chunked/nltk_chunking/model_doc_funnel/chunk_3.txt

To improve the efficiency, we examine the
much-overlooked redundancy in maintaining a full-length token-level presentation, especially for tasks that only
require a single-vector presentation of the sequence.