id	sid	tid	token	lemma	pos
orr-bootleg-2020	1	1	Bootleg	bootleg	NN
orr-bootleg-2020	1	2	:	:	:
orr-bootleg-2020	1	3	Chasing	chase	VBG
orr-bootleg-2020	1	4	the	the	DT
orr-bootleg-2020	1	5	Tail	Tail	NNP
orr-bootleg-2020	1	6	with	with	IN
orr-bootleg-2020	1	7	Self	self	NN
orr-bootleg-2020	1	8	-	-	HYPH
orr-bootleg-2020	1	9	Supervised	supervise	VBN
orr-bootleg-2020	1	10	Named	Named	NNP
orr-bootleg-2020	1	11	Entity	Entity	NNP
orr-bootleg-2020	1	12	Disambiguation	Disambiguation	NNP
orr-bootleg-2020	1	13	Laurel	Laurel	NNP
orr-bootleg-2020	1	14	Orr†	Orr†	NNP
orr-bootleg-2020	1	15	,	,	,
orr-bootleg-2020	1	16	Megan	Megan	NNP
orr-bootleg-2020	1	17	Leszczynski†	Leszczynski†	NNP
orr-bootleg-2020	1	18	,	,	,
orr-bootleg-2020	1	19	Simran	Simran	NNP
orr-bootleg-2020	1	20	Arora†	arora†	ADD
orr-bootleg-2020	1	21	,	,	,
orr-bootleg-2020	1	22	Sen	Sen	NNP
orr-bootleg-2020	1	23	Wu†	wu†	ADD
orr-bootleg-2020	1	24	,	,	,
orr-bootleg-2020	1	25	Neel	Neel	NNP
orr-bootleg-2020	1	26	Guha†	Guha†	NNS
orr-bootleg-2020	1	27	,	,	,
orr-bootleg-2020	1	28	Xiao	Xiao	NNP
orr-bootleg-2020	1	29	Ling‡	Ling‡	NNS
orr-bootleg-2020	1	30	,	,	,
orr-bootleg-2020	1	31	and	and	CC
orr-bootleg-2020	1	32	Christopher	Christopher	NNP
orr-bootleg-2020	1	33	Ré†	ré†	ADD
orr-bootleg-2020	1	34	†Stanford	†Stanford	NNP
orr-bootleg-2020	1	35	University	University	NNP
orr-bootleg-2020	1	36	‡Apple	‡Apple	NNP
orr-bootleg-2020	1	37	{	{	-LRB-
orr-bootleg-2020	1	38	lorr1,mleszczy,simran,senwu,nguha,chrismre}@cs.stanford.edu	lorr1,mleszczy,simran,senwu,nguha,chrismre}@cs.stanford.edu	NNP
orr-bootleg-2020	1	39	,	,	,
orr-bootleg-2020	1	40	xiaoling@apple.com	xiaoling@apple.com	ADD
orr-bootleg-2020	1	41	Abstract	Abstract	NNP
orr-bootleg-2020	1	42	A	a	DT
orr-bootleg-2020	1	43	challenge	challenge	NN
orr-bootleg-2020	1	44	for	for	IN
orr-bootleg-2020	1	45	named	name	VBN
orr-bootleg-2020	1	46	entity	entity	NN
orr-bootleg-2020	1	47	disambiguation	disambiguation	NN
orr-bootleg-2020	1	48	(	(	-LRB-
orr-bootleg-2020	1	49	NED	NED	NNP
orr-bootleg-2020	1	50	)	)	-RRB-
orr-bootleg-2020	1	51	,	,	,
orr-bootleg-2020	1	52	the	the	DT
orr-bootleg-2020	1	53	task	task	NN
orr-bootleg-2020	1	54	of	of	IN
orr-bootleg-2020	1	55	mapping	map	VBG
orr-bootleg-2020	1	56	textual	textual	JJ
orr-bootleg-2020	1	57	mentions	mention	NNS
orr-bootleg-2020	1	58	to	to	IN
orr-bootleg-2020	1	59	entities	entity	NNS
orr-bootleg-2020	1	60	in	in	IN
orr-bootleg-2020	1	61	a	a	DT
orr-bootleg-2020	1	62	knowledge	knowledge	NN
orr-bootleg-2020	1	63	base	base	NN
orr-bootleg-2020	1	64	,	,	,
orr-bootleg-2020	1	65	is	be	VBZ
orr-bootleg-2020	1	66	how	how	WRB
orr-bootleg-2020	1	67	to	to	TO
orr-bootleg-2020	1	68	disambiguate	disambiguate	VB
orr-bootleg-2020	1	69	entities	entity	NNS
orr-bootleg-2020	1	70	that	that	WDT
orr-bootleg-2020	1	71	appear	appear	VBP
orr-bootleg-2020	1	72	rarely	rarely	RB
orr-bootleg-2020	1	73	in	in	IN
orr-bootleg-2020	1	74	the	the	DT
orr-bootleg-2020	1	75	training	training	NN
orr-bootleg-2020	1	76	data	datum	NNS
orr-bootleg-2020	1	77	,	,	,
orr-bootleg-2020	1	78	termed	term	VBN
orr-bootleg-2020	1	79	tail	tail	NN
orr-bootleg-2020	1	80	entities	entity	NNS
orr-bootleg-2020	1	81	.	.	.
orr-bootleg-2020	2	1	Humans	human	NNS
orr-bootleg-2020	2	2	use	use	VBP
orr-bootleg-2020	2	3	subtle	subtle	JJ
orr-bootleg-2020	2	4	reasoning	reasoning	NN
orr-bootleg-2020	2	5	patterns	pattern	NNS
orr-bootleg-2020	2	6	based	base	VBN
orr-bootleg-2020	2	7	on	on	IN
orr-bootleg-2020	2	8	knowledge	knowledge	NN
orr-bootleg-2020	2	9	of	of	IN
orr-bootleg-2020	2	10	entity	entity	NN
orr-bootleg-2020	2	11	facts	fact	NNS
orr-bootleg-2020	2	12	,	,	,
orr-bootleg-2020	2	13	relations	relation	NNS
orr-bootleg-2020	2	14	,	,	,
orr-bootleg-2020	2	15	and	and	CC
orr-bootleg-2020	2	16	types	type	NNS
orr-bootleg-2020	2	17	to	to	TO
orr-bootleg-2020	2	18	disambiguate	disambiguate	VB
orr-bootleg-2020	2	19	unfamiliar	unfamiliar	JJ
orr-bootleg-2020	2	20	entities	entity	NNS
orr-bootleg-2020	2	21	.	.	.
orr-bootleg-2020	3	1	Inspired	inspire	VBN
orr-bootleg-2020	3	2	by	by	IN
orr-bootleg-2020	3	3	these	these	DT
orr-bootleg-2020	3	4	patterns	pattern	NNS
orr-bootleg-2020	3	5	,	,	,
orr-bootleg-2020	3	6	we	-PRON-	PRP
orr-bootleg-2020	3	7	introduce	introduce	VBP
orr-bootleg-2020	3	8	Bootleg	Bootleg	NNP
orr-bootleg-2020	3	9	,	,	,
orr-bootleg-2020	3	10	a	a	DT
orr-bootleg-2020	3	11	self	self	NN
orr-bootleg-2020	3	12	-	-	HYPH
orr-bootleg-2020	3	13	supervised	supervise	VBN
orr-bootleg-2020	3	14	NED	NED	NNP
orr-bootleg-2020	3	15	system	system	NN
orr-bootleg-2020	3	16	that	that	WDT
orr-bootleg-2020	3	17	is	be	VBZ
orr-bootleg-2020	3	18	explicitly	explicitly	RB
orr-bootleg-2020	3	19	grounded	ground	VBN
orr-bootleg-2020	3	20	in	in	IN
orr-bootleg-2020	3	21	reasoning	reason	VBG
orr-bootleg-2020	3	22	patterns	pattern	NNS
orr-bootleg-2020	3	23	for	for	IN
orr-bootleg-2020	3	24	disambiguation	disambiguation	NN
orr-bootleg-2020	3	25	.	.	.
orr-bootleg-2020	4	1	We	-PRON-	PRP
orr-bootleg-2020	4	2	define	define	VBP
orr-bootleg-2020	4	3	core	core	NN
orr-bootleg-2020	4	4	reasoning	reason	VBG
orr-bootleg-2020	4	5	patterns	pattern	NNS
orr-bootleg-2020	4	6	for	for	IN
orr-bootleg-2020	4	7	disambiguation	disambiguation	NN
orr-bootleg-2020	4	8	,	,	,
orr-bootleg-2020	4	9	create	create	VB
orr-bootleg-2020	4	10	a	a	DT
orr-bootleg-2020	4	11	learning	learn	VBG
orr-bootleg-2020	4	12	procedure	procedure	NN
orr-bootleg-2020	4	13	to	to	TO
orr-bootleg-2020	4	14	encourage	encourage	VB
orr-bootleg-2020	4	15	the	the	DT
orr-bootleg-2020	4	16	self	self	NN
orr-bootleg-2020	4	17	-	-	HYPH
orr-bootleg-2020	4	18	supervised	supervise	VBN
orr-bootleg-2020	4	19	model	model	NN
orr-bootleg-2020	4	20	to	to	TO
orr-bootleg-2020	4	21	learn	learn	VB
orr-bootleg-2020	4	22	the	the	DT
orr-bootleg-2020	4	23	patterns	pattern	NNS
orr-bootleg-2020	4	24	,	,	,
orr-bootleg-2020	4	25	and	and	CC
orr-bootleg-2020	4	26	show	show	VB
orr-bootleg-2020	4	27	how	how	WRB
orr-bootleg-2020	4	28	to	to	TO
orr-bootleg-2020	4	29	use	use	VB
orr-bootleg-2020	4	30	weak	weak	JJ
orr-bootleg-2020	4	31	supervision	supervision	NN
orr-bootleg-2020	4	32	to	to	TO
orr-bootleg-2020	4	33	enhance	enhance	VB
orr-bootleg-2020	4	34	the	the	DT
orr-bootleg-2020	4	35	signals	signal	NNS
orr-bootleg-2020	4	36	in	in	IN
orr-bootleg-2020	4	37	the	the	DT
orr-bootleg-2020	4	38	training	training	NN
orr-bootleg-2020	4	39	data	datum	NNS
orr-bootleg-2020	4	40	.	.	.
orr-bootleg-2020	5	1	Encoding	encode	VBG
orr-bootleg-2020	5	2	the	the	DT
orr-bootleg-2020	5	3	reasoning	reasoning	NN
orr-bootleg-2020	5	4	patterns	pattern	NNS
orr-bootleg-2020	5	5	in	in	IN
orr-bootleg-2020	5	6	a	a	DT
orr-bootleg-2020	5	7	simple	simple	JJ
orr-bootleg-2020	5	8	Transformer	transformer	NN
orr-bootleg-2020	5	9	architecture	architecture	NN
orr-bootleg-2020	5	10	,	,	,
orr-bootleg-2020	5	11	Bootleg	Bootleg	NNP
orr-bootleg-2020	5	12	meets	meet	VBZ
orr-bootleg-2020	5	13	or	or	CC
orr-bootleg-2020	5	14	exceeds	exceed	VBZ
orr-bootleg-2020	5	15	state	state	NN
orr-bootleg-2020	5	16	-	-	HYPH
orr-bootleg-2020	5	17	of	of	IN
orr-bootleg-2020	5	18	-	-	HYPH
orr-bootleg-2020	5	19	the	the	DT
orr-bootleg-2020	5	20	-	-	HYPH
orr-bootleg-2020	5	21	art	art	NN
orr-bootleg-2020	5	22	on	on	IN
orr-bootleg-2020	5	23	three	three	CD
orr-bootleg-2020	5	24	NED	NED	NNP
orr-bootleg-2020	5	25	benchmarks	benchmark	NNS
orr-bootleg-2020	5	26	.	.	.
