Content area
The translation of Qur’anic verses poses significant linguistic challenges due to the complexity of Arabic morphology, syntax, and semantics. Additionally, metaphors and stylistic expressions within the Qur’an present further obstacles to accurate translations. This thesis addresses these challenges by developing and evaluating three models tailored for Qur’anic studies: a Text-to-Text Transfer Transformer (T5) translation model, a semantic textual similarity (STS) model, and a metaphor detection model.
The T5 model is built from scratch to translate English Qur’anic verses to Arabic and incorporates a custom Arabic tokenizer and a cleaned dataset of English-Arabic verse pairs. The STS model is employed to measure the semantic alignment between translated verses and their original texts. Finally, the metaphor detection model aims to identify metaphors within the Arabic Qur’anic text to enhance the understanding of figurative language in translations.
The research contributions include a detailed comparison of the proposed models against existing baselines and an error analysis of their performance. The results demonstrate that the T5 model achieves a high degree of accuracy in translating complex Qur’anic verses, with notable success in preserving metaphors and nuanced meanings. The metaphor detection model further highlights the unique stylistic elements of the Qur’an while aiding in semantic interpretation.
This work contributes to the fields of machine translation, computational linguistics, and Qur’anic studies by providing novel approaches to handling the linguistic intricacies of Qur’anic text. Future research directions can expand the dataset, improve metaphor detection techniques, and apply these models to other religious and classical texts.