With the volume of biomedical literature currently increasing at an unparalleled rate, researchers in biomedical sciences are quickly losing the capacity to single-handedly grasp the available information in a diversity of domains. However, owing to the growing body of available texts and open-access policies of an increasing number of publishers, Information Extraction (IE) in the biomedical domain is becoming an important task to aid in the condensation and systematization of scientific information in subject-specific areas and ideally reduce the burden of manual literature research in the biomedical scope. The present manuscript aims to provide a comprehensive survey on the tasks of Biomedical Named Entity Recognition (BioNER) and Biomedical Relation Extraction (BioRE), namely in terms of their evolution, current common methodologies, existing resources, and the challenges and opportunities they provide. Finally, a few possible routes for the development of Biomedical IE are also discussed.

Multimodal models, namely vision-language models, present unique possibilities through the seamless integration of different information mediums for data generation. These models mostly act as a black-box, making them lack transparency and explicability. Reliable results require accountable and trustworthy Artificial Intelligence (AI), namely when in use for critical tasks, such as the automatic generation of medical imaging reports for healthcare diagnosis. By exploring stress-testing techniques, multimodal generative models can become more transparent by disclosing their shortcomings, further supporting their responsible usage in the medical field.

Optical Character Recognition (OCR) systems are designed to extract text from images. While typically optimised for modern documents, they often struggle when applied to historical documents due to older fonts, complex layouts, and physical degradation, which can result in noisy outputs. To reduce OCR errors, post-OCR algorithms are commonly used, however, their development and evaluation require image-transcription pairs. Compared to other European languages, there is a lack of transcribed documents for historical Portuguese, especially for texts predating the 19th century. To address this gap, we introduce Portuguese post-OCR Resources for Text Optimisation (PORTO), a dataset that spans from the 17th to the 20th centuries. PORTO contains 3,782 image-transcription pairs, along with OCR outputs from four different systems, providing a valuable resource for the development and evaluation of OCR and post-OCR methods tailored to historical Portuguese.

This research investigates how to improve machine translation systems for low-resource languages by integrating loanword constraints as external linguistic knowledge.Focusing on the Portuguese-Emakhuwa language pair, which exhibits significant lexical borrowing, we address the challenge of effectively adapting loanwords during the translation process.To tackle this, we propose a novel approach that augments source sentences with loanword constraints, explicitly linking source-language loanwords to their target-language equivalents.Then, we perform supervised fine-tuning on multilingual neural machine translation models and multiple Large Language Models of different sizes.Our results demonstrate that incorporating loanword constraints leads to significant improvements in translation quality as well as in handling loanword adaptation correctly in target languages, as measured by different machine translation metrics.This approach offers a promising direction for improving machine translation performance in low-resource settings characterized by frequent lexical borrowing.

Senyu Li, Jiayi Wang, Felermino D. M. A. Ali, Colin Cherry, Daniel Deutsch, Eleftheria Briakou, Rui Sousa-Silva, Henrique Lopes Cardoso, Pontus Stenetorp, David Ifeoluwa Adelani. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

The accurate identification of loanwords within a given text holds significant potential as a valuable tool for addressing data augmentation and mitigating data sparsity issues. Such identification can improve the performance of various natural language processing tasks, particularly in the context of low-resource languages that lack standardized spelling conventions.This research proposes a supervised method to identify loanwords in Emakhuwa, borrowed from Portuguese. Our methodology encompasses a two-fold approach. Firstly, we employ traditional machine learning algorithms incorporating handcrafted features, including language-specific and similarity-based features. We build upon prior studies to extract similarity features and propose utilizing two external resources: a Sequence-to-Sequence model and a dictionary. This innovative approach allows us to identify loanwords solely by analyzing the target word without prior knowledge about its donor counterpart. Furthermore, we fine-tune the pre-trained CANINE model for the downstream task of loanword detection, which culminates in the impressive achievement of the F1-score of 93%. To the best of our knowledge, this study is the first of its kind focusing on Emakhuwa, and the preliminary results are promising as they pave the way to further advancements.

Question Generation aims to automatically generate questions based on a given input provided as context. A controllable question generation scheme focuses on generating questions with specific attributes, allowing better control. In this study, we propose a few-shot prompting strategy for controlling the generation of question-answer pairs from childrens narrative texts. We aim to control two attributes: the questions explicitness and underlying narrative elements. With empirical evaluation, we show the effectiveness of controlling the generation process by employing few-shot prompting side by side with a reference model. Our experiments highlight instances where the few-shot strategy surpasses the reference model, particularly in scenarios such as semantic closeness evaluation and the diversity and coherency of question-answer pairs. However, these improvements are not always statistically significant. The code is publicly available at github.com/bernardoleite/few-shot-prompting-qg-control.

