Homeostasis in Harmonic Grammar

Second-language (L2) phonology learning presents a unique problem for generative linguistics and neuroscience. A child may acquire multiple phonological systems during a hypothesized "critical period", where language structure self-stabilizes. Learners after this period notice a substantial tendency toward a native grammar. I propose that adult language and learning obey homeostatic principles and can be unified under the concept of a homeostatic space and deviations. Specifically, a unified framework for both can be achieved using Homeostatic Reinforcement Learning. I define a Harmonic Grammar as a homeostatic space, and show that it can model deviance from native constraint weighting as reduction in the space. I apply homeostatic reinforcement learning to a two and four-grammar competition, and show that the learner quickly tends to the L1 grammar. This new union, besides accounting for a traditional linguistic assumption, provides new prospects for cognitive theories uniting physical and abstract mechanisms rather than separating them.