feat: don't split text inside ruby transcriptions#2795
Conversation
birtles
force-pushed
the
dont-split-rt-text
branch
from
9a71097 to
c52b45a
Compare
birtles
force-pushed
the
dont-split-rt-text
branch
from
d19f18c to
b9ced9f
Compare
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 18f77cdf9b
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c7d8ef1da3
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| > { | ||
| let [word, inputLengths] = normalizeInput(input); | ||
| noSplitMask = remapNoSplitMaskToNormalized({ inputLengths, noSplitMask }); | ||
| noSplitMask = addYoonToNoSplitMask({ input: word, noSplitMask }); |
There was a problem hiding this comment.
When the words DB falls back to the flat-file backend, word is converted from katakana to hiragana (kanaToHiragana) later in this function, but the yoon mask is computed earlier from the pre-conversion text. That means katakana inputs like キャ can lose yoon protection in fallback mode and get shortened to き, which is a behavior regression from the previous endsInYoon logic (it ran after conversion inside wordSearch). This affects lookup accuracy specifically while the IndexedDB words DB is unavailable/updating.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Better still, we should make isYoon work on katakana キャ etc.
1 participant
Fixes #2785