Large language models pass most functional-correctness benchmarks, but passing tests says little about whether their code is readable, efficient or secure. Many techniques claim to improve these qualities, yet their evidence is scattered, usually covers one quality dimension and is rarely compared at an equal compute budget. We contribute a systematic review and a controlled re-evaluation. A systematic literature review of 628 primary studies (July 2021 to October 2026), with disclosed, author-reviewed LLM assistance, organizes the techniques by stage, signal source and quality target. Most studies measure only the dimension they target, gains often come with unchanged or unreported correctness, single automated quality signals are gamed, and compute, variance and contamination are rarely reported. A controlled re-evaluation runs ten techniques on four models at matched budgets on correctness, maintainability, efficiency and security benchmarks, extends it to thirteen current models and repository-level tasks, and compares 66 highlighted techniques head-to-head on Gemma 4 31B. At one call, quality prompting raises analyzer issue counts and leaves complexity and judge scores almost unchanged. At matched budgets, best-of-n sampling with an analyzer-weighted ranker matches or beats prompting, planning, self-refinement, a reviewer agent and retrieval on measures the ranker does not optimize, while an analyzer feedback loop and fine-tuning beat it only on analyzer issue counts. Large correctness losses on HumanEval+, a likely contaminated benchmark near its ceiling, do not appear on contests held after the models' training cutoffs. In the head-to-head, 1 of 66 techniques beats best-of-n on its primary target beyond the 95% interval, a rate consistent with chance. Gains should be reported against a budget-matched sampling baseline and on measures independent of the technique's own signal.
Zhuoyuan Hao, Zhuo Li, Wu Li, Fangming Liu, Min Zhang, Jing Li
Recent work improves reasoning in large reasoning models (LRMs) by test-time scaling through self-consistency and parallel thinking, adding generic "thinking tokens", or presenting the question twice, as in re-reading (RE2). However, these approaches either inject task-agnostic tokens or mandate heuristics that do not explain, and often ignore, the spontaneous repetition that many LRMs exhibit at the head of their long chain-of-thought (CoT). In this work, we analyze and harness the model's tendency to restate the question, dubbed Echo of Prompt (EOP), as a front-loaded, compute-shaping mechanism, which can be viewed as self-generated prompt repetition inside the reasoning trace rather than in the input. Specifically, we formalize its probabilistic cost by casting echo removal as rejection-based conditioning, and define the Echo Likelihood Gap $Δ\mathcal{L}$ as a computable proxy linking early repetition to likelihood gains and downstream accuracy. We find that (1) in a causal intervention, reinserting an echo into failed GSM8K traces raises exact match by 7.9-10.4 points for two reasoning models but not for a non-reasoning base model; and (2) in length- and suffix-controlled likelihood analyses and layer-wise attention studies conducted to show that EOP's benefits go beyond verbosity, EOP increases answer to answer-prefix attention in middle layers, consistent with an attention refocusing mechanism. Building on these findings, we develop Echo-Distilled SFT (ED-SFT) to instill an "echo-then-reason" pattern through supervised finetuning, and training-free Echoic Prompting (EP) to re-ground the model mid-trace. We evaluate on GSM8K, MathQA, Hendrycks-MATH, AIME24, and MATH-500 under identical decoding settings and budgets, and find consistent gains over baselines, including thinking-token insertion. Our code is publicly released at https://github.com/hhh2210/echoes-as-anchors.
Language models increasingly create 3D content through code, with practical workflows often relying on editing existing programs rather than regenerating them, for speed and precise control. Such edits require understanding what a program builds and foreseeing how a change reshapes it, yet prior benchmarks mostly evaluate generation or simple edits against a reference. In this paper, we introduce 3DCodeEdit, a benchmark of 2,592 instruction-guided editing tasks over 108 Blender programs that construct objects entirely in code. Its edits range from parameter to structural changes, style redesign, and scene composition, and are judged from renders against human-reviewed checklists. Across 17 models, the average success rate is only 37%, and even the strongest models complete fewer than 60% of structural and style edits. As models become more capable, their failures shift to geometric foresight, where the code runs but builds the wrong shape. We therefore study whether multi-turn feedback can compensate for this missing foresight. Feedback from a separate reviewer model that describes how the renders differ from the request helps most, more than maximum reasoning effort or an agent harness, and can let a 27B model surpass the single-shot rate of a 2.8T-parameter model.
