This paper introduces the new parsing technique "gap-filling LR parsing." This technique enhances the robustness of the generalized LR parsing and is very useful in speech recognition in that the parser can skip the very noisy part of the input The unidentified portion is resolved by the re-utterance of that portion by the speaker, which is parsed very efficiently by using the parse record of the first utterance. Experiments in parsing Japanese spoken! input show the significant effectiveness of this interactive approach.