. and (c) final consonant deletion. respectively. Patterns involving /t/ sounds and /k/ sounds were placed in separate categories because /t/ has been cited within literature as a commonly deleted phoneme in many American English dialects (Guy. 1980; Stockman. 2006) while /k/ deletion is not as commonly documented across dialects. The t/th and f/th substitution patterns were also separately categorized due to the fact that they are distinct "th" substitution patterns variably found in AAE (Pollock. et al. 1998). 57 Table 3 AAE Phonological Patterns Included in the Stimuli Phonological Pattern Abbreviation Used Example Final Deletion of /t/ FDt "Cat” /r<®t/ --> /r<(~)/ Final Deletion Of /I(/ FDR “Duck” /d pk] --> /d (.7/ Final Cluster Reduction FRt "List” /?rlot/ --> / Mo/ of /st/ Final Cluster Reduction FRk “Desk" /580'K/ --> of /sk/ /680/ Final t/th Substitution t/th “Teeth” ltrT/ --> /‘L't‘t'/ Final f/th Substitution f/th “Teeth” /nT/ --> /n¢/ Perceptual salience groupings. These six phonological patterns were categorized into two salience groups. This was done in order to combine different sound patterns in individual sentences while continuing to have a systematic basis for comparison. Evidence from Labov (1966) suggested that if the same phonological feature is perceived multiple times. then a cumulative perceptual effect could result. In terms of Johansson‘s (1978) model of feature weighting discussed earlier. a phonological feature may become increasingly “irritating” to the listener if used repeatedly. Such an effect could increase dialect detectability. Although this was a valid point to consider. it was beyond the scope of this study. Categorization allowed different features to be used within single sentences. thus counteracting any effects from a sole feature being experienced repeatedly in single sentences. Perceptual salience of the sound patterns in non-symbolic syllables was used to group the phonological patterns. As discussed previously. evidence from Steriade (2004) indicated that perceptual salience might be linked to how noticeable a feature would be in 58 running speech and how frequently the same feature would be likely to occur across languages and dialects. The relative perceptual salience of the phonological patterns under investigation was determined from a preliminary experiment. The preliminary experiment is described in detail in Appendix A. In summary. Anglo-European SLP graduate students were presented pairs of two- syllable. non-symbolic units (i.e.. two-syllable nonsense words) that simulated the phonological environment of the variable subject-noun followed by the verb complex “may be” used in the main study’s stimulil. When the two syllable units in the pairs were different. the difference simulated the phonetic difference that occurred in the AAE phonological patterns to be studied. For example. to examine the perceptual salience of final /t/ deletion. the contrasted syllables were [dat me] and [da me]. The [dat] in the first syllable of the first unit simulates a subject-noun with a final consonant /t/. The [da] in the first syllable of the second unit simulates the same word in which the final /t/ was deleted. The [me] in the second syllable of both units represents the first syllable of the verb complex “may be” that immediately followed the subject—nouns in the main study. Participants rated each pair on a 4-point scale (1: exactly the same. 2= almost the same. 3= significantly different. and 4= extremely different). Some pairs of nonsymbolic units were identical (e. g.. [dat me] — [dat me]'). These were considered control pairs and the ratings for these were subtracted from the pairs containing different units (i.e. the 1 Two other phonological features were also tested in the preliminary study. postvocalic lr/ deletion (i.e. in which the vowel sound is not changed: four -) /fo/) and vocalization of /r/ (i.e. in which the final /r/ is changed to a schwa: four -) /fo<—>/). The results showed that these two sound changes were both highly salient. However. they did not provide a dichotomy from which differences could be compared in the main study. Additionally. these two features were the only ones that included vowel changes. Further examination of AAE features including lr/ would be better examined in another study. 59 experimental pairs) to get a perceptual salience rating for each phonological pattern to be studied. The perceptual salience averages were ranked for each pattern. The perceptual salience rankings resulted in the saliency groupings presented in Table 4. See Appendix A for a more detailed presentation of the preliminary study‘s results. Table 4 Ranking of Phonological Patterns for Perceptual Salience in Nonsymbolic Units Salience Grou n Ranki Pattern Mean SD More 1 F Dk 1.88 More t/th 1.87 More FDt 1.65 Less FCt 1.48 Less FCk 1.33 Less f/th 0.35 Note: Means were obtained from the preliminary experiment data by subtracting the perceptual rating of control pairs from experimental pairs for each phonological pattern. Description of the Sentences The stimulus sentences consisted of eight words each (see Table la/b). The sentences each consisted of the following constituents: determiner + subject noun + verb complex + preposition + possessive determiner + object noun + final adverbial. For example. in the less salient group. one sentence was “A cat may be with my sock Monday.” In the more salient group. one sentence was “A guest may be with my desk hAondayf’ The sentence constituents met six criteria: (a) The subject detemriner was the Word a to avoid opportunities for dialectal substitutions of d/A in words such as the. this. 60 or these. (b) Only words that were plamred to have a phonological alternate at some point during the sample contained codas. or syllable-final consonants (the subject-noun. preposition. and the prepositional object-noun) were used. In these cases. only the codas were opportunities for AAE phonological features. All other words had a CV syllable structure. (c) The codas always preceded the consonant /m/ to control for coarticulatory effects (McCarthy. 1986). (d) The verb complex included the words “may be” in all of the sentences. This specific verb complex was chosen because: it allowed a grammatical construction without a bound morpheme attached to a verb stem. and it began with a /m/. thereby allowing comparative environments for the preceding coda consonants of the subject-noun (see criterion 3). (e) The verb complex was followed by the preposition with because it was a word that could be used as a planned phonological alternation in both the more and the less salient stimulus groups (e. g.. with-) [wlt] in the more salient group. and wit/19 [wa] in the less salient group). (it) The preposition with was followed by the possessive determiner my because it began with a /m/ to satisfy criterion #3. and it was CV syllable structure (see criterion #2). Monophthongization of the diphong /a1/ is a feature of AAE; however. the production of “my” was kept consistent across all of the sentences. Both speakers produced the diphthong /aI/ in all readings without being instructed to do so. To summarize the above criteria. the stimulus sentences included both variable words and fixed words. The fixed words were: (a) the noun phrase determiner a. (b) the verb may be. (c) the preposition with. (c) the determiner my within a prepositional phrase. and (5) the adverbial Monday. These words together created the carrier phrase A __ may be with my Monday. as presented in Table 1. 61 ' It The variable words included the (a) subject-noun. and (b) prepositional object- noun. The preposition with varied only in pronunciation (i.e. /wlS/. /wIt/. and /wa/). The actual word remained the same in all sentences (see Table 1b. p. 54). Preparation of the Stimulus Recordings Speakers’ characteristics. The speakers were two African American adult females who spoke AAE and had the bidialectal skill to add or take away AAE phonological features upon cue with practice. The speakers were recruited from a university theater program and a university communications program. They both signed a consent form (see Appendix D). To account for speaker-specific effects. two speakers were used. Although it would have been desirable to record both a male and a female speaker given that gender differences could exist. this procedure would not have permitted a gender effect to be isolated from a speaker effect when using just two speakers. Because gender differences were beyond the scope of this study. two speakers of the same gender were used. Speakers" conditions. The speakers were recorded in a quiet room. A Logitech USB desktop microphone (Model #980186-0403) was placed approximately 6 inches from each speaker's mouth. The microphone was elevated so that the speakers could sit in an upright. relaxed position. The speech samples were digitized using the Wavesurfer software. version 1.8.3 (Sjolander & Beskow. 2004). at a sampling rate of 22.050 Hz. The waveform on the computer screen was displayed for the speakers. They were told to watch the waveform and adjust their vocal volume to make the waveform fill the display window but not “clip off.” The investigator also watched the waveform to monitor the sound level. Both speakers were asked to re-record at least one sample because of too little vocal volume. Prior to the recording. each speaker was instructed to consider the context that would be presented to naturalness raters and the participants of the study (see Verification of Naturalness on page 69 for a description of the context). They were encouraged to practice each sentence before recording it. Once a self-deterrnined. comfortable intonation pattern was produced. the speakers were instructed to read each sentence again with the same intonation pattern. They were recorded reading the sentences in this way three times. These productions were referred to as the neutral readings. The neutral readings were used as the phonetically consistent carrier phrases into which variable words were inserted. Only the recordings of the lexically and phonologically consistent words were used from this reading (i.e. the detemriners a and my. verb complex may be. and final adverbial Monday). The resulting intonation pattern was similar between the two speakers. It included primary word stress on the subject- noun and secondary word stress on the object-noun of the preposition (the two variable word positions). Pitch was raised on these two words. while a lower pitch was used on the other words in the sentences. Immediately following the neutral readings. the speakers read the sentences aloud again using the same intonation pattern. This time. they were instructed to pronounce each word making sure that the final consonants were consistent with the word spellings (see Table 1b). These productions were referred to as the SAE readings. The SAE readings were also recorded three times. Finally. the speakers read the sentences again and made the specific AAE phonological changes on the subject-noun. preposition with. and the object-noun of the 63 preposition (see Table 1b). These productions were referred to as the AAE readings. During this portion of the recording session. both speakers stressed the preposition with while attempting to make the required phonological modifications. The speakers were reminded of the intonation used in the neutral readings and then spoke the sentences without stressing the word with. The AAE readings were also recorded three times. The variable words from the SAE readings and the AAE readings were embedded into the appropriate position of the neutral readings. This was done to maintain some consistency in the carrier phrases across the experimental conditions. It was necessary to keep the carrier phrases phonetically consistent to control for any unplanned vowel changes that might occur as a result of the speakers” attempts to produce variable dialect features during the SAE and AAE recording processes. For example. the sentences including three planned alternations (AAE readings) could have included vowels that were more characteristic of AAE than the sentences containing no planned alternations (neutral and SAE readings). These vowel differences could have affected the perceptual judgments. as was demonstrated in the study by Graff and colleagues (1983) discussed earlier. Therefore all of the carrier phrases were obtained from the neutral readings. Constrzustion ofmttutttlness rating stimulus CDs. In total. 54 sentence recordings were obtained from each speaker. The six different sentence wordings were repeated three times in three different styles (neutral. SAE. and AAE). All of the sentence recordings were spectrographically displayed on a computer monitor using the Wavesurfer software (Sjolander & Beskow. 2004) that was used during the audio recording process. The investigator modified the neutral readings from the two speakers by replacing the acoustic signal that corresponded to the subject noun. preposition with. 64 and the object noun of the preposition with a corresponding acoustic signal in either the AAE readings or the SAE readings. Table 5 presents a description of the source (i.e. neutral. SAE. or AAE) from which each constituent of the stimulus sentences was obtained. 65 Table 5 The Recording Source of Each of the Stimulus Sentence Constituents Constituents of Stimulus Sentences Varia- Fixed Variable Fixed Variable Fixed Variable Fixed bility Part of Deter- Subject- Verb Prepo- Deter- Object-noun Adverbial Speech miner noun complex sition miner Word A * may be with my * Monday 0 Neutral SAE Neutral SAE Neutral SAE Neutral (:3 1 Neutral AAE Neutral SAE Neutral SAE Neutral Gun 6 2", 3 Neutral AAE Neutral AAE Neutral AAE Neutral .9 E 5 Z Note: Recording sources: Neutral = the initial recording by the speakers in which no phonological features were dictated. SAE = the second recording by the speakers in which spelling consistent pronunciation was prescribed. and AAE. = the third recording by the speakers in which specific AAE phonological patterns were prescribed. * Lexically variable word. see Table 1a. p. 53. for specifics. Three lexically and phonologically consistent stimulus sentence recordings were created for each sentence pronunciation (see Table lb.) and compiled onto a compact disc as a prototype of the recordings to be used in the naturalness rating C Ds. These Prototypic recordings were played for a female African American university professor and researcher of AAE phonological patterns. Both she and the investigator listened to 66 the prototypes and determined if they were reasonable representations ofthe AAE phonological patterns to be studied. The only modification that resulted from this listening involved Speaker 1. Perhaps because of her desire to pronounce a clear final consonant in the SAE recordings. Speaker 1’s three productions of final /t/ on the word cat were pronounced with a long aspiration duration (M = 821115. SD = 921115). This was considerably longer than the 80th percentile (62.2ms) of 808 conversational /t/ aspiration durations reported in Greenberg. Hollenback. and Ellis“ (1996) study of phonemic parameters produced in conversation. It was also considerably longer than the /t/ durations produced from Speaker 2 (M = 44ms. SD = 6.9ms). The duration of Speaker 2’s /t/ aspirations on cat were similar to the /t/ aspiration durations associated with the 50th percentile (3891115) in the study by Greenberg and colleagues. Presumably because of this uncharacteristically long /t/ duration. the investigator and the second listener perceived an /s/ rather than a /t/. Therefore. the aspiration duration was shortened to 40 ms by deleting approximately 40 ms from the final portion of the aspiration signal. Upon review of the recording. a final aspirated /t/ was perceived by both the investigator and the second listener. The sets of three complete stimulus sentence recordings were randomized via an online randomization program (Dallal. 2003) in two different permutations. The sets of three recordings of each sentence were embedded onto separate slides of a PowerPoint. version 2002. presentation. Verifying Naturalness ofStimulus Sentences Because the sentences were acoustically modified to control for a number of criteria. it was possible that listeners would perceive some of the recordings as unnatural. 67 Therefore. to ensure that the most natural sounding recordings were used for the study. the three recordings per speaker ofeach of the stimulus sentences underwent the following verification process. Naturalness rater clutracteristics. Six adult AAE speakers (3 men and 3 women). who were within the age range of 18-35. functioned as natia'alness raters. Three of the naturalness raters. one male participant and two female participants. were recruited from a university theater course in which the AAE dialect was discussed regularly throughout the semester. The others were also members of the university community. Prior to beginning the ratings. the investigator listened to the speech ofeach of these participants and verified that each spoke AAE. Each participant signed a consent form (see Appendix D) and passed a hearing screening with procedures identical to those described for the listener participants. The naturalness raters were each compensated $10 for their time. Naturalness rating task. Before making their observations. the naturalness raters were informed that the purpose of their task was to select the most natural sounding stimuli to be used in a study examining the perception of language use in African American commtmities. Half of the naturalness raters were presented with one randomization and the others received the remaining permutation. Distribution of the randomization permutations was counterbalanced across gender. The naturalness raters made their judgments in a university computer laboratory. They viewed the PowerPoint presentations on computer screens and listened to the sets of stimuli through earphones. No more than two naturalness raters participated at one time. All naturalness ratings occurred over a three-hour period of time. When two raters were participating simultaneously. they rated different randomization compilations. 68 The naturalness raters made a total of 36 ratings. Each rating was made after the raters listened to all three phonologically and lexically consistent sentence recordings in each group. They listened to each set and selected the one sentence that was the most “natural sounding”. Each rater advanced the slides of the presentation and determined when the recordings would play by clicking a mouse at a self-determined rate. The investigator conducted a brief orientation session prior to listening to the stimuli to familiarize the listeners with how to adjust the volume and advance the slides. At the beginning of the presentation. they were presented with the following context scenario and instructions: Context Scenario Imagine that the sentences you are about to hear were spoken by the mother ofa teenager. The mother is talking on the phone about her upcoming weekend vacation. She and her husband are leaving their teenage son at home. and they suspect that he is planning to have a wild party. She is very concerned that when they return on Monday. their house will be in shambles. She is envisioning what their home will look like when they return. Each screen will contain a sentence that is read three times. Click on each of the three speaker icons until you have listened to all of the sentences on the page. When you have heard all three sentences on the page. identify the one that sounds the most natural (or similar to the way you might actually hear someone say it in daily conversation). Write the letter of your choice on the form [see Appendix E]. You may listen to the sentences as many times as you wish. The whole process should take you 15-30 minutes to complete. Since all of the sentences in the sets were identical with respect to content and planned phonological alternatives. the naturalness raters were forced to base their choices on suprasegrnental and digitization quality differences present among the three versions of each sentence. Naturalness rating results. A complete presentation of the naturalness rating results is included in Appendix F. Every stimulus item was selected by at least half of the 69 participants. and no stimulus item was selected solely by raters from one stimulus compilation. Twenty—eight percent (10/36) of the stimulus sentences was selected by only halfof the naturalness raters. while the remaining 72% (27/36) was selected by over half of the naturalness raters. Analyzed according to speakers. speaker I received consensus of over halfof the naturalness raters on 83% (15/18) ofthe stimulus sentences. and speaker 2 received consensus ofover halfof the naturalness raters on 61% (7/18) of the stimulus sentences. Two stimulus sentences. one for each speaker. received unanimous support from the naturalness raters. Both of the sentences receiving unanimous support from the naturalness raters involved the phonological pattern t/th. There was one tie on stimulus sentence number 31 (speaker 2) In this case. three ofthe naturalness raters chose recording A and three chose recording B. The investigator listened to the three recordings from stimulus sentence number 31. and judged recording B to be the most natural sounding. Randomization and Cotmterlutlancing ofStimulus Sentences The digital files for each recording selected during the naturalness rating process were embedded into slides of PowerPoint presentations. F our different randomized stimulus compilations were nrade. The sentences were randomized and counterbalanced to control for order effects related to: (a) speakers and (b) tasks (see Table 5). 70 Table 6 Speaker Order and Task Order for Each Stimulus Compilation Compilation Speaker Number Order Percegtual Rating Order 1 1-2 Comprehensibility-Detectability 2 2-1 Comprehensibility-Detectabilit 3 2-1 Detectability-Comprehensibility 4 1-2 Detectability-Comprehensibility Speaker order. For the dialect detectability ratings. recordings for the two speakers were presented to the participants one at a time in blocks of 18 sentences (e.g. 18 sentences spoken by one speaker followed by 18 sentences spoken by another speaker). For the comprehensibility ratings. the stimulus sentences were presented on slides six at a time grouped according to salience and number of planned AAE features. Within these groups of six sentences. three recordings from one speaker were placed on the left side of the screen and three from the other speaker were placed on the right side of the screen. The words “Speaker I” appeared over the left set of sentences. and the words “Speaker 2“ appeared over the right set of sentences. The speakers were segregated and clearly delineated in both tasks to control for the processing costs that could be associated with randomly interrnixing the speakers. Pisoni and Lively showed evidence of increased mental processing load expected to be associated with rapidly switching attention to different voices (Pisoni & Lively. 1995). 