Fusing Part-of-Speech Information in Low-Resource Neural Paraphrase Generation

Table 3

Performance of RNN-based models on ParaNMT datasets.

Dataset	Model	BLEU	ROUGE-1	ROUGE-2	ROUGE-L
ParaNMT50K	base	10.96 (±0.07)	35.12 (±0.13)	14.7 (±0.1)	32.84 (±0.14)
	add	11.27 (±0.1)^†††	35.75 (±0.16)^†††	15.1 (±0.11)^†††	33.44 (±0.14)^†††
	cat	11.24 (±0.07)^†††	35.54 (±0.12)^†††	15.0 (±0.11)^†††	33.28 (±0.13)^†††
	dc	11.09 (±0.16)^†	35.85 (±0.51)^††	15.05 (±0.37)^†	33.56 (±0.5)^††
ParaNMT100K	base	12.35 (±0.13)	38.98 (±0.12)	16.99 (±0.12)	36.29 (±0.1)
	add	12.6 (±0.12)^†††	39.4 (±0.24)^†††	17.36 (±0.16)^†††	36.72 (±0.23)^†††
	cat	12.59 (±0.08)^†††	39.31 (±0.11)^†††	17.26 (±0.09)^†††	36.62 (±0.1)^†††
	dc	12.3 (±0.13)	39.35 (±0.11)^†††	17.17 (±0.1)^††	36.64 (±0.1)^†††