Masoud Jasbi

Eva Portelance, Masoud Jasbi

In mid-20th century, the linguist Noam Chomsky established generative linguistics, and made significant contributions to linguistics, computer science, and cognitive science by developing the computational and philosophical foundations for a theory that defined language as a formal system, instantiated in human minds or artificial machines. These developments in turn ushered a wave of research on symbolic Artificial Intelligence (AI). More recently, a new wave of non-symbolic AI has emerged with neural Language Models (LMs) that exhibit impressive linguistic performance, leading many to question the older approach and wonder about the the compatibility of generative AI and generative linguistics. In this paper, we argue that generative AI is compatible with generative linguistics and reinforces its basic tenets in at least three ways. First, we argue that LMs are formal generative models as intended originally in Chomsky's work on formal language theory. Second, LMs can help develop a program for discovery procedures as defined by Chomsky's "Syntactic Structures". Third, LMs can be a major asset for Chomsky's minimalist approach to Universal Grammar and language acquisition. In turn, generative linguistics can provide the foundation for evaluating and improving LMs as well as other generative computational models of language.

Tovly Deutsch, Masoud Jasbi, Stuart Shieber

Readability assessment aims to automatically classify text by the level appropriate for learning readers. Traditional approaches to this task utilize a variety of linguistically motivated features paired with simple machine learning models. More recent methods have improved performance by discarding these features and utilizing deep learning models. However, it is unknown whether augmenting deep learning models with linguistically motivated features would improve performance further. This paper combines these two approaches with the goal of improving overall model performance and addressing this question. Evaluating on two large readability corpora, we find that, given sufficient training data, augmenting deep learning models with linguistically motivated features does not improve state-of-the-art performance. Our results provide preliminary evidence for the hypothesis that the state-of-the-art deep learning models represent linguistic features of the text related to readability. Future research on the nature of representations formed in these models can shed light on the learned features and their relations to linguistically motivated ones hypothesized in traditional approaches.