Date of Award

Spring 1-1-2019

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

First Advisor

Albert Kim

Second Advisor

Tim Curran

Third Advisor

Matthew Jones

Fourth Advisor

Erik Willcutt

Fifth Advisor

James Martin

Abstract

In conversation or during reading, we sometimes find ourselves making predictions about the identity of an upcoming word or phrase. This phenomenon has been reflected in the results of laboratory experiments that show changes in eye gaze patterns or the Electroencephalogram (EEG) prior to encountering a predicted word. However, questions remain about exactly what is predicted during language comprehension, as well as how often predictions are computed. Existing results most clearly support the prediction of upcoming words’ meanings, while lower-level predictions of a word’s perceptual features are less well supported. Furthermore, most evidence for prediction has come from tasks where participants read language that is designed to be predictable, so the importance of predictive processes for typical language use also remains unclear. Meanwhile, increasing attention has been paid to more general models of brain function that posit prediction and prediction error as representing the two fundamental “units of account” that are used in the passing of information between levels of the neural hierarchy. These predictive coding models imply that predictions during language comprehension must be generated constantly, and at all levels of representation. In this dissertation I describe three experiments that are designed to address the empirical matter of whether low-level word-forms are in fact predicted during comprehension, as predictive coding models demand. In study 1, I show that the unexpected omission of a highly predictable, sentence-embedded visual word still leads to a brain response that reflects the omitted word’s visual length. In study 2 I describe an experiment that suggests study 1’s findings reflect prediction error, rather than word-form prediction itself. Finally, study 3 provides evidence that word-form predictions are not limited to highly predictable contexts. By providing consistent evidence for the predictions of word forms during comprehension, the results suggest prediction during language is not limited to only the most abstract levels of representation, or only the most constraining linguistic contexts. Instead, they suggest the neural machinery underlying word recognition is fundamentally predictive, confirming the predictions of predictive coding theories.

Share

COinS