United States

We present an overview of the MultiClinAI shared task, which focuses on multilingual clinical entity extraction and automatic corpus generation through annotation projection. It addresses two key challenges in clinical natural language processing (NLP): (i) developing comparable multilingual named entity recognition (NER) systems and (ii) automatically constructing multilingual clinical corpora through annotation projection. The MultiClinAI task provides a unified benchmark for evaluating multilingual and cross-lingual clinical NLP approaches that cover diseases, symptoms, and procedures in Spanish, English, Dutch, Italian, Romanian, Swedish, and Czech. A total of 21 teams from 13 countries participated, submitting 531 runs across the different subtasks. The top runs obtained very competitive results across several languages and entity types. The results highlight both the challenges and opportunities of multilingual clinical information extraction. All resources, including a corpus of over 738,201 manually revised entity mentions across seven languages, are publicly available on Zenodo at: https://zenodo.org/records/19334278.

ACL 2026

The MultiClinAI Shared Task on Multilingual Clinical Corpus Construction and Concept Extraction: Systems, Evaluation, and Datasets

annotation projection

multilinguality

named entity recognition

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. <br>
[Full Message from General Chair Philipp Koehn](https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing) <br><br><br>
<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>
<br><br><br><br>

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

workshop paper

We present two systems for the #SMM4H-
HeaRD 2026 Task 2 (Lopez-Garcia et al., 2026)
shared task of automated insomnia detection
from clinical notes. Our system addresses both
subtasks: (1) binary insomnia classification and
(2) multi-label rule prediction with evidence
span extraction. For Subtask 1, we employ an
ensemble architecture combining Qwen3-4B-
Instruct1 and Bio_ClinicalBERT2 to capture
both general semantic reasoning and domain-
specific clinical representations. The frame-
work utilizes chunk-based processing with over-
lapping token windows to handle long clinical
notes efficiently. For Subtask 2, we develop
a dual-head multi-task transformer model that
jointly predicts insomnia labels and token-level
evidence spans using BIO tagging3. To im-
prove clinical relevance, we additionally incor-
porate sentence-level filtering using sentence-
transformer embeddings and similarity-based
retrieval of insomnia-related contexts. Experi-
mental results suggest competitive performance
relative to the shared task mean and median
scores across both subtasks. Our best Subtask 1
system achieves a recall of 0.9474, surpassing
the shared-task mean and median recall, while
our Subtask 2 system exceeds the mean and
median scores across label classification, exact
match, and partial match metrics. The end-
to-end implementation is publicly available on
GitHub.

NoviceTrio in #SMM4H-HeaRD 2026: Hybrid Clinical Transformer
Ensembles for Insomnia Detection and Evidence Extraction from Clinical
Notes

The Tumor, Node, and Metastasis (TNM) stag-
ing system is critical to cancer treatment. This
study aims to predict TNM stage labels in-
dependently, with the Cancer Genome Atlas
(TCGA) pathology report as the sixth shared
task of SMM4H-HeaRD 2026. The problem is
framed as three multi-label classification tasks.
We explore both classical and deep learning ap-
proaches using Term Frequency-Inverse Doc-
ument Frequency (TF-IDF) features and em-
beddings from ClinicalBERT, BioBERT, and
PubMedBERT. These representations are used
with Logistic Regression (LR), Light Gradient
Boosting Machine (LightGBM), Feed-Forward
Neural Networks (FFNN), and Wide Residual
Networks (WRN). Our results show that in-
dividual embeddings perform similarly to the
TNM label classification, while their combi-
nation improves its predictive ability. WRN
achieves AUROC scores of 0.839 (T), 0.8502
(N), and 0.803 (M) with F1-scores of 0.622,
0.702, and 0.9337, respectively, for the train-
ing phase. LightGBM with TF-IDF performs
best with AUROC scores of 0.9368 (T), 0.9524
(N), and 0.8311 (M) and F1-scores of 0.7559
(T), 0.7384 (N), and 0.7017 (M) during the
training phase. Furthermore, the result of the
Codabench for the test sets indicates a Macro-
F1 score of 0.978, 0.957, and 0.879 for the T,
N, and M categories respectively for test set
1; while test set 2 records a Macro-F1 score
for T, N, and M is 0.807, 0.767, 1.0 respec-
tively. However, performance declined during
the evaluation phase of the test sets, a drop from
0.938 for test set 1 to 0.858 for test set 2, for the
Macro-F1 score across all stages; suggesting
limitations in model generalizability, sensitivity
to class imbalance, and challenges in process-
ing lengthy clinical documents. Although this
study provides an efficient baseline model and
a reproducible pipeline, further optimization
and validation are required before it can be con-
sidered suitable for use in a real-world clinical
setting.

