The paper introduces Multilingual GSM‑Symbolic, a large multilingual math dataset with 30,000 question‑answer pairs across 15 languages, designed to evaluate cross‑lingual capability transfer using symbolic templates that generate millions of variations. It quantifies key determinants of transfer—model size, language resource level, reasoning ability, and typological distance—showing that larger models and better reasoning reduce performance gaps between low‑ and high‑resource languages, while typological distance has limited impact. The authors present a predictive framework that explains 92 % of language‑to‑language variation and can estimate unseen language performance within 6 pp, improving to 4.2 pp with only ten target‑language templates.
arXiv:2608. 03803v1 Announce Type: cross Abstract: Multilingual language models are deployed across a hundred or more languages, yet most benchmarks test whether a model can perform a task _in_ a language rather than whether it commands the language itself, conflating fluency with proficiency.
arXiv:2605.29637v2 Announce Type: replace Abstract: Large language models often exhibit a substantial gap between their performance in English and in lower-resourced languages on equivalent knowledge...
The paper re‑examines the GSM‑Symbolic benchmark, arguing that its claim of widespread reasoning failures in 25 LLMs is based on weak statistics. Using bootstrapped Generalised Linear Mixed Models on 20 open‑weight models, only eight show significant performance changes, and a systematic shift toward larger integers in the dataset explains many of these effects. The authors also uncover model‑specific failure modes such as variable binding fragility, arithmetic limits, and dual‑task interference, cautioning against blanket conclusions about LLM reasoning.
arXiv:2609.39640v1 Announce Type: cross Abstract: Cross-lingual transfer describes how knowledge in a source language benefits a target language. Measuring it quantitatively requires broad multilingu...
The study investigates how the choice of tokenizer influences multilingual language models across 54 tokenizers and 123 models. It finds that tokenizer impact is greater for languages with less training data, and that excluding a language from tokenizer training consistently worsens its performance. While reallocating tokenizer training data to lower-resource languages can help, it does not guarantee improvement, and the best tokenizer properties vary by language, enabling predictive screening of tokenizer candidates.
arXiv:2609.13847v1 Announce Type: new Abstract: In this paper, we extend FLORES+ with Portuguese-source evaluation sets for three Mozambican Bantu varieties: Xichangana, Mozambican Nyanja, and Sena....
arXiv:2608.30092v1 Announce Type: cross Abstract: We present Arkios, a 1.04B-parameter dense transformer pretrained from scratch on 150B tokens of bilingual English-Nepali text, using a custom single...
The paper demonstrates that large language model (LLM) evaluators, whether reward‑model based or prompted LLM‑as‑a‑Judge, exhibit significant language bias in multilingual settings. Experiments with semantically identical instruction‑response pairs across 23 languages reveal that lower‑resource languages receive higher scores, a bias that persists across eight open‑weight evaluators and is not detectable by standard pairwise accuracy metrics. The authors link the bias to model uncertainty and language identity, showing it cannot be explained by content difficulty alone.
arXiv:2608. 15964v1 Announce Type: cross Abstract: Language-specific competency (LSC) is the phenomenon of a language model performing better or worse depending on the language of the prompt.
The paper shows that HuggingFace’s ByteLevel pre‑tokenizer, which treats a word as a sequence of Unicode letters, splits abugida scripts at every vowel sign, creating a training‑free lower bound on tokenizer fertility. Across 26 languages, all 17 abugidas exhibit increased token counts (up to 9×), while Latin, Cyrillic, Hangul, and Han remain unchanged. The authors demonstrate that correcting the character class reduces Nepali token counts, improves model performance, and that this issue is widespread in popular HuggingFace models.
arXiv:2610.09152v1 Announce Type: new Abstract: Multilingual LLM benchmarks omit Slovak, a morphologically rich West Slavic language of five million speakers, or cover it only by machine translation....
The paper investigates how to choose language models (teachers) for generating multilingual synthetic data used to fine‑tune smaller student models. By evaluating 10 teacher models across six diverse languages and training 240 students, the authors find that teacher effectiveness is not driven by model size but by data qualities such as prompt diversity, length, and fluency, which explain most of the variance in student performance. Practical guidelines are offered, including matching teacher and student families and using translated prompts to improve outcomes for low‑resource languages.