Summary

The Hungarian UD treebank is derived from the Szeged Dependency Treebank (Vincze et al. 2010).

Introduction

The Hungarian UD treebank is derived from the Szeged Dependency Treebank (Vincze et al. 2010). One of its newspaper sections (Népszava) has been semi-automatically converted to the Universal Dependencies scheme, and linguists have corrected the annotation manually. Hence, it contains texts from the newspaper domain with various topics (politics, economics, sport, culture etc.).

Sentences 1-129 (i.e. 10% of the sentences) form the development set, then sentences 130-1162 (80%) form the training set and finally sentences 1163-1299 (10%) form the test set.

In version 1.3, 500 sentences from the HVG subcorpus of the Szeged Treebank were added to the data, in harmony with the above principles.

Statistical Data

hu-ud-dev.conllu Tree count: 441 Word count: 11418 Token count: 11418 Dep. relations: 48 of which 21 language specific POS tags: 16 Category=value feature pairs: 69

hu-ud-train.conllu Tree count: 910 Word count: 20166 Token count: 20166 Dep. relations: 54 of which 23 language specific POS tags: 16 Category=value feature pairs: 70

hu-ud-test.conllu Tree count: 449 Word count: 10448 Token count: 10448 Dep. relations: 47 of which 20 language specific POS tags: 15 Category=value feature pairs: 68

References

The original Szeged Dependency Treebank is described in the following paper:

Vincze, Veronika; Szauter, Dóra; Almási, Attila; Móra, György; Alexin, Zoltán; Csirik, János 2010: Hungarian Dependency Treebank. In: Proceedings of the Seventh Conference on International Language Resources and Evaluation (LREC'10), Valletta, Malta.

The detailed annotation principles of the Hungarian UD treebank will be discussed in a forthcoming paper.

Contact

The UD version of the treebank has been prepared by Richárd Farkas (rfarkas@inf.u-szeged.hu) and Veronika Vincze (vinczev@inf.u-szeged.hu). For questions and comments please contact them.

Changelog

2022-11-15 v2.11
- The lemma of "volna" is "van".
- Fixed: spurious copulas.
- Fixed: case/mark/cc/aux dependents should be leaves.
- Fixed: only the first element of a goeswith sequence should have morphological annotation.
- Fixed: clause has at most one subject.
- Fixed: structure of orphans in gapping.
- Fixed: pronouns, nouns, verbs, adpositions cannot be advmod.
- Fixed: appos must go left to right.
2022-05-15 v2.10
- Case=Sub now means subessive in UD; sublative relabeled to Case=Sbl.
2021-05-15 v2.8
- Removed superfluous features Number[psor]=None, Number[psed]=None, Person[psor]=None.
- The relations nmod:obl and amod:obl changed to just obl under a VERB, ADJ or ADV.
- The relation nmod:obllvc changed to obl:lvc.
- The relation amod:mode changed to advmod:mode under a VERB, ADJ or ADV (but the UPOS tag of the dependent is still ADJ).
- The undocumented relation amod:obl changed to amod:att (4 occurrences).
- Fixed non-projective punctuation with udapy -s ud.FixPunct.
2018-04-15 v2.2
- Repository renamed from UD_Hungarian to UD_Hungarian-Szeged.
2017-03-01 v2.0
- Data re-split so that there is 10K dev and 10K test for the CoNLL shared task.
- Converted to follow the UD v2 guidelines.
2016-05-15 v1.3
- 500 new sentences added.
2015-11-15 v1.2
- Empty morphological features have been removed.
- Adverbial participles have been annotated as VerbForm=Trans.
- Format of morphological features with more than one values has been corrected.
2015-05-15 v1.1
- Sentences with multiple roots have been corrected.
2015-01-15 v1.0
- First release in UD