The treatment of second language learners' errors has been the subject of much debate among second language acquisition researchers and language teachers. In the framework of the interaction hypothesis (Long, 1996), recasts (i.e., teacher's utterances that reformulate or expand those of the learners while retaining their central meaning) have been investigated in order to determine whether learners' selective attention is directed to problematic features of the target language. There is an ongoing debate between those researchers who claim that recasts are too implicit to be useful in the classroom (Lyster, 1998; Lyster & Ranta, 1997), and those who claim that recasts draw learners' attention to linguistic form, making it salient and thereby creating a context for learning (Doughty, 1999; Doughty & Valera, 1998; Long, Inagaki, & Ortega, 1998; Mackey & Philp, 1998). One problematic issue for the latter concerns the questions whether recasts are in fact perceived by learners, and if their perceptions about the target of the recasts are correct.
This study addresses how learners of Japanese perceive feedback in the language classroom. Fifteen hours of teacher-learner interactions were videotaped in six classrooms, in which Japanese was taught at beginning, beginning-high, and low-intermediate levels at a U.S. university. The videotapes were transcribed, and the recast episodes were categorized by various types of error triggering feedback (morphosyntactic, lexical, phonological, and semantic errors). After the class, nine learners (three from each level) watched the recast episodes on video and recalled their thoughts at the time of feedback. The results indicated that learners were most accurate in their perceptions about phonological feedback and were relatively accurate about lexical feedback, but feedback on lexis and morphosyntax were seldom perceived correctly in the form of recasts.