Amogh Sheth, Andrew Lin, Biruk Assefa, Yi Wen Huang, Yuhao Ge
Large language models (LLMs) excel at multi-step reasoning but incur substantial inference cost. We introduce Causal Attribution Pruning (CAP), a training-free method that identifies critical attention heads by measuring their causal impact on reasoning tasks and uses these head-level scores to guide fine-grained weight pruning. For each attention head, CAP estimates the expected performance degradation when the head is masked during forward passes on a small calibration set of reasoning problems. These causal scores are then converted into weight-level importance values for the corresponding projection matrices. Unlike magnitude-only or activation-based criteria, CAP's interventional measurement directly captures each head's functional contribution, yielding relative accuracy gains of up to 61% over Wanda on ARC-Challenge at 20% sparsity. We evaluate CAP on GSM8K, StrategyQA, and ARC-Challenge using Llama-3-8B-Instruct and Mistral-7B-Instruct at 10%, 20%, and 50% sparsity. At moderate sparsity (10-20%), CAP improves over Wanda in most model-benchmark configurations. with especially large gains on ARC-Challenge for Llama-3. Our results suggest that attention-head-level causal attribution can better preserve reasoning performance on downstream benchmarks than correlational pruning criteria at equivalent sparsity, while remaining limited by coarse MLP attribution at 50% sparsity.
Wenwen Hou, Xinyuan Song, Wenqian Weng, Shiwei Liu
Activation sparsity is a promising approach for accelerating large language model (LLM) inference by reducing computation and memory movement. However, existing activation sparsity methods typically apply uniform sparsity across projections, ignoring the heterogeneous statistical properties of Transformer weights and thereby amplifying performance degradation. In this paper, we propose ActTail, a TopK magnitude-based activation sparsity method with global activation sparsity allocation grounded in Heavy-Tailed Self-Regularization (HT-SR) theory. Specifically, we capture this heterogeneity via the heavy-tail exponent computed from each projection's empirical spectral density (ESD), which is used as a quantitative indicator to assign projection-specific sparsity budgets. Importantly, we provide a theoretical analysis that establishes an explicit relationship between the activation sparsity ratio and the heavy-tail exponent under the HT-SR regime, offering principled guidance for sparsity allocation beyond heuristic design. Experiments on LLaMA and Mistral models show that our method improves both perplexity and downstream task performance at high sparsity compared to uniform allocation. At 80% sparsity, perplexity is reduced by 21.8% on LLaMA-2-7B, 40.1% on LLaMA-2-13B, and 9.4% on Mistral-7B.
GLP-1 medications are increasingly used for weight management, and many people share their experiences online. These narratives show why people take the drugs and what they experience, but they remain unstructured and disconnected from formal pharmacovigilance. We therefore present AskGLP1, a system that transforms online patient narratives into a knowledge graph (KG) and answers questions about medication experiences. Through an LLM-assisted pipeline grounded in the MedDRA terminology, AskGLP1 screens nearly 8.5 million Reddit posts for medication experiences and distills 163,135 of them, with 2,506 WebMD reviews, into a KG of 385,790 drug-side-effect and drug-purpose relations. We find that GLP-1 experiences are source- and drug-specific. Online narratives share the most common gastrointestinal symptoms with FAERS, but they foreground everyday burdens such as fatigue and food aversion, whereas FAERS reports more often include injection-site reactions and selected serious diagnoses such as pancreatitis. We further find that posts report less nausea and vomiting for tirzepatide than for semaglutide, more fatigue for Wegovy than Ozempic and for Zepbound than Mounjaro. Evaluations show that the agent answers with verifiable evidence and declines requests for medical advice. Our findings reveal gaps between what FAERS captures and what people report online, and they call for patient-centered approaches to pharmacovigilance. Code, prompts, and evaluation questions are available at https://github.com/ZhaoqianXue/AskGLP1.
