HyLaT: Efficient Multi-Agent Communication via Hybrid Latent-Text Protocol
2026-05-25 • Computation and Language
Computation and Language
AI summaryⓘ
The authors address a problem in how multiple AI agents talk to each other using large language models. They note that current methods either use easy-to-understand text that can be slow or compact hidden messages that are hard to interpret. To fix this, the authors created HyLaT, a way for agents to send detailed info efficiently through hidden signals while still using short clear text for important points. They trained the agents in two steps to understand and use both kinds of messages. Their tests showed HyLaT cuts down on message length without hurting how well the agents perform tasks.
communication protocollarge language modelsmulti-agent systemslatent-space communicationnatural languagetraining frameworkco-trainingmessage interpretabilitycommunication efficiencyhybrid communication
Authors
Xinyi Mou, Siyuan Wang, Zejun Li, Yulan He, Zhongyu Wei
Abstract
Communication protocol design is a central challenge in large language model-based multi-agent systems. Existing single-channel approaches face an inherent communication trilemma: text-based methods are interpretable but verbose, while latent-space methods are efficient but opaque and limited to unidirectional workflows. Inspired by multi-channel communication theory, we propose HyLaT, a hybrid latent-text communication protocol that transmits elaborate cognitive signals through a latent channel for efficiency, while expressing concise critical signals in natural language to preserve interpretability and precision. We introduce a two-stage training framework combining single-agent hybrid generation learning and multi-agent interactive co-training, enabling agents to generate and interpret hybrid messages across multiple rounds of interaction. Experiments demonstrate that HyLaT reduces communication overhead significantly while maintaining competitive task performance, with strong generalization and robustness across diverse settings.