INTEGRATING ONTOLOGY AND COGNITIVE LINGUISTICS FOR NATURAL LANGUAGE PRODUCTION AND UNDERSTANDING

Ontologies have been designed to capture the semantic knowledge of a domain in a machine understandable form. Current standards for managing ontologies like OWL are lacking in linguistic grounding, and are not able to achieve a clear link with natural language. Bridging this gap, unskilled users could be able to infer the information described in the ontology and it would be possible either producing or parsing utterances about the represented domain automatically. Moreover, as in the case of enterprises, it could be very useful to extract information from external documental corpora that are related to the same domain. Many attempts have been made with the aim to create a natural language interface to ontology but very few of them use grammars during interaction; such interfaces are focused only on verbalizing information contained in the ontology, while it is often necessary to give exhaustive answers to the users’s queries by retrieving data outside of the knowledge base. The work presented in this thesis has been inspired by theories in the field of Cognitive Linguistics, and in particular by the Construction Grammar, to create a grammar-based tool for ontology verbalization that combines OWL ontologies with Fluid Construction Grammar (FCG). Currently, FCG is the only computational implementation of Construction Grammar that performs both production, and parsing using the same set of constructions. The main idea is to compute both lexical and grammatical constructions (the meaning-form couples) in the FCG from the ontology. To achieve this goal, a suitable set of rules based on linguistic typology have been defined to infer semantics and syntax from the RDF triples inside the OWL structure, while combining them as the poles of constructions in the FCG. To allow verbalization from external documents related to the ontology, synonymous constructions have been defined, which are matched to free text. Computing all possible syntactic forms for the same meaning was achieved using linguistic rules. The information retrieval procedure outlined above allows semantic annotation of the text as a side effect. A system for automatic generation of contents for Semantic MediaWiki from standard Wikipedia pages has been implemented in this respect. The combination of OWL with FCG and the integration of external documents in the ontology are the core of the system whose theoretical background, modeling, design, and evaluation with respect to other contributes in this research field form the main subjects of this thesis.