Treebanks
Sentences annotated in Universal Dependencies (CoNLL-U format: words, lemmas, tags, morphological features, dependency tree).
| File | Sentences | Texts | Annotation | License |
|---|---|---|---|---|
mova-silver-en.conllu | 348 | Tatoeba sentences under CC0 (# source says which), example sentences from Mova seeds, and 6 sentences from ACL Anthology papers under CC BY 4.0 | parsed by Mova, checked and corrected by a large language model, with UD validation gates | annotation ours, CC BY 4.0 (because of the ACL sentences; attribution in # source) |
mova-bronze-en.conllu | 326 | example sentences from Mova seeds | parsed by Mova only, not checked | ours, Elamur Cosmic License |
mova-tales-silver-en.conllu (new in 0.3) | 2,087 | sentences of 41 fairy-tale books from Project Gutenberg (public domain in the USA; book id in # sent_id) | parsed by Mova, patched by a large language model (Claude Opus, medium effort), UD 2.18 validation gates | annotation ours, Elamur Cosmic License |
eslspok-mova/ | 2,320 (train 1,856, dev 232, test 232) | UD English ESLSpok | the original with Mova's dialect rules applied: 9 rules changed 523 of 21,312 words | CC BY-SA 4.0, as the original |
How Mova uses it: en trains and tests its tagger and parser on UD
treebanks; the silver and bronze sets test that the grammar seeds really parse
the way they describe. The training-data gate (en::license::commercial_gate)
only accepts sources whose license allows any purpose.
The tales split (how Mova's tales numbers are measured): world conllu-split mova-tales-silver-en.conllu even.conllu odd.conllu — the odd half (1,043 sentences) is the test set, the even half (1,044) is for development;
Mova's self-training splits the even half further and never reads the odd one. LAS on the odd half: 76.85 in 0.2,
77.38 in 0.3.
This page in Markdown → https://md.mova.elamur.ai/free/en/docs/corpus/open/0.3/treebanks-README/