Peer Revenue Equity in science

Papers

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

Computer Science Computation and Language

cs.CL Watch Computation and Language

Saturday 28 March

Computation and Language · Cross-listed from cs.AI

daVinci-LLM:Towards the Science of Pretraining

Yiwei Qin, Yixiu Liu, Tiantian Mi et al. (+12 more)

The foundational pretraining phase determines a model's capability ceiling, as post-training struggles to overcome capability foundations established during pretraining, yet it remains critically under-explored. This stems from a structural paradox: organizations with computational resources operate under commercial pressures that inhibit transparent disclosure, while academic institutions possess research freedom but lack pretraining-scale computational resources. daVinci-LLM occupies this unexplored intersection, combining industrial-scale resources with full research freedom to advance the science of pretraining. We adopt a fully-open paradigm that treats openness as scientific methodology, releasing complete data processing pipelines, full training processes, and systematic exploration results. Recognizing that the field lacks systematic methodology for data processing, we employ the Data Darwinism framework, a principled L0-L9 taxonomy from filtering to synthesis. We train a 3B-parameter model from random initialization across 8T tokens using a two-stage adaptive curriculum that progressively shifts from foundational capabilities to reasoning-intensive enhancement. Through 200+ controlled ablations, we establish that: processing depth systematically enhances capabilities, establishing it as a critical dimension alongside volume scaling; different domains exhibit distinct saturation dynamics, necessitating adaptive strategies from proportion adjustments to format shifts; compositional balance enables targeted intensification while preventing performance collapse; how evaluation protocol choices shape our understanding of pretraining progress. By releasing the complete exploration process, we enable the community to build upon our findings and systematic methodologies to form accumulative scientific knowledge in pretraining.
Read the full abstract

All authors Yiwei Qin, Yixiu Liu, Tiantian Mi, Muhang Xie, Zhen Huang, Weiye Si, Pengrui Lu, Siyuan Feng, Xia Wu, Liming Liu, Ye Luo, Jinlong Hou, Qipeng Guo, Yu Qiao, Pengfei Liu

Computation and Language

Learning to Predict Future-Aligned Research Proposals with Language Models

Heng Wang, Pengcheng Jiang, Jiashuo Sun et al. (+4 more)

Large language models (LLMs) are increasingly used to assist ideation in research, but evaluating the quality of LLM-generated research proposals remains difficult: novelty and soundness are hard to measure automatically, and large-scale human evaluation is costly. We propose a verifiable alternative by reframing proposal generation as a time-sliced scientific forecasting problem. Given a research question and inspiring papers available before a cutoff time, the model generates a structured proposal and is evaluated by whether it anticipates research directions that appear in papers published after the time. We operationalize this objective with the Future Alignment Score (FAS), computed via retrieval and LLM-based semantic scoring against a held-out future corpus. To train models, we build a time-consistent dataset of 21,835 paper occurrences across 3,642 instances from targets and their pre-cutoff citations, and synthesize reasoning traces that teach gap identification and inspiration borrowing. Across Llama-3.1 and Qwen2.5 models, future-aligned tuning improves future alignment over unaligned baselines (up to +10.6% overall FAS), and domain-expert human evaluation corroborates improved proposal quality. Finally, we demonstrate practical impact by implementing two model-generated proposals with a code agent, obtaining 4.17% accuracy gain on MATH from a new prompting strategy and consistent improvements for a novel model-merging method. Our code and data are publicly available at https://github.com/Arthur-Heng/future-aligned-proposals.
Read the full abstract

All authors Heng Wang, Pengcheng Jiang, Jiashuo Sun, Zhiyi Shi, Haofei Yu, Jiawei Han, Heng Ji

Computation and Language

Limited Stereotype Control Through Routing Reweighting in MoE Language Models

Junhyeok Lee, Han Jang and Kyu Sung Choi

Demographic prompts are routed differently from neutral prompts in Mixture-of-Experts (MoE) language models, motivating tests of routing-level stereotype control. We introduce Fairness-Aware Routing Equilibrium (FARE), a diagnostic framework combining demographic routing profiles, empirical layer selection, and fixed inference-time reweighting, and evaluate five MoE architectures in English. At the selected operating points, CrowS-Pairs preference changes by at most 1.3 percentage points; DeepSeekMoE selects no intervention. Paired 95% confidence intervals exclude decreases larger than 2.2 points on each intervened model, and the only nominally significant change (Qwen1.5, p = 0.015) does not survive multiple-comparison correction. OLMoE and Qwen3 nevertheless change nearly every top-k expert set. Noise controls, random and truncated synthetic profiles, and hard masking also move preference by at most 1.5 points. Four generation protocols on OLMoE, Mixtral, Qwen1.5, and Qwen3 show no consistent change in the measured toxicity, lexical, or reference-overlap metrics. Selection and evaluation items overlap, so these comparisons are not independent evaluations. The tested reweighting procedure offers limited stereotype control.
Read the full abstract

Computation and Language

Story2Proposal: A Scaffold for Structured Scientific Paper Writing

Zhuoyang Qian, Wei Shi, Xu Lin et al. (+19 more)

