China

Recent works have suggested that In-Context Learning (ICL) operates in dual modes, i.e. task retrieval (remember learned patterns from pre-training) and task learning (inference-time “learning&quot; from demonstrations). However, disentangling these the two modes remains a challenging goal. We introduce ICL CIPHERS, a class of task reformulations based on substitution ciphers borrowed from classic cryptography. In this approach, a subset of tokens in the in-context inputs are substituted with other (irrelevant) tokens, rendering English sentences less comprehensible to human eye. However, by design, there is a latent, fixed pattern to this substitution, making it reversible. This bijective (reversible) cipher ensures that the task remains a well-defined task in some abstract sense, despite the transformations. It is a curious question if LLMs can solve ICL CIPHERS with a BIJECTIVE mapping, which requires deciphering the latent cipher. We show that LLMs are better at solving ICL CIPHERS with BIJECTIVE mappings than the NON-BIJECTIVE (irreversible) baseline, providing a novel approach to quantify “learning” in ICL. While this gap is small, it is consistent across the board on four datasets and six models. Finally, our interpretability analysis shows evidence that LLMs can internally decode ciphered inputs.

EMNLP 2025

ICL CIPHERS: Quantifying &#39;&#39;Learning&#39;&#39; in In-Context Learning via Substitution Ciphers

task retrieval

in-context learning

task learning

## Welcome to EMNLP 2022!
I am delighted to welcome you to EMNLP 2022! I believe this conference has been complicated beyond any precedent. Over the past year, it’s been thrilling to see the organization team approach each new puzzle with creativity and enthusiasm. We hope that those participating in Abu Dhabi as well as those joining remotely will leave the conference feeling newly inspired by the program and newly connected to our ever-growing community. Following EMNLP 2021 and major NLP conferences since, EMNLP 2022 is “hybrid,” serving both virtual and in-person participants.

Our key innovations for EMNLP 2022 include:

* EMNLP 2022 is “hybrid” in a second sense, as well: we allowed both direct and rolling review paper submissions, building on the pilot experiment of EMNLP 2021, which considered a small number of ARR submissions. 
* Familiar from NAACL but new to EMNLP, we’ve added an industry track.
* During the conference, “portals” will link virtual poster sessions to in-person conference participants during poster sessions each day.
* The first ACL-family conference in the United Arab Emirates.

 *Message from Noah A. Smith, University of Washington and Allen Institute for AI, Seattle, Washington, USA* 
***EMNLP 2022 General Chair***
 
[![](https://assets.underline.io/uploads/markdown_image/1/image/9eec7d4a287ee18c278b08229290aa83.png)](https://drive.google.com/file/d/1OlPv6QBeo62VVTughj2jkiLeyHd1WnUt/view)
 
[![](https://assets.underline.io/uploads/markdown_image/1/image/a3db7a768409f05192210d98601edb25.png)](https://emnlp2022.rocket.chat/)

To access this site you need to register. Please register [here](https://2022.emnlp.org/registration/).

Register here

EMNLP 2022

Welcome!
EMNLP 2022 will take place in Abu Dhabi from December 7th to December 11th, 2022. And it will be held in hybrid mode, both online and offline.

Recent works have suggested that In-Context Learning (ICL) operates in dual modes, i.e. task retrieval (remember learned patterns from pre-training) and task learning (inference-time “learning" from demonstrations). However, disentangling these the two modes remains a challenging goal. We introduce ICL CIPHERS, a class of task reformulations based on substitution ciphers borrowed from classic cryptography. In this approach, a subset of tokens in the in-context inputs are substituted with other (irrelevant) tokens, rendering English sentences less comprehensible to human eye. However, by design, there is a latent, fixed pattern to this substitution, making it reversible. This bijective (reversible) cipher ensures that the task remains a well-defined task in some abstract sense, despite the transformations. It is a curious question if LLMs can solve ICL CIPHERS with a BIJECTIVE mapping, which requires deciphering the latent cipher. We show that LLMs are better at solving ICL CIPHERS with BIJECTIVE mappings than the NON-BIJECTIVE (irreversible) baseline, providing a novel approach to quantify “learning” in ICL. While this gap is small, it is consistent across the board on four datasets and six models. Finally, our interpretability analysis shows evidence that LLMs can internally decode ciphered inputs.

ICL CIPHERS: Quantifying ''Learning'' in In-Context Learning via Substitution Ciphers

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.

