Learning a Manifold of Fonts

I’m a tad skeptical that this sort of statistical approach is going to produce any interesting results, especially given that their choice of fonts to use as input wasn’t very well selected. They simply took all of the preinstalled Windows 7 fonts as their sample, but they didn’t bother to exclude those fonts which aren't intended for Latin usage. That means their sample is going to be heavily biased towards Arial and Times New Roman since most of the Windows non-Latin fonts populate their basic latin/latin-1 blocks with glyphs from those fonts.

Andre

There is a lot of “stylistic” choice (curation) involved in constructing such devices, in deciding where to place the sources on a simple two-axis grid, bearing in mind that PANOSE has way more than two dimensions! 

Here it is in three dimensions, Noordzij cube by David Ross:
https://www.youtube.com/watch?v=_0XX88qVEEc

It breaks down with things like disjunctive forms of /a and /g (one- and two-storey).

It also reminds me of the FontFont poster from a few years ago, black and yellow, in which all their typefaces were arranged in like groups. I couldn’t find a picture of it online; it’s somewhere in my studio…

Usually, this kind of scientific dabbling could use some grounding in type culture, to avoid starting at square one as if nobody had walked these paths before.