Perceptual Assimilation Model Research Articles

O uso de estímulos sintéticos em testes de percepção de vogais de uma língua estrangeira

The development of models like Flege’s Speech Learning Model and Best’s Perceptual Assimilation Model directed Interphonology research to the perception of target language contrasts. These model shave triggered the development of perception tests, often using synthesized stimuli. The present paper discusses the ways previous studies have used these stimuli in vowel perception tests, and also presents the results of a pilot test developed in Brazil. While previous tests were concerned with checking the acoustic cues used in the perception of vowel contrasts, the present paper analyzes a test that can access the categorical boundaries in the perception of English front vowels by Brazilian Portuguese speakers and native English speakers. The stimuli for this test were created with the ASLsynthesizer; innovations can be observed in the test developmentand application.

Theoretical models of non-native speech perception: Implications for perception of foreign-accented speech

Linguistic experience systematically affects speech perception, specifically for non-native phonemes and contrasts. Native language influences are strong in adults, and emerge in infancy. Several theoretical models of non-native speech perception have been proposed, including the Perceptual Assimilation Model (PAM: Best), Speech Learning Model (LSM: Flege), and Native Language Magnet model (NLM: Kuhl). Findings that infants learn statistical distributions of phoneme sequences, and phonetic details, converge on the notion that exposure to specific phonetic patterns affects the perception of phonological structure. A reasonable extrapolation is that perception of even native utterances should be influenced by their pronunciation with either familiar or unfamiliar phonetic structure. That is, if a listener is inexperienced with a given foreign accent, or another native language dialect, perception should be predictably affected when native phonemes/words are produced with those accents. This issue has received modest research attention, which has focused on sentences/words rather than phonemes. Findings suggest that unfamiliar accents do reduce recognition and intelligibility of native sentences and words. The implications of non-native speech models, particularly PAM and SLM, for understanding and studying the impact of foreign accent on perception of native phonemes and contrasts, as well as longer utterances, will be discussed. [Work supported by NIH.]

Perception of syllable-initial and syllable-final nasals in English by Korean and Japanese speakers

In both Korean and English, /m/ and /n/ contrast syllable-initially and /m/, /n/ and /&rhookn;/ contrast syllable-finally. In Japanese, /m/ and /n/ contrast syllable-initially while nasals do not contrast syllable-finally. Experiment 1 of this study investigated Korean and Japanese speakers’ perception of English nasals to examine how learners’ L1 influences the perception of L2 segments. The Japanese speakers had considerable difficulty distinguishing /&rhookn;/ from /n/ syllable-finally. Final /m/-/&rhookn;/ and /m/-/n/ were not particularly problematic for the Japanese speakers, although these contrasts do not exist in their L1. Experiment 2 examined the perceived relation between English /m/, /n/ and /&rhookn;/ and Japanese categories to investigate why final /n/-/&rhookn;/ was especially difficult for the Japanese speakers. In order to examine which Japanese categories are used to represent final nasals in English, Japanese speakers were asked to write English words used in Experiment 1 with the Katakana orthography. It was found that syllable-final /m/ was assimilated to one Japanese category, (see PDF for character), whereas two or more categories were used to classify /n/ and /&rhookn;/. There was a relatively high degree of overlap in classification between /n/ and /&rhookn;/ since both /n/ and /&rhookn;/ were classified with (see PDF for character) /N/ and yH (see PDF for character). The three final contrasts, /n/-/&rhookn;/, /m/-/&rhookn;/ and /m/-/n/ were classified as one of the assimilation types of the Perceptual Assimilation Model (PAM), based on the results from Experiment 2. The syllable-final /m/-/&rhookn;/ and /m/-/n/ contrasts were classified as Uncategorized-Categorized (UC), and the syllable-final /n/-/&rhookn;/ contrast was classified as Both Uncategorizable (UU). PAM expects discrimination to be very good for UC, but poor for UU if the two non-native categories are relatively similar to each other. The results from Experiment 1 for the Japanese speakers were consistent with the PAM prediction.

Learning foreign vowels.

