id	sid	tid	token	lemma	pos
xd07gq7170m	1	1	neural	neural	ADJ
xd07gq7170m	1	2	machine	machine	NOUN
xd07gq7170m	1	3	translation	translation	NOUN
xd07gq7170m	1	4	(	(	PUNCT
xd07gq7170m	1	5	nmt	nmt	PROPN
xd07gq7170m	1	6	)	)	PUNCT
xd07gq7170m	1	7	,	,	PUNCT
xd07gq7170m	1	8	which	which	PRON
xd07gq7170m	1	9	exploits	exploit	VERB
xd07gq7170m	1	10	the	the	DET
xd07gq7170m	1	11	power	power	NOUN
xd07gq7170m	1	12	of	of	ADP
xd07gq7170m	1	13	continuous	continuous	ADJ
xd07gq7170m	1	14	representation	representation	NOUN
xd07gq7170m	1	15	and	and	CCONJ
xd07gq7170m	1	16	neural	neural	ADJ
xd07gq7170m	1	17	networks	network	NOUN
xd07gq7170m	1	18	,	,	PUNCT
xd07gq7170m	1	19	has	have	AUX
xd07gq7170m	1	20	become	become	VERB
xd07gq7170m	1	21	the	the	DET
xd07gq7170m	1	22	de	de	X
xd07gq7170m	1	23	facto	facto	X
xd07gq7170m	1	24	standard	standard	ADJ
xd07gq7170m	1	25	choice	choice	NOUN
xd07gq7170m	1	26	in	in	ADP
xd07gq7170m	1	27	both	both	CCONJ
xd07gq7170m	1	28	academic	academic	ADJ
xd07gq7170m	1	29	research	research	NOUN
xd07gq7170m	1	30	and	and	CCONJ
xd07gq7170m	1	31	industry	industry	NOUN
xd07gq7170m	1	32	.	.	PUNCT
xd07gq7170m	2	1	while	while	SCONJ
xd07gq7170m	2	2	the	the	DET
xd07gq7170m	2	3	early	early	ADJ
xd07gq7170m	2	4	non	non	ADJ
xd07gq7170m	2	5	-	-	ADJ
xd07gq7170m	2	6	attentional	attentional	ADJ
xd07gq7170m	2	7	,	,	PUNCT
xd07gq7170m	2	8	recurrent	recurrent	ADJ
xd07gq7170m	2	9	neural	neural	ADJ
xd07gq7170m	2	10	network	network	NOUN
xd07gq7170m	2	11	-	-	PUNCT
xd07gq7170m	2	12	based	base	VERB
xd07gq7170m	2	13	nmt	nmt	PROPN
xd07gq7170m	2	14	already	already	ADV
xd07gq7170m	2	15	showed	show	VERB
xd07gq7170m	2	16	impressive	impressive	ADJ
xd07gq7170m	2	17	performance	performance	NOUN
xd07gq7170m	2	18	,	,	PUNCT
xd07gq7170m	2	19	with	with	ADP
xd07gq7170m	2	20	the	the	DET
xd07gq7170m	2	21	invention	invention	NOUN
xd07gq7170m	2	22	of	of	ADP
xd07gq7170m	2	23	the	the	DET
xd07gq7170m	2	24	attention	attention	NOUN
xd07gq7170m	2	25	mechanism	mechanism	NOUN
xd07gq7170m	2	26	,	,	PUNCT
xd07gq7170m	2	27	it	it	PRON
xd07gq7170m	2	28	has	have	AUX
xd07gq7170m	2	29	been	be	AUX
xd07gq7170m	2	30	pushed	push	VERB
xd07gq7170m	2	31	further	far	ADV
xd07gq7170m	2	32	and	and	CCONJ
xd07gq7170m	2	33	achieved	achieve	VERB
xd07gq7170m	2	34	state	state	NOUN
xd07gq7170m	2	35	-	-	PUNCT
xd07gq7170m	2	36	of	of	ADP
xd07gq7170m	2	37	-	-	PUNCT
