id	sid	tid	token	lemma	pos
bracis-33603	1	1	ptt5	ptt5	PROPN
bracis-33603	1	2	-	-	PUNCT
bracis-33603	1	3	v2	v2	NOUN
bracis-33603	1	4	:	:	PUNCT
bracis-33603	1	5	a	a	DET
bracis-33603	1	6	closer	close	ADJ
bracis-33603	1	7	look	look	NOUN
bracis-33603	1	8	at	at	ADP
bracis-33603	1	9	 	 	SPACE
bracis-33603	1	10	continued	continue	VERB
bracis-33603	1	11	pretraining	pretraine	VERB
bracis-33603	1	12	of	of	ADP
bracis-33603	1	13	 	 	SPACE
bracis-33603	1	14	t5	t5	PROPN
bracis-33603	1	15	models	model	NOUN
bracis-33603	1	16	for	for	ADP
bracis-33603	1	17	 	 	SPACE
bracis-33603	1	18	the	the	DET
bracis-33603	1	19	 	 	SPACE
bracis-33603	1	20	portuguese	portuguese	ADJ
bracis-33603	1	21	language	language	NOUN
bracis-33603	1	22	|	|	NOUN
bracis-33603	1	23	springer	springer	NOUN
bracis-33603	1	24	nature	nature	PROPN
bracis-33603	1	25	link	link	PROPN
bracis-33603	1	26	(	(	PUNCT
bracis-33603	1	27	formerly	formerly	ADV
bracis-33603	1	28	springerlink	springerlink	NOUN
bracis-33603	1	29	)	)	PUNCT
bracis-33603	1	30	skip	skip	VERB
bracis-33603	1	31	to	to	ADP
bracis-33603	1	32	main	main	ADJ
bracis-33603	1	33	content	content	NOUN
bracis-33603	1	34	advertisement	advertisement	NOUN
bracis-33603	1	35	log	log	NOUN
bracis-33603	1	36	in	in	ADP
bracis-33603	1	37	menu	menu	NOUN
bracis-33603	1	38	find	find	VERB
bracis-33603	1	39	a	a	DET
bracis-33603	1	40	journal	journal	NOUN
bracis-33603	1	41	publish	publish	VERB
bracis-33603	1	42	with	with	ADP
bracis-33603	1	43	us	we	PRON
bracis-33603	1	44	track	track	VERB
bracis-33603	1	45	your	your	PRON
bracis-33603	1	46	research	research	NOUN
bracis-33603	1	47	search	search	NOUN
bracis-33603	1	48	cart	cart	NOUN
bracis-33603	1	49	home	home	NOUN
bracis-33603	2	1	intelligent	intelligent	ADJ
bracis-33603	2	2	systems	system	NOUN
bracis-33603	2	3	conference	conference	NOUN
bracis-33603	2	4	paper	paper	NOUN
bracis-33603	2	5	ptt5	ptt5	PROPN
bracis-33603	2	6	-	-	PUNCT
bracis-33603	2	7	v2	v2	PROPN
bracis-33603	2	8	:	:	PUNCT
bracis-33603	2	9	a	a	DET
bracis-33603	2	10	closer	close	ADJ
bracis-33603	2	11	look	look	NOUN
bracis-33603	2	12	at	at	ADP
bracis-33603	2	13	 	 	SPACE
bracis-33603	2	14	continued	continue	VERB
bracis-33603	2	15	pretraining	pretraine	VERB
bracis-33603	2	16	of	of	ADP
bracis-33603	2	17	 	 	SPACE
bracis-33603	2	18	t5	t5	PROPN
bracis-33603	2	19	models	model	NOUN
bracis-33603	2	20	for	for	ADP
bracis-33603	2	21	 	 	SPACE
bracis-33603	2	22	the	the	DET
bracis-33603	2	23	 	 	SPACE
bracis-33603	2	24	portuguese	portuguese	ADJ
bracis-33603	2	25	language	language	NOUN
bracis-33603	2	26	conference	conference	NOUN
bracis-33603	2	27	paper	paper	NOUN
bracis-33603	2	28	first	first	ADV
bracis-33603	2	29	online	online	ADV
bracis-33603	2	30	:	:	PUNCT
bracis-33603	2	31	30	30	NUM
bracis-33603	2	32	january	january	NOUN
bracis-33603	2	33	2025	2025	NUM
bracis-33603	2	34	pp	pp	NUM
bracis-33603	2	35	324–338	324–338	NUM
bracis-33603	2	36	cite	cite	VERB
bracis-33603	2	37	this	this	DET
bracis-33603	2	38	conference	conference	NOUN
bracis-33603	2	39	paper	paper	NOUN
bracis-33603	2	40	access	access	NOUN
bracis-33603	2	41	provided	provide	VERB
bracis-33603	2	42	by	by	ADP
bracis-33603	2	43	university	university	PROPN
bracis-33603	2	44	of	of	ADP
bracis-33603	2	45	notre	notre	PROPN
bracis-33603	2	46	dame	dame	PROPN
bracis-33603	2	47	hesburgh	hesburgh	PROPN
bracis-33603	2	48	library	library	PROPN
bracis-33603	2	49	download	download	PROPN
bracis-33603	2	50	book	book	NOUN
bracis-33603	2	51	pdf	pdf	PROPN
bracis-33603	2	52	download	download	NOUN
bracis-33603	2	53	book	book	NOUN
bracis-33603	2	54	epub	epub	PROPN
bracis-33603	2	55	intelligent	intelligent	ADJ
bracis-33603	2	56	systems	system	NOUN
bracis-33603	2	57	(	(	PUNCT
bracis-33603	2	58	bracis	bracis	NOUN
bracis-33603	2	59	2024	2024	NUM
bracis-33603	2	60	)	)	PUNCT
bracis-33603	2	61	ptt5	ptt5	NOUN
bracis-33603	2	62	-	-	PUNCT
bracis-33603	2	63	v2	v2	PROPN
bracis-33603	2	64	:	:	PUNCT
bracis-33603	2	65	a	a	DET
bracis-33603	2	66	closer	close	ADJ
bracis-33603	2	67	look	look	NOUN
bracis-33603	2	68	at	at	ADP
bracis-33603	2	69	 	 	SPACE
bracis-33603	2	70	continued	continue	VERB
bracis-33603	2	71	pretraining	pretraine	VERB
bracis-33603	2	72	of	of	ADP
bracis-33603	2	73	 	 	SPACE
bracis-33603	2	74	t5	t5	PROPN
bracis-33603	2	75	models	model	NOUN
bracis-33603	2	76	for	for	ADP
bracis-33603	2	77	 	 	SPACE
bracis-33603	2	78	the	the	DET
bracis-33603	2	79	 	 	SPACE
bracis-33603	2	80	portuguese	portuguese	ADJ
bracis-33603	2	81	language	language	NOUN
bracis-33603	2	82	download	download	NOUN
bracis-33603	2	83	book	book	NOUN
bracis-33603	2	84	pdf	pdf	PROPN
bracis-33603	2	85	download	download	NOUN
bracis-33603	2	86	book	book	PROPN
bracis-33603	2	87	epub	epub	PROPN
bracis-33603	2	88	marcos	marcos	PROPN
bracis-33603	2	89	piau	piau	PROPN
bracis-33603	2	90	  	  	SPACE
bracis-33603	2	91	orcid	orcid	NOUN
bracis-33603	2	92	:	:	PUNCT
bracis-33603	2	93	orcid.org/0009-0001-1490-34769	orcid.org/0009-0001-1490-34769	NOUN
bracis-33603	2	94	,	,	PUNCT
bracis-33603	2	95	roberto	roberto	PROPN
bracis-33603	2	96	lotufo	lotufo	PROPN
bracis-33603	2	97	  	  	SPACE
bracis-33603	2	98	orcid	orcid	NOUN
bracis-33603	2	99	:	:	PUNCT
bracis-33603	2	100	orcid.org/0000-0002-5652-08529,10	orcid.org/0000-0002-5652-08529,10	PROPN
bracis-33603	2	101	&	&	CCONJ
bracis-33603	2	102	rodrigo	rodrigo	PROPN
bracis-33603	2	103	nogueira	nogueira	PROPN
bracis-33603	2	104	  	  	SPACE
bracis-33603	2	105	orcid	orcid	PROPN
bracis-33603	2	106	:	:	PUNCT
bracis-33603	2	107	orcid.org/0000-0002-2600-60359,11	orcid.org/0000-0002-2600-60359,11	PROPN
bracis-33603	2	108	  	  	SPACE
bracis-33603	2	109	part	part	NOUN
bracis-33603	2	110	of	of	ADP
bracis-33603	2	111	the	the	DET
bracis-33603	2	112	book	book	NOUN
bracis-33603	2	113	series	series	NOUN
bracis-33603	2	114	:	:	PUNCT
bracis-33603	2	115	lecture	lecture	NOUN
bracis-33603	2	116	notes	note	NOUN
bracis-33603	2	117	in	in	ADP
bracis-33603	2	118	computer	computer	NOUN
bracis-33603	2	119	science	science	NOUN
bracis-33603	2	120	(	(	PUNCT
bracis-33603	2	121	(	(	PUNCT
bracis-33603	2	122	lnai	lnai	ADJ
bracis-33603	2	123	,	,	PUNCT
bracis-33603	2	124	volume	volume	NOUN
bracis-33603	2	125	15413	15413	NUM
bracis-33603	2	126	)	)	PUNCT
bracis-33603	2	127	)	)	PUNCT
bracis-33603	2	128	included	include	VERB
bracis-33603	2	129	in	in	ADP
bracis-33603	2	130	the	the	DET
bracis-33603	2	131	following	follow	VERB
bracis-33603	2	132	conference	conference	NOUN
bracis-33603	2	133	series	series	NOUN
bracis-33603	2	134	:	:	PUNCT
bracis-33603	2	135	brazilian	brazilian	ADJ
bracis-33603	2	136	conference	conference	NOUN
bracis-33603	2	137	on	on	ADP
bracis-33603	2	138	intelligent	intelligent	ADJ
bracis-33603	2	139	systems	system	NOUN
bracis-33603	2	140	392	392	NUM
bracis-33603	2	141	accesses	access	VERB
bracis-33603	2	142	10	10	NUM
bracis-33603	2	143	citations	citation	NOUN
bracis-33603	2	144	abstract	abstract	ADJ
bracis-33603	2	145	despite	despite	SCONJ
bracis-33603	2	146	advancements	advancement	NOUN
bracis-33603	2	147	in	in	ADP
bracis-33603	2	148	natural	natural	ADJ
bracis-33603	2	149	language	language	NOUN
bracis-33603	2	150	processing	processing	NOUN
bracis-33603	2	151	(	(	PUNCT
bracis-33603	2	152	nlp	nlp	NOUN
bracis-33603	2	153	)	)	PUNCT
bracis-33603	2	154	and	and	CCONJ
bracis-33603	2	155	the	the	DET
bracis-33603	2	156	growing	grow	VERB
bracis-33603	2	157	availability	availability	NOUN
bracis-33603	2	158	of	of	ADP
bracis-33603	2	159	pretrained	pretraine	VERB
bracis-33603	2	160	models	model	NOUN
bracis-33603	2	161	,	,	PUNCT
bracis-33603	2	162	the	the	DET
bracis-33603	2	163	english	english	ADJ
bracis-33603	2	164	language	language	NOUN
bracis-33603	2	165	remains	remain	VERB
bracis-33603	2	166	the	the	DET
bracis-33603	2	167	primary	primary	ADJ
bracis-33603	2	168	focus	focus	NOUN
bracis-33603	2	169	of	of	ADP
bracis-33603	2	170	model	model	NOUN
bracis-33603	2	171	development	development	NOUN
bracis-33603	2	172	.	.	PUNCT
bracis-33603	3	1	continued	continue	VERB
bracis-33603	3	2	pretraining	pretraine	VERB
bracis-33603	3	3	on	on	ADP
bracis-33603	3	4	language	language	NOUN
bracis-33603	3	5	-	-	PUNCT
bracis-33603	3	6	specific	specific	NOUN
bracis-33603	3	7	corpora	corpora	NOUN
bracis-33603	3	8	provides	provide	VERB
bracis-33603	3	9	a	a	DET
bracis-33603	3	10	practical	practical	ADJ
bracis-33603	3	11	solution	solution	NOUN
bracis-33603	3	12	for	for	ADP
bracis-33603	3	13	adapting	adapt	VERB
bracis-33603	3	14	models	model	NOUN
bracis-33603	3	15	to	to	ADP
bracis-33603	3	16	other	other	ADJ
bracis-33603	3	17	languages	language	NOUN
bracis-33603	3	18	.	.	PUNCT
bracis-33603	4	1	however	however	ADV
bracis-33603	4	2	,	,	PUNCT
bracis-33603	4	3	the	the	DET
bracis-33603	4	4	impact	impact	NOUN
bracis-33603	4	5	of	of	ADP
bracis-33603	4	6	different	different	ADJ
bracis-33603	4	7	pretraining	pretraine	VERB
bracis-33603	4	8	settings	setting	NOUN
bracis-33603	4	9	on	on	ADP
bracis-33603	4	10	downstream	downstream	ADJ
bracis-33603	4	11	tasks	task	NOUN
bracis-33603	4	12	remains	remain	VERB
bracis-33603	4	13	underexplored	underexplored	ADJ
bracis-33603	4	14	.	.	PUNCT
bracis-33603	5	1	this	this	DET
bracis-33603	5	2	work	work	NOUN
bracis-33603	5	3	introduces	introduce	VERB
bracis-33603	5	4	ptt5	ptt5	NOUN
bracis-33603	5	5	-	-	PUNCT
bracis-33603	5	6	v2	v2	PROPN
bracis-33603	5	7	,	,	PUNCT
bracis-33603	5	8	investigating	investigate	VERB
bracis-33603	5	9	the	the	DET
bracis-33603	5	10	continued	continued	ADJ
bracis-33603	5	11	pretraining	pretraining	NOUN
bracis-33603	5	12	of	of	ADP
bracis-33603	5	13	t5	t5	PROPN
bracis-33603	5	14	models	model	NOUN
bracis-33603	5	15	for	for	ADP
bracis-33603	5	16	portuguese	portuguese	NOUN
bracis-33603	5	17	.	.	PUNCT
bracis-33603	6	1	we	we	PRON
bracis-33603	6	2	first	first	ADV
bracis-33603	6	3	develop	develop	VERB
bracis-33603	6	4	a	a	DET
bracis-33603	6	5	baseline	baseline	NOUN
bracis-33603	6	6	set	set	NOUN
bracis-33603	6	7	of	of	ADP
bracis-33603	6	8	settings	setting	NOUN
bracis-33603	6	9	and	and	CCONJ
bracis-33603	6	10	pretrain	pretrain	NOUN
bracis-33603	6	11	models	model	NOUN
bracis-33603	6	12	with	with	ADP
bracis-33603	6	13	sizes	size	NOUN
bracis-33603	6	14	up	up	ADP
bracis-33603	6	15	to	to	ADP
bracis-33603	6	16	3b	3b	NUM
bracis-33603	6	17	parameters	parameter	NOUN
bracis-33603	6	18	.	.	PUNCT
bracis-33603	7	1	finetuning	finetune	VERB
bracis-33603	7	2	on	on	ADP
bracis-33603	7	3	three	three	NUM
bracis-33603	7	4	portuguese	portuguese	ADJ
bracis-33603	7	5	downstream	downstream	ADJ
bracis-33603	7	6	tasks	task	NOUN
bracis-33603	7	7	(	(	PUNCT
bracis-33603	7	8	assin2	assin2	PROPN
bracis-33603	7	9	sts	st	NOUN
bracis-33603	7	10	,	,	PUNCT
bracis-33603	7	11	assin2	assin2	PROPN
bracis-33603	7	12	rte	rte	NOUN
bracis-33603	7	13	,	,	PUNCT
bracis-33603	7	14	and	and	CCONJ
bracis-33603	7	15	tweetsentbr	tweetsentbr	NOUN
bracis-33603	7	16	)	)	PUNCT
bracis-33603	7	17	yields	yield	VERB
bracis-33603	7	18	sota	sota	NOUN
bracis-33603	7	19	results	result	NOUN
bracis-33603	7	20	on	on	ADP
bracis-33603	7	21	the	the	DET
bracis-33603	7	22	latter	latter	ADJ
bracis-33603	7	23	two	two	NUM
bracis-33603	7	24	.	.	PUNCT
bracis-33603	8	1	we	we	PRON
bracis-33603	8	2	then	then	ADV
bracis-33603	8	3	explore	explore	VERB
bracis-33603	8	4	the	the	DET
bracis-33603	8	5	effects	effect	NOUN
bracis-33603	8	6	of	of	ADP
bracis-33603	8	7	different	different	ADJ
bracis-33603	8	8	pretraining	pretraine	VERB
bracis-33603	8	9	configurations	configuration	NOUN
bracis-33603	8	10	,	,	PUNCT
bracis-33603	8	11	including	include	VERB
bracis-33603	8	12	quality	quality	NOUN
bracis-33603	8	13	filters	filter	NOUN
bracis-33603	8	14	,	,	PUNCT
bracis-33603	8	15	optimization	optimization	NOUN
bracis-33603	8	16	strategies	strategy	NOUN
bracis-33603	8	17	,	,	PUNCT
bracis-33603	8	18	and	and	CCONJ
bracis-33603	8	19	multi	multi	ADJ
bracis-33603	8	20	-	-	ADJ
bracis-33603	8	21	epoch	epoch	ADJ
bracis-33603	8	22	pretraining	pretraining	NOUN
bracis-33603	8	23	.	.	PUNCT
bracis-33603	9	1	perhaps	perhaps	ADV
bracis-33603	9	2	surprisingly	surprisingly	ADV
bracis-33603	9	3	,	,	PUNCT
bracis-33603	9	4	their	their	PRON
bracis-33603	9	5	impact	impact	NOUN
bracis-33603	9	6	remains	remain	VERB
bracis-33603	9	7	subtle	subtle	ADJ
bracis-33603	9	8	compared	compare	VERB
bracis-33603	9	9	to	to	ADP
bracis-33603	9	10	our	our	PRON
bracis-33603	9	11	baseline	baseline	NOUN
bracis-33603	9	12	.	.	PUNCT
bracis-33603	10	1	we	we	PRON
bracis-33603	10	2	release	release	VERB
bracis-33603	10	3	ptt5	ptt5	PROPN
bracis-33603	10	4	-	-	PUNCT
bracis-33603	10	5	v2	v2	NOUN
bracis-33603	10	6	pretrained	pretraine	VERB
bracis-33603	10	7	checkpoints	checkpoint	NOUN
bracis-33603	10	8	and	and	CCONJ
bracis-33603	10	9	finetuned	finetune	VERB
bracis-33603	10	10	monot5	monot5	NOUN
bracis-33603	10	11	rerankers	reranker	NOUN
bracis-33603	10	12	on	on	ADP
bracis-33603	10	13	huggingface	huggingface	NOUN
bracis-33603	10	14	in	in	ADP
bracis-33603	10	15	their	their	PRON
bracis-33603	10	16	respective	respective	ADJ
bracis-33603	10	17	collections	collection	NOUN
bracis-33603	10	18	at	at	ADP
bracis-33603	10	19	https://huggingface.co/unicamp-dl	https://huggingface.co/unicamp-dl	PROPN
bracis-33603	10	20	.	.	PROPN
bracis-33603	10	21	access	access	NOUN
bracis-33603	10	22	provided	provide	VERB
bracis-33603	10	23	by	by	ADP
bracis-33603	10	24	university	university	PROPN
bracis-33603	10	25	of	of	ADP
bracis-33603	10	26	notre	notre	PROPN
bracis-33603	10	27	dame	dame	PROPN
bracis-33603	10	28	hesburgh	hesburgh	PROPN
bracis-33603	10	29	library	library	PROPN
bracis-33603	10	30	.	.	PUNCT
bracis-33603	11	1	download	download	PROPN
bracis-33603	11	2	conference	conference	NOUN
bracis-33603	11	3	paper	paper	NOUN
bracis-33603	11	4	pdf	pdf	NOUN
bracis-33603	11	5	similar	similar	ADJ
bracis-33603	11	6	content	content	NOUN
bracis-33603	11	7	being	be	AUX
bracis-33603	11	8	viewed	view	VERB
bracis-33603	11	9	by	by	ADP
bracis-33603	11	10	others	other	NOUN
bracis-33603	11	11	sabiá	sabiá	PROPN
bracis-33603	11	12	:	:	PUNCT
bracis-33603	11	13	portuguese	portuguese	ADJ
bracis-33603	11	14	large	large	ADJ
bracis-33603	11	15	language	language	NOUN
bracis-33603	11	16	models	model	NOUN
bracis-33603	11	17	chapter	chapter	NOUN
bracis-33603	11	18	©	©	PROPN
bracis-33603	11	19	2023	2023	NUM
bracis-33603	11	20	pre	pre	ADJ
bracis-33603	11	21	-	-	ADJ
bracis-33603	11	22	trained	train	VERB
bracis-33603	11	23	models	model	NOUN
bracis-33603	11	24	for	for	ADP
bracis-33603	11	25	representation	representation	NOUN
bracis-33603	11	26	learning	learn	VERB
bracis-33603	11	27	chapter	chapter	NOUN
bracis-33603	11	28	©	©	PROPN
bracis-33603	11	29	2023	2023	NUM
bracis-33603	11	30	evaluating	evaluate	VERB
bracis-33603	11	31	the	the	DET
bracis-33603	11	32	 	 	SPACE
bracis-33603	11	33	effectiveness	effectiveness	NOUN
bracis-33603	11	34	of	of	ADP
bracis-33603	11	35	 	 	SPACE
bracis-33603	11	36	pre	pre	ADJ
bracis-33603	11	37	-	-	ADJ
bracis-33603	11	38	trained	train	VERB
bracis-33603	11	39	language	language	NOUN
bracis-33603	11	40	models	model	NOUN
bracis-33603	11	41	in	in	ADP
bracis-33603	11	42	 	 	SPACE
bracis-33603	11	43	predicting	predict	VERB
bracis-33603	11	44	the	the	DET
bracis-33603	11	45	 	 	SPACE
bracis-33603	11	46	helpfulness	helpfulness	NOUN
bracis-33603	11	47	of	of	ADP
bracis-33603	11	48	 	 	SPACE
bracis-33603	11	49	online	online	ADJ
bracis-33603	11	50	product	product	NOUN
bracis-33603	11	51	reviews	review	NOUN
bracis-33603	11	52	chapter	chapter	NOUN
bracis-33603	11	53	©	©	ADP
bracis-33603	11	54	2024	2024	NUM
bracis-33603	11	55	1	1	NUM
bracis-33603	11	56	introduction	introduction	NOUN
bracis-33603	11	57	transformer	transformer	NOUN
bracis-33603	11	58	-	-	PUNCT
bracis-33603	11	59	based	base	VERB
bracis-33603	11	60	pretrained	pretraine	VERB
bracis-33603	11	61	language	language	NOUN
bracis-33603	11	62	models	model	NOUN
bracis-33603	11	63	have	have	AUX
bracis-33603	11	64	established	establish	VERB
bracis-33603	11	65	themselves	themselves	PRON
bracis-33603	11	66	as	as	ADP
bracis-33603	11	67	the	the	DET
bracis-33603	11	68	core	core	NOUN
bracis-33603	11	69	paradigm	paradigm	NOUN
bracis-33603	11	70	in	in	ADP
bracis-33603	11	71	the	the	DET
bracis-33603	11	72	field	field	NOUN
bracis-33603	11	73	of	of	ADP
bracis-33603	11	74	natural	natural	ADJ
bracis-33603	11	75	language	language	NOUN
bracis-33603	11	76	processing	processing	NOUN
bracis-33603	11	77	(	(	PUNCT
bracis-33603	11	78	nlp	nlp	NOUN
bracis-33603	11	79	)	)	PUNCT
bracis-33603	11	80	.	.	PUNCT
bracis-33603	12	1	starting	start	VERB
bracis-33603	12	2	with	with	ADP
bracis-33603	12	3	the	the	DET
bracis-33603	12	4	advent	advent	NOUN
bracis-33603	12	5	of	of	ADP
bracis-33603	12	6	bert	bert	PROPN
bracis-33603	12	7	[	[	X
bracis-33603	12	8	12	12	NUM
bracis-33603	12	9	]	]	PUNCT
bracis-33603	12	10	,	,	PUNCT
bracis-33603	12	11	which	which	PRON
bracis-33603	12	12	popularized	popularize	VERB
bracis-33603	12	13	the	the	DET
bracis-33603	12	14	“	"	PUNCT
bracis-33603	12	15	pretrain	pretrain	NOUN
bracis-33603	12	16	,	,	PUNCT
bracis-33603	12	17	then	then	ADV
bracis-33603	12	18	fine	fine	ADJ
bracis-33603	12	19	-	-	PUNCT
bracis-33603	12	20	tune	tune	NOUN
bracis-33603	12	21	”	"	PUNCT
bracis-33603	12	22	approach	approach	NOUN
bracis-33603	12	23	and	and	CCONJ
bracis-33603	12	24	the	the	DET
bracis-33603	12	25	use	use	NOUN
bracis-33603	12	26	of	of	ADP
bracis-33603	12	27	the	the	DET
bracis-33603	12	28	transformer	transformer	NOUN
bracis-33603	12	29	architecture	architecture	NOUN
bracis-33603	12	30	itself	itself	PRON
bracis-33603	12	31	,	,	PUNCT
bracis-33603	12	32	these	these	DET
bracis-33603	12	33	models	model	NOUN
bracis-33603	12	34	acquire	acquire	VERB
bracis-33603	12	35	a	a	DET
bracis-33603	12	36	general	general	ADJ
bracis-33603	12	37	-	-	PUNCT
bracis-33603	12	38	purpose	purpose	NOUN
bracis-33603	12	39	language	language	NOUN
bracis-33603	12	40	representation	representation	NOUN
bracis-33603	12	41	by	by	ADP
bracis-33603	12	42	unsupervised	unsupervised	ADJ
bracis-33603	12	43	pretraining	pretraine	VERB
bracis-33603	12	44	on	on	ADP
bracis-33603	12	45	extensive	extensive	ADJ
bracis-33603	12	46	corpora	corpus	NOUN
bracis-33603	12	47	of	of	ADP
bracis-33603	12	48	unlabeled	unlabeled	ADJ
bracis-33603	12	49	text	text	NOUN
bracis-33603	12	50	.	.	PUNCT
bracis-33603	13	1	the	the	DET
bracis-33603	13	2	dynamics	dynamic	NOUN
bracis-33603	13	3	of	of	ADP
bracis-33603	13	4	the	the	DET
bracis-33603	13	5	pretraining	pretraine	VERB
bracis-33603	13	6	process	process	NOUN
bracis-33603	13	7	had	have	AUX
bracis-33603	13	8	been	be	AUX
bracis-33603	13	9	studied	study	VERB
bracis-33603	13	10	in	in	ADP
bracis-33603	13	11	depth	depth	NOUN
bracis-33603	13	12	by	by	ADP
bracis-33603	13	13	many	many	ADJ
bracis-33603	13	14	works	work	NOUN
bracis-33603	13	15	,	,	PUNCT
bracis-33603	13	16	like	like	ADP
bracis-33603	13	17	raffel	raffel	NOUN
bracis-33603	13	18	et	et	NOUN
bracis-33603	13	19	al	al	PROPN
bracis-33603	13	20	.	.	PUNCT
bracis-33603	14	1	[	[	X
bracis-33603	14	2	34	34	NUM
bracis-33603	14	3	]	]	PUNCT
bracis-33603	14	4	,	,	PUNCT
bracis-33603	14	5	that	that	PRON
bracis-33603	14	6	introduced	introduce	VERB
bracis-33603	14	7	t5	t5	PROPN
bracis-33603	14	8	,	,	PUNCT
bracis-33603	14	9	and	and	CCONJ
bracis-33603	14	10	scaled	scale	VERB
bracis-33603	14	11	up	up	ADP
bracis-33603	14	12	models	model	NOUN
bracis-33603	14	13	to	to	ADP
bracis-33603	14	14	billions	billion	NOUN
bracis-33603	14	15	of	of	ADP
bracis-33603	14	16	parameters	parameter	NOUN
bracis-33603	14	17	and	and	CCONJ
bracis-33603	14	18	set	set	VERB
bracis-33603	14	19	new	new	ADJ
bracis-33603	14	20	sotas	sota	NOUN
bracis-33603	14	21	across	across	ADP
bracis-33603	14	22	many	many	ADJ
bracis-33603	14	23	tasks	task	NOUN
bracis-33603	14	24	.	.	PUNCT
bracis-33603	15	1	the	the	DET
bracis-33603	15	2	trend	trend	NOUN
bracis-33603	15	3	towards	towards	ADP
bracis-33603	15	4	increasing	increase	VERB
bracis-33603	15	5	model	model	NOUN
bracis-33603	15	6	sizes	size	NOUN
bracis-33603	15	7	and	and	CCONJ
bracis-33603	15	8	datasets	dataset	NOUN
bracis-33603	15	9	to	to	PART
bracis-33603	15	10	improve	improve	VERB
bracis-33603	15	11	performance	performance	NOUN
bracis-33603	15	12	motivated	motivated	ADJ
bracis-33603	15	13	studies	study	NOUN
bracis-33603	15	14	like	like	ADP
bracis-33603	15	15	kaplan	kaplan	PROPN
bracis-33603	15	16	et	et	PROPN
bracis-33603	15	17	al	al	PROPN
bracis-33603	15	18	.	.	PUNCT
bracis-33603	16	1	[	[	X
bracis-33603	16	2	23	23	NUM
bracis-33603	16	3	]	]	PUNCT
bracis-33603	16	4	on	on	ADP
bracis-33603	16	5	scaling	scale	VERB
bracis-33603	16	6	laws	law	NOUN
bracis-33603	16	7	and	and	CCONJ
bracis-33603	16	8	hoffmann	hoffmann	PROPN
bracis-33603	16	9	et	et	PROPN
bracis-33603	16	10	al	al	PROPN
bracis-33603	16	11	.	.	PUNCT
bracis-33603	17	1	[	[	X
bracis-33603	17	2	18	18	NUM
bracis-33603	17	3	]	]	PUNCT
bracis-33603	17	4	,	,	PUNCT
bracis-33603	17	5	who	who	PRON
bracis-33603	17	6	demonstrated	demonstrate	VERB
bracis-33603	17	7	the	the	DET
bracis-33603	17	8	importance	importance	NOUN
bracis-33603	17	9	of	of	ADP
bracis-33603	17	10	training	training	NOUN
bracis-33603	17	11	data	datum	NOUN
bracis-33603	17	12	size	size	NOUN
bracis-33603	17	13	relative	relative	ADJ
bracis-33603	17	14	to	to	ADP
bracis-33603	17	15	model	model	NOUN
bracis-33603	17	16	size	size	NOUN
bracis-33603	17	17	for	for	ADP
bracis-33603	17	18	compute	compute	NOUN
bracis-33603	17	19	-	-	PUNCT
bracis-33603	17	20	optimal	optimal	ADJ
bracis-33603	17	21	training	training	NOUN
bracis-33603	17	22	regimes	regime	NOUN
bracis-33603	17	23	;	;	PUNCT
bracis-33603	17	24	more	more	ADV
bracis-33603	17	25	recently	recently	ADV
bracis-33603	17	26	work	work	VERB
bracis-33603	17	27	from	from	ADP
bracis-33603	17	28	gadre	gadre	PROPN
bracis-33603	17	29	et	et	PROPN
bracis-33603	17	30	al	al	PROPN
bracis-33603	17	31	.	.	PUNCT
bracis-33603	18	1	[	[	X
bracis-33603	18	2	14	14	NUM
bracis-33603	18	3	]	]	PUNCT
bracis-33603	18	4	specifically	specifically	ADV
bracis-33603	18	5	examined	examine	VERB
bracis-33603	18	6	the	the	DET
bracis-33603	18	7	influence	influence	NOUN
bracis-33603	18	8	of	of	ADP
bracis-33603	18	9	extended	extended	ADJ
bracis-33603	18	10	pretraining	pretraine	VERB
bracis-33603	18	11	on	on	ADP
bracis-33603	18	12	downstream	downstream	ADJ
bracis-33603	18	13	task	task	NOUN
bracis-33603	18	14	performance	performance	NOUN
bracis-33603	18	15	.	.	PUNCT
bracis-33603	19	1	despite	despite	SCONJ
bracis-33603	19	2	the	the	DET
bracis-33603	19	3	extensive	extensive	ADJ
bracis-33603	19	4	study	study	NOUN
bracis-33603	19	5	of	of	ADP
bracis-33603	19	6	pretraining	pretraine	VERB
bracis-33603	19	7	dynamics	dynamic	NOUN
bracis-33603	19	8	,	,	PUNCT
bracis-33603	19	9	the	the	DET
bracis-33603	19	10	focus	focus	NOUN
bracis-33603	19	11	has	have	AUX
bracis-33603	19	12	predominantly	predominantly	ADV
bracis-33603	19	13	been	be	AUX
bracis-33603	19	14	on	on	ADP
bracis-33603	19	15	english	english	PROPN
bracis-33603	19	16	,	,	PUNCT
bracis-33603	19	17	leaving	leave	VERB
bracis-33603	19	18	non	non	ADJ
bracis-33603	19	19	-	-	ADJ
bracis-33603	19	20	english	english	ADJ
bracis-33603	19	21	languages	language	NOUN
bracis-33603	19	22	less	less	ADV
bracis-33603	19	23	explored	explore	VERB
bracis-33603	19	24	.	.	PUNCT
bracis-33603	20	1	continued	continue	VERB
bracis-33603	20	2	pretraining	pretraine	VERB
bracis-33603	20	3	presents	present	VERB
bracis-33603	20	4	a	a	DET
bracis-33603	20	5	strategic	strategic	ADJ
bracis-33603	20	6	approach	approach	NOUN
bracis-33603	20	7	to	to	ADP
bracis-33603	20	8	adapting	adapt	VERB
bracis-33603	20	9	these	these	DET
bracis-33603	20	10	models	model	NOUN
bracis-33603	20	11	to	to	ADP
bracis-33603	20	12	additional	additional	ADJ
bracis-33603	20	13	languages	language	NOUN
bracis-33603	20	14	and	and	CCONJ
bracis-33603	20	15	domains	domain	NOUN
bracis-33603	20	16	using	use	VERB
bracis-33603	20	17	significantly	significantly	ADV
bracis-33603	20	18	less	less	ADJ
bracis-33603	20	19	data	datum	NOUN
bracis-33603	20	20	and	and	CCONJ
bracis-33603	20	21	computational	computational	ADJ
bracis-33603	20	22	resources	resource	NOUN
bracis-33603	20	23	than	than	ADP
bracis-33603	20	24	training	training	NOUN
bracis-33603	20	25	from	from	ADP
bracis-33603	20	26	scratch	scratch	NOUN
bracis-33603	20	27	.	.	PUNCT
bracis-33603	21	1	this	this	DET
bracis-33603	21	2	method	method	NOUN
bracis-33603	21	3	involves	involve	VERB
bracis-33603	21	4	further	far	ADV
bracis-33603	21	5	pretraining	pretraine	VERB
bracis-33603	21	6	on	on	ADP
bracis-33603	21	7	language	language	NOUN
bracis-33603	21	8	-	-	PUNCT
bracis-33603	21	9	specific	specific	NOUN
bracis-33603	21	10	corpora	corpora	NOUN
bracis-33603	21	11	,	,	PUNCT
bracis-33603	21	12	which	which	PRON
bracis-33603	21	13	has	have	AUX
bracis-33603	21	14	been	be	AUX
bracis-33603	21	15	shown	show	VERB
bracis-33603	21	16	to	to	PART
bracis-33603	21	17	substantially	substantially	ADV
bracis-33603	21	18	enhance	enhance	VERB
bracis-33603	21	19	model	model	NOUN
bracis-33603	21	20	performance	performance	NOUN
bracis-33603	21	21	on	on	ADP
bracis-33603	21	22	downstream	downstream	ADJ
bracis-33603	21	23	tasks	task	NOUN
bracis-33603	21	24	in	in	ADP
bracis-33603	21	25	the	the	DET
bracis-33603	21	26	target	target	NOUN
bracis-33603	21	27	language	language	NOUN
bracis-33603	21	28	[	[	X
bracis-33603	21	29	5	5	NUM
bracis-33603	21	30	,	,	PUNCT
bracis-33603	21	31	6	6	NUM
bracis-33603	21	32	,	,	PUNCT
bracis-33603	21	33	11	11	NUM
bracis-33603	21	34	,	,	PUNCT
bracis-33603	21	35	25	25	NUM
bracis-33603	21	36	,	,	PUNCT
bracis-33603	21	37	32	32	NUM
bracis-33603	21	38	,	,	PUNCT
bracis-33603	21	39	42	42	NUM
bracis-33603	21	40	]	]	PUNCT
bracis-33603	21	41	.	.	PUNCT
bracis-33603	22	1	however	however	ADV
bracis-33603	22	2	,	,	PUNCT
bracis-33603	22	3	there	there	PRON
bracis-33603	22	4	is	be	VERB
bracis-33603	22	5	a	a	DET
bracis-33603	22	6	lack	lack	NOUN
bracis-33603	22	7	of	of	ADP
bracis-33603	22	8	detailed	detailed	ADJ
bracis-33603	22	9	investigations	investigation	NOUN
bracis-33603	22	10	into	into	ADP
bracis-33603	22	11	how	how	SCONJ
bracis-33603	22	12	different	different	ADJ
bracis-33603	22	13	settings	setting	NOUN
bracis-33603	22	14	during	during	ADP
bracis-33603	22	15	the	the	DET
bracis-33603	22	16	continued	continued	ADJ
bracis-33603	22	17	pretraining	pretraine	VERB
bracis-33603	22	18	phase	phase	NOUN
bracis-33603	22	19	influence	influence	NOUN
bracis-33603	22	20	on	on	ADP
bracis-33603	22	21	downstream	downstream	ADJ
bracis-33603	22	22	tasks	task	NOUN
bracis-33603	22	23	performance	performance	NOUN
bracis-33603	22	24	,	,	PUNCT
bracis-33603	22	25	with	with	ADP
bracis-33603	22	26	most	most	ADJ
bracis-33603	22	27	studies	study	NOUN
bracis-33603	22	28	merely	merely	ADV
bracis-33603	22	29	aiming	aim	VERB
bracis-33603	22	30	for	for	ADP
bracis-33603	22	31	benchmark	benchmark	NOUN
bracis-33603	22	32	-	-	PUNCT
bracis-33603	22	33	leading	lead	VERB
bracis-33603	22	34	results	result	NOUN
bracis-33603	22	35	without	without	ADP
bracis-33603	22	36	a	a	DET
bracis-33603	22	37	thorough	thorough	ADJ
bracis-33603	22	38	examination	examination	NOUN
bracis-33603	22	39	of	of	ADP
bracis-33603	22	40	the	the	DET
bracis-33603	22	41	underlying	underlie	VERB
bracis-33603	22	42	factors	factor	NOUN
bracis-33603	22	43	.	.	PUNCT
bracis-33603	23	1	in	in	ADP
bracis-33603	23	2	this	this	DET
bracis-33603	23	3	work	work	NOUN
bracis-33603	23	4	,	,	PUNCT
bracis-33603	23	5	we	we	PRON
bracis-33603	23	6	study	study	VERB
bracis-33603	23	7	the	the	DET
bracis-33603	23	8	continued	continued	ADJ
bracis-33603	23	9	pretraining	pretraining	NOUN
bracis-33603	23	10	of	of	ADP
bracis-33603	23	11	t5	t5	PROPN
bracis-33603	23	12	models	model	NOUN
bracis-33603	23	13	for	for	ADP
bracis-33603	23	14	the	the	DET
bracis-33603	23	15	portuguese	portuguese	ADJ
bracis-33603	23	16	language	language	NOUN
bracis-33603	23	17	,	,	PUNCT
bracis-33603	23	18	analyzing	analyze	VERB
bracis-33603	23	19	the	the	DET
bracis-33603	23	20	impact	impact	NOUN
bracis-33603	23	21	of	of	ADP
bracis-33603	23	22	various	various	ADJ
bracis-33603	23	23	settings	setting	NOUN
bracis-33603	23	24	on	on	ADP
bracis-33603	23	25	downstream	downstream	ADJ
bracis-33603	23	26	task	task	NOUN
bracis-33603	23	27	performance	performance	NOUN
bracis-33603	23	28	.	.	PUNCT
bracis-33603	24	1	rather	rather	ADV
bracis-33603	24	2	than	than	ADP
bracis-33603	24	3	solely	solely	ADV
bracis-33603	24	4	focusing	focus	VERB
bracis-33603	24	5	on	on	ADP
bracis-33603	24	6	achieving	achieve	VERB
bracis-33603	24	7	state	state	NOUN
bracis-33603	24	8	-	-	PUNCT
bracis-33603	24	9	of	of	ADP
bracis-33603	24	10	-	-	PUNCT
bracis-33603	24	11	the	the	DET
bracis-33603	24	12	-	-	PUNCT
bracis-33603	24	13	art	art	NOUN
bracis-33603	24	14	results	result	NOUN
bracis-33603	24	15	,	,	PUNCT
bracis-33603	24	16	our	our	PRON
bracis-33603	24	17	study	study	NOUN
bracis-33603	24	18	also	also	ADV
bracis-33603	24	19	investigates	investigate	VERB
bracis-33603	24	20	how	how	SCONJ
bracis-33603	24	21	factors	factor	NOUN
bracis-33603	24	22	like	like	ADP
bracis-33603	24	23	model	model	NOUN
bracis-33603	24	24	size	size	NOUN
bracis-33603	24	25	,	,	PUNCT
bracis-33603	24	26	optimization	optimization	NOUN
bracis-33603	24	27	schedules	schedule	NOUN
bracis-33603	24	28	,	,	PUNCT
bracis-33603	24	29	and	and	CCONJ
bracis-33603	24	30	the	the	DET
bracis-33603	24	31	application	application	NOUN
bracis-33603	24	32	of	of	ADP
bracis-33603	24	33	quality	quality	NOUN
bracis-33603	24	34	filters	filter	NOUN
bracis-33603	24	35	in	in	ADP
bracis-33603	24	36	the	the	DET
bracis-33603	24	37	pretraining	pretraine	VERB
bracis-33603	24	38	dataset	dataset	NOUN
bracis-33603	24	39	affect	affect	VERB
bracis-33603	24	40	the	the	DET
bracis-33603	24	41	performance	performance	NOUN
bracis-33603	24	42	.	.	PUNCT
bracis-33603	25	1	we	we	PRON
bracis-33603	25	2	continue	continue	VERB
bracis-33603	25	3	the	the	DET
bracis-33603	25	4	pretraning	pretraning	NOUN
bracis-33603	25	5	of	of	ADP
bracis-33603	25	6	google	google	PROPN
bracis-33603	25	7	’s	’s	PART
bracis-33603	25	8	t5	t5	PROPN
bracis-33603	25	9	with	with	ADP
bracis-33603	25	10	up	up	ADP
bracis-33603	25	11	to	to	PART
bracis-33603	25	12	3	3	NUM
bracis-33603	25	13	billion	billion	NUM
bracis-33603	25	14	parameters	parameter	NOUN
bracis-33603	25	15	on	on	ADP
bracis-33603	25	16	portuguese	portuguese	ADJ
bracis-33603	25	17	texts	text	NOUN
bracis-33603	25	18	.	.	PUNCT
bracis-33603	26	1	by	by	ADP
bracis-33603	26	2	experimenting	experiment	VERB
bracis-33603	26	3	with	with	ADP
bracis-33603	26	4	different	different	ADJ
bracis-33603	26	5	configurations	configuration	NOUN
bracis-33603	26	6	in	in	ADP
bracis-33603	26	7	the	the	DET
bracis-33603	26	8	pretraining	pretraine	VERB
bracis-33603	26	9	stage	stage	NOUN
bracis-33603	26	10	,	,	PUNCT
bracis-33603	26	11	we	we	PRON
bracis-33603	26	12	observe	observe	VERB
bracis-33603	26	13	nuanced	nuanced	ADJ
bracis-33603	26	14	effects	effect	NOUN
bracis-33603	26	15	on	on	ADP
bracis-33603	26	16	downstream	downstream	ADJ
bracis-33603	26	17	tasks	task	NOUN
bracis-33603	26	18	,	,	PUNCT
bracis-33603	26	19	with	with	ADP
bracis-33603	26	20	some	some	DET
bracis-33603	26	21	settings	setting	NOUN
bracis-33603	26	22	only	only	ADV
bracis-33603	26	23	marginally	marginally	ADV
bracis-33603	26	24	outperforming	outperform	VERB
bracis-33603	26	25	the	the	DET
bracis-33603	26	26	baselines	baseline	NOUN
bracis-33603	26	27	.	.	PUNCT
bracis-33603	27	1	our	our	PRON
bracis-33603	27	2	findings	finding	NOUN
bracis-33603	27	3	also	also	ADV
bracis-33603	27	4	suggest	suggest	VERB
bracis-33603	27	5	that	that	SCONJ
bracis-33603	27	6	while	while	SCONJ
bracis-33603	27	7	continued	continue	VERB
bracis-33603	27	8	pretraining	pretraine	VERB
bracis-33603	27	9	enhances	enhance	VERB
bracis-33603	27	10	model	model	NOUN
bracis-33603	27	11	capabilities	capability	NOUN
bracis-33603	27	12	,	,	PUNCT
bracis-33603	27	13	the	the	DET
bracis-33603	27	14	increments	increment	NOUN
bracis-33603	27	15	in	in	ADP
bracis-33603	27	16	performance	performance	NOUN
bracis-33603	27	17	diminish	diminish	NOUN
bracis-33603	27	18	as	as	ADP
bracis-33603	27	19	model	model	NOUN
bracis-33603	27	20	size	size	NOUN
bracis-33603	27	21	increases	increase	NOUN
bracis-33603	27	22	.	.	PUNCT
bracis-33603	28	1	t5	t5	PROPN
bracis-33603	28	2	models	model	NOUN
bracis-33603	28	3	[	[	X
bracis-33603	28	4	34	34	NUM
bracis-33603	28	5	]	]	PUNCT
bracis-33603	28	6	demonstrate	demonstrate	NOUN
bracis-33603	28	7	adaptability	adaptability	NOUN
bracis-33603	28	8	across	across	ADP
bracis-33603	28	9	various	various	ADJ
bracis-33603	28	10	natural	natural	ADJ
bracis-33603	28	11	language	language	NOUN
bracis-33603	28	12	processing	processing	NOUN
bracis-33603	28	13	(	(	PUNCT
bracis-33603	28	14	nlp	nlp	NOUN
bracis-33603	28	15	)	)	PUNCT
bracis-33603	28	16	tasks	task	NOUN
bracis-33603	28	17	due	due	ADP
bracis-33603	28	18	to	to	ADP
bracis-33603	28	19	their	their	PRON
bracis-33603	28	20	encoder	encoder	NOUN
bracis-33603	28	21	-	-	PUNCT
bracis-33603	28	22	decoder	decoder	NOUN
bracis-33603	28	23	architecture	architecture	NOUN
bracis-33603	28	24	.	.	PUNCT
bracis-33603	29	1	this	this	DET
bracis-33603	29	2	structure	structure	NOUN
bracis-33603	29	3	enables	enable	VERB
bracis-33603	29	4	them	they	PRON
bracis-33603	29	5	to	to	PART
bracis-33603	29	6	process	process	VERB
bracis-33603	29	7	text	text	NOUN
bracis-33603	29	8	for	for	ADP
bracis-33603	29	9	both	both	CCONJ
bracis-33603	29	10	understanding	understanding	NOUN
bracis-33603	29	11	and	and	CCONJ
bracis-33603	29	12	generation	generation	NOUN
bracis-33603	29	13	,	,	PUNCT
bracis-33603	29	14	providing	provide	VERB
bracis-33603	29	15	an	an	DET
bracis-33603	29	16	advantage	advantage	NOUN
bracis-33603	29	17	over	over	ADP
bracis-33603	29	18	encoder	encoder	NOUN
bracis-33603	29	19	-	-	PUNCT
bracis-33603	29	20	only	only	ADJ
bracis-33603	29	21	models	model	NOUN
bracis-33603	29	22	like	like	ADP
bracis-33603	29	23	bert	bert	PROPN
bracis-33603	29	24	.	.	PUNCT
