China

Large Language Models (LLMs) have been explored for automating or enhancing penetration testing tasks, but their effectiveness and reliability across diverse attack phases remain open questions. This study presents a comprehensive evaluation of multiple LLM-based agents—from singular to modular—across realistic penetration testing scenarios, analyzing their empirical performance and recurring failure patterns. We further investigate the impact of core functional capabilities on agent success, operationalized through five targeted augmentations: Global Context Memory (GCM), Inter-Agent Messaging (IAM), Context-Conditioned Invocation (CCI), Adaptive Planning (AP), and Real-Time Monitoring (RTM). These interventions respectively support the capabilities of Context Coherence &amp; Retention, Inter-Component Coordination &amp; State Management, Tool Usage Accuracy &amp; Selective Execution, Multi-Step Strategic Planning &amp; Error Detection &amp; Recovery, and Real-Time Dynamic Responsiveness. Our findings reveal that while some architectures natively exhibit select properties, targeted augmentations significantly enhance modular agent performance—particularly in complex, multi-step, and real-time penetration testing scenarios.

EMNLP 2025

From Capabilities to Performance: Evaluating Key Functional Properties of LLM Architectures in Penetration Testing

Large Language Models (LLMs) have been explored for automating or enhancing penetration testing tasks, but their effectiveness and reliability across diverse attack phases remain open questions. This study presents a comprehensive evaluation of multiple LLM-based agents—from singular to modular—across realistic penetration testing scenarios, analyzing their empirical performance and recurring failure patterns. We further investigate the impact of core functional capabilities on agent success, operationalized through five targeted augmentations: Global Context Memory (GCM), Inter-Agent Messaging (IAM), Context-Conditioned Invocation (CCI), Adaptive Planning (AP), and Real-Time Monitoring (RTM). These interventions respectively support the capabilities of Context Coherence & Retention, Inter-Component Coordination & State Management, Tool Usage Accuracy & Selective Execution, Multi-Step Strategic Planning & Error Detection & Recovery, and Real-Time Dynamic Responsiveness. Our findings reveal that while some architectures natively exhibit select properties, targeted augmentations significantly enhance modular agent performance—particularly in complex, multi-step, and real-time penetration testing scenarios.

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) increasingly power mental-health chatbots, yet the field still lacks a scalable, theory-grounded way to decide which model is more effective to deploy. We present ESC-Judge, the first end-to-end evaluation framework that (i) grounds head-to-head comparison of Emotional-Support LLMs (ES-LLMs) in an established psychological theory—Clara Hill’s Exploration–Insight–Action (E-I-A) counselling model—thereby delivering a structured, interpretable lens on performance, and (ii) fully automates the pipeline at scale. ESC-Judge proceeds in three stages: (1) it synthesizes realistic help-seeker roles by sampling empirically salient attributes (stressors, personality, life history); (2) it has two candidate ES-Agents conduct separate sessions with the same role, isolating model-specific strategies; and (3) it asks a specialised judge LLM to issue pairwise preferences across rubric-anchored skills that exhaustively cover the E-I-A spectrum. In our empirical study, ESC-Judge matches PhD-level annotators in 85% of Exploration, 83% of Insight, and 86% of Action decisions, demonstrating human-level reliability at a fraction of the cost. We release all code, prompts, synthetic roles, transcripts, and judgment scripts to catalyze transparent progress in emotionally supportive AI

ESC-Judge: A Framework for Comparing Emotional Support Conversational Agents

Considering deep neural networks as manifold mappers, the pretrain-then-fine-tune paradigm can be interpreted as a two-stage process: pretrain establishes a broad knowledge base, and fine-tune adjusts the model parameters to activate specific neural pathways to align with the target manifold. Although prior fine-tuning approaches demonstrate success, their rigid parameter space limits their ability to dynamically activate appropriate neural pathways, rendering them ill-equipped to adapt flexibly to the diverse and evolving data distributions. In light of this view, we propose a novel approach, Mixture of Expert Prompt Tuning (MEPT), as an effective and efficient manifold-mapping framework. MEPT leverages the Mixture of Experts architecture by integrating multiple prompt experts to adaptively learn diverse and non-stationary data distributions. Empirical evaluations demonstrate that MEPT outperforms several state-of-the-art parameter efficient baselines on SuperGLUE, achieving notable improvements in mean accuracy (e.g., 1.94%) while significantly reducing activated prompts by 79.25%. The effectiveness of MEPT is further supported by theoretical insights from manifold learning and validated through neural activation pathway visualization results.

