Automatic Acquisition of Semantic Elements Based on Statistical Decomposition

Abstract

The implement of machine translation system based on semantic element (SE) requires a large scale semantic element base. An automatic acquisition method for SEs based on statistical decomposition was proposed in this paper. Firstly, .analogical decomposition was applied to decompose abandonable SE and two conditions are found to assure the correctness of decomposition. Secondly, three statistics were designed, namely parameter index, translation index and language model index, to measure the rationality of analogical decomposition. Finally, multi-attribute decision making values for the above three statistics were prioritized to choose the best decomposition. Abandonable SE could be decomposed iteratively thereafter. Experimental results show that the accuracy has reached around 60%.