Qingcheng Zeng

Qingcheng Zeng, Lucas Garay, Peilin Zhou et al. · cambridge, harvard

Large pre-trained models have revolutionized natural language processing (NLP) research and applications, but high training costs and limited data resources have prevented their benefits from being shared equally amongst speakers of all the world's languages. To address issues of cross-linguistic access to such models and reduce energy consumption for sustainability during large-scale model training, this study proposes an effective and energy-efficient framework called GreenPLM that uses bilingual lexicons to directly "translate" pre-trained language models of one language into another at almost no additional cost. We validate this approach in 18 languages' BERT models and show that this framework is comparable to, if not better than, other heuristics with high training costs. In addition, given lightweight continued pre-training on limited data where available, this framework outperforms the original monolingual language models in six out of seven tested languages with up to 200x less pre-training efforts. Aiming at the Leave No One Behind Principle (LNOB), our approach manages to reduce inequalities between languages and energy consumption greatly. We make our codes and models publicly available here: \url{https://github.com/qcznlp/GreenPLMs}

Sheng Jin, Minghao Liu, Yunze Xiao et al.

Knowledge benchmarks for LLMs face three issues: scaling-driven designs that do not operationalize disciplinary representativeness; flat-payment annotation that permits lazy consensus; and unaudited ranking instability under bounded test budgets. We introduce KINA, an 899-item benchmark across 261 fine-grained disciplines, with two formal results. First, we cast representativeness as a coverage-style objective over expert-elicited anchors and operationalize disciplinary representativeness through a proxy, yielding a (1-1/e) greedy approximation (Proposition 1); the guarantee applies to the proxy, not to population representativeness. Second, we prove a bonus-on-bar tournament weakly FOSD-dominates flat payment in released-review quality, with incentive-compatibility threshold B > Delta C / Delta p_min (Theorem 1). Evaluating 42 models from 13 labs, the top model, Gemini-3.1-Pro-Preview, reaches 53.17%, followed by Claude-Opus-4.6 at 49.92% and GPT-5.4 at 48.55%, leaving substantial headroom below saturation. The full leaderboard shows a tiered structure rather than a smooth total order: a small frontier tier lies above 48%, a dense strong-model tier spans roughly 38-45%, and low-performing models remain only modestly above the 10% chance baseline. Tool augmentation adds up to 5.17 points across the five tool-use evaluations, with gains varying substantially across models. We report bootstrap ranking-stability statistics to make bounded-budget variance explicit and to discourage over-interpretation of adjacent ranks.

Qingcheng Zeng, Mingyu Jin, Qinkai Yu et al.

Large Language Models (LLMs) are employed across various high-stakes domains, where the reliability of their outputs is crucial. One commonly used method to assess the reliability of LLMs' responses is uncertainty estimation, which gauges the likelihood of their answers being correct. While many studies focus on improving the accuracy of uncertainty estimations for LLMs, our research investigates the fragility of uncertainty estimation and explores potential attacks. We demonstrate that an attacker can embed a backdoor in LLMs, which, when activated by a specific trigger in the input, manipulates the model's uncertainty without affecting the final output. Specifically, the proposed backdoor attack method can alter an LLM's output probability distribution, causing the probability distribution to converge towards an attacker-predefined distribution while ensuring that the top-1 prediction remains unchanged. Our experimental results demonstrate that this attack effectively undermines the model's self-evaluation reliability in multiple-choice questions. For instance, we achieved a 100 attack success rate (ASR) across three different triggering strategies in four models. Further, we investigate whether this manipulation generalizes across different prompts and domains. This work highlights a significant threat to the reliability of LLMs and underscores the need for future defenses against such attacks. The code is available at https://github.com/qcznlp/uncertainty_attack.

Rui Yang, Qingcheng Zeng, Keen You et al.

This study introduces Ascle, a pioneering natural language processing (NLP) toolkit designed for medical text generation. Ascle is tailored for biomedical researchers and healthcare professionals with an easy-to-use, all-in-one solution that requires minimal programming expertise. For the first time, Ascle evaluates and provides interfaces for the latest pre-trained language models, encompassing four advanced and challenging generative functions: question-answering, text summarization, text simplification, and machine translation. In addition, Ascle integrates 12 essential NLP functions, along with query and search capabilities for clinical databases. The toolkit, its models, and associated data are publicly available via https://github.com/Yale-LILY/MedGen.

Daoyang Li, Haiyan Zhao, Qingcheng Zeng et al.

