A Comparative Analysis of AI-Generated and Human Translations in Urdu Across Selected Genres

Abstract

This study compares the performance of artificial intelligence (AI) and human translators in translating English texts into Urdu across three genres: literary, scientific, and legal. Selected passages from each genre, sourced from a recently published work with its available Urdu translations by human translators, are compared with the translations generated by ChatGPT (OpenAI, 2023) and Snapchat’s My AI (Snap Inc., 2023). The translations are evaluated using Walker’s (2016) quality assessment scale, a 1–4 ranking system adapted from Aziz et al. (2012), to assess fluency, adequacy, cohesion, and pragmatics. Human translators excel in literary and legal texts, accurately handling idioms, tone, metaphors, and legal terms with cultural understanding for clear, relevant Urdu translations. ChatGPT performs better in scientific texts with precise, consistent translations, while Snapchat’s My AI struggles across all genres, often mistranslating figurative language and legal terms, producing unnatural Urdu phrases. The findings highlight the benefits of hybrid approaches: AI tools can support initial drafts, especially for scientific texts, while human translators refine outputs in culturally complex genres. This study offers insights for educators, translators, policymakers, and developers, promoting AI-human collaboration and advocating for AI systems trained on Urdu-specific data to enhance cultural and legal accuracy, thereby improving efficiency and translation quality across genres.

Keywords: Artificial intelligence, human translation, genres, translation, translation evaluation, comparison