Abstract
Legal AI systems typically assume that law has a canonical written form from which rules can be extracted. This paper calls that premise the Written Law Assumption (WLA) and argues that it fails for orally-transmitted customary law, where legal validity derives from social practice, communal recognition, locality, and proof rather than from a stable text. To represent this class of law, we introduce the Contested Norm Structure (CNS), a formal knowledge object that models a customary rule as a belief distribution over possible norm-states. A CNS combines witness testimony, expert opinion, and recognised manuscripts using Dempster-Shafer evidence theory, while preserving source authority, locality, conflict, and uncertainty. The paper establishes four formal results. First, representing a customary rule as a single ground-truth fact necessarily loses information. Second, Dempster-Shafer combination matches the Nigerian Evidence Act's structure for proving custom. Third, judicial notice corresponds to belief convergence across independent cases, formalising Romaine v. Romaine's requirement of frequent use in several cases. Fourth, the repugnancy doctrine in Section 18(3) of the Evidence Act 2011 is a post-combination validity filter, not a pre-filter on evidence. Grounded in Nigerian legal doctrine and leading customary-law cases, the paper offers a constructive representation theory for legal systems whose rules are authoritative without being textually canonical.
Keywords: customary law, knowledge representation, Dempster-Shafer theory, legal AI, African law, legal pluralism, evidence theory, knowledge graphs, judicial notice, oral law