The learnability of linguistic irregularities: A simplicity-based approach

There is currently a significant divide in researchers views on language acquisition. Linguists have often posited that a strong innately born knowledge of language structure is necessary for learning language. This innateness hypothesis proposes that we are equipped at birth with specific knowledge concerning the structure of human language. Recently, a general statistical model centred on a simplicity principle has been shown to be capable of learning many different aspects of language syntax. These results suggest that language learning can be achieved with much weaker innate language specific knowledge than previously supposed and set an upper bound on what is learnable using statistics and syntax alone. However, because it is prohibitively computationally intensive to apply these models to the scope of natural language, these models have only used on specially restricted data sets or artificial language. The goal of our research is to adapt these models so that they are applicable to natural English language. We will then assess the learnability of specific parts of language by applying the simplicity principle to natural English as a whole, providing a quantitative analysis of this central debate in language acquisition.