Machine translation technology has been developing rapidly in general-purpose fields, but accurate translation of specialized terminology is still difficult. In some specialized fields, the accuracy of terminology translation has a significant impact on the overall translation quality, and specialized improvement methods for terminology translation need to be developed. In this paper, we propose an English translation neural network model that incorporates terminology information with the aim of improving the English translation of specialized terminology in computer translation. Taking the Transformer model with self-attention mechanism as the baseline, the model is optimized through three key steps: firstly, the bilingual corpus is customized with terminology dictionaries for terminology segmentation; secondly, the word vectors are trained on the general corpus and the specialized corpus by combining two methods, Glove and Word2vec, respectively, and are used for the initialization of parameters in model embedding layer; and lastly, for the problem of unregistered words, the introduction of the external terminology dictionary for lookup replacement. The experimental results show that on the electrical engineering domain corpus, the BLEU value of the proposed model reaches 35.1%, which is 1.36% higher than that of the baseline model; during the training process, the performance advantage of the present model continues to be stable when more than 10,000 steps are taken; compared with the other seven translation models, the present model obtains the optimal translation effect by increasing the training time by only 5.86%. The experiment proves that the English translation neural network model fusing terminology information can effectively improve the translation quality of specialized terminology, and provides a new solution for machine translation in vertical domains.
