2018 Friday Poster 6485

Friday, November 2, 2018 | Poster Session I, Metcalf Small | 3pm

Modeling developmental changes in infants’ discrimination of English vowels
M. Sundara, C. Mayer

Infants initially discriminate native and non-native contrasts. With perceptual reorganization, discrimination of non-native contrasts reduces while the discrimination of native contrasts improves. Researchers have proposed distributional learning as a domain-general mechanism by which infants’ acquire phonetic categories (e.g., Saffran et al, 1999; Maye et al., 2002). Recent proposals, however, argue for an interactive mechanism where learning words concurrently supplements distributional learning of phonetic categories (e.g., Swingley, 2009; Feldman et al., 2013a). Not only is this interactive mechanism available in the first year of life, its computational implementations outperform ones based on distributional learning alone (Feldman et al., 2013a; Feldman et al., 2013b). We investigated whether distributional learning models or interactive models best account for developmental changes in infant discrimination.

First, we replicated and extended Feldman et al.’s Bayesian distributional and interactive model. Specifically, we enriched their distributional model with additional acoustic cues known to distinguish vowel categories in English, also obtained from the Hillenbrand Corpus. These include duration, dynamic spectral cues, as well as pitch and higher formants. Like in Feldman et al., lexical frequency was obtained from CHILDES. As expected, higher dimension distributional models performed better than low dimension ones; nevertheless, interactive models consistently outperformed their distributional counterparts (Table 1).

Next, we evaluated predictions of the interactive and distributional learning models against infant discrimination data. Data from 4-, and 8-month-olds (n~280), learning (a) only English or

(b) Spanish and English, tested on English (i) /e-ɛ/ (previously published in Sundara & Scutellaro, 2011), (ii) /i-ɪ/ and (iii) /e-ɪ/, were compared. Stimuli were from 8 different female talkers from the Hillenbrand corpus; infants were tested using a visual fixation procedure with a habituation criterion of 50%. Listening times were analyzed using Linear Mixed Effects models. Infants discriminated all vowel pairs, except /e-ɪ/ at 8-months (Figure 1). Discrimination results from 4-month-olds were best correlated with predictions from the high dimension distributional model (model 4), with /e-ɪ/ being least confusable, followed by /i-ɪ/, with /e-ɛ/ being the most confusable pair. Discrimination results from 8-month-olds were also best correlated with predictions from a distributional model, albeit a low dimensional one (model 1). For 8-month- olds, /e-ɪ/ was the most confusable vowel pair, followed by /i-ɪ/, with /e-ɛ/ being the least confusable. In contrast, not only did interactive models outperform infants at both ages, discriminability scores from the interactive models did not correlate with infant performance at either age. Thus, infants’ discrimination data was consistent with the low performing distributional models not the high performing interactive models.

We are now testing discrimination of English /e-ɪ/ by monolingual English and bilingual Spanish and English-learning infants in older infants to model discrimination performance as a function of changing vocabulary size. We aim to demonstrate that infants’ attentional shift towards words is gradual, and results in decrements in discrimination captured by a u-shaped curve.

References

Feldman, N.H., Griffiths, T.L., Goldwater, S., Morgan, J.L. (2013a). A role for the developing lexicon in phonetic category acquisition. Psych. Review, 120(4), 751-778.

Feldman, N. H., Myers, E. B., White, K. S., Griffiths, T. L., & Morgan, J. L. (2013b). “Word- level information influences phonetic learning in adults and infants.” Cognition, 127(3), 427-438.

Maye, J. Werker, J., Gerken, L. (2002). Infant sensitivity to distributional information can affect phonetic discrimination. Cognition, 82, B101-B111.

Saffran, J.R., Johnson, E.K., Aslin, R.N., & Newport, E.L. (1999). Statistical learning of tone sequences by human infants and adults. Cognition, 70(1), 27-52.

Sundara, M. & Scutellaro, A. (2011). Rhythmic distance between languages affects the development of speech perception in bilingual infants. Journal of Phonetics, 39(4), 505-513.

Swingley, D. (2009). Contributions of infant word learning to language development. Phil. Trans. Royal Soc. B, 364, 3617-3622