Peer Revenue Play credits. Real papers.

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

Computer Science Computation and Language

cs.CL Watch Computation and Language

Tuesday 30 December

Computation and Language

Tracing the Flow of Knowledge From Science to Technology Using Deep Learning

Michael E. Rose, Mainak Ghosh, Sebastian Erhardt et al. (+3 more)

We develop a language similarity model suitable for working with patents and scientific publications at the same time. In a horse race-style evaluation, we subject eight language (similarity) models to predict credible Patent-Paper Citations. We find that our Pat-SPECTER model performs best, which is the SPECTER2 model fine-tuned on patents. In two real-world scenarios (separating patent-paper-pairs and predicting patent-paper-pairs) we demonstrate the capabilities of the Pat-SPECTER. We finally test the hypothesis that US patents cite papers that are semantically less similar than in other large jurisdictions, which we posit is because of the duty of candor. The model is open for the academic community and practitioners alike.
Read the full abstract

All authors Michael E. Rose, Mainak Ghosh, Sebastian Erhardt, Cheng Li, Erik Buunk, Dietmar Harhoff

Computation and Language

LAILA: A Large Trait-Based Dataset for Arabic Automated Essay Scoring

May Bashendy, Walid Massoud, Sohaila Eltanbouly et al. (+5 more)

Automated Essay Scoring (AES) has gained increasing attention in recent years, yet research on Arabic AES remains limited due to the lack of publicly available datasets. To address this, we introduce LAILA, the largest publicly available Arabic AES dataset to date, comprising 7,859 essays annotated with holistic and trait-specific scores on seven dimensions: relevance, organization, vocabulary, style, development, mechanics, and grammar. We detail the dataset design, collection, and annotations, and provide benchmark results using state-of-the-art Arabic and English models in prompt-specific and cross-prompt settings. LAILA fills a critical need in Arabic AES research, supporting the development of robust scoring systems.
Read the full abstract

All authors May Bashendy, Walid Massoud, Sohaila Eltanbouly, Salam Albatarni, Marwan Sayed, Abrar Abir, Houda Bouamor, Tamer Elsayed

Computation and Language

MedKGI: Iterative Differential Diagnosis with Medical Knowledge Graphs and Information-Guided Inquiring

Qipeng Wang, Rui Sheng, Yafei Li et al. (+3 more)

Recent advancements in Large Language Models (LLMs) have demonstrated significant promise in clinical diagnosis. However, current models struggle to emulate the iterative, diagnostic hypothesis-driven reasoning of real clinical scenarios. Specifically, current LLMs suffer from three critical limitations: (1) generating hallucinated medical content due to weak grounding in verified knowledge, (2) asking redundant or inefficient questions rather than discriminative ones that hinder diagnostic progress, and (3) losing coherence over multi-turn dialogues, leading to contradictory or inconsistent conclusions. To address these challenges, we propose MedKGI, a diagnostic framework grounded in clinical practices. MedKGI integrates a medical knowledge graph (KG) to constrain reasoning to validated medical ontologies, selects questions based on information gain to maximize diagnostic efficiency, and adopts an OSCE-format structured state to maintain consistent evidence tracking across turns. Experiments on clinical benchmarks show that MedKGI outperforms strong LLM baselines in both diagnostic accuracy and inquiry efficiency, improving dialogue efficiency by 30% on average while maintaining state-of-the-art accuracy.
Read the full abstract

All authors Qipeng Wang, Rui Sheng, Yafei Li, Huamin Qu, Yushi Sun, Min Zhu

Computation and Language

Training Report of TeleChat3-MoE

Xinzhang Liu, Chao Wang, Zhihao Yang et al. (+51 more)

TeleChat3-MoE is the latest series of TeleChat large language models, featuring a Mixture-of-Experts (MoE) architecture with parameter counts ranging from 105 billion to over one trillion,trained end-to-end on Ascend NPU cluster. This technical report mainly presents the underlying training infrastructure that enables reliable and efficient scaling to frontier model sizes. We detail systematic methodologies for operator-level and end-to-end numerical accuracy verification, ensuring consistency across hardware platforms and distributed parallelism strategies. Furthermore, we introduce a suite of performance optimizations, including interleaved pipeline scheduling, attention-aware data scheduling for long-sequence training,hierarchical and overlapped communication for expert parallelism, and DVM-based operator fusion. A systematic parallelization framework, leveraging analytical estimation and integer linear programming, is also proposed to optimize multi-dimensional parallelism configurations. Additionally, we present methodological approaches to cluster-level optimizations, addressing host- and device-bound bottlenecks during large-scale training tasks. These infrastructure advancements yield significant throughput improvements and near-linear scaling on clusters comprising thousands of devices, providing a robust foundation for large-scale language model development on hardware ecosystems.
Read the full abstract

