Machine Translation: An Integrated Approach
Kuang‐Hua Chen, Hsin‐Hsi Chen · 1995
A pure statistics-based machine translation system is usually incapable of processing long sentences and is usually domain dependent. A pure rule-based machine translation system involves many costs in formulating rules. In addition, it is easy to introduce inconsistencies in a rule-based system, when the number of rules increases. Integrating both of approaches will get rid of these disadvantages. In this paper, a new model for machine translation system is proposed. A partial parsing method is adopted and the translation process is performed chunk by chunk. In synthesis module, the words are locally rearranged in chunks according to Markov model. Since the length of a chunk is much shorter than that of a sentence, the disadvantage of Markov model in dealing with long distance phenomena is greatly reduced. The structural transfer is fulfilled using a set of rules; in contrast, lexical transfer is resolved using bilingual constraints. The qualitative and quantitative knowledge is applied interleavingly and cooperatively, so that the advantages of both approaches are kept.