id	sid	tid	token	lemma	pos
bracis-28417	1	1	sabiá	sabiá	PROPN
bracis-28417	1	2	:	:	PUNCT
bracis-28417	2	1	portuguese	portuguese	ADJ
bracis-28417	2	2	large	large	ADJ
bracis-28417	2	3	language	language	NOUN
bracis-28417	2	4	models	model	NOUN
bracis-28417	2	5	|	|	ADV
bracis-28417	2	6	springer	springer	NOUN
bracis-28417	2	7	nature	nature	NOUN
bracis-28417	2	8	link	link	PROPN
bracis-28417	2	9	(	(	PUNCT
bracis-28417	2	10	formerly	formerly	ADV
bracis-28417	2	11	springerlink	springerlink	NOUN
bracis-28417	2	12	)	)	PUNCT
bracis-28417	2	13	skip	skip	VERB
bracis-28417	2	14	to	to	ADP
bracis-28417	2	15	main	main	ADJ
bracis-28417	2	16	content	content	NOUN
bracis-28417	2	17	advertisement	advertisement	NOUN
bracis-28417	2	18	log	log	NOUN
bracis-28417	2	19	in	in	ADP
bracis-28417	2	20	menu	menu	NOUN
bracis-28417	2	21	find	find	VERB
bracis-28417	2	22	a	a	DET
bracis-28417	2	23	journal	journal	NOUN
bracis-28417	2	24	publish	publish	VERB
bracis-28417	2	25	with	with	ADP
bracis-28417	2	26	us	we	PRON
bracis-28417	2	27	track	track	VERB
bracis-28417	2	28	your	your	PRON
bracis-28417	2	29	research	research	NOUN
bracis-28417	2	30	search	search	NOUN
bracis-28417	2	31	cart	cart	NOUN
bracis-28417	2	32	home	home	NOUN
bracis-28417	2	33	intelligent	intelligent	ADJ
bracis-28417	2	34	systems	system	NOUN
bracis-28417	2	35	conference	conference	PROPN
bracis-28417	2	36	paper	paper	NOUN
bracis-28417	2	37	sabiá	sabiá	NOUN
bracis-28417	2	38	:	:	PUNCT
bracis-28417	3	1	portuguese	portuguese	ADJ
bracis-28417	3	2	large	large	ADJ
bracis-28417	3	3	language	language	NOUN
bracis-28417	3	4	models	model	NOUN
bracis-28417	3	5	conference	conference	NOUN
bracis-28417	3	6	paper	paper	NOUN
bracis-28417	3	7	first	first	ADV
bracis-28417	3	8	online	online	ADV
bracis-28417	3	9	:	:	PUNCT
bracis-28417	3	10	12	12	NUM
bracis-28417	3	11	october	october	NOUN
bracis-28417	3	12	2023	2023	NUM
bracis-28417	3	13	pp	pp	ADP
bracis-28417	3	14	226–240	226–240	NUM
bracis-28417	3	15	cite	cite	VERB
bracis-28417	3	16	this	this	DET
bracis-28417	3	17	conference	conference	NOUN
bracis-28417	3	18	paper	paper	NOUN
bracis-28417	3	19	access	access	NOUN
bracis-28417	3	20	provided	provide	VERB
bracis-28417	3	21	by	by	ADP
bracis-28417	3	22	university	university	PROPN
bracis-28417	3	23	of	of	ADP
bracis-28417	3	24	notre	notre	PROPN
bracis-28417	3	25	dame	dame	PROPN
bracis-28417	3	26	hesburgh	hesburgh	PROPN
bracis-28417	3	27	library	library	PROPN
bracis-28417	3	28	download	download	PROPN
bracis-28417	3	29	book	book	NOUN
bracis-28417	3	30	pdf	pdf	PROPN
bracis-28417	3	31	download	download	NOUN
bracis-28417	3	32	book	book	NOUN
bracis-28417	3	33	epub	epub	PROPN
bracis-28417	3	34	intelligent	intelligent	ADJ
bracis-28417	3	35	systems	system	NOUN
bracis-28417	3	36	(	(	PUNCT
bracis-28417	3	37	bracis	bracis	NOUN
bracis-28417	3	38	2023	2023	NUM
bracis-28417	3	39	)	)	PUNCT
bracis-28417	3	40	sabiá	sabiá	NOUN
bracis-28417	3	41	:	:	PUNCT
bracis-28417	3	42	portuguese	portuguese	ADJ
bracis-28417	3	43	large	large	ADJ
bracis-28417	3	44	language	language	NOUN
bracis-28417	3	45	models	model	NOUN
bracis-28417	3	46	download	download	VERB
bracis-28417	3	47	book	book	NOUN
bracis-28417	3	48	pdf	pdf	PROPN
bracis-28417	3	49	download	download	NOUN
bracis-28417	3	50	book	book	PROPN
bracis-28417	3	51	epub	epub	PROPN
bracis-28417	3	52	ramon	ramon	NOUN
bracis-28417	3	53	pires	pire	VERB
bracis-28417	3	54	  	  	SPACE
bracis-28417	3	55	orcid	orcid	NOUN
bracis-28417	3	56	:	:	PUNCT
bracis-28417	3	57	orcid.org/0000-0002-0023-19719	orcid.org/0000-0002-0023-19719	PROPN
bracis-28417	3	58	,	,	PUNCT
bracis-28417	3	59	hugo	hugo	PROPN
bracis-28417	3	60	abonizio	abonizio	PROPN
bracis-28417	3	61	  	  	SPACE
bracis-28417	3	62	orcid	orcid	NOUN
bracis-28417	3	63	:	:	PUNCT
bracis-28417	3	64	orcid.org/0000-0001-5208-02909	orcid.org/0000-0001-5208-02909	ADJ
bracis-28417	3	65	,	,	PUNCT
bracis-28417	3	66	thales	thales	PROPN
bracis-28417	3	67	sales	sales	PROPN
bracis-28417	3	68	almeida	almeida	PROPN
bracis-28417	3	69	  	  	SPACE
bracis-28417	3	70	orcid	orcid	PROPN
bracis-28417	3	71	:	:	PUNCT
bracis-28417	3	72	orcid.org/0009-0006-9568-93319	orcid.org/0009-0006-9568-93319	PROPN
bracis-28417	3	73	&	&	CCONJ
bracis-28417	3	74	…	…	PUNCT
bracis-28417	3	75	rodrigo	rodrigo	PROPN
bracis-28417	3	76	nogueira	nogueira	PROPN
bracis-28417	3	77	  	  	SPACE
bracis-28417	3	78	orcid	orcid	PROPN
bracis-28417	3	79	:	:	PUNCT
bracis-28417	3	80	orcid.org/0000-0002-2600-60359	orcid.org/0000-0002-2600-60359	NOUN
bracis-28417	3	81	  	  	SPACE
bracis-28417	3	82	show	show	VERB
bracis-28417	3	83	authors	author	NOUN
bracis-28417	3	84	part	part	NOUN
bracis-28417	3	85	of	of	ADP
bracis-28417	3	86	the	the	DET
bracis-28417	3	87	book	book	NOUN
bracis-28417	3	88	series	series	NOUN
bracis-28417	3	89	:	:	PUNCT
bracis-28417	3	90	lecture	lecture	NOUN
bracis-28417	3	91	notes	note	NOUN
bracis-28417	3	92	in	in	ADP
bracis-28417	3	93	computer	computer	NOUN
bracis-28417	3	94	science	science	NOUN
bracis-28417	3	95	(	(	PUNCT
bracis-28417	3	96	(	(	PUNCT
bracis-28417	3	97	lnai	lnai	ADJ
bracis-28417	3	98	,	,	PUNCT
bracis-28417	3	99	volume	volume	NOUN
bracis-28417	3	100	14197	14197	NUM
bracis-28417	3	101	)	)	PUNCT
bracis-28417	3	102	)	)	PUNCT
bracis-28417	3	103	included	include	VERB
bracis-28417	3	104	in	in	ADP
bracis-28417	3	105	the	the	DET
bracis-28417	3	106	following	follow	VERB
bracis-28417	3	107	conference	conference	NOUN
bracis-28417	3	108	series	series	NOUN
bracis-28417	3	109	:	:	PUNCT
bracis-28417	3	110	brazilian	brazilian	ADJ
bracis-28417	3	111	conference	conference	NOUN
bracis-28417	3	112	on	on	ADP
bracis-28417	3	113	intelligent	intelligent	ADJ
bracis-28417	3	114	systems	system	NOUN
bracis-28417	3	115	724	724	NUM
bracis-28417	3	116	accesses	access	VERB
bracis-28417	3	117	59	59	NUM
bracis-28417	3	118	citations	citation	NOUN
bracis-28417	3	119	109	109	NUM
bracis-28417	3	120	altmetric	altmetric	ADJ
bracis-28417	3	121	abstract	abstract	NOUN
bracis-28417	3	122	as	as	SCONJ
bracis-28417	3	123	the	the	DET
bracis-28417	3	124	capabilities	capability	NOUN
bracis-28417	3	125	of	of	ADP
bracis-28417	3	126	language	language	NOUN
bracis-28417	3	127	models	model	NOUN
bracis-28417	3	128	continue	continue	VERB
bracis-28417	3	129	to	to	PART
bracis-28417	3	130	advance	advance	VERB
bracis-28417	3	131	,	,	PUNCT
bracis-28417	3	132	it	it	PRON
bracis-28417	3	133	is	be	AUX
bracis-28417	3	134	conceivable	conceivable	ADJ
bracis-28417	3	135	that	that	SCONJ
bracis-28417	3	136	“	"	PUNCT
bracis-28417	3	137	one	one	NUM
bracis-28417	3	138	-	-	PUNCT
bracis-28417	3	139	size	size	NOUN
bracis-28417	3	140	-	-	PUNCT
bracis-28417	3	141	fits	fit	VERB
bracis-28417	3	142	-	-	PUNCT
bracis-28417	3	143	all	all	PRON
bracis-28417	3	144	”	"	PUNCT
bracis-28417	3	145	model	model	NOUN
bracis-28417	3	146	will	will	AUX
bracis-28417	3	147	remain	remain	VERB
bracis-28417	3	148	as	as	ADP
bracis-28417	3	149	the	the	DET
bracis-28417	3	150	main	main	ADJ
bracis-28417	3	151	paradigm	paradigm	NOUN
bracis-28417	3	152	.	.	PUNCT
bracis-28417	4	1	for	for	ADP
bracis-28417	4	2	instance	instance	NOUN
bracis-28417	4	3	,	,	PUNCT
bracis-28417	4	4	given	give	VERB
bracis-28417	4	5	the	the	DET
bracis-28417	4	6	vast	vast	ADJ
bracis-28417	4	7	number	number	NOUN
bracis-28417	4	8	of	of	ADP
bracis-28417	4	9	languages	language	NOUN
bracis-28417	4	10	worldwide	worldwide	ADV
bracis-28417	4	11	,	,	PUNCT
bracis-28417	4	12	many	many	ADJ
bracis-28417	4	13	of	of	ADP
bracis-28417	4	14	which	which	PRON
bracis-28417	4	15	are	be	AUX
bracis-28417	4	16	low	low	ADJ
bracis-28417	4	17	-	-	PUNCT
bracis-28417	4	18	resource	resource	NOUN
bracis-28417	4	19	,	,	PUNCT
bracis-28417	4	20	the	the	DET
bracis-28417	4	21	prevalent	prevalent	ADJ
bracis-28417	4	22	practice	practice	NOUN
bracis-28417	4	23	is	be	AUX
bracis-28417	4	24	to	to	PART
bracis-28417	4	25	pretrain	pretrain	VERB
bracis-28417	4	26	a	a	DET
bracis-28417	4	27	single	single	ADJ
bracis-28417	4	28	model	model	NOUN
bracis-28417	4	29	on	on	ADP
bracis-28417	4	30	multiple	multiple	ADJ
bracis-28417	4	31	languages	language	NOUN
bracis-28417	4	32	.	.	PUNCT
bracis-28417	5	1	in	in	ADP
bracis-28417	5	2	this	this	DET
bracis-28417	5	3	paper	paper	NOUN
bracis-28417	5	4	,	,	PUNCT
bracis-28417	5	5	we	we	PRON
bracis-28417	5	6	add	add	VERB
bracis-28417	5	7	to	to	ADP
bracis-28417	5	8	the	the	DET
bracis-28417	5	9	growing	grow	VERB
bracis-28417	5	10	body	body	NOUN
bracis-28417	5	11	of	of	ADP
bracis-28417	5	12	evidence	evidence	NOUN
bracis-28417	5	13	that	that	PRON
bracis-28417	5	14	challenges	challenge	VERB
bracis-28417	5	15	this	this	DET
bracis-28417	5	16	practice	practice	NOUN
bracis-28417	5	17	,	,	PUNCT
bracis-28417	5	18	demonstrating	demonstrate	VERB
bracis-28417	5	19	that	that	SCONJ
bracis-28417	5	20	monolingual	monolingual	ADJ
bracis-28417	5	21	pretraining	pretraine	VERB
bracis-28417	5	22	on	on	ADP
bracis-28417	5	23	the	the	DET
bracis-28417	5	24	target	target	NOUN
bracis-28417	5	25	language	language	NOUN
bracis-28417	5	26	significantly	significantly	ADV
bracis-28417	5	27	improves	improve	VERB
bracis-28417	5	28	models	model	NOUN
bracis-28417	5	29	already	already	ADV
bracis-28417	5	30	extensively	extensively	ADV
bracis-28417	5	31	trained	train	VERB
bracis-28417	5	32	on	on	ADP
bracis-28417	5	33	diverse	diverse	ADJ
bracis-28417	5	34	corpora	corpus	NOUN
bracis-28417	5	35	.	.	PUNCT
bracis-28417	6	1	more	more	ADV
bracis-28417	6	2	specifically	specifically	ADV
bracis-28417	6	3	,	,	PUNCT
bracis-28417	6	4	we	we	PRON
bracis-28417	6	5	further	far	ADV
bracis-28417	6	6	pretrain	pretrain	NOUN
bracis-28417	6	7	gpt	gpt	PROPN
bracis-28417	6	8	-	-	PUNCT
bracis-28417	6	9	j	j	PROPN
bracis-28417	6	10	and	and	CCONJ
bracis-28417	6	11	llama	llama	NOUN
bracis-28417	6	12	models	model	NOUN
bracis-28417	6	13	on	on	ADP
bracis-28417	6	14	portuguese	portuguese	ADJ
bracis-28417	6	15	texts	text	NOUN
bracis-28417	6	16	using	use	VERB
bracis-28417	6	17	3	3	NUM
bracis-28417	6	18	%	%	NOUN
bracis-28417	6	19	or	or	CCONJ
bracis-28417	6	20	less	less	ADJ
bracis-28417	6	21	of	of	ADP
bracis-28417	6	22	their	their	PRON
bracis-28417	6	23	original	original	ADJ
bracis-28417	6	24	pretraining	pretraine	VERB
bracis-28417	6	25	budget	budget	NOUN
bracis-28417	6	26	.	.	PUNCT
bracis-28417	7	1	few	few	ADJ
bracis-28417	7	2	-	-	PUNCT
bracis-28417	7	3	shot	shot	NOUN
bracis-28417	7	4	evaluations	evaluation	NOUN
bracis-28417	7	5	on	on	ADP
bracis-28417	7	6	poeta	poeta	PROPN
bracis-28417	7	7	,	,	PUNCT
bracis-28417	7	8	a	a	DET
bracis-28417	7	9	suite	suite	NOUN
bracis-28417	7	10	of	of	ADP
bracis-28417	7	11	14	14	NUM
bracis-28417	7	12	portuguese	portuguese	ADJ
bracis-28417	7	13	datasets	dataset	NOUN
bracis-28417	7	14	,	,	PUNCT
bracis-28417	7	15	reveal	reveal	VERB
bracis-28417	7	16	that	that	SCONJ
bracis-28417	7	17	our	our	PRON
bracis-28417	7	18	models	model	NOUN
bracis-28417	7	19	outperform	outperform	VERB
bracis-28417	7	20	english	english	ADJ
bracis-28417	7	21	-	-	PUNCT
bracis-28417	7	22	centric	centric	ADJ
bracis-28417	7	23	and	and	CCONJ
bracis-28417	7	24	multilingual	multilingual	ADJ
bracis-28417	7	25	counterparts	counterpart	NOUN
bracis-28417	7	26	by	by	ADP
bracis-28417	7	27	a	a	DET
bracis-28417	7	28	significant	significant	ADJ
bracis-28417	7	29	margin	margin	NOUN
bracis-28417	7	30	.	.	PUNCT
bracis-28417	8	1	our	our	PRON
bracis-28417	8	2	best	good	ADJ
bracis-28417	8	3	model	model	NOUN
bracis-28417	8	4	,	,	PUNCT
bracis-28417	8	5	sabiá-65b	sabiá-65b	PROPN
bracis-28417	8	6	,	,	PUNCT
bracis-28417	8	7	performs	perform	VERB
bracis-28417	8	8	on	on	ADP
bracis-28417	8	9	par	par	NOUN
bracis-28417	8	10	with	with	ADP
bracis-28417	8	11	gpt-3.5	gpt-3.5	NOUN
bracis-28417	8	12	-	-	PUNCT
bracis-28417	8	13	turbo	turbo	NOUN
bracis-28417	8	14	.	.	PUNCT
bracis-28417	9	1	by	by	ADP
bracis-28417	9	2	evaluating	evaluate	VERB
bracis-28417	9	3	on	on	ADP
bracis-28417	9	4	datasets	dataset	NOUN
bracis-28417	9	5	originally	originally	ADV
bracis-28417	9	6	conceived	conceive	VERB
bracis-28417	9	7	in	in	ADP
bracis-28417	9	8	the	the	DET
bracis-28417	9	9	target	target	NOUN
bracis-28417	9	10	language	language	NOUN
bracis-28417	9	11	as	as	ADV
bracis-28417	9	12	well	well	ADV
bracis-28417	9	13	as	as	ADP
bracis-28417	9	14	translated	translate	VERB
bracis-28417	9	15	ones	one	NOUN
bracis-28417	9	16	,	,	PUNCT
bracis-28417	9	17	we	we	PRON
bracis-28417	9	18	study	study	VERB
bracis-28417	9	19	the	the	DET
bracis-28417	9	20	impact	impact	NOUN
bracis-28417	9	21	of	of	ADP
bracis-28417	9	22	language	language	NOUN
bracis-28417	9	23	-	-	PUNCT
bracis-28417	9	24	specific	specific	ADJ
bracis-28417	9	25	pretraining	pretraining	NOUN
bracis-28417	9	26	in	in	ADP
bracis-28417	9	27	terms	term	NOUN
bracis-28417	9	28	of	of	ADP
bracis-28417	9	29	1	1	NUM
bracis-28417	9	30	)	)	PUNCT
bracis-28417	9	31	capturing	capture	VERB
bracis-28417	9	32	linguistic	linguistic	ADJ
bracis-28417	9	33	nuances	nuance	NOUN
bracis-28417	9	34	and	and	CCONJ
bracis-28417	9	35	structures	structure	NOUN
bracis-28417	9	36	inherent	inherent	ADJ
bracis-28417	9	37	to	to	ADP
bracis-28417	9	38	the	the	DET
bracis-28417	9	39	target	target	NOUN
bracis-28417	9	40	language	language	NOUN
bracis-28417	9	41	,	,	PUNCT
bracis-28417	9	42	and	and	CCONJ
bracis-28417	9	43	2	2	X
bracis-28417	9	44	)	)	PUNCT
bracis-28417	9	45	enriching	enrich	VERB
bracis-28417	9	46	the	the	DET
bracis-28417	9	47	model	model	NOUN
bracis-28417	9	48	’s	’s	PART
bracis-28417	9	49	knowledge	knowledge	NOUN
bracis-28417	9	50	about	about	ADP
bracis-28417	9	51	a	a	DET
bracis-28417	9	52	domain	domain	NOUN
bracis-28417	9	53	or	or	CCONJ
bracis-28417	9	54	culture	culture	NOUN
bracis-28417	9	55	.	.	PUNCT
bracis-28417	10	1	our	our	PRON
bracis-28417	10	2	results	result	NOUN
bracis-28417	10	3	indicate	indicate	VERB
bracis-28417	10	4	that	that	SCONJ
bracis-28417	10	5	most	most	ADJ
bracis-28417	10	6	benefits	benefit	NOUN
bracis-28417	10	7	stem	stem	VERB
bracis-28417	10	8	from	from	ADP
bracis-28417	10	9	the	the	DET
bracis-28417	10	10	domain	domain	NOUN
bracis-28417	10	11	-	-	PUNCT
bracis-28417	10	12	specific	specific	ADJ
bracis-28417	10	13	knowledge	knowledge	NOUN
bracis-28417	10	14	acquired	acquire	VERB
bracis-28417	10	15	through	through	ADP
bracis-28417	10	16	monolingual	monolingual	ADJ
bracis-28417	10	17	pretraining	pretraining	NOUN
bracis-28417	10	18	.	.	PUNCT
bracis-28417	11	1	finally	finally	ADV
bracis-28417	11	2	,	,	PUNCT
bracis-28417	11	3	we	we	PRON
bracis-28417	11	4	show	show	VERB
bracis-28417	11	5	that	that	SCONJ
bracis-28417	11	6	our	our	PRON
bracis-28417	11	7	optimized	optimize	VERB
bracis-28417	11	8	model	model	NOUN
bracis-28417	11	9	for	for	ADP
bracis-28417	11	10	portuguese	portuguese	PROPN
bracis-28417	11	11	demonstrates	demonstrate	VERB
bracis-28417	11	12	a	a	DET
bracis-28417	11	13	reduced	reduce	VERB
bracis-28417	11	14	performance	performance	NOUN
bracis-28417	11	15	in	in	ADP
bracis-28417	11	16	english	english	ADJ
bracis-28417	11	17	tasks	task	NOUN
bracis-28417	11	18	,	,	PUNCT
bracis-28417	11	19	thereby	thereby	ADV
bracis-28417	11	20	substantiating	substantiate	VERB
bracis-28417	11	21	the	the	DET
bracis-28417	11	22	inherent	inherent	ADJ
bracis-28417	11	23	compromise	compromise	NOUN
bracis-28417	11	24	in	in	ADP
bracis-28417	11	25	refining	refining	NOUN
bracis-28417	11	26	models	model	NOUN
bracis-28417	11	27	for	for	ADP
bracis-28417	11	28	specific	specific	ADJ
bracis-28417	11	29	linguistic	linguistic	ADJ
bracis-28417	11	30	domains	domain	NOUN
bracis-28417	11	31	.	.	PUNCT
bracis-28417	12	1	access	access	NOUN
bracis-28417	12	2	provided	provide	VERB
bracis-28417	12	3	by	by	ADP
bracis-28417	12	4	university	university	PROPN
bracis-28417	12	5	of	of	ADP
bracis-28417	12	6	notre	notre	PROPN
bracis-28417	12	7	dame	dame	PROPN
bracis-28417	12	8	hesburgh	hesburgh	PROPN
bracis-28417	12	9	library	library	PROPN
bracis-28417	12	10	.	.	PUNCT
bracis-28417	13	1	download	download	PROPN
bracis-28417	13	2	conference	conference	NOUN
bracis-28417	13	3	paper	paper	NOUN
bracis-28417	13	4	pdf	pdf	NOUN
bracis-28417	13	5	similar	similar	ADJ
bracis-28417	13	6	content	content	NOUN
bracis-28417	13	7	being	be	AUX
bracis-28417	13	8	viewed	view	VERB
bracis-28417	13	9	by	by	ADP
bracis-28417	13	10	others	other	NOUN
bracis-28417	13	11	ptt5	ptt5	PROPN
bracis-28417	13	12	-	-	PUNCT
bracis-28417	13	13	v2	v2	PROPN
bracis-28417	13	14	:	:	PUNCT
bracis-28417	13	15	a	a	DET
bracis-28417	13	16	closer	close	ADJ
bracis-28417	13	17	look	look	NOUN
bracis-28417	13	18	at	at	ADP
bracis-28417	13	19	 	 	SPACE
bracis-28417	13	20	continued	continue	VERB
bracis-28417	13	21	pretraining	pretraine	VERB
bracis-28417	13	22	of	of	ADP
bracis-28417	13	23	 	 	SPACE
bracis-28417	13	24	t5	t5	PROPN
bracis-28417	13	25	models	model	NOUN
bracis-28417	13	26	for	for	ADP
bracis-28417	13	27	 	 	SPACE
bracis-28417	13	28	the	the	DET
bracis-28417	13	29	 	 	SPACE
bracis-28417	13	30	portuguese	portuguese	ADJ
bracis-28417	13	31	language	language	NOUN
bracis-28417	13	32	chapter	chapter	NOUN
bracis-28417	13	33	©	©	PROPN
bracis-28417	13	34	2025	2025	NUM
bracis-28417	13	35	monoand	monoand	NOUN
bracis-28417	13	36	multilingual	multilingual	ADJ
bracis-28417	13	37	gpt-3	gpt-3	NOUN
bracis-28417	13	38	models	model	NOUN
bracis-28417	13	39	for	for	ADP
bracis-28417	13	40	 	 	SPACE
bracis-28417	13	41	hungarian	hungarian	ADJ
bracis-28417	13	42	chapter	chapter	NOUN
bracis-28417	13	43	©	©	PROPN
bracis-28417	13	44	2023	2023	NUM
bracis-28417	13	45	scaling	scaling	NOUN
bracis-28417	13	46	and	and	CCONJ
bracis-28417	13	47	 	 	SPACE
bracis-28417	13	48	adapting	adapt	VERB
bracis-28417	13	49	large	large	ADJ
bracis-28417	13	50	language	language	NOUN
bracis-28417	13	51	models	model	NOUN
bracis-28417	13	52	for	for	ADP
bracis-28417	13	53	 	 	SPACE
bracis-28417	13	54	portuguese	portuguese	ADJ
bracis-28417	13	55	open	open	ADJ
bracis-28417	13	56	information	information	NOUN
bracis-28417	13	57	extraction	extraction	NOUN
bracis-28417	13	58	:	:	PUNCT
bracis-28417	13	59	a	a	DET
bracis-28417	13	60	comparative	comparative	ADJ
bracis-28417	13	61	study	study	NOUN
bracis-28417	13	62	of	of	ADP
bracis-28417	13	63	 	 	SPACE
bracis-28417	13	64	fine	fine	ADV
bracis-28417	13	65	-	-	PUNCT
bracis-28417	13	66	tuning	tuning	NOUN
bracis-28417	13	67	and	and	CCONJ
bracis-28417	13	68	 	 	SPACE
bracis-28417	13	69	lora	lora	PROPN
bracis-28417	13	70	chapter	chapter	NOUN
bracis-28417	13	71	©	©	PROPN
bracis-28417	13	72	2025	2025	NUM
bracis-28417	13	73	explore	explore	VERB
bracis-28417	13	74	related	relate	VERB
bracis-28417	13	75	subjects	subject	NOUN
bracis-28417	13	76	discover	discover	VERB
bracis-28417	13	77	the	the	DET
bracis-28417	13	78	latest	late	ADJ
bracis-28417	13	79	articles	article	NOUN
bracis-28417	13	80	,	,	PUNCT
bracis-28417	13	81	books	book	NOUN
bracis-28417	13	82	and	and	CCONJ
bracis-28417	13	83	news	news	NOUN
bracis-28417	13	84	in	in	ADP
bracis-28417	13	85	related	related	ADJ
bracis-28417	13	86	subjects	subject	NOUN
bracis-28417	13	87	,	,	PUNCT
bracis-28417	13	88	suggested	suggest	VERB
bracis-28417	13	89	using	use	VERB
bracis-28417	13	90	machine	machine	NOUN
bracis-28417	13	91	learning	learning	NOUN
bracis-28417	13	92	.	.	PUNCT
bracis-28417	14	1	african	african	ADJ
bracis-28417	14	2	languages	language	NOUN
bracis-28417	14	3	language	language	NOUN
bracis-28417	14	4	teaching	teaching	NOUN
bracis-28417	14	5	and	and	CCONJ
bracis-28417	14	6	learning	learn	VERB
bracis-28417	14	7	languages	language	NOUN
bracis-28417	14	8	language	language	NOUN
bracis-28417	14	9	education	education	NOUN
bracis-28417	14	10	portuguese	portuguese	ADJ
bracis-28417	14	11	language	language	NOUN
bracis-28417	14	12	policy	policy	NOUN
bracis-28417	14	13	and	and	CCONJ
bracis-28417	14	14	planning	plan	VERB
bracis-28417	14	15	1	1	NUM
bracis-28417	14	16	introduction	introduction	NOUN
bracis-28417	14	17	language	language	NOUN
bracis-28417	14	18	models	model	NOUN
bracis-28417	14	19	have	have	AUX
bracis-28417	14	20	revolutionized	revolutionize	VERB
bracis-28417	14	21	the	the	DET
bracis-28417	14	22	field	field	NOUN
bracis-28417	14	23	of	of	ADP
bracis-28417	14	24	natural	natural	ADJ
bracis-28417	14	25	language	language	NOUN
bracis-28417	14	26	processing	processing	NOUN
bracis-28417	14	27	with	with	ADP
bracis-28417	14	28	their	their	PRON
bracis-28417	14	29	exceptional	exceptional	ADJ
bracis-28417	14	30	ability	ability	NOUN
bracis-28417	14	31	to	to	PART
bracis-28417	14	32	perform	perform	VERB
bracis-28417	14	33	tasks	task	NOUN
bracis-28417	14	34	with	with	ADP
bracis-28417	14	35	minimal	minimal	ADJ
bracis-28417	14	36	supervision	supervision	NOUN
bracis-28417	14	37	.	.	PUNCT
bracis-28417	15	1	although	although	SCONJ
bracis-28417	15	2	primarily	primarily	ADV
bracis-28417	15	3	pretrained	pretraine	VERB
bracis-28417	15	4	on	on	ADP
bracis-28417	15	5	english	english	ADJ
bracis-28417	15	6	-	-	PUNCT
bracis-28417	15	7	centric	centric	NOUN
bracis-28417	15	8	corpora	corpora	NOUN
bracis-28417	15	9	,	,	PUNCT
bracis-28417	15	10	the	the	DET
bracis-28417	15	11	models	model	NOUN
bracis-28417	15	12	have	have	AUX
bracis-28417	15	13	shown	show	VERB
bracis-28417	15	14	impressive	impressive	ADJ
bracis-28417	15	15	multilingual	multilingual	ADJ
bracis-28417	15	16	capabilities	capability	NOUN
bracis-28417	15	17	 	 	SPACE
bracis-28417	16	1	[	[	X
bracis-28417	16	2	10	10	NUM
bracis-28417	16	3	]	]	PUNCT
bracis-28417	16	4	.	.	PUNCT
bracis-28417	17	1	given	give	VERB
bracis-28417	17	2	the	the	DET
bracis-28417	17	3	abundance	abundance	NOUN
bracis-28417	17	4	of	of	ADP
bracis-28417	17	5	languages	language	NOUN
bracis-28417	17	6	worldwide	worldwide	ADV
bracis-28417	17	7	,	,	PUNCT
bracis-28417	17	8	the	the	DET
bracis-28417	17	9	majority	majority	NOUN
bracis-28417	17	10	of	of	ADP
bracis-28417	17	11	which	which	PRON
bracis-28417	17	12	are	be	AUX
bracis-28417	17	13	low	low	ADJ
bracis-28417	17	14	-	-	PUNCT
bracis-28417	17	15	resource	resource	NOUN
bracis-28417	17	16	,	,	PUNCT
bracis-28417	17	17	it	it	PRON
bracis-28417	17	18	has	have	AUX
bracis-28417	17	19	become	become	VERB
bracis-28417	17	20	a	a	DET
bracis-28417	17	21	common	common	ADJ
bracis-28417	17	22	practice	practice	NOUN
bracis-28417	17	23	to	to	PART
bracis-28417	17	24	pretrain	pretrain	VERB
bracis-28417	17	25	single	single	ADJ
bracis-28417	17	26	models	model	NOUN
bracis-28417	17	27	on	on	ADP
bracis-28417	17	28	multiple	multiple	ADJ
bracis-28417	17	29	languages	language	NOUN
bracis-28417	17	30	simultaneously	simultaneously	ADV
bracis-28417	17	31	.	.	PUNCT
bracis-28417	18	1	models	model	NOUN
bracis-28417	18	2	like	like	ADP
bracis-28417	18	3	xlm	xlm	NOUN
bracis-28417	18	4	-	-	PUNCT
bracis-28417	18	5	r	r	NOUN
bracis-28417	18	6	 	 	SPACE
bracis-28417	18	7	[	[	NOUN
bracis-28417	18	8	12	12	NUM
bracis-28417	18	9	]	]	PUNCT
bracis-28417	18	10	,	,	PUNCT
bracis-28417	18	11	mbart	mbart	NOUN
bracis-28417	18	12	 	 	SPACE
bracis-28417	19	1	[	[	X
bracis-28417	19	2	28	28	NUM
bracis-28417	19	3	]	]	X
bracis-28417	19	4	,	,	PUNCT
bracis-28417	19	5	mt5	mt5	ADJ
bracis-28417	19	6	 	 	SPACE
bracis-28417	19	7	[	[	X
bracis-28417	19	8	70	70	NUM
bracis-28417	19	9	]	]	PUNCT
bracis-28417	19	10	,	,	PUNCT
bracis-28417	19	11	and	and	CCONJ
bracis-28417	19	12	bloom	bloom	NOUN
bracis-28417	19	13	 	 	SPACE
bracis-28417	20	1	[	[	X
bracis-28417	20	2	54	54	NUM
bracis-28417	20	3	]	]	PUNCT
bracis-28417	20	4	exemplify	exemplify	VERB
bracis-28417	20	5	this	this	DET
bracis-28417	20	6	approach	approach	NOUN
bracis-28417	20	7	.	.	PUNCT
bracis-28417	21	1	despite	despite	SCONJ
bracis-28417	21	2	the	the	DET
bracis-28417	21	3	success	success	NOUN
bracis-28417	21	4	of	of	ADP
bracis-28417	21	5	these	these	DET
bracis-28417	21	6	multilingual	multilingual	ADJ
bracis-28417	21	7	models	model	NOUN
bracis-28417	21	8	,	,	PUNCT
bracis-28417	21	9	we	we	PRON
bracis-28417	21	10	argue	argue	VERB
bracis-28417	21	11	that	that	SCONJ
bracis-28417	21	12	they	they	PRON
bracis-28417	21	13	may	may	AUX
bracis-28417	21	14	not	not	PART
bracis-28417	21	15	be	be	AUX
bracis-28417	21	16	the	the	DET
bracis-28417	21	17	optimal	optimal	ADJ
bracis-28417	21	18	approach	approach	NOUN
bracis-28417	21	19	for	for	ADP
bracis-28417	21	20	capturing	capture	VERB
bracis-28417	21	21	the	the	DET
bracis-28417	21	22	cultural	cultural	ADJ
bracis-28417	21	23	and	and	CCONJ
bracis-28417	21	24	knowledge	knowledge	NOUN
bracis-28417	21	25	richness	richness	NOUN
bracis-28417	21	26	inherent	inherent	ADJ
bracis-28417	21	27	in	in	ADP
bracis-28417	21	28	individual	individual	ADJ
bracis-28417	21	29	languages	language	NOUN
bracis-28417	21	30	.	.	PUNCT
bracis-28417	22	1	when	when	SCONJ
bracis-28417	22	2	a	a	DET
bracis-28417	22	3	moderately	moderately	ADV
bracis-28417	22	4	-	-	PUNCT
bracis-28417	22	5	sized	sized	ADJ
bracis-28417	22	6	language	language	NOUN
bracis-28417	22	7	-	-	PUNCT
bracis-28417	22	8	specific	specific	NOUN
bracis-28417	22	9	corpus	corpus	NOUN
bracis-28417	22	10	is	be	AUX
bracis-28417	22	11	available	available	ADJ
bracis-28417	22	12	,	,	PUNCT
bracis-28417	22	13	continued	continued	ADJ
bracis-28417	22	14	pretraining	pretraine	VERB
bracis-28417	22	15	could	could	AUX
bracis-28417	22	16	integrate	integrate	VERB
bracis-28417	22	17	the	the	DET
bracis-28417	22	18	missing	miss	VERB
bracis-28417	22	19	knowledge	knowledge	NOUN
bracis-28417	22	20	into	into	ADP
bracis-28417	22	21	the	the	DET
bracis-28417	22	22	model	model	NOUN
bracis-28417	22	23	,	,	PUNCT
bracis-28417	22	24	enhancing	enhance	VERB
bracis-28417	22	25	its	its	PRON
bracis-28417	22	26	performance	performance	NOUN
bracis-28417	22	27	on	on	ADP
bracis-28417	22	28	targeted	target	VERB
bracis-28417	22	29	tasks	task	NOUN
bracis-28417	22	30	.	.	PUNCT
bracis-28417	23	1	to	to	PART
bracis-28417	23	2	test	test	VERB
bracis-28417	23	3	this	this	DET
bracis-28417	23	4	hypothesis	hypothesis	NOUN
bracis-28417	23	5	,	,	PUNCT
bracis-28417	23	6	we	we	PRON
bracis-28417	23	7	extend	extend	VERB
bracis-28417	23	8	the	the	DET
bracis-28417	23	9	pretraining	pretraining	NOUN
bracis-28417	23	10	of	of	ADP
bracis-28417	23	11	english	english	ADJ
bracis-28417	23	12	-	-	PUNCT
bracis-28417	23	13	centric	centric	ADJ
bracis-28417	23	14	models	model	NOUN
bracis-28417	23	15	using	use	VERB
bracis-28417	23	16	portuguese	portuguese	PROPN
bracis-28417	23	17	corpora	corpora	PROPN
bracis-28417	23	18	and	and	CCONJ
bracis-28417	23	19	evaluate	evaluate	VERB
bracis-28417	23	20	their	their	PRON
bracis-28417	23	21	performance	performance	NOUN
bracis-28417	23	22	on	on	ADP
bracis-28417	23	23	an	an	DET
bracis-28417	23	24	extensive	extensive	ADJ
bracis-28417	23	25	range	range	NOUN
bracis-28417	23	26	of	of	ADP
bracis-28417	23	27	portuguese	portuguese	ADJ
bracis-28417	23	28	datasets	dataset	NOUN
bracis-28417	23	29	employing	employ	VERB
bracis-28417	23	30	a	a	DET
bracis-28417	23	31	few	few	ADJ
bracis-28417	23	32	-	-	PUNCT
bracis-28417	23	33	shot	shot	NOUN
bracis-28417	23	34	learning	learning	NOUN
bracis-28417	23	35	approach	approach	NOUN
bracis-28417	23	36	.	.	PUNCT
bracis-28417	24	1	our	our	PRON
bracis-28417	24	2	results	result	NOUN
bracis-28417	24	3	indicate	indicate	VERB
bracis-28417	24	4	that	that	SCONJ
bracis-28417	24	5	,	,	PUNCT
bracis-28417	24	6	even	even	ADV
bracis-28417	24	7	for	for	ADP
bracis-28417	24	8	models	model	NOUN
bracis-28417	24	9	trained	train	VERB
bracis-28417	24	10	beyond	beyond	ADP
bracis-28417	24	11	the	the	DET
bracis-28417	24	12	recommendations	recommendation	NOUN
bracis-28417	24	13	by	by	ADP
bracis-28417	24	14	hoffmann	hoffmann	PROPN
bracis-28417	24	15	et	et	PROPN
bracis-28417	24	16	al	al	PROPN
bracis-28417	24	17	 	 	SPACE
bracis-28417	24	18	[	[	X
bracis-28417	24	19	18	18	NUM
bracis-28417	24	20	]	]	PUNCT
bracis-28417	24	21	,	,	PUNCT
bracis-28417	24	22	this	this	DET
bracis-28417	24	23	additional	additional	ADJ
bracis-28417	24	24	pretraining	pretraine	VERB
bracis-28417	24	25	considerably	considerably	ADV
bracis-28417	24	26	improves	improve	VERB
bracis-28417	24	27	performance	performance	NOUN
bracis-28417	24	28	compared	compare	VERB
bracis-28417	24	29	to	to	ADP
bracis-28417	24	30	multilingual	multilingual	ADJ
bracis-28417	24	31	models	model	NOUN
bracis-28417	24	32	.	.	PUNCT
bracis-28417	25	1	we	we	PRON
bracis-28417	25	2	evaluate	evaluate	VERB
bracis-28417	25	3	our	our	PRON
bracis-28417	25	4	models	model	NOUN
bracis-28417	25	5	on	on	ADP
bracis-28417	25	6	datasets	dataset	NOUN
bracis-28417	25	7	comprising	comprise	VERB
bracis-28417	25	8	texts	text	NOUN
bracis-28417	25	9	originally	originally	ADV
bracis-28417	25	10	created	create	VERB
bracis-28417	25	11	by	by	ADP
bracis-28417	25	12	native	native	ADJ
bracis-28417	25	13	brazilian	brazilian	ADJ
bracis-28417	25	14	portuguese	portuguese	ADJ
bracis-28417	25	15	speakers	speaker	NOUN
bracis-28417	25	16	,	,	PUNCT
bracis-28417	25	17	as	as	ADV
bracis-28417	25	18	well	well	ADV
bracis-28417	25	19	as	as	ADP
bracis-28417	25	20	datasets	dataset	NOUN
bracis-28417	25	21	translated	translate	VERB
bracis-28417	25	22	from	from	ADP
bracis-28417	25	23	english	english	PROPN
bracis-28417	25	24	to	to	ADP
bracis-28417	25	25	portuguese	portuguese	PROPN
bracis-28417	25	26	.	.	PUNCT
bracis-28417	26	1	we	we	PRON
bracis-28417	26	2	observe	observe	VERB
bracis-28417	26	3	improvements	improvement	NOUN
bracis-28417	26	4	across	across	ADP
bracis-28417	26	5	all	all	DET
bracis-28417	26	6	datasets	dataset	NOUN
bracis-28417	26	7	due	due	ADP
bracis-28417	26	8	to	to	ADP
bracis-28417	26	9	the	the	DET
bracis-28417	26	10	portuguese	portuguese	ADJ
bracis-28417	26	11	pretraining	pretraine	VERB
bracis-28417	26	12	,	,	PUNCT
bracis-28417	26	13	with	with	SCONJ
bracis-28417	26	14	the	the	DET
bracis-28417	26	15	gains	gain	NOUN
bracis-28417	26	16	being	be	AUX
bracis-28417	26	17	particularly	particularly	ADV
bracis-28417	26	18	pronounced	pronounce	VERB
bracis-28417	26	19	for	for	ADP
bracis-28417	26	20	datasets	dataset	NOUN
bracis-28417	26	21	created	create	VERB
bracis-28417	26	22	by	by	ADP
bracis-28417	26	23	brazilian	brazilian	ADJ
bracis-28417	26	24	speakers	speaker	NOUN
bracis-28417	26	25	.	.	PUNCT
bracis-28417	27	1	one	one	NUM
bracis-28417	27	2	of	of	ADP
bracis-28417	27	3	the	the	DET
bracis-28417	27	4	largest	large	ADJ
bracis-28417	27	5	improvements	improvement	NOUN
bracis-28417	27	6	was	be	AUX
bracis-28417	27	7	observed	observe	VERB
bracis-28417	27	8	on	on	ADP
bracis-28417	27	9	the	the	DET
bracis-28417	27	10	enem	enem	PROPN
bracis-28417	27	11	dataset	dataset	NOUN
bracis-28417	27	12	 	 	SPACE
bracis-28417	28	1	[	[	X
bracis-28417	28	2	57	57	NUM
bracis-28417	28	3	]	]	PUNCT
bracis-28417	28	4	,	,	PUNCT
bracis-28417	28	5	which	which	PRON
bracis-28417	28	6	is	be	AUX
bracis-28417	28	7	derived	derive	VERB
bracis-28417	28	8	from	from	ADP
bracis-28417	28	9	entrance	entrance	NOUN
bracis-28417	28	10	exams	exam	NOUN
bracis-28417	28	11	used	use	VERB
bracis-28417	28	12	by	by	ADP
bracis-28417	28	13	brazilian	brazilian	ADJ
bracis-28417	28	14	universities	university	NOUN
bracis-28417	28	15	and	and	CCONJ
bracis-28417	28	16	requires	require	VERB
bracis-28417	28	17	extensive	extensive	ADJ
bracis-28417	28	18	knowledge	knowledge	NOUN
bracis-28417	28	19	of	of	ADP
bracis-28417	28	20	the	the	DET
bracis-28417	28	21	country	country	NOUN
bracis-28417	28	22	’s	’s	PART
bracis-28417	28	23	history	history	NOUN
bracis-28417	28	24	,	,	PUNCT
bracis-28417	28	25	geography	geography	NOUN
bracis-28417	28	26	,	,	PUNCT
bracis-28417	28	27	and	and	CCONJ
bracis-28417	28	28	literature	literature	NOUN
bracis-28417	28	29	.	.	PUNCT
bracis-28417	29	1	this	this	DET
bracis-28417	29	2	result	result	NOUN
bracis-28417	29	3	provides	provide	VERB
bracis-28417	29	4	evidence	evidence	NOUN
bracis-28417	29	5	that	that	SCONJ
bracis-28417	29	6	the	the	DET
bracis-28417	29	7	major	major	ADJ
bracis-28417	29	8	contribution	contribution	NOUN
bracis-28417	29	9	of	of	ADP
bracis-28417	29	10	our	our	PRON
bracis-28417	29	11	language	language	NOUN
bracis-28417	29	12	-	-	PUNCT
bracis-28417	29	13	specific	specific	ADJ
bracis-28417	29	14	pretraining	pretraining	NOUN
bracis-28417	29	15	is	be	AUX
bracis-28417	29	16	to	to	PART
bracis-28417	29	17	inject	inject	VERB
bracis-28417	29	18	domain	domain	NOUN
bracis-28417	29	19	-	-	PUNCT
bracis-28417	29	20	specific	specific	ADJ
bracis-28417	29	21	knowledge	knowledge	NOUN
bracis-28417	29	22	about	about	ADP
bracis-28417	29	23	a	a	DET
bracis-28417	29	24	particular	particular	ADJ
bracis-28417	29	25	culture	culture	NOUN
bracis-28417	29	26	as	as	SCONJ
bracis-28417	29	27	opposed	oppose	VERB
bracis-28417	29	28	to	to	PART
bracis-28417	29	29	solely	solely	ADV
bracis-28417	29	30	enhancing	enhance	VERB
bracis-28417	29	31	language	language	NOUN
bracis-28417	29	32	proficiency	proficiency	NOUN
bracis-28417	29	33	.	.	PUNCT
bracis-28417	30	1	2	2	NUM
bracis-28417	30	2	related	relate	VERB
bracis-28417	30	3	work	work	NOUN
bracis-28417	30	4	the	the	DET
bracis-28417	30	5	success	success	NOUN
bracis-28417	30	6	of	of	ADP
bracis-28417	30	7	multilingual	multilingual	ADJ
bracis-28417	30	8	pretraining	pretraine	VERB
bracis-28417	30	9	has	have	AUX
bracis-28417	30	10	been	be	AUX
bracis-28417	30	11	well	well	ADV
bracis-28417	30	12	-	-	PUNCT
bracis-28417	30	13	documented	document	VERB
bracis-28417	30	14	in	in	ADP
bracis-28417	30	15	the	the	DET
bracis-28417	30	16	literature	literature	NOUN
bracis-28417	30	17	,	,	PUNCT
bracis-28417	30	18	with	with	ADP
bracis-28417	30	19	models	model	NOUN
bracis-28417	30	20	such	such	ADJ
bracis-28417	30	21	as	as	ADP
bracis-28417	30	22	byt5	byt5	PROPN
bracis-28417	30	23	 	 	SPACE
bracis-28417	31	1	[	[	X
bracis-28417	31	2	69	69	NUM
bracis-28417	31	3	]	]	X
bracis-28417	31	4	,	,	PUNCT
bracis-28417	31	5	mt5	mt5	ADJ
bracis-28417	31	6	 	 	SPACE
bracis-28417	31	7	[	[	X
bracis-28417	31	8	70	70	NUM
bracis-28417	31	9	]	]	PUNCT
bracis-28417	31	10	,	,	PUNCT
bracis-28417	31	11	xlm	xlm	PROPN
bracis-28417	31	12	-	-	PUNCT
bracis-28417	31	13	r	r	VERB
bracis-28417	31	14	 	 	SPACE
bracis-28417	31	15	[	[	NOUN
bracis-28417	31	16	12	12	NUM
bracis-28417	31	17	]	]	X
bracis-28417	31	18	,	,	PUNCT
bracis-28417	31	19	xglm	xglm	PROPN
bracis-28417	31	20	 	 	SPACE
bracis-28417	32	1	[	[	X
bracis-28417	32	2	27	27	NUM
bracis-28417	32	3	]	]	PUNCT
bracis-28417	32	4	and	and	CCONJ
bracis-28417	32	5	mgpt	mgpt	NOUN
bracis-28417	32	6	 	 	SPACE
bracis-28417	32	7	[	[	X
bracis-28417	32	8	56	56	NUM
bracis-28417	32	9	]	]	PUNCT
bracis-28417	32	10	paving	pave	VERB
bracis-28417	32	11	the	the	DET
bracis-28417	32	12	way	way	NOUN
bracis-28417	32	13	for	for	ADP
bracis-28417	32	14	more	more	ADJ
bracis-28417	32	15	inclusive	inclusive	ADJ
bracis-28417	32	16	language	language	NOUN
bracis-28417	32	17	understanding	understanding	NOUN
bracis-28417	32	18	and	and	CCONJ
bracis-28417	32	19	generation	generation	NOUN
bracis-28417	32	20	by	by	ADP
bracis-28417	32	21	leveraging	leverage	VERB
bracis-28417	32	22	shared	share	VERB
bracis-28417	32	23	knowledge	knowledge	NOUN
bracis-28417	32	24	across	across	ADP
bracis-28417	32	25	multiple	multiple	ADJ
bracis-28417	32	26	languages	language	NOUN
bracis-28417	32	27	.	.	PUNCT
bracis-28417	33	1	however	however	ADV
bracis-28417	33	2	,	,	PUNCT
bracis-28417	33	3	there	there	PRON
bracis-28417	33	4	are	be	VERB
bracis-28417	33	5	limitations	limitation	NOUN
bracis-28417	33	6	to	to	ADP
bracis-28417	33	7	this	this	DET
bracis-28417	33	8	approach	approach	NOUN
bracis-28417	33	9	.	.	PUNCT
bracis-28417	34	1	bloom	bloom	NOUN
bracis-28417	34	2	,	,	PUNCT
bracis-28417	34	3	a	a	DET
bracis-28417	34	4	175b	175b	NUM
bracis-28417	34	5	-	-	PUNCT
bracis-28417	34	6	parameter	parameter	NOUN
bracis-28417	34	7	model	model	NOUN
bracis-28417	34	8	pretrained	pretraine	VERB
bracis-28417	34	9	on	on	ADP
bracis-28417	34	10	46	46	NUM
bracis-28417	34	11	languages	language	NOUN
bracis-28417	34	12	,	,	PUNCT
bracis-28417	34	13	performs	perform	VERB
bracis-28417	34	14	worse	bad	ADJ
bracis-28417	34	15	on	on	ADP
bracis-28417	34	16	english	english	ADJ
bracis-28417	34	17	tasks	task	NOUN
bracis-28417	34	18	compared	compare	VERB
bracis-28417	34	19	to	to	PART
bracis-28417	34	20	opt	opt	VERB
bracis-28417	34	21	 	 	SPACE
bracis-28417	34	22	[	[	X
bracis-28417	34	23	74	74	NUM
bracis-28417	34	24	]	]	PUNCT
bracis-28417	34	25	,	,	PUNCT
bracis-28417	34	26	a	a	DET
bracis-28417	34	27	similarly	similarly	ADV
bracis-28417	34	28	sized	sized	ADJ
bracis-28417	34	29	model	model	NOUN
bracis-28417	34	30	pretrained	pretraine	VERB
bracis-28417	34	31	on	on	ADP
bracis-28417	34	32	english	english	ADJ
bracis-28417	34	33	-	-	PUNCT
bracis-28417	34	34	centric	centric	NOUN
bracis-28417	34	35	corpora	corpora	NOUN
bracis-28417	34	36	using	use	VERB
bracis-28417	34	37	comparable	comparable	ADJ
bracis-28417	34	38	computational	computational	ADJ
bracis-28417	34	39	resources	resource	NOUN
bracis-28417	34	40	and	and	CCONJ
bracis-28417	34	41	data	datum	NOUN
bracis-28417	34	42	size	size	NOUN
bracis-28417	34	43	.	.	PUNCT
bracis-28417	35	1	we	we	PRON
bracis-28417	35	2	conjecture	conjecture	VERB
bracis-28417	35	3	that	that	SCONJ
bracis-28417	35	4	bloom	bloom	NOUN
bracis-28417	35	5	’s	’s	PART
bracis-28417	35	6	underperformance	underperformance	NOUN
bracis-28417	35	7	may	may	AUX
bracis-28417	35	8	be	be	AUX
bracis-28417	35	9	attributed	attribute	VERB
bracis-28417	35	10	to	to	ADP
bracis-28417	35	11	its	its	PRON
bracis-28417	35	12	relatively	relatively	ADV
bracis-28417	35	13	limited	limited	ADJ
bracis-28417	35	14	exposure	exposure	NOUN
bracis-28417	35	15	to	to	ADP
bracis-28417	35	16	english	english	ADJ
bracis-28417	35	17	tokens	token	NOUN
bracis-28417	35	18	during	during	ADP
bracis-28417	35	19	the	the	DET
bracis-28417	35	20	pretraining	pretraine	VERB
bracis-28417	35	21	phase	phase	NOUN
bracis-28417	35	22	.	.	PUNCT
bracis-28417	36	1	consequently	consequently	ADV
bracis-28417	36	2	,	,	PUNCT
bracis-28417	36	3	this	this	DET
bracis-28417	36	4	observation	observation	NOUN
bracis-28417	36	5	suggests	suggest	VERB
bracis-28417	36	6	that	that	SCONJ
bracis-28417	36	7	monolingual	monolingual	ADJ
bracis-28417	36	8	pretraining	pretraine	VERB
bracis-28417	36	9	could	could	AUX
bracis-28417	36	10	offer	offer	VERB