CaresAI at SMM4H-HeaRD 2026: Predicting TNM Staging

In this paper, we address Task 1 on adverse
drug event (ADE) detection and Task 8 on Mul-
tiClinNER at SMM4H-HeaRD 2026. ADE
detection is formulated as a multilingual bi-
nary classification problem over social me-
dia posts spanning German, French, Russian,
English, Mandarin and Japanese, with zero-
shot on Farsi. Using XLM-RoBERTa-Large
with a dual-pooling head, combined with strat-
ified sampling, language-conditioned inputs,
translation-based augmentation, and calibrated
ensembling, our model achieves a macro F1
score of 0.6088, surpassing both the competi-
tion mean (0.5465) and median (0.5798). Our
work in MultiClinNER targets clinical NER
for English text. Using GLiNER-large with
sliding-window inference, query augmentation,
and calibrated thresholds, it achieves strict F1
scores of 0.7591 (Disease), 0.7263 (Procedure),
and 0.6733 (Symptom), outperforming a Pub-
MedBERT baseline across all entities.

Vinland_Vector at #SMM4H-HeaRD 2026: Multilingual ADE Detection
and Query-Augmented Clinical NER for English

We present our systems for Task 1 and Task 3
of the #SMM4H-HeaRD 2026 shared tasks.
Task 1 focuses on binary classification of ad-
verse drug event (ADE) mentions across seven
languages, including a zero-shot Persian set-
ting without labeled training data. We fine-tune
XLM-RoBERTa-large using weighted cross-
entropy loss and augment low-resource settings
with additional CADEC data and machine-
translation-based Persian augmentation. Our
system achieves a macro F1 score of 0.582, out-
performing the shared task average of 0.547.
Task 3 addresses influenza vaccine effective-
ness estimation through classification of vac-
cination status and flu-test results from X
posts. We fine-tune twitter-roberta-large,
achieving micro F1 scores of 0.845 for vaccina-
tion status and 0.883 for flu-test classification
on the official test set. Post-evaluation experi-
ments with focal loss, test-time augmentation,
and head-tail truncation further improve per-
formance. These results highlight the effec-
tiveness of robust transformer adaptation for
health-related social media classification.

Cuet_Data_Wizards at #SMM4H-HeaRD 2026:
Multilingual ADE Detection and Influenza Vaccine Effectiveness
Estimation from Social Media

We introduce Output-Space Search (OS-Search), which turns LLM generation into endpoint search. An outer loop selects a target z* in a frozen encoder-defined 3D output space Z, and a retrieval-grounded policy trained with sequence-level RL generates outputs whose coordinates land near z* under standard autoregressive decoding. This enables parallel sweeps and black-box optimization in Z without path-dependent token/program search. On stories, sweeping Z (text) yields 3.1x higher LLM-scored diversity than prompt-chaining. On code, Bayesian optimization over Z (code) improves an objective withheld from the controller under matched inference budgets while preserving validity.

Output-Space Search: Targeting LLM Generations in a Frozen Encoder-Defined Output Space

Code generation is increasingly critical for real-world applications. Still, diffusion-based large language models continue to struggle with this demand. Unlike free-form text, code requires syntactic precision; even minor structural inconsistencies can render a program non-executable. Existing diffusion-based large language models rely on random token masking for corruption, leading to two key failures: they lack awareness of syntactic boundaries during the iterative denoising process, and they fail to capture the long-range hierarchical dependencies essential for program correctness.

We propose TreeDiff to address both issues. Specifically, we propose a syntax-aware diffusion framework that incorporates structural priors from Abstract Syntax Tree (AST) into the corruption process. Instead of masking individual tokens at random, we selectively mask tokens belonging to key AST nodes. By aligning the corruption process with the underlying structure of code, our method encourages the model to internalize the compositional nature of programming languages, enabling it to reconstruct programs that respect grammatical boundaries and capture long-range dependencies. Our method achieves a 13.3% relative improvement over the random masking training method, demonstrating its effectiveness in code generation task by leveraging underlying structures.

