Working memory constraints on multiple center-embedding.

Gibson’s (1998) theory on the locality of syntactic dependencies claims that multiply center-embedded clauses are unacceptable if they contain a parse-state with at least two long unresolved predicted categories in addition to the toplevel verb. ‘Long unresolved’ means a syntactic prediction spanning at least three intervening new discourse referents. This claim was based on experimental analysis of invented examples. Karlsson (2007b) provided corpus data demonstrating that, contrary to widely accepted views in linguistics and cognitive science, there are well-defined constraints on how many (maximally three) and what types of multiple center-embeddings occur in spoken and written discourse in natural languages. Gibson’s theory of the processing of multiple center-embeddings will be evaluated in the light of Karlsson’s empirical data. The corpus data do not support the idea of a discrete limit on working memory capacity, because almost one third of the extant examples of multiple center-embedding are more complex than Gibson’s acceptability limit stipulates. Spoken language processing complexity is clearly below Gibson’s limit, written language is capable of transgressing it.