id	sid	tid	token	lemma	pos
cord-027316-echxuw74	1	1	key	key	NN
cord-027316-echxuw74	1	2	:	:	:
cord-027316-echxuw74	2	1	cord-027316-echxuw74	cord-027316-echxuw74	NN
cord-027316-echxuw74	2	2	authors	author	NNS
cord-027316-echxuw74	2	3	:	:	:
cord-027316-echxuw74	2	4	Modarresi	Modarresi	NNP
cord-027316-echxuw74	2	5	,	,	,
cord-027316-echxuw74	2	6	Kourosh	Kourosh	NNP
cord-027316-echxuw74	2	7	title	title	NN
cord-027316-echxuw74	2	8	:	:	:
cord-027316-echxuw74	2	9	Detecting	detect	VBG
cord-027316-echxuw74	2	10	the	the	DT
cord-027316-echxuw74	2	11	Most	most	RBS
cord-027316-echxuw74	2	12	Insightful	insightful	JJ
cord-027316-echxuw74	2	13	Parts	part	NNS
cord-027316-echxuw74	2	14	of	of	IN
cord-027316-echxuw74	2	15	Documents	document	NNS
cord-027316-echxuw74	2	16	Using	use	VBG
cord-027316-echxuw74	2	17	a	a	DT
cord-027316-echxuw74	2	18	Regularized	regularized	JJ
cord-027316-echxuw74	2	19	Attention	attention	NN
cord-027316-echxuw74	2	20	-	-	HYPH
cord-027316-echxuw74	2	21	Based	base	VBN
cord-027316-echxuw74	2	22	Model	Model	NNP
cord-027316-echxuw74	2	23	date	date	NN
cord-027316-echxuw74	2	24	:	:	:
cord-027316-echxuw74	2	25	2020	2020	CD
cord-027316-echxuw74	2	26	-	-	HYPH
cord-027316-echxuw74	2	27	05	05	CD
cord-027316-echxuw74	2	28	-	-	HYPH
cord-027316-echxuw74	2	29	22	22	CD
cord-027316-echxuw74	2	30	journal	journal	NN
cord-027316-echxuw74	2	31	:	:	:
cord-027316-echxuw74	3	1	Computational	Computational	NNP
cord-027316-echxuw74	3	2	Science	Science	NNP
cord-027316-echxuw74	3	3	-	-	HYPH
cord-027316-echxuw74	3	4	ICCS	ICCS	NNP
cord-027316-echxuw74	3	5	2020	2020	CD
cord-027316-echxuw74	4	1	DOI	DOI	NNP
cord-027316-echxuw74	4	2	:	:	:
cord-027316-echxuw74	4	3	10.1007/978	10.1007/978	CD
cord-027316-echxuw74	4	4	-	-	HYPH
cord-027316-echxuw74	4	5	3	3	CD
cord-027316-echxuw74	4	6	-	-	HYPH
cord-027316-echxuw74	4	7	030	030	CD
cord-027316-echxuw74	4	8	-	-	HYPH
cord-027316-echxuw74	4	9	50420	50420	CD
cord-027316-echxuw74	4	10	-	-	HYPH
cord-027316-echxuw74	4	11	5_20	5_20	CD
cord-027316-echxuw74	4	12	sha	sha	NNP
cord-027316-echxuw74	4	13	:	:	:
cord-027316-echxuw74	5	1	9fd627849391d6fc245d5fba6852ea8d622756de	9fd627849391d6fc245d5fba6852ea8d622756de	CD
cord-027316-echxuw74	5	2	doc_id	doc_id	CD
cord-027316-echxuw74	5	3	:	:	:
cord-027316-echxuw74	5	4	27316	27316	CD
cord-027316-echxuw74	5	5	cord_uid	cord_uid	NNS
cord-027316-echxuw74	5	6	:	:	:
cord-027316-echxuw74	5	7	echxuw74	echxuw74	.
cord-027316-echxuw74	6	1	Every	every	DT
cord-027316-echxuw74	6	2	individual	individual	JJ
cord-027316-echxuw74	6	3	text	text	NN
cord-027316-echxuw74	6	4	or	or	CC
cord-027316-echxuw74	6	5	document	document	NN
cord-027316-echxuw74	6	6	is	be	VBZ
cord-027316-echxuw74	6	7	generated	generate	VBN
cord-027316-echxuw74	6	8	for	for	IN
cord-027316-echxuw74	6	9	specific	specific	JJ
cord-027316-echxuw74	6	10	purpose(s	purpose(s	NN
cord-027316-echxuw74	6	11	)	)	-RRB-
cord-027316-echxuw74	6	12	.	.	.
cord-027316-echxuw74	7	1	Sometime	sometime	RB
cord-027316-echxuw74	7	2	,	,	,
cord-027316-echxuw74	7	3	the	the	DT
cord-027316-echxuw74	7	4	text	text	NN
cord-027316-echxuw74	7	5	is	be	VBZ
cord-027316-echxuw74	7	6	deployed	deploy	VBN
cord-027316-echxuw74	7	7	to	to	TO
cord-027316-echxuw74	7	8	convey	convey	VB
cord-027316-echxuw74	7	9	a	a	DT
cord-027316-echxuw74	7	10	specific	specific	JJ
cord-027316-echxuw74	7	11	message	message	NN
cord-027316-echxuw74	7	12	about	about	IN
cord-027316-echxuw74	7	13	an	an	DT
cord-027316-echxuw74	7	14	event	event	NN
cord-027316-echxuw74	7	15	or	or	CC
cord-027316-echxuw74	7	16	a	a	DT
cord-027316-echxuw74	7	17	product	product	NN
cord-027316-echxuw74	7	18	.	.	.
cord-027316-echxuw74	8	1	Other	other	JJ
cord-027316-echxuw74	8	2	occasions	occasion	NNS
cord-027316-echxuw74	8	3	,	,	,
cord-027316-echxuw74	8	4	it	-PRON-	PRP
cord-027316-echxuw74	8	5	may	may	MD
cord-027316-echxuw74	8	6	be	be	VB
cord-027316-echxuw74	8	7	communicating	communicate	VBG
cord-027316-echxuw74	8	8	a	a	DT
cord-027316-echxuw74	8	9	scientific	scientific	JJ
cord-027316-echxuw74	8	10	breakthrough	breakthrough	NN
cord-027316-echxuw74	8	11	,	,	,
cord-027316-echxuw74	8	12	development	development	NN
cord-027316-echxuw74	8	13	or	or	CC
cord-027316-echxuw74	8	14	new	new	JJ
cord-027316-echxuw74	8	15	model	model	NN
cord-027316-echxuw74	8	16	and	and	CC
cord-027316-echxuw74	8	17	so	so	RB
cord-027316-echxuw74	8	18	on	on	RB
cord-027316-echxuw74	8	19	.	.	.
cord-027316-echxuw74	9	1	Given	give	VBN
cord-027316-echxuw74	9	2	any	any	DT
cord-027316-echxuw74	9	3	specific	specific	JJ
cord-027316-echxuw74	9	4	objective	objective	NN
cord-027316-echxuw74	9	5	,	,	,
cord-027316-echxuw74	9	6	the	the	DT
cord-027316-echxuw74	9	7	creators	creator	NNS
cord-027316-echxuw74	9	8	and	and	CC
cord-027316-echxuw74	9	9	the	the	DT
cord-027316-echxuw74	9	10	users	user	NNS
cord-027316-echxuw74	9	11	of	of	IN
cord-027316-echxuw74	9	12	documents	document	NNS
cord-027316-echxuw74	9	13	may	may	MD
cord-027316-echxuw74	9	14	like	like	VB
cord-027316-echxuw74	9	15	to	to	TO
cord-027316-echxuw74	9	16	know	know	VB
cord-027316-echxuw74	9	17	which	which	WDT
cord-027316-echxuw74	9	18	part(s	part(s	NNP
cord-027316-echxuw74	9	19	)	)	-RRB-
cord-027316-echxuw74	9	20	of	of	IN
cord-027316-echxuw74	9	21	the	the	DT
cord-027316-echxuw74	9	22	documents	document	NNS
cord-027316-echxuw74	9	23	are	be	VBP
cord-027316-echxuw74	9	24	more	more	RBR
cord-027316-echxuw74	9	25	influential	influential	JJ
cord-027316-echxuw74	9	26	in	in	IN
cord-027316-echxuw74	9	27	conveying	convey	VBG
cord-027316-echxuw74	9	28	their	-PRON-	PRP$
cord-027316-echxuw74	9	29	specific	specific	JJ
cord-027316-echxuw74	9	30	messages	message	NNS
cord-027316-echxuw74	9	31	or	or	CC
cord-027316-echxuw74	9	32	achieving	achieve	VBG
cord-027316-echxuw74	9	33	their	-PRON-	PRP$
cord-027316-echxuw74	9	34	objectives	objective	NNS
cord-027316-echxuw74	9	35	.	.	.
cord-027316-echxuw74	10	1	Understanding	understanding	NN
cord-027316-echxuw74	10	2	which	which	WDT
cord-027316-echxuw74	10	3	parts	part	NNS
cord-027316-echxuw74	10	4	of	of	IN
cord-027316-echxuw74	10	5	a	a	DT
cord-027316-echxuw74	10	6	document	document	NN
cord-027316-echxuw74	10	7	has	have	VBZ
cord-027316-echxuw74	10	8	more	more	JJR
cord-027316-echxuw74	10	9	impact	impact	NN
cord-027316-echxuw74	10	10	on	on	IN
cord-027316-echxuw74	10	11	the	the	DT
cord-027316-echxuw74	10	12	viewer	viewer	NN
cord-027316-echxuw74	10	13	’s	’s	POS
cord-027316-echxuw74	10	14	perception	perception	NN
cord-027316-echxuw74	10	15	would	would	MD
cord-027316-echxuw74	10	16	allow	allow	VB
cord-027316-echxuw74	10	17	the	the	DT
cord-027316-echxuw74	10	18	content	content	NN
cord-027316-echxuw74	10	19	creators	creator	NNS
cord-027316-echxuw74	10	20	to	to	TO
cord-027316-echxuw74	10	21	design	design	VB
cord-027316-echxuw74	10	22	more	more	RBR
cord-027316-echxuw74	10	23	effective	effective	JJ
cord-027316-echxuw74	10	24	content	content	NN
cord-027316-echxuw74	10	25	.	.	.
cord-027316-echxuw74	11	1	Detecting	detect	VBG
cord-027316-echxuw74	11	2	the	the	DT
cord-027316-echxuw74	11	3	more	more	RBR
cord-027316-echxuw74	11	4	impactful	impactful	JJ
cord-027316-echxuw74	11	5	parts	part	NNS
cord-027316-echxuw74	11	6	of	of	IN
cord-027316-echxuw74	11	7	a	a	DT
cord-027316-echxuw74	11	8	content	content	NN
cord-027316-echxuw74	11	9	would	would	MD
cord-027316-echxuw74	11	10	help	help	VB
cord-027316-echxuw74	11	11	content	content	NN
cord-027316-echxuw74	11	12	users	user	NNS
cord-027316-echxuw74	11	13	,	,	,
cord-027316-echxuw74	11	14	such	such	JJ
cord-027316-echxuw74	11	15	as	as	IN
cord-027316-echxuw74	11	16	advertisers	advertiser	NNS
cord-027316-echxuw74	11	17	,	,	,
cord-027316-echxuw74	11	18	to	to	TO
cord-027316-echxuw74	11	19	concentrate	concentrate	VB
cord-027316-echxuw74	11	20	their	-PRON-	PRP$
cord-027316-echxuw74	11	21	efforts	effort	NNS
cord-027316-echxuw74	11	22	more	more	RBR
cord-027316-echxuw74	11	23	on	on	IN
cord-027316-echxuw74	11	24	those	those	DT
cord-027316-echxuw74	11	25	parts	part	NNS
cord-027316-echxuw74	11	26	of	of	IN
cord-027316-echxuw74	11	27	the	the	DT
cord-027316-echxuw74	11	28	content	content	NN
cord-027316-echxuw74	11	29	and	and	CC
cord-027316-echxuw74	11	30	thus	thus	RB
cord-027316-echxuw74	11	31	to	to	TO
cord-027316-echxuw74	11	32	avoid	avoid	VB
cord-027316-echxuw74	11	33	spending	spend	VBG
cord-027316-echxuw74	11	34	resources	resource	NNS
cord-027316-echxuw74	11	35	on	on	IN
cord-027316-echxuw74	11	36	the	the	DT
cord-027316-echxuw74	11	37	rest	rest	NN
cord-027316-echxuw74	11	38	of	of	IN
cord-027316-echxuw74	11	39	the	the	DT
cord-027316-echxuw74	11	40	document	document	NN
cord-027316-echxuw74	11	41	.	.	.
cord-027316-echxuw74	12	1	This	this	DT
cord-027316-echxuw74	12	2	work	work	NN
cord-027316-echxuw74	12	3	uses	use	VBZ
cord-027316-echxuw74	12	4	a	a	DT
cord-027316-echxuw74	12	5	regularized	regularized	JJ
cord-027316-echxuw74	12	6	attention	attention	NN
cord-027316-echxuw74	12	7	-	-	HYPH
cord-027316-echxuw74	12	8	based	base	VBN
cord-027316-echxuw74	12	9	method	method	NN
cord-027316-echxuw74	12	10	to	to	TO
cord-027316-echxuw74	12	11	detect	detect	VB
cord-027316-echxuw74	12	12	the	the	DT
cord-027316-echxuw74	12	13	most	most	RBS
cord-027316-echxuw74	12	14	influential	influential	JJ
cord-027316-echxuw74	12	15	part(s	part(s	NNP
cord-027316-echxuw74	12	16	)	)	-RRB-
cord-027316-echxuw74	12	17	of	of	IN
cord-027316-echxuw74	12	18	any	any	DT
cord-027316-echxuw74	12	19	given	give	VBN
cord-027316-echxuw74	12	20	document	document	NN
cord-027316-echxuw74	12	21	or	or	CC
cord-027316-echxuw74	12	22	text	text	NN
cord-027316-echxuw74	12	23	.	.	.
cord-027316-echxuw74	13	1	The	the	DT
cord-027316-echxuw74	13	2	model	model	NN
cord-027316-echxuw74	13	3	uses	use	VBZ
cord-027316-echxuw74	13	4	an	an	DT
cord-027316-echxuw74	13	5	encoder	encoder	NN
cord-027316-echxuw74	13	6	-	-	HYPH
cord-027316-echxuw74	13	7	decoder	decoder	NN
cord-027316-echxuw74	13	8	architecture	architecture	NN
cord-027316-echxuw74	13	9	based	base	VBN
cord-027316-echxuw74	13	10	on	on	IN
cord-027316-echxuw74	13	11	attention	attention	NN
cord-027316-echxuw74	13	12	-	-	HYPH
cord-027316-echxuw74	13	13	based	base	VBN
cord-027316-echxuw74	13	14	decoder	decoder	NN
cord-027316-echxuw74	13	15	with	with	IN
cord-027316-echxuw74	13	16	regularization	regularization	NN
cord-027316-echxuw74	13	17	applied	apply	VBN
cord-027316-echxuw74	13	18	to	to	IN
cord-027316-echxuw74	13	19	the	the	DT
cord-027316-echxuw74	13	20	corresponding	correspond	VBG
cord-027316-echxuw74	13	21	weights	weight	NNS
cord-027316-echxuw74	13	22	.	.	.
cord-027316-echxuw74	14	1	The	the	DT
cord-027316-echxuw74	14	2	main	main	JJ
cord-027316-echxuw74	14	3	purpose	purpose	NN
cord-027316-echxuw74	14	4	of	of	IN
cord-027316-echxuw74	14	5	NLP	NLP	NNP
cord-027316-echxuw74	14	6	(	(	-LRB-
cord-027316-echxuw74	14	7	Natural	Natural	NNP
cord-027316-echxuw74	14	8	Language	Language	NNP
cord-027316-echxuw74	14	9	Processing	Processing	NNP
cord-027316-echxuw74	14	10	)	)	-RRB-
cord-027316-echxuw74	14	11	and	and	CC
cord-027316-echxuw74	14	12	NLU	NLU	NNP
cord-027316-echxuw74	14	13	(	(	-LRB-
cord-027316-echxuw74	14	14	Natural	Natural	NNP
cord-027316-echxuw74	14	15	Language	Language	NNP
cord-027316-echxuw74	14	16	Understanding	Understanding	NNP
cord-027316-echxuw74	14	17	)	)	-RRB-
cord-027316-echxuw74	14	18	is	be	VBZ
cord-027316-echxuw74	14	19	to	to	TO
cord-027316-echxuw74	14	20	understand	understand	VB
cord-027316-echxuw74	14	21	the	the	DT
cord-027316-echxuw74	14	22	language	language	NN
cord-027316-echxuw74	14	23	.	.	.
cord-027316-echxuw74	15	1	More	more	RBR
cord-027316-echxuw74	15	2	specifically	specifically	RB
cord-027316-echxuw74	15	3	,	,	,
cord-027316-echxuw74	15	4	they	-PRON-	PRP
cord-027316-echxuw74	15	5	are	be	VBP
cord-027316-echxuw74	15	6	focused	focus	VBN
cord-027316-echxuw74	15	7	on	on	IN
cord-027316-echxuw74	15	8	not	not	RB
cord-027316-echxuw74	15	9	just	just	RB
cord-027316-echxuw74	15	10	to	to	TO
cord-027316-echxuw74	15	11	see	see	VB
cord-027316-echxuw74	15	12	the	the	DT
cord-027316-echxuw74	15	13	context	context	NN
cord-027316-echxuw74	15	14	of	of	IN
cord-027316-echxuw74	15	15	text	text	NN
cord-027316-echxuw74	15	16	but	but	CC
cord-027316-echxuw74	15	17	also	also	RB
cord-027316-echxuw74	15	18	to	to	TO
cord-027316-echxuw74	15	19	see	see	VB
cord-027316-echxuw74	15	20	how	how	WRB
cord-027316-echxuw74	15	21	human	human	JJ
cord-027316-echxuw74	15	22	uses	use	VBZ
cord-027316-echxuw74	15	23	the	the	DT
cord-027316-echxuw74	15	24	language	language	NN
cord-027316-echxuw74	15	25	in	in	IN
cord-027316-echxuw74	15	26	daily	daily	JJ
cord-027316-echxuw74	15	27	life	life	NN
cord-027316-echxuw74	15	28	.	.	.
