Error Detection in Japanese Translation Based on Optimized Large Language Models

Cross-language translation have drawn the most significant advancements in the recent decades, particularly with the emergence of large language models(LLM). These methods have obtained promising performance in effectively translating high-resource languages. However, translation quality for linguistic and morphologically complex languages such as Japanese remains a real bottleneck for researchers. The major characteristics of Japanese language includes flexible word order, omission of subjects and its writing style, making it difficult to achieve an accurate translation. To overcome this aforementioned issue, this research paper developed novel framework that integrates the Gorilla Troop Optimization technique to fine-tune the LLM techniques to handle complex language. The complete hybrid structure was trained with the WMT datasets and comprehensive experimentation was conducted by interfacing the baseline architectures such as a LLMA andT5 models. To validate the suggested structure performance measures like accuracy, precision, recall, specificity, BLEU score are validated and compared with the other optimization model. Furthermore, test experiments are conducted to explore the improvisation factors in the existing LLM models. Results demonstrate that the proposed integration model has achieved the translation accuracy of 97.7%, BLEU score of 0.94 and error rate of 0.022 in understanding the complex languages and triggers the LLM models to reduce the translation errors.