ECO Decoding: Entropy-Based Control for Controllability and Fluency in Controllable Dialogue Generation

TL;DR

ECO decoding introduces an entropy-based dynamic control method for dialogue generation, improving controllability and fluency by adjusting attribute influence at each step, outperforming previous fixed-weight approaches.

Contribution

This paper presents ECO decoding, a novel entropy-based method that adaptively manages control strength in dialogue generation, enhancing performance over fixed-weight techniques.

Findings

ECO decoding improves controllability and fluency in dialogue responses.

It outperforms prior methods across multiple datasets and models.

The approach effectively handles multi-attribute generation scenarios.

Abstract

Controllable Dialogue Generation (CDG) enables chatbots to generate responses with desired attributes, and weighted decoding methods have achieved significant success in the CDG task. However, using a fixed constant value to manage the bias of attribute probabilities makes it challenging to find an ideal control strength that satisfies both controllability and fluency. To address this issue, we propose ECO decoding (Entropy-based COntrol), which dynamically adjusts the control strength at each generation step according to the model's entropy in both the language model and attribute classifier probability distributions. Experiments on the DailyDialog and MultiWOZ datasets demonstrate that ECO decoding consistently improves controllability while maintaining fluency and grammaticality, outperforming prior decoding methods across various models and settings. Furthermore, ECO decoding alleviates probability interpolation issues in multi-attribute generation and consequently demonstrates strong performance in both single and multi-attribute scenarios.