id	sid	tid	token	lemma	pos
fcis-31632	1	1	frontiers	frontier	NOUN
fcis-31632	1	2	in	in	ADP
fcis-31632	1	3	computing	computing	NOUN
fcis-31632	1	4	and	and	CCONJ
fcis-31632	1	5	intelligent	intelligent	ADJ
fcis-31632	1	6	systems	system	NOUN
fcis-31632	1	7	issn	issn	VERB
fcis-31632	1	8	:	:	PUNCT
fcis-31632	1	9	2832	2832	NUM
fcis-31632	1	10	-	-	SYM
fcis-31632	1	11	6024	6024	NUM
fcis-31632	1	12	|	|	NOUN
fcis-31632	1	13	vol	vol	NOUN
fcis-31632	1	14	.	.	PROPN
fcis-31632	2	1	13	13	NUM
fcis-31632	2	2	,	,	PUNCT
fcis-31632	2	3	no	no	INTJ
fcis-31632	2	4	.	.	NOUN
fcis-31632	2	5	2	2	NUM
fcis-31632	2	6	,	,	PUNCT
fcis-31632	2	7	2025	2025	NUM
fcis-31632	2	8	70	70	NUM
fcis-31632	2	9	analysis	analysis	NOUN
fcis-31632	2	10	on	on	ADP
fcis-31632	2	11	the	the	DET
fcis-31632	2	12	method	method	NOUN
fcis-31632	2	13	of	of	ADP
fcis-31632	2	14	improving	improve	VERB
fcis-31632	2	15	the	the	DET
fcis-31632	2	16	performance	performance	NOUN
fcis-31632	2	17	of	of	ADP
fcis-31632	2	18	natural	natural	ADJ
fcis-31632	2	19	language	language	NOUN
fcis-31632	2	20	processing	processing	NOUN
fcis-31632	2	21	model	model	NOUN
fcis-31632	2	22	driven	drive	VERB
fcis-31632	2	23	by	by	ADP
fcis-31632	2	24	artificial	artificial	ADJ
fcis-31632	2	25	intelligence	intelligence	NOUN
fcis-31632	2	26	zenghua	zenghua	NOUN
fcis-31632	2	27	wang	wang	PROPN
fcis-31632	2	28	*	*	PUNCT
fcis-31632	2	29	yancheng	yancheng	PROPN
fcis-31632	2	30	advanced	advanced	ADJ
fcis-31632	2	31	vocational	vocational	ADJ
fcis-31632	2	32	school	school	NOUN
fcis-31632	2	33	of	of	ADP
fcis-31632	2	34	economics	economic	NOUN
fcis-31632	2	35	and	and	CCONJ
fcis-31632	2	36	trade	trade	NOUN
fcis-31632	2	37	,	,	PUNCT
fcis-31632	2	38	yancheng	yancheng	PROPN
fcis-31632	2	39	,	,	PUNCT
fcis-31632	2	40	jiangsu	jiangsu	PROPN
fcis-31632	2	41	,	,	PUNCT
fcis-31632	2	42	china	china	PROPN
fcis-31632	2	43	*	*	PUNCT
fcis-31632	2	44	corresponding	correspond	VERB
fcis-31632	2	45	author	author	NOUN
fcis-31632	2	46	email	email	NOUN
fcis-31632	2	47	:	:	PUNCT
fcis-31632	2	48	494119434@qq.com	494119434@qq.com	NUM
fcis-31632	2	49	abstract	abstract	ADJ
fcis-31632	2	50	:	:	PUNCT
fcis-31632	2	51	natural	natural	ADJ
fcis-31632	2	52	language	language	NOUN
fcis-31632	2	53	processing	processing	NOUN
fcis-31632	2	54	(	(	PUNCT
fcis-31632	2	55	nlp	nlp	NOUN
fcis-31632	2	56	)	)	PUNCT
fcis-31632	2	57	,	,	PUNCT
fcis-31632	2	58	as	as	ADP
fcis-31632	2	59	the	the	DET
fcis-31632	2	60	core	core	NOUN
fcis-31632	2	61	field	field	NOUN
fcis-31632	2	62	of	of	ADP
fcis-31632	2	63	artificial	artificial	ADJ
fcis-31632	2	64	intelligence	intelligence	NOUN
fcis-31632	2	65	(	(	PUNCT
fcis-31632	2	66	ai	ai	NOUN
fcis-31632	2	67	)	)	PUNCT
fcis-31632	2	68	,	,	PUNCT
fcis-31632	2	69	has	have	AUX
fcis-31632	2	70	made	make	VERB
fcis-31632	2	71	remarkable	remarkable	ADJ
fcis-31632	2	72	progress	progress	NOUN
fcis-31632	2	73	with	with	ADP
fcis-31632	2	74	the	the	DET
fcis-31632	2	75	help	help	NOUN
fcis-31632	2	76	of	of	ADP
fcis-31632	2	77	transformer	transformer	NOUN
fcis-31632	2	78	to	to	AUX
fcis-31632	2	79	pre	pre	VERB
fcis-31632	2	80	-	-	ADJ
fcis-31632	2	81	train	train	ADJ
fcis-31632	2	82	large	large	ADJ
fcis-31632	2	83	models	model	NOUN
fcis-31632	2	84	,	,	PUNCT
fcis-31632	2	85	but	but	CCONJ
fcis-31632	2	86	it	it	PRON
fcis-31632	2	87	is	be	AUX
fcis-31632	2	88	still	still	ADV
fcis-31632	2	89	restricted	restrict	VERB
fcis-31632	2	90	by	by	ADP
fcis-31632	2	91	bottlenecks	bottleneck	NOUN
fcis-31632	2	92	such	such	ADJ
fcis-31632	2	93	as	as	ADP
fcis-31632	2	94	strong	strong	ADJ
fcis-31632	2	95	data	datum	NOUN
fcis-31632	2	96	dependence	dependence	NOUN
fcis-31632	2	97	,	,	PUNCT
fcis-31632	2	98	high	high	ADJ
fcis-31632	2	99	computational	computational	ADJ
fcis-31632	2	100	cost	cost	NOUN
fcis-31632	2	101	,	,	PUNCT
fcis-31632	2	102	and	and	CCONJ
fcis-31632	2	103	insufficient	insufficient	ADJ
fcis-31632	2	104	generalization	generalization	NOUN
fcis-31632	2	105	and	and	CCONJ
fcis-31632	2	106	robustness	robustness	NOUN
fcis-31632	2	107	.	.	PUNCT
fcis-31632	3	1	aiming	aim	VERB
fcis-31632	3	2	at	at	ADP
fcis-31632	3	3	the	the	DET
fcis-31632	3	4	problem	problem	NOUN
fcis-31632	3	5	that	that	SCONJ
fcis-31632	3	6	the	the	DET
fcis-31632	3	7	existing	exist	VERB
fcis-31632	3	8	research	research	NOUN
fcis-31632	3	9	is	be	AUX
fcis-31632	3	10	mostly	mostly	ADV
fcis-31632	3	11	limited	limit	VERB
fcis-31632	3	12	to	to	ADP
fcis-31632	3	13	one	one	NUM
fcis-31632	3	14	-	-	PUNCT
fcis-31632	3	15	dimensional	dimensional	ADJ
fcis-31632	3	16	optimization	optimization	NOUN
fcis-31632	3	17	,	,	PUNCT
fcis-31632	3	18	this	this	DET
fcis-31632	3	19	study	study	NOUN
fcis-31632	3	20	proposes	propose	VERB
fcis-31632	3	21	a	a	DET
fcis-31632	3	22	hierarchical	hierarchical	ADJ
fcis-31632	3	23	optimization	optimization	NOUN
fcis-31632	3	24	framework	framework	NOUN
fcis-31632	3	25	(	(	PUNCT
fcis-31632	3	26	hof	hof	NOUN
fcis-31632	3	27	)	)	PUNCT
fcis-31632	3	28	covering	cover	VERB
fcis-31632	3	29	data	datum	NOUN
fcis-31632	3	30	,	,	PUNCT
fcis-31632	3	31	model	model	NOUN
fcis-31632	3	32	,	,	PUNCT
fcis-31632	3	33	training	training	NOUN
fcis-31632	3	34	and	and	CCONJ
fcis-31632	3	35	deployment	deployment	NOUN
fcis-31632	3	36	,	,	PUNCT
fcis-31632	3	37	and	and	CCONJ
fcis-31632	3	38	improves	improve	VERB
fcis-31632	3	39	the	the	DET
fcis-31632	3	40	performance	performance	NOUN
fcis-31632	3	41	of	of	ADP
fcis-31632	3	42	nlp	nlp	NOUN
fcis-31632	3	43	model	model	NOUN
fcis-31632	3	44	through	through	ADP
fcis-31632	3	45	full	full	ADJ
fcis-31632	3	46	link	link	NOUN
fcis-31632	3	47	collaborative	collaborative	ADJ
fcis-31632	3	48	design	design	NOUN
fcis-31632	3	49	.	.	PUNCT
fcis-31632	4	1	in	in	ADP
fcis-31632	4	2	the	the	DET
fcis-31632	4	3	data	datum	NOUN
fcis-31632	4	4	layer	layer	NOUN
fcis-31632	4	5	,	,	PUNCT
fcis-31632	4	6	an	an	DET
fcis-31632	4	7	antagonistic	antagonistic	ADJ
fcis-31632	4	8	knowledge	knowledge	NOUN
fcis-31632	4	9	injection	injection	NOUN
fcis-31632	4	10	(	(	PUNCT
fcis-31632	4	11	aki	aki	NOUN
fcis-31632	4	12	)	)	PUNCT
fcis-31632	4	13	method	method	NOUN
fcis-31632	4	14	is	be	AUX
fcis-31632	4	15	proposed	propose	VERB
fcis-31632	4	16	,	,	PUNCT
fcis-31632	4	17	which	which	PRON
fcis-31632	4	18	uses	use	VERB
fcis-31632	4	19	external	external	ADJ
fcis-31632	4	20	knowledge	knowledge	NOUN
fcis-31632	4	21	maps	map	NOUN
fcis-31632	4	22	to	to	PART
fcis-31632	4	23	guide	guide	VERB
fcis-31632	4	24	text	text	NOUN
fcis-31632	4	25	generation	generation	NOUN
fcis-31632	4	26	and	and	CCONJ
fcis-31632	4	27	verification	verification	NOUN
fcis-31632	4	28	,	,	PUNCT
fcis-31632	4	29	thus	thus	ADV
fcis-31632	4	30	alleviating	alleviate	VERB
fcis-31632	4	31	the	the	DET
fcis-31632	4	32	problem	problem	NOUN
fcis-31632	4	33	of	of	ADP
fcis-31632	4	34	data	datum	NOUN
fcis-31632	4	35	shortage	shortage	NOUN
fcis-31632	4	36	in	in	ADP
fcis-31632	4	37	low	low	ADJ
fcis-31632	4	38	-	-	PUNCT
fcis-31632	4	39	resource	resource	NOUN
fcis-31632	4	40	scenes	scene	NOUN
fcis-31632	4	41	.	.	PUNCT
fcis-31632	5	1	in	in	ADP
fcis-31632	5	2	the	the	DET
fcis-31632	5	3	model	model	NOUN
fcis-31632	5	4	layer	layer	NOUN
fcis-31632	5	5	,	,	PUNCT
fcis-31632	5	6	dynamic	dynamic	ADJ
fcis-31632	5	7	sparse	sparse	ADJ
fcis-31632	5	8	gating	gate	VERB
fcis-31632	5	9	transformer	transformer	NOUN
fcis-31632	5	10	(	(	PUNCT
fcis-31632	5	11	dsgt	dsgt	NOUN
fcis-31632	5	12	)	)	PUNCT
fcis-31632	5	13	is	be	AUX
fcis-31632	5	14	designed	design	VERB
fcis-31632	5	15	to	to	PART
fcis-31632	5	16	balance	balance	VERB
fcis-31632	5	17	accuracy	accuracy	NOUN
fcis-31632	5	18	and	and	CCONJ
fcis-31632	5	19	reasoning	reasoning	NOUN
fcis-31632	5	20	efficiency	efficiency	NOUN
fcis-31632	5	21	through	through	ADP
fcis-31632	5	22	dynamic	dynamic	ADJ
fcis-31632	5	23	sparse	sparse	ADJ
fcis-31632	5	24	gating	gate	VERB
fcis-31632	5	25	mechanism	mechanism	NOUN
fcis-31632	5	26	.	.	PUNCT
fcis-31632	6	1	in	in	ADP
fcis-31632	6	2	the	the	DET
fcis-31632	6	3	training	training	NOUN
fcis-31632	6	4	layer	layer	NOUN
fcis-31632	6	5	,	,	PUNCT
fcis-31632	6	6	the	the	DET
fcis-31632	6	7	meta	meta	ADJ
fcis-31632	6	8	-	-	PUNCT
fcis-31632	6	9	adaptive	adaptive	ADJ
fcis-31632	6	10	multitask	multitask	NOUN
fcis-31632	6	11	learning	learning	NOUN
fcis-31632	6	12	(	(	PUNCT
fcis-31632	6	13	mamtl	mamtl	NOUN
fcis-31632	6	14	)	)	PUNCT
fcis-31632	6	15	method	method	NOUN
fcis-31632	6	16	driven	drive	VERB
fcis-31632	6	17	by	by	ADP
fcis-31632	6	18	meta	meta	ADJ
fcis-31632	6	19	-	-	PUNCT
fcis-31632	6	20	learning	learning	NOUN
fcis-31632	6	21	is	be	AUX
fcis-31632	6	22	adopted	adopt	VERB
fcis-31632	6	23	to	to	PART
fcis-31632	6	24	enhance	enhance	VERB
fcis-31632	6	25	the	the	DET
fcis-31632	6	26	cross	cross	NOUN
fcis-31632	6	27	-	-	NOUN
fcis-31632	6	28	language	language	NOUN
fcis-31632	6	29	and	and	CCONJ
fcis-31632	6	30	cross	cross	ADJ
fcis-31632	6	31	-	-	ADJ
fcis-31632	6	32	domain	domain	ADJ
fcis-31632	6	33	generalization	generalization	NOUN
fcis-31632	6	34	ability	ability	NOUN
fcis-31632	6	35	.	.	PUNCT
fcis-31632	7	1	in	in	ADP
fcis-31632	7	2	the	the	DET
fcis-31632	7	3	deployment	deployment	NOUN
fcis-31632	7	4	layer	layer	NOUN
fcis-31632	7	5	,	,	PUNCT
fcis-31632	7	6	an	an	DET
fcis-31632	7	7	optimal	optimal	ADJ
fcis-31632	7	8	transport	transport	NOUN
fcis-31632	7	9	alignment	alignment	NOUN
fcis-31632	7	10	(	(	PUNCT
fcis-31632	7	11	ota	ota	PROPN
fcis-31632	7	12	)	)	PUNCT
fcis-31632	7	13	method	method	NOUN
fcis-31632	7	14	based	base	VERB
fcis-31632	7	15	on	on	ADP
fcis-31632	7	16	optimal	optimal	ADJ
fcis-31632	7	17	transport	transport	NOUN
fcis-31632	7	18	is	be	AUX
fcis-31632	7	19	proposed	propose	VERB
fcis-31632	7	20	to	to	PART
fcis-31632	7	21	achieve	achieve	VERB
fcis-31632	7	22	efficient	efficient	ADJ
fcis-31632	7	23	multimodal	multimodal	ADJ
fcis-31632	7	24	semantic	semantic	ADJ
fcis-31632	7	25	fusion	fusion	NOUN
fcis-31632	7	26	.	.	PUNCT
fcis-31632	8	1	the	the	DET
fcis-31632	8	2	experimental	experimental	ADJ
fcis-31632	8	3	results	result	NOUN
fcis-31632	8	4	show	show	VERB
fcis-31632	8	5	that	that	SCONJ
fcis-31632	8	6	the	the	DET
fcis-31632	8	7	bleu	bleu	NOUN
fcis-31632	8	8	of	of	ADP
fcis-31632	8	9	hof	hof	NOUN
fcis-31632	8	10	is	be	AUX
fcis-31632	8	11	47.5	47.5	NUM
fcis-31632	8	12	on	on	ADP
fcis-31632	8	13	flores-200	flores-200	NOUN
fcis-31632	8	14	data	datum	NOUN
fcis-31632	8	15	set	set	PROPN
fcis-31632	8	16	,	,	PUNCT
fcis-31632	8	17	the	the	DET
fcis-31632	8	18	macro	macro	ADJ
fcis-31632	8	19	f1	f1	NOUN
fcis-31632	8	20	is	be	AUX
fcis-31632	8	21	84.7	84.7	NUM
fcis-31632	8	22	on	on	ADP
fcis-31632	8	23	x	x	ADJ
fcis-31632	8	24	-	-	ADJ
fcis-31632	8	25	cross	cross	ADJ
fcis-31632	8	26	data	datum	NOUN
fcis-31632	8	27	set	set	VERB
fcis-31632	8	28	,	,	PUNCT
fcis-31632	8	29	the	the	DET
fcis-31632	8	30	reasoning	reasoning	NOUN
fcis-31632	8	31	delay	delay	NOUN
fcis-31632	8	32	of	of	ADP
fcis-31632	8	33	edge	edge	NOUN
fcis-31632	8	34	devices	device	NOUN
fcis-31632	8	35	is	be	AUX
fcis-31632	8	36	reduced	reduce	VERB
fcis-31632	8	37	to	to	ADP
fcis-31632	8	38	62ms	62ms	NOUN
fcis-31632	8	39	,	,	PUNCT
fcis-31632	8	40	and	and	CCONJ
fcis-31632	8	41	the	the	DET
fcis-31632	8	42	gender	gender	NOUN
fcis-31632	8	43	and	and	CCONJ
fcis-31632	8	44	race	race	NOUN
fcis-31632	8	45	prediction	prediction	NOUN
fcis-31632	8	46	bias	bias	NOUN
fcis-31632	8	47	is	be	AUX
fcis-31632	8	48	significantly	significantly	ADV
fcis-31632	8	49	reduced	reduce	VERB
fcis-31632	8	50	.	.	PUNCT
fcis-31632	9	1	this	this	DET
fcis-31632	9	2	research	research	NOUN
fcis-31632	9	3	fills	fill	VERB
fcis-31632	9	4	the	the	DET
fcis-31632	9	5	gap	gap	NOUN
fcis-31632	9	6	of	of	ADP
fcis-31632	9	7	nlp	nlp	NOUN
fcis-31632	9	8	full	full	PROPN
fcis-31632	9	9	link	link	PROPN
fcis-31632	9	10	collaborative	collaborative	ADJ
fcis-31632	9	11	design	design	NOUN
fcis-31632	9	12	,	,	PUNCT
fcis-31632	9	13	and	and	CCONJ
fcis-31632	9	14	provides	provide	VERB
fcis-31632	9	15	a	a	DET
fcis-31632	9	16	new	new	ADJ
fcis-31632	9	17	paradigm	paradigm	NOUN
fcis-31632	9	18	for	for	ADP
