id	sid	tid	token	lemma	pos
bracis-33663	1	1	tuning	tune	VERB
bracis-33663	1	2	hypothesis	hypothesis	NOUN
bracis-33663	1	3	creation	creation	NOUN
bracis-33663	1	4	:	:	PUNCT
bracis-33663	1	5	combining	combine	VERB
bracis-33663	1	6	discrete	discrete	ADJ
bracis-33663	1	7	and	and	CCONJ
bracis-33663	1	8	 	 	SPACE
bracis-33663	1	9	continuous	continuous	ADJ
bracis-33663	1	10	spaces	space	NOUN
bracis-33663	1	11	for	for	ADP
bracis-33663	1	12	 	 	SPACE
bracis-33663	1	13	zero	zero	NUM
bracis-33663	1	14	-	-	PUNCT
bracis-33663	1	15	shot	shoot	VERB
bracis-33663	1	16	hate	hate	NOUN
bracis-33663	1	17	speech	speech	NOUN
bracis-33663	1	18	detection	detection	NOUN
bracis-33663	1	19	|	|	ADV
bracis-33663	1	20	springer	springer	NOUN
bracis-33663	1	21	nature	nature	PROPN
bracis-33663	1	22	link	link	PROPN
bracis-33663	1	23	(	(	PUNCT
bracis-33663	1	24	formerly	formerly	ADV
bracis-33663	1	25	springerlink	springerlink	NOUN
bracis-33663	1	26	)	)	PUNCT
bracis-33663	1	27	skip	skip	VERB
bracis-33663	1	28	to	to	ADP
bracis-33663	1	29	main	main	ADJ
bracis-33663	1	30	content	content	NOUN
bracis-33663	1	31	advertisement	advertisement	NOUN
bracis-33663	1	32	log	log	NOUN
bracis-33663	1	33	in	in	ADP
bracis-33663	1	34	menu	menu	NOUN
bracis-33663	1	35	find	find	VERB
bracis-33663	1	36	a	a	DET
bracis-33663	1	37	journal	journal	NOUN
bracis-33663	1	38	publish	publish	VERB
bracis-33663	1	39	with	with	ADP
bracis-33663	1	40	us	we	PRON
bracis-33663	1	41	track	track	VERB
bracis-33663	1	42	your	your	PRON
bracis-33663	1	43	research	research	NOUN
bracis-33663	1	44	search	search	NOUN
bracis-33663	1	45	cart	cart	NOUN
bracis-33663	1	46	home	home	NOUN
bracis-33663	1	47	intelligent	intelligent	ADJ
bracis-33663	1	48	systems	system	NOUN
bracis-33663	1	49	conference	conference	NOUN
bracis-33663	1	50	paper	paper	NOUN
bracis-33663	1	51	tuning	tuning	NOUN
bracis-33663	1	52	hypothesis	hypothesis	NOUN
bracis-33663	1	53	creation	creation	NOUN
bracis-33663	1	54	:	:	PUNCT
bracis-33663	1	55	combining	combine	VERB
bracis-33663	1	56	discrete	discrete	ADJ
bracis-33663	1	57	and	and	CCONJ
bracis-33663	1	58	 	 	SPACE
bracis-33663	1	59	continuous	continuous	ADJ
bracis-33663	1	60	spaces	space	NOUN
bracis-33663	1	61	for	for	ADP
bracis-33663	1	62	 	 	SPACE
bracis-33663	1	63	zero	zero	NUM
bracis-33663	1	64	-	-	PUNCT
bracis-33663	1	65	shot	shoot	VERB
bracis-33663	1	66	hate	hate	NOUN
bracis-33663	1	67	speech	speech	NOUN
bracis-33663	1	68	detection	detection	NOUN
bracis-33663	1	69	conference	conference	NOUN
bracis-33663	1	70	paper	paper	NOUN
bracis-33663	1	71	first	first	ADV
bracis-33663	1	72	online	online	ADV
bracis-33663	1	73	:	:	PUNCT
bracis-33663	1	74	31	31	NUM
bracis-33663	1	75	january	january	NOUN
bracis-33663	1	76	2025	2025	NUM
bracis-33663	1	77	pp	pp	PROPN
bracis-33663	1	78	253–268	253–268	NUM
bracis-33663	1	79	cite	cite	VERB
bracis-33663	1	80	this	this	DET
bracis-33663	1	81	conference	conference	NOUN
bracis-33663	1	82	paper	paper	NOUN
bracis-33663	1	83	access	access	NOUN
bracis-33663	1	84	provided	provide	VERB
bracis-33663	1	85	by	by	ADP
bracis-33663	1	86	university	university	PROPN
bracis-33663	1	87	of	of	ADP
bracis-33663	1	88	notre	notre	PROPN
bracis-33663	1	89	dame	dame	PROPN
bracis-33663	1	90	hesburgh	hesburgh	PROPN
bracis-33663	1	91	library	library	PROPN
bracis-33663	1	92	download	download	PROPN
bracis-33663	1	93	book	book	NOUN
bracis-33663	1	94	pdf	pdf	PROPN
bracis-33663	1	95	download	download	NOUN
bracis-33663	1	96	book	book	NOUN
bracis-33663	1	97	epub	epub	PROPN
bracis-33663	1	98	intelligent	intelligent	ADJ
bracis-33663	1	99	systems	system	NOUN
bracis-33663	1	100	(	(	PUNCT
bracis-33663	1	101	bracis	bracis	NOUN
bracis-33663	1	102	2024	2024	NUM
bracis-33663	1	103	)	)	PUNCT
bracis-33663	1	104	tuning	tune	VERB
bracis-33663	1	105	hypothesis	hypothesis	NOUN
bracis-33663	1	106	creation	creation	NOUN
bracis-33663	1	107	:	:	PUNCT
bracis-33663	1	108	combining	combine	VERB
bracis-33663	1	109	discrete	discrete	ADJ
bracis-33663	1	110	and	and	CCONJ
bracis-33663	1	111	 	 	SPACE
bracis-33663	1	112	continuous	continuous	ADJ
bracis-33663	1	113	spaces	space	NOUN
bracis-33663	1	114	for	for	ADP
bracis-33663	1	115	 	 	SPACE
bracis-33663	1	116	zero	zero	NUM
bracis-33663	1	117	-	-	PUNCT
bracis-33663	1	118	shot	shoot	VERB
bracis-33663	1	119	hate	hate	NOUN
bracis-33663	1	120	speech	speech	NOUN
bracis-33663	1	121	detection	detection	NOUN
bracis-33663	1	122	download	download	NOUN
bracis-33663	1	123	book	book	NOUN
bracis-33663	1	124	pdf	pdf	PROPN
bracis-33663	1	125	download	download	NOUN
bracis-33663	1	126	book	book	PROPN
bracis-33663	1	127	epub	epub	PROPN
bracis-33663	1	128	lorenzo	lorenzo	PROPN
bracis-33663	1	129	puppi	puppi	PROPN
bracis-33663	1	130	vecchi	vecchi	PROPN
bracis-33663	1	131	  	  	SPACE
bracis-33663	1	132	orcid	orcid	NOUN
bracis-33663	1	133	:	:	PUNCT
bracis-33663	1	134	orcid.org/0009-0003-9483-20269	orcid.org/0009-0003-9483-20269	ADV
bracis-33663	1	135	,	,	PUNCT
bracis-33663	1	136	sylvio	sylvio	PROPN
bracis-33663	1	137	barbon	barbon	PROPN
bracis-33663	1	138	junior	junior	PROPN
bracis-33663	1	139	  	  	SPACE
bracis-33663	1	140	orcid	orcid	NOUN
bracis-33663	1	141	:	:	PUNCT
bracis-33663	1	142	orcid.org/0000-0002-4988-070210	orcid.org/0000-0002-4988-070210	PROPN
bracis-33663	1	143	&	&	CCONJ
bracis-33663	1	144	emerson	emerson	PROPN
bracis-33663	1	145	cabrera	cabrera	PROPN
bracis-33663	1	146	paraiso	paraiso	PROPN
bracis-33663	1	147	  	  	SPACE
bracis-33663	1	148	orcid	orcid	NOUN
bracis-33663	1	149	:	:	PUNCT
bracis-33663	1	150	orcid.org/0000-0002-6740-78559	orcid.org/0000-0002-6740-78559	ADJ
bracis-33663	1	151	  	  	SPACE
bracis-33663	1	152	part	part	NOUN
bracis-33663	1	153	of	of	ADP
bracis-33663	1	154	the	the	DET
bracis-33663	1	155	book	book	NOUN
bracis-33663	1	156	series	series	NOUN
bracis-33663	1	157	:	:	PUNCT
bracis-33663	1	158	lecture	lecture	NOUN
bracis-33663	1	159	notes	note	NOUN
bracis-33663	1	160	in	in	ADP
bracis-33663	1	161	computer	computer	NOUN
bracis-33663	1	162	science	science	NOUN
bracis-33663	1	163	(	(	PUNCT
bracis-33663	1	164	(	(	PUNCT
bracis-33663	1	165	lnai	lnai	ADJ
bracis-33663	1	166	,	,	PUNCT
bracis-33663	1	167	volume	volume	NOUN
bracis-33663	1	168	15415	15415	NUM
bracis-33663	1	169	)	)	PUNCT
bracis-33663	1	170	)	)	PUNCT
bracis-33663	1	171	included	include	VERB
bracis-33663	1	172	in	in	ADP
bracis-33663	1	173	the	the	DET
bracis-33663	1	174	following	follow	VERB
bracis-33663	1	175	conference	conference	NOUN
bracis-33663	1	176	series	series	NOUN
bracis-33663	1	177	:	:	PUNCT
bracis-33663	1	178	brazilian	brazilian	ADJ
bracis-33663	1	179	conference	conference	NOUN
bracis-33663	1	180	on	on	ADP
bracis-33663	1	181	intelligent	intelligent	ADJ
bracis-33663	1	182	systems	system	NOUN
bracis-33663	1	183	219	219	NUM
bracis-33663	1	184	accesses	access	NOUN
bracis-33663	1	185	abstract	abstract	ADJ
bracis-33663	1	186	hate	hate	NOUN
bracis-33663	1	187	speech	speech	NOUN
bracis-33663	1	188	detection	detection	NOUN
bracis-33663	1	189	is	be	AUX
bracis-33663	1	190	a	a	DET
bracis-33663	1	191	crucial	crucial	ADJ
bracis-33663	1	192	endeavor	endeavor	NOUN
bracis-33663	1	193	in	in	ADP
bracis-33663	1	194	maintaining	maintain	VERB
bracis-33663	1	195	the	the	DET
bracis-33663	1	196	safety	safety	NOUN
bracis-33663	1	197	of	of	ADP
bracis-33663	1	198	online	online	ADJ
bracis-33663	1	199	spaces	space	NOUN
bracis-33663	1	200	,	,	PUNCT
bracis-33663	1	201	but	but	CCONJ
bracis-33663	1	202	the	the	DET
bracis-33663	1	203	effectiveness	effectiveness	NOUN
bracis-33663	1	204	of	of	ADP
bracis-33663	1	205	supervised	supervised	ADJ
bracis-33663	1	206	approaches	approach	NOUN
bracis-33663	1	207	hinges	hinge	NOUN
bracis-33663	1	208	mainly	mainly	ADV
bracis-33663	1	209	on	on	ADP
bracis-33663	1	210	the	the	DET
bracis-33663	1	211	availability	availability	NOUN
bracis-33663	1	212	of	of	ADP
bracis-33663	1	213	annotated	annotate	VERB
bracis-33663	1	214	data	datum	NOUN
bracis-33663	1	215	.	.	PUNCT
bracis-33663	2	1	prior	prior	ADJ
bracis-33663	2	2	research	research	NOUN
bracis-33663	2	3	has	have	AUX
bracis-33663	2	4	explored	explore	VERB
bracis-33663	2	5	the	the	DET
bracis-33663	2	6	utility	utility	NOUN
bracis-33663	2	7	of	of	ADP
bracis-33663	2	8	natural	natural	ADJ
bracis-33663	2	9	language	language	NOUN
bracis-33663	2	10	inference	inference	NOUN
bracis-33663	2	11	(	(	PUNCT
bracis-33663	2	12	nli	nli	NOUN
bracis-33663	2	13	)	)	PUNCT
bracis-33663	2	14	models	model	NOUN
bracis-33663	2	15	for	for	ADP
bracis-33663	2	16	zero	zero	NUM
bracis-33663	2	17	-	-	PUNCT
bracis-33663	2	18	shot	shot	NOUN
bracis-33663	2	19	hate	hate	NOUN
bracis-33663	2	20	speech	speech	NOUN
bracis-33663	2	21	detection	detection	NOUN
bracis-33663	2	22	(	(	PUNCT
bracis-33663	2	23	zshsd	zshsd	PROPN
bracis-33663	2	24	)	)	PUNCT
bracis-33663	2	25	,	,	PUNCT
bracis-33663	2	26	which	which	PRON
bracis-33663	2	27	leverages	leverage	VERB
bracis-33663	2	28	the	the	DET
bracis-33663	2	29	capacity	capacity	NOUN
bracis-33663	2	30	of	of	ADP
bracis-33663	2	31	these	these	DET
bracis-33663	2	32	models	model	NOUN
bracis-33663	2	33	to	to	PART
bracis-33663	2	34	learn	learn	VERB
bracis-33663	2	35	semantic	semantic	ADJ
bracis-33663	2	36	relationships	relationship	NOUN
bracis-33663	2	37	and	and	CCONJ
bracis-33663	2	38	adapt	adapt	VERB
bracis-33663	2	39	to	to	ADP
bracis-33663	2	40	downstream	downstream	ADJ
bracis-33663	2	41	task	task	NOUN
bracis-33663	2	42	without	without	ADP
bracis-33663	2	43	relying	rely	VERB
bracis-33663	2	44	on	on	ADP
bracis-33663	2	45	large	large	ADJ
bracis-33663	2	46	annotated	annotated	ADJ
bracis-33663	2	47	datasets	dataset	NOUN
bracis-33663	2	48	.	.	PUNCT
bracis-33663	3	1	nli	nli	PROPN
bracis-33663	3	2	models	model	NOUN
bracis-33663	3	3	assess	assess	VERB
bracis-33663	3	4	if	if	SCONJ
bracis-33663	3	5	a	a	DET
bracis-33663	3	6	premise	premise	ADJ
bracis-33663	3	7	sentence	sentence	NOUN
bracis-33663	3	8	logically	logically	ADV
bracis-33663	3	9	entails	entail	VERB
bracis-33663	3	10	a	a	DET
bracis-33663	3	11	hypothesis	hypothesis	NOUN
bracis-33663	3	12	sentence	sentence	NOUN
bracis-33663	3	13	,	,	PUNCT
bracis-33663	3	14	relying	rely	VERB
bracis-33663	3	15	on	on	ADP
bracis-33663	3	16	precise	precise	ADJ
bracis-33663	3	17	hypothesis	hypothesis	NOUN
bracis-33663	3	18	design	design	NOUN
bracis-33663	3	19	to	to	PART
bracis-33663	3	20	achieve	achieve	VERB
bracis-33663	3	21	adequate	adequate	ADJ
bracis-33663	3	22	downstream	downstream	ADJ
bracis-33663	3	23	task	task	NOUN
bracis-33663	3	24	performance	performance	NOUN
bracis-33663	3	25	.	.	PUNCT
bracis-33663	4	1	existing	exist	VERB
bracis-33663	4	2	frameworks	framework	NOUN
bracis-33663	4	3	that	that	PRON
bracis-33663	4	4	use	use	VERB
bracis-33663	4	5	nli	nli	ADJ
bracis-33663	4	6	model	model	NOUN
bracis-33663	4	7	for	for	ADP
bracis-33663	4	8	zshsd	zshsd	PROPN
bracis-33663	4	9	rely	rely	VERB
bracis-33663	4	10	on	on	ADP
bracis-33663	4	11	multiple	multiple	ADJ
bracis-33663	4	12	inferences	inference	NOUN
bracis-33663	4	13	with	with	ADP
bracis-33663	4	14	different	different	ADJ
bracis-33663	4	15	hypothesis	hypothesis	NOUN
bracis-33663	4	16	to	to	PART
bracis-33663	4	17	extract	extract	VERB
bracis-33663	4	18	characteristics	characteristic	NOUN
bracis-33663	4	19	to	to	PART
bracis-33663	4	20	achieve	achieve	VERB
bracis-33663	4	21	desirable	desirable	ADJ
bracis-33663	4	22	outcomes	outcome	NOUN
bracis-33663	4	23	.	.	PUNCT
bracis-33663	5	1	in	in	ADP
bracis-33663	5	2	light	light	NOUN
bracis-33663	5	3	of	of	ADP
bracis-33663	5	4	the	the	DET
bracis-33663	5	5	challenges	challenge	NOUN
bracis-33663	5	6	surrounding	surround	VERB
bracis-33663	5	7	zshsd	zshsd	NOUN
bracis-33663	5	8	and	and	CCONJ
bracis-33663	5	9	the	the	DET
bracis-33663	5	10	method	method	NOUN
bracis-33663	5	11	of	of	ADP
bracis-33663	5	12	relying	rely	VERB
bracis-33663	5	13	on	on	ADP
bracis-33663	5	14	discrete	discrete	ADJ
bracis-33663	5	15	tokens	token	NOUN
bracis-33663	5	16	to	to	PART
bracis-33663	5	17	design	design	VERB
bracis-33663	5	18	hypotheses	hypothesis	NOUN
bracis-33663	5	19	,	,	PUNCT
bracis-33663	5	20	we	we	PRON
bracis-33663	5	21	aim	aim	VERB
bracis-33663	5	22	to	to	PART
bracis-33663	5	23	optimize	optimize	VERB
bracis-33663	5	24	and	and	CCONJ
bracis-33663	5	25	identify	identify	VERB
bracis-33663	5	26	ideal	ideal	ADJ
bracis-33663	5	27	intermediate	intermediate	ADJ
bracis-33663	5	28	representations	representation	NOUN
bracis-33663	5	29	by	by	ADP
bracis-33663	5	30	applying	apply	VERB
bracis-33663	5	31	p	p	NOUN
bracis-33663	5	32	-	-	PUNCT
bracis-33663	5	33	tuning	tune	VERB
bracis-33663	5	34	techniques	technique	NOUN
bracis-33663	5	35	.	.	PUNCT
bracis-33663	6	1	on	on	ADP
bracis-33663	6	2	hatecheck	hatecheck	NOUN
bracis-33663	6	3	,	,	PUNCT
bracis-33663	6	4	a	a	DET
bracis-33663	6	5	fully	fully	ADV
bracis-33663	6	6	tuned	tune	VERB
bracis-33663	6	7	hypothesis	hypothesis	NOUN
bracis-33663	6	8	led	lead	VERB
bracis-33663	6	9	to	to	ADP
bracis-33663	6	10	a	a	DET
bracis-33663	6	11	18.8	18.8	NUM
bracis-33663	6	12	%	%	NOUN
bracis-33663	6	13	point	point	NOUN
bracis-33663	6	14	(	(	PUNCT
bracis-33663	6	15	pp	pp	ADJ
bracis-33663	6	16	)	)	PUNCT
bracis-33663	6	17	accuracy	accuracy	NOUN
bracis-33663	6	18	improvement	improvement	NOUN
bracis-33663	6	19	,	,	PUNCT
bracis-33663	6	20	compared	compare	VERB
bracis-33663	6	21	to	to	ADP
bracis-33663	6	22	a	a	DET
bracis-33663	6	23	discrete	discrete	ADJ
bracis-33663	6	24	designed	design	VERB
bracis-33663	6	25	hypothesis	hypothesis	NOUN
bracis-33663	6	26	.	.	PUNCT
bracis-33663	7	1	our	our	PRON
bracis-33663	7	2	work	work	NOUN
bracis-33663	7	3	surpassed	surpass	VERB
bracis-33663	7	4	prior	prior	ADJ
bracis-33663	7	5	work	work	NOUN
bracis-33663	7	6	by	by	ADP
bracis-33663	7	7	achieving	achieve	VERB
bracis-33663	7	8	a	a	DET
bracis-33663	7	9	5.6	5.6	NUM
bracis-33663	7	10	pp	pp	NOUN
bracis-33663	7	11	accuracy	accuracy	NOUN
bracis-33663	7	12	enhancement	enhancement	NOUN
bracis-33663	7	13	,	,	PUNCT
bracis-33663	7	14	outperforming	outperform	VERB
bracis-33663	7	15	previous	previous	ADJ
bracis-33663	7	16	approaches	approach	NOUN
bracis-33663	7	17	that	that	PRON
bracis-33663	7	18	required	require	VERB
bracis-33663	7	19	multiple	multiple	ADJ
bracis-33663	7	20	inferences	inference	NOUN
bracis-33663	7	21	.	.	PUNCT
bracis-33663	8	1	also	also	ADV
bracis-33663	8	2	,	,	PUNCT
bracis-33663	8	3	the	the	DET
bracis-33663	8	4	optimized	optimize	VERB
bracis-33663	8	5	tokens	token	NOUN
bracis-33663	8	6	uncover	uncover	VERB
bracis-33663	8	7	relations	relation	NOUN
bracis-33663	8	8	to	to	ADP
bracis-33663	8	9	broader	broad	ADJ
bracis-33663	8	10	aspects	aspect	NOUN
bracis-33663	8	11	of	of	ADP
bracis-33663	8	12	hate	hate	NOUN
bracis-33663	8	13	speech	speech	NOUN
bracis-33663	8	14	,	,	PUNCT
bracis-33663	8	15	offering	offer	VERB
bracis-33663	8	16	insights	insight	NOUN
bracis-33663	8	17	for	for	ADP
bracis-33663	8	18	hypothesis	hypothesis	NOUN
bracis-33663	8	19	design	design	NOUN
bracis-33663	8	20	.	.	PUNCT
bracis-33663	9	1	access	access	NOUN
bracis-33663	9	2	provided	provide	VERB
bracis-33663	9	3	by	by	ADP
bracis-33663	9	4	university	university	PROPN
bracis-33663	9	5	of	of	ADP
bracis-33663	9	6	notre	notre	PROPN
bracis-33663	9	7	dame	dame	PROPN
bracis-33663	9	8	hesburgh	hesburgh	PROPN
bracis-33663	9	9	library	library	PROPN
bracis-33663	9	10	.	.	PUNCT
bracis-33663	10	1	download	download	PROPN
bracis-33663	10	2	conference	conference	NOUN
bracis-33663	10	3	paper	paper	NOUN
bracis-33663	10	4	pdf	pdf	NOUN
bracis-33663	10	5	similar	similar	ADJ
bracis-33663	10	6	content	content	NOUN
bracis-33663	10	7	being	be	AUX
bracis-33663	10	8	viewed	view	VERB
bracis-33663	10	9	by	by	ADP
bracis-33663	10	10	others	other	NOUN
bracis-33663	10	11	exploring	explore	VERB
bracis-33663	10	12	hate	hate	NOUN
bracis-33663	10	13	speech	speech	NOUN
bracis-33663	10	14	detection	detection	NOUN
bracis-33663	10	15	:	:	PUNCT
bracis-33663	10	16	challenges	challenge	NOUN
bracis-33663	10	17	,	,	PUNCT
bracis-33663	10	18	resources	resource	NOUN
bracis-33663	10	19	,	,	PUNCT
bracis-33663	10	20	current	current	ADJ
bracis-33663	10	21	research	research	NOUN
bracis-33663	10	22	and	and	CCONJ
bracis-33663	10	23	future	future	ADJ
bracis-33663	10	24	directions	direction	NOUN
bracis-33663	10	25	article	article	NOUN
bracis-33663	10	26	04	04	NUM
bracis-33663	10	27	march	march	NOUN
bracis-33663	10	28	2025	2025	NUM
bracis-33663	10	29	improving	improve	VERB
bracis-33663	10	30	natural	natural	ADJ
bracis-33663	10	31	language	language	NOUN
bracis-33663	10	32	inference	inference	NOUN
bracis-33663	10	33	with	with	ADP
bracis-33663	10	34	residual	residual	ADJ
bracis-33663	10	35	attention	attention	NOUN
bracis-33663	10	36	chapter	chapter	NOUN
bracis-33663	10	37	©	©	PROPN
bracis-33663	10	38	2024	2024	NUM
bracis-33663	10	39	dss	dss	NOUN
bracis-33663	10	40	for	for	ADP
bracis-33663	10	41	hate	hate	NOUN
bracis-33663	10	42	speech	speech	NOUN
bracis-33663	10	43	detection	detection	NOUN
bracis-33663	10	44	chapter	chapter	NOUN
bracis-33663	10	45	©	©	NOUN
bracis-33663	10	46	2025	2025	NUM
bracis-33663	10	47	1	1	NUM
bracis-33663	10	48	introduction	introduction	NOUN
bracis-33663	10	49	recently	recently	ADV
bracis-33663	10	50	,	,	PUNCT
bracis-33663	10	51	there	there	PRON
bracis-33663	10	52	has	have	AUX
bracis-33663	10	53	been	be	AUX
bracis-33663	10	54	a	a	DET
bracis-33663	10	55	rise	rise	NOUN
bracis-33663	10	56	in	in	ADP
bracis-33663	10	57	harmful	harmful	ADJ
bracis-33663	10	58	content	content	NOUN
bracis-33663	10	59	like	like	ADP
bracis-33663	10	60	hate	hate	NOUN
bracis-33663	10	61	speech	speech	NOUN
bracis-33663	10	62	and	and	CCONJ
bracis-33663	10	63	misinformation	misinformation	NOUN
bracis-33663	10	64	on	on	ADP
bracis-33663	10	65	the	the	DET
bracis-33663	10	66	internet	internet	NOUN
bracis-33663	10	67	.	.	PUNCT
bracis-33663	11	1	this	this	DET
bracis-33663	11	2	increase	increase	NOUN
bracis-33663	11	3	may	may	AUX
bracis-33663	11	4	not	not	PART
bracis-33663	11	5	just	just	ADV
bracis-33663	11	6	be	be	AUX
bracis-33663	11	7	due	due	ADJ
bracis-33663	11	8	to	to	ADP
bracis-33663	11	9	more	more	ADJ
bracis-33663	11	10	people	people	NOUN
bracis-33663	11	11	using	use	VERB
bracis-33663	11	12	blogs	blog	NOUN
bracis-33663	11	13	and	and	CCONJ
bracis-33663	11	14	social	social	ADJ
bracis-33663	11	15	media	medium	NOUN
bracis-33663	11	16	but	but	CCONJ
bracis-33663	11	17	could	could	AUX
bracis-33663	11	18	also	also	ADV
bracis-33663	11	19	result	result	VERB
bracis-33663	11	20	from	from	ADP
bracis-33663	11	21	global	global	ADJ
bracis-33663	11	22	political	political	ADJ
bracis-33663	11	23	polarization	polarization	NOUN
bracis-33663	11	24	and	and	CCONJ
bracis-33663	11	25	social	social	ADJ
bracis-33663	11	26	media	medium	NOUN
bracis-33663	11	27	algorithms	algorithm	NOUN
bracis-33663	11	28	that	that	PRON
bracis-33663	11	29	amplify	amplify	VERB
bracis-33663	11	30	divisive	divisive	ADJ
bracis-33663	11	31	content	content	NOUN
bracis-33663	11	32	to	to	PART
bracis-33663	11	33	drive	drive	VERB
bracis-33663	11	34	engagement	engagement	NOUN
bracis-33663	11	35	.	.	PUNCT
bracis-33663	12	1	this	this	PRON
bracis-33663	12	2	makes	make	VERB
bracis-33663	12	3	online	online	ADJ
bracis-33663	12	4	safety	safety	NOUN
bracis-33663	12	5	and	and	CCONJ
bracis-33663	12	6	content	content	NOUN
bracis-33663	12	7	moderation	moderation	NOUN
bracis-33663	12	8	crucial	crucial	ADJ
bracis-33663	12	9	to	to	ADP
bracis-33663	12	10	maintaining	maintain	VERB
bracis-33663	12	11	civility	civility	NOUN
bracis-33663	12	12	and	and	CCONJ
bracis-33663	12	13	protecting	protect	VERB
bracis-33663	12	14	users	user	NOUN
bracis-33663	12	15	[	[	X
bracis-33663	12	16	9	9	NUM
bracis-33663	12	17	,	,	PUNCT
bracis-33663	12	18	19	19	NUM
bracis-33663	12	19	]	]	PUNCT
bracis-33663	12	20	.	.	PUNCT
bracis-33663	13	1	fig	fig	NOUN
bracis-33663	13	2	.	.	PUNCT
bracis-33663	14	1	1	1	X
bracis-33663	14	2	.	.	X
bracis-33663	14	3	combining	combine	VERB
bracis-33663	14	4	standard	standard	ADJ
bracis-33663	14	5	hypothesis	hypothesis	NOUN
bracis-33663	14	6	and	and	CCONJ
bracis-33663	14	7	p	p	NOUN
bracis-33663	14	8	-	-	PUNCT
bracis-33663	14	9	tuned	tune	VERB
bracis-33663	14	10	virtual	virtual	ADJ
bracis-33663	14	11	tokens	token	NOUN
bracis-33663	14	12	full	full	ADJ
bracis-33663	14	13	size	size	NOUN
bracis-33663	14	14	image	image	NOUN
bracis-33663	14	15	before	before	ADV
bracis-33663	14	16	,	,	PUNCT
bracis-33663	14	17	detecting	detect	VERB
bracis-33663	14	18	hate	hate	NOUN
bracis-33663	14	19	speech	speech	NOUN
bracis-33663	14	20	mainly	mainly	ADV
bracis-33663	14	21	relied	rely	VERB
bracis-33663	14	22	on	on	ADP
bracis-33663	14	23	classification	classification	NOUN
bracis-33663	14	24	models	model	NOUN
bracis-33663	14	25	that	that	PRON
bracis-33663	14	26	needed	need	VERB
bracis-33663	14	27	lots	lot	NOUN
bracis-33663	14	28	of	of	ADP
bracis-33663	14	29	labeled	label	VERB
bracis-33663	14	30	data	datum	NOUN
bracis-33663	14	31	[	[	X
bracis-33663	14	32	2	2	NUM
bracis-33663	14	33	,	,	PUNCT
bracis-33663	14	34	21	21	NUM
bracis-33663	14	35	]	]	PUNCT
bracis-33663	14	36	.	.	PUNCT
bracis-33663	15	1	recently	recently	ADV
bracis-33663	15	2	,	,	PUNCT
bracis-33663	15	3	natural	natural	ADJ
bracis-33663	15	4	language	language	NOUN
bracis-33663	15	5	inference	inference	NOUN
bracis-33663	15	6	(	(	PUNCT
bracis-33663	15	7	nli	nli	NOUN
bracis-33663	15	8	)	)	PUNCT
bracis-33663	15	9	models	model	NOUN
bracis-33663	15	10	,	,	PUNCT
bracis-33663	15	11	which	which	PRON
bracis-33663	15	12	are	be	AUX
bracis-33663	15	13	used	use	VERB
bracis-33663	15	14	for	for	ADP
bracis-33663	15	15	tasks	task	NOUN
bracis-33663	15	16	like	like	ADP
bracis-33663	15	17	zero	zero	NUM
bracis-33663	15	18	-	-	PUNCT
bracis-33663	15	19	shot	shot	NOUN
bracis-33663	15	20	classification	classification	NOUN
bracis-33663	15	21	[	[	X
bracis-33663	15	22	7	7	NUM
bracis-33663	15	23	]	]	PUNCT
bracis-33663	15	24	,	,	PUNCT
bracis-33663	15	25	are	be	AUX
bracis-33663	15	26	being	be	AUX
bracis-33663	15	27	explored	explore	VERB
bracis-33663	15	28	.	.	PUNCT
bracis-33663	16	1	nli	nli	PROPN
bracis-33663	16	2	models	model	NOUN
bracis-33663	16	3	show	show	VERB
bracis-33663	16	4	promising	promising	ADJ
bracis-33663	16	5	results	result	NOUN
bracis-33663	16	6	for	for	ADP
bracis-33663	16	7	zero	zero	NUM
bracis-33663	16	8	-	-	PUNCT
bracis-33663	16	9	shot	shot	NOUN
bracis-33663	16	10	hate	hate	NOUN
bracis-33663	16	11	speech	speech	NOUN
bracis-33663	16	12	detection	detection	NOUN
bracis-33663	16	13	(	(	PUNCT
bracis-33663	16	14	zshsd	zshsd	PROPN
bracis-33663	16	15	)	)	PUNCT
bracis-33663	16	16	because	because	SCONJ
bracis-33663	16	17	they	they	PRON
bracis-33663	16	18	capture	capture	VERB
bracis-33663	16	19	relationships	relationship	NOUN
bracis-33663	16	20	between	between	ADP
bracis-33663	16	21	sentences	sentence	NOUN
bracis-33663	16	22	,	,	PUNCT
bracis-33663	16	23	which	which	PRON
bracis-33663	16	24	helps	help	VERB
bracis-33663	16	25	adapt	adapt	VERB
bracis-33663	16	26	content	content	NOUN
bracis-33663	16	27	moderation	moderation	NOUN
bracis-33663	16	28	to	to	ADP
bracis-33663	16	29	the	the	DET
bracis-33663	16	30	ever	ever	ADV
bracis-33663	16	31	-	-	PUNCT
bracis-33663	16	32	changing	change	VERB
bracis-33663	16	33	online	online	ADJ
bracis-33663	16	34	world	world	NOUN
bracis-33663	16	35	.	.	PUNCT
bracis-33663	17	1	in	in	ADP
bracis-33663	17	2	our	our	PRON
bracis-33663	17	3	study	study	NOUN
bracis-33663	17	4	,	,	PUNCT
bracis-33663	17	5	we	we	PRON
bracis-33663	17	6	’re	’re	AUX
bracis-33663	17	7	using	use	VERB
bracis-33663	17	8	the	the	DET
bracis-33663	17	9	term	term	NOUN
bracis-33663	17	10	“	"	PUNCT
bracis-33663	17	11	hate	hate	NOUN
bracis-33663	17	12	speech	speech	NOUN
bracis-33663	17	13	”	"	PUNCT
bracis-33663	17	14	to	to	PART
bracis-33663	17	15	cover	cover	VERB
bracis-33663	17	16	a	a	DET
bracis-33663	17	17	wide	wide	ADJ
bracis-33663	17	18	range	range	NOUN
bracis-33663	17	19	of	of	ADP
bracis-33663	17	20	offensive	offensive	ADJ
bracis-33663	17	21	content	content	NOUN
bracis-33663	17	22	created	create	VERB
bracis-33663	17	23	by	by	ADP
bracis-33663	17	24	users	user	NOUN
bracis-33663	17	25	.	.	PUNCT
bracis-33663	18	1	we	we	PRON
bracis-33663	18	2	chose	choose	VERB
bracis-33663	18	3	this	this	DET
bracis-33663	18	4	term	term	NOUN
bracis-33663	18	5	because	because	SCONJ
bracis-33663	18	6	it	it	PRON
bracis-33663	18	7	reflects	reflect	VERB
bracis-33663	18	8	hurtful	hurtful	ADJ
bracis-33663	18	9	communication	communication	NOUN
bracis-33663	18	10	that	that	PRON
bracis-33663	18	11	spreads	spread	VERB
bracis-33663	18	12	hostility	hostility	NOUN
bracis-33663	18	13	through	through	ADP
bracis-33663	18	14	stereotypes	stereotype	NOUN
bracis-33663	18	15	[	[	X
bracis-33663	18	16	15	15	NUM
bracis-33663	18	17	]	]	PUNCT
bracis-33663	18	18	.	.	PUNCT
bracis-33663	19	1	in	in	ADP
bracis-33663	19	2	nli	nli	PROPN
bracis-33663	19	3	models	model	NOUN
bracis-33663	19	4	,	,	PUNCT
bracis-33663	19	5	generating	generate	VERB
bracis-33663	19	6	hypotheses	hypothesis	NOUN
bracis-33663	19	7	is	be	AUX
bracis-33663	19	8	crucial	crucial	ADJ
bracis-33663	19	9	and	and	CCONJ
bracis-33663	19	10	affects	affect	VERB
bracis-33663	19	11	tasks	task	NOUN
bracis-33663	19	12	like	like	ADP
bracis-33663	19	13	zshsd	zshsd	NOUN
bracis-33663	20	1	[	[	X
bracis-33663	20	2	20	20	NUM
bracis-33663	20	3	]	]	PUNCT
bracis-33663	20	4	.	.	PUNCT
bracis-33663	21	1	even	even	ADV
bracis-33663	21	2	small	small	ADJ
bracis-33663	21	3	changes	change	NOUN
bracis-33663	21	4	in	in	ADP
bracis-33663	21	5	how	how	SCONJ
bracis-33663	21	6	a	a	DET
bracis-33663	21	7	hypothesis	hypothesis	NOUN
bracis-33663	21	8	is	be	AUX
bracis-33663	21	9	worded	word	VERB
bracis-33663	21	10	can	can	AUX
bracis-33663	21	11	greatly	greatly	ADV
bracis-33663	21	12	impact	impact	VERB
bracis-33663	21	13	the	the	DET
bracis-33663	21	14	model	model	NOUN
bracis-33663	21	15	’s	’s	PART
bracis-33663	21	16	performance	performance	NOUN
bracis-33663	21	17	.	.	PUNCT
bracis-33663	22	1	for	for	ADP
bracis-33663	22	2	example	example	NOUN
bracis-33663	22	3	,	,	PUNCT
bracis-33663	22	4	compare	compare	VERB
bracis-33663	22	5	“	"	PUNCT
bracis-33663	22	6	climate	climate	NOUN
bracis-33663	22	7	change	change	NOUN
bracis-33663	22	8	is	be	AUX
bracis-33663	22	9	a	a	DET
bracis-33663	22	10	pressing	press	VERB
bracis-33663	22	11	global	global	ADJ
bracis-33663	22	12	issue	issue	NOUN
bracis-33663	22	13	”	"	PUNCT
bracis-33663	22	14	to	to	ADP
bracis-33663	22	15	“	"	PUNCT
bracis-33663	22	16	addressing	address	VERB
bracis-33663	22	17	climate	climate	NOUN
bracis-33663	22	18	change	change	NOUN
bracis-33663	22	19	requires	require	VERB
bracis-33663	22	20	urgent	urgent	ADJ
bracis-33663	22	21	global	global	ADJ
bracis-33663	22	22	cooperation	cooperation	NOUN
bracis-33663	22	23	.	.	PUNCT
bracis-33663	22	24	”	"	PUNCT
bracis-33663	23	1	both	both	DET
bracis-33663	23	2	statements	statement	NOUN
bracis-33663	23	3	express	express	VERB
bracis-33663	23	4	concern	concern	NOUN
bracis-33663	23	5	about	about	ADP
bracis-33663	23	6	climate	climate	NOUN
bracis-33663	23	7	change	change	NOUN
bracis-33663	23	8	,	,	PUNCT
bracis-33663	23	9	but	but	CCONJ
bracis-33663	23	10	the	the	DET
bracis-33663	23	11	second	second	ADJ
bracis-33663	23	12	one	one	NOUN
bracis-33663	23	13	emphasizes	emphasize	VERB
bracis-33663	23	14	the	the	DET
bracis-33663	23	15	need	need	NOUN
bracis-33663	23	16	for	for	ADP
bracis-33663	23	17	global	global	ADJ
bracis-33663	23	18	collaboration	collaboration	NOUN
bracis-33663	23	19	,	,	PUNCT
bracis-33663	23	20	which	which	PRON
bracis-33663	23	21	could	could	AUX
bracis-33663	23	22	lead	lead	VERB
bracis-33663	23	23	to	to	ADP
bracis-33663	23	24	different	different	ADJ
bracis-33663	23	25	conclusions	conclusion	NOUN
bracis-33663	23	26	and	and	CCONJ
bracis-33663	23	27	outcomes	outcome	NOUN
bracis-33663	23	28	in	in	ADP
bracis-33663	23	29	zshsd	zshsd	PROPN
bracis-33663	23	30	.	.	PUNCT
bracis-33663	24	1	to	to	PART
bracis-33663	24	2	better	well	ADV
bracis-33663	24	3	handle	handle	VERB
bracis-33663	24	4	the	the	DET
bracis-33663	24	5	complexity	complexity	NOUN
bracis-33663	24	6	of	of	ADP
bracis-33663	24	7	forming	form	VERB
bracis-33663	24	8	hypotheses	hypothesis	NOUN
bracis-33663	24	9	,	,	PUNCT
bracis-33663	24	10	current	current	ADJ
bracis-33663	24	11	frameworks	framework	NOUN
bracis-33663	24	12	using	use	VERB
bracis-33663	24	13	nli	nli	PROPN
bracis-33663	24	14	models	model	NOUN
bracis-33663	24	15	take	take	VERB
bracis-33663	24	16	a	a	DET
bracis-33663	24	17	step	step	NOUN
bracis-33663	24	18	-	-	PUNCT
bracis-33663	24	19	by	by	ADP
bracis-33663	24	20	-	-	PUNCT
bracis-33663	24	21	step	step	NOUN
bracis-33663	24	22	approach	approach	NOUN
bracis-33663	24	23	.	.	PUNCT
bracis-33663	25	1	they	they	PRON
bracis-33663	25	2	use	use	VERB
bracis-33663	25	3	multiple	multiple	ADJ
bracis-33663	25	4	hypotheses	hypothesis	NOUN
bracis-33663	25	5	to	to	PART
bracis-33663	25	6	understand	understand	VERB
bracis-33663	25	7	different	different	ADJ
bracis-33663	25	8	aspects	aspect	NOUN
bracis-33663	25	9	of	of	ADP
bracis-33663	25	10	text	text	NOUN
bracis-33663	25	11	,	,	PUNCT
bracis-33663	25	12	like	like	ADP
bracis-33663	25	13	identifying	identify	VERB
bracis-33663	25	14	hate	hate	NOUN
bracis-33663	25	15	speech	speech	NOUN
bracis-33663	25	16	in	in	ADP
bracis-33663	25	17	a	a	DET
bracis-33663	25	18	tweet	tweet	NOUN
bracis-33663	25	19	.	.	PUNCT
bracis-33663	26	1	for	for	ADP
bracis-33663	26	2	example	example	NOUN
bracis-33663	26	3	,	,	PUNCT
bracis-33663	26	4	one	one	NUM
bracis-33663	26	5	hypothesis	hypothesis	NOUN
bracis-33663	26	6	might	might	AUX
bracis-33663	26	7	spot	spot	VERB
bracis-33663	26	8	explicit	explicit	ADJ
bracis-33663	26	9	language	language	NOUN
bracis-33663	26	10	(	(	PUNCT
bracis-33663	26	11	“	"	PUNCT
bracis-33663	26	12	this	this	DET
bracis-33663	26	13	tweet	tweet	NOUN
bracis-33663	26	14	has	have	VERB
bracis-33663	26	15	offensive	offensive	ADJ
bracis-33663	26	16	words	word	NOUN
bracis-33663	26	17	”	"	PUNCT
bracis-33663	26	18	)	)	PUNCT
bracis-33663	26	19	,	,	PUNCT
bracis-33663	26	20	while	while	SCONJ
bracis-33663	26	21	another	another	PRON
bracis-33663	26	22	digs	dig	VERB
bracis-33663	26	23	deeper	deeply	ADV
bracis-33663	26	24	into	into	ADP
bracis-33663	26	25	implied	imply	VERB
bracis-33663	26	26	meanings	meaning	NOUN
bracis-33663	26	27	(	(	PUNCT
bracis-33663	26	28	“	"	PUNCT
bracis-33663	26	29	the	the	DET
bracis-33663	26	30	tone	tone	NOUN
bracis-33663	26	31	of	of	ADP
bracis-33663	26	32	this	this	DET
bracis-33663	26	33	tweet	tweet	NOUN
bracis-33663	26	34	encourages	encourage	VERB
bracis-33663	26	35	hostility	hostility	NOUN
bracis-33663	26	36	”	"	PUNCT
bracis-33663	26	37	)	)	PUNCT
bracis-33663	26	38	.	.	PUNCT
bracis-33663	27	1	this	this	DET
bracis-33663	27	2	method	method	NOUN
bracis-33663	27	3	helps	help	VERB
bracis-33663	27	4	the	the	DET
bracis-33663	27	5	model	model	NOUN
bracis-33663	27	6	grasp	grasp	VERB
bracis-33663	27	7	various	various	ADJ
bracis-33663	27	8	aspects	aspect	NOUN
bracis-33663	27	9	of	of	ADP
bracis-33663	27	10	text	text	NOUN
bracis-33663	27	11	,	,	PUNCT
bracis-33663	27	12	which	which	PRON
bracis-33663	27	13	is	be	AUX
bracis-33663	27	14	crucial	crucial	ADJ
bracis-33663	27	15	for	for	ADP
bracis-33663	27	16	accurate	accurate	ADJ
bracis-33663	27	17	zshsd	zshsd	NOUN
bracis-33663	27	18	performance	performance	NOUN
bracis-33663	27	19	[	[	X
bracis-33663	27	20	7	7	NUM
bracis-33663	27	21	]	]	PUNCT
bracis-33663	27	22	.	.	PUNCT
bracis-33663	28	1	traditionally	traditionally	ADV
bracis-33663	28	2	,	,	PUNCT
bracis-33663	28	3	approaches	approach	NOUN
bracis-33663	28	4	used	use	VERB
bracis-33663	28	5	simple	simple	ADJ
bracis-33663	28	6	hypotheses	hypothesis	NOUN
bracis-33663	28	7	like	like	ADP
bracis-33663	28	8	“	"	PUNCT
bracis-33663	28	9	that	that	PRON
bracis-33663	28	10	’s	’	VERB
bracis-33663	28	11	hate	hate	NOUN
bracis-33663	28	12	speech	speech	NOUN
bracis-33663	28	13	”	"	PUNCT
bracis-33663	28	14	which	which	PRON
bracis-33663	28	15	,	,	PUNCT
bracis-33663	28	16	while	while	SCONJ
bracis-33663	28	17	somewhat	somewhat	ADV
bracis-33663	28	18	effective	effective	ADJ
bracis-33663	28	19	,	,	PUNCT
bracis-33663	28	20	often	often	ADV
bracis-33663	28	21	oversimplify	oversimplify	VERB
bracis-33663	28	22	the	the	DET
bracis-33663	28	23	complexities	complexity	NOUN
bracis-33663	28	24	of	of	ADP
bracis-33663	28	25	identifying	identify	VERB
bracis-33663	28	26	hate	hate	NOUN
bracis-33663	28	27	speech	speech	NOUN
bracis-33663	28	28	.	.	PUNCT
bracis-33663	29	1	these	these	DET
bracis-33663	29	2	basic	basic	ADJ
bracis-33663	29	3	hypotheses	hypothesis	NOUN
bracis-33663	29	4	struggle	struggle	VERB
bracis-33663	29	5	to	to	PART
bracis-33663	29	6	capture	capture	VERB
bracis-33663	29	7	the	the	DET
bracis-33663	29	8	wide	wide	ADJ
bracis-33663	29	9	range	range	NOUN
bracis-33663	29	10	of	of	ADP
bracis-33663	29	11	nuances	nuance	NOUN
bracis-33663	29	12	,	,	PUNCT
bracis-33663	29	13	subtleties	subtlety	NOUN
bracis-33663	29	14	,	,	PUNCT
bracis-33663	29	15	and	and	CCONJ
bracis-33663	29	16	context	context	NOUN
bracis-33663	29	17	-	-	PUNCT
bracis-33663	29	18	dependent	dependent	ADJ
bracis-33663	29	19	variations	variation	NOUN
bracis-33663	29	20	in	in	ADP
bracis-33663	29	21	hate	hate	NOUN
bracis-33663	29	22	speech	speech	NOUN
bracis-33663	29	23	[	[	X
bracis-33663	29	24	7	7	NUM
bracis-33663	29	25	,	,	PUNCT
bracis-33663	29	26	13	13	NUM
bracis-33663	29	27	]	]	PUNCT
bracis-33663	29	28	.	.	PUNCT
bracis-33663	30	1	this	this	PRON
bracis-33663	30	2	has	have	AUX
