TABLE 11.

Comparison of pipeline models with and without cascading errors.

Event role	CNN-BiLSTM-PCNN (no cascading errors)			CNN-BiLSTM-PCNN
	P	R	F 1	P	R	F 1
Acquisition-uses	0.657	0.98	0.951	0.597	0.74	0.661
Acquisition-produces	0.612	1	0.759	0.558	0.898	0.688
Perform analysis-on	0.695	0.842	0.761	0.571	0.667	0.615
Perform analysis-produces	0.692	0.857	0.765	0.5	0.692	0.581
Perform analysis-uses	1	0.478	0.647	0.7	0.333	0.451
Perform experiment-by	0.469	0.958	0.63	0.471	0.696	0.562
Perform experiment-participates in	0.423	0.785	0.55	0.3	0.462	0.364
Perform experiment-uses	0.692	0.947	0.799	0.511	0.649	0.571
Effect-affect	0.886	0.921	0.903	0.75	0.788	0.768
Effect-cause	0.783	0.775	0.779	0.649	0.649	0.649
Deactivate-affect	0.906	1	0.951	0.875	0.966	0.918
Deactivate-cause	0.869	1	0.93	0.905	0.95	0.918
Activate-affect	0.975	0.962	0.968	0.862	0.812	0.836
Activate-cause	0.833	0.882	0.857	0.566	0.625	0.594
ACQ-attribute	0.878	0.935	0.906	0.594	0.92	0.867
SEN-attribute	0.96	0.923	0.941	0.385	0.385	0.385
STP-attribute	0.96	0.827	0.889	1	0.815	0.898