Depressive symptomatology may be reflected in the language used by possible depressive profiles (PDP). This paper investigates to what extent symptoms of depression are manifested in Brazilian Portuguese narrative texts, and whether these can be used to identify relevant linguistic clues related to PDP. Moreover, the relation between these symptoms and PDP is explored, characterising the lexical, syntactic, and psycholinguistic aspects of texts produced by PDP. We found that texts associated with PDPs differed in some of these characteristics from non-PDP texts. The interactions between symptoms and PDP can also shed light on patterns of communication differentiation and the relationship between them. The results of this paper can help to characterise and understand the indicators that can be used to train more bespoke and accurate large language models.

Evaluating open-ended text generation in large language models remains challenging, particularly for non-English languages. We introduce EduBench, a comprehensive Portuguese-language benchmark comprising 3,149 discursive questions from Brazilian university entrance examinations spanning 2015–2025. Unlike multiple-choice or extractive QA benchmarks, EduBench requires extended, argumentative responses across diverse domains, including Humanities, Exact and Natural Sciences, and Languages. Each question includes expert-curated reference answers from official sources, rich metadata, and automated image descriptions to support text-only evaluation. We establish baseline results using nine contemporary models, ranging from 4B-parameter SLMs to state-of-the-art reasoning-capable LLMs, and evaluate them using complementary metrics (BLEU, BERTScore, G-Eval). Our results reveal substantial metric disagreement and highlight the complexity of assessing discursive generation, with models achieving 54–71% alignment with expert answers. We release EduBench publicly to support research on Portuguese NLP and open-ended generation evaluation.

Metaphor is a pervasive phenomenon in language that shapes how people conceptualize and communicate complex ideas. Detecting and interpreting metaphor is not only relevant for linguistic theory but also for many Natural Language Processing (NLP) applications, from machine translation to sentiment analysis, to mention a few. Despite its relevance, no open-source annotated corpus of metaphors exists for one of the world’s most widely spoken languages: Brazilian Portuguese. This paper addresses this gap by presenting an extension of Meta4XNLI, Meta4XNLI-ptBR, with token-level metaphor annotation in Brazilian Portuguese. To achieve this, we propose a pipeline that combines automatic translation via language models with human annotation, following guidelines adapted from MIPVU and Meta4XNLI. The final corpus contains 1,784 human-annotated sentences, of which 42.26% contain at least one metaphorical token. To our knowledge, this is the first open corpus of its kind for Brazilian Portuguese, and it is already freely available.

Figurative language, including multiword expressions and metaphors, provides a sensitive lens on cognitive functioning but remains largely overlooked in computational studies of Alzheimer’s Disease (AD). This work investigates figurative-language patterns in AD and whether they can help in distinguishing AD from non-clinical discourse and whether a neural model encodes comparable linguistic tendencies. We propose a two-step framework that combines relevant linguistic features with neural representations. Figurative expressions are automatically identified using Large Language Models focusing on idiomaticity and metaphor detection. These figurative language indicators are integrated with lexical, syntactic, and readability features and used to train classifiers on the ADReSS dataset. Correlation and proxy-model analyses reveal significant alignment between linguistic indicators and model predictions: participants with AD produce fewer figurative constructions, lower lexical diversity, and more concrete language. The results obtained demonstrate that contextual embeddings implicitly encode linguistic cues associated with cognitive decline and highlight the value of figurative-language metrics for transparent and linguistically grounded clinical NLP.

Electoral debates are influential moments in public discourse, providing candidates with a high-visibility platform to present their proposals, contrast their positions, and engage in exchanges that shape voter decisions. In Brazil, these debates reach a broad and diverse audience, reflecting regional, social, and ideological variations that affect linguistic choices and thematic content. This paper presents CoDEl-BR (Corpus de Debates Eleitorais, in Portuguese), a corpus of transcripts from 22 second-round mayoral debates held in 13 Brazilian state capitals during the 2024 municipal elections. It comprises 2,943 transcript segments totaling approximately 32 hours. Exploratory analyses reveal differences in thematic priorities between candidates and voters’ questions, as well as variations by race and party affiliation. The corpus aims to enable research in discourse and argumentation analysis, stance and sentiment detection, polarization modeling, and other related NLP tasks. We demonstrate that this initial release provides a curated, high-quality subset of debates with significant potential for expansion.

