ud-next — changes after UD 2.18 that will go into 2.19
The ud-next layer of the mova dialect (The Mova dialect: a UD metadialect, treebanks, and converters). With it, future 2.19 changes can already be used.
What was compared.
- Release 2.18 (15.05.2026):
markdown-source/from ud-documentation-v2.18.tgz and ud-tools-v2.18.tgz. - A snapshot of the
docsandtoolsrepositories from 24.09.2026 (shallow clone, no history): {docs,tools}. Thedocstracker — issues: 1000 issues and PRs (#296–#1305), bodies only, without comments. - There are no treebank branches in the snapshot: whether EWT
devalready follows the new guidelines is not visible from these data.
How it was compared.
diff -r -I 'Interlanguage links updated': without this timestamp, which a script rewrites in every language-guideline file, 4680 files would look changed;tools/data/*.jsonwas expanded into "path → value" lines and compared withdiff;udtoolsandvalidate.pywere compared selectively;- numbers for EWT 2.18 — the
en expert-checkengine with counter rules from the seeds, some by grep.
Path notation. docs/… and tools/… — the snapshot. 2.18:docs/… and 2.18:tools/… — the 2.18 archives. #N — an issue or PR in docs.
In brief
- Changes after 2.18 — 11 groups.
- Universal, 6: clarification No. 18 on features and
Exponence; a new definition ofdiscourse; a new validator check; feature layers;expl:*pages; validator data. - English, 2: case of who pronouns; the treebank list.
- Tools — 1.
- Other languages and the site — 2 seeds, one of them on the French subtypes as a borrowing candidate.
- Universal, 6: clarification No. 18 on features and
- Two change English annotation:
- who/whom/whose get
Case, whom gets the lemma who: 24 lemmas and 460 features in EWT 2.18; - adverbs are no longer
discourse: 9 EWT tokens. Two more only legitimize or protect what EWT already does: features from context (No. 18) and non-empty XPOS.
- who/whom/whose get
- Registry.
en/data/ud-registry-en.tsvmatches the validator data of both 2.18 and the snapshot: nothing changed for en (Validator data tools/data/.json: English unchanged). This is the state of 2.18 and also of the 24.09 snapshot. It cannot yet be called the 2.19 state: the data freeze is 01.11.2026, the release is 15.11.2026. - The converter
mova⇄ud-nextis in Converter mova ⇄ ud-next, run with the real engineen convert/en convert-check. The forward one changes 469 words on EWT, the reverse one restores everything except 9discourseon adverbs.
Changes
| seed | level | type | what changed | for English annotation | EWT 2.18 | in the registry |
|---|---|---|---|---|---|---|
| en-wh-case | en | guideline _en/pos/PRON.md | who Case=Nom|Acc by context, whom → lemma who + Acc, whose Case=Gen with a noun; 2.18 TODO closed | yes: lemma and FEATS | whom with lemma whom 23 (+1 whomever); Case on who pronouns 0 of 447; whose without Gen 13 | yes: PRON Case=*, Poss=Yes |
| discourse-scope | universal (en — explicitly) | guideline _u-dep/discourse.md | discourse — by grammatical features; ADV → advmod, prepositional phrase → obl; emoticons — discourse | yes: DEPREL | ADV with discourse 9 (GUM 78); emoticons 123 | yes |
| features-context | universal | CLARIFICATION No. 18, 2.19 | feature value — from context, but the form wins; Exponence in MISC | no: EWT already does this | VBD/VBP get Number/Person from the subject; you/it Case by position; Exponence 0 | not applicable |
| xpos-not-empty | universal | validator udtools 0.2.8 | error empty-string-in-xpos | export gate | empty XPOS 0 | not applicable |
| feat-layer-ids | universal | guideline | a layer identifier — lowercase Latin letters only, now a rule | no | no layers | not applicable |
| expl-pass-pv-docs | universal | documentation | the expl:pass/expl:pv pages changed from Czech to universal | no | — | not present for en |
| validator-data | universal | validator data | 48 languages: 18 new, 7 got new entries, 23 changed; en unchanged | no | — | = 2.18 = snapshot |
| tools-scripts | universal | tools | conllu_convert_uposf_to_xpos.pl, survey_scripts.pl, pages on script (#1032) | no | — | — |
| en-treebank-list | en | documentation | 13 active treebanks (#1297) | no | — | — |
| fr-spoken-subtypes | fr | guideline | discourse:filler, discourse:tag, conj:reform | borrowing candidate | — | not present for en |
| other-languages | other languages, site | various | Armenian (flat:redup), oge, ka, lt, el, es, egy, ru, ps, ta…; new languages; site | no | — | — |
The rules in the seeds en-wh-case, discourse-scope and features-context check annotation for conformity with ud-next. On EWT 2.18 the error rules give 24 + 423 + 13 + 9 violations, and this is the negative control. On EWT after the mova → ud-next converter and on a manual example with correct annotation — 0:
target/release/en expert-check dialects/ud-next/seeds \
$MOVA_DATA/en/ud/ud-treebanks-v2.18/UD_English-EWT/en_ewt-ud-{train,dev,test}.conllu
Proposals for mova on each change are in dialects/mova/.
