Improving Neural Machine Translation with AMR Semantic Graphs

Table 2

Experimental results on Seq2Seq using one- and two-directional LSTMs.

Model	BLEU
	tst2013	tst2015
Song’s method	26.12	23.58
Uni-LSTM-AMR	26.97	24.80
Uni-LSTM-AMR-F	27.21	24.86
Uni-LSTM-AMR-B	26.61	24.66
Bi-LSTM-AMR	29.29	26.41
Bi-LSTM-AMR-F	28.67	26.20
Bi-LSTM-AMR-B	28.36	26.04

The bold values are the highest results for each group of models.