New tools: UPOS+FEATS → XPOS and treebank scripts

Gist. After 2.18 two scripts by Zeman appeared in tools, and two new pages on the site:

• conllu_convert_uposf_to_xpos.pl computes XPOS in a given tagset from UPOS and FEATS via Interset. It is the inverse of conll_convert_tags_to_uposf.pl;
• survey_scripts.pl determines which script (ISO 15924) each treebank is written in. The result is docs/survey-scripts.md;
• docs/languages.md — a list of languages with script codes.

This closed #1032 «How to document writing system used in a treebank?» (closed 07.06.2026, milestone v2.19).

Level. Universal, tools and metadata. Does not affect annotation.

Evidence.

• diff -rq: tools/conllu_convert_uposf_to_xpos.pl and tools/survey_scripts.pl exist only in the snapshot;
• tools/README.md, section «conllu_convert_uposf_to_xpos.pl»: «takes UPOS and FEATS, and based on them it computes XPOS in a specific tagset … The target tagset is a tagset identifier known to Interset; the default value is cs::pdtc»;
• docs/survey-scripts.md: «automatically generated list of scripts that occur in the UD data».

Also in tools: package_ud_release.sh copies via rsync without __pycache__; survey_language_families.pl was updated; the snapshot has no compat/, conllu-formconvert.py, conllu-w2t.py, create_iso_639_3_symlinks.py.

What it means for English annotation. Directly — nothing. For Mova:

• Interset has an en::penn table. So conllu_convert_uposf_to_xpos.pl -t en::penn gives an independent check of UPOS+FEATS ↔ PTB pairs in our en, and also XPOS for treebanks without it. Using it requires Perl and Lingua::Interset. That is an external tool, so only on Mova's say-so;
• the script (Latn) will become a metadata field, and for the treebanks/<language>/… cards one more line.

Registry. Not applicable.

Sources. tools/conllu_convert_uposf_to_xpos.pl, tools/survey_scripts.pl, tools/README.md, docs/survey-scripts.md, docs/languages.md; #1032.

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/dialects/ud-next/seeds/tools-scripts/