ALEF: An Enhanced Approach to Arabic-English Bilingual Translation

Translation between structurally diverse languages, such as Arabic and English, presents significant challenges due to their linguistic and cultural differences. This paper investigates the effectiveness of Facebook’s mBART model, fine-tuned for sequence-to-sequence (seq2seq) translation tasks between Arabic and English, and enhanced through advanced refinement techniques, specifically using GPT-3.5 and GPT-4. We leverage the Alef Dataset, a meticulously curated parallel corpus that captures the linguistic richness, nuances, and contextual accuracy necessary for high-quality translation. The fine-tuned mBART model achieves a BLEU score of 38.97, METEOR score of 58.11, and TER score of 56.33, surpassing popular systems like Google Translate. These results highlight the effectiveness of combining mBART with advanced language models to bridge the translation gap between Arabic and English, providing a robust and context-aware machine translation solution for diverse linguistic contexts.