All authors Xinzhang Liu, Chao Wang, Zhihao Yang, Zhuo Jiang, Xuncheng Zhao, Haoran Wang, Lei Li, Dongdong He, Luobin Liu, Kaizhe Yuan, Han Gao, Zihan Wang, Yitong Yao, Sishi Xiong, Wenmin Deng, Haowei He, Kaidong Yu, Yu Zhao, Ruiyu Fang, Yuhao Jiang, Yingyan Li, Xiaohui Hu, Xi Yu, Jingqi Li, Yanwei Liu, Qingli Li, Xinyu Shi, Junhao Niu, Chengnuo Huang, Yao Xiao, Ruiwen Wang, Fengkai Li, Luwen Pu, Kaipeng Jia, Fubei Yao, Yuyao Huang, Xuewei He, Zhuoru Jiang, Ruiting Song, Rui Xue, Qiyi Xie, Jie Zhang, Zilu Huang, Zhaoxi Zhang, Zhilong Lu, Yanhan Zhang, Yin Zhang, Yanlei Xue, Zhu Yuan, Teng Su, Xin Jiang, Shuangyong Song, Yongxiang Li, Xuelong Li

Computation and Language

Large Emotional World Model

Changhao Song, Yazhou Zhang, Hui Gao et al. (+2 more)

The world is governed by both physical laws and affective dynamics. Physical laws govern state transitions, while affective dynamics shape human actions, decisions, and interactions. A world model that learns only physical laws can approximate the physical world, but not the human world. In this paper, we introduce human emotion as a key state variable in world models, enabling them to capture both future state transitions and their emotional causes. We first construct Emotion-Why-How (EWH), the first world model dataset centered on emotional state transitions, containing 10,850 emotion-aware transition tuples. Each tuple encodes the pre-state, pre-emotion, action, post-emotion, and post-state, supporting reasoning about why actions occur and how emotions reshape future states. Based on EWH, we propose the Large Emotional World Model (LEWM), which factorizes future prediction into two coupled steps: first predicting the future emotional state from the current context, and then conditioning future world-state prediction on the predicted emotion. Experiments show that LEWM brings consistent gains across world-state prediction, emotion understanding, and general reasoning tasks. It achieves up to 45.72% accuracy improvement on EWH, 3.94% on WorldNet, 17.47% F1 improvement on MELD, and a 6.10% gain on specific MMLU categories. These results demonstrate that incorporating emotion into world models enables more realistic simulation of human-centered environments and expands the predictive understanding of intelligent agents.
Read the full abstract

All authors Changhao Song, Yazhou Zhang, Hui Gao, Chang Yang, Peng Zhang

Computation and Language

Activation Steering for Masked Diffusion Language Models

Adi Shnaidman, Erin Feiglin, Osher Yaari et al. (+3 more)

Masked diffusion language models (MDLMs) generate text via iterative masked-token denoising, enabling mask-parallel decoding and distinct controllability and efficiency tradeoffs from autoregressive LLMs. Yet, efficient representation-level mechanisms for inference-time control in MDLMs remain largely unexplored. To address this gap, we introduce an activation steering primitive for MDLMs: we extract a single low-dimensional direction from contrastive prompt sets using one prompt-only forward pass, and apply a global intervention on residual-stream activations throughout reverse diffusion, without performing optimization or altering the diffusion sampling procedure. Using safety refusal as a deployment-relevant case study, we find that refusal behavior in multiple MDLMs is governed by a consistent, approximately one-dimensional activation subspace. Applying the corresponding direction yields large and systematic behavioral shifts and is substantially more effective than prompt-based and optimization-based baselines. We further uncover diffusion-specific accessibility: effective directions can be extracted not only from post-instruction tokens, but also from pre-instruction tokens that are typically ineffective in autoregressive models due to causal attention. Ablations localize maximal leverage to early denoising steps and mid-to-late transformer layers, with early diffusion blocks contributing disproportionately. Finally, in an MDLM trained on English and Chinese, extracted directions transfer strongly between English and Chinese, but do not reliably generalize to an autoregressive architecture, highlighting architecture-dependent representations of safety constraints.
Read the full abstract

All authors Adi Shnaidman, Erin Feiglin, Osher Yaari, Efrat Mentel, Amit Levi, Raz Lapid

Computation and Language · Cross-listed from cs.LG

OptRot: Mitigating Weight Outliers via Data-Free Rotations for Post-Training Quantization

Advait Gadhikar, Riccardo Grazzi and James Hensman

The presence of outliers in Large Language Models (LLMs) weights and activations makes them difficult to quantize. Recent work has leveraged rotations to mitigate these outliers. In this work, we propose methods that learn fusible rotations by minimizing principled and cheap proxy objectives to the weight quantization error. We primarily focus on GPTQ as the quantization method. Our main method is OptRot, which reduces weight outliers simply by minimizing the element-wise fourth power of the rotated weights. We show that OptRot outperforms both Hadamard rotations and more expensive, data-dependent methods like SpinQuant and OSTQuant for weight quantization. It also improves activation quantization in the W4A8 setting. We also propose a data-dependent method, OptRot$^{+}$, that further improves performance by incorporating information on the activation covariance. In the W4A4 setting, we see that both OptRot and OptRot$^{+}$ perform worse, highlighting a trade-off between weight and activation quantization.
Read the full abstract

Computation and Language

Training a Huggingface Model on AWS Sagemaker (Without Tears)

