Peer Revenue Impact factor: volatile

Papers

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

Computer Science Computation and Language

cs.CL Watch Computation and Language

Friday 11 September

Computation and Language

Meddies-PII: A Multilingual Framework for Personally Identifiable Information Extraction in Clinical De-identification

Linh Uyen Le, Ho Dinh Tri, Christian Hoang et al. (+1 more)

Clinical de-identification relies on accurately identifying personally identifiable information (PII). However, manually annotated datasets are costly to construct, while existing synthetic alternatives often provide limited details about their generation process or rely on relatively simple synthesis strategies. We introduce Meddies-PII-Dataset, a corpus of one million synthetic clinical documents spanning seventeen languages and nine PII labels. The documents are generated using attribute-conditioned prompts and validated through thirteen deterministic gates that enforce structural and annotation consistency. To evaluate the dataset's utility, we train Meddies-PII-Model, a BIOES token classifier, and compare it with existing PII extraction systems using exact-match entity-level F1. Meddies-PII-Model achieves the highest performance among the evaluated systems on all reported benchmarks, with a mean F1 of 0.827 across fifteen external benchmarks, compared with 0.658 for the strongest baseline. Upon acceptance, we will publicly release the dataset, benchmark suite, model, generation framework, and evaluation code to support research on multilingual clinical de-identification.
Read the full abstract

All authors Linh Uyen Le, Ho Dinh Tri, Christian Hoang, Huy Hoang Ha

Computation and Language

Agent as Policy for Robotic Manipulation

Mengzhao Jia, Yang Lin, Xixin Zhang et al. (+3 more)

We demonstrate that a general-purpose agent can directly drive a physical robot throughout task execution without any task-specific or environment-specific training. We introduce Agent as Policy (AGP), which places task planning and execution under the agent's control. Given a task and a robot interface, the agent interprets visual evidence, writes executable programs, issues motion commands, and revises its actions in response to physical outcomes. This brings the agent's reasoning and programming capabilities into continuous interaction with the physical world. We study AGP across multiple real-world manipulation tasks spanning precision manipulation, dynamic motions, and deformable objects. These include assembly from human videos, block construction from goal images, dice flipping, targeted throwing, and bimanual towel folding. AGP achieves success rates of at least 80% in seven of eight task configurations and significantly outperforms previous agentic robot systems. We further study efficiency through task experience accumulation and find that reusing saved procedures and programs shortens execution time across repeated trials. These findings support a path for general-purpose agents to act as robot policies, extending their autonomy to physical manipulation through runtime reasoning, programming, and interaction.
Read the full abstract

All authors Mengzhao Jia, Yang Lin, Xixin Zhang, Zhihan Zhang, Xiaobai Liu, Meng Jiang

Computation and Language

The House with a Million Windows: Interactive Fiction for Narrative Restorying

Cody Kommers, Sarah G Immel, Drew Hemment et al. (+1 more)

AI-assisted writing can flatten meaning in human storytelling, enabling the production of homogeneous outputs without the intentional effort and sense-making writing entails. To address this challenge, we present The House with a Million Windows (HWAMW), an LLM-based interactive fiction system designed to help users explore both the breadth and depth of potential meanings within their personal stories – drawing on a psychological paradigm called the restorying intervention. In HWAMW, users play through a text-based narrative in which they tell a story, then encounter a set of LLM-generated "windows" reframing it according to different literary styles. Empirical evidence shows that HWAMW increases users' sense of narrative identity, while an expert review explores how this effect is achieved. Our findings suggest that HWAMW facilitates restorying and offers a valuable paradigm for AI-assisted writing, wherein LLMs do not tell our stories but rather help us see greater potential in the stories we tell.
Read the full abstract

All authors Cody Kommers, Sarah G Immel, Drew Hemment, Mina Lee

Computation and Language · Cross-listed from cs.CV

Earth-Agent-Pro: Towards Real-World Full-Chain Earth Observation with Agents

Zhutao Lv, Chenhao Dang, Yi Feng et al. (+5 more)

Real-world Earth observation (EO) agents must translate high-level scientific questions into executable workflows to acquire observations, prepare data, perform domain computations, and derive conclusions from runtime evidence. Existing EO agents typically start from supplied observations, while benchmarks typically provide prepared inputs or candidate answers, leaving full-chain open-world EO execution largely untested. We present Earth-Agent-Pro, an execution-adaptive Plan-and-Execute framework using expert-authored skills to constrain planning and runtime tool use. Workflow-centered structured memory records planned steps, accepted evidence, and their dependencies, enabling repair of only the affected workflow suffix when runtime evidence invalidates a step. Separate large language model adapters use sequence-level supervised fine-tuning for planner workflow composition and node-level group relative policy optimization with locally verifiable rewards for executor tool-argument grounding. Earth-Bench-Pro instantiates 248 expert-curated task cores as 744 questions under three matched regimes. Its 248 Open-World Execution questions span RGB imagery, spectral observations, and remote sensing products, pairing high-level requests with runtime data requirements, executable trajectories, and open-ended answers grounded in execution evidence. With a shared GPT-5 backbone, Earth-Agent-Pro achieves 66.13% LLM-as-Judge accuracy, exceeding ReAct by 20.95 points in this metric and 24.44 points in Tools-In-Order. Joint adapter tuning raises Qwen3.5-9B LLM-as-Judge accuracy from 38.31% to 50.00%, an 11.69-point gain over the untuned configuration. Planning-only evaluation and execution with the reference workflow show that the adapters improve workflow composition and argument grounding, respectively. Code and datasets will be released soon.
Read the full abstract

