United States

Self-Anchoring Calibration Drift} (SACD), a tendency for large language models (LLMs) to show systematic changes in expressed confidence when building iteratively on their own prior outputs across multi-turn conversations. Through a controlled three-condition study comparing Claude Sonnet 4.6, Gemini 3.1 Pro, and GPT-5.2 across factual, technical, and open-ended domains, we find that SACD is real but multiform: models exhibit distinct self-anchoring signatures ranging from active confidence suppression to calibration improvement suppression, with effects concentrated in open-ended domains. These findings challenge the adequacy of single-turn calibration evaluation for characterizing LLM reliability in realistic multi-turn deployment contexts. Code and data are available at https://github.com/hvardhan878/calibration-drift

ACL 2026

Self-Anchoring Calibration Drift in Large Language Models: How Multi-Turn Conversations Reshape Model Confidence

expected calibration error

self-anchoring

confidence elicitation

multi-turn dialogue

calibration

large language models

evaluation

It is my great pleasure and honor to welcome you to the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026), held in San Diego, California, USA, July 2–7, 2026. ACL 2026 continues our fields tradition of excellence in scholarship, innovation, and inclusivity, and I am deeply grateful to the many volunteers who have worked tirelessly to bring this event to life.

It is exciting to see the many developments in the field reflected at our conference, such as the advancement of large language models (born out of our community) into many application directions. This is not without challenges, partly due to unsolved problems, and partly due to misuse of the technology. All this is reflected well at ACL 2026. The conference also reflects that the field is not only about advancing technology, but also the science of fundamentally understanding language and how it is use by many different speaker communities. <br>
[Full Message from General Chair Philipp Koehn](https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing) <br><br><br>
<html><button style="display: inline-flex; align-items: center; justify-content: center; white-space: nowrap; border-radius: 9999px; font-weight: bold; background: #7c3aed; color: white; font-family: 'Space Grotesk', sans-serif; height: 40px; font-size: 16px; padding: 0 20px; border: none; cursor: pointer" onclick="window.open('https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing')">Conference Handbook</button></html>
<br><br><br><br>

To access this event page, you need to log in with the **email address you registered with**. <br>Access credentials will be sent to your email from Underline -  subject line "Welcome to ACL 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://2026.aclweb.org/registration/) first.

Registration Required

The 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026) will take place in San Diego, California, from July 2 - 7, 2026.

workshop paper

Representing continuous time is a critical and under-explored challenge in modeling temporal event sequences with large language models (LLMs). Various strategies like byte-level representations or calendar tokens have been proposed. However, the optimal approach remains unclear, especially given the diverse statistical distributions of real-world event data, which range from smooth log-normal to discrete, spiky patterns. This paper presents a systematic empirical study of temporal tokenization for modeling event sequences with LLMs, comparing distinct encoding strategies: naive numeric strings, high-precision byte-level representations, human-semantic calendar tokens, classic uniform binning, and adaptive residual scalar quantization. We evaluate these strategies by fine-tuning LLMs on real-world datasets that exemplify these diverse distributions. Our analysis reveals that no single strategy is universally superior; instead, prediction performance depends heavily on aligning the tokenizer with the data's statistical properties, highlighting temporal tokenization as a critical yet often overlooked design dimension in LLM-based event modeling.

Temporal Tokenization Strategies for Event Sequence Modeling with Large Language Models

We present a large-scale human evaluation benchmark for assessing cultural localisation in machine translation produced by state-of-the-art multilingual large language models (LLMs). Existing MT benchmarks emphasise token-level and grammatical accuracy, but often overlook pragmatic and culturally grounded competencies required for real-world localisation. Building on a pilot study of 87 translations across 20 languages, we evaluate 7 multilingual LLMs across 15 target languages with 5 native-speaker raters per language. Raters scored both full-text translations and segment-level instances of culturally nuanced language (idioms, puns, holidays, and culturally embedded concepts) on an ordinal 0–3 quality scale; segment ratings additionally included an NA option for untranslated segments. Across full-text evaluations, mean overall quality is modest (1.68/3): GPT-5 (2.10/3), Claude Sonnet 3.7 (1.97/3), and Mistral Medium 3.1 (1.84/3) form the strongest tier with fewer catastrophic failures. Segment-level results show sharp category effects: holidays (2.20/3) and cultural concepts (2.19/3) translate substantially better than idioms (1.65/3) and puns (1.45/3), and idioms are most likely to be left untranslated. These findings demonstrate a persistent gap between grammatical adequacy and cultural resonance. To our knowledge, this is the first multilingual, human-annotated benchmark focused explicitly on cultural nuance in translation and localisation, highlighting the need for culturally informed training data, improved cross-lingual pragmatics, and evaluation paradigms that better reflect real-world communicative competence.

