conllpp

CoNLLpp is a corrected version of the CoNLL2003 NER dataset where labels of 5.38% of the sentences in the test set have been manually corrected. The training set and development set from CoNLL2003 is included for completeness. One correction on the test set for example, is: {"tokens": ["SOCCER", "-", "JAPAN", "GET", "LUCKY", "WIN", ",", "CHINA", "IN", "SURPRISE", "DEFEAT", "."], "original-ner-tags-in-conll2003": ["O", "O", "B-LOC", "O", "O", "O", "O", "B-PER", "O", "O", "O", "O"], "corrected-ner-tags-in-conllpp": ["O", "O", "B-LOC", "O", "O", "O", "O", "B-LOC", "O", "O", "O", "O"],}