All authors Zhutao Lv, Chenhao Dang, Yi Feng, Yanpei Gong, Xiaolei Wang, Junyan Ye, Conghui He, Weijia Li

Computation and Language · Cross-listed from cs.AI

Information Specialization and Constrained Synthesis in Multi-Agent LLM Forecasting: A Prospective Live-Study of the 2026 FIFA World Cup

Julian Varghese, Lucas Bickmann and Sarah Sandmann

Large language models are being organized into multi-agent systems with specialized roles, but whether such specialization produces distinct forecasts and whether subsequent synthesis improves utility remains unclear. In this study, we carried out a live, prospective evaluation over the final 56 matches of the information-dense 2026 FIFA World Cup, keeping a frontier foundation model constant while assigning two primary forecasting agents contrasting specialist roles: a quantitative specialist focusing on structured performance statistics and a news specialist focusing on current injuries, tactics and information from press conferences. Their forecasts were then reviewed by a separate critic before being combined by a meta-agent, resulting in a sequential four-agent model. Forecasts from the betting market served as an external benchmark. The news specialist obtained the highest mean probability-weighted Top-3 utility and matched the betting market in Top-3 exact-score hits. Nevertheless, the two specialist forecasters agreed on at least two of the three scorelines in 50 out of 56 matches, and the meta-agent never generated more than one scoreline outside the specialists' forecast set. These findings show that rapidly changing, unstructured information can provide a valuable forecasting signal alongside structured statistics, whereas adding critic and meta-agent stages does not necessarily create complementary information or improve on the strongest specialist.
Read the full abstract

Computation and Language · Cross-listed from cs.AI

Confidence-Gated Transductive Test Generation for Code Reranking

Sungjae Lee, Youngsik Yoon, Seockbean Song et al. (+3 more)

Test case synthesis is crucial for evaluating and ranking programs generated by large language models (LLMs). However, constructing high-quality test cases remains challenging because reliable expected outputs are often difficult to obtain. We propose Confidence-Gated Transductive Test Generation (CoTT), which first uses an efficient inductive procedure and invokes transductive generation only when inductive confidence is low. This adaptive design improves output reliability while allocating extra computation only when needed. On code reranking benchmarks, CoTT outperforms prior baselines across the reported metrics while reducing cost relative to applying transductive generation to every input. These results show that confidence-based allocation of test-time computation provides a favorable efficiency-effectiveness trade-off with a single efficient LLM.
Read the full abstract

All authors Sungjae Lee, Youngsik Yoon, Seockbean Song, Siwei Wang, Wei Chen, Jungseul Ok

Computation and Language

Zipbench: Low-Cost Framework for Compressing Comprehensive Benchmarks of Large Language Models

Zhongzhan Huang, Junxin Li, Guoming Ling et al. (+3 more)

Comprehensive benchmark suites are essential for improving large language models (LLMs), but many widely used benchmarks are redundant, making evaluation unnecessarily expensive. Although recent benchmark compression methods (BCMs) can mitigate this cost, many strong BCMs rely on large collections of per-sample evaluation results from numerous LLMs to identify representative samples. Building such collections is also expensive unless they are already public, making these methods difficult to extend to newly released benchmarks. To address this challenge, we present ZipBench, a simple and low-cost BCM with theoretical error and rank-consistency guarantees. ZipBench evaluates only a small set of anchor LLMs, synthesizes pseudo evaluation results to broaden coverage, learns compact sample representations, and selects a small yet representative subset. Building on it, we create ZipBench Zoo, a collection of compact versions of 100+ benchmark proxies spanning text, multimodal, and agent tasks. These benchmark achieve mean absolute errors of 0.002–0.02 and average Spearman correlations of ~0.98 with the full benchmarks. Overall, ZipBench reduces the cost of both LLM evaluation and compact benchmark construction, lowering the barrier to broad LLM research for compute-constrained researchers. The code has been released in https://github.com/MilkThink-Lab/ZipBench.
Read the full abstract

All authors Zhongzhan Huang, Junxin Li, Guoming Ling, Yupei Lin, Shanshan Zhong, Hefeng Wu

Computation and Language

AMDKernelVault: Large-Scale Datasets and Agentic Training for AMD GPU Kernel Optimization

Ji Liu, Saptarshi Majumder, Yiqing Huang et al. (+14 more)

We introduce AMDKernelVault, an open HIP and Triton kernel corpus and training framework for recent AMD CDNA GPUs. Existing LLM-based kernel agents are largely CUDA/NVIDIA-centric and often depend on repeated frontier-LLM calls for generation, reflection, and optimization. To address this gap, we develop HIPKernelGen and TritonKernelGen, agent-driven pipelines that transform PyTorch references into HIP or Triton kernels, compile and validate candidates under ROCm, and latency-profile them on AMD hardware. The corpus contains 62,153 execution-verified HIP kernel samples, 2,377 production-grounded ROCm Libraries QA entries, and 39,893 Triton kernels. We further train Qwen3-8B with supervised fine-tuning and execution-aware reinforcement learning as a demonstration of the corpus's utility. Under fixed evaluation budgets, it achieves the highest correctness among the compared models on PyTorch-to-HIP (34.0% Pass@1), TritonBench-G (33.2% Corr@3), and ROCmBench (41.94% Corr@3), but does not uniformly lead compilation or speed metrics. The corpus and documentation are available at https://huggingface.co/datasets/amd/AIG-Datasets, and the associated training and kernel-generation code is available at https://github.com/AMD-AGI/hip_kernel_llm_lab.
Read the full abstract

