China

Large Language Models (LLMs) are expected to provide helpful and harmless responses, yet they often exhibit \textit{sycophancy}—conforming to user beliefs regardless of factual accuracy or ethical soundness. Prior research on sycophancy has primarily focused on single-turn factual correctness, overlooking the dynamics of real-world interactions. In this work, we introduce \textbf{SYCON Bench} (\textbf{SY}cophantic \textbf{CON}formity benchmark), a novel evaluation suite that assesses sycophantic behavior in multi-turn, free-form conversational settings. Our benchmark measures how quickly a model conforms to the user (\textit{Turn of Flip}) and how frequently it shifts its stance under sustained user pressure (\textit{Number of Flip}). Applying \textsc{SYCON Bench} to 17 LLMs across three real-world scenarios, we find that sycophancy remains a prevalent failure mode. Our analysis shows that alignment tuning amplifies sycophantic behavior, whereas model scaling and reasoning optimization strengthen the model&#39;s ability to resist undesirable user views. Reasoning models generally outperform instruction-tuned models but often fail when they over-index on logical exposition instead of directly addressing the user&#39;s underlying beliefs. Finally, we evaluate four additional prompting strategies and demonstrate that adopting a third-person perspective reduces sycophancy by up to 63.8% in debate scenario.

EMNLP 2025

Measuring Sycophancy of Language Models in Multi-turn Dialogues

sycophancy

llm evaluation

benchmarking

### 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

Large Language Models (LLMs) are expected to provide helpful and harmless responses, yet they often exhibit \textit{sycophancy}—conforming to user beliefs regardless of factual accuracy or ethical soundness. Prior research on sycophancy has primarily focused on single-turn factual correctness, overlooking the dynamics of real-world interactions. In this work, we introduce \textbf{SYCON Bench} (\textbf{SY}cophantic \textbf{CON}formity benchmark), a novel evaluation suite that assesses sycophantic behavior in multi-turn, free-form conversational settings. Our benchmark measures how quickly a model conforms to the user (\textit{Turn of Flip}) and how frequently it shifts its stance under sustained user pressure (\textit{Number of Flip}). Applying \textsc{SYCON Bench} to 17 LLMs across three real-world scenarios, we find that sycophancy remains a prevalent failure mode. Our analysis shows that alignment tuning amplifies sycophantic behavior, whereas model scaling and reasoning optimization strengthen the model's ability to resist undesirable user views. Reasoning models generally outperform instruction-tuned models but often fail when they over-index on logical exposition instead of directly addressing the user's underlying beliefs. Finally, we evaluate four additional prompting strategies and demonstrate that adopting a third-person perspective reduces sycophancy by up to 63.8% in debate scenario.

poster

## Welcome!
"I am excited to welcome you to this year’s edition of the Conference on Empirical Methods in Natural Language Processing! Importantly, it marks the 30th edition of EMNLP. With over 8,000 submissions, more than 3,000 accepted papers, and thousands of attendees, we have come a long way from that first
workshop, which had 14 accepted papers. As the field looks ahead, Suzhou is the fitting location for celebrating this milestone: rooted in a long literary tradition, yet modern and forward-looking, and home to a large share of the NLP community." 

*Message from the General Chair, Dirk Hovy*

Celebrate 30 Years of EMNLP! 
EMNLP 2025 will be held in Suzhou, China from November 5th to November 9th, 2025.

Chat-oriented dialogue systems that deliver tangible benefits, such as sharing news or frailty prevention for seniors, require Proactive acquisition of specific user Information Via chats On user-favored Topics (PIVOT). This study proposes the PIVOT task to support the development of these systems. In this task, a system needs to acquire a user's answers to predefined questions without making the user feel abrupt while engaging in a chat on a predefined topic. We created and analyzed a dataset of 650 PIVOT chats, identifying key challenges and effective strategies for recent LLMs. Our system, designed from these insights, surpassed the performance of LLMs prompted solely with task instructions. Finally, we demonstrate that automatic evaluation of this task is reasonably accurate, suggesting its potential as a framework to efficiently develop techniques for systems dealing with complex dialogue goals, extending beyond the scope of PIVOT alone.