xd07gq7170m	2	38	the	the	DET
xd07gq7170m	2	39	-	-	PUNCT
xd07gq7170m	2	40	art	art	NOUN
xd07gq7170m	2	41	performance	performance	NOUN
xd07gq7170m	2	42	.	.	PUNCT
xd07gq7170m	3	1	however	however	ADV
xd07gq7170m	3	2	,	,	PUNCT
xd07gq7170m	3	3	like	like	ADP
xd07gq7170m	3	4	other	other	ADJ
xd07gq7170m	3	5	neural	neural	ADJ
xd07gq7170m	3	6	networks	network	NOUN
xd07gq7170m	3	7	,	,	PUNCT
xd07gq7170m	3	8	these	these	DET
xd07gq7170m	3	9	nmt	nmt	PROPN
xd07gq7170m	3	10	systems	system	NOUN
xd07gq7170m	3	11	are	be	AUX
xd07gq7170m	3	12	often	often	ADV
xd07gq7170m	3	13	data	data	NOUN
xd07gq7170m	3	14	-	-	PUNCT
xd07gq7170m	3	15	hungry	hungry	ADJ
xd07gq7170m	3	16	and	and	CCONJ
xd07gq7170m	3	17	require	require	VERB
xd07gq7170m	3	18	millions	million	NOUN
xd07gq7170m	3	19	of	of	ADP
xd07gq7170m	3	20	training	training	NOUN
xd07gq7170m	3	21	examples	example	NOUN
xd07gq7170m	3	22	to	to	PART
xd07gq7170m	3	23	achieve	achieve	VERB
xd07gq7170m	3	24	competitive	competitive	ADJ
xd07gq7170m	3	25	results.unfortunately	results.unfortunately	ADV
xd07gq7170m	3	26	,	,	PUNCT
xd07gq7170m	3	27	only	only	ADV
xd07gq7170m	3	28	a	a	DET
xd07gq7170m	3	29	few	few	ADJ
xd07gq7170m	3	30	language	language	NOUN
xd07gq7170m	3	31	pairs	pair	NOUN
xd07gq7170m	3	32	have	have	VERB
xd07gq7170m	3	33	this	this	DET
xd07gq7170m	3	34	privilege	privilege	NOUN
xd07gq7170m	3	35	.	.	PUNCT
xd07gq7170m	4	1	most	most	ADJ
xd07gq7170m	4	2	fall	fall	VERB
xd07gq7170m	4	3	into	into	ADP
xd07gq7170m	4	4	the	the	DET
xd07gq7170m	4	5	low	low	ADJ
xd07gq7170m	4	6	-	-	PUNCT
xd07gq7170m	4	7	resource	resource	NOUN
xd07gq7170m	4	8	domain	domain	NOUN
xd07gq7170m	4	9	and	and	CCONJ
xd07gq7170m	4	10	often	often	ADV
xd07gq7170m	4	11	have	have	VERB
xd07gq7170m	4	12	only	only	ADV
xd07gq7170m	4	13	around	around	ADP
xd07gq7170m	4	14	100k	100k	NUM
xd07gq7170m	4	15	training	training	NOUN
xd07gq7170m	4	16	sentence	sentence	NOUN
xd07gq7170m	4	17	pairs	pair	NOUN
xd07gq7170m	4	18	or	or	CCONJ
xd07gq7170m	4	19	less	less	ADJ
xd07gq7170m	4	20	.	.	PUNCT
xd07gq7170m	5	1	some	some	DET
xd07gq7170m	5	2	examples	example	NOUN
xd07gq7170m	5	3	could	could	AUX
xd07gq7170m	5	4	be	be	AUX
xd07gq7170m	5	5	the	the	DET
xd07gq7170m	5	6	lorelei	lorelei	NOUN
xd07gq7170m	5	7	or	or	CCONJ
xd07gq7170m	5	8	iwslt	iwslt	PROPN
xd07gq7170m	5	9	datasets	dataset	NOUN
xd07gq7170m	5	10	.	.	PUNCT
xd07gq7170m	6	1	this	this	DET
