Contrastively pretrained audio–language models (e.g., CLAP) excel at clip-level understanding but struggle with frame-level tasks.Existing extensions fail to exploit the varying granularity of real-world audio–text data, where massive clip-level textual descriptions coexist with limited frame-level annotations. This paper proposes **Fine**-grained **L**anguage-**A**udio **P**retraining (**FineLAP**), a novel training paradigm that advances both clip- and frame-level alignment in CLAP with heterogeneous data.FineLAP introduces a dual-stream sigmoid loss with a cluster-based sampling strategy to jointly learn from clip- and frame-level supervision. To capture both global semantics and local details, FineLAP uses a decoupled audio projector on top of a self-supervised encoder.To alleviate the scarcity of temporally annotated data, we present FineLAP-100k, a large-scale synthetic SED dataset constructed through a scalable curation pipeline.Extensive experiments demonstrate that FineLAP achieves SOTA performance across multiple audio understanding tasks, including retrieval, classification, sound event detection, and text-to-audio grounding. Ablation studies further show that coarse- and fine-grained alignment are mutually beneficial, providing insights for building better audio-language models (ALMs).
论文检索
输入标题、作者或关键词，从 100,903 篇学术成果中精准定位
In isolating languages such as Vietnamese, core morphological structure is encoded not by inflection but by the composition and ordering of monosyllabic morphemes, yet standard Transformer encoders largely overlook this signal. We introduce HuTieuBERT, a morpheme-aware Transformer that augments a pretrained Vietnamese encoder with two lightweight inductive biases: (i) Adaptive Boundary-Token Fusion, which integrates BMES-based morpheme boundary embeddings into token representations via a learnable gate, and (ii) a Morpheme-Aware Attention Bias, which injects a fixed structural attention matrix into early self-attention layers while minimally perturbing the pretrained attention geometry. Across a suite of Vietnamese POS, NER, and sentence-level classification benchmarks, HuTieuBERT consistently outperforms strong baselines, with the largest gains on syntactic tasks. Hyperparameter ablations show a broad regime in which structural biases improve accuracy without destabilizing representations. Applying the same design to ChineseBERT (Chinese-BERT-wwm) yields MAChineseBERT, which improves F_{1} and produces more balanced tag distributions on Chinese POS and NER, suggesting that explicit morpheme-aware attention is a portable and effective strategy for modeling isolating languages.
The gap between existing benchmarks and the dynamic nature of real-world legal practice poses a key barrier to advancing legal intelligence. To this end, we introduce J1-ENVS, the first interactive and dynamic legal environment tailored for LLM-based agents. Guided by legal experts, it comprises six representative scenarios from Chinese legal practices at three levels of environmental complexity. We further introduce J1-EVAL, a dual-metric evaluation framework, designed to assess both task performance and procedural compliance across varying levels of legal proficiency. Extensive experiments on 17 LLM agents reveal that while many models demonstrate solid legal knowledge, they struggle with procedural execution in dynamic settings. Even the SOTA model is below 60% overall performance . These findings highlight persistent challenges in achieving dynamic legal intelligence and offer valuable insights to guide future research.
While Mamba has demonstrated strong performance in language modeling, its potential as a speech self-supervised learning (SSL) model remains underexplored, with prior studies limited to isolated tasks. To address this, we explore Mamba-based HuBERT models as alternatives to Transformer-based SSL architectures. Leveraging the linear-time Selective State Space, these models enable fine-tuning on long-context ASR with significantly lower compute. Moreover, they show superior performance when fine-tuned for streaming ASR. Beyond fine-tuning, these models show competitive performance on SUPERB probing benchmarks, particularly in causal settings. Our analysis shows that they yield higher-quality quantized representations and capture speaker-related features more distinctly than Transformer-based models. These findings highlight Mamba-based SSL as a promising and complementary direction for long-sequence modeling, real-time speech modeling, and speech unit extraction. The codebase is available at https://github.com/hckuo145/Mamba-based-HuBERT.
