Peer Revenue Play credits. Real papers.

Papers

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

Computer Science Computation and Language

cs.CL Watch Computation and Language

Monday 19 January

Computation and Language

BioPulse-QA: A Dynamic Biomedical Question-Answering Benchmark for Evaluating Factuality, Robustness, and Bias in Large Language Models

Kriti Bhattarai, Vipina K. Keloth, Donald Wright et al. (+3 more)

Objective: Large language models (LLMs) are increasingly applied in biomedical settings, and existing benchmark datasets have played an important role in supporting model development and evaluation. However, these benchmarks often have limitations. Many rely on static or outdated datasets that fail to capture the dynamic, context-rich, and high-stakes nature of biomedical knowledge. They also carry increasing risk of data leakage due to overlap with model pretraining corpora and often overlook critical dimensions such as robustness to linguistic variation and potential demographic biases. Materials and Methods: To address these gaps, we introduce BioPulse-QA, a benchmark that evaluates LLMs on answering questions from newly published biomedical documents including drug labels, trial protocols, and clinical guidelines. BioPulse-QA includes 2,280 expert-verified question answering (QA) pairs and perturbed variants, covering both extractive and abstractive formats. We evaluate four LLMs - GPT-4o, GPT-o1, Gemini-2.0-Flash, and LLaMA-3.1 8B Instruct - released prior to the publication dates of the benchmark documents. Results: GPT-o1 achieves the highest relaxed F1 score (0.92), followed by Gemini-2.0-Flash (0.90) on drug labels. Clinical trials are the most challenging source, with extractive F1 scores as low as 0.36. Discussion and Conclusion: Performance differences are larger for paraphrasing than for typographical errors, while bias testing shows negligible differences. BioPulse-QA provides a scalable and clinically relevant framework for evaluating biomedical LLMs.
Read the full abstract

All authors Kriti Bhattarai, Vipina K. Keloth, Donald Wright, Andrew Loza, Yang Ren, Hua Xu

Sunday 18 January

Computation and Language · Cross-listed from cs.DC

Spark-LLM-Eval: A Distributed Framework for Statistically Rigorous Large Language Model Evaluation

Subhadip Mitra

Evaluating large language models at scale remains a practical bottleneck for many organizations. While existing evaluation frameworks work well for thousands of examples, they struggle when datasets grow to hundreds of thousands or millions of samples. This scale is common when assessing model behavior across diverse domains or conducting comprehensive regression testing. We present Spark-LLM-Eval, a distributed evaluation framework built natively on Apache Spark. The system treats evaluation as a data-parallel problem, partitioningexamplesacrossexecutorsandaggregatingresultswithproperstatistical accounting. Beyond raw throughput, we emphasize statistical rigor: every reported metric includes bootstrap confidence intervals, and model comparisons come with appropriate significance tests (paired t-tests, McNemar's test, or Wilcoxon signed-rank, depending on the metric type). The framework also addresses the cost problem inherent in LLM evaluation through content-addressable response caching backed by Delta Lake, which allows iterating on metric definitions without re-running inference. We describe the system architecture, the statistical methodology, and report benchmark results showing linear scaling with cluster size. The framework and all evaluation code are available as open source.
Read the full abstract

Computation and Language

Visualizing and Benchmarking LLM Factual Hallucination Tendencies via Internal State Analysis and Clustering

Nathan Mao, Varun Kaushik, Shreya Shivkumar et al. (+3 more)

Large Language Models (LLMs) often hallucinate, generating nonsensical or false information that can be especially harmful in sensitive fields such as medicine or law. To study this phenomenon systematically, we introduce FalseCite, a curated dataset designed to capture and benchmark hallucinated responses induced by misleading or fabricated citations. Running GPT-4o-mini, Falcon-7B, and Mistral 7-B through FalseCite, we observed a noticeable increase in hallucination activity for false claims with deceptive citations, especially in GPT-4o-mini. Using the responses from FalseCite, we can also analyze the internal states of hallucinating models, visualizing and clustering the hidden state vectors. From this analysis, we noticed that the hidden state vectors, regardless of hallucination or non-hallucination, tend to trace out a distinct horn-like shape. Our work underscores FalseCite's potential as a foundation for evaluating and mitigating hallucinations in future LLM research.
Read the full abstract

All authors Nathan Mao, Varun Kaushik, Shreya Shivkumar, Parham Sharafoleslami, Kevin Zhu, Sunishchal Dev

Computation and Language · Cross-listed from cs.LG

Large Language Models Predict Functional Outcomes after Acute Ischemic Stroke

Anjali K. Kapoor, Anton Alyakin, Jin Vivian Lee et al. (+8 more)

Accurate prediction of functional outcomes after acute ischemic stroke can inform clinical decision-making and resource allocation. Prior work on modified Rankin Scale (mRS) prediction has relied primarily on structured variables (e.g., age, NIHSS) and conventional machine learning. The ability of large language models (LLMs) to infer future mRS scores directly from routine admission notes remains largely unexplored. We evaluated encoder (BERT, NYUTron) and generative (Llama-3.1-8B, MedGemma-4B) LLMs, in both frozen and fine-tuned settings, for discharge and 90-day mRS prediction using a large, real-world stroke registry. The discharge outcome dataset included 9,485 History and Physical notes and the 90-day outcome dataset included 1,898 notes from the NYU Langone Get With The Guidelines-Stroke registry (2016-2025). Data were temporally split with the most recent 12 months held out for testing. Performance was assessed using exact (7-class) mRS accuracy and binary functional outcome (mRS 0-2 vs. 3-6) accuracy and compared against established structured-data baselines incorporating NIHSS and age. Fine-tuned Llama achieved the highest performance, with 90-day exact mRS accuracy of 33.9% [95% CI, 27.9-39.9%] and binary accuracy of 76.3% [95% CI, 70.7-81.9%]. Discharge performance reached 42.0% [95% CI, 39.0-45.0%] exact accuracy and 75.0% [95% CI, 72.4-77.6%] binary accuracy. For 90-day prediction, Llama performed comparably to structured-data baselines. Fine-tuned LLMs can predict post-stroke functional outcomes from admission notes alone, achieving performance comparable to models requiring structured variable abstraction. Our findings support the development of text-based prognostic tools that integrate seamlessly into clinical workflows without manual data extraction.
Read the full abstract

