Zero-Shot Dialog Generation with Cross-Domain Latent Actions (2018-05-01T00:00:00.000000Z)

TL;DR

Experimental results show that the proposed zero-shot dialog generation method is able to achieve superior performance in learning dialog models that can rapidly adapt their behavior to new domains and suggests promising future research.

Abstract

This paper introduces zero-shot dialog generation (ZSDG), as a step towards neural dialog systems that can instantly generalize to new situations with minimum data. ZSDG requires an end-to-end generative dialog system to generalize to a new domain for which only a domain description is provided and no training dialogs are available. Then a novel learning framework, Action Matching, is proposed. This algorithm can learn a cross-domain embedding space that models the semantics of dialog responses which in turn, enables a neural dialog generation model to generalize to new domains. We evaluate our methods on two datasets, a new synthetic dialog dataset, and an existing human-human multi-domain dialog dataset. Experimental results show that our method is able to achieve superior performance in learning dialog models that can rapidly adapt their behavior to new domains and suggests promising future research.

Authors

M. Eskénazi

7 papers

Tiancheng Zhao

6 papers

TL;DR

Abstract

Authors

References36 items

Zero-Shot Question Generation from Knowledge Graphs for Unseen Predicates and Entity Types

Bootstrapping incremental dialogue systems from minimal data: the generalisation power of dialogue grammars

Towards Zero-Shot Frame Semantic Parsing for Domain Scaling

Zero-Shot Task Generalization with Multi-Task Deep Reinforcement Learning

Generative Encoder-Decoder Models for Task-Oriented Spoken Dialog Systems with Chatting Capability

Key-Value Retrieval Networks for Task-Oriented Dialogue

One-Shot Imitation Learning

A Copy-Augmented Sequence-to-Sequence Architecture Gives Good Performance on Task-Oriented Dialogue

Pointer Sentinel Mixture Models

Towards End-to-End Learning for Dialog State Tracking and Management using Deep Reinforcement Learning

A Hierarchical Latent Variable Encoder-Decoder Model for Generating Dialogues

A Network-based End-to-End Trainable Task-oriented Dialogue System

Incorporating Copying Mechanism in Sequence-to-Sequence Learning

Zero-shot learning of intent embeddings for expansion by convolutional deep structured semantic models

Evaluating Prerequisite Qualities for Learning End-to-End Dialog Systems

Effective Approaches to Attention-based Neural Machine Translation

An embarrassingly simple approach to zero-shot learning

A Hierarchical Neural Autoencoder for Paragraphs and Documents

Adam: A Method for Stochastic Optimization

Recurrent Neural Network Regularization

Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation

Auto-Encoding Variational Bayes

Zero-Shot Learning Through Cross-Modal Transfer

Zero-shot Learning with Semantic Output Codes

Zero-data Learning of New Tasks

A unified architecture for natural language processing: deep neural networks with multitask learning

What else can I do for you?

Learning with Kernels: Support Vector Machines, Regularization, Optimization, and Beyond

Bleu: a Method for Automatic Evaluation of Machine Translation

Long Short-Term Memory

Gaussian Processes for POMDP-Based Dialogue Manager Optimization

What do you need? User (0.83): Hello robot. I am looking for a restaurant. Sys: Which city are you interested in? User (0.75): I am in San Jose. And Korean. Uhm sorry, I am interested in Chinese food

direction. The discourse-level LSTM-RNN and the decoder have hidden size 512

For decoders with copy mechanisms, we also use the pointer loss proposed in (Merity et al., 2016) which minimizes − log( g + (cid:80) kj ∈ I ( x , c ) α kj ) , where I results in all positions of

Sys: QUERY goal: restaurant food-pref: Chinese location

0.08. 30% dropouts are applied at the input and output of recurrent neural networks

Field of Study

Journal Information

Name

Volume