id	sid	tid	token	lemma	pos
ajst-2171	1	1	academic	academic	ADJ
ajst-2171	1	2	journal	journal	NOUN
ajst-2171	1	3	of	of	ADP
ajst-2171	1	4	science	science	NOUN
ajst-2171	1	5	and	and	CCONJ
ajst-2171	1	6	technology	technology	NOUN
ajst-2171	1	7	issn	issn	NOUN
ajst-2171	1	8	:	:	PUNCT
ajst-2171	1	9	2771	2771	NUM
ajst-2171	1	10	-	-	SYM
ajst-2171	1	11	3032	3032	NUM
ajst-2171	1	12	|	|	NOUN
ajst-2171	1	13	vol	vol	NOUN
ajst-2171	1	14	.	.	PROPN
ajst-2171	2	1	3	3	NUM
ajst-2171	2	2	,	,	PUNCT
ajst-2171	2	3	no	no	INTJ
ajst-2171	2	4	.	.	NOUN
ajst-2171	2	5	2	2	NUM
ajst-2171	2	6	,	,	PUNCT
ajst-2171	2	7	2022	2022	NUM
ajst-2171	2	8	191	191	NUM
ajst-2171	2	9	lightweight	lightweight	ADJ
ajst-2171	2	10	text	text	NOUN
ajst-2171	2	11	matching	matching	NOUN
ajst-2171	2	12	method	method	PROPN
ajst-2171	2	13	qingyu	qingyu	PROPN
ajst-2171	2	14	li	li	PROPN
ajst-2171	2	15	,	,	PUNCT
ajst-2171	2	16	yujun	yujun	PROPN
ajst-2171	2	17	zhang	zhang	PROPN
ajst-2171	2	18	*	*	PUNCT
ajst-2171	2	19	school	school	NOUN
ajst-2171	2	20	of	of	ADP
ajst-2171	2	21	computer	computer	NOUN
ajst-2171	2	22	and	and	CCONJ
ajst-2171	2	23	software	software	NOUN
ajst-2171	2	24	engineering	engineering	NOUN
ajst-2171	2	25	,	,	PUNCT
ajst-2171	2	26	university	university	NOUN
ajst-2171	2	27	of	of	ADP
ajst-2171	2	28	science	science	NOUN
ajst-2171	2	29	and	and	CCONJ
ajst-2171	2	30	technology	technology	NOUN
ajst-2171	2	31	liaoning	liaoning	NOUN
ajst-2171	2	32	,	,	PUNCT
ajst-2171	2	33	anshan	anshan	PROPN
ajst-2171	2	34	114051	114051	NUM
ajst-2171	2	35	,	,	PUNCT
ajst-2171	2	36	china	china	PROPN
ajst-2171	2	37	*	*	PUNCT
ajst-2171	2	38	corresponding	correspond	VERB
ajst-2171	2	39	author	author	NOUN
ajst-2171	2	40	abstract	abstract	NOUN
ajst-2171	2	41	:	:	PUNCT
ajst-2171	2	42	this	this	DET
ajst-2171	2	43	paper	paper	NOUN
ajst-2171	2	44	constructs	construct	VERB
ajst-2171	2	45	a	a	DET
ajst-2171	2	46	lightweight	lightweight	ADJ
ajst-2171	2	47	text	text	NOUN
ajst-2171	2	48	matching	matching	NOUN
ajst-2171	2	49	model	model	NOUN
ajst-2171	2	50	.	.	PUNCT
ajst-2171	3	1	this	this	DET
ajst-2171	3	2	model	model	NOUN
ajst-2171	3	3	fully	fully	ADV
ajst-2171	3	4	extracts	extract	VERB
ajst-2171	3	5	the	the	DET
ajst-2171	3	6	features	feature	NOUN
ajst-2171	3	7	of	of	ADP
ajst-2171	3	8	two	two	NUM
ajst-2171	3	9	text	text	NOUN
ajst-2171	3	10	through	through	ADP
ajst-2171	3	11	the	the	DET
ajst-2171	3	12	fusion	fusion	NOUN
ajst-2171	3	13	of	of	ADP
ajst-2171	3	14	double	double	ADJ
ajst-2171	3	15	tower	tower	NOUN
ajst-2171	3	16	and	and	CCONJ
ajst-2171	3	17	interactive	interactive	ADJ
ajst-2171	3	18	methods	method	NOUN
ajst-2171	3	19	,	,	PUNCT
ajst-2171	3	20	and	and	CCONJ
ajst-2171	3	21	then	then	ADV
ajst-2171	3	22	conducts	conduct	VERB
ajst-2171	3	23	deep	deep	ADJ
ajst-2171	3	24	interaction	interaction	NOUN
ajst-2171	3	25	.	.	PUNCT
ajst-2171	4	1	then	then	ADV
ajst-2171	4	2	,	,	PUNCT
ajst-2171	4	3	it	it	PRON
ajst-2171	4	4	classifies	classify	VERB
ajst-2171	4	5	the	the	DET
ajst-2171	4	6	last	last	ADJ
ajst-2171	4	7	extracted	extract	VERB
ajst-2171	4	8	text	text	NOUN
ajst-2171	4	9	features	feature	NOUN
ajst-2171	4	10	through	through	ADP
ajst-2171	4	11	the	the	DET
ajst-2171	4	12	classification	classification	NOUN
ajst-2171	4	13	network	network	NOUN
ajst-2171	4	14	,	,	PUNCT
ajst-2171	4	15	and	and	CCONJ
ajst-2171	4	16	conducts	conduct	VERB
ajst-2171	4	17	training	training	NOUN
ajst-2171	4	18	,	,	PUNCT
ajst-2171	4	19	validation	validation	NOUN
ajst-2171	4	20	and	and	CCONJ
ajst-2171	4	21	testing	testing	NOUN
ajst-2171	4	22	on	on	ADP
ajst-2171	4	23	the	the	DET
ajst-2171	4	24	chinese	chinese	ADJ
ajst-2171	4	25	text	text	NOUN
ajst-2171	4	26	matching	matching	NOUN
ajst-2171	4	27	dataset	dataset	VERB
ajst-2171	4	28	lcqmc	lcqmc	NOUN
ajst-2171	4	29	.	.	PUNCT
ajst-2171	5	1	although	although	SCONJ
ajst-2171	5	2	the	the	DET
ajst-2171	5	3	experimental	experimental	ADJ
ajst-2171	5	4	results	result	NOUN
ajst-2171	5	5	show	show	VERB
ajst-2171	5	6	that	that	SCONJ
ajst-2171	5	7	the	the	DET
ajst-2171	5	8	model	model	NOUN
ajst-2171	5	9	in	in	ADP
ajst-2171	5	10	this	this	DET
ajst-2171	5	11	paper	paper	NOUN
ajst-2171	5	12	is	be	AUX
ajst-2171	5	13	relatively	relatively	ADV
ajst-2171	5	14	good	good	ADJ
ajst-2171	5	15	,	,	PUNCT
ajst-2171	5	16	it	it	PRON
ajst-2171	5	17	still	still	ADV
ajst-2171	5	18	needs	need	VERB
ajst-2171	5	19	to	to	PART
ajst-2171	5	20	be	be	AUX
ajst-2171	5	21	improved	improve	VERB
ajst-2171	5	22	.	.	PUNCT
ajst-2171	6	1	keywords	keyword	NOUN
ajst-2171	6	2	:	:	PUNCT
ajst-2171	6	3	text	text	NOUN
ajst-2171	6	4	matching	matching	NOUN
ajst-2171	6	5	,	,	PUNCT
ajst-2171	6	6	nlp	nlp	ADJ
ajst-2171	6	7	,	,	PUNCT
ajst-2171	6	8	deep	deep	ADJ
ajst-2171	6	9	learning	learning	NOUN
ajst-2171	6	10	.	.	PUNCT
ajst-2171	7	1	1	1	X
ajst-2171	7	2	.	.	X
ajst-2171	7	3	introduction	introduction	NOUN
ajst-2171	7	4	as	as	ADP
ajst-2171	7	5	a	a	DET
ajst-2171	7	6	basic	basic	ADJ
ajst-2171	7	7	and	and	CCONJ
ajst-2171	7	8	core	core	NOUN
ajst-2171	7	9	task	task	NOUN
ajst-2171	7	10	of	of	ADP
ajst-2171	7	11	natural	natural	ADJ
ajst-2171	7	12	language	language	NOUN
ajst-2171	7	13	processing	processing	NOUN
ajst-2171	7	14	[	[	X
ajst-2171	7	15	1	1	NUM
ajst-2171	7	16	]	]	PUNCT
ajst-2171	7	17	,	,	PUNCT
ajst-2171	7	18	text	text	NOUN
ajst-2171	7	19	matching	matching	NOUN
ajst-2171	7	20	can	can	AUX
ajst-2171	7	21	be	be	AUX
ajst-2171	7	22	applied	apply	VERB
ajst-2171	7	23	to	to	ADP
ajst-2171	7	24	many	many	ADJ
ajst-2171	7	25	fields	field	NOUN
ajst-2171	7	26	such	such	ADJ
ajst-2171	7	27	as	as	ADP
ajst-2171	7	28	information	information	NOUN
ajst-2171	7	29	retrieval	retrieval	NOUN
ajst-2171	7	30	,	,	PUNCT
ajst-2171	7	31	retelling	retell	VERB
ajst-2171	7	32	discrimination	discrimination	NOUN
ajst-2171	7	33	,	,	PUNCT
ajst-2171	7	34	etc	etc	X
ajst-2171	7	35	.	.	X
ajst-2171	8	1	nowadays	nowadays	ADV
ajst-2171	8	2	,	,	PUNCT
ajst-2171	8	3	with	with	ADP
ajst-2171	8	4	the	the	DET
ajst-2171	8	5	rapid	rapid	ADJ
ajst-2171	8	6	development	development	NOUN
ajst-2171	8	7	of	of	ADP
ajst-2171	8	8	deep	deep	ADJ
ajst-2171	8	9	learning	learning	NOUN
ajst-2171	8	10	,	,	PUNCT
ajst-2171	8	11	traditional	traditional	ADJ
ajst-2171	8	12	machine	machine	NOUN
ajst-2171	8	13	learning	learning	NOUN
ajst-2171	8	14	has	have	AUX
ajst-2171	8	15	been	be	AUX
ajst-2171	8	16	replaced	replace	VERB
ajst-2171	8	17	.	.	PUNCT
ajst-2171	9	1	more	more	ADJ
ajst-2171	9	2	researchers	researcher	NOUN
ajst-2171	9	3	are	be	AUX
ajst-2171	9	4	committed	commit	VERB
ajst-2171	9	5	to	to	ADP
ajst-2171	9	6	studying	study	VERB
ajst-2171	9	7	efficient	efficient	ADJ
ajst-2171	9	8	text	text	NOUN
ajst-2171	9	9	matching	matching	NOUN
ajst-2171	9	10	models	model	NOUN
ajst-2171	9	11	for	for	ADP
ajst-2171	9	12	deep	deep	ADJ
ajst-2171	9	13	learning	learning	NOUN
ajst-2171	9	14	.	.	PUNCT
ajst-2171	10	1	for	for	ADP
ajst-2171	10	2	example	example	NOUN
ajst-2171	10	3	,	,	PUNCT
ajst-2171	10	4	the	the	DET
ajst-2171	10	5	dssm	dssm	ADJ
ajst-2171	10	6	model	model	NOUN
ajst-2171	10	7	[	[	X
ajst-2171	10	8	2	2	NUM
ajst-2171	10	9	]	]	PUNCT
ajst-2171	10	10	proposed	propose	VERB
ajst-2171	10	11	by	by	ADP
ajst-2171	10	12	huang	huang	PROPN
ajst-2171	10	13	et	et	PROPN
ajst-2171	10	14	al	al	PROPN
ajst-2171	10	15	.	.	PROPN
ajst-2171	10	16	is	be	AUX
ajst-2171	10	17	the	the	DET
ajst-2171	10	18	first	first	ADJ
ajst-2171	10	19	research	research	NOUN
ajst-2171	10	20	result	result	NOUN
ajst-2171	10	21	of	of	ADP
ajst-2171	10	22	deep	deep	ADJ
ajst-2171	10	23	learning	learning	NOUN
ajst-2171	10	24	on	on	ADP
ajst-2171	10	25	text	text	NOUN
ajst-2171	10	26	matching	matching	NOUN
ajst-2171	10	27	.	.	PUNCT
ajst-2171	11	1	this	this	DET
ajst-2171	11	2	model	model	NOUN
ajst-2171	11	3	provides	provide	VERB
ajst-2171	11	4	a	a	DET
ajst-2171	11	5	good	good	ADJ
ajst-2171	11	6	research	research	NOUN
ajst-2171	11	7	idea	idea	NOUN
ajst-2171	11	8	for	for	ADP
ajst-2171	11	9	later	later	ADJ
ajst-2171	11	10	researchers	researcher	NOUN
ajst-2171	11	11	,	,	PUNCT
ajst-2171	11	12	later	later	ADV
ajst-2171	11	13	,	,	PUNCT
ajst-2171	11	14	with	with	ADP
ajst-2171	11	15	the	the	DET
ajst-2171	11	16	popularity	popularity	NOUN
ajst-2171	11	17	of	of	ADP
ajst-2171	11	18	cnn	cnn	PROPN
ajst-2171	11	19	and	and	CCONJ
ajst-2171	11	20	rnn	rnn	PROPN
ajst-2171	11	21	,	,	PUNCT
ajst-2171	11	22	cnn	cnn	PROPN
