Multi-modal knowledge graph reasoning (MMKGR) aims to predict the missing links by exploiting both graph structure information and multi-modal entity contents. Most existing works are designed for a transductive setting, which learns dataset-specific embeddings and struggles to generalize to new KGs. Recent knowledge graph foundation models (KGFMs) improve cross-KG transfer, but they mainly exploit structural patterns and ignore rich multi-modal signals. We address these gaps by proposing a token-based foundation model (TOFU) for MMKGR, which exhibits strong generalization across different MMKGs. TOFU discretizes structural, visual, and textual information into modality-specific tokens. TOFU then employs a hierarchical fusion architecture with mixture-of-message mechanisms, aiming to process these tokens and obtain transferable features for MMKGR. Experimental results on 17 transductive, inductive, and fully-inductive MMKGs show that TOFU generally outperforms strong KGFM and MMKGR baselines, delivering strong performance on unseen MMKGs.
Medical multimodal reasoning models are typically trained on teacher-generated rationales, discarding the valid reasoning in students' own failed trajectories, while general-domain correction methods offer little guidance for errors in image reading, anatomical context, or differential diagnosis. We propose MedEvo, which uses medical knowledge, namely a tag taxonomy and error-type diagnosis, to navigate both trajectory repair and data selection. For each failed rollout, a gold-blind auditor locates the first consequential error and provides a minimal, answer-free hint; the same student retains the valid prefix and regenerates the erroneous suffix. After each round, MedEvo identifies capability gaps across medical tags on a held-out feedback set and prioritizes related questions for new rollouts and repairs. With supervised fine-tuning (SFT) alone, MedEvo raises Qwen3.5-9B mean accuracy across seven benchmarks from 65.57% to 69.12% (47.40% to 54.50% on the multimodal track of MedXpertQA (MedXpertQA-MM)), surpassing teacher-distillation SFT (66.93%). It also raises Qwen3.5-35B and Qwen3.5-397B mean accuracy from 67.91% to 73.04% and from 71.83% to 74.95%, respectively. Our analyses show that failed trajectories contain reasoning worth preserving, that repair success does not imply learning value, and that independent feedback is key to sustained improvement.
Electrocardiogram (ECG) models need supervision that targets evolving weaknesses and remains faithful to waveform evidence. We introduce ECG-RSI, a recursive self-improvement framework combining verified waveform synthesis with feedback-guided intervention search. Signal remeasurement verifies waveform edits; construction errors guide generator repair, while learner errors motivate interventions compared on a separate selection set. Across seven benchmarks, three rounds raise Qwen3.5-35B-A3B's mean question-answering accuracy from 34.57% to 56.04% and mean classification F1 from 24.27 to 34.03. Across the retained trajectory, the preferred supervision shifts from direct reinforcement to confusion-focused contrasts. Continued feedback correction yields diminishing external gains. These results motivate separating error diagnosis from intervention selection and checking whether gains on the feedback set transfer to independent benchmarks.
Long-video generators increasingly retrieve past visual observations to prevent identity drift. The intuition is compelling: more access to history should make future video more consistent. But an image can be faithful to the past and wrong for the present. Does visual memory always improve the generated story? We investigate this question through VMem-Bench, a two-track diagnostic study. Track-A measures which references are selected from 92 real films (16,548 segments), before generation and without generator confounds. Track-B compares four retrieval policies with no-memory and gold-informed last-seen controls on 30 causally generated stories (2,824 segments) using a shared generator. Retrieval increases entity recall from 0.559 to 0.628-0.693 and improves similarity to past appearances, but decreases precision from 0.977 to 0.915-0.938 and forbidden-entity avoidance from 0.968 to 0.845-0.898. It can also reinforce obsolete appearances after the story changes them. Conversely, high long-gap recall can arise without memory when the prompt itself supplies the appearance; an appearance-stripping analysis exposes this confound. These results distinguish historical fidelity from current-story correctness and show why retrieval relevance or final-video consistency alone does not establish memory utility.
Hao Zhang, Lyu Mengsi, Zhuo Chen Beijing Academy of Artificial Intelligence, Yulong Ao Beijing Academy of Artificial Intelligence, Yonghua Lin
Large Language Models (LLMs) demonstrate exceptional capabilities across various tasks, but their deployment is constrained by high computational and memory costs. Model pruning provides an effective means to alleviate these demands. However, existing methods often ignore the characteristics of prefill-decode (PD) disaggregation in practice. In this paper, we propose a pruning method that is highly integrated with PD disaggregation, enabling more precise pruning of blocks. Our approach constructs pruning and distillation sets to perform iterative block removal, obtaining better pruning solutions. Moreover, we analyze the pruning sensitivity of the prefill and decode stages and identify removable blocks specific to each stage, making it well suited for PD disaggregation deployment. Extensive experiments demonstrate our approach consistently achieves strong performance in both PD disaggregation and PD unified (non-PD disaggregation) settings, and can also be extended to other non-block pruning methods. Under the same settings, our method achieves improved performance and faster inference.