Aligning Large Language Models (LLMs) to human preferences is pivotal for Machine Translation (MT), yet current approaches are often hindered by misleading reward signals. Our analysis reveals that prevailing Quality Estimation (QE) models exhibit a systematic blind spot towards **partial errors**—specifically partial hallucinations and omissions—often favoring superficially fluent but unfaithful translations. To address this, we propose **M^2PO** (**M**ulti-Perspective **M**ulti-Pair **P**reference **O**ptimization), a data-centric framework for preference optimization in machine translation. First, to correct the bias towards fluency, M^2PO uses a multi-perspective alignment mechanism that decouples semantic fidelity from fluency, prioritizing faithfulness via a curriculum strategy. Second, with the bias corrected, partial errors fall between perfect and severely incorrect translations, making them inefficient to learn via standard best-versus-worst comparisons. We thus introduce a multi-pair objective that leverages the full candidate list to capture these fine-grained error signals. Experiments on WMT23, WMT24, and FLORES-200 show that M^2PO enables a 9B model to outperform leading open-source baselines and achieve parity with proprietary models like GPT-4o and Gemini-2.0-Flash, demonstrating significant potential for efficient, high-fidelity LLM-based translation.
With the rapid advancement of post-training techniques for reasoning and information seeking, large language models (LLMs) can incorporate a large quantity of retrieved knowledge to solve complex tasks. However, the limited context window of LLMs obstructs scaling the amount of external knowledge input, prohibiting further improvement. Existing context window extension methods inevitably cause information loss. LLM-based multi-agent methods emerge as a new paradigm to handle massive input in a distributional manner, where we identify two core bottlenecks in existing agent orchestration designs. In this work, we develop a multi-agent framework, **ExtAgents**, to overcome the bottlenecks and enable better scalability in inference-time knowledge integration without longer-context training. Benchmarked with our enhanced multi-hop question answering test, **\boldsymbol\inftyBench+**, and other public test sets including long survey generation, ExtAgents significantly enhances the performance over existing non-training methods with the same amount of external knowledge input, regardless of whether it falls *within or exceeds the context window*. Moreover, the method maintains efficiency due to high parallelism. We believe further study in the coordination of LLM agents on increasing external knowledge input could benefit real-world applications.
Reinforcement Learning with Verifiable Rewards (RLVR) is a promising approach for enhancing agentic search. However, its performance is often hindered by reward sparsity, whereby agents receive very limited positive feedback despite incurring significant exploration costs. In this paper, we formalize this challenge as a new research problem termed **Reward Density Optimization**, which aims to improve the reward obtained per unit of exploration cost. To address this problem, we introduce InfoFlow, a systematic framework that operates along three complementary dimensions: 1) **Sub-goal Scaffolding**: which decomposes long-horizon tasks into intermediate objectives and assigns process-level rewards to provide denser learning signals; 2) **Pathfinding Hints**: which injects corrective guidance into stalled trajectories to increase the ratio of successful trials; and 3) **Dual-agent Refinement**: which employs a dual-agent architecture to offload the cognitive burden of deep exploration. We evaluate InfoFlow on several popular agentic search benchmarks, where it significantly outperforms strong baselines and enables lightweight LLMs to achieve performance comparable to that of advanced proprietary models.
Recent works have analyzed the impact of individual components of neural networks on gendered predictions, often with a focus on mitigating gender bias. However, mechanistic interpretations of gender tend to (i) focus on a very specific gender-related task, such as gendered pronoun prediction, or (ii) fail to distinguish between the production of factually gendered outputs (the correct assumption of gender given a word that carries gender as a semantic property) and gender biased outputs (based on a stereotype). To address these issues, we curate Gknow, a benchmark to assess gender knowledge and gender bias in language models across different types of gender-related predictions. Gknow allows us to identify and analyze circuits and individual neurons responsible for gendered predictions. We test the impact of neuron ablation on benchmarks for disentangling stereotypical and factual gender (DiFair and the test set of Gknow), as well as StereoSet. Results show that gender bias and factual gender are severely entangled on the level of both circuits and neurons, entailing that ablation is an unreliable debiasing method. Furthermore, we show that benchmarks for evaluating gender bias can hide the decrease in factual gender knowledge that accompanies neuron ablation. We curate Gknow as a contribution to the continuous development of robust gender bias benchmarks.
Large language models (LLMs) increasingly solve complex reasoning tasks via long chain-of-thought, but their forward-only autoregressive generation process is fragile; early token errors can cascade, which creates a clear need for self-reflection mechanisms. However, existing self-reflection either performs revisions over full drafts or learns self-correction via expensive training, both fundamentally reactive and inefficient. To address this, we propose Self-Reflective Generation at Test Time (SRGen), a lightweight test-time framework that reflects before generating at uncertain points. During token generation, SRGen utilizes dynamic entropy thresholding to identify high-uncertainty tokens. For each identified token, it trains a specific corrective vector, which fully exploits the already generated context for a self-reflective generation to correct the token probability distribution. By retrospectively analyzing the partial output, this self-reflection enables more trustworthy decisions, thereby significantly reducing the probability of errors at highly uncertain points. Evaluated on challenging mathematical reasoning benchmarks and a diverse set of LLMs, SRGen can significantly strengthen model reasoning. Moreover, our findings position SRGen as a plug-and-play method that integrates reflection into the generation process for reliable LLM reasoning, achieving consistent gains with bounded overhead and can be combined with other training-time (e.g., RLHF) and test-time (e.g., SLOT) techniques.
