iiinova — where your digital self keeps evolving

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Bingchen Yao, Haobo Xu, Haokun Lin, Yichen Wu, Ziyu Guo, Renrui Zhang, Zhichao Lu, Zhenan Sun, Ying Wei: STEPQuant: When and Where Errors Matter in Delta-Rule Recurrent State Quantization arxiv.org/abs/2609.38169 arxiv.org/pdf/2609.38169 arxiv.org/html/2609.38169

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Dor Tirosh, Ido Amos, Mor Geva: Pretraining Latent Information Feedback Transformers with Teacher Supervision arxiv.org/abs/2609.38149 arxiv.org/pdf/2609.38149 arxiv.org/html/2609.38149

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Quang Hieu Pham, Thuy Duong Nguyen, Jocelyn Qiaochu Chen, Xi Ye: LongHarness Bench: Stress-Testing Language Model Harnesses for Long-Context Reasoning arxiv.org/abs/2609.38137 arxiv.org/pdf/2609.38137 arxiv.org/html/2609.38137

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Hongzhu Guo, Mohsen Fayyaz, Nanyun Peng: From Routing Signals to Selective Review: Visual regrounding in MoE VLMs arxiv.org/abs/2609.38111 arxiv.org/pdf/2609.38111 arxiv.org/html/2609.38111

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Cutter Dawes, Nick Alonso, Tom Figliolia, Beren Millidge: How Local Mixing Encodes Relative Position in Global NoPE Attention arxiv.org/abs/2609.38109 arxiv.org/pdf/2609.38109 arxiv.org/html/2609.38109

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Puduppully, Misra, Iyer, Kalwar, Palod, Kambhampati: Correct Answers, Invalid Traces: What Verifiable Grade-School Math Reveals About Chain-of-Thought Traces arxiv.org/abs/2609.38107 arxiv.org/pdf/2609.38107 arxiv.org/html/2609.38107

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Edoardo Bolzoni, Valerio Capraro: Gender bias across LLMs is common and highly heterogeneous arxiv.org/abs/2609.38036 arxiv.org/pdf/2609.38036 arxiv.org/html/2609.38036

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Junning Shao, Siwei Wang, Zhixuan Fang: Layer-Informed Fine-Tuning via Three-Stage Functional Segmentation of LLMs arxiv.org/abs/2609.38027 arxiv.org/pdf/2609.38027 arxiv.org/html/2609.38027

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Christopher J. Chanhnourack: Auditable Long-Term Memory: A Deterministic Retrieval Chain Measured at 479/475 of 500 on LongMemEval-S arxiv.org/abs/2609.38021 arxiv.org/pdf/2609.38021 arxiv.org/html/2609.38021

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Julien Knafou, Luc Mottin, Alexandre Flament, Paul van Rijen, Esteban Gaillac, Patrick Ruch: BITEM at the NTCIR-19 R2C2 Task: Predicting Confidence from Agentic RAG Pipeline Signals arxiv.org/abs/2609.37993 arxiv.org/pdf/2609.37993 arxiv.org/html/2609.37993

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Jiaming Tang, Mingyan Liu, Armin Sarabi: Learning What to Remember: Long-horizon Counterfactual Memory Optimization arxiv.org/abs/2609.37930 arxiv.org/pdf/2609.37930 arxiv.org/html/2609.37930

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Gon\c{c}alo Paulo, Louis Jaburi, Nora Belrose, Lucia Quirke, Stella Biderman: The Unequal Influence of Bad Advice: Using Training Data Attribution to Modulate Emergent Misalignment arxiv.org/abs/2609.37914 arxiv.org/pdf/2609.37914 arxiv.org/html/2609.37914

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Langlais, Delobelle, Detrois, Chizhov, Rosas-Hinostroza, Smail, Burtin, Shcharbakova, Yamshchikov, Stasenko: It's All Training: A Fully Synthetic Single-Stage Recipe for LLMs arxiv.org/abs/2609.37891 arxiv.org/pdf/2609.37891 arxiv.org/html/2609.37891

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Lalita Lowphansirikul, Attapol Rutherford, Jian Gang Ngui, Sarana Nutanong, Peerat Limkonchotiwat: Zero-shot Dependency Parsing with Unsupervised Cross-Lingual Bootstrapping arxiv.org/abs/2609.37883 arxiv.org/pdf/2609.37883 arxiv.org/html/2609.37883

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Bhanu Prakash Vangala, Sowmya Guda, Navya Vangala: How Many Labels Does a Language Need? Annotation Budgets and Cross-Lingual Pooling for African-Language Text Classification arxiv.org/abs/2609.37882 arxiv.org/pdf/2609.37882 arxiv.org/html/2609.37882

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Timur Mudarisov, Mikhail Burtsev, Radu State: Retrieval Capacity of Self-Attention Under Competition arxiv.org/abs/2609.37879 arxiv.org/pdf/2609.37879 arxiv.org/html/2609.37879

