xuuHuang / IMTLab

Code for the EMNLP2023 main conference paper "IMTLab: An Open-Source Platform for Building, Evaluating, and Diagnosing Interactive Machine Translation Systems"

IMTLab

IMTLab is an end-to-end interactive machine translation (IMT) evaluation platform that enables researchers to quickly build IMT systems with state-of-the-art models, perform an end-to-end evaluation, and diagnose the weakness of systems.

Requirements

• Python version >= 3.8.0
• Pytorch version >= 1.8.0
• editdistance
• sentencepiece
• flask
• openai

We include fairseq toolkit as a third party library to implement common IMT systems. We recommend users to install fairseq 0.12.2-release from our repository.

cd IMTLab
pip install -e third_party/fairseq

Running IMTLab

First change your configures in config/*.json and src/run.sh. The config/*.json files contain the parameters of the IMT models based on fairseq. And src/run.sh contains the arguments of the environment such as the model type, the policy type, the path of test data, etc. After configuration, simply run the following code.

cd src
bash run.sh

If you want to use the human environment, change the policy type to 5 in src/run.sh and run the above code.

Data

In data/ dir is the test data we used in the experiments which is randomly sampled from the testset of WMT.

The human_exps directory contains the human interaction data collected from our human experiments.