Liling Tan

The development of Large Language Models (LLMs) has primarily been driven by resource-rich research groups and industry partners. Due to the lack of on-premise computing resources required for increasingly complex models, many researchers are turning to cloud services like AWS SageMaker to train Hugging Face models. However, the steep learning curve of cloud platforms often presents a barrier for researchers accustomed to local environments. Existing documentation frequently leaves knowledge gaps, forcing users to seek fragmented information across the web. This demo paper aims to democratize cloud adoption by centralizing the essential information required for researchers to successfully train their first Hugging Face model on AWS SageMaker from scratch.
Read the full abstract

Computation and Language · Cross-listed from cs.CV

Factorized Learning for Temporally Grounded Video-Language Models

Wenzheng Zeng, Difei Gao, Mike Zheng Shou et al. (+1 more)

Recent video-language models have shown great potential for video understanding, but still struggle with accurate temporal grounding for event-level perception. We observe that two main factors in video understanding (i.e., temporal grounding and textual response) form a logical hierarchy: accurate temporal evidence grounding lays the foundation for reliable textual response. However, existing works typically handle these two tasks in a coupled manner without a clear logical structure, leading to sub-optimal objectives. We address this from a factorized learning perspective. We first propose D$^2$VLM, a framework that decouples the learning of these two tasks while also emphasizing their inherent dependency. We adopt a "grounding then answering with evidence referencing" paradigm and introduce evidence tokens for evidence grounding, which emphasize event-level visual semantic capture beyond the focus on timestamp representation in existing works. To further facilitate the learning of these two tasks, we introduce a novel factorized preference optimization (FPO) algorithm. Unlike standard preference optimization, FPO explicitly incorporates probabilistic temporal grounding modeling into the optimization objective, enabling preference learning for both temporal grounding and textual response. We also construct a synthetic dataset to address the lack of suitable datasets for factorized preference learning with explicit temporal grounding. Experiments on various tasks demonstrate the clear advantage of our approach. Our source code is available at https://github.com/nusnlp/d2vlm.
Read the full abstract

All authors Wenzheng Zeng, Difei Gao, Mike Zheng Shou, Hwee Tou Ng

Computation and Language

HY-MT1.5 Technical Report

Mao Zheng, Zheng Li, Tao Chen et al. (+2 more)

In this report, we introduce our latest translation models, HY-MT1.5-1.8B and HY-MT1.5-7B, a new family of machine translation models developed through a holistic training framework tailored for high-performance translation. Our methodology orchestrates a multi-stage pipeline that integrates general and MT-oriented pre-training, supervised fine-tuning, on-policy distillation, and reinforcement learning. HY-MT1.5-1.8B, the 1.8B-parameter model demonstrates remarkable parameter efficiency, comprehensively outperforming significantly larger open-source baselines (e.g., Tower-Plus-72B, Qwen3-32B) and mainstream commercial APIs (e.g., Microsoft Translator, Doubao Translator) in standard Chinese-foreign and English-foreign tasks. It achieves approximately 90% of the performance of ultra-large proprietary models such as Gemini-3.0-Pro, while marginally trailing Gemini-3.0-Pro on WMT25 and Mandarin-minority language benchmarks, it maintains a substantial lead over other competing models. Furthermore, HY-MT1.5-7B establishes a new state-of-the-art for its size class, achieving 95% of Gemini-3.0-Pro's performance on Flores-200 and surpassing it on the challenging WMT25 and Mandarin-minority language test sets. Beyond standard translation, the HY-MT1.5 series supports advanced constraints, including terminology intervention, context-aware translation, and format preservation. Extensive empirical evaluations confirm that both models offer highly competitive, robust solutions for general and specialized translation tasks within their respective parameter scales.
Read the full abstract

All authors Mao Zheng, Zheng Li, Tao Chen, Mingyang Song, Di Wang

Computation and Language

Beyond Hallucinations: A Composite Score for Measuring Reliability in Open-Source Large Language Models

Rohit Kumar Salla, Manoj Saravanan and Shrikar Reddy Kota

Large Language Models (LLMs) like LLaMA, Mistral, and Gemma are increasingly used in decision-critical domains such as healthcare, law, and finance, yet their reliability remains uncertain. They often make overconfident errors, degrade under input shifts, and lack clear uncertainty estimates. Existing evaluations are fragmented, addressing only isolated aspects. We introduce the Composite Reliability Score (CRS), a unified framework that integrates calibration, robustness, and uncertainty quantification into a single interpretable metric. Through experiments on ten leading open-source LLMs across five QA datasets, we assess performance under baselines, perturbations, and calibration methods. CRS delivers stable model rankings, uncovers hidden failure modes missed by single metrics, and highlights that the most dependable systems balance accuracy, robustness, and calibrated uncertainty.
Read the full abstract

Computation and Language · Cross-listed from cs.SD

AHA: Aligning Large Audio-Language Models for Reasoning Hallucinations via Counterfactual Hard Negatives

Yanxi Chen, Wenhui Zhu, Xiwen Chen et al. (+9 more)

