Are there theories that can predict, at least for modern English, how a complex word will be realised phonologically and graphically i.e. how it will be pronounced and spelled, given the spelling and pronunciation of the input / component morphemes? (please consider derivational, inflectional and compounding. if possible, please consider specialist languages especially medical terms)


They aren't really "theories" in the scientific sense, but there are definitely patterns to the ways that complicated/derived words in English are spelled and pronounced. In fact, such words tend to have more "regular" spellings than shorter and more common words—I suppose that could be seen as a specific example of the general principle that frequent words are more likely to be irregular (which is most often applied to the topic of English verb inflection, rather than spelling).

Stedman's Medical Dictionary has sections that touch on the pronunciation of Greek and Latin-derived words in medical terminology. A chapter about this used to be freely available online, but it looks like now you have to pay for it: https://stedmansonline.com/webFiles/Dict-Stedmans28/APP04.pdf

The Wikipedia article "Traditional English pronunciation of Latin" is quite long, but contains a fair amount of information that is relevant to the pronunciation of medical terms.

One complication is that medical students themselves don't necessarily receive detailed instruction in pronunciation "rules" like this, so there isn't always consistent usage of the theoretically regular pronunciation. For example, an ELU poster said that "some medical instructors use /ks/, others use /k/ and a few of them use /s/" in the word buccinator. Another irregular pronunciation I have noticed in use is me-DOOL-a for medulla. You can see more variable words mentioned in this Reddit thread. Basically, you can only count on the pronunciation to be predictable from the spelling if the people who use the terms expect the pronunciation to be predictable from the spelling; otherwise, some people will just guess, which leads to variation, especially in things like the position of stress.

  • For those types of examples, we can train models (which are not really theories) as well as humans do, ie pass the Turing test, which is, as you say, not perfection, that's life, and probably outperform any single human on very large eval sets, of course with more training hours/examples. But much harder for speech synthesis is the contextually-dependent stuff, for example "test_all()", "IEEE", "MUSE", slang, misspellings... – Adam Bittlingmayer Feb 23 '19 at 14:07
  • 1
    (I use software terms, but medical terms should be the same, there as a time when "x-ray" or "CAT scan" was a neologism. And "Mayo" maybe has a different pronunciation in the domain context. Also not clear what's correct, I know how I would pronounce Mehmet Öz, but I'm sure the inhabitants of the Greatest Country on Earth have more creative interpretations.) – Adam Bittlingmayer Feb 23 '19 at 14:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.