MIT Press

Skip to Main Content

Table 1:

Test results of models. “(2M)” / “(20M)” denotes using the corresponding number of automatically labeled Gigaword data instances as additional training data.

Methods	BLEU
PBMT (Pourdamghani et al., 2016)	26.9
Tree2Str (Flanigan et al., 2016)	23.0
TSP (Song et al., 2016)	22.4
S2S+Anon (Konstas et al., 2017)	22.0
GraphLSTM (Song et al., 2018)	23.3
t-GCNSEQ (Damonte and Cohen, 2019)	23.9
g-GCNSEQ (Damonte and Cohen, 2019)	24.4

Transformer	17.7
Graph Transformer	25.9

S2S+Anon (2M) (Konstas et al., 2017)	32.3
S2S+Anon (20M) (Konstas et al., 2017)	33.8
S2S+Copy (2M) (Song et al., 2018)	31.7
GraphLSTM (2M) (Song et al., 2018)	33.6

Transformer (2M)	35.1
Graph Transformer (2M)	36.4

Methods	BLEU
PBMT (Pourdamghani et al., 2016)	26.9
Tree2Str (Flanigan et al., 2016)	23.0
TSP (Song et al., 2016)	22.4
S2S+Anon (Konstas et al., 2017)	22.0
GraphLSTM (Song et al., 2018)	23.3
t-GCNSEQ (Damonte and Cohen, 2019)	23.9
g-GCNSEQ (Damonte and Cohen, 2019)	24.4

Transformer	17.7
Graph Transformer	25.9

S2S+Anon (2M) (Konstas et al., 2017)	32.3
S2S+Anon (20M) (Konstas et al., 2017)	33.8
S2S+Copy (2M) (Song et al., 2018)	31.7
GraphLSTM (2M) (Song et al., 2018)	33.6

Transformer (2M)	35.1
Graph Transformer (2M)	36.4