Durational differences among Japanese homophones as a function of their meanings

Actual pronunciations, i.e. phonetic realizations, of words have been assumed to be free from influence from upper-level information such as semantics. One consequence of the assumption is that phonetic realizations should be determined solely by phonological makeups such as segments which words are made of. Words that are said to have the same pronunciation, such as time and thyme in English, called homophonous words, are therefore predicted to show the same phonetic realization, according to the traditional assumption mentioned earlier. In this talk, this assumption will be challenged by recent evidence with a Japanese conversational corpus, finding that homophonous words are systematically different in their phonetic realizations according to semantics. Different meanings, different realizations. The finding is consistent with Discriminative Lexicon Model (Baayen et al, 2019), a framework in which semantics and phonetics are connected directly without any intermediate linguistic levels.