Distinguished Lectures in Humanities: Using Large Language Models to Augment the Psycholinguistic Toolbox

Abstract

Until recently, there were two main ways of obtaining information about words and expressions. The first was to analyze large text data sets (corpora) and calculate the frequency with which words and phrases occur, as well as the typical contexts in which they occur. The second was to ask participants to provide subjective information about words and phrases, such as the familiarity of the stimuli or the age at which they are typically acquired. The development of large language models has given us a third option. Instead of asking participants for information, we can query large language models. The results show that the information obtained from those models is just as good and often even better than the information obtained from people, especially when the model is tuned to a few thousand stimuli.

About the speaker

Marc BRYSBAERT is Professor of Psychology at Ghent University, Belgium. He is best known for his work collecting word-related information in several languages (including Mandarin Chinese), such as measurements of word frequency, age of acquisition, concreteness, valence, arousal, and word processing times. He served as Editor-in-Chief of Behavioral Research Methods and is a highly cited author.