We explore a new language model inversion problem under strict black-box, zero-shot, and limited data conditions. We propose a novel training-free framework that reconstructs prompts using only a limited number of text outputs from a language model. Existing methods rely on the availability of a large number of outputs for both training and inference, an assumption that is unrealistic in the real world, and they can sometimes produce garbled text. In contrast, our approach, which relies on limited resources, consistently yields coherent and semantically meaningful prompts. Our framework leverages a large language model together with an optimization process inspired by the genetic algorithm to effectively recover prompts. Experimental results on several datasets derived from public sources indicate that our approach achieves high-quality prompt recovery and generates prompts more semantically and functionally aligned with the originals than current state-of-the-art methods. Additionally, use-case studies introduced demonstrate the method's strong potential for generating high-quality text data on perturbed prompts.

Reverse Prompt Engineering: A Zero-Shot, Genetic Algorithm Approach to Language Model Inversion

The ability of large language models (LLMs) to utilize external tools has enabled them to tackle an increasingly diverse range of tasks. However, as the tasks become more complex and long-horizon, the intricate tool utilization process may trigger various unexpected errors. Therefore, how to effectively handle such errors, including identifying, diagnosing, and recovering from them, has emerged as a key research direction for advancing tool learning. In this work, we first extensively analyze the types of errors encountered during the function-calling process on several competitive tool evaluation benchmarks. Based on it, we introduce CRITICTOOL, a comprehensive critique evaluation benchmark specialized for tool learning. Building upon a novel evolutionary strategy for dataset construction, CRITICTOOL holds diverse tool-use errors with varying complexities, which better reflects real-world scenarios. We conduct extensive experiments on CRITICTOOL, and validate the generalization and effectiveness of our constructed benchmark strategy. We also provide an in-depth analysis of the tool reflection ability on various LLMs, offering a new perspective on the field of tool learning in LLMs. Code will be available.

CRITICTOOL: Evaluating Self-Critique Capabilities of Large Language Models in Tool-Calling Error Scenarios

High-quality image captions are essential for improving modality alignment and visual understanding in Large Vision-Language Models (LVLMs). However, the scarcity of ultra-detailed image caption data limits further advancements. This paper presents a systematic pipeline for generating high-quality, ultra-detailed image captions, encompassing both pre-processing and post-processing stages. In the pre-processing stage, we classify and deduplicate images, extract visual information using expert tools, and leverage GPT-4o with structured prompts to generate initial captions. To enhance comprehensiveness, we introduce an expansion strategy based on Large Language Models (LLMs), defining eight descriptive dimensions to refine and extend captions, which serve as seed data for training a proprietary captioner model. In the post-processing stage, we incorporate human error-correction annotations and an active learning-inspired approach to refine low-quality samples. Using high-quality corrected data, we apply Direct Preference Optimization (DPO) and develop a critic-rewrite pipeline, training a sentence-level critic model to mitigate hallucinations. Experimental results demonstrate that our ultra-detailed captions significantly enhance LVLMs' perception and cognitive abilities across multiple vision-language benchmarks.

Enhancing Large Vision-Language Models with Ultra-Detailed Image Caption Generation

Personally identifiable information (PII) anonymization is a high-stakes task that poses a barrier to many open-science data sharing initiatives. While PII identification has made large strides in recent years, in practice, error thresholds and the recall/precision trade-off still limit the uptake of these anonymization pipelines. We present PIIvot, a lighter-weight framework for PII anonymization that leverages knowledge of the data context to simplify the PII detection problem. To demonstrate its effectiveness, we also contribute QATD_2k, the largest open-source real-world tutoring dataset of its kind, to support the demand for quality educational dialogue data.

PIIvot: A Lightweight NLP Anonymization Framework for Question-Anchored Tutoring Dialogues

Large language models (LLMs) are known to perpetuate stereotypes and exhibit biases. Various strategies have been proposed to mitigate these biases, but most work studies biases as a black-box problem without considering how concepts are represented within the model. We adapt techniques from representation engineering to study how the concept of "gender" is represented within LLMs. We introduce a new method that extracts concept representations via probability weighting without labeled data and efficiently selects a steering vector for measuring and manipulating the model's representation. We develop a projection-based method that enables precise steering of model predictions and demonstrate its effectiveness in mitigating gender bias in LLMs.

Unsupervised Concept Vector Extraction for Bias Control in LLMs

News articles often describe the same real-world event in strikingly different ways, shaping perception through framing rather than factual disagreement. However, traditional computational framing approaches often rely on coarse-grained topic classification, limiting their ability to capture subtle, event-level differences in how the same occurrences are presented across sources. We introduce Framing-divergent Event Coreference (FrECo), a novel task that identifies pairs of event mentions referring to the same underlying occurrence but differing in framing across documents to provide a event-centric lens for computational framing analysis. To support this task, we construct the high-agreement and diverse FrECo corpus. We evaluate the FrECo task on the corpus through supervised and preference-based tuning of large language models, providing strong baseline performance. To scale beyond the annotated data, we develop a bootstrapped mining pipeline that iteratively expands the training set with high-confidence FrECo pairs. Our approach enables scalable, interpretable analysis of how media frame the same events differently, offering a new lens for contrastive framing analysis at the event level.

