United States

Generative Reward Models (GenRMs) and LLM-as-a-Judge exhibit deceptive alignment by producing correct judgments for incorrect reasons, as they are trained and evaluated to prioritize \textit{Outcome Accuracy}, which undermines their ability to generalize during RLHF. We introduce \textit{Rationale Consistency}, a fine-grained metric that quantifies the alignment between the model&#39;s reasoning process and human judgment. Our evaluation of frontier models reveals that rationale consistency effectively discriminates among state-of-the-art models and detects deceptive alignment, while outcome accuracy falls short in both respects. To mitigate this gap, we introduce a hybrid signal that combines rationale consistency with outcome accuracy for GenRM training. Our training method achieves state-of-the-art performance on RM-Bench (87.1%) and JudgeBench (82%), surpassing outcome-only baselines by an average of 5%. Using RM during RLHF, our method effectively improves performance as demonstrated on Arena Hard v2, notably yielding a 7% improvement in creative writing tasks. Further analysis confirms that our method escapes the deceptive alignment trap, effectively reversing the decline in rationale consistency observed in outcome-only training.

ACL 2026

Outcome Accuracy is Not Enough: Aligning the Reasoning Process of Reward Models

### Welcome to 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics

Welcome to the 2025 meeting of the Nations of the Americas Chapter of the Association for Computational Linguistics! I am proud to help organize the first NAACL conference to carry the new name of our organization, one that emphasizes inclusion for all of the Americas. I am also pleased to welcome you to Albuquerque, New Mexico, a state whose unique blend of cultural influences will make for an excellent backdrop for NAACL 2025, especially with this year’s special theme on NLP in a Multicultural World. 
**[Continue reading...](https://drive.google.com/file/d/1jX-qGhqVSZZCIrAnJaz798pflu5Irrdn/view?usp=sharing)**

*- Colin Cherry, Google, NAACL 2025 General Chair* 

[![](https://assets.underline.io/markdown_image/1/image/b087f8a4dc5816d6e1a6514e59c59ac3.png)](https://drive.google.com/file/d/1T96GzPqObXrMln2BMByCSXTSizjTg69P/view?usp=sharing)

You need to log in with the email address you registered with. Access credentials have been sent to your email. 

Please be sure to check your spam and other email folders if you do not see an email confirmation right away.

Please log in to explore this event.

To access NAACL 2025 event page you are required to register. Please follow [**this link**](https://2025.naacl.org/registration/registration/) to register. Access will depend on your registration type.

Please register!

NAACL 2025

Welcome to 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics

Generative Reward Models (GenRMs) and LLM-as-a-Judge exhibit deceptive alignment by producing correct judgments for incorrect reasons, as they are trained and evaluated to prioritize \textit{Outcome Accuracy}, which undermines their ability to generalize during RLHF. We introduce \textit{Rationale Consistency}, a fine-grained metric that quantifies the alignment between the model's reasoning process and human judgment. Our evaluation of frontier models reveals that rationale consistency effectively discriminates among state-of-the-art models and detects deceptive alignment, while outcome accuracy falls short in both respects. To mitigate this gap, we introduce a hybrid signal that combines rationale consistency with outcome accuracy for GenRM training. Our training method achieves state-of-the-art performance on RM-Bench (87.1%) and JudgeBench (82%), surpassing outcome-only baselines by an average of 5%. Using RM during RLHF, our method effectively improves performance as demonstrated on Arena Hard v2, notably yielding a 7% improvement in creative writing tasks. Further analysis confirms that our method escapes the deceptive alignment trap, effectively reversing the decline in rationale consistency observed in outcome-only training.

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 how visual artifacts introduced by diffusion-based inpainting affect language generation in vision-language models. We use a two-stage diagnostic setup in which masked image regions are reconstructed and then provided to captioning models, enabling controlled comparisons between captions generated from original and reconstructed inputs. Across multiple datasets, we analyze the relationship between reconstruction fidelity and downstream caption quality. We observe consistent associations between pixel-level and perceptual reconstruction metrics and both lexical and semantic captioning performance. Additional analysis of intermediate visual representations and attention patterns shows that inpainting artifacts lead to systematic, layer-dependent changes in model behavior. Together, these results provide a practical diagnostic framework for examining how visual reconstruction quality influences language generation in multimodal systems.

How Do Inpainting Artifacts Propagate to Language?

Large vision–language models (VLMs) are highly capable, yet often hallucinate by favoring textual prompts over visual evidence. We study this failure mode in a controlled object-counting setting, where the prompt overstates the number of objects in the image (e.g., asking a model to describe \textit{four} waterlilies when only \textit{three} are present). At low object counts, models often correct the overestimation, but as the number of objects increases, they increasingly conform to the prompt regardless of the discrepancy. Through mechanistic analysis of three VLMs, we identify a small set of attention heads whose ablation substantially reduces prompt-induced hallucinations (PIH) by at least 40% without additional training. Across models, PIH-heads mediate prompt copying in model-specific ways. We characterize these differences and show that PIH ablation increases correction toward visual evidence. Our findings offer insights into the internal mechanisms driving prompt-induced hallucinations, revealing model-specific differences in how these behaviors are implemented.