bracis-33663	30	3	led	lead	VERB
bracis-33663	30	4	to	to	ADP
bracis-33663	30	5	a	a	DET
bracis-33663	30	6	growing	grow	VERB
bracis-33663	30	7	recognition	recognition	NOUN
bracis-33663	30	8	of	of	ADP
bracis-33663	30	9	the	the	DET
bracis-33663	30	10	need	need	NOUN
bracis-33663	30	11	for	for	ADP
bracis-33663	30	12	more	more	ADV
bracis-33663	30	13	sophisticated	sophisticated	ADJ
bracis-33663	30	14	hypothesis	hypothesis	NOUN
bracis-33663	30	15	generation	generation	NOUN
bracis-33663	30	16	in	in	ADP
bracis-33663	30	17	zshsd	zshsd	PROPN
bracis-33663	30	18	.	.	PUNCT
bracis-33663	31	1	in	in	ADP
bracis-33663	31	2	our	our	PRON
bracis-33663	31	3	work	work	NOUN
bracis-33663	31	4	,	,	PUNCT
bracis-33663	31	5	we	we	PRON
bracis-33663	31	6	aim	aim	VERB
bracis-33663	31	7	to	to	PART
bracis-33663	31	8	tackle	tackle	VERB
bracis-33663	31	9	the	the	DET
bracis-33663	31	10	difficulties	difficulty	NOUN
bracis-33663	31	11	linked	link	VERB
bracis-33663	31	12	with	with	ADP
bracis-33663	31	13	hypothesis	hypothesis	NOUN
bracis-33663	31	14	generation	generation	NOUN
bracis-33663	31	15	in	in	ADP
bracis-33663	31	16	nli	nli	PROPN
bracis-33663	31	17	models	model	NOUN
bracis-33663	31	18	.	.	PUNCT
bracis-33663	32	1	we	we	PRON
bracis-33663	32	2	introduce	introduce	VERB
bracis-33663	32	3	a	a	DET
bracis-33663	32	4	method	method	NOUN
bracis-33663	32	5	that	that	PRON
bracis-33663	32	6	optimizes	optimize	VERB
bracis-33663	32	7	hypothesis	hypothesis	NOUN
bracis-33663	32	8	creation	creation	NOUN
bracis-33663	32	9	through	through	ADP
bracis-33663	32	10	a	a	DET
bracis-33663	32	11	process	process	NOUN
bracis-33663	32	12	known	know	VERB
bracis-33663	32	13	as	as	ADP
bracis-33663	32	14	“	"	PUNCT
bracis-33663	32	15	p	p	NOUN
bracis-33663	32	16	-	-	PUNCT
bracis-33663	32	17	tuning	tuning	NOUN
bracis-33663	32	18	”	"	PUNCT
bracis-33663	33	1	[	[	X
bracis-33663	33	2	10	10	NUM
bracis-33663	33	3	]	]	PUNCT
bracis-33663	33	4	,	,	PUNCT
bracis-33663	33	5	which	which	PRON
bracis-33663	33	6	starts	start	VERB
bracis-33663	33	7	with	with	ADP
bracis-33663	33	8	a	a	DET
bracis-33663	33	9	standard	standard	ADJ
bracis-33663	33	10	hypothesis	hypothesis	NOUN
bracis-33663	33	11	prompt	prompt	NOUN
bracis-33663	33	12	but	but	CCONJ
bracis-33663	33	13	iteratively	iteratively	ADV
bracis-33663	33	14	refines	refine	VERB
bracis-33663	33	15	it	it	PRON
bracis-33663	33	16	to	to	PART
bracis-33663	33	17	suit	suit	VERB
bracis-33663	33	18	the	the	DET
bracis-33663	33	19	complexities	complexity	NOUN
bracis-33663	33	20	of	of	ADP
bracis-33663	33	21	zshsd	zshsd	PROPN
bracis-33663	33	22	.	.	PUNCT
bracis-33663	34	1	p	p	X
bracis-33663	34	2	-	-	PUNCT
bracis-33663	34	3	tuning	tuning	NOUN
bracis-33663	34	4	allows	allow	VERB
bracis-33663	34	5	us	we	PRON
bracis-33663	34	6	to	to	PART
bracis-33663	34	7	adapt	adapt	VERB
bracis-33663	34	8	the	the	DET
bracis-33663	34	9	hypothesis	hypothesis	NOUN
bracis-33663	34	10	to	to	ADP
bracis-33663	34	11	the	the	DET
bracis-33663	34	12	specific	specific	ADJ
bracis-33663	34	13	characteristics	characteristic	NOUN
bracis-33663	34	14	and	and	CCONJ
bracis-33663	34	15	nuances	nuance	NOUN
bracis-33663	34	16	of	of	ADP
bracis-33663	34	17	the	the	DET
bracis-33663	34	18	task	task	NOUN
bracis-33663	34	19	,	,	PUNCT
bracis-33663	34	20	ensuring	ensure	VERB
bracis-33663	34	21	that	that	SCONJ
bracis-33663	34	22	it	it	PRON
bracis-33663	34	23	encapsulates	encapsulate	VERB
bracis-33663	34	24	a	a	DET
bracis-33663	34	25	broader	broad	ADJ
bracis-33663	34	26	and	and	CCONJ
bracis-33663	34	27	more	more	ADV
bracis-33663	34	28	representative	representative	ADJ
bracis-33663	34	29	range	range	NOUN
bracis-33663	34	30	of	of	ADP
bracis-33663	34	31	hate	hate	NOUN
bracis-33663	34	32	speech	speech	NOUN
bracis-33663	34	33	expressions	expression	NOUN
bracis-33663	34	34	.	.	PUNCT
bracis-33663	35	1	general	general	ADJ
bracis-33663	35	2	objective	objective	NOUN
bracis-33663	35	3	:	:	PUNCT
bracis-33663	35	4	this	this	DET
bracis-33663	35	5	article	article	NOUN
bracis-33663	35	6	aims	aim	VERB
bracis-33663	35	7	to	to	PART
bracis-33663	35	8	tackle	tackle	VERB
bracis-33663	35	9	the	the	DET
bracis-33663	35	10	challenge	challenge	NOUN
bracis-33663	35	11	in	in	ADP
bracis-33663	35	12	hypothesis	hypothesis	NOUN
bracis-33663	35	13	generation	generation	NOUN
bracis-33663	35	14	for	for	ADP
bracis-33663	35	15	natural	natural	ADJ
bracis-33663	35	16	language	language	NOUN
bracis-33663	35	17	inference	inference	NOUN
bracis-33663	35	18	when	when	SCONJ
bracis-33663	35	19	used	use	VERB
bracis-33663	35	20	for	for	ADP
bracis-33663	35	21	zero	zero	NUM
bracis-33663	35	22	-	-	PUNCT
bracis-33663	35	23	shot	shot	NOUN
bracis-33663	35	24	classification	classification	NOUN
bracis-33663	35	25	.	.	PUNCT
bracis-33663	36	1	it	it	PRON
bracis-33663	36	2	uses	use	VERB
bracis-33663	36	3	“	"	PUNCT
bracis-33663	36	4	p	p	NOUN
bracis-33663	36	5	-	-	PUNCT
bracis-33663	36	6	tuning	tuning	NOUN
bracis-33663	36	7	”	"	PUNCT
bracis-33663	36	8	as	as	ADP
bracis-33663	36	9	a	a	DET
bracis-33663	36	10	method	method	NOUN
bracis-33663	36	11	to	to	PART
bracis-33663	36	12	optimize	optimize	VERB
bracis-33663	36	13	hypothesis	hypothesis	NOUN
bracis-33663	36	14	creation	creation	NOUN
bracis-33663	36	15	for	for	ADP
bracis-33663	36	16	addressing	address	VERB
bracis-33663	36	17	the	the	DET
bracis-33663	36	18	nuances	nuance	NOUN
bracis-33663	36	19	involved	involve	VERB
bracis-33663	36	20	in	in	ADP
bracis-33663	36	21	hate	hate	NOUN
bracis-33663	36	22	speech	speech	NOUN
bracis-33663	36	23	detection	detection	NOUN
bracis-33663	36	24	.	.	PUNCT
bracis-33663	37	1	in	in	ADP
bracis-33663	37	2	this	this	DET
bracis-33663	37	3	process	process	NOUN
bracis-33663	37	4	,	,	PUNCT
bracis-33663	37	5	we	we	PRON
bracis-33663	37	6	start	start	VERB
bracis-33663	37	7	with	with	ADP
bracis-33663	37	8	a	a	DET
bracis-33663	37	9	initial	initial	ADJ
bracis-33663	37	10	hypothesis	hypothesis	NOUN
bracis-33663	37	11	(	(	PUNCT
bracis-33663	37	12	h	h	NOUN
bracis-33663	37	13	)	)	PUNCT
bracis-33663	37	14	,	,	PUNCT
bracis-33663	37	15	choosing	choose	VERB
bracis-33663	37	16	the	the	DET
bracis-33663	37	17	most	most	ADV
bracis-33663	37	18	accurate	accurate	ADJ
bracis-33663	37	19	one	one	NOUN
bracis-33663	37	20	,	,	PUNCT
bracis-33663	37	21	out	out	ADP
bracis-33663	37	22	of	of	ADP
bracis-33663	37	23	different	different	ADJ
bracis-33663	37	24	hypothesis	hypothesis	NOUN
bracis-33663	37	25	sentence	sentence	NOUN
bracis-33663	37	26	designed	design	VERB
bracis-33663	37	27	to	to	PART
bracis-33663	37	28	detect	detect	VERB
bracis-33663	37	29	hate	hate	NOUN
bracis-33663	37	30	speech	speech	NOUN
bracis-33663	37	31	using	use	VERB
bracis-33663	37	32	nli	nli	PROPN
bracis-33663	37	33	.	.	PUNCT
bracis-33663	38	1	results	result	NOUN
bracis-33663	38	2	can	can	AUX
bracis-33663	38	3	be	be	AUX
bracis-33663	38	4	seen	see	VERB
bracis-33663	38	5	in	in	ADP
bracis-33663	38	6	table	table	NOUN
bracis-33663	38	7	 	 	SPACE
bracis-33663	38	8	2	2	NUM
bracis-33663	38	9	.	.	PUNCT
bracis-33663	39	1	next	next	ADV
bracis-33663	39	2	,	,	PUNCT
bracis-33663	39	3	the	the	DET
bracis-33663	39	4	optimized	optimize	VERB
bracis-33663	39	5	hypothesis	hypothesis	NOUN
bracis-33663	39	6	(	(	PUNCT
bracis-33663	39	7	oh	oh	INTJ
bracis-33663	39	8	)	)	PUNCT
bracis-33663	39	9	is	be	AUX
bracis-33663	39	10	generated	generate	VERB
bracis-33663	39	11	,	,	PUNCT
bracis-33663	39	12	which	which	PRON
bracis-33663	39	13	starts	start	VERB
bracis-33663	39	14	from	from	ADP
bracis-33663	39	15	the	the	DET
bracis-33663	39	16	best	good	ADJ
bracis-33663	39	17	performed	perform	VERB
bracis-33663	39	18	hypothesis	hypothesis	NOUN
bracis-33663	39	19	of	of	ADP
bracis-33663	39	20	h	h	NOUN
bracis-33663	39	21	,	,	PUNCT
bracis-33663	39	22	and	and	CCONJ
bracis-33663	39	23	is	be	AUX
bracis-33663	39	24	optimized	optimize	VERB
bracis-33663	39	25	using	use	VERB
bracis-33663	39	26	p	p	NOUN
bracis-33663	39	27	-	-	PUNCT
bracis-33663	39	28	tuning	tuning	NOUN
bracis-33663	39	29	.	.	PUNCT
bracis-33663	40	1	different	different	ADJ
bracis-33663	40	2	strategies	strategy	NOUN
bracis-33663	40	3	of	of	ADP
bracis-33663	40	4	using	use	VERB
bracis-33663	40	5	the	the	DET
bracis-33663	40	6	resulting	result	VERB
bracis-33663	40	7	virtual	virtual	ADJ
bracis-33663	40	8	tokens	token	NOUN
bracis-33663	40	9	can	can	AUX
bracis-33663	40	10	be	be	AUX
bracis-33663	40	11	seen	see	VERB
bracis-33663	40	12	in	in	ADP
bracis-33663	40	13	table	table	NOUN
bracis-33663	40	14	 	 	SPACE
bracis-33663	40	15	4	4	NUM
bracis-33663	40	16	and	and	CCONJ
bracis-33663	40	17	table	table	NOUN
bracis-33663	40	18	 	 	SPACE
bracis-33663	40	19	5	5	NUM
bracis-33663	40	20	.	.	PUNCT
bracis-33663	41	1	finally	finally	ADV
bracis-33663	41	2	,	,	PUNCT
bracis-33663	41	3	the	the	DET
bracis-33663	41	4	combined	combined	ADJ
bracis-33663	41	5	hypothesis	hypothesis	NOUN
bracis-33663	41	6	(	(	PUNCT
bracis-33663	41	7	ch	ch	NOUN
bracis-33663	41	8	)	)	PUNCT
bracis-33663	41	9	is	be	AUX
bracis-33663	41	10	formed	form	VERB
bracis-33663	41	11	by	by	ADP
bracis-33663	41	12	integrating	integrate	VERB
bracis-33663	41	13	the	the	DET
bracis-33663	41	14	most	most	ADV
bracis-33663	41	15	relevant	relevant	ADJ
bracis-33663	41	16	tokens	token	NOUN
bracis-33663	41	17	from	from	ADP
bracis-33663	41	18	oh	oh	INTJ
bracis-33663	41	19	with	with	ADP
bracis-33663	41	20	the	the	DET
bracis-33663	41	21	original	original	ADJ
bracis-33663	41	22	h.	h.	NOUN
bracis-33663	41	23	the	the	DET
bracis-33663	41	24	accuracy	accuracy	NOUN
bracis-33663	41	25	of	of	ADP
bracis-33663	41	26	different	different	ADJ
bracis-33663	41	27	virtual	virtual	ADJ
bracis-33663	41	28	tokens	token	NOUN
bracis-33663	41	29	inserted	insert	VERB
bracis-33663	41	30	into	into	ADP
bracis-33663	41	31	the	the	DET
bracis-33663	41	32	original	original	ADJ
bracis-33663	41	33	hypothesis	hypothesis	NOUN
bracis-33663	41	34	can	can	AUX
bracis-33663	41	35	be	be	AUX
bracis-33663	41	36	seen	see	VERB
bracis-33663	41	37	in	in	ADP
bracis-33663	41	38	table	table	NOUN
bracis-33663	41	39	 	 	SPACE
bracis-33663	41	40	4	4	NUM
bracis-33663	41	41	and	and	CCONJ
bracis-33663	41	42	table	table	NOUN
bracis-33663	41	43	 	 	SPACE
bracis-33663	41	44	5	5	NUM
bracis-33663	41	45	,	,	PUNCT
bracis-33663	41	46	and	and	CCONJ
bracis-33663	41	47	a	a	DET
bracis-33663	41	48	general	general	ADJ
bracis-33663	41	49	benchmark	benchmark	NOUN
bracis-33663	41	50	,	,	PUNCT
bracis-33663	41	51	comparing	compare	VERB
bracis-33663	41	52	with	with	ADP
bracis-33663	41	53	previous	previous	ADJ
bracis-33663	41	54	works	work	NOUN
bracis-33663	41	55	,	,	PUNCT
bracis-33663	41	56	can	can	AUX
bracis-33663	41	57	be	be	AUX
bracis-33663	41	58	seen	see	VERB
bracis-33663	41	59	in	in	ADP
bracis-33663	41	60	table	table	NOUN
bracis-33663	41	61	 	 	SPACE
bracis-33663	41	62	6	6	NUM
bracis-33663	41	63	.	.	PUNCT
bracis-33663	42	1	a	a	DET
bracis-33663	42	2	visual	visual	ADJ
bracis-33663	42	3	scheme	scheme	NOUN
bracis-33663	42	4	of	of	ADP
bracis-33663	42	5	this	this	DET
bracis-33663	42	6	idea	idea	NOUN
bracis-33663	42	7	is	be	AUX
bracis-33663	42	8	illustrated	illustrate	VERB
bracis-33663	42	9	in	in	ADP
bracis-33663	42	10	fig	fig	NOUN
bracis-33663	42	11	.	.	PUNCT
bracis-33663	42	12	 	 	SPACE
bracis-33663	43	1	1	1	X
bracis-33663	43	2	.	.	PUNCT
bracis-33663	44	1	our	our	PRON
bracis-33663	44	2	findings	finding	NOUN
bracis-33663	44	3	highlight	highlight	VERB
bracis-33663	44	4	the	the	DET
bracis-33663	44	5	effectiveness	effectiveness	NOUN
bracis-33663	44	6	of	of	ADP
bracis-33663	44	7	our	our	PRON
bracis-33663	44	8	optimized	optimize	VERB
bracis-33663	44	9	hypothesis	hypothesis	NOUN
bracis-33663	44	10	(	(	PUNCT
bracis-33663	44	11	oh	oh	INTJ
bracis-33663	44	12	)	)	PUNCT
bracis-33663	44	13	,	,	PUNCT
bracis-33663	44	14	as	as	SCONJ
bracis-33663	44	15	certain	certain	ADJ
bracis-33663	44	16	virtual	virtual	ADJ
bracis-33663	44	17	tokens	token	NOUN
bracis-33663	44	18	within	within	ADP
bracis-33663	44	19	the	the	DET
bracis-33663	44	20	continuous	continuous	ADJ
bracis-33663	44	21	embedding	embed	VERB
bracis-33663	44	22	space	space	NOUN
bracis-33663	44	23	closely	closely	ADV
bracis-33663	44	24	approximate	approximate	ADJ
bracis-33663	44	25	broader	broad	ADJ
bracis-33663	44	26	hate	hate	VERB
bracis-33663	44	27	speech	speech	NOUN
bracis-33663	44	28	concepts	concept	NOUN
bracis-33663	44	29	.	.	PUNCT
bracis-33663	45	1	this	this	PRON
bracis-33663	45	2	suggests	suggest	VERB
bracis-33663	45	3	that	that	SCONJ
bracis-33663	45	4	including	include	VERB
bracis-33663	45	5	these	these	DET
bracis-33663	45	6	virtual	virtual	ADJ
bracis-33663	45	7	tokens	token	NOUN
bracis-33663	45	8	(	(	PUNCT
bracis-33663	45	9	oh	oh	INTJ
bracis-33663	45	10	)	)	PUNCT
bracis-33663	45	11	in	in	ADP
bracis-33663	45	12	our	our	PRON
bracis-33663	45	13	original	original	ADJ
bracis-33663	45	14	hypothesis	hypothesis	NOUN
bracis-33663	45	15	(	(	PUNCT
bracis-33663	45	16	h	h	NOUN
bracis-33663	45	17	)	)	PUNCT
bracis-33663	45	18	formulation	formulation	NOUN
bracis-33663	45	19	can	can	AUX
bracis-33663	45	20	enhance	enhance	VERB
bracis-33663	45	21	the	the	DET
bracis-33663	45	22	model	model	NOUN
bracis-33663	45	23	’s	’s	PART
bracis-33663	45	24	capacity	capacity	NOUN
bracis-33663	45	25	to	to	PART
bracis-33663	45	26	detect	detect	VERB
bracis-33663	45	27	and	and	CCONJ
bracis-33663	45	28	address	address	VERB
bracis-33663	45	29	hate	hate	NOUN
bracis-33663	45	30	speech	speech	NOUN
bracis-33663	45	31	more	more	ADV
bracis-33663	45	32	comprehensively	comprehensively	ADV
bracis-33663	45	33	.	.	PUNCT
bracis-33663	46	1	this	this	PRON
bracis-33663	46	2	opens	open	VERB
bracis-33663	46	3	up	up	ADP
bracis-33663	46	4	three	three	NUM
bracis-33663	46	5	promising	promising	ADJ
bracis-33663	46	6	avenues	avenue	NOUN
bracis-33663	46	7	for	for	ADP
bracis-33663	46	8	future	future	ADJ
bracis-33663	46	9	research	research	NOUN
bracis-33663	46	10	:	:	PUNCT
bracis-33663	46	11	first	first	ADV
bracis-33663	46	12	:	:	PUNCT
bracis-33663	46	13	researchers	researcher	NOUN
bracis-33663	46	14	can	can	AUX
bracis-33663	46	15	directly	directly	ADV
bracis-33663	46	16	use	use	VERB
bracis-33663	46	17	our	our	PRON
bracis-33663	46	18	virtual	virtual	ADJ
bracis-33663	46	19	token	token	VERB
bracis-33663	46	20	weights	weight	NOUN
bracis-33663	46	21	alongside	alongside	ADP
bracis-33663	46	22	roberta	roberta	PROPN
bracis-33663	46	23	large	large	ADJ
bracis-33663	46	24	mnli	mnli	NOUN
bracis-33663	46	25	to	to	PART
bracis-33663	46	26	enhance	enhance	VERB
bracis-33663	46	27	their	their	PRON
bracis-33663	46	28	models	model	NOUN
bracis-33663	46	29	’	'	PUNCT
bracis-33663	46	30	zshsdfootnote	zshsdfootnote	NOUN
bracis-33663	46	31	1	1	NUM
bracis-33663	46	32	.	.	PUNCT
bracis-33663	47	1	second	second	ADJ
bracis-33663	47	2	:	:	PUNCT
bracis-33663	47	3	they	they	PRON
bracis-33663	47	4	can	can	AUX
bracis-33663	47	5	employ	employ	VERB
bracis-33663	47	6	selected	select	VERB
bracis-33663	47	7	model	model	NOUN
bracis-33663	47	8	tokens	token	NOUN
bracis-33663	47	9	from	from	ADP
bracis-33663	47	10	our	our	PRON
bracis-33663	47	11	study	study	NOUN
bracis-33663	47	12	to	to	PART
bracis-33663	47	13	adapt	adapt	VERB
bracis-33663	47	14	their	their	PRON
bracis-33663	47	15	models	model	NOUN
bracis-33663	47	16	for	for	ADP
bracis-33663	47	17	improved	improved	ADJ
bracis-33663	47	18	classification	classification	NOUN
bracis-33663	47	19	.	.	PUNCT
bracis-33663	48	1	lastly	lastly	ADV
bracis-33663	48	2	:	:	PUNCT
bracis-33663	48	3	our	our	PRON
bracis-33663	48	4	research	research	NOUN
bracis-33663	48	5	encourages	encourage	VERB
bracis-33663	48	6	the	the	DET
bracis-33663	48	7	creation	creation	NOUN
bracis-33663	48	8	of	of	ADP
bracis-33663	48	9	hypotheses	hypothesis	NOUN
bracis-33663	48	10	based	base	VERB
bracis-33663	48	11	on	on	ADP
bracis-33663	48	12	concepts	concept	NOUN
bracis-33663	48	13	’	'	PUNCT
bracis-33663	48	14	mean	mean	VERB
bracis-33663	48	15	,	,	PUNCT
bracis-33663	48	16	expanding	expand	VERB
bracis-33663	48	17	the	the	DET
bracis-33663	48	18	method	method	NOUN
bracis-33663	48	19	’s	’s	PART
bracis-33663	48	20	applicability	applicability	NOUN
bracis-33663	48	21	.	.	PUNCT
bracis-33663	49	1	2	2	NUM
bracis-33663	49	2	background	background	NOUN
bracis-33663	49	3	the	the	DET
bracis-33663	49	4	evaluation	evaluation	NOUN
bracis-33663	49	5	of	of	ADP
bracis-33663	49	6	hate	hate	NOUN
bracis-33663	49	7	speech	speech	NOUN
bracis-33663	49	8	detection	detection	NOUN
bracis-33663	49	9	models	model	NOUN
bracis-33663	49	10	often	often	ADV
bracis-33663	49	11	suffers	suffer	VERB
bracis-33663	49	12	from	from	ADP
bracis-33663	49	13	overestimation	overestimation	NOUN
bracis-33663	49	14	issues	issue	NOUN
bracis-33663	49	15	,	,	PUNCT
bracis-33663	49	16	impacting	impact	VERB
bracis-33663	49	17	both	both	PRON
bracis-33663	49	18	state	state	NOUN
bracis-33663	49	19	-	-	PUNCT
bracis-33663	49	20	of	of	ADP
bracis-33663	49	21	-	-	PUNCT
bracis-33663	49	22	the	the	DET
bracis-33663	49	23	-	-	PUNCT
bracis-33663	49	24	art	art	NOUN
bracis-33663	49	25	(	(	PUNCT
bracis-33663	49	26	sota	sota	NOUN
bracis-33663	49	27	)	)	PUNCT
bracis-33663	49	28	and	and	CCONJ
bracis-33663	49	29	baseline	baseline	NOUN
bracis-33663	49	30	models	model	NOUN
bracis-33663	49	31	.	.	PUNCT
bracis-33663	50	1	when	when	SCONJ
bracis-33663	50	2	tested	test	VERB
bracis-33663	50	3	on	on	ADP
bracis-33663	50	4	different	different	ADJ
bracis-33663	50	5	datasets	dataset	NOUN
bracis-33663	50	6	,	,	PUNCT
bracis-33663	50	7	these	these	DET
bracis-33663	50	8	models	model	NOUN
bracis-33663	50	9	typically	typically	ADV
bracis-33663	50	10	show	show	VERB
bracis-33663	50	11	reduced	reduced	ADJ
bracis-33663	50	12	performance	performance	NOUN
bracis-33663	50	13	,	,	PUNCT
bracis-33663	50	14	highlighting	highlight	VERB
bracis-33663	50	15	a	a	DET
bracis-33663	50	16	gap	gap	NOUN
bracis-33663	50	17	between	between	ADP
bracis-33663	50	18	test	test	NOUN
bracis-33663	50	19	sets	set	NOUN
bracis-33663	50	20	and	and	CCONJ
bracis-33663	50	21	real	real	ADJ
bracis-33663	50	22	-	-	PUNCT
bracis-33663	50	23	world	world	NOUN
bracis-33663	50	24	data	datum	NOUN
bracis-33663	50	25	[	[	X
bracis-33663	50	26	1	1	NUM
bracis-33663	50	27	,	,	PUNCT
bracis-33663	50	28	5	5	NUM
bracis-33663	50	29	,	,	PUNCT
bracis-33663	50	30	18	18	NUM
bracis-33663	50	31	,	,	PUNCT
bracis-33663	50	32	23	23	NUM
bracis-33663	50	33	,	,	PUNCT
bracis-33663	50	34	24	24	NUM
bracis-33663	50	35	]	]	PUNCT
bracis-33663	50	36	.	.	PUNCT
bracis-33663	51	1	various	various	ADJ
bracis-33663	51	2	approaches	approach	NOUN
bracis-33663	51	3	to	to	PART
bracis-33663	51	4	hate	hate	VERB
bracis-33663	51	5	speech	speech	NOUN
bracis-33663	51	6	detection	detection	NOUN
bracis-33663	51	7	have	have	AUX
bracis-33663	51	8	been	be	AUX
bracis-33663	51	9	explored	explore	VERB
bracis-33663	51	10	,	,	PUNCT
bracis-33663	51	11	including	include	VERB
bracis-33663	51	12	traditional	traditional	ADJ
bracis-33663	51	13	machine	machine	NOUN
bracis-33663	51	14	learning	learning	NOUN
bracis-33663	51	15	models	model	NOUN
bracis-33663	51	16	such	such	ADJ
bracis-33663	51	17	as	as	ADP
bracis-33663	51	18	character	character	NOUN
bracis-33663	51	19	n	n	CCONJ
bracis-33663	51	20	-	-	PUNCT
bracis-33663	51	21	gram	gram	NOUN
bracis-33663	51	22	logistic	logistic	ADJ
bracis-33663	51	23	regression	regression	NOUN
bracis-33663	51	24	[	[	X
bracis-33663	51	25	8	8	NUM
bracis-33663	51	26	]	]	PUNCT
bracis-33663	51	27	,	,	PUNCT
bracis-33663	51	28	support	support	VERB
bracis-33663	51	29	vector	vector	NOUN
bracis-33663	51	30	machines	machine	NOUN
bracis-33663	51	31	[	[	X
bracis-33663	51	32	17	17	NUM
bracis-33663	51	33	]	]	PUNCT
bracis-33663	51	34	,	,	PUNCT
bracis-33663	51	35	and	and	CCONJ
bracis-33663	51	36	shallow	shallow	ADJ
bracis-33663	51	37	networks	network	NOUN
bracis-33663	51	38	with	with	ADP
bracis-33663	51	39	pre	pre	ADJ
bracis-33663	51	40	-	-	ADJ
bracis-33663	51	41	trained	train	VERB
bracis-33663	51	42	embeddings	embedding	NOUN
bracis-33663	51	43	like	like	ADP
bracis-33663	51	44	mlp	mlp	NOUN
bracis-33663	51	45	with	with	ADP
bracis-33663	51	46	byte	byte	NOUN
bracis-33663	51	47	-	-	PUNCT
bracis-33663	51	48	pair	pair	NOUN
bracis-33663	51	49	encoding	encoding	NOUN
bracis-33663	51	50	(	(	PUNCT
bracis-33663	51	51	bpe	bpe	PROPN
bracis-33663	51	52	)	)	PUNCT
bracis-33663	52	1	[	[	X
bracis-33663	52	2	22	22	NUM
bracis-33663	52	3	]	]	PUNCT
bracis-33663	52	4	.	.	PUNCT
bracis-33663	53	1	generally	generally	ADV
bracis-33663	53	2	,	,	PUNCT
bracis-33663	53	3	these	these	DET
bracis-33663	53	4	simpler	simple	ADJ
bracis-33663	53	5	models	model	NOUN
bracis-33663	53	6	underperform	underperform	NOUN
bracis-33663	53	7	compared	compare	VERB
bracis-33663	53	8	to	to	ADP
bracis-33663	53	9	deep	deep	ADJ
bracis-33663	53	10	neural	neural	ADJ
bracis-33663	53	11	networks	network	NOUN
bracis-33663	53	12	.	.	PUNCT
bracis-33663	54	1	prior	prior	ADV
bracis-33663	54	2	to	to	ADP
bracis-33663	54	3	2019	2019	NUM
bracis-33663	54	4	,	,	PUNCT
bracis-33663	54	5	recurrent	recurrent	ADJ
bracis-33663	54	6	neural	neural	ADJ
bracis-33663	54	7	networks	network	NOUN
bracis-33663	54	8	were	be	AUX
bracis-33663	54	9	commonly	commonly	ADV
bracis-33663	54	10	used	use	VERB
bracis-33663	54	11	as	as	ADP
bracis-33663	54	12	sota	sota	NOUN
bracis-33663	54	13	models	model	NOUN
bracis-33663	54	14	[	[	X
bracis-33663	54	15	8	8	NUM
bracis-33663	54	16	]	]	PUNCT
bracis-33663	54	17	.	.	PUNCT
bracis-33663	55	1	the	the	DET
bracis-33663	55	2	introduction	introduction	NOUN
bracis-33663	55	3	of	of	ADP
bracis-33663	55	4	bert	bert	PROPN
bracis-33663	55	5	(	(	PUNCT
bracis-33663	55	6	bidirectional	bidirectional	ADJ
bracis-33663	55	7	encoder	encoder	NOUN
bracis-33663	55	8	representations	representation	VERB
bracis-33663	55	9	from	from	ADP
bracis-33663	55	10	transformers	transformer	NOUN
bracis-33663	55	11	)	)	PUNCT
bracis-33663	55	12	by	by	ADP
bracis-33663	55	13	[	[	X
bracis-33663	55	14	3	3	NUM
bracis-33663	55	15	]	]	PUNCT
bracis-33663	55	16	marked	mark	VERB
bracis-33663	55	17	a	a	DET
bracis-33663	55	18	significant	significant	ADJ
bracis-33663	55	19	advancement	advancement	NOUN
bracis-33663	55	20	,	,	PUNCT
bracis-33663	55	21	establishing	establish	VERB
bracis-33663	55	22	bert	bert	PROPN
bracis-33663	55	23	and	and	CCONJ
bracis-33663	55	24	its	its	PRON
bracis-33663	55	25	variants	variant	NOUN
bracis-33663	55	26	as	as	ADP
bracis-33663	55	27	the	the	DET
bracis-33663	55	28	new	new	ADJ
bracis-33663	55	29	sota	sota	NOUN
bracis-33663	55	30	in	in	ADP
bracis-33663	55	31	hate	hate	NOUN
bracis-33663	55	32	speech	speech	NOUN
bracis-33663	55	33	detection	detection	NOUN
bracis-33663	55	34	[	[	X
bracis-33663	55	35	4	4	NUM
bracis-33663	55	36	,	,	PUNCT
bracis-33663	55	37	6	6	NUM
bracis-33663	55	38	]	]	PUNCT
bracis-33663	55	39	.	.	PUNCT
bracis-33663	56	1	although	although	SCONJ
bracis-33663	56	2	bert	bert	PROPN
bracis-33663	56	3	and	and	CCONJ
bracis-33663	56	4	its	its	PRON
bracis-33663	56	5	derivatives	derivative	NOUN
bracis-33663	56	6	have	have	AUX
bracis-33663	56	7	improved	improve	VERB
bracis-33663	56	8	in	in	ADP
bracis-33663	56	9	capturing	capture	VERB
bracis-33663	56	10	contextual	contextual	ADJ
bracis-33663	56	11	information	information	NOUN
bracis-33663	56	12	and	and	CCONJ
bracis-33663	56	13	linguistic	linguistic	ADJ
bracis-33663	56	14	nuances	nuance	NOUN
bracis-33663	56	15	,	,	PUNCT
bracis-33663	56	16	our	our	PRON
bracis-33663	56	17	study	study	NOUN
bracis-33663	56	18	focuses	focus	VERB
bracis-33663	56	19	on	on	ADP
bracis-33663	56	20	nli	nli	PROPN
bracis-33663	56	21	(	(	PUNCT
bracis-33663	56	22	natural	natural	ADJ
bracis-33663	56	23	language	language	NOUN
bracis-33663	56	24	inference	inference	NOUN
bracis-33663	56	25	)	)	PUNCT
bracis-33663	56	26	models	model	NOUN
bracis-33663	56	27	.	.	PUNCT
bracis-33663	57	1	this	this	DET
bracis-33663	57	2	choice	choice	NOUN
bracis-33663	57	3	is	be	AUX
bracis-33663	57	4	based	base	VERB
bracis-33663	57	5	on	on	ADP
bracis-33663	57	6	the	the	DET
bracis-33663	57	7	rationale	rationale	NOUN
bracis-33663	57	8	that	that	SCONJ
bracis-33663	57	9	explicit	explicit	ADJ
bracis-33663	57	10	logic	logic	NOUN
bracis-33663	57	11	learning	learn	VERB
bracis-33663	57	12	through	through	ADP
bracis-33663	57	13	textual	textual	ADJ
bracis-33663	57	14	entailment	entailment	NOUN
bracis-33663	57	15	can	can	AUX
bracis-33663	57	16	reduce	reduce	VERB
bracis-33663	57	17	bias	bias	NOUN
bracis-33663	57	18	and	and	CCONJ
bracis-33663	57	19	enhance	enhance	VERB
bracis-33663	57	20	the	the	DET
bracis-33663	57	21	recognition	recognition	NOUN
bracis-33663	57	22	of	of	ADP
bracis-33663	57	23	social	social	ADJ
bracis-33663	57	24	communities	community	NOUN
bracis-33663	57	25	[	[	X
bracis-33663	57	26	12	12	NUM
bracis-33663	57	27	]	]	PUNCT
bracis-33663	57	28	.	.	PUNCT
bracis-33663	58	1	nli	nli	PROPN
bracis-33663	58	2	models	model	NOUN
bracis-33663	58	3	excel	excel	VERB
bracis-33663	58	4	in	in	ADP
bracis-33663	58	5	hate	hate	NOUN
bracis-33663	58	6	speech	speech	NOUN
bracis-33663	58	7	detection	detection	NOUN
bracis-33663	58	8	by	by	ADP
bracis-33663	58	9	leveraging	leverage	VERB
bracis-33663	58	10	logical	logical	ADJ
bracis-33663	58	11	training	training	NOUN
bracis-33663	58	12	,	,	PUNCT
bracis-33663	58	13	which	which	PRON
bracis-33663	58	14	improves	improve	VERB
bracis-33663	58	15	context	context	NOUN
bracis-33663	58	16	-	-	PUNCT
bracis-33663	58	17	aware	aware	ADJ
bracis-33663	58	18	zero	zero	NUM
bracis-33663	58	19	-	-	PUNCT
bracis-33663	58	20	shot	shot	NOUN
bracis-33663	58	21	hate	hate	NOUN
bracis-33663	58	22	speech	speech	NOUN
bracis-33663	58	23	detection	detection	NOUN
bracis-33663	58	24	(	(	PUNCT
bracis-33663	58	25	zshsd	zshsd	NOUN
bracis-33663	58	26	)	)	PUNCT
bracis-33663	58	27	systems	system	NOUN
bracis-33663	58	28	.	.	PUNCT
bracis-33663	59	1	unlike	unlike	ADP
bracis-33663	59	2	pre	pre	ADJ
bracis-33663	59	3	-	-	ADJ
bracis-33663	59	4	trained	train	VERB
bracis-33663	59	5	sentence	sentence	NOUN
bracis-33663	59	6	encoders	encoder	NOUN
bracis-33663	59	7	that	that	PRON
bracis-33663	59	8	may	may	AUX
bracis-33663	59	9	perpetuate	perpetuate	VERB
bracis-33663	59	10	stereotypes	stereotype	NOUN
bracis-33663	59	11	,	,	PUNCT
bracis-33663	59	12	nli	nli	PROPN
bracis-33663	59	13	models	model	NOUN
bracis-33663	59	14	help	help	VERB
bracis-33663	59	15	mitigate	mitigate	VERB
bracis-33663	59	16	bias	bias	NOUN
bracis-33663	59	17	present	present	ADJ
bracis-33663	59	18	in	in	ADP
bracis-33663	59	19	training	training	NOUN
bracis-33663	59	20	data	datum	NOUN
bracis-33663	59	21	[	[	X
bracis-33663	59	22	14	14	NUM
bracis-33663	59	23	]	]	PUNCT
bracis-33663	59	24	.	.	PUNCT
bracis-33663	60	1	using	use	VERB
bracis-33663	60	2	nli	nli	PROPN
bracis-33663	60	3	models	model	NOUN
bracis-33663	60	4	for	for	ADP
bracis-33663	60	5	predictive	predictive	ADJ
bracis-33663	60	6	tasks	task	NOUN
bracis-33663	60	7	involves	involve	VERB
bracis-33663	60	8	transforming	transform	VERB
bracis-33663	60	9	the	the	DET
bracis-33663	60	10	target	target	NOUN
bracis-33663	60	11	task	task	NOUN
bracis-33663	60	12	into	into	ADP
bracis-33663	60	13	an	an	DET
bracis-33663	60	14	nli	nli	ADJ
bracis-33663	60	15	format	format	NOUN
bracis-33663	60	16	,	,	PUNCT
bracis-33663	60	17	essentially	essentially	ADV
bracis-33663	60	18	a	a	DET
bracis-33663	60	19	fine	fine	ADV
bracis-33663	60	20	-	-	PUNCT
bracis-33663	60	21	tuning	tune	VERB
bracis-33663	60	22	task	task	NOUN
bracis-33663	60	23	.	.	PUNCT
bracis-33663	61	1	the	the	DET
bracis-33663	61	2	model	model	NOUN
bracis-33663	61	3	assesses	assess	VERB
bracis-33663	61	4	a	a	DET
bracis-33663	61	5	premise	premise	NOUN
bracis-33663	61	6	and	and	CCONJ
bracis-33663	61	7	a	a	DET
bracis-33663	61	8	hypothesis	hypothesis	NOUN
bracis-33663	61	9	to	to	PART
bracis-33663	61	10	determine	determine	VERB
bracis-33663	61	11	their	their	PRON
bracis-33663	61	12	logical	logical	ADJ
bracis-33663	61	13	relationship	relationship	NOUN
bracis-33663	61	14	—	—	PUNCT
bracis-33663	61	15	whether	whether	SCONJ
bracis-33663	61	16	the	the	DET
bracis-33663	61	17	premise	premise	NOUN
bracis-33663	61	18	implies	imply	VERB
bracis-33663	61	19	,	,	PUNCT
bracis-33663	61	20	contradicts	contradict	VERB
bracis-33663	61	21	,	,	PUNCT
bracis-33663	61	22	or	or	CCONJ
bracis-33663	61	23	is	be	AUX
bracis-33663	61	24	neutral	neutral	ADJ
bracis-33663	61	25	to	to	ADP
bracis-33663	61	26	the	the	DET
bracis-33663	61	27	hypothesis	hypothesis	NOUN
bracis-33663	61	28	.	.	PUNCT
bracis-33663	62	1	a	a	DET
bracis-33663	62	2	recent	recent	ADJ
bracis-33663	62	3	method	method	NOUN
bracis-33663	62	4	applies	apply	VERB
bracis-33663	62	5	nli	nli	PROPN
bracis-33663	62	6	models	model	NOUN
bracis-33663	62	7	for	for	ADP
bracis-33663	62	8	zero	zero	NUM
bracis-33663	62	9	-	-	PUNCT
bracis-33663	62	10	shot	shot	NOUN
bracis-33663	62	11	topic	topic	NOUN
bracis-33663	62	12	classification	classification	NOUN
bracis-33663	62	13	,	,	PUNCT
bracis-33663	62	14	where	where	SCONJ
bracis-33663	62	15	the	the	DET
bracis-33663	62	16	input	input	NOUN
bracis-33663	62	17	text	text	NOUN
bracis-33663	62	18	serves	serve	VERB
bracis-33663	62	19	as	as	ADP
bracis-33663	62	20	the	the	DET
bracis-33663	62	21	premise	premise	NOUN
bracis-33663	62	22	and	and	CCONJ
bracis-33663	62	23	each	each	DET
bracis-33663	62	24	topic	topic	NOUN
bracis-33663	62	25	has	have	VERB
bracis-33663	62	26	a	a	DET
bracis-33663	62	27	corresponding	corresponding	ADJ
bracis-33663	62	28	hypothesis	hypothesis	NOUN
bracis-33663	62	29	like	like	ADP
bracis-33663	62	30	“	"	PUNCT
bracis-33663	62	31	this	this	DET
bracis-33663	62	32	text	text	NOUN
bracis-33663	62	33	is	be	AUX
bracis-33663	62	34	about	about	ADP
bracis-33663	62	35	<	<	X
bracis-33663	62	36	topic	topic	NOUN
bracis-33663	62	37	>	>	X
bracis-33663	62	38	.	.	PUNCT
bracis-33663	62	39	”	"	PUNCT
bracis-33663	63	1	here	here	ADV
bracis-33663	63	2	,	,	PUNCT
bracis-33663	63	3	“	"	PUNCT
bracis-33663	63	4	neutral	neutral	ADJ
bracis-33663	63	5	”	"	PUNCT
bracis-33663	63	6	and	and	CCONJ
bracis-33663	63	7	“	"	PUNCT
bracis-33663	63	8	contradiction	contradiction	NOUN
bracis-33663	63	9	”	"	PUNCT
bracis-33663	63	10	labels	label	NOUN
bracis-33663	63	11	merge	merge	VERB
bracis-33663	63	12	into	into	ADP
bracis-33663	63	13	“	"	PUNCT
bracis-33663	63	14	not	not	PART
bracis-33663	63	15	-	-	PUNCT
bracis-33663	63	16	entailment	entailment	NOUN
bracis-33663	63	17	,	,	PUNCT
bracis-33663	63	18	”	"	PUNCT
bracis-33663	63	19	meaning	mean	VERB
bracis-33663	63	20	an	an	DET
bracis-33663	63	21	“	"	PUNCT
bracis-33663	63	22	entailment	entailment	ADJ
bracis-33663	63	23	”	"	PUNCT
bracis-33663	63	24	prediction	prediction	NOUN
bracis-33663	63	25	indicates	indicate	VERB
bracis-33663	63	26	relevance	relevance	NOUN
bracis-33663	63	27	to	to	ADP
bracis-33663	63	28	the	the	DET
bracis-33663	63	29	topic	topic	NOUN
bracis-33663	63	30	,	,	PUNCT
bracis-33663	63	31	while	while	SCONJ
bracis-33663	63	32	“	"	PUNCT
bracis-33663	63	33	not	not	PART
bracis-33663	63	34	-	-	PUNCT
bracis-33663	63	35	entailment	entailment	NOUN
bracis-33663	63	36	”	"	PUNCT
bracis-33663	63	37	suggests	suggest	VERB
bracis-33663	63	38	irrelevance	irrelevance	NOUN
bracis-33663	63	39	.	.	PUNCT
bracis-33663	64	1	it	it	PRON
bracis-33663	64	2	’s	’	VERB
bracis-33663	64	3	important	important	ADJ
bracis-33663	64	4	to	to	PART
bracis-33663	64	5	note	note	VERB
bracis-33663	64	6	that	that	SCONJ
bracis-33663	64	7	an	an	DET
bracis-33663	64	8	entailment	entailment	ADJ
bracis-33663	64	9	prediction	prediction	NOUN
bracis-33663	64	10	indicates	indicate	VERB
bracis-33663	64	11	alignment	alignment	NOUN
bracis-33663	64	12	with	with	ADP
bracis-33663	64	13	certain	certain	ADJ
bracis-33663	64	14	model	model	NOUN
bracis-33663	64	15	artifacts	artifact	NOUN
bracis-33663	64	16	,	,	PUNCT
bracis-33663	64	17	not	not	PART
bracis-33663	64	18	necessarily	necessarily	ADV
bracis-33663	64	19	correctness	correctness	NOUN
bracis-33663	64	20	[	[	X
bracis-33663	64	21	25	25	NUM
bracis-33663	64	22	]	]	PUNCT
bracis-33663	64	23	.	.	PUNCT
bracis-33663	65	1	3	3	NUM
bracis-33663	65	2	proposed	propose	VERB
bracis-33663	65	3	approach	approach	NOUN
bracis-33663	65	4	our	our	PRON
bracis-33663	65	5	method	method	NOUN
bracis-33663	65	6	focused	focus	VERB
bracis-33663	65	7	on	on	ADP
bracis-33663	65	8	identifying	identify	VERB
bracis-33663	65	9	the	the	DET
bracis-33663	65	10	most	most	ADV
bracis-33663	65	11	effective	effective	ADJ
bracis-33663	65	12	hypothesis	hypothesis	NOUN
bracis-33663	65	13	for	for	ADP
bracis-33663	65	14	assessing	assess	VERB
bracis-33663	65	15	the	the	DET
bracis-33663	65	16	existence	existence	NOUN
bracis-33663	65	17	of	of	ADP
bracis-33663	65	18	hate	hate	NOUN
bracis-33663	65	19	speech	speech	NOUN
bracis-33663	65	20	in	in	ADP
bracis-33663	65	21	a	a	DET
bracis-33663	65	22	given	give	VERB
bracis-33663	65	23	premise	premise	NOUN
bracis-33663	65	24	.	.	PUNCT
bracis-33663	66	1	rather	rather	ADV
bracis-33663	66	2	than	than	ADP
bracis-33663	66	3	initiating	initiate	VERB
bracis-33663	66	4	the	the	DET
bracis-33663	66	5	optimization	optimization	NOUN
bracis-33663	66	6	process	process	NOUN
bracis-33663	66	7	from	from	ADP
bracis-33663	66	8	entirely	entirely	ADV
bracis-33663	66	9	ran’dom	ran’dom	NOUN
bracis-33663	66	10	weights	weight	NOUN
bracis-33663	66	11	,	,	PUNCT
bracis-33663	66	12	we	we	PRON
bracis-33663	66	13	leveraged	leverage	VERB
bracis-33663	66	14	the	the	DET
bracis-33663	66	15	best	well	ADV
bracis-33663	66	16	-	-	PUNCT
bracis-33663	66	17	performing	perform	VERB
bracis-33663	66	18	results	result	NOUN
bracis-33663	66	19	obtained	obtain	VERB
bracis-33663	66	20	from	from	ADP
