Features from context, form wins; `Exponence` — proposal

Proposal:

accept the principle of clarification No. 18: feature values come from context, and in a conflict the form wins;
wait on Exponence in MISC.

What exactly in mova.

keep the "FEATS from the subject" normalization (en::ud::agreement, The Mova dialect: a UD metadialect, treebanks, and converters) and give it a boundary: where the form is unambiguous, agreement does not override it. VBZ is always Sing|3. VBP is never Sing|3: the team are → Plur|3. Was is always Sing;
put gates udnext.vbz-3sg and udnext.vbp-not-3sg on the en output and on silver data.

Why accept.

This is already EWT practice:
- VBD and VBP get Number/Person from the subject (VBD Sing|3 3499, VBP Plur|3 2515…);
- you and it get Case by position;
- VBZ Sing|3 — 5655 of 5656 (the exception is a token with empty FEATS);
- there is no VBP Sing|3 in the basic tree.
Lossless: export to 2.18 changes nothing.
What it guards against: the subject normalization must not override the form under semantic agreement (the team are, a number of people are). Otherwise it would produce a feature that is not in the standard.

Why wait on Exponence.

There is no value set. Clarification No. 18 says directly: "We are not prepared to standardize the set of values without further community experimentation". There is only a pilot in French (Exponence[Number]=Inherent).
#1233 is still open: "lexical" features on a/each/every and on modals.
For English one split would be useful: Number/Person on VBD/VBP — from context, on VBZ — from the form. But it is computable from XPOS, so there is no need to write it into MISC until there is a standard.

Check. udnext.vbz-3sg, udnext.vbp-not-3sg (Features from context and Exponence in MISC (universal, 2.19)). On EWT 2.18: 1 and 0 violations.

Origin. Features from context and Exponence in MISC (universal, 2.19); docs/changes.md#morphosyntactic-features; #1233.

Mova decides.

Features from context, form wins; Exponence — proposal