Machine Translation of Quranic Verses: A Transformer-Based Approach to Urdu Rendering

Abstract

Translate Quranic Arabic into Urdu is a Challenge due to linguistics and theological differences. While machine translation has advanced significantly, transformer-based Neural Machine Translation (NMT) models have not yet been utilized for Quranic Arabic to Urdu translation. This study addresses this gap by developing a transformer-based model that ensures accurate and context-sensitive translation of Quranic verses. A dataset has been initialized that contains Quranic Arabic text and Urdu translation of respected. I performed preprocessing on the dataset by applying it towards tokenization, stemming, and lemmatization, without compromising the theological nature of the theme. To enrich the model to mine the linguistic and stylistic cues, transformer architectures such as Helsinki NLP/MiarinMT were used with the transfer learning. Finally, the model was evaluated for theological correctness by Islamic scholars, and, secondly, by some automated metrics (BLEU, Rouge, and Cosine Similarity). Results show that the transformer model is a better model by far that provides better translation quality in the sense that meanings are preserved, that is, contextual meaning as well as religious meaning, implying better accessibility to Urdu-speaking Muslims. This research proposes a new approach to the problem of translating sacred texts and solves, albeit theologically correct, otherwise unsolvable problems in Quranic translation, computational linguistics, and AI development. This research introduces a novel approach to Quranic translation, and Future work will explore multimodal learning for deeper contextual understanding.