Computation and Language
Relational Priors as Convergence Pressure in LLM-Based Multi-Agent Systems
Ming Shen, Chao Shang, Sadat Shahriar et al. (+4 more)
Large
language model-based multi-
agent systems (LLM-MAS) are designed through roles, debate protocols, and aggregation rules. These choices create implicit expectations of trust, skepticism, deference, or collaboration. We make inter-
agent relations explicit as signed pairwise priors, rendered in natural
language and added to system
prompts while keeping the task protocol fixed. Across commons governance and multi-
agent debate, these priors change how readily
agents coordinate or agree, a pattern we call convergence pressure. More positive relations generally improve sustainability within the GovSim relational-prior sweep and increase consensus on subjective questions. These improvements over negative relations differ from gains over no-prior prompting. On objective QA, fully positive priors usually yield lower final-answer accuracy than the no-prior baseline, and some conditions produce more frequent but less accurate consensus. Effects depend on model backbone, relation type, and
topology; explicitly neutral relations also produce different outcomes from omitting relational framing. Relational priors are therefore task-specific interventions and diagnostic probes of sensitivity to social framing. Evaluate them against a no-prior baseline using the task's primary metric, report accuracy and consensus correctness alongside agreement on objective tasks, and keep no-prior prompting as the default for accuracy-centric tasks unless validation supports a relational prior.
Read the full abstract ▾
All authors Ming Shen, Chao Shang, Sadat Shahriar, Devang Kulshreshtha, Yi Zhang, Sandesh Swamy, Yanjun Qi
Computation and Language
On the Diversity of Analogy Making in Large Language Models
Yuanhao Shen, Daniel Xavier de Sousa, Caio César Sifuentes Barcelos et al. (+2 more)
Large
Language Models (LLMs) have demonstrated remarkable potential for analogy making, a core cognitive capability that drives novelty and creativity. While prior research has extensively investigated the applications and underlying mechanisms of LLM-based analogy making, its output diversity remains largely unexplored, despite being essential for broadening cross-domain connections and fostering scientific innovation. In this work, we present a comprehensive evaluation of analogy diversity across ten state-of-the-art open- and closed-source LLMs. Our findings highlight a concerning issue of domain homogeneity, a prevalent tendency for LLMs to generate analogies from a narrow set of target domains, limiting both inter-query and intra-model diversity. Furthermore, our analysis reveals a fundamental trade-off in existing LLM diversity-enhancement methods: increasing output diversity often comes at the expense of output quality. Finally, our causal analysis of LLM information flow reveals substantial differences in the model-sensitive regions governing analogy diversity across LLMs, suggesting a potential mechanism for the observed diversity-quality trade-off. To our knowledge, this is among the first studies to systematically investigate output diversity in LLM-based analogy making.
Read the full abstract ▾
All authors Yuanhao Shen, Daniel Xavier de Sousa, Caio César Sifuentes Barcelos, Hongyu Guo, Xiaodan Zhu
Computation and Language · Cross-listed from cs.LG
Agentic Reinforcement Learning with Self-Distilled Reward Shaping
Ranxu Zhang, Guinan Chen, Chenshaodong et al. (+5 more)
Agentic reinforcement learning enables LLM
agents to learn through interaction, but sparse trajectory-level rewards reveal success without identifying which intermediate decisions deserve credit. Training-only privileged skills can provide denser supervision by allowing the same frozen policy snapshot to rescore fixed
tokens from skill-free trajectories while conditioned on task-matched procedural skills. Existing methods, however, do not jointly calibrate teacher scores across interaction steps, relate teacher confidence to realized returns, and integrate the resulting signal into native reward-to-advantage construction. We introduce Agentic Reinforcement Learning with Self-Distilled Reward Shaping (ADRS), a framework for constructing return-associated
token-level credit for multi-turn
language agents. ADRS centers and normalizes privileged
token scores within each step, modulates them with a return-associated Teacher Value Advantage (TVA) gate based on within-group confidence–return association, and incorporates the gated
token signal into native RL credit construction. Together, these components determine what the teacher prefers, when that preference is return-relevant, and how it enters the native reinforcement-learning credit path, while keeping rollouts and inference skill-free. Finally, experiments across three interactive
benchmarks show that ADRS consistently improves performance on long-horizon tasks, with gains persisting across RL backbones, reduced-data settings, unseen tasks, and extended training. For anonymous review, our code is available at the following the link: https://github.com/gitrxh/ADRS-arxiv
Read the full abstract ▾
All authors Ranxu Zhang, Guinan Chen, Chenshaodong, Jinghao Lin, Xiaozhou Xu, Sunzhe, Yanyong Zhang, Chao Wang
Computation and Language · Cross-listed from cs.AI
Reachability Is Not Realization: Tracing the Sources of LLM Benchmark Gains
Yanchao Li, Wanhao Liu, Jiaqing Xie et al. (+4 more)