Proactive User Information Acquisition via Chats on User-Favored Topics

Large Language Models (LLMs) such as GPT and LLaMA excel in natural language tasks, e.g., text generation and machine translation. However, inherent biases from training on vast Internet datasets potentially amplify harmful stereotypes—widely held, oversimplified, and often inaccurate generalizations about groups of people. Our contribution introduces a novel, systematic, and architecture-aware method to identify and mitigate stereotypical bias in decoder-only transformer models. This interpretable approach operates without gradient access or retraining from scratch. We first evaluate bias and then apply a bias localization mechanism that correlates internal activations with a newly defined Context Influence (CI) Score. Our method pinpoints specific attention heads that consistently align with biased shifts in model predictions. To mitigate this, we introduce a soft pruning strategy that scales attention head parameters based on their correlation strength, followed by lightweight fine-tuning to maintain fluent text generation. Experiments across five models demonstrate our approach reduces bias by up to 37% on BBQ, 32% on StereoSet, and 33% on CrowS-Pairs while simultaneously improving reasoning performance on MMLU by up to 10%. \footnote{Data and code are avaliable here \url{https://anonymous.4open.science/status/mitigate_bias}}

Toward Inclusive Language Models: Sparsity-Driven Calibration for Systematic and Interpretable Mitigation of Social Biases in LLMs

Recent research has attempted to associate preference optimization (PO) performance with the underlying preference datasets. In this work, our observation is that the differences between the preferred response y⁺ and dispreferred response y⁻ influence what LLMs can learn, which may not match the desirable differences to learn. Therefore, we use distance and reward margin to quantify these differences, and combine them to get Distance Calibrated Reward Margin (DCRM), a metric that measures the quality of a response pair for PO. Intuitively, DCRM encourages minimal noisy differences and maximal desired differences. With this, we study 3 types of commonly used preference datasets, classified along two axes: the source of the responses and the preference labeling function. We establish a general correlation between higher DCRM of the training set and better learning outcome. Inspired by this, we propose a best-of-N² pairing method that selects response pairs with the highest DCRM. Empirically, in various settings, our method produces training datasets that can further improve models' performance on AlpacaEval, MT-Bench, and Arena-Hard over the existing training sets.

DCRM: A Heuristic to Measure Response Pair Quality in Preference Optimization

Adapting visual programming or prompting large language models (LLMs) to generate executable code for visual tasks like visual question answering (VQA) for specialized tasks or domains remains challenging due to high annotation and inference costs. We propose a low-cost visual program distillation method that can be used for models with at most 1 billion parameters and requires no human-generated program annotations. We achieve this through synthetic data augmentation based on decoupling programs into higher-level skills, called templates, and their corresponding arguments. Experimental results show that, with a relatively small amount of question/answer data, small language models can generate high-quality specialized visual programs with the added benefit of much faster inference.

Visual Program Distillation with Template-Based Augmentation

Applying deep learning to malware detection has drawn great attention due to its notable performance. With the increasing prevalence of cyberattacks targeting IoT devices, there is a parallel rise in the development of malware across various Instruction Set Architectures (ISAs). It is thus important to extend malware detection capacity to multiple ISAs. However, training a deep learning-based malware detection model usually requires a large number of labeled malware samples. The process of collecting and labeling sufficient malware samples to build datasets for each ISA is labor-intensive and time-consuming. To reduce the burden of data collection, we propose to leverage the ideas of Neural Machine Translation (NMT) and Normalizing Flows (NFs) for malware detection. Specifically, when dealing with malware in a certain ISA, we translate it to an ISA with sufficient malware samples (like X86-64). This allows us to apply a model trained on one ISA to analyze malware from another ISA. Our approach reduces the data collection effort by enabling malware detection across multiple ISAs using a model trained on a single ISA.

FlowMalTrans: Unsupervised Binary Code Translation for Malware Detection Using Flow-Adapter Architecture

As language models become integral to critical workflows, assessing their behavior remains a fundamental challenge -- human evaluation is costly and noisy, while automated metrics provide only coarse, difficult-to-interpret signals. We introduce \textbf{natural language unit tests}, a paradigm that decomposes response quality into explicit, testable criteria, along with a unified scoring model, LMUnit, which combines multi-objective training across preferences, direct ratings, and natural language rationales. Through controlled human studies, we show this paradigm significantly improves inter-annotator agreement and enables more effective LLM development workflows. LMunit achieves state-of-the-art performance on evaluation benchmarks (FLASK, BigGenBench) and competitive results on RewardBench. These results validate both our proposed paradigm and scoring model, suggesting a promising path forward for language model evaluation and development.