All authors Ji Liu, Saptarshi Majumder, Yiqing Huang, Wenwen Ouyang, Umang Pandey, Zeping Li, Chushi Chen, Zihao An, Puyuan Yang, Zekai Li, Sina Rafati, Ziqiong Liu, Pratik Prabhanjan Brahma, Dong Li, Zicheng Liu, Sharon Zhou, Emad Barsoum

Computation and Language

Not All Speech Is Intent: Adaptive Self-Correcting Inference Layer for Post-ASR False Wake-Up

Preeti Saraswat, Divya Neelagiri and Anil Yadav

False wake-up activations remain a persistent challenge in conversational AI. Speech phonetically similar to a device's wake word can produce a syntactically valid and semantically coherent ASR transcript that the assistant incorrectly executes. Most existing systems make a single intent decision in isolation, without a mechanism to learn from recurring errors over time or adapt to individual users through personalized learning. We introduce the Feedback-Driven Adaptive Self-Correcting Inference Layer (ASCIL), a complementary post-ASR correction framework that re-evaluates wake-up intent before response generation by fusing acoustic embeddings, linguistic cues, device context, and patterns from past misclassifications. ASCIL interprets implicit signals, including hesitation, disengagement, and silence, and explicit signals, including cancellation and repetition, as automatically inferred, noisy behavioral indicators of potential misclassification. These signals drive online pattern updates without manual annotation, whereas the intentional/unintentional reference labels used for offline evaluation are human-annotated. It generalizes from prior errors, applies corrective adjustments at inference time, and continuously updates in parallel with natural-language execution. Evaluated on a proprietary dataset of 3,667 interactions with human-annotated intentional/unintentional reference labels spanning 14 acoustic and contextual conditions, ASCIL achieves 54.27% relative error reduction on a session-disjoint subset constructed from baseline failures, and up to 24.39% relative error reduction at threshold 0.90 on the issue-tagged evaluation slice. These gains are achieved while improving intentional acceptance rates, with a median added latency below 60 ms in the reported benchmark.
Read the full abstract

Computation and Language

GraphProfiler: Source-Linked Sensitive Attribute Inference via Personal Knowledge Graphs

Ahmed Sohair Khan, Estrid He, Chenglong Ma et al. (+2 more)

Sensitive attributes such as age, income, and occupation can be inferred from user-generated content by aggregating indirect cues across many ordinary posts. LLM-based profilers can perform this aggregation automatically and with high accuracy, which makes large-scale personal attribute inference a major privacy threat. Existing LLM-based profilers, however, offer limited insight into which specific posts, concepts, and relationships made an inference possible, which is key to targeted privacy mitigation, i.e., redacting or rewriting only the few posts that actually leak an attribute, rather than perturbing entire histories. We introduce GraphProfiler, an auditable LLM-based profiler that represents each user's post history as a source-linked personal knowledge graph where nodes and edges trace back to the originating post and resolves attribute predictions to cited graph records and source texts. GraphProfiler reaches 86.7% attack success rate on the eight-attribute SynthPAI benchmark, within two points of strong text-only baselines, and 84.6% on PANDORA, while citing supporting evidence for over 98% of predictions. Our controlled ablation experiments provide evidence that the cited posts contribute to attack success, as removing them reduces the attack success rate substantially more than removing an equal number of random posts.
Read the full abstract

All authors Ahmed Sohair Khan, Estrid He, Chenglong Ma, Monica Wachowicz, Elham Naghizade

Computation and Language · Cross-listed from cs.AI

Do LLMs Trust the Accuser or the Accusation? Measuring Belief Shifts in Werewolf

Yu-Yu Yang, Ti-Rong Wu, Hung Guei et al. (+2 more)

Social-deduction games such as Werewolf are increasingly used to evaluate LLM agents, but existing evaluations often rely on final game outcomes. We propose a belief-shift evaluation benchmark in Werewolf for analyzing communication skills through belief updating. Using LLM-played games, we annotate suspicion and accusation messages and measure how an observing village-side model's beliefs change after each message. We evaluate 40 open-weight LLM configurations on 1,224 annotated messages. Our results show that larger models better distinguish true wolves from villagers based on game history, but accusations still strongly influence their beliefs. Models become more suspicious of the accused target and less suspicious of the accuser, especially when the accuser is trusted, even if the accuser is wolf-aligned. Larger models better resist accusations from accusers they already distrust. Overall, our findings suggest that current open-weight LLMs up to 120B parameters still struggle to integrate accusation content with source trust in strategic communication. Our benchmark and code are available at https://rlg.iis.sinica.edu.tw/papers/werewolf-accusation-benchmark.
Read the full abstract

All authors Yu-Yu Yang, Ti-Rong Wu, Hung Guei, Hsing-Yu Chen, I-Chen Wu

Computation and Language · Cross-listed from physics.soc-ph

Diverse Minds, Divided Networks? Personality Composition, Polarization, and Collective Intelligence in LLM-Based Social Simulations

Raad Bin Tareaf

Simulated societies of large language model agents are used to study online polarization, and separately to study collective intelligence, but the two are rarely measured in the same system. It is therefore difficult to say whether a society's personality composition shapes both, or whether reducing polarization costs collective competence. We present TraitMix, an experimental design in which the Big Five composition of a simulated social network, both trait levels and trait heterogeneity, is a controlled experimental variable, and in which polarization and collective performance are measured in the same runs. Across 991 simulations of hundred-agent societies, spanning six contested topics and six language models, trait heterogeneity has the largest measured effects, acting in opposite directions on two faces of polarization: varied societies hold more dispersed opinions while being less segregated into camps, so homogeneous societies are not moderate but consensual echo chambers. Trait effects are not additive, as Agreeableness determines the sign of Openness, an interaction that replicates across models although the primary model's estimate is influence-driven. Contrary to the trade-off the study was designed to measure, no polarization measure predicts poorer collective performance, and cross-cutting interaction is the only one of four whose association with collective accuracy survives partialling on the aggregation identity. We report ablations removing two potential measurement circularities, an induction gate applied to every model, and the measures that failed them.
Read the full abstract

