Learning IE patterns: a terminology extraction perspective

The large-scale applicability of knowledge-based information access systems such as the ones based on Information Extraction techniques strongly depends on the possibility of automatically acquiring the large amount of knowledge required. However, the basic assumption of the IE paradigm, i.e. that the information need is known in advance, limits inherently its applicability since the resulting IE pattern learning algorithms are not generally conceived for the analysis of large corpora if not driven by a specific information need. Since in the terminological studies the corpora and not the information needs already drive the extraction of the knowledge, they offer many insights and mechanisms to automatically model the knowledge content of a coherent text collection. In this paper, we will present a terminological perspective to the acquisition of IE patterns based on a novel algorithm for estimating the domain relevance of the relations among domain concepts. The algorithm and the representation space will be presented. Before starting the discussion, however, we will describe the overall process of building a domain ontology out from a extensional domain model (i.e. the collected domain corpus). Finally, the results of the application of the algorithm over a large domain corpus will be presented and the resulting ontology is discussed.