Table 2.

Examples of errors made by the NER trained on supervised annotated data

Error type	No	Category	Examples
Partial match	1	Tagged prefix	Probable person to person transmission of novel avian influenza A (H7N9) virus in EasternChina, 2013.
	2	Tagged suffix	Surveillance was conducted in live poultry markets in Fujian, Guangdong, Guangxi, Guiyang, Hunan and YunnanProvinces.
	3	Tagged suffix	University of Ibadan, OyaState, Ibadan and Nigeria.
	4	Unrecognized token	The overwhelming majority (94.2%) of H9N2 influenza viruses were isolated in Asia, with > 65% coming from mainland and Hong Kong of China
False positive	5	Other entities	Phylogenetic analyses show that it is a recombinant virus containing genome segments derived from the Eurasia and North America gene pools.
	6	Other entities	Thus, current G1-like viruses in southern China might have originally been introduced from Middle Eastern countries, or it is also likely that the virus spread the other way around, similar to the transmission of FIG.
	7	Other entities	This work was supported by a Natural Sciences and Engineering Research Council of Canada discovery grant.
	8	Partial annotation	Abbreviations: BJ and Bei, Beijing; Ck, chicken; Dk, duck.
False negative	9	Table entries	Virus Group State of isolation Date of isolation A/chicken/Nigeria/1071-1/2007 EMA1/EMA2-2: 6-R07 Plateau Jan 2 A/chicken/Nigeria/1071-3/2007 EMA2 Sokoto Jan 5.
	10	Unrecognized toponym	The characterization of the swH3N2 / pH1N1 reassortant vi- ruses from swine in the province of Quebec indicates that reassortment of gene segments had occurred between the North American swine H3N2.
	11	Unrecognized toponym	Centers for Disease Control and Prevention, Atlanta, Ga.

Note: Underlined tokens indicate entities recognized by the NER. Italicized tokens are human annotated gold standard entities.