Seizure semiology inferred from clinical descriptions and from video recordings. How accurate are they?