Conversational data collected in domains such as healthcare or social sciences is a valuable resource for research and automated analysis. However, responsible data sharing requires the detection and removal of personally identifiable and sensitive information to protect individual privacy. To support the development and evaluation of automatic de-identification systems, we present DialogPII, a multilingual dataset of synthetic dialogs and speech-derived transcripts for personal information detection. DialogPII covers eight interaction scenarios (emergency calls, medical anamnesis interviews, therapy sessions, insurance communication, customer support, clinical interviews regarding an AI-supported dashboard, police reports, and group therapy discussions), 19 entity types, and 11 languages (English, Arabic, Finnish, French, German, Hindi, Italian, Polish, Portuguese, Spanish, and Turkish). Dialogs were generated semi-automatically using large language models, manually curated for plausibility and diversity, and localized to country- and city-specific contexts. All dialogs were additionally converted to speech via text-to-speech synthesis, transcribed with Whisper, and annotated through automatic projection and manual correction, yielding aligned written and speech-derived resources across all languages. We further release baseline multilingual named entity recognition models and provide technical validation through inter-annotator agreement analysis, translation quality evaluation, annotation projection assessment, and benchmark experiments with transformer-based sequence labeling models.

Argentina has a diverse, yet little-known, Indigenous language heritage. Most of these languages are at risk of disappearing, resulting in a significant loss of world heritage and cultural knowledge. Currently, no unified information on speakers and computational tools is available for these languages. In this work, we present a systematization of the Indigenous languages spoken in Argentina, along with national demographic data on the country's Indigenous population. The languages are classified into seven families: Mapuche, Tupí-Guaraní, Guaycurú, Quechua, Mataco-Mataguaya, Aymara, and Chon. We also provide an introductory survey of the computational resources available for these languages, whether or not they are specifically developed for Argentine varieties.

Hate speech detection deals with many language variants, slang, slurs, expression modalities, and cultural nuances. This outlines the importance of working with specific corpora, when addressing hate speech within the scope of Natural Language Processing, recently revolutionized by the irruption of Large Language Models. This work presents a brief analysis of the performance of large language models in the detection of Hate Speech for Rioplatense Spanish. We performed classification experiments leveraging chain-of-thought reasoning with ChatGPT 3.5, Mixtral, and Aya, comparing their results with those of a state-of-the-art BERT classifier. These experiments outline that, even if large language models show a lower precision compared to the fine-tuned BERT classifier and, in some cases, they find hard-to-get slurs or colloquialisms, they still are sensitive to highly nuanced cases (particularly, homophobic/transphobic hate speech). We make our code and models publicly available for future research.

Information retrieval (IR) is the task of finding relevant documents in response to a user query. Although Spanish is the second most spoken native language, there are few Spanish IR datasets, which limits the development of information access tools for Spanish speakers. We introduce MessIRve, a large-scale Spanish IR dataset with almost 700,000 queries from Google's autocomplete API and relevant documents sourced from Wikipedia. MessIRve's queries reflect diverse Spanish-speaking regions, unlike other datasets that are translated from English or do not consider dialectal variations. The large size of the dataset allows it to cover a wide variety of topics, unlike smaller datasets. We provide a comprehensive description of the dataset, comparisons with existing datasets, and baseline evaluations of prominent IR models. Our contributions aim to advance Spanish IR research and improve information access for Spanish speakers.

OBJECTIVE:Our study aims to enhance epidemic intelligence through event-based surveillance in an emerging pandemic context. We classified electronic health records (EHRs) from La Rioja, Argentina, focusing on predicting COVID-19-related categories in a scenario with limited disease knowledge, evolving symptoms, non-standardized coding practices, and restricted training data due to privacy issues. METHODS:Using natural language processing techniques, we developed rapid, cost-effective methods suitable for implementation with limited resources. We annotated a corpus for training and testing classification models, ranging from simple logistic regression to more complex fine-tuned transformers. RESULTS:The transformer-based, Spanish-adapted models BETO Clínico and RoBERTa Clínico, further pre-trained with an unannotated portion of our corpus, were the best-performing models (F1= 88.13% and 87.01%). A simple logistic regression (LR) model ranked third (F1=85.09%), outperforming more complex models like XGBoost and BiLSTM. Data classified as COVID-confirmed using LR and BETO Clínico exhibit stronger time-series Pearson correlation with official COVID-19 case counts from the National Health Surveillance System (SNVS 2.0) in La Rioja province compared to the correlations observed between the International Code of Diseases (ICD-10) codes and the SNVS 2.0 data (0.840, 0.873, and 0.663, p-values ≤3×10-7). Both models have a good Pearson correlation with ICD-10 codes assigned to the clinical notes for confirmed (0.940 and 0.902) and for suspected cases (0.960 and 0.954), p-values ≤1.7×10-18. CONCLUSION:This study shows that simple, resource-efficient methods can achieve results comparable to complex approaches. BETO Clínico and LR strongly correlate with official data, revealing uncoded confirmed cases at the pandemic's onset. Our results suggest that annotating a smaller set of EHRs and training a simple model may be more cost-effective than manual coding. This points to potentially efficient strategies in public health emergencies, particularly in resource-limited settings, and provides valuable insights for future epidemic response efforts.

