United States

Recent progress in large language models (LLMs) has focused on test-time scaling to improve reasoning via increased inference computation, but often at the cost of efficiency. We revisit test-time behavior and uncover a simple yet underexplored phenomenon: reasoning uncertainty is highly localized—only a small subset of high-entropy tokens dominantly affects output correctness. Motivated by this, we propose Minimal Test-Time Intervention (MTI), a training-free framework that enhances reasoning accuracy and stability with minimal overhead. MTI includes: (i) Selective CFG intervention, applying classifier-free guidance only at uncertain positions; and (ii) Lightweight negative-prompt guidance, reusing the main model’s KV cache to approximate unconditional decoding efficiently. MTI yields consistent gains across general, coding, and STEM tasks—e.g., +9.28% average improvement on six benchmarks for DeepSeek-R1-7B and +11.25% on AIME2024 using Ling-mini-2.0—while remaining highly efficient.

ACL 2026

Less is More: Improving LLM Reasoning with Minimal Test-Time Intervention

### Welcome!
The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. Our Virtual Poster Sessions will take place online Thursday, August 22, 2024.

You are required to register for this event. **Please register [here](https://2024.aclweb.org/registration). **

If you have already registered, please check your inbox for an email from Underline granting you access to ACL 2024 content.

Please register!

ACL 2024

The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. More information will be announced soon.

poster

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. 
[Full Message from General Chair Philipp Koehn](https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing) 
<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>

To access this event page, you need to log in with the **email address you registered with**. 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.

We study reinforcement learning (RL) fine-tuning of large language model (LLM) agents for long-horizon multi-turn tool use, where context length quickly becomes a fundamental bottleneck. Existing multi-turn RL pipelines suffer from degraded instruction following, excessive rollout costs, and most importantly, strict context limits. In this work, to address these challenges, we introduce \emph{summarization-based context management to training}. In specific, it periodically compresses the tool using history by LLM-generated summaries that retain task-relevant information to keep a compact context while enabling the agent to scale beyond the fixed context window. Building on this formulation, we derive a policy gradient representation that seamlessly enables standard LLM RL infrastructures to optimize both tool-use behaviors as well as summarization strategies in an end-to-end fashion. We instantiate this framework with \underline{SU}mmarization augmented \underline{P}olicy \underline{O}ptimization (\texttt{SUPO}), an LLM RL algorithm that enables long-horizon training beyond a fixed context limit. Experiments on interactive function calling and searching tasks demonstrate that \texttt{SUPO} significantly improves the success rate while maintaining the same or even lower working context length compared to baselines. We also demonstrate that for complex searching tasks \texttt{SUPO} can further improve the evaluation performance when scaling test-time maximum round of summarization beyond that of training time.

Beyond the Context Window: Scaling Agentic RL via End-to-end Optimized Context Compression

Large Reasoning Models (LRMs) have emerged as a powerful advancement in multi-step reasoning tasks, offering enhanced transparency and logical consistency through explicit chains of thought (CoT). However, these models introduce novel safety and reliability risks, such as CoT-hijacking and prompt-induced inefficiencies, which are not fully captured by existing evaluation methods. To address this gap, we propose Rt-LRM, a unified benchmark designed to assess the trustworthiness of LRMs. Rt-LRM evaluates three core dimensions: truthfulness, safety and efficiency. Beyond metric-based evaluation, we further introduce the training paradigm as a key analytical perspective to investigate the systematic impact of different training strategies on model trustworthiness. We achieve this by designing a curated suite of 30 reasoning tasks from an observational standpoint. We conduct extensive experiments on 26 models and identify several valuable insights into the trustworthiness of LRMs. For example, LRMs generally face trustworthiness challenges and tend to be more fragile than Large Language Models (LLMs) when encountering reasoning-induced risks. These findings uncover previously underexplored vulnerabilities and highlight the need for more targeted evaluations. In addition, we release a scalable toolbox for standardized trustworthiness research to support future advancements in this important field.

