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.
By Ej Zhou, Lucas Resck, Zheng Hui, Anna Korhonen
The study investigates how the calibration distribution used for Block Influence (BI) pruning of Transformer blocks affects multilingual language models. It finds that English-only calibration underestimates late‑layer importance for Indian languages, leading to suboptimal pruning decisions, while a balanced multilingual calibration mitigates perplexity loss in some models but not universally. The results highlight that pruning strategies must be validated for each architecture and language set rather than relying on a single calibration source.
By Vedant Pandhare, Piyush Dhoka
The paper investigates gender bias in machine translation evaluation metrics using an occupation-balanced subset of GAMBIT+ across seven English‑source language pairs, including a new German extension. It finds that masculine translations tend to receive higher scores and that biases align with stereotypical gender representations, though the strength varies by evaluator and language. The study highlights that assessing bias requires multiple dimensions beyond a single aggregate measure.
By Orfeas Menis Mastromichalakis, Giorgos Filandrianos, Wafaa Mohammed, Giuseppe Attanasio, Chrysoula Zerva
arXiv:2604. 04532v2 Announce Type: replace-cross Abstract: Evaluation language is typically treated as a fixed English default in agentic code benchmarks, yet we show that changing the judge's language can invert backbone rankings.
By Alhasan Mahmood, Samir Abdaljalil, Hasan Kurban
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.
By Tom\'a\v{s} Burkert, Angelika Peljak-{\L}api\'nska, David Zelen\'y
Teochew has a substantial speaker community and exhibits distinctive lexical, syntactic, and pragmatic features, yet textual resources for evaluating large language models remain limited. We present T...
arXiv:2609.18156v1 Announce Type: new
Abstract: Teochew has a substantial speaker community and exhibits distinctive lexical, syntactic, and pragmatic features, yet textual resources for evaluating l...
By Jianan Wu
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.
By Chaymaa Abbas, Nour Shammaa, Mariette Awad
When a tool-using agent is given the same task in a different language, does it still take the same steps? Multilingual evaluation rarely asks: it compares final answers and discards the actions.
arXiv:2608. 14896v1 Announce Type: cross Abstract: Large language models work well on English and behave in poorly understood ways on languages typologically far from it.
By Florian Braun
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...
By Debajyoti Mazumder, Divyansh Pathak, Prashant Kodali, Aditya Joshi, Akshay Agarwal, Jasabanta Patro
The paper introduces a calibrated instrument for rigorously measuring how inference optimizations—such as quantization, early‑exit, and speculative decoding—affect the output quality of large language models. It uses a formally calibrated LLM judge that verifies no systematic bias between statistically equivalent outputs and includes a null condition to ensure measured differences are zero. Applying this method, the authors find that a 4‑bit model is indistinguishable from its 16‑bit counterpart, while 3‑bit quantization and early‑exit techniques incur measurable quality losses that vary by language and task, and that token‑certainty‑based acceptance rules cannot reliably identify impactful errors.
By Jerry Kaplan