harisont / L2-UD

Tools for working with UD treebanks of learner texts (and parallel treebanks in general).

harisont/L2-UD Issues

Reconcile 2l-ud branch with main as much as possible
Updated 15 days ago
Replace gf-ud with deptreehs
Updated 3 months ago1
Hit background should be set row-wise, not cell-wise
Closed 9 months ago1
Background for nonexisting files turns red with Dark Reader enabled, but not otherwise
Closed 9 months ago
Implement simplification that only removes content words forms and/or lemmas
Updated 6 months ago
Do not extract errors whose simplified patterns end up being the same
Closed a year ago
Improve extraction when the problem is a missing token
Updated 15 days ago
Simplify away identical subtrees
Closed a year ago
Simplification of sequence patterns (WO errors)
Updated a year ago1
Prioritize most similar examples
Closed a year ago1
Figure out where `{AND [] -> AND []}` patterns come from
Updated a year ago1
Pattern simplification by keeping only erroneous CoNLL-U fields
Closed a year ago1
Implement basic queries by example
Closed a year ago1
Extracted pattern simplification by removing all fields that are identical both in the L1 and in the L2 part
Closed a year ago1
Incorrect punctuation currently treated as morphosyn error
Closed a year ago
Variables optimization
Updated a year ago
Fix `--conllu` option
Closed a year ago
Update README
Closed a year ago
Query preprocessor does not accept spaces before/after the arrow
Closed a year ago1
Make it so that patterns with no replacement are considered to be L2-only patterns
Closed a year ago
Implement variables
Closed a year ago1
(Re-)implement CoNNL-U (subtree) output
Closed a year ago1
Only highlight part of the sentence matching the pattern, rather than the whole matching subtree
Closed 2 years ago
False positives with `TREE` and `SEQUENCE` patterns
Closed a year ago5
`smallest` should be used in `l2-ud match` too
Closed 2 years ago
Fix linearized output when tokenization divides words
Closed a year ago1
Consider morphosyntactic error only alignments where DEPREL and/or UPOS and/or POS and/or FEATS change but the lemmas (of the content words) are the same
Updated a year ago1
Pruning should be recursive
Closed a year ago
Discrepancy-aware pruning does not take missing subtrees into account
Closed a year ago
When searching for the smallest subtree for each error, there is no check that it includes the "whole" error
Updated a year ago
Patterns are extracted even when identical aligned subtrees only differ in `ID` and/or `HEAD`
Closed a year ago3
No `SEQUENCE`s in extracted patterns
Closed a year ago1
Word order errors not always detected
Closed 2 years ago1
Allow using language-specific POS
Closed 2 years ago
Allow retreival of full sentences with errors highlighted when querying
Closed 2 years ago
False positives in OR queries with two "specular" replacements
Closed 2 years ago1
No parse for some queries where the error is a missing token
Closed a year ago1