United States

Medical question answering is a high-stakes setting where factual errors can have serious consequences. Retrieval-augmented generation (RAG) is widely viewed as a promising solution, and prior work has reported substantial gains for large medical QA models. We revisit this assumption across a broad range of open-weight instruction-tuned models spanning 7B to 72B parameters. Across five models, ten biomedical QA datasets, four retrieval methods, and four retrieval corpora, we find that retrieval yields only small and inconsistent improvements over a no-retrieval baseline, typically within 1--2 points. In contrast, the choice of backbone model has a much larger effect than the choice of retriever or corpus, and expert and layman retrieval sources perform similarly in most settings. These results suggest that the main bottleneck is not retrieval quality alone, but the model&#39;s limited ability to use retrieved evidence effectively.

ACL 2026

When Retrieval Doesn&#39;t Help: A Large-Scale Study of Biomedical RAG

evidence utilization

medical nlp

biomedical question answering

retrieval-augmented generation

large language models

### Welcome to 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics

Welcome to the 2025 meeting of the Nations of the Americas Chapter of the Association for Computational Linguistics! I am proud to help organize the first NAACL conference to carry the new name of our organization, one that emphasizes inclusion for all of the Americas. I am also pleased to welcome you to Albuquerque, New Mexico, a state whose unique blend of cultural influences will make for an excellent backdrop for NAACL 2025, especially with this year’s special theme on NLP in a Multicultural World. 
**[Continue reading...](https://drive.google.com/file/d/1jX-qGhqVSZZCIrAnJaz798pflu5Irrdn/view?usp=sharing)**

*- Colin Cherry, Google, NAACL 2025 General Chair* 

[![](https://assets.underline.io/markdown_image/1/image/b087f8a4dc5816d6e1a6514e59c59ac3.png)](https://drive.google.com/file/d/1T96GzPqObXrMln2BMByCSXTSizjTg69P/view?usp=sharing)

You need to log in with the email address you registered with. Access credentials have been sent to your email. 

Please be sure to check your spam and other email folders if you do not see an email confirmation right away.

Please log in to explore this event.

To access NAACL 2025 event page you are required to register. Please follow [**this link**](https://2025.naacl.org/registration/registration/) to register. Access will depend on your registration type.

Please register!

NAACL 2025

Welcome to 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics

Medical question answering is a high-stakes setting where factual errors can have serious consequences. Retrieval-augmented generation (RAG) is widely viewed as a promising solution, and prior work has reported substantial gains for large medical QA models. We revisit this assumption across a broad range of open-weight instruction-tuned models spanning 7B to 72B parameters. Across five models, ten biomedical QA datasets, four retrieval methods, and four retrieval corpora, we find that retrieval yields only small and inconsistent improvements over a no-retrieval baseline, typically within 1--2 points. In contrast, the choice of backbone model has a much larger effect than the choice of retriever or corpus, and expert and layman retrieval sources perform similarly in most settings. These results suggest that the main bottleneck is not retrieval quality alone, but the model's limited ability to use retrieved evidence effectively.

When Retrieval Doesn't Help: A Large-Scale Study of Biomedical RAG

workshop paper

It is my great pleasure and honor to welcome you to the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026), held in San Diego, California, USA, July 2–7, 2026. ACL 2026 continues our fields 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.

It is exciting to see the many developments in the field reflected at our conference, such as the advancement of large language models (born out of our community) into many application directions. This is not without challenges, partly due to unsolved problems, and partly due to misuse of the technology. All this is reflected well at ACL 2026. The conference also reflects that the field is not only about advancing technology, but also the science of fundamentally understanding language and how it is use by many different speaker communities. 
[Full Message from General Chair Philipp Koehn](https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing) 
<html><button style="display: inline-flex; align-items: center; justify-content: center; white-space: nowrap; border-radius: 9999px; font-weight: bold; background: #7c3aed; color: white; font-family: 'Space Grotesk', sans-serif; height: 40px; font-size: 16px; padding: 0 20px; border: none; cursor: pointer" onclick="window.open('https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing')">Conference Handbook</button></html>

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 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://2026.aclweb.org/registration/) first.

Registration Required

The 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026) will take place in San Diego, California, from July 2 - 7, 2026.

Visual-language models (VLMs) are rapidly advancing on tasks that require visual understanding of text, tables, plots, and diagrams. Yet extracting structured information from text-heavy scientific diagrams remains challenging, as it requires not only OCR but also recovery of layout, grouping, and flow relationships. We study this problem in the context of CONSORT flow diagrams, which summarize participant screening, randomization, follow-up, and analysis in randomized controlled trials. We introduce a 200-example benchmark of PubMed Central diagrams, annotated by a biomedical team specializing in systematic literature reviews and clinical evidence extraction, and evaluate schema-constrained CONSORT extraction across proprietary and open-weight model families. Using structure-aware metrics, we compare single-pass and stepwise extraction strategies. Expert-guided single-pass extraction performs best for proprietary frontier models, with Gemini 3 Pro achieving the strongest overall results, whereas stepwise prompting improves less capable open-weight models on challenging arm-level extraction. These results offer practical deployment guidance and suggest that high-quality schema-constrained extraction is feasible, but not yet solved.

