United States

Supervised Fine-Tuning (SFT) followed by Reinforcement Learning (RL) has emerged as the standard post-training paradigm for large language models (LLMs). However, the conventional SFT process, driven by Cross-Entropy (CE) loss, often induces mode collapse, where models over-concentrate on specific response patterns. This lack of distributional diversity severely restricts the exploration efficiency required for subsequent RL. While recent studies have attempted to improve SFT by replacing CE loss, aiming to preserve diversity or refine the update policy, they fail to adequately balance diversity and accuracy, thereby achieving sub-optimal performance after RL. To address the mode collapse problem, we propose SED-SFT, which adaptively encourages diversity based on the token exploration space. This framework introduces a selective entropy regularization term with a selective masking mechanism into the optimization objective. Extensive experiments across eight mathematical benchmarks demonstrate that SED-SFT significantly enhances generation diversity with a negligible computational overhead increase compared with CE loss, yielding average improvements of 2.06 and 1.20 points in subsequent RL performance over standard CE-based baselines on Llama-3.2-3B-Instruct and Qwen2.5-Math-7B-Instruct, respectively.

ACL 2026

SED-SFT: Selectively Encouraging Diversity in Supervised Fine-Tuning

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.

poster

Embedding fusion has become a widely adopted technique for enhancing performance across various NLP tasks. While prior research suggests that different layers of language models encode distinct representations and that pooling strategies influence performance, there is a lack of systematic analysis regarding the empirical efficacy of these differences or the impact of combining embeddings from multiple models. This study provides a rigorous, empirical evaluation of layer-wise fusion strategies to determine their actual contribution to classification performance. Our findings reveal that the effectiveness of individual layers is more dependent on dataset characteristics than on the model architecture itself. Furthermore, we demonstrate that fusing embeddings from multiple models yields more robust and consistent representations across tasks, with the influence of any single model diminishing as the number of integrated models increases. Notably, experiments on low-resource datasets show that embedding fusion provides particularly significant gains when training data is scarce, highlighting its robustness and adaptability in data-constrained environments. We also analyze the trade-off between performance gains and computational overhead, and discuss which fusion configurations provide the best balance between stability and efficiency.

Beyond Single Representations: Multi-Model Embedding Fusion for Stable Text Classification

Large Language Models (LLMs) are increasingly applied in high-stakes domains such as finance, healthcare, and education, where reliable multi-turn interactions with users are essential. However, existing work on confidence estimation and calibration, a major approach to building trustworthy LLM systems, largely focuses on single-turn settings and overlooks the risks and potential of multi-turn conversations. In this work, we introduce the task of multi-turn calibration to reframe calibration from a static property into a dynamic challenge central to reliable multi-turn conversation, where calibrating model confidence at each turn conditioned on the conversation history is required. We first reveal the risks of this setting: using Expected Calibration Error at turn T (ECE@T), a new metric that tracks calibration dynamics over turns, we show that user feedback (e.g., persuasion) can degrade multi-turn calibration. To address this, we propose MTCal, which minimises ECE@T via a surrogate calibration target, and further leverage calibrated confidence in ConfChat, a decoding strategy that improves both factuality and consistency of the model response in multi-turn interactions. Extensive experiments demonstrate that MTCal achieves outstanding and consistent performance in multi-turn calibration, and ConfChat preserves and even enhances model performance in multi-turn interactions. Our results mark multi-turn calibration as one missing link for scaling LLM calibration toward safe, reliable, and real-world use. The code is available at: https://github.com/petezone/Multiturn-Calibration.

Confidence Should Be Calibrated More Than One Turn Deep

Large language models (LLMs) are fluent but often brittle when interpretation depends on external information (e.g., events or participant roles), as next-token prediction does not explicitly encode situation-level semantic constraints. FrameNet provides a structured account of semantics through its inventory of frames, roles, and relations. We present a scalable framework that injects frame-semantic knowledge into LLMs via LoRA, moving from fact-oriented prompting to principle-oriented supervision over the full FrameNet inventory. The supervision encodes semantic constraints through semantic types, sense-aware definitions, frame relations, and role-annotated examples. To test whether this knowledge generalizes beyond surface cues, we use Natural Language Inference (NLI) as a diagnostic task for event-level reasoning. Experiments on CONFER and SNLI show consistent gains over Meta-Llama-3.1-8B-Instruct in zero-shot and few-shot settings, especially for entailment and contradiction. Complementary semantic role labeling analyses further indicate improved sensitivity to frame, role, and span structure.

