Linguistic Analysis On Machine Translation Outputs

Machine translation (MT), in terms of definition, is automatic translation of human language by computers. The paper begins with the birth and growth of MT and then converts to human translation and MT software, conveying a pre-representation of the whole topic. Earlier MT software is somewhat a failure. But some new MT systems show a certain feasibility in reality, for example: Esopo (a Machine Translation System (MTS) based on Noam Chomsky's Generative-Transformational Grammar), whose aim is to have the quality translation with the minimal interaction from the user and China's Yaxin/CAT translation system, which possesses somewhat a memory function. While enormous tests and enhanced efforts made consistently, they still remain inadequately applicable in society. Why is it so hard to find the ready application? Many factors to be examined in this paper will clarify the difficulty of MT, including words with multiple meanings, sentences with multiple grammatical structures, uncertainty about what a pronoun refers to, simile, metaphor and other problems of grammar, mainly because natural human language is so ambiguous and so full of special constructions and exceptions to rules. Thus this paper proceeds to observe the many MT fails in various ways, including translation theories: ideal translation theory, aesthetic translation theory.... It can't identify clearly some rhetorical phenomena and newborn words etc. consequently the products of MTS are always of low readability and the readers are somewhat unsatisfied and disappointed. But MT does create a wholly new world for the human being because of its high capacity and speed, convenient retrieval and return. It can handle large volumes and maintain automatically consistency. Thus it is ideal for large scale and/or rapid translation of technical documentation, software localization manuals, and real-time translation of weather reports. While in the near future it seems impossible to employ a human untouched machine translation and it is also ridiculous to manly translate a large document without the help of a computer. Thus compromise should be maintained between them. Specifically in the machine translation field, further linguistic research has provided the corpus-based approaches of statistical language model in dealing with natural language processing, which attempts to approach the bottleneck problem in MT so as to perfect machine-aided translation systems.