ACE 2005 Evaluation Corpus

This corpus contains the data used for the ACE 2005 evaluation exercise. Training data files were dually annotated by two annotators working independently. Discrepancies between the two versions of each file were then adjudicated by a senior annotator or team leader, resulting in a gold standard file. After adjudication, TIMEX2 values were normalized (for English only).

The distribution of files across domains in the corpus is as follows:

ACE 2005 Evaluation corpus

Domain	Domain Code	#Docs	#Words	#TIMEX2	Comments
Broadcast Conversation	BC	9	7499 (15%)	142
Broadcast News	BN	74	10049 (20%)	322
Conversational Telephone Speech	CTS	6	7531 (15%)	70
Newswire	NW	34	10410 (20%)	305
Usenet Newsgroups	UN	13	7503 (15%)	167
Weblog	WL	19	7299 (15%)	148
Total		155	50291 (100%)	1154