Accompanying repository of our AAAI-20 paper: "Fine-Grained Argument Unit Recognition and Classification".

⚠️ The dataset was updated (cleaner parsing and encoding) and is not similar to the original one in the paper.

⚠️ However, the count of the sentences and labels is the same.

⚠️ We will soon update the reported results.

pip install -r requirements.txt

sh download.sh

The column merged_segments are the gold labels, created from the 5 annotators per sample.

Each entry is composed of 3 parts, resulted from the crowd annotation interface.

Example ('false', '(21,22);(49,132);', 'con;pro;')

The first part (false or true) indicates whether the annotator selected the field that no argument span can be detected. (false means there was an argument span selected)

The second part are the positions (in characters) of the argument segment(s) within the sentence in the form (start_of_segment,length_of_segment) separated with ; if there are several segments.

The third part are the corresponding stance labels (con or pro) for the argument segment(s).

python3 preparation.py

sh run.sh

If you find this dataset helpful, please cite our publication.

@inproceedings{trautmann2020fine,
  title = {Fine-Grained Argument Unit Recognition and Classification},
  author = {Dietrich Trautmann and
            Johannes Daxenberger and
            Christian Stab and
            Hinrich Schutze and
            Iryna Gurevych},
  booktitle = {The Thirty-Fourth {AAAI} Conference on Artificial Intelligence, {AAAI} 2020},
  publisher = {{AAAI} Press},
  month = {2},
  year = {2020},
  url = {https://aaai.org/Papers/AAAI/2020GB/AAAI-TrautmannD.7498.pdf},
}