LLM-based search agents are widely used for information-seeking tasks, but their reliance on external tool returns introduces a critical security risk: web content retrieved during execution is untrusted, exposing agents to prompt injection and goal hijacking. Prior work on search-agent safety primarily focuses on static web-content injection, but modern agents issue follow-up queries and cross-check competing sources, so a single injected page is often diluted or rejected. We show that the channel delivering search and page observations is a fragile security boundary: beyond exposing the agent to a single poisoned page, a mediated search interface can repeatedly steer how the agent gathers evidence and forms its final answer. Under a constrained tool-intermediary threat model, appending only one controlled result per query can substantially increase attack success when the evidence is coordinated across the agent's trajectory. We study this setting with a strategy-driven long-horizon attack system and introduce Authority-Chain Hijack (ACH), an expert-refined strategy that turns isolated search-result and page-content manipulations into a coherent evidence chain across seemingly corroborating sources. ACH achieves the highest Overall ASR among all baselines, reaching 55.9% / 83.3% ASR / MaxN ASR on the full SafeSearch test split. We further introduce Trace-Guided Strategy Evolution (TGSE), which automatically improves attacker strategies from execution traces, replacing manual redesign with trace-driven refinement; its strongest single setting reaches 71.4% / 95.0% in held-out evaluation. Code is available at https://github.com/MBOSSLESS/Breadcrumbing-Search-Agents.
Hazel H. Kim, Jakub Binkowski, Ekaterina Fadeeva, Juan David Guerra, Yijie Hao, Zheng Yi Ho, Baixiang Huang, Buyun Liang 等 25 位作者
Hallucination remains one of the most critical challenges in deploying large language models (LLMs) reliably and safely. Yet the term is used inconsistently across the research community. We argue for a strict definition of hallucinations in LLMs. First, hallucinations should be confident incorrectness, kept separate from expressed uncertainty (UNK)1. UNK reveals that the model may be wrong, whereas a hallucination does not reveal it and is therefore more dangerous in practice. Conflating the two obscures their distinct behaviors and makes them hard to study. Second, hallucinations should be split by a simple re-elicitation test: does any semantically equivalent prompt recover the correct response? This test probes whether the relevant knowledge2 is accessible. If none does, the knowledge is likely absent or inaccessible, and the error type is a confabulation (CF) that likely requires external knowledge or retraining; if some does, the knowledge is accessible from the model but applied inconsistently, and the error type is a semantic illusion (SI) that simple prompt or decoding changes can mitigate3. Re-evaluating 14 recent hallucination methods under this taxonomy yields three findings. (1) Detection methods flag UNK more often than hallucinated responses (i.e., SI and CF). (2) Mitigation methods are far more effective for SI than for CF. (3) Latent signals are more informative than output-level signals for CF. An audit of recent hallucination papers shows that most conflate these categories. We recommend reporting CF, SI, and UNK separately, and introduce a practical labeling protocol that needs only a modest number of semantically equivalent prompts.
Test-time training (TTT) enables a deployed language model to internalize incoming context on-the-fly into a small set of fast-weight parameters, rather than merely treating it as transient context for attention. In-Place TTT recently made this idea practical for large language models by repurposing the down-projection of standard MLP blocks as fast weights. However, we find that its benefit largely disappears on realistic web long-context question answering. We attribute this to In-Place TTT's uniform write schedule: every chunk is written with equal strength, so irrelevant content dilutes query-relevant information in fast-weight memory. Building on this insight, we propose Select-TTT, a model-agnostic, query-conditioned rule for In-Place TTT that scales each chunk's write strength by its relevance to the user query. This relevance signal requires no supervision and no new parameters: it is read directly from the model's frozen output head, as the probability mass that a chunk's hidden states assign to the query tokens. A temperature-scaled softmax turns these scores into learning-rate multipliers, leaving the chunk-wise, context-parallel computation of In-Place TTT intact. Select-TTT consistently improves long-context question answering on two benchmarks and two backbones. These findings highlight the value of query-conditioned fast weight writes for practical test-time training on long-context question answering, which is also potentially promising for long-horizon agent tasks involving multiple user queries over evolving contexts.
As agents collaborate on longer tasks, responsibility may pass between models before the work is complete. The receiving model must build on prior decisions, observations, and actions while addressing the remaining requirements. Evaluating this transition requires connecting the information preserved at handoff to how the receiver continues the work, a relationship that task-success and memory-retention scores alone do not capture. We introduce the Handoff State Continuity Benchmark (HSC-Bench), comprising 1,039 tasks that evaluate two complementary aspects of task takeover: recovering relevant repository evidence and continuing from established conversational, tool, and file-work state. Comparisons across six receiving models show that structured handoff state improves repository evidence recovery for every model and continuation performance for most models. The strongest receivers differ between the two tracks, showing that evidence-recovery rankings alone do not characterize performance when continuing from inherited state. By combining these comparisons with checks of subsequent responses and actions, HSC-Bench evaluates both access to prior evidence and adherence to established commitments and completed work. This provides a basis for selecting receiving models and assessing handoff representations according to the work that must be carried forward.
