PRO_ASA_06_rev1 - Perceptual assimilation in context and...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Perceptual assimilation in context and isolation James D. Harnsberger, Ph.D (Department of Communication Sciences and Disorders, University of Florida) Ratree Wayland, Ph.D. (Program in Linguistics, University of Florida) ABSTRACT Jenna Silver (Program in Linguistics, University of Florida) E-mail: [email protected] 3aSC5 INTRODUCTION (continued) 0.63 Results Stimulus Attempts to test some of these models’ predictions have had limited success: • Lotto, Kluender, and Holt, 1998; Guenther, 2000; and Lotto, 2000 found that the perceptual magnet effect (NLM) may simply be the byproduct of a phonetic context effect in the discrimination, but not identification, tests. • Harnsberger (1999, 2001a) and Wayland, (submitted) found very limited support for PAM’s predictions concerning the link between perceptual category structure and discrimination in tests involving a wide range of non-native listener groups and stimulus materials. These findings suggest that methodological issues (i.e., the design of cross-language speech perception tests, the variety of speech materials used) must be very carefully considered in testing any proposed model. Talker H1 Present Study H6 Examine perceptual assimilation of non-native speech sounds in isolation and in sentence frames H1 Question: To what extent does the identification and discrimination of non-native speech sounds vary across context? H6 MAIN EXPERIMENT – Discrimination and Identification A. Stimulus Materials H1 Two VOT contrasts, [tSdZ], [kg], consisting of four Hindi consonants [tS dZ k g] in one vowel context [a] produced by two talkers (H1 - female, H6 – male). Stimuli were produced in isolation and in a sentence frame at three speaking rates (slow, normal, fast). H6 H1 B. Subjects Isolation: 20 native speakers of American English, 18 – 25 years of age with no reported hearing problems or fluency with Hindi or languages with similar consonant contrasts. H6 Sentence context at three rates: 30 native speakers of American English. C. Procedures Categorial AXB discrimination, ISI = 1 s Forced-choice orthographic classification test Results Hindi voiced consonants consistently identified with AE voiced, while Hindi voiceless consonants variably identified with both AE consonants (e.g., Hindi [k] identified as both AE [k] and [g]) Discrimination: Talker, Context were significant (Isolation > Sentence Context) Discrimination, Speaking Rate Analysis: Contrast, Talker*Rate were significant (H6 Slow > H6 Fast) Context Isolation Fast Normal Slow Isolation Fast Normal Slow Isolation Fast Normal Slow Isolation Fast Normal Slow Isolation Fast Normal Slow Isolation Fast Normal Slow Isolation Fast Normal Slow Isolation Fast Normal Slow Modal 0.72 PR 51% 64% 71% 62% 53% 68% 84% 57% 96% 88% 89% 86% 94% 94% 95% 95% 53% 54% 72% 67% 63% 77% 76% 64% 99% 93% 92% 90% 99% 94% 98% 97% 1. Hindi voiceless consonants were inconsistently mapped onto their AE voiceless counterparts when appearing in isolation 0.70 H6 k-g 0.60 H6 tS-dZ H1 k-g H1 tS-dZ 0.66 0.66 Normal 0.66 0.63 Notes 2. Hindi voiceless consonants were inconsistently mapped onto their AE voiced counterparts when appearing in sentence context 0.85 0.84 Isolation 0.82 0.73 0 0.2 0.4 0.6 0.8 1 Proportion Correct Figure 2. Mean proportion correct scores for each stimulus pair in four contexts (isolation, fast rate, slow rate, normal rate). Notes 3. Hindi voiced consonants were consistently identified with their voiced AE counterparts. Figure 1. Modal responses (with corresponding proportions) for each stimulus in four contexts (isolation, fast rate, normal rate, slow rate). Notes 0.69 Slow Significant difference, p < 0.01 Examples of methodological variables: Discrimination test type (AXB, AX, Oddity, 4IAX) Stimulus variation in indexical properties -Speaking rate -Speaking style -Talker Use of noise Stimuli in isolation vs. larger prosodic unit Methods Later studies used similar methods and found a much larger range in performance when examining a variety of non-native speech sounds and listener groups. Theoretical models have been proposed to describe and/or predict the discrimination and identification (or “perceptual assimilation”) of non-native contrasts: • Speech Learning Model (SLM) • Native Language Magnet Model (NLM) • Perceptual Assimilation Model (PAM) 0.75 0.65 Sentence stimuli produced at three speaking rates Early cross-language experiments focused on a small set of speech sounds that were widely thought to be difficult for certain language groups. Performance on these tasks was typically poor, and discrimination and identification corresponded. 0.63 Fast MAIN EXPERIMENT (continued) INTRODUCTION Cross-language speech (phonetic) perception concerns the perception of speech stimuli that are unfamiliar to the listener because they occur outside of his/her ambient language environment. AXB Discrimination Cont ext A study was conducted to examine the effects of sentence context and speaking rate in the perceptual assimilation of non-native speech sounds. Native speakers of American English were presented with two voicing contrasts, [k]-[g] and [tS]-[dZ], produced by two Hindi speakers. They appeared word-initially before [a] and within a short sentence frame. These sentences were produced at three speaking rates (slow, normal, and fast). Listeners were administered a categorial AXB discrimination test and a forced-choice identification test. The results were compared to those collected in a previous study that employed the same contrasts appearing in isolated words and produced by the same talkers. The results showed that the discrimination of these speech sounds in a sentence frame was significantly poorer relative to the isolation context. Moreover, the modal native consonant used to classify the Hindi stops varied depending upon the sentence context variable: in isolation, more discriminable uncategorizable assimilation types were elicited, while in a sentence frame, Hindi contrasts tended to be classified using a single English (voiced) consonant. Speaking rate only showed modest effects on proportion of responses represented by the modal native consonants. These results highlight the sensitivity of perceptual patterns elicited in the laboratory to experimental variables. MAIN EXPERIMENT (continued) GENERAL DISCUSSION AND CONCLUSIONS A sentence context, regardless of rate, made the Hindi contrasts more challenging for AE listeners by influencing the identification pattern. -May be a byproduct of the sentence context serving as an additional load on short-term memory. -Attention directed away from fine-grained phonetic detail Within a sentence frame, the Hindi voicing contrasts were perceived more categorically, in the sense that they were more likely to be assimilated to a single AE category. Speaking rate itself had few significant effects on discrimination, and no systematic effect on modal responses in identification. Future work: 1. Use of distracter tasks 2. Manipulate short-term memory load more systematically 3. Correlate optimized tests with L2 training outcomes ACKNOWLEDGMENTS This work was supported by NIH-NIDCD Grant R03-DC5735-2. ...
View Full Document

{[ snackBarMessage ]}

Ask a homework question - tutors are online