THE ALGORITHMIC TRANSLATION OF INDIGENOUS AFRICAN METAPHORS: SEMANTIC LOSS, LINGUISTIC INJUSTICE, AND CULTURAL EROSION IN AI LANGUAGE MODELS

Abstract

The rapid integration of artificial intelligence into translation systems has reshaped global communication, multilingual media circulation, and digital content production in ways that would have been difficult to imagine only a decade ago. Beneath the celebrated speed and efficiency of Large Language Models (LLMs), however, there appears to be a quieter but increasingly significant epistemic problem: the difficulty algorithmic systems encounter when attempting to preserve culturally embedded meanings encoded in indigenous African metaphors. This paper examines how AI-driven translation architectures often flatten, distort, or partially erase the semantic and symbolic complexity of African metaphorical expressions during computational translation processes. Drawing from postcolonial translation theory, cultural linguistics, semiotics, and critical algorithm studies, the study interrogates what may be described as semantic bleaching in AI-mediated language transfer. Through qualitative comparative semantic analysis, selected metaphors from Igbo, Yoruba, Swahili, and Akan linguistic traditions are examined alongside probable AI-generated translations in order to demonstrate how contextual, spiritual, historical, and communal meanings are frequently reduced to literal or Western-oriented approximations. The analysis suggests that algorithmic translation systems are not culturally neutral technologies. Rather, they are shaped by asymmetrical linguistic datasets and enduring epistemic hierarchies that continue to privilege high-resource languages within digital infrastructures. Particular attention is given to African cinema, subtitling practices, screenplay localization, and streaming platforms where semantic loss may gradually weaken narrative authenticity and cultural memory. The paper argues that preserving indigenous African meaning systems requires more than the simple inclusion of additional languages in NLP systems. It calls instead for decolonized AI architectures, culturally grounded training models, and ethically accountable linguistic governance frameworks capable of protecting semantic integrity in an era increasingly mediated by machine translation.