Designing semantic feature spaces for brain-reading
Résumé
We focus on a brain-reading task which consists in discovering a word a person is thinking of based on an fMRI image of their brain. Previous studies have demonstrated the feasibility of this brain-reading task through the design of what has been called a semantic space, i.e. a continuous low dimensional space reflecting the similarity between words. So far the best results have been achieved by carefully designing this semantic space by hand which limits the generalization of such a method. We propose to automatically design several semantic spaces from linguistic resources and to combine them in a principled way and achieve results comparable to that of manually built semantic spaces.