Building a Japanese Parsed Corpus

该项目的目标是建立一个日语分析语料库，并同时改善形态分析器和分析器。在两年的项目期间，我们取得了以下成果：（a）我们增强了我们的形态分析器JUMAN，以处理一个单词串作为一个整体，并发现和输入有问题的固定表达式，被正常的形态分析错误的分析。我们于1996年10月发布了JUMAN的增强版，JUMAN 3.0。(b)我们在KNP语法分析器中加强了对并列结构和从属结构的处理，同时也加强了对几类具有特殊句法功能的短语的处理。1997年3月，我们发布了KNP的增强版KNP 2.0。(c)我们设计了一个鼠标操作的界面来帮助和加快人工对JUMAN和KNP所分配的标签的纠正，并提供了对语料库的检索功能。(d)截至1997年3月，我们已经建立了大约20，000个句子的解析和手动校正的语料库。其中，我们在1997年3月对大约10 000个句子进行了开审。

The goal of the project was to construct a Japnese parsed corpus and to simultaneously improve a morphological analyzer and a parser. In the period of two years' project, we have achieved the following results :(a) We enhanced our morphological analyzer JUMAN to handle a word string as a whole, and to find and enter problematic fixed expressions which were analyzed incorrectly by the normal morphological analysis. We released the enhanced version of JUMAN,JUMAN3.0 in October 1996.(b) We enhanced the treatment of coordination structures and subordinate structures in our parser, KNP.KNP was also enhanced to handle several types of phrases with exceptional sentential functions. We released the enhanced version of KNP,KNP2.0 In March 1997.(c) We made a mouse-based interface to help and to speed up the human correction of tags assigned by JUMAN and KNP.The interface also provides the retrieval functions for the corpus.(d) As of March 1997, we have constructed about 20,000 sentences of parsed and manually-corrected corpus. Out of them, we opened about 10,000 sentences in March 1997.