UTCS Artificial Intelligence
Using Neural Networks to Determine How Word Meaning Varies Across Sentences
This research focuses on the development of computational models to study how the meaning of a word varies between different sentences, using fMRI data. Our hypothesis is based on the assumption that words or concepts are represented as a collection of individual features (attributes) localized on known brain areas/networks. The neural network architecture FGREP is applied to map Concept Attribute Representations (CARs) to brain images in order to understand the fMRI patterns and possibly to find new semantic dimensions (attributes) under the effects of context.
Nora E. Aguirre-Celis
naguirre [at] cs utexas edu
Natural Language Processing (Cognitive)