China

While Vision-Language Models (VLMs) have achieved competitive performance in various tasks, their comprehension of the underlying structure and semantics of a scene remains understudied. To investigate the understanding of VLMs, we study their capability regarding object properties and relations in a controlled and interpretable manner. To this scope, we introduce CIVET, a novel and extensible framework for systemati**C** evaluat**I**on **V**ia controll**E**d s**T**imuli. CIVET addresses the lack of standardized systematic evaluation for assessing VLMs&#39; understanding, enabling researchers to test hypotheses with statistical rigor. With CIVET, we evaluate five state-of-the-art VLMs on exhaustive sets of stimuli, free from annotation noise, dataset-specific biases, and uncontrolled scene complexity. Our findings reveal that 1) current VLMs can accurately recognize only a limited set of basic object properties; 2) their performance heavily depends on the position of the object in the scene; 3) they struggle to understand basic relations among objects. Furthermore, a comparative evaluation with human annotators reveals that VLMs still fall short of achieving human-level accuracy.

EMNLP 2025

CIVET: Systematic Evaluation of Understanding in VLMs

While Vision-Language Models (VLMs) have achieved competitive performance in various tasks, their comprehension of the underlying structure and semantics of a scene remains understudied. To investigate the understanding of VLMs, we study their capability regarding object properties and relations in a controlled and interpretable manner. To this scope, we introduce CIVET, a novel and extensible framework for systemati**C** evaluat**I**on **V**ia controll**E**d s**T**imuli. CIVET addresses the lack of standardized systematic evaluation for assessing VLMs' understanding, enabling researchers to test hypotheses with statistical rigor. With CIVET, we evaluate five state-of-the-art VLMs on exhaustive sets of stimuli, free from annotation noise, dataset-specific biases, and uncontrolled scene complexity. Our findings reveal that 1) current VLMs can accurately recognize only a limited set of basic object properties; 2) their performance heavily depends on the position of the object in the scene; 3) they struggle to understand basic relations among objects. Furthermore, a comparative evaluation with human annotators reveals that VLMs still fall short of achieving human-level accuracy.

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.

We propose a training-free approach to improve sentence embeddings leveraging test-time compute by applying generative text models for data augmentation at inference time. Unlike conventional data augmentation that utilises synthetic training data, our approach does not require access to model parameters or the computational resources typically required for fine-tuning state-of-the-art models. Generatively Augmented Sentence Encoding variates the input text by paraphrasing, summarising, or extracting keywords, followed by pooling the original and synthetic embeddings. Experimental results on the Massive Text Embedding Benchmark for Semantic Textual Similarity (STS) demonstrate performance improvements across a range of embedding models using different generative models for augmentation. We find that generative augmentation leads to larger performance improvements for embedding models with lower baseline performance. These findings suggest that integrating generative augmentation at inference time adds semantic diversity and can enhance the robustness and generalisability of sentence embeddings for embedding models. Our results show that performance gains depend on the embedding model and the dataset.

GASE: Generatively Augmented Sentence Encoding

The generative large language models (LLMs) are increasingly used for data augmentation tasks, where text samples are paraphrased (or generated anew) and then used for downstream model fine-tuning. This is useful, especially for low-resource settings. For better augmentations, LLMs are prompted with examples (few-shot scenarios). Yet, the samples are mostly selected randomly, and a comprehensive overview of the effects of other (more ''informed'') sample selection strategies is lacking. In this work, we compare sample selection strategies existing in the few-shot learning literature and investigate their effects in LLM-based textual augmentation in a low-resource setting. We evaluate this on in-distribution and out-of-distribution model performance. Results indicate that while some ''informed'' selection strategies increase the performance of models, especially for out-of-distribution data, it happens only seldom and with marginal performance increases. Unless further advances are made, a default of random sample selection remains a good option for augmentation practitioners.

