EDU-AP: Elementary Discourse Unit based Argument Parser

This is the implementation of the paper:

EDU-AP: Elementary Discourse Unit based Argument Parser

Sougata Saha, Souvik Das, Rohini Srihari

The 23rd Annual Meeting of the Special Interest Group on Discourse and Dialogue (SIGDIAL 2022)

Abstract

Neural approaches to end-to-end argument mining (AM) are often formulated as dependency parsing (DP), which relies on token-level sequence labeling and intricate post-processing for extracting argumentative structures from text. Although such methods yield reasonable results, operating solely with tokens increases the possibility of discontinuous and overly segmented structures due to minor inconsistencies in token level predictions. In this paper, we propose EDU-AP, an end-to-end argument parser, that alleviates such problems in dependency-based methods by exploiting the intrinsic relationship between elementary discourse units (EDUs) and argumentative discourse units (ADUs) and operates at both token and EDU level granularity. Further, appropriately using contextual information, along with optimizing a novel objective function during training, EDU-AP achieves significant improvements across all four tasks of AM compared to existing dependency-based methods.

Dataset Used:

Persuasive Essays (PE) Corpus: https://tudatalib.ulb.tu-darmstadt.de/handle/tudatalib/2422

Model for EDU Segmentation:

Bi LSTM-CRF based discourse segmenter (NeuralEDUSeg) by Wang et al. (2018): https://github.com/PKU-TANGENT/NeuralEDUSeg

Training and Inference

You can train and evaluate all experiments using the runner.sh script. Example: nohup bash runner.sh 1 12 > log.txt 2>&1 & runs experiment numbers 1 to 12 sequentially. All the different configurations for the experiments can be found in the experiment_config.json file. In order to experiment with different parameters, you can directly execute the run_training.py script. Sample command below:

python -m torch.distributed.run --nnodes=1 --nproc_per_node=4 --master_port 9999 ./run_training.py --batch_size 16 --num_epochs 15 --learning_rate 0.00002 --base_transformer "roberta-base"

Prior to training, please download the PE dataset into a folder named ./data/, and run NeuralEDUSeg on the data to generate the EDUs.

Citation

If you are using this library then do cite:

@inproceedings{saha-etal-2022-edu-ap,
    title = "EDU-AP: Elementary Discourse Unit based Argument Parser",
    author = "Saha, Sougata  and
      Das, Souvik  and
      Srihari, Rohini",
    booktitle = "Proceedings of the 23rd Annual Meeting of the Special Interest Group on Discourse and Dialogue",
    month = sep,
    year = "2022",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.sigdial-1.19",
    pages = "183--192"
}