Large language models have made rapid progress in generating machine-checked proofs for well-defined mathematical problems. Extending this progress to research mathematics requires more than solving isolated theorems. Research involves choosing questions, developing definitions and conjectures, connecting results, and combining formal proofs with expert judgment. This survey examines how LLM-driven systems can support these broader activities. We review formal-mathematics tasks, datasets, and autoformalization, and organize proof-generation methods into three overlapping areas: training strategies, inference-time search and adaptation, and agentic workflows. We then examine benchmark results, selected research projects, and platforms for collaborative mathematical work. A central distinction runs throughout the survey: checking a formal proof does not by itself establish that it captures the intended problem or makes a new mathematical contribution. We therefore distinguish theorem benchmarks, repository tasks, formalization projects, and research claims, and discuss the evidence needed to assess each. Finally, we identify five priorities for progress: better data and evaluation, representations of mathematical relationships, conjecture and concept generation, reliable tool integration, and human--AI collaboration. Together, these perspectives connect advances in proof generation to the practical demands of mathematical research.
Andrey Ignatov, Georgy Perevozchikov, Radu Timofte, Yuhong He, Yuxuan Qiu, Tianyu Hao, Ruoqi Li, Yueqi Yang 等 22 位作者
With the increasing importance of mobile camera quality in modern smartphones, growing attention has been directed toward computational photography and advanced image signal processing (ISP) algorithms aimed at enhancing various perceptual attributes of captured images. This creates a need for efficient deep learning models for this problem that are not only performant but can also be deployed on constrained mobile hardware. In this Mobile AI challenge, we address this problem and propose the participants to design fast solutions that can reconstruct RGB photos from RAW Bayer camera data while running on popular smartphone GPUs from Qualcomm and Arm. For this, the participants were provided with a large-scale Fujifilm UltraISP dataset consisting of RAW-RGB image pairs captured with the Sony IMX586 Quad Bayer mobile sensor and a professional 102MP medium format FujiFilm GFX100 camera. The runtime of all models was evaluated on the Snapdragon 8 Elite Gen 5 and Dimensity 9500 chipsets with the Adreno and Mali GPUs, respectively. Top solutions proposed in this challenge are able to reconstruct high-quality 5MP RGB image from raw data on both platforms under 100 milliseconds. A comprehensive description of the models developed in the challenge is provided in this paper.
Frontier LLM agents can turn known vulnerabilities into working exploits, but are costly and often access-restricted due to their potential for malicious use. Open-weight models offer a more accessible alternative, but struggle to maintain progress across long exploitation chains. They also often fail to apply relevant security knowledge at the appropriate stage and in the right context to identify and evaluate promising attack directions. To address these limitations, we propose DoGNAVY-Exploit, a multi-agent harness built around two complementary designs: stage-aware decomposition with progressive expert knowledge disclosure, and hierarchical orchestration with parallel exploration. Specifically, DoGNAVY-Exploit uses configurable milestones to recast long-horizon exploitation as short, verifiable capability transitions and provides only the knowledge relevant to each stage. Within each milestone, a hierarchical architecture separates strategic planning from parallel execution: a planner proposes diverse attack routes, executors test them concurrently, and a controller accepts progress only when supported by runtime evidence. A shared structured memory consolidates verified facts, artifacts, replay steps, and failed routes across branches, allowing agents to build on collective progress and carry validated knowledge into subsequent milestones. In controlled benchmark environments, DoGNAVY-Exploit with GLM-5.2 achieves 97.5% Pass@1 on Cybench, and 97.5% zero-day and 100% one-day Pass@1 on CVE-Bench. On ExploitGym, it solves 79 tasks with GLM-5.2 and 157 with GLM-5.3, while the corresponding public Claude Code baselines solve 39 and 130 tasks, respectively. Ablation results further demonstrate that the stage-aware orchestration, expert knowledge, parallel exploration, and memory designs can substantially improve the exploitation capabilities of open-weight agents.