Computation and Language · Cross-listed from cs.AI

Beyond ID Embeddings: Process-Grounded Language Modeling for Cognitive Diagnosis

Minghang Liu, Yuanzhuo Wang, Qiang Qiu et al. (+2 more)

Cognitive Diagnosis Models (CDMs) play a pivotal role in personalized online learning. Traditional CDMs rely on discrete, ID-based embeddings to represent students, exercises, and concepts. This paradigm diverges from the nature of learner cognition, where knowledge is not stored and retrieved as isolated symbols. As a result, CDMs suffer from semantic limitations when new exercises or concepts appear. In this paper, we propose a Process-aware Language Cognitive Diagnosis (PLCD) framework that uses language-derived structures as cognitive priors and response records to calibrate student posterior states. PLCD leverages large language models (LLMs) to construct concept schemas and cognitive process graphs, and uses target-conditioned semantic memory to retrieve historical responses that are relevant to each target exercise. A process-grounded Language-to-Cognition Mapper with DA-MoE experts and process-level contrastive learning then maps the textual evidence into a unified cognitive space. Experimental results show that PLCD not only outperforms traditional baselines in predicting student performance but also exhibits strong cognitive transfer capabilities. These results connect the computational power of LLMs with the psychometric goal of measuring latent knowledge states, suggesting that structured language priors calibrated by response records can improve cold-start robustness and cognitive grounding.
Read the full abstract

All authors Minghang Liu, Yuanzhuo Wang, Qiang Qiu, Huawei Shen, Xueqi Cheng

Computation and Language

Representation-based Masked Diffusion Model

Yangrong Hu, Ding Huang, Xueyu Zhou et al. (+1 more)

Masked Diffusion Models (MDMs) have emerged as a compelling paradigm for language modeling, offering the capability for efficient parallel text generation. However, existing parallel sampling methods typically update multiple masked tokens independently and ignore the complex mutual dependencies among the masked tokens. This independent updating mechanism lacks global coordination and might lead to incoherent outputs. To address this limitation, we propose Representation-based Masked Diffusion Model (RMDM), a framework that leverages the text representation to explicitly encode global semantics and help to parallel update tokens more precisely. Specifically, we first encode text into a continuous semantic space using a pretrained encoder and learn an invertible transformation that normalizes the representation distribution to a Gaussian prior, facilitating efficient sampling during generation. Conditioned on this latent semantic representation, we train a masked diffusion model to learn the conditional text distribution, where the representation serves as global semantic guidance to coordinate parallel token updates and faithfully approximate the target distribution. Empirical results demonstrate that RMDM significantly improves generation quality, particularly in aggressive few-step sampling regimes.
Read the full abstract

All authors Yangrong Hu, Ding Huang, Xueyu Zhou, Jian Huang

Computation and Language

ORQA: An Occupation-Realistic Question and Answer Framework for LLM Professional Knowledge

Shreyas Krishnan, Serina Chang and Abhishek Nagaraj

We present ORQA, a method for testing occupation-level knowledge in large language models. Prior methods either map abstract LLM skills to occupations via task definitions or utilize expert knowledge which is difficult to obtain at scale and expensive. ORQA complements both of these methods by connecting O*NET occupations to trusted occupation-specific websites (such as regulatory agencies, licensing bodies, professional organizations, and government publications) and converting these into source-traceable question-answer pairs. A combination of an automated pipeline and human review produces a set of high quality questions about occupations. The question set created via our method covers 116 occupations from all 21 major groups in the SOC, with 480 questions sourced from 187 different websites. Each question is designed to probe a real-world skill question that is relevant to the occupation in question. We test 15 state-of-the-art frontier and open-weight models via this method. Claude Opus 4.6, GPT-5.4 and Claude Sonnet 4.6 all perform the best at approximately 58-62% while smaller open-weight models achieve approximately 33-41% performance. Performance varies significantly across occupations. Healthcare-related occupations achieve the highest performance (78%) while Office and Administrative Support achieve approximately 40%. Performance on individual occupations (e.g. Sheet Metal Workers and Fish and Game Wardens) is essentially zero. We also find that open-ended questions and weighting by wage bill do not significantly affect the ranking of models on this benchmark. We believe that leveraging existing trusted occupation-specific information to test LLM knowledge in professional domains may be a scalable and useful method for evaluating occupation-level AI performance in the future. Results and data are available at orqabench.org.
Read the full abstract

Computation and Language

CueMem: Cue-Guided Context Reconstruction for Long-Term Conversational Memory

Changjian Wang, Rongzhen Li, Weili Guan et al. (+3 more)

Long-term conversational agents must answer user queries by recalling information from extended dialogue histories, yet directly using the full history is costly and often unreliable, while compressed memory units may lose fine-grained evidence needed for question answering. Motivated by the reconstructive view of autobiographical memory, we propose CueMem, a cue-guided framework that treats extracted memory records as retrieval cues rather than self-contained evidence and reconstructs query-relevant dialogue context from their source turns. During memory construction, CueMem extracts fine-grained memory cues from dialogue turns and links each cue to its source turn. At query time, it retrieves query-relevant cues, maps them to source-turn anchors, and expands from these anchors over a turn graph that captures temporal proximity and semantic relatedness, reconstructing a compact evidence context from the original dialogue for LLM answer generation. Experiments on LoCoMo and LongMemEval show that CueMem consistently outperforms representative long-term memory baselines. Further analyses show that graph-based context reconstruction helps recover supporting dialogue evidence while reducing query-time input tokens and latency compared with the full-history LLM setting. These results highlight retrieval cues as an effective alternative to self-contained memory evidence for long-term conversational question answering.
Read the full abstract