Use Random Selection for Now: Investigation of Few-Shot Selection Strategies in LLM-based Text Augmentation

The capabilities of large language models (LLMs) are advancing at an remarkable pace, along with a surge in cloud services that are powered by LLMs. Their convenience has gradually transformed the routines people work. However, for services such as document summarizing, editing, and so on, users need to upload relevant files or context to obtain the desired services, which may inadvertently expose their privacy. This paper aims to address the challenging balance between the convenience of LLMs services and user privacy concerns. Specifically, based on the structural and functional characteristics of LLMs, we have developed a strategy that safeguards user prompt while accessing LLM cloud services, even in scenarios where advanced reconstruction attacks are adopted. We comprehensively evaluate the efficacy of our method across prominent LLM benchmarks. The empirical results show that our method not only effectively thwarts reconstruction attacks but also, in certain tasks, even improves model performance, surpassing the outcomes reported in official model cards.

LLMs are Privacy Erasable

Understanding the core dimensions of conceptual semantics is fundamental to uncovering how meaning is organized in language and the brain. Existing approaches often rely on predefined semantic dimensions that offer only broad representations, overlooking finer conceptual distinctions. This paper proposes a novel framework to investigate the subdimensions underlying coarse-grained semantic dimensions. Specifically, we introduce a textbfDisentangled textbfContinuous textbfSemantic textbfRepresentation textbfModel (textbfDCSRM) that decomposes word embeddings from large language models into multiple sub-embeddings, each encoding specific semantic information. Using these subembeddings, we identify a set of interpretable semantic subdimensions. To assess their neural plausibility, we apply voxel-wise encoding models to map these subdimensions to brain activation. Our work offers a more fine-grained interpretable semantic subdimensions of conceptual meaning. Further analyses reveal that semantic dimensions are structured according to distinct principles, with polarity emerging as a key factor driving their decomposition into subdimensions. The neural correlates of the identified subdimensions support their cognitive and neuroscientific plausibility.

Discovering Semantic Subdimensions through Disentangled Conceptual Representations

Peer review is central to academic publishing, but the growing volume of submissions is straining the process. This motivates the development of computational approaches to support peer review. While each review is tailored to a specific paper, reviewers often make assessments according to certain aspects such as Novelty, which reflect the values of the research community. This alignment creates opportunities for standardizing the reviewing process, improving quality control, and enabling computational support. While prior work has demonstrated the potential of aspect analysis for peer review assistance, the notion of aspect remains poorly formalized. Existing approaches often derive aspects from review forms and guidelines, yet data-driven methods for aspect identification are underexplored. To address this gap, our work takes a bottom-up approach: we propose an operational definition of aspect and develop a data-driven schema for deriving aspects from a corpus of peer reviews. We introduce a dataset of peer reviews augmented with aspects and show how it can be used for community-level review analysis. We further show how the choice of aspects can impact downstream applications, such as LLM-generated review detection. Our results lay a foundation for a principled and data-driven investigation of review aspects, and pave the path for new applications of NLP to support peer review.

Identifying Aspects in Peer Reviews

In the era of prosperity of large language models (LLMs), hallucination remains a serious issue hindering LLMs' expansion and reliability. Predicting the presence (and absence) of certain knowledge in LLMs could aid the hallucination avoidance. However, the token-based generation mode of LLM is different from the knowledge storage structure in the form of triples, which makes it difficult to accurately evaluate the knowledge boundary of LLM. We approach this problem from a novel perspective and, for the first time, introduce the concept of {\em token knowledge} in large language models. Consequently, we propose a token knowledge dataset construction method and use the intermediate states during inference to train probes. This allows us to predict if a specific token will appear in the LLM’s generated sequence, without even generating a single token. Our approach unlocks the model's latent potential, enhancing its accuracy in assessing token knowledge from about 60% to over 90%, with strong out-of-distribution generalization by training on just a few dozen prompts. Finally, we apply KEGT to enhance a state-of-the-art knowledge boundary detection method, achieving improved performance while reducing computational time by over 90%. Furthermore, KEGT enables prevention of hallucinations in certain cases by leveraging its guidance in the token-level knowledge semantic space. Our code is available at https://anonymous.4open.science/r/KEGT.

