Austria

Retrieval Augmented Generation (RAG) complements the knowledge of Large Language Models (LLMs) by leveraging external information to enhance response accuracy for queries. This approach is widely applied in several fields by taking its advantage of injecting the most up-to-date information, and researchers are focusing on understanding and improving this aspect to unlock the full potential of RAG in such high-stakes applications. However, despite the potential of RAG to address these needs, the mechanisms behind the confidence levels of its outputs remain underexplored. Our study focuses on the impact of RAG, specifically examining whether RAG increases the confidence of LLM outputs in the medical domain. We conduct this analysis across various configurations and models. We evaluate confidence by treating the model&#39;s predicted probability as its output and calculating several evaluation metrics which include calibration error method, entropy, best probability, and accuracy. Experimental results across multiple datasets confirmed that certain models possess the capability to judge for themselves whether an inserted document relates to the correct answer. These results suggest that evaluating models based on their output probabilities determine whether they function as generators in the RAG framework. Our approach allows to evaluate whether the models handle retrieved documents. and improving this aspect to unlock the full potential of RAG in such high-stakes applications. However, despite the potential of RAG to address these needs, the mechanisms behind the confidence levels of its outputs remain underexplored. Our study focuses on the impact of RAG, specifically examining whether RAG increases the confidence of LLM outputs in the medical domain. We conduct this analysis across various configurations and models. We evaluate confidence by treating the model&#39;s predicted probability as its output and calculating several evaluation metrics which include calibration error method, entropy, best probability, and accuracy. Experimental results across multiple datasets confirmed that certain models possess the capability to judge for themselves whether an inserted document relates to the correct answer. These results suggest that evaluating models based on their output probabilities determine whether they function as generators in the RAG framework. Our approach allows to evaluate whether the models handle retrieved documents.

ACL 2025

Understanding the Impact of Confidence in Retrieval Augmented Generation: A Case Study in the Medical Domain

## 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*

EMNLP 2025

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

Retrieval Augmented Generation (RAG) complements the knowledge of Large Language Models (LLMs) by leveraging external information to enhance response accuracy for queries. This approach is widely applied in several fields by taking its advantage of injecting the most up-to-date information, and researchers are focusing on understanding and improving this aspect to unlock the full potential of RAG in such high-stakes applications. However, despite the potential of RAG to address these needs, the mechanisms behind the confidence levels of its outputs remain underexplored. Our study focuses on the impact of RAG, specifically examining whether RAG increases the confidence of LLM outputs in the medical domain. We conduct this analysis across various configurations and models. We evaluate confidence by treating the model's predicted probability as its output and calculating several evaluation metrics which include calibration error method, entropy, best probability, and accuracy. Experimental results across multiple datasets confirmed that certain models possess the capability to judge for themselves whether an inserted document relates to the correct answer. These results suggest that evaluating models based on their output probabilities determine whether they function as generators in the RAG framework. Our approach allows to evaluate whether the models handle retrieved documents. and improving this aspect to unlock the full potential of RAG in such high-stakes applications. However, despite the potential of RAG to address these needs, the mechanisms behind the confidence levels of its outputs remain underexplored. Our study focuses on the impact of RAG, specifically examining whether RAG increases the confidence of LLM outputs in the medical domain. We conduct this analysis across various configurations and models. We evaluate confidence by treating the model's predicted probability as its output and calculating several evaluation metrics which include calibration error method, entropy, best probability, and accuracy. Experimental results across multiple datasets confirmed that certain models possess the capability to judge for themselves whether an inserted document relates to the correct answer. These results suggest that evaluating models based on their output probabilities determine whether they function as generators in the RAG framework. Our approach allows to evaluate whether the models handle retrieved documents.

workshop paper

### Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Message from the General Chair: 
*It is my great pleasure and honor to welcome you to the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025), held in beautiful Vienna, Austria, from July 27 to August 1, 2025. ACL2025continues our field’s tradition of excellence in scholarship, innovation, and inclusivity, and I am deeply grateful to the many volunteers who have worked tirelessly to bring this event to life.* 
[Read more](https://drive.google.com/file/d/1GI_hvOpjswAuYdUTromfeDiPpCcqidwg/view?usp=sharing)

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to ACL 2025". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you need to log in with the **email address you registered with**. 

Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Medical text summarization faces significant challenges due to the complexity and domain-specific nature of the language. Although large language models have achieved significant success in general domains, their effectiveness in the medical domain remains limited. This limitation stems from their insufficient understanding of domain-specific terminology and difficulty in interpreting complex medical relationships, which often results in suboptimal summarization quality. To address these challenges, we propose MedSummRAG, a novel retrieval-augmented generation (RAG) framework that integrates external knowledge to enhance summarization. Our approach employs a fine-tuned dense retriever, trained with contrastive learning, to retrieve relevant documents for medical summarization. The retrieved documents are then integrated with the input text to generate high-quality summaries. Experimental results show that MedSummRAG achieves significant improvements in ROUGE scores on both zero/few-shot and fine-tuned language models, outperforming baseline methods. These findings underscore the importance of RAG and domain adaptation of the retriever for medical text summarization. The source code of this paper can be obtained from: https://github.com/guantingluo98/MedSummRAG

MedSummRAG: Domain-Specific Retrieval for Medical Summarization

Social media platforms generate an enormous volume of multi-modal data, yet stress detection research has predominantly relied on text-based analysis. In this work, we propose a novel framework that integrates textual content with synthesized visual cues to enhance stress detection. Using the generative model DALL·E, we synthesize images from social media posts, which are then fused with text through the multi-modal capabilities of a pre-trained CLIP model. Our approach is evaluated on the Dreaddit dataset, where a classifier trained on frozen CLIP features achieves 94.90% accuracy, and full fine-tuning further improves performance to 98.41%. These results underscore the integration of synthesized visuals with textual data not only enhances stress detection but also offers a robust method over traditional text-only methods, paving the way for innovative approaches in mental health monitoring and social media analytics.

Enhancing Stress Detection on Social Media Through Multi-Modal Fusion of Text and Synthesized Visuals

We developed a new methodology of extracting the frequency of a patient's epilepsy seizures from unstructured, free-text outpatient clinic letters by: first, devising a singular unit of measurement for seizure frequency; and second, fine-tuning a generative Large Language Model (LLM) on our bespoke annotated dataset. We measured frequency by the number of seizures per month: one seizure or more requires an integer; and less than one a decimal. This approach enables us to track whether a patient's seizures are improving or not over time. We found fine-tuning improves the F1 score of our best-performing LLM, Ministral-8B-Instruct-2410, by around three times compared to an untrained model. We also found Ministral demonstrated an impressive ability for mathematical reasoning.

Fine-tuning LLMs to Extract Epilepsy Seizure Frequency Data from Health Records

Accurate identification and labeling of biomedical entities, such as diseases, genes, chemical and species, within scientific texts are crucial for understanding complex relationships. We propose Adaptive BERT or AdaBioBERT, a robust named entity recognition (NER) model that builds upon BioBERT (Biomedical Bidirectional Encoded Representation from Transformers) based on an adaptive loss function to learn different types of biomedical token sequence. This adaptive loss function combines the standard Cross Entropy (CE) loss and Conditional Random Field (CRF) loss to optimize both token level accuracy and sequence-level coherence. AdaBioBERT captures rich semantic nuances by leveraging pre-trained contextual embeddings from BioBERT. On the other hand, the CRF loss of AdaBioBERT ensures proper identification of complex multi-token biomedical entities in a sequence and the CE loss can capture the simple unigram entities in a sequence. The empirical analysis on multiple standard biomedical coprora demonstrates that AdaBioBERT performs better than the state of the arts for most of the datasets in terms of macro and micro averaged F1 score.

AdaBioBERT: Adaptive Token Sequence Learning for Biomedical Named Entity Recognition

Animal research, sometimes referred to as preclinical research, plays a vital role in bridging the gap between basic science and clinical applications. However, the rapid increase in publications and the complexity of reported findings make it increasingly difficult for researchers to extract and assess relevant information. While automation through natural language processing (NLP) holds great potential for addressing this challenge, progress is hindered by the absence of high-quality, comprehensive annotated resources specific to preclinical studies. To fill this gap, we introduce PreClinIE, a fully open manually annotated dataset. The corpus consists of abstracts and methods sections from 725 publications, annotated for study rigor indicators (e.g., random allocation) and other study characteristics (e.g., species). We describe the data collection and annotation process, outlining the challenges of working with preclinical literature. By providing this resource, we aim to accelerate the development of NLP tools that enhance literature mining in preclinical research.