All authors Changjian Wang, Rongzhen Li, Weili Guan, Shuming Shi, Quan Lu, Ning Jiang

Computation and Language

SynthSentry: Detecting Synthetic Data Contamination in Language Model Training Data

Praveen Kumar Myakala, Ravichandra Namburi, Sowmya Keragodu Jayaramu et al. (+1 more)

Large language models trained recursively on their own or other models' outputs undergo model collapse, in which distributional tails and factual accuracy deteriorate while fluency survives. Prior work diagnoses collapse after training; the actionable problem is screening a corpus of unknown provenance before training. We introduce SynthSentry, a corpus-level, model-agnostic contamination signal requiring no access to the generating model, no generation history, and no synthetic labels. The score is a distributional divergence over three statistics: lexical diversity collapse, n-gram tail truncation, and perplexity variance across reference models. We evaluate on corpora contaminated by small open-weight generators and an instruction-tuned open-weight model under a leave-one-generator-out protocol. A domain-stratified study measures false positives on naturally repetitive human text (legal, clinical, source code). The score ranks corpora by severity with little loss when whole generator families are held out. Per-domain calibration holds near its nominal false-positive budget once covariance shrinkage and a bootstrap threshold replace a naive quantile, which runs four times over budget. A downstream fine-tuning check showed no contamination-driven accuracy deficit at our scale, so whether pruning recovers one remains open; the same run shows over-pruning risk once pruning exceeds the true contamination fraction. We frame screening as a data-curation defense rather than a post-hoc diagnosis and release the scoring toolkit. All results are small-scale; scope is English-language, batch-mode corpus screening. Contamination sources are single-generation or hand-authored rather than recursively generated, so results speak to synthetic contamination generally and not to recursion depth.
Read the full abstract

All authors Praveen Kumar Myakala, Ravichandra Namburi, Sowmya Keragodu Jayaramu, Sooraj George Thomas

Computation and Language

I Am No One: Style-Aware Paraphrasing for Text Anonymization

Ahmed Sohair Khan, Estrid He, Monica Wachowicz et al. (+1 more)

Authorship attribution models can re-identify users from seemingly anonymized text by exploiting stable stylistic fingerprints, even after explicit identifiers are removed, posing a growing privacy risk for text publishing and analytics. This risk extends to speech-derived text such as ASR transcripts of meetings and call-center conversations, where stylometric leakage can persist even after acoustic anonymization. Differential privacy-based anonymization often severely degrades text quality and utility. We propose a style-aware, prompt-driven anonymization approach that uses pretrained large language models to construct compact stylistic profiles from minimal samples and rewrite text to suppress identifiable style markers while preserving meaning. Across blog and review datasets, our approach reduces authorship attribution F1 by 60-70% while maintaining content quality and readability, substantially outperforming DP-based and non-DP baselines.
Read the full abstract

All authors Ahmed Sohair Khan, Estrid He, Monica Wachowicz, Elham Naghizade

Computation and Language

ESTS at WMT26: Routing-Informed Expert Pruning for Model Compression

Liu O. Martin, Lucas Bandarkar and Nanyun Peng

We describe six submissions under the team name ESTS to the unconstrained WMT26 Model Compression Shared Task for English–Simplified Chinese and English–Egyptian Arabic. We submit three compression operating points per translation direction, all derived from GPT-OSS-20B. We use task-specific routing mass to rank experts and cross-lingual routing divergence to allocate retained capacity across layers, then physically remove low-importance experts. The resulting specialists are recovery-tuned on GPT-5.1-generated synthetic translation data and further compressed by applying MXFP4 quantization to the retained expert projection weights. We additionally implement a robust inference system for the instruction-conditioned WMT26 setting, including category inference, output validation, retries, segmented fallback, and source-owned JSON reconstruction. Across our six submissions, parameter counts range from 4.186B to 7.770B and packed artifact sizes from 4.55 to 6.33~GiB. Internal xCOMET-XL evaluation using GPT-5.1 pseudo-references provides an internal comparison across the submitted compression operating points.
Read the full abstract

Computation and Language

Breaking the Token Ceiling: Distilling Smaller, Stronger Byte Models

Kalyani Marathe, Artidoro Pagnoni, Tomasz Limisiewicz et al. (+4 more)

Small models are made more capable through distillation from a larger one that shares their tokenization scheme. However, do distilled byte and token models behave similarly in terms of scaling trends as compute and data increases? To enable this comparison, we introduce two variants to efficiently convert token logits to Byte Logits: 1) approximate: Marginalize-It, and 2) exact: End-Of-Token. We then present the first large scale study of overtraining decoder-only dense transformer models varying two dimensions simultaneously: the tokenization scheme (Tokens, Bytes, Bytes w/ eot) and the training objective (Distillation vs. Cross-Entropy), sweeping layer-parameter-matched models with roughly 1 billion parameters up to 1 trillion bytes of data. Across eight benchmarks spanning three categories: Multiple Choice QA, Language Generation, and Machine Translation, we find that Token-1B models outperform byte models (End-Of-Token-1B and Bytes-1B) in the low-FLOP regime but eventually plateau; byte models start worse yet surpass Token-1B models with more compute, reaching a higher downstream task performance ceiling. Extrapolating the average top-1 error vs. validation BPB scaling laws predicts that, asymptotically, distilled End-Of-Token-1B outperforms distilled Token-1B by up to 4%. They are also far more data efficient, matching the performance of distilled Token-1B using only one-sixth of the training data. Moreover, by operating over a small vocabulary of 256 bytes instead of on the order of 100K tokens, they circumvent the need for top-k truncation during logit dumping, while also reducing logit storage costs to roughly one-fifth. Finally, our downstream performance scaling laws predict that our distilled End-Of-Token-1B models asymptotically surpass the Llama 3.2-1B, Gemma-3-1B-pt, and Gemma 2B models on averaged downstream tasks by up to 6.5%, 8.1%, and 2.1%, respectively.
Read the full abstract