Diagnosis: Bad Planning & Reasoning. Treatment: SCoPE - Planning for Hybrid Querying over Clinical Trial Data

Tabular data is widely used in important areas such as healthcare and finance, but building accurate models in real-world settings faces three main challenges: protecting data privacy, handling distributed data, and maintaining strong performance. Existing methods do not solve these issues together. Converting tabular data into text for Large Language Models (LLMs) can expose sensitive information, struggle with anonymized features and exact numerical values, and require expensive training while often not outperforming traditional tree-based models. In addition, many real-world datasets are spread across different institutions, making centralized training impossible. We propose a federated framework that connects distributed tabular data with LLM reasoning using decision tree rules as privacy-preserving intermediaries. Each client trains a local Random Forest and shares only extracted rules—feature comparisons and thresholds, without revealing raw data. These rules are combined into a global pool, allowing an LLM to generate a better partitioning rule without accessing any original data, adding an extra layer of privacy. Using this rule, each client learns local gradient-based corrections, which are then aggregated. We also show that this process reduces prediction error. Experiments on 12 datasets, including seven medical tasks, show that our method consistently outperforms federated baselines and achieves results close to centralized models.

Expert-Guided Schema-Based Structured Extraction from CONSORT Diagrams Using Vision-Language Models

The rapid expansion of biomedical literature makes manual identification of novel drug-disease relationships increasingly difficult. Existing approaches have leveraged LLMs to mine abstracts or construct knowledge graphs for drug repurposing. There are two key limitations: finite context windows for capturing macro-level research trends, and single-pass black-box pipelines make it difficult to verify outputs. This paper proposes a pipeline for discovering new drug targets by combining disease and drug research trends using Large Language Models (LLMs). Our method extracts PICO components from PubMed abstracts, normalizing the Population and Intervention Component to ICD and ATC codes, respectively. A temporal frequency delta matrix is constructed to capture publication count shifts across 2013 to 2022, then used to discover novel drug areas. Compared with the abstract-based baseline, our approach showed qualitative signs of generating combinations that were more closely aligned with observed research trends and, in some cases, more clinically plausible. These findings suggest the potential usefulness of structured trend information for LLM-based exploration, although the differences between the two methods were limited and the results remain preliminary. Future work will focus on validating the consistency and reliability of these candidates.

Fast, Accurate, and Local Conversion of MIMIC-IV to OMOP with DBT

Family health history (FHx) offers insight into a person's health and disease risk, but it is largely held within free-text clinical formats that require processing for maximal utility of the data. The rapid deployment of ambient AI scribes and conversational agents in clinical settings necessitates evaluation on dynamic patient-clinician and patient-agent dialogs. To address this gap, we introduce two new datasets of patient FHx dialog documents designed to benchmark information extraction and entity linking. Distinct from clinician-entered datasets, patient-reported dialog data has its own semantic and content characteristics, which need to be studied for more patient-centered healthcare. We contribute a publicly available resource called FHexchange, with new annotations for family members, clinical observations, related entities, and standardized UMLS CUIs, offering the clinical NLP community a robust evaluation bed for emerging generative AI tools.

Exploring Novel Drug Research Area using Large Language Models Based on Research Trends in Biomedical Literature

We present IndicMedDialog, a parallel multi-turn medical dialogue dataset spanning English and nine Indic languages (Assamese, Bengali, Gujarati, Hindi, Marathi, Punjabi, Tamil, Telugu, and Urdu). The dataset extends the MDDial corpus with LLM-generated synthetic consultations, translated using TranslateGemma, verified by native speakers, and refined through a script-aware post-processing pipeline to correct phonetic, lexical, and character-spacing errors introduced during automatic translation. Building on this dataset, we fine-tune IndicMedLM via parameter-efficient adaptation (LoRA) of a quantized small language model, incorporating an optional patient pre-context to personalise multi-turn symptom elicitation. We evaluate IndicMedLM against zero-shot multilingual baselines across ten languages and conduct systematic error analysis, identifying five failure modes: Instruction Drift, Label Collapse, Cross-Domain Confusion, Tokenization Failure, and Paraphrase-over-Label Generation. Results show strong post-processed diagnostic accuracy in Hindi, Marathi, and Bengali, while Assamese, Tamil, and Telugu remain in an extreme failure tier attributable to base-model tokenizer gaps, a finding with direct patient safety implications. Medical expert evaluation confirms the clinical plausibility and safety of the generated consultations.