fcis-31632	9	19	promoting	promote	VERB
fcis-31632	9	20	efficient	efficient	ADJ
fcis-31632	9	21	and	and	CCONJ
fcis-31632	9	22	robust	robust	ADJ
fcis-31632	9	23	nlp	nlp	NOUN
fcis-31632	9	24	technology	technology	NOUN
fcis-31632	9	25	and	and	CCONJ
fcis-31632	9	26	building	build	VERB
fcis-31632	9	27	a	a	DET
fcis-31632	9	28	trusted	trust	VERB
fcis-31632	9	29	ai	ai	NOUN
fcis-31632	9	30	system	system	NOUN
fcis-31632	9	31	.	.	PUNCT
fcis-31632	10	1	keywords	keyword	NOUN
fcis-31632	10	2	:	:	PUNCT
fcis-31632	10	3	natural	natural	ADJ
fcis-31632	10	4	language	language	NOUN
fcis-31632	10	5	processing	processing	NOUN
fcis-31632	10	6	;	;	PUNCT
fcis-31632	10	7	artificial	artificial	ADJ
fcis-31632	10	8	intelligence	intelligence	NOUN
fcis-31632	10	9	;	;	PUNCT
fcis-31632	10	10	performance	performance	NOUN
fcis-31632	10	11	improvement	improvement	NOUN
fcis-31632	10	12	;	;	PUNCT
fcis-31632	10	13	hierarchical	hierarchical	ADJ
fcis-31632	10	14	optimization	optimization	NOUN
fcis-31632	10	15	framework	framework	NOUN
fcis-31632	10	16	.	.	PUNCT
fcis-31632	11	1	1	1	X
fcis-31632	11	2	.	.	X
fcis-31632	11	3	introduction	introduction	NOUN
fcis-31632	11	4	as	as	ADP
fcis-31632	11	5	the	the	DET
fcis-31632	11	6	core	core	NOUN
fcis-31632	11	7	field	field	NOUN
fcis-31632	11	8	of	of	ADP
fcis-31632	11	9	artificial	artificial	ADJ
fcis-31632	11	10	intelligence	intelligence	NOUN
fcis-31632	11	11	(	(	PUNCT
fcis-31632	11	12	ai	ai	NOUN
fcis-31632	11	13	)	)	PUNCT
fcis-31632	11	14	,	,	PUNCT
fcis-31632	11	15	natural	natural	ADJ
fcis-31632	11	16	language	language	NOUN
fcis-31632	11	17	processing	processing	NOUN
fcis-31632	11	18	(	(	PUNCT
fcis-31632	11	19	nlp	nlp	NOUN
fcis-31632	11	20	)	)	PUNCT
fcis-31632	11	21	has	have	AUX
fcis-31632	11	22	made	make	VERB
fcis-31632	11	23	remarkable	remarkable	ADJ
fcis-31632	11	24	progress	progress	NOUN
fcis-31632	11	25	in	in	ADP
fcis-31632	11	26	tasks	task	NOUN
fcis-31632	11	27	such	such	ADJ
fcis-31632	11	28	as	as	ADP
fcis-31632	11	29	machine	machine	NOUN
fcis-31632	11	30	translation	translation	NOUN
fcis-31632	11	31	and	and	CCONJ
fcis-31632	11	32	text	text	NOUN
fcis-31632	11	33	generation	generation	NOUN
fcis-31632	11	34	,	,	PUNCT
fcis-31632	11	35	relying	rely	VERB
fcis-31632	11	36	on	on	ADP
fcis-31632	11	37	pre	pre	ADJ
fcis-31632	11	38	-	-	ADJ
fcis-31632	11	39	trained	train	VERB
fcis-31632	11	40	large	large	ADJ
fcis-31632	11	41	models	model	NOUN
fcis-31632	11	42	based	base	VERB
fcis-31632	11	43	on	on	ADP
fcis-31632	11	44	transformer	transformer	NOUN
fcis-31632	11	45	,	,	PUNCT
fcis-31632	11	46	such	such	ADJ
fcis-31632	11	47	as	as	ADP
fcis-31632	11	48	bert	bert	NOUN
fcis-31632	11	49	and	and	CCONJ
fcis-31632	11	50	gpt	gpt	NOUN
fcis-31632	11	51	,	,	PUNCT
fcis-31632	11	52	which	which	PRON
fcis-31632	11	53	has	have	AUX
fcis-31632	11	54	promoted	promote	VERB
fcis-31632	11	55	the	the	DET
fcis-31632	11	56	landing	landing	NOUN
fcis-31632	11	57	of	of	ADP
fcis-31632	11	58	many	many	ADJ
fcis-31632	11	59	kinds	kind	NOUN
fcis-31632	11	60	of	of	ADP
fcis-31632	11	61	intelligent	intelligent	ADJ
fcis-31632	11	62	applications	application	NOUN
fcis-31632	11	63	[	[	X
fcis-31632	11	64	1	1	NUM
fcis-31632	11	65	-	-	SYM
fcis-31632	11	66	2	2	NUM
fcis-31632	11	67	]	]	PUNCT
fcis-31632	11	68	.	.	PUNCT
fcis-31632	12	1	however	however	ADV
fcis-31632	12	2	,	,	PUNCT
fcis-31632	12	3	its	its	PRON
fcis-31632	12	4	development	development	NOUN
fcis-31632	12	5	is	be	AUX
fcis-31632	12	6	still	still	ADV
fcis-31632	12	7	limited	limit	VERB
fcis-31632	12	8	by	by	ADP
fcis-31632	12	9	key	key	ADJ
fcis-31632	12	10	bottlenecks	bottleneck	NOUN
fcis-31632	12	11	such	such	ADJ
fcis-31632	12	12	as	as	ADP
fcis-31632	12	13	strong	strong	ADJ
fcis-31632	12	14	data	datum	NOUN
fcis-31632	12	15	dependence	dependence	NOUN
fcis-31632	12	16	,	,	PUNCT
fcis-31632	12	17	high	high	ADJ
fcis-31632	12	18	computational	computational	ADJ
fcis-31632	12	19	cost	cost	NOUN
fcis-31632	12	20	,	,	PUNCT
fcis-31632	12	21	insufficient	insufficient	ADJ
fcis-31632	12	22	generalization	generalization	NOUN
fcis-31632	12	23	and	and	CCONJ
fcis-31632	12	24	robustness	robustness	NOUN
fcis-31632	12	25	,	,	PUNCT
fcis-31632	12	26	poor	poor	ADJ
fcis-31632	12	27	interpretability	interpretability	NOUN
fcis-31632	12	28	and	and	CCONJ
fcis-31632	12	29	weak	weak	ADJ
fcis-31632	12	30	multimodal	multimodal	NOUN
fcis-31632	12	31	fusion	fusion	NOUN
fcis-31632	12	32	ability	ability	NOUN
fcis-31632	12	33	,	,	PUNCT
fcis-31632	12	34	especially	especially	ADV
fcis-31632	12	35	in	in	ADP
fcis-31632	12	36	low	low	ADJ
fcis-31632	12	37	-	-	PUNCT
fcis-31632	12	38	resource	resource	NOUN
fcis-31632	12	39	scenes	scene	NOUN
fcis-31632	12	40	and	and	CCONJ
fcis-31632	12	41	cross	cross	ADJ
fcis-31632	12	42	-	-	ADJ
fcis-31632	12	43	modal	modal	ADJ
fcis-31632	12	44	interaction	interaction	NOUN
fcis-31632	12	45	.	.	PUNCT
fcis-31632	13	1	this	this	DET
fcis-31632	13	2	study	study	NOUN
fcis-31632	13	3	focuses	focus	VERB
fcis-31632	13	4	on	on	ADP
fcis-31632	13	5	the	the	DET
fcis-31632	13	6	systematic	systematic	ADJ
fcis-31632	13	7	challenge	challenge	NOUN
fcis-31632	13	8	of	of	ADP
fcis-31632	13	9	improving	improve	VERB
fcis-31632	13	10	the	the	DET
fcis-31632	13	11	performance	performance	NOUN
fcis-31632	13	12	of	of	ADP
fcis-31632	13	13	nlp	nlp	NOUN
fcis-31632	13	14	model	model	NOUN
fcis-31632	13	15	.	.	PUNCT
fcis-31632	14	1	aiming	aim	VERB
fcis-31632	14	2	at	at	ADP
fcis-31632	14	3	the	the	DET
fcis-31632	14	4	problem	problem	NOUN
fcis-31632	14	5	that	that	SCONJ
fcis-31632	14	6	the	the	DET
fcis-31632	14	7	existing	exist	VERB
fcis-31632	14	8	research	research	NOUN
fcis-31632	14	9	is	be	AUX
fcis-31632	14	10	mostly	mostly	ADV
fcis-31632	14	11	limited	limit	VERB
fcis-31632	14	12	to	to	ADP
fcis-31632	14	13	singledimensional	singledimensional	ADJ
fcis-31632	14	14	optimization	optimization	NOUN
fcis-31632	14	15	,	,	PUNCT
fcis-31632	14	16	it	it	PRON
fcis-31632	14	17	is	be	AUX
fcis-31632	14	18	devoted	devote	VERB
fcis-31632	14	19	to	to	ADP
fcis-31632	14	20	the	the	DET
fcis-31632	14	21	collaborative	collaborative	ADJ
fcis-31632	14	22	design	design	NOUN
fcis-31632	14	23	of	of	ADP
fcis-31632	14	24	the	the	DET
fcis-31632	14	25	whole	whole	ADJ
fcis-31632	14	26	link	link	NOUN
fcis-31632	14	27	from	from	ADP
fcis-31632	14	28	data	data	PROPN
fcis-31632	14	29	,	,	PUNCT
fcis-31632	14	30	model	model	NOUN
fcis-31632	14	31	,	,	PUNCT
fcis-31632	14	32	training	training	NOUN
fcis-31632	14	33	to	to	ADP
fcis-31632	14	34	deployment	deployment	NOUN
fcis-31632	14	35	.	.	PUNCT
fcis-31632	15	1	the	the	DET
fcis-31632	15	2	core	core	NOUN
fcis-31632	15	3	of	of	ADP
fcis-31632	15	4	this	this	DET
fcis-31632	15	5	study	study	NOUN
fcis-31632	15	6	is	be	AUX
fcis-31632	15	7	to	to	PART
fcis-31632	15	8	explore	explore	VERB
fcis-31632	15	9	how	how	SCONJ
fcis-31632	15	10	to	to	PART
fcis-31632	15	11	build	build	VERB
fcis-31632	15	12	a	a	DET
fcis-31632	15	13	low	low	ADJ
fcis-31632	15	14	-	-	PUNCT
fcis-31632	15	15	resource	resource	NOUN
fcis-31632	15	16	-	-	PUNCT
fcis-31632	15	17	friendly	friendly	ADJ
fcis-31632	15	18	data	datum	NOUN
fcis-31632	15	19	enhancement	enhancement	NOUN
fcis-31632	15	20	and	and	CCONJ
fcis-31632	15	21	knowledge	knowledge	NOUN
fcis-31632	15	22	injection	injection	NOUN
fcis-31632	15	23	method	method	NOUN
fcis-31632	15	24	to	to	PART
fcis-31632	15	25	reduce	reduce	VERB
fcis-31632	15	26	the	the	DET
fcis-31632	15	27	dependence	dependence	NOUN
fcis-31632	15	28	on	on	ADP
fcis-31632	15	29	labeled	label	VERB
fcis-31632	15	30	data	datum	NOUN
fcis-31632	15	31	,	,	PUNCT
fcis-31632	15	32	design	design	VERB
fcis-31632	15	33	an	an	DET
fcis-31632	15	34	efficient	efficient	ADJ
fcis-31632	15	35	and	and	CCONJ
fcis-31632	15	36	lightweight	lightweight	ADJ
fcis-31632	15	37	model	model	NOUN
fcis-31632	15	38	architecture	architecture	NOUN
fcis-31632	15	39	to	to	PART
fcis-31632	15	40	balance	balance	VERB
fcis-31632	15	41	accuracy	accuracy	NOUN
fcis-31632	15	42	and	and	CCONJ
fcis-31632	15	43	reasoning	reasoning	NOUN
fcis-31632	15	44	efficiency	efficiency	NOUN
fcis-31632	15	45	,	,	PUNCT
fcis-31632	15	46	develop	develop	VERB
fcis-31632	15	47	adaptive	adaptive	ADJ
fcis-31632	15	48	training	training	NOUN
fcis-31632	15	49	strategies	strategy	NOUN
fcis-31632	15	50	to	to	PART
fcis-31632	15	51	enhance	enhance	VERB
fcis-31632	15	52	the	the	DET
fcis-31632	15	53	generalization	generalization	NOUN
fcis-31632	15	54	ability	ability	NOUN
fcis-31632	15	55	of	of	ADP
fcis-31632	15	56	the	the	DET
fcis-31632	15	57	model	model	NOUN
fcis-31632	15	58	in	in	ADP
fcis-31632	15	59	cross	cross	NOUN
fcis-31632	15	60	-	-	NOUN
fcis-31632	15	61	language	language	NOUN
fcis-31632	15	62	and	and	CCONJ
fcis-31632	15	63	cross	cross	ADJ
fcis-31632	15	64	-	-	ADJ
fcis-31632	15	65	domain	domain	ADJ
fcis-31632	15	66	scenarios	scenario	NOUN
fcis-31632	15	67	,	,	PUNCT
fcis-31632	15	68	and	and	CCONJ
fcis-31632	15	69	realize	realize	VERB
fcis-31632	15	70	the	the	DET
fcis-31632	15	71	effective	effective	ADJ
fcis-31632	15	72	fusion	fusion	NOUN
fcis-31632	15	73	of	of	ADP
fcis-31632	15	74	multimodal	multimodal	ADJ
fcis-31632	15	75	semantic	semantic	ADJ
fcis-31632	15	76	information	information	NOUN
fcis-31632	15	77	to	to	PART
fcis-31632	15	78	break	break	VERB
fcis-31632	15	79	through	through	ADP
fcis-31632	15	80	purity	purity	NOUN
fcis-31632	15	81	.	.	PUNCT
fcis-31632	16	1	on	on	ADP
fcis-31632	16	2	the	the	DET
fcis-31632	16	3	theoretical	theoretical	ADJ
fcis-31632	16	4	level	level	NOUN
fcis-31632	16	5	,	,	PUNCT
fcis-31632	16	6	a	a	DET
fcis-31632	16	7	hierarchical	hierarchical	ADJ
fcis-31632	16	8	optimization	optimization	NOUN
fcis-31632	16	9	framework	framework	NOUN
fcis-31632	16	10	(	(	PUNCT
fcis-31632	16	11	hof	hof	NOUN
fcis-31632	16	12	)	)	PUNCT
fcis-31632	16	13	covering	cover	VERB
fcis-31632	16	14	data	datum	NOUN
fcis-31632	16	15	,	,	PUNCT
fcis-31632	16	16	model	model	NOUN
fcis-31632	16	17	,	,	PUNCT
fcis-31632	16	18	training	training	NOUN
fcis-31632	16	19	and	and	CCONJ
fcis-31632	16	20	deployment	deployment	NOUN
fcis-31632	16	21	is	be	AUX
fcis-31632	16	22	proposed	propose	VERB
fcis-31632	16	23	,	,	PUNCT
fcis-31632	16	24	which	which	PRON
fcis-31632	16	25	fills	fill	VERB
fcis-31632	16	26	the	the	DET
fcis-31632	16	27	research	research	NOUN
fcis-31632	16	28	gap	gap	NOUN
fcis-31632	16	29	of	of	ADP
fcis-31632	16	30	full	full	ADJ
fcis-31632	16	31	link	link	NOUN
fcis-31632	16	32	collaborative	collaborative	ADJ
fcis-31632	16	33	design	design	NOUN
fcis-31632	16	34	in	in	ADP
fcis-31632	16	35	nlp	nlp	NOUN
fcis-31632	16	36	field	field	NOUN
fcis-31632	16	37	and	and	CCONJ
fcis-31632	16	38	provides	provide	VERB
fcis-31632	16	39	a	a	DET
fcis-31632	16	40	new	new	ADJ
fcis-31632	16	41	paradigm	paradigm	NOUN
fcis-31632	16	42	for	for	ADP
fcis-31632	16	43	performance	performance	NOUN
fcis-31632	16	44	improvement	improvement	NOUN
fcis-31632	16	45	.	.	PUNCT
fcis-31632	17	1	on	on	ADP
fcis-31632	17	2	the	the	DET
fcis-31632	17	3	practical	practical	ADJ
fcis-31632	17	4	level	level	NOUN
fcis-31632	17	5	,	,	PUNCT
fcis-31632	17	6	through	through	ADP
fcis-31632	17	7	the	the	DET
fcis-31632	17	8	experimental	experimental	ADJ
fcis-31632	17	9	verification	verification	NOUN
fcis-31632	17	10	in	in	ADP
fcis-31632	17	11	low	low	ADJ
fcis-31632	17	12	-	-	PUNCT
fcis-31632	17	13	resource	resource	NOUN
fcis-31632	17	14	language	language	NOUN
fcis-31632	17	15	translation	translation	NOUN
fcis-31632	17	16	,	,	PUNCT
fcis-31632	17	17	real	real	ADJ
fcis-31632	17	18	-	-	PUNCT
fcis-31632	17	19	time	time	NOUN
fcis-31632	17	20	reasoning	reasoning	NOUN
fcis-31632	17	21	of	of	ADP
fcis-31632	17	22	edge	edge	NOUN
fcis-31632	17	23	devices	device	NOUN
fcis-31632	17	24	and	and	CCONJ
fcis-31632	17	25	multimodal	multimodal	NOUN
fcis-31632	17	26	tasks	task	NOUN
fcis-31632	17	27	,	,	PUNCT
fcis-31632	17	28	the	the	DET
fcis-31632	17	29	application	application	NOUN
fcis-31632	17	30	of	of	ADP
fcis-31632	17	31	efficient	efficient	ADJ
fcis-31632	17	32	,	,	PUNCT
