SideControl: Controlled Open-domain Dialogue Generation via Additive Side Networks

TL;DR

SideControl introduces a novel framework for controlling open-domain dialogue generation using additive side networks, achieving better controllability, quality, and sample-efficiency with limited training data.

Contribution

The paper proposes SideControl, a new method that incorporates control signals via a control attributes loss, improving over existing gradient and weighted-decoding approaches.

Findings

Outperforms baseline methods in controllability and quality

Requires fewer training samples for effective control

Demonstrates superior sample-efficiency on benchmark datasets

Abstract

Transformer-based pre-trained language models boost the performance of open-domain dialogue systems. Prior works leverage Transformer-based pre-trained language models to generate texts with desired attributes in two general approaches: (1) gradient-based methods: updating all latent representations of pre-trained models with gradients from attribute models; (2) weighted-decoding methods: re-ranking beam candidates from pre-trained models with attribute functions. However, gradient-based methods lead to high computation cost and can easily get overfitted on small training sets, while weighted-decoding methods are inherently constrained by the low-variance high-bias pre-trained model. In this work, we propose a novel approach to control the generation of Transformer-based pre-trained language models: the SideControl framework, which leverages a novel control attributes loss to incorporate useful control signals, and is shown to perform well with very limited training samples. We evaluate our proposed method on two benchmark open-domain dialogue datasets, and results show that the SideControl framework has better controllability, higher generation quality and better sample-efficiency than existing gradient-based and weighted-decoding baselines.