融合单语翻译记忆的神经机器翻译方法

Neural machine translation method integrating monolingual translation memory

摘要

与以往使用双语语料库作为翻译记忆（Translation Memory，TM）并采用源端相似度搜索进行记忆检索，进而将检索到的相似句对与神经机器翻译（Neural Machine Translation，NMT）模型融合的这种分阶段进行的方法不同，提出一种新的融合框架，即基于跨语言注意力记忆网络的神经机器翻译模型，该模型使用单语翻译记忆即目标语言句子作为TM，并以跨语言的方式执行可学习的检索。该框架具有一定的优势：第一，跨语言注意力记忆网络允许单语句子作为TM，适合于双语语料缺乏的低资源场景；第二，跨语言注意力记忆网络和NMT模型可以为最终的翻译目标进行联合优化，实现一体化训练。实验表明，所提出的方法在4个翻译任务上取得了较好的效果，在双语资源稀缺的专业领域中也表现出其在低资源场景下的有效性。

Abstract

Different from previous researches that used bilingual corpus as TM and source-end similarity search for memory retrieval， a new NMT framework was proposed， which used monolingual translation memory and performed learnable retrieval in a cross-language way. Monolingual translation memory was the use of target language sentences as TM. This framework had certain advantages： firstly， the cross-language memory network allowed monolingual data to be used as TM； secondly， the cross-language memory network and NMT model was jointly optimized for the ultimate translation goal， thus realizing integrated training. Experiments show that the proposed method achieved good results in four translation tasks， and the model also shows its effectiveness in low-resource scenarios.

关键词

神经机器翻译 / 单语翻译记忆 / 跨语言注意力记忆网络 / 低资源领域 / Transformer模型

Key words

neural machine translation / monolingual translation memory / cross-language attention memory network / low-resource scenarios / transformer model

基金资助

教育部人文社会科学研究项目(19YJC740107)

国家自然科学基金(U1908216)

沈阳市科学技术计划(20-202-1-28)