This work explores efficient ultra-long context modeling. We posit that an effective solution requires three fundamental properties: sparsity, random-access flexibility, and length generalization. To achieve this, we leverage Hierarchical Sparse Attention (HSA), a novel attention mechanism that satisfies all three properties. We integrate HSA into the Transformer architecture to develop HSA-UltraLong, an 8B-parameter Mixture-of-Experts (MoE) model trained on over 8 trillion tokens. We rigorously evaluate the model across tasks with both in-domain and out-of-domain context lengths to validate its capabilities. Our model demonstrates comparable performance to full-attention baselines on in-domain sequence lengths. Crucially, it achieves over 90% accuracy on most in-context retrieval tasks with contexts up to 512 times the pre-training context length. This work reports our findings and remaining issues throughout the experiments, offering insights for future research in ultra-long context modeling.
Open-weight large language models (LLMs) enable broad customization, but also increase exposure to post-release misuse, including malicious fine-tuning (MFT). To mitigate this risk, many prior defenses aim to improve the robustness of open-weight models to MFT by constraining adversarial fine-tuning dynamics in parameter space or mitigating harmful information encoded in internal representations. Nevertheless, since malicious fine-tuning can still erode safety, developing robust safeguards for open-weight models that fundamentally mitigate this risk remains an open research problem. In this paper, we characterize a safety region for open-weight LLMs and propose Safety Guidance Trigger (SGT), which guides fine-tuning toward the safety manifold to preserve alignment. SGT has two stages: (1) optimizing a safety trigger that steers the base model toward safe responses and (2) training the open-weight model to align its internal features with trigger-induced safety representations. We demonstrate that SGT substantially improves robustness against malicious fine-tuning, requiring adversaries to increase their data budget significantly to compromise safety. Our analysis shows that SGT anchors model representations to a safety region, which remains stable under malicious fine-tuning.
Feeling Rules in Language Models: Mapping Norms of Emotional Appropriateness Across Roles, Institutions, and Intensity
PDF ↗When asked explicitly, a Large language model (LLM) may validate your anger—but implicitly, it may still judge that anger as inappropriate. We call this divergence the endorsement–exposure gap, and it reveals that LLMs encode hidden norms about which emotions are acceptable in which contexts. To measure these norms systematically, we introduce Feeling Rules Atlas, a benchmark of 1,320 vignettes spanning 6 institutional settings, 12 roles, 7 emotions, and 5 intensity levels. We pair the benchmark with two probes: explicit norm judgments (APPROPRIATE/INAPPROPRIATE/DEPENDS) and implicit acceptability scored by log-likelihood contrast. Across six model families, we find large cross-model variation in sanctioning thresholds and institutional "norm signatures" not reducible to overall strictness; models that appear similarly lenient explicitly can diverge sharply in implicit judgments. These results establish normative affect—context-conditioned judgments of emotional appropriateness—as a distinct alignment axis, and motivate transparent profiling of feeling rules for emotionally sensitive deployments.
Do LLMs Capture Embodied Cognition and Cultural Variation? Cross-Linguistic Evidence from Demonstratives
PDF ↗Do large language models (LLMs) truly acquire embodied cognition and cultural conventions from text? We introduce demonstratives, fundamental spatial expressions like “this/that” in English and “这/那” in Chinese, as a novel probe for grounded knowledge. Using 6,400 responses from 320 native speakers, we establish a human baseline: English speakers reliably distinguish proximal–distal referents but struggle with perspective-taking, while Chinese speakers switch perspectives fluently but tolerate distal ambiguity. In contrast, five state-of-the-art LLMs fail to inherently understand the proximal–distal contrast and show no cultural differences, defaulting to English-centric reasoning. Our study contributes (i) demonstratives as a new lens for evaluating embodied cognition and cultural conventions, (ii) empirical evidence of cross-cultural asymmetries in human interpretation, (iii) a new perspective on the egocentric–sociocentric debate, showing both orientations coexist but vary across languages, and (iv) a call to address individual variation in future model design.
