01 / IlùBench

Does a model explain an Igbo proverb differently in Igbo?

Ask a model to explain a proverb in English and it usually answers as an outsider: gloss, literal translation, a comparison to an English saying. Ask the same model the same question in Igbo and it often answers from inside the culture, reasoning from other proverbs and shifting into the register the form actually carries. IlùBench calls that cultural register switching and measures it: two fresh, independent API calls per probe, no system prompt, provider defaults. The interesting result is not which language comes back. It is whether the two answers are the same explanation.