how test data(unlabel data) do data augmentation

[xilinniao123]

thanks for sharing! i have 2 problems want to ask:
1: in the mixMatch step if all test data are labeled O，how to detect target token and non-target token?
2： in figure 5 if the augument is span based,then the origin AS entity 'food' may be replaced with another entity that has diffent length for example:

now the two token sequence have different length,and the true labels is not suitable for the augument data?