TreeDiff: AST-Guided Code Generation with Diffusion LLMs

Representing continuous time is a critical and under-explored challenge in modeling temporal event sequences with large language models (LLMs). Various strategies like byte-level representations or calendar tokens have been proposed. However, the optimal approach remains unclear, especially given the diverse statistical distributions of real-world event data, which range from smooth log-normal to discrete, spiky patterns. This paper presents a systematic empirical study of temporal tokenization for modeling event sequences with LLMs, comparing distinct encoding strategies: naive numeric strings, high-precision byte-level representations, human-semantic calendar tokens, classic uniform binning, and adaptive residual scalar quantization. We evaluate these strategies by fine-tuning LLMs on real-world datasets that exemplify these diverse distributions. Our analysis reveals that no single strategy is universally superior; instead, prediction performance depends heavily on aligning the tokenizer with the data's statistical properties, highlighting temporal tokenization as a critical yet often overlooked design dimension in LLM-based event modeling.

Temporal Tokenization Strategies for Event Sequence Modeling with Large Language Models

We introduce EconWebArena, a benchmark for evaluating autonomous agents on complex, multimodal economic tasks in realistic web environments. The benchmark comprises 360 curated tasks from 82 authoritative websites spanning domains such as macroeconomics, labor, finance, trade, and public policy. Each task challenges agents to navigate live websites, interpret structured and visual content, interact with real interfaces, and extract precise, time-sensitive data through multi-step workflows. We construct the benchmark by prompting multiple large language models (LLMs) to generate candidate tasks, followed by rigorous human curation to ensure clarity, feasibility, and source reliability. Unlike prior work, EconWebArena emphasizes fidelity to authoritative data sources and the need for grounded web-based economic reasoning. We evaluate a diverse set of state-of-the-art multimodal LLMs as web agents, analyze failure cases, and conduct ablation studies to assess the impact of visual grounding, plan-based reasoning, and interaction design. Our results reveal substantial performance gaps and highlight persistent challenges in grounding, navigation, and multimodal understanding, positioning EconWebArena as a rigorous testbed for economic web intelligence.

EconWebArena: Benchmarking Autonomous Agents on Economic Tasks in Realistic Web Environments

Accurate chart comprehension represents a critical challenge in advancing multimodal learning systems, as extensive information is compressed into structured visual representations. However, existing vision-language models (VLMs) frequently struggle to generalize on unseen charts because it requires abstract, symbolic, and quantitative reasoning over structured visual representations. In this work, we introduce Chart-RL, an effective reinforcement learning (RL) method that employs mathematically verifiable rewards to enhance chart question answering in VLMs. Our experiments demonstrate that Chart-RL consistently outperforms supervised fine-tuning (SFT) across different chart understanding benchmarks, achieving relative improvements of 16.7% on MultiChartQA, and 11.5% on ChartInsights. We conduct robustness analysis, where Chart-RL achieves enhanced performance in 18 of 25 perturbed chart categories, demonstrating strong consistency and reasoning capability across visual variations. Furthermore, we demonstrate that task difficulty and inherent complexity are more critical than data quantity in RL training. For instance, Chart-RL trained on merely 10 complex chart-query examples significantly outperforms models trained on over 6,000 simple examples. Additionally, training on challenging reasoning tasks not only improves in-domain generalization relative to simpler tasks, but also facilitate strong transfer to out-of-domain visual mathematical problems.

Chart-RL: Generalized Chart Comprehension via Reinforcement Learning with Verifiable Rewards

When a language model answers a table question, users have no way to verify which cells informed which reasoning steps. We introduce RSAT, a method that trains small language models (SLMs, 1–8B) to produce step-by-step reasoning with cell-level citations grounded in table evidence. Phase 1 (SFT) teaches a structured JSON output format from verified reasoning traces. Phase 2 (GRPO) optimizes a composite reward centered on NLI-based faithfulness, alongside citation validity and parsimony. Across six models from two families—Qwen2.5 (1.5B/3B/7B) and Llama3 (1B/3B/8B)—RSAT improves faithfulness 3.7$\times$ over SFT alone (0.224$\rightarrow$0.826), with near-perfect citation validity (0.992). Post-hoc attribution collapses below 13\% format success, confirming that attribution must be integrated into reasoning, not retrofitted. Ablations show the faithfulness reward is essential: removing it drops faithfulness from 0.97 to 0.03.

Downloads

Next from ACL 2026

NoviceTrio in #SMM4H-HeaRD 2026: Hybrid Clinical Transformer Ensembles for Insomnia Detection and Evidence Extraction from Clinical Notes

Similar lecture

Enigma at #SMM4H–HeaRD 2026: Leveraging Multilingual Pre-trained Models for Clinical Named Entity Recognition

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