Phraseology at the service of automatic text classification: Computational account of binomials in the company registration discourse with translational implications

The paper problematises text classification potential of phraseology in the context of methodological research paradigms of digital humanities. Specifically, the analysis discussed in the paper aims at verifying whether four structural features of English legal binomials can construct a statistically reliable genre predictive model, whereby legal texts are automatically assigned to three groups of legal genres, processed through company registration, in a court-based procedure. The computational part of the analysis rests in generating a random decision forest. The candidate terms have been extracted from a corpus of authentic texts, referred to as company registration discourse. The quantitatively most marked patterns are subjected to qualitative analysis involving the examination of the linguistic context, the contextual metadata and the genre requirements. The analytical part is preceded by atheoretical account of the concept of digital humanities with reference to related notions and ahistorical overview of how this discipline evolved. The findings related to the predictive capacity of the model derived are discussed in the context of the translation output, the translation quality and the comprehensibility of legal texts. At this stage the discussion draws on the parallel component of the corpus. The author concludes that legal genres covered by the analysis are relatively schematic and the structural profile of binomials makes up part of some of these schemes. Automatic text classification is possible in some cases. The original generic structure of the texts covered by the analysis is distorted in the translation.