 Mechanisms of Prompt-Induced Hallucination in Vision–Language Models

Graph Neural Networks (GNNs) have emerged as powerful tools for learning over structured data, including text-attributed graphs (TAGs), which are common in domains such as citation networks, social platforms, and knowledge graphs. GNNs are not inherently interpretable and thus, many explanation methods have been proposed. However, existing explanation methods often struggle to generate interpretable, fine-grained rationales, especially when node attributes include rich natural language. In this work, we introduce GSPELL, a lightweight, post-hoc framework that uses large language models (LLMs) to generate faithful and interpretable explanations for GNN predictions. GSPELL projects GNN node embeddings into the LLM embedding space and constructs hybrid prompts that interleave soft prompts with textual inputs from the graph structure. This enables the LLM to reason about GNN internal representations and to produce natural-language explanations, along with concise explanation subgraphs. Our experiments across real-world TAG datasets demonstrate that GSPELL achieves a favorable trade-off between fidelity and sparsity, while improving human-centric metrics such as insightfulness. GSPELL sets a new direction for LLM-based explainability in graph learning by aligning GNN internals with human reasoning.

From Nodes to Narratives: Explaining Graph Neural Networks with LLMs and Graph Context

Millions of people use machine translation (MT) tools daily, yet little is known about their perception of what systems can and cannot do. This paper studies users' mental models of speech translation systems through a new framework based on cross-lingual question answering, where users either accept MT output or request professional re-translation to answer questions based on the information presented in a foreign language. By analyzing user behavior and accuracy trends across varying translation qualities, we examine to what extent they can predict where the system is likely to be wrong, and how this mental model evolves. Users develop stronger mental models with practice, especially when they have some knowledge of the source language, primarily by relying on surface-level error cues. Moreover, providing speech transcriptions can help users develop better mental models. Our results show the promise of cross-lingual question answering as a downstream task for studying MT mental models and advancing our understanding of human–AI collaboration.

Measuring User's Mental Models of Speech Translation in Human-AI Collaboration

We present PEAR (Pairwise Evaluation for Automatic Relative Scoring), a supervised quality estimation (QE) metric family that reframes reference-free machine translation (MT) evaluation as a graded pairwise comparison. Given a source segment and two candidate translations, PEAR predicts the direction and magnitude of their quality difference. The metrics are trained using pairwise supervision derived from differences in human judgments, with an additional regularization term that encourages sign inversion under candidate order reversal.
On the WMT24 meta-evaluation benchmark, PEAR outperforms strictly matched single-candidate QE baselines trained with the same data and backbones, isolating the benefit of the proposed pairwise formulation. Despite using substantially fewer parameters than recent large metrics, PEAR surpasses far larger QE models and reference-based metrics. Our analysis further indicates that PEAR yields a less redundant evaluation signal relative to other top metrics. 
Finally, we show that PEAR is an effective utility function for minimum Bayes risk (MBR) decoding, reducing pairwise scoring cost at negligible impact.

PEAR: Pairwise Evaluation for Automatic Relative Scoring in Machine Translation

Federated learning (FL) enables collaborative training across organizational silos without sharing raw data, making it attractive for privacy-sensitive applications. With the rapid adoption of large language models (LLMs), federated fine-tuning of generative LLMs has gained attention as a way to leverage distributed data while preserving confidentiality. However, this setting introduces fundamental challenges: (i) privacy leakage of personally identifiable information (PII) due to LLM memorization, and (ii) a persistent tension between global generalization and local utility under heterogeneous data. Existing defenses, such as data sanitization and differential privacy, reduce leakage but often degrade downstream performance. We propose SecureGate, a privacy-aware federated fine-tuning framework for LLMs that provides fine-grained privacy control without sacrificing utility. SecureGate employs a dual-adapter LoRA architecture: a secure adapter that learns sanitized, globally shareable representations, and a revealing adapter that captures sensitive, organization-specific knowledge. A token-controlled gating module selectively activates these adapters at inference time, enabling controlled information disclosure without retraining. Extensive experiments across multiple LLMs and real-world datasets show that SecureGate improves task utility while substantially reducing PII leakage, achieving up to a 31.66x reduction in inference attack accuracy and a 17.07x reduction in extraction recall for unauthorized requests. Additionally, it maintains 100% routing reliability to the correct adapter and incurs only minimal computational and communication overhead. Code is available at https://github.com/wsu-cyber-security-lab-ai/SecureGate.