Probing techniques for large language models (LLMs) have primarily focused on English, overlooking the vast majority of the world's languages. In this paper, we extend these probing methods to a multilingual context, investigating the behaviors of LLMs across diverse languages. We conduct experiments on several open-source LLM models, analyzing probing accuracy, trends across layers, and similarities between probing vectors for multiple languages. Our key findings reveal: (1) a consistent performance gap between high-resource and low-resource languages, with high-resource languages achieving significantly higher probing accuracy; (2) divergent layer-wise accuracy trends, where high-resource languages show substantial improvement in deeper layers similar to English; and (3) higher representational similarities among high-resource languages, with low-resource languages demonstrating lower similarities both among themselves and with high-resource languages. These results highlight significant disparities in LLMs' multilingual capabilities and emphasize the need for improved modeling of low-resource languages.

Fan Gao, Hang Jiang, Rui Yang et al. · mit

Educational materials such as survey articles in specialized fields like computer science traditionally require tremendous expert inputs and are therefore expensive to create and update. Recently, Large Language Models (LLMs) have achieved significant success across various general tasks. However, their effectiveness and limitations in the education domain are yet to be fully explored. In this work, we examine the proficiency of LLMs in generating succinct survey articles specific to the niche field of NLP in computer science, focusing on a curated list of 99 topics. Automated benchmarks reveal that GPT-4 surpasses its predecessors, inluding GPT-3.5, PaLM2, and LLaMa2 by margins ranging from 2% to 20% in comparison to the established ground truth. We compare both human and GPT-based evaluation scores and provide in-depth analysis. While our findings suggest that GPT-created surveys are more contemporary and accessible than human-authored ones, certain limitations were observed. Notably, GPT-4, despite often delivering outstanding content, occasionally exhibited lapses like missing details or factual errors. At last, we compared the rating behavior between humans and GPT-4 and found systematic bias in using GPT evaluation.

Qingcheng Zeng, Dading Chong, Peilin Zhou et al. · harvard

Accented speech recognition and accent classification are relatively under-explored research areas in speech technology. Recently, deep learning-based methods and Transformer-based pretrained models have achieved superb performances in both areas. However, most accent classification tasks focused on classifying different kinds of English accents and little attention was paid to geographically-proximate accent classification, especially under a low-resource setting where forensic speech science tasks usually encounter. In this paper, we explored three main accent modelling methods combined with two different classifiers based on 105 speaker recordings retrieved from five urban varieties in Northern England. Although speech representations generated from pretrained models generally have better performances in downstream classification, traditional methods like Mel Frequency Cepstral Coefficients (MFCCs) and formant measurements are equipped with specific strengths. These results suggest that in forensic phonetics scenario where data are relatively scarce, a simple modelling method and classifier could be competitive with state-of-the-art pretrained speech models as feature extractors, which could enhance a sooner estimation for the accent information in practices. Besides, our findings also cross-validated a new methodology in quantifying sociophonetic changes.

Mingyu Jin, Yutong Yin, Jingcheng Niu et al.

In this work, we investigate how Large Language Models (LLMs) adapt their internal representations when encountering inputs of increasing difficulty, quantified as the degree of out-of-distribution (OOD) shift. We reveal a consistent and quantifiable phenomenon: as task difficulty increases, whether through harder reasoning questions, longer contexts, or adding answer choices, the last hidden states of LLMs become substantially sparser. In short, \textbf{\textit{the farther the shift, the sparser the representations}}. This sparsity--difficulty relation is observable across diverse models and domains, suggesting that language models respond to unfamiliar or complex inputs by concentrating computation into specialized subspaces in the last hidden state. Through a series of controlled analyses with a learning dynamic explanation, we demonstrate that this sparsity is not incidental but an adaptive mechanism for stabilizing reasoning under OOD. Leveraging this insight, we design \textit{Sparsity-Guided Curriculum In-Context Learning (SG-ICL)}, a strategy that explicitly uses representation sparsity to schedule few-shot demonstrations, leading to considerable performance enhancements. Our study provides new mechanistic insights into how LLMs internalize OOD challenges. The source code is available at the URL: https://github.com/MingyuJ666/sparsityLLM.

Peilin Zhou, Dading Chong, Helin Wang et al.

