Multi-layer analysis of translation corpora: methodological issues and practical implications

The present paper discusses an application of multilingual, multi-layer corpus analysis from translation studies. The concrete context is the empirical testing of hypotheses about the specific properties of translations, such as explicitation, simplification, sanitization or normalization. While some of these assumed properties can be tested using some rather shallow measures that operate at the level of words (e.g., typetoken ratio or lexical density), others require the analysis of more abstract linguistic features at the level of the clause; also, it is necessary to refer to linguistic features across different strata of linguistic organization (semantics, grammar, discourse). We present some selected hypotheses about the nature of translations which have been tested on a bilingual EnglishGerman corpus. We discuss the particular corpus design adopted and the requirements on the techniques needed to extract information from this corpus and show how we use a combination of some standard corpus analysis techniques, such as string-based concordancing and part-of-speech tagging, and semi-automatic corpus analysis tools. We conclude with a summary and a list of issues for future work.