Progress Report
The quest continues to make https://dechifro.org/shavian the best pronunciation tool available for the English language. The final boss is the Collins Scrabble Dictionary, all 279,500 words that can legally be played in a Scrabble tournament. If you've ever watched Will Anderson's YouTube channel, you know how ridiculously over-inclusive this word list is. A common question in his comment section is, "What language is this?"
I refereed the above game between two computer players, allowing NOULD, AIRTH, SERK/CIRQUE, LIN, DOW, and EELIER because they were just too good to refuse.
My translation is 99.97% complete -- only 86 words remain unshaved -- but it's only about 88% correct. I pull 1000 random Scrabble words, look them up in the Collins on-line dictionary, find about 120 that need fixing, end up fixing another 800 in the process, then score 88% again on the next try. The thought of crawling the entire dictionary to find and fix all 33,000 incorrect words is too horrifying to contemplate.
I did, however, verify all the two- and three-letter Scrabble words, of which there are 362 and 3086 respectively.
Our immense scientific vocabulary, far exceeding 100,000 words, comes almost entirely from Greek and Latin, which are five-vowel languages. How do you map sounds from a five-vowel language into a fifteen-vowel language? You might say, duh, use the five vowels that best match the donor language and discard the other ten. Wrong!
What you must do instead is assign two or three sounds to each of the five vowel letters, then have an Enigma-style rotor that assigns one sound to each vowel in a particular word. Always run this rotor from right to left so that each new suffix changes the entire word:
econ 𐑰𐑒𐑪𐑯
economy 𐑦𐑒𐑪𐑯𐑩𐑥𐑦
economic 𐑰𐑒𐑩𐑯𐑪𐑥𐑦𐑒
econometer 𐑰𐑒𐑩𐑯𐑪𐑥𐑦𐑑𐑼
econometric 𐑦𐑒𐑪𐑯𐑩𐑥𐑧𐑑𐑮𐑦𐑒
econometrician 𐑰𐑒𐑩𐑯𐑪𐑥𐑩𐑑𐑮𐑦𐑖𐑩𐑯
This is a common argument against phonetic spelling. It breaks roots like nation/national, chaste/chastity, and thousands of others.