Comparing Verbal Descriptions of Image Memories with Natural Language Processing

Loading...
Thumbnail Image

Date

2023

Authors

Gamez, Julian

Journal Title

Journal ISSN

Volume Title

Publisher

University of Oregon

Abstract

A goal of memory research is to understand how the brain remembers similar events. Analyzing data from human subjects, we explore how competition between memories of images influences their recall by answering the question Does studying images from similar or differently themed categories affect the verbal content used to describe them? The competitive condition was composed of images from a single category (“Pond 1,” “Pond 2”), whereas the non-competitive condition was a set of images from different categories (“Pond 1,” “Library 1”). Specifically, we aimed to quantify how verbal memories of these images varied depending on the study condition. To quantify subjects’ verbal memories, we used natural language processing to map subjects’ descriptions of the images onto points in a high-dimensional “text embedding” space. We performed dimensionality reduction and clustering analyses on these text embeddings and found that semantic representations of images studied in the competitive condition were similarly differentiated compared with those in the non-competitive condition. Our results suggest that verbal memories of images were influenced by the similarity of subjects’ memories and that highly similar memories may push their respective representations away from one another.

Description

7 pages

Keywords

Citation