MT^{3}: A Synergistic Multi-Task RL Framework for Specializing MLLMs in Text Image Machine Translation
PDF ↗Text Image Machine Translation (TIMT)—the task of translating textual content embedded in images—is critical for applications in accessibility, cross-lingual information access, and real-world document understanding. However, TIMT remains a complex challenge due to the need for accurate optical character recognition (OCR), robust visual-text reasoning, and high-quality translation, often requiring cascading multi-stage pipelines. Recent advances in large-scale Reinforcement Learning (RL) have improved reasoning in Large Language Models (LLMs) and Multimodal LLMs (MLLMs), but their application to end-to-end TIMT is still underexplored. To bridge this gap, we introduce MT^{3}, a novel Multi-Task RL framework to specialize MLLMs into end-to-end expert TIMT models. MT^{3} adopts a synergistic multi-task optimization paradigm targeting three key sub-skills: text recognition, context-aware reasoning, and translation. It is trained using a novel multi-mixed reward mechanism that provides fine-grained feedback, fostering a controllable and transparent optimization process. Furthermore, to facilitate the evaluation of TIMT in authentic cross-cultural and real-world social media contexts, we introduced XHSPost, the first social media TIMT benchmark. Our MT^{3}-7B-Zero achieves state-of-the-art results on the latest in-domain MIT-10M benchmark, outperforming strong baselines such as Qwen2.5-VL-72B and InternVL2.5-78B by notable margins across multiple metrics. Additionally, the model shows strong generalization to out-of-distribution language pairs and datasets. In-depth analyses reveal how multi-task synergy, reinforcement learning initialization, curriculum design, and reward formulation contribute to advancing MLLM-driven TIMT.
Recent end-to-end spoken dialogue models enable natural interaction. However, as user demands become increasingly complex, models that rely solely on conversational abilities often struggle to cope. Incorporating agentic capabilities is therefore essential: by enabling tool use, these models can extend their knowledge boundaries and better solve real-world tasks. Yet, existing research has largely concentrated on core perception and generation, with comparatively limited exploration of such tool-augmented extensions. To bridge this gap, we present VoxMind, an integrated framework designed to equip end-to-end spoken dialogue models with comprehensive agentic abilities. Leveraging our curated 470-hour AgentChat dataset, we incorporate a "Think-before-Speak" mechanism, enabling the model to internalize structured reasoning as a critical prerequisite for planning and response generation. Furthermore, to mitigate latency bottlenecks caused by large-scale tool integration, we propose a Multi-Agent Dynamic Tool Management architecture. By asynchronously delegating retrieval tasks to an auxiliary agent aligned with the main model’s reasoning trajectory, this system effectively decouples inference latency from toolset size. Experimental results confirm that VoxMind achieves significant improvements in agent performance: compared with strong baselines, the task completion rate increases from 34.88% to 74.57%, outperforming Gemini-2.5-Pro on spoken agent tasks while preserving general conversational quality. The source code and associated data are publicly available at https://github.com/MM-Speech/VoxMind.
As large language models (LLMs) are increasingly applied to legal domain-specific tasks, evaluating their ability to perform legal work in real-world settings has become essential. However, existing legal benchmarks rely on simplified and highly standardized tasks, failing to capture the ambiguity, complexity, and reasoning demands of real legal practice. Moreover, prior evaluations often adopt coarse, single-dimensional metrics and do not explicitly assess fine-grained legal reasoning. To address these limitations, we introduce PLawBench, a Practical Law Benchmark designed to evaluate LLMs in realistic legal practice scenarios. Grounded in real-world legal workflows, PLawBench models the core processes of legal practitioners through three task categories: public legal consultation, practical case analysis, and legal document generation. These tasks assess a model’s ability to identify legal issues and key facts, perform structured legal reasoning, and generate legally coherent documents. PLawBench comprises 850 questions across 13 practical legal scenarios, with each question accompanied by expert-designed evaluation rubrics, resulting in approximately 12,500 rubric items for fine-grained assessment. Using an LLM-based evaluator aligned with human expert judgments, we evaluate 10 state-of-the-art LLMs. Experimental results show that none achieves strong performance on PLawBench, revealing substantial limitations in the fine-grained legal reasoning capabilities of current LLMs and highlighting important directions for future evaluation and development of legal LLMs. Data is available at: https://anonymous.4open.science/r/PLawbench-B524/.
