A Dynamic Speaker Model for Conversational Interactions

This repository includes the codes and models for the paper A Dynamic Speaker Model for Conversational Interactions for reproducing the experiment results on Switchboard Dialog Act classification.

@InProceedings{Cheng2019NAACL,
  author    = {Hao Cheng and Hao Fang and Mari Ostendorf},
  title     = {A Dynamic Speaker Model for Conversational Interactions},
  booktitle = {Proc. Annual Conference of the North American Chapter of the Association for Computational Linguistics (NAACL)},
  year      = {2019},
  pages     = {2772--2785},
  url       = {https://www.aclweb.org/anthology/N19-1284},
}

Requirements

• Python >= 2.7
• virtualenv

SWDA Data Processing

• Set up the Python virtual environment for data processing:

  virtualenv env/venv_data_processing
  source env/venv_data_processing/bin/activate
  pip install -r data_script/requirement.txt

• Install spacy==2.0.16 and download spacy model.

  pip install spacy==2.0.16
  python -m spacy download en_core_web_sm
  python -m spacy link en_core_web_sm en

• Install nltk==2.0.5. You may get the following error if you directly use pip install nltk==2.0.5.

   urllib2.HTTPError: HTTP Error 403: SSL is required
  

  • Download the package from here. Untar the package, and edit distribute_setup.py: change http to https in line 50

     DEFAULT_URL = "https://pypi.python.org/packages/source/d/distribute/"
    

  • Install the package using pip.

     pip install ./nltk-2.0.5
    

• Download the swda.zip from The Switchboard Dialog Act Corpus and unzip it into data/swda.
• Run the data processing script

  ./process_predictor_data.sh

  This script produces two subdirectories data/swda_user_dialog_dir and data/swda_predictor_dialog_dir which contain the converted data for training the dynamic speaker model and the dialog act tagging model, respectively.

SWDA Dialog Act Tagging Model

• Set up the Python virtual environment for the tagging model.

  virtualenv env/venv_tagging_model
  source env/venv_tagging_model/bin/activate
  pip install -r src/requirement.txt

• Untar the pretrained model

  tar -xzvf swda_model.tgz

• Run the following script to evaluate the model

  ./eval_tagger_model.sh

  This script outputs the evaluation results and dumps the prediction into misc/eval_tagger_model_dir.

• To train your own model, please see ./train_tagger_model.sh for details.

Train and Evaluate Dynamic Speaker Models

Please see ./train_user_model.sh and ./eval_user_model.sh for details.