ajst-2171	11	23	-	-	PUNCT
ajst-2171	11	24	dssm	dssm	PROPN
ajst-2171	11	25	model	model	NOUN
ajst-2171	11	26	[	[	X
ajst-2171	11	27	3	3	NUM
ajst-2171	11	28	]	]	PUNCT
ajst-2171	11	29	and	and	CCONJ
ajst-2171	11	30	lstm	lstm	VERB
ajst-2171	11	31	-	-	PUNCT
ajst-2171	11	32	dssm	dssm	ADJ
ajst-2171	11	33	model	model	NOUN
ajst-2171	11	34	[	[	X
ajst-2171	11	35	4	4	NUM
ajst-2171	11	36	]	]	PUNCT
ajst-2171	11	37	were	be	AUX
ajst-2171	11	38	developed	develop	VERB
ajst-2171	11	39	.	.	PUNCT
ajst-2171	12	1	because	because	SCONJ
ajst-2171	12	2	these	these	DET
ajst-2171	12	3	models	model	NOUN
ajst-2171	12	4	did	do	AUX
ajst-2171	12	5	not	not	PART
ajst-2171	12	6	pay	pay	VERB
ajst-2171	12	7	attention	attention	NOUN
ajst-2171	12	8	to	to	ADP
ajst-2171	12	9	the	the	DET
ajst-2171	12	10	deep	deep	ADJ
ajst-2171	12	11	interaction	interaction	NOUN
ajst-2171	12	12	between	between	ADP
ajst-2171	12	13	texts	text	NOUN
ajst-2171	12	14	,	,	PUNCT
ajst-2171	12	15	researchers	researcher	NOUN
ajst-2171	12	16	later	later	ADV
ajst-2171	12	17	shifted	shift	VERB
ajst-2171	12	18	their	their	PRON
ajst-2171	12	19	focus	focus	NOUN
ajst-2171	12	20	to	to	ADP
ajst-2171	12	21	how	how	SCONJ
ajst-2171	12	22	to	to	PART
ajst-2171	12	23	obtain	obtain	VERB
ajst-2171	12	24	the	the	DET
ajst-2171	12	25	interaction	interaction	NOUN
ajst-2171	12	26	information	information	NOUN
ajst-2171	12	27	between	between	ADP
ajst-2171	12	28	texts	text	NOUN
ajst-2171	12	29	,	,	PUNCT
ajst-2171	12	30	and	and	CCONJ
ajst-2171	12	31	developed	develop	VERB
ajst-2171	12	32	many	many	ADJ
ajst-2171	12	33	excellent	excellent	ADJ
ajst-2171	12	34	models	model	NOUN
ajst-2171	12	35	,	,	PUNCT
ajst-2171	12	36	such	such	ADJ
ajst-2171	12	37	as	as	ADP
ajst-2171	12	38	matchpyramid	matchpyramid	NOUN
ajst-2171	12	39	model	model	NOUN
ajst-2171	12	40	[	[	X
ajst-2171	12	41	5	5	NUM
ajst-2171	12	42	]	]	PUNCT
ajst-2171	12	43	,	,	PUNCT
ajst-2171	12	44	esim	esim	ADJ
ajst-2171	12	45	model	model	NOUN
ajst-2171	13	1	[	[	X
ajst-2171	13	2	6	6	NUM
ajst-2171	13	3	]	]	PUNCT
ajst-2171	13	4	and	and	CCONJ
ajst-2171	13	5	bimpm	bimpm	PROPN
ajst-2171	13	6	model	model	NOUN
ajst-2171	13	7	[	[	X
ajst-2171	13	8	7	7	NUM
ajst-2171	13	9	]	]	PUNCT
ajst-2171	13	10	.	.	PUNCT
ajst-2171	14	1	these	these	DET
ajst-2171	14	2	models	model	NOUN
ajst-2171	14	3	provide	provide	VERB
ajst-2171	14	4	a	a	DET
ajst-2171	14	5	good	good	ADJ
ajst-2171	14	6	idea	idea	NOUN
ajst-2171	14	7	for	for	ADP
ajst-2171	14	8	our	our	PRON
ajst-2171	14	9	research	research	NOUN
ajst-2171	14	10	,	,	PUNCT
ajst-2171	14	11	but	but	CCONJ
ajst-2171	14	12	from	from	ADP
ajst-2171	14	13	the	the	DET
ajst-2171	14	14	current	current	ADJ
ajst-2171	14	15	background	background	NOUN
ajst-2171	14	16	,	,	PUNCT
ajst-2171	14	17	the	the	DET
ajst-2171	14	18	model	model	NOUN
ajst-2171	14	19	performance	performance	NOUN
ajst-2171	14	20	is	be	AUX
ajst-2171	14	21	not	not	PART
ajst-2171	14	22	excellent	excellent	ADJ
ajst-2171	14	23	,	,	PUNCT
ajst-2171	14	24	the	the	DET
ajst-2171	14	25	semantic	semantic	ADJ
ajst-2171	14	26	information	information	NOUN
ajst-2171	14	27	of	of	ADP
ajst-2171	14	28	text	text	NOUN
ajst-2171	14	29	can	can	AUX
ajst-2171	14	30	not	not	PART
ajst-2171	14	31	be	be	AUX
ajst-2171	14	32	extracted	extract	VERB
ajst-2171	14	33	completely	completely	ADV
ajst-2171	14	34	.	.	PUNCT
ajst-2171	15	1	with	with	ADP
ajst-2171	15	2	the	the	DET
ajst-2171	15	3	rise	rise	NOUN
ajst-2171	15	4	of	of	ADP
ajst-2171	15	5	the	the	DET
ajst-2171	15	6	large	large	ADJ
ajst-2171	15	7	pretraining	pretraine	VERB
ajst-2171	15	8	language	language	NOUN
ajst-2171	15	9	model	model	NOUN
ajst-2171	15	10	bert	bert	PROPN
ajst-2171	16	1	[	[	X
ajst-2171	16	2	8	8	NUM
ajst-2171	16	3	]	]	PUNCT
ajst-2171	16	4	,	,	PUNCT
ajst-2171	16	5	pretraining	pretraine	VERB
ajst-2171	16	6	and	and	CCONJ
ajst-2171	16	7	fine	fine	ADJ
ajst-2171	16	8	tuning	tuning	NOUN
ajst-2171	16	9	have	have	AUX
ajst-2171	16	10	become	become	VERB
ajst-2171	16	11	the	the	DET
ajst-2171	16	12	mainstream	mainstream	NOUN
ajst-2171	16	13	research	research	NOUN
ajst-2171	16	14	paradigm	paradigm	NOUN
ajst-2171	16	15	,	,	PUNCT
ajst-2171	16	16	but	but	CCONJ
ajst-2171	16	17	the	the	DET
ajst-2171	16	18	large	large	ADJ
ajst-2171	16	19	pretraining	pretraine	VERB
ajst-2171	16	20	model	model	NOUN
ajst-2171	16	21	has	have	VERB
ajst-2171	16	22	large	large	ADJ
ajst-2171	16	23	parameters	parameter	NOUN
ajst-2171	16	24	and	and	CCONJ
ajst-2171	16	25	long	long	ADJ
ajst-2171	16	26	training	training	NOUN
ajst-2171	16	27	time	time	NOUN
ajst-2171	16	28	.	.	PUNCT
ajst-2171	17	1	how	how	SCONJ
ajst-2171	17	2	to	to	PART
ajst-2171	17	3	find	find	VERB
ajst-2171	17	4	a	a	DET
ajst-2171	17	5	compromise	compromise	NOUN
ajst-2171	17	6	way	way	NOUN
ajst-2171	17	7	to	to	PART
ajst-2171	17	8	get	get	VERB
ajst-2171	17	9	a	a	DET
ajst-2171	17	10	model	model	NOUN
ajst-2171	17	11	with	with	ADP
ajst-2171	17	12	high	high	ADJ
ajst-2171	17	13	performance	performance	NOUN
ajst-2171	17	14	and	and	CCONJ
ajst-2171	17	15	fast	fast	ADJ
ajst-2171	17	16	speed	speed	NOUN
ajst-2171	17	17	has	have	AUX
ajst-2171	17	18	become	become	VERB
ajst-2171	17	19	the	the	DET
ajst-2171	17	20	focus	focus	NOUN
ajst-2171	17	21	of	of	ADP
ajst-2171	17	22	research	research	NOUN
ajst-2171	17	23	.	.	PUNCT
ajst-2171	18	1	this	this	DET
ajst-2171	18	2	paper	paper	NOUN
ajst-2171	18	3	proposes	propose	VERB
ajst-2171	18	4	a	a	DET
ajst-2171	18	5	lightweight	lightweight	ADJ
ajst-2171	18	6	text	text	NOUN
ajst-2171	18	7	matching	matching	NOUN
ajst-2171	18	8	method	method	NOUN
ajst-2171	18	9	,	,	PUNCT
ajst-2171	18	10	which	which	PRON
ajst-2171	18	11	takes	take	VERB
ajst-2171	18	12	less	less	ADJ
ajst-2171	18	13	training	training	NOUN
ajst-2171	18	14	time	time	NOUN
ajst-2171	18	15	than	than	ADP
ajst-2171	18	16	large	large	ADJ
ajst-2171	18	17	pretraining	pretraine	VERB
ajst-2171	18	18	models	model	NOUN
ajst-2171	18	19	.	.	PUNCT
ajst-2171	19	1	2	2	X
ajst-2171	19	2	.	.	X
ajst-2171	19	3	our	our	PRON
ajst-2171	19	4	method	method	NOUN
ajst-2171	19	5	this	this	DET
ajst-2171	19	6	paper	paper	NOUN
ajst-2171	19	7	refers	refer	VERB
ajst-2171	19	8	to	to	ADP
ajst-2171	19	9	the	the	DET
ajst-2171	19	10	design	design	NOUN
ajst-2171	19	11	idea	idea	NOUN
ajst-2171	19	12	of	of	ADP
ajst-2171	19	13	re2	re2	PROPN
ajst-2171	19	14	framework	framework	NOUN
ajst-2171	19	15	[	[	X
ajst-2171	19	16	9	9	NUM
ajst-2171	19	17	]	]	PUNCT
ajst-2171	19	18	,	,	PUNCT
ajst-2171	19	19	uses	use	VERB
ajst-2171	19	20	bert	bert	PROPN
ajst-2171	19	21	model	model	NOUN
ajst-2171	19	22	to	to	PART
ajst-2171	19	23	obtain	obtain	VERB
ajst-2171	19	24	the	the	DET
ajst-2171	19	25	word	word	NOUN
ajst-2171	19	26	embedding	embed	VERB
ajst-2171	19	27	of	of	ADP
ajst-2171	19	28	two	two	NUM
ajst-2171	19	29	texts	text	NOUN
ajst-2171	19	30	,	,	PUNCT
ajst-2171	19	31	and	and	CCONJ
ajst-2171	19	32	then	then	ADV
ajst-2171	19	33	respectively	respectively	ADV
ajst-2171	19	34	input	input	VERB
ajst-2171	19	35	the	the	DET
ajst-2171	19	36	two	two	NUM
ajst-2171	19	37	texts	text	NOUN
ajst-2171	19	38	into	into	ADP
ajst-2171	19	39	the	the	DET
ajst-2171	19	40	encoder	encoder	NOUN
ajst-2171	19	41	layer	layer	NOUN
ajst-2171	19	42	to	to	PART
ajst-2171	19	43	encode	encode	VERB
ajst-2171	19	44	the	the	DET
ajst-2171	19	45	text	text	NOUN
ajst-2171	19	46	,	,	PUNCT
ajst-2171	19	47	and	and	CCONJ
ajst-2171	19	48	then	then	ADV
ajst-2171	19	49	interactively	interactively	ADV
ajst-2171	19	50	align	align	VERB
ajst-2171	19	51	the	the	DET
ajst-2171	19	52	features	feature	NOUN
ajst-2171	19	53	obtained	obtain	VERB
ajst-2171	19	54	from	from	ADP
ajst-2171	19	55	the	the	DET
ajst-2171	19	56	encoder	encoder	NOUN
ajst-2171	19	57	layer	layer	NOUN
ajst-2171	19	58	,	,	PUNCT
ajst-2171	19	59	then	then	ADV
ajst-2171	19	60	fuse	fuse	VERB
ajst-2171	19	61	the	the	DET
ajst-2171	19	62	features	feature	NOUN
ajst-2171	19	63	through	through	ADP
ajst-2171	19	64	the	the	DET
ajst-2171	19	65	fusion	fusion	NOUN
ajst-2171	19	66	layer	layer	NOUN
ajst-2171	19	67	,	,	PUNCT
ajst-2171	19	68	and	and	CCONJ
ajst-2171	19	69	finally	finally	ADV
ajst-2171	19	70	conduct	conduct	VERB
ajst-2171	19	71	classification	classification	NOUN
ajst-2171	19	72	prediction	prediction	NOUN
ajst-2171	19	73	after	after	ADP
ajst-2171	19	74	pooling	pool	VERB
ajst-2171	19	75	.	.	PUNCT
ajst-2171	20	1	the	the	DET