MEPT: Mixture of Expert Prompt Tuning as a Manifold Mapper

Auto-evaluating language models (LMs), *i.e*., using a grader LM to evaluate the candidate LM, is an appealing way to accelerate the evaluation process and the cost associated with it. But this presents a paradox: how can we trust the grader LM, which is presumably weaker than the candidate LM, to assess problems that are beyond the frontier of the capabilities of either model or both? For instance, today's LMs struggle on graduate-level physics and Olympiad-level math, making them unreliable graders in these domains. We show that providing *privileged information* -- such as ground-truth solutions or problem-specific guidelines -- improves automated evaluations on such frontier problems. This approach offers two key advantages. First, it expands the range of problems where LMs graders apply. Specifically, weaker models can now rate the predictions of stronger models. Second, privileged information can be used to devise easier variations of challenging problems which improves the separability of different LMs on tasks where their performance is generally low. With this approach, general-purpose LM graders match the state of the art performance on *RewardBench*, surpassing almost all the specially-tuned models. LM graders also outperform individual human raters on *Vibe-Eval*, and approach human expert graders on Olympiad-level math problems.

Graders Should Cheat: Privileged Information Enables Expert-Level Automated Evaluations

This survey explores the burgeoning field of representation convergence in foundation models, a phenomenon where diverse neural networks, despite differences in architecture, training objectives, and data modalities, increasingly learn aligned representations. We delve into the Platonic Representation Hypothesis, which posits that this convergence reflects a move towards a shared statistical model of reality. Our survey examines empirical evidence for this convergence across single modalities, including vision, language, speech, and graph data, as well as across modalities such as vision-language, speech-language, and graph-language. We further discuss the potential factors driving this convergence, such as scale, task generality, and simplicity bias. Finally, we explore the significant implications of representation convergence for multimodal learning, unified AI systems, and our understanding of intelligence, while also considering the limitations and open questions that remain in this evolving area of research.

Representation Potentials of Foundation Models for Multimodal Alignment: A Survey

Large language models (LLMs) can underpin AI assistants that help users with everyday tasks like making recommendations or performing basic computation. Despite AI assistants' promise, little is known about implicit values they display while completing everyday tasks. Humans may consider values like environmentalism, charity, and diversity. To what extent do LLMs exhibit these values in completing everyday tasks? How do they compare with humans? We answer these questions by auditing how six popular LLMs complete 30 everyday tasks, comparing LLMs to each other and to 100 human crowdworkers from the US. We find LLMs often do not align with humans, nor with other LLMs, in the implicit values exhibited.

Implicit Values Embedded in How Humans and LLMs Complete Subjective Everyday Tasks

Large Language Models (LLMs) increasingly serve diverse global audiences, making it critical for responsible AI deployment across cultures. While recent works have proposed various approaches to enhance cultural alignment in LLMs, a systematic analysis of their evaluation benchmarks remains needed. We propose a novel framework that conceptualizes alignment along three dimensions: Cultural Group (who to align with), Cultural Elements (what to align), and Awareness Scope (how to align). Through this framework, we analyze 105 cultural alignment evaluation benchmarks, revealing significant imbalances: Region (37.9%) and Language (28.9%) dominate Cultural Group representation; Social and Political Relations (25.1%) and Speech and Language (20.9%) concentrate Cultural Elements coverage; and an overwhelming majority (97.1%) of datasets adopt majority-focused Awareness Scope approaches. In a case study examining AI safety evaluation across nine Asian countries (Section 5), we demonstrate how our framework reveals critical gaps between existing benchmarks and real-world cultural biases identified in the study, providing actionable guidance for developing more comprehensive evaluation resources tailored to specific deployment contexts.

