Abstract

. We present in this paper a method to insert new concepts in an existing information extraction system based on a conceptual graph architecture. We use verb patterns as conceptual sub-graphs to characterize unknown terms in sentences. The objective is to help an information extraction system to cope with unknown expressions contained in a specific semantic context. 1 Introduction Internet pushes today the electronic document processing in a new era, where efficient methods to extract information from large collections are more than ever needed [2]. Today, statistical methods for text processing are fast and reliable but yield too often absurd results as no intelligence really controls the final output. Therefore ontologies are used to enlarge queries to words that are similar along semantic relations. One of the problems raised by this kind of approach is the creation and maintenance of these bases of concepts. Several projects have started to provide ready to use electronic ontolo...

Citations