The theoretical goal of this paper is to clarify some central concepts used in corpus-based translation studies. If one is primarily interested in uncovering the essence of translation per se, a distinction should not be made between norm-dependent and potential universal features, but rather talk about laws of translation more widely. The empirical goal is to outline some results on dissimilarities in the frequencies and distributions of three non-finite structures of Finnish (referative, final and temporal constructions) in different language variants: texts originally produced in Finnish and texts translated from English and Russian into Finnish. Evidence is provided in support of a possible universal law that translations tend to under-represent target-language-specific, unique linguistic features and over-represent features that have straightforward translation equivalents (functioning as some kind of stimuli) in the source language. It is a question of interference but not in a negative, but rather a neutral, abstract and statistical sense.