Token Knowledge: A New Perspective For Knowledge in Large Language Models

Recent studies have explored integrating large langucage models (LLMs) into recommendation systems but face several challenges, including training-induced bias and bottlenecks from serialized architecture.To effectively address these issues, we propose a Query-to-Recommendation, a parallel recommendation framework that decouples LLMs from candidate pre-selection and instead enables direct retrieval over the entire item pool. Our framework connects LLMs and recommendation models in a parallel manner, allowing each component to independently utilize its strengths without interfering with the other. In this framework, LLMs are utilized to generate feature-enriched item descriptions and personalized user queries, allowing for capturing diverse preferences and enabling rich semantic matching in a zero-shot manner. To effectively combine the complementary strengths of LLM and collaborative signals, we introduce an adaptive reranking strategy. Extensive experiments demonstrate an improvement in performance up to 57%, while also improving the novelty and diversity of recommendations.

Rethinking LLM-Based Recommendations: A Personalized Query-Driven Parallel Integration

We introduce TR-MTEB, the first large-scale, task-diverse benchmark designed to evaluate sentence embedding models for Turkish. Covering six core tasks as classification, clustering, pair classification, retrieval, bitext mining, and semantic textual similarity, TR-MTEB incorporates 26 high-quality datasets, including native and translated resources. To complement this benchmark, we construct a corpus of 34.2 million weakly supervised Turkish sentence pairs and train two Turkish-specific embedding models using contrastive pretraining and supervised fine-tuning. Evaluation results show that our models, despite being trained on limited resources, achieve competitive performance across most tasks and significantly improve upon baseline monolingual models. All datasets, models, and evaluation pipelines are publicly released to facilitate further research in Turkish natural language processing and low-resource benchmarking.

TR-MTEB: A Comprehensive Benchmark and Embedding Model Suite for Turkish Sentence Representations

Large Language Models (LLMs) have demonstrated significant potential in specialized tasks, including medical problem-solving. However, most studies predominantly focus on English-language contexts. This study introduces a novel benchmark dataset based on Polish medical licensing and specialization exams (LEK, LDEK, PES). The dataset, sourced from publicly available materials provided by the Medical Examination Center and the Chief Medical Chamber, includes Polish medical exam questions, along with a subset of parallel Polish-English corpora professionally translated for foreign candidates. By structuring a benchmark from these exam questions, we evaluate state-of-the-art LLMs, spanning general-purpose, domain-specific, and Polish-specific models, and compare their performance with that of human medical students and doctors. Our analysis shows that while models like GPT-4o achieve near-human performance, challenges persist in cross-lingual translation and domain-specific understanding. These findings highlight disparities in model performance across languages and medical specialties, emphasizing the limitations and ethical considerations of deploying LLMs in clinical practice.

Polish-English medical knowledge transfer: A new benchmark and results

Despite the state-of-the-art performance of Large Language Models (LLMs), these models often suffer from hallucinations, which can undermine their performance in critical applications. In this work, we propose SAFE, a novel framework for detecting and mitigating hallucinations by leveraging Sparse Autoencoders (SAEs). While hallucination detection techniques and SAEs have been explored independently, their synergistic application in a comprehensive system, particularly for hallucination-aware query enrichment, has not been fully investigated. To validate the effectiveness of SAFE, we evaluate it on two models with available SAEs across four diverse cross-domain datasets designed to assess hallucination problems. Empirical results demonstrate that SAFE consistently improves query generation accuracy and mitigates hallucinations across all datasets, achieving accuracy improvements of up to 29.45%.

Downloads

Next from EMNLP 2025

GASE: Generatively Augmented Sentence Encoding

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES