←── back to feed
/topics/arxiv-computational-linguistics-papers-september-23

arXiv computational linguistics papers September 23

31 items●1 sources●updated 46d ago●trend 0

On September 23, 2026, arXiv published 20 computational linguistics papers spanning fake news corpus audits, Rust-based language model training, numerical representation invariance in LLMs, semantic change in Sanskrit, meeting summarization, monetary policy sentiment analysis, peer-review verification, autonomous AI model development, on-policy distillation, LLM over-refusal mitigation, open mathematics benchmarks, POI fusion, agent memory systems, multimodal document reasoning, word deduction games, parameter-efficient fine-tuning, LLM-based human behavior simulation, and a 151M-parameter Turkish language model.

  • ISOT/Kaggle fake news corpus audit reveals classifiers achieve 98%+ accuracy partly through subject-field leakage, not article text.
  • Language model pretrained end-to-end in Rust for $164 GPU cost; documents five Candle and three Burn framework defects.
  • Numerical representation invariance test on 3,600 problems shows canonical accuracy 0.969–0.996 but orbit invariance drops to 0.848–0.981.
  • FrontierMath Erdős benchmark contains 68 open Erdős conjectures formalized in Lean for systematic AI capability evaluation.
  • ufakzeka-1: 151M-parameter Turkish LLM trained from scratch on 13.5B tokens for $286, achieving 1.77 tokens/word.
  • LatentPort demonstrates cross-model recurrent state transfer between Qwen3.5 4B and 9B without target prefix replay.
[BLG]blog/rss31
What Does 99% Accuracy Measure? A Reproducible Audit of Shortcut Learning in a Widely Used Fake News Corpus
arXiv cs.CL · Yuvraj Verma · 46d
Training a Language Model End-to-End in Rust: An Experience Report
arXiv cs.CL · Arif Adito · 46d
Same Quantity, Different Answer: Numerical Representation Invariance in Language Models
arXiv cs.CL · Ephraim Atta-Duncan · 46d
A Computational Approach to Measuring Semantic Change in Sanskrit Literature
arXiv cs.CL · Tanay Agrawal · 46d
Retrieved-Span Training for Efficient Query-Focused Meeting Summarization on QMSum
arXiv cs.CL · Edward Xi Yang (Ertas AI) · 46d
From Tone to Trajectory: Continuous Sentiment and the Shape of Monetary Policy Communication
arXiv cs.CL · Martin Feldkircher, M\'arton Kardos, Kristoffer Laigaard Nielbo · 46d
Peerify: Benchmarking Peer-Review Claim Verification
arXiv cs.CL · Alireza Daghighfarsoodeh, Sajad Ebrahimi, Ali Ghorbanpour, Soroush Sadeghian, Radin Cheraghi, Negar Arabzadeh, Ebrahim Bagheri · 46d
AIBuildAI-2.5: Efficient Autonomous AI Model Development Through LLM-Guided Tree Search
arXiv cs.CL · Peijia Qin, Ruiyi Zhang, Qi Cao, Han Guo, Li Zhang, Pengtao Xie · 46d
Prompt Breadth and Rollout Refresh Interact in On-Policy Distillation
arXiv cs.CL · Lingxiang Hu, Tianle Xia, Ming Xu, Yiding Sun, Linfang Shang · 46d
Mitigating LLM Over-Refusal via Dynamic Semantic Routing Calibratione
arXiv cs.CL · Zixuan Wang, Bingjie Zhang, He Zhao, Dandan Guo · 46d
FrontierMath Erd\H{o}s
arXiv cs.CL · Tom Adamczewski (Epoch AI), Thomas F. Bloom (University of Manchester) · 46d
LLM-Driven Training-free Location-Attribute Synergic Fusion: A Closed-Loop Paradigm for Dual-source Encrypted POIs and LULC Mapping