cord-027316-echxuw74	16	1	Thus	thus	RB
cord-027316-echxuw74	16	2	,	,	,
cord-027316-echxuw74	16	3	among	among	IN
cord-027316-echxuw74	16	4	other	other	JJ
cord-027316-echxuw74	16	5	ways	way	NNS
cord-027316-echxuw74	16	6	of	of	IN
cord-027316-echxuw74	16	7	utilizing	utilize	VBG
cord-027316-echxuw74	16	8	this	this	DT
cord-027316-echxuw74	16	9	,	,	,
cord-027316-echxuw74	16	10	we	-PRON-	PRP
cord-027316-echxuw74	16	11	could	could	MD
cord-027316-echxuw74	16	12	provide	provide	VB
cord-027316-echxuw74	16	13	an	an	DT
cord-027316-echxuw74	16	14	optimal	optimal	JJ
cord-027316-echxuw74	16	15	online	online	NN
cord-027316-echxuw74	16	16	experience	experience	NN
cord-027316-echxuw74	16	17	addressing	address	VBG
cord-027316-echxuw74	16	18	needs	need	NNS
cord-027316-echxuw74	16	19	of	of	IN
cord-027316-echxuw74	16	20	users	user	NNS
cord-027316-echxuw74	16	21	'	'	POS
cord-027316-echxuw74	16	22	digital	digital	JJ
cord-027316-echxuw74	16	23	experience	experience	NN
cord-027316-echxuw74	16	24	.	.	.
cord-027316-echxuw74	17	1	Language	language	NN
cord-027316-echxuw74	17	2	processing	processing	NN
cord-027316-echxuw74	17	3	and	and	CC
cord-027316-echxuw74	17	4	understanding	understanding	NN
cord-027316-echxuw74	17	5	is	be	VBZ
cord-027316-echxuw74	17	6	much	much	RB
cord-027316-echxuw74	17	7	more	more	RBR
cord-027316-echxuw74	17	8	complex	complex	JJ
cord-027316-echxuw74	17	9	than	than	IN
cord-027316-echxuw74	17	10	many	many	JJ
cord-027316-echxuw74	17	11	other	other	JJ
cord-027316-echxuw74	17	12	applications	application	NNS
cord-027316-echxuw74	17	13	in	in	IN
cord-027316-echxuw74	17	14	machine	machine	NN
cord-027316-echxuw74	17	15	learning	learn	VBG
cord-027316-echxuw74	17	16	such	such	JJ
cord-027316-echxuw74	17	17	as	as	IN
cord-027316-echxuw74	17	18	image	image	NN
cord-027316-echxuw74	17	19	classification	classification	NN
cord-027316-echxuw74	17	20	as	as	IN
cord-027316-echxuw74	17	21	NLP	NLP	NNP
cord-027316-echxuw74	17	22	and	and	CC
cord-027316-echxuw74	17	23	NLU	NLU	NNP
cord-027316-echxuw74	17	24	involve	involve	VBP
cord-027316-echxuw74	17	25	deeper	deep	JJR
cord-027316-echxuw74	17	26	context	context	NN
cord-027316-echxuw74	17	27	analysis	analysis	NN
cord-027316-echxuw74	17	28	than	than	IN
cord-027316-echxuw74	17	29	other	other	JJ
cord-027316-echxuw74	17	30	machine	machine	NN
cord-027316-echxuw74	17	31	learning	learning	NN
cord-027316-echxuw74	17	32	applications	application	NNS
cord-027316-echxuw74	17	33	.	.	.
cord-027316-echxuw74	18	1	This	this	DT
cord-027316-echxuw74	18	2	paper	paper	NN
cord-027316-echxuw74	18	3	is	be	VBZ
cord-027316-echxuw74	18	4	written	write	VBN
cord-027316-echxuw74	18	5	as	as	IN
cord-027316-echxuw74	18	6	a	a	DT
cord-027316-echxuw74	18	7	short	short	JJ
cord-027316-echxuw74	18	8	paper	paper	NN
cord-027316-echxuw74	18	9	and	and	CC
cord-027316-echxuw74	18	10	focuses	focus	VBZ
cord-027316-echxuw74	18	11	on	on	IN
cord-027316-echxuw74	18	12	explaining	explain	VBG
cord-027316-echxuw74	18	13	only	only	RB
cord-027316-echxuw74	18	14	the	the	DT
cord-027316-echxuw74	18	15	parts	part	NNS
cord-027316-echxuw74	18	16	that	that	WDT
cord-027316-echxuw74	18	17	are	be	VBP
cord-027316-echxuw74	18	18	contribution	contribution	NN
cord-027316-echxuw74	18	19	of	of	IN
cord-027316-echxuw74	18	20	this	this	DT
cord-027316-echxuw74	18	21	paper	paper	NN
cord-027316-echxuw74	18	22	to	to	IN
cord-027316-echxuw74	18	23	the	the	DT
cord-027316-echxuw74	18	24	state	state	NN
cord-027316-echxuw74	18	25	-	-	HYPH
cord-027316-echxuw74	18	26	of	of	IN
cord-027316-echxuw74	18	27	-	-	HYPH
cord-027316-echxuw74	18	28	the	the	DT
cord-027316-echxuw74	18	29	art	art	NN
cord-027316-echxuw74	18	30	.	.	.
cord-027316-echxuw74	19	1	Thus	thus	RB
cord-027316-echxuw74	19	2	,	,	,
cord-027316-echxuw74	19	3	this	this	DT
cord-027316-echxuw74	19	4	paper	paper	NN
cord-027316-echxuw74	19	5	does	do	VBZ
cord-027316-echxuw74	19	6	not	not	RB
cord-027316-echxuw74	19	7	describe	describe	VB
cord-027316-echxuw74	19	8	the	the	DT
cord-027316-echxuw74	19	9	state	state	NN
cord-027316-echxuw74	19	10	-	-	HYPH
cord-027316-echxuw74	19	11	of	of	IN
cord-027316-echxuw74	19	12	-	-	HYPH
cord-027316-echxuw74	19	13	the	the	DT
cord-027316-echxuw74	19	14	-	-	HYPH
cord-027316-echxuw74	19	15	art	art	NN
cord-027316-echxuw74	19	16	works	work	NNS
cord-027316-echxuw74	19	17	in	in	IN
cord-027316-echxuw74	19	18	details	detail	NNS
cord-027316-echxuw74	19	19	and	and	CC
cord-027316-echxuw74	19	20	uses	use	VBZ
cord-027316-echxuw74	19	21	those	those	DT
cord-027316-echxuw74	19	22	works	work	NNS
cord-027316-echxuw74	19	23	[	[	-LRB-
cord-027316-echxuw74	19	24	2	2	CD
cord-027316-echxuw74	19	25	,	,	,
cord-027316-echxuw74	19	26	4	4	CD
cord-027316-echxuw74	19	27	,	,	,
cord-027316-echxuw74	19	28	5	5	CD
cord-027316-echxuw74	19	29	,	,	,
cord-027316-echxuw74	19	30	8	8	CD
cord-027316-echxuw74	19	31	,	,	,
cord-027316-echxuw74	19	32	53	53	CD
cord-027316-echxuw74	19	33	,	,	,
cord-027316-echxuw74	19	34	60	60	CD
cord-027316-echxuw74	19	35	,	,	,
cord-027316-echxuw74	19	36	66	66	CD
cord-027316-echxuw74	19	37	,	,	,
cord-027316-echxuw74	19	38	70	70	CD
cord-027316-echxuw74	19	39	,	,	,
cord-027316-echxuw74	19	40	74	74	CD
cord-027316-echxuw74	19	41	,	,	,
cord-027316-echxuw74	19	42	84	84	CD
cord-027316-echxuw74	19	43	]	]	-RRB-
cord-027316-echxuw74	19	44	to	to	TO
cord-027316-echxuw74	19	45	build	build	VB
cord-027316-echxuw74	19	46	its	-PRON-	PRP$
cord-027316-echxuw74	19	47	model	model	NN
cord-027316-echxuw74	19	48	as	as	IN
cord-027316-echxuw74	19	49	a	a	DT
cord-027316-echxuw74	19	50	modification	modification	NN
cord-027316-echxuw74	19	51	and	and	CC
cord-027316-echxuw74	19	52	extension	extension	NN
cord-027316-echxuw74	19	53	of	of	IN
cord-027316-echxuw74	19	54	the	the	DT
cord-027316-echxuw74	19	55	state	state	NN
cord-027316-echxuw74	19	56	of	of	IN
cord-027316-echxuw74	19	57	the	the	DT
cord-027316-echxuw74	19	58	art	art	NN
cord-027316-echxuw74	19	59	.	.	.
cord-027316-echxuw74	20	1	Therefore	therefore	RB
cord-027316-echxuw74	20	2	,	,	,
cord-027316-echxuw74	20	3	a	a	DT
cord-027316-echxuw74	20	4	comprehensive	comprehensive	JJ
cord-027316-echxuw74	20	5	set	set	NN
cord-027316-echxuw74	20	6	of	of	IN
cord-027316-echxuw74	20	7	reference	reference	NN
cord-027316-echxuw74	20	8	works	work	NNS
cord-027316-echxuw74	20	9	have	have	VBP
cord-027316-echxuw74	20	10	been	be	VBN
cord-027316-echxuw74	20	11	added	add	VBN
cord-027316-echxuw74	20	12	for	for	IN
cord-027316-echxuw74	20	13	anyone	anyone	NN
cord-027316-echxuw74	20	14	interested	interested	JJ
cord-027316-echxuw74	20	15	in	in	IN
cord-027316-echxuw74	20	16	learning	learn	VBG
cord-027316-echxuw74	20	17	more	more	JJR
cord-027316-echxuw74	20	18	details	detail	NNS
cord-027316-echxuw74	20	19	of	of	IN
cord-027316-echxuw74	20	20	the	the	DT
cord-027316-echxuw74	20	21	previous	previous	JJ
cord-027316-echxuw74	20	22	state	state	NN
cord-027316-echxuw74	20	23	of	of	IN
cord-027316-echxuw74	20	24	the	the	DT
cord-027316-echxuw74	20	25	art	art	NN
cord-027316-echxuw74	20	26	research	research	NN
cord-027316-echxuw74	20	27	[	[	-LRB-
cord-027316-echxuw74	20	28	3	3	CD
cord-027316-echxuw74	20	29	,	,	,
cord-027316-echxuw74	20	30	5	5	CD
cord-027316-echxuw74	20	31	,	,	,
cord-027316-echxuw74	20	32	10	10	CD
cord-027316-echxuw74	20	33	,	,	,
cord-027316-echxuw74	20	34	17	17	CD
cord-027316-echxuw74	20	35	,	,	,
cord-027316-echxuw74	20	36	33	33	CD
cord-027316-echxuw74	20	37	,	,	,
cord-027316-echxuw74	20	38	48	48	CD
cord-027316-echxuw74	20	39	,	,	,
cord-027316-echxuw74	20	40	49	49	CD
cord-027316-echxuw74	20	41	,	,	,
cord-027316-echxuw74	20	42	61	61	CD
cord-027316-echxuw74	20	43	-	-	SYM
cord-027316-echxuw74	20	44	63	63	CD
cord-027316-echxuw74	20	45	,	,	,
cord-027316-echxuw74	20	46	67	67	CD
cord-027316-echxuw74	20	47	-	-	SYM
cord-027316-echxuw74	20	48	73	73	CD
cord-027316-echxuw74	20	49	,	,	,
cord-027316-echxuw74	20	50	76	76	CD
cord-027316-echxuw74	20	51	,	,	,
cord-027316-echxuw74	20	52	77	77	CD
cord-027316-echxuw74	20	53	,	,	,
cord-027316-echxuw74	20	54	90	90	CD
cord-027316-echxuw74	20	55	,	,	,
cord-027316-echxuw74	20	56	91	91	CD
cord-027316-echxuw74	20	57	,	,	,
cord-027316-echxuw74	20	58	93	93	CD
cord-027316-echxuw74	20	59	]	]	-RRB-
cord-027316-echxuw74	20	60	.	.	.
cord-027316-echxuw74	21	1	Deep	deep	JJ
cord-027316-echxuw74	21	2	Learning	Learning	NNP
cord-027316-echxuw74	21	3	has	have	VBZ
cord-027316-echxuw74	21	4	become	become	VBN
cord-027316-echxuw74	21	5	a	a	DT
cord-027316-echxuw74	21	6	main	main	JJ
cord-027316-echxuw74	21	7	model	model	NN
cord-027316-echxuw74	21	8	in	in	IN
cord-027316-echxuw74	21	9	natural	natural	JJ
cord-027316-echxuw74	21	10	language	language	NN
cord-027316-echxuw74	21	11	processing	processing	NN
cord-027316-echxuw74	21	12	applications	application	NNS
cord-027316-echxuw74	21	13	[	[	-LRB-
cord-027316-echxuw74	21	14	6	6	CD
cord-027316-echxuw74	21	15	,	,	,
cord-027316-echxuw74	21	16	7	7	CD
cord-027316-echxuw74	21	17	,	,	,
cord-027316-echxuw74	21	18	11	11	CD
cord-027316-echxuw74	21	19	,	,	,
cord-027316-echxuw74	21	20	22	22	CD
cord-027316-echxuw74	21	21	,	,	,
cord-027316-echxuw74	21	22	38	38	CD
cord-027316-echxuw74	21	23	,	,	,
cord-027316-echxuw74	21	24	55	55	CD
cord-027316-echxuw74	21	25	,	,	,
cord-027316-echxuw74	21	26	64	64	CD
cord-027316-echxuw74	21	27	,	,	,
cord-027316-echxuw74	21	28	71	71	CD
cord-027316-echxuw74	21	29	,	,	,
cord-027316-echxuw74	21	30	75	75	CD
cord-027316-echxuw74	21	31	,	,	,
cord-027316-echxuw74	21	32	78	78	CD
cord-027316-echxuw74	21	33	-	-	SYM
cord-027316-echxuw74	21	34	81	81	CD
cord-027316-echxuw74	21	35	,	,	,
cord-027316-echxuw74	21	36	85	85	CD
cord-027316-echxuw74	21	37	,	,	,
cord-027316-echxuw74	21	38	88	88	CD
cord-027316-echxuw74	21	39	,	,	,
cord-027316-echxuw74	21	40	94	94	CD
cord-027316-echxuw74	21	41	]	]	-RRB-
cord-027316-echxuw74	21	42	.	.	.
cord-027316-echxuw74	22	1	Among	among	IN
cord-027316-echxuw74	22	2	deep	deep	JJ
cord-027316-echxuw74	22	3	learning	learning	NN
cord-027316-echxuw74	22	4	models	model	NNS
cord-027316-echxuw74	22	5	,	,	,
cord-027316-echxuw74	22	6	often	often	RB
cord-027316-echxuw74	22	7	RNN	RNN	NNP
cord-027316-echxuw74	22	8	-	-	HYPH
cord-027316-echxuw74	22	9	based	base	VBN
cord-027316-echxuw74	22	10	models	model	NNS
cord-027316-echxuw74	22	11	like	like	IN
cord-027316-echxuw74	22	12	LSTM	LSTM	NNP
cord-027316-echxuw74	22	13	and	and	CC
cord-027316-echxuw74	22	14	GRU	GRU	NNP
cord-027316-echxuw74	22	15	have	have	VBP
cord-027316-echxuw74	22	16	been	be	VBN
cord-027316-echxuw74	22	17	deployed	deploy	VBN
cord-027316-echxuw74	22	18	for	for	IN
cord-027316-echxuw74	22	19	text	text	NN
cord-027316-echxuw74	22	20	analysis	analysis	NN
cord-027316-echxuw74	22	21	[	[	-LRB-
cord-027316-echxuw74	22	22	9	9	CD
cord-027316-echxuw74	22	23	,	,	,
cord-027316-echxuw74	22	24	13	13	CD
cord-027316-echxuw74	22	25	,	,	,
cord-027316-echxuw74	22	26	16	16	CD
cord-027316-echxuw74	22	27	,	,	,
cord-027316-echxuw74	22	28	23	23	CD
cord-027316-echxuw74	22	29	,	,	,
cord-027316-echxuw74	22	30	32	32	CD
cord-027316-echxuw74	22	31	,	,	,
cord-027316-echxuw74	22	32	39	39	CD
cord-027316-echxuw74	22	33	-	-	SYM
cord-027316-echxuw74	22	34	42	42	CD
cord-027316-echxuw74	22	35	,	,	,
cord-027316-echxuw74	22	36	50	50	CD
cord-027316-echxuw74	22	37	,	,	,
cord-027316-echxuw74	22	38	51	51	CD
cord-027316-echxuw74	22	39	,	,	,
cord-027316-echxuw74	22	40	58	58	CD
cord-027316-echxuw74	22	41	,	,	,
cord-027316-echxuw74	22	42	59	59	CD
cord-027316-echxuw74	22	43	]	]	-RRB-
cord-027316-echxuw74	22	44	.	.	.
cord-027316-echxuw74	23	1	Though	though	RB
cord-027316-echxuw74	23	2	,	,	,
cord-027316-echxuw74	23	3	modified	modify	VBN
cord-027316-echxuw74	23	4	version	version	NN
cord-027316-echxuw74	23	5	of	of	IN
cord-027316-echxuw74	23	6	RNN	RNN	NNP
cord-027316-echxuw74	23	7	like	like	IN
cord-027316-echxuw74	23	8	LSTM	LSTM	NNP
cord-027316-echxuw74	23	9	and	and	CC
cord-027316-echxuw74	23	10	GRU	GRU	NNP
cord-027316-echxuw74	23	11	have	have	VBP
cord-027316-echxuw74	23	12	been	be	VBN
cord-027316-echxuw74	23	13	improvement	improvement	NN
cord-027316-echxuw74	23	14	over	over	IN
cord-027316-echxuw74	23	15	RNN	RNN	NNP
cord-027316-echxuw74	23	16	(	(	-LRB-
cord-027316-echxuw74	23	17	recurrent	recurrent	JJ
cord-027316-echxuw74	23	18	neural	neural	JJ
cord-027316-echxuw74	23	19	networks	network	NNS
cord-027316-echxuw74	23	20	)	)	-RRB-
cord-027316-echxuw74	23	21	in	in	IN
cord-027316-echxuw74	23	22	dealing	deal	VBG
cord-027316-echxuw74	23	23	with	with	IN
cord-027316-echxuw74	23	24	vanishing	vanish	VBG
cord-027316-echxuw74	23	25	gradients	gradient	NNS
cord-027316-echxuw74	23	26	and	and	CC
cord-027316-echxuw74	23	27	long	long	JJ
cord-027316-echxuw74	23	28	-	-	HYPH
cord-027316-echxuw74	23	29	term	term	NN
cord-027316-echxuw74	23	30	memory	memory	NN
cord-027316-echxuw74	23	31	loss	loss	NN
cord-027316-echxuw74	23	32	,	,	,
cord-027316-echxuw74	23	33	still	still	RB
cord-027316-echxuw74	23	34	they	-PRON-	PRP
cord-027316-echxuw74	23	35	suffer	suffer	VBP
cord-027316-echxuw74	23	36	from	from	IN
cord-027316-echxuw74	23	37	many	many	JJ
cord-027316-echxuw74	23	38	deficiencies	deficiency	NNS
cord-027316-echxuw74	23	39	.	.	.
