The paper investigates why large language models sometimes hallucinate when asked about facts in a language different from the one in which the facts were learned. By training small Transformer models on synthetic multilingual datasets, the authors show that the degree of correlation between facts and their learning language (informativeness) and the ease of language identification (extractability) determine whether models develop unified or separate representations across languages. Unified representations enable cross‑lingual fact transfer, while separate representations do not. The study proposes a unifying perspective on cross‑lingual transfer and suggests training methods to promote representational unification.
By Carter Blum, Katja Filippova, Ann Yuan, Asma Ghandeharioun, Julian Zimmert, Fred Zhang, Jessica Hoffmann, Tal Linzen, Martin Wattenberg, Lucas Dixon, Mor Geva
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.
By Alexander Manev
arXiv:2606. 20089v1 Announce Type: cross Abstract: Persian pretrained language models (PLMs) are still limited by the scarcity of large-scale, high-quality pretraining corpora and by insufficient evaluation beyond standard classification and NER tasks.
By Arash Ghafouri, Mahdi Firouzmandi, Hossein Saberi, Mohammad Reza Hasani Ahangar
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.
By Vihari Piratla, Purvam Jain, Darshan Singh, Trevor Cohn, Preethi Jyothi, Partha Talukdar
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.
By Tarek Naous, Anagha Savit, Carlos Rafael Catalan, Geyang Guo, Jaehyeok Lee, Kyungdon Lee, Lheane Marie Dizon, Mengyu Ye, Neel Kothari, Sahajpreet Singh, Sarah Masud, Tanish Patwa, Trung Thanh Tran, Zohaib Khan, Alan Ritter, Tanmoy Chakraborty, Yuki Arase, Keisuke Sakaguchi, JinYeong Bak, Wei Xu
arXiv:2608.28645v1 Announce Type: cross
Abstract: Low-resource languages without an adequate training corpus often use a related, higher-resource language as a scaffold for comprehension. Still, ther...
By Sindhu Shetty, Spurthi Setty, Natan Vidra
The paper introduces Golden-GRPO Injection (GRIN), a three-stage self‑learning framework that uses a mixed‑policy reinforcement learning algorithm to inject knowledge into large language models. GRIN injects a golden answer to provide learning signals even when on‑policy rollouts fail on novel facts, and is evaluated on two new document‑level benchmarks—Blank and Counter—that test novel acquisition and counterfactual overwrite. Experiments show that mixed‑policy RL enables knowledge absorption beyond what supervised fine‑tuning can achieve, with GRIN outperforming SFT and other RL baselines on harder question types while matching them on basic fact recall.
By Zhibo Hou, Fan Zhao, Zhiyu An, Wan Du
arXiv:2608. 13698v1 Announce Type: cross Abstract: Reinforcement Learning with Verifiable Rewards (RLVR), often optimized with Group Relative Policy Optimization (GRPO), has become a central recipe for improving the reasoning capabilities of pretrained language models but current studies remain heavily English-centric.
By Konstantin Dobler, Federico Scozzafava, Jonathan Janke, Mohamed Ali, Simon Lehnerer
The paper investigates why large reasoning language models struggle to transfer parametric knowledge across different scripts. Through observational data and regression analysis on ECLeKTic and MultiLoKo datasets, the authors find that script mismatch—not language family—is the main predictor of transfer failure when controlling for model capability and question difficulty. By providing key entities in the source language and training models to reason about transliteration ambiguities, they demonstrate a reduction in the cross‑script transfer gap, suggesting that post‑training improvements can enhance cross‑lingual knowledge transfer.
By Lucas Bandarkar, Alan Ansell, Trevor Cohn
arXiv:2603.26236v2 Announce Type: replace
Abstract: While multilingual language models successfully transfer factual and syntactic knowledge across languages, it remains unclear whether they process...
By Uri Z. Kialy, Avi Shtarkberg, Ayal Klein
XHotpotQA is a new benchmark for cross‑lingual knowledge composition in multi‑hop question answering. It presents each instance as an evidence‑dependency graph with explicit language assignments for the question, bridge evidence, answer‑bearing evidence, and distractors, and includes 15,661 training and 7,405 validation examples with sentence‑level support supervision. The dataset reveals significant performance drops when evidence spans language boundaries, providing a diagnostic tool for systems that must integrate evidence across languages.
By Iman Barati, Arash Ghafouri, Behrouz Minaei-Bidgoli
The paper introduces Centroid Intervention Fusion (CIF), a framework that merges multiple multilingual intervention projections into a single language-shared operator for inference-time modification of large language models. CIF improves cross-lingual transfer without updating model parameters and achieves up to +3.378 percentage points better performance than prior pairwise intervention baselines across several benchmarks, including low-resource languages. The authors provide code at https://github.com/VRCMF/CIF.git.
By Wei Sun, Marie-Francine Moens