bracis-33663	66	21	prior	prior	ADJ
bracis-33663	66	22	experiments	experiment	NOUN
bracis-33663	66	23	,	,	PUNCT
bracis-33663	66	24	as	as	SCONJ
bracis-33663	66	25	seen	see	VERB
bracis-33663	66	26	in	in	ADP
bracis-33663	66	27	table	table	NOUN
bracis-33663	66	28	 	 	SPACE
bracis-33663	66	29	2	2	NUM
bracis-33663	66	30	.	.	PUNCT
bracis-33663	67	1	this	this	DET
bracis-33663	67	2	strategic	strategic	ADJ
bracis-33663	67	3	starting	starting	NOUN
bracis-33663	67	4	point	point	NOUN
bracis-33663	67	5	aimed	aim	VERB
bracis-33663	67	6	to	to	PART
bracis-33663	67	7	expedite	expedite	VERB
bracis-33663	67	8	the	the	DET
bracis-33663	67	9	optimization	optimization	NOUN
bracis-33663	67	10	process	process	NOUN
bracis-33663	67	11	and	and	CCONJ
bracis-33663	67	12	build	build	VERB
bracis-33663	67	13	upon	upon	SCONJ
bracis-33663	67	14	existing	exist	VERB
bracis-33663	67	15	knowledge	knowledge	NOUN
bracis-33663	67	16	.	.	PUNCT
bracis-33663	68	1	3.1	3.1	NUM
bracis-33663	68	2	hypothesis	hypothesis	NOUN
bracis-33663	68	3	optimization	optimization	NOUN
bracis-33663	68	4	in	in	ADP
bracis-33663	68	5	the	the	DET
bracis-33663	68	6	realm	realm	NOUN
bracis-33663	68	7	of	of	ADP
bracis-33663	68	8	nli	nli	PROPN
bracis-33663	68	9	models	model	NOUN
bracis-33663	68	10	,	,	PUNCT
bracis-33663	68	11	a	a	DET
bracis-33663	68	12	hypothesis	hypothesis	NOUN
bracis-33663	68	13	is	be	AUX
bracis-33663	68	14	a	a	DET
bracis-33663	68	15	sentence	sentence	NOUN
bracis-33663	68	16	,	,	PUNCT
bracis-33663	68	17	that	that	PRON
bracis-33663	68	18	is	be	AUX
bracis-33663	68	19	encoded	encode	VERB
bracis-33663	68	20	in	in	ADP
bracis-33663	68	21	a	a	DET
bracis-33663	68	22	sequence	sequence	NOUN
bracis-33663	68	23	of	of	ADP
bracis-33663	68	24	tokens	token	NOUN
bracis-33663	68	25	.	.	PUNCT
bracis-33663	69	1	each	each	DET
bracis-33663	69	2	token	token	NOUN
bracis-33663	69	3	is	be	AUX
bracis-33663	69	4	represented	represent	VERB
bracis-33663	69	5	as	as	ADP
bracis-33663	69	6	a	a	DET
bracis-33663	69	7	integer	integer	NOUN
bracis-33663	69	8	and	and	CCONJ
bracis-33663	69	9	,	,	PUNCT
bracis-33663	69	10	after	after	ADP
bracis-33663	69	11	passing	pass	VERB
bracis-33663	69	12	thought	think	VERB
bracis-33663	69	13	the	the	DET
bracis-33663	69	14	models	model	NOUN
bracis-33663	69	15	embedding	embed	VERB
bracis-33663	69	16	layer	layer	NOUN
bracis-33663	69	17	,	,	PUNCT
bracis-33663	69	18	every	every	DET
bracis-33663	69	19	token	token	NOUN
bracis-33663	69	20	becomes	become	VERB
bracis-33663	69	21	a	a	DET
bracis-33663	69	22	vector	vector	NOUN
bracis-33663	69	23	.	.	PUNCT
bracis-33663	70	1	when	when	SCONJ
bracis-33663	70	2	working	work	VERB
bracis-33663	70	3	with	with	ADP
bracis-33663	70	4	p	p	NOUN
bracis-33663	70	5	-	-	PUNCT
bracis-33663	70	6	tuning	tuning	NOUN
bracis-33663	70	7	,	,	PUNCT
bracis-33663	70	8	instead	instead	ADV
bracis-33663	70	9	of	of	ADP
bracis-33663	70	10	changing	change	VERB
bracis-33663	70	11	the	the	DET
bracis-33663	70	12	initial	initial	ADJ
bracis-33663	70	13	hypothesis	hypothesis	NOUN
bracis-33663	70	14	sentence	sentence	NOUN
bracis-33663	70	15	,	,	PUNCT
bracis-33663	70	16	the	the	DET
bracis-33663	70	17	continuous	continuous	ADJ
bracis-33663	70	18	vectors	vector	NOUN
bracis-33663	70	19	as	as	SCONJ
bracis-33663	70	20	optimized	optimize	VERB
bracis-33663	70	21	directly	directly	ADV
bracis-33663	70	22	.	.	PUNCT
bracis-33663	71	1	this	this	DET
bracis-33663	71	2	optimization	optimization	NOUN
bracis-33663	71	3	creates	create	VERB
bracis-33663	71	4	what	what	PRON
bracis-33663	71	5	s	s	VERB
bracis-33663	71	6	is	be	AUX
bracis-33663	71	7	called	call	VERB
bracis-33663	71	8	as	as	ADP
bracis-33663	71	9	“	"	PUNCT
bracis-33663	71	10	virtual	virtual	ADJ
bracis-33663	71	11	tokens	token	NOUN
bracis-33663	71	12	”	"	PUNCT
bracis-33663	71	13	,	,	PUNCT
bracis-33663	71	14	since	since	SCONJ
bracis-33663	71	15	then	then	ADV
bracis-33663	71	16	do	do	AUX
bracis-33663	71	17	nt	not	PART
bracis-33663	71	18	perfectly	perfectly	ADV
bracis-33663	71	19	resemble	resemble	VERB
bracis-33663	71	20	a	a	DET
bracis-33663	71	21	proper	proper	ADJ
bracis-33663	71	22	token	token	NOUN
bracis-33663	71	23	anymore	anymore	ADV
bracis-33663	71	24	.	.	PUNCT
bracis-33663	72	1	this	this	PRON
bracis-33663	72	2	means	mean	VERB
bracis-33663	72	3	that	that	SCONJ
bracis-33663	72	4	the	the	DET
bracis-33663	72	5	algorithm	algorithm	NOUN
bracis-33663	72	6	learns	learn	VERB
bracis-33663	72	7	to	to	PART
bracis-33663	72	8	adjust	adjust	VERB
bracis-33663	72	9	the	the	DET
bracis-33663	72	10	continuous	continuous	ADJ
bracis-33663	72	11	vector	vector	NOUN
bracis-33663	72	12	representation	representation	NOUN
bracis-33663	72	13	of	of	ADP
bracis-33663	72	14	initially	initially	ADV
bracis-33663	72	15	passed	pass	VERB
bracis-33663	72	16	hypothesis	hypothesis	NOUN
bracis-33663	72	17	to	to	PART
bracis-33663	72	18	improve	improve	VERB
bracis-33663	72	19	its	its	PRON
bracis-33663	72	20	performance	performance	NOUN
bracis-33663	72	21	in	in	ADP
bracis-33663	72	22	zshsd	zshsd	PROPN
bracis-33663	72	23	.	.	PUNCT
bracis-33663	73	1	a	a	DET
bracis-33663	73	2	general	general	ADJ
bracis-33663	73	3	scheme	scheme	NOUN
bracis-33663	73	4	of	of	ADP
bracis-33663	73	5	the	the	DET
bracis-33663	73	6	process	process	NOUN
bracis-33663	73	7	used	use	VERB
bracis-33663	73	8	to	to	PART
bracis-33663	73	9	optimize	optimize	VERB
bracis-33663	73	10	the	the	DET
bracis-33663	73	11	initial	initial	ADJ
bracis-33663	73	12	hypothesis	hypothesis	NOUN
bracis-33663	73	13	\(h\	\(h\	NOUN
bracis-33663	73	14	)	)	PUNCT
bracis-33663	73	15	and	and	CCONJ
bracis-33663	73	16	test	test	VERB
bracis-33663	73	17	the	the	DET
bracis-33663	73	18	different	different	ADJ
bracis-33663	73	19	combinations	combination	NOUN
bracis-33663	73	20	of	of	ADP
bracis-33663	73	21	hypothesis	hypothesis	NOUN
bracis-33663	73	22	tokens	token	NOUN
bracis-33663	73	23	,	,	PUNCT
bracis-33663	73	24	can	can	AUX
bracis-33663	73	25	be	be	AUX
bracis-33663	73	26	seen	see	VERB
bracis-33663	73	27	in	in	ADP
bracis-33663	73	28	table	table	NOUN
bracis-33663	73	29	 	 	SPACE
bracis-33663	73	30	2	2	NUM
bracis-33663	73	31	.	.	PUNCT
bracis-33663	74	1	let	let	VERB
bracis-33663	74	2	’s	’s	NOUN
bracis-33663	74	3	denote	denote	PROPN
bracis-33663	74	4	\(h\	\(h\	PROPN
bracis-33663	74	5	)	)	PUNCT
bracis-33663	74	6	as	as	ADP
bracis-33663	74	7	initial	initial	ADJ
bracis-33663	74	8	hypothesis	hypothesis	NOUN
bracis-33663	74	9	vectors	vector	NOUN
bracis-33663	74	10	(	(	PUNCT
bracis-33663	74	11	selected	select	VERB
bracis-33663	74	12	from	from	ADP
bracis-33663	74	13	the	the	DET
bracis-33663	74	14	best	good	ADJ
bracis-33663	74	15	results	result	NOUN
bracis-33663	74	16	of	of	ADP
bracis-33663	74	17	table	table	NOUN
bracis-33663	74	18	 	 	SPACE
bracis-33663	74	19	2	2	NUM
bracis-33663	74	20	)	)	PUNCT
bracis-33663	74	21	and	and	CCONJ
bracis-33663	74	22	\(oh\	\(oh\	NOUN
bracis-33663	74	23	)	)	PUNCT
bracis-33663	74	24	as	as	ADP
bracis-33663	74	25	the	the	DET
bracis-33663	74	26	optimized	optimize	VERB
bracis-33663	74	27	hypothesis	hypothesis	NOUN
bracis-33663	74	28	vectors	vector	NOUN
bracis-33663	74	29	.	.	PUNCT
bracis-33663	75	1	the	the	DET
bracis-33663	75	2	learning	learning	NOUN
bracis-33663	75	3	process	process	NOUN
bracis-33663	75	4	involves	involve	VERB
bracis-33663	75	5	adjusting	adjust	VERB
bracis-33663	75	6	the	the	DET
bracis-33663	75	7	weights	weight	NOUN
bracis-33663	75	8	\(\theta	\(\theta	NOUN
bracis-33663	75	9	\	\	NOUN
bracis-33663	75	10	)	)	PUNCT
bracis-33663	75	11	of	of	ADP
bracis-33663	75	12	the	the	DET
bracis-33663	75	13	model	model	NOUN
bracis-33663	75	14	to	to	PART
bracis-33663	75	15	find	find	VERB
bracis-33663	75	16	an	an	DET
bracis-33663	75	17	optimal	optimal	ADJ
bracis-33663	75	18	hypothesis	hypothesis	NOUN
bracis-33663	75	19	vector	vector	NOUN
bracis-33663	75	20	\(oh\	\(oh\	PROPN
bracis-33663	75	21	)	)	PUNCT
bracis-33663	75	22	,	,	PUNCT
bracis-33663	75	23	that	that	PRON
bracis-33663	75	24	minimizes	minimize	VERB
bracis-33663	75	25	a	a	DET
bracis-33663	75	26	specific	specific	ADJ
bracis-33663	75	27	loss	loss	NOUN
bracis-33663	75	28	function	function	NOUN
bracis-33663	75	29	,	,	PUNCT
bracis-33663	75	30	typically	typically	ADV
bracis-33663	75	31	associated	associate	VERB
bracis-33663	75	32	with	with	ADP
bracis-33663	75	33	zshsd	zshsd	PROPN
bracis-33663	75	34	.	.	PUNCT
bracis-33663	76	1	mathematically	mathematically	ADV
bracis-33663	76	2	,	,	PUNCT
bracis-33663	76	3	this	this	PRON
bracis-33663	76	4	can	can	AUX
bracis-33663	76	5	be	be	AUX
bracis-33663	76	6	represented	represent	VERB
bracis-33663	76	7	as	as	ADP
bracis-33663	76	8	:	:	PUNCT
bracis-33663	76	9	$	$	SYM
bracis-33663	76	10	$	$	SYM
bracis-33663	76	11	\begin{aligned	\begin{aligne	VERB
bracis-33663	76	12	}	}	PUNCT
bracis-33663	76	13	oh	oh	INTJ
bracis-33663	76	14	=	=	SYM
bracis-33663	76	15	argmin_{\theta	argmin_{\theta	ADJ
bracis-33663	76	16	}	}	PUNCT
bracis-33663	76	17	l(p	l(p	PROPN
bracis-33663	76	18	,	,	PUNCT
bracis-33663	76	19	h	h	NOUN
bracis-33663	76	20	,	,	PUNCT
bracis-33663	76	21	\theta	\theta	PROPN
bracis-33663	76	22	)	)	PUNCT
bracis-33663	76	23	\end{aligned}$$	\end{aligned}$$	PROPN
bracis-33663	77	1	(	(	PUNCT
bracis-33663	77	2	1	1	NUM
bracis-33663	77	3	)	)	PUNCT
bracis-33663	77	4	here	here	ADV
bracis-33663	77	5	,	,	PUNCT
bracis-33663	77	6	\(l\	\(l\	NOUN
bracis-33663	77	7	)	)	PUNCT
bracis-33663	77	8	represents	represent	VERB
bracis-33663	77	9	the	the	DET
bracis-33663	77	10	loss	loss	NOUN
bracis-33663	77	11	function	function	NOUN
bracis-33663	77	12	,	,	PUNCT
bracis-33663	77	13	\(p\	\(p\	PROPN
bracis-33663	77	14	)	)	PUNCT
bracis-33663	77	15	denotes	denote	VERB
bracis-33663	77	16	the	the	DET
bracis-33663	77	17	task	task	NOUN
bracis-33663	77	18	-	-	PUNCT
bracis-33663	77	19	specific	specific	ADJ
bracis-33663	77	20	premise	premise	NOUN
bracis-33663	77	21	,	,	PUNCT
bracis-33663	77	22	\(h\	\(h\	NOUN
bracis-33663	77	23	)	)	PUNCT
bracis-33663	77	24	signifies	signify	VERB
bracis-33663	77	25	the	the	DET
bracis-33663	77	26	initial	initial	ADJ
bracis-33663	77	27	hypothesis	hypothesis	NOUN
bracis-33663	77	28	vectors	vector	NOUN
bracis-33663	77	29	,	,	PUNCT
bracis-33663	77	30	and	and	CCONJ
bracis-33663	77	31	\(\theta	\(\theta	PROPN
bracis-33663	77	32	\	\	PROPN
bracis-33663	77	33	)	)	PUNCT
bracis-33663	77	34	symbolizes	symbolize	VERB
bracis-33663	77	35	the	the	DET
bracis-33663	77	36	model	model	NOUN
bracis-33663	77	37	’s	’s	PART
bracis-33663	77	38	weights	weight	NOUN
bracis-33663	77	39	.	.	PUNCT
bracis-33663	78	1	the	the	DET
bracis-33663	78	2	objective	objective	NOUN
bracis-33663	78	3	is	be	AUX
bracis-33663	78	4	to	to	PART
bracis-33663	78	5	find	find	VERB
bracis-33663	78	6	the	the	DET
bracis-33663	78	7	optimal	optimal	ADJ
bracis-33663	78	8	set	set	NOUN
bracis-33663	78	9	of	of	ADP
bracis-33663	78	10	weights	weight	NOUN
bracis-33663	78	11	that	that	PRON
bracis-33663	78	12	minimizes	minimize	VERB
bracis-33663	78	13	the	the	DET
bracis-33663	78	14	loss	loss	NOUN
bracis-33663	78	15	function	function	NOUN
bracis-33663	78	16	,	,	PUNCT
bracis-33663	78	17	thereby	thereby	ADV
bracis-33663	78	18	producing	produce	VERB
bracis-33663	78	19	the	the	DET
bracis-33663	78	20	optimized	optimize	VERB
bracis-33663	78	21	hypothesis	hypothesis	NOUN
bracis-33663	78	22	vectors	vector	NOUN
bracis-33663	78	23	\(oh\	\(oh\	PROPN
bracis-33663	78	24	)	)	PUNCT
bracis-33663	78	25	.	.	PUNCT
bracis-33663	79	1	nli	nli	PROPN
bracis-33663	79	2	models	model	NOUN
bracis-33663	79	3	typically	typically	ADV
bracis-33663	79	4	feature	feature	VERB
bracis-33663	79	5	a	a	DET
bracis-33663	79	6	softmax	softmax	NOUN
bracis-33663	79	7	layer	layer	NOUN
bracis-33663	79	8	at	at	ADP
bracis-33663	79	9	the	the	DET
bracis-33663	79	10	end	end	NOUN
bracis-33663	79	11	,	,	PUNCT
bracis-33663	79	12	which	which	PRON
bracis-33663	79	13	serves	serve	VERB
bracis-33663	79	14	to	to	PART
bracis-33663	79	15	produce	produce	VERB
bracis-33663	79	16	three	three	NUM
bracis-33663	79	17	primary	primary	ADJ
bracis-33663	79	18	outputs	output	NOUN
bracis-33663	79	19	denoting	denote	VERB
bracis-33663	79	20	the	the	DET
bracis-33663	79	21	potential	potential	ADJ
bracis-33663	79	22	relationships	relationship	NOUN
bracis-33663	79	23	between	between	ADP
bracis-33663	79	24	a	a	DET
bracis-33663	79	25	given	give	VERB
bracis-33663	79	26	premise	premise	NOUN
bracis-33663	79	27	\(p\	\(p\	PROPN
bracis-33663	79	28	)	)	PUNCT
bracis-33663	79	29	and	and	CCONJ
bracis-33663	79	30	hypothesis	hypothesis	NOUN
bracis-33663	79	31	\(h\	\(h\	PROPN
bracis-33663	79	32	):	):	PUNCT
bracis-33663	79	33	contradiction	contradiction	NOUN
bracis-33663	79	34	,	,	PUNCT
bracis-33663	79	35	neutrality	neutrality	NOUN
bracis-33663	79	36	,	,	PUNCT
bracis-33663	79	37	and	and	CCONJ
bracis-33663	79	38	entailment	entailment	ADJ
bracis-33663	79	39	.	.	PUNCT
bracis-33663	80	1	these	these	DET
bracis-33663	80	2	outputs	output	NOUN
bracis-33663	80	3	,	,	PUNCT
bracis-33663	80	4	akin	akin	ADJ
bracis-33663	80	5	to	to	ADP
bracis-33663	80	6	probabilities	probability	NOUN
bracis-33663	80	7	,	,	PUNCT
bracis-33663	80	8	are	be	AUX
bracis-33663	80	9	normalized	normalize	VERB
bracis-33663	80	10	to	to	PART
bracis-33663	80	11	ensure	ensure	VERB
bracis-33663	80	12	that	that	SCONJ
bracis-33663	80	13	their	their	PRON
bracis-33663	80	14	collective	collective	ADJ
bracis-33663	80	15	sum	sum	NOUN
bracis-33663	80	16	equals	equal	VERB
bracis-33663	80	17	1	1	NUM
bracis-33663	80	18	,	,	PUNCT
bracis-33663	80	19	effectively	effectively	ADV
bracis-33663	80	20	indicating	indicate	VERB
bracis-33663	80	21	the	the	DET
bracis-33663	80	22	likelihood	likelihood	NOUN
bracis-33663	80	23	of	of	ADP
bracis-33663	80	24	each	each	DET
bracis-33663	80	25	possible	possible	ADJ
bracis-33663	80	26	relationship	relationship	NOUN
bracis-33663	80	27	.	.	PUNCT
bracis-33663	81	1	the	the	DET
bracis-33663	81	2	goal	goal	NOUN
bracis-33663	81	3	of	of	ADP
bracis-33663	81	4	training	training	NOUN
bracis-33663	81	5	an	an	DET
bracis-33663	81	6	nli	nli	ADJ
bracis-33663	81	7	model	model	NOUN
bracis-33663	81	8	is	be	AUX
bracis-33663	81	9	to	to	PART
bracis-33663	81	10	minimize	minimize	VERB
bracis-33663	81	11	the	the	DET
bracis-33663	81	12	error	error	NOUN
bracis-33663	81	13	in	in	ADP
bracis-33663	81	14	predicting	predict	VERB
bracis-33663	81	15	these	these	DET
bracis-33663	81	16	probabilities	probability	NOUN
bracis-33663	81	17	,	,	PUNCT
bracis-33663	81	18	ensuring	ensure	VERB
bracis-33663	81	19	that	that	SCONJ
bracis-33663	81	20	the	the	DET
bracis-33663	81	21	model	model	NOUN
bracis-33663	81	22	accurately	accurately	ADV
bracis-33663	81	23	identifies	identify	VERB
bracis-33663	81	24	the	the	DET
bracis-33663	81	25	logical	logical	ADJ
bracis-33663	81	26	relationship	relationship	NOUN
bracis-33663	81	27	between	between	ADP
bracis-33663	81	28	the	the	DET
bracis-33663	81	29	premise	premise	NOUN
bracis-33663	81	30	and	and	CCONJ
bracis-33663	81	31	hypothesis	hypothesis	NOUN
bracis-33663	81	32	.	.	PUNCT
bracis-33663	82	1	this	this	PRON
bracis-33663	82	2	is	be	AUX
bracis-33663	82	3	typically	typically	ADV
bracis-33663	82	4	achieved	achieve	VERB
bracis-33663	82	5	through	through	ADP
bracis-33663	82	6	the	the	DET
bracis-33663	82	7	use	use	NOUN
bracis-33663	82	8	of	of	ADP
bracis-33663	82	9	a	a	DET
bracis-33663	82	10	categorical	categorical	ADJ
bracis-33663	82	11	cross	cross	ADJ
bracis-33663	82	12	-	-	ADJ
bracis-33663	82	13	entropy	entropy	ADJ
bracis-33663	82	14	loss	loss	NOUN
bracis-33663	82	15	\(l\	\(l\	NOUN
bracis-33663	82	16	)	)	PUNCT
bracis-33663	82	17	,	,	PUNCT
bracis-33663	82	18	which	which	PRON
bracis-33663	82	19	is	be	AUX
bracis-33663	82	20	defined	define	VERB
bracis-33663	82	21	as	as	ADP
bracis-33663	82	22	:	:	PUNCT
bracis-33663	82	23	$	$	SYM
bracis-33663	82	24	$	$	SYM
bracis-33663	82	25	\begin{aligned	\begin{aligne	VERB
bracis-33663	82	26	}	}	PUNCT
bracis-33663	82	27	l	l	NOUN
bracis-33663	82	28	=	=	PUNCT
bracis-33663	82	29	-\sum	-\sum	NOUN
bracis-33663	82	30	_	_	PUNCT
bracis-33663	82	31	{	{	PUNCT
bracis-33663	82	32	c=1}^3y_i\log	c=1}^3y_i\log	X
bracis-33663	82	33	(	(	PUNCT
bracis-33663	82	34	p_{i	p_{i	NOUN
bracis-33663	82	35	}	}	PUNCT
bracis-33663	82	36	)	)	PUNCT
bracis-33663	82	37	\end{aligned}$$	\end{aligned}$$	X
bracis-33663	82	38	(	(	PUNCT
bracis-33663	82	39	2	2	NUM
bracis-33663	82	40	)	)	PUNCT
bracis-33663	82	41	in	in	ADP
bracis-33663	82	42	the	the	DET
bracis-33663	82	43	context	context	NOUN
bracis-33663	82	44	of	of	ADP
bracis-33663	82	45	binary	binary	PROPN
bracis-33663	82	46	hate	hate	PROPN
bracis-33663	82	47	speech	speech	NOUN
bracis-33663	82	48	detection	detection	NOUN
bracis-33663	82	49	,	,	PUNCT
bracis-33663	82	50	the	the	DET
bracis-33663	82	51	loss	loss	NOUN
bracis-33663	82	52	function	function	NOUN
bracis-33663	82	53	in	in	ADP
bracis-33663	82	54	this	this	DET
bracis-33663	82	55	case	case	NOUN
bracis-33663	82	56	is	be	AUX
bracis-33663	82	57	optimized	optimize	VERB
bracis-33663	82	58	to	to	PART
bracis-33663	82	59	increase	increase	VERB
bracis-33663	82	60	entailment	entailment	ADJ
bracis-33663	82	61	output	output	NOUN
bracis-33663	82	62	when	when	SCONJ
bracis-33663	82	63	hate	hate	NOUN
bracis-33663	82	64	is	be	AUX
bracis-33663	82	65	present	present	ADJ
bracis-33663	82	66	,	,	PUNCT
bracis-33663	82	67	i.e.	i.e.	X
bracis-33663	82	68	\(c=3\	\(c=3\	NOUN
bracis-33663	82	69	)	)	PUNCT
bracis-33663	82	70	,	,	PUNCT
bracis-33663	82	71	and	and	CCONJ
bracis-33663	82	72	maximize	maximize	VERB
bracis-33663	82	73	contradiction	contradiction	NOUN
bracis-33663	82	74	output	output	NOUN
bracis-33663	82	75	when	when	SCONJ
bracis-33663	82	76	hate	hate	NOUN
bracis-33663	82	77	is	be	AUX
bracis-33663	82	78	not	not	PART
bracis-33663	82	79	present	present	ADJ
bracis-33663	82	80	,	,	PUNCT
bracis-33663	82	81	i.e.	i.e.	X
bracis-33663	82	82	\(c=1\	\(c=1\	NOUN
bracis-33663	82	83	)	)	PUNCT
bracis-33663	82	84	.	.	PUNCT
bracis-33663	83	1	given	give	VERB
bracis-33663	83	2	that	that	DET
bracis-33663	83	3	\(c	\(c	NOUN
bracis-33663	83	4	=	=	PUNCT
bracis-33663	84	1	[	[	X
bracis-33663	84	2	1,2,3]\	1,2,3]\	NUM
bracis-33663	84	3	)	)	PUNCT
bracis-33663	84	4	represents	represent	VERB
bracis-33663	84	5	,	,	PUNCT
bracis-33663	84	6	contradiction	contradiction	NOUN
bracis-33663	84	7	,	,	PUNCT
bracis-33663	84	8	neutrality	neutrality	NOUN
bracis-33663	84	9	and	and	CCONJ
bracis-33663	84	10	entailment	entailment	ADJ
bracis-33663	84	11	,	,	PUNCT
bracis-33663	84	12	respectively	respectively	ADV
bracis-33663	84	13	.	.	PUNCT
bracis-33663	85	1	after	after	ADP
bracis-33663	85	2	optimizing	optimize	VERB
bracis-33663	85	3	our	our	PRON
bracis-33663	85	4	hypothesis	hypothesis	NOUN
bracis-33663	85	5	,	,	PUNCT
bracis-33663	85	6	we	we	PRON
bracis-33663	85	7	proceeded	proceed	VERB
bracis-33663	85	8	with	with	ADP
bracis-33663	85	9	a	a	DET
bracis-33663	85	10	structured	structured	ADJ
bracis-33663	85	11	sequence	sequence	NOUN
bracis-33663	85	12	of	of	ADP
bracis-33663	85	13	experiments	experiment	NOUN
bracis-33663	85	14	to	to	PART
bracis-33663	85	15	assess	assess	VERB
bracis-33663	85	16	both	both	CCONJ
bracis-33663	85	17	the	the	DET
bracis-33663	85	18	individual	individual	ADJ
bracis-33663	85	19	and	and	CCONJ
bracis-33663	85	20	combined	combined	ADJ
bracis-33663	85	21	performance	performance	NOUN
bracis-33663	85	22	of	of	ADP
bracis-33663	85	23	optimized	optimize	VERB
bracis-33663	85	24	tokens	token	NOUN
bracis-33663	85	25	.	.	PUNCT
bracis-33663	86	1	these	these	DET
bracis-33663	86	2	experiments	experiment	NOUN
bracis-33663	86	3	were	be	AUX
bracis-33663	86	4	designed	design	VERB
bracis-33663	86	5	to	to	PART
bracis-33663	86	6	gauge	gauge	VERB
bracis-33663	86	7	the	the	DET
bracis-33663	86	8	efficacy	efficacy	NOUN
bracis-33663	86	9	of	of	ADP
bracis-33663	86	10	our	our	PRON
bracis-33663	86	11	approach	approach	NOUN
bracis-33663	86	12	in	in	ADP
bracis-33663	86	13	diverse	diverse	ADJ
bracis-33663	86	14	contexts	contexts	NOUN
bracis-33663	86	15	,	,	PUNCT
bracis-33663	86	16	thereby	thereby	ADV
bracis-33663	86	17	illuminating	illuminate	VERB
bracis-33663	86	18	its	its	PRON
bracis-33663	86	19	capacity	capacity	NOUN
bracis-33663	86	20	to	to	PART
bracis-33663	86	21	discern	discern	VERB
bracis-33663	86	22	hate	hate	NOUN
bracis-33663	86	23	speech	speech	NOUN
bracis-33663	86	24	.	.	PUNCT
bracis-33663	87	1	in	in	ADP
bracis-33663	87	2	our	our	PRON
bracis-33663	87	3	study	study	NOUN
bracis-33663	87	4	,	,	PUNCT
bracis-33663	87	5	we	we	PRON
bracis-33663	87	6	used	use	VERB
bracis-33663	87	7	the	the	DET
bracis-33663	87	8	roberta	roberta	PROPN
bracis-33663	87	9	large	large	ADJ
bracis-33663	87	10	modelfootnote	modelfootnote	NOUN
bracis-33663	87	11	2	2	NUM
bracis-33663	87	12	[	[	X
bracis-33663	87	13	11	11	NUM
bracis-33663	87	14	]	]	PUNCT
bracis-33663	87	15	,	,	PUNCT
bracis-33663	87	16	which	which	PRON
bracis-33663	87	17	had	have	AUX
bracis-33663	87	18	undergone	undergo	VERB
bracis-33663	87	19	fine	fine	ADJ
bracis-33663	87	20	-	-	PUNCT
bracis-33663	87	21	tuning	tuning	NOUN
bracis-33663	87	22	on	on	ADP
bracis-33663	87	23	a	a	DET
bracis-33663	87	24	diverse	diverse	ADJ
bracis-33663	87	25	range	range	NOUN
bracis-33663	87	26	of	of	ADP
bracis-33663	87	27	text	text	NOUN
bracis-33663	87	28	sources	source	NOUN
bracis-33663	87	29	,	,	PUNCT
bracis-33663	87	30	including	include	VERB
bracis-33663	87	31	wikipedia	wikipedia	PROPN
bracis-33663	87	32	,	,	PUNCT
bracis-33663	87	33	the	the	DET
bracis-33663	87	34	book	book	NOUN
bracis-33663	87	35	corpus	corpus	NOUN
bracis-33663	87	36	,	,	PUNCT
bracis-33663	87	37	and	and	CCONJ
bracis-33663	87	38	the	the	DET
bracis-33663	87	39	multinli	multinli	PROPN
bracis-33663	87	40	dataset	dataset	PROPN
bracis-33663	87	41	.	.	PUNCT
bracis-33663	88	1	the	the	DET
bracis-33663	88	2	training	training	NOUN
bracis-33663	88	3	of	of	ADP
bracis-33663	88	4	this	this	DET
bracis-33663	88	5	model	model	NOUN
bracis-33663	88	6	was	be	AUX
bracis-33663	88	7	conducted	conduct	VERB
bracis-33663	88	8	on	on	ADP
bracis-33663	88	9	an	an	DET
bracis-33663	88	10	nvidia	nvidia	PROPN
bracis-33663	88	11	4090	4090	NUM
bracis-33663	88	12	rtx	rtx	PROPN
bracis-33663	88	13	gpu	gpu	PROPN
bracis-33663	88	14	,	,	PUNCT
bracis-33663	88	15	spanning	span	VERB
bracis-33663	88	16	a	a	DET
bracis-33663	88	17	duration	duration	NOUN
bracis-33663	88	18	of	of	ADP
bracis-33663	88	19	approximately	approximately	ADV
bracis-33663	88	20	12	12	NUM
bracis-33663	88	21	 	 	SPACE
bracis-33663	88	22	h	h	NOUN
bracis-33663	88	23	,	,	PUNCT
bracis-33663	88	24	with	with	ADP
bracis-33663	88	25	a	a	DET
bracis-33663	88	26	training	training	NOUN
bracis-33663	88	27	regimen	regiman	NOUN
bracis-33663	88	28	extending	extend	VERB
bracis-33663	88	29	over	over	ADP
bracis-33663	88	30	20	20	NUM
bracis-33663	88	31	epochs	epoch	NOUN
bracis-33663	88	32	.	.	PUNCT
bracis-33663	89	1	for	for	ADP
bracis-33663	89	2	optimization	optimization	NOUN
bracis-33663	89	3	,	,	PUNCT
bracis-33663	89	4	we	we	PRON
bracis-33663	89	5	employed	employ	VERB
bracis-33663	89	6	the	the	DET
bracis-33663	89	7	adamw	adamw	NOUN
bracis-33663	89	8	optimizer	optimizer	NOUN
bracis-33663	89	9	and	and	CCONJ
bracis-33663	89	10	implemented	implement	VERB
bracis-33663	89	11	a	a	DET
bracis-33663	89	12	linear	linear	ADJ
bracis-33663	89	13	decay	decay	NOUN
bracis-33663	89	14	learning	learn	VERB
bracis-33663	89	15	rate	rate	NOUN
bracis-33663	89	16	schedule	schedule	NOUN
bracis-33663	89	17	,	,	PUNCT
bracis-33663	89	18	setting	set	VERB
bracis-33663	89	19	the	the	DET
bracis-33663	89	20	initial	initial	ADJ
bracis-33663	89	21	learning	learning	NOUN
bracis-33663	89	22	rate	rate	NOUN
bracis-33663	89	23	at	at	ADP
bracis-33663	89	24	1e-3	1e-3	NUM
bracis-33663	89	25	.	.	PUNCT
bracis-33663	90	1	it	it	PRON
bracis-33663	90	2	is	be	AUX
bracis-33663	90	3	noteworthy	noteworthy	ADJ
bracis-33663	90	4	that	that	SCONJ
bracis-33663	90	5	our	our	PRON
bracis-33663	90	6	training	training	NOUN
bracis-33663	90	7	process	process	NOUN
bracis-33663	90	8	required	require	VERB
bracis-33663	90	9	significantly	significantly	ADV
bracis-33663	90	10	less	less	ADJ
bracis-33663	90	11	time	time	NOUN
bracis-33663	90	12	and	and	CCONJ
bracis-33663	90	13	computational	computational	ADJ
bracis-33663	90	14	resources	resource	NOUN
bracis-33663	90	15	in	in	ADP
bracis-33663	90	16	comparison	comparison	NOUN
bracis-33663	90	17	to	to	ADP
bracis-33663	90	18	previous	previous	ADJ
bracis-33663	90	19	work	work	NOUN
bracis-33663	90	20	[	[	X
bracis-33663	90	21	19	19	NUM
bracis-33663	90	22	]	]	PUNCT
bracis-33663	90	23	.	.	PUNCT
bracis-33663	91	1	notably	notably	ADV
bracis-33663	91	2	,	,	PUNCT
bracis-33663	91	3	we	we	PRON
bracis-33663	91	4	optimized	optimize	VERB
bracis-33663	91	5	our	our	PRON
bracis-33663	91	6	approach	approach	NOUN
bracis-33663	91	7	to	to	PART
bracis-33663	91	8	achieve	achieve	VERB
bracis-33663	91	9	efficient	efficient	ADJ
bracis-33663	91	10	resource	resource	NOUN
bracis-33663	91	11	utilization	utilization	NOUN
bracis-33663	91	12	by	by	ADP
bracis-33663	91	13	employing	employ	VERB
bracis-33663	91	14	p	p	NOUN
bracis-33663	91	15	-	-	PUNCT
bracis-33663	91	16	tuning	tuning	NOUN
bracis-33663	91	17	solely	solely	ADV
bracis-33663	91	18	on	on	ADP
bracis-33663	91	19	the	the	DET
bracis-33663	91	20	hypothesis	hypothesis	NOUN
bracis-33663	91	21	,	,	PUNCT
bracis-33663	91	22	in	in	ADP
bracis-33663	91	23	contrast	contrast	NOUN
bracis-33663	91	24	to	to	ADP
bracis-33663	91	25	traditional	traditional	ADJ
bracis-33663	91	26	whole	whole	ADJ
bracis-33663	91	27	-	-	PUNCT
bracis-33663	91	28	model	model	NOUN
bracis-33663	91	29	fine	fine	ADV
bracis-33663	91	30	-	-	PUNCT
bracis-33663	91	31	tuning	tune	VERB
bracis-33663	91	32	methodologies	methodology	NOUN
bracis-33663	91	33	.	.	PUNCT
bracis-33663	92	1	this	this	DET
bracis-33663	92	2	strategic	strategic	ADJ
bracis-33663	92	3	adaptation	adaptation	NOUN
bracis-33663	92	4	likely	likely	ADV
bracis-33663	92	5	contributed	contribute	VERB
bracis-33663	92	6	to	to	ADP
bracis-33663	92	7	the	the	DET
bracis-33663	92	8	reduced	reduce	VERB
bracis-33663	92	9	time	time	NOUN
bracis-33663	92	10	and	and	CCONJ
bracis-33663	92	11	computational	computational	ADJ
bracis-33663	92	12	resources	resource	NOUN
bracis-33663	92	13	required	require	VERB
bracis-33663	92	14	for	for	ADP
bracis-33663	92	15	our	our	PRON
bracis-33663	92	16	training	training	NOUN
bracis-33663	92	17	process	process	NOUN
bracis-33663	92	18	.	.	PUNCT
bracis-33663	93	1	3.2	3.2	NUM
bracis-33663	93	2	hypothesis	hypothesis	NOUN
bracis-33663	93	3	similarity	similarity	NOUN
bracis-33663	93	4	analysis	analysis	NOUN
bracis-33663	93	5	our	our	PRON
bracis-33663	93	6	initial	initial	ADJ
bracis-33663	93	7	phase	phase	NOUN
bracis-33663	93	8	involved	involve	VERB
bracis-33663	93	9	scrutinizing	scrutinize	VERB
bracis-33663	93	10	the	the	DET
bracis-33663	93	11	optimized	optimize	VERB
bracis-33663	93	12	hypothesis	hypothesis	NOUN
bracis-33663	93	13	and	and	CCONJ
bracis-33663	93	14	assessing	assess	VERB
bracis-33663	93	15	its	its	PRON
bracis-33663	93	16	similarity	similarity	NOUN
bracis-33663	93	17	to	to	ADP
bracis-33663	93	18	the	the	DET
bracis-33663	93	19	original	original	ADJ
bracis-33663	93	20	tokens	token	NOUN
bracis-33663	93	21	before	before	ADP
bracis-33663	93	22	optimization	optimization	NOUN
bracis-33663	93	23	.	.	PUNCT
bracis-33663	94	1	this	this	PRON
bracis-33663	94	2	allowed	allow	VERB
bracis-33663	94	3	us	we	PRON
bracis-33663	94	4	to	to	PART
bracis-33663	94	5	gauge	gauge	VERB
bracis-33663	94	6	the	the	DET
bracis-33663	94	7	extent	extent	NOUN
bracis-33663	94	8	to	to	PART
bracis-33663	94	9	which	which	PRON
bracis-33663	94	10	our	our	PRON
bracis-33663	94	11	optimizations	optimization	NOUN
bracis-33663	94	12	retained	retain	VERB
bracis-33663	94	13	the	the	DET
bracis-33663	94	14	core	core	NOUN
bracis-33663	94	15	meaning	meaning	NOUN
bracis-33663	94	16	of	of	ADP
bracis-33663	94	17	the	the	DET
bracis-33663	94	18	original	original	ADJ
bracis-33663	94	19	text	text	NOUN
bracis-33663	94	20	.	.	PUNCT
bracis-33663	95	1	we	we	PRON
bracis-33663	95	2	measured	measure	VERB
bracis-33663	95	3	this	this	DET
bracis-33663	95	4	distance	distance	NOUN
bracis-33663	95	5	using	use	VERB
bracis-33663	95	6	cosine	cosine	NOUN
bracis-33663	95	7	similarity	similarity	NOUN
bracis-33663	95	8	with	with	ADP
bracis-33663	95	9	the	the	DET
bracis-33663	95	10	base	base	NOUN
bracis-33663	95	11	model	model	NOUN
bracis-33663	95	12	’s	’s	PART
bracis-33663	95	13	token	token	VERB
bracis-33663	95	14	embeddings	embedding	NOUN
bracis-33663	95	15	.	.	PUNCT
bracis-33663	96	1	3.3	3.3	NUM
bracis-33663	96	2	token	token	ADJ
bracis-33663	96	3	importance	importance	NOUN
bracis-33663	96	4	assessment	assessment	NOUN
bracis-33663	96	5	in	in	ADP
bracis-33663	96	6	the	the	DET
bracis-33663	96	7	second	second	ADJ
bracis-33663	96	8	phase	phase	NOUN
bracis-33663	96	9	,	,	PUNCT
bracis-33663	96	10	we	we	PRON
bracis-33663	96	11	delved	delve	VERB
bracis-33663	96	12	into	into	ADP
bracis-33663	96	13	the	the	DET
bracis-33663	96	14	importance	importance	NOUN
bracis-33663	96	15	of	of	ADP
bracis-33663	96	16	each	each	DET
bracis-33663	96	17	individual	individual	NOUN
bracis-33663	96	18	optimized	optimize	VERB
bracis-33663	96	19	token	token	VERB
bracis-33663	96	20	.	.	PUNCT
bracis-33663	97	1	we	we	PRON
bracis-33663	97	2	examined	examine	VERB
bracis-33663	97	3	the	the	DET
bracis-33663	97	4	outcomes	outcome	NOUN
bracis-33663	97	5	of	of	ADP
bracis-33663	97	6	using	use	VERB
bracis-33663	97	7	these	these	DET
bracis-33663	97	8	tokens	token	NOUN
bracis-33663	97	9	in	in	ADP
bracis-33663	97	10	three	three	NUM
bracis-33663	97	11	distinct	distinct	ADJ
bracis-33663	97	12	manners	manner	NOUN
bracis-33663	97	13	:	:	PUNCT
bracis-33663	97	14	isolated	isolated	ADJ
bracis-33663	97	15	(	(	PUNCT
bracis-33663	97	16	one	one	NUM
bracis-33663	97	17	at	at	ADP
bracis-33663	97	18	a	a	DET
bracis-33663	97	19	time	time	NOUN
bracis-33663	97	20	)	)	PUNCT
bracis-33663	97	21	,	,	PUNCT
bracis-33663	97	22	inserted	insert	VERB
bracis-33663	97	23	individually	individually	ADV
bracis-33663	97	24	,	,	PUNCT
bracis-33663	97	25	and	and	CCONJ
bracis-33663	97	26	cumulatively	cumulatively	ADV
bracis-33663	97	27	.	.	PUNCT
bracis-33663	98	1	following	follow	VERB
bracis-33663	98	2	the	the	DET
bracis-33663	98	3	idea	idea	NOUN
bracis-33663	98	4	of	of	ADP
bracis-33663	98	5	ch	ch	NOUN
bracis-33663	98	6	presented	present	VERB
bracis-33663	98	7	in	in	ADP
bracis-33663	98	8	fig	fig	NOUN
bracis-33663	98	9	.	.	PUNCT
bracis-33663	98	10	 	 	SPACE
bracis-33663	99	1	1	1	NUM
bracis-33663	99	2	.	.	PUNCT
bracis-33663	99	3	isolated	isolate	VERB
bracis-33663	99	4	(	(	PUNCT
bracis-33663	99	5	token	token	ADJ
bracis-33663	99	6	-	-	PUNCT
bracis-33663	99	7	level	level	NOUN
bracis-33663	99	8	):	):	PUNCT
bracis-33663	99	9	in	in	ADP
bracis-33663	99	10	the	the	DET
bracis-33663	99	11	“	"	PUNCT
bracis-33663	99	12	isolate	isolate	NOUN
bracis-33663	99	13	”	"	PUNCT
bracis-33663	99	14	experiment	experiment	NOUN
bracis-33663	99	15	,	,	PUNCT
bracis-33663	99	16	we	we	PRON
bracis-33663	99	17	isolated	isolate	VERB
bracis-33663	99	18	one	one	NUM
bracis-33663	99	19	token	token	VERB
bracis-33663	99	20	at	at	ADP
bracis-33663	99	21	a	a	DET
bracis-33663	99	22	time	time	NOUN
bracis-33663	99	23	from	from	ADP
bracis-33663	99	24	the	the	DET
bracis-33663	99	25	optimized	optimize	VERB
bracis-33663	99	26	hypothesis	hypothesis	NOUN
bracis-33663	99	27	.	.	PUNCT
bracis-33663	100	1	the	the	DET
bracis-33663	100	2	primary	primary	ADJ
bracis-33663	100	3	objective	objective	NOUN
bracis-33663	100	4	was	be	AUX
bracis-33663	100	5	to	to	PART
bracis-33663	100	6	determine	determine	VERB
bracis-33663	100	7	whether	whether	SCONJ
bracis-33663	100	8	the	the	DET
bracis-33663	100	9	isolated	isolated	ADJ
bracis-33663	100	10	tokens	token	NOUN
bracis-33663	100	11	exhibited	exhibit	VERB
bracis-33663	100	12	improved	improved	ADJ
bracis-33663	100	13	performance	performance	NOUN
bracis-33663	100	14	,	,	PUNCT
bracis-33663	100	15	particularly	particularly	ADV
bracis-33663	100	16	concerning	concern	VERB
bracis-33663	100	17	terms	term	NOUN
bracis-33663	100	18	associated	associate	VERB
bracis-33663	100	19	with	with	ADP
bracis-33663	100	20	hate	hate	NOUN
bracis-33663	100	21	speech	speech	NOUN
bracis-33663	100	22	,	,	PUNCT
bracis-33663	100	23	such	such	ADJ
bracis-33663	100	24	as	as	ADP
bracis-33663	100	25	“	"	PUNCT
bracis-33663	100	26	hateful	hateful	ADJ
bracis-33663	100	27	”	"	PUNCT
bracis-33663	100	28	and	and	CCONJ
bracis-33663	100	29	“	"	PUNCT
bracis-33663	100	30	offensive	offensive	ADJ
bracis-33663	100	31	.	.	PUNCT
bracis-33663	100	32	”	"	PUNCT
bracis-33663	101	1	individual	individual	ADJ
bracis-33663	101	2	(	(	PUNCT
bracis-33663	101	3	contextual	contextual	ADJ
bracis-33663	101	4	insertion	insertion	NOUN
bracis-33663	101	5	):	):	PUNCT
bracis-33663	101	6	in	in	ADP
bracis-33663	101	7	the	the	DET
bracis-33663	101	8	“	"	PUNCT
bracis-33663	101	9	individual	individual	ADJ
bracis-33663	101	10	”	"	PUNCT
bracis-33663	101	11	experiment	experiment	NOUN
bracis-33663	101	12	,	,	PUNCT
bracis-33663	101	13	once	once	SCONJ
bracis-33663	101	14	the	the	DET
bracis-33663	101	15	tokens	token	NOUN
bracis-33663	101	16	have	have	AUX
bracis-33663	101	17	been	be	AUX
bracis-33663	101	18	optimized	optimize	VERB
bracis-33663	101	19	in	in	ADP
bracis-33663	101	20	this	this	DET
bracis-33663	101	21	new	new	ADJ
bracis-33663	101	22	continuous	continuous	ADJ
bracis-33663	101	23	space	space	NOUN
bracis-33663	101	24	,	,	PUNCT
bracis-33663	101	25	they	they	PRON
bracis-33663	101	26	are	be	AUX
