【摘要】 由于维吾尔语具有音变差异，在记录时容易混淆，这对机器翻译的容错能力提出了挑战。从词粒度研究的角度出发，针对维吾尔语中具有明显语音特性的音变词的翻译问题，采用字符级神经网络机器翻译（Character level Neural Network Machine Translation,CharNMT）结构，并结合覆盖率机制进行优化，这一方法不仅提高了音变词的翻译质量，还展现出较好的容错能力。在基础模型取得了1.25 Blue分的提升，并在训练和解码速度方面也具有优势。更多还原

【Abstract】 The Uyghur language’s phonetic variation introduces challenges in machine translation due to the potential for recording errors. This study focuses on the translation of phonetically variant words with unique phonetic attributes from a word granularity perspective. By leveraging the Character level neural Network Machine Translation(CharNMT) framework and integrating it with a coverage attention mechanism,we enhance the translation quality of such words while also showcasing robust fault tolerance. Foundational model achieved a 1.25 point improvement in BLEU score and demonstrated superior training and decoding speeds.更多还原