China

Autoregressive models dominate text generation but suffer from left-to-right decoding constraints that limit efficiency and bidirectional reasoning. Diffusion-based models offer a flexible alternative but face challenges in adapting to discrete text efficiently. We propose LAD (LoRA-Adapted Diffusion), a framework for non-autoregressive generation that adapts LLaMA models for iterative, bidirectional sequence refinement using LoRA adapters. LAD employs a structural denoising objective combining masking with text perturbations (swaps, duplications and span shifts), enabling full sequence editing during generation. We aim to demonstrate that LAD could be a viable and efficient alternative to training diffusion models from scratch, by providing both validation results as well as two interactive demos directly available online: https://ruurdkuiper.github.io/tini-lad/ https://huggingface.co/spaces/Ruurd/tini-lad Inference and training code: https://github.com/RuurdKuiper/lad-code

EMNLP 2025

LAD: LoRA-Adapted Diffusion

demo

## 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) generate human-aligned content under certain safety constraints. However, the current known technique ``jailbreak prompt'' can circumvent safety-aligned measures and induce LLMs to output malicious content. Research on Jailbreaking can help identify vulnerabilities in LLMs and guide the development of robust security frameworks. To circumvent the issue of attack templates becoming obsolete as models evolve, existing methods adopt iterative mutation and dynamic optimization to facilitate more automated jailbreak attacks. However, these methods face two challenges: inefficiency and repetitive optimization, as they overlook the value of past attack experiences. To better integrate past attack experiences to assist current jailbreak attempts, we propose the textbfJailExpert, an automated jailbreak framework, which is the first to achieve a formal representation of experience structure, group experiences based on semantic drift, and support the dynamic updating of the experience pool. Extensive experiments demonstrate that JailExpert significantly improves both attack effectiveness and efficiency. Compared to the current state-of-the-art black-box jailbreak method, JailExpert achieves an average increase of 24% in attack success rate and 2.7 times improvement in attack efficiency.

Stand on The Shoulders of Giants: Building JailExpert from Previous Attack Experience

Building language models (LMs), especially small and medium ones, remains more art than science. While large LMs often improve by sheer scale, it is still unclear why many design choices work. For small LMs, this uncertainty is more limiting: tight parameter budgets make each decision critical, yet researchers still lack systematic, scientific ways to test and refine new ideas. We introduce Pico, a lightweight, modular framework that enables systematic, hypothesis-driven research for small and medium-scale language model development. Pico consists of two libraries that together provide a practical sandbox where researchers can make targeted changes to a model's architecture or training procedures and directly observe their effects on the model's behavior. To support reproducible experimentation, we also release a suite of baseline models, pico-decoder, trained under standardized conditions and open-sourced for the community. Case Studies highlight how Pico can support iterative small LM design and analysis.

Pico: A Modular Framework for Hypothesis-Driven Small Language Model Research

We present BAREC Demo, a web-based system for fine-grained, sentence-level Arabic readability assessment. The demo is part of the Balanced Arabic Readability Evaluation Corpus (BAREC) project, which manually annotated 69,000 sentences (over one million words) from diverse genres and domains using a 19-level readability scale inspired by the Taha/Arabi21 framework, covering reading abilities from kindergarten to postgraduate levels. The project also developed models for automatic readability assessment. The demo provides two main functionalities for educators, content creators, language learners, and researchers: (1) a Search interface to explore the annotated dataset for text selection and resource development, and (2) an Analyze interface, which uses trained models to assign detailed readability labels to Arabic texts at the sentence level. The system and all of its resources are accessible at https://barec.camel-lab.com.

BAREC Demo: Resources and Tools for Sentence-level Arabic Readability Assessment

Information extraction (IE) is fundamental to numerous NLP applications, yet existing solutions often require specialized models for different tasks or rely on computationally expensive large language models. We present GLiNER2, a unified framework that enhances the original GLiNER architecture to support named entity recognition, text classification, and hierarchical structured data extraction within a single efficient model. Built on a fine-tuned encoder architecture, GLiNER2 maintains CPU efficiency and compact size while introducing multi-task composition through an intuitive schema-based interface. Our experiments demonstrate competitive performance across diverse IE tasks with substantial improvements in deployment accessibility compared to LLM-based alternatives. We release GLiNER2 as an open-source library available through pip, complete with pre-trained models and comprehensive documentation.