bracis-33603	30	1	while	while	SCONJ
bracis-33603	30	2	not	not	PART
bracis-33603	30	3	the	the	DET
bracis-33603	30	4	focus	focus	NOUN
bracis-33603	30	5	of	of	ADP
bracis-33603	30	6	this	this	DET
bracis-33603	30	7	work	work	NOUN
bracis-33603	30	8	,	,	PUNCT
bracis-33603	30	9	t5	t5	PROPN
bracis-33603	30	10	’s	’s	PART
bracis-33603	30	11	adaptability	adaptability	NOUN
bracis-33603	30	12	to	to	ADP
bracis-33603	30	13	instruction	instruction	NOUN
bracis-33603	30	14	-	-	PUNCT
bracis-33603	30	15	based	base	VERB
bracis-33603	30	16	fine	fine	ADJ
bracis-33603	30	17	-	-	PUNCT
bracis-33603	30	18	tuning	tuning	NOUN
bracis-33603	30	19	,	,	PUNCT
bracis-33603	30	20	as	as	SCONJ
bracis-33603	30	21	seen	see	VERB
bracis-33603	30	22	in	in	ADP
bracis-33603	30	23	flan	flan	NOUN
bracis-33603	30	24	-	-	PROPN
bracis-33603	30	25	t5	t5	PROPN
bracis-33603	31	1	[	[	X
bracis-33603	31	2	8	8	NUM
bracis-33603	31	3	]	]	PUNCT
bracis-33603	31	4	,	,	PUNCT
bracis-33603	31	5	also	also	ADV
bracis-33603	31	6	enables	enable	VERB
bracis-33603	31	7	effective	effective	ADJ
bracis-33603	31	8	zero	zero	NUM
bracis-33603	31	9	-	-	PUNCT
bracis-33603	31	10	shot	shot	NOUN
bracis-33603	31	11	and	and	CCONJ
bracis-33603	31	12	few	few	ADJ
bracis-33603	31	13	-	-	PUNCT
bracis-33603	31	14	shot	shot	NOUN
bracis-33603	31	15	applications	application	NOUN
bracis-33603	31	16	.	.	PUNCT
bracis-33603	32	1	these	these	DET
bracis-33603	32	2	factors	factor	NOUN
bracis-33603	32	3	,	,	PUNCT
bracis-33603	32	4	combined	combine	VERB
bracis-33603	32	5	with	with	ADP
bracis-33603	32	6	the	the	DET
bracis-33603	32	7	scarcity	scarcity	NOUN
bracis-33603	32	8	of	of	ADP
bracis-33603	32	9	portuguese	portuguese	PROPN
bracis-33603	32	10	pretrained	pretraine	VERB
bracis-33603	32	11	encoder	encoder	NOUN
bracis-33603	32	12	-	-	PUNCT
bracis-33603	32	13	decoder	decoder	NOUN
bracis-33603	32	14	models	model	NOUN
bracis-33603	32	15	,	,	PUNCT
bracis-33603	32	16	motivate	motivate	VERB
bracis-33603	32	17	our	our	PRON
bracis-33603	32	18	choice	choice	NOUN
bracis-33603	32	19	to	to	PART
bracis-33603	32	20	continue	continue	VERB
bracis-33603	32	21	the	the	DET
bracis-33603	32	22	investigation	investigation	NOUN
bracis-33603	32	23	the	the	DET
bracis-33603	32	24	t5	t5	PROPN
bracis-33603	32	25	architecture	architecture	NOUN
bracis-33603	32	26	in	in	ADP
bracis-33603	32	27	this	this	DET
bracis-33603	32	28	study	study	NOUN
bracis-33603	32	29	.	.	PUNCT
bracis-33603	33	1	2	2	NUM
bracis-33603	33	2	related	related	ADJ
bracis-33603	33	3	work	work	NOUN
bracis-33603	33	4	the	the	DET
bracis-33603	33	5	t5	t5	PROPN
bracis-33603	33	6	model	model	NOUN
bracis-33603	33	7	 	 	SPACE
bracis-33603	34	1	[	[	X
bracis-33603	34	2	34	34	NUM
bracis-33603	34	3	]	]	PUNCT
bracis-33603	34	4	is	be	AUX
bracis-33603	34	5	an	an	DET
bracis-33603	34	6	encoder	encoder	NOUN
bracis-33603	34	7	-	-	PUNCT
bracis-33603	34	8	decoder	decoder	NOUN
bracis-33603	34	9	transformer	transformer	NOUN
bracis-33603	34	10	,	,	PUNCT
bracis-33603	34	11	and	and	CCONJ
bracis-33603	34	12	one	one	NUM
bracis-33603	34	13	of	of	ADP
bracis-33603	34	14	its	its	PRON
bracis-33603	34	15	main	main	ADJ
bracis-33603	34	16	innovations	innovation	NOUN
bracis-33603	34	17	was	be	AUX
bracis-33603	34	18	to	to	PART
bracis-33603	34	19	cast	cast	VERB
bracis-33603	34	20	all	all	DET
bracis-33603	34	21	tasks	task	NOUN
bracis-33603	34	22	into	into	ADP
bracis-33603	34	23	a	a	DET
bracis-33603	34	24	text	text	NOUN
bracis-33603	34	25	-	-	PUNCT
bracis-33603	34	26	to	to	ADP
bracis-33603	34	27	-	-	PUNCT
bracis-33603	34	28	text	text	NOUN
bracis-33603	34	29	format	format	NOUN
bracis-33603	34	30	,	,	PUNCT
bracis-33603	34	31	allowing	allow	VERB
bracis-33603	34	32	for	for	ADP
bracis-33603	34	33	a	a	DET
bracis-33603	34	34	unified	unified	ADJ
bracis-33603	34	35	approach	approach	NOUN
bracis-33603	34	36	;	;	PUNCT
bracis-33603	34	37	scaling	scale	VERB
bracis-33603	34	38	up	up	ADP
bracis-33603	34	39	models	model	NOUN
bracis-33603	34	40	to	to	ADP
bracis-33603	34	41	11b	11b	NOUN
bracis-33603	34	42	parameters	parameter	NOUN
bracis-33603	34	43	,	,	PUNCT
bracis-33603	34	44	they	they	PRON
bracis-33603	34	45	consolidated	consolidate	VERB
bracis-33603	34	46	the	the	DET
bracis-33603	34	47	transfer	transfer	NOUN
bracis-33603	34	48	learning	learning	NOUN
bracis-33603	34	49	approach	approach	NOUN
bracis-33603	34	50	,	,	PUNCT
bracis-33603	34	51	setting	set	VERB
bracis-33603	34	52	new	new	ADJ
bracis-33603	34	53	sotas	sota	NOUN
bracis-33603	34	54	for	for	ADP
bracis-33603	34	55	glue	glue	NOUN
bracis-33603	34	56	[	[	X
bracis-33603	34	57	46	46	NUM
bracis-33603	34	58	]	]	PUNCT
bracis-33603	34	59	,	,	PUNCT
bracis-33603	34	60	superglue	superglue	NOUN
bracis-33603	34	61	[	[	X
bracis-33603	34	62	45	45	NUM
bracis-33603	34	63	]	]	PUNCT
bracis-33603	34	64	,	,	PUNCT
bracis-33603	34	65	cnn	cnn	PROPN
bracis-33603	34	66	/	/	SYM
bracis-33603	34	67	daily	daily	ADJ
bracis-33603	34	68	mail	mail	NOUN
bracis-33603	34	69	[	[	X
bracis-33603	34	70	17	17	NUM
bracis-33603	34	71	]	]	PUNCT
bracis-33603	34	72	benchmarks	benchmark	NOUN
bracis-33603	34	73	.	.	PUNCT
bracis-33603	35	1	it	it	PRON
bracis-33603	35	2	was	be	AUX
bracis-33603	35	3	pretrained	pretraine	VERB
bracis-33603	35	4	using	use	VERB
bracis-33603	35	5	the	the	DET
bracis-33603	35	6	“	"	PUNCT
bracis-33603	35	7	span	span	NOUN
bracis-33603	35	8	corruption	corruption	NOUN
bracis-33603	35	9	”	"	PUNCT
bracis-33603	35	10	objective	objective	NOUN
bracis-33603	35	11	over	over	ADP
bracis-33603	35	12	the	the	DET
bracis-33603	35	13	c4	c4	NOUN
bracis-33603	35	14	(	(	PUNCT
bracis-33603	35	15	“	"	PUNCT
bracis-33603	35	16	colossal	colossal	ADJ
bracis-33603	35	17	clean	clean	ADJ
bracis-33603	35	18	crawled	crawl	VERB
bracis-33603	35	19	corpus	corpus	NOUN
bracis-33603	35	20	”	"	PUNCT
bracis-33603	35	21	)	)	PUNCT
bracis-33603	35	22	dataset	dataset	NOUN
bracis-33603	35	23	,	,	PUNCT
bracis-33603	35	24	where	where	SCONJ
bracis-33603	35	25	random	random	ADJ
bracis-33603	35	26	consecutive	consecutive	ADJ
bracis-33603	35	27	spans	span	NOUN
bracis-33603	35	28	in	in	ADP
bracis-33603	35	29	the	the	DET
bracis-33603	35	30	input	input	NOUN
bracis-33603	35	31	are	be	AUX
bracis-33603	35	32	replaced	replace	VERB
bracis-33603	35	33	by	by	ADP
bracis-33603	35	34	special	special	ADJ
bracis-33603	35	35	mask	mask	NOUN
bracis-33603	35	36	tokens	token	NOUN
bracis-33603	35	37	,	,	PUNCT
bracis-33603	35	38	and	and	CCONJ
bracis-33603	35	39	the	the	DET
bracis-33603	35	40	model	model	NOUN
bracis-33603	35	41	is	be	AUX
bracis-33603	35	42	trained	train	VERB
bracis-33603	35	43	to	to	PART
bracis-33603	35	44	predict	predict	VERB
bracis-33603	35	45	these	these	DET
bracis-33603	35	46	corrupted	corrupted	ADJ
bracis-33603	35	47	tokens	token	NOUN
bracis-33603	35	48	(	(	PUNCT
bracis-33603	35	49	models	model	NOUN
bracis-33603	35	50	with	with	ADP
bracis-33603	35	51	up	up	ADP
bracis-33603	35	52	to	to	ADP
bracis-33603	35	53	11b	11b	NOUN
bracis-33603	35	54	parameters	parameter	NOUN
bracis-33603	35	55	)	)	PUNCT
bracis-33603	35	56	.	.	PUNCT
bracis-33603	36	1	building	build	VERB
bracis-33603	36	2	upon	upon	SCONJ
bracis-33603	36	3	this	this	DET
bracis-33603	36	4	foundation	foundation	NOUN
bracis-33603	36	5	,	,	PUNCT
bracis-33603	36	6	mt5	mt5	ADJ
bracis-33603	36	7	[	[	X
bracis-33603	36	8	48	48	NUM
bracis-33603	36	9	]	]	PUNCT
bracis-33603	36	10	extended	extend	VERB
bracis-33603	36	11	the	the	DET
bracis-33603	36	12	t5	t5	PROPN
bracis-33603	36	13	framework	framework	NOUN
bracis-33603	36	14	to	to	ADP
bracis-33603	36	15	multilingual	multilingual	ADJ
bracis-33603	36	16	settings	setting	NOUN
bracis-33603	36	17	,	,	PUNCT
bracis-33603	36	18	having	having	AUX
bracis-33603	36	19	been	be	AUX
bracis-33603	36	20	pretrained	pretraine	VERB
bracis-33603	36	21	on	on	ADP
bracis-33603	36	22	mc4	mc4	PROPN
bracis-33603	36	23	,	,	PUNCT
bracis-33603	36	24	a	a	DET
bracis-33603	36	25	multilingual	multilingual	ADJ
bracis-33603	36	26	dataset	dataset	NOUN
bracis-33603	36	27	covering	cover	VERB
bracis-33603	36	28	101	101	NUM
bracis-33603	36	29	languages	language	NOUN
bracis-33603	36	30	(	(	PUNCT
bracis-33603	36	31	models	model	NOUN
bracis-33603	36	32	up	up	ADP
bracis-33603	36	33	to	to	ADP
bracis-33603	36	34	13b	13b	NOUN
bracis-33603	36	35	parameters	parameter	NOUN
bracis-33603	36	36	)	)	PUNCT
bracis-33603	36	37	.	.	PUNCT
bracis-33603	37	1	ptt5	ptt5	PROPN
bracis-33603	37	2	[	[	X
bracis-33603	37	3	6	6	NUM
bracis-33603	37	4	]	]	PUNCT
bracis-33603	37	5	further	far	ADV
bracis-33603	37	6	adapted	adapt	VERB
bracis-33603	37	7	t5	t5	PROPN
bracis-33603	37	8	for	for	ADP
bracis-33603	37	9	portuguese	portuguese	NOUN
bracis-33603	37	10	by	by	ADP
bracis-33603	37	11	continuing	continue	VERB
bracis-33603	37	12	the	the	DET
bracis-33603	37	13	pretraining	pretraining	NOUN
bracis-33603	37	14	of	of	ADP
bracis-33603	37	15	t5	t5	PROPN
bracis-33603	37	16	models	model	NOUN
bracis-33603	37	17	on	on	ADP
bracis-33603	37	18	the	the	DET
bracis-33603	37	19	brwac	brwac	NOUN
bracis-33603	37	20	dataset	dataset	NOUN
bracis-33603	37	21	[	[	X
bracis-33603	37	22	44	44	NUM
bracis-33603	37	23	]	]	PUNCT
bracis-33603	37	24	.	.	PUNCT
bracis-33603	38	1	this	this	DET
bracis-33603	38	2	approach	approach	NOUN
bracis-33603	38	3	led	lead	VERB
bracis-33603	38	4	to	to	ADP
bracis-33603	38	5	significant	significant	ADJ
bracis-33603	38	6	improvements	improvement	NOUN
bracis-33603	38	7	in	in	ADP
bracis-33603	38	8	downstream	downstream	ADJ
bracis-33603	38	9	portuguese	portuguese	ADJ
bracis-33603	38	10	language	language	NOUN
bracis-33603	38	11	tasks	task	NOUN
bracis-33603	38	12	,	,	PUNCT
bracis-33603	38	13	which	which	PRON
bracis-33603	38	14	were	be	AUX
bracis-33603	38	15	further	far	ADV
bracis-33603	38	16	enhanced	enhance	VERB
bracis-33603	38	17	by	by	ADP
bracis-33603	38	18	a	a	DET
bracis-33603	38	19	portuguese	portuguese	ADJ
bracis-33603	38	20	tokenizer	tokenizer	NOUN
bracis-33603	38	21	.	.	PUNCT
bracis-33603	39	1	for	for	ADP
bracis-33603	39	2	clarity	clarity	NOUN
bracis-33603	39	3	,	,	PUNCT
bracis-33603	39	4	we	we	PRON
bracis-33603	39	5	’ll	’ll	AUX
bracis-33603	39	6	refer	refer	VERB
bracis-33603	39	7	to	to	ADP
bracis-33603	39	8	the	the	DET
bracis-33603	39	9	work	work	NOUN
bracis-33603	39	10	of	of	ADP
bracis-33603	39	11	carmo	carmo	PROPN
bracis-33603	39	12	et	et	PROPN
bracis-33603	39	13	al	al	PROPN
bracis-33603	39	14	.	.	PROPN
bracis-33603	40	1	as	as	ADP
bracis-33603	40	2	ptt5	ptt5	NOUN
bracis-33603	40	3	-	-	PUNCT
bracis-33603	40	4	v1	v1	NOUN
bracis-33603	40	5	.	.	PUNCT
bracis-33603	41	1	other	other	ADJ
bracis-33603	41	2	notable	notable	ADJ
bracis-33603	41	3	international	international	ADJ
bracis-33603	41	4	adaptations	adaptation	NOUN
bracis-33603	41	5	of	of	ADP
bracis-33603	41	6	t5	t5	PROPN
bracis-33603	41	7	/	/	SYM
bracis-33603	41	8	mt5	mt5	NOUN
bracis-33603	41	9	include	include	VERB
bracis-33603	41	10	it5	it5	PROPN
bracis-33603	41	11	[	[	X
bracis-33603	41	12	40	40	NUM
bracis-33603	41	13	]	]	PUNCT
bracis-33603	41	14	(	(	PUNCT
bracis-33603	41	15	italian	italian	NOUN
bracis-33603	41	16	)	)	PUNCT
bracis-33603	41	17	,	,	PUNCT
bracis-33603	41	18	afriteva	afriteva	VERB
bracis-33603	42	1	[	[	X
bracis-33603	42	2	22	22	NUM
bracis-33603	42	3	]	]	PUNCT
bracis-33603	42	4	(	(	PUNCT
bracis-33603	42	5	low	low	ADJ
bracis-33603	42	6	-	-	PUNCT
bracis-33603	42	7	resource	resource	NOUN
bracis-33603	42	8	african	african	ADJ
bracis-33603	42	9	languages	language	NOUN
bracis-33603	42	10	)	)	PUNCT
bracis-33603	42	11	,	,	PUNCT
bracis-33603	42	12	arat5	arat5	PROPN
bracis-33603	43	1	[	[	X
bracis-33603	43	2	29	29	NUM
bracis-33603	43	3	]	]	X
bracis-33603	43	4	(	(	PUNCT
bracis-33603	43	5	arabic	arabic	ADJ
bracis-33603	43	6	)	)	PUNCT
bracis-33603	43	7	,	,	PUNCT
bracis-33603	43	8	and	and	CCONJ
bracis-33603	43	9	plt5	plt5	PROPN
bracis-33603	44	1	[	[	X
bracis-33603	44	2	7	7	X
bracis-33603	44	3	]	]	X
bracis-33603	44	4	(	(	PUNCT
bracis-33603	44	5	polish	polish	PROPN
bracis-33603	44	6	)	)	PUNCT
bracis-33603	44	7	.	.	PUNCT
bracis-33603	45	1	bertimbau	bertimbau	PROPN
bracis-33603	46	1	[	[	X
bracis-33603	46	2	42	42	NUM
bracis-33603	46	3	]	]	PUNCT
bracis-33603	46	4	,	,	PUNCT
bracis-33603	46	5	a	a	DET
bracis-33603	46	6	popular	popular	ADJ
bracis-33603	46	7	adaptation	adaptation	NOUN
bracis-33603	46	8	of	of	ADP
bracis-33603	46	9	the	the	DET
bracis-33603	46	10	bert	bert	PROPN
bracis-33603	46	11	encoder	encoder	NOUN
bracis-33603	46	12	model	model	NOUN
bracis-33603	46	13	,	,	PUNCT
bracis-33603	46	14	remains	remain	VERB
bracis-33603	46	15	influential	influential	ADJ
bracis-33603	46	16	within	within	ADP
bracis-33603	46	17	portuguese	portuguese	ADJ
bracis-33603	46	18	language	language	NOUN
bracis-33603	46	19	modeling	modeling	NOUN
bracis-33603	46	20	.	.	PUNCT
bracis-33603	47	1	others	other	NOUN
bracis-33603	47	2	exploring	explore	VERB
bracis-33603	47	3	encoder	encoder	NOUN
bracis-33603	47	4	architectures	architecture	NOUN
bracis-33603	47	5	include	include	VERB
bracis-33603	47	6	albertina	albertina	PROPN
bracis-33603	47	7	[	[	X
bracis-33603	47	8	37	37	NUM
bracis-33603	47	9	]	]	PUNCT
bracis-33603	47	10	,	,	PUNCT
bracis-33603	47	11	debertinha	debertinha	NOUN
bracis-33603	47	12	[	[	X
bracis-33603	47	13	5	5	NUM
bracis-33603	47	14	]	]	PUNCT
bracis-33603	47	15	,	,	PUNCT
bracis-33603	47	16	the	the	DET
bracis-33603	47	17	work	work	NOUN
bracis-33603	47	18	of	of	ADP
bracis-33603	47	19	gomes	gome	NOUN
bracis-33603	47	20	et	et	PROPN
bracis-33603	47	21	al	al	PROPN
bracis-33603	47	22	.	.	PUNCT
bracis-33603	48	1	[	[	X
bracis-33603	48	2	16	16	NUM
bracis-33603	48	3	]	]	X
bracis-33603	48	4	(	(	PUNCT
bracis-33603	48	5	which	which	PRON
bracis-33603	48	6	pretrains	pretrain	VERB
bracis-33603	48	7	a	a	DET
bracis-33603	48	8	roberta	roberta	PROPN
bracis-33603	48	9	model	model	PROPN
bracis-33603	48	10	)	)	PUNCT
bracis-33603	48	11	,	,	PUNCT
bracis-33603	48	12	and	and	CCONJ
bracis-33603	48	13	de	de	X
bracis-33603	48	14	morais	morais	PROPN
bracis-33603	48	15	et	et	PROPN
bracis-33603	48	16	al	al	PROPN
bracis-33603	48	17	.	.	PUNCT
bracis-33603	49	1	[	[	X
bracis-33603	49	2	28	28	NUM
bracis-33603	49	3	]	]	PUNCT
bracis-33603	49	4	.	.	PUNCT
bracis-33603	50	1	reflecting	reflect	VERB
bracis-33603	50	2	a	a	DET
bracis-33603	50	3	broader	broad	ADJ
bracis-33603	50	4	trend	trend	NOUN
bracis-33603	50	5	,	,	PUNCT
bracis-33603	50	6	numerous	numerous	ADJ
bracis-33603	50	7	recent	recent	ADJ
bracis-33603	50	8	portuguese	portuguese	ADJ
bracis-33603	50	9	models	model	NOUN
bracis-33603	50	10	prioritize	prioritize	VERB
bracis-33603	50	11	decoder	decoder	NOUN
bracis-33603	50	12	-	-	PUNCT
bracis-33603	50	13	only	only	ADJ
bracis-33603	50	14	architectures	architecture	NOUN
bracis-33603	50	15	,	,	PUNCT
bracis-33603	50	16	such	such	ADJ
bracis-33603	50	17	as	as	ADP
bracis-33603	50	18	sabiá	sabiá	NOUN
bracis-33603	50	19	[	[	X
bracis-33603	50	20	32	32	NUM
bracis-33603	50	21	]	]	PUNCT
bracis-33603	50	22	,	,	PUNCT
bracis-33603	50	23	glória	glória	PROPN
bracis-33603	51	1	[	[	X
bracis-33603	51	2	27	27	NUM
bracis-33603	51	3	]	]	PUNCT
bracis-33603	51	4	,	,	PUNCT
bracis-33603	51	5	bode	bode	VERB
bracis-33603	51	6	[	[	X
bracis-33603	51	7	15	15	NUM
bracis-33603	51	8	]	]	PUNCT
bracis-33603	51	9	,	,	PUNCT
bracis-33603	51	10	cabrita	cabrita	NOUN
bracis-33603	52	1	[	[	X
bracis-33603	52	2	25	25	NUM
bracis-33603	52	3	]	]	PUNCT
bracis-33603	52	4	,	,	PUNCT
bracis-33603	53	1	and	and	CCONJ
bracis-33603	53	2	gervásio	gervásio	VERB
bracis-33603	53	3	[	[	X
bracis-33603	53	4	39	39	NUM
bracis-33603	53	5	]	]	PUNCT
bracis-33603	53	6	.	.	PUNCT
bracis-33603	54	1	in	in	ADP
bracis-33603	54	2	the	the	DET
bracis-33603	54	3	encoder	encoder	NOUN
bracis-33603	54	4	-	-	PUNCT
bracis-33603	54	5	decoder	decoder	NOUN
bracis-33603	54	6	space	space	NOUN
bracis-33603	54	7	,	,	PUNCT
bracis-33603	54	8	the	the	DET
bracis-33603	54	9	work	work	NOUN
bracis-33603	54	10	of	of	ADP
bracis-33603	54	11	carmo	carmo	PROPN
bracis-33603	54	12	et	et	PROPN
bracis-33603	54	13	al	al	PROPN
bracis-33603	54	14	.	.	PUNCT
bracis-33603	55	1	(	(	PUNCT
bracis-33603	55	2	ptt5	ptt5	NOUN
bracis-33603	55	3	-	-	PUNCT
bracis-33603	55	4	v1	v1	NOUN
bracis-33603	55	5	)	)	PUNCT
bracis-33603	55	6	explored	explore	VERB
bracis-33603	55	7	adapting	adapt	VERB
bracis-33603	55	8	t5	t5	PROPN
bracis-33603	55	9	models	model	NOUN
bracis-33603	55	10	for	for	ADP
bracis-33603	55	11	portuguese	portuguese	PROPN
bracis-33603	55	12	.	.	PUNCT
bracis-33603	56	1	beyond	beyond	ADP
bracis-33603	56	2	generic	generic	ADJ
bracis-33603	56	3	portuguese	portuguese	ADJ
bracis-33603	56	4	models	model	NOUN
bracis-33603	56	5	,	,	PUNCT
bracis-33603	56	6	several	several	ADJ
bracis-33603	56	7	works	work	NOUN
bracis-33603	56	8	specialize	specialize	VERB
bracis-33603	56	9	in	in	ADP
bracis-33603	56	10	custom	custom	NOUN
bracis-33603	56	11	domains	domain	NOUN
bracis-33603	56	12	:	:	PUNCT
bracis-33603	56	13	de	de	X
bracis-33603	56	14	barros	barros	X
bracis-33603	56	15	et	et	PROPN
bracis-33603	56	16	al	al	PROPN
bracis-33603	56	17	.	.	PUNCT
bracis-33603	57	1	[	[	X
bracis-33603	57	2	2	2	NUM
bracis-33603	57	3	]	]	PUNCT
bracis-33603	57	4	and	and	CCONJ
bracis-33603	57	5	bertabaporu	bertabaporu	NOUN
bracis-33603	58	1	[	[	X
bracis-33603	58	2	10	10	NUM
bracis-33603	58	3	]	]	PUNCT
bracis-33603	58	4	were	be	AUX
bracis-33603	58	5	designed	design	VERB
bracis-33603	58	6	for	for	ADP
bracis-33603	58	7	portuguese	portuguese	ADJ
bracis-33603	58	8	social	social	ADJ
bracis-33603	58	9	media	medium	NOUN
bracis-33603	58	10	data	datum	NOUN
bracis-33603	58	11	,	,	PUNCT
bracis-33603	58	12	while	while	SCONJ
bracis-33603	58	13	bertaú	bertaú	PROPN
bracis-33603	59	1	[	[	X
bracis-33603	59	2	13	13	NUM
bracis-33603	59	3	]	]	PUNCT
bracis-33603	59	4	focuses	focus	VERB
bracis-33603	59	5	on	on	ADP
bracis-33603	59	6	financial	financial	ADJ
bracis-33603	59	7	language	language	NOUN
bracis-33603	59	8	.	.	PUNCT
bracis-33603	60	1	3	3	NUM
bracis-33603	60	2	methodology	methodology	NOUN
bracis-33603	60	3	this	this	DET
bracis-33603	60	4	section	section	NOUN
bracis-33603	60	5	describes	describe	VERB
bracis-33603	60	6	the	the	DET
bracis-33603	60	7	methodology	methodology	NOUN
bracis-33603	60	8	for	for	ADP
bracis-33603	60	9	pretraining	pretraine	VERB
bracis-33603	60	10	and	and	CCONJ
bracis-33603	60	11	evaluating	evaluate	VERB
bracis-33603	60	12	our	our	PRON
bracis-33603	60	13	key	key	ADJ
bracis-33603	60	14	experiments	experiment	NOUN
bracis-33603	60	15	,	,	PUNCT
bracis-33603	60	16	covering	cover	VERB
bracis-33603	60	17	the	the	DET
bracis-33603	60	18	pretraining	pretraine	VERB
bracis-33603	60	19	dataset	dataset	NOUN
bracis-33603	60	20	,	,	PUNCT
bracis-33603	60	21	language	language	NOUN
bracis-33603	60	22	-	-	PUNCT
bracis-33603	60	23	specific	specific	ADJ
bracis-33603	60	24	vocabulary	vocabulary	NOUN
bracis-33603	60	25	,	,	PUNCT
bracis-33603	60	26	model	model	NOUN
bracis-33603	60	27	architectures	architecture	NOUN
bracis-33603	60	28	,	,	PUNCT
bracis-33603	60	29	optimization	optimization	NOUN
bracis-33603	60	30	strategies	strategy	NOUN
bracis-33603	60	31	,	,	PUNCT
bracis-33603	60	32	and	and	CCONJ
bracis-33603	60	33	finetuning	finetuning	NOUN
bracis-33603	60	34	and	and	CCONJ
bracis-33603	60	35	validation	validation	NOUN
bracis-33603	60	36	processes	process	NOUN
bracis-33603	60	37	for	for	ADP
bracis-33603	60	38	downstream	downstream	ADJ
bracis-33603	60	39	tasks	task	NOUN
bracis-33603	60	40	.	.	PUNCT
bracis-33603	61	1	3.1	3.1	NUM
bracis-33603	61	2	unsupervised	unsupervised	ADJ
bracis-33603	61	3	continued	continued	ADJ
bracis-33603	61	4	pretraining	pretraine	VERB
bracis-33603	61	5	as	as	ADP
bracis-33603	61	6	the	the	DET
bracis-33603	61	7	pretraining	pretraine	VERB
bracis-33603	61	8	data	datum	NOUN
bracis-33603	61	9	,	,	PUNCT
bracis-33603	61	10	we	we	PRON
bracis-33603	61	11	utilized	utilize	VERB
bracis-33603	61	12	the	the	DET
bracis-33603	61	13	portuguese	portuguese	ADJ
bracis-33603	61	14	segment	segment	NOUN
bracis-33603	61	15	of	of	ADP
bracis-33603	61	16	the	the	DET
bracis-33603	61	17	mc4	mc4	PROPN
bracis-33603	61	18	dataset	dataset	NOUN
bracis-33603	61	19	(	(	PUNCT
bracis-33603	61	20	hereafter	hereafter	ADV
bracis-33603	61	21	referred	refer	VERB
bracis-33603	61	22	to	to	ADP
bracis-33603	61	23	as	as	ADP
bracis-33603	61	24	mc4	mc4	PROPN
bracis-33603	61	25	-	-	PUNCT
bracis-33603	61	26	pt	pt	NOUN
bracis-33603	61	27	)	)	PUNCT
bracis-33603	61	28	,	,	PUNCT
bracis-33603	61	29	comprising	comprise	VERB
bracis-33603	61	30	approximately	approximately	ADV
bracis-33603	61	31	524	524	NUM
bracis-33603	61	32	 	 	SPACE
bracis-33603	61	33	gb	gb	NOUN
bracis-33603	61	34	of	of	ADP
bracis-33603	61	35	uncompressed	uncompress	VERB
bracis-33603	61	36	text	text	NOUN
bracis-33603	61	37	across	across	ADP
bracis-33603	61	38	169	169	NUM
bracis-33603	61	39	million	million	NUM
bracis-33603	61	40	documents	document	NOUN
bracis-33603	61	41	.	.	PUNCT
bracis-33603	62	1	this	this	DET
bracis-33603	62	2	dataset	dataset	NOUN
bracis-33603	62	3	is	be	AUX
bracis-33603	62	4	significantly	significantly	ADV
bracis-33603	62	5	larger	large	ADJ
bracis-33603	62	6	than	than	ADP
bracis-33603	62	7	the	the	DET
bracis-33603	62	8	one	one	NOUN
bracis-33603	62	9	used	use	VERB
bracis-33603	62	10	for	for	ADP
bracis-33603	62	11	the	the	DET
bracis-33603	62	12	pretraining	pretraining	NOUN
bracis-33603	62	13	of	of	ADP
bracis-33603	62	14	ptt5	ptt5	NOUN
bracis-33603	62	15	-	-	PUNCT
bracis-33603	62	16	v1	v1	NOUN
bracis-33603	62	17	models	model	NOUN
bracis-33603	62	18	,	,	PUNCT
bracis-33603	62	19	which	which	PRON
bracis-33603	62	20	originated	originate	VERB
bracis-33603	62	21	from	from	ADP
bracis-33603	62	22	the	the	DET
bracis-33603	62	23	brwac	brwac	NOUN
bracis-33603	62	24	dataset	dataset	NOUN
bracis-33603	62	25	[	[	X
bracis-33603	62	26	44	44	NUM
bracis-33603	62	27	]	]	PUNCT
bracis-33603	62	28	and	and	CCONJ
bracis-33603	62	29	consisted	consist	VERB
bracis-33603	62	30	of	of	ADP
bracis-33603	62	31	around	around	ADP
bracis-33603	62	32	15	15	NUM
bracis-33603	62	33	 	 	SPACE
bracis-33603	62	34	gb	gb	NOUN
bracis-33603	62	35	of	of	ADP
bracis-33603	62	36	text	text	NOUN
bracis-33603	62	37	from	from	ADP
bracis-33603	62	38	7.4	7.4	NUM
bracis-33603	62	39	million	million	NUM
bracis-33603	62	40	documents	document	NOUN
bracis-33603	62	41	after	after	ADP
bracis-33603	62	42	preprocessing	preprocesse	VERB
bracis-33603	62	43	.	.	PUNCT
bracis-33603	63	1	we	we	PRON
bracis-33603	63	2	adopted	adopt	VERB
bracis-33603	63	3	the	the	DET
bracis-33603	63	4	portuguese	portuguese	ADJ
bracis-33603	63	5	language	language	NOUN
bracis-33603	63	6	vocabulary	vocabulary	NOUN
bracis-33603	63	7	from	from	ADP
bracis-33603	63	8	ptt5	ptt5	NOUN
bracis-33603	63	9	-	-	PUNCT
bracis-33603	63	10	v1	v1	NOUN
bracis-33603	63	11	.	.	PUNCT
bracis-33603	64	1	this	this	DET
bracis-33603	64	2	sentencepiece	sentencepiece	PROPN
bracis-33603	64	3	unigram	unigram	NOUN
bracis-33603	64	4	tokenizer	tokenizer	NOUN
bracis-33603	64	5	[	[	X
bracis-33603	64	6	24	24	NUM
bracis-33603	64	7	]	]	PUNCT
bracis-33603	64	8	,	,	PUNCT
bracis-33603	64	9	comprising	comprise	VERB
bracis-33603	64	10	32,000	32,000	NUM
bracis-33603	64	11	tokens	token	NOUN
bracis-33603	64	12	,	,	PUNCT
bracis-33603	64	13	was	be	AUX
bracis-33603	64	14	trained	train	VERB
bracis-33603	64	15	over	over	ADP
bracis-33603	64	16	a	a	DET
bracis-33603	64	17	corpus	corpus	NOUN
bracis-33603	64	18	of	of	ADP
bracis-33603	64	19	2	2	NUM
bracis-33603	64	20	million	million	NUM
bracis-33603	64	21	documents	document	NOUN
bracis-33603	64	22	from	from	ADP
bracis-33603	64	23	the	the	DET
bracis-33603	64	24	portuguese	portuguese	PROPN
bracis-33603	64	25	wikipedia	wikipedia	PROPN
bracis-33603	64	26	.	.	PUNCT
bracis-33603	65	1	this	this	DET
bracis-33603	65	2	vocabulary	vocabulary	ADJ
bracis-33603	65	3	shares	share	NOUN
bracis-33603	65	4	the	the	DET
bracis-33603	65	5	same	same	ADJ
bracis-33603	65	6	number	number	NOUN
bracis-33603	65	7	of	of	ADP
bracis-33603	65	8	tokens	token	NOUN
bracis-33603	65	9	and	and	CCONJ
bracis-33603	65	10	control	control	NOUN
bracis-33603	65	11	tokens	token	NOUN
bracis-33603	65	12	as	as	ADP
bracis-33603	65	13	t5	t5	PROPN
bracis-33603	65	14	,	,	PUNCT
bracis-33603	65	15	facilitating	facilitate	VERB
bracis-33603	65	16	the	the	DET
bracis-33603	65	17	direct	direct	ADJ
bracis-33603	65	18	use	use	NOUN
bracis-33603	65	19	of	of	ADP
bracis-33603	65	20	google	google	PROPN
bracis-33603	65	21	’s	’s	PART
bracis-33603	65	22	model	model	NOUN
bracis-33603	65	23	checkpoints	checkpoint	NOUN
bracis-33603	65	24	.	.	PUNCT
bracis-33603	66	1	as	as	ADP
bracis-33603	66	2	the	the	DET
bracis-33603	66	3	pretraining	pretraine	VERB
bracis-33603	66	4	objective	objective	NOUN
bracis-33603	66	5	,	,	PUNCT
bracis-33603	66	6	the	the	DET
bracis-33603	66	7	span	span	NOUN
bracis-33603	66	8	corruption	corruption	NOUN
bracis-33603	66	9	task	task	NOUN
bracis-33603	66	10	was	be	AUX
bracis-33603	66	11	employed	employ	VERB
bracis-33603	66	12	,	,	PUNCT
bracis-33603	66	13	utilizing	utilize	VERB
bracis-33603	66	14	batches	batch	NOUN
bracis-33603	66	15	of	of	ADP
bracis-33603	66	16	128	128	NUM
bracis-33603	66	17	sequences	sequence	NOUN
bracis-33603	66	18	of	of	ADP
bracis-33603	66	19	512	512	NUM
bracis-33603	66	20	tokens	token	NOUN
bracis-33603	66	21	(	(	PUNCT
bracis-33603	66	22	65,536	65,536	NUM
bracis-33603	66	23	tokens	token	NOUN
bracis-33603	66	24	)	)	PUNCT
bracis-33603	66	25	a	a	DET
bracis-33603	66	26	methodology	methodology	NOUN
bracis-33603	66	27	consistent	consistent	ADJ
bracis-33603	66	28	with	with	ADP
bracis-33603	66	29	the	the	DET
bracis-33603	66	30	baseline	baseline	ADJ
bracis-33603	66	31	experiment	experiment	NOUN
bracis-33603	66	32	by	by	ADP
bracis-33603	66	33	raffel	raffel	NOUN
bracis-33603	66	34	et	et	PROPN
bracis-33603	66	35	al	al	PROPN
bracis-33603	66	36	.	.	PUNCT
bracis-33603	67	1	[	[	X
bracis-33603	67	2	34	34	NUM
bracis-33603	67	3	]	]	PUNCT
bracis-33603	67	4	.	.	PUNCT
bracis-33603	68	1	adafactor	adafactor	NOUN
bracis-33603	68	2	optimizer	optimizer	NOUN
bracis-33603	68	3	[	[	X
bracis-33603	68	4	41	41	NUM
bracis-33603	68	5	]	]	PUNCT
bracis-33603	68	6	with	with	ADP
bracis-33603	68	7	a	a	DET
bracis-33603	68	8	constant	constant	ADJ
bracis-33603	68	9	learning	learning	NOUN
bracis-33603	68	10	rate	rate	NOUN
bracis-33603	68	11	of	of	ADP
bracis-33603	68	12	0.001	0.001	NUM
bracis-33603	68	13	and	and	CCONJ
bracis-33603	68	14	cross	cross	NOUN
bracis-33603	68	15	-	-	NOUN
bracis-33603	68	16	entropy	entropy	NOUN
bracis-33603	68	17	as	as	SCONJ
bracis-33603	68	18	loss	loss	NOUN
bracis-33603	68	19	was	be	AUX
bracis-33603	68	20	used	use	VERB
bracis-33603	68	21	during	during	ADP
bracis-33603	68	22	the	the	DET
bracis-33603	68	23	entire	entire	ADJ
bracis-33603	68	24	pretraining	pretraine	VERB
bracis-33603	68	25	process	process	NOUN
bracis-33603	68	26	.	.	PUNCT
bracis-33603	69	1	using	use	VERB
bracis-33603	69	2	these	these	DET
bracis-33603	69	3	experimental	experimental	ADJ
bracis-33603	69	4	settings	setting	NOUN
bracis-33603	69	5	,	,	PUNCT
bracis-33603	69	6	we	we	PRON
bracis-33603	69	7	started	start	VERB
bracis-33603	69	8	from	from	ADP
bracis-33603	69	9	google	google	PROPN
bracis-33603	69	10	’s	’s	PART
bracis-33603	69	11	original	original	ADJ
bracis-33603	69	12	checkpoints	checkpoint	NOUN
bracis-33603	69	13	with	with	ADP
bracis-33603	69	14	sizes	size	NOUN
bracis-33603	69	15	from	from	ADP
bracis-33603	69	16	t5	t5	PROPN
bracis-33603	69	17	-	-	PUNCT
bracis-33603	69	18	small	small	ADJ
bracis-33603	69	19	(	(	PUNCT
bracis-33603	69	20	60	60	NUM
bracis-33603	69	21	m	m	PROPN
bracis-33603	69	22	parameters	parameter	NOUN
bracis-33603	69	23	)	)	PUNCT
bracis-33603	69	24	up	up	ADP
bracis-33603	69	25	to	to	ADP
bracis-33603	69	26	t5	t5	PROPN
bracis-33603	69	27	-	-	PUNCT
bracis-33603	69	28	3b	3b	NOUN
bracis-33603	69	29	(	(	PUNCT
bracis-33603	69	30	3b	3b	NUM
bracis-33603	69	31	parameters	parameter	NOUN
bracis-33603	69	32	)	)	PUNCT
bracis-33603	69	33	,	,	PUNCT
bracis-33603	69	34	and	and	CCONJ
bracis-33603	69	35	performed	perform	VERB
bracis-33603	69	36	a	a	DET
bracis-33603	69	37	complete	complete	ADJ
bracis-33603	69	38	epoch	epoch	NOUN
bracis-33603	69	39	of	of	ADP
bracis-33603	69	40	continued	continue	VERB
bracis-33603	69	41	pretraining	pretraine	VERB
bracis-33603	69	42	over	over	ADP
bracis-33603	69	43	the	the	DET
bracis-33603	69	44	mc4	mc4	PROPN
bracis-33603	69	45	-	-	PUNCT
bracis-33603	69	46	pt	pt	NOUN
bracis-33603	69	47	dataset	dataset	NOUN
bracis-33603	69	48	.	.	PUNCT
bracis-33603	70	1	considering	consider	VERB
bracis-33603	70	2	these	these	DET
bracis-33603	70	3	settings	setting	NOUN
bracis-33603	70	4	,	,	PUNCT
bracis-33603	70	5	a	a	DET
bracis-33603	70	6	single	single	ADJ
bracis-33603	70	7	epoch	epoch	NOUN
bracis-33603	70	8	over	over	ADP
bracis-33603	70	9	the	the	DET
bracis-33603	70	10	mc4	mc4	PROPN
bracis-33603	70	11	-	-	PUNCT
bracis-33603	70	12	pt	pt	NOUN
bracis-33603	70	13	dataset	dataset	NOUN
bracis-33603	70	14	comprises	comprise	VERB
bracis-33603	70	15	approximately	approximately	ADV
bracis-33603	70	16	1,764,515	1,764,515	NUM
bracis-33603	70	17	training	training	NOUN
bracis-33603	70	18	steps	step	NOUN
bracis-33603	70	19	and	and	CCONJ
bracis-33603	70	20	116	116	NUM
bracis-33603	70	21	billion	billion	NUM
bracis-33603	70	22	training	training	NOUN
bracis-33603	70	23	tokens	token	NOUN
bracis-33603	70	24	.	.	PUNCT
bracis-33603	71	1	additional	additional	ADJ
bracis-33603	71	2	pretraining	pretraine	VERB
bracis-33603	71	3	experiments	experiment	NOUN
bracis-33603	71	4	are	be	AUX
bracis-33603	71	5	detailed	detail	VERB
bracis-33603	71	6	in	in	ADP
bracis-33603	71	7	sect	sect	NOUN
bracis-33603	71	8	.	.	PUNCT
bracis-33603	71	9	 	 	SPACE
bracis-33603	72	1	5.1	5.1	NUM
bracis-33603	72	2	.	.	PUNCT
bracis-33603	73	1	both	both	PRON
bracis-33603	73	2	pretraining	pretraine	VERB
bracis-33603	73	3	and	and	CCONJ
bracis-33603	73	4	finetuning	finetune	VERB
bracis-33603	73	5	experiments	experiment	NOUN
bracis-33603	73	6	utilized	utilize	VERB
bracis-33603	73	7	tpuv2	tpuv2	PROPN
bracis-33603	73	8	-	-	PUNCT
bracis-33603	73	9	8	8	NUM
bracis-33603	73	10	and	and	CCONJ
bracis-33603	73	11	tpuv3	tpuv3	NOUN
bracis-33603	73	12	-	-	PUNCT
bracis-33603	73	13	8	8	NUM
bracis-33603	73	14	devices	device	NOUN
bracis-33603	73	15	,	,	PUNCT
bracis-33603	73	16	leveraging	leverage	VERB
bracis-33603	73	17	t5	t5	PROPN
bracis-33603	74	1	[	[	X
bracis-33603	74	2	34	34	NUM
bracis-33603	74	3	]	]	PUNCT
bracis-33603	74	4	and	and	CCONJ
bracis-33603	74	5	seqio	seqio	NOUN
bracis-33603	74	6	[	[	X
bracis-33603	74	7	36	36	NUM
bracis-33603	74	8	]	]	PUNCT
bracis-33603	74	9	frameworks	framework	NOUN
bracis-33603	74	10	.	.	PUNCT
bracis-33603	75	1	3.2	3.2	NUM
bracis-33603	75	2	supervised	supervised	ADJ
bracis-33603	75	3	finetuning	finetune	VERB
bracis-33603	75	4	on	on	ADP
bracis-33603	75	5	 	 	SPACE
bracis-33603	75	6	downstream	downstream	ADJ
bracis-33603	75	7	tasks	task	NOUN
bracis-33603	75	8	we	we	PRON
bracis-33603	75	9	assess	assess	VERB
bracis-33603	75	10	the	the	DET
bracis-33603	75	11	impact	impact	NOUN
bracis-33603	75	12	of	of	ADP
bracis-33603	75	13	our	our	PRON
bracis-33603	75	14	pretraining	pretraine	VERB
bracis-33603	75	15	on	on	ADP
bracis-33603	75	16	three	three	NUM
bracis-33603	75	17	portuguese	portuguese	ADJ
bracis-33603	75	18	language	language	NOUN
bracis-33603	75	19	downstream	downstream	ADJ
bracis-33603	75	20	tasks	task	NOUN
bracis-33603	75	21	:	:	PUNCT
bracis-33603	75	22	assin2	assin2	PROPN
bracis-33603	75	23	rte	rte	NOUN
bracis-33603	75	24	,	,	PUNCT
bracis-33603	75	25	assin2	assin2	PROPN
bracis-33603	75	26	sts	st	NOUN
bracis-33603	75	27	,	,	PUNCT
bracis-33603	75	28	and	and	CCONJ
bracis-33603	75	29	tweetsentbr	tweetsentbr	NOUN
bracis-33603	75	30	.	.	PUNCT
bracis-33603	76	1	the	the	DET
bracis-33603	76	2	assin2	assin2	PROPN
bracis-33603	76	3	dataset	dataset	VERB
bracis-33603	77	1	[	[	PUNCT
bracis-33603	77	2	31	31	NUM
bracis-33603	77	3	]	]	PUNCT
bracis-33603	77	4	provides	provide	VERB
bracis-33603	77	5	two	two	NUM
bracis-33603	77	6	tasks	task	NOUN
bracis-33603	77	7	:	:	PUNCT
bracis-33603	77	8	rte	rte	NOUN
bracis-33603	77	9	(	(	PUNCT
bracis-33603	77	10	recognizing	recognize	VERB
bracis-33603	77	11	textual	textual	ADJ
bracis-33603	77	12	entailment	entailment	NOUN
bracis-33603	77	13	)	)	PUNCT
bracis-33603	77	14	,	,	PUNCT
bracis-33603	77	15	which	which	PRON
bracis-33603	77	16	involves	involve	VERB
bracis-33603	77	17	determining	determine	VERB
bracis-33603	77	18	whether	whether	SCONJ
bracis-33603	77	19	one	one	NUM
bracis-33603	77	20	sentence	sentence	NOUN
bracis-33603	77	21	entails	entail	VERB
bracis-33603	77	22	another	another	PRON
bracis-33603	77	23	,	,	PUNCT
bracis-33603	77	24	and	and	CCONJ
bracis-33603	77	25	sts	st	NOUN
bracis-33603	77	26	(	(	PUNCT
bracis-33603	77	27	semantic	semantic	ADJ
bracis-33603	77	28	textual	textual	ADJ
bracis-33603	77	29	similarity	similarity	NOUN
bracis-33603	77	30	)	)	PUNCT
bracis-33603	77	31	,	,	PUNCT
bracis-33603	77	32	which	which	PRON
bracis-33603	77	33	quantifies	quantify	VERB
bracis-33603	77	34	the	the	DET
bracis-33603	77	35	semantic	semantic	ADJ
bracis-33603	77	36	similarity	similarity	NOUN
bracis-33603	77	37	between	between	ADP
bracis-33603	77	38	sentence	sentence	NOUN
bracis-33603	77	39	pairs	pair	NOUN
bracis-33603	77	40	on	on	ADP
bracis-33603	77	41	a	a	DET
bracis-33603	77	42	1–5	1–5	NUM
bracis-33603	77	43	scale	scale	NOUN
bracis-33603	77	44	.	.	PUNCT
bracis-33603	78	1	the	the	DET
bracis-33603	78	2	tweetsentbr	tweetsentbr	NOUN
bracis-33603	78	3	dataset	dataset	VERB
bracis-33603	78	4	[	[	PUNCT
bracis-33603	78	5	4	4	NUM
bracis-33603	78	6	]	]	PUNCT
bracis-33603	78	7	is	be	AUX
bracis-33603	78	8	a	a	DET
bracis-33603	78	9	sentiment	sentiment	NOUN
bracis-33603	78	10	analysis	analysis	NOUN
bracis-33603	78	11	task	task	NOUN
bracis-33603	78	12	for	for	ADP
bracis-33603	78	13	brazilian	brazilian	ADJ
bracis-33603	78	14	portuguese	portuguese	ADJ
bracis-33603	78	15	tweets	tweet	NOUN
bracis-33603	78	16	,	,	PUNCT
bracis-33603	78	17	classifying	classify	VERB