Two hypotheses have recently been put forward to account for listeners' ability to distinguish and learn contrasts between speech sounds in foreign languages. First, Best's Perceptual Assimilation Model and Flege's Speech Learning Model both predict that the ease with which a listener can tell one non-native phoneme from another varies directly with the extent to which these sounds assimilate to different native phonemes (Best, 1994; also Best, McRoberts, & Goodell, 2001; Flege, 1991). Second, Logan, Lively, & Pisoni (1991) have argued that training listeners to identify non-native phonemes teaches them sets of exemplars rather than more abstract distinctive feature values. I report here the results of three sets of experiments designed to test these hypotheses, in which American English listeners were trained to categorize German nonlow vowels. The first set of experiments show that some instances of the same contrast between German vowels are more easily discriminated than others, a result incompatible with the predictions of either Best's or Flege's models, but compatible with the alternative category recognition interpretation. The second set of experiments reveals effects of contextual and speaker variation on listeners' ability to learn [tense] but not [high] contrasts between foreign vowels, and are thus at least partly compatible with an exemplar model of foreign category learning (Pisoni, Lively, & Logan, 1994; also Nosofsky, 1986). The third set of experiments compares the predictions of Nosofsky's (1986) selective attention exemplar model of category learning with those of a feature learning model in tests of listeners' learning the natural classes to which the German vowels belong. The results are mixed: listeners learned the features that define the natural classes of [+/- high] and [+/- back] vowels, but could have learned either the feature that defines the natural classes of [+/- tense] vowels or sets of [+/- tense] exemplars. Natural classes defined by abstract distinctive feature values are thus learnable, even if their membership is phonetically polymorphous.

Infant perception of non-native consonant contrasts that adults assimilate in different ways.

Numerous findings suggest that non-native speech perception undergoes dramatic changes before the infant's first birthday. Yet the nature and cause of these changes remain uncertain. We evaluated the predictions of several theoretical accounts of developmental change in infants' perception of non-native consonant contrasts. Experiment 1 assessed English-learning infants' discrimination of three isiZulu distinctions that American adults had categorized and discriminated quite differently, consistent with the Perceptual Assimilation Model (PAM: Best, 1995; Best et al., 1988). All involved a distinction employing a single articulatory organ, in this case the larynx. Consistent with all theoretical accounts, 6-8 month olds discriminated all contrasts. However, 10-12 month olds performed more poorly on each, consistent with the Articulatory-Organ-matching hypothesis (AO) derived from PAM and Articulatory Phonology (Studdert-Kennedy & Goldstein, 2003), specifically that older infants should show a decline for non-native distinctions involving a single articulatory organ. However, the results may also be open to other interpretations. The converse AO hypothesis, that non-native between-organ distinctions will remain highly discriminable to older infants, was tested in Experiment 2, using a non-native Tigrinya distinction involving lips versus tongue tip. Both ages discriminated this between-organ contrast well, further supporting the AO hypothesis. Implications for theoretical accounts of infant speech perception are discussed.

Effects of perceptual assmilation on the production of English vowels by native Japanese speakers

In our previous studies, it was found that English vowel contrasts that were difficult for Japanese speakers to discriminate were for the most part identified with the same Japanese vowels. This finding complies with one the Perceptual Assimilation Model (Best, 1995) postulates. The present study investigated how perceptual assimilation affects production of English vowels by Japanese speakers. Three experienced Japanese learners in Columbus, Ohio, and five inexperienced Japanese learners in Kobe, Japan produced English vowels, repeating after two different native speakers’ utterances. Each subject heard the same token twice in a different order, so they each produced four tokens of each English vowel. The results support what our previous studies have found, and show the effects of perception on production. The production of the vowels difficult to discriminate tended to be close to each other within subjects’ production vowel space. The production of the vowels identified as good exemplars of Japanese vowels were more stable than those identified as poor exemplars. Thus, the subjects /i/ tokens were almost always identified as /i/, whereas their /æ/ tokens were often identified as /ε/, /ɑ/, or /ʌ/. Experienced learners’ utterances were far more correctly identified.

