This repository contains scripts to train German language models. Check licenses on the sites of the specific datasets, if you want use the these language models.
| Name | Num. Sentences | Site |
|---|---|---|
| EuroParl | 1920208 | http://www.statmt.org/europarl/ |
| News-Commentary | 383764 | https://www.statmt.org/wmt13/translation-task.html |
| Tuda-Text | 7776674 | https://www.inf.uni-hamburg.de/en/inst/ab/lt/resources/data/acoustic-models.html |
In order to reproduce the language models, first the data has to be prepared. The required commands can be found in prepare_data.sh.
For n-gram language models, kenlm is used.
Trained models can be found in the attachements of the releases.