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
arXiv:2610.28066v1 Announce Type: cross
Abstract: Multilingual evaluations often score a model's opening yes or no, a token that also carries model and language answering conventions. We ask when dif...
By Haonan Huang, Yuwei Lu, Chengkun Cai
arXiv:2610.02926v1 Announce Type: new
Abstract: Standard factual benchmarks assume clean monolingual prompts and exact-match scoring, two assumptions that break simultaneously in real-world multiling...
By Riju Marwah, Ritvik Garimella, Khusham Bansal, Atishay Jain, Amit Sheth
arXiv:2607. 20526v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly deployed in settings where fluent but incorrect answers can be costly.
By Matthew ffrench-Constant, Daniel Yang, Xinmeng Huang, Sanyam Kapoor
arXiv:2609.13824v1 Announce Type: new
Abstract: Large language models (LLMs) are increasingly used to evaluate the responses of other language models. This approach, known as LLM-as-a-Judge, is faste...
By Aakash Kumar Tiwari
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
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: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 study investigates how different prompting strategies affect large language model (LLM) judges across multiple languages and tasks. Using a benchmark of 9 prompting strategies, 4 open-weight judges, 27 languages, and 4 tasks, the authors collected over 1 million pointwise scores validated against human judgments. Results show that prompting effectiveness is highly dependent on the judge model, can reverse direction across judges, and becomes less stable in low‑resource languages, indicating that prompt choice must be tailored to the specific evaluation setting.
By Asal Mehrabi, Mohammad Aliannejadi, Clemencia Siro
The paper investigates whether lower stereotype bias scores in multilingual language models truly indicate greater fairness. By auditing ten open models across eight languages, the authors show that lower average scores often arise from translation artifacts rather than genuine bias reduction, as item‑level correlations between English and translated scores are weak. They also identify pipeline errors that can mask bias signals and provide code to replicate their findings.
By Harsh Kumar
The paper investigates whether fine‑tuning improves how language models use supplied Bangladeshi legal text in bilingual question‑answering. Using a hierarchy‑preserving statutory corpus, 2,165 fine‑tuning examples, and a 150‑item control set, the authors evaluate six instruction‑tuned models with multiple LoRA seeds, separating scoring, retrieval, and model effects. Results show that while fine‑tuning can boost overall accuracy, it does not increase the models’ reliance on the governing provision, highlighting the need to disentangle scorer, retriever, and model contributions in legal adaptation studies.
By Moniruzzaman Mahadi, Abrar Mohammed Tanzim Alam, Sayma Siddika Monalisa, Mir Mohammad Asif Abdullah, Swakkhar Shatabda, Md Adnan Arefeen
arXiv:2609.15467v1 Announce Type: cross
Abstract: Adding answer options can lower multiple-choice scores without improving assessment validity. Turkish MMLU Pro examines this distinction using 12,000...
By M. Ali Bayram