Alessandro Lenci

Abstract

This talk will focus on issues concerning the development of ontologies for computational lexical semantics. The top-ontology of semantic types developed within the SIMPLE-project will be illustrated, as an example of a general model for the construction of large-scale, plurilingual, ontology-based lexicons.

In SIMPLE the backbone for lexical representation is given by an ontology of semantic types, each corresponding to a template of structured information, which provides the interpretation of the types themselves. A semantic unit representing a specific word sense is defined as a particular instantiation of a given template, and is thereby assigned a type in the ontology. The sorts of information associated with types are very rich and variegate, spanning from purely lexical information to general world-knowledge. Despite this heterogeneity, the SIMPLE model seems to provide a rich and versatile tool for semantic encoding, thus presenting itself as an authentic general-purpose lexical resource, different aspects of which can be exploited by NLP systems.

The SIMPLE framework also represents an interesting vantage point to evaluate important issues raised by the crucial role of ontologies in computational semantics and language engineering in general: (i) the trade-off between theoretical inquiry in lexical and formal semantics and computational and applicative issues; (ii.) the balance between general-purpose resources and their necessary customization for domain-oriented applications; (iii.) the necessity of establishing solid criteria to carve out the ontology of semantic types; (iv.) the problem of scalability, and dynamic update of ontologies to meet different applicative needs. Some of these points will be addressed, by showing how a multidimensional, modular and open framework for the development of type systems may reduce the gap between general multi-purpose research, and strictly task-oriented application of lexical semantics.