Benchmark gains are often treated as evidence of greater LLM capability. Yet the same gain can reflect different changes in model behavior. A model may reach new answers, or produce answers that were already within reach. Aggregate scores do not distinguish these changes question by question. We establish a question-level audit under fixed budgets, temperatures, and answer formats. A question is realized when the default deployment procedure produces the correct answer. A question is reachable when a specified probe finds that answer within a fixed budget. We first test whether inference-time layer routing can expand reachability. Under a matched budget, random routes match or exceed structured search in all 43 model and task settings. Answer-blind procedures retain almost none of this gain, which instead requires access to the correct answer. We then ask why reachable answers sometimes fail to appear. Across six cases spanning 0.5B to 31B, silencing one identified MLP block repairs 68 to 92 percent of a predefined failure set. We next test whether training closes the gap by expanding reachability. In five of six matched evaluations, deployed performance rises while the reachable ceiling remains flat or falls. For DAPO, the deployed score rises by 14.7 points while the reachable ceiling falls by 13.3 points. Across the settings we audit, realization and reachability therefore do not always change together. Claims of capability expansion should report both realized performance and reachability under matched evaluation conditions. Code is available at https://github.com/LiZaiyuan0619/reachability-not-realization
Read the full abstract ▾
All authors Yanchao Li, Wanhao Liu, Jiaqing Xie, Ben Gao, Yanbo Wang, Tianfan Fu, Yuqiang Li
Computation and Language · Cross-listed from eess.AS
GROW: Group-Relative Advantage-Weighted On-Policy Reinforcement Learning of Autoregressive-Diffusion Text-to-Speech model
Guanrou Yang, Tian Tan, Qian Chen et al. (+8 more)
Reinforcement learning for flow-matching text-to-
speech is complicated by deterministic ODE sampling: trajectory-level policy-gradient methods typically convert the ODE into an SDE and track per-step likelihood ratios, introducing
stochastic perturbations and substantial overhead. We propose GROW, a group-relative advantage-weighted on-policy RL method that acts directly on the standard flow-matching objective. For each
prompt, GROW samples a group of on-policy utterances, separately standardizes intelligibility and speaker-similarity rewards within the group, and combines them to reweight flow-matching regression. A Wasserstein-2 velocity penalty anchors the updated model to a frozen pretrained reference. A group-mean reward baseline is introduced to convert reward weighting into advantage weighting. For strong pretrained TTS models with concentrated rewards, positive exponential weighting is dominated by reward-agnostic self-imitation, whereas a zero-mean signed advantage preserves effective within-group credit assignment. Instantiated on DiTAR and evaluated on LibriSpeech and Seed-TTS EN/ZH, GROW reduces average WER from 2.016 to 1.558 and raises speaker similarity from 0.676 to 0.715 while keeping UTMOS. With 10-NFE training rollouts and 32-NFE evaluation, GROW retains comparable performance while training 2.9x faster than 32-NFE DiTAR-GRPO. We will open-source complete GROW codes, faithful DiTAR reproduction, and all model checkpoints.
Read the full abstract ▾
All authors Guanrou Yang, Tian Tan, Qian Chen, Ziyang Ma, Yakun Song, Zhikang Niu, Qi Chen, Wenming Tu, Haitao Li, Shan Yang, Xie Chen
Computation and Language
ICO: Enhancing Semantic-Shift Jailbreaks via Iterative Context Optimization
Hujian Zhu, Yihao Huang, Felix Juefei-Xu et al. (+5 more)
Foundation models have achieved remarkable success across diverse tasks, but they remain vulnerable. To investigate such vulnerabilities, semantic-shift jailbreaks have recently emerged as a promising attack paradigm. They bypass explicit safety mechanisms by replacing harmful terms in original harmful questions with benign alternatives and leveraging contextual information to induce the target model to reinterpret these alternatives as their corresponding harmful concepts. However, existing semantic-shift jailbreaks often achieve limited effectiveness. In this work, we reveal that this limitation arises from overlooking the semantic-shift capability of contexts. Through systematic analysis, we find that contexts exhibit substantially different abilities in inducing semantic shifts: contexts with stronger semantic-shift capabilities are more likely to guide models toward recovering harmful meanings and achieving successful jailbreaks. Based on this finding, we systematically identify and distill the characteristics of effective contexts and propose a black-box context-aware semantic-shift jailbreak framework with Iterative Context Optimization (ICO). In each iteration, ICO leverages these characteristics and feedback from the target model to optimize contexts. Extensive experiments on three
datasets and eight target foundation models demonstrate that ICO consistently outperforms eight state-of-the-art baselines, achieving an average attack success rate of 74.6%.
Read the full abstract ▾
All authors Hujian Zhu, Yihao Huang, Felix Juefei-Xu, Xinfeng Li, Peng Zeng, Simeng Qin, Qing Guo, Geguang Pu
Computation and Language · Cross-listed from cs.CY
EduClaw-Bench: A Long-Horizon Benchmark for Pedagogical LLM Agents with Simulated Learners
Unggi Lee, Sookbun Lee, Yeil Jeong et al. (+3 more)
Large
language models (LLMs) power educational applications from tutoring to essay scoring, but each is a point solution to a single task, and only recently have these point solutions been integrated into
agents operating over a learning management system (LMS). Yet tutoring is long-horizon, since a learner improves over days and weeks rather than in a single turn, and no
benchmark evaluates an
agent tutor across a sustained relationship. We introduce EduClaw-Bench, a
benchmark that places an
agent tutor in a continuous 30-day relationship with a simulated learner grounded in knowledge tracing (KT), whose knowledge-concept mastery, from a KT model trained on real-student data, drives its answers and is probed for learning gain across 55 scenarios. Each
agent is scored on three primary axes (learning gain, responsiveness, and helpfulness) and two curriculum-design axes (Gagné and Rosenshine), with helpfulness and the curriculum axes judged by a cross-family panel of three LLM judges. Evaluating 10
agent adapters over three base-model tiers yields two findings that single-tier, single-session evaluation cannot reach. First, tutoring quality belongs to the base model and the
agent harness together rather than either alone. Second, almost no combination sustains good tutoring over the full horizon. A calibration check ($\text{ECE}=0.049$) and a live-classroom field study confirm that the simulated learner and its measurements track reality. Our work is a step toward trustworthy AI tutors for future education.
