Words with hyphens guessed wrongly by the Morphology guesser

Question

In the language we work some affixes and clitics are separated from the stem by hyphens. This hyphen is defined in the language settings as word-medial punctuation.

Now in the Morphology field Paratext sometimes makes wrong guesses because it ignores such hyphens. For example, it guesses the morphology of ‹khøuwa-te› as /khøuwat-e/. Both ‹khøuwat› and ‹-e› do exist as morphemes, but here it should be clear because of the hyphen that the morphology should be ‹khøuwa› and ‹-te›, which are also already defined as morphemes.

Is there a setting anywhere I could adjust so that the Paratext Morphology guesser will not treat any character string with a hyphen in it as one morpheme?

If not, I’m going to suggest such a feature.

Thanks,
rfkvg

Answer 1

It does seem that “word-medial punctuation” by definition implies a morpheme break in most cases, so the parser should prioritize a break at those hyphens. If hyphen represents a “phoneme” i a language should it not be included in the list of alphabetical characters?

Answer 2

Forgot to mention that I’m otherwise very impressed with the morphology guesser! It’s an immense time saver!

Comments

Sometimes Paratext needs a few corrections of a similar type in order to learn a pattern and to begin to make good guesses. I have no idea whether the presence of a hyphen is confounding the guesses though.
Does manually correcting hyphenation or morphology in their respective columns in the Wordlist (accessed via the View menu) cause Paratext to learn the rules you’d like it to understand? Do subsequent suggestions from Paratext guess the morphemes more reliably? Or have you tried this and found that Paratext keeps guessing badly?

---

Thanks for that, IanH.
On the whole Paratext is quite good at learning the rules. However, in this specific case both guesses would be equally correct if there wasn’t a hyphen in the orthography. It seems that Paratext is just ignoring such hyphens. It would be great if there was a way to tell Paratext not to ignore them.