Recurrent Neural Networks in Linguistic Theory: Revisiting Pinker and Prince (1988) and the Past Tense Debate

TL;DR

This paper demonstrates that modern neural network architectures, like Encoder-Decoder models, effectively address past linguistic debates and criticisms, suggesting their renewed potential in cognitive and linguistic modeling.

Contribution

The study shows that contemporary neural networks can overcome previous criticisms of neural models in linguistic tasks, prompting a re-evaluation of their role in cognitive science.

Findings

Modern neural networks outperform older models in past tense mapping.

Encoder-Decoder architectures address Pinker and Prince's criticisms.

Results support using neural networks for cognitive modeling.

Abstract

Can advances in NLP help advance cognitive modeling? We examine the role of artificial neural networks, the current state of the art in many common NLP tasks, by returning to a classic case study. In 1986, Rumelhart and McClelland famously introduced a neural architecture that learned to transduce English verb stems to their past tense forms. Shortly thereafter, Pinker & Prince (1988) presented a comprehensive rebuttal of many of Rumelhart and McClelland's claims. Much of the force of their attack centered on the empirical inadequacy of the Rumelhart and McClelland (1986) model. Today, however, that model is severely outmoded. We show that the Encoder-Decoder network architectures used in modern NLP systems obviate most of Pinker and Prince's criticisms without requiring any simplication of the past tense mapping problem. We suggest that the empirical performance of modern networks warrants a re-examination of their utility in linguistic and cognitive modeling.