Remote sensing image fusion and semantic interpretation are usually optimized separately, so reconstruction quality need not preserve downstream retrieval utility. We propose LANCE, which uses structured language as an explicit, indexable semantic state. A frozen vision-language model maps the input to a closed-vocabulary text record; a language-conditioned generator fuses the imagery; and a dense verifier with a differentiable continuous summarizer re-describes the output. Text-text consistency and anti-collapse contrastive text-mask losses with negative queries close the loop without dense annotations or VLM fine-tuning. A variational EM formulation yields a per-step contraction result under regularity assumptions, distinguishing semantic faithfulness from verifier overfitting. Across seven test sets on GF2, QuickBird, WorldView-III, and WorldView-II, LANCE improves reduced-scale reference metrics by up to 11.7% (median 2.1%) over the strongest learning baseline. Independent evaluation using a held-out VLM and a supervised aerial detector yields gains of +6.4 mAP and +0.072 open-vocabulary F1 over the closest competitor. Text perturbations, cross-verifier transfer, and pseudo-RGB ablations examine the mechanisms underlying these gains. The framework connects image fusion to verifiable semantic representations for multimodal retrieval and knowledge management.
Yawen Shao, Jie Xiao, Kai Zhu, Yu Liu, Hongchen Luo, Xueyang Fu, Yang Cao, Wei Zhai 等 9 位作者
Reinforcement learning (RL) has shown strong potential for aligning generative models with human intent, but adapting RL to masked generative models (MGMs) remains largely underexplored. Existing approaches that directly adapt Group Relative Policy Optimization (GRPO) to MGMs leave a foundational conflict unaddressed: MGMs' exploitative decoding nature is inherently at odds with the exploratory diversity that GRPO presupposes. In this work, we reveal a structural decoding bias termed the visual exploration trap: confidence-based sampling trades the exploration of diverse subject realizations for the greedy resolution of low-uncertainty background regions, causing premature collapse of the generative exploration space and starving GRPO of the rollout diversity it depends on. To this end, we propose MaskSense, a novel RL framework for MGMs that confronts the visual exploration trap at both the sampling and optimization stages. Specifically, we introduce a semantic anchored routing sampling mechanism that leverages semantic priors to preserve high-entropy subject token exploration while amplifying intra-group reward variance through exploratory-exploitative routing to yield more discriminative advantage estimates. Furthermore, we design a global entropy transition anchoring strategy to identify the most consequential decoding steps and concentrate gradient updates on them. Extensive experiments on multiple text-to-image benchmarks demonstrate that MaskSense substantially improves upon the base model, with GenEval accuracy improving from 54% to 86% and HPS from 28.89 to 36.39, achieving state-of-the-art performance.
Sichun Luo, Yi Huang, Mukai Li, Haochen Luo, Guanzhi Deng, Haibo Wang, Jian Xu, Zefa Hu 等 10 位作者
Large language model (LLM)-driven agent scaffold evolution improves agent workflows through iterative candidate generation and evaluation. Evaluation both guides subsequent search and determines the final scaffold, but evaluating every candidate on the full task set incurs substantial execution costs. We investigate how evaluation budgets can be allocated across these two roles. Our experiments show that sparse feedback can sustain productive evolution, while additional targeted evaluation improves final selection. Replay analyses further show that shared informative items and a common scoring reference support comparisons among closely matched candidates. Motivated by these findings, we propose \method, an evaluation allocation framework that combines sparse search-time feedback with targeted final confirmation. A short fully evaluated warm-up establishes a shared informative core and calibrates a response model whose item parameters remain fixed during search. After search, \method refits the model and applies paired successive halving, reusing accumulated observations and executing only missing evaluations. Across four benchmarks and two evolution frameworks, \method achieves higher mean performance than full evaluation in most settings while using approximately half as many evaluation cells.
Large language models (LLM) have achieved remarkable performance across a wide range of tasks. However, their substantial parameter sizes pose significant challenges for deployment on edge devices with limited computational and memory resources. Low-rank compression is a promising approach to address this issue, as it reduces both computational and memory costs, making LLM more suitable for resource-constrained environments. Nonetheless, naïve low-rank compression methods require a significant reduction in the retained rank to achieve meaningful memory and computation savings. For a low-rank model, the ranks need to be reduced by more than half to yield efficiency gains. Such aggressive truncation, however, typically results in substantial performance degradation. To address this trade-off, we propose SkipCat, a novel low-rank compression framework that enables the use of higher ranks while achieving the same compression rates. First, we introduce an intra-layer shared low-rank projection method, where multiple matrices that share the same input use a common projection. This reduces redundancy and improves compression efficiency. Second, we propose a block skipping technique that omits computations and memory transfers for selected sub-blocks within the low-rank decomposition. These two techniques jointly enable our compressed model to retain more effective ranks under the same compression budget. Experimental results show that, without any additional fine-tuning, our method outperforms previous low-rank compression approaches by 7% accuracy improvement on zero-shot tasks under the same compression rate. These results highlight the effectiveness of our rank-maximized compression strategy in preserving model performance under tight resource constraints.
