Argument Optimization

This repository contains the code associated with the following paper:

Claim Optimization in Computational Argumentation by Gabriella Skitalinskaya, Maximilian Spliethöver, Henning Wachsmuth

Reproducing results

Models & Data

To obtain access to BART checkpoints, along with BERT, SBERT and SVMRank models, please reach out to Gabriella Skitalinskaya (email can be found in paper). Let us know if you have any questions.

Generation

    pip install torch==1.3.1
    pip install numpy==1.19.2
    pip install --editable .

In the fairseq/data folder you can put the three dataset configurations each containing train.source , train.target , val.source , val.target files.

BPE preprocessing:

      for SPLIT in train val test
      do
        for LANG in source target
        do
          python -m examples.roberta.multiprocessing_bpe_encoder \
          --encoder-json encoder.json \
          --vocab-bpe vocab.bpe \
          --inputs "/data/no_context/$SPLIT.$LANG" \
          --outputs "/data/no_context/$SPLIT.bpe.$LANG" \
          --workers 60 \
          --keep-empty;
        done
      done

Binarizing dataset:

      fairseq-preprocess \
        --source-lang "source" \
        --target-lang "target" \
        --trainpref "/data/no_context/train.bpe" \
        --validpref "/data/no_context/val.bpe" \
        --destdir "/data/no_context" \
        --workers 60 \
        --srcdict dict.txt \
        --tgtdict dict.txt;

Model training:

      TOTAL_NUM_UPDATES=20000  
      WARMUP_UPDATES=500      
      LR=3e-05
      MAX_TOKENS=1024
      UPDATE_FREQ=8
      python train.py data/no_context\
        --restore-file ../bart.large/model.pt \
        --max-tokens $MAX_TOKENS \
        --task translation \
        --source-lang source --target-lang target \
        --truncate-source \
        --truncate-target \
        --layernorm-embedding \
        --share-all-embeddings \
        --share-decoder-input-output-embed \
        --reset-optimizer --reset-dataloader --reset-meters \
        --required-batch-size-multiple 1 \
        --arch bart_large \
        --criterion label_smoothed_cross_entropy \
        --label-smoothing 0.1 \
        --dropout 0.1 --attention-dropout 0.1 \
        --weight-decay 0.01 --optimizer adam --adam-betas "(0.9, 0.999)" --adam-eps 1e-08 \
        --clip-norm 0.1 \
        --lr-scheduler polynomial_decay --lr $LR --total-num-update $TOTAL_NUM_UPDATES --warmup-updates $WARMUP_UPDATES \
        --memory-efficient-fp16 --update-freq $UPDATE_FREQ \
        --save-dir "checkpoint-opt/3types/no_context" \
        --ddp-backend=no_c10d  \
        --skip-invalid-size-inputs-valid-test \
        --max-epoch 10 \
        --find-unused-parameters;

Candidate generation:

Before running SVMRank-related experiments, follow the instructions to installing PySVMRank as described here.

      python get_candidates.py \
        --modeldir checkpoint-opt/3types/no_context \
        --datadir data/no_context \
        --outdir data/no_context

Candidate ranking:

    pip install pandas==1.1.5
    pip install scikit-learn==0.24.2
    pip install sentence-transformers==2.1.0
    pip install transformers==4.11.2
    pip install nltk==3.6.3
    pip install scipy==1.5.4

    python rank.py \
        --mode no_context\
        --candidates checkpoint-opt/3types/no_context/test.hypo \
        --datadir data/no_context/ \
        --svmrank models/svmrank/svmrank_sbert_model.bin \
        --fluency models/fluency/ \
        --argquality models/argument_quality/ \
        --sbertmodel models/sbert/ \
        --outfile results.csv

Citation

If you use this corpus or code in your research, please include the following citation:

@inproceedings{skitalinskaya-etal-2023-claim,
    title = "Claim Optimization in Computational Argumentation",
    author = "Skitalinskaya, Gabriella  and
      Spliethöver, Maximilian  and
      Wachsmuth, Henning",
    booktitle = "Proceedings of the 16th International Natural Language Generation Conference",
    month = sep,
    year = "2023",
    address = "Prague, Czechia",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.inlg-main.10",
    pages = "134--152",
}