Automatic symptom detection from electronic health records is a valuable source for event-based surveillance systems. In this study, we develop tools to automatically detect symptoms associated with febrile illnesses in electronic health records written in Spanish. Therefore, we use a custom corpus, comprising 6228 expertly labeled and approximately 1 million unlabeled health reports. Our approach involved fine-tuning state-of-the-art named entity recognition models, including BiLSTM-CRF and transformer-based models like RoBERTa. We focused on domain-adaptive and task-adaptive models to enhance performance: the former were pretrained on biomedical corpora, while the latter were further pretrained on our unlabeled health reports. Despite computational constraints, our models demonstrated promising results, with RoBERTa-Clinico, a task-adaptive transformer model pretrained in our unlabeled corpus, showing the best micro recall performance (79.30), and 70.83 micro F1 score, which are comparable to results in similar studies. In this way, we contribute to the limited body of work in BioNLP in Spanish.

Social networks and other digital media deal with huge amounts of user-generated contents where hate speech has become a problematic more and more relevant. A great effort has been made to develop automatic tools for its analysis and moderation, at least in its most threatening forms, such as in violent acts against people and groups protected by law. One limitation of current approaches to automatic hate speech detection is the lack of context. The spotlight on isolated messages, without considering any type of conversational context or even the topic being discussed, severely restricts the available information to determine whether a post on a social network should be tagged as hateful or not. In this work, we assess the impact of adding contextual information to the hate speech detection task. We specifically study a subdomain of Twitter data consisting of replies to digital newspapers posts, which provides a natural environment for contextualized hate speech detection. We built a new corpus in Spanish (Rioplatense variant) focused on hate speech associated to the COVID-19 pandemic, annotated using guidelines carefully designed by our interdisciplinary team. Our classification experiments using state-of-the-art transformer-based machine learning techniques show evidence that adding contextual information improves the performance of hate speech detection for two proposed tasks: binary and multi-label prediction, increasing their Macro F1 by 4.2 and 5.5 points, respectively. These results highlight the importance of using contextual information in hate speech detection. Our code, models, and corpus has been made available for further research.

Background: In order to detect threats to public health and to be well-prepared for endemic and pandemic illness outbreaks, countries usually rely on event-based surveillance (EBS) and indicator-based surveillance systems. Event-based surveillance systems are key components of early warning systems and focus on fast capturing of data to detect threat signals through channels other than traditional surveillance. In this study, we develop Natural Language Processing tools that can be used within EBS systems. In particular, we focus on information extraction techniques that enable digital surveillance to monitor Internet data and social media.Results: We created an annotated Spanish corpus from ProMED-mail health reports regarding disease outbreaks in Latin America. The corpus has been used to train algorithms for two information extraction tasks: named entity recognition and relation extraction. The algorithms, based on deep learning and rules, have been applied to recognize diseases, hosts, and geographical locations where a disease is occurring, among other entities and relations. In addition, an in-depth analysis of micro-average F1 metrics shows the suitability of our approaches for both tasks.Conclusions: The annotated corpus and algorithms presented could leverage the development of automated tools for extracting information from news and health reports written in Spanish. Moreover, this framework could be useful within EBS systems to support the early detection of Latin American disease outbreaks.

In this paper, we provide an overview of the ninth annual edition of the CLEF eHealth evaluation lab. CLEF eHealth 2021 continues our evaluation resource building efforts around the easing and support of patients, their next-of-kins, health care professionals, and health scientists in understanding, accessing, and authoring electronic health information in a multilingual setting. The 2021 lab offered two tasks: Task 1 on multilingual Information Extraction (IE), this year extending to a corpus of Spanish radiology reports; and Task 2 on Consumer Health Search (CHS) that builds on the previous year's Information Retrieval (IR) tasks. In total, 11 teams took part in these tasks (7 in Task 1 on IE and 4 in Task 2 on IR). Herein, we describe the resources created for these tasks and the evaluation methodology adopted, and we provide a brief summary of the participants of this year's challenges as well as the results obtained. As in previous years, the organizers have made data, tools, and more specific overview papers associated with the lab tasks available for future research and development.

Motivated by the ever increasing difficulties faced by laypeople in retrieving and digesting valid and relevant information to make health-centred decisions, the CLEF eHealth lab series has offered shared tasks to the community in the fields of Information Extraction (IE), management, and Information Retrieval (IR) since 2013. These tasks have attracted large participation and led to statistically significant improvements in processing quality. In 2021, CLEF eHealth is calling for participants to contribute to the following two tasks: Task 1 on IE focuses on IE from noisy text. Participants will identify and classify Named Entities in written ultrasonography reports, containing misspellings and inconsistencies, from a major public hospital in Argentina. Identified entities will then have to be classified, which can be very challenging as it requires to handle lexical variations. Task 2 is a novel extension of the most popular and established task on consumer health search (CHS), aiming at retrieving relevant, understandable, and credible information for patients and their next-of-kins. In this paper we describe recent advances in the fields of IE and IR, and the subsequent offerings of this years CLEF eHealth lab challenges.

