Since language is not "abcdefghijklmnopqrstuvwxyz" or even "the quick brown fox jumps over the lazy dog", I agree that considering real words/phrases (along with their frequencies) is of primary importance.

Unfortunately, it doesn't seem like most type designers -even those in high esteem- bother with linguistics. I guess type design is undervalued enough that it's simply not cost-effective to spend too much time making things just right; readers -and even most type users- can't consciously detect problems that occur as a result of linguistic ignorance (which of course doesn't mean there's no point - is there a point in designing new fonts to begin with? Readers can't tell the difference between Times and Garamond). So you have to do it for personal satisfaction, or for love of the craft.

For the past ~3 years I've been collecting works that provide data and/or insight into real language, and I'd recommend two types of references to those interested is raising their game (this includes me, since I'm far from taking advantage of -or even just absorbing- most of what I've collected):

1. What usually happens: plain word and pair/triplet/etc. frequencies, in various languages. The archetype is "Computational analysis of present-day American English" by Kucera and Francis, 1967.

2. What *can* happen: improbable but potentially problematic language, for which I can site "Making the Alphabet Dance" by Eckler, 1996. Among the many cool things in there is a list of real words that contain every single possible letterpair in English (like "calfkill" for "fk").

Considering real language in parallel to the conventional things like individual glyphs, preset lists of kerning pairs, control strings, pangrams, and extremely rare problematic words (like "wyvern") helps your font work better when it's actually used, as opposed to when you use it in your controlled laboratory.

hhp