71 The order ofthe speakers was varied across the compilationsz. Specifically. during the dialect detectability ratings. two of the compilations had eighteen sentences read by speaker I first. and two had eighteen sentences read by speaker 2 first. For the comprehensibility ratings. two of the compilations had speaker I first on every slide. and two had speaker 2 first on every slide. Within each compilation. the speaker order was kept constant between the two rating tasks. That is. if speaker I was first in the detectability presentation. then speaker I would also be presented first in the comprehensibility presentation. Task order. Two of the compilations had the participants rate comprehensibility first. while the remaining two had the participants rate dialect detectability first. This counterbalancing of task order is similar to what was done by Munro and Derwing (1995b). They reported no order effects. In contrast. Gass and Varonis (1984) reported that listeners found sentences to be more intelligible after they became familiar with the speakers and context. This finding suggested that there could be a difference between the groups of listeners who rate comprehensibility first and those who rate comprehensibility last. Sentence randomization. Niedzielski (1999‘) found that listener expectations can affect the way that sentences are perceived. If listeners hear few dialect features in the beginning. they may prej udge the speaker as a non-dialect user. Therefore. the sentences that follow may be perceived differently from the scenario in which many features are 2 The actual speaker order was counterbalanced across the compilations. However. in the comprehensibility rating task the words “Speaker 1" always appeared on the left side of the screen. and the words “Speaker 2“ always appeared on the right side of the screen. regardless of which speaker was actually being played. Likewise. in the dialect detectability rating task. the words “Speaker 1” always preceded the first set of sentences and “Speaker 2” always preceded the second set of sentences. heard in the beginning. To control for effects of presentation order. the sentences read by each speaker were randomized (Dallal. 2003) for each presentation. Since each participant heard the eighteen sentences read twice during the dialect detectability ratings (once by each of the two speakers). there were a total of eight randomized permutations for the dialect detectability ratings (two permutations for each of the four compilations). For the comprehensibility ratings. the items were presented in six groups ofsix sentences each (see Table 2). The six sentence groups contained three sentences read by one speaker and the same three sentences read by the other speaker. The three sentences were randomized within each sub-group. and the groups of six sentences were randomized with separate permutations in the four different stimulus compilations. Collection of Data Presentation ofStimu/i The experiment took place in a public school office space with individual computers at different cubicles. The listener participants attended to a PowerPoint presentation similar to the ones used by the naturalness raters. They viewed the presentations individually on personal computers. No more than four listeners participated at one time over a span of three weeks. Each listener was randomly assigned to one of the four stimulus compilations. such that each compilation was rated by four listeners. They each listened to the stimulus items through earphones. The investigator instructed the participants on adjusting the volume and advancing the slides before they began the presentation. They each indicated their perceptual judgments on two computer-readable “bubble” forms (one for comprehensibility ratings and one for detectability ratings) using a #2 pencil. After completing the tasks. they enclosed the 73 form in an envelope and gave it to the investigator. They did not provide any identifying information on the form. Rating Tasks Two phases of the rating tasks will be described: 1) task orientation and familiarity. and 2) stimulus presentation. Task orientation amljanziliarization. Prior to each rating task. the participants viewed a set of instructional slides that oriented them to the task and context of the sentences (the same context provided to the naturalness raters. presented on page 68). Each rating task was preceded by a familiarity slide. The familiarity slide was designed to provide a consistent point of reference for all of the participants. This was done by allowing them to experience a neutral set of speech patterns (i.e.. no AAE features present) before listening to various pronunciation patterns. For example. if some of the participants were presented with a stimulus sentence containing three highly salient AAE features. then those listeners might perceive the following sentences differently than those presented with fewer highly salient features. Without hearing the sentences devoid of AAE features first. it would have been unclear. in these instances. as to what in their individual repertoire of experiences they were basing their judgments on. By presenting the sentences without any planned AAE features prior to the rating task. it was assumed that when the first item was presented. each listener made herjudgrnents based on the sentences without AAE phonological features present. The familiarization phase also oriented participants to each speaker’s voice to decrease processing load. As discussed previously. if the speaker‘s voice changed 74 without orientation. then there could be increased processing load associated with switching attention to another speaker (Pisoni & Lively. 1995). Thus. the familiarization task also served to clearly delineate to listeners when they were about to hear a different speaker and to familiarize them with the voices they were about to hear. Speaker familiarization was slightly different for the two rating tasks. Prior to the comprehensibility rating task. the participants clicked on a set of icons over which read “Speaker 1.” They then listened (but did not rate) the six sentences read by the first speaker on the compilation that contained no planned AAE phonological features. To the right of the “Speaker 1” icons. they clicked on a set of icons over which read “Speaker 2.” They then listened to all of the sentences read by the other speaker that contained no planned AAE phonological features. The sentences read by the different speakers were clearly labeled as “Speaker 1" and "Speaker 2“ in the comprehensibility ratings that followed. An orthographic transcription ofeach sentence was provided. Prior to rating 18 sentences read by each speaker for the dialect detectability rating task. the participants listened to the same six sentences used for “Speaker 1” familiarization in the comprehensibility rating task. They then continued to rate the 18 sentences read by the first speaker. After rating the 18 sentences read by the first speaker. they then heard the same “Speaker 2” familiarization recordings used in the comprehensibility rating task. They then continued by rating the 18 sentences read by the second. speaker. Prior to the familiarization task. the following instructions were provided in the PowerPoint presentations for the two tasks. Although the following descriptions depict the scenario in which comprehensibility ratings occur first. the two task descriptions were 75 reversed when the detectability ratings occurred first. Comprehensibility. The following instructions were provided to the participants on the PowerPoint presentations prior to rating comprehensibility. Task 1 1. You are about to listen to six groups of six sentences each. Three of the sentences in each group will be read by one speaker. and three will be read by another speaker. 2. After listening to both speakers in each group of sentences. you will rate the speakers regarding how easily understandable you believe they would be to people in the general population. 3. Anytime you see the ‘11 icon. click on it and listen to the audio clip. 4. Click on the arrow on the screen to advance to the next group of sentences. Making Your Ratings - Imagine that you have these two speakers as clients. and you need to determine (based on the overall sound of their speech) how understandable they are likely to be to people in the general population. 0 After you hear all of the sentences spoken by the two speakers. fill in the bubble on the response sheet that corresponds to a letter from A to E (A= Very difficult to understand. E= Very easy to understand). 0 In this task you are rating all of the recordings from the two speakers as a whole. You are not rating the speakers individually. If you feel that there is a difference between how easily understandable the two speakers are. then think of an average rating for the two. 0 Make sure that the number on the form matches the one on the computer screen and make sure that you fill in the bubble completely. - If you need to listen to the sentences again. you may do so as many times as you wish. 0 If you make a mistake and need to go back. right-click somewhere on the screen and select ‘previous.’ Then continue as before. - After you finish rating the sentence groups. you will be asked to rate the same sentences again in another way. But you’ll learn about that later. Additional Information ° The whole experiment (including both ratings) should take about 25-35 minutes to complete. 0 You may stop this experiment at anytime for whatever reason and discontinue your participation in the study without any further questions.” Before the listeners began their ratings. the same scenario given to the naturalness raters (see page 68) was presented to orient them to the lexical and odd 76 pragmatic characteristics of the sentences they were about to hear. The familiarization slide was then presented. After completing the familiarization slide. the listeners continued to the first comprehensibility rating. On each screen. the item number was presented at the top of the page. Below that. a table similar to Figure 4 was shown in which the listeners clicked on each speaker icon to hear the speech recording. Figure 4 Example Stimulus Table for Comprehensibility Ratings Sentence Speaker 1 Speaker 2 A tooth may be with my list Monday. (i 5 (Q A desk may be with my teeth Monday. (if; (E A guest may be with my desk Monday. (E (E A question was displayed beneath the table asking. “How understandable do you think that these speakers would be to people in the general population?" The rating scale ofA through E was presented under the question with “Very difficult to understand” located under the “A” and “Very easy to understand” located under the E. The participants rated the sentences by filling in the appropriate bubbles on a cornputer- readable response sheet. The investigator supervised the participants to ensure that no unexpected problems arose during the comprehensibility ratings. 77 The orthographic transcription of the sentences was included for several reasons. First. the SLP listeners were judging whether the speaker was likely to be understood by the general population. Such a judgment required the SLPs to make their decision based upon what the person was trying to say. Second. the inclusion of the orthographic transcription somewhat replicated relational phonological analysis often used in diagnostic sessions. Third. providing a transcript of the sentences to be rated controlled for the fact that intelligibility is likely to increase as the experience oflistening to a speaker increases. Therefore. later sentences could have had an increase in comprehensibility ratings that were more based on the experience listening to the individual speakers and less based on the presence or absence of AAE phonological patterns. Finally. providing a transcript ensured that the SLPs‘ ratings of projected comprehensibility were not altered because of reduced intelligibility. Due to the fact that the sentences were phonologically controlled. the stimulus sentences were semantically odd. If no transcript was provided. it would be unclear whether the SLPs really understood the intended sentence at all. The ratings obtained could have been for sentences that the listeners thought they were hearing. rather than the actual sentences intended. Thus. the inclusion of an orthographic transcript ensured that the SLPs rated the intended sentence with the dialect features under investigation and that they didn’t rate misinterpreted sentences resulting from unintelligibility. After they listened to all of the sentences. they viewed a screen that read as follows: END OF SECTION ONE. ° Now take a 5-minute Break. 0 You may put your response sheet in the envelope. 0 After 5 minutes. click on the “next" arrow to continue. 78 Dialect Detectability. The following instructions were included for the rating of dialect detectability: Task 2 ° The speakers of the sentences you have heard so far3 speak a dialect called African American English. a dialect that is commonly spoken by many African Americans. 0 You are about to hear the same sentences again. 0 This time you will rate them regarding how noticeable the use of African American English is. 0 You will rate each sentence individually. rather than in groups this time. ’ You will make a total of 36 ratings for this task. Making Your Ratings 1) After listening to each sentence. consider how noticeable the use ofAfrican American English (AAE) is in each sentence. 2) Base your judgment not on the overall message of the sentence. but rather. on how the whole sentence sounds to you. 3) Indicate your responses by filling in a bubble next to each item number that corresponds to a scale from A to E: A: Not noticeable at all. E= Extremely noticeable. Additional Information 1) After you finish making these ratings. you are done with the experiment. 2) However. as previously mentioned. you may stop this experiment at any point. for any reason. without further explanation. and without penalty.” The listeners were then presented with another familiarization slide that oriented them to the sentences that they heard during the detectability ratings and the sound of the speaker's voice that followed. The slides were formatted such that the item number was located at the top of each slide. The orthographic transcription of the sentence was centered on the screen with a speaker icon located below. The rating scale was presented on each slide directly under the question “How noticeable is the dialect of African 3 The words “you have heard so far” were replaced with the words “you are about to hear" when the dialect detectability task was first. Similar modifications were made that between the instructions for the two tasks with respect to their ordering on different compilations. 79 American English in this recording?” The listeners indicated their responses on a different computer-readable sheet. Upon completing the ratings. the listeners were presented with the following slide: “END OF EXPERIMENT. Thank you for participating in this study! Please place your response sheet in the provided envelope and hand it to Greg” Intra-Rater Reliability Four participants (one from each compilation. 25% of the participants) were randomly asked to rate the same stimulus compilation that they had rated previously. The second rating occurred two weeks after the first presentation to determine intra-rater reliability for both tasks. During the second rating. the reliability raters participated in the same environment and on the same computer that they used in the first rating. Data Processing The computerized response sheets were given to University Scoring Office at Michigan State University for processing. The scoring office sent the investigator a data file over electronic mail. The data were imported into Excel 2002 for Windows and SPSS. version 12.0. for analysis. Data Analysis The analysis of the data occurred in two phases: (a) compilation comparisons. (b) task-specific comparisons. Compilation Comparisons The compilations were compared using a one-way Analysis of Variance (ANOVA). This was followed by exhaustive multiple comparisons using paired t-tests to determine the source of the differences among the four different compilations. Type I 80 error rates were controlled using the Bonferroni method. Ratings from identical stimuli among the compilations were matched during the paired t-tests. The overall comparison among the compilations were followed by: (a) a compilation comparison specific to the dialect detectability task and (b) a compilation comparison specific to the comprehensibility task. Similar to the overall compilation comparisons. one-way ANOVAs were used for these analyses. and exhaustive multiple comparisons using Bonferroni paired t-tests followed when differences were detected in the omnibus ANOVA tests. T ask-Specific Comparison of Test Variables Three within-subjects factors were included in the omnibus analyses of the dialect detectability data: (a) speaker (2 levels). (b) frequency ('3 levels). and (c) salience (2 levels). Due to differences in the methods specific to the comprehensibility ratings. the speaker factor was not included. Therefore. the dialect detectability data were analyzed using a parametric 2 x3 x 2 ANOVA. and the comprehensibility data were analyzed using a parametric 3 x 2 ANOVA. Both ANOVAs also included the between-subjects factor compilation. Such analyses were followed by paired t-tests to detect the origin of differences revealed by the ANOVAs. When t-tests were used to compare more than two levels of a variable. the Type I error rates were controlled using the Bonferroni method. Some statisticians have suggested that nonparametric statistics should be used when analyzing ordinal. Likert scale data in lieu of parametric statistics (Schiavetti & Metz. 1997); however. Thomas (1998) demonstrated that parametric statistical analyses could be used reliably with Likert scale data. as was generated by this study. 81 Repeated-measures statistics were used because the comparisons made in this study were most often between specific stimuli. not between distinct groups of participants. Therefore. the comparisons between specific stimuli within and across the compilations were analyzed as within-subjects factors. When differences among the compilations were analyzed. the compilation variable was maintained as a between- subjects factor. This study‘s results were informed by the use oftwo different effect size measures. When the results of a repeated measures ANOVA were reported. the effect size measurement partial eta square (ifp) was used. This measurement is the proportion of total variation in the data attributable to the effect of the variable itself. excluding other factor effects from the total nonerror variation. The partial eta square measure ranges from 0-1 (Pierce. Block. & Aguinis. 2004). When the results of paired t-tests were reported. the effect size measure was C ohen‘s d (Cohen. 1988). This measure is calculated by dividing the mean difference by the standard deviation. C ohen‘s standard was used for interpretation. as follows: .20< d< .50 = “small”. .50.80 = “large”. CHAPTER III RESULTS This dissertation sought to determine ifAnglo-European American SLPs‘ perceptions of dialect detectability and comprehensibility were influenced by the number (frequency) and type (salience) of AAE phonological features present within the sentence. This overall objective was divided into two measurements: (a) dialect detectability. and (b) comprehensibility. The research questions for these two components are: a) Dialect Detectability: Are Anglo-European American SLPs’ perceptions of dialect detectability influenced by the number and type of AAE phonological features included in the sentence? b) Comprehensibility: Are Anglo European American SLPs’ perceptions of comprehensibility influenced by the number and type ofAAE phonological features included in the sentence? This section of the dissertation begins by reporting intra-rater reliability for randomly selected participants. Next. a comparison follows to report differences among the four different stimulus compilations. Detection ofdifferences among the compilations was required to determine how the compilations may be pooled for the subsequent analyses of the experimental independent variables. Effects specific to the dependent variables. dialect detectability and conzprehensibility. are reported as the the principal findings. 83 Intra-rater Reliability One participant from each stimulus compilation was randomly selected to complete her ratings again. Each reliability participant rated all items a second time two weeks after the first time on the same computer and in the same enviromnent. The ratings associated with the 42 individual stimulus items for each of four reliability participants were matched for time 1 and time 2. The ratings were analyzed for association between time I and time 2 using a paired samples correlation test. Hayes and Hatch (1999) argued that using correlation measures for reliability testing may be superior to using percentage of agreement because: (a) percentage of agreement does not automatically consider effects due to chance. and (b) chance agreements are strongly influenced by data variability. data plateaus. and by scoring of near agreement as absolute agreement. The item-matched correlation comparison resulted in an N equal to 168 for the comparison over both tasks (42 stimulus sentences times 4 participants). When the ratings specific to dialect detectability and comprehensibility were analyzed together. the test-retest reliability ratings were highly correlated. r(168)= .90. p<.001. The analysis revealed that 8 % of the variance was accounted for by the association between time 1 and time 2 ratings. r3: .81. In a follow-up reliability analysis. separate item-matched correlation comparisons were preformed specific to dialect detectability ratings and comprehensibility ratings. In the dialect detectability reliability analysis. only the 36 dialect detectability ratings generated by each of the 4 reliability participants were included. reducing the N to from 168 to 144. Time 1 and time 2 measures specific to the dialect detectability ratings were 84 also highly correlated. r( 144) = .92. p<.001. This analysis revealed that 85% of the variance was accounted for by the association between time 1 and time 2 ratings. r3: .85. In the reliability analysis of comprehensibility. only the 6 comprehensibility ratings for each of the 4 participants were included. resulting in an N: 24. Time 1 and time 2 comprehensibility ratings were significantly. but not as highly correlated as the detectability ratings. r(24.)= .740. p<.001. This analysis revealed that 5 % of the variance was accounted for by the association between time 1 and time 2 ratings. r3: .55. Due to the reduced reliability revealed in the comprehensibility data. the ratings were analyzed for difference using a paired t-test. This analysis showed that the ratings were not significantly different. t(23)=-.419. p=.679. Compilation Comparisons Compilation was the only between-subjects factor in this study. Therefore. each of the four compilations included four of the sixteen total participants. Before analysis of the independent variables could be done. it was necessary to determine if there were substantive differences among the ratings elicited by the four compilations. If the ratings were similar among the compilations. then the data associated with the different compilations could be pooled in the subsequent analyses. If any of the compilations elicited statistically different ratings for either of the two tasks. then only those statistically similar could be pooled in the subsequent analyses. Compilation differences were analyzed: (a) over both tasks. (b) specific to comprehensibility ratings. and (c) specific to dialect detectability. 