bracis-33603	78	18	them	they	PRON
bracis-33603	78	19	as	as	ADP
bracis-33603	78	20	positive	positive	ADJ
bracis-33603	78	21	,	,	PUNCT
bracis-33603	78	22	negative	negative	ADJ
bracis-33603	78	23	,	,	PUNCT
bracis-33603	78	24	or	or	CCONJ
bracis-33603	78	25	neutral	neutral	ADJ
bracis-33603	78	26	.	.	PUNCT
bracis-33603	79	1	tables	table	NOUN
bracis-33603	79	2	 	 	SPACE
bracis-33603	79	3	1	1	NUM
bracis-33603	79	4	and	and	CCONJ
bracis-33603	79	5	2	2	NUM
bracis-33603	79	6	show	show	VERB
bracis-33603	79	7	further	further	ADJ
bracis-33603	79	8	details	detail	NOUN
bracis-33603	79	9	and	and	CCONJ
bracis-33603	79	10	examples	example	NOUN
bracis-33603	79	11	for	for	ADP
bracis-33603	79	12	each	each	DET
bracis-33603	79	13	task	task	NOUN
bracis-33603	79	14	.	.	PUNCT
bracis-33603	79	15	table	table	NOUN
bracis-33603	80	1	1	1	NUM
bracis-33603	80	2	.	.	PUNCT
bracis-33603	80	3	downstream	downstream	ADJ
bracis-33603	80	4	tasks	task	NOUN
bracis-33603	80	5	datasets.full	datasets.full	NUM
bracis-33603	80	6	size	size	NOUN
bracis-33603	80	7	table	table	NOUN
bracis-33603	80	8	table	table	NOUN
bracis-33603	80	9	2	2	NUM
bracis-33603	80	10	.	.	PUNCT
bracis-33603	80	11	downstream	downstream	ADJ
bracis-33603	80	12	tasks	task	NOUN
bracis-33603	80	13	inputs	input	NOUN
bracis-33603	80	14	and	and	CCONJ
bracis-33603	80	15	targetsfull	targetsfull	ADJ
bracis-33603	80	16	size	size	NOUN
bracis-33603	80	17	table	table	NOUN
bracis-33603	80	18	we	we	PRON
bracis-33603	80	19	finetuned	finetune	VERB
bracis-33603	80	20	the	the	DET
bracis-33603	80	21	pretrained	pretraine	VERB
bracis-33603	80	22	models	model	NOUN
bracis-33603	80	23	over	over	ADP
bracis-33603	80	24	100	100	NUM
bracis-33603	80	25	epochs	epoch	NOUN
bracis-33603	80	26	with	with	ADP
bracis-33603	80	27	batches	batch	NOUN
bracis-33603	80	28	of	of	ADP
bracis-33603	80	29	128	128	NUM
bracis-33603	80	30	sequences	sequence	NOUN
bracis-33603	80	31	and	and	CCONJ
bracis-33603	80	32	a	a	DET
bracis-33603	80	33	maximum	maximum	ADJ
bracis-33603	80	34	length	length	NOUN
bracis-33603	80	35	of	of	ADP
bracis-33603	80	36	512	512	NUM
bracis-33603	80	37	tokens	token	NOUN
bracis-33603	80	38	,	,	PUNCT
bracis-33603	80	39	using	use	VERB
bracis-33603	80	40	adafactor	adafactor	NOUN
bracis-33603	80	41	as	as	ADP
bracis-33603	80	42	the	the	DET
bracis-33603	80	43	optimizer	optimizer	NOUN
bracis-33603	80	44	with	with	ADP
bracis-33603	80	45	a	a	DET
bracis-33603	80	46	constant	constant	ADJ
bracis-33603	80	47	learning	learning	NOUN
bracis-33603	80	48	rate	rate	NOUN
bracis-33603	80	49	of	of	ADP
bracis-33603	80	50	0.001	0.001	NUM
bracis-33603	80	51	.	.	PUNCT
bracis-33603	81	1	the	the	DET
bracis-33603	81	2	model	model	NOUN
bracis-33603	81	3	checkpoint	checkpoint	NOUN
bracis-33603	81	4	yielding	yield	VERB
bracis-33603	81	5	the	the	DET
bracis-33603	81	6	best	good	ADJ
bracis-33603	81	7	performance	performance	NOUN
bracis-33603	81	8	on	on	ADP
bracis-33603	81	9	the	the	DET
bracis-33603	81	10	validation	validation	NOUN
bracis-33603	81	11	set	set	NOUN
bracis-33603	81	12	was	be	AUX
bracis-33603	81	13	selected	select	VERB
bracis-33603	81	14	for	for	ADP
bracis-33603	81	15	testing	testing	NOUN
bracis-33603	81	16	,	,	PUNCT
bracis-33603	81	17	and	and	CCONJ
bracis-33603	81	18	greedy	greedy	ADJ
bracis-33603	81	19	decoding	decoding	NOUN
bracis-33603	81	20	was	be	AUX
bracis-33603	81	21	utilized	utilize	VERB
bracis-33603	81	22	as	as	ADP
bracis-33603	81	23	the	the	DET
bracis-33603	81	24	decoding	decode	VERB
bracis-33603	81	25	method	method	NOUN
bracis-33603	81	26	.	.	PUNCT
bracis-33603	82	1	because	because	SCONJ
bracis-33603	82	2	tweetsentbr	tweetsentbr	NOUN
bracis-33603	82	3	lacks	lack	VERB
bracis-33603	82	4	a	a	DET
bracis-33603	82	5	validation	validation	NOUN
bracis-33603	82	6	set	set	NOUN
bracis-33603	82	7	,	,	PUNCT
bracis-33603	82	8	we	we	PRON
bracis-33603	82	9	reserved	reserve	VERB
bracis-33603	82	10	10	10	NUM
bracis-33603	82	11	%	%	NOUN
bracis-33603	82	12	of	of	ADP
bracis-33603	82	13	the	the	DET
bracis-33603	82	14	training	training	NOUN
bracis-33603	82	15	data	datum	NOUN
bracis-33603	82	16	for	for	ADP
bracis-33603	82	17	validation	validation	NOUN
bracis-33603	82	18	and	and	CCONJ
bracis-33603	82	19	used	use	VERB
bracis-33603	82	20	the	the	DET
bracis-33603	82	21	remaining	remain	VERB
bracis-33603	82	22	90	90	NUM
bracis-33603	82	23	%	%	NOUN
bracis-33603	82	24	for	for	ADP
bracis-33603	82	25	training	training	NOUN
bracis-33603	82	26	.	.	PUNCT
bracis-33603	83	1	all	all	DET
bracis-33603	83	2	tasks	task	NOUN
bracis-33603	83	3	were	be	AUX
bracis-33603	83	4	approached	approach	VERB
bracis-33603	83	5	using	use	VERB
bracis-33603	83	6	a	a	DET
bracis-33603	83	7	text	text	NOUN
bracis-33603	83	8	-	-	PUNCT
bracis-33603	83	9	to	to	ADP
bracis-33603	83	10	-	-	PUNCT
bracis-33603	83	11	text	text	NOUN
bracis-33603	83	12	format	format	NOUN
bracis-33603	83	13	.	.	PUNCT
bracis-33603	84	1	specifically	specifically	ADV
bracis-33603	84	2	for	for	ADP
bracis-33603	84	3	the	the	DET
bracis-33603	84	4	assin2	assin2	PROPN
bracis-33603	84	5	sts	st	NOUN
bracis-33603	84	6	task	task	NOUN
bracis-33603	84	7	,	,	PUNCT
bracis-33603	84	8	which	which	PRON
bracis-33603	84	9	involves	involve	VERB
bracis-33603	84	10	the	the	DET
bracis-33603	84	11	prediction	prediction	NOUN
bracis-33603	84	12	of	of	ADP
bracis-33603	84	13	continuous	continuous	ADJ
bracis-33603	84	14	values	value	NOUN
bracis-33603	84	15	in	in	ADP
bracis-33603	84	16	the	the	DET
bracis-33603	84	17	range	range	NOUN
bracis-33603	84	18	between	between	ADP
bracis-33603	84	19	1	1	NUM
bracis-33603	84	20	to	to	PART
bracis-33603	84	21	5	5	NUM
bracis-33603	84	22	,	,	PUNCT
bracis-33603	84	23	we	we	PRON
bracis-33603	84	24	adopted	adopt	VERB
bracis-33603	84	25	the	the	DET
bracis-33603	84	26	strategy	strategy	NOUN
bracis-33603	84	27	from	from	ADP
bracis-33603	84	28	raffel	raffel	PROPN
bracis-33603	84	29	et	et	PROPN
bracis-33603	84	30	al	al	PROPN
bracis-33603	84	31	.	.	PUNCT
bracis-33603	85	1	[	[	X
bracis-33603	85	2	34	34	NUM
bracis-33603	85	3	]	]	PUNCT
bracis-33603	85	4	,	,	PUNCT
bracis-33603	85	5	by	by	ADP
bracis-33603	85	6	rounding	round	VERB
bracis-33603	85	7	the	the	DET
bracis-33603	85	8	target	target	NOUN
bracis-33603	85	9	scores	score	NOUN
bracis-33603	85	10	to	to	ADP
bracis-33603	85	11	the	the	DET
bracis-33603	85	12	nearest	near	ADJ
bracis-33603	85	13	0.2	0.2	NUM
bracis-33603	85	14	increment	increment	NOUN
bracis-33603	85	15	and	and	CCONJ
bracis-33603	85	16	converting	convert	VERB
bracis-33603	85	17	these	these	PRON
bracis-33603	85	18	to	to	ADP
bracis-33603	85	19	strings	string	NOUN
bracis-33603	85	20	,	,	PUNCT
bracis-33603	85	21	thus	thus	ADV
bracis-33603	85	22	framing	frame	VERB
bracis-33603	85	23	it	it	PRON
bracis-33603	85	24	as	as	ADP
bracis-33603	85	25	a	a	DET
bracis-33603	85	26	multiclass	multiclass	ADJ
bracis-33603	85	27	classification	classification	NOUN
bracis-33603	85	28	problem	problem	NOUN
bracis-33603	85	29	compatible	compatible	ADJ
bracis-33603	85	30	with	with	ADP
bracis-33603	85	31	the	the	DET
bracis-33603	85	32	text	text	NOUN
bracis-33603	85	33	-	-	PUNCT
bracis-33603	85	34	to	to	ADP
bracis-33603	85	35	-	-	PUNCT
bracis-33603	85	36	text	text	NOUN
bracis-33603	85	37	format	format	NOUN
bracis-33603	85	38	.	.	PUNCT
bracis-33603	86	1	to	to	PART
bracis-33603	86	2	compare	compare	VERB
bracis-33603	86	3	the	the	DET
bracis-33603	86	4	quality	quality	NOUN
bracis-33603	86	5	of	of	ADP
bracis-33603	86	6	the	the	DET
bracis-33603	86	7	new	new	ADJ
bracis-33603	86	8	checkpoints	checkpoint	NOUN
bracis-33603	86	9	against	against	ADP
bracis-33603	86	10	existing	exist	VERB
bracis-33603	86	11	alternatives	alternative	NOUN
bracis-33603	86	12	,	,	PUNCT
bracis-33603	86	13	we	we	PRON
bracis-33603	86	14	also	also	ADV
bracis-33603	86	15	use	use	VERB
bracis-33603	86	16	the	the	DET
bracis-33603	86	17	same	same	ADJ
bracis-33603	86	18	finetuning	finetuning	NOUN
bracis-33603	86	19	procedure	procedure	NOUN
bracis-33603	86	20	on	on	ADP
bracis-33603	86	21	google	google	PROPN
bracis-33603	86	22	t5	t5	PROPN
bracis-33603	86	23	and	and	CCONJ
bracis-33603	86	24	mt5	mt5	ADJ
bracis-33603	86	25	models	model	NOUN
bracis-33603	86	26	.	.	PUNCT
bracis-33603	87	1	3.3	3.3	NUM
bracis-33603	87	2	monoptt5	monoptt5	NOUN
bracis-33603	87	3	rerankers	reranker	NOUN
bracis-33603	87	4	to	to	PART
bracis-33603	87	5	evaluate	evaluate	VERB
bracis-33603	87	6	the	the	DET
bracis-33603	87	7	adaptability	adaptability	NOUN
bracis-33603	87	8	of	of	ADP
bracis-33603	87	9	the	the	DET
bracis-33603	87	10	ptt5	ptt5	NOUN
bracis-33603	87	11	-	-	PUNCT
bracis-33603	87	12	v2	v2	NOUN
bracis-33603	87	13	models	model	NOUN
bracis-33603	87	14	for	for	ADP
bracis-33603	87	15	information	information	NOUN
bracis-33603	87	16	retrieval	retrieval	NOUN
bracis-33603	87	17	tasks	task	NOUN
bracis-33603	87	18	,	,	PUNCT
bracis-33603	87	19	specifically	specifically	ADV
bracis-33603	87	20	passage	passage	NOUN
bracis-33603	87	21	reranking	reranking	NOUN
bracis-33603	87	22	,	,	PUNCT
bracis-33603	87	23	we	we	PRON
bracis-33603	87	24	trained	train	VERB
bracis-33603	87	25	monot5	monot5	NOUN
bracis-33603	87	26	rerankers	reranker	NOUN
bracis-33603	88	1	[	[	X
bracis-33603	88	2	30	30	NUM
bracis-33603	88	3	]	]	PUNCT
bracis-33603	88	4	using	use	VERB
bracis-33603	88	5	checkpoints	checkpoint	NOUN
bracis-33603	88	6	generated	generate	VERB
bracis-33603	88	7	as	as	SCONJ
bracis-33603	88	8	described	describe	VERB
bracis-33603	88	9	in	in	ADP
bracis-33603	88	10	sect	sect	NOUN
bracis-33603	88	11	.	.	PUNCT
bracis-33603	88	12	 	 	SPACE
bracis-33603	89	1	3.1	3.1	NUM
bracis-33603	89	2	.	.	PUNCT
bracis-33603	90	1	we	we	PRON
bracis-33603	90	2	named	name	VERB
bracis-33603	90	3	these	these	DET
bracis-33603	90	4	models	model	NOUN
bracis-33603	90	5	monoptt5	monoptt5	PROPN
bracis-33603	90	6	.	.	PUNCT
bracis-33603	91	1	monot5	monot5	PROPN
bracis-33603	91	2	rerankers	reranker	NOUN
bracis-33603	91	3	are	be	AUX
bracis-33603	91	4	used	use	VERB
bracis-33603	91	5	for	for	ADP
bracis-33603	91	6	passage	passage	NOUN
bracis-33603	91	7	reranking	reranking	NOUN
bracis-33603	91	8	,	,	PUNCT
bracis-33603	91	9	a	a	DET
bracis-33603	91	10	two	two	NUM
bracis-33603	91	11	-	-	PUNCT
bracis-33603	91	12	stage	stage	NOUN
bracis-33603	91	13	process	process	NOUN
bracis-33603	91	14	:	:	PUNCT
bracis-33603	91	15	first	first	ADV
bracis-33603	91	16	,	,	PUNCT
bracis-33603	91	17	a	a	DET
bracis-33603	91	18	less	less	ADV
bracis-33603	91	19	computationally	computationally	ADV
bracis-33603	91	20	expensive	expensive	ADJ
bracis-33603	91	21	method	method	NOUN
bracis-33603	91	22	like	like	SCONJ
bracis-33603	91	23	bm25	bm25	PROPN
bracis-33603	91	24	generates	generate	VERB
bracis-33603	91	25	an	an	DET
bracis-33603	91	26	initial	initial	ADJ
bracis-33603	91	27	set	set	NOUN
bracis-33603	91	28	of	of	ADP
bracis-33603	91	29	relevant	relevant	ADJ
bracis-33603	91	30	documents	document	NOUN
bracis-33603	91	31	for	for	ADP
bracis-33603	91	32	a	a	DET
bracis-33603	91	33	given	give	VERB
bracis-33603	91	34	query	query	NOUN
bracis-33603	91	35	;	;	PUNCT
bracis-33603	91	36	the	the	DET
bracis-33603	91	37	reranker	reranker	NOUN
bracis-33603	91	38	model	model	NOUN
bracis-33603	91	39	then	then	ADV
bracis-33603	91	40	reranks	rerank	VERB
bracis-33603	91	41	a	a	DET
bracis-33603	91	42	subset	subset	NOUN
bracis-33603	91	43	of	of	ADP
bracis-33603	91	44	these	these	DET
bracis-33603	91	45	documents	document	NOUN
bracis-33603	91	46	to	to	PART
bracis-33603	91	47	improve	improve	VERB
bracis-33603	91	48	relevance	relevance	NOUN
bracis-33603	91	49	ordering	ordering	NOUN
bracis-33603	91	50	.	.	PUNCT
bracis-33603	92	1	during	during	ADP
bracis-33603	92	2	training	training	NOUN
bracis-33603	92	3	,	,	PUNCT
bracis-33603	92	4	the	the	DET
bracis-33603	92	5	model	model	NOUN
bracis-33603	92	6	learns	learn	VERB
bracis-33603	92	7	in	in	ADP
bracis-33603	92	8	a	a	DET
bracis-33603	92	9	supervised	supervised	ADJ
bracis-33603	92	10	text	text	NOUN
bracis-33603	92	11	-	-	PUNCT
bracis-33603	92	12	to	to	ADP
bracis-33603	92	13	-	-	PUNCT
bracis-33603	92	14	text	text	NOUN
bracis-33603	92	15	manner	manner	NOUN
bracis-33603	92	16	to	to	PART
bracis-33603	92	17	generate	generate	VERB
bracis-33603	92	18	tokens	token	NOUN
bracis-33603	92	19	corresponding	correspond	VERB
bracis-33603	92	20	to	to	ADP
bracis-33603	92	21	relevant	relevant	ADJ
bracis-33603	92	22	and	and	CCONJ
bracis-33603	92	23	non	non	ADJ
bracis-33603	92	24	-	-	ADJ
bracis-33603	92	25	relevant	relevant	ADJ
bracis-33603	92	26	labels	label	NOUN
bracis-33603	92	27	.	.	PUNCT
bracis-33603	93	1	for	for	ADP
bracis-33603	93	2	inference	inference	NOUN
bracis-33603	93	3	,	,	PUNCT
bracis-33603	93	4	we	we	PRON
bracis-33603	93	5	greedy	greedy	ADJ
bracis-33603	93	6	decode	decode	VERB
bracis-33603	93	7	a	a	DET
bracis-33603	93	8	single	single	ADJ
bracis-33603	93	9	token	token	NOUN
bracis-33603	93	10	and	and	CCONJ
bracis-33603	93	11	calculate	calculate	VERB
bracis-33603	93	12	the	the	DET
bracis-33603	93	13	softmax	softmax	NOUN
bracis-33603	93	14	over	over	ADP
bracis-33603	93	15	the	the	DET
bracis-33603	93	16	logits	logit	NOUN
bracis-33603	93	17	of	of	ADP
bracis-33603	93	18	the	the	DET
bracis-33603	93	19	two	two	NUM
bracis-33603	93	20	possible	possible	ADJ
bracis-33603	93	21	tokens	token	NOUN
bracis-33603	93	22	,	,	PUNCT
bracis-33603	93	23	using	use	VERB
bracis-33603	93	24	the	the	DET
bracis-33603	93	25	probability	probability	NOUN
bracis-33603	93	26	of	of	ADP
bracis-33603	93	27	the	the	DET
bracis-33603	93	28	positive	positive	ADJ
bracis-33603	93	29	class	class	NOUN
bracis-33603	93	30	as	as	ADP
bracis-33603	93	31	the	the	DET
bracis-33603	93	32	relevance	relevance	NOUN
bracis-33603	93	33	score	score	NOUN
bracis-33603	93	34	.	.	PUNCT
bracis-33603	94	1	we	we	PRON
bracis-33603	94	2	adapted	adapt	VERB
bracis-33603	94	3	the	the	DET
bracis-33603	94	4	input	input	NOUN
bracis-33603	94	5	and	and	CCONJ
bracis-33603	94	6	target	target	NOUN
bracis-33603	94	7	format	format	NOUN
bracis-33603	94	8	for	for	ADP
bracis-33603	94	9	the	the	DET
bracis-33603	94	10	portuguese	portuguese	ADJ
bracis-33603	94	11	language	language	NOUN
bracis-33603	94	12	to	to	ADP
bracis-33603	94	13	the	the	DET
bracis-33603	94	14	structure	structure	NOUN
bracis-33603	94	15	"	"	PUNCT
bracis-33603	94	16	pergunta	pergunta	VERB
bracis-33603	94	17	:	:	PUNCT
bracis-33603	94	18	{	{	PUNCT
bracis-33603	94	19	query	query	NOUN
bracis-33603	94	20	}	}	PUNCT
bracis-33603	94	21	documento	documento	NOUN
bracis-33603	94	22	:	:	PUNCT
bracis-33603	94	23	{	{	PUNCT
bracis-33603	94	24	document	document	NOUN
bracis-33603	94	25	}	}	PUNCT
bracis-33603	94	26	relevante	relevante	PROPN
bracis-33603	94	27	:	:	PUNCT
bracis-33603	94	28	"	"	PUNCT
bracis-33603	94	29	,	,	PUNCT
bracis-33603	94	30	assigning	assign	VERB
bracis-33603	94	31	the	the	DET
bracis-33603	94	32	tokens	token	NOUN
bracis-33603	94	33	“	"	PUNCT
bracis-33603	94	34	sim	sim	NOUN
bracis-33603	94	35	”	"	PUNCT
bracis-33603	94	36	(	(	PUNCT
bracis-33603	94	37	relevant	relevant	ADJ
bracis-33603	94	38	)	)	PUNCT
bracis-33603	94	39	and	and	CCONJ
bracis-33603	94	40	“	"	PUNCT
bracis-33603	94	41	não	não	NOUN
bracis-33603	94	42	”	"	PUNCT
bracis-33603	94	43	(	(	PUNCT
bracis-33603	94	44	non	non	ADJ
bracis-33603	94	45	-	-	ADJ
bracis-33603	94	46	relevant	relevant	ADJ
bracis-33603	94	47	)	)	PUNCT
bracis-33603	94	48	.	.	PUNCT
bracis-33603	95	1	this	this	DET
bracis-33603	95	2	format	format	NOUN
bracis-33603	95	3	is	be	AUX
bracis-33603	95	4	applied	apply	VERB
bracis-33603	95	5	during	during	ADP
bracis-33603	95	6	both	both	DET
bracis-33603	95	7	training	training	NOUN
bracis-33603	95	8	and	and	CCONJ
bracis-33603	95	9	inference	inference	NOUN
bracis-33603	95	10	,	,	PUNCT
bracis-33603	95	11	regardless	regardless	ADV
bracis-33603	95	12	of	of	ADP
bracis-33603	95	13	the	the	DET
bracis-33603	95	14	input	input	NOUN
bracis-33603	95	15	language	language	NOUN
bracis-33603	95	16	.	.	PUNCT
bracis-33603	96	1	the	the	DET
bracis-33603	96	2	training	training	NOUN
bracis-33603	96	3	data	datum	NOUN
bracis-33603	96	4	originated	originate	VERB
bracis-33603	96	5	from	from	ADP
bracis-33603	96	6	the	the	DET
bracis-33603	96	7	mmarco	mmarco	NOUN
bracis-33603	96	8	dataset	dataset	VERB
bracis-33603	97	1	[	[	PUNCT
bracis-33603	97	2	3	3	NUM
bracis-33603	97	3	]	]	PUNCT
bracis-33603	97	4	,	,	PUNCT
bracis-33603	97	5	a	a	DET
bracis-33603	97	6	translated	translate	VERB
bracis-33603	97	7	version	version	NOUN
bracis-33603	97	8	of	of	ADP
bracis-33603	97	9	the	the	DET
bracis-33603	97	10	ms	ms	PROPN
bracis-33603	97	11	marco	marco	PROPN
bracis-33603	97	12	passage	passage	PROPN
bracis-33603	97	13	retrieval	retrieval	NOUN
bracis-33603	97	14	dataset	dataset	NOUN
bracis-33603	98	1	[	[	X
bracis-33603	98	2	1	1	NUM
bracis-33603	98	3	]	]	PUNCT
bracis-33603	98	4	,	,	PUNCT
bracis-33603	98	5	originally	originally	ADV
bracis-33603	98	6	in	in	ADP
bracis-33603	98	7	english	english	PROPN
bracis-33603	98	8	,	,	PUNCT
bracis-33603	98	9	to	to	ADP
bracis-33603	98	10	13	13	NUM
bracis-33603	98	11	languages	language	NOUN
bracis-33603	98	12	,	,	PUNCT
bracis-33603	98	13	including	include	VERB
bracis-33603	98	14	portuguese	portuguese	PROPN
bracis-33603	98	15	.	.	PUNCT
bracis-33603	99	1	the	the	DET
bracis-33603	99	2	training	training	NOUN
bracis-33603	99	3	subset	subset	VERB
bracis-33603	99	4	consists	consist	VERB
bracis-33603	99	5	of	of	ADP
bracis-33603	99	6	triples	triple	NOUN
bracis-33603	99	7	(	(	PUNCT
bracis-33603	99	8	query	query	NOUN
bracis-33603	99	9	,	,	PUNCT
bracis-33603	99	10	relevant	relevant	ADJ
bracis-33603	99	11	passage	passage	NOUN
bracis-33603	99	12	,	,	PUNCT
bracis-33603	99	13	non	non	ADJ
bracis-33603	99	14	-	-	ADJ
bracis-33603	99	15	relevant	relevant	ADJ
bracis-33603	99	16	passage	passage	NOUN
bracis-33603	99	17	)	)	PUNCT
bracis-33603	99	18	,	,	PUNCT
bracis-33603	99	19	which	which	PRON
bracis-33603	99	20	we	we	PRON
bracis-33603	99	21	split	split	VERB
bracis-33603	99	22	into	into	ADP
bracis-33603	99	23	two	two	NUM
bracis-33603	99	24	training	training	NOUN
bracis-33603	99	25	example	example	NOUN
bracis-33603	99	26	pairs	pair	NOUN
bracis-33603	99	27	,	,	PUNCT
bracis-33603	99	28	with	with	ADP
bracis-33603	99	29	each	each	DET
bracis-33603	99	30	pair	pair	NOUN
bracis-33603	99	31	containing	contain	VERB
bracis-33603	99	32	the	the	DET
bracis-33603	99	33	query	query	NOUN
bracis-33603	99	34	matched	match	VERB
bracis-33603	99	35	to	to	ADP
bracis-33603	99	36	one	one	NUM
bracis-33603	99	37	passage	passage	NOUN
bracis-33603	99	38	–	–	PUNCT
bracis-33603	99	39	either	either	CCONJ
bracis-33603	99	40	relevant	relevant	ADJ
bracis-33603	99	41	or	or	CCONJ
bracis-33603	99	42	non	non	ADJ
bracis-33603	99	43	-	-	ADJ
bracis-33603	99	44	relevant	relevant	ADJ
bracis-33603	99	45	–	–	PUNCT
bracis-33603	99	46	thus	thus	ADV
bracis-33603	99	47	creating	create	VERB
bracis-33603	99	48	one	one	NUM
bracis-33603	99	49	example	example	NOUN
bracis-33603	99	50	for	for	ADP
bracis-33603	99	51	each	each	DET
bracis-33603	99	52	label	label	NOUN
bracis-33603	99	53	.	.	PUNCT
bracis-33603	100	1	we	we	PRON
bracis-33603	100	2	created	create	VERB
bracis-33603	100	3	a	a	DET
bracis-33603	100	4	bilingual	bilingual	ADJ
bracis-33603	100	5	portuguese	portuguese	ADJ
bracis-33603	100	6	-	-	PUNCT
bracis-33603	100	7	english	english	ADJ
bracis-33603	100	8	training	training	NOUN
bracis-33603	100	9	dataset	dataset	VERB
bracis-33603	100	10	by	by	ADP
bracis-33603	100	11	randomly	randomly	ADV
bracis-33603	100	12	assigning	assign	VERB
bracis-33603	100	13	one	one	NUM
bracis-33603	100	14	of	of	ADP
bracis-33603	100	15	the	the	DET
bracis-33603	100	16	two	two	NUM
bracis-33603	100	17	languages	language	NOUN
bracis-33603	100	18	to	to	ADP
bracis-33603	100	19	each	each	DET
bracis-33603	100	20	training	training	NOUN
bracis-33603	100	21	triplet	triplet	NOUN
bracis-33603	100	22	.	.	PUNCT
bracis-33603	101	1	this	this	DET
bracis-33603	101	2	“	"	PUNCT
bracis-33603	101	3	translate	translate	ADJ
bracis-33603	101	4	-	-	PUNCT
bracis-33603	101	5	train	train	NOUN
bracis-33603	101	6	”	"	PUNCT
bracis-33603	101	7	approach	approach	NOUN
bracis-33603	101	8	[	[	X
bracis-33603	101	9	9	9	NUM
bracis-33603	101	10	,	,	PUNCT
bracis-33603	101	11	20	20	NUM
bracis-33603	101	12	,	,	PUNCT
bracis-33603	101	13	48	48	NUM
bracis-33603	101	14	]	]	PUNCT
bracis-33603	101	15	leverages	leverage	VERB
bracis-33603	101	16	synthetic	synthetic	ADJ
bracis-33603	101	17	data	datum	NOUN
bracis-33603	101	18	augmentation	augmentation	NOUN
bracis-33603	101	19	by	by	ADP
bracis-33603	101	20	integrating	integrate	VERB
bracis-33603	101	21	machine	machine	NOUN
bracis-33603	101	22	translations	translation	NOUN
bracis-33603	101	23	with	with	ADP
bracis-33603	101	24	original	original	ADJ
bracis-33603	101	25	text	text	NOUN
bracis-33603	101	26	data	datum	NOUN
bracis-33603	101	27	to	to	PART
bracis-33603	101	28	substantially	substantially	ADV
bracis-33603	101	29	expand	expand	VERB
bracis-33603	101	30	the	the	DET
bracis-33603	101	31	available	available	ADJ
bracis-33603	101	32	training	training	NOUN
bracis-33603	101	33	material	material	NOUN
bracis-33603	101	34	in	in	ADP
bracis-33603	101	35	the	the	DET
bracis-33603	101	36	target	target	NOUN
bracis-33603	101	37	language	language	NOUN
bracis-33603	101	38	.	.	PUNCT
bracis-33603	102	1	prior	prior	ADJ
bracis-33603	102	2	research	research	NOUN
bracis-33603	102	3	[	[	X
bracis-33603	102	4	3	3	NUM
bracis-33603	102	5	,	,	PUNCT
bracis-33603	102	6	9	9	NUM
bracis-33603	102	7	,	,	PUNCT
bracis-33603	102	8	38	38	NUM
bracis-33603	102	9	]	]	PUNCT
bracis-33603	102	10	has	have	AUX
bracis-33603	102	11	shown	show	VERB
bracis-33603	102	12	the	the	DET
bracis-33603	102	13	effectiveness	effectiveness	NOUN
bracis-33603	102	14	of	of	ADP
bracis-33603	102	15	this	this	DET
bracis-33603	102	16	bilingual	bilingual	ADJ
bracis-33603	102	17	training	training	NOUN
bracis-33603	102	18	strategy	strategy	NOUN
bracis-33603	102	19	,	,	PUNCT
bracis-33603	102	20	motivating	motivate	VERB
bracis-33603	102	21	our	our	PRON
bracis-33603	102	22	adoption	adoption	NOUN
bracis-33603	102	23	of	of	ADP
bracis-33603	102	24	this	this	DET
bracis-33603	102	25	method	method	NOUN
bracis-33603	102	26	.	.	PUNCT
bracis-33603	103	1	the	the	DET
bracis-33603	103	2	models	model	NOUN
bracis-33603	103	3	were	be	AUX
bracis-33603	103	4	trained	train	VERB
bracis-33603	103	5	for	for	ADP
bracis-33603	103	6	100k	100k	NUM
bracis-33603	103	7	steps	step	NOUN
bracis-33603	103	8	with	with	ADP
bracis-33603	103	9	batch	batch	NOUN
bracis-33603	103	10	sizes	size	NOUN
bracis-33603	103	11	of	of	ADP
bracis-33603	103	12	128	128	NUM
bracis-33603	103	13	sequences	sequence	NOUN
bracis-33603	103	14	and	and	CCONJ
bracis-33603	103	15	a	a	DET
bracis-33603	103	16	maximum	maximum	ADJ
bracis-33603	103	17	length	length	NOUN
bracis-33603	103	18	of	of	ADP
bracis-33603	103	19	512	512	NUM
bracis-33603	103	20	tokens	token	NOUN
bracis-33603	103	21	,	,	PUNCT
bracis-33603	103	22	utilizing	utilize	VERB
bracis-33603	103	23	adafactor	adafactor	NOUN
bracis-33603	103	24	with	with	ADP
bracis-33603	103	25	a	a	DET
bracis-33603	103	26	constant	constant	ADJ
bracis-33603	103	27	learning	learning	NOUN
bracis-33603	103	28	rate	rate	NOUN
bracis-33603	103	29	of	of	ADP
bracis-33603	103	30	0.001	0.001	NUM
bracis-33603	103	31	as	as	ADP
bracis-33603	103	32	the	the	DET
bracis-33603	103	33	optimizer	optimizer	NOUN
bracis-33603	103	34	.	.	PUNCT
bracis-33603	104	1	instances	instance	NOUN
bracis-33603	104	2	exceeding	exceed	VERB
bracis-33603	104	3	the	the	DET
bracis-33603	104	4	maximum	maximum	ADV
bracis-33603	104	5	token	token	ADJ
bracis-33603	104	6	length	length	NOUN
bracis-33603	104	7	were	be	AUX
bracis-33603	104	8	excluded	exclude	VERB
bracis-33603	104	9	from	from	ADP
bracis-33603	104	10	the	the	DET
bracis-33603	104	11	training	training	NOUN
bracis-33603	104	12	dataset	dataset	NOUN
bracis-33603	104	13	;	;	PUNCT
bracis-33603	104	14	these	these	PRON
bracis-33603	104	15	constituted	constitute	VERB
bracis-33603	104	16	approximately	approximately	ADV
bracis-33603	104	17	0.01	0.01	NUM
bracis-33603	104	18	%	%	NOUN
bracis-33603	104	19	of	of	ADP
bracis-33603	104	20	the	the	DET
bracis-33603	104	21	training	training	NOUN
bracis-33603	104	22	samples	sample	NOUN
bracis-33603	104	23	and	and	CCONJ
bracis-33603	104	24	were	be	AUX
bracis-33603	104	25	predominantly	predominantly	ADV
bracis-33603	104	26	attributed	attribute	VERB
bracis-33603	104	27	to	to	ADP
bracis-33603	104	28	noisy	noisy	ADJ
bracis-33603	104	29	translation	translation	NOUN
bracis-33603	104	30	data	datum	NOUN
bracis-33603	104	31	.	.	PUNCT
bracis-33603	105	1	given	give	VERB
bracis-33603	105	2	the	the	DET
bracis-33603	105	3	significant	significant	ADJ
bracis-33603	105	4	computational	computational	ADJ
bracis-33603	105	5	resources	resource	NOUN
bracis-33603	105	6	required	require	VERB
bracis-33603	105	7	for	for	ADP
bracis-33603	105	8	training	train	VERB
bracis-33603	105	9	these	these	DET
bracis-33603	105	10	rerankers	reranker	NOUN
bracis-33603	105	11	,	,	PUNCT
bracis-33603	105	12	we	we	PRON
bracis-33603	105	13	focused	focus	VERB
bracis-33603	105	14	exclusively	exclusively	ADV
bracis-33603	105	15	on	on	ADP
bracis-33603	105	16	models	model	NOUN
bracis-33603	105	17	based	base	VERB
bracis-33603	105	18	on	on	ADP
bracis-33603	105	19	the	the	DET
bracis-33603	105	20	main	main	ADJ
bracis-33603	105	21	ptt5	ptt5	NOUN
bracis-33603	105	22	-	-	PUNCT
bracis-33603	105	23	v2	v2	NOUN
bracis-33603	105	24	checkpoints	checkpoint	NOUN
bracis-33603	105	25	.	.	PUNCT
bracis-33603	106	1	to	to	PART
bracis-33603	106	2	evaluate	evaluate	VERB
bracis-33603	106	3	the	the	DET
bracis-33603	106	4	rerankers	reranker	NOUN
bracis-33603	106	5	,	,	PUNCT
bracis-33603	106	6	we	we	PRON
bracis-33603	106	7	first	first	ADV
bracis-33603	106	8	used	use	VERB
bracis-33603	106	9	bm25	bm25	PROPN
bracis-33603	106	10	to	to	PART
bracis-33603	106	11	generate	generate	VERB
bracis-33603	106	12	an	an	DET
bracis-33603	106	13	initial	initial	ADJ
bracis-33603	106	14	set	set	NOUN
bracis-33603	106	15	of	of	ADP
bracis-33603	106	16	relevant	relevant	ADJ
bracis-33603	106	17	documentsfootnote	documentsfootnote	NOUN
bracis-33603	106	18	1	1	NUM
bracis-33603	106	19	and	and	CCONJ
bracis-33603	106	20	then	then	ADV
bracis-33603	106	21	rerank	rerank	VERB
bracis-33603	106	22	the	the	DET
bracis-33603	106	23	top-1000	top-1000	NOUN
bracis-33603	106	24	documents	document	NOUN
bracis-33603	106	25	.	.	PUNCT
bracis-33603	107	1	retrieval	retrieval	NOUN
bracis-33603	107	2	metrics	metric	NOUN
bracis-33603	107	3	are	be	AUX
bracis-33603	107	4	calculated	calculate	VERB
bracis-33603	107	5	by	by	ADP
bracis-33603	107	6	comparing	compare	VERB
bracis-33603	107	7	this	this	DET
bracis-33603	107	8	ordered	order	VERB
bracis-33603	107	9	list	list	NOUN
bracis-33603	107	10	with	with	ADP
bracis-33603	107	11	the	the	DET
bracis-33603	107	12	relevance	relevance	NOUN
bracis-33603	107	13	judgments	judgment	NOUN
bracis-33603	107	14	from	from	ADP
bracis-33603	107	15	each	each	DET
bracis-33603	107	16	dataset	dataset	NOUN
bracis-33603	107	17	.	.	PUNCT
bracis-33603	108	1	we	we	PRON
bracis-33603	108	2	consider	consider	VERB
bracis-33603	108	3	two	two	NUM
bracis-33603	108	4	retrieval	retrieval	ADJ
bracis-33603	108	5	scenarios	scenario	NOUN
bracis-33603	108	6	:	:	PUNCT
bracis-33603	108	7	in	in	ADP
bracis-33603	108	8	-	-	PUNCT
bracis-33603	108	9	domain	domain	NOUN
bracis-33603	108	10	(	(	PUNCT
bracis-33603	108	11	using	use	VERB
bracis-33603	108	12	the	the	DET
bracis-33603	108	13	“	"	PUNCT
bracis-33603	108	14	small	small	ADJ
bracis-33603	108	15	dev	dev	NOUN
bracis-33603	108	16	”	"	PUNCT
bracis-33603	108	17	set	set	NOUN
bracis-33603	108	18	of	of	ADP
bracis-33603	108	19	6,980	6,980	NUM
bracis-33603	108	20	queries	query	NOUN
bracis-33603	108	21	from	from	ADP
bracis-33603	108	22	the	the	DET
bracis-33603	108	23	mmarco	mmarco	ADJ
bracis-33603	108	24	-	-	PUNCT
bracis-33603	108	25	pt	pt	NOUN
bracis-33603	108	26	dataset	dataset	NOUN
bracis-33603	108	27	)	)	PUNCT
bracis-33603	108	28	and	and	CCONJ
bracis-33603	108	29	zero	zero	NUM
bracis-33603	108	30	-	-	PUNCT
bracis-33603	108	31	shot	shot	NOUN
bracis-33603	108	32	(	(	PUNCT
bracis-33603	108	33	using	use	VERB
bracis-33603	108	34	the	the	DET
bracis-33603	108	35	portuguese	portuguese	ADJ
bracis-33603	108	36	subset	subset	NOUN
bracis-33603	108	37	of	of	ADP
bracis-33603	108	38	249	249	NUM
bracis-33603	108	39	annotated	annotate	VERB
bracis-33603	108	40	queries	query	NOUN
bracis-33603	108	41	from	from	ADP
bracis-33603	108	42	mrobust	mrobust	PROPN
bracis-33603	108	43	[	[	X
bracis-33603	108	44	21	21	NUM
bracis-33603	108	45	]	]	PUNCT
bracis-33603	108	46	)	)	PUNCT
bracis-33603	108	47	.	.	PUNCT
bracis-33603	109	1	due	due	ADP
bracis-33603	109	2	to	to	ADP
bracis-33603	109	3	the	the	DET
bracis-33603	109	4	longer	long	ADJ
bracis-33603	109	5	document	document	NOUN
bracis-33603	109	6	length	length	NOUN
bracis-33603	109	7	in	in	ADP
bracis-33603	109	8	mrobust	mrobust	PROPN
bracis-33603	109	9	,	,	PUNCT
bracis-33603	109	10	we	we	PRON
bracis-33603	109	11	segmented	segment	VERB
bracis-33603	109	12	documents	document	NOUN
bracis-33603	109	13	into	into	ADP
bracis-33603	109	14	sliding	slide	VERB
bracis-33603	109	15	sentence	sentence	NOUN
bracis-33603	109	16	windows	window	NOUN
bracis-33603	109	17	using	use	VERB
bracis-33603	109	18	a	a	DET
bracis-33603	109	19	spacy	spacy	NOUN
bracis-33603	109	20	[	[	X
bracis-33603	109	21	19	19	NUM
bracis-33603	109	22	]	]	X
bracis-33603	109	23	sentencizer	sentencizer	NOUN
bracis-33603	109	24	pipeline	pipeline	NOUN
bracis-33603	109	25	,	,	PUNCT
bracis-33603	109	26	with	with	ADP
bracis-33603	109	27	a	a	DET
bracis-33603	109	28	maximum	maximum	ADJ
bracis-33603	109	29	length	length	NOUN
bracis-33603	109	30	of	of	ADP
bracis-33603	109	31	8	8	NUM
bracis-33603	109	32	and	and	CCONJ
bracis-33603	109	33	a	a	DET
bracis-33603	109	34	stride	stride	NOUN
bracis-33603	109	35	of	of	ADP
bracis-33603	109	36	4	4	NUM
bracis-33603	109	37	sentences	sentence	NOUN
bracis-33603	109	38	to	to	PART
bracis-33603	109	39	mitigate	mitigate	VERB
bracis-33603	109	40	truncation	truncation	NOUN
bracis-33603	109	41	during	during	ADP
bracis-33603	109	42	reranking	reranking	NOUN
bracis-33603	109	43	.	.	PUNCT
bracis-33603	110	1	table	table	NOUN
bracis-33603	110	2	3	3	NUM
bracis-33603	110	3	.	.	PUNCT
bracis-33603	110	4	main	main	ADJ
bracis-33603	110	5	results	result	NOUN
bracis-33603	110	6	after	after	ADP
bracis-33603	110	7	finetuning	finetune	VERB
bracis-33603	110	8	.	.	PUNCT
bracis-33603	111	1	for	for	ADP
bracis-33603	111	2	supervised	supervised	ADJ
bracis-33603	111	3	sotas	sota	NOUN
bracis-33603	111	4	,	,	PUNCT
bracis-33603	111	5	model	model	NOUN
bracis-33603	111	6	sizes	size	NOUN
bracis-33603	111	7	are	be	AUX
bracis-33603	111	8	as	as	SCONJ
bracis-33603	111	9	follows	follow	VERB
bracis-33603	111	10	:	:	PUNCT
bracis-33603	112	1	[	[	X
bracis-33603	112	2	38	38	NUM
bracis-33603	112	3	]	]	PUNCT
bracis-33603	112	4	(	(	PUNCT
bracis-33603	112	5	335	335	NUM
bracis-33603	112	6	m	m	NOUN
bracis-33603	112	7	)	)	PUNCT
bracis-33603	112	8	,	,	PUNCT
bracis-33603	113	1	[	[	X
bracis-33603	113	2	37	37	NUM
bracis-33603	113	3	]	]	X
bracis-33603	113	4	(	(	PUNCT
bracis-33603	113	5	900	900	NUM
bracis-33603	113	6	m	m	NUM
bracis-33603	113	7	)	)	PUNCT
bracis-33603	113	8	,	,	PUNCT
bracis-33603	113	9	[	[	X
bracis-33603	113	10	2	2	NUM
bracis-33603	113	11	]	]	X
bracis-33603	113	12	(	(	PUNCT
bracis-33603	113	13	110	110	NUM
bracis-33603	113	14	m	m	NOUN
bracis-33603	113	15	)	)	PUNCT
bracis-33603	113	16	,	,	PUNCT
bracis-33603	113	17	[	[	X
bracis-33603	113	18	38	38	NUM
bracis-33603	113	19	]	]	PUNCT
bracis-33603	113	20	,	,	PUNCT
bracis-33603	113	21	and	and	CCONJ
bracis-33603	113	22	[	[	X
bracis-33603	113	23	21	21	NUM
bracis-33603	113	24	]	]	X
bracis-33603	113	25	(	(	PUNCT
bracis-33603	113	26	580	580	NUM
bracis-33603	113	27	m	m	NOUN
bracis-33603	113	28	)	)	PUNCT
bracis-33603	113	29	.	.	PUNCT
bracis-33603	114	1	npm	npm	PROPN
bracis-33603	114	2	results	result	NOUN
bracis-33603	114	3	exclude	exclude	VERB
bracis-33603	114	4	retrieval	retrieval	NOUN
bracis-33603	114	5	tasks.full	tasks.full	DET
bracis-33603	114	6	size	size	NOUN
bracis-33603	114	7	table	table	NOUN
bracis-33603	114	8	4	4	NUM
bracis-33603	114	9	main	main	ADJ
bracis-33603	114	10	results	result	NOUN
bracis-33603	114	11	table	table	NOUN
bracis-33603	114	12	 	 	SPACE
bracis-33603	114	13	3	3	NUM
bracis-33603	114	14	shows	show	VERB
bracis-33603	114	15	the	the	DET
bracis-33603	114	16	results	result	NOUN
bracis-33603	114	17	on	on	ADP
bracis-33603	114	18	the	the	DET
bracis-33603	114	19	downstream	downstream	ADJ
bracis-33603	114	20	tasks	task	NOUN
bracis-33603	114	21	considered	consider	VERB
bracis-33603	114	22	.	.	PUNCT
bracis-33603	115	1	in	in	ADP
bracis-33603	115	2	the	the	DET
bracis-33603	115	3	assin2	assin2	PROPN
bracis-33603	115	4	rte	rte	NOUN
bracis-33603	115	5	task	task	NOUN
bracis-33603	115	6	,	,	PUNCT
bracis-33603	115	7	our	our	PRON
bracis-33603	115	8	3b	3b	NOUN
bracis-33603	115	9	sets	set	VERB
bracis-33603	115	10	a	a	DET
bracis-33603	115	11	new	new	ADJ
bracis-33603	115	12	sota	sota	NOUN
bracis-33603	115	13	,	,	PUNCT
bracis-33603	115	14	surpassing	surpass	VERB
bracis-33603	115	15	the	the	DET
bracis-33603	115	16	current	current	ADJ
bracis-33603	115	17	one	one	NUM
bracis-33603	115	18	by	by	ADP
bracis-33603	115	19	0.61	0.61	NUM
bracis-33603	115	20	f1	f1	ADJ
bracis-33603	115	21	-	-	PUNCT
bracis-33603	115	22	macro	macro	NOUN
bracis-33603	115	23	points	point	NOUN
bracis-33603	115	24	.	.	PUNCT
bracis-33603	116	1	for	for	SCONJ
bracis-33603	116	2	the	the	DET
bracis-33603	116	3	tweetsentbr	tweetsentbr	NOUN
bracis-33603	116	4	dataset	dataset	VERB
bracis-33603	116	5	,	,	PUNCT
bracis-33603	116	6	we	we	PRON
bracis-33603	116	7	achieved	achieve	VERB
bracis-33603	116	8	better	well	ADJ
bracis-33603	116	9	performance	performance	NOUN
bracis-33603	116	10	than	than	ADP
bracis-33603	116	11	current	current	ADJ
