Large Language Models (LLMs) are increas- ingly integral to information dissemination and decision-making processes. Given their grow- ing societal influence, understanding potential biases, particularly within the political domain, is crucial to prevent undue influence on public opinion and democratic processes. This work investigates political bias and stereotype propa- gation across eight prominent LLMs using the two-dimensional Political Compass Test (PCT). Initially, the PCT is employed to assess the in- herent political leanings of these models. Sub- sequently, persona prompting with the PCT is used to explore explicit stereotypes across vari- ous social dimensions. In a final step, implicit stereotypes are uncovered by evaluating mod- els with multilingual versions of the PCT. Key findings reveal a consistent left-leaning polit- ical alignment across all investigated models. Furthermore, while the nature and extent of stereotypes vary considerably between models, implicit stereotypes elicited through language variation are more pronounced than those iden- tified via explicit persona prompting. Interest- ingly, for most models, implicit and explicit stereotypes show a notable alignment, suggest- ing a degree of transparency or "awareness" regarding their inherent biases. This study un- derscores the complex interplay of political bias and stereotypes in LLMs.

Scientific knowledge discovery increasingly relies on large language models, yet many existing scholarly assistants depend on proprietary systems with tens or hundreds of billions of parameters. Such reliance limits reproducibility and accessibility for the research community. In this work, we ask a simple question: do we need bigger models for scientific applications? Specifically, we investigate to what extent carefully designed retrieval pipelines can compensate for reduced model scale in scientific applications. We design a lightweight retrieval-augmented framework that performs task-aware routing to select specialized retrieval strategies based on the input query. The system further integrates evidence from full-text scientific papers and structured scholarly metadata, and employs compact instruction-tuned language models to generate responses with citations. We evaluate the framework across several scholarly tasks, focusing on scholarly question answering (QA), including single- and multi-document scenarios, as well as biomedical QA under domain shift and scientific text compression. Our findings demonstrate that retrieval and model scale are complementary rather than interchangeable. While retrieval design can partially compensate for smaller models, model capacity remains important for complex reasoning tasks. This work highlights retrieval and task-aware design as key factors for building practical and reproducible scholarly assistants.

This paper presents a systematic investigation into the constrained generation capabilities of large language models (LLMs) in producing Songci, a classical Chinese poetry form characterized by strict structural, tonal, and rhyme constraints defined by Cipai templates. We first develop a comprehensive, multi-faceted evaluation framework that includes: (i) a formal conformity score, (ii) automated quality assessment using LLMs, (iii) human evaluation, and (iv) classification-based probing tasks. Using this framework, we evaluate the generative performance of 18 LLMs, including 3 proprietary models and 15 open-source models across 4 families, under five prompting strategies: zero-shot, one-shot, completion-based, instruction-based, and chain-of-thought. Finally, we propose a Generate-Critic architecture in which the evaluation framework functions as an automated critic. Leveraging the critic’s feedback as a scoring function for best-of-N selection, we fine-tune 3 lightweight open-source LLMs via supervised fine-tuning (SFT), resulting in improvements of up to 5.88% in formal conformity. Our findings offer new insights into the generative strengths and limitations of LLMs in producing culturally significant and formally constrained literary texts.

We present SemRepo, an RDF knowledge graph comprising over 81 million triples describing nearly 200,000 GitHub repositories associated with scientific research. SemRepo captures repository-level metadata, such as contributors, issues, and programming languages, and interlinks this information with external scholarly knowledge graphs. In particular, repository authors are linked to their profiles in SemOpenAlex, repositories are connected to scholarly publications in LPWC, and research artifacts, such as datasets and experiments, are linked via MLSea-KG. This integration enables queries that span publications and their scholarly artifacts, which are typically fragmented across separate platforms. SemRepo supports analyses that are difficult to perform with existing resources in isolation, including provenance reconstruction across repositories and publications, as well as the systematic identification of risks to research reproducibility and software sustainability. By unifying research software with its scholarly context in a single graph, SemRepo provides an important infrastructure for large-scale analysis of software within the broader scientific research ecosystem.

