Polish Dependency Bank 2.0 (PDB 2.0)

PDB 2.0 is an extended version of Składnica zależnościowa (the first Polish dependency treebank). It consists of 22,152 trees and 350,001 tokens (i.e. 15.8 tokens per sentence on average). There are four parts of PDB 2.0:

1. NKJP1M-based trees (14K)
2. Projection-based trees (4K)
3. CDScorpus-based trees (2K)
4. OTHER trees (2K)

The PDB sentences contain some problematic linguistic phenomena, e.g. ellipsis, comparative constructions, constructions with the bi-functional subordinating conjunction JAKO, directed speech, interpolations and comments, nominative noun phrases used in the vocative function and many others. The Polish dependency relation types are as follows: abbrev_punct, adjunct, adjunct_compar, adjunct_qt, adjunct_rc, aglt, app, aux, comp, comp_ag, comp_fin, comp_inf, cond, conjunct, imp, item, mwe, ne, neg, obj, obj_th, pd, pre_coord, punct, refl, root, subj, vocative. Descriptions of Polish dependency relation types are at http://zil.ipipan.waw.pl/PDB/DepRelTypes. Some dependents are annotated with semantic roles, e.g. Beneficiary/Recipient.

Download: The updated version of Składnica zależnościowa (the first version of PDB). If you wish to get the entire PDB corpus (22K sentences annotated with the dependency trees) please contact alina <at> ipipan.waw.pl (replace <at> with @).

PDB in Universal Dependencies format (PDB-UD)

PDB-UD is a conversion of PDB in the UD-like format. It is an extended and corrected version of the Polish UD treebank (the release 2.1). PDB-UD contains enhanced graphs, i.e. trees with enhanced edges encoding shared dependents of coordinated elements, e.g. Dziewczynka śpiewa i tańczy (The girl sings and dances), and shared governors of coordinated elements, e.g. Dziewczynka i chłopiec śpiewają (A girl and a boy sing). The Polish dependency types are listed here.

PDB-UD trees were used in two shared tasks: LAW-MWE-CxG-2018 and PolEval 2018.

Download: PDB-UD is publicly available at http://git.nlp.ipipan.waw.pl/alina/PDBUD.

Download: Alternatively, you can download PDB-UD trees from the UD repository.

PDB-trained COMBO’s models

Natural language preprocessing models estimated on PDB and PDB-UD are available at http://zil.ipipan.waw.pl/PDB/COMBO.

Publications

Alina Wróblewska. Polish Dependency Parser Trained on an Automatically Induced Dependency Bank. PhD thesis, Institute of Computer Science, Polish Academy of Sciences, Warsaw, 2014. [PDF]

Alina Wróblewska. Extended and Enhanced Polish Dependency Bank in Universal Dependencies Format. In Proceedings of the Second Workshop on Universal Dependencies (UDW 2018), pages 173–182, 2018. [PDF]

Alina Wróblewska. Towards the Conversion of National Corpus of Polish to Universal Dependencies. In Proceedings of the 12th Language Resources and Evaluation Conference, pages 5308–5315, Marseille, France, 2020. European Language Resources Association. [PDF]

Acknowledgements

The creation of PDB was supported by grant no POIG.01.01.02-14-013/09 from Innovative Economy Operational Programme co-financed by the European Union (European Regional Development Fund) and by the grant from the Polish Ministry of Science and Higher Education as part of the investment in the CLARIN-PL research infrastructure (2016-2020).

Licence

The resources are distributed under the CC BY-NC-SA 4.0 licence.

Contact

Any questions, comments? Please send them to Alina Wróblewska (alina@ipipan.waw.pl).