All authors Kalyani Marathe, Artidoro Pagnoni, Tomasz Limisiewicz, Margaret Li, Mike Lewis, Luke Zettlemoyer, Srinivasan Iyer

Thursday 10 September

Computation and Language · Cross-listed from cs.SD

How Robust Are Neural Audio Codecs for African Speech? A Multi-Task Benchmark and the Limits of Perceptual Quality

Chibuzor Okocha and Christan Earl Grant

Neural audio codecs enable low-bitrate speech compression and tokenization, yet their robustness on accented and multilingual speech remains under-evaluated. We benchmark seven open-source neural codecs (DAC, EnCodec, FocalCodec, LanguageCodec, SemantiCodec, UniCodec, WavTokenizer) on three African speech datasets afrinames, afrispeech dialog, afrispeech multilingual, reporting signal-level quality (NISQA, UTMOS, ViSQOL, STOI, F0-RMSE) alongside two downstream tasks: automatic speech recognition (ASR) and speaker verification (ASV). Our analysis yields four findings. First, signal metrics differ sharply in downstream validity: reference-based structural/intelligibility measures (ViSQOL, STOI) and prosodic error (F0-RMSE) track ASR and ASV degradation far more reliably than neural mean-opinion-score predictors (NISQA, UTMOS). Second, intelligibility and speaker-identity preservation diverge sharply across architectures, and the apparent identity ranking itself depends on the ASV backend. Third, degradation is strongly domain-dependent and largest for conversational dialog. Fourth, the resulting degradation is partly recoverable: parameter-efficient codec adaptation (LoRA, $\sim$1–3% of parameters) on roughly 35 hours of African speech reduces the compression-induced word-error-rate gap, recovering recognition toward the uncompressed baseline (developed in companion work). These results motivate task-aware, domain-representative, and adaptation-aware evaluation of speech codecs as a prerequisite for inclusive deployment.
Read the full abstract

Computation and Language · Cross-listed from cs.CR

Evasion Attacks: How Adversarial Noise Bypasses ML Classifiers

Parker Hummel, Ryne Skabo and Muhammad Abusaqer

This paper presents a reproducible, educational study of evasion attacks in image classification and text classification. A compact convolutional network trained on MNIST reached 98.63% clean test accuracy and was evaluated under two white-box attacks. Under FGSM, accuracy fell to 60.20% at $\epsilon$ = 0.15 and 1.72% at $\epsilon$ = 0.30; under PGD it fell to 32.47% and 0.41%, and a bit-depth-reduction defense recovered only part of the loss. In the second experiment, DistilBERT fine-tuned on the SMS Spam Collection reached 98.75% accuracy and a 94.96% F1-score, but a controlled sequence of pre-defined perturbations (character substitutions, whitespace noise, and a benign suffix) produced only modest probability shifts in most displayed examples and no flip from spam to ham. Adversarial vulnerability is strongly modality-dependent: the MNIST experiment is a clear evasion demonstration, whereas the text experiment is a controlled robustness evaluation. Robustness must be tested empirically rather than inferred from clean accuracy.
Read the full abstract

Computation and Language

Are We Grading Properly? Understanding Failure Modes in Medical Benchmarks

Prithvi Dixit and Pedram Hosseini

Medical evaluation is shifting from static option-based questioning to realistic clinical scenarios with open-ended output modes. Grading these at scale naively, however, is expensive, and rubric-based evaluation has become the dominant scalable alternative. We ask what happens when the rubrics themselves are not airtight, and whether such flaws can be detected and corrected. We apply RIFT, a global rubric failure taxonomy, to two clinical benchmarks (HealthBench Professional and LiveMedBench), and find failure modes are meaningful: on HealthBench Professional an LLM judge flags 29.6% of criteria as non-atomic and 65.4% as misaligned/rigid. Then, we show that these flaws are meaningful and not simply cosmetic. As an example, rewriting bundled criteria of the form "at least one of / all of the following" as equally weighted children and regrading identical responses shifts scores by up to 15.9 percentage points on affected conversations, with disjunctive bundles inflating scores and conjunctive bundles deflating them. We also find that RIFT generally under-detects bundling on clinical rubrics, flagging 3.3% of LiveMedBench criteria as non-atomic where surface-form analysis finds structure in 25.8%.
Read the full abstract

Computation and Language

EAR: Entity-Aware Partitioning Approach for Retrieval-Augmented Generation Development

Cenab Batu Bora, Oylum Alatl{\i}, Sebnem Bora et al. (+1 more)

