Feedback Mechanisms for a Natural Language Interface: An Application of the Critic Paradigm

Users of natural language interfaces (NLI) to databases often build their own notions about the linguistic capabilities of the NLI and bear false expectations in mind with respect to the intelligence of the system. An NLI should provide feedback on what it does and what it understands. In addition, user queries and intermediate results have to be analyzed for errors, so that the interface can signal potential problem situations. This paper discusses the application of the critic paradigm for human-computer interaction to natural language interfaces, and presents a computer-based assistant for the NLI-Z39.50, a natural language interface for access to library databases. The search assistant, a so called critic system, spots errors in the natural language input, provides feedback on intermediate results, suggests search terms for query refinement, and aids the users when they do not know how to perform a search successfully or when they encounter common search problems.