论文检索

Adam Štorek, Mukur Gupta, Samira Hajizadeh, Prashast Srivastava, Suman Jana

Large language models (LLMs) are increasingly deployed for understanding large codebases, but whether they understand operational semantics of long code context or rely on pattern matching shortcuts remains unclear. We distinguish between lexical recall (retrieving code verbatim) and semantic recall (understanding operational semantics). Evaluating 10 state-of-the-art LLMs, we find that while frontier models achieve near-perfect, position-independent lexical recall, semantic recall degrades severely when code is centrally positioned in long contexts. We introduce semantic recall sensitivity to measure whether tasks require understanding of code’s operational semantics vs. permit pattern matching shortcuts. Through a novel counterfactual measurement method, we show that models rely heavily on pattern matching shortcuts to solve existing code understanding benchmarks. We propose a new task SemTrace, which achieves high semantic recall sensitivity through unpredictable operations; LLMs’ accuracy exhibits severe positional effects, with median accuracy drops of 92.73% versus CRUXEval’s 53.36% as the relevant code snippet approaches the middle of the input code context. Our findings suggest current evaluations substantially underestimate semantic recall failures in long context code understanding.

Jie He, Nan Hu, Wanqiu Long, Jiaoyan Chen, Jeff Z. Pan

Retrieval-Augmented Generation (RAG) enhances large language models (LLMs) by integrating external knowledge, enabling them to tackle knowledge-intensive tasks. However, limited research has explored how LLMs effectively leverage RAG techniques for multi-hop question answering (QA), particularly when handling knowledge with with varying degrees of familiarity. In this paper, we introduce MINTQA (Multi-hop Question Answering on New and Tail Knowledge), a benchmark designed to evaluate multi-hop QA performance on questions involving 10,479 question-answer pairs for evaluating old/new knowledge and 17,887 pairs for assessing popular/unpopular knowledge, with each question equipped with corresponding sub-questions and answers. This benchmark primarily evaluates the multi-hop reasoning ability of LLMs and their capacity to handle knowledge with varying levels of familiarity during the reasoning process. We evaluate 22 state-of-the-art LLMs using three distinct QA strategies: LLM-based parameterized knowledge QA, direct RAG-enhanced QA, and multi-hop RAG-enhanced QA. Our experiments reveal key challenges in how LLMs handle knowledge with different familiarity and offer insights into improving their multi-hop reasoning capabilities when combined with RAG techniques.

Niclas Doll, Jasper Schulze Buschhoff, Shalaka Satheesh, Hammam Abdelwahab, Héctor Allende-Cid, Katrin Klug

This paper narrows the performance gap between small, specialized models and significantly larger general-purpose models through domain adaptation via continual pre-training and merging. We address the scarcity of specialized non-English data by constructing a high-quality German medical corpus (FineMed-de) from FineWeb2. This corpus is used to continually pre-train and merge three well-known LLMs (ranging from 7B to 24B parameters), creating the DeFineMed model family. A comprehensive evaluation confirms that specialization dramatically enhances 7B model performance on German medical benchmarks. Furthermore, the pairwise win-rate analysis of the Qwen2.5-based models demonstrates an approximately 3.5-fold increase in the win-rate against the much larger Mistral-Small-24B-Instruct through domain adaptation. This evidence positions specialized 7B models as a competitive, resource-efficient solution for complex medical instruction-following tasks. While model merging successfully restores instruction-following abilities, a subsequent failure mode analysis reveals inherent trade-offs, including the introduction of language mixing and increased verbosity, highlighting the need for more targeted fine-tuning in future work. This research provides a robust, compliant methodology for developing specialized LLMs, serving as the foundation for practical use in German-speaking healthcare contexts.

