This paper reports an eye-tracking study which tested the hypothesis that if phonological coding occurs in reading kanji for comprehension, skilled Japanese readers should notice more nonhomophonic errors than homophonic errors which were inserted into Japanese texts. Participants' error-recognition rates and eye-movement patterns on the errors supported this hypothesis. Furthermore, reading time was strongly affected by the existence of nonhomophonic errors. In addition, the sound differences between the errors and their correct versions had significant effects on the error-noticing behaviors, while the word and character frequencies had no significant effect, and the content variables had only weak effects. These data suggested a similarity rather than a difference between processing alphabetic and nonalphabetic scripts.

Key words: kanji, phonological coding, eye movements, reading processes