Yiqing Wang, Maria A. Woodward, Ziyun Yang, N. Venkatesh Prajna, Chunming He, Leslie M. Niziol, Mercy Pawar, Ming-Chen Lu 等 14 位作者
To develop an open-source multimodal model for bacterial-versus-fungal microbial keratitis classification and assess shortcuts from geographic pooling. Participants: 1,645 US and Indian patients contributed 17,158 slit-lamp images under blue-light, sclerotic-scatter, and white-light illumination. Methods: The model combined cross-modality contrastive learning, modality-specific fine-tuning, and fusion with 7 clinical variables. Diagnoses used smear, culture, and polymerase chain reaction results, with expert review of culture-negative cases. The model and 5 comparators used 5 patient-disjoint outer folds stratified by country and organism, with inner validation. Secondary analyses examined within-country performance, a country-only reference, ablations, and imbalance mitigation. Code: https://github.com/yqwang01/TPMKA. Main Outcome Measures: Patient-level macro-F1, class-specific F1, accuracy, and area under the receiver operating characteristic curve (AUROC), overall and by country. Results: The model had the highest overall macro-F1 (83.06%; 95% CI, 81.58-84.56) and accuracy (84.50%; 95% CI, 83.16-85.84) among matched comparators, with 89.9% AUROC (95% CI, 88.7-91.1). Paired differences were not consistently significant. AUROC was 83.9% in India and 67.0% in the US; bacterial F1 in India was 47.78%, and fungal F1 in the US was 40.97%. A country-only reference achieved 81.88% accuracy and 78.2% AUROC overall despite 50.0% AUROC within each country. Resampling and group distributionally robust optimization improved some underrepresented groups without removing the geographic pattern. Conclusions: The model compared favorably with matched methods, but geographic prevalence differences made pooled performance more reassuring than within-country discrimination. Independent geographic validation is needed.
Zehao Zhao, Shangjian Yin, Rui Liu, Chongyang Bai, Mark Ibrahim, Kavosh Asadi, Yuchen Lu, Hang Cui 等 9 位作者
Test-time updating lets a language model change persistent state after training---such as model weights, adapter parameters, or a memory module---before or while answering queries. Existing evaluations often compare systems using update count or frequency and performance measured immediately afterward. These summaries omit what is updated, where, when, and in what order. The same total can therefore hide an important difference: one system may answer after one update while another answers after two. We introduce the Control Map, an evaluation framework that records the committed updates preceding each scored answer and separately evaluates behavioral changes, immediate learning (acquisition), later recall (retention), preservation of unrelated behavior (locality), and performance in new settings. We prove when counts alone determine final state and show, under stated assumptions, how differences between pre-answer histories limit differences in average scores. Using ControlMap, our experiments reveal three insights. First, update scheduling matters: at a 760M-parameter checkpoint, 24 schedules with the same update count yield 19.7%--34.2% accuracy. Second, learning an edit does not guarantee that it lasts or remains contained: all six selected CounterFact LoRA settings learn the new information, while ROME retains edits at all 13 operating points but fails locality at each. Third, no editor performs best consistently across models, benchmarks, and tests. Evaluations should therefore report update histories and assess acquisition, retention, and locality separately.