Although Large Audio-Language Models (LALMs) deliver state-of-the-art (SOTA) performance, they frequently suffer from hallucinations, e.g. generating text not grounded in the audio input. We analyze these grounding failures and identify a distinct taxonomy: Event Omission, False Event Identity, Temporal Relation Error, and Quantitative Temporal Error. To address this, we introduce the AHA (Audio Hallucination Alignment) framework. By leveraging counterfactual hard negative mining, our pipeline constructs a high-quality preference dataset that forces models to distinguish strict acoustic evidence from linguistically plausible fabrications. Additionally, we establish AHA-Eval, a diagnostic benchmark designed to rigorously test these fine-grained temporal reasoning capabilities. We apply this data to align Qwen2.5-Omni. The resulting model, Qwen-Audio-AHA, achieves a 13.7% improvement on AHA-Eval. Crucially, this benefit generalizes beyond our diagnostic set. Our model shows substantial gains on public benchmarks, including 1.3% on MMAU-Test and 1.6% on MMAR, outperforming latest SOTA methods. The model and dataset are open-sourced at https://github.com/LLM-VLM-GSL/AHA.
Read the full abstract

All authors Yanxi Chen, Wenhui Zhu, Xiwen Chen, Zhipeng Wang, Xin Li, Peijie Qiu, Hao Wang, Xuanzhao Dong, Yujian Xiong, Anderson Schneider, Yuriy Nevmyvaka, Yalin Wang

Computation and Language · Cross-listed from cs.CR

Jailbreaking Attacks vs. Content Safety Filters: How Far Are We in the LLM Safety Arms Race?

Yuan Xin, Dingfan Chen, Linyi Yang et al. (+2 more)

As large language models (LLMs) are increasingly deployed, ensuring their safe use is paramount. Jailbreaking, adversarial prompts that bypass model alignment to trigger harmful outputs, present significant risks, with existing studies reporting high success rates in evading common LLMs. However, previous evaluations have focused solely on the models, neglecting the full deployment pipeline, which typically incorporates additional safety mechanisms like content moderation filters. To address this gap, we present the first systematic evaluation of jailbreak attacks targeting LLM safety alignment, assessing their success across the full inference pipeline, including both input and output filtering stages. Our findings yield two key insights: first, nearly all evaluated jailbreak techniques can be detected by at least one safety filter, suggesting that prior assessments may have overestimated the practical success of these attacks; second, while safety filters are effective in detection, there remains room to better balance recall and precision to further optimize protection and user experience. We highlight critical gaps and call for further refinement of detection accuracy and usability in LLM safety systems.
Read the full abstract

All authors Yuan Xin, Dingfan Chen, Linyi Yang, Michael Backes, Xiao Zhang

Computation and Language

iCLP: Large Language Model Reasoning with Implicit Cognition Latent Planning

Sijia Chen and Di Niu

Large language models (LLMs), when guided by explicit textual plans, can perform reliable step-by-step reasoning during problem-solving. However, generating accurate and effective textual plans remains challenging due to LLM hallucinations and the high diversity of task-specific questions. To address this, we draw inspiration from human Implicit Cognition (IC), the subconscious process by which decisions are guided by compact, generalized patterns learned from past experiences without requiring explicit verbalization. We propose iCLP, a novel framework that enables LLMs to adaptively generate latent plans (LPs), which are compact encodings of effective reasoning instructions. iCLP first distills explicit plans from existing step-by-step reasoning trajectories. It then learns discrete representations of these plans via a vector-quantized autoencoder coupled with a codebook. Finally, by fine-tuning LLMs on paired latent plans and corresponding reasoning steps, the models learn to perform implicit planning during reasoning. Experimental results on mathematical reasoning and code generation tasks demonstrate that, with iCLP, LLMs can plan in latent space while reasoning in language space. This approach yields significant improvements in both accuracy and efficiency and, crucially, demonstrates strong cross-domain generalization while preserving the interpretability of chain-of-thought reasoning.
Read the full abstract

Computation and Language

WISE: Web Information Satire and Fakeness Evaluation

Gaurab Chhetri, Subasish Das and Tausif Islam Chowdhury

Distinguishing fake or untrue news from satire or humor poses a unique challenge due to their overlapping linguistic features and divergent intent. This study develops WISE (Web Information Satire and Fakeness Evaluation) framework which benchmarks eight lightweight transformer models alongside two baseline models on a balanced dataset of 20,000 samples from Fakeddit, annotated as either fake news or satire. Using stratified 5-fold cross-validation, we evaluate models across comprehensive metrics including accuracy, precision, recall, F1-score, ROC-AUC, PR-AUC, MCC, Brier score, and Expected Calibration Error. Our evaluation reveals that MiniLM, a lightweight model, achieves the highest accuracy (87.58%) among all models, while RoBERTa-base achieves the highest ROC-AUC (95.42%) and strong accuracy (87.36%). DistilBERT offers an excellent efficiency-accuracy trade-off with 86.28% accuracy and 93.90% ROC-AUC. Statistical tests confirm significant performance differences between models, with paired t-tests and McNemar tests providing rigorous comparisons. Our findings highlight that lightweight models can match or exceed baseline performance, offering actionable insights for deploying misinformation detection systems in real-world, resource-constrained settings.
Read the full abstract

2512.24000 · 30 Dec · This is the author's preprint. Accepted to WEB&GRAPH 2026 (co-located with WSDM 2026), Boise, Idaho, USA, Feb 26, 2026. Final version will appear in WSDM 2026 Companion Proceedings. Conf: https://wsdm-conference.org/2026/ Workshop: https://aiimlab.org/events/WSDM_2026_WEB_and_GRAPH_2026_Workshop_on_Web_and_Graphs_Responsible_Intelligence_and_Social_Media.html

