Research On Interactive Machine Translation Technique Based On Positive Feedback

The technology of machine translation uses a computer to translate one natural language into another natural language,realizing the rapid conversion between different languages,and has developed rapidly since it was proposed.However,the translation provided by state-of-the-art automatic MT systems still cannot meet the quality requirements of many practical applications,and the production of correct translation needs the participation of human translators.Under such circumstances,researchers changed the focus from fully automatic machine translation to computer-assisted translation,in which the translators perform post-editing on the MT output with the help of auxiliary tools.Interactive Machine Translation(IMT)is an important research topic in the field of computer-assisted translation.It guides the computer to decode and improves the quality of the output translation through the interaction between the MT system and the translator,achieving the combination of the high efficiency of the MT system and the high accuracy of the translator.Currently,state-of-the-art IMT system takes the prefix validated by the translator as the only constraint to guide decoding.The interaction mode is limited and the interaction efficiency is low.This paper improves the IMT method from two aspects of interaction mode and decoding algorithm.In terms of interaction mode,the translators are allowed to select the correct translation of source phrases from the phrase table before translation.A re-ranking algorithm was proposed to improve the diversity of the phrase table and expand the space for candidate translations.An interface was designed according to the cognitive process of the translator to improve the user experience.In terms of decoding algorithm,the source phrase and the correct translation are taken as positive feedback information together with the prefix,which participates in guiding the decoding process and improves the accuracy of translation hypotheses evaluation and filtration.Experimentation with real users is conducted on the LDC Chinese-English parallel corpora.Average translation time is used as the main evaluation metric.Evaluation results on different corpora show that the translation time decreased by 16.80% on average,indicating that our method can reduce the user cognitive burden and improve the translation efficiency compared with the traditional IMT method.