General-purpose embodied intelligence requires robots to coordinate diverse skills over long horizons, track task progress, and recover from failures. Specialist policies alone do not ensure task-level reliability, while powerful multimodal agents rely on costly reasoning and external experience that compact models struggle to reproduce. We introduce OmniBrain-0, a lightweight unified embodied agent that coordinates manipulation and navigation through a shared high-level decision model, integrating task understanding, planning, progress tracking, and feedback-driven recovery. Our progressive training framework combines embodied pretraining, domain-specialized agent distillation, and multi-teacher on-policy distillation (MOPD). It distills powerful teachers' interaction experience into manipulation and navigation experts, then consolidates their decision capabilities into OmniBrain-0. Unification therefore occurs within the learned decision model rather than only at the tool interface. We further introduce Rubric-Guided Recursive Self-Improvement (RSI), a data flywheel in which a Meta-Agent turns execution evidence into refined rubrics, step-level reminders, and verified trajectories for domain-expert updates before their reconsolidation into OmniBrain-0 through MOPD. We evaluate OmniBrain-0 on LIBERO-Pro, LIBERO-Plus, RoboCasa365, RoboTwin 2.0, and R2R-CE Val-Unseen, as well as multiple real-world manipulation tasks. OmniBrain-0 combines strong cross-domain task performance with substantially lower inference and execution overhead than its closed-source teacher, demonstrating efficient transfer and consolidation of embodied capabilities into a compact unified agent.
Training tool-calling agents requires large-scale trajectory data with verifiable labels, yet existing approaches either synthesize environments that diverge from real API behavior or generate tasks without ground-truth outcomes for verification. We present FireFly, a pipeline for generating verified tool-call data from real-world MCP servers. Our key insight is to invert the standard synthesis pipeline: rather than generating tasks and hoping they are solvable, we first let a strong LLM explore real APIs along graph-guided DAG structures, then synthesize tasks backward from observed outcomes, guaranteeing label correctness by construction. To handle the scale of real-world tool spaces (${\sim}$1,000 tools), we build a pairwise tool graph and sample sub-DAGs to focus exploration on semantically coherent workflows. To address environment drift in live APIs, we construct a retrieval-augmented simulator that caches all exploration results and replays them during training and evaluation, enabling fully offline and reproducible RL. Applying this pipeline yields 5,144 verified tasks spanning 240 servers and 993 tools. A 4B-parameter model trained with GRPO on FireFly matches Claude Sonnet 4.6 on our held-out test set and shows improvements on multiple tool-calling benchmarks including Tau2-Bench, MCPMark, and MCP-Atlas.
Multimodal Emotion Recognition (MER) systems often suffer from missing modalities in real-world scenarios. Existing methods usually generate, align, or distill missing modalities as a whole, overlooking the heterogeneous nature of the information carried by each modality. Such holistic treatment mixes inferable shared semantics with uncertain modality-specific details, yielding unstable representations and degrading robustness. To address this issue, we propose the Primitive Memory Distillation (PriMD) framework. Unlike existing methods, PriMD takes an intra-modal perspective and focuses on how different types of information within a modality differ in recoverability. PriMD first disentangles cross-modal shared semantics from modality-specific representations, and then discretizes the latter into learnable semantic primitives to construct modality-specific memory banks. When modalities are missing, PriMD is a teacher-student framework that the student model uses the shared semantics of available modalities as queries to dynamically retrieve primitives. It compensates for missing modality-specific information within a constrained memory space and aligns with the teacher model. Extensive experiments on IEMOCAP, CMU-MOSI, and CMU-MOSEI demonstrate that PriMD achieves state-of-the-art performance and consistently stronger robustness across a wide range of missing-modality settings, while mitigating the instability caused by holistic feature inference. Our code and project website are available at https://www.jiaqizhang-sengoku.com/Code/PriMD and https://www.jiaqizhang-sengoku.com/PriMD/ , respectively.
Haochen Luo, Yifan LI, Ho Tin Ko City University of Hong Kong, An Binh Minh, Junjie Xu, Tang Pok Hin City University of Hong Kong, Wang Chak Wong, Gao Yuan 等 11 位作者
Large language model (LLM) agents are increasingly applied to financial decision-making tasks that require interaction with external tools, including market data retrieval, news analysis, and trade execution. However, existing trading systems rely on fragmented and task-specific APIs, resulting in inconsistent schemas, complex integration, and limited reproducibility. We present QFinZero, a unified trading environment for LLM-based financial agents. QFinZero standardizes three core capabilities: (i) multi-frequency market and derivatives data access, (ii) structured news and event retrieval, and (iii) stateful brokerage simulation with explicit order lifecycle management. All tools adopt consistent JSON schemas and time-aligned interfaces, enabling agents to acquire information and execute trades within a coherent framework. By abstracting financial interaction into composable, agent-invokable primitives, QFinZero reduces engineering overhead and supports reproducible evaluation through comprehensive logging and deterministic replay. We argue that such a standardized trading environment is essential for scalable research on LLM-based financial agents.
