Quality Classification and Evaluation of Human-Machine Composite Translations of Scientific Text Based on KPCA

In this paper, the kernel principal component analysis (KPCA) is applied to perform the translation quality classification and evaluation of human-machine composite subject of scientific text. Firstly, four different translations are quantified by means of questionnaire survey according to the basic standards of Chinese-English translation. Then, the quantitative data is evaluated by Gaussian kernel function and polynomial kernel function. The results show that, on the one hand, the translation qualities of machine translation, professional translator and scientific researcher approximate to form an equilateral triangle in two-dimensional evaluation space, which indicates that the qualities of the above three translations is independent of each other in terms of evaluation space, and the translation quality of computer-aided scientific researcher is closest to the translation quality of professional translator; on the other hand, when the evaluation space dimension is reduced to one-dimensional, Gaussian kernel function can still get similar result, but polynomial kernel function gives different result when its order is greater than a certain threshold.