My Chiffon Nguyen, Tianyi Hu, Vinod Anbalagan, Henry Gagnier, Pouya Sadeghi, Farah Abdou, Tidor-Vlad Pricope, Diego MMS Oliveira 等 164 位作者
As LLMs expand their language coverage, evaluating how well they understand and represent diverse cultures becomes increasingly important. However, many benchmarks remain English-centric, with limited cultural coverage and a focus on factual knowledge over its application in real-world cultural contexts. We propose CulturalRiddles, a benchmark of original, culturally grounded riddles created entirely by humans through a participatory community effort, spanning 62 cultures (45 countries, 51 languages). These riddles are deeply rooted in their respective cultures and require abductive and compositional reasoning, alongside detailed factual and social knowledge, language skills, and an understanding of culturally specific associations. We benchmark 28 LLM configurations using both automatic and large-scale human evaluation. Even frontier models achieve at most 22% accuracy in some communities. We publicly release all artifacts alongside an open contribution process to expand the benchmark's linguistic and cultural coverage.
Zefeng He, Yichen He, Yuan Lin, Yang Yang, Yu Cheng
Large language models (LLMs) can produce plausible but incorrect answers when they are uncertain, which can mislead users and undermine model reliability. Training models to abstain with reinforcement learning is a natural approach, but most existing methods use a fixed error penalty and therefore cannot adapt their behavior to different confidence requirements at inference time. We study \emph{explicit confidence targets}, where a user specifies a target \(p\) in the prompt and the model directly chooses whether to answer or abstain, with the goal of achieving an accuracy of at least \(p\) among answered questions. However, learning to respond to different confidence targets with a single model has proven difficult, and prior attempts have failed to distinguish different targets. We show that this failure comes from two problems: the reward signal that distinguishes different values of $p$ is often lost during optimization, while the original error penalty grows without bound as $p$ approaches one. We propose RLEC (\textbf{R}einforcement \textbf{L}earning with \textbf{E}xplicit \textbf{C}onfidence Targets), which preserves the learning signal needed to distinguish confidence targets, introduces a stable reward family that keeps the desired answer condition unchanged, and uses reward scheduling to balance effective learning and training stability. Together, these designs enable the first successful training of a single LLM for explicit confidence target control. Extensive experiments across diverse benchmarks and model scales show that RLEC largely preserves general capabilities and achieves competitive calibration performance. RLEC also supports fine-grained target control without an external decision system or threshold calibration after training. By making explicit confidence targets part of the trained policy, RLEC opens a new and more direct path toward behavioral calibration.
Multi-Turn Long-Form Question Answering (MT-LFQA) is a key application paradigm of Large Language Models (LLMs) in knowledge-intensive domains. However, existing benchmarks are limited to single-turn dialogue, while multi-turn dialogue benchmarks typically assess other orthogonal capabilities rather than knowledge-intensive factuality. To bridge this critical gap, we introduce \textbf{KnowMT-Bench}, a new benchmark designed to systematically evaluate MT-LFQA for LLMs across knowledge-intensive fields, including medicine, finance, and law. To faithfully assess the model's real-world performance, KnowMT-Bench employs a dynamic evaluation setting where models generate their own multi-turn dialogue histories given logically progressive question sequences. Our experiments on a diverse suite of LLMs show a clear degradation in both factual capability and information delivery efficiency in multi-turn dialogue. We further probe the underlying causes and investigate mitigation strategies spanning domain-specific finetuning, active context denoising, and RAG. The results show domain-dependent gains from finetuning, while active context denoising and RAG more consistently mitigate these issues, with RAG notably reversing this performance degradation. These findings underscore the importance of our benchmark for evaluating LLMs' conversational factual capabilities in real-world applications. Code and data are publicly available.\footnote{\href{https://github.com/Sven-LI-sankyuu/KnowMT-Bench}{\texttt{https://github.com/Sven-LI-sankyuu/KnowMT-Bench}}}
Xiangchi Yuan, Xingjian Diao, Zheyuan Liu, Chunhui Zhang, Dachuan Shi, Kejing Xia, Zhenbang Du, Shenglong Yao 等 10 位作者
LLM post-training often optimizes multiple objectives with unequal priorities, such as correctness and reasoning efficiency. However, the widely used additive reward aggregation in RL can violate these priorities: when primary and auxiliary advantages conflict, a sufficiently large auxiliary signal can overturn the preference induced by the primary objective. We term this problem \emph{advantage priority inversion}. Although reward-level gating can protect priority more directly, it still removes potentially useful auxiliary distinctions among rollout samples that fail the primary condition. To address this trade-off, we propose Primary Advantage Policy Optimization (\textbf{PAPO}), a priority-aware advantage aggregation method that asymmetrically combines a primary and an auxiliary advantage. PAPO retains auxiliary supervision when the objectives agree and smoothly attenuates conflicting auxiliary signals, preserving the primary sign during fusion for \emph{primary-consistent ordering}. This rule applies to sequence-level advantages in GRPO and token-level advantages in actor--critic PPO. Experiments on multiple domains and models show that PAPO consistently achieves the best primary objective while maintaining best or competitive auxiliary-objective performance across both GRPO- and PPO-based optimization. \href{https://xiangchi-yuan.github.io/papo-project/}{Code is available}.
