Corpus-based contrastive studies and AI-generated translations

Abstract This article addresses the issue of using AI-generated translations to perform contrastive analysis. The aim is to establish whether bidirectional translation corpora (and by extension human translators) have become superfluous, given the success of AI-generated translations. To investigate this, results from a previous study of English bring and Norwegian bringe based on bidirectional data are compared with results from a study based on AI-generated data. The AI-generated translations show a markedly higher Mutual Correspondence between the verbs than the human translations. This AI-translation effect may give an inaccurate picture of how equivalent the verbs really are. In cases where AI and human translations deviate, the latter are characterised by semantically more specific verbs. The study highlights some features only available in bidirectional corpora, including the possibility of taking individual variation into account. The findings suggest that bidirectional corpora (and human translators) still have a role to play in contrastive studies.