All authors Anjali K. Kapoor, Anton Alyakin, Jin Vivian Lee, Eunice Yang, Annelene M. Schulze, Krithik Vishwanath, Jinseok Lee, Yindalon Aphinyanaphongs, Howard Riina, Jennifer A. Frontera, Eric Karl Oermann

Computation and Language

Construct, Align, and Reason: Large Ontology Models for Enterprise Knowledge Management

Yao Zhang and Hongyin Zhu

Enterprise-scale knowledge management faces significant challenges in integrating multi-source heterogeneous data and enabling effective semantic reasoning. Traditional knowledge graphs often struggle with implicit relationship discovery and lack sufficient semantic understanding for complex question answering. To address these limitations, we introduce a unified construct–align–reason framework, the large ontology model (LOM). We first build a dual-layer enterprise ontology from structured databases and unstructured text, subsequently fusing these sources into a comprehensive enterprise ontology. To enable instruction-aligned reasoning, we propose a unified three-stage training pipeline: ontology instruction fine-tuning to improve structural understanding; text-ontology grounding to strengthen node semantic encoding; and multi-task instruction tuning on ontology-language pairs with curriculum learning to enhance semantic reasoning and generation. We also construct comprehensive training and evaluation datasets covering diverse ontology reasoning tasks. On this benchmark, our 4B-parameter LOM achieves 89.47% accuracy and outperforms DeepSeek-V3.2 on complex graph reasoning, indicating effective fusion of ontology structure and language.
Read the full abstract

Computation and Language

Guided by the Plan: Enhancing Faithful Autoregressive Text-to-Audio Generation with Guided Decoding

Juncheng Wang, Zhe Hu, Chao Xu et al. (+5 more)

Autoregressive (AR) models excel at generating temporally coherent audio by producing tokens sequentially, yet they often falter in faithfully following complex textual prompts, especially those describing complex sound events. We uncover a surprising capability in AR audio generators: their early prefix tokens implicitly encode global semantic attributes of the final output, such as event count and sound-object category, revealing a form of implicit planning. Building on this insight, we propose Plan-Critic, a lightweight auxiliary model trained with a Generalized Advantage Estimation (GAE)-inspired objective to predict final instruction-following quality from partial generations. At inference time, Plan-Critic enables guided exploration: it evaluates candidate prefixes early, prunes low-fidelity trajectories, and reallocates computation to high-potential planning seeds. Our Plan-Critic-guided sampling achieves up to a 10-point improvement in CLAP score over the AR baseline-establishing a new state of the art in AR text-to-audio generation-while maintaining computational parity with standard best-of-N decoding. This work bridges the gap between causal generation and global semantic alignment, demonstrating that even strictly autoregressive models can plan ahead.
Read the full abstract

All authors Juncheng Wang, Zhe Hu, Chao Xu, Siyue Ren, Yuxiang Feng, Yang Liu, Baigui Sun, Shujun Wang

Computation and Language

Disagreement as Data: Reasoning Trace Analytics in Multi-Agent Systems

Elham Tajik, Conrad Borchers, Bahar Shahrokhian et al. (+4 more)

Learning analytics researchers often analyze qualitative student data such as coded annotations or interview transcripts to understand learning processes. With the rise of generative AI, fully automated and human-AI workflows have emerged as promising methods for analysis. However, methodological standards to guide such workflows remain limited. In this study, we propose that reasoning traces generated by large language model (LLM) agents, especially within multi-agent systems, constitute a novel and rich form of process data to enhance interpretive practices in qualitative coding. We apply cosine similarity to LLM reasoning traces to systematically detect, quantify, and interpret disagreements among agents, reframing disagreement as a meaningful analytic signal. Analyzing nearly 10,000 instances of agent pairs coding human tutoring dialog segments, we show that LLM agents' semantic reasoning similarity robustly differentiates consensus from disagreement and correlates with human coding reliability. Qualitative analysis guided by this metric reveals nuanced instructional sub-functions within codes and opportunities for conceptual codebook refinement. By integrating quantitative similarity metrics with qualitative review, our method has the potential to improve and accelerate establishing inter-rater reliability during coding by surfacing interpretive ambiguity, especially when LLMs collaborate with humans. We discuss how reasoning-trace disagreements represent a valuable new class of analytic signals advancing methodological rigor and interpretive depth in educational research.
Read the full abstract

All authors Elham Tajik, Conrad Borchers, Bahar Shahrokhian, Sebastian Simon, Ali Keramati, Sonika Pal, Sreecharan Sankaranarayanan

Computation and Language

A Cloud-based Multi-Agentic Workflow for Science

Anurag Acharya, Timothy Vega, Rizwan A. Ashraf et al. (+3 more)

