A Diversity-Promoting Objective Function for Neural Conversation Models

Abstract

Sequence-to-sequence neural network models for generation of conversational responses tend to generate safe, commonplace responses (e.g., "I don't know") regardless of the input. We suggest that the traditional objective function, i.e., the likelihood of output (response) given input (message) is unsuited to response generation tasks. Instead we propose using Maximum Mutual Information (MMI) as the objective function in neural models. Experimental results demonstrate that the proposed MMI models produce more diverse, interesting, and appropriate responses, yielding substantive gains in BLEU scores on two conversational datasets and in human evaluations.

In. Proc of NAACL 2016

References in corpus (3)

• Sequence to Sequence Learning with Neural Networks
• A Neural Conversational Model
• Attention with Intention for a Neural Network Conversation Model

Cited by in corpus (39)

• A Survey on Dialogue Systems: Recent Advances and New Frontiers
• Emotional Chatting Machine: Emotional Conversation Generation with Internal and External Memory
• Personalized Dialogue Generation with Diversified Traits
• Low-Resource Knowledge-Grounded Dialogue Generation
• Generating Persona Consistent Dialogues by Exploiting Natural Language Inference
• Generative Deep Neural Networks for Dialogue: A Short Review
• One Chatbot Per Person: Creating Personalized Chatbots based on Implicit User Profiles
• Speaking the Same Language: Matching Machine to Human Captions by Adversarial Training
• Latent Intention Dialogue Models
• Context-aware Captions from Context-agnostic Supervision
• A Hybrid Retrieval-Generation Neural Conversation Model
• Learning Implicit User Profiles for Personalized Retrieval-Based Chatbot
• Gunrock 2.0: A User Adaptive Social Conversational System
• Improving Conversational Recommender Systems via Knowledge Graph based Semantic Fusion
• Emotion-aware Chat Machine: Automatic Emotional Response Generation for Human-like Emotional Interaction
• Neural Personalized Response Generation as Domain Adaptation
• Shaping the Narrative Arc: An Information-Theoretic Approach to Collaborative Dialogue
• Retrieval-Augmented Transformer-XL for Close-Domain Dialog Generation
• Chat More If You Like: Dynamic Cue Words Planning to Flow Longer Conversations
• Hierarchical Context Tagging for Utterance Rewriting
• Learning to Ask Questions in Open-domain Conversational Systems with Typed Decoders
• Why Do Neural Response Generation Models Prefer Universal Replies?
• Beyond task success: A closer look at jointly learning to see, ask, and GuessWhat
• Affective Neural Response Generation
• Hybrid Generative-Retrieval Transformers for Dialogue Domain Adaptation
• Target Guided Emotion Aware Chat Machine
• Domain Agnostic Real-Valued Specificity Prediction
• An Ensemble Dialogue System for Facts-Based Sentence Generation
• Relevance-Promoting Language Model for Short-Text Conversation
• Controllable Dialogue Generation with Disentangled Multi-grained Style Specification and Attribute Consistency Reward
• Which Kind Is Better in Open-domain Multi-turn Dialog,Hierarchical or Non-hierarchical Models? An Empirical Study
• A Face-to-Face Neural Conversation Model
• Automatic, Personalized, and Flexible Playlist Generation using Reinforcement Learning
• A Sentiment-Controllable Topic-to-Essay Generator with Topic Knowledge Graph
• Improving Matching Models with Hierarchical Contextualized Representations for Multi-turn Response Selection
• Do Encoder Representations of Generative Dialogue Models Encode Sufficient Information about the Task ?
• Turn-level Dialog Evaluation with Dialog-level Weak Signals for Bot-Human Hybrid Customer Service Systems
• Learning Natural Language Generation from Scratch
• Predicting User Engagement Status for Online Evaluation of Intelligent Assistants