📄 Proceedings of the Third Workshop on Universal Dependencies (UDW, SyntaxFest 2019)

Summary

This is the proceedings volume of the third Universal Dependencies workshop (UDW 2019), the annual meeting of the UD project community. The workshop took place in Paris in late August 2019 as part of the first SyntaxFest. Depling, TLT and Quasy, a new workshop on quantitative syntax, were held alongside it. In the preface, the editors explain why these events were brought together: annotated corpora are increasingly needed both for machine learning and for the empirical testing of linguistic theories. At that time, UD already contained about 120 treebanks for 80 languages in a single annotation scheme. All four events shared a common review process. The UDW papers describe new treebanks for low-resource languages, a large German corpus and corrections to existing annotations. Several papers discuss contested points of the scheme: parataxis, nested coordination and ellipsis. There are also tools, typological studies and parsing experiments. For Mova, this volume is useful as an overview of practices for building UD resources for low-resource languages and for resolving difficult annotation cases.

Key points

• SyntaxFest 2019 ran in Paris from 26 to 30 August and brought together four events. UDW was held on 29–30 August, and its program chairs were Alexandre Rademaker and Francis Tyers.
• Reviewing was shared. Of 40 submitted long papers, 28 were accepted, 6 of them for UDW. Of 62 short papers, 41 were accepted, 9 of them for UDW. Overall, about 70% of long papers and 66% of short papers were accepted.
• More than 110 participants registered. Adam Przepiórkowski gave an invited talk on arguments and adjuncts.
• New resources include HDT-UD, a very large UD treebank for German. UD corpora for Wolof, Mbyá Guaraní, Occitan (building on Romance UD corpora) and Karelian were also presented, along with an overview of Uralic UD corpora.
• Methodological papers:
  • updated annotations of a Turkish treebank;
  • testing Greenberg's word order universals on UD data;
  • a recursive LSTM representation of trees for arc-standard parsing;
  • the graphical editor ConlluEditor.