Nigerian Academic Writing Corpus: Pre-AI Benchmark for AI-Generated Text Detection

A curated corpus of pre-AI era (2005–2022) Nigerian academic writing paired with AI-generated equivalents for training and evaluating AI-generated text detectors calibrated to Nigerian English.

The dataset contains two components: (1) 4,239 authentic academic texts extracted from Covenant University and University of Ibadan open-access repositories, spanning theses, dissertations, journal articles, and student projects across multiple disciplines; and (2) 6,000 AI-generated academic essays produced by six major large language models (GPT-4o-mini, GPT-4o, Claude Sonnet, Gemini 2.5 Flash, Grok 3 Mini, and DeepSeek Chat) using topics drawn from the human corpus.

All human texts predate the launch of ChatGPT (November 2022), ensuring zero AI contamination in the baseline. The paired structure enables controlled experiments for binary classification (human vs AI), multi-class source attribution (identifying which LLM generated a text), and cross-dialectal evaluation of AI detection systems on African English.