bracis-28417	36	11	supplementary	supplementary	ADJ
bracis-28417	36	12	advantages	advantage	NOUN
bracis-28417	36	13	.	.	PUNCT
bracis-28417	37	1	in	in	ADP
bracis-28417	37	2	support	support	NOUN
bracis-28417	37	3	of	of	ADP
bracis-28417	37	4	this	this	DET
bracis-28417	37	5	hypothesis	hypothesis	NOUN
bracis-28417	37	6	,	,	PUNCT
bracis-28417	37	7	models	model	NOUN
bracis-28417	37	8	with	with	ADP
bracis-28417	37	9	hundreds	hundred	NOUN
bracis-28417	37	10	of	of	ADP
bracis-28417	37	11	millions	million	NOUN
bracis-28417	37	12	of	of	ADP
bracis-28417	37	13	parameters	parameter	NOUN
bracis-28417	37	14	pretrained	pretraine	VERB
bracis-28417	37	15	on	on	ADP
bracis-28417	37	16	monolingual	monolingual	ADJ
bracis-28417	37	17	texts	text	NOUN
bracis-28417	37	18	have	have	AUX
bracis-28417	37	19	demonstrated	demonstrate	VERB
bracis-28417	37	20	gains	gain	NOUN
bracis-28417	37	21	over	over	ADP
bracis-28417	37	22	multilingual	multilingual	ADJ
bracis-28417	37	23	counterparts	counterpart	NOUN
bracis-28417	37	24	 	 	SPACE
bracis-28417	38	1	[	[	X
bracis-28417	38	2	2	2	NUM
bracis-28417	38	3	,	,	PUNCT
bracis-28417	38	4	6,7,8	6,7,8	NUM
bracis-28417	38	5	,	,	PUNCT
bracis-28417	38	6	21	21	NUM
bracis-28417	38	7	,	,	PUNCT
bracis-28417	38	8	24	24	NUM
bracis-28417	38	9	,	,	PUNCT
bracis-28417	38	10	25	25	NUM
bracis-28417	38	11	,	,	PUNCT
bracis-28417	38	12	32	32	NUM
bracis-28417	38	13	,	,	PUNCT
bracis-28417	38	14	36	36	NUM
bracis-28417	38	15	,	,	PUNCT
bracis-28417	38	16	52	52	NUM
bracis-28417	38	17	,	,	PUNCT
bracis-28417	38	18	59	59	NUM
bracis-28417	38	19	]	]	PUNCT
bracis-28417	38	20	.	.	PUNCT
bracis-28417	39	1	additionally	additionally	ADV
bracis-28417	39	2	,	,	PUNCT
bracis-28417	39	3	research	research	NOUN
bracis-28417	39	4	has	have	AUX
bracis-28417	39	5	indicated	indicate	VERB
bracis-28417	39	6	that	that	SCONJ
bracis-28417	39	7	language	language	NOUN
bracis-28417	39	8	adaptation	adaptation	NOUN
bracis-28417	39	9	is	be	AUX
bracis-28417	39	10	beneficial	beneficial	ADJ
bracis-28417	39	11	even	even	ADV
bracis-28417	39	12	for	for	ADP
bracis-28417	39	13	low	low	ADJ
bracis-28417	39	14	-	-	PUNCT
bracis-28417	39	15	resource	resource	NOUN
bracis-28417	39	16	languages	language	NOUN
bracis-28417	39	17	 	 	SPACE
bracis-28417	40	1	[	[	X
bracis-28417	40	2	4	4	NUM
bracis-28417	40	3	,	,	PUNCT
bracis-28417	40	4	13	13	NUM
bracis-28417	40	5	,	,	PUNCT
bracis-28417	40	6	38	38	NUM
bracis-28417	40	7	,	,	PUNCT
bracis-28417	40	8	72	72	NUM
bracis-28417	40	9	]	]	PUNCT
bracis-28417	40	10	.	.	PUNCT
bracis-28417	41	1	however	however	ADV
bracis-28417	41	2	,	,	PUNCT
bracis-28417	41	3	there	there	PRON
bracis-28417	41	4	is	be	VERB
bracis-28417	41	5	a	a	DET
bracis-28417	41	6	limited	limited	ADJ
bracis-28417	41	7	number	number	NOUN
bracis-28417	41	8	of	of	ADP
bracis-28417	41	9	published	publish	VERB
bracis-28417	41	10	research	research	NOUN
bracis-28417	41	11	articles	article	NOUN
bracis-28417	41	12	with	with	ADP
bracis-28417	41	13	comprehensive	comprehensive	ADJ
bracis-28417	41	14	evaluations	evaluation	NOUN
bracis-28417	41	15	of	of	ADP
bracis-28417	41	16	the	the	DET
bracis-28417	41	17	benefits	benefit	NOUN
bracis-28417	41	18	of	of	ADP
bracis-28417	41	19	continued	continue	VERB
bracis-28417	41	20	pretraining	pretraine	VERB
bracis-28417	41	21	at	at	ADP
bracis-28417	41	22	the	the	DET
bracis-28417	41	23	multi	multi	ADJ
bracis-28417	41	24	-	-	ADJ
bracis-28417	41	25	billion	billion	NUM
bracis-28417	41	26	-	-	PUNCT
bracis-28417	41	27	parameter	parameter	NOUN
bracis-28417	41	28	scale	scale	NOUN
bracis-28417	41	29	 	 	SPACE
bracis-28417	42	1	[	[	X
bracis-28417	42	2	22	22	NUM
bracis-28417	42	3	,	,	PUNCT
bracis-28417	42	4	50	50	NUM
bracis-28417	42	5	,	,	PUNCT
bracis-28417	42	6	73	73	NUM
bracis-28417	42	7	]	]	PUNCT
bracis-28417	42	8	.	.	PUNCT
bracis-28417	43	1	through	through	ADP
bracis-28417	43	2	this	this	DET
bracis-28417	43	3	study	study	NOUN
bracis-28417	43	4	,	,	PUNCT
bracis-28417	43	5	we	we	PRON
bracis-28417	43	6	contribute	contribute	VERB
bracis-28417	43	7	to	to	ADP
bracis-28417	43	8	the	the	DET
bracis-28417	43	9	literature	literature	NOUN
bracis-28417	43	10	by	by	ADP
bracis-28417	43	11	demonstrating	demonstrate	VERB
bracis-28417	43	12	the	the	DET
bracis-28417	43	13	effectiveness	effectiveness	NOUN
bracis-28417	43	14	of	of	ADP
bracis-28417	43	15	continued	continue	VERB
bracis-28417	43	16	language	language	NOUN
bracis-28417	43	17	-	-	PUNCT
bracis-28417	43	18	specific	specific	ADJ
bracis-28417	43	19	pretraining	pretraining	NOUN
bracis-28417	43	20	for	for	ADP
bracis-28417	43	21	portuguese	portuguese	ADJ
bracis-28417	43	22	language	language	NOUN
bracis-28417	43	23	models	model	NOUN
bracis-28417	43	24	up	up	ADP
bracis-28417	43	25	to	to	ADP
bracis-28417	43	26	the	the	DET
bracis-28417	43	27	65b	65b	NUM
bracis-28417	43	28	-	-	PUNCT
bracis-28417	43	29	parameter	parameter	NOUN
bracis-28417	43	30	scale	scale	NOUN
bracis-28417	43	31	.	.	PUNCT
bracis-28417	44	1	the	the	DET
bracis-28417	44	2	question	question	NOUN
bracis-28417	44	3	concerning	concern	VERB
bracis-28417	44	4	whether	whether	SCONJ
bracis-28417	44	5	it	it	PRON
bracis-28417	44	6	is	be	AUX
bracis-28417	44	7	advantageous	advantageous	ADJ
bracis-28417	44	8	to	to	PART
bracis-28417	44	9	train	train	VERB
bracis-28417	44	10	models	model	NOUN
bracis-28417	44	11	for	for	ADP
bracis-28417	44	12	specific	specific	ADJ
bracis-28417	44	13	languages	language	NOUN
bracis-28417	44	14	is	be	AUX
bracis-28417	44	15	closely	closely	ADV
bracis-28417	44	16	associated	associate	VERB
bracis-28417	44	17	with	with	ADP
bracis-28417	44	18	the	the	DET
bracis-28417	44	19	question	question	NOUN
bracis-28417	44	20	of	of	ADP
bracis-28417	44	21	whether	whether	SCONJ
bracis-28417	44	22	it	it	PRON
bracis-28417	44	23	is	be	AUX
bracis-28417	44	24	beneficial	beneficial	ADJ
bracis-28417	44	25	to	to	PART
bracis-28417	44	26	train	train	NOUN
bracis-28417	44	27	models	model	NOUN
bracis-28417	44	28	for	for	ADP
bracis-28417	44	29	particular	particular	ADJ
bracis-28417	44	30	domains	domain	NOUN
bracis-28417	44	31	of	of	ADP
bracis-28417	44	32	knowledge	knowledge	NOUN
bracis-28417	44	33	.	.	PUNCT
bracis-28417	45	1	recent	recent	ADJ
bracis-28417	45	2	studies	study	NOUN
bracis-28417	45	3	,	,	PUNCT
bracis-28417	45	4	such	such	ADJ
bracis-28417	45	5	as	as	ADP
bracis-28417	45	6	minerva	minerva	PROPN
bracis-28417	45	7	 	 	SPACE
bracis-28417	45	8	[	[	X
bracis-28417	45	9	26	26	NUM
bracis-28417	45	10	]	]	PUNCT
bracis-28417	45	11	and	and	CCONJ
bracis-28417	45	12	galactica	galactica	PROPN
bracis-28417	45	13	 	 	SPACE
bracis-28417	46	1	[	[	X
bracis-28417	46	2	62	62	NUM
bracis-28417	46	3	]	]	PUNCT
bracis-28417	46	4	,	,	PUNCT
bracis-28417	46	5	have	have	AUX
bracis-28417	46	6	shown	show	VERB
bracis-28417	46	7	that	that	SCONJ
bracis-28417	46	8	domain	domain	NOUN
bracis-28417	46	9	-	-	PUNCT
bracis-28417	46	10	specific	specific	ADJ
bracis-28417	46	11	pretraining	pretraining	NOUN
bracis-28417	46	12	can	can	AUX
bracis-28417	46	13	lead	lead	VERB
bracis-28417	46	14	to	to	ADP
bracis-28417	46	15	significant	significant	ADJ
bracis-28417	46	16	improvements	improvement	NOUN
bracis-28417	46	17	,	,	PUNCT
bracis-28417	46	18	even	even	ADV
bracis-28417	46	19	with	with	ADP
bracis-28417	46	20	a	a	DET
bracis-28417	46	21	smaller	small	ADJ
bracis-28417	46	22	pretraining	pretraine	VERB
bracis-28417	46	23	corpus	corpus	NOUN
bracis-28417	46	24	compared	compare	VERB
bracis-28417	46	25	to	to	ADP
bracis-28417	46	26	large	large	ADJ
bracis-28417	46	27	-	-	PUNCT
bracis-28417	46	28	scale	scale	NOUN
bracis-28417	46	29	,	,	PUNCT
bracis-28417	46	30	general	general	ADJ
bracis-28417	46	31	-	-	PUNCT
bracis-28417	46	32	purpose	purpose	NOUN
bracis-28417	46	33	pretraining	pretraine	VERB
bracis-28417	46	34	corpora	corpora	PROPN
bracis-28417	46	35	.	.	PUNCT
bracis-28417	47	1	analogously	analogously	ADV
bracis-28417	47	2	,	,	PUNCT
bracis-28417	47	3	fu	fu	PROPN
bracis-28417	47	4	et	et	PROPN
bracis-28417	47	5	al	al	PROPN
bracis-28417	47	6	.	.	PUNCT
bracis-28417	47	7	 	 	SPACE
bracis-28417	48	1	[	[	X
bracis-28417	48	2	15	15	NUM
bracis-28417	48	3	]	]	PUNCT
bracis-28417	48	4	demonstrated	demonstrate	VERB
bracis-28417	48	5	the	the	DET
bracis-28417	48	6	feasibility	feasibility	NOUN
bracis-28417	48	7	of	of	ADP
bracis-28417	48	8	specializing	specialize	VERB
bracis-28417	48	9	smaller	small	ADJ
bracis-28417	48	10	models	model	NOUN
bracis-28417	48	11	to	to	PART
bracis-28417	48	12	perform	perform	VERB
bracis-28417	48	13	multi	multi	ADJ
bracis-28417	48	14	-	-	ADJ
bracis-28417	48	15	step	step	ADJ
bracis-28417	48	16	reasoning	reasoning	NOUN
bracis-28417	48	17	,	,	PUNCT
bracis-28417	48	18	a	a	DET
bracis-28417	48	19	capability	capability	NOUN
bracis-28417	48	20	typically	typically	ADV
bracis-28417	48	21	exclusive	exclusive	ADJ
bracis-28417	48	22	to	to	ADP
bracis-28417	48	23	models	model	NOUN
bracis-28417	48	24	with	with	ADP
bracis-28417	48	25	at	at	ADP
bracis-28417	48	26	least	least	ADJ
bracis-28417	48	27	50b	50b	NUM
bracis-28417	48	28	parameters	parameter	NOUN
bracis-28417	48	29	,	,	PUNCT
bracis-28417	48	30	at	at	ADP
bracis-28417	48	31	the	the	DET
bracis-28417	48	32	expense	expense	NOUN
bracis-28417	48	33	of	of	ADP
bracis-28417	48	34	diminished	diminished	ADJ
bracis-28417	48	35	performance	performance	NOUN
bracis-28417	48	36	in	in	ADP
bracis-28417	48	37	other	other	ADJ
bracis-28417	48	38	,	,	PUNCT
bracis-28417	48	39	more	more	ADV
bracis-28417	48	40	general	general	ADJ
bracis-28417	48	41	tasks	task	NOUN
bracis-28417	48	42	.	.	PUNCT
bracis-28417	49	1	pretraining	pretraine	VERB
bracis-28417	49	2	with	with	ADP
bracis-28417	49	3	a	a	DET
bracis-28417	49	4	combination	combination	NOUN
bracis-28417	49	5	of	of	ADP
bracis-28417	49	6	general	general	ADJ
bracis-28417	49	7	and	and	CCONJ
bracis-28417	49	8	domain	domain	NOUN
bracis-28417	49	9	-	-	PUNCT
bracis-28417	49	10	specific	specific	ADJ
bracis-28417	49	11	corpora	corpora	NOUN
bracis-28417	49	12	can	can	AUX
bracis-28417	49	13	potentially	potentially	ADV
bracis-28417	49	14	enhance	enhance	VERB
bracis-28417	49	15	performance	performance	NOUN
bracis-28417	49	16	in	in	ADP
bracis-28417	49	17	specialized	specialized	ADJ
bracis-28417	49	18	tasks	task	NOUN
bracis-28417	49	19	without	without	ADP
bracis-28417	49	20	compromising	compromise	VERB
bracis-28417	49	21	effectiveness	effectiveness	NOUN
bracis-28417	49	22	in	in	ADP
bracis-28417	49	23	general	general	ADJ
bracis-28417	49	24	-	-	PUNCT
bracis-28417	49	25	purpose	purpose	NOUN
bracis-28417	49	26	tasks	task	NOUN
bracis-28417	49	27	,	,	PUNCT
bracis-28417	49	28	albeit	albeit	SCONJ
bracis-28417	49	29	at	at	ADP
bracis-28417	49	30	the	the	DET
bracis-28417	49	31	cost	cost	NOUN
bracis-28417	49	32	of	of	ADP
bracis-28417	49	33	increased	increase	VERB
bracis-28417	49	34	computational	computational	ADJ
bracis-28417	49	35	demands	demand	NOUN
bracis-28417	49	36	.	.	PUNCT
bracis-28417	50	1	for	for	ADP
bracis-28417	50	2	example	example	NOUN
bracis-28417	50	3	,	,	PUNCT
bracis-28417	50	4	bloomberggpt	bloomberggpt	NOUN
bracis-28417	50	5	 	 	SPACE
bracis-28417	51	1	[	[	X
bracis-28417	51	2	68	68	NUM
bracis-28417	51	3	]	]	X
bracis-28417	51	4	,	,	PUNCT
bracis-28417	51	5	a	a	DET
bracis-28417	51	6	50b	50b	NUM
bracis-28417	51	7	-	-	PUNCT
bracis-28417	51	8	parameter	parameter	NOUN
bracis-28417	51	9	model	model	NOUN
bracis-28417	51	10	pretrained	pretraine	VERB
bracis-28417	51	11	on	on	ADP
bracis-28417	51	12	heterogeneous	heterogeneous	ADJ
bracis-28417	51	13	corpus	corpus	NOUN
bracis-28417	51	14	in	in	ADP
bracis-28417	51	15	which	which	PRON
bracis-28417	51	16	more	more	ADJ
bracis-28417	51	17	than	than	ADP
bracis-28417	51	18	half	half	NOUN
bracis-28417	51	19	of	of	ADP
bracis-28417	51	20	texts	text	NOUN
bracis-28417	51	21	are	be	AUX
bracis-28417	51	22	from	from	ADP
bracis-28417	51	23	the	the	DET
bracis-28417	51	24	financial	financial	ADJ
bracis-28417	51	25	domain	domain	NOUN
bracis-28417	51	26	,	,	PUNCT
bracis-28417	51	27	exhibits	exhibit	VERB
bracis-28417	51	28	comparable	comparable	ADJ
bracis-28417	51	29	performance	performance	NOUN
bracis-28417	51	30	to	to	ADP
bracis-28417	51	31	opt-66b	opt-66b	NOUN
bracis-28417	51	32	in	in	ADP
bracis-28417	51	33	general	general	ADJ
bracis-28417	51	34	tasks	task	NOUN
bracis-28417	51	35	.	.	PUNCT
bracis-28417	52	1	however	however	ADV
bracis-28417	52	2	,	,	PUNCT
bracis-28417	52	3	bloomberggpt	bloomberggpt	PROPN
bracis-28417	52	4	’s	’s	PART
bracis-28417	52	5	pretraining	pretraine	VERB
bracis-28417	52	6	dataset	dataset	NOUN
bracis-28417	52	7	is	be	AUX
bracis-28417	52	8	three	three	NUM
bracis-28417	52	9	times	time	NOUN
bracis-28417	52	10	larger	large	ADJ
bracis-28417	52	11	,	,	PUNCT
bracis-28417	52	12	and	and	CCONJ
bracis-28417	52	13	consequently	consequently	ADV
bracis-28417	52	14	used	use	VERB
bracis-28417	52	15	more	more	ADJ
bracis-28417	52	16	computational	computational	ADJ
bracis-28417	52	17	resources	resource	NOUN
bracis-28417	52	18	.	.	PUNCT
bracis-28417	53	1	rather	rather	ADV
bracis-28417	53	2	than	than	ADP
bracis-28417	53	3	pursuing	pursue	VERB
bracis-28417	53	4	a	a	DET
bracis-28417	53	5	single	single	ADJ
bracis-28417	53	6	model	model	NOUN
bracis-28417	53	7	that	that	PRON
bracis-28417	53	8	performs	perform	VERB
bracis-28417	53	9	well	well	ADV
bracis-28417	53	10	across	across	ADP
bracis-28417	53	11	multiple	multiple	ADJ
bracis-28417	53	12	domains	domain	NOUN
bracis-28417	53	13	,	,	PUNCT
bracis-28417	53	14	gururangan	gururangan	VERB
bracis-28417	53	15	et	et	PROPN
bracis-28417	53	16	al	al	PROPN
bracis-28417	53	17	.	.	PUNCT
bracis-28417	53	18	 	 	SPACE
bracis-28417	54	1	[	[	X
bracis-28417	54	2	17	17	NUM
bracis-28417	54	3	]	]	X
bracis-28417	54	4	propose	propose	VERB
bracis-28417	54	5	an	an	DET
bracis-28417	54	6	alternative	alternative	ADJ
bracis-28417	54	7	approach	approach	NOUN
bracis-28417	54	8	:	:	PUNCT
bracis-28417	54	9	using	use	VERB
bracis-28417	54	10	multiple	multiple	ADJ
bracis-28417	54	11	expert	expert	NOUN
bracis-28417	54	12	models	model	NOUN
bracis-28417	54	13	,	,	PUNCT
bracis-28417	54	14	each	each	PRON
bracis-28417	54	15	trained	train	VERB
bracis-28417	54	16	on	on	ADP
bracis-28417	54	17	a	a	DET
bracis-28417	54	18	domain	domain	NOUN
bracis-28417	54	19	-	-	PUNCT
bracis-28417	54	20	specific	specific	ADJ
bracis-28417	54	21	subset	subset	NOUN
bracis-28417	54	22	within	within	ADP
bracis-28417	54	23	a	a	DET
bracis-28417	54	24	broader	broad	ADJ
bracis-28417	54	25	,	,	PUNCT
bracis-28417	54	26	diverse	diverse	ADJ
bracis-28417	54	27	dataset	dataset	NOUN
bracis-28417	54	28	,	,	PUNCT
bracis-28417	54	29	to	to	PART
bracis-28417	54	30	function	function	VERB
bracis-28417	54	31	as	as	ADP
bracis-28417	54	32	a	a	DET
bracis-28417	54	33	single	single	ADJ
bracis-28417	54	34	general	general	ADJ
bracis-28417	54	35	-	-	PUNCT
bracis-28417	54	36	purpose	purpose	NOUN
bracis-28417	54	37	model	model	NOUN
bracis-28417	54	38	.	.	PUNCT
bracis-28417	55	1	their	their	PRON
bracis-28417	55	2	models	model	NOUN
bracis-28417	55	3	outperform	outperform	VERB
bracis-28417	55	4	dense	dense	ADJ
bracis-28417	55	5	ones	one	NOUN
bracis-28417	55	6	across	across	ADP
bracis-28417	55	7	various	various	ADJ
bracis-28417	55	8	domain	domain	NOUN
bracis-28417	55	9	-	-	PUNCT
bracis-28417	55	10	specific	specific	ADJ
bracis-28417	55	11	tasks	task	NOUN
bracis-28417	55	12	,	,	PUNCT
bracis-28417	55	13	at	at	ADP
bracis-28417	55	14	the	the	DET
bracis-28417	55	15	expense	expense	NOUN
bracis-28417	55	16	of	of	ADP
bracis-28417	55	17	an	an	DET
bracis-28417	55	18	increased	increase	VERB
bracis-28417	55	19	parameter	parameter	NOUN
bracis-28417	55	20	count	count	NOUN
bracis-28417	55	21	,	,	PUNCT
bracis-28417	55	22	consequently	consequently	ADV
bracis-28417	55	23	leading	lead	VERB
bracis-28417	55	24	to	to	ADP
bracis-28417	55	25	larger	large	ADJ
bracis-28417	55	26	memory	memory	NOUN
bracis-28417	55	27	requirements	requirement	NOUN
bracis-28417	55	28	for	for	ADP
bracis-28417	55	29	efficient	efficient	ADJ
bracis-28417	55	30	inference.footnote	inference.footnote	NUM
bracis-28417	55	31	1	1	NUM
bracis-28417	55	32	3	3	NUM
bracis-28417	55	33	methodology	methodology	NOUN
bracis-28417	55	34	in	in	ADP
bracis-28417	55	35	this	this	DET
bracis-28417	55	36	section	section	NOUN
bracis-28417	55	37	,	,	PUNCT
bracis-28417	55	38	we	we	PRON
bracis-28417	55	39	outline	outline	VERB
bracis-28417	55	40	the	the	DET
bracis-28417	55	41	pretraining	pretraine	VERB
bracis-28417	55	42	data	datum	NOUN
bracis-28417	55	43	and	and	CCONJ
bracis-28417	55	44	training	training	NOUN
bracis-28417	55	45	details	detail	NOUN
bracis-28417	55	46	used	use	VERB
bracis-28417	55	47	to	to	PART
bracis-28417	55	48	build	build	VERB
bracis-28417	55	49	our	our	PRON
bracis-28417	55	50	models	model	NOUN
bracis-28417	55	51	,	,	PUNCT
bracis-28417	55	52	including	include	VERB
bracis-28417	55	53	data	datum	NOUN
bracis-28417	55	54	sources	source	NOUN
bracis-28417	55	55	,	,	PUNCT
bracis-28417	55	56	preprocessing	preprocesse	VERB
bracis-28417	55	57	techniques	technique	NOUN
bracis-28417	55	58	,	,	PUNCT
bracis-28417	55	59	architectures	architecture	NOUN
bracis-28417	55	60	,	,	PUNCT
bracis-28417	55	61	hyperparameters	hyperparameter	NOUN
bracis-28417	55	62	,	,	PUNCT
bracis-28417	55	63	and	and	CCONJ
bracis-28417	55	64	optimization	optimization	NOUN
bracis-28417	55	65	methods	method	NOUN
bracis-28417	55	66	.	.	PUNCT
bracis-28417	56	1	3.1	3.1	NUM
bracis-28417	56	2	pretraining	pretraine	VERB
bracis-28417	56	3	data	datum	NOUN
bracis-28417	56	4	the	the	DET
bracis-28417	56	5	pretraining	pretraine	VERB
bracis-28417	56	6	data	datum	NOUN
bracis-28417	56	7	is	be	AUX
bracis-28417	56	8	derived	derive	VERB
bracis-28417	56	9	from	from	ADP
bracis-28417	56	10	the	the	DET
bracis-28417	56	11	portuguese	portuguese	PROPN
bracis-28417	56	12	subset	subset	NOUN
bracis-28417	56	13	of	of	ADP
bracis-28417	56	14	the	the	DET
bracis-28417	56	15	clueweb	clueweb	NOUN
bracis-28417	56	16	2022	2022	NUM
bracis-28417	56	17	dataset	dataset	VERB
bracis-28417	56	18	 	 	SPACE
bracis-28417	56	19	[	[	X
bracis-28417	56	20	40	40	NUM
bracis-28417	56	21	,	,	PUNCT
bracis-28417	56	22	41	41	NUM
bracis-28417	56	23	]	]	PUNCT
bracis-28417	56	24	.	.	PUNCT
bracis-28417	57	1	to	to	PART
bracis-28417	57	2	increase	increase	VERB
bracis-28417	57	3	the	the	DET
bracis-28417	57	4	datasets	dataset	NOUN
bracis-28417	57	5	’s	’s	PART
bracis-28417	57	6	quality	quality	NOUN
bracis-28417	57	7	,	,	PUNCT
bracis-28417	57	8	we	we	PRON
bracis-28417	57	9	apply	apply	VERB
bracis-28417	57	10	the	the	DET
bracis-28417	57	11	quality	quality	NOUN
bracis-28417	57	12	filters	filter	NOUN
bracis-28417	57	13	from	from	ADP
bracis-28417	57	14	massivetext	massivetext	ADJ
bracis-28417	57	15	 	 	SPACE
bracis-28417	58	1	[	[	X
bracis-28417	58	2	45	45	NUM
bracis-28417	58	3	]	]	PUNCT
bracis-28417	58	4	,	,	PUNCT
bracis-28417	58	5	modifying	modify	VERB
bracis-28417	58	6	them	they	PRON
bracis-28417	58	7	to	to	PART
bracis-28417	58	8	accommodate	accommodate	VERB
bracis-28417	58	9	the	the	DET
bracis-28417	58	10	specific	specific	ADJ
bracis-28417	58	11	requirements	requirement	NOUN
bracis-28417	58	12	of	of	ADP
bracis-28417	58	13	the	the	DET
bracis-28417	58	14	portuguese	portuguese	ADJ
bracis-28417	58	15	language	language	NOUN
bracis-28417	58	16	.	.	PUNCT
bracis-28417	59	1	we	we	PRON
bracis-28417	59	2	normalize	normalize	VERB
bracis-28417	59	3	the	the	DET
bracis-28417	59	4	text	text	NOUN
bracis-28417	59	5	with	with	ADP
bracis-28417	59	6	ftfyfootnote	ftfyfootnote	NOUN
bracis-28417	59	7	2	2	NUM
bracis-28417	59	8	,	,	PUNCT
bracis-28417	59	9	convert	convert	VERB
bracis-28417	59	10	wikitexts	wikitext	NOUN
bracis-28417	59	11	into	into	ADP
bracis-28417	59	12	human	human	ADJ
bracis-28417	59	13	-	-	PUNCT
bracis-28417	59	14	readable	readable	ADJ
bracis-28417	59	15	texts	text	NOUN
bracis-28417	59	16	,	,	PUNCT
bracis-28417	59	17	and	and	CCONJ
bracis-28417	59	18	exclude	exclude	VERB
bracis-28417	59	19	documents	document	NOUN
bracis-28417	59	20	containing	contain	VERB
bracis-28417	59	21	less	less	ADJ
bracis-28417	59	22	than	than	ADP
bracis-28417	59	23	200	200	NUM
bracis-28417	59	24	unique	unique	ADJ
bracis-28417	59	25	tokens	token	NOUN
bracis-28417	59	26	.	.	PUNCT
bracis-28417	60	1	these	these	DET
bracis-28417	60	2	quality	quality	NOUN
bracis-28417	60	3	filters	filter	NOUN
bracis-28417	60	4	are	be	AUX
bracis-28417	60	5	primarily	primarily	ADV
bracis-28417	60	6	designed	design	VERB
bracis-28417	60	7	for	for	ADP
bracis-28417	60	8	web	web	NOUN
bracis-28417	60	9	pages	page	NOUN
bracis-28417	60	10	and	and	CCONJ
bracis-28417	60	11	may	may	AUX
bracis-28417	60	12	not	not	PART
bracis-28417	60	13	seamlessly	seamlessly	ADV
bracis-28417	60	14	transfer	transfer	VERB
bracis-28417	60	15	to	to	ADP
bracis-28417	60	16	other	other	ADJ
bracis-28417	60	17	domains	domain	NOUN
bracis-28417	60	18	.	.	PUNCT
bracis-28417	61	1	there	there	PRON
bracis-28417	61	2	is	be	VERB
bracis-28417	61	3	potential	potential	ADJ
bracis-28417	61	4	for	for	ADP
bracis-28417	61	5	improvement	improvement	NOUN
bracis-28417	61	6	by	by	ADP
bracis-28417	61	7	employing	employ	VERB
bracis-28417	61	8	more	more	ADV
bracis-28417	61	9	automated	automate	VERB
bracis-28417	61	10	methods	method	NOUN
bracis-28417	61	11	;	;	PUNCT
bracis-28417	61	12	however	however	ADV
bracis-28417	61	13	,	,	PUNCT
bracis-28417	61	14	this	this	DET
bracis-28417	61	15	study	study	NOUN
bracis-28417	61	16	did	do	AUX
bracis-28417	61	17	not	not	PART
bracis-28417	61	18	explore	explore	VERB
bracis-28417	61	19	such	such	ADJ
bracis-28417	61	20	approaches	approach	NOUN
bracis-28417	61	21	due	due	ADP
bracis-28417	61	22	to	to	ADP
bracis-28417	61	23	the	the	DET
bracis-28417	61	24	resource	resource	NOUN
bracis-28417	61	25	-	-	PUNCT
bracis-28417	61	26	intensive	intensive	ADJ
bracis-28417	61	27	nature	nature	NOUN
bracis-28417	61	28	of	of	ADP
bracis-28417	61	29	pretraining	pretraine	VERB
bracis-28417	61	30	experiments	experiment	NOUN
bracis-28417	61	31	.	.	PUNCT
bracis-28417	62	1	following	follow	VERB
bracis-28417	62	2	the	the	DET
bracis-28417	62	3	cleaning	cleaning	NOUN
bracis-28417	62	4	process	process	NOUN
bracis-28417	62	5	,	,	PUNCT
bracis-28417	62	6	all	all	DET
bracis-28417	62	7	documents	document	NOUN
bracis-28417	62	8	are	be	AUX
bracis-28417	62	9	concatenated	concatenate	VERB
bracis-28417	62	10	using	use	VERB
bracis-28417	62	11	an	an	DET
bracis-28417	62	12	end	end	NOUN
bracis-28417	62	13	-	-	PUNCT
bracis-28417	62	14	of	of	ADP
bracis-28417	62	15	-	-	PUNCT
bracis-28417	62	16	sequence	sequence	NOUN
bracis-28417	62	17	token	token	VERB
bracis-28417	62	18	as	as	ADP
bracis-28417	62	19	a	a	DET
bracis-28417	62	20	separator	separator	NOUN
bracis-28417	62	21	,	,	PUNCT
bracis-28417	62	22	and	and	CCONJ
bracis-28417	62	23	then	then	ADV
bracis-28417	62	24	tokenized	tokenize	VERB
bracis-28417	62	25	.	.	PUNCT
bracis-28417	63	1	the	the	DET
bracis-28417	63	2	gpt	gpt	PROPN
bracis-28417	63	3	-	-	PUNCT
bracis-28417	63	4	j	j	NOUN
bracis-28417	63	5	tokenizer	tokenizer	NOUN
bracis-28417	63	6	,	,	PUNCT
bracis-28417	63	7	which	which	PRON
bracis-28417	63	8	is	be	AUX
bracis-28417	63	9	identical	identical	ADJ
bracis-28417	63	10	to	to	ADP
bracis-28417	63	11	the	the	DET
bracis-28417	63	12	gpt-2	gpt-2	NOUN
bracis-28417	63	13	tokenizer	tokenizer	NOUN
bracis-28417	63	14	 	 	SPACE
bracis-28417	64	1	[	[	X
bracis-28417	64	2	44	44	NUM
bracis-28417	64	3	]	]	PUNCT
bracis-28417	64	4	,	,	PUNCT
bracis-28417	64	5	produces	produce	VERB
bracis-28417	64	6	7.8	7.8	NUM
bracis-28417	64	7	billion	billion	NUM
bracis-28417	64	8	tokens	token	NOUN
bracis-28417	64	9	,	,	PUNCT
bracis-28417	64	10	while	while	SCONJ
bracis-28417	64	11	the	the	DET
bracis-28417	64	12	llama	llama	NOUN
bracis-28417	64	13	tokenizer	tokenizer	NOUN
bracis-28417	64	14	produces	produce	VERB
bracis-28417	64	15	7.3	7.3	NUM
bracis-28417	64	16	billion	billion	NUM
bracis-28417	64	17	tokens	token	NOUN
bracis-28417	64	18	.	.	PUNCT
bracis-28417	65	1	the	the	DET
bracis-28417	65	2	discrepancy	discrepancy	NOUN
bracis-28417	65	3	in	in	ADP
bracis-28417	65	4	the	the	DET
bracis-28417	65	5	total	total	ADJ
bracis-28417	65	6	number	number	NOUN
bracis-28417	65	7	of	of	ADP
bracis-28417	65	8	tokens	token	NOUN
bracis-28417	65	9	is	be	AUX
bracis-28417	65	10	primarily	primarily	ADV
bracis-28417	65	11	due	due	ADJ
bracis-28417	65	12	to	to	ADP
bracis-28417	65	13	the	the	DET
bracis-28417	65	14	different	different	ADJ
bracis-28417	65	15	tokenization	tokenization	NOUN
bracis-28417	65	16	strategies	strategy	NOUN
bracis-28417	65	17	each	each	DET
bracis-28417	65	18	model	model	NOUN
bracis-28417	65	19	employs	employ	VERB
bracis-28417	65	20	,	,	PUNCT
bracis-28417	65	21	byte	byte	NOUN
bracis-28417	65	22	-	-	PUNCT
bracis-28417	65	23	level	level	NOUN
bracis-28417	65	24	bpe	bpe	PROPN
bracis-28417	65	25	and	and	CCONJ
bracis-28417	65	26	bpe	bpe	PROPN
bracis-28417	65	27	based	base	VERB
bracis-28417	65	28	on	on	ADP
bracis-28417	65	29	sentencepiece	sentencepiece	PROPN
bracis-28417	65	30	 	 	SPACE
bracis-28417	66	1	[	[	X
bracis-28417	66	2	23	23	NUM
bracis-28417	66	3	]	]	PUNCT
bracis-28417	66	4	,	,	PUNCT
bracis-28417	66	5	respectively	respectively	ADV
bracis-28417	66	6	along	along	ADP
bracis-28417	66	7	with	with	ADP
bracis-28417	66	8	the	the	DET
bracis-28417	66	9	variation	variation	NOUN
bracis-28417	66	10	of	of	ADP
bracis-28417	66	11	the	the	DET
bracis-28417	66	12	vocabularies	vocabulary	NOUN
bracis-28417	66	13	used	use	VERB
bracis-28417	66	14	by	by	ADP
bracis-28417	66	15	each	each	DET
bracis-28417	66	16	tokenizer	tokenizer	NOUN
bracis-28417	66	17	.	.	PUNCT
bracis-28417	67	1	we	we	PRON
bracis-28417	67	2	extended	extend	VERB
bracis-28417	67	3	the	the	DET
bracis-28417	67	4	training	training	NOUN
bracis-28417	67	5	of	of	ADP
bracis-28417	67	6	three	three	NUM
bracis-28417	67	7	models	model	NOUN
bracis-28417	67	8	llama	llama	PROPN
bracis-28417	67	9	7b	7b	NOUN
bracis-28417	67	10	and	and	CCONJ
bracis-28417	67	11	65b	65b	NUM
bracis-28417	67	12	 	 	SPACE
bracis-28417	67	13	[	[	PUNCT
bracis-28417	67	14	63	63	NUM
bracis-28417	67	15	]	]	PUNCT
bracis-28417	67	16	as	as	ADV
bracis-28417	67	17	well	well	ADV
bracis-28417	67	18	as	as	ADP
bracis-28417	67	19	gpt	gpt	NOUN
bracis-28417	67	20	-	-	PUNCT
bracis-28417	67	21	j	j	NOUN
bracis-28417	67	22	 	 	SPACE
bracis-28417	68	1	[	[	X
bracis-28417	68	2	66	66	NUM
bracis-28417	68	3	]	]	PUNCT
bracis-28417	68	4	originally	originally	ADV
bracis-28417	68	5	trained	train	VERB
bracis-28417	68	6	on	on	ADP
bracis-28417	68	7	english	english	ADJ
bracis-28417	68	8	-	-	PUNCT
bracis-28417	68	9	centric	centric	NOUN
bracis-28417	68	10	corpora	corpora	NOUN
bracis-28417	68	11	,	,	PUNCT
bracis-28417	68	12	on	on	ADP
bracis-28417	68	13	portuguese	portuguese	ADJ
bracis-28417	68	14	texts	text	NOUN
bracis-28417	68	15	;	;	PUNCT
bracis-28417	68	16	these	these	DET
bracis-28417	68	17	further	further	ADJ
bracis-28417	68	18	pretrained	pretraine	VERB
bracis-28417	68	19	models	model	NOUN
bracis-28417	68	20	from	from	ADP
bracis-28417	68	21	llama	llama	NOUN
bracis-28417	68	22	are	be	AUX
bracis-28417	68	23	denoted	denote	VERB
bracis-28417	68	24	as	as	ADP
bracis-28417	68	25	sabiá	sabiá	NOUN
bracis-28417	68	26	,	,	PUNCT
bracis-28417	68	27	while	while	SCONJ
bracis-28417	68	28	the	the	DET
bracis-28417	68	29	one	one	NOUN
bracis-28417	68	30	derived	derive	VERB
bracis-28417	68	31	from	from	ADP
bracis-28417	68	32	gpt	gpt	NOUN
bracis-28417	68	33	-	-	PUNCT
bracis-28417	68	34	j	j	PROPN
bracis-28417	68	35	is	be	AUX
bracis-28417	68	36	referred	refer	VERB
bracis-28417	68	37	to	to	ADP
bracis-28417	68	38	as	as	SCONJ
bracis-28417	68	39	sabiá-j.footnote	sabiá-j.footnote	PROPN
bracis-28417	68	40	3	3	NUM
bracis-28417	68	41	3.2	3.2	NUM
bracis-28417	68	42	sabiá	sabiá	NOUN
bracis-28417	68	43	models	model	NOUN
bracis-28417	68	44	the	the	DET
bracis-28417	68	45	llama	llama	PROPN
bracis-28417	68	46	7b	7b	NOUN
bracis-28417	68	47	and	and	CCONJ
bracis-28417	68	48	65b	65b	NUM
bracis-28417	68	49	models	model	NOUN
bracis-28417	68	50	are	be	AUX
bracis-28417	68	51	decoder	decoder	NOUN
bracis-28417	68	52	-	-	PUNCT
bracis-28417	68	53	only	only	ADV
bracis-28417	68	54	transformer	transformer	NOUN
bracis-28417	68	55	models	model	NOUN
bracis-28417	68	56	 	 	SPACE
bracis-28417	69	1	[	[	X
bracis-28417	69	2	64	64	NUM
bracis-28417	69	3	]	]	PUNCT
bracis-28417	69	4	with	with	ADP
bracis-28417	69	5	a	a	DET
bracis-28417	69	6	similar	similar	ADJ
bracis-28417	69	7	architecture	architecture	NOUN
bracis-28417	69	8	to	to	ADP
bracis-28417	69	9	palm	palm	NOUN
bracis-28417	69	10	’s	’s	NOUN
bracis-28417	69	11	 	 	SPACE
bracis-28417	70	1	[	[	X
bracis-28417	70	2	10	10	NUM
bracis-28417	70	3	]	]	PUNCT
bracis-28417	70	4	.	.	PUNCT
bracis-28417	71	1	the	the	DET
bracis-28417	71	2	models	model	NOUN
bracis-28417	71	3	were	be	AUX
bracis-28417	71	4	trained	train	VERB
bracis-28417	71	5	using	use	VERB
bracis-28417	71	6	a	a	DET
bracis-28417	71	7	causal	causal	ADJ
bracis-28417	71	8	language	language	NOUN
bracis-28417	71	9	modeling	model	VERB
bracis-28417	71	10	objective	objective	NOUN
bracis-28417	71	11	on	on	ADP
bracis-28417	71	12	a	a	DET
bracis-28417	71	13	massive	massive	ADJ
bracis-28417	71	14	dataset	dataset	NOUN
bracis-28417	71	15	sourced	source	VERB
bracis-28417	71	16	from	from	ADP
bracis-28417	71	17	webpages	webpage	NOUN
bracis-28417	71	18	,	,	PUNCT
bracis-28417	71	19	code	code	NOUN
bracis-28417	71	20	,	,	PUNCT
bracis-28417	71	21	books	book	NOUN
bracis-28417	71	22	,	,	PUNCT
bracis-28417	71	23	and	and	CCONJ
bracis-28417	71	24	scientific	scientific	ADJ
bracis-28417	71	25	papers	paper	NOUN
bracis-28417	71	26	.	.	PUNCT
bracis-28417	72	1	the	the	DET
bracis-28417	72	2	7b	7b	PROPN
bracis-28417	72	3	model	model	NOUN
bracis-28417	72	4	was	be	AUX
bracis-28417	72	5	trained	train	VERB
bracis-28417	72	6	on	on	ADP
bracis-28417	72	7	1	1	NUM
bracis-28417	72	8	trillion	trillion	NUM
bracis-28417	72	9	tokens	token	NOUN
bracis-28417	72	10	and	and	CCONJ
bracis-28417	72	11	the	the	DET
bracis-28417	72	12	65b	65b	ADJ
bracis-28417	72	13	model	model	NOUN
bracis-28417	72	14	was	be	AUX
bracis-28417	72	15	trained	train	VERB
bracis-28417	72	16	on	on	ADP
bracis-28417	72	17	1.4	1.4	NUM
bracis-28417	72	18	trillion	trillion	NUM
bracis-28417	72	19	tokens	token	NOUN
bracis-28417	72	20	.	.	PUNCT
bracis-28417	73	1	while	while	SCONJ
bracis-28417	73	2	the	the	DET
bracis-28417	73	3	majority	majority	NOUN
bracis-28417	73	4	of	of	ADP
bracis-28417	73	5	the	the	DET
bracis-28417	73	6	corpus	corpus	NOUN
bracis-28417	73	7	is	be	AUX
bracis-28417	73	8	in	in	ADP
bracis-28417	73	9	english	english	PROPN
bracis-28417	73	10	,	,	PUNCT
bracis-28417	73	11	it	it	PRON
bracis-28417	73	12	also	also	ADV
bracis-28417	73	13	includes	include	VERB
bracis-28417	73	14	an	an	DET
bracis-28417	73	15	unspecified	unspecified	ADJ
bracis-28417	73	16	amount	amount	NOUN
bracis-28417	73	17	of	of	ADP
bracis-28417	73	18	portuguese	portuguese	ADJ
bracis-28417	73	19	text	text	NOUN
bracis-28417	73	20	.	.	PUNCT
bracis-28417	74	1	starting	start	VERB
bracis-28417	74	2	from	from	ADP
bracis-28417	74	3	the	the	DET
bracis-28417	74	4	llama	llama	NOUN
bracis-28417	74	5	weights	weight	NOUN
bracis-28417	74	6	,	,	PUNCT
bracis-28417	74	7	we	we	PRON
bracis-28417	74	8	train	train	VERB
bracis-28417	74	9	the	the	DET
bracis-28417	74	10	sabiá	sabiá	NOUN
bracis-28417	74	11	models	model	NOUN
bracis-28417	74	12	on	on	ADP
bracis-28417	74	13	our	our	PRON
bracis-28417	74	14	portuguese	portuguese	ADJ
bracis-28417	74	15	dataset	dataset	NOUN
bracis-28417	74	16	(	(	PUNCT
bracis-28417	74	17	see	see	VERB
bracis-28417	74	18	sect	sect	NOUN
bracis-28417	74	19	.	.	PUNCT
bracis-28417	74	20	 	 	SPACE
bracis-28417	74	21	3.1	3.1	NUM
bracis-28417	74	22	)	)	PUNCT
bracis-28417	74	23	using	use	VERB
bracis-28417	74	24	the	the	DET
bracis-28417	74	25	t5x	t5x	NOUN
bracis-28417	74	26	and	and	CCONJ
bracis-28417	74	27	seqio	seqio	NOUN
bracis-28417	74	28	frameworks	framework	NOUN
bracis-28417	74	29	 	 	SPACE
bracis-28417	75	1	[	[	X
bracis-28417	75	2	48	48	NUM
bracis-28417	75	3	]	]	PUNCT
bracis-28417	75	4	.	.	PUNCT
bracis-28417	76	1	adhering	adhere	VERB
bracis-28417	76	2	closely	closely	ADV
bracis-28417	76	3	to	to	ADP
bracis-28417	76	4	the	the	DET
bracis-28417	76	5	hyperparameters	hyperparameter	NOUN
bracis-28417	76	6	used	use	VERB
bracis-28417	76	7	by	by	ADP
bracis-28417	76	8	palm	palm	NOUN
bracis-28417	76	9	,	,	PUNCT
bracis-28417	76	10	we	we	PRON
bracis-28417	76	11	use	use	VERB
bracis-28417	76	12	the	the	DET
bracis-28417	76	13	adafactor	adafactor	NOUN
bracis-28417	76	14	optimizer	optimizer	NOUN
bracis-28417	76	15	 	 	SPACE
bracis-28417	77	1	[	[	X
bracis-28417	77	2	55	55	NUM
bracis-28417	77	3	]	]	PUNCT
bracis-28417	77	4	without	without	ADP
bracis-28417	77	5	factorization	factorization	NOUN
bracis-28417	77	6	,	,	PUNCT
bracis-28417	77	7	a	a	DET
bracis-28417	77	8	first	first	ADJ
bracis-28417	77	9	-	-	PUNCT
bracis-28417	77	10	order	order	NOUN
bracis-28417	77	11	momentum	momentum	NOUN
bracis-28417	77	12	\(\beta	\(\beta	NOUN
bracis-28417	77	13	_	_	NOUN
bracis-28417	77	14	1	1	NUM
bracis-28417	77	15	=	=	SYM
bracis-28417	77	16	0.9\	0.9\	NOUN
bracis-28417	77	17	)	)	PUNCT
bracis-28417	77	18	,	,	PUNCT
bracis-28417	77	19	and	and	CCONJ
bracis-28417	77	20	a	a	DET
bracis-28417	77	21	second	second	ADJ
bracis-28417	77	22	-	-	PUNCT
bracis-28417	77	23	order	order	NOUN
bracis-28417	77	24	momentum	momentum	NOUN
bracis-28417	77	25	\(\beta	\(\beta	NOUN
bracis-28417	77	26	_	_	NOUN
bracis-28417	77	27	2	2	NUM
bracis-28417	77	28	=	=	SYM
bracis-28417	77	29	1	1	NUM
bracis-28417	77	30	k^{-0.8}\	k^{-0.8}\	PROPN
bracis-28417	77	31	)	)	PUNCT
bracis-28417	77	32	,	,	PUNCT
bracis-28417	77	33	where	where	SCONJ
bracis-28417	77	34	k	k	PROPN
bracis-28417	77	35	represents	represent	VERB
bracis-28417	77	36	the	the	DET
bracis-28417	77	37	step	step	NOUN
bracis-28417	77	38	number	number	NOUN
bracis-28417	77	39	.	.	PUNCT
bracis-28417	78	1	we	we	PRON
bracis-28417	78	2	apply	apply	VERB
bracis-28417	78	3	global	global	ADJ
bracis-28417	78	4	norm	norm	NOUN
bracis-28417	78	5	clipping	clip	VERB
bracis-28417	78	6	at	at	ADP
bracis-28417	78	7	1.0	1.0	NUM
bracis-28417	78	8	and	and	CCONJ
bracis-28417	78	9	dynamic	dynamic	ADJ
bracis-28417	78	10	weight	weight	NOUN
bracis-28417	78	11	decay	decay	NOUN
bracis-28417	78	12	of	of	ADP
bracis-28417	78	13	\(lr^2\	\(lr^2\	NOUN
bracis-28417	78	14	)	)	PUNCT
bracis-28417	78	15	,	,	PUNCT
bracis-28417	78	16	with	with	ADP
bracis-28417	78	17	lr	lr	NOUN
bracis-28417	78	18	denoting	denote	VERB
bracis-28417	78	19	the	the	DET
bracis-28417	78	20	current	current	ADJ
bracis-28417	78	21	learning	learning	NOUN
bracis-28417	78	22	rate	rate	NOUN
bracis-28417	78	23	.	.	PUNCT
bracis-28417	79	1	besides	besides	SCONJ
bracis-28417	79	2	the	the	DET
bracis-28417	79	3	standard	standard	ADJ
bracis-28417	79	4	causal	causal	ADJ
bracis-28417	79	5	language	language	NOUN
bracis-28417	79	6	modeling	modeling	NOUN
bracis-28417	79	7	loss	loss	NOUN
bracis-28417	79	8	,	,	PUNCT
bracis-28417	79	9	we	we	PRON
bracis-28417	79	10	use	use	VERB
bracis-28417	79	11	an	an	DET
bracis-28417	79	12	auxiliary	auxiliary	ADJ
bracis-28417	79	13	loss	loss	NOUN
bracis-28417	79	14	of	of	ADP
bracis-28417	79	15	\(10^{-4	\(10^{-4	NOUN
bracis-28417	79	16	}	}	PUNCT
bracis-28417	79	17	\log	\log	PROPN
bracis-28417	79	18	^2	^2	PUNCT
bracis-28417	79	19	(	(	PUNCT
bracis-28417	79	20	\sum	\sum	NOUN
bracis-28417	79	21	_	_	PUNCT
bracis-28417	79	22	i	i	PRON
bracis-28417	79	23	e^{z_i})\	e^{z_i})\	PROPN
bracis-28417	79	24	)	)	PUNCT
bracis-28417	79	25	,	,	PUNCT
bracis-28417	79	26	where	where	SCONJ
bracis-28417	79	27	z	z	NOUN
bracis-28417	79	28	are	be	AUX
bracis-28417	79	29	the	the	DET
bracis-28417	79	30	logits	logit	NOUN
bracis-28417	79	31	,	,	PUNCT
bracis-28417	79	32	to	to	PART
bracis-28417	79	33	decrease	decrease	VERB
bracis-28417	79	34	the	the	DET
bracis-28417	79	35	likelihood	likelihood	NOUN
bracis-28417	79	36	of	of	ADP
bracis-28417	79	37	loss	loss	NOUN
bracis-28417	79	38	spikes	spike	NOUN
bracis-28417	79	39	at	at	ADP
bracis-28417	79	40	the	the	DET
bracis-28417	79	41	65b	65b	NUM
bracis-28417	79	42	-	-	PUNCT
bracis-28417	79	43	parameter	parameter	NOUN
bracis-28417	79	44	scale	scale	NOUN
bracis-28417	79	45	.	.	PUNCT
bracis-28417	80	1	the	the	DET
bracis-28417	80	2	learning	learning	NOUN
bracis-28417	80	3	rate	rate	NOUN
bracis-28417	80	4	is	be	AUX
bracis-28417	80	5	linearly	linearly	ADV
bracis-28417	80	6	increased	increase	VERB
bracis-28417	80	7	from	from	ADP
bracis-28417	80	8	0	0	NUM
bracis-28417	80	9	to	to	ADP
bracis-28417	80	10	1e-3	1e-3	PROPN
bracis-28417	80	11	over	over	ADP
bracis-28417	80	12	the	the	DET
bracis-28417	80	13	initial	initial	ADJ
bracis-28417	80	14	1,000	1,000	NUM
bracis-28417	80	15	steps	step	NOUN
bracis-28417	80	16	,	,	PUNCT
bracis-28417	80	17	followed	follow	VERB
bracis-28417	80	18	by	by	ADP
bracis-28417	80	19	a	a	DET
bracis-28417	80	20	constant	constant	ADJ
bracis-28417	80	21	learning	learning	NOUN
bracis-28417	80	22	rate	rate	NOUN
bracis-28417	80	23	of	of	ADP
bracis-28417	80	24	1e-3	1e-3	PROPN
bracis-28417	80	25	for	for	ADP
bracis-28417	80	26	an	an	DET
bracis-28417	80	27	additional	additional	ADJ
bracis-28417	80	28	9,000	9,000	NUM
bracis-28417	80	29	steps	step	NOUN
bracis-28417	80	30	.	.	PUNCT
bracis-28417	81	1	the	the	DET
bracis-28417	81	2	models	model	NOUN
bracis-28417	81	3	were	be	AUX
bracis-28417	81	4	trained	train	VERB
bracis-28417	81	5	on	on	ADP
bracis-28417	81	6	a	a	DET
bracis-28417	81	7	tpu	tpu	NOUN
bracis-28417	81	8	v2	v2	PROPN
bracis-28417	81	9	-	-	PUNCT
bracis-28417	81	10	512	512	NUM
bracis-28417	81	11	,	,	PUNCT
bracis-28417	81	12	using	use	VERB
bracis-28417	81	13	batches	batch	NOUN
bracis-28417	81	14	of	of	ADP
bracis-28417	81	15	512	512	NUM
bracis-28417	81	16	sequences	sequence	NOUN
bracis-28417	81	17	,	,	PUNCT
bracis-28417	81	18	each	each	PRON
bracis-28417	81	19	containing	contain	VERB
bracis-28417	81	20	2048	2048	NUM
bracis-28417	81	21	tokens	token	NOUN
bracis-28417	81	22	.	.	PUNCT
bracis-28417	82	1	we	we	PRON
bracis-28417	82	2	utilized	utilize	VERB
bracis-28417	82	3	gradient	gradient	NOUN