85 Compilation Comparisons Over Both Tasks A one-way ANOVA was used to compare the data elicited by the four different compilations. The ratings for the 42 individual stimulus items in each of the different compilations were compared for the 16 subjects (N = 672). This omnibus analysis revealed significant differences among the four stimulus compilations. F (3. 668') = 4.272. .4. p = .005. Due to the significance of this test. a multiple comparison analysis was used to detect the origin of the differences. In this analysis. the 42 ratings from the four participants in each compilation (N—=l68) were matched according to item similarity. The multiple comparisons analyses revealed that the ratings for three of six comparisons were significantly different. Table 7 shows the mean difference between the ratings for each paired compilation analysis with the standard deviation of the differences. The t-score is displayed with its probability value and Cohen's d. The Bonferroni technique was used to reduce the alpha level from .05 to .008 to control for Type I error. The differences revealed by these analyses suggested that the characteristics of different compilations (i.e.. task ordering and/or speaker ordering) may have influenced the ratings of comprehensibility and/or detectability. The ratings for comprehensibility elicited by each of the compilations were then analyzed for differences separately from the ratings for dialect detectability. 86 Table 7 Differences Among the Four Compilations Over Both Perceptual Rating Tasks (Four different listener ratings for 42 stimulus sentences in each compilation. N=l68) t(l67)= 2988*. p = .003. d= .23. small effect size Stimulus Compilation 2 Compilation 3 Compilation 4 Compilation 2-1. C-D 2-1. D-C 1-2. D-C Compilation 1 Md=.530. SDd= 1.07 Md: .208. SDd=1.52 Md: .393. SDd=1.30 1-2. C-D t( 167) = -6.406*. t(l67)= -1 .781. t(l67)= -3.918*. p<.001. d= .49 small p = .077 (ns. p<.001. d= .30. effect size p>.008). small effect size cl: .14 Compilation 2 Md= .321. SDd= Md: .137. SDd=l .12 2-1. C-D 1.40 t(l67)= 1.583 (ns). p = .115 (ns. p>.008_). d= .12 Compilation 3 2-1. D-C Aid: .185. SDtl=1.77 t(l67)= -l .354 (ns). p = .178 (ns. p>.008). d= .10 Abbreviations: Md= Mean Difference. SDd= Standard Deviation of the Difference. C = Comprehensibility Task. D= Dialect Detectability Task. 1: Speaker 1. 2= Speaker 2. Compilation labels denote task order followed by speaker order (e.g. 1-2. C-D = Speaker 1 followed by Speaker 2. C omprehensibility followed by Dialect Detectability) * Differences are significant at alpha of .05 (p<.008 Bonferroni adjustment). Compilation Comparisons Specific to Comprehensibility For the compilation comparison specific to comprehensibility. a one-way ANOVA was performed with the four-level. between-subjects factor ofcompilation. In this case. only the six ratings generated during the comprehensibility task were included from each of the 16 participants (N=96). The four compilations did not elicit statistically 87 different ratings for comprehensibility. F ( 3. 92): 2.462. [F .068. This indicated that the differences revealed in the overall comparison of the compilations originated primarily in the dialect detectability data. Compilation Comparisons Specific to Dialect Detectability For the analysis of the compilation differences associated with the dialect detectability data. a one-way ANOVA with the four-level. between-subj ects factor compilation was used. The 36 ratings from each of the 16 participants were analyzed across the four compilations (N= 576). The dialect detectability ratings among the four compilations were statistically different. F (3. 572) = 5.246. p = .001. To determine the origin of the difference. an exhaustive multiple comparisons analysis was done using paired t-tests. matched for item similarity across compilations. Table 8 shows the mean differences ofthe dialect detectability ratings between the various compilations with the corresponding standard deviations of the differences. The t-scores are reported with the probability values and C ohen‘s d scores. Although some of the compilation differences had p-values less than .05. the Bonferroni-adj usted alpha is .008 to control for Type I error. Therefore. only those t-scores with corresponding probabilities less than .008 are considered significant. The summarized data in Table 8 shows that compilation l was statistically different from both 2 and 4. but not statistically different from compilation 3. Interestingly. compilations 2 and 4 were very similar. both eliciting the same mean dialect detectability score. While the mean scores in compilations 1 and 3 were slightly (not significantly) different. the standard deviations for the two were the same. 88 Table 8 Differences Among the Four Compilations Specific to the Dialect Detectability Task (Four different listener ratings for 36 stimulus sentences in each compilation. N=144) t(l43)= -6.515*. p<.001. d= .54 “medium” effect size t(143)= -2.301. p=.023 (ns. p>.008) d= .19 Stimulus Compilation 2 Compilation 3 Compilation 4 Compilation 2-1. C-D 2-1. D-C 1-2. D-C Compilation 1 Md; .576. SDd= Mcfi .285. SDd= Md: .576. SD: 1.24 1-2. C-D 1.06 1.49 t(143)= -5.562*. p<.001. d= .46 "small” effect size Compilation 2 2-1. C -D Md= .292. SD: 1.44 ((143): 2.434 (ns). p = .016 (ns. p>.008). d= .20 Md: .000. SD: 1.10 t (143): .000. p = 1.00 (ns. p>.008). (1:.00 Compilation 3 2-1. D-C Md: .292. 1.79 t(143)= -1.961 (ns). p = .052 (ns. p>.008). d: .16 Abbreviations: Md: Mean Difference. SDd= Standard Deviation ofthe Difference. C = C omprehensibility Task. D= Dialect Detectability Task. l= Speaker 1. 2= Speaker 2. Compilation labels denote task order followed by speaker order (e.g. 1-2. C -D = Speaker 1 followed by Speaker 2. C omprehensibility followed by Dialect Detectability) * Differences are significant at alpha of .05 (p<.008 Bonferroni adjustment). Since the compilations were counterbalanced according to speaker order and task order. it was possible that one or both of these factors caused different dialect detectability ratings. The compilations were ranked according to their mean detectability ratings. and the characteristics of each compilation were presented alongside the ranking 89 (see Table 9.). The ranking revealed that neither speaker order nor task order had a consistent effect. yet an interaction could be inferred between these two variables. Task order was significant when speaker I was presented first (detectability-comprehensibility > comprehensibility-detectability). Speaker order was significant when comprehensibility ratings were first (speaker order 2-1 > speaker order 1-2). No . . . . . . . . . 4 theoretical justification for such an interaction could be interred . Table 9 Compilation Characteristics and Rank Ordering According to Average Dialect Detectability Ratings Rank Compilation Speaker Mean N Number Order Percefltual Rating Order Detectability SD 1 2 2-1 Comprehensibility-Detectability 3.22 1.47 144 2 4 1-2 Detectability Comprehensibility 3.22 1.57 144 3 3 2-1 Detectability-ComprehensibilitL 2.93 1.37 144 4 1 1-2 Comprehensibility-Deteetability 2.65 1.37 144 (.‘ompilt‘ttion Groupings Resulting Front (.‘ompilc‘ttion Differences Although the reason for the differences across compilations was not determined. the data associated with the four compilations were grouped according to statistical 4 It is possible that despite the use ofcontrolled exposure to phonologically consistent recordings across the compilations and randomized orderings. stimulus ordering effects occurred nonetheless. To test the possibility that early exposure to minimally or maximally detectable features influenced the ratings for the rest of the sample. the first five stimuli presented in the detectability section of each compilation were analyzed for differences. The detectability ratings were not statistically different within the first five stimulus items across groups. F (3. 76): .353. p= .787. The stimuli were further analyzed qualitatively. The first five stimulus items occurring on each compilation were coded regarding the frequency of features contained in each item and the relative salience of the features. There were no consistent qualitative differences in the groupings of stimuli pertaining to frequency or salience. 90 similarity. To determine statistical similarity. a hierarchical cluster analysis was used. The squared Euclidian distance between the average dialect detectability ratings created two groups. Compilations 2 and 4 received the highest ratings and comprised one group; compilations 1 and 3 received the lowest ratings and comprised a second group. Compilations 2 and 4 were designated as the high detectability group: compilations l and 3 were designated as the low detectability group. These two groups of 288 ratings (36 ratings from eight participants) were analyzed separately during the analysis of dialect detectability. In the separate analyses for these two data groupings. the between-subjects factor of stimulus compilation had two levels rather than four (N=144 for each level). As stated previously. no significant differences were detected among the compilations in the comprehensibility data. Therefore. the six comprehensibility ratings generated from all 16 participants were analyzed together (96 total ratings). Thus. the between-subject factor of compilation had four levels in the comprehensibility analysis (N=24 for each level). Principal Findings: Dialect Detectability Ratings Description of F actors Does the number ofAAEt/‘eatru'es and/or the perceptual salience of the AAE features in a stin-zulus sentence affect the degree to which the listeners detect an AAE dialect? The analysis used to answer this research question involved two methodological variables (compilation and speaker) and two experimental variables (frequency and salience). Of the two methodological variables. compilation was a between-subjects factor and speaker was a within-subjects factor. Both of the experimental variables were within—subj ects factors. 91 The methodological. within-subj ects factor speaker (2 levels) referred to the two different speakers that were included in the study. These two speakers have been labeled speaker I and speaker 2. The experimental variables. frequency (3 levels) and salience (2 levels). were both within-subjects factors. Frequency referred to the number of dialect featm‘es that were included in each of the stimulus sentences (0. 1. or 3.). Salience referred to the relative perceptual salience of the features included in the stimulus sentence (low or high). As described in Chapter II. phonological features were grouped into categories of low salience and high salience according to the results of the preliminary experiment examining the perceptual salience of the phonological patterns presented in nonsymbolic syllables (see Appendix A). The three within-subj ects factors described above yielded a 2 x 3 x 2 within- subjects ANOVA. with a two-level between-subjects factor. compilation. ANOVAs of this type were used as omnibus tests to detect differences and interactions among these four factors in both the high detectability group analysis and the low detectability group analyses. Due to the fact that each participant received the same stimulus items. the within-subjects comparisons in the analyses were between stimulus items. not between participants. When differences were revealed by the omnibus test. a paired t-test was done that matched similar items between comparison stimulus groups. In which case. if more than two groups were compared. the alpha was adjusted using the Bonferroni method. The following dialect detectability findings are organized as such: (a) high detectability group results. (b) low detectability group results. (c) post hoc comparison of features. and (d) summary of dialect detectability results. High Detectability Group Results Frequency effects. As hypothesized there was a significant main effect for frequency in the high detectability group. F (2. 22)= 155.057. p<.001. indicating that as the number of AAE features increased. so did the dialect detectability ratings. This effect is displayed graphically in Figure 5. This effect was large. ifp= .876. indicating that almost 90% of the total effect was accounted for by the variable itself. Table 10 shows the differences between the average dialect detectability ratings analyzed for the three frequency levels. As hypothesized. the stimuli containing three AAE features were rated as more detectable than stimuli containing just one feature. t(95) = -9.3 56. p<.001. Stimuli containing one feature were significantly more AAE-detectable than were those containing no features. t(95)= -11.675. p<.001. While both comparisons had large effect sizes. the effect of the 0-1 frequency difference (d= 1.19) was larger than the effect of the 1-3 frequency difference (d= .96). There was not a significant frequency-speaker interaction. F (2. 44) = 1.254. p = .294. indicating that the frequency effect was consistent across the speakers. Likewise. there was not a significant frequency-compilation interaction. F (2. 44) = .177. p: .839. indicating that the frequency effect did not differ significantly across the two compilations in the high detectability group. 93 Figure 5 Average Dialect Detectability Scores in the High Detectability Group Obtained in Relation to the Frequency and Salience of AAE Features (N=l6) 01 I .p 1 Average Detectability Ratings '1’ ‘t’ T l l O 1 l 3 Number of AAE Features Salience — Low - - - High Dialect Detectability Rating Scale: I: Not noticeable at all. 5: Extremely noticeable 94 Table 10 Sentences in the High Detectability Group With 0. 1. and 3 Features Analyzed for Dialect Detectability (N= 96) Frequency = 0 Frequency = 1 Frequency = 3 M: 1.71. M: 3.45. M: 4.51. SD= 1.035 SD= 1.204 SD= .649 Difference 0-1 Difference 1-3 t(95) = -9.356 t(95)= -11.675 p<.001. d= 1.19 p<.001. d= .96 Salience effl’cts. Although there was a significant main effect for salience. F (1. 22) = 8.008. 1): .01 . in the high detectability group (see Figure 5). the effect is qualified by a significant salience-speaker interaction. In other words. the salience effect varied for the two different speakers. F (1. 22)= 20.129. p<.001. Specifically. speaker 2 elicited a significant salience effect. t(_7l) = -4.512. p<.001. with a medium effect size. d= .53. while speaker I did not elicit a significant salience effect. t(71) = .462. p = .645. There was not a significant salience-compilation interaction. F (1. 22)= .249. p=.622. F requency-sal ience interaction effects. There was no interaction between the frequency and salience variables in the high detectability group. F (2. 44)= .740. p=.483. Low Detectability Group Results Frequency effects. There was a significant main effect for frequency in the low detectability group. F (2. 44)= 159.135. p<.001. This effect was qualified by a significant frequency-speaker interaction. F (2. 44)= 6.306. p= .004. and a significant frequency- cornpilation interaction. F (I. 22)= 6.709. p=.003. These results indicated that the 95 frequency effects were different between the two speakers and the two compilations in the low detectability group. Figure 6 illustrates that even though there were methodological interactions between frequency and the two methodological variables. they did not affect the overall direction of the effects. In all cases. increases in frequency of AAE features resulted in increases in dialect detectability scores. To determine the origin ofthe frequency—speaker and fi'equency-compilation interactions. subsequent analyses were done. F irst. the analyses related to the frequency— speaker interaction are reported. Then the analyses related to the frequency-compilation interaction are reported. To analyze how the frequency effects by the two different speakers differed. the 12 ratings from the 8 participants elicited by the 2 different speakers in the low detectability group were analyzed separately (N= 96). These two speaker-specific analyses were done using two separate. 2 x 3 ANOVAs which included the within- subjects factors of salience and frequency and excluded the between-subjects factor compilation. Both speaker I. F (2. 94)= 75.583. p<.001 . and speaker 2. F (2. 94)= 150.849. p<.001 elicited frequency effects. While both speakers elicited large frequency effect sizes for the low detectability group. speaker 2 had a larger effect size. if}: .852. than did speaker I. (72,): .617. 96 Figure 6 Average Dialect Detectability Ratings for the Two Speakers and the Two Compilations in the Low Detectability Group Organized by Frequency and Salience Speaker 1 Speaker 2 35— Salience 35‘ Salience g to“; .g Low 624— ' '9 §4-— .. - - High Q .é‘ E '5 £3 __ S ._l .9.» 3 a 3 3 8 o __ t” _. 32 32 0 0 > > < < 1-- 1— l l I l l l O 1 3 0 1 3 Number of Features Number of Features Compilation 1 Compilation 3 U) 5— Salience 35— Salience 5’ .5 Low .5 LOW '05 H' 11 iii - a: - ' lg n. 4__ - - - Huh 4— r = n 9 m .33 3 g 3 o a: 2- 8,2— ? e ‘1’ 2 2 1__ < 1.... l l l l 1 I O 1 3 1 2 3 Number of Features Number of Features Dialect Detectability Rating Scale: l= Not noticeable at all. 5= Extremely noticeable 97 The ratings elicited by the two speakers were examined separately for the different frequency levels using paired t-tests with a Bonferroni-adjusted alpha of .025. Table 1 1 is organized by speaker and shows the mean dialect detectability ratings for the different frequency levels and the differences between the ratings associated with each adjacent frequency level. Both speakers elicited a greater effect when the frequency of dialectal features changed from zero to one than from one to three. However. speaker 2 elicited a greater range of detectability changes across all frequency levels than did speaker I. Table 11 Speaker C ornparison for the Mean Change in Dialect Detectability Scores Across the Different Frequency Levels in the Low Detectability Group (N= 48) Speaker 1 Speaker 2 Frequency Frequency Frequency Frequency Frequency Frequency 0 1 3 0 1 3 M: 1.67. M: 2.79. M: 3.54. M: 1.54. M= 3.15. M: 4.04. SD= .975 SD= 1.166 SD= .944 SD=1.031 SD= 1.185 SD= .798 Difference 0-1 Difference 1-3 Difference 0-1 Difference 1-3 [(47): -7.323. p<.001. d= 1.06 t(47)= 4803. p<.001. d= .69 t(47)= -10.419. p<.001. d= 1.50 t(47)= —6.859. p<.001. d: .99 A multiple comparisons analysis using paired t-tests was done to compare the differences between the two different speakers at the three different frequency levels. In this analysis. six ratings were generated by the eight participants for each of the three different frequency levels (N=48). These ratings were matched according to stimulus 98 similarity and were compared between speakers. Specifically. the 48 ratings elicited by speaker I at the three-feature level were compared to the 48 elicited by speaker 2 at the three-feature level. The same comparison between speakers was done for the one and the zero feature levels. Speaker 2 elicited higher detectability ratings when three AAE features were used than did speaker 1. t(47)= -3.958. p<.001. The effect size of this difference was medium. d= .57. In fact. the occurrence of three features was the only frequency level that differed significantly between the speakers. They were not statistically different for stimuli containing zero features. t(47)= 1.137. [F .261. or one AAE feature. t(47)= -2.1-4. p=.036 (not significant at Bonferroni-adjusted alpha of .025). Analyses were also done to determine the source of the frequency-compilation interaction, F (1. 22)= 6.709. p=.003. When the ratings elicited by the two compilations were analyzed separately for the three levels of frequency using 2 x 3 ANOVAs (only including the within-subj ects factors of salience and frequency). both compilation 1, F ( 2. 46)= 149.352. p<.001. and compilation 3. F(2. 46)= 81.414.p<.001. elicited frequency effects. However. compilation l elicited a greater effect size. ([21,: .869. than did compilation 3. (ff .634. A multiple comparisons analysis similar to the one used to examine speaker differences across the frequency levels was used to examine compilation differences across the frequency levels. The data presented in Table 12 shows that both compilations elicited a greater change in detectability scores when the stimuli increased from zero features to one feature than they did when the frequency increased from one feature to three features. These similarities indicate that. although the frequency effects were 99 quantitatively different between the two low detectability group compilations. they were not qualitatively different. Both compilations elicited similar types of frequency effects. Table 12 Compilation Comparison for the Mean Change in Dialect Detectability Scores Across the A ,4 Different Frequency Levels in the Low Detectability Group (N= 48) Compilation 1 Compilation 3 I F req= O F req= 1 F req= 3 Freq: 0 F req= 1 F req= 3 ‘ M: 1.21. M: 2.90. M: 3.83. M: 2.00. M: 3.04. M= 3.75. SD= .504 SD= 1.077 SD= .808 SD=1.203 SD= 1.288 SD= 1.000 Difference 0-1 Difference 1-3 Difference 0-1 Difference 1-3 t(47)= -10.876. t(47)= ~6.015. t(47)= -7.148. t(47)= -5.464. p<.001. d= 1.57 p<.001. d= .89 p<.001. d: 1.03 p<.001. cf: .79 Salience efflcts. The low detectability group had a significant main effect for salience. F (l. 22) = 14.749. p= .001. The effect size for the low detectability group was medium. (72,): .401 Approximately 4 ' % ofthe salience effect elicited by the low detectability group could be accounted for by the variable itself. There was not a salience-compilation interaction. F(l . 22)= 2.880.p=.104. There was also no salience-speaker interaction. F (1, 22)=.757. p=.394. as there was in the high detectability group. F(l. 22)= 20.129.p<.001. F requemjv by salience interaclicm. Similar to the high detectability group. the low detectability group did not yield a frequency-salience interaction. F ('2. 44)= .084. p=.919. 100 Post Hoc Comparisons ofFeaturesfbr Detectability Ratings A post-hoc analysis was done comparing the dialect detectability scores associated with the six individual AAE dialect features used in the detectability stimuli: (a) final consonant deletion of/k/ (F Dk). (b) final consonant deletion of /t/ (F Dt). (c) final cluster reduction of /k/ ( F Rk). (d) final cluster reduction of /t/ (F Rt). (e) final t/0 substitution (t/th). and (1) final f/0 substitution (f/th). Due to the fact that there were six total features analyzed. the results would have been too complex and beyond the scope of this dissertation to explore differences between rankings according to the methodological variables compilation and speaker. Therefore. to simplify these post hoc results. all methodological factors were pooled in the analyses. These post hoc analyses aimed to answer the following questions: 1) Which dialect features elicited the highest dialect detectability ratings and which elicited the lowest? In other words. what is the ranking of the dialect features with regard to each one’s influence on dialect detectability ratings? 