Assimilation and discrimination of Canadian French vowels by English-speaking adults

According to the Perceptual Assimilation Model (PAM) discriminability of non-native contrasts depends on the perceived similarity to native phonetic categories [Best (1994)]. With respect to vowel contrasts, it has been claimed that PAM predictions hold so long as context-specific realizations of a given contrast are considered [Strange et al. (2001); Levy (2002)]. To further assess these claims, English adults were tested on perception of Canadian French tense high vowels, /i/, /y/, /u/ and their respective context-conditioned variants [I], [Y], and [U]. Adults completed 2 tasks with both citation and sentence context stimuli; stimuli for both tasks were natural /bVs/ tokens, produced by male and female talkers. For each vowel, subjects completed an identification and rating task (using English vowel response categories) which provided data to assess the assimilation pattern for four non-native contrasts: /i-y/, /y-u/, /I-Y/, and /Y-U/. Discrimination was assessed for each contrast using a categorical AXB task, in which each token is produced by a different talker. The ability to predict discrimination differences from assimilation data was assessed. The findings were examined together with acoustic analysis to determine whether acoustic differences and assimilation differences are equally predictive of relative discriminability.

Perception of English vowels by native Korean and Japanese speakers

Native speakers of Korean and Japanese were assessed in their ability to perceive English vowels categorically. These two languages were chosen because they differ in the size of their vowel inventories with Korean containing more vowels than Japanese. The Korean vowel space is presumably partitioned more so than the Japanese and this may facilitate Korean speakers’ discrimination of English. This hypothesis is in keeping with Best’s Perceptual Assimilation Model (PAM). PAM states that non-native speakers’ discrimination of non-native sounds can be predicted by differing degrees based on whether the new non-native sounds are assimilated to one or more native phonemic categories. Further, discrimination is mediated by how well the non-native category fits the native category. Both subject groups completed the following tasks: the Categorical Discrimination Test (CDT, designed by J. Flege) which tested their discrimination of English vowels; and an assimilation task where subjects equated English vowels to their native vowels and rated each for category goodness. A’ scores were calculated for each contrast in the CDT. In addition, confusion matrices were calculated for each group. For both subject groups, the English vowels assimilated to a single vowel category were found to be more difficult to discriminate.

On the relationship between identification and discrimination of non-native nasal consonants.

To examine the relationship between the identification and discrimination of non-native sounds, nasal consonants varying in place of articulation from Malayalam, Marathi, and Oriya were presented in two experiments to seven listener groups varying in their native nasal consonant inventory: Malayalam, Marathi, Punjabi, Tamil, Oriya, Bengali, and American English. The experiments consisted of a categorial AXB discrimination test and a forced-choice identification test with category goodness ratings. The identification test results were used to classify the non-native contrasts as one of five "assimilation types" of the Perceptual Assimilation Model (PAM) that are predicted to vary in their relative discriminability: two-category (TC), uncategorizable-categorizable (UC), both uncategorizable (UU), category-goodness (CG), and single-category (SC). The results showed that the mean percent correct discrimination scores of the assimilation types, but not the range of scores, were accurately predicted. Furthermore, differences in category goodness ratings in the CG and SC assimilations that were predicted to correlate with discrimination showed a weak, but significant correlation (r= 0.3 1, p<0.05). The implications of the results for models of cross-language speech perception were discussed, and an alternative model of cross-language speech perception was outlined, in which the discriminability of non-native contrasts is a function of the similarity of non-native sounds to each other in a multidimensional, phonologized perceptual space.

Discrimination of non-native consonant contrasts varying in perceptual assimilation to the listener's native phonological system.

Classic non-native speech perception findings suggested that adults have difficulty discriminating segmental distinctions that are not employed contrastively in their own language. However, recent reports indicate a gradient of performance across non-native contrasts, ranging from near-chance to near-ceiling. Current theoretical models argue that such variations reflect systematic effects of experience with phonetic properties of native speech. The present research addressed predictions from Best's perceptual assimilation model (PAM), which incorporates both contrastive phonological and noncontrastive phonetic influences from the native language in its predictions about discrimination levels for diverse types of non-native contrasts. We evaluated the PAM hypotheses that discrimination of a non-native contrast should be near-ceiling if perceived as phonologically equivalent to a native contrast, lower though still quite good if perceived as a phonetic distinction between good versus poor exemplars of a single native consonant, and much lower if both non-native segments are phonetically equivalent in goodness of fit to a single native consonant. Two experiments assessed native English speakers' perception of Zulu and Tigrinya contrasts expected to fit those criteria. Findings supported the PAM predictions, and provided evidence for some perceptual differentiation of phonological, phonetic, and nonlinguistic information in perception of non-native speech. Theoretical implications for non-native speech perception are discussed, and suggestions are made for further research.