Open discussions where decisions are expected before 2.19
About the milestone. 129 issues have the v2.19 milestone: 8 closed, 121 open. For 73 of the open ones the last update is 22.05.2026, and they were created before 2.18. This is a mass carry-over from v2.18, so the milestone by itself promises nothing. Activity after 22.05 (or creation after the 2.18 release) — in 47. Below are those of them that touch English or universal rules. The dump contains no decisions, because it has no comments.
| # | topic | activity | what it touches in English (EWT 2.18) | for mova |
|---|---|---|---|---|
| #1118 | Mood for modal AUX in English (Pot/Nec? will — Irr or Tense=Fut?) | 45 comments, 27.05 | MD with VerbForm=Fin without Mood — 4049; warning verbform-fin-without-mood (Modal MD: VerbForm=Fin without Mood) | wait |
| #1155 | restrict this warning to VERB | 11, 22.05 | the same 4049 MD (AUX) | wait |
| #1233 | when to put "lexical" features (examples from EWT and GUM) | 20, 07.07 | modals with VerbForm=Fin; a/each/every/another/both without Number (5348 a); Exponence — partly answered by No. 18 | wait |
| #1252 | ExtPos in FEATS or in MISC; PROPN boundaries | 32, 11.06 | EWT writes ExtPos (7 values in the registry); moving it to MISC would change the registry and Feat in en | wait |
| #1176 | dependents of a nominalized adjective — nmod, not obl («decided in a Core Group conversation») | 0, 22.05 | ADJ in nominal position without cop: with nmod 83, with obl 53 (several of these children) | wait for the text; the candidate rule obl → nmod is simple |
| #1274 | on what: case or mark | 7, 10.07 | by a rough counter, a preposition as mark before a wh-clause: EWT 69, GUM 17; as case on a wh-word: EWT 92, GUM 246 | wait |
| #1248 | both X and Y, tanto … como | 9, 25.05 | EWT: both — CCONJ cc:preconj 58, the rest DET or ADV; and — CCONJ | wait |
| #1254 | participles in adjective position: ADJ or VERB | 5, 10.06 | in EWT the boundary is drawn by the PTB tag: VBN and VBG are always VERB or AUX (VBN with amod — 585), lexicalized -ed forms have JJ and ADJ (739) | wait |
| #776 | Lang= for code-switching | 26, 24.05 | EWT: Foreign=Yes — 61 tokens; flat:foreign — 0 | wait |
| #1273, #1280, #1289, #1290, #1300, #1282 | UniDive, spoken language: segmentation, co-constructions of several speakers, tokenization and morphology, parataxis, tag questions, backchannels, metadata | 07–09.2026 | ESLSpok, the spoken part of GUM, CHILDES; see French subtypes for spoken language (another language, a borrowing candidate) | wait; the guideline fork is not merged yet |
| #1291 | one language — different scripts | 1, 13.08 | not applicable | — |
English issues in milestone v2.19 without activity after 22.05 (16 with the lang:English label, activity only in #1118 and #1274):
- #308 tough;
- #455 dates, times, addresses;
- #647
aux:passfor get; - #658 coordination of auxiliaries;
- #864 reported speech;
- #876 EUD;
- #923 be easy to VERB;
- #1009 independent possessives;
- #1040 names and numbers;
- #1070 Espresso;
- #1086 everyone at X;
- #1091 all too;
- #1099 internet slang;
- #1176 (above).
Related universal ones:
- #1192 relative that: PRON or SCONJ, 14 comments;
- #1208 days and months: PROPN or NOUN, 68 comments;
- #1024 clausal
appos, 36; - #1066 subjects with non-clausal relations, 35;
- #1064
nmod:possanddet:poss: a proposal to forbiddet:possin the validator. Matches our normalizationdet:poss→nmod:poss.
Closed in milestone v2.19 (8):
- #1032 — treebank script, see
tools-scripts.md; - #1250 — Thai verbal clauses without conjunctions;
- #1270, #1279, #1286, #1287, #1299, #1301 — the site and Middle Persian.
Outside the milestone, PR #1297 (the list of English treebanks, en-treebank-list.md) and #1302 (EUD tools) were closed. Neither of them changes English annotation.
This page in Markdown → https://md.mova.elamur.ai/free/en/docs/dialects/ud-next/