Construction of an annotated corpus to support biomedical information extraction

BMC Bioinformatics

Table 7 Most common semantic role patterns

AGENT	THEME	Other	Count (%)
AGENT	THEME		947 (30.88)
	THEME		693 (22.60)
	THEME	DESCRIPTIVE-THEME	119 (3.88)
	THEME	LOCATION	117 (3.81)
AGENT	THEME	MANNER	113 (3.68)
AGENT		DESTINATION	113 (3.68)
	THEME	MANNER	112 (3.65)
AGENT	THEME	LOCATION	64 (2.09)
AGENT			59 (1.92)
AGENT		DESCRIPTIVE-AGENT	51 (1.66)
	THEME	CONDITION	47 (1.53)
		MANNER	42 (1.37)
AGENT	THEME	CONDITION	38 (1.24)
		SOURCE	36 (1.17)
	THEME	PURPOSE	31 (1.01)

The patterns shown are independent of their ordering in the text. If the roles of AGENT and/or THEME are present in the pattern, this is indicated in the 1^st and 2^nd columns, respectively. The 3^rd column shows any other role present in the pattern (the most common patterns all have a maximum of one role which is not AGENT and/or THEME). The final column shows the total number of events that have been annotated with each pattern in the corpus, together with the percentage of all events in the corpus that this figure represents.

ISSN: 1471-2105