CogSci Colloquium: Tal Linzen (NYU)

The Role of Language Models in Linguistics 

Abstract: What should cognitive scientists make of the linguistic skills of systems like ChatGPT? I’ll start this talk with the somewhat disappointing answer “not as much as some people might think”. Fortunately, the remainder of the talk will be less bleak: I will show that the neural network architectures that these systems are based on can, in fact, be helpful in addressing debates in cognitive science, if used with care; I will focus on poverty of the stimulus arguments in language acquisition and the role of prediction in language comprehension.