Computation and Language
Fair Fact-Checking: Closing the Cross-Lingual Gap in LLM Factual Judgement with RoSh
Muhammad Ahmad, Fatemeh Seyedin, Adrian Weller et al. (+2 more)
Misinformation on social media remains a critical problem, and more and more people settle it by asking a
language model instead of a fact checker. Whether models judge such claims reliably is debated; whether they judge them equally well in every
language people ask in has gone almost unasked. We test eight models from five families, 3B to 70B, on 1,500 encyclopedic factual claims that exist in identical form in eight
languages. English is judged better than every other
language on every model, and the gap is widest on the smallest ones, where Llama-3B on Arabic is no better than guessing. Existing remedies retrain on more multilingual data or fit an unconstrained map between
language representations, and neither asks whether the model already holds the answer and simply fails to say it. It largely does: a linear probe recovers the truth from the very activations the model fails to express. We propose RoSh, a per-
language shift and rotation of the residual stream, computed in closed form at three layers, with no training and no weight modified. It improves every model and closes 75% of the gap on average, helping most where the model was worst: Arabic on Llama-3B goes from chance to nearly the English level, and a fifth fewer of the claims answered correctly in English are lost in translation. What remains is no longer a read-out failure: afterwards the head recovers as much of what is encoded outside English as it does in English. An unconstrained map fitted on the same pairs falls below the untouched baseline, so the orthogonality constraint is doing the work, and every model clears a scrambled-correspondence control and ten further controls. On the two
benchmarks of the closest inference-time method, latent-space intervention, run with its own data and metric code, RoSh's gains are five to thirteen times larger.
Read the full abstract ▾
All authors Muhammad Ahmad, Fatemeh Seyedin, Adrian Weller, Dongwon Lee, Mahmoudreza Babaei
Computation and Language
Rewarding Novel Deductions: Solver-guided Process Supervision for Logical Reasoning
Muhammad Asif Ali, Wenqing Wang, Huan Wang et al. (+1 more)
Logical
reasoning remains a major challenge for large
language models (LLMs), particularly on structured problems that require precise constraint tracking, consistency preservation, and multi-step deduction. This challenge is especially acute for small-scale LLMs, which are more prone to producing inconsistent, redundant, or brittle
reasoning trajectories. Existing approaches for improving logical
reasoning largely optimize for final-answer correctness, providing only weak supervision over the intermediate
reasoning process. In this work, we propose SPRING: (Solver-guided Process Rewards for Novel LogIcal
ReasoNing Step Generation). SPRING uses SMT solver as a training-time verifier of intermediate
reasoning steps to provide process-level supervision. It introduces the notion of a novel
reasoning step, namely, a step that is logically valid, consistent with the evolving
reasoning state, and not already implied by previously accepted non-contradictory deductions. Based on this solver-based assessment, it designs process rewards that encourage novel inferential progress while penalizing contradictory and uninformative
reasoning steps. Evaluation across three logical
reasoning benchmarks, ZebraLogic, AR-LSAT, and Knights and Knaves, and four LLMs shows that SPRING consistently outperforms base LLMs, outcome-only reward baselines, and Logic-LM. On ZebraLogic, SPRING improves puzzle accuracy by up to 49.71 and 15.43 points over the base LLM and strongest outcome-only baseline, respectively. On AR-LSAT, it improves overall accuracy by up to 64.93 and 12.14 points, respectively. On Knights and Knaves, SPRING achieves up to 93.14 puzzle accuracy and 96.05 person accuracy.
Read the full abstract ▾
All authors Muhammad Asif Ali, Wenqing Wang, Huan Wang, Mohammad Raza
Computation and Language · Cross-listed from cs.LG
When Can Attention Heads Be Statically Defined?
Weixian Waylon Li, Yintao Tai, Marcio Fonseca et al. (+1 more)
Some
attention heads learn similar patterns across inputs. Reusing these patterns could reduce training cost by avoiding repeated query-key score computation and softmax. Through controlled pretraining comparisons, we identify Selective
Attention Freezing (SAF), which selects heads with low
attention-pattern variance and replaces their
attention weights with fitted post-softmax means halfway through training. We represent these fixed patterns with absolute-position and relative-distance preferences, reducing storage from quadratic to linear in sequence length. A fused kernel reconstructs the patterns and executes ordinary-
attention and replaced heads together. At matched training-
token budgets, replacing 25% of
attention heads gives 1.056x faster post-replacement optimiser updates at 124M parameters and 4K context, with a 0.77% perplexity increase. At 1B and 8K context, post-replacement updates are 1.068x faster on four GPUs including communication, with a 0.51% perplexity increase. The resulting models also accelerate long-input finetuning and causal prefill. After associative-recall adaptation, the 124M model with 25% replacement generalises to more key-value pairs at a fixed length better than ordinary
attention and two
pruning controls.
Read the full abstract ▾
All authors Weixian Waylon Li, Yintao Tai, Marcio Fonseca, Shay B. Cohen
Computation and Language · Cross-listed from cs.AI
After the Fix: Transfer of Corrected Agent Experience
Yanfei Zhang and Xu Lin
Does repairing an episode make its experience a better
memory for the next task? We transfer the same failed source before and after accepted repair to a fixed target, alongside independent execution. Our 3,300 runs cover 100 ThinkingBox pairs and the same 100 APEX pairs with and without source-state inheritance, under eleven conditions. ThinkingBox's Full/Skill/Hybrid correction gains are 44/29/32 percentage points, with corrected performance 25/22/18 points above independence; inference weakens at the task-family level. Yet 12 of Full's 15-point larger correction gap over Skill come from worse uncorrected performance, not better corrected
memory. Moreover, 22 of Full's 46 upward transitions restore observed baseline success. Neither APEX regime establishes comparable aggregate correction benefits. Action evidence connects workflow gains with reusable obligations and convention conflicts with source-local choices. Text APEX's accepted execution reaches 52% versus its summary's 40%, without robust global/group-level superiority or an estab- lished advantage over independence. Smaller handoffs reduce input but increase calls. The value of repairing experience is therefore distinct from the value of reusing it:
memory updates require both a previous-version reference and a fresh-start reference.
