Resource-Size matters: Improving Neural Named Entity Recognition with\n Optimized Large Corpora

This study improves the performance of neural named entity recognition by a\nmargin of up to 11% in F-score on the example of a low-resource language like\nGerman, thereby outperforming existing baselines and establishing a new\nstate-of-the-art on each single open-source dataset. Rather than designing\ndeeper and wider hybrid neural architectures, we gather all available resources\nand perform a detailed optimization and grammar-dependent morphological\nprocessing consisting of lemmatization and part-of-speech tagging prior to\nexposing the raw data to any training process. We test our approach in a\nthreefold monolingual experimental setup of a) single, b) joint, and c)\noptimized training and shed light on the dependency of downstream-tasks on the\nsize of corpora used to compute word embeddings.\n