SecureGate: Learning When to Reveal PII Safely via Token-Gated Dual-Adapters for Federated LLMs

Diffusion-based large language models offer a non-autoregressive alternative for text generation, but enabling them to perform complex reasoning remains challenging. Reinforcement learning has recently emerged as an effective post-training strategy for improving their performance; however, existing methods rely primarily on outcome-based rewards, which provide no direct supervision over the denoising process and often result in poorly structured reasoning that is difficult to interpret and inconsistently supports the final prediction. To address this limitation, we introduce \emph{denoising process reward}, a process-level reinforcement signal defined over the denoising trajectory of diffusion language models. This reward is obtained by estimating the contribution of intermediate denoising intervals to the final task outcome, encouraging the model to favor reasoning trajectories that consistently guide generation toward correct predictions. We further propose an efficient stochastic estimator that reuses standard training rollouts, enabling practical process-level supervision at scale. Experiments on challenging reasoning benchmarks demonstrate that our approach yields consistent improvements in reasoning stability, interpretability, and overall task performance.

 Advancing Reasoning in Diffusion Language Models with Denoising Process Rewards

Reinforcement Learning from Human Feedback (RLHF) is central to aligning Large Language Models (LLMs), yet it introduces a critical vulnerability: an imperfect Reward Model (RM) can become a single point of failure when it fails to penalize unsafe behaviors. While existing red-teaming approaches primarily target policy-level weaknesses, they overlook what we term systemic weaknesses cases where both the core LLM and the RM fail in tandem.

We present ARES, a framework that systematically discovers and mitigates such dual vulnerabilities. ARES employs a ``Safety Mentor'' that dynamically composes semantically coherent adversarial prompts by combining structured component types (topics, personas, tactics, goals) and generates corresponding malicious and safe responses. This dual-targeting approach exposes weaknesses in both the core LLM and the RM simultaneously. Using the vulnerabilities gained, ARES implements a two-stage repair process: first fine-tuning the RM to better detect harmful content, then leveraging the improved RM to optimize the core model. Experiments across multiple adversarial safety benchmarks demonstrate that ARES substantially enhances safety robustness while preserving model capabilities, establishing a new paradigm for comprehensive RLHF safety alignment.

ARES: Adaptive Red-Teaming and End-to-End Repair of Policy-Reward System

Large language models (LLMs) show significant potential in healthcare, prompting numerous benchmarks to evaluate their capabilities. However, concerns persist regarding the reliability of these benchmarks, which often lack clinical fidelity, robust data management, and safety-oriented evaluation metrics. To address these shortcomings, we introduce MedCheck, the first lifecycle-oriented assessment framework specifically designed for medical benchmarks. Our framework deconstructs benchmark development into five stages from design to governance, and provides a comprehensive checklist of 46 medically-tailored criteria. Using MedCheck, we conducted an in-depth empirical evaluation of 56 medical LLM benchmarks. Our analysis uncovers widespread, systemic issues, including a profound disconnect from clinical practice, a crisis of data integrity due to unmitigated contamination risks, and a systematic neglect of safety-critical evaluation dimensions like model robustness and uncertainty awareness. Based on these findings, MedCheck is both a diagnostic tool for existing benchmarks and an actionable guideline for a more standardized, reliable, and transparent approach to evaluating AI in healthcare.

Beyond the Leaderboard: Rethinking Medical Benchmarks for Large Language Models

As large language models (LLMs) increasingly imitate personal writing styles, personalization has become a key challenge for machine-generated text (MGT) detection. Yet personalized MGT detection remains largely underexplored. In this work, we introduce StyloBench, the first benchmark for evaluating detector robustness under personalization, built from literary and blog texts paired with their LLM-generated imitations. Experiments across diverse detectors show pronounced performance instability under personalization, with frequent inversions relative to general-domain behavior. To better understand this limitation, we conduct an in-depth analysis and attribute it to a feature-inversion trap, i.e., features that are effective for separating human-written text (HWT) from MGT in general flip their effect in personalized contexts, ultimately misleading detectors. Motivated by this, we propose StyloCheck, a diagnostic framework for predicting detector robustness under personalization. StyloCheck identifies the inverted features and quantifies detector dependence using perturbed texts pronounced in the features. In our experiments, StyloCheck predicts both the direction and magnitude of cross-domain performance shifts with an 85% correlation to actual outcomes. We hope this work will raise awareness of the structural risks introduced by personalization and motivate more robust approaches to personalized MGT detection. The code is available at: https://github.com/mbzuai-nlp/Personalized_MGT_Detect

Downloads

Next from ACL 2026

How Do Inpainting Artifacts Propagate to Language?

Similar lecture

Self-Training Meets Consistency: Improving LLMs’ Reasoning with Consistency-Driven Rationale Evaluation