orr-bootleg-2020	6	1	We	-PRON-	PRP
orr-bootleg-2020	6	2	further	further	RB
orr-bootleg-2020	6	3	show	show	VBP
orr-bootleg-2020	6	4	that	that	IN
orr-bootleg-2020	6	5	the	the	DT
orr-bootleg-2020	6	6	learned	learn	VBN
orr-bootleg-2020	6	7	representations	representation	NNS
orr-bootleg-2020	6	8	from	from	IN
orr-bootleg-2020	6	9	Bootleg	Bootleg	NNP
orr-bootleg-2020	6	10	successfully	successfully	RB
orr-bootleg-2020	6	11	transfer	transfer	VBP
orr-bootleg-2020	6	12	to	to	IN
orr-bootleg-2020	6	13	other	other	JJ
orr-bootleg-2020	6	14	non	non	JJ
orr-bootleg-2020	6	15	-	-	JJ
orr-bootleg-2020	6	16	disambiguation	disambiguation	JJ
orr-bootleg-2020	6	17	tasks	task	NNS
orr-bootleg-2020	6	18	that	that	WDT
orr-bootleg-2020	6	19	require	require	VBP
orr-bootleg-2020	6	20	entity	entity	NN
orr-bootleg-2020	6	21	-	-	HYPH
orr-bootleg-2020	6	22	based	base	VBN
orr-bootleg-2020	6	23	knowledge	knowledge	NN
orr-bootleg-2020	6	24	:	:	:
orr-bootleg-2020	6	25	we	-PRON-	PRP
orr-bootleg-2020	6	26	set	set	VBP
orr-bootleg-2020	6	27	a	a	DT
orr-bootleg-2020	6	28	new	new	JJ
orr-bootleg-2020	6	29	state	state	NN
orr-bootleg-2020	6	30	-	-	HYPH
orr-bootleg-2020	6	31	of-	of-	VBN
orr-bootleg-2020	6	32	the	the	DT
orr-bootleg-2020	6	33	-	-	HYPH
orr-bootleg-2020	6	34	art	art	NN
orr-bootleg-2020	6	35	in	in	IN
orr-bootleg-2020	6	36	the	the	DT
orr-bootleg-2020	6	37	popular	popular	JJ
orr-bootleg-2020	6	38	TACRED	TACRED	NNP
orr-bootleg-2020	6	39	relation	relation	NN
orr-bootleg-2020	6	40	extraction	extraction	NN
orr-bootleg-2020	6	41	task	task	NN
orr-bootleg-2020	6	42	by	by	IN
orr-bootleg-2020	6	43	1.0	1.0	CD
orr-bootleg-2020	6	44	F1	F1	NNP
orr-bootleg-2020	6	45	points	point	NNS
orr-bootleg-2020	6	46	and	and	CC
orr-bootleg-2020	6	47	demonstrate	demonstrate	VB
orr-bootleg-2020	6	48	up	up	RP
orr-bootleg-2020	6	49	to	to	TO
orr-bootleg-2020	6	50	8	8	CD
orr-bootleg-2020	6	51	%	%	NN
orr-bootleg-2020	6	52	performance	performance	NN
orr-bootleg-2020	6	53	lift	lift	NN
orr-bootleg-2020	6	54	in	in	IN
orr-bootleg-2020	6	55	highly	highly	RB
orr-bootleg-2020	6	56	optimized	optimize	VBN
orr-bootleg-2020	6	57	production	production	NN
orr-bootleg-2020	6	58	search	search	NN
orr-bootleg-2020	6	59	and	and	CC
orr-bootleg-2020	6	60	assistant	assistant	NN
orr-bootleg-2020	6	61	tasks	task	NNS
orr-bootleg-2020	6	62	at	at	IN
orr-bootleg-2020	6	63	a	a	DT
orr-bootleg-2020	6	64	major	major	JJ
orr-bootleg-2020	6	65	technology	technology	NN
orr-bootleg-2020	6	66	company	company	NN
orr-bootleg-2020	6	67	.	.	.
orr-bootleg-2020	7	1	1	1	CD
orr-bootleg-2020	7	2	Introduction	Introduction	NNP
orr-bootleg-2020	7	3	Knowledge	knowledge	NN
orr-bootleg-2020	7	4	-	-	HYPH
orr-bootleg-2020	7	5	aware	aware	JJ
orr-bootleg-2020	7	6	deep	deep	JJ
orr-bootleg-2020	7	7	learning	learning	NN
orr-bootleg-2020	7	8	models	model	NNS
orr-bootleg-2020	7	9	have	have	VBP
orr-bootleg-2020	7	10	recently	recently	RB
orr-bootleg-2020	7	11	led	lead	VBN
orr-bootleg-2020	7	12	to	to	IN
orr-bootleg-2020	7	13	significant	significant	JJ
orr-bootleg-2020	7	14	progress	progress	NN
orr-bootleg-2020	7	15	in	in	IN
orr-bootleg-2020	7	16	fields	field	NNS
orr-bootleg-2020	7	17	ranging	range	VBG
orr-bootleg-2020	7	18	from	from	IN
orr-bootleg-2020	7	19	natural	natural	JJ
orr-bootleg-2020	7	20	language	language	NN
orr-bootleg-2020	7	21	understanding	understanding	NN
orr-bootleg-2020	7	22	[	[	-LRB-
orr-bootleg-2020	7	23	38	38	CD
orr-bootleg-2020	7	24	,	,	,
orr-bootleg-2020	7	25	41	41	CD
orr-bootleg-2020	7	26	]	]	-RRB-
orr-bootleg-2020	7	27	to	to	TO
orr-bootleg-2020	7	28	computer	computer	NN
orr-bootleg-2020	7	29	vision	vision	NN
orr-bootleg-2020	7	30	[	[	-LRB-
orr-bootleg-2020	7	31	56	56	CD
orr-bootleg-2020	7	32	]	]	-RRB-
orr-bootleg-2020	7	33	.	.	.
orr-bootleg-2020	8	1	Incorporating	incorporate	VBG
orr-bootleg-2020	8	2	explicit	explicit	JJ
orr-bootleg-2020	8	3	knowledge	knowledge	NN
orr-bootleg-2020	8	4	allows	allow	VBZ
orr-bootleg-2020	8	5	for	for	IN
orr-bootleg-2020	8	6	models	model	NNS
orr-bootleg-2020	8	7	to	to	TO
orr-bootleg-2020	8	8	better	better	RB
orr-bootleg-2020	8	9	recall	recall	VB
orr-bootleg-2020	8	10	factual	factual	JJ
orr-bootleg-2020	8	11	information	information	NN
orr-bootleg-2020	8	12	about	about	IN
orr-bootleg-2020	8	13	specific	specific	JJ
orr-bootleg-2020	8	14	entities	entity	NNS
orr-bootleg-2020	8	15	[	[	-LRB-
orr-bootleg-2020	8	16	38	38	CD
orr-bootleg-2020	8	17	]	]	-RRB-
orr-bootleg-2020	8	18	.	.	.
orr-bootleg-2020	9	1	Despite	despite	IN
orr-bootleg-2020	9	2	these	these	DT
orr-bootleg-2020	9	3	successes	success	NNS
orr-bootleg-2020	9	4	,	,	,
orr-bootleg-2020	9	5	a	a	DT
orr-bootleg-2020	9	6	persistent	persistent	JJ
orr-bootleg-2020	9	7	challenge	challenge	NN
orr-bootleg-2020	9	8	that	that	WDT
orr-bootleg-2020	9	9	recent	recent	JJ
orr-bootleg-2020	9	10	works	work	NNS
orr-bootleg-2020	9	11	continue	continue	VBP
orr-bootleg-2020	9	12	to	to	TO
orr-bootleg-2020	9	13	identify	identify	VB
orr-bootleg-2020	9	14	is	be	VBZ
orr-bootleg-2020	9	15	how	how	WRB
orr-bootleg-2020	9	16	to	to	TO
orr-bootleg-2020	9	17	leverage	leverage	VB
orr-bootleg-2020	9	18	knowledge	knowledge	NN
orr-bootleg-2020	9	19	for	for	IN
orr-bootleg-2020	9	20	low	low	JJ
orr-bootleg-2020	9	21	-	-	HYPH
orr-bootleg-2020	9	22	resource	resource	NN
orr-bootleg-2020	9	23	regimes	regime	NNS
orr-bootleg-2020	9	24	,	,	,
orr-bootleg-2020	9	25	such	such	JJ
orr-bootleg-2020	9	26	as	as	IN
orr-bootleg-2020	9	27	tail	tail	NN
orr-bootleg-2020	9	28	examples	example	NNS
orr-bootleg-2020	9	29	that	that	WDT
orr-bootleg-2020	9	30	appear	appear	VBP
orr-bootleg-2020	9	31	rarely	rarely	RB
orr-bootleg-2020	9	32	(	(	-LRB-
orr-bootleg-2020	9	33	if	if	IN
orr-bootleg-2020	9	34	at	at	RB
orr-bootleg-2020	9	35	all	all	RB
orr-bootleg-2020	9	36	)	)	-RRB-
orr-bootleg-2020	9	37	in	in	IN
orr-bootleg-2020	9	38	the	the	DT
orr-bootleg-2020	9	39	training	training	NN
orr-bootleg-2020	9	40	data	datum	NNS
orr-bootleg-2020	9	41	[	[	-LRB-
orr-bootleg-2020	9	42	16	16	CD
orr-bootleg-2020	9	43	]	]	-RRB-
orr-bootleg-2020	9	44	.	.	.
orr-bootleg-2020	10	1	In	in	IN
orr-bootleg-2020	10	2	this	this	DT
orr-bootleg-2020	10	3	work	work	NN
orr-bootleg-2020	10	4	,	,	,
orr-bootleg-2020	10	5	we	-PRON-	PRP
orr-bootleg-2020	10	6	study	study	VBP
orr-bootleg-2020	10	7	knowledge	knowledge	NN
orr-bootleg-2020	10	8	incorporation	incorporation	NN
orr-bootleg-2020	10	9	in	in	IN
orr-bootleg-2020	10	10	the	the	DT
orr-bootleg-2020	10	11	context	context	NN
orr-bootleg-2020	10	12	of	of	IN
orr-bootleg-2020	10	13	named	name	VBN
orr-bootleg-2020	10	14	entity	entity	NN
orr-bootleg-2020	10	15	disambiguation	disambiguation	NN
orr-bootleg-2020	10	16	(	(	-LRB-
orr-bootleg-2020	10	17	NED	NED	NNP
orr-bootleg-2020	10	18	)	)	-RRB-
orr-bootleg-2020	10	19	to	to	TO
orr-bootleg-2020	10	20	better	well	RBR
orr-bootleg-2020	10	21	disambiguate	disambiguate	VB
orr-bootleg-2020	10	22	the	the	DT
orr-bootleg-2020	10	23	long	long	JJ
orr-bootleg-2020	10	24	tail	tail	NN
orr-bootleg-2020	10	25	of	of	IN
orr-bootleg-2020	10	26	entities	entity	NNS
orr-bootleg-2020	10	27	that	that	WDT
orr-bootleg-2020	10	28	occur	occur	VBP
orr-bootleg-2020	10	29	infrequently	infrequently	RB
orr-bootleg-2020	10	30	during	during	IN
orr-bootleg-2020	10	31	training.1	training.1	ADD
orr-bootleg-2020	10	32	Humans	human	NNS
orr-bootleg-2020	10	33	disambiguate	disambiguate	VBP
orr-bootleg-2020	10	34	by	by	IN
orr-bootleg-2020	10	35	leveraging	leverage	VBG
orr-bootleg-2020	10	36	subtle	subtle	JJ
orr-bootleg-2020	10	37	reasoning	reasoning	NN
orr-bootleg-2020	10	38	over	over	IN
orr-bootleg-2020	10	39	entity	entity	NN
orr-bootleg-2020	10	40	-	-	HYPH
orr-bootleg-2020	10	41	based	base	VBN
orr-bootleg-2020	10	42	knowledge	knowledge	NN
orr-bootleg-2020	10	43	to	to	TO
orr-bootleg-2020	10	44	map	map	VB
orr-bootleg-2020	10	45	strings	string	NNS
orr-bootleg-2020	10	46	to	to	IN
orr-bootleg-2020	10	47	entities	entity	NNS
orr-bootleg-2020	10	48	in	in	IN
orr-bootleg-2020	10	49	a	a	DT
orr-bootleg-2020	10	50	knowledge	knowledge	NN
orr-bootleg-2020	10	51	base	base	NN
orr-bootleg-2020	10	52	.	.	.