xd07gq7170m	6	2	lack	lack	NOUN
xd07gq7170m	6	3	of	of	ADP
xd07gq7170m	6	4	data	datum	NOUN
xd07gq7170m	6	5	is	be	AUX
xd07gq7170m	6	6	particularly	particularly	ADV
xd07gq7170m	6	7	critical	critical	ADJ
xd07gq7170m	6	8	for	for	ADP
xd07gq7170m	6	9	the	the	DET
xd07gq7170m	6	10	early	early	ADJ
xd07gq7170m	6	11	recurrent	recurrent	ADJ
xd07gq7170m	6	12	neural	neural	ADJ
xd07gq7170m	6	13	network	network	NOUN
xd07gq7170m	6	14	-	-	PUNCT
xd07gq7170m	6	15	based	base	VERB
xd07gq7170m	6	16	nmt	nmt	PROPN
xd07gq7170m	6	17	systems	system	NOUN
xd07gq7170m	6	18	,	,	PUNCT
xd07gq7170m	6	19	which	which	PRON
xd07gq7170m	6	20	are	be	AUX
xd07gq7170m	6	21	often	often	ADV
xd07gq7170m	6	22	outperformed	outperform	VERB
xd07gq7170m	6	23	by	by	ADP
xd07gq7170m	6	24	the	the	DET
xd07gq7170m	6	25	traditional	traditional	ADJ
xd07gq7170m	6	26	phrase	phrase	NOUN
xd07gq7170m	6	27	-	-	PUNCT
xd07gq7170m	6	28	based	base	VERB
xd07gq7170m	6	29	machine	machine	NOUN
xd07gq7170m	6	30	translation	translation	NOUN
xd07gq7170m	6	31	(	(	PUNCT
xd07gq7170m	6	32	pbmt	pbmt	ADJ
xd07gq7170m	6	33	)	)	PUNCT
xd07gq7170m	6	34	ones	one	NOUN
xd07gq7170m	6	35	in	in	ADP
xd07gq7170m	6	36	low	low	ADJ
xd07gq7170m	6	37	-	-	PUNCT
xd07gq7170m	6	38	resource	resource	NOUN
xd07gq7170m	6	39	scenarios	scenario	NOUN
xd07gq7170m	6	40	.	.	PUNCT
xd07gq7170m	7	1	it	it	PRON
xd07gq7170m	7	2	also	also	ADV
xd07gq7170m	7	3	means	mean	VERB
xd07gq7170m	7	4	more	more	ADV
xd07gq7170m	7	5	advanced	advanced	ADJ
xd07gq7170m	7	6	techniques	technique	NOUN
xd07gq7170m	7	7	are	be	AUX
xd07gq7170m	7	8	required	require	VERB
xd07gq7170m	7	9	to	to	PART
xd07gq7170m	7	10	effectively	effectively	ADV
xd07gq7170m	7	11	train	train	VERB
xd07gq7170m	7	12	a	a	DET
xd07gq7170m	7	13	good	good	ADJ
xd07gq7170m	7	14	nmt	nmt	PROPN
xd07gq7170m	7	15	system	system	NOUN
xd07gq7170m	7	16	for	for	ADP
xd07gq7170m	7	17	low	low	ADJ
xd07gq7170m	7	18	-	-	PUNCT
xd07gq7170m	7	19	resource	resource	NOUN
xd07gq7170m	7	20	languages.in	languages.in	NUM
xd07gq7170m	7	21	this	this	DET
xd07gq7170m	7	22	dissertation	dissertation	NOUN
xd07gq7170m	7	23	,	,	PUNCT
xd07gq7170m	7	24	i	i	PRON
xd07gq7170m	7	25	will	will	AUX
xd07gq7170m	7	26	show	show	VERB
xd07gq7170m	7	27	that	that	SCONJ
xd07gq7170m	7	28	with	with	ADP
xd07gq7170m	7	29	better	well	ADJ
xd07gq7170m	7	30	use	use	NOUN
xd07gq7170m	7	31	of	of	ADP
