The study of the ideas of historical texts requires expertise identifying the context, the precise meaning of the statements and the establishment of their filiations. Purely statistical counts cannot consider the nuances induced by the style, the absence or the presence of a word which would change the meaning of a sentence, the context, etc.

This Exploratory Action Apollon aims to automate such a task by combining reinforcement learning approaches and human expertise to create the first lexicon of ideas from the corpus of Aristotle’s Politics, one of the founding works of political science, then to apply these methods to other corpora and to the automatic generation of texts.