Lukas Helff, Ahmad Omar, Felix Friedrich, Antonia Wüst, Hikaru Shindo, Rupert Mitchell, Tim Woydt, Patrick Schramowski, Wolfgang Stammer, Kristian Kersting

We introduce SLR, an end-to-end framework for systematic evaluation and training of Large Language Models (LLMs) via Scalable Logical Reasoning. Given a user’s task specification, SLR automatically synthesizes (i) an instruction prompt for an inductive reasoning task, (ii) a validation program, executable on model outputs to provide verifiable rewards, and (iii) the latent ground-truth rule. This process is fully automated, scalable, requires no human annotations, and offers precise control over task difficulty. Using SLR, we create SLR-Bench, a benchmark comprising 19k prompts organized into 20 curriculum levels that progressively increase in relational, arithmetic, and recursive complexity. Large-scale evaluation reveals that contemporary LLMs readily produce syntactically valid rules, yet often fail at correct logical inference. Recent reasoning LLMs demonstrate improved performance but incur very high test-time computation, with costs exceeding 300 for just 1,000 prompts. Finally, curriculum learning via SLR doubles Llama-3-8B accuracy on SLR-Bench, achieving parity with Gemini-Flash-Thinking at a fraction of computational cost. Moreover, these reasoning capabilities generalize to a wide range of established benchmarks, underscoring the effectiveness of SLR for downstream reasoning.

Jiho Choi, Seojeong Park, Seongjong Song, Hyunjung Shim

Automating scientific poster generation requires hierarchical document understanding and coherent content-layout planning.Existing methods often rely on flat summarization or optimize content and layout separately.As a result, they often suffer from information loss, weak logical flow, and poor visual balance.We present PosterForest, a training-free framework for scientific poster generation.Our method introduces the Poster Tree, a structured intermediate representation that captures document hierarchy and visual-textual semantics across multiple levels.Building on this representation, content and layout agents perform hierarchical reasoning and recursive refinement, progressively optimizing the poster from global organization to local composition.This joint optimization improves semantic coherence, logical flow, and visual harmony.Experiments show that PosterForest outperforms prior methods in both automatic and human evaluations, without additional training or domain-specific supervision.

Chenming Tang, Yutong Yang, Kexue Wang, Yunfang Wu

Knowledge editing aims to modify outdated knowledge in language models efficiently while retaining their original capabilities. Mainstream datasets for knowledge editing are predominantly static and fail to keep in pace with the evolving real-world knowledge. In this work, we introduce CRAFT, an ever-evolving real-world dataset for knowledge editing. It evaluates models on temporal locality, common-sense locality, composite portability and alias portability, providing a comprehensive and challenging evaluation for knowledge editing, on which previous methods hardly achieve balanced performance. Towards flexible real-time knowledge editing, we propose KEDAS, a novel paradigm of knowledge editing alignment featuring diverse edit augmentation and self-adaptive post-alignment inference, exhibiting significant performance gain on both CRAFT and traditional datasets compared to previous methods. We hope this work may serve as a catalyst for shifting the focus of knowledge editing from static update to dynamic evolution.

Seungyoon Lee, Minhyuk Kim, Seongtae Hong, Youngjoon Jang, Dongsuk Oh, Heuiseok Lim

Existing multilingual embedding models often encounter challenges in cross-lingual scenarios due to imbalanced linguistic resources and less consideration of cross-lingual alignment during training. Although standardized contrastive learning approaches for cross-lingual adaptation are widely adopted, they may struggle to capture fundamental alignment between languages and degrade performance in well-aligned languages such as English. To address these challenges, we propose Cross-Lingual Enhancement in RetrievAl via Reverse-training (CLEAR), a novel loss function utilizing a reverse training scheme to improve retrieval performance across diverse cross-lingual retrieval scenarios. CLEAR leverages an English passage as a bridge to strengthen alignments between the target language and English, ensuring robust performance in the cross-lingual retrieval task. Our extensive experiments demonstrate that CLEAR achieves notable improvements in cross-lingual scenarios, with gains up to 15%, particularly in low-resource languages, while minimizing performance degradation in English. Furthermore, our findings highlight that CLEAR offers promising effectiveness even in multilingual training, suggesting its potential for broad application and scalability. We release the code at https://github.com/dltmddbs100/CLEAR.

