On Translation Unit:Machine Translation Approach

Translation unit is a crucial subject in translation theory and practice.Scholars at home and abroad have paid much attention to this issue for a long time.They have carried out a fierce discussion regarding the nature of translation unit and the operational unit in translating,but a general review of previous studies has found that different explanations on this problem resulted in different conclusions.What's more,controversy still exists till today.Machine translation aims to use computer and its software system to realize the translation of text or speech from one natural language(source language)to another(target language).As an important research field of translation,especially the important part of translation practice,studies that combine machine translation with translation theory are far from enough.The idea of translation unit comes from translation theory and requires to be used for instructing actual translated texts as well.An investigation from the perspective of machine translation will be able to provide a relatively real and objective verification platform for the research of translation unit.Based on the above research background,this thesis attempts to take machine translation approach as the cut-in point.With a mixed method of qualitative research and quantitative research,statistical analysis on translation unit in 134 machine translation systems has been conducted.First of all,through an introduction of the existing research achievements and the application of theoretical framework,this study has indicated that in the real translation situation,translation unit should be built on the hierarchy of linguistic units.Second,through the collection of machine translation systems that have contested in the translation evaluation competitions,this study has made a preliminary description on the quantitative distribution of translation units in machine translation systems.Finally,through the emphasis on the desirability and limitations of translating at various linguistic levels in human translation,this study has confirmed the deficiency of translation unit selection by translators,which in turn enhances the rationality of translation unit selection in machine translation systems.Through the presentation of research design,this thesis primarily answers the following questions: 1)In theory,translation unit can appear in the form of any linguistic units;then in practice,what is the distribution tendency of operational unit in machine translation systems? 2)Whether there is a centralized choice on translation unit in machine translation systems? If it does,which linguistic levels can be chosen as translation unit? 3)What is the rationale and merits of translation unit selection from the viewpoint of machine translation? In comparison to studies on translation unit in human translation,what revelations can be offered in this study?After the theoretical exposition and data analysis,major findings that have been obtained are as follows: 1)The distribution of the units of operation in machine translation systems tends to be on or below the sentence level.2)The basic and most frequently-used translation unit of machine translation systems is determined at the phrase level.3)Taking phrase level as the fundamental translation unit has two advantages: firstly,word level as the basic translation unit has rich meanings and flexible forms,while phrase can provide specific translation context for single word,which is helpful to word sense disambiguation;secondly,linguistic units that are on or above the sentence level have intricate structure.If these levels are seen as the basic translation unit,it would lead to a low repetition rate and recognition degree,which would further make translation efficiency worse.However,phrase level often has a moderate length,and the repeated word co-occurrence gives a reliable statistical support for phrase as a whole to be translated.This study has established a close link between theory of translation unit and concrete practice of machine translation systems,which creates a new attempt to study translation unit.The future research direction lies in two aspects.For one thing,the source of machine translation systems that serve as research samples is narrow and the quantity of those systems is limited.More extensive source and larger number of research samples should be included to intensify the persuasion of statistical results.For another,the current research mainly concentrates on how to determine phrase level as the translation unit and why phrase level should be selected as the translation unit in machine translation systems.Later researches can make a deeper exploration on how to use machine translation systems to extract various kinds of effective phrasal units.By doing that,it will lay foundation for building large translation database and providebetter service for later translation practice.