Read the full abstract ▾
Computation and Language
The Model Knows When to Stop: Training-Free Early Stopping for Long-Context Reading
Muath Alyobi, Mohamed Eltahir, Almoayyad Abuljdail et al. (+3 more)
Language models often process long inputs sequentially in chunks, but continuing to read after sufficient evidence has been acquired wastes computation. Existing stopping mechanisms either learn sufficiency from internal activations or train an exit gate, while a simpler alternative asks the model whether it has read enough. We introduce Answer-Convergence Stopping (ACS), a training-free stopping rule that measures rather than asks. After each chunk, it probes the frozen model's current answer state and stops when that state is both confident and stable. The rule requires only output-side generation and
token log probabilities, has no trained components, and uses one shared configuration across models and
benchmarks. Because a stopping policy can save computation simply by stopping too early, we evaluate the stopping decision itself using evidence position where available. On the full LongBench-v2 with two frontier models, ACS is the only stopping policy that matches or exceeds full-reading accuracy. Furthermore, across 250 S-NIAH questions, the premature stopping rate for ACS across five models from two families ranges from 0% to 12%, compared to 8.4% to 45.6% for the verbalized gate. Taken together, ACS reveals that by properly utilizing the output signals of frozen models, we can achieve favorable behaviors like adaptive stopping without the need for additional training.
Read the full abstract ▾
All authors Muath Alyobi, Mohamed Eltahir, Almoayyad Abuljdail, Riyadh Almutawa, Tanveer Hussain, Naeemullah Khan
Computation and Language
In-game Toxic Detection: Bi-directional Representations with Attention Residuals
Yuanzhe Jia
In-game toxic
language has emerged as a critical concern in the gaming industry and community. While several frameworks and models for online game toxicity analysis have been proposed, detecting toxicity in player chat utterances remains a formidable challenge: stemming not only from the extremely short length of such utterances but also from the heavy reliance on game slang, abbreviations, and domain-specific jargon, which generic
language models are poorly suited to recognize. This paper presents a shared task for in-game toxic
language detection built upon real-world in-game chat data, and proposes the best-preforming model for the toxic
language slot filling: Bi-directional Representations with
Attention Residuals (BRAR). Experimental results demonstrate that BRAR effectively captures the global context and outperforms the existing baselines on slot filling.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.AI
Nudgeability: Reasoning Models Follow Confidence Signals Without Tracking Their Own Competence
Rohit Saxena and Utkarsh Upadhyay
Reasoning language models that can call tools must decide during inference whether to answer unaided or delegate. Any self-reflection mechanism for this must answer three questions: where the reflective signal comes from (verbal reports, output distributions, hidden states, a separate predictor), how it is presented to the model (numerical prediction, confidence
token,
prompt injection), and whether it changes the model's subsequent action. We isolate the third question. At a fixed point in otherwise identical
reasoning trajectories, we insert a single first-person sentence expressing either confidence or doubt; the model then continues
reasoning and chooses whether to answer directly or call a tool. Comparing these counterfactual continuations measures the causal effect of the reflective signal on delegation. We call this behavioral response Nudgeability and measure it along two dimensions: sensitivity, how strongly confidence and doubt change delegation rates, and targeting, whether delegation increases for problems the model cannot solve unaided and decreases for those it can. Across nine small-to-medium open-weight
reasoning models from three families (Qwen, Gemma, and GLM) and two tasks, models are consistently sensitive: doubt increases delegation and confidence decreases it, with a median confidence-to-doubt swing of 20.6 percentage points, and 53 to 70 points for the larger provider-served models. This responsiveness is poorly targeted: a median 42% of induced flips are well-targeted, only a +2 percentage-point lift over a random-selection baseline. Confidence
language is thus a strong control surface for delegation, but current models use it only weakly in accordance with their actual competence. Nudgeability offers a simple, post-training-free way to evaluate both sensitivity and targeting as endogenous self-reflection mechanisms mature.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.CV
Rethinking Latent Visual Reasoning: Grounding Latent Reasoning in Visual Evidence
Xi Xiao, Tianchen Zhao, Youngeun Kim et al. (+10 more)
Latent visual
reasoning (LVR) enables
multimodal large
language models (MLLMs) to perform intermediate computation in continuous latent
tokens rather than expressing every
reasoning step in words. However, unlike textual CoT, latent
reasoning is not directly observable, making it difficult to supervise what latent
tokens learn. In this work, we first conduct a thorough analysis of latent-
token behavior and identify a latent evidence-credit gap: latent
tokens respond only weakly to image perturbations that alter the correct answer. We hypothesize that this issue stems from the lack of explicit supervision during GRPO training. These findings suggest that a final-answer reward provides too little guidance on what visual evidence to preserve or how credit should be assigned across latent
tokens. To bridge this gap, we propose ReaLVR, which brings visual-evidence supervision to the model's own free-running latent trajectories. ReaLVR contrasts correct and model-generated wrong answers to determine where stronger supervision is needed, and relevant and mismatched visual evidence to specify what to preserve. Across three model families, ReaLVR consistently outperforms evaluated LVR baselines, achieving the highest five-task average of 63.7% on Qwen2.5-VL-7B. Crucially, we are the first to scale visual
reasoning in latent space, showing that our framework continues to deliver robust improvements at frontier model scales up to 235B. Further analyses show more question-sensitive latent-
token positions, stronger
alignment with relevant visual regions, and greater fixed-context dependence on the most attended latent
tokens.
