Natural language query system

Joris Pelemans

The students of the master in artificial intelligence study, option speech and language technology, at the KULeuven, Belgium, have to do a 3 month internship in a company. This presentation will discuss my work in the company "Language and Computing".

The goal of the internship was to develop a natural language querying system to retrieve instance information from a drug database. The database is integrated in the company ontology by means of a module which maps data from the database onto ontology concepts. I was asked to build a user interface that allows the user to extract relevant concepts by means of natural language, using this mapping information. Knowing the size of this project, and its limited duration, the task was limited to developing only a prototype system.

The drug database contains information about a number of drugs; more specific it contains a number of attributes attached to these drugs, e.g. the company that makes the drug, the location of this company, the kind of package the drug is in, the size of this package, the way the drug is ingested, ...

A typed-in query like "drugs made in a company based in Spain." would then have to retrieve the data meeting these constraints.

The system was built using the principles of Systemic Functional Grammar (SFG), both in syntax as in semantics. SFG uses system networks to build computational grammars and is based on the ideas of Michael Halliday. Halliday's grammar used a constituency-based formalism, while the grammar used in L&C was transformed into a dependency formalism by Mick O'Donnell.