Date of Award

10-19-2015

Degree Type

Thesis

Degree Name

M.Sc.

Department

Computer Science

First Advisor

Frost, Richard

Keywords

dbpedia, language, natural, query, semantic, triplestore

Rights

CC BY-NC-ND 4.0

Abstract

DBpedia is a triple-based binary-relational database which contains 3 billion, and counting, facts derived from Wikipedia. Ideally, individuals should be able to access semantic web triple-store data through natural-language queries. Several attempts have been made to create natural-language (NL) query interfaces to DBpedia. However, no one has yet built a wide-coverage natural-language query processor for DBpedia. DBpedia does not currently encode contextual data representing the time, location or other properties of binary-relationships. This means that NL queries cannot contain prepositional phrases such as the phrase "in 2004" in the query: "what film was directed by Clint Eastwood in 2004". Existing NL query interfaces to DBpedia cannot handle prepositional phrases; they are also unable to be extended to do so when used with triple-stores other than DBpedia, which can accommodate contextual data. In this thesis, we investigate an alternative approach to querying DBpedia in which NL queries are treated as expressions of the lambda calculus which are evaluated directly with respect to the triple-store using a compositional and extensible denotational semantics of English.

Share

COinS