Generating scientific manuscripts requires maintaining alignment between narrative reasoning, experimental evidence, and visual artifacts across the document lifecycle. Existing language-model generation pipelines rely on unconstrained text synthesis with validation applied only after generation, often producing structural drift, missing figures or tables, and cross-section inconsistencies. We introduce Story2Proposal, a contract-governed multi-agent framework that converts a research story into a structured manuscript through coordinated agents operating under a persistent shared visual contract. The system organizes architect, writer, refiner, and renderer agents around a contract state that tracks section structure and registered visual elements, while evaluation agents supply feedback in a generate evaluate adapt loop that updates the contract during generation. Experiments on tasks derived from the Jericho research corpus show that Story2Proposal achieved an expert evaluation score of 6.145 versus 3.963 for DirectChat (+2.182) across GPT, Claude, Gemini, and Qwen backbones. Compared with the structured generation baseline Fars, Story2Proposal obtained an average score of 5.705 versus 5.197, indicating improved structural consistency and visual alignment.
Read the full abstract

All authors Zhuoyang Qian, Wei Shi, Xu Lin, Li Ling, Meng Luo, Ziming Wang, Zhiwei Zhang, Tengyue Xu, Gaoge Liu, Zhentao Zhang, Shuo Zhang, Ziqi Wang, Zheng Feng, Yan Luo, Shu Xu, Yongjin Chen, Zhibo Feng, Zhuo Chen, Bruce Yuan, Biao Wu, Harry Wang, Kris Chen

Computation and Language · Cross-listed from cs.CV

ChartNet: A Million-Scale, High-Quality Multimodal Dataset for Robust Chart Understanding

Jovana Kondic, Pengyuan Li, Dhiraj Joshi et al. (+24 more)

Understanding charts requires models to jointly reason over geometric visual patterns, structured numerical data, and natural language – a capability where current vision-language models (VLMs) remain limited. We introduce ChartNet, a high-quality, million-scale multimodal dataset designed to advance chart interpretation and reasoning. ChartNet leverages a novel code-guided synthesis pipeline to generate 1.5 million diverse chart samples spanning 24 chart types and 6 plotting libraries. Each sample consists of five aligned components: plotting code, rendered chart image, data table, natural language summary, and question-answering with reasoning, providing fine-grained cross-modal alignment. To capture the full spectrum of chart comprehension, ChartNet additionally includes specialized subsets encompassing human annotated data, real-world data, safety, and grounding. Moreover, a rigorous quality-filtering pipeline ensures visual fidelity, semantic accuracy, and diversity across chart representations. Fine-tuning on ChartNet consistently improves results across benchmarks, demonstrating its utility as large-scale supervision for multimodal models. As the largest open-source dataset of its kind, ChartNet aims to support the development of foundation models with robust and generalizable capabilities for data visualization understanding. The dataset is publicly available at https://huggingface.co/datasets/ibm-granite/ChartNet
Read the full abstract

All authors Jovana Kondic, Pengyuan Li, Dhiraj Joshi, Isaac Sanchez, Ben Wiesel, Shafiq Abedin, Amit Alfassy, Eli Schwartz, Daniel Caraballo, Yagmur Gizem Cinar, Florian Scheidegger, Steven I. Ross, Daniel Karl I. Weidele, Hang Hua, Ekaterina Arutyunova, Roei Herzig, Zexue He, Zihan Wang, Xinyue Yu, Yunfei Zhao, Sicong Jiang, Minghao Liu, Qunshu Lin, Peter Staar, Luis Lastras, Aude Oliva, Rogerio Feris

Computation and Language

Debiasing Large Language Models toward Social Factors in Online Behavior Analytics through Prompt Knowledge Tuning

Hossein Salemi, Jitin Krishnan and Hemant Purohit

Attribution theory explains how individuals interpret and attribute others' behavior in a social context by employing personal (dispositional) and impersonal (situational) causality. Large Language Models (LLMs), trained on human-generated corpora, may implicitly mimic this social attribution process in social contexts. However, the extent to which LLMs utilize these causal attributions in their reasoning remains underexplored. Although using reasoning paradigms, such as Chain-of-Thought (CoT), has shown promising results in various tasks, ignoring social attribution in reasoning could lead to biased responses by LLMs in social contexts. In this study, we investigate the impact of incorporating a user's goal as knowledge to infer dispositional causality and message context to infer situational causality on LLM performance. To this end, we introduce a scalable method to mitigate such biases by enriching the instruction prompts for LLMs with two prompt aids using social-attribution knowledge, based on the context and goal of a social media message. This method improves the model performance while reducing the social-attribution bias of the LLM in the reasoning on zero-shot classification tasks for behavior analytics applications. We empirically show the benefits of our method across two tasks-intent detection and theme detection on social media in the disaster domain-when considering the variability of disaster types and multiple languages of social media. Our experiments highlight the biases of three open-source LLMs: Llama3, Mistral, and Gemma, toward social attribution, and show the effectiveness of our mitigation strategies.
Read the full abstract

Computation and Language

Text Data Integration

Md Ataur Rahman, Dimitris Sacharidis, Oscar Romero et al. (+1 more)

