chunked/nltk_chunking/model_doc_funnel/chunk_2.txt

The abstract from the paper is the following:
With the success of language pretraining, it is highly desirable to develop more efficient architectures of good
scalability that can exploit the abundant unlabeled data at a lower cost.