Natural Language Processing has already proven to be an effective tool for helping in the identification of mental health disorders in text. However, most studies limit themselves to a binary classification setup or base their label set on pre-established resources. By doing so, they don’t explicitly model many common ways users can express their depression online, limiting our understanding of what kind of depression signs such models can accurately classify. This study evaluates how machine learning techniques deal with the classification of a fine-grained set of 21 depression signs in social media posts from Brazilian undergraduate students. We found out that model performance is not necessarily driven by a depression sign’s frequency on social media posts, since evaluated machine learning techniques struggle to classify the majority of signs of depression typically present in posts. Thus, model performance seems to be more related to the inherent difficulty of identifying a given sign than with its occurrence frequency.

Pre-trained language models have achieved remarkable results on several NLP tasks. Most of them adopt masked language modeling to learn representations by randomly masking tokens and predicting them based on their context. However, this random selection of tokens to be masked is inefficient to learn some language patterns as it may not consider linguistic information that can be helpful for many NLP tasks, such as multimodal machine translation (MMT). Hence, we propose three novel masking strategies for cross-lingual visual pre-training - more informed visual masking, more informed textual masking, and more informed visual and textual masking - each one focusing on learning different linguistic patterns. We apply them to Vision Translation Language Modelling for video subtitles (Sato et al., 2022) and conduct extensive experiments on the Portuguese-English MMT task. The results show that our masking approaches yield significant improvements over the original random masking strategy for downstream MMT performance. Our models outperform the MMT baseline and we achieve state-of-the-art accuracy (52.70 in terms of BLEU score) on the How2 dataset, indicating that more informed masking helps in acquiring an understanding of specific language structures and has great potential for language understanding.

Humans constantly deal with multimodal information, that is, data from different modalities, such as texts and images. In order for machines to process information similarly to humans, they must be able to process multimodal data and understand the joint relationship between these modalities. This paper describes the work performed on the VTLM (Visual Translation Language Modelling) framework from (Caglayan et al., 2021) to test its generalization ability for other language pairs and corpora. We use the multimodal and multilingual corpus How2 (Sanabria et al., 2018) in three parallel streams with aligned English-Portuguese-Visual information to investigate the effectiveness of the model for this new language pair and in more complex scenarios, where the sentence associated with each image is not a simple description of it. Our experiments on the Portuguese-English multimodal translation task using the How2 dataset demonstrate the efficacy of cross-lingual visual pretraining. We achieved a BLEU score of 51.8 and a METEOR score of 78.0 on the test set, outperforming the MMT baseline by about 14 BLEU and 14 METEOR. The good BLEU and METEOR values obtained for this new language pair, regarding the original English-German VTLM, establish the suitability of the model to other languages.

Machine Translation (MT) is one of the most important natural language processing applications. Independently of the applied MT approach, a MT system automatically generates an equivalent version (in some target language) of an input sentence (in some source language). Recently, a new MT approach has been proposed: neural machine translation (NMT). NMT systems have already outperformed traditional phrase-based statistical machine translation (PBSMT) systems for some pairs of languages. However, any MT approach outputs errors. In this work we present a comparative study of MT errors generated by a NMT system and a PBSMT system trained on the same English – Brazilian Portuguese parallel corpus. This is the first study of this kind involving NMT for Brazilian Portuguese. Furthermore, the analyses and conclusions presented here point out the specific problems of NMT outputs in relation to PBSMT ones and also give lots of insights into how to implement automatic post-editing for a NMT system. Finally, the corpora annotated with MT errors generated by both PBSMT and NMT systems are also available.

We propose a method for joint unsupervised discovery of multiword expressions (MWEs) and their translations from parallel corpora. First, we apply independent monolingual MWE extraction in source and target languages simultaneously. Then, we calculate translation probability, association score and distributional similarity of co-occurring pairs. Finally, we rank all translations of a given MWE using a linear combination of these features. Preliminary experiments on light verb constructions show promising results.

Nowadays we are facing a growing demand for semantic knowledge in computational applications, particularly in Natural Language Processing (NLP). However, there aren’t sufficient human resources to produce that knowledge at the same rate of its demand. Considering the Portuguese language, which has few resources in the semantic area, the situation is even more alarming. Aiming to solve that problem, this work investigates how some semantic relations can be automatically extracted from Portuguese texts. The two main approaches investigated here are based on (i) textual patterns and (ii) machine learning algorithms. Thus, this work investigates how and to which extent these two approaches can be applied to the automatic extraction of seven binary semantic relations (is-a, part-of, location-of, effect-of, property-of, made-of and used-for) in Portuguese texts. The results indicate that machine learning, in particular Support Vector Machines, is a promising technique for the task, although textual patterns presented better results for the used-for relation.