Abstract PDF $BNCH ×1$DATA ×1$TRFM ×1

Computation and Language

Fantastic Reasoning Behaviors and Where to Find Them: Unsupervised Discovery of the Reasoning Process

Zhenyu Zhang, Shujian Zhang, John Lambert et al. (+6 more)

Despite the growing reasoning capabilities of recent large language models (LLMs), their internal mechanisms during the reasoning process remain underexplored. Prior approaches often rely on human-defined concepts (e.g., overthinking, reflection) at the word level to analyze reasoning in a supervised manner. However, such methods are limited, as it is infeasible to capture the full spectrum of potential reasoning behaviors, many of which are difficult to define in token space. In this work, we propose an unsupervised framework (namely, RISE: Reasoning behavior Interpretability via Sparse auto-Encoder) for discovering reasoning vectors, which we define as directions in the activation space that encode distinct reasoning behaviors. By segmenting chain-of-thought traces into sentence-level 'steps' and training sparse auto-encoders (SAEs) on step-level activations, we uncover disentangled features corresponding to interpretable behaviors such as reflection and backtracking. Visualization and clustering analyses show that these behaviors occupy separable regions in the decoder column space. Moreover, targeted interventions on SAE-derived vectors can controllably amplify or suppress specific reasoning behaviors, altering inference trajectories without retraining. Beyond behavior-specific disentanglement, SAEs capture structural properties such as response length, revealing clusters of long versus short reasoning traces. More interestingly, SAEs enable the discovery of novel behaviors beyond human supervision. We demonstrate the ability to control response confidence by identifying confidence-related vectors in the SAE decoder space. These findings underscore the potential of unsupervised latent discovery for both interpreting and controllably steering reasoning in LLMs.
Read the full abstract

All authors Zhenyu Zhang, Shujian Zhang, John Lambert, Wenxuan Zhou, Zhangyang Wang, Mingqing Chen, Andrew Hard, Rajiv Mathews, Lun Wang

Computation and Language

CEC-Zero: Zero-Supervision Character Error Correction with Self-Generated Rewards

Zhiming Lin, Kai Zhao, Sophie Zhang et al. (+2 more)

Large-scale Chinese spelling correction (CSC) remains critical for real-world text processing, yet existing LLMs and supervised methods lack robustness to novel errors and rely on costly annotations. We introduce CEC-Zero, a zero-supervision reinforcement learning framework that addresses this by enabling LLMs to correct their own mistakes. CEC-Zero synthesizes errorful inputs from clean text, computes cluster-consensus rewards via semantic similarity and candidate agreement, and optimizes the policy with PPO. It outperforms supervised baselines by 10–13 F$_1$ points and strong LLM fine-tunes by 5–8 points across 9 benchmarks, with theoretical guarantees of unbiased rewards and convergence. CEC-Zero establishes a label-free paradigm for robust, scalable CSC, unlocking LLM potential in noisy text pipelines.
Read the full abstract

All authors Zhiming Lin, Kai Zhao, Sophie Zhang, Peilai Yu, Canran Xiao

Computation and Language

Efficient Context Scaling with LongCat ZigZag Attention

Chen Zhang, Yang Bai, Jiahuan Li et al. (+19 more)

We introduce LongCat ZigZag Attention (LoZA), which is a sparse attention scheme designed to transform any existing full-attention models into sparse versions with rather limited compute budget. In long-context scenarios, LoZA can achieve significant speed-ups both for prefill-intensive (e.g., retrieval-augmented generation) and decode-intensive (e.g., tool-integrated reasoning) cases. Specifically, by applying LoZA to LongCat-Flash during mid-training, we serve LongCat-Flash-Exp as a long-context foundation model that can swiftly process up to 1 million tokens, enabling efficient long-term reasoning and long-horizon agentic capabilities.
Read the full abstract

All authors Chen Zhang, Yang Bai, Jiahuan Li, Anchun Gui, Keheng Wang, Feifan Liu, Guanyu Wu, Yuwei Jiang, Defei Bu, Li Wei, Haihang Jing, Hongyin Tang, Xin Chen, Xiangzhou Huang, Fengcun Li, Rongxiang Weng, Yulei Qian, Yifan Lu, Yerui Sun, Jingang Wang, Yuchen Xie, Xunliang Cai

Computation and Language

HGMEM: Hypergraph-based Working Memory to Improve Multi-step RAG for Long-Context Complex Relational Modeling

Chulun Zhou, Chunkang Zhang, Guoxin Yu et al. (+4 more)