ajst-2171	20	2	overall	overall	ADJ
ajst-2171	20	3	frame	frame	NOUN
ajst-2171	20	4	diagram	diagram	NOUN
ajst-2171	20	5	is	be	AUX
ajst-2171	20	6	shown	show	VERB
ajst-2171	20	7	in	in	ADP
ajst-2171	20	8	figure	figure	NOUN
ajst-2171	20	9	1	1	NUM
ajst-2171	20	10	.	.	PUNCT
ajst-2171	20	11	figure	figure	NOUN
ajst-2171	20	12	1	1	NUM
ajst-2171	20	13	.	.	PUNCT
ajst-2171	21	1	overall	overall	ADJ
ajst-2171	21	2	frame	frame	NOUN
ajst-2171	21	3	diagram	diagram	NOUN
ajst-2171	21	4	2.1	2.1	NUM
ajst-2171	21	5	.	.	PUNCT
ajst-2171	22	1	encoder	encoder	NOUN
ajst-2171	22	2	in	in	ADP
ajst-2171	22	3	this	this	DET
ajst-2171	22	4	paper	paper	NOUN
ajst-2171	22	5	,	,	PUNCT
ajst-2171	22	6	the	the	DET
ajst-2171	22	7	encoder	encoder	NOUN
ajst-2171	22	8	layer	layer	NOUN
ajst-2171	22	9	uses	use	VERB
ajst-2171	22	10	one	one	NUM
ajst-2171	22	11	-	-	PUNCT
ajst-2171	22	12	dimensional	dimensional	ADJ
ajst-2171	22	13	convolutions	convolution	NOUN
ajst-2171	22	14	with	with	ADP
ajst-2171	22	15	convolution	convolution	NOUN
ajst-2171	22	16	kernels	kernel	NOUN
ajst-2171	22	17	of	of	ADP
ajst-2171	22	18	2	2	NUM
ajst-2171	22	19	,	,	PUNCT
ajst-2171	22	20	3	3	NUM
ajst-2171	22	21	and	and	CCONJ
ajst-2171	22	22	4	4	NUM
ajst-2171	22	23	to	to	PART
ajst-2171	22	24	encode	encode	VERB
ajst-2171	22	25	the	the	DET
ajst-2171	22	26	word	word	NOUN
ajst-2171	22	27	embedding	embed	VERB
ajst-2171	22	28	ea	ea	NOUN
ajst-2171	22	29	and	and	CCONJ
ajst-2171	22	30	eb	eb	PROPN
ajst-2171	22	31	obtained	obtain	VERB
ajst-2171	22	32	from	from	ADP
ajst-2171	22	33	the	the	DET
ajst-2171	22	34	embedding	embed	VERB
ajst-2171	22	35	layer	layer	NOUN
ajst-2171	22	36	to	to	PART
ajst-2171	22	37	obtain	obtain	VERB
ajst-2171	22	38	text	text	NOUN
ajst-2171	22	39	features	feature	NOUN
ajst-2171	22	40	a1	a1	PROPN
ajst-2171	22	41	,	,	PUNCT
ajst-2171	22	42	a2	a2	PROPN
ajst-2171	22	43	,	,	PUNCT
ajst-2171	22	44	a3	a3	NOUN
ajst-2171	22	45	,	,	PUNCT
ajst-2171	22	46	b1	b1	NOUN
ajst-2171	22	47	,	,	PUNCT
ajst-2171	22	48	b2	b2	NOUN
ajst-2171	22	49	,	,	PUNCT
ajst-2171	22	50	and	and	CCONJ
ajst-2171	22	51	b3	b3	PROPN
ajst-2171	22	52	of	of	ADP
ajst-2171	22	53	192	192	NUM
ajst-2171	22	54	different	different	ADJ
ajst-2171	22	55	scales	scale	NOUN
ajst-2171	22	56	.	.	PUNCT
ajst-2171	23	1	the	the	DET
ajst-2171	23	2	calculation	calculation	NOUN
ajst-2171	23	3	formula	formula	NOUN
ajst-2171	23	4	is	be	AUX
ajst-2171	23	5	as	as	SCONJ
ajst-2171	23	6	follows	follow	VERB
ajst-2171	23	7	:	:	PUNCT
ajst-2171	23	8	(	(	PUNCT
ajst-2171	23	9	,	,	PUNCT
ajst-2171	23	10	)	)	PUNCT
ajst-2171	24	1	i	i	PRON
ajst-2171	24	2	a	a	DET
ajst-2171	24	3	ia	ia	PROPN
ajst-2171	24	4	conv	conv	PROPN
ajst-2171	24	5	e	e	PROPN
ajst-2171	24	6	k	k	PROPN
ajst-2171	24	7	(	(	PUNCT
ajst-2171	24	8	1	1	NUM
ajst-2171	24	9	)	)	PUNCT
ajst-2171	24	10	(	(	PUNCT
ajst-2171	24	11	,	,	PUNCT
ajst-2171	24	12	)	)	PUNCT
ajst-2171	25	1	i	i	PRON
ajst-2171	25	2	b	b	NUM
ajst-2171	25	3	ib	ib	PROPN
ajst-2171	25	4	conv	conv	PROPN
ajst-2171	25	5	e	e	PROPN
ajst-2171	25	6	k	k	PROPN
ajst-2171	25	7	(	(	PUNCT
ajst-2171	25	8	2	2	NUM
ajst-2171	25	9	)	)	PUNCT
ajst-2171	25	10	where	where	SCONJ
ajst-2171	25	11	conv	conv	NOUN
ajst-2171	25	12	represents	represent	VERB
ajst-2171	25	13	one	one	NUM
ajst-2171	25	14	-	-	PUNCT
ajst-2171	25	15	dimensional	dimensional	ADJ
ajst-2171	25	16	convolution	convolution	NOUN
ajst-2171	25	17	operation	operation	NOUN
ajst-2171	25	18	,	,	PUNCT
ajst-2171	25	19	and	and	CCONJ
ajst-2171	25	20	k	k	PROPN
ajst-2171	25	21	represents	represent	VERB
ajst-2171	25	22	the	the	DET
ajst-2171	25	23	size	size	NOUN
ajst-2171	25	24	of	of	ADP
ajst-2171	25	25	convolution	convolution	NOUN
ajst-2171	25	26	kernel	kernel	NOUN
ajst-2171	25	27	.	.	PUNCT
ajst-2171	26	1	2.2	2.2	NUM
ajst-2171	26	2	.	.	PUNCT
ajst-2171	26	3	alignment	alignment	NOUN
ajst-2171	26	4	align	align	VERB
ajst-2171	26	5	the	the	DET
ajst-2171	26	6	features	feature	NOUN
ajst-2171	26	7	of	of	ADP
ajst-2171	26	8	the	the	DET
ajst-2171	26	9	two	two	NUM
ajst-2171	26	10	pieces	piece	NOUN
ajst-2171	26	11	of	of	ADP
ajst-2171	26	12	text	text	NOUN
ajst-2171	26	13	obtained	obtain	VERB
ajst-2171	26	14	from	from	ADP
ajst-2171	26	15	the	the	DET
ajst-2171	26	16	encoder	encoder	NOUN
ajst-2171	26	17	layer	layer	NOUN
ajst-2171	26	18	.	.	PUNCT
ajst-2171	27	1	first	first	ADV
ajst-2171	27	2	,	,	PUNCT
ajst-2171	27	3	construct	construct	VERB
ajst-2171	27	4	the	the	DET
ajst-2171	27	5	interaction	interaction	NOUN
ajst-2171	27	6	matrix	matrix	NOUN
ajst-2171	27	7	attn	attn	NOUN
ajst-2171	27	8	of	of	ADP
ajst-2171	27	9	the	the	DET
ajst-2171	27	10	two	two	NUM
ajst-2171	27	11	pieces	piece	NOUN
ajst-2171	27	12	of	of	ADP
ajst-2171	27	13	text	text	NOUN
ajst-2171	27	14	,	,	PUNCT
ajst-2171	27	15	and	and	CCONJ
ajst-2171	27	16	then	then	ADV
ajst-2171	27	17	perform	perform	VERB
ajst-2171	27	18	the	the	DET
ajst-2171	27	19	softmax	softmax	NOUN
ajst-2171	27	20	operation	operation	NOUN
ajst-2171	27	21	on	on	ADP
ajst-2171	27	22	the	the	DET
ajst-2171	27	23	interaction	interaction	NOUN
ajst-2171	27	24	matrix	matrix	NOUN
ajst-2171	27	25	respectively	respectively	ADV
ajst-2171	27	26	,	,	PUNCT
ajst-2171	27	27	and	and	CCONJ
ajst-2171	27	28	then	then	ADV
ajst-2171	27	29	multiply	multiply	VERB
ajst-2171	27	30	the	the	DET
ajst-2171	27	31	corresponding	correspond	VERB
ajst-2171	27	32	text	text	NOUN
ajst-2171	27	33	feature	feature	NOUN
ajst-2171	27	34	matrix	matrix	NOUN
ajst-2171	27	35	to	to	PART
ajst-2171	27	36	obtain	obtain	VERB
ajst-2171	27	37	the	the	DET
ajst-2171	27	38	aligned	aligned	ADJ
ajst-2171	27	39	matrix	matrix	NOUN
ajst-2171	27	40	.	.	PUNCT
ajst-2171	28	1	as	as	SCONJ
ajst-2171	28	2	the	the	DET
ajst-2171	28	3	encoder	encoder	NOUN
ajst-2171	28	4	layer	layer	NOUN
ajst-2171	28	5	uses	use	VERB
ajst-2171	28	6	three	three	NUM
ajst-2171	28	7	different	different	ADJ
ajst-2171	28	8	convolution	convolution	NOUN
ajst-2171	28	9	kernels	kernel	NOUN
ajst-2171	28	10	to	to	PART
ajst-2171	28	11	extract	extract	VERB
ajst-2171	28	12	features	feature	NOUN
ajst-2171	28	13	from	from	ADP
ajst-2171	28	14	two	two	NUM
ajst-2171	28	15	pieces	piece	NOUN
ajst-2171	28	16	of	of	ADP
ajst-2171	28	17	text	text	NOUN
ajst-2171	28	18	,	,	PUNCT
ajst-2171	28	19	we	we	PRON
ajst-2171	28	20	align	align	VERB
ajst-2171	28	21	the	the	DET
ajst-2171	28	22	features	feature	NOUN
ajst-2171	28	23	corresponding	correspond	VERB
ajst-2171	28	24	to	to	ADP
ajst-2171	28	25	two	two	NUM
ajst-2171	28	26	pieces	piece	NOUN
ajst-2171	28	27	of	of	ADP
ajst-2171	28	28	text	text	NOUN
ajst-2171	28	29	to	to	PART
ajst-2171	28	30	get	get	VERB
ajst-2171	28	31	align	align	ADJ
ajst-2171	28	32	_	_	PUNCT
ajst-2171	28	33	a1	a1	NOUN
ajst-2171	28	34	,	,	PUNCT
ajst-2171	28	35	align	align	ADJ
ajst-2171	28	36	_	_	SYM
ajst-2171	28	37	a2	a2	NOUN
ajst-2171	28	38	,	,	PUNCT
ajst-2171	28	39	align	align	ADJ
ajst-2171	28	40	_	_	PRON
ajst-2171	28	41	a3	a3	NOUN
ajst-2171	28	42	,	,	PUNCT
ajst-2171	28	43	align	align	ADJ
ajst-2171	28	44	_	_	PUNCT
ajst-2171	29	1	b1	b1	NOUN
ajst-2171	29	2	,	,	PUNCT
ajst-2171	29	3	align	align	ADJ
ajst-2171	29	4	_	_	PRON
ajst-2171	29	5	b2	b2	NOUN
ajst-2171	29	6	,	,	PUNCT
ajst-2171	29	7	align	align	ADJ
ajst-2171	29	8	_	_	PUNCT
ajst-2171	29	9	b3	b3	NOUN
ajst-2171	30	1	,	,	PUNCT
ajst-2171	30	2	the	the	DET
ajst-2171	30	3	calculation	calculation	NOUN
ajst-2171	30	4	formula	formula	NOUN
ajst-2171	30	5	is	be	AUX
ajst-2171	30	6	as	as	SCONJ
ajst-2171	30	7	follows	follow	VERB
ajst-2171	30	8	:	:	PUNCT
ajst-2171	30	9	*	*	PUNCT
ajst-2171	30	10	t	t	X
ajst-2171	31	1	i	i	PRON
ajst-2171	31	2	i	i	PRON
ajst-2171	32	1	i	i	PRON
ajst-2171	32	2	iattn	iattn	VERB
ajst-2171	32	3	a	a	DET
ajst-2171	32	4	b	b	NOUN
ajst-2171	32	5	t	t	PROPN
ajst-2171	32	6	(	(	PUNCT
ajst-2171	32	7	3	3	NUM
ajst-2171	32	8	)	)	PUNCT
ajst-2171	32	9	_	_	PUNCT
ajst-2171	33	1	max	max	PROPN
ajst-2171	33	2	(	(	PUNCT
ajst-2171	33	3	)	)	PUNCT
ajst-2171	34	1	*	*	PUNCT
ajst-2171	34	2	i	i	PRON
ajst-2171	34	3	i	i	PRON
ajst-2171	34	4	ialign	ialign	VERB
ajst-2171	34	5	a	a	DET
ajst-2171	34	6	soft	soft	ADJ
ajst-2171	34	7	attn	attn	NOUN
ajst-2171	34	8	b	b	NOUN
ajst-2171	34	9	(	(	PUNCT
ajst-2171	34	10	4	4	NUM
ajst-2171	34	11	)	)	PUNCT
ajst-2171	34	12	_	_	PRON
ajst-2171	35	1	max	max	PROPN
ajst-2171	35	2	(	(	PUNCT
ajst-2171	35	3	)	)	PUNCT
ajst-2171	36	1	*	*	PUNCT
ajst-2171	36	2	t	t	X
ajst-2171	37	1	i	i	PRON
