Table 4.

The hyperparameters used in the experiments

Hyperparameter name	Value
Word embedding dimensionality	200
Position embedding dimensionality	20
Mini batch size	64
Latent variable size	32
Encoder Bi-LSTM hidden units	600
1st Decoder CNNs hidden units	300
2nd Decoder CNNs hidden units	600
3rd Decoder CNNs hidden units	1000
Classifier CNNs hidden units	300
Encoder dropout rate after embedding	0.5
Decoder dropout rate before output	0.5
Classifier dropout rate after embedding	0.5
Classifier dropout rate before output	0.5