Modified output, second language (L2) learners' reformulation of their own utterances, has been attracting researchers' interest as an important component of learner interactions, and as a manifestation of interlanguage development and psycholinguistic processing. The output hypothesis (Swain, 1985, 1993, 1995, 2005) claims that the act of production constitutes part of the process of L2 learning in terms of noticing, hypothesis testing and metalinguistic functions. This hypothesis has been used as a theoretical framework to investigate the relationship between modified output and L2 learning (e.g., McDonough, 2001, 2005; Nobuyoshi Ellis, 1993; O'Relly, Flatiz, Kromrey, 2001; Takashima Ellis, 1999). However, the empirical evidence from these studies does not yet appear to confirm unequivocally that the production of modified output facilitates L2 learning. The present study further explored the impact of modified output on L2 learning, by means of an experimental pre-test, post-test and delayed-post design. The production of modified output was triggered by one type of implicit feedback, clarification requests. The data were collected from 28 undergraduate students who were learning Japanese as a foreign language. The target linguistic feature was the negation of adjectives in Japanese, and a total of 1,011 negations were elicited and analysed. The impact of modified output on L2 learning was measured in two different aspects of potential outcomes of modified output (i.e., grammatical accuracy and interlanguage development). In addition, the study investigated whether the non-targetlike forms which participants previously modified were then produced in the subsequent situations of use. The output hypothesis was originally framed in terms of the relationship between output and grammatical accuracy, but the findings of the current study suggest that production of modified output in response to clarification requests may facilitate the progress of interlanguage development towards targetlike use even when its immediate impact on grammatical accuracy may not be observed. Therefore, the present study lends at least partial support to the claim of the output hypothesis. The results did not clearly demonstrate whether or not production of modified output might sensitise learners to avoid the use of the same non-targetlike form that they have previously modified. This indicates a possibility of the limited role of production of modified output in L2 learning, and suggests that the follow-up feedback to learners' modified output may be necessary to maximise the impact of modified output in facilitating L2 learning.