Sense-making software for crime investigation: how to combine stories and arguments?1

Crime investigation is a difficult and laborious process, especially in large cases that involve a mass of unstructured evidence of which crime investigators have to make sense. Because of the difficulty of this task and the high costs of mistakes, it is worth investigating how crime investigators could benefit from support tools. Current professional state-of-the-art software for managing and visualizing evidence, such as Analyst's Notebook 2 , has the limitation that it does not allow for expressing the reasons why certain pieces of evidence support or attack a certain hypothesis. Therefore, in this paper a reasoning model for proof-of-concept software is proposed that offers crime investigators the ability to visualize their reasoning about a case. Subsequently, an implementation of this model in the AVERs (Argument Visualization for Evidential Reasoning based on stories) prototype system is presented. Crime investigators can use this sense-making tool to construct possible stories about what happened and link these stories with the available evidence through arguments. Arguably, such software has good chances of being useful while based on sound theories of the reasoning involved in crime investigations. To be usable in practice and to improve the quality of crime investigations, our software design should be based on a natural and rationally well-founded theory of reasoning about evidence. To this end, two well-known accounts of reasoning about evidence, namely the argumentative approach as advocated in the New Evidence Theory (NET) [1] and the story based approach of Anchored Narratives Theory (ANT) [2] will be combined. The former takes its inspiration in Wigmores charting method, in which arguments from evidence to conclusions can be visualized. The latter stresses the importance of constructing stories about what might have happened in a case and of " anchoring " them in commonsense generalizations. In our opinion, both approaches have their shortcomings: NET does not allow for the comparison of stories and ANT does not give an accurate account of how stories can be connected to evidence. Our aim in this paper is to formalize and combine the two theories; this solves some of the problems of the separate theories and clarifies the relations between stories and arguments. The model should incorporate two ways of reasoning with causal knowledge, that is, from cause to effect (causal generalizations, e.g. fire causes smoke) and from effect to cause (evidential generalizations, e.g. smoke is evidence for fire). In our model, the different stories about what …