Presenter: Jeremy Ellman
This paper addresses the research question is whether, and how a machine can identify people’s intention and justify a purpose of that intention from natural language text alone.
We describes an innovative computer model, ELVIE (Extended Legal Violence Inference Engine), that can analyse people’s intention from legal case report texts in court cases describing violent assaults between individuals. ELVIE tries to identify whether protagonists intended to cause harm or were acting in self-defence. Participants’ behaviours are found from the text using natural language analysis and processing. These are then interpreted as combinations of three semantic roles: action, item, and part of body. Using these in context, the degree of harm of each action can be determined. In differentiating the degrees of harm, we propose a method for justifying whether self-defence motivated the action. This is based on the legal concept of proportionality.
Identifying an intention to offend is an aspect of artificial intelligence and law (AIL). In AIL, much research has tried to produce a supportive model or application that may reduce the legal decision-making process time. ELVIE was designed to support information regarding “Who intends to harm whom, by what action, with what item, in what part of the body, and with the purpose of self-defence”. Such information could be useful as part of the process of decision making by lawyers or even by court judges
ELVIE uses the text of Court of Appeal case reports to analyse the intent of an offence and whether the purpose behind that intention could be self-defence. The analysis uses rule-based inference techniques to identify whether an action in a real assault case report could be an intentional offence. If it were, a method is proposed to determine the purpose of self-defence for each action. The result of the analysis is in the form of “Who intends to harm whom, by what action, with what item, in what part of the body, and whether it was with the purpose of self-defence”. The method is based on the concept of proportionality of offence, where an offence is a physical act by one protagonist against another and proportionality is the relative degree of force used by participants.
The performance of the system was evaluated using 40 criminal cases collected from the UK Court of Appeals and tested against their verdicts. The experiment achieved 80% accuracy on both processes of identifying intention of offence and justifying the purpose of self-defence.