Read the full abstract ▾
All authors Xi Xiao, Tianchen Zhao, Youngeun Kim, Zhuowei Li, Linghan Xu, Jiaye Wu, Zheng Zhang, Xiang Xu, Xuanbai Chen, Farhan Tejani, Jakub Zablocki, Julia Xu, Yifan Xing
Computation and Language
RoPE is Dead, Long Live RoPE: Towards Scalable Data-aware Positional Encodings
Jarod Lévy, Mathurin Videau, Jad Yehya et al. (+3 more)
Transformers process
tokens without any inherent notion of order, making positional encoding a fundamental requirement rather than an architectural refinement. Rotary Position
Embedding (RoPE) has become the default positional encoding in modern
language models, yet it is heavily biased toward nearby
tokens. Existing alternatives have been evaluated under different settings, leaving the literature fragmented and without a clear replacement. We bring structure to this landscape by examining a specific weakness of RoPE: its slow frequency bands, whose wavelengths exceed the training context and expose models to unseen angles during extrapolation. We therefore introduce Data aware RoPE (DaRoPE), which preserves standard RoPE on the fast bands but replaces absolute position on the slow bands with bounded coordinates learned from contextual representations. Therefore, the slow-band geometry depends on the data rather than only on positional distance. We compare representative encodings under matched conditions across synthetic tasks, symbolic music, genomics, neural signals, and
language models spanning 124M to 50B parameters. Across these experiments, DaRoPE leads on non-text
benchmarks, mitigates recency bias, while remaining best or on par in
language modeling and length extrapolation. Moreover, the learned coordinates also make the mechanism interpretable, revealing how
attention layers leverage contextual information beyond
token distance. Together, these results support DaRoPE as the best overall default among the evaluated methods, when there is no domain-specific reasons to prefer another.
Read the full abstract ▾
All authors Jarod Lévy, Mathurin Videau, Jad Yehya, Jean-Rémi King, Stéphane d'Ascoli, Thomas Moreau
Computation and Language · Cross-listed from cs.CV
ActionLens: Diagnosing Spatial-Temporal Binding Failures in Vision-Language Models
Gueter Josmy Faure, Min-Hung Chen, Hao Ping Wang et al. (+3 more)
Video-capable
vision-
language models score above 80% on popular
benchmarks yet struggle with spatial-temporal binding: associating the right action with the right person at the right moment. We introduce ActionLens, a diagnostic
benchmark of 6,701 multiple-choice
video questions spanning five targeted diagnostics: transition detection, actor-specific identification, concurrent action binding, directed interaction
reasoning, and gaze detection. Ground-truth answers are derived deterministically from 1.58 million per-second, per-person annotations. Fourteen rounds of human quality engineering raised answer clarity from 53% to above 90% human accuracy. Across 20 VLMs, the full-set leader scores 68.8%; on the human-reviewed subset, it scores 65.9% versus 91.0% for the pooled human reference. Gaze detection remains near chance against 89.6% human accuracy. On actor disambiguation, reference-interface controls show that relational descriptions recover 5.55–13.25 points over static coordinates, confirming a substantial numeric-parsing penalty; yet visual boxes still lead every model by 1.15–6.50 points, exposing a residual unboxed actor-resolution gap. A binding-trap analysis shows models systematically select the wrong actor's action. ActionLens provides diagnostic measurements of these distinct failure modes across model families and scales for direct comparison. We release all data, code, and evaluation scripts at https://anonymous.4open.science/r/lmms-eval-2276
Read the full abstract ▾
All authors Gueter Josmy Faure, Min-Hung Chen, Hao Ping Wang, Timothée Lardy, Hung-Ting Su, Winston H. Hsu
Computation and Language
Papers Without Code: Availability of GitHub Repositories Linked in *CL Publications
Selina Meyer and Michael Roth
Source code and data published at computational linguistics (*CL) venues are increasingly being shared via GitHub. While this generally is a favourable development for the accessibility and potential reusability of research artifacts in natural
language processing (NLP), the long-term availability of such repositories has not been evaluated. In this squib, we discuss the availability of repositories linked in papers published in the Computational Linguistics (CL) journal as well as at ACL and its co-located events over the past ten years. Contrary to our expectations, we find that GitHub repositories linked in more recent ACL publications are unavailable at similar rates as in older publications, in parts due to an increase in empty and placeholder repositories. Similar trends hold for other *CL venues, but not for platforms other than GitHub.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.CR
How to Tame a Multi-Headed Hydra? Adaptive Multi-Category Safety Steering for Large Language Models
Chenxi Wang, Ruiyang Huang, Li Huang et al. (+1 more)
As large
language models (LLMs) become increasingly widespread, preventing unsafe responses to harmful
prompts is essential for their safe deployment. Activation steering offers an approach to improving LLM safety by modifying internal activations during inference without updating model parameters. However, a single
prompt can involve multiple harm categories, and steering toward safety in one category may leave harmful content from another unaddressed. Despite advances in adaptive steering, existing methods do not explicitly coordinate steering direction and strength when multiple harm categories co-occur within a single
prompt. To address this problem, we propose CAM-Steer, a Category-Adaptive Multi-category Safety Steering framework. Specifically, it estimates the risk associated with each harm category by comparing the current hidden state with safe and unsafe prototypes. The estimated risks are then used to combine the safety directions for different harm categories into a single steering direction and to determine the strength of the intervention. Finally, it rotates the hidden state along the composed steering direction, with the rotation angle determined by the estimated risks, while preserving the hidden-state norm. Experiments across three LLM backbones and seven harm categories show that CAM-Steer outperforms the evaluated baselines in average defense success rate, including when categories co-occur. Further analyses support its component designs and informative risk scores, with negligible inference overhead.
