BoN Appetit Team at LeWiDi-2025: Best-of-N Test-time Scaling Can Not Stomach Annotation Disagreements (Yet)

Test-time scaling is a family of techniques to improve LLM outputs at inference time by performing extra computation. To the best of our knowledge, test-time scaling has been limited to domains with verifiably correct answers, like mathematics and coding. We transfer test-time scaling to the LeWiDi-2025 tasks to evaluate annotation disagreements. We experiment with three test-time scaling methods: two benchmark algorithms (Model Averaging and Majority Voting), and a Best-of-N sampling method. The two benchmark methods improve LLM performance consistently on the LeWiDi tasks, but the Best-of-N method does not. Our experiments suggest that the Best-of-N method does not currently transfer from mathematics to LeWiDi tasks, and we analyze potential reasons for this gap.

Standard-to-Dialect Transfer Trends Differ across Text and Speech: A Case Study on Intent and Topic Classification in German Dialects

Research on cross-dialectal transfer from a standard to a non-standard dialect variety has typically focused on text data. However, dialects are primarily spoken, and non-standard spellings are known to cause issues in text processing. We compare standard-to-dialect transfer in three settings: text models, speech models, and cascaded systems where speech first gets automatically transcribed and then further processed by a text model. In our experiments, we focus on German and multiple German dialects in the context of written and spoken intent and topic classification. To that end, we release the first dialectal audio intent classification dataset. We find that the speech-only setup provides the best results on the dialect data while the text-only setup works best on the standard data. While the cascaded systems lag behind the text-only models for German, they perform relatively well on the dialectal data if the transcription system generates normalized, standard-like output.

Is It Thinking or Cheating? Detecting Implicit Reward Hacking by Measuring Reasoning Effort

Reward hacking, where a reasoning model exploits loopholes in a reward function to achieve high rewards without solving the intended task, poses a significant threat. This behavior may be explicit, i.e. verbalized in the model's chain-of-thought (CoT), or implicit, where the CoT appears benign thus bypasses CoT monitors. To detect implicit reward hacking, we propose TRACE (Truncated Reasoning AUC Evaluation). Our key observation is that hacking occurs when exploiting the loophole is easier than solving the actual task. This means that the model is using less `effort' than required to achieve high reward. TRACE quantifies effort by measuring how early a model's reasoning becomes sufficient to pass a verifier. We progressively truncate a model's CoT at various lengths, force the model to answer, and measure the verifier-passing rate at each cutoff. A hacking model, which takes a shortcut, will achieve a high passing rate with only a small fraction of its CoT, yielding a large area under the accuracy-vs-length curve. TRACE achieves over 65% gains over our strongest 72B CoT monitor in math reasoning, and over 30% gains over a 32B monitor in coding. We further show that TRACE can discover unknown loopholes during training. Overall, TRACE offers a scalable unsupervised approach for oversight where current monitoring methods prove ineffective.

Evaluating Large Language Models for Cross-Lingual Retrieval

Multi-stage information retrieval (IR) has become a widely-adopted paradigm in search. While Large Language Models (LLMs) have been extensively evaluated as second-stage reranking models for monolingual IR, a systematic large-scale comparison is still lacking for cross-lingual IR (CLIR). Moreover, while prior work shows that LLM-based rerankers improve CLIR performance, their evaluation setup relies on lexical retrieval with machine translation (MT) for the first stage. This is not only prohibitively expensive but also prone to error propagation across stages. Our evaluation on passage-level and document-level CLIR reveals that further gains can be achieved with multilingual bi-encoders as first-stage retrievers and that the benefits of translation diminishes with stronger reranking models. We further show that pairwise rerankers based on instruction-tuned LLMs perform competitively with listwise rerankers. To the best of our knowledge, we are the first to study the interaction between retrievers and rerankers in two-stage CLIR with LLMs. Our findings reveal that, without MT, current state-of-the-art rerankers fall severely short when directly applied in CLIR.

Revisiting Active Learning under (Human) Label Variation

Access to high-quality labeled data remains a limiting factor in applied supervised learning. While label variation (LV), i.e., differing labels for the same instance, is common, especially in natural language processing, annotation frameworks often still rest on the assumption of a single ground truth. This overlooks human label variation (HLV), the occurrence of plausible differences in annotations, as an informative signal. Similarly, active learning (AL), a popular approach to optimizing the use of limited annotation budgets in training ML models, often relies on at least one of several simplifying assumptions, which rarely hold in practice when acknowledging HLV. In this paper, we examine foundational assumptions about truth and label nature, highlighting the need to decompose observed LV into signal (e.g., HLV) and noise (e.g., annotation error). We survey how the AL and (H)LV communities have addressed -- or neglected -- these distinctions and propose a conceptual framework for incorporating HLV throughout the AL loop, including instance selection, annotator choice, and label representation. We further discuss the integration of large language models (LLM) as annotators. Our work aims to lay a conceptual foundation for HLV-aware active learning, better reflecting the complexities of real-world annotation.