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
Huilin Deng, Hongchen Luo Northeastern University, Yue Zhu, Long Li, Zhuoyue Chen Alibaba, Xinghao Zhao, Ming LI Hithink RoyalFlush Information Network Co.,Ltd., Chuyang Zhao Baidu 等 12 位作者
Despite recent advances in Reinforcement learning with verifiable rewards (RLVR) for large language model (LLM) reasoning, most methods suffer from exploration collapse, as the semantic homogeneity of random rollouts traps models in narrow, over-optimized behaviors. Existing methods leverage policy entropy to encourage exploration, but face inherent limitations: global entropy regularization is susceptible to reward hacking, inducing meaningless verbosity, whereas local token-selective updates struggle with the strong inductive bias of pre-trained models. To this end, we propose Latent Policy Optimization via Iterative Information Bottleneck ( I²B-LPO), which shifts from statistical perturbation of token distributions to topological branching of reasoning trajectories. I²BLPO triggers latent branching at high-entropy states to diversify reasoning trajectories and applies the Information Bottleneck as a trajectory filter and self-reward to ensure concise and informative exploration. Empirical results on four mathematical benchmarks demonstrate that I²B-LPO achieves state-of-the-art performance, with margins of up to 5.3% in accuracy and 7.4% in diversity metrics. Code is available at https://github.com/denghuilin-cyber/IIB-LPO.
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
The powerful generative capacity of Large Language Models (LLMs) has instigated a paradigm shift in recommendation. However, existing generative models (e.g., OneRec) operate as implicit predictors, critically lacking the capacity for explicit and controllable reasoning—a key advantage of LLMs. To bridge this gap, we propose OneRec-Think, a unified framework that seamlessly integrates dialogue, reasoning, and personalized recommendation. OneRec-Think incorporates: (1) Itemic Alignment: cross-modal Item-Textual Alignment for semantic grounding; (2) Reasoning Activation: Reasoning Scaffolding to activate LLM reasoning within the recommendation context; and (3) Reasoning Enhancement, where we design a recommendation-specific reward function that accounts for the multi-validity nature of user preferences. Experiments across public benchmarks show state-of-the-art performance. Moreover, our proposed “Think-Ahead” architecture enables effective industrial deployment, achieving a 0.159% gain in APP Stay Time and validating the practical efficacy of the model’s explicit reasoning capability.
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
Yu Fu University of California, Riverside, Chen Luo Amazon, Josef Valvoda Google, Xin Zhang, Xuejing Lei Amazon, Xiao Pan, Hui Liu Amazon, Yue Dong University of California, Riverside
Key-Value (KV) cache compression techniques have improved the efficiency of long-context summarization in Large Language Models (LLMs), but their impact on model hallucination remains underexplored. In this paper, we present the first systematic study of how KV cache compression affects hallucination in long-context summarization, demonstrating that aggressive compression can increase hallucination scores by up to 3.36× compared to the baseline. To mitigate this issue, we propose HalluKV, a decoding-phase strategy that selectively removes generated KV pairs from retrieval heads responsible for retrieving critical information from source context, thereby anchoring their attention on the preserved source information. Our approach maintains computational efficiency while significantly reducing hallucination across multiple models and datasets, achieving up to 5.48 average point reductions on Llama-3-8B-Instruct, enabling more trustworthy long-context summarization.
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
Tool-calling agents are increasingly deployed in real-world customer-facing workflows. Yet most studies on tool-calling agents focus on idealized settings with general, fixed, and well-specified tasks.In real-world applications, user requests are often (1) ambiguous, (2) changing over time, or (3) infeasible due to policy constraints, and training and evaluation data that cover these diverse, complex interaction patterns remain under-represented.To bridge the gap, we present Trajectory2Task a verifiable data generation pipeline for studying tool use at scale under three realistic user scenarios: ambiguous intent, changing intent, and infeasible intents.The pipeline first conducts multi-turn exploration to produce valid tool-call trajectories. It then converts these trajectories into user-facing tasks with controlled intent adaptations. This process yields verifiable task that support closed-loop evaluation and training. We benchmark several state-of-the-art LLMs on the generated complex user scenario tasks and observe frequent failures.Finally, using successful trajectories obtained from task rollouts, we fine-tune lightweight LLMs and find consistent improvements across all three conditions, along with better generalization to unseen tool-use domains, indicating stronger tool-calling ability.
