Data

The data for this study have been extracted from the bi-directional Dutch-English component^[1] of the Dutch Parallel Corpus or DPC (Macken et al. 2011; Paulussen et al. 2013), a sentence-aligned corpus of translated texts. The texts that constitute the DPC were published between 1997 and 2009, and belong to five text types: fictional and non-fictional literature, journalistic texts, instructive texts, administrative texts, and external communication (such as press releases, brochures and corporate magazines). The corpus has been balanced proportionally for translation direction as well as for text type, resulting in 500,000 words for each text type for each translation direction (e.g. Dutch-English and English-Dutch), which amounts to a corpus of 5 million words for the purposes of this study. All instances of so and dus were extracted automatically from the corpus (together with the aligned target-text sentences), and subsequently checked manually to remove any double entries and irrelevant tokens. The latter especially pertained to those instances where so is an adverb of manner or degree, a pro-form, part of a fixed expression, or a marker of purpose (see Sect. 1). As this is a bi-directional corpus, different angles can be looked at. It is not only possible to consider how so has been translated from English into Dutch (and dus from Dutch into English), but also backtranslations have been taken into account, which means that I have searched for so in English target texts and traced the correspondences for these tokens in the Dutch source texts (and vice versa for dus).

The main drawback of working with a corpus of translated texts is their inherent bias for written registers while pragmatic markers are rather more typical of spoken registers. As Johansson (2006) aptly points out, though, the target texts are the result of a thorough process of translation in which translators have independently interpreted source texts, and “[w]hat we are studying is the result of this interpretation (and recreation) process” (2006: 117). This can shed light on how a pragmatic marker functions in the source language, and how this function can be conveyed in the target language. Moreover, many texts in the DPC may have appeared in written mode, but were either meant to be spoken or otherwise reflect spoken language. Many administrative texts, for example, are transcripts from meetings at the European Parliament or official speeches by government ministers. The fiction component of the corpus in its turn contains many instances of direct speech (which has already proved a valuable resource in studies by Aijmer and Simon-Vandenbergen 2003; Johansson 2006; Degand 2009; Denturck and Vandepitte 2009), as do the journalistic texts, many of which are interviews or columns. It should also be noted that, contrary to markers with a predominant interpersonal function (such as well or you know), so and dus are highly frequent both in spoken and written language. Nonetheless, it should be borne in mind that (i) some functions of these pragmatic markers may not be attested in the present corpus because they are particular to specific spoken registers, and that (ii) the texts show a bias towards more formal registers.

• [1] The DPC also has a bidirectional Dutch-French component, but this was not included in the present study.