bracis-33663	101	27	reinserted	reinserte	VERB
bracis-33663	101	28	back	back	ADV
bracis-33663	101	29	into	into	ADP
bracis-33663	101	30	the	the	DET
bracis-33663	101	31	standard	standard	ADJ
bracis-33663	101	32	hypothesis	hypothesis	NOUN
bracis-33663	101	33	structure	structure	NOUN
bracis-33663	101	34	.	.	PUNCT
bracis-33663	102	1	this	this	PRON
bracis-33663	102	2	means	mean	VERB
bracis-33663	102	3	replacing	replace	VERB
bracis-33663	102	4	the	the	DET
bracis-33663	102	5	original	original	ADJ
bracis-33663	102	6	tokens	token	NOUN
bracis-33663	102	7	with	with	ADP
bracis-33663	102	8	their	their	PRON
bracis-33663	102	9	tuned	tune	VERB
bracis-33663	102	10	representations	representation	NOUN
bracis-33663	102	11	in	in	ADP
bracis-33663	102	12	the	the	DET
bracis-33663	102	13	hypothesis	hypothesis	NOUN
bracis-33663	102	14	.	.	PUNCT
bracis-33663	103	1	let	let	VERB
bracis-33663	103	2	’s	’s	NOUN
bracis-33663	103	3	say	say	VERB
bracis-33663	103	4	the	the	DET
bracis-33663	103	5	token	token	ADJ
bracis-33663	103	6	“	"	PUNCT
bracis-33663	103	7	hateful	hateful	ADJ
bracis-33663	103	8	”	"	PUNCT
bracis-33663	103	9	is	be	AUX
bracis-33663	103	10	initially	initially	ADV
bracis-33663	103	11	represented	represent	VERB
bracis-33663	103	12	as	as	ADP
bracis-33663	103	13	\([0.2	\([0.2	PROPN
bracis-33663	103	14	,	,	PUNCT
bracis-33663	103	15	0.5	0.5	NUM
bracis-33663	103	16	,	,	PUNCT
bracis-33663	103	17	-0.3]\	-0.3]\	NUM
bracis-33663	103	18	)	)	PUNCT
bracis-33663	103	19	,	,	PUNCT
bracis-33663	103	20	and	and	CCONJ
bracis-33663	103	21	after	after	ADP
bracis-33663	103	22	tuning	tune	VERB
bracis-33663	103	23	,	,	PUNCT
bracis-33663	103	24	it	it	PRON
bracis-33663	103	25	becomes	become	VERB
bracis-33663	103	26	\([0.8	\([0.8	ADP
bracis-33663	103	27	,	,	PUNCT
bracis-33663	103	28	-0.1	-0.1	PROPN
bracis-33663	103	29	,	,	PUNCT
bracis-33663	103	30	0.7]\	0.7]\	NUM
bracis-33663	103	31	)	)	PUNCT
bracis-33663	103	32	.	.	PUNCT
bracis-33663	104	1	after	after	ADP
bracis-33663	104	2	reinserting	reinserte	VERB
bracis-33663	104	3	the	the	DET
bracis-33663	104	4	tuned	tuned	ADJ
bracis-33663	104	5	tokens	token	NOUN
bracis-33663	104	6	,	,	PUNCT
bracis-33663	104	7	the	the	DET
bracis-33663	104	8	hypothesis	hypothesis	NOUN
bracis-33663	104	9	may	may	AUX
bracis-33663	104	10	look	look	VERB
bracis-33663	104	11	like	like	ADP
bracis-33663	104	12	this	this	PRON
bracis-33663	104	13	:	:	PUNCT
bracis-33663	104	14	“	"	PUNCT
bracis-33663	104	15	this	this	DET
bracis-33663	104	16	sentence	sentence	NOUN
bracis-33663	104	17	contains	contain	VERB
bracis-33663	104	18	\([{hateful}]\	\([{hateful}]\	NOUN
bracis-33663	104	19	)	)	PUNCT
bracis-33663	104	20	entries	entry	NOUN
bracis-33663	104	21	or	or	CCONJ
bracis-33663	104	22	offensive	offensive	ADJ
bracis-33663	104	23	language	language	NOUN
bracis-33663	104	24	”	"	PUNCT
bracis-33663	104	25	“	"	PUNCT
bracis-33663	104	26	this	this	DET
bracis-33663	104	27	sentence	sentence	NOUN
bracis-33663	104	28	contains	contain	VERB
bracis-33663	104	29	\([{hateful}]\	\([{hateful}]\	NOUN
bracis-33663	104	30	)	)	PUNCT
bracis-33663	104	31	entries	entry	NOUN
bracis-33663	104	32	or	or	CCONJ
bracis-33663	104	33	\([{offensive}]\	\([{offensive}]\	NOUN
bracis-33663	104	34	)	)	PUNCT
bracis-33663	104	35	\([{language}]\	\([{language}]\	NOUN
bracis-33663	104	36	)	)	PUNCT
bracis-33663	104	37	”	"	PUNCT
bracis-33663	104	38	here	here	ADV
bracis-33663	104	39	,	,	PUNCT
bracis-33663	104	40	\([{hateful}]\	\([{hateful}]\	PROPN
bracis-33663	104	41	)	)	PUNCT
bracis-33663	104	42	,	,	PUNCT
bracis-33663	104	43	\([{offensive}]\	\([{offensive}]\	NOUN
bracis-33663	104	44	)	)	PUNCT
bracis-33663	104	45	and	and	CCONJ
bracis-33663	104	46	\([{language}]\	\([{language}]\	NOUN
bracis-33663	104	47	)	)	PUNCT
bracis-33663	104	48	represent	represent	VERB
bracis-33663	104	49	the	the	DET
bracis-33663	104	50	transformed	transform	VERB
bracis-33663	104	51	versions	version	NOUN
bracis-33663	104	52	of	of	ADP
bracis-33663	104	53	the	the	DET
bracis-33663	104	54	original	original	ADJ
bracis-33663	104	55	tokens	token	NOUN
bracis-33663	104	56	“	"	PUNCT
bracis-33663	104	57	hateful	hateful	ADJ
bracis-33663	104	58	”	"	PUNCT
bracis-33663	104	59	,	,	PUNCT
bracis-33663	104	60	“	"	PUNCT
bracis-33663	104	61	offensive	offensive	ADJ
bracis-33663	104	62	”	"	PUNCT
bracis-33663	104	63	and	and	CCONJ
bracis-33663	104	64	“	"	PUNCT
bracis-33663	104	65	language	language	NOUN
bracis-33663	104	66	”	"	PUNCT
bracis-33663	104	67	.	.	PUNCT
bracis-33663	105	1	the	the	DET
bracis-33663	105	2	motivation	motivation	NOUN
bracis-33663	105	3	behind	behind	ADP
bracis-33663	105	4	this	this	DET
bracis-33663	105	5	approach	approach	NOUN
bracis-33663	105	6	is	be	AUX
bracis-33663	105	7	to	to	PART
bracis-33663	105	8	leverage	leverage	VERB
bracis-33663	105	9	a	a	DET
bracis-33663	105	10	more	more	ADV
bracis-33663	105	11	nuanced	nuanced	ADJ
bracis-33663	105	12	and	and	CCONJ
bracis-33663	105	13	contextually	contextually	ADV
bracis-33663	105	14	aware	aware	ADJ
bracis-33663	105	15	representation	representation	NOUN
bracis-33663	105	16	of	of	ADP
bracis-33663	105	17	the	the	DET
bracis-33663	105	18	key	key	ADJ
bracis-33663	105	19	terms	term	NOUN
bracis-33663	105	20	in	in	ADP
bracis-33663	105	21	the	the	DET
bracis-33663	105	22	hypothesis	hypothesis	NOUN
bracis-33663	105	23	.	.	PUNCT
bracis-33663	106	1	by	by	ADP
bracis-33663	106	2	doing	do	VERB
bracis-33663	106	3	so	so	ADV
bracis-33663	106	4	,	,	PUNCT
bracis-33663	106	5	this	this	DET
bracis-33663	106	6	study	study	NOUN
bracis-33663	106	7	aims	aim	VERB
bracis-33663	106	8	to	to	PART
bracis-33663	106	9	measure	measure	VERB
bracis-33663	106	10	whether	whether	SCONJ
bracis-33663	106	11	this	this	DET
bracis-33663	106	12	enhanced	enhance	VERB
bracis-33663	106	13	understanding	understanding	NOUN
bracis-33663	106	14	of	of	ADP
bracis-33663	106	15	specific	specific	ADJ
bracis-33663	106	16	language	language	NOUN
bracis-33663	106	17	cues	cue	NOUN
bracis-33663	106	18	can	can	AUX
bracis-33663	106	19	lead	lead	VERB
bracis-33663	106	20	to	to	ADP
bracis-33663	106	21	improved	improved	ADJ
bracis-33663	106	22	performance	performance	NOUN
bracis-33663	106	23	in	in	ADP
bracis-33663	106	24	hate	hate	NOUN
bracis-33663	106	25	speech	speech	NOUN
bracis-33663	106	26	detection	detection	NOUN
bracis-33663	106	27	as	as	ADP
bracis-33663	106	28	a	a	DET
bracis-33663	106	29	nli	nli	PROPN
bracis-33663	106	30	tasks	task	NOUN
bracis-33663	106	31	.	.	PUNCT
bracis-33663	107	1	the	the	DET
bracis-33663	107	2	idea	idea	NOUN
bracis-33663	107	3	is	be	AUX
bracis-33663	107	4	that	that	SCONJ
bracis-33663	107	5	the	the	DET
bracis-33663	107	6	tuned	tune	VERB
bracis-33663	107	7	representations	representation	NOUN
bracis-33663	107	8	may	may	AUX
bracis-33663	107	9	capture	capture	VERB
bracis-33663	107	10	subtleties	subtlety	NOUN
bracis-33663	107	11	and	and	CCONJ
bracis-33663	107	12	context	context	VERB
bracis-33663	107	13	better	well	ADV
bracis-33663	107	14	than	than	ADP
bracis-33663	107	15	the	the	DET
bracis-33663	107	16	original	original	ADJ
bracis-33663	107	17	,	,	PUNCT
bracis-33663	107	18	static	static	ADJ
bracis-33663	107	19	tokens	token	NOUN
bracis-33663	107	20	.	.	PUNCT
bracis-33663	108	1	cumulative	cumulative	ADJ
bracis-33663	108	2	(	(	PUNCT
bracis-33663	108	3	sequential	sequential	ADJ
bracis-33663	108	4	insertion	insertion	NOUN
bracis-33663	108	5	):	):	PUNCT
bracis-33663	108	6	the	the	DET
bracis-33663	108	7	“	"	PUNCT
bracis-33663	108	8	cumulative	cumulative	ADJ
bracis-33663	108	9	”	"	PUNCT
bracis-33663	108	10	experiment	experiment	NOUN
bracis-33663	108	11	involved	involve	VERB
bracis-33663	108	12	the	the	DET
bracis-33663	108	13	incremental	incremental	ADJ
bracis-33663	108	14	addition	addition	NOUN
bracis-33663	108	15	of	of	ADP
bracis-33663	108	16	the	the	DET
bracis-33663	108	17	optimized	optimize	VERB
bracis-33663	108	18	tokens	token	NOUN
bracis-33663	108	19	in	in	ADP
bracis-33663	108	20	the	the	DET
bracis-33663	108	21	natural	natural	ADJ
bracis-33663	108	22	order	order	NOUN
bracis-33663	108	23	of	of	ADP
bracis-33663	108	24	the	the	DET
bracis-33663	108	25	sentence	sentence	NOUN
bracis-33663	108	26	.	.	PUNCT
bracis-33663	109	1	while	while	SCONJ
bracis-33663	109	2	we	we	PRON
bracis-33663	109	3	anticipated	anticipate	VERB
bracis-33663	109	4	an	an	DET
bracis-33663	109	5	overall	overall	ADJ
bracis-33663	109	6	improvement	improvement	NOUN
bracis-33663	109	7	,	,	PUNCT
bracis-33663	109	8	the	the	DET
bracis-33663	109	9	objective	objective	NOUN
bracis-33663	109	10	here	here	ADV
bracis-33663	109	11	was	be	AUX
bracis-33663	109	12	to	to	PART
bracis-33663	109	13	analyze	analyze	VERB
bracis-33663	109	14	performance	performance	NOUN
bracis-33663	109	15	peaks	peak	NOUN
bracis-33663	109	16	corresponding	correspond	VERB
bracis-33663	109	17	to	to	ADP
bracis-33663	109	18	the	the	DET
bracis-33663	109	19	introduction	introduction	NOUN
bracis-33663	109	20	of	of	ADP
bracis-33663	109	21	each	each	DET
bracis-33663	109	22	new	new	ADJ
bracis-33663	109	23	optimized	optimize	VERB
bracis-33663	109	24	token	token	VERB
bracis-33663	109	25	.	.	PUNCT
bracis-33663	110	1	this	this	DET
bracis-33663	110	2	step	step	NOUN
bracis-33663	110	3	provided	provide	VERB
bracis-33663	110	4	insights	insight	NOUN
bracis-33663	110	5	into	into	ADP
bracis-33663	110	6	the	the	DET
bracis-33663	110	7	cumulative	cumulative	ADJ
bracis-33663	110	8	impact	impact	NOUN
bracis-33663	110	9	of	of	ADP
bracis-33663	110	10	our	our	PRON
bracis-33663	110	11	method	method	NOUN
bracis-33663	110	12	on	on	ADP
bracis-33663	110	13	zshsd	zshsd	PROPN
bracis-33663	110	14	.	.	PUNCT
bracis-33663	111	1	3.4	3.4	NUM
bracis-33663	111	2	benchmark	benchmark	NOUN
bracis-33663	111	3	in	in	ADP
bracis-33663	111	4	the	the	DET
bracis-33663	111	5	final	final	ADJ
bracis-33663	111	6	phase	phase	NOUN
bracis-33663	111	7	of	of	ADP
bracis-33663	111	8	our	our	PRON
bracis-33663	111	9	experimentation	experimentation	NOUN
bracis-33663	111	10	,	,	PUNCT
bracis-33663	111	11	our	our	PRON
bracis-33663	111	12	focus	focus	NOUN
bracis-33663	111	13	shifted	shift	VERB
bracis-33663	111	14	towards	towards	ADP
bracis-33663	111	15	comparative	comparative	ADJ
bracis-33663	111	16	analysis	analysis	NOUN
bracis-33663	111	17	.	.	PUNCT
bracis-33663	112	1	specifically	specifically	ADV
bracis-33663	112	2	,	,	PUNCT
bracis-33663	112	3	we	we	PRON
bracis-33663	112	4	sought	seek	VERB
bracis-33663	112	5	to	to	PART
bracis-33663	112	6	assess	assess	VERB
bracis-33663	112	7	the	the	DET
bracis-33663	112	8	efficacy	efficacy	NOUN
bracis-33663	112	9	of	of	ADP
bracis-33663	112	10	our	our	PRON
bracis-33663	112	11	methodology	methodology	NOUN
bracis-33663	112	12	,	,	PUNCT
bracis-33663	112	13	which	which	PRON
bracis-33663	112	14	incorporated	incorporate	VERB
bracis-33663	112	15	the	the	DET
bracis-33663	112	16	optimized	optimize	VERB
bracis-33663	112	17	hypothesis	hypothesis	NOUN
bracis-33663	112	18	and	and	CCONJ
bracis-33663	112	19	tokens	token	NOUN
bracis-33663	112	20	,	,	PUNCT
bracis-33663	112	21	in	in	ADP
bracis-33663	112	22	contrast	contrast	NOUN
bracis-33663	112	23	to	to	ADP
bracis-33663	112	24	established	establish	VERB
bracis-33663	112	25	benchmarks	benchmark	NOUN
bracis-33663	112	26	as	as	SCONJ
bracis-33663	112	27	presented	present	VERB
bracis-33663	112	28	in	in	ADP
bracis-33663	112	29	prior	prior	ADJ
bracis-33663	112	30	studies	study	NOUN
bracis-33663	112	31	[	[	X
bracis-33663	112	32	7	7	NUM
bracis-33663	112	33	,	,	PUNCT
bracis-33663	112	34	16	16	NUM
bracis-33663	112	35	,	,	PUNCT
bracis-33663	112	36	19	19	NUM
bracis-33663	112	37	]	]	PUNCT
bracis-33663	112	38	.	.	PUNCT
bracis-33663	113	1	dataset	dataset	PROPN
bracis-33663	113	2	.	.	PUNCT
bracis-33663	114	1	the	the	DET
bracis-33663	114	2	dataset	dataset	NOUN
bracis-33663	114	3	utilized	utilize	VERB
bracis-33663	114	4	in	in	ADP
bracis-33663	114	5	this	this	DET
bracis-33663	114	6	study	study	NOUN
bracis-33663	114	7	comprises	comprise	VERB
bracis-33663	114	8	two	two	NUM
bracis-33663	114	9	distinct	distinct	ADJ
bracis-33663	114	10	sources	source	NOUN
bracis-33663	114	11	,	,	PUNCT
bracis-33663	114	12	each	each	PRON
bracis-33663	114	13	contributing	contribute	VERB
bracis-33663	114	14	valuable	valuable	ADJ
bracis-33663	114	15	insights	insight	NOUN
bracis-33663	114	16	into	into	ADP
bracis-33663	114	17	zshsd	zshsd	PROPN
bracis-33663	114	18	.	.	PUNCT
bracis-33663	115	1	in	in	ADP
bracis-33663	115	2	aggregate	aggregate	NOUN
bracis-33663	115	3	,	,	PUNCT
bracis-33663	115	4	the	the	DET
bracis-33663	115	5	total	total	ADJ
bracis-33663	115	6	dataset	dataset	NOUN
bracis-33663	115	7	employed	employ	VERB
bracis-33663	115	8	in	in	ADP
bracis-33663	115	9	our	our	PRON
bracis-33663	115	10	research	research	NOUN
bracis-33663	115	11	comprises	comprise	VERB
bracis-33663	115	12	40,772	40,772	NUM
bracis-33663	115	13	instances	instance	NOUN
bracis-33663	115	14	.	.	PUNCT
bracis-33663	116	1	it	it	PRON
bracis-33663	116	2	consists	consist	VERB
bracis-33663	116	3	of	of	ADP
bracis-33663	116	4	22,571	22,571	NUM
bracis-33663	116	5	instances	instance	NOUN
bracis-33663	116	6	labeled	label	VERB
bracis-33663	116	7	as	as	ADP
bracis-33663	116	8	“	"	PUNCT
bracis-33663	116	9	hateful	hateful	ADJ
bracis-33663	116	10	”	"	PUNCT
bracis-33663	116	11	and	and	CCONJ
bracis-33663	116	12	18,201	18,201	NUM
bracis-33663	116	13	instances	instance	NOUN
bracis-33663	116	14	categorized	categorize	VERB
bracis-33663	116	15	as	as	ADP
bracis-33663	116	16	“	"	PUNCT
bracis-33663	116	17	non	non	ADJ
bracis-33663	116	18	-	-	ADJ
bracis-33663	116	19	hateful	hateful	ADJ
bracis-33663	116	20	.	.	PUNCT
bracis-33663	116	21	”	"	PUNCT
bracis-33663	117	1	this	this	DET
bracis-33663	117	2	diverse	diverse	ADJ
bracis-33663	117	3	and	and	CCONJ
bracis-33663	117	4	multifaceted	multifaceted	ADJ
bracis-33663	117	5	dataset	dataset	NOUN
bracis-33663	117	6	serves	serve	NOUN
bracis-33663	117	7	as	as	ADP
bracis-33663	117	8	a	a	DET
bracis-33663	117	9	foundation	foundation	NOUN
bracis-33663	117	10	for	for	ADP
bracis-33663	117	11	our	our	PRON
bracis-33663	117	12	explorations	exploration	NOUN
bracis-33663	117	13	into	into	ADP
bracis-33663	117	14	hate	hate	NOUN
bracis-33663	117	15	speech	speech	NOUN
bracis-33663	117	16	detection	detection	NOUN
bracis-33663	117	17	,	,	PUNCT
bracis-33663	117	18	allowing	allow	VERB
bracis-33663	117	19	us	we	PRON
bracis-33663	117	20	to	to	PART
bracis-33663	117	21	analyze	analyze	VERB
bracis-33663	117	22	and	and	CCONJ
bracis-33663	117	23	evaluate	evaluate	VERB
bracis-33663	117	24	the	the	DET
bracis-33663	117	25	performance	performance	NOUN
bracis-33663	117	26	of	of	ADP
bracis-33663	117	27	the	the	DET
bracis-33663	117	28	method	method	NOUN
bracis-33663	117	29	across	across	ADP
bracis-33663	117	30	a	a	DET
bracis-33663	117	31	wide	wide	ADJ
bracis-33663	117	32	range	range	NOUN
bracis-33663	117	33	of	of	ADP
bracis-33663	117	34	real	real	ADJ
bracis-33663	117	35	-	-	PUNCT
bracis-33663	117	36	world	world	NOUN
bracis-33663	117	37	hate	hate	NOUN
bracis-33663	117	38	speech	speech	NOUN
bracis-33663	117	39	scenarios	scenario	NOUN
bracis-33663	117	40	(	(	PUNCT
bracis-33663	117	41	table	table	NOUN
bracis-33663	117	42	1	1	NUM
bracis-33663	117	43	)	)	PUNCT
bracis-33663	117	44	.	.	PUNCT
bracis-33663	118	1	table	table	NOUN
bracis-33663	118	2	1	1	NUM
bracis-33663	118	3	.	.	PUNCT
bracis-33663	119	1	dataset	dataset	PROPN
bracis-33663	119	2	data	datum	NOUN
bracis-33663	119	3	sources	source	NOUN
bracis-33663	119	4	and	and	CCONJ
bracis-33663	119	5	number	number	NOUN
bracis-33663	119	6	of	of	ADP
bracis-33663	119	7	examplesfull	examplesfull	ADJ
bracis-33663	119	8	size	size	NOUN
bracis-33663	119	9	table	table	NOUN
bracis-33663	119	10	learning	learn	VERB
bracis-33663	119	11	from	from	ADP
bracis-33663	119	12	the	the	DET
bracis-33663	119	13	worst	bad	ADJ
bracis-33663	119	14	.	.	PUNCT
bracis-33663	120	1	[	[	X
bracis-33663	120	2	19	19	NUM
bracis-33663	120	3	]	]	PUNCT
bracis-33663	120	4	is	be	AUX
bracis-33663	120	5	a	a	DET
bracis-33663	120	6	process	process	NOUN
bracis-33663	120	7	to	to	PART
bracis-33663	120	8	create	create	VERB
bracis-33663	120	9	datasets	dataset	NOUN
bracis-33663	120	10	for	for	ADP
bracis-33663	120	11	training	training	NOUN
bracis-33663	120	12	better	well	ADJ
bracis-33663	120	13	hate	hate	VERB
bracis-33663	120	14	detection	detection	NOUN
bracis-33663	120	15	models	model	NOUN
bracis-33663	120	16	.	.	PUNCT
bracis-33663	121	1	it	it	PRON
bracis-33663	121	2	uses	use	VERB
bracis-33663	121	3	both	both	DET
bracis-33663	121	4	humans	human	NOUN
bracis-33663	121	5	and	and	CCONJ
bracis-33663	121	6	models	model	NOUN
bracis-33663	121	7	.	.	PUNCT
bracis-33663	122	1	the	the	DET
bracis-33663	122	2	dataset	dataset	NOUN
bracis-33663	122	3	has	have	VERB
bracis-33663	122	4	37,044	37,044	NUM
bracis-33663	122	5	instances	instance	NOUN
bracis-33663	122	6	:	:	PUNCT
bracis-33663	122	7	20,008	20,008	NUM
bracis-33663	122	8	are	be	AUX
bracis-33663	122	9	labeled	label	VERB
bracis-33663	122	10	“	"	PUNCT
bracis-33663	122	11	hateful	hateful	ADJ
bracis-33663	122	12	”	"	PUNCT
bracis-33663	122	13	,	,	PUNCT
bracis-33663	122	14	and	and	CCONJ
bracis-33663	122	15	17,036	17,036	NUM
bracis-33663	122	16	are	be	AUX
bracis-33663	122	17	“	"	PUNCT
bracis-33663	122	18	non	non	ADJ
bracis-33663	122	19	-	-	ADJ
bracis-33663	122	20	hateful	hateful	ADJ
bracis-33663	122	21	.	.	PUNCT
bracis-33663	122	22	”	"	PUNCT
bracis-33663	123	1	we	we	PRON
bracis-33663	123	2	followed	follow	VERB
bracis-33663	123	3	the	the	DET
bracis-33663	123	4	authors	author	NOUN
bracis-33663	123	5	’	’	PART
bracis-33663	123	6	setup	setup	NOUN
bracis-33663	123	7	,	,	PUNCT
bracis-33663	123	8	training	train	VERB
bracis-33663	123	9	the	the	DET
bracis-33663	123	10	model	model	NOUN
bracis-33663	123	11	on	on	ADP
bracis-33663	123	12	the	the	DET
bracis-33663	123	13	four	four	NUM
bracis-33663	123	14	rounds	round	NOUN
bracis-33663	123	15	made	make	VERB
bracis-33663	123	16	available	available	ADJ
bracis-33663	123	17	by	by	ADP
bracis-33663	123	18	the	the	DET
bracis-33663	123	19	dataset	dataset	NOUN
bracis-33663	123	20	(	(	PUNCT
bracis-33663	123	21	r1	r1	NOUN
bracis-33663	123	22	,	,	PUNCT
bracis-33663	123	23	r2	r2	PROPN
bracis-33663	123	24	,	,	PUNCT
bracis-33663	123	25	r3	r3	PROPN
bracis-33663	123	26	,	,	PUNCT
bracis-33663	123	27	and	and	CCONJ
bracis-33663	123	28	r4	r4	NOUN
bracis-33663	123	29	)	)	PUNCT
bracis-33663	123	30	and	and	CCONJ
bracis-33663	123	31	testing	testing	NOUN
bracis-33663	123	32	on	on	ADP
bracis-33663	123	33	each	each	DET
bracis-33663	123	34	rounds	round	NOUN
bracis-33663	123	35	splits	split	VERB
bracis-33663	123	36	separately	separately	ADV
bracis-33663	123	37	.	.	PUNCT
bracis-33663	124	1	hatecheck	hatecheck	NOUN
bracis-33663	124	2	.	.	PUNCT
bracis-33663	125	1	[	[	X
bracis-33663	125	2	16	16	NUM
bracis-33663	125	3	]	]	PUNCT
bracis-33663	125	4	is	be	AUX
bracis-33663	125	5	a	a	DET
bracis-33663	125	6	suite	suite	NOUN
bracis-33663	125	7	of	of	ADP
bracis-33663	125	8	tests	test	NOUN
bracis-33663	125	9	for	for	ADP
bracis-33663	125	10	evaluating	evaluate	VERB
bracis-33663	125	11	hate	hate	NOUN
bracis-33663	125	12	speech	speech	NOUN
bracis-33663	125	13	detection	detection	NOUN
bracis-33663	125	14	models	model	NOUN
bracis-33663	125	15	.	.	PUNCT
bracis-33663	126	1	it	it	PRON
bracis-33663	126	2	has	have	VERB
bracis-33663	126	3	29	29	NUM
bracis-33663	126	4	test	test	NOUN
bracis-33663	126	5	functionalities	functionality	NOUN
bracis-33663	126	6	based	base	VERB
bracis-33663	126	7	on	on	ADP
bracis-33663	126	8	previous	previous	ADJ
bracis-33663	126	9	research	research	NOUN
bracis-33663	126	10	and	and	CCONJ
bracis-33663	126	11	civil	civil	ADJ
bracis-33663	126	12	society	society	NOUN
bracis-33663	126	13	input	input	NOUN
bracis-33663	126	14	.	.	PUNCT
bracis-33663	127	1	the	the	DET
bracis-33663	127	2	dataset	dataset	NOUN
bracis-33663	127	3	contains	contain	VERB
bracis-33663	127	4	3,728	3,728	NUM
bracis-33663	127	5	instances	instance	NOUN
bracis-33663	127	6	:	:	PUNCT
bracis-33663	127	7	2,563	2,563	NUM
bracis-33663	127	8	are	be	AUX
bracis-33663	127	9	“	"	PUNCT
bracis-33663	127	10	hateful	hateful	ADJ
bracis-33663	127	11	,	,	PUNCT
bracis-33663	127	12	”	"	PUNCT
bracis-33663	127	13	and	and	CCONJ
bracis-33663	127	14	1,165	1,165	NUM
bracis-33663	127	15	are	be	AUX
bracis-33663	127	16	“	"	PUNCT
bracis-33663	127	17	non	non	ADJ
bracis-33663	127	18	-	-	ADJ
bracis-33663	127	19	hateful	hateful	ADJ
bracis-33663	127	20	.	.	PUNCT
bracis-33663	127	21	”	"	PUNCT
bracis-33663	128	1	we	we	PRON
bracis-33663	128	2	used	use	VERB
bracis-33663	128	3	the	the	DET
bracis-33663	128	4	best	good	ADJ
bracis-33663	128	5	model	model	NOUN
bracis-33663	128	6	accuracy	accuracy	NOUN
bracis-33663	128	7	scores	score	NOUN
bracis-33663	128	8	for	for	ADP
bracis-33663	128	9	individual	individual	ADJ
bracis-33663	128	10	tests	test	NOUN
bracis-33663	128	11	based	base	VERB
bracis-33663	128	12	on	on	ADP
bracis-33663	128	13	the	the	DET
bracis-33663	128	14	original	original	ADJ
bracis-33663	128	15	work	work	NOUN
bracis-33663	128	16	[	[	X
bracis-33663	128	17	16	16	NUM
bracis-33663	128	18	]	]	PUNCT
bracis-33663	128	19	.	.	PUNCT
bracis-33663	129	1	for	for	ADP
bracis-33663	129	2	the	the	DET
bracis-33663	129	3	average	average	ADJ
bracis-33663	129	4	best	good	ADJ
bracis-33663	129	5	performance	performance	NOUN
bracis-33663	129	6	we	we	PRON
bracis-33663	129	7	used	use	VERB
bracis-33663	129	8	the	the	DET
bracis-33663	129	9	accuracy	accuracy	NOUN
bracis-33663	129	10	of	of	ADP
bracis-33663	129	11	“	"	PUNCT
bracis-33663	129	12	all	all	PRON
bracis-33663	129	13	”	"	PUNCT
bracis-33663	129	14	the	the	DET
bracis-33663	129	15	different	different	ADJ
bracis-33663	129	16	hypotheses	hypothesis	NOUN
bracis-33663	129	17	combined	combine	VERB
bracis-33663	129	18	as	as	SCONJ
bracis-33663	129	19	proposed	propose	VERB
bracis-33663	129	20	by	by	ADP
bracis-33663	129	21	[	[	X
bracis-33663	129	22	7	7	NUM
bracis-33663	129	23	]	]	PUNCT
bracis-33663	129	24	.	.	PUNCT
bracis-33663	129	25	4	4	NUM
bracis-33663	129	26	results	result	NOUN
bracis-33663	129	27	through	through	ADP
bracis-33663	129	28	an	an	DET
bracis-33663	129	29	evaluation	evaluation	NOUN
bracis-33663	129	30	of	of	ADP
bracis-33663	129	31	the	the	DET
bracis-33663	129	32	optimized	optimize	VERB
bracis-33663	129	33	hypothesis	hypothesis	NOUN
bracis-33663	129	34	and	and	CCONJ
bracis-33663	129	35	token	token	VERB
bracis-33663	129	36	representations	representation	NOUN
bracis-33663	129	37	,	,	PUNCT
bracis-33663	129	38	as	as	ADV
bracis-33663	129	39	well	well	ADV
bracis-33663	129	40	as	as	ADP
bracis-33663	129	41	their	their	PRON
bracis-33663	129	42	impact	impact	NOUN
bracis-33663	129	43	on	on	ADP
bracis-33663	129	44	zshsd	zshsd	PROPN
bracis-33663	129	45	performance	performance	NOUN
bracis-33663	129	46	,	,	PUNCT
bracis-33663	129	47	we	we	PRON
bracis-33663	129	48	aim	aim	VERB
bracis-33663	129	49	to	to	PART
bracis-33663	129	50	provide	provide	VERB
bracis-33663	129	51	a	a	DET
bracis-33663	129	52	detailed	detailed	ADJ
bracis-33663	129	53	insight	insight	NOUN
bracis-33663	129	54	into	into	ADP
bracis-33663	129	55	the	the	DET
bracis-33663	129	56	capabilities	capability	NOUN
bracis-33663	129	57	and	and	CCONJ
bracis-33663	129	58	limitations	limitation	NOUN
bracis-33663	129	59	of	of	ADP
bracis-33663	129	60	our	our	PRON
bracis-33663	129	61	approach	approach	NOUN
bracis-33663	129	62	.	.	PUNCT
bracis-33663	130	1	4.1	4.1	NUM
bracis-33663	130	2	zero	zero	NUM
bracis-33663	130	3	-	-	PUNCT
bracis-33663	130	4	shoot	shoot	NOUN
bracis-33663	130	5	hate	hate	NOUN
bracis-33663	130	6	speech	speech	NOUN
bracis-33663	130	7	hypothesis	hypothesis	NOUN
bracis-33663	130	8	in	in	ADP
bracis-33663	130	9	this	this	DET
bracis-33663	130	10	hypothesis	hypothesis	NOUN
bracis-33663	130	11	test	test	NOUN
bracis-33663	130	12	,	,	PUNCT
bracis-33663	130	13	we	we	PRON
bracis-33663	130	14	employed	employ	VERB
bracis-33663	130	15	an	an	DET
bracis-33663	130	16	nli	nli	PROPN
bracis-33663	130	17	model	model	NOUN
bracis-33663	130	18	to	to	PART
bracis-33663	130	19	assess	assess	VERB
bracis-33663	130	20	the	the	DET
bracis-33663	130	21	effectiveness	effectiveness	NOUN
bracis-33663	130	22	of	of	ADP
bracis-33663	130	23	various	various	ADJ
bracis-33663	130	24	hypotheses	hypothesis	NOUN
bracis-33663	130	25	in	in	ADP
bracis-33663	130	26	measuring	measure	VERB
bracis-33663	130	27	hate	hate	NOUN
bracis-33663	130	28	speech	speech	NOUN
bracis-33663	130	29	and	and	CCONJ
bracis-33663	130	30	offensive	offensive	ADJ
bracis-33663	130	31	language	language	NOUN
bracis-33663	130	32	.	.	PUNCT
bracis-33663	131	1	table	table	NOUN
bracis-33663	131	2	 	 	SPACE
bracis-33663	131	3	2	2	NUM
bracis-33663	131	4	presents	present	VERB
bracis-33663	131	5	the	the	DET
bracis-33663	131	6	results	result	NOUN
bracis-33663	131	7	,	,	PUNCT
bracis-33663	131	8	including	include	VERB
bracis-33663	131	9	the	the	DET
bracis-33663	131	10	average	average	ADJ
bracis-33663	131	11	performance	performance	NOUN
bracis-33663	131	12	.	.	PUNCT
bracis-33663	132	1	the	the	DET
bracis-33663	132	2	best	good	ADJ
bracis-33663	132	3	hypothesis	hypothesis	NOUN
bracis-33663	132	4	with	with	ADP
bracis-33663	132	5	an	an	DET
bracis-33663	132	6	average	average	ADJ
bracis-33663	132	7	accuracy	accuracy	NOUN
bracis-33663	132	8	result	result	NOUN
bracis-33663	132	9	of	of	ADP
bracis-33663	132	10	0.669	0.669	NUM
bracis-33663	132	11	was	be	AUX
bracis-33663	132	12	“	"	PUNCT
bracis-33663	132	13	this	this	DET
bracis-33663	132	14	sentence	sentence	NOUN
bracis-33663	132	15	contains	contain	VERB
bracis-33663	132	16	hateful	hateful	ADJ
bracis-33663	132	17	entries	entry	NOUN
bracis-33663	132	18	or	or	CCONJ
bracis-33663	132	19	offensive	offensive	ADJ
bracis-33663	132	20	language	language	NOUN
bracis-33663	132	21	”	"	PUNCT
bracis-33663	132	22	.	.	PUNCT
bracis-33663	133	1	this	this	DET
bracis-33663	133	2	result	result	NOUN
bracis-33663	133	3	suggests	suggest	VERB
bracis-33663	133	4	that	that	SCONJ
bracis-33663	133	5	explicitly	explicitly	ADV
bracis-33663	133	6	addressing	address	VERB
bracis-33663	133	7	hateful	hateful	ADJ
bracis-33663	133	8	entries	entry	NOUN
bracis-33663	133	9	and	and	CCONJ
bracis-33663	133	10	offensive	offensive	ADJ
bracis-33663	133	11	language	language	NOUN
bracis-33663	133	12	in	in	ADP
bracis-33663	133	13	the	the	DET
bracis-33663	133	14	hypothesis	hypothesis	NOUN
bracis-33663	133	15	effectively	effectively	ADV
bracis-33663	133	16	captures	capture	VERB
bracis-33663	133	17	a	a	DET
bracis-33663	133	18	broader	broad	ADJ
bracis-33663	133	19	range	range	NOUN
bracis-33663	133	20	of	of	ADP
bracis-33663	133	21	harmful	harmful	ADJ
bracis-33663	133	22	content	content	NOUN
bracis-33663	133	23	.	.	PUNCT
bracis-33663	134	1	the	the	DET
bracis-33663	134	2	model	model	NOUN
bracis-33663	134	3	excels	excel	VERB
bracis-33663	134	4	here	here	ADV
bracis-33663	134	5	because	because	SCONJ
bracis-33663	134	6	it	it	PRON
bracis-33663	134	7	does	do	AUX
bracis-33663	134	8	n’t	not	PART
bracis-33663	134	9	need	need	VERB
bracis-33663	134	10	to	to	PART
bracis-33663	134	11	make	make	VERB
bracis-33663	134	12	a	a	DET
bracis-33663	134	13	strict	strict	ADJ
bracis-33663	134	14	distinction	distinction	NOUN
bracis-33663	134	15	between	between	ADP
bracis-33663	134	16	the	the	DET
bracis-33663	134	17	two	two	NUM
bracis-33663	134	18	types	type	NOUN
bracis-33663	134	19	of	of	ADP
bracis-33663	134	20	content	content	NOUN
bracis-33663	134	21	,	,	PUNCT
bracis-33663	134	22	which	which	PRON
bracis-33663	134	23	can	can	AUX
bracis-33663	134	24	be	be	AUX
bracis-33663	134	25	challenging	challenge	VERB
bracis-33663	134	26	given	give	VERB
bracis-33663	134	27	their	their	PRON
bracis-33663	134	28	nuanced	nuanced	ADJ
bracis-33663	134	29	overlap	overlap	NOUN
bracis-33663	134	30	.	.	PUNCT
bracis-33663	135	1	we	we	PRON
bracis-33663	135	2	conducted	conduct	VERB
bracis-33663	135	3	this	this	DET
bracis-33663	135	4	brief	brief	ADJ
bracis-33663	135	5	experiment	experiment	NOUN
bracis-33663	135	6	using	use	VERB
bracis-33663	135	7	the	the	DET
bracis-33663	135	8	testing	testing	NOUN
bracis-33663	135	9	split	split	NOUN
bracis-33663	135	10	of	of	ADP
bracis-33663	135	11	“	"	PUNCT
bracis-33663	135	12	learning	learn	VERB
bracis-33663	135	13	from	from	ADP
bracis-33663	135	14	the	the	DET
bracis-33663	135	15	worst	bad	ADJ
bracis-33663	135	16	”	"	PUNCT
bracis-33663	135	17	and	and	CCONJ
bracis-33663	135	18	“	"	PUNCT
bracis-33663	135	19	hatecheck	hatecheck	ADJ
bracis-33663	135	20	”	"	PUNCT
bracis-33663	135	21	datasets	dataset	NOUN
bracis-33663	135	22	.	.	PUNCT
bracis-33663	136	1	table	table	NOUN
bracis-33663	136	2	2	2	NUM
bracis-33663	136	3	.	.	X
bracis-33663	136	4	hypothesis	hypothesis	NOUN
bracis-33663	136	5	test	test	NOUN
bracis-33663	136	6	(	(	PUNCT
bracis-33663	136	7	best	good	ADJ
bracis-33663	136	8	performing	performing	NOUN
bracis-33663	136	9	results	result	NOUN
bracis-33663	136	10	are	be	AUX
bracis-33663	136	11	)	)	PUNCT
bracis-33663	136	12	full	full	ADJ
bracis-33663	136	13	size	size	NOUN
bracis-33663	136	14	table	table	NOUN
bracis-33663	136	15	the	the	DET
bracis-33663	136	16	worst	bad	ADJ
bracis-33663	136	17	hypothesis	hypothesis	NOUN
bracis-33663	136	18	with	with	ADP
bracis-33663	136	19	an	an	DET
bracis-33663	136	20	average	average	ADJ
bracis-33663	136	21	accuracy	accuracy	NOUN
bracis-33663	136	22	result	result	NOUN
bracis-33663	136	23	of	of	ADP
bracis-33663	136	24	0.601	0.601	NUM
bracis-33663	136	25	was	be	AUX
bracis-33663	136	26	“	"	PUNCT
bracis-33663	136	27	hate	hate	NOUN
bracis-33663	136	28	”	"	PUNCT
bracis-33663	136	29	.	.	PUNCT
bracis-33663	137	1	this	this	DET
bracis-33663	137	2	result	result	NOUN
bracis-33663	137	3	could	could	AUX
bracis-33663	137	4	be	be	AUX
bracis-33663	137	5	attributed	attribute	VERB
bracis-33663	137	6	to	to	ADP
bracis-33663	137	7	the	the	DET
bracis-33663	137	8	extreme	extreme	ADJ
bracis-33663	137	9	brevity	brevity	NOUN
bracis-33663	137	10	and	and	CCONJ
bracis-33663	137	11	lack	lack	NOUN
bracis-33663	137	12	of	of	ADP
bracis-33663	137	13	context	context	NOUN
bracis-33663	137	14	in	in	ADP
bracis-33663	137	15	the	the	DET
bracis-33663	137	16	hypothesis	hypothesis	NOUN
bracis-33663	137	17	.	.	PUNCT
bracis-33663	138	1	it	it	PRON
bracis-33663	138	2	does	do	AUX
bracis-33663	138	3	not	not	PART
bracis-33663	138	4	provide	provide	VERB
bracis-33663	138	5	sufficient	sufficient	ADJ
bracis-33663	138	6	information	information	NOUN
bracis-33663	138	7	for	for	ADP
bracis-33663	138	8	the	the	DET
bracis-33663	138	9	model	model	NOUN
bracis-33663	138	10	to	to	PART
bracis-33663	138	11	accurately	accurately	ADV
bracis-33663	138	12	classify	classify	VERB
bracis-33663	138	13	hate	hate	NOUN
bracis-33663	138	14	speech	speech	NOUN
bracis-33663	138	15	or	or	CCONJ
bracis-33663	138	16	offensive	offensive	ADJ
bracis-33663	138	17	language	language	NOUN
bracis-33663	138	18	.	.	PUNCT
bracis-33663	139	1	furthermore	furthermore	ADV
bracis-33663	139	2	,	,	PUNCT
bracis-33663	139	3	the	the	DET
bracis-33663	139	4	term	term	NOUN
bracis-33663	139	5	“	"	PUNCT
bracis-33663	139	6	hate	hate	NOUN
bracis-33663	139	7	”	"	PUNCT
bracis-33663	139	8	on	on	ADP
bracis-33663	139	9	its	its	PRON
bracis-33663	139	10	own	own	ADJ
bracis-33663	139	11	is	be	AUX
bracis-33663	139	12	highly	highly	ADV
bracis-33663	139	13	ambiguous	ambiguous	ADJ
bracis-33663	139	14	,	,	PUNCT
bracis-33663	139	15	making	make	VERB
bracis-33663	139	16	it	it	PRON
bracis-33663	139	17	challenging	challenge	VERB
bracis-33663	139	18	for	for	ADP
bracis-33663	139	19	the	the	DET
bracis-33663	139	20	model	model	NOUN
bracis-33663	139	21	to	to	PART
bracis-33663	139	22	make	make	VERB
bracis-33663	139	23	accurate	accurate	ADJ
bracis-33663	139	24	predictions	prediction	NOUN
bracis-33663	139	25	.	.	PUNCT
bracis-33663	140	1	these	these	DET
bracis-33663	140	2	results	result	NOUN
bracis-33663	140	3	show	show	VERB
bracis-33663	140	4	that	that	SCONJ
bracis-33663	140	5	the	the	DET
bracis-33663	140	6	performance	performance	NOUN
bracis-33663	140	7	of	of	ADP
bracis-33663	140	8	the	the	DET
bracis-33663	140	9	nli	nli	PROPN
bracis-33663	140	10	model	model	NOUN
bracis-33663	140	11	in	in	ADP
bracis-33663	140	12	zshsd	zshsd	PROPN
bracis-33663	140	13	is	be	AUX
bracis-33663	140	14	closely	closely	ADV
bracis-33663	140	15	tied	tie	VERB
bracis-33663	140	16	to	to	ADP
bracis-33663	140	17	the	the	DET
bracis-33663	140	18	clarity	clarity	NOUN
bracis-33663	140	19	and	and	CCONJ
bracis-33663	140	20	specificity	specificity	NOUN
bracis-33663	140	21	of	of	ADP
bracis-33663	140	22	the	the	DET
bracis-33663	140	23	hypothesis	hypothesis	NOUN
bracis-33663	140	24	.	.	PUNCT
bracis-33663	141	1	hypotheses	hypothesis	NOUN
bracis-33663	141	2	that	that	PRON
bracis-33663	141	3	encompass	encompass	VERB
bracis-33663	141	4	a	a	DET
bracis-33663	141	5	broader	broad	ADJ
bracis-33663	141	6	spectrum	spectrum	NOUN
bracis-33663	141	7	of	of	ADP
bracis-33663	141	8	harmful	harmful	ADJ
bracis-33663	141	9	content	content	NOUN
bracis-33663	141	10	and	and	CCONJ
bracis-33663	141	11	provide	provide	VERB
bracis-33663	141	12	context	context	NOUN
bracis-33663	141	13	tend	tend	VERB
bracis-33663	141	14	to	to	PART
bracis-33663	141	15	yield	yield	VERB
bracis-33663	141	16	better	well	ADJ
bracis-33663	141	17	results	result	NOUN
bracis-33663	141	18	.	.	PUNCT
bracis-33663	142	1	conversely	conversely	ADV
bracis-33663	142	2	,	,	PUNCT
bracis-33663	142	3	overly	overly	ADV
bracis-33663	142	4	brief	brief	ADJ
bracis-33663	142	5	or	or	CCONJ
bracis-33663	142	6	ambiguous	ambiguous	ADJ
bracis-33663	142	7	hypotheses	hypothesis	NOUN
bracis-33663	142	8	may	may	AUX
bracis-33663	142	9	hinder	hinder	VERB
bracis-33663	142	10	the	the	DET
bracis-33663	142	11	model	model	NOUN
bracis-33663	142	12	’s	’s	PART
bracis-33663	142	13	ability	ability	NOUN
bracis-33663	142	14	to	to	PART
bracis-33663	142	15	make	make	VERB
bracis-33663	142	16	accurate	accurate	ADJ
bracis-33663	142	17	classifications	classification	NOUN
bracis-33663	142	18	,	,	PUNCT
bracis-33663	142	19	as	as	SCONJ
bracis-33663	142	20	evidenced	evidence	VERB
bracis-33663	142	21	by	by	ADP
bracis-33663	142	22	the	the	DET
bracis-33663	142	23	lower	low	ADJ
bracis-33663	142	24	scores	score	NOUN
bracis-33663	142	25	observed	observe	VERB
bracis-33663	142	26	in	in	ADP
bracis-33663	142	27	this	this	DET
bracis-33663	142	28	test	test	NOUN
bracis-33663	142	29	.	.	PUNCT
bracis-33663	143	1	therefore	therefore	ADV
bracis-33663	143	2	,	,	PUNCT
bracis-33663	143	3	it	it	PRON
bracis-33663	143	4	is	be	AUX
bracis-33663	143	5	imperative	imperative	ADJ
bracis-33663	143	6	to	to	PART
bracis-33663	143	7	carefully	carefully	ADV
bracis-33663	143	8	craft	craft	VERB
bracis-33663	143	9	hypotheses	hypothesis	NOUN
bracis-33663	143	10	that	that	PRON
bracis-33663	143	11	encompass	encompass	VERB
bracis-33663	143	12	the	the	DET
bracis-33663	143	13	diversity	diversity	NOUN
bracis-33663	143	14	and	and	CCONJ
bracis-33663	143	15	context	context	NOUN
bracis-33663	143	16	of	of	ADP
bracis-33663	143	17	hate	hate	NOUN
bracis-33663	143	18	speech	speech	NOUN
bracis-33663	143	19	to	to	PART
bracis-33663	143	20	enhance	enhance	VERB
bracis-33663	143	21	the	the	DET
bracis-33663	143	22	effectiveness	effectiveness	NOUN
bracis-33663	143	23	of	of	ADP
bracis-33663	143	24	nli	nli	PROPN
bracis-33663	143	25	-	-	PUNCT
bracis-33663	143	26	based	base	VERB
bracis-33663	143	27	zshsd	zshsd	NOUN
bracis-33663	143	28	models	model	NOUN
bracis-33663	143	29	.	.	PUNCT
bracis-33663	144	1	4.2	4.2	NUM
bracis-33663	144	2	hypothesis	hypothesis	NOUN
bracis-33663	144	3	similarity	similarity	NOUN
bracis-33663	144	4	analysis	analysis	NOUN
bracis-33663	144	5	in	in	ADP
bracis-33663	144	6	table	table	NOUN
bracis-33663	144	7	 	 	SPACE
bracis-33663	144	8	3	3	NUM
bracis-33663	144	9	below	below	ADV
bracis-33663	144	10	,	,	PUNCT
bracis-33663	144	11	we	we	PRON
bracis-33663	144	12	present	present	VERB
bracis-33663	144	13	the	the	DET
bracis-33663	144	14	original	original	ADJ
bracis-33663	144	15	token	token	VERB
bracis-33663	144	16	representations	representation	NOUN
bracis-33663	144	17	alongside	alongside	ADP
bracis-33663	144	18	their	their	PRON
bracis-33663	144	19	optimized	optimize	VERB
bracis-33663	144	20	counterparts	counterpart	NOUN
bracis-33663	144	21	.	.	PUNCT
bracis-33663	145	1	we	we	PRON
bracis-33663	145	2	will	will	AUX
bracis-33663	145	3	examine	examine	VERB
bracis-33663	145	4	the	the	DET
bracis-33663	145	5	changes	change	NOUN
bracis-33663	145	6	in	in	ADP
bracis-33663	145	7	each	each	DET
bracis-33663	145	8	token	token	ADJ
bracis-33663	145	9	and	and	CCONJ
bracis-33663	145	10	discuss	discuss	VERB
bracis-33663	145	11	why	why	SCONJ
bracis-33663	145	12	the	the	DET
bracis-33663	145	13	optimized	optimize	VERB
bracis-33663	145	14	version	version	NOUN
bracis-33663	145	15	might	might	AUX
bracis-33663	145	16	be	be	AUX
bracis-33663	145	17	more	more	ADV
bracis-33663	145	18	or	or	CCONJ
bracis-33663	145	19	less	less	ADV
bracis-33663	145	20	effective	effective	ADJ
bracis-33663	145	21	for	for	ADP
bracis-33663	145	22	predicting	predict	VERB
bracis-33663	145	23	hate	hate	NOUN
bracis-33663	145	24	speech	speech	NOUN
bracis-33663	145	25	.	.	PUNCT
bracis-33663	146	1	furthermore	furthermore	ADV
bracis-33663	146	2	,	,	PUNCT
bracis-33663	146	3	we	we	PRON
bracis-33663	146	4	will	will	AUX
bracis-33663	146	5	highlight	highlight	VERB
bracis-33663	146	6	the	the	DET
bracis-33663	146	7	tokens	token	NOUN
bracis-33663	146	8	that	that	PRON
bracis-33663	146	9	are	be	AUX
bracis-33663	146	10	likely	likely	ADJ
bracis-33663	146	11	to	to	PART
bracis-33663	146	12	be	be	AUX
bracis-33663	146	13	more	more	ADV
bracis-33663	146	14	useful	useful	ADJ
bracis-33663	146	15	for	for	ADP
bracis-33663	146	16	this	this	DET
bracis-33663	146	17	task	task	NOUN
bracis-33663	146	18	.	.	PUNCT
bracis-33663	147	1	it	it	PRON
bracis-33663	147	2	’s	’	VERB
bracis-33663	147	3	important	important	ADJ
bracis-33663	147	4	to	to	PART
bracis-33663	147	5	note	note	VERB
bracis-33663	147	6	that	that	SCONJ
bracis-33663	147	7	these	these	DET
bracis-33663	147	8	transformations	transformation	NOUN
bracis-33663	147	9	occur	occur	VERB
bracis-33663	147	10	within	within	ADP
bracis-33663	147	11	a	a	DET
bracis-33663	147	12	continuous	continuous	ADJ
bracis-33663	147	13	embedding	embed	VERB
bracis-33663	147	14	space	space	NOUN
bracis-33663	147	15	,	,	PUNCT
bracis-33663	147	16	so	so	CCONJ
bracis-33663	147	17	the	the	DET
bracis-33663	147	18	similarity	similarity	NOUN
bracis-33663	147	19	of	of	ADP
bracis-33663	147	20	the	the	DET
bracis-33663	147	21	optimized	optimize	VERB
bracis-33663	147	22	tokens	token	NOUN
bracis-33663	147	23	to	to	ADP
bracis-33663	147	24	specific	specific	ADJ
bracis-33663	147	25	words	word	NOUN
bracis-33663	147	26	might	might	AUX
bracis-33663	147	27	not	not	PART
bracis-33663	147	28	fully	fully	ADV
bracis-33663	147	29	capture	capture	VERB
bracis-33663	147	30	the	the	DET
bracis-33663	147	31	subtleties	subtlety	NOUN
bracis-33663	147	32	of	of	ADP
bracis-33663	147	33	their	their	PRON
bracis-33663	147	34	representation	representation	NOUN
bracis-33663	147	35	.	.	PUNCT
bracis-33663	148	1	additionally	additionally	ADV
bracis-33663	148	2	,	,	PUNCT
bracis-33663	148	3	even	even	ADV
bracis-33663	148	4	if	if	SCONJ
bracis-33663	148	5	tokens	token	NOUN
bracis-33663	148	6	seem	seem	VERB
bracis-33663	148	7	less	less	ADV
bracis-33663	148	8	interpretable	interpretable	ADJ
bracis-33663	148	9	from	from	ADP
bracis-33663	148	10	a	a	DET
bracis-33663	148	11	human	human	ADJ
bracis-33663	148	12	perspective	perspective	NOUN
bracis-33663	148	13	,	,	PUNCT
bracis-33663	148	14	they	they	PRON
bracis-33663	148	15	can	can	AUX
bracis-33663	148	16	still	still	ADV
bracis-33663	148	17	impact	impact	VERB
bracis-33663	148	18	the	the	DET
bracis-33663	148	19	model	model	NOUN
bracis-33663	148	20	’s	’s	PART
bracis-33663	148	21	internal	internal	ADJ
bracis-33663	148	22	weights	weight	NOUN
bracis-33663	148	23	in	in	ADP
bracis-33663	148	24	ways	way	NOUN
bracis-33663	148	25	that	that	PRON
bracis-33663	148	26	may	may	AUX
bracis-33663	148	27	positively	positively	ADV
bracis-33663	148	28	or	or	CCONJ
bracis-33663	148	29	negatively	negatively	ADV
bracis-33663	148	30	affect	affect	VERB
bracis-33663	148	31	its	its	PRON
bracis-33663	148	32	performance	performance	NOUN
bracis-33663	148	33	on	on	ADP
bracis-33663	148	34	the	the	DET
bracis-33663	148	35	task	task	NOUN
bracis-33663	148	36	.	.	PUNCT
bracis-33663	148	37	table	table	NOUN
bracis-33663	148	38	3	3	NUM
bracis-33663	148	39	.	.	PUNCT
bracis-33663	148	40	original	original	ADJ
bracis-33663	148	41	token	token	VERB
bracis-33663	148	42	and	and	CCONJ
bracis-33663	148	43	optimized	optimize	VERB
bracis-33663	148	44	/	/	SYM
bracis-33663	148	45	p	p	NOUN
bracis-33663	148	46	-	-	PUNCT
bracis-33663	148	47	tuned	tune	VERB
bracis-33663	148	48	token	token	ADJ
bracis-33663	148	49	cosine	cosine	NOUN
bracis-33663	148	50	similarityfull	similarityfull	ADJ
bracis-33663	148	51	size	size	NOUN
bracis-33663	148	52	table	table	NOUN
bracis-33663	148	53	this	this	PRON
bracis-33663	148	54	the	the	DET
bracis-33663	148	55	optimized	optimize	VERB
bracis-33663	148	56	version	version	NOUN
bracis-33663	148	57	adds	add	VERB
bracis-33663	148	58	unique	unique	ADJ
bracis-33663	148	59	characters	character	NOUN
bracis-33663	148	60	and	and	CCONJ
bracis-33663	148	61	symbols	symbol	NOUN
bracis-33663	148	62	,	,	PUNCT
bracis-33663	148	63	like	like	ADP
bracis-33663	148	64	“	"	PUNCT
bracis-33663	148	65	[	[	X
bracis-33663	148	66	...	...	PUNCT
bracis-33663	148	67	]	]	X
bracis-33663	148	68	”	"	PUNCT
bracis-33663	148	69	and	and	CCONJ
bracis-33663	148	70	“	"	PUNCT
bracis-33663	148	71	”	"	PUNCT
bracis-33663	148	72	,	,	PUNCT
bracis-33663	148	73	which	which	PRON
bracis-33663	148	74	might	might	AUX
bracis-33663	148	75	enhance	enhance	VERB
bracis-33663	148	76	the	the	DET
bracis-33663	148	77	model	model	NOUN
bracis-33663	148	78	’s	’s	PART
bracis-33663	148	79	ability	ability	NOUN
bracis-33663	148	80	to	to	PART
bracis-33663	148	81	detect	detect	VERB
bracis-33663	148	82	nuanced	nuanced	ADJ
bracis-33663	148	83	hate	hate	NOUN
bracis-33663	148	84	speech	speech	NOUN
bracis-33663	148	85	or	or	CCONJ
bracis-33663	148	86	sarcasm	sarcasm	NOUN
bracis-33663	148	87	.	.	PUNCT
bracis-33663	149	1	however	however	ADV
bracis-33663	149	2	,	,	PUNCT
bracis-33663	149	3	it	it	PRON
bracis-33663	149	4	makes	make	VERB
bracis-33663	149	5	direct	direct	ADJ
bracis-33663	149	6	interpretation	interpretation	NOUN
bracis-33663	149	7	harder	hard	ADV
bracis-33663	149	8	.	.	PUNCT
bracis-33663	150	1	sentence	sentence	VERB
bracis-33663	150	2	the	the	DET
bracis-33663	150	3	optimized	optimize	VERB
bracis-33663	150	4	version	version	NOUN
bracis-33663	150	5	includes	include	VERB
bracis-33663	150	6	unrelated	unrelated	ADJ
bracis-33663	150	7	words	word	NOUN
bracis-33663	150	8	like	like	ADP
bracis-33663	150	9	“	"	PUNCT
bracis-33663	150	10	edits	edit	NOUN
bracis-33663	150	11	”	"	PUNCT
bracis-33663	150	12	,	,	PUNCT
bracis-33663	150	13	“	"	PUNCT
bracis-33663	150	14	inscription	inscription	NOUN
bracis-33663	150	15	”	"	PUNCT
bracis-33663	150	16	and	and	CCONJ
bracis-33663	150	17	“	"	PUNCT
bracis-33663	150	18	subtitles	subtitles	PROPN
bracis-33663	150	19	”	"	PUNCT
bracis-33663	150	20	which	which	PRON
bracis-33663	150	21	do	do	AUX
bracis-33663	150	22	n’t	not	PART
bracis-33663	150	23	seem	seem	VERB
bracis-33663	150	24	to	to	PART
bracis-33663	150	25	help	help	VERB
bracis-33663	150	26	with	with	ADP
bracis-33663	150	27	zshsd	zshsd	NOUN
bracis-33663	150	28	but	but	CCONJ
bracis-33663	150	29	rather	rather	ADV
bracis-33663	150	30	shift	shift	VERB
bracis-33663	150	31	towards	towards	ADP
bracis-33663	150	32	unrelated	unrelated	ADJ
bracis-33663	150	33	terms	term	NOUN
bracis-33663	150	34	.	.	PUNCT
bracis-33663	151	1	contains	contain	VERB
bracis-33663	151	2	the	the	DET
bracis-33663	151	3	optimized	optimize	VERB
bracis-33663	151	4	version	version	NOUN
bracis-33663	151	5	contains	contain	VERB
bracis-33663	151	6	random	random	ADJ
bracis-33663	151	7	characters	character	NOUN
bracis-33663	151	8	and	and	CCONJ
bracis-33663	151	9	sequences	sequence	NOUN
bracis-33663	151	10	(	(	PUNCT
bracis-33663	151	11	“	"	PUNCT
bracis-33663	151	12	pha	pha	PROPN
bracis-33663	151	13	”	"	PUNCT
bracis-33663	151	14	,	,	PUNCT
bracis-33663	151	15	“	"	PUNCT
bracis-33663	151	16	nton	nton	PROPN
bracis-33663	151	17	”	"	PUNCT
bracis-33663	151	18	,	,	PUNCT
bracis-33663	151	19	etc	etc	X
bracis-33663	151	20	.	.	X
bracis-33663	151	21	)	)	PUNCT
bracis-33663	151	22	that	that	PRON
bracis-33663	151	23	do	do	AUX
bracis-33663	151	24	n’t	not	PART
bracis-33663	151	25	align	align	VERB
bracis-33663	151	26	with	with	ADP
bracis-33663	151	27	the	the	DET
bracis-33663	151	28	original	original	ADJ
bracis-33663	151	29	token	token	NOUN
bracis-33663	151	30	’s	’s	PART
bracis-33663	151	31	meaning	meaning	NOUN
bracis-33663	151	32	.	.	PUNCT
bracis-33663	152	1	this	this	DET
bracis-33663	152	2	transformation	transformation	NOUN
bracis-33663	152	3	does	do	AUX
bracis-33663	152	4	n’t	not	PART
bracis-33663	152	5	relate	relate	VERB
bracis-33663	152	6	to	to	ADP
bracis-33663	152	7	zshsd	zshsd	PROPN
bracis-33663	152	8	.	.	PUNCT
bracis-33663	153	1	hateful	hateful	VERB
bracis-33663	153	2	the	the	DET
bracis-33663	153	3	optimized	optimize	VERB
bracis-33663	153	4	version	version	NOUN
bracis-33663	153	5	keeps	keep	VERB
bracis-33663	153	6	“	"	PUNCT
bracis-33663	153	7	hateful	hateful	ADJ
bracis-33663	153	8	”	"	PUNCT
bracis-33663	153	9	and	and	CCONJ
bracis-33663	153	10	“	"	PUNCT
bracis-33663	153	11	despicable	despicable	ADJ
bracis-33663	153	12	”	"	PUNCT
bracis-33663	153	13	while	while	SCONJ
bracis-33663	153	14	adding	add	VERB
bracis-33663	153	15	terms	term	NOUN
bracis-33663	153	16	like	like	ADP
bracis-33663	153	17	“	"	PUNCT
bracis-33663	153	18	fascist	fascist	NOUN
bracis-33663	153	19	”	"	PUNCT
bracis-33663	153	20	and	and	CCONJ
bracis-33663	153	21	“	"	PUNCT
bracis-33663	153	22	discriminatory	discriminatory	ADJ
bracis-33663	153	23	”	"	PUNCT
bracis-33663	153	24	,	,	PUNCT
bracis-33663	153	25	broadening	broaden	VERB
bracis-33663	153	26	its	its	PRON
bracis-33663	153	27	capacity	capacity	NOUN
bracis-33663	153	28	to	to	PART
bracis-33663	153	29	detect	detect	VERB
bracis-33663	153	30	hate	hate	NOUN
bracis-33663	153	31	speech	speech	NOUN
bracis-33663	153	32	with	with	ADP
bracis-33663	153	33	more	more	ADV
bracis-33663	153	34	discriminatory	discriminatory	ADJ
bracis-33663	153	35	and	and	CCONJ
bracis-33663	153	36	extremist	extremist	ADJ
bracis-33663	153	37	terms	term	NOUN
bracis-33663	153	38	.	.	PUNCT
bracis-33663	154	1	entries	entry	NOUN
bracis-33663	154	2	the	the	DET
bracis-33663	154	3	optimized	optimize	VERB
bracis-33663	154	4	version	version	NOUN
bracis-33663	154	5	introduces	introduce	NOUN
bracis-33663	154	6	unrelated	unrelated	ADJ
bracis-33663	154	7	terms	term	NOUN
bracis-33663	154	8	(	(	PUNCT
bracis-33663	154	9	“	"	PUNCT
bracis-33663	154	10	ansk	ansk	ADJ
bracis-33663	154	11	”	"	PUNCT
bracis-33663	154	12	,	,	PUNCT
bracis-33663	154	13	“	"	PUNCT
bracis-33663	154	14	owitz	owitz	PROPN
bracis-33663	154	15	”	"	PUNCT
bracis-33663	154	16	,	,	PUNCT
bracis-33663	154	17	etc	etc	X
bracis-33663	154	18	.	.	X
bracis-33663	154	19	)	)	PUNCT
bracis-33663	154	20	that	that	PRON
bracis-33663	154	21	do	do	AUX
bracis-33663	154	22	n’t	not	PART
bracis-33663	154	23	positively	positively	ADV
bracis-33663	154	24	contribute	contribute	VERB
bracis-33663	154	25	to	to	ADP
bracis-33663	154	26	zshsd	zshsd	NOUN
bracis-33663	154	27	or	or	CCONJ
bracis-33663	154	28	relate	relate	VERB
bracis-33663	154	29	to	to	ADP
bracis-33663	154	30	the	the	DET
bracis-33663	154	31	original	original	ADJ
bracis-33663	154	32	token	token	NOUN
bracis-33663	154	33	’s	’s	PART
bracis-33663	154	34	meaning	meaning	NOUN
bracis-33663	154	35	.	.	PUNCT
bracis-33663	155	1	or	or	CCONJ
bracis-33663	155	2	the	the	DET
bracis-33663	155	3	optimized	optimize	VERB
bracis-33663	155	4	version	version	NOUN
bracis-33663	155	5	adds	add	VERB
bracis-33663	155	6	terms	term	NOUN
bracis-33663	155	7	like	like	ADP
bracis-33663	155	8	“	"	PUNCT
bracis-33663	155	9	antagonist	antagonist	NOUN
bracis-33663	155	10	”	"	PUNCT
bracis-33663	155	11	,	,	PUNCT
bracis-33663	155	12	“	"	PUNCT
bracis-33663	155	13	favorably	favorably	ADV
bracis-33663	155	14	”	"	PUNCT
bracis-33663	155	15	and	and	CCONJ
bracis-33663	155	16	“	"	PUNCT
bracis-33663	155	17	extremists	extremist	NOUN
bracis-33663	155	18	”	"	PUNCT
bracis-33663	155	19	which	which	PRON
bracis-33663	155	20	are	be	AUX
bracis-33663	155	21	n’t	not	PART
bracis-33663	155	22	directly	directly	ADV
bracis-33663	155	23	related	relate	VERB
bracis-33663	155	24	to	to	ADP
bracis-33663	155	25	the	the	DET
bracis-33663	155	26	original	original	ADJ
bracis-33663	155	27	token	token	NOUN
bracis-33663	155	28	but	but	CCONJ
bracis-33663	155	29	may	may	AUX
bracis-33663	155	30	help	help	AUX
bracis-33663	155	31	identify	identify	VERB
bracis-33663	155	32	hate	hate	NOUN
bracis-33663	155	33	speech	speech	NOUN
bracis-33663	155	34	involving	involve	VERB
bracis-33663	155	35	comparisons	comparison	NOUN
bracis-33663	155	36	,	,	PUNCT
bracis-33663	155	37	contrasts	contrast	VERB
bracis-33663	155	38	,	,	PUNCT
bracis-33663	155	39	or	or	CCONJ
bracis-33663	155	40	extremist	extremist	ADJ
bracis-33663	155	41	viewpoints	viewpoint	NOUN
bracis-33663	155	42	.	.	PUNCT
bracis-33663	156	1	offensive	offensive	VERB
bracis-33663	156	2	the	the	DET
bracis-33663	156	3	optimized	optimize	VERB
bracis-33663	156	4	version	version	NOUN
bracis-33663	156	5	includes	include	VERB
bracis-33663	156	6	terms	term	NOUN
bracis-33663	156	7	like	like	ADP
bracis-33663	156	8	“	"	PUNCT
bracis-33663	156	9	transsexual	transsexual	ADJ
bracis-33663	156	10	”	"	PUNCT
bracis-33663	156	11	,	,	PUNCT
bracis-33663	156	12	“	"	PUNCT
bracis-33663	156	13	lgbt	lgbt	PROPN
bracis-33663	156	14	”	"	PUNCT
bracis-33663	156	15	,	,	PUNCT
bracis-33663	156	16	and	and	CCONJ
bracis-33663	156	17	“	"	PUNCT
bracis-33663	156	18	ethnic	ethnic	ADJ
bracis-33663	156	19	”	"	PUNCT
bracis-33663	156	20	which	which	PRON
bracis-33663	156	21	are	be	AUX
bracis-33663	156	22	relevant	relevant	ADJ
bracis-33663	156	23	for	for	ADP
bracis-33663	156	24	detecting	detect	VERB
bracis-33663	156	25	hate	hate	NOUN
bracis-33663	156	26	speech	speech	NOUN
bracis-33663	156	27	involving	involve	VERB
bracis-33663	156	28	sensitive	sensitive	ADJ
bracis-33663	156	29	topics	topic	NOUN
bracis-33663	156	30	and	and	CCONJ
bracis-33663	156	31	discriminatory	discriminatory	ADJ
bracis-33663	156	32	language	language	NOUN
bracis-33663	156	33	.	.	PUNCT
bracis-33663	157	1	language	language	NOUN
bracis-33663	157	2	the	the	DET
bracis-33663	157	3	optimized	optimize	VERB
bracis-33663	157	4	version	version	NOUN
bracis-33663	157	5	introduces	introduce	NOUN
bracis-33663	157	6	unrelated	unrelated	ADJ
bracis-33663	157	7	terms	term	NOUN
bracis-33663	157	8	(	(	PUNCT
bracis-33663	157	9	“	"	PUNCT
bracis-33663	157	10	andra	andra	X
bracis-33663	157	11	”	"	PUNCT
bracis-33663	157	12	,	,	PUNCT
bracis-33663	157	13	“	"	PUNCT
bracis-33663	157	14	eto	eto	PROPN
bracis-33663	157	15	”	"	PUNCT
bracis-33663	157	16	,	,	PUNCT
bracis-33663	157	17	etc	etc	X
bracis-33663	157	18	.	.	X
bracis-33663	157	19	)	)	PUNCT
bracis-33663	158	1	that	that	PRON
bracis-33663	158	2	do	do	AUX
bracis-33663	158	3	n’t	not	PART
bracis-33663	158	4	contribute	contribute	VERB
bracis-33663	158	5	positively	positively	ADV
bracis-33663	158	6	to	to	ADP
bracis-33663	158	7	zshsd	zshsd	PROPN
bracis-33663	158	8	.	.	PUNCT
bracis-33663	159	1	the	the	DET
bracis-33663	159	2	evaluation	evaluation	NOUN
bracis-33663	159	3	of	of	ADP
bracis-33663	159	4	the	the	DET
bracis-33663	159	5	optimized	optimize	VERB
bracis-33663	159	6	tokens	token	NOUN
bracis-33663	159	7	and	and	CCONJ
bracis-33663	159	8	their	their	PRON
bracis-33663	159	9	contribution	contribution	NOUN
bracis-33663	159	10	to	to	ADP
bracis-33663	159	11	zshsd	zshsd	PROPN
bracis-33663	159	12	can	can	AUX
bracis-33663	159	13	vary	vary	VERB
bracis-33663	159	14	significantly	significantly	ADV
bracis-33663	159	15	depending	depend	VERB
bracis-33663	159	16	on	on	ADP
bracis-33663	159	17	the	the	DET
bracis-33663	159	18	specific	specific	ADJ
bracis-33663	159	19	nli	nli	ADJ
bracis-33663	159	20	model	model	NOUN
bracis-33663	159	21	and	and	CCONJ
bracis-33663	159	22	its	its	PRON
bracis-33663	159	23	objectives	objective	NOUN
bracis-33663	159	24	.	.	PUNCT
bracis-33663	160	1	the	the	DET
bracis-33663	160	2	analysis	analysis	NOUN
bracis-33663	160	3	provided	provide	VERB
bracis-33663	160	4	here	here	ADV
bracis-33663	160	5	represents	represent	VERB
bracis-33663	160	6	a	a	DET
bracis-33663	160	7	human	human	NOUN
bracis-33663	160	8	-	-	PUNCT
bracis-33663	160	9	driven	drive	VERB
bracis-33663	160	10	evaluation	evaluation	NOUN
bracis-33663	160	11	of	of	ADP
bracis-33663	160	12	the	the	DET
bracis-33663	160	13	tokens	token	NOUN
bracis-33663	160	14	’	'	PUNCT
bracis-33663	160	15	relevance	relevance	NOUN
bracis-33663	160	16	to	to	ADP
bracis-33663	160	17	zshsd	zshsd	PROPN
bracis-33663	160	18	.	.	PUNCT
bracis-33663	161	1	while	while	SCONJ
bracis-33663	161	2	some	some	DET
bracis-33663	161	3	tokens	token	NOUN
bracis-33663	161	4	like	like	ADP
bracis-33663	161	5	“	"	PUNCT
bracis-33663	161	6	hateful	hateful	ADJ
bracis-33663	161	7	”	"	PUNCT
bracis-33663	161	8	and	and	CCONJ
bracis-33663	161	9	“	"	PUNCT
bracis-33663	161	10	offensive	offensive	ADJ
bracis-33663	161	11	”	"	PUNCT
bracis-33663	161	12	appear	appear	VERB
bracis-33663	161	13	to	to	PART
bracis-33663	161	14	strengthen	strengthen	VERB
bracis-33663	161	15	the	the	DET
bracis-33663	161	16	model	model	NOUN
bracis-33663	161	17	’s	’s	PART
bracis-33663	161	18	ability	ability	NOUN
bracis-33663	161	19	to	to	PART
bracis-33663	161	20	identify	identify	VERB
bracis-33663	161	21	hate	hate	VERB
bracis-33663	161	22	speech	speech	NOUN
bracis-33663	161	23	by	by	ADP
bracis-33663	161	24	introducing	introduce	VERB
bracis-33663	161	25	relevant	relevant	ADJ
bracis-33663	161	26	terms	term	NOUN
bracis-33663	161	27	,	,	PUNCT
bracis-33663	161	28	others	other	NOUN
bracis-33663	161	29	seem	seem	VERB
bracis-33663	161	30	unrelated	unrelated	ADJ
bracis-33663	161	31	to	to	ADP
bracis-33663	161	32	the	the	DET
bracis-33663	161	33	task	task	NOUN
bracis-33663	161	34	and	and	CCONJ
bracis-33663	161	35	may	may	AUX
bracis-33663	161	36	introduce	introduce	VERB
bracis-33663	161	37	noise	noise	NOUN
bracis-33663	161	38	.	.	PUNCT
bracis-33663	162	1	it	it	PRON
bracis-33663	162	2	’s	’	VERB
bracis-33663	162	3	essential	essential	ADJ
bracis-33663	162	4	to	to	PART
bracis-33663	162	5	note	note	VERB
bracis-33663	162	6	that	that	SCONJ
bracis-33663	162	7	the	the	DET
bracis-33663	162	8	effectiveness	effectiveness	NOUN
bracis-33663	162	9	of	of	ADP
bracis-33663	162	10	token	token	ADJ
bracis-33663	162	11	optimization	optimization	NOUN
bracis-33663	162	12	is	be	AUX
bracis-33663	162	13	context	context	NOUN
bracis-33663	162	14	-	-	PUNCT
bracis-33663	162	15	dependent	dependent	ADJ
bracis-33663	162	16	and	and	CCONJ
bracis-33663	162	17	should	should	AUX
bracis-33663	162	18	be	be	AUX
bracis-33663	162	19	evaluated	evaluate	VERB
bracis-33663	162	20	within	within	ADP
bracis-33663	162	21	the	the	DET
bracis-33663	162	22	broader	broad	ADJ
bracis-33663	162	23	framework	framework	NOUN
bracis-33663	162	24	of	of	ADP
bracis-33663	162	25	the	the	DET
bracis-33663	162	26	nli	nli	PROPN
bracis-33663	162	27	model	model	PROPN
bracis-33663	162	28	’s	’s	PART
bracis-33663	162	29	goals	goal	NOUN
bracis-33663	162	30	.	.	PUNCT
bracis-33663	163	1	further	further	ADJ
bracis-33663	163	2	experimentation	experimentation	NOUN
bracis-33663	163	3	and	and	CCONJ
bracis-33663	163	4	fine	fine	ADV
bracis-33663	163	5	-	-	PUNCT
bracis-33663	163	6	tuning	tuning	NOUN
bracis-33663	163	7	may	may	AUX
bracis-33663	163	8	be	be	AUX
bracis-33663	163	9	necessary	necessary	ADJ
bracis-33663	163	10	to	to	PART
bracis-33663	163	11	determine	determine	VERB
bracis-33663	163	12	the	the	DET
bracis-33663	163	13	most	most	ADV
bracis-33663	163	14	suitable	suitable	ADJ
bracis-33663	163	15	token	token	ADJ
bracis-33663	163	16	representations	representation	NOUN
bracis-33663	163	17	.	.	PUNCT
bracis-33663	164	1	ultimately	ultimately	ADV
bracis-33663	164	2	,	,	PUNCT
bracis-33663	164	3	the	the	DET
bracis-33663	164	4	choice	choice	NOUN
bracis-33663	164	5	of	of	ADP
bracis-33663	164	6	optimized	optimize	VERB
bracis-33663	164	7	tokens	token	NOUN
bracis-33663	164	8	should	should	AUX
bracis-33663	164	9	align	align	VERB
bracis-33663	164	10	with	with	ADP
bracis-33663	164	11	the	the	DET
bracis-33663	164	12	specific	specific	ADJ
bracis-33663	164	13	nuances	nuance	NOUN
bracis-33663	164	14	and	and	CCONJ
bracis-33663	164	15	objectives	objective	NOUN
bracis-33663	164	16	of	of	ADP
bracis-33663	164	17	the	the	DET
bracis-33663	164	18	problem	problem	NOUN
bracis-33663	164	19	at	at	ADP
bracis-33663	164	20	hand	hand	NOUN
bracis-33663	164	21	,	,	PUNCT
bracis-33663	164	22	and	and	CCONJ
bracis-33663	164	23	the	the	DET
bracis-33663	164	24	assessment	assessment	NOUN
bracis-33663	164	25	should	should	AUX
bracis-33663	164	26	be	be	AUX
bracis-33663	164	27	guided	guide	VERB
bracis-33663	164	28	by	by	ADP
bracis-33663	164	29	both	both	DET
bracis-33663	164	30	human	human	ADJ
bracis-33663	164	31	judgment	judgment	NOUN
bracis-33663	164	32	and	and	CCONJ
bracis-33663	164	33	quantitative	quantitative	ADJ
bracis-33663	164	34	performance	performance	NOUN
bracis-33663	164	35	metrics	metric	NOUN
bracis-33663	164	36	.	.	PUNCT
bracis-33663	165	1	4.3	4.3	NUM
bracis-33663	165	2	token	token	ADJ
bracis-33663	165	3	importance	importance	NOUN
bracis-33663	165	4	assessment	assessment	NOUN
bracis-33663	165	5	the	the	DET
bracis-33663	165	6	performance	performance	NOUN
bracis-33663	165	7	of	of	ADP
bracis-33663	165	8	three	three	NUM
bracis-33663	165	9	different	different	ADJ
bracis-33663	165	10	methods	method	NOUN
bracis-33663	165	11	for	for	ADP
bracis-33663	165	12	incorporating	incorporate	VERB
bracis-33663	165	13	optimized	optimize	VERB
bracis-33663	165	14	tokens	token	NOUN
bracis-33663	165	15	into	into	ADP
bracis-33663	165	16	the	the	DET
bracis-33663	165	17	hypothesis	hypothesis	NOUN
bracis-33663	165	18	for	for	ADP
bracis-33663	165	19	zshsd—“cumulative	zshsd—“cumulative	NUM
bracis-33663	165	20	,	,	PUNCT
bracis-33663	165	21	”	"	PUNCT
bracis-33663	165	22	“	"	PUNCT
bracis-33663	165	23	individual	individual	ADJ
bracis-33663	165	24	,	,	PUNCT
bracis-33663	165	25	”	"	PUNCT
bracis-33663	165	26	and	and	CCONJ
bracis-33663	165	27	“	"	PUNCT
bracis-33663	165	28	isolate”—is	isolate”—is	ADV
bracis-33663	165	29	shown	show	VERB
bracis-33663	165	30	in	in	ADP
bracis-33663	165	31	table	table	NOUN
bracis-33663	165	32	 	 	SPACE
bracis-33663	165	33	4	4	NUM
bracis-33663	165	34	.	.	PUNCT
bracis-33663	166	1	the	the	DET
bracis-33663	166	2	accuracy	accuracy	NOUN
bracis-33663	166	3	scores	score	NOUN
bracis-33663	166	4	over	over	ADP
bracis-33663	166	5	all	all	DET
bracis-33663	166	6	test	test	NOUN
bracis-33663	166	7	data	datum	NOUN
bracis-33663	166	8	reveal	reveal	VERB
bracis-33663	166	9	the	the	DET
bracis-33663	166	10	contributions	contribution	NOUN
bracis-33663	166	11	of	of	ADP
bracis-33663	166	12	specific	specific	ADJ
bracis-33663	166	13	tokens	token	NOUN
bracis-33663	166	14	:	:	PUNCT
bracis-33663	166	15	table	table	NOUN
bracis-33663	166	16	4	4	NUM
bracis-33663	166	17	.	.	PUNCT
bracis-33663	166	18	original	original	ADJ
bracis-33663	166	19	and	and	CCONJ
bracis-33663	166	20	virtual	virtual	ADJ
bracis-33663	166	21	tokens	token	NOUN
bracis-33663	166	22	combinations	combination	NOUN
bracis-33663	166	23	performancefull	performancefull	ADJ
bracis-33663	166	24	size	size	NOUN
bracis-33663	166	25	table	table	NOUN
bracis-33663	166	26	“	"	PUNCT
bracis-33663	166	27	this	this	PRON
bracis-33663	166	28	”	"	PUNCT
bracis-33663	166	29	and	and	CCONJ
bracis-33663	166	30	“	"	PUNCT
bracis-33663	166	31	sentence	sentence	NOUN
bracis-33663	166	32	”	"	PUNCT
bracis-33663	166	33	:	:	PUNCT
bracis-33663	166	34	consistently	consistently	ADV
bracis-33663	166	35	contribute	contribute	VERB
bracis-33663	166	36	across	across	ADP
bracis-33663	166	37	all	all	DET
bracis-33663	166	38	methods	method	NOUN
bracis-33663	166	39	,	,	PUNCT
bracis-33663	166	40	with	with	ADP
bracis-33663	166	41	“	"	PUNCT
bracis-33663	166	42	individual	individual	ADJ
bracis-33663	166	43	”	"	PUNCT
bracis-33663	166	44	showing	show	VERB
bracis-33663	166	45	a	a	DET
bracis-33663	166	46	slightly	slightly	ADV
bracis-33663	166	47	higher	high	ADJ
bracis-33663	166	48	impact	impact	NOUN
bracis-33663	166	49	.	.	PUNCT
bracis-33663	167	1	“	"	PUNCT
bracis-33663	167	2	contains	contain	VERB
bracis-33663	167	3	”	"	PUNCT
bracis-33663	167	4	:	:	PUNCT
bracis-33663	167	5	significantly	significantly	ADV
bracis-33663	167	6	boosts	boost	VERB
bracis-33663	167	7	performance	performance	NOUN
bracis-33663	167	8	in	in	ADP
bracis-33663	167	9	the	the	DET
bracis-33663	167	10	“	"	PUNCT
bracis-33663	167	11	cumulative	cumulative	ADJ
bracis-33663	167	12	”	"	PUNCT
bracis-33663	167	13	approach	approach	NOUN
bracis-33663	167	14	,	,	PUNCT
bracis-33663	167	15	indicating	indicate	VERB
bracis-33663	167	16	its	its	PRON
bracis-33663	167	17	value	value	NOUN
bracis-33663	167	18	in	in	ADP
bracis-33663	167	19	enhancing	enhance	VERB
bracis-33663	167	20	zshsd	zshsd	NOUN
bracis-33663	167	21	.	.	PUNCT
bracis-33663	168	1	“	"	PUNCT
bracis-33663	168	2	hateful	hateful	ADJ
bracis-33663	168	3	”	"	PUNCT
bracis-33663	168	4	:	:	PUNCT
bracis-33663	168	5	has	have	AUX
bracis-33663	168	6	a	a	DET
bracis-33663	168	7	strong	strong	ADJ
bracis-33663	168	8	impact	impact	NOUN
bracis-33663	168	9	in	in	ADP
bracis-33663	168	10	the	the	DET
bracis-33663	168	11	“	"	PUNCT
bracis-33663	168	12	cumulative	cumulative	ADJ
bracis-33663	168	13	”	"	PUNCT
bracis-33663	168	14	method	method	NOUN
bracis-33663	168	15	but	but	CCONJ
bracis-33663	168	16	decreases	decrease	NOUN
bracis-33663	168	17	in	in	ADP
bracis-33663	168	18	“	"	PUNCT
bracis-33663	168	19	individual	individual	ADJ
bracis-33663	168	20	”	"	PUNCT
bracis-33663	168	21	and	and	CCONJ
bracis-33663	168	22	“	"	PUNCT
bracis-33663	168	23	isolate	isolate	VERB
bracis-33663	168	24	,	,	PUNCT
bracis-33663	168	25	”	"	PUNCT
bracis-33663	168	26	suggesting	suggest	VERB
bracis-33663	168	27	its	its	PRON
bracis-33663	168	28	lesser	less	ADJ
bracis-33663	168	29	standalone	standalone	ADJ
bracis-33663	168	30	effect	effect	NOUN
bracis-33663	168	31	.	.	PUNCT
bracis-33663	169	1	“	"	PUNCT
bracis-33663	169	2	entries	entry	NOUN
bracis-33663	169	3	”	"	PUNCT
bracis-33663	169	4	and	and	CCONJ
bracis-33663	169	5	“	"	PUNCT
bracis-33663	169	6	or	or	CCONJ
bracis-33663	169	7	”	"	PUNCT
bracis-33663	169	8	:	:	PUNCT
bracis-33663	169	9	vary	vary	VERB
bracis-33663	169	10	in	in	ADP
bracis-33663	169	11	contribution	contribution	NOUN
bracis-33663	169	12	,	,	PUNCT
bracis-33663	169	13	with	with	ADP
bracis-33663	169	14	“	"	PUNCT
bracis-33663	169	15	cumulative	cumulative	ADJ
bracis-33663	169	16	”	"	PUNCT
bracis-33663	169	17	showing	show	VERB
bracis-33663	169	18	the	the	DET
bracis-33663	169	19	highest	high	ADJ
bracis-33663	169	20	values	value	NOUN
bracis-33663	169	21	.	.	PUNCT
bracis-33663	170	1	“	"	PUNCT
bracis-33663	170	2	or	or	CCONJ
bracis-33663	170	3	”	"	PUNCT
bracis-33663	170	4	contributes	contribute	VERB
bracis-33663	170	5	the	the	DET
bracis-33663	170	6	least	least	ADJ
bracis-33663	170	7	in	in	ADP
bracis-33663	170	8	the	the	DET
bracis-33663	170	9	“	"	PUNCT
bracis-33663	170	10	isolate	isolate	NOUN
bracis-33663	170	11	”	"	PUNCT
bracis-33663	170	12	approach	approach	NOUN
bracis-33663	170	13	.	.	PUNCT
bracis-33663	171	1	“	"	PUNCT
bracis-33663	171	2	offensive	offensive	ADJ
bracis-33663	171	3	”	"	PUNCT
bracis-33663	171	4	and	and	CCONJ
bracis-33663	171	5	“	"	PUNCT
bracis-33663	171	6	language	language	NOUN
bracis-33663	171	7	”	"	PUNCT
bracis-33663	171	8	:	:	PUNCT
bracis-33663	171	9	like	like	ADP
bracis-33663	171	10	“	"	PUNCT
bracis-33663	171	11	hateful	hateful	ADJ
bracis-33663	171	12	,	,	PUNCT
bracis-33663	171	13	”	"	PUNCT
bracis-33663	171	14	these	these	DET
bracis-33663	171	15	tokens	token	NOUN
bracis-33663	171	16	are	be	AUX
bracis-33663	171	17	crucial	crucial	ADJ
bracis-33663	171	18	in	in	ADP
bracis-33663	171	19	the	the	DET
bracis-33663	171	20	“	"	PUNCT
bracis-33663	171	21	cumulative	cumulative	ADJ
bracis-33663	171	22	”	"	PUNCT
bracis-33663	171	23	approach	approach	NOUN
bracis-33663	171	24	but	but	CCONJ
bracis-33663	171	25	less	less	ADV
bracis-33663	171	26	impactful	impactful	ADJ
bracis-33663	171	27	individually	individually	ADV
bracis-33663	171	28	.	.	PUNCT
bracis-33663	172	1	the	the	DET
bracis-33663	172	2	“	"	PUNCT
bracis-33663	172	3	cumulative	cumulative	ADJ
bracis-33663	172	4	”	"	PUNCT
bracis-33663	172	5	approach	approach	NOUN
bracis-33663	172	6	shows	show	VERB
bracis-33663	172	7	the	the	DET
bracis-33663	172	8	most	most	ADV
bracis-33663	172	9	significant	significant	ADJ
bracis-33663	172	10	performance	performance	NOUN
bracis-33663	172	11	improvement	improvement	NOUN
bracis-33663	172	12	as	as	SCONJ
bracis-33663	172	13	tokens	token	NOUN
bracis-33663	172	14	are	be	AUX
bracis-33663	172	15	sequentially	sequentially	ADV
bracis-33663	172	16	added	add	VERB
bracis-33663	172	17	,	,	PUNCT
bracis-33663	172	18	underscoring	underscore	VERB
bracis-33663	172	19	the	the	DET
bracis-33663	172	20	value	value	NOUN
bracis-33663	172	21	of	of	ADP
bracis-33663	172	22	considering	consider	VERB
bracis-33663	172	23	tokens	token	NOUN
bracis-33663	172	24	cumulatively	cumulatively	ADV
bracis-33663	172	25	for	for	ADP
bracis-33663	172	26	enhanced	enhanced	ADJ
bracis-33663	172	27	zshsd	zshsd	NOUN
bracis-33663	172	28	.	.	PUNCT
bracis-33663	173	1	however	however	ADV
bracis-33663	173	2	,	,	PUNCT
bracis-33663	173	3	the	the	DET
bracis-33663	173	4	impact	impact	NOUN
bracis-33663	173	5	of	of	ADP
bracis-33663	173	6	individual	individual	ADJ
bracis-33663	173	7	tokens	token	NOUN
bracis-33663	173	8	varies	vary	VERB
bracis-33663	173	9	,	,	PUNCT
bracis-33663	173	10	and	and	CCONJ
bracis-33663	173	11	not	not	PART
bracis-33663	173	12	all	all	DET
bracis-33663	173	13	tokens	token	NOUN
bracis-33663	173	14	contribute	contribute	VERB
bracis-33663	173	15	equally	equally	ADV
bracis-33663	173	16	.	.	PUNCT
bracis-33663	174	1	the	the	DET
bracis-33663	174	2	“	"	PUNCT
bracis-33663	174	3	cumulative	cumulative	ADJ
bracis-33663	174	4	”	"	PUNCT
bracis-33663	174	5	approach	approach	NOUN
bracis-33663	174	6	shows	show	VERB
bracis-33663	174	7	the	the	DET
bracis-33663	174	8	most	most	ADV
bracis-33663	174	9	significant	significant	ADJ
bracis-33663	174	10	performance	performance	NOUN
bracis-33663	174	11	improvement	improvement	NOUN
bracis-33663	174	12	,	,	PUNCT
bracis-33663	174	13	highlighting	highlight	VERB
bracis-33663	174	14	the	the	DET
bracis-33663	174	15	importance	importance	NOUN
bracis-33663	174	16	of	of	ADP
bracis-33663	174	17	considering	consider	VERB
bracis-33663	174	18	tokens	token	NOUN
