Nigerian slang is often difficult to translate because it tends to be informal, idiomatic, and culturally salient. This study examines five artificial intelligence (AI) models with respect to human translators in their ability to translate the quality of Nigerian slang expressions, adapting three theoretical frameworks: Relevance, Equivalence, and Polysystem. 31 common words or phrases of Nigerian slang were translated by ChatGPT, Grok, Copilot, Gemini, DeepSeek, and professional human translators. To rate the translated words or phrases, bilingual experts rated each successful translation of the slang on a 5-point Likert scale across three theories rated on pragmatic intent, semantic meaning, and cultural appropriateness. Human translations rated higher consistently across all three dimensions of pragmatic intent, cultural fit, and semantic meaning. Human translations were able to best preserve the meaning, context, and relevant cultural fit of the Nigerian slang words or phrases being translated. From highest to lowest relevance, of AI models were Grok, ChatGPT, Gemini, Copilot, and then DeepSeek. Statistically significant differences were confirmed between human and AI model outputs, with greater differences between human translation outputs compared to AI model outputs in pragmatic and cultural dimensions. This study shows that AI models are rapidly improving but are currently incapable of fully considering the cultural and contextual dimensions of translating Nigerian slang expressions, pointing to the necessity of having human agents involved in informal language translation projects. To improve future translation quality, AI model outputs can be improved through a hybrid human-AI workflow and training datasets enriched with culturally relevant material.