bracis-28417	82	4	checkpointing	checkpointing	NOUN
bracis-28417	82	5	,	,	PUNCT
bracis-28417	82	6	also	also	ADV
bracis-28417	82	7	known	know	VERB
bracis-28417	82	8	as	as	ADP
bracis-28417	82	9	rematerialization	rematerialization	NOUN
bracis-28417	82	10	,	,	PUNCT
bracis-28417	82	11	to	to	PART
bracis-28417	82	12	enable	enable	VERB
bracis-28417	82	13	the	the	DET
bracis-28417	82	14	use	use	NOUN
bracis-28417	82	15	of	of	ADP
bracis-28417	82	16	larger	large	ADJ
bracis-28417	82	17	batches	batch	NOUN
bracis-28417	82	18	,	,	PUNCT
bracis-28417	82	19	thereby	thereby	ADV
bracis-28417	82	20	increasing	increase	VERB
bracis-28417	82	21	tpu	tpu	NOUN
bracis-28417	82	22	utilization	utilization	NOUN
bracis-28417	82	23	.	.	PUNCT
bracis-28417	83	1	for	for	ADP
bracis-28417	83	2	the	the	DET
bracis-28417	83	3	7b	7b	PROPN
bracis-28417	83	4	model	model	NOUN
bracis-28417	83	5	,	,	PUNCT
bracis-28417	83	6	this	this	DET
bracis-28417	83	7	configuration	configuration	NOUN
bracis-28417	83	8	results	result	VERB
bracis-28417	83	9	in	in	ADP
bracis-28417	83	10	a	a	DET
bracis-28417	83	11	throughput	throughput	NOUN
bracis-28417	83	12	of	of	ADP
bracis-28417	83	13	124,000	124,000	NUM
bracis-28417	83	14	tokens	token	NOUN
bracis-28417	83	15	/	/	SYM
bracis-28417	83	16	sec	sec	PROPN
bracis-28417	83	17	,	,	PUNCT
bracis-28417	83	18	corresponding	correspond	VERB
bracis-28417	83	19	to	to	ADP
bracis-28417	83	20	a	a	DET
bracis-28417	83	21	model	model	NOUN
bracis-28417	83	22	flops	flop	NOUN
bracis-28417	83	23	utilization	utilization	NOUN
bracis-28417	83	24	(	(	PUNCT
bracis-28417	83	25	mfu	mfu	NOUN
bracis-28417	83	26	)	)	PUNCT
bracis-28417	83	27	 	 	SPACE
bracis-28417	84	1	[	[	X
bracis-28417	84	2	10	10	NUM
bracis-28417	84	3	]	]	PUNCT
bracis-28417	84	4	of	of	ADP
bracis-28417	84	5	45.2	45.2	NUM
bracis-28417	84	6	%	%	NOUN
bracis-28417	84	7	,	,	PUNCT
bracis-28417	84	8	excluding	exclude	VERB
bracis-28417	84	9	the	the	DET
bracis-28417	84	10	self	self	NOUN
bracis-28417	84	11	-	-	PUNCT
bracis-28417	84	12	attention	attention	NOUN
bracis-28417	84	13	operations	operation	NOUN
bracis-28417	84	14	.	.	PUNCT
bracis-28417	85	1	for	for	ADP
bracis-28417	85	2	the	the	DET
bracis-28417	85	3	65b	65b	ADJ
bracis-28417	85	4	model	model	NOUN
bracis-28417	85	5	,	,	PUNCT
bracis-28417	85	6	we	we	PRON
bracis-28417	85	7	achieve	achieve	VERB
bracis-28417	85	8	a	a	DET
bracis-28417	85	9	throughput	throughput	NOUN
bracis-28417	85	10	of	of	ADP
bracis-28417	85	11	14,000	14,000	NUM
bracis-28417	85	12	tokens	token	NOUN
bracis-28417	85	13	/	/	SYM
bracis-28417	85	14	sec	sec	PROPN
bracis-28417	85	15	,	,	PUNCT
bracis-28417	85	16	resulting	result	VERB
bracis-28417	85	17	in	in	ADP
bracis-28417	85	18	an	an	DET
bracis-28417	85	19	mfu	mfu	NOUN
bracis-28417	85	20	of	of	ADP
bracis-28417	85	21	47.4	47.4	NUM
bracis-28417	85	22	%	%	NOUN
bracis-28417	85	23	.	.	PUNCT
bracis-28417	86	1	the	the	DET
bracis-28417	86	2	resulting	result	VERB
bracis-28417	86	3	models	model	NOUN
bracis-28417	86	4	were	be	AUX
bracis-28417	86	5	trained	train	VERB
bracis-28417	86	6	on	on	ADP
bracis-28417	86	7	a	a	DET
bracis-28417	86	8	total	total	NOUN
bracis-28417	86	9	of	of	ADP
bracis-28417	86	10	10.4	10.4	NUM
bracis-28417	86	11	billion	billion	NUM
bracis-28417	86	12	tokens	token	NOUN
bracis-28417	86	13	,	,	PUNCT
bracis-28417	86	14	or	or	CCONJ
bracis-28417	86	15	1.52	1.52	NUM
bracis-28417	86	16	epochs	epoch	NOUN
bracis-28417	86	17	of	of	ADP
bracis-28417	86	18	the	the	DET
bracis-28417	86	19	portuguese	portuguese	ADJ
bracis-28417	86	20	dataset	dataset	NOUN
bracis-28417	86	21	.	.	PUNCT
bracis-28417	87	1	this	this	PRON
bracis-28417	87	2	equals	equal	VERB
bracis-28417	87	3	to	to	ADP
bracis-28417	87	4	10,000	10,000	NUM
bracis-28417	87	5	training	training	NOUN
bracis-28417	87	6	steps	step	NOUN
bracis-28417	87	7	,	,	PUNCT
bracis-28417	87	8	which	which	PRON
bracis-28417	87	9	is	be	AUX
bracis-28417	87	10	the	the	DET
bracis-28417	87	11	same	same	ADJ
bracis-28417	87	12	amount	amount	NOUN
bracis-28417	87	13	used	use	VERB
bracis-28417	87	14	to	to	PART
bracis-28417	87	15	train	train	VERB
bracis-28417	87	16	sabiá	sabiá	NOUN
bracis-28417	87	17	.	.	PUNCT
bracis-28417	88	1	we	we	PRON
bracis-28417	88	2	noticed	notice	VERB
bracis-28417	88	3	improvements	improvement	NOUN
bracis-28417	88	4	in	in	ADP
bracis-28417	88	5	few	few	ADJ
bracis-28417	88	6	-	-	PUNCT
bracis-28417	88	7	shot	shot	NOUN
bracis-28417	88	8	tasks	task	NOUN
bracis-28417	88	9	beyond	beyond	ADP
bracis-28417	88	10	one	one	NUM
bracis-28417	88	11	epoch	epoch	NOUN
bracis-28417	88	12	,	,	PUNCT
bracis-28417	88	13	which	which	PRON
bracis-28417	88	14	corroborates	corroborate	VERB
bracis-28417	88	15	results	result	NOUN
bracis-28417	88	16	from	from	ADP
bracis-28417	88	17	taylor	taylor	PROPN
bracis-28417	88	18	et	et	PROPN
bracis-28417	88	19	al	al	PROPN
bracis-28417	88	20	.	.	PUNCT
bracis-28417	88	21	 	 	SPACE
bracis-28417	89	1	[	[	X
bracis-28417	89	2	62	62	NUM
bracis-28417	89	3	]	]	PUNCT
bracis-28417	89	4	.	.	PUNCT
bracis-28417	90	1	however	however	ADV
bracis-28417	90	2	,	,	PUNCT
bracis-28417	90	3	due	due	ADP
bracis-28417	90	4	to	to	ADP
bracis-28417	90	5	the	the	DET
bracis-28417	90	6	high	high	ADJ
bracis-28417	90	7	costs	cost	NOUN
bracis-28417	90	8	of	of	ADP
bracis-28417	90	9	pretraining	pretraine	VERB
bracis-28417	90	10	,	,	PUNCT
bracis-28417	90	11	we	we	PRON
bracis-28417	90	12	did	do	AUX
bracis-28417	90	13	not	not	PART
bracis-28417	90	14	continue	continue	VERB
bracis-28417	90	15	training.footnote	training.footnote	NUM
bracis-28417	90	16	4	4	NUM
bracis-28417	90	17	3.3	3.3	NUM
bracis-28417	90	18	sabiá	sabiá	NOUN
bracis-28417	90	19	-	-	PUNCT
bracis-28417	90	20	j	j	PROPN
bracis-28417	90	21	the	the	DET
bracis-28417	90	22	gpt	gpt	PROPN
bracis-28417	90	23	-	-	PUNCT
bracis-28417	90	24	j	j	PROPN
bracis-28417	90	25	model	model	NOUN
bracis-28417	90	26	is	be	AUX
bracis-28417	90	27	a	a	DET
bracis-28417	90	28	6b	6b	NUM
bracis-28417	90	29	-	-	PUNCT
bracis-28417	90	30	parameter	parameter	NOUN
bracis-28417	90	31	decoder	decoder	NOUN
bracis-28417	90	32	-	-	PUNCT
bracis-28417	90	33	only	only	ADV
bracis-28417	90	34	transformer	transformer	NOUN
bracis-28417	90	35	model	model	NOUN
bracis-28417	90	36	whose	whose	DET
bracis-28417	90	37	architecture	architecture	NOUN
bracis-28417	90	38	and	and	CCONJ
bracis-28417	90	39	training	training	NOUN
bracis-28417	90	40	hyperparameters	hyperparameter	NOUN
bracis-28417	90	41	closely	closely	ADV
bracis-28417	90	42	follow	follow	VERB
bracis-28417	90	43	gpt-3	gpt-3	NOUN
bracis-28417	90	44	6.7b	6.7b	NOUN
bracis-28417	90	45	.	.	PUNCT
bracis-28417	91	1	the	the	DET
bracis-28417	91	2	main	main	ADJ
bracis-28417	91	3	differences	difference	NOUN
bracis-28417	91	4	reside	reside	VERB
bracis-28417	91	5	on	on	ADP
bracis-28417	91	6	computing	compute	VERB
bracis-28417	91	7	the	the	DET
bracis-28417	91	8	mlp	mlp	NOUN
bracis-28417	91	9	and	and	CCONJ
bracis-28417	91	10	self	self	NOUN
bracis-28417	91	11	-	-	PUNCT
bracis-28417	91	12	attention	attention	NOUN
bracis-28417	91	13	in	in	ADP
bracis-28417	91	14	parallel	parallel	NOUN
bracis-28417	91	15	,	,	PUNCT
bracis-28417	91	16	applying	apply	VERB
bracis-28417	91	17	attention	attention	NOUN
bracis-28417	91	18	head	head	NOUN
bracis-28417	91	19	with	with	ADP
bracis-28417	91	20	dimension	dimension	NOUN
bracis-28417	91	21	256	256	NUM
bracis-28417	91	22	(	(	PUNCT
bracis-28417	91	23	twice	twice	ADV
bracis-28417	91	24	larger	large	ADJ
bracis-28417	91	25	than	than	ADP
bracis-28417	91	26	gpt-3	gpt-3	NOUN
bracis-28417	91	27	6.7b	6.7b	NOUN
bracis-28417	91	28	)	)	PUNCT
bracis-28417	91	29	,	,	PUNCT
bracis-28417	91	30	and	and	CCONJ
bracis-28417	91	31	using	use	VERB
bracis-28417	91	32	rotary	rotary	ADJ
bracis-28417	91	33	positional	positional	ADJ
bracis-28417	91	34	embedding	embed	VERB
bracis-28417	91	35	(	(	PUNCT
bracis-28417	91	36	rope	rope	NOUN
bracis-28417	91	37	)	)	PUNCT
bracis-28417	91	38	 	 	SPACE
bracis-28417	92	1	[	[	X
bracis-28417	92	2	61	61	NUM
bracis-28417	92	3	]	]	PUNCT
bracis-28417	92	4	.	.	PUNCT
bracis-28417	93	1	gpt	gpt	PROPN
bracis-28417	93	2	-	-	PUNCT
bracis-28417	93	3	j	j	PROPN
bracis-28417	93	4	was	be	AUX
bracis-28417	93	5	trained	train	VERB
bracis-28417	93	6	on	on	ADP
bracis-28417	93	7	400b	400b	PROPN
bracis-28417	93	8	tokens	token	NOUN
bracis-28417	93	9	from	from	ADP
bracis-28417	93	10	the	the	DET
bracis-28417	93	11	pile	pile	ADJ
bracis-28417	93	12	dataset	dataset	NOUN
bracis-28417	93	13	 	 	SPACE
bracis-28417	94	1	[	[	X
bracis-28417	94	2	16	16	NUM
bracis-28417	94	3	]	]	PUNCT
bracis-28417	94	4	,	,	PUNCT
bracis-28417	94	5	whose	whose	DET
bracis-28417	94	6	97.4	97.4	NUM
bracis-28417	94	7	%	%	NOUN
bracis-28417	94	8	tokens	token	NOUN
bracis-28417	94	9	are	be	AUX
bracis-28417	94	10	in	in	ADP
bracis-28417	94	11	english	english	PROPN
bracis-28417	94	12	.	.	PUNCT
bracis-28417	95	1	we	we	PRON
bracis-28417	95	2	begin	begin	VERB
bracis-28417	95	3	training	train	VERB
bracis-28417	95	4	sabiá	sabiá	PROPN
bracis-28417	95	5	-	-	PUNCT
bracis-28417	95	6	j	j	PROPN
bracis-28417	95	7	from	from	ADP
bracis-28417	95	8	the	the	DET
bracis-28417	95	9	released	release	VERB
bracis-28417	95	10	gpt	gpt	NOUN
bracis-28417	95	11	-	-	PUNCT
bracis-28417	95	12	j	j	NOUN
bracis-28417	95	13	checkpoint	checkpoint	NOUN
bracis-28417	95	14	,	,	PUNCT
bracis-28417	95	15	footnote	footnote	VERB
bracis-28417	95	16	5	5	NUM
bracis-28417	95	17	using	use	VERB
bracis-28417	95	18	the	the	DET
bracis-28417	95	19	mesh	mesh	NOUN
bracis-28417	95	20	-	-	PUNCT
bracis-28417	95	21	transformer	transformer	NOUN
bracis-28417	95	22	-	-	PUNCT
bracis-28417	95	23	jax	jax	NOUN
bracis-28417	95	24	framework	framework	NOUN
bracis-28417	95	25	 	 	SPACE
bracis-28417	96	1	[	[	X
bracis-28417	96	2	65	65	NUM
bracis-28417	96	3	]	]	PUNCT
bracis-28417	96	4	and	and	CCONJ
bracis-28417	96	5	adamw	adamw	PROPN
bracis-28417	96	6	optimizer	optimizer	NOUN
bracis-28417	96	7	 	 	SPACE
bracis-28417	97	1	[	[	X
bracis-28417	97	2	30	30	NUM
bracis-28417	97	3	]	]	PUNCT
bracis-28417	97	4	with	with	ADP
bracis-28417	97	5	a	a	DET
bracis-28417	97	6	weight	weight	NOUN
bracis-28417	97	7	decay	decay	NOUN
bracis-28417	97	8	of	of	ADP
bracis-28417	97	9	0.1	0.1	NUM
bracis-28417	97	10	.	.	PUNCT
bracis-28417	98	1	we	we	PRON
bracis-28417	98	2	start	start	VERB
bracis-28417	98	3	the	the	DET
bracis-28417	98	4	pretraining	pretraine	VERB
bracis-28417	98	5	by	by	ADP
bracis-28417	98	6	warming	warm	VERB
bracis-28417	98	7	up	up	ADP
bracis-28417	98	8	the	the	DET
bracis-28417	98	9	learning	learning	NOUN
bracis-28417	98	10	rate	rate	NOUN
bracis-28417	98	11	until	until	ADP
bracis-28417	98	12	1.2e-5	1.2e-5	NUM
bracis-28417	98	13	over	over	ADP
bracis-28417	98	14	13,500	13,500	NUM
bracis-28417	98	15	steps	step	NOUN
bracis-28417	98	16	,	,	PUNCT
bracis-28417	98	17	followed	follow	VERB
bracis-28417	98	18	by	by	ADP
bracis-28417	98	19	a	a	DET
bracis-28417	98	20	cosine	cosine	NOUN
bracis-28417	98	21	annealing	anneal	VERB
bracis-28417	98	22	decay	decay	NOUN
bracis-28417	98	23	during	during	ADP
bracis-28417	98	24	135,518	135,518	NUM
bracis-28417	98	25	steps	step	NOUN
bracis-28417	98	26	until	until	ADP
bracis-28417	98	27	the	the	DET
bracis-28417	98	28	end	end	NOUN
bracis-28417	98	29	learning	learn	VERB
bracis-28417	98	30	rate	rate	NOUN
bracis-28417	98	31	of	of	ADP
bracis-28417	98	32	2.4e-6	2.4e-6	NUM
bracis-28417	98	33	,	,	PUNCT
bracis-28417	98	34	and	and	CCONJ
bracis-28417	98	35	kept	keep	VERB
bracis-28417	98	36	it	it	PRON
bracis-28417	98	37	constant	constant	ADJ
bracis-28417	98	38	from	from	ADP
bracis-28417	98	39	there	there	ADV
bracis-28417	98	40	on	on	ADP
bracis-28417	98	41	.	.	PUNCT
bracis-28417	99	1	we	we	PRON
bracis-28417	99	2	train	train	VERB
bracis-28417	99	3	on	on	ADP
bracis-28417	99	4	a	a	DET
bracis-28417	99	5	tpu	tpu	NOUN
bracis-28417	99	6	v3	v3	PROPN
bracis-28417	99	7	-	-	PUNCT
bracis-28417	99	8	8	8	NUM
bracis-28417	99	9	using	use	VERB
bracis-28417	99	10	an	an	DET
bracis-28417	99	11	effective	effective	ADJ
bracis-28417	99	12	batch	batch	NOUN
bracis-28417	99	13	size	size	NOUN
bracis-28417	99	14	of	of	ADP
bracis-28417	99	15	32	32	NUM
bracis-28417	99	16	sequences	sequence	NOUN
bracis-28417	99	17	of	of	ADP
bracis-28417	99	18	2048	2048	NUM
bracis-28417	99	19	tokens	token	NOUN
bracis-28417	99	20	.	.	PUNCT
bracis-28417	100	1	this	this	DET
bracis-28417	100	2	results	result	VERB
bracis-28417	100	3	in	in	ADP
bracis-28417	100	4	a	a	DET
bracis-28417	100	5	throughput	throughput	NOUN
bracis-28417	100	6	of	of	ADP
bracis-28417	100	7	5,200	5,200	NUM
bracis-28417	100	8	tokens	token	NOUN
bracis-28417	100	9	/	/	SYM
bracis-28417	100	10	sec	sec	PROPN
bracis-28417	100	11	,	,	PUNCT
bracis-28417	100	12	corresponding	correspond	VERB
bracis-28417	100	13	to	to	ADP
bracis-28417	100	14	a	a	DET
bracis-28417	100	15	mfu	mfu	NOUN
bracis-28417	100	16	of	of	ADP
bracis-28417	100	17	44.5	44.5	NUM
bracis-28417	100	18	%	%	NOUN
bracis-28417	100	19	without	without	ADP
bracis-28417	100	20	self	self	NOUN
bracis-28417	100	21	-	-	PUNCT
bracis-28417	100	22	attention	attention	NOUN
bracis-28417	100	23	.	.	PUNCT
bracis-28417	101	1	the	the	DET
bracis-28417	101	2	model	model	NOUN
bracis-28417	101	3	was	be	AUX
bracis-28417	101	4	trained	train	VERB
bracis-28417	101	5	for	for	ADP
bracis-28417	101	6	18	18	NUM
bracis-28417	101	7	 	 	SPACE
bracis-28417	101	8	d	d	NOUN
bracis-28417	101	9	on	on	ADP
bracis-28417	101	10	7.8b	7.8b	PROPN
bracis-28417	101	11	tokens	token	NOUN
bracis-28417	101	12	,	,	PUNCT
bracis-28417	101	13	or	or	CCONJ
bracis-28417	101	14	one	one	NUM
bracis-28417	101	15	epoch	epoch	NOUN
bracis-28417	101	16	of	of	ADP
bracis-28417	101	17	the	the	DET
bracis-28417	101	18	portuguese	portuguese	NOUN
bracis-28417	101	19	dataset.footnote	dataset.footnote	VERB
bracis-28417	101	20	6	6	NUM
bracis-28417	101	21	4	4	NUM
bracis-28417	101	22	evaluation	evaluation	NOUN
bracis-28417	101	23	on	on	ADP
bracis-28417	101	24	 	 	SPACE
bracis-28417	101	25	poeta	poeta	NOUN
bracis-28417	101	26	we	we	PRON
bracis-28417	101	27	evaluate	evaluate	VERB
bracis-28417	101	28	the	the	DET
bracis-28417	101	29	sabiá	sabiá	NOUN
bracis-28417	101	30	models	model	NOUN
bracis-28417	101	31	on	on	ADP
bracis-28417	101	32	the	the	DET
bracis-28417	101	33	portuguese	portuguese	ADJ
bracis-28417	101	34	evaluation	evaluation	NOUN
bracis-28417	101	35	tasks	task	NOUN
bracis-28417	101	36	(	(	PUNCT
bracis-28417	101	37	poeta	poeta	NOUN
bracis-28417	101	38	)	)	PUNCT
bracis-28417	101	39	benchmark	benchmark	NOUN
bracis-28417	101	40	,	,	PUNCT
bracis-28417	101	41	which	which	PRON
bracis-28417	101	42	comprises	comprise	VERB
bracis-28417	101	43	14	14	NUM
bracis-28417	101	44	downstream	downstream	ADJ
bracis-28417	101	45	nlp	nlp	NOUN
bracis-28417	101	46	datasets	dataset	NOUN
bracis-28417	101	47	in	in	ADP
bracis-28417	101	48	portuguese	portuguese	PROPN
bracis-28417	101	49	:	:	PUNCT
bracis-28417	101	50	assin	assin	PROPN
bracis-28417	101	51	2	2	NUM
bracis-28417	101	52	rte	rte	NOUN
bracis-28417	101	53	and	and	CCONJ
bracis-28417	101	54	sts	st	VERB
bracis-28417	101	55	 	 	SPACE
bracis-28417	102	1	[	[	X
bracis-28417	102	2	47	47	NUM
bracis-28417	102	3	]	]	PUNCT
bracis-28417	102	4	,	,	PUNCT
bracis-28417	102	5	enem	enem	PROPN
bracis-28417	102	6	challenge	challenge	NOUN
bracis-28417	102	7	 	 	SPACE
bracis-28417	103	1	[	[	X
bracis-28417	103	2	57	57	NUM
bracis-28417	103	3	]	]	PUNCT
bracis-28417	103	4	,	,	PUNCT
bracis-28417	103	5	enem	enem	PROPN
bracis-28417	103	6	2022	2022	NUM
bracis-28417	103	7	 	 	SPACE
bracis-28417	104	1	[	[	X
bracis-28417	104	2	37	37	NUM
bracis-28417	104	3	]	]	PUNCT
bracis-28417	104	4	,	,	PUNCT
bracis-28417	104	5	faquad	faquad	ADV
bracis-28417	104	6	 	 	SPACE
bracis-28417	105	1	[	[	X
bracis-28417	105	2	53	53	NUM
bracis-28417	105	3	]	]	PUNCT
bracis-28417	105	4	,	,	PUNCT
bracis-28417	105	5	tweetsentbr	tweetsentbr	NOUN
bracis-28417	105	6	 	 	SPACE
bracis-28417	106	1	[	[	X
bracis-28417	106	2	5	5	NUM
bracis-28417	106	3	]	]	PUNCT
bracis-28417	106	4	,	,	PUNCT
bracis-28417	106	5	ag	ag	PROPN
bracis-28417	106	6	news	news	PROPN
bracis-28417	106	7	 	 	SPACE
bracis-28417	107	1	[	[	X
bracis-28417	107	2	75	75	NUM
bracis-28417	107	3	]	]	PUNCT
bracis-28417	107	4	,	,	PUNCT
bracis-28417	107	5	imdb	imdb	PROPN
bracis-28417	107	6	 	 	SPACE
bracis-28417	108	1	[	[	X
bracis-28417	108	2	31	31	NUM
bracis-28417	108	3	]	]	PUNCT
bracis-28417	108	4	,	,	PUNCT
bracis-28417	109	1	massive	massive	ADJ
bracis-28417	109	2	 	 	SPACE
bracis-28417	109	3	[	[	X
bracis-28417	109	4	14	14	NUM
bracis-28417	109	5	]	]	PUNCT
bracis-28417	109	6	,	,	PUNCT
bracis-28417	109	7	mkqa	mkqa	PROPN
bracis-28417	109	8	 	 	SPACE
bracis-28417	110	1	[	[	X
bracis-28417	110	2	29	29	NUM
bracis-28417	110	3	]	]	PUNCT
bracis-28417	110	4	,	,	PUNCT
bracis-28417	110	5	boolq	boolq	ADV
bracis-28417	110	6	 	 	SPACE
bracis-28417	111	1	[	[	X
bracis-28417	111	2	11	11	NUM
bracis-28417	111	3	]	]	PUNCT
bracis-28417	111	4	,	,	PUNCT
bracis-28417	111	5	sst2	sst2	NOUN
bracis-28417	111	6	 	 	SPACE
bracis-28417	111	7	[	[	X
bracis-28417	111	8	58	58	NUM
bracis-28417	111	9	]	]	PUNCT
bracis-28417	111	10	,	,	PUNCT
bracis-28417	111	11	wsc	wsc	PROPN
bracis-28417	111	12	 	 	SPACE
bracis-28417	112	1	[	[	X
bracis-28417	112	2	33	33	NUM
bracis-28417	112	3	]	]	PUNCT
bracis-28417	112	4	,	,	PUNCT
bracis-28417	112	5	and	and	CCONJ
bracis-28417	112	6	bluex	bluex	NOUN
bracis-28417	112	7	 	 	SPACE
bracis-28417	112	8	[	[	X
bracis-28417	112	9	1	1	NUM
bracis-28417	112	10	]	]	PUNCT
bracis-28417	112	11	.	.	PUNCT
bracis-28417	113	1	half	half	NOUN
bracis-28417	113	2	of	of	ADP
bracis-28417	113	3	them	they	PRON
bracis-28417	113	4	(	(	PUNCT
bracis-28417	113	5	assin	assin	PROPN
bracis-28417	113	6	2	2	NUM
bracis-28417	113	7	rte	rte	NOUN
bracis-28417	113	8	and	and	CCONJ
bracis-28417	113	9	sts	st	NOUN
bracis-28417	113	10	,	,	PUNCT
bracis-28417	113	11	bluex	bluex	PROPN
bracis-28417	113	12	,	,	PUNCT
bracis-28417	113	13	enem	enem	PROPN
bracis-28417	113	14	challenge	challenge	NOUN
bracis-28417	113	15	,	,	PUNCT
bracis-28417	113	16	enem	enem	PROPN
bracis-28417	113	17	2022	2022	NUM
bracis-28417	113	18	,	,	PUNCT
bracis-28417	113	19	faquad	faquad	ADJ
bracis-28417	113	20	,	,	PUNCT
bracis-28417	113	21	and	and	CCONJ
bracis-28417	113	22	tweetsentbr	tweetsentbr	NOUN
bracis-28417	113	23	)	)	PUNCT
bracis-28417	113	24	were	be	AUX
bracis-28417	113	25	originally	originally	ADV
bracis-28417	113	26	written	write	VERB
bracis-28417	113	27	in	in	ADP
bracis-28417	113	28	portuguese	portuguese	PROPN
bracis-28417	113	29	,	,	PUNCT
bracis-28417	113	30	and	and	CCONJ
bracis-28417	113	31	the	the	DET
bracis-28417	113	32	remaining	remain	VERB
bracis-28417	113	33	ones	one	NOUN
bracis-28417	113	34	were	be	AUX
bracis-28417	113	35	either	either	CCONJ
bracis-28417	113	36	manually	manually	ADV
bracis-28417	113	37	or	or	CCONJ
bracis-28417	113	38	automatically	automatically	ADV
bracis-28417	113	39	translated	translate	VERB
bracis-28417	113	40	into	into	ADP
bracis-28417	113	41	portuguese	portuguese	PROPN
bracis-28417	113	42	from	from	ADP
bracis-28417	113	43	their	their	PRON
bracis-28417	113	44	originals	original	NOUN
bracis-28417	113	45	in	in	ADP
bracis-28417	113	46	english	english	PROPN
bracis-28417	113	47	.	.	PUNCT
bracis-28417	114	1	we	we	PRON
bracis-28417	114	2	refer	refer	VERB
bracis-28417	114	3	to	to	ADP
bracis-28417	114	4	the	the	DET
bracis-28417	114	5	first	first	ADJ
bracis-28417	114	6	group	group	NOUN
bracis-28417	114	7	as	as	ADP
bracis-28417	114	8	“	"	PUNCT
bracis-28417	114	9	native	native	ADJ
bracis-28417	114	10	”	"	PUNCT
bracis-28417	114	11	datasets	dataset	NOUN
bracis-28417	114	12	and	and	CCONJ
bracis-28417	114	13	the	the	DET
bracis-28417	114	14	second	second	ADJ
bracis-28417	114	15	group	group	NOUN
bracis-28417	114	16	as	as	ADP
bracis-28417	114	17	“	"	PUNCT
bracis-28417	114	18	translated	translate	VERB
bracis-28417	114	19	”	"	PUNCT
bracis-28417	114	20	datasets.footnote	datasets.footnote	PROPN
bracis-28417	114	21	7	7	NUM
bracis-28417	114	22	the	the	DET
bracis-28417	114	23	models	model	NOUN
bracis-28417	114	24	were	be	AUX
bracis-28417	114	25	evaluated	evaluate	VERB
bracis-28417	114	26	in	in	ADP
bracis-28417	114	27	a	a	DET
bracis-28417	114	28	few	few	ADJ
bracis-28417	114	29	-	-	PUNCT
bracis-28417	114	30	shot	shot	NOUN
bracis-28417	114	31	manner	manner	NOUN
bracis-28417	114	32	using	use	VERB
bracis-28417	114	33	the	the	DET
bracis-28417	114	34	maximum	maximum	ADJ
bracis-28417	114	35	number	number	NOUN
bracis-28417	114	36	of	of	ADP
bracis-28417	114	37	examples	example	NOUN
bracis-28417	114	38	that	that	PRON
bracis-28417	114	39	fits	fit	VERB
bracis-28417	114	40	into	into	ADP
bracis-28417	114	41	a	a	DET
bracis-28417	114	42	2048	2048	NUM
bracis-28417	114	43	-	-	PUNCT
bracis-28417	114	44	token	token	VERB
bracis-28417	114	45	context	context	NOUN
bracis-28417	114	46	for	for	ADP
bracis-28417	114	47	each	each	DET
bracis-28417	114	48	task	task	NOUN
bracis-28417	114	49	.	.	PUNCT
bracis-28417	115	1	we	we	PRON
bracis-28417	115	2	used	use	VERB
bracis-28417	115	3	the	the	DET
bracis-28417	115	4	gpt-2	gpt-2	NOUN
bracis-28417	115	5	tokenizer	tokenizer	NOUN
bracis-28417	115	6	as	as	ADP
bracis-28417	115	7	a	a	DET
bracis-28417	115	8	reference	reference	NOUN
bracis-28417	115	9	because	because	SCONJ
bracis-28417	115	10	it	it	PRON
bracis-28417	115	11	results	result	VERB
bracis-28417	115	12	in	in	ADP
bracis-28417	115	13	more	more	ADJ
bracis-28417	115	14	tokens	token	NOUN
bracis-28417	115	15	.	.	PUNCT
bracis-28417	116	1	this	this	PRON
bracis-28417	116	2	allowed	allow	VERB
bracis-28417	116	3	us	we	PRON
bracis-28417	116	4	to	to	PART
bracis-28417	116	5	comfortably	comfortably	ADV
bracis-28417	116	6	fit	fit	VERB
bracis-28417	116	7	prompts	prompt	NOUN
bracis-28417	116	8	tokenized	tokenize	VERB
bracis-28417	116	9	with	with	ADP
bracis-28417	116	10	other	other	ADJ
bracis-28417	116	11	tokenizers	tokenizer	NOUN
bracis-28417	116	12	.	.	PUNCT
bracis-28417	117	1	to	to	PART
bracis-28417	117	2	evaluate	evaluate	VERB
bracis-28417	117	3	the	the	DET
bracis-28417	117	4	models	model	NOUN
bracis-28417	117	5	,	,	PUNCT
bracis-28417	117	6	we	we	PRON
bracis-28417	117	7	manually	manually	ADV
bracis-28417	117	8	select	select	VERB
bracis-28417	117	9	a	a	DET
bracis-28417	117	10	set	set	NOUN
bracis-28417	117	11	of	of	ADP
bracis-28417	117	12	few	few	ADJ
bracis-28417	117	13	-	-	PUNCT
bracis-28417	117	14	shot	shot	NOUN
bracis-28417	117	15	examples	example	NOUN
bracis-28417	117	16	for	for	ADP
bracis-28417	117	17	each	each	DET
bracis-28417	117	18	dataset	dataset	NOUN
bracis-28417	117	19	on	on	ADP
bracis-28417	117	20	poeta	poeta	PROPN
bracis-28417	117	21	.	.	PUNCT
bracis-28417	118	1	depending	depend	VERB
bracis-28417	118	2	on	on	ADP
bracis-28417	118	3	the	the	DET
bracis-28417	118	4	dataset	dataset	NOUN
bracis-28417	118	5	,	,	PUNCT
bracis-28417	118	6	these	these	DET
bracis-28417	118	7	examples	example	NOUN
bracis-28417	118	8	are	be	AUX
bracis-28417	118	9	balanced	balance	VERB
bracis-28417	118	10	by	by	ADP
bracis-28417	118	11	class	class	NOUN
bracis-28417	118	12	(	(	PUNCT
bracis-28417	118	13	except	except	SCONJ
bracis-28417	118	14	for	for	ADP
bracis-28417	118	15	faquad	faquad	ADJ
bracis-28417	118	16	,	,	PUNCT
bracis-28417	118	17	bluex	bluex	PROPN
bracis-28417	118	18	,	,	PUNCT
bracis-28417	118	19	enem	enem	PROPN
bracis-28417	118	20	challenge	challenge	NOUN
bracis-28417	118	21	,	,	PUNCT
bracis-28417	118	22	enem	enem	PROPN
bracis-28417	118	23	2022	2022	NUM
bracis-28417	118	24	,	,	PUNCT
bracis-28417	118	25	mkqa	mkqa	NOUN
bracis-28417	118	26	,	,	PUNCT
bracis-28417	118	27	and	and	CCONJ
bracis-28417	118	28	wsc	wsc	PROPN
bracis-28417	118	29	)	)	PUNCT
bracis-28417	118	30	.	.	PUNCT
bracis-28417	119	1	for	for	ADP
bracis-28417	119	2	each	each	DET
bracis-28417	119	3	test	test	NOUN
bracis-28417	119	4	example	example	NOUN
bracis-28417	119	5	,	,	PUNCT
bracis-28417	119	6	the	the	DET
bracis-28417	119	7	prompts	prompt	NOUN
bracis-28417	119	8	are	be	AUX
bracis-28417	119	9	built	build	VERB
bracis-28417	119	10	with	with	ADP
bracis-28417	119	11	the	the	DET
bracis-28417	119	12	selected	select	VERB
bracis-28417	119	13	few	few	ADJ
bracis-28417	119	14	-	-	PUNCT
bracis-28417	119	15	shot	shot	NOUN
bracis-28417	119	16	examples	example	NOUN
bracis-28417	119	17	in	in	ADP
bracis-28417	119	18	alternating	alternate	VERB
bracis-28417	119	19	order	order	NOUN
bracis-28417	119	20	.	.	PUNCT
bracis-28417	120	1	each	each	DET
bracis-28417	120	2	task	task	NOUN
bracis-28417	120	3	on	on	ADP
bracis-28417	120	4	poeta	poeta	PROPN
bracis-28417	120	5	has	have	VERB
bracis-28417	120	6	a	a	DET
bracis-28417	120	7	particular	particular	ADJ
bracis-28417	120	8	instruction	instruction	NOUN
bracis-28417	120	9	that	that	PRON
bracis-28417	120	10	is	be	AUX
bracis-28417	120	11	placed	place	VERB
bracis-28417	120	12	at	at	ADP
bracis-28417	120	13	the	the	DET
bracis-28417	120	14	beginning	beginning	NOUN
bracis-28417	120	15	of	of	ADP
bracis-28417	120	16	the	the	DET
bracis-28417	120	17	prompt	prompt	NOUN
bracis-28417	120	18	.	.	PUNCT
bracis-28417	121	1	following	follow	VERB
bracis-28417	121	2	srivastava	srivastava	PROPN
bracis-28417	121	3	et	et	PROPN
bracis-28417	121	4	al	al	PROPN
bracis-28417	121	5	.	.	PUNCT
bracis-28417	121	6	 	 	SPACE
bracis-28417	122	1	[	[	X
bracis-28417	122	2	60	60	NUM
bracis-28417	122	3	]	]	PUNCT
bracis-28417	122	4	,	,	PUNCT
bracis-28417	122	5	we	we	PRON
bracis-28417	122	6	adopt	adopt	VERB
bracis-28417	122	7	the	the	DET
bracis-28417	122	8	normalized	normalize	VERB
bracis-28417	122	9	preferred	preferred	ADJ
bracis-28417	122	10	metric	metric	ADJ
bracis-28417	122	11	(	(	PUNCT
bracis-28417	122	12	npm	npm	PROPN
bracis-28417	122	13	)	)	PUNCT
bracis-28417	122	14	as	as	ADP
bracis-28417	122	15	our	our	PRON
bracis-28417	122	16	primary	primary	ADJ
bracis-28417	122	17	evaluation	evaluation	NOUN
bracis-28417	122	18	measure	measure	NOUN
bracis-28417	122	19	:	:	PUNCT
bracis-28417	122	20	$	$	SYM
bracis-28417	122	21	$	$	SYM
bracis-28417	122	22	\begin{aligned	\begin{aligne	VERB
bracis-28417	122	23	}	}	PUNCT
bracis-28417	122	24	\texttt	\texttt	PROPN
bracis-28417	122	25	{	{	PUNCT
bracis-28417	122	26	npm	npm	PROPN
bracis-28417	122	27	}	}	PUNCT
bracis-28417	122	28	=	=	SYM
bracis-28417	122	29	\frac{1}{n	\frac{1}{n	NOUN
bracis-28417	122	30	}	}	PUNCT
bracis-28417	122	31	\sum	\sum	NOUN
bracis-28417	122	32	_	_	PUNCT
bracis-28417	122	33	{	{	PUNCT
bracis-28417	122	34	i=1}^n	i=1}^n	NOUN
bracis-28417	122	35	100	100	NUM
bracis-28417	122	36	\times	\time	NOUN
bracis-28417	122	37	\frac{\texttt	\frac{\texttt	NUM
bracis-28417	122	38	{	{	PUNCT
bracis-28417	122	39	[	[	PUNCT
bracis-28417	122	40	raw	raw	ADJ
bracis-28417	122	41	preferred	preferred	ADJ
bracis-28417	122	42	metric]}_i	metric]}_i	NOUN
bracis-28417	123	1	\texttt	\texttt	PROPN
bracis-28417	123	2	{	{	PUNCT
bracis-28417	123	3	[	[	X
bracis-28417	123	4	random	random	ADJ
bracis-28417	123	5	score]}_i}{\texttt	score]}_i}{\texttt	NOUN
bracis-28417	123	6	{	{	PUNCT
bracis-28417	123	7	[	[	X
bracis-28417	123	8	high	high	ADJ
bracis-28417	123	9	score]}_i	score]}_i	NOUN
bracis-28417	124	1	\texttt	\texttt	ADP
bracis-28417	124	2	{	{	PUNCT
bracis-28417	124	3	[	[	X
bracis-28417	124	4	random	random	ADJ
bracis-28417	124	5	score]}_i	score]}_i	NOUN
bracis-28417	124	6	}	}	PUNCT
bracis-28417	124	7	\end{aligned}$$	\end{aligned}$$	X
bracis-28417	124	8	(	(	PUNCT
bracis-28417	124	9	1	1	NUM
bracis-28417	124	10	)	)	PUNCT
bracis-28417	124	11	where	where	SCONJ
bracis-28417	124	12	n	n	PRON
bracis-28417	124	13	is	be	AUX
bracis-28417	124	14	the	the	DET
bracis-28417	124	15	number	number	NOUN
bracis-28417	124	16	of	of	ADP
bracis-28417	124	17	evaluation	evaluation	NOUN
bracis-28417	124	18	datasets	dataset	NOUN
bracis-28417	124	19	,	,	PUNCT
bracis-28417	124	20	\(\texttt	\(\texttt	X
bracis-28417	124	21	{	{	PUNCT
bracis-28417	124	22	[	[	PUNCT
bracis-28417	124	23	raw	raw	ADJ
bracis-28417	124	24	preferred	preferred	ADJ
bracis-28417	124	25	metric]}_i\	metric]}_i\	NOUN
bracis-28417	124	26	)	)	PUNCT
bracis-28417	124	27	is	be	AUX
bracis-28417	124	28	the	the	DET
bracis-28417	124	29	score	score	NOUN
bracis-28417	124	30	obtained	obtain	VERB
bracis-28417	124	31	by	by	ADP
bracis-28417	124	32	the	the	DET
bracis-28417	124	33	model	model	NOUN
bracis-28417	124	34	on	on	ADP
bracis-28417	124	35	the	the	DET
bracis-28417	124	36	i	i	PROPN
bracis-28417	124	37	-	-	PUNCT
bracis-28417	124	38	th	th	X
bracis-28417	124	39	dataset	dataset	NOUN
bracis-28417	124	40	,	,	PUNCT
bracis-28417	124	41	\(\texttt	\(\texttt	NOUN
bracis-28417	124	42	{	{	PUNCT
bracis-28417	124	43	[	[	X
bracis-28417	124	44	random	random	ADJ
bracis-28417	124	45	score]}_i\	score]}_i\	NOUN
bracis-28417	124	46	)	)	PUNCT
bracis-28417	124	47	is	be	AUX
bracis-28417	124	48	the	the	DET
bracis-28417	124	49	score	score	NOUN
bracis-28417	124	50	of	of	ADP
bracis-28417	124	51	a	a	DET
bracis-28417	124	52	random	random	ADJ
bracis-28417	124	53	model	model	NOUN
bracis-28417	124	54	(	(	PUNCT
bracis-28417	124	55	e.g.	e.g.	ADV
bracis-28417	124	56	,	,	PUNCT
bracis-28417	124	57	50	50	NUM
bracis-28417	124	58	%	%	NOUN
bracis-28417	124	59	for	for	ADP
bracis-28417	124	60	a	a	DET
bracis-28417	124	61	binary	binary	ADJ
bracis-28417	124	62	classification	classification	NOUN
bracis-28417	124	63	task	task	NOUN
bracis-28417	124	64	)	)	PUNCT
bracis-28417	124	65	and	and	CCONJ
bracis-28417	124	66	\(\texttt	\(\texttt	NOUN
bracis-28417	124	67	{	{	PUNCT
bracis-28417	124	68	[	[	PUNCT
bracis-28417	124	69	high	high	ADJ
bracis-28417	124	70	score]}_i\	score]}_i\	NOUN
bracis-28417	124	71	)	)	PUNCT
bracis-28417	124	72	is	be	AUX
bracis-28417	124	73	the	the	DET
bracis-28417	124	74	highest	high	ADJ
bracis-28417	124	75	possible	possible	ADJ
bracis-28417	124	76	score	score	NOUN
bracis-28417	124	77	on	on	ADP
bracis-28417	124	78	that	that	DET
bracis-28417	124	79	dataset	dataset	NOUN
bracis-28417	124	80	,	,	PUNCT
bracis-28417	124	81	which	which	PRON
bracis-28417	124	82	is	be	AUX
bracis-28417	124	83	either	either	CCONJ
bracis-28417	124	84	1	1	NUM
bracis-28417	124	85	or	or	CCONJ
bracis-28417	124	86	100	100	NUM
bracis-28417	124	87	.	.	PUNCT
bracis-28417	125	1	the	the	DET
bracis-28417	125	2	preferred	preferred	ADJ
bracis-28417	125	3	metric	metric	ADJ
bracis-28417	125	4	and	and	CCONJ
bracis-28417	125	5	random	random	ADJ
bracis-28417	125	6	score	score	NOUN
bracis-28417	125	7	for	for	ADP
bracis-28417	125	8	each	each	DET
bracis-28417	125	9	dataset	dataset	NOUN
bracis-28417	125	10	are	be	AUX
bracis-28417	125	11	presented	present	VERB
bracis-28417	125	12	in	in	ADP
bracis-28417	125	13	table	table	NOUN
bracis-28417	125	14	 	 	SPACE
bracis-28417	125	15	1	1	NUM
bracis-28417	125	16	.	.	PUNCT
bracis-28417	126	1	the	the	DET
bracis-28417	126	2	rationale	rationale	NOUN
bracis-28417	126	3	behind	behind	ADP
bracis-28417	126	4	employing	employ	VERB
bracis-28417	126	5	npm	npm	PROPN
bracis-28417	126	6	rather	rather	ADV
bracis-28417	126	7	than	than	ADP
bracis-28417	126	8	a	a	DET
bracis-28417	126	9	straightforward	straightforward	ADJ
bracis-28417	126	10	average	average	NOUN
bracis-28417	126	11	across	across	ADP
bracis-28417	126	12	all	all	DET
bracis-28417	126	13	datasets	dataset	NOUN
bracis-28417	126	14	is	be	AUX
bracis-28417	126	15	to	to	PART
bracis-28417	126	16	mitigate	mitigate	VERB
bracis-28417	126	17	the	the	DET
bracis-28417	126	18	undue	undue	ADJ
bracis-28417	126	19	influence	influence	NOUN
bracis-28417	126	20	of	of	ADP
bracis-28417	126	21	datasets	dataset	NOUN
bracis-28417	126	22	with	with	ADP
bracis-28417	126	23	inherently	inherently	ADV
bracis-28417	126	24	high	high	ADJ
bracis-28417	126	25	scores	score	NOUN
bracis-28417	126	26	,	,	PUNCT
bracis-28417	126	27	such	such	ADJ
bracis-28417	126	28	as	as	ADP
bracis-28417	126	29	binary	binary	ADJ
bracis-28417	126	30	classification	classification	NOUN
bracis-28417	126	31	datasets	dataset	NOUN
bracis-28417	126	32	,	,	PUNCT
bracis-28417	126	33	which	which	PRON
bracis-28417	126	34	could	could	AUX
bracis-28417	126	35	otherwise	otherwise	ADV
bracis-28417	126	36	outweigh	outweigh	VERB
bracis-28417	126	37	datasets	dataset	NOUN
bracis-28417	126	38	characterized	characterize	VERB
bracis-28417	126	39	by	by	ADP
bracis-28417	126	40	lower	low	ADJ
bracis-28417	126	41	scores	score	NOUN
bracis-28417	126	42	.	.	PUNCT
bracis-28417	127	1	table	table	NOUN
bracis-28417	127	2	1	1	NUM
bracis-28417	127	3	.	.	PUNCT
bracis-28417	128	1	a	a	DET
bracis-28417	128	2	summary	summary	NOUN
bracis-28417	128	3	of	of	ADP
bracis-28417	128	4	the	the	DET
bracis-28417	128	5	datasets	dataset	NOUN
bracis-28417	128	6	constituting	constitute	VERB
bracis-28417	128	7	the	the	DET
bracis-28417	128	8	poeta	poeta	NOUN
bracis-28417	128	9	benchmark.full	benchmark.full	NOUN
bracis-28417	128	10	size	size	NOUN
bracis-28417	128	11	table	table	NOUN
bracis-28417	128	12	5	5	NUM
bracis-28417	128	13	results	result	VERB
bracis-28417	128	14	the	the	DET
bracis-28417	128	15	main	main	ADJ
bracis-28417	128	16	results	result	NOUN
bracis-28417	128	17	can	can	AUX
bracis-28417	128	18	be	be	AUX
bracis-28417	128	19	found	find	VERB
bracis-28417	128	20	in	in	ADP
bracis-28417	128	21	table	table	NOUN
bracis-28417	128	22	 	 	SPACE
bracis-28417	128	23	2	2	NUM
bracis-28417	128	24	.	.	NOUN
bracis-28417	128	25	models	model	NOUN
bracis-28417	128	26	such	such	ADJ
bracis-28417	128	27	as	as	ADP
bracis-28417	128	28	bloomz	bloomz	ADJ
bracis-28417	128	29	,	,	PUNCT
bracis-28417	128	30	xglm	xglm	PROPN
bracis-28417	128	31	and	and	CCONJ
bracis-28417	128	32	bertin	bertin	PROPN
bracis-28417	128	33	-	-	PUNCT
bracis-28417	128	34	gpt	gpt	NOUN
bracis-28417	128	35	struggled	struggle	VERB
bracis-28417	128	36	to	to	PART
bracis-28417	128	37	generate	generate	VERB
bracis-28417	128	38	answers	answer	NOUN
bracis-28417	128	39	in	in	ADP
bracis-28417	128	40	portuguese	portuguese	PROPN
bracis-28417	128	41	.	.	PUNCT
bracis-28417	129	1	to	to	PART
bracis-28417	129	2	address	address	VERB
bracis-28417	129	3	this	this	DET
bracis-28417	129	4	issue	issue	NOUN
bracis-28417	129	5	,	,	PUNCT
bracis-28417	129	6	we	we	PRON
bracis-28417	129	7	adopted	adopt	VERB
bracis-28417	129	8	an	an	DET
bracis-28417	129	9	approach	approach	NOUN
bracis-28417	129	10	akin	akin	ADJ
bracis-28417	129	11	to	to	ADP
bracis-28417	129	12	that	that	PRON
bracis-28417	129	13	used	use	VERB
bracis-28417	129	14	by	by	ADP
bracis-28417	129	15	the	the	DET
bracis-28417	129	16	xglm	xglm	PROPN
bracis-28417	129	17	authors	author	NOUN
bracis-28417	129	18	:	:	PUNCT
bracis-28417	129	19	by	by	ADP
bracis-28417	129	20	calculating	calculate	VERB
bracis-28417	129	21	the	the	DET
bracis-28417	129	22	likelihood	likelihood	NOUN
bracis-28417	129	23	of	of	ADP
bracis-28417	129	24	each	each	DET
bracis-28417	129	25	candidate	candidate	NOUN
bracis-28417	129	26	answer	answer	NOUN
bracis-28417	129	27	string	string	NOUN
bracis-28417	129	28	based	base	VERB
bracis-28417	129	29	on	on	ADP
bracis-28417	129	30	the	the	DET
bracis-28417	129	31	input	input	NOUN
bracis-28417	129	32	text	text	NOUN
bracis-28417	129	33	and	and	CCONJ
bracis-28417	129	34	subsequently	subsequently	ADV
bracis-28417	129	35	selecting	select	VERB
bracis-28417	129	36	the	the	DET
bracis-28417	129	37	class	class	NOUN
bracis-28417	129	38	with	with	ADP
bracis-28417	129	39	the	the	DET
bracis-28417	129	40	highest	high	ADJ
bracis-28417	129	41	probability	probability	NOUN
bracis-28417	129	42	.	.	PUNCT
bracis-28417	130	1	for	for	ADP
bracis-28417	130	2	faquad	faquad	ADJ
bracis-28417	130	3	,	,	PUNCT
bracis-28417	130	4	the	the	DET
bracis-28417	130	5	only	only	ADJ
bracis-28417	130	6	dataset	dataset	NOUN
bracis-28417	130	7	in	in	ADP
bracis-28417	130	8	the	the	DET
bracis-28417	130	9	benchmark	benchmark	NOUN
bracis-28417	130	10	without	without	ADP
bracis-28417	130	11	predetermined	predetermine	VERB
bracis-28417	130	12	candidate	candidate	NOUN
bracis-28417	130	13	answers	answer	NOUN
bracis-28417	130	14	,	,	PUNCT
bracis-28417	130	15	we	we	PRON
bracis-28417	130	16	allowed	allow	VERB
bracis-28417	130	17	the	the	DET
bracis-28417	130	18	models	model	NOUN
bracis-28417	130	19	to	to	PART
bracis-28417	130	20	generate	generate	VERB
bracis-28417	130	21	answers	answer	NOUN
bracis-28417	130	22	in	in	ADP
bracis-28417	130	23	their	their	PRON
bracis-28417	130	24	original	original	ADJ
bracis-28417	130	25	format	format	NOUN
bracis-28417	130	26	.	.	PUNCT
bracis-28417	131	1	we	we	PRON
bracis-28417	131	2	observe	observe	VERB
bracis-28417	131	3	that	that	SCONJ
bracis-28417	131	4	the	the	DET
bracis-28417	131	5	llama	llama	PROPN
bracis-28417	131	6	baselines	baseline	NOUN
bracis-28417	131	7	significantly	significantly	ADV
bracis-28417	131	8	outperform	outperform	VERB
bracis-28417	131	9	models	model	NOUN
bracis-28417	131	10	of	of	ADP
bracis-28417	131	11	equivalent	equivalent	ADJ
bracis-28417	131	12	size	size	NOUN
bracis-28417	131	13	trained	train	VERB
bracis-28417	131	14	with	with	ADP
bracis-28417	131	15	fewer	few	ADJ
bracis-28417	131	16	tokens	token	NOUN
bracis-28417	131	17	,	,	PUNCT
bracis-28417	131	18	such	such	ADJ
bracis-28417	131	19	as	as	ADP
bracis-28417	131	20	galactica	galactica	PROPN
bracis-28417	131	21	and	and	CCONJ
bracis-28417	131	22	opt	opt	VERB
bracis-28417	131	23	.	.	PUNCT
bracis-28417	132	1	furthermore	furthermore	ADV
bracis-28417	132	2	,	,	PUNCT
bracis-28417	132	3	despite	despite	SCONJ
bracis-28417	132	4	being	be	AUX
bracis-28417	132	5	trained	train	VERB
bracis-28417	132	6	on	on	ADP
bracis-28417	132	7	english	english	ADJ
bracis-28417	132	8	-	-	PUNCT
bracis-28417	132	9	centric	centric	NOUN
bracis-28417	132	10	corpora	corpora	NOUN
bracis-28417	132	11	,	,	PUNCT
bracis-28417	132	12	llama-7b	llama-7b	X
bracis-28417	132	13	surpasses	surpass	VERB
bracis-28417	132	14	multilingual	multilingual	ADJ
bracis-28417	132	15	bloom	bloom	NOUN
bracis-28417	132	16	and	and	CCONJ
bracis-28417	132	17	xglm	xglm	NOUN
bracis-28417	132	18	of	of	ADP
bracis-28417	132	19	similar	similar	ADJ
bracis-28417	132	20	sizes	size	NOUN
bracis-28417	132	21	.	.	PUNCT
bracis-28417	133	1	the	the	DET
bracis-28417	133	2	sabiá	sabiá	NOUN
bracis-28417	133	3	models	model	NOUN
