Poster C83, Sunday, March 26, 5:00 – 7:00 pm, Pacific Concourse
Investigating semantic representations in brain with fMRI and LSA
Sverker Sikström1, Johan Mårtensson1; 1Department of psychology, Lund university
Participants studied words, and pictures describing the words, while being scanned with functional magnetic resonance imagining. Latent semantic analysis (LSA), based on Google-ngrams, was used to generate semantic representations of the words used as stimuli material. Functional data analysis was conducted by using MVPA/RSA as outlined in Kriegskorte, Mur & Bandettini (2008). Fibertracking was performed by creating estimations of native space tracts between the distinct brain regions that were activated during the fMRI task. This was done in FSL/FDT. Individual ROIs from cortical areas were estimated using the Freesurfer software package and used as starting regions for the MVPA/RSA analysis. Connectivity indices along with semantic dimensions were compared for representational similarity using RSA. The results showed that the semantic similarity between words using LSA predicted the representation similarity in the brain using RSA.
Topic Area: LANGUAGE: Semantic