Shortcut to filter out SRL mistakes
Closed this issue · 2 comments
elliottash commented
SRL sometimes makes mistakes, where non-verbs are identified as the verb, and the resulting annotations are garbage.
A shortcut to automatically filter these out, would be to run a part of speech tagger on the sentence. When SRL tags non-verbs as the verb, flag that sentence so it can be dropped easily.
-- in process_rules
function, add an option to filter for POS tags by semantic role.
-- by default, only allow for V* (verbs) for the verb semantic role.
-- could also add a prefab for agents and patients, keeping only adjectives and nouns (but probably not as default).
elliottash commented
@Dbhasin1 can make tests for this issue before it's merged into the master branch.
PinchOfData commented
Done in relatio v0.3