Native speakers rated outputs of LLMs fine-tuned on Nigerian Pidgin English as understandable and context-aware, though no non-fine-tuned baseline was evaluated¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · checked against its sources 2026-10-20 (automated, 1 of 1 entry) · not reviewed by a person

Subclaims¶

q2 i? Bilingual English–Pidgin speakers consistently rated outputs from the fine-tuned models as understandable and context-aware. The authors read this as improved linguistic alignment relative to non-adapted baselines, but the evaluation compared only three quantisation levels of the fine-tuned model with a GPT-4o reference (which rated highest), with no non-fine-tuned version of the model, so it cannot show that fine-tuning caused the improvement. → Nwogo 2026

Evidence¶

Nwogo 2026¶

Nwogo, E. U., Ihianle, I. K., Machado, P., Bird, J. J., Lotfi, A., Shuaib, A. A., Akinwumi, I. I., & Oluranti, J. (2026). An AI-Based Adaptive Learning Platform for Multilingual and Low-Resource Educational Contexts: A Case Study on Nigeria. arXiv. https://arxiv.org/abs/2608.15738 doi:10.48550/arxiv.2608.15738

q2 · i? · design · r2

Human-centred cultural and linguistic evaluation: a structured survey was distributed to 50 native Nigerian Pidgin speakers via Google Forms, "yielding 18 valid responses", who rated fluency, coherence, relevance and cultural appropriateness on a five-point Likert scale. Responses to 14 prompts came from the fine-tuned model at three quantisation levels (Q4, Q5, Q8) and from OpenAI GPT-4o as a reference; no non-fine-tuned model was rated, so the comparison in the authors' sentence below was not tested.

"Across responses, participants consistently rated outputs from the fine-tuned models as understandable and context- aware, confirming that fine-tuning on Nigerian Pidgin data improves linguistic alignment relative to non-adapted base- lines."

Discussion¶

Related Claims¶

• Higher-bit quantisation (8-bit) better preserves structural coherence and semantic alignment in a Pidgin-fine-tuned LLM, achieving the highest ROUGE-L (0.3120) and BERTScore (0.8590) among the quantised models — related
• Prompting experiments show LLMs perform badly at generating non-US national standard varieties and conflate Nigerian English with Nigerian Pidgin — related
• Automated grading systems may produce systematically inequitable outcomes for international and non-Western students — related
• Current LLMs reproduce linguistic hierarchies by being built on dominant standard English varieties, producing a monolithic English that ignores regional variation — related
• The OpenAI gpt-4o reference model outperformed all quantised models on human-rated coherence, fluency, relevance and cultural appropriateness, serving as an upper benchmark — related
• In the fine-tuning experiments, increasing base model size up to 32B parameters improved performance, and larger LLMs consistently outperformed smaller MLMs — related