United States

Benchmark-based evaluation is the de facto standard for comparing large language models (LLMs). However, its reliability is increasingly threatened by test set contamination, where test samples or their close variants leak into training data and artificially inflate reported performance. To address this issue, prior work has explored two main lines of mitigation. One line attempts to identify and remove contaminated benchmark items before evaluation, but this inevitably alters the evaluation set itself and becomes unreliable when contamination is moderate or severe. The other line preserves the benchmark and instead suppresses contaminated behavior at evaluation time; however, such interventions often interfere with normal inference and lead to noticeable performance degradation on clean inputs. We propose DeconIEP, a decontamination framework that operates entirely during evaluation by applying small, bounded perturbations in the input embedding space. Guided by a relatively less-contaminated reference model, DeconIEP learns an instance-adaptive perturbation generator that steers the evaluated model away from memorization-driven shortcut pathways. Across multiple open-weight LLMs and benchmarks, extensive empirical results show that DeconIEP achieves strong decontamination effectiveness while incurring only minimal degradation in benign utility.

ACL 2026

When Benchmarks Leak: Inference-Time Decontamination for LLMs

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.

**Online Registration Form**: https://acl.swoogo.com/acl2026

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

Pretrained large language models (LLMs) are typically reused as indivisible artifacts, adapted, merged, or ensembled as a whole. In this study, we show that LLMs can instead be structurally recomposed as modular building blocks to create new architectures without access to original training data. We introduce architecture-level reassembly as a new reuse paradigm, in which Transformer blocks from heterogeneous models are treated as reusable components. This idea is formalized through a two-dimensional reassembly space that supports both vertical recombination across depth and horizontal composition within layers. To make this space tractable, we propose a chromosome-based architectural encoding and perform a bi-level multi-objective evolutionary optimization over vertical structure and horizontal composition. To resolve representation incompatibility across heterogeneous blocks, we introduce lightweight glue layers trained via data-free knowledge distillation, enabling valid information flow without modifying pretrained parameters. Our results demonstrate that architecture-level reassembly unlocks a new dimension of flexibility in model reuse, pointing toward a modular and evolutionary view of LLM design.

 Building LLMs Like LEGO: Two-dimensional Architecture Reassembly of Large Language Models

We present T$^\star$, a simple TraceRL-based curriculum for progressive block-size scaling in masked diffusion language models (MDMs).
Starting from an AR-initialized small-block MDM, T$^\star$ gradually increases the block size while re-optimizing the denoising policy at each stage, enabling higher-parallelism decoding with limited degradation on math reasoning benchmarks. Across two SDAR scales and three benchmarks, T$^\star$ consistently outperforms direct large-block TraceRL and is substantially more stable during training. Our schedule analysis suggests that the learned policy does not simply revert to a strictly left-to-right order; instead, it retains block-size-specific non-monotone updates while improving accuracy.

 T$^\star$: Progressive Block Scaling for Masked Diffusion Language Models Through Trajectory Aware Reinforcement Learning

Large language models pretrained on general-domain corpora often exhibit tokenization inefficiencies when applied to specialized domains. Although continual pretraining for domain adaptation partially alleviate performance degradation, it does not resolve the fundamental vocabulary mismatch. To address this gap, we introduce a targeted parameter-efficient domain adaptation approach that combines vocabulary adaptation with pretraining for LLM-based text summarization. Our unified framework augments pretrained tokenizers with domain-specific tokens while selectively replacing under-trained and unreachable tokens to limit parameter growth. We evaluate our approach on Llama-3.1-8B and Qwen2.5-7B across legal and medical summarization tasks on a challenge-oriented evaluation protocol focused on expert-driven text and summaries which typically has higher concentration of over-fragmented Out-of-Vocabulary (**OOV**) words. The vocabulary adaptation algorithm enhances the overall quality of the summarization model by improving semantic similarity between the generated summaries and their references. In addition, the adapted model produces summaries that incorporate more appropriate novel and domain-specific words, leading to improved coherence, relevance, and faithfulness. We further observe that our proposed approach significantly reduce training time by 35-55% over continual pretraining and reduce parameter counts up to 37% w.r.t expansion-only methods. We make codebase publicly available at [url](https://github.com/gb-kgp/VocabReplace-Then-Expand).