Multi-step retrieval-augmented generation (RAG) has become a widely adopted strategy for enhancing large language models (LLMs) on tasks that demand global comprehension and intensive reasoning. Although many RAG systems incorporate a working memory to consolidate information, existing designs primarily function as a passive storage for isolated facts. This static nature overlooks crucial high-order correlations among primitive facts, thereby limiting models' capacity for multi-step reasoning and resulting in fragmented reasoning and weak global sense-making within extended contexts. We introduce HGMem, a hypergraph-based working memory system, extending the concept of memory beyond simple storage into a dynamic, expressive structure for complex reasoning and global understanding. In our approach, memory is represented as a hypergraph where hyperedges correspond to distinct memory units, enabling the progressive formation of high-order interactions within memory. This mechanism connects facts and thoughts around the focal problem, evolving the memory into an integrated and situated knowledge structure that provides strong propositions for deeper reasoning. We evaluate HGMem on several challenging global sense-making benchmarks. Extensive experiments and in-depth analyses demonstrate that our method consistently improves multi-step RAG and substantially outperforms strong baseline systems across diverse datasets.
Read the full abstract

All authors Chulun Zhou, Chunkang Zhang, Guoxin Yu, Fandong Meng, Jie Zhou, Wai Lam, Mo Yu

Computation and Language

Disentangling Learning from Judgment: Representation Learning for Open Response Analytics

Conrad Borchers, Manit Patel, Seiyon M. Lee et al. (+1 more)

Open-ended responses are central to learning, yet automated scoring often conflates what students wrote with how teachers grade. We present an analytics-first framework that separates content signals from rater tendencies, making judgments visible and auditable via analytics. Using de-identified ASSISTments mathematics responses, we model teacher histories as dynamic priors and represent text with sentence embeddings. We apply centroid normalization and response-problem embedding differences, and explicitly model teacher effects with priors to reduce problem- and teacher-related confounds. Temporally-validated linear models quantify the contributions of each signal, and model disagreements surface observations for qualitative inspection. Results show that teacher priors heavily influence grade predictions; the strongest results arise when priors are combined with content embeddings (AUC~0.815), while content-only models remain above chance but substantially weaker (AUC~0.626). Adjusting for rater effects sharpens the selection of features derived from content representations, retaining more informative embedding dimensions and revealing cases where semantic evidence supports understanding as opposed to surface-level differences in how students respond. The contribution presents a practical pipeline that transforms embeddings from mere features into learning analytics for reflection, enabling teachers and researchers to examine where grading practices align (or conflict) with evidence of student reasoning and learning.
Read the full abstract

All authors Conrad Borchers, Manit Patel, Seiyon M. Lee, Anthony F. Botelho

Monday 29 December

Computation and Language · Cross-listed from cs.AI

An Agentic LLM Framework for Adverse Media Screening in AML Compliance

Pavel Chernakov, Sasan Jafarnejad and Raphaël Frank

Adverse media screening is a critical component of anti-money laundering (AML) and know-your-customer (KYC) compliance processes in financial institutions. Traditional approaches rely on keyword-based searches that generate high false-positive rates or require extensive manual review. We present an agentic system that leverages Large Language Models (LLMs) with Retrieval-Augmented Generation (RAG) to automate adverse media screening. Our system implements a multi-step approach where an LLM agent searches the web, retrieves and processes relevant documents, and computes an Adverse Media Index (AMI) score for each subject. We evaluate our approach using multiple LLM backends on a dataset comprising Politically Exposed Persons (PEPs), persons from regulatory watchlists, and sanctioned persons from OpenSanctions and clean names from academic sources, demonstrating the system's ability to distinguish between high-risk and low-risk individuals.
Read the full abstract

Computation and Language · Cross-listed from cs.HC

Althea: The Fact-Checking–Metalearning Tradeoff in AI-Assisted Verification

Svetlana Churina, Kokil Jaidka, Anab Maulana Barik et al. (+5 more)

Fact-checking systems must be scalable and epistemically trustworthy. We introduce Althea, a retrieval-augmented system for user-driven claim evaluation that matches standard pipelines on AVeriTeC while improving supported/refuted discrimination. A longitudinal survey experiment (N=961) treats a ten-day follow-up as a fading test: after modeling a verification procedure, we remove the system and ask whether users reproduce it unaided, testing metalearning rather than one-time accuracy. We compare two AI-assisted treatments, Exploratory (guided reasoning) and Summary (synthesized verdicts), against two baselines, unrelated news and Self-search. The treatments yield the strongest immediate accuracy and confidence gains but do not survive the fading test: on unseen claims they perform no better than news, while Self-search, with no procedure to fade, retains a large advantage. This reveals a factchecking-metalearning tradeoff: conditions that most improve immediate accuracy are least likely to produce metalearning, cautioning against treating AI-delivered verdicts as a source of durable literacy gains.
Read the full abstract

All authors Svetlana Churina, Kokil Jaidka, Anab Maulana Barik, Harshit Aneja, Cai Yang, Insyirah Binte Imam Mujtahid, Wynne Hsu, Mong Li Lee

Computation and Language · Cross-listed from cs.RO

Leveraging Adaptive Group Negotiation for Heterogeneous Multi-Robot Collaboration with Large Language Models

Siqi Song, Xuanbing Xie, Zonglin Li et al. (+3 more)