Frame-Semantic Knowledge Injection for Event-Level Inference in LLMs

Retrieval-augmented generation(RAG) systems depend on retrieval modules to supply grounding evidence for large language models. While hybrid approaches combining sparse and dense retrievers improve performance, most rely on fixed weights that ignore query-specific and corpus-specific variation. Similarly, query expansion has long been used to enrich recall, but its integration with original queries is usually static and can introduce noise. We present QuDAR, a dual-perspective adaptive retrieval framework that adapts along two perspectives: retriever type (sparse vs. dense) and query format (original vs.expanded). Leveraging margin-derived confidence (e.g., top-1--top-2 score gaps) and blind LLM-based relevance scoring, QuDAR dynamically assigns query-specific weights, fusing lexical specificity with semantic breadth while mitigating noise. QuDAR is lightweight, retriever-agnostic, and broadly applicable. Experiments show consistent gains over static baselines, improving overall retrieval quality and yielding more stable performance across queries.

 QuDAR: Query-Wise Dual-Perspective Adaptive Retrieval

The growing sequence length of large language models poses significant challenges for key-value (KV) caches. Existing state-of-the-art cache eviction methods primarily analyze the inference behavior of attention heads in successful retrieval-reasoning cases, often overlooking diverse behaviors in failure cases, such as bias and distraction. This oversight limits the potential to leverage heterogeneous head behaviors for improved eviction performance. Inspired by the confusion matrix, we introduce an Attention Behavior Matrix to comprehensively analyze attention head behaviors in both success and failure scenarios. By maximizing the signal-to-noise ratio — strengthening valid reasoning pathways in success cases while inhibiting noise from bias and distraction in failure cases — we propose REtrieval-reAsoning and Logic-constructed (REAL) KV cache eviction, the first method to leverage multi-behavior analysis. Comprehensive evaluations show that REAL achieves remarkable performance across various models and benchmarks; notably, on LongBench v2, it achieves comparable accuracy to the strongest baseline, HeadKV-R2, while requiring 32x less space. By offering a novel perspective on behavior analysis, we pave the way for a shift from success-only to comprehensive, failure-aware methods in long-context modeling. Our code is available at https://github.com/yonseicasl/REAL.

REAL: REtrieval-reAsoning and Logic-constructed Attention Behaviors for Long-Context KV Cache Compression

The rapid proliferation of large audio models (LAMs) demands efficient approaches for model comparison, yet comprehensive benchmarks are costly. To fill this gap, we investigate whether minimal subsets can reliably evaluate LAMs while reducing costs and data redundancy. Analyzing 10 subset selection methods with 18 audio models across 40 tasks covering major LAM evaluation dimensions, we show that subsets of just 50 examples (0.3% of data) can achieve over 0.93 Pearson correlation with full benchmark scores. To understand how well these scores align with what practitioners ultimately care about—user satisfaction—we collect 776 human preference ratings from realistic voice assistant conversations, finding that both subsets and full benchmark achieve only 0.85 correlation with human. To better predict preferences, we trained regression models on these selected subsets, achieving 0.98 correlation—outperforming regression models trained on both random subsets and the full benchmark. This demonstrates that in regression modeling, well-curated subsets outpredict the full benchmark, showing quality over quantity. We open-source these regression-weighted subsets as the HUMANS benchmark, an efficient proxy for LAM evaluation that captures both benchmark performance and user preferences.