Read the full abstract ▾
All authors Chenxi Wang, Ruiyang Huang, Li Huang, Yifan Wu
Computation and Language · Cross-listed from cs.LG
Low-Confidence Remasking Traps Flexibility: Realizing Arbitrary-Order Potential for Diverse Rollouts in Diffusion LLMs
Moongyu Jeon, Dongjae Jeon, Bumjun Kim et al. (+2 more)
Masked
diffusion language models support arbitrary-order generation, suggesting a natural way to produce diverse outputs. However, recent work argues that this flexibility reduces diversity by delaying high-uncertainty
tokens that can lead to different generation paths. We trace this diversity loss not to arbitrary-order generation itself, but largely to low-confidence remasking (LCR), a widely used decoding rule. At each step, LCR samples a
token at every masked position but commits only the sampled
token with the highest probability, filtering out the rest. We show that this mechanism can exponentially suppress lower-probability
tokens as more positions compete, and observe the same suppression in LLaDA. In contrast, top-probability position selection (TPP), which has often been conflated with LCR under the shared label confidence-based decoding, avoids this diversity loss. TPP first selects the position whose most likely
token has the highest probability, then samples directly from that position's distribution. Replacing LCR with TPP restores diversity and yields Pass@$k$ comparable to left-to-right decoding, suggesting that the reported diversity loss stems largely from LCR's filtering rather than from generating high-confidence positions first. To further exploit order flexibility, we introduce
Entropy-Guided Initialization (EGI), which samples the first
token at the highest-
entropy position and then follows TPP. This simple modification further improves rollout diversity and solution coverage beyond left-to-right decoding, with gains extending to downstream policy optimization, highlighting the potential of arbitrary-order generation for diverse rollouts.
Read the full abstract ▾
All authors Moongyu Jeon, Dongjae Jeon, Bumjun Kim, Mingyu Kim, Albert No
Computation and Language
CARDAMOM: A Micro-Dialectal Arabic Speech Dataset for ASR
Bashar Talafha, Samar M. Magdy, Aisha Alansari et al. (+36 more)
We present Cardamom, a micro-dialectal Arabic
speech dataset designed to support fine-grained evaluation and adaptation of automatic
speech recognition (ASR) systems. Community-curated by native speakers familiar with the represented varieties, Cardamom contains approximately 40 hours of transcribed YouTube
speech spanning 21 micro-dialects across Egypt, Jordan, Lebanon, Mauritania, Palestine, and Saudi Arabia. Each segment is annotated with one or more operational micro-dialect labels, code-switching information, and utterance-level perceived gender, enabling analysis of sub-country variation that is obscured by conventional country-level labels. We describe the collection and annotation process, motivate the micro-dialect inventory linguistically, and
benchmark four multilingual ASR systems in zero-shot and adapted settings. The strongest zero-shot system obtains 43.47% aggregate WER, with particularly high error rates on Mauritanian and Lebanese varieties; adaptation on Cardamom reduces its WER to 35.21%. Audio-based identification experiments further show that the annotations provide a learnable prediction target, with a dedicated classifier reaching 85.57% accuracy on 21-way micro-dialect identification. Cardamom provides a resource for studying localized dialectal variation and developing Arabic
speech systems with broader regional coverage.
Read the full abstract ▾
All authors Bashar Talafha, Samar M. Magdy, Aisha Alansari, Alaa Alkhawaldeh, Abdurrahman Juma, Sharaf Makahleh, Nour Gamal, Omar Attia, Hanaa Kurdi, Najwa Rizk, Maysa Anaya, Hessah Altimyat, Layal Alhazmi, Shumukh Alotaibi, Hajar Alhadaris, Rayan Alomari, Rahaf Almalaq, Malak Alkhorasani, Sara alghamdi, Rahaf Alshamrani, Nsrin Ashraf, Ibrahim Jaradat, Nada Qardahji, Yasmin Zaraket, Elmoukhtar Brahim, Sidi Ebeidy, Oumoulmouminin Mahmoud, Yahjeb Bouha Khatraty, Meya Haroune, Mohammad Ghaddar, Mohamad Eldirany, Rashed Alamoush, Tala Chhaytle, Nuha Albadi, Yahya El Hadj, Hamzah Luqman, Fadi A. Zaraket, Mustafa Jarrar, Muhammad Abdul-Mageed
Computation and Language · Cross-listed from cs.SE
The Last Mile Is the File: OfficeEditBench for Preservation-Aware Office Editing
Zhiwen Wu and Chengxu Wu
A small Office edit creates two obligations: propagate every required update and leave protected state untouched. Updating too little leaves dependencies inconsistent; updating too much changes content the user did not authorize. We introduce OfficeEditBench, a 170-task
benchmark for change-scoped maintenance of spreadsheets, presentations, and documents. Task contracts specify required updates, protected state, native structures, and applicable interaction requirements. Across 510 archived task-system outcomes from WorkBuddy, Doubao, and Codex, we distinguish file delivery, target completion, and verifier-defined acceptance. Hard package-valid delivery ranges from 92% to 100%, yet no selected output satisfies the complete contract. Case analysis highlights why local correctness is insufficient: an updated value can lose its generating formula, a revised rule can fail to reach related conclusions, and a new deadline can omit a retained prerequisite. These mechanisms connect artifact-level checks to the continued maintainability of Office files. We analyze maintenance failures while distinguishing frozen automatic verdicts from human acceptability. OfficeEditBench provides a testbed for completing required changes while preserving the logic and scope of existing work.