This paper introduces a comprehensive collection of NLP resources for Emakhuwa, Mozambique's most widely spoken language.The resources include the first manually translated news bitext corpus between Portuguese and Emakhuwa, news topic classification datasets, and monolingual data.We detail the process and challenges of acquiring this data and present benchmark results for machine translation and news topic classification tasks.Our evaluation examines the impact of different data types-originally clean text, postcorrected OCR, and back-translated data-and the effects of fine-tuning from pre-trained models, including those focused on African languages.Our benchmarks demonstrate good performance in news topic classification and promising results in machine translation.We fine-tuned multilingual encoder-decoder models using real and synthetic data and evaluated them on our test set and the FLORES evaluation sets.The results highlight the importance of incorporating more data and potential for future improvements.All models, code, and datasets are available in the https://huggingface.

The task of Question Generation (QG) has attracted the interest of the natural language processing community in recent years. QG aims to automatically generate well-formed questions from an input (e.g., text), which can be especially relevant for computer-supported educational platforms. Recent work relies on large-scale question-answering (QA) datasets (in English) to train and build the QG systems. However, large-scale quality QA datasets are not widely available for lower-resourced languages. In this respect, this research addresses the task of QG in a lower-resourced language Portuguese using a traditional rule-based approach for generating wh-questions. We perform a feasibility analysis of the approach through a comprehensive evaluation supported by two studies: (1) comparing the similarity between machine-generated and human-authored questions using automatic metrics, and (2) comparing the perceived quality of machine-generated questions to those elaborated by humans. Although the results show that rule-based generated questions fall short in quality compared to those authored by humans, they also suggest that a rule-based approach remains a feasible alternative to neural-based techniques when these are not viable. The code is publicly available at https://github.com/bernardoleite/question-generation-portuguese.

Abstract The study of argumentation is transversal to several research domains, from philosophy to linguistics, from the law to computer science and artificial intelligence. In discourse analysis, several distinct models have been proposed to harness argumentation, each with a different focus or aim. To analyze the use of argumentation in natural language, several corpora annotation efforts have been carried out, with a more or less explicit grounding on one of such theoretical argumentation models. In fact, given the recent growing interest in argument mining applications, argument-annotated corpora are crucial to train machine learning models in a supervised way. However, the proliferation of such corpora has led to a wide disparity in the granularity of the argument annotations employed. In this paper, we review the most relevant theoretical argumentation models, after which we survey argument annotation projects closely following those theoretical models. We also highlight the main simplifications that are often introduced in practice. Furthermore, we glimpse other annotation efforts that are not so theoretically grounded but instead follow a shallower approach. It turns out that most argument annotation projects make their own assumptions and simplifications, both in terms of the textual genre they focus on and in terms of adapting the adopted theoretical argumentation model for their own agenda. Issues of compatibility among argument-annotated corpora are discussed by looking at the problem from a syntactical, semantic, and practical perspective. Finally, we discuss current and prospective applications of models that take advantage of argument-annotated corpora.

The Rapid development and the emergence of technologies capable of producing real-time data opened new horizons to both planning and optimization of vehicle routes [4]. In this dissertation, the Autoridade de Segurança Alimentar e Económica (ASAE) operation's scenario will be explored and analyzed as a case study to the problem. ASAE is a Portuguese administrative authority specialized in food security and economic auditing and is responsible to regulate thousands of economic entities in the Portuguese territory. ASAE inspections are usually done by brigades using vehicles to inspect economic operators, taking into account their timetables. Previous work on this topic led to the implementation of an inspection route optimization module capable of defining and assigning routes to inspect economic operators, seeking to maximize a utility function. Using optimization algorithms, inspection routes are calculated for each brigade, with information regarding specific map paths and inspection schedules. The approach used does not take into consideration the dynamic properties of real-life scenarios, as the precalculated operation plan is not reviewed in real-time. This work aims to study the dynamic properties of ASAE's operational environment and proposes a solution to efficiently review the precalculated inspection routes and apply the required changes in an appropriate time frame. Vehicle routing problems (VRP) are optimization problems where the aim is to calculate the set of optimized routes for a vehicle fleet, from a starting point to several interesting locations. Dynamic vehicle routing problem (DVRP) is a variant of VRP that makes use of real-time information to calculate the most optimized set of routes at a certain moment [39]. DVRP is a challenging problem because its scope is real-time, meaning that decisions sometimes must be made in short time windows, preventing the use of complex algorithms that require long computational times [10]. The typical approach to this problem is to initially calculate the routes for the whole fleet and dynamically revise the defined operations plan in real-time, once a disruption occurs. This work will model the problem as a DVRP and will compare the performance of heuristics and other modern optimization techniques, proposing a solution that will reduce the impact of disruptions on inspection routes. An optimized operations plan will reduce the time required for inspections, allowing massive economic savings, while reducing a company's ecological footstep. The work can eventually be scaled and used in other institutions, such as GNR or PSP in Portugal, that operate similarly.