As Large Language Models (LLMs) become ubiquitous across various scientific domains, their lack of ability to perform complex tasks like running simulations or to make complex decisions limits their utility. LLM-based agents bridge this gap due to their ability to call external resources and tools and thus are now rapidly gaining popularity. However, coming up with a workflow that can balance the models, cloud providers, and external resources is very challenging, making implementing an agentic system more of a hindrance than a help. In this work, we present a domain-agnostic, model-independent workflow for an agentic framework that can act as a scientific assistant while being run entirely on cloud. Built with a supervisor agent marshaling an array of agents with individual capabilities, our framework brings together straightforward tasks like literature review and data analysis with more complex ones like simulation runs. We describe the framework here in full, including a proof-of-concept system we built to accelerate the study of Catalysts, which is highly important in the field of Chemistry and Material Science. We report the cost to operate and use this framework, including the breakdown of the cost by services use. We also evaluate our system on a custom-curated synthetic benchmark and a popular Chemistry benchmark, and also perform expert validation of the system. The results show that our system is able to route the task to the correct agent 90% of the time and successfully complete the assigned task 97.5% of the time for the synthetic tasks and 91% of the time for real-world tasks, while still achieving better or comparable accuracy to most frontier models, showing that this is a viable framework for other scientific domains to replicate.
Read the full abstract

All authors Anurag Acharya, Timothy Vega, Rizwan A. Ashraf, Anshu Sharma, Derek Parker, Robert Rallo

Computation and Language · Cross-listed from cs.SD

SSVD-O: Parameter-Efficient Fine-Tuning with Structured SVD for Speech Recognition

Pu Wang, Shinji Watanabe and Hugo Van hamme

Parameter-efficient fine-tuning (PEFT) is a scalable approach for adapting large speech foundation models to new domains. While methods such as LoRA and its state-of-the-art variants reduce adaptation costs, they typically allocate parameters uniformly across model subspaces, which limits their efficiency and scalability in speech applications. Building on our prior work, this paper introduces SSVD-Outer (SSVD-O), an extension of the structured SVD-guided (SSVD) fine-tuning method. SSVD-O combines input acoustic feature space-associated inner transformations with output semantic feature space-associated outer transformations to enable scalable and balanced adaptation. We conduct the first systematic analysis of parameter budget allocation across model subspaces in PEFT for automatic speech recognition (ASR), and investigate the trade-off between learning and forgetting under constrained resources. SSVD-O is benchmarked against LoRA, DoRA, PiSSA, and SSVD on domain-shifted ASR tasks, including child speech and regional accents, across model scales from 0.1B to 2B within the ESPnet framework. Experimental results show that SSVD-O consistently narrows the performance gap to full fine-tuning while improving generalization and mitigating catastrophic forgetting.
Read the full abstract

Computation and Language · Cross-listed from cs.LG

Dissecting Linear Recurrent Models: How Different Gating Strategies Drive Selectivity and Generalization

Younes Bouhadjar, Maxime Fabre, Felix Schmidt et al. (+1 more)

Linear recurrent neural networks have emerged as efficient alternatives to the original Transformer's softmax attention mechanism, thanks to their highly parallelizable training and constant memory and computation requirements at inference. Iterative refinements of these models have introduced an increasing number of architectural mechanisms, leading to increased complexity and computational costs. Nevertheless, systematic direct comparisons among these models remain limited. Existing benchmark tasks are either too simplistic to reveal substantial differences or excessively resource-intensive for experimentation. In this work, we propose a refined taxonomy of linear recurrent models and introduce SelectivBench, a set of lightweight and customizable synthetic benchmark tasks for systematically evaluating sequence models. SelectivBench specifically evaluates selectivity in sequence models at small to medium scale, such as the capacity to focus on relevant inputs while ignoring context-based distractors. It employs rule-based grammars to generate sequences with adjustable complexity, incorporating irregular gaps that intentionally violate transition rules. Evaluations of linear recurrent models on SelectivBench reveal performance patterns consistent with results from large-scale language tasks. Our analysis clarifies the roles of essential architectural features: gating and rapid forgetting mechanisms facilitate recall, in-state channel mixing is unnecessary for selectivity, but critical for generalization, and softmax attention remains dominant due to its memory capacity scaling with sequence length. Our benchmark enables targeted, efficient exploration of linear recurrent models and provides a controlled setting for studying behaviors observed in large-scale evaluations. Code is available at https://github.com/symseqbench/selectivbench
Read the full abstract

All authors Younes Bouhadjar, Maxime Fabre, Felix Schmidt, Emre Neftci

Computation and Language

Evaluating Contextually Mediated Factual Recall in Multilingual Large Language Models

Yihong Liu, Bingyu Xiong and Hinrich Schütze

Large language models (LLMs) can recall a wide range of factual knowledge across languages. However, existing factual recall evaluations primarily assess fact retrieval in isolation, where the queried entity is explicitly named and the fact is requested directly. In natural language use, facts are often accessed through context, where the relevant entity is introduced only indirectly. In this work, we study contextually mediated factual recall, asking whether LLMs can reliably retrieve factual knowledge when the target entity is embedded in a naturalistic context rather than queried explicitly, across languages. We construct controlled prompts that preserve the underlying fact while introducing referential mediation through contextual sentences. To disentangle contextual effects from name-specific associations, we further compare performance using synthetic names and real names across languages. Evaluating multiple model families in five languages, we find that contextual mediation consistently degrades factual recall, with substantial variation across relations. Larger models are more robust to contextual mediation, exhibiting a reduced performance gap relative to direct queries, while the effect of real names and name origin is mixed and unsystematic. These findings highlight a gap between isolated factual recall and context-dependent language understanding in multilingual LLMs.
Read the full abstract

