Study On Translation Evaluation For Science & Technology Literature Based On Model Of Neural Network

Translation evaluation is an important part of translation theory, which contributes a lot to the development of theory. However, it is also found to be a less-developed part in the whole translation research. Through summing up the literature on translation standard associated with science and technology area, the dissertation proposes five factors throwing much influence on translation quality, including faithfulness and precision, expressiveness and intelligibility, logical clarity, proper use of terms and the principle of catering to professional readers. As is discussed by translation researchers, translation evaluation is agreed to be a mathematical system characterized by complexity and non-linear nature since these factors are inner connected in a way devoid of evidence. Neural Network is a better tool in resolving complex non-linear problems. In the study, the author makes the five factors an index system in order to get a quantified data. First, they are classified into five scales, 'excellent, good, medium, pass, failure', with each representing a digital interval. Then a questionnaire is conducted by marking five translated versions according to the offered indexes by thirty participants consisting of translation theorists, translators and professional readers. Then the author uses Back Propagation network to train the evaluation model with the data gained from these readers. By self-learning and adjusting priorities in translation evaluation, the model is finally capable of producing synthetical assessment on translated versions. The results obtained by means of two different methods are compared and analyzed in the dissertation too. It is found that the difference between them is quite big. Only 60% of them are concord, which means the traditional evaluation by human is less objective and lower in accuracy and fails to produce answer close to the fact. On the contrary, due to the objectivity and artificial intelligence possessed by NN, the result with this approach is proved to be more reliable and accurate.