Debopam Sanyal, Hongjie Chen, Riya Elizabeth Kuriyan, Arth Tak, Divya Kothandaraman, Akshay Mehra, Joshua Kimball, Alexey Tumanov
Pretrained transformers spend a large share of their inference compute inside self-attention, yet the per-head query-key interaction is structurally low-rank by construction. We ask a diagnostic question: across layers and architectures, which heads actually use the full rank budget on real data, and which do not? We propose SVD-Attention, a post-hoc method that decomposes the per-head interaction matrix $M_h$ and scores every singular mode by both its spectral magnitude and its activation variance across tokens on a small calibration set. The score keeps the modes the head can amplify and that real data actually activates, and discards the rest; the resulting factorization is exact for the retained modes and needs no finetuning. Across seven pretrained models and five task types, the diagnostic finds a task-dependent pattern: redundancy sits in the early to middle layers for sentiment and image classification and for generation, and in the late layers for dense prediction and extractive question answering. It is also predictive: truncating the layers it flags costs at most $1.75$ F1, while truncating the same number of layers elsewhere in the same model costs up to $50$ F1. SVD-Attention is an analysis tool rather than a compression method: acting on it reduces inference FLOPs by $0.8$-$1.5\%$ on its own and by $8$-$11\%$ when composed with weight pruning, with $0$-$4\%$ change in task quality.
Jia Wang, Yuqing Li, Jiangnan Li, Zheng Lin, Mo Yu
Locating relevant evidence in long inputs is a core capability of Multimodal Large Language Models. Prior studies attribute this capability to retrieval heads, a sparse set of attention heads that transfer information from the context to the output. However, these analyses are confined to a limited set of modalities and the corresponding models. Moreover, how retrieval heads can be leveraged to enhance the model calls for further exploration. In this paper, we ask whether this mechanism expands together with the input modalities of the model. Taking audio-visual input as a newly added modality, we construct paired inputs in which the same answer is either spoken or shown on screen. With the evidence content fixed, retrieval heads remain sparse in the new modality and are shared to a large extent across modalities. Across 18 language, vision-language, and omni-modal checkpoints, omni-modal models largely preserve the retrieval heads of their backbones. Ablating these heads eliminates nearly all audio-visual retrieval. The heads shared across modalities make up fewer than 2\% of all heads, yet ablating them also degrades document understanding, spatial perception, and visual mathematics. Building on these findings, we turn retrieval heads into a target of optimization by training their attention with a ranking objective. This raises the page Recall@1 of Qwen3-VL-8B on MMDocIR by 6.03 points and improves document retrieval across three model families. Functional attention heads can thus serve both as a lens on multimodal models and as a target for improving them.
Vishal Sankar Ram, Jason Kushner, Manas Paldhe, Youngseo Son
Administrative tasks in the healthcare domain share linguistic commonalities, but it can be time-consuming to manually design LLM prompts for each use case. When calling health insurers, interactive voice response (IVR) systems cause delays in patient care and increase provider burnout due to complex routing and long hold times. Thus, IVR navigation models can offer significant time savings and reduce barriers to care. We propose a production-quality automated LLM pipeline which leverages a small number of human-labeled ground truth datasets to transfer specialized prompts from one task to another; specifically, we perform a cross-task transfer of our IVR navigation logic, adapting the prompt from reaching the claims department to reaching the patient benefit department. Our approach reduces prompt complexity by up to 80% and obtains 82% turn-level accuracy in real-world industrial healthcare settings, surpassing a human-designed prompt at 79%.
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 6: Industry Track) · 2026 · 正式发表
Jian Gao, Richeng Xuan Tencent, Zhaolu Kang, Dingshi Liao, Wenxin Huang, Zongmou Huang, Yangdi Xu, Bowen Qin Beijing Academy of Artificial Intelligence 等 12 位作者
The rapid advancement of large language models (LLMs) has not been matched by their evaluation in low-resource languages, especially Southeast Asian languages like Lao. To fill this gap, we introduce LaoBench, the first large-scale, high-quality, and multidimensional benchmark for assessing LLM language understanding and reasoning in Lao. LaoBench contains 17,000+ expert-curated samples across three dimensions: culturally grounded knowledge application, curriculum-aligned K12 education, and bilingual translation among Lao, Chinese, and English. It includes open-source and held-out subsets, where the held-out portion enables secure black-box evaluation via a controlled service to improve fairness and data security. We construct LaoBench with a hybrid pipeline that combines expert authoring with agent-assisted verification, ensuring linguistic accuracy, cultural relevance, and educational validity. We evaluate diverse state-of-the-art open-source and closed-source LLMs, and find that even strong multilingual models lag behind human experts, particularly in culturally grounded reasoning and translation fidelity. We hope LaoBench will catalyze research on Lao and other underrepresented Southeast Asian languages for more inclusive multilingual evaluation.