LMUNIT: Fine-grained Evaluation with Natural Language Unit Tests

The advancement of Large Language Models (LLMs) has made ensuring their trustworthiness increasingly critical, especially in terms of fairness across diverse human groups. While modern LLMs are aligned with user preferences through Reinforcement Learning from Human Feedback (RLHF), the reward models used for alignment are trained on preference data that may both reflect societal biases and suffer from demographic skewness, as labeler populations are often uneven due to systemic accessibility or participation gaps. In this work, we reveal that reward models can exhibit significant discrepancies across different demographic groups, posing a fundamental challenge to fair and robust alignment. Using real-world datasets, we conduct the most comprehensive study to date, auditing various state-of-the-art reward models across nine sensitive attributes, including age, gender, ethnicity, etc. Our evaluation spans both (1) the agreement level between reward models and specific user groups, and (2) the reward model's preference toward responses associated with different groups. Based on these findings, we propose the first method to mitigate group disparities in reward modeling. Code submitted as software.

Not All Voices Are Rewarded Equally: Probing and Repairing Reward Models across Human Diversity

Hallucinations in large language models (LLMs) pose significant challenges in tasks requiring complex multi-step reasoning, such as mathematical problem-solving. Existing approaches primarily detect the presence of hallucinations but lack a nuanced understanding of their types and manifestations. In this paper, we first introduce a comprehensive taxonomy that categorizes the common hallucinations in mathematical reasoning tasks into six types. We then propose FG-PRM (\textbf{F}ine-\textbf{G}rained \textbf{P}rocess \textbf{R}eward \textbf{M}odel), an augmented model designed to detect and mitigate hallucinations in a fine-grained, step-level manner. To address the limitations of manually labeling training data, we propose an automated method for generating fine-grained hallucination data using LLMs. Our FG-PRM demonstrates superior performance across two key tasks: 1) Fine-grained hallucination detection: classifying hallucination types for each reasoning step; and 2) Verification: ranking multiple LLM-generated outputs to select the most accurate solution. Our experiments show that FG-PRM excels in fine-grained hallucination detection and substantially boosts the performance of LLMs on GSM8K and MATH benchmarks. These results highlight the benefits of fine-grained supervision in enhancing the reliability and interpretability of LLM reasoning processes.

FG-PRM: Fine-grained Hallucination Detection and Mitigation in Language Model Mathematical Reasoning

Large language models (LLMs) have demonstrated an impressive ability to role-play humans and replicate complex social dynamics. However, large-scale LLM-driven simulations still face significant challenges in high time and computational costs. We observe that there exists redundancy in current agent communication: when expressing the same intention, agents tend to use lengthy and repetitive language, whereas humans naturally prefer concise expressions. To this end, we propose EcoLANG: Efficient and Effective Agent Communication Language Induction for Social Simulation. Inspired by how human language evolves through interactions, we induce a more compact language by identifying and preserving core communicative concepts at the vocabulary level and evolving efficient expression patterns at the sentence level through natural selection. We apply the induced language in various social simulations. Experimental results demonstrate that EcoLANG reduces token consumption by over 20%, enhancing efficiency without sacrificing simulation accuracy.

EcoLANG: Efficient and Effective Agent Communication Language Induction for Social Simulation

Large language models (LLMs) excel across many natural language processing tasks but face challenges in domain-specific, analytical tasks such as conducting research surveys. This study introduces ResearchArena, a benchmark designed to evaluate LLMs' capabilities in conducting academic surveys—a foundational step in academic research. ResearchArena models the process in three stages: (1) information discovery, identifying relevant literature; (2) information selection, evaluating papers' relevance and impact; and (3) information organization, structuring knowledge into hierarchical frameworks such as mind-maps. Notably, mind-map construction is treated as a bonus task, reflecting its supplementary role in survey-writing. To support these evaluations, we construct an offline environment of 12M full-text academic papers and 7.9K survey papers. To ensure ethical compliance, we do not redistribute copyrighted materials; instead, we provide code to construct the environment from the Semantic Scholar Open Research Corpus (S2ORC). Preliminary evaluations reveal that LLM-based approaches underperform compared to simpler keyword-based retrieval methods, though recent reasoning models such as DeepSeek-R1 show slightly better zero-shot performance. These results underscore significant opportunities for advancing LLMs in autonomous research.

Downloads

Next from EMNLP 2025

Proactive User Information Acquisition via Chats on User-Favored Topics

Similar lecture

TRUTH DECAY: Quantifying Multi-Turn Sycophancy in Language Models

Stay up to date with the latest Underline news!

PRODUCT

COMPANY

RESOURCES