Genre induction from a linguistic approach

Abstract

In most current work on genre, a set of genre categories needs to be predetermined. However, there are some cases where such predetermined genres cannot be clearly identified. Popular science, for instance, is a broad register carrying several specific purposes within it, suggesting that there are several genres of popular science, but it is unclear what these genres are. This paper introduces a linguistic approach to reveal hidden genres. For 600 written popular science texts from a variety of sources and disciplines, linguistic features were analysed using a range of computer programs and a cluster analysis conducted. The analysis produced four clusters with shared linguistic features, representing text types. The association of these text types with key features, functional relations, dominant sources, and prototypical members of each cluster helps us to induce genres on the basis of communicative purposes, a traditional criterion in identifying genres. Whether the produced text types are equivalent to genres was evaluated with a test set of data. The proposed approach achieves more than 70 % accuracy. The approach appears applicable for identifying genres of popular science and has pedagogical implications.