China

Long-range tasks demand reasoning over long inputs. However, existing solutions are limited, e.g., long-context models require large compute budgets, parameter-efficient fine-tuning (PEFT) needs training data, and retrieval-augmented generation (RAG) entails complex task-specific designs. Though in-context approaches overcome many of these issues, methods with short-context LLMs are inefficient, trading context for processing more tokens. We introduce **PRISM**, a highly token-efficient in-context method based on structured schemas that outperforms baselines on diverse tasks with **4x shorter contexts**. This approach produces concise outputs and efficiently leverages key-value (KV) caches to **reduce costs by up to 54%**. PRISM scales down to tiny contexts without increasing costs or sacrificing quality, and generalizes to new tasks with minimal effort by generating schemas from task descriptions.

EMNLP 2025

PRISM: Efficient Long-Range Reasoning With Short-Context LLMs

nlp in resource-constrained settings

prompting

## 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*





<br>

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

poster

Miscalibration in Large Language Models (LLMs) undermines their reliability, highlighting the need for accurate confidence estimation. We introduce CCPS (Calibrating LLM Confidence by Probing Perturbed Representation Stability), a novel method analyzing internal representational stability in LLMs. CCPS applies targeted adversarial perturbations to final hidden states, extracts features reflecting the model's response to these perturbations, and uses a lightweight classifier to predict answer correctness. CCPS was evaluated on LLMs from 8B to 32B parameters (covering Llama, Qwen, and Mistral architectures) using MMLU and MMLU-Pro benchmarks in both multiple-choice and open-ended formats. Our results show that CCPS significantly outperforms current approaches. Across four LLMs and three MMLU variants, CCPS reduces Expected Calibration Error by approximately 55% and Brier score by 21%, while increasing accuracy by 5 percentage points, Area Under the Precision-Recall Curve by 4 percentage points, and Area Under the Receiver Operating Characteristic Curve by 6 percentage points, all relative to the strongest prior method. CCPS delivers an efficient, broadly applicable, and more accurate solution for estimating LLM confidence, thereby improving their trustworthiness.

Calibrating LLM Confidence by Probing Perturbed Representation Stability

Unlearning evaluation has traditionally followed the retrieval paradigm, where adversaries attempt to extract residual knowledge of an unlearning target by issuing queries to a language model. However, the absence of retrievable knowledge does not necessarily prevent an adversary from inferring which targets have been intentionally unlearned in the post-training optimization. Such inferences can still pose significant privacy risks, as they may reveal the sensitive data in the model's training set and the internal policies of model creators. To quantify such privacy risks, we propose a new evaluation framework **Forensic Unlearning Membership Attacks (FUMA)**, drawing on principles from membership inference attacks. FUMA assesses whether unlearning leaves behind detectable artifacts that can be exploited to infer membership in the forget set. Specifically, we evaluate four major optimization-based unlearning methods on 258 models across diverse unlearned settings and show that examples in the forget set can be identified up to 99% accuracy. This highlights privacy risks not covered in existing retrieval-based benchmarks. We conclude by discussing recommendations to mitigate these vulnerabilities.

Identifying Unlearned Data in LLMs via Membership Inference Attacks

As the deployment of AI models shifts towards edge devices, developing efficient sequence models has become critical. State-space models (SSMs), particularly Mamba, have emerged as strong rivals to Transformers due to their linear-time complexity and impressive performance across a range of tasks. However, their large parameter counts still hinder their use in resource-constrained environments. To address this, we propose a novel unstructured pruning framework specifically tailored for Mamba, achieving up to 70% parameter reduction with only a 3–9% drop in performance. Unlike pruning techniques designed for Transformers, our approach leverages Mamba's unique recurrent dynamics by incorporating pruning based on both weight and gradient importance to preserve critical parameters, a gradual pruning schedule to maintain model stability, and a global strategy to optimize parameter allocation across the model. Extensive experiments on the WikiText-103, Long Range Arena, and ETT benchmarks demonstrate significant efficiency gains, including 1.77× faster inference and a 46% reduction in memory usage. Our component analysis confirms Mamba's robustness to pruning, highlighting the framework's potential for enabling practical deployment while underscoring the need for careful evaluation to avoid introducing biases in sensitive applications.

Efficient Unstructured Pruning of Mamba State-Space Models for Resource-Constrained Environments

Unlearning has emerged as a critical capability for large language models (LLMs) to support data privacy, regulatory compliance, and ethical AI deployment. Recent techniques often rely on obfuscation by injecting incorrect or irrelevant information to suppress knowledge. Such methods effectively constitute knowledge addition rather than true removal, often leaving models vulnerable to probing. In this paper, we formally distinguish unlearning from obfuscation and introduce a probing-based evaluation framework to assess whether existing approaches genuinely remove targeted information. Moreover, we propose DF-MCQ, a novel unlearning method that flattens the model predictive distribution over automatically generated multiple-choice questions using KL-divergence, effectively removing knowledge about target individuals and triggering appropriate refusal behaviour. Experimental results demonstrate that DF-MCQ achieves unlearning with over 90% refusal rate and a random choice-level uncertainty that is much higher than obfuscation on probing questions.

Unlearning vs. Obfuscation: Are We Truly Removing Knowledge?

Instruct models, obtained from various instruction tuning or post-training steps, are commonly deemed superior and more usable than their base counterpart. While the model gains instruction following ability, instruction tun- ing may lead to forgetting the knowledge from pre-training or it may encourage the model being overly conversational or verbose. This, in turn, can lead to degradation of in-context few-shot learning performance. In this work, we study the performance trajectory between base and instruct models by scaling down the strength of instruction-tuning via the partial adaption method. We show that, across several model families and model sizes, reducing the strength of instruction-tuning results in material improvement on a few-shot in-context learning benchmark covering a variety of classic natural language tasks. This comes at the cost of losing some degree of instruction following ability as measured by AlpacaEval. Our study shines light on the potential trade-off between in-context learning and instruction following abilities that is worth considering in practice.