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
Jiujiang Guo, Zhengliang Guo, Kai Wang, Meiyang Wang Hebei University of Science and Technology, Dehua Peng, Shaozu Yuan, Chengyin Hu, Shuan Ai Jilin Business and Technology College 等 9 位作者
Existing knowledge graph completion research is gradually shifting from representing logical semantics of static facts to modeling evolving semantics of temporal facts, yet lacks collaborative modeling of both within a unified framework. To this end, we use concept of snapshots to decompose fact features into two complementary mechanisms: (a) intra-snapshot semantic coupling, where entities and relations exhibit snapshot-specific meanings through multidimensional interactions; (b) trans-snapshot evolutionary synergy, where relations between entities evolve across snapshots and manifest varying states. These snapshot mechanisms jointly reveal underlying logic of facts. To track them, we propose TeCES, a framework for high-fidelity modeling of evolving snapshots. TeCES embeds facts into a 2-grade geometric algebra (GA) system to capture complex semantics via multilevel structures. Temporal information is attached to each entity for mapping into snapshot spaces, while relations and timestamps are reconfigured into composite GA representations. Geometric products enable multidimensional interactions, revealing relation state changes over time. Lastly, the head entity at each snapshot combines with fused temporal-relational representation via geometric product to approximate the target tail entity at multiple levels. Overall, TeCES supports joint modeling of evolving snapshots within a lightweight GA system and significantly outperforms SOTA models on six benchmarks.
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
Xianda Zheng University of Auckland, Zijian Huang University of Auckland, Meng-Fen Chiang, Jiamou Liu The University of Auckland, Yuan Fang Singapore Management University, Michael J. Witbrock University of Auckland, Kaiqi Zhao Harbin Institute of Technology, Shenzhen
Explicit knowledge conflicts, where retrieved contexts contain contradictory information, have become increasingly prevalent as Large Language Models (LLMs) integrate diverse data sources. The core challenge lies in the complexity of entangled narratives and the heterogeneity of conflict cases, which impose excessive demands on the reasoning capabilities of standard models. To address this, we propose Knowledge Conflict Reasoning (KCR), a framework that adjudicates conflicts by structuring the underlying logic. KCR first disentangles conflicting contexts into distinct sets of reasoning traces, utilizing both textual and graph-based representations, to simplify comprehension. It then employs a Reinforcement Learning with Verifiable Rewards (RLVR) paradigm, guiding the model to internalize a reasoning process that maximizes logical consistency while actively suppressing spurious reasoning paths derived from contradictory contexts. Extensive experiments demonstrate that KCR yields substantial improvements: a KCR-enhanced 7B model surpasses the performance of baselines equipped with top-tier closed-source models such as GPT-4o and GPT-5.1.
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
Lukas Helff Technische Universität Darmstadt, Ahmad Omar, Felix Friedrich, Antonia Wüst Technische Universität Darmstadt, Hikaru Shindo SB Intuitions, Rupert Mitchell Technische Universität Darmstadt, Tim Woydt, Patrick Schramowski German Research Center for AI 等 10 位作者
We introduce SLR, an end-to-end framework for systematic evaluation and training of Large Language Models (LLMs) via Scalable Logical Reasoning. Given a user’s task specification, SLR automatically synthesizes (i) an instruction prompt for an inductive reasoning task, (ii) a validation program, executable on model outputs to provide verifiable rewards, and (iii) the latent ground-truth rule. This process is fully automated, scalable, requires no human annotations, and offers precise control over task difficulty. Using SLR, we create SLR-Bench, a benchmark comprising 19k prompts organized into 20 curriculum levels that progressively increase in relational, arithmetic, and recursive complexity. Large-scale evaluation reveals that contemporary LLMs readily produce syntactically valid rules, yet often fail at correct logical inference. Recent reasoning LLMs demonstrate improved performance but incur very high test-time computation, with costs exceeding 300 for just 1,000 prompts. Finally, curriculum learning via SLR doubles Llama-3-8B accuracy on SLR-Bench, achieving parity with Gemini-Flash-Thinking at a fraction of computational cost. Moreover, these reasoning capabilities generalize to a wide range of established benchmarks, underscoring the effectiveness of SLR for downstream reasoning.
