MaltParser’s engmalt model (trained with Stanford Dependencies) was used to parse SEMAFOR’s training data for the argument identification model used in this demo. The frame identification model was trained with parses from the TurboParser configuration described above.
This demo will not spend more than 20 seconds per query, so it may fail on very long input.
Though these parsers approach or attain state-of-the-art performance on standard benchmarks,
they rarely predict completely accurate analyses—as can be seen for many of the provided example sentences.