arXivDaily arXiv每日学术速递 周一至周五更新

期刊&会议

Annual Meeting of the Association for Computational Linguistics · 会议 · Natural Language Processing

2026-07-31 至 2026-07-31 共收录 6
2607.28282 2026-07-31 cs.CL cs.LG 新提交

(Towards) Scalable Reliable Automated Evaluation with Large Language Models

面向可扩展的可靠自动化评估:基于大语言模型

Bertil Braun, Martin Forell

机构 * KIT(卡尔斯鲁厄理工学院)

AI总结 本研究提出一种基于多LLM两两比较与Elo评分的自动化评估框架,可近似专家评估,减少人工干预,实现对LLM输出的高效可靠评估。

Comments 17 pages. Published in the Proceedings of the Fourth Workshop on Generation, Evaluation and Metrics (GEM2) at ACL 2025

Journal ref Proceedings of the Fourth Workshop on Generation, Evaluation and Metrics (GEM2), pages 320-336, Association for Computational Linguistics, 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.28263 2026-07-31 cs.CL 新提交

Understanding Is Done Early: A Depth Division of Labor in Large Language Models and Its Use for Unbounded-Context Memory

理解在早期完成:大语言模型的深度分工及其在无界上下文记忆中的应用

Hanzuo Liu, Xuan Qi, Chunyu Liu, Haotian Zhong, Yulong Wang, Rayying, Key, Alex Lamb, Mingyu Gao

机构 * Tsinghua University(清华大学) Tencent(腾讯)

AI总结 该研究提出CoMem方法,利用大语言模型的深度分工构建无界上下文记忆,在RULER、LoCoMo等基准上性能优于全上下文KV-Direct,内存占用低、预填充速度快,证明长上下文记忆可沿层轴组织。

Comments 19 pages, 4 figures, 27 tables. Submitted to ACL Rolling Review

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.27379 2026-07-31 cs.CL 新提交

HSS-Synth: Humanities and Social Sciences Data Synthesis for LLMs

HSS-Synth:面向大语言模型的人文社科数据合成

Ru Peng, Tianyu Zhao, Xijun Gu, Zhiting Fan, Haokai Xu, Jinyang Zhang, Yawen Zeng, Yihong Zhuang, Kexin Yang, Junyang Lin, Dayiheng Liu, Junbo Zhao

机构 * Zhejiang University(浙江大学) Inclusion AI, Ant Group(蚂蚁集团Inclusion AI) Peking University(北京大学) Qwen Team, Alibaba Group(阿里巴巴集团通义千问团队)

AI总结 本文针对LLM的HSS数据稀缺问题,提出以学科为中心的HSS-Synth流水线,生成23.7万指令微调样本,使Qwen3-8B-Base达SOTA并提升相关能力。

Comments ACL Findings 2026 Paper

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.00086 2026-07-31 cs.CL cs.AI 版本更新

NorBERTo: A ModernBERT Model Trained for Portuguese with 331 Billion Tokens Corpus

NorBERTo:一个基于ModernBERT架构的葡萄牙语现代编码器模型,训练于331十亿个token语料库

Enzo S. N. Silva, Pablo B. Costa, Raphael C. Vlasman, Rosimeire P. Costa, Henrique L. P. Silva, Lucas F. A. O. Pellicer, Guilherme Rinaldo, Renato A. Almeida, Darian S. R. Rabbani, Cinthya O. Oestreich, Vinicius F. Caridá

机构 * Itaú Unibanco ICTi

AI总结 NorBERTo基于ModernBERT架构,利用Aurora-PT语料库训练,具备长上下文支持和高效注意力机制,在葡萄牙语语义相似性、文本蕴含和分类任务中表现优异,达到最高F1和准确率。

Comments This article has already undergone formal submission, review, acceptance, and publication in the proceedings of PROPOR 2026: Proceedings of the 17th International Conference on Computational Processing of Portuguese, Vol. 1. The published version is available in the ACL Anthology at https://aclanthology.org/2026.propor-1.18/ 11 pages, 9 tables, 2 figures

Journal ref Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1

详情

展开后加载摘要…

URL PDF HTML 收藏
2507.18046 2026-07-31 cs.CV cs.AI 版本更新

Enhancing Scene Transition Awareness in Video Generation via Post-Training

通过后训练增强视频生成中的场景转换意识

Hanwen Shen, Jiajie Lu, Yupeng Cao, Xiaonan Yang

机构 * Stevens Institute of Technology(史蒂文斯理工学院)

AI总结 本文提出TAV数据集,通过后训练提升视频生成中场景转换的理解能力,改善多场景生成效果并保持图像质量。

Journal ref Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (2025) 706-721

详情

展开后加载摘要…

URL PDF HTML 收藏
2601.06599 2026-07-31 cs.CL cs.AI 版本更新

How Context Shapes Truth: Geometric Transformations of Statement-level Truth Representations in LLMs

上下文如何塑造真相:LLMs中语句级真相表示的几何变换

Shivam Adarsh, Maria Maistro, Christina Lioma

机构 * University of Copenhagen(哥本哈根大学)

AI总结 研究LLMs中上下文如何改变真相向量,发现早期层正交、中层收敛,上下文增加向量幅度,大模型通过方向变化区分相关与无关上下文。

Comments ACL 2026 (Main)

详情

展开后加载摘要…

URL PDF HTML 收藏