Variational Latent-State GPT for Semi-supervised Task-Oriented Dialog Systems

Recently, two approaches, fine-tuning large pre-trained language models and variational training, have attracted significant interests, separately, for semi-supervised end-to-end taskoriented dialog (TOD) systems. In this paper, we propose Variational Latent-State GPT model (VLS-GPT), which is the first to combine the strengths of the two approaches. Among many options of models, we propose the generative model and the inference model for variational learning of the end-to-end TOD system, both as auto-regressive language models based on GPT-2, which can be further trained over a mix of labeled and unlabeled dialog data in a semi-supervised manner. We develop the strategy of sampling-then-forward-computation, which successfully overcomes the memory explosion issue of using GPT in variational learning and speeds up training. Semisupervised TOD experiments are conducted on two benchmark multi-domain datasets of different languages MultiWOZ2.1 and CrossWOZ. VLS-GPT is shown to significantly outperform both supervised-only and semi-supervised baselines.

[1]  Xiaojin Zhu,et al.  --1 CONTENTS , 2006 .

[2]  Ivan Vulić,et al.  Hello, It’s GPT-2 - How Can I Help You? Towards the Use of Pretrained Language Models for Task-Oriented Dialogue Systems , 2019, EMNLP.

[3]  Min-Yen Kan,et al.  Sequicity: Simplifying Task-oriented Dialogue Systems with Single Sequence-to-Sequence Architectures , 2018, ACL.

[4]  Xiaojun Quan,et al.  UBAR: Towards Fully End-to-End Task-Oriented Dialog Systems with GPT-2 , 2020, ArXiv.

[5]  Gökhan Tür,et al.  Flexibly-Structured Model for Task-Oriented Dialogues , 2019, SIGdial.

[6]  Hua Wu,et al.  PLATO: Pre-trained Dialogue Generation Model with Discrete Latent Variable , 2020, ACL.

[7]  Tsung-Hsien Wen,et al.  Latent Intention Dialogue Models , 2017, ICML.

[8]  Zhijian Ou,et al.  A Probabilistic End-To-End Task-Oriented Dialog Model with Latent Belief States towards Semi-Supervised Learning , 2020, EMNLP.

[9]  Stefan Ultes,et al.  MultiWOZ - A Large-Scale Multi-Domain Wizard-of-Oz Dataset for Task-Oriented Dialogue Modelling , 2018, EMNLP.

[10]  Tsung-Hsien Wen,et al.  Neural Belief Tracker: Data-Driven Dialogue State Tracking , 2016, ACL.

[11]  Kee-Eung Kim,et al.  End-to-End Neural Pipeline for Goal-Oriented Dialogue Systems using GPT-2 , 2020, ACL.

[12]  Quoc V. Le,et al.  Sequence to Sequence Learning with Neural Networks , 2014, NIPS.

[13]  Lukasz Kaiser,et al.  Attention is All you Need , 2017, NIPS.

[14]  Bing Liu,et al.  An End-to-End Trainable Neural Network Model with Belief Tracking for Task-Oriented Dialog , 2017, INTERSPEECH.

[15]  Ondrej Dusek,et al.  AuGPT: Dialogue with Pre-trained Language Models and Data Augmentation , 2021, ArXiv.

[16]  Ilya Sutskever,et al.  Language Models are Unsupervised Multitask Learners , 2019 .

[17]  Zheng Zhang,et al.  CrossWOZ: A Large-Scale Chinese Cross-Domain Task-Oriented Dialogue Dataset , 2020, Transactions of the Association for Computational Linguistics.

[18]  Zhijian Ou,et al.  Task-Oriented Dialog Systems that Consider Multiple Appropriate Responses under the Same Context , 2019, AAAI.

[19]  Nurul Lubis,et al.  TripPy: A Triple Copy Strategy for Value Independent Neural Dialog State Tracking , 2020, SIGdial.

[20]  Baolin Peng,et al.  Soloist: Building Task Bots at Scale with Transfer Learning and Machine Teaching , 2021, Transactions of the Association for Computational Linguistics.

[21]  Maxine Eskénazi,et al.  Rethinking Action Spaces for Reinforcement Learning in End-to-end Dialog Agents with Latent Variable Models , 2019, NAACL.

[22]  Ming-Wei Chang,et al.  BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding , 2019, NAACL.

[23]  Maxine Eskénazi,et al.  Structured Fusion Networks for Dialog , 2019, SIGdial.

[24]  Mihail Eric,et al.  MultiWOZ 2. , 2019 .

[25]  Byeongchang Kim,et al.  Sequential Latent Knowledge Selection for Knowledge-Grounded Dialogue , 2020, ICLR.

[26]  Ben Poole,et al.  Categorical Reparameterization with Gumbel-Softmax , 2016, ICLR.

[27]  David Vandyke,et al.  A Network-based End-to-End Trainable Task-oriented Dialogue System , 2016, EACL.

[28]  Zhijian Ou,et al.  Paraphrase Augmented Task-Oriented Dialog Generation , 2020, ACL.

[29]  Yoshua Bengio,et al.  Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation , 2013, ArXiv.

[30]  Richard Socher,et al.  A Simple Language Model for Task-Oriented Dialogue , 2020, NeurIPS.

[31]  Max Welling,et al.  Auto-Encoding Variational Bayes , 2013, ICLR.

[32]  Zhaochun Ren,et al.  Explicit State Tracking with Semi-Supervisionfor Neural Dialogue Generation , 2018, CIKM.

[33]  Alec Radford,et al.  Improving Language Understanding by Generative Pre-Training , 2018 .