基于Web的无指导译文消歧词模型与N-gram模型及对比研究

发布时间：2018-02-27 09:34

本文关键词： 计算语言学无指导译文消歧词模型 N-gram模型 Page Count 双语词汇Web相关度搜索引擎对比研究模型方法上下文　出处：《电子与信息学报》2009年12期 　论文类型：期刊论文

【摘要】：该文提出了基于Web的无指导译文消歧的词模型及N-gram模型方法,并在尽可能相同的条件下进行了比较。两种方法均利用搜索引擎统计不同搜索片段在Web上的Page Count作为主要消歧信息。词模型定义了汉语词汇与英语词汇之间的双语词汇Web相关度,根据汉语上下文词汇与英语译文之间的相关度进行消歧;N-gram模型首先假设不同语义下的多义词N-gram序列行为模式不同,从而可对多义词不同语义类下词汇在实例中的N-gram序列进行统计与分析以进行消歧。两个模型的性能均超过了在国际语义评测SemEval2007的task#5上可比较的最好无指导系统。对这两个模型进行试验对比可发现N-gram模型性能优于词模型,也表明组合两类模型的结果有进一步提升消歧性能的潜力。
[Abstract]:In this paper, a word model and N-gram model of undirected translation disambiguation based on Web are proposed. Under the same conditions as possible, the two methods use the search engine to calculate the Page Count of different search segments on Web as the main disambiguation information. The word model defines the double between Chinese and English words. Web relevance, The disambiguation N-gram model based on the correlation between the Chinese context vocabulary and the English translation assumes that the N-gram sequences of polysemous words have different behavior patterns under different semantics. The N-gram sequences of polysemous words in different semantic categories can be statistically analyzed and analyzed for disambiguation. The performance of the two models is better than the best undirected system which can be compared on the task#5 of international semantic evaluation SemEval2007. The performance of N-gram model is better than that of word model. It is also shown that the results of the combined two kinds of models have the potential to further improve the disambiguation performance.
【作者单位】：北京大学计算语言学研究所;哈尔滨工业大学计算机科学与技术学院;
【基金】：国家重点基础研究发展计划(2004CB318102)资助课题
【分类号】：TP391.1

【相似文献】