Modeling Legislation Using Natural Language Processing

$EVWUDFW This paper describes the possibilities of the translation of legislation, which is written in natural language, into a formal language, i.e. UML/OCL. The tool OPAL (Object-oriented Parsing and Analysis of Legislation) is developed to support the automatic modelling of legislation with the use of appropriate NLP techniques. The aim is not to perform this modelling in a batch fashion from legislation to final model, but interactively in dialogue with the knowledge engineer. The main components of OPAL are a parser (based on a chart-parser algorithm) and a model generator. A special component called modelling interface is added to OPAL to give the knowledge engineer the possibility to keep track of the modelling process and to make adjustments to the final model. 1Introduction The application of legislation is equivalent to judging the consequences of a given situation (the case) within the context of the law. Two things are needed to judge these consequences, the law and a description of the case. The ideal situation would be that a machine performs the judgement of the juridical consequences. Before this is possible, the machine at least must be able to read and interpret the law and the case. If we take a closer look at the law, we see that it is written in natural language, namely Dutch in this research. There are some aspects, in addition to arbitrary texts in natural language, that make legal texts a bit more formal. In (Dutch) legislation for example fixed language constructs are used extensively. Legislation is also organised in a hierarchical structure. The fact that legislation is written in natural language causes some difficulties. The most important problem is that natural language is ambiguous, which entails that an expression in natural language can have multiple meanings. This problem can arise at word level, for example 'bank', and at phrase or sentence level, for example 'the man sees the woman with the telescope'. Another problem that arises is that natural language, and legislation as well, contains vague and unclear notions like 'almost' and 'for the most part'. So before law enforcement can be automated, legislation has to be translated into a language that does not have aforementioned problems and can be read by a computer, for example a specification language. An extra benefit from the translation of legislation into a specification language is that ambiguous constructs in the law can be detected at an early …