Statistical comparability: Methodological caveats

Abstract

The notion of comparable corpora implies the notion of comparability. The present paper aims at explicating this notion with respect to statistical methods because statistical comparison requires the use of statistical tests, which again require certain properties of the data under analysis. Linguistic data, however, do not automatically meet these requirements. In corpus linguistics and other linguistic fields, statistical methods are often applied without any previous check of their applicability. The paper will give some warnings and show some examples of corresponding test procedures. A number of other frequently used terms and concepts, such as representativeness, homogeneity, and balanced corpora, play a central role in corpus-linguistic argumentations and will be analysed in the paper, too, as they concern compilation and use of comparable corpora.