Colloquium: On Syntactic Copying

January 20, 3:30-5pm, Cobb 201
Greg Kobele, Humboldt-Universität zu Berlin

One of the central projects of linguistics is to determine what the defining characteristics of human language are. Theoretical computational linguists contribute to this project by classifying natural languages in terms of the kind of abstract computational resources needed to recognize all and only their grammatical sentences. Today, the dominant perspective is that natural languages belong to the `mildly context-sensitive' family of languages. In this talk, I demonstrate on the basis of data from the Nigerian language Yoruba that this perspective cannot be maintained. Like many other West African languages, Yoruba has a verbal relative clause construction which involves copying of verbal material into the head of the relative clause. What makes Yoruba so interesting from the computational linguist's perspective is the fact that the copied material can in principle itself contain copies, which takes Yoruba outside the purview of mild context-sensitivity. I propose a new, still restrictive, formal constraint on the complexity of natural languages; that they be contained in the class of parallel multiple context-free languages, which are recognizable in polynomial time.