Is There a Patent Genre? A Computational Text Analysis of French Patents, 1903–1940

Patents are legal documents that grant temporary exclusive rights over inventions and describe their technical features. They have become an important source for studying innovation and technological change, including through computational analyses of patent texts. Yet patent applications are not only repositories of technical knowledge, but also shaped by institutional and linguistic conventions that structure their writing. This paper approaches patents as a rhetorical genre by analysing a corpus of French patent descriptions from 1903–1940. Using computational text analysis, it examines both the stability and variation of patent language. First, we analyse the persistence of frequent bigrams over time to identify repeating lexical patterns. Second, we measure lexical differences between patents from different countries filed in France using Jensen–Shannon distance. The results show a stable core of word constructions in patent descriptions. We also observe differences between countries that show little systematic convergence over time. We find persistent closeness within linguistic regions, for instance between US and UK patents and among German-language or Romance-language countries. This points to the persistence of shared linguistic conventions and patent practices that survive translation into French. The findings highlight the importance of approaching patents as historically situated textual artifacts.