Incorporating Diverse Perspectives in Cultural Alignment: Survey of Evaluation Benchmarks Through A Three-Dimensional Framework

Graphical User Interface (GUI) agents are expected to precisely operate on the screens of digital devices. Existing GUI agents merely depend on current visual observations and plain-text action history, ignoring the significance of history screens. To mitigate this issue, we propose UI-Hawk, a multi-modal GUI agent specially designed to process screen streams encountered during GUI navigation. UI-Hawk incorporates a history-aware visual encoder and an efficient resampler to handle the screen sequences. To acquire a better understanding of screen streams, we define four fundamental tasks—UI grounding, UI referring, screen question answering, and screen summarization. We develop an automated data curation method to generate the corresponding training data for UI-Hawk. Along with the efforts above, we have also created a benchmark FunUI to quantitatively evaluate the fundamental screen understanding ability of MLLMs. Extensive experiments on FunUI and GUI navigation benchmarks consistently validate that screen stream understanding is essential for GUI tasks.

UI-Hawk: Unleashing the Screen Stream Understanding for Mobile GUI Agents

Relation extraction (RE) aims to identify semantic relations between entities in unstructured text. Although recent work extends traditional RE to multimodal scenarios, most approaches still adopt classification-based paradigms with fused multimodal features, representing relations as discrete labels. This paradigm has two significant limitations: (1) it overlooks structural constraints like entity types and positional cues, and (2) it lacks semantic expressiveness for fine-grained relation understanding. We propose **R**etrieval **O**ver **C**lassification (ROC), a novel framework that reformulates multimodal RE as a retrieval task driven by relation semantics. ROC integrates entity type and positional information through a multimodal encoder, expands relation labels into natural language descriptions using a large language model, and aligns entity-relation pairs via semantic similarity-based contrastive learning. Experiments show that our method achieves state-of-the-art performance on the benchmark datasets MNRE and MORE and exhibits stronger robustness and interpretability.

Retrieval over Classification: Integrating Relation Semantics for Multimodal Relation Extraction

In real practice, questions are typically complex and knowledge-intensive, requiring Large Language Models (LLMs) to recognize the multifaceted nature of the question and reason across multiple information sources. Iterative and adaptive retrieval, where LLMs decide when and what to retrieve based on their reasoning, has been shown to be a promising approach to resolve complex, knowledge-intensive questions. However, the performance of such retrieval frameworks is limited by the accumulation of reasoning errors and misaligned retrieval results. To overcome these limitations, we propose TreeRare (Syntax Tree-Guided Retrieval and Reasoning, a framework that utilizes syntax trees to guide information retrieval and reasoning for question answering. Following the principle of compositionality, TreeRare traverses the syntax tree in a bottom-up fashion, and in each node, it generates subcomponent-based queries and retrieves relevant passages to resolve localized uncertainty. A subcomponent question answering module then synthesizes these passages into concise, context-aware evidence. Finally, TreeRare aggregates the evidence across the tree to form a final answer. Experiments across five question answering datasets involving ambiguous or multi-hop reasoning demonstrate that TreeRare achieves substantial improvements over existing state-of-the-art methods.

TreeRare: Syntax Tree-Guided Retrieval and Reasoning for Knowledge-Intensive Question Answering

A lack of demographic context in existing toxic speech datasets limits our understanding of how different age groups communicate online. In collaboration with funk, a German public service content network, this research introduces the first large-scale German dataset annotated for toxicity and enriched with platform-provided age estimates. The dataset includes 3,024 human-annotated and 30,024 LLM-annotated anonymized comments from Instagram, TikTok, and YouTube. To ensure relevance, comments were consolidated using predefined toxic keywords, resulting in 16.7\% labeled as problematic. The annotation pipeline combined human expertise with state-of-the-art language models, identifying key categories such as insults, disinformation, and criticism of broadcasting fees. The dataset reveals age-based differences in toxic speech patterns, with younger users favoring expressive language and older users more often engaging in disinformation and devaluation. This resource provides new opportunities for studying linguistic variation across demographics and supports the development of more equitable and age-aware content moderation systems.

Downloads

Next from EMNLP 2025

ESC-Judge: A Framework for Comparing Emotional Support Conversational Agents

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES