akutuzov / mrptask

http://mrp.nlpl.eu/ /net/work/projects/mrptask

Download training data from LDC:

• https://catalog.ldc.upenn.edu/organization/downloads
  • Log-in as Dan. He is the contact point for Charles University, and only he can download the data.
• Update of UCCA training data can be downloaded directly from the shared task website.
• UDPipe-analyzed training data available as Companion Data.

TO DO

• Ask organizers for the companion data we want to use:
  • Universal Dependencies English treebanks (EWT, GUM, ParTUT, PUD, LinES) release 2.4
  • Penn Treebank 3 constituent annotation
  • PropBank/NomBank
    • English Web Treebank Propbank
    • OntoNotes Release 5.0
    • Update on Github
  • CzEngVallex
  • Prague Czech-English Dependency Treebank 2.0 (at least the English part, including tectogrammatical trees)
  • Raw texts and word embeddings from the CoNLL 2017 shared task
  • FasText pre-trained word vectors
  • BERT pre-trained models
• Collect the initial data on ÚFAL network: /net/work/projects/mrptask (Dan).
• Look for suitable implementations of graph parsers (Kira).
• Look for possibilities of "UD enhancement" (Dan).

Next meeting

Friday April 5 afternoon

Potentially useful stuff

• Stanford CoreNLP includes a converter from the Penn Treebank constituents to UD v1
• Context-sensitive mapping of PTB POS tags to Universal POS tags