Dynamic Collaboration of Multi-Language Models based on Minimal Complete Semantic Units

2025-08-26

AI Summary

This paper addresses semantic misalignment in multilingual large language model (LLM) collaborative inference caused by lexical misalignment across languages. To tackle this, we propose a dynamic collaboration framework grounded in the Minimum Intact Conceptual Unit (MICU). Methodologically: (1) MICU is formally defined as the fundamental cross-lingual semantic alignment unit; (2) a distribution-distance-driven, token-level dynamic model selection mechanism is introduced; and (3) an autoregressive collaborative generation paradigm is established, integrating multi-model token distributions while enforcing semantic alignment. Extensive evaluation on multilingual reasoning benchmarks—including XNLI and XCOPA—demonstrates consistent and significant improvements over existing collaborative approaches, achieving an average +3.2% accuracy gain. Results confirm the framework’s strong cross-lingual generalization and robustness. The implementation is publicly available.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Multiagent Systems: Coordination and Collaboration

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

This paper investigates the enhancement of reasoning capabilities in language models through token-level multi-model collaboration. Our approach selects the optimal tokens from the next token distributions provided by multiple models to perform autoregressive reasoning. Contrary to the assumption that more models yield better results, we introduce a distribution distance-based dynamic selection strategy (DDS) to optimize the multi-model collaboration process. To address the critical challenge of vocabulary misalignment in multi-model collaboration, we propose the concept of minimal complete semantic units (MCSU), which is simple yet enables multiple language models to achieve natural alignment within the linguistic space. Experimental results across various benchmarks demonstrate the superiority of our method. The code will be available at https://github.com/Fanye12/DDS.

Problem

• Optimizing token selection from multiple language models
• Addressing vocabulary misalignment in multi-model collaboration
• Enhancing reasoning through dynamic model selection strategy

Innovation

• Token-level multi-model collaboration for reasoning
• Dynamic selection strategy using distribution distance
• Minimal complete semantic units for vocabulary alignment