cord-027316-echxuw74	24	1	As	as	IN
cord-027316-echxuw74	24	2	a	a	DT
cord-027316-echxuw74	24	3	specific	specific	JJ
cord-027316-echxuw74	24	4	example	example	NN
cord-027316-echxuw74	24	5	,	,	,
cord-027316-echxuw74	24	6	a	a	DT
cord-027316-echxuw74	24	7	RNN	RNN	NNP
cord-027316-echxuw74	24	8	-	-	HYPH
cord-027316-echxuw74	24	9	based	base	VBN
cord-027316-echxuw74	24	10	encoder	encoder	NN
cord-027316-echxuw74	24	11	-	-	HYPH
cord-027316-echxuw74	24	12	decoder	decoder	NN
cord-027316-echxuw74	24	13	architecture	architecture	NN
cord-027316-echxuw74	24	14	uses	use	VBZ
cord-027316-echxuw74	24	15	the	the	DT
cord-027316-echxuw74	24	16	encoded	encode	VBN
cord-027316-echxuw74	24	17	vector	vector	NN
cord-027316-echxuw74	24	18	(	(	-LRB-
cord-027316-echxuw74	24	19	feature	feature	NN
cord-027316-echxuw74	24	20	vector	vector	NN
cord-027316-echxuw74	24	21	)	)	-RRB-
cord-027316-echxuw74	24	22	,	,	,
cord-027316-echxuw74	24	23	computed	compute	VBN
cord-027316-echxuw74	24	24	at	at	IN
cord-027316-echxuw74	24	25	the	the	DT
cord-027316-echxuw74	24	26	end	end	NN
cord-027316-echxuw74	24	27	of	of	IN
cord-027316-echxuw74	24	28	encoder	encoder	NN
cord-027316-echxuw74	24	29	,	,	,
cord-027316-echxuw74	24	30	as	as	IN
cord-027316-echxuw74	24	31	the	the	DT
cord-027316-echxuw74	24	32	input	input	NN
cord-027316-echxuw74	24	33	to	to	IN
cord-027316-echxuw74	24	34	the	the	DT
cord-027316-echxuw74	24	35	decoder	decoder	NN
cord-027316-echxuw74	24	36	and	and	CC
cord-027316-echxuw74	24	37	uses	use	VBZ
cord-027316-echxuw74	24	38	this	this	DT
cord-027316-echxuw74	24	39	vector	vector	NN
cord-027316-echxuw74	24	40	as	as	IN
cord-027316-echxuw74	24	41	a	a	DT
cord-027316-echxuw74	24	42	compressed	compress	VBN
cord-027316-echxuw74	24	43	representation	representation	NN
cord-027316-echxuw74	24	44	of	of	IN
cord-027316-echxuw74	24	45	all	all	PDT
cord-027316-echxuw74	24	46	the	the	DT
cord-027316-echxuw74	24	47	data	datum	NNS
cord-027316-echxuw74	24	48	and	and	CC
cord-027316-echxuw74	24	49	information	information	NN
cord-027316-echxuw74	24	50	from	from	IN
cord-027316-echxuw74	24	51	encoder	encoder	NN
cord-027316-echxuw74	24	52	(	(	-LRB-
cord-027316-echxuw74	24	53	input	input	NN
cord-027316-echxuw74	24	54	)	)	-RRB-
cord-027316-echxuw74	24	55	.	.	.
cord-027316-echxuw74	25	1	This	this	DT
cord-027316-echxuw74	25	2	ignores	ignore	VBZ
cord-027316-echxuw74	25	3	the	the	DT
cord-027316-echxuw74	25	4	possibility	possibility	NN
cord-027316-echxuw74	25	5	of	of	IN
cord-027316-echxuw74	25	6	looking	look	VBG
cord-027316-echxuw74	25	7	at	at	RB
cord-027316-echxuw74	25	8	all	all	DT
cord-027316-echxuw74	25	9	previous	previous	JJ
cord-027316-echxuw74	25	10	sequences	sequence	NNS
cord-027316-echxuw74	25	11	of	of	IN
cord-027316-echxuw74	25	12	the	the	DT
cord-027316-echxuw74	25	13	encoder	encoder	NN
cord-027316-echxuw74	25	14	and	and	CC
cord-027316-echxuw74	25	15	thus	thus	RB
cord-027316-echxuw74	25	16	suffers	suffer	VBZ
cord-027316-echxuw74	25	17	from	from	IN
cord-027316-echxuw74	25	18	information	information	NN
cord-027316-echxuw74	25	19	bottleneck	bottleneck	NN
cord-027316-echxuw74	25	20	leading	lead	VBG
cord-027316-echxuw74	25	21	to	to	IN
cord-027316-echxuw74	25	22	low	low	JJ
cord-027316-echxuw74	25	23	precision	precision	NN
cord-027316-echxuw74	25	24	,	,	,
cord-027316-echxuw74	25	25	especially	especially	RB
cord-027316-echxuw74	25	26	for	for	IN
cord-027316-echxuw74	25	27	texts	text	NNS
cord-027316-echxuw74	25	28	of	of	IN
cord-027316-echxuw74	25	29	medium	medium	JJ
cord-027316-echxuw74	25	30	or	or	CC
cord-027316-echxuw74	25	31	long	long	JJ
cord-027316-echxuw74	25	32	sequences	sequence	NNS
cord-027316-echxuw74	25	33	.	.	.
cord-027316-echxuw74	26	1	To	to	TO
cord-027316-echxuw74	26	2	address	address	VB
cord-027316-echxuw74	26	3	this	this	DT
cord-027316-echxuw74	26	4	problem	problem	NN
cord-027316-echxuw74	26	5	,	,	,
cord-027316-echxuw74	26	6	global	global	JJ
cord-027316-echxuw74	26	7	attention	attention	NN
cord-027316-echxuw74	26	8	-	-	HYPH
cord-027316-echxuw74	26	9	based	base	VBN
cord-027316-echxuw74	26	10	model	model	NN
cord-027316-echxuw74	26	11	[	[	-LRB-
cord-027316-echxuw74	26	12	2	2	CD
cord-027316-echxuw74	26	13	,	,	,
cord-027316-echxuw74	26	14	5	5	CD
cord-027316-echxuw74	26	15	]	]	-RRB-
cord-027316-echxuw74	26	16	where	where	WRB
cord-027316-echxuw74	26	17	each	each	DT
cord-027316-echxuw74	26	18	of	of	IN
cord-027316-echxuw74	26	19	the	the	DT
cord-027316-echxuw74	26	20	encoder	encoder	NN
cord-027316-echxuw74	26	21	sequence	sequence	NN
cord-027316-echxuw74	26	22	uses	use	VBZ
cord-027316-echxuw74	26	23	all	all	DT
cord-027316-echxuw74	26	24	of	of	IN
cord-027316-echxuw74	26	25	the	the	DT
cord-027316-echxuw74	26	26	encoder	encoder	NN
cord-027316-echxuw74	26	27	sequences	sequence	NNS
cord-027316-echxuw74	26	28	.	.	.
cord-027316-echxuw74	27	1	Figure	figure	NN
cord-027316-echxuw74	27	2	1	1	CD
cord-027316-echxuw74	27	3	shows	show	VBZ
cord-027316-echxuw74	27	4	an	an	DT
cord-027316-echxuw74	27	5	attention	attention	NN
cord-027316-echxuw74	27	6	-	-	HYPH
cord-027316-echxuw74	27	7	based	base	VBN
cord-027316-echxuw74	27	8	model	model	NN
cord-027316-echxuw74	27	9	.	.	.
cord-027316-echxuw74	28	1	Where	where	WRB
cord-027316-echxuw74	28	2	i	i	PRP
cord-027316-echxuw74	28	3	=	=	VBZ
cord-027316-echxuw74	28	4	1	1	CD
cord-027316-echxuw74	28	5	:	:	:
cord-027316-echxuw74	28	6	n	n	CC
cord-027316-echxuw74	28	7	is	be	VBZ
cord-027316-echxuw74	28	8	the	the	DT
cord-027316-echxuw74	28	9	encoder	encoder	NN
cord-027316-echxuw74	28	10	sequences	sequence	NNS
cord-027316-echxuw74	28	11	and	and	CC
cord-027316-echxuw74	28	12	,	,	,
cord-027316-echxuw74	28	13	t	t	NN
cord-027316-echxuw74	28	14	=	=	SYM
cord-027316-echxuw74	28	15	1	1	CD
cord-027316-echxuw74	29	1	:	:	:
cord-027316-echxuw74	29	2	m	m	NNP
cord-027316-echxuw74	29	3	represents	represent	VBZ
cord-027316-echxuw74	29	4	the	the	DT
cord-027316-echxuw74	29	5	decoder	decoder	NN
cord-027316-echxuw74	29	6	sequences	sequence	NNS
cord-027316-echxuw74	29	7	.	.	.
cord-027316-echxuw74	30	1	Each	each	DT
cord-027316-echxuw74	30	2	of	of	IN
cord-027316-echxuw74	30	3	the	the	DT
cord-027316-echxuw74	30	4	encoder	encoder	NN
cord-027316-echxuw74	30	5	states	state	VBZ
cord-027316-echxuw74	30	6	looks	look	VBZ
cord-027316-echxuw74	30	7	into	into	IN
cord-027316-echxuw74	30	8	the	the	DT
cord-027316-echxuw74	30	9	data	datum	NNS
cord-027316-echxuw74	30	10	from	from	IN
cord-027316-echxuw74	30	11	all	all	PDT
cord-027316-echxuw74	30	12	the	the	DT
cord-027316-echxuw74	30	13	encoder	encoder	NN
cord-027316-echxuw74	30	14	sequences	sequence	NNS
cord-027316-echxuw74	30	15	with	with	IN
cord-027316-echxuw74	30	16	specific	specific	JJ
cord-027316-echxuw74	30	17	attention	attention	NN
cord-027316-echxuw74	30	18	measured	measure	VBN
cord-027316-echxuw74	30	19	by	by	IN
cord-027316-echxuw74	30	20	the	the	DT
cord-027316-echxuw74	30	21	weights	weight	NNS
cord-027316-echxuw74	30	22	.	.	.
cord-027316-echxuw74	31	1	Each	each	DT
cord-027316-echxuw74	31	2	weight	weight	NN
cord-027316-echxuw74	31	3	,	,	,
cord-027316-echxuw74	31	4	w	w	NNP
cord-027316-echxuw74	31	5	ti	ti	NNP
cord-027316-echxuw74	31	6	,	,	,
cord-027316-echxuw74	31	7	indicates	indicate	VBZ
cord-027316-echxuw74	31	8	the	the	DT
cord-027316-echxuw74	31	9	attention	attention	NN
cord-027316-echxuw74	31	10	decoder	decoder	NN
cord-027316-echxuw74	31	11	network	network	NN
cord-027316-echxuw74	31	12	t	t	NN
cord-027316-echxuw74	31	13	pays	pay	VBZ
cord-027316-echxuw74	31	14	for	for	IN
cord-027316-echxuw74	31	15	the	the	DT
cord-027316-echxuw74	31	16	encoder	encoder	NN
cord-027316-echxuw74	31	17	network	network	NN
cord-027316-echxuw74	31	18	i.	i.	NN
cord-027316-echxuw74	32	1	These	these	DT
cord-027316-echxuw74	32	2	weights	weight	NNS
cord-027316-echxuw74	32	3	are	be	VBP
cord-027316-echxuw74	32	4	dependent	dependent	JJ
cord-027316-echxuw74	32	5	on	on	IN
cord-027316-echxuw74	32	6	the	the	DT
cord-027316-echxuw74	32	7	previous	previous	JJ
cord-027316-echxuw74	32	8	decoder	decoder	NN
cord-027316-echxuw74	32	9	and	and	CC
cord-027316-echxuw74	32	10	output	output	NN
cord-027316-echxuw74	32	11	states	state	NNS
cord-027316-echxuw74	32	12	and	and	CC
cord-027316-echxuw74	32	13	present	present	JJ
cord-027316-echxuw74	32	14	encoder	encoder	NN
cord-027316-echxuw74	32	15	state	state	NN
cord-027316-echxuw74	32	16	as	as	IN
cord-027316-echxuw74	32	17	shown	show	VBN
cord-027316-echxuw74	32	18	in	in	IN
cord-027316-echxuw74	32	19	Fig	Fig	NNP
cord-027316-echxuw74	32	20	.	.	.
cord-027316-echxuw74	33	1	2	2	LS
cord-027316-echxuw74	33	2	.	.	.
cord-027316-echxuw74	34	1	Given	give	VBN
cord-027316-echxuw74	34	2	the	the	DT
cord-027316-echxuw74	34	3	complexity	complexity	NN
cord-027316-echxuw74	34	4	of	of	IN
cord-027316-echxuw74	34	5	these	these	DT
cord-027316-echxuw74	34	6	dependencies	dependency	NNS
cord-027316-echxuw74	34	7	,	,	,
cord-027316-echxuw74	34	8	a	a	DT
cord-027316-echxuw74	34	9	neural	neural	JJ
cord-027316-echxuw74	34	10	network	network	NN
cord-027316-echxuw74	34	11	model	model	NN
cord-027316-echxuw74	34	12	is	be	VBZ
cord-027316-echxuw74	34	13	used	use	VBN
cord-027316-echxuw74	34	14	to	to	TO
cord-027316-echxuw74	34	15	compute	compute	VB
cord-027316-echxuw74	34	16	these	these	DT
cord-027316-echxuw74	34	17	weights	weight	NNS
cord-027316-echxuw74	34	18	.	.	.
cord-027316-echxuw74	35	1	Two	two	CD
cord-027316-echxuw74	35	2	layers	layer	NNS
cord-027316-echxuw74	35	3	(	(	-LRB-
cord-027316-echxuw74	35	4	1024	1024	CD
cord-027316-echxuw74	35	5	)	)	-RRB-
cord-027316-echxuw74	35	6	of	of	IN
cord-027316-echxuw74	35	7	fully	fully	RB
cord-027316-echxuw74	35	8	connected	connect	VBN
cord-027316-echxuw74	35	9	layers	layer	NNS
cord-027316-echxuw74	35	10	and	and	CC
cord-027316-echxuw74	35	11	ReLU	relu	NN
cord-027316-echxuw74	35	12	activation	activation	NN
cord-027316-echxuw74	35	13	function	function	NN
cord-027316-echxuw74	35	14	is	be	VBZ
cord-027316-echxuw74	35	15	used	use	VBN
cord-027316-echxuw74	35	16	.	.	.
cord-027316-echxuw74	36	1	Where	where	WRB
cord-027316-echxuw74	36	2	H	h	NN
cord-027316-echxuw74	36	3	is	be	VBZ
cord-027316-echxuw74	36	4	the	the	DT
cord-027316-echxuw74	36	5	state	state	NN
cord-027316-echxuw74	36	6	of	of	IN
cord-027316-echxuw74	36	7	the	the	DT
cord-027316-echxuw74	36	8	encoder	encoder	NN
cord-027316-echxuw74	36	9	networks	network	NNS
cord-027316-echxuw74	36	10	,	,	,
cord-027316-echxuw74	36	11	s	s	VBZ
cord-027316-echxuw74	36	12	t−1	t−1	WP
cord-027316-echxuw74	36	13	is	be	VBZ
cord-027316-echxuw74	36	14	the	the	DT
cord-027316-echxuw74	36	15	previous	previous	JJ
cord-027316-echxuw74	36	16	state	state	NN
cord-027316-echxuw74	36	17	of	of	IN
cord-027316-echxuw74	36	18	the	the	DT
cord-027316-echxuw74	36	19	decoder	decoder	NN
cord-027316-echxuw74	36	20	and	and	CC
cord-027316-echxuw74	36	21	v	v	LS
cord-027316-echxuw74	36	22	t−1	t−1	NN
cord-027316-echxuw74	36	23	is	be	VBZ
cord-027316-echxuw74	36	24	the	the	DT
cord-027316-echxuw74	36	25	previous	previous	JJ
cord-027316-echxuw74	36	26	decoder	decoder	NN
cord-027316-echxuw74	36	27	output	output	NN
cord-027316-echxuw74	36	28	.	.	.
cord-027316-echxuw74	37	1	Also	also	RB
cord-027316-echxuw74	37	2	,	,	,
cord-027316-echxuw74	37	3	W	w	NN
cord-027316-echxuw74	37	4	t	t	NN
cord-027316-echxuw74	37	5	is	be	VBZ
cord-027316-echxuw74	37	6	the	the	DT
cord-027316-echxuw74	37	7	weights	weight	NNS
cord-027316-echxuw74	37	8	of	of	IN
cord-027316-echxuw74	37	9	the	the	DT
cord-027316-echxuw74	37	10	encoder	encoder	NN
cord-027316-echxuw74	37	11	state	state	NN
cord-027316-echxuw74	37	12	t.	t.	NNP
cord-027316-echxuw74	38	1	Since	since	IN
cord-027316-echxuw74	38	2	W	W	NNP
cord-027316-echxuw74	38	3	t	t	NN
cord-027316-echxuw74	38	4	are	be	VBP
cord-027316-echxuw74	38	5	the	the	DT
cord-027316-echxuw74	38	6	output	output	NN
cord-027316-echxuw74	38	7	from	from	IN
cord-027316-echxuw74	38	8	softmax	softmax	NNP
cord-027316-echxuw74	38	9	function	function	NN
cord-027316-echxuw74	38	10	,	,	,
cord-027316-echxuw74	38	11	then	then	RB
cord-027316-echxuw74	38	12	,	,	,
cord-027316-echxuw74	38	13	This	this	DT
cord-027316-echxuw74	38	14	section	section	NN
cord-027316-echxuw74	38	15	overviews	overview	NNS
cord-027316-echxuw74	38	16	of	of	IN
cord-027316-echxuw74	38	17	the	the	DT
cord-027316-echxuw74	38	18	contribution	contribution	NN
cord-027316-echxuw74	38	19	of	of	IN
cord-027316-echxuw74	38	20	this	this	DT
cord-027316-echxuw74	38	21	paper	paper	NN
cord-027316-echxuw74	38	22	and	and	CC
cord-027316-echxuw74	38	23	explains	explain	VBZ
cord-027316-echxuw74	38	24	the	the	DT
cord-027316-echxuw74	38	25	extension	extension	NN
cord-027316-echxuw74	38	26	made	make	VBN
cord-027316-echxuw74	38	27	over	over	IN
cord-027316-echxuw74	38	28	the	the	DT
cord-027316-echxuw74	38	29	state	state	NN
cord-027316-echxuw74	38	30	-	-	HYPH
cord-027316-echxuw74	38	31	of	of	IN
cord-027316-echxuw74	38	32	-	-	HYPH
cord-027316-echxuw74	38	33	the	the	DT
cord-027316-echxuw74	38	34	-	-	HYPH
cord-027316-echxuw74	38	35	art	art	NN
cord-027316-echxuw74	38	36	model	model	NN
cord-027316-echxuw74	38	37	.	.	.
cord-027316-echxuw74	39	1	A	a	DT
cord-027316-echxuw74	39	2	major	major	JJ
cord-027316-echxuw74	39	3	point	point	NN
cord-027316-echxuw74	39	4	of	of	IN
cord-027316-echxuw74	39	5	attention	attention	NN
cord-027316-echxuw74	39	6	for	for	IN
cord-027316-echxuw74	39	7	many	many	JJ
cord-027316-echxuw74	39	8	texts	text	NNS
cord-027316-echxuw74	39	9	related	relate	VBN
cord-027316-echxuw74	39	10	analysis	analysis	NN
cord-027316-echxuw74	39	11	is	be	VBZ
cord-027316-echxuw74	39	12	to	to	TO
cord-027316-echxuw74	39	13	determine	determine	VB
cord-027316-echxuw74	39	14	which	which	WDT
cord-027316-echxuw74	39	15	part(s	part(s	NNP
cord-027316-echxuw74	39	16	)	)	-RRB-
cord-027316-echxuw74	39	17	of	of	IN
cord-027316-echxuw74	39	18	the	the	DT
cord-027316-echxuw74	39	19	input	input	NN
cord-027316-echxuw74	39	20	text	text	NN
cord-027316-echxuw74	39	21	has	have	VBZ
cord-027316-echxuw74	39	22	had	have	VBN
cord-027316-echxuw74	39	23	more	more	JJR
cord-027316-echxuw74	39	24	impact	impact	NN
cord-027316-echxuw74	39	25	in	in	IN
cord-027316-echxuw74	39	26	determining	determine	VBG
cord-027316-echxuw74	39	27	the	the	DT
cord-027316-echxuw74	39	28	output	output	NN
cord-027316-echxuw74	39	29	.	.	.
cord-027316-echxuw74	40	1	he	-PRON-	PRP
cord-027316-echxuw74	40	2	length	length	NN
cord-027316-echxuw74	40	3	of	of	IN
cord-027316-echxuw74	40	4	input	input	NN
cord-027316-echxuw74	40	5	text	text	NN
cord-027316-echxuw74	40	6	could	could	MD
cord-027316-echxuw74	40	7	be	be	VB
cord-027316-echxuw74	40	8	very	very	RB
cord-027316-echxuw74	40	9	long	long	JJ
cord-027316-echxuw74	40	10	combining	combining	NN
cord-027316-echxuw74	40	11	of	of	IN
cord-027316-echxuw74	40	12	potentially	potentially	RB
cord-027316-echxuw74	40	13	hundreds	hundred	NNS
cord-027316-echxuw74	40	14	and	and	CC
cord-027316-echxuw74	40	15	thousands	thousand	NNS
cord-027316-echxuw74	40	16	of	of	IN
cord-027316-echxuw74	40	17	words	word	NNS
cord-027316-echxuw74	40	18	or	or	CC
cord-027316-echxuw74	40	19	sequences	sequence	NNS
cord-027316-echxuw74	40	20	,	,	,
cord-027316-echxuw74	40	21	i.e.	i.e.	FW
cord-027316-echxuw74	40	22	,	,	,
cord-027316-echxuw74	40	23	n	n	LS
cord-027316-echxuw74	40	24	could	could	MD
cord-027316-echxuw74	40	25	be	be	VB
cord-027316-echxuw74	40	26	very	very	RB
cord-027316-echxuw74	40	27	large	large	JJ
cord-027316-echxuw74	40	28	number	number	NN
cord-027316-echxuw74	40	29	.	.	.
cord-027316-echxuw74	41	1	Thus	thus	RB
cord-027316-echxuw74	41	2	,	,	,
cord-027316-echxuw74	41	3	there	there	EX
cord-027316-echxuw74	41	4	are	be	VBP
cord-027316-echxuw74	41	5	many	many	JJ
cord-027316-echxuw74	41	6	weights	weight	NNS
cord-027316-echxuw74	41	7	(	(	-LRB-
cord-027316-echxuw74	41	8	w	w	NNP
cord-027316-echxuw74	41	9	ti	ti	NNP
cord-027316-echxuw74	41	10	)	)	-RRB-
cord-027316-echxuw74	41	11	in	in	IN
cord-027316-echxuw74	41	12	determining	determine	VBG
cord-027316-echxuw74	41	13	any	any	DT
cord-027316-echxuw74	41	14	part	part	NN
cord-027316-echxuw74	41	15	of	of	IN
cord-027316-echxuw74	41	16	output	output	NN
cord-027316-echxuw74	41	17	v	v	NNP
cord-027316-echxuw74	41	18	t	t	NN
cord-027316-echxuw74	41	19	,	,	,
cord-027316-echxuw74	41	20	and	and	CC
cord-027316-echxuw74	41	21	also	also	RB
cord-027316-echxuw74	41	22	since	since	IN
cord-027316-echxuw74	41	23	many	many	JJ
cord-027316-echxuw74	41	24	of	of	IN
cord-027316-echxuw74	41	25	these	these	DT
cord-027316-echxuw74	41	26	weights	weight	NNS
cord-027316-echxuw74	41	27	are	be	VBP
cord-027316-echxuw74	41	28	correlated	correlate	VBN
cord-027316-echxuw74	41	29	,	,	,
cord-027316-echxuw74	41	30	it	-PRON-	PRP
cord-027316-echxuw74	41	31	's	be	VBZ
cord-027316-echxuw74	41	32	difficult	difficult	JJ
cord-027316-echxuw74	41	33	to	to	TO
cord-027316-echxuw74	41	34	determine	determine	VB
cord-027316-echxuw74	41	35	the	the	DT
cord-027316-echxuw74	41	36	significance	significance	NN
cord-027316-echxuw74	41	37	of	of	IN
cord-027316-echxuw74	41	38	any	any	DT
cord-027316-echxuw74	41	39	input	input	NN
cord-027316-echxuw74	41	40	sequence	sequence	NN
cord-027316-echxuw74	41	41	in	in	IN
cord-027316-echxuw74	41	42	computing	compute	VBG
cord-027316-echxuw74	41	43	any	any	DT
cord-027316-echxuw74	41	44	output	output	NN
cord-027316-echxuw74	41	45	sequence	sequence	NN
cord-027316-echxuw74	41	46	v	v	NN
cord-027316-echxuw74	41	47	t	t	NN
cord-027316-echxuw74	41	48	.	.	.
cord-027316-echxuw74	42	1	To	to	TO
cord-027316-echxuw74	42	2	make	make	VB
cord-027316-echxuw74	42	3	these	these	DT
cord-027316-echxuw74	42	4	dependencies	dependency	NNS
cord-027316-echxuw74	42	5	clearer	clear	JJR
cord-027316-echxuw74	42	6	and	and	CC
cord-027316-echxuw74	42	7	to	to	TO
cord-027316-echxuw74	42	8	recognize	recognize	VB
cord-027316-echxuw74	42	9	the	the	DT
cord-027316-echxuw74	42	10	most	most	RBS
cord-027316-echxuw74	42	11	significant	significant	JJ
cord-027316-echxuw74	42	12	input	input	NN
cord-027316-echxuw74	42	13	sequences	sequence	NNS
cord-027316-echxuw74	42	14	for	for	IN
cord-027316-echxuw74	42	15	any	any	DT
cord-027316-echxuw74	42	16	output	output	NN
cord-027316-echxuw74	42	17	sequence	sequence	NN
cord-027316-echxuw74	42	18	,	,	,
cord-027316-echxuw74	42	19	we	-PRON-	PRP
cord-027316-echxuw74	42	20	apply	apply	VBP
cord-027316-echxuw74	42	21	a	a	DT
cord-027316-echxuw74	42	22	zero	zero	CD
cord-027316-echxuw74	42	23	-	-	HYPH
cord-027316-echxuw74	42	24	norm	norm	NN
cord-027316-echxuw74	42	25	penalty	penalty	NN
cord-027316-echxuw74	42	26	to	to	TO
cord-027316-echxuw74	42	27	make	make	VB
cord-027316-echxuw74	42	28	the	the	DT
cord-027316-echxuw74	42	29	corresponding	corresponding	JJ
cord-027316-echxuw74	42	30	weight	weight	NN
cord-027316-echxuw74	42	31	vector	vector	NN
cord-027316-echxuw74	42	32	to	to	TO
cord-027316-echxuw74	42	33	become	become	VB
cord-027316-echxuw74	42	34	a	a	DT
cord-027316-echxuw74	42	35	sparse	sparse	JJ
cord-027316-echxuw74	42	36	vector	vector	NN
cord-027316-echxuw74	42	37	.	.	.
cord-027316-echxuw74	43	1	To	to	TO
cord-027316-echxuw74	43	2	achieve	achieve	VB
cord-027316-echxuw74	43	3	the	the	DT
cord-027316-echxuw74	43	4	desired	desire	VBN
cord-027316-echxuw74	43	5	sparsity	sparsity	NN
cord-027316-echxuw74	43	6	,	,	,
cord-027316-echxuw74	43	7	zero	zero	CD
cord-027316-echxuw74	43	8	-	-	HYPH
cord-027316-echxuw74	43	9	norm	norm	NN
cord-027316-echxuw74	43	10	(	(	-LRB-
cord-027316-echxuw74	43	11	L	l	NN
cord-027316-echxuw74	43	12	0	0	CD
cord-027316-echxuw74	43	13	)	)	-RRB-
cord-027316-echxuw74	43	14	is	be	VBZ
cord-027316-echxuw74	43	15	applied	apply	VBN
cord-027316-echxuw74	43	16	to	to	TO
cord-027316-echxuw74	43	17	make	make	VB
cord-027316-echxuw74	43	18	any	any	DT
cord-027316-echxuw74	43	19	corresponding	correspond	VBG
cord-027316-echxuw74	43	20	W	w	NN
cord-027316-echxuw74	43	21	t	t	NN
cord-027316-echxuw74	43	22	vector	vector	NN
cord-027316-echxuw74	43	23	very	very	RB
cord-027316-echxuw74	43	24	sparse	sparse	JJ
cord-027316-echxuw74	43	25	as	as	IN
cord-027316-echxuw74	43	26	the	the	DT
cord-027316-echxuw74	43	27	penalty	penalty	NN
cord-027316-echxuw74	43	28	leads	lead	VBZ
cord-027316-echxuw74	43	29	to	to	IN
cord-027316-echxuw74	43	30	minimization	minimization	NN
cord-027316-echxuw74	43	31	of	of	IN
cord-027316-echxuw74	43	32	the	the	DT
cord-027316-echxuw74	43	33	number	number	NN
cord-027316-echxuw74	43	34	of	of	IN
cord-027316-echxuw74	43	35	non	non	JJ
cord-027316-echxuw74	43	36	-	-	JJ
cord-027316-echxuw74	43	37	zero	zero	JJ
cord-027316-echxuw74	43	38	entries	entry	NNS
cord-027316-echxuw74	43	39	in	in	IN
cord-027316-echxuw74	43	40	W	w	NN
cord-027316-echxuw74	43	41	t	t	NN
cord-027316-echxuw74	43	42	.	.	.
cord-027316-echxuw74	44	1	The	the	DT
cord-027316-echxuw74	44	2	process	process	NN
cord-027316-echxuw74	44	3	is	be	VBZ
cord-027316-echxuw74	44	4	implemented	implement	VBN
cord-027316-echxuw74	44	5	by	by	IN
cord-027316-echxuw74	44	6	imposing	impose	VBG
cord-027316-echxuw74	44	7	the	the	DT
cord-027316-echxuw74	44	8	constraint	constraint	NN
cord-027316-echxuw74	44	9	of	of	IN
cord-027316-echxuw74	44	10	,	,	,
cord-027316-echxuw74	44	11	Since	since	IN
cord-027316-echxuw74	44	12	L	l	NN
cord-027316-echxuw74	44	13	0	0	CD
cord-027316-echxuw74	44	14	is	be	VBZ
cord-027316-echxuw74	44	15	computationally	computationally	RB
cord-027316-echxuw74	44	16	intractable	intractable	JJ
cord-027316-echxuw74	44	17	,	,	,
cord-027316-echxuw74	44	18	we	-PRON-	PRP
cord-027316-echxuw74	44	19	could	could	MD
cord-027316-echxuw74	44	20	use	use	VB
cord-027316-echxuw74	44	21	surrogate	surrogate	JJ
cord-027316-echxuw74	44	22	norms	norm	NNS
cord-027316-echxuw74	44	23	such	such	JJ
cord-027316-echxuw74	44	24	as	as	IN
cord-027316-echxuw74	44	25	L	l	NN
cord-027316-echxuw74	44	26	1	1	CD
cord-027316-echxuw74	44	27	norm	norm	NN
cord-027316-echxuw74	44	28	or	or	CC
cord-027316-echxuw74	44	29	Euclidean	euclidean	JJ
cord-027316-echxuw74	44	30	norm	norm	NN
cord-027316-echxuw74	44	31	,	,	,
cord-027316-echxuw74	44	32	L	l	NN
cord-027316-echxuw74	44	33	2	2	CD
cord-027316-echxuw74	44	34	.	.	.
cord-027316-echxuw74	45	1	To	to	TO
cord-027316-echxuw74	45	2	impose	impose	VB
cord-027316-echxuw74	45	3	sparsity	sparsity	NN
cord-027316-echxuw74	45	4	,	,	,
cord-027316-echxuw74	45	5	the	the	DT
cord-027316-echxuw74	45	6	L	l	NN
cord-027316-echxuw74	45	7	1	1	CD
cord-027316-echxuw74	45	8	norm	norm	NN
cord-027316-echxuw74	45	9	,	,	,
cord-027316-echxuw74	45	10	LASSO	LASSO	NNP
cord-027316-echxuw74	45	11	[	[	-LRB-
cord-027316-echxuw74	45	12	8	8	CD
cord-027316-echxuw74	45	13	,	,	,
cord-027316-echxuw74	45	14	14	14	CD
cord-027316-echxuw74	45	15	,	,	,
cord-027316-echxuw74	45	16	15	15	CD
cord-027316-echxuw74	45	17	,	,	,
cord-027316-echxuw74	45	18	18	18	CD
cord-027316-echxuw74	45	19	,	,	,
cord-027316-echxuw74	45	20	21	21	CD
cord-027316-echxuw74	45	21	]	]	-RRB-
cord-027316-echxuw74	45	22	is	be	VBZ
cord-027316-echxuw74	45	23	used	use	VBN
cord-027316-echxuw74	45	24	in	in	IN
cord-027316-echxuw74	45	25	this	this	DT
cord-027316-echxuw74	45	26	work	work	NN
cord-027316-echxuw74	45	27	,	,	,
cord-027316-echxuw74	45	28	Or	or	CC
cord-027316-echxuw74	45	29	,	,	,
cord-027316-echxuw74	45	30	As	as	IN
cord-027316-echxuw74	45	31	the	the	DT
cord-027316-echxuw74	45	32	penalty	penalty	NN
cord-027316-echxuw74	45	33	function	function	NN
cord-027316-echxuw74	45	34	to	to	TO
cord-027316-echxuw74	45	35	enforce	enforce	VB
cord-027316-echxuw74	45	36	sparsity	sparsity	NN
cord-027316-echxuw74	45	37	on	on	IN
cord-027316-echxuw74	45	38	the	the	DT
cord-027316-echxuw74	45	39	weight	weight	NN
cord-027316-echxuw74	45	40	vectors	vector	NNS
cord-027316-echxuw74	45	41	.	.	.
cord-027316-echxuw74	46	1	This	this	DT
cord-027316-echxuw74	46	2	penalty	penalty	NN
cord-027316-echxuw74	46	3	,	,	,
cord-027316-echxuw74	46	4	β	β	NN
cord-027316-echxuw74	46	5	W	w	NN
cord-027316-echxuw74	46	6	t	t	NN
cord-027316-echxuw74	46	7	1	1	CD
cord-027316-echxuw74	46	8	,	,	,
cord-027316-echxuw74	46	9	is	be	VBZ
cord-027316-echxuw74	46	10	the	the	DT
cord-027316-echxuw74	46	11	first	first	JJ
cord-027316-echxuw74	46	12	extension	extension	NN
cord-027316-echxuw74	46	13	to	to	IN
cord-027316-echxuw74	46	14	the	the	DT
cord-027316-echxuw74	46	15	attention	attention	NN
cord-027316-echxuw74	46	16	model	model	NN
cord-027316-echxuw74	46	17	[	[	-LRB-
cord-027316-echxuw74	46	18	2	2	CD
cord-027316-echxuw74	46	19	,	,	,
cord-027316-echxuw74	46	20	5	5	CD
cord-027316-echxuw74	46	21	]	]	-RRB-
cord-027316-echxuw74	46	22	.	.	.
cord-027316-echxuw74	47	1	Here	here	RB
cord-027316-echxuw74	47	2	,	,	,
cord-027316-echxuw74	47	3	β	β	NN
cord-027316-echxuw74	47	4	is	be	VBZ
cord-027316-echxuw74	47	5	the	the	DT
cord-027316-echxuw74	47	6	regularization	regularization	NN
cord-027316-echxuw74	47	7	parameter	parameter	NN
cord-027316-echxuw74	47	8	which	which	WDT
cord-027316-echxuw74	47	9	is	be	VBZ
cord-027316-echxuw74	47	10	set	set	VBN
cord-027316-echxuw74	47	11	as	as	IN
cord-027316-echxuw74	47	12	a	a	DT
cord-027316-echxuw74	47	13	hyperparameter	hyperparameter	NN
cord-027316-echxuw74	47	14	where	where	WRB
cord-027316-echxuw74	47	15	its	-PRON-	PRP$
cord-027316-echxuw74	47	16	value	value	NN
cord-027316-echxuw74	47	17	is	be	VBZ
cord-027316-echxuw74	47	18	set	set	VBN
cord-027316-echxuw74	47	19	before	before	IN
cord-027316-echxuw74	47	20	learning	learn	VBG
cord-027316-echxuw74	47	21	.	.	.
