English spelling has spent centuries earning its reputation as a practical joke, and Mark Rosenfelder’s old Zompist page responds with the deeply Internet-era reaction: fine, let’s model the damn thing.
Visit Zompist’s English spelling page
The page, dated 2000, is titled “Hou tu pranownse Inglish.” Its argument is not that English spelling is elegant. It is that the system contains far more regularity than people usually give it credit for.
Rosenfelder lays out sound values, dialect assumptions, letter combinations, vowel rules, consonant rules, exceptions, historical leftovers, and the famous troublemakers such as gh and ough.
He also takes a swing at George Bernard Shaw’s famous “ghoti” joke, arguing that the proposed pronunciation only works by ignoring where English spelling rules actually permit those letter values.
The page then does something that makes it especially interesting as an old Web artifact: it turns the argument into a computational test.
Make the computer pronounce it
Rosenfelder assembled a sample lexicon of more than 5,000 English words and a set of rules for his Sound Change Applier. According to the page, the rules generated 59 percent of pronunciations perfectly and 85 percent either perfectly or with only relatively minor errors.
That does not make English spelling sane.
It does demonstrate that “English spelling is completely random” is not a particularly useful explanation.
The page is also full of old-browser archaeology. It discusses Unicode support as something the reader’s browser may or may not handle correctly and explains notation choices partly in terms of what HTML could reliably display.
That mixture is classic early specialist Web publishing: a serious subject, a personal voice, hand-built technical notation, downloadable data, and the assumption that an interested reader will happily scroll for a very long time.
Modern search results tend to chop questions like this into isolated answers: why is ough weird, why is knight spelled that way, why does c have two sounds?
Rosenfelder tried to put the machinery in one place.
CacheRat’s 1,967 Ancient Web Domains research list includes pages like this because old personal sites often preserve the full argument instead of optimizing each paragraph into a separate answer box.
English spelling is still a mess.
At least this mess comes with documentation.