This paper presents an approach to improve the performance of the IBM Translation Model for Vietnamese Sign Language (VSL) automatic translation. The main contribution is the optimization of the EM algorithm in the IBM model by adding an α coefficient to match the strings in the alignment process. The results show that they only converge to 1 after 2 iterations with the test set experiment. The BLEU translation quality score of the model is described. Data enrichment shows a significant improvement in translation performance compared to the original data for both models. Specifically, for the translation model on the improved IBM model, the BLEU score increases from 42.31 to 60.32; For statistical translation on the improved IBM model, the BLEU score increases from 48.75 to 76.25. This shows that the use of enriched data can help improve translation quality. With the improved IBM model statistical model, the translation performance is better than the IBM model translation model when using the original data; higher translation performance and significant improvement when using enriched data. Additionally, the high BLEU score can be attributed to the model's language convergence, where the translation model remains relatively consistent with most language units being identical in both languages.This finding suggests that the proposed approach effectively enhances the IBM Translation Model for VSL automatic translation, leading to more accurate and fluent translations.