A phonetic model of non-native spoken word processing

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 1480–1490 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 1480 A phonetic model of non-native spoken word processing Yevgen Matusevych School of Informatics University of Edinburgh yevgen.matusevych@ed.ac.uk Herman Kamper E&E Engineering Stellenbosch University kamperh@sun.ac.za Thomas Schatz Department of Linguistics & UMIACS University of Maryland tschatz@umd.edu Naomi H. Feldman Department of Linguistics & UMIACS University of Maryland nhf@umd.edu Sharon Goldwater School of Informatics University of Edinburgh sgwater@inf.ed.ac.uk Abstract Non-native speakers show difficulties with spoken word processing. Many studies at- tribute these difficulties to imprecise phonolog- ical encoding of words in the lexical memory. We test an alternative hypothesis: that some of these difficulties can arise from the non-native speakers’ phonetic perception. We train a com- putational model of phonetic learning, which has no access to phonology, on either one or two languages. We first show that the model exhibits predictable behaviors on phone-level and word-level discrimination tasks. We then test the model on a spoken word processing task, showing that phonology may not be nec- essary to explain some of the word processing effects observed in non-native speakers. We run an additional analysis of the model’s lexi- cal representation space, showing that the two training languages are not fully separated in that space, similarly to the languages of a bilin- gual human speaker. 1 Introduction Compared to native speakers, non-native speakers perform differently in a variety of tasks related to auditory language processing, both at the phone and at the word level. At the phone level, these tasks usually require speakers to compare individ- ual phones (e.g., phone discrimination or identifica- tion), while spoken word processing tasks usually test the implicit activation of a certain word in the memory (e.g., lexical priming, word translation). In some cases, non-native speakers’ behavior is consistent across the tasks: lower performance in spoken word processing tasks is directly associated with difficult phone contrasts. For example, upon hearing a word rock, Japanese speakers activate both rock and lock in their lexical memory (Cutler and Otake, 2004), probably because they find it difficult to discriminate the English [ô]–[l] phone contrast (Miyawaki et al., 1975). In other cases, however, non-native speakers’ behavior in spoken word processing tasks cannot be explained by dif- ficult phone contrasts (Cook et al., 2016; Amen- gual, 2016; Darcy et al., 2012). For example, in a translation task native English speakers may con- fuse Russian words moloko [m@ë2"ko] (‘milk’) and molotok [m@ë2"tok] (‘hammer’), even though this pair of words does not have a difficult phone con- trast (Cook et al., 2016). This dissociation between the behavior in phone discrimination vs. spoken word processing tasks has been attributed to different kinds of representa- tions involved. On the one hand, thanks to phonetic knowledge, speakers recognize individual phones in a given language. On the other hand, speak- ers store phonological representations of the words they know in their mental lexicon and use those rep- resentations to recognize spoken …

Content truncated to 500 words