Analysis of Strategies and Skills of English Translation Based on Coverage Mechanism

This is the authors' abstract. We don't add key points for this paper.

In order to alleviate the problem of over translation and missing translation in NMT, based on the consistency and complementarity of information stored in different covering models, a multicoverage fusion model is proposed, which uses coverage vector and coverage score to guide the attention mechanism at the same time. First, the concept level definitions of words are covered. Then, two kinds of translation history information stored in the cover vector and cover score are used to guide the calculation of the attention score at the same time. Finally, the dual attention decoding method based on the fusion coverage mechanism is adopted. The experimental results show that the multicoverage fusion model can improve the translation quality of NMT.