Corpus Guided Sense Cluster Analysis: a methodology for ontology development (with examples from the spatial domain)

The paper explores problems in ontology construction that arise due to the complex mapping between language and meaning. A new methodology is proposed, which combines a definitional approach using formal logic, with corpus-based statistical analysis of the use of terminology in natural language text. Underlying the approach is a semantic theory in which the notion of sense cluster plays a central role. Rather than having a single precise definition, the referent of a conceptual term is taken to be a sense cluster, modelled by a probability distribution over a set of precise definitions. This style of semantic specification pays heed to insights into the nature of language coming from philosophers such as Wittgenstein and his followers; but it also provides a framework supporting rigorous formal ontology development, which is often regarded as incompatible with the view of language suggested by Wittgenstein (in his later works). Although the methodology is quite general, this paper will mainly draw its examples from the domain of spatial properties and relations, and will examine the complex correspondence between the spatial vocabulary of natural language and logically defined geometrical constraints.