The past ten years have witnessed the rapid development of text-based intent detection, whose benchmark performances have already been taken to a remarkable level by deep learning techniques. However, automatic speech recognition (ASR) errors are inevitable in real-world applications due to the environment noise, unique speech patterns and etc, leading to sharp performance drop in state-of-the-art text-based intent detection models. Essentially, this phenomenon is caused by the semantic drift brought by ASR errors and most existing works tend to focus on designing new model structures to reduce its impact, which is at the expense of versatility and flexibility. Different from previous one-piece model, in this paper, we propose a novel and agile framework called CR-ID for ASR error robust intent detection with two plug-and-play modules, namely semantic drift calibration module (SDCM) and phonemic refinement module (PRM), which are both model-agnostic and thus could be easily integrated to any existing intent detection models without modifying their structures. Experimental results on SNIPS dataset show that, our proposed CR-ID framework achieves competitive performance and outperform all the baseline methods on ASR outputs, which verifies that CR-ID can effectively alleviate the semantic drift caused by ASR errors.

Rui Yang, Huitao Li, Weihao Xuan et al.

Despite continuous advances in medical technology, the global distribution of health care resources remains uneven. The development of large language models (LLMs) has transformed the landscape of medicine and holds promise for improving health care quality and expanding access to medical information globally. However, existing LLMs are primarily trained on high-resource languages, limiting their applicability in global medical scenarios. To address this gap, we constructed GlobMed, a large multilingual medical dataset, containing over 500,000 entries spanning 12 languages, including four low-resource languages. Building on this, we established GlobMed-Bench, which systematically assesses 56 state-of-the-art proprietary and open-weight LLMs across multiple multilingual medical tasks, revealing significant performance disparities across languages, particularly for low-resource languages. Additionally, we introduced GlobMed-LLMs, a suite of multilingual medical LLMs trained on GlobMed, with parameters ranging from 1.7B to 8B. GlobMed-LLMs achieved an average performance improvement of over 40% relative to baseline models, with a more than threefold increase in performance on low-resource languages. Together, these resources provide an important foundation for advancing the equitable development and application of LLMs globally, enabling broader language communities to benefit from technological advances.

Mukai Li, Qingcheng Zeng, Tianqing Fang et al.

As large language model agents tackle increasingly complex long-horizon tasks, effective post-training becomes critical. Prior work faces fundamental challenges: outcome-only rewards fail to precisely attribute credit to intermediate steps, estimated step-level rewards introduce systematic noise, and Monte Carlo sampling approaches for step reward estimation incur prohibitive computational cost. Inspired by findings that only a small fraction of high-entropy tokens drive effective RL for reasoning, we propose Critical Step Optimization (CSO), which focuses preference learning on verified critical steps, decision points where alternate actions demonstrably flip task outcomes from failure to success. Crucially, our method starts from failed policy trajectories rather than expert demonstrations, directly targeting the policy model's weaknesses. We use a process reward model (PRM) to identify candidate critical steps, leverage expert models to propose high-quality alternatives, then continue execution from these alternatives using the policy model itself until task completion. Only alternatives that the policy successfully executes to correct outcomes are verified and used as DPO training data, ensuring both quality and policy reachability. This yields fine-grained, verifiable supervision at critical decisions while avoiding trajectory-level coarseness and step-level noise. Experiments on GAIA-Text-103 and XBench-DeepSearch show that CSO achieves 37% and 26% relative improvement over the SFT baseline and substantially outperforms other post-training methods, while requiring supervision at only 16% of trajectory steps. This demonstrates the effectiveness of selective verification-based learning for agent post-training.

Qingcheng Zeng, Yuheng Lu, Zeqi Zhou et al.

Code-switching is a pervasive linguistic phenomenon in global communication, yet modern information retrieval systems remain predominantly designed for, and evaluated within, monolingual contexts. To bridge this critical disconnect, we present a holistic study dedicated to code-switching IR. We introduce CSR-L (Code-Switching Retrieval benchmark-Lite), constructing a dataset via human annotation to capture the authentic naturalness of mixed-language queries. Our evaluation across statistical, dense, and late-interaction paradigms reveals that code-switching acts as a fundamental performance bottleneck, degrading the effectiveness of even robust multilingual models. We demonstrate that this failure stems from substantial divergence in the embedding space between pure and code-switched text. Scaling this investigation, we propose CS-MTEB, a comprehensive benchmark covering 11 diverse tasks, where we observe performance declines of up to 27%. Finally, we show that standard multilingual techniques like vocabulary expansion are insufficient to resolve these deficits completely. These findings underscore the fragility of current systems and establish code-switching as a crucial frontier for future IR optimization.

Jiayu Liu, Rui Wang, Qing Zong et al.

