SkładnicaMWE

SkładnicaMWE is a constituency version of the Składnica treebank annotated with various types of multiword expressions. It was created within the PhD thesis work by Jakub Waszczuk, and partly funded by the IC1207 COST action PARSEME.

Some aspects of its construction, contents and use have been described in:

• SAVARY, A., WASZCZUK, J., (2017): "Projecting multiword expression resources on a Polish treebank", in the Proceedings of the 6th Workshop on Balto-Slavic Natural Language Processing (BSNLP 2017), 4 April 2017, Valencia, Spain.

The pre-annotation was performed by automatically projecting 3 Polish MWE resources:

• named entity layer of the National Corpus of Polish,

• SEJF, the grammatical lexicon of Polish nominal, adjectival and adverbial MWEs,

• Walenty, a Polish valence dictionary with phraseological component (2015 version).

All automatic pre-annotation results were manually validated.

The treebank contains about 2,000 MWE annotations in about 9,000 constituency trees, with the following distribution:

• 1,304 multiword named entities (e.g. Buenos Aires, Ministerstwo Pracy i Polityki Socjalnej),

• 368 nominal, adjectival and adverbial compounds (e.g. związki zawodowe, jedyny w swoim rodzaju, przede wszystkim),

• 365 verbal MWEs (e.g. wejść w życie, pominąć milczeiem, zależć za skórę, udzielić rady).

Authors

• Jakub Waszczuk

• Agata Savary

License

The data are available under the GPLv3 license.

Available resources

• SkładnicaMWE v 1.0 -- a version of Składnica (containing only the correct parses) with MWE annotations, in a custom XML format. Token identifiers are compatible with the original Składnica corpus.

Future work

• Repeating the automatic mapping and manual validation with more recent versions of Składnica and of Walenty.

• Enhancing the lexicon projection to include more fine-grained syntactic constraints.
• Enhancing the annotation schema towards a standard format.
• Linking the MWE occurrences in the treebank with their entries in lexicons.