Multi-robot collaboration tasks often require heterogeneous robots to work together over long horizons under spatial constraints and environmental uncertainties. Although Large Language Models (LLMs) excel at reasoning and planning, their potential for coordinated control has not been fully explored. Inspired by human teamwork, we present CLiMRS (Cooperative Large-Language-Model-Driven Heterogeneous Multi-Robot System), an adaptive group negotiation framework among LLMs for multi-robot collaboration. This framework pairs each robot with an LLM agent and dynamically forms subgroups through a general proposal planner. Within each subgroup, a subgroup manager leads perception-driven multi-LLM discussions to get commands for actions. Feedback is provided by both robot execution outcomes and environment changes. This grouping-planning-execution-feedback loop enables efficient planning and robust execution. To evaluate these capabilities, we introduce CLiMBench, a heterogeneous multi-robot benchmark of challenging assembly tasks. Our experiments show that CLiMRS surpasses the best baseline, achieving over 40% higher efficiency on complex tasks without sacrificing success on simpler ones. Overall, our results demonstrate that leveraging human-inspired group formation and negotiation principles significantly enhances the efficiency of heterogeneous multi-robot collaboration. Our code is available here: https://github.com/song-siqi/CLiMRS.
Read the full abstract

All authors Siqi Song, Xuanbing Xie, Zonglin Li, Yuqiang Li, Shijie Wang, Biqing Qi

Computation and Language

An Empirical Analysis of Fine-Tuning Large Language Models on Bioinformatics Literature: PRSGPT and BioStarsGPT

Muhammad Muneeb and David B. Ascher

Large language models (LLMs) often lack specialized knowledge for complex bioinformatics applications. We present a reproducible pipeline for fine-tuning LLMs on specialized bioinformatics data, demonstrated through two use cases: PRSGPT, focused on polygenic risk score (PRS) tools, and BioStarsGPT, trained on community forum discussions. The nine-step pipeline integrates diverse data sources, structured preprocessing, prompt-based question-answer (QA) generation (via Google Gemini), natural language inference (NLI) for quality control, semantic deduplication, clustering-based data splitting, and parameter-efficient fine-tuning using LoRA. We fine-tuned three LLMs (LLaMA-3.2-3B, Qwen2.5-7B, Gemma) and benchmarked them on over 14 lexical and semantic metrics. Qwen2.5-7B emerged as the best performer, with BLEU-4 and ROUGE-1 improvements of 82% and 70% for PRSGPT and 6% and 18% for BioStarsGPT, respectively. The open-source datasets produced include over 28,000 QA pairs for PRSGPT and 154,282 for BioStarsGPT. Human evaluation of PRSGPT yielded 61.9% accuracy on the PRS tools comparison task, comparable to Google Gemini (61.4%), but with richer methodological detail and accurate citations. BioStarsGPT demonstrated 59% conceptual accuracy across 142 curated bioinformatics questions. Our pipeline enables scalable, domain-specific fine-tuning of LLMs. It enables privacy-preserving, locally deployable bioinformatics assistants, explores their practical applications, and addresses the challenges, limitations, and mitigation strategies associated with their development and use.
Read the full abstract

Computation and Language

Benchmarking Cross-Lingual Semantic Alignment in Multilingual Embeddings

Wen G. Gong

With hundreds of multilingual embedding models available, practitioners lack clear guidance on which provide genuine cross-lingual semantic alignment versus task performance through language-specific patterns. Task-driven benchmarks (MTEB) may mask fundamental alignment shortcomings. We introduce Semantic Affinity (SA), a bounded (between 0 and 1) metric measuring inter-lingual to intra-lingual spread ratio using cosine distance, combined with PHATE visualization in our Semanscope framework. Benchmarking 13 models across 4 datasets (52 experiments) reveals a three-tier structure: (1) Top BERT models (LaBSE SA = 0.70, USE SA = 0.68, S-BERT SA = 0.68) achieve strong alignment via translation-pair supervision; (2) LLM embeddings plateau at SA between 0.55 and 0.61 regardless of 0.6 B to 8 B scale; (3) MLM-only BERT models (mBERT, XLM-R, SA < 0.50) fail despite more than 100 language training. Training objective, not architecture or scale, determines alignment. Oracle Bone primitives (1200 BCE) expose semantic drift-models learn corpus patterns rather than cognitive primitives. This work provides semantic benchmarking to help practitioners select quality multilingual embeddings from hundreds of available models, showing cross-lingual alignment requires explicit translation supervision, not merely model scale or multilingual data.
Read the full abstract

Computation and Language

Geometric Patterns of Meaning: A PHATE Manifold Analysis of Multi-lingual Embeddings

Wen G Gong

We introduce a multi-level analysis framework for examining semantic geometry in multilingual embeddings, implemented through Semanscope (a visualization tool that applies PHATE manifold learning across four linguistic levels). Analysis of diverse datasets spanning sub-character components, alphabetic systems, semantic domains, and numerical concepts reveals systematic geometric patterns and critical limitations in current embedding models. At the sub-character level, purely structural elements (Chinese radicals) exhibit geometric collapse, highlighting model failures to distinguish semantic from structural components. At the character level, different writing systems show distinct geometric signatures. At the word level, content words form clustering-branching patterns across 20 semantic domains in English, Chinese, and German. Arabic numbers organize through spiral trajectories rather than clustering, violating standard distributional semantics assumptions. These findings establish PHATE manifold learning as an essential analytic tool not only for studying geometric structure of meaning in embedding space, but also for validating the effectiveness of embedding models in capturing semantic relationships.
Read the full abstract

