Research On Automatic Quantification Of Translation Criterion

Translation criterion, as the term indicates itself, refers to the criterion that is observed in the translation activity. As one of the key issues in translation theory, translation criterion is often regarded as the scale for identifying the translation quality. Therefore, translation criterion is the principle as well as the objective pursued by translators. To be exact, translation is the conceptual generalization of the language and cultural observation, which regulates human translation activity. It possesses the direct instruction for translation practice since it provides principles for translators to choose between the various translation strategies and methods.Originated from the perception to the translation essence, translation criterion is the dialectical unity of universality and individuality with various beliefs. Translation criterion may be a convention for a special group of people related in translation in a specific period as well as the summarization of a single personal experience.A trend in the translation criterion study in late 20th century, which comes with the broader revolution of the theory and methodology in academic circles, is the systematic investigation of various elements involved in translation process. And one characterizing feature in this trend is the impersonal and quantitative strategy prevailing in the literature.Focused on the quantitative study of translation criterion, this dissertation proposes the automatic translation criterion quantification as a new cross-disciplinary solution to the formalization and quantification of the translation criterion. The dissertation first discloses the scientific and quantitative trend in translation criterion through a diachronic contrastive study of the translation criterion development between China and the west. The feasibility of the automatic quantification is probed via a survey of the current translation evaluation models and automatic machine translation evaluation, with the practical parameters and technology possibilities enlightened. Introducing a variety of natural language processing techniques, this dissertation further probes the automatic acquisition of translation quality features, revealing their effectiveness by correlation analysis. Finally, an automatic translation criterion quantification model is statistically established via multi-linear regression, which is proved superior to the state-of-art automatic machine translation evaluation technology in correlation to human evaluation by ten-fold cross validation.The research of this dissertation is carried in respect to the following four aspects in detail:1) To deal with"the most historic issue"in translation theory, this dissertation first conducts a diachronic contrastive exploration on the progresses of the translation criterion in both China and the west. Though they assume different traces of development, the translation criterion research in China and west share some similar stages like the argument for direct or free translation, the shift from source text focus to target reader orientation and so on. It is revealed that the translation criterion is not only influenced by the social cultures, but also confined by the science and technology available. In this sense, the current diversified, impersonal and quantitative traits of the translation criterion from multi-disciplinary discussion is brought and dominated by the information age and its distinctive cross-disciplinary researches. This part solves the theoretical premise of automatic translation criterion quantification.2) To disclose the most impersonal elements in existing translation criterion, this dissertation then makes a comprehensive survey on various translation evaluation models.It is discovered that the translation evaluation arguments have shifted from human-centered quality prescription into translating process oriented quantitative description, with abundant parameters and measures suggested. This dissertation also touches upon the automatic evaluation for machine translation, discussing its success and limitation by modeling translation criterion by simple string similarities. This part clarifies the feasibility of automatic quantification of translation criterion.3) The chief difficulty of the quantification automation remains in the automatic acquisition of translation criterion related parameters. This dissertation employs the technology of natural language processing with substantial implementation and adaptation for automatic translation information processing. With such tools as Chinese word segmentation, bilingual sentence alignment, English Chinese automatic word alignment, language modeling and parsing, etc., various word and sentence level features are obtained and further validated for their correlation to human evaluation statistically. This part paves the way for automatic translation criterion quantification in technology and in translation quality related features.4) The key issue to automatic quantification lies with the empirical modeling of the translation criterion. Enlightened by the corpus linguistics, this dissertation solved the data resource by constructing a translation testing sample corpus. Then the multi-linear regression model is suggested to best decide the weight of each translation quality feature. The achieved model is examined by ten-fold cross validation, proving a consistent superior performance over BLEU method with a 0.739 Pearson coefficient.To sum up, this dissertation attacks the issue of automatic quantification of translation criterion as a cross-discipline, probing the mathematical modeling of the translation criterion by means of computational linguistics and natural language technology. It introduces the computer technology into translation criterion study and provides a new impersonal solution to the quantitative analysis of translation criterion. Although the findings of the dissertation deserve further verification, they are of great prospects in its application in translation study and translation testing.