Seminar in Computational Linguistics

The Zipfian principle of least effort (frequent forms are likely to be shorter) is assumed to manifest itself at various language levels. It has, for instance, been claimed that more frequent members of morphological oppositions are on average shorter (Hawkins 2010), e.g. singular forms are usually shorter than plural forms. In this talk, I present the first (to my knowledge) large-scale test of this assumption. Using Universal Dependencies corpora, I test whether the assumption holds across 60 languages and 56 morphological features. Most often, there indeed is a significant negative correlation between frequency and length, but it is not universal, and the proportion of such correlations varies across features, parts of speech and languages.

 

Hawkins, John A. "Processing efficiency and complexity in typological patterns." In The Oxford handbook of linguistic typology. 2010.