Frames: A Corpus for Adding Memory to Goal-Oriented Dialogue Systems

TL;DR

The paper introduces the Frames dataset, a large corpus of human dialogues, to explore memory's role in goal-oriented dialogue systems, and proposes a new task called frame tracking to improve dialogue state management.

Contribution

It provides a new dataset and task for studying memory in dialogue systems, along with a baseline model for frame tracking.

Findings

Frames dataset contains 1369 dialogues with 15 turns on average.

Frame tracking extends traditional state tracking to multiple states.

Baseline model demonstrates the feasibility of the proposed task.

Abstract

This paper presents the Frames dataset (Frames is available at http://datasets.maluuba.com/Frames), a corpus of 1369 human-human dialogues with an average of 15 turns per dialogue. We developed this dataset to study the role of memory in goal-oriented dialogue systems. Based on Frames, we introduce a task called frame tracking, which extends state tracking to a setting where several states are tracked simultaneously. We propose a baseline model for this task. We show that Frames can also be used to study memory in dialogue management and information presentation through natural language generation.