{To Follow}

This reminds me of the image John Hudson posted a few times (last time here: http://typophile.com/node/91898): a collection of Baskerville’s lowercase a sorts from a single page.

Full size
Thanks for the link, that's exactly what I'm talking about. (Pasted it above.)

Also related is Koch's Neuland. In some ways it's exactly what I'm imagining, though less drastic. Here's the different cuts of R:

Here's a thought, as we're all molding our letters with copy & paste. What if you drew one alphabet, avoiding the use of copy & paste—just judging by eye. Then take a some letters, maybe d, i, Q, and E*, and redraw the rest of the alphabet. Probably halfway through, remove your DNA letters, and redrew those too. Do it a few times, maybe with different letters for DNA. I wonder what you'd come up with.

I'm sorry, why exactly are we doing this?

I'm sorry, why exactly are we doing this?

1. Because it's interesting. What if you made the same font twice—what would be different?
2. To make text 'livelier', by making slightly different versions of each glyph. In a way emulate artifacts of older methods of printing.

I've thought this out, though, and that it may make more sense to make alternates based on letter frequency. The glyphs that are used more often are the ones that "need" more alternates. A piece of type that is used more gets worn more. Or, has more sorts made.

Does it seem like a pointless idea?

You’d also have to factor in the most noticeable opportunities for divergence—i.e. those letters which appear most often next to, or close to, each other.

With that in mind, I developed a pseudo-random feature that requires only one alternate.
The theory is that because there are virtually no instances in which a letter appears three times in a row, two letters may be coded so that the closest they appear is separated by four letters, far enough apart to not be present simultaneously in the fovea, which is the only circumstance in which artifacts are detectable in fluent reading.

http://ilovetypography.com/2011/04/01/engaging-contextuality/
(Discussing Fontesque Pro.)

But it's definitely not just how many can occur in a row; I think it's how many can be seen in a fixation (assuming such detailed visual memory dissipates with a saccade). This is why I like LettError's scheme of 4 alternates (you can't see more than that in one fixation).

hhp

How many words is it possible to see in a fixation? Isn't the average around 2-3 words, with good readers able to take in up to around 5? That's quite a block to optimize for.

Do you have more info on LettError's scheme?

In any case, I'm making a script to count glyph pair occurrence in a given document (actually, it's almost done). It would take little effort to make it count trios, or larger groups. With little modification, I could count occurrence of glyph pairs near one another. For example, how often e occurs next to i with three glyphs between.

I was unclear: it's possible to see a number of words during a given fixation, but only the 3-4 letters within the fovea are clear enough for gentle variations to be relevant.

What your script does has been done before, but there's certainly plenty of room for type-design-savvy refinement and focus!

In LettError's scheme there's no randomization - it simply cycles through the choices (which cannot be called "pseudo-random" BTW).

hhp

What if you made the same font twice

There is so much work to do, so many ideas to realize, why would I make the same font twice? You say it would be interesting. It wouldn't be interesting to me.

http://processtypefoundry.com/fonts/lingua/
http://www.volcanokit.com/volcanokit3/ligCounter/

James: I get your point. Actually, I think I may have to agree with you. Making alternate characters by "accident" isn't nearly as interesting or useful as purposefully making variations. The idea still sounds interesting—but the results could be trivial. Which might make the idea less interesting.

Thanks for the links, Nick.

It's always interesting to find that others have thought my thoughts before. Not that's it's a bad thing. I'm not just interested in the frequency of single letters and pairs but also larger groups and adjacencies.

On the other hand to me making letters look "accidental" is... blasphemous. What would be cool is to use true random numbers* to generate -or at least vary- glyph shapes.

* See for example: random.org

hhp

Right. But how do you do that in Opentype programming? I don't want a Beowolf approach. Even with a "true" random, some guiding has to be done to make the letters excellent rather than just interesting.