Partial DRS, an Intermediate Representation, to Generate SPARQL Queries from NL Questions

Abstract

This paper presents a tool to build an intermediate representation, Partial DRS, a Language-Independent Partial Discourse Representation Structure (DRS), that can be used to generate SPARQL queries from Natural Language (NL) Questions. The Partial DRS builder consists of two main steps: creating a Universal Dependencies (UD) parse and transforming the UD parse using a set of rules into a graph with discourse referents as edges and syntactic roles as arcs labels. The Partial DRS of a natural language sentence can be interpreted within a target application domain modelled as a domain ontology, enabling the semantic interpretation of the syntactic roles represented by the graph arc labels. The domain ontology uses the annotations on Object and Data Properties, and Classes to represent natural language and domain-dependent information. These annotations include syntactic role preferences and vocabulary. An evaluation of the proposed system is presented, showing very promising results in the specific DBLP domain.