xd07gq7170m	7	32	training	training	NOUN
xd07gq7170m	7	33	data	datum	NOUN
xd07gq7170m	7	34	,	,	PUNCT
xd07gq7170m	7	35	better	well	ADJ
xd07gq7170m	7	36	normalization	normalization	NOUN
xd07gq7170m	7	37	or	or	CCONJ
xd07gq7170m	7	38	improved	improved	ADJ
xd07gq7170m	7	39	model	model	NOUN
xd07gq7170m	7	40	architecture	architecture	NOUN
xd07gq7170m	7	41	,	,	PUNCT
xd07gq7170m	7	42	we	we	PRON
xd07gq7170m	7	43	can	can	AUX
xd07gq7170m	7	44	in	in	ADP
xd07gq7170m	7	45	fact	fact	NOUN
xd07gq7170m	7	46	build	build	VERB
xd07gq7170m	7	47	a	a	DET
xd07gq7170m	7	48	competitive	competitive	ADJ
xd07gq7170m	7	49	nmt	nmt	NOUN
xd07gq7170m	7	50	system	system	NOUN
xd07gq7170m	7	51	with	with	ADP
xd07gq7170m	7	52	only	only	ADJ
xd07gq7170m	7	53	limited	limited	ADJ
xd07gq7170m	7	54	data	datum	NOUN
xd07gq7170m	7	55	at	at	ADP
xd07gq7170m	7	56	hand	hand	NOUN
xd07gq7170m	7	57	.	.	PUNCT
xd07gq7170m	8	1	on	on	ADP
xd07gq7170m	8	2	data	datum	NOUN
xd07gq7170m	8	3	,	,	PUNCT
xd07gq7170m	8	4	i	i	PRON
xd07gq7170m	8	5	propose	propose	VERB
xd07gq7170m	8	6	two	two	NUM
xd07gq7170m	8	7	simple	simple	ADJ
xd07gq7170m	8	8	methods	method	NOUN
xd07gq7170m	8	9	to	to	PART
xd07gq7170m	8	10	improve	improve	VERB
xd07gq7170m	8	11	nmt	nmt	PROPN
xd07gq7170m	8	12	performance	performance	NOUN
xd07gq7170m	8	13	by	by	ADP
xd07gq7170m	8	14	better	well	ADV
xd07gq7170m	8	15	exploiting	exploit	VERB
xd07gq7170m	8	16	training	training	NOUN
xd07gq7170m	8	17	resources	resource	NOUN
xd07gq7170m	8	18	.	.	PUNCT
xd07gq7170m	9	1	the	the	DET
xd07gq7170m	9	2	first	first	ADJ
xd07gq7170m	9	3	approach	approach	NOUN
xd07gq7170m	9	4	makes	make	VERB
xd07gq7170m	9	5	use	use	NOUN
xd07gq7170m	9	6	of	of	ADP
xd07gq7170m	9	7	the	the	DET
xd07gq7170m	9	8	relationship	relationship	NOUN
xd07gq7170m	9	9	between	between	ADP
xd07gq7170m	9	10	languages	language	NOUN
xd07gq7170m	9	11	for	for	ADP
xd07gq7170m	9	12	better	well	ADJ
xd07gq7170m	9	13	transfer	transfer	NOUN
xd07gq7170m	9	14	learning	learn	VERB
xd07gq7170m	9	15	from	from	ADP
xd07gq7170m	9	16	one	one	NUM
xd07gq7170m	9	17	language	language	NOUN
xd07gq7170m	9	18	pair	pair	NOUN
xd07gq7170m	9	19	to	to	ADP
xd07gq7170m	9	20	the	the	DET
xd07gq7170m	9	21	other	other	ADJ
xd07gq7170m	9	22	.	.	PUNCT
xd07gq7170m	10	1	the	the	DET
xd07gq7170m	10	2	second	second	ADJ
xd07gq7170m	10	3	one	one	NOUN
xd07gq7170m	10	4	is	be	AUX
xd07gq7170m	10	5	a	a	DET