cord-027316-echxuw74	48	1	Higher	high	JJR
cord-027316-echxuw74	48	2	constraint	constraint	NN
cord-027316-echxuw74	48	3	leads	lead	VBZ
cord-027316-echxuw74	48	4	to	to	IN
cord-027316-echxuw74	48	5	higher	high	JJR
cord-027316-echxuw74	48	6	sparsity	sparsity	NN
cord-027316-echxuw74	48	7	with	with	IN
cord-027316-echxuw74	48	8	higher	high	JJR
cord-027316-echxuw74	48	9	added	add	VBN
cord-027316-echxuw74	48	10	regularization	regularization	NN
cord-027316-echxuw74	48	11	biased	bias	VBN
cord-027316-echxuw74	48	12	error	error	NN
cord-027316-echxuw74	48	13	and	and	CC
cord-027316-echxuw74	48	14	lower	low	JJR
cord-027316-echxuw74	48	15	values	value	NNS
cord-027316-echxuw74	48	16	of	of	IN
cord-027316-echxuw74	48	17	the	the	DT
cord-027316-echxuw74	48	18	regularization	regularization	NN
cord-027316-echxuw74	48	19	parameter	parameter	NN
cord-027316-echxuw74	48	20	leads	lead	VBZ
cord-027316-echxuw74	48	21	to	to	TO
cord-027316-echxuw74	48	22	lower	lower	VB
cord-027316-echxuw74	48	23	sparsity	sparsity	NN
cord-027316-echxuw74	48	24	and	and	CC
cord-027316-echxuw74	48	25	lesser	less	JJR
cord-027316-echxuw74	48	26	regularization	regularization	NN
cord-027316-echxuw74	48	27	bias	bias	NN
cord-027316-echxuw74	48	28	.	.	.
cord-027316-echxuw74	49	1	The	the	DT
cord-027316-echxuw74	49	2	main	main	JJ
cord-027316-echxuw74	49	3	goal	goal	NN
cord-027316-echxuw74	49	4	of	of	IN
cord-027316-echxuw74	49	5	this	this	DT
cord-027316-echxuw74	49	6	work	work	NN
cord-027316-echxuw74	49	7	is	be	VBZ
cord-027316-echxuw74	49	8	to	to	TO
cord-027316-echxuw74	49	9	find	find	VB
cord-027316-echxuw74	49	10	out	out	RP
cord-027316-echxuw74	49	11	which	which	WDT
cord-027316-echxuw74	49	12	parts	part	NNS
cord-027316-echxuw74	49	13	of	of	IN
cord-027316-echxuw74	49	14	encoder	encoder	NN
cord-027316-echxuw74	49	15	sequences	sequence	NNS
cord-027316-echxuw74	49	16	are	be	VBP
cord-027316-echxuw74	49	17	most	most	RBS
cord-027316-echxuw74	49	18	critical	critical	JJ
cord-027316-echxuw74	49	19	in	in	IN
cord-027316-echxuw74	49	20	determining	determine	VBG
cord-027316-echxuw74	49	21	and	and	CC
cord-027316-echxuw74	49	22	computing	compute	VBG
cord-027316-echxuw74	49	23	any	any	DT
cord-027316-echxuw74	49	24	output	output	NN
cord-027316-echxuw74	49	25	.	.	.
cord-027316-echxuw74	50	1	The	the	DT
cord-027316-echxuw74	50	2	output	output	NN
cord-027316-echxuw74	50	3	could	could	MD
cord-027316-echxuw74	50	4	be	be	VB
cord-027316-echxuw74	50	5	a	a	DT
cord-027316-echxuw74	50	6	word	word	NN
cord-027316-echxuw74	50	7	,	,	,
cord-027316-echxuw74	50	8	a	a	DT
cord-027316-echxuw74	50	9	sentence	sentence	NN
cord-027316-echxuw74	50	10	or	or	CC
cord-027316-echxuw74	50	11	any	any	DT
cord-027316-echxuw74	50	12	other	other	JJ
cord-027316-echxuw74	50	13	subsequence	subsequence	NN
cord-027316-echxuw74	50	14	.	.	.
cord-027316-echxuw74	51	1	The	the	DT
cord-027316-echxuw74	51	2	goal	goal	NN
cord-027316-echxuw74	51	3	is	be	VBZ
cord-027316-echxuw74	51	4	critical	critical	JJ
cord-027316-echxuw74	51	5	especially	especially	RB
cord-027316-echxuw74	51	6	in	in	IN
cord-027316-echxuw74	51	7	application	application	NN
cord-027316-echxuw74	51	8	such	such	JJ
cord-027316-echxuw74	51	9	as	as	IN
cord-027316-echxuw74	51	10	machine	machine	NN
cord-027316-echxuw74	51	11	translation	translation	NN
cord-027316-echxuw74	51	12	,	,	,
cord-027316-echxuw74	51	13	image	image	NN
cord-027316-echxuw74	51	14	captioning	captioning	NN
cord-027316-echxuw74	51	15	,	,	,
cord-027316-echxuw74	51	16	sentiment	sentiment	NN
cord-027316-echxuw74	51	17	analysis	analysis	NN
cord-027316-echxuw74	51	18	,	,	,
cord-027316-echxuw74	51	19	topic	topic	NN
cord-027316-echxuw74	51	20	modeling	modeling	NN
cord-027316-echxuw74	51	21	and	and	CC
cord-027316-echxuw74	51	22	predictive	predictive	JJ
cord-027316-echxuw74	51	23	modeling	modeling	NN
cord-027316-echxuw74	51	24	such	such	JJ
cord-027316-echxuw74	51	25	as	as	IN
cord-027316-echxuw74	51	26	time	time	NN
cord-027316-echxuw74	51	27	series	series	NNP
cord-027316-echxuw74	51	28	analysis	analysis	NN
cord-027316-echxuw74	51	29	and	and	CC
cord-027316-echxuw74	51	30	prediction	prediction	NN
cord-027316-echxuw74	51	31	.	.	.
cord-027316-echxuw74	52	1	To	to	TO
cord-027316-echxuw74	52	2	add	add	VB
cord-027316-echxuw74	52	3	another	another	DT
cord-027316-echxuw74	52	4	layer	layer	NN
cord-027316-echxuw74	52	5	of	of	IN
cord-027316-echxuw74	52	6	regularization	regularization	NN
cord-027316-echxuw74	52	7	,	,	,
cord-027316-echxuw74	52	8	this	this	DT
cord-027316-echxuw74	52	9	work	work	NN
cord-027316-echxuw74	52	10	imposes	impose	VBZ
cord-027316-echxuw74	52	11	embedding	embed	VBG
cord-027316-echxuw74	52	12	error	error	NN
cord-027316-echxuw74	52	13	penalty	penalty	NN
cord-027316-echxuw74	52	14	to	to	IN
cord-027316-echxuw74	52	15	the	the	DT
cord-027316-echxuw74	52	16	objective	objective	JJ
cord-027316-echxuw74	52	17	function	function	NN
cord-027316-echxuw74	52	18	(	(	-LRB-
cord-027316-echxuw74	52	19	usually	usually	RB
cord-027316-echxuw74	52	20	,	,	,
cord-027316-echxuw74	52	21	cross	cross	NN
cord-027316-echxuw74	52	22	entropy	entropy	NN
cord-027316-echxuw74	52	23	)	)	-RRB-
cord-027316-echxuw74	52	24	.	.	.
cord-027316-echxuw74	53	1	This	this	DT
cord-027316-echxuw74	53	2	added	add	VBN
cord-027316-echxuw74	53	3	penalty	penalty	NN
cord-027316-echxuw74	53	4	also	also	RB
cord-027316-echxuw74	53	5	helps	help	VBZ
cord-027316-echxuw74	53	6	to	to	TO
cord-027316-echxuw74	53	7	address	address	VB
cord-027316-echxuw74	53	8	the	the	DT
cord-027316-echxuw74	53	9	"	"	``
cord-027316-echxuw74	53	10	coverage	coverage	NN
cord-027316-echxuw74	53	11	problem	problem	NN
cord-027316-echxuw74	53	12	"	"	''
cord-027316-echxuw74	53	13	(	(	-LRB-
cord-027316-echxuw74	53	14	the	the	DT
cord-027316-echxuw74	53	15	phenomenon	phenomenon	NN
cord-027316-echxuw74	53	16	of	of	IN
cord-027316-echxuw74	53	17	often	often	RB
cord-027316-echxuw74	53	18	observed	observe	VBN
cord-027316-echxuw74	53	19	dropping	drop	VBG
cord-027316-echxuw74	53	20	or	or	CC
cord-027316-echxuw74	53	21	frequently	frequently	RB
cord-027316-echxuw74	53	22	repeating	repeat	VBG
cord-027316-echxuw74	53	23	words	word	NNS
cord-027316-echxuw74	53	24	--or	--or	.
cord-027316-echxuw74	53	25	any	any	DT
cord-027316-echxuw74	53	26	other	other	JJ
cord-027316-echxuw74	53	27	subsequence	subsequence	NN
cord-027316-echxuw74	53	28	--by	--by	VBZ
cord-027316-echxuw74	53	29	the	the	DT
cord-027316-echxuw74	53	30	network	network	NN
cord-027316-echxuw74	53	31	)	)	-RRB-
cord-027316-echxuw74	53	32	.	.	.
cord-027316-echxuw74	54	1	The	the	DT
cord-027316-echxuw74	54	2	embedding	embed	VBG
cord-027316-echxuw74	54	3	regularization	regularization	NN
cord-027316-echxuw74	54	4	is	be	VBZ
cord-027316-echxuw74	54	5	,	,	,
cord-027316-echxuw74	54	6	α	α	NNP
cord-027316-echxuw74	54	7	Embedding	embedding	NN
cord-027316-echxuw74	54	8	Error	error	NN
cord-027316-echxuw74	54	9	2	2	CD
cord-027316-echxuw74	54	10	(	(	-LRB-
cord-027316-echxuw74	54	11	6	6	CD
cord-027316-echxuw74	54	12	)	)	-RRB-
cord-027316-echxuw74	55	1	Input	input	NN
cord-027316-echxuw74	55	2	to	to	IN
cord-027316-echxuw74	55	3	any	any	DT
cord-027316-echxuw74	55	4	model	model	NN
cord-027316-echxuw74	55	5	has	have	VBZ
cord-027316-echxuw74	55	6	to	to	TO
cord-027316-echxuw74	55	7	be	be	VB
cord-027316-echxuw74	55	8	a	a	DT
cord-027316-echxuw74	55	9	number	number	NN
cord-027316-echxuw74	55	10	and	and	CC
cord-027316-echxuw74	55	11	hence	hence	RB
cord-027316-echxuw74	55	12	the	the	DT
cord-027316-echxuw74	55	13	raw	raw	JJ
cord-027316-echxuw74	55	14	input	input	NN
cord-027316-echxuw74	55	15	of	of	IN
cord-027316-echxuw74	55	16	words	word	NNS
cord-027316-echxuw74	55	17	or	or	CC
cord-027316-echxuw74	55	18	text	text	NN
cord-027316-echxuw74	55	19	sequence	sequence	NN
cord-027316-echxuw74	55	20	needs	need	VBZ
cord-027316-echxuw74	55	21	to	to	TO
cord-027316-echxuw74	55	22	be	be	VB
cord-027316-echxuw74	55	23	transformed	transform	VBN
cord-027316-echxuw74	55	24	to	to	IN
cord-027316-echxuw74	55	25	continuous	continuous	JJ
cord-027316-echxuw74	55	26	numbers	number	NNS
cord-027316-echxuw74	55	27	.	.	.
cord-027316-echxuw74	56	1	This	this	DT
cord-027316-echxuw74	56	2	is	be	VBZ
cord-027316-echxuw74	56	3	done	do	VBN
cord-027316-echxuw74	56	4	by	by	IN
cord-027316-echxuw74	56	5	using	use	VBG
cord-027316-echxuw74	56	6	one	one	CD
cord-027316-echxuw74	56	7	-	-	HYPH
cord-027316-echxuw74	56	8	hot	hot	JJ
cord-027316-echxuw74	56	9	encoding	encoding	NN
cord-027316-echxuw74	56	10	of	of	IN
cord-027316-echxuw74	56	11	the	the	DT
cord-027316-echxuw74	56	12	words	word	NNS
cord-027316-echxuw74	56	13	and	and	CC
cord-027316-echxuw74	56	14	then	then	RB
cord-027316-echxuw74	56	15	using	use	VBG
cord-027316-echxuw74	56	16	embedding	embedding	NN
cord-027316-echxuw74	56	17	as	as	IN
cord-027316-echxuw74	56	18	shown	show	VBN
cord-027316-echxuw74	56	19	in	in	IN
cord-027316-echxuw74	56	20	Fig	Fig	NNP
cord-027316-echxuw74	56	21	.	.	.
cord-027316-echxuw74	57	1	3	3	LS
cord-027316-echxuw74	57	2	.	.	.
cord-027316-echxuw74	58	1	Whereu	Whereu	NNP
cord-027316-echxuw74	59	1	i	i	PRP
cord-027316-echxuw74	59	2	is	be	VBZ
cord-027316-echxuw74	59	3	the	the	DT
cord-027316-echxuw74	59	4	raw	raw	JJ
cord-027316-echxuw74	59	5	input	input	NN
cord-027316-echxuw74	59	6	text	text	NN
cord-027316-echxuw74	59	7	,	,	,
cord-027316-echxuw74	59	8	is	be	VBZ
cord-027316-echxuw74	59	9	the	the	DT
cord-027316-echxuw74	59	10	one	one	CD
cord-027316-echxuw74	59	11	-	-	HYPH
cord-027316-echxuw74	59	12	hot	hot	JJ
cord-027316-echxuw74	59	13	encoding	encoding	NN
cord-027316-echxuw74	59	14	representation	representation	NN
cord-027316-echxuw74	59	15	of	of	IN
cord-027316-echxuw74	59	16	the	the	DT
cord-027316-echxuw74	59	17	raw	raw	JJ
cord-027316-echxuw74	59	18	input	input	NN
cord-027316-echxuw74	59	19	and	and	CC
cord-027316-echxuw74	59	20	u	u	NN
cord-027316-echxuw74	60	1	i	i	PRP
cord-027316-echxuw74	60	2	is	be	VBZ
cord-027316-echxuw74	60	3	the	the	DT
cord-027316-echxuw74	60	4	embedding	embedding	NN
cord-027316-echxuw74	60	5	of	of	IN
cord-027316-echxuw74	60	6	the	the	DT
cord-027316-echxuw74	60	7	i	i	NN
cord-027316-echxuw74	60	8	-	-	HYPH
cord-027316-echxuw74	60	9	th	th	RP
cord-027316-echxuw74	60	10	input	input	NN
cord-027316-echxuw74	60	11	or	or	CC
cord-027316-echxuw74	60	12	sequence	sequence	NN
cord-027316-echxuw74	60	13	.	.	.
cord-027316-echxuw74	61	1	Also	also	RB
cord-027316-echxuw74	61	2	,	,	,
cord-027316-echxuw74	61	3	α	α	NN
cord-027316-echxuw74	61	4	is	be	VBZ
cord-027316-echxuw74	61	5	the	the	DT
cord-027316-echxuw74	61	6	regularization	regularization	NN
cord-027316-echxuw74	61	7	parameter	parameter	NN
cord-027316-echxuw74	61	8	.	.	.
cord-027316-echxuw74	62	1	The	the	DT
cord-027316-echxuw74	62	2	idea	idea	NN
cord-027316-echxuw74	62	3	of	of	IN
cord-027316-echxuw74	62	4	embedding	embed	VBG
cord-027316-echxuw74	62	5	is	be	VBZ
cord-027316-echxuw74	62	6	based	base	VBN
cord-027316-echxuw74	62	7	on	on	IN
cord-027316-echxuw74	62	8	that	that	DT
cord-027316-echxuw74	62	9	embedding	embedding	NN
cord-027316-echxuw74	62	10	should	should	MD
cord-027316-echxuw74	62	11	preserve	preserve	VB
cord-027316-echxuw74	62	12	word	word	NN
cord-027316-echxuw74	62	13	similarities	similarity	NNS
cord-027316-echxuw74	62	14	,	,	,
cord-027316-echxuw74	62	15	i.e.	i.e.	FW
cord-027316-echxuw74	62	16	,	,	,
cord-027316-echxuw74	62	17	the	the	DT
cord-027316-echxuw74	62	18	words	word	NNS
cord-027316-echxuw74	62	19	that	that	WDT
cord-027316-echxuw74	62	20	are	be	VBP
cord-027316-echxuw74	62	21	synonyms	synonyms	NN
cord-027316-echxuw74	62	22	before	before	IN
cord-027316-echxuw74	62	23	embedding	embedding	NN
cord-027316-echxuw74	62	24	,	,	,
cord-027316-echxuw74	62	25	should	should	MD
cord-027316-echxuw74	62	26	remain	remain	VB
cord-027316-echxuw74	62	27	synonyms	synonyms	NN
cord-027316-echxuw74	62	28	after	after	IN
cord-027316-echxuw74	62	29	embedding	embed	VBG
cord-027316-echxuw74	62	30	.	.	.
cord-027316-echxuw74	63	1	Using	use	VBG
cord-027316-echxuw74	63	2	this	this	DT
cord-027316-echxuw74	63	3	concept	concept	NN
cord-027316-echxuw74	63	4	of	of	IN
cord-027316-echxuw74	63	5	embedding	embedding	NN
cord-027316-echxuw74	63	6	,	,	,
cord-027316-echxuw74	63	7	the	the	DT
cord-027316-echxuw74	63	8	scaled	scale	VBN
cord-027316-echxuw74	63	9	embedding	embedding	NN
cord-027316-echxuw74	63	10	error	error	NN
cord-027316-echxuw74	63	11	is	be	VBZ
cord-027316-echxuw74	63	12	,	,	,
cord-027316-echxuw74	63	13	Or	or	CC
cord-027316-echxuw74	63	14	,	,	,
cord-027316-echxuw74	63	15	after	after	IN
cord-027316-echxuw74	63	16	scaling	scale	VBG
cord-027316-echxuw74	63	17	the	the	DT
cord-027316-echxuw74	63	18	embedding	embedding	NN
cord-027316-echxuw74	63	19	error	error	NN
cord-027316-echxuw74	63	20	,	,	,
cord-027316-echxuw74	63	21	Which	which	WDT
cord-027316-echxuw74	63	22	could	could	MD
cord-027316-echxuw74	63	23	be	be	VB
cord-027316-echxuw74	63	24	re	re	VBN
cord-027316-echxuw74	63	25	-	-	VBN
cord-027316-echxuw74	63	26	written	write	VBN
cord-027316-echxuw74	63	27	,	,	,
cord-027316-echxuw74	63	28	using	use	VBG
cord-027316-echxuw74	63	29	a	a	DT
cord-027316-echxuw74	63	30	regularization	regularization	NN
cord-027316-echxuw74	63	31	parameter	parameter	NN
cord-027316-echxuw74	63	32	(	(	-LRB-
cord-027316-echxuw74	63	33	α	α	NN
cord-027316-echxuw74	63	34	)	)	-RRB-
cord-027316-echxuw74	63	35	,	,	,
cord-027316-echxuw74	63	36	as	as	IN
cord-027316-echxuw74	63	37	,	,	,
cord-027316-echxuw74	63	38	Where	where	WRB
cord-027316-echxuw74	63	39	L	L	NNP
cord-027316-echxuw74	63	40	is	be	VBZ
cord-027316-echxuw74	63	41	the	the	DT
cord-027316-echxuw74	63	42	measure	measure	NN
cord-027316-echxuw74	63	43	or	or	CC
cord-027316-echxuw74	63	44	metric	metric	NN
cord-027316-echxuw74	63	45	of	of	IN
cord-027316-echxuw74	63	46	similarity	similarity	NN
cord-027316-echxuw74	63	47	of	of	IN
cord-027316-echxuw74	63	48	words	word	NNS
cord-027316-echxuw74	63	49	representations	representation	NNS
cord-027316-echxuw74	63	50	.	.	.
