GERNERMED++: Transfer Learning in German Medical NLP

TL;DR

This paper introduces GERNERMED++, a transfer learning-based German medical NLP model for named entity recognition, leveraging multilingual techniques and pretrained language models to improve performance and accessibility for the German medical research community.

Contribution

GERNERMED++ is a new open, publicly available German medical NER model that outperforms previous models by combining transfer learning, word-alignment, and neural machine translation.

Findings

Achieved strong NER performance on German medical texts

Provided an open model based on English data without legal restrictions

Enhanced baseline for German medical NLP research

Abstract

We present a statistical model for German medical natural language processing trained for named entity recognition (NER) as an open, publicly available model. The work serves as a refined successor to our first GERNERMED model which is substantially outperformed by our work. We demonstrate the effectiveness of combining multiple techniques in order to achieve strong results in entity recognition performance by the means of transfer-learning on pretrained deep language models (LM), word-alignment and neural machine translation. Due to the sparse situation on open, public medical entity recognition models for German texts, this work offers benefits to the German research community on medical NLP as a baseline model. Since our model is based on public English data, its weights are provided without legal restrictions on usage and distribution. The sample code and the statistical model is available at: https://github.com/frankkramer-lab/GERNERMED-pp