Research And Application On Dynamic Word Alignment For Interactive Translation

As a fundamental corpus processing technology, word alignment has an importantapplication value in machine translation, term extraction, cross-language informationretrieval and lexicography. The target of word alignment processing is receiving thecorresponding relationship of words level from bilingual parallel corpus automatically on thebasis of sentence alignment.Word alignment measure based on statistics requires a large scale of bilingual corpus asinput, it is difficult to avoid the problem of data sparse and algorithm time overhead. Thispaper presents an efficient word alignment algorithm based on bidirectional dictionary andsemantic similarity calculation in order to satisfy the demand for real-time alignment ofsentence or paragraph level. This paper realizes a real-time method of word alignment, whichuses the English-Chinese dictionary and Chinese-English dictionary for word alignment,whose basic idea is bidirectional integration.The paper uses the method of dynamic block segmentation and matching based ondictionary driven, which needn’t segmentation processing to Chinese sentence. The methodavoid the problem effectively that dictionary cannot be used in alignment due to improperChinese word segmentation and increase the translation coverage of the bilingual dictionary.In addition, the algorithm effectively solved the case of identical English nodes and crossedChinese positions, and many to many problems by taking maximum matching conflictresolution principle, closest matching principle and pruning disambiguation strategy.Combined with the diversity of language expression and the flexibility of translation inactual translation, it can be found that it is impossible for dictionary to fully include theexplanation of words. Aimed at the problem, after the bidirectional integration, the algorithmimproved the recall rate of word alignment obviously, by making the similarity extendingalignment based on HowNet to the block which is not aligned on semantic level.Compared with the standard algorithm, the experimental results show that the accuracyrate and recall rate can be effectively improved by this alignment method, which has better adaptability on the aspect of small-scale bilingual corpus and real-time alignment suchas interactive translation.