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
Xiaoyuan Wu Carnegie Mellon University, Roshni Kaushik Fujitsu Research and Development Center Co. Ltm., Wenkai Li Fujitsu Research and Development Center Co. Ltm., Lujo Bauer Carnegie Mellon University, Koichi Onoue
Large language models (LLMs) are rapidly being adopted for tasks like draftingemails, summarizing meetings, and answering health questions. In thesesettings, users may need to share private information (e.g., contactdetails, health records). To evaluate LLMs’ ability to identify and redactsuch information, prior work introduced real-life, scenario-based benchmarks(e.g., ConfAIde, PrivacyLens) and found that LLMs can leak privateinformation in complex scenarios. However, these evaluations relied on proxy LLMs to judge the helpfulnessand privacy-preservation quality of LLM responses, rather than directlymeasuring users’ perceptions. To understand how users perceive the helpfulness and privacy-preservationquality of LLM responses to privacy-sensitive scenarios, we conducted auser study (n=94) using 90 PrivacyLens scenarios. We found that users hadlow agreement with each other when evaluating identical LLM responses. Incontrast, five proxy LLMs reached high agreement, yet each proxy LLM hadlow correlation with users’ evaluations. These results indicate that proxy LLMs cannot accurately estimate users’ wide range of perceptions of utility and privacy inprivacy-sensitive scenarios. We discuss the need for more user-centeredstudies to measure LLMs’ ability to help users while preserving privacy,and for improving alignment between LLMs and users in estimating perceivedprivacy and utility.
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
Adrián Gude Universidad de La Coruña, Roi Santos-Rios, Francis Bond Palacký University Olomouc, Dan Flickinger, Carlos Gómez-Rodríguez Universidade da Coruña, Olga Zamaraeva Universidad de La Coruña
This study contributes to a growing line of research in comparing LLM-generated texts with human-authored text, in this case, English news text. We focus in particular on the evaluation of syntactic properties through formal grammar frameworks. Our analysis compares two generations of LLMs in the context of two human-authored English news datasets from two different years. Employing the Head-Driven Phrase Structure Grammar (HPSG) formalism, we investigate the distributions of syntactic structures and lexical types of AI-generated texts and contrast them with the corresponding distributions in the human-authored New Your Times (NYT) articles. We use diversity metrics from ecology and information theory to quantify variation in grammatical constructions and lexical types. Our results show that, while English news text has changed little in the given time frame, newer, instruction-tuned LLMs display reduced syntactic and, especially, lexical diversity compared to older, non-instruction-tuned models. These findings point to future work in studying effects of instruction tuning, which, while enhancing coherence and adherence to prompts, may narrow the expressive range of model output.
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
Viktoriia A. Chekalina FusionBrain Lab, Gerasin Timofey Independent, Andrey Kuznetsov Innopolis University and FusionBrain Lab, Evgeny Frolov Higher School of Economics, Higher School of Economics and AXXX
Quantization has shown strong results in preserving model quality under compression. However, under aggressive bit-width reductions, even quantization may require additional information to prevent performance degradation. A natural source of it is second-order curvature information, captured by the Hessian. Since the Hessian of the model layers is prohibitively large, direct computation is infeasible, making structured parameterizations and approximations crucial in practice.In this work, we propose efficient Kronecker-factored approximation yielding state-of-the-art performance when integrated into existing quantization schemes. Evaluations on the LLaMA and Qwen model families show near-baseline quality at 4-bit compression and only a 5–6% degradation at 2-bit. Moreover, our method substantially accelerates the most expensive component in second-order quantization – Hessian parameterization – achieving up to a 10× speedup over prior approaches.