Computation and Language

Benchmarking Concept-Spilling Across Languages in LLMs

Ilia Badanin, Daniil Dzenhaliou and Imanol Schlag

Multilingual Large Language Models (LLMs) exhibit remarkable cross-lingual abilities, yet often exhibit a systematic bias toward the representations from other languages, resulting in semantic interference when generating content in non-English languages$-$a phenomenon we define as language spilling. This paper presents a novel comparative framework for evaluating multilingual semantic robustness by systematically measuring how models handle polysemous words across languages. Our methodology provides a relative measure of model performance: when required to generate exactly five meanings, both strong and weak models may resort to meanings from dominant languages, but semantically stronger models do so later in the generation sequence, producing more true meanings from the target language before failing, while weaker models resort to dominant-language meanings earlier in the sequence. We evaluate a diverse set of open and closed multilingual LLMs using a structured meaning generation task across nine languages, employing a carefully curated benchmark of 100 high-polysemy English words. Our findings reveal significant variation in semantic robustness across both models and languages, providing a principled ranking system for model comparison without requiring definitive causal attribution of error sources. We contribute both a scalable comparative benchmark for multilingual semantic evaluation and a rigorous validation pipeline$-$critical tools for developing more linguistically balanced AI systems.
Read the full abstract

Computation and Language · Cross-listed from cs.AI

MemeLens: Multilingual Multitask VLMs for Memes

Ali Ezzat Shahroor, Mohamed Bayan Kmainasi, Abul Hasnat et al. (+4 more)

Memes are a dominant medium for online communication and manipulation because meaning emerges from interactions between embedded text, imagery, and cultural context. Existing meme research is distributed across tasks (e.g., hate, misogyny, propaganda, sentiment, humour) and languages, which limits cross-domain generalization. To address this gap, we propose MemeLens, a unified multilingual, multitask explanation-enhanced Vision-Language Model (VLM) for meme understanding. We consolidate $38$ public meme datasets, filter and map dataset-specific labels into a shared taxonomy of $20$ tasks spanning harm, targets, figurative/pragmatic intent, and affect. We present a comprehensive empirical analysis across modeling paradigms, task categories, and datasets. Our findings suggest that robust meme understanding requires multimodal training, varies substantially across semantic categories, and remains sensitive to over-specialization when models are fine-tuned on individual datasets rather than trained in a unified setting. We make the experimental resources (https://github.com/MohamedBayan/MemeLens), model (https://huggingface.co/QCRI/MemeLens-VLM) and datasets (https://huggingface.co/datasets/QCRI/MemeLens) publicly available to the community.
Read the full abstract

All authors Ali Ezzat Shahroor, Mohamed Bayan Kmainasi, Abul Hasnat, Dimitar Dimitrov, Giovanni Da San Martino, Preslav Nakov, Firoj Alam

Computation and Language · Cross-listed from cs.AI

A Survey of Agentic Reasoning for Large Language Models: Towards Recursively Self-Improving and Collective Agents

Tianxin Wei, Ting-Wei Li, Zhining Liu et al. (+26 more)

Reasoning is a fundamental cognitive process underlying inference, problem-solving, and decision-making. While large language models (LLMs) demonstrate strong reasoning capabilities in closed-world settings, they struggle in open-ended and dynamic environments. Agentic reasoning marks a paradigm shift by reframing LLMs as autonomous agents that plan, act, and learn through continual interaction. In this survey, we organize agentic reasoning along three complementary dimensions. First, we characterize environmental dynamics through three layers: foundational agentic reasoning, which establishes core single-agent capabilities including planning, tool use, and search in stable environments; self-evolving agentic reasoning, which studies how agents refine these capabilities through feedback, memory, and adaptation; and collective multi-agent reasoning, which extends intelligence to collaborative settings involving coordination, knowledge sharing, and shared goals. Across these layers, we distinguish in-context reasoning, which scales test-time interaction through structured orchestration, from post-training reasoning, which optimizes behaviors via reinforcement learning and supervised fine-tuning. We further review representative agentic reasoning frameworks across real-world applications and benchmarks, including science, robotics, healthcare, autonomous research, and mathematics. This survey synthesizes agentic reasoning methods into a unified roadmap bridging thought and action, and outlines open challenges and future directions, including personalization, long-horizon interaction, world modeling, scalable multi-agent training, and governance for real-world deployment.
Read the full abstract

All authors Tianxin Wei, Ting-Wei Li, Zhining Liu, Xuying Ning, Ze Yang, Jiaru Zou, Zhichen Zeng, Ruizhong Qiu, Xiao Lin, Dongqi Fu, Zihao Li, Mengting Ai, Duo Zhou, Wenxuan Bao, Yunzhe Li, Gaotang Li, Cheng Qian, Yu Wang, Xiangru Tang, Yin Xiao, Liri Fang, Hui Liu, Xianfeng Tang, Yuji Zhang, Chi Wang, Jiaxuan You, Heng Ji, Hanghang Tong, Jingrui He

Computation and Language

Improving Low-Resource Machine Translation via Round-Trip Reinforcement Learning

Ahmed Attia and Alham Fikri Aji

Low-resource machine translation (MT) has gained increasing attention as parallel data from low-resource language communities is collected, but many approaches for improving low-resource MT remain underexplored. We investigate a self-supervised reinforcement learning fine-tuning for translation in low-resource settings using round-trip bootstrapping with the No Language Left Behind (NLLB) family of models. Our approach translates English into a target low-resource language and then back into English, using a combination of chrF++ and BLEU as the reward function on the reconstructed English sentences. Using the NLLB-MD dataset, we evaluate both the 600M and 1.3B parameter NLLB models and observe consistent improvements for the following languages: Central Aymara, Friulian, Wolof, Dyula, Bhojpuri and Russian. Qualitative inspection of translation outputs indicates increased fluency and semantic fidelity. We argue that our method can further benefit from scale, enabling models to increasingly leverage their pretrained knowledge and continue self-improving. Code available at: https://github.com/Copticoder/MT-via-Round-Trip-RL
Read the full abstract