bracis-33603	116	12	finetuned	finetune	VERB
bracis-33603	116	13	sotas	sota	NOUN
bracis-33603	116	14	with	with	ADP
bracis-33603	116	15	ptt5	ptt5	PROPN
bracis-33603	116	16	-	-	PUNCT
bracis-33603	116	17	v2	v2	NOUN
bracis-33603	116	18	-	-	PUNCT
bracis-33603	116	19	large	large	ADJ
bracis-33603	116	20	and	and	CCONJ
bracis-33603	116	21	ptt5	ptt5	NOUN
bracis-33603	116	22	-	-	PUNCT
bracis-33603	116	23	v2	v2	NOUN
bracis-33603	116	24	-	-	PUNCT
bracis-33603	116	25	3b	3b	NOUN
bracis-33603	116	26	,	,	PUNCT
bracis-33603	116	27	by	by	ADP
bracis-33603	116	28	0.52	0.52	NUM
bracis-33603	116	29	and	and	CCONJ
bracis-33603	116	30	1.54	1.54	NUM
bracis-33603	116	31	f1	f1	NOUN
bracis-33603	116	32	-	-	PUNCT
bracis-33603	116	33	macro	macro	NOUN
bracis-33603	116	34	points	point	NOUN
bracis-33603	116	35	,	,	PUNCT
bracis-33603	116	36	respectively	respectively	ADV
bracis-33603	116	37	,	,	PUNCT
bracis-33603	116	38	but	but	CCONJ
bracis-33603	116	39	our	our	PRON
bracis-33603	116	40	results	result	NOUN
bracis-33603	116	41	are	be	AUX
bracis-33603	116	42	worse	bad	ADJ
bracis-33603	116	43	when	when	SCONJ
bracis-33603	116	44	comparing	compare	VERB
bracis-33603	116	45	to	to	ADP
bracis-33603	116	46	gpt-4	gpt-4	VERB
bracis-33603	116	47	.	.	PUNCT
bracis-33603	117	1	we	we	PRON
bracis-33603	117	2	highlight	highlight	VERB
bracis-33603	117	3	that	that	SCONJ
bracis-33603	117	4	our	our	PRON
bracis-33603	117	5	ptt5	ptt5	NOUN
bracis-33603	117	6	-	-	PUNCT
bracis-33603	117	7	v2	v2	PROPN
bracis-33603	117	8	were	be	AUX
bracis-33603	117	9	trained	train	VERB
bracis-33603	117	10	exclusively	exclusively	ADV
bracis-33603	117	11	on	on	ADP
bracis-33603	117	12	each	each	DET
bracis-33603	117	13	task	task	NOUN
bracis-33603	117	14	training	training	NOUN
bracis-33603	117	15	data	datum	NOUN
bracis-33603	117	16	using	use	VERB
bracis-33603	117	17	the	the	DET
bracis-33603	117	18	text	text	NOUN
bracis-33603	117	19	-	-	PUNCT
bracis-33603	117	20	to	to	ADP
bracis-33603	117	21	-	-	PUNCT
bracis-33603	117	22	text	text	NOUN
bracis-33603	117	23	framework	framework	NOUN
bracis-33603	117	24	without	without	ADP
bracis-33603	117	25	any	any	DET
bracis-33603	117	26	data	datum	NOUN
bracis-33603	117	27	augmentation	augmentation	NOUN
bracis-33603	117	28	or	or	CCONJ
bracis-33603	117	29	adaptation	adaptation	NOUN
bracis-33603	117	30	to	to	ADP
bracis-33603	117	31	the	the	DET
bracis-33603	117	32	model	model	NOUN
bracis-33603	117	33	’s	’s	PART
bracis-33603	117	34	architecture	architecture	NOUN
bracis-33603	117	35	,	,	PUNCT
bracis-33603	117	36	unlike	unlike	ADP
bracis-33603	117	37	the	the	DET
bracis-33603	117	38	works	work	NOUN
bracis-33603	117	39	of	of	ADP
bracis-33603	117	40	[	[	X
bracis-33603	117	41	2	2	NUM
bracis-33603	117	42	]	]	PUNCT
bracis-33603	117	43	and	and	CCONJ
bracis-33603	117	44	[	[	X
bracis-33603	117	45	38	38	NUM
bracis-33603	117	46	]	]	PUNCT
bracis-33603	117	47	,	,	PUNCT
bracis-33603	117	48	which	which	PRON
bracis-33603	117	49	held	hold	VERB
bracis-33603	117	50	the	the	DET
bracis-33603	117	51	sota	sota	NOUN
bracis-33603	117	52	for	for	ADP
bracis-33603	117	53	tweetsentbr	tweetsentbr	NOUN
bracis-33603	117	54	and	and	CCONJ
bracis-33603	117	55	assin2	assin2	PROPN
bracis-33603	117	56	rte	rte	NOUN
bracis-33603	117	57	.	.	PUNCT
bracis-33603	118	1	in	in	ADP
bracis-33603	118	2	the	the	DET
bracis-33603	118	3	assin2	assin2	PROPN
bracis-33603	118	4	sts	st	NOUN
bracis-33603	118	5	task	task	NOUN
bracis-33603	118	6	,	,	PUNCT
bracis-33603	118	7	our	our	PRON
bracis-33603	118	8	models	model	NOUN
bracis-33603	118	9	did	do	AUX
bracis-33603	118	10	not	not	PART
bracis-33603	118	11	surpass	surpass	VERB
bracis-33603	118	12	the	the	DET
bracis-33603	118	13	current	current	ADJ
bracis-33603	118	14	sota	sota	NOUN
bracis-33603	118	15	;	;	PUNCT
bracis-33603	118	16	regardless	regardless	ADV
bracis-33603	118	17	,	,	PUNCT
bracis-33603	118	18	ptt5	ptt5	NOUN
bracis-33603	118	19	-	-	PUNCT
bracis-33603	118	20	v2	v2	PROPN
bracis-33603	118	21	still	still	ADV
bracis-33603	118	22	shows	show	VERB
bracis-33603	118	23	better	well	ADJ
bracis-33603	118	24	performance	performance	NOUN
bracis-33603	118	25	than	than	ADP
bracis-33603	118	26	mt5	mt5	ADJ
bracis-33603	118	27	and	and	CCONJ
bracis-33603	118	28	t5	t5	PROPN
bracis-33603	118	29	models	model	NOUN
bracis-33603	118	30	with	with	ADP
bracis-33603	118	31	approximate	approximate	ADJ
bracis-33603	118	32	sizes	size	NOUN
bracis-33603	118	33	,	,	PUNCT
bracis-33603	118	34	and	and	CCONJ
bracis-33603	118	35	this	this	PRON
bracis-33603	118	36	is	be	AUX
bracis-33603	118	37	also	also	ADV
bracis-33603	118	38	the	the	DET
bracis-33603	118	39	only	only	ADJ
bracis-33603	118	40	task	task	NOUN
bracis-33603	118	41	where	where	SCONJ
bracis-33603	118	42	a	a	DET
bracis-33603	118	43	smaller	small	ADJ
bracis-33603	118	44	ptt5	ptt5	NOUN
bracis-33603	118	45	-	-	PUNCT
bracis-33603	118	46	v2	v2	PROPN
bracis-33603	118	47	model	model	NOUN
bracis-33603	118	48	(	(	PUNCT
bracis-33603	118	49	ptt5	ptt5	NOUN
bracis-33603	118	50	-	-	PUNCT
bracis-33603	118	51	v2	v2	NOUN
bracis-33603	118	52	-	-	PUNCT
bracis-33603	118	53	large	large	NOUN
bracis-33603	118	54	)	)	PUNCT
bracis-33603	118	55	has	have	VERB
bracis-33603	118	56	better	well	ADJ
bracis-33603	118	57	performance	performance	NOUN
bracis-33603	118	58	of	of	ADP
bracis-33603	118	59	a	a	DET
bracis-33603	118	60	large	large	ADJ
bracis-33603	118	61	one	one	NUM
bracis-33603	118	62	ptt5	ptt5	NOUN
bracis-33603	118	63	-	-	PUNCT
bracis-33603	118	64	v2	v2	NOUN
bracis-33603	118	65	-	-	PUNCT
bracis-33603	118	66	3b	3b	NOUN
bracis-33603	118	67	.	.	PUNCT
bracis-33603	119	1	in	in	ADP
bracis-33603	119	2	addition	addition	NOUN
bracis-33603	119	3	to	to	ADP
bracis-33603	119	4	using	use	VERB
bracis-33603	119	5	individual	individual	ADJ
bracis-33603	119	6	metrics	metric	NOUN
bracis-33603	119	7	for	for	ADP
bracis-33603	119	8	each	each	DET
bracis-33603	119	9	task	task	NOUN
bracis-33603	119	10	,	,	PUNCT
bracis-33603	119	11	we	we	PRON
bracis-33603	119	12	also	also	ADV
bracis-33603	119	13	incorporate	incorporate	VERB
bracis-33603	119	14	the	the	DET
bracis-33603	119	15	normalized	normalize	VERB
bracis-33603	119	16	preferred	preferred	ADJ
bracis-33603	119	17	metric	metric	ADJ
bracis-33603	119	18	(	(	PUNCT
bracis-33603	119	19	npm	npm	PROPN
bracis-33603	119	20	)	)	PUNCT
bracis-33603	120	1	[	[	X
bracis-33603	120	2	43	43	NUM
bracis-33603	120	3	]	]	PUNCT
bracis-33603	120	4	to	to	PART
bracis-33603	120	5	facilitate	facilitate	VERB
bracis-33603	120	6	the	the	DET
bracis-33603	120	7	evaluation	evaluation	NOUN
bracis-33603	120	8	of	of	ADP
bracis-33603	120	9	the	the	DET
bracis-33603	120	10	overall	overall	ADJ
bracis-33603	120	11	performance	performance	NOUN
bracis-33603	120	12	of	of	ADP
bracis-33603	120	13	a	a	DET
bracis-33603	120	14	pretrained	pretraine	VERB
bracis-33603	120	15	model	model	NOUN
bracis-33603	120	16	across	across	ADP
bracis-33603	120	17	multiple	multiple	ADJ
bracis-33603	120	18	tasks	task	NOUN
bracis-33603	120	19	.	.	PUNCT
bracis-33603	121	1	the	the	DET
bracis-33603	121	2	npm	npm	PROPN
bracis-33603	121	3	normalizes	normalize	VERB
bracis-33603	121	4	a	a	DET
bracis-33603	121	5	task	task	NOUN
bracis-33603	121	6	’s	’s	PART
bracis-33603	121	7	preferred	preferred	ADJ
bracis-33603	121	8	metric	metric	NOUN
bracis-33603	121	9	(	(	PUNCT
bracis-33603	121	10	e.g.	e.g.	ADJ
bracis-33603	121	11	,	,	PUNCT
bracis-33603	121	12	f1	f1	NOUN
bracis-33603	121	13	-	-	PUNCT
bracis-33603	121	14	macro	macro	NOUN
bracis-33603	121	15	for	for	ADP
bracis-33603	121	16	assin2	assin2	PROPN
bracis-33603	121	17	rte	rte	PROPN
bracis-33603	121	18	)	)	PUNCT
bracis-33603	121	19	,	,	PUNCT
bracis-33603	121	20	assigning	assign	VERB
bracis-33603	121	21	a	a	DET
bracis-33603	121	22	value	value	NOUN
bracis-33603	121	23	of	of	ADP
bracis-33603	121	24	0	0	NUM
bracis-33603	121	25	to	to	PART
bracis-33603	121	26	represent	represent	VERB
bracis-33603	121	27	random	random	ADJ
bracis-33603	121	28	performance	performance	NOUN
bracis-33603	121	29	and	and	CCONJ
bracis-33603	121	30	100	100	NUM
bracis-33603	121	31	to	to	PART
bracis-33603	121	32	denote	denote	VERB
bracis-33603	121	33	maximum	maximum	ADJ
bracis-33603	121	34	performance	performance	NOUN
bracis-33603	121	35	.	.	PUNCT
bracis-33603	122	1	below	below	ADV
bracis-33603	122	2	is	be	AUX
bracis-33603	122	3	the	the	DET
bracis-33603	122	4	equation	equation	NOUN
bracis-33603	122	5	used	use	VERB
bracis-33603	122	6	to	to	ADP
bracis-33603	122	7	calculating	calculate	VERB
bracis-33603	122	8	the	the	DET
bracis-33603	122	9	npm	npm	NOUN
bracis-33603	122	10	for	for	ADP
bracis-33603	122	11	a	a	DET
bracis-33603	122	12	given	give	VERB
bracis-33603	122	13	model	model	NOUN
bracis-33603	122	14	and	and	CCONJ
bracis-33603	122	15	set	set	VERB
bracis-33603	122	16	n	n	PROPN
bracis-33603	122	17	of	of	ADP
bracis-33603	122	18	tasks	task	NOUN
bracis-33603	122	19	:	:	PUNCT
bracis-33603	122	20	$	$	SYM
bracis-33603	122	21	$	$	SYM
bracis-33603	122	22	\begin{aligned	\begin{aligne	VERB
bracis-33603	122	23	}	}	PUNCT
bracis-33603	122	24	{	{	PUNCT
bracis-33603	122	25	\text	\text	PROPN
bracis-33603	122	26	{	{	PUNCT
bracis-33603	122	27	npm	npm	PROPN
bracis-33603	122	28	}	}	PUNCT
bracis-33603	122	29	}	}	PUNCT
bracis-33603	122	30	=	=	SYM
bracis-33603	122	31	\frac{100}{n	\frac{100}{n	PROPN
bracis-33603	122	32	}	}	PUNCT
bracis-33603	122	33	\sum	\sum	NOUN
bracis-33603	122	34	_	_	PUNCT
bracis-33603	122	35	{	{	PUNCT
bracis-33603	122	36	i=1}^{n	i=1}^{n	ADJ
bracis-33603	122	37	}	}	PUNCT
bracis-33603	122	38	\frac{{\text	\frac{{\text	NOUN
bracis-33603	122	39	{	{	PUNCT
bracis-33603	122	40	[	[	PUNCT
bracis-33603	122	41	raw	raw	ADJ
bracis-33603	122	42	preferred	preferred	ADJ
bracis-33603	122	43	metric]}}_i	metric]}}_i	PROPN
bracis-33603	122	44	{	{	PUNCT
bracis-33603	122	45	\text	\text	NOUN
bracis-33603	122	46	{	{	PUNCT
bracis-33603	122	47	[	[	X
bracis-33603	122	48	random	random	ADJ
bracis-33603	122	49	score]}}_i}{{\text	score]}}_i}{{\text	NOUN
bracis-33603	122	50	{	{	PUNCT
bracis-33603	122	51	[	[	X
bracis-33603	122	52	maximum	maximum	ADJ
bracis-33603	122	53	score]}}_i	score]}}_i	VERB
bracis-33603	122	54	-	-	PUNCT
bracis-33603	122	55	{	{	PUNCT
bracis-33603	122	56	\text	\text	NOUN
bracis-33603	122	57	{	{	PUNCT
bracis-33603	122	58	[	[	X
bracis-33603	122	59	random	random	ADJ
bracis-33603	122	60	score]}}_i	score]}}_i	NOUN
bracis-33603	122	61	}	}	PUNCT
bracis-33603	122	62	\end{aligned}$$	\end{aligned}$$	X
bracis-33603	122	63	(	(	PUNCT
bracis-33603	122	64	1	1	NUM
bracis-33603	122	65	)	)	PUNCT
bracis-33603	122	66	given	give	VERB
bracis-33603	122	67	that	that	SCONJ
bracis-33603	122	68	monoptt5	monoptt5	PROPN
bracis-33603	122	69	rerankers	reranker	NOUN
bracis-33603	122	70	were	be	AUX
bracis-33603	122	71	exclusively	exclusively	ADV
bracis-33603	122	72	trained	train	VERB
bracis-33603	122	73	starting	start	VERB
bracis-33603	122	74	from	from	ADP
bracis-33603	122	75	ptt5	ptt5	PROPN
bracis-33603	122	76	-	-	PUNCT
bracis-33603	122	77	v2	v2	PROPN
bracis-33603	122	78	pretrained	pretraine	VERB
bracis-33603	122	79	checkpoints	checkpoint	NOUN
bracis-33603	122	80	,	,	PUNCT
bracis-33603	122	81	retrieval	retrieval	NOUN
bracis-33603	122	82	tasks	task	NOUN
bracis-33603	122	83	were	be	AUX
bracis-33603	122	84	excluded	exclude	VERB
bracis-33603	122	85	from	from	ADP
bracis-33603	122	86	this	this	DET
bracis-33603	122	87	evaluation	evaluation	NOUN
bracis-33603	122	88	.	.	PUNCT
bracis-33603	123	1	therefore	therefore	ADV
bracis-33603	123	2	,	,	PUNCT
bracis-33603	123	3	we	we	PRON
bracis-33603	123	4	only	only	ADV
bracis-33603	123	5	considered	consider	VERB
bracis-33603	123	6	assin2	assin2	PROPN
bracis-33603	123	7	rte	rte	NOUN
bracis-33603	123	8	,	,	PUNCT
bracis-33603	123	9	assin2	assin2	PROPN
bracis-33603	123	10	sts	st	NOUN
bracis-33603	123	11	,	,	PUNCT
bracis-33603	123	12	and	and	CCONJ
bracis-33603	123	13	tweetsentbr	tweetsentbr	NOUN
bracis-33603	123	14	tasks	task	VERB
bracis-33603	123	15	.	.	PUNCT
bracis-33603	124	1	for	for	ADP
bracis-33603	124	2	each	each	DET
bracis-33603	124	3	model	model	NOUN
bracis-33603	124	4	,	,	PUNCT
bracis-33603	124	5	we	we	PRON
bracis-33603	124	6	calculate	calculate	VERB
bracis-33603	124	7	its	its	PRON
bracis-33603	124	8	aggregate	aggregate	ADJ
bracis-33603	124	9	performance	performance	NOUN
bracis-33603	124	10	by	by	ADP
bracis-33603	124	11	first	first	ADV
bracis-33603	124	12	determining	determine	VERB
bracis-33603	124	13	the	the	DET
bracis-33603	124	14	npm	npm	NOUN
bracis-33603	124	15	for	for	ADP
bracis-33603	124	16	each	each	DET
bracis-33603	124	17	task	task	NOUN
bracis-33603	124	18	and	and	CCONJ
bracis-33603	124	19	then	then	ADV
bracis-33603	124	20	computing	compute	VERB
bracis-33603	124	21	the	the	DET
bracis-33603	124	22	average	average	NOUN
bracis-33603	124	23	of	of	ADP
bracis-33603	124	24	these	these	DET
bracis-33603	124	25	values	value	NOUN
bracis-33603	124	26	.	.	PUNCT
bracis-33603	125	1	our	our	PRON
bracis-33603	125	2	ptt5	ptt5	NOUN
bracis-33603	125	3	-	-	PUNCT
bracis-33603	125	4	v2	v2	NOUN
bracis-33603	125	5	models	model	NOUN
bracis-33603	125	6	have	have	VERB
bracis-33603	125	7	higher	high	ADJ
bracis-33603	125	8	npm	npm	NOUN
bracis-33603	125	9	values	value	NOUN
bracis-33603	125	10	than	than	ADP
bracis-33603	125	11	t5	t5	PROPN
bracis-33603	125	12	and	and	CCONJ
bracis-33603	125	13	mt5	mt5	ADJ
bracis-33603	125	14	models	model	NOUN
bracis-33603	125	15	with	with	ADP
bracis-33603	125	16	considerably	considerably	ADV
bracis-33603	125	17	more	more	ADJ
bracis-33603	125	18	parameters	parameter	NOUN
bracis-33603	125	19	:	:	PUNCT
bracis-33603	125	20	for	for	ADP
bracis-33603	125	21	example	example	NOUN
bracis-33603	125	22	,	,	PUNCT
bracis-33603	125	23	ptt5	ptt5	NOUN
bracis-33603	125	24	-	-	PUNCT
bracis-33603	125	25	v2	v2	NOUN
bracis-33603	125	26	-	-	PUNCT
bracis-33603	125	27	base	base	NOUN
bracis-33603	125	28	is	be	AUX
bracis-33603	125	29	only	only	ADV
bracis-33603	125	30	surpassed	surpass	VERB
bracis-33603	125	31	by	by	ADP
bracis-33603	125	32	t5	t5	PROPN
bracis-33603	125	33	-	-	PUNCT
bracis-33603	125	34	3b	3b	NOUN
bracis-33603	125	35	(	(	PUNCT
bracis-33603	125	36	\(\sim	\(\sim	NOUN
bracis-33603	125	37	\)13.6x	\)13.6x	VERB
bracis-33603	125	38	larger	large	ADJ
bracis-33603	125	39	)	)	PUNCT
bracis-33603	125	40	and	and	CCONJ
bracis-33603	125	41	t5	t5	PROPN
bracis-33603	125	42	-	-	PUNCT
bracis-33603	125	43	xl	xl	PROPN
bracis-33603	125	44	(	(	PUNCT
bracis-33603	125	45	\(\sim	\(\sim	NOUN
bracis-33603	125	46	\)16.81x	\)16.81x	X
bracis-33603	125	47	larger	large	ADJ
bracis-33603	125	48	)	)	PUNCT
bracis-33603	125	49	;	;	PUNCT
bracis-33603	125	50	this	this	DET
bracis-33603	125	51	performance	performance	NOUN
bracis-33603	125	52	gap	gap	NOUN
bracis-33603	125	53	,	,	PUNCT
bracis-33603	125	54	however	however	ADV
bracis-33603	125	55	,	,	PUNCT
bracis-33603	125	56	is	be	AUX
bracis-33603	125	57	most	most	ADV
bracis-33603	125	58	pronounced	pronounce	VERB
bracis-33603	125	59	in	in	ADP
bracis-33603	125	60	smaller	small	ADJ
bracis-33603	125	61	models	model	NOUN
bracis-33603	125	62	,	,	PUNCT
bracis-33603	125	63	narrowing	narrow	VERB
bracis-33603	125	64	as	as	ADP
bracis-33603	125	65	model	model	NOUN
bracis-33603	125	66	size	size	NOUN
bracis-33603	125	67	increases	increase	NOUN
bracis-33603	125	68	.	.	PUNCT
bracis-33603	126	1	a	a	DET
bracis-33603	126	2	similar	similar	ADJ
bracis-33603	126	3	result	result	NOUN
bracis-33603	126	4	was	be	AUX
bracis-33603	126	5	also	also	ADV
bracis-33603	126	6	observed	observe	VERB
bracis-33603	126	7	by	by	ADP
bracis-33603	126	8	xue	xue	PROPN
bracis-33603	126	9	et	et	PROPN
bracis-33603	126	10	al	al	PROPN
bracis-33603	126	11	.	.	PUNCT
bracis-33603	127	1	[	[	X
bracis-33603	127	2	48	48	NUM
bracis-33603	127	3	]	]	PUNCT
bracis-33603	127	4	,	,	PUNCT
bracis-33603	127	5	which	which	PRON
bracis-33603	127	6	analyzed	analyze	VERB
bracis-33603	127	7	the	the	DET
bracis-33603	127	8	performance	performance	NOUN
bracis-33603	127	9	of	of	ADP
bracis-33603	127	10	t5	t5	PROPN
bracis-33603	127	11	and	and	CCONJ
bracis-33603	127	12	mt5	mt5	ADJ
bracis-33603	127	13	models	model	NOUN
bracis-33603	127	14	performance	performance	NOUN
bracis-33603	127	15	on	on	ADP
bracis-33603	127	16	the	the	DET
bracis-33603	127	17	squad	squad	ADJ
bracis-33603	127	18	benchmark	benchmark	NOUN
bracis-33603	127	19	[	[	X
bracis-33603	127	20	35	35	NUM
bracis-33603	127	21	]	]	PUNCT
bracis-33603	127	22	,	,	PUNCT
bracis-33603	127	23	observing	observe	VERB
bracis-33603	127	24	a	a	DET
bracis-33603	127	25	performance	performance	NOUN
bracis-33603	127	26	gap	gap	NOUN
bracis-33603	127	27	between	between	ADP
bracis-33603	127	28	t5	t5	PROPN
bracis-33603	127	29	-	-	PUNCT
bracis-33603	127	30	small	small	ADJ
bracis-33603	127	31	and	and	CCONJ
bracis-33603	127	32	t5	t5	PROPN
bracis-33603	127	33	-	-	PUNCT
bracis-33603	127	34	base	base	NOUN
bracis-33603	127	35	vs	vs	ADP
bracis-33603	127	36	mt5	mt5	ADJ
bracis-33603	127	37	models	model	NOUN
bracis-33603	127	38	of	of	ADP
bracis-33603	127	39	equivalent	equivalent	ADJ
bracis-33603	127	40	sizes	size	NOUN
bracis-33603	127	41	,	,	PUNCT
bracis-33603	127	42	which	which	PRON
bracis-33603	127	43	is	be	AUX
bracis-33603	127	44	diminished	diminish	VERB
bracis-33603	127	45	starting	start	VERB
bracis-33603	127	46	from	from	ADP
bracis-33603	127	47	t5	t5	PROPN
bracis-33603	127	48	-	-	PUNCT
bracis-33603	127	49	large	large	ADJ
bracis-33603	127	50	.	.	PUNCT
bracis-33603	128	1	this	this	DET
bracis-33603	128	2	performance	performance	NOUN
bracis-33603	128	3	gap	gap	NOUN
bracis-33603	128	4	observed	observe	VERB
bracis-33603	128	5	on	on	ADP
bracis-33603	128	6	smaller	small	ADJ
bracis-33603	128	7	model	model	NOUN
bracis-33603	128	8	sizes	size	NOUN
bracis-33603	128	9	is	be	AUX
bracis-33603	128	10	advantageous	advantageous	ADJ
bracis-33603	128	11	when	when	SCONJ
bracis-33603	128	12	we	we	PRON
bracis-33603	128	13	consider	consider	VERB
bracis-33603	128	14	environments	environment	NOUN
bracis-33603	128	15	constrained	constrain	VERB
bracis-33603	128	16	by	by	ADP
bracis-33603	128	17	computational	computational	ADJ
bracis-33603	128	18	resources	resource	NOUN
bracis-33603	128	19	,	,	PUNCT
bracis-33603	128	20	increasing	increase	VERB
bracis-33603	128	21	the	the	DET
bracis-33603	128	22	maximum	maximum	ADJ
bracis-33603	128	23	attainable	attainable	ADJ
bracis-33603	128	24	level	level	NOUN
bracis-33603	128	25	of	of	ADP
bracis-33603	128	26	performance	performance	NOUN
bracis-33603	128	27	;	;	PUNCT
bracis-33603	128	28	additionally	additionally	ADV
bracis-33603	128	29	,	,	PUNCT
bracis-33603	128	30	a	a	DET
bracis-33603	128	31	language	language	NOUN
bracis-33603	128	32	-	-	PUNCT
bracis-33603	128	33	specific	specific	ADJ
bracis-33603	128	34	tokenizer	tokenizer	NOUN
bracis-33603	128	35	reduces	reduce	VERB
bracis-33603	128	36	text	text	NOUN
bracis-33603	128	37	splitting	splitting	NOUN
bracis-33603	128	38	into	into	ADP
bracis-33603	128	39	fewer	few	ADJ
bracis-33603	128	40	tokens	token	NOUN
bracis-33603	128	41	,	,	PUNCT
bracis-33603	128	42	leading	lead	VERB
bracis-33603	128	43	to	to	ADP
bracis-33603	128	44	lower	low	ADJ
bracis-33603	128	45	latency	latency	NOUN
bracis-33603	128	46	and	and	CCONJ
bracis-33603	128	47	the	the	DET
bracis-33603	128	48	potential	potential	NOUN
bracis-33603	128	49	to	to	PART
bracis-33603	128	50	accommodate	accommodate	VERB
bracis-33603	128	51	more	more	ADJ
bracis-33603	128	52	text	text	NOUN
bracis-33603	128	53	within	within	ADP
bracis-33603	128	54	the	the	DET
bracis-33603	128	55	same	same	ADJ
bracis-33603	128	56	maximum	maximum	ADJ
bracis-33603	128	57	token	token	ADJ
bracis-33603	128	58	context	context	NOUN
bracis-33603	128	59	window	window	NOUN
bracis-33603	128	60	.	.	PUNCT
bracis-33603	129	1	interestingly	interestingly	ADV
bracis-33603	129	2	,	,	PUNCT
bracis-33603	129	3	mt5	mt5	ADJ
bracis-33603	129	4	models	model	NOUN
bracis-33603	129	5	tend	tend	VERB
bracis-33603	129	6	to	to	PART
bracis-33603	129	7	show	show	VERB
bracis-33603	129	8	lower	low	ADJ
bracis-33603	129	9	npm	npm	PROPN
bracis-33603	129	10	values	value	NOUN
bracis-33603	129	11	,	,	PUNCT
bracis-33603	129	12	except	except	SCONJ
bracis-33603	129	13	in	in	ADP
bracis-33603	129	14	the	the	DET
bracis-33603	129	15	3	3	NUM
bracis-33603	129	16	billion	billion	NUM
bracis-33603	129	17	parameter	parameter	NOUN
bracis-33603	129	18	range	range	NOUN
bracis-33603	129	19	,	,	PUNCT
bracis-33603	129	20	where	where	SCONJ
bracis-33603	129	21	they	they	PRON
bracis-33603	129	22	slightly	slightly	ADV
bracis-33603	129	23	outperform	outperform	VERB
bracis-33603	129	24	t5	t5	PROPN
bracis-33603	129	25	models	model	NOUN
bracis-33603	129	26	.	.	PUNCT
bracis-33603	130	1	for	for	ADP
bracis-33603	130	2	the	the	DET
bracis-33603	130	3	retrieval	retrieval	NOUN
bracis-33603	130	4	tasks	task	NOUN
bracis-33603	130	5	,	,	PUNCT
bracis-33603	130	6	our	our	PRON
bracis-33603	130	7	monoptt5	monoptt5	NOUN
bracis-33603	130	8	rerankers	reranker	NOUN
bracis-33603	130	9	were	be	AUX
bracis-33603	130	10	able	able	ADJ
bracis-33603	130	11	to	to	PART
bracis-33603	130	12	set	set	VERB
bracis-33603	130	13	new	new	ADJ
bracis-33603	130	14	sotas	sota	NOUN
bracis-33603	130	15	for	for	ADP
bracis-33603	130	16	both	both	CCONJ
bracis-33603	130	17	mmarco	mmarco	ADJ
bracis-33603	130	18	-	-	PUNCT
bracis-33603	130	19	pt	pt	ADJ
bracis-33603	130	20	and	and	CCONJ
bracis-33603	130	21	mrobust	mrobust	ADV
bracis-33603	130	22	-	-	PUNCT
bracis-33603	130	23	pt	pt	NOUN
bracis-33603	130	24	.	.	PROPN
bracis-33603	130	25	for	for	ADP
bracis-33603	130	26	the	the	DET
bracis-33603	130	27	mmarco	mmarco	ADJ
bracis-33603	130	28	-	-	PUNCT
bracis-33603	130	29	pt	pt	NOUN
bracis-33603	130	30	dataset	dataset	NOUN
bracis-33603	130	31	,	,	PUNCT
bracis-33603	130	32	models	model	NOUN
bracis-33603	130	33	starting	start	VERB
bracis-33603	130	34	from	from	ADP
bracis-33603	130	35	t5	t5	PROPN
bracis-33603	130	36	-	-	PUNCT
bracis-33603	130	37	base	base	NOUN
bracis-33603	130	38	size	size	NOUN
bracis-33603	130	39	were	be	AUX
bracis-33603	130	40	able	able	ADJ
bracis-33603	130	41	to	to	PART
bracis-33603	130	42	surpass	surpass	VERB
bracis-33603	130	43	the	the	DET
bracis-33603	130	44	current	current	ADJ
bracis-33603	130	45	sota	sota	NOUN
bracis-33603	130	46	;	;	PUNCT
bracis-33603	130	47	the	the	DET
bracis-33603	130	48	3	3	NUM
bracis-33603	130	49	billion	billion	NUM
bracis-33603	130	50	parameters	parameter	NOUN
bracis-33603	130	51	reranker	reranker	NOUN
bracis-33603	130	52	obtained	obtain	VERB
bracis-33603	130	53	a	a	DET
bracis-33603	130	54	gain	gain	NOUN
bracis-33603	130	55	of	of	ADP
bracis-33603	130	56	+0.026	+0.026	ADJ
bracis-33603	130	57	points	point	NOUN
bracis-33603	130	58	of	of	ADP
bracis-33603	130	59	mrr@10	mrr@10	PROPN
bracis-33603	130	60	.	.	PUNCT
bracis-33603	131	1	in	in	ADP
bracis-33603	131	2	the	the	DET
bracis-33603	131	3	mrobust	mrobust	ADV
bracis-33603	131	4	-	-	PUNCT
bracis-33603	131	5	pt	pt	NOUN
bracis-33603	131	6	,	,	PUNCT
bracis-33603	131	7	our	our	PRON
bracis-33603	131	8	large	large	ADJ
bracis-33603	131	9	and	and	CCONJ
bracis-33603	131	10	3b	3b	NUM
bracis-33603	131	11	rerankers	reranker	NOUN
bracis-33603	131	12	surpassed	surpass	VERB
bracis-33603	131	13	the	the	DET
bracis-33603	131	14	current	current	ADJ
bracis-33603	131	15	sota	sota	NOUN
bracis-33603	131	16	by	by	ADP
bracis-33603	131	17	+0.071	+0.071	PROPN
bracis-33603	131	18	and	and	CCONJ
bracis-33603	131	19	+0.121	+0.121	PRON
bracis-33603	131	20	in	in	ADP
bracis-33603	131	21	terms	term	NOUN
bracis-33603	131	22	of	of	ADP
bracis-33603	131	23	ndgc@20	ndgc@20	NOUN
bracis-33603	131	24	,	,	PUNCT
bracis-33603	131	25	respectively	respectively	ADV
bracis-33603	131	26	.	.	PUNCT
bracis-33603	132	1	a	a	DET
bracis-33603	132	2	more	more	ADV
bracis-33603	132	3	in	in	ADP
bracis-33603	132	4	-	-	PUNCT
bracis-33603	132	5	depth	depth	NOUN
bracis-33603	132	6	analysis	analysis	NOUN
bracis-33603	132	7	of	of	ADP
bracis-33603	132	8	the	the	DET
bracis-33603	132	9	retrieval	retrieval	NOUN
bracis-33603	132	10	tasks	task	NOUN
bracis-33603	132	11	is	be	AUX
bracis-33603	132	12	conducted	conduct	VERB
bracis-33603	132	13	on	on	ADP
bracis-33603	132	14	sect	sect	NOUN
bracis-33603	132	15	.	.	PUNCT
bracis-33603	132	16	 	 	SPACE
bracis-33603	133	1	5.2	5.2	NUM
bracis-33603	133	2	.	.	PUNCT
bracis-33603	134	1	5	5	NUM
bracis-33603	134	2	ablations	ablation	VERB
bracis-33603	134	3	5.1	5.1	NUM
bracis-33603	134	4	additional	additional	ADJ
bracis-33603	134	5	pretraining	pretraine	VERB
bracis-33603	134	6	experiments	experiment	NOUN
bracis-33603	134	7	this	this	DET
bracis-33603	134	8	sections	section	NOUN
bracis-33603	134	9	includes	include	VERB
bracis-33603	134	10	pretraining	pretraine	VERB
bracis-33603	134	11	experiments	experiment	NOUN
bracis-33603	134	12	additional	additional	ADJ
bracis-33603	134	13	to	to	ADP
bracis-33603	134	14	those	those	PRON
bracis-33603	134	15	described	describe	VERB
bracis-33603	134	16	in	in	ADP
bracis-33603	134	17	sect	sect	NOUN
bracis-33603	134	18	.	.	PUNCT
bracis-33603	134	19	 	 	SPACE
bracis-33603	135	1	3.1	3.1	NUM
bracis-33603	135	2	.	.	PUNCT
bracis-33603	135	3	comparison	comparison	NOUN
bracis-33603	135	4	with	with	ADP
bracis-33603	135	5	ptt5	ptt5	NOUN
bracis-33603	135	6	-	-	PUNCT
bracis-33603	135	7	v1	v1	NOUN
bracis-33603	135	8	:	:	PUNCT
bracis-33603	135	9	given	give	VERB
bracis-33603	135	10	the	the	DET
bracis-33603	135	11	title	title	NOUN
bracis-33603	135	12	of	of	ADP
bracis-33603	135	13	our	our	PRON
bracis-33603	135	14	work	work	NOUN
bracis-33603	135	15	,	,	PUNCT
bracis-33603	135	16	a	a	DET
bracis-33603	135	17	pertinent	pertinent	ADJ
bracis-33603	135	18	question	question	NOUN
bracis-33603	135	19	arises	arise	VERB
bracis-33603	135	20	:	:	PUNCT
bracis-33603	135	21	how	how	SCONJ
bracis-33603	135	22	do	do	AUX
bracis-33603	135	23	ptt5	ptt5	NOUN
bracis-33603	135	24	-	-	PUNCT
bracis-33603	135	25	v2	v2	NOUN
bracis-33603	135	26	models	model	NOUN
bracis-33603	135	27	compare	compare	VERB
bracis-33603	135	28	with	with	ADP
bracis-33603	135	29	the	the	DET
bracis-33603	135	30	work	work	NOUN
bracis-33603	135	31	in	in	ADP
bracis-33603	135	32	ptt5	ptt5	NOUN
bracis-33603	135	33	-	-	PUNCT
bracis-33603	135	34	v1	v1	NOUN
bracis-33603	135	35	?	?	PUNCT
bracis-33603	136	1	a	a	DET
bracis-33603	136	2	few	few	ADJ
bracis-33603	136	3	key	key	ADJ
bracis-33603	136	4	differences	difference	NOUN
bracis-33603	136	5	exist	exist	VERB
bracis-33603	136	6	in	in	ADP
bracis-33603	136	7	the	the	DET
bracis-33603	136	8	pretraining	pretraining	NOUN
bracis-33603	136	9	of	of	ADP
bracis-33603	136	10	ptt5	ptt5	NOUN
bracis-33603	136	11	-	-	PUNCT
bracis-33603	136	12	v1	v1	NOUN
bracis-33603	136	13	.	.	PUNCT
bracis-33603	137	1	notably	notably	ADV
bracis-33603	137	2	,	,	PUNCT
bracis-33603	137	3	it	it	PRON
bracis-33603	137	4	utilized	utilize	VERB
bracis-33603	137	5	brwac	brwac	PROPN
bracis-33603	137	6	,	,	PUNCT
bracis-33603	137	7	a	a	DET
bracis-33603	137	8	considerably	considerably	ADV
bracis-33603	137	9	smaller	small	ADJ
bracis-33603	137	10	dataset	dataset	NOUN
bracis-33603	137	11	,	,	PUNCT
bracis-33603	137	12	and	and	CCONJ
bracis-33603	137	13	a	a	DET
bracis-33603	137	14	slightly	slightly	ADV
bracis-33603	137	15	different	different	ADJ
bracis-33603	137	16	pretraining	pretraine	VERB
bracis-33603	137	17	objective	objective	NOUN
bracis-33603	137	18	(	(	PUNCT
bracis-33603	137	19	denoising	denoising	NOUN
bracis-33603	137	20	,	,	PUNCT
bracis-33603	137	21	where	where	SCONJ
bracis-33603	137	22	some	some	DET
bracis-33603	137	23	input	input	NOUN
bracis-33603	137	24	tokens	token	NOUN
bracis-33603	137	25	are	be	AUX
bracis-33603	137	26	masked	mask	VERB
bracis-33603	137	27	and	and	CCONJ
bracis-33603	137	28	the	the	DET
bracis-33603	137	29	model	model	NOUN
bracis-33603	137	30	is	be	AUX
bracis-33603	137	31	trained	train	VERB
bracis-33603	137	32	to	to	PART
bracis-33603	137	33	predict	predict	VERB
bracis-33603	137	34	the	the	DET
bracis-33603	137	35	original	original	ADJ
bracis-33603	137	36	text	text	NOUN
bracis-33603	137	37	,	,	PUNCT
bracis-33603	137	38	rather	rather	ADV
bracis-33603	137	39	than	than	ADP
bracis-33603	137	40	span	span	VERB
bracis-33603	137	41	corruption	corruption	NOUN
bracis-33603	137	42	)	)	PUNCT
bracis-33603	137	43	.	.	PUNCT
bracis-33603	138	1	additionally	additionally	ADV
bracis-33603	138	2	,	,	PUNCT
bracis-33603	138	3	ptt5	ptt5	NOUN
bracis-33603	138	4	-	-	PUNCT
bracis-33603	138	5	v1	v1	NOUN
bracis-33603	138	6	employed	employ	VERB
bracis-33603	138	7	models	model	NOUN
bracis-33603	138	8	ranging	range	VERB
bracis-33603	138	9	from	from	ADP
bracis-33603	138	10	t5	t5	PROPN
bracis-33603	138	11	-	-	PUNCT
bracis-33603	138	12	small	small	ADJ
bracis-33603	138	13	to	to	ADP
bracis-33603	138	14	t5	t5	PROPN
bracis-33603	138	15	-	-	PUNCT
bracis-33603	138	16	large	large	NOUN
bracis-33603	138	17	with	with	ADP
bracis-33603	138	18	an	an	DET
bracis-33603	138	19	adafactor	adafactor	NOUN
bracis-33603	138	20	optimizer	optimizer	NOUN
bracis-33603	138	21	using	use	VERB
bracis-33603	138	22	a	a	DET
bracis-33603	138	23	learning	learning	NOUN
bracis-33603	138	24	rate	rate	NOUN
bracis-33603	138	25	of	of	ADP
bracis-33603	138	26	0.003	0.003	NUM
bracis-33603	138	27	(	(	PUNCT
bracis-33603	138	28	threefold	threefold	ADV
bracis-33603	138	29	larger	large	ADJ
bracis-33603	138	30	than	than	ADP
bracis-33603	138	31	our	our	PRON
bracis-33603	138	32	setting	setting	NOUN
bracis-33603	138	33	)	)	PUNCT
bracis-33603	138	34	.	.	PUNCT
bracis-33603	139	1	ptt5	ptt5	NOUN
bracis-33603	139	2	-	-	PUNCT
bracis-33603	139	3	v1	v1	NOUN
bracis-33603	139	4	also	also	ADV
bracis-33603	139	5	explored	explore	VERB
bracis-33603	139	6	both	both	DET
bracis-33603	139	7	t5	t5	PROPN
bracis-33603	139	8	’s	’s	PART
bracis-33603	139	9	original	original	ADJ
bracis-33603	139	10	english	english	ADJ
bracis-33603	139	11	vocabulary	vocabulary	NOUN
bracis-33603	139	12	and	and	CCONJ
bracis-33603	139	13	a	a	DET
bracis-33603	139	14	portuguese	portuguese	ADJ
bracis-33603	139	15	language	language	NOUN
bracis-33603	139	16	-	-	PUNCT
bracis-33603	139	17	specific	specific	ADJ
bracis-33603	139	18	tokenizer	tokenizer	NOUN
bracis-33603	139	19	.	.	PUNCT
bracis-33603	140	1	in	in	ADP
bracis-33603	140	2	contrast	contrast	NOUN
bracis-33603	140	3	,	,	PUNCT
bracis-33603	140	4	ptt5	ptt5	NOUN
bracis-33603	140	5	-	-	PUNCT
bracis-33603	140	6	v2	v2	PROPN
bracis-33603	140	7	exclusively	exclusively	ADV
bracis-33603	140	8	uses	use	VERB
bracis-33603	140	9	the	the	DET
bracis-33603	140	10	latter	latter	ADJ
bracis-33603	140	11	.	.	PUNCT
bracis-33603	141	1	to	to	PART
bracis-33603	141	2	ensure	ensure	VERB
bracis-33603	141	3	a	a	DET
bracis-33603	141	4	fair	fair	ADJ
bracis-33603	141	5	comparison	comparison	NOUN
bracis-33603	141	6	,	,	PUNCT
bracis-33603	141	7	we	we	PRON
bracis-33603	141	8	finetuned	finetune	VERB
bracis-33603	141	9	the	the	DET
bracis-33603	141	10	ptt5	ptt5	NOUN
bracis-33603	141	11	-	-	PUNCT
bracis-33603	141	12	v1	v1	NOUN
bracis-33603	141	13	checkpoints	checkpoint	NOUN
bracis-33603	141	14	following	follow	VERB
bracis-33603	141	15	the	the	DET
bracis-33603	141	16	methodology	methodology	NOUN
bracis-33603	141	17	outlined	outline	VERB
bracis-33603	141	18	in	in	ADP
bracis-33603	141	19	sect	sect	NOUN
bracis-33603	141	20	.	.	PUNCT
bracis-33603	141	21	 	 	SPACE
bracis-33603	142	1	3.2	3.2	NUM
bracis-33603	142	2	.	.	PUNCT
bracis-33603	143	1	figure	figure	NOUN
bracis-33603	143	2	 	 	SPACE
bracis-33603	143	3	1	1	NUM
bracis-33603	143	4	presents	present	VERB
bracis-33603	143	5	the	the	DET
bracis-33603	143	6	npm	npm	PROPN
bracis-33603	143	7	values	value	NOUN
bracis-33603	143	8	for	for	ADP
bracis-33603	143	9	both	both	PRON
bracis-33603	143	10	ptt5	ptt5	NOUN
bracis-33603	143	11	-	-	PUNCT
bracis-33603	143	12	v1	v1	NOUN
bracis-33603	143	13	and	and	CCONJ
bracis-33603	143	14	ptt5	ptt5	NOUN
bracis-33603	143	15	-	-	PUNCT
bracis-33603	143	16	v2	v2	PROPN
bracis-33603	143	17	,	,	PUNCT
bracis-33603	143	18	alongside	alongside	ADP
bracis-33603	143	19	comparisons	comparison	NOUN
bracis-33603	143	20	to	to	ADP
bracis-33603	143	21	mt5	mt5	ADJ
bracis-33603	143	22	and	and	CCONJ
bracis-33603	143	23	t5	t5	PROPN
bracis-33603	143	24	models	model	NOUN
bracis-33603	143	25	.	.	PUNCT
bracis-33603	144	1	the	the	DET
bracis-33603	144	2	data	datum	NOUN
bracis-33603	144	3	corroborates	corroborate	VERB
bracis-33603	144	4	the	the	DET
bracis-33603	144	5	enhancements	enhancement	NOUN
bracis-33603	144	6	achievable	achievable	ADJ
bracis-33603	144	7	through	through	ADP
bracis-33603	144	8	monolingual	monolingual	ADJ
bracis-33603	144	9	pretraining	pretraine	VERB
bracis-33603	144	10	on	on	ADP
bracis-33603	144	11	the	the	DET
bracis-33603	144	12	target	target	NOUN
bracis-33603	144	13	language	language	NOUN
bracis-33603	144	14	,	,	PUNCT
bracis-33603	144	15	which	which	PRON
bracis-33603	144	16	is	be	AUX
bracis-33603	144	17	further	far	ADV
bracis-33603	144	18	augmented	augment	VERB
bracis-33603	144	19	by	by	ADP
bracis-33603	144	20	employing	employ	VERB
bracis-33603	144	21	a	a	DET
bracis-33603	144	22	dedicated	dedicated	ADJ
bracis-33603	144	23	tokenizer	tokenizer	NOUN
bracis-33603	144	24	.	.	PUNCT
bracis-33603	145	1	in	in	ADP
bracis-33603	145	2	comparison	comparison	NOUN
bracis-33603	145	3	between	between	ADP
bracis-33603	145	4	the	the	DET
bracis-33603	145	5	two	two	NUM
bracis-33603	145	6	ptt5	ptt5	PROPN
bracis-33603	145	7	iterations	iteration	NOUN
bracis-33603	145	8	,	,	PUNCT
bracis-33603	145	9	a	a	DET
bracis-33603	145	10	performance	performance	NOUN
bracis-33603	145	11	disparity	disparity	NOUN
bracis-33603	145	12	favoring	favor	VERB
bracis-33603	145	13	ptt5	ptt5	PROPN
bracis-33603	145	14	-	-	PUNCT
bracis-33603	145	15	v2	v2	PROPN
bracis-33603	145	16	is	be	AUX
bracis-33603	145	17	apparent	apparent	ADJ
bracis-33603	145	18	for	for	ADP
bracis-33603	145	19	the	the	DET
bracis-33603	145	20	small	small	ADJ
bracis-33603	145	21	and	and	CCONJ
bracis-33603	145	22	large	large	ADJ
bracis-33603	145	23	sizes	size	NOUN
bracis-33603	145	24	,	,	PUNCT