xd07gq7170m	10	6	simple	simple	ADJ
xd07gq7170m	10	7	data	datum	NOUN
xd07gq7170m	10	8	augmentation	augmentation	NOUN
xd07gq7170m	10	9	via	via	ADP
xd07gq7170m	10	10	concatenation	concatenation	NOUN
xd07gq7170m	10	11	which	which	PRON
xd07gq7170m	10	12	can	can	AUX
xd07gq7170m	10	13	yield	yield	VERB
xd07gq7170m	10	14	on	on	ADP
xd07gq7170m	10	15	average	average	ADJ
xd07gq7170m	10	16	+1	+1	NOUN
xd07gq7170m	10	17	bleu	bleu	NOUN
xd07gq7170m	10	18	on	on	ADP
xd07gq7170m	10	19	several	several	ADJ
xd07gq7170m	10	20	language	language	NOUN
xd07gq7170m	10	21	pairs	pair	NOUN
xd07gq7170m	10	22	.	.	PUNCT
xd07gq7170m	11	1	on	on	ADP
xd07gq7170m	11	2	normalization	normalization	NOUN
xd07gq7170m	11	3	,	,	PUNCT
xd07gq7170m	11	4	i	i	PRON
xd07gq7170m	11	5	show	show	VERB
xd07gq7170m	11	6	how	how	SCONJ
xd07gq7170m	11	7	simple	simple	ADJ
xd07gq7170m	11	8	l2	l2	NOUN
xd07gq7170m	11	9	-	-	PUNCT
xd07gq7170m	11	10	based	base	VERB
xd07gq7170m	11	11	normalization	normalization	NOUN
xd07gq7170m	11	12	at	at	ADP
xd07gq7170m	11	13	word	word	NOUN
xd07gq7170m	11	14	embedding	embedding	NOUN
xd07gq7170m	11	15	and	and	CCONJ
xd07gq7170m	11	16	hidden	hide	VERB
xd07gq7170m	11	17	state	state	NOUN
xd07gq7170m	11	18	levels	level	NOUN
xd07gq7170m	11	19	can	can	AUX
xd07gq7170m	11	20	significantly	significantly	ADV
xd07gq7170m	11	21	improve	improve	VERB
xd07gq7170m	11	22	translation	translation	NOUN
xd07gq7170m	11	23	for	for	ADP
xd07gq7170m	11	24	low	low	ADJ
xd07gq7170m	11	25	-	-	PUNCT
xd07gq7170m	11	26	resource	resource	NOUN
xd07gq7170m	11	27	languages	language	NOUN
xd07gq7170m	11	28	,	,	PUNCT
xd07gq7170m	11	29	with	with	ADP
xd07gq7170m	11	30	a	a	DET
xd07gq7170m	11	31	particular	particular	ADJ
xd07gq7170m	11	32	focus	focus	NOUN
xd07gq7170m	11	33	on	on	ADP
xd07gq7170m	11	34	rare	rare	ADJ
xd07gq7170m	11	35	words	word	NOUN
xd07gq7170m	11	36	.	.	PUNCT
xd07gq7170m	12	1	finally	finally	ADV
xd07gq7170m	12	2	,	,	PUNCT
xd07gq7170m	12	3	on	on	ADP
xd07gq7170m	12	4	model	model	NOUN
xd07gq7170m	12	5	architecture	architecture	NOUN
xd07gq7170m	12	6	,	,	PUNCT
xd07gq7170m	12	7	i	i	PRON
xd07gq7170m	12	8	investigate	investigate	VERB
xd07gq7170m	12	9	three	three	NUM
xd07gq7170m	12	10	simple	simple	ADJ
xd07gq7170m	12	11	yet	yet	ADV
xd07gq7170m	12	12	effective	effective	ADJ
xd07gq7170m	12	13	changes	change	NOUN
xd07gq7170m	12	14	.	.	PUNCT
