add sciarg dataset

[ArneBinder]

In this PR, we add SciArg - an argumentation-mining dataset, according to the description in #10.

Note: This requires pie-dataset>=0.4.0 (in detail: #57, #58, #59, #60).

TODO:

• complete tests
• complete PIE dataset card
• edit HF dataset card, if needed: see https://huggingface.co/datasets/DFKI-SLT/brat/commit/bb8c37d84ddf2da1e691d226c55fef48fd8149b5
• maybe re-think adhust how parts_of_same relations and dataset variant name=merge_fragmented_spans play together @ArneBinder
  • overwrite configs with a single default one, but with parameter merge_fragmented_spans=true (and set DOCUMENT_CONVERTERS instead of overwriting document_converters)
  • [ ] add entry for ...With LabeledMultiSpan... to DOCUMENT_CONVERTERS will be added in a follow-up PR

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (e8d556b) 94.57% compared to head (8011ece) 94.72%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #61      +/-   ##
==========================================
+ Coverage   94.57%   94.72%   +0.14%     
==========================================
  Files          18       19       +1     
  Lines        1272     1289      +17     
==========================================
+ Hits         1203     1221      +18     
+ Misses         69       68       -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[idalr]

Regarding label counts, there are discrepancies between

• what's reported in L et al., 2018, p. 43),
• previous report in pie-document-level (here; labels counted in TextDocumentWith...), and
• in the current HF dataset card (labels counted in BratDocument's).

Possibly caused by the difference in label assignment during the document extraction and/or conversion processes.

[ArneBinder]

The dataset cards got a bit mixed up, see below. In short: we do not have a HF sciarg dataset, so we can not have a HF dataset card for it.

EDIT: Please also put the content of your above comment to a fitting location in the PIE dataset card.