Looking back to move forward. Challenges related to deceitful parallel texts and slippery terms

Abstract

In this article we will address some pitfalls and shortcomings related to semi-automatic term extraction and an uncritical reliance on translational equivalents in parallel texts. After a short description of the KB-N project and its corpus material and semi-automatic term extraction tool, we shall go on to give examples of faulty equivalents in translated texts resulting from adaptations to an external context. Next, succeeding a comparison of different views of “termhood”, we shall give examples of how even highly domain-focal terms can go unnoticed by an extraction tool. Finally, we shall present two case studies: verbalisation and clipping. The former study shows how semi-automatic extraction may fail completely and manual extraction be impeded owing to linguistic choices made by a translator in representing certain key concepts. The latter study shows how the absence of full-term realisations in a text can cause confusion as to which concepts are involved if the text is handled exclusively by an extraction tool. The overall conclusion is that there are characteristics of texts which are beyond the control of regular semi-automatic extraction tools, and that human intervention is indispensable.