discourse only for interjections, particles, numbering and emoticons — proposal
Proposal: accept the new definition of discourse (discourse: boundary by word class (universal, with an English example)). Do not introduce MISC marks for the 9 EWT adverbs.
What exactly in mova.
- ADV is never
discourse, butadvmod: though, so, also, maybe, btw; - a prepositional phrase is never
discourse, butobl(in other words); discourseremains for INTJ (oh, well, like, um, yes), item numbering (NUM, 1., (a)), emoticons (SYM), thanks, ps.
Why.
- More consistent. The criterion is the word class, visible without interpreting pragmatics. In EWT though is
advmod42 times anddiscourse3; actually isadvmod83 times out of 83. In GUM so isdiscourse57 times andadvmod622. - Lossless into the standard. The 2.18 guideline already excluded adverbs ("non-adverbial discourse markers", discourse on adverbs). So
advmodis the correct export both in 2.18 and inud-next. - What we pay. The reverse run of EWT
mova → 2.18diverges from gold in 9 tokens (train 4, dev 1, test 4), in GUM — 78. These are divergences from gold errors, not from the guideline, so it is not worth preserving them with a MISC mark. In the reverse-run report they should be shown as a separate line.
Evidence from the data (EWT 2.18): discourse — 1066: INTJ 775, SYM 123, NUM 113, NOUN 27, VERB 12, ADV 9, PROPN 3, ADJ 3, X 1. Prepositional phrases with discourse — 0.
Converters.
- EWT, GUM →
mova: rulemova.ud-next.discourse-adv(Converter mova ⇄ ud-next); mova→ 2.18 /ud-next: no change;- DEPS: in EWT
N:discoursewill remain. An action for DEPS or an EUD rebuild is needed (seeconvert.md).
Check. Gates udnext.discourse-not-adv and udnext.discourse-not-pp (error).
Origin. discourse: boundary by word class (universal, with an English example), discourse on adverbs.
Mova decides.
This page in Markdown → https://md.mova.elamur.ai/free/en/docs/dialects/mova/seeds/discourse-adverbs/