bracis-33603	145	25	with	with	ADP
bracis-33603	145	26	the	the	DET
bracis-33603	145	27	largest	large	ADJ
bracis-33603	145	28	gap	gap	NOUN
bracis-33603	145	29	observed	observe	VERB
bracis-33603	145	30	for	for	ADP
bracis-33603	145	31	large	large	ADJ
bracis-33603	145	32	;	;	PUNCT
bracis-33603	145	33	however	however	ADV
bracis-33603	145	34	,	,	PUNCT
bracis-33603	145	35	models	model	NOUN
bracis-33603	145	36	of	of	ADP
bracis-33603	145	37	the	the	DET
bracis-33603	145	38	base	base	NOUN
bracis-33603	145	39	size	size	NOUN
bracis-33603	145	40	exhibit	exhibit	VERB
bracis-33603	145	41	marginally	marginally	ADV
bracis-33603	145	42	superior	superior	ADJ
bracis-33603	145	43	performance	performance	NOUN
bracis-33603	145	44	in	in	ADP
bracis-33603	145	45	the	the	DET
bracis-33603	145	46	ptt5	ptt5	NOUN
bracis-33603	145	47	-	-	PUNCT
bracis-33603	145	48	v1	v1	NOUN
bracis-33603	145	49	variant	variant	NOUN
bracis-33603	145	50	.	.	PUNCT
bracis-33603	146	1	surprisingly	surprisingly	ADV
bracis-33603	146	2	,	,	PUNCT
bracis-33603	146	3	mt5	mt5	ADJ
bracis-33603	146	4	’s	’s	PART
bracis-33603	146	5	performance	performance	NOUN
bracis-33603	146	6	lags	lag	VERB
bracis-33603	146	7	behind	behind	ADP
bracis-33603	146	8	all	all	DET
bracis-33603	146	9	other	other	ADJ
bracis-33603	146	10	models	model	NOUN
bracis-33603	146	11	,	,	PUNCT
bracis-33603	146	12	including	include	VERB
bracis-33603	146	13	the	the	DET
bracis-33603	146	14	monolingual	monolingual	ADJ
bracis-33603	146	15	english	english	PROPN
bracis-33603	146	16	t5	t5	PROPN
bracis-33603	146	17	,	,	PUNCT
bracis-33603	146	18	except	except	SCONJ
bracis-33603	146	19	in	in	ADP
bracis-33603	146	20	the	the	DET
bracis-33603	146	21	3b	3b	NUM
bracis-33603	146	22	parameter	parameter	NOUN
bracis-33603	146	23	range	range	NOUN
bracis-33603	146	24	,	,	PUNCT
bracis-33603	146	25	where	where	SCONJ
bracis-33603	146	26	it	it	PRON
bracis-33603	146	27	closely	closely	ADV
bracis-33603	146	28	matches	match	VERB
bracis-33603	146	29	ptt5	ptt5	NOUN
bracis-33603	146	30	-	-	PUNCT
bracis-33603	146	31	v2	v2	PROPN
bracis-33603	146	32	and	and	CCONJ
bracis-33603	146	33	surpasses	surpasse	NOUN
bracis-33603	146	34	t5	t5	PROPN
bracis-33603	146	35	.	.	PUNCT
bracis-33603	147	1	fig	fig	PROPN
bracis-33603	147	2	.	.	PUNCT
bracis-33603	148	1	1	1	X
bracis-33603	148	2	.	.	X
bracis-33603	148	3	npm	npm	PROPN
bracis-33603	148	4	vs	vs	ADP
bracis-33603	148	5	parameters	parameter	NOUN
bracis-33603	148	6	for	for	ADP
bracis-33603	148	7	varying	vary	VERB
bracis-33603	148	8	t5	t5	PROPN
bracis-33603	148	9	configurations	configuration	NOUN
bracis-33603	148	10	.	.	PUNCT
bracis-33603	149	1	full	full	ADJ
bracis-33603	149	2	size	size	NOUN
bracis-33603	149	3	image	image	NOUN
bracis-33603	149	4	quality	quality	NOUN
bracis-33603	149	5	filters	filter	NOUN
bracis-33603	149	6	:	:	PUNCT
bracis-33603	149	7	in	in	ADP
bracis-33603	149	8	our	our	PRON
bracis-33603	149	9	primary	primary	ADJ
bracis-33603	149	10	experiments	experiment	NOUN
bracis-33603	149	11	,	,	PUNCT
bracis-33603	149	12	we	we	PRON
bracis-33603	149	13	utilized	utilize	VERB
bracis-33603	149	14	the	the	DET
bracis-33603	149	15	entire	entire	ADJ
bracis-33603	149	16	mc4	mc4	PROPN
bracis-33603	149	17	-	-	PUNCT
bracis-33603	149	18	pt	pt	NOUN
bracis-33603	149	19	dataset	dataset	NOUN
bracis-33603	149	20	,	,	PUNCT
bracis-33603	149	21	containing	contain	VERB
bracis-33603	149	22	approximately	approximately	ADV
bracis-33603	149	23	116	116	NUM
bracis-33603	149	24	billion	billion	NUM
bracis-33603	149	25	training	training	NOUN
bracis-33603	149	26	tokens	token	NOUN
bracis-33603	149	27	,	,	PUNCT
bracis-33603	149	28	in	in	ADP
bracis-33603	149	29	the	the	DET
bracis-33603	149	30	pretraining	pretraine	VERB
bracis-33603	149	31	phase	phase	NOUN
bracis-33603	149	32	.	.	PUNCT
bracis-33603	150	1	in	in	ADP
bracis-33603	150	2	this	this	DET
bracis-33603	150	3	additional	additional	ADJ
bracis-33603	150	4	experiment	experiment	NOUN
bracis-33603	150	5	,	,	PUNCT
bracis-33603	150	6	we	we	PRON
bracis-33603	150	7	consider	consider	VERB
bracis-33603	150	8	the	the	DET
bracis-33603	150	9	massivetext	massivetext	ADJ
bracis-33603	150	10	quality	quality	NOUN
bracis-33603	150	11	filters	filter	NOUN
bracis-33603	150	12	[	[	X
bracis-33603	150	13	33	33	NUM
bracis-33603	150	14	]	]	PUNCT
bracis-33603	150	15	to	to	PART
bracis-33603	150	16	investigate	investigate	VERB
bracis-33603	150	17	the	the	DET
bracis-33603	150	18	impact	impact	NOUN
bracis-33603	150	19	of	of	ADP
bracis-33603	150	20	this	this	DET
bracis-33603	150	21	filtering	filtering	NOUN
bracis-33603	150	22	process	process	NOUN
bracis-33603	150	23	on	on	ADP
bracis-33603	150	24	downstream	downstream	ADJ
bracis-33603	150	25	tasks	task	NOUN
bracis-33603	150	26	.	.	PUNCT
bracis-33603	151	1	applying	apply	VERB
bracis-33603	151	2	these	these	DET
bracis-33603	151	3	filters	filter	NOUN
bracis-33603	151	4	to	to	ADP
bracis-33603	151	5	mc4	mc4	PROPN
bracis-33603	151	6	-	-	PUNCT
bracis-33603	151	7	pt	pt	PROPN
bracis-33603	151	8	reduces	reduce	VERB
bracis-33603	151	9	the	the	DET
bracis-33603	151	10	number	number	NOUN
bracis-33603	151	11	of	of	ADP
bracis-33603	151	12	training	training	NOUN
bracis-33603	151	13	tokens	token	NOUN
bracis-33603	151	14	to	to	ADP
bracis-33603	151	15	approximately	approximately	ADV
bracis-33603	151	16	82	82	NUM
bracis-33603	151	17	billion	billion	NUM
bracis-33603	151	18	,	,	PUNCT
bracis-33603	151	19	a	a	DET
bracis-33603	151	20	reduction	reduction	NOUN
bracis-33603	151	21	of	of	ADP
bracis-33603	151	22	about	about	ADV
bracis-33603	151	23	30	30	NUM
bracis-33603	151	24	%	%	NOUN
bracis-33603	151	25	.	.	PUNCT
bracis-33603	152	1	this	this	DET
bracis-33603	152	2	experiment	experiment	NOUN
bracis-33603	152	3	was	be	AUX
bracis-33603	152	4	restricted	restrict	VERB
bracis-33603	152	5	to	to	ADP
bracis-33603	152	6	models	model	NOUN
bracis-33603	152	7	of	of	ADP
bracis-33603	152	8	t5	t5	PROPN
bracis-33603	152	9	-	-	PUNCT
bracis-33603	152	10	base	base	NOUN
bracis-33603	152	11	size	size	NOUN
bracis-33603	152	12	,	,	PUNCT
bracis-33603	152	13	keeping	keep	VERB
bracis-33603	152	14	the	the	DET
bracis-33603	152	15	same	same	ADJ
bracis-33603	152	16	batch	batch	NOUN
bracis-33603	152	17	size	size	NOUN
bracis-33603	152	18	and	and	CCONJ
bracis-33603	152	19	optimization	optimization	NOUN
bracis-33603	152	20	strategy	strategy	NOUN
bracis-33603	152	21	used	use	VERB
bracis-33603	152	22	in	in	ADP
bracis-33603	152	23	the	the	DET
bracis-33603	152	24	main	main	ADJ
bracis-33603	152	25	experiments	experiment	NOUN
bracis-33603	152	26	.	.	PUNCT
bracis-33603	153	1	figure	figure	NOUN
bracis-33603	153	2	 	 	SPACE
bracis-33603	153	3	2	2	NUM
bracis-33603	153	4	shows	show	VERB
bracis-33603	153	5	the	the	DET
bracis-33603	153	6	effect	effect	NOUN
bracis-33603	153	7	of	of	ADP
bracis-33603	153	8	using	use	VERB
bracis-33603	153	9	massivetext	massivetext	ADJ
bracis-33603	153	10	’s	’s	PART
bracis-33603	153	11	quality	quality	NOUN
bracis-33603	153	12	filters	filter	NOUN
bracis-33603	153	13	on	on	ADP
bracis-33603	153	14	downstream	downstream	ADJ
bracis-33603	153	15	task	task	NOUN
bracis-33603	153	16	performance	performance	NOUN
bracis-33603	153	17	,	,	PUNCT
bracis-33603	153	18	measured	measure	VERB
bracis-33603	153	19	in	in	ADP
bracis-33603	153	20	terms	term	NOUN
bracis-33603	153	21	of	of	ADP
bracis-33603	153	22	npm	npm	PROPN
bracis-33603	153	23	.	.	PUNCT
bracis-33603	154	1	pretraining	pretraine	VERB
bracis-33603	154	2	with	with	ADP
bracis-33603	154	3	the	the	DET
bracis-33603	154	4	filtered	filter	VERB
bracis-33603	154	5	dataset	dataset	NOUN
bracis-33603	154	6	shows	show	VERB
bracis-33603	154	7	an	an	DET
bracis-33603	154	8	upward	upward	ADJ
bracis-33603	154	9	trend	trend	NOUN
bracis-33603	154	10	in	in	ADP
bracis-33603	154	11	performance	performance	NOUN
bracis-33603	154	12	,	,	PUNCT
bracis-33603	154	13	which	which	PRON
bracis-33603	154	14	continues	continue	VERB
bracis-33603	154	15	without	without	ADP
bracis-33603	154	16	saturation	saturation	NOUN
bracis-33603	154	17	up	up	ADP
bracis-33603	154	18	to	to	ADP
bracis-33603	154	19	the	the	DET
bracis-33603	154	20	duration	duration	NOUN
bracis-33603	154	21	of	of	ADP
bracis-33603	154	22	one	one	NUM
bracis-33603	154	23	mc4	mc4	PROPN
bracis-33603	154	24	-	-	PUNCT
bracis-33603	154	25	pt	pt	ADJ
bracis-33603	154	26	epoch	epoch	NOUN
bracis-33603	154	27	,	,	PUNCT
bracis-33603	154	28	which	which	PRON
bracis-33603	154	29	is	be	AUX
bracis-33603	154	30	the	the	DET
bracis-33603	154	31	last	last	ADJ
bracis-33603	154	32	point	point	NOUN
bracis-33603	154	33	in	in	ADP
bracis-33603	154	34	the	the	DET
bracis-33603	154	35	plot	plot	NOUN
bracis-33603	154	36	.	.	PUNCT
bracis-33603	155	1	it	it	PRON
bracis-33603	155	2	’s	’	VERB
bracis-33603	155	3	important	important	ADJ
bracis-33603	155	4	to	to	PART
bracis-33603	155	5	note	note	VERB
bracis-33603	155	6	that	that	SCONJ
bracis-33603	155	7	one	one	NUM
bracis-33603	155	8	epoch	epoch	NOUN
bracis-33603	155	9	in	in	ADP
bracis-33603	155	10	the	the	DET
bracis-33603	155	11	filtered	filter	VERB
bracis-33603	155	12	dataset	dataset	NOUN
bracis-33603	155	13	has	have	VERB
bracis-33603	155	14	fewer	few	ADJ
bracis-33603	155	15	steps	step	NOUN
bracis-33603	155	16	than	than	ADP
bracis-33603	155	17	the	the	DET
bracis-33603	155	18	full	full	ADJ
bracis-33603	155	19	dataset	dataset	NOUN
bracis-33603	155	20	,	,	PUNCT
bracis-33603	155	21	so	so	SCONJ
bracis-33603	155	22	at	at	ADP
bracis-33603	155	23	one	one	NUM
bracis-33603	155	24	full	full	ADJ
bracis-33603	155	25	dataset	dataset	NOUN
bracis-33603	155	26	epoch	epoch	NOUN
bracis-33603	155	27	,	,	PUNCT
bracis-33603	155	28	data	datum	NOUN
bracis-33603	155	29	is	be	AUX
bracis-33603	155	30	repeated	repeat	VERB
bracis-33603	155	31	for	for	ADP
bracis-33603	155	32	the	the	DET
bracis-33603	155	33	filtered	filter	VERB
bracis-33603	155	34	dataset	dataset	NOUN
bracis-33603	155	35	.	.	PUNCT
bracis-33603	156	1	despite	despite	SCONJ
bracis-33603	156	2	the	the	DET
bracis-33603	156	3	upward	upward	ADJ
bracis-33603	156	4	trend	trend	NOUN
bracis-33603	156	5	favoring	favor	VERB
bracis-33603	156	6	the	the	DET
bracis-33603	156	7	pretraining	pretraine	VERB
bracis-33603	156	8	with	with	ADP
bracis-33603	156	9	the	the	DET
bracis-33603	156	10	filtered	filter	VERB
bracis-33603	156	11	dataset	dataset	NOUN
bracis-33603	156	12	,	,	PUNCT
bracis-33603	156	13	the	the	DET
bracis-33603	156	14	difference	difference	NOUN
bracis-33603	156	15	in	in	ADP
bracis-33603	156	16	terms	term	NOUN
bracis-33603	156	17	of	of	ADP
bracis-33603	156	18	npm	npm	PROPN
bracis-33603	156	19	is	be	AUX
bracis-33603	156	20	small	small	ADJ
bracis-33603	156	21	at	at	ADP
bracis-33603	156	22	the	the	DET
bracis-33603	156	23	one	one	NUM
bracis-33603	156	24	mc4	mc4	PROPN
bracis-33603	156	25	-	-	PUNCT
bracis-33603	156	26	pt	pt	PROPN
bracis-33603	156	27	epoch	epoch	PROPN
bracis-33603	156	28	mark	mark	NOUN
bracis-33603	156	29	.	.	PUNCT
bracis-33603	157	1	fig	fig	NOUN
bracis-33603	157	2	.	.	PUNCT
bracis-33603	158	1	2	2	X
bracis-33603	158	2	.	.	X
bracis-33603	158	3	effect	effect	NOUN
bracis-33603	158	4	of	of	ADP
bracis-33603	158	5	using	use	VERB
bracis-33603	158	6	quality	quality	NOUN
bracis-33603	158	7	filters	filter	NOUN
bracis-33603	158	8	on	on	ADP
bracis-33603	158	9	mc4	mc4	PROPN
bracis-33603	158	10	-	-	PUNCT
bracis-33603	158	11	pt	pt	PROPN
bracis-33603	158	12	.	.	PROPN
bracis-33603	158	13	vertical	vertical	PROPN
bracis-33603	158	14	lines	line	NOUN
bracis-33603	158	15	indicate	indicate	VERB
bracis-33603	158	16	the	the	DET
bracis-33603	158	17	number	number	NOUN
bracis-33603	158	18	of	of	ADP
bracis-33603	158	19	pretraining	pretraine	VERB
bracis-33603	158	20	tokens	token	NOUN
bracis-33603	158	21	for	for	ADP
bracis-33603	158	22	each	each	DET
bracis-33603	158	23	dataset	dataset	NOUN
bracis-33603	158	24	considered	consider	VERB
bracis-33603	158	25	.	.	PUNCT
bracis-33603	159	1	full	full	ADJ
bracis-33603	159	2	size	size	NOUN
bracis-33603	159	3	image	image	NOUN
bracis-33603	159	4	pretraining	pretraine	VERB
bracis-33603	159	5	optimization	optimization	NOUN
bracis-33603	159	6	strategy	strategy	NOUN
bracis-33603	159	7	:	:	PUNCT
bracis-33603	159	8	in	in	ADP
bracis-33603	159	9	our	our	PRON
bracis-33603	159	10	exploration	exploration	NOUN
bracis-33603	159	11	of	of	ADP
bracis-33603	159	12	optimization	optimization	NOUN
bracis-33603	159	13	strategies	strategy	NOUN
bracis-33603	159	14	,	,	PUNCT
bracis-33603	159	15	we	we	PRON
bracis-33603	159	16	initially	initially	ADV
bracis-33603	159	17	employed	employ	VERB
bracis-33603	159	18	adafactor	adafactor	NOUN
bracis-33603	159	19	with	with	ADP
bracis-33603	159	20	a	a	DET
bracis-33603	159	21	constant	constant	ADJ
bracis-33603	159	22	learning	learning	NOUN
bracis-33603	159	23	rate	rate	NOUN
bracis-33603	159	24	.	.	PUNCT
bracis-33603	160	1	this	this	DET
bracis-33603	160	2	ablation	ablation	NOUN
bracis-33603	160	3	extends	extend	VERB
bracis-33603	160	4	our	our	PRON
bracis-33603	160	5	investigation	investigation	NOUN
bracis-33603	160	6	to	to	ADP
bracis-33603	160	7	the	the	DET
bracis-33603	160	8	“	"	PUNCT
bracis-33603	160	9	inverse	inverse	ADJ
bracis-33603	160	10	square	square	NOUN
bracis-33603	160	11	root	root	NOUN
bracis-33603	160	12	”	"	PUNCT
bracis-33603	160	13	learning	learn	VERB
bracis-33603	160	14	rate	rate	NOUN
bracis-33603	160	15	schedule	schedule	NOUN
bracis-33603	160	16	,	,	PUNCT
bracis-33603	160	17	as	as	SCONJ
bracis-33603	160	18	utilized	utilize	VERB
bracis-33603	160	19	by	by	ADP
bracis-33603	160	20	[	[	X
bracis-33603	160	21	34	34	NUM
bracis-33603	160	22	]	]	PUNCT
bracis-33603	160	23	in	in	ADP
bracis-33603	160	24	their	their	PRON
bracis-33603	160	25	final	final	ADJ
bracis-33603	160	26	pretraining	pretraine	VERB
bracis-33603	160	27	experiments	experiment	NOUN
bracis-33603	160	28	.	.	PUNCT
bracis-33603	161	1	this	this	DET
bracis-33603	161	2	learning	learning	NOUN
bracis-33603	161	3	rate	rate	NOUN
bracis-33603	161	4	schedule	schedule	NOUN
bracis-33603	161	5	computes	compute	VERB
bracis-33603	161	6	the	the	DET
bracis-33603	161	7	rate	rate	NOUN
bracis-33603	161	8	as	as	ADP
bracis-33603	161	9	\(\frac{1}{\sqrt{\max	\(\frac{1}{\sqrt{\max	PROPN
bracis-33603	161	10	(	(	PUNCT
bracis-33603	161	11	n	n	CCONJ
bracis-33603	161	12	,	,	PUNCT
bracis-33603	161	13	k)}}\	k)}}\	NOUN
bracis-33603	161	14	)	)	PUNCT
bracis-33603	161	15	,	,	PUNCT
bracis-33603	161	16	where	where	SCONJ
bracis-33603	161	17	n	n	PRON
bracis-33603	161	18	represents	represent	VERB
bracis-33603	161	19	the	the	DET
bracis-33603	161	20	current	current	ADJ
bracis-33603	161	21	step	step	NOUN
bracis-33603	161	22	,	,	PUNCT
bracis-33603	161	23	and	and	CCONJ
bracis-33603	161	24	k	k	PROPN
bracis-33603	161	25	is	be	AUX
bracis-33603	161	26	the	the	DET
bracis-33603	161	27	number	number	NOUN
bracis-33603	161	28	of	of	ADP
bracis-33603	161	29	warm	warm	ADJ
bracis-33603	161	30	-	-	PUNCT
bracis-33603	161	31	up	up	ADP
bracis-33603	161	32	steps	step	NOUN
bracis-33603	161	33	.	.	PUNCT
bracis-33603	162	1	raffel	raffel	VERB
bracis-33603	162	2	et	et	NOUN
bracis-33603	162	3	al	al	PROPN
bracis-33603	162	4	.	.	PUNCT
bracis-33603	163	1	[	[	X
bracis-33603	163	2	34	34	NUM
bracis-33603	163	3	]	]	PUNCT
bracis-33603	163	4	used	use	VERB
bracis-33603	163	5	\(k=10,000\	\(k=10,000\	PROPN
bracis-33603	163	6	)	)	PUNCT
bracis-33603	163	7	,	,	PUNCT
bracis-33603	163	8	which	which	PRON
bracis-33603	163	9	sets	set	VERB
bracis-33603	163	10	the	the	DET
bracis-33603	163	11	learning	learning	NOUN
bracis-33603	163	12	rate	rate	NOUN
bracis-33603	163	13	of	of	ADP
bracis-33603	163	14	0.01	0.01	NUM
bracis-33603	163	15	for	for	ADP
bracis-33603	163	16	the	the	DET
bracis-33603	163	17	initial	initial	ADJ
bracis-33603	163	18	10k	10k	NOUN
bracis-33603	163	19	steps	step	NOUN
bracis-33603	163	20	,	,	PUNCT
bracis-33603	163	21	subsequently	subsequently	ADV
bracis-33603	163	22	decreasing	decrease	VERB
bracis-33603	163	23	exponentially	exponentially	ADV
bracis-33603	163	24	.	.	PUNCT
bracis-33603	164	1	the	the	DET
bracis-33603	164	2	learning	learning	NOUN
bracis-33603	164	3	rate	rate	NOUN
bracis-33603	164	4	at	at	ADP
bracis-33603	164	5	the	the	DET
bracis-33603	164	6	end	end	NOUN
bracis-33603	164	7	of	of	ADP
bracis-33603	164	8	the	the	DET
bracis-33603	164	9	pretraining	pretraining	NOUN
bracis-33603	164	10	,	,	PUNCT
bracis-33603	164	11	which	which	PRON
bracis-33603	164	12	consisted	consist	VERB
bracis-33603	164	13	of	of	ADP
bracis-33603	164	14	around	around	ADV
bracis-33603	164	15	1	1	NUM
bracis-33603	164	16	million	million	NUM
bracis-33603	164	17	steps	step	NOUN
bracis-33603	164	18	,	,	PUNCT
bracis-33603	164	19	was	be	AUX
bracis-33603	164	20	close	close	ADJ
bracis-33603	164	21	to	to	ADP
bracis-33603	164	22	0.001	0.001	NUM
bracis-33603	164	23	,	,	PUNCT
bracis-33603	164	24	the	the	DET
bracis-33603	164	25	same	same	ADJ
bracis-33603	164	26	one	one	NOUN
bracis-33603	164	27	used	use	VERB
bracis-33603	164	28	during	during	ADP
bracis-33603	164	29	the	the	DET
bracis-33603	164	30	finetuning	finetuning	NOUN
bracis-33603	164	31	.	.	PUNCT
bracis-33603	165	1	figure	figure	NOUN
bracis-33603	165	2	 	 	SPACE
bracis-33603	165	3	3	3	NUM
bracis-33603	165	4	illustrates	illustrate	VERB
bracis-33603	165	5	the	the	DET
bracis-33603	165	6	difference	difference	NOUN
bracis-33603	165	7	between	between	ADP
bracis-33603	165	8	these	these	DET
bracis-33603	165	9	optimization	optimization	NOUN
bracis-33603	165	10	strategies	strategy	NOUN
bracis-33603	165	11	.	.	PUNCT
bracis-33603	166	1	fig	fig	NOUN
bracis-33603	166	2	.	.	PUNCT
bracis-33603	167	1	3	3	X
bracis-33603	167	2	.	.	X
bracis-33603	167	3	learning	learn	VERB
bracis-33603	167	4	rate	rate	NOUN
bracis-33603	167	5	schedules	schedule	NOUN
bracis-33603	167	6	:	:	PUNCT
bracis-33603	167	7	constant	constant	ADJ
bracis-33603	167	8	vs.	vs.	ADP
bracis-33603	167	9	inverse	inverse	NOUN
bracis-33603	167	10	square	square	PROPN
bracis-33603	167	11	root	root	NOUN
bracis-33603	167	12	scheduler	scheduler	NOUN
bracis-33603	167	13	as	as	ADP
bracis-33603	167	14	a	a	DET
bracis-33603	167	15	function	function	NOUN
bracis-33603	167	16	of	of	ADP
bracis-33603	167	17	training	training	NOUN
bracis-33603	167	18	steps	step	NOUN
bracis-33603	167	19	.	.	PUNCT
bracis-33603	168	1	full	full	ADJ
bracis-33603	168	2	size	size	NOUN
bracis-33603	168	3	image	image	NOUN
bracis-33603	168	4	fig	fig	NOUN
bracis-33603	168	5	.	.	PUNCT
bracis-33603	169	1	4	4	X
bracis-33603	169	2	.	.	X
bracis-33603	169	3	effect	effect	NOUN
bracis-33603	169	4	of	of	ADP
bracis-33603	169	5	scheduler	scheduler	NOUN
bracis-33603	169	6	used	use	VERB
bracis-33603	169	7	during	during	ADP
bracis-33603	169	8	pretraining	pretraine	VERB
bracis-33603	169	9	.	.	PUNCT
bracis-33603	170	1	epochs	epoch	NOUN
bracis-33603	170	2	are	be	AUX
bracis-33603	170	3	relative	relative	ADJ
bracis-33603	170	4	to	to	ADP
bracis-33603	170	5	mc4	mc4	PROPN
bracis-33603	170	6	-	-	PUNCT
bracis-33603	170	7	pt	pt	PROPN
bracis-33603	170	8	.	.	NOUN
bracis-33603	170	9	full	full	ADJ
bracis-33603	170	10	size	size	NOUN
bracis-33603	170	11	image	image	NOUN
bracis-33603	170	12	attempting	attempt	VERB
bracis-33603	170	13	to	to	PART
bracis-33603	170	14	closely	closely	ADV
bracis-33603	170	15	mirror	mirror	VERB
bracis-33603	170	16	the	the	DET
bracis-33603	170	17	t5	t5	PROPN
bracis-33603	170	18	pretraining	pretraine	VERB
bracis-33603	170	19	recipe	recipe	NOUN
bracis-33603	170	20	,	,	PUNCT
bracis-33603	170	21	we	we	PRON
bracis-33603	170	22	applied	apply	VERB
bracis-33603	170	23	this	this	DET
bracis-33603	170	24	identical	identical	ADJ
bracis-33603	170	25	schedule	schedule	NOUN
bracis-33603	170	26	in	in	ADP
bracis-33603	170	27	our	our	PRON
bracis-33603	170	28	preliminary	preliminary	ADJ
bracis-33603	170	29	experiments	experiment	NOUN
bracis-33603	170	30	.	.	PUNCT
bracis-33603	171	1	however	however	ADV
bracis-33603	171	2	,	,	PUNCT
bracis-33603	171	3	we	we	PRON
bracis-33603	171	4	observed	observe	VERB
bracis-33603	171	5	a	a	DET
bracis-33603	171	6	rapid	rapid	ADJ
bracis-33603	171	7	overshoot	overshoot	NOUN
bracis-33603	171	8	in	in	ADP
bracis-33603	171	9	the	the	DET
bracis-33603	171	10	training	training	NOUN
bracis-33603	171	11	losses	loss	NOUN
bracis-33603	171	12	for	for	ADP
bracis-33603	171	13	t5	t5	PROPN
bracis-33603	171	14	-	-	PUNCT
bracis-33603	171	15	large	large	ADJ
bracis-33603	171	16	t5	t5	PROPN
bracis-33603	171	17	-	-	PUNCT
bracis-33603	171	18	3b	3b	NOUN
bracis-33603	171	19	within	within	ADP
bracis-33603	171	20	hours	hour	NOUN
bracis-33603	171	21	;	;	PUNCT
bracis-33603	171	22	adjusting	adjust	VERB
bracis-33603	171	23	n	n	CCONJ
bracis-33603	171	24	only	only	ADV
bracis-33603	171	25	delayed	delay	VERB
bracis-33603	171	26	the	the	DET
bracis-33603	171	27	overshoot	overshoot	NOUN
bracis-33603	171	28	without	without	ADP
bracis-33603	171	29	averting	avert	VERB
bracis-33603	171	30	it	it	PRON
bracis-33603	171	31	.	.	PUNCT
bracis-33603	172	1	transitioning	transition	VERB
bracis-33603	172	2	to	to	ADP
bracis-33603	172	3	a	a	DET
bracis-33603	172	4	constant	constant	ADJ
bracis-33603	172	5	learning	learning	NOUN
bracis-33603	172	6	rate	rate	NOUN
bracis-33603	172	7	of	of	ADP
bracis-33603	172	8	0.001	0.001	NUM
bracis-33603	172	9	solved	solve	VERB
bracis-33603	172	10	the	the	DET
bracis-33603	172	11	overshoot	overshoot	NOUN
bracis-33603	172	12	issue	issue	NOUN
bracis-33603	172	13	,	,	PUNCT
bracis-33603	172	14	leading	lead	VERB
bracis-33603	172	15	to	to	ADP
bracis-33603	172	16	stable	stable	ADJ
bracis-33603	172	17	pretraining	pretraine	VERB
bracis-33603	172	18	loss	loss	NOUN
bracis-33603	172	19	across	across	ADP
bracis-33603	172	20	all	all	DET
bracis-33603	172	21	model	model	NOUN
bracis-33603	172	22	sizes	size	NOUN
bracis-33603	172	23	and	and	CCONJ
bracis-33603	172	24	simplifying	simplify	VERB
bracis-33603	172	25	our	our	PRON
bracis-33603	172	26	experimental	experimental	ADJ
bracis-33603	172	27	setup	setup	NOUN
bracis-33603	172	28	.	.	PUNCT
bracis-33603	173	1	because	because	SCONJ
bracis-33603	173	2	overshoot	overshoot	NOUN
bracis-33603	173	3	was	be	AUX
bracis-33603	173	4	observed	observe	VERB
bracis-33603	173	5	only	only	ADV
bracis-33603	173	6	in	in	ADP
bracis-33603	173	7	the	the	DET
bracis-33603	173	8	larger	large	ADJ
bracis-33603	173	9	models	model	NOUN
bracis-33603	173	10	,	,	PUNCT
bracis-33603	173	11	and	and	CCONJ
bracis-33603	173	12	also	also	ADV
bracis-33603	173	13	given	give	VERB
bracis-33603	173	14	the	the	DET
bracis-33603	173	15	computational	computational	ADJ
bracis-33603	173	16	costs	cost	NOUN
bracis-33603	173	17	associated	associate	VERB
bracis-33603	173	18	with	with	ADP
bracis-33603	173	19	pretraining	pretraine	VERB
bracis-33603	173	20	the	the	DET
bracis-33603	173	21	larger	large	ADJ
bracis-33603	173	22	models	model	NOUN
bracis-33603	173	23	,	,	PUNCT
bracis-33603	173	24	we	we	PRON
bracis-33603	173	25	performed	perform	VERB
bracis-33603	173	26	additional	additional	ADJ
bracis-33603	173	27	pretraining	pretraine	VERB
bracis-33603	173	28	experiments	experiment	NOUN
bracis-33603	173	29	with	with	ADP
bracis-33603	173	30	“	"	PUNCT
bracis-33603	173	31	inverse	inverse	ADJ
bracis-33603	173	32	square	square	NOUN
bracis-33603	173	33	root	root	NOUN
bracis-33603	173	34	”	"	PUNCT
bracis-33603	173	35	scheduler	scheduler	NOUN
bracis-33603	173	36	for	for	ADP
bracis-33603	173	37	t5	t5	PROPN
bracis-33603	173	38	-	-	PUNCT
bracis-33603	173	39	small	small	ADJ
bracis-33603	173	40	and	and	CCONJ
bracis-33603	173	41	t5	t5	PROPN
bracis-33603	173	42	-	-	PUNCT
bracis-33603	173	43	base	base	NOUN
bracis-33603	173	44	models	model	NOUN
bracis-33603	173	45	only	only	ADV
bracis-33603	173	46	.	.	PUNCT
bracis-33603	174	1	number	number	NOUN
bracis-33603	174	2	of	of	ADP
bracis-33603	174	3	pretraining	pretraine	VERB
bracis-33603	174	4	epochs	epoch	NOUN
bracis-33603	174	5	:	:	PUNCT
bracis-33603	174	6	in	in	ADP
bracis-33603	174	7	the	the	DET
bracis-33603	174	8	primary	primary	ADJ
bracis-33603	174	9	set	set	NOUN
bracis-33603	174	10	of	of	ADP
bracis-33603	174	11	experiments	experiment	NOUN
bracis-33603	174	12	,	,	PUNCT
bracis-33603	174	13	the	the	DET
bracis-33603	174	14	mc4	mc4	PROPN
bracis-33603	174	15	-	-	PUNCT
bracis-33603	174	16	pt	pt	NOUN
bracis-33603	174	17	dataset	dataset	NOUN
bracis-33603	174	18	was	be	AUX
bracis-33603	174	19	fully	fully	ADV
bracis-33603	174	20	utilized	utilize	VERB
bracis-33603	174	21	for	for	ADP
bracis-33603	174	22	pretraining	pretraine	VERB
bracis-33603	174	23	over	over	ADP
bracis-33603	174	24	one	one	NUM
bracis-33603	174	25	epoch	epoch	NOUN
bracis-33603	174	26	.	.	PUNCT
bracis-33603	175	1	to	to	PART
bracis-33603	175	2	explore	explore	VERB
bracis-33603	175	3	the	the	DET
bracis-33603	175	4	influence	influence	NOUN
bracis-33603	175	5	of	of	ADP
bracis-33603	175	6	the	the	DET
bracis-33603	175	7	number	number	NOUN
bracis-33603	175	8	of	of	ADP
bracis-33603	175	9	pretraining	pretraine	VERB
bracis-33603	175	10	epochs	epoch	NOUN
bracis-33603	175	11	on	on	ADP
bracis-33603	175	12	downstream	downstream	ADJ
bracis-33603	175	13	task	task	NOUN
bracis-33603	175	14	performance	performance	NOUN
bracis-33603	175	15	,	,	PUNCT
bracis-33603	175	16	we	we	PRON
bracis-33603	175	17	conducted	conduct	VERB
bracis-33603	175	18	experiments	experiment	NOUN
bracis-33603	175	19	over	over	ADP
bracis-33603	175	20	various	various	ADJ
bracis-33603	175	21	epochs	epoch	NOUN
bracis-33603	175	22	,	,	PUNCT
bracis-33603	175	23	including	include	VERB
bracis-33603	175	24	partial	partial	ADJ
bracis-33603	175	25	epochs	epoch	NOUN
bracis-33603	175	26	(	(	PUNCT
bracis-33603	175	27	0.25	0.25	NUM
bracis-33603	175	28	,	,	PUNCT
bracis-33603	175	29	0.5	0.5	NUM
bracis-33603	175	30	,	,	PUNCT
bracis-33603	175	31	and	and	CCONJ
bracis-33603	175	32	0.75	0.75	NUM
bracis-33603	175	33	of	of	ADP
bracis-33603	175	34	an	an	DET
bracis-33603	175	35	epoch	epoch	NOUN
bracis-33603	175	36	)	)	PUNCT
bracis-33603	175	37	.	.	PUNCT
bracis-33603	176	1	considering	consider	VERB
bracis-33603	176	2	the	the	DET
bracis-33603	176	3	significant	significant	ADJ
bracis-33603	176	4	time	time	NOUN
bracis-33603	176	5	and	and	CCONJ
bracis-33603	176	6	computational	computational	ADJ
bracis-33603	176	7	resources	resource	NOUN
bracis-33603	176	8	required	require	VERB
bracis-33603	176	9	for	for	ADP
bracis-33603	176	10	extended	extended	ADJ
bracis-33603	176	11	pretraining	pretraining	NOUN
bracis-33603	176	12	,	,	PUNCT
bracis-33603	176	13	especially	especially	ADV
bracis-33603	176	14	with	with	ADP
bracis-33603	176	15	larger	large	ADJ
bracis-33603	176	16	models	model	NOUN
bracis-33603	176	17	,	,	PUNCT
bracis-33603	176	18	we	we	PRON
bracis-33603	176	19	limited	limit	VERB
bracis-33603	176	20	the	the	DET
bracis-33603	176	21	pretraining	pretraining	NOUN
bracis-33603	176	22	of	of	ADP
bracis-33603	176	23	the	the	DET
bracis-33603	176	24	t5	t5	PROPN
bracis-33603	176	25	-	-	PUNCT
bracis-33603	176	26	large	large	ADJ
bracis-33603	176	27	model	model	NOUN
bracis-33603	176	28	to	to	ADP
bracis-33603	176	29	two	two	NUM
bracis-33603	176	30	epochs	epoch	NOUN
bracis-33603	176	31	and	and	CCONJ
bracis-33603	176	32	the	the	DET
bracis-33603	176	33	t5	t5	PROPN
bracis-33603	176	34	-	-	PUNCT
bracis-33603	176	35	3b	3b	PROPN
bracis-33603	176	36	model	model	NOUN
bracis-33603	176	37	to	to	ADP
bracis-33603	176	38	a	a	DET
bracis-33603	176	39	single	single	ADJ
bracis-33603	176	40	epoch	epoch	NOUN
bracis-33603	176	41	.	.	PUNCT
bracis-33603	177	1	the	the	DET
bracis-33603	177	2	reduced	reduce	VERB
bracis-33603	177	3	epoch	epoch	NOUN
bracis-33603	177	4	duration	duration	NOUN
bracis-33603	177	5	for	for	ADP
bracis-33603	177	6	the	the	DET
bracis-33603	177	7	t5	t5	PROPN
bracis-33603	177	8	-	-	PUNCT
bracis-33603	177	9	small	small	ADJ
bracis-33603	177	10	and	and	CCONJ
bracis-33603	177	11	t5	t5	PROPN
bracis-33603	177	12	-	-	PUNCT
bracis-33603	177	13	base	base	NOUN
bracis-33603	177	14	models	model	NOUN
bracis-33603	177	15	enabled	enable	VERB
bracis-33603	177	16	more	more	ADV
bracis-33603	177	17	extensive	extensive	ADJ
bracis-33603	177	18	pretraining	pretraine	VERB
bracis-33603	177	19	periods	period	NOUN
bracis-33603	177	20	for	for	ADP
bracis-33603	177	21	these	these	DET
bracis-33603	177	22	configurations	configuration	NOUN
bracis-33603	177	23	.	.	PUNCT
bracis-33603	178	1	in	in	ADP
bracis-33603	178	2	fig	fig	NOUN
bracis-33603	178	3	.	.	PUNCT
bracis-33603	178	4	 	 	SPACE
bracis-33603	178	5	4	4	NUM
bracis-33603	178	6	,	,	PUNCT
bracis-33603	178	7	the	the	DET
bracis-33603	178	8	npm	npm	PROPN
bracis-33603	178	9	values	value	NOUN
bracis-33603	178	10	for	for	ADP
bracis-33603	178	11	t5	t5	PROPN
bracis-33603	178	12	-	-	PUNCT
bracis-33603	178	13	base	base	NOUN
bracis-33603	178	14	are	be	AUX
bracis-33603	178	15	shown	show	VERB
bracis-33603	178	16	with	with	ADP
bracis-33603	178	17	the	the	DET
bracis-33603	178	18	constant	constant	ADJ
bracis-33603	178	19	and	and	CCONJ
bracis-33603	178	20	the	the	DET
bracis-33603	178	21	inverse	inverse	NOUN
bracis-33603	178	22	square	square	PROPN
bracis-33603	178	23	root	root	NOUN
bracis-33603	178	24	scheduler	scheduler	NOUN
bracis-33603	178	25	,	,	PUNCT
bracis-33603	178	26	across	across	ADP
bracis-33603	178	27	a	a	DET
bracis-33603	178	28	varying	vary	VERB
bracis-33603	178	29	number	number	NOUN
bracis-33603	178	30	of	of	ADP
bracis-33603	178	31	pretraining	pretraine	VERB
bracis-33603	178	32	epochs	epoch	NOUN
bracis-33603	178	33	.	.	PUNCT
bracis-33603	179	1	it	it	PRON
bracis-33603	179	2	is	be	AUX
bracis-33603	179	3	observed	observe	VERB
bracis-33603	179	4	that	that	SCONJ
bracis-33603	179	5	there	there	PRON
bracis-33603	179	6	is	be	VERB
bracis-33603	179	7	a	a	DET
bracis-33603	179	8	difference	difference	NOUN
bracis-33603	179	9	between	between	ADP
bracis-33603	179	10	these	these	DET
bracis-33603	179	11	two	two	NUM
bracis-33603	179	12	optimization	optimization	NOUN
bracis-33603	179	13	strategies	strategy	NOUN
bracis-33603	179	14	:	:	PUNCT
bracis-33603	179	15	the	the	DET
bracis-33603	179	16	inverse	inverse	ADJ
bracis-33603	179	17	square	square	PROPN
bracis-33603	179	18	root	root	NOUN
bracis-33603	179	19	scheduler	scheduler	NOUN
bracis-33603	179	20	has	have	VERB
bracis-33603	179	21	the	the	DET
bracis-33603	179	22	advantage	advantage	NOUN
bracis-33603	179	23	for	for	ADP
bracis-33603	179	24	up	up	ADP
bracis-33603	179	25	to	to	PART
bracis-33603	179	26	two	two	NUM
bracis-33603	179	27	epochs	epoch	NOUN
bracis-33603	179	28	;	;	PUNCT
bracis-33603	179	29	afterwards	afterwards	ADV
bracis-33603	179	30	,	,	PUNCT
bracis-33603	179	31	the	the	DET
bracis-33603	179	32	constant	constant	ADJ
bracis-33603	179	33	learning	learning	NOUN
bracis-33603	179	34	rate	rate	NOUN
bracis-33603	179	35	takes	take	VERB
bracis-33603	179	36	the	the	DET
bracis-33603	179	37	upper	upper	ADJ
bracis-33603	179	38	hand	hand	NOUN
bracis-33603	179	39	,	,	PUNCT
bracis-33603	179	40	and	and	CCONJ
bracis-33603	179	41	by	by	ADP
bracis-33603	179	42	the	the	DET
bracis-33603	179	43	last	last	ADJ
bracis-33603	179	44	epoch	epoch	NOUN
bracis-33603	179	45	considered	consider	VERB
bracis-33603	179	46	,	,	PUNCT
bracis-33603	179	47	they	they	PRON
bracis-33603	179	48	reach	reach	VERB
bracis-33603	179	49	the	the	DET
bracis-33603	179	50	same	same	ADJ
bracis-33603	179	51	value	value	NOUN
bracis-33603	179	52	.	.	PUNCT
bracis-33603	180	1	an	an	DET
bracis-33603	180	2	increasing	increase	VERB
bracis-33603	180	3	trend	trend	NOUN
bracis-33603	180	4	in	in	ADP
bracis-33603	180	5	the	the	DET
bracis-33603	180	6	values	value	NOUN
bracis-33603	180	7	of	of	ADP
bracis-33603	180	8	npm	npm	PROPN
bracis-33603	180	9	is	be	AUX
bracis-33603	180	10	also	also	ADV
bracis-33603	180	11	noted	note	VERB
bracis-33603	180	12	for	for	ADP
bracis-33603	180	13	more	more	ADJ
bracis-33603	180	14	epochs	epoch	NOUN
bracis-33603	180	15	.	.	PUNCT
bracis-33603	181	1	5.2	5.2	NUM
bracis-33603	181	2	monoptt5	monoptt5	NOUN
bracis-33603	181	3	rerankers	reranker	NOUN
bracis-33603	181	4	the	the	DET
bracis-33603	181	5	information	information	NOUN
bracis-33603	181	6	retrieval	retrieval	NOUN
bracis-33603	181	7	tasks	task	NOUN
bracis-33603	181	8	reported	report	VERB
bracis-33603	181	9	in	in	ADP
bracis-33603	181	10	3	3	NUM
bracis-33603	181	11	represents	represent	VERB
bracis-33603	181	12	the	the	DET
bracis-33603	181	13	performance	performance	NOUN
bracis-33603	181	14	of	of	ADP
bracis-33603	181	15	our	our	PRON
bracis-33603	181	16	monoptt5	monoptt5	NOUN
bracis-33603	181	17	experiments	experiment	NOUN
bracis-33603	181	18	,	,	PUNCT
bracis-33603	181	19	developed	develop	VERB
bracis-33603	181	20	with	with	ADP
bracis-33603	181	21	the	the	DET
bracis-33603	181	22	methodology	methodology	NOUN
bracis-33603	181	23	describe	describe	VERB
bracis-33603	181	24	 	 	SPACE
bracis-33603	181	25	in	in	ADP
bracis-33603	181	26	sect	sect	NOUN
bracis-33603	181	27	.	.	PUNCT
bracis-33603	181	28	 	 	SPACE
bracis-33603	181	29	3.3	3.3	NUM
bracis-33603	181	30	;	;	PUNCT
bracis-33603	181	31	in	in	ADP
bracis-33603	181	32	this	this	DET
bracis-33603	181	33	section	section	NOUN
bracis-33603	181	34	we	we	PRON
bracis-33603	181	35	also	also	ADV
bracis-33603	181	36	report	report	VERB
bracis-33603	181	37	the	the	DET
bracis-33603	181	38	results	result	NOUN
bracis-33603	181	39	for	for	ADP
bracis-33603	181	40	other	other	ADJ
bracis-33603	181	41	approaches	approach	NOUN
bracis-33603	181	42	,	,	PUNCT
bracis-33603	181	43	including	include	VERB
bracis-33603	181	44	bm25	bm25	PROPN
bracis-33603	181	45	,	,	PUNCT
bracis-33603	181	46	and	and	CCONJ
bracis-33603	181	47	dense	dense	ADJ
bracis-33603	181	48	retrieval	retrieval	NOUN
bracis-33603	181	49	using	use	VERB
bracis-33603	181	50	multilingual	multilingual	NOUN
bracis-33603	181	51	-	-	PUNCT
bracis-33603	181	52	e5	e5	NOUN
bracis-33603	182	1	[	[	X
bracis-33603	182	2	47	47	NUM
bracis-33603	182	3	]	]	ADJ
bracis-33603	182	4	models	model	NOUN
bracis-33603	182	5	.	.	PUNCT
bracis-33603	183	1	the	the	DET
bracis-33603	183	2	dense	dense	ADJ
bracis-33603	183	3	models	model	NOUN
bracis-33603	183	4	are	be	AUX
