From Close to Distant and Back: How to Read with the Help of Machines

Digital Humanities (DH) is a variegate field of study that combines a humongous number of different interactions between humanist disciplines and the use of the computer. In recent years researchers have noticed a common trend among these different approaches characterized by the adoption of quantitative methods in the study of digital sources [1]. In particular, due to Franco Moretti’s definition of “distant reading” as a group of text mining approaches proposed in opposition to traditional-hermeneutic analysis [2], the DH community is dividing itself in two opposite factions [3]. Central to this division is the idea that computational methods seem to move in the direction of making the work of the humanist irrelevant for the production of insights. Starting with these assumptions, the purpose of our paper is twofold: first, we intend to stress how text mining methods will always need a strong support from the humanist, and second, we argue about the usefulness and necessity of advanced text mining approaches in the DH. We would like to think of a humanist research involving computational techniques as a three steps process. The first step is a “close reading”, which includes selecting a specific case study, crafting the initial features, and labeling of the training corpus. The second step is a “distant reading” since it involves performing a computational analysis. The third step is another “close reading”, which consists of evaluation and interpretation of the results and the use of these results in a humanities research. At the same time, we think that researchers should not renounce text mining approaches, but should instead experiment with advanced methods such as the ones belonging to the family of deep learning [5]. Deep learning techniques essentially perform representation learning, and therefore permit to automatically analyze text as a multilayered set of encoded features.