An acquisition-based evaluation metric for competing syntactic theories

Abstract

This paper addresses the puzzle of choosing among competing syntactic theories, which are contrasting theories that can nevertheless all properly account for the utterances of some language. To do so, I develop an evaluation metric that is based on language learnability and acquisition. In particular, such an acquisition-based metric considers speed of acquiring a target grammar, as well as relative accuracy of acquisition. In applying such a metric, I propose using the kinds of novel data that can be generated via computational modeling for syntactic learning. I then demonstrate how this evaluation metric can be used effectively by reporting on learning simulations for a case study, which focuses on two competing theories that arise in the literature for verb raising in Swedish. © Institute for the Study of Language and Information.