bracis-28417	133	4	demonstrate	demonstrate	VERB
bracis-28417	133	5	considerable	considerable	ADJ
bracis-28417	133	6	improvement	improvement	NOUN
bracis-28417	133	7	in	in	ADP
bracis-28417	133	8	npm	npm	NOUN
bracis-28417	133	9	compared	compare	VERB
bracis-28417	133	10	to	to	ADP
bracis-28417	133	11	their	their	PRON
bracis-28417	133	12	respective	respective	ADJ
bracis-28417	133	13	baseline	baseline	NOUN
bracis-28417	133	14	models	model	NOUN
bracis-28417	133	15	.	.	PUNCT
bracis-28417	134	1	these	these	DET
bracis-28417	134	2	npm	npm	PROPN
bracis-28417	134	3	gains	gain	NOUN
bracis-28417	134	4	are	be	AUX
bracis-28417	134	5	more	more	ADV
bracis-28417	134	6	substantial	substantial	ADJ
bracis-28417	134	7	for	for	ADP
bracis-28417	134	8	the	the	DET
bracis-28417	134	9	smaller	small	ADJ
bracis-28417	134	10	sabiá	sabiá	NOUN
bracis-28417	134	11	-	-	PUNCT
bracis-28417	134	12	j	j	PROPN
bracis-28417	134	13	and	and	CCONJ
bracis-28417	134	14	sabiá-7b	sabiá-7b	ADV
bracis-28417	134	15	models	model	NOUN
bracis-28417	134	16	.	.	PUNCT
bracis-28417	135	1	notably	notably	ADV
bracis-28417	135	2	,	,	PUNCT
bracis-28417	135	3	sabiá-65b	sabiá-65b	PROPN
bracis-28417	135	4	marginally	marginally	ADV
bracis-28417	135	5	outperforms	outperform	VERB
bracis-28417	135	6	openai	openai	PROPN
bracis-28417	135	7	’s	’s	PART
bracis-28417	135	8	gpt-3.5	gpt-3.5	NOUN
bracis-28417	135	9	-	-	PUNCT
bracis-28417	135	10	turbo	turbo	NOUN
bracis-28417	135	11	,	,	PUNCT
bracis-28417	135	12	which	which	PRON
bracis-28417	135	13	serves	serve	VERB
bracis-28417	135	14	as	as	ADP
bracis-28417	135	15	the	the	DET
bracis-28417	135	16	base	base	NOUN
bracis-28417	135	17	model	model	NOUN
bracis-28417	135	18	for	for	ADP
bracis-28417	135	19	chatgpt	chatgpt	NOUN
bracis-28417	135	20	.	.	PUNCT
bracis-28417	136	1	table	table	NOUN
bracis-28417	136	2	2	2	NUM
bracis-28417	136	3	.	.	X
bracis-28417	137	1	few	few	ADJ
bracis-28417	137	2	-	-	PUNCT
bracis-28417	137	3	shot	shot	NOUN
bracis-28417	137	4	npm	npm	NOUN
bracis-28417	137	5	results	result	NOUN
bracis-28417	137	6	on	on	ADP
bracis-28417	137	7	the	the	DET
bracis-28417	137	8	poeta	poeta	NOUN
bracis-28417	137	9	benchmark.full	benchmark.full	NOUN
bracis-28417	137	10	size	size	NOUN
bracis-28417	137	11	table	table	NOUN
bracis-28417	137	12	through	through	ADP
bracis-28417	137	13	our	our	PRON
bracis-28417	137	14	portuguese	portuguese	ADJ
bracis-28417	137	15	pretraining	pretraine	VERB
bracis-28417	138	1	,	,	PUNCT
bracis-28417	138	2	we	we	PRON
bracis-28417	138	3	observed	observe	VERB
bracis-28417	138	4	that	that	SCONJ
bracis-28417	138	5	the	the	DET
bracis-28417	138	6	improvement	improvement	NOUN
bracis-28417	138	7	in	in	ADP
bracis-28417	138	8	npm	npm	PROPN
bracis-28417	138	9	was	be	AUX
bracis-28417	138	10	higher	high	ADJ
bracis-28417	138	11	in	in	ADP
bracis-28417	138	12	native	native	ADJ
bracis-28417	138	13	datasets	dataset	NOUN
bracis-28417	138	14	than	than	ADP
bracis-28417	138	15	that	that	PRON
bracis-28417	138	16	in	in	ADP
bracis-28417	138	17	translated	translate	VERB
bracis-28417	138	18	datasets	dataset	NOUN
bracis-28417	138	19	.	.	PUNCT
bracis-28417	139	1	for	for	ADP
bracis-28417	139	2	sabiá-65b	sabiá-65b	PROPN
bracis-28417	139	3	,	,	PUNCT
bracis-28417	139	4	improvements	improvement	NOUN
bracis-28417	139	5	over	over	ADP
bracis-28417	139	6	llama-65b	llama-65b	PROPN
bracis-28417	139	7	were	be	AUX
bracis-28417	139	8	mostly	mostly	ADV
bracis-28417	139	9	from	from	ADP
bracis-28417	139	10	the	the	DET
bracis-28417	139	11	native	native	ADJ
bracis-28417	139	12	subset	subset	NOUN
bracis-28417	139	13	.	.	PUNCT
bracis-28417	140	1	we	we	PRON
bracis-28417	140	2	hypothesize	hypothesize	VERB
bracis-28417	140	3	that	that	SCONJ
bracis-28417	140	4	this	this	PRON
bracis-28417	140	5	is	be	AUX
bracis-28417	140	6	due	due	ADJ
bracis-28417	140	7	to	to	ADP
bracis-28417	140	8	the	the	DET
bracis-28417	140	9	“	"	PUNCT
bracis-28417	140	10	mechanistic	mechanistic	ADJ
bracis-28417	140	11	”	"	PUNCT
bracis-28417	140	12	nature	nature	NOUN
bracis-28417	140	13	of	of	ADP
bracis-28417	140	14	translated	translate	VERB
bracis-28417	140	15	datasets	dataset	NOUN
bracis-28417	140	16	:	:	PUNCT
bracis-28417	140	17	since	since	SCONJ
bracis-28417	140	18	they	they	PRON
bracis-28417	140	19	were	be	AUX
bracis-28417	140	20	translated	translate	VERB
bracis-28417	140	21	from	from	ADP
bracis-28417	140	22	english	english	PROPN
bracis-28417	140	23	,	,	PUNCT
bracis-28417	140	24	the	the	DET
bracis-28417	140	25	baseline	baseline	PROPN
bracis-28417	140	26	model	model	NOUN
bracis-28417	140	27	already	already	ADV
bracis-28417	140	28	possesses	possess	VERB
bracis-28417	140	29	the	the	DET
bracis-28417	140	30	knowledge	knowledge	NOUN
bracis-28417	140	31	needed	need	VERB
bracis-28417	140	32	to	to	PART
bracis-28417	140	33	solve	solve	VERB
bracis-28417	140	34	them	they	PRON
bracis-28417	140	35	and	and	CCONJ
bracis-28417	140	36	gains	gain	VERB
bracis-28417	140	37	little	little	ADJ
bracis-28417	140	38	from	from	ADP
bracis-28417	140	39	learning	learn	VERB
bracis-28417	140	40	the	the	DET
bracis-28417	140	41	linguistic	linguistic	ADJ
bracis-28417	140	42	,	,	PUNCT
bracis-28417	140	43	syntactic	syntactic	ADJ
bracis-28417	140	44	,	,	PUNCT
bracis-28417	140	45	and	and	CCONJ
bracis-28417	140	46	grammatical	grammatical	ADJ
bracis-28417	140	47	knowledge	knowledge	NOUN
bracis-28417	140	48	of	of	ADP
bracis-28417	140	49	the	the	DET
bracis-28417	140	50	target	target	NOUN
bracis-28417	140	51	language	language	NOUN
bracis-28417	140	52	.	.	PUNCT
bracis-28417	141	1	for	for	ADP
bracis-28417	141	2	instance	instance	NOUN
bracis-28417	141	3	,	,	PUNCT
bracis-28417	141	4	to	to	PART
bracis-28417	141	5	answer	answer	VERB
bracis-28417	141	6	the	the	DET
bracis-28417	141	7	question	question	NOUN
bracis-28417	141	8	“	"	PUNCT
bracis-28417	141	9	does	do	AUX
bracis-28417	141	10	p	p	PROPN
bracis-28417	141	11	o	o	PROPN
bracis-28417	141	12	box	box	NOUN
bracis-28417	141	13	come	come	VERB
bracis-28417	141	14	before	before	ADP
bracis-28417	141	15	street	street	NOUN
bracis-28417	141	16	address	address	NOUN
bracis-28417	141	17	”	"	PUNCT
bracis-28417	141	18	(	(	PUNCT
bracis-28417	141	19	boolq	boolq	PROPN
bracis-28417	141	20	dataset	dataset	PROPN
bracis-28417	141	21	)	)	PUNCT
bracis-28417	141	22	,	,	PUNCT
bracis-28417	141	23	the	the	DET
bracis-28417	141	24	model	model	NOUN
bracis-28417	141	25	gains	gain	VERB
bracis-28417	141	26	little	little	ADJ
bracis-28417	141	27	from	from	ADP
bracis-28417	141	28	additional	additional	ADJ
bracis-28417	141	29	pretraining	pretraining	NOUN
bracis-28417	141	30	on	on	ADP
bracis-28417	141	31	a	a	DET
bracis-28417	141	32	portuguese	portuguese	ADJ
bracis-28417	141	33	corpus	corpus	NOUN
bracis-28417	141	34	as	as	SCONJ
bracis-28417	141	35	it	it	PRON
bracis-28417	141	36	is	be	AUX
bracis-28417	141	37	unlikely	unlikely	ADJ
bracis-28417	141	38	that	that	SCONJ
bracis-28417	141	39	the	the	DET
bracis-28417	141	40	corpus	corpus	NOUN
bracis-28417	141	41	would	would	AUX
bracis-28417	141	42	provide	provide	VERB
bracis-28417	141	43	new	new	ADJ
bracis-28417	141	44	information	information	NOUN
bracis-28417	141	45	regarding	regard	VERB
bracis-28417	141	46	the	the	DET
bracis-28417	141	47	formatting	formatting	NOUN
bracis-28417	141	48	of	of	ADP
bracis-28417	141	49	us	we	PRON
bracis-28417	141	50	mailing	mail	VERB
bracis-28417	141	51	addresses	address	NOUN
bracis-28417	141	52	that	that	SCONJ
bracis-28417	141	53	the	the	DET
bracis-28417	141	54	model	model	NOUN
bracis-28417	141	55	has	have	AUX
bracis-28417	141	56	not	not	PART
bracis-28417	141	57	already	already	ADV
bracis-28417	141	58	encountered	encounter	VERB
bracis-28417	141	59	during	during	ADP
bracis-28417	141	60	its	its	PRON
bracis-28417	141	61	initial	initial	ADJ
bracis-28417	141	62	english	english	ADJ
bracis-28417	141	63	-	-	PUNCT
bracis-28417	141	64	centric	centric	ADJ
bracis-28417	141	65	pretraining	pretraining	NOUN
bracis-28417	141	66	.	.	PUNCT
bracis-28417	142	1	conversely	conversely	ADV
bracis-28417	142	2	,	,	PUNCT
bracis-28417	142	3	language	language	NOUN
bracis-28417	142	4	-	-	PUNCT
bracis-28417	142	5	specific	specific	ADJ
bracis-28417	142	6	pretraining	pretraine	VERB
bracis-28417	142	7	introduces	introduce	NOUN
bracis-28417	142	8	the	the	DET
bracis-28417	142	9	specific	specific	ADJ
bracis-28417	142	10	knowledge	knowledge	NOUN
bracis-28417	142	11	required	require	VERB
bracis-28417	142	12	to	to	PART
bracis-28417	142	13	solve	solve	VERB
bracis-28417	142	14	tasks	task	NOUN
bracis-28417	142	15	in	in	ADP
bracis-28417	142	16	the	the	DET
bracis-28417	142	17	native	native	ADJ
bracis-28417	142	18	subset	subset	NOUN
bracis-28417	142	19	.	.	PUNCT
bracis-28417	143	1	although	although	SCONJ
bracis-28417	143	2	gpt	gpt	PROPN
bracis-28417	143	3	-	-	PUNCT
bracis-28417	143	4	j	j	NOUN
bracis-28417	143	5	exhibited	exhibit	VERB
bracis-28417	143	6	lower	low	ADJ
bracis-28417	143	7	few	few	ADJ
bracis-28417	143	8	-	-	PUNCT
bracis-28417	143	9	shot	shot	NOUN
bracis-28417	143	10	performance	performance	NOUN
bracis-28417	143	11	in	in	ADP
bracis-28417	143	12	english	english	ADJ
bracis-28417	143	13	tasks	task	NOUN
bracis-28417	143	14	relative	relative	ADJ
bracis-28417	143	15	to	to	ADP
bracis-28417	143	16	llama	llama	NOUN
bracis-28417	143	17	,	,	PUNCT
bracis-28417	143	18	we	we	PRON
bracis-28417	143	19	use	use	VERB
bracis-28417	143	20	it	it	PRON
bracis-28417	143	21	in	in	ADP
bracis-28417	143	22	this	this	DET
bracis-28417	143	23	study	study	NOUN
bracis-28417	143	24	to	to	PART
bracis-28417	143	25	illustrate	illustrate	VERB
bracis-28417	143	26	that	that	SCONJ
bracis-28417	143	27	not	not	PART
bracis-28417	143	28	only	only	ADV
bracis-28417	143	29	highly	highly	ADV
bracis-28417	143	30	optimized	optimize	VERB
bracis-28417	143	31	models	model	NOUN
bracis-28417	143	32	like	like	ADP
bracis-28417	143	33	llama	llama	NOUN
bracis-28417	143	34	can	can	AUX
bracis-28417	143	35	benefit	benefit	VERB
bracis-28417	143	36	from	from	ADP
bracis-28417	143	37	extended	extended	ADJ
bracis-28417	143	38	pretraining	pretraining	NOUN
bracis-28417	143	39	.	.	PUNCT
bracis-28417	144	1	we	we	PRON
bracis-28417	144	2	chose	choose	VERB
bracis-28417	144	3	not	not	PART
bracis-28417	144	4	to	to	PART
bracis-28417	144	5	use	use	VERB
bracis-28417	144	6	bloom-7.1b	bloom-7.1b	NOUN
bracis-28417	144	7	as	as	ADP
bracis-28417	144	8	our	our	PRON
bracis-28417	144	9	initial	initial	ADJ
bracis-28417	144	10	checkpoint	checkpoint	NOUN
bracis-28417	144	11	for	for	ADP
bracis-28417	144	12	pretraining	pretraine	VERB
bracis-28417	144	13	due	due	ADP
bracis-28417	144	14	to	to	ADP
bracis-28417	144	15	its	its	PRON
bracis-28417	144	16	inferior	inferior	ADJ
bracis-28417	144	17	performance	performance	NOUN
bracis-28417	144	18	compared	compare	VERB
bracis-28417	144	19	to	to	ADP
bracis-28417	144	20	gpt	gpt	NOUN
bracis-28417	144	21	-	-	PUNCT
bracis-28417	144	22	j	j	NOUN
bracis-28417	144	23	in	in	ADP
bracis-28417	144	24	preliminary	preliminary	ADJ
bracis-28417	144	25	few	few	ADJ
bracis-28417	144	26	-	-	PUNCT
bracis-28417	144	27	shot	shot	NOUN
bracis-28417	144	28	experiments	experiment	NOUN
bracis-28417	144	29	on	on	ADP
bracis-28417	144	30	three	three	NUM
bracis-28417	144	31	portuguese	portuguese	ADJ
bracis-28417	144	32	datasets	dataset	NOUN
bracis-28417	144	33	.	.	PUNCT
bracis-28417	145	1	however	however	ADV
bracis-28417	145	2	,	,	PUNCT
bracis-28417	145	3	we	we	PRON
bracis-28417	145	4	later	later	ADV
bracis-28417	145	5	discovered	discover	VERB
bracis-28417	145	6	that	that	SCONJ
bracis-28417	145	7	its	its	PRON
bracis-28417	145	8	performance	performance	NOUN
bracis-28417	145	9	on	on	ADP
bracis-28417	145	10	poeta	poeta	PROPN
bracis-28417	145	11	surpassed	surpass	VERB
bracis-28417	145	12	gpt	gpt	PROPN
bracis-28417	145	13	-	-	PUNCT
bracis-28417	145	14	j	j	NOUN
bracis-28417	145	15	’s	’s	NOUN
bracis-28417	145	16	.	.	PUNCT
bracis-28417	146	1	nonetheless	nonetheless	ADV
bracis-28417	146	2	,	,	PUNCT
bracis-28417	146	3	bloom	bloom	NOUN
bracis-28417	146	4	still	still	ADV
bracis-28417	146	5	exhibits	exhibit	VERB
bracis-28417	146	6	lower	low	ADJ
bracis-28417	146	7	performance	performance	NOUN
bracis-28417	146	8	compared	compare	VERB
bracis-28417	146	9	to	to	ADP
bracis-28417	146	10	llama	llama	PROPN
bracis-28417	146	11	.	.	PUNCT
bracis-28417	147	1	analogous	analogous	ADJ
bracis-28417	147	2	to	to	ADP
bracis-28417	147	3	sabiá	sabiá	NOUN
bracis-28417	147	4	-	-	PUNCT
bracis-28417	147	5	j	j	PROPN
bracis-28417	147	6	,	,	PUNCT
bracis-28417	147	7	bertin	bertin	PROPN
bracis-28417	147	8	-	-	PUNCT
bracis-28417	147	9	gpt	gpt	NOUN
bracis-28417	147	10	is	be	AUX
bracis-28417	147	11	a	a	DET
bracis-28417	147	12	model	model	NOUN
bracis-28417	147	13	pretrained	pretraine	VERB
bracis-28417	147	14	on	on	ADP
bracis-28417	147	15	spanish	spanish	ADJ
bracis-28417	147	16	text	text	NOUN
bracis-28417	147	17	starting	start	VERB
bracis-28417	147	18	from	from	ADP
bracis-28417	147	19	the	the	DET
bracis-28417	147	20	gpt	gpt	PROPN
bracis-28417	147	21	-	-	PUNCT
bracis-28417	147	22	j	j	NOUN
bracis-28417	147	23	weights	weight	NOUN
bracis-28417	147	24	.	.	PUNCT
bracis-28417	148	1	since	since	SCONJ
bracis-28417	148	2	spanish	spanish	ADJ
bracis-28417	148	3	and	and	CCONJ
bracis-28417	148	4	portuguese	portuguese	ADJ
bracis-28417	148	5	are	be	AUX
bracis-28417	148	6	similar	similar	ADJ
bracis-28417	148	7	languages	language	NOUN
bracis-28417	148	8	,	,	PUNCT
bracis-28417	148	9	it	it	PRON
bracis-28417	148	10	is	be	AUX
bracis-28417	148	11	reasonable	reasonable	ADJ
bracis-28417	148	12	to	to	PART
bracis-28417	148	13	expect	expect	VERB
bracis-28417	148	14	that	that	SCONJ
bracis-28417	148	15	bertin	bertin	NOUN
bracis-28417	148	16	-	-	PUNCT
bracis-28417	148	17	gpt	gpt	NOUN
bracis-28417	148	18	would	would	AUX
bracis-28417	148	19	perform	perform	VERB
bracis-28417	148	20	better	well	ADJ
bracis-28417	148	21	than	than	ADP
bracis-28417	148	22	its	its	PRON
bracis-28417	148	23	baseline	baseline	NOUN
bracis-28417	148	24	model	model	NOUN
bracis-28417	148	25	.	.	PUNCT
bracis-28417	149	1	nevertheless	nevertheless	ADV
bracis-28417	149	2	,	,	PUNCT
bracis-28417	149	3	the	the	DET
bracis-28417	149	4	observed	observe	VERB
bracis-28417	149	5	npm	npm	NOUN
bracis-28417	149	6	for	for	ADP
bracis-28417	149	7	bertin	bertin	NOUN
bracis-28417	149	8	-	-	PUNCT
bracis-28417	149	9	gpt	gpt	NOUN
bracis-28417	149	10	is	be	AUX
bracis-28417	149	11	only	only	ADV
bracis-28417	149	12	slightly	slightly	ADV
bracis-28417	149	13	higher	high	ADJ
bracis-28417	149	14	than	than	ADP
bracis-28417	149	15	gpt	gpt	NOUN
bracis-28417	149	16	-	-	PUNCT
bracis-28417	149	17	j	j	NOUN
bracis-28417	149	18	’s	’s	NOUN
bracis-28417	149	19	.	.	PUNCT
bracis-28417	150	1	a	a	DET
bracis-28417	150	2	noteworthy	noteworthy	ADJ
bracis-28417	150	3	comparison	comparison	NOUN
bracis-28417	150	4	involves	involve	VERB
bracis-28417	150	5	galactica	galactica	PROPN
bracis-28417	150	6	,	,	PUNCT
bracis-28417	150	7	a	a	DET
bracis-28417	150	8	model	model	NOUN
bracis-28417	150	9	pretrained	pretraine	VERB
bracis-28417	150	10	on	on	ADP
bracis-28417	150	11	scientific	scientific	ADJ
bracis-28417	150	12	text	text	NOUN
bracis-28417	150	13	,	,	PUNCT
bracis-28417	150	14	predominantly	predominantly	ADV
bracis-28417	150	15	in	in	ADP
bracis-28417	150	16	english	english	PROPN
bracis-28417	150	17	,	,	PUNCT
bracis-28417	150	18	and	and	CCONJ
bracis-28417	150	19	a	a	DET
bracis-28417	150	20	similarly	similarly	ADV
bracis-28417	150	21	-	-	PUNCT
bracis-28417	150	22	sized	sized	ADJ
bracis-28417	150	23	opt	opt	NOUN
bracis-28417	150	24	model	model	NOUN
bracis-28417	150	25	,	,	PUNCT
bracis-28417	150	26	which	which	PRON
bracis-28417	150	27	utilized	utilize	VERB
bracis-28417	150	28	comparable	comparable	ADJ
bracis-28417	150	29	pretraining	pretraine	VERB
bracis-28417	150	30	compute	compute	NOUN
bracis-28417	150	31	but	but	CCONJ
bracis-28417	150	32	was	be	AUX
bracis-28417	150	33	pretrained	pretraine	VERB
bracis-28417	150	34	on	on	ADP
bracis-28417	150	35	a	a	DET
bracis-28417	150	36	larger	large	ADJ
bracis-28417	150	37	and	and	CCONJ
bracis-28417	150	38	more	more	ADV
bracis-28417	150	39	diverse	diverse	ADJ
bracis-28417	150	40	english	english	ADJ
bracis-28417	150	41	-	-	PUNCT
bracis-28417	150	42	centric	centric	NOUN
bracis-28417	150	43	corpus	corpus	NOUN
bracis-28417	150	44	.	.	PUNCT
bracis-28417	151	1	in	in	ADP
bracis-28417	151	2	their	their	PRON
bracis-28417	151	3	study	study	NOUN
bracis-28417	151	4	,	,	PUNCT
bracis-28417	151	5	the	the	DET
bracis-28417	151	6	authors	author	NOUN
bracis-28417	151	7	demonstrate	demonstrate	VERB
bracis-28417	151	8	that	that	SCONJ
bracis-28417	151	9	galactica	galactica	PROPN
bracis-28417	151	10	performs	perform	VERB
bracis-28417	151	11	on	on	ADP
bracis-28417	151	12	par	par	NOUN
bracis-28417	151	13	with	with	ADP
bracis-28417	151	14	opt	opt	NOUN
bracis-28417	151	15	on	on	ADP
bracis-28417	151	16	english	english	ADJ
bracis-28417	151	17	tasks	task	NOUN
bracis-28417	151	18	and	and	CCONJ
bracis-28417	151	19	largely	largely	ADV
bracis-28417	151	20	outperforms	outperform	VERB
bracis-28417	151	21	opt	opt	NOUN
bracis-28417	151	22	on	on	ADP
bracis-28417	151	23	scientific	scientific	ADJ
bracis-28417	151	24	-	-	PUNCT
bracis-28417	151	25	related	relate	VERB
bracis-28417	151	26	tasks	task	NOUN
bracis-28417	151	27	.	.	PUNCT
bracis-28417	152	1	conversely	conversely	ADV
bracis-28417	152	2	,	,	PUNCT
bracis-28417	152	3	opt	opt	VERB
bracis-28417	152	4	significantly	significantly	ADV
bracis-28417	152	5	outperforms	outperform	VERB
bracis-28417	152	6	galactica	galactica	PROPN
bracis-28417	152	7	in	in	ADP
bracis-28417	152	8	portuguese	portuguese	ADJ
bracis-28417	152	9	tasks	task	NOUN
bracis-28417	152	10	.	.	PUNCT
bracis-28417	153	1	this	this	DET
bracis-28417	153	2	result	result	NOUN
bracis-28417	153	3	underscores	underscore	VERB
bracis-28417	153	4	the	the	DET
bracis-28417	153	5	trade	trade	NOUN
bracis-28417	153	6	-	-	PUNCT
bracis-28417	153	7	offs	off	NOUN
bracis-28417	153	8	associated	associate	VERB
bracis-28417	153	9	with	with	ADP
bracis-28417	153	10	domain	domain	NOUN
bracis-28417	153	11	-	-	PUNCT
bracis-28417	153	12	specific	specific	ADJ
bracis-28417	153	13	specialization	specialization	NOUN
bracis-28417	153	14	,	,	PUNCT
bracis-28417	153	15	which	which	PRON
bracis-28417	153	16	often	often	ADV
bracis-28417	153	17	entails	entail	VERB
bracis-28417	153	18	diminished	diminished	ADJ
bracis-28417	153	19	performance	performance	NOUN
bracis-28417	153	20	in	in	ADP
bracis-28417	153	21	other	other	ADJ
bracis-28417	153	22	tasks	task	NOUN
bracis-28417	153	23	.	.	PUNCT
bracis-28417	154	1	bloomz	bloomz	ADJ
bracis-28417	154	2	 	 	SPACE
bracis-28417	155	1	[	[	X
bracis-28417	155	2	35	35	NUM
bracis-28417	155	3	]	]	X
bracis-28417	155	4	,	,	PUNCT
bracis-28417	155	5	a	a	DET
bracis-28417	155	6	multilingual	multilingual	ADJ
bracis-28417	155	7	instruction	instruction	NOUN
bracis-28417	155	8	-	-	PUNCT
bracis-28417	155	9	tuned	tune	VERB
bracis-28417	155	10	model	model	NOUN
bracis-28417	155	11	,	,	PUNCT
bracis-28417	155	12	demonstrated	demonstrate	VERB
bracis-28417	155	13	superior	superior	ADJ
bracis-28417	155	14	performance	performance	NOUN
bracis-28417	155	15	compared	compare	VERB
bracis-28417	155	16	to	to	ADP
bracis-28417	155	17	its	its	PRON
bracis-28417	155	18	baseline	baseline	ADJ
bracis-28417	155	19	bloom	bloom	NOUN
bracis-28417	155	20	model	model	NOUN
bracis-28417	155	21	,	,	PUNCT
bracis-28417	155	22	rivaling	rival	VERB
bracis-28417	155	23	llama	llama	NOUN
bracis-28417	155	24	of	of	ADP
bracis-28417	155	25	equivalent	equivalent	ADJ
bracis-28417	155	26	size.footnote	size.footnote	NUM
bracis-28417	155	27	8	8	NUM
bracis-28417	155	28	nevertheless	nevertheless	ADV
bracis-28417	155	29	,	,	PUNCT
bracis-28417	155	30	our	our	PRON
bracis-28417	155	31	approach	approach	NOUN
bracis-28417	155	32	of	of	ADP
bracis-28417	155	33	pretraining	pretraine	VERB
bracis-28417	155	34	in	in	ADP
bracis-28417	155	35	portuguese	portuguese	PROPN
bracis-28417	155	36	appears	appear	VERB
bracis-28417	155	37	to	to	PART
bracis-28417	155	38	yield	yield	VERB
bracis-28417	155	39	superior	superior	ADJ
bracis-28417	155	40	results	result	NOUN
bracis-28417	155	41	,	,	PUNCT
bracis-28417	155	42	as	as	SCONJ
bracis-28417	155	43	sabiá	sabiá	NOUN
bracis-28417	155	44	-	-	PUNCT
bracis-28417	155	45	j	j	PROPN
bracis-28417	155	46	surpasses	surpass	VERB
bracis-28417	155	47	bloomz	bloomz	ADJ
bracis-28417	155	48	despite	despite	SCONJ
bracis-28417	155	49	originating	originate	VERB
bracis-28417	155	50	from	from	ADP
bracis-28417	155	51	a	a	DET
bracis-28417	155	52	lower	lower	ADV
bracis-28417	155	53	-	-	PUNCT
bracis-28417	155	54	performing	perform	VERB
bracis-28417	155	55	baseline	baseline	NOUN
bracis-28417	155	56	model	model	NOUN
bracis-28417	155	57	.	.	PUNCT
bracis-28417	156	1	we	we	PRON
bracis-28417	156	2	envision	envision	VERB
bracis-28417	156	3	continued	continued	ADJ
bracis-28417	156	4	pretraining	pretraine	VERB
bracis-28417	156	5	and	and	CCONJ
bracis-28417	156	6	instruction	instruction	NOUN
bracis-28417	156	7	tuning	tuning	NOUN
bracis-28417	156	8	as	as	ADP
bracis-28417	156	9	complementary	complementary	ADJ
bracis-28417	156	10	techniques	technique	NOUN
bracis-28417	156	11	to	to	PART
bracis-28417	156	12	be	be	AUX
bracis-28417	156	13	combined	combine	VERB
bracis-28417	156	14	in	in	ADP
bracis-28417	156	15	future	future	ADJ
bracis-28417	156	16	research	research	NOUN
bracis-28417	156	17	.	.	PUNCT
bracis-28417	157	1	5.1	5.1	NUM
bracis-28417	157	2	results	result	NOUN
bracis-28417	157	3	per	per	ADP
bracis-28417	157	4	 	 	SPACE
bracis-28417	157	5	dataset	dataset	NOUN
bracis-28417	157	6	table	table	NOUN
bracis-28417	157	7	 	 	SPACE
bracis-28417	157	8	3	3	NUM
bracis-28417	157	9	presents	present	VERB
bracis-28417	157	10	the	the	DET
bracis-28417	157	11	results	result	NOUN
bracis-28417	157	12	per	per	ADP
bracis-28417	157	13	poeta	poeta	PROPN
bracis-28417	157	14	dataset	dataset	VERB
bracis-28417	157	15	for	for	ADP
bracis-28417	157	16	sabiá	sabiá	NOUN
bracis-28417	157	17	models	model	NOUN
bracis-28417	157	18	,	,	PUNCT
bracis-28417	157	19	their	their	PRON
bracis-28417	157	20	baselines	baseline	NOUN
bracis-28417	157	21	,	,	PUNCT
bracis-28417	157	22	and	and	CCONJ
bracis-28417	157	23	for	for	ADP
bracis-28417	157	24	the	the	DET
bracis-28417	157	25	supervised	supervised	ADJ
bracis-28417	157	26	state	state	NOUN
bracis-28417	157	27	-	-	PUNCT
bracis-28417	157	28	of	of	ADP
bracis-28417	157	29	-	-	PUNCT
bracis-28417	157	30	the	the	DET
bracis-28417	157	31	-	-	PUNCT
bracis-28417	157	32	art	art	NOUN
bracis-28417	157	33	.	.	PUNCT
bracis-28417	158	1	the	the	DET
bracis-28417	158	2	sota	sota	PROPN
bracis-28417	158	3	results	result	NOUN
bracis-28417	158	4	reported	report	VERB
bracis-28417	158	5	for	for	SCONJ
bracis-28417	158	6	the	the	DET
bracis-28417	158	7	translated	translate	VERB
bracis-28417	158	8	datasets	dataset	NOUN
bracis-28417	158	9	were	be	AUX
bracis-28417	158	10	obtained	obtain	VERB
bracis-28417	158	11	using	use	VERB
bracis-28417	158	12	their	their	PRON
bracis-28417	158	13	original	original	ADJ
bracis-28417	158	14	english	english	ADJ
bracis-28417	158	15	versions	version	NOUN
bracis-28417	158	16	 	 	SPACE
bracis-28417	159	1	[	[	X
bracis-28417	159	2	46	46	NUM
bracis-28417	159	3	,	,	PUNCT
bracis-28417	159	4	51	51	NUM
bracis-28417	159	5	,	,	PUNCT
bracis-28417	159	6	71	71	NUM
bracis-28417	159	7	,	,	PUNCT
bracis-28417	159	8	76	76	NUM
bracis-28417	159	9	]	]	PUNCT
bracis-28417	159	10	.	.	PUNCT
bracis-28417	160	1	since	since	SCONJ
bracis-28417	160	2	the	the	DET
bracis-28417	160	3	poeta	poeta	PROPN
bracis-28417	160	4	benchmark	benchmark	NOUN
bracis-28417	160	5	excludes	exclude	VERB
bracis-28417	160	6	unanswerable	unanswerable	ADJ
bracis-28417	160	7	examples	example	NOUN
bracis-28417	160	8	of	of	ADP
bracis-28417	160	9	the	the	DET
bracis-28417	160	10	mkqa	mkqa	PROPN
bracis-28417	160	11	dataset	dataset	PROPN
bracis-28417	160	12	,	,	PUNCT
bracis-28417	160	13	we	we	PRON
bracis-28417	160	14	decided	decide	VERB
bracis-28417	160	15	not	not	PART
bracis-28417	160	16	to	to	PART
bracis-28417	160	17	include	include	VERB
bracis-28417	160	18	the	the	DET
bracis-28417	160	19	sota	sota	ADJ
bracis-28417	160	20	result	result	NOUN
bracis-28417	160	21	for	for	ADP
bracis-28417	160	22	this	this	DET
bracis-28417	160	23	dataset	dataset	NOUN
bracis-28417	160	24	.	.	PUNCT
bracis-28417	161	1	in	in	ADP
bracis-28417	161	2	more	more	ADV
bracis-28417	161	3	challenging	challenging	ADJ
bracis-28417	161	4	datasets	dataset	NOUN
bracis-28417	161	5	,	,	PUNCT
bracis-28417	161	6	such	such	ADJ
bracis-28417	161	7	as	as	ADP
bracis-28417	161	8	enem	enem	PROPN
bracis-28417	161	9	challenge	challenge	NOUN
bracis-28417	161	10	,	,	PUNCT
bracis-28417	161	11	enem	enem	PROPN
bracis-28417	161	12	2022	2022	NUM
bracis-28417	161	13	,	,	PUNCT
bracis-28417	161	14	and	and	CCONJ
bracis-28417	161	15	bluex	bluex	PROPN
bracis-28417	161	16	,	,	PUNCT
bracis-28417	161	17	which	which	PRON
bracis-28417	161	18	are	be	AUX
bracis-28417	161	19	derived	derive	VERB
bracis-28417	161	20	from	from	ADP
bracis-28417	161	21	admission	admission	NOUN
bracis-28417	161	22	exams	exam	NOUN
bracis-28417	161	23	to	to	ADP
bracis-28417	161	24	brazilian	brazilian	ADJ
bracis-28417	161	25	universities	university	NOUN
bracis-28417	161	26	,	,	PUNCT
bracis-28417	161	27	we	we	PRON
bracis-28417	161	28	see	see	VERB
bracis-28417	161	29	the	the	DET
bracis-28417	161	30	most	most	ADV
bracis-28417	161	31	significant	significant	ADJ
bracis-28417	161	32	gains	gain	NOUN
bracis-28417	161	33	due	due	ADP
bracis-28417	161	34	to	to	ADP
bracis-28417	161	35	language	language	NOUN
bracis-28417	161	36	-	-	PUNCT
bracis-28417	161	37	specific	specific	ADJ
bracis-28417	161	38	pretraining	pretraining	NOUN
bracis-28417	161	39	.	.	PUNCT
bracis-28417	162	1	substantial	substantial	ADJ
bracis-28417	162	2	improvements	improvement	NOUN
bracis-28417	162	3	are	be	AUX
bracis-28417	162	4	also	also	ADV
bracis-28417	162	5	observed	observe	VERB
bracis-28417	162	6	in	in	ADP
bracis-28417	162	7	tweetsentbr	tweetsentbr	NOUN
bracis-28417	162	8	,	,	PUNCT
bracis-28417	162	9	a	a	DET
bracis-28417	162	10	dataset	dataset	NOUN
bracis-28417	162	11	containing	contain	VERB
bracis-28417	162	12	tweets	tweet	NOUN
bracis-28417	162	13	with	with	ADP
bracis-28417	162	14	an	an	DET
bracis-28417	162	15	abundance	abundance	NOUN
bracis-28417	162	16	of	of	ADP
bracis-28417	162	17	slang	slang	PROPN
bracis-28417	162	18	and	and	CCONJ
bracis-28417	162	19	references	reference	NOUN
bracis-28417	162	20	to	to	ADP
bracis-28417	162	21	brazilian	brazilian	ADJ
bracis-28417	162	22	popular	popular	ADJ
bracis-28417	162	23	culture	culture	NOUN
bracis-28417	162	24	.	.	PUNCT
bracis-28417	163	1	we	we	PRON
bracis-28417	163	2	hypothesize	hypothesize	VERB
bracis-28417	163	3	that	that	SCONJ
bracis-28417	163	4	this	this	DET
bracis-28417	163	5	pretraining	pretraine	VERB
bracis-28417	163	6	imparts	impart	VERB
bracis-28417	163	7	specific	specific	ADJ
bracis-28417	163	8	knowledge	knowledge	NOUN
bracis-28417	163	9	about	about	ADP
bracis-28417	163	10	the	the	DET
bracis-28417	163	11	country	country	NOUN
bracis-28417	163	12	’s	’s	PART
bracis-28417	163	13	culture	culture	NOUN
bracis-28417	163	14	,	,	PUNCT
bracis-28417	163	15	literature	literature	NOUN
bracis-28417	163	16	,	,	PUNCT
bracis-28417	163	17	and	and	CCONJ
bracis-28417	163	18	geography	geography	NOUN
bracis-28417	163	19	that	that	PRON
bracis-28417	163	20	is	be	AUX
bracis-28417	163	21	less	less	ADV
bracis-28417	163	22	frequently	frequently	ADV
bracis-28417	163	23	encountered	encounter	VERB
bracis-28417	163	24	and	and	CCONJ
bracis-28417	163	25	learned	learn	VERB
bracis-28417	163	26	during	during	ADP
bracis-28417	163	27	the	the	DET
bracis-28417	163	28	original	original	ADJ
bracis-28417	163	29	pretraining	pretraine	VERB
bracis-28417	163	30	with	with	ADP
bracis-28417	163	31	more	more	ADV
bracis-28417	163	32	diverse	diverse	ADJ
bracis-28417	163	33	texts	text	NOUN
bracis-28417	163	34	.	.	PUNCT
bracis-28417	164	1	table	table	NOUN
bracis-28417	164	2	3	3	NUM
bracis-28417	164	3	.	.	PUNCT
bracis-28417	164	4	results	result	NOUN
bracis-28417	164	5	per	per	ADP
bracis-28417	164	6	dataset	dataset	NOUN
bracis-28417	164	7	.	.	PUNCT
bracis-28417	165	1	\(^1\	\(^1\	PROPN
bracis-28417	165	2	)	)	PUNCT
bracis-28417	166	1	[	[	X
bracis-28417	166	2	49	49	NUM
bracis-28417	166	3	]	]	X
bracis-28417	166	4	;	;	PUNCT
bracis-28417	166	5	\(^2\	\(^2\	X
bracis-28417	166	6	)	)	PUNCT
bracis-28417	167	1	[	[	X
bracis-28417	167	2	9	9	NUM
bracis-28417	167	3	]	]	PUNCT
bracis-28417	167	4	;	;	PUNCT
bracis-28417	167	5	\(^3\	\(^3\	ADJ
bracis-28417	167	6	)	)	PUNCT
bracis-28417	168	1	[	[	X
bracis-28417	168	2	34	34	NUM
bracis-28417	168	3	]	]	X
bracis-28417	168	4	;	;	PUNCT
bracis-28417	168	5	\(^4\	\(^4\	NOUN
bracis-28417	168	6	)	)	PUNCT
bracis-28417	169	1	[	[	X
bracis-28417	169	2	3	3	NUM
bracis-28417	169	3	]	]	PUNCT
bracis-28417	169	4	;	;	PUNCT
bracis-28417	169	5	\(^5\	\(^5\	NOUN
bracis-28417	169	6	)	)	PUNCT
bracis-28417	170	1	[	[	X
bracis-28417	170	2	71	71	NUM
bracis-28417	170	3	]	]	X
bracis-28417	170	4	;	;	PUNCT
bracis-28417	170	5	\(^6\	\(^6\	NOUN
bracis-28417	170	6	)	)	PUNCT
bracis-28417	171	1	[	[	X
bracis-28417	171	2	76	76	NUM
bracis-28417	171	3	]	]	X
bracis-28417	171	4	;	;	PUNCT
bracis-28417	171	5	\(^7\	\(^7\	NOUN
bracis-28417	171	6	)	)	PUNCT
bracis-28417	172	1	[	[	X
bracis-28417	172	2	46	46	NUM
bracis-28417	172	3	]	]	X
bracis-28417	172	4	;	;	PUNCT
bracis-28417	172	5	\(^8\	\(^8\	NOUN
bracis-28417	172	6	)	)	PUNCT
bracis-28417	173	1	[	[	X
bracis-28417	173	2	51].full	51].full	NUM
bracis-28417	173	3	size	size	NOUN
bracis-28417	173	4	table	table	NOUN
bracis-28417	173	5	certain	certain	ADJ
bracis-28417	173	6	capabilities	capability	NOUN
bracis-28417	173	7	only	only	ADV
bracis-28417	173	8	emerge	emerge	VERB
bracis-28417	173	9	at	at	ADP
bracis-28417	173	10	scale	scale	NOUN
bracis-28417	173	11	,	,	PUNCT
bracis-28417	173	12	as	as	SCONJ
bracis-28417	173	13	evidenced	evidence	VERB
bracis-28417	173	14	by	by	ADP
bracis-28417	173	15	[	[	X
bracis-28417	173	16	67	67	NUM
bracis-28417	173	17	]	]	PUNCT
bracis-28417	173	18	.	.	PUNCT
bracis-28417	174	1	for	for	ADP
bracis-28417	174	2	example	example	NOUN
bracis-28417	174	3	,	,	PUNCT
bracis-28417	174	4	6	6	NUM
bracis-28417	174	5	-	-	PUNCT
bracis-28417	174	6	7b	7b	NOUN
bracis-28417	174	7	models	model	NOUN
bracis-28417	174	8	perform	perform	VERB
bracis-28417	174	9	close	close	ADV
bracis-28417	174	10	to	to	ADP
bracis-28417	174	11	the	the	DET
bracis-28417	174	12	random	random	ADJ
bracis-28417	174	13	baseline	baseline	NOUN
bracis-28417	174	14	in	in	ADP
bracis-28417	174	15	datasets	dataset	NOUN
bracis-28417	174	16	such	such	ADJ
bracis-28417	174	17	as	as	ADP
bracis-28417	174	18	assin	assin	PROPN
bracis-28417	174	19	2	2	NUM
bracis-28417	174	20	rte	rte	NOUN
bracis-28417	174	21	and	and	CCONJ
bracis-28417	174	22	sts	st	NOUN
bracis-28417	174	23	,	,	PUNCT
bracis-28417	174	24	and	and	CCONJ
bracis-28417	174	25	wsc	wsc	PROPN
bracis-28417	174	26	.	.	PUNCT
bracis-28417	175	1	however	however	ADV
bracis-28417	175	2	,	,	PUNCT
bracis-28417	175	3	at	at	ADP
bracis-28417	175	4	the	the	DET
bracis-28417	175	5	65b	65b	ADJ
bracis-28417	175	6	scale	scale	NOUN
bracis-28417	175	7	,	,	PUNCT
bracis-28417	175	8	we	we	PRON
bracis-28417	175	9	observe	observe	VERB
bracis-28417	175	10	substantial	substantial	ADJ
bracis-28417	175	11	improvements	improvement	NOUN
bracis-28417	175	12	,	,	PUNCT
bracis-28417	175	13	approaching	approach	VERB
bracis-28417	175	14	or	or	CCONJ
bracis-28417	175	15	surpassing	surpass	VERB
bracis-28417	175	16	state	state	NOUN
bracis-28417	175	17	-	-	PUNCT
bracis-28417	175	18	of	of	ADP
bracis-28417	175	19	-	-	PUNCT
bracis-28417	175	20	the	the	DET
bracis-28417	175	21	-	-	PUNCT
bracis-28417	175	22	art	art	NOUN
bracis-28417	175	23	supervised	supervise	VERB
bracis-28417	175	24	models	model	NOUN
bracis-28417	175	25	on	on	ADP
bracis-28417	175	26	the	the	DET
bracis-28417	175	27	assin	assin	PROPN
bracis-28417	175	28	2	2	NUM
bracis-28417	175	29	rte	rte	NOUN
bracis-28417	175	30	and	and	CCONJ
bracis-28417	175	31	faquad	faquad	ADJ
bracis-28417	175	32	datasets	dataset	NOUN
bracis-28417	175	33	.	.	PUNCT
bracis-28417	176	1	gpt-4	gpt-4	VERB
bracis-28417	176	2	 	 	SPACE
bracis-28417	176	3	[	[	X
bracis-28417	176	4	39	39	NUM
bracis-28417	176	5	]	]	PUNCT
bracis-28417	176	6	results	result	NOUN
bracis-28417	176	7	indicate	indicate	VERB
bracis-28417	176	8	that	that	SCONJ
bracis-28417	176	9	there	there	PRON
bracis-28417	176	10	is	be	VERB
bracis-28417	176	11	still	still	ADV
bracis-28417	176	12	room	room	NOUN
bracis-28417	176	13	for	for	ADP
bracis-28417	176	14	improvement	improvement	NOUN
bracis-28417	176	15	for	for	ADP
bracis-28417	176	16	sabiá-65b	sabiá-65b	PROPN
bracis-28417	176	17	in	in	ADP
bracis-28417	176	18	the	the	DET
bracis-28417	176	19	majority	majority	NOUN
bracis-28417	176	20	of	of	ADP
bracis-28417	176	21	the	the	DET
bracis-28417	176	22	datasets	dataset	NOUN
bracis-28417	176	23	evaluated	evaluate	VERB
bracis-28417	176	24	in	in	ADP
bracis-28417	176	25	this	this	DET
bracis-28417	176	26	work	work	NOUN
bracis-28417	176	27	.	.	PUNCT
bracis-28417	177	1	nevertheless	nevertheless	ADV
bracis-28417	177	2	,	,	PUNCT
bracis-28417	177	3	sabiá-65b	sabiá-65b	PROPN
bracis-28417	177	4	performs	perform	VERB
bracis-28417	177	5	on	on	ADP
bracis-28417	177	6	par	par	NOUN
bracis-28417	177	7	with	with	ADP
bracis-28417	177	8	gpt-4	gpt-4	NUM
bracis-28417	177	9	in	in	ADP
bracis-28417	177	10	datasets	dataset	NOUN
bracis-28417	177	11	such	such	ADJ
bracis-28417	177	12	as	as	ADP
bracis-28417	177	13	assin	assin	PROPN
bracis-28417	177	14	2	2	NUM
bracis-28417	177	15	rte	rte	PROPN
bracis-28417	177	16	,	,	PUNCT
bracis-28417	177	17	enem	enem	NOUN
bracis-28417	177	18	challenge	challenge	NOUN
bracis-28417	177	19	,	,	PUNCT
bracis-28417	177	20	and	and	CCONJ
bracis-28417	177	21	faquad	faquad	ADJ
bracis-28417	177	22	.	.	PUNCT
bracis-28417	178	1	5.2	5.2	NUM
bracis-28417	178	2	data	datum	NOUN
bracis-28417	178	3	contamination	contamination	VERB
bracis-28417	178	4	the	the	DET
bracis-28417	178	5	pretraining	pretraine	VERB
bracis-28417	178	6	data	datum	NOUN
bracis-28417	178	7	for	for	ADP
bracis-28417	178	8	sabiá	sabiá	NOUN
bracis-28417	178	9	models	model	NOUN
bracis-28417	178	10	were	be	AUX
bracis-28417	178	11	collected	collect	VERB
bracis-28417	178	12	up	up	ADP
bracis-28417	178	13	until	until	ADP
bracis-28417	178	14	february	february	PROPN
bracis-28417	178	15	2022	2022	NUM
bracis-28417	178	16	.	.	PUNCT
bracis-28417	179	1	since	since	SCONJ
bracis-28417	179	2	enem	enem	PROPN
bracis-28417	179	3	2022	2022	NUM
bracis-28417	179	4	was	be	AUX
bracis-28417	179	5	publicly	publicly	ADV
bracis-28417	179	6	released	release	VERB
bracis-28417	179	7	in	in	ADP
bracis-28417	179	8	november	november	PROPN
bracis-28417	179	9	2022	2022	NUM
bracis-28417	179	10	,	,	PUNCT
bracis-28417	179	11	the	the	DET
bracis-28417	179	12	model	model	NOUN
bracis-28417	179	13	could	could	AUX
bracis-28417	179	14	not	not	PART
bracis-28417	179	15	have	have	VERB
bracis-28417	179	16	access	access	NOUN
bracis-28417	179	17	to	to	ADP
bracis-28417	179	18	the	the	DET
bracis-28417	179	19	answers	answer	NOUN
bracis-28417	179	20	for	for	ADP
bracis-28417	179	21	the	the	DET
bracis-28417	179	22	questions	question	NOUN
bracis-28417	179	23	present	present	ADJ
bracis-28417	179	24	within	within	ADP
bracis-28417	179	25	its	its	PRON
bracis-28417	179	26	pretraining	pretraine	VERB
bracis-28417	179	27	data	datum	NOUN
bracis-28417	179	28	.	.	PUNCT
bracis-28417	180	1	consequently	consequently	ADV
bracis-28417	180	2	,	,	PUNCT
bracis-28417	180	3	the	the	DET
bracis-28417	180	4	improvements	improvement	NOUN
bracis-28417	180	5	observed	observe	VERB
bracis-28417	180	6	at	at	ADP
bracis-28417	180	7	least	least	ADJ
bracis-28417	180	8	for	for	ADP
bracis-28417	180	9	enem	enem	PROPN
bracis-28417	180	10	2022	2022	NUM
bracis-28417	180	11	,	,	PUNCT
bracis-28417	180	12	which	which	PRON
bracis-28417	180	13	were	be	AUX
bracis-28417	180	14	higher	high	ADJ
bracis-28417	180	15	than	than	ADP
bracis-28417	180	16	the	the	DET
bracis-28417	180	17	average	average	NOUN
bracis-28417	180	18	of	of	ADP
bracis-28417	180	19	the	the	DET
bracis-28417	180	20	datasets	dataset	NOUN
bracis-28417	180	21	,	,	PUNCT
bracis-28417	180	22	can	can	AUX
bracis-28417	180	23	not	not	PART
bracis-28417	180	24	be	be	AUX
bracis-28417	180	25	attributed	attribute	VERB
bracis-28417	180	26	to	to	ADP
bracis-28417	180	27	data	datum	NOUN
bracis-28417	180	28	contamination	contamination	NOUN
bracis-28417	180	29	.	.	PUNCT
bracis-28417	181	1	however	however	ADV
bracis-28417	181	2	,	,	PUNCT
bracis-28417	181	3	for	for	ADP
bracis-28417	181	4	the	the	DET
bracis-28417	181	5	other	other	ADJ
bracis-28417	181	6	datasets	dataset	NOUN
bracis-28417	181	7	,	,	PUNCT
bracis-28417	181	8	the	the	DET
bracis-28417	181	9	possibility	possibility	NOUN
bracis-28417	181	10	of	of	ADP
bracis-28417	181	11	data	datum	NOUN
bracis-28417	181	12	contamination	contamination	NOUN
bracis-28417	181	13	can	can	AUX
bracis-28417	181	14	not	not	PART
bracis-28417	181	15	be	be	AUX
bracis-28417	181	16	ruled	rule	VERB
bracis-28417	181	17	out	out	ADP
bracis-28417	181	18	.	.	PUNCT
bracis-28417	182	1	5.3	5.3	NUM
bracis-28417	182	2	ablation	ablation	NOUN
bracis-28417	182	3	:	:	PUNCT
bracis-28417	182	4	english	english	ADJ
bracis-28417	182	5	datasets	dataset	NOUN
bracis-28417	182	6	in	in	ADP
bracis-28417	182	7	this	this	DET
bracis-28417	182	8	ablation	ablation	NOUN
bracis-28417	182	9	study	study	NOUN
bracis-28417	182	10	,	,	PUNCT
bracis-28417	182	11	we	we	PRON
bracis-28417	182	12	investigate	investigate	VERB
bracis-28417	182	13	the	the	DET
bracis-28417	182	14	potential	potential	ADJ
bracis-28417	182	15	impact	impact	NOUN
bracis-28417	182	16	of	of	ADP
bracis-28417	182	17	portuguese	portuguese	NOUN
bracis-28417	182	18	pretraining	pretraine	VERB
bracis-28417	182	19	on	on	ADP
bracis-28417	182	20	the	the	DET
bracis-28417	182	21	performance	performance	NOUN
