A Deep Learning Approach to Semantic Clarity in Urdu Translations of the Holy Quran

Abstract

The Holy Quran holds profound significance from both religious and linguistic perspectives yet its Urdu translations face difficulties in preserving the original meaning because of ambiguous words which create interpretation challenges for speakers and listeners. This research tackles translation ambiguity in the Urdu translations of the Holy Quran authored by Maulana Abul A’ala Maududi and Fateh Muhammad Jalandhry by applying Word Sense Disambiguation methods with deep learning algorithms. A model based on multilingual BERT identifies ambiguous word senses for Surah Al-Baqarah in particular. The dataset features Surah Al-Baqarah's complete Urdu translation together with a Sense Inventory that contains 3 to 8 senses for 50 frequently used Urdu ambiguous words which are collected from GitHub repository. Sequence classification frameworks within BERT receive contextual embeddings during fine-tuning. The evaluation framework includes the determination of F1 scores alongside confusion matrix analysis and classification report assessment. The model achieved an F1-score of 0.82 when identifying the most frequent sense while reaching an average F1-score of 0.62 across eight predefined sense labels. A sense prediction system functions to improve word sense matching thereby leading to more precise translations. The proposed research makes significant contributions to computational linguistics and Quranic studies by delivering an expandable method that solves word sense ambiguity while offering important insights to help translators and scholars improve their understanding of how context affects meaning within translated texts.