PreClinIE: An Annotated Corpus for Information Extraction in Preclinical Studies

Many language models (LMs) in the literature claim excellent zero-shot and/or few-shot capabilities for named entity recognition (NER) and relation extraction (RE) tasks and assert their ability to generalize beyond their training datasets. However, these claims have yet to be tested across different model architectures. This paper presents a performance evaluation of zero-shot relation triplet extraction (NER followed by RE of the entities) for both small and large LMs, utilizing 13,867 texts from 61 biomedical corpora and encompassing 151 unique entity types. This comprehensive evaluation offers valuable insights into the practical applicability and performance of LMs within the intricate domain of biomedical relation triplet extraction, highlighting their effectiveness in managing a diverse range of relations and entity types. Gemini 1.5 Pro, the largest LM included in the study, was the top-performing zero-shot model, achieving an average partial match micro F1 of 0.492 for NER, followed closely by SciLitLLM 1.5 14B with a score of 0.475. Fine-tuned models generally outperformed others on the corpora they were trained on, even in a few-shot setting, but struggled to generalize across all datasets with similar entity types. No models achieved an F1 score above 0.5 for the RTE task on any dataset, and their scores fluctuated based on the specific class of entity and the dataset involved. This observation highlights that there is still large room for improvement on the zero-shot utility of LMs in biomedical RTE applications.

Benchmarking zero-shot biomedical relation triplet extraction across language model architectures

Extractive question answering over clinical text is a crucial need to help deal with the deluge of clinical text generated in hospitals. While encoder models (e.g., BERT) have been popular for this reading comprehension--style question answering task, recently encoder-decoder models (e.g., T5) are on the rise. There is also the emergence of preference optimization techniques to align decoder-only LLMs with human preferences. In this paper, we combine encoder-decoder models with the direct preference optimization (DPO) method for the RadQA radiology question answering task. Our approach achieves a 12--15 F1 point improvement over previous state-of-the-art models. To the best of our knowledge, this effort is the first to show that DPO method also works for reading comprehension via novel heuristics to generate preference data without human inputs.

RadQA-DPO: A Radiology Question Answering System with Encoder-Decoder Models Enhanced by Direct Preference Optimization

This paper presents a pipeline for mitigating gender bias in large language models (LLMs) used in medical literature by neutralizing gendered occupational pronouns. A set of 379,000 PubMed abstracts from 1965-1980 was processed to identify and modify pronouns tied to professions. We developed a BERT-based model, "Modern Occupational Bias Elimination with Refined Training," or "MOBERT," trained on these neutralized abstracts, and compared it with "1965BERT," trained on the original dataset. MOBERT achieved a 70% inclusive replacement rate, while 1965BERT reached only 4%. A further analysis of MOBERT revealed that pronoun replacement accuracy correlated with the frequency of occupational terms in the training data. We propose expanding the dataset and refining the pipeline to improve performance and ensure more equitable language modeling in medical applications.

Gender-Neutral Large Language Models for Medical Applications: Reducing Bias in PubMed Abstracts

Large Language Models (LLMs) have approached human-level performance in text generation and summarization, yet their application in clinical settings remains constrained by potential inaccuracies that could lead to serious consequences. This work addresses the critical safety weaknesses in medical documentation systems by focusing on detecting subtle errors that require specialized medical expertise. We introduce a novel multi-agent debating framework that achieves 78.8\% accuracy on medical error detection, significantly outperforming both single-agent approaches and previous multi-agent systems. Our framework leverages specialized LLM agents with asymmetric access to complementary medical knowledge sources (Mayo Clinic and WebMD), engaging them in structured debate to identify inaccuracies in clinical notes. A judge agent evaluates these arguments based solely on their medical reasoning quality, with agent-specific performance metrics incorporated as feedback for developing situation-specific trust models.

Error Detection in Medical Note through Multi Agent Debate

Biomedical entity linking models disambiguate mentions in text by matching them with unique biomedical concepts. This problem is commonly addressed using a two-stage pipeline comprising an inexpensive candidate generator, which filters a subset of suitable entities for a mention, and a costly but precise reranker that provides the final matching between the mention and the concept. With the goal of applying two-stage entity linking at scale, we explore the construction of effective cross-encoder reranker models, capable of scoring multiple mention-entity pairs simultaneously. Through experiments on four entity linking datasets, we show that our cross-encoder models provide between 2.7 to 36.97 times faster training speeds and 3.42 to 26.47 times faster inference speeds than a base cross-encoder model capable of scoring only one entity, while achieving similar accuracy (differences between -3.42% to 2.76% Acc@1).

Downloads

Next from ACL 2025

MedSummRAG: Domain-Specific Retrieval for Medical Summarization

Similar lecture

InfoGain-RAG: Boosting Retrieval-Augmented Generation through Document Information Gain-based Reranking and Filtering