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 models (LLMs) have achieved remarkable success across a wide range of natural language processing tasks, yet their performance remains heavily biased toward high-resource languages. Tibetan, despite its cultural significance and large speaker population, is still substantially underrepresented. In this work, we present a comprehensive pipeline for advancing Tibetan language modeling through large-scale data curation and continual pre-training. We construct a 72 GB high-quality Tibetan corpus, the largest to date, and adapt Qwen2.5-7B through balanced multilingual continual pre-training with Tibetan, Chinese, and English, followed by multilingual instruction tuning. To further scale capacity efficiently, we extend the dense model to a 50B-A10B Mixture-of-Experts architecture. Due to the absence of standardized Tibetan benchmarks, we build multiple evaluation datasets via high-quality translation and human verification. Experimental results show that both dense and MoE models consistently outperform existing open-source and Tibetan-focused models of similar scale across diverse tasks. Our work advances Tibetan-centric LLM research and provides transferable insights for extending LLMs to other low-resource languages. We will release the model weights, evaluation benchmarks, and detailed data processing documentation in the follow-up.
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
Yichi Zhang Zhejiang University, Zhuo Chen Zhejiang University, Lingbing Guo Nanjing University, Jun Xu, Mengshu Sun, Zhizhen Liu, Lei Liang, Wen Zhang Zhejiang University 等 9 位作者
Question answering (QA) with reference texts is a classic application scenario for large language models (LLMs), where high standards for the credibility and traceability of generated answers are crucial. Many existing approaches focus on generating multi-level citations linked to specific references within the answer, making it verifiable and trustworthy. However, they often overlook key challenges such as citation granularity, the awareness of unknown information, and the adoption of effective training strategies. In this paper, we introduce Knowledge-informed Citation (KFC), which addresses these issues through a novel data construction pipeline, a new benchmark, and an innovative training strategy. With approximately 42K samples spanning 19 distinct domains, KFC includes both traditional citations referencing known entity-level information and specialized citations referring to unknown knowledge in the given question. This structure provides a more granular approach to citations, guiding the model to recognize and explicitly indicate unknown information, thus enhancing the quality and credibility of the response. Additionally, we propose a self-correction paradigm, Self-KFC, designed to fine-tune LLMs by refining poorly cited answers into more accurate ones, making it particularly suitable for citation-dependent scenarios. We present comprehensive experimental results to demonstrate the effectiveness and generalization of Self-KFC on the KFC benchmark.
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
Mengzhuo Chen, Junjie Wang Institute of Software, Chinese Academy of Sciences, Fangwen Mu, Yawen Wang, Zhe Liu Institute of Software Chinese Academy of Sciences, Huanxiang Feng, Qing Wang
Failure attribution, i.e., identifying the responsible agent and decisive step of a failure, is particularly challenging in LLM-based multi-agent systems (MAS) due to their natural-language reasoning, nondeterministic outputs, and intricate interaction dynamics. A reliable benchmark is therefore essential to guide and evaluate attribution techniques. Yet existing benchmarks rely on partially observable traces that capture only agent outputs, omitting the inputs and context that developers actually use when debugging. We argue that attribution should be studied under full execution observability, aligning with real-world developer-facing scenarios where complete traces, rather than only outputs, are accessible for diagnosis. To this end, we introduce TraceElephant, a benchmark designed for failure attribution with full execution traces and reproducible environments. We then systematically evaluate failure attribution techniques across various configurations. Specifically, full traces improve attribution accuracy by up to 76.5% over a partial-observation counterpart, confirming that missing inputs obscure many failure causes. TraceElephant provides a foundation for follow-up failure attribution research, promoting evaluation practices that reflect real-world debugging and supporting the development of more transparent MASs.
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
Zhuo Chen, Minghao Li, Xiaoqian Ma Xiaohongshu, Siqi Fan University of Electronic Science and Technology of China, Xiusheng Huang, Zhang Liujie Xiaohongshu, Weihang Chen
Singular Value Decomposition (SVD) enables hardware-agnostic LLM compression via low-rank approximation, yet optimal rank allocation remains a bottleneck. Existing methods predominantly derive layer importance from performance-oriented proxies. Yet, these metrics fail to distinguish between representational importance and structural compressibility, consequently obscuring the fine-grained influence of spectral distribution shape. We demonstrate this disconnect through spectral analysis, revealing that layers with similar information capacity can exhibit markedly different singular value decay behaviors, corresponding to varying degrees of redundancy in the spectral tail. This imperfect coupling implies that allocation strategies driven solely by importance leave significant compression opportunities underexploited. To address this gap, we propose HiSVD, a hierarchical rank allocation framework with two stages: (1) Capacity-Anchored Baseline Allocation, which preserves representational stability by aligning rank budgets with information capacity; and (2) Redundancy-Aware Refinement, which modulates this baseline using tail redundancy to penalize structural excess. Experiments on LLMs demonstrate that HiSVD achieves superior compression efficiency, significantly outperforming state-of-the-art baselines by effectively exploiting this spectral heterogeneity.
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