Title: Large Language Model-Based Chatbots In Selected African Linguistic Contexts
Authors: Chukwuma Livinus Ndububa
Volume: 10
Issue: 6
Pages: 374-390
Publication Date: 2026/06/28
Abstract:
In the digital era, large language model-based chatbots increasingly shape knowledge and communication in many languages. Programmers adapt chatbots to diverse contexts, including different languages and purposes. However, given that chatbots' efficiency is shaped by the scope and quality of pretraining data, there are concerns about underperformance in specific linguistic contexts. Thus, this study investigated how these chatbots accommodate, misaccommodate, or show non-accommodation in widely spoken African languages, and the implications for relational trust, sociocultural representation, communicative efficiency, and intercultural intelligibility. Based on Giles' (1973) Communication Accommodation Theory (CAT) and a mixed-methods approach within a comparative research design, specific linguistic outputs of six prominent LLM-based chatbots in 20 widely spoken African languages were purposively analysed. The study found that ChatGPT demonstrates the highest level of convergence (successful accommodation) across the investigated areas, while Perplexity shows the weakest performance, frequently exhibiting misaccommodation through wrong-language outputs, partial coherence, and hallucinated responses; Grok also displays notable misaccommodation and, in several instances, explicitly indicates inability, reflecting cases of non-accommodation. It was further found that Kanuri is the least effectively handled among the sampled languages, with frequent misaccommodation and occasional non-accommodation. These results have implications for reliability, representation, efficiency, and intelligibility in African linguistic contexts. Thus, the researcher suggests that LLM-based chatbot developers improve training data for underrepresented African languages, while users should remain cautious of these limitations to avoid uncritical reliance on inaccurate outputs. This study contributes to research on LLM-based chatbots and the representation of African languages in global digital communication systems. While the study achieved its objectives, there is room for further studies to investigate other linguistic areas such as pragmatics, idioms, and discourse structures.