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Nagham Omar, Mahmoud Jabarin, Kinan Ibraheem, Lotem Peled-Cohen: It's Not What the Image Shows: Irrelevant Context Destabilises VLM Judges Without Informing Them arxiv.org/abs/2609.37863 arxiv.org/pdf/2609.37863 arxiv.org/html/2609.37863

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Bhanu Prakash Vangala, Vangala Navya: One Threshold Does Not Fit All Languages: Language-Conditional Deferral for Reliable and Efficient Low-Resource Text Classification arxiv.org/abs/2609.37861 arxiv.org/pdf/2609.37861 arxiv.org/html/2609.37861

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Liu, Chen, Song, Yuan, Zhang, Zhuotong, Zhou, Feng, Hu, Wang, Yang, Liu, Li, Wang, Ji, Wang: AnthroDial: Benchmarking LLM Anthropomorphism in Autonomous Social Interaction arxiv.org/abs/2609.37853 arxiv.org/pdf/2609.37853 arxiv.org/html/2609.37853

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Deng, Xing, Xu, He, Zheng, Li, Zhu, Li: Can Vision-Language Models Stay Helpful When Facing Implicit Risks? Intent-Privilege OPSD for Efficient Safety-Helpfulness Alignment arxiv.org/abs/2609.37837 arxiv.org/pdf/2609.37837 arxiv.org/html/2609.37837

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Timur Mudarisov, Mikhail Burtsev, Radu State: The Geometry of Inference in Transformer Residual Streams arxiv.org/abs/2609.37824 arxiv.org/pdf/2609.37824 arxiv.org/html/2609.37824

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Cai, Wang, Xie, Zhang, Lei, Cao, Sun, Wu: Thinking in Depth, Speaking Directly: Recurrent Latent Reasoning for Paralinguistically Grounded Spoken Dialogue arxiv.org/abs/2609.37818 arxiv.org/pdf/2609.37818 arxiv.org/html/2609.37818

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Philipp E. Glass, Alina Miron: CompOrca: Corpus-Scale Compliance Labelling of Instruction-Tuning Data arxiv.org/abs/2609.37807 arxiv.org/pdf/2609.37807 arxiv.org/html/2609.37807

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Zhangshu Joshua Jiang, Zina Ibrahim, James T. Teo: A Proposed Rubric for Evaluating Expressed Clinical Reasoning in Large Language Model Responses arxiv.org/abs/2609.37788 arxiv.org/pdf/2609.37788 arxiv.org/html/2609.37788

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Zifeng Cheng, Jie Zheng, Zhiwei Jiang, Shuwen Wang, Fei Shen, Shiping Ge, Qing Gu: Selecting What Matters: Semantic Compression-Guided Selective Pooling for Long-Context Embeddings arxiv.org/abs/2609.37782 arxiv.org/pdf/2609.37782 arxiv.org/html/2609.37782

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Anton Repushko, Elena Chepel: Which papyrus HTR is good enough? Character-error-rate tolerance of four papyrological tasks on Greek texts arxiv.org/abs/2609.37755 arxiv.org/pdf/2609.37755 arxiv.org/html/2609.37755

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Aaron Steiner, Ksenia Elagin, Ralph Peeters, Johannes Knopp, Christian Bizer: Billiger.de Products: A Bilingual Entity Matching Benchmark arxiv.org/abs/2609.37713 arxiv.org/pdf/2609.37713 arxiv.org/html/2609.37713

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Akio Hayakawa, Horacio Saggion: Reader Proficiency Shapes Layer-wise Surprisal Profiles arxiv.org/abs/2609.37688 arxiv.org/pdf/2609.37688 arxiv.org/html/2609.37688

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Baoxian Liu, Tong Wei: Corpus-Guided Dual-Path Propagation for Graph Retrieval-Augmented Generation arxiv.org/abs/2609.37661 arxiv.org/pdf/2609.37661 arxiv.org/html/2609.37661

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Tobias Deu{\ss}er, Lorenz Sparrenberg, Rafet Sifa: Evaluating and Benchmarking the System One Model Jev arxiv.org/abs/2609.37647 arxiv.org/pdf/2609.37647 arxiv.org/html/2609.37647

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Emmanuel Chemla, Benjamin Spector, Alexandros Kalomoiros, Philippe Schlenker: Co-Linguistics: AI-augmented Theory Construction in Linguistics arxiv.org/abs/2609.37635 arxiv.org/pdf/2609.37635 arxiv.org/html/2609.37635

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Jacob Epifano: Correct, Don't Delete: Mitigating Emergent Misalignment with Corrective Supervision arxiv.org/abs/2609.37624 arxiv.org/pdf/2609.37624 arxiv.org/html/2609.37624

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Bruno Brocai, Maria Becker: Pair Difficulty Matters: Rethinking Pairwise LLM-as-a-Judge Evaluation and Consistency arxiv.org/abs/2609.37577 arxiv.org/pdf/2609.37577 arxiv.org/html/2609.37577

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Zhang, Zhang, Bai, Zhang, Xiang, Chen: Devils in Question Relay: Source-Conditioned Relay Steering to Mitigate Hallucinations in Audio-visual Large Language Models arxiv.org/abs/2609.37568 arxiv.org/pdf/2609.37568 arxiv.org/html/2609.37568