Cross-linguistic perceptual assimilation of American English /r/–/l/ by Korean speakers and Japanese speakers

According to the Perceptual Assimilation Model (PAM) [Best et al., JEP:HPP 14, 345–360 (1988)], the difficulties that second-language learners have in distinguishing non-native phoneme contrasts is predictable from the way learners assimilate the non-native sounds into their native phonological system. The present study examined Korean and Japanese adults’ perceptual assimilation of American English (AE) /r/–/l/ contrasts in first two experiments. In experiment 1 (assimilation task), subjects listened to nonsense AE words contrasting /r/ and /l/ in five syllabic positions, and chose one of the native orthographies displayed visually. In experiment 2 (rating task), they listened to the same nonsense words, and rated the ‘‘goodness-of-fit’’ of each stimulus to the presented native orthography on the scale from 1 (least fit) to 7 (best fit). Then, in experiment 3 (identification task), the ability of Korean and Japanese to identify both real AE words and nonsense words was measured, and a significant effect of position was found in both language groups. When the assimilation and rating data from experiments 1 and 2 were interpreted in terms of the PAM categories, the predicted levels of identification accuracy were in good agreement with the identification results in experiment 3.

Effect of linguistic background on accuracy of discrimination task and reaction time

This study examines how American English (AE) speakers categorize Hindi laryngeal and place of articulation contrasts. These results are interpreted with respect to the Perceptual Assimilation Model (PAM) [Best, ‘‘A direct realist view of cross-language speech perception,’’ in Speech Perception and Linguistic Experience: Issues in Cross Language Research (York Press, Inc., 1995), pp. 171–204]. Accuracy and reaction time data were obtained for discrimination and identification of Hindi bilabial and dental phonation contrasts and the retroflex-dental place contrast. Subjects included both AE speakers and a control Hindi speaker group. Discrimination accuracy averaged across the subjects for the phonation contrasts was nativelike while accuracy for the retroflex-dental contrast was significantly lower but above chance. Overall, the reaction time was inversely proportional to discrimination accuracy. For some subjects discrimination performance for the retroflex-dental contrast was nativelike. However, the reaction time for those subjects was significantly higher than that of other speakers. In general, the reaction time was higher for the retroflex-dental contrast than for the other contrasts.

Native-language phonetic and phonological constraints on perception of non-native speech contrasts

Adults discriminate many non-native speech contrasts poorly, especially from an unfamiliar language that is not (yet) an L2. The Perceptual Assimilation Model (PAM) [Best et al., JEP: HPP 14, 34 560 (1988)] posits that this difficulty stems from knowledge of both the phonological functions and the phonetic details of native speech segments. Thus, discrimination of an unfamiliar contrast depends not only on whether it resembles a native phonological contrast, but also on perceived goodness of fit between the non-native segments and native phonetic category(s). Cross-language comparisons support PAMs prediction that perceptual assimilation of contrasting non-native segments to a single native category yields poorer discrimination than assimilation to two categories (TC contrasts). Discrimination is worst if both phones are equally similar to a single category (SC), substantially better if they differ in category goodness (CG), near ceiling for TC assimilations, and good to excellent for consonants that fail to be assimilated as speech, instead being perceived as nonspeech events (non-assimilable: NA). Recent findings indicate that, as predicted, SC, CG, and TC assimilations are associated with preferential activation of left hemisphere language regions, whereas NA stimuli yields bilateral brain activation. Other findings with fluent bilinguals indicate a persisting L1 effect on non-native consonant discrimination even if their L2 is acquired prior to 5 years. [Work supported by NICHHD and NIDCD.]