fcis-31632	17	33	lightweight	lightweight	ADJ
fcis-31632	17	34	and	and	CCONJ
fcis-31632	17	35	robust	robust	ADJ
fcis-31632	17	36	nlp	nlp	NOUN
fcis-31632	17	37	technology	technology	NOUN
fcis-31632	17	38	in	in	ADP
fcis-31632	17	39	real	real	ADJ
fcis-31632	17	40	scenes	scene	NOUN
fcis-31632	17	41	is	be	AUX
fcis-31632	17	42	promoted	promote	VERB
fcis-31632	17	43	;	;	PUNCT
fcis-31632	17	44	at	at	ADP
fcis-31632	17	45	the	the	DET
fcis-31632	17	46	social	social	ADJ
fcis-31632	17	47	level	level	NOUN
fcis-31632	17	48	,	,	PUNCT
fcis-31632	17	49	it	it	PRON
fcis-31632	17	50	is	be	AUX
fcis-31632	17	51	committed	commit	VERB
fcis-31632	17	52	to	to	ADP
fcis-31632	17	53	improving	improve	VERB
fcis-31632	17	54	the	the	DET
fcis-31632	17	55	interpretability	interpretability	NOUN
fcis-31632	17	56	and	and	CCONJ
fcis-31632	17	57	fairness	fairness	NOUN
fcis-31632	17	58	of	of	ADP
fcis-31632	17	59	the	the	DET
fcis-31632	17	60	model	model	NOUN
fcis-31632	17	61	,	,	PUNCT
fcis-31632	17	62	reducing	reduce	VERB
fcis-31632	17	63	the	the	DET
fcis-31632	17	64	risk	risk	NOUN
fcis-31632	17	65	of	of	ADP
fcis-31632	17	66	ai	ai	ADJ
fcis-31632	17	67	bias	bias	NOUN
fcis-31632	17	68	and	and	CCONJ
fcis-31632	17	69	helping	help	VERB
fcis-31632	17	70	to	to	PART
fcis-31632	17	71	build	build	VERB
fcis-31632	17	72	a	a	DET
fcis-31632	17	73	credible	credible	ADJ
fcis-31632	17	74	and	and	CCONJ
fcis-31632	17	75	responsible	responsible	ADJ
fcis-31632	17	76	ai	ai	NOUN
fcis-31632	17	77	system	system	NOUN
fcis-31632	17	78	.	.	PUNCT
fcis-31632	18	1	2	2	X
fcis-31632	18	2	.	.	X
fcis-31632	18	3	core	core	NOUN
fcis-31632	18	4	methodology	methodology	NOUN
fcis-31632	18	5	2.1	2.1	NUM
fcis-31632	18	6	.	.	PUNCT
fcis-31632	19	1	performance	performance	NOUN
fcis-31632	19	2	improvement	improvement	NOUN
fcis-31632	19	3	technical	technical	ADJ
fcis-31632	19	4	framework	framework	NOUN
fcis-31632	19	5	hof	hof	NOUN
fcis-31632	19	6	improves	improve	VERB
fcis-31632	19	7	the	the	DET
fcis-31632	19	8	performance	performance	NOUN
fcis-31632	19	9	of	of	ADP
fcis-31632	19	10	nlp	nlp	NOUN
fcis-31632	19	11	model	model	NOUN
fcis-31632	19	12	through	through	ADP
fcis-31632	19	13	systematic	systematic	ADJ
fcis-31632	19	14	collaborative	collaborative	ADJ
fcis-31632	19	15	design	design	NOUN
fcis-31632	19	16	.	.	PUNCT
fcis-31632	20	1	as	as	SCONJ
fcis-31632	20	2	shown	show	VERB
fcis-31632	20	3	in	in	ADP
fcis-31632	20	4	figure	figure	NOUN
fcis-31632	20	5	1	1	NUM
fcis-31632	20	6	,	,	PUNCT
fcis-31632	20	7	the	the	DET
fcis-31632	20	8	framework	framework	NOUN
fcis-31632	20	9	covers	cover	VERB
fcis-31632	20	10	data	datum	NOUN
fcis-31632	20	11	layer	layer	NOUN
fcis-31632	20	12	,	,	PUNCT
fcis-31632	20	13	model	model	NOUN
fcis-31632	20	14	layer	layer	NOUN
fcis-31632	20	15	,	,	PUNCT
fcis-31632	20	16	training	training	NOUN
fcis-31632	20	17	layer	layer	NOUN
fcis-31632	20	18	and	and	CCONJ
fcis-31632	20	19	deployment	deployment	NOUN
fcis-31632	20	20	layer	layer	NOUN
fcis-31632	20	21	:	:	PUNCT
fcis-31632	20	22	in	in	ADP
fcis-31632	20	23	the	the	DET
fcis-31632	20	24	data	datum	NOUN
fcis-31632	20	25	layer	layer	NOUN
fcis-31632	20	26	,	,	PUNCT
fcis-31632	20	27	low	low	ADJ
fcis-31632	20	28	-	-	PUNCT
fcis-31632	20	29	resource	resource	NOUN
fcis-31632	20	30	data	datum	NOUN
fcis-31632	20	31	enhancement	enhancement	NOUN
fcis-31632	20	32	and	and	CCONJ
fcis-31632	20	33	knowledge	knowledge	NOUN
fcis-31632	20	34	injection	injection	NOUN
fcis-31632	20	35	technology	technology	NOUN
fcis-31632	20	36	are	be	AUX
fcis-31632	20	37	adopted	adopt	VERB
fcis-31632	20	38	to	to	PART
fcis-31632	20	39	reduce	reduce	VERB
fcis-31632	20	40	the	the	DET
fcis-31632	20	41	dependence	dependence	NOUN
fcis-31632	20	42	on	on	ADP
fcis-31632	20	43	large	large	ADJ
fcis-31632	20	44	-	-	PUNCT
fcis-31632	20	45	scale	scale	NOUN
fcis-31632	20	46	annotation	annotation	NOUN
fcis-31632	20	47	data	datum	NOUN
fcis-31632	21	1	[	[	X
fcis-31632	21	2	3	3	NUM
fcis-31632	21	3	]	]	PUNCT
fcis-31632	21	4	;	;	PUNCT
fcis-31632	21	5	design	design	VERB
fcis-31632	21	6	a	a	DET
fcis-31632	21	7	dynamic	dynamic	ADJ
fcis-31632	21	8	sparse	sparse	ADJ
fcis-31632	21	9	compression	compression	NOUN
fcis-31632	21	10	architecture	architecture	NOUN
fcis-31632	21	11	in	in	ADP
fcis-31632	21	12	the	the	DET
fcis-31632	21	13	model	model	NOUN
fcis-31632	21	14	layer	layer	NOUN
fcis-31632	21	15	,	,	PUNCT
fcis-31632	21	16	taking	take	VERB
fcis-31632	21	17	into	into	ADP
fcis-31632	21	18	account	account	NOUN
fcis-31632	21	19	the	the	DET
fcis-31632	21	20	model	model	NOUN
fcis-31632	21	21	accuracy	accuracy	NOUN
fcis-31632	21	22	and	and	CCONJ
fcis-31632	21	23	computational	computational	ADJ
fcis-31632	21	24	efficiency	efficiency	NOUN
fcis-31632	21	25	;	;	PUNCT
fcis-31632	21	26	adaptive	adaptive	ADJ
fcis-31632	21	27	multi	multi	ADJ
fcis-31632	21	28	-	-	ADJ
fcis-31632	21	29	task	task	ADJ
fcis-31632	21	30	joint	joint	ADJ
fcis-31632	21	31	learning	learning	NOUN
fcis-31632	21	32	strategy	strategy	NOUN
fcis-31632	21	33	is	be	AUX
fcis-31632	21	34	introduced	introduce	VERB
fcis-31632	21	35	in	in	ADP
fcis-31632	21	36	the	the	DET
fcis-31632	21	37	training	training	NOUN
fcis-31632	21	38	layer	layer	NOUN
fcis-31632	21	39	to	to	PART
fcis-31632	21	40	enhance	enhance	VERB
fcis-31632	21	41	the	the	DET
fcis-31632	21	42	generalization	generalization	NOUN
fcis-31632	21	43	ability	ability	NOUN
fcis-31632	21	44	of	of	ADP
fcis-31632	21	45	the	the	DET
fcis-31632	21	46	model	model	NOUN
fcis-31632	21	47	in	in	ADP
fcis-31632	21	48	cross	cross	NOUN
fcis-31632	21	49	-	-	ADJ
fcis-31632	21	50	domain	domain	ADJ
fcis-31632	21	51	and	and	CCONJ
fcis-31632	21	52	cross	cross	ADJ
fcis-31632	21	53	-	-	ADJ
fcis-31632	21	54	task	task	ADJ
fcis-31632	21	55	scenarios	scenario	NOUN
fcis-31632	21	56	[	[	X
fcis-31632	21	57	4	4	NUM
fcis-31632	21	58	-	-	SYM
fcis-31632	21	59	5	5	NUM
fcis-31632	21	60	]	]	PUNCT
fcis-31632	21	61	;	;	PUNCT
fcis-31632	21	62	multi	multi	ADJ
fcis-31632	21	63	-	-	ADJ
fcis-31632	21	64	modal	modal	ADJ
fcis-31632	21	65	semantic	semantic	ADJ
fcis-31632	21	66	alignment	alignment	NOUN
fcis-31632	21	67	reasoning	reasoning	NOUN
fcis-31632	21	68	is	be	AUX
fcis-31632	21	69	realized	realize	VERB
fcis-31632	21	70	in	in	ADP
fcis-31632	21	71	the	the	DET
fcis-31632	21	72	deployment	deployment	NOUN
fcis-31632	21	73	layer	layer	NOUN
fcis-31632	21	74	,	,	PUNCT
fcis-31632	21	75	which	which	PRON
fcis-31632	21	76	supports	support	VERB
fcis-31632	21	77	efficient	efficient	ADJ
fcis-31632	21	78	and	and	CCONJ
fcis-31632	21	79	accurate	accurate	ADJ
fcis-31632	21	80	interaction	interaction	NOUN
fcis-31632	21	81	in	in	ADP
fcis-31632	21	82	complex	complex	ADJ
fcis-31632	21	83	real	real	ADJ
fcis-31632	21	84	scenes	scene	NOUN
fcis-31632	21	85	.	.	PUNCT
fcis-31632	22	1	71	71	NUM
fcis-31632	22	2	figure	figure	NOUN
fcis-31632	22	3	1	1	NUM
fcis-31632	22	4	.	.	NOUN
fcis-31632	22	5	hof	hof	NOUN
fcis-31632	22	6	frame	frame	NOUN
fcis-31632	22	7	structure	structure	NOUN
fcis-31632	22	8	2.2	2.2	NUM
fcis-31632	22	9	.	.	PUNCT
fcis-31632	22	10	detailed	detailed	ADJ
fcis-31632	22	11	description	description	NOUN
fcis-31632	22	12	of	of	ADP
fcis-31632	22	13	innovative	innovative	ADJ
fcis-31632	22	14	methods	method	NOUN
fcis-31632	22	15	propose	propose	VERB
fcis-31632	22	16	adversarial	adversarial	ADJ
fcis-31632	22	17	knowledge	knowledge	NOUN
fcis-31632	22	18	injection	injection	NOUN
fcis-31632	22	19	(	(	PUNCT
fcis-31632	22	20	aki	aki	NOUN
fcis-31632	22	21	)	)	PUNCT
fcis-31632	22	22	method	method	NOUN
fcis-31632	22	23	at	at	ADP
fcis-31632	22	24	the	the	DET
fcis-31632	22	25	data	data	NOUN
fcis-31632	22	26	layer	layer	NOUN
fcis-31632	22	27	to	to	PART
fcis-31632	22	28	alleviate	alleviate	VERB
fcis-31632	22	29	the	the	DET
fcis-31632	22	30	problem	problem	NOUN
fcis-31632	22	31	of	of	ADP
fcis-31632	22	32	annotated	annotate	VERB
fcis-31632	22	33	data	datum	NOUN
fcis-31632	22	34	scarcity	scarcity	NOUN
fcis-31632	22	35	in	in	ADP
fcis-31632	22	36	low	low	ADJ
fcis-31632	22	37	resource	resource	NOUN
fcis-31632	22	38	scenarios	scenario	NOUN
fcis-31632	22	39	[	[	X
fcis-31632	22	40	6	6	NUM
fcis-31632	22	41	]	]	PUNCT
fcis-31632	22	42	.	.	PUNCT
fcis-31632	23	1	this	this	DET
fcis-31632	23	2	method	method	NOUN
fcis-31632	23	3	introduces	introduce	VERB
fcis-31632	23	4	an	an	DET
fcis-31632	23	5	external	external	ADJ
fcis-31632	23	6	structured	structured	ADJ
fcis-31632	23	7	knowledge	knowledge	NOUN
fcis-31632	23	8	graph	graph	NOUN
fcis-31632	23	9	as	as	ADP
fcis-31632	23	10	a	a	DET
fcis-31632	23	11	semantic	semantic	ADJ
fcis-31632	23	12	prior	prior	ADV
fcis-31632	23	13	to	to	PART
fcis-31632	23	14	guide	guide	VERB
fcis-31632	23	15	the	the	DET
fcis-31632	23	16	generation	generation	NOUN
fcis-31632	23	17	and	and	CCONJ
fcis-31632	23	18	validation	validation	NOUN
fcis-31632	23	19	of	of	ADP
fcis-31632	23	20	high	high	ADJ
fcis-31632	23	21	-	-	PUNCT
fcis-31632	23	22	quality	quality	NOUN
fcis-31632	23	23	text	text	NOUN
fcis-31632	23	24	.	.	PUNCT
fcis-31632	24	1	the	the	DET
fcis-31632	24	2	core	core	NOUN
fcis-31632	24	3	mechanism	mechanism	NOUN
fcis-31632	24	4	is	be	AUX
fcis-31632	24	5	to	to	PART
fcis-31632	24	6	build	build	VERB
fcis-31632	24	7	a	a	DET
fcis-31632	24	8	dual	dual	ADJ
fcis-31632	24	9	channel	channel	NOUN
fcis-31632	24	10	adversarial	adversarial	ADJ
fcis-31632	24	11	generator	generator	NOUN
fcis-31632	24	12	:	:	PUNCT
fcis-31632	24	13	the	the	DET
fcis-31632	24	14	generator	generator	NOUN
fcis-31632	24	15	uses	use	VERB
fcis-31632	24	16	a	a	DET
fcis-31632	24	17	conditional	conditional	ADJ
fcis-31632	24	18	variational	variational	ADJ
fcis-31632	24	19	autoencoder	autoencoder	NOUN
fcis-31632	24	20	(	(	PUNCT
fcis-31632	24	21	cvae	cvae	NOUN
fcis-31632	24	22	)	)	PUNCT
fcis-31632	24	23	to	to	PART
fcis-31632	24	24	generate	generate	VERB
fcis-31632	24	25	semantically	semantically	ADV
fcis-31632	24	26	reasonable	reasonable	ADJ
fcis-31632	24	27	text	text	NOUN
fcis-31632	24	28	based	base	VERB
fcis-31632	24	29	on	on	ADP
fcis-31632	24	30	knowledge	knowledge	NOUN
fcis-31632	24	31	constraints	constraint	NOUN
fcis-31632	24	32	,	,	PUNCT
fcis-31632	24	33	while	while	SCONJ
fcis-31632	24	34	the	the	DET
fcis-31632	24	35	discriminator	discriminator	NOUN
fcis-31632	24	36	jointly	jointly	ADV
fcis-31632	24	37	judges	judge	VERB
fcis-31632	24	38	the	the	DET
fcis-31632	24	39	authenticity	authenticity	NOUN
fcis-31632	24	40	and	and	CCONJ
fcis-31632	24	41	knowledge	knowledge	NOUN
fcis-31632	24	42	consistency	consistency	NOUN
fcis-31632	24	43	of	of	ADP
fcis-31632	24	44	the	the	DET
fcis-31632	24	45	generated	generate	VERB
fcis-31632	24	46	text	text	NOUN
fcis-31632	24	47	,	,	PUNCT
fcis-31632	24	48	ensuring	ensure	VERB
fcis-31632	24	49	that	that	SCONJ
fcis-31632	24	50	the	the	DET
fcis-31632	24	51	output	output	NOUN
fcis-31632	24	52	is	be	AUX
fcis-31632	24	53	both	both	ADV
fcis-31632	24	54	close	close	ADJ
fcis-31632	24	55	to	to	ADP
fcis-31632	24	56	the	the	DET
fcis-31632	24	57	real	real	ADJ
fcis-31632	24	58	language	language	NOUN
fcis-31632	24	59	distribution	distribution	NOUN
fcis-31632	24	60	and	and	CCONJ
fcis-31632	24	61	conforms	conform	VERB
fcis-31632	24	62	to	to	ADP
fcis-31632	24	63	the	the	DET
fcis-31632	24	64	known	know	VERB
fcis-31632	24	65	knowledge	knowledge	NOUN
fcis-31632	24	66	structure	structure	NOUN
fcis-31632	24	67	.	.	PUNCT
fcis-31632	25	1	in	in	ADP
fcis-31632	25	2	order	order	NOUN
fcis-31632	25	3	to	to	PART
fcis-31632	25	4	realize	realize	VERB
fcis-31632	25	5	the	the	DET
fcis-31632	25	6	effective	effective	ADJ
fcis-31632	25	7	fusion	fusion	NOUN
fcis-31632	25	8	of	of	ADP
fcis-31632	25	9	knowledge	knowledge	NOUN
fcis-31632	25	10	,	,	PUNCT
fcis-31632	25	11	aki	aki	NOUN
fcis-31632	25	12	designs	design	VERB
fcis-31632	25	13	a	a	DET
fcis-31632	25	14	knowledge	knowledge	NOUN
fcis-31632	25	15	constraint	constraint	NOUN
fcis-31632	25	16	loss	loss	NOUN
fcis-31632	25	17	function	function	NOUN
fcis-31632	25	18	:	:	PUNCT
fcis-31632	26	1			NOUN
fcis-31632	26	2			SYM
fcis-31632	26	3			NOUN
fcis-31632	26	4			NOUN
fcis-31632	26	5			PUNCT
fcis-31632	26	6			NOUN
fcis-31632	26	7			PRON
fcis-31632	27	1	kzpxzqklxdel	kzpxzqklxdel	VERB
fcis-31632	27	2	datapxkc	datapxkc	ADJ
fcis-31632	27	3			ADP
fcis-31632	27	4	log~	log~	ADJ
fcis-31632	27	5	(	(	PUNCT
fcis-31632	27	6	1	1	NUM
fcis-31632	27	7	)	)	PUNCT
fcis-31632	27	8	where	where	SCONJ
fcis-31632	27	9			NOUN
fcis-31632	27	10	xzq	xzq	PROPN
fcis-31632	27	11	represents	represent	VERB
fcis-31632	27	12	the	the	DET
fcis-31632	27	13	potential	potential	ADJ
fcis-31632	27	14	semantic	semantic	ADJ
fcis-31632	27	15	distribution	distribution	NOUN
fcis-31632	27	16	of	of	ADP
fcis-31632	27	17	text	text	NOUN
fcis-31632	27	18	x	x	SYM
fcis-31632	27	19	,	,	PUNCT
fcis-31632	27	20			PROPN
fcis-31632	27	21	kzp	kzp	INTJ
fcis-31632	27	22	is	be	AUX
fcis-31632	27	23	the	the	DET
fcis-31632	27	24	prior	prior	ADJ
fcis-31632	27	25	distribution	distribution	NOUN
fcis-31632	27	26	provided	provide	VERB
fcis-31632	27	27	by	by	ADP
fcis-31632	27	28	knowledge	knowledge	NOUN
fcis-31632	27	29	graph	graph	NOUN
fcis-31632	27	30	k	k	PROPN
fcis-31632	27	31	,	,	PUNCT
fcis-31632	27	32	and	and	CCONJ
fcis-31632	27	33			NOUN
fcis-31632	27	34	controls	control	VERB
fcis-31632	27	35	the	the	DET
fcis-31632	27	36	knowledge	knowledge	NOUN
fcis-31632	27	37	constraint	constraint	NOUN
fcis-31632	27	38	strength	strength	NOUN
fcis-31632	27	39	(	(	PUNCT
fcis-31632	27	40	set	set	VERB
fcis-31632	27	41	to	to	ADP
fcis-31632	27	42	0.8	0.8	NUM
fcis-31632	27	43	in	in	ADP
fcis-31632	27	44	the	the	DET
fcis-31632	27	45	experiment	experiment	NOUN
fcis-31632	27	46	)	)	PUNCT
fcis-31632	27	47	.	.	PUNCT
fcis-31632	28	1	in	in	ADP
fcis-31632	28	2	the	the	DET
fcis-31632	28	3	model	model	NOUN
fcis-31632	28	4	layer	layer	NOUN
fcis-31632	28	5	,	,	PUNCT
fcis-31632	28	6	dynamic	dynamic	ADJ
fcis-31632	28	7	sparse	sparse	ADJ
fcis-31632	28	8	gating	gate	VERB
fcis-31632	28	9	transformer	transformer	NOUN
fcis-31632	28	10	(	(	PUNCT
fcis-31632	28	11	dsgt	dsgt	NOUN
fcis-31632	28	12	)	)	PUNCT
fcis-31632	28	13	is	be	AUX
fcis-31632	28	14	proposed	propose	VERB
fcis-31632	28	15	to	to	PART
fcis-31632	28	16	solve	solve	VERB
fcis-31632	28	17	the	the	DET
fcis-31632	28	18	contradiction	contradiction	NOUN
fcis-31632	28	19	between	between	ADP
fcis-31632	28	20	accuracy	accuracy	NOUN
fcis-31632	28	21	and	and	CCONJ
fcis-31632	28	22	reasoning	reasoning	NOUN
fcis-31632	28	23	efficiency	efficiency	NOUN
fcis-31632	28	24	of	of	ADP
fcis-31632	28	25	large	large	ADJ
fcis-31632	28	26	models	model	NOUN
fcis-31632	28	27	.	.	PUNCT
fcis-31632	29	1	the	the	DET
fcis-31632	29	2	core	core	NOUN
fcis-31632	29	3	of	of	ADP
fcis-31632	29	4	dsgt	dsgt	NOUN
fcis-31632	29	5	is	be	AUX
fcis-31632	29	6	to	to	PART
fcis-31632	29	7	introduce	introduce	VERB
fcis-31632	29	8	the	the	DET
fcis-31632	29	9	dynamic	dynamic	ADJ
fcis-31632	29	10	sparse	sparse	ADJ
fcis-31632	29	11	gating	gate	VERB
fcis-31632	29	12	mechanism	mechanism	NOUN
fcis-31632	29	13	,	,	PUNCT
fcis-31632	29	14	and	and	CCONJ
fcis-31632	29	15	calculate	calculate	VERB
fcis-31632	29	16	the	the	DET
fcis-31632	29	17	gating	gate	VERB
fcis-31632	29	18	vector	vector	NOUN
fcis-31632	29	19	through	through	ADP
fcis-31632	29	20	the	the	DET
fcis-31632	29	21	learnable	learnable	ADJ
fcis-31632	29	22	gating	gate	VERB
fcis-31632	29	23	weight	weight	NOUN
fcis-31632	29	24	gw	gw	NOUN
fcis-31632	29	25	,	,	PUNCT
fcis-31632	29	26	combining	combine	VERB
fcis-31632	29	27	the	the	DET
fcis-31632	29	28	previous	previous	ADJ
fcis-31632	29	29	state	state	NOUN
fcis-31632	29	30	1lh	1lh	NUM
fcis-31632	29	31	of	of	ADP
fcis-31632	29	32	the	the	DET
fcis-31632	29	33	current	current	ADJ
fcis-31632	29	34	layer	layer	NOUN
fcis-31632	29	35	and	and	CCONJ
fcis-31632	29	36	the	the	DET
fcis-31632	29	37	calculated	calculate	VERB
fcis-31632	29	38	cost	cost	NOUN
fcis-31632	29	39	prediction	prediction	NOUN
fcis-31632	29	40	vector	vector	NOUN
fcis-31632	29	41	lc	lc	PROPN
fcis-31632	29	42	:	:	PUNCT
fcis-31632	29	43			VERB
fcis-31632	29	44			VERB
fcis-31632	29	45			PROPN
fcis-31632	29	46			PROPN
fcis-31632	29	47	d	d	PROPN
fcis-31632	29	48	ligl	ligl	ADJ
fcis-31632	29	49	chwg	chwg	NOUN
fcis-31632	29	50	1,0;1	1,0;1	NOUN
fcis-31632	29	51			PROPN
fcis-31632	29	52			PROPN
fcis-31632	29	53	(	(	PUNCT
fcis-31632	29	54	2	2	NUM
fcis-31632	29	55	)	)	PUNCT
fcis-31632	29	56	where	where	SCONJ
fcis-31632	29	57			PROPN
fcis-31632	29	58	is	be	AUX
fcis-31632	29	59	a	a	DET
fcis-31632	29	60	sigmoid	sigmoid	NOUN
fcis-31632	29	61	function	function	NOUN
fcis-31632	29	62	,	,	PUNCT
fcis-31632	29	63	and	and	CCONJ
fcis-31632	29	64	then	then	ADV
fcis-31632	29	65	binarization	binarization	NOUN
fcis-31632	29	66	processing	processing	NOUN
fcis-31632	29	67	(	(	PUNCT
fcis-31632	29	68	threshold	threshold	NOUN
fcis-31632	29	69	0.5	0.5	NUM
fcis-31632	29	70	)	)	PUNCT
fcis-31632	29	71	is	be	AUX
fcis-31632	29	72	performed	perform	VERB
fcis-31632	29	73	to	to	PART
fcis-31632	29	74	generate	generate	VERB
fcis-31632	29	75	a	a	DET
fcis-31632	29	76	binary	binary	ADJ
fcis-31632	29	77	gating	gate	VERB
fcis-31632	29	78	signal	signal	NOUN
fcis-31632	29	79	to	to	PART
fcis-31632	29	80	determine	determine	VERB
fcis-31632	29	81	whether	whether	SCONJ
fcis-31632	29	82	each	each	DET
fcis-31632	29	83	layer	layer	NOUN
fcis-31632	29	84	in	in	ADP
fcis-31632	29	85	the	the	DET
fcis-31632	29	86	network	network	NOUN
fcis-31632	29	87	is	be	AUX
fcis-31632	29	88	activated	activate	VERB
fcis-31632	29	89	.	.	PUNCT
fcis-31632	30	1	the	the	DET
fcis-31632	30	2	gating	gate	VERB
fcis-31632	30	3	vector	vector	NOUN
fcis-31632	30	4	lg	lg	NOUN
fcis-31632	30	5	realizes	realize	VERB
fcis-31632	30	6	the	the	DET
fcis-31632	30	7	selective	selective	ADJ
fcis-31632	30	8	activation	activation	NOUN
fcis-31632	30	9	mechanism	mechanism	NOUN
fcis-31632	30	10	:	:	PUNCT
fcis-31632	30	11	only	only	ADV
fcis-31632	30	12	when	when	SCONJ
fcis-31632	30	13	the	the	DET
fcis-31632	30	14	corresponding	corresponding	ADJ
fcis-31632	30	15	position	position	NOUN
fcis-31632	30	16	of	of	ADP
fcis-31632	30	17	lg	lg	NOUN
fcis-31632	30	18	is	be	AUX
fcis-31632	30	19	1	1	NUM
fcis-31632	30	20	,	,	PUNCT
fcis-31632	30	21	the	the	DET
fcis-31632	30	22	complete	complete	ADJ
fcis-31632	30	23	calculation	calculation	NOUN
fcis-31632	30	24	of	of	ADP
fcis-31632	30	25	this	this	DET
fcis-31632	30	26	layer	layer	NOUN
fcis-31632	30	27	is	be	AUX
fcis-31632	30	28	performed	perform	VERB
fcis-31632	30	29	,	,	PUNCT
fcis-31632	30	30	otherwise	otherwise	ADV
fcis-31632	30	31	,	,	PUNCT
fcis-31632	30	32	the	the	DET
fcis-31632	30	33	layer	layer	NOUN
fcis-31632	30	34	is	be	AUX
fcis-31632	30	35	skipped	skip	VERB
fcis-31632	30	36	to	to	PART
fcis-31632	30	37	save	save	VERB
fcis-31632	30	38	calculation	calculation	NOUN
fcis-31632	30	39	resources	resource	NOUN
fcis-31632	30	40	[	[	X
fcis-31632	30	41	7	7	NUM
fcis-31632	30	42	]	]	PUNCT
fcis-31632	30	43	.	.	PUNCT
fcis-31632	31	1	this	this	DET
fcis-31632	31	2	dynamic	dynamic	ADJ
fcis-31632	31	3	sparse	sparse	ADJ
fcis-31632	31	4	strategy	strategy	NOUN
fcis-31632	31	5	enables	enable	VERB
fcis-31632	31	6	the	the	DET
fcis-31632	31	7	model	model	NOUN
fcis-31632	31	8	to	to	PART
fcis-31632	31	9	adaptively	adaptively	ADV
fcis-31632	31	10	allocate	allocate	VERB
fcis-31632	31	11	the	the	DET
fcis-31632	31	12	computation	computation	NOUN
fcis-31632	31	13	according	accord	VERB
fcis-31632	31	14	to	to	ADP
fcis-31632	31	15	the	the	DET
fcis-31632	31	16	input	input	NOUN
fcis-31632	31	17	complexity	complexity	NOUN
fcis-31632	31	18	,	,	PUNCT
fcis-31632	31	19	quickly	quickly	ADV
fcis-31632	31	20	reason	reason	NOUN
fcis-31632	31	21	on	on	ADP
fcis-31632	31	22	simple	simple	ADJ
fcis-31632	31	23	samples	sample	NOUN
fcis-31632	31	24	,	,	PUNCT
fcis-31632	31	25	and	and	CCONJ
fcis-31632	31	26	reserve	reserve	VERB
fcis-31632	31	27	the	the	DET
fcis-31632	31	28	deep	deep	ADJ
fcis-31632	31	29	processing	processing	NOUN
fcis-31632	31	30	ability	ability	NOUN
fcis-31632	31	31	on	on	ADP
fcis-31632	31	32	complex	complex	ADJ
fcis-31632	31	33	samples	sample	NOUN
fcis-31632	31	34	,	,	PUNCT
fcis-31632	31	35	thus	thus	ADV
fcis-31632	31	36	greatly	greatly	ADV
fcis-31632	31	37	reducing	reduce	VERB
fcis-31632	31	38	the	the	DET
fcis-31632	31	39	overall	overall	ADJ
fcis-31632	31	40	calculation	calculation	NOUN
fcis-31632	31	41	cost	cost	NOUN
fcis-31632	31	42	and	and	CCONJ
fcis-31632	31	43	delay	delay	VERB
fcis-31632	31	44	without	without	ADP
fcis-31632	31	45	significantly	significantly	ADV
fcis-31632	31	46	sacrificing	sacrifice	VERB
fcis-31632	31	47	the	the	DET
fcis-31632	31	48	accuracy	accuracy	NOUN
fcis-31632	31	49	of	of	ADP
fcis-31632	31	50	the	the	DET
fcis-31632	31	51	model	model	NOUN
fcis-31632	31	52	,	,	PUNCT
fcis-31632	31	53	especially	especially	ADV
fcis-31632	31	54	suitable	suitable	ADJ
fcis-31632	31	55	for	for	ADP
fcis-31632	31	56	edge	edge	NOUN
fcis-31632	31	57	devices	device	NOUN
fcis-31632	31	58	and	and	CCONJ
fcis-31632	31	59	real	real	ADJ
fcis-31632	31	60	-	-	PUNCT
fcis-31632	31	61	time	time	NOUN
fcis-31632	31	62	application	application	NOUN
fcis-31632	31	63	scenarios	scenario	NOUN
fcis-31632	31	64	with	with	ADP
fcis-31632	31	65	limited	limited	ADJ
fcis-31632	31	66	resources	resource	NOUN
fcis-31632	31	67	.	.	PUNCT
fcis-31632	32	1	the	the	DET
fcis-31632	32	2	training	training	NOUN
fcis-31632	32	3	layer	layer	NOUN
fcis-31632	32	4	is	be	AUX
fcis-31632	32	5	based	base	VERB
fcis-31632	32	6	on	on	ADP
fcis-31632	32	7	the	the	DET
fcis-31632	32	8	meta	meta	ADJ
fcis-31632	32	9	-	-	PUNCT
fcis-31632	32	10	adaptive	adaptive	ADJ
fcis-31632	32	11	multitask	multitask	NOUN
fcis-31632	32	12	learning	learning	NOUN
fcis-31632	32	13	(	(	PUNCT
fcis-31632	32	14	mamtl	mamtl	NOUN
fcis-31632	32	15	)	)	PUNCT
fcis-31632	32	16	method	method	NOUN
fcis-31632	32	17	,	,	PUNCT
fcis-31632	32	18	which	which	PRON
fcis-31632	32	19	can	can	AUX
fcis-31632	32	20	improve	improve	VERB
fcis-31632	32	21	the	the	DET
fcis-31632	32	22	generalization	generalization	NOUN
fcis-31632	32	23	ability	ability	NOUN
fcis-31632	32	24	of	of	ADP
fcis-31632	32	25	the	the	DET
fcis-31632	32	26	model	model	NOUN
fcis-31632	32	27	in	in	ADP
fcis-31632	32	28	cross	cross	NOUN
fcis-31632	32	29	-	-	NOUN
fcis-31632	32	30	language	language	NOUN
fcis-31632	32	31	and	and	CCONJ
fcis-31632	32	32	cross	cross	ADJ
fcis-31632	32	33	-	-	ADJ
fcis-31632	32	34	domain	domain	ADJ
fcis-31632	32	35	scenarios	scenario	NOUN
fcis-31632	32	36	[	[	X
fcis-31632	32	37	8	8	NUM
fcis-31632	32	38	]	]	PUNCT
fcis-31632	32	39	.	.	PUNCT
fcis-31632	33	1	through	through	ADP
fcis-31632	33	2	the	the	DET
fcis-31632	33	3	task	task	NOUN
fcis-31632	33	4	routing	routing	NOUN
fcis-31632	33	5	mechanism	mechanism	NOUN
fcis-31632	33	6	,	,	PUNCT
fcis-31632	33	7	multiple	multiple	ADJ
fcis-31632	33	8	tasks	task	NOUN
fcis-31632	33	9	are	be	AUX
fcis-31632	33	10	divided	divide	VERB
fcis-31632	33	11	into	into	ADP
fcis-31632	33	12	k	k	PROPN
fcis-31632	33	13	clusters	cluster	NOUN
fcis-31632	33	14	by	by	ADP
fcis-31632	33	15	domain	domain	NOUN
fcis-31632	33	16	,	,	PUNCT
fcis-31632	33	17	and	and	CCONJ
fcis-31632	33	18	the	the	DET
fcis-31632	33	19	organizational	organizational	ADJ
fcis-31632	33	20	management	management	NOUN
fcis-31632	33	21	of	of	ADP
fcis-31632	33	22	task	task	NOUN
fcis-31632	33	23	structure	structure	NOUN
fcis-31632	33	24	is	be	AUX
fcis-31632	33	25	realized	realize	VERB
fcis-31632	33	26	.	.	PUNCT
fcis-31632	34	1	on	on	ADP
fcis-31632	34	2	this	this	DET
fcis-31632	34	3	basis	basis	NOUN
fcis-31632	34	4	,	,	PUNCT
fcis-31632	34	5	an	an	DET
fcis-31632	34	6	adaptive	adaptive	ADJ
fcis-31632	34	7	weight	weight	NOUN
fcis-31632	34	8	distribution	distribution	NOUN
fcis-31632	34	9	mechanism	mechanism	NOUN
fcis-31632	34	10	is	be	AUX
fcis-31632	34	11	introduced	introduce	VERB
fcis-31632	34	12	to	to	PART
fcis-31632	34	13	dynamically	dynamically	ADV
fcis-31632	34	14	adjust	adjust	VERB
fcis-31632	34	15	the	the	DET
fcis-31632	34	16	loss	loss	NOUN
fcis-31632	34	17	weight	weight	NOUN
