A large-scale grammar is needed in parsing a variety of sentences, but it is difficult to build it manually. On the other hand, it is possible to build a large-scale grammar by deriving it from a large-scale tree parsed corpus(hereinafter, only abbriviated to a corpus). But because a great number of parse trees are created by using a grammar derived from a corpus, parsing accuracy gets worse, and to make matters worse, it takes a long time and a large amount of memory to parse sentences. For the above reasons, the corpus and the grammar should be modified to decrease the number of parse trees. This paper proposes the method to build a large-scale context free grammar that creates smaller number of parse trees by leaving the analysis of the structures that cannot be solved if no semantic information is considered, and' shows the effectiveness of the grammar experimentally. We expect that it is possible to build a practical large-scale grammar that decreases the number of parse trees by using our method.
展开▼