orr-bootleg-2020	11	1	For	for	IN
orr-bootleg-2020	11	2	example	example	NN
orr-bootleg-2020	11	3	,	,	,
orr-bootleg-2020	11	4	in	in	IN
orr-bootleg-2020	11	5	the	the	DT
orr-bootleg-2020	11	6	sentence	sentence	NN
orr-bootleg-2020	11	7	“	"	``
orr-bootleg-2020	11	8	Where	where	WRB
orr-bootleg-2020	11	9	is	be	VBZ
orr-bootleg-2020	11	10	Lincoln	Lincoln	NNP
orr-bootleg-2020	11	11	in	in	IN
orr-bootleg-2020	11	12	Logan	Logan	NNP
orr-bootleg-2020	11	13	County	County	NNP
orr-bootleg-2020	11	14	?	?	.
orr-bootleg-2020	11	15	”	"	''
orr-bootleg-2020	11	16	,	,	,
orr-bootleg-2020	11	17	resolving	resolve	VBG
orr-bootleg-2020	11	18	the	the	DT
orr-bootleg-2020	11	19	mention	mention	NN
orr-bootleg-2020	11	20	“	"	``
orr-bootleg-2020	11	21	Lincoln	Lincoln	NNP
orr-bootleg-2020	11	22	”	"	''
orr-bootleg-2020	11	23	to	to	IN
orr-bootleg-2020	11	24	“	"	``
orr-bootleg-2020	11	25	Lincoln	Lincoln	NNP
orr-bootleg-2020	11	26	,	,	,
orr-bootleg-2020	11	27	IL	IL	NNP
orr-bootleg-2020	11	28	”	"	''
orr-bootleg-2020	11	29	requires	require	VBZ
orr-bootleg-2020	11	30	reasoning	reason	VBG
orr-bootleg-2020	11	31	about	about	IN
orr-bootleg-2020	11	32	relations	relation	NNS
orr-bootleg-2020	11	33	because	because	IN
orr-bootleg-2020	11	34	“	"	``
orr-bootleg-2020	11	35	Lincoln	Lincoln	NNP
orr-bootleg-2020	11	36	,	,	,
orr-bootleg-2020	11	37	IL”—not	IL”—not	:
orr-bootleg-2020	11	38	“	"	``
orr-bootleg-2020	11	39	Lincoln	Lincoln	NNP
orr-bootleg-2020	11	40	,	,	,
orr-bootleg-2020	11	41	NE	NE	NNP
orr-bootleg-2020	11	42	”	"	''
orr-bootleg-2020	11	43	or	or	CC
orr-bootleg-2020	11	44	“	"	``
orr-bootleg-2020	11	45	Abraham	Abraham	NNP
orr-bootleg-2020	11	46	Lincoln”—is	lincoln”—i	VBD
orr-bootleg-2020	11	47	the	the	DT
orr-bootleg-2020	11	48	capital	capital	NN
orr-bootleg-2020	11	49	of	of	IN
orr-bootleg-2020	11	50	Logan	Logan	NNP
orr-bootleg-2020	11	51	County	County	NNP
orr-bootleg-2020	11	52	.	.	.
orr-bootleg-2020	12	1	Previous	previous	JJ
orr-bootleg-2020	12	2	NED	NED	NNP
orr-bootleg-2020	12	3	systems	system	NNS
orr-bootleg-2020	12	4	disambiguate	disambiguate	VBP
orr-bootleg-2020	12	5	by	by	IN
orr-bootleg-2020	12	6	memorizing	memorize	VBG
orr-bootleg-2020	12	7	co	co	NNS
orr-bootleg-2020	12	8	-	-	NNS
orr-bootleg-2020	12	9	occurrences	occurrence	NNS
orr-bootleg-2020	12	10	between	between	IN
orr-bootleg-2020	12	11	entities	entity	NNS
orr-bootleg-2020	12	12	and	and	CC
orr-bootleg-2020	12	13	textual	textual	JJ
orr-bootleg-2020	12	14	context	context	NN
orr-bootleg-2020	12	15	in	in	IN
orr-bootleg-2020	12	16	a	a	DT
orr-bootleg-2020	12	17	self	self	NN
orr-bootleg-2020	12	18	-	-	HYPH
orr-bootleg-2020	12	19	supervised	supervise	VBN
orr-bootleg-2020	12	20	manner	manner	NN
orr-bootleg-2020	12	21	[	[	-LRB-
orr-bootleg-2020	12	22	16	16	CD
orr-bootleg-2020	12	23	,	,	,
orr-bootleg-2020	12	24	51	51	CD
orr-bootleg-2020	12	25	]	]	-RRB-
orr-bootleg-2020	12	26	.	.	.
orr-bootleg-2020	13	1	The	the	DT
orr-bootleg-2020	13	2	self	self	NN
orr-bootleg-2020	13	3	-	-	HYPH
orr-bootleg-2020	13	4	supervision	supervision	NN
orr-bootleg-2020	13	5	is	be	VBZ
orr-bootleg-2020	13	6	critical	critical	JJ
orr-bootleg-2020	13	7	to	to	IN
orr-bootleg-2020	13	8	building	build	VBG
orr-bootleg-2020	13	9	a	a	DT
orr-bootleg-2020	13	10	model	model	NN
orr-bootleg-2020	13	11	that	that	WDT
orr-bootleg-2020	13	12	is	be	VBZ
orr-bootleg-2020	13	13	easy	easy	JJ
orr-bootleg-2020	13	14	to	to	TO
orr-bootleg-2020	13	15	maintain	maintain	VB
orr-bootleg-2020	13	16	and	and	CC
orr-bootleg-2020	13	17	does	do	VBZ
orr-bootleg-2020	13	18	not	not	RB
orr-bootleg-2020	13	19	require	require	VB
orr-bootleg-2020	13	20	expensive	expensive	JJ
orr-bootleg-2020	13	21	hand	hand	NN
orr-bootleg-2020	13	22	-	-	HYPH
orr-bootleg-2020	13	23	curated	curate	VBN
orr-bootleg-2020	13	24	features	feature	NNS
orr-bootleg-2020	13	25	.	.	.
orr-bootleg-2020	14	1	However	however	RB
orr-bootleg-2020	14	2	,	,	,
orr-bootleg-2020	14	3	these	these	DT
orr-bootleg-2020	14	4	approaches	approach	NNS
orr-bootleg-2020	14	5	struggle	struggle	VBP
orr-bootleg-2020	14	6	to	to	TO
orr-bootleg-2020	14	7	handle	handle	VB
orr-bootleg-2020	14	8	tail	tail	NN
orr-bootleg-2020	14	9	entities	entity	NNS
orr-bootleg-2020	14	10	:	:	:
orr-bootleg-2020	14	11	a	a	DT
orr-bootleg-2020	14	12	baseline	baseline	JJ
orr-bootleg-2020	14	13	SotA	SotA	NNP
orr-bootleg-2020	14	14	model	model	NN
orr-bootleg-2020	14	15	from	from	IN
orr-bootleg-2020	14	16	[	[	-LRB-
orr-bootleg-2020	14	17	16	16	CD
orr-bootleg-2020	14	18	]	]	-RRB-
orr-bootleg-2020	14	19	achieves	achieve	VBZ
orr-bootleg-2020	14	20	less	less	JJR
orr-bootleg-2020	14	21	than	than	IN
orr-bootleg-2020	14	22	28	28	CD
orr-bootleg-2020	14	23	F1	f1	NN
orr-bootleg-2020	14	24	points	point	NNS
orr-bootleg-2020	14	25	over	over	IN
orr-bootleg-2020	14	26	the	the	DT
orr-bootleg-2020	14	27	tail	tail	NN
orr-bootleg-2020	14	28	,	,	,
orr-bootleg-2020	14	29	compared	compare	VBN
orr-bootleg-2020	14	30	to	to	IN
orr-bootleg-2020	14	31	86	86	CD
orr-bootleg-2020	14	32	F1	f1	NN
orr-bootleg-2020	14	33	points	point	NNS
orr-bootleg-2020	14	34	over	over	IN
orr-bootleg-2020	14	35	all	all	DT
orr-bootleg-2020	14	36	entities	entity	NNS
orr-bootleg-2020	14	37	.	.	.
orr-bootleg-2020	15	1	Despite	despite	IN
orr-bootleg-2020	15	2	their	-PRON-	PRP$
orr-bootleg-2020	15	3	rarity	rarity	NN
orr-bootleg-2020	15	4	in	in	IN
orr-bootleg-2020	15	5	training	training	NN
orr-bootleg-2020	15	6	data	datum	NNS
orr-bootleg-2020	15	7	,	,	,
orr-bootleg-2020	15	8	many	many	JJ
orr-bootleg-2020	15	9	real	real	JJ
orr-bootleg-2020	15	10	-	-	HYPH
orr-bootleg-2020	15	11	world	world	NN
orr-bootleg-2020	15	12	entities	entity	NNS
orr-bootleg-2020	15	13	are	be	VBP
orr-bootleg-2020	15	14	tail	tail	NN
orr-bootleg-2020	15	15	entities	entity	NNS
orr-bootleg-2020	15	16	:	:	:
orr-bootleg-2020	15	17	89	89	CD
orr-bootleg-2020	15	18	%	%	NN
orr-bootleg-2020	15	19	of	of	IN
orr-bootleg-2020	15	20	entities	entity	NNS
orr-bootleg-2020	15	21	in	in	IN
orr-bootleg-2020	15	22	the	the	DT
orr-bootleg-2020	15	23	Wikidata	Wikidata	NNP
orr-bootleg-2020	15	24	knowledge	knowledge	NN
orr-bootleg-2020	15	25	base	base	NN
orr-bootleg-2020	15	26	do	do	VBP
orr-bootleg-2020	15	27	not	not	RB
orr-bootleg-2020	15	28	have	have	VB
orr-bootleg-2020	15	29	Wikipedia	wikipedia	JJ
orr-bootleg-2020	15	30	pages	page	NNS
orr-bootleg-2020	15	31	to	to	TO
orr-bootleg-2020	15	32	serve	serve	VB
orr-bootleg-2020	15	33	as	as	IN
orr-bootleg-2020	15	34	a	a	DT
orr-bootleg-2020	15	35	source	source	NN
orr-bootleg-2020	15	36	of	of	IN
orr-bootleg-2020	15	37	textual	textual	JJ
orr-bootleg-2020	15	38	training	training	NN
orr-bootleg-2020	15	39	data	datum	NNS
orr-bootleg-2020	15	40	.	.	.
