[Retracted] Analyzing the Effect of Masking Length Distribution of MLM: An Evaluation Framework and Case Study on Chinese MRC Datasets

Table 8

The number of contexts of each MRC dataset.

Dataset	Context #	Train context #	Dev context #	Test context #	Unit of the context
CoQA	8399	7199	500	700	Passage
CLOTH	7131	5513	805	813	Passage
DREAM	6444	3869	1288	1287	Dialogue
Qangaroo-M	2508	1620	342	546	Passage
TQA	1076	666	200	210	Lesson
MovieQA	548	362	77	109	Movie
SQuAD1.1	536	442	48	46	Article
SQuAD2.0	505	442	35	28	Article
Short span	20,519	15,265	2770	2484	Passage
Long span	10,484	7756	1163	1565	Passage
Short cloze	58,500	40,500	9000	9000	Passage
Long cloze	58,500	40,500	9000	9000	Passage