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

2025-11-03

AI Summary

In controllable dialogue generation (CDG), fixed control strength often compromises attribute accuracy and linguistic fluency. To address this, we propose a dynamic weighted decoding mechanism that jointly leverages language model and attribute classifier entropy to adaptively modulate control strength at each decoding step. By measuring uncertainty in the output probability distribution via entropy, our method enables fine-grained, context-aware attribute intervention—unified across both single- and multi-attribute settings. Unlike conventional static interpolation strategies, it eliminates manual hyperparameter tuning, thereby enhancing both control precision and generation naturalness. Experiments on DailyDialog and MultiWOZ demonstrate that our approach maintains grammatical correctness and fluency while improving controllability metrics by 12.7% on average and boosting multi-attribute collaborative consistency by 9.3%, significantly outperforming existing decoding methods.

Technology Category

• Natural Language Processing: Conversational AI/Dialog Systems
• Planning, Routing, and Scheduling: Planning with Language Models
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

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.

Problem

• Dynamic adjustment of control strength in dialogue generation
• Balancing controllability and fluency in chatbot responses
• Solving fixed bias issues in multi-attribute dialogue generation

Innovation

• Dynamic control strength adjustment based on entropy
• Improves controllability while maintaining fluency
• Works for both single and multi-attribute generation