Language ambiguity is the core challenge in machine translation.To overcome the limitation of existing neural machine translation (NMT) models that are prone to semantic deviations in complex contexts, this paper proposes a multi-engine architecture based on the attention mechanism.It achieves ambiguity resolution by dynamically integrating the context modelling of neural networks, the grammar constraints of the rule engine, and the historical knowledge of the statistical engine.Experiments on public datasets such as workshop on machine translation and discourse in machine translation show that this system improves the bilingual evaluation understudy score by 2.1 points in the English-German translation task, increases the accuracy of ambiguous phrase translation by 15.7%, and its semantic coherence (normalised discounted cumulative gain) is significantly better than the mainstream baseline models.This research provides an effective solution for building a robust and accurate context-aware translation system.
Ziao Zhang (Thu,) studied this question.