orr-bootleg-2020	16	1	However	however	RB
orr-bootleg-2020	16	2	,	,	,
orr-bootleg-2020	16	3	to	to	TO
orr-bootleg-2020	16	4	achieve	achieve	VB
orr-bootleg-2020	16	5	60	60	CD
orr-bootleg-2020	16	6	F1	f1	NN
orr-bootleg-2020	16	7	points	point	NNS
orr-bootleg-2020	16	8	on	on	IN
orr-bootleg-2020	16	9	disambiguation	disambiguation	NN
orr-bootleg-2020	16	10	,	,	,
orr-bootleg-2020	16	11	we	-PRON-	PRP
orr-bootleg-2020	16	12	find	find	VBP
orr-bootleg-2020	16	13	that	that	IN
orr-bootleg-2020	16	14	the	the	DT
orr-bootleg-2020	16	15	prior	prior	JJ
orr-bootleg-2020	16	16	SotA	SotA	NNP
orr-bootleg-2020	16	17	baseline	baseline	NNP
orr-bootleg-2020	16	18	model	model	NN
orr-bootleg-2020	16	19	should	should	MD
orr-bootleg-2020	16	20	see	see	VB
orr-bootleg-2020	16	21	an	an	DT
orr-bootleg-2020	16	22	entity	entity	NN
orr-bootleg-2020	16	23	1In	1in	CD
orr-bootleg-2020	16	24	this	this	DT
orr-bootleg-2020	16	25	work	work	NN
orr-bootleg-2020	16	26	,	,	,
orr-bootleg-2020	16	27	we	-PRON-	PRP
orr-bootleg-2020	16	28	define	define	VBP
orr-bootleg-2020	16	29	tail	tail	NN
orr-bootleg-2020	16	30	entities	entity	NNS
orr-bootleg-2020	16	31	as	as	IN
orr-bootleg-2020	16	32	those	those	DT
orr-bootleg-2020	16	33	occurring	occur	VBG
orr-bootleg-2020	16	34	10	10	CD
orr-bootleg-2020	16	35	or	or	CC
orr-bootleg-2020	16	36	fewer	few	JJR
orr-bootleg-2020	16	37	times	time	NNS
orr-bootleg-2020	16	38	in	in	IN
orr-bootleg-2020	16	39	the	the	DT
orr-bootleg-2020	16	40	training	training	NN
orr-bootleg-2020	16	41	data	datum	NNS
orr-bootleg-2020	16	42	.	.	.
orr-bootleg-2020	17	1	1	1	CD
orr-bootleg-2020	17	2	ar	ar	NNP
orr-bootleg-2020	17	3	X	X	NNP
orr-bootleg-2020	17	4	iv	iv	NN
orr-bootleg-2020	17	5	:	:	:
orr-bootleg-2020	17	6	2	2	CD
orr-bootleg-2020	17	7	01	01	CD
orr-bootleg-2020	17	8	0	0	CD
orr-bootleg-2020	17	9	.	.	.
orr-bootleg-2020	18	1	10	10	CD
orr-bootleg-2020	18	2	36	36	CD
orr-bootleg-2020	18	3	3v	3v	NNS
orr-bootleg-2020	18	4	3	3	CD
orr-bootleg-2020	18	5	[	[	-LRB-
orr-bootleg-2020	18	6	cs	cs	NNP
orr-bootleg-2020	18	7	.C	.C	.
orr-bootleg-2020	18	8	L	l	NN
orr-bootleg-2020	18	9	]	]	-RRB-
orr-bootleg-2020	18	10	2	2	CD
orr-bootleg-2020	18	11	3	3	CD
orr-bootleg-2020	18	12	O	o	NN
orr-bootleg-2020	18	13	ct	ct	IN
orr-bootleg-2020	18	14	2	2	CD
orr-bootleg-2020	18	15	02	02	CD
orr-bootleg-2020	18	16	0	0	CD
orr-bootleg-2020	18	17	How	how	WRB
orr-bootleg-2020	18	18	tall	tall	JJ
orr-bootleg-2020	18	19	is	be	VBZ
orr-bootleg-2020	18	20	Lincoln	Lincoln	NNP
orr-bootleg-2020	18	21	?	?	.
orr-bootleg-2020	19	1	The	the	DT
orr-bootleg-2020	19	2	Core	Core	NNP
orr-bootleg-2020	19	3	Reasoning	Reasoning	NNP
orr-bootleg-2020	19	4	Patterns	Patterns	NNPS
orr-bootleg-2020	19	5	of	of	IN
orr-bootleg-2020	19	6	NED	NED	NNP
orr-bootleg-2020	19	7	Type	Type	NNP
orr-bootleg-2020	19	8	Affordance	Affordance	NNP
orr-bootleg-2020	19	9	people	people	NNS
orr-bootleg-2020	19	10	have	have	VBP
orr-bootleg-2020	19	11	“	"	``
orr-bootleg-2020	19	12	heights	height	NNS
orr-bootleg-2020	19	13	”	"	''
orr-bootleg-2020	19	14	Lincoln	Lincoln	NNP
orr-bootleg-2020	19	15	,	,	,
orr-bootleg-2020	19	16	NE	NE	NNP
orr-bootleg-2020	19	17	Abraham	Abraham	NNP
orr-bootleg-2020	19	18	Lincoln	Lincoln	NNP
orr-bootleg-2020	19	19	Lincoln	Lincoln	NNP
orr-bootleg-2020	19	20	Motor	Motor	NNP
orr-bootleg-2020	19	21	Lincoln	Lincoln	NNP
orr-bootleg-2020	19	22	,	,	,
orr-bootleg-2020	19	23	IL	IL	NNP
orr-bootleg-2020	19	24	LOC	LOC	NNP
orr-bootleg-2020	19	25	PER	per	IN
orr-bootleg-2020	19	26	ORG	ORG	NNP
orr-bootleg-2020	19	27	LOC	LOC	NNP
orr-bootleg-2020	19	28	co	co	NN
orr-bootleg-2020	19	29	-	-	NN
orr-bootleg-2020	19	30	occurrence	occurrence	NN
orr-bootleg-2020	19	31	with	with	IN
orr-bootleg-2020	19	32	"	"	``
orr-bootleg-2020	19	33	Nebraska	Nebraska	NNP
orr-bootleg-2020	19	34	"	"	''
orr-bootleg-2020	19	35	Lincoln	Lincoln	NNP
orr-bootleg-2020	19	36	,	,	,
orr-bootleg-2020	19	37	NE	NE	NNP
orr-bootleg-2020	19	38	Abraham	Abraham	NNP
orr-bootleg-2020	19	39	Lincoln	Lincoln	NNP
orr-bootleg-2020	19	40	Lincoln	Lincoln	NNP
orr-bootleg-2020	19	41	Motor	Motor	NNP
orr-bootleg-2020	19	42	Lincoln	Lincoln	NNP
orr-bootleg-2020	19	43	,	,	,
orr-bootleg-2020	19	44	IL	IL	NNP
orr-bootleg-2020	19	45	Where	where	WRB
orr-bootleg-2020	19	46	is	be	VBZ
orr-bootleg-2020	19	47	Lincoln	Lincoln	NNP
orr-bootleg-2020	19	48	in	in	IN
orr-bootleg-2020	19	49	Logan	Logan	NNP
orr-bootleg-2020	19	50	County	County	NNP
orr-bootleg-2020	19	51	?	?	.
orr-bootleg-2020	20	1	"	"	``
orr-bootleg-2020	20	2	capital	capital	NN
orr-bootleg-2020	20	3	-	-	HYPH
orr-bootleg-2020	20	4	of	of	IN
orr-bootleg-2020	20	5	"	"	``
orr-bootleg-2020	20	6	relation	relation	NN
orr-bootleg-2020	20	7	Lincoln	Lincoln	NNP
orr-bootleg-2020	20	8	,	,	,
orr-bootleg-2020	20	9	NE	NE	NNP
orr-bootleg-2020	20	10	Abraham	Abraham	NNP
orr-bootleg-2020	20	11	Lincoln	Lincoln	NNP
orr-bootleg-2020	20	12	Lincoln	Lincoln	NNP
orr-bootleg-2020	20	13	Motor	Motor	NNP
orr-bootleg-2020	20	14	Lincoln	Lincoln	NNP
orr-bootleg-2020	20	15	,	,	,
orr-bootleg-2020	20	16	IL	IL	NNP
orr-bootleg-2020	20	17	Logan	Logan	NNP
orr-bootleg-2020	20	18	County	County	NNP
orr-bootleg-2020	20	19	,	,	,
orr-bootleg-2020	20	20	IL	IL	NNP
orr-bootleg-2020	20	21	Logan	Logan	NNP
orr-bootleg-2020	20	22	County	County	NNP
orr-bootleg-2020	20	23	,	,	,
orr-bootleg-2020	20	24	OK	OK	NNP
orr-bootleg-2020	20	25	Logan	Logan	NNP
orr-bootleg-2020	20	26	County	County	NNP
orr-bootleg-2020	20	27	,	,	,
orr-bootleg-2020	20	28	OH	OH	NNP
orr-bootleg-2020	20	29	Is	be	VBZ
orr-bootleg-2020	20	30	a	a	DT
orr-bootleg-2020	20	31	Lincoln	Lincoln	NNP
orr-bootleg-2020	20	32	or	or	CC
orr-bootleg-2020	20	33	Ford	Ford	NNP
orr-bootleg-2020	20	34	more	more	RBR
orr-bootleg-2020	20	35	expensive	expensive	JJ
orr-bootleg-2020	20	36	?	?	.
orr-bootleg-2020	21	1	consistent	consistent	JJ
orr-bootleg-2020	21	2	"	"	``
orr-bootleg-2020	21	3	car	car	NN
orr-bootleg-2020	21	4	"	"	''
orr-bootleg-2020	21	5	types	type	VBZ
orr-bootleg-2020	21	6	Lincoln	Lincoln	NNP
orr-bootleg-2020	21	7	,	,	,
orr-bootleg-2020	21	8	NE	NE	NNP
orr-bootleg-2020	21	9	Abraham	Abraham	NNP
orr-bootleg-2020	21	10	Lincoln	Lincoln	NNP
orr-bootleg-2020	21	11	Lincoln	Lincoln	NNP
orr-bootleg-2020	21	12	Motor	Motor	NNP
orr-bootleg-2020	21	13	Lincoln	Lincoln	NNP
orr-bootleg-2020	21	14	,	,	,
orr-bootleg-2020	21	15	IL	IL	NNP
orr-bootleg-2020	21	16	Ford	Ford	NNP
orr-bootleg-2020	21	17	Motor	Motor	NNP
orr-bootleg-2020	21	18	Ford	Ford	NNP
orr-bootleg-2020	21	19	,	,	,
orr-bootleg-2020	21	20	Australia	Australia	NNP
orr-bootleg-2020	21	21	Henry	Henry	NNP
orr-bootleg-2020	21	22	Ford	Ford	NNP
orr-bootleg-2020	21	23	KG	KG	NNP
orr-bootleg-2020	21	24	Relations	Relations	NNPS
orr-bootleg-2020	21	25	Entity	Entity	NNP
orr-bootleg-2020	21	26	Memorization	Memorization	NNP
orr-bootleg-2020	21	27	Type	Type	NNP
orr-bootleg-2020	21	28	Consistency	Consistency	NNP
orr-bootleg-2020	21	29	Increasing	Increasing	NNP
orr-bootleg-2020	21	30	generality	generality	NN
orr-bootleg-2020	21	31	of	of	IN
orr-bootleg-2020	21	32	pattern	pattern	NN
orr-bootleg-2020	21	33	Where	where	WRB
orr-bootleg-2020	21	34	is	be	VBZ
orr-bootleg-2020	21	35	Lincoln	Lincoln	NNP
orr-bootleg-2020	21	36	Nebraska	Nebraska	NNP
orr-bootleg-2020	21	37	?	?	.
