Linguistics Seminar Talk - Tal Linzen

Title:  Event polysemy and the illocutionary/descriptive distinction

Abstract  

Prediction of upcoming events has emerged as a unifying framework for understanding human cognition. Concurrently, deep learning models trained to predict upcoming words have proved to be an effective foundation for artificial intelligence. The combination of these two developments presents a prospect for a unified framework for human sentence comprehension. We present an evaluation of this hypothesis using reading times from 2000 participants, who read a diverse set of syntactically complex English sentences. We find that standard LSTM and transformer language models drastically underpredicted human processing difficulty and left much of the item-wise variability unaccounted for. The discrepancy was reduced, but not eliminated, when we considered a model that assigns a higher weight to syntax than is necessarily for the word prediction objective. We conclude that humans’ next word predictions differ from those of standard language models, and that prediction error (surprisal) at the word level is unlikely to be a sufficient account of human sentence reading patterns.

Tal Linzen is an Associate Professor of Linguistics and Data Science at New York University and a Research Scientist at Google. Before moving to NYU in 2020, he was a faculty member at Johns Hopkins University, and before that, a postdoctoral researcher at the École Normale Supérieure in Paris. He received his PhD from NYU in 2015. At NYU, he directs the Computation and Psycholinguistics Lab, which studies the connections between machine learning and human language comprehension and acquisition. He has received a Google Faculty Award and a National Science Foundation CAREER award.