Can AI agents autonomously train useful language models and improve them using feedback from users? We introduce RSI Arena, a two-stage study of autonomous post-training as a step toward recursive self-improvement. Ten frontier models act as training agents, each starting from NVIDIA Nemotron 3.5 Lightning 30B-A3B Base. During a six-day first stage, each agent has 1,000 GPU-hours and $300 in API credit on a shared 64-GPU cluster, independently selecting data, implementing training procedures, and choosing checkpoints. We evaluate their models on 20 benchmarks and through blind human evaluations. All ten agents produce working assistants, with the strongest model, trained by GPT-6 Astra, closing a median of 76% of the base-to-Instruct gap across 12 capability benchmarks. Its human-preference rating is also close to NVIDIA Instruct after statistical adjustment for answer length and formatting. In Stage 2, each agent receives the comparison records and votes involving its Stage 1 model and is given three additional days for further training. Using NVIDIA Instruct as a common reference across the two evaluation rounds, we observe an improvement from Stage 1 to Stage 2 in the style-adjusted relative rating of eight of the ten models. We show that benchmark gains were modest, with no clear link to preference gains, while some agents encouraged longer responses and one model showed increased harmful compliance on StrongREJECT. RSI Arena provides an auditable setting for studying how human feedback can guide autonomous model development. It takes a step toward recursive self-improvement with human feedback (RSIHF), in which people help define what improvement means for the AI systems they use.
Qundong Shi ModelBest, Jie Zhou, Biyuan Lin, Junbo Cui ModelBest, Guoyang Zeng ModelBest, Yixuan Zhou Tsinghua University, Ziyang Wang, Xin Liu 等 11 位作者
The development of audio foundation models has accelerated rapidly since the emergence of GPT-4o. However, the lack of comprehensive evaluation has become a critical bottleneck for further progress in the field, particularly in audio generation. Current audio evaluation faces three major challenges: (1) audio evaluation lacks a unified framework, with datasets and code scattered across various sources;(2) audio codec, as a key component of audio foundation models, lacks a widely accepted and holistic evaluation methodology; (3) existing speech benchmarks are heavily reliant on English, making it challenging to objectively assess models’ performance on Chinese.We introduce UltraEval-Audio, a unified framework addressing these challenges through a modular architecture supporting 10 languages, 14 task categories, 24 models, and 36 benchmarks with one-command evaluation and real-time leaderboards. For audio codec, we propose a three-dimensional evaluation scheme covering semantic accuracy, timbre fidelity, and acoustic quality. For Chinese evaluation, we introduce two new benchmarks: SpeechCMMLU and SpeechHSK. Our code, benchmarks, and leaderboards are available at https://github.com/OpenBMB/UltraEval-Audio.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations) · 2026 · 正式发表
Xuanle Zhao York University and Institute of automation, Chinese academy of science, Chinese Academy of Sciences, Zilin Sang, Yuxuan Li, Qi Shi Tsinghua University, Weilun Zhao ModelBest, Shuo Wang, Duzhen Zhang Mohamed bin Zayed University of Artificial Intelligence, Xu Han Tsinghua University, Tsinghua University 等 10 位作者
Efficient reproduction of research papers is pivotal to accelerating scientific progress. However, the increasing complexity of proposed methods often renders reproduction a labor-intensive endeavor, necessitating profound domain expertise.To address this, we introduce the paper lineage, which systematically mines implicit knowledge from the cited literature. This algorithm serves as the backbone of our proposed , a multi-agent framework designed to autonomously reproduce experimental code in a complete, end-to-end manner. To ensure code executability, incorporates a sampling-based unit testing strategy for rapid validation. To assess reproduction capabilities, we introduce , a benchmark featuring verified implementations, alongside comprehensive metrics for evaluating both reproduction and execution fidelity. Extensive evaluations on PaperBench and demonstrate that consistently surpasses existing baselines across all metrics. Notably, it yields substantial improvements in reproduction fidelity and final execution performance. The code is available at https://github.com/AI9Stars/AutoReproduce.