Perceptual assimilation of Hindi dental and retroflex stop-consonants by native English and Japanese speakers

Previous research has shown that English speakers have great difficulty distinguishing the dental and retroflex stop-consonants of the Hindi language. However, native Japanese speakers have somewhat less difficulty perceiving this contrast even though it is not employed in the Japanese language. Best and colleagues have attributed similar differences in the perception of non-native speech sounds to differences in the assimilation of such sounds to native-language speech categories [Best, McRoberts, and Sithole, J. Exp. Psych: Human Percept. Perform. 14 (1988)]. According to Best’s Perceptual Assimilation Model (PAM), four patterns of assimilation have been described which are purportedly predictive of perceptual difficulty. To determine if this model could be applied to the present case, native speakers of English and Japanese transcribed multiple instances of the Hindi consonants in different voicing/manner classes, produced with different vowels. Marked differences were found between the responses of the two language groups which were partially dependent upon the voicing/manner class of the contrast. Interestingly, an assimilation pattern emerged which was different from those discussed in PAM. Implications of these findings for PAM and for perceptual training of non-native speech sounds will be discussed. [Work supported in part by NIDCD.]

The Development of Speech Perception: The Transition from Speech Sounds to Spoken Words

Introduction: developing theories of speech perception - constraints from developmental data, Judith C. Goodman et al. Part 1 Innate sensory mechanisms and constraints on learning: observations on speech perception, its development, and the search for a mechanism, Joanne L. Miller and Peter D. Eimas the importance of childhood to language acquisition - evidence from American sign language, Rachel I. Mayberry. Part 2 Perceptual learning of phonological systems: cross-language speech perception - developmental change does not involve loss, Janet F. Werker perceptual learning of non-native speech contrasts - implications for theories of speech perception, David B. Pisoni et al the emergence of native-language phonological influences in infants - a perceptual assimilation model, Catherine T. Best. Part 3 Interactions of linguistic levels - influences on perceptual development: infant speech perception and the development of the mental lexicon, Peter W. Jusczyk sentential processes in early child language - evidence from the perception and production of function morphemes, LouAnn Gerken learning to hear speech as spoken language, Howard C. Nusbaum and Judith C. Goodman.

The identification and discrimination of a non-native nasal series by Bengali and American English listeners

Predictions based on the perceptual assimilation model (PAM) (Best, 1995) concerning the perception of non-native contrasts rely primarily on three factors: the contrast in question, the inventory of native categories, and the degree of gestural similarity between them. To determine if these factors are sufficient for predicting listeners’ difficulties, nasal contrasts from three languages, Malayalam, Marathi, and Oriya, varying in place of articulation (bilabial, interdental/dental, alveolar, and retroflex), syllabic position, and vowel context, were presented in an AXB discrimination test and an identification test with category-goodness ratings to sixteen speakers each of American English (AE) and Bengali. AE and Bengali were selected for their similar nasal series ([m], [n], and [■]). However, Bengali differs from AE in its use of dental and retroflex oral stops. PAM should predict equivalent performance for both subject groups. However, a pilot of this study demonstrated poorer performance by Bengali listeners on dental-retroflex (34.6% mean correct to AE 53.6%) and interdental-alveolar (33.2% to AE 69.6%). The implications of these results for phonological and phonetic factors in the perception of non-native contrasts will be discussed.

Non-native speech perception as a window on the native phonological system and its development

The perceptual assimilation model (PAM) [Best etal ., JEP: HPP 14, 345–360 (1988)] posits that perception of non-native contrasts is constrained by experience with both the phonological functions and the phonetic details of familiar native contrasts. Adults’ discrimination of an unfamiliar contrast depends not only on whether its members resemble a known phonological contrast, but also on perceived ‘‘goodness of fit’’ to native phonetic categories. Adult findings support PAM’s prediction that perceptual assimilation of contrasting non-native consonants or vowels to a single native category yields poorer discrimination than assimilation to two categories (TC contrasts). Discrimination is worst if both phones are equally similar to a single category (SC), substantially better if they differ in category goodness (CG), near ceiling for TC assimilations. PAM predictions have focused on adults’ first encounters with unfamiliar contrasts. However, non-native speech perception also provides a window on infants’ emerging knowledge about phonetic details and phonological organization in native speech. Native phonetic experience influences discrimination of unfamiliar non-native contrasts by 10 months, but systematic phonological knowledge is not evident until later. PAM can also be extended to phonetic and phonological aspects of second language (L2) learning developmentally. [Work supported by NICHHD and NIDCD.]

