On an emotional node: modeling sentiment in graphs of action verbs

Neuroimaging studies have over the past decades established that language is grounded in sensorimotor areas of the brain. Not only action verbs related to face and hand motion but also emotional expressions activate premotor systems in the brain. Hypothesizing that patterns of neural activation might be reflected in the latent semantics of words, we apply hierarchical clustering and network graph analysis to quantify the interaction of emotion and motion related action verbs based on two large-scale text corpora. Comparing the word topologies to neural networks we suggest that the co-activation of associated word forms in the brain resemble the latent semantics of action verbs, which may in turn reflect parameters of force and spatial differentiation underlying action based language.