Towards a logic-based approach to legal case-based reasoning

The thesis develops formal, logic-based models of legal case-based reasoning (CBR). Broadly, CBR is a reasoning paradigm in which decisions or conclusions are drawn by analogy with previously decided cases, collected in a case base. It constitutes a central form of legal reasoning in common law systems, and its formalization is increasingly significant for the theoretical foundations of AI and Law. Although several formal frameworks for legal CBR have been proposed, they typically rest on strong assumptions that limit their applicability to real legal practice. These include: A1) that legal cases can be represented as sets of case features (e.g. factors) favoring one of the parties, where the directionality of the features is known a priori; A2) that factual relevance alone determines whether a precedent constrains a decision; and A3) that no conflicts among precedents arise in the decision of new cases. This thesis systematically explores the consequences of abandoning these assumptions. First, it investigates whether it is possible to identify ex post which features of cases behave as factors and determine their directionality. Second, it examines which precedents are binding for a given case, considering both factual relevance and the jurisdictional relations among the deciding courts. Finally, it acknowledges the possibility of conflicts among binding precedents and analyses how such conflicts can be resolved in a legally grounded way. The proposed account is formalized within the Modal Logic of Binary Input Classifiers (BCL) along with suitable extensions introduced in this work.