Department of Linguistics
University of Pennsylvania
3401 Walnut St. Room 401B, or via Zoom
Distributional learning of recursive structures
Languages differ regarding the depth, structure, and syntactic domains of recursive structures. For example, English allows infinite free embedding of the prenominal possessive ‘-s’ to express ownership, whereas the postnominal possessive ‘of’ is much more restricted; in contrast, German allows the post-nominal ‘von’ (of) but not the prenominal ‘-s’ to embed freely. Therefore, while the ability for recursion is considered universally available, speakers need to learn from experience which specific structures allow free embedding and which do not. In this talk, I will present a proposal on how recursive structures should be formulated, which leads to a theory of how they are acquired by children. I propose that recursion derives from structural substitutability: A possessive structure such as ‘N2’s-N1’ is recursive if all nouns that can appear in one position of N1 or N2 can also appear in the other. Therefore, recursive structures can be acquired from one-level data distributionally as a special instance of productivity in language. A series of corpus analyses and artificial language learning experiments will be presented in support of the proposal.