Constructing An Integrated System： Rule-based And Statistics-based Machine Translation

Machine translation, since the proposing of its funding theories, has gone through ahistory of more than60years. Prevalent approaches to machine translation have comeunder two categories: Rule-based and Statistical machine translation. However,neither proves to be a success for the fact that Rule-based Machine Translation hasbeen relying too much on the grammar rules preprogrammed in its designated modulefor parsing while Statistical Machine Translation bases every decision making onempirical knowledge during roaming procedures. In other words, Rule-based machinetranslation systems are more modular than procedural and Statistical machinetranslation vice versa.This dissertation is grounded on Chomsky grammar to illustrate the merits of modulesin Rule-based machine translation and on Nida’s Towards a Science ofTranslating(Nida1964) to elucidate the strength of procedures in Statistical machinetranslation. This dissertation will decompose the strength and weakness of themodules in Rule-based Machine Translation and of the procedures in StatisticalMachine Translation and integrate the merits of the two to establish an optimizedalgorithm with “library”,“parser” and “crawling mechanism” to better process naturallanguage and the translation of it. This dissertation applies the classic grammar inlibrary to Nida’s translation procedures to resolve the uncertainties in parsing ofStatistical Machine Translation and the deficiencies of natural language processing inRule-based Machine Translation. In the concluding part, this paper illustrates a trendof Machine Translation’s future development which is partially based on parser andlibrary to process and on crawling mechanism for reference materials. Last but notleast, a prospect in system design as well as peripheral device integration is alsobriefed.