SWORD is a new benchmark that tests large language models’ ability to reject factually incorrect statements across eight major languages by distorting Wikidata triples. The benchmark reveals that models often perform better on semantically plausible distortions than on random ones, indicating a reliance on distributional familiarity rather than true factual verification. It also shows significant performance drops for East Asian languages, with gaps up to 28 percentage points, highlighting asymmetric multilingual factual reasoning capabilities.
arXiv:2607. 19243v1 Announce Type: cross Abstract: Although Large Language Models (LLMs) demonstrate remarkable multilingual fluency, their internal knowledge representations remain disproportionately biased toward high-resource languages.
arXiv:2510. 15551v2 Announce Type: replace-cross Abstract: Any piece of knowledge is usually expressed in one or a handful of natural languages on the web or in any large corpus.
The paper presents an empirical study of factual errors in human-written text, focusing on corrections in newspaper articles to build a taxonomy of common mistakes such as kanji misconversions and unit errors. It evaluates large language models’ ability to detect these errors, finding that even advanced models like GPT‑5.4 achieve only a 52% word‑level F1 score on synthetic data, underscoring the difficulty of the task. The work highlights the gap in research on factual error detection in human writing compared to LLM hallucinations.
Camellia is a new benchmark that tests cultural bias in large language models (LLMs) across nine Asian languages and six Asian cultures. It contains 19,530 manually annotated entities linked to Asian or Western cultures and 2,173 masked social‑media contexts for these entities. Using Camellia, the authors evaluate four multilingual LLMs on cultural context adaptation, sentiment association, and entity extractive QA, finding that models struggle with cultural adaptation, exhibit differing biases across regions and families, and have difficulty understanding context in some Asian languages.
The paper investigates how adjectival modifiers affect the semantic plausibility of events, using the Adept benchmark of 16,000 English sentence pairs that differ by a single adjective. Experiments show that sentence transformers, despite being conceptually suited to the task, underperform compared to models like RoBERTa. The authors provide an error analysis and discuss the implications of their findings for future work on balancing training and test data.
The study investigates how exposing large language models to Arabic translations of English benchmark tasks can hide data contamination from standard English-only detection methods. By testing four instruction-tuned models with increasing Arabic exposure, the authors find that traditional probes like TS-Guessing and Min-K%++ fail to detect contamination, yet English performance still improves. They propose a new Translation-Aware Contamination Detection (TACD) method that uses cross-lingual consistency and choice reordering to reveal contamination effects that are otherwise invisible to English-only diagnostics.
The paper introduces the first large‑scale benchmark dataset of Bangla idioms, along with a synthetic multiple‑choice question set for idiom meaning identification. It evaluates recent large language models on three idiom‑related tasks—paraphrasing, idiom span detection, and meaning identification—using zero‑shot and few‑shot prompting. Results show significant variability across models, with Phi‑4‑mini‑instruct best at paraphrasing, Kimi‑K2‑32b‑instruct excelling at span detection, and Gemini‑2.5‑flash leading in meaning identification.
The study investigates whether large language models (LLMs) are more prone to errors when they doubt the plausibility of input data, a phenomenon termed context‑memory conflict. Using non‑English and low‑resource language datasets, the authors generate text from factual, counterfactual, and fictional RDF triples in English, Czech, Slovak, and Upper Sorbian, and evaluate faithfulness with both human annotations and an LLM judge (Kimi K3). Contrary to expectations, the results show only a weak context‑memory conflict: counterfactual inputs receive slightly lower faithfulness scores than factual ones, and the choice of LLM judge can significantly affect perceived conflict strength.
The paper investigates whether multilingual language models transfer factual knowledge from one language to another during continued pretraining. Using an English-pretrained model continued on Persian data with systematically removed facts, the authors create SIFT, a dataset of 500 triples across 20 relations, split by cultural origin. Their findings indicate that factual transfer is minimal, especially for Persian-related facts, and that simple removal strategies or easy negative candidate sets can overestimate transfer.
PROOF is a benchmark that profiles the reliability of object-level facts in instruction-tuned language models by converting a frozen Wikidata snapshot into 18,486 English multiple-choice questions grounded in 11,779 semantic facts across 101 classes, 392 properties, and 14 domains. Each question includes an explicit "I don't know" option, a "No correct option" control, and nine controlled formulations, with 1,849 questions designed as no-correct-option traps. The study evaluates 18 open-weight model deployments on 166,374 prompts, revealing wide variability in factual accuracy, sensitivity to wording changes, and the impact of decoder perturbations.
The paper presents a unified evaluation of cross‑lingual consistency (CLC) enhancement methods for multilingual language models, covering inference‑time interventions and post‑training approaches across three model families and three closed‑form benchmarks. Results indicate that post‑training methods, especially direct distribution alignment, consistently improve CLC across all model‑dataset combinations, while other methods are more sensitive to answer format and language coverage. The study also examines the impact of CLC enhancement on culturally diverse question answering, finding no systematic degradation in controlled settings but occasional accuracy drops in open‑ended generation, particularly for non‑English responses.
The paper introduces JSFactCheckBench, a Japanese benchmark that explicitly defines the three key design decisions for evaluating large language model (LLM) fact-checkers: the questions the claims answer, the conditions under which each claim is judged, and the sources considered as evidence. It contains 1,913 claims derived from LLM answers to questions about misleading Japanese social‑media content, each annotated for factuality at a reference time and for importance to the question, using evidence from preselected Web sources. Experiments with six LLMs reveal that models are less accurate on claims that directly answer the question than on supplementary claims, that accuracy drops when models must locate relevant content in longer or multiple sources, and that all four open models perform worse when the surrounding question and response are included—weaknesses that are only apparent because the benchmark records the conditions behind each label.