Learning Discourse-level Diversity for Neural Dialog Models using Conditional Variational Autoencoders

Abstract

While recent neural encoder-decoder models have shown great promise in modeling open-domain conversations, they often generate dull and generic responses. Unlike past work that has focused on diversifying the output of the decoder at word-level to alleviate this problem, we present a novel framework based on conditional variational autoencoders that captures the discourse-level diversity in the encoder. Our model uses latent variables to learn a distribution over potential conversational intents and generates diverse responses using only greedy decoders. We have further developed a novel variant that is integrated with linguistic prior knowledge for better performance. Finally, the training procedure is improved by introducing a bag-of-word loss. Our proposed models have been validated to generate significantly more diverse responses than baseline approaches and exhibit competence in discourse-level decision-making.

arXiv commentAppeared in ACL2017 proceedings as a long paper. Correct a calculation mistake in Table 1 E-bow & A-bow and results into higher scores

References in corpus (16)

• Adam: A Method for Stochastic Optimization
• Sequence to Sequence Learning with Neural Networks
• Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling
• Stochastic Backpropagation and Approximate Inference in Deep Generative Models
• Recurrent Neural Network Regularization
• A Neural Conversational Model
• Dialogue Act Modeling for Automatic Tagging and Recognition of Conversational Speech
• How NOT To Evaluate Your Dialogue System: An Empirical Study of Unsupervised Evaluation Metrics for Dialogue Response Generation
• Deep Reinforcement Learning for Dialogue Generation
• Fine-grained Analysis of Sentence Embeddings Using Auxiliary Prediction Tasks
• A Persona-Based Neural Conversation Model
• Towards End-to-End Learning for Dialog State Tracking and Management using Deep Reinforcement Learning
• Sequence-to-Sequence Learning as Beam-Search Optimization
• Attribute2Image: Conditional Image Generation from Visual Attributes
• A Neural Network Approach to Context-Sensitive Generation of Conversational Responses
• The Influence of Context on Dialogue Act Recognition

Cited by in corpus (72)

• DialogWAE: Multimodal Response Generation with Conditional Wasserstein Auto-Encoder
• Generating Persona Consistent Dialogues by Exploiting Natural Language Inference
• Breaking the Softmax Bottleneck: A High-Rank RNN Language Model
• Topic-Guided Variational Autoencoders for Text Generation
• ControlVAE: Controllable Variational Autoencoder
• Zero-Resource Knowledge-Grounded Dialogue Generation
• Topic-based Evaluation for Conversational Bots
• Deep Learning Based Chatbot Models
• CG-BERT: Conditional Text Generation with BERT for Generalized Few-shot Intent Detection
• Eval all, trust a few, do wrong to none: Comparing sentence generation models
• Z-Forcing: Training Stochastic Recurrent Networks
• How higher goals are constructed and collapse under stress: a hierarchical Bayesian control systems perspective
• Towards Explainable and Controllable Open Domain Dialogue Generation with Dialogue Acts
• STAR: A Schema-Guided Dialog Dataset for Transfer Learning
• Learning to Select Knowledge for Response Generation in Dialog Systems
• Unsupervised Discrete Sentence Representation Learning for Interpretable Neural Dialog Generation
• Gunrock 2.0: A User Adaptive Social Conversational System
• An Adversarial Approach to High-Quality, Sentiment-Controlled Neural Dialogue Generation
• OpenViDial: A Large-Scale, Open-Domain Dialogue Dataset with Visual Contexts
• MojiTalk: Generating Emotional Responses at Scale
• FlipDial: A Generative Model for Two-Way Visual Dialogue
• Dirichlet Variational Autoencoder for Text Modeling
• Local Contextual Attention with Hierarchical Structure for Dialogue Act Recognition
• Persona-Aware Tips Generation
• End-to-End Trainable Non-Collaborative Dialog System
• Generative Encoder-Decoder Models for Task-Oriented Spoken Dialog Systems with Chatting Capability
• Unsupervised Dialog Structure Learning
• EmpDG: Multiresolution Interactive Empathetic Dialogue Generation
• Decoupling Strategy and Generation in Negotiation Dialogues
• KECRS: Towards Knowledge-Enriched Conversational Recommendation System
• Policy-Driven Neural Response Generation for Knowledge-Grounded Dialogue Systems
• Chat More If You Like: Dynamic Cue Words Planning to Flow Longer Conversations
• RefNet: A Reference-aware Network for Background Based Conversation
• Generating Multiple Diverse Responses for Short-Text Conversation
• Probing Neural Dialog Models for Conversational Understanding
• Discovering Dialog Structure Graph for Open-Domain Dialog Generation
• Dialogue Response Ranking Training with Large-Scale Human Feedback Data
• Report from the NSF Future Directions Workshop, Toward User-Oriented Agents: Research Directions and Challenges
• OpenViDial 2.0: A Larger-Scale, Open-Domain Dialogue Generation Dataset with Visual Contexts
• Why Do Neural Response Generation Models Prefer Universal Replies?
• Polite Dialogue Generation Without Parallel Data
• Composed Variational Natural Language Generation for Few-shot Intents
• mu-Forcing: Training Variational Recurrent Autoencoders for Text Generation
• Why are Sequence-to-Sequence Models So Dull? Understanding the Low-Diversity Problem of Chatbots
• Hierarchical CVAE for Fine-Grained Hate Speech Classification
• Latent Topic Conversational Models
• The Style-Content Duality of Attractiveness: Learning to Write Eye-Catching Headlines via Disentanglement
• A Batch Normalized Inference Network Keeps the KL Vanishing Away
• OpenMEVA: A Benchmark for Evaluating Open-ended Story Generation Metrics
• Aiming to Know You Better Perhaps Makes Me a More Engaging Dialogue Partner
• UNION: An Unreferenced Metric for Evaluating Open-ended Story Generation
• Relevance-Promoting Language Model for Short-Text Conversation
• ControlVAE: Tuning, Analytical Properties, and Performance Analysis
• Controllable and Diverse Text Generation in E-commerce
• Semantic-Enhanced Explainable Finetuning for Open-Domain Dialogues
• EnsembleGAN: Adversarial Learning for Retrieval-Generation Ensemble Model on Short-Text Conversation
• A Sentiment-Controllable Topic-to-Essay Generator with Topic Knowledge Graph
• A Simple and Efficient Multi-Task Learning Approach for Conditioned Dialogue Generation
• Toward Interpretability of Dual-Encoder Models for Dialogue Response Suggestions
• Posterior-GAN: Towards Informative and Coherent Response Generation with Posterior Generative Adversarial Network
• Who Responded to Whom: The Joint Effects of Latent Topics and Discourse in Conversation Structure
• Incremental Learning from Scratch for Task-Oriented Dialogue Systems
• Automatically Select Emotion for Response via Personality-affected Emotion Transition
• Enhancing audio quality for expressive Neural Text-to-Speech
• TransSent: Towards Generation of Structured Sentences with Discourse Marker
• Conversational Semantic Role Labeling
• Improving Robustness of Neural Dialog Systems in a Data-Efficient Way with Turn Dropout
• Incorporating Relevant Knowledge in Context Modeling and Response Generation
• Evidence-Aware Inferential Text Generation with Vector Quantised Variational AutoEncoder
• Diversifying Topic-Coherent Response Generation for Natural Multi-turn Conversations
• Domain-Constrained Advertising Keyword Generation
• Pseudo Siamese Network for Few-shot Intent Generation