Abstract: Both speech production and comprehension involve the representation of homophones in lexical access. The shared representation model by Levelt and the independent representation model by Caramazza for homophones were derived from the Two-Step Activation Model and Independent Network Model, respectively, in studies of speech production. These two models were tested with word frequency effect in language experiments and phonological treatments on patients. The article reviews the advance in related studies and explores the differences of the two models for homophone representation, suggesting that the representation of homophones has much to do with language difference, processing paradigms, perceptual modalities, and etc. From the studies on speech comprehension (speech perception and word recognition), however, it shows that the two representation models for speech production have difficulties explaining the representation of homophones, especially of Chinese homophones. Suggestions are made for feasible models of homophone representations according to the findings of resent studies on speech comprehension.

Key words: homophone, representation, speech production, speech comprehension, Chinese homophone