Read the full abstract ▾
All authors Unggi Lee, Sookbun Lee, Yeil Jeong, Eunjoo Lee, Minchul Shin, Hoilym Kwon
Computation and Language
Aligning Large Vision-Language Models at Test Time: A Trajectory-Guided Structured Sampling Approach
Tianbao Jiang, Weicong Ni, Gerard de Melo et al. (+1 more)
Post-training reinforcement learning (RL)
algorithms are commonly used to align large
vision-
language models (LVLMs) with human intent and the requirements of visual
reasoning tasks. However, existing RL-based
alignment methods are often resource-intensive and encounter mismatches between training objectives and inference-time distributions. To bridge this gap, we propose a novel test-time
alignment approach that leverages trajectory-guided structured sampling for dynamic inference-time refinement, achieving better
alignment with visual grounding and ensuring logical consistency. Our approach begins with curating a
reasoning memory bank via a trajectory learning
algorithm, which decomposes complex question solving into ordered sequences of predefined
reasoning patterns. It subsequently accomplishes inference-time
alignment by first collecting trajectories from
reasoning memory bank to establish a global structural
reasoning prior, and then using an iterative Markov Chain Monte Carlo (MCMC)
algorithm for localized multi-objective refinement of the
reasoning trace. Experiments across multiple
multimodal reasoning datasets demonstrate that our approach significantly improves accuracy without incurring prohibitive inference overhead. These results establish trajectory-guided test-time sampling as a scalable and effective alternative to traditional post-training
alignment, particularly for complex visual
reasoning tasks.
Read the full abstract ▾
All authors Tianbao Jiang, Weicong Ni, Gerard de Melo, Linlin Wang
Computation and Language · Cross-listed from cs.AI
Evidence-Grounded Multimodal Knowledge Graph Construction for Multi-Lecture Educational Reasoning
Sahil Al Farib, Momota Ahsana Meem, Sheikh Redwanul Islam et al. (+1 more)
Lecture
videos distribute knowledge across
speech, slide text, diagrams, equations, and presentation order, which transcript-only
retrieval does not fully preserve. This paper presents an evidence-grounded
multimodal pipeline that transcribes lectures, selects semantic anchors, applies optical character recognition (OCR), and uses a
vision-
language model to extract only concepts and typed relationships supported by transcript, OCR, or visual evidence. Mentions are validated and canonicalized into a provenance-rich knowledge graph. On three neural-network lectures, the pipeline processed 3,118 frames, 756 transcript segments, and 559 anchors. It retained 1,022 concept and 312 relationship mentions, yielding 172 canonical concepts and 282 relationships with 90.38% endpoint coverage. A preliminary three question
retrieval test achieved 100% top-1 and top-3 accuracy and 100% mean top-5 recall. The contribution is an auditable construction method rather than a state-of-the-art performance claim.
Read the full abstract ▾
All authors Sahil Al Farib, Momota Ahsana Meem, Sheikh Redwanul Islam, Md. Tanvir Raihan
Computation and Language
ANCHOR-RE: An Agentic Biomedical Relation Extraction Framework with Retrieval Augmentation and Error-Driven Verification
Shufan Ming, Yikun Han, Gibong Hong et al. (+2 more)
Biomedical relation extraction (BioRE) supports knowledge graph construction and hypothesis generation. In real-world settings, most candidate entity pairs are unrelated. Symbolic systems such as SemRep offer high precision and interpretability but limited recall, whereas large
language models (LLMs) can produce false-positive relations. We developed ANCHOR-RE, an agentic framework that combines retrieved demonstrations, ontology-guided semantic constraints, external biomedical knowledge, and error-driven verification to improve BioRE reliability without task-specific fine-tuning. We evaluated ANCHOR-RE on SemRepGS, DDI, and ChemProt using proprietary and open-weight LLMs. To assess generalizability while reducing potential pretraining contamination, we also conducted a post-cutoff temporal precision evaluation using 100 biomedical articles published in 2026. With the proprietary backbone, ANCHOR-RE improved micro-F1 over direct prompting from 0.654 to 0.676 on SemRepGS, from 0.769 to 0.881 on DDI, and from 0.937 to 0.941 on ChemProt, outperforming previously reported inference-only methods on DDI and ChemProt. Gains were also observed with the open-weight backbone. Additional experiments demonstrated cross-family model compatibility and sensitivity to external knowledge quality. Manual assessment of 500 randomly sampled positive predictions from the post-cutoff set yielded 69% precision. These results show that contextual evidence and error-driven verification can improve BioRE reliability across
benchmarks, model families, and newly published literature, supporting ANCHOR-RE as a practical training-free approach to biomedical literature mining.
Read the full abstract ▾
All authors Shufan Ming, Yikun Han, Gibong Hong, Rui Zhang, Halil Kilicoglu
Computation and Language
Internalizing Academic Writing Workflows for Introduction Generation via Struct-Aware Policy Learning
Meicong Zhang, Tiancheng Su, Jiahao Cheng et al. (+3 more)
Generating a rigorous paper introduction with large
language models (LLMs) remains challenging, since it requires coordinating background, gap identification, method and contribution within a coherent narrative. Existing solutions externalize this process as multi-stage
prompts or
agent workflows which are expensive and vulnerable to cross-stage drift. We propose StructPO, a struct-aware policy learning framework that internalizes the entire multi-stage writing workflow into a single-pass policy controlled by explicit stage
tokens. StructPO introduces struct-aware credit assignment to decouple local stage quality from global coherence and refinement-guided optimization to internalize revision behavior into the first-pass policy. Experiments show that StructPO improves semantic
alignment, structural rationality and inference efficiency over workflow-based baselines, generalizes to out-of-domain settings, and remains competitive with GPT-5.1 in human evaluation when scaled to Qwen3-32B. These results show that internalizing academic writing workflows through fine-grained policy optimization offers a viable alternative to costly external orchestration.