This paper provides an overview of SpRadIE, the Multilingual Information Extraction Task of CLEF eHealth 2021 evaluation lab. The challenge targets information extraction from Spanish radiology reports, and aims at providing a standard evaluation framework to contribute to the advancement in the field of clinical natural language processing in Spanish. Overall seven different teams participated, trying to detect seven named entities and hedge cues. Information extraction from radiology reports has particular challenges, such as domain specific language, telegraphic style, abundance of non-standard abbreviations and a large number of discontinuous, as well as overlapping entities. Participants addressed these challenges using a variety of different classifiers and introduced multiple solutions. The most successful approaches rely on multiple neural classifiers in order to deal with overlapping entities. As a result of the challenge, a manually annotated dataset of radiology reports in Spanish has been made available. To our knowledge this is the first public challenge for named entity recognition and hedge cue detection for radiology reports in Spanish.

In this paper we present an annotated corpus which can be used for training and testing algorithms to automatically extract information about diseases outbreaks from news and health reports. We also propose initial approaches to extract information from it. The corpus has been constructed with two main tasks in mind. The first one, to extract entities about outbreaks such as disease, host, location among others. The second one, to retrieve relations among entities, for instance, in such geographic location fifteen cases of a given disease were reported. Overall, our goal is to offer resources and tools to perform an automated analysis so as to support early detection of disease outbreaks and therefore diminish their spreading.

Named Entity Recognition in the clinical domain and in languages different from English has the difficulty of the absence of complete dictionaries, the informality of texts, the polysemy of terms, the lack of accordance in the boundaries of an entity, the scarcity of corpora and of other resources available. We present a Named Entity Recognition method for poorly resourced languages. The method was tested with Spanish radiology reports and compared with a conditional random fields system.

The objective of this study is to develop a methodology for the automatic detection of negated findings in radiological reports which takes into account semantic and syntactic descriptions, as well as morphological and syntactic analysis rules. In order to achieve this goal, a series of rules for processing lexical and syntactic information was elaborated. This required development of an electronic dictionary of medical terminology and computerized grammar. Computational framework was carried out with NooJ, a free software developed by Silberztein, which has various utilities for treating natural language. Results show that the detection of negated findings improves if lexical-grammatical information is added.

We present a methodology for the automatic recognition of negated findings in radiological reports considering morphological, syntactic, and semantic information. In order to achieve this goal, a series of rules for processing lexical and syntactic information was elaborated. This required development of an electronic dictionary of medical terminology and informatics grammars. Pertinent information for the assembly of the specialized dictionary was extracted from the ontology SNOMED CT and a medical dictionary (RANM, 2012). Likewise, a general language dictionary was also included. Lexicon-Grammar (LG), proposed by Gross (1975; Cahiers de l'institut de linguistique de Louvain, 24. 23-41 1998), was used to set up the database, which allowed an exhaustive description of the argument structure of predicates projected by lexical units. Computational framework was carried out with NooJ, a free software developed by Silberztein (Silberztein and Noo 2018, 2016), which has various utilities for treating natural language, such as morphological and syntactic grammar, as well as dictionaries. This methodology was compared with a Spanish version of NegEx (Chapman et al. Journal of Biomedical Informatics, 34(5):301-310 2001; Stricker 2016). Results show that there are minimal differences in favor of the algorithm developed using NooJ, but the quality and specificity of the data improves if lexical-grammatical information is added.

Radiology reports express the results of a radiology study and contain information about anatomical entities, findings, measures and impressions of the medical doctor.The use of information extraction techniques can help physicians to access this information in order to understand data and to infer further knowledge.Supervised machine learning methods are very popular to address information extraction, but are usually domain and language dependent.To train new classification models, annotated data is required.Moreover, annotated data is also required as an evaluation resource of information extraction algorithms.However, one major drawback of processing clinical data is the low availability of annotated datasets.For this reason we performed a manual annotation of radiology reports written in Spanish.This paper presents the corpus, the annotation schema, the annotation guidelines and further insight of the data.

A semantic tagger aiming to detect relevant entities in Arabic medical documents and tagging them with their appropriate semantic class is presented. The system takes profit of a Multilingual Framework covering four languages (Arabic, English, French, and Spanish), in a way that resources available for each language can be used to improve the results of the others, this is specially important for less resourced languages as Arabic. The approach has been evaluated against Wikipedia pages of the four languages belonging to the medical domain. The core of the system is the definition of a base tagset consisting of the three most represented classes in SNOMED-CT taxonomy and the learning of a binary classifier for each semantic category in the tagset and each language, using a distant learning approach over three widely used knowledge resources, namely Wikipedia, Dbpedia, and SNOMED-CT.

This paper presents a new annotated corpus of 513 anonymized radiology reports written in Spanish. Reports were manually annotated with entities, negation and uncertainty terms and relations. The corpus was conceived as an evaluation resource for named entity recognition and relation extraction algorithms, and as input for the use of supervised methods. Biomedical annotated resources are scarce due to confidentiality issues and associated costs. This work provides some guidelines that could help other researchers to undertake similar tasks.