Large language models (LLMs) excel at function calling, but inference scaling has been explored mainly for unstructured generation. We propose an inference-scaling framework for structured outputs that combines fine-grained beam search with ToolPRM, a process reward model scoring each intra-call decision (function name and argument filling). We build the first fine-grained intra-call supervision dataset via function masking, rollout collection, and step-level annotation. ToolPRM outperforms outcome and coarse-grained reward models in predictive accuracy and yields consistent test-time gains on multiple function-calling benchmarks. We further show that structured generation follows “explore more but retain less”, since early JSON errors are unrecoverable.
论文检索
输入标题、作者或关键词，从 12,319 篇学术成果中精准定位
While Large Language Models have significantly advanced Text2SQL generation, a critical semantic gap persists: syntactically valid queries can still misinterpret user intent. To mitigate this challenge, we propose GBV-SQL, a multi-agent framework that introduces Guided Generation with SQL2Text Back-translation Validation. In particular, a dedicated validator translates generated SQL back into natural language and checks whether its logic is aligned with the original question. Beyond the method itself, we also conduct a systematic audit of benchmark quality and introduce a typology of “Gold Errors” in Text2SQL datasets. Our analysis shows that benchmark issues can coexist with strong execution accuracy and can substantially affect evaluation outcomes. On the challenging BIRD benchmark, GBV-SQL achieves 63.23% execution accuracy, a 5.8% absolute improvement over the Deepseek-v3-based MAC-SQL setting. Under manually audited benchmark corrections, GBV-SQL reaches 96.5% (dev) and 97.6% (test) on Spider, and 90.42% on repaired BIRD dev, providing a diagnostic view of model behavior under improved gold quality. Our work contributes both a practical framework for semantic validation and an empirical analysis of benchmark integrity in Text2SQL evaluation.
Your Reasoning Benchmark May Not Test Reasoning: Revealing Perception Bottleneck in Abstract Reasoning Benchmarks
PDF ↗Reasoning benchmarks such as the Abstraction and Reasoning Corpus (ARC) and ARC-AGI are widely used to assess progress in artificial intelligence and are often interpreted as probes of core, so-called “fluid” reasoning abilities. Despite their apparent simplicity for humans, these tasks remain challenging for frontier vision-language models (VLMs), a gap commonly attributed to deficiencies in machine reasoning. We challenge this interpretation and hypothesize that the gap arises primarily from limitations in visual perception rather than from shortcomings in inductive reasoning.To verify this hypothesis, we introduce a two-stage experimental pipeline that explicitly separates perception and reasoning. In the perception stage, each image is independently converted into a natural-language description, while in the reasoning stage a model induces and applies rules using these descriptions. This design prevents leakage of cross-image inductive signals and isolates reasoning from perception bottlenecks. Across three ARC-style datasets, Mini-ARC, ACRE, and Bongard-LOGO, we show that the perception capability is the dominant factor underlying the observed performance gap by comparing the two-stage pipeline with against standard end-to-end one-stage evaluation. Manual inspection of reasoning traces in the VLM outputs further reveals that approximately 80 percent of model failures stem from perception errors. Together, these results demonstrate that ARC-style benchmarks conflate perceptual and reasoning challenges and that observed performance gaps may overstate deficiencies in machine reasoning. Our findings underscore the need for evaluation protocols that disentangle perception from reasoning when assessing progress in machine intelligence.
Phone recognition (PR) serves as the atomic interface for language-agnostic modeling for cross-lingual speech processing and phonetic analysis. Despite prolonged efforts in developing PR systems, current evaluations only measure surface-level transcription accuracy. We introduce PRiSM, the first open-source benchmark designed to expose blind spots in phonetic perception through intrinsic and extrinsic evaluation of PR systems. PRiSM standardizes transcription-based evaluation and assesses downstream utility in clinical, educational, and multilingual settings with transcription and representation probes. We find that diverse language exposure during training is key to PR performance, encoder-CTC models are the most stable, and specialized PR systems still outperform LALMs. PRiSM releases code, recipes, and datasets to move the field toward multilingual speech models with robust phonetic ability.
