We analyzed a Japanese lexical database to investigate the structure of the lexical environment based on the hypothesis that the lexical environment is optimized for the functioning of verbal working memory. Our prediction was that, as a consequence of the cultural transmission of language, low-imageable meanings tend to be represented by frequent phonological patterns in the current vocabulary rather than infrequent phonological patterns. This prediction was based on two findings of previous laboratory studies on verbal working memory. (1) The quality of phonological (phonemic and accent) representations in verbal working memory depends on phonological regularity knowledge; therefore, short-term phonological representations are less robust for words with infrequent phonological patterns. (2) Phonological representations are underpinned by contributions from semantic knowledge; therefore, phonological representations of highly imageable words are more robust than those for low-imageable words. Our database analyses show that nouns with less imageable meanings tend to be associated with more frequent phonological patterns in Japanese vocabulary. This lexical structure can maintain the quality of phonological representations in verbal working memory through contributions of semantic and phonological regularity knowledge. Larger semantic contributions compensate for the less robust phonological representations of infrequent phonological forms. The quality of phonological representations is preserved by phonological regularity knowledge when larger semantic contributions are not expected.