Phonological Typology and Learnability
Generative linguistics assumes that language learning is possible because the space of possible gram- mars (the typology) has non-trivial structure that can be exploited by a learner. What is this structure? Within the framework of constraint-based phonology, the typology is defined through a substantive theory of the phonological constraints plus a formal theory of their mode of interaction. Learnability guarantees have so far been derived from formal properties of the mode of constraint interaction, with- out substantive phonological assumptions about the constraints. Our project charts the limits of this ”substance-free” approach and lays the foundations of a new ”substance-full” alternative. The complex- ity of various subcomponents of the learning problem is probed through intractability results, showing that these tasks are too hard for a learner to succeed, if (s)he uses only the structure provided by the mode of constraint interaction. Additional structure is needed, provided by phonologically substantive and typologically supported restrictions on the constraints (whether innate or induced). We focus on three phenomena: variation, chain-shifts, and segmental phonotactics. Which constraint theories account for the typological restrictions observed in these three phenomena? How can the learnability implications of these substantial constraint theories be extracted? We aim at analytical guarantees, rather than just simulation results. The complementarity between the phonological expertise of the MIT team and the learnability expertise of the French team will allow our project to break new ground in the learnability of phonology, developing the generative axiom of a connection between learnability and typology to a new level of mathematical sophistication.