In The Science of Word Recognition, Kevin Larson, a psychologist on Microsoft’s ClearType team, reviews word recognition and reading acquisition.
It turns out that the techies on the team were using an old but oft-cited paper by Bouma as the basis for their work. It suggested the word-shape model of recognition — which is no longer considered valid. The parallel-letter model better reflects how we really read.
I love some of the experiments in the literature, like asking subjects to proofread documents with a few intentional misspellings — some of which change the word’s shape without changing the letter’s shape (than versus tnan), some of which change the letter’s shape without changing the word’s shape (than versus tdan), etc.
(Hat tip to Slashdot.)