bracis-28417	182	22	of	of	ADP
bracis-28417	182	23	the	the	DET
bracis-28417	182	24	model	model	NOUN
bracis-28417	182	25	in	in	ADP
bracis-28417	182	26	english	english	ADJ
bracis-28417	182	27	datasets	dataset	NOUN
bracis-28417	182	28	.	.	PUNCT
bracis-28417	183	1	we	we	PRON
bracis-28417	183	2	evaluated	evaluate	VERB
bracis-28417	183	3	the	the	DET
bracis-28417	183	4	llama-7b	llama-7b	PROPN
bracis-28417	183	5	and	and	CCONJ
bracis-28417	183	6	the	the	DET
bracis-28417	183	7	sabiá-7b	sabiá-7b	ADJ
bracis-28417	183	8	models	model	NOUN
bracis-28417	183	9	in	in	ADP
bracis-28417	183	10	english	english	ADJ
bracis-28417	183	11	multiple	multiple	ADJ
bracis-28417	183	12	-	-	PUNCT
bracis-28417	183	13	choice	choice	NOUN
bracis-28417	183	14	tasks	task	NOUN
bracis-28417	183	15	.	.	PUNCT
bracis-28417	184	1	for	for	ADP
bracis-28417	184	2	simplicity	simplicity	NOUN
bracis-28417	184	3	,	,	PUNCT
bracis-28417	184	4	we	we	PRON
bracis-28417	184	5	employed	employ	VERB
bracis-28417	184	6	a	a	DET
bracis-28417	184	7	few	few	ADJ
bracis-28417	184	8	-	-	PUNCT
bracis-28417	184	9	shot	shot	NOUN
bracis-28417	184	10	evaluation	evaluation	NOUN
bracis-28417	184	11	setup	setup	NOUN
bracis-28417	184	12	with	with	ADP
bracis-28417	184	13	10	10	NUM
bracis-28417	184	14	randomly	randomly	ADV
bracis-28417	184	15	selected	select	VERB
bracis-28417	184	16	examples	example	NOUN
bracis-28417	184	17	(	(	PUNCT
bracis-28417	184	18	dynamic	dynamic	ADV
bracis-28417	184	19	-	-	PUNCT
bracis-28417	184	20	sampled	sample	VERB
bracis-28417	184	21	prompt	prompt	NOUN
bracis-28417	184	22	)	)	PUNCT
bracis-28417	184	23	.	.	PUNCT
bracis-28417	185	1	importantly	importantly	ADV
bracis-28417	185	2	,	,	PUNCT
bracis-28417	185	3	we	we	PRON
bracis-28417	185	4	did	do	AUX
bracis-28417	185	5	not	not	PART
bracis-28417	185	6	incorporate	incorporate	VERB
bracis-28417	185	7	any	any	DET
bracis-28417	185	8	descriptions	description	NOUN
bracis-28417	185	9	or	or	CCONJ
bracis-28417	185	10	include	include	VERB
bracis-28417	185	11	portuguese	portuguese	ADJ
bracis-28417	185	12	keywords	keyword	NOUN
bracis-28417	185	13	to	to	PART
bracis-28417	185	14	delimit	delimit	VERB
bracis-28417	185	15	the	the	DET
bracis-28417	185	16	few	few	ADJ
bracis-28417	185	17	-	-	PUNCT
bracis-28417	185	18	shot	shot	NOUN
bracis-28417	185	19	examples	example	NOUN
bracis-28417	185	20	.	.	PUNCT
bracis-28417	186	1	we	we	PRON
bracis-28417	186	2	also	also	ADV
bracis-28417	186	3	restricted	restrict	VERB
bracis-28417	186	4	all	all	DET
bracis-28417	186	5	the	the	DET
bracis-28417	186	6	datasets	dataset	NOUN
bracis-28417	186	7	to	to	ADP
bracis-28417	186	8	350	350	NUM
bracis-28417	186	9	test	test	NOUN
bracis-28417	186	10	examples	example	NOUN
bracis-28417	186	11	.	.	PUNCT
bracis-28417	187	1	following	follow	VERB
bracis-28417	187	2	llama	llama	PROPN
bracis-28417	187	3	’s	’s	NOUN
bracis-28417	187	4	 	 	SPACE
bracis-28417	187	5	[	[	X
bracis-28417	187	6	63	63	NUM
bracis-28417	187	7	]	]	PUNCT
bracis-28417	187	8	approach	approach	NOUN
bracis-28417	187	9	,	,	PUNCT
bracis-28417	187	10	given	give	VERB
bracis-28417	187	11	the	the	DET
bracis-28417	187	12	provided	provide	VERB
bracis-28417	187	13	context	context	NOUN
bracis-28417	187	14	,	,	PUNCT
bracis-28417	187	15	we	we	PRON
bracis-28417	187	16	select	select	VERB
bracis-28417	187	17	the	the	DET
bracis-28417	187	18	answer	answer	NOUN
bracis-28417	187	19	with	with	ADP
bracis-28417	187	20	the	the	DET
bracis-28417	187	21	highest	high	ADJ
bracis-28417	187	22	likelihood	likelihood	NOUN
bracis-28417	187	23	normalized	normalize	VERB
bracis-28417	187	24	by	by	ADP
bracis-28417	187	25	the	the	DET
bracis-28417	187	26	number	number	NOUN
bracis-28417	187	27	of	of	ADP
bracis-28417	187	28	characters	character	NOUN
bracis-28417	187	29	.	.	PUNCT
bracis-28417	188	1	the	the	DET
bracis-28417	188	2	results	result	NOUN
bracis-28417	188	3	in	in	ADP
bracis-28417	188	4	table	table	NOUN
bracis-28417	188	5	 	 	SPACE
bracis-28417	188	6	4	4	NUM
bracis-28417	188	7	indicate	indicate	VERB
bracis-28417	188	8	that	that	SCONJ
bracis-28417	188	9	the	the	DET
bracis-28417	188	10	sabiá-7b	sabiá-7b	ADJ
bracis-28417	188	11	model	model	NOUN
bracis-28417	188	12	exhibits	exhibit	VERB
bracis-28417	188	13	a	a	DET
bracis-28417	188	14	slightly	slightly	ADV
bracis-28417	188	15	reduced	reduce	VERB
bracis-28417	188	16	performance	performance	NOUN
bracis-28417	188	17	in	in	ADP
bracis-28417	188	18	english	english	ADJ
bracis-28417	188	19	tasks	task	NOUN
bracis-28417	188	20	compared	compare	VERB
bracis-28417	188	21	to	to	ADP
bracis-28417	188	22	the	the	DET
bracis-28417	188	23	baseline	baseline	NOUN
bracis-28417	188	24	.	.	PUNCT
bracis-28417	189	1	this	this	DET
bracis-28417	189	2	result	result	NOUN
bracis-28417	189	3	corroborates	corroborate	VERB
bracis-28417	189	4	our	our	PRON
bracis-28417	189	5	premise	premise	NOUN
bracis-28417	189	6	that	that	SCONJ
bracis-28417	189	7	model	model	NOUN
bracis-28417	189	8	specialization	specialization	NOUN
bracis-28417	189	9	invariably	invariably	ADV
bracis-28417	189	10	entails	entail	VERB
bracis-28417	189	11	a	a	DET
bracis-28417	189	12	balancing	balance	VERB
bracis-28417	189	13	act	act	NOUN
bracis-28417	189	14	,	,	PUNCT
bracis-28417	189	15	where	where	SCONJ
bracis-28417	189	16	improvements	improvement	NOUN
bracis-28417	189	17	in	in	ADP
bracis-28417	189	18	one	one	NUM
bracis-28417	189	19	domain	domain	NOUN
bracis-28417	189	20	frequently	frequently	ADV
bracis-28417	189	21	coincide	coincide	VERB
bracis-28417	189	22	with	with	ADP
bracis-28417	189	23	degradation	degradation	NOUN
bracis-28417	189	24	in	in	ADP
bracis-28417	189	25	another	another	PRON
bracis-28417	189	26	.	.	PUNCT
bracis-28417	190	1	table	table	NOUN
bracis-28417	190	2	4	4	NUM
bracis-28417	190	3	.	.	PUNCT
bracis-28417	190	4	results	result	NOUN
bracis-28417	190	5	in	in	ADP
bracis-28417	190	6	english	english	ADJ
bracis-28417	190	7	datasets.full	datasets.full	NUM
bracis-28417	190	8	size	size	NOUN
bracis-28417	190	9	table	table	NOUN
bracis-28417	190	10	6	6	NUM
bracis-28417	190	11	limitations	limitation	NOUN
bracis-28417	190	12	owing	owe	VERB
bracis-28417	190	13	to	to	ADP
bracis-28417	190	14	the	the	DET
bracis-28417	190	15	financial	financial	ADJ
bracis-28417	190	16	constraints	constraint	NOUN
bracis-28417	190	17	associated	associate	VERB
bracis-28417	190	18	with	with	ADP
bracis-28417	190	19	pretraining	pretraine	VERB
bracis-28417	190	20	and	and	CCONJ
bracis-28417	190	21	,	,	PUNCT
bracis-28417	190	22	more	more	ADV
bracis-28417	190	23	significantly	significantly	ADV
bracis-28417	190	24	,	,	PUNCT
bracis-28417	190	25	the	the	DET
bracis-28417	190	26	manual	manual	ADJ
bracis-28417	190	27	labor	labor	NOUN
bracis-28417	190	28	involved	involve	VERB
bracis-28417	190	29	in	in	ADP
bracis-28417	190	30	collecting	collect	VERB
bracis-28417	190	31	and	and	CCONJ
bracis-28417	190	32	curating	curate	VERB
bracis-28417	190	33	evaluation	evaluation	NOUN
bracis-28417	190	34	datasets	dataset	NOUN
bracis-28417	190	35	,	,	PUNCT
bracis-28417	190	36	experiments	experiment	NOUN
bracis-28417	190	37	were	be	AUX
bracis-28417	190	38	conducted	conduct	VERB
bracis-28417	190	39	exclusively	exclusively	ADV
bracis-28417	190	40	in	in	ADP
bracis-28417	190	41	portuguese	portuguese	NOUN
bracis-28417	190	42	.	.	PUNCT
bracis-28417	191	1	given	give	VERB
bracis-28417	191	2	that	that	SCONJ
bracis-28417	191	3	our	our	PRON
bracis-28417	191	4	models	model	NOUN
bracis-28417	191	5	started	start	VERB
bracis-28417	191	6	pretraining	pretraine	VERB
bracis-28417	191	7	from	from	ADP
bracis-28417	191	8	english	english	PROPN
bracis-28417	191	9	-	-	PUNCT
bracis-28417	191	10	pretrained	pretraine	VERB
bracis-28417	191	11	models	model	NOUN
bracis-28417	191	12	and	and	CCONJ
bracis-28417	191	13	that	that	SCONJ
bracis-28417	191	14	portuguese	portuguese	ADJ
bracis-28417	191	15	and	and	CCONJ
bracis-28417	191	16	english	english	PROPN
bracis-28417	191	17	exhibit	exhibit	VERB
bracis-28417	191	18	relatively	relatively	ADV
bracis-28417	191	19	close	close	ADJ
bracis-28417	191	20	linguistic	linguistic	ADJ
bracis-28417	191	21	proximity	proximity	NOUN
bracis-28417	191	22	,	,	PUNCT
bracis-28417	191	23	we	we	PRON
bracis-28417	191	24	anticipate	anticipate	VERB
bracis-28417	191	25	that	that	SCONJ
bracis-28417	191	26	other	other	ADJ
bracis-28417	191	27	researchers	researcher	NOUN
bracis-28417	191	28	conducting	conduct	VERB
bracis-28417	191	29	further	far	ADV
bracis-28417	191	30	pretraining	pretraine	VERB
bracis-28417	191	31	on	on	ADP
bracis-28417	191	32	languages	language	NOUN
bracis-28417	191	33	closely	closely	ADV
bracis-28417	191	34	related	relate	VERB
bracis-28417	191	35	to	to	ADP
bracis-28417	191	36	english	english	PROPN
bracis-28417	191	37	will	will	AUX
bracis-28417	191	38	observe	observe	VERB
bracis-28417	191	39	comparable	comparable	ADJ
bracis-28417	191	40	improvements	improvement	NOUN
bracis-28417	191	41	in	in	ADP
bracis-28417	191	42	their	their	PRON
bracis-28417	191	43	target	target	NOUN
bracis-28417	191	44	tasks	task	NOUN
bracis-28417	191	45	.	.	PUNCT
bracis-28417	192	1	however	however	ADV
bracis-28417	192	2	,	,	PUNCT
bracis-28417	192	3	determining	determine	VERB
bracis-28417	192	4	whether	whether	SCONJ
bracis-28417	192	5	the	the	DET
bracis-28417	192	6	benefits	benefit	NOUN
bracis-28417	192	7	of	of	ADP
bracis-28417	192	8	this	this	DET
bracis-28417	192	9	method	method	NOUN
bracis-28417	192	10	persist	persist	NOUN
bracis-28417	192	11	for	for	ADP
bracis-28417	192	12	languages	language	NOUN
bracis-28417	192	13	more	more	ADV
bracis-28417	192	14	distant	distant	ADJ
bracis-28417	192	15	from	from	ADP
bracis-28417	192	16	english	english	PROPN
bracis-28417	192	17	remains	remain	VERB
bracis-28417	192	18	an	an	DET
bracis-28417	192	19	open	open	ADJ
bracis-28417	192	20	research	research	NOUN
bracis-28417	192	21	question	question	NOUN
bracis-28417	192	22	.	.	PUNCT
bracis-28417	193	1	portuguese	portuguese	PROPN
bracis-28417	193	2	is	be	AUX
bracis-28417	193	3	a	a	DET
bracis-28417	193	4	language	language	NOUN
bracis-28417	193	5	with	with	ADP
bracis-28417	193	6	an	an	DET
bracis-28417	193	7	abundance	abundance	NOUN
bracis-28417	193	8	of	of	ADP
bracis-28417	193	9	high	high	ADJ
bracis-28417	193	10	-	-	PUNCT
bracis-28417	193	11	quality	quality	NOUN
bracis-28417	193	12	web	web	NOUN
bracis-28417	193	13	-	-	PUNCT
bracis-28417	193	14	based	base	VERB
bracis-28417	193	15	texts	text	NOUN
bracis-28417	193	16	.	.	PUNCT
bracis-28417	194	1	thus	thus	ADV
bracis-28417	194	2	,	,	PUNCT
bracis-28417	194	3	the	the	DET
bracis-28417	194	4	gains	gain	NOUN
bracis-28417	194	5	observed	observe	VERB
bracis-28417	194	6	with	with	ADP
bracis-28417	194	7	the	the	DET
bracis-28417	194	8	proposed	propose	VERB
bracis-28417	194	9	method	method	NOUN
bracis-28417	194	10	may	may	AUX
bracis-28417	194	11	not	not	PART
bracis-28417	194	12	necessarily	necessarily	ADV
bracis-28417	194	13	extend	extend	VERB
bracis-28417	194	14	to	to	ADP
bracis-28417	194	15	low	low	ADJ
bracis-28417	194	16	-	-	PUNCT
bracis-28417	194	17	resource	resource	NOUN
bracis-28417	194	18	languages	language	NOUN
bracis-28417	194	19	with	with	ADP
bracis-28417	194	20	limited	limited	ADJ
bracis-28417	194	21	availability	availability	NOUN
bracis-28417	194	22	of	of	ADP
bracis-28417	194	23	quality	quality	NOUN
bracis-28417	194	24	texts	text	NOUN
bracis-28417	194	25	.	.	PUNCT
bracis-28417	195	1	in	in	ADP
bracis-28417	195	2	such	such	ADJ
bracis-28417	195	3	cases	case	NOUN
bracis-28417	195	4	,	,	PUNCT
bracis-28417	195	5	parameter	parameter	NOUN
bracis-28417	195	6	-	-	PUNCT
bracis-28417	195	7	efficient	efficient	ADJ
bracis-28417	195	8	methods	method	NOUN
bracis-28417	195	9	 	 	SPACE
bracis-28417	196	1	[	[	X
bracis-28417	196	2	19	19	NUM
bracis-28417	196	3	,	,	PUNCT
bracis-28417	196	4	42	42	NUM
bracis-28417	196	5	,	,	PUNCT
bracis-28417	196	6	43	43	NUM
bracis-28417	196	7	]	]	PUNCT
bracis-28417	196	8	could	could	AUX
bracis-28417	196	9	be	be	AUX
bracis-28417	196	10	advantageous	advantageous	ADJ
bracis-28417	196	11	,	,	PUNCT
bracis-28417	196	12	as	as	SCONJ
bracis-28417	196	13	evidenced	evidence	VERB
bracis-28417	196	14	by	by	ADP
bracis-28417	196	15	yong	yong	PROPN
bracis-28417	196	16	et	et	PROPN
bracis-28417	196	17	al	al	PROPN
bracis-28417	196	18	.	.	PUNCT
bracis-28417	196	19	 	 	SPACE
bracis-28417	197	1	[	[	X
bracis-28417	197	2	72	72	NUM
bracis-28417	197	3	]	]	PUNCT
bracis-28417	197	4	.	.	PUNCT
bracis-28417	198	1	we	we	PRON
bracis-28417	198	2	did	do	AUX
bracis-28417	198	3	not	not	PART
bracis-28417	198	4	use	use	VERB
bracis-28417	198	5	these	these	DET
bracis-28417	198	6	techniques	technique	NOUN
bracis-28417	198	7	in	in	ADP
bracis-28417	198	8	this	this	DET
bracis-28417	198	9	study	study	NOUN
bracis-28417	198	10	due	due	ADP
bracis-28417	198	11	to	to	ADP
bracis-28417	198	12	the	the	DET
bracis-28417	198	13	training	training	NOUN
bracis-28417	198	14	costs	cost	NOUN
bracis-28417	198	15	,	,	PUNCT
bracis-28417	198	16	which	which	PRON
bracis-28417	198	17	are	be	AUX
bracis-28417	198	18	approximately	approximately	ADV
bracis-28417	198	19	equivalent	equivalent	ADJ
bracis-28417	198	20	to	to	ADP
bracis-28417	198	21	training	train	VERB
bracis-28417	198	22	the	the	DET
bracis-28417	198	23	entire	entire	ADJ
bracis-28417	198	24	model.footnote	model.footnote	NOUN
bracis-28417	198	25	9	9	NUM
bracis-28417	198	26	7	7	NUM
bracis-28417	198	27	conclusion	conclusion	NOUN
bracis-28417	198	28	in	in	ADP
bracis-28417	198	29	this	this	DET
bracis-28417	198	30	study	study	NOUN
bracis-28417	198	31	,	,	PUNCT
bracis-28417	198	32	we	we	PRON
bracis-28417	198	33	contributed	contribute	VERB
bracis-28417	198	34	to	to	ADP
bracis-28417	198	35	the	the	DET
bracis-28417	198	36	expanding	expand	VERB
bracis-28417	198	37	body	body	NOUN
bracis-28417	198	38	of	of	ADP
bracis-28417	198	39	scientific	scientific	ADJ
bracis-28417	198	40	evidence	evidence	NOUN
bracis-28417	198	41	that	that	SCONJ
bracis-28417	198	42	specializing	specialize	VERB
bracis-28417	198	43	models	model	NOUN
bracis-28417	198	44	for	for	ADP
bracis-28417	198	45	individual	individual	ADJ
bracis-28417	198	46	languages	language	NOUN
bracis-28417	198	47	leads	lead	VERB
bracis-28417	198	48	to	to	ADP
bracis-28417	198	49	improvements	improvement	NOUN
bracis-28417	198	50	,	,	PUNCT
bracis-28417	198	51	even	even	ADV
bracis-28417	198	52	when	when	SCONJ
bracis-28417	198	53	the	the	DET
bracis-28417	198	54	baseline	baseline	NOUN
bracis-28417	198	55	model	model	NOUN
bracis-28417	198	56	is	be	AUX
bracis-28417	198	57	large	large	ADJ
bracis-28417	198	58	and	and	CCONJ
bracis-28417	198	59	extensively	extensively	ADV
bracis-28417	198	60	trained	train	VERB
bracis-28417	198	61	.	.	PUNCT
bracis-28417	199	1	we	we	PRON
bracis-28417	199	2	achieved	achieve	VERB
bracis-28417	199	3	this	this	PRON
bracis-28417	199	4	for	for	ADP
bracis-28417	199	5	the	the	DET
bracis-28417	199	6	portuguese	portuguese	ADJ
bracis-28417	199	7	language	language	NOUN
bracis-28417	199	8	utilizing	utilize	VERB
bracis-28417	199	9	a	a	DET
bracis-28417	199	10	near	near	ADJ
bracis-28417	199	11	state	state	NOUN
bracis-28417	199	12	-	-	PUNCT
bracis-28417	199	13	of	of	ADP
bracis-28417	199	14	-	-	PUNCT
bracis-28417	199	15	the	the	DET
bracis-28417	199	16	-	-	PUNCT
bracis-28417	199	17	art	art	NOUN
bracis-28417	199	18	model	model	NOUN
bracis-28417	199	19	with	with	ADP
bracis-28417	199	20	65	65	NUM
bracis-28417	199	21	billion	billion	NUM
bracis-28417	199	22	parameters	parameter	NOUN
bracis-28417	199	23	.	.	PUNCT
bracis-28417	200	1	given	give	VERB
bracis-28417	200	2	the	the	DET
bracis-28417	200	3	relatively	relatively	ADV
bracis-28417	200	4	low	low	ADJ
bracis-28417	200	5	pretraining	pretraine	VERB
bracis-28417	200	6	cost	cost	NOUN
bracis-28417	200	7	and	and	CCONJ
bracis-28417	200	8	significant	significant	ADJ
bracis-28417	200	9	performance	performance	NOUN
bracis-28417	200	10	gains	gain	NOUN
bracis-28417	200	11	observed	observe	VERB
bracis-28417	200	12	,	,	PUNCT
bracis-28417	200	13	we	we	PRON
bracis-28417	200	14	foresee	foresee	VERB
bracis-28417	200	15	a	a	DET
bracis-28417	200	16	future	future	ADJ
bracis-28417	200	17	landscape	landscape	NOUN
bracis-28417	200	18	consisting	consist	VERB
bracis-28417	200	19	of	of	ADP
bracis-28417	200	20	a	a	DET
bracis-28417	200	21	diverse	diverse	ADJ
bracis-28417	200	22	array	array	NOUN
bracis-28417	200	23	of	of	ADP
bracis-28417	200	24	models	model	NOUN
bracis-28417	200	25	,	,	PUNCT
bracis-28417	200	26	each	each	PRON
bracis-28417	200	27	tailored	tailor	VERB
bracis-28417	200	28	to	to	ADP
bracis-28417	200	29	a	a	DET
bracis-28417	200	30	specific	specific	ADJ
bracis-28417	200	31	domain	domain	NOUN
bracis-28417	200	32	,	,	PUNCT
bracis-28417	200	33	rather	rather	ADV
bracis-28417	200	34	than	than	ADP
bracis-28417	200	35	a	a	DET
bracis-28417	200	36	single	single	ADJ
bracis-28417	200	37	,	,	PUNCT
bracis-28417	200	38	all	all	PRON
bracis-28417	200	39	-	-	PUNCT
bracis-28417	200	40	encompassing	encompass	VERB
bracis-28417	200	41	model	model	NOUN
bracis-28417	200	42	.	.	PUNCT
bracis-28417	201	1	notes	note	NOUN
bracis-28417	201	2	1.to	1.to	PRON
bracis-28417	201	3	serve	serve	VERB
bracis-28417	201	4	their	their	PRON
bracis-28417	201	5	ensemble	ensemble	ADJ
bracis-28417	201	6	with	with	ADP
bracis-28417	201	7	a	a	DET
bracis-28417	201	8	low	low	ADJ
bracis-28417	201	9	latency	latency	NOUN
bracis-28417	201	10	,	,	PUNCT
bracis-28417	201	11	the	the	DET
bracis-28417	201	12	weights	weight	NOUN
bracis-28417	201	13	for	for	ADP
bracis-28417	201	14	each	each	DET
bracis-28417	201	15	expert	expert	NOUN
bracis-28417	201	16	must	must	AUX
bracis-28417	201	17	be	be	AUX
bracis-28417	201	18	kept	keep	VERB
bracis-28417	201	19	in	in	ADP
bracis-28417	201	20	gpu	gpu	NOUN
bracis-28417	201	21	memory	memory	NOUN
bracis-28417	201	22	.	.	PUNCT
bracis-28417	202	1	2.ftfy	2.ftfy	NUM
bracis-28417	202	2	normalization	normalization	NOUN
bracis-28417	202	3	fixes	fix	VERB
bracis-28417	202	4	mojibakes	mojibake	NOUN
bracis-28417	202	5	and	and	CCONJ
bracis-28417	202	6	remove	remove	VERB
bracis-28417	202	7	remnant	remnant	ADJ
bracis-28417	202	8	html	html	NOUN
bracis-28417	202	9	tags	tag	NOUN
bracis-28417	202	10	.	.	PUNCT
bracis-28417	203	1	3.sabiá	3.sabiá	NUM
bracis-28417	203	2	is	be	AUX
bracis-28417	203	3	a	a	DET
bracis-28417	203	4	tribute	tribute	NOUN
bracis-28417	203	5	to	to	ADP
bracis-28417	203	6	the	the	DET
bracis-28417	203	7	eponymous	eponymous	ADJ
bracis-28417	203	8	bird	bird	NOUN
bracis-28417	203	9	,	,	PUNCT
bracis-28417	203	10	renowned	renowne	VERB
bracis-28417	203	11	for	for	ADP
bracis-28417	203	12	its	its	PRON
bracis-28417	203	13	diverse	diverse	ADJ
bracis-28417	203	14	and	and	CCONJ
bracis-28417	203	15	intricate	intricate	ADJ
bracis-28417	203	16	vocalizations	vocalization	NOUN
bracis-28417	203	17	.	.	PUNCT
bracis-28417	204	1	4.considering	4.considere	VERB
bracis-28417	204	2	the	the	DET
bracis-28417	204	3	on	on	ADP
bracis-28417	204	4	-	-	PUNCT
bracis-28417	204	5	demand	demand	NOUN
bracis-28417	204	6	pricing	pricing	NOUN
bracis-28417	204	7	of	of	ADP
bracis-28417	204	8	384	384	NUM
bracis-28417	204	9	usd	usd	NOUN
bracis-28417	204	10	per	per	ADP
bracis-28417	204	11	hour	hour	NOUN
bracis-28417	204	12	for	for	ADP
bracis-28417	204	13	a	a	DET
bracis-28417	204	14	tpu	tpu	NOUN
bracis-28417	204	15	v2	v2	PROPN
bracis-28417	204	16	-	-	PUNCT
bracis-28417	204	17	512	512	NUM
bracis-28417	204	18	,	,	PUNCT
bracis-28417	204	19	pretraining	pretraine	VERB
bracis-28417	204	20	sabiá-7b	sabiá-7b	ADV
bracis-28417	204	21	and	and	CCONJ
bracis-28417	204	22	sabiá-65b	sabiá-65b	PROPN
bracis-28417	204	23	costs	cost	VERB
bracis-28417	204	24	approximately	approximately	ADV
bracis-28417	204	25	9,000	9,000	NUM
bracis-28417	204	26	and	and	CCONJ
bracis-28417	204	27	80,000	80,000	NUM
bracis-28417	204	28	usd	usd	NOUN
bracis-28417	204	29	,	,	PUNCT
bracis-28417	204	30	respectively	respectively	ADV
bracis-28417	204	31	.	.	PUNCT
bracis-28417	205	1	5.https://huggingface.co/eleutherai/gpt-j-6b	5.https://huggingface.co/eleutherai/gpt-j-6b	NUM
bracis-28417	205	2	.	.	PUNCT
bracis-28417	206	1	6.due	6.due	NUM
bracis-28417	206	2	to	to	ADP
bracis-28417	206	3	constraints	constraint	NOUN
bracis-28417	206	4	in	in	ADP
bracis-28417	206	5	our	our	PRON
bracis-28417	206	6	hardware	hardware	NOUN
bracis-28417	206	7	budget	budget	NOUN
bracis-28417	206	8	,	,	PUNCT
bracis-28417	206	9	this	this	DET
bracis-28417	206	10	model	model	NOUN
bracis-28417	206	11	was	be	AUX
bracis-28417	206	12	trained	train	VERB
bracis-28417	206	13	with	with	ADP
bracis-28417	206	14	fewer	few	ADJ
bracis-28417	206	15	tokens	token	NOUN
bracis-28417	206	16	compared	compare	VERB
bracis-28417	206	17	to	to	AUX
bracis-28417	206	18	sabiá	sabiá	VERB
bracis-28417	206	19	.	.	PUNCT
bracis-28417	207	1	7.the	7.the	DET
bracis-28417	207	2	massive	massive	ADJ
bracis-28417	207	3	dataset	dataset	ADJ
bracis-28417	207	4	underwent	underwent	NOUN
bracis-28417	207	5	manual	manual	ADJ
bracis-28417	207	6	translation	translation	NOUN
bracis-28417	207	7	and	and	CCONJ
bracis-28417	207	8	localization	localization	NOUN
bracis-28417	207	9	;	;	PUNCT
bracis-28417	207	10	however	however	ADV
bracis-28417	207	11	,	,	PUNCT
bracis-28417	207	12	given	give	VERB
bracis-28417	207	13	that	that	SCONJ
bracis-28417	207	14	the	the	DET
bracis-28417	207	15	original	original	ADJ
bracis-28417	207	16	text	text	NOUN
bracis-28417	207	17	was	be	AUX
bracis-28417	207	18	composed	compose	VERB
bracis-28417	207	19	in	in	ADP
bracis-28417	207	20	english	english	PROPN
bracis-28417	207	21	,	,	PUNCT
bracis-28417	207	22	it	it	PRON
bracis-28417	207	23	has	have	AUX
bracis-28417	207	24	been	be	AUX
bracis-28417	207	25	categorized	categorize	VERB
bracis-28417	207	26	as	as	ADP
bracis-28417	207	27	a	a	DET
bracis-28417	207	28	translated	translate	VERB
bracis-28417	207	29	dataset	dataset	NOUN
bracis-28417	207	30	.	.	PUNCT
bracis-28417	208	1	8.this	8.this	NUM
bracis-28417	208	2	model	model	NOUN
bracis-28417	208	3	was	be	AUX
bracis-28417	208	4	used	use	VERB
bracis-28417	208	5	in	in	ADP
bracis-28417	208	6	the	the	DET
bracis-28417	208	7	experiments	experiment	NOUN
bracis-28417	208	8	:	:	PUNCT
bracis-28417	208	9	https://huggingface.co/bigscience/bloomz-7b1-mt	https://huggingface.co/bigscience/bloomz-7b1-mt	X
bracis-28417	208	10	.	.	PUNCT
bracis-28417	209	1	9.although	9.although	NUM
bracis-28417	209	2	parameter	parameter	NOUN
bracis-28417	209	3	-	-	PUNCT
bracis-28417	209	4	efficient	efficient	ADJ
bracis-28417	209	5	methods	method	NOUN
bracis-28417	209	6	adjust	adjust	VERB
bracis-28417	209	7	only	only	ADV
bracis-28417	209	8	a	a	DET
bracis-28417	209	9	fraction	fraction	NOUN
bracis-28417	209	10	of	of	ADP
bracis-28417	209	11	the	the	DET
bracis-28417	209	12	weights	weight	NOUN
bracis-28417	209	13	,	,	PUNCT
bracis-28417	209	14	they	they	PRON
bracis-28417	209	15	use	use	VERB
bracis-28417	209	16	only	only	ADV
bracis-28417	209	17	marginally	marginally	ADV
bracis-28417	209	18	fewer	few	ADJ
bracis-28417	209	19	training	training	NOUN
bracis-28417	209	20	flops	flop	NOUN
bracis-28417	209	21	,	,	PUNCT
bracis-28417	209	22	as	as	SCONJ
bracis-28417	209	23	activations	activation	NOUN
bracis-28417	209	24	and	and	CCONJ
bracis-28417	209	25	gradients	gradient	NOUN
bracis-28417	209	26	are	be	AUX
bracis-28417	209	27	computed	compute	VERB
bracis-28417	209	28	for	for	ADP
bracis-28417	209	29	the	the	DET
bracis-28417	209	30	entire	entire	ADJ
bracis-28417	209	31	model	model	NOUN
bracis-28417	209	32	.	.	PUNCT
bracis-28417	210	1	for	for	ADP
bracis-28417	210	2	instance	instance	NOUN
bracis-28417	210	3	,	,	PUNCT
bracis-28417	210	4	lora	lora	PROPN
bracis-28417	210	5	 	 	SPACE
bracis-28417	211	1	[	[	X
bracis-28417	211	2	20	20	NUM
bracis-28417	211	3	]	]	PUNCT
bracis-28417	211	4	,	,	PUNCT
bracis-28417	211	5	a	a	DET
bracis-28417	211	6	parameter	parameter	NOUN
bracis-28417	211	7	-	-	PUNCT
bracis-28417	211	8	efficient	efficient	ADJ
bracis-28417	211	9	method	method	NOUN
bracis-28417	211	10	,	,	PUNCT
bracis-28417	211	11	improves	improve	VERB
bracis-28417	211	12	training	train	VERB
bracis-28417	211	13	throughput	throughput	NOUN
bracis-28417	211	14	of	of	ADP
bracis-28417	211	15	a	a	DET
bracis-28417	211	16	gpt-3	gpt-3	NUM
bracis-28417	211	17	175b	175b	NUM
bracis-28417	211	18	model	model	NOUN
bracis-28417	211	19	by	by	ADP
bracis-28417	211	20	only	only	ADV
bracis-28417	211	21	nearly	nearly	ADV
bracis-28417	211	22	32	32	NUM
bracis-28417	211	23	%	%	NOUN
bracis-28417	211	24	.	.	PUNCT
bracis-28417	212	1	references	reference	NOUN
bracis-28417	212	2	almeida	almeida	PROPN
bracis-28417	212	3	,	,	PUNCT
bracis-28417	212	4	t.s	t.s	PROPN
bracis-28417	212	5	.	.	PROPN
bracis-28417	212	6	,	,	PUNCT
bracis-28417	212	7	laitz	laitz	PROPN
bracis-28417	212	8	,	,	PUNCT
bracis-28417	212	9	t.	t.	PROPN
bracis-28417	212	10	,	,	PUNCT
bracis-28417	212	11	bonás	bonás	PROPN
bracis-28417	212	12	,	,	PUNCT
bracis-28417	212	13	g.k	g.k	PROPN
bracis-28417	212	14	.	.	PROPN
bracis-28417	212	15	,	,	PUNCT
bracis-28417	212	16	nogueira	nogueira	PROPN
bracis-28417	212	17	,	,	PUNCT
bracis-28417	212	18	r.	r.	PROPN
bracis-28417	212	19	:	:	PUNCT
bracis-28417	212	20	bluex	bluex	PROPN
bracis-28417	212	21	:	:	PUNCT
bracis-28417	212	22	a	a	DET
bracis-28417	212	23	benchmark	benchmark	NOUN
bracis-28417	212	24	based	base	VERB
bracis-28417	212	25	on	on	ADP
bracis-28417	212	26	brazilian	brazilian	ADJ
bracis-28417	212	27	leading	lead	VERB
bracis-28417	212	28	universities	university	NOUN
bracis-28417	212	29	entrance	entrance	NOUN
bracis-28417	212	30	exams	exam	NOUN
bracis-28417	212	31	.	.	PUNCT
bracis-28417	213	1	to	to	PART
bracis-28417	213	2	appear	appear	VERB
bracis-28417	213	3	(	(	PUNCT
bracis-28417	213	4	2023	2023	NUM
bracis-28417	213	5	)	)	PUNCT
bracis-28417	213	6	google	google	NOUN
bracis-28417	213	7	scholar	scholar	NOUN
bracis-28417	213	8	  	  	SPACE
bracis-28417	213	9	antoun	antoun	NOUN
bracis-28417	213	10	,	,	PUNCT
bracis-28417	213	11	w.	w.	PROPN
bracis-28417	213	12	,	,	PUNCT
bracis-28417	213	13	baly	baly	PROPN
bracis-28417	213	14	,	,	PUNCT
bracis-28417	213	15	f.	f.	PROPN
bracis-28417	213	16	,	,	PUNCT
bracis-28417	213	17	hajj	hajj	PROPN
bracis-28417	213	18	,	,	PUNCT
bracis-28417	213	19	h.	h.	PROPN
bracis-28417	213	20	:	:	PUNCT
bracis-28417	213	21	arabert	arabert	PROPN
bracis-28417	213	22	:	:	PUNCT
bracis-28417	213	23	transformer	transformer	NOUN
bracis-28417	213	24	-	-	PUNCT
bracis-28417	213	25	based	base	VERB
bracis-28417	213	26	model	model	NOUN
bracis-28417	213	27	for	for	ADP
bracis-28417	213	28	arabic	arabic	ADJ
bracis-28417	213	29	language	language	NOUN
bracis-28417	213	30	understanding	understanding	NOUN
bracis-28417	213	31	.	.	PUNCT
bracis-28417	214	1	in	in	ADP
bracis-28417	214	2	:	:	PUNCT
bracis-28417	214	3	proceedings	proceeding	NOUN
bracis-28417	214	4	of	of	ADP
bracis-28417	214	5	the	the	DET
bracis-28417	214	6	4th	4th	ADJ
bracis-28417	214	7	workshop	workshop	NOUN
bracis-28417	214	8	on	on	ADP
bracis-28417	214	9	open	open	ADJ
bracis-28417	214	10	-	-	PUNCT
bracis-28417	214	11	source	source	NOUN
bracis-28417	214	12	arabic	arabic	ADJ
bracis-28417	214	13	corpora	corpora	NOUN
bracis-28417	214	14	and	and	CCONJ
bracis-28417	214	15	processing	processing	NOUN
bracis-28417	214	16	tools	tool	NOUN
bracis-28417	214	17	,	,	PUNCT
bracis-28417	214	18	with	with	ADP
bracis-28417	214	19	a	a	DET
bracis-28417	214	20	shared	share	VERB
bracis-28417	214	21	task	task	NOUN
bracis-28417	214	22	on	on	ADP
bracis-28417	214	23	offensive	offensive	ADJ
bracis-28417	214	24	language	language	NOUN
bracis-28417	214	25	detection	detection	NOUN
bracis-28417	214	26	.	.	PUNCT
bracis-28417	215	1	pp	pp	ADJ
bracis-28417	215	2	.	.	PUNCT
bracis-28417	216	1	9–15	9–15	PROPN
bracis-28417	216	2	.	.	PUNCT
bracis-28417	217	1	european	european	ADJ
bracis-28417	217	2	language	language	PROPN
bracis-28417	217	3	resource	resource	NOUN
bracis-28417	217	4	association	association	PROPN
bracis-28417	217	5	,	,	PUNCT
bracis-28417	217	6	marseille	marseille	PROPN
bracis-28417	217	7	,	,	PUNCT
bracis-28417	217	8	france	france	PROPN
bracis-28417	217	9	(	(	PUNCT
bracis-28417	217	10	2020	2020	NUM
bracis-28417	217	11	)	)	PUNCT
bracis-28417	217	12	google	google	PROPN
bracis-28417	217	13	scholar	scholar	NOUN
bracis-28417	217	14	  	  	SPACE
bracis-28417	217	15	barros	barros	PROPN
bracis-28417	217	16	,	,	PUNCT
bracis-28417	217	17	t.m.d	t.m.d	NOUN
bracis-28417	217	18	.	.	PROPN
bracis-28417	217	19	,	,	PUNCT
bracis-28417	217	20	et	et	PROPN
bracis-28417	217	21	al	al	PROPN
bracis-28417	217	22	.	.	PROPN
bracis-28417	217	23	:	:	PUNCT
bracis-28417	218	1	employing	employ	VERB
bracis-28417	218	2	transformers	transformer	NOUN
bracis-28417	218	3	and	and	CCONJ
bracis-28417	218	4	emoji	emoji	NOUN
bracis-28417	218	5	to	to	PART
bracis-28417	218	6	perform	perform	VERB
bracis-28417	218	7	sentiment	sentiment	NOUN
bracis-28417	218	8	classification	classification	NOUN
bracis-28417	218	9	of	of	ADP
bracis-28417	218	10	social	social	ADJ
bracis-28417	218	11	media	medium	NOUN
bracis-28417	218	12	texts	text	NOUN
bracis-28417	218	13	:	:	PUNCT
bracis-28417	218	14	utilizando	utilizando	PROPN
bracis-28417	218	15	transformers	transformer	NOUN
bracis-28417	218	16	e	e	PROPN
bracis-28417	218	17	emoji	emoji	PROPN
bracis-28417	218	18	na	na	PROPN
bracis-28417	218	19	classificação	classificação	PROPN
bracis-28417	218	20	de	de	PROPN
bracis-28417	218	21	sentimento	sentimento	PROPN
bracis-28417	218	22	de	de	X
bracis-28417	218	23	textos	textos	PROPN
bracis-28417	218	24	oriundos	oriundo	NOUN
bracis-28417	218	25	de	de	PROPN
bracis-28417	218	26	redes	redes	PROPN
bracis-28417	218	27	sociais	sociais	PROPN
bracis-28417	218	28	(	(	PUNCT
bracis-28417	218	29	2021	2021	NUM
bracis-28417	218	30	)	)	PUNCT
bracis-28417	218	31	google	google	PROPN
bracis-28417	218	32	scholar	scholar	NOUN
bracis-28417	218	33	  	  	SPACE
bracis-28417	218	34	bhattacharjee	bhattacharjee	PROPN
bracis-28417	218	35	,	,	PUNCT
bracis-28417	218	36	a.	a.	PROPN
bracis-28417	218	37	,	,	PUNCT
bracis-28417	218	38	et	et	PROPN
bracis-28417	218	39	al	al	PROPN
bracis-28417	218	40	.	.	PROPN
bracis-28417	218	41	:	:	PUNCT
bracis-28417	219	1	banglabert	banglabert	NOUN
bracis-28417	219	2	:	:	PUNCT
bracis-28417	219	3	language	language	NOUN
bracis-28417	219	4	model	model	NOUN
bracis-28417	219	5	pretraining	pretraine	VERB
bracis-28417	219	6	and	and	CCONJ
bracis-28417	219	7	benchmarks	benchmark	NOUN
bracis-28417	219	8	for	for	ADP
bracis-28417	219	9	low	low	ADJ
bracis-28417	219	10	-	-	PUNCT
bracis-28417	219	11	resource	resource	NOUN
bracis-28417	219	12	language	language	NOUN
bracis-28417	219	13	understanding	understanding	NOUN
bracis-28417	219	14	evaluation	evaluation	NOUN
bracis-28417	219	15	in	in	ADP
bracis-28417	219	16	bangla	bangla	PROPN
bracis-28417	219	17	.	.	PUNCT
bracis-28417	220	1	in	in	ADP
bracis-28417	220	2	:	:	PUNCT
bracis-28417	220	3	findings	finding	NOUN
bracis-28417	220	4	of	of	ADP
bracis-28417	220	5	the	the	DET
bracis-28417	220	6	association	association	NOUN
bracis-28417	220	7	for	for	ADP
bracis-28417	220	8	computational	computational	ADJ
bracis-28417	220	9	linguistics	linguistic	NOUN
bracis-28417	220	10	:	:	PUNCT
bracis-28417	220	11	naacl	naacl	PROPN
bracis-28417	220	12	2022	2022	NUM
bracis-28417	220	13	,	,	PUNCT
bracis-28417	220	14	pp	pp	PROPN
bracis-28417	220	15	.	.	PUNCT
bracis-28417	220	16	1318–1327	1318–1327	NUM
bracis-28417	220	17	.	.	PUNCT
bracis-28417	221	1	association	association	PROPN
bracis-28417	221	2	for	for	ADP
bracis-28417	221	3	computational	computational	ADJ
bracis-28417	221	4	linguistics	linguistic	NOUN
bracis-28417	221	5	,	,	PUNCT
bracis-28417	221	6	seattle	seattle	PROPN
bracis-28417	221	7	,	,	PUNCT
bracis-28417	221	8	united	united	PROPN
bracis-28417	221	9	states	states	PROPN
bracis-28417	221	10	(	(	PUNCT
bracis-28417	221	11	2022	2022	NUM
bracis-28417	221	12	)	)	PUNCT
bracis-28417	221	13	.	.	PUNCT
bracis-28417	222	1	https://doi.org/10.18653/v1/2022.findings-naacl.98	https://doi.org/10.18653/v1/2022.findings-naacl.98	PRON
bracis-28417	222	2	brum	brum	NOUN
bracis-28417	222	3	,	,	PUNCT
bracis-28417	222	4	h.	h.	PROPN
bracis-28417	222	5	,	,	PUNCT
bracis-28417	222	6	volpe	volpe	PROPN
bracis-28417	222	7	nunes	nunes	PROPN
bracis-28417	222	8	,	,	PUNCT
bracis-28417	222	9	m.d.g	m.d.g	NOUN
bracis-28417	222	10	.	.	PUNCT
bracis-28417	222	11	:	:	PUNCT
bracis-28417	223	1	building	build	VERB
bracis-28417	223	2	a	a	DET
bracis-28417	223	3	sentiment	sentiment	NOUN
bracis-28417	223	4	corpus	corpus	NOUN
bracis-28417	223	5	of	of	ADP
bracis-28417	223	6	tweets	tweet	NOUN
bracis-28417	223	7	in	in	ADP
bracis-28417	223	8	brazilian	brazilian	ADJ
bracis-28417	223	9	portuguese	portuguese	PROPN
bracis-28417	223	10	.	.	PUNCT
bracis-28417	224	1	in	in	ADP
bracis-28417	224	2	:	:	PUNCT
bracis-28417	224	3	proceedings	proceeding	NOUN
bracis-28417	224	4	of	of	ADP
bracis-28417	224	5	the	the	DET
bracis-28417	224	6	eleventh	eleventh	ADJ
bracis-28417	224	7	international	international	ADJ
bracis-28417	224	8	conference	conference	NOUN
bracis-28417	224	9	on	on	ADP
bracis-28417	224	10	language	language	NOUN
bracis-28417	224	11	resources	resource	NOUN
bracis-28417	224	12	and	and	CCONJ
bracis-28417	224	13	evaluation	evaluation	NOUN
bracis-28417	224	14	(	(	PUNCT
bracis-28417	224	15	lrec	lrec	PROPN
bracis-28417	224	16	2018	2018	NUM
bracis-28417	224	17	)	)	PUNCT
bracis-28417	224	18	.	.	PUNCT
bracis-28417	225	1	european	european	ADJ
bracis-28417	225	2	language	language	PROPN
bracis-28417	225	3	resources	resource	NOUN
bracis-28417	225	4	association	association	NOUN
bracis-28417	225	5	(	(	PUNCT
bracis-28417	225	6	elra	elra	ADV
bracis-28417	225	7	)	)	PUNCT
bracis-28417	225	8	,	,	PUNCT
bracis-28417	225	9	miyazaki	miyazaki	PROPN
bracis-28417	225	10	,	,	PUNCT
bracis-28417	225	11	japan	japan	PROPN
bracis-28417	225	12	(	(	PUNCT
bracis-28417	225	13	may	may	PROPN
bracis-28417	225	14	2018	2018	NUM
bracis-28417	225	15	)	)	PUNCT
bracis-28417	225	16	google	google	PROPN
bracis-28417	225	17	scholar	scholar	NOUN
bracis-28417	225	18	  	  	SPACE
bracis-28417	225	19	cañete	cañete	NOUN
bracis-28417	225	20	,	,	PUNCT
bracis-28417	225	21	j.	j.	PROPN
bracis-28417	225	22	,	,	PUNCT
bracis-28417	225	23	chaperon	chaperon	PROPN
bracis-28417	225	24	,	,	PUNCT
bracis-28417	225	25	g.	g.	PROPN
bracis-28417	225	26	,	,	PUNCT
bracis-28417	225	27	fuentes	fuentes	PROPN
bracis-28417	225	28	,	,	PUNCT
bracis-28417	225	29	r.	r.	PROPN
bracis-28417	225	30	,	,	PUNCT
bracis-28417	225	31	ho	ho	PROPN
bracis-28417	225	32	,	,	PUNCT
bracis-28417	225	33	j.h	j.h	PROPN
bracis-28417	225	34	.	.	PROPN
bracis-28417	225	35	,	,	PUNCT
bracis-28417	225	36	kang	kang	PROPN
bracis-28417	225	37	,	,	PUNCT
bracis-28417	225	38	h.	h.	PROPN
bracis-28417	225	39	,	,	PUNCT
bracis-28417	225	40	pérez	pérez	NOUN
bracis-28417	225	41	,	,	PUNCT
bracis-28417	225	42	j.	j.	PROPN
bracis-28417	225	43	:	:	PUNCT
bracis-28417	225	44	spanish	spanish	ADJ
bracis-28417	225	45	pre	pre	ADJ
bracis-28417	225	46	-	-	ADJ
bracis-28417	225	47	trained	train	VERB
bracis-28417	225	48	bert	bert	NOUN
bracis-28417	225	49	model	model	NOUN
bracis-28417	225	50	and	and	CCONJ
bracis-28417	225	51	evaluation	evaluation	NOUN
bracis-28417	225	52	data	datum	NOUN
bracis-28417	225	53	.	.	PUNCT
bracis-28417	226	1	in	in	ADP
bracis-28417	226	2	:	:	PUNCT
bracis-28417	226	3	pml4dc	pml4dc	VERB
bracis-28417	226	4	at	at	ADP
bracis-28417	226	5	iclr	iclr	NOUN
bracis-28417	226	6	2020	2020	NUM
bracis-28417	226	7	(	(	PUNCT
bracis-28417	226	8	2020	2020	NUM
bracis-28417	226	9	)	)	PUNCT
bracis-28417	226	10	google	google	PROPN
bracis-28417	226	11	scholar	scholar	NOUN
bracis-28417	226	12	  	  	SPACE
bracis-28417	226	13	carmo	carmo	PROPN
bracis-28417	226	14	,	,	PUNCT
bracis-28417	226	15	d.	d.	PROPN
bracis-28417	226	16	,	,	PUNCT
bracis-28417	226	17	piau	piau	NOUN
bracis-28417	226	18	,	,	PUNCT
bracis-28417	226	19	m.	m.	NOUN
bracis-28417	226	20	,	,	PUNCT
bracis-28417	226	21	campiotti	campiotti	PROPN
bracis-28417	226	22	,	,	PUNCT
bracis-28417	226	23	i.	i.	PROPN
bracis-28417	226	24	,	,	PUNCT
bracis-28417	226	25	nogueira	nogueira	PROPN
bracis-28417	226	26	,	,	PUNCT
bracis-28417	226	27	r.	r.	PROPN
bracis-28417	226	28	,	,	PUNCT
bracis-28417	226	29	lotufo	lotufo	PROPN
bracis-28417	226	30	,	,	PUNCT
bracis-28417	226	31	r.	r.	PROPN
bracis-28417	226	32	:	:	PUNCT
bracis-28417	226	33	ptt5	ptt5	PROPN
bracis-28417	226	34	:	:	PUNCT
bracis-28417	226	35	pretraining	pretraine	VERB
bracis-28417	226	36	and	and	CCONJ
bracis-28417	226	37	validating	validate	VERB
bracis-28417	226	38	the	the	DET
bracis-28417	226	39	t5	t5	PROPN
bracis-28417	226	40	model	model	NOUN
bracis-28417	226	41	on	on	ADP
bracis-28417	226	42	brazilian	brazilian	ADJ
bracis-28417	226	43	portuguese	portuguese	ADJ
bracis-28417	226	44	data	datum	NOUN
bracis-28417	226	45	.	.	PUNCT
bracis-28417	227	1	arxiv	arxiv	PROPN
bracis-28417	227	2	preprint	preprint	PROPN
bracis-28417	227	3	arxiv:2008.09144	arxiv:2008.09144	ADP
bracis-28417	227	4	(	(	PUNCT
bracis-28417	227	5	2020	2020	NUM
bracis-28417	227	6	)	)	PUNCT
bracis-28417	227	7	chan	chan	PROPN
bracis-28417	227	8	,	,	PUNCT
bracis-28417	227	9	b.	b.	PROPN
bracis-28417	227	10	,	,	PUNCT
bracis-28417	227	11	schweter	schweter	ADV
bracis-28417	227	12	,	,	PUNCT
bracis-28417	227	13	s.	s.	PROPN
bracis-28417	227	14	,	,	PUNCT
bracis-28417	227	15	möller	möller	NOUN
bracis-28417	227	16	,	,	PUNCT
bracis-28417	227	17	t.	t.	PROPN
bracis-28417	227	18	:	:	PUNCT
bracis-28417	227	19	german	german	PROPN
bracis-28417	227	20	’s	’s	PART
bracis-28417	227	21	next	next	ADJ
bracis-28417	227	22	language	language	NOUN
bracis-28417	227	23	model	model	NOUN
bracis-28417	227	24	.	.	PUNCT
bracis-28417	228	1	in	in	ADP
bracis-28417	228	2	:	:	PUNCT
bracis-28417	228	3	proceedings	proceeding	NOUN
bracis-28417	228	4	of	of	ADP
bracis-28417	228	5	the	the	DET
bracis-28417	228	6	28th	28th	ADJ
bracis-28417	228	7	international	international	ADJ
bracis-28417	228	8	conference	conference	NOUN
bracis-28417	228	9	on	on	ADP
bracis-28417	228	10	computational	computational	ADJ
bracis-28417	228	11	linguistics	linguistic	NOUN
bracis-28417	228	12	,	,	PUNCT
