UD_German-GSD
Treebank card from UD release 2.18: metadata and texts come from UD_German-GSD/README.md; size is counted by streaming the .conllu files and checked against the release stats.xml. Generated by tools/treebanks build; do not edit by hand. Papers: Papers: UD_German-GSD; seeds: seeds/; converter: convert.md.
| Field | Value |
|---|---|
| UD name | UD_German-GSD |
| Language | German (de) |
| Family | IE |
| Treebank code | de_gsd |
| UD page | universaldependencies.org/treebanks/de gsd/index.html |
| License | CC BY-SA 4.0 |
| Genres | news, reviews, wiki |
| Text in data | yes |
| Parallel | no |
| In UD since | UD v1.0 |
| Authors | Slav Petrov; Wolfgang Seeker; Ryan McDonald; Joakim Nivre; Daniel Zeman; Adriane Boyd; Verena Blaschke |
| Contact | zeman@ufal.mff.cuni.cz |
| Development | in the UD repo (here) |
Annotation layers
| Layer | Origin |
|---|---|
| Lemmas | automatic |
| UPOS | converted from manual |
| XPOS | automatic |
| Features | automatic |
| Relations | converted from manual |
UD scale: manual native — annotated by hand directly in UD; converted from manual — converted from manual annotation in another scheme; converted with corrections — converted and partly corrected by hand; automatic with corrections — automatic, with partial manual corrections; automatic — automatic; not available — the layer is absent.
Size
| Split | Sentences | Tokens | Words | Multiword tokens | Empty nodes |
|---|---|---|---|---|---|
| train | 13 813 | 259 167 | 263 777 | 4 610 | 0 |
| dev | 799 | 12 316 | 12 480 | 164 | 0 |
| test | 977 | 16 225 | 16 499 | 274 | 0 |
| total | 15 589 | 287 708 | 292 756 | 5 048 | 0 |
Tokens are surface tokens (Spanish vámonos is one), words are syntactic words (vamos + nos); multiword tokens are those split into several words; empty nodes are the N.M nodes of the enhanced graph.
Summary
The German UD is converted from the content head version of the universal dependency treebank v2.0 (legacy).
Introduction
The ACL 2013 paper (github.com/ryanmcd/uni-dep-tb/blob/master/ACL2013.pdf, McDonald et al.) describes version 1.0 of the corpus, of which there are 2200 train/800 dev/1000 test sentences in German. According to the paper they consist of Reviews and News genres (the news data being from the TIGER Treebank, Reviews presumably from Google).
The subsequent 2.0 release has more data: 14118 train/799 dev/977 test sentences. Some of the sentences in 1.0 turned out to be duplicated across splits, which was fixed for 2.0. There is no indication in the READMEs of where the new German sentences came from.
Based on the above and the mappings in not-to-release/ud-tiger-mapping.txt, it appears that the genres are:
train: Reviews=s1-s1500, News=s1501-s2200, Web=s2201-s14118 By searching for a selection of sentences in the s2201-s14118 range, i.e. the new ones in version 2.0, it looks like they are from Wikipedia and other websites. dev: Reviews=s1-s500, News=s501-s799 test: Reviews=s1-s301, News=s302-s977
This page in Markdown → https://md.mova.elamur.ai/free/en/docs/treebanks/de/gsd/