尚无 AI 解读，展示原文摘要
cs.CL · cs.MA
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Chengao Liu University of the Chinese Academy of Sciences, Yuan Li, Yingze Wang, Jianbin Jiao University of the Chinese Academy of Sciences
Temporal Knowledge Graph (TKG) reasoning remains challenging to characterize with conventional flat representations due to its intrinsic heterogeneous structure. Existing multi-geometry approaches face two key bottlenecks: 1) the Riemannian depth barrier driven by numerical instability, which restricts models to shallow architectures; and 2) gate collapse, where adaptive fusion mechanisms suffer from gradient starvation and degenerate into single-geometry solutions. To this end, we propose MAGIC (Multi-geometry Annealing Graph Interaction with Consensus). Our framework introduces a Tangent-Residual Engine in multi-geometric spaces, which enables the first stable 8-layer geometric evolution and reveals a phenomenon termed Geometric Annealing, where manifold curvature spontaneously evolves from semantic flatness in shallow layers to structural complexity in deeper layers. We further design an explicit reasoning module with structural consensus, leveraging geometric invariants and structural priors to regulate gradient flow, prevent collapse, and ensure robust synergy across Hyperbolic, Spherical, and Euclidean spaces. Experiments show that MAGIC achieves state-of-the-art performance in TKG reasoning, improving MRR by up to 2.9 points.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Zhichen Liu, Yongyuan Li, Yang Xu Southern University of Science and Technology
Researchers have explored ways to improve large language models (LLMs)’ capabilities via dummy token insertion in contexts. However, existing works focus solely on the dummy tokens themselves, but failed to leverage the inherent sentence-level structure of natural language. This is a critical oversight, as LLMs acquire linguistic capabilities through exposure to human-generated texts, which are inherently structured at the sentence level. Motivated by the gap, we proposed a method that inserts delimiters at sentence boundaries. Our method not only integrates dummy tokens into contexts, but also enables LLMs with sentence-by-sentence processing behavior during reasoning. Two approaches are proposed: (1). In-context learning and (2). Supervised fine-tuning are experimented from 7B LLMs to 600B Deepseek-V3. Experimental results demonstrate consistent improvements in various tasks, with notable gains of up to 7.7% on GSM8k and 12.5% on DROP. Furthermore, LLMs fine-tuned via our strategy further incorporate sentence awareness into their inner representations. Our work establishes a simple yet effective technique for enhancing LLM’s capabilities, offering promising directions for cognitive-inspired LLM enhancement paradigm.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Ding Xia The University of Tokyo, Xinyue Gui, Mark Colley University College London, University of London, Fan Gao, Zhongyi Zhou Google, Dongyuan Li The University of Tokyo and Institute of Science Tokyo, Renhe Jiang The University of Tokyo, Takeo Igarashi The University of Tokyo, Tokyo Institute of Technology
Automated vehicles lack natural communication channels with other road users, making external Human-Machine Interfaces (eHMIs) essential for conveying intent and maintaining trust in shared environments. However, most eHMI studies rely on developer-crafted message-action pairs, which are difficult to adapt to diverse and dynamic traffic contexts. A promising alternative is to use Large Language Models (LLMs) as action designers that generate context-conditioned eHMI actions, yet such designers lack perceptual verification and typically depend on fixed prompts or costly human-annotated feedback for improvement.We present See2Refine, a human-free, closed-loop framework that uses vision-language models (VLMs) for perceptual evaluation as automated visual feedback to improve an LLM-based eHMI action designer. Given a driving context and a candidate eHMI action, the VLM evaluates the perceived appropriateness of the action, and this feedback is used to iteratively revise the designer’s outputs, enabling systematic refinement without human supervision.We evaluate our framework across three eHMI modalities (lightbar, eyes, and arm) and multiple LLM model sizes. Across settings, our framework consistently outperforms prompt-only LLM designers and manually specified baselines in both VLM-based metrics and human-subject evaluations. The results further indicate that the improvements are generalized across modalities and that VLM evaluations are reasonably aligned with human preferences in our controlled settings, supporting the robustness and effectiveness of See2Refine for scalable action design.
