[PS-1.1] Division of labour between semantics and modality-specific systems in Japanese pitch accent processing: Convergent evidence from experimental psychology, developmental psychology, and computational modelling

Ueno, T. 1 , Tanida, Y. 2 , Saito, S. 2, 1 & Lambon Ralph, M. 1

1 NARU, School of Psychological Sciences, University of Manchester
2 Department of Cognitive Psychology in Education, Kyoto Univerisity

The division-of-labour computational models assume that language/cognitive processing is an emergent property of interactivity between semantics and ?surface?-level representations (Patterson & Lambon Ralph, 1999; Plaut, et al. 1996; Seidenberg, 1997; Ueno et al, 2011). These models potentially offer a powerful, overarching framework that could apply to various kinds of language/cognitive activities. If so, then they should also generalize beyond English to other languages, including to language-specific features that does not exist in English. A Japanese corpus analysis revealed that both typical and atypical accent forms co-exist in Japanese (forming a language-specific quasi-regular domain). We predicted that there should be a clearer semantic impact on the processing of words with an atypical pitch accent. In turn, if word meaning is intrinsically weak (e.g., low word imageability), then more typical pitch accent errors should emerge (a form of regularization). Consistent with these predictions, in two different language tasks (single-word repetition/reading), participants exhibited (1) reduced pitch accent accuracy and/or slower reaction time for words with intrinsically-weaker semantic representations, and (2) frequent accent errors reflecting the more typical pattern found in Japanese. We explored these finding in the context of a neuroanatomically-constrained computational model of spoken language processing (Ueno et al., 2011) This model includes both a dorsal pathway (the sound-motor mapping) and a ventral pathway (realization of sound-meaning-motor mappings). This model was trained to repeat comprehend, and name 1710 Japanese tri-mora words. The model recaptured the accent typicality effect on nonword repetition accuracy in both adults and children, and the impact of semantics (stronger input for the semantic system) on repetition times for words with atypical accent patterns.