Learning Faster with Better Tokens: Parameter-Efficient Vocabulary Adaptation for Specialized Text Summarization

Large Language Models (LLMs) have been widely applied in various domains such as education and healthcare, making safety assurance crucial. Jailbreak attacks, a method used in red-teaming, can help evaluate and improve the defensive strategies of LLMs. However, existing jailbreak methods often overlook the semantic differences across categories of harmful questions, leading to inconsistent success rates and reduced overall attack effectiveness. We propose the first category-aware jailbreak framework, SHARP, which incorporates the semantic category of harmful questions into prompt generation. Trained on a verified jailbreak dataset, SHARP enables the model to learn category-specific semantic features and adaptively generate prompts that bypass safety mechanisms. The method combines two-stage LoRA fine-tuning, and DPO-based reinforcement learning to optimize both attack success and category alignment. Experiments show that SHARP significantly improves attack success rates and achieves better cross-category robustness compared to the state-of-the-art (SOTA) baselines, providing an efficient and scalable tool for evaluating LLM safety.

SHARP: Self-adaptive Harmful Category-aware Prompt Generation for Black-box Jailbreaking

Recently, AI-driven interactions with computing devices have advanced from basic prototype tools to sophisticated, LLM-based systems that emulate human-like operations in graphical user interfaces. We are now witnessing the emergence of Computer-Using Agents (CUAs), capable of autonomously performing tasks such as navigating desktop applications, web pages, and mobile apps. However, as these agents grow in capability, they also introduce novel safety and security risks. Vulnerabilities in LLM-driven reasoning, with the added complexity of integrating multiple software components and multimodal inputs, further complicate the security landscape. In this paper, we present a systematization of knowledge on the safety and security threats of CUAs. We conduct a comprehensive literature review and distill our findings along four research objectives: (i) define the CUA that suits safety analysis; (ii) categorize current safety threats among CUAs; (iii) propose a comprehensive taxonomy of existing defensive strategies; (iv) summarize prevailing benchmarks, datasets, and evaluation metrics used to assess the safety and performance of CUAs. Building on these insights, our work provides future researchers with a structured foundation for exploring unexplored vulnerabilities and offers practitioners actionable guidance in designing and deploying secure Computer-Using Agents.

JARVIS or Ultron? A Survey on the Safety and Security Threats of Computer-Using Agents

Human–AI collaboration in knowledge-intensive tasks such as fact-checking requires understanding model uncertainty in multi‑document reasoning amid conflicting/agreeing evidence. Yet, existing methods only express uncertainty as numbers or hedges without revealing which evidence conflicts cause the uncertainty, leaving users unable to resolve disagreements. We present CLUE (**C**onflict-&Agreement-aware **L**anguage-model **U**ncertainty **E**xplanations), a plug‑and‑play white‑box framework that, to our knowledge, is the first to generate natural‑language explanations of model uncertainty grounded in conflicting/agreeing evidence. CLUE (i) identifies span‑level claim–evidence and inter‑evidence relations that signal conflict or agreement without supervision, and (ii) uses these relations to steer explanation generation, articulating how they drive the model's uncertainty. Across three language models and two fact‑checking datasets, CLUE produces explanations that more faithfully track model uncertainty and better align with the model’s fact‑checking decisions than span‑agnostic explanation prompting; human raters also judge them more helpful, more informative, less redundant, and more logically consistent with the input. By explicitly tying uncertainty to evidence conflicts and agreements, CLUE supports practical fact‑checking and other tasks that require reasoning over complex, conflicting information.