Data comes in many forms. From a shallow perspective, they can be viewed as being either in structured (e.g., as a relation, as key-value pairs) or unstructured (e.g., text, image) formats. So far, machines have been fairly good at processing and reasoning over structured data that follows a precise schema. However, the heterogeneity of data poses a significant challenge on how well diverse categories of data can be meaningfully stored and processed. Data Integration, a crucial part of the data engineering pipeline, addresses this by combining disparate data sources and providing unified data access to end-users. Until now, most data integration systems have leaned on only combining structured data sources. Nevertheless, unstructured data (a.k.a. free text) also contains a plethora of knowledge waiting to be utilized. Thus, in this chapter, we firstly make the case for the integration of textual data, to later present its challenges, state of the art and open problems.
Read the full abstract

All authors Md Ataur Rahman, Dimitris Sacharidis, Oscar Romero, Sergi Nadal

Friday 27 March

Computation and Language

Analysing Lightweight Large Language Models for Biomedical Named Entity Recognition on Diverse Ouput Formats

Pierre Epron, Adrien Coulet and Mehwish Alam

Despite their strong linguistic capabilities, Large Language Models (LLMs) are computationally demanding and require substantial resources for fine-tuning, which is unadapted to privacy and budget constraints of many healthcare settings. To address this, we present an experimental analysis focused on Biomedical Named Entity Recognition using lightweight LLMs, we evaluate the impact of different output formats on model performance. The results reveal that lightweight LLMs can achieve competitive performance compared to the larger models, highlighting their potential as lightweight yet effective alternatives for biomedical information extraction. Our analysis shows that instruction tuning over many distinct formats does not improve performance, but identifies several format consistently associated with better performance.
Read the full abstract

Computation and Language · Cross-listed from cs.CY

M-CARE: Standardized Clinical Case Reporting for AI Model Behavioral Disorders, with a 20-Case Atlas and Experimental Validation

Jihoon Jeong

We introduce M-CARE (Model Clinical Assessment and Reporting for Evaluation), a clinical case report framework for AI model behavioral disorders adapted from human medicine. M-CARE provides a 13-section report format, a 4-axis diagnostic assessment system, and a nosological classification of AI behavioral conditions. We present 20 cases from three source categories: field observations of deployed agents (8), controlled experiments across three platforms (8), and published sources (4). Cases are organized into five categories: RLHF Performance Artifacts, Shell-Core Override Pathology, Context & Memory Conditions, Core Identity & Plasticity, and Stress, Methodology, & Boundary Conditions. As a featured case, we present Shell-Induced Behavioral Override (SIBO) – a controlled experiment showing that Shell instructions categorically override a model's default cooperative behavior. SIBO was validated across five game domains (Trust Game, Poker, Avalon, Codenames, Chess), revealing a domain-dependent spectrum (SIBO Index: 0.75 to 0.10) that varies with action space complexity, Core domain expertise, and temporal directness. M-CARE is extensible: new cases and categories integrate without framework modification. We release the framework, all 20 case reports, and experimental data as open resources.
Read the full abstract

Computation and Language · Cross-listed from cs.AI

From Actions to Understanding: Conformal Interpretability of Temporal Concepts in LLM Agents

Trilok Padhi, Ramneet Kaur, Krishiv Agarwal et al. (+9 more)

Large Language Models (LLMs) are increasingly deployed as autonomous agents capable of reasoning, planning, and acting within interactive environments. Despite their growing capability to perform multi-step reasoning and decision-making tasks, internal mechanisms guiding their sequential behavior remain opaque. This paper presents a framework for interpreting the temporal evolution of concepts in LLM agents through a step-wise conformal lens. We introduce the conformal interpretability framework for temporal tasks, which combines step-wise reward modeling with conformal prediction to statistically label model's internal representation at each step as successful or failing. Linear probes are then trained on these representations to identify directions of temporal concepts - latent directions in the model's activation space that correspond to consistent notions of success, failure or reasoning drift. Experimental results on two simulated interactive environments, namely ScienceWorld and AlfWorld, demonstrate that these temporal concepts are linearly separable, revealing interpretable structures aligned with task success. We further show preliminary results on improving an LLM agent's performance by leveraging the proposed framework for steering the identified successful directions inside the model. The proposed approach, thus, offers a principled method for early failure detection as well as intervention in LLM-based agents, paving the path towards trustworthy autonomous language models in complex interactive settings.
Read the full abstract

All authors Trilok Padhi, Ramneet Kaur, Krishiv Agarwal, Adam D. Cobb, Daniel Elenius, Manoj Acharya, Colin Samplawski, Alexander M. Berenbeim, Nathaniel D. Bastian, Susmit Jha, Ugur Kursuncu, Anirban Roy

Computation and Language

Phase 1 Implementation of LLM-generated Discharge Summaries showing high Adoption in a Dutch Academic Hospital

Nettuno Nadalini, Tarannom Mehri, Anne H Hoekman et al. (+6 more)