Read the full abstract ▾
Computation and Language
RGDT-Bench: Benchmarking LLM Reasoning for Rule-Governed Decisions and Their Justifications
Jianpeng Zhao, Haihua Xu, Haoyang Zhang et al. (+7 more)
We study
reasoning in Rule-Governed Decision Tasks (RGDTs), where models apply external rules to case facts and justify decisions, as required in policy, contract, and compliance settings. Beyond the deductive capability emphasized by standard mathematical and logical
reasoning tasks, RGDTs require interpreting rules and their applicability, assessing conditions from evidence, combining judgments under rules and exceptions, and providing checkable justifications. These demands motivate a
benchmark assessing both decisions and their stated grounds. We introduce RGDT-Bench, providing 202.1K condition-level supervision slots across four task tracks and eight supported task-probe combinations that vary access to supporting information. Label-blind extraction and deterministic checks produce labels for warrant completeness: source-referenced coverage and consistency of stated decision grounds. The
benchmark attributes failures to four process layers: rule use, condition, evidence, and aggregation, and checks the final outcome. Among evaluable correct responses, warrant incompleteness averages 40.2% across six evaluated LLMs and supported task-probe combinations. Such warrant incompleteness poses potential safety risks and remains difficult to detect: the best of seventeen existing evaluators reaches only 57.69% (random: 50%) task-averaged area under the receiver operating characteristic curve (AUROC). To address this difficulty, we train a simple reward model with warrant supervision. It achieves 69.24% task-averaged AUROC among correct answers, exceeding the matched outcome-supervised baseline by 10.37 pp (percentage points) and the best existing evaluator by 11.55 pp. Beyond completeness assessment, the model outperforms both outcome-supervised baselines across nearly all response-selection comparisons, supporting RGDT-Bench's warrant supervision for RGDT
reasoning.
Read the full abstract ▾
All authors Jianpeng Zhao, Haihua Xu, Haoyang Zhang, Shuang Qian, Yixiang Tang, Xintao Wang, Kun Sun, Pei Wu, Shuhan Zhong, Pengyang Wang
Computation and Language
When Words Fall Short: Iterative Synergy Between Verbalized Reasoning and Hidden Features for LLM Confidence Estimation
Yekun Xu, Ante Wang, Jingyi Ren et al. (+3 more)
Confidence estimation is crucial for developing trustworthy large
language models (LLMs), with most methods following
estimator-based or verbalization-based paradigms. While recent research increasingly focuses on improving verbalized self-reports of confidence, we challenge the prevailing view that this approach surpasses independent confidence
estimators. Our empirical study shows that a dedicated confidence
estimator can substantially outperform verbalized confidence, indicating that LLMs' internal representations contain richer confidence signals. Building on this finding, we propose Iterative Policy-
Estimator Training (IPoET), a framework that synergizes the complementary strengths of verbalized
reasoning traces and informative representations. IPoET alternates policy optimization with
estimator updating, integrating
estimator-derived confidence feedback into policy learning and refreshing the
estimator on new policy rollouts. Experiments across diverse
datasets and Qwen and Llama backbones demonstrate that, by iteratively exploiting richer hidden features and adapting to the evolving policy distribution, IPoET consistently outperforms both
estimator- and verbalization-based baselines in-domain and achieves superior or comparable results across all out-of-domain metrics. For more details, refer to https://github.com/xyk829/ipoet.
Read the full abstract ▾
All authors Yekun Xu, Ante Wang, Jingyi Ren, Xuanyi Chen, Weizhi Ma, Yang Liu
Computation and Language
Unbiased Top-$k$ Estimation for On-Policy Distillation
Linjian Meng, Siyuan Gan, YuHan Li et al. (+5 more)
On-policy
distillation (OPD) is becoming an important component of large
language model (LLM) post-training for transferring the
reasoning capability of a strong teacher LLM to a weaker student LLM. OPD trains the student by minimizing the reverse KL divergence between the teacher and the student via rollouts generated by the student's policy. However, estimating the gradient of the reverse KL divergence in OPD remains a challenge. Using only the sampled
token from the student-generated rollout is computationally cheap but provides limited distributional supervision, which will degrade accuracy. In addition, using the full vocabulary provides complete distributional supervision but is computationally expensive. Therefore, recent works propose Top-$k$ OPD (TK-OPD) that use selected top-$k$
tokens, which provides richer distributional supervision than sampled-
token estimation at substantially lower computational cost than full-vocabulary estimation. Unfortunately, using only the selected top-$k$
tokens induces bias, leading to accuracy degradation, as the probability mass outside the selected top-$k$
tokens is discarded. To address the bias of TK-OPD, we propose Tail-Corrected Top-$k$ On-Policy
Distillation (TT-OPD). It preserves the advantages of TK-OPD, including rich distributional supervision and low computational cost, while providing an unbiased
estimator of the gradient of the reverse KL divergence. The key insight of TT-OPD is to use not only the selected top-$k$
tokens, but also the sampled
token from the student-generated rollout, thereby recovering the discarded probability mass in expectation, avoiding the bias. Experimental results demonstrate that TT-OPD significantly outperforms other tested OPD variants.
Read the full abstract ▾
All authors Linjian Meng, Siyuan Gan, YuHan Li, Xiran Wang, Ziyang Ding, Ditang Gou, Yiming Wu, Zhen Zhao
Computation and Language
Wanqi Zhou, Jiawei Lu, Yang Wang et al. (+4 more)
Long-term
memory is essential for
language agents to maintain coherent and effective behavior over extended, multi-session interactions. Existing
memory systems mainly use
retrieval at read time, while write-time
memory formation still relies on direct extraction or compression. However, when future information needs are unknown, compressing an entire interaction in one pass can overlook locally important details that may matter later. To this end, we introduce RIME, a
retrieval-induced
memory framework that shifts
memory construction from monolithic compression toward evidence-centered integration. RIME uses generic self-questions to retrieve focused dialogue evidence and grounds
memory formation in both the retrieved evidence and relevant historical
memories, which are jointly reconciled into an evolving
memory bank with temporal and provenance information. At inference time, compressed
memory serves as the primary rather than the sole source of evidence: when it cannot support an answer, RIME retrieves relevant source dialogue together with its local context to recover information omitted during
memory formation, without resorting to full-history processing. Extensive experiments on LoCoMo with Qwen3-235B-A22B and GPT-5.6 Sol show that RIME consistently achieves the best performance across all three quality metrics among the compared methods, while requiring substantially fewer query-time LLM
tokens.
