Interactive question answering and constraint relaxation in spoken dialogue systems

Abstract

We explore the relationship between question answering and constraint relaxation in spoken dialog systems. We develop dialogue strategies for selecting and presenting information succinctly. In particular, we describe methods for dealing with the results of database queries in informationseeking dialogs. Our goal is to structure the dialogue in such a way that the user is neither overwhelmed with information nor left uncertain as to how to refine the query further. We present evaluation results obtained from a user study involving 20 subjects in a restaurant selection task.

Interactive Question Answering

“…For example, a partial name detector maps a partial name to its full variant and the edit region detection algorithm is responsible for tracking and removing hesitations, repetitions and repair regions. In the CHAT system (Varges, Weng and Pon-Barry, 2007) several preprocessing components such as edit region detection algorithm, a partial name identifier, a domain-specific shallow semantic parser and a domain-general deep structural parser are included. These modules help the system to deal with irregularities in the spoken input and possible mistakes of the speech recognizer.…”

Using domain knowledge for fostering the collaborative ability of a web dialogue system

“…We have defined a set of relaxation rules (similar to the ones proposed in [12]) that are applied over the domain concepts represented in the ontology.…”

Correlation-based query relaxation for example-based dialog modeling

“…If constraints on the query are too strict, it may fail to produce an answer. Constraints on queries that fail to produce an answer (failed queries) are relaxed to satisfy more tuples so that queries can return relevant answers [1][2][3][4].…”

“…RELATED WORK Some approaches to query relaxation have been proposed which relax constraints on failed queries by selecting a subset of query constraints. For example, a query relaxation method using domain-independent strategies was used to relax failed queries to find more contents (e.g., restaurants and MP3 songs) for information presentation in a spoken dialog system [1]. In addition, automated query relaxation algorithm was proposed that uses domain knowledge extracted from a small dataset to generate successful queries that are highly similar to the failed query [2].…”

“…If constraints on the query are too strict, it may fail to produce an answer. Constraints on queries that fail to produce an answer (failed queries) are relaxed to satisfy more tuples so that queries can return relevant answers [1][2][3][4].We here suggest a simple correlation-based algorithm for relaxing constraints on failed queries for an example-based dialog modeling (EBDM) framework [5]. In the EBDM framework, a query (e.g., a set of semantic and discourse constraints) returns semantically-similar dialog examples to predict the next system action.…”