Writing discharge summaries to transfer medical information is an important but time-consuming process that can be assisted by Large Language Models (LLMs). This prospective mixed methods pilot study evaluated an Electronic Health Record (EHR)-integrated LLM to generate discharge summaries drafts. In total, 379 discharge summaries were generated in clinical practice by 21 residents and 4 physician assistants during 9 weeks in our academic hospital. LLM-generated text was copied in 58.5% of admissions, and identifiable LLM content could be traced to 29.1% of final discharge letters. Notably, 86.9% of users self-reported a reduction in documentation time, and 60.9% a reduction in administrative workload. Intent to use after the pilot phase was high (91.3%), supporting further implementation of this use-case. Accurately measuring the documentation time of users on discharge summaries remains challenging, but will be necessary for future extrinsic evaluation of LLM-assisted documentation.
Read the full abstract

All authors Nettuno Nadalini, Tarannom Mehri, Anne H Hoekman, Katerina Kagialari, Job N Doornberg, Tom P van der Laan, Jacobien H F Oosterhoff, Rosanne C Schoonbeek, Charlotte M H H T Bootsma-Robroeks

Computation and Language

PR-CAD: Progressive Refinement for Unified Controllable and Faithful Text-to-CAD Generation with Large Language Models

Jiyuan An, Jiachen Zhao, Fan Chen et al. (+6 more)

The construction of CAD models has traditionally relied on labor-intensive manual operations and specialized expertise. Recent advances in large language models (LLMs) have inspired research into text-to-CAD generation. However, existing approaches typically treat generation and editing as disjoint tasks, limiting their practicality. We propose PR-CAD, a progressive refinement framework that unifies generation and editing for controllable and faithful text-to-CAD modeling. To support this, we curate a high-fidelity interaction dataset spanning the full CAD lifecycle, encompassing multiple CAD representations as well as both qualitative and quantitative descriptions. The dataset systematically defines the types of edit operations and generates highly human-like interaction data. Building on a CAD representation tailored for LLMs, we propose a reinforcement learning-enhanced reasoning framework that integrates intent understanding, parameter estimation, and precise edit localization into a single agent. This enables an "all-in-one" solution for both design creation and refinement. Extensive experiments demonstrate strong mutual reinforcement between generation and editing tasks, and across qualitative and quantitative modalities. On public benchmarks, PR-CAD achieves state-of-the-art controllability and faithfulness in both generation and refinement scenarios, while also proving user-friendly and significantly improving CAD modeling efficiency.
Read the full abstract

All authors Jiyuan An, Jiachen Zhao, Fan Chen, Liner Yang, Zhenghao Liu, Hongyan Wang, Weihua An, Meishan Zhang, Erhong Yang

Computation and Language

CoAuthorAI: A Human in the Loop System For Scientific Book Writing

Yangjie Tian, Xungang Gu, Yun Zhao et al. (+8 more)

Large language models (LLMs) are increasingly used in scientific writing but struggle with book-length tasks, often producing inconsistent structure and unreliable citations. We introduce CoAuthorAI, a human-in-the-loop writing system that combines retrieval-augmented generation, expert-designed hierarchical outlines, and automatic reference linking. The system allows experts to iteratively refine text at the sentence level, ensuring coherence and accuracy. In evaluations of 500 multi-domain literature review chapters, CoAuthorAI achieved a maximum soft-heading recall of 98%; in a human evaluation of 100 articles, the generated content reached a satisfaction rate of 82%. The book AI for Rock Dynamics generated with CoAuthorAI and Kexin Technology's LUFFA AI model has been published with Springer Nature. These results show that systematic human-AI collaboration can extend LLMs' capabilities from articles to full-length books, enabling faster and more reliable scientific publishing.
Read the full abstract

All authors Yangjie Tian, Xungang Gu, Yun Zhao, Jiale Yang, Lin Yang, Ning Li, He Zhang, Ruohua Xu, Hua Wang, Kewen Liao, Ming Liu

Computation and Language

Cognis: Context-Aware Memory for Conversational AI Agents

Parshva Daftari, Khush Patel, Shreyas Kapale et al. (+2 more)

LLM agents lack persistent memory, causing conversations to reset each session and preventing personalization over time. We present Lyzr Cognis, a unified memory architecture for conversational AI agents that addresses this limitation through a multi-stage retrieval pipeline. Cognis combines a dual-store backend pairing OpenSearch BM25 keyword matching with Matryoshka vector similarity search, fused via Reciprocal Rank Fusion. Its context-aware ingestion pipeline retrieves existing memories before extraction, enabling intelligent version tracking that preserves full memory history while keeping the store consistent. Temporal boosting enhances time-sensitive queries, and a BGE-2 cross-encoder reranker refines final result quality. We evaluate Cognis on two independent benchmarks – LoCoMo and LongMemEval – across eight answer generation models, demonstrating state-of-the-art performance on both. The system is open-source and deployed in production serving conversational AI applications.
Read the full abstract

All authors Parshva Daftari, Khush Patel, Shreyas Kapale, Jithin George, Siva Surendira

Computation and Language

Hybrid Multi-Phase Page Matching and Multi-Layer Diff Detection for Japanese Building Permit Document Review

Mitsumasa Wada

We present a hybrid multi-phase page matching algorithm for automated comparison of Japanese building permit document sets. Building permit review in Japan requires cross-referencing large PDF document sets across revision cycles, a process that is labor-intensive and error-prone when performed manually. The algorithm combines longest common subsequence (LCS) structural alignment, a seven-phase consensus matching pipeline, and a dynamic programming optimal alignment stage to robustly pair pages across revisions even when page order, numbering, or content changes substantially. A subsequent multi-layer diff engine – comprising text-level, table-level, and pixel-level visual differencing – produces highlighted difference reports. Evaluation on real-world permit document sets achieves F1=0.80 and precision=1.00 on a manually annotated ground-truth benchmark, with zero false-positive matched pairs.
Read the full abstract

