MIT Press

Table 5:

Distribution (%) of types of required prior knowledge based on a subset of test and development sets of C³, Chinese free-form abstractive dataset DuReader (He et al., 2017), and English free-form multiple-choice datasets RACE (Lai et al., 2017) and DREAM (Sun et al., 2019a). Answering a question may require more than one type of prior knowledge.

	C_M³	C_D³	C³	RACE	DREAM	DuReader
Matching	12.0	14.3	13.2	14.7	8.7	62.0
Prior knowledge	88.0	85.7	86.8	85.3	91.3	38.0
◇ Linguistic	49.0	30.7	39.8	47.3	40.0	22.0
◇ Domain-specific	0.7	1.0	0.8	0.0	0.0	16.0
◇ General world	50.7	64.0	57.3	43.3	57.3	0.0
Arithmetic	3.0	4.7	3.8	3.3	1.3	0.0
Connotation	1.3	5.3	3.3	2.0	5.3	0.0
Cause-effect	14.0	6.7	10.3	2.7	3.3	0.0
Implication	17.7	20.3	19.0	24.0	26.7	0.0
Part-whole	5.0	5.0	5.0	2.7	7.3	0.0
Precondition	2.7	4.3	3.5	2.7	1.3	0.0
Scenario	9.6	24.3	17.0	7.3	21.3	0.0
Other	3.3	0.3	1.8	2.0	0.7	0.0

Single sentence	50.7	22.7	36.7	24.0	12.0	14.6
Multiple sentences	47.0	77.0	62.0	75.3	88.0	68.7
Independent	2.3	0.3	1.3	0.7	0.0	16.7

# of annotated instances	300	300	600	150	150	150

	C_M³	C_D³	C³	RACE	DREAM	DuReader
Matching	12.0	14.3	13.2	14.7	8.7	62.0
Prior knowledge	88.0	85.7	86.8	85.3	91.3	38.0
◇ Linguistic	49.0	30.7	39.8	47.3	40.0	22.0
◇ Domain-specific	0.7	1.0	0.8	0.0	0.0	16.0
◇ General world	50.7	64.0	57.3	43.3	57.3	0.0
Arithmetic	3.0	4.7	3.8	3.3	1.3	0.0
Connotation	1.3	5.3	3.3	2.0	5.3	0.0
Cause-effect	14.0	6.7	10.3	2.7	3.3	0.0
Implication	17.7	20.3	19.0	24.0	26.7	0.0
Part-whole	5.0	5.0	5.0	2.7	7.3	0.0
Precondition	2.7	4.3	3.5	2.7	1.3	0.0
Scenario	9.6	24.3	17.0	7.3	21.3	0.0
Other	3.3	0.3	1.8	2.0	0.7	0.0

Single sentence	50.7	22.7	36.7	24.0	12.0	14.6
Multiple sentences	47.0	77.0	62.0	75.3	88.0	68.7
Independent	2.3	0.3	1.3	0.7	0.0	16.7

# of annotated instances	300	300	600	150	150	150

View Large

This Feature Is Available To Subscribers Only