globalwordnet / globalwordnet/semcor
Fix nested closing-quote order swap ('"' should be '\'"')
- Dominant language
- Python
- Stars
- 0
- Forks
- 0
- Avg merge
- 2d 22h
- Merged PRs (30d)
- 24
Description
When a double-quoted span contains a nested single-quoted word/phrase and both close at the same point, Brown's real text always closes the *inner* single quote before the *outer* double quote -- e.g. `elected'".`, never `elected"'.`. This corpus had 42 raw instances of the reverse (wrong) order.
Verified each individually against `src/semcor/brown-nolines.txt` via context-window matching (reconstruct the target word with the pair swapped back, require a unique context-confirmed match in the reference). **27 confirmed** as genuine order bugs and fixed via `semcor-fix-quote-order` / `src/semcor/quote-order-fixes.yaml`.
15 left unfixed for a follow-up:
- 5 are sentence-initial with no preceding word to reconstruct a target from
- the rest involve dialect apostrophes or other irregular preceding tokens that don't reconstruct cleanly
Also extended #67/#67-followup's single-quote-gap fix (`semcor-fix-single-quote-gap`): the scan had a blind spot where the word after a misattached opening quote is itself immediately followed by a flush contraction suffix rendered as one word in the reference (`You` + `'re` -> `You're`), e.g. `think,' You're` for Brown's `think, 'You're`. Extending the match target through flush-adjacent tokens found 5 more confirmed instances (120 total now, up from 115).
Both found while investigating a user-reported example in `data/press_reportage/br-a06.yaml` that combined one instance of each bug in the same sentence.
Fixes applied directly to `verify-brown-corpus` per the project's convention of not opening a PR per sub-issue of #5.
Contributor guide
No contributing guide indexed for this repository
Research direction
Start with verify-brown-corpus, semcor-fix-quote-order, and src/semcor/quote-order-fixes.yaml, then inspect the remaining 15 cases against src/semcor/brown-nolines.txt. Confirm each candidate with context-window matching and update the corpus fixes only when the reference proves the order or attachment is wrong; done means all remaining cases have a documented resolution.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- data, tooling
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Active
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100