Word Alignment Model Based on Maximum Entropy in Foreign Language Translation
Chen Jun · 2020 5th International Conference on Smart Grid and Electrical Automation (ICSGEA) · 2020
Automatic mining of bilingual knowledge from large-scale real bilingual corpus has become a very important way to acquire translation knowledge. Therefore, this paper proposes a word alignment algorithm to acquire translation knowledge manually. First, a hybrid heuristic word alignment algorithm is implemented on the Chinese and English parallel corpus of clause alignment. Then, through a lot of research and practice of existing methods, a maximum word alignment model with noise training is proposed. The English and Chinese word entities are extracted from English and Chinese parallel corpora to form multiple candidate entity equivalent pairs, and the eigenvalues of multiple features between each candidate equivalent pair are calculated. Finally, using the entity equivalent pair alignment module, the candidate word entity equivalent pairs are aligned using the maximum entropy model, to obtain the final set of named entity equivalent pairs. The experimental results show that the maximum entropy alignment model achieves better alignment accuracy in the noise training environment.