ajst-2171	37	2	i	i	PRON
ajst-2171	37	3	ialign	ialign	VERB
ajst-2171	37	4	b	b	NUM
ajst-2171	37	5	soft	soft	ADJ
ajst-2171	37	6	attn	attn	NOUN
ajst-2171	37	7	a	a	NOUN
ajst-2171	37	8	(	(	PUNCT
ajst-2171	37	9	5	5	NUM
ajst-2171	37	10	)	)	PUNCT
ajst-2171	37	11	2.3	2.3	NUM
ajst-2171	37	12	.	.	PUNCT
ajst-2171	38	1	fusion	fusion	VERB
ajst-2171	38	2	the	the	DET
ajst-2171	38	3	feature	feature	NOUN
ajst-2171	38	4	matrix	matrix	NOUN
ajst-2171	38	5	obtained	obtain	VERB
ajst-2171	38	6	from	from	ADP
ajst-2171	38	7	the	the	DET
ajst-2171	38	8	alignment	alignment	NOUN
ajst-2171	38	9	layer	layer	NOUN
ajst-2171	38	10	and	and	CCONJ
ajst-2171	38	11	the	the	DET
ajst-2171	38	12	feature	feature	NOUN
ajst-2171	38	13	matrix	matrix	NOUN
ajst-2171	38	14	information	information	NOUN
ajst-2171	38	15	obtained	obtain	VERB
ajst-2171	38	16	from	from	ADP
ajst-2171	38	17	the	the	DET
ajst-2171	38	18	encoder	encoder	NOUN
ajst-2171	38	19	layer	layer	NOUN
ajst-2171	38	20	are	be	AUX
ajst-2171	38	21	fused	fuse	VERB
ajst-2171	38	22	to	to	PART
ajst-2171	38	23	obtain	obtain	VERB
ajst-2171	38	24	multiple	multiple	ADJ
ajst-2171	38	25	semantic	semantic	ADJ
ajst-2171	38	26	features	feature	NOUN
ajst-2171	38	27	.	.	PUNCT
ajst-2171	39	1	after	after	SCONJ
ajst-2171	39	2	the	the	DET
ajst-2171	39	3	two	two	NUM
ajst-2171	39	4	matrices	matrix	NOUN
ajst-2171	39	5	are	be	AUX
ajst-2171	39	6	spliced	splice	VERB
ajst-2171	39	7	,	,	PUNCT
ajst-2171	39	8	they	they	PRON
ajst-2171	39	9	pass	pass	VERB
ajst-2171	39	10	through	through	ADP
ajst-2171	39	11	the	the	DET
ajst-2171	39	12	full	full	ADJ
ajst-2171	39	13	connection	connection	NOUN
ajst-2171	39	14	layer	layer	NOUN
ajst-2171	39	15	.	.	PUNCT
ajst-2171	40	1	here	here	ADV
ajst-2171	40	2	we	we	PRON
ajst-2171	40	3	choose	choose	VERB
ajst-2171	40	4	three	three	NUM
ajst-2171	40	5	splicing	splicing	NOUN
ajst-2171	40	6	methods	method	NOUN
ajst-2171	40	7	.	.	PUNCT
ajst-2171	41	1	the	the	DET
ajst-2171	41	2	calculation	calculation	NOUN
ajst-2171	41	3	formula	formula	NOUN
ajst-2171	41	4	of	of	ADP
ajst-2171	41	5	text	text	NOUN
ajst-2171	41	6	a	a	PRON
ajst-2171	41	7	is	be	AUX
ajst-2171	41	8	as	as	SCONJ
ajst-2171	41	9	follows	follow	VERB
ajst-2171	41	10	,	,	PUNCT
ajst-2171	41	11	while	while	SCONJ
ajst-2171	41	12	the	the	DET
ajst-2171	41	13	calculation	calculation	NOUN
ajst-2171	41	14	formula	formula	NOUN
ajst-2171	41	15	of	of	ADP
ajst-2171	41	16	text	text	NOUN
ajst-2171	41	17	b	b	NOUN
ajst-2171	41	18	is	be	AUX
ajst-2171	41	19	the	the	DET
ajst-2171	41	20	same	same	ADJ
ajst-2171	41	21	.	.	PUNCT
ajst-2171	42	1	after	after	ADP
ajst-2171	42	2	fusion	fusion	NOUN
ajst-2171	42	3	operation	operation	NOUN
ajst-2171	42	4	,	,	PUNCT
ajst-2171	42	5	the	the	DET
ajst-2171	42	6	fused	fuse	VERB
ajst-2171	42	7	features	feature	NOUN
ajst-2171	42	8	fai	fai	X
ajst-2171	42	9	and	and	CCONJ
ajst-2171	42	10	fbi	fbi	PROPN
ajst-2171	42	11	are	be	AUX
ajst-2171	42	12	obtained	obtain	VERB
ajst-2171	42	13	:	:	PUNCT
ajst-2171	42	14	1	1	NUM
ajst-2171	42	15	(	(	PUNCT
ajst-2171	42	16	[	[	PUNCT
ajst-2171	42	17	,	,	PUNCT
ajst-2171	42	18	_	_	NOUN
ajst-2171	42	19	]	]	PUNCT
ajst-2171	42	20	)	)	PUNCT
ajst-2171	43	1	f	f	X
ajst-2171	43	2	x	x	PUNCT
ajst-2171	43	3	linear	linear	ADJ
ajst-2171	43	4	x	x	INTJ
ajst-2171	43	5	align	align	VERB
ajst-2171	43	6	x	x	PUNCT
ajst-2171	44	1	(	(	PUNCT
ajst-2171	44	2	6	6	NUM
ajst-2171	44	3	)	)	SYM
ajst-2171	44	4	2	2	NUM
ajst-2171	44	5	(	(	PUNCT
ajst-2171	44	6	[	[	PUNCT
ajst-2171	44	7	,	,	PUNCT
ajst-2171	44	8	_	_	NOUN
ajst-2171	44	9	]	]	PUNCT
ajst-2171	44	10	)	)	PUNCT
ajst-2171	45	1	f	f	X
ajst-2171	45	2	x	x	PUNCT
ajst-2171	45	3	linear	linear	ADJ
ajst-2171	45	4	x	x	X
ajst-2171	45	5	x	x	PUNCT
ajst-2171	45	6	align	align	VERB
ajst-2171	45	7	x	x	PUNCT
ajst-2171	46	1			NOUN
ajst-2171	46	2	(	(	PUNCT
ajst-2171	46	3	7	7	NUM
ajst-2171	46	4	)	)	PUNCT
ajst-2171	46	5	3	3	NUM
ajst-2171	46	6	(	(	PUNCT
ajst-2171	46	7	[	[	PUNCT
ajst-2171	46	8	,	,	PUNCT
ajst-2171	46	9	*	*	PUNCT
ajst-2171	46	10	_	_	NOUN
ajst-2171	46	11	]	]	PUNCT
ajst-2171	46	12	)	)	PUNCT
ajst-2171	46	13	f	f	X
ajst-2171	47	1	x	x	PUNCT
ajst-2171	47	2	linear	linear	ADJ
ajst-2171	47	3	x	x	X
ajst-2171	47	4	x	x	PUNCT
ajst-2171	47	5	align	align	VERB
ajst-2171	47	6	x	x	PUNCT
ajst-2171	48	1	(	(	PUNCT
ajst-2171	48	2	8)	8)	NUM
ajst-2171	48	3	[	[	PUNCT
ajst-2171	48	4	1	1	NUM
ajst-2171	48	5	,	,	PUNCT
ajst-2171	48	6	2	2	NUM
ajst-2171	48	7	,	,	PUNCT
ajst-2171	48	8	3	3	NUM
ajst-2171	48	9	]	]	PUNCT
ajst-2171	48	10	i	i	PRON
ajst-2171	49	1	i	i	PRON
ajst-2171	50	1	i	i	PRON
ajst-2171	50	2	ifa	ifa	VERB
ajst-2171	50	3	f	f	PROPN
ajst-2171	51	1	a	a	DET
ajst-2171	51	2	f	f	X
ajst-2171	51	3	a	a	DET
ajst-2171	51	4	f	f	NOUN
ajst-2171	51	5	a	a	NOUN
ajst-2171	51	6	(	(	PUNCT
ajst-2171	51	7	9	9	NUM
ajst-2171	51	8	)	)	PUNCT
ajst-2171	51	9	[	[	PUNCT
ajst-2171	51	10	1	1	NUM
ajst-2171	51	11	,	,	PUNCT
ajst-2171	51	12	2	2	NUM
ajst-2171	51	13	,	,	PUNCT
ajst-2171	51	14	3	3	NUM
ajst-2171	51	15	]	]	PUNCT
ajst-2171	51	16	i	i	PRON
ajst-2171	52	1	i	i	PRON
ajst-2171	53	1	i	i	PRON
ajst-2171	53	2	ifb	ifb	PROPN
ajst-2171	53	3	f	f	PROPN
ajst-2171	53	4	b	b	PROPN
ajst-2171	53	5	f	f	PROPN
ajst-2171	53	6	b	b	PROPN
ajst-2171	53	7	f	f	PROPN
ajst-2171	53	8	b	b	PROPN
ajst-2171	53	9	(	(	PUNCT
ajst-2171	53	10	10	10	NUM
ajst-2171	53	11	)	)	PUNCT
ajst-2171	53	12	2.4	2.4	NUM
ajst-2171	53	13	.	.	PUNCT
ajst-2171	54	1	pooling	pool	VERB
ajst-2171	54	2	the	the	DET
ajst-2171	54	3	feature	feature	NOUN
ajst-2171	54	4	information	information	NOUN
ajst-2171	54	5	of	of	ADP
ajst-2171	54	6	the	the	DET
ajst-2171	54	7	two	two	NUM
ajst-2171	54	8	texts	text	NOUN
ajst-2171	54	9	obtained	obtain	VERB
ajst-2171	54	10	from	from	ADP
ajst-2171	54	11	the	the	DET
ajst-2171	54	12	previous	previous	ADJ
ajst-2171	54	13	layer	layer	NOUN
ajst-2171	54	14	is	be	AUX
ajst-2171	54	15	pooled	pool	VERB
ajst-2171	54	16	to	to	ADP
ajst-2171	54	17	the	the	DET
ajst-2171	54	18	maximum	maximum	NOUN
ajst-2171	54	19	to	to	PART
ajst-2171	54	20	obtain	obtain	VERB
ajst-2171	54	21	more	more	ADV
ajst-2171	54	22	accurate	accurate	ADJ
ajst-2171	54	23	text	text	NOUN
ajst-2171	54	24	information	information	NOUN
ajst-2171	54	25	a1	a1	NOUN
ajst-2171	54	26	,	,	PUNCT
ajst-2171	54	27	a2	a2	PROPN
ajst-2171	54	28	,	,	PUNCT
ajst-2171	54	29	a3	a3	NOUN
ajst-2171	54	30	,	,	PUNCT
ajst-2171	54	31	b1	b1	NOUN
ajst-2171	54	32	,	,	PUNCT
ajst-2171	54	33	b2	b2	NOUN
ajst-2171	54	34	,	,	PUNCT
ajst-2171	54	35	b3	b3	PROPN
ajst-2171	54	36	.	.	PUNCT
ajst-2171	55	1	the	the	DET
ajst-2171	55	2	calculation	calculation	NOUN
ajst-2171	55	3	formula	formula	NOUN
ajst-2171	55	4	is	be	AUX
ajst-2171	55	5	as	as	SCONJ
ajst-2171	55	6	follows	follow	VERB
ajst-2171	55	7	:	:	PUNCT
ajst-2171	55	8	max	max	PROPN
ajst-2171	55	9	(	(	PUNCT
ajst-2171	55	10	)	)	PUNCT
ajst-2171	55	11	i	i	PRON
ajst-2171	55	12	ia	ia	VERB
ajst-2171	55	13	pooling	pool	VERB
ajst-2171	55	14	fa	fa	PROPN
ajst-2171	55	15	(	(	PUNCT
ajst-2171	55	16	11	11	NUM
ajst-2171	55	17	)	)	PUNCT
ajst-2171	55	18	max	max	NOUN
ajst-2171	55	19	(	(	PUNCT
ajst-2171	55	20	)	)	PUNCT
ajst-2171	55	21	i	i	PRON
ajst-2171	55	22	ib	ib	VERB
ajst-2171	55	23	pooling	pool	VERB
ajst-2171	55	24	fb	fb	PROPN
ajst-2171	55	25	(	(	PUNCT
ajst-2171	55	26	12	12	NUM
ajst-2171	55	27	)	)	PUNCT
ajst-2171	55	28	2.5	2.5	NUM
ajst-2171	55	29	.	.	PUNCT
ajst-2171	56	1	prediction	prediction	NOUN
ajst-2171	56	2	this	this	DET
ajst-2171	56	3	layer	layer	NOUN
ajst-2171	56	4	is	be	AUX
ajst-2171	56	5	used	use	VERB
ajst-2171	56	6	to	to	PART
ajst-2171	56	7	classify	classify	VERB
ajst-2171	56	8	the	the	DET
ajst-2171	56	9	final	final	ADJ
ajst-2171	56	10	features	feature	NOUN
ajst-2171	56	11	of	of	ADP
ajst-2171	56	12	two	two	NUM
ajst-2171	56	13	pieces	piece	NOUN
ajst-2171	56	14	of	of	ADP
ajst-2171	56	15	text	text	NOUN
ajst-2171	56	16	.	.	PUNCT
ajst-2171	57	1	first	first	ADV
ajst-2171	57	2	,	,	PUNCT
ajst-2171	57	3	the	the	DET
ajst-2171	57	4	features	feature	NOUN
ajst-2171	57	5	are	be	AUX
ajst-2171	57	6	spliced	splice	VERB
ajst-2171	57	7	,	,	PUNCT
ajst-2171	57	8	and	and	CCONJ
ajst-2171	57	9	then	then	ADV
ajst-2171	57	10	the	the	DET
ajst-2171	57	11	final	final	ADJ
ajst-2171	57	12	classification	classification	NOUN
ajst-2171	57	13	information	information	NOUN
ajst-2171	57	14	is	be	AUX
ajst-2171	57	15	obtained	obtain	VERB
ajst-2171	57	16	through	through	ADP