Putting HUMANS first: Efficient LAM Evaluation with Human Preference Alignment

The linguistic proximity between Galician, Portuguese, and Spanish results in a lexical overlap that often conceals semantic interference. This is particularly evident in false friends, posing a challenge for NLP systems.
In this work, we assess whether state-of-the-art language models can identify and process false friends among these languages. We introduce six cross-lingual datasets --created manually or using semi-automatic methods, with all instances being carefully verified-- covering cognates and false friends. We evaluate a broad range of encoder and decoder models of varying sizes via zero-shot and few-shot settings. Our results highlight the challenging nature of the task, but also show the clear progress made by LLMs in recent years, particularly those of a larger size, with smaller language models struggling on the task. Notably, unlike other tasks where language distance poses additional challenges, we find that linguistic proximity itself introduces errors: closely related language pairs tend to perform worse, reflecting the challenge of semantic discrimination due to lexical overlap.

False Friends or Cognates? A Cross-lingual Semantic Ambiguity Evaluation for Galician, Portuguese and Spanish

Answering natural language queries over relational data often requires retrieving and reasoning over multiple tables, yet most retrievers optimize only for query–table relevance and ignore table–table compatibility. We introduce REaR (Retrieve, Expand and Refine), a three-stage, LLM-free framework that separates semantic relevance from structural joinability for efficient, high-fidelity multi-table retrieval. REaR (i) retrieves query-aligned tables, (ii) expands these with structurally joinable tables via fast, precomputed column-embedding comparisons, and (iii) refines them by pruning noisy or weakly related candidates. Empirically, REaR is retriever-agnostic and consistently improves dense/ sparse retrievers on complex table QA datasets (BIRD, MMQA, and Spider) by improving both multi-table retrieval quality and downstream SQL execution. Despite being LLM-free, it delivers performance competitive with state-of-the-art LLM-augmented retrieval systems (e.g., ARM) while achieving much lower latency and cost. Ablations confirm complementary gains from expansion and refinement, underscoring REaR as a practical, scalable building block for table-based downstream tasks (e.g., Text-to-SQL).

 REaR : Retrieve, Expand and Refine for Effective Multitable Retrieval

Smart Contracts are the foundation of Decentralized Finance (DeFi), executing financial logic without trusted intermediaries. Recent advances in large language models (LLMs) have substantially lowered the barrier to smart contract development by enabling code generation from natural language. However, because smart contracts are immutable and directly manage financial assets, this accessibility introduces a critical trust gap: generated contracts are easy to produce but hard to trust. To bridge this gap, we present LeVer, the first trustworthy smart contract synthesis framework that integrates LLM-based generation with Lean-based auto-formalization and Verification. LeVer employs a closed-loop multi-agent architecture to iteratively generate, verify, attack, and repair contracts, providing both formal guarantees and empirical robustness. To facilitate the adoption of automated formal verification in smart contract generation and audition, we open-source our framework and datasets at: https://github.com/gl-bowei/LeVer

Towards Trustworthy Smart Contract Synthesis: A Multi-Agent Framework with Lean-Based Verification

Visual scale recognition is a fundamental aspect for humans to perceive physical quantities in the real world, and it is crucial for enabling human-like intelligence in multimodal large language models (MLLMs). However, existing benchmarks typically focus on a single type of quantity (e.g., time) or a specific format (e.g., dials), lacking a comprehensive evaluation of scale recognition capabilities. To address these problems, we propose ScaleBench, a visual scale recognition benchmark built using images from COCO, Open Images, and Flickr, designed to comprehensively evaluate the scale recognition capabilities of MLLMs. To ensure high data quality, we develop detailed annotation guidelines and procedures, resulting in a total of 6,574 annotated samples. Based on this benchmark, we evaluate multiple closed-source and open-source MLLMs. Experimental results reveal that the best-performing model achieves only 42.60% accuracy, far lower than the 97.40% of humans. Furthermore, we conduct in-depth experimental analyses and provide future research directions. Our benchmark and implementation codes are available at https://github.com/Sonder-hang/ScaleBench.

Downloads

Next from ACL 2026

Beyond Single Representations: Multi-Model Embedding Fusion for Stable Text Classification

Similar lecture

Learning While Staying Curious: Entropy-Preserving Supervised Fine-Tuning via Adaptive Self-Distillation for Large Reasoning Models