Read the full abstract ▾
All authors Wanqi Zhou, Jiawei Lu, Yang Wang, Zhaolong Xing, Zhen Chen, Ai Han, Haoyue Shi
Computation and Language
AgentHop: A Diagnostic Benchmark for Agentic Multi-Hop Scientific Question Answering
Chanhee Park, Jeongho Yoon, Sungbin Han et al. (+2 more)
Agentic tasks require a large
language model to interact with the world, navigating information and gathering evidence across multiple steps with restricted resources. Due to this complexity, agentic task failures arise from various sources, and pinpointing these failure causes is essential to diagnose and improve agentic systems. Existing
benchmarks, however, tend to focus on a single leaderboard score, leaving the underlying failure modes opaque. To fill this gap, we introduce AgentHop, a diagnostic
benchmark of 1,011 multiple-choice questions paired with a controlled seven-tool sandbox under fixed
token, turn, and tool-call constraints. AgentHop reveals model vulnerabilities by dissecting a single accuracy score along four axes of
agent operation:
retrieval, synthesis, tool-call, and resource management. Across 19 models, we find that behavior clusters by model family, with tool-call signatures revealing distinct family fingerprints: GPT models commit early, Anthropic and GLM checkpoints verify before committing, DeepSeek and Kimi over-search, and Gemini-3 Pro stays balanced. Decomposed axes further expose within-family structure: Claude Opus 4.6 and Sonnet 4.6 land within one accuracy point yet diverge on
retrieval-versus-synthesis emphasis, with Opus retrieving more and Sonnet synthesizing better. We release the full
benchmark set and the harness to support diagnostic
agent benchmarking.
Read the full abstract ▾
All authors Chanhee Park, Jeongho Yoon, Sungbin Han, Hyeonseok Moon, Heuiseok Lim
Computation and Language · Cross-listed from cs.LG
LLMs as Adaptive Meta-Solvers: Strategy-Diverse RL for Industrial-Scale Optimization
Shihao Zhang, Weiting Liu, Siyu Shao et al. (+4 more)
Scaling LLM-based optimization from textbook-scale instances to real-world, industrial tasks remains a critical open challenge. Existing approaches are predominantly evaluated on small, self-contained textual problems and often commit to a solver-integrated paradigm, limiting their ability to handle the scale and structural diversity of practical optimization workloads. In this work, we propose a practical framework for training open-source LLMs to tackle real-world, industrial-scale optimization. We first show empirically that solver-integrated
reasoning, exact combinatorial
algorithm, and heuristic search exhibit complementary strengths across different problem structures and scales. Motivated by this, we introduce Strategy-Diverse Reinforcement Learning (SDRL), which trains LLMs as adaptive optimization meta-solvers. SDRL leverages this complementarity through a correctness-gated hierarchical diversity reward that promotes robust exploration across varying strategies and within each strategy, effectively preventing premature strategy collapse. We further introduce a mixed-format training scheme that jointly supports both self-contained textual problems and file-grounded instances. Across comprehensive evaluations, our framework outperforms existing fine-tuned methods and frontier models including DeepSeek-V4-Pro and GPT-5.5, both on average across
benchmarks and on industrial-scale optimization tasks.
Read the full abstract ▾
All authors Shihao Zhang, Weiting Liu, Siyu Shao, Yitian Chen, Jianfeng Feng, Dongdong Ge, Yinyu Ye
Computation and Language
Zero-Shot Cue-Grounded Topic Segmentation of Spoken Documents
Suhwan Choi, Myeongho Jeon and Myungjoo Kang
Topic segmentation structures spoken documents into coherent sections, facilitating navigation and downstream understanding. The appropriate granularity can vary substantially, ranging from broad thematic shifts to fine-grained subtopics. Existing LLM-based segmenters, however, often struggle to adapt to this variation, causing them to either merge distinct subtopics or over-segment coherent themes. To address this, we introduce Cue-Grounded Segmentation (CGS), a training-free framework that operates without any task-specific supervision. CGS first identifies phrases that explicitly signal the start of a new topic and uses their sentence positions as segment boundaries. When such cues are insufficient, it falls back to semantic segmentation, guided by the document structure inferred during cue extraction. Across six
benchmarks and six LLM backbones, CGS consistently outperforms existing baselines, remains robust to noisy ASR transcripts, and achieves these gains with low API cost on proprietary models.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.LG
Coding Agent Memory Post-training: Unlocking the Memory Potential of Pre-trained File Operations for Long-Horizon Tasks via Reinforcement Learning
Lirui Luo, Kelong Mao, Heming Xia et al. (+10 more)
Language-model
agents increasingly tackle long-horizon tasks whose interaction histories exceed the model's active context. Recent work has begun to use reinforcement learning to make
memory control part of the policy, often relying on predefined
memory tools within domain-specific training environments of relatively short horizons. This setup ties learned
memory behavior to environment-specific interfaces that lie outside the base model's pre-training and must be learned from scratch, so even after post-training,
agents struggle to use
memory in long-horizon tasks. To address these limitations, we introduce Coding
Agent Memory Gym (CAMG), a suite of long-horizon agentic-RL environments spanning Shop, Coding, DeepResearch, and AutoResearch. Alongside each environment's native task interface, CAMG provides executable shell access and an episode-persistent workspace, enabling
agents to create, revise, search, and reuse files as
memory throughout an episode. We also introduce CAMG-RL, which trains a single policy jointly across all four environments with fully asynchronous PPO, learning this file-based
memory behavior directly from downstream task reward, and we train CAMG-RL-4B and CAMG-RL-9B from Qwen3.5 models of matching size. On SWE-bench Verified and MLE-bench Lite, CAMG-RL-4B and CAMG-RL-9B are competitive with Qwen3.5-35B-A3B and Qwen3.5-122B-A10B, respectively.