xd07gq7170m	13	1	first	first	ADV
xd07gq7170m	13	2	,	,	PUNCT
xd07gq7170m	13	3	i	i	PRON
xd07gq7170m	13	4	propose	propose	VERB
xd07gq7170m	13	5	a	a	DET
xd07gq7170m	13	6	simple	simple	ADJ
xd07gq7170m	13	7	lexical	lexical	ADJ
xd07gq7170m	13	8	module	module	NOUN
xd07gq7170m	13	9	which	which	PRON
xd07gq7170m	13	10	can	can	AUX
xd07gq7170m	13	11	alleviate	alleviate	VERB
xd07gq7170m	13	12	the	the	DET
xd07gq7170m	13	13	mistranslation	mistranslation	NOUN
xd07gq7170m	13	14	issue	issue	NOUN
xd07gq7170m	13	15	for	for	ADP
xd07gq7170m	13	16	rare	rare	ADJ
xd07gq7170m	13	17	words	word	NOUN
xd07gq7170m	13	18	.	.	PUNCT
xd07gq7170m	14	1	second	second	ADV
xd07gq7170m	14	2	,	,	PUNCT
xd07gq7170m	14	3	i	i	PRON
xd07gq7170m	14	4	study	study	VERB
xd07gq7170m	14	5	the	the	DET
xd07gq7170m	14	6	transformer	transformer	NOUN
xd07gq7170m	14	7	model	model	NOUN
xd07gq7170m	14	8	and	and	CCONJ
xd07gq7170m	14	9	show	show	VERB
xd07gq7170m	14	10	how	how	SCONJ
xd07gq7170m	14	11	a	a	DET
xd07gq7170m	14	12	simple	simple	ADJ
xd07gq7170m	14	13	rearrangement	rearrangement	NOUN
xd07gq7170m	14	14	of	of	ADP
xd07gq7170m	14	15	its	its	PRON
xd07gq7170m	14	16	components	component	NOUN
xd07gq7170m	14	17	can	can	AUX
xd07gq7170m	14	18	improve	improve	VERB
xd07gq7170m	14	19	both	both	DET
xd07gq7170m	14	20	training	training	NOUN
xd07gq7170m	14	21	and	and	CCONJ
xd07gq7170m	14	22	performance	performance	NOUN
xd07gq7170m	14	23	for	for	ADP
xd07gq7170m	14	24	low	low	ADJ
xd07gq7170m	14	25	-	-	PUNCT
xd07gq7170m	14	26	resource	resource	NOUN
xd07gq7170m	14	27	languages	language	NOUN
xd07gq7170m	14	28	.	.	PUNCT
xd07gq7170m	15	1	lastly	lastly	ADV
xd07gq7170m	15	2	,	,	PUNCT
xd07gq7170m	15	3	i	i	PRON
xd07gq7170m	15	4	explore	explore	VERB
xd07gq7170m	15	5	the	the	DET
xd07gq7170m	15	6	untied	untied	ADJ
xd07gq7170m	15	7	positional	positional	ADJ
xd07gq7170m	15	8	attention	attention	NOUN
xd07gq7170m	15	9	in	in	ADP
xd07gq7170m	15	10	transformer	transformer	NOUN
xd07gq7170m	15	11	and	and	CCONJ
xd07gq7170m	15	12	demonstrate	demonstrate	VERB
xd07gq7170m	15	13	how	how	SCONJ
xd07gq7170m	15	14	it	it	PRON
xd07gq7170m	15	15	can	can	AUX
xd07gq7170m	15	16	improve	improve	VERB
xd07gq7170m	15	17	both	both	DET
xd07gq7170m	15	18	performance	performance	NOUN
xd07gq7170m	15	19	and	and	CCONJ
xd07gq7170m	15	20	interpretability	interpretability	NOUN
xd07gq7170m	15	21	.	.	PUNCT