GLiNER2: Schema-Driven Multi-Task Learning for Structured Information Extraction

Multi-agent debate (MAD) has demonstrated the ability to augment collective intelligence by scaling test-time compute and leveraging expertise. Current frameworks for multi-agent debate are often tailored towards tool use, lack integrated evaluation, or provide restricted configurability of agent personas, response generators, discussion paradigms, and decision protocols. We introduce MALLM (Multi-Agent Large Language Models), an open-source framework that enables systematic analysis of MAD components. MALLM offers more than 144 unique configurations of MAD, including (1) agent personas (e.g., Expert, Personality), (2) response generators (e.g., Critical, Reasoning), (3) discussion paradigms (e.g., Memory, Relay), and (4) decision protocols (e.g., Voting, Consensus). MALLM uses simple configuration files to define a debate. Furthermore, MALLM can load any textual Huggingface dataset (e.g., MMLU-Pro, WinoGrande) and provides an evaluation pipeline for easy comparison of MAD configurations. MALLM is tailored towards researchers and provides a window into the heart of multi-agent debate, facilitating the understanding of its components and their interplay.

MALLM: Multi-Agent Large Language Models Framework

Evaluating LLMs with a single prompt has proven unreliable, with small changes leading to significant performance differences. However, generating the prompt variations needed for a more robust multi-prompt evaluation is challenging, limiting its adoption in practice. To address this, we introduce PromptSuite, a framework that enables the automatic generation of various prompts. PromptSuite is flexible – working out of the box on a wide range of tasks and benchmarks. It follows a modular prompt design, allowing controlled perturbations to each component, and is extensible, supporting the addition of new components and perturbation types. Through a series of case studies, we show that PromptSuite provides meaningful variations to support strong evaluation practices. It is available through both a Python API, and a user-friendly web interface.

PromptSuite: A Task-Agnostic Framework for Multi-Prompt Generation

The proliferation of transformer-based language models has revolutionized NLP domain while simultaneously introduced significant challenges regarding model transparency and trustworthiness. The complexity of achieving explainable systems in this domain is evidenced by the extensive array of explanation methods and evaluation metrics developed by researchers. To address the challenge of selecting optimal explainability approaches, we present o-mega, a hyperparameter optimization tool designed to automatically identify the most effective explainable AI methods and their configurations within the semantic matching domain. We evaluate o-mega on a post-claim matching pipeline using a curated dataset of social media posts paired with refuting claims. Our tool systematically explores different explainable methods and their hyperparameters, demonstrating improved transparency in automated fact-checking systems. As a result, such automated optimization of explanation methods can significantly enhance the interpretability of claim-matching models in critical applications such as misinformation detection, contributing to more trustworthy and transparent AI systems.

o-MEGA: Optimized Methods for Explanation Generation and Analysis

Traditional neural network training typically follows fixed, predefined optimization recipes, lacking the flexibility to dynamically respond to instabilities or emerging training issues. In this paper, we introduce Interactive Training, an open-source framework that enables real-time, feedback-driven intervention during neural network training by human experts or automated AI agents. At its core, Interactive Training uses a control server to mediate communication between users or agents and the ongoing training process, allowing users to dynamically adjust optimizer hyperparameters, training data, and model checkpoints. Through three case studies, we demonstrate that Interactive Training achieves superior training stability, reduced sensitivity to initial hyperparameters, and improved adaptability to evolving user needs, paving the way toward a future training paradigm where AI agents autonomously monitor training logs, proactively resolves instabilities, and optimizes training dynamics.

Interactive Training: Feedback-Driven Neural Network Optimization

Large Language Model (LLM) agents hold promise for a flexible and scalable alternative to traditional business process automation, but struggle to reliably follow complex company policies. In this study we introduce a deterministic, transparent, and modular framework for enforcing business policy adherence in agentic workflows. Our method operates in two phases: (1) an offline buildtime stage that compiles policy documents into verifiable guard code associated with tool use, and (2) a runtime integration where these guards ensure compliance before each agent action. We demonstrate our approach on the challenging τ-bench Airlines domain, showing encouraging preliminary results in policy enforcement, and further outline key challenges for real-world deployments.

Towards Enforcing Company Policy Adherence in Agentic Workflows

Multi-agent systems built on language models have shown strong performance on collaborative reasoning tasks. However, existing evaluations focus only on the correctness of the final output, overlooking how inefficient communication and poor coordination contribute to redundant reasoning and higher computational costs. We introduce **GEMMAS**, a graph-based evaluation framework that analyzes the internal collaboration process by modeling agent interactions as a directed acyclic graph. To capture collaboration quality, we propose two process-level metrics: Information Diversity Score (IDS) to measure semantic variation in inter-agent messages, and Unnecessary Path Ratio (UPR) to quantify redundant reasoning paths. We evaluate GEMMAS across five benchmarks and highlight results on GSM8K, where systems with only a 2.1% difference in accuracy differ by 12.8% in IDS and 80% in UPR, revealing substantial variation in internal collaboration. These findings demonstrate that outcome-only metrics are insufficient for evaluating multi-agent performance and highlight the importance of process-level diagnostics in designing more interpretable and resource-efficient collaborative AI systems.

Downloads

Next from EMNLP 2025

Stand on The Shoulders of Giants: Building JailExpert from Previous Attack Experience

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES