Use words, not constructions!
Thomas Proisl · International Journal of Corpus Linguistics · 2022
Abstract The aim of collostructional analysis or, more precisely, simple collexeme analysis, is to quantify the statistical association between a constructioncand a lexemelthat occurs in a particular slot of the construction. The analysis is based on2×2contingency tables that ought to represent a cross-classification of the units of analysis. So far, the units of analysis have been identified either as all constructions in the corpus or all instances of a classCof constructions to which constructioncbelongs. In practice, it is often not possible or feasible to identify these constructions. Therefore, the sample size is typically approximated by heuristic estimates. The bottom-right cell of the contingency table is most affected by these approximations. I suggest that the units of analysis be defined on the word level, instead, as the classWof word forms that satisfy the restrictions on the collexeme slot ofc.