On the Evaluation of Dialogue Systems with Next Utterance Classification

TL;DR

This paper evaluates the Next-Utterance-Classification task for dialogue systems by comparing human and machine performance, confirming its feasibility and potential as an evaluation method for automated dialogue system development.

Contribution

It provides the first validation of NUC as an evaluation method by analyzing human performance across domains and expertise levels, and compares it with state-of-the-art systems.

Findings

Humans outperform chance in NUC tasks.

Performance varies across domains and expertise levels.

Automated systems match novice performance but lag behind experts.

Abstract

An open challenge in constructing dialogue systems is developing methods for automatically learning dialogue strategies from large amounts of unlabelled data. Recent work has proposed Next-Utterance-Classification (NUC) as a surrogate task for building dialogue systems from text data. In this paper we investigate the performance of humans on this task to validate the relevance of NUC as a method of evaluation. Our results show three main findings: (1) humans are able to correctly classify responses at a rate much better than chance, thus confirming that the task is feasible, (2) human performance levels vary across task domains (we consider 3 datasets) and expertise levels (novice vs experts), thus showing that a range of performance is possible on this type of task, (3) automated dialogue systems built using state-of-the-art machine learning methods have similar performance to the human novices, but worse than the experts, thus confirming the utility of this class of tasks for driving further research in automated dialogue systems.