Read the full abstract ▾
All authors Meicong Zhang, Tiancheng Su, Jiahao Cheng, Guoxiu He, Xinqi Tao, Dejia Song
Computation and Language · Cross-listed from cs.CR
DP-MemView: A Memory Interface for Attribute-Level Transcript Privacy in Long-Term LLM Agents
Jong Wook Kim, Byoungjae Min, Kennedy Edemacu et al. (+3 more)
Long-term
memory enables persistent personalization in LLM
agents, but repeated
memory-conditioned responses can cumulatively reveal protected attributes even when they are never stated explicitly. We formalize this threat as adaptive transcript
privacy and introduce DP-MemView, a differentially private interface that privately selects public response-conditioning views and exposes those views—rather than raw
memory—to the response LLM. Each private selection is charged to every protected attribute whose
memory group intersects the read set. Per-attribute ledgers block any selection that would exceed its cap and return a fixed generic view instead. Under an explicit interface contract, we prove pure B_a-DP for the entire adaptive transcript. We also extend the result to stores that differ across multiple protected groups and bound how much observing the transcript can change an adversary's prior odds. We evaluate the online and preallocated modes with three response LLMs on a controlled adjacent-store
benchmark and a public-corpus transfer track. Both modes keep transcript distinguishability near chance while preserving target-required personalization and overall response quality. Further diagnostics show that removing key safeguards causes mismatched output support, missing ledger charges, revealing side channels, or growing long-horizon leakage.
Read the full abstract ▾
All authors Jong Wook Kim, Byoungjae Min, Kennedy Edemacu, Yoonhyuk Choi, Sae-Hong Cho, Beakcheol Jang
Computation and Language
From SQL Errors to Concept Gaps: An AI-Powered Knowledge Graph Analytics Platform for Personalized Feedback
Abdulrahman AlRabah, Weijian Zhou, Xing Gao et al. (+1 more)
This innovative practice full paper describes an AI-powered knowledge graph platform that connects SQL errors to conceptual gaps in undergraduate and graduate database systems courses. Students learning Structured Query
Language (SQL) frequently struggle with semantic errors that reflect conceptual misunderstandings rather than syntax mistakes. A query may execute yet return incorrect results due to gaps spanning related concepts; misusing NATURAL JOIN in place of an explicit subquery reflects intertwined misunderstandings of JOIN, GROUP BY, and HAVING. Autograding systems detect correctness but provide surface-level feedback without connecting errors to the conceptual structure of the course. Educational knowledge graph research has shown the value of structured concept representations for curriculum analysis and adaptive learning, but these approaches have not been applied to diagnosing SQL misconceptions from student submissions. We present a platform that automatically extracts course concepts and relations from instructional materials, links them to student submission traces through a graph database, and classifies errors at the concept level. We evaluate the platform across two database systems courses at two universities, one using real student submissions and one using simulated submissions, through an expert study with five participants and an automated evaluation using an LLM as a judge. Results show that 95.7% of extracted nodes were rated as at least somewhat valid and 63.8% of triplets were rated fully correct. Expert feedback confirmed that the generated graphs align with instructor mental models and that mapping errors to course concepts provides actionable diagnostic insight; evaluating impact on student learning remains future work.
Read the full abstract ▾
All authors Abdulrahman AlRabah, Weijian Zhou, Xing Gao, Abdussalam Alawini
Computation and Language · Cross-listed from cs.LG
Convex-Hull-Neighborhood Smooth Dual Generalization: Controlling Local Correction Propagation in Offline RL
Yi Yang, Zhennan Chen, Mingfeng Lv et al. (+3 more)
Offline reinforcement learning (offline RL) can benefit from nearby out-of-distribution (OOD) actions, but estimation errors at these actions may be amplified by bootstrapping. Existing regularization and local-generalization methods control either the admissible OOD region or the influence of generalized targets, often through separate mechanisms. We propose Convex Hull Neighborhood Smooth Dual Generalization (CSDG), which expresses the Bellman backup as an in-sample value target plus a CHN-local correction. This formulation makes the generalized contribution explicit and separates it from the in-sample reference path. The correction is obtained by smoothing in-sample-oriented and OOD-oriented candidates sampled at different perturbation radii. A mixture coefficient lambda scales its contribution to each backup, while the recursive discount remains gamma. Under boundedness and fixed perturbation kernels, we derive an exact one-step correction identity, a time-varying iterate bound, and a fixed-point bound that depends only on the branch discrepancy at the fixed point. We further characterize the implicit policies induced by the idealized operators and give a conditional non-degradation criterion. The practical
algorithm approximates these quantities using asymmetric bounded noise and expectile regression, without exact support classification or an additional pessimistic OOD penalty. Experiments on Gym-MuJoCo and AntMaze show strong aggregate performance and stable value estimation. Code is available at: https://github.com/YOUNG-fnxm/CSDG
Read the full abstract ▾
All authors Yi Yang, Zhennan Chen, Mingfeng Lv, Hanlei Li, Zhengsen Ruan, Lvqing Yang
Computation and Language
HomoEnsNER: Does Language Alignment Outperform Architectural Complexity in Gujarati Named Entity Recognition?
Chandrakant K. Bhogayata
Named Entity Recognition (NER) for Gujarati remains underexplored, hindered by the absence of capitalization cues, rich morphology, lexical ambiguity, and free word order. Prior ensemble work has emphasized architectural diversity by combining heterogeneous classifiers,
multilingual encoders, or classical sequence models, rather than exploiting
language-aligned monolingual pretraining. This study asks whether, for a low-resource, morphologically rich
language like Gujarati, a homogeneous ensemble of a single monolingual encoder outperforms such architectural diversity. We propose HomoEnsNER, a homogeneous ensemble of five independently fine-tuned GujaratiBERT models combined via majority voting, evaluated against a single GujaratiBERT baseline and six heterogeneous alternatives, including combinations with MuRIL-base, MuRIL-large, IndicBERT, mBERT, BiLSTM, CRF, and a stacked BiLSTM-CRF-GujaratiBERT architecture. All eight models were trained under a consistent budget and evaluated using entity-level F1 on the Naamapadam Gujarati test split. HomoEnsNER achieved the highest F1 (0.8442), surpassing the baseline (0.8347) and every heterogeneous alternative (lowest: 0.7855), indicating that
language alignment is a more effective, budget-conscious ensembling strategy than architectural complexity for low-resource Indian
language NER.