2) Was the perceptual salience classification accurate as measured by the actual detectability of features observed in this study? In other words. was each individual feature included in the high salience group more detectable than each of the features included in the low salience group? Ranking (go/:feamres. To determine each feature’s individual detectability value. a calculation was done. The detectability ratings elicited by sentences containing just one feature were compared to sentences that had an unfulfilled opportunity for that same feature to occur (i.e.. the control stimulus). This new calculation was called the mean D score. For example. the mean D score for F Dk was obtained by subtracting the perceived dialect detectability ratings of sentences that had an unfulfilled opportunity for F Dk (and 101 no other planned features) from the ratings of sentences that had one instance of F Dk (and no other planned AAE features). The mean D score associated with each phonological feature was ranked. Paired t-tests matched for stimulus similarity were used to determine if the differences between the zero-frequency levels were significantly different from the one-frequency levels for each dialect feature. In addition to these values. the effect sizes for the differences between the two compared frequency levels for each dialect feature were reported in Table 13. The Bonferroni-adj usted alpha was .008. Table 13 Ranking of Features by Mean D Scores (N=32 for each feature) Ranking Feature Mean D- SD 1 (31) p Cohen’s d value score (M/SD) 1 /k/-deletion 2.19 1.091 -11.346 <.001 2.01 (large) 2 f/th 1.94 1.014 -10.809 <.001 1.91 (large) substitution 3 /sk/-cluster 1.66 .902 -10.388 <.001 1.84 (large) reduction 4 /st/ cluster 1.06 .948 -6.338 <.001 1.12 (large) reduction 5 t/th 1.56 1.480 -5.973 <.001 1.06 (large) substitution 6 t-dclction .91 1.729 -2.964 .003 .52 (medium) Note: Mean D scores were derived by subtracting the detectability ratings for the zero frequency level from the detectability ratings for the one frequency level. In this ranking ofphonological features. it is particularly interesting to examine the top three and bottom three ranked features (see Table 13). The bottom three features in the ranking. all involved a /t/ in the final position (F Rt. t/th. and F Dt). The top three features did not involve a /t/ (F Dk. f/th. and F Rk). The phoneme /t/ has been cited as especially prone to deletion in many dialects. including those considered “standard” (Guy. 1980; Stockman. 2006). It is possible that final /t/ is so prone to deletion among many American speakers that the listeners didn’t perceive its presence or absence as much as the other features that were devoid of a final /t/ alternation. F urthennore, the environment for final /t/ was preconsonantal. Such an environment may be especially prone to deletion of final /t/ by a wide array of American English speakers (Guy. 1980). Perceptual salience classification and detectability. The next analysis aimed to determine ifthe salience classification was accurate for each feature. The mean D scores for each high salience feature were matched to the mean D scores for a phonetically comparable low salience feature. The comparable features matched in this analysis were: 1) FDk (High)--FRk (Low); 2) F Dt (High)--F Rt (Low). and 3) t/th (High)--f/th (Low). Paired t-tests were used to determine if there were significant detectability differences between the components of each pair (see Table 14). The Bonferroni-adjusted alpha was .02. Only the F Dk and FRk comparisons reached significance. as shown in Table 14. The other two comparisons. although not significant. tended in the opposite direction than was expected. In both cases. the features classified as low salience had higher mean D scores than did the phonetically comparable high salience features (cf. 1.94 and 1.56 for f/th and t/th. respectively). Table 14 Comparison of Mean D Scores for Comparable Features in the High and Low Salience Groups (N=32) Salience Feature Mean D SD Difference SD t( 31) p .531 1.107 2.715 .010* Low FRk 1.66 .902 High FDt .91 1.721 -.156 1.936 -.456 .651 Low F Rt 1.06 .948 (ns) High t/th 1.56 1.480 -.375 1.385 -1.531 .136 Low f/th 1 .94 1 .014 (ns) Note: Mean D scores were derived by subtracting the detectability ratings for the zero frequency level from the detectability ratings for the one frequency level. * Significant. p<.02 (.05 alpha adjusted per the Bonferroni method) Summary of Dialect Detectability Results The data supported the hypothesis that the number of dialect features included in the sentence would increase the perception of dialect detectability. The issue of perceptual salience was not as clear. In some instances. the perceptual salience of the phonological patterns had a moderate effect on the ratings of dialect detectability. However. for speaker 1 in the low detectability group. there was no salience effect. These inconsistent findings may be partially explained by the post hoc analysis. The post hoc analysis revealed that the perceptual salience of some phonological patterns did not 104 explain the dialect detectability ratings elicited in the study. These findings will be examined further in the upcoming discussion. Principal Findings: Comprehensil)ility Ratings Does the number of AAE features and/or the perceptual salience of the AAE features in a stimulus sentence affect the degree to which the listeners deem a speaker as comprehensible to the general population? In answering this question. it is important to consider that comprehensibility was measured in a way that was different from dialect detectability. During the rating of comprehensibility. the listeners were asked to assign one rating to both speakers. so the methodological. within-subjects factor speaker does not exist in the comprehensibility analysis. All other factors (salience. frequency. and compilation) remained. As discussed previously. the different stimulus compilations did not elicit significantly different comprehensibility ratings. F (3. 92): 2.46. p= .068. However. compilation was retained as a between-subjects factor in the analysis to determine if either of the within-subjects factors interacted with compilation. Therefore. a 2 x 3 repeated measures ANOVA with a four-level between-subjects factor. compilation. was used to analyze the comprehensibility ratings. As occurred in the dialect detectability analyses. paired t-tests adjusted using the Bonferroni technique were done to investigate differences between levels of variables revealed as significant in the omnibus ANOVA. Frequency Effects There was a main effect for frequency in the comprehensibility data. F (2. 24): 50.356. p<.001. Figure 7 shows that as the number features increased. the perceived comprehensibility ratings decreased. This effect was large. ([3,): .808. and indicated that 105 over 80% of the total effect was accounted for by the variable itself. There was no frequency-compilation interaction. F (6. 24): .600. [F .195. indicating that the frequency effect was consistent across all four stimulus compilations. Figure 7 Average C omprehensibility Scores Elicited from Stimuli Obtained in Relation to the Frequency and Salience of AAE Features (N=16) Salience — Low - - - High 01 l A l Average Comprehensibility Ratings 0: l .\._ .‘-..“ 2.— 1— l I T 0 1 3 Number of Features Comprehensibility Rating Scale: 1: Very difficult to understand. 5: Very easy to understand 106 A paired t-test was done comparing the stimuli associated with the different frequency levels. To control for Type I error. the alpha was adjusted to .025 using the Bonferroni technique. The comparisons revealed that stimuli containing 3 features were rated significantly less comprehensible than stimuli containing 1 feature. t(3 l) = 3.215. p<.001. Stimuli containing 1 feature were significantly less comprehensible than those containing no features. t(3 l )= 8.530. p<.001. The effect size of the 0-1 difference (d= 1.51) was much larger than the effect of the 1-3 difference (d= .57). This finding was similar to the frequency effects in the detectability ratings. yet magnified. The inclusion of one dialectal feature contrasted with no features apparently had a more extreme impact on the SLPs perceptions of comprehensibility than did the inclusion of three features contrasted with one feature. The effect from 0-1 feature may indicate the listeners“ awareness of AAE presence in contrast to AAE absence. The effect from 1-3 features. may reflect less awareness of AAE presence and reflect awareness of increasing use of AAE. Salience Effects There was a significant main effect for salience. F (1.12): 35.593. p<.001. Figure 8 presents these results graphically. Those features classified as less salient elicited lower comprehensibility ratings than those features classified as more salient. ((47): 5.51 l_. p<.001. This effect was large. 1131,: .748. and indicated that almost three-quarters of the overall effect was accounted for by the variable itself. The salience effect did not significantly differ across the four different stimulus compilations. F (3. 12)= .926. [F .458. 107 Frequency by Sal ience Interaction Unlike the detectability ratings. the comprehensibility ratings revealed a significant frequency-salience interaction effect. F ( 2. 24): 4.926. p= .016. This interaction may be viewed graphically in Figure 8. To analyze this interaction. paired t- tests were used to make the following comparisons: pair 1) high salience/zero features with low salience/zero features; pair 2) high salience/one feature with low salience/one feature; and pair 3‘) high salience/three features with low salience/three features. Type I error was controlled using the Bonferroni-adjusted alpha of .02. There were no Significant differences between the components of pair 1. t(15_)= 1.861. p= .083. indicating that the mere opporttutity for high or low salience features was not enough to generate a salience effect. In pair 2. a salience effect was elicited when one AAE feature was included. t(15)= 4.472. p<.001. Specifically. the higher salience features generated lower ratings of comprehensibility. The effect of this difference was large. (1: 1.12. In pair 3, the salience effect was maintained when three AAE features were included. t(15)= 3503. p=.003. While the effect size associated with three features was still considered large, d= .88. it was smaller than the effect elicited when one feature was included (d= 1.12), Sumn'lanv of C omprehensihility Results The data associated with the perception ofcomprehensibility yielded the fOIIOWing conclusions: (a) the Anglo-European American SLP listenersjudged sentences to be less comprehensible to the general public when they included a greater number of AAE features. and (b) the listeners judged the high salient features as having a greater unpaCt on comprehensibility than the low salience features. In addition to these main 108 effects. there was an interaction between frequency and salience. Specifically. as the number of AAE features increased. the ones with higher perceptual salience had a greater impact on perceptions of comprehensibility than the ones with lower perceptual salience. l/l’ritten Comments Regarding Study After completing all of the ratings. the participants were asked to write any “feelings. comments. or questions about the study.” Six of the sixteen (37.5%) did so. Two participants commented about the tasks of rating detectability and comprehensibility. One of these participants stated. “Detectability is a very different question. because someone can be very understandable. but also have a very noticeable accent.” She also commented that “consonants are not the only things that influence detectability—there is something about the vowels that carry the accent.” The other person commented that “final consonant deletion is difficult to understand anytime.” The other participants commented that they were not accustomed to viewing the features included in the study as dialectal. Rather they were apt to label such features as Phonological disorders. One commented. “It was a different experience to think about it as ‘African American English’ and not 'we’ve got to enroll this kid!” Another PaniCipant commented that. “It is somewhat difficult to separate/rate how noticeable the dlalect is from what. in my students. I would consider an “articulation error’.” 109 CHAPTER IV DISCUSSION This section begins with a discussion of the conclusions pertaining to the main research questions that were informed by the dialect detectability and comprehensibility data. It continues with a discussion of the observations related to the methodological variables of the study. Next. it addresses the implications of the study's findings for theory. applied research. and clinical practices. It then concludes by outlining areas for future research. Gradient Perception ofAAE Dialect by SLPs The purpose of this dissertation was to better understand how AAE use is perceived among Anglo-European American SLPs. Their perceptions were of interest because of their high degree of professional power in determining who is a normal or disordered speaker. The two experimental independent variables.fiequency and Perceptual salience. were investigated as factors that could influence perceptions of dialect detectability and comprehensibility in SLP listeners relatively unfamiliar with AAE. Frequency of Dialect Features The data from this study supported the hypothesis that an increased number of AAE dialect features present in an utterance will elicit increased ratings of dialect detectability and decreased ratings of comprehensibility from non-AAE speaking SLPs. In Other words. if many AAE features (of any type) are used within sentences and the person‘s speech is evaluated by an SLP with little exposure to the dialect. the speaker will likely be identified as a highly detectable AAE speaker with compromised 110 comprehensibility in the general population. The threshold for a significant change of either comprehensibility or dialect detectability ratings. however. was just one feature. That is. it only took one feature in an eight-word sentence to increase dialect detectability ratings and decrease the perceived comprehensibility of the sentence. Although. the subsequent inclusion oftwo additional dialect features did change both ratings. the change from one feature to three features did not affect scores as much as the change from zero features to one such feature. Apparently. all that was needed for the SLPs in the study to detect that the speaker spoke AAE was one phonological feature. regardless of the type of feature used. This shift in categorization from a “non-AAE user“ to “AAE user” created the greatest change in ratings. As additional features were added. no shift in categorization was needed. The speaker was perceived to be on the continuum of AAE use from that point on. The same type of threshold effect occurred with the comprehensibility ratings. One feature was enough to significantly decrease the SLPs‘ projections of how understandable the person would be to the general population. We may conclude. therefore. that presence is of greater weight thanfrequencr’. Perceptual Sal ience of Dialect Features Based upon the expectations of Steriade (2004). perceptual salience was expected to influence comprehensibility and dialect detectability ratings. Thus. the categorization of the AAE phonological patterns examined was based on that very factor. However. perceptual salience did not appear to influence the ratings of both tasks in the same way. Comprehensil)ilit)y'zulgrrzents. Perceptual salience appeared to accurately classify the features for the comprehensibility ratings (i.e.. the group of features with high 111 perceptual salience elicited lower comprehensibility ratings than the features with low perceptual salience). However. given that the sentences were rated as a group for comprehensibility. a true feature-by-feature comparison could not be made. F urthemtore. the features were already grouped according to perceptual salience in the very stimuli that were rated. The listeners were not free in this task to base their judgments on other groupings. as they were in rating the dialect detectability task. Nevertheless. the data suggest that SLPs may have considered the perceptual salience of the features in detemtining which sentences would be comprehensible to the general population and which ones would not. One particular construct in the methodology could have aided these results—the orthographic transcription of the auditory stimuli. In rating the projected comprehensibility of the sentences. the SLPs were presented with an orthographic transcription which corresponded to the auditory stimuli. This type of task is different from other studies that examined comprehensibility. For instance. the studies that examined perception of nonnative accents. (e. g.. Derwing & Munro. 1997; Munro & Derwing. 1995a. 1995b). did not provide an orthographic representation of the auditory stimuli. However. SLP listeners in this study may have been less linguistically naive than the listeners from the general population who participated in second language accent perception studies. SLPs have specific training in phonetics and phonology that the participants in the other studies would not have had. This additional knowledge of the SLPs created several questions regarding what they would base their judgments on. Would the SLPs recall and use their knowledge about phonetics and phonology in making their ratings. or would they not? If they did use such knowledge to make their judgments. on what phonetic target would they base their decisions? Due to these questions. some added stability was required. However. the stability that was added was not unlike the inherent stability in the SLPs’ actual practice. Specifically. it was similar to the relational analysis (comparing the client's productions to an intended target or model) that many of them would use in assessing the phonological patterns of their clients. The inclusion of the orthographic transcription created a need for an additional modification from the procedures used in the majority of second language accent perception studies. That is. the SLPs were not asked to base their comprehensibility ratings on whether or not they themselves understood the sentence. Instead. they were instructed to base theirjudgments on whether or not the people speaking would be understandable to others in the general population. Presumably. this modification required each SLP to base her judgments on a self-determined paradigm—perhaps the very paradigm that she would have used with her clients. In fact. one participant remarked that “final consonant deletion is always difficult to understand." Such a statement appears to reveal a portion of the paradigm on which she based on her ratings (i.e.. If the client omits a final consonant. then the comprehensibility rating should be decreased). Indeed. some clinical researchers have developed guidelines concerning what phonological patterns are likely to impact a given client’s intelligibility in the general population. Such guidelines may have been instrumental in assisting some of participants with the creation of their rating paradigm. One popular phonological disorders assessment. The Hodson Assessment of Phonological Processes-3 (’HAPP-3; Hodson. 2004) weights final consonant deletions more heavily than other phonological patterns due to its expected negative impact on intelligibility. It is possible that paradigms similar to the one used in the HAPP-3 were used in making the comprehensibility judgments. Dialect detectability jm‘lgments. The categories of perceptual salience did not explain the dialect detectability scores as consistently as they did the comprehensibility scores. The high salience features did elicit higher dialect detectability scores than the low salience features in most cases. However. when the dialect detectability ratings were examined post hoc for each feature individually. their ranking suggested that the perceptual salience did not entirely explain the ratings (refer to Table 14. p. 103). In fact. the feature that received the lowest perceptual salience ratings in the preliminary experiment involving nonsyrnbolic syllables (f/th substitution). elicited the second highest dialect detectability ratings in the main study. This discrepancy was allowed due to the fact that the features were not pre-grouped as they had been in the comprehensibility task. The data from the dialect detectability task were gathered differently from the comprehensibility task. Because of this difference. the participants were free to group the dialect features as they wished. Such a difference limits the extent to which the data generated by the two tasks can be compared. However. the tasks were also inherently different perceptual judgments. This point was acknowledged in writing by a participant after completing the experiment. "Detectability is a very different question. because someone can be very understandable. but also have a very noticeable accent.” One difference between the two tasks concerns the prerequisite knowledge of the SLPs in making informed ratings. While comprehensibility could be determined without any specific knowledge of common AAE features. the dialect detectability task required 114 some degree of awareness ofphonological features associated with the AAE dialect. In fact. one participant commented on her own lack of AAE phonological awareness. "It was a different experience to think about it as ‘African American English“ and not ‘we've got to enroll this kid!”9 The SLP participants were selected based on their lack of familiarity with AAE. They indicated that they did not speak AAE themselves. and they did not converse regularly with a person who spoke AAE. If they only heard AAE spoken infrequently. then those features used most commonly by AAE speakers in the general population may have been rated higher in detectability. In the terms of Preston's (1996a) work. they may have rated those utterances as highly AAE detectable that included features highly available to them and for which they could readily cite as characteristically AAE (i.e. detail). Irvine (2001) described features such as these as iconic. For example. if the SLPs associated the f/th substitution pattern as particularly characteristic of AAE. then this one feature might elicit higher detectability ratings. regardless of its perceptual salience in nonsyrnbolic units. If Steriade’s (2004) expectation that less salient features will be used more frequently by speakers in all languages and dialects is correct. then the least perceptually salient features may have higher iconicity by virtue of their ubiquitous use in the language of AAE speakers. In this case. the participants of the study should have rated the low salience features as having higher dialect detectability in all cases. This scenario did not manifest either. As a group. the high salience features often elicited significantly higher dialect detectability ratings than the low salience features. In fact. the final consonant deletion of /k/ received the highest mean D-score. indicating that 115 this feature created the greatest increase in dialect detectability ratings of all the featru‘es studied. Final consonant deletion of /k/ was also the most highly perceptually salient. Additionally. it was identified less frequently than th substitutions and cluster reduction in the Craig and colleagues (2003) study. Stockman (In Press) further identified that final /k/ deletion occurred significantly less frequently than final /t/ deletion in a corpus of AAE speaking children. It appears that a combination of the two factors occurred: (a) ubiquitous use by AAE speakers and (b) perceptual salience. That is. those features that were so commonly used by AAE speakers received inflated detectability ratings due to their heightened iconicity with the dialect for the listeners. In this case. the f/th substitution pattern as the highest ranked of the low salience features would be the most iconic (second highest mean