Abstract:

In our talk we shall look into semantic relations and give suggestions to what role they should play in OntoQuery, both theoretically, in relation to the ontology, as well as in relation to the elaboration of a prototype.

During the last decade, attempts have been made to exploit among others the Princetown WordNet Database in Information Retrieval, applying in particular the synonymy relation represented via the so-called synsets (a synset is a set of words with the same part-of-speech that can be interchanged in a certain context). Enriching queries with semantically related terms as well as comparing queries and documents via conceptual distance measures are two of the aspects that have been addressed here. In EuroWordNet this has been brought a little further (cf. Gonzalo et al. 1998), partly by extending the idea of a synset so that nouns and verbs do not form separate networks. This means that synonymy and hyponymy relations can be further exploited as cross-part-of-speech relations. Furthermore, the so-called topic relation, where words can be related via domain labels, has increased the utility of the network with respect to Information Retrieval.

In our talk we propose to be a little more ambitious with respect to the application of semantic relations in Information Retrieval. As an introduction to this, we shall first compare how semantic relations are used within two different fields, namely lexical semantics and terminology. For each field one or two representatives are used as source; for the field of lexical semantics two large-scale, state-of-the-art lexicon projects in Europe are taken as representatives, namely EuroWordNet and SIMPLE. As regards the field of terminology, the set of commonly used relations at the Copenhagen Business School as well as the relations applied by Nuopponen will be discussed. Finally, we relate these relations to those defined within Formal Ontology by Nilsson (1999).

Secondly, we shall present a proposal of how semantic relations can be grouped into coherent classes and ordered hierarchically. The aim of this ordering of the relations is again to facilitate search. The relations are to be used in the semantic representation of complex NPs from the texts in the database, which form descriptions of the contents of the texts. Likewise, in the analysis of NL queries, the relations will be part of the semantic representation. The hierarchy of relations can be used in a more refined conceptual distance measure, which can be applied when comparing queries with the descriptions in the database. In relation to this, a weighting of the relations will furthermore be proposed with respect to their plausible importance in the area of IR. For instance synonymy and hyponymy will most presumably receive a high weighting in the search calculations, whereas for instance role relations are most presumable less significant for search and should therefore have less weight in the calculations.