bracis-28417	228	13	pp	pp	ADV
bracis-28417	228	14	.	.	PUNCT
bracis-28417	229	1	6788–6796	6788–6796	NUM
bracis-28417	229	2	.	.	PUNCT
bracis-28417	230	1	international	international	ADJ
bracis-28417	230	2	committee	committee	PROPN
bracis-28417	230	3	on	on	ADP
bracis-28417	230	4	computational	computational	ADJ
bracis-28417	230	5	linguistics	linguistic	NOUN
bracis-28417	230	6	,	,	PUNCT
bracis-28417	230	7	barcelona	barcelona	PROPN
bracis-28417	230	8	,	,	PUNCT
bracis-28417	230	9	spain	spain	PROPN
bracis-28417	230	10	(	(	PUNCT
bracis-28417	230	11	online	online	PROPN
bracis-28417	230	12	)	)	PUNCT
bracis-28417	230	13	(	(	PUNCT
bracis-28417	230	14	2020	2020	NUM
bracis-28417	230	15	)	)	PUNCT
bracis-28417	230	16	.	.	PUNCT
bracis-28417	231	1	https://doi.org/10.18653/v1/2020.coling-main.598	https://doi.org/10.18653/v1/2020.coling-main.598	PROPN
bracis-28417	231	2	chaves	chaves	PROPN
bracis-28417	231	3	rodrigues	rodrigues	PROPN
bracis-28417	231	4	,	,	PUNCT
bracis-28417	231	5	r.	r.	PROPN
bracis-28417	231	6	,	,	PUNCT
bracis-28417	231	7	tanti	tanti	NOUN
bracis-28417	231	8	,	,	PUNCT
bracis-28417	231	9	m.	m.	NOUN
bracis-28417	231	10	,	,	PUNCT
bracis-28417	231	11	agerri	agerri	PROPN
bracis-28417	231	12	,	,	PUNCT
bracis-28417	231	13	r.	r.	PROPN
bracis-28417	231	14	:	:	PUNCT
bracis-28417	231	15	evaluation	evaluation	NOUN
bracis-28417	231	16	of	of	ADP
bracis-28417	231	17	portuguese	portuguese	ADJ
bracis-28417	231	18	language	language	NOUN
bracis-28417	231	19	models	model	NOUN
bracis-28417	231	20	(	(	PUNCT
bracis-28417	231	21	2023	2023	NUM
bracis-28417	231	22	)	)	PUNCT
bracis-28417	231	23	.	.	PUNCT
bracis-28417	232	1	https://doi.org/10.5281/zenodo.7781848	https://doi.org/10.5281/zenodo.7781848	PROPN
bracis-28417	232	2	,	,	PUNCT
bracis-28417	232	3	https://github.com/ruanchaves/eplm	https://github.com/ruanchaves/eplm	NOUN
bracis-28417	232	4	chowdhery	chowdhery	NOUN
bracis-28417	232	5	,	,	PUNCT
bracis-28417	232	6	a.	a.	NOUN
bracis-28417	232	7	,	,	PUNCT
bracis-28417	232	8	et	et	PROPN
bracis-28417	232	9	al	al	PROPN
bracis-28417	232	10	.	.	PUNCT
bracis-28417	232	11	:	:	PUNCT
bracis-28417	233	1	palm	palm	NOUN
bracis-28417	233	2	:	:	PUNCT
bracis-28417	233	3	scaling	scale	VERB
bracis-28417	233	4	language	language	NOUN
bracis-28417	233	5	modeling	modeling	NOUN
bracis-28417	233	6	with	with	ADP
bracis-28417	233	7	pathways	pathway	NOUN
bracis-28417	233	8	.	.	PUNCT
bracis-28417	234	1	arxiv	arxiv	PROPN
bracis-28417	234	2	preprint	preprint	VERB
bracis-28417	234	3	arxiv:2204.02311	arxiv:2204.02311	NOUN
bracis-28417	234	4	(	(	PUNCT
bracis-28417	234	5	2022	2022	NUM
bracis-28417	234	6	)	)	PUNCT
bracis-28417	234	7	clark	clark	PROPN
bracis-28417	234	8	,	,	PUNCT
bracis-28417	234	9	c.	c.	PROPN
bracis-28417	234	10	,	,	PUNCT
bracis-28417	234	11	lee	lee	PROPN
bracis-28417	234	12	,	,	PUNCT
bracis-28417	234	13	k.	k.	PROPN
bracis-28417	234	14	,	,	PUNCT
bracis-28417	234	15	chang	chang	PROPN
bracis-28417	234	16	,	,	PUNCT
bracis-28417	234	17	m.w	m.w	PROPN
bracis-28417	234	18	.	.	PROPN
bracis-28417	234	19	,	,	PUNCT
bracis-28417	234	20	kwiatkowski	kwiatkowski	PROPN
bracis-28417	234	21	,	,	PUNCT
bracis-28417	234	22	t.	t.	PROPN
bracis-28417	234	23	,	,	PUNCT
bracis-28417	234	24	collins	collins	PROPN
bracis-28417	234	25	,	,	PUNCT
bracis-28417	234	26	m.	m.	NOUN
bracis-28417	234	27	,	,	PUNCT
bracis-28417	234	28	toutanova	toutanova	NOUN
bracis-28417	234	29	,	,	PUNCT
bracis-28417	234	30	k.	k.	PROPN
bracis-28417	234	31	:	:	PUNCT
bracis-28417	235	1	boolq	boolq	ADV
bracis-28417	235	2	:	:	PUNCT
bracis-28417	235	3	exploring	explore	VERB
bracis-28417	235	4	the	the	DET
bracis-28417	235	5	surprising	surprising	ADJ
bracis-28417	235	6	difficulty	difficulty	NOUN
bracis-28417	235	7	of	of	ADP
bracis-28417	235	8	natural	natural	ADJ
bracis-28417	235	9	yes	yes	NOUN
bracis-28417	235	10	/	/	SYM
bracis-28417	235	11	no	no	PRON
bracis-28417	235	12	questions	question	NOUN
bracis-28417	235	13	.	.	PUNCT
bracis-28417	236	1	in	in	ADP
bracis-28417	236	2	:	:	PUNCT
bracis-28417	236	3	proceedings	proceeding	NOUN
bracis-28417	236	4	of	of	ADP
bracis-28417	236	5	the	the	DET
bracis-28417	236	6	2019	2019	NUM
bracis-28417	236	7	conference	conference	NOUN
bracis-28417	236	8	of	of	ADP
bracis-28417	236	9	the	the	DET
bracis-28417	236	10	north	north	ADJ
bracis-28417	236	11	american	american	ADJ
bracis-28417	236	12	chapter	chapter	NOUN
bracis-28417	236	13	of	of	ADP
bracis-28417	236	14	the	the	DET
bracis-28417	236	15	association	association	NOUN
bracis-28417	236	16	for	for	ADP
bracis-28417	236	17	computational	computational	ADJ
bracis-28417	236	18	linguistics	linguistic	NOUN
bracis-28417	236	19	:	:	PUNCT
bracis-28417	236	20	human	human	ADJ
bracis-28417	236	21	language	language	NOUN
bracis-28417	236	22	technologies	technology	NOUN
bracis-28417	236	23	,	,	PUNCT
bracis-28417	236	24	volume	volume	NOUN
bracis-28417	236	25	1	1	NUM
bracis-28417	236	26	(	(	PUNCT
bracis-28417	236	27	long	long	ADJ
bracis-28417	236	28	and	and	CCONJ
bracis-28417	236	29	short	short	ADJ
bracis-28417	236	30	papers	paper	NOUN
bracis-28417	236	31	)	)	PUNCT
bracis-28417	236	32	,	,	PUNCT
bracis-28417	236	33	pp	pp	PROPN
bracis-28417	236	34	.	.	PUNCT
bracis-28417	237	1	2924–2936	2924–2936	NUM
bracis-28417	237	2	.	.	PUNCT
bracis-28417	237	3	association	association	NOUN
bracis-28417	237	4	for	for	ADP
bracis-28417	237	5	computational	computational	ADJ
bracis-28417	237	6	linguistics	linguistic	NOUN
bracis-28417	237	7	,	,	PUNCT
bracis-28417	237	8	minneapolis	minneapolis	PROPN
bracis-28417	237	9	,	,	PUNCT
bracis-28417	237	10	minnesota	minnesota	PROPN
bracis-28417	237	11	(	(	PUNCT
bracis-28417	237	12	2019	2019	NUM
bracis-28417	237	13	)	)	PUNCT
bracis-28417	237	14	.	.	PUNCT
bracis-28417	238	1	https://doi.org/10.18653/v1/n19-1300	https://doi.org/10.18653/v1/n19-1300	NOUN
bracis-28417	238	2	conneau	conneau	PROPN
bracis-28417	238	3	,	,	PUNCT
bracis-28417	238	4	a.	a.	PROPN
bracis-28417	238	5	,	,	PUNCT
bracis-28417	238	6	et	et	PROPN
bracis-28417	238	7	al	al	PROPN
bracis-28417	238	8	.	.	PROPN
bracis-28417	238	9	:	:	PUNCT
bracis-28417	239	1	unsupervised	unsupervised	ADJ
bracis-28417	239	2	cross	cross	ADJ
bracis-28417	239	3	-	-	ADJ
bracis-28417	239	4	lingual	lingual	ADJ
bracis-28417	239	5	representation	representation	NOUN
bracis-28417	239	6	learning	learn	VERB
bracis-28417	239	7	at	at	ADP
bracis-28417	239	8	scale	scale	NOUN
bracis-28417	239	9	.	.	PUNCT
bracis-28417	240	1	in	in	ADP
bracis-28417	240	2	:	:	PUNCT
bracis-28417	240	3	proceedings	proceeding	NOUN
bracis-28417	240	4	of	of	ADP
bracis-28417	240	5	the	the	DET
bracis-28417	240	6	58th	58th	ADJ
bracis-28417	240	7	annual	annual	ADJ
bracis-28417	240	8	meeting	meeting	NOUN
bracis-28417	240	9	of	of	ADP
bracis-28417	240	10	the	the	DET
bracis-28417	240	11	association	association	NOUN
bracis-28417	240	12	for	for	ADP
bracis-28417	240	13	computational	computational	ADJ
bracis-28417	240	14	linguistics	linguistic	NOUN
bracis-28417	240	15	,	,	PUNCT
bracis-28417	240	16	pp	pp	X
bracis-28417	240	17	.	.	PUNCT
bracis-28417	240	18	8440–8451	8440–8451	NUM
bracis-28417	240	19	(	(	PUNCT
bracis-28417	240	20	2020	2020	NUM
bracis-28417	240	21	)	)	PUNCT
bracis-28417	240	22	google	google	PROPN
bracis-28417	240	23	scholar	scholar	NOUN
bracis-28417	240	24	  	  	SPACE
bracis-28417	240	25	ebrahimi	ebrahimi	PROPN
bracis-28417	240	26	,	,	PUNCT
bracis-28417	240	27	a.	a.	PROPN
bracis-28417	240	28	,	,	PUNCT
bracis-28417	240	29	kann	kann	PROPN
bracis-28417	240	30	,	,	PUNCT
bracis-28417	240	31	k.	k.	PROPN
bracis-28417	240	32	:	:	PUNCT
bracis-28417	240	33	how	how	SCONJ
bracis-28417	240	34	to	to	PART
bracis-28417	240	35	adapt	adapt	VERB
bracis-28417	240	36	your	your	PRON
bracis-28417	240	37	pretrained	pretraine	VERB
bracis-28417	240	38	multilingual	multilingual	ADJ
bracis-28417	240	39	model	model	NOUN
bracis-28417	240	40	to	to	ADP
bracis-28417	240	41	1600	1600	NUM
bracis-28417	240	42	languages	language	NOUN
bracis-28417	240	43	.	.	PUNCT
bracis-28417	241	1	in	in	ADP
bracis-28417	241	2	:	:	PUNCT
bracis-28417	241	3	proceedings	proceeding	NOUN
bracis-28417	241	4	of	of	ADP
bracis-28417	241	5	the	the	DET
bracis-28417	241	6	59th	59th	ADJ
bracis-28417	241	7	annual	annual	ADJ
bracis-28417	241	8	meeting	meeting	NOUN
bracis-28417	241	9	of	of	ADP
bracis-28417	241	10	the	the	DET
bracis-28417	241	11	association	association	NOUN
bracis-28417	241	12	for	for	ADP
bracis-28417	241	13	computational	computational	ADJ
bracis-28417	241	14	linguistics	linguistic	NOUN
bracis-28417	241	15	and	and	CCONJ
bracis-28417	241	16	the	the	DET
bracis-28417	241	17	11th	11th	ADJ
bracis-28417	241	18	international	international	ADJ
bracis-28417	241	19	joint	joint	ADJ
bracis-28417	241	20	conference	conference	NOUN
bracis-28417	241	21	on	on	ADP
bracis-28417	241	22	natural	natural	ADJ
bracis-28417	241	23	language	language	NOUN
bracis-28417	241	24	processing	processing	NOUN
bracis-28417	241	25	(	(	PUNCT
bracis-28417	241	26	volume	volume	NOUN
bracis-28417	241	27	1	1	NUM
bracis-28417	241	28	:	:	PUNCT
bracis-28417	241	29	long	long	ADJ
bracis-28417	241	30	papers	paper	NOUN
bracis-28417	241	31	)	)	PUNCT
bracis-28417	241	32	,	,	PUNCT
bracis-28417	241	33	pp	pp	ADP
bracis-28417	241	34	.	.	PUNCT
bracis-28417	242	1	4555–4567	4555–4567	NUM
bracis-28417	242	2	.	.	PUNCT
bracis-28417	243	1	association	association	NOUN
bracis-28417	243	2	for	for	ADP
bracis-28417	243	3	computational	computational	ADJ
bracis-28417	243	4	linguistics	linguistic	NOUN
bracis-28417	243	5	,	,	PUNCT
bracis-28417	243	6	online	online	ADJ
bracis-28417	243	7	(	(	PUNCT
bracis-28417	243	8	aug	aug	PROPN
bracis-28417	243	9	2021	2021	NUM
bracis-28417	243	10	)	)	PUNCT
bracis-28417	243	11	.	.	PUNCT
bracis-28417	244	1	10.18653	10.18653	NUM
bracis-28417	244	2	/	/	SYM
bracis-28417	244	3	v1/2021.acl	v1/2021.acl	VERB
bracis-28417	244	4	-	-	PUNCT
bracis-28417	244	5	long.351	long.351	NOUN
bracis-28417	244	6	google	google	PROPN
bracis-28417	244	7	scholar	scholar	NOUN
bracis-28417	244	8	  	  	SPACE
bracis-28417	244	9	fitzgerald	fitzgerald	NOUN
bracis-28417	244	10	,	,	PUNCT
bracis-28417	244	11	j.	j.	PROPN
bracis-28417	244	12	,	,	PUNCT
bracis-28417	244	13	et	et	PROPN
bracis-28417	244	14	al	al	PROPN
bracis-28417	244	15	.	.	PUNCT
bracis-28417	244	16	:	:	PUNCT
bracis-28417	245	1	massive	massive	ADJ
bracis-28417	245	2	:	:	PUNCT
bracis-28417	245	3	a	a	DET
bracis-28417	245	4	1m	1m	NUM
bracis-28417	245	5	-	-	PUNCT
bracis-28417	245	6	example	example	NOUN
bracis-28417	245	7	multilingual	multilingual	ADJ
bracis-28417	245	8	natural	natural	ADJ
bracis-28417	245	9	language	language	NOUN
bracis-28417	245	10	understanding	understanding	NOUN
bracis-28417	245	11	dataset	dataset	VERB
bracis-28417	245	12	with	with	ADP
bracis-28417	245	13	51	51	NUM
bracis-28417	245	14	typologically	typologically	ADV
bracis-28417	245	15	-	-	PUNCT
bracis-28417	245	16	diverse	diverse	ADJ
bracis-28417	245	17	languages	language	NOUN
bracis-28417	245	18	(	(	PUNCT
bracis-28417	245	19	2022	2022	NUM
bracis-28417	245	20	)	)	PUNCT
bracis-28417	245	21	google	google	PROPN
bracis-28417	245	22	scholar	scholar	NOUN
bracis-28417	245	23	  	  	SPACE
bracis-28417	245	24	fu	fu	NOUN
bracis-28417	245	25	,	,	PUNCT
bracis-28417	245	26	y.	y.	PROPN
bracis-28417	245	27	,	,	PUNCT
bracis-28417	245	28	peng	peng	PROPN
bracis-28417	245	29	,	,	PUNCT
bracis-28417	245	30	h.	h.	PROPN
bracis-28417	245	31	,	,	PUNCT
bracis-28417	245	32	ou	ou	PROPN
bracis-28417	245	33	,	,	PUNCT
bracis-28417	245	34	l.	l.	PROPN
bracis-28417	245	35	,	,	PUNCT
bracis-28417	245	36	sabharwal	sabharwal	PROPN
bracis-28417	245	37	,	,	PUNCT
bracis-28417	245	38	a.	a.	NOUN
bracis-28417	245	39	,	,	PUNCT
bracis-28417	245	40	khot	khot	ADJ
bracis-28417	245	41	,	,	PUNCT
bracis-28417	245	42	t.	t.	PROPN
bracis-28417	245	43	:	:	PUNCT
bracis-28417	245	44	specializing	specialize	VERB
bracis-28417	245	45	smaller	small	ADJ
bracis-28417	245	46	language	language	NOUN
bracis-28417	245	47	models	model	NOUN
bracis-28417	245	48	towards	towards	ADP
bracis-28417	245	49	multi	multi	ADJ
bracis-28417	245	50	-	-	ADJ
bracis-28417	245	51	step	step	ADJ
bracis-28417	245	52	reasoning	reasoning	NOUN
bracis-28417	245	53	.	.	PUNCT
bracis-28417	246	1	arxiv	arxiv	PROPN
bracis-28417	246	2	preprint	preprint	NOUN
bracis-28417	246	3	arxiv:2301.12726	arxiv:2301.12726	NOUN
bracis-28417	246	4	(	(	PUNCT
bracis-28417	246	5	2023	2023	NUM
bracis-28417	246	6	)	)	PUNCT
bracis-28417	246	7	gao	gao	PROPN
bracis-28417	246	8	,	,	PUNCT
bracis-28417	246	9	l.	l.	PROPN
bracis-28417	246	10	,	,	PUNCT
bracis-28417	246	11	et	et	PROPN
bracis-28417	246	12	al	al	PROPN
bracis-28417	246	13	.	.	PUNCT
bracis-28417	246	14	:	:	PUNCT
bracis-28417	247	1	the	the	DET
bracis-28417	247	2	pile	pile	NOUN
bracis-28417	247	3	:	:	PUNCT
bracis-28417	247	4	an	an	DET
bracis-28417	247	5	800	800	NUM
bracis-28417	247	6	gb	gb	NOUN
bracis-28417	247	7	dataset	dataset	NOUN
bracis-28417	247	8	of	of	ADP
bracis-28417	247	9	diverse	diverse	ADJ
bracis-28417	247	10	text	text	NOUN
bracis-28417	247	11	for	for	ADP
bracis-28417	247	12	language	language	NOUN
bracis-28417	247	13	modeling	modeling	NOUN
bracis-28417	247	14	.	.	PUNCT
bracis-28417	248	1	arxiv	arxiv	PROPN
bracis-28417	248	2	preprint	preprint	VERB
bracis-28417	248	3	arxiv:2101.00027	arxiv:2101.00027	PROPN
bracis-28417	248	4	(	(	PUNCT
bracis-28417	248	5	2020	2020	NUM
bracis-28417	248	6	)	)	PUNCT
bracis-28417	248	7	gururangan	gururangan	NOUN
bracis-28417	248	8	,	,	PUNCT
bracis-28417	248	9	s.	s.	PROPN
bracis-28417	248	10	,	,	PUNCT
bracis-28417	248	11	et	et	PROPN
bracis-28417	248	12	al	al	PROPN
bracis-28417	248	13	.	.	PROPN
bracis-28417	248	14	:	:	PUNCT
bracis-28417	249	1	scaling	scale	VERB
bracis-28417	249	2	expert	expert	ADJ
bracis-28417	249	3	language	language	NOUN
bracis-28417	249	4	models	model	NOUN
bracis-28417	249	5	with	with	ADP
bracis-28417	249	6	unsupervised	unsupervised	ADJ
bracis-28417	249	7	domain	domain	NOUN
bracis-28417	249	8	discovery	discovery	NOUN
bracis-28417	249	9	.	.	PUNCT
bracis-28417	250	1	arxiv	arxiv	PROPN
bracis-28417	250	2	preprint	preprint	VERB
bracis-28417	250	3	arxiv:2303.14177	arxiv:2303.14177	PRON
bracis-28417	250	4	(	(	PUNCT
bracis-28417	250	5	2023	2023	NUM
bracis-28417	250	6	)	)	PUNCT
bracis-28417	250	7	hoffmann	hoffmann	PROPN
bracis-28417	250	8	,	,	PUNCT
bracis-28417	250	9	j.	j.	PROPN
bracis-28417	250	10	,	,	PUNCT
bracis-28417	250	11	et	et	PROPN
bracis-28417	250	12	al	al	PROPN
bracis-28417	250	13	.	.	PROPN
bracis-28417	250	14	:	:	PUNCT
bracis-28417	251	1	training	train	VERB
bracis-28417	251	2	compute	compute	NOUN
bracis-28417	251	3	-	-	PUNCT
bracis-28417	251	4	optimal	optimal	ADJ
bracis-28417	251	5	large	large	ADJ
bracis-28417	251	6	language	language	NOUN
bracis-28417	251	7	models	model	NOUN
bracis-28417	251	8	.	.	PUNCT
bracis-28417	252	1	arxiv	arxiv	PROPN
bracis-28417	252	2	preprint	preprint	NOUN
bracis-28417	252	3	arxiv:2203.15556	arxiv:2203.15556	NOUN
bracis-28417	252	4	(	(	PUNCT
bracis-28417	252	5	2022	2022	NUM
bracis-28417	252	6	)	)	PUNCT
bracis-28417	252	7	houlsby	houlsby	ADJ
bracis-28417	252	8	,	,	PUNCT
bracis-28417	252	9	n.	n.	NOUN
bracis-28417	252	10	,	,	PUNCT
bracis-28417	252	11	et	et	PROPN
bracis-28417	252	12	al	al	PROPN
bracis-28417	252	13	.	.	PUNCT
bracis-28417	252	14	:	:	PUNCT
bracis-28417	253	1	parameter	parameter	NOUN
bracis-28417	253	2	-	-	PUNCT
bracis-28417	253	3	efficient	efficient	ADJ
bracis-28417	253	4	transfer	transfer	NOUN
bracis-28417	253	5	learning	learning	NOUN
bracis-28417	253	6	for	for	ADP
bracis-28417	253	7	nlp	nlp	NOUN
bracis-28417	253	8	.	.	PUNCT
bracis-28417	254	1	in	in	ADP
bracis-28417	254	2	:	:	PUNCT
bracis-28417	254	3	international	international	ADJ
bracis-28417	254	4	conference	conference	NOUN
bracis-28417	254	5	on	on	ADP
bracis-28417	254	6	machine	machine	NOUN
bracis-28417	254	7	learning	learning	NOUN
bracis-28417	254	8	,	,	PUNCT
bracis-28417	254	9	pp	pp	ADJ
bracis-28417	254	10	.	.	PUNCT
bracis-28417	255	1	2790–2799	2790–2799	NUM
bracis-28417	255	2	.	.	PUNCT
bracis-28417	256	1	pmlr	pmlr	NOUN
bracis-28417	256	2	(	(	PUNCT
bracis-28417	256	3	2019	2019	NUM
bracis-28417	256	4	)	)	PUNCT
bracis-28417	256	5	google	google	PROPN
bracis-28417	256	6	scholar	scholar	NOUN
bracis-28417	256	7	  	  	SPACE
bracis-28417	256	8	hu	hu	PROPN
bracis-28417	256	9	,	,	PUNCT
bracis-28417	256	10	e.j	e.j	PROPN
bracis-28417	256	11	.	.	PROPN
bracis-28417	256	12	,	,	PUNCT
bracis-28417	256	13	et	et	PROPN
bracis-28417	256	14	al	al	PROPN
bracis-28417	256	15	.	.	PUNCT
bracis-28417	256	16	:	:	PUNCT
bracis-28417	257	1	lora	lora	NOUN
bracis-28417	257	2	:	:	PUNCT
bracis-28417	257	3	low	low	ADJ
bracis-28417	257	4	-	-	PUNCT
bracis-28417	257	5	rank	rank	NOUN
bracis-28417	257	6	adaptation	adaptation	NOUN
bracis-28417	257	7	of	of	ADP
bracis-28417	257	8	large	large	ADJ
bracis-28417	257	9	language	language	NOUN
bracis-28417	257	10	models	model	NOUN
bracis-28417	257	11	.	.	PUNCT
bracis-28417	258	1	in	in	ADP
bracis-28417	258	2	:	:	PUNCT
bracis-28417	258	3	international	international	ADJ
bracis-28417	258	4	conference	conference	NOUN
bracis-28417	258	5	on	on	ADP
bracis-28417	258	6	learning	learn	VERB
bracis-28417	258	7	representations	representation	NOUN
bracis-28417	258	8	(	(	PUNCT
bracis-28417	258	9	2022	2022	NUM
bracis-28417	258	10	)	)	PUNCT
bracis-28417	258	11	.	.	PUNCT
bracis-28417	259	1	https://openreview.net/forum?id=nzevkeefyf9	https://openreview.net/forum?id=nzevkeefyf9	PROPN
bracis-28417	259	2	kalyan	kalyan	PROPN
bracis-28417	259	3	,	,	PUNCT
bracis-28417	259	4	k.s	k.s	PROPN
bracis-28417	259	5	.	.	PROPN
bracis-28417	259	6	,	,	PUNCT
bracis-28417	259	7	rajasekharan	rajasekharan	NOUN
bracis-28417	259	8	,	,	PUNCT
bracis-28417	259	9	a.	a.	NOUN
bracis-28417	259	10	,	,	PUNCT
bracis-28417	259	11	sangeetha	sangeetha	PROPN
bracis-28417	259	12	,	,	PUNCT
bracis-28417	259	13	s.	s.	PROPN
bracis-28417	259	14	:	:	PUNCT
bracis-28417	259	15	ammus	ammus	VERB
bracis-28417	259	16	:	:	PUNCT
bracis-28417	259	17	a	a	DET
bracis-28417	259	18	survey	survey	NOUN
bracis-28417	259	19	of	of	ADP
bracis-28417	259	20	transformer	transformer	NOUN
bracis-28417	259	21	-	-	PUNCT
bracis-28417	259	22	based	base	VERB
bracis-28417	259	23	pretrained	pretraine	VERB
bracis-28417	259	24	models	model	NOUN
bracis-28417	259	25	in	in	ADP
bracis-28417	259	26	natural	natural	ADJ
bracis-28417	259	27	language	language	NOUN
bracis-28417	259	28	processing	processing	NOUN
bracis-28417	259	29	.	.	PUNCT
bracis-28417	260	1	arxiv	arxiv	PROPN
bracis-28417	260	2	preprint	preprint	VERB
bracis-28417	260	3	arxiv:2108.05542	arxiv:2108.05542	PROPN
bracis-28417	260	4	(	(	PUNCT
bracis-28417	260	5	2021	2021	NUM
bracis-28417	260	6	)	)	PUNCT
bracis-28417	260	7	kim	kim	PROPN
bracis-28417	260	8	,	,	PUNCT
bracis-28417	260	9	b.	b.	PROPN
bracis-28417	260	10	,	,	PUNCT
bracis-28417	260	11	et	et	PROPN
bracis-28417	260	12	al	al	PROPN
bracis-28417	260	13	.	.	PROPN
bracis-28417	260	14	:	:	PUNCT
bracis-28417	261	1	what	what	DET
bracis-28417	261	2	changes	change	NOUN
bracis-28417	261	3	can	can	AUX
bracis-28417	261	4	large	large	ADJ
bracis-28417	261	5	-	-	PUNCT
bracis-28417	261	6	scale	scale	NOUN
bracis-28417	261	7	language	language	NOUN
bracis-28417	261	8	models	model	NOUN
bracis-28417	261	9	bring	bring	VERB
bracis-28417	261	10	?	?	PUNCT
bracis-28417	262	1	intensive	intensive	ADJ
bracis-28417	262	2	study	study	NOUN
bracis-28417	262	3	on	on	ADP
bracis-28417	262	4	hyperclova	hyperclova	NOUN
bracis-28417	262	5	:	:	PUNCT
bracis-28417	262	6	billions	billion	NOUN
bracis-28417	262	7	-	-	PUNCT
bracis-28417	262	8	scale	scale	NOUN
bracis-28417	262	9	korean	korean	ADJ
bracis-28417	262	10	generative	generative	NOUN
bracis-28417	262	11	pretrained	pretraine	VERB
bracis-28417	262	12	transformers	transformer	NOUN
bracis-28417	262	13	.	.	PUNCT
bracis-28417	263	1	in	in	ADP
bracis-28417	263	2	:	:	PUNCT
bracis-28417	263	3	proceedings	proceeding	NOUN
bracis-28417	263	4	of	of	ADP
bracis-28417	263	5	the	the	DET
bracis-28417	263	6	2021	2021	NUM
bracis-28417	263	7	conference	conference	NOUN
bracis-28417	263	8	on	on	ADP
bracis-28417	263	9	empirical	empirical	ADJ
bracis-28417	263	10	methods	method	NOUN
bracis-28417	263	11	in	in	ADP
bracis-28417	263	12	natural	natural	ADJ
bracis-28417	263	13	language	language	NOUN
bracis-28417	263	14	processing	processing	NOUN
bracis-28417	263	15	,	,	PUNCT
bracis-28417	263	16	pp	pp	ADJ
bracis-28417	263	17	.	.	PUNCT
bracis-28417	264	1	3405–3424	3405–3424	NUM
bracis-28417	264	2	(	(	PUNCT
bracis-28417	264	3	2021k	2021k	NUM
bracis-28417	264	4	google	google	PROPN
bracis-28417	264	5	scholar	scholar	NOUN
bracis-28417	264	6	  	  	SPACE
bracis-28417	264	7	kudo	kudo	PROPN
bracis-28417	264	8	,	,	PUNCT
bracis-28417	264	9	t.	t.	PROPN
bracis-28417	264	10	,	,	PUNCT
bracis-28417	264	11	richardson	richardson	PROPN
bracis-28417	264	12	,	,	PUNCT
bracis-28417	264	13	j.	j.	PROPN
bracis-28417	264	14	:	:	PUNCT
bracis-28417	264	15	sentencepiece	sentencepiece	PROPN
bracis-28417	264	16	:	:	PUNCT
bracis-28417	264	17	a	a	DET
bracis-28417	264	18	simple	simple	ADJ
bracis-28417	264	19	and	and	CCONJ
bracis-28417	264	20	language	language	NOUN
bracis-28417	264	21	independent	independent	PROPN
bracis-28417	264	22	subword	subword	PROPN
bracis-28417	264	23	tokenizer	tokenizer	NOUN
bracis-28417	264	24	and	and	CCONJ
bracis-28417	264	25	detokenizer	detokenizer	VERB
bracis-28417	264	26	for	for	ADP
bracis-28417	264	27	neural	neural	ADJ
bracis-28417	264	28	text	text	NOUN
bracis-28417	264	29	processing	processing	NOUN
bracis-28417	264	30	.	.	PUNCT
bracis-28417	265	1	in	in	ADP
bracis-28417	265	2	:	:	PUNCT
bracis-28417	265	3	proceedings	proceeding	NOUN
bracis-28417	265	4	of	of	ADP
bracis-28417	265	5	the	the	DET
bracis-28417	265	6	2018	2018	NUM
bracis-28417	265	7	conference	conference	NOUN
bracis-28417	265	8	on	on	ADP
bracis-28417	265	9	empirical	empirical	ADJ
bracis-28417	265	10	methods	method	NOUN
bracis-28417	265	11	in	in	ADP
bracis-28417	265	12	natural	natural	ADJ
bracis-28417	265	13	language	language	NOUN
bracis-28417	265	14	processing	processing	NOUN
bracis-28417	265	15	:	:	PUNCT
bracis-28417	265	16	system	system	NOUN
bracis-28417	265	17	demonstrations	demonstration	NOUN
bracis-28417	265	18	,	,	PUNCT
bracis-28417	265	19	pp	pp	ADV
bracis-28417	265	20	.	.	PUNCT
bracis-28417	265	21	66–71	66–71	NOUN
bracis-28417	265	22	.	.	PUNCT
bracis-28417	266	1	association	association	NOUN
bracis-28417	266	2	for	for	ADP
bracis-28417	266	3	computational	computational	ADJ
bracis-28417	266	4	linguistics	linguistic	NOUN
bracis-28417	266	5	,	,	PUNCT
bracis-28417	266	6	brussels	brussels	PROPN
bracis-28417	266	7	,	,	PUNCT
bracis-28417	266	8	belgium	belgium	NOUN
bracis-28417	266	9	(	(	PUNCT
bracis-28417	266	10	2018	2018	NUM
bracis-28417	266	11	)	)	PUNCT
bracis-28417	266	12	.	.	PUNCT
bracis-28417	267	1	https://doi.org/10.18653/v1/d18-2012	https://doi.org/10.18653/v1/d18-2012	PROPN
bracis-28417	267	2	le	le	PROPN
bracis-28417	267	3	,	,	PUNCT
bracis-28417	267	4	h.	h.	PROPN
bracis-28417	267	5	,	,	PUNCT
bracis-28417	267	6	et	et	PROPN
bracis-28417	267	7	al	al	PROPN
bracis-28417	267	8	.	.	PROPN
bracis-28417	267	9	:	:	PUNCT
bracis-28417	268	1	flaubert	flaubert	PROPN
bracis-28417	268	2	:	:	PUNCT
bracis-28417	268	3	unsupervised	unsupervised	ADJ
bracis-28417	268	4	language	language	NOUN
bracis-28417	268	5	model	model	NOUN
bracis-28417	268	6	pre	pre	NOUN
bracis-28417	268	7	-	-	NOUN
bracis-28417	268	8	training	training	NOUN
bracis-28417	268	9	for	for	ADP
bracis-28417	268	10	french	french	NOUN
bracis-28417	268	11	.	.	PUNCT
bracis-28417	269	1	in	in	ADP
bracis-28417	269	2	:	:	PUNCT
bracis-28417	269	3	proceedings	proceeding	NOUN
bracis-28417	269	4	of	of	ADP
bracis-28417	269	5	the	the	DET
bracis-28417	269	6	twelfth	twelfth	ADJ
bracis-28417	269	7	language	language	NOUN
bracis-28417	269	8	resources	resource	NOUN
bracis-28417	269	9	and	and	CCONJ
bracis-28417	269	10	evaluation	evaluation	NOUN
bracis-28417	269	11	conference	conference	NOUN
bracis-28417	269	12	,	,	PUNCT
bracis-28417	269	13	pp	pp	X
bracis-28417	269	14	.	.	PUNCT
bracis-28417	270	1	2479–2490	2479–2490	NUM
bracis-28417	270	2	.	.	PUNCT
bracis-28417	271	1	european	european	ADJ
bracis-28417	271	2	language	language	PROPN
bracis-28417	271	3	resources	resource	NOUN
bracis-28417	271	4	association	association	PROPN
bracis-28417	271	5	,	,	PUNCT
bracis-28417	271	6	marseille	marseille	PROPN
bracis-28417	271	7	,	,	PUNCT
bracis-28417	271	8	france	france	PROPN
bracis-28417	271	9	(	(	PUNCT
bracis-28417	271	10	2020	2020	NUM
bracis-28417	271	11	)	)	PUNCT
bracis-28417	271	12	google	google	PROPN
bracis-28417	271	13	scholar	scholar	NOUN
bracis-28417	271	14	  	  	SPACE
bracis-28417	271	15	lee	lee	PROPN
bracis-28417	271	16	,	,	PUNCT
bracis-28417	271	17	h.	h.	PROPN
bracis-28417	271	18	,	,	PUNCT
bracis-28417	271	19	yoon	yoon	PROPN
bracis-28417	271	20	,	,	PUNCT
bracis-28417	271	21	j.	j.	PROPN
bracis-28417	271	22	,	,	PUNCT
bracis-28417	271	23	hwang	hwang	PROPN
bracis-28417	271	24	,	,	PUNCT
bracis-28417	271	25	b.	b.	PROPN
bracis-28417	271	26	,	,	PUNCT
bracis-28417	271	27	joe	joe	PROPN
bracis-28417	271	28	,	,	PUNCT
bracis-28417	271	29	s.	s.	PROPN
bracis-28417	271	30	,	,	PUNCT
bracis-28417	271	31	min	min	PROPN
bracis-28417	271	32	,	,	PUNCT
bracis-28417	271	33	s.	s.	PROPN
bracis-28417	271	34	,	,	PUNCT
bracis-28417	271	35	gwon	gwon	VERB
bracis-28417	271	36	,	,	PUNCT
bracis-28417	271	37	y.	y.	NOUN
bracis-28417	271	38	:	:	PUNCT
bracis-28417	271	39	korealbert	korealbert	ADV
bracis-28417	271	40	:	:	PUNCT
bracis-28417	271	41	pretraining	pretraine	VERB
bracis-28417	271	42	a	a	DET
bracis-28417	271	43	lite	lite	ADJ
bracis-28417	271	44	bert	bert	PROPN
bracis-28417	271	45	model	model	NOUN
bracis-28417	271	46	for	for	ADP
bracis-28417	271	47	korean	korean	ADJ
bracis-28417	271	48	language	language	NOUN
bracis-28417	271	49	understanding	understanding	NOUN
bracis-28417	271	50	.	.	PUNCT
bracis-28417	272	1	in	in	ADP
bracis-28417	272	2	:	:	PUNCT
bracis-28417	272	3	2020	2020	NUM
bracis-28417	272	4	25th	25th	NOUN
bracis-28417	272	5	international	international	ADJ
bracis-28417	272	6	conference	conference	NOUN
bracis-28417	272	7	on	on	ADP
bracis-28417	272	8	pattern	pattern	NOUN
bracis-28417	272	9	recognition	recognition	NOUN
bracis-28417	272	10	(	(	PUNCT
bracis-28417	272	11	icpr	icpr	PROPN
bracis-28417	272	12	)	)	PUNCT
bracis-28417	272	13	,	,	PUNCT
bracis-28417	272	14	pp	pp	PROPN
bracis-28417	272	15	.	.	PUNCT
bracis-28417	272	16	5551–5557	5551–5557	X
bracis-28417	272	17	.	.	PUNCT
bracis-28417	273	1	ieee	ieee	NOUN
bracis-28417	273	2	(	(	PUNCT
bracis-28417	273	3	2021	2021	NUM
bracis-28417	273	4	)	)	PUNCT
bracis-28417	273	5	google	google	NOUN
bracis-28417	273	6	scholar	scholar	NOUN
bracis-28417	273	7	  	  	SPACE
bracis-28417	273	8	lewkowycz	lewkowycz	NOUN
bracis-28417	273	9	,	,	PUNCT
bracis-28417	273	10	a.	a.	NOUN
bracis-28417	273	11	,	,	PUNCT
bracis-28417	273	12	et	et	PROPN
bracis-28417	273	13	al	al	PROPN
bracis-28417	273	14	.	.	PROPN
bracis-28417	273	15	:	:	PUNCT
bracis-28417	273	16	solving	solve	VERB
bracis-28417	273	17	quantitative	quantitative	ADJ
bracis-28417	273	18	reasoning	reasoning	NOUN
bracis-28417	273	19	problems	problem	NOUN
bracis-28417	273	20	with	with	ADP
bracis-28417	273	21	language	language	NOUN
bracis-28417	273	22	models	model	NOUN
bracis-28417	273	23	.	.	PUNCT
bracis-28417	274	1	arxiv	arxiv	PROPN
bracis-28417	274	2	preprint	preprint	NOUN
bracis-28417	274	3	arxiv:2206.14858	arxiv:2206.14858	NOUN
bracis-28417	274	4	(	(	PUNCT
bracis-28417	274	5	2022	2022	NUM
bracis-28417	274	6	)	)	PUNCT
bracis-28417	274	7	lin	lin	PROPN
bracis-28417	274	8	,	,	PUNCT
bracis-28417	274	9	x.v	x.v	PROPN
bracis-28417	274	10	.	.	PROPN
bracis-28417	274	11	,	,	PUNCT
bracis-28417	274	12	et	et	PROPN
bracis-28417	274	13	al	al	PROPN
bracis-28417	274	14	.	.	PUNCT
bracis-28417	274	15	:	:	PUNCT
bracis-28417	275	1	few	few	ADJ
bracis-28417	275	2	-	-	PUNCT
bracis-28417	275	3	shot	shot	NOUN
bracis-28417	275	4	learning	learning	NOUN
bracis-28417	275	5	with	with	ADP
bracis-28417	275	6	multilingual	multilingual	ADJ
bracis-28417	275	7	generative	generative	ADJ
bracis-28417	275	8	language	language	NOUN
bracis-28417	275	9	models	model	NOUN
bracis-28417	275	10	.	.	PUNCT
bracis-28417	276	1	in	in	ADP
bracis-28417	276	2	:	:	PUNCT
bracis-28417	276	3	proceedings	proceeding	NOUN
bracis-28417	276	4	of	of	ADP
bracis-28417	276	5	the	the	DET
bracis-28417	276	6	2022	2022	NUM
bracis-28417	276	7	conference	conference	NOUN
bracis-28417	276	8	on	on	ADP
bracis-28417	276	9	empirical	empirical	ADJ
bracis-28417	276	10	methods	method	NOUN
bracis-28417	276	11	in	in	ADP
bracis-28417	276	12	natural	natural	ADJ
bracis-28417	276	13	language	language	NOUN
bracis-28417	276	14	processing	processing	NOUN
bracis-28417	276	15	,	,	PUNCT
bracis-28417	276	16	pp	pp	ADP
bracis-28417	276	17	.	.	PUNCT
bracis-28417	277	1	9019–9052	9019–9052	NUM
bracis-28417	277	2	(	(	PUNCT
bracis-28417	277	3	2022	2022	NUM
bracis-28417	277	4	)	)	PUNCT
bracis-28417	277	5	google	google	PROPN
bracis-28417	277	6	scholar	scholar	NOUN
bracis-28417	277	7	  	  	SPACE
bracis-28417	277	8	liu	liu	PROPN
bracis-28417	277	9	,	,	PUNCT
bracis-28417	277	10	y.	y.	PROPN
bracis-28417	277	11	,	,	PUNCT
bracis-28417	277	12	et	et	PROPN
bracis-28417	277	13	al	al	PROPN
bracis-28417	277	14	.	.	PUNCT
bracis-28417	277	15	:	:	PUNCT
bracis-28417	278	1	multilingual	multilingual	ADJ
bracis-28417	278	2	denoising	denoise	VERB
bracis-28417	278	3	pre	pre	NOUN
bracis-28417	278	4	-	-	NOUN
bracis-28417	278	5	training	training	NOUN
bracis-28417	278	6	for	for	ADP
bracis-28417	278	7	neural	neural	ADJ
bracis-28417	278	8	machine	machine	NOUN
bracis-28417	278	9	translation	translation	NOUN
bracis-28417	278	10	.	.	PUNCT
bracis-28417	279	1	trans	trans	PROPN
bracis-28417	279	2	.	.	PUNCT
bracis-28417	280	1	assoc	assoc	PROPN
bracis-28417	280	2	.	.	PUNCT
bracis-28417	281	1	comput	comput	PROPN
bracis-28417	281	2	.	.	PUNCT
bracis-28417	282	1	linguist	linguist	NOUN
bracis-28417	282	2	.	.	PUNCT
bracis-28417	283	1	8	8	NUM
bracis-28417	283	2	,	,	PUNCT
bracis-28417	283	3	726–742	726–742	NUM
bracis-28417	283	4	(	(	PUNCT
bracis-28417	283	5	2020	2020	NUM
bracis-28417	283	6	)	)	PUNCT
bracis-28417	283	7	article	article	NOUN
bracis-28417	283	8	  	  	SPACE
bracis-28417	283	9	google	google	PROPN
bracis-28417	283	10	scholar	scholar	NOUN
bracis-28417	283	11	  	  	SPACE
bracis-28417	283	12	longpre	longpre	PROPN
bracis-28417	283	13	,	,	PUNCT
bracis-28417	283	14	s.	s.	PROPN
bracis-28417	283	15	,	,	PUNCT
bracis-28417	283	16	lu	lu	PROPN
bracis-28417	283	17	,	,	PUNCT
bracis-28417	283	18	y.	y.	PROPN
bracis-28417	283	19	,	,	PUNCT
bracis-28417	283	20	daiber	daiber	PROPN
bracis-28417	283	21	,	,	PUNCT
bracis-28417	283	22	j.	j.	PROPN
bracis-28417	283	23	:	:	PUNCT
bracis-28417	283	24	mkqa	mkqa	PROPN
bracis-28417	283	25	:	:	PUNCT
bracis-28417	283	26	a	a	DET
bracis-28417	283	27	linguistically	linguistically	ADV
bracis-28417	283	28	diverse	diverse	ADJ
bracis-28417	283	29	benchmark	benchmark	NOUN
bracis-28417	283	30	for	for	ADP
bracis-28417	283	31	multilingual	multilingual	ADJ
bracis-28417	283	32	open	open	ADJ
bracis-28417	283	33	domain	domain	NOUN
bracis-28417	283	34	question	question	NOUN
bracis-28417	283	35	answering	answering	NOUN
bracis-28417	283	36	.	.	PUNCT
bracis-28417	284	1	trans	trans	PROPN
bracis-28417	284	2	.	.	PUNCT
bracis-28417	285	1	assoc	assoc	PROPN
bracis-28417	285	2	.	.	PUNCT
bracis-28417	286	1	comput	comput	PROPN
bracis-28417	286	2	.	.	PUNCT
bracis-28417	287	1	linguist	linguist	NOUN
bracis-28417	287	2	.	.	PUNCT
bracis-28417	288	1	9	9	NUM
bracis-28417	288	2	,	,	PUNCT
bracis-28417	288	3	1389–1406	1389–1406	NUM
bracis-28417	288	4	(	(	PUNCT
bracis-28417	288	5	2021	2021	NUM
bracis-28417	288	6	)	)	PUNCT
bracis-28417	288	7	article	article	NOUN
bracis-28417	288	8	  	  	SPACE
bracis-28417	288	9	google	google	PROPN
bracis-28417	288	10	scholar	scholar	NOUN
bracis-28417	288	11	  	  	SPACE
bracis-28417	288	12	loshchilov	loshchilov	PROPN
bracis-28417	288	13	,	,	PUNCT
bracis-28417	288	14	i.	i.	NOUN
bracis-28417	288	15	,	,	PUNCT
bracis-28417	288	16	hutter	hutter	NOUN
bracis-28417	288	17	,	,	PUNCT
bracis-28417	288	18	f.	f.	PROPN
bracis-28417	288	19	:	:	PUNCT
bracis-28417	288	20	decoupled	decouple	VERB
bracis-28417	288	21	weight	weight	NOUN
bracis-28417	288	22	decay	decay	NOUN
bracis-28417	288	23	regularization	regularization	NOUN
bracis-28417	288	24	.	.	PUNCT
bracis-28417	289	1	in	in	ADP
bracis-28417	289	2	:	:	PUNCT
bracis-28417	289	3	international	international	ADJ
bracis-28417	289	4	conference	conference	NOUN
bracis-28417	289	5	on	on	ADP
bracis-28417	289	6	learning	learn	VERB
bracis-28417	289	7	representations	representation	NOUN
bracis-28417	289	8	(	(	PUNCT
bracis-28417	289	9	2019	2019	NUM
bracis-28417	289	10	)	)	PUNCT
bracis-28417	289	11	google	google	PROPN
bracis-28417	289	12	scholar	scholar	NOUN
bracis-28417	289	13	  	  	SPACE
bracis-28417	289	14	maas	maas	PROPN
bracis-28417	289	15	,	,	PUNCT
bracis-28417	289	16	a.l	a.l	PROPN
bracis-28417	289	17	.	.	PROPN
bracis-28417	289	18	,	,	PUNCT
bracis-28417	289	19	daly	daly	PROPN
bracis-28417	289	20	,	,	PUNCT
bracis-28417	289	21	r.e	r.e	PROPN
bracis-28417	289	22	.	.	PROPN
bracis-28417	289	23	,	,	PUNCT
bracis-28417	289	24	pham	pham	PROPN
bracis-28417	289	25	,	,	PUNCT
bracis-28417	289	26	p.t	p.t	PROPN
bracis-28417	289	27	.	.	PROPN
bracis-28417	289	28	,	,	PUNCT
bracis-28417	289	29	huang	huang	PROPN
bracis-28417	289	30	,	,	PUNCT
bracis-28417	289	31	d.	d.	PROPN
bracis-28417	289	32	,	,	PUNCT
bracis-28417	289	33	ng	ng	PROPN
bracis-28417	289	34	,	,	PUNCT
bracis-28417	289	35	a.y	a.y	PROPN
bracis-28417	289	36	.	.	PROPN
bracis-28417	289	37	,	,	PUNCT
bracis-28417	289	38	potts	potts	PROPN
bracis-28417	289	39	,	,	PUNCT
bracis-28417	289	40	c.	c.	PROPN
bracis-28417	289	41	:	:	PUNCT
bracis-28417	289	42	learning	learn	VERB
bracis-28417	289	43	word	word	NOUN
bracis-28417	289	44	vectors	vector	NOUN
bracis-28417	289	45	for	for	ADP
bracis-28417	289	46	sentiment	sentiment	NOUN
bracis-28417	289	47	analysis	analysis	NOUN
bracis-28417	289	48	.	.	PUNCT
bracis-28417	290	1	in	in	ADP
bracis-28417	290	2	:	:	PUNCT
bracis-28417	290	3	proceedings	proceeding	NOUN
bracis-28417	290	4	of	of	ADP
bracis-28417	290	5	the	the	DET
bracis-28417	290	6	49th	49th	ADJ
bracis-28417	290	7	annual	annual	ADJ
bracis-28417	290	8	meeting	meeting	NOUN
bracis-28417	290	9	of	of	ADP
bracis-28417	290	10	the	the	DET
bracis-28417	290	11	association	association	NOUN
bracis-28417	290	12	for	for	ADP
bracis-28417	290	13	computational	computational	ADJ
bracis-28417	290	14	linguistics	linguistic	NOUN
bracis-28417	290	15	:	:	PUNCT
bracis-28417	290	16	human	human	ADJ
bracis-28417	290	17	language	language	NOUN
bracis-28417	290	18	technologies	technology	NOUN
bracis-28417	290	19	,	,	PUNCT
bracis-28417	290	20	pp	pp	ADP
bracis-28417	290	21	.	.	PUNCT
bracis-28417	291	1	142–150	142–150	NUM
bracis-28417	291	2	.	.	PUNCT
bracis-28417	292	1	association	association	NOUN
bracis-28417	292	2	for	for	ADP
bracis-28417	292	3	computational	computational	ADJ
bracis-28417	292	4	linguistics	linguistic	NOUN
bracis-28417	292	5	,	,	PUNCT
bracis-28417	292	6	portland	portland	PROPN
bracis-28417	292	7	,	,	PUNCT
bracis-28417	292	8	oregon	oregon	PROPN
bracis-28417	292	9	,	,	PUNCT
bracis-28417	292	10	usa	usa	PROPN
bracis-28417	292	11	(	(	PUNCT
bracis-28417	292	12	2011	2011	NUM
bracis-28417	292	13	)	)	PUNCT
bracis-28417	292	14	google	google	PROPN
bracis-28417	292	15	scholar	scholar	NOUN
bracis-28417	292	16	  	  	SPACE
bracis-28417	292	17	martin	martin	PROPN
bracis-28417	292	18	,	,	PUNCT
bracis-28417	292	19	l.	l.	PROPN
bracis-28417	292	20	,	,	PUNCT
bracis-28417	292	21	et	et	PROPN
bracis-28417	292	22	al	al	PROPN
bracis-28417	292	23	.	.	PROPN
bracis-28417	292	24	:	:	PUNCT
bracis-28417	293	1	camembert	camembert	NOUN
bracis-28417	293	2	:	:	PUNCT
bracis-28417	293	3	a	a	DET
bracis-28417	293	4	tasty	tasty	ADJ
bracis-28417	293	5	french	french	ADJ
bracis-28417	293	6	language	language	NOUN
bracis-28417	293	7	model	model	NOUN
bracis-28417	293	8	.	.	PUNCT
bracis-28417	294	1	in	in	ADP
bracis-28417	294	2	:	:	PUNCT
bracis-28417	294	3	proceedings	proceeding	NOUN
bracis-28417	294	4	of	of	ADP
bracis-28417	294	5	the	the	DET
bracis-28417	294	6	58th	58th	ADJ
bracis-28417	294	7	annual	annual	ADJ
bracis-28417	294	8	meeting	meeting	NOUN
bracis-28417	294	9	of	of	ADP
bracis-28417	294	10	the	the	DET
bracis-28417	294	11	association	association	NOUN
bracis-28417	294	12	for	for	ADP
bracis-28417	294	13	computational	computational	ADJ
bracis-28417	294	14	linguistics	linguistic	NOUN
bracis-28417	294	15	.	.	PUNCT
bracis-28417	295	1	pp	pp	ADJ
bracis-28417	295	2	.	.	PUNCT
bracis-28417	296	1	7203–7219	7203–7219	NUM
bracis-28417	296	2	.	.	PUNCT
bracis-28417	297	1	association	association	NOUN
bracis-28417	297	2	for	for	ADP
bracis-28417	297	3	computational	computational	ADJ
bracis-28417	297	4	linguistics	linguistic	NOUN
bracis-28417	297	5	,	,	PUNCT
bracis-28417	297	6	online	online	ADV
bracis-28417	297	7	(	(	PUNCT
bracis-28417	297	8	2020	2020	NUM
bracis-28417	297	9	)	)	PUNCT
bracis-28417	297	10	.	.	PUNCT
bracis-28417	298	1	https://doi.org/10.18653/v1/2020.acl-main.645	https://doi.org/10.18653/v1/2020.acl-main.645	PROPN
bracis-28417	298	2	de	de	X
bracis-28417	298	3	melo	melo	PROPN
bracis-28417	298	4	,	,	PUNCT
bracis-28417	298	5	g.	g.	PROPN
bracis-28417	298	6	,	,	PUNCT
bracis-28417	298	7	imaizumi	imaizumi	PROPN
bracis-28417	298	8	,	,	PUNCT
bracis-28417	298	9	v.	v.	PROPN
bracis-28417	298	10	,	,	PUNCT
bracis-28417	298	11	cozman	cozman	NOUN
bracis-28417	298	12	,	,	PUNCT
bracis-28417	298	13	f.	f.	PROPN
bracis-28417	298	14	:	:	PROPN
bracis-28417	298	15	winograd	winograd	PROPN
bracis-28417	298	16	schemas	schema	NOUN
bracis-28417	298	17	in	in	ADP
bracis-28417	298	18	portuguese	portuguese	PROPN
bracis-28417	298	19	.	.	PUNCT
bracis-28417	299	1	in	in	ADP
bracis-28417	299	2	:	:	PUNCT
bracis-28417	299	3	anais	anais	PROPN
bracis-28417	299	4	do	do	AUX