orr-bootleg-2020	22	1	Up	up	IN
orr-bootleg-2020	22	2	to	to	IN
orr-bootleg-2020	22	3	100x	100x	NNS
orr-bootleg-2020	22	4	more	more	JJR
orr-bootleg-2020	22	5	data	datum	NNS
orr-bootleg-2020	22	6	needed	need	VBN
orr-bootleg-2020	22	7	to	to	TO
orr-bootleg-2020	22	8	recover	recover	VB
orr-bootleg-2020	22	9	performance	performance	NN
orr-bootleg-2020	22	10	of	of	IN
orr-bootleg-2020	22	11	Bootleg	Bootleg	NNP
orr-bootleg-2020	22	12	over	over	IN
orr-bootleg-2020	22	13	the	the	DT
orr-bootleg-2020	22	14	tail	tail	NN
orr-bootleg-2020	22	15	Overcoming	overcome	VBG
orr-bootleg-2020	22	16	the	the	DT
orr-bootleg-2020	22	17	Long	Long	NNP
orr-bootleg-2020	22	18	Tail	Tail	NNP
orr-bootleg-2020	22	19	of	of	IN
orr-bootleg-2020	22	20	NED	NED	NNP
orr-bootleg-2020	22	21	Baseline	Baseline	NNP
orr-bootleg-2020	22	22	Bootleg	Bootleg	NNP
orr-bootleg-2020	22	23	Tail	Tail	NNP
orr-bootleg-2020	22	24	Torso	Torso	NNP
orr-bootleg-2020	22	25	Head	Head	NNP
orr-bootleg-2020	22	26	Unseen	Unseen	NNP
orr-bootleg-2020	22	27	0	0	CD
orr-bootleg-2020	22	28	~100x	~100x	NNP
orr-bootleg-2020	22	29	F1	f1	NN
orr-bootleg-2020	22	30	0	0	CD
orr-bootleg-2020	22	31	0.2	0.2	CD
orr-bootleg-2020	22	32	0.4	0.4	CD
orr-bootleg-2020	22	33	0.6	0.6	CD
orr-bootleg-2020	22	34	0.8	0.8	CD
orr-bootleg-2020	22	35	1.0	1.0	CD
orr-bootleg-2020	22	36	Number	Number	NNP
orr-bootleg-2020	22	37	entity	entity	NN
orr-bootleg-2020	22	38	occurrences	occurrence	NNS
orr-bootleg-2020	22	39	in	in	IN
orr-bootleg-2020	22	40	training	train	VBG
orr-bootleg-2020	22	41	1	1	CD
orr-bootleg-2020	22	42	102	102	CD
orr-bootleg-2020	22	43	104	104	CD
orr-bootleg-2020	22	44	106	106	CD
orr-bootleg-2020	22	45	Figure	figure	NN
orr-bootleg-2020	22	46	1	1	CD
orr-bootleg-2020	22	47	:	:	:
orr-bootleg-2020	22	48	(	(	-LRB-
orr-bootleg-2020	22	49	Left	left	NN
orr-bootleg-2020	22	50	)	)	-RRB-
orr-bootleg-2020	22	51	shows	show	VBZ
orr-bootleg-2020	22	52	the	the	DT
orr-bootleg-2020	22	53	four	four	CD
orr-bootleg-2020	22	54	reasoning	reason	VBG
orr-bootleg-2020	22	55	patterns	pattern	NNS
orr-bootleg-2020	22	56	for	for	IN
orr-bootleg-2020	22	57	disambiguation	disambiguation	NN
orr-bootleg-2020	22	58	.	.	.
orr-bootleg-2020	23	1	The	the	DT
orr-bootleg-2020	23	2	correct	correct	JJ
orr-bootleg-2020	23	3	entity	entity	NN
orr-bootleg-2020	23	4	is	be	VBZ
orr-bootleg-2020	23	5	bolded	bolde	VBN
orr-bootleg-2020	23	6	.	.	.
orr-bootleg-2020	24	1	(	(	-LRB-
orr-bootleg-2020	24	2	Right	right	UH
orr-bootleg-2020	24	3	)	)	-RRB-
orr-bootleg-2020	24	4	shows	show	VBZ
orr-bootleg-2020	24	5	F1	f1	NN
orr-bootleg-2020	24	6	versus	versus	IN
orr-bootleg-2020	24	7	number	number	NN
orr-bootleg-2020	24	8	of	of	IN
orr-bootleg-2020	24	9	times	time	NNS
orr-bootleg-2020	24	10	an	an	DT
orr-bootleg-2020	24	11	entity	entity	NN
orr-bootleg-2020	24	12	was	be	VBD
orr-bootleg-2020	24	13	seen	see	VBN
orr-bootleg-2020	24	14	in	in	IN
orr-bootleg-2020	24	15	training	train	VBG
orr-bootleg-2020	24	16	data	datum	NNS
orr-bootleg-2020	24	17	for	for	IN
orr-bootleg-2020	24	18	a	a	DT
orr-bootleg-2020	24	19	baseline	baseline	JJ
orr-bootleg-2020	24	20	NED	NED	NNP
orr-bootleg-2020	24	21	model	model	NN
orr-bootleg-2020	24	22	compared	compare	VBN
orr-bootleg-2020	24	23	to	to	IN
orr-bootleg-2020	24	24	Bootleg	Bootleg	NNP
orr-bootleg-2020	24	25	across	across	IN
orr-bootleg-2020	24	26	the	the	DT
orr-bootleg-2020	24	27	head	head	NN
orr-bootleg-2020	24	28	,	,	,
orr-bootleg-2020	24	29	torso	torso	VB
orr-bootleg-2020	24	30	,	,	,
orr-bootleg-2020	24	31	tail	tail	NN
orr-bootleg-2020	24	32	,	,	,
orr-bootleg-2020	24	33	and	and	CC
orr-bootleg-2020	24	34	unseen	unseen	JJ
orr-bootleg-2020	24	35	.	.	.
orr-bootleg-2020	25	1	on	on	IN
orr-bootleg-2020	25	2	-	-	HYPH
orr-bootleg-2020	25	3	the	the	DT
orr-bootleg-2020	25	4	-	-	HYPH
orr-bootleg-2020	25	5	order	order	NN
orr-bootleg-2020	25	6	-	-	HYPH
orr-bootleg-2020	25	7	of	of	IN
orr-bootleg-2020	25	8	100	100	CD
orr-bootleg-2020	25	9	times	time	NNS
orr-bootleg-2020	25	10	during	during	IN
orr-bootleg-2020	25	11	training	training	NN
orr-bootleg-2020	25	12	(	(	-LRB-
orr-bootleg-2020	25	13	Figure	figure	NN
orr-bootleg-2020	25	14	1	1	CD
orr-bootleg-2020	25	15	(	(	-LRB-
orr-bootleg-2020	25	16	right	right	UH
orr-bootleg-2020	25	17	)	)	-RRB-
orr-bootleg-2020	25	18	)	)	-RRB-
orr-bootleg-2020	25	19	.	.	.
orr-bootleg-2020	26	1	This	this	DT
orr-bootleg-2020	26	2	presents	present	VBZ
orr-bootleg-2020	26	3	a	a	DT
orr-bootleg-2020	26	4	scalability	scalability	NN
orr-bootleg-2020	26	5	challenge	challenge	NN
orr-bootleg-2020	26	6	as	as	IN
orr-bootleg-2020	26	7	there	there	EX
orr-bootleg-2020	26	8	are	be	VBP
orr-bootleg-2020	26	9	15x	15x	NNS
orr-bootleg-2020	26	10	more	more	JJR
orr-bootleg-2020	26	11	entities	entity	NNS
orr-bootleg-2020	26	12	in	in	IN
orr-bootleg-2020	26	13	Wikidata	Wikidata	NNP
orr-bootleg-2020	26	14	than	than	IN
orr-bootleg-2020	26	15	in	in	IN
orr-bootleg-2020	26	16	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	26	17	,	,	,
orr-bootleg-2020	26	18	the	the	DT
orr-bootleg-2020	26	19	majority	majority	NN
orr-bootleg-2020	26	20	of	of	IN
orr-bootleg-2020	26	21	which	which	WDT
orr-bootleg-2020	26	22	are	be	VBP
orr-bootleg-2020	26	23	tail	tail	NN
orr-bootleg-2020	26	24	entities	entity	NNS
orr-bootleg-2020	26	25	.	.	.
orr-bootleg-2020	27	1	For	for	IN
orr-bootleg-2020	27	2	the	the	DT
orr-bootleg-2020	27	3	model	model	NN
orr-bootleg-2020	27	4	to	to	TO
orr-bootleg-2020	27	5	observe	observe	VB
orr-bootleg-2020	27	6	each	each	DT
orr-bootleg-2020	27	7	of	of	IN
orr-bootleg-2020	27	8	these	these	DT
orr-bootleg-2020	27	9	tail	tail	NN
orr-bootleg-2020	27	10	entities	entity	NNS
orr-bootleg-2020	27	11	100x	100x	NNS
orr-bootleg-2020	27	12	,	,	,
orr-bootleg-2020	27	13	the	the	DT
orr-bootleg-2020	27	14	training	training	NN
orr-bootleg-2020	27	15	data	datum	NNS
orr-bootleg-2020	27	16	would	would	MD
orr-bootleg-2020	27	17	need	need	VB
orr-bootleg-2020	27	18	to	to	TO
orr-bootleg-2020	27	19	be	be	VB
orr-bootleg-2020	27	20	scaled	scale	VBN
orr-bootleg-2020	27	21	by	by	IN
orr-bootleg-2020	27	22	1,500x	1,500x	CD
orr-bootleg-2020	27	23	the	the	DT
orr-bootleg-2020	27	24	size	size	NN
orr-bootleg-2020	27	25	of	of	IN
orr-bootleg-2020	27	26	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	27	27	.	.	.
orr-bootleg-2020	28	1	Prior	prior	JJ
orr-bootleg-2020	28	2	approaches	approach	VBZ
orr-bootleg-2020	28	3	struggle	struggle	NN
orr-bootleg-2020	28	4	with	with	IN
orr-bootleg-2020	28	5	the	the	DT
orr-bootleg-2020	28	6	tail	tail	NN
orr-bootleg-2020	28	7	,	,	,
orr-bootleg-2020	28	8	yet	yet	CC
orr-bootleg-2020	28	9	industry	industry	NN
orr-bootleg-2020	28	10	applications	application	NNS
orr-bootleg-2020	28	11	such	such	JJ
orr-bootleg-2020	28	12	as	as	IN
orr-bootleg-2020	28	13	search	search	NN
orr-bootleg-2020	28	14	and	and	CC
orr-bootleg-2020	28	15	voice	voice	NN
orr-bootleg-2020	28	16	assistants	assistant	NNS
orr-bootleg-2020	28	17	are	be	VBP
orr-bootleg-2020	28	18	known	know	VBN
orr-bootleg-2020	28	19	to	to	TO
orr-bootleg-2020	28	20	be	be	VB
orr-bootleg-2020	28	21	tail	tail	NN
orr-bootleg-2020	28	22	-	-	HYPH
orr-bootleg-2020	28	23	heavy	heavy	NN
orr-bootleg-2020	28	24	[	[	-LRB-
orr-bootleg-2020	28	25	4	4	CD
orr-bootleg-2020	28	26	,	,	,
orr-bootleg-2020	28	27	20	20	CD
orr-bootleg-2020	28	28	]	]	-RRB-
orr-bootleg-2020	28	29	.	.	.
orr-bootleg-2020	29	1	Given	give	VBN
orr-bootleg-2020	29	2	the	the	DT
orr-bootleg-2020	29	3	requirement	requirement	NN
orr-bootleg-2020	29	4	for	for	IN
orr-bootleg-2020	29	5	high	high	JJ
orr-bootleg-2020	29	6	quality	quality	NN
orr-bootleg-2020	29	7	tail	tail	NN
orr-bootleg-2020	29	8	disambiguation	disambiguation	NN
orr-bootleg-2020	29	9	,	,	,
orr-bootleg-2020	29	10	major	major	JJ
orr-bootleg-2020	29	11	technology	technology	NN
orr-bootleg-2020	29	12	companies	company	NNS
orr-bootleg-2020	29	13	continue	continue	VBP
orr-bootleg-2020	29	14	to	to	TO
orr-bootleg-2020	29	15	press	press	VB
orr-bootleg-2020	29	16	on	on	IN
orr-bootleg-2020	29	17	this	this	DT
orr-bootleg-2020	29	18	challenge	challenge	NN
orr-bootleg-2020	29	19	[	[	-LRB-
orr-bootleg-2020	29	20	29	29	CD
orr-bootleg-2020	29	21	,	,	,
orr-bootleg-2020	29	22	39	39	CD
orr-bootleg-2020	29	23	]	]	-RRB-
orr-bootleg-2020	29	24	.	.	.
