Computational modelling
In this strand of the project we have used modelling techniques to understand patterns that may be associated with defectivity or overabundance.
Recent work, with colleagues in Tübingen (), has used high-dimensional vectors (embeddings) derived from corpora as a basis for comparing observed paradigms of Russian nouns that are classified as defective in the genitive plural with nouns that are not defective in this paradigm cell.
Our results showed that nouns classified as defective in the genitive plural had lower semantic transparency, meaning that the inflected forms of these lexemes are more idiosyncratic when compared with other lexemes. This also means that a simple model that computes meaning based on information from the embeddings of these defective nouns performs less precisely than it does for other lexemes. Defectivity in the genitive plural of nouns investigated here is often assumed to arise because of issues associated with the rules that generate forms, yet our results show that the relationship with semantics is stronger than previously anticipated.