Detecting modification of biomedical events using a deep parsing approach

Table 1 Results over development set

Mod	RMRS from	Extra	Gold			UTurku
			R	P	F	R	P	F
$S PECULATION$	-	$W_{+ 3}^{- 3}$	42.9	55.4	48.3	19.0	33.3	24.2
$S PECULATION$	RASP	-	16.7	66.7	26.7	5.5	26.1	9.0
$S PECULATION$	ERG	-	20.2	68.0	31.2	10.7	56.2	18.0
$S PECULATION$	fb(ERG,RASP)	-	25.0	61.8	35.6	13.1	50.0	20.8
$S PECULATION$	cb(ERG,RASP)	-	15.5	59.1	24.5	10.7	52.9	17.8
$S PECULATION$	ERG	$W_{+ 3}^{- 3}$	45.2	59.4	51.3	20.2	34.0	25.4
$S PECULATION$	fb(ERG,RASP)	$W_{+ 3}^{- 3}$	45.2	60.3	51.7	16.7	31.1	21.7
$S PECULATION$	cb(ERG,RASP)	$W_{+ 3}^{- 3}$	40.5	54.8	46.6	19.0	32.0	23.9
$N EGATION$	-	Negex	32.7	32.7	32.7	22.6	17.8	19.9
$N EGATION$	-	$W_{+ 3}^{- 4}$	51.8	54.8	53.3	25.4	33.7	29.0
$N EGATION$	RASP	-	12.7	35.9	18.8	5.4	26.1	9.0
$N EGATION$	ERG	-	26.4	72.5	38.7	15.4	34.0	21.2
$N EGATION$	fb(ERG,RASP)	-	35.4	66.1	46.1	17.3	34.6	23.0
$N EGATION$	cb(ERG,RASP)	-	29.1	64.0	40.0	13.6	34.1	19.5
$N EGATION$	ERG	$W_{+ 3}^{- 4}$	45.4	48.5	47.0	18.2	26.3	21.5
$N EGATION$	fb(ERG,RASP)	$W_{+ 3}^{- 4}$	44.6	66.2	53.3	19.1	33.3	24.3
$N EGATION$	cb(ERG,RASP)	$W_{+ 3}^{- 4}$	50.9	59.0	54.6	21.8	32.0	26.0

Results over the development data using gold-standard Task 1 annotations and the UTurku Task 1 system ("fb" = fallback strategy, where we use the first source if possible, otherwise the second; "cb" = use undifferentiated RMRSs from each source to create feature vectors).

ISSN: 1472-6947