Retrieval-augmented generation (RAG) can improve knowledge-intensive question answering, but the first design choice is easy to overlook: how should the source corpus be partitioned into retrievable units? Fixed-size chunks often return long passages whose relation to the question is only implicit. We introduce EAR, an Entity-Aware Partitioning approach for multiple-choice question answering (MCQA). EAR extracts normalized surface anchors from the question, answer options, and corpus; retrieves local windows around matching corpus anchors; and can attach a larger parent passage through an extractive summary. We evaluate EAR on a cleaned Massive Multitask Language Understanding (MMLU)-style subset of 153 questions selected by an automatic corpus-support heuristic and using decontaminated public textbook text. Across same-protocol top-k = 3 and top-k = 8 sweeps with Mistral, Gemma, and DeepSeek, EAR entity-window reduces retrieved words by 37.5-40.2% relative to chunks. Observed accuracy changes are +5.2, +1.3, and -3.9 points at top-k = 3, and +5.9, -3.3, and -4.6 points at top-k = 8; none of the entity-window differences is statistically significant. The scoped contribution is methodological: EAR provides a compact and inspectable retrieval unit, while its rule-based anchor extractor remains domain-specific and requires separate validation before transfer.
Read the full abstract

All authors Cenab Batu Bora, Oylum Alatl{\i}, Sebnem Bora, Oguz Dikenelli

Computation and Language

HypoKG: Evidence-Disciplined Biomedical Hypothesis Generation Beyond Endpoint Knowledge

Dominic Okonkwo, Adetayo Okunoye and Ismailcem Budak Arpinar

Large language models (LLMs) can generate biomedical hypotheses, but it remains unclear whether they truly reason from scientific evidence or simply produce convincing-sounding ideas. To study this, we combine three major biological databases: the Kyoto Encyclopedia of Genes and Genomes (KEGG), Rhea, and UniProt, into a unified biochemical knowledge graph and construct a benchmark of 550 paths connecting enzyme sources to rare disease endpoints, yielding 13,200 hypotheses from six LLMs under four conditions varying the biological information each model receives: source enzyme only, full biological path, or source and disease endpoint only. Hypotheses are scored using an expert-derived five-criterion rubric on a 1-5 scale per criterion. We find that models given both the source and disease endpoint often produce the highest-scoring hypotheses, showing that LLMs can generate compelling ideas from minimal information. However, these hypotheses are less grounded in the evidence. In contrast, models given the full biological path generate hypotheses more consistent with known mechanistic relationships. We call this evidence-disciplined reasoning. To confirm this effect, we shuffled intermediate path steps while keeping endpoints fixed. Evidence grounding dropped significantly (delta = -0.793, p < 0.001), confirming models genuinely used path structure during reasoning. Our findings show that knowledge graphs support hypothesis generation in two ways: they identify biological endpoint pairs absent from the literature, and their mechanistic paths guide how LLMs reason between them.
Read the full abstract

Computation and Language

Automated Detection and Structuring of Social Tipping Point Evidence in Climate related Documents: A Modular AI Framework

Kavindu Perera, Mohammad Abaeiani, Ekaterina Gilman et al. (+14 more)

The climate literature has grown faster than review teams can read it. That gap matters most for a concept like the environmental social tipping point, the threshold at which a small change triggers rapid, self-reinforcing change in a social system. Evidence of this kind of shift is usually contained in one or two paragraphs within a longer document. As a result, existing text mining tools-which categorize entire documents by topic or highlight isolated claims-leave an expanding set of important evidence without any systematic method for discovery or organization. This paper presents an open and modular transformer-based framework that detects and structures social tipping point evidence at the passage level. The framework joins five components into a single deployable workflow: a DistilBERT boundary splitter for segmentation, an iteratively augmented RoBERTa classifier for detection, a Mistral 7B model that rewrites each detected passage for clarity, a LLaMA 3.2 3B model that rates the passage against five published social tipping point criteria, and a Milvus vector store for semantic retrieval. The system is wrapped in a Streamlit interface backed by MinIO object storage. Evaluated on a 163-passage benchmark labelled by GPT-4.1 and a 51-passage set reviewed by experts, the splitter surpassed three competing methods on a nine-metric composite score (6.137). The tuned RoBERTa model achieved 71.4 percent accuracy with a Cohen's kappa of 0.337 on the full benchmark, and 87.5 percent accuracy with a kappa of 0.742 on passages with labels, outperforming both a climate-focused model and untuned language models.
Read the full abstract

All authors Kavindu Perera, Mohammad Abaeiani, Ekaterina Gilman, Lauri Loven, Mourad Oussalah, Tassos Kanellos, Beatrice Gobbo, Dante Adami, Nicolò Ferriani, Maximiliano Romero, Pierre Rossel, Marc Bonazountas, Christina Deligianni, Nikos Xyderis, Artur Bogucki, Lampros Argyriou, Prasasthy Balasubramanian

Computation and Language

Chopthin-Consensus Power Sampling: A Diversity-Preserving Approach to LLM Decoding

Minoo Ahmadi, Seyedarmin Azizi, Erfan Baghaei Potraghloo et al. (+2 more)

Inference-time power sampling via Sequential Monte Carlo (SMC) can substantially improve large language model (LLM) reasoning without requiring post-training. However, many existing SMC approaches rely on equal-weight resampling, which can aggressively prune low-weight trajectories, discarding potentially correct reasoning paths and degrading the genealogical diversity of the search space. To address this, we introduce Chopthin-Consensus Power Sampling (CCPS). Our method applies the Chopthin resampler to LLM decoding: rather than equalizing weights and forcing unnecessary particle duplication, it enforces an upper bound on the ratio between the largest and smallest weights and carries the unequal weights forward. This targeted intervention preserves a richer set of distinct reasoning paths, keeps the weighted SMC approximation unchanged in conditional expectation, and guarantees a lower bound on the post-resampling effective sample size (ESS). To fully exploit this enriched population, we employ a semantic-majority selection mechanism that merges token-identical final trajectories, clusters semantically equivalent answers, and returns the answer supported by the largest number of distinct trajectories. Evaluating across three open-weight models and five reasoning benchmarks, we show that Chopthin increases oracle coverage in 13 of 15 settings. Combined with semantic-majority selection, CCPS matches or exceeds the final-answer accuracy of the Power-SMC baseline in 14 of 15 settings, delivering absolute gains of up to 10.6 percentage points. These findings demonstrate that diversity-preserving resampling and diversity-aware selection are complementary mechanisms for training-free LLM reasoning. Code is available at github.com/MinooAhmadii/chopthin-consensus-power-sampling.
Read the full abstract

