L'évaluation (de l'évaluation)+ de la diversité lexicale

Abstract

While measures of lexical diversity are systematically evaluated from the point of view of their robustness, they are rarely and rather informally evaluated from the point of view of their sensitivity. This paper proposes a method for the evaluation of sensitivity based on a text generation algorithm which makes it possible to control the degree of lexical diversity of the generated data. The method is illustrated by means of a comparison between two measures relying on two different ways of applying a resampling strategy for the evaluation of lexical diversity.