The increasing adoption of large language models (LLMs) has raised serious concerns about their reliability and trustworthiness. As a result, a growing body of research focuses on evidence-based text generation with LLMs, aiming to link model outputs to supporting evidence to ensure traceability and verifiability. However, the field is fragmented due to inconsistent terminology, isolated evaluation practices, and a lack of unified benchmarks. To bridge this gap, we systematically analyze 134 papers, introduce a unified taxonomy of evidence-based text generation with LLMs, and investigate 300 evaluation metrics across seven key dimensions. Thereby, we focus on approaches that use citations, attribution, or quotations for evidence-based text generation. Building on this, we examine the distinctive characteristics and representative methods in the field. Finally, we highlight open challenges and outline promising directions for future work.

Graph Neural Networks (GNNs) have substantially advanced the field of recommender systems. However, despite the creation of more than a thousand knowledge graphs (KGs) under the W3C standard RDF, their rich semantic information has not yet been fully leveraged in GNN-based recommender systems. To address this gap, we propose a comprehensive integration of RDF KGs with GNNs that utilizes both the topological information from RDF object properties and the content information from RDF datatype properties. Our main focus is an in-depth evaluation of various GNNs, analyzing how different semantic feature initializations and types of graph structure heterogeneity influence their performance in recommendation tasks. Through experiments across multiple recommendation scenarios involving multi-million-node RDF graphs, we demonstrate that harnessing the semantic richness of RDF KGs significantly improves recommender systems and lays the groundwork for GNN-based recommender systems for the Linked Open Data cloud. The code and data are available on our GitHub repository: https://github.com/davidlamprecht/rdf-gnn-recommendation

Scientific claims shared on social media are often difficult to verify and may contribute to the spread of misinformation. To address this challenge, automated fact verification systems require scientific claim-source retrieval, the task of identifying the source publication underlying a given claim. However, claims often differ substantially from their source publications in language, style, and specificity, making retrieval challenging. We present a comparative study of scientific claim-source retrieval on the CheckThat! 2026 benchmark across sparse and dense retrieval models. Our results show that translating claims into English outperforms both original and bilingual claim representations, while incorporating publication metadata provides additional retrieval gains by capturing indirect source references. In addition, we analyze four style transfer approaches and find that they improve retrieval performance for most models, although the optimal style depends on the underlying retrieval objective. Finally, we investigate similarity- and signal-based re-ranking approaches, introducing three novel re-ranking models based on attribution, entity overlap, and verification-based reasoning. Verification-based re-ranking yields additional gains beyond semantic similarity and achieves the best overall performance with an MRR@5 of 0.758.

Large Language Models (LLMs) are commonly used in Question Answering (QA) settings, increasingly in the natural sciences if not science at large. Reliable Uncertainty Quantification (UQ) is critical for the trustworthy uptake of generated answers. Existing UQ approaches remain weakly validated in scientific QA, a domain relying on fact-retrieval and reasoning capabilities. We introduce the first large-scale benchmark for evaluating UQ metrics in reasoning-demanding QA studying calibration of UQ methods, providing an extensible open-source framework to reproducibly assess calibration. Our study spans up to 20 large language models of base, instruction-tuned and reasoning variants. Our analysis covers seven scientific QA datasets, including both multiple-choice and arithmetic question answering tasks, using prompting to emulate an open question answering setting. We evaluate and compare methods representative of prominent approaches on a total of 685,000 long-form responses, spanning different reasoning complexities representative of domain-specific tasks. At the token level, we find that instruction tuning induces strong probability mass polarization, reducing the reliability of token-level confidences as estimates of uncertainty. Models further fine-tuned for reasoning are exposed to the same effect, but the reasoning process appears to mitigate it depending on the provider. At the sequence level, we show that verbalized approaches are systematically biased and poorly correlated with correctness, while answer frequency (consistency across samples) yields the most reliable calibration. In the wake of our analysis, we study and report the misleading effect of relying exclusively on ECE as a sole measure for judging performance of UQ methods on benchmark datasets. Our findings expose critical limitations of current UQ methods for LLMs and standard practices in benchmarking thereof.