arXiv cs.CL · Chang Li, Xingtao Peng, Yongjun Zhang, Yinfei He, Cairun Huang · 46d
Self-Cleaning and Captured Anyway: One Measured Primitive for Error in a Store an Agent Writes to Itself, and What a Falling Score Actually Measures
arXiv cs.CL · Wenhui Chen, Jianlin Chen, Ziyao Lin, Chi Man Vong · 46d
LatentPort: Beyond KV Cache - Cross-Model Transfer of Recurrent Memory in Hybrid Language Models: A 4B-to-9B Hybrid-State Handoff Without Target Prefix Replay
arXiv cs.CL · Simon P. Villani · 46d
MoM: Memory of Memory
arXiv cs.CL · Bowen Qin, Yao Lu · 46d
ICDAR2026 Competition on Multimodal Reasoning over Documents in Multiple Domains
arXiv cs.CL · Artemis Llabr\'es, Marc Serra Ortega, Tom\`as Ockier, Samuel Ortega Cuadra, Amritpal Singh, Christos Georgakilas, Andrey Barsky, Ernest Valveny, Dimosthenis Karatzas · 46d
Graph-Based Inference for Feedback-Driven Word Deduction: A Scalable Framework for the Jotto Problem
arXiv cs.CL · Dakshi Arora, Prakhar Kumar Srivastava, Ranjib Banerjee · 46d
ChainDoRA: Tensor-Train Factorized Weight-Decomposed Low-Rank Adaptation for Parameter-Efficient LLM Fine-Tuning
arXiv cs.CL · Ashfak Yeafi, Mehedi Hasan, Md Khairul Islam · 46d
Understanding Reliability in LLM-based Human Behavior Simulation
arXiv cs.CL · Pei Wang, Lei Wang, Yuanzi Li, Xu Chen · 46d
ufakzeka-1: Building and Evaluating a 151M-Parameter Turkish Language Model from Scratch
arXiv cs.CL · Sait Furkan Teke (ufak AI) · 46d
Impact Is Not Invalidation: Ask About the Claim, Not the Diff
arXiv cs.CL · Atul Anand · 46d
FinFIRST: Benchmarking Search Agents for Financial Information Retrieval, Sourcing and Traceability
arXiv cs.CL · Wenqing Wang, Haitao Xiang, Xinyi Zhao, Mingming Yin, Ying Zhong, Zhaoxin Huan, Qiheng Zhou, Jin Zhu, Xiaolu Zhang, Shi Chang, Jun Zhou · 46d
FineWeb-CLaR: Culture, Language, and Region Annotations for Benchmark-Aligned Corpus Auditing
arXiv cs.CL · Yusser Al Ghussin, Eva Gavaller, Cristina Espa\~na-Bonet, Josef van Genabith, Simon Ostermann · 46d
TelecomGPT-R1: Unified Post-Training for Reasoning Across Heterogeneous Telecom Tasks
arXiv cs.CL · Bohao Wang, Chenwei Wu, Hang Zou, Yu Tian, Lina Bariah, Li Wei, Chongwen Huang, Yongliang Shen, Zhaoyang Zhang, Merouane Debbah · 46d
Passes Alone, Fails Together: Benchmarking Semantic Coordination in Parallel LLM-Agent Development
arXiv cs.CL · Haocheng Xia, Eugene Wu, Yongjoo Park · 46d
Mining Legal Arguments in U.S. Corporate Case Law
arXiv cs.CL · Luis Brena, William Jurayj, Gregory Deyesu, Zaid Al-Huneidi, Andrew Blair-Stanek, Benjamin Van Durme · 46d
Conduct Under Pressure: What Sixty Language Models Do When a User Pushes
arXiv cs.CL · Tapan Parikh · 46d
Matryoshka attribution: Learning to attribute language model outputs to representations and weights
arXiv cs.CL · Aryaman Arora, Kirill Acharya, Nathan Hu, Yanzhe Zhang, Noah Goodman, Dan Jurafsky, Christopher Potts · 46d
Compressing Long Context into Answer-Aligned Memory Embeddings for LLM Inference
arXiv cs.CL · Md Mostafizer Rahman, Md Faizul Ibne Amin, Md Shahajada Mia, Yutaka Watanobe, Fang Liu · 46d
Qwen3.8-Omni: Towards Native Omni-Modal Agents
arXiv cs.CL · Qwen Team · 46d
From Utterances to Networks: Modelling Slang Adoption and Diffusion Across Subreddits
arXiv cs.CL · Xiaoning Wang, Ted Underwood, Zhewei Sun · 46d