Identifying Text Genres Using Phrasal Verbs

Understanding the textual distinction between spokenness-informality and writtenness-formality serves many purposes. It can facilitate text mining, improve parser accuracy, offer better appraisals of student writing, and may also facilitate better interpretations of experimental data. Previous studies of such textual variation (e.g., Biber, 1988, Louwerse et al., 2004) have failed to produce a simple and effective method for computationally distinguishing these text types. Indeed, Biber (1988) using 67 lexical features could not determine any spoken/written dimension and Louwerse et al. (2004) using over 200 textual indices could not identify a formal/informal dimension. In this study, we tested the hypothesis that phrasal verbs could distinguish such text-types because the presence of this verb construction is often claimed to be indicative of both spoken and less formal discourse (e.g., McWhorter, 2001). To test our hypotheses, we used Coh-Metrix (Graesser et al., 2004) to calculate the incidences of various phrasal verbs forms across two corpora: the Biber Corpus (Biber, 1988; Louwerse et al., 2004); and a larger, yet structurally identical second corpus. For the spoken/written distinction, we used texts identified in the Louwerse et al. first dimension (LSWD). For the formal/informal distinction, we used tests identified in the Biber fifth dimension (BFID).