Seeing the Same Story Differently: Framing‑Divergent Event Coreference for Computational Framing Analysis

In-context learning (ICL) enables large language models (LLMs) to perform new tasks using only a few demonstrations. However, in Named Entity Recognition (NER), existing ICL methods typically rely on task-agnostic semantic similarity for demonstration retrieval, which often yields less relevant examples and leads to inferior results. We introduce DEER, a training-free ICL approach that enables LLMs to make more informed entity predictions through the use of label-grounded statistics. DEER leverages token-level statistics from training labels to identify tokens most informative for entity recognition, enabling entity-focused demonstrations. It further uses these statistics to detect and refine error-prone tokens through a targeted reflection step. Evaluated on five NER datasets across four LLMs, DEER consistently outperforms existing ICL methods and achieves performance comparable to supervised fine-tuning. Further analyses demonstrate that DEER improves example retrieval, remains effective on both seen and unseen entities, and exhibits strong robustness in low-resource settings.

LLMs are Better Than You Think: Label-Guided In-Context Learning for Named Entity Recognition

Multimodal Large Language Models (MLLMs) have demonstrated remarkable performance in complex multimodal tasks. While MLLMs excel at visual perception and reasoning in third-person and egocentric videos, they are prone to hallucinations, generating coherent yet inaccurate responses. We present EGOILLUSION, a first benchmark to evaluate MLLM hallucinations in egocentric videos. EGOILLUSION comprises 1,400 videos paired with 8,000 human-annotated open and closed-ended questions designed to trigger hallucinations in both visual and auditory cues in egocentric videos. Evaluations across ten MLLMs reveal significant challenges, including powerful models like GPT-4o and Gemini, achieving only 59% accuracy. EGOILLUSION lays the foundation in developing robust benchmarks to evaluate the effectiveness of MLLMs and spurs the development of better egocentric MLLMs with reduced hallucination rates. Our benchmark will be open-sourced for reproducibility

EGOILLUSION: Benchmarking Hallucinations in Egocentric Video Understanding

The rapid progress of Large Language Models (LLMs) has empowered omni models to act as voice assistants capable of understanding spoken dialogues. These models can process multimodal inputs beyond text, such as speech and visual data, enabling more context-aware interactions. However, current benchmarks fall short in comprehensively evaluating how well these models generate context-aware responses, particularly when it comes to implicitly understanding fine-grained speech characteristics, such as pitch, emotion, timbre, and volume or the environmental acoustic context such as background sounds. Additionally, they inadequately assess the ability of models to align paralinguistic cues with complementary visual signals to inform their responses. To address these gaps, we introduce MultiVox, the first omni voice assistant benchmark designed to evaluate the ability of voice assistants to integrate spoken and visual cues including paralinguistic speech features for truly multimodal understanding. Specifically, MultiVox includes 1000 human-annotated and recorded speech dialogues that encompass diverse paralinguistic features and a range of visual cues such as images and videos. Our evaluation on 9 state-of-the-art models reveals that, although humans excel at these tasks, current models consistently struggle to produce contextually grounded responses. Our benchmark will be open-sourced.

MULTIVOX: A Benchmark for Evaluating Voice Assistants for Multimodal Interactions

With the growing adoption of machine learning models in critical domains, techniques for explaining differences between models have become essential for trust, debugging, and informed deployment. Previous approaches address this by identifying input transformations that cause divergent predictions or by learning joint surrogate models to align and contrast behaviors. These methods often require access to training data and do not produce natural language explanations. In this paper, we introduce SLED, a framework that generates faithful natural language explanations of when and how two ML models converge or diverge in their predictions. SLED first uses gradient-based optimization to synthesize input samples that highlight divergence and convergence patterns, and then leverages a large language model (LLM) to generate explanations grounded in these synthetic samples. Across both text-based (3 tasks, 7 models) and structured (10 tasks, 4 models) classification tasks, we show that SLED explanations are 18--24% more faithful than the strongest baselines. User studies also indicate that SLED explanations achieve a real-world simulatability of 63.5%. Importantly, SLED requires minimal access to training data and generalizes well to real-world samples, enabling transparent and data-efficient model comparison.

Downloads

Next from EMNLP 2025

Reverse Prompt Engineering: A Zero-Shot, Genetic Algorithm Approach to Language Model Inversion

Similar lecture

Segmenting Numerical Substitution Ciphers