bracis-33603	183	5	used	use	VERB
bracis-33603	183	6	as	as	ADP
bracis-33603	183	7	a	a	DET
bracis-33603	183	8	single	single	ADJ
bracis-33603	183	9	stage	stage	NOUN
bracis-33603	183	10	retrieval	retrieval	NOUN
bracis-33603	183	11	system	system	NOUN
bracis-33603	183	12	without	without	ADP
bracis-33603	183	13	reranking	reranking	NOUN
bracis-33603	183	14	;	;	PUNCT
bracis-33603	183	15	dense	dense	ADJ
bracis-33603	183	16	indexing	indexing	NOUN
bracis-33603	183	17	and	and	CCONJ
bracis-33603	183	18	retrieval	retrieval	NOUN
bracis-33603	183	19	is	be	AUX
bracis-33603	183	20	performed	perform	VERB
bracis-33603	183	21	with	with	ADP
bracis-33603	183	22	a100	a100	PROPN
bracis-33603	183	23	and	and	CCONJ
bracis-33603	183	24	v100	v100	PROPN
bracis-33603	183	25	gpus	gpu	NOUN
bracis-33603	183	26	on	on	ADP
bracis-33603	183	27	google	google	PROPN
bracis-33603	183	28	colab	colab	PROPN
bracis-33603	183	29	,	,	PUNCT
bracis-33603	183	30	leveraging	leverage	VERB
bracis-33603	183	31	the	the	DET
bracis-33603	183	32	pyserini	pyserini	NOUN
bracis-33603	183	33	framework	framework	NOUN
bracis-33603	183	34	.	.	PUNCT
bracis-33603	184	1	for	for	ADP
bracis-33603	184	2	mrobust	mrobust	ADV
bracis-33603	184	3	-	-	PUNCT
bracis-33603	184	4	pt	pt	NOUN
bracis-33603	184	5	,	,	PUNCT
bracis-33603	184	6	which	which	PRON
bracis-33603	184	7	contain	contain	VERB
bracis-33603	184	8	longer	long	ADJ
bracis-33603	184	9	documents	document	NOUN
bracis-33603	184	10	,	,	PUNCT
bracis-33603	184	11	we	we	PRON
bracis-33603	184	12	mitigate	mitigate	VERB
bracis-33603	184	13	document	document	NOUN
bracis-33603	184	14	truncation	truncation	NOUN
bracis-33603	184	15	by	by	ADP
bracis-33603	184	16	using	use	VERB
bracis-33603	184	17	the	the	DET
bracis-33603	184	18	same	same	ADJ
bracis-33603	184	19	splitting	splitting	NOUN
bracis-33603	184	20	strategy	strategy	NOUN
bracis-33603	184	21	described	describe	VERB
bracis-33603	184	22	in	in	ADP
bracis-33603	184	23	the	the	DET
bracis-33603	184	24	sect	sect	NOUN
bracis-33603	184	25	.	.	PUNCT
bracis-33603	184	26	 	 	SPACE
bracis-33603	184	27	3.3	3.3	NUM
bracis-33603	184	28	,	,	PUNCT
bracis-33603	184	29	using	use	VERB
bracis-33603	184	30	the	the	DET
bracis-33603	184	31	maximum	maximum	ADJ
bracis-33603	184	32	score	score	NOUN
bracis-33603	184	33	among	among	ADP
bracis-33603	184	34	the	the	DET
bracis-33603	184	35	document	document	NOUN
bracis-33603	184	36	segments	segment	NOUN
bracis-33603	184	37	as	as	ADP
bracis-33603	184	38	the	the	DET
bracis-33603	184	39	document	document	NOUN
bracis-33603	184	40	score	score	NOUN
bracis-33603	184	41	.	.	PUNCT
bracis-33603	185	1	figures	figure	NOUN
bracis-33603	185	2	 	 	SPACE
bracis-33603	185	3	5	5	NUM
bracis-33603	185	4	and	and	CCONJ
bracis-33603	185	5	6	6	NUM
bracis-33603	185	6	are	be	AUX
bracis-33603	185	7	used	use	VERB
bracis-33603	185	8	to	to	PART
bracis-33603	185	9	illustrate	illustrate	VERB
bracis-33603	185	10	the	the	DET
bracis-33603	185	11	discussion	discussion	NOUN
bracis-33603	185	12	presented	present	VERB
bracis-33603	185	13	in	in	ADP
bracis-33603	185	14	this	this	DET
bracis-33603	185	15	section	section	NOUN
bracis-33603	185	16	.	.	PUNCT
bracis-33603	186	1	results	result	NOUN
bracis-33603	186	2	in	in	ADP
bracis-33603	186	3	table	table	NOUN
bracis-33603	186	4	 	 	SPACE
bracis-33603	186	5	3	3	NUM
bracis-33603	186	6	only	only	ADV
bracis-33603	186	7	shows	show	VERB
bracis-33603	186	8	the	the	DET
bracis-33603	186	9	effectiveness	effectiveness	NOUN
bracis-33603	186	10	in	in	ADP
bracis-33603	186	11	each	each	DET
bracis-33603	186	12	retrieval	retrieval	NOUN
bracis-33603	186	13	task	task	NOUN
bracis-33603	186	14	of	of	ADP
bracis-33603	186	15	our	our	PRON
bracis-33603	186	16	monoptt5	monoptt5	NOUN
bracis-33603	186	17	models	model	NOUN
bracis-33603	186	18	and	and	CCONJ
bracis-33603	186	19	the	the	DET
bracis-33603	186	20	sota	sota	NOUN
bracis-33603	186	21	competitors	competitor	NOUN
bracis-33603	186	22	.	.	PUNCT
bracis-33603	187	1	for	for	ADP
bracis-33603	187	2	the	the	DET
bracis-33603	187	3	in	in	ADP
bracis-33603	187	4	-	-	PUNCT
bracis-33603	187	5	domain	domain	NOUN
bracis-33603	187	6	retrieval	retrieval	NOUN
bracis-33603	187	7	task	task	NOUN
bracis-33603	187	8	,	,	PUNCT
bracis-33603	187	9	the	the	DET
bracis-33603	187	10	mmarco	mmarco	ADJ
bracis-33603	187	11	-	-	PUNCT
bracis-33603	187	12	pt	pt	NOUN
bracis-33603	187	13	dataset	dataset	NOUN
bracis-33603	187	14	,	,	PUNCT
bracis-33603	187	15	we	we	PRON
bracis-33603	187	16	note	note	VERB
bracis-33603	187	17	that	that	SCONJ
bracis-33603	187	18	bm25	bm25	PROPN
bracis-33603	187	19	is	be	AUX
bracis-33603	187	20	easily	easily	ADV
bracis-33603	187	21	surpassed	surpass	VERB
bracis-33603	187	22	by	by	ADP
bracis-33603	187	23	all	all	DET
bracis-33603	187	24	alternatives	alternative	NOUN
bracis-33603	187	25	considered	consider	VERB
bracis-33603	187	26	,	,	PUNCT
bracis-33603	187	27	and	and	CCONJ
bracis-33603	187	28	the	the	DET
bracis-33603	187	29	effectiveness	effectiveness	NOUN
bracis-33603	187	30	figures	figure	NOUN
bracis-33603	187	31	for	for	ADP
bracis-33603	187	32	monoptt5	monoptt5	NOUN
bracis-33603	187	33	rerankers	reranker	NOUN
bracis-33603	187	34	and	and	CCONJ
bracis-33603	187	35	multilingual	multilingual	ADJ
bracis-33603	187	36	-	-	PUNCT
bracis-33603	187	37	e5	e5	NOUN
bracis-33603	187	38	models	model	NOUN
bracis-33603	187	39	are	be	AUX
bracis-33603	187	40	similar	similar	ADJ
bracis-33603	187	41	when	when	SCONJ
bracis-33603	187	42	consider	consider	VERB
bracis-33603	187	43	the	the	DET
bracis-33603	187	44	size	size	NOUN
bracis-33603	187	45	range	range	NOUN
bracis-33603	187	46	in	in	ADP
bracis-33603	187	47	common	common	ADJ
bracis-33603	187	48	,	,	PUNCT
bracis-33603	187	49	and	and	CCONJ
bracis-33603	187	50	monoptt5	monoptt5	NOUN
bracis-33603	187	51	rerankers	reranker	NOUN
bracis-33603	187	52	effectiveness	effectiveness	NOUN
bracis-33603	187	53	is	be	AUX
bracis-33603	187	54	above	above	ADP
bracis-33603	187	55	sota	sota	NOUN
bracis-33603	187	56	starting	start	VERB
bracis-33603	187	57	from	from	ADP
bracis-33603	187	58	models	model	NOUN
bracis-33603	187	59	of	of	ADP
bracis-33603	187	60	size	size	NOUN
bracis-33603	187	61	t5	t5	PROPN
bracis-33603	187	62	-	-	PUNCT
bracis-33603	187	63	base	base	NOUN
bracis-33603	187	64	.	.	PUNCT
bracis-33603	188	1	for	for	ADP
bracis-33603	188	2	mrobust	mrobust	ADV
bracis-33603	188	3	-	-	PUNCT
bracis-33603	188	4	pt	pt	NOUN
bracis-33603	188	5	,	,	PUNCT
bracis-33603	188	6	representing	represent	VERB
bracis-33603	188	7	a	a	DET
bracis-33603	188	8	zero	zero	NUM
bracis-33603	188	9	-	-	PUNCT
bracis-33603	188	10	shot	shot	NOUN
bracis-33603	188	11	setting	setting	NOUN
bracis-33603	188	12	,	,	PUNCT
bracis-33603	188	13	bm25	bm25	PROPN
bracis-33603	188	14	is	be	AUX
bracis-33603	188	15	only	only	ADV
bracis-33603	188	16	surpassed	surpass	VERB
bracis-33603	188	17	by	by	ADP
bracis-33603	188	18	the	the	DET
bracis-33603	188	19	mt5	mt5	ADJ
bracis-33603	188	20	reranker	reranker	NOUN
bracis-33603	188	21	from	from	ADP
bracis-33603	188	22	jeronymo	jeronymo	PROPN
bracis-33603	188	23	et	et	PROPN
bracis-33603	188	24	al	al	PROPN
bracis-33603	188	25	.	.	PUNCT
bracis-33603	189	1	[	[	X
bracis-33603	189	2	21	21	NUM
bracis-33603	189	3	]	]	PUNCT
bracis-33603	189	4	,	,	PUNCT
bracis-33603	189	5	and	and	CCONJ
bracis-33603	189	6	monoptt5	monoptt5	NOUN
bracis-33603	189	7	models	model	NOUN
bracis-33603	189	8	starting	start	VERB
bracis-33603	189	9	from	from	ADP
bracis-33603	189	10	t5	t5	PROPN
bracis-33603	189	11	-	-	PUNCT
bracis-33603	189	12	base	base	NOUN
bracis-33603	189	13	size	size	NOUN
bracis-33603	189	14	.	.	PUNCT
bracis-33603	190	1	fig	fig	NOUN
bracis-33603	190	2	.	.	PUNCT
bracis-33603	191	1	5	5	X
bracis-33603	191	2	.	.	X
bracis-33603	191	3	retrieval	retrieval	NOUN
bracis-33603	191	4	results	result	NOUN
bracis-33603	191	5	on	on	ADP
bracis-33603	191	6	mmarco	mmarco	NOUN
bracis-33603	191	7	-	-	PUNCT
bracis-33603	191	8	pt	pt	PROPN
bracis-33603	191	9	.	.	PROPN
bracis-33603	191	10	mcolbert	mcolbert	NOUN
bracis-33603	191	11	and	and	CCONJ
bracis-33603	191	12	mt5	mt5	ADJ
bracis-33603	191	13	values	value	NOUN
bracis-33603	191	14	are	be	AUX
bracis-33603	191	15	from	from	ADP
bracis-33603	191	16	bonifacio	bonifacio	PROPN
bracis-33603	191	17	et	et	PROPN
bracis-33603	191	18	al	al	PROPN
bracis-33603	191	19	.	.	PUNCT
bracis-33603	192	1	[	[	X
bracis-33603	192	2	3	3	NUM
bracis-33603	192	3	]	]	PUNCT
bracis-33603	192	4	.	.	PUNCT
bracis-33603	193	1	total	total	ADJ
bracis-33603	193	2	size	size	NOUN
bracis-33603	193	3	excludes	exclude	VERB
bracis-33603	193	4	embedding	embed	VERB
bracis-33603	193	5	parameters	parameter	NOUN
bracis-33603	193	6	.	.	PUNCT
bracis-33603	194	1	full	full	ADJ
bracis-33603	194	2	size	size	NOUN
bracis-33603	194	3	image	image	NOUN
bracis-33603	194	4	fig	fig	NOUN
bracis-33603	194	5	.	.	PUNCT
bracis-33603	195	1	6	6	NUM
bracis-33603	195	2	.	.	X
bracis-33603	195	3	retrieval	retrieval	NOUN
bracis-33603	195	4	results	result	NOUN
bracis-33603	195	5	on	on	ADP
bracis-33603	195	6	mrobust	mrobust	ADJ
bracis-33603	195	7	-	-	PUNCT
bracis-33603	195	8	pt	pt	PROPN
bracis-33603	195	9	.	.	PROPN
bracis-33603	195	10	mcolbert	mcolbert	NOUN
bracis-33603	195	11	and	and	CCONJ
bracis-33603	195	12	mt5	mt5	ADJ
bracis-33603	195	13	values	value	NOUN
bracis-33603	195	14	are	be	AUX
bracis-33603	195	15	from	from	ADP
bracis-33603	195	16	jeronymo	jeronymo	PROPN
bracis-33603	195	17	et	et	PROPN
bracis-33603	195	18	al	al	PROPN
bracis-33603	195	19	.	.	PUNCT
bracis-33603	196	1	[	[	X
bracis-33603	196	2	21	21	NUM
bracis-33603	196	3	]	]	PUNCT
bracis-33603	196	4	.	.	PUNCT
bracis-33603	197	1	total	total	ADJ
bracis-33603	197	2	size	size	NOUN
bracis-33603	197	3	excludes	exclude	VERB
bracis-33603	197	4	embedding	embed	VERB
bracis-33603	197	5	parameters	parameter	NOUN
bracis-33603	197	6	.	.	PUNCT
bracis-33603	198	1	full	full	ADJ
bracis-33603	198	2	size	size	NOUN
bracis-33603	198	3	image	image	NOUN
bracis-33603	198	4	6	6	NUM
bracis-33603	198	5	conclusion	conclusion	NOUN
bracis-33603	198	6	in	in	ADP
bracis-33603	198	7	this	this	DET
bracis-33603	198	8	study	study	NOUN
bracis-33603	198	9	,	,	PUNCT
bracis-33603	198	10	we	we	PRON
bracis-33603	198	11	introduced	introduce	VERB
bracis-33603	198	12	ptt5	ptt5	PROPN
bracis-33603	198	13	-	-	PUNCT
bracis-33603	198	14	v2	v2	PROPN
bracis-33603	198	15	,	,	PUNCT
bracis-33603	198	16	exploring	explore	VERB
bracis-33603	198	17	the	the	DET
bracis-33603	198	18	continued	continued	ADJ
bracis-33603	198	19	pretraining	pretraining	NOUN
bracis-33603	198	20	of	of	ADP
bracis-33603	198	21	t5	t5	PROPN
bracis-33603	198	22	models	model	NOUN
bracis-33603	198	23	for	for	ADP
bracis-33603	198	24	the	the	DET
bracis-33603	198	25	portuguese	portuguese	ADJ
bracis-33603	198	26	language	language	NOUN
bracis-33603	198	27	.	.	PUNCT
bracis-33603	199	1	we	we	PRON
bracis-33603	199	2	pretrained	pretraine	VERB
bracis-33603	199	3	t5	t5	PROPN
bracis-33603	199	4	models	model	NOUN
bracis-33603	199	5	using	use	VERB
bracis-33603	199	6	a	a	DET
bracis-33603	199	7	portuguese	portuguese	ADJ
bracis-33603	199	8	language	language	NOUN
bracis-33603	199	9	tokenizer	tokenizer	NOUN
bracis-33603	199	10	,	,	PUNCT
bracis-33603	199	11	over	over	ADP
bracis-33603	199	12	a	a	DET
bracis-33603	199	13	portuguese	portuguese	ADJ
bracis-33603	199	14	language	language	NOUN
bracis-33603	199	15	corpus	corpus	NOUN
bracis-33603	199	16	.	.	PUNCT
bracis-33603	200	1	the	the	DET
bracis-33603	200	2	finetuned	finetune	VERB
bracis-33603	200	3	models	model	NOUN
bracis-33603	200	4	achieved	achieve	VERB
bracis-33603	200	5	sota	sota	NOUN
bracis-33603	200	6	on	on	ADP
bracis-33603	200	7	assin2	assin2	PROPN
bracis-33603	200	8	rte	rte	PROPN
bracis-33603	200	9	and	and	CCONJ
bracis-33603	200	10	tweetsentbr	tweetsentbr	NOUN
bracis-33603	200	11	datasets	dataset	NOUN
bracis-33603	200	12	,	,	PUNCT
bracis-33603	200	13	two	two	NUM
bracis-33603	200	14	of	of	ADP
bracis-33603	200	15	the	the	DET
bracis-33603	200	16	three	three	NUM
bracis-33603	200	17	downstream	downstream	ADJ
bracis-33603	200	18	tasks	task	NOUN
bracis-33603	200	19	considered	consider	VERB
bracis-33603	200	20	.	.	PUNCT
bracis-33603	201	1	additionally	additionally	ADV
bracis-33603	201	2	,	,	PUNCT
bracis-33603	201	3	we	we	PRON
bracis-33603	201	4	applied	apply	VERB
bracis-33603	201	5	these	these	DET
bracis-33603	201	6	pretrained	pretraine	VERB
bracis-33603	201	7	checkpoints	checkpoint	NOUN
bracis-33603	201	8	to	to	PART
bracis-33603	201	9	develop	develop	VERB
bracis-33603	201	10	monot5	monot5	NOUN
bracis-33603	201	11	rerankers	reranker	NOUN
bracis-33603	201	12	customized	customize	VERB
bracis-33603	201	13	for	for	ADP
bracis-33603	201	14	the	the	DET
bracis-33603	201	15	portuguese	portuguese	ADJ
bracis-33603	201	16	language	language	NOUN
bracis-33603	201	17	,	,	PUNCT
bracis-33603	201	18	achieving	achieve	VERB
bracis-33603	201	19	top	top	ADJ
bracis-33603	201	20	performance	performance	NOUN
bracis-33603	201	21	on	on	ADP
bracis-33603	201	22	the	the	DET
bracis-33603	201	23	mmarco	mmarco	NOUN
bracis-33603	201	24	-	-	PUNCT
bracis-33603	201	25	pt	pt	PROPN
bracis-33603	201	26	and	and	CCONJ
bracis-33603	201	27	mrobust	mrobust	ADJ
bracis-33603	201	28	-	-	PUNCT
bracis-33603	201	29	pt	pt	NOUN
bracis-33603	201	30	datasets	dataset	NOUN
bracis-33603	201	31	.	.	PUNCT
bracis-33603	202	1	our	our	PRON
bracis-33603	202	2	main	main	ADJ
bracis-33603	202	3	results	result	NOUN
bracis-33603	202	4	supports	support	VERB
bracis-33603	202	5	the	the	DET
bracis-33603	202	6	evidence	evidence	NOUN
bracis-33603	202	7	of	of	ADP
bracis-33603	202	8	a	a	DET
bracis-33603	202	9	performance	performance	NOUN
bracis-33603	202	10	gap	gap	NOUN
bracis-33603	202	11	favoring	favor	VERB
bracis-33603	202	12	monolingual	monolingual	ADJ
bracis-33603	202	13	models	model	NOUN
bracis-33603	202	14	over	over	ADP
bracis-33603	202	15	english	english	PROPN
bracis-33603	202	16	-	-	PUNCT
bracis-33603	202	17	focused	focused	ADJ
bracis-33603	202	18	and	and	CCONJ
bracis-33603	202	19	multilingual	multilingual	ADJ
bracis-33603	202	20	models	model	NOUN
bracis-33603	202	21	,	,	PUNCT
bracis-33603	202	22	a	a	DET
bracis-33603	202	23	gap	gap	NOUN
bracis-33603	202	24	that	that	PRON
bracis-33603	202	25	narrows	narrow	VERB
bracis-33603	202	26	as	as	ADP
bracis-33603	202	27	model	model	NOUN
bracis-33603	202	28	capacity	capacity	NOUN
bracis-33603	202	29	increases	increase	NOUN
bracis-33603	202	30	.	.	PUNCT
bracis-33603	203	1	this	this	PRON
bracis-33603	203	2	underscores	underscore	VERB
bracis-33603	203	3	the	the	DET
bracis-33603	203	4	importance	importance	NOUN
bracis-33603	203	5	of	of	ADP
bracis-33603	203	6	language	language	NOUN
bracis-33603	203	7	-	-	PUNCT
bracis-33603	203	8	specific	specific	ADJ
bracis-33603	203	9	pretraining	pretraining	NOUN
bracis-33603	203	10	,	,	PUNCT
bracis-33603	203	11	and	and	CCONJ
bracis-33603	203	12	our	our	PRON
bracis-33603	203	13	analysis	analysis	NOUN
bracis-33603	203	14	of	of	ADP
bracis-33603	203	15	pretraining	pretraine	VERB
bracis-33603	203	16	settings	setting	NOUN
bracis-33603	203	17	suggests	suggest	VERB
bracis-33603	203	18	that	that	SCONJ
bracis-33603	203	19	while	while	SCONJ
bracis-33603	203	20	data	datum	NOUN
bracis-33603	203	21	filtering	filtering	NOUN
bracis-33603	203	22	,	,	PUNCT
bracis-33603	203	23	optimization	optimization	NOUN
bracis-33603	203	24	strategies	strategy	NOUN
bracis-33603	203	25	,	,	PUNCT
bracis-33603	203	26	and	and	CCONJ
bracis-33603	203	27	pretraining	pretraine	VERB
bracis-33603	203	28	duration	duration	NOUN
bracis-33603	203	29	can	can	AUX
bracis-33603	203	30	offer	offer	VERB
bracis-33603	203	31	incremental	incremental	ADJ
bracis-33603	203	32	improvements	improvement	NOUN
bracis-33603	203	33	,	,	PUNCT
bracis-33603	203	34	the	the	DET
bracis-33603	203	35	overall	overall	ADJ
bracis-33603	203	36	effects	effect	NOUN
bracis-33603	203	37	were	be	AUX
bracis-33603	203	38	limited	limit	VERB
bracis-33603	203	39	in	in	ADP
bracis-33603	203	40	comparison	comparison	NOUN
bracis-33603	203	41	to	to	ADP
bracis-33603	203	42	our	our	PRON
bracis-33603	203	43	baseline	baseline	NOUN
bracis-33603	203	44	settings	setting	NOUN
bracis-33603	203	45	and	and	CCONJ
bracis-33603	203	46	the	the	DET
bracis-33603	203	47	core	core	NOUN
bracis-33603	203	48	pretraining	pretraine	VERB
bracis-33603	203	49	recipe	recipe	NOUN
bracis-33603	203	50	remained	remain	VERB
bracis-33603	203	51	robust	robust	ADJ
bracis-33603	203	52	.	.	PUNCT
bracis-33603	204	1	notes	note	VERB
bracis-33603	204	2	1.all	1.all	ADJ
bracis-33603	204	3	applications	application	NOUN
bracis-33603	204	4	of	of	ADP
bracis-33603	204	5	bm25	bm25	PROPN
bracis-33603	204	6	in	in	ADP
bracis-33603	204	7	this	this	DET
bracis-33603	204	8	work	work	NOUN
bracis-33603	204	9	use	use	NOUN
bracis-33603	204	10	pyserini	pyserini	NOUN
bracis-33603	204	11	’s	’s	PART
bracis-33603	204	12	implementation	implementation	NOUN
bracis-33603	205	1	[	[	X
bracis-33603	205	2	26	26	NUM
bracis-33603	205	3	]	]	PUNCT
bracis-33603	205	4	with	with	ADP
bracis-33603	205	5	default	default	NOUN
bracis-33603	205	6	parameters	parameter	NOUN
bracis-33603	205	7	\(k_1=0.9\	\(k_1=0.9\	PROPN
bracis-33603	205	8	)	)	PUNCT
bracis-33603	205	9	and	and	CCONJ
bracis-33603	205	10	\(b=0.4\	\(b=0.4\	NUM
bracis-33603	205	11	)	)	PUNCT
bracis-33603	205	12	.	.	PUNCT
bracis-33603	206	1	references	reference	NOUN
bracis-33603	206	2	bajaj	bajaj	PROPN
bracis-33603	206	3	,	,	PUNCT
bracis-33603	206	4	p.	p.	PROPN
bracis-33603	206	5	,	,	PUNCT
bracis-33603	206	6	et	et	PROPN
bracis-33603	206	7	 	 	SPACE
bracis-33603	206	8	al	al	PROPN
bracis-33603	206	9	.	.	PROPN
bracis-33603	206	10	:	:	PUNCT
bracis-33603	207	1	ms	ms	PROPN
bracis-33603	207	2	marco	marco	PROPN
bracis-33603	207	3	:	:	PUNCT
bracis-33603	207	4	a	a	DET
bracis-33603	207	5	human	human	ADJ
bracis-33603	207	6	generated	generate	VERB
bracis-33603	207	7	machine	machine	NOUN
bracis-33603	207	8	reading	reading	NOUN
bracis-33603	207	9	comprehension	comprehension	NOUN
bracis-33603	207	10	dataset	dataset	NOUN
bracis-33603	207	11	(	(	PUNCT
bracis-33603	207	12	2018	2018	NUM
bracis-33603	207	13	)	)	PUNCT
bracis-33603	207	14	google	google	PROPN
bracis-33603	207	15	scholar	scholar	NOUN
bracis-33603	207	16	  	  	SPACE
bracis-33603	207	17	de	de	PROPN
bracis-33603	207	18	 	 	SPACE
bracis-33603	207	19	barros	barros	PROPN
bracis-33603	207	20	,	,	PUNCT
bracis-33603	207	21	t.m	t.m	PROPN
bracis-33603	207	22	.	.	PROPN
bracis-33603	207	23	,	,	PUNCT
bracis-33603	207	24	pedrini	pedrini	PROPN
bracis-33603	207	25	,	,	PUNCT
bracis-33603	207	26	h.	h.	PROPN
bracis-33603	207	27	,	,	PUNCT
bracis-33603	207	28	dias	dias	PROPN
bracis-33603	207	29	,	,	PUNCT
bracis-33603	207	30	z.	z.	PROPN
bracis-33603	207	31	:	:	PUNCT
bracis-33603	207	32	leveraging	leverage	VERB
bracis-33603	207	33	emoji	emoji	NOUN
bracis-33603	207	34	to	to	PART
bracis-33603	207	35	improve	improve	VERB
bracis-33603	207	36	sentiment	sentiment	NOUN
bracis-33603	207	37	classification	classification	NOUN
bracis-33603	207	38	of	of	ADP
bracis-33603	207	39	tweets	tweet	NOUN
bracis-33603	207	40	.	.	PUNCT
bracis-33603	208	1	in	in	ADP
bracis-33603	208	2	:	:	PUNCT
bracis-33603	208	3	proceedings	proceeding	NOUN
bracis-33603	208	4	of	of	ADP
bracis-33603	208	5	the	the	DET
bracis-33603	208	6	36th	36th	ADJ
bracis-33603	208	7	annual	annual	ADJ
bracis-33603	208	8	acm	acm	NOUN
bracis-33603	208	9	symposium	symposium	NOUN
bracis-33603	208	10	on	on	ADP
bracis-33603	208	11	applied	apply	VERB
bracis-33603	208	12	computing	computing	NOUN
bracis-33603	208	13	,	,	PUNCT
bracis-33603	208	14	sac	sac	NOUN
bracis-33603	208	15	2021	2021	NUM
bracis-33603	208	16	,	,	PUNCT
bracis-33603	208	17	pp	pp	ADJ
bracis-33603	208	18	.	.	PUNCT
bracis-33603	209	1	845–852	845–852	NUM
bracis-33603	209	2	.	.	PUNCT
bracis-33603	209	3	association	association	NOUN
bracis-33603	209	4	for	for	ADP
bracis-33603	209	5	computing	compute	VERB
bracis-33603	209	6	machinery	machinery	NOUN
bracis-33603	209	7	,	,	PUNCT
bracis-33603	209	8	new	new	PROPN
bracis-33603	209	9	york	york	PROPN
bracis-33603	209	10	,	,	PUNCT
bracis-33603	209	11	ny	ny	PROPN
bracis-33603	209	12	,	,	PUNCT
bracis-33603	209	13	usa	usa	PROPN
bracis-33603	209	14	(	(	PUNCT
bracis-33603	209	15	2021	2021	NUM
bracis-33603	209	16	)	)	PUNCT
bracis-33603	209	17	.	.	PUNCT
bracis-33603	210	1	https://doi.org/10.1145/3412841.3441960	https://doi.org/10.1145/3412841.3441960	PROPN
bracis-33603	210	2	bonifacio	bonifacio	PROPN
bracis-33603	210	3	,	,	PUNCT
bracis-33603	210	4	l.	l.	PROPN
bracis-33603	210	5	,	,	PUNCT
bracis-33603	210	6	et	et	PROPN
bracis-33603	210	7	 	 	SPACE
bracis-33603	210	8	al	al	PROPN
bracis-33603	210	9	.	.	PROPN
bracis-33603	210	10	:	:	PUNCT
bracis-33603	211	1	mmarco	mmarco	ADJ
bracis-33603	211	2	:	:	PUNCT
bracis-33603	211	3	a	a	DET
bracis-33603	211	4	multilingual	multilingual	ADJ
bracis-33603	211	5	version	version	NOUN
bracis-33603	211	6	of	of	ADP
bracis-33603	211	7	the	the	DET
bracis-33603	211	8	ms	ms	PROPN
bracis-33603	211	9	marco	marco	PROPN
bracis-33603	211	10	passage	passage	NOUN
bracis-33603	211	11	ranking	rank	VERB
bracis-33603	211	12	dataset	dataset	NOUN
bracis-33603	211	13	(	(	PUNCT
bracis-33603	211	14	2022	2022	NUM
bracis-33603	211	15	)	)	PUNCT
bracis-33603	211	16	google	google	NOUN
bracis-33603	211	17	scholar	scholar	NOUN
bracis-33603	211	18	  	  	SPACE
bracis-33603	211	19	brum	brum	NOUN
bracis-33603	211	20	,	,	PUNCT
bracis-33603	211	21	h.	h.	PROPN
bracis-33603	211	22	,	,	PUNCT
bracis-33603	211	23	volpe	volpe	PROPN
bracis-33603	211	24	 	 	SPACE
bracis-33603	211	25	nunes	nune	NOUN
bracis-33603	211	26	,	,	PUNCT
bracis-33603	211	27	m.d.g	m.d.g	NOUN
bracis-33603	211	28	.	.	PUNCT
bracis-33603	211	29	:	:	PUNCT
bracis-33603	212	1	building	build	VERB
bracis-33603	212	2	a	a	DET
bracis-33603	212	3	sentiment	sentiment	NOUN
bracis-33603	212	4	corpus	corpus	NOUN
bracis-33603	212	5	of	of	ADP
bracis-33603	212	6	tweets	tweet	NOUN
bracis-33603	212	7	in	in	ADP
bracis-33603	212	8	brazilian	brazilian	ADJ
bracis-33603	212	9	portuguese	portuguese	PROPN
bracis-33603	212	10	.	.	PUNCT
bracis-33603	213	1	in	in	ADP
bracis-33603	213	2	:	:	PUNCT
bracis-33603	213	3	calzolari	calzolari	PROPN
bracis-33603	213	4	,	,	PUNCT
bracis-33603	213	5	n.	n.	PROPN
bracis-33603	213	6	,	,	PUNCT
bracis-33603	213	7	et	et	PROPN
bracis-33603	213	8	al	al	PROPN
bracis-33603	213	9	.	.	PUNCT
bracis-33603	214	1	(	(	PUNCT
bracis-33603	214	2	eds	ed	NOUN
bracis-33603	214	3	.	.	PUNCT
bracis-33603	214	4	)	)	PUNCT
bracis-33603	214	5	proceedings	proceeding	NOUN
bracis-33603	214	6	of	of	ADP
bracis-33603	214	7	the	the	DET
bracis-33603	214	8	eleventh	eleventh	ADJ
bracis-33603	214	9	international	international	ADJ
bracis-33603	214	10	conference	conference	NOUN
bracis-33603	214	11	on	on	ADP
bracis-33603	214	12	language	language	NOUN
bracis-33603	214	13	resources	resource	NOUN
bracis-33603	214	14	and	and	CCONJ
bracis-33603	214	15	evaluation	evaluation	NOUN
bracis-33603	214	16	,	,	PUNCT
bracis-33603	214	17	lrec	lrec	PROPN
bracis-33603	214	18	2018	2018	NUM
bracis-33603	214	19	,	,	PUNCT
bracis-33603	214	20	may	may	PROPN
bracis-33603	214	21	2018	2018	NUM
bracis-33603	214	22	.	.	PUNCT
bracis-33603	215	1	european	european	ADJ
bracis-33603	215	2	language	language	PROPN
bracis-33603	215	3	resources	resource	NOUN
bracis-33603	215	4	association	association	NOUN
bracis-33603	215	5	(	(	PUNCT
bracis-33603	215	6	elra	elra	ADV
bracis-33603	215	7	)	)	PUNCT
bracis-33603	215	8	,	,	PUNCT
bracis-33603	215	9	miyazaki	miyazaki	PROPN
bracis-33603	215	10	,	,	PUNCT
bracis-33603	215	11	japan	japan	PROPN
bracis-33603	215	12	(	(	PUNCT
bracis-33603	215	13	2018	2018	NUM
bracis-33603	215	14	)	)	PUNCT
bracis-33603	215	15	google	google	PROPN
bracis-33603	215	16	scholar	scholar	NOUN
bracis-33603	215	17	  	  	SPACE
bracis-33603	215	18	campiotti	campiotti	ADJ
bracis-33603	215	19	,	,	PUNCT
bracis-33603	215	20	i.	i.	PROPN
bracis-33603	215	21	,	,	PUNCT
bracis-33603	215	22	rodrigues	rodrigues	PROPN
bracis-33603	215	23	,	,	PUNCT
bracis-33603	215	24	m.	m.	NOUN
bracis-33603	215	25	,	,	PUNCT
bracis-33603	215	26	albuquerque	albuquerque	NOUN
bracis-33603	215	27	,	,	PUNCT
bracis-33603	215	28	y.	y.	PROPN
bracis-33603	215	29	,	,	PUNCT
bracis-33603	215	30	azevedo	azevedo	PROPN
bracis-33603	215	31	,	,	PUNCT
bracis-33603	215	32	r.	r.	PROPN
bracis-33603	215	33	,	,	PUNCT
bracis-33603	215	34	andrade	andrade	PROPN
bracis-33603	215	35	,	,	PUNCT
bracis-33603	215	36	a.	a.	NOUN
bracis-33603	215	37	:	:	PUNCT
bracis-33603	215	38	debertinha	debertinha	NOUN
bracis-33603	215	39	:	:	PUNCT
bracis-33603	215	40	a	a	DET
bracis-33603	215	41	multistep	multistep	ADJ
bracis-33603	215	42	approach	approach	NOUN
bracis-33603	215	43	to	to	PART
bracis-33603	215	44	adapt	adapt	VERB
bracis-33603	215	45	debertav3	debertav3	NOUN
bracis-33603	215	46	xsmall	xsmall	NOUN
bracis-33603	215	47	for	for	ADP
bracis-33603	215	48	brazilian	brazilian	ADJ
bracis-33603	215	49	portuguese	portuguese	ADJ
bracis-33603	215	50	natural	natural	ADJ
bracis-33603	215	51	language	language	NOUN
bracis-33603	215	52	processing	processing	NOUN
bracis-33603	215	53	task	task	NOUN
bracis-33603	215	54	(	(	PUNCT
bracis-33603	215	55	2023	2023	NUM
bracis-33603	215	56	)	)	PUNCT
bracis-33603	215	57	google	google	PROPN
bracis-33603	215	58	scholar	scholar	NOUN
bracis-33603	215	59	  	  	SPACE
bracis-33603	215	60	carmo	carmo	PROPN
bracis-33603	215	61	,	,	PUNCT
bracis-33603	215	62	d.	d.	PROPN
bracis-33603	215	63	,	,	PUNCT
bracis-33603	215	64	piau	piau	NOUN
bracis-33603	215	65	,	,	PUNCT
bracis-33603	215	66	m.	m.	NOUN
bracis-33603	215	67	,	,	PUNCT
bracis-33603	215	68	campiotti	campiotti	PROPN
bracis-33603	215	69	,	,	PUNCT
bracis-33603	215	70	i.	i.	PROPN
bracis-33603	215	71	,	,	PUNCT
bracis-33603	215	72	nogueira	nogueira	PROPN
bracis-33603	215	73	,	,	PUNCT
bracis-33603	215	74	r.	r.	PROPN
bracis-33603	215	75	,	,	PUNCT
bracis-33603	215	76	lotufo	lotufo	PROPN
bracis-33603	215	77	,	,	PUNCT
bracis-33603	215	78	r.	r.	PROPN
bracis-33603	215	79	:	:	PUNCT
bracis-33603	215	80	ptt5	ptt5	PROPN
bracis-33603	215	81	:	:	PUNCT
bracis-33603	215	82	pretraining	pretraine	VERB
bracis-33603	215	83	and	and	CCONJ
bracis-33603	215	84	validating	validate	VERB
bracis-33603	215	85	the	the	DET
bracis-33603	215	86	t5	t5	PROPN
bracis-33603	215	87	model	model	NOUN
bracis-33603	215	88	on	on	ADP
bracis-33603	215	89	brazilian	brazilian	ADJ
bracis-33603	215	90	portuguese	portuguese	ADJ
bracis-33603	215	91	data	datum	NOUN
bracis-33603	215	92	(	(	PUNCT
bracis-33603	215	93	2020	2020	NUM
bracis-33603	215	94	)	)	PUNCT
bracis-33603	215	95	google	google	PROPN
bracis-33603	215	96	scholar	scholar	NOUN
bracis-33603	215	97	  	  	SPACE
bracis-33603	215	98	chrabrowa	chrabrowa	PROPN
bracis-33603	215	99	,	,	PUNCT
bracis-33603	215	100	a.	a.	PROPN
bracis-33603	215	101	,	,	PUNCT
bracis-33603	215	102	et	et	PROPN
bracis-33603	215	103	 	 	SPACE
bracis-33603	215	104	al	al	PROPN
bracis-33603	215	105	.	.	PROPN
bracis-33603	215	106	:	:	PUNCT
bracis-33603	216	1	evaluation	evaluation	NOUN
bracis-33603	216	2	of	of	ADP
bracis-33603	216	3	transfer	transfer	NOUN
bracis-33603	216	4	learning	learn	VERB
bracis-33603	216	5	for	for	ADP
bracis-33603	216	6	polish	polish	NOUN
bracis-33603	216	7	with	with	ADP
bracis-33603	216	8	a	a	DET
bracis-33603	216	9	text	text	NOUN
bracis-33603	216	10	-	-	PUNCT
bracis-33603	216	11	to	to	ADP
bracis-33603	216	12	-	-	PUNCT
bracis-33603	216	13	text	text	NOUN
bracis-33603	216	14	model	model	NOUN
bracis-33603	216	15	(	(	PUNCT
bracis-33603	216	16	2022	2022	NUM
bracis-33603	216	17	)	)	PUNCT
bracis-33603	216	18	google	google	PROPN
bracis-33603	216	19	scholar	scholar	NOUN
bracis-33603	216	20	  	  	SPACE
bracis-33603	216	21	chung	chung	PROPN
bracis-33603	216	22	,	,	PUNCT
bracis-33603	216	23	h.w	h.w	PROPN
bracis-33603	216	24	.	.	PROPN
bracis-33603	216	25	,	,	PUNCT
bracis-33603	216	26	et	et	PROPN
bracis-33603	216	27	 	 	SPACE
bracis-33603	216	28	al	al	PROPN
bracis-33603	216	29	.	.	PROPN
bracis-33603	216	30	:	:	PUNCT
bracis-33603	216	31	scaling	scale	VERB
bracis-33603	216	32	instruction	instruction	NOUN
bracis-33603	216	33	-	-	PUNCT
bracis-33603	216	34	finetuned	finetune	VERB
bracis-33603	216	35	language	language	NOUN
bracis-33603	216	36	models	model	NOUN
bracis-33603	216	37	(	(	PUNCT
bracis-33603	216	38	2022	2022	NUM
bracis-33603	216	39	)	)	PUNCT
bracis-33603	216	40	google	google	NOUN
bracis-33603	216	41	scholar	scholar	NOUN
bracis-33603	216	42	  	  	SPACE
bracis-33603	216	43	conneau	conneau	NOUN
bracis-33603	216	44	,	,	PUNCT
bracis-33603	216	45	a.	a.	NOUN
bracis-33603	216	46	,	,	PUNCT
bracis-33603	216	47	et	et	PROPN
bracis-33603	216	48	 	 	SPACE
bracis-33603	216	49	al	al	PROPN
bracis-33603	216	50	.	.	PROPN
bracis-33603	216	51	:	:	PUNCT
bracis-33603	217	1	unsupervised	unsupervised	ADJ
bracis-33603	217	2	cross	cross	ADJ
bracis-33603	217	3	-	-	ADJ
bracis-33603	217	4	lingual	lingual	ADJ
bracis-33603	217	5	representation	representation	NOUN
bracis-33603	217	6	learning	learn	VERB
bracis-33603	217	7	at	at	ADP
bracis-33603	217	8	scale	scale	NOUN
bracis-33603	217	9	(	(	PUNCT
bracis-33603	217	10	2020	2020	NUM
bracis-33603	217	11	)	)	PUNCT
bracis-33603	217	12	google	google	PROPN
bracis-33603	217	13	scholar	scholar	NOUN
bracis-33603	217	14	  	  	SPACE
bracis-33603	217	15	costa	costa	PROPN
bracis-33603	217	16	,	,	PUNCT
bracis-33603	217	17	p.b	p.b	PROPN
bracis-33603	217	18	.	.	PROPN
bracis-33603	217	19	,	,	PUNCT
bracis-33603	217	20	pavan	pavan	PROPN
bracis-33603	217	21	,	,	PUNCT
bracis-33603	217	22	m.c	m.c	PROPN
bracis-33603	217	23	.	.	PROPN
bracis-33603	217	24	,	,	PUNCT
bracis-33603	217	25	santos	santos	PROPN
bracis-33603	217	26	,	,	PUNCT
bracis-33603	217	27	w.r	w.r	PROPN
bracis-33603	217	28	.	.	PROPN
bracis-33603	217	29	,	,	PUNCT
bracis-33603	217	30	silva	silva	PROPN
bracis-33603	217	31	,	,	PUNCT
bracis-33603	217	32	s.c	s.c	PROPN
bracis-33603	217	33	.	.	PROPN
bracis-33603	217	34	,	,	PUNCT
bracis-33603	217	35	paraboni	paraboni	PROPN
bracis-33603	217	36	,	,	PUNCT
bracis-33603	217	37	i.	i.	PROPN
bracis-33603	217	38	:	:	PUNCT
bracis-33603	217	39	bertabaporu	bertabaporu	NOUN
bracis-33603	217	40	:	:	PUNCT
bracis-33603	217	41	assessing	assess	VERB
bracis-33603	217	42	a	a	DET
bracis-33603	217	43	genre	genre	NOUN
bracis-33603	217	44	-	-	PUNCT
bracis-33603	217	45	specific	specific	ADJ
bracis-33603	217	46	language	language	NOUN
bracis-33603	217	47	model	model	NOUN
bracis-33603	217	48	for	for	ADP
bracis-33603	217	49	portuguese	portuguese	ADJ
bracis-33603	217	50	nlp	nlp	NOUN
bracis-33603	217	51	.	.	PUNCT
bracis-33603	218	1	in	in	ADP
bracis-33603	218	2	:	:	PUNCT
bracis-33603	218	3	mitkov	mitkov	PROPN
bracis-33603	218	4	,	,	PUNCT
bracis-33603	218	5	r.	r.	PROPN
bracis-33603	218	6	,	,	PUNCT
bracis-33603	218	7	angelova	angelova	PROPN
bracis-33603	218	8	,	,	PUNCT
bracis-33603	218	9	g.	g.	PROPN
bracis-33603	218	10	(	(	PUNCT
bracis-33603	218	11	eds	eds	PROPN
bracis-33603	218	12	.	.	PUNCT
bracis-33603	218	13	)	)	PUNCT
bracis-33603	219	1	proceedings	proceeding	NOUN
bracis-33603	219	2	of	of	ADP
bracis-33603	219	3	the	the	DET
bracis-33603	219	4	14th	14th	ADJ
bracis-33603	219	5	international	international	ADJ
bracis-33603	219	6	conference	conference	NOUN
bracis-33603	219	7	on	on	ADP
bracis-33603	219	8	recent	recent	ADJ
bracis-33603	219	9	advances	advance	NOUN
bracis-33603	219	10	in	in	ADP
bracis-33603	219	11	natural	natural	ADJ
bracis-33603	219	12	language	language	NOUN
bracis-33603	219	13	processing	processing	NOUN
bracis-33603	219	14	,	,	PUNCT
bracis-33603	219	15	september	september	PROPN
bracis-33603	219	16	2023	2023	NUM
bracis-33603	219	17	,	,	PUNCT
bracis-33603	219	18	pp	pp	ADJ
bracis-33603	219	19	.	.	PUNCT
bracis-33603	220	1	217–223	217–223	NUM
bracis-33603	220	2	.	.	PUNCT
bracis-33603	220	3	incoma	incoma	PROPN
bracis-33603	220	4	ltd	ltd	PROPN
bracis-33603	220	5	.	.	PROPN
bracis-33603	220	6	,	,	PUNCT
bracis-33603	220	7	shoumen	shoumen	PROPN
bracis-33603	220	8	,	,	PUNCT
bracis-33603	220	9	bulgaria	bulgaria	PROPN
bracis-33603	220	10	,	,	PUNCT
bracis-33603	220	11	varna	varna	NOUN
bracis-33603	220	12	,	,	PUNCT
bracis-33603	220	13	bulgaria	bulgaria	PROPN
bracis-33603	220	14	(	(	PUNCT
bracis-33603	220	15	2023	2023	NUM
bracis-33603	220	16	)	)	PUNCT
bracis-33603	220	17	google	google	NOUN
bracis-33603	220	18	scholar	scholar	NOUN
bracis-33603	220	19	  	  	SPACE
bracis-33603	220	20	cui	cui	NOUN
bracis-33603	220	21	,	,	PUNCT
bracis-33603	220	22	y.	y.	PROPN
bracis-33603	220	23	,	,	PUNCT
bracis-33603	220	24	yang	yang	PROPN
bracis-33603	220	25	,	,	PUNCT
bracis-33603	220	26	z.	z.	PROPN
bracis-33603	220	27	,	,	PUNCT
bracis-33603	220	28	yao	yao	PROPN
bracis-33603	220	29	,	,	PUNCT
bracis-33603	220	30	x.	x.	NOUN
bracis-33603	220	31	:	:	PUNCT
bracis-33603	220	32	efficient	efficient	ADJ
bracis-33603	220	33	and	and	CCONJ
bracis-33603	220	34	effective	effective	ADJ
bracis-33603	220	35	text	text	NOUN
bracis-33603	220	36	encoding	encoding	NOUN
bracis-33603	220	37	for	for	ADP
bracis-33603	220	38	chinese	chinese	ADJ
bracis-33603	220	39	llama	llama	NOUN
bracis-33603	220	40	and	and	CCONJ
bracis-33603	220	41	alpaca	alpaca	NOUN
bracis-33603	220	42	(	(	PUNCT
bracis-33603	220	43	2024	2024	NUM
bracis-33603	220	44	)	)	PUNCT
bracis-33603	220	45	google	google	NOUN
bracis-33603	220	46	scholar	scholar	NOUN
bracis-33603	220	47	  	  	SPACE
bracis-33603	220	48	devlin	devlin	PROPN
bracis-33603	220	49	,	,	PUNCT
bracis-33603	220	50	j.	j.	PROPN
bracis-33603	220	51	,	,	PUNCT
bracis-33603	220	52	chang	chang	PROPN
bracis-33603	220	53	,	,	PUNCT
bracis-33603	220	54	m.w	m.w	PROPN
bracis-33603	220	55	.	.	PROPN
bracis-33603	220	56	,	,	PUNCT
bracis-33603	220	57	lee	lee	PROPN
bracis-33603	220	58	,	,	PUNCT
bracis-33603	220	59	k.	k.	PROPN
bracis-33603	220	60	,	,	PUNCT
bracis-33603	220	61	toutanova	toutanova	PROPN
bracis-33603	220	62	,	,	PUNCT
bracis-33603	220	63	k.	k.	PROPN
bracis-33603	220	64	:	:	PUNCT
bracis-33603	221	1	bert	bert	PROPN