cord-027316-echxuw74	64	1	Here	here	RB
cord-027316-echxuw74	64	2	,	,	,
cord-027316-echxuw74	64	3	for	for	IN
cord-027316-echxuw74	64	4	all	all	DT
cord-027316-echxuw74	64	5	similarity	similarity	NN
cord-027316-echxuw74	64	6	measures	measure	NNS
cord-027316-echxuw74	64	7	,	,	,
cord-027316-echxuw74	64	8	both	both	CC
cord-027316-echxuw74	64	9	Euclidean	euclidean	JJ
cord-027316-echxuw74	64	10	norm	norm	NN
cord-027316-echxuw74	64	11	and	and	CC
cord-027316-echxuw74	64	12	cosine	cosine	NN
cord-027316-echxuw74	64	13	similarity	similarity	NN
cord-027316-echxuw74	64	14	(	(	-LRB-
cord-027316-echxuw74	64	15	dissimilarity	dissimilarity	NN
cord-027316-echxuw74	64	16	)	)	-RRB-
cord-027316-echxuw74	64	17	have	have	VBP
cord-027316-echxuw74	64	18	been	be	VBN
cord-027316-echxuw74	64	19	used	use	VBN
cord-027316-echxuw74	64	20	.	.	.
cord-027316-echxuw74	65	1	In	in	IN
cord-027316-echxuw74	65	2	this	this	DT
cord-027316-echxuw74	65	3	work	work	NN
cord-027316-echxuw74	65	4	,	,	,
cord-027316-echxuw74	65	5	the	the	DT
cord-027316-echxuw74	65	6	embedding	embedding	NN
cord-027316-echxuw74	65	7	error	error	NN
cord-027316-echxuw74	65	8	using	use	VBG
cord-027316-echxuw74	65	9	the	the	DT
cord-027316-echxuw74	65	10	Euclidean	euclidean	JJ
cord-027316-echxuw74	65	11	norm	norm	NN
cord-027316-echxuw74	65	12	is	be	VBZ
cord-027316-echxuw74	65	13	used	use	VBN
cord-027316-echxuw74	65	14	,	,	,
cord-027316-echxuw74	65	15	Alternatively	alternatively	RB
cord-027316-echxuw74	65	16	,	,	,
cord-027316-echxuw74	65	17	we	-PRON-	PRP
cord-027316-echxuw74	65	18	could	could	MD
cord-027316-echxuw74	65	19	include	include	VB
cord-027316-echxuw74	65	20	the	the	DT
cord-027316-echxuw74	65	21	embedding	embedding	NN
cord-027316-echxuw74	65	22	error	error	NN
cord-027316-echxuw74	65	23	of	of	IN
cord-027316-echxuw74	65	24	the	the	DT
cord-027316-echxuw74	65	25	output	output	NN
cord-027316-echxuw74	65	26	sequence	sequence	NN
cord-027316-echxuw74	65	27	in	in	IN
cord-027316-echxuw74	65	28	Eq	Eq	NNP
cord-027316-echxuw74	65	29	.	.	.
cord-027316-echxuw74	66	1	(	(	-LRB-
cord-027316-echxuw74	66	2	10	10	CD
cord-027316-echxuw74	66	3	)	)	-RRB-
cord-027316-echxuw74	66	4	.	.	.
cord-027316-echxuw74	67	1	When	when	WRB
cord-027316-echxuw74	67	2	the	the	DT
cord-027316-echxuw74	67	3	input	input	NN
cord-027316-echxuw74	67	4	sequence	sequence	NN
cord-027316-echxuw74	67	5	(	(	-LRB-
cord-027316-echxuw74	67	6	or	or	CC
cord-027316-echxuw74	67	7	the	the	DT
cord-027316-echxuw74	67	8	dictionary	dictionary	NN
cord-027316-echxuw74	67	9	)	)	-RRB-
cord-027316-echxuw74	67	10	is	be	VBZ
cord-027316-echxuw74	67	11	too	too	RB
cord-027316-echxuw74	67	12	long	long	JJ
cord-027316-echxuw74	67	13	,	,	,
cord-027316-echxuw74	67	14	to	to	TO
cord-027316-echxuw74	67	15	prevent	prevent	VB
cord-027316-echxuw74	67	16	high	high	JJ
cord-027316-echxuw74	67	17	computational	computational	JJ
cord-027316-echxuw74	67	18	complexity	complexity	NN
cord-027316-echxuw74	67	19	of	of	IN
cord-027316-echxuw74	67	20	computing	compute	VBG
cord-027316-echxuw74	67	21	similarity	similarity	NN
cord-027316-echxuw74	67	22	of	of	IN
cord-027316-echxuw74	67	23	each	each	DT
cord-027316-echxuw74	67	24	specific	specific	JJ
cord-027316-echxuw74	67	25	word	word	NN
cord-027316-echxuw74	67	26	with	with	IN
cord-027316-echxuw74	67	27	all	all	DT
cord-027316-echxuw74	67	28	other	other	JJ
cord-027316-echxuw74	67	29	words	word	NNS
cord-027316-echxuw74	67	30	,	,	,
cord-027316-echxuw74	67	31	we	-PRON-	PRP
cord-027316-echxuw74	67	32	choose	choose	VBP
cord-027316-echxuw74	67	33	a	a	DT
cord-027316-echxuw74	67	34	random	random	JJ
cord-027316-echxuw74	67	35	(	(	-LRB-
cord-027316-echxuw74	67	36	uniform	uniform	JJ
cord-027316-echxuw74	67	37	)	)	-RRB-
cord-027316-echxuw74	67	38	sample	sample	NN
cord-027316-echxuw74	67	39	of	of	IN
cord-027316-echxuw74	67	40	the	the	DT
cord-027316-echxuw74	67	41	input	input	NN
cord-027316-echxuw74	67	42	sequences	sequence	NNS
cord-027316-echxuw74	67	43	to	to	TO
cord-027316-echxuw74	67	44	compute	compute	VB
cord-027316-echxuw74	67	45	the	the	DT
cord-027316-echxuw74	67	46	embedding	embedding	NN
cord-027316-echxuw74	67	47	error	error	NN
cord-027316-echxuw74	67	48	.	.	.
cord-027316-echxuw74	68	1	The	the	DT
cord-027316-echxuw74	68	2	regularization	regularization	NN
cord-027316-echxuw74	68	3	parameter	parameter	NN
cord-027316-echxuw74	68	4	,	,	,
cord-027316-echxuw74	68	5	α	α	NN
cord-027316-echxuw74	68	6	,	,	,
cord-027316-echxuw74	68	7	is	be	VBZ
cord-027316-echxuw74	68	8	computed	compute	VBN
cord-027316-echxuw74	68	9	using	use	VBG
cord-027316-echxuw74	68	10	cross	cross	NN
cord-027316-echxuw74	68	11	validation	validation	NN
cord-027316-echxuw74	68	12	[	[	-LRB-
cord-027316-echxuw74	68	13	26	26	CD
cord-027316-echxuw74	68	14	]	]	-RRB-
cord-027316-echxuw74	68	15	[	[	-LRB-
cord-027316-echxuw74	68	16	27	27	CD
cord-027316-echxuw74	68	17	]	]	-RRB-
cord-027316-echxuw74	68	18	[	[	-LRB-
cord-027316-echxuw74	68	19	28	28	CD
cord-027316-echxuw74	68	20	]	]	-RRB-
cord-027316-echxuw74	68	21	[	[	-LRB-
cord-027316-echxuw74	68	22	29	29	CD
cord-027316-echxuw74	68	23	]	]	-RRB-
cord-027316-echxuw74	69	1	[	[	-LRB-
cord-027316-echxuw74	69	2	30	30	CD
cord-027316-echxuw74	69	3	]	]	-RRB-
cord-027316-echxuw74	69	4	[	[	-LRB-
cord-027316-echxuw74	69	5	31	31	CD
cord-027316-echxuw74	69	6	]	]	-RRB-
cord-027316-echxuw74	69	7	.	.	.
cord-027316-echxuw74	70	1	Alternatively	alternatively	RB
cord-027316-echxuw74	70	2	,	,	,
cord-027316-echxuw74	70	3	adaptive	adaptive	JJ
cord-027316-echxuw74	70	4	regularization	regularization	NN
cord-027316-echxuw74	70	5	parameters	parameter	NNS
cord-027316-echxuw74	70	6	[	[	-LRB-
cord-027316-echxuw74	70	7	82	82	CD
cord-027316-echxuw74	70	8	,	,	,
cord-027316-echxuw74	70	9	83	83	CD
cord-027316-echxuw74	70	10	]	]	-RRB-
cord-027316-echxuw74	70	11	could	could	MD
cord-027316-echxuw74	70	12	be	be	VB
cord-027316-echxuw74	70	13	used	use	VBN
cord-027316-echxuw74	70	14	.	.	.
cord-027316-echxuw74	71	1	This	this	DT
cord-027316-echxuw74	71	2	model	model	NN
cord-027316-echxuw74	71	3	was	be	VBD
cord-027316-echxuw74	71	4	applied	apply	VBN
cord-027316-echxuw74	71	5	on	on	IN
cord-027316-echxuw74	71	6	Wikipedia	Wikipedia	NNP
cord-027316-echxuw74	71	7	datasets	dataset	NNS
cord-027316-echxuw74	71	8	for	for	IN
cord-027316-echxuw74	71	9	English	english	JJ
cord-027316-echxuw74	71	10	-	-	HYPH
cord-027316-echxuw74	71	11	German	german	JJ
cord-027316-echxuw74	71	12	translation	translation	NN
cord-027316-echxuw74	71	13	(	(	-LRB-
cord-027316-echxuw74	71	14	one	one	CD
cord-027316-echxuw74	71	15	-	-	HYPH
cord-027316-echxuw74	71	16	way	way	NN
cord-027316-echxuw74	71	17	translation	translation	NN
cord-027316-echxuw74	71	18	)	)	-RRB-
cord-027316-echxuw74	71	19	with	with	IN
cord-027316-echxuw74	71	20	1000	1000	CD
cord-027316-echxuw74	71	21	sentences	sentence	NNS
cord-027316-echxuw74	71	22	.	.	.
cord-027316-echxuw74	72	1	The	the	DT
cord-027316-echxuw74	72	2	idea	idea	NN
cord-027316-echxuw74	72	3	was	be	VBD
cord-027316-echxuw74	72	4	to	to	TO
cord-027316-echxuw74	72	5	determine	determine	VB
cord-027316-echxuw74	72	6	which	which	WDT
cord-027316-echxuw74	72	7	specific	specific	JJ
cord-027316-echxuw74	72	8	input	input	NN
cord-027316-echxuw74	72	9	word	word	NN
cord-027316-echxuw74	72	10	(	(	-LRB-
cord-027316-echxuw74	72	11	in	in	IN
cord-027316-echxuw74	72	12	English	English	NNP
cord-027316-echxuw74	72	13	)	)	-RRB-
cord-027316-echxuw74	72	14	is	be	VBZ
cord-027316-echxuw74	72	15	the	the	DT
cord-027316-echxuw74	72	16	most	most	RBS
cord-027316-echxuw74	72	17	important	important	JJ
cord-027316-echxuw74	72	18	one	one	NN
cord-027316-echxuw74	72	19	for	for	IN
cord-027316-echxuw74	72	20	the	the	DT
cord-027316-echxuw74	72	21	corresponding	corresponding	JJ
cord-027316-echxuw74	72	22	German	german	JJ
cord-027316-echxuw74	72	23	translation	translation	NN
cord-027316-echxuw74	72	24	.	.	.
cord-027316-echxuw74	73	1	The	the	DT
cord-027316-echxuw74	73	2	results	result	NNS
cord-027316-echxuw74	73	3	were	be	VBD
cord-027316-echxuw74	73	4	often	often	RB
cord-027316-echxuw74	73	5	an	an	DT
cord-027316-echxuw74	73	6	almost	almost	RB
cord-027316-echxuw74	73	7	diagonal	diagonal	JJ
cord-027316-echxuw74	73	8	weight	weight	NN
cord-027316-echxuw74	73	9	matrix	matrix	NN
cord-027316-echxuw74	73	10	,	,	,
cord-027316-echxuw74	73	11	with	with	IN
cord-027316-echxuw74	73	12	few	few	JJ
cord-027316-echxuw74	73	13	non	non	JJ
cord-027316-echxuw74	73	14	-	-	JJ
cord-027316-echxuw74	73	15	zero	zero	CD
cord-027316-echxuw74	73	16	off	off	IN
cord-027316-echxuw74	73	17	diagonal	diagonal	JJ
cord-027316-echxuw74	73	18	entries	entry	NNS
cord-027316-echxuw74	73	19	,	,	,
cord-027316-echxuw74	73	20	indicating	indicate	VBG
cord-027316-echxuw74	73	21	the	the	DT
cord-027316-echxuw74	73	22	significance	significance	NN
cord-027316-echxuw74	73	23	of	of	IN
cord-027316-echxuw74	73	24	the	the	DT
cord-027316-echxuw74	73	25	corresponding	corresponding	JJ
cord-027316-echxuw74	73	26	word(s	word(s	NN
cord-027316-echxuw74	73	27	)	)	-RRB-
cord-027316-echxuw74	73	28	in	in	IN
cord-027316-echxuw74	73	29	the	the	DT
cord-027316-echxuw74	73	30	original	original	JJ
cord-027316-echxuw74	73	31	language	language	NN
cord-027316-echxuw74	73	32	(	(	-LRB-
cord-027316-echxuw74	73	33	English	English	NNP
cord-027316-echxuw74	73	34	)	)	-RRB-
cord-027316-echxuw74	73	35	.	.	.
cord-027316-echxuw74	74	1	Since	since	IN
cord-027316-echxuw74	74	2	the	the	DT
cord-027316-echxuw74	74	3	model	model	NN
cord-027316-echxuw74	74	4	is	be	VBZ
cord-027316-echxuw74	74	5	an	an	DT
cord-027316-echxuw74	74	6	unsupervised	unsupervised	JJ
cord-027316-echxuw74	74	7	approach	approach	NN
cord-027316-echxuw74	74	8	,	,	,
cord-027316-echxuw74	74	9	it	-PRON-	PRP
cord-027316-echxuw74	74	10	's	be	VBZ
cord-027316-echxuw74	74	11	hard	hard	JJ
cord-027316-echxuw74	74	12	to	to	TO
cord-027316-echxuw74	74	13	evaluate	evaluate	VB
cord-027316-echxuw74	74	14	its	-PRON-	PRP$
cord-027316-echxuw74	74	15	performance	performance	NN
cord-027316-echxuw74	74	16	without	without	IN
cord-027316-echxuw74	74	17	using	use	VBG
cord-027316-echxuw74	74	18	domain	domain	NN
cord-027316-echxuw74	74	19	knowledge	knowledge	NN
cord-027316-echxuw74	74	20	.	.	.
cord-027316-echxuw74	75	1	The	the	DT
cord-027316-echxuw74	75	2	next	next	JJ
cord-027316-echxuw74	75	3	step	step	NN
cord-027316-echxuw74	75	4	in	in	IN
cord-027316-echxuw74	75	5	this	this	DT
cord-027316-echxuw74	75	6	work	work	NN
cord-027316-echxuw74	75	7	would	would	MD
cord-027316-echxuw74	75	8	be	be	VB
cord-027316-echxuw74	75	9	to	to	TO
cord-027316-echxuw74	75	10	develop	develop	VB
cord-027316-echxuw74	75	11	a	a	DT
cord-027316-echxuw74	75	12	unified	unified	JJ
cord-027316-echxuw74	75	13	and	and	CC
cord-027316-echxuw74	75	14	interpretable	interpretable	JJ
cord-027316-echxuw74	75	15	metric	metric	NN
cord-027316-echxuw74	75	16	for	for	IN
cord-027316-echxuw74	75	17	automatic	automatic	JJ
cord-027316-echxuw74	75	18	testing	testing	NN
cord-027316-echxuw74	75	19	and	and	CC
cord-027316-echxuw74	75	20	evaluation	evaluation	NN
cord-027316-echxuw74	75	21	of	of	IN
cord-027316-echxuw74	75	22	the	the	DT
cord-027316-echxuw74	75	23	model	model	NN
cord-027316-echxuw74	75	24	without	without	IN
cord-027316-echxuw74	75	25	using	use	VBG
cord-027316-echxuw74	75	26	any	any	DT
cord-027316-echxuw74	75	27	domain	domain	NN
cord-027316-echxuw74	75	28	knowledge	knowledge	NN
cord-027316-echxuw74	75	29	and	and	CC
cord-027316-echxuw74	75	30	also	also	RB
cord-027316-echxuw74	75	31	to	to	TO
cord-027316-echxuw74	75	32	apply	apply	VB
cord-027316-echxuw74	75	33	the	the	DT
cord-027316-echxuw74	75	34	model	model	NN
cord-027316-echxuw74	75	35	to	to	IN
cord-027316-echxuw74	75	36	other	other	JJ
cord-027316-echxuw74	75	37	applications	application	NNS
cord-027316-echxuw74	75	38	such	such	JJ
cord-027316-echxuw74	75	39	as	as	IN
cord-027316-echxuw74	75	40	sentiment	sentiment	NN
cord-027316-echxuw74	75	41	analysis	analysis	NN
cord-027316-echxuw74	75	42	.	.	.
