An Evaluation Of Output Of Domestic MT Software

As the name indicates, Machine Translation (MT) is the transfer of meaning from one natural (human) language to another by computer or with the aid of the computer. In this sense, MT can be considered as a subject which straddles linguistics, mathematics and computer science. MT plays an increasing important role in this information-explosive age. After having undergone a long and winding way, MT has grown from the laboratory novelty to a commonly used commercial tool. In the field of MT, there emerge two main schools, including ationalist and mpiricist The former is rule-based MT (RBMT) and the latter consists of example-based MT (EBMT) and statistics-based MT (SBMT). Simultaneously, much progress has been made in the quality of MT. However, progress does not mean success, for the output of MT is still far from being satisfactory. As a result, there remains a debate both at home and abroad as to where MT is going and whether MT can replace human translation. To answer the question, we should first find out how well MT has been developed in terms of its output quality and what problems there are in MT. Based on the idea, the author starts with an evaluation of the quality of the domestic MT software in the aspects of smoothness and faithfulness. According to the statistical results, the MT performs well in dealing with the sentences with similar sentence structures and order; while the MT performs poorly in handling the following issues of ambiguities, preposition, conjunction and more difficult sentences. By investigating the causes of the results, the author finds out that there exist two pairs of unsolvable contradictions in MT when observing the issues from the aspects of the nature of translation and linguistic characteristics. Of course, we do not exclude the factors of software programs. The contradictions include 1) the flexible use of language and the rigid deduction of rules; 2) the limited rules and the unlimited language. The two pairs of contradictions constitute the unbreakable limitation of the currently rule-based MT. It follows that the interactive MT, which has long been ignored, should be regarded as the main stream of MT at present. The author also puts forward her solution to optimize the currently rule-based MT system, that is, to visualize such a system which is specialization-oriented and rule-based and takes in the basic idea of example-based. During the interactive MT, the improved sentences are put into machine translation memory system for future use. As the corpus increases, the quality of MT system will be improved and human workload will be decreased.