An Industry Specific Natural Language Interface Tool

Student: Justin Michael Conaghan

Supervisor: Dr. Marcus Gallagher

Category: Software Engineering Thesis Project

The national electricity market involves large amounts of data of great interest to participants and analysers alike. This information is most useful when made easily accessible and available in an intuitive and easy to use manner. This thesis investigates the technologies involved in the use of natural language as an interface to the market data and the feasibility of developing such a product. The aim is to produce an easy to use program that can supply information quickly and effectively for both experts and novices.

There are many aspects to developing a natural language interface. From the initial recognition of words to deciphering the correct response to return from the database, parsing techniques are very numerous and varying in their approach and effectiveness. This thesis examines briefly every aspect of constructing a natural language interface for the national electricity market, from human computer interaction to statistical analysis of language.

One of the major difficulties in developing a natural language interface is in removing ambiguity from an input. Inferring the intent of a users query is one of the prime areas of concern. In recent times statistical approaches to many aspects of processing natural language have come to the fore. An approach to the inference engine (disambiguation of queries) of the interface using Bayesian networks is investigated and experimented with and produces some promising preliminary results.

Poster Presentation (PDF)

Thesis Document (PDF)