Domain-Level Reasoning for Dialogue Systems

Author: Dirk Bühler

Status: completed

Description:

Motivated by the need to make the human-machine information-seeking dialogue as efficient and user-friendly as possible the subject of this thesis is to propose a logic-based reasoning component for a Spoken Language Dialogue Systems architecture. It enables the integration of isolated services and devices into a unified domain model that serves as the basis for user assistance. Such assistance seems especially relevant in mobile environments, since an efficient and intelligent interaction is assumed to reduce the distraction of the driver and thus the safety risk.

The core component, called Problem Assistant, supports the Dialogue Management component. Specifically, the assistantaims at enabling hypothetical reasoning by maintaining concurrent knowledge contexts (possible worlds), signalling inferences and choice points that may initiate a negotiation dialogue between the user and the system, and detecting conflicts that arise from a user's specifications (goals) and suggesting possible solutions.

The core processing is finite model generation. This inference technique tries to find solutions that fit both the user's constraints and are consistent with the Problem Assistant's rule base. Since the assistant interactively generates transparent information about its inference process, the approach envisaged in this thesis will provide the basis for incremental explanation dialogues and collaborative conflict resolution.

Further Information

PhD_Description.pdf

Presentation

Link to Document