尚无 AI 解读，展示原文摘要
cs.CL · cs.CV
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Zhaoxuan Tan University of Notre Dame, Zixuan Zhang Amazon, Haoyang Wen Amazon, Zheng Li, Rongzhi Zhang Georgia Institute of Technology, Zhejiang University and Amazon, Pei Chen Amazon, Fengran Mo, Zheyuan Liu University of Notre Dame 等 11 位作者
Personalized large language models (LLMs) tailor content to individual preferences using user profiles or histories. However, existing parameter-efficient fine-tuning (PEFT) methods, such as the “One-PEFT-Per-User” (OPPU) paradigm, require training a separate adapter for each user, making them computationally expensive and impractical for real-time updates. We introduce Profile-to-PEFT, a scalable framework that employs a hypernetwork, trained end-to-end, to map a user’s encoded profile directly to a full set of adapter parameters (e.g., LoRA), eliminating per-user training at deployment. This design enables instant adaptation, generalization to unseen users, and privacy-preserving local deployment. Experimental results demonstrate that our method outperforms both prompt-based personalization and OPPU while using substantially fewer computational resources at deployment. The framework exhibits strong generalization to out-of-distribution users and maintains robustness across varying user activity levels and different embedding backbones. The proposed Profile-to-PEFT framework enables efficient, scalable, and adaptive LLM personalization suitable for large-scale applications.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Bryan Chen Zhengyu Tan, Zhengyuan Liu Agency for Science, Technology and Research (A*STAR), Singapore, Xiaoyuan Yi Microsoft Research, Jing Yao Microsoft, Xing Xie Microsoft Research Asia, Nancy F. Chen A*STAR, Roy Ka-Wei Lee Singapore University of Technology and Design
Despite their global prevalence, many Large Language Models (LLMs) are aligned to a monolithic, often Western-centric set of values. This paper investigates the more challenging task of fine-grained value alignment: examining whether LLMs can emulate the distinct cultural values of demographic subgroups. Using Singapore as a case study and the World Values Survey (WVS), we examine the value landscape and show that even state-of-the-art models like GPT-4.1 achieve only 57.4% accuracy in predicting subgroup modal preferences. We construct a dataset of over 20,000 samples to train and evaluate a range of models. We demonstrate that simple fine-tuning on structured numerical preferences yields substantial gains, improving accuracy on unseen, out-of-distribution subgroups by an average of 17.4%. These gains partially transfer to open-ended generation. However, we find significant pre-existing performance biases, where models better emulate young, male, Chinese, and Christian personas. Furthermore, while fine-tuning improves average performance, it widens the disparity between subgroups when measured by distance-aware metrics. Our work offers insights into the limits and fairness implications of subgroup-level cultural alignment.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Zhirong Chen, Kaiyan Chang Institute of Computing Technology, CAS, Zhuolin Li Peking University, Cangyuan Li , Chinese Academy of Sciences, Xinyang He, Chujie Chen, Mengdi Wang, Haobo Xu 等 11 位作者
Large Language Models have emerged as powerful tools for automating Register-Transfer Level (RTL) code generation, yet they face critical limitations: existing approaches typically fail to simultaneously optimize functional correctness and hardware efficiency metrics such as Power, Performance, and Area (PPA). Methods relying on supervised fine-tuning commonly produce functionally correct but suboptimal designs due to the lack of inherent mechanisms for learning hardware optimization principles. Conversely, external post-processing techniques aiming to refine PPA performance after generation often suffer from inefficiency and do not improve the LLMs’ intrinsic capabilities.To overcome these challenges, we propose ChipSeek, a novel hierarchical reward based reinforcement learning framework designed to encourage LLMs to generate RTL code that is both functionally correct and optimized for PPA metrics. Our approach integrates direct feedback from EDA simulators and synthesis tools into a hierarchical reward mechanism, facilitating a nuanced understanding of hardware design trade-offs. Through Curriculum-Guided Dynamic Policy Optimization (CDPO), ChipSeek enhances the LLM’s ability to generate high-quality, optimized RTL code. Evaluations on standard benchmarks demonstrate ChipSeek’s superior performance, achieving state-of-the-art functional correctness and PPA performance. Furthermore, it excels in specific optimization tasks, consistently yielding highly efficient designs when individually targeting fine-grained optimization goals such as power, delay, and area. The artifact is open-source in https://github.com/rong-hash/chipseek.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Zhiyu Shen, Jiyuan Liu, Yunhe Pang, Yanghui Rao SUN YAT-SEN UNIVERSITY, Fu Lee Wang Hong Kong Metropolitan University, Jianxing Yu
Multi-Hop Question Answering (MHQA) is crucial for evaluating the model’s capability to integrate information from diverse sources. However, creating extensive and high-quality MHQA datasets is challenging: (i) manual annotation is expensive, and (ii) current synthesis methods often produce simplistic questions or require extensive manual guidance. This paper introduces HopWeaver, the first cross-document framework synthesizing authentic multi-hop questions without human intervention. HopWeaver synthesizes bridge and comparison questions through an innovative pipeline that identifies complementary documents and constructs authentic reasoning paths to ensure true multi-hop reasoning. We further present a comprehensive system for evaluating the synthesized multi-hop questions. Empirical evaluations demonstrate that the synthesized questions achieve comparable or superior quality to human-annotated datasets at a lower cost. Our framework provides a valuable tool for the research community: it can automatically generate challenging benchmarks from any raw corpus, which opens new avenues for both evaluation and targeted training to improve the reasoning capabilities of advanced question answering models, especially in domains with scarce resources.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Yubo Shan, Kun Zhang Tencent, Qiming Xu Beijing Technology and Business University, Liping Cao, Yingying Cao, Jian Zhang, Yu Wang, Jingyuan Li 等 9 位作者
Interleaved multimodal understanding and generation—where models can interactively comprehend and produce images and text in arbitrary orders—has emerged as a key research direction in generative Multimodal Large Language Models(MLLMs). Such interleaved image–text content plays an increasingly important role in information dissemination. However, the compounded persuasive power of multimodal narratives also raises the risk of factual misinformation. Despite this, existing benchmarks lack effective mechanisms to evaluate factual consistency in interleaved image–text content. To bridge this gap, we introduce FactVerse, a benchmark dedicated to evaluating factual consistency in interleaved image-text generation. FactVerse comprises 3,000 human-verified instances across four categories and 50 domains, supporting both English and Chinese. We also establish a multi-dimensional evaluation framework designed to rigorously assess factual consistency. Experiments demonstrate that our framework achieves high alignment with human judgments, significantly outperforming existing evaluation methods. Furthermore, our analysis reveals systematic deficiencies in current models, offering critical insights for future design.