bracis-33603	221	2	:	:	PUNCT
bracis-33603	221	3	pre	pre	ADJ
bracis-33603	221	4	-	-	NOUN
bracis-33603	221	5	training	training	NOUN
bracis-33603	221	6	of	of	ADP
bracis-33603	221	7	deep	deep	ADJ
bracis-33603	221	8	bidirectional	bidirectional	ADJ
bracis-33603	221	9	transformers	transformer	NOUN
bracis-33603	221	10	for	for	ADP
bracis-33603	221	11	language	language	NOUN
bracis-33603	221	12	understanding	understanding	NOUN
bracis-33603	221	13	(	(	PUNCT
bracis-33603	221	14	2019	2019	NUM
bracis-33603	221	15	)	)	PUNCT
bracis-33603	221	16	google	google	PROPN
bracis-33603	221	17	scholar	scholar	NOUN
bracis-33603	221	18	  	  	SPACE
bracis-33603	221	19	finardi	finardi	PROPN
bracis-33603	221	20	,	,	PUNCT
bracis-33603	221	21	p.	p.	NOUN
bracis-33603	221	22	,	,	PUNCT
bracis-33603	221	23	viegas	viegas	PROPN
bracis-33603	221	24	,	,	PUNCT
bracis-33603	221	25	j.d	j.d	PROPN
bracis-33603	221	26	.	.	PROPN
bracis-33603	221	27	,	,	PUNCT
bracis-33603	221	28	ferreira	ferreira	PROPN
bracis-33603	221	29	,	,	PUNCT
bracis-33603	221	30	g.t	g.t	PROPN
bracis-33603	221	31	.	.	PROPN
bracis-33603	221	32	,	,	PUNCT
bracis-33603	221	33	mansano	mansano	PROPN
bracis-33603	221	34	,	,	PUNCT
bracis-33603	221	35	a.f	a.f	PROPN
bracis-33603	221	36	.	.	PROPN
bracis-33603	221	37	,	,	PUNCT
bracis-33603	221	38	caridá	caridá	PROPN
bracis-33603	221	39	,	,	PUNCT
bracis-33603	221	40	v.f	v.f	PROPN
bracis-33603	221	41	.	.	PROPN
bracis-33603	221	42	:	:	PUNCT
bracis-33603	222	1	bertaú	bertaú	PROPN
bracis-33603	222	2	:	:	PUNCT
bracis-33603	222	3	itaú	itaú	PROPN
bracis-33603	222	4	bert	bert	PROPN
bracis-33603	222	5	for	for	ADP
bracis-33603	222	6	digital	digital	ADJ
bracis-33603	222	7	customer	customer	NOUN
bracis-33603	222	8	service	service	NOUN
bracis-33603	222	9	(	(	PUNCT
bracis-33603	222	10	2021	2021	NUM
bracis-33603	222	11	)	)	PUNCT
bracis-33603	222	12	google	google	NOUN
bracis-33603	222	13	scholar	scholar	NOUN
bracis-33603	222	14	  	  	SPACE
bracis-33603	222	15	gadre	gadre	NOUN
bracis-33603	222	16	,	,	PUNCT
bracis-33603	222	17	s.y	s.y	PROPN
bracis-33603	222	18	.	.	PROPN
bracis-33603	222	19	,	,	PUNCT
bracis-33603	222	20	et	et	PROPN
bracis-33603	222	21	 	 	SPACE
bracis-33603	222	22	al	al	PROPN
bracis-33603	222	23	.	.	PROPN
bracis-33603	222	24	:	:	PUNCT
bracis-33603	223	1	language	language	NOUN
bracis-33603	223	2	models	model	NOUN
bracis-33603	223	3	scale	scale	VERB
bracis-33603	223	4	reliably	reliably	ADV
bracis-33603	223	5	with	with	ADP
bracis-33603	223	6	over	over	ADP
bracis-33603	223	7	-	-	PUNCT
bracis-33603	223	8	training	training	NOUN
bracis-33603	223	9	and	and	CCONJ
bracis-33603	223	10	on	on	ADP
bracis-33603	223	11	downstream	downstream	ADJ
bracis-33603	223	12	tasks	task	NOUN
bracis-33603	223	13	(	(	PUNCT
bracis-33603	223	14	2024	2024	NUM
bracis-33603	223	15	)	)	PUNCT
bracis-33603	223	16	google	google	PROPN
bracis-33603	223	17	scholar	scholar	NOUN
bracis-33603	223	18	  	  	SPACE
bracis-33603	223	19	garcia	garcia	PROPN
bracis-33603	223	20	,	,	PUNCT
bracis-33603	223	21	g.l	g.l	PROPN
bracis-33603	223	22	.	.	PROPN
bracis-33603	223	23	,	,	PUNCT
bracis-33603	223	24	et	et	PROPN
bracis-33603	223	25	 	 	SPACE
bracis-33603	223	26	al	al	PROPN
bracis-33603	223	27	.	.	PROPN
bracis-33603	223	28	:	:	PUNCT
bracis-33603	224	1	introducing	introduce	VERB
bracis-33603	224	2	bode	bode	NOUN
bracis-33603	224	3	:	:	PUNCT
bracis-33603	224	4	a	a	DET
bracis-33603	224	5	fine	fine	ADV
bracis-33603	224	6	-	-	PUNCT
bracis-33603	224	7	tuned	tune	VERB
bracis-33603	224	8	large	large	ADJ
bracis-33603	224	9	language	language	NOUN
bracis-33603	224	10	model	model	NOUN
bracis-33603	224	11	for	for	ADP
bracis-33603	224	12	portuguese	portuguese	ADJ
bracis-33603	224	13	prompt	prompt	NOUN
bracis-33603	224	14	-	-	PUNCT
bracis-33603	224	15	based	base	VERB
bracis-33603	224	16	task	task	NOUN
bracis-33603	224	17	(	(	PUNCT
bracis-33603	224	18	2024	2024	NUM
bracis-33603	224	19	)	)	PUNCT
bracis-33603	224	20	google	google	NOUN
bracis-33603	224	21	scholar	scholar	NOUN
bracis-33603	224	22	  	  	SPACE
bracis-33603	224	23	gomes	gome	NOUN
bracis-33603	224	24	,	,	PUNCT
bracis-33603	224	25	j.r.s	j.r.s	PROPN
bracis-33603	224	26	.	.	PUNCT
bracis-33603	224	27	,	,	PUNCT
bracis-33603	224	28	et	et	PROPN
bracis-33603	224	29	 	 	SPACE
bracis-33603	224	30	al	al	PROPN
bracis-33603	224	31	.	.	PROPN
bracis-33603	224	32	:	:	PUNCT
bracis-33603	225	1	deep	deep	ADJ
bracis-33603	225	2	learning	learn	VERB
bracis-33603	225	3	brasil	brasil	PROPN
bracis-33603	225	4	at	at	ADP
bracis-33603	225	5	absapt	absapt	NOUN
bracis-33603	225	6	2022	2022	NUM
bracis-33603	225	7	:	:	PUNCT
bracis-33603	225	8	portuguese	portuguese	ADJ
bracis-33603	225	9	transformer	transformer	NOUN
bracis-33603	225	10	ensemble	ensemble	ADJ
bracis-33603	225	11	approaches	approach	NOUN
bracis-33603	225	12	(	(	PUNCT
bracis-33603	225	13	2023	2023	NUM
bracis-33603	225	14	)	)	PUNCT
bracis-33603	225	15	google	google	PROPN
bracis-33603	225	16	scholar	scholar	NOUN
bracis-33603	225	17	  	  	SPACE
bracis-33603	225	18	hermann	hermann	PROPN
bracis-33603	225	19	,	,	PUNCT
bracis-33603	225	20	k.m	k.m	PROPN
bracis-33603	225	21	.	.	PROPN
bracis-33603	225	22	,	,	PUNCT
bracis-33603	225	23	et	et	PROPN
bracis-33603	225	24	 	 	SPACE
bracis-33603	225	25	al	al	PROPN
bracis-33603	225	26	.	.	PROPN
bracis-33603	225	27	:	:	PUNCT
bracis-33603	226	1	teaching	teach	VERB
bracis-33603	226	2	machines	machine	NOUN
bracis-33603	226	3	to	to	PART
bracis-33603	226	4	read	read	VERB
bracis-33603	226	5	and	and	CCONJ
bracis-33603	226	6	comprehend	comprehend	VERB
bracis-33603	226	7	(	(	PUNCT
bracis-33603	226	8	2015	2015	NUM
bracis-33603	226	9	)	)	PUNCT
bracis-33603	226	10	google	google	PROPN
bracis-33603	226	11	scholar	scholar	NOUN
bracis-33603	226	12	  	  	SPACE
bracis-33603	226	13	hoffmann	hoffmann	PROPN
bracis-33603	226	14	,	,	PUNCT
bracis-33603	226	15	j.	j.	PROPN
bracis-33603	226	16	,	,	PUNCT
bracis-33603	226	17	et	et	PROPN
bracis-33603	226	18	 	 	SPACE
bracis-33603	226	19	al	al	PROPN
bracis-33603	226	20	.	.	PROPN
bracis-33603	226	21	:	:	PUNCT
bracis-33603	227	1	training	train	VERB
bracis-33603	227	2	compute	compute	NOUN
bracis-33603	227	3	-	-	PUNCT
bracis-33603	227	4	optimal	optimal	ADJ
bracis-33603	227	5	large	large	ADJ
bracis-33603	227	6	language	language	NOUN
bracis-33603	227	7	models	model	NOUN
bracis-33603	227	8	(	(	PUNCT
bracis-33603	227	9	2022	2022	NUM
bracis-33603	227	10	)	)	PUNCT
bracis-33603	227	11	google	google	NOUN
bracis-33603	227	12	scholar	scholar	NOUN
bracis-33603	227	13	  	  	SPACE
bracis-33603	227	14	honnibal	honnibal	ADJ
bracis-33603	227	15	,	,	PUNCT
bracis-33603	227	16	m.	m.	NOUN
bracis-33603	227	17	,	,	PUNCT
bracis-33603	227	18	montani	montani	NOUN
bracis-33603	227	19	,	,	PUNCT
bracis-33603	227	20	i.	i.	PROPN
bracis-33603	227	21	,	,	PUNCT
bracis-33603	227	22	landeghem	landeghem	PROPN
bracis-33603	227	23	,	,	PUNCT
bracis-33603	227	24	s.v	s.v	PROPN
bracis-33603	227	25	.	.	PROPN
bracis-33603	227	26	,	,	PUNCT
bracis-33603	227	27	boyd	boyd	PROPN
bracis-33603	227	28	,	,	PUNCT
bracis-33603	227	29	a.	a.	NOUN
bracis-33603	227	30	:	:	PUNCT
bracis-33603	227	31	spacy	spacy	NOUN
bracis-33603	227	32	:	:	PUNCT
bracis-33603	227	33	industrial	industrial	ADJ
bracis-33603	227	34	-	-	PUNCT
bracis-33603	227	35	strength	strength	NOUN
bracis-33603	227	36	natural	natural	ADJ
bracis-33603	227	37	language	language	NOUN
bracis-33603	227	38	processing	processing	NOUN
bracis-33603	227	39	in	in	ADP
bracis-33603	227	40	python	python	NOUN
bracis-33603	227	41	(	(	PUNCT
bracis-33603	227	42	2020	2020	NUM
bracis-33603	227	43	,	,	PUNCT
bracis-33603	227	44	to	to	PART
bracis-33603	227	45	appear	appear	VERB
bracis-33603	227	46	)	)	PUNCT
bracis-33603	227	47	.	.	PUNCT
bracis-33603	228	1	https://doi.org/10.5281/zenodo.1212303	https://doi.org/10.5281/zenodo.1212303	PROPN
bracis-33603	228	2	hu	hu	PROPN
bracis-33603	228	3	,	,	PUNCT
bracis-33603	228	4	j.	j.	PROPN
bracis-33603	228	5	,	,	PUNCT
bracis-33603	228	6	ruder	ruder	PROPN
bracis-33603	228	7	,	,	PUNCT
bracis-33603	228	8	s.	s.	PROPN
bracis-33603	228	9	,	,	PUNCT
bracis-33603	228	10	siddhant	siddhant	NOUN
bracis-33603	228	11	,	,	PUNCT
bracis-33603	228	12	a.	a.	NOUN
bracis-33603	228	13	,	,	PUNCT
bracis-33603	228	14	neubig	neubig	ADJ
bracis-33603	228	15	,	,	PUNCT
bracis-33603	228	16	g.	g.	PROPN
bracis-33603	228	17	,	,	PUNCT
bracis-33603	228	18	firat	firat	PROPN
bracis-33603	228	19	,	,	PUNCT
bracis-33603	228	20	o.	o.	PROPN
bracis-33603	228	21	,	,	PUNCT
bracis-33603	228	22	johnson	johnson	PROPN
bracis-33603	228	23	,	,	PUNCT
bracis-33603	228	24	m.	m.	NOUN
bracis-33603	228	25	:	:	PUNCT
bracis-33603	228	26	xtreme	xtreme	PROPN
bracis-33603	228	27	:	:	PUNCT
bracis-33603	228	28	a	a	DET
bracis-33603	228	29	massively	massively	ADV
bracis-33603	228	30	multilingual	multilingual	ADJ
bracis-33603	228	31	multi	multi	ADJ
bracis-33603	228	32	-	-	ADJ
bracis-33603	228	33	task	task	ADJ
bracis-33603	228	34	benchmark	benchmark	NOUN
bracis-33603	228	35	for	for	ADP
bracis-33603	228	36	evaluating	evaluate	VERB
bracis-33603	228	37	cross	cross	ADJ
bracis-33603	228	38	-	-	ADJ
bracis-33603	228	39	lingual	lingual	ADJ
bracis-33603	228	40	generalization	generalization	NOUN
bracis-33603	228	41	(	(	PUNCT
bracis-33603	228	42	2020	2020	NUM
bracis-33603	228	43	)	)	PUNCT
bracis-33603	228	44	google	google	PROPN
bracis-33603	228	45	scholar	scholar	NOUN
bracis-33603	228	46	  	  	SPACE
bracis-33603	228	47	jeronymo	jeronymo	NOUN
bracis-33603	228	48	,	,	PUNCT
bracis-33603	228	49	v.	v.	PROPN
bracis-33603	228	50	,	,	PUNCT
bracis-33603	228	51	nascimento	nascimento	PROPN
bracis-33603	228	52	,	,	PUNCT
bracis-33603	228	53	m.	m.	NOUN
bracis-33603	228	54	,	,	PUNCT
bracis-33603	228	55	lotufo	lotufo	PROPN
bracis-33603	228	56	,	,	PUNCT
bracis-33603	228	57	r.	r.	PROPN
bracis-33603	228	58	,	,	PUNCT
bracis-33603	228	59	nogueira	nogueira	PROPN
bracis-33603	228	60	,	,	PUNCT
bracis-33603	228	61	r.	r.	PROPN
bracis-33603	228	62	:	:	PUNCT
bracis-33603	228	63	mrobust04	mrobust04	PROPN
bracis-33603	228	64	:	:	PUNCT
bracis-33603	228	65	a	a	DET
bracis-33603	228	66	multilingual	multilingual	ADJ
bracis-33603	228	67	version	version	NOUN
bracis-33603	228	68	of	of	ADP
bracis-33603	228	69	the	the	DET
bracis-33603	228	70	trec	trec	PROPN
bracis-33603	228	71	robust	robust	ADJ
bracis-33603	228	72	2004	2004	NUM
bracis-33603	228	73	benchmark	benchmark	NOUN
bracis-33603	228	74	(	(	PUNCT
bracis-33603	228	75	2022	2022	NUM
bracis-33603	228	76	)	)	PUNCT
bracis-33603	228	77	google	google	PROPN
bracis-33603	228	78	scholar	scholar	NOUN
bracis-33603	228	79	  	  	SPACE
bracis-33603	228	80	jude	jude	NOUN
bracis-33603	228	81	 	 	SPACE
bracis-33603	228	82	ogundepo	ogundepo	NOUN
bracis-33603	228	83	,	,	PUNCT
bracis-33603	228	84	o.	o.	PROPN
bracis-33603	228	85	,	,	PUNCT
bracis-33603	228	86	oladipo	oladipo	NOUN
bracis-33603	228	87	,	,	PUNCT
bracis-33603	228	88	a.	a.	NOUN
bracis-33603	228	89	,	,	PUNCT
bracis-33603	228	90	adeyemi	adeyemi	ADV
bracis-33603	228	91	,	,	PUNCT
bracis-33603	228	92	m.	m.	NOUN
bracis-33603	228	93	,	,	PUNCT
bracis-33603	228	94	ogueji	ogueji	PROPN
bracis-33603	228	95	,	,	PUNCT
bracis-33603	228	96	k.	k.	PROPN
bracis-33603	228	97	,	,	PUNCT
bracis-33603	228	98	lin	lin	PROPN
bracis-33603	228	99	,	,	PUNCT
bracis-33603	228	100	j.	j.	PROPN
bracis-33603	228	101	:	:	PUNCT
bracis-33603	228	102	afriteva	afriteva	PROPN
bracis-33603	228	103	:	:	PUNCT
bracis-33603	228	104	extending	extend	VERB
bracis-33603	228	105	“	"	PUNCT
bracis-33603	228	106	small	small	ADJ
bracis-33603	228	107	data	datum	NOUN
bracis-33603	228	108	”	"	PUNCT
bracis-33603	228	109	pretraining	pretraine	VERB
bracis-33603	228	110	approaches	approach	NOUN
bracis-33603	228	111	to	to	ADP
bracis-33603	228	112	sequence	sequence	NOUN
bracis-33603	228	113	-	-	PUNCT
bracis-33603	228	114	to	to	ADP
bracis-33603	228	115	-	-	PUNCT
bracis-33603	228	116	sequence	sequence	NOUN
bracis-33603	228	117	models	model	NOUN
bracis-33603	228	118	.	.	PUNCT
bracis-33603	229	1	in	in	ADP
bracis-33603	229	2	:	:	PUNCT
bracis-33603	229	3	cherry	cherry	NOUN
bracis-33603	229	4	,	,	PUNCT
bracis-33603	229	5	c.	c.	PROPN
bracis-33603	229	6	,	,	PUNCT
bracis-33603	229	7	(	(	PUNCT
bracis-33603	229	8	eds	ed	NOUN
bracis-33603	229	9	.	.	PUNCT
bracis-33603	229	10	)	)	PUNCT
bracis-33603	229	11	proceedings	proceeding	NOUN
bracis-33603	229	12	of	of	ADP
bracis-33603	229	13	the	the	DET
bracis-33603	229	14	third	third	ADJ
bracis-33603	229	15	workshop	workshop	NOUN
bracis-33603	229	16	on	on	ADP
bracis-33603	229	17	deep	deep	ADJ
bracis-33603	229	18	learning	learning	NOUN
bracis-33603	229	19	for	for	ADP
bracis-33603	229	20	low	low	ADJ
bracis-33603	229	21	-	-	PUNCT
bracis-33603	229	22	resource	resource	NOUN
bracis-33603	229	23	natural	natural	ADJ
bracis-33603	229	24	language	language	NOUN
bracis-33603	229	25	processing	processing	NOUN
bracis-33603	229	26	,	,	PUNCT
bracis-33603	229	27	july	july	PROPN
bracis-33603	229	28	2022	2022	NUM
bracis-33603	229	29	,	,	PUNCT
bracis-33603	229	30	pp	pp	ADJ
bracis-33603	229	31	.	.	PUNCT
bracis-33603	230	1	126–135	126–135	NUM
bracis-33603	230	2	.	.	X
bracis-33603	230	3	association	association	NOUN
bracis-33603	230	4	for	for	ADP
bracis-33603	230	5	computational	computational	ADJ
bracis-33603	230	6	linguistics	linguistic	NOUN
bracis-33603	230	7	,	,	PUNCT
bracis-33603	230	8	hybrid	hybrid	NOUN
bracis-33603	230	9	(	(	PUNCT
bracis-33603	230	10	2022	2022	NUM
bracis-33603	230	11	)	)	PUNCT
bracis-33603	230	12	.	.	PUNCT
bracis-33603	231	1	https://doi.org/10.18653/v1/2022.deeplo-1.14	https://doi.org/10.18653/v1/2022.deeplo-1.14	PROPN
bracis-33603	231	2	kaplan	kaplan	PROPN
bracis-33603	231	3	,	,	PUNCT
bracis-33603	231	4	j.	j.	PROPN
bracis-33603	231	5	,	,	PUNCT
bracis-33603	231	6	mccandlish	mccandlish	PROPN
bracis-33603	231	7	,	,	PUNCT
bracis-33603	231	8	s.	s.	PROPN
bracis-33603	231	9	,	,	PUNCT
bracis-33603	231	10	et	et	PROPN
bracis-33603	231	11	 	 	SPACE
bracis-33603	231	12	al	al	PROPN
bracis-33603	231	13	.	.	PROPN
bracis-33603	231	14	:	:	PUNCT
bracis-33603	231	15	scaling	scale	VERB
bracis-33603	231	16	laws	law	NOUN
bracis-33603	231	17	for	for	ADP
bracis-33603	231	18	neural	neural	ADJ
bracis-33603	231	19	language	language	NOUN
bracis-33603	231	20	models	model	NOUN
bracis-33603	231	21	(	(	PUNCT
bracis-33603	231	22	2020	2020	NUM
bracis-33603	231	23	)	)	PUNCT
bracis-33603	231	24	google	google	PROPN
bracis-33603	231	25	scholar	scholar	NOUN
bracis-33603	231	26	  	  	SPACE
bracis-33603	231	27	kudo	kudo	PROPN
bracis-33603	231	28	,	,	PUNCT
bracis-33603	231	29	t.	t.	PROPN
bracis-33603	231	30	,	,	PUNCT
bracis-33603	231	31	richardson	richardson	PROPN
bracis-33603	231	32	,	,	PUNCT
bracis-33603	231	33	j.	j.	PROPN
bracis-33603	231	34	:	:	PUNCT
bracis-33603	231	35	sentencepiece	sentencepiece	PROPN
bracis-33603	231	36	:	:	PUNCT
bracis-33603	231	37	a	a	DET
bracis-33603	231	38	simple	simple	ADJ
bracis-33603	231	39	and	and	CCONJ
bracis-33603	231	40	language	language	NOUN
bracis-33603	231	41	independent	independent	PROPN
bracis-33603	231	42	subword	subword	PROPN
bracis-33603	231	43	tokenizer	tokenizer	NOUN
bracis-33603	231	44	and	and	CCONJ
bracis-33603	231	45	detokenizer	detokenizer	VERB
bracis-33603	231	46	for	for	ADP
bracis-33603	231	47	neural	neural	ADJ
bracis-33603	231	48	text	text	NOUN
bracis-33603	231	49	processing	processing	NOUN
bracis-33603	231	50	.	.	PUNCT
bracis-33603	232	1	arxiv	arxiv	PROPN
bracis-33603	232	2	preprint	preprint	VERB
bracis-33603	232	3	arxiv:1808.06226	arxiv:1808.06226	NUM
bracis-33603	232	4	(	(	PUNCT
bracis-33603	232	5	2018	2018	NUM
bracis-33603	232	6	)	)	PUNCT
bracis-33603	232	7	larcher	larcher	ADV
bracis-33603	232	8	,	,	PUNCT
bracis-33603	232	9	c.	c.	PROPN
bracis-33603	232	10	,	,	PUNCT
bracis-33603	232	11	piau	piau	NOUN
bracis-33603	232	12	,	,	PUNCT
bracis-33603	232	13	m.	m.	NOUN
bracis-33603	232	14	,	,	PUNCT
bracis-33603	232	15	finardi	finardi	PROPN
bracis-33603	232	16	,	,	PUNCT
bracis-33603	232	17	p.	p.	NOUN
bracis-33603	232	18	,	,	PUNCT
bracis-33603	232	19	gengo	gengo	PROPN
bracis-33603	232	20	,	,	PUNCT
bracis-33603	232	21	p.	p.	PROPN
bracis-33603	232	22	,	,	PUNCT
bracis-33603	232	23	esposito	esposito	PROPN
bracis-33603	232	24	,	,	PUNCT
bracis-33603	232	25	p.	p.	PROPN
bracis-33603	232	26	,	,	PUNCT
bracis-33603	232	27	caridá	caridá	NOUN
bracis-33603	232	28	,	,	PUNCT
bracis-33603	232	29	v.	v.	ADV
bracis-33603	232	30	:	:	PUNCT
bracis-33603	232	31	cabrita	cabrita	NOUN
bracis-33603	232	32	:	:	PUNCT
bracis-33603	232	33	closing	close	VERB
bracis-33603	232	34	the	the	DET
bracis-33603	232	35	gap	gap	NOUN
bracis-33603	232	36	for	for	ADP
bracis-33603	232	37	foreign	foreign	ADJ
bracis-33603	232	38	languages	language	NOUN
bracis-33603	232	39	(	(	PUNCT
bracis-33603	232	40	2023	2023	NUM
bracis-33603	232	41	)	)	PUNCT
bracis-33603	232	42	google	google	PROPN
bracis-33603	232	43	scholar	scholar	NOUN
bracis-33603	232	44	  	  	SPACE
bracis-33603	232	45	lin	lin	PROPN
bracis-33603	232	46	,	,	PUNCT
bracis-33603	232	47	j.	j.	PROPN
bracis-33603	232	48	,	,	PUNCT
bracis-33603	232	49	ma	ma	PROPN
bracis-33603	232	50	,	,	PUNCT
bracis-33603	232	51	x.	x.	PROPN
bracis-33603	232	52	,	,	PUNCT
bracis-33603	232	53	lin	lin	PROPN
bracis-33603	232	54	,	,	PUNCT
bracis-33603	232	55	s.c	s.c	PROPN
bracis-33603	232	56	.	.	PROPN
bracis-33603	232	57	,	,	PUNCT
bracis-33603	232	58	yang	yang	PROPN
bracis-33603	232	59	,	,	PUNCT
bracis-33603	232	60	j.h	j.h	PROPN
bracis-33603	232	61	.	.	PROPN
bracis-33603	232	62	,	,	PUNCT
bracis-33603	232	63	pradeep	pradeep	PROPN
bracis-33603	232	64	,	,	PUNCT
bracis-33603	232	65	r.	r.	PROPN
bracis-33603	232	66	,	,	PUNCT
bracis-33603	232	67	nogueira	nogueira	PROPN
bracis-33603	232	68	,	,	PUNCT
bracis-33603	232	69	r.	r.	PROPN
bracis-33603	232	70	:	:	PUNCT
bracis-33603	232	71	pyserini	pyserini	PROPN
bracis-33603	232	72	:	:	PUNCT
bracis-33603	232	73	a	a	DET
bracis-33603	232	74	python	python	NOUN
bracis-33603	232	75	toolkit	toolkit	NOUN
bracis-33603	232	76	for	for	ADP
bracis-33603	232	77	reproducible	reproducible	ADJ
bracis-33603	232	78	information	information	NOUN
bracis-33603	232	79	retrieval	retrieval	NOUN
bracis-33603	232	80	research	research	NOUN
bracis-33603	232	81	with	with	ADP
bracis-33603	232	82	sparse	sparse	ADJ
bracis-33603	232	83	and	and	CCONJ
bracis-33603	232	84	dense	dense	ADJ
bracis-33603	232	85	representations	representation	NOUN
bracis-33603	232	86	.	.	PUNCT
bracis-33603	233	1	in	in	ADP
bracis-33603	233	2	:	:	PUNCT
bracis-33603	233	3	proceedings	proceeding	NOUN
bracis-33603	233	4	of	of	ADP
bracis-33603	233	5	the	the	DET
bracis-33603	233	6	44th	44th	ADJ
bracis-33603	233	7	annual	annual	ADJ
bracis-33603	233	8	international	international	PROPN
bracis-33603	233	9	acm	acm	PROPN
bracis-33603	233	10	sigir	sigir	PROPN
bracis-33603	233	11	conference	conference	PROPN
bracis-33603	233	12	on	on	ADP
bracis-33603	233	13	research	research	NOUN
bracis-33603	233	14	and	and	CCONJ
bracis-33603	233	15	development	development	NOUN
bracis-33603	233	16	in	in	ADP
bracis-33603	233	17	information	information	NOUN
bracis-33603	233	18	retrieval	retrieval	NOUN
bracis-33603	233	19	,	,	PUNCT
bracis-33603	233	20	sigir	sigir	NOUN
bracis-33603	233	21	2021	2021	NUM
bracis-33603	233	22	,	,	PUNCT
bracis-33603	233	23	pp	pp	ADV
bracis-33603	233	24	.	.	PUNCT
bracis-33603	234	1	2356–2362	2356–2362	NUM
bracis-33603	234	2	(	(	PUNCT
bracis-33603	234	3	2021	2021	NUM
bracis-33603	234	4	)	)	PUNCT
bracis-33603	234	5	google	google	PROPN
bracis-33603	234	6	scholar	scholar	NOUN
bracis-33603	234	7	  	  	SPACE
bracis-33603	234	8	lopes	lopes	PROPN
bracis-33603	234	9	,	,	PUNCT
bracis-33603	234	10	r.	r.	PROPN
bracis-33603	234	11	,	,	PUNCT
bracis-33603	234	12	magalhães	magalhães	NOUN
bracis-33603	234	13	,	,	PUNCT
bracis-33603	234	14	j.	j.	PROPN
bracis-33603	234	15	,	,	PUNCT
bracis-33603	234	16	semedo	semedo	PROPN
bracis-33603	234	17	,	,	PUNCT
bracis-33603	234	18	d.	d.	PROPN
bracis-33603	234	19	:	:	PUNCT
bracis-33603	234	20	glória	glória	PROPN
bracis-33603	234	21	–	–	PUNCT
bracis-33603	234	22	a	a	DET
bracis-33603	234	23	generative	generative	NOUN
bracis-33603	234	24	and	and	CCONJ
bracis-33603	234	25	open	open	VERB
bracis-33603	234	26	large	large	ADJ
bracis-33603	234	27	language	language	NOUN
bracis-33603	234	28	model	model	NOUN
bracis-33603	234	29	for	for	ADP
bracis-33603	234	30	portuguese	portuguese	PROPN
bracis-33603	234	31	(	(	PUNCT
bracis-33603	234	32	2024	2024	NUM
bracis-33603	234	33	)	)	PUNCT
bracis-33603	234	34	google	google	NOUN
bracis-33603	234	35	scholar	scholar	NOUN
bracis-33603	234	36	  	  	SPACE
bracis-33603	234	37	de	de	PROPN
bracis-33603	234	38	 	 	SPACE
bracis-33603	234	39	morais	morais	PROPN
bracis-33603	234	40	,	,	PUNCT
bracis-33603	234	41	l.p	l.p	PROPN
bracis-33603	234	42	.	.	PROPN
bracis-33603	234	43	,	,	PUNCT
bracis-33603	234	44	da	da	PROPN
bracis-33603	234	45	 	 	SPACE
bracis-33603	234	46	silva	silva	PROPN
bracis-33603	234	47	 	 	SPACE
bracis-33603	234	48	soares	soares	PROPN
bracis-33603	234	49	,	,	PUNCT
bracis-33603	234	50	a.	a.	NOUN
bracis-33603	234	51	,	,	PUNCT
bracis-33603	234	52	da	da	NOUN
bracis-33603	234	53	 	 	SPACE
bracis-33603	234	54	cm	cm	NOUN
bracis-33603	234	55	 	 	SPACE
bracis-33603	234	56	borges	borge	NOUN
bracis-33603	234	57	,	,	PUNCT
bracis-33603	234	58	v.	v.	ADV
bracis-33603	234	59	,	,	PUNCT
bracis-33603	234	60	da	da	ADJ
bracis-33603	234	61	 	 	SPACE
bracis-33603	234	62	silva	silva	NOUN
bracis-33603	234	63	,	,	PUNCT
bracis-33603	234	64	n.f.f	n.f.f	PROPN
bracis-33603	234	65	.	.	PROPN
bracis-33603	234	66	,	,	PUNCT
bracis-33603	234	67	pereira	pereira	PROPN
bracis-33603	234	68	,	,	PUNCT
bracis-33603	234	69	f.s	f.s	PROPN
bracis-33603	234	70	.	.	PROPN
bracis-33603	234	71	:	:	PUNCT
bracis-33603	235	1	sub	sub	ADJ
bracis-33603	235	2	-	-	ADJ
bracis-33603	235	3	language	language	ADJ
bracis-33603	235	4	sentiment	sentiment	NOUN
bracis-33603	235	5	analysis	analysis	NOUN
bracis-33603	235	6	in	in	ADP
bracis-33603	235	7	whatsapp	whatsapp	ADJ
bracis-33603	235	8	domain	domain	NOUN
bracis-33603	235	9	with	with	ADP
bracis-33603	235	10	deep	deep	ADJ
bracis-33603	235	11	learning	learning	NOUN
bracis-33603	235	12	approaches	approach	NOUN
bracis-33603	235	13	.	.	PUNCT
bracis-33603	236	1	revista	revista	PROPN
bracis-33603	236	2	de	de	X
bracis-33603	236	3	sistemas	sistemas	PROPN
bracis-33603	236	4	de	de	PROPN
bracis-33603	236	5	informaçao	informaçao	PROPN
bracis-33603	236	6	da	da	PROPN
bracis-33603	236	7	fsma	fsma	PROPN
bracis-33603	236	8	1(31	1(31	NUM
bracis-33603	236	9	)	)	PUNCT
bracis-33603	236	10	,	,	PUNCT
bracis-33603	236	11	32–47	32–47	NUM
bracis-33603	236	12	(	(	PUNCT
bracis-33603	236	13	2023	2023	NUM
bracis-33603	236	14	)	)	PUNCT
bracis-33603	236	15	google	google	NOUN
bracis-33603	236	16	scholar	scholar	NOUN
bracis-33603	236	17	  	  	SPACE
bracis-33603	236	18	nagoudi	nagoudi	NOUN
bracis-33603	236	19	,	,	PUNCT
bracis-33603	236	20	e.m.b	e.m.b	PROPN
bracis-33603	236	21	.	.	PROPN
bracis-33603	236	22	,	,	PUNCT
bracis-33603	236	23	elmadany	elmadany	PROPN
bracis-33603	236	24	,	,	PUNCT
bracis-33603	236	25	a.	a.	NOUN
bracis-33603	236	26	,	,	PUNCT
bracis-33603	236	27	abdul	abdul	PROPN
bracis-33603	236	28	-	-	PUNCT
bracis-33603	236	29	mageed	mageed	NOUN
bracis-33603	236	30	,	,	PUNCT
bracis-33603	236	31	m.	m.	NOUN
bracis-33603	236	32	:	:	PUNCT
bracis-33603	236	33	arat5	arat5	PROPN
bracis-33603	236	34	:	:	PUNCT
bracis-33603	236	35	text	text	NOUN
bracis-33603	236	36	-	-	PUNCT
bracis-33603	236	37	to	to	ADP
bracis-33603	236	38	-	-	PUNCT
bracis-33603	236	39	text	text	NOUN
bracis-33603	236	40	transformers	transformer	NOUN
bracis-33603	236	41	for	for	ADP
bracis-33603	236	42	arabic	arabic	ADJ
bracis-33603	236	43	language	language	NOUN
bracis-33603	236	44	generation	generation	NOUN
bracis-33603	236	45	(	(	PUNCT
bracis-33603	236	46	2022	2022	NUM
bracis-33603	236	47	)	)	PUNCT
bracis-33603	236	48	google	google	PROPN
bracis-33603	236	49	scholar	scholar	NOUN
bracis-33603	236	50	  	  	SPACE
bracis-33603	236	51	nogueira	nogueira	PROPN
bracis-33603	236	52	,	,	PUNCT
bracis-33603	236	53	r.	r.	PROPN
bracis-33603	236	54	,	,	PUNCT
bracis-33603	236	55	jiang	jiang	PROPN
bracis-33603	236	56	,	,	PUNCT
bracis-33603	236	57	z.	z.	PROPN
bracis-33603	236	58	,	,	PUNCT
bracis-33603	236	59	pradeep	pradeep	PROPN
bracis-33603	236	60	,	,	PUNCT
bracis-33603	236	61	r.	r.	PROPN
bracis-33603	236	62	,	,	PUNCT
bracis-33603	236	63	lin	lin	PROPN
bracis-33603	236	64	,	,	PUNCT
bracis-33603	236	65	j.	j.	PROPN
bracis-33603	236	66	:	:	PUNCT
bracis-33603	236	67	document	document	NOUN
bracis-33603	236	68	ranking	rank	VERB
bracis-33603	236	69	with	with	ADP
bracis-33603	236	70	a	a	DET
bracis-33603	236	71	pretrained	pretraine	VERB
bracis-33603	236	72	sequence	sequence	NOUN
bracis-33603	236	73	-	-	PUNCT
bracis-33603	236	74	to	to	ADP
bracis-33603	236	75	-	-	PUNCT
bracis-33603	236	76	sequence	sequence	NOUN
bracis-33603	236	77	model	model	NOUN
bracis-33603	236	78	.	.	PUNCT
bracis-33603	237	1	in	in	ADP
bracis-33603	237	2	:	:	PUNCT
bracis-33603	237	3	cohn	cohn	PROPN
bracis-33603	237	4	,	,	PUNCT
bracis-33603	237	5	t.	t.	PROPN
bracis-33603	237	6	,	,	PUNCT
bracis-33603	237	7	he	he	PRON
bracis-33603	237	8	,	,	PUNCT
bracis-33603	237	9	y.	y.	PROPN
bracis-33603	237	10	,	,	PUNCT
bracis-33603	237	11	liu	liu	PROPN
bracis-33603	237	12	,	,	PUNCT
bracis-33603	237	13	y.	y.	PROPN
bracis-33603	237	14	(	(	PUNCT
bracis-33603	237	15	eds	eds	PROPN
bracis-33603	237	16	.	.	PUNCT
bracis-33603	237	17	)	)	PUNCT
bracis-33603	238	1	findings	finding	NOUN
bracis-33603	238	2	of	of	ADP
bracis-33603	238	3	the	the	DET
bracis-33603	238	4	association	association	NOUN
bracis-33603	238	5	for	for	ADP
bracis-33603	238	6	computational	computational	ADJ
bracis-33603	238	7	linguistics	linguistic	NOUN
bracis-33603	238	8	,	,	PUNCT
bracis-33603	238	9	emnlp	emnlp	ADP
bracis-33603	238	10	2020	2020	NUM
bracis-33603	238	11	,	,	PUNCT
bracis-33603	238	12	online	online	INTJ
bracis-33603	238	13	,	,	PUNCT
bracis-33603	238	14	november	november	PROPN
bracis-33603	238	15	2020	2020	NUM
bracis-33603	238	16	,	,	PUNCT
bracis-33603	238	17	pp	pp	ADV
bracis-33603	238	18	.	.	PUNCT
bracis-33603	239	1	708–718	708–718	NUM
bracis-33603	239	2	.	.	PUNCT
bracis-33603	239	3	association	association	NOUN
bracis-33603	239	4	for	for	ADP
bracis-33603	239	5	computational	computational	ADJ
bracis-33603	239	6	linguistics	linguistic	NOUN
bracis-33603	239	7	(	(	PUNCT
bracis-33603	239	8	2020	2020	NUM
bracis-33603	239	9	)	)	PUNCT
bracis-33603	239	10	.	.	PUNCT
bracis-33603	240	1	https://doi.org/10.18653/v1/2020.findings-emnlp.63	https://doi.org/10.18653/v1/2020.findings-emnlp.63	PROPN
bracis-33603	240	2	oliveira	oliveira	PROPN
bracis-33603	240	3	,	,	PUNCT
bracis-33603	240	4	h.g	h.g	PROPN
bracis-33603	240	5	.	.	PROPN
bracis-33603	240	6	,	,	PUNCT
bracis-33603	240	7	real	real	ADJ
bracis-33603	240	8	,	,	PUNCT
bracis-33603	240	9	l.	l.	PROPN
bracis-33603	240	10	,	,	PUNCT
bracis-33603	240	11	fonseca	fonseca	PROPN
bracis-33603	240	12	,	,	PUNCT
bracis-33603	240	13	e.	e.	PROPN
bracis-33603	240	14	(	(	PUNCT
bracis-33603	240	15	eds	eds	PROPN
bracis-33603	240	16	.	.	PUNCT
bracis-33603	240	17	):	):	PUNCT
bracis-33603	240	18	proceedings	proceeding	NOUN
bracis-33603	240	19	of	of	ADP
bracis-33603	240	20	the	the	DET
bracis-33603	240	21	assin	assin	PROPN
bracis-33603	240	22	2	2	NUM
bracis-33603	240	23	shared	share	VERB
bracis-33603	240	24	task	task	NOUN
bracis-33603	240	25	:	:	PUNCT
bracis-33603	240	26	evaluating	evaluate	VERB
bracis-33603	240	27	semantic	semantic	ADJ
bracis-33603	240	28	textual	textual	ADJ
bracis-33603	240	29	similarity	similarity	NOUN
bracis-33603	240	30	and	and	CCONJ
bracis-33603	240	31	textual	textual	ADJ
bracis-33603	240	32	entailment	entailment	NOUN
bracis-33603	240	33	in	in	ADP
bracis-33603	240	34	portuguese	portuguese	PROPN
bracis-33603	240	35	,	,	PUNCT
bracis-33603	240	36	extended	extend	VERB
bracis-33603	240	37	semantic	semantic	ADJ
bracis-33603	240	38	web	web	NOUN
bracis-33603	240	39	conference	conference	NOUN
bracis-33603	240	40	,	,	PUNCT
bracis-33603	240	41	no	no	INTJ
bracis-33603	240	42	.	.	PUNCT
bracis-33603	240	43	 	 	SPACE
bracis-33603	240	44	2583	2583	NUM
bracis-33603	240	45	in	in	ADP
bracis-33603	240	46	ceur	ceur	NOUN
bracis-33603	240	47	workshop	workshop	NOUN
bracis-33603	240	48	proceedings	proceeding	NOUN
bracis-33603	240	49	(	(	PUNCT
bracis-33603	240	50	2020	2020	NUM
bracis-33603	240	51	)	)	PUNCT
bracis-33603	240	52	google	google	PROPN
bracis-33603	240	53	scholar	scholar	NOUN
bracis-33603	240	54	  	  	SPACE
bracis-33603	240	55	pires	pire	NOUN
bracis-33603	240	56	,	,	PUNCT
bracis-33603	240	57	r.	r.	PROPN
bracis-33603	240	58	,	,	PUNCT
bracis-33603	240	59	abonizio	abonizio	PROPN
bracis-33603	240	60	,	,	PUNCT
bracis-33603	240	61	h.	h.	PROPN
bracis-33603	240	62	,	,	PUNCT
bracis-33603	240	63	almeida	almeida	PROPN
bracis-33603	240	64	,	,	PUNCT
bracis-33603	240	65	t.s	t.s	PROPN
bracis-33603	240	66	.	.	PROPN
bracis-33603	240	67	,	,	PUNCT
bracis-33603	240	68	nogueira	nogueira	PROPN
bracis-33603	240	69	,	,	PUNCT
bracis-33603	240	70	r.	r.	PROPN
bracis-33603	240	71	:	:	PUNCT
bracis-33603	240	72	sabiá	sabiá	NOUN
bracis-33603	240	73	:	:	PUNCT
bracis-33603	240	74	portuguese	portuguese	VERB
bracis-33603	240	75	large	large	ADJ
bracis-33603	240	76	language	language	NOUN
bracis-33603	240	77	models	model	NOUN
bracis-33603	240	78	,	,	PUNCT
bracis-33603	240	79	pp	pp	ADV
bracis-33603	240	80	.	.	PUNCT
bracis-33603	241	1	226–240	226–240	NUM
bracis-33603	241	2	.	.	PUNCT
bracis-33603	241	3	springer	springer	NOUN
bracis-33603	241	4	,	,	PUNCT
bracis-33603	241	5	heidelberg	heidelberg	PROPN
bracis-33603	241	6	(	(	PUNCT
bracis-33603	241	7	2023	2023	NUM
bracis-33603	241	8	)	)	PUNCT
bracis-33603	241	9	.	.	PUNCT
bracis-33603	242	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-33603	242	2	rae	rae	PROPN
bracis-33603	242	3	,	,	PUNCT
bracis-33603	242	4	j.w	j.w	PROPN
bracis-33603	242	5	.	.	PROPN
bracis-33603	242	6	,	,	PUNCT
bracis-33603	242	7	et	et	PROPN
bracis-33603	242	8	 	 	SPACE
bracis-33603	242	9	al	al	PROPN
bracis-33603	242	10	.	.	PROPN
bracis-33603	242	11	:	:	PUNCT
bracis-33603	242	12	scaling	scale	VERB
bracis-33603	242	13	language	language	NOUN
bracis-33603	242	14	models	model	NOUN
bracis-33603	242	15	:	:	PUNCT
bracis-33603	242	16	methods	method	NOUN
bracis-33603	242	17	,	,	PUNCT
bracis-33603	242	18	analysis	analysis	NOUN
bracis-33603	242	19	&	&	CCONJ
bracis-33603	242	20	insights	insight	NOUN
bracis-33603	242	21	from	from	ADP
bracis-33603	242	22	training	train	VERB
bracis-33603	242	23	gopher	gopher	NOUN
bracis-33603	242	24	(	(	PUNCT
bracis-33603	242	25	2022	2022	NUM
bracis-33603	242	26	)	)	PUNCT
bracis-33603	242	27	google	google	NOUN
bracis-33603	242	28	scholar	scholar	NOUN
bracis-33603	242	29	  	  	SPACE
bracis-33603	242	30	raffel	raffel	NOUN
bracis-33603	242	31	,	,	PUNCT
bracis-33603	242	32	c.	c.	PROPN
bracis-33603	242	33	,	,	PUNCT
bracis-33603	242	34	et	et	PROPN
bracis-33603	242	35	 	 	SPACE
bracis-33603	242	36	al	al	PROPN
bracis-33603	242	37	.	.	PROPN
bracis-33603	242	38	:	:	PUNCT
bracis-33603	242	39	exploring	explore	VERB
bracis-33603	242	40	the	the	DET
bracis-33603	242	41	limits	limit	NOUN
bracis-33603	242	42	of	of	ADP
bracis-33603	242	43	transfer	transfer	NOUN
bracis-33603	242	44	learning	learn	VERB
bracis-33603	242	45	with	with	ADP
bracis-33603	242	46	a	a	DET
bracis-33603	242	47	unified	unified	ADJ
bracis-33603	242	48	text	text	NOUN
bracis-33603	242	49	-	-	PUNCT
bracis-33603	242	50	to	to	ADP
bracis-33603	242	51	-	-	PUNCT
bracis-33603	242	52	text	text	NOUN
bracis-33603	242	53	transformer	transformer	NOUN
bracis-33603	242	54	.	.	PUNCT
bracis-33603	243	1	arxiv	arxiv	PROPN
bracis-33603	243	2	preprint	preprint	VERB
bracis-33603	243	3	arxiv:1910.10683	arxiv:1910.10683	NOUN
bracis-33603	243	4	(	(	PUNCT
bracis-33603	243	5	2019	2019	NUM
bracis-33603	243	6	)	)	PUNCT
bracis-33603	243	7	rajpurkar	rajpurkar	NOUN
bracis-33603	243	8	,	,	PUNCT
bracis-33603	243	9	p.	p.	PROPN
bracis-33603	243	10	,	,	PUNCT
bracis-33603	243	11	zhang	zhang	PROPN
bracis-33603	243	12	,	,	PUNCT
bracis-33603	243	13	j.	j.	PROPN
bracis-33603	243	14	,	,	PUNCT
bracis-33603	243	15	lopyrev	lopyrev	PROPN
bracis-33603	243	16	,	,	PUNCT
bracis-33603	243	17	k.	k.	PROPN
bracis-33603	243	18	,	,	PUNCT
bracis-33603	243	19	liang	liang	PROPN
bracis-33603	243	20	,	,	PUNCT
bracis-33603	243	21	p.	p.	NOUN
bracis-33603	243	22	:	:	PUNCT
bracis-33603	244	1	squad	squad	NOUN
bracis-33603	244	2	:	:	PUNCT
bracis-33603	244	3	100,000	100,000	NUM
bracis-33603	244	4	+	+	NUM
bracis-33603	244	5	questions	question	NOUN
bracis-33603	244	6	for	for	ADP
bracis-33603	244	7	machine	machine	NOUN
bracis-33603	244	8	comprehension	comprehension	NOUN
bracis-33603	244	9	of	of	ADP
bracis-33603	244	10	text	text	NOUN
bracis-33603	244	11	(	(	PUNCT
bracis-33603	244	12	2016	2016	NUM
bracis-33603	244	13	)	)	PUNCT
bracis-33603	244	14	google	google	PROPN
bracis-33603	244	15	scholar	scholar	NOUN
bracis-33603	244	16	  	  	SPACE
bracis-33603	244	17	roberts	roberts	PROPN
bracis-33603	244	18	,	,	PUNCT
bracis-33603	244	19	a.	a.	PROPN
bracis-33603	244	20	,	,	PUNCT
bracis-33603	244	21	et	et	PROPN
bracis-33603	244	22	 	 	SPACE
bracis-33603	244	23	al	al	PROPN
bracis-33603	244	24	.	.	PROPN
bracis-33603	244	25	:	:	PUNCT
bracis-33603	244	26	scaling	scale	VERB
bracis-33603	244	27	up	up	ADP
bracis-33603	244	28	models	model	NOUN
bracis-33603	244	29	and	and	CCONJ
bracis-33603	244	30	data	datum	NOUN
bracis-33603	244	31	with	with	ADP
bracis-33603	244	32	t5x	t5x	NOUN
bracis-33603	244	33	and	and	CCONJ
bracis-33603	244	34	seqio	seqio	NOUN
bracis-33603	244	35	(	(	PUNCT
bracis-33603	244	36	2022	2022	NUM
bracis-33603	244	37	)	)	PUNCT
bracis-33603	244	38	google	google	PROPN
bracis-33603	244	39	scholar	scholar	NOUN
bracis-33603	244	40	  	  	SPACE
bracis-33603	244	41	rodrigues	rodrigue	NOUN
bracis-33603	244	42	,	,	PUNCT
bracis-33603	244	43	j.	j.	PROPN
bracis-33603	244	44	,	,	PUNCT