Qianli Wang, Van Bach Nguyen, Yihong Liu, Fedor Splitt, Nils Feldhus, Christin Seifert, Hinrich Schuetze, Sebastian Möller, Vera Schmitt

Counterfactuals refer to minimally edited inputs that cause a model’s prediction to change, serving as a promising approach to explaining the model’s behavior. Large language models (LLMs) excel at generating English counterfactuals and demonstrate multilingual proficiency. However, their effectiveness in generating multilingual counterfactuals remains unclear. To this end, we conduct a comprehensive study on multilingual counterfactuals. We first conduct automatic evaluations on both directly generated counterfactuals in the target languages and those derived via English translation across six languages. Although translation-based counterfactuals offer higher validity than their directly generated counterparts, they demand substantially more modifications and still fall short of matching the quality of the original English counterfactuals. Second, we find the patterns of edits applied to high-resource European-language counterfactuals to be remarkably similar, suggesting that cross-lingual perturbations follow common strategic principles. Third, we reveal that multilingual counterfactual data augmentation (CDA) yields larger model performance improvements than cross-lingual CDA, especially for lower-resource languages. Yet, the imperfections of the generated counterfactuals limit gains in model performance and robustness. Finally, we identify and categorize four main types of errors that consistently appear in the generated counterfactuals across languages.

Yizhen Yuan, Rui Kong, Dongze Li, Yuanchun Li, Yunxin Liu

Large language models (LLMs) integrated with retrieval-augmented generation (RAG) have become a dominant framework for building intelligent assistants. In real-world applications such as ChatGPT with web search, the retrieved document often comes from diverse, potentially unreliable sources and may contain inconsistent claims. Unlike traditional search engines that rely on users to manually compare information, LLM-based systems typically feed all retrieved content into the model’s context, requiring LLMs to autonomously identify, differentiate, and reason over conflicting viewpoints. Unlike mainstream LLM evaluation tasks like math and code generation that are primarily focused on reasoning with factual context, question-answering with multi-source references requires fundamentally different capabilities to identify and reason over knowledge contradictions. In this paper, we introduce ConfRAG, a benchmark for evaluating LLMs’ reasoning capability over real-world conflicting documents retrieved from the web. It consists of 1,814 real-world questions, each paired with an average of 9.58 retrieved paragraphs from heterogeneous online sources. A total of 57.2% of the questions exhibit explicit contradictions. We further propose three structured evaluation tasks, answer clustering, answer coverage, and reason coverage, to quantify a model’s ability to organize and explain contradictory content. Experiments with state-of-the-art models such as GPT-4.1 and Claude-3-7-Sonnet reveal substantial performance gaps, highlighting the need for more targeted research in contradiction-aware question answering. To the best of our knowledge, ConfRAG is the first benchmark specifically designed to evaluate contradiction-aware reasoning on real-world long web documents.

Myunghoon Kang, Dahyun Jung, Suhyune Son, Seonmin Koo, Changwoo Chun, Daniel Rim, Haeyoung Kwon, Yuna Hur, Heuiseok Lim

Recent advancements in table-based question answering (table QA) have been driven by the development of table-specific reasoning strategies for leveraging large language models. Previous works employ sub-table-based reasoning, which involves matching query-relevant table values and aggregating them into sub-tables for precise reasoning. However, these approaches are limited to scenarios with query-relevant single tables, failing to handle real-world table QA settings that involve noisy multi-table sets. To address the challenges of real-world table QA, we propose **EASE**: **E**ntity-**A**ware **S**ub-table Generation for R**E**al-world Multi-table QA framework. Given a noisy multi-table set, EASE first extracts key entities from the question to construct a sub-table schema. It then populates this schema by utilizing a selected set of column values from the noisy multi-table set, thereby facilitating efficient and effective sub-table-based reasoning. We introduce a Noisy Multi-table QA dataset and conduct extensive experiments to evaluate EASE’s effectiveness on real-world table QA. Our results demonstrate that EASE effectively filters out irrelevant information while incorporating pertinent table values, leading to efficient and effective performance on real-world table QA. Our dataset can be found https://github.com/Metalchaos8527/ease_noisy_multi-table_qa.git

