Natural Language to Structured Query Language (NL2SQL) systems have advanced with large language models (LLMs), yet evaluations remain centered on English, leaving open how these systems behave across languages and language varieties. We aim to contribute to this by analyzing two state-of-the-art LLMs (LLaMA 3.3 70B and Qwen 3.6 35B, both zero-shot) on the WikiSQL benchmark in English, Brazilian Portuguese (PT-BR), and European Portuguese (PT-PT), keeping database schemas and inference configuration fixed. Translations were validated through distance metrics (cosine similarity 0.966), contrastive linguistic analysis confirming varietal features (e.g., cleft constructions 36.8× more frequent in PT-PT), and multi-model judgment (85% / 81% semantic equivalence, 95.5% PT-BR and 96.0% PT-PT inter-judge agreement). Results show a marked drop from English (47.25% EX) to Portuguese (PT-BR: 40.03%; PT-PT: 39.34%), with the PT-BR vs. PT-PT difference reaching statistical significance (p = 0.019). English-only successes outnumber Portuguese-only successes nearly 10:1. A replication on a second model (Qwen 3.6 35B) confirms the English-Portuguese gap but suggests that intralinguistic effects may be model-dependent. Stratified analyses show semantic complexity and quantification amplify degradation, particularly for COUNT queries. These findings point to persistent English bias in NL2SQL and measurable effects of intralinguistic variation, even between closely related varieties.

This paper presents EpiCorpus-BR, the first annotated corpus for Named Entity Recognition (NER) in Brazilian Portuguese epidemiological surveillance documents, covering the annual SIREVA-SUS reports (2013–2024) and complementary documents from the Brazilian Ministry of Health. The corpus comprises 8,314 textual units across 22 documents (6,358 narrative text sentences and 1,956 table rows), with 4,159 entity mentions in the silver set. The tagset consists of eight specialized categories (PATOGENO, SOROTIPO, FAIXA_ETARIA, LOCAL, PERIODO_TEMPORAL, METODO_LAB, METRICA_EPI, MANIFESTACAO_CLINICA), annotated via few-shot prompting with GPT-4o-mini and manually reviewed in a stratified sample of 280 units (180 text sentences and 100 table rows) by two independent annotators (κ = 0.76). Strict IOB2 evaluation with seqeval yields a combined micro-F1 of 0.77. METRICA_EPI is absent from narrative text but reaches F1 = 0.89 on table rows, which shows why the tabular sub-corpus must be included in the evaluation. A BERTimbau baseline fine-tuned on the silver corpus, with the gold held out, reaches micro-F1 = 0.73, showing that the corpus supports training a dedicated NER model. The corpus and code are publicly available.

We present an ongoing research project focused on the construction of a Universal Dependencies (UD) corpus of Portuguese epidemiological reports derived from documents published within the Brazilian public health system. We describe findings and challenges to build such a corpus from PDF reports processed through a controlled document extraction pipeline that contrasts layout-aware extraction with raw PDF text extraction, explicitly addressing the impact of tabular content on downstream syntactic analysis. Narrative text is annotated using multiple UD parsers for Portuguese, including widely used and state-of-the-art tools, and their outputs are systematically compared using descriptive structural indicators and targeted qualitative inspection. Our analysis highlights domain-specific challenges in epidemiological texts and shows that document extraction and representation choices have a stronger effect on parsing behavior than parser selection alone. Based on these findings, we identify robust preprocessing configurations and discuss design choices for a UD-epidemiological corpus to support future research on syntactic parsing, domain adaptation, and downstream natural language processing tasks in epidemiology and public health.

Although large language models have transformed natural language processing, their computational costs create accessibility barriers for low-resource languages such as Brazilian Portuguese. This work presents a systematic evaluation of Parameter-Efficient Fine-Tuning (PEFT) and quantization techniques applied to BERTimbau for Question Answering on SQuAD-BR, the Brazilian Portuguese translation of SQuAD v1.We evaluate 40 configurations combining four PEFT methods (LoRA, DoRA, QLoRA, QDoRA) across two model sizes (Base: 110M and Large: 335M parameters). Our findings reveal three critical insights: (1) LoRA achieves 95.8% of baseline performance on BERTimbau-Large while reducing training time by 73.5% (F1 = 81.32 vs. 84.86); (2) higher learning rates (2e-4) substantially improve PEFT performance, with F1 gains of up to +19.71 points compared to standard rates; and (3) larger models show twice the quantization resilience (loss of 4.83 vs. 9.56 F1 points).These results demonstrate that encoder-based models can be efficiently fine-tuned for extractive Brazilian Portuguese question answering with substantially lower computational cost than large generative LLMs, promoting more sustainable approaches aligned with Green AI principles. An exploratory evaluation of Tucano and Sabiá on the same benchmark shows that although generative models can achieve competitive F1 scores with LoRA fine-tuning, they require up to 4.2 times more GPU memory and three times more training time than BERTimbau-Base, reinforcing the efficiency advantage of smaller encoder-based architectures for this task.

Automatic summarization of financial news in Portuguese lacks reliable reference-free evaluation metrics. While LLM-as-a-Judge approaches are gaining traction, their correlation with human perception in specialized domains remains under-explored. This work evaluates the efficacy of Question Answering (QA) based metrics against a direct LLM-as-a-Judge baseline for Portuguese financial news. We propose a pipeline comparing Lexical, Binary, and Semantic (LLM-based) QA scoring methods, validated against a human ground truth of 50 news items annotated for Faithfulness and Completeness. Our results show that granular QA metrics significantly outperform the monolithic LLM-Judge in evaluating Completeness, with QA-Binary achieving the highest rank correlation (ρ ≈ 0.49 with pessimistic human aggregation). For Faithfulness, we observe a strong ceiling effect in human evaluation, yet the Semantic QA metric demonstrated a “super-human” ability to detect subtle hallucinations (e.g., temporal shifts) missed by annotators. We conclude that decomposing evaluation into atomic QA pairs is superior to holistic judging for the Portuguese financial domain.

The rapid dissemination of digital information has exposed financial markets to the risks of disinformation. Although numerous methods exist to detect fake news, they predominantly focus on textual features, often neglecting the significant role of image-based content. This paper introduces a novel framework for detecting financial fake news in Brazilian Portuguese by bridging this gap. The proposed system integrates Natural Language Processing (NLP) with an image-to-text classification strategy: using a Tesseract-based OCR, the system extracts text from images and processes it using the unified pipeline used for text classification. Experiments on Fake.BR, FakeRecogna corpus and BBC News Brasil show that our approach achieves 98% accuracy using BERTimbau Fine Tuned on financial news. These findings underscore the critical importance of analyzing visual text and demonstrate the multimodal strategy is effective for disinformation detection.

This study analyzes texts from multiple sources, including social media and news portals, to observe how different sectors of Brazilian society discuss the antimicrobial resistance. The main goal is to support epidemiological surveillance and public policy decisions through computational tools. Three datasets were used: tweets collected between 2008 and 2025 (64,225 documents), news articles from G1 (4,363 documents), and official government publications (.gov.br, 1,515 documents). These sources enable comparative analysis between informal discourse (social media) and institutional or journalistic discourse (official and media outlets). The study applies and compares topic modeling techniques, particularly those designed for Short Text Topic Modeling (STTM), such as GSDMM and BERTopic, to identify discursive trends, semantic patterns, and emerging topics related to antimicrobial resistance. By exploring these distinct contexts, this work demonstrates the potential of Natural Language Processing (NLP) and AI methods as instruments for integrated analysis of public health data in both informal and formal environments.