MIT Press

Table 6:

Document-level machine translation on En-De and Zh-En. (×) The randomly initialized Doc-MT model cannot produce translations aligned to the original sentences, so only document evaluation is possible.

(a) Sentence- and Document-level BLEU scores on En-De					(b) Document-level BLEU scores on Zh-En
Model	Random		mBART25		Model	Random	mBART25	HAN (2018)
Model	s-BLEU	d-BLEU	s-BLEU	d-BLEU	Model	d-BLEU	d-BLEU	d-BLEU
Sent-MT	34.5	35.9	36.4	38.0	Sent-MT	22.0	28.4	–
Doc-MT	×	7.7	37.1	38.5	Doc-MT	3.2	29.6	24.0

(a) Sentence- and Document-level BLEU scores on En-De					(b) Document-level BLEU scores on Zh-En
Model	Random		mBART25		Model	Random	mBART25	HAN (2018)
Model	s-BLEU	d-BLEU	s-BLEU	d-BLEU	Model	d-BLEU	d-BLEU	d-BLEU
Sent-MT	34.5	35.9	36.4	38.0	Sent-MT	22.0	28.4	–
Doc-MT	×	7.7	37.1	38.5	Doc-MT	3.2	29.6	24.0