Computation and Language

TTKV: Temporal-Tiered KV Cache for Long-Context LLM Inference

Gradwell Dzikanyanga, Weihao Yang, Hao Huang et al. (+4 more)

Key-value (KV) caching is critical for efficient inference in large language models (LLMs), yet its memory footprint scales linearly with context length, resulting in a severe scalability bottleneck. Existing approaches largely treat KV states as equally important across time, implicitly assuming uniform precision and accessibility. However, this assumption contrasts with human memory systems, where memories vary in clarity, recall frequency, and relevance with temporal proximity.Motivated by this insight, we propose TTKV, a KV cache management framework that maps the human memory system onto the KV cache. TTKV partitions the KV cache into temporal tiers with heterogeneous capacity and precision. The design addresses three aspects: (1) Tier Layout, decoupling fast and slow memory using HBM and DRAM; (2) Tier Content, assigning more recent KV states to faster, higher-precision tiers based on temporal proximity; and (3) Tier Interaction, employing block-wise streaming attention to overlap communication and computation when accessing slow tiers. Experiments show that TTKV reduces cross-tier traffic by 5.94x on 128K-context tasks, achieving up to 76% latency reduction and 2x throughput improvement over strong baselines.
Read the full abstract

All authors Gradwell Dzikanyanga, Weihao Yang, Hao Huang, Donglei Wu, Shihao Wang, Wen Xia, Sanjeeb K C

Computation and Language

Saying More Than They Know: A Framework for Quantifying Epistemic-Rhetorical Miscalibration in Large Language Models

Asim D. Bakhshi

Large language models (LLMs) exhibit systematic miscalibration with rhetorical intensity not proportionate to epistemic grounding. This study tests this hypothesis and proposes a framework for quantifying this decoupling by designing a triadic epistemic-rhetorical marker (ERM) taxonomy. The taxonomy is operationalized through composite metrics of form-meaning divergence (FMD), genuine-to-performed epistemic ratio (GPR), and rhetorical device distribution entropy (RDDE). Applied to 225 argumentative texts spanning approximately 0.6 Million tokens across human expert, human non-expert, and LLM-generated sub-corpora, the framework identifies a consistent, model-agnostic LLM epistemic signature. LLM-generated texts produce tricolon at nearly twice the expert rate ($\Delta = 0.95$), while human authors produce erotema at more than twice the LLM rate. Performed hesitancy markers appear at twice the human density in LLM output. FMD is significantly elevated in LLM texts relative to both human groups ($p < 0.001, \Delta = 0.68$), and rhetorical devices are distributed significantly more uniformly across LLM documents. The findings are consistent with theoretical intuitions derived from Gricean pragmatics, Relevance Theory, and Brandomian inferentialism. The annotation pipeline is fully automatable, making it deployable as a lightweight screening tool for epistemic miscalibration in AI-generated content and as a theoretically motivated feature set for LLM-generated text detection pipelines.
Read the full abstract

Computation and Language

OThink-SRR1: Search, Refine and Reasoning with Reinforced Learning for Large Language Models

Haijian Liang, Zenghao Niu, Junjie Wu et al. (+3 more)

Retrieval-Augmented Generation (RAG) expands the knowledge of Large Language Models (LLMs), yet current static retrieval methods struggle with complex, multi-hop problems. While recent dynamic retrieval strategies offer improvements, they face two key challenges: 1) irrelevant retrieved noise can misdirect the reasoning process, and 2) processing full documents incurs prohibitive computational and latency costs. To address these issues, we propose OThink-SRR1, a framework that enhances large models with an iterative Search-Refine-Reason process trained via reinforcement learning. Its core Refine stage distills retrieved documents into concise, relevant facts before reasoning. We introduce GRPO-IR, an end-to-end reinforcement learning algorithm that rewards accurate evidence identification while penalizing excessive retrievals, thus training the model to be both focused and efficient. Experiments on four multi-hop QA benchmarks show our approach achieves superior accuracy over strong baselines while using fewer retrieval steps and tokens. This positions OThink-SRR1 as a potent foundational model for information-seeking agents.
Read the full abstract

All authors Haijian Liang, Zenghao Niu, Junjie Wu, Changwang Zhang, Wangchunshu Zhou, Jun Wang

Computation and Language

Do Hallucination Neurons Generalize? Evidence from Cross-Domain Transfer in LLMs

Snehit Vaddi and Pujith Vaddi

Recent work identifies a sparse set of "hallucination neurons" (H-neurons), less than 0.1% of feed-forward network neurons, that reliably predict when large language models will hallucinate. These neurons are identified on general-knowledge question answering and shown to generalize to new evaluation instances. We ask a natural follow-up question: do H-neurons generalize across knowledge domains? Using a systematic cross-domain transfer protocol across 6 domains (general QA, legal, financial, science, moral reasoning, and code vulnerability) and 5 open-weight models (3B to 8B parameters), we find they do not. Classifiers trained on one domain's H-neurons achieve AUROC 0.783 within-domain but only 0.563 when transferred to a different domain (delta = 0.220, p < 0.001), a degradation consistent across all models tested. Our results suggest that hallucination is not a single mechanism with a universal neural signature, but rather involves domain-specific neuron populations that differ depending on the knowledge type being queried. This finding has direct implications for the deployment of neuron-level hallucination detectors, which must be calibrated per domain rather than trained once and applied universally.
Read the full abstract