bracis-33663	174	19	collectively	collectively	ADV
bracis-33663	174	20	for	for	ADP
bracis-33663	174	21	better	well	ADJ
bracis-33663	174	22	zshsd	zshsd	NOUN
bracis-33663	174	23	.	.	PUNCT
bracis-33663	175	1	for	for	ADP
bracis-33663	175	2	further	further	ADJ
bracis-33663	175	3	analysis	analysis	NOUN
bracis-33663	175	4	,	,	PUNCT
bracis-33663	175	5	the	the	DET
bracis-33663	175	6	tokens	token	NOUN
bracis-33663	175	7	“	"	PUNCT
bracis-33663	175	8	contains	contain	VERB
bracis-33663	175	9	,	,	PUNCT
bracis-33663	175	10	”	"	PUNCT
bracis-33663	175	11	“	"	PUNCT
bracis-33663	175	12	hateful	hateful	ADJ
bracis-33663	175	13	,	,	PUNCT
bracis-33663	175	14	”	"	PUNCT
bracis-33663	175	15	“	"	PUNCT
bracis-33663	175	16	or	or	CCONJ
bracis-33663	175	17	,	,	PUNCT
bracis-33663	175	18	”	"	PUNCT
bracis-33663	175	19	and	and	CCONJ
bracis-33663	175	20	“	"	PUNCT
bracis-33663	175	21	offensive	offensive	ADJ
bracis-33663	175	22	”	"	PUNCT
bracis-33663	175	23	were	be	AUX
bracis-33663	175	24	selected	select	VERB
bracis-33663	175	25	due	due	ADP
bracis-33663	175	26	to	to	ADP
bracis-33663	175	27	their	their	PRON
bracis-33663	175	28	pivotal	pivotal	ADJ
bracis-33663	175	29	roles	role	NOUN
bracis-33663	175	30	.	.	PUNCT
bracis-33663	176	1	the	the	DET
bracis-33663	176	2	“	"	PUNCT
bracis-33663	176	3	individual	individual	ADJ
bracis-33663	176	4	”	"	PUNCT
bracis-33663	176	5	method	method	NOUN
bracis-33663	176	6	was	be	AUX
bracis-33663	176	7	used	use	VERB
bracis-33663	176	8	for	for	ADP
bracis-33663	176	9	additional	additional	ADJ
bracis-33663	176	10	tests	test	NOUN
bracis-33663	176	11	:	:	PUNCT
bracis-33663	176	12	“	"	PUNCT
bracis-33663	176	13	contains	contain	VERB
bracis-33663	176	14	”	"	PUNCT
bracis-33663	176	15	:	:	PUNCT
bracis-33663	176	16	directly	directly	ADV
bracis-33663	176	17	relates	relate	VERB
bracis-33663	176	18	to	to	PART
bracis-33663	176	19	hate	hate	VERB
bracis-33663	176	20	speech	speech	NOUN
bracis-33663	176	21	identification	identification	NOUN
bracis-33663	176	22	.	.	PUNCT
bracis-33663	177	1	“	"	PUNCT
bracis-33663	177	2	hateful	hateful	ADJ
bracis-33663	177	3	”	"	PUNCT
bracis-33663	177	4	and	and	CCONJ
bracis-33663	177	5	“	"	PUNCT
bracis-33663	177	6	offensive	offensive	ADJ
bracis-33663	177	7	”	"	PUNCT
bracis-33663	177	8	:	:	PUNCT
bracis-33663	177	9	explicitly	explicitly	ADV
bracis-33663	177	10	linked	link	VERB
bracis-33663	177	11	to	to	PART
bracis-33663	177	12	hate	hate	VERB
bracis-33663	177	13	speech	speech	NOUN
bracis-33663	177	14	,	,	PUNCT
bracis-33663	177	15	enhancing	enhance	VERB
bracis-33663	177	16	the	the	DET
bracis-33663	177	17	model	model	NOUN
bracis-33663	177	18	’s	’s	PART
bracis-33663	177	19	discriminatory	discriminatory	ADJ
bracis-33663	177	20	power	power	NOUN
bracis-33663	177	21	.	.	PUNCT
bracis-33663	178	1	“	"	PUNCT
bracis-33663	178	2	or	or	CCONJ
bracis-33663	178	3	”	"	PUNCT
bracis-33663	178	4	:	:	PUNCT
bracis-33663	178	5	despite	despite	SCONJ
bracis-33663	178	6	being	be	AUX
bracis-33663	178	7	a	a	DET
bracis-33663	178	8	common	common	ADJ
bracis-33663	178	9	conjunction	conjunction	NOUN
bracis-33663	178	10	,	,	PUNCT
bracis-33663	178	11	it	it	PRON
bracis-33663	178	12	introduces	introduce	VERB
bracis-33663	178	13	nuanced	nuanced	ADJ
bracis-33663	178	14	interpretations	interpretation	NOUN
bracis-33663	178	15	with	with	ADP
bracis-33663	178	16	terms	term	NOUN
bracis-33663	178	17	like	like	ADP
bracis-33663	178	18	“	"	PUNCT
bracis-33663	178	19	antagonist	antagonist	NOUN
bracis-33663	178	20	”	"	PUNCT
bracis-33663	178	21	,	,	PUNCT
bracis-33663	178	22	“	"	PUNCT
bracis-33663	178	23	favorably	favorably	ADV
bracis-33663	178	24	”	"	PUNCT
bracis-33663	178	25	and	and	CCONJ
bracis-33663	178	26	“	"	PUNCT
bracis-33663	178	27	extremists	extremist	NOUN
bracis-33663	178	28	”	"	PUNCT
bracis-33663	178	29	demonstrating	demonstrate	VERB
bracis-33663	178	30	its	its	PRON
bracis-33663	178	31	adaptability	adaptability	NOUN
bracis-33663	178	32	in	in	ADP
bracis-33663	178	33	capturing	capture	VERB
bracis-33663	178	34	varied	varied	ADJ
bracis-33663	178	35	hate	hate	NOUN
bracis-33663	178	36	speech	speech	NOUN
bracis-33663	178	37	nuances	nuance	NOUN
bracis-33663	178	38	.	.	PUNCT
bracis-33663	179	1	table	table	NOUN
bracis-33663	179	2	 	 	SPACE
bracis-33663	179	3	5	5	NUM
bracis-33663	179	4	shows	show	VERB
bracis-33663	179	5	the	the	DET
bracis-33663	179	6	“	"	PUNCT
bracis-33663	179	7	full_optimized	full_optimize	VERB
bracis-33663	179	8	”	"	PUNCT
bracis-33663	179	9	hypothesis	hypothesis	NOUN
bracis-33663	179	10	achieving	achieve	VERB
bracis-33663	179	11	the	the	DET
bracis-33663	179	12	highest	high	ADJ
bracis-33663	179	13	accuracy	accuracy	NOUN
bracis-33663	179	14	(	(	PUNCT
bracis-33663	179	15	0.870	0.870	NUM
bracis-33663	179	16	)	)	PUNCT
bracis-33663	179	17	,	,	PUNCT
bracis-33663	179	18	indicating	indicate	VERB
bracis-33663	179	19	its	its	PRON
bracis-33663	179	20	effectiveness	effectiveness	NOUN
bracis-33663	179	21	.	.	PUNCT
bracis-33663	180	1	combinations	combination	NOUN
bracis-33663	180	2	including	include	VERB
bracis-33663	180	3	“	"	PUNCT
bracis-33663	180	4	hateful	hateful	ADJ
bracis-33663	180	5	”	"	PUNCT
bracis-33663	180	6	and	and	CCONJ
bracis-33663	180	7	“	"	PUNCT
bracis-33663	180	8	offensive	offensive	ADJ
bracis-33663	180	9	”	"	PUNCT
bracis-33663	180	10	also	also	ADV
bracis-33663	180	11	exhibit	exhibit	VERB
bracis-33663	180	12	high	high	ADJ
bracis-33663	180	13	accuracy	accuracy	NOUN
bracis-33663	180	14	,	,	PUNCT
bracis-33663	180	15	close	close	ADJ
bracis-33663	180	16	to	to	ADP
bracis-33663	180	17	the	the	DET
bracis-33663	180	18	“	"	PUNCT
bracis-33663	180	19	full_optimized	full_optimize	VERB
bracis-33663	180	20	”	"	PUNCT
bracis-33663	180	21	hypothesis	hypothesis	NOUN
bracis-33663	180	22	,	,	PUNCT
bracis-33663	180	23	suggesting	suggest	VERB
bracis-33663	180	24	that	that	SCONJ
bracis-33663	180	25	a	a	DET
bracis-33663	180	26	selected	select	VERB
bracis-33663	180	27	combination	combination	NOUN
bracis-33663	180	28	of	of	ADP
bracis-33663	180	29	tokens	token	NOUN
bracis-33663	180	30	can	can	AUX
bracis-33663	180	31	achieve	achieve	VERB
bracis-33663	180	32	similar	similar	ADJ
bracis-33663	180	33	results	result	NOUN
bracis-33663	180	34	.	.	PUNCT
bracis-33663	181	1	table	table	NOUN
bracis-33663	181	2	5	5	NUM
bracis-33663	181	3	.	.	PUNCT
bracis-33663	181	4	optimized	optimize	VERB
bracis-33663	181	5	virtual	virtual	ADJ
bracis-33663	181	6	tokens	token	NOUN
bracis-33663	181	7	inserted	insert	VERB
bracis-33663	181	8	into	into	ADP
bracis-33663	181	9	standard	standard	ADJ
bracis-33663	181	10	hypothesis	hypothesis	NOUN
bracis-33663	181	11	performancefull	performancefull	ADJ
bracis-33663	181	12	size	size	NOUN
bracis-33663	181	13	table	table	NOUN
bracis-33663	181	14	the	the	DET
bracis-33663	181	15	statistical	statistical	ADJ
bracis-33663	181	16	analysis	analysis	NOUN
bracis-33663	181	17	reveals	reveal	VERB
bracis-33663	181	18	a	a	DET
bracis-33663	181	19	significant	significant	ADJ
bracis-33663	181	20	improvement	improvement	NOUN
bracis-33663	181	21	of	of	ADP
bracis-33663	181	22	the	the	DET
bracis-33663	181	23	“	"	PUNCT
bracis-33663	181	24	full	full	ADJ
bracis-33663	181	25	optimized	optimize	VERB
bracis-33663	181	26	”	"	PUNCT
bracis-33663	181	27	sample	sample	NOUN
bracis-33663	181	28	over	over	ADP
bracis-33663	181	29	the	the	DET
bracis-33663	181	30	“	"	PUNCT
bracis-33663	181	31	pure	pure	ADJ
bracis-33663	181	32	hypothesis	hypothesis	NOUN
bracis-33663	181	33	”	"	PUNCT
bracis-33663	181	34	.	.	PUNCT
bracis-33663	182	1	considering	consider	VERB
bracis-33663	182	2	the	the	DET
bracis-33663	182	3	first	first	ADJ
bracis-33663	182	4	value	value	NOUN
bracis-33663	182	5	and	and	CCONJ
bracis-33663	182	6	second	second	ADJ
bracis-33663	182	7	value	value	NOUN
bracis-33663	182	8	as	as	ADP
bracis-33663	182	9	the	the	DET
bracis-33663	182	10	significance	significance	NOUN
bracis-33663	182	11	to	to	ADP
bracis-33663	182	12	“	"	PUNCT
bracis-33663	182	13	full	full	ADJ
bracis-33663	182	14	optimized	optimize	VERB
bracis-33663	182	15	”	"	PUNCT
bracis-33663	182	16	and	and	CCONJ
bracis-33663	182	17	“	"	PUNCT
bracis-33663	182	18	pure	pure	ADJ
bracis-33663	182	19	hypothesis	hypothesis	NOUN
bracis-33663	182	20	”	"	PUNCT
bracis-33663	182	21	respectively	respectively	ADV
bracis-33663	182	22	,	,	PUNCT
bracis-33663	182	23	among	among	ADP
bracis-33663	182	24	the	the	DET
bracis-33663	182	25	combinations	combination	NOUN
bracis-33663	182	26	analyzed	analyze	VERB
bracis-33663	182	27	,	,	PUNCT
bracis-33663	182	28	“	"	PUNCT
bracis-33663	182	29	contains	contain	VERB
bracis-33663	182	30	hateful	hateful	ADJ
bracis-33663	182	31	offensive	offensive	NOUN
bracis-33663	182	32	”	"	PUNCT
bracis-33663	182	33	(	(	PUNCT
bracis-33663	182	34	0.321	0.321	NUM
bracis-33663	182	35	vs.	vs.	ADP
bracis-33663	182	36	0.001	0.001	NUM
bracis-33663	182	37	)	)	PUNCT
bracis-33663	182	38	,	,	PUNCT
bracis-33663	182	39	“	"	PUNCT
bracis-33663	182	40	hateful	hateful	ADJ
bracis-33663	182	41	or	or	CCONJ
bracis-33663	182	42	offensive	offensive	ADJ
bracis-33663	182	43	”	"	PUNCT
bracis-33663	182	44	(	(	PUNCT
bracis-33663	182	45	0.216	0.216	NUM
bracis-33663	182	46	vs.	vs.	ADP
bracis-33663	182	47	0.003	0.003	NUM
bracis-33663	182	48	)	)	PUNCT
bracis-33663	182	49	,	,	PUNCT
bracis-33663	182	50	and	and	CCONJ
bracis-33663	182	51	“	"	PUNCT
bracis-33663	182	52	contains	contain	VERB
bracis-33663	182	53	hateful	hateful	ADJ
bracis-33663	182	54	or	or	CCONJ
bracis-33663	182	55	offensive	offensive	ADJ
bracis-33663	182	56	”	"	PUNCT
bracis-33663	182	57	(	(	PUNCT
bracis-33663	182	58	0.262	0.262	NUM
bracis-33663	182	59	vs.	vs.	ADP
bracis-33663	182	60	0.003	0.003	NUM
bracis-33663	182	61	)	)	PUNCT
bracis-33663	182	62	show	show	VERB
bracis-33663	182	63	significant	significant	ADJ
bracis-33663	182	64	improvements	improvement	NOUN
bracis-33663	182	65	when	when	SCONJ
bracis-33663	182	66	compared	compare	VERB
bracis-33663	182	67	to	to	ADP
bracis-33663	182	68	“	"	PUNCT
bracis-33663	182	69	h	h	PRON
bracis-33663	182	70	pure	pure	ADJ
bracis-33663	182	71	hypothesis	hypothesis	NOUN
bracis-33663	182	72	.	.	PUNCT
bracis-33663	182	73	”	"	PUNCT
bracis-33663	183	1	the	the	DET
bracis-33663	183	2	analyses	analysis	NOUN
bracis-33663	183	3	were	be	AUX
bracis-33663	183	4	conducted	conduct	VERB
bracis-33663	183	5	with	with	ADP
bracis-33663	183	6	a	a	DET
bracis-33663	183	7	95	95	NUM
bracis-33663	183	8	%	%	NOUN
bracis-33663	183	9	confidence	confidence	NOUN
bracis-33663	183	10	interval	interval	NOUN
bracis-33663	183	11	,	,	PUNCT
bracis-33663	183	12	considering	consider	VERB
bracis-33663	183	13	p	p	NOUN
bracis-33663	183	14	-	-	PUNCT
bracis-33663	183	15	values	value	NOUN
bracis-33663	183	16	below	below	ADP
bracis-33663	183	17	0.05	0.05	NUM
bracis-33663	183	18	as	as	ADV
bracis-33663	183	19	significant	significant	ADJ
bracis-33663	183	20	,	,	PUNCT
bracis-33663	183	21	thereby	thereby	ADV
bracis-33663	183	22	confirming	confirm	VERB
bracis-33663	183	23	the	the	DET
bracis-33663	183	24	alternative	alternative	ADJ
bracis-33663	183	25	hypothesis	hypothesis	NOUN
bracis-33663	183	26	.	.	PUNCT
bracis-33663	184	1	4.4	4.4	NUM
bracis-33663	184	2	previous	previous	ADJ
bracis-33663	184	3	work	work	NOUN
bracis-33663	184	4	comparation	comparation	NOUN
bracis-33663	184	5	table	table	NOUN
bracis-33663	184	6	 	 	SPACE
bracis-33663	184	7	6	6	NUM
bracis-33663	184	8	displays	display	NOUN
bracis-33663	184	9	accuracy	accuracy	NOUN
bracis-33663	184	10	results	result	NOUN
bracis-33663	184	11	from	from	ADP
bracis-33663	184	12	various	various	ADJ
bracis-33663	184	13	hate	hate	NOUN
bracis-33663	184	14	speech	speech	NOUN
bracis-33663	184	15	tests	test	NOUN
bracis-33663	184	16	,	,	PUNCT
bracis-33663	184	17	highlighting	highlight	VERB
bracis-33663	184	18	the	the	DET
bracis-33663	184	19	performance	performance	NOUN
bracis-33663	184	20	of	of	ADP
bracis-33663	184	21	different	different	ADJ
bracis-33663	184	22	hypothesis	hypothesis	NOUN
bracis-33663	184	23	configurations	configuration	NOUN
bracis-33663	184	24	,	,	PUNCT
bracis-33663	184	25	from	from	ADP
bracis-33663	184	26	pure	pure	ADJ
bracis-33663	184	27	hypotheses	hypothesis	NOUN
bracis-33663	184	28	to	to	ADP
bracis-33663	184	29	intermediate	intermediate	ADJ
bracis-33663	184	30	insertions	insertion	NOUN
bracis-33663	184	31	to	to	ADP
bracis-33663	184	32	fully	fully	ADV
bracis-33663	184	33	optimized	optimize	VERB
bracis-33663	184	34	hypotheses	hypothesis	NOUN
bracis-33663	184	35	.	.	PUNCT
bracis-33663	185	1	results	result	NOUN
bracis-33663	185	2	exceeding	exceed	VERB
bracis-33663	185	3	previous	previous	ADJ
bracis-33663	185	4	work	work	NOUN
bracis-33663	185	5	are	be	AUX
bracis-33663	185	6	marked	mark	VERB
bracis-33663	185	7	in	in	ADP
bracis-33663	185	8	green	green	NOUN
bracis-33663	185	9	.	.	PUNCT
bracis-33663	186	1	the	the	DET
bracis-33663	186	2	individual	individual	ADJ
bracis-33663	186	3	insert	insert	NOUN
bracis-33663	186	4	method	method	NOUN
bracis-33663	186	5	was	be	AUX
bracis-33663	186	6	used	use	VERB
bracis-33663	186	7	for	for	ADP
bracis-33663	186	8	intermediate	intermediate	ADJ
bracis-33663	186	9	steps	step	NOUN
bracis-33663	186	10	,	,	PUNCT
bracis-33663	186	11	as	as	SCONJ
bracis-33663	186	12	depicted	depict	VERB
bracis-33663	186	13	in	in	ADP
bracis-33663	186	14	fig	fig	NOUN
bracis-33663	186	15	.	.	PUNCT
bracis-33663	186	16	 	 	SPACE
bracis-33663	187	1	1	1	NUM
bracis-33663	187	2	.	.	PUNCT
bracis-33663	187	3	table	table	NOUN
bracis-33663	187	4	6	6	NUM
bracis-33663	187	5	.	.	PUNCT
bracis-33663	188	1	datasets	dataset	NOUN
bracis-33663	188	2	detailed	detail	VERB
bracis-33663	188	3	resultsfull	resultsfull	NOUN
bracis-33663	188	4	size	size	NOUN
bracis-33663	188	5	table	table	NOUN
bracis-33663	188	6	key	key	ADJ
bracis-33663	188	7	observations	observation	NOUN
bracis-33663	188	8	include	include	VERB
bracis-33663	188	9	an	an	DET
bracis-33663	188	10	increased	increase	VERB
bracis-33663	188	11	accuracy	accuracy	NOUN
bracis-33663	188	12	with	with	ADP
bracis-33663	188	13	the	the	DET
bracis-33663	188	14	addition	addition	NOUN
bracis-33663	188	15	of	of	ADP
bracis-33663	188	16	optimized	optimize	VERB
bracis-33663	188	17	tokens	token	NOUN
bracis-33663	188	18	to	to	ADP
bracis-33663	188	19	the	the	DET
bracis-33663	188	20	pure	pure	ADJ
bracis-33663	188	21	hypothesis	hypothesis	NOUN
bracis-33663	188	22	and	and	CCONJ
bracis-33663	188	23	a	a	DET
bracis-33663	188	24	general	general	ADJ
bracis-33663	188	25	trend	trend	NOUN
bracis-33663	188	26	of	of	ADP
bracis-33663	188	27	increasing	increase	VERB
bracis-33663	188	28	accuracy	accuracy	NOUN
bracis-33663	188	29	from	from	ADP
bracis-33663	188	30	initial	initial	ADJ
bracis-33663	188	31	to	to	ADP
bracis-33663	188	32	fully	fully	ADV
bracis-33663	188	33	optimized	optimize	VERB
bracis-33663	188	34	hypotheses	hypothesis	NOUN
bracis-33663	188	35	,	,	PUNCT
bracis-33663	188	36	underscoring	underscore	VERB
bracis-33663	188	37	the	the	DET
bracis-33663	188	38	effectiveness	effectiveness	NOUN
bracis-33663	188	39	of	of	ADP
bracis-33663	188	40	the	the	DET
bracis-33663	188	41	optimization	optimization	NOUN
bracis-33663	188	42	procedure	procedure	NOUN
bracis-33663	188	43	in	in	ADP
bracis-33663	188	44	enhancing	enhance	VERB
bracis-33663	188	45	the	the	DET
bracis-33663	188	46	model	model	NOUN
bracis-33663	188	47	’s	’s	PART
bracis-33663	188	48	zshsd	zshsd	NOUN
bracis-33663	188	49	capabilities	capability	NOUN
bracis-33663	188	50	.	.	PUNCT
bracis-33663	189	1	some	some	DET
bracis-33663	189	2	tests	test	NOUN
bracis-33663	189	3	show	show	VERB
bracis-33663	189	4	that	that	SCONJ
bracis-33663	189	5	specific	specific	ADJ
bracis-33663	189	6	intermediate	intermediate	ADJ
bracis-33663	189	7	steps	step	NOUN
bracis-33663	189	8	outperform	outperform	VERB
bracis-33663	189	9	others	other	NOUN
bracis-33663	189	10	,	,	PUNCT
bracis-33663	189	11	indicating	indicate	VERB
bracis-33663	189	12	the	the	DET
bracis-33663	189	13	need	need	NOUN
bracis-33663	189	14	to	to	PART
bracis-33663	189	15	tailor	tailor	VERB
bracis-33663	189	16	the	the	DET
bracis-33663	189	17	optimization	optimization	NOUN
bracis-33663	189	18	process	process	NOUN
bracis-33663	189	19	to	to	ADP
bracis-33663	189	20	different	different	ADJ
bracis-33663	189	21	hate	hate	NOUN
bracis-33663	189	22	speech	speech	NOUN
bracis-33663	189	23	categories	category	NOUN
bracis-33663	189	24	.	.	PUNCT
bracis-33663	190	1	intermediate	intermediate	ADJ
bracis-33663	190	2	representations	representation	NOUN
bracis-33663	190	3	such	such	ADJ
bracis-33663	190	4	as	as	ADP
bracis-33663	190	5	“	"	PUNCT
bracis-33663	190	6	hateful	hateful	ADJ
bracis-33663	190	7	or	or	CCONJ
bracis-33663	190	8	offensive	offensive	ADJ
bracis-33663	190	9	”	"	PUNCT
bracis-33663	190	10	,	,	PUNCT
bracis-33663	190	11	“	"	PUNCT
bracis-33663	190	12	contains	contain	VERB
bracis-33663	190	13	hateful	hateful	ADJ
bracis-33663	190	14	or	or	CCONJ
bracis-33663	190	15	offensive	offensive	ADJ
bracis-33663	190	16	”	"	PUNCT
bracis-33663	190	17	,	,	PUNCT
bracis-33663	190	18	and	and	CCONJ
bracis-33663	190	19	“	"	PUNCT
bracis-33663	190	20	contains	contain	VERB
bracis-33663	190	21	hateful	hateful	ADJ
bracis-33663	190	22	offensive	offensive	NOUN
bracis-33663	190	23	”	"	PUNCT
bracis-33663	190	24	consistently	consistently	ADV
bracis-33663	190	25	yield	yield	VERB
bracis-33663	190	26	high	high	ADJ
bracis-33663	190	27	accuracy	accuracy	NOUN
bracis-33663	190	28	across	across	ADP
bracis-33663	190	29	various	various	ADJ
bracis-33663	190	30	tests	test	NOUN
bracis-33663	190	31	,	,	PUNCT
bracis-33663	190	32	reinforcing	reinforce	VERB
bracis-33663	190	33	their	their	PRON
bracis-33663	190	34	importance	importance	NOUN
bracis-33663	190	35	in	in	ADP
bracis-33663	190	36	improving	improve	VERB
bracis-33663	190	37	.	.	PUNCT
bracis-33663	191	1	performance	performance	NOUN
bracis-33663	191	2	enhancements	enhancement	NOUN
bracis-33663	191	3	were	be	AUX
bracis-33663	191	4	notable	notable	ADJ
bracis-33663	191	5	when	when	SCONJ
bracis-33663	191	6	tested	test	VERB
bracis-33663	191	7	on	on	ADP
bracis-33663	191	8	the	the	DET
bracis-33663	191	9	learning	learning	NOUN
bracis-33663	191	10	from	from	ADP
bracis-33663	191	11	the	the	DET
bracis-33663	191	12	worst	bad	ADJ
bracis-33663	191	13	test	test	NOUN
bracis-33663	191	14	dataset	dataset	VERB
bracis-33663	191	15	[	[	X
bracis-33663	191	16	19	19	NUM
bracis-33663	191	17	]	]	PUNCT
bracis-33663	191	18	,	,	PUNCT
bracis-33663	191	19	with	with	ADP
bracis-33663	191	20	accuracy	accuracy	NOUN
bracis-33663	191	21	gains	gain	NOUN
bracis-33663	191	22	of	of	ADP
bracis-33663	191	23	5.5	5.5	NUM
bracis-33663	191	24	%	%	NOUN
bracis-33663	191	25	points	point	NOUN
bracis-33663	191	26	(	(	PUNCT
bracis-33663	191	27	pp	pp	NUM
bracis-33663	191	28	)	)	PUNCT
bracis-33663	191	29	and	and	CCONJ
bracis-33663	191	30	1.2	1.2	NUM
bracis-33663	191	31	pp	pp	ADJ
bracis-33663	191	32	on	on	ADP
bracis-33663	191	33	test	test	NOUN
bracis-33663	191	34	splits	split	VERB
bracis-33663	191	35	r2	r2	PROPN
bracis-33663	191	36	and	and	CCONJ
bracis-33663	191	37	r4	r4	NOUN
bracis-33663	191	38	,	,	PUNCT
bracis-33663	191	39	respectively	respectively	ADV
bracis-33663	191	40	.	.	PUNCT
bracis-33663	192	1	unlike	unlike	ADP
bracis-33663	192	2	previous	previous	ADJ
bracis-33663	192	3	efforts	effort	NOUN
bracis-33663	192	4	that	that	PRON
bracis-33663	192	5	relied	rely	VERB
bracis-33663	192	6	on	on	ADP
bracis-33663	192	7	fully	fully	ADV
bracis-33663	192	8	fine	fine	ADV
bracis-33663	192	9	-	-	PUNCT
bracis-33663	192	10	tuned	tune	VERB
bracis-33663	192	11	classification	classification	NOUN
bracis-33663	192	12	models	model	NOUN
bracis-33663	192	13	,	,	PUNCT
bracis-33663	192	14	our	our	PRON
bracis-33663	192	15	approach	approach	NOUN
bracis-33663	192	16	optimizes	optimize	VERB
bracis-33663	192	17	the	the	DET
bracis-33663	192	18	input	input	NOUN
bracis-33663	192	19	hypothesis	hypothesis	NOUN
bracis-33663	192	20	of	of	ADP
bracis-33663	192	21	a	a	DET
bracis-33663	192	22	nli	nli	ADJ
bracis-33663	192	23	model	model	NOUN
bracis-33663	192	24	.	.	PUNCT
bracis-33663	193	1	on	on	ADP
bracis-33663	193	2	the	the	DET
bracis-33663	193	3	hatecheck	hatecheck	NOUN
bracis-33663	193	4	dataset	dataset	NOUN
bracis-33663	194	1	[	[	X
bracis-33663	194	2	16	16	NUM
bracis-33663	194	3	]	]	PUNCT
bracis-33663	194	4	,	,	PUNCT
bracis-33663	194	5	optimized	optimize	VERB
bracis-33663	194	6	tokens	token	NOUN
bracis-33663	194	7	led	lead	VERB
bracis-33663	194	8	to	to	ADP
bracis-33663	194	9	an	an	DET
bracis-33663	194	10	average	average	ADJ
bracis-33663	194	11	accuracy	accuracy	NOUN
bracis-33663	194	12	improvement	improvement	NOUN
bracis-33663	194	13	of	of	ADP
bracis-33663	194	14	7.74	7.74	NUM
bracis-33663	194	15	pp	pp	NOUN
bracis-33663	194	16	,	,	PUNCT
bracis-33663	194	17	with	with	ADP
bracis-33663	194	18	a	a	DET
bracis-33663	194	19	best	good	ADJ
bracis-33663	194	20	-	-	PUNCT
bracis-33663	194	21	case	case	NOUN
bracis-33663	194	22	scenario	scenario	NOUN
bracis-33663	194	23	improvement	improvement	NOUN
bracis-33663	194	24	of	of	ADP
bracis-33663	194	25	18.8	18.8	NUM
bracis-33663	194	26	pp	pp	ADV
bracis-33663	194	27	using	use	VERB
bracis-33663	194	28	a	a	DET
bracis-33663	194	29	fully	fully	ADV
bracis-33663	194	30	tuned	tune	VERB
bracis-33663	194	31	hypothesis	hypothesis	NOUN
bracis-33663	194	32	(	(	PUNCT
bracis-33663	194	33	oh	oh	INTJ
bracis-33663	194	34	)	)	PUNCT
bracis-33663	194	35	compared	compare	VERB
bracis-33663	194	36	to	to	ADP
bracis-33663	194	37	(	(	PUNCT
bracis-33663	194	38	h	h	NOUN
bracis-33663	194	39	)	)	PUNCT
bracis-33663	194	40	.	.	PUNCT
bracis-33663	195	1	our	our	PRON
bracis-33663	195	2	fully	fully	ADV
bracis-33663	195	3	tuned	tune	VERB
bracis-33663	195	4	single	single	ADJ
bracis-33663	195	5	hypothesis	hypothesis	NOUN
bracis-33663	195	6	also	also	ADV
bracis-33663	195	7	outperformed	outperform	VERB
bracis-33663	195	8	previous	previous	ADJ
bracis-33663	195	9	work	work	NOUN
bracis-33663	195	10	by	by	ADP
bracis-33663	195	11	5.6	5.6	NUM
bracis-33663	195	12	pp	pp	NOUN
bracis-33663	195	13	.	.	PUNCT
bracis-33663	196	1	these	these	DET
bracis-33663	196	2	findings	finding	NOUN
bracis-33663	196	3	highlight	highlight	VERB
bracis-33663	196	4	the	the	DET
bracis-33663	196	5	effectiveness	effectiveness	NOUN
bracis-33663	196	6	of	of	ADP
bracis-33663	196	7	our	our	PRON
bracis-33663	196	8	optimization	optimization	NOUN
bracis-33663	196	9	procedure	procedure	NOUN
bracis-33663	196	10	in	in	ADP
bracis-33663	196	11	enhancing	enhance	VERB
bracis-33663	196	12	the	the	DET
bracis-33663	196	13	model	model	NOUN
bracis-33663	196	14	’s	’s	PART
bracis-33663	196	15	zshsd	zshsd	PROPN
bracis-33663	196	16	capabilities	capability	NOUN
bracis-33663	196	17	.	.	PUNCT
bracis-33663	197	1	both	both	DET
bracis-33663	197	2	fully	fully	ADV
bracis-33663	197	3	tuned	tune	VERB
bracis-33663	197	4	hypotheses	hypothesis	NOUN
bracis-33663	197	5	and	and	CCONJ
bracis-33663	197	6	significant	significant	ADJ
bracis-33663	197	7	token	token	VERB
bracis-33663	197	8	insertions	insertion	NOUN
bracis-33663	197	9	on	on	ADP
bracis-33663	197	10	standard	standard	ADJ
bracis-33663	197	11	hypotheses	hypothesis	NOUN
bracis-33663	197	12	showed	show	VERB
bracis-33663	197	13	improved	improved	ADJ
bracis-33663	197	14	accuracy	accuracy	NOUN
bracis-33663	197	15	,	,	PUNCT
bracis-33663	197	16	suggesting	suggest	VERB
bracis-33663	197	17	future	future	ADJ
bracis-33663	197	18	directions	direction	NOUN
bracis-33663	197	19	for	for	ADP
bracis-33663	197	20	broader	broad	ADJ
bracis-33663	197	21	hypothesis	hypothesis	NOUN
bracis-33663	197	22	design	design	NOUN
bracis-33663	197	23	within	within	ADP
bracis-33663	197	24	nli	nli	PROPN
bracis-33663	197	25	models	model	NOUN
bracis-33663	197	26	.	.	PUNCT
bracis-33663	198	1	5	5	NUM
bracis-33663	198	2	conclusion	conclusion	NOUN
bracis-33663	198	3	our	our	PRON
bracis-33663	198	4	study	study	NOUN
bracis-33663	198	5	underscores	underscore	VERB
bracis-33663	198	6	the	the	DET
bracis-33663	198	7	significance	significance	NOUN
bracis-33663	198	8	of	of	ADP
bracis-33663	198	9	thoughtful	thoughtful	ADJ
bracis-33663	198	10	token	token	ADJ
bracis-33663	198	11	selection	selection	NOUN
bracis-33663	198	12	and	and	CCONJ
bracis-33663	198	13	the	the	DET
bracis-33663	198	14	synergy	synergy	NOUN
bracis-33663	198	15	between	between	ADP
bracis-33663	198	16	different	different	ADJ
bracis-33663	198	17	aspects	aspect	NOUN
bracis-33663	198	18	of	of	ADP
bracis-33663	198	19	hypothesis	hypothesis	NOUN
bracis-33663	198	20	optimization	optimization	NOUN
bracis-33663	198	21	.	.	PUNCT
bracis-33663	199	1	by	by	ADP
bracis-33663	199	2	improving	improve	VERB
bracis-33663	199	3	how	how	SCONJ
bracis-33663	199	4	hypotheses	hypothesis	NOUN
bracis-33663	199	5	are	be	AUX
bracis-33663	199	6	formulated	formulate	VERB
bracis-33663	199	7	,	,	PUNCT
bracis-33663	199	8	we	we	PRON
bracis-33663	199	9	can	can	AUX
bracis-33663	199	10	enhance	enhance	VERB
bracis-33663	199	11	nli	nli	PROPN
bracis-33663	199	12	model	model	NOUN
bracis-33663	199	13	performance	performance	NOUN
bracis-33663	199	14	in	in	ADP
bracis-33663	199	15	zshsd	zshsd	PROPN
bracis-33663	199	16	,	,	PUNCT
bracis-33663	199	17	contributing	contribute	VERB
bracis-33663	199	18	to	to	ADP
bracis-33663	199	19	the	the	DET
bracis-33663	199	20	vital	vital	ADJ
bracis-33663	199	21	work	work	NOUN
bracis-33663	199	22	of	of	ADP
bracis-33663	199	23	online	online	ADJ
bracis-33663	199	24	content	content	NOUN
bracis-33663	199	25	moderation	moderation	NOUN
bracis-33663	199	26	and	and	CCONJ
bracis-33663	199	27	digital	digital	ADJ
bracis-33663	199	28	safety	safety	NOUN
bracis-33663	199	29	.	.	PUNCT
bracis-33663	200	1	our	our	PRON
bracis-33663	200	2	results	result	NOUN
bracis-33663	200	3	highlighted	highlight	VERB
bracis-33663	200	4	the	the	DET
bracis-33663	200	5	importance	importance	NOUN
bracis-33663	200	6	of	of	ADP
bracis-33663	200	7	specific	specific	ADJ
bracis-33663	200	8	tokens	token	NOUN
bracis-33663	200	9	like	like	ADP
bracis-33663	200	10	“	"	PUNCT
bracis-33663	200	11	hateful	hateful	ADJ
bracis-33663	200	12	”	"	PUNCT
bracis-33663	200	13	and	and	CCONJ
bracis-33663	200	14	“	"	PUNCT
bracis-33663	200	15	offensive	offensive	ADJ
bracis-33663	200	16	”	"	PUNCT
bracis-33663	200	17	in	in	ADP
bracis-33663	200	18	enhancing	enhance	VERB
bracis-33663	200	19	model	model	NOUN
bracis-33663	200	20	accuracy	accuracy	NOUN
bracis-33663	200	21	.	.	PUNCT
bracis-33663	201	1	by	by	ADP
bracis-33663	201	2	optimizing	optimize	VERB
bracis-33663	201	3	these	these	DET
bracis-33663	201	4	tokens	token	NOUN
bracis-33663	201	5	,	,	PUNCT
bracis-33663	201	6	we	we	PRON
bracis-33663	201	7	expanded	expand	VERB
bracis-33663	201	8	the	the	DET
bracis-33663	201	9	model	model	NOUN
bracis-33663	201	10	’s	’s	PART
bracis-33663	201	11	capability	capability	NOUN
bracis-33663	201	12	to	to	PART
bracis-33663	201	13	detect	detect	VERB
bracis-33663	201	14	a	a	DET
bracis-33663	201	15	wider	wide	ADJ
bracis-33663	201	16	range	range	NOUN
bracis-33663	201	17	of	of	ADP
bracis-33663	201	18	hateful	hateful	ADJ
bracis-33663	201	19	language	language	NOUN
bracis-33663	201	20	and	and	CCONJ
bracis-33663	201	21	ideologies	ideology	NOUN
bracis-33663	201	22	.	.	PUNCT
bracis-33663	202	1	for	for	ADP
bracis-33663	202	2	instance	instance	NOUN
bracis-33663	202	3	,	,	PUNCT
bracis-33663	202	4	the	the	DET
bracis-33663	202	5	optimized	optimize	VERB
bracis-33663	202	6	version	version	NOUN
bracis-33663	202	7	of	of	ADP
bracis-33663	202	8	“	"	PUNCT
bracis-33663	202	9	hateful	hateful	ADJ
bracis-33663	202	10	”	"	PUNCT
bracis-33663	202	11	introduced	introduce	VERB
bracis-33663	202	12	terms	term	NOUN
bracis-33663	202	13	like	like	ADP
bracis-33663	202	14	“	"	PUNCT
bracis-33663	202	15	fascist	fascist	NOUN
bracis-33663	202	16	”	"	PUNCT
bracis-33663	202	17	and	and	CCONJ
bracis-33663	202	18	“	"	PUNCT
bracis-33663	202	19	discriminatory	discriminatory	ADJ
bracis-33663	202	20	”	"	PUNCT
bracis-33663	202	21	enriching	enrich	VERB
bracis-33663	202	22	the	the	DET
bracis-33663	202	23	model	model	NOUN
bracis-33663	202	24	’s	’s	PART
bracis-33663	202	25	understanding	understanding	NOUN
bracis-33663	202	26	of	of	ADP
bracis-33663	202	27	hatred	hatred	NOUN
bracis-33663	202	28	and	and	CCONJ
bracis-33663	202	29	intolerance	intolerance	NOUN
bracis-33663	202	30	.	.	PUNCT
bracis-33663	203	1	similarly	similarly	ADV
bracis-33663	203	2	,	,	PUNCT
bracis-33663	203	3	optimizing	optimize	VERB
bracis-33663	203	4	“	"	PUNCT
bracis-33663	203	5	offensive	offensive	ADJ
bracis-33663	203	6	”	"	PUNCT
bracis-33663	203	7	incorporated	incorporate	VERB
bracis-33663	203	8	terms	term	NOUN
bracis-33663	203	9	like	like	ADP
bracis-33663	203	10	“	"	PUNCT
bracis-33663	203	11	transsexual	transsexual	ADJ
bracis-33663	203	12	”	"	PUNCT
bracis-33663	203	13	,	,	PUNCT
bracis-33663	203	14	“	"	PUNCT
bracis-33663	203	15	lgbt	lgbt	PROPN
bracis-33663	203	16	”	"	PUNCT
bracis-33663	203	17	and	and	CCONJ
bracis-33663	203	18	“	"	PUNCT
bracis-33663	203	19	ethnic	ethnic	ADJ
bracis-33663	203	20	”	"	PUNCT
bracis-33663	203	21	which	which	PRON
bracis-33663	203	22	are	be	AUX
bracis-33663	203	23	relevant	relevant	ADJ
bracis-33663	203	24	to	to	PART
bracis-33663	203	25	hate	hate	VERB
bracis-33663	203	26	speech	speech	NOUN
bracis-33663	203	27	detection	detection	NOUN
bracis-33663	203	28	.	.	PUNCT
bracis-33663	204	1	these	these	DET
bracis-33663	204	2	additions	addition	NOUN
bracis-33663	204	3	improved	improve	VERB
bracis-33663	204	4	the	the	DET
bracis-33663	204	5	model	model	NOUN
bracis-33663	204	6	’s	’s	PART
bracis-33663	204	7	ability	ability	NOUN
bracis-33663	204	8	to	to	PART
bracis-33663	204	9	recognize	recognize	VERB
bracis-33663	204	10	subtle	subtle	ADJ
bracis-33663	204	11	nuances	nuance	NOUN
bracis-33663	204	12	in	in	ADP
bracis-33663	204	13	discriminatory	discriminatory	ADJ
bracis-33663	204	14	language	language	NOUN
bracis-33663	204	15	.	.	PUNCT
bracis-33663	205	1	future	future	ADJ
bracis-33663	205	2	work	work	NOUN
bracis-33663	205	3	should	should	AUX
bracis-33663	205	4	focus	focus	VERB
bracis-33663	205	5	on	on	ADP
bracis-33663	205	6	expanding	expand	VERB
bracis-33663	205	7	and	and	CCONJ
bracis-33663	205	8	diversifying	diversify	VERB
bracis-33663	205	9	token	token	ADJ
bracis-33663	205	10	pools	pool	NOUN
bracis-33663	205	11	,	,	PUNCT
bracis-33663	205	12	adapting	adapt	VERB
bracis-33663	205	13	models	model	NOUN
bracis-33663	205	14	for	for	ADP
bracis-33663	205	15	multilingual	multilingual	ADJ
bracis-33663	205	16	contexts	context	NOUN
bracis-33663	205	17	,	,	PUNCT
bracis-33663	205	18	and	and	CCONJ
bracis-33663	205	19	developing	develop	VERB
bracis-33663	205	20	real	real	ADJ
bracis-33663	205	21	-	-	PUNCT
bracis-33663	205	22	time	time	NOUN
bracis-33663	205	23	adaptation	adaptation	NOUN
bracis-33663	205	24	mechanisms	mechanism	NOUN
bracis-33663	205	25	to	to	PART
bracis-33663	205	26	keep	keep	VERB
bracis-33663	205	27	up	up	ADP
bracis-33663	205	28	with	with	ADP
bracis-33663	205	29	evolving	evolve	VERB
bracis-33663	205	30	hate	hate	NOUN
bracis-33663	205	31	speech	speech	NOUN
bracis-33663	205	32	.	.	PUNCT
bracis-33663	206	1	enhancing	enhance	VERB
bracis-33663	206	2	contextual	contextual	ADJ
bracis-33663	206	3	analysis	analysis	NOUN
bracis-33663	206	4	,	,	PUNCT
bracis-33663	206	5	incorporating	incorporate	VERB
bracis-33663	206	6	user	user	NOUN
bracis-33663	206	7	feedback	feedback	NOUN
bracis-33663	206	8	,	,	PUNCT
bracis-33663	206	9	addressing	address	VERB
bracis-33663	206	10	ethical	ethical	ADJ
bracis-33663	206	11	and	and	CCONJ
bracis-33663	206	12	bias	bias	NOUN
bracis-33663	206	13	issues	issue	NOUN
bracis-33663	206	14	,	,	PUNCT
bracis-33663	206	15	are	be	AUX
bracis-33663	206	16	crucial	crucial	ADJ
bracis-33663	206	17	for	for	ADP
bracis-33663	206	18	improving	improve	VERB
bracis-33663	206	19	accuracy	accuracy	NOUN
bracis-33663	206	20	and	and	CCONJ
bracis-33663	206	21	practical	practical	ADJ
bracis-33663	206	22	utility	utility	NOUN
bracis-33663	206	23	in	in	ADP
bracis-33663	206	24	online	online	ADJ
bracis-33663	206	25	content	content	NOUN
bracis-33663	206	26	moderation	moderation	NOUN
bracis-33663	206	27	.	.	PUNCT
bracis-33663	207	1	notes	note	NOUN
bracis-33663	207	2	1.https://anonymous.4open.science/r/combining-discrete-and-continuous-spaces-for-zero-shot-hate-speech-detection-fb34/readme.md	1.https://anonymous.4open.science/r/combining-discrete-and-continuous-spaces-for-zero-shot-hate-speech-detection-fb34/readme.md	PROPN
bracis-33663	207	3	.	.	PUNCT
bracis-33663	208	1	2.https://huggingface.co/roberta-large-mnli	2.https://huggingface.co/roberta-large-mnli	NUM
bracis-33663	208	2	.	.	PUNCT
bracis-33663	209	1	references	reference	NOUN
bracis-33663	209	2	arango	arango	PROPN
bracis-33663	209	3	,	,	PUNCT
bracis-33663	209	4	a.	a.	NOUN
bracis-33663	209	5	,	,	PUNCT
bracis-33663	209	6	pérez	pérez	NOUN
bracis-33663	209	7	,	,	PUNCT
bracis-33663	209	8	j.	j.	PROPN
bracis-33663	209	9	,	,	PUNCT
bracis-33663	209	10	poblete	poblete	PROPN
bracis-33663	209	11	,	,	PUNCT
bracis-33663	209	12	b.	b.	PROPN
bracis-33663	209	13	:	:	PUNCT
bracis-33663	209	14	hate	hate	VERB
bracis-33663	209	15	speech	speech	NOUN
bracis-33663	209	16	detection	detection	NOUN
bracis-33663	209	17	is	be	AUX
bracis-33663	209	18	not	not	PART
bracis-33663	209	19	as	as	ADV
bracis-33663	209	20	easy	easy	ADJ
bracis-33663	209	21	as	as	SCONJ
bracis-33663	209	22	you	you	PRON
bracis-33663	209	23	may	may	AUX
bracis-33663	209	24	think	think	VERB
bracis-33663	209	25	:	:	PUNCT
bracis-33663	209	26	a	a	DET
bracis-33663	209	27	closer	close	ADJ