Accurately assessing model confidence is essential for deploying large language models (LLMs) in mission-critical factual domains. While retrieval-augmented generation (RAG) is widely adopted to improve grounding, confidence calibration in RAG settings remains poorly understood. We conduct a systematic study across four benchmarks, revealing that LLMs exhibit poor calibration performance due to noisy retrieved contexts. Specifically, contradictory or irrelevant evidence tends to inflate the model's false certainty, leading to severe overconfidence. To address this, we propose NAACL Rules (Noise-AwAre Confidence CaLibration Rules) to provide a principled foundation for resolving overconfidence under noise. We further design NAACL, a noise-aware calibration framework that synthesizes supervision from about 2K HotpotQA examples guided by these rules. By performing supervised fine-tuning (SFT) with this data, NAACL equips models with intrinsic noise awareness without relying on stronger teacher models. Empirical results show that NAACL yields substantial gains, improving ECE scores by 10.9% in-domain and 8.0% out-of-domain. By bridging the gap between retrieval noise and verbal calibration, NAACL paves the way for both accurate and epistemically reliable LLMs.

Qingcheng Zeng, Puxuan Yu, Aman Mehta et al.

Instruction-following information retrieval (IF-IR) studies retrieval systems that must not only find documents relevant to a query, but also obey explicit user constraints such as required attributes, exclusions, or output preferences. However, most retrievers are trained primarily for semantic relevance and often fail to distinguish documents that match the topic from those that satisfy the instruction. We propose a dual-view data synthesis strategy based on polarity reversal: given a query, a document that is relevant under the instruction, and a hard negative that matches the query but violates the instruction, we prompt an LLM to generate a complementary instruction under which the two documents swap relevance labels. By presenting the same document pair under complementary instructions that invert their relevance labels, the training signal forces the retriever to reconsider the same candidate set through the instruction, rather than relying on fixed topical cues. On a 305M-parameter encoder, our method improves performance on the FollowIR benchmark by 45%, surpassing general-purpose embedding models of comparable or larger scale. Through head-to-head comparisons at matched data budgets, we further show that data diversity and instruction supervision play complementary roles: the former preserves general retrieval quality, while the latter improves instruction sensitivity. These results highlight the value of targeted data synthesis for building retrieval systems that are both broadly capable and instruction-aware.

Qingcheng Zeng, An-Ran Li

Irony is a ubiquitous figurative language in daily communication. Previously, many researchers have approached irony from linguistic, cognitive science, and computational aspects. Recently, some progress have been witnessed in automatic irony processing due to the rapid development in deep neural models in natural language processing (NLP). In this paper, we will provide a comprehensive overview of computational irony, insights from linguistic theory and cognitive science, as well as its interactions with downstream NLP tasks and newly proposed multi-X irony processing perspectives.

Zezhong Jin, Dading Zhong, Xiao Song et al.

Fine tuning self supervised pretrained models using pseudo labels can effectively improve speech recognition performance. But, low quality pseudo labels can misguide decision boundaries and degrade performance. We propose a simple yet effective strategy to filter low quality pseudo labels to alleviate this problem. Specifically, pseudo-labels are produced over the entire training set and filtered via average probability scores calculated from the model output. Subsequently, an optimal percentage of utterances with high probability scores are considered reliable training data with trustworthy labels. The model is iteratively updated to correct the unreliable pseudo labels to minimize the effect of noisy labels. The process above is repeated until unreliable pseudo abels have been adequately corrected. Extensive experiments on LibriSpeech show that these filtered samples enable the refined model to yield more correct predictions, leading to better ASR performances under various experimental settings.

Weihao Xuan, Qingcheng Zeng, Heli Qi et al.

Autonomous agents based on large language models (LLMs) are rapidly evolving to handle multi-turn tasks, but ensuring their trustworthiness remains a critical challenge. A fundamental pillar of this trustworthiness is calibration, which refers to an agent's ability to express confidence that reliably reflects its actual performance. While calibration is well-established for static models, its dynamics in tool-integrated agentic workflows remain underexplored. In this work, we systematically investigate verbalized calibration in tool-use agents, revealing a fundamental confidence dichotomy driven by tool type. Specifically, our pilot study identifies that evidence tools (e.g., web search) systematically induce severe overconfidence due to inherent noise in retrieved information, while verification tools (e.g., code interpreters) can ground reasoning through deterministic feedback and mitigate miscalibration. To robustly improve calibration across tool types, we propose a reinforcement learning (RL) fine-tuning framework that jointly optimizes task accuracy and calibration, supported by a holistic benchmark of reward designs. We demonstrate that our trained agents not only achieve superior calibration but also exhibit robust generalization from local training environments to noisy web settings and to distinct domains such as mathematical reasoning. Our results highlight the necessity of domain-specific calibration strategies for tool-use agents. More broadly, this work establishes a foundation for building self-aware agents that can reliably communicate uncertainty in high-stakes, real-world deployments.

