Computerized Translation Incorporating Grammar Awareness and Tag Smoothing
Xiaoyan Lv · Advances in transdisciplinary engineering · 2025
With the rapid development of artificial intelligence and neural machine intelligence algorithms translation technology, improving the adaptability of translation models to grammatical properties and translation diversity has become a key research direction. The study, in order to explore a new approach to computer translation that integrates syntax-aware and label-smoothing strategies in intelligent algorithms, captures explicit syntactic features and implicit semantic information through a dynamic expert hybrid model with the Transformer architecture that combines component attention, and introduces label smoothing to reduce the overconfidence phenomenon of the model. Experimental results show that this new method has the highest bilingual evaluation replacement score of 49.3 and exhibits a translation fitness with a mean value of 0.6 in two types of public translation datasets. Quantitative data found that the new method had the most 91.98% translation accuracy, the highest grammatical consistency score of 9.13, and the lowest average translation latency of 0.88 seconds. The results show that the proposed method achieves a balance between quality and efficiency in multilingual translation tasks, and provides a new direction for further improving the performance of neural machine translation.