ajst-2171	57	17	two	two	NUM
ajst-2171	57	18	fc	fc	PROPN
ajst-2171	57	19	connection	connection	NOUN
ajst-2171	57	20	layers	layer	NOUN
ajst-2171	57	21	,	,	PUNCT
ajst-2171	57	22	the	the	DET
ajst-2171	57	23	calculation	calculation	NOUN
ajst-2171	57	24	formula	formula	NOUN
ajst-2171	57	25	is	be	AUX
ajst-2171	57	26	as	as	SCONJ
ajst-2171	57	27	follows	follow	VERB
ajst-2171	57	28	:	:	PUNCT
ajst-2171	57	29	1	1	NUM
ajst-2171	57	30	2	2	NUM
ajst-2171	57	31	3a	3a	NUM
ajst-2171	57	32	a	a	DET
ajst-2171	57	33	a	a	DET
ajst-2171	57	34	a	a	NOUN
ajst-2171	57	35			X
ajst-2171	57	36			X
ajst-2171	57	37	(	(	PUNCT
ajst-2171	57	38	13	13	NUM
ajst-2171	57	39	)	)	SYM
ajst-2171	57	40	1	1	NUM
ajst-2171	57	41	2	2	NUM
ajst-2171	57	42	3b	3b	NOUN
ajst-2171	57	43	b	b	PROPN
ajst-2171	57	44	b	b	PROPN
ajst-2171	57	45	b	b	PROPN
ajst-2171	57	46			PROPN
ajst-2171	57	47			X
ajst-2171	57	48	(	(	PUNCT
ajst-2171	57	49	14	14	NUM
ajst-2171	57	50	)	)	PUNCT
ajst-2171	57	51	(	(	PUNCT
ajst-2171	57	52	(	(	PUNCT
ajst-2171	57	53	[	[	PUNCT
ajst-2171	57	54	,	,	PUNCT
ajst-2171	57	55	]	]	X
ajst-2171	57	56	)	)	PUNCT
ajst-2171	57	57	pred	pre	VERB
ajst-2171	57	58	linear	linear	PROPN
ajst-2171	57	59	linear	linear	PROPN
ajst-2171	57	60	a	a	DET
ajst-2171	57	61	b	b	NOUN
ajst-2171	57	62	(	(	PUNCT
ajst-2171	57	63	15	15	NUM
ajst-2171	57	64	)	)	PUNCT
ajst-2171	57	65	3	3	NUM
ajst-2171	57	66	.	.	X
ajst-2171	57	67	experiment	experiment	NOUN
ajst-2171	57	68	3.1	3.1	NUM
ajst-2171	57	69	.	.	PUNCT
ajst-2171	58	1	dataset	dataset	VERB
ajst-2171	58	2	in	in	ADP
ajst-2171	58	3	this	this	DET
ajst-2171	58	4	paper	paper	NOUN
ajst-2171	58	5	,	,	PUNCT
ajst-2171	58	6	lcqmc[10	lcqmc[10	PROPN
ajst-2171	58	7	]	]	PUNCT
ajst-2171	58	8	,	,	PUNCT
ajst-2171	58	9	a	a	DET
ajst-2171	58	10	large	large	ADJ
ajst-2171	58	11	chinese	chinese	ADJ
ajst-2171	58	12	text	text	NOUN
ajst-2171	58	13	matching	matching	NOUN
ajst-2171	58	14	dataset	dataset	NOUN
ajst-2171	58	15	,	,	PUNCT
ajst-2171	58	16	is	be	AUX
ajst-2171	58	17	selected	select	VERB
ajst-2171	58	18	for	for	ADP
ajst-2171	58	19	experiments	experiment	NOUN
ajst-2171	58	20	.	.	PUNCT
ajst-2171	59	1	it	it	PRON
ajst-2171	59	2	contains	contain	VERB
ajst-2171	59	3	two	two	NUM
ajst-2171	59	4	pieces	piece	NOUN
ajst-2171	59	5	of	of	ADP
ajst-2171	59	6	text	text	NOUN
ajst-2171	59	7	and	and	CCONJ
ajst-2171	59	8	a	a	DET
ajst-2171	59	9	label	label	NOUN
ajst-2171	59	10	.	.	PUNCT
ajst-2171	60	1	if	if	SCONJ
ajst-2171	60	2	two	two	NUM
ajst-2171	60	3	pieces	piece	NOUN
ajst-2171	60	4	of	of	ADP
ajst-2171	60	5	text	text	NOUN
ajst-2171	60	6	are	be	AUX
ajst-2171	60	7	judged	judge	VERB
ajst-2171	60	8	to	to	PART
ajst-2171	60	9	match	match	VERB
ajst-2171	60	10	,	,	PUNCT
ajst-2171	60	11	the	the	DET
ajst-2171	60	12	label	label	NOUN
ajst-2171	60	13	is	be	AUX
ajst-2171	60	14	1	1	NUM
ajst-2171	60	15	,	,	PUNCT
ajst-2171	60	16	and	and	CCONJ
ajst-2171	60	17	if	if	SCONJ
ajst-2171	60	18	not	not	PART
ajst-2171	60	19	,	,	PUNCT
ajst-2171	60	20	the	the	DET
ajst-2171	60	21	label	label	NOUN
ajst-2171	60	22	is	be	AUX
ajst-2171	60	23	0	0	NUM
ajst-2171	60	24	.	.	PROPN
ajst-2171	60	25	3.2	3.2	NUM
ajst-2171	60	26	.	.	PUNCT
ajst-2171	61	1	experimental	experimental	ADJ
ajst-2171	61	2	environment	environment	NOUN
ajst-2171	61	3	and	and	CCONJ
ajst-2171	61	4	parameter	parameter	NOUN
ajst-2171	61	5	settings	setting	NOUN
ajst-2171	61	6	this	this	DET
ajst-2171	61	7	experiment	experiment	NOUN
ajst-2171	61	8	is	be	AUX
ajst-2171	61	9	implemented	implement	VERB
ajst-2171	61	10	with	with	ADP
ajst-2171	61	11	pytorch	pytorch	NOUN
ajst-2171	61	12	and	and	CCONJ
ajst-2171	61	13	trained	train	VERB
ajst-2171	61	14	on	on	ADP
ajst-2171	61	15	nvidia	nvidia	PROPN
ajst-2171	61	16	geforce	geforce	NOUN
ajst-2171	61	17	rtx	rtx	PROPN
ajst-2171	61	18	3090	3090	NUM
ajst-2171	61	19	24	24	NUM
ajst-2171	61	20	g.	g.	NOUN
ajst-2171	61	21	in	in	ADP
ajst-2171	61	22	the	the	DET
ajst-2171	61	23	model	model	NOUN
ajst-2171	61	24	,	,	PUNCT
ajst-2171	61	25	bert	bert	PROPN
ajst-2171	61	26	selects	select	VERB
ajst-2171	61	27	the	the	DET
ajst-2171	61	28	chinese_wwm_ext	chinese_wwm_ext	PROPN
ajst-2171	62	1	[	[	X
ajst-2171	62	2	11	11	NUM
ajst-2171	62	3	]	]	PUNCT
ajst-2171	62	4	pretained	pretaine	VERB
ajst-2171	62	5	by	by	ADP
ajst-2171	62	6	cui	cui	PROPN
ajst-2171	62	7	et	et	PROPN
ajst-2171	62	8	al	al	PROPN
ajst-2171	62	9	.	.	PUNCT
ajst-2171	63	1	the	the	DET
ajst-2171	63	2	text	text	NOUN
ajst-2171	63	3	length	length	NOUN
ajst-2171	63	4	is	be	AUX
ajst-2171	63	5	32	32	NUM
ajst-2171	63	6	,	,	PUNCT
ajst-2171	63	7	the	the	DET
ajst-2171	63	8	word	word	NOUN
ajst-2171	63	9	emebdding	emebdde	VERB
ajst-2171	63	10	dimension	dimension	NOUN
ajst-2171	63	11	is	be	AUX
ajst-2171	63	12	768	768	NUM
ajst-2171	63	13	,	,	PUNCT
ajst-2171	63	14	the	the	DET
ajst-2171	63	15	hidden	hide	VERB
ajst-2171	63	16	layer	layer	NOUN
ajst-2171	63	17	dimension	dimension	NOUN
ajst-2171	63	18	is	be	AUX
ajst-2171	63	19	384	384	NUM
ajst-2171	63	20	,	,	PUNCT
ajst-2171	63	21	and	and	CCONJ
ajst-2171	63	22	the	the	DET
ajst-2171	63	23	batch_size	batch_size	NOUN
ajst-2171	63	24	is	be	AUX
ajst-2171	63	25	128	128	NUM
ajst-2171	63	26	,	,	PUNCT
ajst-2171	63	27	the	the	DET
ajst-2171	63	28	lr	lr	NOUN
ajst-2171	63	29	of	of	ADP
ajst-2171	63	30	bert	bert	PROPN
ajst-2171	63	31	is	be	AUX
ajst-2171	63	32	2e-5	2e-5	NUM
ajst-2171	63	33	and	and	CCONJ
ajst-2171	63	34	the	the	DET
ajst-2171	63	35	other	other	ADJ
ajst-2171	63	36	part	part	NOUN
ajst-2171	63	37	is	be	AUX
ajst-2171	63	38	0.001	0.001	NUM
ajst-2171	63	39	,	,	PUNCT
ajst-2171	63	40	the	the	DET
ajst-2171	63	41	dropout	dropout	NOUN
ajst-2171	63	42	is	be	AUX
ajst-2171	63	43	0.2	0.2	NUM
ajst-2171	63	44	,	,	PUNCT
ajst-2171	63	45	and	and	CCONJ
ajst-2171	63	46	epoch	epoch	NOUN
ajst-2171	63	47	is	be	AUX
ajst-2171	63	48	20	20	NUM
ajst-2171	63	49	.	.	PUNCT
ajst-2171	63	50	3.3	3.3	NUM
ajst-2171	63	51	.	.	PUNCT
ajst-2171	64	1	evaluation	evaluation	NOUN
ajst-2171	64	2	criteria	criterion	NOUN
ajst-2171	64	3	in	in	ADP
ajst-2171	64	4	this	this	DET
ajst-2171	64	5	paper	paper	NOUN
ajst-2171	64	6	,	,	PUNCT
ajst-2171	64	7	the	the	DET
ajst-2171	64	8	acc	acc	PROPN
ajst-2171	64	9	and	and	CCONJ
ajst-2171	64	10	f1	f1	NOUN
ajst-2171	64	11	are	be	AUX
ajst-2171	64	12	selected	select	VERB
ajst-2171	64	13	to	to	PART
ajst-2171	64	14	evaluate	evaluate	VERB
ajst-2171	64	15	the	the	DET
ajst-2171	64	16	performance	performance	NOUN
ajst-2171	64	17	of	of	ADP
ajst-2171	64	18	the	the	DET
ajst-2171	64	19	model	model	NOUN
ajst-2171	64	20	.	.	PUNCT
ajst-2171	65	1	the	the	DET
ajst-2171	65	2	formula	formula	NOUN
ajst-2171	65	3	is	be	AUX
ajst-2171	65	4	as	as	SCONJ
ajst-2171	65	5	follows	follow	VERB
ajst-2171	65	6	:	:	PUNCT
ajst-2171	65	7	(	(	PUNCT
ajst-2171	65	8	)	)	PUNCT
ajst-2171	65	9	(	(	PUNCT
ajst-2171	65	10	)	)	PUNCT
ajst-2171	66	1	p	p	NOUN
ajst-2171	66	2	n	n	NUM
ajst-2171	66	3	p	p	NOUN
ajst-2171	66	4	n	n	PROPN
ajst-2171	66	5	p	p	NOUN
ajst-2171	66	6	n	n	ADP
ajst-2171	66	7	t	t	PROPN
ajst-2171	66	8	t	t	PROPN
ajst-2171	66	9	acc	acc	PROPN
ajst-2171	67	1	t	t	PROPN
ajst-2171	67	2	t	t	PROPN
ajst-2171	67	3	f	f	PROPN
ajst-2171	67	4	f	f	PROPN
ajst-2171	67	5			PROPN
ajst-2171	67	6			PROPN
ajst-2171	67	7			PROPN
ajst-2171	67	8			PUNCT
ajst-2171	67	9			X
ajst-2171	67	10	(	(	PUNCT
ajst-2171	67	11	16	16	NUM
ajst-2171	67	12	)	)	PUNCT
ajst-2171	67	13	2	2	NUM
ajst-2171	67	14	*	*	PUNCT
ajst-2171	67	15	*	*	PUNCT
ajst-2171	67	16	1	1	X
ajst-2171	67	17	)	)	PUNCT
ajst-2171	67	18	(	(	PUNCT
ajst-2171	67	19	)	)	PUNCT
ajst-2171	67	20	(	(	PUNCT
ajst-2171	67	21	)	)	PUNCT
ajst-2171	68	1	(	(	PUNCT
ajst-2171	68	2	)	)	PUNCT
ajst-2171	68	3	p	p	X
ajst-2171	69	1	p	p	X
ajst-2171	69	2	p	p	X
ajst-2171	69	3	p	p	PROPN
ajst-2171	69	4	p	p	NOUN
ajst-2171	69	5	n	n	CCONJ
ajst-2171	69	6	p	p	NOUN
ajst-2171	69	7	r	r	NOUN
ajst-2171	69	8	t	t	NOUN
ajst-2171	69	9	t	t	NOUN
ajst-2171	69	10	f	f	X
ajst-2171	69	11	p	p	NOUN
ajst-2171	69	12	r	r	NOUN
ajst-2171	69	13	p	p	NOUN
ajst-2171	69	14	r	r	NOUN
ajst-2171	69	15	t	t	NOUN
ajst-2171	70	1	f	f	X
ajst-2171	70	2	f	f	PROPN
ajst-2171	71	1	f	f	PROPN
ajst-2171	71	2			PROPN
ajst-2171	72	1			PROPN
ajst-2171	72	2			PROPN
ajst-2171	72	3			ADV
ajst-2171	72	4			PUNCT
ajst-2171	72	5			PUNCT
ajst-2171	72	6	(	(	PUNCT
ajst-2171	72	7	，	，	X
ajst-2171	72	8	(	(	PUNCT
ajst-2171	72	9	17	17	NUM
ajst-2171	72	10	)	)	PUNCT
ajst-2171	72	11	tp	tp	NOUN