While large language models (LLMs) have increasingly been applied to hate speech detoxification, the prompts often trigger safety alerts, causing LLMs to refuse the task. In this study, we systematically investigate false refusal behavior in hate speech detoxification and analyze the contextual and linguistic biases that trigger such refusals. We evaluate nine LLMs on both English and multilingual datasets, our results show that LLMs disproportionately refuse inputs with higher semantic toxicity and those targeting specific groups, particularly nationality, religion, and political ideology. Although multilingual datasets exhibit lower overall false refusal rates than English datasets, models still display systematic, language-dependent biases toward certain targets. Based on these findings, we propose a simple cross-translation strategy, translating English hate speech into Chinese for detoxification and back, which substantially reduces false refusals while preserving the original content, providing an effective and lightweight mitigation approach.

Multilingual scientific claim-source retrieval aims to identify the scientific publication supporting a claim shared on social media. This task is challenging because claims often differ from source publications in terms of language, wording, and level of detail, which weakens the connection between claims and their underlying evidence. In this paper, we present our approach for the CheckThat! 2026 Lab Task 1: Source Retrieval for Scientific Web Claims. We propose a multi-stage retrieval framework for multilingual scientific claim-source retrieval that combines structured claim and source representations with progressive candidate refinement. To address multilingual retrieval challenges, the framework employs bilingual claim representations, metadata-enhanced source representations, and language-specific adaptation of dense retrieval models. Building on this setup, a first-stage retriever generates an initial pool of candidate sources, after which similarity-based re-ranking improves the ranking of highly relevant sources and verification-based re-ranking identifies the candidate source that best supports the claim using verification signals. Our approach achieves an average MRR@5 score of 0.7628 across English, German, and French claims, ranking first on the CheckThat! 2026 leaderboard.

Large Language Models (LLMs) are increasingly employed as AI tutors due to their scalability and potential for personalized instruction. However, off-the-shelf LLMs often underperform in educational settings: they frequently reveal answers too readily, fail to adapt their responses to student uncertainty, and remain vulnerable to emotionally manipulative prompts. To address these challenges, we introduce CoDAE, a framework that adapts LLMs for educational use through Chain-of-Thought (CoT) data augmentation. We collect real-world dialogues between students and a ChatGPT-based tutor and enrich them using CoT prompting to promote step-by-step reasoning and pedagogically aligned guidance. Furthermore, we design targeted dialogue cases to explicitly mitigate three key limitations: over-compliance, low response adaptivity, and threat vulnerability. We fine-tune four open-source LLMs on different variants of the augmented datasets and evaluate them in simulated educational scenarios using both automatic metrics and LLM-as-a-judge assessments. Our results show that models fine-tuned with CoDAE deliver more pedagogically appropriate guidance, better support reasoning processes, and effectively resist premature answer disclosure.

We study how large language models recall relational knowledge during text generation, with a focus on identifying latent representations suitable for relation classification via linear probes. Prior work shows how attention heads and MLPs interact to resolve subject, predicate, and object, but it remains unclear which representations support faithful linear relation classification and why some relation types are easier to capture linearly than others. We systematically evaluate different latent representations derived from attention head and MLP contributions, showing that per-head attention contributions to the residual stream are comparatively strong features for linear relation classification. Feature attribution analyses of the trained probes, as well as characteristics of the different relation types, reveal clear correlations between probe accuracy and relation specificity, entity connectedness, and how distributed the signal on which the probe relies is across attention heads. Finally, we show how token-level feature attribution of probe predictions can be used to reveal probe behavior in further detail.