bracis-28417	299	5	xvi	xvi	NOUN
bracis-28417	299	6	encontro	encontro	VERB
bracis-28417	299	7	nacional	nacional	ADJ
bracis-28417	299	8	de	de	X
bracis-28417	299	9	inteligência	inteligência	PROPN
bracis-28417	299	10	artificial	artificial	PROPN
bracis-28417	299	11	e	e	NOUN
bracis-28417	299	12	computacional	computacional	NOUN
bracis-28417	299	13	,	,	PUNCT
bracis-28417	299	14	pp	pp	ADJ
bracis-28417	299	15	.	.	PUNCT
bracis-28417	300	1	787–798	787–798	NUM
bracis-28417	300	2	.	.	PUNCT
bracis-28417	301	1	sbc	sbc	NOUN
bracis-28417	301	2	(	(	PUNCT
bracis-28417	301	3	2019	2019	NUM
bracis-28417	301	4	)	)	PUNCT
bracis-28417	301	5	google	google	PROPN
bracis-28417	301	6	scholar	scholar	NOUN
bracis-28417	301	7	  	  	SPACE
bracis-28417	301	8	moraes	moraes	PROPN
bracis-28417	301	9	,	,	PUNCT
bracis-28417	301	10	g.	g.	PROPN
bracis-28417	301	11	,	,	PUNCT
bracis-28417	301	12	bonifácio	bonifácio	PROPN
bracis-28417	301	13	,	,	PUNCT
bracis-28417	301	14	l.h	l.h	PROPN
bracis-28417	301	15	.	.	PROPN
bracis-28417	301	16	,	,	PUNCT
bracis-28417	301	17	rodrigues	rodrigues	PROPN
bracis-28417	301	18	de	de	PROPN
bracis-28417	301	19	souza	souza	PROPN
bracis-28417	301	20	,	,	PUNCT
bracis-28417	301	21	l.	l.	PROPN
bracis-28417	301	22	,	,	PUNCT
bracis-28417	301	23	nogueira	nogueira	PROPN
bracis-28417	301	24	,	,	PUNCT
bracis-28417	301	25	r.	r.	PROPN
bracis-28417	301	26	,	,	PUNCT
bracis-28417	301	27	lotufo	lotufo	PROPN
bracis-28417	301	28	,	,	PUNCT
bracis-28417	301	29	r.	r.	PROPN
bracis-28417	301	30	:	:	PUNCT
bracis-28417	301	31	a	a	DET
bracis-28417	301	32	cost	cost	NOUN
bracis-28417	301	33	-	-	PUNCT
bracis-28417	301	34	benefit	benefit	NOUN
bracis-28417	301	35	analysis	analysis	NOUN
bracis-28417	301	36	of	of	ADP
bracis-28417	301	37	cross	cross	ADJ
bracis-28417	301	38	-	-	ADJ
bracis-28417	301	39	lingual	lingual	ADJ
bracis-28417	301	40	transfer	transfer	NOUN
bracis-28417	301	41	methods	method	NOUN
bracis-28417	301	42	.	.	PUNCT
bracis-28417	302	1	arxiv	arxiv	PROPN
bracis-28417	302	2	preprint	preprint	VERB
bracis-28417	302	3	arxiv:2105.06813	arxiv:2105.06813	PROPN
bracis-28417	302	4	(	(	PUNCT
bracis-28417	302	5	2021	2021	NUM
bracis-28417	302	6	)	)	PUNCT
bracis-28417	302	7	.	.	PUNCT
bracis-28417	303	1	https://arxiv.org/abs/2105.06813	https://arxiv.org/abs/2105.06813	PROPN
bracis-28417	303	2	muennighoff	muennighoff	PROPN
bracis-28417	303	3	,	,	PUNCT
bracis-28417	303	4	n.	n.	PROPN
bracis-28417	303	5	,	,	PUNCT
bracis-28417	303	6	et	et	PROPN
bracis-28417	303	7	al	al	PROPN
bracis-28417	303	8	.	.	PROPN
bracis-28417	303	9	:	:	PUNCT
bracis-28417	303	10	crosslingual	crosslingual	ADJ
bracis-28417	303	11	generalization	generalization	NOUN
bracis-28417	303	12	through	through	ADP
bracis-28417	303	13	multitask	multitask	ADJ
bracis-28417	303	14	finetuning	finetuning	NOUN
bracis-28417	303	15	(	(	PUNCT
bracis-28417	303	16	2022	2022	NUM
bracis-28417	303	17	)	)	PUNCT
bracis-28417	303	18	google	google	NOUN
bracis-28417	303	19	scholar	scholar	NOUN
bracis-28417	303	20	  	  	SPACE
bracis-28417	303	21	nguyen	nguyen	NOUN
bracis-28417	303	22	,	,	PUNCT
bracis-28417	303	23	d.q	d.q	PROPN
bracis-28417	303	24	.	.	PROPN
bracis-28417	303	25	,	,	PUNCT
bracis-28417	303	26	tuan	tuan	PROPN
bracis-28417	303	27	nguyen	nguyen	PROPN
bracis-28417	303	28	,	,	PUNCT
bracis-28417	303	29	a.	a.	NOUN
bracis-28417	303	30	:	:	PUNCT
bracis-28417	303	31	phobert	phobert	NOUN
bracis-28417	303	32	:	:	PUNCT
bracis-28417	303	33	pre	pre	ADJ
bracis-28417	303	34	-	-	ADJ
bracis-28417	303	35	trained	train	VERB
bracis-28417	303	36	language	language	NOUN
bracis-28417	303	37	models	model	NOUN
bracis-28417	303	38	for	for	ADP
bracis-28417	303	39	vietnamese	vietnamese	NOUN
bracis-28417	303	40	.	.	PUNCT
bracis-28417	304	1	in	in	ADP
bracis-28417	304	2	:	:	PUNCT
bracis-28417	304	3	findings	finding	NOUN
bracis-28417	304	4	of	of	ADP
bracis-28417	304	5	the	the	DET
bracis-28417	304	6	association	association	NOUN
bracis-28417	304	7	for	for	ADP
bracis-28417	304	8	computational	computational	ADJ
bracis-28417	304	9	linguistics	linguistic	NOUN
bracis-28417	304	10	:	:	PUNCT
bracis-28417	304	11	emnlp	emnlp	PROPN
bracis-28417	304	12	2020	2020	NUM
bracis-28417	304	13	,	,	PUNCT
bracis-28417	304	14	pp	pp	ADV
bracis-28417	304	15	.	.	PUNCT
bracis-28417	305	1	1037–1042	1037–1042	NUM
bracis-28417	305	2	.	.	PUNCT
bracis-28417	305	3	association	association	NOUN
bracis-28417	305	4	for	for	ADP
bracis-28417	305	5	computational	computational	ADJ
bracis-28417	305	6	linguistics	linguistic	NOUN
bracis-28417	305	7	,	,	PUNCT
bracis-28417	305	8	online	online	ADV
bracis-28417	305	9	(	(	PUNCT
bracis-28417	305	10	2020	2020	NUM
bracis-28417	305	11	)	)	PUNCT
bracis-28417	305	12	.	.	PUNCT
bracis-28417	306	1	https://doi.org/10.18653/v1/2020.findings-emnlp.92	https://doi.org/10.18653/v1/2020.findings-emnlp.92	PROPN
bracis-28417	306	2	nunes	nunes	PROPN
bracis-28417	306	3	,	,	PUNCT
bracis-28417	306	4	d.	d.	PROPN
bracis-28417	306	5	,	,	PUNCT
bracis-28417	306	6	primi	primi	PROPN
bracis-28417	306	7	,	,	PUNCT
bracis-28417	306	8	r.	r.	PROPN
bracis-28417	306	9	,	,	PUNCT
bracis-28417	306	10	pires	pire	NOUN
bracis-28417	306	11	,	,	PUNCT
bracis-28417	306	12	r.	r.	PROPN
bracis-28417	306	13	,	,	PUNCT
bracis-28417	306	14	lotufo	lotufo	PROPN
bracis-28417	306	15	,	,	PUNCT
bracis-28417	306	16	r.	r.	PROPN
bracis-28417	306	17	,	,	PUNCT
bracis-28417	306	18	nogueira	nogueira	PROPN
bracis-28417	306	19	,	,	PUNCT
bracis-28417	306	20	r.	r.	PROPN
bracis-28417	306	21	:	:	PUNCT
bracis-28417	306	22	evaluating	evaluate	VERB
bracis-28417	306	23	gpt-3.5	gpt-3.5	PUNCT
bracis-28417	306	24	and	and	CCONJ
bracis-28417	306	25	gpt-4	gpt-4	NOUN
bracis-28417	306	26	models	model	NOUN
bracis-28417	306	27	on	on	ADP
bracis-28417	306	28	brazilian	brazilian	ADJ
bracis-28417	306	29	university	university	NOUN
bracis-28417	306	30	admission	admission	NOUN
bracis-28417	306	31	exams	exam	NOUN
bracis-28417	306	32	(	(	PUNCT
bracis-28417	306	33	2023	2023	NUM
bracis-28417	306	34	)	)	PUNCT
bracis-28417	306	35	google	google	NOUN
bracis-28417	306	36	scholar	scholar	NOUN
bracis-28417	306	37	  	  	SPACE
bracis-28417	306	38	ogueji	ogueji	NOUN
bracis-28417	306	39	,	,	PUNCT
bracis-28417	306	40	k.	k.	PROPN
bracis-28417	306	41	,	,	PUNCT
bracis-28417	306	42	zhu	zhu	PROPN
bracis-28417	306	43	,	,	PUNCT
bracis-28417	306	44	y.	y.	PROPN
bracis-28417	306	45	,	,	PUNCT
bracis-28417	306	46	lin	lin	PROPN
bracis-28417	306	47	,	,	PUNCT
bracis-28417	306	48	j.	j.	PROPN
bracis-28417	306	49	:	:	PUNCT
bracis-28417	306	50	small	small	ADJ
bracis-28417	306	51	data	datum	NOUN
bracis-28417	306	52	?	?	PUNCT
bracis-28417	307	1	no	no	DET
bracis-28417	307	2	problem	problem	NOUN
bracis-28417	307	3	!	!	PUNCT
bracis-28417	308	1	exploring	explore	VERB
bracis-28417	308	2	the	the	DET
bracis-28417	308	3	viability	viability	NOUN
bracis-28417	308	4	of	of	ADP
bracis-28417	308	5	pretrained	pretraine	VERB
bracis-28417	308	6	multilingual	multilingual	ADJ
bracis-28417	308	7	language	language	NOUN
bracis-28417	308	8	models	model	NOUN
bracis-28417	308	9	for	for	ADP
bracis-28417	308	10	low	low	ADV
bracis-28417	308	11	-	-	PUNCT
bracis-28417	308	12	resourced	resource	VERB
bracis-28417	308	13	languages	language	NOUN
bracis-28417	308	14	.	.	PUNCT
bracis-28417	309	1	in	in	ADP
bracis-28417	309	2	:	:	PUNCT
bracis-28417	309	3	proceedings	proceeding	NOUN
bracis-28417	309	4	of	of	ADP
bracis-28417	309	5	the	the	DET
bracis-28417	309	6	1st	1st	ADJ
bracis-28417	309	7	workshop	workshop	NOUN
bracis-28417	309	8	on	on	ADP
bracis-28417	309	9	multilingual	multilingual	ADJ
bracis-28417	309	10	representation	representation	NOUN
bracis-28417	309	11	learning	learning	NOUN
bracis-28417	309	12	,	,	PUNCT
bracis-28417	309	13	pp	pp	ADJ
bracis-28417	309	14	.	.	PUNCT
bracis-28417	310	1	116–126	116–126	NUM
bracis-28417	310	2	.	.	PUNCT
bracis-28417	310	3	association	association	NOUN
bracis-28417	310	4	for	for	ADP
bracis-28417	310	5	computational	computational	ADJ
bracis-28417	310	6	linguistics	linguistic	NOUN
bracis-28417	310	7	,	,	PUNCT
bracis-28417	310	8	punta	punta	PROPN
bracis-28417	310	9	cana	cana	PROPN
bracis-28417	310	10	,	,	PUNCT
bracis-28417	310	11	dominican	dominican	PROPN
bracis-28417	310	12	republic	republic	NOUN
bracis-28417	310	13	(	(	PUNCT
bracis-28417	310	14	2021	2021	NUM
bracis-28417	310	15	)	)	PUNCT
bracis-28417	310	16	google	google	NOUN
bracis-28417	310	17	scholar	scholar	NOUN
bracis-28417	310	18	  	  	SPACE
bracis-28417	310	19	openai	openai	PROPN
bracis-28417	310	20	:	:	PUNCT
bracis-28417	310	21	gpt-4	gpt-4	NUM
bracis-28417	310	22	technical	technical	ADJ
bracis-28417	310	23	report	report	NOUN
bracis-28417	310	24	(	(	PUNCT
bracis-28417	310	25	2023	2023	NUM
bracis-28417	310	26	)	)	PUNCT
bracis-28417	310	27	google	google	NOUN
bracis-28417	310	28	scholar	scholar	NOUN
bracis-28417	310	29	  	  	SPACE
bracis-28417	310	30	overwijk	overwijk	NOUN
bracis-28417	310	31	,	,	PUNCT
bracis-28417	310	32	a.	a.	NOUN
bracis-28417	310	33	,	,	PUNCT
bracis-28417	310	34	xiong	xiong	PROPN
bracis-28417	310	35	,	,	PUNCT
bracis-28417	310	36	c.	c.	PROPN
bracis-28417	310	37	,	,	PUNCT
bracis-28417	310	38	callan	callan	PROPN
bracis-28417	310	39	,	,	PUNCT
bracis-28417	310	40	j.	j.	PROPN
bracis-28417	310	41	:	:	PUNCT
bracis-28417	310	42	clueweb	clueweb	PROPN
bracis-28417	310	43	22	22	NUM
bracis-28417	310	44	:	:	SYM
bracis-28417	310	45	10	10	NUM
bracis-28417	310	46	billion	billion	NUM
bracis-28417	310	47	web	web	NOUN
bracis-28417	310	48	documents	document	NOUN
bracis-28417	310	49	with	with	ADP
bracis-28417	310	50	rich	rich	ADJ
bracis-28417	310	51	information	information	NOUN
bracis-28417	310	52	.	.	PUNCT
bracis-28417	311	1	in	in	ADP
bracis-28417	311	2	:	:	PUNCT
bracis-28417	311	3	proceedings	proceeding	NOUN
bracis-28417	311	4	of	of	ADP
bracis-28417	311	5	the	the	DET
bracis-28417	311	6	45th	45th	ADJ
bracis-28417	311	7	international	international	PROPN
bracis-28417	311	8	acm	acm	PROPN
bracis-28417	311	9	sigir	sigir	PROPN
bracis-28417	311	10	conference	conference	PROPN
bracis-28417	311	11	on	on	ADP
bracis-28417	311	12	research	research	NOUN
bracis-28417	311	13	and	and	CCONJ
bracis-28417	311	14	development	development	NOUN
bracis-28417	311	15	in	in	ADP
bracis-28417	311	16	information	information	NOUN
bracis-28417	311	17	retrieval	retrieval	NOUN
bracis-28417	311	18	,	,	PUNCT
bracis-28417	311	19	pp	pp	ADP
bracis-28417	311	20	.	.	PUNCT
bracis-28417	312	1	3360–3362	3360–3362	NUM
bracis-28417	312	2	(	(	PUNCT
bracis-28417	312	3	2022	2022	NUM
bracis-28417	312	4	)	)	PUNCT
bracis-28417	312	5	google	google	NOUN
bracis-28417	312	6	scholar	scholar	NOUN
bracis-28417	312	7	  	  	SPACE
bracis-28417	312	8	overwijk	overwijk	NOUN
bracis-28417	312	9	,	,	PUNCT
bracis-28417	312	10	a.	a.	NOUN
bracis-28417	312	11	,	,	PUNCT
bracis-28417	312	12	xiong	xiong	PROPN
bracis-28417	312	13	,	,	PUNCT
bracis-28417	312	14	c.	c.	PROPN
bracis-28417	312	15	,	,	PUNCT
bracis-28417	312	16	liu	liu	PROPN
bracis-28417	312	17	,	,	PUNCT
bracis-28417	312	18	x.	x.	PROPN
bracis-28417	312	19	,	,	PUNCT
bracis-28417	312	20	vandenberg	vandenberg	PROPN
bracis-28417	312	21	,	,	PUNCT
bracis-28417	312	22	c.	c.	PROPN
bracis-28417	312	23	,	,	PUNCT
bracis-28417	312	24	callan	callan	PROPN
bracis-28417	312	25	,	,	PUNCT
bracis-28417	312	26	j.	j.	PROPN
bracis-28417	312	27	:	:	PUNCT
bracis-28417	312	28	clueweb	clueweb	PROPN
bracis-28417	312	29	22	22	NUM
bracis-28417	312	30	:	:	SYM
bracis-28417	312	31	10	10	NUM
bracis-28417	312	32	billion	billion	NUM
bracis-28417	312	33	web	web	NOUN
bracis-28417	312	34	documents	document	NOUN
bracis-28417	312	35	with	with	ADP
bracis-28417	312	36	visual	visual	ADJ
bracis-28417	312	37	and	and	CCONJ
bracis-28417	312	38	semantic	semantic	ADJ
bracis-28417	312	39	information	information	NOUN
bracis-28417	312	40	(	(	PUNCT
bracis-28417	312	41	2022	2022	NUM
bracis-28417	312	42	)	)	PUNCT
bracis-28417	312	43	google	google	PROPN
bracis-28417	312	44	scholar	scholar	NOUN
bracis-28417	312	45	  	  	SPACE
bracis-28417	312	46	pfeiffer	pfeiffer	PROPN
bracis-28417	312	47	,	,	PUNCT
bracis-28417	312	48	j.	j.	PROPN
bracis-28417	312	49	,	,	PUNCT
bracis-28417	312	50	kamath	kamath	PROPN
bracis-28417	312	51	,	,	PUNCT
bracis-28417	312	52	a.	a.	PROPN
bracis-28417	312	53	,	,	PUNCT
bracis-28417	312	54	rücklé	rücklé	PROPN
bracis-28417	312	55	,	,	PUNCT
bracis-28417	312	56	a.	a.	PROPN
bracis-28417	312	57	,	,	PUNCT
bracis-28417	312	58	cho	cho	PROPN
bracis-28417	312	59	,	,	PUNCT
bracis-28417	312	60	k.	k.	PROPN
bracis-28417	312	61	,	,	PUNCT
bracis-28417	312	62	gurevych	gurevych	PROPN
bracis-28417	312	63	,	,	PUNCT
bracis-28417	312	64	i.	i.	NOUN
bracis-28417	312	65	:	:	PUNCT
bracis-28417	312	66	adapterfusion	adapterfusion	NOUN
bracis-28417	312	67	:	:	PUNCT
bracis-28417	312	68	non	non	ADJ
bracis-28417	312	69	-	-	ADJ
bracis-28417	312	70	destructive	destructive	ADJ
bracis-28417	312	71	task	task	NOUN
bracis-28417	312	72	composition	composition	NOUN
bracis-28417	312	73	for	for	ADP
bracis-28417	312	74	transfer	transfer	NOUN
bracis-28417	312	75	learning	learning	NOUN
bracis-28417	312	76	.	.	PUNCT
bracis-28417	313	1	in	in	ADP
bracis-28417	313	2	:	:	PUNCT
bracis-28417	313	3	proceedings	proceeding	NOUN
bracis-28417	313	4	of	of	ADP
bracis-28417	313	5	the	the	DET
bracis-28417	313	6	16th	16th	ADJ
bracis-28417	313	7	conference	conference	NOUN
bracis-28417	313	8	of	of	ADP
bracis-28417	313	9	the	the	DET
bracis-28417	313	10	european	european	ADJ
bracis-28417	313	11	chapter	chapter	NOUN
bracis-28417	313	12	of	of	ADP
bracis-28417	313	13	the	the	DET
bracis-28417	313	14	association	association	NOUN
bracis-28417	313	15	for	for	ADP
bracis-28417	313	16	computational	computational	ADJ
bracis-28417	313	17	linguistics	linguistic	NOUN
bracis-28417	313	18	:	:	PUNCT
bracis-28417	313	19	main	main	ADJ
bracis-28417	313	20	volume	volume	NOUN
bracis-28417	313	21	,	,	PUNCT
bracis-28417	313	22	pp	pp	X
bracis-28417	313	23	.	.	PUNCT
bracis-28417	314	1	487–503	487–503	NUM
bracis-28417	314	2	.	.	PUNCT
bracis-28417	315	1	association	association	NOUN
bracis-28417	315	2	for	for	ADP
bracis-28417	315	3	computational	computational	ADJ
bracis-28417	315	4	linguistics	linguistic	NOUN
bracis-28417	315	5	,	,	PUNCT
bracis-28417	315	6	online	online	ADV
bracis-28417	315	7	(	(	PUNCT
bracis-28417	315	8	2021	2021	NUM
bracis-28417	315	9	)	)	PUNCT
bracis-28417	315	10	.	.	PUNCT
bracis-28417	316	1	https://doi.org/10.18653/v1/2021.eacl-main.39	https://doi.org/10.18653/v1/2021.eacl-main.39	PROPN
bracis-28417	316	2	pfeiffer	pfeiffer	PROPN
bracis-28417	316	3	,	,	PUNCT
bracis-28417	316	4	j.	j.	PROPN
bracis-28417	316	5	,	,	PUNCT
bracis-28417	316	6	vulić	vulić	PROPN
bracis-28417	316	7	,	,	PUNCT
bracis-28417	316	8	i.	i.	NOUN
bracis-28417	316	9	,	,	PUNCT
bracis-28417	316	10	gurevych	gurevych	PROPN
bracis-28417	316	11	,	,	PUNCT
bracis-28417	316	12	i.	i.	NOUN
bracis-28417	316	13	,	,	PUNCT
bracis-28417	316	14	ruder	ruder	PROPN
bracis-28417	316	15	,	,	PUNCT
bracis-28417	316	16	s.	s.	PROPN
bracis-28417	316	17	:	:	PUNCT
bracis-28417	316	18	mad	mad	ADJ
bracis-28417	316	19	-	-	PUNCT
bracis-28417	316	20	x	x	NOUN
bracis-28417	316	21	:	:	PUNCT
bracis-28417	316	22	an	an	DET
bracis-28417	316	23	adapter	adapter	NOUN
bracis-28417	316	24	-	-	PUNCT
bracis-28417	316	25	based	base	VERB
bracis-28417	316	26	framework	framework	NOUN
bracis-28417	316	27	for	for	ADP
bracis-28417	316	28	multi	multi	ADJ
bracis-28417	316	29	-	-	ADJ
bracis-28417	316	30	task	task	ADJ
bracis-28417	316	31	cross	cross	ADJ
bracis-28417	316	32	-	-	ADJ
bracis-28417	316	33	lingual	lingual	ADJ
bracis-28417	316	34	transfer	transfer	NOUN
bracis-28417	316	35	.	.	PUNCT
bracis-28417	317	1	arxiv	arxiv	PROPN
bracis-28417	317	2	preprint	preprint	VERB
bracis-28417	317	3	arxiv:2005.00052	arxiv:2005.00052	PROPN
bracis-28417	317	4	(	(	PUNCT
bracis-28417	317	5	2020	2020	NUM
bracis-28417	317	6	)	)	PUNCT
bracis-28417	318	1	radford	radford	PROPN
bracis-28417	318	2	,	,	PUNCT
bracis-28417	318	3	a.	a.	PROPN
bracis-28417	318	4	,	,	PUNCT
bracis-28417	318	5	wu	wu	PROPN
bracis-28417	318	6	,	,	PUNCT
bracis-28417	318	7	j.	j.	PROPN
bracis-28417	318	8	,	,	PUNCT
bracis-28417	318	9	child	child	NOUN
bracis-28417	318	10	,	,	PUNCT
bracis-28417	318	11	r.	r.	PROPN
bracis-28417	318	12	,	,	PUNCT
bracis-28417	318	13	luan	luan	PROPN
bracis-28417	318	14	,	,	PUNCT
bracis-28417	318	15	d.	d.	PROPN
bracis-28417	318	16	,	,	PUNCT
bracis-28417	318	17	amodei	amodei	PROPN
bracis-28417	318	18	,	,	PUNCT
bracis-28417	318	19	d.	d.	PROPN
bracis-28417	318	20	,	,	PUNCT
bracis-28417	318	21	sutskever	sutskever	PROPN
bracis-28417	318	22	,	,	PUNCT
bracis-28417	318	23	i.	i.	PROPN
bracis-28417	318	24	,	,	PUNCT
bracis-28417	318	25	et	et	PROPN
bracis-28417	318	26	al	al	PROPN
bracis-28417	318	27	.	.	PUNCT
bracis-28417	318	28	:	:	PUNCT
bracis-28417	319	1	language	language	NOUN
bracis-28417	319	2	models	model	NOUN
bracis-28417	319	3	are	be	AUX
bracis-28417	319	4	unsupervised	unsupervised	ADJ
bracis-28417	319	5	multitask	multitask	ADJ
bracis-28417	319	6	learners	learner	NOUN
bracis-28417	319	7	.	.	PUNCT
bracis-28417	320	1	openai	openai	PROPN
bracis-28417	320	2	blog	blog	NOUN
bracis-28417	320	3	1(8	1(8	NUM
bracis-28417	320	4	)	)	PUNCT
bracis-28417	320	5	,	,	PUNCT
bracis-28417	320	6	9	9	NUM
bracis-28417	320	7	(	(	PUNCT
bracis-28417	320	8	2019	2019	NUM
bracis-28417	320	9	)	)	PUNCT
bracis-28417	320	10	google	google	PROPN
bracis-28417	320	11	scholar	scholar	NOUN
bracis-28417	320	12	  	  	SPACE
bracis-28417	320	13	rae	rae	PROPN
bracis-28417	320	14	,	,	PUNCT
bracis-28417	320	15	j.w	j.w	PROPN
bracis-28417	320	16	.	.	PROPN
bracis-28417	320	17	,	,	PUNCT
bracis-28417	320	18	et	et	PROPN
bracis-28417	320	19	al	al	PROPN
bracis-28417	320	20	.	.	PUNCT
bracis-28417	320	21	:	:	PUNCT
bracis-28417	321	1	scaling	scale	VERB
bracis-28417	321	2	language	language	NOUN
bracis-28417	321	3	models	model	NOUN
bracis-28417	321	4	:	:	PUNCT
bracis-28417	321	5	methods	method	NOUN
bracis-28417	321	6	,	,	PUNCT
bracis-28417	321	7	analysis	analysis	NOUN
bracis-28417	321	8	&	&	CCONJ
bracis-28417	321	9	insights	insight	NOUN
bracis-28417	321	10	from	from	ADP
bracis-28417	321	11	training	train	VERB
bracis-28417	321	12	gopher	gopher	NOUN
bracis-28417	321	13	.	.	PUNCT
bracis-28417	322	1	arxiv	arxiv	PROPN
bracis-28417	322	2	preprint	preprint	NOUN
bracis-28417	322	3	arxiv:2112.11446	arxiv:2112.11446	NOUN
bracis-28417	322	4	(	(	PUNCT
bracis-28417	322	5	2021	2021	NUM
bracis-28417	322	6	)	)	PUNCT
bracis-28417	322	7	raffel	raffel	NOUN
bracis-28417	322	8	,	,	PUNCT
bracis-28417	322	9	c.	c.	PROPN
bracis-28417	322	10	,	,	PUNCT
bracis-28417	322	11	et	et	PROPN
bracis-28417	322	12	al	al	PROPN
bracis-28417	322	13	.	.	PROPN
bracis-28417	322	14	:	:	PUNCT
bracis-28417	322	15	exploring	explore	VERB
bracis-28417	322	16	the	the	DET
bracis-28417	322	17	limits	limit	NOUN
bracis-28417	322	18	of	of	ADP
bracis-28417	322	19	transfer	transfer	NOUN
bracis-28417	322	20	learning	learn	VERB
bracis-28417	322	21	with	with	ADP
bracis-28417	322	22	a	a	DET
bracis-28417	322	23	unified	unified	ADJ
bracis-28417	322	24	text	text	NOUN
bracis-28417	322	25	-	-	PUNCT
bracis-28417	322	26	to	to	ADP
bracis-28417	322	27	-	-	PUNCT
bracis-28417	322	28	text	text	NOUN
bracis-28417	322	29	transformer	transformer	NOUN
bracis-28417	322	30	.	.	PUNCT
bracis-28417	323	1	j.	j.	PROPN
bracis-28417	323	2	mach	mach	PROPN
bracis-28417	323	3	.	.	PUNCT
bracis-28417	324	1	learn	learn	VERB
bracis-28417	324	2	.	.	PUNCT
bracis-28417	325	1	res	re	NOUN
bracis-28417	325	2	.	.	PUNCT
bracis-28417	326	1	21(1	21(1	NUM
bracis-28417	326	2	)	)	PUNCT
bracis-28417	327	1	,	,	PUNCT
bracis-28417	327	2	5485–5551	5485–5551	NUM
bracis-28417	327	3	(	(	PUNCT
bracis-28417	327	4	2020	2020	NUM
bracis-28417	327	5	)	)	PUNCT
bracis-28417	327	6	mathscinet	mathscinet	NOUN
bracis-28417	327	7	  	  	SPACE
bracis-28417	327	8	google	google	PROPN
bracis-28417	327	9	scholar	scholar	NOUN
bracis-28417	327	10	  	  	SPACE
bracis-28417	327	11	real	real	ADJ
bracis-28417	327	12	,	,	PUNCT
bracis-28417	327	13	l.	l.	PROPN
bracis-28417	327	14	,	,	PUNCT
bracis-28417	327	15	fonseca	fonseca	PROPN
bracis-28417	327	16	,	,	PUNCT
bracis-28417	327	17	e.	e.	PROPN
bracis-28417	327	18	,	,	PUNCT
bracis-28417	327	19	gonçalo	gonçalo	PROPN
bracis-28417	327	20	oliveira	oliveira	PROPN
bracis-28417	327	21	,	,	PUNCT
bracis-28417	327	22	h.	h.	PROPN
bracis-28417	327	23	:	:	PUNCT
bracis-28417	327	24	the	the	DET
bracis-28417	327	25	assin	assin	PROPN
bracis-28417	327	26	2	2	NUM
bracis-28417	327	27	shared	share	VERB
bracis-28417	327	28	task	task	NOUN
bracis-28417	327	29	:	:	PUNCT
bracis-28417	327	30	a	a	DET
bracis-28417	327	31	quick	quick	ADJ
bracis-28417	327	32	overview	overview	NOUN
bracis-28417	327	33	.	.	PUNCT
bracis-28417	328	1	in	in	ADP
bracis-28417	328	2	:	:	PUNCT
bracis-28417	328	3	quaresma	quaresma	PROPN
bracis-28417	328	4	,	,	PUNCT
bracis-28417	328	5	p.	p.	PROPN
bracis-28417	328	6	,	,	PUNCT
bracis-28417	328	7	vieira	vieira	PROPN
bracis-28417	328	8	,	,	PUNCT
bracis-28417	328	9	r.	r.	PROPN
bracis-28417	328	10	,	,	PUNCT
bracis-28417	328	11	aluísio	aluísio	NOUN
bracis-28417	328	12	,	,	PUNCT
bracis-28417	328	13	s.	s.	PROPN
bracis-28417	328	14	,	,	PUNCT
bracis-28417	328	15	moniz	moniz	PROPN
bracis-28417	328	16	,	,	PUNCT
bracis-28417	328	17	h.	h.	PROPN
bracis-28417	328	18	,	,	PUNCT
bracis-28417	328	19	batista	batista	PROPN
bracis-28417	328	20	,	,	PUNCT
bracis-28417	328	21	f.	f.	PROPN
bracis-28417	328	22	,	,	PUNCT
bracis-28417	328	23	gonçalves	gonçalves	PROPN
bracis-28417	328	24	,	,	PUNCT
bracis-28417	328	25	t.	t.	PROPN
bracis-28417	328	26	(	(	PUNCT
bracis-28417	328	27	eds	eds	PROPN
bracis-28417	328	28	.	.	PUNCT
bracis-28417	328	29	)	)	PUNCT
bracis-28417	328	30	propor	propor	ADJ
bracis-28417	328	31	2020	2020	NUM
bracis-28417	328	32	.	.	PUNCT
bracis-28417	329	1	lncs	lncs	PROPN
bracis-28417	329	2	(	(	PUNCT
bracis-28417	329	3	lnai	lnai	ADJ
bracis-28417	329	4	)	)	PUNCT
bracis-28417	329	5	,	,	PUNCT
bracis-28417	329	6	vol	vol	NOUN
bracis-28417	329	7	.	.	PROPN
bracis-28417	329	8	12037	12037	NUM
bracis-28417	329	9	,	,	PUNCT
bracis-28417	329	10	pp	pp	ADP
bracis-28417	329	11	.	.	PUNCT
bracis-28417	330	1	406–412	406–412	NUM
bracis-28417	330	2	.	.	PUNCT
bracis-28417	330	3	springer	springer	NOUN
bracis-28417	330	4	,	,	PUNCT
bracis-28417	330	5	cham	cham	PROPN
bracis-28417	330	6	(	(	PUNCT
bracis-28417	330	7	2020	2020	NUM
bracis-28417	330	8	)	)	PUNCT
bracis-28417	330	9	.	.	PUNCT
bracis-28417	331	1	https://doi.org/10.1007/978-3-030-41505-1_39	https://doi.org/10.1007/978-3-030-41505-1_39	PROPN
bracis-28417	331	2	chapter	chapter	NOUN
bracis-28417	331	3	  	  	SPACE
bracis-28417	331	4	google	google	PROPN
bracis-28417	331	5	scholar	scholar	NOUN
bracis-28417	331	6	  	  	SPACE
bracis-28417	331	7	roberts	roberts	PROPN
bracis-28417	331	8	,	,	PUNCT
bracis-28417	331	9	a.	a.	PROPN
bracis-28417	331	10	,	,	PUNCT
bracis-28417	331	11	et	et	PROPN
bracis-28417	331	12	al	al	PROPN
bracis-28417	331	13	.	.	PUNCT
bracis-28417	331	14	:	:	PUNCT
bracis-28417	332	1	scaling	scale	VERB
bracis-28417	332	2	up	up	ADP
bracis-28417	332	3	models	model	NOUN
bracis-28417	332	4	and	and	CCONJ
bracis-28417	332	5	data	datum	NOUN
bracis-28417	332	6	with	with	ADP
bracis-28417	332	7	t5x	t5x	NOUN
bracis-28417	332	8	and	and	CCONJ
bracis-28417	332	9	seqio	seqio	NOUN
bracis-28417	332	10	.	.	PUNCT
bracis-28417	333	1	arxiv	arxiv	PROPN
bracis-28417	333	2	preprint	preprint	NOUN
bracis-28417	333	3	arxiv:2203.17189	arxiv:2203.17189	PROPN
bracis-28417	333	4	13	13	NUM
bracis-28417	333	5	(	(	PUNCT
bracis-28417	333	6	2022	2022	NUM
bracis-28417	333	7	)	)	PUNCT
bracis-28417	333	8	rosa	rosa	PROPN
bracis-28417	333	9	,	,	PUNCT
bracis-28417	333	10	g.m	g.m	PROPN
bracis-28417	333	11	.	.	PROPN
bracis-28417	333	12	,	,	PUNCT
bracis-28417	333	13	bonifacio	bonifacio	PROPN
bracis-28417	333	14	,	,	PUNCT
bracis-28417	333	15	l.h	l.h	PROPN
bracis-28417	333	16	.	.	PROPN
bracis-28417	333	17	,	,	PUNCT
bracis-28417	333	18	de	de	PROPN
bracis-28417	333	19	souza	souza	PROPN
bracis-28417	333	20	,	,	PUNCT
bracis-28417	333	21	l.r	l.r	PROPN
bracis-28417	333	22	.	.	PROPN
bracis-28417	333	23	,	,	PUNCT
bracis-28417	333	24	lotufo	lotufo	PROPN
bracis-28417	333	25	,	,	PUNCT
bracis-28417	333	26	r.	r.	PROPN
bracis-28417	333	27	,	,	PUNCT
bracis-28417	333	28	nogueira	nogueira	PROPN
bracis-28417	333	29	,	,	PUNCT
bracis-28417	333	30	r.	r.	PROPN
bracis-28417	333	31	:	:	PUNCT
bracis-28417	333	32	a	a	DET
bracis-28417	333	33	cost	cost	NOUN
bracis-28417	333	34	-	-	PUNCT
bracis-28417	333	35	benefit	benefit	NOUN
bracis-28417	333	36	analysis	analysis	NOUN
bracis-28417	333	37	of	of	ADP
bracis-28417	333	38	cross	cross	ADJ
bracis-28417	333	39	-	-	ADJ
bracis-28417	333	40	lingual	lingual	ADJ
bracis-28417	333	41	transfer	transfer	NOUN
bracis-28417	333	42	methods	method	NOUN
bracis-28417	333	43	.	.	PUNCT
bracis-28417	334	1	arxiv	arxiv	PROPN
bracis-28417	334	2	preprint	preprint	VERB
bracis-28417	334	3	arxiv:2105.06813	arxiv:2105.06813	PROPN
bracis-28417	334	4	(	(	PUNCT
bracis-28417	334	5	2021	2021	NUM
bracis-28417	334	6	)	)	PUNCT
bracis-28417	334	7	la	la	PROPN
bracis-28417	334	8	rosa	rosa	PROPN
bracis-28417	334	9	,	,	PUNCT
bracis-28417	334	10	j.d	j.d	PROPN
bracis-28417	334	11	.	.	PROPN
bracis-28417	334	12	,	,	PUNCT
bracis-28417	334	13	fernández	fernández	PROPN
bracis-28417	334	14	,	,	PUNCT
bracis-28417	334	15	a.	a.	NOUN
bracis-28417	334	16	:	:	PUNCT
bracis-28417	334	17	zero	zero	NUM
bracis-28417	334	18	-	-	PUNCT
bracis-28417	334	19	shot	shot	NOUN
bracis-28417	334	20	reading	reading	NOUN
bracis-28417	334	21	comprehension	comprehension	NOUN
bracis-28417	334	22	and	and	CCONJ
bracis-28417	334	23	reasoning	reasoning	NOUN
bracis-28417	334	24	for	for	ADP
bracis-28417	334	25	spanish	spanish	ADJ
bracis-28417	334	26	with	with	ADP
bracis-28417	334	27	bertin	bertin	PROPN
bracis-28417	334	28	gpt	gpt	PROPN
bracis-28417	334	29	-	-	PUNCT
bracis-28417	334	30	j-6b	j-6b	PROPN
bracis-28417	334	31	.	.	PUNCT
bracis-28417	335	1	in	in	ADP
bracis-28417	335	2	:	:	PUNCT
bracis-28417	335	3	y	y	PROPN
bracis-28417	335	4	gómez	gómez	NOUN
bracis-28417	335	5	,	,	PUNCT
bracis-28417	335	6	m.m	m.m	PROPN
bracis-28417	335	7	.	.	PROPN
bracis-28417	335	8	,	,	PUNCT
bracis-28417	335	9	(	(	PUNCT
bracis-28417	335	10	eds	ed	NOUN
bracis-28417	335	11	.	.	PUNCT
bracis-28417	335	12	)	)	PUNCT
bracis-28417	335	13	proceedings	proceeding	NOUN
bracis-28417	335	14	of	of	ADP
bracis-28417	335	15	the	the	DET
bracis-28417	335	16	iberian	iberian	PROPN
bracis-28417	335	17	languages	languages	PROPN
bracis-28417	335	18	evaluation	evaluation	NOUN
bracis-28417	335	19	forum	forum	NOUN
bracis-28417	335	20	(	(	PUNCT
bracis-28417	335	21	iberlef	iberlef	PROPN
bracis-28417	335	22	2022	2022	NUM
bracis-28417	335	23	)	)	PUNCT
bracis-28417	335	24	.	.	PUNCT
bracis-28417	336	1	ceur	ceur	PROPN
bracis-28417	336	2	workshop	workshop	NOUN
bracis-28417	336	3	proceedings	proceeding	NOUN
bracis-28417	336	4	(	(	PUNCT
bracis-28417	336	5	2022	2022	NUM
bracis-28417	336	6	)	)	PUNCT
bracis-28417	336	7	google	google	NOUN
bracis-28417	336	8	scholar	scholar	NOUN
bracis-28417	336	9	  	  	SPACE
bracis-28417	336	10	sakaguchi	sakaguchi	ADJ
bracis-28417	336	11	,	,	PUNCT
bracis-28417	336	12	k.	k.	PROPN
bracis-28417	336	13	,	,	PUNCT
bracis-28417	336	14	bras	bras	PROPN
bracis-28417	336	15	,	,	PUNCT
bracis-28417	336	16	r.l	r.l	PROPN
bracis-28417	336	17	.	.	PROPN
bracis-28417	336	18	,	,	PUNCT
bracis-28417	336	19	bhagavatula	bhagavatula	PROPN
bracis-28417	336	20	,	,	PUNCT
bracis-28417	336	21	c.	c.	NOUN
bracis-28417	336	22	,	,	PUNCT
bracis-28417	336	23	choi	choi	NOUN
bracis-28417	336	24	,	,	PUNCT
bracis-28417	336	25	y.	y.	NOUN
bracis-28417	336	26	:	:	PUNCT
bracis-28417	336	27	winogrande	winogrande	PROPN
bracis-28417	336	28	:	:	PUNCT
bracis-28417	336	29	an	an	DET
bracis-28417	336	30	adversarial	adversarial	ADJ
bracis-28417	336	31	winograd	winograd	ADJ
bracis-28417	336	32	schema	schema	NOUN
bracis-28417	336	33	challenge	challenge	NOUN
bracis-28417	336	34	at	at	ADP
bracis-28417	336	35	scale	scale	NOUN
bracis-28417	336	36	.	.	PUNCT
bracis-28417	337	1	commun	commun	PROPN
bracis-28417	337	2	.	.	PUNCT
bracis-28417	338	1	acm	acm	PROPN
bracis-28417	338	2	64(9	64(9	NUM
bracis-28417	338	3	)	)	PUNCT
bracis-28417	338	4	,	,	PUNCT
bracis-28417	338	5	99–106	99–106	NUM
bracis-28417	338	6	(	(	PUNCT
bracis-28417	338	7	2021	2021	NUM
bracis-28417	338	8	)	)	PUNCT
bracis-28417	338	9	article	article	NOUN
bracis-28417	338	10	  	  	SPACE
bracis-28417	338	11	google	google	PROPN
bracis-28417	338	12	scholar	scholar	NOUN
bracis-28417	338	13	  	  	SPACE
bracis-28417	338	14	sarti	sarti	NOUN
bracis-28417	338	15	,	,	PUNCT
bracis-28417	338	16	g.	g.	PROPN
bracis-28417	338	17	,	,	PUNCT
bracis-28417	338	18	nissim	nissim	PROPN
bracis-28417	338	19	,	,	PUNCT
bracis-28417	338	20	m.	m.	NOUN
bracis-28417	338	21	:	:	PUNCT
bracis-28417	338	22	it5	it5	NOUN
bracis-28417	338	23	:	:	PUNCT
bracis-28417	338	24	large	large	ADJ
bracis-28417	338	25	-	-	PUNCT
bracis-28417	338	26	scale	scale	NOUN
bracis-28417	338	27	text	text	NOUN
bracis-28417	338	28	-	-	PUNCT
bracis-28417	338	29	to	to	ADP
bracis-28417	338	30	-	-	PUNCT
bracis-28417	338	31	text	text	NOUN
bracis-28417	338	32	pretraining	pretraine	VERB
bracis-28417	338	33	for	for	ADP
bracis-28417	338	34	italian	italian	ADJ
bracis-28417	338	35	language	language	NOUN
bracis-28417	338	36	understanding	understanding	NOUN
bracis-28417	338	37	and	and	CCONJ
bracis-28417	338	38	generation	generation	NOUN
bracis-28417	338	39	.	.	PUNCT
bracis-28417	339	1	arxiv	arxiv	PROPN
bracis-28417	339	2	preprint	preprint	NOUN
bracis-28417	339	3	arxiv:2203.03759	arxiv:2203.03759	PROPN
bracis-28417	339	4	(	(	PUNCT
bracis-28417	339	5	2022	2022	NUM
bracis-28417	339	6	)	)	PUNCT
bracis-28417	339	7	sayama	sayama	NOUN
bracis-28417	339	8	,	,	PUNCT
bracis-28417	339	9	h.f	h.f	PROPN
bracis-28417	339	10	.	.	PROPN
bracis-28417	339	11	,	,	PUNCT
bracis-28417	339	12	araujo	araujo	PROPN
bracis-28417	339	13	,	,	PUNCT
bracis-28417	339	14	a.v	a.v	PROPN
bracis-28417	339	15	.	.	PROPN
bracis-28417	339	16	,	,	PUNCT
bracis-28417	339	17	fernandes	fernandes	PROPN
bracis-28417	339	18	,	,	PUNCT
bracis-28417	339	19	e.r	e.r	PROPN
bracis-28417	339	20	.	.	PROPN
bracis-28417	339	21	:	:	PUNCT
bracis-28417	340	1	faquad	faquad	ADJ
bracis-28417	340	2	:	:	PUNCT
bracis-28417	340	3	reading	read	VERB
bracis-28417	340	4	comprehension	comprehension	NOUN
bracis-28417	340	5	dataset	dataset	VERB
bracis-28417	340	6	in	in	ADP
bracis-28417	340	7	the	the	DET
bracis-28417	340	8	domain	domain	NOUN
bracis-28417	340	9	of	of	ADP
bracis-28417	340	10	brazilian	brazilian	ADJ
bracis-28417	340	11	higher	high	ADJ
bracis-28417	340	12	education	education	NOUN
bracis-28417	340	13	.	.	PUNCT
bracis-28417	341	1	in	in	ADP
bracis-28417	341	2	:	:	PUNCT
bracis-28417	341	3	2019	2019	NUM
bracis-28417	341	4	8th	8th	ADJ
bracis-28417	341	5	brazilian	brazilian	ADJ
bracis-28417	341	6	conference	conference	NOUN
bracis-28417	341	7	on	on	ADP
bracis-28417	341	8	intelligent	intelligent	ADJ
bracis-28417	341	9	systems	system	NOUN
bracis-28417	341	10	(	(	PUNCT
bracis-28417	341	11	bracis	bracis	NOUN
bracis-28417	341	12	)	)	PUNCT
bracis-28417	341	13	,	,	PUNCT
bracis-28417	341	14	pp	pp	ADP
bracis-28417	341	15	.	.	PUNCT
bracis-28417	342	1	443–448	443–448	NUM
bracis-28417	342	2	(	(	PUNCT
bracis-28417	342	3	2019	2019	NUM
bracis-28417	342	4	)	)	PUNCT
bracis-28417	342	5	.	.	PUNCT
bracis-28417	343	1	https://doi.org/10.1109/bracis.2019.00084	https://doi.org/10.1109/bracis.2019.00084	PROPN
bracis-28417	343	2	scao	scao	PROPN
bracis-28417	343	3	,	,	PUNCT
bracis-28417	343	4	t.l	t.l	PROPN
bracis-28417	343	5	.	.	PROPN
bracis-28417	343	6	,	,	PUNCT
bracis-28417	343	7	et	et	PROPN
bracis-28417	343	8	al	al	PROPN
bracis-28417	343	9	.	.	PROPN
bracis-28417	343	10	:	:	PUNCT
bracis-28417	344	1	bloom	bloom	NOUN
bracis-28417	344	2	:	:	PUNCT
bracis-28417	344	3	a	a	DET
bracis-28417	344	4	176b	176b	ADJ
bracis-28417	344	5	-	-	PUNCT
bracis-28417	344	6	parameter	parameter	NOUN
bracis-28417	344	7	open	open	ADJ
bracis-28417	344	8	-	-	PUNCT
bracis-28417	344	9	access	access	NOUN
bracis-28417	344	10	multilingual	multilingual	ADJ
bracis-28417	344	11	language	language	NOUN
bracis-28417	344	12	model	model	NOUN
bracis-28417	344	13	.	.	PUNCT
bracis-28417	345	1	arxiv	arxiv	PROPN
bracis-28417	345	2	preprint	preprint	VERB
bracis-28417	345	3	arxiv:2211.05100	arxiv:2211.05100	NOUN
bracis-28417	345	4	(	(	PUNCT
bracis-28417	345	5	2022	2022	NUM
bracis-28417	345	6	)	)	PUNCT
bracis-28417	345	7	shazeer	shazeer	NOUN
bracis-28417	345	8	,	,	PUNCT
bracis-28417	345	9	n.	n.	NOUN
bracis-28417	345	10	,	,	PUNCT
bracis-28417	345	11	stern	stern	ADJ
bracis-28417	345	12	,	,	PUNCT
bracis-28417	345	13	m.	m.	NOUN
bracis-28417	345	14	:	:	PUNCT
bracis-28417	345	15	adafactor	adafactor	NOUN
bracis-28417	345	16	:	:	PUNCT
bracis-28417	345	17	adaptive	adaptive	ADJ
bracis-28417	345	18	learning	learning	NOUN
bracis-28417	345	19	rates	rate	NOUN
bracis-28417	345	20	with	with	ADP
bracis-28417	345	21	sublinear	sublinear	NOUN
bracis-28417	345	22	memory	memory	NOUN
bracis-28417	345	23	cost	cost	NOUN
bracis-28417	345	24	.	.	PUNCT
bracis-28417	346	1	in	in	ADP
bracis-28417	346	2	:	:	PUNCT
bracis-28417	346	3	international	international	ADJ
bracis-28417	346	4	conference	conference	NOUN
bracis-28417	346	5	on	on	ADP
bracis-28417	346	6	machine	machine	NOUN
bracis-28417	346	7	learning	learning	NOUN
bracis-28417	346	8	,	,	PUNCT
bracis-28417	346	9	pp	pp	ADP
bracis-28417	346	10	.	.	PUNCT
bracis-28417	347	1	4596–4604	4596–4604	NUM
bracis-28417	347	2	.	.	PUNCT
bracis-28417	348	1	pmlr	pmlr	NOUN
bracis-28417	348	2	(	(	PUNCT
bracis-28417	348	3	2018	2018	NUM
bracis-28417	348	4	)	)	PUNCT
bracis-28417	348	5	google	google	PROPN
bracis-28417	348	6	scholar	scholar	NOUN
bracis-28417	348	7	  	  	SPACE
bracis-28417	348	8	shliazhko	shliazhko	NOUN
bracis-28417	348	9	,	,	PUNCT
bracis-28417	348	10	o.	o.	PROPN
bracis-28417	348	11	,	,	PUNCT
bracis-28417	348	12	fenogenova	fenogenova	PROPN
bracis-28417	348	13	,	,	PUNCT
bracis-28417	348	14	a.	a.	PROPN
bracis-28417	348	15	,	,	PUNCT
bracis-28417	348	16	tikhonova	tikhonova	PROPN
bracis-28417	348	17	,	,	PUNCT
bracis-28417	348	18	m.	m.	NOUN
bracis-28417	348	19	,	,	PUNCT
bracis-28417	348	20	mikhailov	mikhailov	PROPN
bracis-28417	348	21	,	,	PUNCT
bracis-28417	348	22	v.	v.	PROPN
bracis-28417	348	23	,	,	PUNCT
bracis-28417	348	24	kozlova	kozlova	PROPN
bracis-28417	348	25	,	,	PUNCT
bracis-28417	348	26	a.	a.	PROPN
bracis-28417	348	27	,	,	PUNCT
bracis-28417	348	28	shavrina	shavrina	NOUN
bracis-28417	348	29	,	,	PUNCT
bracis-28417	348	30	t.	t.	NOUN
bracis-28417	348	31	:	:	PUNCT
bracis-28417	348	32	mgpt	mgpt	NOUN
bracis-28417	348	33	:	:	PUNCT
bracis-28417	348	34	few	few	ADJ
bracis-28417	348	35	-	-	PUNCT
bracis-28417	348	36	shot	shot	NOUN
bracis-28417	348	37	learners	learner	NOUN
bracis-28417	348	38	go	go	VERB
bracis-28417	348	39	multilingual	multilingual	ADJ
bracis-28417	348	40	.	.	PUNCT
bracis-28417	349	1	arxiv	arxiv	PROPN
bracis-28417	349	2	preprint	preprint	VERB
bracis-28417	349	3	arxiv:2204.07580	arxiv:2204.07580	NOUN
bracis-28417	349	4	(	(	PUNCT
bracis-28417	349	5	2022	2022	NUM
bracis-28417	349	6	)	)	PUNCT
bracis-28417	349	7	silveira	silveira	PROPN
bracis-28417	349	8	,	,	PUNCT
bracis-28417	349	9	i.c	i.c	PROPN
bracis-28417	349	10	.	.	PROPN
bracis-28417	349	11	,	,	PUNCT
bracis-28417	349	12	maua	maua	PROPN
bracis-28417	349	13	,	,	PUNCT
bracis-28417	349	14	d.d	d.d	PROPN
bracis-28417	349	15	.	.	PROPN
bracis-28417	349	16	:	:	PUNCT
bracis-28417	349	17	advances	advance	VERB
bracis-28417	349	18	in	in	ADP
bracis-28417	349	19	automatically	automatically	ADV
bracis-28417	349	20	solving	solve	VERB
bracis-28417	349	21	the	the	DET
bracis-28417	349	22	enem	enem	NOUN
bracis-28417	349	23	.	.	PUNCT
bracis-28417	350	1	in	in	ADP
bracis-28417	350	2	:	:	PUNCT
bracis-28417	350	3	2018	2018	NUM
bracis-28417	350	4	7th	7th	ADJ
bracis-28417	350	5	brazilian	brazilian	ADJ
bracis-28417	350	6	conference	conference	NOUN
bracis-28417	350	7	on	on	ADP
bracis-28417	350	8	intelligent	intelligent	ADJ
bracis-28417	350	9	systems	system	NOUN
bracis-28417	350	10	(	(	PUNCT
bracis-28417	350	11	bracis	bracis	NOUN
bracis-28417	350	12	)	)	PUNCT
bracis-28417	350	13	,	,	PUNCT
bracis-28417	350	14	pp	pp	PROPN
bracis-28417	350	15	.	.	PUNCT
bracis-28417	350	16	43–48	43–48	NUM
bracis-28417	350	17	.	.	PUNCT
bracis-28417	350	18	ieee	ieee	NOUN
bracis-28417	350	19	computer	computer	NOUN
bracis-28417	350	20	society	society	NOUN
bracis-28417	350	21	,	,	PUNCT
bracis-28417	350	22	los	los	PROPN
bracis-28417	350	23	alamitos	alamitos	PROPN
bracis-28417	350	24	,	,	PUNCT
bracis-28417	350	25	ca	ca	PROPN
bracis-28417	350	26	,	,	PUNCT
bracis-28417	350	27	usa	usa	PROPN
bracis-28417	350	28	(	(	PUNCT
bracis-28417	350	29	oct	oct	PROPN
bracis-28417	350	30	2018	2018	NUM
bracis-28417	350	31	)	)	PUNCT
bracis-28417	350	32	.	.	PUNCT
bracis-28417	351	1	https://doi.org/10.1109/bracis.2018.00016	https://doi.org/10.1109/bracis.2018.00016	PROPN
bracis-28417	351	2	socher	socher	NOUN
bracis-28417	351	3	,	,	PUNCT
bracis-28417	351	4	r.	r.	PROPN
bracis-28417	351	5	,	,	PUNCT
bracis-28417	351	6	et	et	PROPN
bracis-28417	351	7	al	al	PROPN
bracis-28417	351	8	.	.	PUNCT
bracis-28417	351	9	:	:	PUNCT
bracis-28417	352	1	recursive	recursive	ADJ
bracis-28417	352	2	deep	deep	ADJ
bracis-28417	352	3	models	model	NOUN
bracis-28417	352	4	for	for	ADP
bracis-28417	352	5	semantic	semantic	ADJ
bracis-28417	352	6	compositionality	compositionality	NOUN
bracis-28417	352	7	over	over	ADP
bracis-28417	352	8	a	a	DET
bracis-28417	352	9	sentiment	sentiment	NOUN
bracis-28417	352	10	treebank	treebank	VERB
bracis-28417	352	11	.	.	PUNCT
bracis-28417	353	1	in	in	ADP
bracis-28417	353	2	:	:	PUNCT
bracis-28417	353	3	proceedings	proceeding	NOUN
bracis-28417	353	4	of	of	ADP
bracis-28417	353	5	the	the	DET
bracis-28417	353	6	2013	2013	NUM
bracis-28417	353	7	conference	conference	NOUN
bracis-28417	353	8	on	on	ADP
bracis-28417	353	9	empirical	empirical	ADJ
bracis-28417	353	10	methods	method	NOUN
