Extended conceptual retrieval
Radboud Winkels, Alexander Boer, D.J.B. Bosscher · UvA-DARE (University of Amsterdam) · 2000
Access to electronic sources - on the internet and on CD-ROMs - is typically handled in the same way as information retrieval from documents in general (e.g. (1)). In traditional text retrieval, documents are indexed by the literal strings that appear in them, and searching is based on string matching techniques, typically enhanced with a query language based on boolean and proximity connectives to combine search strings. When the strings in the search expression match the strings in the document, the document is assumed to be relevant and is returned. For problems, key word matching is limiting, irrespective of the query language, because it requires the user to know the right phrases in the law to be able to use it successfully. Moreover, the quantity and quality of the search result leaves much to be desired. In conceptual retrieval (CR), the idea is that documents are indexed by the concepts denoted by strings rather than the strings themselves. In this case, the retrieval mechanism is intended to return those documents that refer to the concepts in the search expression (cf. (2)). If these concepts are somehow related to each other, typically in type hierarchies, documents can be returned that refer to more abstract concepts than the one a user asks for (e.g. (3)). Although this idea successfully extends retrieval to implied concepts, it is not at all a real problem solving tool; It returns information conceptually related to the question, while the user is actually interested in information related to the answer. The alternative approach to information systems explicitly aims to assist the user in solving legal problems. The layman's conceptual view on law in the continental tradition is quite straightforward; It divides knowledge into two orthogonal types: rules and cases. The pivotal problem in this view is to assess whether a case complies with (or deviates from) the rules. Rules are typically interpreted as deontic sentences and cases as situations or events potentially regulated by rules. This perspective explains i.a. the attention in AI & Law for modal deontic logics, and, more recently, generic and reusable task models and ontologies for assessment (4,5). The latter approach, coined Legal Information Serving by us (6), aims to provide KADS-like blueprints for the development of advanced information retrieval systems. In the CLIME project, of which this paper is a result, we develop a Legal Information Server (LIS) using such a blueprint (7).