交互式机器翻译中译文查询行为的预测技术

以交互式机器翻译(IMT)为研究背景,针对用户在翻译过程中调用双语查词而导致的鼠标-键盘频繁切换的问题,提出了一种面向翻译查词行为的预测模型。该模型将查词行为转化为当前翻译条件下的译文选择问题,利用对齐模型、翻译模型和语言模型实现了高准确率的查词行为预测。在人工双语对齐语料的测试中,该方法预测准确率约为64.99%,特别对占有比例较高的名词预测正确率可达71.43%,能有效减少人工翻译中的重复性和机械性操作,改善了用户使用交互翻译系统的翻译体验,从而提高翻译效率。

Prediction technology of translation query behavior in interactive machine translation

To deal with the problems of frequent switches between mouse and keyboard when the users call the bilingual words searching in the process of translation, a prediction model oriented word searching behavior was proposed based on Interactive Machine Translation (IMT). This behavior of word searching was transformed into the selection problems under the condition of current translation, and the prediction of the word searching behavior was achieved with a high accuracy using the alignment model, translation model and language model. In the tests of artificial bilingual alignment corpus, the prediction accuracy of the approach is about 64.99%, and especially for the prediction of noun, the prediction accuracycan reach 71.43%. The approach effectively reduces the repeatability and mechanical operations in the manual translation, and improves the user's translation experience when using the interactive translation system, which also improves the translation efficiency.