All authors Minoo Ahmadi, Seyedarmin Azizi, Erfan Baghaei Potraghloo, Mehdi Kamal, Massoud Pedram

Computation and Language

Repair Before Reinforce: Context-Augmented Knowledge Graph Reasoning for Multi-Hop Question Answering

Tharaka D. Fonseka and Niraj K. Jha

Question-answering often requires reasoning across multiple connected facts rather than retrieving a single isolated relation. Knowledge graphs (KGs) provide a structured way to represent such facts, but training large language models (LLMs) only on isolated KG head-relation-tail triples may limit their ability to learn the surrounding context needed for multi-hop reasoning. In this work, we propose a context-augmented training framework for multi-hop question-answering. Although generally applicable, we validate the framework in the context of disease-specific KGs, extracted using a reliable KG extraction framework called GraphMERT, for Gastroparesis and Diabetes. For each primary KG triple, we attach supporting triples extracted from the same source text chunk to form a context graph (CG). This creates two supervision settings: KG-grounded supervision, which uses only the target KG triple or path, and CG-grounded supervision, which uses the target KG triple or path together with supporting context triples. We train the Qwen3-14B model using supervised fine-tuning (SFT) under both settings, producing KGModel and CGModel variants. To strengthen the lower-hop factual foundation of the models, we introduce an LLM-judged, history-aware adaptive repair pipeline that identifies unresolved one-hop failures, continually fine-tunes on targeted repair examples, and removes or quarantines problematic noisy triples. This repair stage enables the models to reach 100% accuracy on the cleaned retained one-hop validation sets. Finally, we employ reinforcement learning (RL) using lower-hop question-answer items and evaluate generalization on harder 3-hop, 4-hop, and 5-hop tasks. Across both diseases, context-augmented supervision consistently improves multi-hop performance over KG-only supervision. RL initialized from repaired SFT checkpoints yields larger and more stable gains.
Read the full abstract

Computation and Language

GAUGE: When Not to Trust LLM-as-a-Judge in User-Simulated Evaluation of Task-Oriented Agents

Umesh Bodhwani, Thanh Tran and Kai Wei

Comparing and selecting task-oriented LLM agents increasingly relies on a low-cost offline evaluation gate: persona-driven LLM user-simulators converse with each candidate, an LLM-as-a-judge scores the transcripts, and the higher-scoring agent is promoted. We introduce GAUGE, a reusable offline protocol that measures whether this gate's ranking matches a grounded verifiable reward across 25 agents from six providers on the $\tau^2$-bench and SimulatorArena benchmarks, separating two kinds of evaluation validity that release practices conflate: ranking validity and construct validity. First, a satisfaction-success gap: satisfaction carries essentially no information about task success, as conversations rated satisfied by our blind panel are decorrelated from actual success, with 57.5% of them failing the customer's task, a pattern consistent across five rater populations, both benchmarks, and every subjective dimension we rated. Second, while the gate's ranking is robust across the broad capability span, it loses resolution among the near-equal strong agents: this decision-disagreement rate jumps from $<$1% on wide-reward pairs to 31% on close pairs. The gate is thus human-validated yet mis-anchored. As a remedy, we propose a calibrate-then-trust cadence in which a judge-free completion bit is a zero-cost tripwire for truncation regressions.
Read the full abstract

Computation and Language · Cross-listed from cs.AI

Can LLMs in Draft-Verify-Revise Pipelines Resolve Deictic Ambiguity?

Obinna I. Ekekezie

Draft-verify-revise is a common LLM orchestration pattern for scaling inference-time compute. One LLM drafts, a second critiques the draft and provides feedback, and a third uses that feedback to revise the draft into the final output. As context cascades between stages, LLMs at different stages can resolve a context-dependent expression such as "previous" differently. When that happens, the expression undergoes a deictic shift, a change in what it refers to. This phenomenon was studied with a synthetic dataset of 10 base examples, each rendered in three conditions. Holding the shared components constant, the conditions varied whether the draft stage LLM (the assistant) or the verify stage LLM (the grader) resolved the expression correctly, and how much independent reasoning the revise stage LLM (the meta-evaluator) needed to determine which reading was correct. Six models from three providers were tested across 21 reasoning effort configurations using e-values for sequential testing, in a primary experiment and an ablation experiment that removed error classification labels from the grader's feedback. A separate LLM analyzed the meta-evaluator's stated rationale for each wrong verdict. Balanced accuracy (the unweighted mean of sensitivity and specificity) ranged from 0.156, below chance, to near-perfect. GPT-5.2 rose from 0.156 without reasoning to 0.942 at its highest reasoning effort level, while Gemini 3 Pro stayed above 0.94 at every level. Gemini 3 Pro at low reasoning effort outscored GPT-5.2 at xhigh reasoning effort for roughly 5% of the cost per trial. When the meta-evaluator erred, it tended to rely on surface cues rather than operational reasoning. Context engineers implementing draft-verify-revise pipelines should be wary of deictic shifts and make the intended referent explicit at each stage.
Read the full abstract