一个基于格语法和转换策略的日汉机器翻译系统.docVIP

一个基于格语法和转换策略的日汉机器翻译系统 陈家骏，戴新宇，尹存燕，王启祥 （南京大学计算机软件新技术国家重点实验室， 南京大学计算机科学与技术系，南京 210093） E-mail: chenjj@nju.edu.cn 摘 要 本文介绍一个基于规则和转换翻译策略的日汉机器翻译系统的设计和实现。该系统的日语分析采用了基于短语结构文法和格语法的句法、语义分析技术。在句法分析中识别出动词短语时，利用动词格框架进行格短语的格角色识别。分析规则采用了复杂特征集和合一运算，并按层次进行设计。日语分析结果为带格角色标记的日语分析树。基于得到的日语分析树，系统采用了转换/生成一体化的汉语生成策略，按深度优先遍历分析树转换/生成汉语。另外，该翻译系统在基于规则的主框架之上，还辅助以翻译记忆的方法。本文的系统参加了863组织的三次机器翻译评测，其中，在2005年的评测中，自动评测的结果（NIST）为：6.3052（对话）和6.7836（篇章）。 关键词：机器翻译，日语，汉语，转换，规则，格语法 中图分类号：TP391.2 A Japanese-to-Chinese Machine Translation System Based on Case Grammar and Transformation Strategy Chen Jia-Jun, Dai Xin-Yu, Yin Cun-Yan, Wang Qi-Xiang (State Key Laboratory for Novel Software Technology, Department of Computer Science Technology Nanjing University, Nanjing 210093) E-mail: chenjj@nju.edu.cn Abstract This paper describes the design and implementation of a Japanese-to-Chinese machine translation system. In the system, rule-based transformation is adopted as the main translation strategy. In the Japanese syntactic and semantic analysis, phrase structure grammar and case grammar are used. As soon as verb phrases are recognized during Japanese analysis, the determinations of semantic role of each case phrase are carried out based on verb case frames. Analysis rules consist of features sets and unification operations, and are arranged in several levels. The analysis results are syntactic trees with semantic roles attached to each case phrase. By traversing the analysis trees on depth-first, transformation and generation are performed as a single procedure to produce corresponding Chinese translations. Translation memory techniques are also adopted in this system. The system took part three times in the MT evaluations organized by national 863 high-tech projects. The evaluation results (NIST) in 2005 are 6.3052(dialog) and 6.7836(essay). Keywords: Machine Translation, Japanese, Chinese, Transformation, Rules, Case Grammar 概述 从基本策略上，机器翻译可分为：直译、转换翻译和中间语言三种体系结构。直译适合于语言结构比较相近的语言之间的翻译；转换则适合于语言结构有一定差距的语言之间