Building a large-scale Japanese Combinatory Categorial Grammar by extracting lexical entries from corpora

本研究以日文的实用词汇化文法为研究对象，提出一种从语料库中（半）自动提取词汇条目来构建大规模文法的算法。当句子由于语言的复杂性而变长时，解析性能就会下降。在CCG、HPSG和DRT的框架下，我们分析了日语中的复杂谓词、复杂句子和关系从句等语言问题，并提出了一种基于语言形式化的人工句子处理策略，该策略也可用于从标注语料库中归纳出CCG语法。

This research project investigated a Japanese practical lexicalized grammar and proposed an algorithm of building a large-scale grammar by(semi-) automatic extraction of lexical entries from corpora. When sentences become longer because of some linguistic complexity the parsing performance deteriorates. Under the framework of CCG, HPSG and DRT, we analyzed such linguistic matters as complex predicates, complex sentences and relative clauses in Japanese, and showed a human sentence processing strategy based on a linguistic formalization, which is also available for inducing a CCG grammar from an annotated corpus.