Shared Tasks

Cross-Framework Meaning Representation Parsing @CoNLL-2020

One of the five frameworks in the MRP task is Discourse Representation Graphs (DRGs). The DRGs are obtained by converting the DRSs from the Parallel Meaning Bank into graphs as illustrated by the example. For the parsing task, participating systems are expected to produce DRGs for input raw texts.

Note that the task is not limited to English. It also contains cross-lingual meaning representation pasring as a subtask. The German language is represented from the DRGs side. The shared task page contains more details about the cross-framework and cross-lingual tasks.

DRS Parsing @IWCS-2019

The first shared task on parsing Discourse Representation Structures (DRSs) took place in 2019. This shared task combined:

  • Semantic role labelling
  • Reference resolution
  • Scope detection
  • Named entity tagging
  • Word sense disambigation
  • Predicate-argument structure prediction
  • Presupposition projection

Yes, all of these into one big challenge!

The DRSs were drawn from the Parallel Meaning Bank. In this first episode of DRS parsing, participating systems were tested mainly on short, open-domain sentences, to make the threshold for participation as low as possible. An example of input-output of the DRS parsing is shown below, where a standard, box-formatted DRS is serialized as a set of clauses which is suitable for machine learning and evaluation.

For more details you can still visit the CodaLab page of the shared task.

The outcome of the shared task and all participating systems was presented at IWCS-2019 in Gothenburg, Sweden. The publications are made available in the ACL Anthology.