Minghao Guo, Qingcheng Zeng, Xujiang Zhao et al.

Large Language Models (LLMs) excel at many reasoning tasks but struggle with knowledge-intensive queries due to their inability to dynamically access up-to-date or domain-specific information. Retrieval-Augmented Generation (RAG) has emerged as a promising solution, enabling LLMs to ground their responses in external sources. However, existing RAG methods lack fine-grained control over both the query and source sides, often resulting in noisy retrieval and shallow reasoning. In this work, we introduce DeepSieve, an agentic RAG framework that incorporates information sieving via LLM-as-a-knowledge-router. DeepSieve decomposes complex queries into structured sub-questions and recursively routes each to the most suitable knowledge source, filtering irrelevant information through a multi-stage distillation process. Our design emphasizes modularity, transparency, and adaptability, leveraging recent advances in agentic system design. Experiments on multi-hop QA tasks across heterogeneous sources demonstrate improved reasoning depth, retrieval precision, and interpretability over conventional RAG approaches. Our codes are available at https://github.com/MinghoKwok/DeepSieve.

Ziqi Gao, Yaotian Zhu, Qingcheng Zeng et al.

Probing studies what information is encoded in a frozen LLM's layer representations by training a lightweight predictor on top of them. Beyond analysis, probes are often used operationally in probe-then-steer pipelines: a learned concept vector is extracted from a probe and injected via additive activation steering by adding it to a layer representation during the forward pass. The effectiveness of this pipeline hinges on estimating concept vectors that are accurate, directionally stable under ablation, and inexpensive to obtain. Motivated by these desiderata, we propose RAPTOR (Ridge-Adaptive Logistic Probe), a simple L2-regularized logistic probe whose validation-tuned ridge strength yields concept vectors from normalized weights. Across extensive experiments on instruction-tuned LLMs and human-written concept datasets, RAPTOR matches or exceeds strong baselines in accuracy while achieving competitive directional stability and substantially lower training cost; these quantitative results are supported by qualitative downstream steering demonstrations. Finally, using the Convex Gaussian Min-max Theorem (CGMT), we provide a mechanistic characterization of ridge logistic regression in an idealized Gaussian teacher-student model in the high-dimensional few-shot regime, explaining how penalty strength mediates probe accuracy and concept-vector stability and yielding structural predictions that qualitatively align with trends observed on real LLM embeddings.

Suet-Ying Lam, Qingcheng Zeng, Jingyi Wu et al.

Whether large language models (LLMs) process language similarly to humans has been the subject of much theoretical and practical debate. We examine this question through the lens of the production-interpretation distinction found in human sentence processing and evaluate the extent to which instruction-tuned LLMs replicate this distinction. Using an empirically documented asymmetry between pronoun production and interpretation in humans for implicit causality verbs as a testbed, we find that some LLMs do quantitatively and qualitatively reflect human-like asymmetries between production and interpretation. We demonstrate that whether this behavior holds depends upon both model size-with larger models more likely to reflect human-like patterns and the choice of meta-linguistic prompts used to elicit the behavior. Our codes and results are available at https://github.com/LingMechLab/Production-Interpretation_Asymmetries_ACL2025.

Mingyu Jin, Qinkai Yu, Jingyuan Huang et al.

Large language models (LLMs) have shown remarkable performances across a wide range of tasks. However, the mechanisms by which these models encode tasks of varying complexities remain poorly understood. In this paper, we explore the hypothesis that LLMs process concepts of varying complexities in different layers, introducing the idea of "Concept Depth" to suggest that more complex concepts are typically acquired in deeper layers. Specifically, we categorize concepts based on their level of abstraction, defining them in the order of increasing complexity within factual, emotional, and inferential tasks. We conduct extensive probing experiments using layer-wise representations across various LLM families (Gemma, LLaMA, Qwen) on various datasets spanning the three domains of tasks. Our findings reveal that models could efficiently conduct probing for simpler tasks in shallow layers, and more complex tasks typically necessitate deeper layers for accurate understanding. Additionally, we examine how external factors, such as adding noise to the input and quantizing the model weights, might affect layer-wise representations. Our findings suggest that these factors can impede the development of a conceptual understanding of LLMs until deeper layers are explored. We hope that our proposed concept and experimental insights will enhance the understanding of the mechanisms underlying LLMs. Our codes are available at https://github.com/Luckfort/CD.

Suet-Ying Lam, Qingcheng Zeng, Kexun Zhang et al.

