Sammlung: DGS-Korpus-Tools

The Public DGS Corpus is an annotated corpus of conversations in German Sign Language (DGS). We have released several scripts released that relate directly to the corpus. They are mainly connected to the OpenPose data that has been published for each transcript of the corpus.

For more information on our OpenPose data and the scripts related to it, see the project note “OpenPose in the Public DGS Corpus”.

The following scripts are available:

OpenPose Wrapper

Usually, OpenPose stores its data as individual files per frame. For the Public DGS Corpus we defined a wrapper format that combines these into a single file and adds some additional meta-information. This script takes care of this process.

OpenPose Frame Extractor

The OpenPose data of each transcript uses the DGS-Korpus OpenPose wrapper format to collect all of its information in a single file. This script converts such a wrapper file back into the one-frame-per-file format used by the OpenPose demo.

OpenPose Error Corrector

For the total perspective of each corpus transcript, which shows both informants and the moderator, the OpenPose output was postprocessed to make several improvements. This script performs those postprocessing steps.