orr-bootleg-2020	30	1	Instead	instead	RB
orr-bootleg-2020	30	2	of	of	IN
orr-bootleg-2020	30	3	scaling	scale	VBG
orr-bootleg-2020	30	4	the	the	DT
orr-bootleg-2020	30	5	training	training	NN
orr-bootleg-2020	30	6	data	datum	NNS
orr-bootleg-2020	30	7	until	until	IN
orr-bootleg-2020	30	8	co	co	NNS
orr-bootleg-2020	30	9	-	-	NNS
orr-bootleg-2020	30	10	occurrences	occurrence	NNS
orr-bootleg-2020	30	11	between	between	IN
orr-bootleg-2020	30	12	tail	tail	NN
orr-bootleg-2020	30	13	entities	entity	NNS
orr-bootleg-2020	30	14	and	and	CC
orr-bootleg-2020	30	15	text	text	NN
orr-bootleg-2020	30	16	can	can	MD
orr-bootleg-2020	30	17	be	be	VB
orr-bootleg-2020	30	18	memorized	memorize	VBN
orr-bootleg-2020	30	19	,	,	,
orr-bootleg-2020	30	20	we	-PRON-	PRP
orr-bootleg-2020	30	21	define	define	VBP
orr-bootleg-2020	30	22	a	a	DT
orr-bootleg-2020	30	23	principled	principle	VBN
orr-bootleg-2020	30	24	set	set	NN
orr-bootleg-2020	30	25	of	of	IN
orr-bootleg-2020	30	26	reasoning	reason	VBG
orr-bootleg-2020	30	27	patterns	pattern	NNS
orr-bootleg-2020	30	28	for	for	IN
orr-bootleg-2020	30	29	entity	entity	NN
orr-bootleg-2020	30	30	disambiguation	disambiguation	NN
orr-bootleg-2020	30	31	across	across	IN
orr-bootleg-2020	30	32	the	the	DT
orr-bootleg-2020	30	33	head	head	NN
orr-bootleg-2020	30	34	and	and	CC
orr-bootleg-2020	30	35	tail	tail	NN
orr-bootleg-2020	30	36	.	.	.
orr-bootleg-2020	31	1	When	when	WRB
orr-bootleg-2020	31	2	humans	human	NNS
orr-bootleg-2020	31	3	disambiguate	disambiguate	VBP
orr-bootleg-2020	31	4	entities	entity	NNS
orr-bootleg-2020	31	5	,	,	,
orr-bootleg-2020	31	6	they	-PRON-	PRP
orr-bootleg-2020	31	7	leverage	leverage	VBP
orr-bootleg-2020	31	8	signals	signal	NNS
orr-bootleg-2020	31	9	from	from	IN
orr-bootleg-2020	31	10	context	context	NN
orr-bootleg-2020	31	11	as	as	RB
orr-bootleg-2020	31	12	well	well	RB
orr-bootleg-2020	31	13	as	as	IN
orr-bootleg-2020	31	14	from	from	IN
orr-bootleg-2020	31	15	entity	entity	NN
orr-bootleg-2020	31	16	relations	relation	NNS
orr-bootleg-2020	31	17	and	and	CC
orr-bootleg-2020	31	18	types	type	NNS
orr-bootleg-2020	31	19	.	.	.
orr-bootleg-2020	32	1	For	for	IN
orr-bootleg-2020	32	2	example	example	NN
orr-bootleg-2020	32	3	,	,	,
orr-bootleg-2020	32	4	resolving	resolve	VBG
orr-bootleg-2020	32	5	“	"	``
orr-bootleg-2020	32	6	Lincoln	Lincoln	NNP
orr-bootleg-2020	32	7	”	"	''
orr-bootleg-2020	32	8	in	in	IN
orr-bootleg-2020	32	9	the	the	DT
orr-bootleg-2020	32	10	text	text	NN
orr-bootleg-2020	32	11	“	"	``
orr-bootleg-2020	32	12	How	how	WRB
orr-bootleg-2020	32	13	tall	tall	JJ
orr-bootleg-2020	32	14	is	be	VBZ
orr-bootleg-2020	32	15	Lincoln	Lincoln	NNP
orr-bootleg-2020	32	16	?	?	.
orr-bootleg-2020	32	17	”	"	''
orr-bootleg-2020	32	18	to	to	IN
orr-bootleg-2020	32	19	“	"	``
orr-bootleg-2020	32	20	Abraham	Abraham	NNP
orr-bootleg-2020	32	21	Lincoln	Lincoln	NNP
orr-bootleg-2020	32	22	”	"	''
orr-bootleg-2020	32	23	requires	require	VBZ
orr-bootleg-2020	32	24	reasoning	reason	VBG
orr-bootleg-2020	32	25	that	that	IN
orr-bootleg-2020	32	26	people	people	NNS
orr-bootleg-2020	32	27	,	,	,
orr-bootleg-2020	32	28	not	not	RB
orr-bootleg-2020	32	29	locations	location	NNS
orr-bootleg-2020	32	30	or	or	CC
orr-bootleg-2020	32	31	car	car	NN
orr-bootleg-2020	32	32	companies	company	NNS
orr-bootleg-2020	32	33	,	,	,
orr-bootleg-2020	32	34	have	have	VBP
orr-bootleg-2020	32	35	heights	height	NNS
orr-bootleg-2020	32	36	—	—	:
orr-bootleg-2020	32	37	a	a	DT
orr-bootleg-2020	32	38	type	type	NN
orr-bootleg-2020	32	39	affordance	affordance	NN
orr-bootleg-2020	32	40	pattern	pattern	NN
orr-bootleg-2020	32	41	.	.	.
orr-bootleg-2020	33	1	These	these	DT
orr-bootleg-2020	33	2	core	core	NN
orr-bootleg-2020	33	3	patterns	pattern	NNS
orr-bootleg-2020	33	4	apply	apply	VBP
orr-bootleg-2020	33	5	to	to	IN
orr-bootleg-2020	33	6	both	both	DT
orr-bootleg-2020	33	7	head	head	NN
orr-bootleg-2020	33	8	and	and	CC
orr-bootleg-2020	33	9	tail	tail	NN
orr-bootleg-2020	33	10	examples	example	NNS
orr-bootleg-2020	33	11	with	with	IN
orr-bootleg-2020	33	12	high	high	JJ
orr-bootleg-2020	33	13	coverage	coverage	NN
orr-bootleg-2020	33	14	and	and	CC
orr-bootleg-2020	33	15	involve	involve	VBP
orr-bootleg-2020	33	16	reasoning	reason	VBG
orr-bootleg-2020	33	17	over	over	IN
orr-bootleg-2020	33	18	entity	entity	NN
orr-bootleg-2020	33	19	facts	fact	NNS
orr-bootleg-2020	33	20	,	,	,
orr-bootleg-2020	33	21	relations	relation	NNS
orr-bootleg-2020	33	22	,	,	,
orr-bootleg-2020	33	23	and	and	CC
orr-bootleg-2020	33	24	types	type	NNS
orr-bootleg-2020	33	25	,	,	,
orr-bootleg-2020	33	26	information	information	NN
orr-bootleg-2020	33	27	which	which	WDT
orr-bootleg-2020	33	28	is	be	VBZ
orr-bootleg-2020	33	29	available	available	JJ
orr-bootleg-2020	33	30	for	for	IN
orr-bootleg-2020	33	31	both	both	DT
orr-bootleg-2020	33	32	head	head	NN
orr-bootleg-2020	33	33	and	and	CC
orr-bootleg-2020	33	34	tail	tail	NN
orr-bootleg-2020	33	35	in	in	IN
orr-bootleg-2020	33	36	structured	structured	JJ
orr-bootleg-2020	33	37	data	datum	NNS
orr-bootleg-2020	33	38	sources	source	NNS
orr-bootleg-2020	33	39	.	.	.
orr-bootleg-2020	34	1	2	2	LS
orr-bootleg-2020	34	2	Thus	thus	RB
orr-bootleg-2020	34	3	,	,	,
orr-bootleg-2020	34	4	we	-PRON-	PRP
orr-bootleg-2020	34	5	hypothesize	hypothesize	VBP
orr-bootleg-2020	34	6	that	that	IN
orr-bootleg-2020	34	7	these	these	DT
orr-bootleg-2020	34	8	patterns	pattern	NNS
orr-bootleg-2020	34	9	assembled	assemble	VBN
orr-bootleg-2020	34	10	from	from	IN
orr-bootleg-2020	34	11	the	the	DT
orr-bootleg-2020	34	12	structured	structure	VBN
orr-bootleg-2020	34	13	resources	resource	NNS
orr-bootleg-2020	34	14	can	can	MD
orr-bootleg-2020	34	15	be	be	VB
orr-bootleg-2020	34	16	learned	learn	VBN
orr-bootleg-2020	34	17	over	over	IN
orr-bootleg-2020	34	18	training	training	NN
orr-bootleg-2020	34	19	data	datum	NNS
orr-bootleg-2020	34	20	and	and	CC
orr-bootleg-2020	34	21	generalize	generalize	VB
orr-bootleg-2020	34	22	to	to	IN
orr-bootleg-2020	34	23	the	the	DT
orr-bootleg-2020	34	24	tail	tail	NN
orr-bootleg-2020	34	25	.	.	.
orr-bootleg-2020	35	1	In	in	IN
orr-bootleg-2020	35	2	this	this	DT
orr-bootleg-2020	35	3	work	work	NN
orr-bootleg-2020	35	4	,	,	,
orr-bootleg-2020	35	5	we	-PRON-	PRP
orr-bootleg-2020	35	6	introduce	introduce	VBP
orr-bootleg-2020	35	7	Bootleg	Bootleg	NNP
orr-bootleg-2020	35	8	,	,	,
orr-bootleg-2020	35	9	an	an	DT
orr-bootleg-2020	35	10	open	open	JJ
orr-bootleg-2020	35	11	-	-	HYPH
orr-bootleg-2020	35	12	source	source	NN
orr-bootleg-2020	35	13	,	,	,
orr-bootleg-2020	35	14	self	self	NN
orr-bootleg-2020	35	15	-	-	HYPH
orr-bootleg-2020	35	16	supervised	supervise	VBN
orr-bootleg-2020	35	17	NED	NED	NNP
orr-bootleg-2020	35	18	system	system	NN
orr-bootleg-2020	35	19	designed	design	VBN
orr-bootleg-2020	35	20	to	to	TO
orr-bootleg-2020	35	21	succeed	succeed	VB
orr-bootleg-2020	35	22	on	on	IN
orr-bootleg-2020	35	23	head	head	NN
orr-bootleg-2020	35	24	and	and	CC
orr-bootleg-2020	35	25	tail	tail	NN
orr-bootleg-2020	35	26	entities	entity	NNS
orr-bootleg-2020	35	27	.	.	.
orr-bootleg-2020	36	1	3	3	CD
orr-bootleg-2020	36	2	Bootleg	Bootleg	NNP
orr-bootleg-2020	36	3	encodes	encode	VBZ
orr-bootleg-2020	36	4	the	the	DT
orr-bootleg-2020	36	5	entity	entity	NN
orr-bootleg-2020	36	6	,	,	,
orr-bootleg-2020	36	7	relation	relation	NN
orr-bootleg-2020	36	8	,	,	,
orr-bootleg-2020	36	9	and	and	CC
orr-bootleg-2020	36	10	type	type	NN
orr-bootleg-2020	36	11	signals	signal	NNS
orr-bootleg-2020	36	12	as	as	IN
orr-bootleg-2020	36	13	embedding	embed	VBG
orr-bootleg-2020	36	14	inputs	input	NNS
orr-bootleg-2020	36	15	to	to	IN
orr-bootleg-2020	36	16	a	a	DT
orr-bootleg-2020	36	17	simple	simple	JJ
orr-bootleg-2020	36	18	stacked	stack	VBN
orr-bootleg-2020	36	19	Transformer	Transformer	NNP
orr-bootleg-2020	36	20	architecture	architecture	NN
orr-bootleg-2020	36	21	.	.	.