Read the full abstract ▾
Computation and Language
What Language Does and What the Evidence Supports: A Functional Role Taxonomy and Evidence Audit of Language Grounding in Embodied Agents
Yifan Guo, Chenghao Li, Zhu Wang et al. (+6 more)
Foundation models place
language throughout
embodied agents, but its presence does not show what it contributes or how well that contribution is grounded. This survey separates these two questions. We define five non-exclusive functional roles for
language: Specification,
Embodied Representation, Action Orchestration, Grounding Regulation, and Execution Coupling. For each role, we trace the path from linguistic content to its
embodied consumer and identify the observations or interventions that can test the claimed responsibility. Applying this framework to the reviewed literature reveals a recurring gap between functional use and evidential support. Interpretable or revised linguistic intermediates may be incorrect, go unused, or fail to affect later behavior. Even when actions are directly conditioned on
language, system-level success does not by itself isolate
language's contribution. We therefore evaluate grounding claim by claim, asking whether the reported evidence supports the specific responsibility assigned to
language. Using role claims rather than architectures as the unit of comparison allows us to compare modular and end-to-end
embodied agents without extending conclusions beyond the reported evidence.
Read the full abstract ▾
All authors Yifan Guo, Chenghao Li, Zhu Wang, Wei Xu, Yu Li, Yulong Zhu, Zhuo Sun, Bin Guo, Zhiwen Yu
Computation and Language
VIVID: A Culturally Grounded Benchmark Exposing the Figurative Language Gap in Vietnamese NLP
Tu Tran Do, Nhat Ngoc Nguyen, Khanh-Tung Tran et al. (+3 more)
We present VIVID (Vietnamese Idioms for Validation and Interpretation Depth), the first systematic
benchmark for evaluating culturally grounded figurative
language understanding in Vietnamese. VIVID comprises 1,636 idioms and proverbs annotated with five complexity traits (literal expressions, pragmatic nuances, Sino-Vietnamese terms, uncommon vocabulary, folk knowledge) and seven semantic themes. We establish an evaluation framework combining generative and discriminative tasks, proposing an LLM-as-a-Judge approach with aspect-based prompting validated against human judgment (Cohen's kappa = 0.792). Evaluating eight state-of-the-art models reveals critical gaps: Vietnamese-specialized models drastically underperform
multilingual systems (VinaLLaMA-7B: 0.13 vs. GPT-4o: 2.46), and even top models achieve less than 50% of maximum scores. Notably, few-shot prompting does not universally improve performance, with GPT-4o exhibiting degradation due to stylistic overfitting. Our analysis exposes systematic failures including literal over-interpretation, lexical gaps, and pragmatic flattening, demonstrating that current models lack cultural competence for nuanced figurative interpretation. VIVID provides an essential tool for advancing figurative
language understanding in culturally rich contexts.
Read the full abstract ▾
All authors Tu Tran Do, Nhat Ngoc Nguyen, Khanh-Tung Tran, Hoang D. Nguyen, Tu Minh Phuong, Long Hoang Dang
Computation and Language · Cross-listed from cs.LG
SMOPD: Multi-Reward Reinforcement Learning via Specialize-and-Merge Online Policy Distillation
Wen Wang, Jiahua Bao, Tu Yongsiqi et al. (+8 more)
We aim to improve model performance in multi-reward reinforcement learning training process. Existing Group reward-Decoupled Normalization Policy Optimization (GDPO) has mitigated the issue of reward signals masking one another during direct scalarization by normalizing each reward dimension separately before aggregation. However, our experiments show that GDPO still struggles to balance reward signals with different granularities. Specifically, in some particular training tasks, the model may receive a dense reward that assigns fine-grained scores ranging from 0.1 to 1.0, together with a sparse reward that provides only binary feedback of either 0 or 1. In such cases, we find that the sparse reward may provide an insufficient optimization signal, preventing its corresponding capability from being effectively reinforced. Therefore, how can we strengthen the optimization signal from the sparse reward without sacrificing the capability already learned from the fine-grained reward? To overcome this limitation, we propose Specialize-and-Merge Online Policy
Distillation (SMOPD), a two-stage training method for multi-reward optimization. Stage1-Specialize: SMOPD first employs reward-priority configurations to train multiple reward-specialized teachers, allowing each reward to be learned under conditions where its signal can effectively drive optimization. Stage2-Merge: SMOPD then utilizes online policy
distillation to combine the reward-specialized capabilities of these teachers into a single student policy, while maintaining balanced task-level optimization. To validate our method, we conduct experiments on two multi-reward settings: complementary rewards(tool-calling accuracy and format) and conflicting rewards (helpful and harmless rewards). Based on above settings, SMOPD outperforms GDPO across 1.5B, 3B and 7B backbones.
Read the full abstract ▾
All authors Wen Wang, Jiahua Bao, Tu Yongsiqi, Yihao Liu, Haotian Zhou, Haoxuan Ma, Mengyu Zhou, Wenkui Fan, Junwei He, Xiaoxi Jiang, Guanjun Jiang
Computation and Language
Scalable Frequency- and Length-Aware Subdocument Deduplication for Large Language Model Pretraining
Hai Wang, Chenhao Wang, Qifeng Cai et al. (+6 more)
Large-scale pretraining corpora contain substantial duplicate content. Although document-level deduplication is widely used, removing subdocument-level redundancy remains challenging. At corpus scale, suffix-array-based methods are commonly applied independently within shards, leaving cross-shard duplicates undetected and making the resulting retention behavior sensitive to the sharding configuration. Hash-based methods enable global exact duplicate counting, but often rely on fixed copy-retention policies that cannot accommodate heterogeneous repetition patterns. We propose a scalable subdocument deduplication framework that decouples duplicate detection from copy retention. It identifies duplicate groups through natural-boundary
segmentation, normalized exact hashing, and distributed aggregation, and then applies an explicit frequency- and length-aware retention policy that allocates an adaptive copy budget to each group, retaining more copies of low-frequency or short repetitions while more aggressively deleting high-frequency or long ones. Experiments on FineWeb-Edu and a code-containing web corpus show that models trained on data processed by our method achieve the best overall performance among the evaluated settings. These results underscore the importance of explicit copy-retention control.
