# GermanWordEmbeddings

A toolkit for training and evaluating German-language word embeddings, covering obtaining and preprocessing German text corpora, training models with gensim's word2vec, and evaluating them against generated test sets.

GermanWordEmbeddings applies deep learning via gensim's word2vec implementation to German corpora, since the authors note that most existing word embedding research and tooling focuses on English text. A bundled shell script automates downloading the toolkit and its corpus files and running training and evaluation end to end, though the README warns this can take a very long time.

Documented corpus sources include German Wikipedia dumps and shuffled German news text from a statistical machine translation dataset spanning 2007 to 2013; the project's own pretrained models were built from German Wikipedia and 2013 news data. A WikipediaExtractor script and shell commands strip the raw Wikipedia XML dump down to plain text before training, and Jupyter notebooks in the repository show example script output for users who want to inspect how each step works without running the full pipeline themselves.

Website: https://devmount.github.io/GermanWordEmbeddings/