Chachereau, NicolasSrinivasa Desikan, Bhargav2022-11-102022-11-102022-11-102022-11-04https://infoscience.epfl.ch/handle/20.500.14299/192177Patents have long been an obvious and important source for documenting technological change. For instance, economists and economic historians have counted patents granted in certain countries or certain industries, while historians of science and technology have selected and read some of them (e.g. to study the thinking process of inventors). In recent years, the digitization of historical patents has opened up new opportunities, allowing to ‘read’ at large scale, i.e. study many patents but still use their full textual content rather than only count them. Our own digital history project analyzes a very large corpus, all 1.3 million patents issued in the United States between 1836 and 1920, with two major aims: categorizing patents into coherent technical categories; identifying discourses of safety, reflexivity, and environmental concern in technological innovation. In our presentation, we will discuss the rationale for our research questions, describe our experiments with various methods of natural language processing, and our early results. We will then move to an open discussion on the specific dynamic of our interdisciplinary project. Indeed, our team brings together historians with technical skills and computer scientists with an interest or a background in the humanities and social sciences. As such, we are in an ideal position to reflect upon the interaction between the two research strands. In our project, digital methods have influenced the historical approach and made themes such as source criticism more prominent than in many other historical research projects. Conversely, the need for interpretability of the results, as well as the specific nature of the historical documents constrained the use of the digital methods.text::conference output::conference presentation