Read the full abstract ▾
All authors Hai Wang, Chenhao Wang, Qifeng Cai, Yixiu Liu, Miao Peng, Nuo Chen, Yuanlin Tu, Chengcheng Xu, Feng Zhang
Computation and Language · Cross-listed from cs.CV
GSTEP: Global Spatio-Temporal Density-Driven Visual Token Pruning for Efficient Video Large Language Models
Mengjie Zhang, Qihui Zhu, Tao Zhang et al. (+10 more)
Video large
language models (VideoLLMs) achieve strong
video understanding performance, but their inference remains expensive due to the large number of redundant spatio-temporal visual
tokens in long
videos. Existing
token pruning methods alleviate this cost by reducing redundant
tokens, yet most of them rely on segment-level local
pruning, where
videos are partitioned into isolated segments and
tokens are selected independently within each segment. Such designs may under-preserve short but semantically dense segments and discard
tokens that appear non-salient locally but remain critical from a global perspective. To address this issue, we propose GSTEP (Global Spatio-Temporal Density
Pruning), a plug-and-play
pruning framework that models
video as a continuous spatio-temporal information flow. GSTEP constructs a
token-level spatio-temporal density by combining a continuous temporal density, obtained from a smoothed centered frame-level change signal, with intra-frame spatial density, and then performs global
token sampling by jointly balancing information density and coverage. Extensive experiments on multiple VideoLLMs and public
benchmarks demonstrate that GSTEP consistently achieves strong accuracy-efficiency trade-offs and generalizes well across model architectures and evaluation settings. On LLaVA-OneVision-7B, GSTEP prunes 75% of visual
tokens, preserves up to 100.2% of the original average performance across
benchmarks, and achieves a 1.17 end-to-end speedup.
Read the full abstract ▾
All authors Mengjie Zhang, Qihui Zhu, Tao Zhang, Shuangwu Chen, Huihuang Qin, Yu Guo, Shenghao Ye, Zijian Wen, Yunpeng Hou, Dong Jin, Xiaobin Tan, Huasen He, Jian Yang
Computation and Language
PAMT: Process-Aligned Reinforcement Learning for Multi-Domain Machine Translation
Yongshi Ye, Biao Fu, Chongxuan Huang et al. (+2 more)
Multi-domain machine translation (MDMT) requires more than fluent generation: it demands domain-sensitive translation decisions such as domain disambiguation, terminology control, and stylistic adaptation. Large
reasoning models (LRMs) make such decisions explicit through intermediate translation steps, but our analysis across 15 domains and four translation directions shows that this explicit
reasoning is double-edged: it improves long-form and high-difficulty translation, yet often drifts in terminology-intensive and stylistically constrained settings. We trace this failure to a credit-assignment bottleneck: existing methods optimize final outputs or coarse trajectories, but cannot identify which translation steps actually help the final translation. To address this, we propose PAMT, a process-aligned training framework that combines cold-start domain-aware Long-CoT supervision with reinforcement learning. PAMT uses sequence-level format and outcome rewards for the final translation, together with a step-level process reward that measures how much each explicit translation step increases the likelihood of the reference translation. Across two backbones, PAMT improves over base models, outperforms MT-specialized baselines on average, and remains competitive with strong LLMs/LRMs across in-domain, OOD, and
multilingual settings.
Read the full abstract ▾
All authors Yongshi Ye, Biao Fu, Chongxuan Huang, Yidong Chen, Xiaodong Shi
Computation and Language · Cross-listed from cs.CR
AI Security Leaderboard: Methodology, Results and Minimal Standard
Jasper Timm, Lukas Struppek, Ziwei Xu et al. (+12 more)
The AI Security Leaderboard is an independent
benchmark that ranks the safeguards of frontier AI models from least to most secure. It tests models against the FAR$.$AI Minimal Standard for Safeguards, which represents a minimum bar for security: meeting it does not guarantee a secure model, but failing to meet it guarantees a lack of state-of-the-art security. Version 1.0 covers severe misuse requests across chemical, biological, radiological,
nuclear, and explosive (CBRNE) threats and offensive cybersecurity. In this report, we tested four leading models for universal jailbreaks in the context of this minimal standard, and found more than a hundredfold difference in security. Claude Fable 5 and GPT-5.6 Sol held against every attack we ran, with no universal jailbreak found; we estimate they would likely cost more than \$14,200 to jailbreak, if it is possible with this methodology at all. Meanwhile, we found hundreds of universal jailbreaks for Grok 4.5 and Gemini 3.1 Pro; each broke for under \$300, with universal jailbreaks in Grok's weakest domain, cybersecurity, accessible for as little as \$24. The gap is fixable: every weakness we found belongs to a known class of attack that already has a defense deployed in production models. The leaderboard will be updated on a rolling basis as new models are released, and the evaluation methodology and Minimal Standard will be periodically revised to take into account the latest capabilities and the state-of-the-art in safeguards. The leaderboard is available at leaderboard.far.ai.
