融合检索技术的译文推荐系统

发布时间：2018-06-11 10:14

本文选题：信息检索 + 机器翻译　；参考：《哈尔滨工程大学学报》2017年03期

【摘要】：本文将基于单语语料的检索技术运用到机器翻译中,构建了一个汉英译文推荐系统,解决传统方法双语料库构建代价高昂的问题,同时提高最终译文的流畅性。译文推荐系统包括查询翻译和信息检索两部分:查询翻译根据给定的一组中文,生成N-best英文结果;信息检索评价目标语言与候选译文的相似程度。系统综合两部分得分返回推荐译文。考虑到N-best结果与候选译文的词序一致性,采用Levenshtein距离使得排序结果更加合理。在英汉数据集上的实验表明:在不同n阶语言模型下,译文推荐系统都有很好的表现,加入Levenshtein距离取得了最高70.83%的f测度值。
[Abstract]:In this paper, we apply monolingual corpus retrieval technology to machine translation and construct a Chinese-English translation recommendation system to solve the expensive problem of traditional dual-corpus construction and to improve the fluency of the final translation. The translation recommendation system includes two parts: query translation and information retrieval: query translation generates N-best English results according to a given set of Chinese; information retrieval evaluates the similarity between target language and candidate translation. The system synthesizes the score of the two parts and returns the recommended translation. Considering the consistency between N-best result and candidate translation, Levenshtein distance is used to make the result more reasonable. The experiments on English and Chinese datasets show that the translation recommendation systems have a good performance under different n-order language models, and the maximum f measure value is 70.83% when Levenshtein distance is added.
【作者单位】：北京工业大学计算机学院;
【基金】：国家自然科学基金项目(61133003)
【分类号】：TP391.3

【参考文献】