Generating normative rules with ILP techniques

This paper describes the use of inductive logic programming techniques in a legislative drafting environment. The drafter supplies a set of positive and negative situations. These situations are transformed into normative rules, i.e. rules that oblige, forbid or permit behaviour. Drafters may impose four diierent types of requirements on the rule generation process, in order to obtain alternative legislative architectures. Legislative drafting is the process in which the government or other legislative bodies decide to formulate new laws, and teams of drafters deene them in terms of (normative) goals, and subsequently codify their goals as law texts 1. Drafting regulations may vary from drafting a contract, a simple regulation for the local library, to legislative drafting under the responsibility of a particular Ministry. In particular for the latter version of this task, the stakes are high, and (semi) automated tools may be welcomed. However , legislative drafting is a complex and empirically not well studied task. 2 Moreover, methodological support is almost absent. There are a few operative systems that ooer automated support for legislative drafting. The systems that support information retrieval and editing tools are systems that provide textual support for drafting new legislation, for instance the LEDA system Voermans, 1995]. Textual support for legislative drafting can be ooered by relatively simple systems (databases, hypertext systems, ooce systems), and concerns information retrieval of relevant texts or electronic communication between drafters. Other legal knowledge-based systems ooer consistency checking of regulations Although textual support provides legislative drafters with intelligent editors that embed the legislative format, and information retrieval provides the relevant background infor-1 Legislative theory on drafting procedures describes general requirements to the drafting process and the composition of new regulations 2 Other terms are: design or modelling tasks, see for instance Breuker & van deVelde, 1994]. mation, drafters still spend large amounts of time codifying their ideas into rules. The approach described in this paper proposes a structured approach for codiication: First, one models the domain that is to be regulated in terms of a (simplistic) world ontology. A world description consists of agents, objects and actions. 3 For instance, for regulating traac behaviour one has to deene types of agents (e.g. cars, trams, cyclists), objects (e.g. road, traac light), and actions (e.g. overtake left, overtake right). The types are organized in abstraction hierarchies, e.g. types of vehicles, where car, tram and bicycle are subtypes of vehicle. Next, a situation …