Cascaded Span Extraction and Response Generation for Document-Grounded Dialog

TL;DR

This paper presents a cascaded approach for span extraction and response generation in document-grounded dialogs, improving accuracy in agent response prediction by focusing on relevant document spans.

Contribution

The paper introduces a cascaded model that grounds response generation on predicted spans, enhancing performance in goal-oriented document-grounded dialog tasks.

Findings

Significant improvements over baseline in span extraction accuracy.

Enhanced response generation quality using span-based grounding.

Effective ensemble methods for span prediction.

Abstract

This paper summarizes our entries to both subtasks of the first DialDoc shared task which focuses on the agent response prediction task in goal-oriented document-grounded dialogs. The task is split into two subtasks: predicting a span in a document that grounds an agent turn and generating an agent response based on a dialog and grounding document. In the first subtask, we restrict the set of valid spans to the ones defined in the dataset, use a biaffine classifier to model spans, and finally use an ensemble of different models. For the second subtask, we use a cascaded model which grounds the response prediction on the predicted span instead of the full document. With these approaches, we obtain significant improvements in both subtasks compared to the baseline.