Closed grantlemons closed 1 month ago
Actually, what's happening here is that the classifier doesn't include rules of capital letters. It is fixable--and I'm on it.
Just to summarize what I took from us discussing this the other day, the issue you've pointed out with not considering capital letters is a different, though related issue.
This is specifically about the way letter pronunciation in initialisms sometimes begin with vowels even when the letter itself is a consonant. i.e. the letter L is pronounced [el]
, so initialisms beginning with L begin phonetically with a vowel and so are referred to with an
not a
.
When we discussed this, I think you said you would just consider all capital letters like this (I don't really remember), but this isn't really the approach either. C
, for example does not phonetically begin with a vowel.
You could resolve this by just hard coding by the letter, but that feels a bit jank.
I recently came across this as well. ("a/an SSH connection")
I don't know of any way to fix it other than hard-coding "weird-named letters":
Consonants with vowel sounds: F, H, L, M, N, R, S, X Vowels with consonant sounds: U, Y(?)
I've committed a temporary fix--explicitly treating apparent acronyms/initialisms in a different way.
If you compile from master
you can give it a go.
I think this one's good to go. Reopen if there are any further issues.
Example:
an LLM
is valid, but haper suggestsa LLM
because L is a consonant, even though phonetically it begins with a vowel.Probably not fixable.