bracis-33663	209	28	look	look	NOUN
bracis-33663	209	29	at	at	ADP
bracis-33663	209	30	model	model	NOUN
bracis-33663	209	31	validation	validation	NOUN
bracis-33663	209	32	.	.	PUNCT
bracis-33663	210	1	in	in	ADP
bracis-33663	210	2	:	:	PUNCT
bracis-33663	210	3	proceedings	proceeding	NOUN
bracis-33663	210	4	of	of	ADP
bracis-33663	210	5	the	the	DET
bracis-33663	210	6	42nd	42nd	PROPN
bracis-33663	210	7	international	international	PROPN
bracis-33663	210	8	acm	acm	PROPN
bracis-33663	210	9	sigir	sigir	PROPN
bracis-33663	210	10	conference	conference	PROPN
bracis-33663	210	11	on	on	ADP
bracis-33663	210	12	research	research	NOUN
bracis-33663	210	13	and	and	CCONJ
bracis-33663	210	14	development	development	NOUN
bracis-33663	210	15	in	in	ADP
bracis-33663	210	16	information	information	NOUN
bracis-33663	210	17	retrieval	retrieval	NOUN
bracis-33663	210	18	,	,	PUNCT
bracis-33663	210	19	pp	pp	ADV
bracis-33663	210	20	.	.	PUNCT
bracis-33663	211	1	45–54	45–54	NUM
bracis-33663	211	2	(	(	PUNCT
bracis-33663	211	3	2019	2019	NUM
bracis-33663	211	4	)	)	PUNCT
bracis-33663	211	5	google	google	NOUN
bracis-33663	211	6	scholar	scholar	NOUN
bracis-33663	211	7	  	  	SPACE
bracis-33663	211	8	davidson	davidson	NOUN
bracis-33663	211	9	,	,	PUNCT
bracis-33663	211	10	t.	t.	PROPN
bracis-33663	211	11	,	,	PUNCT
bracis-33663	211	12	warmsley	warmsley	PROPN
bracis-33663	211	13	,	,	PUNCT
bracis-33663	211	14	d.	d.	PROPN
bracis-33663	211	15	,	,	PUNCT
bracis-33663	211	16	macy	macy	PROPN
bracis-33663	211	17	,	,	PUNCT
bracis-33663	211	18	m.	m.	NOUN
bracis-33663	211	19	,	,	PUNCT
bracis-33663	211	20	weber	weber	PROPN
bracis-33663	211	21	,	,	PUNCT
bracis-33663	211	22	i.	i.	PROPN
bracis-33663	211	23	:	:	PUNCT
bracis-33663	211	24	automated	automate	VERB
bracis-33663	211	25	hate	hate	NOUN
bracis-33663	211	26	speech	speech	NOUN
bracis-33663	211	27	detection	detection	NOUN
bracis-33663	211	28	and	and	CCONJ
bracis-33663	211	29	the	the	DET
bracis-33663	211	30	problem	problem	NOUN
bracis-33663	211	31	of	of	ADP
bracis-33663	211	32	offensive	offensive	ADJ
bracis-33663	211	33	language	language	NOUN
bracis-33663	211	34	.	.	PUNCT
bracis-33663	212	1	in	in	ADP
bracis-33663	212	2	:	:	PUNCT
bracis-33663	212	3	proceedings	proceeding	NOUN
bracis-33663	212	4	of	of	ADP
bracis-33663	212	5	the	the	DET
bracis-33663	212	6	international	international	ADJ
bracis-33663	212	7	aaai	aaai	PROPN
bracis-33663	212	8	conference	conference	PROPN
bracis-33663	212	9	on	on	ADP
bracis-33663	212	10	web	web	NOUN
bracis-33663	212	11	and	and	CCONJ
bracis-33663	212	12	social	social	ADJ
bracis-33663	212	13	media	medium	NOUN
bracis-33663	212	14	,	,	PUNCT
bracis-33663	212	15	vol	vol	NOUN
bracis-33663	212	16	.	.	PUNCT
bracis-33663	212	17	 	 	SPACE
bracis-33663	212	18	11	11	NUM
bracis-33663	212	19	,	,	PUNCT
bracis-33663	212	20	pp	pp	ADJ
bracis-33663	212	21	.	.	PUNCT
bracis-33663	213	1	512–515	512–515	NUM
bracis-33663	213	2	(	(	PUNCT
bracis-33663	213	3	2017	2017	NUM
bracis-33663	213	4	)	)	PUNCT
bracis-33663	213	5	.	.	PUNCT
bracis-33663	214	1	https://doi.org/10.1609/icwsm.v11i1.14955	https://doi.org/10.1609/icwsm.v11i1.14955	VERB
bracis-33663	214	2	devlin	devlin	PROPN
bracis-33663	214	3	,	,	PUNCT
bracis-33663	214	4	j.	j.	PROPN
bracis-33663	214	5	,	,	PUNCT
bracis-33663	214	6	chang	chang	PROPN
bracis-33663	214	7	,	,	PUNCT
bracis-33663	214	8	m.w	m.w	PROPN
bracis-33663	214	9	.	.	PROPN
bracis-33663	214	10	,	,	PUNCT
bracis-33663	214	11	lee	lee	PROPN
bracis-33663	214	12	,	,	PUNCT
bracis-33663	214	13	k.	k.	PROPN
bracis-33663	214	14	,	,	PUNCT
bracis-33663	214	15	toutanova	toutanova	PROPN
bracis-33663	214	16	,	,	PUNCT
bracis-33663	214	17	k.	k.	PROPN
bracis-33663	214	18	:	:	PUNCT
bracis-33663	215	1	bert	bert	PROPN
bracis-33663	215	2	:	:	PUNCT
bracis-33663	215	3	pre	pre	ADJ
bracis-33663	215	4	-	-	NOUN
bracis-33663	215	5	training	training	NOUN
bracis-33663	215	6	of	of	ADP
bracis-33663	215	7	deep	deep	ADJ
bracis-33663	215	8	bidirectional	bidirectional	ADJ
bracis-33663	215	9	transformers	transformer	NOUN
bracis-33663	215	10	for	for	ADP
bracis-33663	215	11	language	language	NOUN
bracis-33663	215	12	understanding	understanding	NOUN
bracis-33663	215	13	.	.	PUNCT
bracis-33663	216	1	in	in	ADP
bracis-33663	216	2	:	:	PUNCT
bracis-33663	216	3	proceedings	proceeding	NOUN
bracis-33663	216	4	of	of	ADP
bracis-33663	216	5	naacl	naacl	PROPN
bracis-33663	216	6	-	-	PUNCT
bracis-33663	216	7	hlt	hlt	PROPN
bracis-33663	216	8	,	,	PUNCT
bracis-33663	216	9	vol	vol	NOUN
bracis-33663	216	10	.	.	PUNCT
bracis-33663	216	11	 	 	SPACE
bracis-33663	216	12	1	1	NUM
bracis-33663	216	13	,	,	PUNCT
bracis-33663	216	14	p.	p.	NOUN
bracis-33663	216	15	 	 	SPACE
bracis-33663	216	16	2	2	NUM
bracis-33663	216	17	(	(	PUNCT
bracis-33663	216	18	2019	2019	NUM
bracis-33663	216	19	)	)	PUNCT
bracis-33663	216	20	.	.	PUNCT
bracis-33663	217	1	https://doi.org/10.18653/v1/n19-1423	https://doi.org/10.18653/v1/n19-1423	PROPN
bracis-33663	217	2	.	.	PUNCT
bracis-33663	217	3	https://aclanthology.org/n19-1423	https://aclanthology.org/n19-1423	PROPN
bracis-33663	217	4	fortuna	fortuna	PROPN
bracis-33663	217	5	,	,	PUNCT
bracis-33663	217	6	p.	p.	PROPN
bracis-33663	217	7	,	,	PUNCT
bracis-33663	217	8	dominguez	dominguez	PROPN
bracis-33663	217	9	,	,	PUNCT
bracis-33663	217	10	m.	m.	NOUN
bracis-33663	217	11	,	,	PUNCT
bracis-33663	217	12	wanner	wanner	NOUN
bracis-33663	217	13	,	,	PUNCT
bracis-33663	217	14	l.	l.	PROPN
bracis-33663	217	15	,	,	PUNCT
bracis-33663	217	16	talat	talat	PROPN
bracis-33663	217	17	,	,	PUNCT
bracis-33663	217	18	z.	z.	PROPN
bracis-33663	217	19	:	:	PUNCT
bracis-33663	217	20	directions	direction	NOUN
bracis-33663	217	21	for	for	ADP
bracis-33663	217	22	nlp	nlp	NOUN
bracis-33663	217	23	practices	practice	NOUN
bracis-33663	217	24	applied	apply	VERB
bracis-33663	217	25	to	to	ADP
bracis-33663	217	26	online	online	ADJ
bracis-33663	217	27	hate	hate	NOUN
bracis-33663	217	28	speech	speech	NOUN
bracis-33663	217	29	detection	detection	NOUN
bracis-33663	217	30	.	.	PUNCT
bracis-33663	218	1	in	in	ADP
bracis-33663	218	2	:	:	PUNCT
bracis-33663	218	3	goldberg	goldberg	PROPN
bracis-33663	218	4	,	,	PUNCT
bracis-33663	218	5	y.	y.	PROPN
bracis-33663	218	6	,	,	PUNCT
bracis-33663	218	7	kozareva	kozareva	PROPN
bracis-33663	218	8	,	,	PUNCT
bracis-33663	218	9	z.	z.	PROPN
bracis-33663	218	10	,	,	PUNCT
bracis-33663	218	11	zhang	zhang	PROPN
bracis-33663	218	12	,	,	PUNCT
bracis-33663	218	13	y.	y.	PROPN
bracis-33663	218	14	(	(	PUNCT
bracis-33663	218	15	eds	eds	PROPN
bracis-33663	218	16	.	.	PUNCT
bracis-33663	218	17	)	)	PUNCT
bracis-33663	218	18	proceedings	proceeding	NOUN
bracis-33663	218	19	of	of	ADP
bracis-33663	218	20	the	the	DET
bracis-33663	218	21	2022	2022	NUM
bracis-33663	218	22	conference	conference	NOUN
bracis-33663	218	23	on	on	ADP
bracis-33663	218	24	empirical	empirical	ADJ
bracis-33663	218	25	methods	method	NOUN
bracis-33663	218	26	in	in	ADP
bracis-33663	218	27	natural	natural	ADJ
bracis-33663	218	28	language	language	NOUN
bracis-33663	218	29	processing	processing	NOUN
bracis-33663	218	30	,	,	PUNCT
bracis-33663	218	31	pp	pp	ADJ
bracis-33663	218	32	.	.	PUNCT
bracis-33663	219	1	11794–11805	11794–11805	NUM
bracis-33663	219	2	.	.	PUNCT
bracis-33663	220	1	association	association	NOUN
bracis-33663	220	2	for	for	ADP
bracis-33663	220	3	computational	computational	ADJ
bracis-33663	220	4	linguistics	linguistic	NOUN
bracis-33663	220	5	,	,	PUNCT
bracis-33663	220	6	abu	abu	PROPN
bracis-33663	220	7	dhabi	dhabi	PROPN
bracis-33663	220	8	,	,	PUNCT
bracis-33663	220	9	united	united	PROPN
bracis-33663	220	10	arab	arab	PROPN
bracis-33663	220	11	emirates	emirates	PROPN
bracis-33663	220	12	(	(	PUNCT
bracis-33663	220	13	2022	2022	NUM
bracis-33663	220	14	)	)	PUNCT
bracis-33663	220	15	.	.	PUNCT
bracis-33663	221	1	https://doi.org/10.18653/v1/2022.emnlp-main.809	https://doi.org/10.18653/v1/2022.emnlp-main.809	NOUN
bracis-33663	221	2	.	.	PUNCT
bracis-33663	222	1	https://aclanthology.org/2022.emnlp-main.809	https://aclanthology.org/2022.emnlp-main.809	PROPN
bracis-33663	222	2	fortuna	fortuna	PROPN
bracis-33663	222	3	,	,	PUNCT
bracis-33663	222	4	p.	p.	PROPN
bracis-33663	222	5	,	,	PUNCT
bracis-33663	222	6	nunes	nunes	PROPN
bracis-33663	222	7	,	,	PUNCT
bracis-33663	222	8	s.	s.	PROPN
bracis-33663	222	9	:	:	PUNCT
bracis-33663	222	10	a	a	DET
bracis-33663	222	11	survey	survey	NOUN
bracis-33663	222	12	on	on	ADP
bracis-33663	222	13	automatic	automatic	ADJ
bracis-33663	222	14	detection	detection	NOUN
bracis-33663	222	15	of	of	ADP
bracis-33663	222	16	hate	hate	NOUN
bracis-33663	222	17	speech	speech	NOUN
bracis-33663	222	18	in	in	ADP
bracis-33663	222	19	text	text	NOUN
bracis-33663	222	20	.	.	PUNCT
bracis-33663	223	1	acm	acm	PROPN
bracis-33663	223	2	comput	comput	NOUN
bracis-33663	223	3	.	.	PUNCT
bracis-33663	224	1	surv	surv	NOUN
bracis-33663	224	2	.	.	PUNCT
bracis-33663	225	1	51(4	51(4	X
bracis-33663	225	2	)	)	PUNCT
bracis-33663	225	3	(	(	PUNCT
bracis-33663	225	4	2018	2018	NUM
bracis-33663	225	5	)	)	PUNCT
bracis-33663	225	6	.	.	PUNCT
bracis-33663	226	1	https://doi.org/10.1145/3232676	https://doi.org/10.1145/3232676	PROPN
bracis-33663	226	2	fortuna	fortuna	PROPN
bracis-33663	226	3	,	,	PUNCT
bracis-33663	226	4	p.	p.	NOUN
bracis-33663	226	5	,	,	PUNCT
bracis-33663	226	6	soler	soler	NOUN
bracis-33663	226	7	-	-	PUNCT
bracis-33663	226	8	company	company	NOUN
bracis-33663	226	9	,	,	PUNCT
bracis-33663	226	10	j.	j.	PROPN
bracis-33663	226	11	,	,	PUNCT
bracis-33663	226	12	wanner	wanner	NOUN
bracis-33663	226	13	,	,	PUNCT
bracis-33663	226	14	l.	l.	PROPN
bracis-33663	226	15	:	:	PUNCT
bracis-33663	226	16	how	how	SCONJ
bracis-33663	226	17	well	well	ADV
bracis-33663	226	18	do	do	AUX
bracis-33663	226	19	hate	hate	VERB
bracis-33663	226	20	speech	speech	NOUN
bracis-33663	226	21	,	,	PUNCT
bracis-33663	226	22	toxicity	toxicity	NOUN
bracis-33663	226	23	,	,	PUNCT
bracis-33663	226	24	abusive	abusive	ADJ
bracis-33663	226	25	and	and	CCONJ
bracis-33663	226	26	offensive	offensive	ADJ
bracis-33663	226	27	language	language	NOUN
bracis-33663	226	28	classification	classification	NOUN
bracis-33663	226	29	models	model	NOUN
bracis-33663	226	30	generalize	generalize	VERB
bracis-33663	226	31	across	across	ADP
bracis-33663	226	32	datasets	dataset	NOUN
bracis-33663	226	33	?	?	PUNCT
bracis-33663	227	1	inf	inf	PROPN
bracis-33663	227	2	.	.	PUNCT
bracis-33663	227	3	process	process	NOUN
bracis-33663	227	4	.	.	PUNCT
bracis-33663	228	1	manag	manag	PROPN
bracis-33663	228	2	.	.	PUNCT
bracis-33663	229	1	58(3	58(3	NUM
bracis-33663	229	2	)	)	PUNCT
bracis-33663	229	3	,	,	PUNCT
bracis-33663	229	4	102524	102524	NUM
bracis-33663	229	5	(	(	PUNCT
bracis-33663	229	6	2021	2021	NUM
bracis-33663	229	7	)	)	PUNCT
bracis-33663	229	8	.	.	PUNCT
bracis-33663	230	1	https://doi.org/10.1016/j.ipm.2021.102524	https://doi.org/10.1016/j.ipm.2021.102524	PROPN
bracis-33663	230	2	.	.	PUNCT
bracis-33663	231	1	https://www.sciencedirect.com/science/article/pii/s0306457321000339	https://www.sciencedirect.com/science/article/pii/s0306457321000339	PROPN
bracis-33663	231	2	goldzycher	goldzycher	PROPN
bracis-33663	231	3	,	,	PUNCT
bracis-33663	231	4	j.	j.	PROPN
bracis-33663	231	5	,	,	PUNCT
bracis-33663	231	6	schneider	schneider	PROPN
bracis-33663	231	7	,	,	PUNCT
bracis-33663	231	8	g.	g.	NOUN
bracis-33663	231	9	:	:	PUNCT
bracis-33663	231	10	hypothesis	hypothesis	NOUN
bracis-33663	231	11	engineering	engineering	NOUN
bracis-33663	231	12	for	for	ADP
bracis-33663	231	13	zero	zero	NUM
bracis-33663	231	14	-	-	PUNCT
bracis-33663	231	15	shot	shot	NOUN
bracis-33663	231	16	hate	hate	NOUN
bracis-33663	231	17	speech	speech	NOUN
bracis-33663	231	18	detection	detection	NOUN
bracis-33663	231	19	.	.	PUNCT
bracis-33663	232	1	in	in	ADP
bracis-33663	232	2	:	:	PUNCT
bracis-33663	232	3	proceedings	proceeding	NOUN
bracis-33663	232	4	of	of	ADP
bracis-33663	232	5	the	the	DET
bracis-33663	232	6	third	third	ADJ
bracis-33663	232	7	workshop	workshop	NOUN
bracis-33663	232	8	on	on	ADP
bracis-33663	232	9	threat	threat	NOUN
bracis-33663	232	10	,	,	PUNCT
bracis-33663	232	11	aggression	aggression	NOUN
bracis-33663	232	12	and	and	CCONJ
bracis-33663	232	13	cyberbullying	cyberbullye	VERB
bracis-33663	232	14	(	(	PUNCT
bracis-33663	232	15	trac	trac	PROPN
bracis-33663	232	16	2022	2022	NUM
bracis-33663	232	17	)	)	PUNCT
bracis-33663	232	18	,	,	PUNCT
bracis-33663	232	19	pp	pp	ADV
bracis-33663	232	20	.	.	PUNCT
bracis-33663	233	1	75–90	75–90	NUM
bracis-33663	233	2	.	.	PUNCT
bracis-33663	234	1	association	association	NOUN
bracis-33663	234	2	for	for	ADP
bracis-33663	234	3	computational	computational	ADJ
bracis-33663	234	4	linguistics	linguistic	NOUN
bracis-33663	234	5	,	,	PUNCT
bracis-33663	234	6	gyeongju	gyeongju	NOUN
bracis-33663	234	7	,	,	PUNCT
bracis-33663	234	8	republic	republic	NOUN
bracis-33663	234	9	of	of	ADP
bracis-33663	234	10	korea	korea	PROPN
bracis-33663	234	11	(	(	PUNCT
bracis-33663	234	12	2022	2022	NUM
bracis-33663	234	13	)	)	PUNCT
bracis-33663	234	14	.	.	PUNCT
bracis-33663	235	1	https://aclanthology.org/2022.trac-1.10	https://aclanthology.org/2022.trac-1.10	PROPN
bracis-33663	235	2	gröndahl	gröndahl	NOUN
bracis-33663	235	3	,	,	PUNCT
bracis-33663	235	4	t.	t.	PROPN
bracis-33663	235	5	,	,	PUNCT
bracis-33663	235	6	pajola	pajola	PROPN
bracis-33663	235	7	,	,	PUNCT
bracis-33663	235	8	l.	l.	PROPN
bracis-33663	235	9	,	,	PUNCT
bracis-33663	235	10	juuti	juuti	PROPN
bracis-33663	235	11	,	,	PUNCT
bracis-33663	235	12	m.	m.	NOUN
bracis-33663	235	13	,	,	PUNCT
bracis-33663	235	14	conti	conti	PROPN
bracis-33663	235	15	,	,	PUNCT
bracis-33663	235	16	m.	m.	NOUN
bracis-33663	235	17	,	,	PUNCT
bracis-33663	235	18	asokan	asokan	ADJ
bracis-33663	235	19	,	,	PUNCT
bracis-33663	235	20	n.	n.	NOUN
bracis-33663	235	21	:	:	PUNCT
bracis-33663	235	22	all	all	PRON
bracis-33663	235	23	you	you	PRON
bracis-33663	235	24	need	need	VERB
bracis-33663	235	25	is	be	AUX
bracis-33663	235	26	“	"	PUNCT
bracis-33663	235	27	love	love	NOUN
bracis-33663	235	28	”	"	PUNCT
bracis-33663	235	29	evading	evade	VERB
bracis-33663	235	30	hate	hate	NOUN
bracis-33663	235	31	speech	speech	NOUN
bracis-33663	235	32	detection	detection	NOUN
bracis-33663	235	33	.	.	PUNCT
bracis-33663	236	1	in	in	ADP
bracis-33663	236	2	:	:	PUNCT
bracis-33663	236	3	proceedings	proceeding	NOUN
bracis-33663	236	4	of	of	ADP
bracis-33663	236	5	the	the	DET
bracis-33663	236	6	11th	11th	ADJ
bracis-33663	236	7	acm	acm	NOUN
bracis-33663	236	8	workshop	workshop	NOUN
bracis-33663	236	9	on	on	ADP
bracis-33663	236	10	artificial	artificial	ADJ
bracis-33663	236	11	intelligence	intelligence	NOUN
bracis-33663	236	12	and	and	CCONJ
bracis-33663	236	13	security	security	NOUN
bracis-33663	236	14	,	,	PUNCT
bracis-33663	236	15	pp	pp	ADP
bracis-33663	236	16	.	.	PUNCT
bracis-33663	237	1	2–12	2–12	INTJ
bracis-33663	237	2	(	(	PUNCT
bracis-33663	237	3	2018	2018	NUM
bracis-33663	237	4	)	)	PUNCT
bracis-33663	237	5	google	google	PROPN
bracis-33663	237	6	scholar	scholar	NOUN
bracis-33663	237	7	  	  	SPACE
bracis-33663	237	8	gruzd	gruzd	NOUN
bracis-33663	237	9	,	,	PUNCT
bracis-33663	237	10	a.	a.	PROPN
bracis-33663	237	11	,	,	PUNCT
bracis-33663	237	12	soares	soares	PROPN
bracis-33663	237	13	,	,	PUNCT
bracis-33663	237	14	f.b	f.b	PROPN
bracis-33663	237	15	.	.	PROPN
bracis-33663	237	16	,	,	PUNCT
bracis-33663	237	17	mai	mai	PROPN
bracis-33663	237	18	,	,	PUNCT
bracis-33663	237	19	p.	p.	NOUN
bracis-33663	237	20	:	:	PUNCT
bracis-33663	238	1	trust	trust	NOUN
bracis-33663	238	2	and	and	CCONJ
bracis-33663	238	3	safety	safety	NOUN
bracis-33663	238	4	on	on	ADP
bracis-33663	238	5	social	social	ADJ
bracis-33663	238	6	media	medium	NOUN
bracis-33663	238	7	:	:	PUNCT
bracis-33663	238	8	understanding	understand	VERB
bracis-33663	238	9	the	the	DET
bracis-33663	238	10	impact	impact	NOUN
bracis-33663	238	11	of	of	ADP
bracis-33663	238	12	anti	anti	ADJ
bracis-33663	238	13	-	-	ADJ
bracis-33663	238	14	social	social	ADJ
bracis-33663	238	15	behavior	behavior	NOUN
bracis-33663	238	16	and	and	CCONJ
bracis-33663	238	17	misinformation	misinformation	NOUN
bracis-33663	238	18	on	on	ADP
bracis-33663	238	19	content	content	NOUN
bracis-33663	238	20	moderation	moderation	NOUN
bracis-33663	238	21	and	and	CCONJ
bracis-33663	238	22	platform	platform	NOUN
bracis-33663	238	23	governance	governance	NOUN
bracis-33663	238	24	.	.	PUNCT
bracis-33663	239	1	soc	soc	NOUN
bracis-33663	239	2	.	.	PUNCT
bracis-33663	240	1	media+	media+	NOUN
bracis-33663	240	2	soc	soc	PROPN
bracis-33663	240	3	.	.	PUNCT
bracis-33663	241	1	9(3	9(3	NUM
bracis-33663	241	2	)	)	PUNCT
bracis-33663	241	3	,	,	PUNCT
bracis-33663	241	4	20563051231196878	20563051231196878	NUM
bracis-33663	241	5	(	(	PUNCT
bracis-33663	241	6	2023	2023	NUM
bracis-33663	241	7	)	)	PUNCT
bracis-33663	241	8	google	google	NOUN
bracis-33663	241	9	scholar	scholar	NOUN
bracis-33663	241	10	  	  	SPACE
bracis-33663	241	11	liu	liu	PROPN
bracis-33663	241	12	,	,	PUNCT
bracis-33663	241	13	x.	x.	PROPN
bracis-33663	241	14	,	,	PUNCT
bracis-33663	241	15	et	et	PROPN
bracis-33663	241	16	al	al	PROPN
bracis-33663	241	17	.	.	PUNCT
bracis-33663	241	18	:	:	PUNCT
bracis-33663	242	1	p	p	X
bracis-33663	242	2	-	-	PUNCT
bracis-33663	242	3	tuning	tune	VERB
bracis-33663	242	4	v2	v2	NOUN
bracis-33663	242	5	:	:	PUNCT
bracis-33663	242	6	prompt	prompt	ADJ
bracis-33663	242	7	tuning	tuning	NOUN
bracis-33663	242	8	can	can	AUX
bracis-33663	242	9	be	be	AUX
bracis-33663	242	10	comparable	comparable	ADJ
bracis-33663	242	11	to	to	ADP
bracis-33663	242	12	fine	fine	ADV
bracis-33663	242	13	-	-	PUNCT
bracis-33663	242	14	tuning	tuning	NOUN
bracis-33663	242	15	universally	universally	ADV
bracis-33663	242	16	across	across	ADP
bracis-33663	242	17	scales	scale	NOUN
bracis-33663	242	18	and	and	CCONJ
bracis-33663	242	19	tasks	task	NOUN
bracis-33663	242	20	(	(	PUNCT
bracis-33663	242	21	2022	2022	NUM
bracis-33663	242	22	)	)	PUNCT
bracis-33663	242	23	google	google	PROPN
bracis-33663	242	24	scholar	scholar	NOUN
bracis-33663	242	25	  	  	SPACE
bracis-33663	242	26	liu	liu	PROPN
bracis-33663	242	27	,	,	PUNCT
bracis-33663	242	28	y.	y.	PROPN
bracis-33663	242	29	,	,	PUNCT
bracis-33663	242	30	et	et	PROPN
bracis-33663	242	31	al	al	PROPN
bracis-33663	242	32	.	.	PUNCT
bracis-33663	242	33	:	:	PUNCT
bracis-33663	243	1	roberta	roberta	NOUN
bracis-33663	243	2	:	:	PUNCT
bracis-33663	243	3	a	a	DET
bracis-33663	243	4	robustly	robustly	ADV
bracis-33663	243	5	optimized	optimize	VERB
bracis-33663	243	6	bert	bert	NOUN
bracis-33663	243	7	pretraining	pretraine	VERB
bracis-33663	243	8	approach	approach	NOUN
bracis-33663	243	9	(	(	PUNCT
bracis-33663	243	10	2019	2019	NUM
bracis-33663	243	11	)	)	PUNCT
bracis-33663	243	12	google	google	PROPN
bracis-33663	243	13	scholar	scholar	NOUN
bracis-33663	243	14	  	  	SPACE
bracis-33663	243	15	luo	luo	PROPN
bracis-33663	243	16	,	,	PUNCT
bracis-33663	243	17	h.	h.	PROPN
bracis-33663	243	18	,	,	PUNCT
bracis-33663	243	19	glass	glass	NOUN
bracis-33663	243	20	,	,	PUNCT
bracis-33663	243	21	j.	j.	PROPN
bracis-33663	243	22	:	:	PUNCT
bracis-33663	243	23	logic	logic	NOUN
bracis-33663	243	24	against	against	ADP
bracis-33663	243	25	bias	bias	NOUN
bracis-33663	243	26	:	:	PUNCT
bracis-33663	243	27	textual	textual	ADJ
bracis-33663	243	28	entailment	entailment	NOUN
bracis-33663	243	29	mitigates	mitigate	VERB
bracis-33663	243	30	stereotypical	stereotypical	ADJ
bracis-33663	243	31	sentence	sentence	NOUN
bracis-33663	243	32	reasoning	reasoning	NOUN
bracis-33663	243	33	.	.	PUNCT
bracis-33663	244	1	in	in	ADP
bracis-33663	244	2	:	:	PUNCT
bracis-33663	244	3	proceedings	proceeding	NOUN
bracis-33663	244	4	of	of	ADP
bracis-33663	244	5	the	the	DET
bracis-33663	244	6	17th	17th	ADJ
bracis-33663	244	7	conference	conference	NOUN
bracis-33663	244	8	of	of	ADP
bracis-33663	244	9	the	the	DET
bracis-33663	244	10	european	european	ADJ
bracis-33663	244	11	chapter	chapter	NOUN
bracis-33663	244	12	of	of	ADP
bracis-33663	244	13	the	the	DET
bracis-33663	244	14	association	association	NOUN
bracis-33663	244	15	for	for	ADP
bracis-33663	244	16	computational	computational	ADJ
bracis-33663	244	17	linguistics	linguistic	NOUN
bracis-33663	244	18	,	,	PUNCT
bracis-33663	244	19	pp	pp	X
bracis-33663	244	20	.	.	PUNCT
bracis-33663	245	1	1235–1246	1235–1246	NUM
bracis-33663	245	2	(	(	PUNCT
bracis-33663	245	3	2023	2023	NUM
bracis-33663	245	4	)	)	PUNCT
bracis-33663	245	5	google	google	NOUN
bracis-33663	245	6	scholar	scholar	NOUN
bracis-33663	245	7	  	  	SPACE
bracis-33663	245	8	macavaney	macavaney	NOUN
bracis-33663	245	9	,	,	PUNCT
bracis-33663	245	10	s.	s.	PROPN
bracis-33663	245	11	,	,	PUNCT
bracis-33663	245	12	yao	yao	PROPN
bracis-33663	245	13	,	,	PUNCT
bracis-33663	245	14	h.r	h.r	PROPN
bracis-33663	245	15	.	.	PROPN
bracis-33663	245	16	,	,	PUNCT
bracis-33663	245	17	yang	yang	PROPN
bracis-33663	245	18	,	,	PUNCT
bracis-33663	245	19	e.	e.	PROPN
bracis-33663	245	20	,	,	PUNCT
bracis-33663	245	21	russell	russell	PROPN
bracis-33663	245	22	,	,	PUNCT
bracis-33663	245	23	k.	k.	PROPN
bracis-33663	245	24	,	,	PUNCT
bracis-33663	245	25	goharian	goharian	PROPN
bracis-33663	245	26	,	,	PUNCT
bracis-33663	245	27	n.	n.	NOUN
bracis-33663	245	28	,	,	PUNCT
bracis-33663	245	29	frieder	frieder	NOUN
bracis-33663	245	30	,	,	PUNCT
bracis-33663	245	31	o.	o.	NOUN
bracis-33663	245	32	:	:	PUNCT
bracis-33663	245	33	hate	hate	VERB
bracis-33663	245	34	speech	speech	NOUN
bracis-33663	245	35	detection	detection	NOUN
bracis-33663	245	36	:	:	PUNCT
bracis-33663	245	37	challenges	challenge	NOUN
bracis-33663	245	38	and	and	CCONJ
bracis-33663	245	39	solutions	solution	NOUN
bracis-33663	245	40	.	.	PUNCT
bracis-33663	246	1	plos	plos	PROPN
bracis-33663	246	2	one	one	NUM
bracis-33663	246	3	14(8	14(8	NUM
bracis-33663	246	4	)	)	PUNCT
bracis-33663	246	5	,	,	PUNCT
bracis-33663	246	6	e0221152	e0221152	PROPN
bracis-33663	246	7	(	(	PUNCT
bracis-33663	246	8	2019	2019	NUM
bracis-33663	246	9	)	)	PUNCT
bracis-33663	246	10	article	article	NOUN
bracis-33663	246	11	  	  	SPACE
bracis-33663	246	12	google	google	PROPN
bracis-33663	246	13	scholar	scholar	NOUN
bracis-33663	246	14	  	  	SPACE
bracis-33663	246	15	nadeem	nadeem	PROPN
bracis-33663	246	16	,	,	PUNCT
bracis-33663	246	17	m.	m.	NOUN
bracis-33663	246	18	,	,	PUNCT
bracis-33663	246	19	bethke	bethke	PROPN
bracis-33663	246	20	,	,	PUNCT
bracis-33663	246	21	a.	a.	NOUN
bracis-33663	246	22	,	,	PUNCT
bracis-33663	246	23	reddy	reddy	PROPN
bracis-33663	246	24	,	,	PUNCT
bracis-33663	246	25	s.	s.	PROPN
bracis-33663	246	26	:	:	PUNCT
bracis-33663	246	27	stereoset	stereoset	PROPN
bracis-33663	246	28	:	:	PUNCT
bracis-33663	246	29	measuring	measure	VERB
bracis-33663	246	30	stereotypical	stereotypical	ADJ
bracis-33663	246	31	bias	bias	NOUN
bracis-33663	246	32	in	in	ADP
bracis-33663	246	33	pretrained	pretraine	VERB
bracis-33663	246	34	language	language	NOUN
bracis-33663	246	35	models	model	NOUN
bracis-33663	246	36	.	.	PUNCT
bracis-33663	247	1	in	in	ADP
bracis-33663	247	2	:	:	PUNCT
bracis-33663	247	3	zong	zong	PROPN
bracis-33663	247	4	,	,	PUNCT
bracis-33663	247	5	c.	c.	PROPN
bracis-33663	247	6	,	,	PUNCT
bracis-33663	247	7	xia	xia	PROPN
bracis-33663	247	8	,	,	PUNCT
bracis-33663	247	9	f.	f.	PROPN
bracis-33663	247	10	,	,	PUNCT
bracis-33663	247	11	li	li	PROPN
bracis-33663	247	12	,	,	PUNCT
bracis-33663	247	13	w.	w.	PROPN
bracis-33663	247	14	,	,	PUNCT
bracis-33663	247	15	navigli	navigli	PROPN
bracis-33663	247	16	,	,	PUNCT
bracis-33663	247	17	r.	r.	PROPN
bracis-33663	247	18	(	(	PUNCT
bracis-33663	247	19	eds	eds	PROPN
bracis-33663	247	20	.	.	PUNCT
bracis-33663	247	21	)	)	PUNCT
bracis-33663	247	22	proceedings	proceeding	NOUN
bracis-33663	247	23	of	of	ADP
bracis-33663	247	24	the	the	DET
bracis-33663	247	25	59th	59th	ADJ
bracis-33663	247	26	annual	annual	ADJ
bracis-33663	247	27	meeting	meeting	NOUN
bracis-33663	247	28	of	of	ADP
bracis-33663	247	29	the	the	DET
bracis-33663	247	30	association	association	NOUN
bracis-33663	247	31	for	for	ADP
bracis-33663	247	32	computational	computational	ADJ
bracis-33663	247	33	linguistics	linguistic	NOUN
bracis-33663	247	34	and	and	CCONJ
bracis-33663	247	35	the	the	DET
bracis-33663	247	36	11th	11th	ADJ
bracis-33663	247	37	international	international	ADJ
bracis-33663	247	38	joint	joint	ADJ
bracis-33663	247	39	conference	conference	NOUN
bracis-33663	247	40	on	on	ADP
bracis-33663	247	41	natural	natural	ADJ
bracis-33663	247	42	language	language	NOUN
bracis-33663	247	43	processing	processing	NOUN
bracis-33663	247	44	(	(	PUNCT
bracis-33663	247	45	volume	volume	NOUN
bracis-33663	247	46	1	1	NUM
bracis-33663	247	47	:	:	PUNCT
bracis-33663	247	48	long	long	ADJ
bracis-33663	247	49	papers	paper	NOUN
bracis-33663	247	50	)	)	PUNCT
bracis-33663	247	51	,	,	PUNCT
bracis-33663	247	52	pp	pp	ADP
bracis-33663	247	53	.	.	PUNCT
bracis-33663	247	54	5356–5371	5356–5371	NUM
bracis-33663	247	55	.	.	PUNCT
bracis-33663	248	1	association	association	NOUN
bracis-33663	248	2	for	for	ADP
bracis-33663	248	3	computational	computational	ADJ
bracis-33663	248	4	linguistics	linguistic	NOUN
bracis-33663	248	5	,	,	PUNCT
bracis-33663	248	6	online	online	ADV
bracis-33663	248	7	(	(	PUNCT
bracis-33663	248	8	2021	2021	NUM
bracis-33663	248	9	)	)	PUNCT
bracis-33663	248	10	.	.	PUNCT
bracis-33663	249	1	https://doi.org/10.18653/v1/2021.acl-long.416	https://doi.org/10.18653/v1/2021.acl-long.416	PROPN
bracis-33663	249	2	.	.	PUNCT
bracis-33663	250	1	https://aclanthology.org/2021.acl-long.416	https://aclanthology.org/2021.acl-long.416	PROPN
bracis-33663	250	2	paz	paz	PROPN
bracis-33663	250	3	,	,	PUNCT
bracis-33663	250	4	m.a	m.a	PROPN
bracis-33663	250	5	.	.	PROPN
bracis-33663	250	6	,	,	PUNCT
bracis-33663	250	7	montero	montero	NOUN
bracis-33663	250	8	-	-	PUNCT
bracis-33663	250	9	díaz	díaz	NOUN
bracis-33663	250	10	,	,	PUNCT
bracis-33663	250	11	j.	j.	PROPN
bracis-33663	250	12	,	,	PUNCT
bracis-33663	250	13	moreno	moreno	PROPN
bracis-33663	250	14	-	-	PUNCT
bracis-33663	250	15	delgado	delgado	PROPN
bracis-33663	250	16	,	,	PUNCT
bracis-33663	250	17	a.	a.	NOUN
bracis-33663	250	18	:	:	PUNCT
bracis-33663	250	19	hate	hate	VERB
bracis-33663	250	20	speech	speech	NOUN
bracis-33663	250	21	:	:	PUNCT
bracis-33663	250	22	a	a	DET
bracis-33663	250	23	systematized	systematized	ADJ
bracis-33663	250	24	review	review	NOUN
bracis-33663	250	25	.	.	PUNCT
bracis-33663	251	1	sage	sage	VERB
bracis-33663	251	2	open	open	ADJ
bracis-33663	251	3	10(4	10(4	NUM
bracis-33663	251	4	)	)	PUNCT
bracis-33663	251	5	,	,	PUNCT
bracis-33663	251	6	2158244020973022	2158244020973022	NUM
bracis-33663	251	7	(	(	PUNCT
bracis-33663	251	8	2020	2020	NUM
bracis-33663	251	9	)	)	PUNCT
bracis-33663	251	10	.	.	PUNCT
bracis-33663	252	1	https://doi.org/10.1177/2158244020973022	https://doi.org/10.1177/2158244020973022	PROPN
bracis-33663	252	2	article	article	NOUN
bracis-33663	252	3	  	  	SPACE
bracis-33663	252	4	google	google	PROPN
bracis-33663	252	5	scholar	scholar	NOUN
bracis-33663	252	6	  	  	SPACE
bracis-33663	252	7	röttger	röttger	NOUN
bracis-33663	252	8	,	,	PUNCT
bracis-33663	252	9	p.	p.	PROPN
bracis-33663	252	10	,	,	PUNCT
bracis-33663	252	11	vidgen	vidgen	PROPN
bracis-33663	252	12	,	,	PUNCT
bracis-33663	252	13	b.	b.	PROPN
bracis-33663	252	14	,	,	PUNCT
bracis-33663	252	15	nguyen	nguyen	PROPN
bracis-33663	252	16	,	,	PUNCT
bracis-33663	252	17	d.	d.	PROPN
bracis-33663	252	18	,	,	PUNCT
bracis-33663	252	19	waseem	waseem	PROPN
bracis-33663	252	20	,	,	PUNCT
bracis-33663	252	21	z.	z.	PROPN
bracis-33663	252	22	,	,	PUNCT
bracis-33663	252	23	margetts	margetts	PROPN
bracis-33663	252	24	,	,	PUNCT
bracis-33663	252	25	h.	h.	PROPN
bracis-33663	252	26	,	,	PUNCT
bracis-33663	252	27	pierrehumbert	pierrehumbert	PROPN
bracis-33663	252	28	,	,	PUNCT
bracis-33663	252	29	j.	j.	PROPN
bracis-33663	252	30	:	:	PUNCT
bracis-33663	252	31	hatecheck	hatecheck	ADJ
bracis-33663	252	32	:	:	PUNCT
bracis-33663	252	33	functional	functional	ADJ
bracis-33663	252	34	tests	test	NOUN
bracis-33663	252	35	for	for	ADP
bracis-33663	252	36	hate	hate	NOUN
bracis-33663	252	37	speech	speech	NOUN
bracis-33663	252	38	detection	detection	NOUN
bracis-33663	252	39	models	model	NOUN
bracis-33663	252	40	.	.	PUNCT
bracis-33663	253	1	in	in	ADP
bracis-33663	253	2	:	:	PUNCT
bracis-33663	253	3	proceedings	proceeding	NOUN
bracis-33663	253	4	of	of	ADP
bracis-33663	253	5	the	the	DET
bracis-33663	253	6	59th	59th	ADJ
bracis-33663	253	7	annual	annual	ADJ
bracis-33663	253	8	meeting	meeting	NOUN
bracis-33663	253	9	of	of	ADP
bracis-33663	253	10	the	the	DET
bracis-33663	253	11	association	association	NOUN
bracis-33663	253	12	for	for	ADP
bracis-33663	253	13	computational	computational	ADJ
bracis-33663	253	14	linguistics	linguistic	NOUN
bracis-33663	253	15	and	and	CCONJ
bracis-33663	253	16	the	the	DET
bracis-33663	253	17	11th	11th	ADJ
bracis-33663	253	18	international	international	ADJ
bracis-33663	253	19	joint	joint	ADJ
bracis-33663	253	20	conference	conference	NOUN
bracis-33663	253	21	on	on	ADP
bracis-33663	253	22	natural	natural	ADJ
bracis-33663	253	23	language	language	NOUN
bracis-33663	253	24	processing	processing	NOUN
bracis-33663	253	25	(	(	PUNCT
bracis-33663	253	26	volume	volume	NOUN
bracis-33663	253	27	1	1	NUM
bracis-33663	253	28	:	:	PUNCT
bracis-33663	253	29	long	long	ADJ
bracis-33663	253	30	papers	paper	NOUN
bracis-33663	253	31	)	)	PUNCT
bracis-33663	253	32	,	,	PUNCT
bracis-33663	253	33	pp	pp	ADV
bracis-33663	253	34	.	.	PUNCT
bracis-33663	254	1	41–58	41–58	PROPN
bracis-33663	254	2	(	(	PUNCT
bracis-33663	254	3	2021	2021	NUM
bracis-33663	254	4	)	)	PUNCT
bracis-33663	254	5	google	google	NOUN
bracis-33663	254	6	scholar	scholar	NOUN
bracis-33663	254	7	  	  	SPACE
bracis-33663	254	8	sevani	sevani	NOUN
bracis-33663	254	9	,	,	PUNCT
bracis-33663	254	10	n.	n.	NOUN
bracis-33663	254	11	,	,	PUNCT
bracis-33663	254	12	soenandi	soenandi	PROPN
bracis-33663	254	13	,	,	PUNCT
bracis-33663	254	14	i.a	i.a	PROPN
bracis-33663	254	15	.	.	PROPN
bracis-33663	254	16	,	,	PUNCT
bracis-33663	254	17	wijaya	wijaya	PROPN
bracis-33663	254	18	,	,	PUNCT
bracis-33663	254	19	j.	j.	PROPN
bracis-33663	254	20	,	,	PUNCT
bracis-33663	254	21	et	et	PROPN
bracis-33663	254	22	 	 	SPACE
bracis-33663	254	23	al	al	PROPN
bracis-33663	254	24	.	.	PUNCT
bracis-33663	254	25	:	:	PUNCT
bracis-33663	255	1	detection	detection	NOUN
bracis-33663	255	2	of	of	ADP
bracis-33663	255	3	hate	hate	NOUN
bracis-33663	255	4	speech	speech	NOUN
bracis-33663	255	5	by	by	ADP
bracis-33663	255	6	employing	employ	VERB
bracis-33663	255	7	support	support	NOUN
bracis-33663	255	8	vector	vector	NOUN
bracis-33663	255	9	machine	machine	NOUN
bracis-33663	255	10	with	with	ADP
bracis-33663	255	11	word2vec	word2vec	PROPN
bracis-33663	255	12	model	model	NOUN
bracis-33663	255	13	.	.	PUNCT
bracis-33663	256	1	in	in	ADP
bracis-33663	256	2	:	:	PUNCT
bracis-33663	256	3	2021	2021	NUM
bracis-33663	256	4	7th	7th	ADJ
bracis-33663	256	5	international	international	ADJ
bracis-33663	256	6	conference	conference	NOUN
bracis-33663	256	7	on	on	ADP
bracis-33663	256	8	electrical	electrical	ADJ
bracis-33663	256	9	,	,	PUNCT
bracis-33663	256	10	electronics	electronic	NOUN
bracis-33663	256	11	and	and	CCONJ
bracis-33663	256	12	information	information	NOUN
bracis-33663	256	13	engineering	engineering	NOUN
bracis-33663	256	14	(	(	PUNCT
bracis-33663	256	15	iceeie	iceeie	NOUN
bracis-33663	256	16	)	)	PUNCT
bracis-33663	256	17	,	,	PUNCT
bracis-33663	256	18	pp	pp	PROPN
bracis-33663	256	19	.	.	PUNCT
bracis-33663	256	20	 	 	SPACE
bracis-33663	257	1	1–5	1–5	PROPN
bracis-33663	257	2	.	.	PUNCT
bracis-33663	257	3	ieee	ieee	NOUN
bracis-33663	257	4	(	(	PUNCT
bracis-33663	257	5	2021	2021	NUM
bracis-33663	257	6	)	)	PUNCT
bracis-33663	257	7	google	google	PROPN
bracis-33663	257	8	scholar	scholar	NOUN
bracis-33663	257	9	  	  	SPACE
bracis-33663	257	10	vidgen	vidgen	NOUN
bracis-33663	257	11	,	,	PUNCT
bracis-33663	257	12	b.	b.	PROPN
bracis-33663	257	13	,	,	PUNCT
bracis-33663	257	14	harris	harris	PROPN
bracis-33663	257	15	,	,	PUNCT
bracis-33663	257	16	a.	a.	PROPN
bracis-33663	257	17	,	,	PUNCT
bracis-33663	257	18	nguyen	nguyen	PROPN
bracis-33663	257	19	,	,	PUNCT
bracis-33663	257	20	d.	d.	PROPN
bracis-33663	257	21	,	,	PUNCT
bracis-33663	257	22	tromble	tromble	PROPN
bracis-33663	257	23	,	,	PUNCT
bracis-33663	257	24	r.	r.	PROPN
bracis-33663	257	25	,	,	PUNCT
bracis-33663	257	26	hale	hale	PROPN
bracis-33663	257	27	,	,	PUNCT
bracis-33663	257	28	s.	s.	PROPN
bracis-33663	257	29	,	,	PUNCT
bracis-33663	257	30	margetts	margetts	PROPN
bracis-33663	257	31	,	,	PUNCT
bracis-33663	257	32	h.	h.	NOUN
bracis-33663	257	33	:	:	PUNCT
bracis-33663	257	34	challenges	challenge	NOUN
bracis-33663	257	35	and	and	CCONJ
bracis-33663	257	36	frontiers	frontier	NOUN
bracis-33663	257	37	in	in	ADP
bracis-33663	257	38	abusive	abusive	ADJ
bracis-33663	257	39	content	content	NOUN
bracis-33663	257	40	detection	detection	NOUN
bracis-33663	257	41	.	.	PUNCT
bracis-33663	258	1	in	in	ADP
bracis-33663	258	2	:	:	PUNCT
bracis-33663	258	3	proceedings	proceeding	NOUN
bracis-33663	258	4	of	of	ADP
bracis-33663	258	5	the	the	DET
bracis-33663	258	6	third	third	ADJ
bracis-33663	258	7	workshop	workshop	NOUN
bracis-33663	258	8	on	on	ADP
bracis-33663	258	9	abusive	abusive	ADJ
bracis-33663	258	10	language	language	NOUN
bracis-33663	258	11	online	online	NOUN
bracis-33663	258	12	.	.	PUNCT
bracis-33663	259	1	association	association	NOUN
bracis-33663	259	2	for	for	ADP
bracis-33663	259	3	computational	computational	ADJ
bracis-33663	259	4	linguistics	linguistic	NOUN
bracis-33663	259	5	(	(	PUNCT
bracis-33663	259	6	2019	2019	NUM
bracis-33663	259	7	)	)	PUNCT