Computation and Language

DoPE: Decoy Oriented Perturbation Encapsulation Human-Readable, AI-Hostile Documents for Academic Integrity

Ashish Raj Shekhar, Shiven Agarwal, Priyanuj Bordoloi et al. (+3 more)

Multimodal Large Language Models (MLLMs) can directly consume exam documents, threatening conventional assessments and academic integrity. We present DoPE (Decoy-Oriented Perturbation Encapsulation), a document-layer defense framework that embeds semantic decoys into PDF/HTML assessments to exploit render-parse discrepancies in MLLM pipelines. By instrumenting exams at authoring time, DoPE provides model-agnostic prevention (stop or confound automated solving) and detection (flag blind AI reliance) without relying on conventional one-shot classifiers. We formalize prevention and detection tasks, and introduce FewSoRT-Q, an LLM-guided pipeline that generates question-level semantic decoys and FewSoRT-D to encapsulate them into watermarked documents. We evaluate on Integrity-Bench, a novel benchmark of 1826 exams (PDF+HTML) derived from public QA datasets and OpenCourseWare. Against black-box MLLMs from OpenAI and Anthropic, DoPE yields strong empirical gains: a 91.4% detection rate at an 8.7% false-positive rate using an LLM-as-Judge verifier, and prevents successful completion or induces decoy-aligned failures in 96.3% of attempts. We release Integrity-Bench, our toolkit, and evaluation code to enable reproducible study of document-layer defenses for academic integrity.
Read the full abstract

All authors Ashish Raj Shekhar, Shiven Agarwal, Priyanuj Bordoloi, Yash Shah, Tejas Anvekar, Vivek Gupta

Computation and Language · Cross-listed from cs.SD

Multi-Task Instruction Tuning via Data Scheduling for Low-Resource Arabic SpeechLLMs

Hunzalah Hassan Bhatti, Firoj Alam and Shammur Absar Chowdhury

Audio large language models (LLMs) enable unified speech understanding and generation, but adapting them to linguistically complex and dialect-rich settings such as Arabic-English remains challenging. We present a controlled study of multi-task instruction tuning for an Arabic-centric audio LLM across generative tasks, including automatic speech recognition (ASR) and speech and text summarization, as well as discriminative tasks, including dialect identification (DID) and speech emotion recognition (SER), in a resource-constrained setting. To support end-to-end Arabic speech summarization, we introduce AraMega-SSum, the first Arabic speech summarization dataset designed for training and benchmarking Arabic-centric audio LLMs. We compare four training strategies: (i) Uniform Mixing (UM), (ii) Task-Progressive Curriculum (TPC), (iii) Aligner-Based Diverse Sampling (ADS) for training-time batch construction, and (iv) a two-stage TPC->ADS strategy. Our results reveal a clear efficiency-robustness trade-off. TPC achieves the strongest performance on generative tasks, including ASR and summarization. ADS improves paralinguistic tasks but reduces generative stability when used alone. The two-stage TPC->ADS strategy provides the best overall balance, achieving the strongest DID and SER performance while outperforming large proprietary models such as Gemini-2.5-Pro on discriminative tasks. We will publicly release AraMega-SSum together with all experimental resources to support future research in Arabic speech understanding.
Read the full abstract

Computation and Language

Capability-Aware Early-Stage Research Idea Evaluation

Renlong Jie, Chen Chu and Zhen Wang

Predicting the outcomes of research ideas at their conceptual stage (i.e. before significant resources are committed) holds great potential for optimizing scientific resource allocation and research planning. While existing methods rely heavily on finished manuscripts or peer reviews, we propose a novel capability-aware framework that predicts paper acceptance and ratings using only author information and research ideas, without requiring full text or experimental results. Our approach integrates author information, (inferred) capability presentation, and research ideas through a three-way transformer architecture with flexible fusion mechanisms. We also introduce a two-stage architecture for learning the capability representation given the author information and idea. Experiments show that our method significantly outperform the single-way models by finetuning bert-base and bert-large, and the capability predicting significantly increase the predictive accuracy of the final model. The proposed method can be applied in both early-stage research outcome prediction and scientific resource allocation.
Read the full abstract

Computation and Language

Knowing When to Abstain: Medical LLMs Under Clinical Uncertainty

Sravanthi Machcha, Sushrita Yerra, Sahil Gupta et al. (+4 more)

Current evaluation of large language models (LLMs) overwhelmingly prioritizes accuracy; however, in real-world and safety-critical applications, the ability to abstain when uncertain is equally vital for trustworthy deployment. We introduce MedAbstain, a unified benchmark and evaluation protocol for abstention in medical multiple-choice question answering (MCQA) – a discrete-choice setting that generalizes to agentic action selection – integrating conformal prediction, adversarial question perturbations, and explicit abstention options. Our systematic evaluation of both open- and closed-source LLMs reveals that even state-of-the-art, high-accuracy models often fail to abstain with uncertain. Notably, providing explicit abstention options consistently increases model uncertainty and safer abstention, far more than input perturbations, while scaling model size or advanced prompting brings little improvement. These findings highlight the central role of abstention mechanisms for trustworthy LLM deployment and offer practical guidance for improving safety in high-stakes applications.
Read the full abstract