Computation and Language · Cross-listed from cs.CY

La norme technique comme catalyseur de transfert de connaissances : la francophonie a l'œuvre dans le domaine de l'éducation

Mokhtar Ben Henda

Standards are adopted in a wide range of fields, both technical and industrial, as well as socio-economic, cultural and linguistic. They are presented explicitly as laws and regulations, technical and industrial standards or implicitly in the form of unwritten social standards. However, in a globalization marked by a very fine mosaic of socio-cultural identities, the question arises in relation to the construction of global, transparent and coherent systems in which considerable work of consensus is necessary to ensure all types of transfers and their local adaptations. The focus here is on the global education ecosystem which develops its own standards for the transfer of knowledge and socio-cultural values through learning, teaching and training. Subcommittee 36 of the International Organization for Standardization is one of the structures of this ecosystem in which the Francophonie participates to develop international standards for distance education on the basis of universal consensus.
Read the full abstract

Computation and Language · Cross-listed from cs.LG

Probing the Limits of Compressive Memory: A Study of Infini-Attention in Small-Scale Pretraining

Ruizhe Huang, Kexuan Zhang, Yihao Fang et al. (+1 more)

This study investigates small-scale pretraining for Small Language Models (SLMs) to enable efficient use of limited data and compute, improve accessibility in low-resource settings and reduce costs. To enhance long-context extrapolation in compact models, we focus on Infini-attention, which builds a compressed memory from past segments while preserving local attention. In our work, we conduct an empirical study using 300M-parameter LLaMA models pretrained with Infini-attention. The model demonstrates training stability and outperforms the baseline in long-context retrieval. We identify the balance factor as a key part of the model performance, and we found that retrieval accuracy drops with repeated memory compressions over long sequences. Even so, Infini-attention still effectively compensates for the SLM's limited parameters. Particularly, despite performance degradation at a 16,384-token context, the Infini-attention model achieves up to 31% higher accuracy than the baseline. Our findings suggest that achieving robust long-context capability in SLMs benefits from architectural memory like Infini-attention.
Read the full abstract

All authors Ruizhe Huang, Kexuan Zhang, Yihao Fang, Baifeng Yu

Computation and Language · Cross-listed from cs.LG

Trellis: Learning to Compress Key-Value Memory in Attention Models

Mahdi Karami, Ali Behrouz, Praneeth Kacham et al. (+1 more)

Transformers, while powerful, suffer from quadratic computational complexity and the ever-growing Key-Value (KV) cache of the attention mechanism. This paper introduces Trellis, a novel Transformer architecture with bounded memory that learns how to compress its key-value memory dynamically at test time. Trellis replaces the standard KV cache with a fixed-size memory and train a two-pass recurrent compression mechanism to store new keys and values into memory. To achieve this, it leverages an online gradient descent procedure with a forget gate, enabling the compressed memory to be updated recursively while learning to retain important contextual information from incoming tokens at test time. Extensive experiments on language modeling, common-sense reasoning, recall-intensive tasks, and time series show that the proposed architecture outperforms strong baselines. Notably, its performance gains increase as the sequence length grows, highlighting its potential for long-context applications.
Read the full abstract

All authors Mahdi Karami, Ali Behrouz, Praneeth Kacham, Vahab Mirrokni

Computation and Language · Cross-listed from cs.AI

The Drill-Down and Fabricate Test (DDFT): A Protocol for Measuring Epistemic Robustness in Language Models

Rahul Baxi

Current language model evaluations measure what models know under ideal conditions but not how robustly they know it under realistic stress. Static benchmarks like MMLU and TruthfulQA cannot distinguish a model that lacks knowledge from one whose verification mechanisms collapse when information degrades or adversaries probe for weaknesses. We introduce the Drill-Down and Fabricate Test (DDFT), a protocol that measures epistemic robustness: a model's ability to maintain factual accuracy under progressive semantic compression and adversarial fabrication. We propose a two-system cognitive model comprising a Semantic System that generates fluent text and an Epistemic Verifier that validates factual accuracy. Our findings, based on evaluating 9 frontier models across 8 knowledge domains at 5 compression levels (1,800 turn-level evaluations), reveal that epistemic robustness is orthogonal to conventional design paradigms. Neither parameter count (r=0.083, p=0.832) nor architectural type (r=0.153, p=0.695) significantly predicts robustness, suggesting it emerges from training methodology and verification mechanisms distinct from current approaches. Error detection capability strongly predicts overall robustness (rho=-0.817, p=0.007), indicating this is the critical bottleneck. We find that flagship models exhibit brittleness despite their scale, while smaller models can achieve robust performance, challenging assumptions about the relationship between model size and reliability. The DDFT framework provides both theoretical foundation and practical tools for assessing epistemic robustness before deployment in critical applications.
Read the full abstract

2512.23850 · 29 Dec · This version strengthens the theoretical and empirical grounding of the CI metric, including explicit analysis of structural dependencies and ranking stability under ablations (e.g., excluding Turn 4). Claims regarding scale and robustness are revised to avoid overgeneralization. The evaluation protocol, jury methodology, and limitations are expanded to clarify assumptions and boundary conditions

Abstract PDF $LANG ×2$ADVR ×1$BNCH ×1