Read the full abstract ▾
All authors Jasper Timm, Lukas Struppek, Ziwei Xu, Grace Cheong, Oscar Mata, Dan Zhao, Mick Yang, Isadora De Andrade, Xiaojun Jia, Yiming Li, Samuel Bauer, Heather McIntyre, Adam Gleave, Edward Yee, Kellin Pelrine
Computation and Language
CVPO: Enhancing LLM Reinforcement Learning Reasoning via Value-Variance Adaptation and Dynamic Curriculum Learning
Ziqi Jia, Yalu Ouyang, Bo Pang et al. (+5 more)
Reinforcement learning (RL) has emerged as an effective method for enhancing the
reasoning capabilities of large
language models (LLMs). However, existing methods suffer from insufficient precision in feedback on generated answer trajectories and exhibit the phenomenon of problem difficulty drift. To address these challenges, we propose CVPO - Curriculum-guided Value-Variance Policy Optimization. At the response trajectory level, we find that
token-level value-variance correlates with exploration intensity. Our theoretical analysis shows this variance bounds policy update magnitude. We then use the estimated trajectory value-variance to quantify the intrinsic randomness in generation. Based on this, we design a variance-aware advantage adjustment mechanism for different reward types. At the question level, we introduce a dynamic curriculum weighting method that adapts to question difficulty. This helps the model focus on tasks matched to its current ability during each training stage. Experimental results show our method outperforms strong value-based baselines like VAPO. It achieves better performance and stronger exploration, enabling more accurate and robust
reasoning in
language models across various math tasks.
Read the full abstract ▾
All authors Ziqi Jia, Yalu Ouyang, Bo Pang, Panpan Li, Hangfei Xu, Shengzhao Wen, Shiyong Li, Yanpeng Wang
Computation and Language
Activation-Guided Neuron Intervention to Induce Alzheimer's-Related Computational Language Phenotypes in a Large Language Model
Rui He, Ercong Nie, Hong Jiang et al. (+3 more)
Changes in spontaneous
speech provide an early signal of cognitive dysfunction in Alzheimer's disease (AD) that large
language models (LLMs) can detect. However, detection alone cannot establish whether the underlying model representations contribute functionally to behavior. We introduce an activation-guided intervention framework using Qwen3-8B. The framework identifies feed-forward
neurons with higher activation rates for AD than control transcripts and modulates their output contributions during generation by scaling the corresponding down-projection weights. This yielded nine edited variants differing in intervention direction, magnitude, and scope. The original and edited models completed the same 12-turn neuropsychological battery, assessed through blinded human ratings and computational linguistic measures. Amplifying AD-associated
neurons produced graded impairments in story recall, verbal fluency, working
memory, procedural discourse, scene construction, and coreference resolution. Attenuation largely preserved performance and selectively improved several outcomes. Amplification also reduced lexical surprisal, idea density, syntactic complexity, and discourse quantity, broadly paralleling changes reported in human AD
speech. These findings show that
neurons identified solely from clinical
language differences can influence behavior across multiple cognitive domains, providing proof of concept for an AD-related computational phenotype and a controlled framework for experimentally examining links between
language and broader cognitive dysfunction.
Read the full abstract ▾
All authors Rui He, Ercong Nie, Hong Jiang, Iris E. Sommer, Philipp Homan, Wolfram Hinzen
Computation and Language
SeqLLM: Augmenting LLMs with Behavioral-Sequence Modeling for High-Stakes Decisions at WeChat Pay
Guilin Li, Jiaxing Zhang, Matthias Hwai Yong Tan et al. (+2 more)
Merchant risk control at large payment platforms screens tens of millions of merchants daily, where false positives harm legitimate merchants and false negatives leave harmful activity undetected. The hardest cases require jointly understanding a merchant's textual profile and long behavioral sequence. Large
language models (LLMs) excel at text but cannot natively model such sequences, while adapting them often causes catastrophic forgetting. We present SeqLLM, a framework that adds behavioral-sequence modeling to a pretrained LLM while preserving its
language ability. SeqLLM combines three components: a compact discrete vocabulary that represents behavioral events as native
tokens; a lightweight projector, trained with a two-stage
alignment curriculum, that grounds these
tokens in the LLM's semantic space; and prefix-guided capability injection, which acquires sequence-modeling ability through task-prefixed supervised fine-tuning rather than continual pre-training. SeqLLM is deployed at WeChat Pay, screening millions of merchants daily. Against the production DeepSeek-based LLM baseline, it raises screening precision from 92.0% to 97.5%. Its pretrained behavior-
token embeddings also improve Precision@Top-0.01% by 26.8 percentage points in a production fraud detector serving billion-scale transaction traffic. Beyond payments, SeqLLM achieves state-of-the-art results on public recommendation
benchmarks. On MovieLens and Amazon, it surpasses the strong User-LLM baseline by up to 32% relative Recall@5 while retaining markedly stronger
language ability. On RecIF, it improves Pass@32 by 14.2% over the full OneRec-8B pipeline using only one-fifth of its GPU-days.
Read the full abstract ▾
All authors Guilin Li, Jiaxing Zhang, Matthias Hwai Yong Tan, Bo Wang, Weiran Huang
Computation and Language · Cross-listed from cs.CV
PDD-RRG: Posterior Diagnostic Decision for Study-level Radiology Report Generation
Yang Yu, Yiming Ji, Bin Dai et al. (+4 more)
Automatic radiology report generation (RRG) aims to simulate the workflow of radiologists, assisting them in clinical diagnosis. However, existing methods often fall short in utilizing all information relevant to the examination, as is typically done in clinical practice. Although some works attempt to incorporate multi-view images and historical data, these additional inputs may sometimes lead to avoidable diagnostic errors on the contrary. To address these challenges, we introduce a decision-making stage after report generation for the first time and propose a Posterior Diagnostic Decision framework (PDD-RRG) to integrate potentially conflicting diagnoses. Specifically, we create various subsets of input data and utilize an existing RRG model to generate reports from different perspectives. Then the
Bayesian posterior probability and the learned thresholds for each clinical observation are calculated to obtain an aggregated diagnostic conclusion, which is subsequently used to refine the generated report. Experiments on MIMIC-CXR demonstrate that our proposed PDD-RRG can effectively enhance the clinical efficacy of existing RRG models without any retraining.
