discourse on adverbs

Gist. Already the 2.18 guideline admits into English discourse only non-adverbial discourse markers: well, like, but not actually. EWT has 9 adverbs with discourse, and GUM has 78. After 2.18 the exception for adverbs is written out explicitly: discourse: boundary by word class (universal, with an English example).

2.18 guideline. _en/dep/discourse.md refers to the universal page. There, 2.18:docs/_u-dep/discourse.md:11–12: «interjections (oh, uh-huh, Welcome), fillers (um, ah), and non-adverbial discourse markers (well, like, but not you know or actually)».

EWT 2.18 data (engine and grep):

• discourse in total — 1066: INTJ 775, SYM 123 (emoticons), NUM 113 (item numbering), NOUN 27 (thanks, ps), VERB 12 (mean, idk), ADV 9, PROPN 3, ADJ 3, X 1;
• ADV with discourse (train 4, dev 1, test 4): though ×3, So ×2, Also, Maybe, btw, FTW. For though this is an exception: ADV though is advmod 42 times and discourse 3;
• actually — advmod 83 times, discourse never; in other words — obl 5 times.

Other 2.18 treebanks. GUM: ADV with discourse — 78, of which so 57, anyways 6, now 5. At the same time so with advmod — 622. LinES — 15, ParTUT — 3, LittlePrince — 3, GENTLE — 1.

Where the discrepancy comes from. The boundary between a marker and an adverb is drawn by meaning, not by word class. After 2.18 the guideline explicitly gave priority to word class.

Sources. 2.18:docs/_u-dep/discourse.md, 2.18:docs/_en/dep/discourse.md.

rule: ud218.discourse-adv
what: ADV with discourse — 2.18 admits only non-adverbial markers
match: d[rel=discourse, upos=ADV]
require: not d[upos=ADV]
severity: warn
source: UD 2.18 docs/_u-dep/discourse.md:11–12

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/dialects/ud-2.18/seeds/discourse-adverbs/