Zhiyuan Fan, Guanqiao Chen, Yanyi Huang, Mingkuan Zhao, Dadi Guo, Yi R. Fung

Large language models (LLMs) have shown strong performance on hard reasoning and general instruction-following tasks. However, when sampling multiple outputs for the same prompt, they often produce highly homogeneous, repetitive responses, resulting in inefficient exploration. This limits the gains from test-time scaling and constrains the upper bound of RL training. We attribute this issue in part to supervised fine-tuning (SFT): when a single prompt is paired with multiple reference responses, the model is trained to generate diverse outputs under the same prior condition, which induces optimization interference and can lead to diversity collapse. To address this, we propose Prefix-Conditioned SFT (P-SFT), a simple yet effective method that constructs semantically consistent yet distributionally distinct prior contents to different responses, thereby projecting the instruction into distinct latent regions to establish diverse prior distributions and decouple the one-to-many mapping. Experiments on large reasoning language models show that our approach improves absolute performance by 5.3% and increases generation diversity by 198.3% on average, while substantially enhancing output diversity and test-time scaling. Notably, even without any additional training, our prefixing strategy can be applied at inference time alone and still yields significant gains in both diversity and reasoning performance for instruction-tuned LLMs and reasoning-enhanced models.

Mengshi Chen, Yuxiang Sun, Tengchao Li, Jianwei Wang, Kai Wang, Xuemin Lin, Ying Zhang, Wenjie Zhang

Data preparation is a critical step in enhancing the usability of tabular data and thus boosts downstream data-driven tasks. Traditional methods often face challenges in capturing the intricate relationships within tables and adapting to the tasks involved. Recent advances in Language Models (LMs), especially in Large Language Models (LLMs), offer new opportunities to automate and support tabular data preparation. However, why LMs suit tabular data preparation (i.e., how their capabilities match task demands) and how to use them effectively across phases still remain to be systematically explored. In this survey, we systematically analyze the role of LMs in enhancing tabular data preparation processes, focusing on four core phases: data acquisition, integration, cleaning, and transformation. For each phase, we present an integrated analysis of how LMs can be combined with other components for different preparation tasks, highlight key advancements, and outline prospective pipelines.

Daria Kryvosheieva, Andrea Gregor de Varda, Evelina Fedorenko, Greta Tuckute

Large language models (LLMs) can reliably distinguish grammatical from ungrammatical sentences, but how grammatical knowledge is represented within the model remains an open question. We investigate whether different syntactic phenomena recruit shared or distinct components in LLMs. Using a functional localization approach inspired by cognitive neuroscience, we identify the LLM units most responsive to 67 English syntactic phenomena in seven open-weight models. These units are consistently recruited across sentence instances and causally support model performance. Critically, different types of syntactic agreement (e.g., subject-verb, anaphor, determiner-noun) recruit overlapping sets of units, suggesting that agreement constitutes a meaningful functional category in LLMs. This pattern holds in Russian and Chinese, and further, across languages: in a cross-lingual analysis of 57 languages, syntactically similar languages share more units for subject-verb agreement. Taken together, these findings reveal that syntactic agreement—a critical marker of syntactic dependencies—constitutes a meaningful category within LLMs’ representational spaces.