Inferring causal relationships between variable pairs is crucial for understanding multivariate interactions in complex systems. Knowledge-based causal discovery -which involves inferring causal relationships by reasoning over the metadata of variables (e.g., names or textual context)-offers a compelling alternative to traditional methods that rely on observational data. However, existing methods using Large Language Models (LLMs) often produce unstable and inconsistent results, compromising their reliability for causal inference. To address this, we introduce a novel approach that integrates Knowledge Graphs (KGs) with LLMs to enhance knowledge-based causal discovery. Our approach identifies informative metapath -based subgraphs within KGs and further refines the selection of these subgraphs using Learning-to-Rank-based models. The top-ranked subgraphs are then incorporated into zero-shot prompts, improving the effectiveness of LLMs in inferring the causal relationship. Extensive experiments on biomedical and open-domain datasets demonstrate that our method outperforms most baselines by up to 44.4 points in F1 scores, evaluated across diverse LLMs and KGs. Our code and datasets are available on GitHub. https://github.com/susantiyuni/path-to-causality

Large Language Models (LLMs) exhibit strong linguistic capabilities, but little is known about how they encode psycholinguistic knowledge across languages. We investigate whether and how LLMs exhibit human-like psycholinguistic responses under different linguistic identities using two tasks: sound symbolism and word valence. We evaluate two models, Llama-3.3-70B-Instruct and Qwen2.5-72B-Instruct, under monolingual and bilingual prompting in English, Dutch, and Chinese. Behaviorally, both models adjust their outputs based on prompted language identity, with Qwen showing greater sensitivity and sharper distinctions between Dutch and Chinese. Probing analysis reveals that psycholinguistic signals become more decodable in deeper layers, with Chinese prompts yielding stronger and more stable valence representations than Dutch. Our results demonstrate that language identity conditions both output behavior and internal representations in LLMs, providing new insights into their application as models of cross-linguistic cognition.

Hate speech detection is a socially sensitive and inherently subjective task, with judgments often varying based on personal traits. While prior work has examined how socio-demographic factors influence annotation, the impact of personality traits on Large Language Models (LLMs) remains largely unexplored. In this paper, we present the first comprehensive study on the role of persona prompts in hate speech classification, focusing on MBTI-based traits. A human annotation survey confirms that MBTI dimensions significantly affect labeling behavior. Extending this to LLMs, we prompt four open-source models with MBTI personas and evaluate their outputs across three hate speech datasets. Our analysis uncovers substantial persona-driven variation, including inconsistencies with ground truth, inter-persona disagreement, and logit-level biases. These findings highlight the need to carefully define persona prompts in LLM-based annotation workflows, with implications for fairness and alignment with human values.

Cross-lingual Named Entity Recognition (NER) leverages knowledge transfer between languages to identify and classify named entities, making it particularly useful for low-resource languages. We show that the data-based cross-lingual transfer method is an effective technique for crosslingual NER and can outperform multilingual language models for low-resource languages. This paper introduces two key enhancements to the annotation projection step in cross-lingual NER for low-resource languages. First, we explore refining word alignments using back-translation to improve accuracy. Second, we present a novel formalized projection approach of matching source entities with extracted target candidates. Through extensive experiments on two datasets spanning 57 languages, we demonstrated that our approach surpasses existing projectionbased methods in low-resource settings. These findings highlight the robustness of projection-based data transfer as an alternative to model-based methods for crosslingual named entity recognition in lowresource languages.

Large language models (LLMs) frequently produce false refusals, declining benign requests that contain terms resembling unsafe queries. We address this challenge by introducing two comprehensive benchmarks: the Exaggerated Safety Benchmark (XSB) for single-turn prompts, annotated with "Focus" keywords that identify refusal-inducing triggers, and the Multi-turn Scenario-based Exaggerated Safety Benchmark (MS-XSB), which systematically evaluates refusal calibration in realistic, context-rich dialog settings. Our benchmarks reveal that exaggerated refusals persist across diverse recent LLMs and are especially pronounced in complex, multi-turn scenarios. To mitigate these failures, we leverage post-hoc explanation methods to identify refusal triggers and deploy three lightweight, model-agnostic approaches, ignore-word instructions, prompt rephrasing, and attention steering, at inference time, all without retraining or parameter access. Experiments on four instruction-tuned Llama models demonstrate that these strategies substantially improve compliance on safe prompts while maintaining robust safety protections. Our findings establish a reproducible framework for diagnosing and mitigating exaggerated refusals, highlighting practical pathways to safer and more helpful LLM deployments.