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
Abbas Ghaddar Huawei Technologies Canada Co., Ltd., Ivan Kobyzev Noah's Ark Lab, Huawei Technologies Canada Co., Boxing Chen Huawei Technologies Ltd., Yufei Cui
Post-training hybridization of large language models (LLMs) often replaces quadratic self-attention with sliding-window attention (SWA) to reduce KV cache usage and improve latency. Existing hybridization schemes are typically defined either at the layer level (e.g., interleaving) or at the head level via static rankings from local to global. Layer-level schemes ignore that local and global dependencies are routed through heads within the same layer, while static head-level rankings suffer from entanglement: a head’s local/global behavior can change after hybridization. We propose BOSCH, Black-box Binary Optimization for Short-context Head Selection, a training-free method that formulates the problem as a Large Neighborhood Search and decomposes it into three subproblems: (i) layer-importance detection via small-budget black-box probes, (ii) adaptive per-layer SWA-ratio assignment based on these sensitivities, and (iii) grouped head-level optimization within ratio buckets. Extensive experiments on 4 LLMs ranging from 1.7B to 30B parameters, across 4 SWA ratios, show that BOSCH consistently outperforms layer-level heuristics and 6 strong static head-level methods, with larger gains at higher SWA ratios. Under continual pretraining, BOSCH recover original long-context performance faster and to a higher level. Analysis of the selected heads reveals substantial turnover for BOSCH across different SWA ratios, underscoring the importance of performing head-level selection for each target ratio rather than relying on fixed locality rankings.
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
Alessandro Corona Mendozza, Anders Søgaard Copenhagen University
We investigate belief-like representations in decoder-only autoregressive LLMs using linear controlled probes on residual stream activations and single attention heads. Following Herrmann and Levinstein’s (2025) criteria (Accuracy, Use, Coherence, and Uniformity) we find that large models exhibit strong truth sensitivity (Accuracy), and steering activations along probe directions reliably changes downstream behavior (Use). Coherence, measured via calibrated probes and cross-dataset probing, is moderate across models, while training on diverse data yields domain-consistent truth directions (Uniformity). The results are particularly encouraging at the head level and align with some standard philosophical accounts of belief, e.g., minimal functionalism, supporting the view that LLMs can maintain propositional attitudes under such theoretical frameworks.
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
Sibo Xiao, Fu Jinyuan, Zhongle Xie Zhejiang University, Lidan Shou Zhejiang University
Accelerating the inference of large language models (LLMs) has been a critical challenge in generative AI. Speculative decoding (SD) substantially improves LLM inference efficiency. However, its utility is limited by a fundamental constraint: the draft and target models must share the same vocabulary, thus limiting the herd of available draft models and often necessitating the training of a new model from scratch. Inspired by Dynamic Time Warping (DTW), a classic algorithm for aligning time series, we propose the algorithm TokenTiming for universal speculative decoding. It operates by re-encoding the draft token sequence to get a new target token sequence, and then uses DTW to build a mapping to transfer the probability distributions for speculative sampling. Benefiting from this, our method accommodates mismatched vocabularies and works with any off-the-shelf models without retraining and modification. We conduct comprehensive experiments on various tasks, demonstrating 1.57x speedup. This work enables a universal approach for draft model selection, making SD a more versatile and practical tool for LLM acceleration.
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
Zihan Wu, Jie Xu, Yun Peng The Chinese University of Hong Kong, Chun Yong Chong Monash University Malaysia, Xiaohua Jia
Large Language Models (LLMs) struggle to automate real-world vulnerability detection due to two key limitations: the heterogeneity of vulnerability patterns undermines the effectiveness of a single unified model, and manual prompt engineering for massive weakness categories is unscalable.To address these challenges, we propose MulVul, a retrieval-augmented multi-agent framework designed for precise and broad-coverage vulnerability detection. MulVul adopts a coarse-to-fine strategy: a Router agent first predicts the top- coarse categories and then forwards the input to specialized Detector agents, which identify the exact vulnerability types. Both agents use evidence retrieved from vulnerability knowledge bases to mitigate hallucinations. Crucially, to automate the generation of specialized prompts, we design Cross-Model Prompt Evolution, a prompt optimization mechanism where a generator LLM iteratively refines candidate prompts while a distinct executor LLM validates their effectiveness. This decoupling mitigates the self-correction bias inherent in single-model optimization. Evaluated on 130 CWE types, MulVul achieves 34.79% Macro-F1, outperforming the best baseline by 41.5%. Ablation studies validate cross-model prompt evolution, which boosts performance by 51.6% over manual prompts by effectively handling diverse vulnerability patterns.
