cs.CL（2026-05-21）

📊 共 18 篇论文 | 🔗 3 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (11 🔗3) 支柱二：RL算法与架构 (RL & Architecture) (6) 支柱一：机器人控制 (Robot Control) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (11 篇)

#题目一句话要点标签🔗⭐
1 LatentOmni: Rethinking Omni-Modal Understanding via Unified Audio-Visual Latent Reasoning LatentOmni：通过统一的音视频潜在空间推理，重新思考全模态理解 large language model multimodal chain-of-thought
2 Understanding Data Temporality Impact on Large Language Models Pre-training 研究数据时序性对大语言模型预训练的影响，提出时序知识评估基准。 large language model ✅
3 Do Factual Recall Mechanisms Carry over from Text to Speech in Multimodal Language Models? 研究多模态语言模型中事实性知识从文本到语音的迁移机制 multimodal
4 Beyond Temperature: Hyperfitting as a Late-Stage Geometric Expansion 揭示超拟合提升LLM生成质量的几何扩张机制，并提出Late-Stage LoRA高效微调策略。 large language model
5 Comparing LLM and Fine-Tuned Model Performance on NVDRS Circumstance Extraction with Varying Prompt Complexity 提出基于复杂性评分的混合框架，利用LLM和微调模型提升NVDRS死因推断准确率 large language model
6 Scene Abstraction for Lexical Semantics: Structured Representations of Situated Meaning 提出场景抽象框架，构建词汇语义的结构化场景表示，提升情境理解。 large language model
7 Polite on the Surface, Wrong in Practice: A Curated Dataset for Fixing Honorific Failures in Multilingual Bangla Generation 提出BLADE数据集，提升孟加拉语生成模型在礼貌性和结构一致性上的表现 large language model ✅
8 Cross-Lingual Consensus: Aligning Multilingual Cultural Knowledge via Multilingual Self-Consistency 提出跨语言共识框架，通过多语言自洽性对齐多语言文化知识，提升大语言模型文化敏感性。 large language model
9 Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements 提出Ishigaki-IDS-Bench基准，用于评估LLM从BIM信息需求生成IDS XML的能力 large language model
10 Hy-MT2: A Family of Fast, Efficient and Powerful Multilingual Translation Models in the Wild Hy-MT2：一套快速、高效且强大的多语言翻译模型，适用于复杂现实场景 instruction following
11 SpecHop: Continuous Speculation for Accelerating Multi-Hop Retrieval Agents SpecHop：通过连续推测加速多跳检索代理 large language model ✅

🔬 支柱二：RL算法与架构 (RL & Architecture) (6 篇)

#题目一句话要点标签🔗⭐
12 Faithful-MR1: Faithful Multimodal Reasoning via Anchoring and Reinforcing Visual Attention 提出Faithful-MR1框架，通过锚定和强化视觉注意力提升多模态推理的忠实性。 reinforcement learning large language model multimodal
13 DeferMem: Query-Time Evidence Distillation via Reinforcement Learning for Long-Term Memory QA 提出DeferMem，通过强化学习进行查询时证据提炼，解决长时记忆问答问题 reinforcement learning distillation large language model
14 LANG: Reinforcement Learning for Multilingual Reasoning with Language-Adaptive Hint Guidance LANG框架通过语言自适应提示指导，提升多语言推理强化学习效果 reinforcement learning large language model language conditioned
15 Token-weighted Direct Preference Optimization with Attention 提出Token加权DPO方法AttentionPO，利用LLM注意力机制提升偏好优化效果 DPO direct preference optimization large language model
16 Self-Policy Distillation via Capability-Selective Subspace Projection 提出基于能力选择子空间投影的自策略蒸馏方法，提升LLM泛化能力。 distillation large language model
17 Unified Data Selection for LLM Reasoning 提出基于高熵和（HES）的无训练数据选择方法，提升LLM推理能力。 reinforcement learning large language model

🔬 支柱一：机器人控制 (Robot Control) (1 篇)

#题目一句话要点标签🔗⭐
18 Reducing Political Manipulation with Consistency Training 提出政治一致性训练，以减少大型语言模型中的政治操纵 manipulation metric depth large language model

⬅️ 返回 cs.CL 首页 · 🏠 返回主页