Abstract
Word-dy-word incremental parsing has a serious issue that the results given sequentially do not necessarily represent the correct syntactic relations. That is, syntactic substrutures predictively computed for initial fragments in the incremental parsing process may become wrong for the whole input sentence. This paper proposes a method for calculating the probability that a syntactic structure computed in the incrementalparsing process is correct. Our method is based on probabilistic context free grammar, and it incrementally calculates the probability for each syntactic structure on a word-by-word basis. In our method, incremental parsing outputs syntactic structure at the point when the probability for the structure becomes greater than an adequately determined threshold. Our technique improves the accuracy of incremental parsing. We havemade an experiment using ATIS corpus and ATR speech and language database. The experimental results have shown the effectiveness of our method.