Read the full abstract ▾
All authors Lirui Luo, Kelong Mao, Heming Xia, Rongqing Li, Xinwei Yang, Luyu Chen, Kieran Wong, Yudong Guo, Xinrui Wang, Jiayin Zhu, Simiu Gu, Sulong Xu, Cong Fang
Computation and Language
Reciprocal Guidance: Orchestrating Draft and Verify Budgets for Advancing the Diffusion-AR Self-Speculation Frontier
Linye Wei, Shutian Zheng, Haoyu Zeng et al. (+1 more)
Diffusion drafting with autoregressive (AR) verification has emerged as a promising paradigm for efficient speculative decoding. Recent self-speculation models, represented by Nemotron-Labs-
Diffusion, further simplify the speculative pipeline by unifying drafting and verification within a shared backbone, while enabling longer acceptance lengths. However, the Pareto frontier between aggregate and per-request throughput remains underexplored. At low concurrency, sequential draft-verify execution requires two model forward passes per round, limiting the effective
tokens per forward (TPF). By contrast, at high concurrency, longer drafts incur increasingly expensive computation, forcing individual requests to operate under constrained speculation budgets and preventing full exploitation of the full-backbone drafter. Our key observation indicates that drafting and verification exhibit reciprocal predictability. Draft logits can anticipate likely verification mismatches, while recent verification outcomes predict future drafting utility and suitable block sizes. Building on this observation, we introduce Reciprocal Guidance (RecGuide), a runtime draft-verify orchestration framework that adapts speculative decoding to varying serving loads. RecGuide exploits spare compute capacity through verification-overlapped drafting at low concurrency, while dynamically allocating request-specific draft block sizes as the workload becomes increasingly compute-intensive. Experiments across a wide range of concurrency levels demonstrate consistent throughput improvements over vanilla self-speculation, achieving up to $1.8\times$ speedup.
Read the full abstract ▾
All authors Linye Wei, Shutian Zheng, Haoyu Zeng, Meng Li
Computation and Language
Look Before You Select: Rethinking Vocabulary Sparsification in On-Policy Distillation
Yongliang Miao, Shuang Liu, Yanguang Liu et al. (+2 more)
On-policy
distillation (OPD) uses teacher correction on student-generated responses. Full-vocabulary correction can provide important corrections even for
tokens that the student assigns low probability, but backpropagating through all
token logits becomes
memory-intensive for long sequences. Existing
memory-saving approaches estimate corrections from sampled
tokens or restrict supervision to the student's TopK
tokens, introducing sampling noise or changing the full-vocabulary correction. We introduce SparseOPD, which uses full-vocabulary teacher correction to determine which corrections matter before selecting the
token logits to differentiate. SparseOPD first constructs the full-vocabulary correction without retaining its backward graph, then selects
tokens by correction magnitude rather than student probability. Signed residual compensation preserves the total promoting and suppressing correction mass, while correction-aware budget allocation distributes the sparse support across positions. Finally, the update backpropagates only through the selected
token logits. Across six task–scale settings spanning mathematics, chemistry QA, and
multimodal reasoning, SparseOPD outperforms Sampled
Token and TopK in task-average accuracy and matches or exceeds Full Vocabulary. Gradient cosine similarity reaches 99% on 4B mathematics, while 8K full-parameter profiling shows 70.5% lower backward
memory.
Read the full abstract ▾
All authors Yongliang Miao, Shuang Liu, Yanguang Liu, Yandong Bai, Mengnan Du
Computation and Language
Just-In-Time Agent Memory with Runtime Agentic Research
Bingyu Yan, Chaofan Li, Hongjin Qian et al. (+3 more)
Memory is critical for AI
agents. Many existing
agent-
memory systems follow an Ahead-of-Time (AOT) design, constructing
memory before a specific request arrives. While this reduces online serving cost, such request-agnostic
memory construction can discard fine-grained information that later becomes important. To address this limitation, we propose Just-In-Time
Agent Memory (JAM), a trainable framework for query-conditioned context construction at runtime. A Memorizer preserves complete raw histories in a hierarchical page-store with compact navigational summaries, while a Researcher iteratively retrieves, inspects, and integrates evidence for each request. To train these
memory-use behaviors, we introduce
Memory-Gym, an evidence-grounded data synthesis pipeline covering nine task types across six domains, and optimize the Researcher through verified-trajectory supervised fine-tuning followed by Hint-guided Group Relative Policy Optimization. We demonstrate the effectiveness of JAM across a variety of
benchmarks on
agent memory and long-context processing, where it achieves stronger task performance than AOT-style
memory systems while remaining substantially more efficient than prior trained agentic
memory approaches. To support reproducibility and future research, we release our anonymized source code at https://github.com/VectorSpaceLab/general-agentic-
memory.