Yifu Chen, Shengpeng Ji, Zhengqing Liu, Qian Chen, Wen Wang, Ziqing Wang, Yangzhuo Li, Tianle Liang, Zhou Zhao

Achieving seamless, human-like interaction remains a key challenge for full-duplex spoken dialogue models (SDMs). Reinforcement learning (RL) has substantially enhanced text- and vision-language models, while well-designed reward signals are crucial for the performance of RL. We consider RL a promising strategy to address the key challenge for SDMs. However, a fundamental barrier persists: prevailing automated metrics for assessing interaction quality rely on superficial proxies, such as behavioral statistics or timing-prediction accuracy, failing to provide reliable reward signals for RL. On the other hand, human evaluations, despite their richness, remain costly, inconsistent, and difficult to scale. We tackle this critical barrier by proposing a Dual-Axis Generative Reward Model, which is trained to understand complex interaction dynamics using a detailed taxonomy and an annotated dataset, produces a single score and, crucially, provides separate evaluations for semantic quality and interaction timing. Such dual outputs furnish precise diagnostic feedback for SDMs and deliver a dependable, instructive reward signal suitable for online reinforcement learning. Our model achieves state-of-the-art performance on interaction-quality assessment across a wide spectrum of datasets, spanning synthetic dialogues and complex real-world interactions.

Louie Hong Yao, Vishesh Anand, Yuan Zhuang, Tianyu Jiang

Rhetorical questions are asked not to seek information but to persuade or signal stance. How large language models internally represent them remains unclear. We analyze rhetorical questions in LLM representations using linear probes on two social-media datasets with different discourse contexts, and find that rhetorical signals emerge early and are most stably captured by last-token representations. Rhetorical questions are linearly separable from information-seeking questions within datasets, and remain detectable under cross-dataset transfer, reaching AUROC around 0.7-0.8. However, we demonstrate that transferability does not simply imply a shared representation. Probes trained on different datasets produce different rankings when applied to the same target corpus, with overlap among the top-ranked instances often below 0.2. Qualitative analysis shows that these divergences correspond to distinct rhetorical phenomena: some probes capture discourse-level rhetorical stance embedded in extended argumentation, while others emphasize localized, syntax-driven interrogative acts. Together, these findings suggest that rhetorical questions in LLM representations are encoded by multiple linear directions emphasizing different cues, rather than a single shared direction.

Chung-ju Huang, Huiqiang Zhao, Yuanpeng He, Lijian Li, Wenpin Jiao, Zhi Jin, Peixuan Chen, Leye Wang

The increasing reliance on cloud-hosted Large Language Models (LLMs) exposes sensitive client data, such as prompts and responses, to potential privacy breaches by service providers.Existing approaches fail to ensure privacy, maintain model performance, and preserve computational efficiency simultaneously.To address this challenge, we propose Talaria, a confidential inference framework that partitions the LLM pipeline between a client-verified Confidential Virtual Machine (CVM) and the public cloud to protect client data without compromising the cloud’s model intellectual property or inference quality.The interaction between the CVM and the cloud is secured by our Reversible Masked Outsourcing (ReMO) protocol, which uses a hybrid masking technique to reversibly obscure intermediate data before outsourcing computations.Extensive evaluations show that Talaria can defend against state-of-the-art token inference attacks, reducing token reconstruction accuracy from over 97.5% to an average of 1.34%, all while being a lossless mechanism that guarantees output identical to the original model without significantly decreasing efficiency and scalability.To the best of our knowledge, this is the first work that ensures clients’ prompts and responses remain inaccessible to the cloud, while also preserving model privacy, performance, and efficiency.

