The aim of European semiotics is to provide a scientific foundation for the humanities. This objective is extremely hard to reach as meaning is a vast object, treated in very different ways by critics, philosophers and social scientists. To find regularities in meaning-making practices is therefore a task potentially capable of reorganizing a disperse galaxy of heterogeneous works. Today, digital methods are a way to approach human data via statistical and computational tools, and they seem to provide a new common foundation for the study of meaning in its rich variety (Rogers, Moretti, Manovich). The question is: can statistical tools alone provide a methodological ground for the humanities? Can the humanities simply become a specialized area in computer science and data mining? In our view, computational tools are an invaluable technical and conceptual instrument for the ends of semiotics, making analysis more controllable and objective, but at the same time they do not replace human interpretation. We will quickly present the results of an analysis of a Reddit corpus about medical topics. We inductively identified patterns of speech acts structuring how conversational exchanges are brought forward. For this analysis we realized an experimental annotation involving 12 subjects and applied a technique, called process mining, for the inductive identification of regularities in human activities (van der Aalst). Process mining starts from digital traces to find patterns; in our case patterns were conversational regularities of speech acts (Sacks, Schegloff, Austin, Searle, Sbisà). Our results allowed us to identify some preferential patterns in digital asynchronous conversations, and even to isolate one precise conversational preference that had never been observed before. In this talk we will compare our results with other similar ones realized with computational tools on textual corpora, so to highlight the heuristic potential and the limitations of these methodologies for semiotic analysis. Particularly, we will make reference to the concepts of “analysis” and “interpretation” as they have been defined by A. Greimas and U. Eco. In fact, statistical methods promise to allow for pure analysis and explanation, working on textual immanence without the need of interpretation. Still they remain anchored to human understanding, for guiding hypotheses, reading results, and so turning data into signs. One of the most important tasks for the epistemology of semiotics today is then to determine how to include statistical methods in interpretative practices without being unaware of the latter.