Research On Example-Based Automatic Machine Translation For English-Chinese Patent

Example-Based Machine Translation (EBMT) method was first proposed by Japanese scholar Nagao Makoto, which had been one of main methods of machine translation after several years’development. The main characteristics of EBMT are taking examples base as main knowledge resource, decomposing sentence into segments by some strategies, combining the translation of segments into the translation of sentence. These methods made the translation result more reliable, but the scale of example base is an important factor in EBMT. The characteristics of patent make it possible to meet the need of EBMT translation using a certain number of examples.This paper presents an implementation method of example based automatic translation of the English Chinese patent, the main work includes:Firstly, this paper presents a sentence segmenting method in patent corpus, which takes independent translation pairs as segmentation units. This method effectively reduces the length of the sentence, decreases the complexity of syntactic analysis and word alignment. Experiments show that the method is effective to improve the performance of the translation system.Secondly, this paper presents a method to compute similarity of sentences, which bases on multi-feather fusion. The method fully considers the key factors in the process of creating the target language, such as words, syntax, verbs and so on; and it improves the quality of the target language’s generation through feather fusion.Finally, this paper implements a translation generation method based on examples, which divides the process of translation into four steps: runtime sentences segmenting, examples filtering, unmatched parts processing and examples revaluing. This method uses viterbi algorithm to decode, and chooses the final translation result from multiple translation results.