Read the full abstract ▾
All authors Bingyu Yan, Chaofan Li, Hongjin Qian, Shuqi Lu, Chaozhuo Li, Zheng Liu
Computation and Language
When Harness Beats Scale, and When Reading Beats Both
Ivan Bondarenko and Nikolay O. Nikitin
We describe our system for DocSem, the document-grounded quantitative
reasoning shared task at DocInsights 2026, and analyze why it succeeded on labeled data and failed on the test set. The pipeline pairs hybrid block
retrieval with Program-of-Thoughts (PoT) generation executed in a sandboxed interpreter, self-consistency sampling, and entity enrichment from chunk-level knowledge graphs. On our held-out split, application architecture moved the metrics far more than model scale did: PoT added 0.282 joint accuracy to a compact 7B model but at most 0.005 to a 72B model, and a 27B model with the full harness matched the 72B (0.884 vs.\ 0.873) at roughly 2.7$\times$ fewer parameters and a quarter of the CO$_2$. We read this through a distinction between world knowledge, which scales steeply with parameters, and
language knowledge, which scales gently, and show that structured-output training makes a compact model harness-ready rather than merely small. On the raster, watermarked test PDFs the same system collapsed to 13.58% joint (rank 149 of 163); a controlled re-rendering of the validation set reproduces the OCR half of the collapse while bounding what the simulation misses. Auditing the physical nature of evaluation inputs precedes architecture, and the leaderboard's bimodality is consistent with reading quality, not
reasoning, having separated the field.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.LG
FORGE: Form-Optimal Routing of Grounded Evidence for Frozen LLM Agents
Xi Xiao, Yunbei Zhang, Chen Liu et al. (+7 more)
In agentic AI systems, frozen foundation models are increasingly deployed as closed-weight API endpoints, making downstream adaptation possible only through the inputs and inference procedures surrounding the model. As a result, for each input query, two coupled decisions largely determine both answer quality and
token cost: what evidence to provide and how much
reasoning budget to allocate. Fixed defaults along these axes are often suboptimal, misallocating support form or
reasoning depth on roughly 80% of queries in our analysis. To address this challenge, we propose FORGE, a unified framework for adapting frozen models through per-query routing over a joint action space that spans both support form and thinking depth. Under an
entropy-regularized, cost-aware utility objective, we derive a closed-form Boltzmann routing target and instantiate the policy as a lightweight 269K-parameter factorized router. The routing policy is trained around the frozen host, without any weight access, through a three-stage pipeline: offline arm enumeration, supervised Kullback-Leibler (KL)
distillation from the Boltzmann target, and Group Relative Policy Optimization (GRPO) refinement with host feedback. Across 5 knowledge-intensive
benchmarks and 8 frozen backbones ranging from 7B to 671B parameters, FORGE improves accuracy at 42-45% lower
token cost on both main hosts, transfers zero-shot across hosts at lower
token cost, and composes with intrinsic thinking budgets where available.
Read the full abstract ▾
All authors Xi Xiao, Yunbei Zhang, Chen Liu, Lin Zhao, Jialin Chen, Tianchen Zhao, Xiang Xu, Youngeun Kim, Tianyang Wang, Min Xu
Computation and Language · Cross-listed from cs.LG
Commutator Memory: Sparse, Path-Local Reading and Steering in Language Models
John Sweeney
Gradient updates on different data generally do not commute: training a
language model on two data sources in opposite orders gives different weights, even with the same data and total exposure. Loss or
benchmark deltas show that the models differ, not where. We ask whether this path dependence leaves a parametric training-history
memory: a weight component that flips sign when the two sources are swapped, is localized in output space, changes the held-out loss gap between the two orders under targeted interventions, and reveals which trained model came from which order. For one small SGD step of size $\eta$ on each of sources $A$ and $B$, the weight difference $\theta_{AB}-\theta_{BA}$ is, to leading order, $\eta^2 b_{AB}$, where $b_{AB}=H_Bg_A-H_Ag_B$ is the Lie bracket of the two gradient fields at the base model. We define commutator
memory by projecting the bracket through the logits into one score per vocabulary
token; the scores sum to the bracket's prediction of the gap. The scores are localized: on three models, the same readout of the measured $\theta_{AB}-\theta_{BA}$, or of a bracket from disjoint batches, shares 82-99% of the original top-20
tokens, versus 35-49% for norm-matched random directions. They are causally actionable: in Qwen-3-4B SFT, downweighting the ten
tokens with the largest predicted share of the gap closes a median 32% of the measured gap, while frequency-matched
tokens with near-zero scores have almost no effect. The weights themselves carry the component: projecting the difference between the two trained models onto $b_{AB}$ identifies which came from which order in 92% of cases across four LLMs (chance 50%). Controlled tests also cover matched-batch DPO, a frozen-rollout GRPO-style objective, and an AdamW endpoint check. The
memory is defined per source pair, not per example, and its projection on $b_{AB}$ decays with further training.
Read the full abstract ▾
Computation and Language
CRISP: Cultural Reward Modeling for Implicit Situated Propriety
Zekun Yuan, Yangfan Ye, Baohang Li et al. (+6 more)
As large
language models (LLMs) are increasingly deployed across countries and regions, the ability to recognize and respond appropriately to diverse cultural contexts becomes increasingly important. However, existing research has largely focused on cultural knowledge or tasks with predefined response spaces, while open-ended culturally situated behavior remains comparatively underexplored. In this work, we introduce CRISP-RM, a culturally situated reward model that assigns rewards according to cultural appropriateness in open-ended social scenarios. During policy optimization, we further introduce Norm Grounding Supervision (NGS), providing guidance that enhances the policy's sensitivity to relevant cultural norms. To construct culturally situated data, we employ a collaborative multi-
agent framework that instantiates implicit cultural norms into diverse social scenarios and further curate NormCompass as a dedicated testbed. We conduct comprehensive experiments to evaluate the effectiveness of CRISP-RM in both reward modeling and policy optimization. Best-of-\(N\) experiments show that CRISP-RM consistently outperforms strong general reward models. During GRPO policy optimization, CRISP-RM generally improves culturally situated behavior, while incorporating NGS yields further gains. Further analyses demonstrate the advantages of CRISP-RM in distinguishing culturally appropriate behavior beyond superficial fluency and politeness, while NGS provides complementary gains during policy optimization by improving norm grounding.
Read the full abstract ▾
All authors Zekun Yuan, Yangfan Ye, Baohang Li, Shuaibo Zhao, Zekun Zhou, Ziming Li, Qichen Hong, Kun Chen, Xiaocheng Feng