Deriving semantic classes of Italian adjectives via word embeddings: a large-scale investigation

This paper investigates the application of word embeddings to derive semantic classes for Italian adjectives. Adjectives were clustered using UMAP for dimensionality reduction and K-means for clustering. Semantic categories such as “Relational”, “Descriptive”, “Evaluative”, “Membership”, and “Physical/HealthRelated” were tested by employing predefined prototypical adjectives for each class. The precision and recall of the classification were analyzed, revealing high accuracy for some classes (e.g., “Evaluative”), but challenges in distinguishing more nuanced categories such as “Descriptive”. Furthermore, cluster overlaps were visualized using KDE and quantified using KNN, , highlighting semantic intermingling between groups, especially between the “Descriptive” and “Evaluative” categories. Finally, a comparison with Wordnet’s adjective categories was provided.