Robust belief state space representation for statistical dialogue managers using deep autoencoders
File(s)
Author(s)
Lygerakis, Fotios
Diakoloulas, Vassilios
Lagoudakis, Michail
Kotti, Margarita
Type
Conference Paper
Abstract
Statistical Dialogue Systems (SDS) have proved their humon-gous potential over the past few years. However, the lack ofefficient and robust representations of the belief state (BS)space refrains them from revealing their full potential. Thereis a great need for automatic BS representations, which willreplace the old hand-crafted, variable-length ones. To tacklethose problems, we introduce a novel use of Autoencoders(AEs). Our goal is to obtain a low-dimensional, fixed-length,and compact, yet robust representation of the BS space. Weinvestigate the use of dense AE, Denoising AE (DAE) andVariational Denoising AE (VDAE), which we combine withGP-SARSA to learn dialogue policies in the PyDial toolkit.In this framework, the BS is normally represented in a rela-tively compact, but still redundant summary space which isobtained through a heuristic mapping of the original masterspace. We show that all the proposed AE-based represen-tations consistently outperform the summary BS representa-tion. Especially, as the Semantic Error Rate (SER) increases,the DAE/VDAE-based representations obtain state-of-the-artand sample efficient performance.
Date Issued
2020-02-20
Date Acceptance
2019-09-13
Citation
2020
Publisher
IEEE
Copyright Statement
© 2020 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.
Identifier
https://ieeexplore.ieee.org/document/9003871
Source
IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) 2019
Publication Status
Published
Start Date
2019-12-14
Finish Date
2019-12-18
Coverage Spatial
Sentosa, Singapore
Date Publish Online
2020-02-20