bracis-33603	244	45	et	et	PROPN
bracis-33603	244	46	 	 	SPACE
bracis-33603	244	47	al	al	PROPN
bracis-33603	244	48	.	.	PROPN
bracis-33603	244	49	:	:	PUNCT
bracis-33603	244	50	advancing	advance	VERB
bracis-33603	244	51	neural	neural	ADJ
bracis-33603	244	52	encoding	encoding	NOUN
bracis-33603	244	53	of	of	ADP
bracis-33603	244	54	portuguese	portuguese	PROPN
bracis-33603	244	55	with	with	ADP
bracis-33603	244	56	transformer	transformer	NOUN
bracis-33603	244	57	albertina	albertina	PROPN
bracis-33603	244	58	pt-	pt-	X
bracis-33603	244	59	*	*	PUNCT
bracis-33603	244	60	(	(	PUNCT
bracis-33603	244	61	2023	2023	NUM
bracis-33603	244	62	)	)	PUNCT
bracis-33603	244	63	google	google	PROPN
bracis-33603	244	64	scholar	scholar	NOUN
bracis-33603	244	65	  	  	SPACE
bracis-33603	244	66	rosa	rosa	PROPN
bracis-33603	244	67	,	,	PUNCT
bracis-33603	244	68	g.m	g.m	PROPN
bracis-33603	244	69	.	.	PROPN
bracis-33603	244	70	,	,	PUNCT
bracis-33603	244	71	bonifacio	bonifacio	PROPN
bracis-33603	244	72	,	,	PUNCT
bracis-33603	244	73	l.h	l.h	PROPN
bracis-33603	244	74	.	.	PROPN
bracis-33603	244	75	,	,	PUNCT
bracis-33603	244	76	de	de	NOUN
bracis-33603	244	77	 	 	SPACE
bracis-33603	244	78	souza	souza	PROPN
bracis-33603	244	79	,	,	PUNCT
bracis-33603	244	80	l.r	l.r	PROPN
bracis-33603	244	81	.	.	PROPN
bracis-33603	244	82	,	,	PUNCT
bracis-33603	244	83	lotufo	lotufo	PROPN
bracis-33603	244	84	,	,	PUNCT
bracis-33603	244	85	r.	r.	PROPN
bracis-33603	244	86	,	,	PUNCT
bracis-33603	244	87	nogueira	nogueira	PROPN
bracis-33603	244	88	,	,	PUNCT
bracis-33603	244	89	r.	r.	PROPN
bracis-33603	244	90	:	:	PUNCT
bracis-33603	244	91	a	a	DET
bracis-33603	244	92	cost	cost	NOUN
bracis-33603	244	93	-	-	PUNCT
bracis-33603	244	94	benefit	benefit	NOUN
bracis-33603	244	95	analysis	analysis	NOUN
bracis-33603	244	96	of	of	ADP
bracis-33603	244	97	cross	cross	ADJ
bracis-33603	244	98	-	-	ADJ
bracis-33603	244	99	lingual	lingual	ADJ
bracis-33603	244	100	transfer	transfer	NOUN
bracis-33603	244	101	methods	method	NOUN
bracis-33603	244	102	(	(	PUNCT
bracis-33603	244	103	2021	2021	NUM
bracis-33603	244	104	)	)	PUNCT
bracis-33603	244	105	google	google	NOUN
bracis-33603	244	106	scholar	scholar	NOUN
bracis-33603	244	107	  	  	SPACE
bracis-33603	244	108	santos	santo	NOUN
bracis-33603	244	109	,	,	PUNCT
bracis-33603	244	110	r.	r.	PROPN
bracis-33603	244	111	,	,	PUNCT
bracis-33603	244	112	silva	silva	PROPN
bracis-33603	244	113	,	,	PUNCT
bracis-33603	244	114	j.	j.	PROPN
bracis-33603	244	115	,	,	PUNCT
bracis-33603	244	116	gomes	gomes	PROPN
bracis-33603	244	117	,	,	PUNCT
bracis-33603	244	118	l.	l.	PROPN
bracis-33603	244	119	,	,	PUNCT
bracis-33603	244	120	rodrigues	rodrigues	PROPN
bracis-33603	244	121	,	,	PUNCT
bracis-33603	244	122	j.	j.	PROPN
bracis-33603	244	123	,	,	PUNCT
bracis-33603	244	124	branco	branco	PROPN
bracis-33603	244	125	,	,	PUNCT
bracis-33603	244	126	a.	a.	NOUN
bracis-33603	244	127	:	:	PUNCT
bracis-33603	244	128	advancing	advance	VERB
bracis-33603	244	129	generative	generative	NOUN
bracis-33603	244	130	ai	ai	VERB
bracis-33603	244	131	for	for	ADP
bracis-33603	244	132	portuguese	portuguese	ADJ
bracis-33603	244	133	with	with	ADP
bracis-33603	244	134	open	open	ADJ
bracis-33603	244	135	decoder	decoder	NOUN
bracis-33603	244	136	gervásio	gervásio	NOUN
bracis-33603	244	137	pt	pt	PROPN
bracis-33603	244	138	*	*	X
bracis-33603	244	139	(	(	PUNCT
bracis-33603	244	140	2024	2024	NUM
bracis-33603	244	141	)	)	PUNCT
bracis-33603	244	142	google	google	PROPN
bracis-33603	244	143	scholar	scholar	NOUN
bracis-33603	244	144	  	  	SPACE
bracis-33603	244	145	sarti	sarti	NOUN
bracis-33603	244	146	,	,	PUNCT
bracis-33603	244	147	g.	g.	PROPN
bracis-33603	244	148	,	,	PUNCT
bracis-33603	244	149	nissim	nissim	PROPN
bracis-33603	244	150	,	,	PUNCT
bracis-33603	244	151	m.	m.	NOUN
bracis-33603	244	152	:	:	PUNCT
bracis-33603	244	153	it5	it5	NOUN
bracis-33603	244	154	:	:	PUNCT
bracis-33603	244	155	large	large	ADJ
bracis-33603	244	156	-	-	PUNCT
bracis-33603	244	157	scale	scale	NOUN
bracis-33603	244	158	text	text	NOUN
bracis-33603	244	159	-	-	PUNCT
bracis-33603	244	160	to	to	ADP
bracis-33603	244	161	-	-	PUNCT
bracis-33603	244	162	text	text	NOUN
bracis-33603	244	163	pretraining	pretraine	VERB
bracis-33603	244	164	for	for	ADP
bracis-33603	244	165	italian	italian	ADJ
bracis-33603	244	166	language	language	NOUN
bracis-33603	244	167	understanding	understanding	NOUN
bracis-33603	244	168	and	and	CCONJ
bracis-33603	244	169	generation	generation	NOUN
bracis-33603	244	170	(	(	PUNCT
bracis-33603	244	171	2022	2022	NUM
bracis-33603	244	172	)	)	PUNCT
bracis-33603	244	173	google	google	NOUN
bracis-33603	244	174	scholar	scholar	NOUN
bracis-33603	244	175	  	  	SPACE
bracis-33603	244	176	shazeer	shazeer	NOUN
bracis-33603	244	177	,	,	PUNCT
bracis-33603	244	178	n.	n.	NOUN
bracis-33603	244	179	,	,	PUNCT
bracis-33603	244	180	stern	stern	ADJ
bracis-33603	244	181	,	,	PUNCT
bracis-33603	244	182	m.	m.	NOUN
bracis-33603	244	183	:	:	PUNCT
bracis-33603	244	184	adafactor	adafactor	NOUN
bracis-33603	244	185	:	:	PUNCT
bracis-33603	244	186	adaptive	adaptive	ADJ
bracis-33603	244	187	learning	learning	NOUN
bracis-33603	244	188	rates	rate	NOUN
bracis-33603	244	189	with	with	ADP
bracis-33603	244	190	sublinear	sublinear	NOUN
bracis-33603	244	191	memory	memory	NOUN
bracis-33603	244	192	cost	cost	NOUN
bracis-33603	244	193	(	(	PUNCT
bracis-33603	244	194	2018	2018	NUM
bracis-33603	244	195	)	)	PUNCT
bracis-33603	244	196	google	google	PROPN
bracis-33603	244	197	scholar	scholar	NOUN
bracis-33603	244	198	  	  	SPACE
bracis-33603	244	199	souza	souza	PROPN
bracis-33603	244	200	,	,	PUNCT
bracis-33603	244	201	f.	f.	PROPN
bracis-33603	244	202	,	,	PUNCT
bracis-33603	244	203	nogueira	nogueira	PROPN
bracis-33603	244	204	,	,	PUNCT
bracis-33603	244	205	r.	r.	PROPN
bracis-33603	244	206	,	,	PUNCT
bracis-33603	244	207	lotufo	lotufo	PROPN
bracis-33603	244	208	,	,	PUNCT
bracis-33603	244	209	r.	r.	PROPN
bracis-33603	244	210	:	:	PUNCT
bracis-33603	244	211	bertimbau	bertimbau	PROPN
bracis-33603	244	212	:	:	PUNCT
bracis-33603	244	213	pretrained	pretraine	VERB
bracis-33603	244	214	bert	bert	NOUN
bracis-33603	244	215	models	model	NOUN
bracis-33603	244	216	for	for	ADP
bracis-33603	244	217	brazilian	brazilian	ADJ
bracis-33603	244	218	portuguese	portuguese	NOUN
bracis-33603	244	219	.	.	PUNCT
bracis-33603	245	1	in	in	ADP
bracis-33603	245	2	:	:	PUNCT
bracis-33603	245	3	cerri	cerri	PROPN
bracis-33603	245	4	,	,	PUNCT
bracis-33603	245	5	r.	r.	PROPN
bracis-33603	245	6	,	,	PUNCT
bracis-33603	245	7	prati	prati	PROPN
bracis-33603	245	8	,	,	PUNCT
bracis-33603	245	9	r.c	r.c	PROPN
bracis-33603	245	10	.	.	PROPN
bracis-33603	245	11	(	(	PUNCT
bracis-33603	245	12	eds	ed	NOUN
bracis-33603	245	13	.	.	PUNCT
bracis-33603	245	14	)	)	PUNCT
bracis-33603	245	15	lncs	lncs	PROPN
bracis-33603	245	16	,	,	PUNCT
bracis-33603	245	17	pp	pp	X
bracis-33603	245	18	.	.	PUNCT
bracis-33603	246	1	403–417	403–417	NUM
bracis-33603	246	2	.	.	PUNCT
bracis-33603	246	3	springer	springer	NOUN
bracis-33603	246	4	,	,	PUNCT
bracis-33603	246	5	cham	cham	PROPN
bracis-33603	246	6	(	(	PUNCT
bracis-33603	246	7	2020	2020	NUM
bracis-33603	246	8	)	)	PUNCT
bracis-33603	246	9	.	.	PUNCT
bracis-33603	247	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-33603	247	2	chapter	chapter	NOUN
bracis-33603	247	3	  	  	SPACE
bracis-33603	247	4	google	google	PROPN
bracis-33603	247	5	scholar	scholar	NOUN
bracis-33603	247	6	  	  	SPACE
bracis-33603	247	7	srivastava	srivastava	PROPN
bracis-33603	247	8	,	,	PUNCT
bracis-33603	247	9	a.	a.	PROPN
bracis-33603	247	10	,	,	PUNCT
bracis-33603	247	11	et	et	PROPN
bracis-33603	247	12	 	 	SPACE
bracis-33603	247	13	al	al	PROPN
bracis-33603	247	14	.	.	PROPN
bracis-33603	247	15	:	:	PUNCT
bracis-33603	248	1	beyond	beyond	ADP
bracis-33603	248	2	the	the	DET
bracis-33603	248	3	imitation	imitation	NOUN
bracis-33603	248	4	game	game	NOUN
bracis-33603	248	5	:	:	PUNCT
bracis-33603	248	6	quantifying	quantify	VERB
bracis-33603	248	7	and	and	CCONJ
bracis-33603	248	8	extrapolating	extrapolate	VERB
bracis-33603	248	9	the	the	DET
bracis-33603	248	10	capabilities	capability	NOUN
bracis-33603	248	11	of	of	ADP
bracis-33603	248	12	language	language	NOUN
bracis-33603	248	13	models	model	NOUN
bracis-33603	248	14	(	(	PUNCT
bracis-33603	248	15	2023	2023	NUM
bracis-33603	248	16	)	)	PUNCT
bracis-33603	248	17	google	google	PROPN
bracis-33603	248	18	scholar	scholar	NOUN
bracis-33603	248	19	  	  	SPACE
bracis-33603	248	20	wagner	wagner	NOUN
bracis-33603	248	21	 	 	SPACE
bracis-33603	248	22	filho	filho	PROPN
bracis-33603	248	23	,	,	PUNCT
bracis-33603	248	24	j.a	j.a	PROPN
bracis-33603	248	25	.	.	PROPN
bracis-33603	248	26	,	,	PUNCT
bracis-33603	248	27	wilkens	wilkens	PROPN
bracis-33603	248	28	,	,	PUNCT
bracis-33603	248	29	r.	r.	PROPN
bracis-33603	248	30	,	,	PUNCT
bracis-33603	248	31	idiart	idiart	NOUN
bracis-33603	248	32	,	,	PUNCT
bracis-33603	248	33	m.	m.	NOUN
bracis-33603	248	34	,	,	PUNCT
bracis-33603	248	35	villavicencio	villavicencio	NOUN
bracis-33603	248	36	,	,	PUNCT
bracis-33603	248	37	a.	a.	NOUN
bracis-33603	248	38	:	:	PUNCT
bracis-33603	248	39	the	the	DET
bracis-33603	248	40	brwac	brwac	PROPN
bracis-33603	248	41	corpus	corpus	PROPN
bracis-33603	248	42	:	:	PUNCT
bracis-33603	248	43	a	a	DET
bracis-33603	248	44	new	new	ADJ
bracis-33603	248	45	open	open	ADJ
bracis-33603	248	46	resource	resource	NOUN
bracis-33603	248	47	for	for	ADP
bracis-33603	248	48	brazilian	brazilian	ADJ
bracis-33603	248	49	portuguese	portuguese	NOUN
bracis-33603	248	50	.	.	PUNCT
bracis-33603	249	1	in	in	ADP
bracis-33603	249	2	:	:	PUNCT
bracis-33603	249	3	proceedings	proceeding	NOUN
bracis-33603	249	4	of	of	ADP
bracis-33603	249	5	the	the	DET
bracis-33603	249	6	eleventh	eleventh	ADJ
bracis-33603	249	7	international	international	ADJ
bracis-33603	249	8	conference	conference	NOUN
bracis-33603	249	9	on	on	ADP
bracis-33603	249	10	language	language	NOUN
bracis-33603	249	11	resources	resource	NOUN
bracis-33603	249	12	and	and	CCONJ
bracis-33603	249	13	evaluation	evaluation	NOUN
bracis-33603	249	14	,	,	PUNCT
bracis-33603	249	15	lrec	lrec	PROPN
bracis-33603	249	16	2018	2018	NUM
bracis-33603	249	17	(	(	PUNCT
bracis-33603	249	18	2018	2018	NUM
bracis-33603	249	19	)	)	PUNCT
bracis-33603	249	20	google	google	PROPN
bracis-33603	249	21	scholar	scholar	NOUN
bracis-33603	249	22	  	  	SPACE
bracis-33603	249	23	wang	wang	PROPN
bracis-33603	249	24	,	,	PUNCT
bracis-33603	249	25	a.	a.	PROPN
bracis-33603	249	26	,	,	PUNCT
bracis-33603	249	27	et	et	PROPN
bracis-33603	249	28	 	 	SPACE
bracis-33603	249	29	al	al	PROPN
bracis-33603	249	30	.	.	PROPN
bracis-33603	249	31	:	:	PUNCT
bracis-33603	249	32	superglue	superglue	NOUN
bracis-33603	249	33	:	:	PUNCT
bracis-33603	249	34	a	a	DET
bracis-33603	249	35	stickier	sticki	ADJ
bracis-33603	249	36	benchmark	benchmark	NOUN
bracis-33603	249	37	for	for	ADP
bracis-33603	249	38	general	general	ADJ
bracis-33603	249	39	-	-	PUNCT
bracis-33603	249	40	purpose	purpose	NOUN
bracis-33603	249	41	language	language	NOUN
bracis-33603	249	42	understanding	understanding	NOUN
bracis-33603	249	43	systems	system	NOUN
bracis-33603	249	44	(	(	PUNCT
bracis-33603	249	45	2020	2020	NUM
bracis-33603	249	46	)	)	PUNCT
bracis-33603	249	47	google	google	PROPN
bracis-33603	249	48	scholar	scholar	NOUN
bracis-33603	249	49	  	  	SPACE
bracis-33603	249	50	wang	wang	PROPN
bracis-33603	249	51	,	,	PUNCT
bracis-33603	249	52	a.	a.	PROPN
bracis-33603	249	53	,	,	PUNCT
bracis-33603	249	54	singh	singh	PROPN
bracis-33603	249	55	,	,	PUNCT
bracis-33603	249	56	a.	a.	PROPN
bracis-33603	249	57	,	,	PUNCT
bracis-33603	249	58	michael	michael	PROPN
bracis-33603	249	59	,	,	PUNCT
bracis-33603	249	60	j.	j.	PROPN
bracis-33603	249	61	,	,	PUNCT
bracis-33603	249	62	hill	hill	PROPN
bracis-33603	249	63	,	,	PUNCT
bracis-33603	249	64	f.	f.	PROPN
bracis-33603	249	65	,	,	PUNCT
bracis-33603	249	66	levy	levy	NOUN
bracis-33603	249	67	,	,	PUNCT
bracis-33603	249	68	o.	o.	PROPN
bracis-33603	249	69	,	,	PUNCT
bracis-33603	249	70	bowman	bowman	PROPN
bracis-33603	249	71	,	,	PUNCT
bracis-33603	249	72	s.r	s.r	PROPN
bracis-33603	249	73	.	.	PROPN
bracis-33603	249	74	:	:	PUNCT
bracis-33603	250	1	glue	glue	NOUN
bracis-33603	250	2	:	:	PUNCT
bracis-33603	250	3	a	a	DET
bracis-33603	250	4	multi	multi	ADJ
bracis-33603	250	5	-	-	ADJ
bracis-33603	250	6	task	task	ADJ
bracis-33603	250	7	benchmark	benchmark	NOUN
bracis-33603	250	8	and	and	CCONJ
bracis-33603	250	9	analysis	analysis	NOUN
bracis-33603	250	10	platform	platform	NOUN
bracis-33603	250	11	for	for	ADP
bracis-33603	250	12	natural	natural	ADJ
bracis-33603	250	13	language	language	NOUN
bracis-33603	250	14	understanding	understanding	NOUN
bracis-33603	250	15	(	(	PUNCT
bracis-33603	250	16	2019	2019	NUM
bracis-33603	250	17	)	)	PUNCT
bracis-33603	250	18	google	google	PROPN
bracis-33603	250	19	scholar	scholar	NOUN
bracis-33603	250	20	  	  	SPACE
bracis-33603	250	21	wang	wang	PROPN
bracis-33603	250	22	,	,	PUNCT
bracis-33603	250	23	l.	l.	PROPN
bracis-33603	250	24	,	,	PUNCT
bracis-33603	250	25	yang	yang	PROPN
bracis-33603	250	26	,	,	PUNCT
bracis-33603	250	27	n.	n.	PROPN
bracis-33603	250	28	,	,	PUNCT
bracis-33603	250	29	huang	huang	PROPN
bracis-33603	250	30	,	,	PUNCT
bracis-33603	250	31	x.	x.	PROPN
bracis-33603	250	32	,	,	PUNCT
bracis-33603	250	33	yang	yang	PROPN
bracis-33603	250	34	,	,	PUNCT
bracis-33603	250	35	l.	l.	PROPN
bracis-33603	250	36	,	,	PUNCT
bracis-33603	250	37	majumder	majumder	PROPN
bracis-33603	250	38	,	,	PUNCT
bracis-33603	250	39	r.	r.	PROPN
bracis-33603	250	40	,	,	PUNCT
bracis-33603	250	41	wei	wei	PROPN
bracis-33603	250	42	,	,	PUNCT
bracis-33603	250	43	f.	f.	PROPN
bracis-33603	250	44	:	:	PUNCT
bracis-33603	250	45	multilingual	multilingual	ADJ
bracis-33603	250	46	e5	e5	NOUN
bracis-33603	250	47	text	text	NOUN
bracis-33603	250	48	embeddings	embedding	NOUN
bracis-33603	250	49	:	:	PUNCT
bracis-33603	250	50	a	a	DET
bracis-33603	250	51	technical	technical	ADJ
bracis-33603	250	52	report	report	NOUN
bracis-33603	250	53	(	(	PUNCT
bracis-33603	250	54	2024	2024	NUM
bracis-33603	250	55	)	)	PUNCT
bracis-33603	250	56	google	google	PROPN
bracis-33603	250	57	scholar	scholar	NOUN
bracis-33603	250	58	  	  	SPACE
bracis-33603	250	59	xue	xue	PROPN
bracis-33603	250	60	,	,	PUNCT
bracis-33603	250	61	l.	l.	PROPN
bracis-33603	250	62	,	,	PUNCT
bracis-33603	250	63	et	et	PROPN
bracis-33603	250	64	 	 	SPACE
bracis-33603	250	65	al	al	PROPN
bracis-33603	250	66	.	.	PROPN
bracis-33603	250	67	:	:	PUNCT
bracis-33603	251	1	mt5	mt5	ADJ
bracis-33603	251	2	:	:	PUNCT
bracis-33603	251	3	a	a	DET
bracis-33603	251	4	massively	massively	ADV
bracis-33603	251	5	multilingual	multilingual	ADJ
bracis-33603	251	6	pre	pre	ADJ
bracis-33603	251	7	-	-	ADJ
bracis-33603	251	8	trained	train	VERB
bracis-33603	251	9	text	text	NOUN
bracis-33603	251	10	-	-	PUNCT
bracis-33603	251	11	to	to	ADP
bracis-33603	251	12	-	-	PUNCT
bracis-33603	251	13	text	text	NOUN
bracis-33603	251	14	transformer	transformer	NOUN
bracis-33603	251	15	(	(	PUNCT
bracis-33603	251	16	2021	2021	NUM
bracis-33603	251	17	)	)	PUNCT
bracis-33603	251	18	google	google	NOUN
bracis-33603	251	19	scholar	scholar	NOUN
bracis-33603	251	20	  	  	SPACE
bracis-33603	251	21	download	download	NOUN
bracis-33603	251	22	references	reference	NOUN
bracis-33603	251	23	acknowledgements	acknowledgement	NOUN
bracis-33603	251	24	we	we	PRON
bracis-33603	251	25	thank	thank	VERB
bracis-33603	251	26	google	google	PROPN
bracis-33603	251	27	for	for	ADP
bracis-33603	251	28	the	the	DET
bracis-33603	251	29	tpu	tpu	PROPN
bracis-33603	251	30	grant	grant	NOUN
bracis-33603	251	31	through	through	ADP
bracis-33603	251	32	the	the	DET
bracis-33603	251	33	trc	trc	PROPN
bracis-33603	251	34	program	program	NOUN
bracis-33603	251	35	.	.	PUNCT
bracis-33603	252	1	r	r	NOUN
bracis-33603	252	2	lotufo	lotufo	NOUN
bracis-33603	252	3	is	be	AUX
bracis-33603	252	4	partially	partially	ADV
bracis-33603	252	5	supported	support	VERB
bracis-33603	252	6	by	by	ADP
bracis-33603	252	7	cnpq	cnpq	NOUN
bracis-33603	252	8	(	(	PUNCT
bracis-33603	252	9	the	the	DET
bracis-33603	252	10	brazilian	brazilian	ADJ
bracis-33603	252	11	national	national	PROPN
bracis-33603	252	12	council	council	PROPN
bracis-33603	252	13	for	for	ADP
bracis-33603	252	14	scientific	scientific	ADJ
bracis-33603	252	15	and	and	CCONJ
bracis-33603	252	16	technological	technological	ADJ
bracis-33603	252	17	development	development	NOUN
bracis-33603	252	18	)	)	PUNCT
bracis-33603	252	19	under	under	ADP
bracis-33603	252	20	grant	grant	NOUN
bracis-33603	252	21	313047/2022	313047/2022	NUM
bracis-33603	252	22	-	-	SYM
bracis-33603	252	23	7	7	NUM
bracis-33603	252	24	.	.	PUNCT
bracis-33603	252	25	author	author	NOUN
bracis-33603	252	26	information	information	NOUN
bracis-33603	252	27	authors	author	NOUN
bracis-33603	252	28	and	and	CCONJ
bracis-33603	252	29	affiliations	affiliation	NOUN
bracis-33603	252	30	school	school	NOUN
bracis-33603	252	31	of	of	ADP
bracis-33603	252	32	electrical	electrical	ADJ
bracis-33603	252	33	and	and	CCONJ
bracis-33603	252	34	computing	compute	VERB
bracis-33603	252	35	engineering	engineering	NOUN
bracis-33603	252	36	,	,	PUNCT
bracis-33603	252	37	state	state	NOUN
bracis-33603	252	38	university	university	PROPN
bracis-33603	252	39	of	of	ADP
bracis-33603	252	40	campinas	campinas	PROPN
bracis-33603	252	41	(	(	PUNCT
bracis-33603	252	42	unicamp	unicamp	PROPN
bracis-33603	252	43	)	)	PUNCT
bracis-33603	252	44	,	,	PUNCT
bracis-33603	252	45	campinas	campinas	PROPN
bracis-33603	252	46	,	,	PUNCT
bracis-33603	252	47	brazil	brazil	PROPN
bracis-33603	252	48	marcos	marcos	PROPN
bracis-33603	252	49	piau	piau	PROPN
bracis-33603	252	50	,	,	PUNCT
bracis-33603	252	51	 	 	SPACE
bracis-33603	252	52	roberto	roberto	PROPN
bracis-33603	252	53	lotufo	lotufo	PROPN
bracis-33603	252	54	 	 	SPACE
bracis-33603	252	55	&	&	CCONJ
bracis-33603	252	56	 	 	SPACE
bracis-33603	252	57	rodrigo	rodrigo	PROPN
bracis-33603	252	58	nogueira	nogueira	PROPN
bracis-33603	252	59	neuralmind	neuralmind	PROPN
bracis-33603	252	60	,	,	PUNCT
bracis-33603	252	61	campinas	campinas	PROPN
bracis-33603	252	62	,	,	PUNCT
bracis-33603	252	63	brazil	brazil	PROPN
bracis-33603	252	64	roberto	roberto	PROPN
bracis-33603	252	65	lotufo	lotufo	PROPN
bracis-33603	252	66	maritaca	maritaca	PROPN
bracis-33603	252	67	ai	ai	PROPN
bracis-33603	252	68	,	,	PUNCT
bracis-33603	252	69	campinas	campinas	PROPN
bracis-33603	252	70	,	,	PUNCT
bracis-33603	252	71	brazil	brazil	PROPN
bracis-33603	252	72	rodrigo	rodrigo	PROPN
bracis-33603	252	73	nogueira	nogueira	PROPN
bracis-33603	252	74	authors	authors	PROPN
bracis-33603	252	75	marcos	marcos	PROPN
bracis-33603	252	76	piauview	piauview	PROPN
bracis-33603	252	77	author	author	NOUN
bracis-33603	252	78	publications	publication	NOUN
bracis-33603	252	79	search	search	NOUN
bracis-33603	252	80	author	author	NOUN
bracis-33603	252	81	on	on	ADP
bracis-33603	252	82	:	:	PUNCT
bracis-33603	252	83	pubmed	pubmed	PROPN
bracis-33603	252	84	 	 	SPACE
bracis-33603	252	85	google	google	PROPN
bracis-33603	252	86	scholar	scholar	PROPN
bracis-33603	252	87	roberto	roberto	PROPN
bracis-33603	252	88	lotufoview	lotufoview	PROPN
bracis-33603	252	89	author	author	NOUN
bracis-33603	252	90	publications	publication	NOUN
bracis-33603	252	91	search	search	NOUN
bracis-33603	252	92	author	author	NOUN
bracis-33603	252	93	on	on	ADP
bracis-33603	252	94	:	:	PUNCT
bracis-33603	252	95	pubmed	pubmed	PROPN
bracis-33603	252	96	 	 	SPACE
bracis-33603	252	97	google	google	PROPN
bracis-33603	252	98	scholar	scholar	PROPN
bracis-33603	252	99	rodrigo	rodrigo	PROPN
bracis-33603	252	100	nogueiraview	nogueiraview	PROPN
bracis-33603	252	101	author	author	NOUN
bracis-33603	252	102	publications	publication	VERB
bracis-33603	252	103	search	search	NOUN
bracis-33603	252	104	author	author	NOUN
bracis-33603	252	105	on	on	ADP
bracis-33603	252	106	:	:	PUNCT
bracis-33603	252	107	pubmed	pubmed	PROPN
bracis-33603	252	108	 	 	SPACE
bracis-33603	252	109	google	google	PROPN
bracis-33603	252	110	scholar	scholar	NOUN
bracis-33603	252	111	corresponding	correspond	VERB
bracis-33603	252	112	author	author	NOUN
bracis-33603	252	113	correspondence	correspondence	NOUN
bracis-33603	252	114	to	to	ADP
bracis-33603	252	115	marcos	marcos	PROPN
bracis-33603	252	116	piau	piau	PROPN
bracis-33603	252	117	.	.	PUNCT
bracis-33603	253	1	editor	editor	NOUN
bracis-33603	253	2	information	information	NOUN
bracis-33603	253	3	editors	editor	NOUN
bracis-33603	253	4	and	and	CCONJ
bracis-33603	253	5	affiliations	affiliation	NOUN
bracis-33603	253	6	universidade	universidade	VERB
bracis-33603	253	7	federal	federal	PROPN
bracis-33603	253	8	fluminense	fluminense	PROPN
bracis-33603	253	9	,	,	PUNCT
bracis-33603	253	10	niterói	niterói	PROPN
bracis-33603	253	11	,	,	PUNCT
bracis-33603	253	12	brazil	brazil	PROPN
bracis-33603	253	13	aline	aline	PROPN
bracis-33603	253	14	paes	paes	PROPN
bracis-33603	253	15	instituto	instituto	PROPN
bracis-33603	253	16	tecnológico	tecnológico	PROPN
bracis-33603	253	17	de	de	PROPN
bracis-33603	253	18	aeronáutica	aeronáutica	PROPN
bracis-33603	253	19	,	,	PUNCT
bracis-33603	253	20	são	são	PROPN
bracis-33603	253	21	josé	josé	PROPN
bracis-33603	253	22	dos	dos	PROPN
bracis-33603	253	23	campos	campos	PROPN
bracis-33603	253	24	,	,	PUNCT
bracis-33603	253	25	brazil	brazil	PROPN
bracis-33603	253	26	filipe	filipe	PROPN
bracis-33603	253	27	a.	a.	PROPN
bracis-33603	253	28	n.	n.	PROPN
bracis-33603	253	29	verri	verri	PROPN
bracis-33603	253	30	rights	right	NOUN
bracis-33603	253	31	and	and	CCONJ
bracis-33603	253	32	permissions	permission	NOUN
bracis-33603	253	33	reprints	reprint	NOUN
bracis-33603	253	34	and	and	CCONJ
bracis-33603	253	35	permissions	permission	VERB
bracis-33603	253	36	copyright	copyright	NOUN
bracis-33603	253	37	information	information	NOUN
bracis-33603	253	38	©	©	ADP
bracis-33603	253	39	2025	2025	NUM
bracis-33603	253	40	the	the	DET
bracis-33603	253	41	author(s	author(s	NOUN
bracis-33603	253	42	)	)	PUNCT
bracis-33603	253	43	,	,	PUNCT
bracis-33603	253	44	under	under	ADP
bracis-33603	253	45	exclusive	exclusive	ADJ
bracis-33603	253	46	license	license	NOUN
bracis-33603	253	47	to	to	ADP
bracis-33603	253	48	springer	springer	NOUN
bracis-33603	253	49	nature	nature	PROPN
bracis-33603	253	50	switzerland	switzerland	PROPN
bracis-33603	253	51	ag	ag	PROPN
bracis-33603	253	52	about	about	ADP
bracis-33603	253	53	this	this	DET
bracis-33603	253	54	paper	paper	NOUN
bracis-33603	253	55	cite	cite	VERB
bracis-33603	253	56	this	this	DET
bracis-33603	253	57	paper	paper	NOUN
bracis-33603	253	58	piau	piau	NOUN
bracis-33603	253	59	,	,	PUNCT
bracis-33603	253	60	m.	m.	NOUN
bracis-33603	253	61	,	,	PUNCT
bracis-33603	253	62	lotufo	lotufo	PROPN
bracis-33603	253	63	,	,	PUNCT
bracis-33603	253	64	r.	r.	PROPN
bracis-33603	253	65	,	,	PUNCT
bracis-33603	253	66	nogueira	nogueira	PROPN
bracis-33603	253	67	,	,	PUNCT
bracis-33603	253	68	r.	r.	PROPN
bracis-33603	253	69	(	(	PUNCT
bracis-33603	253	70	2025	2025	NUM
bracis-33603	253	71	)	)	PUNCT
bracis-33603	253	72	.	.	PUNCT
bracis-33603	254	1	ptt5	ptt5	PROPN
bracis-33603	254	2	-	-	PUNCT
bracis-33603	254	3	v2	v2	PROPN
bracis-33603	254	4	:	:	PUNCT
bracis-33603	254	5	a	a	DET
bracis-33603	254	6	closer	close	ADJ
bracis-33603	254	7	look	look	NOUN
bracis-33603	254	8	at	at	ADP
bracis-33603	254	9	 	 	SPACE
bracis-33603	254	10	continued	continue	VERB
bracis-33603	254	11	pretraining	pretraine	VERB
bracis-33603	254	12	of	of	ADP
bracis-33603	254	13	 	 	SPACE
bracis-33603	254	14	t5	t5	PROPN
bracis-33603	254	15	models	model	NOUN
bracis-33603	254	16	for	for	ADP
bracis-33603	254	17	 	 	SPACE
bracis-33603	254	18	the	the	DET
bracis-33603	254	19	 	 	SPACE
bracis-33603	254	20	portuguese	portuguese	ADJ
bracis-33603	254	21	language	language	NOUN
bracis-33603	254	22	.	.	PUNCT
bracis-33603	255	1	in	in	ADP
bracis-33603	255	2	:	:	PUNCT
bracis-33603	255	3	paes	paes	PROPN
bracis-33603	255	4	,	,	PUNCT
bracis-33603	255	5	a.	a.	PROPN
bracis-33603	255	6	,	,	PUNCT
bracis-33603	255	7	verri	verri	NOUN
bracis-33603	255	8	,	,	PUNCT
bracis-33603	255	9	f.a.n	f.a.n	ADJ
bracis-33603	255	10	.	.	PUNCT
bracis-33603	255	11	(	(	PUNCT
bracis-33603	255	12	eds	ed	NOUN
bracis-33603	255	13	)	)	PUNCT
bracis-33603	255	14	intelligent	intelligent	ADJ
bracis-33603	255	15	systems	system	NOUN
bracis-33603	255	16	.	.	PUNCT
bracis-33603	255	17	bracis	bracis	PROPN
bracis-33603	255	18	2024	2024	NUM
bracis-33603	255	19	.	.	PUNCT
bracis-33603	256	1	lecture	lecture	NOUN
bracis-33603	256	2	notes	note	NOUN
bracis-33603	256	3	in	in	ADP
bracis-33603	256	4	computer	computer	NOUN
bracis-33603	256	5	science	science	NOUN
bracis-33603	256	6	(	(	PUNCT
bracis-33603	256	7	)	)	PUNCT
bracis-33603	256	8	,	,	PUNCT
bracis-33603	256	9	vol	vol	NOUN
bracis-33603	256	10	15413	15413	NUM
bracis-33603	256	11	.	.	PUNCT
bracis-33603	257	1	springer	springer	NOUN
bracis-33603	257	2	,	,	PUNCT
bracis-33603	257	3	cham	cham	PROPN
bracis-33603	257	4	.	.	PUNCT
bracis-33603	258	1	https://doi.org/10.1007/978-3-031-79032-4_23	https://doi.org/10.1007/978-3-031-79032-4_23	PROPN
bracis-33603	258	2	download	download	NOUN
bracis-33603	258	3	citation	citation	NOUN
bracis-33603	258	4	.ris	.ris	PUNCT
bracis-33603	259	1	.enw	.enw	PROPN
bracis-33603	259	2	.bib	.bib	PUNCT
bracis-33603	260	1	doi	doi	PROPN
bracis-33603	260	2	:	:	PUNCT
bracis-33603	261	1	https://doi.org/10.1007/978-3-031-79032-4_23	https://doi.org/10.1007/978-3-031-79032-4_23	PROPN
bracis-33603	261	2	published	publish	VERB
bracis-33603	261	3	:	:	PUNCT
bracis-33603	261	4	30	30	NUM
bracis-33603	261	5	january	january	PROPN
bracis-33603	261	6	2025	2025	NUM
bracis-33603	261	7	publisher	publisher	NOUN
bracis-33603	261	8	name	name	NOUN
bracis-33603	261	9	:	:	PUNCT
bracis-33603	261	10	springer	springer	NOUN
bracis-33603	261	11	,	,	PUNCT
bracis-33603	261	12	cham	cham	PROPN
bracis-33603	261	13	print	print	PROPN
bracis-33603	261	14	isbn	isbn	PROPN
bracis-33603	261	15	:	:	PUNCT
bracis-33603	261	16	978	978	NUM
bracis-33603	261	17	-	-	SYM
bracis-33603	261	18	3	3	NUM
bracis-33603	261	19	-	-	PUNCT
bracis-33603	261	20	031	031	NUM
bracis-33603	261	21	-	-	PUNCT
bracis-33603	261	22	79031	79031	NUM
bracis-33603	261	23	-	-	SYM
bracis-33603	261	24	7	7	NUM
bracis-33603	261	25	online	online	ADJ
bracis-33603	261	26	isbn	isbn	NOUN
bracis-33603	261	27	:	:	PUNCT
bracis-33603	261	28	978	978	NUM
bracis-33603	261	29	-	-	SYM
bracis-33603	261	30	3	3	NUM
bracis-33603	261	31	-	-	PUNCT
bracis-33603	261	32	031	031	NUM
bracis-33603	261	33	-	-	PUNCT
bracis-33603	261	34	79032	79032	NUM
bracis-33603	261	35	-	-	SYM
bracis-33603	261	36	4	4	NUM
bracis-33603	261	37	ebook	ebook	NOUN
bracis-33603	261	38	packages	package	NOUN
bracis-33603	261	39	:	:	PUNCT
bracis-33603	261	40	computer	computer	NOUN
bracis-33603	261	41	sciencecomputer	sciencecomputer	NOUN
bracis-33603	261	42	science	science	NOUN
bracis-33603	261	43	(	(	PUNCT
bracis-33603	261	44	r0	r0	NOUN
bracis-33603	261	45	)	)	PUNCT
bracis-33603	261	46	share	share	VERB
bracis-33603	261	47	this	this	DET
bracis-33603	261	48	paper	paper	NOUN
bracis-33603	261	49	anyone	anyone	PRON
bracis-33603	261	50	you	you	PRON
bracis-33603	261	51	share	share	VERB
bracis-33603	261	52	the	the	DET
bracis-33603	261	53	following	follow	VERB
bracis-33603	261	54	link	link	NOUN
bracis-33603	261	55	with	with	ADP
bracis-33603	261	56	will	will	AUX
bracis-33603	261	57	be	be	AUX
bracis-33603	261	58	able	able	ADJ
bracis-33603	261	59	to	to	PART
bracis-33603	261	60	read	read	VERB
bracis-33603	261	61	this	this	DET
bracis-33603	261	62	content	content	NOUN
bracis-33603	261	63	:	:	PUNCT
bracis-33603	261	64	get	get	VERB
bracis-33603	261	65	shareable	shareable	ADJ
bracis-33603	261	66	linksorry	linksorry	NOUN
bracis-33603	261	67	,	,	PUNCT
bracis-33603	261	68	a	a	DET
bracis-33603	261	69	shareable	shareable	ADJ
bracis-33603	261	70	link	link	NOUN
bracis-33603	261	71	is	be	AUX
bracis-33603	261	72	not	not	PART
bracis-33603	261	73	currently	currently	ADV
bracis-33603	261	74	available	available	ADJ
bracis-33603	261	75	for	for	ADP
bracis-33603	261	76	this	this	DET
bracis-33603	261	77	article	article	NOUN
bracis-33603	261	78	.	.	PUNCT
bracis-33603	262	1	copy	copy	VERB
bracis-33603	262	2	shareable	shareable	ADJ
bracis-33603	262	3	link	link	NOUN
bracis-33603	262	4	to	to	PART
bracis-33603	262	5	clipboard	clipboard	NOUN
bracis-33603	262	6	provided	provide	VERB
bracis-33603	262	7	by	by	ADP
bracis-33603	262	8	the	the	DET
bracis-33603	262	9	springer	springer	NOUN
bracis-33603	262	10	nature	nature	PROPN
bracis-33603	262	11	sharedit	sharedit	PROPN
bracis-33603	262	12	content	content	NOUN
bracis-33603	262	13	-	-	PUNCT
bracis-33603	262	14	sharing	share	VERB
bracis-33603	262	15	initiative	initiative	NOUN
bracis-33603	262	16	keywords	keyword	NOUN
bracis-33603	262	17	portuguese	portuguese	PROPN
bracis-33603	262	18	t5	t5	PROPN
bracis-33603	262	19	continued	continue	VERB
bracis-33603	262	20	pretraining	pretraine	VERB
bracis-33603	262	21	information	information	NOUN
bracis-33603	262	22	retrieval	retrieval	NOUN
bracis-33603	262	23	publish	publish	NOUN
bracis-33603	262	24	with	with	ADP
bracis-33603	262	25	us	us	PROPN
bracis-33603	262	26	policies	policy	NOUN
bracis-33603	262	27	and	and	CCONJ
bracis-33603	262	28	ethics	ethic	NOUN
bracis-33603	262	29	search	search	NOUN
bracis-33603	262	30	search	search	NOUN
bracis-33603	262	31	by	by	ADP
bracis-33603	262	32	keyword	keyword	NOUN
bracis-33603	262	33	or	or	CCONJ
bracis-33603	262	34	author	author	NOUN
bracis-33603	262	35	search	search	NOUN
bracis-33603	262	36	navigation	navigation	NOUN
bracis-33603	262	37	find	find	VERB
bracis-33603	262	38	a	a	DET
bracis-33603	262	39	journal	journal	NOUN
bracis-33603	262	40	publish	publish	VERB
bracis-33603	262	41	with	with	ADP
bracis-33603	262	42	us	we	PRON
bracis-33603	262	43	track	track	VERB
bracis-33603	262	44	your	your	PRON
bracis-33603	262	45	research	research	NOUN
bracis-33603	262	46	discover	discover	VERB
bracis-33603	262	47	content	content	NOUN
bracis-33603	262	48	journals	journal	NOUN
bracis-33603	262	49	a	a	DET
bracis-33603	262	50	-	-	PUNCT
bracis-33603	262	51	z	z	NOUN
bracis-33603	262	52	books	book	NOUN
bracis-33603	262	53	a	a	DET
bracis-33603	262	54	-	-	PUNCT
bracis-33603	262	55	z	z	NOUN
bracis-33603	262	56	publish	publish	NOUN
bracis-33603	262	57	with	with	ADP
bracis-33603	262	58	us	us	PROPN
bracis-33603	262	59	journal	journal	PROPN
bracis-33603	262	60	finder	finder	PROPN
bracis-33603	262	61	publish	publish	VERB
bracis-33603	262	62	your	your	PRON
bracis-33603	262	63	research	research	NOUN
bracis-33603	262	64	language	language	NOUN
bracis-33603	262	65	editing	edit	VERB
bracis-33603	262	66	open	open	ADJ
bracis-33603	262	67	access	access	NOUN
bracis-33603	262	68	publishing	publishing	NOUN
bracis-33603	262	69	products	product	NOUN
bracis-33603	262	70	and	and	CCONJ
bracis-33603	262	71	services	service	NOUN
bracis-33603	262	72	our	our	PRON
bracis-33603	262	73	products	product	NOUN
bracis-33603	262	74	librarians	librarian	VERB
bracis-33603	262	75	societies	society	NOUN
bracis-33603	262	76	partners	partner	NOUN
bracis-33603	262	77	and	and	CCONJ
bracis-33603	262	78	advertisers	advertiser	NOUN
bracis-33603	262	79	our	our	PRON
bracis-33603	262	80	brands	brand	NOUN
bracis-33603	262	81	springer	springer	NOUN
bracis-33603	262	82	nature	nature	PROPN
bracis-33603	262	83	portfolio	portfolio	PROPN
bracis-33603	262	84	bmc	bmc	PROPN
bracis-33603	262	85	palgrave	palgrave	PROPN
bracis-33603	262	86	macmillan	macmillan	PROPN
bracis-33603	262	87	apress	apress	PROPN
bracis-33603	262	88	discover	discover	VERB
bracis-33603	262	89	your	your	PRON
bracis-33603	262	90	privacy	privacy	NOUN
bracis-33603	262	91	choices	choice	NOUN
bracis-33603	262	92	/	/	SYM
bracis-33603	262	93	manage	manage	NOUN
bracis-33603	262	94	cookies	cookie	NOUN
bracis-33603	262	95	your	your	PRON
bracis-33603	262	96	us	us	PROPN
bracis-33603	263	1	state	state	NOUN
bracis-33603	263	2	privacy	privacy	NOUN
bracis-33603	263	3	rights	right	NOUN
bracis-33603	263	4	accessibility	accessibility	NOUN
bracis-33603	263	5	statement	statement	NOUN
bracis-33603	263	6	terms	term	NOUN
bracis-33603	263	7	and	and	CCONJ
bracis-33603	263	8	conditions	condition	NOUN
bracis-33603	263	9	privacy	privacy	NOUN
bracis-33603	263	10	policy	policy	NOUN
bracis-33603	263	11	help	help	NOUN
bracis-33603	263	12	and	and	CCONJ
bracis-33603	263	13	support	support	VERB
bracis-33603	263	14	legal	legal	ADJ
bracis-33603	263	15	notice	notice	NOUN
bracis-33603	263	16	cancel	cancel	VERB
bracis-33603	263	17	contracts	contract	NOUN
bracis-33603	263	18	here	here	ADV
bracis-33603	263	19	129.74.145.123	129.74.145.123	NUM
bracis-33603	263	20	hesburgh	hesburgh	PROPN
bracis-33603	263	21	library	library	PROPN
bracis-33603	263	22	er	er	INTJ
bracis-33603	263	23	unit	unit	NOUN
bracis-33603	263	24	(	(	PUNCT
bracis-33603	263	25	3005732405	3005732405	NUM
bracis-33603	263	26	)	)	PUNCT
bracis-33603	263	27	northeast	northeast	ADJ
bracis-33603	263	28	research	research	NOUN
bracis-33603	263	29	libraries	library	NOUN
bracis-33603	263	30	(	(	PUNCT
bracis-33603	263	31	nerl	nerl	PROPN
bracis-33603	263	32	)	)	PUNCT
bracis-33603	263	33	(	(	PUNCT
bracis-33603	263	34	8200828607	8200828607	NUM
bracis-33603	263	35	)	)	PUNCT
bracis-33603	263	36	nerl	nerl	VERB
bracis-33603	263	37	ta	ta	X
bracis-33603	263	38	account	account	NOUN
bracis-33603	263	39	(	(	PUNCT
bracis-33603	263	40	3006206169	3006206169	NUM
bracis-33603	263	41	)	)	PUNCT
bracis-33603	263	42	university	university	NOUN
bracis-33603	263	43	of	of	ADP
bracis-33603	263	44	notre	notre	PROPN
bracis-33603	263	45	dame	dame	PROPN
bracis-33603	263	46	hesburgh	hesburgh	PROPN
bracis-33603	263	47	library	library	NOUN
bracis-33603	263	48	(	(	PUNCT
bracis-33603	263	49	3000184373	3000184373	NUM
bracis-33603	263	50	)	)	PUNCT
bracis-33603	264	1	©	©	ADP
bracis-33603	264	2	2025	2025	NUM
bracis-33603	264	3	springer	springer	NOUN
bracis-33603	264	4	nature	nature	NOUN