Download examples/output_fix_map.examples.tsv from UyghurReferenceTexts/uyghur-normalizer: direct link, hf CLI and curl.
- Browser
- Download file 2.38 kB
-
https://huggingface.co/datasets/UyghurReferenceTexts/uyghur-normalizer/resolve/main/examples/output_fix_map.examples.tsv
- Command line
-
hf download hf://datasets/UyghurReferenceTexts/uyghur-normalizer/examples/output_fix_map.examples.tsv
-
curl -L -o output_fix_map.examples.tsv https://huggingface.co/datasets/UyghurReferenceTexts/uyghur-normalizer/resolve/main/examples/output_fix_map.examples.tsv
2.38 kB
| # output_fix_map.tsv — OUTPUT-SIDE ONLY. Never used by build_vNN.py. | |
| # | |
| # These are forms the MODEL INVENTS. They do not appear in the training corpus, | |
| # so they cannot be fixed by a build-side map and must not be added to | |
| # calque_map.tsv — doing so would make the corpus compliance check report | |
| # violations for forms that were never there. | |
| # | |
| # Established by the v13 sweep: a rule was in calque_map.tsv, corpus compliance | |
| # was 100.0% (zero occurrences), and every checkpoint still emitted the wrong | |
| # form anyway. See v13_provenance.md §3. | |
| # | |
| # 2026-08-17 — the Cold War rules were REMOVED from this file. All three forms | |
| # (سوغۇق ئۇرۇش، سوغۇق مۇناسىۋەتلەر ئۇرۇشى، سوغۇقچىلىق ئۇرۇشى) turned out to be | |
| # attested in published Uyghur; that is register variation among valid synonyms, | |
| # not a calque, and a normalizer has no business picking between them. Variant | |
| # preference belongs in the eval synonym map, not here. | |
| # | |
| # NOT loaded by default. Pass it explicitly: | |
| # python uyghur_normalize.py --in out.txt --fix-map normalizer/output_fix_map.tsv | |
| # | |
| # Format: variant<TAB>standard[<TAB>flags]. Longest LHS wins. Suffix-aware. | |
| # Applied AFTER calque/era/imla_loan, so right-hand sides must already be in | |
| # correct 2009/2011 imla — nothing runs after this pass to clean them. | |
| # | |
| # FLAGS | |
| # nosfx fire only on the bare form. Use when the replacement is ALREADY | |
| # inflected, so appending a captured suffix would double it. | |
| # | |
| # EVIDENCE COLUMN: only add a rule here after seeing the form in actual model | |
| # output. Do not add speculatively. | |
| # ---- fault line ---- | |
| # v13 ep8 -9.3 chrF. Model produces two competing literal calques. | |
| پارچىلىنىش لىنىيەسى ئۈزۈك سەپ | |
| پارچىلىنىش سىزىقى ئۈزۈك سەپ | |
| يېرىق سىزىقى ئۈزۈك سەپ | |
| # ---- nation state ---- | |
| # v13 ep8 -13.1 chrF, the single largest regression. Number-agreement error: | |
| # the model collapses the plural to a singular possessive compound. | |
| # NOTE: singular «مىللىي دۆلەت» is correct where the English IS singular. | |
| # This rule assumes the plural, which is the attested failure. If you see it | |
| # fire on a genuine singular, narrow it or remove it. | |
| # nosfx: the replacement is already plural; a captured suffix would stack on it. | |
| مىللەت دۆلىتى مىللىي دۆلەتلەر nosfx | |