Luís Filipe Cunha

Abstract
In healthcare, diagnoses usually rely on physician expertise. However, complex cases may benefit from consulting similar past clinical reports cases. In this paper, we present MedLink (http://medlink.inesctec.pt), a tool that given a free-text medical report, retrieves and ranks relevant clinical case reports published in health conferences and journals, aiming to support clinical decision-making, particularly in challenging or complex diagnoses. To this regard, we trained two BERT models on the sentence similarity task: a bi-encoder for retrieval and a cross-encoder for reranking. To evaluate our approach, we used 10 medical reports and asked a physician to rank the top 10 most relevant published case reports for each one. Our results show that MedLink’s ranking model achieved NDCG@10 of 0.747. Our demo also includes the visualization of clinical entities (using a NER model) and the production of a textual explanation (using a LLM) to ease comparison and contrasting between reports. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

Abstract
Manual text annotation is a complex and time-consuming task. However, recent advancements demonstrate that such a task can be accelerated with automated pre-annotation. In this paper, we present a methodology to improve the efficiency of manual text annotation by leveraging LLMs for text pre-annotation. For this purpose, we train a BERT model for a token classification task and integrate it into the INCEpTION annotation tool to generate span-level suggestions for human annotators. To assess the usefulness of our approach, we conducted an experiment where an experienced linguist annotated plain text both with and without our model’s pre-annotations. Our results show that the model-assisted approach reduces annotation time by nearly 23%. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

Abstract
Electronic Health Records (EHRs) contain vast amounts of unstructured narrative text, posing challenges for organization, curation, and automated information extraction in clinical and research settings. Developing effective annotation schemes is crucial for training extraction models, yet it remains complex for both human experts and Large Language Models (LLMs). This study compares human- and LLM-generated annotation schemes and guidelines through an experimental framework. In the first phase, both a human expert and an LLM created annotation schemes based on predefined criteria. In the second phase, experienced annotators applied these schemes following the guidelines. In both cases, the results were qualitatively evaluated using Likert scales. The findings indicate that the human-generated scheme is more comprehensive, coherent, and clear compared to those produced by the LLM. These results align with previous research suggesting that while LLMs show promising performance with respect to text annotation, the same does not apply to the development of annotation schemes, and human validation remains essential to ensure accuracy and reliability. © 2025 Copyright for this paper by its authors.

Abstract
The capabilities of the most recent language models have increased the interest in integrating them into real-world applications. However, the fact that these models generate plausible, yet incorrect text poses a constraint when considering their use in several domains. Healthcare is a prime example of a domain where text-generative trustworthiness is a hard requirement to safeguard patient well-being. In this paper, we present Physio, a chat-based application for physical rehabilitation. Physio is capable of making an initial diagnosis while citing reliable health sources to support the information provided. Furthermore, drawing upon external knowledge databases, Physio can recommend rehabilitation exercises and over-the-counter medication for symptom relief. By combining these features, Physio can leverage the power of generative models for language processing while also conditioning its response on dependable and verifiable sources. A live demo of Physio is available at https://physio.inesctec.pt.

Abstract
Event extraction is an NLP task that commonly involves identifying the central word (trigger) for an event and its associated arguments in text. ACE-2005 is widely recognised as the standard corpus in this field. While other corpora, like PropBank, primarily focus on annotating predicate-argument structure, ACE-2005 provides comprehensive information about the overall event structure and semantics. However, its limited language coverage restricts its usability. This paper introduces ACE-2005-PT, a corpus created by translating ACE-2005 into Portuguese, with European and Brazilian variants. To speed up the process of obtaining ACE-2005-PT, we rely on automatic translators. This, however, poses some challenges related to automatically identifying the correct alignments between multi-word annotations in the original text and in the corresponding translated sentence. To achieve this, we developed an alignment pipeline that incorporates several alignment techniques: lemmatization, fuzzy matching, synonym matching, multiple translations and a BERT-based word aligner. To measure the alignment effectiveness, a subset of annotations from the ACE-2005-PT corpus was manually aligned by a linguist expert. This subset was then compared against our pipeline results which achieved exact and relaxed match scores of 70.55% and 87.55% respectively. As a result, we successfully generated a Portuguese version of the ACE-2005 corpus, which has been accepted for publication by LDC.