orr-bootleg-2020	37	1	The	the	DT
orr-bootleg-2020	37	2	key	key	JJ
orr-bootleg-2020	37	3	challenges	challenge	NNS
orr-bootleg-2020	37	4	we	-PRON-	PRP
orr-bootleg-2020	37	5	face	face	VBP
orr-bootleg-2020	37	6	are	be	VBP
orr-bootleg-2020	37	7	understanding	understand	VBG
orr-bootleg-2020	37	8	how	how	WRB
orr-bootleg-2020	37	9	to	to	TO
orr-bootleg-2020	37	10	use	use	VB
orr-bootleg-2020	37	11	knowledge	knowledge	NN
orr-bootleg-2020	37	12	for	for	IN
orr-bootleg-2020	37	13	NED	NED	NNP
orr-bootleg-2020	37	14	,	,	,
orr-bootleg-2020	37	15	designing	design	VBG
orr-bootleg-2020	37	16	a	a	DT
orr-bootleg-2020	37	17	model	model	NN
orr-bootleg-2020	37	18	that	that	WDT
orr-bootleg-2020	37	19	learns	learn	VBZ
orr-bootleg-2020	37	20	those	those	DT
orr-bootleg-2020	37	21	patterns	pattern	NNS
orr-bootleg-2020	37	22	,	,	,
orr-bootleg-2020	37	23	and	and	CC
orr-bootleg-2020	37	24	fully	fully	RB
orr-bootleg-2020	37	25	extracting	extract	VBG
orr-bootleg-2020	37	26	the	the	DT
orr-bootleg-2020	37	27	useful	useful	JJ
orr-bootleg-2020	37	28	knowledge	knowledge	NN
orr-bootleg-2020	37	29	signals	signal	NNS
orr-bootleg-2020	37	30	from	from	IN
orr-bootleg-2020	37	31	the	the	DT
orr-bootleg-2020	37	32	training	training	NN
orr-bootleg-2020	37	33	data	datum	NNS
orr-bootleg-2020	37	34	:	:	:
orr-bootleg-2020	37	35	•	•	NNP
orr-bootleg-2020	37	36	Tail	Tail	NNP
orr-bootleg-2020	37	37	Reasoning	reasoning	NN
orr-bootleg-2020	37	38	:	:	:
orr-bootleg-2020	37	39	Humans	human	NNS
orr-bootleg-2020	37	40	use	use	VBP
orr-bootleg-2020	37	41	subtle	subtle	JJ
orr-bootleg-2020	37	42	reasoning	reasoning	NN
orr-bootleg-2020	37	43	patterns	pattern	NNS
orr-bootleg-2020	37	44	to	to	TO
orr-bootleg-2020	37	45	disambiguate	disambiguate	VB
orr-bootleg-2020	37	46	different	different	JJ
orr-bootleg-2020	37	47	entities	entity	NNS
orr-bootleg-2020	37	48	,	,	,
orr-bootleg-2020	37	49	especially	especially	RB
orr-bootleg-2020	37	50	unfamiliar	unfamiliar	JJ
orr-bootleg-2020	37	51	tail	tail	NN
orr-bootleg-2020	37	52	entities	entity	NNS
orr-bootleg-2020	37	53	.	.	.
orr-bootleg-2020	38	1	The	the	DT
orr-bootleg-2020	38	2	first	first	JJ
orr-bootleg-2020	38	3	challenge	challenge	NN
orr-bootleg-2020	38	4	is	be	VBZ
orr-bootleg-2020	38	5	characterizing	characterize	VBG
orr-bootleg-2020	38	6	these	these	DT
orr-bootleg-2020	38	7	reasoning	reasoning	NN
orr-bootleg-2020	38	8	patterns	pattern	NNS
orr-bootleg-2020	38	9	and	and	CC
orr-bootleg-2020	38	10	understanding	understand	VBG
orr-bootleg-2020	38	11	their	-PRON-	PRP$
orr-bootleg-2020	38	12	coverage	coverage	NN
orr-bootleg-2020	38	13	over	over	IN
orr-bootleg-2020	38	14	the	the	DT
orr-bootleg-2020	38	15	tail	tail	NN
orr-bootleg-2020	38	16	.	.	.
orr-bootleg-2020	39	1	•	•	VB
orr-bootleg-2020	39	2	Poor	Poor	NNP
orr-bootleg-2020	39	3	Tail	Tail	NNP
orr-bootleg-2020	39	4	Generalization	generalization	NN
orr-bootleg-2020	39	5	:	:	:
orr-bootleg-2020	39	6	We	-PRON-	PRP
orr-bootleg-2020	39	7	find	find	VBP
orr-bootleg-2020	39	8	that	that	IN
orr-bootleg-2020	39	9	a	a	DT
orr-bootleg-2020	39	10	model	model	NN
orr-bootleg-2020	39	11	trained	train	VBN
orr-bootleg-2020	39	12	using	use	VBG
orr-bootleg-2020	39	13	standard	standard	JJ
orr-bootleg-2020	39	14	regularization	regularization	NN
orr-bootleg-2020	39	15	and	and	CC
orr-bootleg-2020	39	16	a	a	DT
orr-bootleg-2020	39	17	combination	combination	NN
orr-bootleg-2020	39	18	of	of	IN
orr-bootleg-2020	39	19	entity	entity	NN
orr-bootleg-2020	39	20	,	,	,
orr-bootleg-2020	39	21	type	type	NN
orr-bootleg-2020	39	22	and	and	CC
orr-bootleg-2020	39	23	relation	relation	NN
orr-bootleg-2020	39	24	information	information	NN
orr-bootleg-2020	39	25	performs	perform	VBZ
orr-bootleg-2020	39	26	10	10	CD
orr-bootleg-2020	39	27	F1	f1	NN
orr-bootleg-2020	39	28	points	point	NNS
orr-bootleg-2020	39	29	worse	bad	JJR
orr-bootleg-2020	39	30	on	on	IN
orr-bootleg-2020	39	31	disambiguating	disambiguate	VBG
orr-bootleg-2020	39	32	unseen	unseen	JJ
orr-bootleg-2020	39	33	entities	entity	NNS
orr-bootleg-2020	39	34	compared	compare	VBN
orr-bootleg-2020	39	35	to	to	IN
orr-bootleg-2020	39	36	the	the	DT
orr-bootleg-2020	39	37	two	two	CD
orr-bootleg-2020	39	38	models	model	NNS
orr-bootleg-2020	39	39	which	which	WDT
orr-bootleg-2020	39	40	respectively	respectively	RB
orr-bootleg-2020	39	41	use	use	VBP
orr-bootleg-2020	39	42	only	only	JJ
orr-bootleg-2020	39	43	type	type	NN
orr-bootleg-2020	39	44	and	and	CC
orr-bootleg-2020	39	45	only	only	JJ
orr-bootleg-2020	39	46	relation	relation	NN
orr-bootleg-2020	39	47	information	information	NN
orr-bootleg-2020	39	48	.	.	.
orr-bootleg-2020	40	1	We	-PRON-	PRP
orr-bootleg-2020	40	2	find	find	VBP
orr-bootleg-2020	40	3	this	this	DT
orr-bootleg-2020	40	4	performance	performance	NN
orr-bootleg-2020	40	5	drop	drop	NN
orr-bootleg-2020	40	6	is	be	VBZ
orr-bootleg-2020	40	7	due	due	JJ
orr-bootleg-2020	40	8	to	to	IN
orr-bootleg-2020	40	9	the	the	DT
orr-bootleg-2020	40	10	model	model	NN
orr-bootleg-2020	40	11	’s	’s	POS
orr-bootleg-2020	40	12	over	over	NN
orr-bootleg-2020	40	13	-	-	HYPH
orr-bootleg-2020	40	14	reliance	reliance	NN
orr-bootleg-2020	40	15	on	on	IN
orr-bootleg-2020	40	16	discriminative	discriminative	JJ
orr-bootleg-2020	40	17	textual	textual	NN
orr-bootleg-2020	40	18	and	and	CC
orr-bootleg-2020	40	19	entity	entity	NN
orr-bootleg-2020	40	20	features	feature	NNS
orr-bootleg-2020	40	21	compared	compare	VBN
orr-bootleg-2020	40	22	to	to	IN
orr-bootleg-2020	40	23	more	more	JJR
orr-bootleg-2020	40	24	general	general	JJ
orr-bootleg-2020	40	25	type	type	NN
orr-bootleg-2020	40	26	and	and	CC
orr-bootleg-2020	40	27	relation	relation	NN
orr-bootleg-2020	40	28	features	feature	NNS
orr-bootleg-2020	40	29	.	.	.
orr-bootleg-2020	41	1	•	•	NNP
orr-bootleg-2020	41	2	Underutilized	Underutilized	NNP
orr-bootleg-2020	41	3	Data	Data	NNP
orr-bootleg-2020	41	4	:	:	:
orr-bootleg-2020	41	5	Self	self	NN
orr-bootleg-2020	41	6	-	-	HYPH
orr-bootleg-2020	41	7	supervised	supervise	VBN
orr-bootleg-2020	41	8	models	model	NNS
orr-bootleg-2020	41	9	improve	improve	VBP
orr-bootleg-2020	41	10	with	with	IN
orr-bootleg-2020	41	11	more	more	JJR
orr-bootleg-2020	41	12	training	training	NN
orr-bootleg-2020	41	13	data	datum	NNS
orr-bootleg-2020	41	14	[	[	-LRB-
orr-bootleg-2020	41	15	7	7	CD
orr-bootleg-2020	41	16	]	]	-RRB-
orr-bootleg-2020	41	17	.	.	.
orr-bootleg-2020	42	1	However	however	RB
orr-bootleg-2020	42	2	,	,	,
orr-bootleg-2020	42	3	only	only	RB
orr-bootleg-2020	42	4	a	a	DT
orr-bootleg-2020	42	5	2We	2We	NNS
orr-bootleg-2020	42	6	find	find	VBP
orr-bootleg-2020	42	7	that	that	DT
orr-bootleg-2020	42	8	type	type	NN
orr-bootleg-2020	42	9	affordance	affordance	NN
orr-bootleg-2020	42	10	patterns	pattern	NNS
orr-bootleg-2020	42	11	apply	apply	VBP
orr-bootleg-2020	42	12	to	to	IN
orr-bootleg-2020	42	13	over	over	IN
orr-bootleg-2020	42	14	84	84	CD
orr-bootleg-2020	42	15	%	%	NN
orr-bootleg-2020	42	16	of	of	IN
orr-bootleg-2020	42	17	all	all	DT
orr-bootleg-2020	42	18	examples	example	NNS
orr-bootleg-2020	42	19	,	,	,
orr-bootleg-2020	42	20	including	include	VBG
orr-bootleg-2020	42	21	tail	tail	NN
orr-bootleg-2020	42	22	examples	example	NNS
orr-bootleg-2020	42	23	,	,	,
orr-bootleg-2020	42	24	while	while	IN
orr-bootleg-2020	42	25	KG	kg	NN
orr-bootleg-2020	42	26	relation	relation	NN
orr-bootleg-2020	42	27	patterns	pattern	NNS
orr-bootleg-2020	42	28	apply	apply	VBP
orr-bootleg-2020	42	29	to	to	IN
orr-bootleg-2020	42	30	over	over	IN
orr-bootleg-2020	42	31	27	27	CD
orr-bootleg-2020	42	32	%	%	NN
orr-bootleg-2020	42	33	of	of	IN
orr-bootleg-2020	42	34	all	all	DT
orr-bootleg-2020	42	35	examples	example	NNS
orr-bootleg-2020	42	36	and	and	CC
orr-bootleg-2020	42	37	type	type	NN
orr-bootleg-2020	42	38	consistency	consistency	NN
orr-bootleg-2020	42	39	applies	apply	VBZ
orr-bootleg-2020	42	40	to	to	IN
orr-bootleg-2020	42	41	over	over	IN
orr-bootleg-2020	42	42	8	8	CD
orr-bootleg-2020	42	43	%	%	NN
orr-bootleg-2020	42	44	of	of	IN
orr-bootleg-2020	42	45	all	all	DT
orr-bootleg-2020	42	46	examples	example	NNS
orr-bootleg-2020	42	47	.	.	.