While a large body of literature suggests that large language models (LLMs) acquire rich linguistic representations, little is known about whether they adapt to linguistic biases in a human-like way. The present study probes this question by asking whether LLMs display human-like referential biases using stimuli and procedures from real psycholinguistic experiments. Recent psycholinguistic studies suggest that humans adapt their referential biases with recent exposure to referential patterns; closely replicating three relevant psycholinguistic experiments from Johnson & Arnold (2022) in an in-context learning (ICL) framework, we found that InstructGPT adapts its pronominal interpretations in response to the frequency of referential patterns in the local discourse, though in a limited fashion: adaptation was only observed relative to syntactic but not semantic biases. By contrast, FLAN-UL2 fails to generate meaningful patterns. Our results provide further evidence that contemporary LLMs discourse representations are sensitive to syntactic patterns in the local context but less so to semantic patterns. Our data and code are available at \url{https://github.com/zkx06111/llm_priming}.

Rui Yang, Haoran Liu, Edison Marrese-Taylor et al.

Large language models (LLMs) have demonstrated impressive generative capabilities with the potential to innovate in medicine. However, the application of LLMs in real clinical settings remains challenging due to the lack of factual consistency in the generated content. In this work, we develop an augmented LLM framework, KG-Rank, which leverages a medical knowledge graph (KG) along with ranking and re-ranking techniques, to improve the factuality of long-form question answering (QA) in the medical domain. Specifically, when receiving a question, KG-Rank automatically identifies medical entities within the question and retrieves the related triples from the medical KG to gather factual information. Subsequently, KG-Rank innovatively applies multiple ranking techniques to refine the ordering of these triples, providing more relevant and precise information for LLM inference. To the best of our knowledge, KG-Rank is the first application of KG combined with ranking models in medical QA specifically for generating long answers. Evaluation on four selected medical QA datasets demonstrates that KG-Rank achieves an improvement of over 18% in ROUGE-L score. Additionally, we extend KG-Rank to open domains, including law, business, music, and history, where it realizes a 14% improvement in ROUGE-L score, indicating the effectiveness and great potential of KG-Rank.

Weihao Xuan, Rui Yang, Heli Qi et al.

Existing large language model (LLM) evaluation benchmarks primarily focus on English, while current multilingual tasks lack parallel questions that specifically assess cross-linguistic reasoning abilities. This dual limitation makes it challenging to comprehensively assess LLMs' performance in the multilingual setting. To fill this gap, we introduce MMLU-ProX, a comprehensive benchmark covering 29 languages, built on an English benchmark. Each language version consists of 11,829 identical questions, enabling direct cross-linguistic comparisons. Additionally, to meet efficient evaluation needs, we provide a lite version containing 658 questions per language. To ensure the high quality of MMLU-ProX, we employ a rigorous development process that involves multiple powerful LLMs for translation, followed by expert review to ensure accurate expression, consistent terminology, and cultural relevance. Building on this, we systematically evaluate 36 state-of-the-art LLMs, including reasoning-enhanced and multilingual-optimized LLMs. The results reveal significant disparities in the multilingual capabilities of LLMs: While they perform well in high-resource languages, their performance declines markedly in low-resource languages, with gaps of up to 24.3%. Through MMLU-ProX, we aim to advance the development of more inclusive AI systems and promote equitable access to technology across global contexts.

Shulin Huang, Linyi Yang, Yan Song et al.

Evaluating large language models (LLMs) poses significant challenges, particularly due to issues of data contamination and the leakage of correct answers. To address these challenges, we introduce ThinkBench, a novel evaluation framework designed to evaluate LLMs' reasoning capability robustly. ThinkBench proposes a dynamic data generation method for constructing out-of-distribution (OOD) datasets and offers an OOD dataset that contains 2,912 samples drawn from reasoning tasks. ThinkBench unifies the evaluation of reasoning models and non-reasoning models. We evaluate 16 LLMs and 4 PRMs under identical experimental conditions and show that most of the LLMs' performance are far from robust and they face a certain level of data leakage. By dynamically generating OOD datasets, ThinkBench effectively provides a reliable evaluation of LLMs and reduces the impact of data contamination.

Philipp Schoenegger, Francesco Salvi, Jiacheng Liu et al. · oxford

We directly compare the persuasion capabilities of a frontier large language model (LLM; Claude Sonnet 3.5) against incentivized human persuaders in an interactive, real-time conversational quiz setting. In this preregistered, large-scale incentivized experiment, participants (quiz takers) completed an online quiz where persuaders (either humans or LLMs) attempted to persuade quiz takers toward correct or incorrect answers. We find that LLM persuaders achieved significantly higher compliance with their directional persuasion attempts than incentivized human persuaders, demonstrating superior persuasive capabilities in both truthful (toward correct answers) and deceptive (toward incorrect answers) contexts. We also find that LLM persuaders significantly increased quiz takers' accuracy, leading to higher earnings, when steering quiz takers toward correct answers, and significantly decreased their accuracy, leading to lower earnings, when steering them toward incorrect answers. Overall, our findings suggest that AI's persuasion capabilities already exceed those of humans that have real-money bonuses tied to performance. Our findings of increasingly capable AI persuaders thus underscore the urgency of emerging alignment and governance frameworks.

