a domain-adapted temporal expression recognizer for the English Voice Assistant domain

To refer to DA-Time in any publication, please cite the following paper:

Alam, T., Zarcone, A., & Padó, S. (2021). New Domain, Major Effort? How Much Data is Necessary to Adapt a Temporal Tagger To the Voice Assistant Domain. In Proceedings of the 14th International Conference on Computational Semantics (IWCS 2021).

@inproceedings{alam2021:new,
    author       = {Touhidul Alam and Alessandra Zarcone and Sebastian Pad\'{o}},
    title        = {{New Domain, Major Effort? How Much Data is Necessary to Adapt a Temporal Tagger To the Voice Assistant Domain}},
    year         = 2021,
    booktitle    = {{Proceedings of the 14th International Conference on Computational Semantics (IWCS 2021)}}
    }

• Clone the repository

  git clone https://github.com/audiolabs/DA-Time.git
  

• Install Python >= 3.7.3
• Make sure to have virtualenv installed by executing

  python -m pip install --user virtualenv
  

• Create the virtual environment

  python -m venv <path-to-env>
  

• Source the environment

  source <path-to-env>/bin/activate
  

• Install the following packages

  pip install -r requirements.txt
  

• Download Spacy English model

  python -m spacy download en_core_web_sm
  

• Download Stanford CoreNLP 4.0.0 (English model)

  https://stanfordnlp.github.io/CoreNLP/download.html
  

• Make sure to check the necessary experimental setup from the config.yaml file.

• To run the model with different mode (Training/testing/fine-tuning) run the following command

  python main.py exec --mode=<train/test/ft>
  

• [Optional] Parameter for multiple run (these will overwrite configs value)

  python main.py exec --mode=<train/test/ft> [--run=<run>] [--amount=<amount>] [--emb=<emb>]
  

  Here, run indicates #of run, amount is to define fine-tune data amount%, emb is the model embedding. To run multiple times, edit and use the script:

  ./RUN.sh
  

• To plot different data statistics run the following command

  python main.py plot
  

• To predict on a trained model, run the following command with n-number of sentences as argument

  python main.py predict "when is our tomorrow's meeting?" "Make a schedule from Friday to Monday"
  

• To simplify text with the CoreNLP run the following command in terminal

  java -mx4g -cp "*" edu.stanford.nlp.pipeline.StanfordCoreNLPServer -port 9000 -timeout 15000000
  

• In parallel, run the simplifier script with this commands:

  cd recognition
  python simplifier.py