Chengwu Liu, Yichun Yin, Ye Yuan, Jiaxuan Xie, Botao Li, Siqi Li, Jianhao Shen, Yan Xu, Lifeng Shang, Ming Zhang

Most ATP benchmarks embed the final answer within the formal statement — a convention we call "Easy Mode" — a design that simplifies the task relative to what human competitors face and may lead to optimistic estimates of model capability.We call the stricter, more realistic setting "Hard Mode": the system must independently discover the answer before constructing a formal proof.To enable Hard Mode research, we make two contributions.First, we release MiniF2F-Hard and FIMO-Hard, expert-reannotated Hard Mode variants of two widely-used ATP benchmarks.Second, we introduce Discover And Prove (DAP), an agentic framework that uses LLM natural-language reasoning with explicit self-reflection to discover answers, then rewrites Hard Mode statements into Easy Mode ones for existing ATP provers.DAP sets the state of the art: on CombiBench it raises solved problems from 7 (previous SOTA, Pass@16) to 10; on PutnamBench it is the first system to formally prove 36 theorems in Hard Mode — while simultaneously revealing that state-of-the-art LLMs exceed 80% answer accuracy on the same problems where formal provers manage under 10%, exposing a substantial gap that Hard Mode benchmarks are uniquely suited to measure.

Jimin Jung, MyoungJin Kim, Jaehyung Seo, Heuiseok Lim

The Plain Writing Act in the United States requires government documents to be written in clear and simple language. However, existing summarization systems struggle to address diverse linguistic and cognitive barriers among general readers. We propose NRLB (No Reader Left Behind), a unified multi-agent framework for plain language summarization that simulates three representative reader groups: elementary school students, non-native speakers, and readers with attention deficits. NRLB integrates template-based planning with an iterative feedback loop guided by simulated readers and domain expert revision to address comprehension barriers such as unknown terms, missing contexts, and confusing sentences. Evaluations across multiple datasets demonstrate consistent improvements in both readability and factuality. Human evaluation further supports these findings, with annotator preference rates ranging from 55% to 76%, highlighting NRLB’s ability to generate summaries that are both faithful to the source and accessible to a wide range of readers.

Pan Lu, Bowen Chen, Sheng Liu, Rahul Thapa, Joseph Boen, James Zou

Solving complex reasoning tasks may involve visual understanding, domain knowledge retrieval, numerical calculation, and multi-step reasoning. Existing methods augment large language models (LLMs) with external tools but are restricted to specialized domains, limited tool types, or require additional training data. In this paper, we introduce OctoTools, a training-free, user-friendly, and easily extensible multi-agent framework designed to tackle complex reasoning across diverse domains. OctoTools introduces standardized tool cards to encapsulate tool functionality, a planner for both high-level and low-level planning, and an executor to carry out tool usage. We validate OctoTools’ generality across 16 diverse tasks (including MathVista, MMLU-Pro, MedQA, and GAIA-Text), achieving substantial average accuracy gains of 9.3% over GPT-4o. Furthermore, OctoTools also outperforms AutoGen, GPT-Functions, and LangChain by up to 10.6% when given the same set of tools. Through comprehensive analysi, ablations, and robustness tests with compact backbones and noisy tool environments, OctoTools demonstrates advantages in task planning, effective tool usage, and multi-step problem solving. Code, demos, and visualization are publicly available at https://octotools.github.io/.

Hongjun Jeong, Minji Kim, Heesoo Jung, Ko Keun Kim, Hogun Park

Knowledge-grounded Question Answering (QA) aims to provide answers to structured queries or natural language questions by leveraging Knowledge Graphs (KGs). Existing approaches are mainly divided into Knowledge Graph Question Answering (KGQA) and Complex Query Answering (CQA). Both approaches have limitations: the first struggles to utilize KG context effectively when essential triplets related to the questions are missing in the given KGs, while the second depends on structured first-order logic queries. To overcome these limitations, we propose a novel framework termed Aqua-QA. Aqua-QAapproximates query graphs from natural language questions, enabling reasoning over KGs. We evaluate Aqua-QA on challenging QA tasks where KGs are incomplete in the context of QA, and complex logical reasoning is required to answer natural language questions. Experimental results on these datasets demonstrate that Aqua-QA outperforms existing methods, showcasing its effectiveness in handling complex reasoning tasks in knowledge-grounded QA settings.