French subtypes for spoken language (another language, a borrowing candidate)

Gist. After 2.18, French documents three subtypes for spoken treebanks:

• discourse:filler — a filled pause (euh ≈ um);
• discourse:tag — a tag question (…, hein ?, …, non ?, tu t'en souviens ?);
• conj:reform — a reformulation of an already completed utterance: une maison en banlieue, la banlieue parisienne. An incomplete utterance is, as before, reparandum.

Level. Language-specific (fr). For English this is a candidate for borrowing into mova. A related open universal discussion is #1290 (the UniDive spoken-language working group: parataxis, tag questions, backchannels).

Now (snapshot 24.09.2026). New files docs/_fr/dep/discourse-filler.md, docs/_fr/dep/discourse-tag.md, docs/_fr/dep/conj-reform.md (not in 2.18). Changed docs/_fr/dep/dislocated-mod.md (only a typo in an example).

Evidence. diff -rq, «Only in docs/_fr/dep». In tools/data/deprels.json the permitted relations for fr changed (see Validator data tools/data/.json: English unchanged).

What it means for English annotation.

• EWT has no such distinction. Fillers (um, uh) are INTJ with discourse, together with oh, yes, wow: INTJ with discourse — 775. A tag question is rare in EWT: grep found one, …, aren't they? (train), and it is parataxis;
• for spoken English treebanks (ESLSpok, the spoken part of GUM, CHILDES) the subtypes would make it possible to distinguish a filler from an interjection and a tag from a parenthetical clause;
• the en registry has no such subtypes, so they must be dropped on export to UD. A subtype can be dropped without loss: discourse:filler → discourse.

Registry. Not allowed for en; allowed for fr in the snapshot.

Sources. docs/_fr/dep/*.md; #1290 (grew.fr/spoken-language-guidelines/workgroups/spoken-data/specific_syntax.html), #1280 (co-constructions by several speakers), #1273 (segmentation of spoken language).

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