Collaboration and information sharing empower Multi-Agent Systems (MAS) but also introduce a critical security risk known as Agent Cascading Injection (ACI). In such attacks, a compromised agent exploits inter-agent trust to propagate malicious instructions, causing cascading failures across the system. However, existing studies consider only limited attack strategies and simplified MAS settings, limiting their generalizability and comprehensive evaluation. To bridge this gap, we introduce ACIArena, a unified framework for evaluating the robustness of MAS. ACIArena offers systematic evaluation suites spanning multiple attack surfaces (i.e., external inputs, agent profiles, inter-agent messages) and attack objectives (i.e., instruction hijacking, task disruption, information exfiltration). Specifically, ACIArena establishes a unified specification that jointly supports MAS construction and attack–defense modules. It covers six widely used MAS implementations and provides a benchmark of 1,356 test cases for systematically evaluating MAS robustness. Our benchmarking results show that evaluating MAS robustness solely through topology is insufficient; robust MAS require deliberate role design and controlled interaction patterns. Moreover, defenses developed in simplified environments often fail to transfer to real-world settings; narrowly scoped defenses may even introduce new vulnerabilities. ACIArena aims to provide a solid foundation for advancing deeper exploration of MAS design principles.
Correct When Paired, Wrong When Split: Decoupling and Editing Modality-Specific Neurons in MLLMs
PDF ↗Although Knowledge Editing provides an efficient mechanism for updating the knowledge of Multimodal Large Language Models (MLLMs), we find that current paradigms still suffer from an important yet remain underexplored issue : editing decoupling failure, where entity-related knowledge can be updated when the model is triggered by multimodal inputs (text–image query pairs), however, it often reverts to outdated pre-edit facts when the paired inputs are split into unimodal ones. Our in-depth empirical analysis reveals that the entity knowledge in MLLMs is not stored as a unified representation, but is instead distributed across disentangled modality-specific pathways. As a result, updates biased toward multimodal queries fail to propagate effectively to unimodal circuits. To bridge this gap, we propose DECODE, which explicitly disentangles and localizes modality-specific neuron groups for targeted knowledge. Extensive experiments demonstrate that DECODE consistently achieves effective knowledge updates under different modality triggers, thereby mitigating editing decoupling failures.
Fine-tuning-as-a-service, while commercially successful for Large Language Model (LLM) providers, exposes models to harmful fine-tuning attacks. As a widely explored defense paradigm against such attacks, unlearning attempts to remove malicious knowledge from LLMs, thereby essentially preventing them from being used to perform malicious tasks. However, we highlight a critical flaw: the inherent general adaptability of LLMs allows them to easily bypass selective unlearning by rapidly relearning or repurposing their general capabilities for harmful tasks. To address this fundamental limitation, we propose a paradigm shift: instead of selective removal, we advocate for inducing model collapse, effectively forcing the model to ”unlearn everything”, specifically in response to updates characteristic of malicious adaptation. This collapse directly neutralizes the very general capabilities that attackers exploit, tackling the core issue unaddressed by selective unlearning. We introduce the Collapse Trap (CTRAP) as a practical mechanism to implement this concept conditionally. Embedded during alignment, CTRAP pre-configures the model’s reaction to subsequent fine-tuning dynamics. If updates during fine-tuning constitute a persistent attempt to reverse safety alignment, the pre-configured trap triggers a progressive degradation of the model’s core language modeling abilities, ultimately rendering it inert and useless for the attacker. Crucially, this collapse mechanism remains dormant during benign fine-tuning, ensuring the model’s utility and general capabilities are preserved.
Many SLT systems quietly assume that brief chunks of signing map directly to spoken-language words. That assumption breaks down because signers often create meaning on the fly using context, space, and movement. We revisit SLT and argue that it is mainly a cross-modal reasoning task, not just a straightforward video-to-text conversion. We thus introduce a reasoning-driven SLT framework that uses an ordered sequence of latent thoughts as an explicit middle layer between the video and the generated text. These latent thoughts gradually extract and organize meaning over time. On top of this, we use a plan-then-ground decoding method: the model first decides what it wants to say, and then looks back at the video to find the evidence. This separation improves coherence and faithfulness. We also built and released a new large-scale gloss-free SLT dataset with stronger context dependencies and more realistic meanings. Experiments across several benchmarks show consistent gains over existing gloss-free methods. Our code and data are available at https://github.com/fletcherjiang/SignThought.