chunked/nltk_chunking/model_doc_bit/chunk_14.txt

The authors show that the combination of both is useful for training with large batch sizes, and has a significant
impact on transfer learning.