Quranic text translation in the age of AI: A tripartite framework of linguistic, theological, and ethical dimensions

Abstract

This study evaluates the theological integrity, linguistic accuracy, and ethical implications of AI-generated English translations of ten verses from the Quran, utilising Yusuf Ali’s authoritative translation (1934) as a benchmark for comparison. This study employs a three-pronged approach—linguistic, theological, and ethical—to analyse the outputs of four AI models (Qwen2.5-Max, QuillBot, DeepSeek, ChatGPT). It aims to assess their ability to uphold theological accuracy, navigate linguistic complexities, and avoid cultural biases. The findings reveal consistent deficiencies: AI translations emphasise fluency at the expense of theological accuracy, especially in allegorical (mutashabihat) verses, such as Q. 3:7, and introduce sectarian or gender biases in contentious verses, like Q. 4:34. A thorough examination uncovers significant departures from traditional understandings, even though the average lexical similarity (66.8) suggested by BLEU scores remains moderate. The average BLEU score of 66.8 across the ten verses highlights a moderate level of lexical similarity while also indicating significant doctrinal differences. The study culminates in a recommendation for interdisciplinary collaboration, aiming to integrate AI systems with Islamic hermeneutics (tafsir) and ethical principles.