bracis-28417	353	11	in	in	ADP
bracis-28417	353	12	natural	natural	ADJ
bracis-28417	353	13	language	language	NOUN
bracis-28417	353	14	processing	processing	NOUN
bracis-28417	353	15	,	,	PUNCT
bracis-28417	353	16	pp	pp	ADJ
bracis-28417	353	17	.	.	PUNCT
bracis-28417	354	1	1631–1642	1631–1642	NUM
bracis-28417	354	2	.	.	PUNCT
bracis-28417	355	1	association	association	NOUN
bracis-28417	355	2	for	for	ADP
bracis-28417	355	3	computational	computational	ADJ
bracis-28417	355	4	linguistics	linguistic	NOUN
bracis-28417	355	5	,	,	PUNCT
bracis-28417	355	6	seattle	seattle	PROPN
bracis-28417	355	7	,	,	PUNCT
bracis-28417	355	8	washington	washington	PROPN
bracis-28417	355	9	,	,	PUNCT
bracis-28417	355	10	usa	usa	PROPN
bracis-28417	355	11	(	(	PUNCT
bracis-28417	355	12	2013	2013	NUM
bracis-28417	355	13	)	)	PUNCT
bracis-28417	355	14	google	google	PROPN
bracis-28417	355	15	scholar	scholar	NOUN
bracis-28417	355	16	  	  	SPACE
bracis-28417	355	17	souza	souza	PROPN
bracis-28417	355	18	,	,	PUNCT
bracis-28417	355	19	f.	f.	PROPN
bracis-28417	355	20	,	,	PUNCT
bracis-28417	355	21	nogueira	nogueira	PROPN
bracis-28417	355	22	,	,	PUNCT
bracis-28417	355	23	r.	r.	PROPN
bracis-28417	355	24	,	,	PUNCT
bracis-28417	355	25	lotufo	lotufo	PROPN
bracis-28417	355	26	,	,	PUNCT
bracis-28417	355	27	r.	r.	PROPN
bracis-28417	355	28	:	:	PUNCT
bracis-28417	355	29	bertimbau	bertimbau	PROPN
bracis-28417	355	30	:	:	PUNCT
bracis-28417	355	31	pretrained	pretraine	VERB
bracis-28417	355	32	bert	bert	NOUN
bracis-28417	355	33	models	model	NOUN
bracis-28417	355	34	for	for	ADP
bracis-28417	355	35	brazilian	brazilian	ADJ
bracis-28417	355	36	portuguese	portuguese	NOUN
bracis-28417	355	37	.	.	PUNCT
bracis-28417	356	1	in	in	ADP
bracis-28417	356	2	:	:	PUNCT
bracis-28417	356	3	cerri	cerri	PROPN
bracis-28417	356	4	,	,	PUNCT
bracis-28417	356	5	r.	r.	PROPN
bracis-28417	356	6	,	,	PUNCT
bracis-28417	356	7	prati	prati	PROPN
bracis-28417	356	8	,	,	PUNCT
bracis-28417	356	9	r.c	r.c	PROPN
bracis-28417	356	10	.	.	PROPN
bracis-28417	356	11	(	(	PUNCT
bracis-28417	356	12	eds	ed	NOUN
bracis-28417	356	13	.	.	PUNCT
bracis-28417	356	14	)	)	PUNCT
bracis-28417	356	15	bracis	bracis	PROPN
bracis-28417	356	16	2020	2020	NUM
bracis-28417	356	17	.	.	PUNCT
bracis-28417	357	1	lncs	lncs	PROPN
bracis-28417	357	2	(	(	PUNCT
bracis-28417	357	3	lnai	lnai	ADJ
bracis-28417	357	4	)	)	PUNCT
bracis-28417	357	5	,	,	PUNCT
bracis-28417	357	6	vol	vol	NOUN
bracis-28417	357	7	.	.	PUNCT
bracis-28417	357	8	12319	12319	NUM
bracis-28417	357	9	,	,	PUNCT
bracis-28417	357	10	pp	pp	ADP
bracis-28417	357	11	.	.	PUNCT
bracis-28417	358	1	403–417	403–417	NUM
bracis-28417	358	2	.	.	PUNCT
bracis-28417	358	3	springer	springer	NOUN
bracis-28417	358	4	,	,	PUNCT
bracis-28417	358	5	cham	cham	PROPN
bracis-28417	358	6	(	(	PUNCT
bracis-28417	358	7	2020	2020	NUM
bracis-28417	358	8	)	)	PUNCT
bracis-28417	358	9	.	.	PUNCT
bracis-28417	359	1	https://doi.org/10.1007/978-3-030-61377-8_28	https://doi.org/10.1007/978-3-030-61377-8_28	PROPN
bracis-28417	359	2	chapter	chapter	NOUN
bracis-28417	359	3	  	  	SPACE
bracis-28417	359	4	google	google	PROPN
bracis-28417	359	5	scholar	scholar	NOUN
bracis-28417	359	6	  	  	SPACE
bracis-28417	359	7	srivastava	srivastava	PROPN
bracis-28417	359	8	,	,	PUNCT
bracis-28417	359	9	a.	a.	PROPN
bracis-28417	359	10	,	,	PUNCT
bracis-28417	359	11	et	et	PROPN
bracis-28417	359	12	al	al	PROPN
bracis-28417	359	13	.	.	PUNCT
bracis-28417	359	14	:	:	PUNCT
bracis-28417	360	1	beyond	beyond	ADP
bracis-28417	360	2	the	the	DET
bracis-28417	360	3	imitation	imitation	NOUN
bracis-28417	360	4	game	game	NOUN
bracis-28417	360	5	:	:	PUNCT
bracis-28417	360	6	quantifying	quantify	VERB
bracis-28417	360	7	and	and	CCONJ
bracis-28417	360	8	extrapolating	extrapolate	VERB
bracis-28417	360	9	the	the	DET
bracis-28417	360	10	capabilities	capability	NOUN
bracis-28417	360	11	of	of	ADP
bracis-28417	360	12	language	language	NOUN
bracis-28417	360	13	models	model	NOUN
bracis-28417	360	14	.	.	PUNCT
bracis-28417	361	1	arxiv	arxiv	PROPN
bracis-28417	361	2	preprint	preprint	NOUN
bracis-28417	361	3	arxiv:2206.04615	arxiv:2206.04615	NOUN
bracis-28417	361	4	(	(	PUNCT
bracis-28417	361	5	2022	2022	NUM
bracis-28417	361	6	)	)	PUNCT
bracis-28417	361	7	su	su	PROPN
bracis-28417	361	8	,	,	PUNCT
bracis-28417	361	9	j.	j.	PROPN
bracis-28417	361	10	,	,	PUNCT
bracis-28417	361	11	lu	lu	PROPN
bracis-28417	361	12	,	,	PUNCT
bracis-28417	361	13	y.	y.	PROPN
bracis-28417	361	14	,	,	PUNCT
bracis-28417	361	15	pan	pan	PROPN
bracis-28417	361	16	,	,	PUNCT
bracis-28417	361	17	s.	s.	PROPN
bracis-28417	361	18	,	,	PUNCT
bracis-28417	361	19	wen	wen	PROPN
bracis-28417	361	20	,	,	PUNCT
bracis-28417	361	21	b.	b.	PROPN
bracis-28417	361	22	,	,	PUNCT
bracis-28417	361	23	liu	liu	PROPN
bracis-28417	361	24	,	,	PUNCT
bracis-28417	361	25	y.	y.	NOUN
bracis-28417	361	26	:	:	PUNCT
bracis-28417	362	1	roformer	roformer	NOUN
bracis-28417	362	2	:	:	PUNCT
bracis-28417	362	3	enhanced	enhanced	ADJ
bracis-28417	362	4	transformer	transformer	NOUN
bracis-28417	362	5	with	with	ADP
bracis-28417	362	6	rotary	rotary	ADJ
bracis-28417	362	7	position	position	NOUN
bracis-28417	362	8	embedding	embed	VERB
bracis-28417	362	9	.	.	PUNCT
bracis-28417	363	1	arxiv	arxiv	PROPN
bracis-28417	363	2	preprint	preprint	VERB
bracis-28417	363	3	arxiv:2104.09864	arxiv:2104.09864	NOUN
bracis-28417	363	4	(	(	PUNCT
bracis-28417	363	5	2021	2021	NUM
bracis-28417	363	6	)	)	PUNCT
bracis-28417	363	7	taylor	taylor	PROPN
bracis-28417	363	8	,	,	PUNCT
bracis-28417	363	9	r.	r.	PROPN
bracis-28417	363	10	et	et	PROPN
bracis-28417	363	11	al	al	PROPN
bracis-28417	363	12	.	.	PROPN
bracis-28417	363	13	:	:	PUNCT
bracis-28417	364	1	galactica	galactica	NOUN
bracis-28417	364	2	:	:	PUNCT
bracis-28417	364	3	a	a	DET
bracis-28417	364	4	large	large	ADJ
bracis-28417	364	5	language	language	NOUN
bracis-28417	364	6	model	model	NOUN
bracis-28417	364	7	for	for	ADP
bracis-28417	364	8	science	science	NOUN
bracis-28417	364	9	.	.	PUNCT
bracis-28417	365	1	arxiv	arxiv	PROPN
bracis-28417	365	2	preprint	preprint	VERB
bracis-28417	365	3	arxiv:2211.09085	arxiv:2211.09085	NUM
bracis-28417	365	4	(	(	PUNCT
bracis-28417	365	5	2022	2022	NUM
bracis-28417	365	6	)	)	PUNCT
bracis-28417	365	7	touvron	touvron	PROPN
bracis-28417	365	8	,	,	PUNCT
bracis-28417	365	9	h.	h.	PROPN
bracis-28417	365	10	,	,	PUNCT
bracis-28417	365	11	et	et	PROPN
bracis-28417	365	12	al	al	PROPN
bracis-28417	365	13	.	.	PUNCT
bracis-28417	365	14	:	:	PUNCT
bracis-28417	366	1	llama	llama	NOUN
bracis-28417	366	2	:	:	PUNCT
bracis-28417	366	3	open	open	ADJ
bracis-28417	366	4	and	and	CCONJ
bracis-28417	366	5	efficient	efficient	ADJ
bracis-28417	366	6	foundation	foundation	NOUN
bracis-28417	366	7	language	language	NOUN
bracis-28417	366	8	models	model	NOUN
bracis-28417	366	9	.	.	PUNCT
bracis-28417	367	1	arxiv	arxiv	PROPN
bracis-28417	367	2	preprint	preprint	PROPN
bracis-28417	367	3	arxiv:2302.13971	arxiv:2302.13971	PROPN
bracis-28417	367	4	(	(	PUNCT
bracis-28417	367	5	2023	2023	NUM
bracis-28417	367	6	)	)	PUNCT
bracis-28417	367	7	vaswani	vaswani	NOUN
bracis-28417	367	8	,	,	PUNCT
bracis-28417	367	9	a.	a.	NOUN
bracis-28417	367	10	,	,	PUNCT
bracis-28417	367	11	et	et	PROPN
bracis-28417	367	12	al	al	PROPN
bracis-28417	367	13	.	.	PUNCT
bracis-28417	367	14	:	:	PUNCT
bracis-28417	368	1	attention	attention	NOUN
bracis-28417	368	2	is	be	AUX
bracis-28417	368	3	all	all	PRON
bracis-28417	368	4	you	you	PRON
bracis-28417	368	5	need	need	VERB
bracis-28417	368	6	.	.	PUNCT
bracis-28417	369	1	in	in	ADP
bracis-28417	369	2	:	:	PUNCT
bracis-28417	369	3	advances	advance	NOUN
bracis-28417	369	4	in	in	ADP
bracis-28417	369	5	neural	neural	ADJ
bracis-28417	369	6	information	information	NOUN
bracis-28417	369	7	processing	processing	NOUN
bracis-28417	369	8	systems	system	NOUN
bracis-28417	369	9	30	30	NUM
bracis-28417	369	10	(	(	PUNCT
bracis-28417	369	11	2017	2017	NUM
bracis-28417	369	12	)	)	PUNCT
bracis-28417	369	13	google	google	PROPN
bracis-28417	369	14	scholar	scholar	NOUN
bracis-28417	369	15	  	  	SPACE
bracis-28417	369	16	wang	wang	PROPN
bracis-28417	369	17	,	,	PUNCT
bracis-28417	369	18	b.	b.	PROPN
bracis-28417	369	19	:	:	PUNCT
bracis-28417	369	20	mesh	mesh	NOUN
bracis-28417	369	21	-	-	PUNCT
bracis-28417	369	22	transformer	transformer	NOUN
bracis-28417	369	23	-	-	PUNCT
bracis-28417	369	24	jax	jax	NOUN
bracis-28417	369	25	:	:	PUNCT
bracis-28417	369	26	model	model	ADJ
bracis-28417	369	27	-	-	PUNCT
bracis-28417	369	28	parallel	parallel	ADJ
bracis-28417	369	29	implementation	implementation	NOUN
bracis-28417	369	30	of	of	ADP
bracis-28417	369	31	transformer	transformer	ADJ
bracis-28417	369	32	language	language	NOUN
bracis-28417	369	33	model	model	NOUN
bracis-28417	369	34	with	with	ADP
bracis-28417	369	35	jax	jax	PROPN
bracis-28417	369	36	.	.	PUNCT
bracis-28417	370	1	https://github.com/kingoflolz/mesh-transformer-jax	https://github.com/kingoflolz/mesh-transformer-jax	PROPN
bracis-28417	370	2	(	(	PUNCT
bracis-28417	370	3	2021	2021	NUM
bracis-28417	370	4	)	)	PUNCT
bracis-28417	370	5	wang	wang	PROPN
bracis-28417	370	6	,	,	PUNCT
bracis-28417	370	7	b.	b.	PROPN
bracis-28417	370	8	,	,	PUNCT
bracis-28417	370	9	komatsuzaki	komatsuzaki	NOUN
bracis-28417	370	10	,	,	PUNCT
bracis-28417	370	11	a.	a.	NOUN
bracis-28417	370	12	:	:	PUNCT
bracis-28417	370	13	gpt	gpt	PROPN
bracis-28417	370	14	-	-	PUNCT
bracis-28417	370	15	j-6b	j-6b	PROPN
bracis-28417	370	16	:	:	PUNCT
bracis-28417	370	17	a	a	DET
bracis-28417	370	18	6	6	NUM
bracis-28417	370	19	billion	billion	NUM
bracis-28417	370	20	parameter	parameter	NOUN
bracis-28417	370	21	autoregressive	autoregressive	ADJ
bracis-28417	370	22	language	language	NOUN
bracis-28417	370	23	model	model	NOUN
bracis-28417	370	24	(	(	PUNCT
bracis-28417	370	25	2021	2021	NUM
bracis-28417	370	26	)	)	PUNCT
bracis-28417	370	27	google	google	PROPN
bracis-28417	370	28	scholar	scholar	NOUN
bracis-28417	370	29	  	  	SPACE
bracis-28417	370	30	wei	wei	PROPN
bracis-28417	370	31	,	,	PUNCT
bracis-28417	370	32	j.	j.	PROPN
bracis-28417	370	33	,	,	PUNCT
bracis-28417	370	34	et	et	PROPN
bracis-28417	370	35	al	al	PROPN
bracis-28417	370	36	.	.	PUNCT
bracis-28417	370	37	:	:	PUNCT
bracis-28417	370	38	emergent	emergent	ADJ
bracis-28417	370	39	abilities	ability	NOUN
bracis-28417	370	40	of	of	ADP
bracis-28417	370	41	large	large	ADJ
bracis-28417	370	42	language	language	NOUN
bracis-28417	370	43	models	model	NOUN
bracis-28417	370	44	.	.	PUNCT
bracis-28417	371	1	transactions	transaction	NOUN
bracis-28417	371	2	on	on	ADP
bracis-28417	371	3	machine	machine	NOUN
bracis-28417	371	4	learning	learn	VERB
bracis-28417	371	5	research	research	NOUN
bracis-28417	371	6	(	(	PUNCT
bracis-28417	371	7	2022	2022	NUM
bracis-28417	371	8	)	)	PUNCT
bracis-28417	371	9	,	,	PUNCT
bracis-28417	371	10	survey	survey	NOUN
bracis-28417	371	11	certification	certification	NOUN
bracis-28417	371	12	google	google	PROPN
bracis-28417	371	13	scholar	scholar	PROPN
bracis-28417	371	14	  	  	SPACE
bracis-28417	371	15	wu	wu	PROPN
bracis-28417	371	16	,	,	PUNCT
bracis-28417	371	17	s.	s.	PROPN
bracis-28417	371	18	,	,	PUNCT
bracis-28417	371	19	et	et	PROPN
bracis-28417	371	20	al	al	PROPN
bracis-28417	371	21	.	.	PUNCT
bracis-28417	371	22	:	:	PUNCT
bracis-28417	372	1	bloomberggpt	bloomberggpt	NOUN
bracis-28417	372	2	:	:	PUNCT
bracis-28417	372	3	a	a	DET
bracis-28417	372	4	large	large	ADJ
bracis-28417	372	5	language	language	NOUN
bracis-28417	372	6	model	model	NOUN
bracis-28417	372	7	for	for	ADP
bracis-28417	372	8	finance	finance	NOUN
bracis-28417	372	9	(	(	PUNCT
bracis-28417	372	10	2023	2023	NUM
bracis-28417	372	11	)	)	PUNCT
bracis-28417	372	12	google	google	NOUN
bracis-28417	372	13	scholar	scholar	NOUN
bracis-28417	372	14	  	  	SPACE
bracis-28417	372	15	xue	xue	PROPN
bracis-28417	372	16	,	,	PUNCT
bracis-28417	372	17	l.	l.	PROPN
bracis-28417	372	18	,	,	PUNCT
bracis-28417	372	19	et	et	PROPN
bracis-28417	372	20	al	al	PROPN
bracis-28417	372	21	.	.	PUNCT
bracis-28417	372	22	:	:	PUNCT
bracis-28417	373	1	byt5	byt5	PROPN
bracis-28417	373	2	:	:	PUNCT
bracis-28417	373	3	towards	towards	ADP
bracis-28417	373	4	a	a	DET
bracis-28417	373	5	token	token	ADJ
bracis-28417	373	6	-	-	PUNCT
bracis-28417	373	7	free	free	ADJ
bracis-28417	373	8	future	future	NOUN
bracis-28417	373	9	with	with	ADP
bracis-28417	373	10	pre	pre	ADJ
bracis-28417	373	11	-	-	ADJ
bracis-28417	373	12	trained	train	VERB
bracis-28417	373	13	byte	byte	NOUN
bracis-28417	373	14	-	-	PUNCT
bracis-28417	373	15	to	to	ADP
bracis-28417	373	16	-	-	PUNCT
bracis-28417	373	17	byte	byte	NOUN
bracis-28417	373	18	models	model	NOUN
bracis-28417	373	19	.	.	PUNCT
bracis-28417	374	1	trans	trans	PROPN
bracis-28417	374	2	.	.	PUNCT
bracis-28417	375	1	assoc	assoc	PROPN
bracis-28417	375	2	.	.	PUNCT
bracis-28417	376	1	comput	comput	PROPN
bracis-28417	376	2	.	.	PUNCT
bracis-28417	377	1	linguist	linguist	NOUN
bracis-28417	377	2	.	.	PUNCT
bracis-28417	378	1	10	10	NUM
bracis-28417	378	2	,	,	PUNCT
bracis-28417	378	3	291–306	291–306	NUM
bracis-28417	378	4	(	(	PUNCT
bracis-28417	378	5	2022	2022	NUM
bracis-28417	378	6	)	)	PUNCT
bracis-28417	378	7	article	article	NOUN
bracis-28417	378	8	  	  	SPACE
bracis-28417	378	9	google	google	PROPN
bracis-28417	378	10	scholar	scholar	NOUN
bracis-28417	378	11	  	  	SPACE
bracis-28417	378	12	xue	xue	PROPN
bracis-28417	378	13	,	,	PUNCT
bracis-28417	378	14	l.	l.	PROPN
bracis-28417	378	15	,	,	PUNCT
bracis-28417	378	16	et	et	PROPN
bracis-28417	378	17	al	al	PROPN
bracis-28417	378	18	.	.	PUNCT
bracis-28417	378	19	:	:	PUNCT
bracis-28417	379	1	mt5	mt5	ADJ
bracis-28417	379	2	:	:	PUNCT
bracis-28417	379	3	a	a	DET
bracis-28417	379	4	massively	massively	ADV
bracis-28417	379	5	multilingual	multilingual	ADJ
bracis-28417	379	6	pre	pre	ADJ
bracis-28417	379	7	-	-	ADJ
bracis-28417	379	8	trained	train	VERB
bracis-28417	379	9	text	text	NOUN
bracis-28417	379	10	-	-	PUNCT
bracis-28417	379	11	to	to	ADP
bracis-28417	379	12	-	-	PUNCT
bracis-28417	379	13	text	text	NOUN
bracis-28417	379	14	transformer	transformer	NOUN
bracis-28417	379	15	.	.	PUNCT
bracis-28417	380	1	arxiv	arxiv	PROPN
bracis-28417	380	2	preprint	preprint	VERB
bracis-28417	380	3	arxiv:2010.11934	arxiv:2010.11934	NOUN
bracis-28417	380	4	(	(	PUNCT
bracis-28417	380	5	2020	2020	NUM
bracis-28417	380	6	)	)	PUNCT
bracis-28417	380	7	yang	yang	PROPN
bracis-28417	380	8	,	,	PUNCT
bracis-28417	380	9	z.	z.	PROPN
bracis-28417	380	10	,	,	PUNCT
bracis-28417	380	11	dai	dai	PROPN
bracis-28417	380	12	,	,	PUNCT
bracis-28417	380	13	z.	z.	PROPN
bracis-28417	380	14	,	,	PUNCT
bracis-28417	380	15	yang	yang	PROPN
bracis-28417	380	16	,	,	PUNCT
bracis-28417	380	17	y.	y.	PROPN
bracis-28417	380	18	,	,	PUNCT
bracis-28417	380	19	carbonell	carbonell	PROPN
bracis-28417	380	20	,	,	PUNCT
bracis-28417	380	21	j.	j.	PROPN
bracis-28417	380	22	,	,	PUNCT
bracis-28417	380	23	salakhutdinov	salakhutdinov	PROPN
bracis-28417	380	24	,	,	PUNCT
bracis-28417	380	25	r.	r.	PROPN
bracis-28417	380	26	,	,	PUNCT
bracis-28417	380	27	le	le	X
bracis-28417	380	28	,	,	PUNCT
bracis-28417	380	29	q.v	q.v	PROPN
bracis-28417	380	30	.	.	PROPN
bracis-28417	380	31	:	:	PUNCT
bracis-28417	381	1	xlnet	xlnet	PROPN
bracis-28417	381	2	:	:	PUNCT
bracis-28417	381	3	generalized	generalize	VERB
bracis-28417	381	4	autoregressive	autoregressive	ADJ
bracis-28417	381	5	pretraining	pretraine	VERB
bracis-28417	381	6	for	for	ADP
bracis-28417	381	7	language	language	NOUN
bracis-28417	381	8	understanding	understanding	NOUN
bracis-28417	381	9	.	.	PUNCT
bracis-28417	382	1	curran	curran	PROPN
bracis-28417	382	2	associates	associates	PROPN
bracis-28417	382	3	inc	inc	PROPN
bracis-28417	382	4	.	.	PROPN
bracis-28417	382	5	,	,	PUNCT
bracis-28417	382	6	red	red	ADJ
bracis-28417	382	7	hook	hook	NOUN
bracis-28417	382	8	,	,	PUNCT
bracis-28417	382	9	ny	ny	PROPN
bracis-28417	382	10	,	,	PUNCT
bracis-28417	382	11	usa	usa	PROPN
bracis-28417	382	12	(	(	PUNCT
bracis-28417	382	13	2019	2019	NUM
bracis-28417	382	14	)	)	PUNCT
bracis-28417	382	15	google	google	PROPN
bracis-28417	382	16	scholar	scholar	NOUN
bracis-28417	382	17	  	  	SPACE
bracis-28417	382	18	yong	yong	PROPN
bracis-28417	382	19	,	,	PUNCT
bracis-28417	382	20	z.x	z.x	PROPN
bracis-28417	382	21	.	.	PROPN
bracis-28417	382	22	,	,	PUNCT
bracis-28417	382	23	et	et	PROPN
bracis-28417	382	24	al	al	PROPN
bracis-28417	382	25	.	.	PUNCT
bracis-28417	382	26	:	:	PUNCT
bracis-28417	383	1	bloom+	bloom+	X
bracis-28417	383	2	1	1	NUM
bracis-28417	383	3	:	:	PUNCT
bracis-28417	383	4	adding	add	VERB
bracis-28417	383	5	language	language	NOUN
bracis-28417	383	6	support	support	NOUN
bracis-28417	383	7	to	to	PART
bracis-28417	383	8	bloom	bloom	VERB
bracis-28417	383	9	for	for	ADP
bracis-28417	383	10	zero	zero	NUM
bracis-28417	383	11	-	-	PUNCT
bracis-28417	383	12	shot	shot	NOUN
bracis-28417	383	13	prompting	prompting	NOUN
bracis-28417	383	14	.	.	PUNCT
bracis-28417	384	1	arxiv	arxiv	PROPN
bracis-28417	384	2	preprint	preprint	VERB
bracis-28417	384	3	arxiv:2212.09535	arxiv:2212.09535	PROPN
bracis-28417	384	4	(	(	PUNCT
bracis-28417	384	5	2022	2022	NUM
bracis-28417	384	6	)	)	PUNCT
bracis-28417	384	7	zeng	zeng	PROPN
bracis-28417	384	8	,	,	PUNCT
bracis-28417	384	9	a.	a.	PROPN
bracis-28417	384	10	,	,	PUNCT
bracis-28417	384	11	et	et	PROPN
bracis-28417	384	12	al	al	PROPN
bracis-28417	384	13	.	.	PROPN
bracis-28417	384	14	:	:	PUNCT
bracis-28417	385	1	glm-130b	glm-130b	ADV
bracis-28417	385	2	:	:	PUNCT
bracis-28417	385	3	an	an	DET
bracis-28417	385	4	open	open	ADJ
bracis-28417	385	5	bilingual	bilingual	ADJ
bracis-28417	385	6	pre	pre	ADJ
bracis-28417	385	7	-	-	ADJ
bracis-28417	385	8	trained	train	VERB
bracis-28417	385	9	model	model	NOUN
bracis-28417	385	10	.	.	PUNCT
bracis-28417	386	1	arxiv	arxiv	PROPN
bracis-28417	386	2	preprint	preprint	NOUN
bracis-28417	386	3	arxiv:2210.02414	arxiv:2210.02414	NUM
bracis-28417	386	4	(	(	PUNCT
bracis-28417	386	5	2022	2022	NUM
bracis-28417	386	6	)	)	PUNCT
bracis-28417	386	7	zhang	zhang	PROPN
bracis-28417	386	8	,	,	PUNCT
bracis-28417	386	9	s.	s.	PROPN
bracis-28417	386	10	,	,	PUNCT
bracis-28417	386	11	et	et	PROPN
bracis-28417	386	12	al	al	PROPN
bracis-28417	386	13	.	.	PROPN
bracis-28417	386	14	:	:	PUNCT
bracis-28417	387	1	opt	opt	PROPN
bracis-28417	387	2	:	:	PUNCT
bracis-28417	387	3	open	open	ADJ
bracis-28417	387	4	pre	pre	ADJ
bracis-28417	387	5	-	-	ADJ
bracis-28417	387	6	trained	train	VERB
bracis-28417	387	7	transformer	transformer	NOUN
bracis-28417	387	8	language	language	NOUN
bracis-28417	387	9	models	model	NOUN
bracis-28417	387	10	.	.	PUNCT
bracis-28417	388	1	arxiv	arxiv	PROPN
bracis-28417	388	2	preprint	preprint	NOUN
bracis-28417	388	3	arxiv:2205.01068	arxiv:2205.01068	PROPN
bracis-28417	388	4	(	(	PUNCT
bracis-28417	388	5	2022	2022	NUM
bracis-28417	388	6	)	)	PUNCT
bracis-28417	388	7	zhang	zhang	PROPN
bracis-28417	388	8	,	,	PUNCT
bracis-28417	388	9	x.	x.	PROPN
bracis-28417	388	10	,	,	PUNCT
bracis-28417	388	11	zhao	zhao	PROPN
bracis-28417	388	12	,	,	PUNCT
bracis-28417	388	13	j.j	j.j	PROPN
bracis-28417	388	14	.	.	PROPN
bracis-28417	388	15	,	,	PUNCT
bracis-28417	388	16	lecun	lecun	ADJ
bracis-28417	388	17	,	,	PUNCT
bracis-28417	388	18	y.	y.	NOUN
bracis-28417	388	19	:	:	PUNCT
bracis-28417	388	20	character	character	NOUN
bracis-28417	388	21	-	-	PUNCT
bracis-28417	388	22	level	level	NOUN
bracis-28417	388	23	convolutional	convolutional	ADJ
bracis-28417	388	24	networks	network	NOUN
bracis-28417	388	25	for	for	ADP
bracis-28417	388	26	text	text	NOUN
bracis-28417	388	27	classification	classification	NOUN
bracis-28417	388	28	.	.	PUNCT
bracis-28417	389	1	in	in	ADP
bracis-28417	389	2	:	:	PUNCT
bracis-28417	389	3	nips	nip	NOUN
bracis-28417	389	4	(	(	PUNCT
bracis-28417	389	5	2015	2015	NUM
bracis-28417	389	6	)	)	PUNCT
bracis-28417	389	7	google	google	PROPN
bracis-28417	389	8	scholar	scholar	NOUN
bracis-28417	389	9	  	  	SPACE
bracis-28417	389	10	zoph	zoph	NOUN
bracis-28417	389	11	,	,	PUNCT
bracis-28417	389	12	b.	b.	PROPN
bracis-28417	389	13	:	:	PUNCT
bracis-28417	389	14	designing	design	VERB
bracis-28417	389	15	effective	effective	ADJ
bracis-28417	389	16	sparse	sparse	ADJ
bracis-28417	389	17	expert	expert	NOUN
bracis-28417	389	18	models	model	NOUN
bracis-28417	389	19	.	.	PUNCT
bracis-28417	390	1	in	in	ADP
bracis-28417	390	2	:	:	PUNCT
bracis-28417	390	3	2022	2022	NUM
bracis-28417	390	4	ieee	ieee	NOUN
bracis-28417	390	5	international	international	PROPN
bracis-28417	390	6	parallel	parallel	ADJ
bracis-28417	390	7	and	and	CCONJ
bracis-28417	390	8	distributed	distributed	ADJ
bracis-28417	390	9	processing	processing	NOUN
bracis-28417	390	10	symposium	symposium	NOUN
bracis-28417	390	11	workshops	workshop	NOUN
bracis-28417	390	12	(	(	PUNCT
bracis-28417	390	13	ipdpsw	ipdpsw	NOUN
bracis-28417	390	14	)	)	PUNCT
bracis-28417	390	15	,	,	PUNCT
bracis-28417	390	16	p.	p.	NOUN
bracis-28417	390	17	1044	1044	NUM
bracis-28417	390	18	.	.	PUNCT
bracis-28417	391	1	ieee	ieee	NOUN
bracis-28417	391	2	(	(	PUNCT
bracis-28417	391	3	2022	2022	NUM
bracis-28417	391	4	)	)	PUNCT
bracis-28417	391	5	google	google	PROPN
bracis-28417	391	6	scholar	scholar	NOUN
bracis-28417	391	7	  	  	SPACE
bracis-28417	391	8	download	download	NOUN
bracis-28417	391	9	references	reference	NOUN
bracis-28417	391	10	acknowledgments	acknowledgment	NOUN
bracis-28417	391	11	we	we	PRON
bracis-28417	391	12	thank	thank	VERB
bracis-28417	391	13	google	google	PROPN
bracis-28417	391	14	cloud	cloud	PROPN
bracis-28417	391	15	for	for	ADP
bracis-28417	391	16	the	the	DET
bracis-28417	391	17	generous	generous	ADJ
bracis-28417	391	18	tpu	tpu	PROPN
bracis-28417	391	19	grant	grant	PROPN
bracis-28417	391	20	.	.	PUNCT
bracis-28417	392	1	author	author	NOUN
bracis-28417	392	2	information	information	NOUN
bracis-28417	392	3	authors	author	NOUN
bracis-28417	392	4	and	and	CCONJ
bracis-28417	392	5	affiliations	affiliation	NOUN
bracis-28417	392	6	maritaca	maritaca	PROPN
bracis-28417	392	7	ai	ai	VERB
bracis-28417	392	8	,	,	PUNCT
bracis-28417	392	9	campinas	campinas	PROPN
bracis-28417	392	10	,	,	PUNCT
bracis-28417	392	11	brazil	brazil	PROPN
bracis-28417	392	12	ramon	ramon	NOUN
bracis-28417	392	13	pires	pire	NOUN
bracis-28417	392	14	,	,	PUNCT
bracis-28417	392	15	 	 	SPACE
bracis-28417	392	16	hugo	hugo	PROPN
bracis-28417	392	17	abonizio	abonizio	PROPN
bracis-28417	392	18	,	,	PUNCT
bracis-28417	392	19	 	 	SPACE
bracis-28417	392	20	thales	thales	PROPN
bracis-28417	392	21	sales	sales	PROPN
bracis-28417	392	22	almeida	almeida	PROPN
bracis-28417	392	23	 	 	SPACE
bracis-28417	392	24	&	&	CCONJ
bracis-28417	392	25	 	 	SPACE
bracis-28417	392	26	rodrigo	rodrigo	PROPN
bracis-28417	392	27	nogueira	nogueira	PROPN
bracis-28417	392	28	authors	author	NOUN
bracis-28417	392	29	ramon	ramon	NOUN
bracis-28417	392	30	piresview	piresview	NOUN
bracis-28417	392	31	author	author	NOUN
bracis-28417	392	32	publications	publication	NOUN
bracis-28417	392	33	search	search	NOUN
bracis-28417	392	34	author	author	NOUN
bracis-28417	392	35	on	on	ADP
bracis-28417	392	36	:	:	PUNCT
bracis-28417	392	37	pubmed	pubmed	PROPN
bracis-28417	392	38	 	 	SPACE
bracis-28417	392	39	google	google	PROPN
bracis-28417	392	40	scholar	scholar	PROPN
bracis-28417	392	41	hugo	hugo	PROPN
bracis-28417	392	42	abonizioview	abonizioview	PROPN
bracis-28417	392	43	author	author	NOUN
bracis-28417	392	44	publications	publication	NOUN
bracis-28417	392	45	search	search	NOUN
bracis-28417	392	46	author	author	NOUN
bracis-28417	392	47	on	on	ADP
bracis-28417	392	48	:	:	PUNCT
bracis-28417	392	49	pubmed	pubmed	PROPN
bracis-28417	392	50	 	 	SPACE
bracis-28417	392	51	google	google	PROPN
bracis-28417	392	52	scholar	scholar	NOUN
bracis-28417	392	53	thales	thales	PROPN
bracis-28417	392	54	sales	sale	NOUN
bracis-28417	392	55	almeidaview	almeidaview	NOUN
bracis-28417	392	56	author	author	NOUN
bracis-28417	392	57	publications	publication	VERB
bracis-28417	392	58	search	search	NOUN
bracis-28417	392	59	author	author	NOUN
bracis-28417	392	60	on	on	ADP
bracis-28417	392	61	:	:	PUNCT
bracis-28417	392	62	pubmed	pubmed	PROPN
bracis-28417	392	63	 	 	SPACE
bracis-28417	392	64	google	google	PROPN
bracis-28417	392	65	scholar	scholar	PROPN
bracis-28417	392	66	rodrigo	rodrigo	PROPN
bracis-28417	392	67	nogueiraview	nogueiraview	PROPN
bracis-28417	392	68	author	author	NOUN
bracis-28417	392	69	publications	publication	VERB
bracis-28417	392	70	search	search	NOUN
bracis-28417	392	71	author	author	NOUN
bracis-28417	392	72	on	on	ADP
bracis-28417	392	73	:	:	PUNCT
bracis-28417	392	74	pubmed	pubmed	PROPN
bracis-28417	392	75	 	 	SPACE
bracis-28417	392	76	google	google	PROPN
bracis-28417	392	77	scholar	scholar	NOUN
bracis-28417	392	78	corresponding	correspond	VERB
bracis-28417	392	79	author	author	NOUN
bracis-28417	392	80	correspondence	correspondence	NOUN
bracis-28417	392	81	to	to	ADP
bracis-28417	392	82	ramon	ramon	NOUN
bracis-28417	392	83	pires	pire	NOUN
bracis-28417	392	84	.	.	PUNCT
bracis-28417	393	1	editor	editor	NOUN
bracis-28417	393	2	information	information	NOUN
bracis-28417	393	3	editors	editor	NOUN
bracis-28417	393	4	and	and	CCONJ
bracis-28417	393	5	affiliations	affiliation	NOUN
bracis-28417	393	6	federal	federal	PROPN
bracis-28417	393	7	university	university	PROPN
bracis-28417	393	8	of	of	ADP
bracis-28417	393	9	são	são	PROPN
bracis-28417	393	10	carlos	carlos	PROPN
bracis-28417	393	11	,	,	PUNCT
bracis-28417	393	12	são	são	PROPN
bracis-28417	393	13	carlos	carlos	PROPN
bracis-28417	393	14	,	,	PUNCT
bracis-28417	393	15	brazil	brazil	PROPN
bracis-28417	393	16	murilo	murilo	PROPN
bracis-28417	393	17	c.	c.	PROPN
bracis-28417	393	18	naldi	naldi	PROPN
bracis-28417	393	19	centro	centro	PROPN
bracis-28417	393	20	universitario	universitario	PROPN
bracis-28417	393	21	da	da	PROPN
bracis-28417	393	22	fei	fei	PROPN
bracis-28417	393	23	,	,	PUNCT
bracis-28417	393	24	são	são	PROPN
bracis-28417	393	25	bernardo	bernardo	PROPN
bracis-28417	393	26	do	do	AUX
bracis-28417	393	27	campo	campo	PROPN
bracis-28417	393	28	,	,	PUNCT
bracis-28417	393	29	brazil	brazil	PROPN
bracis-28417	393	30	reinaldo	reinaldo	PROPN
bracis-28417	393	31	a.	a.	PROPN
bracis-28417	393	32	c.	c.	PROPN
bracis-28417	393	33	bianchi	bianchi	PROPN
bracis-28417	393	34	rights	right	NOUN
bracis-28417	393	35	and	and	CCONJ
bracis-28417	393	36	permissions	permission	NOUN
bracis-28417	393	37	reprints	reprint	NOUN
bracis-28417	393	38	and	and	CCONJ
bracis-28417	393	39	permissions	permission	VERB
bracis-28417	393	40	copyright	copyright	NOUN
bracis-28417	393	41	information	information	NOUN
bracis-28417	393	42	©	©	ADP
bracis-28417	393	43	2023	2023	NUM
bracis-28417	393	44	the	the	DET
bracis-28417	393	45	author(s	author(s	NOUN
bracis-28417	393	46	)	)	PUNCT
bracis-28417	393	47	,	,	PUNCT
bracis-28417	393	48	under	under	ADP
bracis-28417	393	49	exclusive	exclusive	ADJ
bracis-28417	393	50	license	license	NOUN
bracis-28417	393	51	to	to	ADP
bracis-28417	393	52	springer	springer	NOUN
bracis-28417	393	53	nature	nature	PROPN
bracis-28417	393	54	switzerland	switzerland	PROPN
bracis-28417	393	55	ag	ag	PROPN
bracis-28417	393	56	about	about	ADP
bracis-28417	393	57	this	this	DET
bracis-28417	393	58	paper	paper	NOUN
bracis-28417	393	59	cite	cite	VERB
bracis-28417	393	60	this	this	DET
bracis-28417	393	61	paper	paper	NOUN
bracis-28417	393	62	pires	pire	NOUN
bracis-28417	393	63	,	,	PUNCT
bracis-28417	393	64	r.	r.	PROPN
bracis-28417	393	65	,	,	PUNCT
bracis-28417	393	66	abonizio	abonizio	PROPN
bracis-28417	393	67	,	,	PUNCT
bracis-28417	393	68	h.	h.	PROPN
bracis-28417	393	69	,	,	PUNCT
bracis-28417	393	70	almeida	almeida	PROPN
bracis-28417	393	71	,	,	PUNCT
bracis-28417	393	72	t.s	t.s	PROPN
bracis-28417	393	73	.	.	PROPN
bracis-28417	393	74	,	,	PUNCT
bracis-28417	393	75	nogueira	nogueira	PROPN
bracis-28417	393	76	,	,	PUNCT
bracis-28417	393	77	r.	r.	PROPN
bracis-28417	393	78	(	(	PUNCT
bracis-28417	393	79	2023	2023	NUM
bracis-28417	393	80	)	)	PUNCT
bracis-28417	393	81	.	.	PUNCT
bracis-28417	394	1	sabiá	sabiá	PROPN
bracis-28417	394	2	:	:	PUNCT
bracis-28417	395	1	portuguese	portuguese	VERB
bracis-28417	395	2	large	large	ADJ
bracis-28417	395	3	language	language	NOUN
bracis-28417	395	4	models	model	NOUN
bracis-28417	395	5	.	.	PUNCT
bracis-28417	396	1	in	in	ADP
bracis-28417	396	2	:	:	PUNCT
bracis-28417	396	3	naldi	naldi	PROPN
bracis-28417	396	4	,	,	PUNCT
bracis-28417	396	5	m.c	m.c	PROPN
bracis-28417	396	6	.	.	PROPN
bracis-28417	396	7	,	,	PUNCT
bracis-28417	396	8	bianchi	bianchi	PROPN
bracis-28417	396	9	,	,	PUNCT
bracis-28417	396	10	r.a.c	r.a.c	ADP
bracis-28417	396	11	.	.	PUNCT
bracis-28417	396	12	(	(	PUNCT
bracis-28417	396	13	eds	ed	NOUN
bracis-28417	396	14	)	)	PUNCT
bracis-28417	396	15	intelligent	intelligent	ADJ
bracis-28417	396	16	systems	system	NOUN
bracis-28417	396	17	.	.	PUNCT
bracis-28417	397	1	bracis	bracis	PROPN
bracis-28417	397	2	2023	2023	NUM
bracis-28417	397	3	.	.	PUNCT
bracis-28417	398	1	lecture	lecture	NOUN
bracis-28417	398	2	notes	note	NOUN
bracis-28417	398	3	in	in	ADP
bracis-28417	398	4	computer	computer	NOUN
bracis-28417	398	5	science	science	NOUN
bracis-28417	398	6	(	(	PUNCT
bracis-28417	398	7	)	)	PUNCT
bracis-28417	398	8	,	,	PUNCT
bracis-28417	398	9	vol	vol	NOUN
bracis-28417	398	10	14197	14197	NUM
bracis-28417	398	11	.	.	PUNCT
bracis-28417	399	1	springer	springer	NOUN
bracis-28417	399	2	,	,	PUNCT
bracis-28417	399	3	cham	cham	PROPN
bracis-28417	399	4	.	.	PUNCT
bracis-28417	400	1	https://doi.org/10.1007/978-3-031-45392-2_15	https://doi.org/10.1007/978-3-031-45392-2_15	PROPN
bracis-28417	400	2	download	download	NOUN
bracis-28417	400	3	citation	citation	NOUN
bracis-28417	400	4	.ris	.ris	PUNCT
bracis-28417	401	1	.enw	.enw	PROPN
bracis-28417	401	2	.bib	.bib	PUNCT
bracis-28417	402	1	doi	doi	PROPN
bracis-28417	402	2	:	:	PUNCT
bracis-28417	402	3	https://doi.org/10.1007/978-3-031-45392-2_15	https://doi.org/10.1007/978-3-031-45392-2_15	VERB
bracis-28417	402	4	published	publish	VERB
bracis-28417	402	5	:	:	PUNCT
bracis-28417	402	6	12	12	NUM
bracis-28417	402	7	october	october	PROPN
bracis-28417	402	8	2023	2023	NUM
bracis-28417	402	9	publisher	publisher	NOUN
bracis-28417	402	10	name	name	NOUN
bracis-28417	402	11	:	:	PUNCT
bracis-28417	402	12	springer	springer	NOUN
bracis-28417	402	13	,	,	PUNCT
bracis-28417	402	14	cham	cham	PROPN
bracis-28417	402	15	print	print	PROPN
bracis-28417	402	16	isbn	isbn	PROPN
bracis-28417	402	17	:	:	PUNCT
bracis-28417	402	18	978	978	NUM
bracis-28417	402	19	-	-	SYM
bracis-28417	402	20	3	3	NUM
bracis-28417	402	21	-	-	PUNCT
bracis-28417	402	22	031	031	NUM
bracis-28417	402	23	-	-	PUNCT
bracis-28417	402	24	45391	45391	NUM
bracis-28417	402	25	-	-	SYM
bracis-28417	402	26	5	5	NUM
bracis-28417	402	27	online	online	ADJ
bracis-28417	402	28	isbn	isbn	NOUN
bracis-28417	402	29	:	:	PUNCT
bracis-28417	402	30	978	978	NUM
bracis-28417	402	31	-	-	SYM
bracis-28417	402	32	3	3	NUM
bracis-28417	402	33	-	-	PUNCT
bracis-28417	402	34	031	031	NUM
bracis-28417	402	35	-	-	PUNCT
bracis-28417	402	36	45392	45392	NUM
bracis-28417	402	37	-	-	SYM
bracis-28417	402	38	2	2	NUM
bracis-28417	402	39	ebook	ebook	NOUN
bracis-28417	402	40	packages	package	NOUN
bracis-28417	402	41	:	:	PUNCT
bracis-28417	402	42	computer	computer	NOUN
bracis-28417	402	43	sciencecomputer	sciencecomputer	NOUN
bracis-28417	402	44	science	science	NOUN
bracis-28417	402	45	(	(	PUNCT
bracis-28417	402	46	r0	r0	NOUN
bracis-28417	402	47	)	)	PUNCT
bracis-28417	402	48	share	share	VERB
bracis-28417	402	49	this	this	DET
bracis-28417	402	50	paper	paper	NOUN
bracis-28417	402	51	anyone	anyone	PRON
bracis-28417	402	52	you	you	PRON
bracis-28417	402	53	share	share	VERB
bracis-28417	402	54	the	the	DET
bracis-28417	402	55	following	follow	VERB
bracis-28417	402	56	link	link	NOUN
bracis-28417	402	57	with	with	ADP
bracis-28417	402	58	will	will	AUX
bracis-28417	402	59	be	be	AUX
bracis-28417	402	60	able	able	ADJ
bracis-28417	402	61	to	to	PART
bracis-28417	402	62	read	read	VERB
bracis-28417	402	63	this	this	DET
bracis-28417	402	64	content	content	NOUN
bracis-28417	402	65	:	:	PUNCT
bracis-28417	402	66	get	get	VERB
bracis-28417	402	67	shareable	shareable	ADJ
bracis-28417	402	68	linksorry	linksorry	NOUN
bracis-28417	402	69	,	,	PUNCT
bracis-28417	402	70	a	a	DET
bracis-28417	402	71	shareable	shareable	ADJ
bracis-28417	402	72	link	link	NOUN
bracis-28417	402	73	is	be	AUX
bracis-28417	402	74	not	not	PART
bracis-28417	402	75	currently	currently	ADV
bracis-28417	402	76	available	available	ADJ
bracis-28417	402	77	for	for	ADP
bracis-28417	402	78	this	this	DET
bracis-28417	402	79	article	article	NOUN
bracis-28417	402	80	.	.	PUNCT
bracis-28417	403	1	copy	copy	VERB
bracis-28417	403	2	shareable	shareable	ADJ
bracis-28417	403	3	link	link	NOUN
bracis-28417	403	4	to	to	PART
bracis-28417	403	5	clipboard	clipboard	NOUN
bracis-28417	403	6	provided	provide	VERB
bracis-28417	403	7	by	by	ADP
bracis-28417	403	8	the	the	DET
bracis-28417	403	9	springer	springer	NOUN
bracis-28417	403	10	nature	nature	PROPN
bracis-28417	403	11	sharedit	sharedit	PROPN
bracis-28417	403	12	content	content	NOUN
bracis-28417	403	13	-	-	PUNCT
bracis-28417	403	14	sharing	share	VERB
bracis-28417	403	15	initiative	initiative	NOUN
bracis-28417	403	16	publish	publish	VERB
bracis-28417	403	17	with	with	ADP
bracis-28417	403	18	us	us	PROPN
bracis-28417	403	19	policies	policy	NOUN
bracis-28417	403	20	and	and	CCONJ
bracis-28417	403	21	ethics	ethic	NOUN
bracis-28417	403	22	search	search	NOUN
bracis-28417	403	23	search	search	NOUN
bracis-28417	403	24	by	by	ADP
bracis-28417	403	25	keyword	keyword	NOUN
bracis-28417	403	26	or	or	CCONJ
bracis-28417	403	27	author	author	NOUN
bracis-28417	403	28	search	search	NOUN
bracis-28417	403	29	navigation	navigation	NOUN
bracis-28417	403	30	find	find	VERB
bracis-28417	403	31	a	a	DET
bracis-28417	403	32	journal	journal	NOUN
bracis-28417	403	33	publish	publish	VERB
bracis-28417	403	34	with	with	ADP
bracis-28417	403	35	us	we	PRON
bracis-28417	403	36	track	track	VERB
bracis-28417	403	37	your	your	PRON
bracis-28417	403	38	research	research	NOUN
bracis-28417	403	39	discover	discover	VERB
bracis-28417	403	40	content	content	NOUN
bracis-28417	403	41	journals	journal	NOUN
bracis-28417	403	42	a	a	DET
bracis-28417	403	43	-	-	PUNCT
bracis-28417	403	44	z	z	NOUN
bracis-28417	403	45	books	book	NOUN
bracis-28417	403	46	a	a	DET
bracis-28417	403	47	-	-	PUNCT
bracis-28417	403	48	z	z	NOUN
bracis-28417	403	49	publish	publish	NOUN
bracis-28417	403	50	with	with	ADP
bracis-28417	403	51	us	us	PROPN
bracis-28417	403	52	journal	journal	PROPN
bracis-28417	403	53	finder	finder	PROPN
bracis-28417	403	54	publish	publish	VERB
bracis-28417	403	55	your	your	PRON
bracis-28417	403	56	research	research	NOUN
bracis-28417	403	57	language	language	NOUN
bracis-28417	403	58	editing	edit	VERB
bracis-28417	403	59	open	open	ADJ
bracis-28417	403	60	access	access	NOUN
bracis-28417	403	61	publishing	publishing	NOUN
bracis-28417	403	62	products	product	NOUN
bracis-28417	403	63	and	and	CCONJ
bracis-28417	403	64	services	service	NOUN
bracis-28417	403	65	our	our	PRON
bracis-28417	403	66	products	product	NOUN
bracis-28417	403	67	librarians	librarian	VERB
bracis-28417	403	68	societies	society	NOUN
bracis-28417	403	69	partners	partner	NOUN
bracis-28417	403	70	and	and	CCONJ
bracis-28417	403	71	advertisers	advertiser	NOUN
bracis-28417	403	72	our	our	PRON
bracis-28417	403	73	brands	brand	NOUN
bracis-28417	403	74	springer	springer	NOUN
bracis-28417	403	75	nature	nature	PROPN
bracis-28417	403	76	portfolio	portfolio	PROPN
bracis-28417	403	77	bmc	bmc	PROPN
bracis-28417	403	78	palgrave	palgrave	PROPN
bracis-28417	403	79	macmillan	macmillan	PROPN
bracis-28417	403	80	apress	apress	PROPN
bracis-28417	403	81	discover	discover	VERB
bracis-28417	403	82	your	your	PRON
bracis-28417	403	83	privacy	privacy	NOUN
bracis-28417	403	84	choices	choice	NOUN
bracis-28417	403	85	/	/	SYM
bracis-28417	403	86	manage	manage	NOUN
bracis-28417	403	87	cookies	cookie	NOUN
bracis-28417	403	88	your	your	PRON
bracis-28417	403	89	us	us	PROPN
bracis-28417	404	1	state	state	NOUN
bracis-28417	404	2	privacy	privacy	NOUN
bracis-28417	404	3	rights	right	NOUN
bracis-28417	404	4	accessibility	accessibility	NOUN
bracis-28417	404	5	statement	statement	NOUN
bracis-28417	404	6	terms	term	NOUN
bracis-28417	404	7	and	and	CCONJ
bracis-28417	404	8	conditions	condition	NOUN
bracis-28417	404	9	privacy	privacy	NOUN
bracis-28417	404	10	policy	policy	NOUN
bracis-28417	404	11	help	help	NOUN
bracis-28417	404	12	and	and	CCONJ
bracis-28417	404	13	support	support	VERB
bracis-28417	404	14	legal	legal	ADJ
bracis-28417	404	15	notice	notice	NOUN
bracis-28417	404	16	cancel	cancel	VERB
bracis-28417	404	17	contracts	contract	NOUN
bracis-28417	404	18	here	here	ADV
bracis-28417	404	19	129.74.145.123	129.74.145.123	NUM
bracis-28417	404	20	hesburgh	hesburgh	PROPN
bracis-28417	404	21	library	library	PROPN
bracis-28417	404	22	er	er	INTJ
bracis-28417	404	23	unit	unit	NOUN
bracis-28417	404	24	(	(	PUNCT
bracis-28417	404	25	3005732405	3005732405	NUM
bracis-28417	404	26	)	)	PUNCT
bracis-28417	404	27	northeast	northeast	ADJ
bracis-28417	404	28	research	research	NOUN
bracis-28417	404	29	libraries	library	NOUN
bracis-28417	404	30	(	(	PUNCT
bracis-28417	404	31	nerl	nerl	PROPN
bracis-28417	404	32	)	)	PUNCT
bracis-28417	404	33	(	(	PUNCT
bracis-28417	404	34	8200828607	8200828607	NUM
bracis-28417	404	35	)	)	PUNCT
bracis-28417	404	36	nerl	nerl	VERB
bracis-28417	404	37	ta	ta	X
bracis-28417	404	38	account	account	NOUN
bracis-28417	404	39	(	(	PUNCT
bracis-28417	404	40	3006206169	3006206169	NUM
bracis-28417	404	41	)	)	PUNCT
bracis-28417	404	42	university	university	NOUN
bracis-28417	404	43	of	of	ADP
bracis-28417	404	44	notre	notre	PROPN
bracis-28417	404	45	dame	dame	PROPN
bracis-28417	404	46	hesburgh	hesburgh	PROPN
bracis-28417	404	47	library	library	NOUN
bracis-28417	404	48	(	(	PUNCT
bracis-28417	404	49	3000184373	3000184373	NUM
bracis-28417	404	50	)	)	PUNCT
bracis-28417	405	1	©	©	ADP
bracis-28417	405	2	2025	2025	NUM
bracis-28417	405	3	springer	springer	NOUN
bracis-28417	405	4	nature	nature	NOUN
