China

Providing constructive feedback to paper authors is a core component of peer review. With reviewers increasingly having less time to perform reviews, automated support systems are required to ensure high reviewing quality, thus making the feedback in reviews useful for authors. To this end, we identify four key aspects of review comments (individual points in weakness sections of reviews) that drive the utility for authors: Actionability, Grounding &amp; Specificity, Verifiability, and Helpfulness. To enable evaluation and development of models assessing review comments, we introduce the RevUtil dataset. We collect 1,430 human-labeled review comments and scale our data with 10k synthetically labeled comments for training purposes. The synthetic data additionally contains rationales, i.e., explanations for the aspect score of a review comment. Employing the RevUtil dataset, we benchmark fine-tuned models for assessing review comments on these aspects and generating rationales. Our experiments demonstrate that these fine-tuned models achieve agreement levels with humans comparable to, and in some cases exceeding, those of powerful closed models like GPT-4o. Our analysis further reveals that GPT-4-generated reviews generally underperform human reviews on our four aspects.

EMNLP 2025

The Good, the Bad and the Constructive: Automatically Measuring Peer Review&#39;s Utility for Authors

Providing constructive feedback to paper authors is a core component of peer review. With reviewers increasingly having less time to perform reviews, automated support systems are required to ensure high reviewing quality, thus making the feedback in reviews useful for authors. To this end, we identify four key aspects of review comments (individual points in weakness sections of reviews) that drive the utility for authors: Actionability, Grounding & Specificity, Verifiability, and Helpfulness. To enable evaluation and development of models assessing review comments, we introduce the RevUtil dataset. We collect 1,430 human-labeled review comments and scale our data with 10k synthetically labeled comments for training purposes. The synthetic data additionally contains rationales, i.e., explanations for the aspect score of a review comment. Employing the RevUtil dataset, we benchmark fine-tuned models for assessing review comments on these aspects and generating rationales. Our experiments demonstrate that these fine-tuned models achieve agreement levels with humans comparable to, and in some cases exceeding, those of powerful closed models like GPT-4o. Our analysis further reveals that GPT-4-generated reviews generally underperform human reviews on our four aspects.

The Good, the Bad and the Constructive: Automatically Measuring Peer Review's Utility for Authors

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."<br>

*Message from the General Chair, Dirk Hovy*

[**Link to Conference Handbook**](https://drive.google.com/file/d/1johU5QqVVYO4RfH7QcIORr7qrVBdzdwC/view?usp=sharing)





<br>

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

Large Language Models (LLMs) require robust evaluation. However, existing frameworks often rely on curated datasets that, once public, may be accessed by newer LLMs. This creates a risk of data leakage, where test sets inadvertently become part of training data, compromising evaluation fairness and integrity. To mitigate this issue, we propose Behave as Claimed (BaC), a novel evaluation framework inspired by counterfactual reasoning. BaC constructs a ``what-if'' scenario where LLMs respond to counterfactual questions about how they would behave if the input were manipulated. We refer to these responses as claims, which are verifiable by observing the LLMs' actual behavior when given the manipulated input. BaC dynamically generates and verifies counterfactual questions using various few-shot in-context learning evaluation datasets, reducing their susceptibility to data leakage. Moreover, BaC provides a more challenging evaluation paradigm for LLMs. LLMs must thoroughly understand the prompt, the task, and the consequences of their responses to achieve better performance. We evaluate several state-of-the-art LLMs and find that, while most perform well on the original datasets, they struggle with BaC. This suggests that LLMs usually fail to align their claims with their actual behavior and that high performance on standard datasets may be less stable than previously assumed.

Do LLMs Behave as Claimed? Investigating How LLMs Follow Their Own Claims using Counterfactual Questions

The impact of Large Language Models (LLMs) has extended into literary domains. However, existing evaluation metrics prioritize mechanical accuracy over artistic expression and tend to overrate machine translation as being superior to human translation from experienced professionals. In the long run, this bias could result in an irreversible decline in translation quality and cultural authenticity. In response to the urgent need for a specialized literary evaluation metric, we introduce LiTransProQA, a novel, reference-free, LLM-based question-answering framework designed for literary translation evaluation. LiTransProQA uniquely integrates insights from professional literary translators and researchers, focusing on critical elements in literary quality assessment such as literary devices, cultural understanding, and authorial voice. Our extensive evaluation shows that while literary-finetuned XCOMET-XL yields marginal gains, LiTransProQA substantially outperforms current metrics, achieving up to 0.07 gain in correlation and surpassing the best state-of-the-art metrics by over 15 points in adequacy assessments. Incorporating professional translator insights as weights further improves performance, highlighting the value of translator inputs. Notably, LiTransProQA reaches human-level evaluation performance comparable to trained student evaluators. It shows broad applicability to open-source models like LLaMA3.3-70b and Qwen2.5-32b, indicating its potential as an accessible and training-free tool for evaluating literary translations that require local processing due to copyright or ethical considerations.

