ParTUT — known annotation errors

Gist. ParTUT is an automatic conversion from TUT via USD with manual corrections. The last corrections were in 2.5. So alongside the Italian customs there are traces of the conversion: lemmas, labels and features that match neither EWT nor ParTUT itself. The converter does not fix them.

Examples and numbers.

• Pronoun lemmas in lowercase and in the oblique case: me → i (9, lowercase), us → us (28), them → them (1), themselves → they (5), our/ours → us (60 DET + 2 PRON). EWT: I, we, they, themselves, our.
• An adjacent noun as obj of the head noun — 15 (ParTUT — a noun before a noun — nmod, not compound): a noun does not take obj.
• Possessive PRON among DET — 2 of 617: an inconsistency in the custom (ParTUT — possessives as DET, by without obl:agent, flat:foreign, when as mark).
• Number on PROPN — only 82 of 2230: some proper nouns have Number, the rest do not.
• Passive subject as nsubj (tb.en.pass-subj) — 1 of 453.
• obl on a noun without a copula — 0 of 11. Here ParTUT already agrees with UD 2.18.

In UD. For merging, ParTUT's main problem is not errors but customs: nmod in compound nouns (862), features without Case and Voice, possessive DET. The errors (≈ 120) are within the noise.

Sources. README UD_English-ParTUT (changelog: last corrections — 2.5); rules in English UD 2.18 treebanks — comparative seed.

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/treebanks/en/partut/seeds/errors/