Extending new language in NLLB-200: language informal Malagasy

This study focuses on integrating informal Malagasy into the NLLB-200 model for machine translation. The model underwent supervised pretraining, which quickly led to improved performance, marked by a significant reduction in both loss and perplexity. This step allowed the model to effectively adapt to the unique linguistic structures of Malagasy. The evaluation of key translation metrics such as BLEU, ROUGE, and BertScore showed that the model produces high-quality translations, combining fluency with semantic coherence. Although the BLEU score was moderate, the ROUGE and BertScore results revealed a remarkable level of lexical and semantic fidelity. This work highlights the importance of developing translation systems that can handle low-resource languages, which are often overlooked by traditional technologies. The study also demonstrates the model’s ability to grasp the nuances of informal Malagasy, resulting in significant improvements over existing translation tools. In conclusion, this approach emphasizes the need to include informal languages in translation systems, paving the way for more inclusive and linguistically tailored applications.