tli doesn't map ' to a proper IPA symbol about g2p HOT 7 CLOSED

Reading up on the Tlingit orthography on Wikipedia (https://en.wikipedia.org/wiki/Tlingit_alphabet), it appears ʼ (\u02BC) would be more appropriate, the ejective stop character (https://en.wikipedia.org/wiki/Dental_and_alveolar_ejective_stops).
Panphon does not know it, though, so I'm guessing in tli-ipa -> eng-ipa, we should manually map \u02BC to the full glottal stop, so that we get HH in eng-arpabet.

from g2p.

Hi @joanise , thanks for this. Yes, you're right, it marks ejectives - frustratingly, I believe periods mark glottal stops. Ejectives should get mapped to their plain unvoiced stop equivalents, so k\u02BC should go to k. Panphon does recognize the character, but will just remove it in the generate-mapping step:

>>> import panphon.distance
>>> dst = panphon.distance.Distance()
>>> dst.weighted_feature_edit_distance("kʼ", "k")
0.125

I suppose the thing to do would be either to replace all ' with \u02BC, or to add individual rules for each ejective (if we thought ' might be used legitimately in punctuation). I think the individual rules option might be the best.

from g2p.

Ah, I see! thanks for the explanations. We need to have ' somewhere in the g2p, because otherwise the tokenizer will not consider it a word character. g2p generate-mapping does not like having \u02BC in a rule on its own, though. I think a reasonable solution is to have a rule for each ejective plosive, as you suggest, in tli-ipa -> eng-ipa, and I would do it manually rather than through generate-mapping, for simplicity's sake.

As for the tokenizer, maybe the simplest solution is to have the no-op rule ',' in tli_equiv.csv mapping ' to itself. Then the tokenizer would include the ASCII ' in the alphabet it's looking for.

from g2p.

I take it back, g2p generate-mapping was quite happy to generate this rule:

    {
        "in": "\u02bc",
        "out": "",
        "context_before": "",
        "context_after": ""
    }

and maybe that's just the simple solution we should implement, and add the rule(s) mapping ' to \u02BC as appropriate in tli_to_ipa.csv.

from g2p.

About the dot mapping to the glottal stop, how does the language mark it's end of sentence punctuation? Is there ambiguity?

from g2p.

Next question, can the glottal stop occur at the end of a word? I could disambiguate by saying only map the . if it's not word-final.

from g2p.

Fixed by PR #82

from g2p.