id	author	title	date	pages	extension	mime	words	sentences	flesch	summary	cache	txt
work_ygxu3etm6zehvdovnjdt3htauu	Zhaopeng Tu	Context Gates for Neural Machine Translation	2017	14	.pdf	application/pdf	6533	631	65	In neural machine translation (NMT), generation of a target word depends on both source Table 1: Source and target contexts are highly correlated to translation adequacy and fluency, respectively. 5src and 5tgt denote halving the contributions from the source and target contexts when generating the translation, respectively. contribution from the source context, the result further loses its adequacy by missing the partial translation "in the first two months of this year". each decoding step, NMT treats the source and target contexts equally, and thus ignores the different control the contributions of source and target contexts on the generation of target words (decoding) previous decoding state ti−1 and the previously generated word yi−1 constitute the target context.3 whether source and target contexts correlate to translation adequacy and fluency. each source word is translated.4 The decoding state implicitly models the notion of "coverage" by recurrently reading the time-dependent	./cache/work_ygxu3etm6zehvdovnjdt3htauu.pdf	./txt/work_ygxu3etm6zehvdovnjdt3htauu.txt