The aim of this study is to analyse the corpus of end-of-year speeches of the Presidents of the Italian Republic from a diachronic perspective in order to identify groups of words that share the same pattern and depict main topic trends. The procedure adopted for the recognition of the dynamics of word frequencies moves from a statistical learning perspective and envisages decisions that concern the normalization of occurrences, the smoothing of the trajectories, and the curve clustering. In resulting clusters, emerging topics as well as those that have disappeared over years are clearly visible but, above all, the individual trait of the President stands out as the most relevant element that determines the contents of his discourses.
