China

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&#39; 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.

EMNLP 2025

MetaFaith: Faithful Natural Language Uncertainty Expression in LLMs

safety and alignment

explanation faithfulness

free-text/natural language explanations

calibration/uncertainty

benchmarking

evaluation

## Welcome to EMNLP 2024! 
We are excited to welcome you to one of the most prominent conferences in the field of Natural Language Processing. This year, EMNLP 2024 is being held in a hybrid format,
offering both virtual and in-person participation in beautiful Miami. Due to a record-breaking number of submissions, we've expanded the total number of accepted papers to accommodate more cutting-edge research from around the globe.
### [Conference Handbook](https://drive.google.com/file/d/1WPROgxjLAC96AJL7Ugy0tEnYm7dkrbHt/view?usp=sharing)

You are required to register for this event. **Please register [here](https://2024.emnlp.org/registration/).** The EMNLP 2024 event page on Underline will be open to public one week prior to the event.

Please register!

EMNLP 2024

EMNLP 2024 will take place in Miami, Florida from Nov 12th to Nov 16th, 2024, at the Hyatt Regency Miami Hote and on Underline for remote participants.

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.

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.

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.

Where Confabulation Lives: Latent Feature Discovery in LLMs

Recent work has demonstrated that using chain of thought (CoT), on soft-reasoning problems such as analytical and commonsense reasoning, often yields limited or even negative performance gains. CoT can also be unfaithful to the model’s actual reasoning. This paper investigates dynamics and unfaithfulness of CoT in soft-reasoning tasks across instruction-tuned, reasoning and reasoning-distilled models. Our findings show that distilled‑reasoning models rely heavily on CoT for these tasks, while instruction‑tuned and reasoning models often use it post‑hoc. Additionally, we find that CoT can steer model predictions without faithfully reflecting reasoning, indicating a disconnect between CoT influence and faithfulness.

Analysing Chain of Thought Dynamics: Active Guidance or Unfaithful Post-hoc Rationalisation?

Reasoning over temporal and numerical data, such as time series, is a crucial aspect of fact-checking. While many systems have recently been developed to handle this form of evidence, their evaluation remains limited by existing datasets, which often lack structured evidence, provide insufficient justifications for verdicts, or rely on synthetic claims. In this paper, we introduce TSVer, a new benchmark dataset for fact verification focusing on temporal and numerical reasoning with time-series evidence. TSVer contains 287 real-world claims sourced from 25 fact-checking organizations and a curated database of 152 time series covering diverse domains. Each claim is annotated with time frames across all pertinent time series, along with a verdict and justifications reflecting how the evidence is used to reach the verdict. Using an LLM-assisted multi-step annotation process, we improve the quality of our annotations and achieve an inter-annotator agreement of 0.709 on verdicts. We also develop a baseline for verifying claims against time-series evidence and show that even the state-of-the-art reasoning models like Gemini-2.5-Pro are challenged by time series, achieving a 63.37 accuracy score on verdicts and an Ev2R score of 48.63 on verdict justifications.

TSVer: A Benchmark for Fact Verification Against Time-Series Evidence

Sparse Autoencoders (SAEs) provide potential for uncovering structured, human-interpretable representations in Large Language Models (LLMs), making them a crucial tool for transparent and controllable AI systems. We systematically analyze SAE for interpretable feature extraction from LLMs in safety-critical classification tasks. Our framework evaluates (1) model-layer selection and scaling properties, (2) SAE architectural configurations, including width and pooling strategies, and (3) the effect of binarizing continuous SAE activations. SAE-derived features achieve macro F1 > 0.8, outperforming hidden-state and BoW baselines while demonstrating cross-model transfer from Gemma 2 2B to 9B-IT models. These features generalize in a zero-shot manner to cross-lingual toxicity detection and visual classification tasks. Our analysis highlights the significant impact of pooling strategies and binarization thresholds, showing that binarization offers an efficient alternative to traditional feature selection while maintaining or improving performance. These findings establish new best practices for SAE-based interpretability and enable scalable, transparent deployment of LLMs in real-world applications.

Sparse Autoencoder Features for Classifications and Transferability

Figurative language conveys stance, emotion, and social nuance, making its appropriate use essential in dialogue. While large language models (LLMs) often succeed in recognizing figurative expressions at the sentence level, their ability to use them coherently in conversation remains uncertain. We introduce FLUID QA, the first multilingual benchmark that evaluates figurative usage in dialogue across English, Korean, and Chinese. Each item embeds figurative choices into multi-turn contexts. To support interpretation, we include FLUTE-bi, a sentence-level diagnostic task. Results reveal a persistent gap: models that perform well on FLUTE-bi frequently fail on FLUID QA, especially in sarcasm and metaphor. These errors reflect systematic rhetorical confusion and limited discourse reasoning. FLUID QA provides a scalable framework for assessing usage-level figurative competence across languages.

