Controlled Text Generation for Data Augmentation in Intelligent Artificial Agents

TL;DR

This paper presents a novel controlled text generation method using conditional variational auto-encoders to augment training data for intelligent agents, significantly improving intent classification performance in low-resource scenarios.

Contribution

It introduces a new approach with conditional variational auto-encoders for data augmentation, outperforming previous controlled text generation methods and requiring only direct optimization.

Findings

Achieves up to 5% absolute F-score improvement in intent classification.

Works effectively with limited training data.

Outperforms previous controlled text generation techniques.

Abstract

Data availability is a bottleneck during early stages of development of new capabilities for intelligent artificial agents. We investigate the use of text generation techniques to augment the training data of a popular commercial artificial agent across categories of functionality, with the goal of faster development of new functionality. We explore a variety of encoder-decoder generative models for synthetic training data generation and propose using conditional variational auto-encoders. Our approach requires only direct optimization, works well with limited data and significantly outperforms the previous controlled text generation techniques. Further, the generated data are used as additional training samples in an extrinsic intent classification task, leading to improved performance by up to 5\% absolute f-score in low-resource cases, validating the usefulness of our approach.