D scores of all of the features). In contrast. the perceptual salience of the deleted final /k/ was high enough that it didn‘t matter if the SLPs associated it with AAE or not. Its highly perceived use was enough to increase dialect detectability ratings and produce an overall perceptual salience effect for much of the dialect detectability data. As for the three features receiving the lowest dialect detectability ratings. they all involved a final /t/ (t/th. F Dt. and F Rt). In the final position of words it is produced variable in many dialects of English. These variations include reduction to a glottal stop or omitted entirely in many dialects of English. even those considered standard (Guy. 1980; Stockman. In Press). It is regularly deleted in the final position of words when occurring in a consonant cluster (i.e. fast-9 lfzes/) in standard varieties of American English. particularly when a consonant follows. as it did in this study. Taking these facts into account. it is understandable that features involving final /t/ would receive lower 116 dialect detectability ratings than other features involving other consonants. It is possible that when a feature involved a final /t/. the listeners had a wider range of acceptance due to the great variability of final /t/ production among speakers of many dialects ofEnglish. Sana-may ofperceptual salience contribution. This study did not ask the SLPs how much they associated each phonological pattern with the AAE dialect. However. the possibility exists that the listeners might have associated some of the features as characteristically AAE. and those features may have influenced their decisions about AAE detectability more than others regarding a sentence’s AAE detectability. This seemed to contrast with the rating of comprehensibility. As one of the participants implied in her written comment. the listeners might have considered some features as contributing significantly to the detectability of the dialect without having a negative impact on comprehensibility in the general population. Conversely. some features may be deemed to affect comprehensibility in the general population but not be iconic enough to contribute to the dialect‘s detectability. In Figure 2. p.11. a model was generated to illustrate how remarks from the general population about how intelligible and noticeable a person is may contribute to the overall perception of dialect density. In this model the two factors existed along a single dimension. or plane. If a speaker was unintelligible to the listener and spoke with a highly noticeable dialect. then he/she likely would be characterized as speaking with a “very thick" accent/dialect. Conversely. if the speaker was highly intelligible and spoke with a dialect that was not noticeable to the listener. then the speaker likely would be identified as speaking no accent/dialect. Figrue 8 reframes this model with the terminology used most frequently in this study. For example. very intelligible and 117 unintelligible correspond to high comprehensibility and low comprehensibility. respectively. Likewise. not noticeable and very noticeable correspond to low detectabilitv and high detectability. The scale was converted from one reflecting the statements made from members of the general population to one reflecting the general concept of perceptual dialect density. Figure 8 Original Single Dimensional Model of Perceived Dialect Density High C omprehensibility Low Comprehensibility Low Detectability High Detectability 0 Low Density High Density By conflating judgments of comprehensibility and dialect detectability. this single dimensional model did not take into account the inherent differences between the two judgments that were reflected in both the quantitative data and the written comments in this study. Dialect detectability and comprehensibility may not operate along the same dimension. Instead. the factors of comprehensibility and detectability may operate along two intersecting dimensions. as depicted in Figure 9. In this revised model. the perception of dialect is multidimensional. capturing an additional layer of complexity. The revised. multi-dimensional model shows that both dialect detectability and comprehensibility contribute to the perception of dialect indicated in statements commonly heard from members of the general population. Both comprehensibility and detectability function as separate factors in the gradient perception of dialect. 118 Figure 9 Multidimensional Model of Perceived. Dialect Density The four quadrants represent perceptual qualifiers that might be used by listeners to categorize speakers who have the given coordinates of comprehensibility and dialect detectability. High Comprehensibility Accent/Dialect Unidentifiable Accent/Dialect Strongly Identifiable AND Easy to Understand BUT Easy to Understand Example: Example: Standard English speaker “without " US Bostonian an identifiable regional/etlmic 1 Standard English Dialect accent/dialect Low Detectability High Detectability . _ . Accent/Dialect Strongly Identifiable AccentiDralect Unrdentrfiable BUT Difficult to Understand AND Difficult to Understand Etample: Example: African A merican English Dialect Ocracoke English Dialectfrom North Carolina ’s Outer Banks Low Comprehensibility The revised model creates four quadrants that correspond to categories in which a speaker might qualify according to the degree that his/her dialect is deemed comprehensible and detectable. Ifa person’s dialect is both highly detectable and incomprehensible to a listener. then he/she is likely to be perceived as a nonstandard speaker with a very strong. or heavy dialect. Such is the case for speakers of socially stigmatized dialects (e. g. African American English) who use a significant number of features that are both highly identifiable with the dialect and perceptually salient. Conversely. if the person's dialect is not highly detectable and highly comprehensible. then he/she may be perceived as a stamlard English speaker "without" an accent/dialect. Alternatively. a person may fall into two other categories. One category corresponds to 119 speakers with a highly detectable but comprehensible dialect. This category might include speakers of distinctive dialects that either are highly familiar to the listener (e. g.. regional dialects from the listener‘s own geographical area or highly prestigious. yet uncommon dialects (e.g.. Brahmin dialect of Boston). Conversely. a speaker may also be judged to be incomprehensible. but speak without a great degree of dialect detectability. Perhaps the dialect is one to which the listener has never been exposed (e. g.. Ocracoke English from the Outer Banks ofNorth Carolina). The use of these less identifiable dialect features could cause the perceptual judgments to be weighted more in the area of comprehensibility limitations and less in the area of dialect detectability. This model helps to explain the interrelationship between two perceptual variables on which global judgments about a person‘s dialect are based. However. other factors may also influence these gestalt dialect judgments. Therefore. additional research should be done to identify whether other factors contribute to the holistic notion of dialect perception. Methodological Variables The methodological factors. compilation (4 levels) and speaker (2 levels). did not create drastically different results but did indicate that there is some variability inherent in the process of perceptually rating dialect. The variability present in the four stimulus compilations and the two different speakers will be discussed. Differences Among Compilations As described earlier. the compilations included the variables of speaker order and task order. Differences among the compilations were present only in the detectability ratings. These ratings did not reflect a consistent main effect for speaker order or task A rE'L order. Instead there seemed to be an interaction between these two variables. In essence. the order of the tasks created differences only when speaker I was first; speaker order created differences only when the comprehensibility task was first. Two factors. other than the intended variables inherent in the compilation C Ds. could have caused these compilation differences: (a) unintended clustering of highly salient AAE phonological patterns in the first few stimuli on some compilations. and (b) unintended clustering of inexperienced or highly experienced SLPs in one or more compilations. Niedzielski‘s (1999) observations suggested that individuals may make conclusions about an entire sample based on their initial experiences with a speaker. Therefore. the initial five stimulus items on each of the compilations were reviewed for unintended clustering of either high or low numbers of AAE features. No clear clustering of frequencies could account for the compilation differences on the compilations. The second possible factor. experience level of the four groups of SLPs. was also examined. In this analysis. the experience levels of the SLPs were analyzed according to compilations. The years experience for the SLPs included in the different compilation groups are presented in Table 15. l m. m—. . Table 15 Years Experience as SLP for Participants Organized by Stimulus Compilation (N=16) Compilation 1 2 3 4 ID# 15g9i13 21611014 3 711115418112j16 Years 629|9j33 7i18 5 21 1410(13 4 317,516 Experience l ' . l J l Descriptive M: 19.25. M= 12.75. M= 10.25. M= 5.25. Statistics SD = 13.72. SD = 7.93 SD = 4.50 SD = 1.71 Range: 27 Range2 16 Range: 10 Range= 4 The average experience values for the participants in each compilation were tested for differences using the nonparametric Kruskal-Wallis H Test for multiple groups. Although the average experience level in group 4 appeared significantly lower than the average experience level in group 1 (cf.. M= 27 and M: 4 for compilations 1 and 4. respectively). the difference was not statistically significant at the .05 level. Kruskal- Wallis H (3) = 4.669. p= .198. However. even if the difference in experience levels observed among the compilations were significantly different. the differences could not consistently account for the compilation differences. While such a difference might have explained why compilations 1 and 4 were placed in separate detectability groups based on the participants. ratings. it would not explain why compilation 4 (the least experienced group) would have been so similar to compilation 2 (the second most experienced group) as to have had the same mean dialect detectability rating (M=3.22). As Flege and Fletcher (1992) described. there are many factors that may influence perceptual ratings of second language accent detectability. Perhaps the rating of dialect is r10 different in this sense. It is possible that several factors combined to affect the ratings. or there are factors inherent in the compilations that have not yet been identified in research on cross-dialect detectability. Further examination of the factors influencing dialect detectability needs to be done before a clear explanation of the differences among the compilations can be developed. Differences Between Speakers There were some inconsistent differences among the dialect detectability ratings elicited by the speakers. These differences were not qualitative. except in the high detectability group when one speaker elicited a salience effect and the other did not. Such slight speaker differences were to be expected. Use of subtle segmental and suprasegrnental speech characteristics may vary widely not only between different speakers but within the same speaker. This individual variability is complicated by the fact that listeners may also vary in how such characteristics may affect perception. Theoretical and Research Implications The findings of this study have implications for five specific theoretical and research paradigms: (a) Steven‘s Law and general sensory perception. (b) Johannson‘s (1978) model of nonnative accent feature-specific perceptual weighting. (c) Steriade’s (2004) P-map model, (d) Prototype-based models of speech perception. and (e) Washington and C rai g‘s (1994. 1998) process for the calculation of dialect rate (density). Implications for these paradigms will be discussed in the following paragraphs. Steven 's Law and General Senson Perception The notion that perception of a stimulus will increase as its magnitude increases has been accepted in general perceptual theory for many years. It was articulated by Stevens (195 7) in Steven‘s Law. which expressed that the degree to which sensory stimulation is perceived is directly related to the magnitude of the sensory input. Such perception is represented with the formula: S = kl 8. The S represents the intensity of the stimulus perceived. The k represents a constant. and 1 represents the magnitude of the stimulus. The exponent a corresponds to the unique effects that the wide array of sensory experiences have on the process of perception. The exponent is specific to different types of stimuli. The perception of a dialect may be similar to other sensory perceptions. and therefore. may be informed by the principles on which Steven‘s Law (1957) is based. Framed in accordance with the variables included in this study. S represents perceived dialect density. 1 represents thefi'eauencv variable. and the exponent a may represent the different types of dialect features involved. Considering Steven‘s Law. it is not surprising that the variable that most directly communicated the magnitude of the dialect. frequency. had such large effects on the ratings of both dialect detectability and comprehensibility. The reduced. but present effect of the type of features used in this study indicates that the values of the exponent a in this specific perceptual experience may be slightly only slightly different for the different dialect features included. Johannson 's F circtors for Non-m'itive Accent Feature Perceptibility Johannson (1978) suggested that two factors influence the degree to which a feature contributes to the detection of a second language accent: (a) the degree to which it impacts speech intelligibility. and (b) the degree to which it is “irritating” to the listener. Since Johannson was interested in nonnative speech patterns. he discussed his findings in terms of the types of mistakes made by speakers in speaking a language that was not their native language. Dialect features may be analogous to the second language 124 mistakes addressed by Johannson. In particular. both are variations from what may be familiar to listeners. While particular AAE dialect features may be irritating to some listeners. the features may not need to be irritating to increase the perception ofdialect density. If the factor of feature irritation mentioned by Johannson is analogous or related in some way to dialect detectability. then the data in this study may support the general idea of Johannson’s argument. That is that awareness of a dialect feature may be a factor in detemrining the degree to which the dialect will be perceived on a scale of density. The model proposed by the investigator (see Figure 9) to explain the results of this study was similar to Johannson’s (1978) model in the sense that there are two factors that influence the perception of language variation. One is the degree to which the variety is understandable to the listener. and the other relates to the degree to which a feature calls attention to itself. In this way. Johannson’s model is supported by the data in this study. However. with respect to dialects. Johannson’s factor irritation may be replaced with the broader factor dialect detectability, which may be affected in part by listener irritation. dialect iconicity. or perceptual salience. Steriade 's P-Map Steriade (2004) asserted that the features with low perceptual salience may be used less frequently by speakers than the features with high perceptual salience. The preliminary study involving perceptual salience of AAE phonological patterns in nonsymbolic units appeared to support Steriade’s P-map expectations. This was verified by the phonological patterns documented from language samples of school-aged AAE- speaking children in Craig and colleagues (2003) study. The features that received the lowest ratings of perceptual salience in the preliminary study were the same ones used 125 most often by school-aged AAE-speaking children in the work from Craig and colleagues (2003). Stated simply. low perceptual salience results in increased use. Increased use of a feature by dialect speakers is likely to cause increased association with the dialect. or increased feature iconicity. and iconicity may be but one factor in dialect perception (Irvine. 2001). The perceptual salience classification obtained from the preliminary study described in this dissertation was derived from Steriade‘s (2004) suggestions concerning the P-map. As described previously the P-map is a representation of the perceptual salience ofcontrasting phonological features. Such perceptual salience did appear to influence comprehensibility ratings. Steriade argues that speakers avoid perceptually salient phonological patterns by seeking the alternation which changes the underlying representation‘s (UR) form minimally. This is done to comply with the phonotactics of the language/dialect being spoken. while still maintaining the closest approximation to the UR as possible. As alternations become perceptually distinct from the UR. the signal may be compromised for wider and wider audiences. Since those very features that were ranked high in perceptual salience were also deemed to have the greatest impact on perceptions of comprehensibility. the comprehensibility data are in accordance with the. expectations expressed by Steriade (2004). F urthemrore. many of these high salience features did not appear to be identified strongly with AAE as evidenced by their contribution to the dialect detectability results. By this interpretation of the data. the dialect detectability results may also support Steriade’s assertion that features that are highly perceptually salient will be used less frequently by speakers. This may explain why many of the high salience features were not associated with the dialect by listeners who were relatively inexperienced with the dialect. Prototrpe-Based Models of Speech Perception Prototype-based theories of speech perception maintain that listeners match auditory input to an internal prototype. For example. Kuhl’s (1994) Native Language Magnet Theory states that perception of speech input is based on a person’s own phonological system. Phonernes are perceived in categories that are developed from the listener‘s own linguistic experiences. lfa listener hears a phonetic feature that falls within the listener‘s perceptual boundaries of a particular phonemic prototype. then the feature will be perceived as that phoneme. This is the case even when the speaker intended a different phoneme to be perceived. This study did not allow for such phonemic misclassifications to affect intelligibility. Because the orthographic representation of the sentence was included. the listeners knew the intended phonemic targets. However. such misclassification could have affected the degree of perceived comprehensibility. If the perceived phonemes did not match those expected from the orthographic transcription. then it is reasonable to assume that the listener would have classified the speaker as having low comprehensibility in the general population. Such findings would support Kuhl‘s (1994) Native Language Magnet Theory. Trends in the data also support the Fuzzy Logic Model of Speech Perception (F LMSP; Massaro. 1988). F LMSP suggests that when speech is heard. the listener will go through three operations of phonemic decoding. First. a feature is either noticed or not noticed. If noticed. then the feature is matched with prototypes stored in the listener‘s memory. Finally. the feature is classified according to its similarity with the stored prototypes. Such classification is based uponfiiz; i (i.e.. approximate) values assigned subconsciously by the listeners (e. g.. 0= definitely not a match. 1= definitely a match. .5= ambiguous). The F LMSP would predict that the most perceptually salient phonological patterns would have received the lowest fuzzy values. That is. they would be the furthest away from matching the prototypes that corresponded to what was expected from the orthographic representations. Such mismatched multiplied by the frequency of unfamiliar or unexpected dialect features would further decrease the fuzzy values for the entire sentence. In the comprehensibility data. there were both main effects for frequency and main effects for salience. F urthermore. the interaction indicated a greater cumulative effect as high salience features were added than when low salience features were added. These findings support the F LMSP. Implications for Research Methodologr There are several ways in which dialect density has been calculated in child language research (Oetting and McDonald 2002). The calculation usually entails summing the number of dialect features and dividing by some unit of language; however. the formula has rarely been used in conjunction with perceptual evidence. The study‘s data support the notion that the number of phonological features affects perception of AAE detectability and comprehensibility. However. the fomrula for dialect density calculation may be too simplistic to adequately represent cross-dialectal perceptual experiences. The calculation of dialect density is analogous to the calculation of physical density. including the assumption that each component of the formula is equally 128 weighted. Dialect density calculations do not typically consider the types of features involved. This study"s data suggest that the specific types of features occurring in a speech sample should be considered before an accurate representation of cross-dialectal perception is approximated. The data also revealed another limitation on the use of dialect density calculation pertaining to dialect perception. That is. the dialect density calculation assumes each additional dialect feature will affect perceptual judgments equally. Throughout the data in this study. the inclusion of one feature consistently elicited greater differences over no features than three features did over just one feature. In this sense. counting each feature equally may assume that each feature is of equal weight. which may not be the case perceptually. The use ofdialect density calculation methods may be adequate in categorizing participants for the purposes of expressive language data analysis; however. such categorization may be limited to quantifying dialectal expression. Models used to make inferences regarding the perception of dialect in speakers by listeners. who do not speak the dialect. should consider the specific types of features involved in the person’s speech as well as the fact that the relationship between the dialect density average and listener perception may not be linear. Practical Implications The implications of this study are relevant for: (a) SLPs currently practicing in the public schools. (b) academic programs devoted to training SLPs. and (c) SLPs engaging in procedures of accent modification. 