Interest in argument mining has resulted in an increasing number of argument annotated corpora. However, most focus on English texts with explicit argumentative discourse markers, such as persuasive essays or legal documents. Conversely, we report on the first extensive and consolidated Portuguese argument annotation project focused on opinion articles. We briefly describe the annotation guidelines based on a multi-layered process and analyze the manual annotations produced, highlighting the main challenges of this textual genre. We then conduct a comprehensive inter-annotator agreement analysis, including argumentative discourse units, their classes and relations, and resulting graphs. This analysis reveals that each of these aspects tackles very different kinds of challenges. We observe differences in annotator profiles, motivating our aim of producing a non-aggregated corpus containing the insights of every annotator. We note that the interpretation and identification of token-level arguments is challenging; nevertheless, tasks that focus on higher-level components of the argument structure can obtain considerable agreement. We lay down perspectives on corpus usage, exploiting its multi-faceted nature.

To overcome the lack of annotated resources in less-resourced languages, unsupervised language adaptation methods have been explored.Based on multilingual word embeddings, Adversarial Training has been successfully employed in a variety of tasks and languages.With recent neural language models, empirical analysis on the task of natural language inference suggests that more challenging auxiliary tasks for Adversarial Training should be formulated to further improve language adaptation.We propose rethinking such auxiliary tasks for language adaptation.

Art. 15, 15 S.

The use of AI in recruitment is growing and there is AI software that reads jobs' descriptions in order to select the best candidates for these jobs. However, it is not uncommon for these descriptions to contain inconsistencies such as contradictions and ambiguities, which confuses job candidates and fools the AI algorithm. In this paper, we present a model based on natural language processing (NLP), machine learning (ML), and rules to detect these inconsistencies in the description of language requirements and to alert the recruiter to them, before the job posting is published. We show that the use of an hybrid model based on ML techniques and a set of domain-specific rules to extract the language details from sentences achieves high performance in the detection of inconsistencies.

Abstract Knowledge Engineering has become essential in the fields of Medical and Health Care with emphasis for helping citizens to improve their health and quality of life. This includes individual methods and techniques in health‐related knowledge acquisition and representation and their application in the construction of intelligent systems capable of using the acquired information to improve the patients' health and/or quality of life. Haemato‐oncological diseases can provide significant disability and suffering, with severe symptoms and psychological distress. They can create difficulties in fulfilling professional, family and social roles, affecting an individual's quality of life. Health related quality of life (HRQoL) is a subjective concept but there is also an objective component related to physiological indicators. Some of these physiological indicators can be easily assessed by wearable technology such heart rate variability (HRV). This paper introduces an intelligent system to assess, in real‐time, potential HRV indices, that can predict HRQoL in lymphoma patients throughout chemotherapy treatment and to account the individuals' variability. The system is based on wearable technology and intelligent processing of the patients' biometric information to assess some quality of life related parameters. A longitudinal study was conducted among 16 lymphoma patients using this intelligent system. Mixed‐effect regression models were performed to investigate predictors for and time effects on HRQoL. There were no significant changes in all HRQoL domains over time. Some quality of life domains revealed similar time trends as HRV indices. These HRV indices also have a significant effect on the domains of quality of life.