All authors Sravanthi Machcha, Sushrita Yerra, Sahil Gupta, Aishwarya Sahoo, Sharmin Sultana, Hong Yu, Zonghai Yao

Computation and Language

Incentivizing In-depth Reasoning over Long Contexts with Process Advantage Shaping

Miao Peng, Weizhou Shen, Nuo Chen et al. (+3 more)

Reinforcement Learning with Verifiable Rewards (RLVR) has proven effective in enhancing LLMs short-context reasoning, but its performance degrades in long-context scenarios that require both precise grounding and robust long-range reasoning. We identify the "almost-there" phenomenon in long-context reasoning, where trajectories are largely correct but fail at the final step, and attribute this failure to two factors: (1) the lack of high reasoning density in long-context QA data that push LLMs beyond mere grounding toward sophisticated multi-hop reasoning; and (2) the loss of valuable learning signals during long-context RL training due to the indiscriminate penalization of partially correct trajectories with incorrect outcomes. To overcome this bottleneck, we propose DeepReasonQA, a KG-driven synthesis framework that controllably constructs high-difficulty, multi-hop long-context QA pairs with inherent reasoning chains. Building on this, we introduce Long-context Process Advantage Shaping (LongPAS), a simple yet effective method that performs fine-grained credit assignment by evaluating reasoning steps along Validity and Relevance dimensions, which captures critical learning signals from "almost-there" trajectories. Experiments on three long-context reasoning benchmarks show that our approach substantially outperforms RLVR baselines and matches frontier LLMs while using far fewer parameters. Further analysis confirms the effectiveness of our methods in strengthening long-context reasoning while maintaining stable RL training.
Read the full abstract

All authors Miao Peng, Weizhou Shen, Nuo Chen, Chenliang Li, Ming Yan, Jia Li

Computation and Language · Cross-listed from cs.CR

Privacy-Preserving Federated Learning with Verifiable Fairness Guarantees

Mohammed Himayath Ali, Mohammed Aqib Abdullah, Syed Muneer Hussain et al. (+2 more)

Federated learning enables collaborative model training across distributed institutions without centralizing sensitive data; however, ensuring algorithmic fairness across heterogeneous data distributions while preserving privacy remains fundamentally unresolved. This paper introduces CryptoFair-FL, a novel cryptographic framework providing the first verifiable fairness guarantees for federated learning systems under formal security definitions. The proposed approach combines additively homomorphic encryption with secure multi-party computation to enable privacy-preserving verification of demographic parity and equalized odds metrics without revealing protected attribute distributions or individual predictions. A novel batched verification protocol reduces computational complexity from BigO(n^2) to BigO(n \log n) while maintaining (\dparam, \deltap)-differential privacy with dparam = 0.5 and deltap = 10^{-6}. Theoretical analysis establishes information-theoretic lower bounds on the privacy cost of fairness verification, demonstrating that the proposed protocol achieves near-optimal privacy-fairness tradeoffs. Comprehensive experiments across four benchmark datasets (MIMIC-IV healthcare records, Adult Income, CelebA, and a novel FedFair-100 benchmark) demonstrate that CryptoFair-FL reduces fairness violations from 0.231 to 0.031 demographic parity difference while incurring only 2.3 times computational overhead compared to standard federated averaging. The framework successfully defends against attribute inference attacks, maintaining adversarial success probability below 0.05 across all tested configurations. These results establish a practical pathway for deploying fairness-aware federated learning in regulated industries requiring both privacy protection and algorithmic accountability.
Read the full abstract

All authors Mohammed Himayath Ali, Mohammed Aqib Abdullah, Syed Muneer Hussain, Mohammed Mudassir Uddin, Shahnawaz Alam

Computation and Language

System-Mediated Attention Imbalances Make Vision-Language Models Say Yes

Tsan Tsai Chan, Varsha Suresh, Anisha Saha et al. (+2 more)

Vision-language model (VLM) hallucination is commonly linked to imbalanced allocation of attention across input modalities: system, image and text. However, existing mitigation strategies tend towards an image-centric interpretation of these imbalances, often prioritising increased image attention while giving less consideration to the roles of the other modalities. In this study, we evaluate a more holistic, system-mediated account, which attributes these imbalances to functionally redundant system weights that reduce attention to image and textual inputs. We show that this framework offers a useful empirical perspective on the yes-bias, a common form of hallucination in which VLMs indiscriminately respond `yes'. Causally redistributing attention from the system modality to image and textual inputs substantially suppresses this bias, often outperforming existing approaches. We further present evidence suggesting that system-mediated attention imbalances contribute to the yes-bias by encouraging a default reliance on coarse input representations, which are effective for some tasks but ill-suited to others. Taken together, these findings firmly establish system attention as a key factor in VLM hallucination and highlight its potential as a lever for mitigation.
Read the full abstract

All authors Tsan Tsai Chan, Varsha Suresh, Anisha Saha, Michael Hahn, Vera Demberg

Computation and Language

Legal Experts Disagree With Rationale Extraction Techniques for Explaining ECtHR Case Outcome Classification

