Improving Neural Machine Translation with AMR Semantic Graphs

Table 6

Experimental results of aggregators on LightConvSeq2Seq-AMR.

Model	tst2013			tst2015
	MA	MP	GCN-A	MA	MP	GCN-A
LightConvSeq2Seq-AMR	28.20	28.46	27.76	25.49	25.05	25.33
LightConvSeq2Seq-AMR-F	27.82	27.71	27.59	25.96	25.27	25.39
LightConvSeq2Seq-AMR-B	28.25	27.84	28.27	25.66	25.67	25.52

The bold values are the highest results when evaluating each model on aggregators (i.e., MA, MP, and GCN-A) for testset (i.e., tst2013 or tst2015).