experiment with visualization

[garfieldnate]

Just read through "Analogy in suffx rivalry: the case of English -ity and -ness", and Sabine has an interesting graph-based visualization which clusters together items which share their most important gang. It was really useful in her analyses, and I wonder if we can come up with something that accounts from even more detail.

With the same graph-based visualization (a Fruchterman-Reingold diagram), we could come up with some new similarity metrics based off of the effect sizes of different gangs, or maybe more granularly based on the effect size (number of points) of each exemplar.

If we assign a vector to each item based on the important of every observed gang or of every other exemplar, we could then run this through PCA to see if there's anything interesting, or a high-dimensional visualization algorithm like t-SNE.