Mahammad Namazov, Tomáš Koref and Ivan Habernal

Interpretability is critical for applications of large language models (LLMs) in the legal domain, where trust and transparency are essential. A central NLP task in this setting is legal outcome prediction, where models forecast whether a court will find a violation of a given right. We study this task on decisions from the European Court of Human Rights (ECtHR), introducing a new ECtHR dataset with carefully curated positive (violation) and negative (non-violation) cases. Existing works propose both task-specific approaches and model-agnostic techniques to explain downstream performance, but it remains unclear which techniques best explain legal outcome prediction. To address this, we propose a comparative analysis framework for model-agnostic interpretability methods. We focus on two rationale extraction techniques that justify model outputs with concise, human-interpretable text fragments from the input. We evaluate faithfulness via normalized sufficiency and comprehensiveness metrics, and plausibility via legal expert judgments of the extracted rationales. We also assess the feasibility of using LLM-as-a-Judge, using these expert evaluations as reference. Our experiments on the new ECtHR dataset show that models' "reasons" for predicting violations differ substantially from those of legal experts, despite strong faithfulness scores. The source code of our experiments is publicly available at https://github.com/trusthlt/IntEval.
Read the full abstract

Computation and Language · Cross-listed from cs.CR

De-Anonymization at Scale via Tournament-Style Attribution

Lirui Zhang and Huishuai Zhang

As LLMs rapidly advance and enter real-world use, their privacy implications are increasingly important. We study an authorship de-anonymization threat: using LLMs to link anonymous documents to their authors, potentially compromising settings such as double-blind peer review. We propose De-Anonymization at Scale (DAS), a large language model-based method for attributing authorship among tens of thousands of candidate texts. DAS uses a sequential progression strategy: it randomly partitions the candidate corpus into fixed-size groups, prompts an LLM to select the text most likely written by the same author as a query text, and iteratively re-queries the surviving candidates to produce a ranked top-k list. To make this practical at scale, DAS adds a dense-retrieval prefilter to shrink the search space and a majority-voting style aggregation over multiple independent runs to improve robustness and ranking precision. Experiments on anonymized review data show DAS can recover same-author texts from pools of tens of thousands with accuracy well above chance, demonstrating a realistic privacy risk for anonymous platforms. On standard authorship benchmarks (Enron emails and blog posts), DAS also improves both accuracy and scalability over prior approaches, highlighting a new LLM-enabled de-anonymization vulnerability.
Read the full abstract

Computation and Language

NADIR: Differential Attention Flow for Non-Autoregressive Transliteration in Indic Languages

Lakshya Tomar, Vinayak Abrol and Puneet Agarwal

In this work, we argue that not all sequence-to-sequence tasks require the strong inductive biases of autoregressive (AR) models. Tasks like multilingual transliteration, code refactoring, grammatical correction or text normalization often rely on local dependencies where the full modeling capacity of AR models can be overkill, creating a trade-off between their high accuracy and high inference latency. While non-autoregressive (NAR) models offer speed, they typically suffer from hallucinations and poor length control. To explore this trade-off, we focus on the multilingual transliteration task in Indic languages and introduce NADIR, a novel NAR architecture designed to strike a balance between speed and accuracy. NADIR integrates a Differential Transformer and a Mixture-of-Experts mechanism, enabling it to robustly model complex character mappings without sequential dependencies. NADIR achieves over a 13x speed-up compared to the state-of-the-art AR baseline. It maintains a competitive mean Character Error Rate of 15.78%, compared to 14.44% for the AR model and 21.88% for a standard NAR equivalent. Importantly, NADIR reduces Repetition errors by 49.53%, Substitution errors by 24.45%, Omission errors by 32.92%, and Insertion errors by 16.87%. This work provides a practical blueprint for building fast and reliable NAR systems, effectively bridging the gap between AR accuracy and the demands of real-time, large-scale deployment.
Read the full abstract

Computation and Language

LR-DWM: Efficient Watermarking for Diffusion Language Models

Ofek Raban, Ethan Fetaya and Gal Chechik

Watermarking (WM) is a critical mechanism for detecting and attributing AI-generated content. Current WM methods for Large Language Models (LLMs) are predominantly tailored for autoregressive (AR) models: They rely on tokens being generated sequentially, and embed stable signals within the generated sequence based on the previously sampled text. Diffusion Language Models (DLMs) generate text via non-sequential iterative denoising, which requires significant modification to use WM methods designed for AR models. Recent work proposed to watermark DLMs by inverting the process when needed, but suffers significant computational or memory overhead. We introduce Left-Right Diffusion Watermarking (LR-DWM), a scheme that biases the generated token based on both left and right neighbors, when they are available. LR-DWM incurs minimal runtime and memory overhead, remaining close to the non-watermarked baseline DLM while enabling reliable statistical detection under standard evaluation settings. Our results demonstrate that DLMs can be watermarked efficiently, achieving high detectability with negligible computational and memory overhead.
Read the full abstract

Computation and Language

A Scalable Entity-Based Framework for Auditing Bias in Large Language Models

Akram Elbouanani, Aboubacar Tuo and Adrian Popescu

Existing approaches to bias evaluation in large language models (LLMs) trade ecological validity for statistical control, relying either on artificial prompts that poorly reflect real-world use or on naturalistic tasks that lack scale and rigor. We introduce a scalable bias-auditing framework that uses named entities as controlled probes to measure systematic disparities in model behavior. Synthetic data enables us to construct diverse, controlled inputs, and we show that it reliably reproduces bias patterns observed in natural text, supporting its use for large-scale analysis. Using this framework, we conduct the largest bias audit to date, comprising 1.9 billion data points across multiple entity types, tasks, languages, models, and prompting strategies. We find consistent patterns: models penalize right-wing politicians and favor left-wing politicians, prefer Western and wealthier countries over the Global South, favor Western companies, and penalize firms in the defense and pharmaceutical sectors. While instruction tuning reduces bias, increasing model scale amplifies it, and prompting in Chinese or Russian does not mitigate Western-aligned preferences. These findings highlight the need for systematic bias auditing before deploying LLMs in high-stakes applications. Our framework is extensible to other domains and tasks, and we make it publicly available to support future work.
Read the full abstract