129 Implicationsfor SLPs in the Public Schools Many aspects of the dissertation‘s study were designed to replicate in a controlled. experimental way the circumstances that might occur if an actual AAE- speaking child were evaluated by an SLP with little exposure to AAE. The study is limited in its ability to predict the effects of such a scenario for two reasons: (a) It was not naturalistic. and (b) it did not involve children. In spite of these limitations. the findings may illuminate some of the possible outcomes of such a scenario. The first practical question to be answered is “Who would have been labeled as speech disordered by SLPs who were using the traditional interpretation of Van Riper’s (1978) definition?“ Van Riper’s definition of speech disordered includes speech that: (a) calls attention to itself. (b) interferes with communication. or (c) places an emotional burden on the speaker. While the last criterion could not be inferred from the data in this study. the first two criteria may be directly informed. The detectability ratings correspond to criterion a. and the comprehensibility ratings correspond to criterion b. The data suggest that speakers who include a greater number of AAE phonological features will be deemed less comprehensible to the general population and more detectable as an AAE-speaker. The data also suggest that speakers who include features with greater perceptual salience will be judged less comprehensible to the general population —— specifically. deletion of /k/. deletion of /t/. and t/th substitution. Furthermore. the findings imply that those individuals will likely be identified as using detectable AAE who: (a) use features that are so ubiquitous in AAE speakers that their use is associated strongly with AAE (e.g. f/th substitution) or (b) use features that are extremely perceptually salient (e.g. k-deletion). While judgments of SLPs practicing in the public schools may be infomred by the data in this study. it is unclear what actions would be most commonly taken in response to the judgments. Such actions are detemrined by how the SLPs interpret the definition of speech disordered. If the SLP interprets the Van Riper (1978) definition traditionally. then the decision of the SLP is clear. Children identified as either less comprehensible to the general population and/or identified as using a highly detectable form of AAE are to be identified as speech disordered. Alternatively. Payne and Taylor (2005) suggested that SLPs with a more accurate perspective toward communication might interpret the traditional definition of speech disorder with the understanding that all communication is socio-culturally based. Individuals are speech disordered. whose speech draws attention to itself or is difficult to understand within the communityfi'om r-vhich they learned their language (i.e.. their home speech community). A non-AAE-speaking SLP is not from the speech community of his/her AAE-speaking client. Therefore. they must observe the client within the home speech community. interview the child‘s primary caregivers. and/or have an understanding of the normal dialectal patterns common in the speech community. With this basic understanding. the SLP will likely use the evidence of the highly detectable AAE features to further examine the adequacy of the child‘s cormnunication in relation to his/her speech community. Two of the three most detectable features in this study were identified by Craig and colleagues (2003) to be the features most frequently used by school-aged AAE-speaking children (th substitutions and final cluster reduction). This suggests that AAE-speakers commonly use features that will be detected and identified as characteristically AAE by SLPs. who are relatively unfamiliar with the 131 ‘. dialect. If the SLP interprets the Van Riper definition in a socio-cultural context. as Payne and Taylor suggested. then the child will likely receive less biased assessment procedures based on the perceptual judgments used as dependent variables in this study. Implicationsfor Instructors of SLPs These findings. in light of the Payne and Taylor (2005) suggestion, place heavy responsibility on the institutions providing academic and clinical training to prospective SLPs. The institutions must instill a socio-cultural view of communication in their students from the very beginning to prevent misdiagnoses with minority dialect-speaking children. Such a view requires all discussion of communication. both normal and disordered. to be in a socio-cultural context. Although such education is now required by ASHA (1999). there are few guidelines in place for this education. In fact. Stockman. Boult. and Robinson (In Preparation) discovered that a majority of general course instructors in communication disorders programs reported that they spend a minimum amount of time discussing multicultural issues in their courses. This was the case. even though a majority identified themselves as strongly committed to infusing multicultural issues into their courses. Implicationsfor SLPs Engaging in Accent Modification The findings ofthis study also have implications for those SLPs engaging in accent modification. If an AAE-speaking individual seeks to speak a dialect that is closer to Standard American English in particular social situations. then the phonological features that would be of greatest detectability would be worthy of targeting. Such features might include: (a) final k-deletion. (b) f/th substitution. and (c) sk-cluster reduction. Alternatively. a general goal to reduce the overall frequency of AAE feature use would be affective in increasing the perception ofcomprehensibility and decreasing the perception of dialect detectability. However. as addressed by ASHA (l983b). accent modification must be done with contextually-based goals. That is. the situations in which a standard dialect would be desired are addressed along with the understanding that the standard dialect should not replace all communicative styles. Study Limitations and Future Research The present study is one of the first studies to examine factors that might influence how SLPs perceive the AAE dialect. Much more research needs to be conducted to further explain how SLPs perceive other nonmainstream dialects. The discussion of study limitations and future research is organized in the following categories: (a) listener variation. (lb) stimulus variation. (c) task variation. and (d) speaker variation. Listener l«"ariation The listeners in the study were Anglo-European American SLPs from Michigan who had little exposure to AAE. It is unknown how well the results would generalize to other listeners. such as AAE speaking SLPs. non-AAE speaking SLPs who are familiar with AAE. or SLPs who speak other stigmatized varieties of English. Such research is needed to determine how personal experience with different language varieties influences perceptual judgments ofdialect detectability and comprehensibility. Stimulus Variation The stimuli involved in this study included just a few commonly cited AAE features in the final position of words. Further research should focus on how other 133 dialect patterns impact perceptual judgments ofcomprehensibility and dialect detectability. Furthermore. studies involving videotaped stimuli may yield different results. For example. the t/th substitution pattern wasjudged to have high perceptual salience in nonsymbolic syllables in the preliminary experiment compared to the f/th substitution. The relative perceptual salience of these features may differ with videotaped stimuli. The t/th substitution. for example. involves only tongue tip variation and has subtle visual differences. The f/th substitution. on the other hand. involves variation of lip movement. which creates a more visually salient difference. Given the fact that SLPs generally assess their clients face to face. this would be an important addition to subsequent studies to determine if different perceptual ratings might vary across visual and auditory stimulus presentations. Task Variation The tasks involved were constructed based on testimonials from nonlinguists in the general population (Alvarez and Kolker. 1986; Niedzielski & Preston. 2004). These lay interviews included statements about dialect comprehensibility and detectability juxtaposed with perceptual statements about dialect density. It was assumed that SLPs would have similar thoughts about dialects. but more empirical evidence of this is needed. SLPs are different from the general population in that they receive a significant amount of information regarding phonological patterns in their academic and clinical education and are likely to have experience assessing and treating phonological disorders. That is. the SLPs had experience analyzing phonology. This is experience that people in the general population are not likely to have. This experience may have caused the SLPs 134 to have heightened rnetalinguistic awareness of the phonological patterns present in each sentence. Such awareness could indicate that further research with SLPs may be conducted with more explicit statement of its aims. For example. SLPs could rate phonological features regarding how highly they associate such features with AAE and. conversely. how greatly they believe the features impact speech intelligibility in the general population. Such an explicitly worded task might demystify much of the thought processes that occurred in this study. In short. what are the folk linguistic theories of SLPs? Additional research also should be conducted that examines the comprehensibility or intelligibility of AAE utterances by the SLPs themselves. rather than the current study that examined perceptions of how comprehensible others might find the stimuli. Although the comprehensibility methods satisfied the purposes of the dissertation's study. the actual difficulty of the SLPs in understanding AAE speech samples is also of interest. For example. the SLP’s own perceptions of intelligibility may correlate higher with the likelihood of misdiagnoses than the SLP's perceptions of comprehensibility to others. Speaker Variation The stimuli in this study were spoken by two adult female AAE speakers. It would have been ideal to include child speakers; however. the speaker's task was judged to be too difficult for a child to do. Now there is some indication regarding the perceptual paradigms that may be used in creating perceptual judgments ofdialect density. The findings of this study may be used to help inform methodology and interpretation of results in subsequent studies using actual child language samples. A Conclusions The data from this study suggest that SLPs will judge speakers as more AAE detectable and less comprehensible to the general population who use more features in a linguistic unit. Furthermore. if speakers use features with greater perceptual salience. then they may be viewed by SLPs as less comprehensible to the general population. At least two factors appeared to influence the degree to which specific types of features contributed to ratings of AAE detectability: (a) perceptual salience and (b) feature iconicity. The findings support several theoretical and research paradigms. however. they suggest that the methods for calculating dialect density production may not generalize to perceptual experiences. F urtherrnore. the findings have several implications for SLPs in traditional practice and those doing accent modification work. Additionally. the study‘s findings highlighted the importance of instilling a socio-cultural view ofcornmunication in SLPs from the beginning of their training. This study was among the first to examine SLPs“ perceptions of dialect use. Much more work needs to be done to illuminate the multifaceted puzzle of dialect density Perception by SLPs and by other professionals with the power to evaluate the communicative adequacy of people. APPENDICES 137 Appendix A Description of Pilot Studv A bstract African American students are commonly misdiagnosed as speech-language impaired by speech-language pathologists (SLPs) in the schools. Although the degree to which the dialect of African American English (AAE) is noticed in these students by SLPs could be a key to this disparity. little research on this area has been done. It is likely that some dialectal features are more noticeable than others. The purpose ofthis project was to determine which phonological features associated with AAE were most noticeable to SLP graduate students who do not speak the dialect. Selected features that are common among AAE speakers were included in nonsense syllables and paired against a Standard American English equivalent. These pairs of different sounds were embedded within a list of recordings that included control pairs that were phonetically similar. The stimuli were recorded by three female speakers and presented over earphones using a participant—guided Power Point presentation. Participants rated the perceived difference between the two sounds in the pair on a 4- point ordinal scale. The participants rated the following phonological features as the most salient. respectively: (a) post-vocalic r alternations. (b) t/th substitution. (c) final consonant deletion. (d) final cluster reduction. and (e) f/th substitution. Appendix A (C ont'd) Introduction This pilot study was designed to determine how noticeable selected phonological features associated with AAE were to SLP graduate students who were unfamiliar with the AAE dialect. The phonological features were placed in nonsense words to determine their perceptual salience apart from a linguistic. or meaningful. context. The study examined the following features (See Table 1): Table 1 The Phonological Processes and Affected Phonemes Used in the Pilot Study. Phonological Process Affected Phonemes Example Final Consonant Deletion /t/ and /k/ fat->fa. tack-) ta Final Cluster Reduction st and sk fast-) fOs or ask-)Os Final th substitution f/th and t/th with-)wif. wit Final r alternation vowel/r and deleted r four-) foe. fo Fourteen participants (13 female and 1 male) listened to 84 pairs of nonsense words spoken by three different female voices. The final consonant of the first syllable varied. while the rest ofthe word remained constant (i.e.. da_me. datrne. dafme. datlrrne. etc.) This syllable structure was used to simulate the phonetic environment that occurred in the dissertation study stimuli. Some of the pairs contained the same nonsense word twice (e. g.. datrne datme). and some contained different nonsense words (e.g. dafrne datrne; see Table 2'). The pairs were blocked by phonological process so that the listeners heard nonsense pairs associated with only one phonological process at a time. Appendix A (C ont‘d) The order of presentation was randomized twice across phonological patterns. speakers. and token presentation. In essence. there were two separate randomized orderings. Table 2 Nonsense words used in the pilot study. F C D C R th sub final r_ datme dastrne dathme dorrne dakme daskrne dafrne dome dame dasrne datrne dawrne The same [me] recording was used for each of the nonsense words within the four groups. Each of these nonsense words were then paired in the following manner with 1.0 seconds of silence generated between the tokens within the pairs and .5 seconds of silence generated before each recording (see Table 2): As the pairs of nonsense words were presented to the participants. the participants indicated the degree to which they felt the two nonsense words in each pair were similar or different ( l= exactly the same. 2: almost the same. 3: significantly different. 4= extremely different). This scale is similar to the one used by Preston (1993) in which participants were asked to detennine the extent to which various dialects in the US. differed from their own (I: no difference. 2: slightly different. 3: different. 4: unintelligibly different). The modification was made to include a forced dichotomy of 140 \ Appendix A (C ont’d) same-different. Results The ratings for the experimental pairs (i.e.. the ones that were different) were subtracted from the control pair that represented the Standard English production. Thus. to test the saliency of final /t/ deletion. the perceptual ratings for the [datme da me] pair and the [da_me datme] pair were subtracted from the ratings associated with the [datme datme] control for each participant. This resulted in a perceptual salience rating for each affected sound under each process. The ratings for each of the two comparison sounds under each process were then tested for difference using the Wilcoxin Signed Rank Test and adjusted for Type 1 error using the Bonferroni method. None of the compared sounds under the phonological processes was found to be statistically significant except for the th substitution process. In this case. t/th substitution was significantly more salient than f/th substitution. The different phonological processes were tested for differences using the Friedman Test. The results showed that the phonological processes received significantly different saliency ratings with the following ranking: Final r processes >>t/th substitution>>Final Consonant Deletion>>Final Cluster Reduction>>flth substitution. Post hoe analysis using the Friedman Test revealed that the final r processes were significantly different from all other processes. and final consonant deletion and t/th substitution were significantly different from final cluster reduction and f/th 141 Appendix A (Cont‘d) substitution. Due to the fact that final r patterns appeared to differ from all of the other. it was excluded from the final stimuli used in the dissertation study. Otherwise. data corresponding to the processes of final consonant deletion differed from the process of final cluster reduction. This line divided the processes into two categories: 1) more salient (t/th substitution and final consonant deletion of t and k) and 2) less salient (f/th substitution and Final Cluster Reduction of st and sk). This categorization will be used to group the sounds to be tested into the two groups of more salient and less salient for comparison. 142 Appendix B Qualifying Questionnaire I. How many years have you been practicing as an SLP? 2. How many clients have you had in the last two years who spoke African American English? a. <3 b. 3-5 c.>5 ‘5 3. How frequently over the past three years have you communicated with someone who speaks AAE? a. < 5% of the time. b. 5-15% of the time. c. 15-25% ofthe time. d. > 25% of the time. 4. Do you speak African American English? Yes No Appendix C Human Subjects Application APPLICATION FOR INITIAL REVIEW APPROVAL OF A PROJECT INVOLVING HUMAN SUBJECTS Biomedical, Health Sciences Institutional Review Board (BIRB) Social Science, Behavioral, Education Institutional Review Board (SIRB) 202 Olds Hall, Michigan State University East Lansing, MI 48824-1047 Phone: (517) 355-2180 Fax: (517) 432-4503 E-mail: irb@msu.edu Office Hours: M-F (8:00 A.M.-5:OO PM.) Jacki—687316 ‘ ID# 1022218 3 ‘11_-_i.'.- Fu-"nLL'.-‘. .J 71a. " ; RespOnsible-PAr-OieCt Investigator: ' Name: Ida STOCKMAN i ID#: xxx-xx-9247 l Department: Audiology College: COM ARTS 8: SCI Academic Rank: Professor ' Mailing Address: 371 Com. Arts and Sciences 1 Phone: 3-6764 '; Fax: 517-432-2370 Email: stockma1@msu.edu .1b. gSeciOflndia'rylnvestigatorzi T Name: Gregory Robinson 3 lD#: xxx-xx-5545 E Department: COMMUNICATION § College: COMM ARTS & $01 I Academic Rank: NA Mailing Address: 1028 Parker 144 Lansing, MI 48912 ‘ Phone: 371-4345 Fax: , Email: robin465@msu.edu 1c. Additional Investigators: T I I : Peter R. LAPINE i Dennis PRESTON j Brad RAKERD Q2". " anaemia.“ A ‘ “ lName: ID#: 1 Department: 9 College: : Academic Rank: i Mailing Address: 2 Phone: ; Fax: Email: 3‘. "I friendlier-eject: p'E'nEE‘p’ndné’dE KER—ICAN AMERICAN—ENv—GLISHHDIALECT DENSITY? ‘ BY SPEECH-LANGUAGE PATHOLOGISTS IN PREDOMINANTLY WHITE SCHOOL DISTRICTS :4. j Hfidveyodiévér’iecelvéd PreflliminaryApprovalfor—thisuprOjecit?’ * No [5. iCategory OfflReviewM ”CT i i ' ' 72.751217 " ? i .EXPEDITED 86. ”TIE-{mg project being conducted to fulfrlltTrCrequrrements of an i ”PhD-h “ if I education/training program? : Dissertation 7a. r Funding: i i i - Projec—tis not 145 7d _ ' 8a. 8b. €10.“ E11. “12. '13. :14. :15. :16. _.... _ _ _...___...h_ _.._ I The protection Of human subjects Often requires resources be dedicated fOr things such as the consent process (space, personnel), the performance of the research funded 7 (trained personnel interacting with subjects), care Of subject Issues or injuries (counseling, medical care), confidentiality of data (space, equipment) and other ' monetary and non-monetary resources. Describe the resources that are available for ‘ this project for the protection Of human subjects. 1 The secondary investigator (Gregory Robinson) will be present during all data gathering tO ensure that all rights are protected The data will be compiled and encoded through the University scoring Office to maintain data integrity. List all sites where this research will be conducted Ionia ISD Office Oyer Speech- ”Language -Hearing Clinic .-qu ..A .‘.4__...__-..-...-... Do any Of these sites have their own IRB? EDO you have any related project that were approved by an MSU IRB? I (a) IRB Numbers: i021555 I Have you submitted this to anOther |RB(s)? ‘ ls anOther institution(s) relying on MSU' s IRB as the IRB of record? __..,.__,_;__ -m.__- __._._V_._-_..~_ Are you using an FDA approved drug/devrceldragnostic test? Are you using an FDA approved drug/device/diagnostic test for a non FDA approved indication? Has this protocol been submitted t6 the FDA or are there plans to submit It to the FDA? Does this prOject involve the use Of Materials of Human Origin (e. g. 1 human blood or tissue)? .-_ ,-,_,_.._.....-......__s ...w— Research Category Education Research FF Survey/Interview 17 AudiONideO Recording p- Oral History I" I Internet-based I— ' Analysis Of Existing Data {— I International Research I— Yes Yes Yes Yes Yes Yes Yes v-" - ‘fimafi ,.--w.-.-.'. ...._--. ,2 -..E. “w —.-—...—._ Gene Transfer Research [- Fetal Research Medical Records Stem Cell Research Medical Imaging 71—1—17 Oncology Clinical Trial (specify below) I— Surgical '— Therapeutic 146 Yes Yes Yes Yes Yes Yes *YES :NO (NO NO .NO NOW V NO a... ......._ ...mw- ..-——__._1...r_m_-__-. .. . ..-...- . ._ ...... -- ....... _--._..__.. . ‘-_.__._. _-__.-.._..-.__. Prevention Other r_NO ' 17. Project Description (AbstraCt) . i I Dialect density” is used by language researchers tO describe inter-speaker differences within 5 dialect groups. When descriptions like these are used by lay persons, they often coincide with E statements regarding comprehensibility (how easy tO understand the speaker was) or dialect I Edetectability (how noticeable the dialect was). These judgments may affect speech- language assessment and intervention. For example, a speech- language pathologist (SLP) may be E more likely tO mistake a speaker of African American English (AAE) for a speech-language ‘ disorder in the public schools, particularly if the student is judged to be low in E comprehensibility and high in dialect detectability. Since phonological features may be the most noticeable, this study aims to answer the following question: How are judgments of comprehensibility and dialect detectability influenced by the number and type of AAE _ phonological features present within sentences? The hypothesis: sentences with more E features and more salient features will impact judgments the greatest. 