Qingcheng Zeng, Weihao Xuan, Leyang Cui et al.

Large reasoning models (LRMs) have recently demonstrated impressive capabilities in complex reasoning tasks by leveraging increased test-time computation and exhibiting behaviors reminiscent of human-like self-reflection. While LRMs show a clear capacity for valuable self-reflection, how this ability interacts with other model behaviors remains underexplored. We investigate this connection by analyzing verbalized confidence, how models articulate their certainty, as a lens into the nature of self-reflection in LRMs. We find that supervised fine-tuning on reasoning traces (i.e., distillation) and reinforcement learning can improve verbalized calibration in reasoning-intensive settings in a progressive, laddered fashion. However, our results also indicate that reasoning models may possess a diminished awareness of their own knowledge boundaries, as evidenced by significantly lower "I don't know" response rates on factuality benchmarks. Moreover, we examine the relationship between verbalized confidence and reasoning chains, finding that models tend to express higher confidence when providing shorter or less elaborate reasoning. Our findings highlight how reasoning-oriented training can enhance performance in reasoning-centric tasks while potentially incurring a "reasoning tax," a cost reflected in the model's reduced ability to accurately recognize the limits of its own knowledge in small-scale models. More broadly, our work showcases how this erosion of knowledge boundaries can compromise model faithfulness, as models grow more confident without a commensurate understanding of when they should abstain.

Guangyang Zeng, Qingcheng Zeng, Xinghan Li et al.

Given 2D point correspondences between an image pair, inferring the camera motion is a fundamental issue in the computer vision community. The existing works generally set out from the epipolar constraint and estimate the essential matrix, which is not optimal in the maximum likelihood (ML) sense. In this paper, we dive into the original measurement model with respect to the rotation matrix and normalized translation vector and formulate the ML problem. We then propose a two-step algorithm to solve it: In the first step, we estimate the variance of measurement noises and devise a consistent estimator based on bias elimination; In the second step, we execute a one-step Gauss-Newton iteration on manifold to refine the consistent estimate. We prove that the proposed estimate owns the same asymptotic statistical properties as the ML estimate: The first is consistency, i.e., the estimate converges to the ground truth as the point number increases; The second is asymptotic efficiency, i.e., the mean squared error of the estimate converges to the theoretical lower bound -- Cramer-Rao bound. In addition, we show that our algorithm has linear time complexity. These appealing characteristics endow our estimator with a great advantage in the case of dense point correspondences. Experiments on both synthetic data and real images demonstrate that when the point number reaches the order of hundreds, our estimator outperforms the state-of-the-art ones in terms of estimation accuracy and CPU time.

Aaron Tu, Weihao Xuan, Heli Qi et al. · gatech

Reinforcement learning with verifiable rewards (RLVR) is a practical and scalable approach to enhancing large language models in areas such as math, code, and other structured tasks. Two questions motivate this paper: how much of the reported gains survive under strictly parity-controlled evaluation, and whether RLVR is cost-free or exacts a measurable tax. We argue that progress is real, but gains are often overstated due to three forces - an RLVR tax, evaluation pitfalls, and data contamination. Using a partial-prompt contamination audit and matched-budget reproductions across base and RL models, we show that several headline gaps shrink or vanish under clean, parity-controlled evaluation. We then propose a tax-aware training and evaluation protocol that co-optimizes accuracy, grounding, and calibrated abstention and standardizes budgeting and provenance checks. Applied to recent RLVR setups, this protocol yields more reliable estimates of reasoning gains and, in several cases, revises prior conclusions. Our position is constructive: RLVR is valuable and industry-ready; we advocate keeping its practical benefits while prioritizing reliability, safety, and measurement.

Zhaoqian Xue, Guanhong Liu, Kai Wei et al.