LiTransProQA: An LLM-based Literary Translation Evaluation Metric with Professional Question Answering

Mechanistic interpretability research faces a gap between analyzing simple circuits in toy tasks and discovering features in large models. To bridge this gap, we propose text-to-SQL generation as an ideal task to study, as it combines the formal structure of toy tasks with real-world complexity. We introduce TinySQL, a synthetic dataset, progressing from basic to advanced SQL operations, and train models ranging from 33M to 1B parameters to establish a comprehensive testbed for interpretability. We apply multiple complementary interpretability techniques, including Edge Attribution Patching and Sparse Autoencoders, to identify minimal circuits and components supporting SQL generation. We compare circuits for different SQL subskills, evaluating their minimality, reliability, and identifiability. Finally, we conduct a layerwise logit lens analysis to reveal how models compose SQL queries across layers: from intent recognition to schema resolution to structured generation. Our work provides a robust framework for probing and comparing interpretability methods in a structured, progressively complex setting.

TinySQL: A Progressive Text-to-SQL Dataset for Mechanistic Interpretability Research

Natural language explanations (NLEs) are commonly used to provide plausible free-text explanations of a model's reasoning about its predictions. However, recent work has questioned their faithfulness, as they may not accurately reflect the model’s internal reasoning process regarding its predicted answer. In contrast, highlight explanations--input fragments critical for the model's predicted answers--exhibit measurable faithfulness. Building on this foundation, we propose G-Tex, a Graph-Guided Textual Explanation Generation framework designed to enhance the faithfulness of NLEs. Specifically, highlight explanations are first extracted as faithful cues reflecting the model's reasoning logic toward answer prediction. They are subsequently encoded through a graph neural network layer to guide the NLE generation, which aligns the generated explanations with the model’s underlying reasoning toward the predicted answer. Experiments on T5 and BART using three reasoning datasets show that G-Tex improves NLE faithfulness by up to 12.18% compared to baseline methods. Additionally, G-Tex generates NLEs with greater semantic and lexical similarity to human-written ones. Human evaluations show that G-Tex can decrease redundant content and enhance the overall quality of NLEs. Our work presents a novel method for explicitly guiding NLE generation to enhance faithfulness, serving as a foundation for addressing broader criteria in NLE and generated text.

Graph-Guided Textual Explanation Generation Framework

The ability of large language models (LLMs) to validate their output and identify potential errors is crucial for ensuring robustness and reliability. However, current research indicates that LLMs struggle with self-correction, encountering significant challenges in detecting errors. While studies have explored methods to enhance self-correction in LLMs, relatively little attention has been given to understanding the models' internal mechanisms underlying error detection. In this paper, we present a mechanistic analysis of error detection in LLMs, focusing on simple arithmetic problems. Through circuit analysis, we identify the computational subgraphs responsible for detecting arithmetic errors across four smaller-sized LLMs. Our findings reveal that all models heavily rely on textitconsistency headstextemdashattention heads that assess surface-level alignment of numerical values in arithmetic solutions. Moreover, we observe that the models' internal arithmetic computation primarily occurs in higher layers, whereas validation takes place in middle layers, before the final arithmetic results are fully encoded. This structural dissociation between arithmetic computation and validation seems to explain why smaller-sized LLMs struggle to detect even simple arithmetic errors.

The Validation Gap: A Mechanistic Analysis of How Language Models Compute Arithmetic but Fail to Validate It

We investigate whether large language models encode latent knowledge of frame semantics, focusing on frame identification—a core challenge in Frame Semantic Parsing that involves selecting the appropriate semantic frame for a target word in context. Using the FrameNet lexical resource, we evaluate models under prompt-based inference and observe that they can perform frame disambiguation effectively even without explicit supervision. To assess the impact of task-specific training, we fine tune the model on FrameNet data and fine tuning substantially improves in-domain accuracy while generalizing well to out-of-domain benchmarks. Further analysis reveals that the model can generate semantically coherent frame definitions and remains partially robust to corrupted frame labels, suggesting an internalized understanding of frame semantics beyond surface cues.

