Table 4:

Reference-based metrics. BLEU and SARI calculated using the human-generated reference sentences.

	BLEU	SARI
n-gram	66.31	33.40
GPT-1	68.19	33.57
GPT-2	66.45	33.40
NTS	70.17	27.67
ClinicalNTS	68.22	30.14
PhraseTable	53.37	27.70