Access to health resources is a critical determinant of public well-being and societal resilience, particularly during public health crises when demand for medical services and preventive care surges. However, disparities in accessibility persist across demographic and geographic groups, raising concerns about equity. Traditional survey methods often fall short due to limitations in coverage, cost, and timeliness. This study leverages crowdsourced data from Google Maps reviews, applying advanced natural language processing techniques, specifically ModernBERT, to extract insights on public perceptions of health resource accessibility in the United States during the COVID-19 pandemic. Additionally, we employ Partial Least Squares regression to examine the relationship between accessibility perceptions and key socioeconomic and demographic factors including political affiliation, racial composition, and educational attainment. Our findings reveal that public perceptions of health resource accessibility varied significantly across the U.S., with disparities peaking during the pandemic and slightly easing post-crisis. Political affiliation, racial demographics, and education levels emerged as key factors shaping these perceptions. These findings underscore the need for targeted interventions and policy measures to address inequities, fostering a more inclusive healthcare infrastructure that can better withstand future public health challenges.

Qingcheng Zeng, Guanhong Liu, Zhaoqian Xue et al.

On July 13, 2024, at the Trump rally in Pennsylvania, someone attempted to assassinate Republican Presidential Candidate Donald Trump. This attempt sparked a large-scale discussion on social media. We collected posts from X (formerly known as Twitter) one week before and after the assassination attempt and aimed to model the short-term effects of such a ``shock'' on public opinions and discussion topics. Specifically, our study addresses three key questions: first, we investigate how public sentiment toward Donald Trump shifts over time and across regions (RQ1) and examine whether the assassination attempt itself significantly affects public attitudes, independent of the existing political alignments (RQ2). Finally, we explore the major themes in online conversations before and after the crisis, illustrating how discussion topics evolved in response to this politically charged event (RQ3). By integrating large language model-based sentiment analysis, difference-in-differences modeling, and topic modeling techniques, we find that following the attempt the public response was broadly sympathetic to Trump rather than polarizing, despite baseline ideological and regional disparities.

Grace LeFevre, Qingcheng Zeng, Adam Leif et al.

The social impact of Natural Language Processing (NLP) is increasingly important, with a rising community focus on initiatives related to NLP for Social Good (NLP4SG). Indeed, in recent years, almost 20% of all papers in the ACL Anthology address topics related to social good as defined by the UN Sustainable Development Goals (Adauto et al., 2023). In this study, we take an author- and venue-level perspective to map the landscape of NLP4SG, quantifying the proportion of work addressing social good concerns both within and beyond the ACL community, by both core ACL contributors and non-ACL authors. With this approach we discover two surprising facts about the landscape of NLP4SG. First, ACL authors are dramatically more likely to do work addressing social good concerns when publishing in venues outside of ACL. Second, the vast majority of publications using NLP techniques to address concerns of social good are done by non-ACL authors in venues outside of ACL. We discuss the implications of these findings on agenda-setting considerations for the ACL community related to NLP4SG.

Kefan Yu, Qingcheng Zeng, Weihao Xuan et al.

Current large language models (LLMs) have demonstrated emerging capabilities in social intelligence tasks, including implicature resolution and theory-of-mind reasoning, both of which require substantial pragmatic understanding. However, how LLMs acquire this pragmatic competence throughout the training process remains poorly understood. In this work, we introduce ALTPRAG, a dataset grounded in the pragmatic concept of alternatives, to evaluate whether LLMs at different training stages can accurately infer nuanced speaker intentions. Each instance pairs two equally plausible yet pragmatically divergent continuations and requires the model to (i) infer the speaker's intended meaning and (ii) explain when and why a speaker would choose one utterance over its alternative, thus directly probing pragmatic competence through contrastive reasoning. We systematically evaluate 22 LLMs across 3 key training stages: after pre-training, supervised fine-tuning (SFT), and preference optimization, to examine the development of pragmatic competence. Our results show that even base models exhibit notable sensitivity to pragmatic cues, which improves consistently with increases in model and data scale. Additionally, SFT and RLHF contribute further gains, particularly in cognitive-pragmatic scenarios. These findings highlight pragmatic competence as an emergent and compositional property of LLM training and offer new insights for aligning models with human communicative norms.

Haoding Meng, Qingcheng Zeng, Xiaoyang Fang et al.

Automatic or semi-automatic conversion of protocols specifying steps in performing a lab procedure into machine-readable format benefits biological research a lot. These noisy, dense, and domain-specific lab protocols processing draws more and more interests with the development of deep learning. This paper presents our teamwork on WNUT 2020 shared task-1: wet lab entity extract, that we conducted studies in several models, including a BiLSTM CRF model and a Bert case model which can be used to complete wet lab entity extraction. And we mainly discussed the performance differences of \textbf{Bert case} under different situations such as \emph{transformers} versions, case sensitivity that may don't get enough attention before.