Protecting public figures from online abuse requires models that go beyond post-level classification to determine whether abuse is directed at a designated target, characterize the abuse intent, and extract textual evidence. We introduce a Target-Aware Multilingual Abuse (TAMA), benchmark of 9,386 X (Twitter) posts aimed at public figures, with aligned supervision for (i) tri-class target detection, (ii) 12-way fine-grained abuse type classification, and (iii) phrase-level abusive spans localization. To exploit the hierarchical coupling of these tasks, we propose Cascaded-MTL, a dependency-aware multi-task framework that conditions downstream predictions on upstream beliefs via three lightweight modules: Cross-Task Feature Fusion (CTF), Task-Adaptive Gating (TAG), and Label-Guided Span Detection (LGSD). Experiments across three multilingual encoders show that Cascaded-MTL consistently yields higher average F1 than single-task and standard multi-task training and delivers robust gains on type classification and span localization. The code and the dataset are released here: https://github.com/zgjiangtoby/CASCADED-MTL
Achieving robust generalization in speech deepfake detection (SDD) remains a primary challenge, as models often fail to detect unseen forgery methods. While research has focused on model-centric and algorithm-centric solutions, the impact of data composition is often underexplored. This paper proposes a data-centric approach, analyzing the SDD data landscape from two practical perspectives: constructing a single dataset and aggregating multiple datasets. To address the first perspective, we conduct a large-scale empirical study to characterize the data scaling laws for SDD, quantifying the impact of source and generator diversity. To address the second, we propose the Diversity-Optimized Sampling Strategy (DOSS), a principled framework for mixing heterogeneous data with two implementations: DOSS-Select (pruning) and DOSS-Weight (re-weighting). Our experiments show that DOSS-Select outperforms the naive aggregation baseline while using only 3% of the total available data. Furthermore, our final model, trained on a 12k-hour curated data pool using the optimal DOSS-Weight strategy, achieves state-of-the-art performance, outperforming large-scale baselines with greater data and model efficiency on both public benchmarks and a new challenge set of various commercial APIs.
DiZiNER: Disagreement-guided Instruction Refinement via Simulating Pilot Annotation for Zero-shot Named Entity Recognition
PDF ↗Large language models (LLMs) have advanced information extraction (IE) by enabling zero-shot and few-shot named entity recognition (NER), yet their generative outputs still show persistent and systematic errors. Despite progress through instruction fine-tuning, zero-shot NER still lags far behind supervised systems. These recurring errors mirror inconsistencies observed in early-stage human annotation processes that resolve disagreements through pilot annotation. Motivated by this analogy, we introduce DiZiNER (Disagreement-guided Instruction Refinement via Pilot Annotation Simulation for Zero-shot Named Entity Recognition), a framework that simulates the pilot annotation process, employing LLMs to act as both annotators and supervisors. Multiple heterogeneous LLMs annotate shared texts, and a supervisor model analyzes inter-model disagreements to refine task instructions. Across 18 benchmarks, DiZiNER achieves zero-shot SOTA results on 14 datasets, improving prior bests by +8.0 F1 and reducing the zero-shot to supervised gap by over +11 points. It also consistently outperforms its supervisor, GPT-5 mini, indicating that improvements stem from disagreement-guided instruction refinement rather than model capacity. Pairwise agreement between models shows a strong correlation with NER performance, further supporting this finding.
OpenRubrics: Towards Scalable Synthetic Rubric Generation for Reward Modeling and LLM Alignment
PDF ↗Reward modeling lies at the core of reinforcement learning from human feedback (RLHF), yet most existing reward models rely on scalar or pairwise judgments that fail to capture the multifaceted nature of human preferences. Recent studies have explored rubrics-as-rewards (RaR) that uses structured criteria to capture multiple dimensions of response quality. However, producing rubrics that are both reliable and scalable remains a key challenge. In this work, we introduce OpenRubrics, a diverse, large-scale collection of (prompt, rubric) pairs for training rubric-generation and rubric-based reward models. To elicit discriminative and comprehensive evaluation signals, we introduce Contrastive Rubric Generation (CRG), which derives both hard rules (explicit constraints) and principles (implicit qualities) by contrasting preferred and rejected responses. We further remove noisy rubrics via preserving preference–label consistency. Across multiple reward-modeling benchmarks, our rubric-based reward model, Rubric-RM, surpasses strong size-matched baselines by 8.4%. These gains transfer to policy models on instruction-following and biomedical benchmarks.The model weights and datasets are publicly available at https://huggingface.co/OpenRubrics.
