基于LSTM的交互式神经机器翻译方法研究

Research on Interactive Neural Machine Translation Method Based on LSTM

神经机器翻译系统在过去几年取得了很好的结果,但是仍然是有缺陷的。机器翻译系统的输出必须在后期编辑阶段由翻译人员进行校对。交互机器翻译促进了人机协作,提高生产力。在这项工作中,我们将神经机器翻译集成到交互式机器翻译框架中。利用seq2seq框架的特性,在解码阶段将翻译人员的纠正信息融合进机器翻译系统,在保持现有信息的情况进行重新解码翻译。并且利用翻译人员的的先验知识对神经翻译系统进行增量训练,以提高机器翻译系统的表现。

Neural machine translation systems have achieved good results in the past few years,but they are still flawed.The output of the machine translation system must be proofread by translators in the later editing stage.Interactive machine translation promotes human-machine collaboration and increases productivity.In this work,we integrate neural machine translation into an interactive machine translation framework.Using the characteristics of the seq2seq framework,the translator’s correction information is integrated into the machine translation system during the decoding stage,and the existing information is re-decoded and translated.And use the translator's prior knowledge to incrementally train the neural translation system to improve the performance of the machine translation system.