FLUID QA: A Multilingual Benchmark for Figurative Language Usage in Dialogue across English, Chinese, and Korean

Large language models (LLMs) often behave inconsistently across inputs, indicating uncertainty and motivating the need for its quantification in high-stakes settings. Prior work on calibration and uncertainty quantification often focuses on individual models, overlooking the potential of model diversity. We hypothesize that LLMs make complementary predictions due to differences in training and the Zipfian nature of language, and that aggregating their outputs leads to more reliable uncertainty estimates. To leverage this, we propose MUSE (Multi-LLM Uncertainty via Subset Ensembles), a simple information-theoretic method that uses Jensen-Shannon Divergence to identify and aggregate well-calibrated subsets of LLMs. Experiments on binary prediction tasks demonstrate improved calibration and predictive performance compared to single-model and naïve ensemble baselines.

Simple Yet Effective: An Information-Theoretic Approach to Multi-LLM Uncertainty Quantification

Large language models (LLMs) demonstrate proficiency across numerous computational tasks, yet their inner workings remain unclear. In theory, the combination of causal self-attention and multilayer perceptron allows every token to access and compute information based on all preceding tokens. In practice, to what extent are such operations present? In this paper, on mental math tasks (i.e., direct math calculation via next-token prediction without explicit reasoning), we investigate this question in three steps: inhibiting input-specific token computations in the initial layers, restricting the routes of information transfer in the next few layers, and forcing all computation to happen at the last token in the remaining layers. With two proposed techniques, Context-Aware Mean Ablation (CAMA) and Attention-Based Peeking (ABP), we identify an All-for-One subgraph (AF1) with high accuracy on a wide variety of mental math tasks, where meaningful computation occurs very late (in terms of layer depth) and only at the last token, which receives information of other tokens in few specific layers. Experiments show that this circuit is sufficient and necessary for high model performance, transfers across different models, and works on a variety of input styles. Ablations on different CAMA and ABP alternatives reveal their unique advantages over other methods, which may be of independent interest.

All for One: LLMs Solve Mental Math at the Last Token With Information Transferred From Other Tokens

Misgendering is the act of referring to someone by a gender that does not match their chosen identity. It marginalizes and undermines a person's sense of self, causing significant harm. English-based approaches have clear-cut approaches to avoiding misgendering, such as the use of the pronoun ``they''. However, other languages pose unique challenges due to both grammatical and cultural constructs. In this work we develop methodologies to assess and mitigate misgendering across 42 languages and dialects using a participatory-design approach to design effective and appropriate guardrails across all languages. We test these guardrails in a standard LLM-based application (meeting transcript summarization), where both the data generation and the annotation steps followed a human-in-the-loop approach. We find that the proposed guardrails are very effective in reducing misgendering rates across all languages in the summaries generated, and without incurring loss of quality. Our human-in-the-loop approach demonstrates a method to feasibly scale inclusive and responsible AI-based solutions across multiple languages and cultures. We release the guardrails and synthetic dataset encompassing 42 languages, along with human and LLM-judge evaluations, to encourage further research on this subject.

A Multilingual, Culture-First Approach to Addressing Misgendering in LLM Applications

In this work, we compile textbf\texttt{DroidCollection}, the most extensive open data suite for training and evaluating machine-generated code detectors, comprising over a million code samples, seven programming languages, outputs from 43 coding models, and over three real-world coding domains. Alongside fully AI-generated samples, our collection includes human-AI co-authored code, as well as adversarial samples explicitly crafted to evade detection. Subsequently, we develop textbf\texttt{DroidDetect}, a suite of encoder-only detectors trained using a multi-task objective over textttDroidCollection. Our experiments show that existing detectors' performance fails to generalise to diverse coding domains and programming languages outside of their narrow training data. Additionally, we demonstrate that while most detectors are easily compromised by humanising the output distributions using superficial prompting and alignment approaches, this problem can be easily amended by training on a small amount of adversarial data. Finally, we demonstrate the effectiveness of metric learning and uncertainty-based resampling as means to enhance detector training on possibly noisy distributions.

Downloads

Next from EMNLP 2025

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

Similar lecture

Does Alignment Tuning Really Break LLMs’ Internal Confidence?