“Be My Cheese?”: Cultural Nuance Benchmarking for Machine Translation in Multilingual LLMs

The field of natural language generation has accumulated a rich ecosystem of automatic evaluation metrics, yet it lacks a coherent theory of what those metrics are actually measuring. Drawing on measurement theory from the quantitative social sciences, this paper argues that current NLG evaluation practices suffer from a fundamental construct validity problem: metrics are treated as proxies for output quality without explicit specification of the underlying constructs they are meant to operationalize. We examine four dominant evaluation paradigms (reference-based metrics, embedding-based metrics, LLM-as-judge, and human evaluation) and demonstrate that each conflates construct definition with operationalization. Building on a long psychometric tradition reaching back to Cronbach and Meehl (1955) and on recent NLP work that has begun to apply this tradition to bias measurement, dialogue evaluation, and benchmark design, we propose that the field adopt a measurement modeling perspective for NLG evaluation. We borrow the concepts of construct validity, reliability, and consequential validity as a foundation for more principled evaluation, and we outline a preliminary taxonomy of NLG quality constructs as a starting point for this work.

Position: What Are We Measuring? Rethinking Evaluation in Natural Language Generation

Post-training methods such as supervised fine-tuning and preference optimization are widely used to align large language models, yet how their benefits distribute across architectural components and transfer across tasks and prompts remains unclear. In this work, we analyze component-level transfer in a Mixture-of-Experts language model by selectively replacing routers, attention modules, and expert networks between two post-trained Mixture of Experts models trained with different post-training recipes and dataset mixtures. Starting from a SFT+DPO checkpoint, we systematically replace its components (routers, attention, experts) with those from a Tulu3 checkpoint and evaluate the impact of each replacement and their combinations on mathematical and scientific reasoning and a general-purpose classification task under zero-shot, few-shot and Chain of Thought prompting. We find strong component-specific specialization: expert networks account for most gains on mathematical and scientific reasoning, while attention mechanisms consistently outweigh expert transfer on general tasks and router transfer alone provides minimal benefit or harms performance. Prompting strategy further modulates these effects, with expert transfer degrading zero-shot science performance but improving few-shot reasoning. Strategically combining components from different model versions can in some cases match or exceed the performance of the best available model, motivating principled techniques for composing post-trained models into task- and prompt-specific systems without additional training.

Component Transfer Can Exceed Full Model Performance: Investigating Post-Trained Mixture-of-Experts

Large language models (LLMs) are increasingly evaluated using LLM-as-a-Judge frameworks, yet existing approaches often rely on fixed evaluation criteria that assume universal notions of answer quality. In expert domains, however, evaluation standards vary across communities and reflect domain-specific priorities. We introduce CAMEL, a framework for learning Community-Aligned Multi-metric Evaluator LLMs directly from large-scale human preference data. Using question–answer interactions from Stack Exchange communities, CAMEL discovers community-specific evaluation metrics, generates rubric-based scoring criteria, and learns metric importance weights that reproduce observed preference orderings. This process converts implicit community preferences into structured supervision for training evaluator models. The resulting evaluator produces interpretable reports that include metric-level scores, rubric-grounded explanations, and a calibrated overall score aligned with community norms. Experiments across seen and unseen communities show that CAMEL improves preference-order prediction compared to prompting-based judges and general-purpose evaluation models. In addition, a human-subject study with domain practitioners demonstrates that CAMEL’s metric-based evaluation reports are strongly preferred for their transparency, consistency, and diagnostic usefulness. Our results highlight the importance of community-aligned evaluation design and suggest that preference data from real-world expert communities can be leveraged to build more interpretable and human-aligned LLM evaluators.

CAMEL: Learning Community-Aligned Metrics and Weights for LLM Evaluation

Extractive QA tasks are commonly evaluated using Exact Match (EM) and F1-score, but these metrics often fail to reflect true model performance. Recent studies have proposed using large language models (LLMs) as judges (LLM-as-a-judge), yet they often lack comprehensive evaluation across datasets and overlook key factors such as sensitivity to answer types, prompt variations, and self-preference bias. In this work, we conduct a systematic study of LLM-as-a-judge across four extractive QA datasets and various prompt variations, assessing multiple LLM families in both answering and judging roles. Our results show that LLM-as-a-judge judgments correlate much more strongly with human evaluations than EM (0.22) and F1 (0.40), achieving correlations up to 0.85 with open-source models. Further analysis reveals that LLM-as-a-judge performs particularly well on number-related answers but faces challenges with more complex types, such as job titles. Contrary to findings in other NLP tasks, we observe no self-preference bias, even when the same model serves as both QA model and judge. Finally, we find that prompt phrasing has minimal impact, and zero-shot, context-free judging often yields the best evaluation performance.

