ALGORITHMIC CATECHISMS:

Abstract

This paper examines how generative AI recontextualises Namwanga oral proverbs within algorithmically produced religious and moral discourse. Namwanga, spoken in Isoka and Nakonde Districts, has a rich tradition of polyvalent proverbs that encode ecological knowledge, social satire, and indirect critique. Using prompt-response analysis across three LLMs (GPT-4, Claude 3, and Gemini Advanced), the paper shows that when queried to "explain Namwanga proverbs about obedience," models consistently strip proverbs of their ironic or situational layers, re-embedding them as conservative, authoritarian tropes (e.g., "Akasuba talifya mwibula" – The sun never dies in the morning, flattened from caution against premature judgment into uncritical elder obedience). The researcher terms this algorithmic catechism: the production of univocal, norm-reinforcing meanings from inherently multivocal oral genres. The paper further traces how AI-generated Namwanga religious texts (used by two Pentecostal congregations in Nakonde) displace elder-led hermeneutic traditions, in which proverb interpretation requires knowledge of the speaker, occasion, and audience. Contrastive analysis with human-translated proverbs shows that AI outputs increase internal consistency but decrease sociolinguistic register variation by 73%. The study concludes that the algorithmic production of meaning in minoritised languages does not simply replicate bias but actively reorders oral epistemology. For Zambian language preservation, this demands not more data but critical hermeneutic annotation.