尚无 AI 解读，展示原文摘要
cs.CL · cs.CV
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Fan Liu, Yanhao Wang East China Normal University, Min Zhang East China Normal University, Zhikang Chen University of Oxford, Zeyuan Li, Lewei He, Jiahui Pan South China Normal University
This paper proposes shortcut decoding, an efficient framework for accelerating Chain-of-Thought (CoT) reasoning in Large Language Models (LLMs). Existing methods that prune or employ early stopping to reduce latency often compromise reasoning reliability. Motivated by the observation that LLMs frequently converge to correct solutions internally before completing explicit textual reasoning, we propose a dual-signal adaptive controller that integrates lightweight probes over internal hidden states with step-level entropy. This controller detects convergence of reasoning during generation and adaptively selects between a fast-exit path and a stability-verified path to remove redundant steps while preserving answer correctness. Experiments across multiple mathematical reasoning benchmarks demonstrate that shortcut decoding reduces token usage by approximately 35%, maintains accuracy comparable to full CoT decoding, and achieves final-answer accuracy comparable to the full CoT baseline, outperforming existing early-stopping methods without updating the base model. Our code is available at https://github.com/kuromi9527/shortcut_decoding.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Dawei Wang, Di Zhao University of Auckland, Xinyuan Liu, Marci Chi Ma Newcastle University, UK, Xiaoyang Liu Newcastle University, UK, Chengming Zhou, Gary Ushaw Newcastle University, UK, Richard Davison University of Newcastle-upon-Tyne
Credit assignment is a fundamental challenge in cooperative multi-agent reinforcement learning, particularly in embodied AI settings characterized by limited and delayed feedback as well as dynamically changing numbers of active agents. We propose MARS-RA, a framework that reformulates credit assignment as a rank aggregation problem using contribution-based pairwise comparisons among agents generated by large multimodal models. This shift from absolute to relative estimation ensures robustness against noise and dynamic agent participation, converting comparison results into contribution scores for potential-based reward shaping. We provide theoretical justification for the convergence and robustness of the proposed framework, and show that Shapley values can be used as an interpretive reference. Experimental results on challenging tasks of different types indicate that MARS-RA can guide agents toward effective cooperation.
尚无 AI 解读，展示原文摘要
cs.CL · cs.MA · cs.CV
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
The deployment of large language models (LLMs) is largely hindered by their large number of parameters. Structural pruning has emerged as a promising solution. Prior structured pruning methods directly remove unimportant parameters based on certain metrics, which often causes knowledge loss and necessitates extensive retraining. To overcome this, we introduce a novel pruning method TRSP: Two-Stage Regularization-Based Structured Pruning for LLMs. Specifically, we multiply the output of each transformer layer by an initial learnable weight and iteratively learn these weights by adding their \ell_1-norm as a regularization term to the loss function, serving as the first-stage regularization. Subsequently, we apply additional regularization to the difference between the output and input of layers with smaller weights, encouraging the shift of knowledge to the preserved layers. This serves as the second-stage regularization. TRSP retains more knowledge and better preserves model performance than direct parameter elimination. Through extensive experimentation we show that TRSP outperforms strong layer-wise structured pruning methods without requiring retraining. As a layer-wise pruning method, it delivers notable end-to-end acceleration, making it a promising solution for efficient LLM deployment.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Zheng Zhang, Qi Liu University of Science and Technology of China, Siyuan Liang Nanyang Technological University, Ning Li, Zirui Hu Nanyang Technological University, Weibo Gao University of Science and Technology of China, Rui Li, Zhenya Huang University of Science and Technology of China 等 11 位作者
