Monday, April 23, 2007

Crunching the numbers

It's the space between the letters. In the May/June issue of Stanford Magazine, a profile of famous ur-computer scientist Donald Knuth, who has over the course of his career written but not finished a mammoth multivolume book called "The Art of Computer Programming."

What interests me most in the article is the discussion of Knuth's exacting views about typography. Knuth was getting all hot and bothered by the poor quality of the printing in scientific textbooks. They tended, for one, to feature letters so closely spaced as to form dark spots on the page, which stop the eye from scanning the page quickly and smoothly.

Knuth took it upon himself to find a solution, even though his typesetting experience was limited to a high school job at a print shop and some experiments with the offset press his father had kept in the basement. Typesetting was no longer a manual craft. "It had changed into a problem of bits, zeroes and ones," he says. "You put the one where you want ink on the page and zero where you don't want ink. So I figured, okay, I'm good at zeroes and ones."

In 1977, Knuth halted research on his books for what he expected to be a one-year hiatus. Instead, it took 10.

...

Knuth, trying to train his programmer's brain to think like an artist's, wanted to create a program that would understand why each stroke in a typeface would be pleasing to the eye. "I wanted to try to capture the intelligence of the design, not just the outcome of the design," he says. For example, how do you insert line breaks into a paragraph so there isn't too much space between words and so that most of the lines don't end in hyphens? Although this seems like an aesthetic challenge to be solved by human taste, Knuth says, computers do it well. "This is a combinatorial problem," he explains. "There might be a thousand ways to break a paragraph into lines and each way has a score."

So Knuth built computer programs to count and score those thousand ways and pick the best one. His software is used in the bulk of scientific publishing today.

[via Metafilter]

Posted by Evan Hughes at 06:41 PM

Print | E-mail to a friend | Permalink | Subscribe via rss