Automated Fact-Checking has largely focused on verifying general knowledge against static corpora, overlooking high-stakes domains like law where truth is evolving and technically complex. We introduce CaseFacts, a benchmark for verifying colloquial legal claims against U.S. Supreme Court precedents. Unlike existing resources that map formal texts to formal texts, CaseFacts challenges systems to bridge the semantic gap between layperson assertions and technical jurisprudence while accounting for temporal validity. The dataset consists of 6,294 claims categorized as Supported, Refuted, or Overruled. We construct this benchmark using a multi-stage pipeline that leverages Large Language Models (LLMs) to synthesize claims from expert case summaries, employing a novel semantic similarity heuristic to efficiently identify and verify complex legal overrulings. Experiments with state-of-the-art LLMs reveal that the task remains challenging; notably, augmenting models with unrestricted web search degrades performance compared to closed-book baselines due to the retrieval of noisy, non-authoritative precedents. We release CaseFacts to spur research into legal fact verification systems.
MultiFinBen: Benchmarking Large Language Models for Multilingual and Multimodal Financial Application
PDF ↗Real-world financial analysis involves information across multiple languages and modalities, from reports and news to scanned filings and meeting recordings. Yet most existing evaluations of LLMs in finance remain text-only, monolingual, and largely saturated by current models. To bridge these gaps, we present MultiFinBen, the first expert-annotated multilingual (five languages) and multimodal (text, vision, audio) benchmark for evaluating LLMs in realistic financial contexts. MultiFinBen introduces two new task families: multilingual financial reasoning, which tests cross-lingual evidence integration from filings and news, and financial OCR, which extracts structured text from scanned documents containing tables and charts. Rather than aggregating all available datasets, we apply a structured, difficulty-aware selection based on advanced model performance, ensuring balanced challenge and removing redundant tasks. Evaluating 21 leading LLMs shows that even frontier multimodal models like GPT-4o achieve only 46.01% overall, stronger on vision and audio but dropping sharply in multilingual settings. These findings expose persistent limitations in multilingual, multimodal, and expert-level financial reasoning. All datasets, evaluation scripts, and leaderboards are publicly released.
Many applications of LLM-based text regression require predicting a full conditional distribution rather than a single point value. We study distributional regression under empirical-quantile supervision, where each input is paired with multiple observed quantile outcomes, and the target distribution is represented by a dense grid of quantiles. We address two key limitations of current approaches: the lack of local grounding for distribution estimates, and the reliance on shared representations that create an indirect bottleneck between inputs and quantile outputs. In this paper, we introduce Quantile Token Regression, which, to our knowledge, is the first work to insert dedicated quantile tokens into the input sequence, enabling direct input-output pathways for each quantile through self-attention. We further augment these quantile tokens with retrieval, incorporating semantically similar neighbor instances and their empirical distributions to ground predictions with local evidence from similar instances. We also provide the first theoretical analysis of loss functions for quantile regression, clarifying which distributional objectives each optimizes. Experiments on the Inside Airbnb and StackSample benchmark datasets with LLMs ranging from 1.7B to 14B parameters show that quantile tokens with neighbors consistently outperform baselines (\sim4 points lower MAPE and 2\times narrower prediction intervals), with especially large gains on smaller and more challenging datasets where quantile tokens produce substantially sharper and more accurate distributions.
ViDoRe V3: A Comprehensive Evaluation of Retrieval Augmented Generation in Complex Real-World Scenarios
PDF ↗Retrieval-Augmented Generation (RAG) pipelines must address challenges beyond simple single-document retrieval, such as interpreting visual elements (tables, charts, images), synthesizing information across documents, and providing accurate source grounding. Existing benchmarks fail to capture this complexity, often focusing on textual data, single-document comprehension, or evaluating retrieval and generation in isolation. We introduce ViDoRe V3, a comprehensive multimodal RAG benchmark featuring multi-type queries over visually rich document corpora. It covers 10 datasets across diverse professional domains, comprising ~26,000 document pages paired with 3,099 human-verified queries, each available in 6 languages. Through 12,000 hours of human annotation effort, we provide high-quality annotations for retrieval relevance, bounding box localization, and verified reference answers. Our evaluation of state-of-the-art RAG pipelines reveals that visual retrievers outperform textual ones, late-interaction models and textual reranking substantially improve performance, and hybrid or purely visual contexts enhance answer generation quality. However, current models still struggle with non-textual elements, open-ended queries, and fine-grained visual grounding. To encourage progress in addressing these challenges, the benchmark is released under a commercially permissive license.
