融合主题的汉越冶金领域统计机器翻译方法

Statistical Machine Translation Method Integrating Topic for Chinese-Vietnamese Metallurgy Field

为有效利用领域知识提高汉越冶金领域的机器翻译效果,利用术语与主题的分布关系约束术语的选择,提出融合主题的统计机器翻译方法。建立术语翻译模型,抽取短语概率表时将术语对源语言文档的主题信息融入到翻译概率表中,利用主题分布描述术语与主题之间的关系,同时使用冶金领域语料训练翻译模型,将2个模型融合到基于短语的汉越翻译系统解码过程中,指导选择出最符合领域特性的术语及译文。融合模型前后的对比实验结果表明,与基于短语的机器翻译方法相比,该方法可有效提高汉越冶金领域的翻译性能。

In order to effectively use domain knowledge to improve the effect of Chinese-Vietnamese metallurgy field translation,this paper proposes a method of statistical machine translation,which uses the terms and topics＇ distribution relation to constrain the choice of terms. It constructs term translation model in the process of phrase probability table extraction, the topic information of the source language document for the term pair is incorporated into the translation probability table, and the relationship between the term and the topic is described by topic distribution. Based on the translation model of metallurgy field corpus training, the two models are integrated into the decoding process of Chinese- Vietnamese phrase-based translation system to guide the selection of terms and translations that best match the domain characteristics. The contrastive experiments are carried out before and after the fusion model. The results show that, compared with the phrase-based machine translation method, the proposed method can increase translation performance in the Chinese-Vietnamese metallurgy field effectively.