Red Teaming Large Reasoning Models

As Large Language Model applications (LLM apps) become widespread, system prompts that determine app behavior are increasingly regarded as intellectual property, raising concerns about leakage. Recent studies show that this threat is no longer theoretical, revealing the prevalence of cloned apps replicating system prompts from others on real-world platforms. These clones pose risks of copyright infringement and malicious misuse, highlighting the need for early and reliable detection. In this paper, we propose PROMPRINT, a novel fingerprinting approach for detecting cloned LLM apps without exposing their system prompts. Motivated by the insight that different system prompts yield distinct responses to the same query, PROMPRINT optimizes queries that induce the LLM to generate a specific first token associated with the given system prompt, resulting in distinctive query–first-token pairs. Experiments on four instruction-tuned LLMs show that generated pairs effectively identify the corresponding system prompts, achieving over 74% probability of generating the target token while remaining below 2.2% on average under other prompts. Furthermore, we demonstrate that our fingerprinting remains robust to partial system prompt modifications and effective under the injection of adversarial instructions.

PROMPRINT: Prompt Fingerprinting via First-Token Response for LLM App Cloning Detection

Large language models leverage both parametric knowledge acquired during pretraining and in-context knowledge provided at inference time. Crucially, when these sources conflict, models arbitrate based on their internal confidence, preferring parametric knowledge for high-confidence facts while deferring to context for less familiar ones. However, the training conditions that give rise to these fundamental behaviors remain unclear. Here we conduct controlled experiments using synthetic corpora to identify the specific data properties that shape knowledge utilization. Our results reveal a counterintuitive finding: the robust, balanced use of both knowledge sources is an emergent property that requires the co-occurrence of three factors typically considered detrimental, including (i) intra-document repetition, (ii) a moderate degree of intra-document inconsistency, and (iii) a skewed knowledge distribution. We further show that these dynamics arise in real-world language model pretraining and analyze how post-training procedures reshape arbitration strategies. Together, our findings provide empirical guidance for designing training data that supports the reliable integration of parametric and in-context knowledge in language models.

 How Training Data Shapes the Use of Parametric and In-Context Knowledge in Language Models

In high-precision scenarios, vision language models suffer from Linguistic Priors Hallucination. When processing familiar text, models tend to over-rely on internal parametric knowledge, effectively "reciting" the content rather than "reading" the image. In this paper, we first systematically investigate this phenomenon by constructing the GlitchText Probing Dataset. We discover that the model's reliance on visual grounding diminishes significantly as the generation length increases. To mitigate this, we propose PAR (Positional Perturbation and Attention Recycling), a training-free, inference-time intervention framework. PAR consists of two parts: (1) Positional Perturbation (PP) injects structured phase noise into the rotary positional embeddings; (2) Foveal Attention Recycling (FAR) detects over-confident linguistic priors and dynamically redistributes attention mass back to important visual regions. Extensive experiments across state-of-the-art models, demonstrate that PAR significantly reduces hallucination rates (reducing CER by 12%), particularly in long-context scenarios, while maintaining robust generalization on standard benchmarks.

 PAR: Training-Free Positional Perturbation and Attention Recycling for Faithful OCR

Building effective knowledge graphs (KGs) for Retrieval-Augmented Generation (RAG) is pivotal for advancing question answering (QA) systems. However, its effectiveness is hindered by a fundamental disconnect: the knowledge graph (KG) construction process is decoupled from its downstream application, yielding suboptimal graph structures. To bridge this gap, we introduce AutoGraph-R1, the first framework to directly optimize KG construction for task performance using Reinforcement Learning (RL). AutoGraph-R1 trains an LLM constructor by framing graph generation as a policy learning problem, where the reward is derived from the graph's functional utility in a RAG pipeline. We design two novel, task-aware reward functions, one for graphs as knowledge carriers and another as knowledge indices. Across multiple QA benchmarks, AutoGraph-R1 consistently enables graph RAG methods to achieve significant performance gains over using task-agnostic baseline graphs. Our work shows it is possible to close the loop between construction and application, shifting the paradigm from building intrinsically `good` graphs to building demonstrably `useful` ones.

