Perceptual epenthesis is the perception of illusory vowels in consonantal sequences that violate native phonotactics. The consensus has been that each language has a single, predictable candidate for perceptual epenthesis, that vowel which is most minimal (i.e., shortest and/or quietest). However, recent studies have shown that alternate epenthetic vowels can be perceived when the perceptual epenthesis of the minimal vowel would violate native co-occurrence restrictions. We propose a potential explanation for these observed patterns: speech perception, and thus also vowel perceptual epenthesis, is modulated by transitional probability whereby epenthetic vowels must conform to the language specific expectations of the listener. To test this explanation, we present two experiments examining perceptual epenthesis of two Japanese vowels—/u/ and /i/—against their transitional probability in CV sequences. In Experiment 1, Japanese listeners assigned VCCV tokens to VCuCV and VCiCV categories. In Experiment 2, participants discriminated VCCV tokens from VCuCV and VCiCV tokens. The results show that sequences where /i/ is transitionally probable are more likely to elicit /i/ perceptual epenthesis.

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.

This study constructs machine learning algorithms that are trained to classify samples using sound symbolism, and then it reports on an experiment designed to measure their understanding against human participants. Random forests are trained using the names of Pokémon, which are fictional video game characters, and their evolutionary status. Pokémon undergo evolution when certain in-game conditions are met. Evolution changes the appearance, abilities, and names of Pokémon. In the first experiment, we train three random forests using the sounds that make up the names of Japanese, Chinese, and Korean Pokémon to classify Pokémon into pre-evolution and post-evolution categories. We then train a fourth random forest using the results of an elicitation experiment whereby Japanese participants named previously unseen Pokémon. In Experiment 2, we reproduce those random forests with name length as a feature and compare the performance of the random forests against humans in a classification experiment whereby Japanese participants classified the names elicited in Experiment 1 into pre-and post-evolution categories. Experiment 2 reveals an issue pertaining to overfitting in Experiment 1 which we resolve using a novel cross-validation method. The results show that the random forests are efficient learners of systematic sound-meaning correspondence patterns and can classify samples with greater accuracy than the human participants.

IntroductionThis paper presents a cross-linguistic study of sound symbolism, analysing a six-language corpus of all Pokémon names available as of January 2022. It tests the effects of labial consonants and voiced plosives on a Pokémon attribute known as friendship. Friendship is a mechanic in the core series of Pokémon video games that arguably reflects how friendly each Pokémon is.MethodPoisson regression is used to examine the relationship between the friendship mechanic and the number of times /p/, /b/, /d/, /m/, /g/, and /w/ occur in the names of English, Japanese, Korean, Chinese, German, and French Pokémon.ResultsBilabial plosives, /p/ and /b/, typically represent high friendship values in Pokémon names while /m/, /d/, and /g/ typically represent low friendship values. No association is found for /w/ in any language.DiscussionMany of the previously known cases of cross-linguistic sound symbolic patterns can be explained by the relationship between how sounds in words are articulated and the physical qualities of the referents. This study, however, builds upon the underexplored relationship between sound symbolism and abstract qualities.

When listeners are exposed to non-native speech, they sometimes perceptually repair the incoming signal to better adhere to the phonotactics of their native language. The present study examines two repair strategies-perceptual assimilation and epenthesis-in two experiments designed to test how Japanese listeners perceive non-homorganic coda [m]. In the first experiment, Japanese listeners categorized tokens that contain a medial coda [m] or a word-final coda [m] into categories represented by the Japanese Hiragana orthography with /mu/ or /ɴ/ in the target position. In Experiment 2, participants discriminated between tokens with coda [m] and tokens with [mu] or [n] in the target position in a series of AXB discrimination tests. The results show that Japanese listeners employ both perceptual epenthesis and perceptual assimilation when exposed to coda [m] sequences but favor assimilation over epenthesis, particularly in the word-final position.