ajst-2171	72	12	represents	represent	VERB
ajst-2171	72	13	the	the	DET
ajst-2171	72	14	positive	positive	ADJ
ajst-2171	72	15	sample	sample	NOUN
ajst-2171	72	16	predicted	predict	VERB
ajst-2171	72	17	by	by	ADP
ajst-2171	72	18	the	the	DET
ajst-2171	72	19	model	model	NOUN
ajst-2171	72	20	as	as	ADP
ajst-2171	72	21	a	a	DET
ajst-2171	72	22	positive	positive	ADJ
ajst-2171	72	23	class	class	NOUN
ajst-2171	72	24	,	,	PUNCT
ajst-2171	72	25	tn	tn	PROPN
ajst-2171	72	26	represents	represent	VERB
ajst-2171	72	27	the	the	DET
ajst-2171	72	28	negative	negative	ADJ
ajst-2171	72	29	sample	sample	NOUN
ajst-2171	72	30	predicted	predict	VERB
ajst-2171	72	31	by	by	ADP
ajst-2171	72	32	the	the	DET
ajst-2171	72	33	model	model	NOUN
ajst-2171	72	34	as	as	ADP
ajst-2171	72	35	a	a	DET
ajst-2171	72	36	negative	negative	ADJ
ajst-2171	72	37	class	class	NOUN
ajst-2171	72	38	,	,	PUNCT
ajst-2171	72	39	fp	fp	X
ajst-2171	72	40	represents	represent	VERB
ajst-2171	72	41	the	the	DET
ajst-2171	72	42	negative	negative	ADJ
ajst-2171	72	43	sample	sample	NOUN
ajst-2171	72	44	predicted	predict	VERB
ajst-2171	72	45	by	by	ADP
ajst-2171	72	46	the	the	DET
ajst-2171	72	47	model	model	NOUN
ajst-2171	72	48	as	as	ADP
ajst-2171	72	49	a	a	DET
ajst-2171	72	50	positive	positive	ADJ
ajst-2171	72	51	class	class	NOUN
ajst-2171	72	52	,	,	PUNCT
ajst-2171	72	53	and	and	CCONJ
ajst-2171	72	54	fn	fn	NOUN
ajst-2171	72	55	represents	represent	VERB
ajst-2171	72	56	the	the	DET
ajst-2171	72	57	positive	positive	ADJ
ajst-2171	72	58	sample	sample	NOUN
ajst-2171	72	59	predicted	predict	VERB
ajst-2171	72	60	by	by	ADP
ajst-2171	72	61	the	the	DET
ajst-2171	72	62	model	model	NOUN
ajst-2171	72	63	as	as	ADP
ajst-2171	72	64	a	a	DET
ajst-2171	72	65	negative	negative	ADJ
ajst-2171	72	66	class	class	NOUN
ajst-2171	72	67	.	.	PUNCT
ajst-2171	73	1	3.4	3.4	NUM
ajst-2171	73	2	.	.	PUNCT
ajst-2171	74	1	experimental	experimental	ADJ
ajst-2171	74	2	result	result	NOUN
ajst-2171	74	3	the	the	DET
ajst-2171	74	4	acc	acc	PROPN
ajst-2171	74	5	and	and	CCONJ
ajst-2171	74	6	f1	f1	PROPN
ajst-2171	74	7	value	value	NOUN
ajst-2171	74	8	of	of	ADP
ajst-2171	74	9	the	the	DET
ajst-2171	74	10	model	model	NOUN
ajst-2171	74	11	obtained	obtain	VERB
ajst-2171	74	12	after	after	ADP
ajst-2171	74	13	training	training	NOUN
ajst-2171	74	14	,	,	PUNCT
ajst-2171	74	15	validation	validation	NOUN
ajst-2171	74	16	and	and	CCONJ
ajst-2171	74	17	test	test	NOUN
ajst-2171	74	18	on	on	ADP
ajst-2171	74	19	lcqmc	lcqmc	PROPN
ajst-2171	74	20	dataset	dataset	NOUN
ajst-2171	74	21	are	be	AUX
ajst-2171	74	22	84.14	84.14	NUM
ajst-2171	74	23	%	%	NOUN
ajst-2171	74	24	and	and	CCONJ
ajst-2171	74	25	85.62	85.62	NUM
ajst-2171	74	26	%	%	NOUN
ajst-2171	74	27	respectively	respectively	ADV
ajst-2171	74	28	.	.	PUNCT
ajst-2171	75	1	compared	compare	VERB
ajst-2171	75	2	with	with	ADP
ajst-2171	75	3	the	the	DET
ajst-2171	75	4	experimental	experimental	ADJ
ajst-2171	75	5	results	result	NOUN
ajst-2171	75	6	in	in	ADP
ajst-2171	75	7	recent	recent	ADJ
ajst-2171	75	8	literatures	literature	NOUN
ajst-2171	75	9	,	,	PUNCT
ajst-2171	75	10	the	the	DET
ajst-2171	75	11	method	method	NOUN
ajst-2171	75	12	in	in	ADP
ajst-2171	75	13	this	this	DET
ajst-2171	75	14	paper	paper	NOUN
ajst-2171	75	15	is	be	AUX
ajst-2171	75	16	relatively	relatively	ADV
ajst-2171	75	17	good	good	ADJ
ajst-2171	75	18	.	.	PUNCT
ajst-2171	76	1	the	the	DET
ajst-2171	76	2	experimental	experimental	ADJ
ajst-2171	76	3	comparison	comparison	NOUN
ajst-2171	76	4	results	result	NOUN
ajst-2171	76	5	are	be	AUX
ajst-2171	76	6	shown	show	VERB
ajst-2171	76	7	in	in	ADP
ajst-2171	76	8	the	the	DET
ajst-2171	76	9	following	following	NOUN
ajst-2171	76	10	,	,	PUNCT
ajst-2171	76	11	see	see	VERB
ajst-2171	76	12	table	table	NOUN
ajst-2171	76	13	1	1	NUM
ajst-2171	76	14	.	.	PUNCT
ajst-2171	76	15	table	table	NOUN
ajst-2171	76	16	1	1	NUM
ajst-2171	76	17	.	.	PUNCT
ajst-2171	77	1	experiment	experiment	NOUN
ajst-2171	77	2	result	result	PROPN
ajst-2171	77	3	model	model	PROPN
ajst-2171	77	4	acc	acc	PROPN
ajst-2171	77	5	(	(	PUNCT
ajst-2171	77	6	%	%	INTJ
ajst-2171	77	7	)	)	PUNCT
ajst-2171	77	8	f1	f1	NOUN
ajst-2171	77	9	-	-	PUNCT
ajst-2171	77	10	score	score	NOUN
ajst-2171	77	11	(	(	PUNCT
ajst-2171	77	12	%	%	INTJ
ajst-2171	77	13	)	)	PUNCT
ajst-2171	77	14	tedcsa[12	tedcsa[12	NOUN
ajst-2171	77	15	]	]	X
ajst-2171	77	16	79.40	79.40	NUM
ajst-2171	77	17	-	-	PUNCT
ajst-2171	77	18	catsnet[13	catsnet[13	NOUN
ajst-2171	77	19	]	]	PUNCT
ajst-2171	77	20	83.15	83.15	NUM
ajst-2171	77	21	-	-	PUNCT
ajst-2171	77	22	our	our	PRON
ajst-2171	77	23	method	method	NOUN
ajst-2171	77	24	84.14	84.14	NUM
ajst-2171	77	25	85.62	85.62	NUM
ajst-2171	77	26	193	193	NUM
ajst-2171	77	27	although	although	SCONJ
ajst-2171	77	28	the	the	DET
ajst-2171	77	29	performance	performance	NOUN
ajst-2171	77	30	of	of	ADP
ajst-2171	77	31	the	the	DET
ajst-2171	77	32	model	model	NOUN
ajst-2171	77	33	in	in	ADP
ajst-2171	77	34	this	this	DET
ajst-2171	77	35	paper	paper	NOUN
ajst-2171	77	36	can	can	AUX
ajst-2171	77	37	not	not	PART
ajst-2171	77	38	be	be	AUX
ajst-2171	77	39	compared	compare	VERB
ajst-2171	77	40	with	with	ADP
ajst-2171	77	41	the	the	DET
ajst-2171	77	42	large	large	ADJ
ajst-2171	77	43	pretraining	pretraine	VERB
ajst-2171	77	44	model	model	NOUN
ajst-2171	77	45	,	,	PUNCT
ajst-2171	77	46	it	it	PRON
ajst-2171	77	47	takes	take	VERB
ajst-2171	77	48	about	about	ADV
ajst-2171	77	49	20	20	NUM
ajst-2171	77	50	minutes	minute	NOUN
ajst-2171	77	51	to	to	PART
ajst-2171	77	52	train	train	VERB
ajst-2171	77	53	each	each	DET
ajst-2171	77	54	epoch	epoch	NOUN
ajst-2171	77	55	,	,	PUNCT
ajst-2171	77	56	with	with	ADP
ajst-2171	77	57	small	small	ADJ
ajst-2171	77	58	parameters	parameter	NOUN
ajst-2171	77	59	and	and	CCONJ
ajst-2171	77	60	high	high	ADJ
ajst-2171	77	61	efficiency	efficiency	NOUN
ajst-2171	77	62	.	.	PUNCT
ajst-2171	78	1	4	4	X
ajst-2171	78	2	.	.	X
ajst-2171	78	3	conclusion	conclusion	NOUN
ajst-2171	78	4	the	the	DET
ajst-2171	78	5	model	model	NOUN
ajst-2171	78	6	proposed	propose	VERB
ajst-2171	78	7	in	in	ADP
ajst-2171	78	8	this	this	DET
ajst-2171	78	9	paper	paper	NOUN
ajst-2171	78	10	is	be	AUX
ajst-2171	78	11	a	a	DET
ajst-2171	78	12	lightweight	lightweight	ADJ
ajst-2171	78	13	text	text	NOUN
ajst-2171	78	14	matching	matching	NOUN
ajst-2171	78	15	model	model	NOUN
ajst-2171	78	16	.	.	PUNCT
ajst-2171	79	1	the	the	DET
ajst-2171	79	2	model	model	NOUN
ajst-2171	79	3	is	be	AUX
ajst-2171	79	4	evaluated	evaluate	VERB
ajst-2171	79	5	on	on	ADP
ajst-2171	79	6	the	the	DET
ajst-2171	79	7	chinese	chinese	ADJ
ajst-2171	79	8	text	text	NOUN
ajst-2171	79	9	matching	matching	NOUN
ajst-2171	79	10	dataset	dataset	NOUN
ajst-2171	79	11	.	.	PUNCT
ajst-2171	80	1	the	the	DET
ajst-2171	80	2	results	result	NOUN
ajst-2171	80	3	show	show	VERB
ajst-2171	80	4	that	that	SCONJ
ajst-2171	80	5	the	the	DET
ajst-2171	80	6	operation	operation	NOUN
ajst-2171	80	7	of	of	ADP
ajst-2171	80	8	using	use	VERB
ajst-2171	80	9	convolution	convolution	NOUN
ajst-2171	80	10	with	with	ADP
ajst-2171	80	11	different	different	ADJ
ajst-2171	80	12	convolution	convolution	NOUN
ajst-2171	80	13	cores	core	NOUN
ajst-2171	80	14	to	to	PART
ajst-2171	80	15	extract	extract	VERB
ajst-2171	80	16	features	feature	NOUN
ajst-2171	80	17	and	and	CCONJ
ajst-2171	80	18	align	align	VERB
ajst-2171	80	19	them	they	PRON
ajst-2171	80	20	respectively	respectively	ADV
ajst-2171	80	21	is	be	AUX
ajst-2171	80	22	more	more	ADV
ajst-2171	80	23	sufficient	sufficient	ADJ
ajst-2171	80	24	to	to	PART
ajst-2171	80	25	mine	mine	VERB
ajst-2171	80	26	the	the	DET
ajst-2171	80	27	semantic	semantic	ADJ
ajst-2171	80	28	information	information	NOUN
ajst-2171	80	29	of	of	ADP
ajst-2171	80	30	the	the	DET
ajst-2171	80	31	text	text	NOUN
ajst-2171	80	32	.	.	PUNCT
ajst-2171	81	1	although	although	SCONJ
ajst-2171	81	2	the	the	DET
ajst-2171	81	3	effect	effect	NOUN
ajst-2171	81	4	of	of	ADP
ajst-2171	81	5	this	this	DET
ajst-2171	81	6	lightweight	lightweight	ADJ
ajst-2171	81	7	model	model	NOUN
ajst-2171	81	8	is	be	AUX
ajst-2171	81	9	still	still	ADV
ajst-2171	81	10	some	some	DET
ajst-2171	81	11	distance	distance	NOUN
ajst-2171	81	12	from	from	ADP
ajst-2171	81	13	large	large	ADJ
ajst-2171	81	14	pretraining	pretraine	VERB
ajst-2171	81	15	model	model	NOUN
ajst-2171	81	16	,	,	PUNCT
ajst-2171	81	17	the	the	DET