Read the full abstract ▾
All authors Yang Yu, Yiming Ji, Bin Dai, Dong Zhang, Zhiyong Zhou, Shoushan Li, Yakang Dai
Computation and Language
PI-Mem: Pushing Long-Context Reasoning to 3.6M Tokens with Parallel-Iterative Memory
Dawei Liu, Haixu Song, Shuang Cheng et al. (+9 more)
Long-context
reasoning remains a critical bottleneck for large
language models, as recent recurrent-
memory approaches face two inherent challenges: sequential chunk-wise updates can overwrite early critical evidence with later irrelevant content, and serial inter-chunk dependencies limit parallelism and cause latency to increase with context length. To address these issues, we propose PI-Mem (Parallel-Iterative
Memory), a mechanism that processes all chunks in parallel and iteratively refines a shared
memory over a bounded number of turns. In each turn, PI-Mem reads all chunks in parallel conditioned on the current
memory, selects new or complementary evidence from each chunk, and merges the selected evidence into a compact shared
memory for the next turn. To discourage redundant turns, we optimize the workflow through reinforcement learning with an auxiliary turn-efficiency reward, enabling the model to adaptively exit once sufficient evidence has been accumulated. We evaluate PI-Mem with Qwen3.5-35B-A3B and Qwen2.5-7B on the HotpotQA
benchmark across context lengths up to 3.6 million
tokens and find that it outperforms the recurrent-
memory baseline by +6.25 and +7.81 absolute points while achieving 6.1$\times$ and 2.1$\times$ inference speedups, respectively. These results demonstrate that PI-Mem breaks the accuracy–efficiency trade-off in long-context
reasoning and provides a scalable approach to complex multi-hop question answering over extremely long documents.
Read the full abstract ▾
All authors Dawei Liu, Haixu Song, Shuang Cheng, Shijie Wang, Haozheng Hou, Kaifeng Liu, Ermo Hua, Zhonghang Yuan, Zhijie Zhong, Yuchen Fan, Biqing Qi, Bowen Zhou
Computation and Language
Emulate or Estimate? The Divergent Strengths of Base and Post-Trained Language Models for Opinion Simulation
Seth Grief-Albert, Jessica Bo, Difan Jiao et al. (+1 more)
Large
language models are increasingly used to simulate human opinions, but prior work reports conflicting results: some studies find promising
alignment with human survey data, while others find persona collapse and weak demographic sensitivity. We propose that much of this conflict stems from conflating two distinct tasks. We call the first task emulation, in which models generate individual responses that aggregate into a population distribution. We call the second task estimation, in which models directly predict the population distribution. Evaluating six matched base and post-trained models on the Pew American Trends Panel, we find that base models are the stronger emulators: they produce response distributions closer to human ground truth and better preserve demographic structure. Post-trained models are generally the stronger
estimators, producing more accurate distributional predictions when asked directly. We argue that model selection for human simulation should be guided by whether the task requires generating text or predicting distributions.
Read the full abstract ▾
All authors Seth Grief-Albert, Jessica Bo, Difan Jiao, Ashton Anderson
Computation and Language
Beyond Accuracy: A Multidimensional Evaluation of Statistical Reasoning in Large Language Models
Monnie McGee, Mateo Langston Smith and Julian Cabrera
Statistical
reasoning is multidimensional, yet evaluations of large
language models (LLMs) typically emphasize response accuracy while overlooking how models construct and communicate statistical explanations. This study demonstrates the value of a multidimensional evaluation by combining response accuracy, response behavior, structural topic modeling, and lexical similarity analysis. The framework is applied to explanations generated by 15 current-generation LLMs responding to 90 questions drawn from four statistics examinations spanning high school, undergraduate, and graduate levels. Accuracy varied substantially across models, ranging from 55% to 78%. In contrast, structural topic modeling revealed a common conceptual organization of statistical
reasoning across all models, while lexical similarity analysis identified modest but consistent vendor-specific differences in explanatory style. Models developed by the same vendor (e.g. Anthropic, OpenAI) produced explanations that were slightly more similar than models from different vendors. These findings demonstrate that statistical
reasoning in contemporary LLMs cannot be characterized by accuracy alone and illustrate how complementary analyses of response behavior and model-generated explanations provide a more comprehensive evaluation of statistical
reasoning in generative AI.
Read the full abstract ▾
Computation and Language
Language Models Encode the Contextual Truth of Propositions
Rupak Sarkar, Pritika Ramu and Rachel Rudinger
Prior work has shown that LLMs encode the truth of factual propositions along linear directions in activation space. It's unclear how these representations extend to contextual truth: propositions whose truth is determined by in-context evidence rather than world knowledge. We show that LLMs maintain a linear representation of contextual truth that persists across structurally different output policies, even when the output doesn't require the model to determine a proposition's truth, and show causal evidence via steering experiments. Using the transcripts from a collaborative
vision-
language task that requires two LLMs to maintain a shared common ground, we show that truth representations of a proposition are significantly swayed by partner assertions about that proposition, even when the LLM has enough evidence to determine its truth. We find evidence that propositions near the decision boundary are more susceptible to having their truth shifted through partner assertions. Separating representation from output distinguish two forms of sycophancy that output behavior alone cannot: the model may accommodate a false proposition while continuing to represent it as false, or shift its representation across the boundary. The latter is 2.59x more common when the model agrees by restating the false claim explicitly than when it agrees implicitly.
Read the full abstract ▾