2 Certified SLPs from predominantly White school districts in mid-Michigan will listen to . individual sentences containing varying amounts and types Of AAE phonological features. I The SLPs will rate each sentence regarding perceived dialect detectability and E comprehensibility. The ratings for the different sentences will be compared tO determine how 3 the number and type Of features present in each sentence contributed tO the SLPs’ - judgments l 18. EProcedures i The procedures will be organized according to the three different groups of subjects used in i this study: 1) listeners, 2) speakers. and 3) naturalness raters. i The lIsteners Will be gIven a hearing screening and answer a brief qualifyIng questionnaire. E After qualifying for the experiment. they will view a Power Point presentation at a personal ; computer. The Power Point presentation will contain sound clips that contain sentences read i with different types and occurrences Of phonological features commonly associated with ; AAE. The participants will listen tO the sound clips Over earphones tO block out ambient ' noise They will read instructions that orient them to the experiment tasks and the pragmatic , Icontext for the recorded sentences They will then listen to sound clips presented in individual I sentences or in groups Of sentences. They will rating the clips regarding 1) how . . understandable they believe the speakers would be tO the general population (1= Very easy to understand, 5: Very difficult tO understand). and 2) how noticeable the use Of African _ American English is in each sentence (1= Not noticeable. 5= Extremely noticeable). They will i record their responses onto a computer-readable “bubble" sheet with a nO. 2 pencil. After completing all Of the ratings. each participant will place his/her response sheet into an envelope and seal it. E The speakers will be recorded using a headset microphone and digitized using Wavesurfer I software (Sjolander & Beskow, 2004) at a sampling rate Of 22,050 Hz. Each speaker will be ‘ instructed to think about the context that will be presented to the listeners and naturalness raters (See Verification Of Naturalness for a description Of the context) They will then be . l encouraged to practice each sentence before recording. Once a self-determined comfortable ' I :intonation pattern is produced, the speakers will be instructed tO read each sentence again . with the same intonation pattern. They will be recorded during this reading. but will not be 'E instructed tO use any specific AAE phonological features. These sentences will be used as I phonetically consistent carrier phrases. Only the recordings Of the non-variable words will be I used from this reading. Immediately following this first recorded reading, the speakers read ' the sentences aloud again. This time, they make specific phonological changes that either correspond to an SAE production or one Of the conditions for AAE productions (varying in 147 A _ +— . .. -_...__...,_..— ..w— P..— -_-. ,_ A . - number and type Of features included). The variable words from these recordings will be , i embedded into the appropriate position Of the previously recorded carrier phrases. I I The naturalness raters will be given a hearing screening. If they pass the screening, they will ' then be presented with sound clips presented in a Power Point Presentation. Prior to the : Iratings, the naturalness raters will view a series Of slides that orient them to their task and the pragmatic context for the sentences they will be hearing. Each slide will include three . different recordings that include exactly the same wording and planned phonological features. ' I The naturalness raters will listen to the three recordings on each slide tO determine which ‘ recording out of the three presented sounded the most natural, ” or the most like what would actually be heard in conversation. They will be rating a total of 36 slides. --~——‘:vw-' --M—d__.m____r__.-., 19. I Does your investigathn involve incomplete disclosure Of the ' NO I research purpose or deception Of the subjects? ' an." 1"". “ “-7 2%. Subject Population I Listeners—16 White adult female Speech-Language Pathologists in predominantly White ‘ school districts. Speakers—2 adult Female African Americans I Naturalness Raters—6 adult African Americans (3 men and 3 women) I20b. IAge range ofsubjects: VI18tO 65 I205. {IEQIJEI} populationmayinclude " i I Target Populatlon: lncidentallnclusion: I— Minors I— Minors . I— PregnantWomen I; PregnantWomen I“ Women Of Childbearing Age l’" Women Of Childbearing Age I" Institutionalized Persons [— Institutionalized Persons ’ I— Students [7 Students . In Low Income Persons [.7 Low Income Persons E [‘7 Minorities '7 Minorities ; I— Prisoners l— Prisoners l— HlV/AlDSlndividuals '— HlV/AlDSlndividuals 3 [— Psychiatric Patients '— Psychiatric Patients i [— Incompetent Persons I— Incompetent Persons I20d. ExpeCted number Of subjects (Including con—t‘rOl-sI—I I I E24- I20em— Justify your sample—vs'i:#w MM“ .__..._.._..- w“ ”I W - I . Listeners= 16 E Speakers= 2 . Naturalness Raters= 6 ' Power analysis for repeated measures ANOVA suggested at least a sample size of 9 for Power- - ..80 Four CD compilations, require at least 4 subjects with each compilation. "H“--. ._._. w-—_ __ _._..___ k.— -.._.4.___..__.__._.__ ., I48 ...fi...___.m. ”in. ,_._..._..__._ ._... ... ..._ .‘._.. ___..__._.._. ._._. _ ,,__,,_.. , -.A.” , ,_ .2... .2...... r, _, i, 20f. Describe the criteria for the inclusion of subjects. E 1) Listeners—Certified SLP, White, working in predominantly White school district, minimal ' exposure to African American English, at least 2 years experience as SLP. 2) Speakers— . Adult female, African American, Native Speakers of AAE, Able to include or not include AAE J features in speech upon cue. 3) Naturalness Raters—African American adults, native ‘ speakers of AAE __.._..._--.. ...__ -- ..h-....... _._ _ ._ __.. u.—__ -...-.-—. mar... . -....- _ _ ..-._- —...- . -.-_-.. _. ___.-... 20g. * Describe the criteria for the exclusion of subjects: E 1) Listeners—not certified, not White, non-native English speaker, significant daily exposure 3 to AAE, less than 2 years SLP experience, failed hearing test. 2) Speakers—Not female, not , E African American, not native speakers of AAE, Unable to include or not include AAE features ' E upon cue. 3) Naturalness Raters—not African American, not native speakers of AAE, failed 1‘ hearing test. ___r ._-.—-_. m...“ _. .._........__._ v 20h(1) How will the subjects be recruited? E ; The participants will be recruited from school districts in a rural mid-Michigan county that has a student population that is 90% White/Caucasian-American and less than 5% African EAmerican as indicated by the county‘s census from the preceding school year 2003 (Annie E. E ; Casey Foundation). The researcher will attend a county-wide SLP meeting. During the E meeting the study will be described and a sign—up sheet will be distributed to collect contact . 3 information. One month prior to the experiment, the SLPs will be contacted via telephone and ' personal interaction for scheduling The SLPs will be compensated monetarily for their time. The speakers will be recruited from a university theater program. EEThe naturalness raters will be recruited through personal contacts ’2°h(2) EWiIl an advertisement be used? ‘ N O ' ' zoi. » Are you associated with the subjects? gYES' E (a) Nature of the association and what measures you are taking to E protect subjects' rights, including safeguards against any coercion. I work with many of the listeners as a fellow Speech- -Language Pathologists i am not associated with the speakers or naturalness raters. E 20j. EWill someone receive payment for recruiting ”the subjects? - NO :20k._ Will the research subjects be compensated? YES N L (1) Details concerning payment, including the amount and schedule of payments including any conditions: g Listeners: $15 (one time after completion of tasks or discontinuation) E Speakers: $20 (one time after completion of tasks) g Naturalness Raters: $10 (one time after completion of tasks or E discontinuation) 20L” l Will the subjects incur additional financial costs as a result of their I NO E E participation in this study? ‘ 20m. ‘ Will this research be conducted with subjects in another country? f NO 20h. EWill this research be conducted with subjects in the U. S. from an E YES ethnicE group of sub-group or other non-mainstream minorities 149 m ...fi..-_.——..—. _.._.“5 ...- .__.... ##7 Q (including non- English speakers)? (1) What country/sub-group are they from? African Americans I (2) Does the different cultural context present any problems or risks ; that need to be addressed? If so, describe the issues and how you will address them. .No 21a. Risks and Benefits for subjects: Describe and assess any potential risks (physical, ' psychological, social, legal, economic) and assess the likelihood and seriousness of 3 such risks. , Minimal risks associated with discomfort wearing headphones or viewing a computer screen . 21b. I Describe procedures for protecting against or minimizing potential risks and provide ‘ . an assessment of their likely effectiveness. I Direct Supervision from secondary investigator Gregory Robinson. 21c. IAssess the potential benefits (if any) to be gained by the sujbects in this study, as well as benefits which may accrue to society in general as a result of the planned work. I i ; Benefits to subjects? Payment, hearing screening 1 Benefits to Society? Understanding of perceptual obstacles that may impact the accuracy of I . speech-language assessments with African American clients when conducted by White SLPs 1 I and may help inform goals and procedures of intervention with clients seeking dialect I modification services from SLPs. The findings may also help inform theories of speech I perception. 22a. 1 How will the subject' 3 privacy be protected? Include a description of who will be 3 interacting with the subjects or accessing and abstracting data from the subject's records (academic, medical, etc.) and where the study will take place. For example, will , 1 individuals not associated with the research study be present during the consent I process and the conduct of the study? 1 No identifying information (except school district) responses pieced in sealed envelopes .—~» ”4---... ____..__...._....... ._._.- m- —.~__._.... ..._..___..__....._ w_....__ _‘ 22b. f Where will the data be stored and for how long? The paper responses will be stored in my personal office. Upon receipt, the data will be entered into my person computer. The data will be stored on my personal computer located in my personal office and will be backed up through my personal, password-protected e-mail account as well as a compact disc The compact disc and the paper response sheets will be kept indefinitely In my personal research archival files. .-__- .....—.. q..- ~.—.—.-—.-o+.—.—..-___ .— -__.-.I-. .u—. ., -—-. 22c. IWho will have access to the research data? [ Access to the data will be limited to me and my dissertation committee members: Ida ‘ Stockman, Peter Lapine, Dennis Preston, and Brad Rakerd. 22d. How will you ensure the confidentiality and/or anonymity of the research data? Include , a description of the procedures and safeguards you will use, Including If identifying [50 Iaiaim'auan will be stored-with the data. ‘ The security and integrity of the data will be ensured by backing up the files in three ‘ locations: on a personal web-based e-mail account. a compact disc. and a file on the hard , ‘ drive of my personal computer The paper documents will be placed in my personal office in a 1 file cabinet The office will be locked when I am not present.’ ..___ __ ___;-A_A_ _-___..,._.‘ s M... ____—__H. ._._,.__. .___ _. , 23. Does this project involve protected health information as defined by NO HIPAA? I No 24. §(a) Select appropriate consent option. 7 Approval of a consent form and process . 4—‘—. (b) Consent Procedures: l Gregory Robinson, will personally hand the , consent form to each participant. The participants will sign a consent form T (See appendix) that will state that there are minimal risks associated with 5 the experiment and will outline their rights. The participants will be asked i to read the form carefully and ask me any questions about it that they ' might have. I will ask them if they have any questions about the form before they begin the experiment. i 25a. mi Have you or will you or a member of your immediate family receive, NO ‘ " from the sponsor of the research, financial or other forms of - ' compensation? : 25b. Do or will you or a member of your immediate family have a ;' NO significant financial interest in the company/agencylt” rm that is to . sponsor the research? 25c i Are you submitting FDA form 3454 or 3455 (Conflict of Interest)? NO 26a. i When would you prefer to begin this project? ,4/15/2005 26b. iEstlmated duration of project: ; 04/15/2005 ADDITIONAL DOCUMENTS/ATTACHMENTS 01. 4/5/2005 Consent Form (i022218_ 4- 05- 05_ Consent Forms_ DIssertatIon pdf) i 02. 4/5/2005 Other (i022218_4-05-05_METHOD_Human_Subjects.pdf) 03. 4/25/2005 Consent Form (i022218_4-22-05_Consent_Forms_Dissertation.pdf) 151 Appendix D (.‘onsem F arms .S‘peakers Perceptions of African American English Dialect Density by Anglo-European American Speech-Language Pathologists Consent Form You are being asked to participate in a study that examines the ability of speech-language pathologists to comprehend and detect African American English (AAE) pronunciation patterns. In this study. you will be asked to read a list of six sentences multiple times. During your readings of the sentences. you will be asked to pronounce various words in the sentences in different ways. You will be encouraged to practice reading the sentences aloud. Your readings will be recorded through a headset microphone. You will be allowed to take a break in the middle, if you like. The entire process should take approximately 1.5 to 2 hours to complete. The risks associated with this experiment are minimal and involve primarily the discomfort with wearing a headset. If you feel any discomfort in any way, please notify Greg immediately to alleviate your discomfort. Your participation in this study is completely voluntary. You will receive $20 for your participation. If you choose to discontinue your participation at any point, for any reason. you will still receive your payment. Your privacy will be protected to the maximum extent allowable by law. Your identity will be kept confidential to the other research participants. The following people will have access to the data and recordings associated with the experiment: Gregory Robinson. Ida Stockman, Peter Lapine, Brad Rakerd. and Dennis Preston. If you have any questions about this study, please contact the investigator, Gregory C. Robinson, at 517-371-4345 or by e-mail at robin465@msu.edu. If you have questions or concems regarding your rights as a study participant, or are dissatisfied at any time with any aspect of this study, you may contact — anonymously, if you wish —Peter Vasilenko, Ph.D., Chair of the University Committee on Research Involving Human Subjects (UCRIHS) by phone: (517) 355-2180. fax: (517) 432-4503. e- mail: ucrihsfiimsuedu. or regular mail: 202 Olds Hall, East Lansing, MI 48824. Your signature below indicates your voluntary agreement to participate in this study. Printed Name Signature Date Appendix D Continued Consent Forms Ar’atzn'uliless Raters Perceptions of African American English Dialect Density by Anglo-European American Speech-Language Pathologists Consent Form You are being asked to participate in a study that examines the ability of speech-language pathologists to comprehend and detect African American English (AAE) pronunciation patterns. Before beginning the experiment. you will be given a hearing screening which will show whether you can hear tones presented over earphones. You will get the results of the screening immediately after its completion. After passing the hearing screening, you may begin the experiment. In this study. you will view a Power Point presentation during which you will listen to sentences through earphones. After listening to each group of sentences, you will be asked to indicate which one out of the group was the most natural sounding. There will be a total of 72 groups of sentences. You will be asked to take a 5 minute break in the middle. The entire experiment should take approximately an hour to complete. The risks associated with this experiment are minimal and involve primarily the discomfort with wearing a headset and visual discomfort associated with looking at a computer screen during the extent of the experiment. If you feel any discomfort in any way, please notify Greg immediately to alleviate your discomfort. Your participation in this study is completely voluntary. You will receive $10 for your participation. After you pass the hearing screening, if you choose to discontinue your participation at any point, for any reason, you will still receive your payment. Your privacy will be protected to the maximum extent allowable by law. You will not be asked to provide any identifying information on the response sheet except the school district in which you work. After you complete your responses, you may place your response sheet into the envelope provided so that your responses can not be associated with your identity. The following people will have access to the data: Gregory Robinson (investigator), and his dissertation committee members: Ida Stockman, Peter Lapine, Brad Rakerd, and Dennis Preston. If you have any questions about this study, please contact the investigator, Gregory C. Robinson, at 517-371-4345 or by e-mail at robin465@msu.edu. If you have questions or concems regarding your rights as a study participant, or are dissatisfied at any time with any aspect of this study, you may contact — anonymously, if you wish —Peter Vasilenko, Ph.D., Chair of the University Committee on Research Involving Human Subjects (UCRIHS) by phone: (517) 355-2180. fax: (517) 432—4503, e- mail: ucrihs@msu.edu, or regular mail: 202 Olds Hall, East Lansing, MI 48824. Your signature below indicates your voluntary agreement to participate in this study. Printed Name Signature Date 153 Appendix D C ()ntinued Listeners Perceptions of African American English Dialect Density by Anglo-European American Speech-Language Pathologists Consent Form You are being asked to panicipate in a study that examines the ability of speech-language pathologists to comprehend and detect African American English (AAE) pronunciation patterns. Before beginning the experiment, you will be given a hearing screening during which will show whether you are able to hear tones presented over earphones. You will get the results of the screening immediately after its completion. After the hearing screening, you will be asked to answer a brief questionnaire (4 questions) that will determine if you qualify for the study. After passing the hearing screening and qualifyinO, you may begin the experiment. In this study, you will view a Power Point presentation on a computer during which you will listen to sentences through earphones. After listening to each sentence, you will be asked to make two different perceptual ratings about the sentence. There will be a total of 42 sound clips. You will be asked to take a 5 minute break between tasks. The entire experiment should take approximately 30 minutes to complete. The risks associated with this experiment are minimal. You may experience some discomfort wearing the earphones and visual discomfort associated with looking at a computer screen during the course of the experiment. If you feel any discomfort, please notify Greg immediately to alleviate your discomfort. Your participation in this study is completely voluntary. You will receive $15 for your participation. After qualifying for the study, if you choose to discontinue your participation at any point. for any reason, you will still receive your payment. Your privacy will be protected to the maximum extent allowable by law. You will not be asked to provide any identifying information on the response sheet except the school district in which you work. After you complete your responses, you may place your response sheet into the envelope provided so that your responses can not be associated with your identity. The following people will have access to the data: Gregory Robinson, the investigator, and his dissertation committee members: Ida Stockman, Peter Lapine, Brad Rakerd, and Dennis Preston. If you have any questions about this study. please contact the investigator, Gregory C. Robinson. at 517-371-4345 or by e-mail at robin465@msu.edu. If you have questions or concems regarding your rights as a study participant, or are dissatisfied at any time with any aspect of this study, you may contact — anonymously, if you wish —Peter Vasilenko, Ph.D., Chair of the University Committee on Research Involving Human Subjects (UCRIHS) by phone: (517) 355-2180, fax: (517)432-4503. e- mail: ucrihs@msu.edu, or regular mail: 202 Olds Hall, East Lansing, MI 48824. Your signature below indicates your voluntary agreement to participate in this study. Printed Name Signature Date Appendix E Naturalness Rating Response Form Which sentence sounded the most like you might hear someone say it in real life? Write a letter a. b, or c corresponding to the sentence that you felt sounded the most natural. I 19. 2 20. 3 21. 4 22 5 23 6 24. 7 25. 8 26. 9 27. 10. 28. 11. 29. 12 30 13. 31. 14. 32. 15. __ 33. 16. 34. 17. 35. 18. 36. 155 Naturalness Rating Results Appendix F The naturalness raters selected one sentence out of each set of three recordings (A, B. or C) that sounded the most natural to them. The table displays the “result,” or the recording that was chosen most frequently by the six raters for each stimulus sentence. Phonological Features: F Dk=Final Deletion of /k/, F Dt=F inal Deletion of /t/, FRk=Final Cluster Reduction of sk, FRt=Final Cluster Reduction of st, t/th=Substitution of [t] for /0/, and f/th=Substitution of [f] for /0/. NR = Naturalness Rater. f/th f/th f/th Stimulus Sentence Compilation 1 Compilation 2 Results Type of No. of Speak- Fea- Fea- NR NR NR NR NR NR Re- To- # er tures tures 1 2 3 1 2 3 suit tai 1 1 FDk 0 A A A C C A A 4 2 1 FDk 1 A C C C B B C 3 3 1 FDk 3 A B B C B A B 3 4 1 F Dt 0 C C C C B B C 4 5 1 FDt 1 B C C B C C C 4 6 1 FDt 3 B C C C C C C 5 7 1 t/th 0 A B A A A B A 4 8 1 t/th 1 B A B B B C B 4 9 1 t/th 3 B B B B B B B 6 10 1 FRk 0 A C C B C C C 4 11 1 F Rk 1 C B C C C A C 4 12 1 FRk 3 A B B B C C B 3 13 1 F Rt 0 C B C C B C C 4 14 1 FRt 1 A B C C C B C 3 15 1 FRt 3 A A C C C B C 3 16 1 f/th O A C B A B A A 3 17 1 f/th 1 C C A A A A A 4 18 1 f/th 3 B B C A C C C 3 19 2 F OK 0 A B C C C C C 4 20 2 FDk 1 A A C C B C C 3 21 2 FDk 3 B B A B B A B 4 22 2 FDt O A A C C C C C 4 23 2 F Dt 1 C C C C B B C 4 24 2 FDt 3 C A C A A A A 4 25 2 t/th 0 B B B B B B B 6 26 2 t/th 1 B B C C C A C 3 27 2 t/th 3 A B B A B B B 4 28 2 F Rk O C B C C B C C 4 __29 2 F Rk 1 B B C C C C C 4 3O 2 FRk 3 C B C C B C C 4 31 2 FRt O A B A B B A B 3 32 2 FRt 1 A C A C C C C 4 33 2 FRt 3 C C B B C C C 4 34 2 O A C A C C C C 4 35 2 1 C C B C B C C 4 36 2 3 C B C B B B B 4 156 REFERENCES Adams, C. L. ( I981). The influence of counselor race, attire, and speech style on Black juvenile delinquents' perception of counselor attractiveness and ability to be warm, empathic. and genuine. Dissertation Abstracts International, 42(3-A). 1324. Alvarez, L., & Kolker, A. (Writer). & L. Alvarez & A. Kolker (Director) (1986). American Tongues [Motion Picture]. In L. Alvarez & A. Kolker (Producer). United States: Center for New American Media, Inc. Anderson. R. T. (2002). Practical assessment strategies with Hispanic students. In A. E. Brice (Ed.), The Hispanic child: Speech, language, culture and education. (pp. 143-184). Boston, MA: Allyn and Bacon Press. Annie E. Casey Foundation. (2003). KIDS Count Data. Baltimore. MD: Author. ASHA. (1983a). Implications ofthe position on social dialects. Asha, 25(9). 