open-sesame
open-sesame copied to clipboard
Question on the reported results.
Is the reported frameid result (89.74 on dev and 86.55 on test) using the gold target labels? And what's the frameid result if targetid module is used instead of gold target labels?
It seems that frameid is trained and evaluated only on sentences which have frames, and the sentences which have no frame are ignored. This result seems not be able to reflect the performance on sentences which we do not know whether there exist frames?