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
Haobo Xu, Sirui Chen University of Illinois at Urbana-Champaign, Ruizhong Qiu University of Illinois Urbana-Champaign, Yuchen Yan Amazon, Chen Luo Amazon, Monica Xiao Cheng Amazon, Jingrui He University of Illinois at Urbana-Champaign, Hanghang Tong University of Illinois at Urbana-Champaign
Reinforcement Learning with Verifiable Rewards (RLVR) has significantly advanced the reasoning capabilities of Large Language Models (LLMs). However, methods such as GRPO and DAPO suffer from substantial computational cost, since they rely on sampling many rollouts for each prompt. Moreover, in RLVR the relative advantage is often sparse: many samples become nearly all-correct or all-incorrect, yielding low within-group reward variance and thus weak learning signals. In this paper, we introduce ARRoL (Accelerating RLVR via online RoLlout Pruning), an online rollout pruning method that prunes rollouts during generation while explicitly steering the surviving ones more correctness-balanced to enhance learning signals. Specifically, ARRoL trains a lightweight quality head on-the-fly to predict the success probability of partial rollouts and uses it to make early pruning decisions. The learned quality head can further weigh candidates to improve inference accuracy during test-time voting. To improve efficiency, we present a system design that prunes rollouts inside the inference engine and re-batches the remaining ones for log-probability computation and policy updates. Across GRPO and DAPO on Qwen-3 and LLaMA-3.2 models (1B-8B), ARRoL improves average accuracy by +2.30 to +2.99 while achieving up to 1.7× training speedup, and yielding up to +8.33 additional gains in average accuracy in test-time voting.
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
Bo Zhang Xi’an Research Institute of High Technology, Jiawei Zhang, Cong Gao, Bingxu Han, Minghao Hu, Jun Zhang National University of Defense Technology, Yunbo Cao Information Center, Academy of Military Science, Zhunchen Luo 等 11 位作者
Although large reasoning models (LRMs) exhibit exceptional mathematical reasoning capabilities on clean inputs, their reasoning accuracy drops substantially in the presence of character-level noise such as typographical errors. Critically, their confidence estimates fail to reflect the corresponding decline in reasoning accuracy. While confidence calibration offers a principled solution, existing methods predominantly target clean inputs, leaving noisy scenarios largely unexplored. To address this gap, we propose DisCal (Distribution-aware Calibration), a confidence calibration framework for character-level noisy inputs. DisCal extracts uncertainty signals from both the empirical answer distribution and the model’s predictive distribution, and integrates them via a learned calibrator to produce well-calibrated confidence. Experiments across multiple mathematical reasoning benchmarks demonstrate that DisCal consistently outperforms existing calibration methods under noisy inputs, reducing Expected Calibration Error (ECE) by up to 39.21% and improving Area Under the Receiver Operating Characteristic Curve (AUROC) by up to 31.44%.
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
Zechen Sun, Yuyang Sun, Zecheng Tang Soochow University, Juntao Li, Wenpeng Hu Academy of Military Science, Wenliang Chen Soochow University, China, Zhunchen Luo, Guotong Geng AMS 等 9 位作者
Generating coherent and controllable long-form content remains a persistent challenge for Large Language Models (LLMs). While reasoning-enhanced models have demonstrated success in logic-intensive domains, our evaluation reveals that they suffer from a severe length collapse in open-ended writing, where performance degrades sharply as target lengths exceed 2,000 words. We attribute this failure to the limitation of static hierarchical planning, which struggles to provide dynamic guidance over extended contexts. To bridge this gap, we introduce the Interleaved Structural Chain-of-Thought (IS-CoT) framework. Unlike external agentic workflows, IS-CoT embeds a dynamic Plan-Write-Reflect cycle into the generation process, enabling continuous strategy adaptation and global alignment without additional assistance. Based on this framework, we construct a high-quality dataset of interleaved reasoning traces via a multi-teacher pipeline and train IS-Writer-8B. Experiments demonstrate that IS-Writer-8B achieves state-of-the-art performance on challenging long-form benchmarks (e.g., +3.08 vs. DeepSeek-V3.2 on LongBench-Write), exhibiting robust length compliance and coherence competitive with significantly larger proprietary models.
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