Reassessing Extractive QA Datasets at Scale: LLM-as-a-Judge and In-Depth Analyses

Indic languages, spoken by over 1.5 billion people, pose unique challenges for NLP due to their cultural richness, linguistic diversity, and structural complexity. We present IndicMMLU-Pro, a comprehensive benchmark extending the MMLU-Pro framework to nine major Indic languages: Hindi, Bengali, Gujarati, Marathi, Kannada, Punjabi, Tamil, Telugu, and Urdu. Covering a wide range of tasks in comprehension, reasoning, and generation, IndicMMLU-Pro offers a standardized evaluation framework to advance AI model development in Indic contexts. This paper details the benchmark’s design, taxonomy, and data curation, and establishes baseline results using state-of-the-art multilingual models. As an open resource IndicMMLU-Pro aims to accelerate progress in Indic language technologies and support inclusive research in multilingual NLP.

IndicMMLU-Pro: Benchmarking Indic Large Language Models on Multi-Task Language Understanding

We design experiments to identify where Large Language Models (LLMs) struggle when answering complex questions. Our focus is on two key stages, mirroring the human QA process: 1) \textit{question decomposition}, where the model breaks down a complex question into sub-questions and 2) \textit{subproblem solving}, where it addresses each sub-question to obtain the final response. We preprocess and expand three multi-hop datasets to create experimental datasets featuring explicit and implicit multi-hop questions, crowdsourced and templated questions, and varying numbers of hops. Our results show that larger models (Llama 3.1 70B and o1) excel at decomposing explicit multi-hop questions but struggle with implicit ones, while smaller models (e.g., Llama 3.1 8B) have difficulty with both. In the sub-problem solving stage, all models perform well on simple questions with context. Furthermore, we found no correlation between accuracy in the question decomposition stage and final QA performance (direct response), highlighting a key difference between human and LLM reasoning.

Identifying Where Large Language Models Struggle in Answering Complex Questions

Tool-calling agents are evaluated on tool selection, parameter accuracy, and scope recognition, yet LLM trajectory assessments remain inherently *post-hoc*. Disconnected from the active execution loop, such assessments identify errors that are usually addressed through prompt-tuning or retraining, and fundamentally cannot course-correct the agent in real time. To close this gap, we move evaluation into the execution loop at *inference time*: a specialized reviewer agent evaluates provisional tool calls *prior to* execution, shifting the paradigm from post-hoc recovery to proactive evaluation and error mitigation. In practice, this architecture establishes a clear separation of concerns between the primary execution agent and a secondary review agent. As with any multi-agent system, the reviewer can introduce new errors while correcting others, yet no prior work to our knowledge has systematically measured this tradeoff. To quantify this tradeoff, we introduce *Helpfulness-Harmfulness metrics*: helpfulness measures the percentage of base agent errors that feedback corrects; harmfulness measures the percentage of correct responses that feedback degrades. These metrics directly inform reviewer design by revealing whether a given model or prompt provides net positive value. We evaluate our approach on BFCL (single-turn) and $\tau^2$-Bench (multi-turn stateful scenarios), achieving +5.5\% on irrelevance detection and +7.1\% on multi-turn tasks. Our metrics reveal that reviewer model choice is critical: the reasoning model o3-mini achieves a 3:1 benefit-to-risk ratio versus 2.1:1 for GPT-4o. Automated prompt optimization via GEPA provides an additional +1.5--2.8\%. Together, these results demonstrate a core advantage of separating execution and review: the reviewer can be systematically improved through model selection and prompt optimization, without retraining the base agent.

Reinforced Agent: Inference-Time Feedback for Tool-Calling Agents

Automated grading of open-ended student answers is a natural setting for **LLM-as-evaluator** research, yet existing methods break down when students take diverse solution approaches to the same problem. We present **DynoGrader**, a human-in-the-loop (HITL) grading system that grounds every LLM judgment in explicit evidence quotes and applies approach-specific rubrics, along with **DynoBench**, a benchmark of 12,820 human-graded responses spanning two university courses. Spending under 9 minutes of TA time per question on lightweight HITL approach discovery (avg. 12 out of 120 unreviewed submissions), DynoGrader reduces MAE by 5.2% over the strongest LLM-evaluator baseline ($p < 0.001$)—recovering 90% of the gain from fully expert-annotated approaches, with no expert pre-annotation required. *This work is under review at ACL 2026.*

Downloads

Next from ACL 2026

Temporal Tokenization Strategies for Event Sequence Modeling with Large Language Models

Similar lecture

Confidence Should Be Calibrated More Than One Turn Deep

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from ACL 2026

Temporal Tokenization Strategies for Event Sequence Modeling with Large Language Models

Similar lecture

Confidence Should Be Calibrated More Than One Turn Deep

Downloads