Abstract

This paper is an attempt to assess empirically whether, as often stated in the literature, inflection is "semantically more regular" than derivation. We reformulate the observation in terms of the stability of syntactic and semantic contrasts within a morphological relation: inflectional contrasts are hypothesized to be more stable than derivational contrasts. We then propose an operational definition of contrasts between words as offset vectors in a distributional vector space of the kind familiar from distributional semantics. In the empirical part of the paper, we show that French data does validate the hypothesis for all pairings of an inflectional and a derivational relation that we were able to investigate.