Executive summary on the state of the art in Semantic Role Labelling

This is a follow-up to the T3: Semantic Role Labeling: Past, Present and Future tutorial by Lluís Màrquez.

A lenient evaluation of the state-of-the-art F1 in Semantic Role Labelling is around 80%, but the quality degrades by ~10% when switching to a test set from a new knowledge domain.

The most promising approach is the construction of a joint system of syntactic and semantic labeling parsers that operate in parallel on the same stream of input tokens.

The field is plagues by the complexity and low performance of the tools.