Table 16

Counts for argument candidates for the two test partitions on the end-to-end semantic role labeling task. The Predicted non-arguments line indicates how many candidate arguments are classified as non-arguments by the argument identification classifier. The Incompatible with SPs line indicates how many candidates were classified as arguments but cannot be modeled by our current SPs (i.e., they are not noun phrases or prepositional phrases containing a noun phrase as the second child). Lastly, the Compatible with SPs line lists how many candidates were both classified as likely arguments and can be modeled by the SPs.

WSJ test
Brown test
Predicted non-arguments 158,310 184,958
Incompatible with SPs 5,739 11,167
Compatible with SPs 7,691 7,867
Total 171,740 203,992

WSJ test
Brown test
Predicted non-arguments 158,310 184,958
Incompatible with SPs 5,739 11,167
Compatible with SPs 7,691 7,867
Total 171,740 203,992
Close Modal