通过 Mixture-of-Thought 学习推理以实现逻辑推理
Learning to Reason via Mixture-of-Thought for Logical Reasoning
AI总结：
针对 LLM 训练仅依赖单一推理模态的问题，本文提出融合自然语言、代码和真值表的 MoT 两阶段框架，在 FOLIO 与 ProofWriter 上最高提升平均准确率 11.7 个百分点。
摘要：
人类自然地利用多种推理模态来学习和解决逻辑问题，即自然语言、代码和符号逻辑等不同的表示形式。相比之下，大多数现有的基于 LLM 的方法在训练期间只使用单一推理模态，通常是自然语言。尽管一些方法在推理时探索了模态选择或增强，但训练过程仍然对模态无感知，限制了模态之间的协同作用。为填补这一空白，我们提出 Mixture-of-Thought（MoT）框架，使 LLM 能够跨三种互补模态进行推理：自然语言、代码，以及新引入的符号模态 truth-table（真值表），该模态系统地枚举逻辑情况，并在一定程度上缓解自然语言推理中的关键失败模式。MoT 采用两阶段设计：（1）自进化 MoT 训练，从跨模态的、经过筛选的自生成推理依据中联合学习；（2）MoT 推理，充分利用三种模态的协同作用以产生更好的预测。在包括 FOLIO 和 ProofWriter 在内的逻辑推理基准上的实验表明，我们的 MoT 框架持续且显著优于采用单模态 chain-of-thought 方法的强大 LLM 基线，实现了最高 +11.7 个百分点的平均准确率提升。进一步分析表明，MoT 框架对训练和推理阶段均有益；它在更困难的逻辑推理问题上尤其有效；并且不同模态贡献了互补优势，其中 truth-table 推理有助于克服自然语言推断中的关键瓶颈。
英文摘要：
Human beings naturally utilize multiple reasoning modalities to learn and solve logical problems, i.e., different representational formats such as natural language, code, and symbolic logic. In contrast, most existing LLM-based approaches operate with a single reasoning modality during training, typically natural language. Although some methods explored modality selection or augmentation at inference time, the training process remains modality-blind, limiting synergy among modalities. To fill in this gap, we propose Mixture-of-Thought (MoT), a framework that enables LLMs to reason across three complementary modalities: natural language, code, and a newly introduced symbolic modality, truth-table, which systematically enumerates logical cases and partially mitigates key failure modes in natural language reasoning. MoT adopts a two-phase design: (1) self-evolving MoT training, which jointly learns from filtered, self-generated rationales across modalities; and (2) MoT inference, which fully leverages the synergy of three modalities to produce better predictions. Experiments on logical reasoning benchmarks including FOLIO and ProofWriter demonstrate that our MoT framework consistently and significantly outperforms strong LLM baselines with single-modality chain-of-thought approaches, achieving up to +11.7pp average accuracy gain. Further analyses show that our MoT framework benefits both training and inference stages; that it is particularly effective on harder logical reasoning problems; and that different modalities contribute complementary strengths, with truth-table reasoning helping to overcome key bottlenecks in natural language inference.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
