MAMMOTH for Sign Language Translation

This project adapts the 🦣 MAMMOTH toolkit, built on top of OpenNMT-py, for sign language translation, leveraging the Phoenix2014T dataset. This work contributes to the advancement of neural machine translation in the domain of sign language using modular, research-friendly tools from Helsinki-NLP.

Installation (install.sh)

1. Installation is the same as provided by mammoth library. In addition, sentencepiece and sacrebleu are also installed.

Project Structure

• configs dir: the configuration json file
• data dir: includes the Phoenix2014T sign language translation dataset
• vocabs dir: includes the tokenizer vocabulary

Usage

1. Download the Phoenix2014T dataset files:

$ ./download.sh

2. Run:

sbatch job.sh