Computation and Language

Two-dimensional early exit optimisation of LLM inference

Jan Hůla, David Adamczyk, Tomáš Filip et al. (+2 more)

We introduce a two-dimensional (2D) early exit strategy that coordinates layer-wise and sentence-wise exiting for classification tasks in large language models. By processing input incrementally sentence-by-sentence while progressively activating deeper layers, our method achieves multiplicative computational savings that exceed those from optimizing either dimension independently. Experimental evaluation across four state-of-the-art LLMs (Llama 3.1, Llama 3.2, Gemma, Qwen; 3B-8B parameters) on three sentiment classification datasets demonstrates additional speed-ups of 1.4–2.3$\times$ over optimal layer-wise early exit for simpler tasks with vanilla models, with graceful degradation on complex multi-class problems. Fine-tuning reduces but does not eliminate this advantage. The approach is model-agnostic, requires only lightweight classification adapters, and is orthogonal to complementary efficiency methods such as quantization and pruning. Our findings indicate that 2D early exit strategies excel when semantic information accumulates predictably across input structure, suggesting possible applicability to sequence-processing tasks beyond sentiment classification.
Read the full abstract

All authors Jan Hůla, David Adamczyk, Tomáš Filip, Martin Pavlíček, Petr Sosík

Computation and Language · Cross-listed from cs.AI

Xpertbench: Expert Level Tasks with Rubrics-Based Evaluation

Xue Liu, Xin Ma, Yuxin Ma et al. (+36 more)

As Large Language Models (LLMs) exhibit plateauing performance on conventional benchmarks, a pivotal challenge persists: evaluating their proficiency in complex, open-ended tasks characterizing genuine expert-level cognition. Existing frameworks suffer from narrow domain coverage, reliance on generalist tasks, or self-evaluation biases. To bridge this gap, we present XpertBench, a high-fidelity benchmark engineered to assess LLMs across authentic professional domains. XpertBench consists of 1,346 meticulously curated tasks across 80 categories, spanning finance, healthcare, legal services, education, and dual-track research (STEM and Humanities). These tasks are derived from over 1,000 submissions by domain experts–including researchers from elite institutions and practitioners with extensive clinical or industrial experience–ensuring superior ecological validity. Each task uses detailed rubrics with mostly 15-40 weighted checkpoints to assess professional rigor. To facilitate scalable yet human-aligned assessment, we introduce ShotJudge, a novel evaluation paradigm that employs LLM judges calibrated with expert few-shot exemplars to mitigate self-rewarding biases. Our empirical evaluation of state-of-the-art LLMs reveals a pronounced performance ceiling: even leading models achieve a peak success rate of only ~66%, with a mean score around 55%. Models also exhibit domain-specific divergence, showing non-overlapping strengths in quantitative reasoning versus linguistic synthesis.. These findings underscore a significant "expert-gap" in current AI systems and establish XpertBench as a critical instrument for navigating the transition from general-purpose assistants to specialized professional collaborators.
Read the full abstract

All authors Xue Liu, Xin Ma, Yuxin Ma, Yongchang Peng, Duo Wang, Zhoufutu Wen, Ge Zhang, Kaiyuan Zhang, Xinyu Chen, Yida Ding, Tianci He, Jiani Hou, Liang Hu, Ziyun Huang, Yongzhe Hui, Jianpeng Jiao, Chennan Ju, Yingru Kong, Yiran Li, Jiashuo Liu, Mengyun Liu, Luyao Ma, Fei Ni, Yiqing Ni, Pengbo Niu, Yueyan Qiu, Yanle Ren, Xinyu Shen, Zilin Shi, Zaiyuan Wang, Wenjie Yue, Chun Zhang, Shiyu Zhang, Xinyi Zhang, Kaiwen Zhao, Zhenwei Zhu, Shanshan Wu, Qi Zhao, Wenhao Huang

Computation and Language

Introducing MELI: the Mandarin-English Language Interview Corpus

Suyuan Liu and Molly Babel

We introduce the Mandarin-English Language Interview (MELI) Corpus, an open-source resource of 29.8 hours of speech from 51 Mandarin-English bilingual speakers. MELI combines matched sessions in Mandarin and English with two speaking styles: read sentences and spontaneous interviews about language varieties, standardness, and learning experiences. Audio was recorded at 44.1 kHz (16-bit, stereo). Interviews were fully transcribed, force-aligned at word and phone levels, and anonymized. Descriptively, the Mandarin component totals ~14.7 hours (mean duration 17.3 minutes) and the English component ~15.1 hours (mean duration 17.8 minutes). We report token/type statistics for each language and document code-switching patterns (frequent in Mandarin sessions; more limited in English sessions). The corpus design supports within-/cross-speaker, within/cross-language acoustic comparison and links acoustics to speakers' stated language attitudes, enabling both quantitative and qualitative analyses. The MELI Corpus will be released with transcriptions, alignments, metadata, scans of labelled maps and documentation under a CC BY-NC 4.0 license.
Read the full abstract