Computation and Language

TaxoBench: Can Deep Research Agents Retrieve and Organize? Evaluating the Synthesis Gap with Expert Taxonomies

Ming Zhang, Jiabao Zhuang, Wenqing Jing et al. (+15 more)

Deep Research Agents increasingly automate survey writing, yet existing benchmarks do not jointly test whether they retrieve the papers experts consider essential and organize those papers into paper-grounded taxonomies. We introduce TaxoBench, a benchmark built from 72 highly cited LLM surveys, 3,815 cited papers, and their expert-authored taxonomies. TaxoBench evaluates systems in two settings: Deep Research mode measures end-to-end retrieval and organization from a topic, while Bottom-Up mode provides the expert paper set and isolates organization. We evaluate leaf-level assignments with ARI and V-Measure and hierarchy-level structure with US-TED, US-NTED, and Sem-Path. Across 7 Deep Research Agents and 16 LLM configurations, the best agent retrieves only 20.92% of expert-cited papers, and none of 70 standard Bottom-Up runs reaches the experts' average depth of 4.86. A controlled probe shows that models which match this depth do so by fragmenting the taxonomy, reducing alignment with the expert reference. We further find that raw Sem-Path remains near a no-organization floor even when a newer model generation gains 3.68 pp ARI; after depth matching, humans lead on all 10 matched surveys by 13.27 pp. These results identify retrieval and hierarchical organization as separate bottlenecks and show why hierarchy metrics must be calibrated before they are used to compare models.
Read the full abstract

All authors Ming Zhang, Jiabao Zhuang, Wenqing Jing, Kexin Tan, Ziyu Kong, Jingyi Deng, Yujiong Shen, Yuhui Wang, Zhenghao Xiang, Qiyuan Peng, Mingqi Wu, Long Ma, Zhangyue Yin, Shihan Dou, Maxm Pan, Tao Gui, Qi Zhang, Xuanjing Huang

Computation and Language · Cross-listed from cs.CR

Zero-Shot Embedding Drift Detection: A Lightweight Defense Against Prompt Injections in LLMs

Anirudh Sekar, Mrinal Agarwal, Rachel Sharma et al. (+4 more)

Prompt injection attacks have become an increasing vulnerability for LLM applications, where adversarial prompts exploit indirect input channels such as emails or user-generated content to circumvent alignment safeguards and induce harmful or unintended outputs. Despite advances in alignment, even state-of-the-art LLMs remain broadly vulnerable to adversarial prompts, underscoring the urgent need for robust, productive, and generalizable detection mechanisms beyond inefficient, model-specific patches. In this work, we propose Zero-Shot Embedding Drift Detection (ZEDD), a lightweight, low-engineering-overhead framework that identifies both direct and indirect prompt injection attempts by quantifying semantic shifts in embedding space between benign and suspect inputs. ZEDD operates without requiring access to model internals, prior knowledge of attack types, or task-specific retraining, enabling efficient zero-shot deployment across diverse LLM architectures. Our method uses adversarial-clean prompt pairs and measures embedding drift via cosine similarity to capture subtle adversarial manipulations inherent to real-world injection attacks. To ensure robust evaluation, we assemble and re-annotate the comprehensive LLMail-Inject dataset spanning five injection categories derived from publicly available sources. Extensive experiments demonstrate that embedding drift is a robust and transferable signal, outperforming traditional methods in detection accuracy and operational efficiency. With greater than 93% accuracy in classifying prompt injections across model architectures like Llama 3, Qwen 2, and Mistral and a false positive rate of <3%, our approach offers a lightweight, scalable defense layer that integrates into existing LLM pipelines, addressing a critical gap in securing LLM-powered systems to withstand adaptive adversarial threats.
Read the full abstract

All authors Anirudh Sekar, Mrinal Agarwal, Rachel Sharma, Akitsugu Tanaka, Jasmine Zhang, Arjun Damerla, Kevin Zhu

Computation and Language · Cross-listed from cs.MA

Rethinking the Value of Multi-Agent Workflow: A Strong Single Agent Baseline

Jiawei Xu, Arief Koesdwiady, Sisong Bei et al. (+8 more)

Recent advances in LLM-based multi-agent systems (MAS) show that workflows composed of multiple LLM agents with distinct roles, tools, and communication patterns can outperform single-LLM baselines on complex tasks. However, most frameworks are homogeneous, where all agents share the same base LLM and differ only in prompts, tools, and positions in the workflow. This raises the question of whether such workflows can be simulated by a single agent through multi-turn conversations. We investigate this across seven benchmarks spanning coding, mathematics, general question answering, domain-specific reasoning, and real-world planning and tool use. Our results show that a single agent can reach the performance of homogeneous workflows with an efficiency advantage from KV cache reuse, and can even match the performance of an automatically optimized heterogeneous workflow. Building on this finding, we propose OneFlow, an algorithm that automatically tailors workflows for single-agent execution, reducing inference costs compared to existing automatic multi-agent design frameworks without trading off accuracy. These results position the single-LLM implementation of multi-agent workflows as a strong baseline for MAS research. We also note that single-LLM methods cannot capture heterogeneous workflows due to the lack of KV cache sharing across different LLMs, highlighting future opportunities in developing truly heterogeneous multi-agent systems.
Read the full abstract

All authors Jiawei Xu, Arief Koesdwiady, Sisong Bei, Yan Han, Baixiang Huang, Dakuo Wang, Yutong Chen, Zheshen Wang, Peihao Wang, Pan Li, Ying Ding