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
Zhihao Xu, Fuzhen Yang, Liang Lin, Xiting Wang Renmin University of China
Recent advancements in Large Reasoning Models (LRMs) have showcased strong performance across various reasoning tasks by leveraging System-2 thinking capabilities. However, existing studies indicate that this reasoning ability alone does not reliably transfer to the general alignment domain. Inspired by cognitive science and how humans solve tasks, we argue that LRMs must be equipped with metacognitive knowledge to fully utilize their System-2 capabilities. In this paper, we propose Priority-Aware Metacognition (PAM), which guides the model to first identify the top-level human preference (e.g., harmlessness) as a means of understanding the alignment task’s nature, and then apply other kinds of metacognitive knowledge to better monitor and regulate the model’s thinking process. We implement PAM via a two-stage pipeline: a cold-start phase that collects structured metacognitive knowledge based on Flavell’s theoretical framework, and a preference-optimization phase that further reinforces such metacognition. Extensive experiments validate the effectiveness of PAM. Under the same training pipelines, PAM consistently yields higher performance, improving general domain alignment performance by ~10 points on the helpfulness and harmless benchmarks. Code is available at https://anonymous.4open.science/r/PAM-RM-02DF.
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
Sikuan Yan Ludwig-Maximilians-Universität München and Huawei Technologies Ltd., Xiufeng Yang, Zuchao Huang, Ercong Nie Shanghai Jiaotong University, Zifeng Ding, Zonggen Li University of Hong Kong, Xiaowen Ma, Jinhe Bi 等 13 位作者
Large Language Models (LLMs) have demonstrated impressive capabilities across a wide range of NLP tasks, but they remain fundamentally stateless, constrained by limited context windows that hinder long-horizon reasoning. Recent efforts to address this limitation often augment LLMs with an external memory bank, yet most existing pipelines are static and heuristic-driven, lacking a learned mechanism for deciding what to store, update, or retrieve. We present Memory-R1, a reinforcement learning (RL) framework that equips LLMs with the ability to actively manage and utilize external memory through two specialized agents: a Memory Manager that learns structured operations, including ADD, UPDATE, DELETE, and NOOP; and an Answer Agent that pre-selects and reasons over relevant entries. Both agents are fine-tuned with outcome-driven RL (PPO and GRPO), enabling adaptive memory management with minimal supervision. With only 152 training QA pairs, Memory-R1 outperforms strong baselines and generalizes across diverse question types, three benchmarks (LoCoMo, MSC, LongMemEval), and multiple model scales (3B–14B).
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
Bowen Zeng, Feiyang Ren, Jun Zhang Qwen, Xiaoling Gu Hangzhou Dianzi University, Ke Chen Zhejiang University, Lidan Shou Zhejiang University, Huan Li Zhejiang University
Multimodal Large Language Models (MLLMs) have advanced unified reasoning over text, images, and videos, but their inference is hindered by the rapid growth of key–value (KV) caches. Each visual input expands into thousands of tokens, causing caches to scale linearly with context length and remain resident in GPU memory throughout decoding, which leads to prohibitive memory overhead and latency even on high-end GPUs. A common solution is to compress caches under a fixed allocated budget at different granularities: token-level uniformly discards less important tokens, layer-level varies retention across layers, and head-level redistributes budgets across heads. Yet these approaches stop at allocation and overlook the heterogeneous behaviors of attention heads that require distinct compression strategies. We propose HybridKV, a hybrid KV cache compression framework that integrates complementary strategies in three stages: heads are first classified into static or dynamic types using text-centric attention; then a top-down budget allocation scheme hierarchically assigns KV budgets; finally, static heads are compressed by text-prior pruning and dynamic heads by chunk-wise retrieval. Experiments on 11 multimodal benchmarks with Qwen2.5-VL-7B show that HybridKV reduces KV cache memory by up to 7.9\times and achieves 1.52\times faster decoding, with almost no performance drop or even higher relative to the full-cache MLLM.
