Setting the environment

NLTK downloads

>>> import nltk
>>> nltk.download('punkt')
>>> nltk.download('averaged_perceptron_tagger')
>>> nltk.download('wordnet')

Using the Stanford Parser (with nltk 3.2.1)

1. Download the stanford parser from the official site.
2. Unzip the file in STANFORD_FOLDER
3. Run:

$ export STANFORDTOOLSDIR=STANFORD_FOLDER
$ export CLASSPATH=$STANFORDTOOLSDIR/stanford-parser-full-XXXX-XX-XX/stanford-parser.jar:$STANFORDTOOLSDIR/stanford-parser-full-XXXX-XX-XX/stanford-parser-3.6.0-models.jar:$STANFORDTOOLSDIR/stanford-parser-full-XXXX-XX-XX/slf4j-api.jar

Experiments

Using the sequential classifiers

For brat annotations, run the following pipeline:

$ cd preprocess
$ python process_arg_essays_for_conll.py --input_dirpath INPUT_DIRPATH --output_file PICKLED_DOCUMENTS

INPUT_DIRPATH must contain the .txt and .ann files to process. This will generate a pickled file PICKLED_DOCUMENTS with an internal representation of the text files.

$ cd ../experiments
$ python crf_baseline.py --input_filename PICKLED_DOCUMENTS

Now check the file logs/log-crf for your results!

Using the neural classifiers

To run the script, clone the repository https://github.com/UKPLab/emnlp2017-bilstm-cnn-crf.git under the name ukplab_nets and add it the path to PYTHONPATH.

export PYTHONPATH=$PYTHONPATH:/home/.../path_to_ukplab_nets:/home/.../path_to_ukplab_nets/ukplab_nets

Install Keras 2.1.5 and Tensorflow 1.7 with pip

pip install keras==2.1.5

Running preprocess

To run the preprocess use

python -m preprocess.ukpnets_process

Running test

python -m models.test_filename

or use nose directly (you wont see the output):

nosetests models.test_arg_bilstm
nosetests models.test_arg_bilstm --nologcapture
nosetests models.test_arg_bilstm:AttArgBiLSTMTest.test_predict_attention