Predicting Lexical Pitch-Accent Change From Properties of Verbal Working Memory: Evidence From a Family Record of the Keihan Dialect of Japanese

Research using linguistic databases has hypothesized that lexical form-meaning mappings are shaped by cognitive pressures. However, diachronic evidence for such constraints remains relatively limited. A useful window for diachronic investigation is provided by the Japanese lexical pitch accent because it is part of the spoken form but is not encoded in the orthography, which may make its change sensitive to cognitive pressures. The purpose of this study was to test whether intergenerational change of lexical pitch-accent patterns in the Keihan dialect exhibits item-level selectivity predicted from properties of verbal working memory-accentual forms should be less stable when they are atypical in the linguistic environment and when their associated meanings are less imageable. I analyzed 7,321 multimoraic nouns from a large-scale Keihan dialect database in which accent types were recorded for four members of a single family across successive generations. Bayesian discrete-time survival models predicted the accent change for each lexical item while controlling multiple psycholinguistic variables. The major findings were that as predicted, accent patterns were more likely to change when they were atypical and when meanings were less imageable and that imageability provided the strongest protection for atypical accent patterns. In conclusion, lexical prosodic change is not random but is selectively biased by item-level properties of form and meaning, consistent with cognitive constraints on maintaining short-term prosodic representations in verbal working memory. Notably, the large lexical sample provides a strong basis for generalization at the item level, but generalization across speakers and languages requires replication using additional records.