Hybrid Combination of Machine Translation with Part-of-Speech Analysis

Abstract

Hypothesis combination is a main method to improve the performance of machine translation (MT) system. The state-of-the-arts strategies include sentence-level and word-level methods, which has its own advantages and disadvantages. And, the current strategies mainly depends on the statistical method with little guidance from the rich linguistic knowledge. This paper propose hybrid framework to combine the ability of the sentence-level and word-level methods. In word-level stage, the method select the well translated words according to its part-of-speech and translation ability of this part-of-speech of the MT system which generate this word. The experimental results with different MT systems proves the effectiveness of this approach.