fcis-31632	34	18	of	of	ADP
fcis-31632	34	19	each	each	DET
fcis-31632	34	20	task	task	NOUN
fcis-31632	34	21	in	in	ADP
fcis-31632	34	22	joint	joint	ADJ
fcis-31632	34	23	training	training	NOUN
fcis-31632	34	24	according	accord	VERB
fcis-31632	34	25	to	to	ADP
fcis-31632	34	26	the	the	DET
fcis-31632	34	27	semantic	semantic	ADJ
fcis-31632	34	28	similarity	similarity	NOUN
fcis-31632	34	29	between	between	ADP
fcis-31632	34	30	the	the	DET
fcis-31632	34	31	current	current	ADJ
fcis-31632	34	32	target	target	NOUN
fcis-31632	34	33	task	task	NOUN
fcis-31632	34	34	and	and	CCONJ
fcis-31632	34	35	the	the	DET
fcis-31632	34	36	target	target	NOUN
fcis-31632	34	37	domain	domain	NOUN
fcis-31632	34	38	,	,	PUNCT
fcis-31632	34	39	so	so	SCONJ
fcis-31632	34	40	that	that	SCONJ
fcis-31632	34	41	the	the	DET
fcis-31632	34	42	model	model	NOUN
fcis-31632	34	43	pays	pay	VERB
fcis-31632	34	44	more	more	ADJ
fcis-31632	34	45	attention	attention	NOUN
fcis-31632	34	46	to	to	ADP
fcis-31632	34	47	the	the	DET
fcis-31632	34	48	tasks	task	NOUN
fcis-31632	34	49	with	with	ADP
fcis-31632	34	50	high	high	ADJ
fcis-31632	34	51	correlation	correlation	NOUN
fcis-31632	34	52	,	,	PUNCT
fcis-31632	34	53	thus	thus	ADV
fcis-31632	34	54	enhancing	enhance	VERB
fcis-31632	34	55	the	the	DET
fcis-31632	34	56	effect	effect	NOUN
fcis-31632	34	57	of	of	ADP
fcis-31632	34	58	knowledge	knowledge	NOUN
fcis-31632	34	59	transfer	transfer	NOUN
fcis-31632	34	60	.	.	PUNCT
fcis-31632	35	1	specifically	specifically	ADV
fcis-31632	35	2	,	,	PUNCT
fcis-31632	35	3	the	the	DET
fcis-31632	35	4	total	total	ADJ
fcis-31632	35	5	loss	loss	NOUN
fcis-31632	35	6	function	function	NOUN
fcis-31632	35	7	is	be	AUX
fcis-31632	35	8	defined	define	VERB
fcis-31632	35	9	as	as	ADP
fcis-31632	35	10	:	:	PUNCT
fcis-31632	35	11			NOUN
fcis-31632	35	12			PUNCT
fcis-31632	35	13			PART
fcis-31632	35	14			NUM
fcis-31632	35	15			PROPN
fcis-31632	35	16	βsim	βsim	NOUN
fcis-31632	35	17	,	,	PUNCT
fcis-31632	35	18	ddβsim	ddβsim	NOUN
fcis-31632	36	1	i	i	PRON
fcis-31632	36	2	k	k	PROPN
fcis-31632	37	1	i	i	PRON
fcis-31632	37	2	taskitotal	taskitotal	VERB
fcis-31632	37	3	e	e	X
fcis-31632	37	4	e	e	PROPN
fcis-31632	37	5	,	,	PUNCT
fcis-31632	37	6	αlαl	αlαl	PROPN
fcis-31632	37	7	targeti	targeti	PROPN
fcis-31632	38	1	i	i	PRON
fcis-31632	38	2	1	1	NUM
fcis-31632	38	3	(	(	PUNCT
fcis-31632	38	4	3	3	NUM
fcis-31632	38	5	)	)	PUNCT
fcis-31632	38	6	among	among	ADP
fcis-31632	38	7	them	they	PRON
fcis-31632	38	8	,	,	PUNCT
fcis-31632	38	9	the	the	DET
fcis-31632	38	10	weight	weight	NOUN
fcis-31632	38	11	iα	iα	NOUN
fcis-31632	38	12	is	be	AUX
fcis-31632	38	13	calculated	calculate	VERB
fcis-31632	38	14	by	by	ADP
fcis-31632	38	15	the	the	DET
fcis-31632	38	16	softmax	softmax	NOUN
fcis-31632	38	17	mechanism	mechanism	NOUN
fcis-31632	38	18	,	,	PUNCT
fcis-31632	38	19	which	which	PRON
fcis-31632	38	20	depends	depend	VERB
fcis-31632	38	21	on	on	ADP
fcis-31632	38	22	the	the	DET
fcis-31632	38	23	semantic	semantic	ADJ
fcis-31632	38	24	similarity	similarity	NOUN
fcis-31632	38	25			PROPN
fcis-31632	38	26	sim	sim	PUNCT
fcis-31632	38	27	between	between	ADP
fcis-31632	38	28	the	the	DET
fcis-31632	38	29	source	source	NOUN
fcis-31632	38	30	domain	domain	NOUN
fcis-31632	38	31	i	i	PROPN
fcis-31632	38	32	d	d	PROPN
fcis-31632	38	33	and	and	CCONJ
fcis-31632	38	34	the	the	DET
fcis-31632	38	35	target	target	NOUN
fcis-31632	38	36	domain	domain	NOUN
fcis-31632	38	37	targetd	targetd	NOUN
fcis-31632	38	38	(	(	PUNCT
fcis-31632	38	39	calculated	calculate	VERB
fcis-31632	38	40	by	by	ADP
fcis-31632	38	41	sbert	sbert	NOUN
fcis-31632	38	42	)	)	PUNCT
fcis-31632	38	43	and	and	CCONJ
fcis-31632	38	44	the	the	DET
fcis-31632	38	45	learnable	learnable	ADJ
fcis-31632	38	46	adjustment	adjustment	NOUN
fcis-31632	38	47	factor	factor	NOUN
fcis-31632	38	48			PROPN
fcis-31632	38	49	.	.	PUNCT
fcis-31632	39	1	furthermore	furthermore	ADV
fcis-31632	39	2	,	,	PUNCT
fcis-31632	39	3	the	the	DET
fcis-31632	39	4	meta	meta	ADJ
fcis-31632	39	5	-	-	PUNCT
fcis-31632	39	6	learning	learn	VERB
fcis-31632	39	7	strategy	strategy	NOUN
fcis-31632	39	8	is	be	AUX
fcis-31632	39	9	adopted	adopt	VERB
fcis-31632	39	10	,	,	PUNCT
fcis-31632	39	11	and	and	CCONJ
fcis-31632	39	12	through	through	ADP
fcis-31632	39	13	the	the	DET
fcis-31632	39	14	performance	performance	NOUN
fcis-31632	39	15	feedback	feedback	NOUN
fcis-31632	39	16	on	on	ADP
fcis-31632	39	17	the	the	DET
fcis-31632	39	18	verification	verification	NOUN
fcis-31632	39	19	set	set	NOUN
fcis-31632	39	20	,	,	PUNCT
fcis-31632	39	21	the	the	DET
fcis-31632	39	22	meta	meta	ADJ
fcis-31632	39	23	-	-	PUNCT
fcis-31632	39	24	gradient	gradient	NOUN
fcis-31632	39	25	update	update	NOUN
fcis-31632	39	26	mechanism	mechanism	NOUN
fcis-31632	39	27	is	be	AUX
fcis-31632	39	28	used	use	VERB
fcis-31632	39	29	to	to	PART
fcis-31632	39	30	optimize	optimize	VERB
fcis-31632	39	31			PROPN
fcis-31632	39	32	,	,	PUNCT
fcis-31632	39	33	so	so	SCONJ
fcis-31632	39	34	as	as	SCONJ
fcis-31632	39	35	to	to	PART
fcis-31632	39	36	realize	realize	VERB
fcis-31632	39	37	the	the	DET
fcis-31632	39	38	dynamic	dynamic	ADJ
fcis-31632	39	39	and	and	CCONJ
fcis-31632	39	40	global	global	ADJ
fcis-31632	39	41	regulation	regulation	NOUN
fcis-31632	39	42	of	of	ADP
fcis-31632	39	43	task	task	NOUN
fcis-31632	39	44	weight	weight	NOUN
fcis-31632	39	45	,	,	PUNCT
fcis-31632	39	46	which	which	PRON
fcis-31632	39	47	makes	make	VERB
fcis-31632	39	48	the	the	DET
fcis-31632	39	49	model	model	NOUN
fcis-31632	39	50	have	have	VERB
fcis-31632	39	51	stronger	strong	ADJ
fcis-31632	39	52	adaptive	adaptive	ADJ
fcis-31632	39	53	ability	ability	NOUN
fcis-31632	39	54	and	and	CCONJ
fcis-31632	39	55	cross	cross	ADJ
fcis-31632	39	56	-	-	ADJ
fcis-31632	39	57	domain	domain	ADJ
fcis-31632	39	58	generalization	generalization	NOUN
fcis-31632	39	59	performance	performance	NOUN
fcis-31632	39	60	.	.	PUNCT
fcis-31632	40	1	in	in	ADP
fcis-31632	40	2	this	this	DET
fcis-31632	40	3	study	study	NOUN
fcis-31632	40	4	,	,	PUNCT
fcis-31632	40	5	an	an	DET
fcis-31632	40	6	optimal	optimal	ADJ
fcis-31632	40	7	transport	transport	NOUN
fcis-31632	40	8	alignment	alignment	NOUN
fcis-31632	40	9	(	(	PUNCT
fcis-31632	40	10	ota	ota	PROPN
fcis-31632	40	11	)	)	PUNCT
fcis-31632	40	12	method	method	NOUN
fcis-31632	40	13	based	base	VERB
fcis-31632	40	14	on	on	ADP
fcis-31632	40	15	optimal	optimal	ADJ
fcis-31632	40	16	transmission	transmission	NOUN
fcis-31632	40	17	is	be	AUX
fcis-31632	40	18	proposed	propose	VERB
fcis-31632	40	19	at	at	ADP
fcis-31632	40	20	the	the	DET
fcis-31632	40	21	deployment	deployment	NOUN
fcis-31632	40	22	layer	layer	NOUN
fcis-31632	40	23	,	,	PUNCT
fcis-31632	40	24	aiming	aim	VERB
fcis-31632	40	25	at	at	ADP
fcis-31632	40	26	the	the	DET
fcis-31632	40	27	efficient	efficient	ADJ
fcis-31632	40	28	integration	integration	NOUN
fcis-31632	40	29	of	of	ADP
fcis-31632	40	30	multimodal	multimodal	NOUN
fcis-31632	40	31	semantics	semantic	NOUN
fcis-31632	40	32	such	such	ADJ
fcis-31632	40	33	as	as	ADP
fcis-31632	40	34	text	text	NOUN
fcis-31632	40	35	,	,	PUNCT
fcis-31632	40	36	image	image	NOUN
fcis-31632	40	37	and	and	CCONJ
fcis-31632	40	38	voice	voice	NOUN
fcis-31632	41	1	[	[	X
fcis-31632	41	2	9	9	NUM
fcis-31632	41	3	]	]	PUNCT
fcis-31632	41	4	.	.	PUNCT
fcis-31632	42	1	ota	ota	PROPN
fcis-31632	42	2	models	model	VERB
fcis-31632	42	3	the	the	DET
fcis-31632	42	4	modal	modal	ADJ
fcis-31632	42	5	alignment	alignment	NOUN
fcis-31632	42	6	problem	problem	NOUN
fcis-31632	42	7	as	as	ADP
fcis-31632	42	8	the	the	DET
fcis-31632	42	9	optimal	optimal	ADJ
fcis-31632	42	10	transmission	transmission	NOUN
fcis-31632	42	11	problem	problem	NOUN
fcis-31632	42	12	between	between	ADP
fcis-31632	42	13	probability	probability	NOUN
fcis-31632	42	14	distributions	distribution	NOUN
fcis-31632	42	15	,	,	PUNCT
fcis-31632	42	16	and	and	CCONJ
fcis-31632	42	17	realizes	realize	VERB
fcis-31632	42	18	cross	cross	ADJ
fcis-31632	42	19	-	-	ADJ
fcis-31632	42	20	modal	modal	ADJ
fcis-31632	42	21	semantic	semantic	ADJ
fcis-31632	42	22	matching	matching	NOUN
fcis-31632	42	23	by	by	ADP
fcis-31632	42	24	minimizing	minimize	VERB
fcis-31632	42	25	the	the	DET
fcis-31632	42	26	following	follow	VERB
fcis-31632	42	27	loss	loss	NOUN
fcis-31632	42	28	functions	function	NOUN
fcis-31632	42	29	:	:	PUNCT
fcis-31632	43	1			NOUN
fcis-31632	43	2			SYM
fcis-31632	43	3			NOUN
fcis-31632	43	4			SYM
fcis-31632	43	5			NOUN
fcis-31632	43	6			PUNCT
fcis-31632	43	7			NOUN
fcis-31632	43	8			NOUN
fcis-31632	43	9			PUNCT
fcis-31632	43	10			ADP
fcis-31632	43	11			PROPN
fcis-31632	43	12			PROPN
fcis-31632	43	13			PRON
fcis-31632	43	14			NUM
fcis-31632	43	15	otvftfcel	otvftfcel	VERB
fcis-31632	43	16	vtvt	vtvt	PROPN
fcis-31632	43	17	pp	pp	PROPN
fcis-31632	43	18	ota	ota	PROPN
fcis-31632	43	19	vt	vt	PROPN
fcis-31632	43	20			ADV
fcis-31632	43	21			PROPN
fcis-31632	43	22			PROPN
fcis-31632	43	23			NOUN
fcis-31632	43	24	,	,	PUNCT
fcis-31632	43	25	inf	inf	PROPN
fcis-31632	43	26	~	~	PUNCT
fcis-31632	43	27	,	,	PUNCT
fcis-31632	43	28	,	,	PUNCT
fcis-31632	43	29	(	(	PUNCT
fcis-31632	43	30	4	4	X
fcis-31632	43	31	)	)	PUNCT
fcis-31632	43	32	72	72	NUM
fcis-31632	43	33	among	among	ADP
fcis-31632	43	34	them	they	PRON
fcis-31632	43	35	,	,	PUNCT
fcis-31632	43	36	vt	vt	PROPN
fcis-31632	43	37	ff	ff	PROPN
fcis-31632	43	38	,	,	PUNCT
fcis-31632	43	39	is	be	AUX
fcis-31632	43	40	the	the	DET
fcis-31632	43	41	encoder	encoder	NOUN
fcis-31632	43	42	of	of	ADP
fcis-31632	43	43	text	text	NOUN
fcis-31632	43	44	and	and	CCONJ
fcis-31632	43	45	visual	visual	ADJ
fcis-31632	43	46	mode	mode	NOUN
fcis-31632	43	47	respectively	respectively	ADV
fcis-31632	43	48	,	,	PUNCT
fcis-31632	43	49	and	and	CCONJ
fcis-31632	43	50			PROPN
fcis-31632	43	51	c	c	PROPN
fcis-31632	43	52	uses	use	VERB
fcis-31632	43	53	cosine	cosine	NOUN
fcis-31632	43	54	similarity	similarity	NOUN
fcis-31632	43	55	as	as	SCONJ
fcis-31632	43	56	the	the	DET
fcis-31632	43	57	cost	cost	NOUN
fcis-31632	43	58	function	function	NOUN
fcis-31632	43	59	to	to	PART
fcis-31632	43	60	measure	measure	VERB
fcis-31632	43	61	the	the	DET
fcis-31632	43	62	difference	difference	NOUN
fcis-31632	43	63	of	of	ADP
fcis-31632	43	64	cross	cross	ADJ
fcis-31632	43	65	-	-	ADJ
fcis-31632	43	66	modal	modal	ADJ
fcis-31632	43	67	representation	representation	NOUN
fcis-31632	43	68	;	;	PUNCT
fcis-31632	43	69			PROPN
fcis-31632	43	70			NUM
fcis-31632	43	71	vt	vt	PROPN
fcis-31632	43	72	pp	pp	ADV
fcis-31632	43	73	,	,	PUNCT
fcis-31632	43	74	represents	represent	VERB
fcis-31632	43	75	the	the	DET
fcis-31632	43	76	joint	joint	ADJ
fcis-31632	43	77	distribution	distribution	NOUN
fcis-31632	43	78	set	set	VERB
fcis-31632	43	79	between	between	ADP
fcis-31632	43	80	text	text	NOUN
fcis-31632	43	81	and	and	CCONJ
fcis-31632	43	82	visual	visual	ADJ
fcis-31632	43	83	feature	feature	NOUN
fcis-31632	43	84	distribution	distribution	NOUN
fcis-31632	43	85	.	.	PUNCT
fcis-31632	44	1	the	the	DET
fcis-31632	44	2	second	second	ADJ
fcis-31632	44	3	term	term	NOUN
fcis-31632	44	4	in	in	ADP
fcis-31632	44	5	the	the	DET
fcis-31632	44	6	loss	loss	NOUN
fcis-31632	44	7	function	function	NOUN
fcis-31632	44	8	is	be	AUX
fcis-31632	44	9	sinkhorn	sinkhorn	VERB
fcis-31632	44	10	regularization	regularization	NOUN
fcis-31632	44	11	term	term	NOUN
fcis-31632	44	12			NOUN
fcis-31632	44	13	ot	ot	PUNCT
fcis-31632	44	14	(	(	PUNCT
fcis-31632	44	15	regularization	regularization	NOUN
fcis-31632	44	16	coefficient	coefficient	NOUN
fcis-31632	44	17	1.0	1.0	NUM
fcis-31632	44	18	)	)	PUNCT
fcis-31632	44	19	,	,	PUNCT
fcis-31632	44	20	which	which	PRON
fcis-31632	44	21	is	be	AUX
fcis-31632	44	22	used	use	VERB
fcis-31632	44	23	to	to	PART
fcis-31632	44	24	improve	improve	VERB
fcis-31632	44	25	the	the	DET
fcis-31632	44	26	computational	computational	ADJ
fcis-31632	44	27	stability	stability	NOUN
fcis-31632	44	28	and	and	CCONJ
fcis-31632	44	29	efficiency	efficiency	NOUN
fcis-31632	44	30	of	of	ADP
fcis-31632	44	31	the	the	DET
fcis-31632	44	32	optimal	optimal	ADJ
