一种混合策略的专利机器翻译系统研究

Hybrid-strategy method combining semantic analysis with rule-based MT for patent machine translation

针对专利文本翻译中的复杂语句,提出了一种基于混合策略的方法,融合语义分析技术和基于规则的翻译技术,来提高专利翻译的效果。利用语义分析技术,重点解决句子中心动词识别和句子中有嵌套结构存在的名称短语的分析,把语义分析结果输入到基于规则的翻译系统中,用以改善翻译的效果。测试结果表明,融合后的翻译系统,BLEU值提高了9.8%。该方法已经集成到了国家知识产权局的在线汉英机器翻译系统中,有效地提高了专利翻译的效果和翻译效率。

Focusing on the complex sentence translation in patent machine translation, this paper presents a hybrid method combining semantic analysis with rule-based MT. The semantic analysis is used to deal with the main verb recognition and the boundary detection of long NP. The result of semantic analysis is input to the rule-based MT system to improve the performance of translation. The result of the experiments shows that this hybrid-strategy method can effectively improve the BLEU score 9.8%. The method has been integrated into an online patent MT system running in SIPO.