Assimilation of non-native vowel contrasts to the American English vowel system.

The perceptual assimilation model (PAM) [Best et al., JEP:HPP 14, 345–360 (1988)] predicts that two non-native sounds that are assimilated to the same nature category will be harder for listeners to discriminate between than sounds that are assimilated to two different native categories (TC contrasts); how difficult will depend on whether they are equally good (or bad) exemplars of the single native category (SC contrasts) or not (CG contrasts). PAM was based on studies of consonant perception, as were subsequent tests of the model. This study extends the model to non-native vowels. American listeners performed keyword identification [W. Strange and T. L. Gottfried, J. Acoust. Soc. Am. 68, 1622–1625 (1979)] and categorical AXB discrimination tasks using six non-native vowel contrasts, Norwegian /i–y/, /i–■/, French /o–õ/, /œ–y/, /œ–■/, and Thai /■–■/. Assimilation patterns for a particular vowel contrast, inferred from keyword results, were more variable than in consonant studies but nonetheless strongly related to discrimination performance: TC contrasts were better discriminated than CG contrasts, which in turn were better discriminated than SC contrasts. Moreover, listeners who assimilated a particular contrast in TC fashion were better able to discriminate it than listeners who assimilated it in CG or SC fashion. [Work supported by NIH.]

Effects of bilingualism on non-native phonetic contrasts.

Monolingual listeners often have difficulties discriminating non-native contrasts. The Perceptual Assimilation Model (PAM) [C. T. Best, G. W. McRoberts, and N. M. Sithole, JEP:HPP 14, 345–360 (1988)] predicts poor discrimination if two non-native sounds assimilated equally well to a single native category (SC), good discrimination if a category goodness (CG) difference is perceived, and excellent discrimination if sounds are assimilated to two different categories (TC). However, implications for bilingual listeners are uncertain. Do bilinguals maintain two separate phonological systems, or do the systems interact to enhance or inhibit non-native discrimination? In this study, monolingual English and bilingual Spanish/English listeners discriminated and categorized Spanish [b–p], English [b–ph], and three Xhosa bilabial stop contrasts. Both groups discriminated English [b–ph] excellently, as did bilinguals for Spanish [b–p], which monolinguals discriminated less well, consistent with CG assimilation. Compatible with PAM, monolinguals showed excellent TC performance on Xhosa unaspirated/aspirated stops; bilinguals’ lower discrimination indicated CG assimilation. On Xhosa implosive/pre-nasalized stops, monolinguals showed CG assimilation and greatly outperformed bilinguals, who showed an SC pattern. Conversely, on Xhosa implosive/plosive stops monolinguals showed SC assimilation; bilingual discrimination was higher, showing GC assimilation. [Work supported by NIH.]

Individual differences and the acquisition of new phonetic categories

Adults’ acquisition of non-native phonetic categories occurs within the context of an individual’s existing phonology. The present work examines (1) the process of acquiring a new phonetic category, (2) the impact of the new phonetic category on nearby, previously existing categories, and (3) transfer of learning to novel contexts. Monolingual American English speakers were required to learn the Hindi voiced, unaspirated, dental stop consonant. First, listeners were asked to identify and judge the goodness of stimuli on two synthetic continua spanning a range from Hindi dental to American English alveolar stop consonants. One continuum was voiced, the other voiceless. In addition, listeners judged the similarities between all possible pairs of stimuli on each continuum. After this perceptual mapping procedure, subjects participated in a two-alternative, forced-choice training program using only voiced stimuli. Progress was monitored throughout training. Following training, the mapping procedure was repeated with both the voiced and voiceless continua. Results are discussed from the theoretical perspective of the nonlinear dynamical approach to learning and transfer in motor behavior presented by Zanone and Kelso [in Swinnen, Heuer, Massion, &amp; Casaer, 461–490 (1994)] and with respect to Best’s perceptual assimilation model and Kuhl’s perceptual magnet effect. [Work supported by NIDCD and NIMH.]