ajst-2171	81	18	training	training	NOUN
ajst-2171	81	19	time	time	NOUN
ajst-2171	81	20	of	of	ADP
ajst-2171	81	21	the	the	DET
ajst-2171	81	22	model	model	NOUN
ajst-2171	81	23	is	be	AUX
ajst-2171	81	24	short	short	ADJ
ajst-2171	81	25	and	and	CCONJ
ajst-2171	81	26	the	the	DET
ajst-2171	81	27	number	number	NOUN
ajst-2171	81	28	of	of	ADP
ajst-2171	81	29	parameters	parameter	NOUN
ajst-2171	81	30	is	be	AUX
ajst-2171	81	31	small	small	ADJ
ajst-2171	81	32	,	,	PUNCT
ajst-2171	81	33	it	it	PRON
ajst-2171	81	34	is	be	AUX
ajst-2171	81	35	convenient	convenient	ADJ
ajst-2171	81	36	for	for	ADP
ajst-2171	81	37	online	online	ADJ
ajst-2171	81	38	use	use	NOUN
ajst-2171	81	39	,	,	PUNCT
ajst-2171	81	40	so	so	CCONJ
ajst-2171	81	41	the	the	DET
ajst-2171	81	42	next	next	ADJ
ajst-2171	81	43	step	step	NOUN
ajst-2171	81	44	is	be	AUX
ajst-2171	81	45	to	to	PART
ajst-2171	81	46	train	train	VERB
ajst-2171	81	47	a	a	DET
ajst-2171	81	48	text	text	NOUN
ajst-2171	81	49	matching	matching	NOUN
ajst-2171	81	50	model	model	NOUN
ajst-2171	81	51	with	with	ADP
ajst-2171	81	52	higher	high	ADJ
ajst-2171	81	53	performance	performance	NOUN
ajst-2171	81	54	than	than	ADP
ajst-2171	81	55	the	the	DET
ajst-2171	81	56	large	large	ADJ
ajst-2171	81	57	pretraining	pretraine	VERB
ajst-2171	81	58	model	model	NOUN
ajst-2171	81	59	under	under	ADP
ajst-2171	81	60	the	the	DET
ajst-2171	81	61	premise	premise	NOUN
ajst-2171	81	62	of	of	ADP
ajst-2171	81	63	ensuring	ensure	VERB
ajst-2171	81	64	lightweight	lightweight	NOUN
ajst-2171	81	65	.	.	PUNCT
ajst-2171	82	1	references	reference	NOUN
ajst-2171	82	2	[	[	X
ajst-2171	82	3	1	1	X
ajst-2171	82	4	]	]	PUNCT
ajst-2171	82	5	pang	pang	PROPN
ajst-2171	82	6	liang	liang	PROPN
ajst-2171	82	7	,	,	PUNCT
ajst-2171	82	8	lan	lan	PROPN
ajst-2171	82	9	yanyan	yanyan	PROPN
ajst-2171	82	10	,	,	PUNCT
ajst-2171	82	11	xu	xu	PROPN
ajst-2171	82	12	jun	jun	PROPN
ajst-2171	82	13	,	,	PUNCT
ajst-2171	82	14	guo	guo	PROPN
ajst-2171	82	15	jiafeng	jiafeng	PROPN
ajst-2171	82	16	,	,	PUNCT
ajst-2171	82	17	wan	wan	PROPN
ajst-2171	82	18	shengxian	shengxian	PROPN
ajst-2171	82	19	,	,	PUNCT
ajst-2171	82	20	cheng	cheng	PROPN
ajst-2171	82	21	xueqi	xueqi	PROPN
ajst-2171	82	22	.	.	PUNCT
ajst-2171	83	1	overview	overview	NOUN
ajst-2171	83	2	of	of	ADP
ajst-2171	83	3	deep	deep	ADJ
ajst-2171	83	4	text	text	NOUN
ajst-2171	83	5	matching	matching	NOUN
ajst-2171	83	6	[	[	X
ajst-2171	83	7	j	j	X
ajst-2171	83	8	]	]	X
ajst-2171	83	9	.	.	PUNCT
ajst-2171	84	1	journal	journal	PROPN
ajst-2171	84	2	of	of	ADP
ajst-2171	84	3	computer	computer	NOUN
ajst-2171	84	4	science	science	NOUN
ajst-2171	84	5	,	,	PUNCT
ajst-2171	84	6	2017,40	2017,40	NUM
ajst-2171	84	7	(	(	PUNCT
ajst-2171	84	8	04	04	NUM
ajst-2171	84	9	):	):	PUNCT
ajst-2171	84	10	985	985	NUM
ajst-2171	84	11	-	-	SYM
ajst-2171	84	12	1003	1003	NUM
ajst-2171	84	13	.	.	PUNCT
ajst-2171	85	1	[	[	X
ajst-2171	85	2	2	2	X
ajst-2171	85	3	]	]	X
ajst-2171	85	4	huang	huang	PROPN
ajst-2171	85	5	p	p	PROPN
ajst-2171	85	6	s	s	PROPN
ajst-2171	85	7	,	,	PUNCT
ajst-2171	85	8	he	he	PRON
ajst-2171	85	9	x	x	PROPN
ajst-2171	85	10	,	,	PUNCT
ajst-2171	85	11	gao	gao	PROPN
ajst-2171	85	12	j	j	PROPN
ajst-2171	85	13	,	,	PUNCT
ajst-2171	85	14	et	et	PROPN
ajst-2171	85	15	al	al	PROPN
ajst-2171	85	16	.	.	PUNCT
ajst-2171	86	1	learning	learn	VERB
ajst-2171	86	2	deep	deep	ADJ
ajst-2171	86	3	structured	structured	ADJ
ajst-2171	86	4	semantic	semantic	ADJ
ajst-2171	86	5	models	model	NOUN
ajst-2171	86	6	for	for	ADP
ajst-2171	86	7	web	web	NOUN
ajst-2171	86	8	search	search	NOUN
ajst-2171	86	9	using	use	VERB
ajst-2171	86	10	clickthrough	clickthrough	NOUN
ajst-2171	86	11	data[c]//proceedings	data[c]//proceeding	NOUN
ajst-2171	86	12	of	of	ADP
ajst-2171	86	13	the	the	DET
ajst-2171	86	14	22nd	22nd	PROPN
ajst-2171	86	15	acm	acm	PROPN
ajst-2171	86	16	international	international	ADJ
ajst-2171	86	17	conference	conference	NOUN
ajst-2171	86	18	on	on	ADP
ajst-2171	86	19	information	information	NOUN
ajst-2171	86	20	&	&	CCONJ
ajst-2171	86	21	knowledge	knowledge	PROPN
ajst-2171	86	22	management	management	PROPN
ajst-2171	86	23	.	.	PUNCT
ajst-2171	87	1	2013	2013	NUM
ajst-2171	87	2	:	:	PUNCT
ajst-2171	87	3	2333	2333	NUM
ajst-2171	87	4	-	-	SYM
ajst-2171	87	5	2338	2338	NUM
ajst-2171	87	6	.	.	PUNCT
ajst-2171	88	1	[	[	X
ajst-2171	88	2	3	3	X
ajst-2171	88	3	]	]	X
ajst-2171	88	4	shen	shen	PROPN
ajst-2171	88	5	y	y	PROPN
ajst-2171	88	6	,	,	PUNCT
ajst-2171	88	7	he	he	PRON
ajst-2171	88	8	x	x	PROPN
ajst-2171	88	9	,	,	PUNCT
ajst-2171	88	10	gao	gao	PROPN
ajst-2171	88	11	j	j	PROPN
ajst-2171	88	12	,	,	PUNCT
ajst-2171	88	13	et	et	PROPN
ajst-2171	88	14	al	al	PROPN
ajst-2171	88	15	.	.	PUNCT
ajst-2171	89	1	a	a	DET
ajst-2171	89	2	latent	latent	ADJ
ajst-2171	89	3	semantic	semantic	ADJ
ajst-2171	89	4	model	model	NOUN
ajst-2171	89	5	with	with	ADP
ajst-2171	89	6	convolutional	convolutional	ADJ
ajst-2171	89	7	-	-	PUNCT
ajst-2171	89	8	pooling	pool	VERB
ajst-2171	89	9	structure	structure	NOUN
ajst-2171	89	10	for	for	ADP
ajst-2171	89	11	information	information	NOUN
ajst-2171	89	12	retrieval[c]//proceedings	retrieval[c]//proceeding	NOUN
ajst-2171	89	13	of	of	ADP
ajst-2171	89	14	the	the	DET
ajst-2171	89	15	23rd	23rd	ADJ
ajst-2171	89	16	acm	acm	PROPN
ajst-2171	89	17	international	international	ADJ
ajst-2171	89	18	conference	conference	NOUN
ajst-2171	89	19	on	on	ADP
ajst-2171	89	20	conference	conference	NOUN
ajst-2171	89	21	on	on	ADP
ajst-2171	89	22	information	information	NOUN
ajst-2171	89	23	and	and	CCONJ
ajst-2171	89	24	knowledge	knowledge	NOUN
ajst-2171	89	25	management	management	NOUN
ajst-2171	89	26	.	.	PUNCT
ajst-2171	90	1	2014	2014	NUM
ajst-2171	90	2	:	:	PUNCT
ajst-2171	91	1	101	101	NUM
ajst-2171	91	2	-	-	SYM
ajst-2171	91	3	110	110	NUM
ajst-2171	91	4	.	.	PUNCT
ajst-2171	92	1	[	[	X
ajst-2171	92	2	4	4	X
ajst-2171	92	3	]	]	X
ajst-2171	92	4	palangi	palangi	NOUN
ajst-2171	92	5	h	h	PROPN
ajst-2171	92	6	,	,	PUNCT
ajst-2171	92	7	deng	deng	PROPN
ajst-2171	92	8	l	l	PROPN
ajst-2171	92	9	,	,	PUNCT
ajst-2171	92	10	shen	shen	PROPN
ajst-2171	92	11	y	y	PROPN
ajst-2171	92	12	,	,	PUNCT
ajst-2171	92	13	et	et	PROPN
ajst-2171	92	14	al	al	PROPN
ajst-2171	92	15	.	.	PUNCT
ajst-2171	92	16	semantic	semantic	ADJ
ajst-2171	92	17	modelling	modelling	NOUN
ajst-2171	92	18	with	with	ADP
ajst-2171	92	19	long	long	ADJ
ajst-2171	92	20	-	-	PUNCT
ajst-2171	92	21	short	short	ADJ
ajst-2171	92	22	-	-	PUNCT
ajst-2171	92	23	term	term	NOUN
ajst-2171	92	24	memory	memory	NOUN
ajst-2171	92	25	for	for	ADP
ajst-2171	92	26	information	information	NOUN
ajst-2171	92	27	retrieval[j	retrieval[j	PROPN
ajst-2171	92	28	]	]	PUNCT
ajst-2171	92	29	.	.	PUNCT
ajst-2171	93	1	arxiv	arxiv	PROPN
ajst-2171	93	2	preprint	preprint	PROPN
ajst-2171	93	3	arxiv:1412.6629	arxiv:1412.6629	NOUN
ajst-2171	93	4	,	,	PUNCT
ajst-2171	93	5	2014	2014	NUM
ajst-2171	93	6	.	.	PUNCT
ajst-2171	94	1	[	[	X
ajst-2171	94	2	5	5	X
ajst-2171	94	3	]	]	PUNCT
ajst-2171	94	4	pang	pang	NOUN
ajst-2171	94	5	l	l	NOUN
ajst-2171	94	6	,	,	PUNCT
ajst-2171	94	7	lan	lan	PROPN
ajst-2171	94	8	y	y	PROPN
ajst-2171	94	9	,	,	PUNCT
ajst-2171	94	10	guo	guo	PROPN
ajst-2171	94	11	j	j	PROPN
ajst-2171	94	12	,	,	PUNCT
ajst-2171	94	13	et	et	PROPN
ajst-2171	94	14	al	al	PROPN
ajst-2171	94	15	.	.	PROPN
ajst-2171	94	16	text	text	NOUN
ajst-2171	94	17	matching	matching	NOUN
ajst-2171	94	18	as	as	ADP
ajst-2171	94	19	image	image	NOUN
ajst-2171	94	20	recognition[c]//proceedings	recognition[c]//proceeding	NOUN
ajst-2171	94	21	of	of	ADP
ajst-2171	94	22	the	the	DET
ajst-2171	94	23	aaai	aaai	PROPN
ajst-2171	94	24	conference	conference	NOUN
ajst-2171	94	25	on	on	ADP
ajst-2171	94	26	artificial	artificial	ADJ
ajst-2171	94	27	intelligence	intelligence	NOUN
ajst-2171	94	28	.	.	PUNCT
ajst-2171	95	1	2016	2016	NUM
ajst-2171	95	2	,	,	PUNCT
ajst-2171	95	3	30(1	30(1	NUM
ajst-2171	95	4	)	)	PUNCT
ajst-2171	95	5	.	.	PUNCT
ajst-2171	96	1	[	[	X
ajst-2171	96	2	6	6	NUM
ajst-2171	96	3	]	]	X
ajst-2171	96	4	chen	chen	PROPN
ajst-2171	96	5	q	q	X
ajst-2171	96	6	,	,	PUNCT
ajst-2171	96	7	zhu	zhu	PROPN
ajst-2171	96	8	x	x	X
ajst-2171	96	9	,	,	PUNCT
ajst-2171	96	10	ling	ling	PROPN
ajst-2171	96	11	z	z	PROPN
ajst-2171	96	12	,	,	PUNCT
ajst-2171	96	13	et	et	PROPN
ajst-2171	96	14	al	al	PROPN
ajst-2171	96	15	.	.	PROPN
ajst-2171	96	16	enhanced	enhance	VERB
ajst-2171	96	17	lstm	lstm	NOUN
ajst-2171	96	18	for	for	ADP
