First steps in building a model for the retrieval of court decisions

Abstract

The MOSAIC project investigates a retrieval model for court decisions based on structured and unstructured (natural language) information in legal cases. This paper focuses on how relevant information in court decisions can function as a key for retrieval and on the automated construction of case representations. Techniques of automated concept learning and rhetorical structure identification are among the most promising ones. # Automated retrieval from large document collections was one of the earliest applications of computer science to law. In 1961 the US Air Force contracted with the University of Pittsburgh for building a full text retrieval system for legal documents. As a result, finding legal information through electronics (FLITE) system saw its first productive use in 1964. A few years later, the US Department of Justice developed the JURIS system, which has been in use since 1971. More recently, commercial systems such as LEXIS-NEXIS and Westlaw, which offer interactive retrieval through terminals at the customer's office, have gained widespread acceptance in circles of legal professionals. In the European Union, there are many databases of court decisions, most of which can be consulted by any citizen via the World Wide Web. Present-day retrieval systems allow users to express their query with a set of key terms. In some systems, key terms can be used in combination with Boolean operators. The result of such a search is a list of documents. These are usually sorted by 'relevance', which most of the time is simply computed as a function of the frequency of occurrence of the search terms in the documents. Documents are returned as being relevant if they contain the query terms or if keywords that were manually assigned to them match the query terms. Thus, current commercial retrieval systems for searching court decisions either rely on a manually built index of cases or on a full-text search. In manual indexing, texts or textual passages are linked to the concepts of a pre-defined thesaurus or classification scheme; case texts are linked by means of citation links or common descriptors. The disadvantages of manual indexing are the huge cost}a problem that will only aggravate with the current growth of the number of cases}and the large amount of inconsistencies between different indexers. In a full-text search, each term (except for stopwords z ) acts as a search key. The major disadvantage of a full-text search is that its retrieval results are quite unreliable, since the occurrence of a particular key word or key phrase in a text is no guarantee that the text is a relevant output for the search request. This is a fundamental and rather problematic issue, which bears on most retrieval tasks but is especially important when retrieving legal cases. You cannot search for meaning in a case solely by taking into account word occurrences. Another disadvantage of full-text searches is that they often overwhelm users with documents that are not or only marginally relevant.z Stopwords are small words}like articl...

Applications of Rhetorical Structure Theory

“…In one project, documents are partly analyzed using RST, in an attempt to capture more information from texts than traditional keyword-based indexing allows (Haouam and Marir, 2003;Marir and Haouam, 2002). Moens and de Busser (2002) propose a system for creating legal summaries, based partly on the identification of rhetorical structure in court decisions. Shinmori et al (2002) extract the most important claim in Japanese patent applications by analyzing the rhetorical structure of the patent description.…”

Combining relations for information extraction from free text

“…The word 'additionally' connects the event in the second sentence to the entity 'ABC Co.' in the first sentence. Fourth, Moens and De Busser (2002) reported that discourse segments tend to be in a fixed order for structured texts such as court decisions or news. Hence, analysis of discourse order may reduce the variability of possible relations between entities.…”

Extractive summarisation of legal texts

“…In order to perform summarisation, it is necessary to look at other features which may be characteristic of texts in general and legal texts in particular. These can then serve to build a model for the creation of legal summaries (Moens and Busser, 2002). In our project, we are developing an automatic summarisation system based on the approach of Teufel and Moens (2002;.…”