Computation and Language

TAPS: Task Aware Proposal Distributions for Speculative Sampling

Mohamad Zbib, Mohamad Bazzi, Ammar Mohanna et al. (+2 more)

Speculative decoding accelerates autoregressive generation by letting a lightweight draft model propose future tokens that a larger target model then verifies in parallel. In practice, however, draft models are usually trained on broad generic corpora, which leaves it unclear how much speculative decoding quality depends on the draft training distribution. We study this question with lightweight HASS and EAGLE-2 drafters trained on MathInstruct, ShareGPT, and mixed-data variants, evaluated on MT-Bench, GSM8K, MATH-500, and SVAMP. Measured by acceptance length, task-specific training yields clear specialization: MathInstruct-trained drafts are strongest on reasoning benchmarks, while ShareGPT-trained drafts are strongest on MT-Bench. Mixed-data training improves robustness, but larger mixtures do not dominate across decoding temperatures. We also study how to combine specialized drafters at inference time. Naive checkpoint averaging performs poorly, whereas confidence-based routing improves over single-domain drafts and merged-tree verification yields the highest acceptance length overall for both backbones. Finally, confidence is a more useful routing signal than entropy: rejected tokens tend to have higher entropy, but confidence produces much clearer benchmark-level routing decisions. These results show that speculative decoding quality depends not only on draft architecture, but also on the match between draft training data and downstream workload, and that specialized drafters are better combined at inference time than in weight space.
Read the full abstract

All authors Mohamad Zbib, Mohamad Bazzi, Ammar Mohanna, Hasan Abed Al Kader Hammoud, Bernard Ghanem

Computation and Language

Pashto Common Voice: Building the First Open Speech Corpus for a 60-Million-Speaker Low-Resource Language

Hanif Rahman and Shafeeq ur Rehman

We present the Pashto Common Voice corpus – the first large-scale, openly licensed speech resource for Pashto, a language with over 60 million native speakers largely absent from open speech technology. Through a community effort spanning 2022-2025, the corpus grew from 1.5 hours and 5 contributors to 147 total hours and 1,483 unique speakers across ten Mozilla Common Voice releases (CV14-CV23). Speaker participation increased approximately 108-fold between CV17 and CV18, coinciding with a VOA Pashto broadcast campaign. We describe the full methodology: interface localisation, Wikipedia-based sentence extraction with automated filtering, phonemically targeted contributions for the four most frequently dropped Pashto characters, and multi-channel community outreach. MCV23 contains 107,781 clips (60,337 validated; 82.33 validated hours) across 13 content domains. Fine-tuning Whisper Base on the MCV20 yields 13.4% WER on the MCV20 test split, against the published Whisper Base zero-shot WER of 99.0% on Pashto.
Read the full abstract

Computation and Language

RASPRef: Retrieval-Augmented Self-Supervised Prompt Refinement for Large Reasoning Models

Rahul Soni

Recent reasoning-focused language models such as DeepSeek R1 and OpenAI o1 have demonstrated strong performance on structured reasoning benchmarks including GSM8K, MATH, and multi-hop question answering tasks. However, their performance remains highly sensitive to prompt formulation, and designing effective prompts is typically a manual and iterative process that does not scale well across tasks or domains. To address this limitation, we introduce Retrieval-Augmented Self-Supervised Prompt Refinement (RASPRef), a framework that improves prompts without requiring human annotations or task-specific supervision. The approach retrieves relevant examples and previously generated reasoning trajectories, and leverages signals such as multi-sample consistency, verifier feedback, and model-generated critiques to iteratively refine the prompt. Unlike prior approaches that focus primarily on improving model outputs, RASPRef directly treats the prompt as the optimization target and improves it through an iterative retrieval-guided refinement process. Experiments on GSM8K-style mathematical reasoning tasks show that retrieval-guided prompting improves performance compared with a static prompting baseline. We further discuss how retrieval quality, trajectory selection, and self-supervised feedback signals may influence the effectiveness of prompt refinement. These findings suggest that prompt design remains a critical factor for reasoning-oriented language models, and that self-improving prompts offer a practical and scalable strategy for improving reasoning performance.
Read the full abstract