 AutoGraph-R1: End-to-End Reinforcement Learning for Knowledge Graph Construction

Standard Large Language Model (LLM) pre-training typically treats corpora as flattened token sequences, often overlooking the real-world context that humans naturally rely on to contextualize information. To bridge this gap, we introduce Knowledge Coordinate Conditioning (KoCo), a simple method that maps every document into a three-dimensional semantic coordinate. By prepending these coordinates as textual prefixes for pre-training, we aim to equip the model with explicit contextual awareness to learn the documents within the real-world knowledge structure. Experiment results demonstrate that KoCo significantly enhances performance across 10 downstream tasks and accelerates pre-training convergence by approximately 30%. Furthermore, our analysis indicates that explicitly modeling knowledge coordinates helps the model distinguish stable facts from noise, effectively mitigating hallucination in generated outputs.

 KoCo: Conditioning Language Model Pre-training on Knowledge Coordinates

Alignment techniques often inadvertently induce sycophancy in LLMs. While prior studies studied this behaviour in direct-answer settings, the role of Chain-of-Thought (CoT) reasoning remains under-explored: does it serve as a logical constraint that mitigates sycophancy, or a tool for post-hoc rationalization that masks it? We evaluate a range of models across objective and subjective tasks to investigate the issue.

Results show that reasoning generally reduces sycophancy in final decisions but also masks sycophancy in some samples, where models construct deceptive justifications through logical inconsistencies, calculation errors, and one-sided arguments etc. Furthermore, LLMs are more prone to sycophancy in subjective tasks and under authority-bias. Our mechanistic analysis reveals that the tendency of sycophancy in LLMs is dynamic during the reasoning process rather than being pre-determined at the input.

Good Arguments Against the People Pleasers: How Reasoning Mitigates (Yet Masks) LLM Sycophancy

Debate has emerged as a promising oversight mechanism for Large Language Models (LLMs) amid rising systemic complexity, particularly where models outperform human evaluators. Yet, Debate provides little verifiable evidence for its final judgments, and its scalability remains largely unexplored. 

To make oversight grounded and scale as capabilities extend, we introduce an Agentic Oversight framework. By using Dialectic Argumentation as a reasoning function, we extend this paradigm to multilingual and multimodal spaces. We employ a weak-to-strong oversight approach based on two expert models that evaluate and defend contesting answers, while a third blind judge determines the winner using Dialectic Argumentation. Experts argue only for belief-consistent answers, founding the Debate on disagreements. We experimented with six tasks on our framework in both multilingual and multimodal scenarios, and dialectic argumentation consistently outperforms single-expert baselines. Moreover, we show that dialectic judgements from a weaker model deliver argument-mediated supervision that, via fine-tuning, instils unsupervised reasoning signals in expert models.

Agentic Oversight via Dialectic Reasoning

Bargaining is often regarded as a logical arena rather than an art or a matter of intuition, yet Large Language Models (LLMs) still struggle to navigate it due to limited strategic depth and difficulty adapting to complex human factors. Current benchmarks rarely capture this limitation. To bridge this gap, we present a utility feedback centric framework. Our contributions are: (i) AgoraBench, a new benchmark spanning nine challenging settings (e.g., deception, monopoly) that supports diverse strategy modeling; (ii) human-aligned, economically grounded metrics derived from utility theory. This is operationalized via agent utility, negotiation power, and acquisition ratio that implicitly measure how well the negotiation aligns with human preference and (iii) a human preference grounded dataset with learning pipeline that strengthens LLMs’ bargaining ability through both prompting and finetuning. Empirical results indicate that baseline LLM strategies often diverge from human preferences, while our mechanism substantially improves negotiation performance, yielding deeper strategic behavior and stronger opponent awareness.

Downloads

Next from ACL 2026

Beyond the Context Window: Scaling Agentic RL via End-to-end Optimized Context Compression

Similar lecture

Multi-Task Inference: Can Large Language Models Follow Multiple Instructions at Once?