An approach to querying text sources based on extracting and evaluating semantic content, given a formal ontology for the text domain, is described in this paper. Queries take the form of natural language expressions, and the system is primarily intended to retrieve text segments whose semantic content matches the content of noun phrases in the query phrase. The approach is based on a fully automatic generation of descriptors of natural language text. Querying uses the generalization/specialization relations for ranking matches, rather than reformulating queries.
After a short section comparing their approach with other systems, the authors discuss their choice of a feasible ontological representation formalism, in sections 3 and 4. In particular, the descriptors, and their embedding in the ontology by means of a notion of ontological grammar, are introduced. Section 4 addresses the disambiguation of noun phrases, whereas section 5 introduces the notion of semantic roles represented in terms of binary semantic relations. As an application domain for the prototype, the authors use the field of nutrition. Section 6 considers examples of noun phrases from nutrition texts, and their mappings onto descriptors. Finally, sections 7 and 8 describe the query-matching process, and the prototype implementation, respectively. Section 9 concludes the paper, and offers some challenges for the future.
The method was developed in the OntoQuery project, and is described in several other papers by the same authors, profusely cited in the text. This detracts from the readability of the paper. While the method is interesting, the authors provide no well-specified experiments justifying their approach. For example, a comparison with other methods of information retrieval would be beneficial.