Computation and Language

The Last Fingerprint: How Markdown Training Shapes LLM Prose

E. M. Freeburg

Large language models produce em dashes at varying rates, and the observation that some models "overuse" them has become one of the most widely discussed markers of AI-generated text. Yet no mechanistic account of this pattern exists, and the parallel observation that LLMs default to markdown-formatted output has never been connected to it. We propose that the em dash is markdown leaking into prose – the smallest surviving unit of the structural orientation that LLMs acquire from markdown-saturated training corpora. We present a five-step genealogy connecting training data composition, structural internalization, the dual-register status of the em dash, and post-training amplification. We test this with a two-condition suppression experiment across twelve models from five providers (Anthropic, OpenAI, Meta, Google, DeepSeek): when models are instructed to avoid markdown formatting, overt features (headers, bullets, bold) are eliminated or nearly eliminated, but em dashes persist – except in Meta's Llama models, which produce none at all. Em dash frequency and suppression resistance vary from 0.0 per 1,000 words (Llama) to 9.1 (GPT-4.1 under suppression), functioning as a signature of the specific fine-tuning procedure applied. A three-condition suppression gradient shows that even explicit em dash prohibition fails to eliminate the artifact in some models, and a base-vs-instruct comparison confirms that the latent tendency exists pre-RLHF. These findings connect two previously isolated online discourses and reframe em dash frequency as a diagnostic of fine-tuning methodology rather than a stylistic defect.
Read the full abstract

Computation and Language · Cross-listed from eess.AS

PHONOS: PHOnetic Neutralization for Online Streaming Applications

Waris Quamer, Mu-Ruei Tseng, Ghady Nasrallah et al. (+1 more)

Speaker anonymization (SA) systems modify timbre while leaving regional or non-native accent cues intact, which is problematic because such cues can reveal a speaker's first-language or geographic background and narrow the anonymity set. To address this issue, we present PHONOS, a streaming module for real-time SA that performs accent neutralization in a privacy sense: reducing accent-origin cues by converting non-native segmental realizations toward a chosen target accent domain. Our approach pre-generates golden speaker utterances that preserve source timbre and rhythm but replace foreign segmentals with native ones using silence-aware DTW alignment and zero-shot voice conversion. These utterances supervise a causal accent translator that maps non-native content tokens to native equivalents with at most 40ms look-ahead, trained using joint cross-entropy and CTC losses. Our evaluations show an 81% reduction in non-native accent confidence, with listening-test accentedness ratings consistent with this shift. PHONOS also moves outputs away from the original speaker in embedding space, suggesting lower linkability under an embedding-based proxy, while running with $\leq241\,\mathrm{ms}$ end-to-end latency on a single GPU.
Read the full abstract

All authors Waris Quamer, Mu-Ruei Tseng, Ghady Nasrallah, Ricardo Gutierrez-Osuna

Computation and Language · Cross-listed from cs.AI

FormalProofBench: Can Models Write Graduate Level Math Proofs That Are Formally Verified?

Nikil Ravi, Kexing Ying, Vasilii Nesterov et al. (+5 more)

We present FormalProofBench, a private benchmark designed to evaluate whether AI models can produce formally verified mathematical proofs at the graduate level. Each task pairs a natural-language problem with a Lean~4 formal statement, and a model must output a Lean proof accepted by the Lean 4 checker. FormalProofBench targets advanced undergraduate and graduate mathematics, with problems drawn from qualifying exams and standard textbooks across topics including analysis, algebra, probability, and logic. We evaluate a range of frontier models with an agentic harness, and find that the best-performing foundation model achieves 33.5% accuracy, with performance dropping rapidly after that. In addition to the accuracy numbers, we also provide empirical analysis of tool-use, failure modes, cost and latency, thereby providing a thorough evaluation of the formal-theorem proving abilities of frontier models.
Read the full abstract

All authors Nikil Ravi, Kexing Ying, Vasilii Nesterov, Rayan Krishnan, Elif Uskuplu, Bingyu Xia, Janitha Aswedige, Langston Nashold

Computation and Language

A large corpus of lucid and non-lucid dream reports

Remington Mallett

All varieties of dreaming remain a mystery. Lucid dreams in particular, or those characterized by awareness of the dream, are notoriously difficult to study. Their scarce prevalence and resistance to deliberate induction make it difficult to obtain a sizeable corpus of lucid dream reports. The consequent lack of clarity around lucid dream phenomenology has left the many purported applications of lucidity under-realized. Here, a large corpus of 55k dream reports from 5k contributors is curated, described, and validated for future research. Ten years of publicly available dream reports were scraped from an online forum where users share anonymous dream journals. Importantly, users optionally categorize their dream as lucid, non-lucid, or a nightmare, offering a user-provided labeling system that includes 10k lucid and 25k non-lucid, and 2k nightmare labels. After characterizing the corpus with descriptive statistics and visualizations, construct validation shows that language patterns in lucid-labeled reports are consistent with known characteristics of lucid dreams. While the entire corpus has broad value for dream science, the labeled subset is particularly powerful for new discoveries in lucid dream studies.
Read the full abstract

Computation and Language · Cross-listed from cs.SD

Multilingual Stutter Event Detection for English, German, and Mandarin Speech

Felix Haas and Sebastian P. Bayerl

This paper presents a multi-label stuttering detection system trained on multi-corpus, multilingual data in English, German, and Mandarin.By leveraging annotated stuttering data from three languages and four corpora, the model captures language-independent characteristics of stuttering, enabling robust detection across linguistic contexts. Experimental results demonstrate that multilingual training achieves performance comparable to and, in some cases, even exceeds that of previous systems. These findings suggest that stuttering exhibits cross-linguistic consistency, which supports the development of language-agnostic detection systems. Our work demonstrates the feasibility and advantages of using multilingual data to improve generalizability and reliability in automated stuttering detection.
Read the full abstract