Extracting and Applying Legal Rules from Precedent Cases

Blass, Joseph

doi:https://doi.org/10.21985/n2-775z-xn78

Work

Extracting and Applying Legal Rules from Precedent Cases

Public

Download PDF

Download All Files (.zip)

In the Western Common Law tradition, legal decisions constrain and guide future cases involving the same legal issues. Although the legal academy disagrees about the specific nature of such precedential reasoning, including about the role of analogy in legal reasoning, there is ample evidence from cognitive science that analogical learning and reasoning are ubiquitous in human cognition. This thesis presents a computational approach to performing legal precedential reasoning and argumentation using analogical learning and reasoning, grounded in research in Artificial Intelligence, Cognitive Science, and Law. The thesis presents a dataset of historical Illinois intentional tort cases upon which the rest of the thesis is trained and tested, and an algorithm for supervised analogical learning that is useful in instances where it is hard to discern whether an analogical match is useful. It demonstrates that this algorithm can learn, from across a body of case law, schemas that capture the legal information governing those cases. The thesis then presents three algorithms for legal reasoning and prediction using such learned legal schemas: one that reasons directly by analogy from prior cases and legal schemas to a new case, one that reasons about the analogies drawn from legal schemas to a new case, and one that converts those schemas into logical rules and reasons about the new case using logic. It also presents a legal argumentation system adapted from the rule-reasoning system. The thesis demonstrates that abstract legal information can be captured through a process of analogical learning, that analogical reasoning can be used to resolve common law cases, and that schemas induced through an analogical learning process can be converted into rules useful for rule-based legal reasoning.

Creator