Dialog State Tracking with Reinforced Data Augmentation

Yichun Yin,Lifeng Shang,Xin Jiang,Xiao Chen,Qun Liu

DOI: https://doi.org/10.48550/arXiv.1908.07795

2019-11-18

Abstract:Neural dialog state trackers are generally limited due to the lack of quantity and diversity of annotated training data. In this paper, we address this difficulty by proposing a reinforcement learning (RL) based framework for data augmentation that can generate high-quality data to improve the neural state tracker. Specifically, we introduce a novel contextual bandit generator to learn fine-grained augmentation policies that can generate new effective instances by choosing suitable replacements for the specific context. Moreover, by alternately learning between the generator and the state tracker, we can keep refining the generative policies to generate more high-quality training data for neural state tracker. Experimental results on the WoZ and MultiWoZ (restaurant) datasets demonstrate that the proposed framework significantly improves the performance over the state-of-the-art models, especially with limited training data.

Computation and Language,Artificial Intelligence,Machine Learning

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the performance limitation in dialog state tracking (DST) due to the insufficient quantity and diversity of annotated training data. Specifically, neural dialog state trackers are usually limited by the lack of a sufficient quantity and diversity of annotated training data. To solve this difficult problem, the author proposes a data - enhancement framework based on reinforcement learning (RL), which can generate high - quality data to improve neural state trackers. By introducing a novel contextual bandit generator, this framework can learn fine - grained data - enhancement strategies, which can generate new valid instances by selecting appropriate replacements in specific contexts. In addition, by alternating learning between the generator and the state tracker, the generation strategy can be continuously optimized, thereby generating more high - quality training data for the neural state tracker. Experimental results show that this framework significantly improves performance on the WoZ and MultiWoZ (restaurant) datasets, especially in cases where the training data is limited.

Dialog State Tracking with Reinforced Data Augmentation

Hagan: Hierarchical Attentive Adversarial Learning For Task-Oriented Dialogue System

Deep Reinforcement Learning for Dialogue Generation

Plan, Generate and Complicate: Improving Low-resource Dialogue State Tracking via Easy-to-Difficult Zero-shot Data Augmentation

Learning Towards Selective Data Augmentation for Dialogue Generation.

Contextual Data Augmentation for Task-Oriented Dialog Systems

Variational Hierarchical Dialog Autoencoder for Dialog State Tracking Data Augmentation

Non-Autoregressive Dialog State Tracking

Rewarding What Matters: Step-by-Step Reinforcement Learning for Task-Oriented Dialogue

Insufficient Data Can Also Rock！Learning to Converse Using Smaller Data with Augmentation

Dialog State Tracking Using Long Short-Term Memory Neural Networks.

Data Augmentation for Retrieval- and Generation-Based Dialog Systems

Cascaded Deep Neural Network Models for Dialog State Tracking

Dialogue Distillation: Open-Domain Dialogue Augmentation Using Unpaired Data

N-Shot Learning for Augmenting Task-Oriented Dialogue State Tracking

Effective Data Augmentation Approaches to End-to-End Task-Oriented Dialogue

Towards Automatic Data Augmentation for Disordered Speech Recognition

A Unified Generation Approach for Robust Dialogue State Tracking

Stabilized In-Context Learning with Pre-trained Language Models for Few Shot Dialogue State Tracking

Diverse and Effective Synthetic Data Generation for Adaptable Zero-Shot Dialogue State Tracking

Automatically Learning Data Augmentation Policies for Dialogue Tasks