Explaining Sources of Uncertainty in Automated Fact-Checking

Legal QA benchmarks have predominantly focused on case law, overlooking the unique challenges of statute-centric regulatory reasoning. In statutory domains, relevant evidence is distributed across hierarchically linked documents, creating a statutory retrieval gap where conventional retrievers fail and models often hallucinate under incomplete context. We introduce SearchFireSafety, a structure- and safety-aware benchmark for statute-centric legal QA. Instantiated on fire-safety regulations as a representative case, the benchmark evaluates whether models can retrieve hierarchically fragmented evidence and safely abstain when statutory context is insufficient. SearchFireSafety adopts a dual-track evaluation framework combining real-world questions that require citation-aware retrieval and synthetic partial-context scenarios that stress-test hallucination and refusal behavior. Experiments across multiple large language models show that graph-guided retrieval substantially improves performance, but also reveal a critical safety trade-off: domain-adapted models are more likely to hallucinate when key statutory evidence is missing. Our findings highlight the need for benchmarks that jointly evaluate hierarchical retrieval and model safety in statute-centric regulatory settings.

Evaluating Structure-Aware Retrieval and Safety in Statute-Centric Legal QA

We introduce SkMTEB, the first comprehensive MTEB-style text embedding benchmark for Slovak, a low-resource West Slavic language, comprising 31 datasets across 7 task types---nearly 4$\times$ the depth of existing multilingual benchmark coverage for Slovak. Our evaluation of 31 embedding models reveals that large instruction-tuned multilingual models achieve the strongest performance, while existing Slovak-specific models trained for NLU tasks transfer poorly to embedding tasks. To address the need for efficient, locally-deployable Slovak embeddings, we develop \texttt{e5-sk-small} (45M parameters) and \texttt{e5-sk-large} (365M) by applying vocabulary trimming and fine-tuning to Multilingual E5 models. Despite size reductions of up to 62\%, our open-source models achieve competitive performance with proprietary APIs while remaining locally deployable for semantic search and retrieval-augmented generation (RAG). We release the benchmark, models, datasets, and code openly, hoping our approach offers a replicable path for other under-resourced languages.

 SkMTEB: Slovak Massive Text Embedding Benchmark and Model Adaptation

Coreference resolution is typically evaluated using aggregate statistical metrics such as CoNLL-F1
, which measure structural overlap between predicted and gold clusters. While widely used, these metrics offer limited diagnostic insights, penalizing errors without revealing whether a system struggles with specific semantic categories, such as people, locations, or events, and making it difficult to interpret model capabilities or derive actionable improvements. We address this gap by introducing a semantically-enhanced evaluation framework for coreference resolution. Our approach overlays Concept and Named Entity Recognition (CNER) onto coreference outputs, assigning semantic labels to nominal mentions and propagating them to entire coreference clusters. This enables the computation of typed scores aimed at evaluating mention extraction and linking capabilities stratified by semantic class. Across our experiments on OntoNotes, LitBank, and PreCo, we show that our framework uncovers systematic weaknesses that remain obscured by aggregate metrics. Furthermore, we show that these diagnostics can be used to design targeted, low-cost data augmentation strategies, achieving measurable out-of-domain improvements.

 Interpretable Coreference Resolution Evaluation Using Explicit Semantics

Language models are increasingly deployed in interactive settings where users reason about facts over time rather than in isolation.
In such scenarios, correct behavior requires models to maintain and update implicit temporal assumptions established earlier in a conversation. We study this challenge through the lens of temporal scope stability: the ability to preserve, override, or transfer time-scoped factual context across dialogue turns. We introduce ChronoScope, a large-scale diagnostic benchmark designed to isolate temporal scope behavior in controlled multi-turn interactions, comprising over one million deterministically generated question chains grounded in Wikidata. ChronoScope evaluates whether models can correctly retain inferred temporal scope when follow-up questions omit explicit time references, spanning implicit carryover, explicit scope switching, cross-entity transfer, and longer temporal trajectories.
Through extensive evaluation of state-of-the-art language models, we find that temporal scope stability is frequently violated in controlled multi-turn settings, with models often drifting toward present-day assumptions despite correct underlying knowledge.
These failures intensify with interaction length and persist even under oracle context conditions, revealing a gap between single-turn factual accuracy and coherent temporal reasoning under sequential interaction.
We make our dataset and evaluation suite publicly available at \url{https://github.com/yashkumaratri/ChronoScope}.

Downloads

Next from ACL 2026

Building LLMs Like LEGO: Two-dimensional Architecture Reassembly of Large Language Models

Similar lecture

Controllable Contamination Detection for Reliable LLM Evaluation with Statistical Guarantees