Skip to content

Latest commit

 

History

History
40 lines (27 loc) · 958 Bytes

readme.md

File metadata and controls

40 lines (27 loc) · 958 Bytes

Notes on the shared task

This repo has the relevant files to recreate our entry for the SIGMORPHON 2021 G2P shared task.

Everything was run using OpenNMT-py in a docker image.

Configuring the docker image

Build an image like this:

docker run -it --gpus all --name nmtpy \
	-v ~/somefolder:/mh \
	pytorch/pytorch:1.7.0-cuda11.0-cudnn8-runtime

The somefolder should contain the 2021-task1-main folder from the SIGMORPHON 2021 github repo.

Inside the image, install opennmt-py like this:

pip install OpenNMT-py

Some additional convenient stuff:

apt update
apt install vim
apt install wget

Other files here

  • final.sh: master batch file; this calls everything else.
  • big.yaml: configuration file for OpenNMT, invoked by final.sh
  • mfinal.py: data munging, called by final.sh
  • stats.py: does the $t$-tests for the paper
  • wer.py: does the Monte Carlo estimates of word-level accuracy