面向专利领域的汉英机器翻译融合系统  

A Hybrid System for Chinese-English Patent Machine Translation

作　　者：李洪政[1] 赵凯 胡韧奋[1] 蒋宏飞 朱筠[1] 晋耀红 LI HongZheng ZHAO Kai HU RenFen JIANG HongFei ZHU Yun JIN YaoHong(Institute of Chinese Information Processing, Beijing Normal University, Beijing 100875, China Beijing Qihu Technology Co., Ltd., Beijing 100015, China Beijing Dinfo Technology Co., Ltd., Beijing 100107, China)

机构地区：[1]北京师范大学中文信息处理研究所,北京100875 [2]北京奇虎科技有限公司,北京100015 [3]中科鼎富(北京)科技发展有限公司,北京100107

出　　处：《情报工程》2017年第3期105-115,共11页Technology Intelligence Engineering

基　　金：国家高技术研究发展计划基金项目"海量文本多层次知识表示及中文文本理解应用系统研制"(2012AA011104)的资助

摘　　要：面向专利领域的机器翻译近年来已成为机器翻译的重要应用领域之一。本文提出了一个汉英专利文本机器翻译融合系统,该系统以规则系统为主导搭建,并把规则翻译方法和基于短语的统计翻译系统相结合。在融合系统中,规则系统主要负责源语言的分析和转换阶段的处理,生成相应的源语言句法分析树与转换树,并确定目标语言的基本句法框架。统计翻译系统则在目标语生成阶段根据生成的目标语句法结构寻找合适的对译词形,并产生最终的候选译文。通过利用自动评测指标对融合系统进行测试,融合系统的结果均优于单个规则系统和统计系统的结果,表明了融合方法的有效性和可行性,可以改善系统的翻译性能,提高翻译质量。Machine translation towards patent domain has translation in recent years. This paper presented a nove become one important application of machine hybrid system, which combines rule-based machine translation （RBMT） with phrase-based statistical machine translation （SMT）, to translate Chinese patent texts into English. The hybrid architecture is basically guided by the RBMT engine which processes source language parsing and transformation, generating proper syntactic trees for the target language. In the generation stage, the SMT subsystem provides proper lexical selection according to the generated grammatical structures and produces final translation. According to the experimental evaluation, the hybrid approach outperforms each individual system over sets of automatic evaluation metrics, indicating that the proposed method performs well in improving translation results.

关 键 词：专利 规则方法 统计方法 融合系统 机器翻译 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]