Automatic Construction of a Sense-disambiguated Multilingual Dictionary

提出了一种从现有双语词典中生成多语种词典的新方法，为多语种机器翻译和信息检索提供了必要条件。不同语言中有一个或多个共同翻译的两个词是否真正是彼此翻译的，是根据从文本语料库中提取的上下文之间的相似性来判断的。以英汉双语词典和日语、汉语报纸语料库为基础，进行了英、日、中三语词典的编撰实验，结果表明该方法的有效性，准确率达到83%，F-score达到73%。

A novel method was developed for producing a multilingual dictionary, which is indispensable for multilingual machine translation and information retrieval, from a set of existing bilingual dictionaries. Whether two words in different languages having one or more translations in common are genuinely translations of one another is judged based on the similarity between their contexts extracted from text corpora. Its effectiveness was demonstrated through an experiment on producing an English-Japanese-Chinese trilingual dictionary from English-Japanese and English -Chinese bilingual dictionaries and Japanese and Chinese newspaper corpora, in which 83% precision and 73% F-score were attained.