基于连接关系的汉语词典信息的推断

Chinese Dictionary Data inference with Adjacency Relations

日本和印度尼西亚、泰国、中国、马来西亚正在合作开发多国语言机器翻译系统。本文介绍这个项目中的正在研究开发的汉语词典开发辅助系统的词典信息推断功能。 这个项目从1987年开始,计划用6年时间,采用中间语言方式,实现5国语言间的互相翻译,打算利用本词典开发辅助系统进行编制的汉语词典,将主要用于这个项目中的汉语分析、汉语生成以及中文输入。 本汉语词典开发辅助系统,以大量语料为基础,收集单词,推断能从例句导出的词典信息,还可以检验人工完成的词典记述并补充词典信息。本文叙述了从分析大量语料入手,在提取词典中尚未登录的单词(形态素)的同时,建立KWIC(上下文中关键词)索引,井以这种大规模的KWIC为基础,根据源于连接关系的束缚条件推断语法特征尚不清楚的汉语单词的语法属性的方法。

This research concerns the dictionary data inference function of Chinese dictionary development support system, a part of a multilingual translation system now being developed in the cooperative research project by Japan, Indonesia,Thailand, the People's Republic of China and Malaysia.This six years project began in 198T, with the goal of translating between five countries languages through the interlingua method.The Chinese dictionary produced with the dictionary development support system will be use in Chinese analysis, Chinese generation, and Chinese input.The system is used to verify manually-entered dictionary enteries and supplement dictionary data.The system takes a large text and infers dictionary data inferable from word collections and sentences.In this paper.we describe the inference method: a large Chinese text is analyzed; words unregistered in the dictionary are output and a large KWIC is generated. Based on this KWIC, grammatical attributes for words which have none are infered, based on the constraints from adjacency relations