The increasing use of large language models in mental health applications calls for principled evaluation frameworks that assess alignment with psychotherapeutic best practices beyond surface-level fluency. While recent systems exhibit conversational competence, they lack structured mechanisms to evaluate adherence to core therapeutic principles. In this paper, we study the problem of evaluating AI-generated therapist-like responses for clinically grounded appropriateness and effectiveness. We assess each therapists utterance along six therapeutic principles: non-judgmental acceptance, warmth, respect for autonomy, active listening, reflective understanding, and situational appropriateness using a fine-grained ordinal scale. We introduce FAITH-M, a benchmark annotated with expert-assigned ordinal ratings, and propose CARE, a multi-stage evaluation framework that integrates intra-dialogue context, contrastive exemplar retrieval, and knowledge-distilled chain-of-thought reasoning. Experiments show that CARE achieves an F-1 score of 63.34 versus the strong baseline Qwen3 F-1 score of 38.56 which is a 64.26% improvement, which also serves as its backbone, indicating that gains arise from structured reasoning and contextual modeling rather than backbone capacity alone. Expert assessment and external dataset evaluations further demonstrate robustness under domain shift, while highlighting challenges in modeling implicit clinical nuance. Overall, CARE provides a clinically grounded framework for evaluating therapeutic fidelity in AI mental health systems.
Hallucinations in machine translation (MT)—outputs that may be fluent yet unfaithful to the source content—remain a critical obstacle. They hinder the reliable deployment of MT systems in real-world applications. Despite growing attention to this phenomenon, progress has been constrained by the lack of large-scale, high-quality benchmarks dedicated to hallucination detection. We introduce HAT (Hallucination Annotation for Translation), a novel dataset designed to advance research on this problem. HAT comprises 350,959 span-level annotated samples across 38 language pairs, with approximately 8,000–10,000 samples per pair partitioned into training, development, and test sets. Annotations were produced by professional translators under rigorous quality control protocols to ensure reliability. We provide a detailed analysis of hallucination distributions and establish benchmark performance using a diverse set of baselines, including automatic MT evaluation metrics as well as large language models. By providing the first large-scale, systematically annotated resource for hallucination detection in MT, HAT enables the development of more faithful translation models and lays the groundwork for future research on building trustworthy machine translation systems.
The “LLM-as-a-judge” paradigm has become a standard method for evaluating open-ended generation. To address the quadratic scalability costs of pairwise comparisons, popular benchmarks like Arena-Hard and AlpacaEval compare all models against a single anchor. However, despite its widespread use, the impact of anchor selection on the reliability of the results remains largely unexplored. In this work, we systematically investigate the effect of anchor selection by evaluating 22 different anchors on the Arena-Hard-v2.0 dataset. We find that the choice of anchor is critical: a poor anchor can dramatically reduce correlation with human rankings. We identify that common anchor choices (best-performing and worst-performing models) make poor anchors. Because these extreme anchors are consistently better or worse than all other models, they are seldom indicative of the relative ranking of the models. We further quantify the effect size of anchor selection, showing it is comparable to the selection of a judge model. We conclude with actionable recommendations. First, we conduct a power analysis, and compute sufficient benchmark sizes for anchor-based evaluation, finding that standard benchmark sizes are insufficient for pairwise evaluation and fail to distinguish between competitive models reliably. Second, we provide guidelines for selecting informative anchors to ensure reliable and efficient evaluation practices.
Hallucinations remain a critical challenge for large language models (LLMs), particularly in Retrieval-Augmented Generation (RAG) settings where models may generate outputs unsupported by the provided context. To address this, we introduce TOHA, a TOpology-based HAllucination detector, which leverages a topological divergence metric to quantify the structural properties of graphs induced by attention matrices. Examining the topological divergence between prompt and response subgraphs in RAG settings reveals consistent patterns: higher divergence values in specific attention heads correlate with unfaithful outputs, independent of the dataset. Extensive experiments — including evaluations on question answering and summarization tasks — show that our approach achieves state-of-the-art or competitive results on several benchmarks while requiring minimal annotated data and computational resources. Our findings indicate that the topological structure of attention matrices provides an efficient and robust metric for assessing the correctness of LLM’s responses.
