Features from context and Exponence in MISC (universal, 2.19)

Gist. A new clarification for 2.19 — No. 18, CLARIFICATION, June 2026:

1. Feature values are taken from context when the form itself does not show them, but words of this class usually express the feature. Examples are syncretism and agreement.
2. If the form and the context give different values, as with notional agreement, the form wins.
3. How exactly a feature is expressed — by inflection, context or lemma — a treebank may indicate in MISC with the Exponence attribute, for example Exponence[Number]=Inherent. The value set is not standardized.
4. The validator will not change.

Level. Universal; changes.md, type CLARIFICATION, in force from 2.19.

Before (2.18). The general rule docs/_u-overview/morphology.md:112: a feature value may depend «on the context of the given sentence where the word is used». There was nothing about conflicts between form and context or about Exponence.

Now (snapshot 24.09.2026). docs/changes.md:27 (table row «18 | 2026-Jun | 2.19 | CLARIFICATION | Morphosyntactic Features») and :46–67:

• «Context should be used to disambiguate the feature value when the wordform itself is underspecified with respect to the feature, but that feature is characteristically expressed by words in the same class.»;
• «(If wordform and context suggest conflicting values of a feature, e.g. with notional agreement, the wordform-based feature is to be preferred.) This was affirmed unanimously by the Core Group.»;
• «We recommend Exponence as the name for the MISC attribute, as already piloted in French … We are not prepared to standardize the set of values»;
• «There are currently no plans for validator changes arising from these clarifications.»

Evidence. diff 2.18:docs/changes.md ↔ docs/changes.md. Discussions:

• #985 — Exponence, not in the dump;
• #1233 «When do we need to instantiate a (lexical) morphosyntactic feature?» (Kahane, open, updated 07.07.2026). The examples there are from EWT and GUM: modals with an "inherent" VerbForm=Fin; a, each, every without Number, although demonstratives have it;
• the paper Kahane et al. 2025, 2025.tlt-1.18.

What it means for English annotation. EWT 2.18 already annotates this way, i.e. the clarification legitimizes existing practice:

• a finite verb takes Number/Person from the subject where the form does not show them. VBD: 3499 Sing|3, 1262 Sing|1, 1120 Plur|3…; VBP: 2515 Plur|3, 2440 Sing|1, 1192 Sing|2…;
• Case on you and it — from position: you Nom 2096 / Acc 651, it Nom 1492 / Acc 784;
• the form wins. VBZ has Sing|3 5655 times of 5656; the exception is the token s with empty FEATS (newsgroup-groups.google.com_AgeingMonkeys_37131d1864a0b950_ENG_20051114_080100-0001). Was is always Sing. There is no VBP with Sing|3 in the basic tree. There is only the EUD empty node 16.1 have in answers-20111108082831AAco5PI_ans-0002; the engine does not read empty nodes;
• EWT has no Exponence. #1233 remains open: whether to put "lexical" features on a/each/every (Number) and on modals (VerbForm, Mood, see Modal MD: VerbForm=Fin without Mood).

For mova this confirms our own "FEATS from the subject" normalization (en::ud::agreement, The Mova dialect: a UD metadialect, treebanks, and converters). At the same time it sets a boundary for it: where the form is unambiguous, agreement does not override it (the team are → Plur).

Registry. Unchanged (Exponence lives in MISC; the FEATS registry does not cover it).

Sources. docs/changes.md#morphosyntactic-features; #1233; 2025.tlt-1.18; the UD survey digest «future», section 2.2 («Exponence (from 2.19, per #985)»).

rule: udnext.vbp-not-3sg
what: VBP is never 3rd person singular — in a conflict of form and context the form wins
match: v[xpos=VBP]
require: not v[feats.Number=Sing, feats.Person=3]
severity: error
source: UD docs/changes.md:53–56 (No. 18, 2.19)

rule: udnext.vbz-3sg
what: VBZ — always 3rd person singular, even with notional agreement
match: v[xpos=VBZ, !feats.Typo]
require: v[feats.Number=Sing, feats.Person=3]
severity: error
source: UD docs/changes.md:53–56 (No. 18, 2.19)

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