Artificial Intelligence (AI) has seen numerous applications in the area of Education. Through the use of educational technologies such as Intelligent Tutoring Systems (ITS), learning possibilities have increased significantly. One of the main challenges for the widespread use of ITS is the ability to automatically generate questions. Bearing in mind that the act of questioning has been shown to improve the students learning outcomes, Automatic Question Generation (AQG) has proven to be one of the most important applications for optimizing this process. We present a tool for generating factual questions in Portuguese by proposing three distinct approaches. The first one performs a syntax-based analysis of a given text by using the information obtained from Part-of-speech tagging (PoS) and Named Entity Recognition (NER). The second approach carries out a semantic analysis of the sentences, through Semantic Role Labeling (SRL). The last method extracts the inherent dependencies within sentences using Dependency Parsing. All of these methods are possible thanks to Natural Language Processing (NLP) techniques. For evaluation, we have elaborated a pilot test that was answered by Portuguese teachers. The results verify the potential of these different approaches, opening up the possibility to use them in a teaching environment.

Esta dissertação tem como objetivo o desenvolvimento de uma arquitetura genérica de alto nível para o desenvolvimento de agentes capazes de eficientemente jogar jogos com um misto de competição e cooperação. Técnicas tradicionais utilizadas no contexto dos jogos incluem estratégias de pesquisa como o Branch & Bound assim como abordagens de Monte-Carlo. Contudo estas técnicas são difíceis de aplicar a esta categoria de jogos, devido aos frequentemente grandes espaços de pesquisa e à dificuldade em calcular os valores das posições e movimentos dos jogadores. Neste trabalho propômos uma arquitetura de agentes genérica que aborda os temas da negociação, confiança e modelação de oponentes, simplificando o desenvolvimento de agentes capazes de jogar estes jogos eficientemente. Esta arquitetura está dividida em quatro módulos independentes, inspirando-se na estrutura de uma nação em tempo de guerra, o Presidente, o Departamento Estratégico, o Departamento de Relações Externas e o Departamento de Inteligência. Demonstramos as aplicações desta arquitetura instanciando-a usando dois jogos diferentes, o Diplomacy e o Werewolves of Miller's Hollow, e testando os agentes obtidos numa variedade de cenários contra agentes existentes. Os resultados obtidos mostram que a arquitetura é genérica o suficiente para ser aplicada numa grande variedade de jogos, e a inclusão de negociação, confiança e modelação de oponentes permite obter agentes mais eficazes.

Normative environments are used to regulate multi-agent interactions, by providing means for monitoring and enforcing agents' compliance with their commitments. In business encounters, agents representing business entities make contracts including no

Normative environments have been proposed to regulate agent interaction in open multi-agent systems. However, most approaches rely on pre-imposed regulations that agents are subject to. Taking a different stance, we focus on a normative framework that assists agents in establishing by themselves their own commitment norms. With that aim in mind, a model of norm defeasibility is presented that enables exploiting and adapting a normative background to different extents. We formalize the normative state using first-order logic and define rules and norms operating on that state. A suitable semantics regarding the use of norms within a hierarchical context structure is given, based on norm activation conflict and defeasibility.

Norms and institutions have been proposed to regulate multi-agent interactions. However, agents are intrinsically autonomous, and may thus decide whether to comply with norms. On the other hand, besides institutional norms, agents may adopt new norms by establishing commitments with other agents. In this paper, we address these issues by considering an electronic institution that monitors the compliance to norms in an evolving normative framework: norms are used both to regulate an existing environment and to define contracts that make agents' commitments explicit. In particular, we consider the creation of virtual organizations in which agents commit to certain cooperation efforts regulated by appropriate norms. The supervision of norm fulfillment is based on the notion of institutional reality, which is constructed by assigning powers to agents enacting institutional roles. Constitutive rules make a connection between the illocutions of those agents and institutional facts, certifying the occurrence of associated external transactions. Contract specification is based on conditional prescription of obligations. Contract monitoring relies on rules for detecting the fulfillment and violation of those obligations. The implementation of our normative institutional environment is supported by a rule-based inference engine.

Electronic Institutions are comprehensive frameworks that may effectively help in the collaborative work of virtual organization activities. This paper focuses on an effort to create e-contracting and ontology-based services in the context of Electronic Institutions. The e-contracting services provide automatic specification of business agreements by formalizing them through e-contracts, plus the procedures for enforcing them. Moreover, ontology-based services enable the interoperability between agents representing organizations using different ontologies. Ontology services provide useful advices on how to negotiate specific items, leading to appropriate conversations and making agreements possible. We believe that the rendering of these services will provide a level of trust and normative behavior allowing the creation, through electronic institutions, of dynamic virtual organizations and their operation. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.