Metonymy and metaphor often co-occur in natural language, yet computational work has studied them largely in isolation. We introduce a framework that transforms a literal sentence into three figurative variants: metonymic, metaphoric, and hybrid. Using this framework, we construct MetFuse, the first dedicated dataset of figurative fusion between metonymy and metaphor, containing 1,000 human-verified meaning-aligned quadruplets totaling 4,000 sentences. Extrinsic experiments on eight existing benchmarks show that augmenting training data with MetFuse consistently improves both metonymy and metaphor classification, with hybrid examples yielding the largest gains on metonymy tasks. Using this dataset, we also analyze how the presence of one figurative type impacts another. Our findings show that both human annotators and large language models better identify metonymy in hybrid sentences than in metonymy-only sentences, demonstrating that the presence of a metaphor makes a metonymic noun more explicit.
Emotion is a central dimension of spoken communication, yet, we still lack a mechanistic account of how modern large audio-language models (LALMs) encode it internally. We present the first neuron-level interpretability study of emotion-sensitive neurons (ESNs) in LALMs and provide causal evidence supporting the existence of such units in Qwen2.5-Omni, Kimi-Audio, and Audio Flamingo 3. Across these three widely used open-source models, we compare frequency-, entropy-, mean-deviation-, and contrast-based neuron selectors on multiple emotion recognition benchmarks. Using inference-time interventions, we reveal a consistent emotion-specific signature: deactivating neurons selected for a given emotion disproportionately degrades recognition of that emotion while largely preserving other classes, whereas targeted steering amplifies these units to bias predictions toward the target emotion. These effects arise with modest amounts of identification data and scale systematically with intervention strength. We further observe that ESNs exhibit non-uniform layer-wise clustering with partial cross-dataset transfer. Taken together, our results offer a causal, neuron-level account of emotion decisions in LALMs and highlight targeted neuron interventions as an actionable handle for controllable affective behaviors.
Grammaticality and likelihood are distinct notions in human language. Pretrained language models (LMs), which are probabilistic models of language fitted to maximize corpus likelihood, generate grammatically well-formed text and discriminate well between grammatical and ungrammatical sentences in tightly controlled minimal pairs. However, their string probabilities do not sharply discriminate between grammatical and ungrammatical sentences overall. But do LMs implicitly acquire a grammaticality distinction distinct from string probability? We explore this question through studying pretrained LMs’ internal representations, by training a linear probe on a dataset of grammatical and (synthetic) ungrammatical sentences obtained by applying perturbations to a naturalistic text corpus. We find that this simple grammaticality probe generalizes to human-curated grammaticality judgment benchmarks and outperforms LM probability-based grammaticality judgments. When applied to semantic plausibility benchmarks, however, in which both members of a minimal pair are grammatical and differ in only plausibility, the probe performs worse than string probability. The English-trained probe also exhibits nontrivial cross-lingual generalization, outperforming string probabilities on grammaticality benchmarks in numerous other languages. Additionally, probe scores correlate only weakly with string probabilities. These results collectively suggest that pretrained LMs acquire to some extent an implicit grammaticality distinction within their hidden layers.
Motion sensor time-series are central to Human Activity Recognition (HAR), yet conventional approaches are constrained to fixed activity sets and typically require costly parameter retraining to adapt to new behaviors. While Large Language Models (LLMs) offer promising open-set reasoning capabilities, applying them directly to numerical time-series often leads to hallucinations and weak grounding. To address this challenge, we propose ZARA (Zero-training Activity Reasoning Agents), a knowledge- and retrieval-augmented agentic framework for motion time-series reasoning in a training-free inference setting. Rather than relying on black-box projections, ZARA distills reference data into a statistically grounded textual knowledge base that transforms implicit signal patterns into verifiable natural-language priors. Guided by retrieved evidence, ZARA iteratively selects discriminative cues and performs grounded reasoning over candidate activities. Extensive experiments on eight benchmarks show that ZARA generalizes robustly to unseen subjects and across datasets, demonstrating strong transferability across heterogeneous sensor domains. These results mark a step toward trustworthy, plug-and-play motion understanding beyond dataset-specific artifacts. Our code is available at https://github.com/zechenli03/ZARA.