Speech-language assessment of stuttering is traditionally manual, subjective, and time-consuming. This paper presents the development of software for automatic detection and classification of stuttering-related disfluencies in Brazilian Portuguese, aiming to support clinical assessment. The system follows a two-stage hybrid approach. In the first stage, it applies deterministic algorithms based on automatic speech recognition (ASR) and temporal information to identify simple disfluencies, such as repetitions and pauses. In the second stage, it employs a hierarchical architecture combining a Kohonen network (Self-Organizing Map, SOM) and a Multilayer Perceptron (MLP) to classify complex disfluencies, specifically blocks and prolongations, using acoustic features. Because no publicly available annotated resources exist for this task in Brazilian Portuguese, we built a initial dataset annotated by specialists. The system achieved 89.5% accuracy in classifying complex disfluencies, with a Matthews Correlation Coeficient (MCC) of 0.812. These results indicate the feasibility of the tool as decision support for clinical assessment and establish a baseline for future research.

Humor processing remains a complex challenge in Natural Language Processing, particularly the task of pun location, which involves identifying the specific ”pivot word” that creates linguistic ambiguity. This paper presents a novel two-stage approach for token-level pun location in Portuguese, addressing the scarcity of research in this language. The first stage uses an ensemble of traditional classifiers to filter out non-pun sentences, thereby reducing class imbalance. The second stage employs a pre-trained BERT encoder combined with a Mixture-of-Experts (MoE) layer to capture specialized linguistic features for token classification. We validate our approach on the Puntuguese corpus, achieving an F-score of 0.74 without requiring post-processing heuristics. Interpretability analyses demonstrate that the MoE experts learn to specialize in distinct mechanisms, such as punchline detection and morphological patterns, thereby confirming the model’s capacity to capture the nuances of humor.

The proliferation of fake news in digital environments poses serious challenges to democratic processes, particularly in morphologically rich languages such as Portuguese. While most existing approaches focus on stylistic cues or propagation patterns in social networks, this paper proposes an automated fake news verification methodology grounded in Knowledge Graphs (KGs). Instead of treating news as raw text, we represent each article as a set of factual events encoded as semantic triples of subject, predicate, and object. A proprietary knowledge graph is built from Brazilian data sources, and a verification algorithm is introduced to estimate the veracity of news articles based on graph connectivity evidence. Experimental results confirm the feasibility of the proposed approach and highlight its inherent explainability as a key advantage over deep learning black-box models. Error analysis further indicates that the main limitation stems from the syntactic complexity of Open Information Extraction in Portuguese, suggesting that improvements at this extraction stage are essential to increase system robustness.