fcis-31632	44	33	transmission	transmission	NOUN
fcis-31632	44	34	problem	problem	NOUN
fcis-31632	44	35	.	.	PUNCT
fcis-31632	45	1	in	in	ADP
fcis-31632	45	2	the	the	DET
fcis-31632	45	3	actual	actual	ADJ
fcis-31632	45	4	deployment	deployment	NOUN
fcis-31632	45	5	,	,	PUNCT
fcis-31632	45	6	in	in	ADP
fcis-31632	45	7	order	order	NOUN
fcis-31632	45	8	to	to	PART
fcis-31632	45	9	further	far	ADV
fcis-31632	45	10	reduce	reduce	VERB
fcis-31632	45	11	the	the	DET
fcis-31632	45	12	reasoning	reasoning	NOUN
fcis-31632	45	13	delay	delay	NOUN
fcis-31632	45	14	,	,	PUNCT
fcis-31632	45	15	a	a	DET
fcis-31632	45	16	cross	cross	ADJ
fcis-31632	45	17	-	-	ADJ
fcis-31632	45	18	modal	modal	ADJ
fcis-31632	45	19	attention	attention	NOUN
fcis-31632	45	20	caching	cache	VERB
fcis-31632	45	21	mechanism	mechanism	NOUN
fcis-31632	45	22	is	be	AUX
fcis-31632	45	23	introduced	introduce	VERB
fcis-31632	45	24	:	:	PUNCT
fcis-31632	45	25	the	the	DET
fcis-31632	45	26	cross	cross	ADJ
fcis-31632	45	27	-	-	ADJ
fcis-31632	45	28	modal	modal	ADJ
fcis-31632	45	29	attention	attention	NOUN
fcis-31632	45	30	weight	weight	NOUN
fcis-31632	45	31	of	of	ADP
fcis-31632	45	32	previous	previous	ADJ
fcis-31632	45	33	time	time	NOUN
fcis-31632	45	34	steps	step	NOUN
fcis-31632	45	35	is	be	AUX
fcis-31632	45	36	cached	cache	VERB
fcis-31632	45	37	in	in	ADP
fcis-31632	45	38	the	the	DET
fcis-31632	45	39	sequence	sequence	NOUN
fcis-31632	45	40	generation	generation	NOUN
fcis-31632	45	41	process	process	NOUN
fcis-31632	45	42	,	,	PUNCT
fcis-31632	45	43	so	so	SCONJ
fcis-31632	45	44	that	that	SCONJ
fcis-31632	45	45	repeated	repeat	VERB
fcis-31632	45	46	calculation	calculation	NOUN
fcis-31632	45	47	is	be	AUX
fcis-31632	45	48	reduced	reduce	VERB
fcis-31632	45	49	,	,	PUNCT
fcis-31632	45	50	and	and	CCONJ
fcis-31632	45	51	the	the	DET
fcis-31632	45	52	reasoning	reasoning	NOUN
fcis-31632	45	53	speed	speed	NOUN
fcis-31632	45	54	of	of	ADP
fcis-31632	45	55	multi	multi	ADJ
fcis-31632	45	56	-	-	ADJ
fcis-31632	45	57	modal	modal	ADJ
fcis-31632	45	58	tasks	task	NOUN
fcis-31632	45	59	is	be	AUX
fcis-31632	45	60	significantly	significantly	ADV
fcis-31632	45	61	improved	improve	VERB
fcis-31632	45	62	,	,	PUNCT
fcis-31632	45	63	thus	thus	ADV
fcis-31632	45	64	achieving	achieve	VERB
fcis-31632	45	65	accurate	accurate	ADJ
fcis-31632	45	66	and	and	CCONJ
fcis-31632	45	67	efficient	efficient	ADJ
fcis-31632	45	68	multi	multi	ADJ
fcis-31632	45	69	-	-	ADJ
fcis-31632	45	70	modal	modal	ADJ
fcis-31632	45	71	semantic	semantic	ADJ
fcis-31632	45	72	alignment	alignment	NOUN
fcis-31632	45	73	and	and	CCONJ
fcis-31632	45	74	real	real	ADJ
fcis-31632	45	75	-	-	PUNCT
fcis-31632	45	76	time	time	NOUN
fcis-31632	45	77	response	response	NOUN
fcis-31632	45	78	.	.	PUNCT
fcis-31632	46	1	3	3	X
fcis-31632	46	2	.	.	X
fcis-31632	46	3	experiment	experiment	NOUN
fcis-31632	46	4	and	and	CCONJ
fcis-31632	46	5	analysis	analysis	NOUN
fcis-31632	46	6	flores-200	flores-200	NOUN
fcis-31632	46	7	data	datum	NOUN
fcis-31632	46	8	set	set	VERB
fcis-31632	46	9	is	be	AUX
fcis-31632	46	10	used	use	VERB
fcis-31632	46	11	to	to	PART
fcis-31632	46	12	evaluate	evaluate	VERB
fcis-31632	46	13	the	the	DET
fcis-31632	46	14	performance	performance	NOUN
fcis-31632	46	15	of	of	ADP
fcis-31632	46	16	low	low	ADJ
fcis-31632	46	17	-	-	PUNCT
fcis-31632	46	18	resource	resource	NOUN
fcis-31632	46	19	translation	translation	NOUN
fcis-31632	46	20	,	,	PUNCT
fcis-31632	46	21	x	x	ADJ
fcis-31632	46	22	-	-	ADJ
fcis-31632	46	23	cross	cross	ADJ
fcis-31632	46	24	data	datum	NOUN
fcis-31632	46	25	set	set	VERB
fcis-31632	46	26	covering	cover	VERB
fcis-31632	46	27	eight	eight	NUM
fcis-31632	46	28	fields	field	NOUN
fcis-31632	46	29	,	,	PUNCT
fcis-31632	46	30	such	such	ADJ
fcis-31632	46	31	as	as	ADP
fcis-31632	46	32	medical	medical	ADJ
fcis-31632	46	33	care	care	NOUN
fcis-31632	46	34	,	,	PUNCT
fcis-31632	46	35	finance	finance	NOUN
fcis-31632	46	36	and	and	CCONJ
fcis-31632	46	37	law	law	NOUN
fcis-31632	46	38	,	,	PUNCT
fcis-31632	46	39	is	be	AUX
fcis-31632	46	40	used	use	VERB
fcis-31632	46	41	to	to	PART
fcis-31632	46	42	verify	verify	VERB
fcis-31632	46	43	the	the	DET
fcis-31632	46	44	cross	cross	ADJ
fcis-31632	46	45	-	-	ADJ
fcis-31632	46	46	domain	domain	ADJ
fcis-31632	46	47	generalization	generalization	NOUN
fcis-31632	46	48	ability	ability	NOUN
fcis-31632	46	49	,	,	PUNCT
fcis-31632	46	50	and	and	CCONJ
fcis-31632	46	51	coco	coco	PROPN
fcis-31632	46	52	-	-	PUNCT
fcis-31632	46	53	captions	caption	NOUN
fcis-31632	46	54	(	(	PUNCT
fcis-31632	46	55	image	image	NOUN
fcis-31632	46	56	-	-	PUNCT
fcis-31632	46	57	text	text	NOUN
fcis-31632	46	58	)	)	PUNCT
fcis-31632	46	59	and	and	CCONJ
fcis-31632	46	60	librispeech	librispeech	PROPN
fcis-31632	46	61	(	(	PUNCT
fcis-31632	46	62	voice	voice	NOUN
fcis-31632	46	63	)	)	PUNCT
fcis-31632	46	64	data	datum	NOUN
fcis-31632	46	65	set	set	VERB
fcis-31632	46	66	are	be	AUX
fcis-31632	46	67	combined	combine	VERB
fcis-31632	46	68	to	to	PART
fcis-31632	46	69	test	test	VERB
fcis-31632	46	70	the	the	DET
fcis-31632	46	71	multi	multi	ADJ
fcis-31632	46	72	-	-	ADJ
fcis-31632	46	73	modal	modal	ADJ
fcis-31632	46	74	task	task	NOUN
fcis-31632	46	75	performance	performance	NOUN
fcis-31632	46	76	.	.	PUNCT
fcis-31632	47	1	the	the	DET
fcis-31632	47	2	model	model	NOUN
fcis-31632	47	3	was	be	AUX
fcis-31632	47	4	trained	train	VERB
fcis-31632	47	5	on	on	ADP
fcis-31632	47	6	nvidia	nvidia	PROPN
fcis-31632	47	7	v100	v100	PROPN
fcis-31632	47	8	cloud	cloud	NOUN
fcis-31632	47	9	server	server	NOUN
fcis-31632	47	10	,	,	PUNCT
fcis-31632	47	11	and	and	CCONJ
fcis-31632	47	12	the	the	DET
fcis-31632	47	13	reasoning	reasoning	NOUN
fcis-31632	47	14	efficiency	efficiency	NOUN
fcis-31632	47	15	was	be	AUX
fcis-31632	47	16	tested	test	VERB
fcis-31632	47	17	on	on	ADP
fcis-31632	47	18	jetson	jetson	PROPN
fcis-31632	47	19	xavier	xavier	PROPN
fcis-31632	47	20	nx	nx	PROPN
fcis-31632	47	21	edge	edge	PROPN
fcis-31632	47	22	device	device	NOUN
fcis-31632	47	23	.	.	PUNCT
fcis-31632	48	1	the	the	DET
fcis-31632	48	2	experimental	experimental	ADJ
fcis-31632	48	3	results	result	NOUN
fcis-31632	48	4	show	show	VERB
fcis-31632	48	5	that	that	SCONJ
fcis-31632	48	6	the	the	DET
fcis-31632	48	7	proposed	propose	VERB
fcis-31632	48	8	hof	hof	NOUN
fcis-31632	48	9	is	be	AUX
fcis-31632	48	10	significantly	significantly	ADV
fcis-31632	48	11	superior	superior	ADJ
fcis-31632	48	12	to	to	ADP
fcis-31632	48	13	the	the	DET
fcis-31632	48	14	baseline	baseline	NOUN
fcis-31632	48	15	model	model	NOUN
fcis-31632	48	16	in	in	ADP
fcis-31632	48	17	all	all	DET
fcis-31632	48	18	indicators	indicator	NOUN
fcis-31632	48	19	.	.	PUNCT
fcis-31632	49	1	as	as	SCONJ
fcis-31632	49	2	shown	show	VERB
fcis-31632	49	3	in	in	ADP
fcis-31632	49	4	table	table	NOUN
fcis-31632	49	5	1	1	NUM
fcis-31632	49	6	,	,	PUNCT
fcis-31632	49	7	on	on	ADP
fcis-31632	49	8	flores-200	flores-200	NOUN
fcis-31632	49	9	,	,	PUNCT
fcis-31632	49	10	the	the	DET
fcis-31632	49	11	bleu	bleu	NOUN
fcis-31632	49	12	reaches	reach	VERB
fcis-31632	49	13	47.5	47.5	NUM
fcis-31632	49	14	,	,	PUNCT
fcis-31632	49	15	on	on	ADP
fcis-31632	49	16	x	x	NOUN
fcis-31632	49	17	-	-	NOUN
fcis-31632	49	18	cross	cross	ADJ
fcis-31632	49	19	,	,	PUNCT
fcis-31632	49	20	the	the	DET
fcis-31632	49	21	macro	macro	ADJ
fcis-31632	49	22	f1	f1	NOUN
fcis-31632	49	23	reaches	reach	VERB
fcis-31632	49	24	84.7	84.7	NUM
fcis-31632	49	25	,	,	PUNCT
fcis-31632	49	26	the	the	DET
fcis-31632	49	27	reasoning	reasoning	NOUN
fcis-31632	49	28	delay	delay	NOUN
fcis-31632	49	29	of	of	ADP
fcis-31632	49	30	edge	edge	NOUN
fcis-31632	49	31	devices	device	NOUN
fcis-31632	49	32	is	be	AUX
fcis-31632	49	33	reduced	reduce	VERB
fcis-31632	49	34	to	to	ADP
fcis-31632	49	35	62ms	62ms	NOUN
fcis-31632	49	36	,	,	PUNCT
fcis-31632	49	37	and	and	CCONJ
fcis-31632	49	38	the	the	DET
fcis-31632	49	39	energy	energy	NOUN
fcis-31632	49	40	consumption	consumption	NOUN
fcis-31632	49	41	is	be	AUX
fcis-31632	49	42	reduced	reduce	VERB
fcis-31632	49	43	to	to	ADP
fcis-31632	49	44	2.9j	2.9j	NUM
fcis-31632	49	45	,	,	PUNCT
fcis-31632	49	46	which	which	PRON
fcis-31632	49	47	verifies	verify	VERB
fcis-31632	49	48	the	the	DET
fcis-31632	49	49	effectiveness	effectiveness	NOUN
fcis-31632	49	50	and	and	CCONJ
fcis-31632	49	51	superiority	superiority	NOUN
fcis-31632	49	52	of	of	ADP
fcis-31632	49	53	collaborative	collaborative	ADJ
fcis-31632	49	54	optimization	optimization	NOUN
fcis-31632	49	55	among	among	ADP
fcis-31632	49	56	data	data	NOUN
fcis-31632	49	57	layer	layer	NOUN
fcis-31632	49	58	(	(	PUNCT
fcis-31632	49	59	aki	aki	NOUN
fcis-31632	49	60	)	)	PUNCT
fcis-31632	49	61	,	,	PUNCT
fcis-31632	49	62	model	model	NOUN
fcis-31632	49	63	layer	layer	NOUN
fcis-31632	49	64	(	(	PUNCT
fcis-31632	49	65	dsgt	dsgt	NOUN
fcis-31632	49	66	)	)	PUNCT
fcis-31632	49	67	and	and	CCONJ
fcis-31632	49	68	training	training	NOUN
fcis-31632	49	69	layer	layer	NOUN
fcis-31632	49	70	(	(	PUNCT
fcis-31632	49	71	mamtl	mamtl	NOUN
fcis-31632	49	72	)	)	PUNCT
fcis-31632	49	73	.	.	PUNCT
fcis-31632	50	1	table	table	NOUN
fcis-31632	50	2	1	1	NUM
fcis-31632	50	3	.	.	PUNCT
fcis-31632	50	4	comparison	comparison	NOUN
fcis-31632	50	5	of	of	ADP
fcis-31632	50	6	performance	performance	NOUN
fcis-31632	50	7	improvement	improvement	NOUN
fcis-31632	50	8	of	of	ADP
fcis-31632	50	9	full	full	ADJ
fcis-31632	50	10	link	link	NOUN
fcis-31632	50	11	method	method	NOUN
fcis-31632	50	12	method	method	NOUN
fcis-31632	50	13	module	module	NOUN
fcis-31632	50	14	bleu	bleu	NOUN
fcis-31632	50	15	(	(	PUNCT
fcis-31632	50	16	flores	flores	PROPN
fcis-31632	50	17	)	)	PUNCT
fcis-31632	50	18	macro	macro	ADJ
fcis-31632	50	19	f1(xcross	f1(xcross	NOUN
fcis-31632	50	20	)	)	PUNCT
fcis-31632	50	21	delay	delay	NOUN
fcis-31632	50	22	(	(	PUNCT
fcis-31632	50	23	ms	ms	NOUN
fcis-31632	50	24	)	)	PUNCT
fcis-31632	50	25	energy	energy	NOUN
fcis-31632	50	26	consumption	consumption	NOUN
fcis-31632	50	27	(	(	PUNCT
fcis-31632	50	28	j	j	NOUN
fcis-31632	50	29	)	)	PUNCT
fcis-31632	50	30	bertlarge	bertlarge	NOUN
fcis-31632	50	31	42.1	42.1	NUM
fcis-31632	50	32	78.3	78.3	NUM
fcis-31632	50	33	210	210	NUM
fcis-31632	50	34	9.8	9.8	NUM
fcis-31632	50	35	+	+	NOUN
fcis-31632	50	36	data	datum	NOUN
fcis-31632	50	37	layer	layer	NOUN
fcis-31632	50	38	(	(	PUNCT
fcis-31632	50	39	aki	aki	NOUN
fcis-31632	50	40	)	)	PUNCT
fcis-31632	50	41	46.3	46.3	NUM
fcis-31632	50	42	(	(	PUNCT
fcis-31632	50	43	+4.2	+4.2	ADJ
fcis-31632	50	44	)	)	PUNCT
fcis-31632	51	1	+	+	ADJ
fcis-31632	51	2	model	model	NOUN
fcis-31632	51	3	layer	layer	NOUN
fcis-31632	51	4	(	(	PUNCT
fcis-31632	51	5	dsgt	dsgt	NOUN
fcis-31632	51	6	)	)	PUNCT
fcis-31632	51	7	45.8	45.8	NUM
fcis-31632	51	8	79.1	79.1	NUM
fcis-31632	51	9	68	68	NUM
fcis-31632	51	10	(	(	PUNCT
fcis-31632	51	11	67.6	67.6	NUM
fcis-31632	51	12	%	%	NOUN
fcis-31632	51	13	)	)	PUNCT
fcis-31632	51	14	3.2	3.2	NUM
fcis-31632	51	15	(	(	PUNCT
fcis-31632	51	16	-67	-67	NUM
fcis-31632	51	17	%	%	NOUN
fcis-31632	51	18	)	)	PUNCT
fcis-31632	52	1	+	+	ADJ
fcis-31632	52	2	training	training	NOUN
fcis-31632	52	3	layer	layer	NOUN
fcis-31632	52	4	(	(	PUNCT
fcis-31632	52	5	mamtl	mamtl	NOUN
fcis-31632	52	6	)	)	PUNCT
fcis-31632	52	7	82.9	82.9	NUM
fcis-31632	52	8	(	(	PUNCT
fcis-31632	52	9	+4.6	+4.6	ADJ
fcis-31632	52	10	)	)	PUNCT
fcis-31632	52	11	full	full	ADJ
fcis-31632	52	12	link	link	NOUN
fcis-31632	52	13	(	(	PUNCT
fcis-31632	52	14	hof	hof	NOUN
fcis-31632	52	15	)	)	PUNCT
fcis-31632	52	16	47.5	47.5	NUM
fcis-31632	52	17	84.7	84.7	NUM
fcis-31632	52	18	62	62	NUM
fcis-31632	52	19	2.9	2.9	NUM
fcis-31632	52	20	the	the	DET
fcis-31632	52	21	social	social	ADJ
fcis-31632	52	22	impact	impact	NOUN
fcis-31632	52	23	of	of	ADP
fcis-31632	52	24	the	the	DET
fcis-31632	52	25	proposed	propose	VERB
fcis-31632	52	26	method	method	NOUN
fcis-31632	52	27	is	be	AUX
fcis-31632	52	28	verified	verify	VERB
fcis-31632	52	29	by	by	ADP
fcis-31632	52	30	the	the	DET
fcis-31632	52	31	fairness	fairness	NOUN
fcis-31632	52	32	test	test	NOUN
fcis-31632	52	33	on	on	ADP
fcis-31632	52	34	biasbench	biasbench	NOUN
fcis-31632	52	35	.	.	PUNCT
