Japanese co-occurrence restrictions influence second language perception

Abstract

Most current models of nonnative speech perception (e.g., extended perceptual assimilation model, PAM-L2, Best & Tyler, 2007; speech learning model, Flege, 1995; native language magnet model, Kuhl, 1993) base their predictions on the native/nonnative status of individual phonetic/phonological segments. This paper demonstrates that the phonotactic properties of Japanese influence the perception of natively contrasting consonants and suggests that phonotactic influence must be formally incorporated in these models. We first propose that by extending the perceptual categories outlined in PAM-L2 to incorporate sequences of sounds, we can account for the effects of differences in native and nonnative phonotactics on nonnative and cross-language segmental perception. In addition, we test predictions based on such an extension in two perceptual experiments. In Experiment 1, Japanese listeners categorized and rated vowel–consonant–vowel strings in combinations that either obeyed or violated Japanese phonotactics. The participants categorized phonotactically illegal strings to the perceptually nearest (legal) categories. In Experiment 2, participants discriminated the same strings in AXB discrimination tests. Our results show that Japanese listeners are more accurate and have faster response times when discriminating between legal strings than between legal and illegal strings. These findings expose serious shortcomings in currently accepted nonnative perception models, which offer no framework for the influence of native language phonotactics.

Japanese coda [m] elicits both perceptual assimilation and epenthesis

“…This weakening is not uniform across restrictions; some restrictions are adhered to in most recent loanwords (e.g., */si/ and */tu/) while others are adhered to infrequently (e.g., */ti/) [3]. This difference is reflected in the above study [1], whereby tokens that contain infrequently adhered to restrictions are easier to discriminate from licit tokens (e.g., /iti/ -/itʃi/) compared to tokens that contain restrictions that are adhered to in most recent loans (e.g., /isi/ -/iʃi/). Perceptual epenthesis involves illusory vowels which typically split unattested consonant clusters.…”

“…In the framework of PAM-L2, this is known as either 'single category' assimilation-when both non-native phones match equally well to the assimilated category-or 'category goodness' assimilation-when sounds at issue vary in how well they are perceived to fit. A recent extension of PAM-L2 expands this model, proposing that phonotactically conditioned repair follows a similar pattern to those outlined in PAM-L2 [1]. Kilpatrick et al propose to incorporate transitional probability into PAM-L2 to predict and account for the perceptual difficulties listeners experience when exposed to sequences of speech that are unattested (or are improbable) in their L1 [1].…”

“…A recent extension of PAM-L2 expands this model, proposing that phonotactically conditioned repair follows a similar pattern to those outlined in PAM-L2 [1]. Kilpatrick et al propose to incorporate transitional probability into PAM-L2 to predict and account for the perceptual difficulties listeners experience when exposed to sequences of speech that are unattested (or are improbable) in their L1 [1]. This model proposes that unattested sequences will sometimes be assimilated to sequences that adhere to L1 phonotactics.…”

“…Perceptual assimilation is typically studied and discussed in terms of non-overlapping phonemic and phonetic inventories. However, perceptual assimilation can also occur when listeners are exposed to L2 speech that adheres to L1 phonemic and phonetic inventories but violates L1 phonotactics [e.g., 1,2]. For example, Japanese maintains several co-occurrence restrictions which limit how consonants and vowels may combine in CV sequences (e.g., */si/, */ti/ and */tu/) and when exposed to L2 speech that violates these restrictions, Japanese listeners sometimes repair the signal by assimilating one of the sounds into an unexpected phonemic category (e.g., /si/ is confused with /ʃi/) [1].…”

“…However, perceptual assimilation can also occur when listeners are exposed to L2 speech that adheres to L1 phonemic and phonetic inventories but violates L1 phonotactics [e.g., 1,2]. For example, Japanese maintains several co-occurrence restrictions which limit how consonants and vowels may combine in CV sequences (e.g., */si/, */ti/ and */tu/) and when exposed to L2 speech that violates these restrictions, Japanese listeners sometimes repair the signal by assimilating one of the sounds into an unexpected phonemic category (e.g., /si/ is confused with /ʃi/) [1]. In AXB discrimination tests, Japanese listeners are significantly less accurate and slower to respond to trials when testing the discriminability of tokens when one token violates co-occurrence restrictions (e.g., /isi/ -/iʃi/) compared to when both tokens adhere to L1 phonotactics (e.g., /usu/ -/uʃu/).…”