自动评价机器翻译译文质量的一种方法

Auto evaluation of machine translation based on BLEU method

针对机器翻译译文质量的评价存在人工评价成本高 ,可重用性差的问题 ,提出了采用BLEU方法代替人工评价 .分析了BLEU方法对汉语译文评价的效果 .根据汉语的特点 ,做了汉字模型和词模型的评价实验 ,比较了人工评价和BLEU方法评价效果 ,结果表明 ,基于BLEU的词模型的性能好于汉字模型 .

Manual evaluation of machine translation is usually costly and non-repeatable, so BLEU evaluation method is proposed to replace conventional evaluation. The effect of BLEU evaluation has been analyzed.. A character-based model and word-based model are realized for the specific feature of the Chinese language. Meanwhile, the manual evaluation result is taken as a criterion for the efficacy of the BLEU method. Results show that the word-based model is appropriate for Chinese translation evaluation.