Table 8.

Hyperparameters for the Span-based Event Extractor.

Parameter	Value
Maximum sentence length, n	30
Maximum span length, M	6
Top-K spans per classifier	sentence token count
Batch size	100
Number of epochs	100
Learning rate	0.001
Optimizer	Adam
Maximum gradient L2-norm	100
BERT embedding dropout	0.3
bi-LSTM hidden size, $v_h$	200
bi-LSTM activation function	tanh
bi-LSTM dropout	0.3
Span classifier projection size, $v_s$	100
Span classifier activation function	ReLU
Span classifier dropout	0.3
Role classifier projection size, $v_r$	100
Role classifier activation function	ReLU
Role classifier dropout	0.3