cord-027316-echxuw74	75	43	 	 	_SP
cord-027316-echxuw74	76	1	Inverse	Inverse	NNP
cord-027316-echxuw74	76	2	Problems	problem	NNS
cord-027316-echxuw74	76	3	:	:	:
cord-027316-echxuw74	77	1	Principles	principle	NNS
cord-027316-echxuw74	77	2	and	and	CC
cord-027316-echxuw74	77	3	Applications	Applications	NNPS
cord-027316-echxuw74	77	4	in	in	IN
cord-027316-echxuw74	77	5	Geophysics	Geophysics	NNP
cord-027316-echxuw74	77	6	,	,	,
cord-027316-echxuw74	77	7	Technology	Technology	NNP
cord-027316-echxuw74	77	8	,	,	,
cord-027316-echxuw74	77	9	and	and	CC
cord-027316-echxuw74	77	10	Medicine	Medicine	NNP
cord-027316-echxuw74	77	11	Polosukhin	Polosukhin	NNP
cord-027316-echxuw74	77	12	:	:	:
cord-027316-echxuw74	77	13	attention	attention	NN
cord-027316-echxuw74	77	14	is	be	VBZ
cord-027316-echxuw74	77	15	all	all	DT
cord-027316-echxuw74	77	16	you	-PRON-	PRP
cord-027316-echxuw74	77	17	need	need	VBP
cord-027316-echxuw74	77	18	Domain	domain	NN
cord-027316-echxuw74	77	19	adaptation	adaptation	NN
cord-027316-echxuw74	77	20	via	via	IN
cord-027316-echxuw74	77	21	pseudo	pseudo	JJ
cord-027316-echxuw74	77	22	in	in	IN
cord-027316-echxuw74	77	23	-	-	HYPH
cord-027316-echxuw74	77	24	domain	domain	NN
cord-027316-echxuw74	77	25	data	data	NN
cord-027316-echxuw74	77	26	selection	selection	NN
cord-027316-echxuw74	77	27	Multiple	multiple	JJ
cord-027316-echxuw74	77	28	object	object	NN
cord-027316-echxuw74	77	29	recognition	recognition	NN
cord-027316-echxuw74	77	30	with	with	IN
cord-027316-echxuw74	77	31	visual	visual	JJ
cord-027316-echxuw74	77	32	attention	attention	NN
cord-027316-echxuw74	77	33	Neural	neural	JJ
cord-027316-echxuw74	77	34	machine	machine	NN
cord-027316-echxuw74	77	35	translation	translation	NN
cord-027316-echxuw74	77	36	by	by	IN
cord-027316-echxuw74	77	37	jointly	jointly	RB
cord-027316-echxuw74	77	38	learning	learn	VBG
cord-027316-echxuw74	77	39	to	to	TO
cord-027316-echxuw74	77	40	align	align	VB
cord-027316-echxuw74	77	41	and	and	CC
cord-027316-echxuw74	77	42	translate	translate	VB
cord-027316-echxuw74	77	43	The	the	DT
cord-027316-echxuw74	77	44	dropout	dropout	NN
cord-027316-echxuw74	77	45	learning	learning	NN
cord-027316-echxuw74	77	46	algorithm	algorithm	NN
cord-027316-echxuw74	77	47	Deep	deep	JJ
cord-027316-echxuw74	77	48	Learning	Learning	NNP
cord-027316-echxuw74	78	1	an	an	DT
cord-027316-echxuw74	78	2	Unsupervised	Unsupervised	NNP
cord-027316-echxuw74	78	3	Feature	feature	NN
cord-027316-echxuw74	78	4	Learning	learn	VBG
cord-027316-echxuw74	78	5	NIPS	nip	NNS
cord-027316-echxuw74	78	6	2012	2012	CD
cord-027316-echxuw74	79	1	Workshop	Workshop	NNP
cord-027316-echxuw74	79	2	NESTA	NESTA	NNP
cord-027316-echxuw74	79	3	,	,	,
cord-027316-echxuw74	79	4	a	a	DT
cord-027316-echxuw74	79	5	fast	fast	JJ
cord-027316-echxuw74	79	6	and	and	CC
cord-027316-echxuw74	79	7	accurate	accurate	JJ
cord-027316-echxuw74	79	8	first	first	JJ
cord-027316-echxuw74	79	9	-	-	HYPH
cord-027316-echxuw74	79	10	order	order	NN
cord-027316-echxuw74	79	11	method	method	NN
cord-027316-echxuw74	79	12	for	for	IN
cord-027316-echxuw74	79	13	sparse	sparse	JJ
cord-027316-echxuw74	79	14	recovery	recovery	NN
cord-027316-echxuw74	80	1	Learning	learn	VBG
cord-027316-echxuw74	80	2	long	long	JJ
cord-027316-echxuw74	80	3	-	-	HYPH
cord-027316-echxuw74	80	4	term	term	NN
cord-027316-echxuw74	80	5	de	de	JJ
cord-027316-echxuw74	80	6	-	-	NN
cord-027316-echxuw74	80	7	pendencies	pendencie	NNS
cord-027316-echxuw74	80	8	with	with	IN
cord-027316-echxuw74	80	9	gradient	gradient	NN
cord-027316-echxuw74	80	10	descent	descent	NN
cord-027316-echxuw74	80	11	is	be	VBZ
cord-027316-echxuw74	80	12	difficult	difficult	JJ
cord-027316-echxuw74	81	1	A	a	DT
cord-027316-echxuw74	81	2	neural	neural	JJ
cord-027316-echxuw74	81	3	probabilistic	probabilistic	JJ
cord-027316-echxuw74	81	4	language	language	NN
cord-027316-echxuw74	81	5	model	model	NN
cord-027316-echxuw74	81	6	Theano	Theano	NNP
cord-027316-echxuw74	81	7	:	:	:
cord-027316-echxuw74	82	1	a	a	DT
cord-027316-echxuw74	82	2	CPU	cpu	NN
cord-027316-echxuw74	82	3	and	and	CC
cord-027316-echxuw74	82	4	GPU	GPU	NNP
cord-027316-echxuw74	82	5	math	math	NN
cord-027316-echxuw74	82	6	expression	expression	NN
cord-027316-echxuw74	82	7	compiler	compiler	NN
cord-027316-echxuw74	82	8	Audio	audio	NN
cord-027316-echxuw74	82	9	chord	chord	NN
cord-027316-echxuw74	82	10	recognition	recognition	NN
cord-027316-echxuw74	82	11	with	with	IN
cord-027316-echxuw74	82	12	recurrent	recurrent	JJ
cord-027316-echxuw74	82	13	neural	neural	JJ
cord-027316-echxuw74	82	14	networks	network	NNS
cord-027316-echxuw74	83	1	A	a	DT
cord-027316-echxuw74	83	2	singular	singular	JJ
cord-027316-echxuw74	83	3	value	value	NN
cord-027316-echxuw74	83	4	thresholding	thresholding	NN
cord-027316-echxuw74	83	5	algorithm	algorithm	NN
cord-027316-echxuw74	83	6	for	for	IN
cord-027316-echxuw74	83	7	matrix	matrix	NN
cord-027316-echxuw74	83	8	completion	completion	NN
cord-027316-echxuw74	83	9	Exact	Exact	NNP
cord-027316-echxuw74	83	10	matrix	matrix	NN
cord-027316-echxuw74	83	11	completion	completion	NN
cord-027316-echxuw74	83	12	via	via	IN
cord-027316-echxuw74	83	13	convex	convex	JJ
cord-027316-echxuw74	83	14	optimization	optimization	NN
cord-027316-echxuw74	83	15	Compressive	Compressive	NNP
cord-027316-echxuw74	83	16	sampling	sampling	NN
cord-027316-echxuw74	83	17	Long	long	JJ
cord-027316-echxuw74	83	18	short	short	JJ
cord-027316-echxuw74	83	19	-	-	HYPH
cord-027316-echxuw74	83	20	term	term	NN
cord-027316-echxuw74	83	21	memory	memory	NN
cord-027316-echxuw74	83	22	-	-	HYPH
cord-027316-echxuw74	83	23	networks	network	NNS
cord-027316-echxuw74	83	24	for	for	IN
cord-027316-echxuw74	83	25	machine	machine	NN
cord-027316-echxuw74	83	26	reading	read	VBG
cord-027316-echxuw74	83	27	Learning	learn	VBG
cord-027316-echxuw74	83	28	phrase	phrase	NN
cord-027316-echxuw74	83	29	representations	representation	NNS
cord-027316-echxuw74	83	30	using	use	VBG
cord-027316-echxuw74	83	31	RNN	RNN	NNP
cord-027316-echxuw74	83	32	encoder	encoder	NN
cord-027316-echxuw74	83	33	-	-	HYPH
cord-027316-echxuw74	83	34	decoder	decoder	NN
cord-027316-echxuw74	83	35	for	for	IN
cord-027316-echxuw74	83	36	statistical	statistical	JJ
cord-027316-echxuw74	83	37	machine	machine	NN
cord-027316-echxuw74	83	38	translation	translation	NN
cord-027316-echxuw74	83	39	Framewise	Framewise	NNP
cord-027316-echxuw74	83	40	phoneme	phoneme	NN
cord-027316-echxuw74	83	41	classification	classification	NN
cord-027316-echxuw74	83	42	with	with	IN
cord-027316-echxuw74	83	43	bidirectional	bidirectional	JJ
cord-027316-echxuw74	83	44	LSTM	LSTM	NNP
cord-027316-echxuw74	83	45	and	and	CC
cord-027316-echxuw74	83	46	other	other	JJ
cord-027316-echxuw74	83	47	neural	neural	JJ
cord-027316-echxuw74	83	48	network	network	NN
cord-027316-echxuw74	83	49	architectures	architecture	NNS
cord-027316-echxuw74	84	1	Generating	generate	VBG
cord-027316-echxuw74	84	2	sequences	sequence	NNS
cord-027316-echxuw74	84	3	with	with	IN
cord-027316-echxuw74	84	4	recurrent	recurrent	JJ
cord-027316-echxuw74	84	5	neural	neural	JJ
cord-027316-echxuw74	84	6	networks	network	NNS
cord-027316-echxuw74	85	1	The	the	DT
cord-027316-echxuw74	85	2	Elements	Elements	NNPS
cord-027316-echxuw74	85	3	of	of	IN
cord-027316-echxuw74	85	4	Statistical	Statistical	NNP
cord-027316-echxuw74	85	5	Learning	Learning	NNP
cord-027316-echxuw74	85	6	;	;	:
cord-027316-echxuw74	85	7	Data	Data	NNP
cord-027316-echxuw74	85	8	miNing	miNing	NNP
cord-027316-echxuw74	85	9	,	,	,
cord-027316-echxuw74	85	10	Inference	Inference	NNP
cord-027316-echxuw74	85	11	and	and	CC
cord-027316-echxuw74	85	12	Prediction	Prediction	NNP
cord-027316-echxuw74	85	13	Handwritten	Handwritten	NNP
cord-027316-echxuw74	85	14	digit	digit	NN
cord-027316-echxuw74	85	15	recognition	recognition	NN
cord-027316-echxuw74	85	16	via	via	IN
cord-027316-echxuw74	85	17	deformable	deformable	JJ
cord-027316-echxuw74	85	18	prototypes	prototype	NNS
cord-027316-echxuw74	85	19	Gene	Gene	NNP
cord-027316-echxuw74	85	20	Shaving	Shaving	NNP
cord-027316-echxuw74	85	21	'	'	''
cord-027316-echxuw74	85	22	as	as	IN
cord-027316-echxuw74	85	23	a	a	DT
cord-027316-echxuw74	85	24	method	method	NN
cord-027316-echxuw74	85	25	for	for	IN
cord-027316-echxuw74	85	26	identifying	identify	VBG
cord-027316-echxuw74	85	27	distinct	distinct	JJ
cord-027316-echxuw74	85	28	sets	set	NNS
cord-027316-echxuw74	85	29	of	of	IN
cord-027316-echxuw74	85	30	genes	gene	NNS
cord-027316-echxuw74	85	31	with	with	IN
cord-027316-echxuw74	85	32	similar	similar	JJ
cord-027316-echxuw74	85	33	expression	expression	NN
cord-027316-echxuw74	85	34	patterns	pattern	NNS
cord-027316-echxuw74	85	35	Matrix	Matrix	NNP
cord-027316-echxuw74	85	36	Completion	Completion	NNP
cord-027316-echxuw74	85	37	via	via	IN
cord-027316-echxuw74	85	38	Iterative	Iterative	NNP
cord-027316-echxuw74	85	39	Soft	Soft	NNP
cord-027316-echxuw74	85	40	-	-	HYPH
cord-027316-echxuw74	85	41	Thresholded	Thresholded	NNP
cord-027316-echxuw74	85	42	SVD	SVD	NNP
cord-027316-echxuw74	85	43	Package	Package	NNP
cord-027316-echxuw74	85	44	'	'	``
cord-027316-echxuw74	85	45	impute	impute	NN
cord-027316-echxuw74	85	46	'	'	''
cord-027316-echxuw74	85	47	.	.	.
cord-027316-echxuw74	86	1	CRAN	CRAN	NNP
cord-027316-echxuw74	86	2	Multilingual	Multilingual	NNP
cord-027316-echxuw74	86	3	distributed	distribute	VBN
cord-027316-echxuw74	86	4	representations	representation	NNS
cord-027316-echxuw74	86	5	without	without	IN
cord-027316-echxuw74	86	6	word	word	NN
cord-027316-echxuw74	86	7	alignment	alignment	NN
cord-027316-echxuw74	86	8	Advances	advance	VBZ
cord-027316-echxuw74	86	9	in	in	IN
cord-027316-echxuw74	86	10	natural	natural	JJ
cord-027316-echxuw74	86	11	language	language	NN
cord-027316-echxuw74	86	12	processing	processing	NN
cord-027316-echxuw74	87	1	Long	long	JJ
cord-027316-echxuw74	87	2	short	short	JJ
cord-027316-echxuw74	87	3	-	-	HYPH
cord-027316-echxuw74	87	4	term	term	NN
cord-027316-echxuw74	87	5	memory	memory	NN
cord-027316-echxuw74	87	6	Gradient	Gradient	NNP
cord-027316-echxuw74	87	7	flow	flow	NN
cord-027316-echxuw74	87	8	in	in	IN
cord-027316-echxuw74	87	9	recurrent	recurrent	JJ
cord-027316-echxuw74	87	10	nets	net	NNS
cord-027316-echxuw74	87	11	:	:	:
cord-027316-echxuw74	88	1	the	the	DT
cord-027316-echxuw74	88	2	difficulty	difficulty	NN
cord-027316-echxuw74	88	3	of	of	IN
cord-027316-echxuw74	88	4	learning	learn	VBG
cord-027316-echxuw74	88	5	long	long	JJ
cord-027316-echxuw74	88	6	-	-	HYPH
cord-027316-echxuw74	88	7	term	term	NN
cord-027316-echxuw74	88	8	dependencies	dependency	NNS
cord-027316-echxuw74	88	9	Regularization	Regularization	NNP
cord-027316-echxuw74	88	10	for	for	IN
cord-027316-echxuw74	88	11	Applied	Applied	NNP
cord-027316-echxuw74	88	12	Inverse	Inverse	NNP
cord-027316-echxuw74	88	13	and	and	CC
cord-027316-echxuw74	88	14	Ill	Ill	NNP
cord-027316-echxuw74	88	15	-	-	HYPH
cord-027316-echxuw74	88	16	Posed	pose	VBN
cord-027316-echxuw74	88	17	problems	problem	NNS
cord-027316-echxuw74	88	18	Compositional	compositional	JJ
cord-027316-echxuw74	88	19	attention	attention	NN
cord-027316-echxuw74	88	20	networks	network	NNS
cord-027316-echxuw74	88	21	for	for	IN
cord-027316-echxuw74	88	22	machine	machine	NN
cord-027316-echxuw74	88	23	reasoning	reason	VBG
cord-027316-echxuw74	88	24	Two	two	CD
cord-027316-echxuw74	88	25	case	case	NN
cord-027316-echxuw74	88	26	studies	study	NNS
cord-027316-echxuw74	88	27	in	in	IN
cord-027316-echxuw74	88	28	the	the	DT
cord-027316-echxuw74	88	29	application	application	NN
cord-027316-echxuw74	88	30	of	of	IN
cord-027316-echxuw74	88	31	principal	principal	JJ
cord-027316-echxuw74	88	32	component	component	NN
cord-027316-echxuw74	88	33	Principal	Principal	NNP
cord-027316-echxuw74	88	34	Component	Component	NNP
cord-027316-echxuw74	88	35	Analysis	Analysis	NNP
cord-027316-echxuw74	88	36	Rotation	Rotation	NNP
cord-027316-echxuw74	88	37	of	of	IN
cord-027316-echxuw74	88	38	principal	principal	JJ
cord-027316-echxuw74	88	39	components	component	NNS
cord-027316-echxuw74	88	40	:	:	:
cord-027316-echxuw74	88	41	choice	choice	NN
cord-027316-echxuw74	88	42	of	of	IN
cord-027316-echxuw74	88	43	normalization	normalization	NN
cord-027316-echxuw74	88	44	constraints	constraint	NNS
cord-027316-echxuw74	88	45	A	a	DT
cord-027316-echxuw74	88	46	modified	modify	VBN
cord-027316-echxuw74	88	47	principal	principal	JJ
cord-027316-echxuw74	88	48	component	component	NN
cord-027316-echxuw74	88	49	technique	technique	NN
cord-027316-echxuw74	88	50	based	base	VBN
cord-027316-echxuw74	88	51	on	on	IN
cord-027316-echxuw74	88	52	the	the	DT
cord-027316-echxuw74	88	53	LASSO	LASSO	NNP
cord-027316-echxuw74	88	54	Recurrent	Recurrent	NNP
cord-027316-echxuw74	88	55	continuous	continuous	JJ
cord-027316-echxuw74	88	56	translation	translation	NN
cord-027316-echxuw74	88	57	models	model	NNS
cord-027316-echxuw74	88	58	Statistical	statistical	JJ
cord-027316-echxuw74	88	59	Machine	machine	NN
cord-027316-echxuw74	89	1	Translation	translation	NN
cord-027316-echxuw74	89	2	Structured	structured	JJ
cord-027316-echxuw74	89	3	attention	attention	NN
cord-027316-echxuw74	89	4	networks	network	NNS
cord-027316-echxuw74	90	1	Statistical	statistical	JJ
cord-027316-echxuw74	90	2	phrase	phrase	NN
cord-027316-echxuw74	90	3	-	-	HYPH
cord-027316-echxuw74	90	4	based	base	VBN
cord-027316-echxuw74	90	5	translation	translation	NN
cord-027316-echxuw74	90	6	Learning	learn	VBG
cord-027316-echxuw74	90	7	phrase	phrase	NN
cord-027316-echxuw74	90	8	representations	representation	NNS
cord-027316-echxuw74	90	9	using	use	VBG
cord-027316-echxuw74	90	10	RNN	RNN	NNP
cord-027316-echxuw74	90	11	encoder	encoder	NN
cord-027316-echxuw74	90	12	-	-	HYPH
cord-027316-echxuw74	90	13	decoder	decoder	NN
cord-027316-echxuw74	90	14	for	for	IN
cord-027316-echxuw74	90	15	statistical	statistical	JJ
cord-027316-echxuw74	90	16	machine	machine	NN
cord-027316-echxuw74	90	17	translation	translation	NN
cord-027316-echxuw74	91	1	Conditional	conditional	JJ
cord-027316-echxuw74	91	2	random	random	JJ
cord-027316-echxuw74	91	3	fields	field	NNS
cord-027316-echxuw74	91	4	:	:	:
cord-027316-echxuw74	92	1	Probabilistic	probabilistic	JJ
cord-027316-echxuw74	92	2	models	model	NNS
cord-027316-echxuw74	92	3	for	for	IN
cord-027316-echxuw74	92	4	segmenting	segment	VBG
cord-027316-echxuw74	92	5	and	and	CC
cord-027316-echxuw74	92	6	labeling	label	VBG
cord-027316-echxuw74	92	7	sequence	sequence	NN
cord-027316-echxuw74	92	8	data	datum	NNS
cord-027316-echxuw74	92	9	Neural	Neural	NNP
cord-027316-echxuw74	92	10	Networks	Networks	NNPS
cord-027316-echxuw74	92	11	:	:	:
cord-027316-echxuw74	93	1	Tricks	trick	NNS
cord-027316-echxuw74	93	2	of	of	IN
cord-027316-echxuw74	93	3	the	the	DT
cord-027316-echxuw74	93	4	Trade	Trade	NNP
cord-027316-echxuw74	93	5	A	a	DT
cord-027316-echxuw74	93	6	structured	structured	JJ
cord-027316-echxuw74	93	7	self	self	NN
cord-027316-echxuw74	93	8	-	-	HYPH
cord-027316-echxuw74	93	9	attentive	attentive	JJ
cord-027316-echxuw74	93	10	sentence	sentence	NN
cord-027316-echxuw74	93	11	embedding	embed	VBG
cord-027316-echxuw74	93	12	Effective	effective	JJ
cord-027316-echxuw74	93	13	Approaches	approach	NNS
cord-027316-echxuw74	93	14	to	to	IN
cord-027316-echxuw74	93	15	attention	attention	NN
cord-027316-echxuw74	93	16	-	-	HYPH
cord-027316-echxuw74	93	17	based	base	VBN
cord-027316-echxuw74	93	18	neural	neural	JJ
cord-027316-echxuw74	93	19	machine	machine	NN
cord-027316-echxuw74	93	20	translation	translation	NN
cord-027316-echxuw74	93	21	Learning	learn	VBG
cord-027316-echxuw74	93	22	to	to	TO
cord-027316-echxuw74	93	23	recognize	recognize	VB
cord-027316-echxuw74	93	24	features	feature	NNS
cord-027316-echxuw74	93	25	of	of	IN
cord-027316-echxuw74	93	26	valid	valid	JJ
cord-027316-echxuw74	93	27	textual	textual	JJ
cord-027316-echxuw74	93	28	entailments	entailment	NNS
cord-027316-echxuw74	93	29	Natural	natural	JJ
cord-027316-echxuw74	93	30	logic	logic	NN
cord-027316-echxuw74	93	31	for	for	IN
cord-027316-echxuw74	93	32	textual	textual	JJ
cord-027316-echxuw74	93	33	inference	inference	NN
cord-027316-echxuw74	93	34	Encyclopedia	Encyclopedia	NNP
cord-027316-echxuw74	93	35	of	of	IN
cord-027316-echxuw74	93	36	Language	Language	NNP
cord-027316-echxuw74	93	37	&	&	CC
cord-027316-echxuw74	93	38	Linguistics	Linguistics	NNP
cord-027316-echxuw74	93	39	Introduction	Introduction	NNP
cord-027316-echxuw74	93	40	to	to	IN
cord-027316-echxuw74	93	41	information	information	NN
cord-027316-echxuw74	93	42	retrieval	retrieval	NN
cord-027316-echxuw74	94	1	The	the	DT
cord-027316-echxuw74	94	2	Stanford	Stanford	NNP
cord-027316-echxuw74	94	3	CoreNLP	CoreNLP	NNP
cord-027316-echxuw74	94	4	Natural	Natural	NNP
cord-027316-echxuw74	94	5	Language	Language	NNP
cord-027316-echxuw74	94	6	Processing	Processing	NNP
cord-027316-echxuw74	94	7	Toolkit	Toolkit	NNP
cord-027316-echxuw74	94	8	.	.	.
