Using linguistic information and machine learning techniques to identify entities from juridical documents

Abstract

Information extraction from legal documents is an important and open problem. A mixed approach, using linguistic information and machine learning techniques, is described in this paper. In this approach, top-level legal concepts are identified and used for document classification using Support Vector Machines. Named entities, such as, locations, organizations, dates, and document references, are identified using semantic information from the output of a natural language parser. This information, legal concepts and named entities, may be used to populate a simple ontology, allowing the enrichment of documents and the creation of high-level legal information retrieval systems. The proposed methodology was applied to a corpus of legal documents - from the EUR-Lex site - and it was evaluated. The obtained results were quite good and indicate this may be a promising approach to the legal information extraction problem. © 2010 Springer-Verlag Berlin Heidelberg.