25-27. ASHA. (1983b). Social dialects. Asha. 25(9). 23-24. ASHA. (1995). We can do better. Recruiting, retaining. and graduating African American students. Asha, 3 7(2), 54-56. ASHA. (1999, May 2004). Standard 3.0 Curriculum (Academic and Clinical Education). Retrieved December 17, 2004, 2004, from http://www.asha.orgr’abOIIL/academia’accreditation/standards.htm ASHA. (2001). Scope ofpractice in speech-lung:rage pathology. Rockville. MD: Author. ASHA. (2005. February 8, 2006). Highlights and trends: ASHA member counts for 2005. Retrieved February 25. 2006 from http://\\\nv.asha.org/memberslrescarch/rcports/member-counts.htm Bailey. G.. & Thomas. E. R. (1998). Some aspects of African American Vernacular English phonology. In S. Mufwene. J. Rickford, G. Bailey & J. Baugh (Eds). African American English (pp. 85-109). London: Routledge. Battle. D. E. (2002.). Communication disorders in multicultural populations (3rd ed.). Boston, MA: Butterworth—Heinemann. Baugh. J. (1983). Black street speech. Austin. TX: University of Texas Press. Belton, M. A. (1984). The effect of therapist dialect upon the Black English speaking patient's perception of therapist empathy, warmth, genuineness, and expertlse. Dissertation Abstracts International, 44(8-B), 2547. Brennan. E., & Brennan. J. (1981a). Accent scaling and language attitudes: Reaction to Mexican-American English speech. Language and Speech, 24, 207-221. Brennan, E., & Brennan, J. (1981b). Measurements of accent and attitude toward Mexican American speech. Journal of Psycholinguistic Research. 10, 487-501. Brice. A. E. (2002). Normal Processes of second language acquisition. In A. E. Brice , (Ed). The Hispanic child: Speech, language, culture and education. (pp. 37-53). Boston, MA: Allyn and Bacon Press. Brook, G. L. (51970). The ltmguage of Dickens. London: Andre Deutsch. Burda. A. N., Scherz, J. A., Hageman, C. F., & Edwards, H. T. (2003). Age and Understanding Speakers with Spanish or Taiwanese Accents. Perceptual and Motor Skills, 97(1), 11-20. C heng. L. L. (1987). Assessment procedures. In L. L. Cheng (Ed.), Assessing Asian language performance (pp. 135-159). Rockville, MD: Aspen Pubhshers. Cohen. J. (1988). Statistical power analysisfor the behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Earlbaum Associates. Craig. H. K., Thompson, C. A., Washington. J. A., & Potter, S. L. (2003). Phonological features of child African American English. Journal of Speech-Langm‘tge- Hearing Research, 46(3), 623-635. Craig, H. K., & Washington, J. A. (2000). An assessment battery for identifying language impairment in African American children. Journal of Speech-Language-Hearmg Research, 43(2), 366-379. Craig, H. K.. & Washington, J. A. (2002). Oral language expectations for African American preschoolers and kindergartners. American Journal of Speech- Language-Pathway, l 1 (1), 59-70. Craig, H. K., & Washington, J. A. (2004). Grade-related changes in the production of African American English. Journal of Speech-Language-Hearing Research. 47(2), 450-463. Craig. H. K., Washington, J. A., & Thompson-Porter, C. (1998a). Average C-unit lengths in the discourse of African American children from low-income, urban homes. Journal QfSpeech-Language-Hearing Research, 41(2). 433-444. Craig. H. K., Washington, J. A.. & Thompson-Porter. C. (1998b). Performances of young African American children on two comprehension tasks. Journal of'Speech- L(mgrurge-Hearing Research, 41(2), 445—45 7. Dale. C .V. & Gurevitz. M. (1995. October 17) Legal analvsis of proposals to make English the official language ofthe United States. Washington, DC: Congressional Research Service. Dallal. G. E. (2003. July 27, 2003). Randomization. com. Retrieved January 12, 2005. from http://wuw.randomization.com de Bardelaben. G. (1980). Examiner's democratic values, Black dialect and the clinical perception of intelligence. Dissertation Abstracts International. 41(1-B), 346-347. Derwing. T. M., & Munro, M. J. (1997). Accent, intelligibility and comprehensibility: Evidence from four Lls. Studies in second language acquisition, 19(1), 1-16. Edwards. J. R. (1982). Language attitudes and their implications among English speakers. In E. B. Ryan & H. Giles (Eds), Attitudes toward language variation (pp. 20-33). London: Arnold. Fasold. R. W. (1981). The relation between Black and White speech in the South. American Speech, 56(3), 163-189. Fasold. R. W. (1999, May). Ebonic need not be English. Paper presented at the Georgetown University Round Table, Washington DC. Flege. J. E.. & Fletcher. K. L. (1992). Talker and listener effects on degree of perceived foreign accent. Journal ofthe Acoustic Society of America, 91 (1), 370-389. F lege. J. E., & Hammond, R. H. (1982). Mimicry of non-distinctive phonetic features between language varieties. Studies in Second Language A cquisition.. 5, 1-17. Gass. S.. & Varonis. M. (1984). The effect of familiarity on the comprehensibility ofnon- native speech. Language Learning, 34, 65-90. Gerken. K. C.. & Deichmann. J. W. (1979). The ability of listeners to report oral responses of Black and White children. Language. Speech, and Hearing Services in Schools, 10(1), 35-46. Googlc. (2005). Google search engine. Retrieved February 15. 2005. 2005. from http://\mvw.google.com 159 Greenberg, S., Hollenback, J. and Ellis. D. (1996) Insights into spoken language gleaned from phonetic transcription of the Switchboard corpus. In Proceedings ofthe Fourth International Conference on Spoken Language (ICSLP), Philadelphia. 824-27 Graff, D., Labov, W., & Harris, W. (1983). Testing listeners' reactions to phonological markers of ethnic identity: A new method for sociolinguistic research. In D. Sankoff (Ed), Diversity and diachrony (pp. 45-58). Amsterdam/Philadelphia: Benjamins. Gutierrez-Clellen. V. F.. C alderon, J .. & Weismer, S. E. (2004). Verbal working memory in bilingual children. Journal (y‘Speech. Lang-zu‘tge, and Hearing Research. 4 7(4), 863-876. Guy. G. (1980) Variation in the group and the individual: The case of final stop deletion. In W. Labov (Ed), Locating language in time and space (pp. 1-36). New York: Academic Press. Hayes, J .R., & Hatch, J .A. (1999) Issues in measuring reliability: Correlation versus percentage of agreement. Written Communication, 16(3), 354-367. Hemingway, B. L., Montague, J. C ., & Bradley, R. H. (1981). Preliminary data on revision of a sentence repetition test for language screening with Black first-grade children. Language, Speech. and Hearing Services in the Schools. 12, 145-152. Hodson, B.W. (2004) Hodson Assessment o/‘Phonological Processes (3rd edition: HAPP- 3). Austin. TX: Pro-Ed. Irvine, J. T. (2001 ). “Style” as distinctiveness: The culture and ideology of linguistic differentiation. In J .R. Rickford & P. Eckert (Eds) Style and sociolinguistic variation (pp. 21-43). New York: Cambridge University Press. J ohansson, S. (1978). Studies of error gravity--Native reactions to errors produced by Swedish learners of English (Gothenburg Studies in English No 44). Gothenburg. Sweden: Department of English, University of Gothenburg. Johnson, F. L., & Buttny, R. (1982). White listeners' responses to "sounding black" and "sounding white": The effects of message content on judgments about language. Communication Monographs, 49(1), 33-49. Johnston. J. R., & Wong, A. M.-Y. (2002). Cultural differences in beliefs and practices concerning talk to children. Journal of Speech, Language, and Hearing Research, 45(5). 916-926. 160 Junker. D. A.. & Stockman. I. J. (2002). Expressive vocabulary of German-English bilingual toddlers. American Journal ofSpeech Language Pathologv, 1 l, 381- 394. Kent. R. D., Miolo. G., & Bloedel. S. (1994‘). The intelligibility ofchildren's speech: A review of evaluation procedures. American Journal ofSpeech Language Pathologv. 3, 81—95. Klee. T.. Stokes. S. F ., Wong. A. M.-Y., F letcher, P., & Gavin. W. J. (2004). Utterance length and lexical diversity in Cantonese-speaking children with and without specific language impairment. Journal of Speech. Language, and Hearing Research. 47(6), 1396-1410. Kuhl. P. K. ( 1994). Learning and representation in speech and language. Current Opinion in Neurobiologv, 4(6), 812-822. Labov, W. (1966). The social stratification ofEnglish in A-"ew York C ity. Philadelphia: The Center for Applied Linguistics. Labov. W. & Ash. S. (1997). Understanding Birmingham. In Bernstein, Nunnally & Sabino (Eds). Language variety in the south revisited (pp. 508-5 73). Tuscaloosa: University of Alabama Lambert, W. E., Hodgson. R., Gardner. R. C.. & Fillenbaum, S. (1960). Evaluational reactions to spoken languages. Journal ofAbnormal and Social Psychologi-r. 60. 44-51. Lane. H. (1963). Foreign accent and speech distortion. Journal ofthe Acoustical Society ofAmerica, 35, 451-453. Leonard. L. B. (1989). Language leamability and specific language impairment in children. Applied Psycholinguistics, 10. 179-202. Leonard. L. B. (1999,). The study of language acquisition across languages. In 0. L. Taylor & L. B. Leonard (Eds), Language acquisition across North America: C ross-cultural and cross-ling:tistic perspectives (pp. 21-38). San Diego: Singular Publishing Group. Levey. S. (2004). Speech-language pathology students' awareness of linguistic and cultural differences. Perspectives on issues in higher education (American Speech-Language-Hearing Association Division 10), 7(2). 2-5. Linguistic Society of America. (1997). Resolution of the Linguistic Society of America on the Oakland "Ebonics" issue. The Modern Language Journal, 81, 14. Luelsdorff. P. (1975). A segmental phonology ofB/ack English. The Hague: Mouton. 161 Malmkjaer, K. (Ed). (1991). The linguistic encyclopedia. New York and London: Routledge. Marston, A. D. (1974). The effect of American regional dialects upon speaker credibility and perceived personality. Dissertation Abstracts Internationc'tl, 34 (9-A, Pt 2), 6154-6155. Massaro, W. (.1998) Perceiving talkingfaces (from speech perception to a behavioral principle). Cambridge: The MIT Press. McCarthy, J. (1986). OCP Effects: Gemination and antigemination. Language Inquiry. 17, 207-263. Mulac. A.. & Rudd, M. J. (1977). Effects of selected American regional dialects upon regional audience members. Communication Monographs, 44(3), 185-195. Mufwene, S. (1998) What research on creole genesis can contribute to historical linguistics. In M. Schmid, J. Austin, and D. Stein (Eds) Historical linguistics 1997 (pp. 315-338). Amsterdam: John Benjamins. Munro. M. J ., & Derwing, T. M. (1995a). Foreign accent, comprehensibility, and intelligibility in the speech of second language learners. Language Learning, 49 (Supplement 1), 285-310. Munro. M- J ., & Derwing. T. M. (1995b). Processing time, accent, and comprehensibility in the perception of native and foreign-accented speech. Language and Speech. 3 8 (Part 3), 289-306. Munro, M. J., Derwing. T. M., & F lege. J. E. (1999). Canadians in Alabama: A perceptual study of dialect acquisition in adults. Journal of Phonetics, 27(4). 385- 403. Nabeleck, A.. & Donahue, A. (1984). Perception of consonants in reverberation by native and non-native listenders. Journal of the Acoustical Society of America, 75 . 632- 634. Nelson, N. W., & McRoskey, R. L. (1978). Comprehension of Standard English at varied speaking rates by children whose major dialect is Black English. Journal of Communication Disorders, 11(1), 37-50. Nicolosi, L.. Harryman, E.. & Kresheck, J. (2003). T erminologv in communication disorders (5th ed.). Baltimore, MD: Lippincott Williams & Wilkins. Niedzielski, N. (1999). The effect of social information on the perception of sociolinguistic variables. Journal ofLanguage and Social Psychology, 18(1). 62- 85. 162 Niedzielski, N.. & Preston. D. (2003). Folk Linguistics. Berlin: Mouton de Gruyter. Oetting. J. B.. & McDonald. J. L. (2001). Nonmainstream dialect use and specific language impairment. Journal of Speech. Language, Hearing Research. 44(1). 207-223. Oetting, J. B., & McDonald. J. L. (2002). Methods for characterizing participants' nomnainstream dialect use in child language research. Journal ofSpeech- Lt‘tnguage-Hearing Research, 45(3). 505-518. Payne. K.T., & Taylor. CL. (2005). Multicultural differences in human communication disorders. In N. B. Anderson & G.H. Shames (Eds). Human con-mumication disorders (7"1 edition). Boston: Pearson. Pena. E.. Iglesias. A.. & Licz, C. (2001). Reducing test bias through dynamic assessment of children's word learning ability. American Journal of Speech-Language Pathologv, 10, 138-154. Pierce. C .A.. Block. R.A. & Aguinas. H. (2004) Cautionary note on reporting eta-squared values from multifactor ANOVA designs. Educational and Psychological Measurement, 64(6), 916-924. Pisoni. D. B., & Lively, S. E. (1995). Variability and invariance in speech perception: A new look at some old problems in perceptual learning. In W. Strange (Ed.), Speech perception and linguistic experience: Issues in cross-language research (pp. 433-459). Baltimore: York Press. Plichta. B.. & Rakerd. B. (2002. October). Perceptions oft/a/jronting in two Michigan dialects. Paper presented at the NWAV Conference, Stanford, CA. Pollock. K. E., Bailey. G., Berni. M. C., Fletcher, J. M., Hinton, L.N., Johnson. 1.. & Weaver, R. (1998, March 17, 2001). Phonological features of African American Vernacular English (AA VE). Retrieved October 24, 2004, from http:x’t’wvnv.rehabmed.ualbena.ca/spa/phonology/features.htm Preston. D. (1996a). W haddayaknow: The modes of folk linguistic awareness. Language A ut'areness. 5 ( l ). 40-74. Preston. D. ( 1996b). Where the worst English is spoken. In E. Schneider (Ed), Focus on the USA (pp. 297-360). Amsterdam: John Benjamins. Preston, D. (1997). The South: The touchstone. In Bernstein. Nunnally & Sabino (Eds). Language variety in the south revisited (pp. 31 1-351). Tuscaloosa: University of Alabama. 163 Preston, D. (1999). A language attitude approach to the perception of regional variety. In D. R. Preston (ed.). Handbook ofperceptual dialectology (pp. 359-73.). Amsterdam: John Benjamins, Preston, D., & Howe, G. M. (1987). Computerized generalizations of mental dialect maps. In Keith Denning (Ed.), Variation in language: NWA V-XV at Stanford (pp. 361-378). Stanford: Department of Linguistics, Stanford University. Preston. D., & Robinson. G. (2005). Dialect Perception and Attitudes to Variation. In M. J. Ball (Ed), Clinical Sociolinguistics (pp. 133-150). Oxford: Blackwell. Prince, A. & Smolensky. P. (1993): Optimality Theory: Constraint Interaction in Generative Grt‘tmmar. Rutgers University Center for Cognitive Science Technical Report 2. Pumell, T., Idsardi. W., & Baugh. J. (1999). Perceptual and phonetic experiments on American English dialect identification. Journal of Language and Social Psychology, 18.1 (Special issue: Attitudes, Perception, and Linguistic Features). Qi. C. H., Kaiser, A. P., Milan, S. E.. Yzquierdo, Z., & Hancock, T. B. (2003). The perfomiance of low-income, African American children on the Preschool Language Scale-3. Journal of Speech-Language and Hearing Research, 46(3 ). 576-590. Rickford, J .R., & Rickford, R.J. (2000) Spoken soul: The story ofB/ack English. New York: John Wiley & Sons. Robinson. P. (1973). Testing the second-language competence of children and adults. English Language Teaching. 27, 190-199. Ryan, E., Carranza, M., & Moffte, R. (1977). Reactions toward varying degrees of accentedness in the speech of Spanish-English bilinguals. Language and Speech. 20, 267-273. Schiavetti, N. & Metz, DE. (1997) Evaluating research in communicative disorders (5m edition) Boston: Pearson. Schmid, P. M., & Yeni-Komshian, G. H. (1999). The effects of speaker accent and target predictability on perception of mispronunciations. Journal of Speech-Lang and Hearing Research, 42(1), 56-64. Schraeder, T., Quinn, M., Stockman, I.J., & Miller, J. (1999) Authentic assessment as an approach to preschool speech-language screening. American Journal ofSpeech- Language Pathology), 6, 195-200. 164 Skutnabb-Kangas, T. (2000). Linguistic genocide in education-or worldwide diversity and human rights? Mahwah. NJ: Lawrence Erlbaum Associates. Seymour. H. N., & Bland, L. (1991). A minority perspective in the diagnosis of child language disorders. Clinical Communication Disorders, 1(1). 39-50. Seymour, H. N., & Pearson, B. Z. (2004). Steps in designing and implementing an innovative assessment instrument. Seminars in Speech and Language, 25 ( 1 ), 27- 31. Seymour, H. N., & Ralabate, P. K. (1985). The acquisition of a phonologic feature of Black English. Journal ofCommunication Disorders, 18(2), 139-148. Seymour, H. N., Roeper, T.. & deVilliers, J. G. (2003). Diagnostic evaluation of language variation (DEL V) criterion-referenced San Antonio, TX: The Psychological Corporation. Seymour. H. N.. & Seymour, C. M. (1981). Black English and standard American English contrasts in consonantal development of four and five-year old children. Journal ofSpeech and Hearing Disorders, 46(3). 274-280. Sjolander. K., & Beskow. J. (2004). W avesurfer (Version 1.7.1). Stockholm, Sweden: Centre for Speech Technology. Smith. E. A. (1993). Cultural and linguistic factors in worker notification to blue collar and no-collar African-Americans. American Journal of Industrial Medicine, 23(1), 37-42. Smitherman. G. (1977). T alkin' and testitfvin': The language of Black America. Boston: Houghton-Mifflin. Smitherman. G. (1994*). Black talk: Words and phrases/rom the hood to the amen corner. Boston: Houghton-Mifflin. Stein. S. J. (1981). C Iient's perception of counselor trustworthiness, expertness. and attractiveness as a function of counselor race and dialect. Dissertation Abstracts Internatiomzl, 42(6-B), 2552-25 53. Steriade, D. (2004). The Phonology ofPerceptibility Effects: the P-map and its consequences for constraint organization. Retrieved 11-20-04, 2004, from wuvulinguisticsucla.edu/people/ steriade/papers/P-map_for_phonology.doc Stockman, I. J. (in press) Alveolar bias in the final consonant deletion patterns of African American children. Language, Speech. and Hearing Services in the Schools. Stockman, I. J. (1986). Language acquisition in culturally diverse populations: the Black child as a case study. In O. L. Taylor (Ed), Nature of communication disorders in culturally and linguistically diverse populations (pp. 1 17-155). San Diego. CA: College-Hill Press. Stockman, I. J. (1996). Phonological development and disorders in African American children. In A. G. Kamhi, K. E. Pollock & J. L. Harris (Eds), Communication development and disorders in African American children (pp. 1 17-153). Baltimore: Paul H. Brooks. Stockman, I. J. (1997a). The promises and pitfalls of language sample analysis as an assessment tool for linguistic minority children. Language . Speech. and Hearing Services in the Schools, 27, 355-366. Stockman, I. J. (1997b). The sociopolitical construction of science: Evidence from the language research of African American children. Archives ofthe W. E. B. Du Bois Distinguished Visiting Lecturer Series. New York. NY: The Graduate School and University Center of The City University of New York. Stockman. I. J. (2000). The new Peabody Picture Vocabulary Test - III: An illusion of unbiased assessment. Language , Speech. and Hearing Services in the Schools. 31. 340-353. Stockman, I. J. (2004). Clinical phonology issues in serving speakers of African- American Vernacular English. In R. D. Kent (Ed). The MIT Encyclopedia of Communication Disorders (pp. 158-160). Cambridge, MA: The MIT Press. Stockman. I. J.. Boult. J. R., & Robinson. G. (2004). Multicultural issues in academic and clinical education: A cultural mosaic. ASHA Leader (20), 6-7. Stockman, I.J., Guillory. B., Newkirk, B., & Seibert, M. (in submission) Regional variation among African American preschoolers in standardized language test scores and dialect density. Stockman, I. J ., & Schraeder, T. (1999). The concept ofa minimal competence core shows assessment promise. Asha, 41(4), 50-51. Stockman, I. J ., & Vaughn-Cooke, F. B. (1991). Child language acquisition research: A neglected area of focus in the Black African diaspora. In M. Cummings, L. Niles & 0. Taylor (Eds), Handbook on communications and development in Africa and the African diaspora (4th ed, pp. 82-129). Needham, MA: Ginn Press. Taylor. 0. L. (1999). Cultural issues and language acquisition. In 0. L. Taylor & L. B. Leonard (Eds). Language acquisition across North America: Cross-Cultural and C toss-linguistic Perspectives (pp. 21-38). San Diego: Singular Publishing Group. 166 Terrell. S. L., & Jackson, R. S. (2002). African Americans in the Americas. In D. E. Battle (Ed). C ommunicc‘ttion disorders in multicultuml populations (3rd ed, pp. 33-70.). Boston, MA: Butterworth-Heinemann. Thomas. J.E. (1998) Properties of the ANOVA F test as applied to ordered categorical data : a simulation-based study. Unpublished doctoral dissertation, Baylor University. US. Department of Special Education Programs. (2004). IDEAdata. org. Retrieved October 10, 2004, 2004, from http://vmvv.ideadata.org/index.html Van Riper. C. (1978’). Speech correction. Englewood C liffs, NJ: Prentice Hall. Voi gt. D. A. (1994). The effects of the dialect of speaker and qualifications of the speaker on rater's perceptions of a job applicant. Dissertation Abstracts International: Section B: The Sciences and Engineering. 5 4(7-B). 3886. Washington. J. A.. & Craig. H. K. (1992). Performances of low-income. African American preschool and kindergarten children on the Peabody Picture Vocabulary Tests-Revised. Language , Speech. and Hearing Services in the Sclmols. 23. 329-333. Washington, I. A.. & Craig, H. K. (1994.). Dialectal forms during discourse of poor. urban. African American preschoolers. Journal of Speech and Hearing Research. 37(4). 816-823. Washington. I. A.. & Craig, H. K. (1998). Socioeconomic status and gender influences on children's dialectal variations. Journal of Speech-Language and Hearing Research. 41(3). 618-626. Washington. I. A.. C raig, H. K., & Kushmaul, A. J. (.1998). Variable use of African American English across two language sampling context. Journal ofSpeech- Language and Hearing Research, 41(5). 1 1 15 -I 124. Westby. C. (1990). There's no such thing a culture-free testing. Texas Journal of Audiologv and Speech Pathology. 16(1). 4-5. Wilson. F. W ., Wilson, J. R., & C oleman, T. J. (2000). Culturally appropriate assessment: issues and strategies. In T. J. Coleman (Ed), Clinical management of communication disorders in children (pp. 101-127). Needham Heights. MA: Allyn and Bacon. Wolfram, W. (1986). Language variation in the United States. In Orlando L. Taylor (ed). Communication problems in culturally and linguistically diverse populations. College-Hill Press. 73-115. 167 Wolfram. W., & Schilling-Estes. N. (.1998). American English. Malden. MA: Blackwell Publishers. Wong. A. M.-Y., Leonard. L. B., Fletcher. P., & Stokes. S. F. (2004). Questions without movement: A study ofCantonese-speaking children with and without specific language impairment. Journal of Speech. Language. and Hearing Research. 47(6), 1440-1453. Wyatt. T. A. (2002). Assessing the communicative abilities ofclients from diverse cultural and language backgrounds. In D. E. Battle (Ed), Communication disorders in multicultural populations. (3rd ed. pp. 415-459). Boston, MA: Butterworth-Heineman. 168