mova — Mova's dialect
Mova has its own dialect based on the latest UD at any given time — so future 2.19 changes can already be used — and also based on alternative UD dialects, whose changes are pulled into the dialect when they are judged good.
What mova is (details in The Mova dialect: a UD metadialect, treebanks, and converters):
- The internal dialect into which all treebanks and representations are converted: UD versions, the customs of individual treebanks, SUD, SD, classic PTB conversions. The
enannotator learns on it. From it we convert to any target, and quality is measured on each treebank's gold data through the reverse converter. - Base — UD 2.18 with EWT conventions (registry
en/data/ud-registry-en.tsv). - Layers on top of the base:
ud-next— changes after 2.18 that will go into 2.19 (dialects/ud-next/);- borrowings from alternative dialects;
- own normalizations:
det:poss→nmod:poss,:tmod/:npmod→:unmarked, FEATS from the subject.
- Selection rule: take whatever is more consistent and more informative and can be converted to the standard without loss. A lossy target is only an export target, never the base.
This folder holds the decisions on ud-next changes and on discrepancies between the 2.18 guidelines and the EWT data, each as a separate seed in seeds/. Decided 2026-09-26: every proposal in the table is taken as proposed. Accepted: 11; "wait" — watched until the 2.19 data (freeze 01.11.2026); "do not accept" — no. Principle: changes from other dialects are pulled into this one when they are judged good.
Decisions
| topic | origin | decision | why, briefly | effect on export |
|---|---|---|---|---|
| wh-case | ud-next, _en/pos/PRON.md | accept | case shows the role of who in the clause; lemma who — as in GUM, GENTLE, LinES, ParTUT | in 2.18 — drop Case, lemma from form; identity on EWT |
| discourse-adverbs | ud-next _u-dep/discourse.md; 2.18 discrepancy | accept | the boundary is by word class; 2.18 already excluded ADV | 9 EWT tokens (GUM — 78) will diverge from gold, but not from the guideline |
| features-context | ud-next, changes.md No. 18 | accept the principle; wait on Exponence | EWT already does this; form wins over semantic agreement; Exponence values do not exist yet | no change |
| xpos-not-empty | ud-next, validator 0.2.8 | accept | otherwise a 2.19 validator error | writer gate |
| feat-layer-ids | ud-next, feat-layers.md | accept | a layer is only [a-z]+ | does not apply to en |
| spoken-subtypes | ud-next, fr; #1290 | wait | no universal decision and no English gold data | the subtype is dropped without loss |
| no-effect | ud-next: expl:*, validator data, tools, treebank list, other languages; 2.18: v1 labels | accept as is / not applicable; do not take uposf→xpos into the pipeline | no effect on English annotation | — |
| iobj-sole | 2.18 discrepancy | accept the 2.12 amendment, as EWT does | iobj with ask/tell/allow X to is more informative than obj; the lexicon decides the boundary | identity |
| to-phrase-obl | 2.18 discrepancy | accept obl | 1285 : 0 in EWT; nmod in the text is a v1 leftover | identity |
| imperative-do | 2.18 discrepancy | accept the EWT custom (Imp|Fin) | the same for Go! and Don't go!; 98 of 106 | 8 EWT exceptions will diverge; the letter of VerbForm.md is derived deterministically |
| subjunctive | 2.18 discrepancy | accept Mood=Sub | a clause feature; 57 in EWT, present in GUM and ParTUT | identity; the letter of Mood.md is derived deterministically |
| modal-mood | 2.18 discrepancy; #1118 | wait | Mood=Pot/Nec is outside the en registry; no decision | no change |
| depictives | 2.18 discrepancy | accept advcl | one answer instead of three; 22 in EWT | identity |
| clausal-predicate | 2.18 discrepancy | accept nsubj:outer | 257 in EWT, 0 of the old analysis | identity |
Summary:
- accept — 11;
- "as is" or "not applicable" — 5 (in
no-effect); - wait — 3:
Exponence, spoken-language subtypes,Moodof modals; - do not accept — 1: the
uposf→xposscript in the pipeline.
Only 3 proposals change mova annotation compared with EWT 2.18:
- who pronouns: 24 lemmas, 460 features;
- adverbs with
discourse: 9 tokens; - imperatives with
Inf: 8 tokens.
The rest confirm that mova follows the EWT data where the _en text is out of date.
Rules for the first two are in Converter mova ⇄ ud-next. They were run with the en convert engine on EWT 2.18: the forward one changes 469 words, the reverse one restores everything except 9 discourse. The rule for the third is in Main verb with imperative do — Mood=Imp|VerbForm=Fin — proposal, for EWT ↔ mova — converter; the engine changes exactly 8 words with it.
Outside the decisions, under watch. #1176: dependents of a nominalized adjective are nmod, not obl. The Core Group decision has been declared but not yet written into the guidelines. In EWT 2.18, ADJ in nominal position has nmod 83 times and obl 53. When it lands in docs/, it will be one more ud-next change with a simple rule obl → nmod. Other open discussions are in ud-next — changes after UD 2.18 that will go into 2.19.
This page in Markdown → https://md.mova.elamur.ai/free/en/docs/dialects/mova/