Building a stronger case: combining evidence and law in scenario-based Bayesian networks

Abstract

Existing approaches to modelling legal cases in Bayesian networks focus either on correctly representing an empirical probabilistic model of evidence traces, or on modeling alternative scenarios that can explain what happened in a case. However, neither approach legally interprets, or qualifies, aspects of a scenario as a normative legal fact. Hence, the fact that a Bayesian network representing a scenario assigns a high posterior probability to a certain victim having been killed by a certain suspect, does not imply that that suspect is guilty of murder in the legal sense, because the events in the scenario cannot be qualified as legal facts. This paper proposes an architecture for concrete legal fact idioms that qualify events in a narrative Bayesian network. This bridges the gap between the real world and the normative legal world through so-called counts-as rules. By modeling the legal facts explicitly in the Bayesian network, we can show whether a narrative completes one or more legal fact idioms. This is demonstrated using a case study. The proposed architecture may help judges and lawyers decide on which narratives they should investigate further and which narratives are stronger than others with regard to both the evidence and the legal facts.