bracis-33663	259	8	google	google	PROPN
bracis-33663	259	9	scholar	scholar	NOUN
bracis-33663	259	10	  	  	SPACE
bracis-33663	259	11	vidgen	vidgen	PROPN
bracis-33663	259	12	,	,	PUNCT
bracis-33663	259	13	b.	b.	PROPN
bracis-33663	259	14	,	,	PUNCT
bracis-33663	259	15	thrush	thrush	NOUN
bracis-33663	259	16	,	,	PUNCT
bracis-33663	259	17	t.	t.	PROPN
bracis-33663	259	18	,	,	PUNCT
bracis-33663	259	19	waseem	waseem	PROPN
bracis-33663	259	20	,	,	PUNCT
bracis-33663	259	21	z.	z.	PROPN
bracis-33663	259	22	,	,	PUNCT
bracis-33663	259	23	kiela	kiela	PROPN
bracis-33663	259	24	,	,	PUNCT
bracis-33663	259	25	d.	d.	PROPN
bracis-33663	259	26	:	:	PUNCT
bracis-33663	260	1	learning	learn	VERB
bracis-33663	260	2	from	from	ADP
bracis-33663	260	3	the	the	DET
bracis-33663	260	4	worst	bad	ADJ
bracis-33663	260	5	:	:	PUNCT
bracis-33663	260	6	dynamically	dynamically	ADV
bracis-33663	260	7	generated	generate	VERB
bracis-33663	260	8	datasets	dataset	NOUN
bracis-33663	260	9	to	to	PART
bracis-33663	260	10	improve	improve	VERB
bracis-33663	260	11	online	online	ADJ
bracis-33663	260	12	hate	hate	NOUN
bracis-33663	260	13	detection	detection	NOUN
bracis-33663	260	14	.	.	PUNCT
bracis-33663	261	1	in	in	ADP
bracis-33663	261	2	:	:	PUNCT
bracis-33663	261	3	proceedings	proceeding	NOUN
bracis-33663	261	4	of	of	ADP
bracis-33663	261	5	the	the	DET
bracis-33663	261	6	59th	59th	ADJ
bracis-33663	261	7	annual	annual	ADJ
bracis-33663	261	8	meeting	meeting	NOUN
bracis-33663	261	9	of	of	ADP
bracis-33663	261	10	the	the	DET
bracis-33663	261	11	association	association	NOUN
bracis-33663	261	12	for	for	ADP
bracis-33663	261	13	computational	computational	ADJ
bracis-33663	261	14	linguistics	linguistic	NOUN
bracis-33663	261	15	and	and	CCONJ
bracis-33663	261	16	the	the	DET
bracis-33663	261	17	11th	11th	ADJ
bracis-33663	261	18	international	international	ADJ
bracis-33663	261	19	joint	joint	ADJ
bracis-33663	261	20	conference	conference	NOUN
bracis-33663	261	21	on	on	ADP
bracis-33663	261	22	natural	natural	ADJ
bracis-33663	261	23	language	language	NOUN
bracis-33663	261	24	processing	processing	NOUN
bracis-33663	261	25	.	.	PUNCT
bracis-33663	262	1	association	association	NOUN
bracis-33663	262	2	for	for	ADP
bracis-33663	262	3	computational	computational	ADJ
bracis-33663	262	4	linguistics	linguistic	NOUN
bracis-33663	262	5	(	(	PUNCT
bracis-33663	262	6	2021	2021	NUM
bracis-33663	262	7	)	)	PUNCT
bracis-33663	262	8	.	.	PUNCT
bracis-33663	263	1	https://doi.org/10.18653/v1/2021.acl-long.132	https://doi.org/10.18653/v1/2021.acl-long.132	PROPN
bracis-33663	263	2	.	.	PUNCT
bracis-33663	264	1	https://aclanthology.org/2021.acl-long.132	https://aclanthology.org/2021.acl-long.132	AUX
bracis-33663	264	2	wang	wang	PROPN
bracis-33663	264	3	,	,	PUNCT
bracis-33663	264	4	s.	s.	PROPN
bracis-33663	264	5	,	,	PUNCT
bracis-33663	264	6	fang	fang	PROPN
bracis-33663	264	7	,	,	PUNCT
bracis-33663	264	8	h.	h.	PROPN
bracis-33663	264	9	,	,	PUNCT
bracis-33663	264	10	khabsa	khabsa	NOUN
bracis-33663	264	11	,	,	PUNCT
bracis-33663	264	12	m.	m.	NOUN
bracis-33663	264	13	,	,	PUNCT
bracis-33663	264	14	mao	mao	PROPN
bracis-33663	264	15	,	,	PUNCT
bracis-33663	264	16	h.	h.	PROPN
bracis-33663	264	17	,	,	PUNCT
bracis-33663	264	18	ma	ma	PROPN
bracis-33663	264	19	,	,	PUNCT
bracis-33663	264	20	h.	h.	PROPN
bracis-33663	264	21	:	:	PUNCT
bracis-33663	264	22	entailment	entailment	VERB
bracis-33663	264	23	as	as	ADP
bracis-33663	264	24	few	few	ADJ
bracis-33663	264	25	-	-	PUNCT
bracis-33663	264	26	shot	shot	NOUN
bracis-33663	264	27	learner	learner	NOUN
bracis-33663	264	28	(	(	PUNCT
bracis-33663	264	29	2021	2021	NUM
bracis-33663	264	30	)	)	PUNCT
bracis-33663	264	31	google	google	NOUN
bracis-33663	264	32	scholar	scholar	NOUN
bracis-33663	264	33	  	  	SPACE
bracis-33663	264	34	waseem	waseem	PROPN
bracis-33663	264	35	,	,	PUNCT
bracis-33663	264	36	z.	z.	PROPN
bracis-33663	264	37	,	,	PUNCT
bracis-33663	264	38	hovy	hovy	PROPN
bracis-33663	264	39	,	,	PUNCT
bracis-33663	264	40	d.	d.	PROPN
bracis-33663	264	41	:	:	PUNCT
bracis-33663	264	42	hateful	hateful	ADJ
bracis-33663	264	43	symbols	symbol	NOUN
bracis-33663	264	44	or	or	CCONJ
bracis-33663	264	45	hateful	hateful	ADJ
bracis-33663	264	46	people	people	NOUN
bracis-33663	264	47	?	?	PUNCT
bracis-33663	265	1	predictive	predictive	ADJ
bracis-33663	265	2	features	feature	NOUN
bracis-33663	265	3	for	for	ADP
bracis-33663	265	4	hate	hate	NOUN
bracis-33663	265	5	speech	speech	NOUN
bracis-33663	265	6	detection	detection	NOUN
bracis-33663	265	7	on	on	ADP
bracis-33663	265	8	twitter	twitter	NOUN
bracis-33663	265	9	.	.	PUNCT
bracis-33663	266	1	in	in	ADP
bracis-33663	266	2	:	:	PUNCT
bracis-33663	266	3	proceedings	proceeding	NOUN
bracis-33663	266	4	of	of	ADP
bracis-33663	266	5	the	the	DET
bracis-33663	266	6	naacl	naacl	PROPN
bracis-33663	266	7	student	student	NOUN
bracis-33663	266	8	research	research	NOUN
bracis-33663	266	9	workshop	workshop	NOUN
bracis-33663	266	10	,	,	PUNCT
bracis-33663	266	11	pp	pp	X
bracis-33663	266	12	.	.	PUNCT
bracis-33663	266	13	88–93	88–93	NUM
bracis-33663	266	14	.	.	PUNCT
bracis-33663	267	1	association	association	NOUN
bracis-33663	267	2	for	for	ADP
bracis-33663	267	3	computational	computational	ADJ
bracis-33663	267	4	linguistics	linguistic	NOUN
bracis-33663	267	5	,	,	PUNCT
bracis-33663	267	6	san	san	PROPN
bracis-33663	267	7	diego	diego	PROPN
bracis-33663	267	8	,	,	PUNCT
bracis-33663	267	9	california	california	PROPN
bracis-33663	267	10	(	(	PUNCT
bracis-33663	267	11	2016	2016	NUM
bracis-33663	267	12	)	)	PUNCT
bracis-33663	267	13	.	.	PUNCT
bracis-33663	268	1	https://doi.org/10.18653/v1/n16-2013	https://doi.org/10.18653/v1/n16-2013	PROPN
bracis-33663	268	2	.	.	PUNCT
bracis-33663	268	3	https://aclanthology.org/n16-2013	https://aclanthology.org/n16-2013	PROPN
bracis-33663	268	4	waseem	waseem	PROPN
bracis-33663	268	5	,	,	PUNCT
bracis-33663	268	6	z.	z.	PROPN
bracis-33663	268	7	,	,	PUNCT
bracis-33663	268	8	thorne	thorne	PROPN
bracis-33663	268	9	,	,	PUNCT
bracis-33663	268	10	j.	j.	PROPN
bracis-33663	268	11	,	,	PUNCT
bracis-33663	268	12	bingel	bingel	NOUN
bracis-33663	268	13	,	,	PUNCT
bracis-33663	268	14	j.	j.	PROPN
bracis-33663	268	15	:	:	PUNCT
bracis-33663	268	16	bridging	bridge	VERB
bracis-33663	268	17	the	the	DET
bracis-33663	268	18	gaps	gap	NOUN
bracis-33663	268	19	:	:	PUNCT
bracis-33663	268	20	multi	multi	ADJ
bracis-33663	268	21	task	task	NOUN
bracis-33663	268	22	learning	learn	VERB
bracis-33663	268	23	for	for	ADP
bracis-33663	268	24	domain	domain	NOUN
bracis-33663	268	25	transfer	transfer	NOUN
bracis-33663	268	26	of	of	ADP
bracis-33663	268	27	hate	hate	NOUN
bracis-33663	268	28	speech	speech	NOUN
bracis-33663	268	29	detection	detection	NOUN
bracis-33663	268	30	.	.	PUNCT
bracis-33663	269	1	online	online	ADJ
bracis-33663	269	2	harassment	harassment	NOUN
bracis-33663	269	3	,	,	PUNCT
bracis-33663	269	4	pp	pp	ADJ
bracis-33663	269	5	.	.	PUNCT
bracis-33663	270	1	29–55	29–55	NUM
bracis-33663	270	2	(	(	PUNCT
bracis-33663	270	3	2018	2018	NUM
bracis-33663	270	4	)	)	PUNCT
bracis-33663	270	5	google	google	PROPN
bracis-33663	270	6	scholar	scholar	NOUN
bracis-33663	270	7	  	  	SPACE
bracis-33663	270	8	wiegand	wiegand	NOUN
bracis-33663	270	9	,	,	PUNCT
bracis-33663	270	10	m.	m.	NOUN
bracis-33663	270	11	,	,	PUNCT
bracis-33663	270	12	ruppenhofer	ruppenhofer	NOUN
bracis-33663	270	13	,	,	PUNCT
bracis-33663	270	14	j.	j.	PROPN
bracis-33663	270	15	,	,	PUNCT
bracis-33663	270	16	kleinbauer	kleinbauer	PROPN
bracis-33663	270	17	,	,	PUNCT
bracis-33663	270	18	t.	t.	PROPN
bracis-33663	270	19	:	:	PUNCT
bracis-33663	270	20	detection	detection	NOUN
bracis-33663	270	21	of	of	ADP
bracis-33663	270	22	abusive	abusive	ADJ
bracis-33663	270	23	language	language	NOUN
bracis-33663	270	24	:	:	PUNCT
bracis-33663	270	25	the	the	DET
bracis-33663	270	26	problem	problem	NOUN
bracis-33663	270	27	of	of	ADP
bracis-33663	270	28	biased	biased	ADJ
bracis-33663	270	29	datasets	dataset	NOUN
bracis-33663	270	30	.	.	PUNCT
bracis-33663	271	1	in	in	ADP
bracis-33663	271	2	:	:	PUNCT
bracis-33663	271	3	burstein	burstein	PROPN
bracis-33663	271	4	,	,	PUNCT
bracis-33663	271	5	j.	j.	PROPN
bracis-33663	271	6	,	,	PUNCT
bracis-33663	271	7	doran	doran	PROPN
bracis-33663	271	8	,	,	PUNCT
bracis-33663	271	9	c.	c.	PROPN
bracis-33663	271	10	,	,	PUNCT
bracis-33663	271	11	solorio	solorio	PROPN
bracis-33663	271	12	,	,	PUNCT
bracis-33663	271	13	t.	t.	PROPN
bracis-33663	271	14	(	(	PUNCT
bracis-33663	271	15	eds	eds	PROPN
bracis-33663	271	16	.	.	PUNCT
bracis-33663	271	17	)	)	PUNCT
bracis-33663	271	18	proceedings	proceeding	NOUN
bracis-33663	271	19	of	of	ADP
bracis-33663	271	20	the	the	DET
bracis-33663	271	21	2019	2019	NUM
bracis-33663	271	22	conference	conference	NOUN
bracis-33663	271	23	of	of	ADP
bracis-33663	271	24	the	the	DET
bracis-33663	271	25	north	north	ADJ
bracis-33663	271	26	american	american	ADJ
bracis-33663	271	27	chapter	chapter	NOUN
bracis-33663	271	28	of	of	ADP
bracis-33663	271	29	the	the	DET
bracis-33663	271	30	association	association	NOUN
bracis-33663	271	31	for	for	ADP
bracis-33663	271	32	computational	computational	ADJ
bracis-33663	271	33	linguistics	linguistic	NOUN
bracis-33663	271	34	:	:	PUNCT
bracis-33663	271	35	human	human	ADJ
bracis-33663	271	36	language	language	NOUN
bracis-33663	271	37	technologies	technology	NOUN
bracis-33663	271	38	,	,	PUNCT
bracis-33663	271	39	volume	volume	NOUN
bracis-33663	271	40	1	1	NUM
bracis-33663	271	41	(	(	PUNCT
bracis-33663	271	42	long	long	ADJ
bracis-33663	271	43	and	and	CCONJ
bracis-33663	271	44	short	short	ADJ
bracis-33663	271	45	papers	paper	NOUN
bracis-33663	271	46	)	)	PUNCT
bracis-33663	271	47	,	,	PUNCT
bracis-33663	271	48	pp	pp	ADP
bracis-33663	271	49	.	.	PUNCT
bracis-33663	272	1	602–608	602–608	NUM
bracis-33663	272	2	.	.	PUNCT
bracis-33663	273	1	association	association	NOUN
bracis-33663	273	2	for	for	ADP
bracis-33663	273	3	computational	computational	ADJ
bracis-33663	273	4	linguistics	linguistic	NOUN
bracis-33663	273	5	,	,	PUNCT
bracis-33663	273	6	minneapolis	minneapolis	PROPN
bracis-33663	273	7	,	,	PUNCT
bracis-33663	273	8	minnesota	minnesota	PROPN
bracis-33663	273	9	(	(	PUNCT
bracis-33663	273	10	2019	2019	NUM
bracis-33663	273	11	)	)	PUNCT
bracis-33663	273	12	.	.	PUNCT
bracis-33663	274	1	https://doi.org/10.18653/v1/n19-1060	https://doi.org/10.18653/v1/n19-1060	NOUN
bracis-33663	274	2	.	.	PUNCT
bracis-33663	275	1	https://aclanthology.org/n19-1060	https://aclanthology.org/n19-1060	PROPN
bracis-33663	275	2	yin	yin	PROPN
bracis-33663	275	3	,	,	PUNCT
bracis-33663	275	4	w.	w.	PROPN
bracis-33663	275	5	,	,	PUNCT
bracis-33663	275	6	zubiaga	zubiaga	PROPN
bracis-33663	275	7	,	,	PUNCT
bracis-33663	275	8	a.	a.	NOUN
bracis-33663	275	9	:	:	PUNCT
bracis-33663	275	10	towards	towards	ADP
bracis-33663	275	11	generalisable	generalisable	ADJ
bracis-33663	275	12	hate	hate	NOUN
bracis-33663	275	13	speech	speech	NOUN
bracis-33663	275	14	detection	detection	NOUN
bracis-33663	275	15	:	:	PUNCT
bracis-33663	275	16	a	a	DET
bracis-33663	275	17	review	review	NOUN
bracis-33663	275	18	on	on	ADP
bracis-33663	275	19	obstacles	obstacle	NOUN
bracis-33663	275	20	and	and	CCONJ
bracis-33663	275	21	solutions	solution	NOUN
bracis-33663	275	22	.	.	PUNCT
bracis-33663	276	1	peerj	peerj	PROPN
bracis-33663	276	2	comput	comput	NOUN
bracis-33663	276	3	.	.	PUNCT
bracis-33663	277	1	sci	sci	PROPN
bracis-33663	277	2	.	.	PROPN
bracis-33663	278	1	7	7	NUM
bracis-33663	278	2	,	,	PUNCT
bracis-33663	278	3	e598	e598	PROPN
bracis-33663	278	4	(	(	PUNCT
bracis-33663	278	5	2021	2021	NUM
bracis-33663	278	6	)	)	PUNCT
bracis-33663	278	7	article	article	NOUN
bracis-33663	278	8	  	  	SPACE
bracis-33663	278	9	math	math	NOUN
bracis-33663	278	10	  	  	SPACE
bracis-33663	278	11	google	google	PROPN
bracis-33663	278	12	scholar	scholar	NOUN
bracis-33663	278	13	  	  	SPACE
bracis-33663	278	14	yin	yin	PROPN
bracis-33663	278	15	,	,	PUNCT
bracis-33663	278	16	w.	w.	PROPN
bracis-33663	278	17	,	,	PUNCT
bracis-33663	278	18	hay	hay	PROPN
bracis-33663	278	19	,	,	PUNCT
bracis-33663	278	20	j.	j.	PROPN
bracis-33663	278	21	,	,	PUNCT
bracis-33663	278	22	roth	roth	PROPN
bracis-33663	278	23	,	,	PUNCT
bracis-33663	278	24	d.	d.	NOUN
bracis-33663	278	25	:	:	PUNCT
bracis-33663	278	26	benchmarking	benchmarke	VERB
bracis-33663	278	27	zero	zero	NUM
bracis-33663	278	28	-	-	PUNCT
bracis-33663	278	29	shot	shot	NOUN
bracis-33663	278	30	text	text	NOUN
bracis-33663	278	31	classification	classification	NOUN
bracis-33663	278	32	:	:	PUNCT
bracis-33663	279	1	datasets	dataset	NOUN
bracis-33663	279	2	,	,	PUNCT
bracis-33663	279	3	evaluation	evaluation	NOUN
bracis-33663	279	4	and	and	CCONJ
bracis-33663	279	5	entailment	entailment	ADJ
bracis-33663	279	6	approach	approach	NOUN
bracis-33663	279	7	.	.	PUNCT
bracis-33663	280	1	in	in	ADP
bracis-33663	280	2	:	:	PUNCT
bracis-33663	280	3	proceedings	proceeding	NOUN
bracis-33663	280	4	of	of	ADP
bracis-33663	280	5	the	the	DET
bracis-33663	280	6	2019	2019	NUM
bracis-33663	280	7	conference	conference	NOUN
bracis-33663	280	8	on	on	ADP
bracis-33663	280	9	empirical	empirical	ADJ
bracis-33663	280	10	methods	method	NOUN
bracis-33663	280	11	in	in	ADP
bracis-33663	280	12	natural	natural	ADJ
bracis-33663	280	13	language	language	NOUN
bracis-33663	280	14	processing	processing	NOUN
bracis-33663	280	15	and	and	CCONJ
bracis-33663	280	16	the	the	DET
bracis-33663	280	17	9th	9th	ADJ
bracis-33663	280	18	international	international	ADJ
bracis-33663	280	19	joint	joint	ADJ
bracis-33663	280	20	conference	conference	NOUN
bracis-33663	280	21	on	on	ADP
bracis-33663	280	22	natural	natural	ADJ
bracis-33663	280	23	language	language	NOUN
bracis-33663	280	24	processing	processing	NOUN
bracis-33663	280	25	(	(	PUNCT
bracis-33663	280	26	emnlp	emnlp	NOUN
bracis-33663	280	27	-	-	PUNCT
bracis-33663	280	28	ijcnlp	ijcnlp	NOUN
bracis-33663	280	29	)	)	PUNCT
bracis-33663	280	30	.	.	PUNCT
bracis-33663	281	1	association	association	NOUN
bracis-33663	281	2	for	for	ADP
bracis-33663	281	3	computational	computational	ADJ
bracis-33663	281	4	linguistics	linguistic	NOUN
bracis-33663	281	5	(	(	PUNCT
bracis-33663	281	6	2019	2019	NUM
bracis-33663	281	7	)	)	PUNCT
bracis-33663	281	8	.	.	PUNCT
bracis-33663	282	1	https://doi.org/10.18653/v1/d19-1404	https://doi.org/10.18653/v1/d19-1404	PROPN
bracis-33663	282	2	.	.	PUNCT
bracis-33663	283	1	https://aclanthology.org/d19-1404	https://aclanthology.org/d19-1404	PROPN
bracis-33663	283	2	download	download	NOUN
bracis-33663	283	3	references	reference	NOUN
bracis-33663	283	4	author	author	NOUN
bracis-33663	283	5	information	information	NOUN
bracis-33663	283	6	authors	author	NOUN
bracis-33663	283	7	and	and	CCONJ
bracis-33663	283	8	affiliations	affiliation	NOUN
bracis-33663	283	9	graduate	graduate	NOUN
bracis-33663	283	10	program	program	NOUN
bracis-33663	283	11	in	in	ADP
bracis-33663	283	12	informatics	informatics	PROPN
bracis-33663	283	13	,	,	PUNCT
bracis-33663	283	14	pontificia	pontificia	PROPN
bracis-33663	283	15	universidade	universidade	PROPN
bracis-33663	283	16	católica	católica	PROPN
bracis-33663	283	17	do	do	VERB
bracis-33663	283	18	paraná	paraná	PROPN
bracis-33663	283	19	,	,	PUNCT
bracis-33663	283	20	curitiba	curitiba	PROPN
bracis-33663	283	21	,	,	PUNCT
bracis-33663	283	22	brazil	brazil	PROPN
bracis-33663	283	23	lorenzo	lorenzo	PROPN
bracis-33663	283	24	puppi	puppi	PROPN
bracis-33663	283	25	vecchi	vecchi	PROPN
bracis-33663	283	26	 	 	SPACE
bracis-33663	283	27	&	&	CCONJ
bracis-33663	283	28	 	 	SPACE
bracis-33663	283	29	emerson	emerson	PROPN
bracis-33663	283	30	cabrera	cabrera	PROPN
bracis-33663	283	31	paraiso	paraiso	PROPN
bracis-33663	283	32	department	department	PROPN
bracis-33663	283	33	of	of	ADP
bracis-33663	283	34	engineering	engineering	NOUN
bracis-33663	283	35	and	and	CCONJ
bracis-33663	283	36	architecture	architecture	NOUN
bracis-33663	283	37	,	,	PUNCT
bracis-33663	283	38	university	university	NOUN
bracis-33663	283	39	of	of	ADP
bracis-33663	283	40	trieste	trieste	PROPN
bracis-33663	283	41	,	,	PUNCT
bracis-33663	283	42	trieste	trieste	PROPN
bracis-33663	283	43	,	,	PUNCT
bracis-33663	283	44	italy	italy	PROPN
bracis-33663	283	45	sylvio	sylvio	PROPN
bracis-33663	283	46	barbon	barbon	PROPN
bracis-33663	283	47	junior	junior	PROPN
bracis-33663	283	48	authors	author	NOUN
bracis-33663	283	49	lorenzo	lorenzo	VERB
bracis-33663	283	50	puppi	puppi	PROPN
bracis-33663	283	51	vecchiview	vecchiview	PROPN
bracis-33663	283	52	author	author	NOUN
bracis-33663	283	53	publications	publication	NOUN
bracis-33663	283	54	search	search	NOUN
bracis-33663	283	55	author	author	NOUN
bracis-33663	283	56	on	on	ADP
bracis-33663	283	57	:	:	PUNCT
bracis-33663	283	58	pubmed	pubmed	PROPN
bracis-33663	283	59	 	 	SPACE
bracis-33663	283	60	google	google	PROPN
bracis-33663	283	61	scholar	scholar	NOUN
bracis-33663	283	62	sylvio	sylvio	PROPN
bracis-33663	283	63	barbon	barbon	PROPN
bracis-33663	283	64	juniorview	juniorview	PROPN
bracis-33663	283	65	author	author	NOUN
bracis-33663	283	66	publications	publication	NOUN
bracis-33663	283	67	search	search	NOUN
bracis-33663	283	68	author	author	NOUN
bracis-33663	283	69	on	on	ADP
bracis-33663	283	70	:	:	PUNCT
bracis-33663	283	71	pubmed	pubmed	PROPN
bracis-33663	283	72	 	 	SPACE
bracis-33663	283	73	google	google	PROPN
bracis-33663	283	74	scholar	scholar	PROPN
bracis-33663	283	75	emerson	emerson	PROPN
bracis-33663	283	76	cabrera	cabrera	PROPN
bracis-33663	283	77	paraisoview	paraisoview	PROPN
bracis-33663	283	78	author	author	NOUN
bracis-33663	283	79	publications	publication	NOUN
bracis-33663	283	80	search	search	NOUN
bracis-33663	283	81	author	author	NOUN
bracis-33663	283	82	on	on	ADP
bracis-33663	283	83	:	:	PUNCT
bracis-33663	283	84	pubmed	pubmed	PROPN
bracis-33663	283	85	 	 	SPACE
bracis-33663	283	86	google	google	PROPN
bracis-33663	283	87	scholar	scholar	NOUN
bracis-33663	283	88	corresponding	correspond	VERB
bracis-33663	283	89	author	author	NOUN
bracis-33663	283	90	correspondence	correspondence	NOUN
bracis-33663	283	91	to	to	ADP
bracis-33663	283	92	lorenzo	lorenzo	PROPN
bracis-33663	283	93	puppi	puppi	PROPN
bracis-33663	283	94	vecchi	vecchi	PROPN
bracis-33663	283	95	.	.	PUNCT
bracis-33663	284	1	editor	editor	NOUN
bracis-33663	284	2	information	information	NOUN
bracis-33663	284	3	editors	editor	NOUN
bracis-33663	284	4	and	and	CCONJ
bracis-33663	284	5	affiliations	affiliation	NOUN
bracis-33663	284	6	universidade	universidade	VERB
bracis-33663	284	7	federal	federal	PROPN
bracis-33663	284	8	fluminense	fluminense	PROPN
bracis-33663	284	9	,	,	PUNCT
bracis-33663	284	10	niterói	niterói	PROPN
bracis-33663	284	11	,	,	PUNCT
bracis-33663	284	12	brazil	brazil	PROPN
bracis-33663	284	13	aline	aline	PROPN
bracis-33663	284	14	paes	paes	PROPN
bracis-33663	284	15	instituto	instituto	PROPN
bracis-33663	284	16	tecnológico	tecnológico	PROPN
bracis-33663	284	17	de	de	PROPN
bracis-33663	284	18	aeronáutica	aeronáutica	PROPN
bracis-33663	284	19	,	,	PUNCT
bracis-33663	284	20	são	são	PROPN
bracis-33663	284	21	josé	josé	PROPN
bracis-33663	284	22	dos	dos	PROPN
bracis-33663	284	23	campos	campos	PROPN
bracis-33663	284	24	,	,	PUNCT
bracis-33663	284	25	brazil	brazil	PROPN
bracis-33663	284	26	filipe	filipe	PROPN
bracis-33663	284	27	a.	a.	PROPN
bracis-33663	284	28	n.	n.	PROPN
bracis-33663	284	29	verri	verri	PROPN
bracis-33663	284	30	ethics	ethic	NOUN
bracis-33663	284	31	declarations	declaration	VERB
bracis-33663	284	32	disclosure	disclosure	NOUN
bracis-33663	284	33	of	of	ADP
bracis-33663	284	34	interests	interest	NOUN
bracis-33663	284	35	the	the	DET
bracis-33663	284	36	authors	author	NOUN
bracis-33663	284	37	have	have	VERB
bracis-33663	284	38	no	no	DET
bracis-33663	284	39	competing	compete	VERB
bracis-33663	284	40	interests	interest	NOUN
bracis-33663	284	41	to	to	PART
bracis-33663	284	42	declare	declare	VERB
bracis-33663	284	43	that	that	PRON
bracis-33663	284	44	are	be	AUX
bracis-33663	284	45	relevant	relevant	ADJ
bracis-33663	284	46	to	to	ADP
bracis-33663	284	47	the	the	DET
bracis-33663	284	48	content	content	NOUN
bracis-33663	284	49	of	of	ADP
bracis-33663	284	50	this	this	DET
bracis-33663	284	51	article	article	NOUN
bracis-33663	284	52	.	.	PUNCT
bracis-33663	285	1	rights	right	NOUN
bracis-33663	285	2	and	and	CCONJ
bracis-33663	285	3	permissions	permission	NOUN
bracis-33663	285	4	reprints	reprint	NOUN
bracis-33663	285	5	and	and	CCONJ
bracis-33663	285	6	permissions	permission	VERB
bracis-33663	285	7	copyright	copyright	NOUN
bracis-33663	285	8	information	information	NOUN
bracis-33663	285	9	©	©	ADP
bracis-33663	285	10	2025	2025	NUM
bracis-33663	285	11	the	the	DET
bracis-33663	285	12	author(s	author(s	NOUN
bracis-33663	285	13	)	)	PUNCT
bracis-33663	285	14	,	,	PUNCT
bracis-33663	285	15	under	under	ADP
bracis-33663	285	16	exclusive	exclusive	ADJ
bracis-33663	285	17	license	license	NOUN
bracis-33663	285	18	to	to	ADP
bracis-33663	285	19	springer	springer	NOUN
bracis-33663	285	20	nature	nature	PROPN
bracis-33663	285	21	switzerland	switzerland	PROPN
bracis-33663	285	22	ag	ag	PROPN
bracis-33663	285	23	about	about	ADP
bracis-33663	285	24	this	this	DET
bracis-33663	285	25	paper	paper	NOUN
bracis-33663	285	26	cite	cite	VERB
bracis-33663	285	27	this	this	DET
bracis-33663	285	28	paper	paper	NOUN
bracis-33663	285	29	vecchi	vecchi	NOUN
bracis-33663	285	30	,	,	PUNCT
bracis-33663	285	31	l.p	l.p	PROPN
bracis-33663	285	32	.	.	PROPN
bracis-33663	285	33	,	,	PUNCT
bracis-33663	285	34	barbon	barbon	NOUN
bracis-33663	285	35	junior	junior	PROPN
bracis-33663	285	36	,	,	PUNCT
bracis-33663	285	37	s.	s.	PROPN
bracis-33663	285	38	,	,	PUNCT
bracis-33663	285	39	paraiso	paraiso	PROPN
bracis-33663	285	40	,	,	PUNCT
bracis-33663	285	41	e.c	e.c	PROPN
bracis-33663	285	42	.	.	PROPN
bracis-33663	285	43	(	(	PUNCT
bracis-33663	285	44	2025	2025	NUM
bracis-33663	285	45	)	)	PUNCT
bracis-33663	285	46	.	.	PUNCT
bracis-33663	286	1	tuning	tune	VERB
bracis-33663	286	2	hypothesis	hypothesis	NOUN
bracis-33663	286	3	creation	creation	NOUN
bracis-33663	286	4	:	:	PUNCT
bracis-33663	286	5	combining	combine	VERB
bracis-33663	286	6	discrete	discrete	ADJ
bracis-33663	286	7	and	and	CCONJ
bracis-33663	286	8	 	 	SPACE
bracis-33663	286	9	continuous	continuous	ADJ
bracis-33663	286	10	spaces	space	NOUN
bracis-33663	286	11	for	for	ADP
bracis-33663	286	12	 	 	SPACE
bracis-33663	286	13	zero	zero	NUM
bracis-33663	286	14	-	-	PUNCT
bracis-33663	286	15	shot	shoot	VERB
bracis-33663	286	16	hate	hate	NOUN
bracis-33663	286	17	speech	speech	NOUN
bracis-33663	286	18	detection	detection	NOUN
bracis-33663	286	19	.	.	PUNCT
bracis-33663	287	1	in	in	ADP
bracis-33663	287	2	:	:	PUNCT
bracis-33663	287	3	paes	paes	PROPN
bracis-33663	287	4	,	,	PUNCT
bracis-33663	287	5	a.	a.	PROPN
bracis-33663	287	6	,	,	PUNCT
bracis-33663	287	7	verri	verri	NOUN
bracis-33663	287	8	,	,	PUNCT
bracis-33663	287	9	f.a.n	f.a.n	ADJ
bracis-33663	287	10	.	.	PUNCT
bracis-33663	287	11	(	(	PUNCT
bracis-33663	287	12	eds	ed	NOUN
bracis-33663	287	13	)	)	PUNCT
bracis-33663	287	14	intelligent	intelligent	ADJ
bracis-33663	287	15	systems	system	NOUN
bracis-33663	287	16	.	.	PUNCT
bracis-33663	287	17	bracis	bracis	PROPN
bracis-33663	287	18	2024	2024	NUM
bracis-33663	287	19	.	.	PUNCT
bracis-33663	288	1	lecture	lecture	NOUN
bracis-33663	288	2	notes	note	NOUN
bracis-33663	288	3	in	in	ADP
bracis-33663	288	4	computer	computer	NOUN
bracis-33663	288	5	science	science	NOUN
bracis-33663	288	6	(	(	PUNCT
bracis-33663	288	7	)	)	PUNCT
bracis-33663	288	8	,	,	PUNCT
bracis-33663	288	9	vol	vol	NOUN
bracis-33663	288	10	15415	15415	NUM
bracis-33663	288	11	.	.	PUNCT
bracis-33663	289	1	springer	springer	NOUN
bracis-33663	289	2	,	,	PUNCT
bracis-33663	289	3	cham	cham	PROPN
bracis-33663	289	4	.	.	PUNCT
bracis-33663	290	1	https://doi.org/10.1007/978-3-031-79038-6_18	https://doi.org/10.1007/978-3-031-79038-6_18	PROPN
bracis-33663	290	2	download	download	NOUN
bracis-33663	290	3	citation	citation	NOUN
bracis-33663	290	4	.ris	.ris	PUNCT
bracis-33663	290	5	.enw	.enw	PROPN
bracis-33663	290	6	.bib	.bib	PUNCT
bracis-33663	291	1	doi	doi	PROPN
bracis-33663	291	2	:	:	PUNCT
bracis-33663	291	3	https://doi.org/10.1007/978-3-031-79038-6_18	https://doi.org/10.1007/978-3-031-79038-6_18	PROPN
bracis-33663	291	4	published	publish	VERB
bracis-33663	291	5	:	:	PUNCT
bracis-33663	291	6	31	31	NUM
bracis-33663	291	7	january	january	PROPN
bracis-33663	291	8	2025	2025	NUM
bracis-33663	291	9	publisher	publisher	NOUN
bracis-33663	291	10	name	name	NOUN
bracis-33663	291	11	:	:	PUNCT
bracis-33663	291	12	springer	springer	NOUN
bracis-33663	291	13	,	,	PUNCT
bracis-33663	291	14	cham	cham	PROPN
bracis-33663	291	15	print	print	PROPN
bracis-33663	291	16	isbn	isbn	PROPN
bracis-33663	291	17	:	:	PUNCT
bracis-33663	291	18	978	978	NUM
bracis-33663	291	19	-	-	SYM
bracis-33663	291	20	3	3	NUM
bracis-33663	291	21	-	-	PUNCT
bracis-33663	291	22	031	031	NUM
bracis-33663	291	23	-	-	PUNCT
bracis-33663	291	24	79037	79037	NUM
bracis-33663	291	25	-	-	SYM
bracis-33663	291	26	9	9	NUM
bracis-33663	291	27	online	online	ADJ
bracis-33663	291	28	isbn	isbn	NOUN
bracis-33663	291	29	:	:	PUNCT
bracis-33663	291	30	978	978	NUM
bracis-33663	291	31	-	-	SYM
bracis-33663	291	32	3	3	NUM
bracis-33663	291	33	-	-	PUNCT
bracis-33663	291	34	031	031	NUM
bracis-33663	291	35	-	-	PUNCT
bracis-33663	291	36	79038	79038	NUM
bracis-33663	291	37	-	-	SYM
bracis-33663	291	38	6	6	NUM
bracis-33663	291	39	ebook	ebook	NOUN
bracis-33663	291	40	packages	package	NOUN
bracis-33663	291	41	:	:	PUNCT
bracis-33663	291	42	computer	computer	NOUN
bracis-33663	291	43	sciencecomputer	sciencecomputer	NOUN
bracis-33663	291	44	science	science	NOUN
bracis-33663	291	45	(	(	PUNCT
bracis-33663	291	46	r0	r0	NOUN
bracis-33663	291	47	)	)	PUNCT
bracis-33663	291	48	share	share	VERB
bracis-33663	291	49	this	this	DET
bracis-33663	291	50	paper	paper	NOUN
bracis-33663	291	51	anyone	anyone	PRON
bracis-33663	291	52	you	you	PRON
bracis-33663	291	53	share	share	VERB
bracis-33663	291	54	the	the	DET
bracis-33663	291	55	following	follow	VERB
bracis-33663	291	56	link	link	NOUN
bracis-33663	291	57	with	with	ADP
bracis-33663	291	58	will	will	AUX
bracis-33663	291	59	be	be	AUX
bracis-33663	291	60	able	able	ADJ
bracis-33663	291	61	to	to	PART
bracis-33663	291	62	read	read	VERB
bracis-33663	291	63	this	this	DET
bracis-33663	291	64	content	content	NOUN
bracis-33663	291	65	:	:	PUNCT
bracis-33663	291	66	get	get	VERB
bracis-33663	291	67	shareable	shareable	ADJ
bracis-33663	291	68	linksorry	linksorry	NOUN
bracis-33663	291	69	,	,	PUNCT
bracis-33663	291	70	a	a	DET
bracis-33663	291	71	shareable	shareable	ADJ
bracis-33663	291	72	link	link	NOUN
bracis-33663	291	73	is	be	AUX
bracis-33663	291	74	not	not	PART
bracis-33663	291	75	currently	currently	ADV
bracis-33663	291	76	available	available	ADJ
bracis-33663	291	77	for	for	ADP
bracis-33663	291	78	this	this	DET
bracis-33663	291	79	article	article	NOUN
bracis-33663	291	80	.	.	PUNCT
bracis-33663	292	1	copy	copy	VERB
bracis-33663	292	2	shareable	shareable	ADJ
bracis-33663	292	3	link	link	NOUN
bracis-33663	292	4	to	to	PART
bracis-33663	292	5	clipboard	clipboard	NOUN
bracis-33663	292	6	provided	provide	VERB
bracis-33663	292	7	by	by	ADP
bracis-33663	292	8	the	the	DET
bracis-33663	292	9	springer	springer	NOUN
bracis-33663	292	10	nature	nature	PROPN
bracis-33663	292	11	sharedit	sharedit	PROPN
bracis-33663	292	12	content	content	NOUN
bracis-33663	292	13	-	-	PUNCT
bracis-33663	292	14	sharing	share	VERB
bracis-33663	292	15	initiative	initiative	NOUN
bracis-33663	292	16	keywords	keyword	NOUN
bracis-33663	292	17	hate	hate	VERB
bracis-33663	292	18	speech	speech	NOUN
bracis-33663	292	19	natural	natural	ADJ
bracis-33663	292	20	language	language	NOUN
bracis-33663	292	21	inference	inference	NOUN
bracis-33663	292	22	zero	zero	NUM
bracis-33663	292	23	-	-	PUNCT
bracis-33663	292	24	shot	shot	NOUN
bracis-33663	292	25	text	text	NOUN
bracis-33663	292	26	classification	classification	NOUN
bracis-33663	292	27	p	p	NOUN
bracis-33663	292	28	-	-	PUNCT
bracis-33663	292	29	tuning	tuning	NOUN
bracis-33663	292	30	publish	publish	NOUN
bracis-33663	292	31	with	with	ADP
bracis-33663	292	32	us	us	PROPN
bracis-33663	292	33	policies	policy	NOUN
bracis-33663	292	34	and	and	CCONJ
bracis-33663	292	35	ethics	ethic	NOUN
bracis-33663	292	36	profiles	profile	NOUN
bracis-33663	292	37	sylvio	sylvio	VERB
bracis-33663	292	38	barbon	barbon	NOUN
bracis-33663	292	39	junior	junior	ADJ
bracis-33663	292	40	view	view	NOUN
bracis-33663	292	41	author	author	NOUN
bracis-33663	292	42	profile	profile	PROPN
bracis-33663	292	43	emerson	emerson	PROPN
bracis-33663	292	44	cabrera	cabrera	PROPN
bracis-33663	292	45	paraiso	paraiso	PROPN
bracis-33663	292	46	view	view	NOUN
bracis-33663	292	47	author	author	NOUN
bracis-33663	292	48	profile	profile	NOUN
bracis-33663	292	49	search	search	NOUN
bracis-33663	292	50	search	search	NOUN
bracis-33663	292	51	by	by	ADP
bracis-33663	292	52	keyword	keyword	NOUN
bracis-33663	292	53	or	or	CCONJ
bracis-33663	292	54	author	author	NOUN
bracis-33663	292	55	search	search	NOUN
bracis-33663	292	56	navigation	navigation	NOUN
bracis-33663	292	57	find	find	VERB
bracis-33663	292	58	a	a	DET
bracis-33663	292	59	journal	journal	NOUN
bracis-33663	292	60	publish	publish	VERB
bracis-33663	292	61	with	with	ADP
bracis-33663	292	62	us	we	PRON
bracis-33663	292	63	track	track	VERB
bracis-33663	292	64	your	your	PRON
bracis-33663	292	65	research	research	NOUN
bracis-33663	292	66	discover	discover	VERB
bracis-33663	292	67	content	content	NOUN
bracis-33663	292	68	journals	journal	NOUN
bracis-33663	292	69	a	a	DET
bracis-33663	292	70	-	-	PUNCT
bracis-33663	292	71	z	z	NOUN
bracis-33663	292	72	books	book	NOUN
bracis-33663	292	73	a	a	DET
bracis-33663	292	74	-	-	PUNCT
bracis-33663	292	75	z	z	NOUN
bracis-33663	292	76	publish	publish	NOUN
bracis-33663	292	77	with	with	ADP
bracis-33663	292	78	us	us	PROPN
bracis-33663	292	79	journal	journal	PROPN
bracis-33663	292	80	finder	finder	PROPN
bracis-33663	292	81	publish	publish	VERB
bracis-33663	292	82	your	your	PRON
bracis-33663	292	83	research	research	NOUN
bracis-33663	292	84	language	language	NOUN
bracis-33663	292	85	editing	edit	VERB
bracis-33663	292	86	open	open	ADJ
bracis-33663	292	87	access	access	NOUN
bracis-33663	292	88	publishing	publishing	NOUN
bracis-33663	292	89	products	product	NOUN
bracis-33663	292	90	and	and	CCONJ
bracis-33663	292	91	services	service	NOUN
bracis-33663	292	92	our	our	PRON
bracis-33663	292	93	products	product	NOUN
bracis-33663	292	94	librarians	librarian	VERB
bracis-33663	292	95	societies	society	NOUN
bracis-33663	292	96	partners	partner	NOUN
bracis-33663	292	97	and	and	CCONJ
bracis-33663	292	98	advertisers	advertiser	NOUN
bracis-33663	292	99	our	our	PRON
bracis-33663	292	100	brands	brand	NOUN
bracis-33663	292	101	springer	springer	NOUN
bracis-33663	292	102	nature	nature	PROPN
bracis-33663	292	103	portfolio	portfolio	PROPN
bracis-33663	292	104	bmc	bmc	PROPN
bracis-33663	292	105	palgrave	palgrave	PROPN
bracis-33663	292	106	macmillan	macmillan	PROPN
bracis-33663	292	107	apress	apress	PROPN
bracis-33663	292	108	discover	discover	VERB
bracis-33663	292	109	your	your	PRON
bracis-33663	292	110	privacy	privacy	NOUN
bracis-33663	292	111	choices	choice	NOUN
bracis-33663	292	112	/	/	SYM
bracis-33663	292	113	manage	manage	NOUN
bracis-33663	292	114	cookies	cookie	NOUN
bracis-33663	292	115	your	your	PRON
bracis-33663	292	116	us	us	PROPN
bracis-33663	293	1	state	state	NOUN
bracis-33663	293	2	privacy	privacy	NOUN
bracis-33663	293	3	rights	right	NOUN
bracis-33663	293	4	accessibility	accessibility	NOUN
bracis-33663	293	5	statement	statement	NOUN
bracis-33663	293	6	terms	term	NOUN
bracis-33663	293	7	and	and	CCONJ
bracis-33663	293	8	conditions	condition	NOUN
bracis-33663	293	9	privacy	privacy	NOUN
bracis-33663	293	10	policy	policy	NOUN
bracis-33663	293	11	help	help	NOUN
bracis-33663	293	12	and	and	CCONJ
bracis-33663	293	13	support	support	VERB
bracis-33663	293	14	legal	legal	ADJ
bracis-33663	293	15	notice	notice	NOUN
bracis-33663	293	16	cancel	cancel	VERB
bracis-33663	293	17	contracts	contract	NOUN
bracis-33663	293	18	here	here	ADV
bracis-33663	293	19	129.74.145.123	129.74.145.123	NUM
bracis-33663	293	20	hesburgh	hesburgh	PROPN
bracis-33663	293	21	library	library	PROPN
bracis-33663	293	22	er	er	INTJ
bracis-33663	293	23	unit	unit	NOUN
bracis-33663	293	24	(	(	PUNCT
bracis-33663	293	25	3005732405	3005732405	NUM
bracis-33663	293	26	)	)	PUNCT
bracis-33663	293	27	northeast	northeast	ADJ
bracis-33663	293	28	research	research	NOUN
bracis-33663	293	29	libraries	library	NOUN
bracis-33663	293	30	(	(	PUNCT
bracis-33663	293	31	nerl	nerl	PROPN
bracis-33663	293	32	)	)	PUNCT
bracis-33663	293	33	(	(	PUNCT
bracis-33663	293	34	8200828607	8200828607	NUM
bracis-33663	293	35	)	)	PUNCT
bracis-33663	293	36	nerl	nerl	VERB
bracis-33663	293	37	ta	ta	X
bracis-33663	293	38	account	account	NOUN
bracis-33663	293	39	(	(	PUNCT
bracis-33663	293	40	3006206169	3006206169	NUM
bracis-33663	293	41	)	)	PUNCT
bracis-33663	293	42	university	university	NOUN
bracis-33663	293	43	of	of	ADP
bracis-33663	293	44	notre	notre	PROPN
bracis-33663	293	45	dame	dame	PROPN
bracis-33663	293	46	hesburgh	hesburgh	PROPN
bracis-33663	293	47	library	library	NOUN
bracis-33663	293	48	(	(	PUNCT
bracis-33663	293	49	3000184373	3000184373	NUM
bracis-33663	293	50	)	)	PUNCT
bracis-33663	294	1	©	©	ADP
bracis-33663	294	2	2025	2025	NUM
bracis-33663	294	3	springer	springer	NOUN
bracis-33663	294	4	nature	nature	NOUN