Evaluation Should Not Ignore Variation: On the Impact of Reference Set Choice on Summarization Metrics

Human language production exhibits remarkable richness and variation, reflecting diverse communication styles and intents. However, this variation is often overlooked in summarization evaluation. While having multiple reference summaries is known to improve correlation with human judgments, the impact of using different reference sets on reference-based metrics has not been systematically investigated. This work examines the sensitivity of widely used reference-based metrics in relation to the choice of reference sets, analyzing three diverse multi-reference summarization datasets: SummEval, GUMSum, and DUC2004. We demonstrate that many popular metrics exhibit significant instability. This instability is particularly concerning for n-gram-based metrics like ROUGE, where model rankings vary depending on the reference sets, undermining the reliability of model comparisons. We also collect human judgments on LLM outputs for genre-diverse data and examine their correlation with metrics to supplement existing findings beyond newswire summaries, finding weak-to-no correlation. Taken together, we recommend incorporating reference set variation into summarization evaluation to enhance consistency alongside correlation with human judgments, especially when evaluating LLMs.

Do LLMs Give Psychometrically Plausible Responses in Educational Assessments?

Knowing how test takers answer items in educational assessments is essential for test development, to evaluate item quality, and to improve test validity. However, this process usually requires extensive pilot studies with human participants. If large language models (LLMs) exhibit human-like response behavior to test items, this could open up the possibility of using them as pilot participants to accelerate test development. In this paper, we evaluate the human-likeness or psychometric plausibility of responses from 18 instruction-tuned LLMs with two publicly available datasets of multiple-choice test items across three subjects: reading, U.S. history, and economics. Our methodology builds on two theoretical frameworks from psychometrics which are commonly used in educational assessment, classical test theory and item response theory. The results show that while larger models are excessively confident, their response distributions can be more human-like when calibrated with temperature scaling. In addition, we find that LLMs tend to correlate better with humans in reading comprehension items compared to other subjects. However, the correlations are not very strong overall, indicating that LLMs should not be used for piloting educational assessments in a zero-shot setting.

Threading the Needle: Reweaving Chain-of-Thought Reasoning to Explain Human Label Variation

The recent rise of reasoning-tuned Large Language Models (LLMs)--which generate chains of thought (CoTs) before giving the final answer--has attracted significant attention and offers new opportunities for gaining insights into human label variation, which refers to plausible differences in how multiple annotators label the same data instance. Prior work has shown that LLM-generated explanations can help align model predictions with human label distributions, but typically adopt a reverse paradigm: producing explanations based on given answers. In contrast, CoTs provide a forward reasoning path that may implicitly embed rationales for each answer option, before generating the answers. We thus propose a novel LLM-based pipeline enriched with linguistically-grounded discourse segmenters to extract supporting and opposing statements for each answer option from CoTs with improved accuracy. We also propose a rank-based HLV evaluation framework that prioritizes the ranking of answers over exact scores, which instead favor direct comparison of label distributions. Our method outperforms a direct generation method as well as baselines on three datasets, and shows better alignment of ranking methods with humans, highlighting the effectiveness of our approach.

LiTEx: A Linguistic Taxonomy of Explanations for Understanding Within-Label Variation in Natural Language Inference

There is increasing evidence of Human Label Variation (HLV) in Natural Language Inference (NLI), where annotators assign different labels to the same premise-hypothesis pair. However, within-label variation--cases where annotators agree on the same label but provide divergent reasoning--poses an additional and mostly overlooked challenge. Several NLI datasets contain highlighted words in the NLI item as explanations, but the same spans on the NLI item can be highlighted for different reasons, as evidenced by free-text explanations, which offer a window into annotators' reasoning. To systematically understand this problem and gain insight into the rationales behind NLI labels, we introduce LITEX, a linguistically-informed taxonomy for categorizing free-text explanations. Using this taxonomy, we annotate a subset of the e-SNLI dataset, validate the taxonomy's reliability, and analyze how it aligns with NLI labels, highlights, and explanations. We further assess the taxonomy's usefulness in explanation generation, demonstrating that conditioning generation on LITEX yields explanations that are linguistically closer to human explanations than those generated using only labels or highlights. Our approach thus not only captures within-label variation but also shows how taxonomy-guided generation for reasoning can bridge the gap between human and model explanations more effectively than existing strategies.

MAKIEval: A Multilingual Automatic WiKidata-based Framework for Cultural Awareness Evaluation for LLMs

Large language models (LLMs) are used globally across many languages, but their English-centric pretraining raises concerns about cross-lingual disparities for cultural awareness, often resulting in biased outputs. However, comprehensive multilingual evaluation remains challenging due to limited benchmarks and questionable translation quality. To better assess these disparities, we introduce MAKIEval, an automatic multilingual framework for evaluating cultural awareness in LLMs across languages, regions, and topics. MAKIEval evaluates open-ended text generation, capturing how models express culturally grounded knowledge in natural language. Leveraging Wikidata's multilingual structure as a cross-lingual anchor, it automatically identifies cultural entities in model outputs and links them to structured knowledge, enabling scalable, language-agnostic evaluation without manual annotation or translation. We then introduce four metrics that capture complementary dimensions of cultural awareness: granularity, diversity, cultural specificity, and consensus across languages. We assess 7 LLMs developed from different parts of the world, encompassing both open-source and proprietary systems, across 13 languages, 19 countries and regions, and 6 culturally salient topics (e.g., food, clothing). Notably, we find that models tend to exhibit stronger cultural awareness in English, suggesting that English prompts more effectively activate culturally grounded knowledge. We publicly release our code and data.