docs: GLM-5.3-Flash attribution correction — wrong haraqat, not convention - #116
Merged
Conversation
…nvention 0.125pp The first reading (orthography-shaped) was wrong; the per-position decomposition with convention normalization and controls localizes the regression to wrong haraqat at ~4x GLM-5.2's rate — which matches our r7 teacher's to within 0.02pp. Missing 1.01%, extra 0.20%, the whole U+0670 effect 0.125pp (310 marks, zero in GT; controls <=0.009pp). Method note: apply the decomposition to any future LLM row before interpreting its DER.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Corrects the paper's protocol note and PUBLICATION-NOTES to match the measured attribution (rababa #69): wrong haraqat 10.05 percent of positions vs GLM-5.2's 2.64 (matching our r7 teacher's 2.62 to 0.02pp); missing 1.01, extra 0.20; the entire dagger-alif U+0670 convention effect is 0.125pp under position-derived rules with r7 and GLM-5.2 as <=0.009pp controls. Stronger dedicated-model finding: the newest generalist lost classical-Arabic mark knowledge its predecessor had. Method (decomposition + convention normalization with controls) noted as standard for any future LLM row.