Predicting perceptions of the lexical richness of short French, German, and Portuguese texts using text-based indices

Abstract

We investigated how well readers' perceptions of the lexical richness of short texts can be predicted on the basis of automatically computable indices of the texts' lexical properties. 3,060 French, German and Portuguese texts (between 9 and 284 words long) written by 8- to 10-year-olds were rated for their lexical richness by between 3 and 18 uninstructed raters, and over 150 indices were derived from these texts. We found that the ratings could to a substantial degree be predicted on the basis of these indices and that the accuracy with which the ratings of shorter texts could be predicted was comparable to that of longer texts. For French and German, the greatest predictive power was attained by opaque models with scores of predictors, but models with fewer predictors based on a 6-dimensional framework of lexical richness perception or even with a single, easily computed predictor, Guiraud's index, fared only slightly worse.