End-to-end latent-variable task-oriented dialogue system with exact log-likelihood optimization

Haotian XU, Haiyun PENG, Haoran XIE, Erik CAMBRIA, Liuyang ZHOU*, Weiguo ZHENG

*Corresponding author for this work

Research output: Journal PublicationsJournal Article (refereed)

1 Scopus Citations

Abstract

We propose an end-to-end dialogue model based on a hierarchical encoder-decoder, which employed a discrete latent variable to learn underlying dialogue intentions. The system is able to model the structure of utterances dominated by statistics of the language and the dependencies among utterances in dialogues without manual dialogue state design. We argue that the latent discrete variable interprets the intentions that guide machine responses generation. We also propose a model which can be refined autonomously with reinforcement learning, due to that intention selection at each dialogue turn can be formulated as a sequential decision-making process. Our experiments show that exact MLE optimized model is much more robust than neural variational inference on dialogue success rate with limited BLEU sacrifice.

Original languageEnglish
Pages (from-to)1-14
Number of pages14
JournalWorld Wide Web
Early online date7 Jun 2019
DOIs
Publication statusE-pub ahead of print - 7 Jun 2019
Externally publishedYes

Fingerprint

Maximum likelihood estimation
Reinforcement learning
Decision making
Statistics
Experiments

Bibliographical note

This work was supported by the Shenzhen Science and Technology Innovation Committee with the project name of Intelligent Question Answering Robot, under grant NO. CKCY20170508121036342.

Keywords

  • Dialogue intention
  • Dialogue model
  • Hierarchical encoder-decoder
  • Log-likelihood optimization

Cite this

XU, Haotian ; PENG, Haiyun ; XIE, Haoran ; CAMBRIA, Erik ; ZHOU, Liuyang ; ZHENG, Weiguo. / End-to-end latent-variable task-oriented dialogue system with exact log-likelihood optimization. In: World Wide Web. 2019 ; pp. 1-14.
@article{6848ed5ed37645aab156866159f49a78,
title = "End-to-end latent-variable task-oriented dialogue system with exact log-likelihood optimization",
abstract = "We propose an end-to-end dialogue model based on a hierarchical encoder-decoder, which employed a discrete latent variable to learn underlying dialogue intentions. The system is able to model the structure of utterances dominated by statistics of the language and the dependencies among utterances in dialogues without manual dialogue state design. We argue that the latent discrete variable interprets the intentions that guide machine responses generation. We also propose a model which can be refined autonomously with reinforcement learning, due to that intention selection at each dialogue turn can be formulated as a sequential decision-making process. Our experiments show that exact MLE optimized model is much more robust than neural variational inference on dialogue success rate with limited BLEU sacrifice.",
keywords = "Dialogue intention, Dialogue model, Hierarchical encoder-decoder, Log-likelihood optimization",
author = "Haotian XU and Haiyun PENG and Haoran XIE and Erik CAMBRIA and Liuyang ZHOU and Weiguo ZHENG",
note = "This work was supported by the Shenzhen Science and Technology Innovation Committee with the project name of Intelligent Question Answering Robot, under grant NO. CKCY20170508121036342.",
year = "2019",
month = "6",
day = "7",
doi = "10.1007/s11280-019-00688-8",
language = "English",
pages = "1--14",
journal = "World Wide Web",
issn = "1386-145X",
publisher = "Springer New York",

}

End-to-end latent-variable task-oriented dialogue system with exact log-likelihood optimization. / XU, Haotian; PENG, Haiyun; XIE, Haoran; CAMBRIA, Erik; ZHOU, Liuyang; ZHENG, Weiguo.

In: World Wide Web, 07.06.2019, p. 1-14.

Research output: Journal PublicationsJournal Article (refereed)

TY - JOUR

T1 - End-to-end latent-variable task-oriented dialogue system with exact log-likelihood optimization

AU - XU, Haotian

AU - PENG, Haiyun

AU - XIE, Haoran

AU - CAMBRIA, Erik

AU - ZHOU, Liuyang

AU - ZHENG, Weiguo

N1 - This work was supported by the Shenzhen Science and Technology Innovation Committee with the project name of Intelligent Question Answering Robot, under grant NO. CKCY20170508121036342.

PY - 2019/6/7

Y1 - 2019/6/7

N2 - We propose an end-to-end dialogue model based on a hierarchical encoder-decoder, which employed a discrete latent variable to learn underlying dialogue intentions. The system is able to model the structure of utterances dominated by statistics of the language and the dependencies among utterances in dialogues without manual dialogue state design. We argue that the latent discrete variable interprets the intentions that guide machine responses generation. We also propose a model which can be refined autonomously with reinforcement learning, due to that intention selection at each dialogue turn can be formulated as a sequential decision-making process. Our experiments show that exact MLE optimized model is much more robust than neural variational inference on dialogue success rate with limited BLEU sacrifice.

AB - We propose an end-to-end dialogue model based on a hierarchical encoder-decoder, which employed a discrete latent variable to learn underlying dialogue intentions. The system is able to model the structure of utterances dominated by statistics of the language and the dependencies among utterances in dialogues without manual dialogue state design. We argue that the latent discrete variable interprets the intentions that guide machine responses generation. We also propose a model which can be refined autonomously with reinforcement learning, due to that intention selection at each dialogue turn can be formulated as a sequential decision-making process. Our experiments show that exact MLE optimized model is much more robust than neural variational inference on dialogue success rate with limited BLEU sacrifice.

KW - Dialogue intention

KW - Dialogue model

KW - Hierarchical encoder-decoder

KW - Log-likelihood optimization

UR - http://www.scopus.com/inward/record.url?scp=85067260059&partnerID=8YFLogxK

U2 - 10.1007/s11280-019-00688-8

DO - 10.1007/s11280-019-00688-8

M3 - Journal Article (refereed)

AN - SCOPUS:85067260059

SP - 1

EP - 14

JO - World Wide Web

JF - World Wide Web

SN - 1386-145X

ER -