Support Cantonese

[jctian98]

Add Cantonese support, which mainly follows Chinese implementation.
Key features:
(1) The Cantonese-to-Jyutping dictionary from CC-Canto.
(2) Jyutping-to-IPA dictionary from Wiki

Other important points:
(1) In the Jyutping-to-IPA rules, the m and ng have different IPA parsing that are dependent on context. Specifically, they can (1) represent isolated syllables; and (2) as a consonant. So we first check (1)
(2) The order of Jyutping-to-IPA rules matters: jyutping items with long characters are checked first.

I haven't found a good test example for Chinese, maybe because it needs file downloading.
As suggestion is welcome

[jctian98]

There should be some issues with the CC-Canto dictionary: some isolated characters are not included, even though there are many words built with it. It seems this issue doesn't happen in the Chinese dictionary but is in the Cantonese dictionary.

E.g., "今" is missing, but many words with "今" is included. Parsing "今天" will fail as "今tʰin˥"

[jctian98]

Update after discussion with @dmort27

When parsing a word with several characters, also add the Jyutping of each character into the dictionary to avoid OOV problem. This will definitely introduce some issues as some characters may have more than one pronunciation.
I manually checked the output of test case and think they are reasonable (although with some modifications with the original ones).

I still get an issue with Jyutping -> IPA mapping. E.g.,
(1) jat1 -> jɐt̚1 when applying at -> ɐt̚ / _
(2) jɐt̚1 -> jɐtʰ̚1 when applying t -> tʰ / _
(3) jɐtʰ̚1 -> jɐtʰ̚˥ when applying 1 -> ˥ / _

The step (2) is totally unexpected, as the t here is already IPA rather than Jyutping. But I'm not sure how to avoid this phenomenon. Any advice on it?

[kalvinchang]

is t really supposed to turn into aspirated t in all environments ?

[jctian98]

No. so that the step (2) is not expected. the rule t -> tʰ / _ is mis-used here as it takes the IPA t as a Jyutping character and transforms it again.

[kalvinchang]

maybe make t -> tʰ word-initial?

seems like u already covered the cases where [t] is word-final

[juice500ml]

Can you also update the README.md too? Thanks!

[juice500ml]

Closing & reopening PR for triggering CI

[juice500ml]

@jctian98 , can u resolve merge conflicts? Also, there is slight change in coding style of download.py, you might want to change it.

[jctian98]

I added some special characters to indicate the transformed IPAs and then avoid applying rules repetitively.

I think this PR is ready to review now. @dmort27

[juice500ml]

It seems that the above issue have been added in tests, and been resolved. Merging the PR. Thanks a lot everyone!