Figure 11.

Distributions over verb classes built from clustering on the semantics of the object. Clockwise from the top left: the heatmap shows the pairwise Jensen-Shannon Divergence (top half) and statistical significance (bottom half), as well as the homogeneity (diagonal). The dendrogram shows hierarchical clustering based on cosine difference between each subdomain's JSD values. The scatter plot is colored according to the best K-means clustering (determined by the Gap statistic) projected onto the first two principal components (normalized).