ajst-2171	96	19	natural	natural	ADJ
ajst-2171	96	20	language	language	NOUN
ajst-2171	96	21	inference[j	inference[j	NOUN
ajst-2171	96	22	]	]	PUNCT
ajst-2171	96	23	.	.	PUNCT
ajst-2171	97	1	arxiv	arxiv	PROPN
ajst-2171	97	2	preprint	preprint	PROPN
ajst-2171	97	3	arxiv:1609.06038	arxiv:1609.06038	PROPN
ajst-2171	97	4	,	,	PUNCT
ajst-2171	97	5	2016	2016	NUM
ajst-2171	97	6	.	.	PUNCT
ajst-2171	98	1	[	[	X
ajst-2171	98	2	7	7	X
ajst-2171	98	3	]	]	X
ajst-2171	98	4	wang	wang	PROPN
ajst-2171	98	5	z	z	PROPN
ajst-2171	98	6	,	,	PUNCT
ajst-2171	98	7	hamza	hamza	PROPN
ajst-2171	98	8	w	w	PROPN
ajst-2171	98	9	,	,	PUNCT
ajst-2171	98	10	florian	florian	PROPN
ajst-2171	98	11	r.	r.	PROPN
ajst-2171	98	12	bilateral	bilateral	PROPN
ajst-2171	98	13	multi	multi	ADJ
ajst-2171	98	14	-	-	NOUN
ajst-2171	98	15	perspective	perspective	ADJ
ajst-2171	98	16	matching	matching	NOUN
ajst-2171	98	17	for	for	ADP
ajst-2171	98	18	natural	natural	ADJ
ajst-2171	98	19	language	language	NOUN
ajst-2171	98	20	sentences[j	sentences[j	NOUN
ajst-2171	98	21	]	]	PUNCT
ajst-2171	98	22	.	.	PUNCT
ajst-2171	99	1	arxiv	arxiv	PROPN
ajst-2171	99	2	preprint	preprint	NOUN
ajst-2171	99	3	arxiv:1702.03814	arxiv:1702.03814	NOUN
ajst-2171	99	4	,	,	PUNCT
ajst-2171	99	5	2017	2017	NUM
ajst-2171	99	6	.	.	PUNCT
ajst-2171	100	1	[	[	X
ajst-2171	100	2	8	8	NUM
ajst-2171	100	3	]	]	SYM
ajst-2171	100	4	yang	yang	PROPN
ajst-2171	100	5	runqi	runqi	PROPN
ajst-2171	100	6	,	,	PUNCT
ajst-2171	100	7	zhang	zhang	PROPN
ajst-2171	100	8	jianhai	jianhai	PROPN
ajst-2171	100	9	,	,	PUNCT
ajst-2171	100	10	gao	gao	PROPN
ajst-2171	100	11	xing	xing	PROPN
ajst-2171	100	12	,	,	PUNCT
ajst-2171	100	13	ji	ji	PROPN
ajst-2171	100	14	feng	feng	PROPN
ajst-2171	100	15	,	,	PUNCT
ajst-2171	100	16	chen	chen	PROPN
ajst-2171	100	17	haiqing	haiqing	PROPN
ajst-2171	100	18	.	.	PUNCT
ajst-2171	101	1	(	(	PUNCT
ajst-2171	101	2	2019	2019	NUM
ajst-2171	101	3	)	)	PUNCT
ajst-2171	101	4	.	.	PUNCT
ajst-2171	102	1	simple	simple	ADJ
ajst-2171	102	2	and	and	CCONJ
ajst-2171	102	3	effective	effective	ADJ
ajst-2171	102	4	text	text	NOUN
ajst-2171	102	5	matching	matching	NOUN
ajst-2171	102	6	with	with	ADP
ajst-2171	102	7	richer	rich	ADJ
ajst-2171	102	8	alignment	alignment	NOUN
ajst-2171	102	9	features	feature	NOUN
ajst-2171	102	10	.	.	PUNCT
ajst-2171	103	1	4699	4699	NUM
ajst-2171	103	2	-	-	SYM
ajst-2171	103	3	4709	4709	NUM
ajst-2171	103	4	.	.	PUNCT
ajst-2171	104	1	10.18653	10.18653	NUM
ajst-2171	104	2	/	/	SYM
ajst-2171	104	3	v1	v1	NOUN
ajst-2171	104	4	/	/	SYM
ajst-2171	104	5	p19	p19	NOUN
ajst-2171	104	6	-	-	PUNCT
ajst-2171	104	7	1465	1465	NUM
ajst-2171	104	8	.	.	PUNCT
ajst-2171	105	1	[	[	X
ajst-2171	105	2	9	9	NUM
ajst-2171	105	3	]	]	PUNCT
ajst-2171	105	4	devlin	devlin	PROPN
ajst-2171	105	5	j	j	PROPN
ajst-2171	105	6	,	,	PUNCT
ajst-2171	105	7	chang	chang	PROPN
ajst-2171	105	8	m	m	PROPN
ajst-2171	105	9	w	w	PROPN
ajst-2171	105	10	,	,	PUNCT
ajst-2171	105	11	lee	lee	PROPN
ajst-2171	105	12	k	k	PROPN
ajst-2171	105	13	,	,	PUNCT
ajst-2171	105	14	et	et	PROPN
ajst-2171	105	15	al	al	PROPN
ajst-2171	105	16	.	.	PUNCT
ajst-2171	105	17	bert	bert	PROPN
ajst-2171	105	18	:	:	PUNCT
ajst-2171	105	19	pre	pre	ADJ
ajst-2171	105	20	-	-	NOUN
ajst-2171	105	21	training	training	NOUN
ajst-2171	105	22	of	of	ADP
ajst-2171	105	23	deep	deep	ADJ
ajst-2171	105	24	bidirectional	bidirectional	ADJ
ajst-2171	105	25	transformers	transformer	NOUN
ajst-2171	105	26	for	for	ADP
ajst-2171	105	27	language	language	NOUN
ajst-2171	105	28	understanding[j	understanding[j	NOUN
ajst-2171	105	29	]	]	PUNCT
ajst-2171	105	30	.	.	PUNCT
ajst-2171	106	1	arxiv	arxiv	PROPN
ajst-2171	106	2	preprint	preprint	PROPN
ajst-2171	106	3	arxiv:1810.04805	arxiv:1810.04805	PROPN
ajst-2171	106	4	,	,	PUNCT
ajst-2171	106	5	2018	2018	NUM
ajst-2171	106	6	.	.	PUNCT
ajst-2171	107	1	[	[	X
ajst-2171	107	2	10	10	NUM
ajst-2171	107	3	]	]	X
ajst-2171	107	4	lcqmc	lcqmc	NOUN
ajst-2171	107	5	:	:	PUNCT
ajst-2171	107	6	a	a	DET
ajst-2171	107	7	large	large	ADJ
ajst-2171	107	8	-	-	PUNCT
ajst-2171	107	9	scale	scale	NOUN
ajst-2171	107	10	chinese	chinese	ADJ
ajst-2171	107	11	question	question	NOUN
ajst-2171	107	12	matching	match	VERB
ajst-2171	107	13	corpus[c]//proceedings	corpus[c]//proceeding	NOUN
ajst-2171	107	14	of	of	ADP
ajst-2171	107	15	the	the	DET
ajst-2171	107	16	27th	27th	ADJ
ajst-2171	107	17	international	international	ADJ
ajst-2171	107	18	conference	conference	NOUN
ajst-2171	107	19	on	on	ADP
ajst-2171	107	20	computational	computational	ADJ
ajst-2171	107	21	linguistics	linguistic	NOUN
ajst-2171	107	22	.	.	PUNCT
ajst-2171	108	1	2018	2018	NUM
ajst-2171	108	2	:	:	PUNCT
ajst-2171	108	3	1952	1952	NUM
ajst-2171	108	4	-	-	SYM
ajst-2171	108	5	1962	1962	NUM
ajst-2171	108	6	.	.	PUNCT
ajst-2171	109	1	[	[	X
ajst-2171	109	2	11	11	NUM
ajst-2171	109	3	]	]	PUNCT
ajst-2171	109	4	cui	cui	X
ajst-2171	109	5	y	y	PROPN
ajst-2171	109	6	,	,	PUNCT
ajst-2171	109	7	che	che	PROPN
ajst-2171	109	8	w	w	PROPN
ajst-2171	109	9	,	,	PUNCT
ajst-2171	109	10	liu	liu	PROPN
ajst-2171	109	11	t	t	PROPN
ajst-2171	109	12	,	,	PUNCT
ajst-2171	109	13	et	et	PROPN
ajst-2171	109	14	al	al	PROPN
ajst-2171	109	15	.	.	PUNCT
ajst-2171	109	16	pre	pre	VERB
ajst-2171	109	17	-	-	VERB
ajst-2171	109	18	training	training	NOUN
ajst-2171	109	19	with	with	ADP
ajst-2171	109	20	whole	whole	ADJ
ajst-2171	109	21	word	word	NOUN
ajst-2171	109	22	masking	mask	VERB
ajst-2171	109	23	for	for	ADP
ajst-2171	109	24	chinese	chinese	ADJ
ajst-2171	109	25	bert[j	bert[j	NOUN
ajst-2171	109	26	]	]	PUNCT
ajst-2171	109	27	.	.	PUNCT
ajst-2171	110	1	ieee	ieee	PROPN
ajst-2171	110	2	/	/	SYM
ajst-2171	110	3	acm	acm	PROPN
ajst-2171	110	4	transactions	transaction	NOUN
ajst-2171	110	5	on	on	ADP
ajst-2171	110	6	audio	audio	NOUN
ajst-2171	110	7	,	,	PUNCT
ajst-2171	110	8	speech	speech	NOUN
ajst-2171	110	9	,	,	PUNCT
ajst-2171	110	10	and	and	CCONJ
ajst-2171	110	11	language	language	NOUN
ajst-2171	110	12	processing	processing	NOUN
ajst-2171	110	13	,	,	PUNCT
ajst-2171	110	14	2021	2021	NUM
ajst-2171	110	15	,	,	PUNCT
ajst-2171	110	16	29	29	NUM
ajst-2171	110	17	:	:	SYM
ajst-2171	110	18	3504	3504	NUM
ajst-2171	110	19	-	-	SYM
ajst-2171	110	20	3514	3514	NUM
ajst-2171	110	21	.	.	PUNCT
ajst-2171	111	1	[	[	X
ajst-2171	111	2	12	12	NUM
ajst-2171	111	3	]	]	PUNCT
ajst-2171	111	4	x.	x.	PROPN
ajst-2171	111	5	zhang	zhang	PROPN
ajst-2171	111	6	,	,	PUNCT
ajst-2171	111	7	f.	f.	PROPN
ajst-2171	111	8	lei	lei	PROPN
ajst-2171	111	9	and	and	CCONJ
ajst-2171	111	10	s.	s.	PROPN
ajst-2171	111	11	yu	yu	PROPN
ajst-2171	111	12	,	,	PUNCT
ajst-2171	111	13	"	"	PUNCT
ajst-2171	111	14	self	self	NOUN
ajst-2171	111	15	-	-	PUNCT
ajst-2171	111	16	attention	attention	NOUN
ajst-2171	111	17	based	base	VERB
ajst-2171	111	18	text	text	NOUN
ajst-2171	111	19	matching	matching	NOUN
ajst-2171	111	20	model	model	NOUN
ajst-2171	111	21	with	with	ADP
ajst-2171	111	22	generative	generative	ADJ
ajst-2171	111	23	pre	pre	ADJ
ajst-2171	111	24	-	-	NOUN
ajst-2171	111	25	training	training	ADJ
ajst-2171	111	26	,	,	PUNCT
ajst-2171	111	27	"	"	PUNCT
ajst-2171	111	28	2021	2021	NUM
ajst-2171	111	29	,	,	PUNCT
ajst-2171	111	30	pp	pp	ADP
ajst-2171	111	31	.	.	PUNCT
ajst-2171	112	1	8491,doi:10.1109	8491,doi:10.1109	ADJ
ajst-2171	112	2	/	/	SYM
ajst-2171	112	3	dasc	dasc	NOUN
ajst-2171	112	4	-	-	PUNCT
ajst-2171	112	5	picom	picom	NOUN
ajst-2171	112	6	-	-	PUNCT
ajst-2171	112	7	cbdcomcyberscitech52372.2021.00027	cbdcomcyberscitech52372.2021.00027	PROPN
ajst-2171	112	8	.	.	PUNCT
ajst-2171	113	1	[	[	X
ajst-2171	113	2	13	13	NUM
ajst-2171	113	3	]	]	X
ajst-2171	113	4	zhen	zhen	PROPN
ajst-2171	113	5	wang	wang	PROPN
ajst-2171	113	6	,	,	PUNCT
ajst-2171	113	7	xiangxie	xiangxie	PROPN
ajst-2171	113	8	zhang	zhang	PROPN
ajst-2171	113	9	,	,	PUNCT
ajst-2171	113	10	yicong	yicong	PROPN
ajst-2171	113	11	tan	tan	PROPN
ajst-2171	113	12	.	.	PUNCT
ajst-2171	114	1	chinese	chinese	ADJ
ajst-2171	114	2	sentences	sentence	VERB
ajst-2171	114	3	similarity	similarity	NOUN
ajst-2171	114	4	via	via	ADP
ajst-2171	114	5	cross	cross	ADJ
ajst-2171	114	6	-	-	ADJ
ajst-2171	114	7	attention	attention	ADJ
ajst-2171	114	8	based	base	VERB
ajst-2171	114	9	siamese	siamese	ADJ
ajst-2171	114	10	network.2021	network.2021	PROPN
ajst-2171	114	11	,	,	PUNCT
ajst-2171	114	12	https://doi.org/10.48550/arxiv.2104.08787	https://doi.org/10.48550/arxiv.2104.08787	PROPN