Forgotten Words: Benchmarking NeoBERT for Dementia Detection in Low-Resource Conversational Filipino and English Speech

Three-dimensional Computed Tomography (3D CT) is a cornerstone of precision medicine. Most AI diagnostic models analyze large num bers of CTslices uniformly, treating all slices as equally important. While this has partly accel erated radiologists'workflows, it overlooks that clinically relevant information is often sparsely distributed throughout a volume. Without tar geted or weighted processing, fine-grained cues may be missed and substantial computation wasted on diagnostically uninformative slices. Wepropose aradiologist-simulating framework for selective and efficient 3D CT interpreta tion. Evaluated on a 3D CT dataset covering eight thoracic lesion types, it was compared with state-of-the-art multimodal large language models such as GPT-4o and supervised visual backbones including ViT and ResNet-50. Us ing accuracy, F1-score, AUC, and blind radiolo gist assessment, Screen-CLIP achieved an AUC of 0.87 and F1-score of 0.82, surpassing ViT Base (AUC: 0.84). For report generation, our method outperformed M3D across all metrics, reaching a BLEU-Avg of 29.03, and achieved the highest average Doctors' Score (6.16/10) in a preliminary human evaluation.

IndicMedDialog: A Parallel Multi-Turn Medical Dialogue Dataset for Accessible Healthcare in Indic Languages

This paper presents an overview of the ClinicalSkillQA 2026 shared task, which was organized with the BioNLP Workshop at ACL 2026. The goal of this shared task is to evaluate continuous perception and procedural reasoning in clinical skill assessment by requiring systems to reconstruct the correct temporal order of shuffled clinical key frames and generate rationales grounded in clinical workflow knowledge. The benchmark contains 200 test-only instances sampled from clinical skill videos, covering three emergency-care procedures. Each instance is annotated with the ground-truth temporal order and an expert-verified rationale. A total of seven teams participated in the task, collectively making 90 submissions, with four teams providing system description papers. Systems are evaluated using Task Accuracy, Pairwise Accuracy, and BERTScore, which measure exact sequence reconstruction, local temporal consistency, and rationale quality, respectively. In this paper, we describe the task setup, dataset construction, and evaluation criteria. We further summarize the methodologies adopted by participating teams and present a comprehensive analysis of the submitted systems. The official results suggest that current models still struggle with continuous perception and procedural reasoning, especially when they must integrate visual evidence, temporal structure, and clinical workflow knowledge.

Overview of the MedGenVidQA 2026 Shared Task on Medical Generative Video Question Answering

We build an ensemble of 10 transformer encoders for the MedExACT 2026 shared task on medical decision span detection. The ensemble is diversified along three training directions: encoder initialization (including domain-adaptive pre-training on clinical text), loss function, and data augmentation with LLM-generated synthetic notes and silver-labeled clinical documents. Greedy forward search selects the combination with the highest validation final score. A BERT-based boundary refiner is applied to the ensemble's predicted spans to correct offset errors before submission.

Otter at MedExAct2026: Diverse Encoder Ensemble for Medical Decision Span Detection

We present our system for the PsyDefDetect shared task, which focuses on detecting and classifying psychological defense mechanisms in peer emotional support conversations. Our core contribution is a hierarchical classification framework that structures prediction as a coarse-to-fine pipeline over a clinically validated label 
 hierarchy, grounded in the Defense Mechanism Rating Scales (DMRS). Through systematic experimentation with flat fine-tuning, few-shot prompting, and hierarchical classification, we demonstrate that explicitly modelling the structured relationships among defense levels offers a more effective alternative to flat classification, achieving a macro F1 of 0.23 on the official test set.

KCL-Cogstack at PsyDefDetect: A Hierarchical Approach to Detecting Defense Mechanisms in Supportive Dialogue

This paper presents an ensemble of Qwen3.5-4B language models for extracting medical decisions from discharge summaries in the MedDec dataset. The models were trained to annotate discharge summaries with inline XML-like tags. Three different training strategies were used including dynamic fine-tuning, reinforcement learning, and pseudo-label augmentation. By combining predictions based on inter-model agreement, the system improved performance across evaluation metrics, achieving an overall F1 of 0.5942 and ranking second on the test leaderboard. The results also showed stable performance across demographic groups, suggesting fairness for underrepresented populations.

Downloads

Next from ACL 2026

Diagnosis: Bad Planning & Reasoning. Treatment: SCoPE - Planning for Hybrid Querying over Clinical Trial Data

Similar lecture

Rationale-Guided Retrieval Augmented Generation for Medical Question Answering

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from ACL 2026

Diagnosis: Bad Planning & Reasoning. Treatment: SCoPE - Planning for Hybrid Querying over Clinical Trial Data

Similar lecture

Rationale-Guided Retrieval Augmented Generation for Medical Question Answering

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Downloads