Table 2: 
Distribution of answer (or correct answer option) types in three kinds of reading comprehension data sets. Statistics of other data sets come from Reddy et al. (2018), Kočiskỳ et al. (2018), and Lai et al. (2017).
SQuADNarrativeQACoQARACEDREAM (this work)
Answer type extractive abstractive abstractive multiple-choice multiple-choice 
Source document type written text written text written text written text dialogue 
# of source documents 536 1,572 8,399 27,933 6,444 
Average answer length 3.2 4.7 2.7 5.3 5.3 
Extractive (%) 100.0 73.6 66.8 13.0 16.3 
Abstractive (%) 0.0 26.4 33.2 87.0 83.7 
SQuADNarrativeQACoQARACEDREAM (this work)
Answer type extractive abstractive abstractive multiple-choice multiple-choice 
Source document type written text written text written text written text dialogue 
# of source documents 536 1,572 8,399 27,933 6,444 
Average answer length 3.2 4.7 2.7 5.3 5.3 
Extractive (%) 100.0 73.6 66.8 13.0 16.3 
Abstractive (%) 0.0 26.4 33.2 87.0 83.7 
Close Modal

or Create an Account

Close Modal
Close Modal