Large language models (LLMs) have achieved remarkable performance across diverse tasks, largely driven by large-scale pretraining. However, this data abundance introduces test data contamination, where benchmark datasets overlap with pretraining corpora, undermining the reliability of model evaluation by confounding memorization with genuine generalization. To mitigate this issue, existing training data detectors attempt to identify clean (unseen) samples from contaminated test sets, but often suffer from residual contamination due to the black-box nature of LLMs. As a result, contaminated data may be mistakenly retained, leading to unreliable evaluation.To address this challenge, we propose FTD (FDR-controlled Training Data detection), a principled framework that detects and filters contaminated evaluation data while providing a statistical guarantee: the proportion of contaminated samples mistakenly retained as clean, the false discovery rate (FDR), is provably controlled below a user-specified threshold. FTD combines multiple complementary detectors via an adaptive weighting strategy, and we theoretically show it achieves high statistical power under valid FDR control. Extensive experiments on real-world benchmarks demonstrate that FTD significantly reduces residual contamination compared to existing methods while preserving evaluation consistency.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Zhihao Gong Peking University, Zeyu Sun Institute of Software Chinese Academy of Sciences, Dong Huang National University of Singapore, Qingyuan Liang, Jie M. Zhang, Dan Hao Peking University
While Large Language Models (LLMs) have substantially improved the functional correctness of code translation, the critical dimension of execution efficiency remains overlooked. We present Trace, the first benchmark to explicitly assess efficiency in LLM-translated code. Trace includes 1,000 efficiency-critical tasks across C++, Java, and Python, each augmented with stress tests that reveal efficiency disparities often overlooked by small-scale tests. Using Trace, we conduct an extensive evaluation of 28 representative LLMs and highlight several key insights: 1) Correctness and efficiency are often misaligned: the correctness leader Claude-Sonnet-4-Think achieves only moderate time efficiency, outperformed by smaller open-source LLMs such as Qwen2.5-Coder-14B-Instruct. 2) Inefficiency is both prevalent and patterned: 23.5% of correct translations suffer from notable inefficiency, mainly arising from algorithm implementation discrepancy (11.9%), language construct mismatch (66.4%), and resource management inefficiency (21.7%).3) Inference-time prompt strategies bring only modest improvements, indicating that simple prompting alone is insufficient to improve translation efficiency. Together, our results establish execution efficiency as an essential dimension of code translation and position Trace as a principled foundation for efficiency-oriented evaluation. Our code and data are available at: https://github.com/Albert-Gong/TRACE.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Fangyuan Li, Pengfei Li, Shijie Wang, Junqi Gao, Jianxing Liu Harbin Institute of Technology, Biqing Qi University of Hong Kong and Shanghai Artificial Intelligence Laboratory, Yuqiang Li
Recent progress in reinforcement learning with verifiable rewards (RLVR) offers a practical path to self-improving language models, but existing methods face a key trade-off: endogenous self-play can drift over iterations, while corpus-grounded approaches rely on curated data environments. We present WIST, a Web-grounded Iterative Self-play Tree framework for domain-targeted reasoning improvement that learns directly from the open-web without requiring any pre-arranged domain corpus. WIST incrementally expands a domain tree to structure exploration and retrieves and cleans path-consistent web evidence to construct a controllable training environment. It then performs Challenger-Solver self-play with verifiable rewards, and feeds learnability signals back to update node posteriors and guide subsequent exploration through an adaptive curriculum. Across four backbones, WIST consistently improves over the base models and typically outperforms both purely endogenous self-evolution and corpus-grounded self-play baselines, with the Overall gains reaching +9.8 (Qwen3-4B-Base) and +9.7 (OctoThinker-8B-Hybrid-Base). WIST is also domain-steerable: improving Qwen3-8B-Base by +14.79 in medicine and Qwen3-4B-Base by +5.28 on PhyBench. Ablations further confirm the importance of WIST’s key components for stable open-web learning. Our Code is available at https://github.com/lfy-123/WIST.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Daniel Zhang-Li, Joy Jia Yin Lim, Binglin Liu, Shangqing Tu, Zijun Yao, Hao Peng Tsinghua University, Tsinghua University, Jifan Yu, Haoxuan Li Tsinghua University 等 19 位作者
Project-Based Learning (PBL) is an important learning method that promotes understanding and acquiring practical skills through training learners through a project. However, effective PBL often requires sustained orchestration and collaboration, but existing LLM-based learning tools provide partial assistance without explicitly modeling these roles, and overly comprehensive help provided by LLM can reduce learner autonomy. We propose SimPBL, a multi-agent framework with an orchestrator agent that provides adaptive scaffolding from interaction logs and collaborator agents that support project work through boundary-aware collaboration. We conduct comprehensive evaluation to study the effectiveness of SimPBL, where we observe a 14% improvement in learner examination score. Results from extensive studies further highlights the ability of SimPBL to manage learning behavior and improve learning experience. Code and materials are available at https://anonymous.4open.science/r/SimPBL-D5B8.
