UniGDD: A Unified Generative Framework for Goal-Oriented Document-Grounded Dialogue

TL;DR

This paper introduces UniGDD, a unified generative framework for goal-oriented document-grounded dialogue that sequentially generates grounding knowledge and responses, improving coherence and reducing error propagation.

Contribution

It proposes a novel unified generative approach with multi-task learning and temperature scheduling to enhance dialogue response quality.

Findings

Outperforms existing pipeline methods in accuracy.

Effectively reduces error propagation in dialogue generation.

Demonstrates robustness with relevant document information.

Abstract

The goal-oriented document-grounded dialogue aims at responding to the user query based on the dialogue context and supporting document. Existing studies tackle this problem by decomposing it into two sub-tasks: knowledge identification and response generation. However, such pipeline methods would unavoidably suffer from the error propagation issue. This paper proposes to unify these two sub-tasks via sequentially generating the grounding knowledge and the response. We further develop a prompt-connected multi-task learning strategy to model the characteristics and connections of different tasks and introduce linear temperature scheduling to reduce the negative effect of irrelevant document information. Experimental results demonstrate the effectiveness of our framework.