SimulEval is a general evaluation framework for simultaneous translation on text and speech.

• python >= 3.7.0

git clone git@github.com:fairinternal/SimulEval.git
cd SimulEval
pip install -e .

Following is the evaluation of a dummy agent which operates wait-k (k = 3) policy and generates random words until the length of the generated words is the same as the number of all the source words. A tutorial can be found here.

cd examples
simuleval \
  --agent dummy/dummy_waitk_text_agent.py \
  --source data/src.txt \
  --target data/tgt.txt

SimulEval is licensed under Creative Commons BY-SA 4.0.

Please cite as:

@inproceedings{simuleval2020,
  title = {Simuleval: An evaluation toolkit for simultaneous translation},
  author = {Xutai Ma, Mohammad Javad Dousti, Changhan Wang, Jiatao Gu, Juan Pino},
  booktitle = {Proceedings of the EMNLP},
  year = {2020},
}