Artificial Intelligence Meets Human Expertise: A Case Study in Translating Violence Against Women Reports

This study investigates the comparative effectiveness of human translation (HT) and machine translation (MT) in translating United Nations (UN) documents on violence against women (VAW) from English to Arabic. The research uses Na Pham’s (2005) theoretical model to evaluate translations based on dynamic equivalence, cohesion, and cultural adaptability. Experiments involving terms and paragraphs analyzed semantic accuracy, cultural fidelity, and contextual coherence in translations by UN professionals and AI models, including ChatGPT-3.5 and GPT-4. Findings reveal that while HT excels in cultural nuance and textual cohesion, MT demonstrates significant improvements in term-level accuracy but struggles with contextual consistency and cultural adaptation. However, it often fails to maintain consistency between isolated term translations and their contextual use, and frequently violates Arabic grammatical norms such as adjective-noun agreement and gendered pronouns. The study emphasises the necessity of integrating advanced linguistic and cultural frameworks into AI systems to enhance translation reliability for sensitive topics. Recommendations are proposed for refining MT tools to support global advocacy efforts stressing the need for culturally aware training systems and integration of language-specific grammatical frameworks to improve semantic fidelity and contextual reliability.