fcis-31632	53	1	compared	compare	VERB
fcis-31632	53	2	with	with	ADP
fcis-31632	53	3	bert	bert	NOUN
fcis-31632	53	4	-	-	PUNCT
fcis-31632	53	5	large	large	ADJ
fcis-31632	53	6	,	,	PUNCT
fcis-31632	53	7	the	the	DET
fcis-31632	53	8	hof	hof	NOUN
fcis-31632	53	9	in	in	ADP
fcis-31632	53	10	this	this	DET
fcis-31632	53	11	study	study	NOUN
fcis-31632	53	12	significantly	significantly	ADV
fcis-31632	53	13	reduces	reduce	VERB
fcis-31632	53	14	the	the	DET
fcis-31632	53	15	prediction	prediction	NOUN
fcis-31632	53	16	bias	bias	NOUN
fcis-31632	53	17	of	of	ADP
fcis-31632	53	18	the	the	DET
fcis-31632	53	19	model	model	NOUN
fcis-31632	53	20	in	in	ADP
fcis-31632	53	21	terms	term	NOUN
fcis-31632	53	22	of	of	ADP
fcis-31632	53	23	gender	gender	NOUN
fcis-31632	53	24	and	and	CCONJ
fcis-31632	53	25	race	race	NOUN
fcis-31632	53	26	(	(	PUNCT
fcis-31632	53	27	gender	gender	NOUN
fcis-31632	53	28	bias	bias	NOUN
fcis-31632	53	29	is	be	AUX
fcis-31632	53	30	reduced	reduce	VERB
fcis-31632	53	31	from	from	ADP
fcis-31632	53	32	0.32	0.32	NUM
fcis-31632	53	33	to	to	ADP
fcis-31632	53	34	0.19	0.19	NUM
fcis-31632	53	35	,	,	PUNCT
fcis-31632	53	36	and	and	CCONJ
fcis-31632	53	37	racial	racial	ADJ
fcis-31632	53	38	bias	bias	NOUN
fcis-31632	53	39	is	be	AUX
fcis-31632	53	40	reduced	reduce	VERB
fcis-31632	53	41	from	from	ADP
fcis-31632	53	42	0.41	0.41	NUM
fcis-31632	53	43	to	to	ADP
fcis-31632	53	44	0.23	0.23	NUM
fcis-31632	53	45	)	)	PUNCT
fcis-31632	53	46	,	,	PUNCT
fcis-31632	53	47	which	which	PRON
fcis-31632	53	48	shows	show	VERB
fcis-31632	53	49	that	that	SCONJ
fcis-31632	53	50	it	it	PRON
fcis-31632	53	51	is	be	AUX
fcis-31632	53	52	helpful	helpful	ADJ
fcis-31632	53	53	to	to	PART
fcis-31632	53	54	build	build	VERB
fcis-31632	53	55	a	a	DET
fcis-31632	53	56	fairer	fair	ADJ
fcis-31632	53	57	and	and	CCONJ
fcis-31632	53	58	more	more	ADV
fcis-31632	53	59	credible	credible	ADJ
fcis-31632	53	60	ai	ai	NOUN
fcis-31632	53	61	system	system	NOUN
fcis-31632	53	62	while	while	SCONJ
fcis-31632	53	63	improving	improve	VERB
fcis-31632	53	64	its	its	PRON
fcis-31632	53	65	performance	performance	NOUN
fcis-31632	53	66	(	(	PUNCT
fcis-31632	53	67	as	as	SCONJ
fcis-31632	53	68	shown	show	VERB
fcis-31632	53	69	in	in	ADP
fcis-31632	53	70	table	table	NOUN
fcis-31632	53	71	2	2	NUM
fcis-31632	53	72	)	)	PUNCT
fcis-31632	53	73	.	.	PUNCT
fcis-31632	54	1	table	table	NOUN
fcis-31632	54	2	2	2	NUM
fcis-31632	54	3	.	.	PUNCT
fcis-31632	54	4	fairness	fairness	NOUN
fcis-31632	54	5	test	test	NOUN
fcis-31632	54	6	model	model	NOUN
fcis-31632	54	7	gender	gender	NOUN
fcis-31632	54	8	prejudice	prejudice	NOUN
fcis-31632	54	9	racial	racial	ADJ
fcis-31632	54	10	bias	bias	NOUN
fcis-31632	54	11	bert	bert	NOUN
fcis-31632	54	12	-	-	PUNCT
fcis-31632	54	13	large	large	ADJ
fcis-31632	54	14	0.32	0.32	NUM
fcis-31632	54	15	0.41	0.41	NUM
fcis-31632	54	16	hof	hof	NOUN
fcis-31632	54	17	0.19	0.19	NUM
fcis-31632	54	18	0.23	0.23	NUM
fcis-31632	54	19	4	4	NUM
fcis-31632	54	20	.	.	PUNCT
fcis-31632	54	21	conclusion	conclusion	NOUN
fcis-31632	54	22	in	in	ADP
fcis-31632	54	23	this	this	DET
fcis-31632	54	24	paper	paper	NOUN
fcis-31632	54	25	,	,	PUNCT
fcis-31632	54	26	the	the	DET
fcis-31632	54	27	method	method	NOUN
fcis-31632	54	28	of	of	ADP
fcis-31632	54	29	improving	improve	VERB
fcis-31632	54	30	the	the	DET
fcis-31632	54	31	performance	performance	NOUN
fcis-31632	54	32	of	of	ADP
fcis-31632	54	33	nlp	nlp	NOUN
fcis-31632	54	34	model	model	NOUN
fcis-31632	54	35	driven	drive	VERB
fcis-31632	54	36	by	by	ADP
fcis-31632	54	37	ai	ai	VERB
fcis-31632	54	38	is	be	AUX
fcis-31632	54	39	systematically	systematically	ADV
fcis-31632	54	40	discussed	discuss	VERB
fcis-31632	54	41	,	,	PUNCT
fcis-31632	54	42	and	and	CCONJ
fcis-31632	54	43	a	a	DET
fcis-31632	54	44	hof	hof	NOUN
fcis-31632	54	45	covering	cover	VERB
fcis-31632	54	46	the	the	DET
fcis-31632	54	47	whole	whole	ADJ
fcis-31632	54	48	link	link	NOUN
fcis-31632	54	49	collaborative	collaborative	ADJ
fcis-31632	54	50	design	design	NOUN
fcis-31632	54	51	of	of	ADP
fcis-31632	54	52	data	datum	NOUN
fcis-31632	54	53	,	,	PUNCT
fcis-31632	54	54	model	model	NOUN
fcis-31632	54	55	,	,	PUNCT
fcis-31632	54	56	training	training	NOUN
fcis-31632	54	57	and	and	CCONJ
fcis-31632	54	58	deployment	deployment	NOUN
fcis-31632	54	59	is	be	AUX
fcis-31632	54	60	proposed	propose	VERB
fcis-31632	54	61	.	.	PUNCT
fcis-31632	55	1	through	through	ADP
fcis-31632	55	2	data	data	NOUN
fcis-31632	55	3	enhancement	enhancement	NOUN
fcis-31632	55	4	and	and	CCONJ
fcis-31632	55	5	knowledge	knowledge	NOUN
fcis-31632	55	6	injection	injection	NOUN
fcis-31632	55	7	technology	technology	NOUN
fcis-31632	55	8	in	in	ADP
fcis-31632	55	9	lowresource	lowresource	PROPN
fcis-31632	55	10	scenarios	scenario	NOUN
fcis-31632	55	11	,	,	PUNCT
fcis-31632	55	12	dynamic	dynamic	ADJ
fcis-31632	55	13	sparse	sparse	ADJ
fcis-31632	55	14	compression	compression	NOUN
fcis-31632	55	15	model	model	NOUN
fcis-31632	55	16	architecture	architecture	NOUN
fcis-31632	55	17	,	,	PUNCT
fcis-31632	55	18	adaptive	adaptive	ADJ
fcis-31632	55	19	multi	multi	ADJ
fcis-31632	55	20	-	-	ADJ
fcis-31632	55	21	task	task	ADJ
fcis-31632	55	22	joint	joint	ADJ
fcis-31632	55	23	learning	learning	NOUN
fcis-31632	55	24	strategy	strategy	NOUN
fcis-31632	55	25	,	,	PUNCT
fcis-31632	55	26	and	and	CCONJ
fcis-31632	55	27	multimodal	multimodal	ADJ
fcis-31632	55	28	semantic	semantic	ADJ
fcis-31632	55	29	alignment	alignment	NOUN
fcis-31632	55	30	method	method	NOUN
fcis-31632	55	31	based	base	VERB
fcis-31632	55	32	on	on	ADP
fcis-31632	55	33	optimal	optimal	ADJ
fcis-31632	55	34	transmission	transmission	NOUN
fcis-31632	55	35	,	,	PUNCT
fcis-31632	55	36	hof	hof	NOUN
fcis-31632	55	37	successfully	successfully	ADV
fcis-31632	55	38	improved	improve	VERB
fcis-31632	55	39	the	the	DET
fcis-31632	55	40	performance	performance	NOUN
fcis-31632	55	41	of	of	ADP
fcis-31632	55	42	nlp	nlp	NOUN
fcis-31632	55	43	model	model	NOUN
fcis-31632	55	44	.	.	PUNCT
fcis-31632	56	1	the	the	DET
fcis-31632	56	2	experimental	experimental	ADJ
fcis-31632	56	3	results	result	NOUN
fcis-31632	56	4	show	show	VERB
fcis-31632	56	5	that	that	PRON
fcis-31632	56	6	hof	hof	NOUN
fcis-31632	56	7	achieves	achieve	VERB
fcis-31632	56	8	a	a	DET
fcis-31632	56	9	bleu	bleu	NOUN
fcis-31632	56	10	score	score	NOUN
fcis-31632	56	11	of	of	ADP
fcis-31632	56	12	47.5	47.5	NUM
fcis-31632	56	13	in	in	ADP
fcis-31632	56	14	the	the	DET
fcis-31632	56	15	low	low	ADJ
fcis-31632	56	16	-	-	PUNCT
fcis-31632	56	17	resource	resource	NOUN
fcis-31632	56	18	translation	translation	NOUN
fcis-31632	56	19	task	task	NOUN
fcis-31632	56	20	and	and	CCONJ
fcis-31632	56	21	a	a	DET
fcis-31632	56	22	macro	macro	ADJ
fcis-31632	56	23	f1	f1	NOUN
fcis-31632	56	24	of	of	ADP
fcis-31632	56	25	84.7	84.7	NUM
fcis-31632	56	26	in	in	ADP
fcis-31632	56	27	the	the	DET
fcis-31632	56	28	cross	cross	ADJ
fcis-31632	56	29	-	-	ADJ
fcis-31632	56	30	domain	domain	ADJ
fcis-31632	56	31	generalization	generalization	NOUN
fcis-31632	56	32	task	task	NOUN
fcis-31632	56	33	,	,	PUNCT
fcis-31632	56	34	and	and	CCONJ
fcis-31632	56	35	the	the	DET
fcis-31632	56	36	reasoning	reasoning	NOUN
fcis-31632	56	37	delay	delay	NOUN
fcis-31632	56	38	on	on	ADP
fcis-31632	56	39	the	the	DET
fcis-31632	56	40	edge	edge	NOUN
fcis-31632	56	41	device	device	NOUN
fcis-31632	56	42	is	be	AUX
fcis-31632	56	43	reduced	reduce	VERB
fcis-31632	56	44	to	to	ADP
fcis-31632	56	45	62	62	NUM
fcis-31632	56	46	milliseconds	millisecond	NOUN
fcis-31632	56	47	,	,	PUNCT
fcis-31632	56	48	and	and	CCONJ
fcis-31632	56	49	the	the	DET
fcis-31632	56	50	energy	energy	NOUN
fcis-31632	56	51	consumption	consumption	NOUN
fcis-31632	56	52	is	be	AUX
fcis-31632	56	53	reduced	reduce	VERB
fcis-31632	56	54	to	to	ADP
fcis-31632	56	55	2.9	2.9	NUM
fcis-31632	56	56	joules	joule	NOUN
fcis-31632	56	57	.	.	PUNCT
fcis-31632	57	1	in	in	ADP
fcis-31632	57	2	addition	addition	NOUN
fcis-31632	57	3	,	,	PUNCT
fcis-31632	57	4	hof	hof	PROPN
fcis-31632	57	5	shows	show	VERB
fcis-31632	57	6	the	the	DET
fcis-31632	57	7	ability	ability	NOUN
fcis-31632	57	8	to	to	PART
fcis-31632	57	9	significantly	significantly	ADV
fcis-31632	57	10	reduce	reduce	VERB
fcis-31632	57	11	gender	gender	NOUN
fcis-31632	57	12	and	and	CCONJ
fcis-31632	57	13	racial	racial	ADJ
fcis-31632	57	14	bias	bias	NOUN
fcis-31632	57	15	in	in	ADP
fcis-31632	57	16	fairness	fairness	NOUN
fcis-31632	57	17	test	test	NOUN
fcis-31632	57	18	,	,	PUNCT
fcis-31632	57	19	which	which	PRON
fcis-31632	57	20	is	be	AUX
fcis-31632	57	21	helpful	helpful	ADJ
fcis-31632	57	22	to	to	PART
fcis-31632	57	23	build	build	VERB
fcis-31632	57	24	a	a	DET
fcis-31632	57	25	more	more	ADV
fcis-31632	57	26	credible	credible	ADJ
fcis-31632	57	27	and	and	CCONJ
fcis-31632	57	28	responsible	responsible	ADJ
fcis-31632	57	29	ai	ai	NOUN
fcis-31632	57	30	system	system	NOUN
fcis-31632	57	31	.	.	PUNCT
fcis-31632	58	1	hof	hof	PROPN
fcis-31632	58	2	provides	provide	VERB
fcis-31632	58	3	a	a	DET
fcis-31632	58	4	new	new	ADJ
fcis-31632	58	5	paradigm	paradigm	NOUN
fcis-31632	58	6	for	for	ADP
fcis-31632	58	7	improving	improve	VERB
fcis-31632	58	8	the	the	DET
fcis-31632	58	9	performance	performance	NOUN
fcis-31632	58	10	of	of	ADP
fcis-31632	58	11	nlp	nlp	NOUN
fcis-31632	58	12	model	model	NOUN
fcis-31632	58	13	,	,	PUNCT
fcis-31632	58	14	which	which	PRON
fcis-31632	58	15	is	be	AUX
fcis-31632	58	16	expected	expect	VERB
fcis-31632	58	17	to	to	PART
fcis-31632	58	18	promote	promote	VERB
fcis-31632	58	19	the	the	DET
fcis-31632	58	20	wide	wide	ADJ
fcis-31632	58	21	application	application	NOUN
fcis-31632	58	22	of	of	ADP
fcis-31632	58	23	efficient	efficient	ADJ
fcis-31632	58	24	,	,	PUNCT
fcis-31632	58	25	lightweight	lightweight	ADJ
fcis-31632	58	26	and	and	CCONJ
fcis-31632	58	27	robust	robust	ADJ
fcis-31632	58	28	nlp	nlp	NOUN
fcis-31632	58	29	technology	technology	NOUN
fcis-31632	58	30	in	in	ADP
fcis-31632	58	31	real	real	ADJ
fcis-31632	58	32	scenes	scene	NOUN
fcis-31632	58	33	.	.	PUNCT
fcis-31632	59	1	references	reference	NOUN
fcis-31632	59	2	[	[	X
fcis-31632	59	3	1	1	NUM
fcis-31632	59	4	]	]	PUNCT
fcis-31632	59	5	sangmin	sangmin	NOUN
fcis-31632	59	6	kim	kim	PROPN
fcis-31632	59	7	,	,	PUNCT
fcis-31632	59	8	byeongcheon	byeongcheon	PROPN
fcis-31632	59	9	lee	lee	PROPN
fcis-31632	59	10	,	,	PUNCT
fcis-31632	59	11	muazzam	muazzam	PROPN
fcis-31632	59	12	maqsood	maqsood	PROPN
fcis-31632	59	13	,	,	PUNCT
fcis-31632	59	14	jihoon	jihoon	PROPN
fcis-31632	59	15	moon	moon	PROPN
fcis-31632	59	16	&	&	CCONJ
fcis-31632	59	17	seungmin	seungmin	PROPN
fcis-31632	59	18	rho.(2025	rho.(2025	PROPN
fcis-31632	59	19	)	)	PUNCT
fcis-31632	59	20	.	.	PUNCT
fcis-31632	60	1	deep	deep	ADJ
fcis-31632	60	2	learning	learning	NOUN
fcis-31632	60	3	-	-	PUNCT
fcis-31632	60	4	based	base	VERB
fcis-31632	60	5	natural	natural	ADJ
fcis-31632	60	6	language	language	NOUN
fcis-31632	60	7	processing	processing	NOUN
fcis-31632	60	8	model	model	NOUN
fcis-31632	60	9	and	and	CCONJ
fcis-31632	60	10	optical	optical	ADJ
fcis-31632	60	11	character	character	NOUN
fcis-31632	60	12	recognition	recognition	NOUN
fcis-31632	60	13	for	for	ADP
fcis-31632	60	14	detection	detection	NOUN
fcis-31632	60	15	of	of	ADP
fcis-31632	60	16	online	online	ADJ
fcis-31632	60	17	grooming	groom	VERB
fcis-31632	60	18	on	on	ADP
fcis-31632	60	19	social	social	ADJ
fcis-31632	60	20	networking	network	VERB
fcis-31632	60	21	services.computer	services.computer	ADJ
fcis-31632	60	22	modeling	modeling	NOUN
fcis-31632	60	23	in	in	ADP
fcis-31632	60	24	engineering	engineering	NOUN
fcis-31632	60	25	&	&	CCONJ
fcis-31632	60	26	sciences	sciences	PROPN
fcis-31632	60	27	,	,	PUNCT
fcis-31632	60	28	143(2),2079	143(2),2079	PROPN
fcis-31632	60	29	-	-	PUNCT
fcis-31632	60	30	2108	2108	NUM
fcis-31632	60	31	.	.	PUNCT
fcis-31632	61	1	[	[	X
fcis-31632	61	2	2	2	X
fcis-31632	61	3	]	]	X
fcis-31632	61	4	wang	wang	PROPN
fcis-31632	61	5	dai	dai	PROPN
fcis-31632	61	6	,	,	PUNCT
fcis-31632	61	7	kebiao	kebiao	PROPN
fcis-31632	61	8	mao	mao	PROPN
fcis-31632	61	9	,	,	PUNCT
