Optimizing Dialogue Management with Reinforcement Learning: Experiments with the NJFun System

TL;DR

This paper introduces a reinforcement learning method to automatically optimize dialogue policies in spoken dialogue systems, demonstrated through the NJFun system that provides New Jersey activity information, resulting in measurable performance improvements.

Contribution

It presents a novel reinforcement learning approach for dialogue policy optimization applied to a real-world system, addressing technical challenges and demonstrating empirical benefits.

Findings

Reinforcement learning improved NJFun's system performance.

Empirical evaluation showed measurable enhancements.

The approach effectively adapts dialogue policies in practice.

Abstract

Designing the dialogue policy of a spoken dialogue system involves many nontrivial choices. This paper presents a reinforcement learning approach for automatically optimizing a dialogue policy, which addresses the technical challenges in applying reinforcement learning to a working dialogue system with human users. We report on the design, construction and empirical evaluation of NJFun, an experimental spoken dialogue system that provides users with access to information about fun things to do in New Jersey. Our results show that by optimizing its performance via reinforcement learning, NJFun measurably improves system performance.