cord-027316-echxuw74	95	1	Computer	computer	NN
cord-027316-echxuw74	95	2	Science	Science	NNP
cord-027316-echxuw74	95	3	,	,	,
cord-027316-echxuw74	95	4	ACL	ACL	NNP
cord-027316-echxuw74	95	5	Computational	computational	JJ
cord-027316-echxuw74	95	6	linguistics	linguistics	NN
cord-027316-echxuw74	95	7	and	and	CC
cord-027316-echxuw74	95	8	deep	deep	JJ
cord-027316-echxuw74	95	9	learning	learning	NN
cord-027316-echxuw74	96	1	Differentiating	differentiate	VBG
cord-027316-echxuw74	96	2	language	language	NN
cord-027316-echxuw74	96	3	usage	usage	NN
cord-027316-echxuw74	96	4	through	through	IN
cord-027316-echxuw74	96	5	topic	topic	NN
cord-027316-echxuw74	96	6	models	model	NNS
cord-027316-echxuw74	97	1	Effective	effective	JJ
cord-027316-echxuw74	97	2	approaches	approach	NNS
cord-027316-echxuw74	97	3	to	to	IN
cord-027316-echxuw74	97	4	attention	attention	NN
cord-027316-echxuw74	97	5	based	base	VBN
cord-027316-echxuw74	97	6	neural	neural	JJ
cord-027316-echxuw74	97	7	machine	machine	NN
cord-027316-echxuw74	97	8	translation	translation	NN
cord-027316-echxuw74	97	9	Application	Application	NNP
cord-027316-echxuw74	97	10	of	of	IN
cord-027316-echxuw74	97	11	DNN	DNN	NNP
cord-027316-echxuw74	97	12	for	for	IN
cord-027316-echxuw74	97	13	Modern	Modern	NNP
cord-027316-echxuw74	97	14	Data	Data	NNP
cord-027316-echxuw74	97	15	with	with	IN
cord-027316-echxuw74	97	16	two	two	CD
cord-027316-echxuw74	97	17	Examples	example	NNS
cord-027316-echxuw74	97	18	:	:	:
cord-027316-echxuw74	97	19	Recommender	Recommender	NNP
cord-027316-echxuw74	97	20	Systems	Systems	NNPS
cord-027316-echxuw74	97	21	&	&	CC
cord-027316-echxuw74	97	22	User	User	NNP
cord-027316-echxuw74	97	23	Recognition	Recognition	NNP
cord-027316-echxuw74	97	24	.	.	.
cord-027316-echxuw74	98	1	Deep	Deep	NNP
cord-027316-echxuw74	98	2	Learning	Learning	NNP
cord-027316-echxuw74	98	3	Summit	Summit	NNP
cord-027316-echxuw74	98	4	Standardization	Standardization	NNP
cord-027316-echxuw74	98	5	of	of	IN
cord-027316-echxuw74	98	6	featureless	featureless	JJ
cord-027316-echxuw74	98	7	variables	variable	NNS
cord-027316-echxuw74	98	8	for	for	IN
cord-027316-echxuw74	98	9	machine	machine	NN
cord-027316-echxuw74	98	10	learning	learning	NN
cord-027316-echxuw74	98	11	models	model	NNS
cord-027316-echxuw74	98	12	using	use	VBG
cord-027316-echxuw74	98	13	natural	natural	JJ
cord-027316-echxuw74	98	14	language	language	NN
cord-027316-echxuw74	98	15	processing	processing	NN
cord-027316-echxuw74	98	16	Generalized	generalize	VBN
cord-027316-echxuw74	98	17	variable	variable	JJ
cord-027316-echxuw74	98	18	conversion	conversion	NN
cord-027316-echxuw74	98	19	using	use	VBG
cord-027316-echxuw74	98	20	k	k	NN
cord-027316-echxuw74	98	21	-	-	HYPH
cord-027316-echxuw74	98	22	means	mean	VBZ
cord-027316-echxuw74	98	23	clustering	clustering	NN
cord-027316-echxuw74	98	24	and	and	CC
cord-027316-echxuw74	98	25	web	web	NN
cord-027316-echxuw74	98	26	scraping	scrape	VBG
cord-027316-echxuw74	98	27	An	an	DT
cord-027316-echxuw74	98	28	efficient	efficient	JJ
cord-027316-echxuw74	98	29	deep	deep	JJ
cord-027316-echxuw74	98	30	learning	learning	NN
cord-027316-echxuw74	98	31	model	model	NN
cord-027316-echxuw74	98	32	for	for	IN
cord-027316-echxuw74	98	33	recommender	recommender	NN
cord-027316-echxuw74	98	34	systems	system	NNS
cord-027316-echxuw74	98	35	Effectiveness	Effectiveness	NNP
cord-027316-echxuw74	98	36	of	of	IN
cord-027316-echxuw74	98	37	Representation	Representation	NNP
cord-027316-echxuw74	98	38	Learning	learn	VBG
cord-027316-echxuw74	98	39	for	for	IN
cord-027316-echxuw74	98	40	the	the	DT
cord-027316-echxuw74	98	41	Analysis	Analysis	NNP
cord-027316-echxuw74	98	42	of	of	IN
cord-027316-echxuw74	98	43	Human	Human	NNP
cord-027316-echxuw74	98	44	Behavior	Behavior	NNP
cord-027316-echxuw74	99	1	An	an	DT
cord-027316-echxuw74	99	2	evaluation	evaluation	NN
cord-027316-echxuw74	99	3	metric	metric	NN
cord-027316-echxuw74	99	4	for	for	IN
cord-027316-echxuw74	99	5	content	content	NN
cord-027316-echxuw74	99	6	providing	provide	VBG
cord-027316-echxuw74	99	7	models	model	NNS
cord-027316-echxuw74	99	8	,	,	,
cord-027316-echxuw74	99	9	recommendation	recommendation	NN
cord-027316-echxuw74	99	10	systems	system	NNS
cord-027316-echxuw74	99	11	,	,	,
cord-027316-echxuw74	99	12	and	and	CC
cord-027316-echxuw74	99	13	online	online	JJ
cord-027316-echxuw74	99	14	campaigns	campaign	NNS
cord-027316-echxuw74	100	1	Combined	Combined	NNP
cord-027316-echxuw74	100	2	Loss	Loss	NNP
cord-027316-echxuw74	100	3	Function	Function	NNP
cord-027316-echxuw74	100	4	for	for	IN
cord-027316-echxuw74	100	5	Deep	Deep	NNP
cord-027316-echxuw74	100	6	Convolutional	Convolutional	NNP
cord-027316-echxuw74	100	7	Neural	Neural	NNP
cord-027316-echxuw74	100	8	Networks	Networks	NNPS
cord-027316-echxuw74	101	1	A	a	DT
cord-027316-echxuw74	101	2	Randomized	randomized	JJ
cord-027316-echxuw74	101	3	Algorithm	algorithm	NN
cord-027316-echxuw74	101	4	for	for	IN
cord-027316-echxuw74	101	5	the	the	DT
cord-027316-echxuw74	101	6	Selection	Selection	NNP
cord-027316-echxuw74	101	7	of	of	IN
cord-027316-echxuw74	101	8	Regularization	Regularization	NNP
cord-027316-echxuw74	101	9	Parameter	Parameter	NNP
cord-027316-echxuw74	101	10	.	.	.
cord-027316-echxuw74	102	1	Inverse	Inverse	NNP
cord-027316-echxuw74	102	2	Problem	Problem	NNP
cord-027316-echxuw74	102	3	Symposium	Symposium	NNP
cord-027316-echxuw74	103	1	A	a	DT
cord-027316-echxuw74	103	2	local	local	JJ
cord-027316-echxuw74	103	3	regularization	regularization	NN
cord-027316-echxuw74	103	4	method	method	NN
cord-027316-echxuw74	103	5	using	use	VBG
cord-027316-echxuw74	103	6	multiple	multiple	JJ
cord-027316-echxuw74	103	7	regularization	regularization	NN
cord-027316-echxuw74	103	8	levels	level	NNS
cord-027316-echxuw74	103	9	A	a	DT
cord-027316-echxuw74	103	10	decomposable	decomposable	JJ
cord-027316-echxuw74	103	11	attention	attention	NN
cord-027316-echxuw74	103	12	model	model	NN
cord-027316-echxuw74	104	1	On	on	IN
cord-027316-echxuw74	104	2	the	the	DT
cord-027316-echxuw74	104	3	difficulty	difficulty	NN
cord-027316-echxuw74	104	4	of	of	IN
cord-027316-echxuw74	104	5	training	train	VBG
cord-027316-echxuw74	104	6	recurrent	recurrent	JJ
cord-027316-echxuw74	104	7	neural	neural	JJ
cord-027316-echxuw74	104	8	networks	network	NNS
cord-027316-echxuw74	104	9	.	.	.
cord-027316-echxuw74	105	1	In	in	IN
cord-027316-echxuw74	105	2	:	:	:
cord-027316-echxuw74	105	3	ICML	ICML	NNP
cord-027316-echxuw74	106	1	On	on	IN
cord-027316-echxuw74	106	2	the	the	DT
cord-027316-echxuw74	106	3	difficulty	difficulty	NN
cord-027316-echxuw74	106	4	of	of	IN
cord-027316-echxuw74	106	5	training	train	VBG
cord-027316-echxuw74	106	6	recurrent	recurrent	JJ
cord-027316-echxuw74	106	7	neural	neural	JJ
cord-027316-echxuw74	106	8	networks	network	NNS
cord-027316-echxuw74	106	9	How	how	WRB
cord-027316-echxuw74	106	10	to	to	TO
cord-027316-echxuw74	106	11	construct	construct	VB
cord-027316-echxuw74	106	12	deep	deep	JJ
cord-027316-echxuw74	106	13	recurrent	recurrent	JJ
cord-027316-echxuw74	106	14	neural	neural	JJ
cord-027316-echxuw74	106	15	networks	network	NNS
cord-027316-echxuw74	106	16	Fast	fast	JJ
cord-027316-echxuw74	106	17	curvature	curvature	NN
cord-027316-echxuw74	106	18	matrix	matrix	NN
cord-027316-echxuw74	106	19	-	-	HYPH
cord-027316-echxuw74	106	20	vector	vector	NN
cord-027316-echxuw74	106	21	products	product	NNS
cord-027316-echxuw74	106	22	for	for	IN
cord-027316-echxuw74	106	23	second	second	JJ
cord-027316-echxuw74	106	24	-	-	HYPH
cord-027316-echxuw74	106	25	order	order	NN
cord-027316-echxuw74	106	26	gradient	gradient	NN
cord-027316-echxuw74	106	27	descent	descent	NN
cord-027316-echxuw74	106	28	Bidirectional	bidirectional	JJ
cord-027316-echxuw74	106	29	recurrent	recurrent	JJ
cord-027316-echxuw74	106	30	neural	neural	JJ
cord-027316-echxuw74	106	31	networks	network	NNS
cord-027316-echxuw74	107	1	Continuous	continuous	JJ
cord-027316-echxuw74	107	2	space	space	NN
cord-027316-echxuw74	107	3	translation	translation	NN
cord-027316-echxuw74	107	4	models	model	NNS
cord-027316-echxuw74	107	5	for	for	IN
cord-027316-echxuw74	107	6	phrase	phrase	NN
cord-027316-echxuw74	107	7	-	-	HYPH
cord-027316-echxuw74	107	8	based	base	VBN
cord-027316-echxuw74	107	9	statistical	statistical	JJ
cord-027316-echxuw74	107	10	machine	machine	NN
cord-027316-echxuw74	107	11	translation	translation	NN
cord-027316-echxuw74	107	12	Continuous	continuous	JJ
cord-027316-echxuw74	107	13	space	space	NN
cord-027316-echxuw74	107	14	language	language	NN
cord-027316-echxuw74	107	15	models	model	NNS
cord-027316-echxuw74	107	16	for	for	IN
cord-027316-echxuw74	107	17	statistical	statistical	JJ
cord-027316-echxuw74	107	18	machine	machine	NN
cord-027316-echxuw74	107	19	translation	translation	NN
cord-027316-echxuw74	107	20	Sequence	sequence	NN
cord-027316-echxuw74	107	21	to	to	TO
cord-027316-echxuw74	107	22	sequence	sequence	NN
cord-027316-echxuw74	107	23	learning	learn	VBG
cord-027316-echxuw74	107	24	with	with	IN
cord-027316-echxuw74	107	25	neural	neural	JJ
cord-027316-echxuw74	107	26	networks	network	NNS
cord-027316-echxuw74	107	27	Google	Google	NNP
cord-027316-echxuw74	107	28	's	's	POS
cord-027316-echxuw74	107	29	neural	neural	JJ
cord-027316-echxuw74	107	30	machine	machine	NN
cord-027316-echxuw74	107	31	translation	translation	NN
cord-027316-echxuw74	107	32	system	system	NN
cord-027316-echxuw74	107	33	:	:	:
cord-027316-echxuw74	107	34	bridging	bridge	VBG
cord-027316-echxuw74	107	35	the	the	DT
cord-027316-echxuw74	107	36	gap	gap	NN
cord-027316-echxuw74	107	37	between	between	IN
cord-027316-echxuw74	107	38	human	human	NN
cord-027316-echxuw74	107	39	and	and	CC
cord-027316-echxuw74	107	40	machine	machine	NN
cord-027316-echxuw74	107	41	translation	translation	NN
cord-027316-echxuw74	107	42	ADADELTA	ADADELTA	NNP
cord-027316-echxuw74	107	43	:	:	:
cord-027316-echxuw74	107	44	an	an	DT
cord-027316-echxuw74	107	45	adaptive	adaptive	JJ
cord-027316-echxuw74	107	46	learning	learning	NN
cord-027316-echxuw74	107	47	rate	rate	NN
cord-027316-echxuw74	107	48	method	method	NN