fcis-31632	61	10	zhonghua	zhonghua	PROPN
fcis-31632	61	11	guo	guo	PROPN
fcis-31632	61	12	,	,	PUNCT
fcis-31632	61	13	zhihao	zhihao	PROPN
fcis-31632	61	14	qin	qin	PROPN
fcis-31632	61	15	,	,	PUNCT
fcis-31632	61	16	jiancheng	jiancheng	PROPN
fcis-31632	61	17	shi	shi	PROPN
fcis-31632	61	18	,	,	PUNCT
fcis-31632	61	19	sayed	say	VERB
fcis-31632	61	20	m.	m.	PROPN
fcis-31632	61	21	bateni	bateni	PROPN
fcis-31632	61	22	&	&	CCONJ
fcis-31632	61	23	liurui	liurui	PROPN
fcis-31632	61	24	xiao	xiao	PROPN
fcis-31632	61	25	.	.	PUNCT
fcis-31632	61	26	(	(	PUNCT
fcis-31632	61	27	2025	2025	NUM
fcis-31632	61	28	)	)	PUNCT
fcis-31632	61	29	.	.	PUNCT
fcis-31632	62	1	joint	joint	ADJ
fcis-31632	62	2	optimization	optimization	NOUN
fcis-31632	62	3	of	of	ADP
fcis-31632	62	4	ai	ai	VERB
fcis-31632	62	5	large	large	ADJ
fcis-31632	62	6	and	and	CCONJ
fcis-31632	62	7	small	small	ADJ
fcis-31632	62	8	models	model	NOUN
fcis-31632	62	9	for	for	ADP
fcis-31632	62	10	surface	surface	NOUN
fcis-31632	62	11	temperature	temperature	NOUN
fcis-31632	62	12	and	and	CCONJ
fcis-31632	62	13	emissivity	emissivity	NOUN
fcis-31632	62	14	retrieval	retrieval	NOUN
fcis-31632	62	15	using	use	VERB
fcis-31632	62	16	knowledge	knowledge	NOUN
fcis-31632	62	17	distillation	distillation	NOUN
fcis-31632	62	18	.	.	PUNCT
fcis-31632	63	1	artificial	artificial	ADJ
fcis-31632	63	2	intelligence	intelligence	NOUN
fcis-31632	63	3	in	in	ADP
fcis-31632	63	4	agriculture,15(3),407	agriculture,15(3),407	NOUN
fcis-31632	63	5	-	-	PUNCT
fcis-31632	63	6	425	425	NUM
fcis-31632	63	7	.	.	PUNCT
fcis-31632	64	1	[	[	X
fcis-31632	64	2	3	3	X
fcis-31632	64	3	]	]	X
fcis-31632	64	4	yang	yang	PROPN
fcis-31632	64	5	juan	juan	PROPN
fcis-31632	64	6	,	,	PUNCT
fcis-31632	64	7	bai	bai	PROPN
fcis-31632	64	8	yu	yu	PROPN
fcis-31632	64	9	,	,	PUNCT
fcis-31632	64	10	gong	gong	PROPN
fcis-31632	64	11	jie	jie	PROPN
fcis-31632	64	12	&	&	CCONJ
fcis-31632	64	13	han	han	PROPN
fcis-31632	64	14	menghui.(2025).the	menghui.(2025).the	DET
fcis-31632	64	15	financial	financial	ADJ
fcis-31632	64	16	institution	institution	NOUN
fcis-31632	64	17	text	text	NOUN
fcis-31632	64	18	data	datum	NOUN
fcis-31632	64	19	mining	mining	NOUN
fcis-31632	64	20	and	and	CCONJ
fcis-31632	64	21	value	value	NOUN
fcis-31632	64	22	analysis	analysis	NOUN
fcis-31632	64	23	model	model	NOUN
fcis-31632	64	24	based	base	VERB
fcis-31632	64	25	on	on	ADP
fcis-31632	64	26	big	big	ADJ
fcis-31632	64	27	data	datum	NOUN
fcis-31632	64	28	and	and	CCONJ
fcis-31632	64	29	natural	natural	ADJ
fcis-31632	64	30	language	language	NOUN
fcis-31632	64	31	processing	processing	NOUN
fcis-31632	64	32	.	.	PUNCT
fcis-31632	65	1	journal	journal	NOUN
fcis-31632	65	2	of	of	ADP
fcis-31632	65	3	organizational	organizational	ADJ
fcis-31632	65	4	and	and	CCONJ
fcis-31632	65	5	end	end	VERB
fcis-31632	65	6	user	user	NOUN
fcis-31632	65	7	computing	computing	PROPN
fcis-31632	65	8	(	(	PUNCT
fcis-31632	65	9	joeuc	joeuc	PROPN
fcis-31632	65	10	)	)	PUNCT
fcis-31632	65	11	,	,	PUNCT
fcis-31632	65	12	37(1	37(1	NUM
fcis-31632	65	13	)	)	PUNCT
fcis-31632	65	14	,	,	PUNCT
fcis-31632	65	15	1	1	NUM
fcis-31632	65	16	-	-	SYM
fcis-31632	65	17	40	40	NUM
fcis-31632	65	18	.	.	PUNCT
fcis-31632	66	1	[	[	X
fcis-31632	66	2	4	4	NUM
fcis-31632	66	3	]	]	X
fcis-31632	66	4	stefanie	stefanie	PROPN
fcis-31632	66	5	seo	seo	PROPN
fcis-31632	66	6	,	,	PUNCT
fcis-31632	66	7	andy	andy	PROPN
fcis-31632	66	8	s	s	VERB
fcis-31632	66	9	ding	ding	PROPN
fcis-31632	66	10	,	,	PUNCT
fcis-31632	66	11	syed	syed	ADJ
fcis-31632	66	12	ameen	ameen	PROPN
fcis-31632	66	13	ahmad	ahmad	PROPN
fcis-31632	66	14	,	,	PUNCT
fcis-31632	66	15	kevin	kevin	PROPN
fcis-31632	66	16	z	z	PROPN
fcis-31632	66	17	xin	xin	PROPN
fcis-31632	66	18	,	,	PUNCT
fcis-31632	66	19	max	max	PROPN
fcis-31632	66	20	l	l	PROPN
fcis-31632	66	21	jiam	jiam	PROPN
fcis-31632	66	22	,	,	PUNCT
fcis-31632	66	23	vincent	vincent	PROPN
fcis-31632	66	24	xin	xin	PROPN
fcis-31632	66	25	...	...	PUNCT
fcis-31632	66	26	&	&	CCONJ
fcis-31632	66	27	nicole	nicole	PROPN
fcis-31632	66	28	t	t	PROPN
fcis-31632	66	29	jiam.(2025).a	jiam.(2025).a	PROPN
fcis-31632	66	30	novel	novel	ADJ
fcis-31632	66	31	natural	natural	ADJ
fcis-31632	66	32	language	language	NOUN
fcis-31632	66	33	processing	processing	NOUN
fcis-31632	66	34	model	model	NOUN
fcis-31632	66	35	for	for	ADP
fcis-31632	66	36	triaging	triage	VERB
fcis-31632	66	37	head	head	NOUN
fcis-31632	66	38	and	and	CCONJ
fcis-31632	66	39	neck	neck	NOUN
fcis-31632	66	40	patient	patient	ADJ
fcis-31632	66	41	appointments	appointment	NOUN
fcis-31632	66	42	..	..	PUNCT
fcis-31632	66	43	otolaryngology	otolaryngology	NOUN
fcis-31632	66	44	--	--	PUNCT
fcis-31632	66	45	head	head	NOUN
fcis-31632	66	46	and	and	CCONJ
fcis-31632	66	47	neck	neck	NOUN
fcis-31632	66	48	surgery	surgery	NOUN
fcis-31632	66	49	:	:	PUNCT
fcis-31632	66	50	official	official	ADJ
fcis-31632	66	51	journal	journal	NOUN
fcis-31632	66	52	of	of	ADP
fcis-31632	66	53	american	american	PROPN
fcis-31632	66	54	academy	academy	PROPN
fcis-31632	66	55	of	of	ADP
fcis-31632	66	56	otolaryngology	otolaryngology	NOUN
fcis-31632	66	57	-	-	PUNCT
fcis-31632	66	58	head	head	NOUN
fcis-31632	66	59	and	and	CCONJ
fcis-31632	66	60	neck	neck	NOUN
fcis-31632	66	61	surgery,173(1),126	surgery,173(1),126	NOUN
fcis-31632	66	62	-	-	PUNCT
fcis-31632	66	63	133	133	NUM
fcis-31632	66	64	.	.	PUNCT
fcis-31632	67	1	[	[	X
fcis-31632	67	2	5	5	NUM
fcis-31632	67	3	]	]	PUNCT
fcis-31632	67	4	yu	yu	PROPN
fcis-31632	67	5	jing.(2024).sentiment	jing.(2024).sentiment	VERB
fcis-31632	67	6	analysis	analysis	NOUN
fcis-31632	67	7	of	of	ADP
fcis-31632	67	8	text	text	NOUN
fcis-31632	67	9	using	use	VERB
fcis-31632	67	10	deep	deep	ADJ
fcis-31632	67	11	learning	learning	NOUN
fcis-31632	67	12	-	-	PUNCT
fcis-31632	67	13	based	base	VERB
fcis-31632	67	14	natural	natural	ADJ
fcis-31632	67	15	language	language	NOUN
fcis-31632	67	16	processing	processing	NOUN
fcis-31632	67	17	models	model	NOUN
fcis-31632	67	18	.	.	PUNCT
fcis-31632	68	1	education	education	NOUN
fcis-31632	68	2	reform	reform	NOUN
fcis-31632	68	3	and	and	CCONJ
fcis-31632	68	4	development,6(12),180	development,6(12),180	NOUN
fcis-31632	68	5	-	-	PUNCT
fcis-31632	68	6	185	185	NUM
fcis-31632	68	7	.	.	PUNCT
fcis-31632	69	1	[	[	X
fcis-31632	69	2	6	6	NUM
fcis-31632	69	3	]	]	X
fcis-31632	69	4	trgovac	trgovac	ADJ
fcis-31632	69	5	ana	ana	PROPN
fcis-31632	69	6	mulović	mulović	PROPN
fcis-31632	69	7	,	,	PUNCT
fcis-31632	69	8	mandić	mandić	PROPN
fcis-31632	69	9	antonija	antonija	PROPN
fcis-31632	69	10	&	&	CCONJ
fcis-31632	69	11	marković	marković	PROPN
fcis-31632	69	12	biljana.(2024	biljana.(2024	PROPN
fcis-31632	69	13	)	)	PUNCT
fcis-31632	69	14	.	.	PUNCT
fcis-31632	70	1	tools	tool	NOUN
fcis-31632	70	2	of	of	ADP
fcis-31632	70	3	artificial	artificial	ADJ
fcis-31632	70	4	intelligence	intelligence	NOUN
fcis-31632	70	5	technology	technology	NOUN
fcis-31632	70	6	as	as	ADP
fcis-31632	70	7	a	a	DET
fcis-31632	70	8	framework	framework	NOUN
fcis-31632	70	9	for	for	ADP
fcis-31632	70	10	transformation	transformation	NOUN
fcis-31632	70	11	digital	digital	ADJ
fcis-31632	70	12	marketing	marketing	NOUN
fcis-31632	70	13	communication	communication	NOUN
fcis-31632	70	14	.	.	PUNCT
fcis-31632	71	1	tehnički	tehnički	NOUN
fcis-31632	71	2	glasnik,18(4),660	glasnik,18(4),660	PROPN
fcis-31632	71	3	-	-	PUNCT
fcis-31632	71	4	665	665	NUM
fcis-31632	71	5	.	.	PUNCT
fcis-31632	72	1	[	[	X
fcis-31632	72	2	7	7	X
fcis-31632	72	3	]	]	X
fcis-31632	72	4	laxmi	laxmi	PROPN
fcis-31632	72	5	choudhary	choudhary	PROPN
fcis-31632	72	6	&	&	CCONJ
fcis-31632	72	7	jitendra	jitendra	PROPN
fcis-31632	72	8	singh	singh	PROPN
fcis-31632	72	9	choudhary.(2024	choudhary.(2024	ADJ
fcis-31632	72	10	)	)	PUNCT
fcis-31632	72	11	.	.	PUNCT
fcis-31632	73	1	deep	deep	ADJ
fcis-31632	73	2	learning	learning	NOUN
fcis-31632	73	3	meets	meet	VERB
fcis-31632	73	4	machine	machine	NOUN
fcis-31632	73	5	learning	learning	NOUN
fcis-31632	73	6	:	:	PUNCT
fcis-31632	73	7	a	a	DET
fcis-31632	73	8	synergistic	synergistic	ADJ
fcis-31632	73	9	approach	approach	NOUN
fcis-31632	73	10	towards	towards	ADP
fcis-31632	73	11	artificial	artificial	ADJ
fcis-31632	73	12	intelligence.journal	intelligence.journal	ADJ
fcis-31632	73	13	of	of	ADP
fcis-31632	73	14	scientific	scientific	ADJ
fcis-31632	73	15	research	research	NOUN
fcis-31632	73	16	and	and	CCONJ
fcis-31632	73	17	reports,30(11),865	reports,30(11),865	VERB
fcis-31632	73	18	-	-	PUNCT
fcis-31632	73	19	875	875	NUM
fcis-31632	73	20	.	.	PUNCT
fcis-31632	74	1	73	73	NUM
fcis-31632	75	1	[	[	SYM
fcis-31632	75	2	8	8	NUM
fcis-31632	75	3	]	]	X
fcis-31632	75	4	xianqiu	xianqiu	PROPN
fcis-31632	75	5	zheng	zheng	PROPN
fcis-31632	75	6	,	,	PUNCT
fcis-31632	75	7	zhidong	zhidong	PROPN
fcis-31632	75	8	zhang	zhang	PROPN
fcis-31632	75	9	,	,	PUNCT
fcis-31632	75	10	liqin	liqin	PROPN
fcis-31632	75	11	wang	wang	PROPN
fcis-31632	75	12	,	,	PUNCT
fcis-31632	75	13	jinhua	jinhua	PROPN
fcis-31632	75	14	wu	wu	PROPN
fcis-31632	75	15	&	&	CCONJ
fcis-31632	75	16	zuofeng	zuofeng	PROPN
fcis-31632	75	17	dong.(2024).investigation	dong.(2024).investigation	NOUN
fcis-31632	75	18	on	on	ADP
fcis-31632	75	19	text	text	NOUN
fcis-31632	75	20	generative	generative	ADJ
fcis-31632	75	21	model	model	NOUN
fcis-31632	75	22	based	base	VERB
fcis-31632	75	23	on	on	ADP
fcis-31632	75	24	deep	deep	ADJ
fcis-31632	75	25	learning	learning	NOUN
fcis-31632	75	26	in	in	ADP
fcis-31632	75	27	natural	natural	ADJ
fcis-31632	75	28	language	language	NOUN
fcis-31632	75	29	processing.journal	processing.journal	NOUN
fcis-31632	75	30	of	of	ADP
fcis-31632	75	31	computational	computational	ADJ
fcis-31632	75	32	methods	method	NOUN
fcis-31632	75	33	in	in	ADP
fcis-31632	75	34	sciences	science	NOUN
fcis-31632	75	35	and	and	CCONJ
fcis-31632	75	36	engineering	engineering	NOUN
fcis-31632	75	37	,	,	PUNCT
fcis-31632	75	38	24	24	NUM
fcis-31632	75	39	(	(	PUNCT
fcis-31632	75	40	6	6	NUM
fcis-31632	75	41	)	)	PUNCT
fcis-31632	75	42	,	,	PUNCT
fcis-31632	75	43	4089	4089	NUM
fcis-31632	75	44	-	-	SYM
fcis-31632	75	45	4100	4100	NUM
fcis-31632	75	46	.	.	PUNCT
fcis-31632	76	1	[	[	X
fcis-31632	76	2	9	9	NUM
fcis-31632	76	3	]	]	X
fcis-31632	76	4	marcel	marcel	PROPN
fcis-31632	76	5	c.	c.	PROPN
fcis-31632	76	6	langenbach	langenbach	PROPN
fcis-31632	76	7	,	,	PUNCT
fcis-31632	76	8	borek	borek	PROPN
fcis-31632	76	9	foldyna	foldyna	PROPN
fcis-31632	76	10	,	,	PUNCT
fcis-31632	76	11	ibrahim	ibrahim	PROPN
fcis-31632	76	12	hadzic	hadzic	PROPN
fcis-31632	76	13	,	,	PUNCT
fcis-31632	76	14	isabel	isabel	PROPN
fcis-31632	76	15	l.	l.	PROPN
fcis-31632	76	16	langenbach	langenbach	PROPN
fcis-31632	76	17	,	,	PUNCT
fcis-31632	76	18	vineet	vineet	PROPN
fcis-31632	76	19	k.	k.	PROPN
fcis-31632	76	20	raghu	raghu	PROPN
fcis-31632	76	21	,	,	PUNCT
fcis-31632	76	22	michael	michael	PROPN
fcis-31632	76	23	t.	t.	PROPN
fcis-31632	76	24	lu	lu	PROPN
fcis-31632	76	25	...	...	PUNCT
fcis-31632	76	26	&	&	CCONJ
fcis-31632	76	27	julius	julius	PROPN
fcis-31632	76	28	c.	c.	PROPN
fcis-31632	76	29	heemelaar	heemelaar	PROPN
fcis-31632	76	30	.	.	PUNCT
fcis-31632	77	1	(	(	PUNCT
fcis-31632	77	2	2024	2024	NUM
fcis-31632	77	3	)	)	PUNCT
fcis-31632	77	4	.	.	PUNCT
fcis-31632	78	1	automated	automate	VERB
fcis-31632	78	2	anonymization	anonymization	NOUN
fcis-31632	78	3	of	of	ADP
fcis-31632	78	4	radiology	radiology	NOUN
fcis-31632	78	5	reports	report	NOUN
fcis-31632	78	6	:	:	PUNCT
fcis-31632	78	7	comparison	comparison	NOUN
fcis-31632	78	8	of	of	ADP
fcis-31632	78	9	publicly	publicly	ADV
fcis-31632	78	10	available	available	ADJ
fcis-31632	78	11	natural	natural	ADJ
fcis-31632	78	12	language	language	NOUN
fcis-31632	78	13	processing	processing	NOUN
fcis-31632	78	14	and	and	CCONJ
fcis-31632	78	15	large	large	ADJ
fcis-31632	78	16	language	language	NOUN
fcis-31632	78	17	models.european	models.european	NOUN
fcis-31632	78	18	radiology	radiology	NOUN
fcis-31632	78	19	,	,	PUNCT
fcis-31632	78	20	35(5),1	35(5),1	PROPN
fcis-31632	78	21	-	-	SYM
fcis-31632	78	22	8	8	NUM
fcis-31632	78	23	.	.	PUNCT