Cohesion and explicitation in an English-German translation corpus

In translation studies, cohesive features as indicators for explicitation have been analysed either in an example-based way (Blum-Kulka 1986) or as concordances in monolingually comparable corpora of raw text (cf. several contributions in Laviosa (ed.) 1998, Olohan & Baker 2000). In spite of the insight gained from this line of research, we argue that where explicitation is investigated without taking into account the source texts, the interpretation of results remains restricted and problematic. Work on translations against a more linguistic background has addressed some of these restrictions and problems (cf. relevant work as in Johansson & Oksefjell (eds.) 1998, Fabricius-Hansen 1999); the focus of these research interests and methodologies is however different from, and partly complementary to, ours with respect to corpus architecture, querying techniques and underlying linguistic modelling (for which cf. Hansen 2003, Neumann 2003, Steiner 2001, 2005a,b,c, Teich 2003). The basic assumption for the analysis of explicitation in the present paper is that the element explicitated in the target text has to be present implicitly in a linguistically traceable way in the source text and vice versa for implicitated elements. Explicitation is thus defined as a relationship and a process between instantiated and aligned pieces of translated texts. Furthermore, we stratify the notion of explicitation according to the linguistic levels of lexicogrammar (not in focus in this paper) and cohesion. As this stratification is still too abstract to be directly quantifiable on linguistic data in an electronic corpus, a series of further micro-level operationalisations is undertaken which are meant to bring the relevant phenomena down to an empirically accessible level. Our investigation of explicitation and implicitation of cohesion markers in translations is based on a cross-linguistic corpus containing statistically meaningful and representative samples (cf. Biber 1993) of German and English registerially parallel texts from 8 registers annotated with parts of speech, morphology, phrase structure and grammatical functions. In addition to these two sub-corpora, two further sub-copora have been compiled consisting of translations of the samples from the first two sub-corpora into the respective other language, yielding 4 sub-corpora. The overall corpus comprises 1 million words (approx. 250 000 for each of the four sub-corpora) plus 68,000 words in register-neutral (cross-register) reference corpora in both languages. A characteristic feature of our corpus is the alignment of source and target texts on different linguistically motivated layers: we not only align sentences (which is state of the art in Translation Memories; e.g. Johansson et al. 1996) and words (which is state of the art in Machine Translation; cf. Och & Ney 2003) but also clauses. One of the methodological principles for the compilation of the resource is the distinction between lexico-grammatical/ cohesive annotation of source and target language texts (including