Improving Instruct Models for Free: A Study on Partial Adaptation

Large Language Models (LLMs) are rapidly surpassing human knowledge in many domains. While improving these models traditionally relies on costly human data, recent self-rewarding mechanisms have shown that LLMs can improve by judging their own responses instead of relying on human labelers. However, existing methods have primarily focused on improving model responses rather than judgment capabilities, resulting in rapid saturation during iterative training. To address this issue, we introduce a novel Meta-Rewarding step to the self-improvement process, where the model judges its own judgements and uses that feedback to refine its judgment skills. Surprisingly, this unsupervised approach improves the model's ability to judge and follow instructions, as demonstrated by a win rate improvement of Llama-3-8B-Instruct from 22.9% to 39.4% on AlpacaEval 2, and 20.6% to 29.1% on Arena-Hard. These results strongly suggest the potential for self-improving models without human supervision.

Meta-Rewarding Language Models: Self-Improving Alignment with LLM-as-a-Meta-Judge

Quantifying identity fusion---the psychological merging of self with another entity or abstract target (e.g., a religious group, political party, ideology, value, brand, belief, etc.)---is vital for understanding a wide range of group‑based human behaviors. We introduce the Cognitive Linguistic Identity Fusion Score (CLIFS), a novel metric that integrates cognitive linguistics with large language models (LLMs), which builds on implicit metaphor detection. Unlike traditional pictorial and verbal scales, which require controlled surveys or direct field contact, CLIFS delivers fully automated, scalable assessments while maintaining strong alignment with the established verbal measure. In benchmarks, CLIFS outperforms both existing automated approaches and human annotation. As a proof of concept, we apply CLIFS to violence risk assessment to demonstrate that it can improve violence risk assessment by more than 240%. Building on our identification of a new NLP task and early success, we underscore the need to develop larger, more diverse datasets encompassing additional fusion-target domains and cultural backgrounds to enhance generalizability and further advance this emerging area.

Cognitive Linguistic Identity Fusion Score (CLIFS): A Scalable Cognition‑Informed Approach to Quantifying Identity Fusion from Text

Visual Language Models have demonstrated remarkable capabilities across various tasks, including visual question answering and image captioning. However, most models rely on text-based instructions, limiting their effectiveness in natural human-machine interactions. Moreover, the quality of language models primarily depends on reasoning and prompting techniques, such as chain-of-thought, which remain underexplored when using speech instructions. To address these challenges, we propose SilVar, an end-to-end multimodal model that leverages speech instructions for reasoning-based visual question answering. Additionally, we investigate reasoning techniques at different levels, including conversational, simple, and complex speech instructions. SilVar is built upon CLIP, Whisper, and LLaMA 3.1-8B, enabling more intuitive interactions by allowing users to provide verbal or text-based instructions. To this end, we introduce a new dataset designed to challenge models with speech-based reasoning tasks for object localization. This dataset enhances the model’s ability to process and explain visual scenes from spoken input, moving beyond simple object recognition to reasoning-based interactions. To the best of our knowledge, SilVar is the first open-source, speech-driven VLM. Despite the challenges posed by speech-based instructions, experiments show that our speech-driven multimodal model performs on par with text-based models of similar size on the MMMU and ScienceQA benchmarks, demonstrating its potential in scenarios where text input is unavailable or not preferred, such as self-driving cars and medical surgery. We believe SilVar will inspire the next generation of multimodal reasoning models, advancing toward expert artificial general intelligence.

SilVar: Speech-Driven Multimodal Model for Reasoning Visual Question Answering and Object Localization

In response to adversarial text attacks, attack detection models have been proposed and shown to successfully identify text modified by adversaries. Attack detection models can be leveraged to provide an additional check for NLP models and give signals for human input. However, the reliability of these models has not yet been thoroughly explored. Thus, we propose and test an attack, \RedHerring. RedHerring aims to make attack detection models unreliable by modifying a text to cause the detection model to predict an attack, while keeping the classifier correct. This creates a tension between the classifier and detector. If a human sees that the detector is giving an ``incorrect'' prediction, but the classifier a correct one, then the human will see the detector as unreliable. We test this novel threat model on 4 datasets against 3 detectors defending 4 classifiers. We find that RedHerring is able to drop detection accuracy between 20 - 71 points, while maintaining (or improving) classifier accuracy. As an initial defense, we propose a simple confidence check which requires no retraining of the classifier or detector and increases detection accuracy greatly. This novel threat model offers new insights into how adversaries may target detection models.

RedHerring Attack: Testing the Reliability of Attack Detection

Language models (LMs), like other neural networks, often favor shortcut heuristics based on surface-level patterns. LMs in particular behave like n-gram models early in training, but eventually learn hierarchical syntax to correctly apply grammatical rules out-of-distribution (OOD). This paper uses case studies of English grammar to explore how complex, diverse training data drives OOD generalization. We construct a framework that unifies our understanding of random variation with training dynamics, rule selection with memorization, and data diversity with complexity. We show that these factors are nuanced, and that intermediate levels of diversity and complexity lead to inconsistent behavior across random seeds and to unstable training dynamics. Our findings emphasize the critical role of training data in shaping generalization patterns and illuminate how competing model strategies lead to inconsistent training outcomes.

Downloads

Next from EMNLP 2025

Calibrating LLM Confidence by Probing Perturbed Representation Stability

Similar lecture

Long-context Language Models Fail in Basic Retrieval Tasks Without Sufficient Reasoning Steps