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
LLMs have static pre-trained knowledge, leading to obsolescence and hallucinations. Knowledge Editing (KE) addresses these issues and typically requires multi-layer modifications. However, existing state-of-the-art methods, largely following the Locate–Select–Assign–Edit (LSAE) paradigm, rely on fixed-layer selection and uniform residual assignment, ignoring the heterogeneous causal efficacy of different layers. To bridge this, we propose \textbf{CAKE} (\textbf{C}ausal-Guided \textbf{A}daptive \textbf{K}nowledge \textbf{E}diting), a collaborative editing method within the more general Locate–Weight–Assign–Edit (LWAE) paradigm that: (1) selectively identifies critical layers via causal tracing scores; and (2) adaptively allocates editing burdens based on causal weights rather than uniform assumptions. We formulate residual assignment as a constrained quadratic optimization problem and derive a solution for optimal residual allocation, showing that aligning edits with causal efficacy mitigates recursive error accumulation. Furthermore, we establish a generalized weight shift error bound, under which existing paradigms emerge as special, restricted cases. Experimental results demonstrate that CAKE achieves SOTA performance with comparable overhead, validating the superiority of causal-guided adaptation.
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
Large language model (LLM) agents that invoke external tools must make sequences of interdependent decisions, yet existing uncertainty quantification (UQ) methods treat each step in isolation, ignoring how confidence evolves and compounds across a full task trajectory.We propose a framework for trajectory-level confidence analysis in the tool-use agent setting. The thesis pursues three aims: (1) estimating action-level confidence by adapting step-wise UQ to the heterogeneous think-act-observe cycles of tool-using agents; (2) aggregating the diverse action space into semantically coherent action types to enable meaningful trajectory-level analysis; and (3) discovering temporal patterns in the resulting confidence trajectories that reliably predict task success or failure.We ground the work in standard tool-use benchmarks and expect the framework to expose early warning signals for agent failure and offer interpretable diagnostic tools for understanding when and why LLM agents lose confidence, with improved calibration of multi-step agentic pipelines as a secondary benefit.
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 4: Student Research Workshop) · 2026 · 正式发表
Recent works on vision-language-action (VLA) models have made great progress in exploring action tokenizers that convert continuous control signals into discrete tokens to align with LLM/VLM training paradigms.These approaches typically train a single tokenizer over entire manipulation trajectories, which often comprise multiple distinct skills and thus pose a challenging optimization trade-off.To address this issue, we introduce MoEActok, a novel action tokenizer that employs a mixture-of-experts (MoE) quantizer to produce skill-aware discrete representations for VLA models. MoEActok utilizes a clustering-driven MoE VQ-VAE mechanism in which each expert specializes in a particular skill.The key components are: (a) an action-skill decoupling strategy that uses k-means clustering to group action chunks, aligning clusters having similar skills; (b) a skill-aware training paradigm that augments VLA models with skill-conditioned context, improving skill grounding; and (c) an adapter that projects shared encoder representations into skill-specific latent spaces for specialized quantization, and subsequently harmonize the heterogeneous quantized representations back into a unified space for coherent reconstruction by the shared decoder.We evaluate MoEActok-based VLA models against multiple prior action tokenizer baselines in the RoboTwin and Simpler-Env simulators, and further assess zero-shot transfer on three real-world tasks. Across both simulated and real-world settings, MoEActok-based VLA substantially outperforms existing discrete tokenization methods.
尚无 AI 解读，展示原文摘要
cs.CV · cs.CL
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2026 · 正式发表
AI 速览：提出一种通过记忆 token 与分段循环机制扩展 Transformer 长序列处理能力的记忆增强架构 RMT。
Transformer-based models show their effectiveness across multiple domains and tasks. The self-attention allows to combine information from all sequence elements into context-aware representations. However, global and local information has to be stored mostly in the same element-wise representations. Moreover, the length of an input sequence is limited by quadratic computational complexity of self-attention. In this work, we propose and study a memory-augmented segment-level recurrent Transformer (RMT). Memory allows to store and process local and global information as well as to pass information between segments of the long sequence with the help of recurrence. We implement a memory mechanism with no changes to Transformer model by adding special memory tokens to the input or output sequence. Then the model is trained to control both memory operations and sequence representations processing. Results of experiments show that RMT performs on par with the Transformer-XL on language modeling for smaller memory sizes and outperforms it for tasks that require longer sequence processing. We show that adding memory tokens to Tr-XL is able to improve its performance. This makes Recurrent Memory Transformer a promising architecture for applications that require learning of long-term dependencies and general purpose in memory processing, such as algorithmic tasks and reasoning.