Automatic Machine Translation Evaluation Based on Sentence Structure Information

Abstract

Automatic evaluation of machine translation plays an important role in improving the performance of machine translation systems. In this paper, we firstly introduce three traditional methods of automatic evaluation, including BLEU, NIST and WER. All these methods are based on surface layer information of translations like vocabularies, so we do some studies on the evaluation method using the information of sentence structure. Because the Hierarchical Network of Concepts (HNC) theory thinks that sentence category and format transformations are two most important links in machine translation, we do some researches about sentence category and format transformations, and get the sentence structure information which is composed of sentence category information and format information of every sentence in the bilingual (Chinese and English) translation corpora. Then, considering the traditional methods above, we propose the method of automatic evaluation based on the information of sentence structure and have proved it effective by experiment.