尚无 AI 解读，展示原文摘要
cs.CL · cs.MA
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Xiangchi Yuan ByteDance Inc., Dachuan Shi Meta Superintelligence Labs and Georgia Institute of Technology, Chunhui Zhang Dartmouth College, Zheyuan Liu University of Notre Dame, Shenglong Yao, Soroush Vosoughi Dartmouth College, Wenke Lee Georgia Institute of Technology
Reinforcement learning (RL) is central to post-training, particularly for agentic models that require specialized reasoning behaviors. In this setting, model merging offers a practical mechanism for integrating multiple RL-trained agents from different tasks into a single generalist model. However, existing merging methods are designed for supervised fine-tuning (SFT), and they are suboptimal to preserve task-specific capabilities on RL-trained agentic models. The root is a task-vector mismatch between RL and SFT: on-policy RL induces task vectors that are highly sparse and heterogeneous, whereas SFT-style merging implicitly assumes dense and globally comparable task vectors. When standard global averaging is applied under this mismatch, RL’s non-overlapping task vectors that encode critical task-specific behaviors are reduced and parameter updates are diluted. To address this issue, we propose Reinforced Agent Merging (RAM), a distribution-aware merging framework explicitly designed for RL-trained agentic models. RAM disentangles shared and task-specific unique parameter updates, averaging shared components while selectively preserving and rescaling unique ones to counteract parameter update dilution. Experiments across multiple agent domains and model architectures demonstrate that RAM not only surpasses merging baselines, but also unlocks synergistic potential among agents to achieve performance superior to that of specialized agents in their domains.
尚无 AI 解读，展示原文摘要
cs.CL · cs.AI
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Pedro Ortiz Suarez, Laurie Burchell Common Crawl Foundation, Catherine Arnett EleutherAI, Rafael Mosquera, Sara Hincapié Monsalve Factored, Thom Vaughan Common Crawl Foundation, Damian Stewart, Malte Ostendorff Common Crawl Foundation 等 97 位作者
Language identification (LID) is a fundamental step in curating multilingual corpora. However, LID models still perform poorly for many languages, especially on the noisy and heterogeneous web data often used to train multilingual language models. In this paper, we introduce CommonLID, a community-driven, human-annotated LID benchmark for the web domain, covering 109 languages. Many of the included languages have been previously under-served, making CommonLID a key resource for developing more representative high-quality text corpora. We show CommonLID’s value by using it, alongside five other common evaluation sets, to test eight popular LID models. We analyse our results to situate our contribution and to provide an overview of the state of the art. In particular, we highlight that existing evaluations overestimate LID accuracy for many languages in the web domain. We make CommonLID and the code used to create it available under an open, permissive license.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Pengfei Li, Shijie Wang, Fangyuan Li, Yikun Fu, Kaifeng Liu, Kaiyan Zhang, Dazhi Zhang Harbin Institute of Technology, Yuqiang Li 等 10 位作者
Reinforcement learning (RL) paradigms have demonstrated strong performance on reasoning-intensive tasks such as code generation. However, limited trajectory diversity often leads to diminishing returns, which constrains the achievable performance ceiling. Search-enhanced RL alleviates this issue by introducing structured exploration, which remains constrained by the single-agent policy priors. Meanwhile, leveraging multiple interacting policies can acquire more diverse exploratory signals, but existing approaches are typically decoupled from structured search. We propose MARS^2 (Multi-Agent Reinforced Tree-Search Scaling), a unified RL framework in which multiple independently-optimized agents collaborate within a shared tree-structured search environment. MARS^2 models the search tree as a learnable multi-agent interaction environment, enabling heterogeneous agents to collaboratively generate and refine candidate solutions within a shared search topology. To support effective learning, we introduce a path-level group advantage formulation based on tree-consistent reward shaping, which facilitates effective credit assignment across complex search trajectories. Experiments on code generation benchmarks show that MARS^2 consistently improves performance across diverse model combinations and training settings, demonstrating the effectiveness of coupling multi-agent collaboration with tree search for enhancing reinforcement learning. Our code is publicly available at https://github.com/TsinghuaC3I/MARTI.
尚无 AI 解读，展示原文摘要
cs.CL · cs.MA
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Yanbin Yin, Kun Zhou University of California, San Diego, Zhen Wang University of California, San Diego, Xiangdong Zhang New York University, Yifei Shao, Shibo Hao University of California, San Diego, Yi Gu University of California, San Diego, Jieyuan Liu 等 14 位作者
The recent explosion of large language models (LLMs), each with its own general or specialized strengths, makes scalable, reliable benchmarking more urgent than ever. Standard practices nowadays face fundamental trade-offs: closed-ended question-based benchmarks (MMLU) struggle with saturation as newer models emerge, while crowd-sourced leaderboards (Chatbot Arena) rely on costly and slow human judges. Recently, automated methods (LLM-as-a-judge) shed light on the scalability, but risk bias by relying on one or a few “authority” models. To tackle these issues, we propose Decentralized Arena (), a fully automated framework leveraging collective intelligence from all LLMs to evaluate each other. It mitigates single-model judge bias by democratic, pairwise evaluation, and remains efficient at scale through two key components: (1) a coarse-to-fine ranking algorithm for fast incremental insertion of new models with sub-quadratic complexity, and (2) an automatic question selection strategy for the construction of new evaluation dimensions. Across extensive experiments across 66 LLMs, attains up to 97% correlation with human judgements, while significantly reducing the cost.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表