Sound and meaning: On the duration of Japanese homophones

Different words can have the same sounds and yet different meanings. They are called homophones. In this paper, we investigate whether the meaning of homophones affects their duration, and how this effect is distributed over the sounds of the word, based on Japanese. Japanese is of particular interest because of its phonemic length contrasts, which could constrain subphonemic durational differences. Using the Discriminative Lexicon Model, we mapped the meanings (from fastText) onto word forms and derived two measures: unconditional semantic support, which represents how well the meaning of a word predicts its sounds as a whole, and conditional semantic support, which represents how this association between word meaning and form unfolds from the beginning to the end of the word. We found that, at the word level, stronger semantic support is associated with longer word duration. At the sound level, in contrast, semantic support is interacted with position in the word. Sounds earlier in the word receive more support and are realized with longer duration, while this effect diminishes and potentially even reverses for sounds later in the word. We discuss our findings in the light of morphological studies, information-theoretic approaches, and alternative theories of the mental lexicon.