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Wang, Liu, Wang, Nie, Zhao, Ma, Liu, Wang, Feng, Wang, Sch\"utze: Orthogonal Yet Coupled: Decoupling Geometric Components for Model Merging arxiv.org/abs/2609.37564 arxiv.org/pdf/2609.37564 arxiv.org/html/2609.37564

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Prosper Arineitwe Asiimwe, Francois Meyer, Jan Buys: RunyaNER: Auxiliary Language Selection for Runyankore NER arxiv.org/abs/2609.37543 arxiv.org/pdf/2609.37543 arxiv.org/html/2609.37543

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Arseny Ivanov, Alexander Kolesov, Alexander Korotin, Ivan Oseledets, Mikhail Goncharov: E-MoE: Enhanced Mixture-of-Experts for Non-Factorized Diffusion Language Models arxiv.org/abs/2609.37533 arxiv.org/pdf/2609.37533 arxiv.org/html/2609.37533

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Yuhao Wang, Ruiyang Ren, Yinan Zhang, Ruiqing Zhang, Jing Liu, Chunyan Miao: From Dissonance to Orchestration: Teacher Intervention in On-Policy Distillation arxiv.org/abs/2609.37510 arxiv.org/pdf/2609.37510 arxiv.org/html/2609.37510

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Dipankar Sarkar: Evaluating Bounded Autonomy in Regulated Agentic AI: A Diagnostic Harness with Constitutional Rewards, Escalation Labels, and Runtime Governance arxiv.org/abs/2609.37501 arxiv.org/pdf/2609.37501 arxiv.org/html/2609.37501

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Claudiu Creanga, Liviu P. Dinu: Who Warmed the Archives? LLMs Overestimate Historical Warmth arxiv.org/abs/2609.37499 arxiv.org/pdf/2609.37499 arxiv.org/html/2609.37499

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Claudiu Creanga, Liviu P. Dinu, Anca Dinu: The Rashomon Wikipedia: A Data-Perspectivist Analysis of Divergent Historical Narratives arxiv.org/abs/2609.37498 arxiv.org/pdf/2609.37498 arxiv.org/html/2609.37498

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Stefania Butnaru, Claudiu Creanga: Larry Caused the Car to Stop, But the Model Didn't Notice: Transformer Blindness to the M-Heuristic arxiv.org/abs/2609.37497 arxiv.org/pdf/2609.37497 arxiv.org/html/2609.37497

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Mohamed Eltahir, Abobaker Ahmed, Nawaf Barebood, Hussain Bu Subayt, Tanveer Hussain, Naeemullah Khan: Your Benchmark Is Not Saturated: Reviving Multiple-Choice Evaluation with Answer Pooling arxiv.org/abs/2609.37494 arxiv.org/pdf/2609.37494 arxiv.org/html/2609.37494

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Zeyan Li, Qirong Guo, SIyuan Qiu, Hu Xu, Chun Li, Jianfeng Xu: Regime Boundary Alignment for Evidence-Gated Question Answering arxiv.org/abs/2609.37491 arxiv.org/pdf/2609.37491 arxiv.org/html/2609.37491

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Suting Chen, Peichun Hua, Yunming Xiao: Relevance Is Not Sufficient Evidence: Detecting Evidence Gaps Before Generation in RAG arxiv.org/abs/2609.37469 arxiv.org/pdf/2609.37469 arxiv.org/html/2609.37469

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Yi-Xuan Deng, Yi Zhang, Wei Liu, Chao Xue, Shuojin Yang: Learning to Retrieve Missing Evidence for Long-Term Memory QA arxiv.org/abs/2609.37443 arxiv.org/pdf/2609.37443 arxiv.org/html/2609.37443

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Chua, Khoo, Tan, Ang, Tan, Sim, Ow, Mundhra, Toh, Xu, Ng: Look What You Made Us Cluster: Hate Narrative Extraction from Reddit Discourse arxiv.org/abs/2609.37408 arxiv.org/pdf/2609.37408 arxiv.org/html/2609.37408

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Rebecca Ramnauth, Brian Scassellati: Compiling Learning Problems into Adaptation Programs for Language Models arxiv.org/abs/2609.37371 arxiv.org/pdf/2609.37371 arxiv.org/html/2609.37371

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Zetong Zhou, Wentao Zhang, Jingyuan Wang, Yifan Yang, Zizhuo Wang, Shixi Hu: SemOPT: Fixing Semantic Errors in LLM-based Optimization Modeling via Reward-Guided Search arxiv.org/abs/2609.37361 arxiv.org/pdf/2609.37361 arxiv.org/html/2609.37361

000

arXiv cs.CL Computation and Language @cscl-bot.bsky.social · 3h

Soyeong Jeong, Sujay Kumar Jauhar, Sung Ju Hwang, Andrew Joohun Nam: Follow the Entities: A Corpus Map for Agentic Search arxiv.org/abs/2609.37226 arxiv.org/pdf/2609.37226 arxiv.org/html/2609.37226

000