Do LLMs Encode Frame Semantics? Evidence from Frame Identification

Safety fine-tuning algorithms reduce harmful outputs in language models, yet their mechanisms remain under-explored. Direct Preference Optimization (DPO) is a popular choice of algorithm, but prior explanations—attributing its effects solely to dampened toxic neurons in the MLP layers—are incomplete. In this study, we analyse four language models (Llama-3.1-8B, Gemma-2-2B, Mistral-7B, GPT-2-Medium) and show that toxic neurons only account for 2.5% to 24% of DPO's effects across models. Instead, DPO induces distributed activation shifts across all MLP neurons to create a net toxicity reduction. We attribute this reduction to four neuron groups—two aligned with reducing toxicity and two promoting anti-toxicity—whose combined effects replicate DPO across models. To further validate this understanding, we develop an activation editing method that mimics DPO through distributed shifts along a toxicity representation. This method outperforms DPO in reducing toxicity while preserving perplexity, without requiring any weight updates. Our work provides a mechanistic understanding of DPO and introduces an efficient, tuning-free alternative for safety fine-tuning.

How Does DPO Reduce Toxicity? A Mechanistic Neuron-Level Analysis

A critical component in the trustworthiness of LLMs is reliable uncertainty communication, yet LLMs often use assertive language when conveying false claims, leading to over-reliance and eroded trust. We present the first systematic study of _faithful confidence calibration_ of LLMs, benchmarking models' ability to use linguistic expressions of uncertainty that _faithfully reflect_ their intrinsic uncertainty, across a comprehensive array of models, datasets, and prompting strategies. Our results demonstrate that LLMs largely fail at this task, and that existing interventions are insufficient: standard prompt approaches provide only marginal gains, and existing, factuality-based calibration techniques can even harm faithful calibration. To address this critical gap, we introduce MetaFaith, a novel prompt-based calibration approach inspired by human metacognition. We show that MetaFaith robustly improves faithful calibration across diverse models and task domains, enabling up to 61% improvement in faithfulness and achieving an 83% win rate over original generations as judged by humans.

MetaFaith: Faithful Natural Language Uncertainty Expression in LLMs

One of the core goals of crisis counseling services is to support emotional de-escalation of the individual in crisis, by reducing intense negative emotional affect and emotional dysregulation. The science of crisis intervention has been impeded, however, by a lack of quantitative approaches that allow for detailed analysis of emotion in crisis conversations. In order to measure de-escalation at scale (millions of text-based conversations), lightweight models are needed that can assign not just binary sentiment predictions but quantitative scores to capture graded change in emotional valence. Accordingly, we developed a transformer-based emotional valence scoring model fit for crisis conversations, BERT-EV, that assigns numerical emotional valence scores to rate the intensity of expressed negative versus positive emotion. This transformer-based model can run on modest hardware configurations, allowing it to scale affordably and efficiently to a massive corpus of crisis conversations. We evaluated model performance on a corpus of hand-scored social media messages, and found that BERT-EV outperforms existing dictionary-based standard tools in the field, as well as other transformer-based implementations and an LLM in accurately matching scores from human annotators. Finally, we show that trends in these emotional valence scores can be used to assess emotional de-escalation during crisis conversations, with sufficient turn-by-turn granularity to help identify helpful vs. detrimental crisis counselor statements.

Assessing effective de-escalation of crisis conversations using transformer-based models and trend statistics

Hallucination remains a critical failure mode of large language models (LLMs), undermining their trustworthiness in real-world applications. In this work, we focus on confabulation, a foundational aspect of hallucination where the model fabricates facts about unknown entities. We introduce a targeted dataset designed to isolate and analyze this behavior across diverse prompt types. Using this dataset, and building on recent progress in interpreting LLM internals, we extract latent directions associated with confabulation using sparse projections. A simple vector-based steering method demonstrates that these directions can modulate model behavior with minimal disruption, shedding light on the inner representations that drive factual and non-factual output. Our findings contribute to a deeper mechanistic understanding of LLMs and pave the way toward more trustworthy and controllable generation. We release the code and dataset at https://anonymous.4open.science/r/Confabulation-discovery.

Downloads

Next from EMNLP 2025

Do LLMs Behave as Claimed? Investigating How LLMs Follow Their Own Claims using Counterfactual Questions

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES