Formality in mediated and non-mediated discourse: Bringing together human judgements and corpus-driven detection

Several works have suggested that both interpreting and translation tend to favour linguistic choices typically considered more formal. Observations concerning the degree of formality of these forms of language mediation, however, have been made within studies mostly focusing on different phenomena, such as standardization or conventionalization. In this paper, we report on a quantitative and qualitative analysis focusing specifically on formality in interpreted language. We take into account native and interpreted speeches delivered at the European Parliament (EP) and collected in the English subcorpora of the EPIC and EPTIC corpora. We compare the examined English varieties to one another, taking into account the mode of delivery of the speeches, i.e. whether they were read out from a written text or delivered impromptu. Our hypothesis is that interpretations of speeches read at the EP are located at the far end of the formality spectrum compared to speeches delivered impromptu by native English MEPs. Unlike previous work, we rely on formality indicators identified by triangulating human judgements and specific linguistic features derived bottom-up from a corpus, based on the MuPDAR[F] approach. The analysis provides partial support for the hypothesis, showing that interpreted texts are generally predicted as being characterised by a high level of formality, irrespective of the actual mode of delivery of their source text. We conclude by commenting on the linguistic features that were found to contribute the most to making interpreted speeches diverge from native ones.