orr-bootleg-2020	43	1	In	in	IN
orr-bootleg-2020	43	2	Wikidata	Wikidata	NNP
orr-bootleg-2020	43	3	,	,	,
orr-bootleg-2020	43	4	75	75	CD
orr-bootleg-2020	43	5	%	%	NN
orr-bootleg-2020	43	6	of	of	IN
orr-bootleg-2020	43	7	entities	entity	NNS
orr-bootleg-2020	43	8	that	that	WDT
orr-bootleg-2020	43	9	are	be	VBP
orr-bootleg-2020	43	10	not	not	RB
orr-bootleg-2020	43	11	in	in	IN
orr-bootleg-2020	43	12	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	43	13	have	have	VB
orr-bootleg-2020	43	14	type	type	NN
orr-bootleg-2020	43	15	or	or	CC
orr-bootleg-2020	43	16	knowledge	knowledge	NN
orr-bootleg-2020	43	17	graph	graph	NN
orr-bootleg-2020	43	18	connectivity	connectivity	NN
orr-bootleg-2020	43	19	signals	signal	NNS
orr-bootleg-2020	43	20	,	,	,
orr-bootleg-2020	43	21	and	and	CC
orr-bootleg-2020	43	22	among	among	IN
orr-bootleg-2020	43	23	tail	tail	NN
orr-bootleg-2020	43	24	entities	entity	NNS
orr-bootleg-2020	43	25	,	,	,
orr-bootleg-2020	43	26	88	88	CD
orr-bootleg-2020	43	27	%	%	NN
orr-bootleg-2020	43	28	are	be	VBP
orr-bootleg-2020	43	29	in	in	IN
orr-bootleg-2020	43	30	non	non	JJ
orr-bootleg-2020	43	31	-	-	JJ
orr-bootleg-2020	43	32	tail	tail	JJ
orr-bootleg-2020	43	33	type	type	NN
orr-bootleg-2020	43	34	categories	category	NNS
orr-bootleg-2020	43	35	and	and	CC
orr-bootleg-2020	43	36	90	90	CD
orr-bootleg-2020	43	37	%	%	NN
orr-bootleg-2020	43	38	are	be	VBP
orr-bootleg-2020	43	39	in	in	IN
orr-bootleg-2020	43	40	non	non	JJ
orr-bootleg-2020	43	41	-	-	JJ
orr-bootleg-2020	43	42	tail	tail	JJ
orr-bootleg-2020	43	43	relation	relation	NN
orr-bootleg-2020	43	44	categories	category	NNS
orr-bootleg-2020	43	45	.	.	.
orr-bootleg-2020	44	1	3Bootleg	3Bootleg	NNP
orr-bootleg-2020	44	2	is	be	VBZ
orr-bootleg-2020	44	3	open	open	JJ
orr-bootleg-2020	44	4	-	-	HYPH
orr-bootleg-2020	44	5	source	source	NN
orr-bootleg-2020	44	6	at	at	IN
orr-bootleg-2020	44	7	http://hazyresearch.stanford.edu/bootleg	http://hazyresearch.stanford.edu/bootleg	ADD
orr-bootleg-2020	44	8	2	2	CD
orr-bootleg-2020	44	9	http://hazyresearch.stanford.edu/bootleg	http://hazyresearch.stanford.edu/bootleg	SYM
orr-bootleg-2020	44	10	limited	limited	JJ
orr-bootleg-2020	44	11	portion	portion	NN
orr-bootleg-2020	44	12	of	of	IN
orr-bootleg-2020	44	13	the	the	DT
orr-bootleg-2020	44	14	standard	standard	JJ
orr-bootleg-2020	44	15	NED	NED	NNP
orr-bootleg-2020	44	16	training	training	NN
orr-bootleg-2020	44	17	dataset	dataset	NN
orr-bootleg-2020	44	18	,	,	,
orr-bootleg-2020	44	19	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	44	20	,	,	,
orr-bootleg-2020	44	21	is	be	VBZ
orr-bootleg-2020	44	22	useful	useful	JJ
orr-bootleg-2020	44	23	:	:	:
orr-bootleg-2020	44	24	Wikipedia	wikipedia	NN
orr-bootleg-2020	44	25	lacks	lack	VBZ
orr-bootleg-2020	44	26	labels	label	NNS
orr-bootleg-2020	44	27	[	[	-LRB-
orr-bootleg-2020	44	28	19	19	CD
orr-bootleg-2020	44	29	]	]	-RRB-
orr-bootleg-2020	44	30	and	and	CC
orr-bootleg-2020	44	31	we	-PRON-	PRP
orr-bootleg-2020	44	32	find	find	VBP
orr-bootleg-2020	44	33	that	that	IN
orr-bootleg-2020	44	34	an	an	DT
orr-bootleg-2020	44	35	estimated	estimate	VBN
orr-bootleg-2020	44	36	68	68	CD
orr-bootleg-2020	44	37	%	%	NN
orr-bootleg-2020	44	38	of	of	IN
orr-bootleg-2020	44	39	entities	entity	NNS
orr-bootleg-2020	44	40	in	in	IN
orr-bootleg-2020	44	41	the	the	DT
orr-bootleg-2020	44	42	dataset	dataset	NN
orr-bootleg-2020	44	43	are	be	VBP
orr-bootleg-2020	44	44	not	not	RB
orr-bootleg-2020	44	45	labeled.4	labeled.4	DT
orr-bootleg-2020	44	46	Bootleg	Bootleg	NNP
orr-bootleg-2020	44	47	addresses	address	VBZ
orr-bootleg-2020	44	48	these	these	DT
orr-bootleg-2020	44	49	challenges	challenge	NNS
orr-bootleg-2020	44	50	through	through	IN
orr-bootleg-2020	44	51	three	three	CD
orr-bootleg-2020	44	52	contributions	contribution	NNS
orr-bootleg-2020	44	53	:	:	:
orr-bootleg-2020	44	54	•	•	XX
orr-bootleg-2020	44	55	Reasoning	Reasoning	NNP
orr-bootleg-2020	44	56	Patterns	Patterns	NNPS
orr-bootleg-2020	44	57	for	for	IN
orr-bootleg-2020	44	58	Disambiguation	Disambiguation	NNP
orr-bootleg-2020	44	59	:	:	:
orr-bootleg-2020	44	60	We	-PRON-	PRP
orr-bootleg-2020	44	61	contribute	contribute	VBP
orr-bootleg-2020	44	62	a	a	DT
orr-bootleg-2020	44	63	principled	principle	VBN
orr-bootleg-2020	44	64	set	set	NN
orr-bootleg-2020	44	65	of	of	IN
orr-bootleg-2020	44	66	core	core	JJ
orr-bootleg-2020	44	67	disambiguation	disambiguation	NN
orr-bootleg-2020	44	68	patterns	pattern	NNS
orr-bootleg-2020	44	69	for	for	IN
orr-bootleg-2020	44	70	NED	NED	NNP
orr-bootleg-2020	44	71	(	(	-LRB-
orr-bootleg-2020	44	72	Figure	Figure	NNP
orr-bootleg-2020	44	73	1	1	CD
orr-bootleg-2020	44	74	(	(	-LRB-
orr-bootleg-2020	44	75	left))—entity	left))—entity	NNP
orr-bootleg-2020	44	76	memorization	memorization	NN
orr-bootleg-2020	44	77	,	,	,
orr-bootleg-2020	44	78	type	type	NN
orr-bootleg-2020	44	79	consistency	consistency	NN
orr-bootleg-2020	44	80	,	,	,
orr-bootleg-2020	44	81	KG	kg	NN
orr-bootleg-2020	44	82	relation	relation	NN
orr-bootleg-2020	44	83	,	,	,
orr-bootleg-2020	44	84	and	and	CC
orr-bootleg-2020	44	85	type	type	NN
orr-bootleg-2020	44	86	affordance	affordance	NN
orr-bootleg-2020	44	87	—	—	:
orr-bootleg-2020	44	88	and	and	CC
orr-bootleg-2020	44	89	show	show	VBP
orr-bootleg-2020	44	90	that	that	IN
orr-bootleg-2020	44	91	on	on	IN
orr-bootleg-2020	44	92	slices	slice	NNS
orr-bootleg-2020	44	93	of	of	IN
orr-bootleg-2020	44	94	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	44	95	examples	example	NNS
orr-bootleg-2020	44	96	exemplifying	exemplify	VBG
orr-bootleg-2020	44	97	each	each	DT
orr-bootleg-2020	44	98	pattern	pattern	NN
orr-bootleg-2020	44	99	,	,	,
orr-bootleg-2020	44	100	Bootleg	Bootleg	NNP
orr-bootleg-2020	44	101	provides	provide	VBZ
orr-bootleg-2020	44	102	a	a	DT
orr-bootleg-2020	44	103	lift	lift	NN
orr-bootleg-2020	44	104	over	over	IN
orr-bootleg-2020	44	105	the	the	DT
orr-bootleg-2020	44	106	baseline	baseline	NN
orr-bootleg-2020	44	107	SotA	SotA	NNP
orr-bootleg-2020	44	108	model	model	NN
orr-bootleg-2020	44	109	on	on	IN
orr-bootleg-2020	44	110	tail	tail	NN
orr-bootleg-2020	44	111	examples	example	NNS
orr-bootleg-2020	44	112	by	by	IN
orr-bootleg-2020	44	113	18	18	CD
orr-bootleg-2020	44	114	F1	f1	NN
orr-bootleg-2020	44	115	,	,	,
orr-bootleg-2020	44	116	56	56	CD
orr-bootleg-2020	44	117	F1	f1	NN
orr-bootleg-2020	44	118	,	,	,
orr-bootleg-2020	44	119	62	62	CD
orr-bootleg-2020	44	120	F1	f1	NN
orr-bootleg-2020	44	121	,	,	,
orr-bootleg-2020	44	122	and	and	CC
orr-bootleg-2020	44	123	45	45	CD
orr-bootleg-2020	44	124	F1	f1	NN
orr-bootleg-2020	44	125	points	point	NNS
orr-bootleg-2020	44	126	respectively	respectively	RB
orr-bootleg-2020	44	127	.	.	.
orr-bootleg-2020	45	1	Overall	overall	RB
orr-bootleg-2020	45	2	,	,	,
orr-bootleg-2020	45	3	using	use	VBG
orr-bootleg-2020	45	4	these	these	DT
orr-bootleg-2020	45	5	patterns	pattern	NNS
orr-bootleg-2020	45	6	,	,	,
orr-bootleg-2020	45	7	Bootleg	Bootleg	NNP
orr-bootleg-2020	45	8	meets	meet	VBZ
orr-bootleg-2020	45	9	or	or	CC
orr-bootleg-2020	45	10	exceeds	exceed	VBZ
orr-bootleg-2020	45	11	state	state	NN
orr-bootleg-2020	45	12	-	-	HYPH
orr-bootleg-2020	45	13	of	of	IN
orr-bootleg-2020	45	14	-	-	HYPH
orr-bootleg-2020	45	15	the	the	DT
orr-bootleg-2020	45	16	-	-	HYPH
orr-bootleg-2020	45	17	art	art	NN
orr-bootleg-2020	45	18	performance	performance	NN
orr-bootleg-2020	45	19	on	on	IN
orr-bootleg-2020	45	20	three	three	CD
orr-bootleg-2020	45	21	NED	NED	NNP
orr-bootleg-2020	45	22	benchmarks	benchmark	NNS
orr-bootleg-2020	45	23	and	and	CC
orr-bootleg-2020	45	24	outperforms	outperform	VBZ
orr-bootleg-2020	45	25	the	the	DT
orr-bootleg-2020	45	26	prior	prior	JJ
orr-bootleg-2020	45	27	SotA	SotA	NNP
orr-bootleg-2020	45	28	by	by	IN
orr-bootleg-2020	45	29	more	more	JJR
orr-bootleg-2020	45	30	than	than	IN
orr-bootleg-2020	45	31	40	40	CD
orr-bootleg-2020	45	32	F1	f1	NN
orr-bootleg-2020	45	33	points	point	NNS
orr-bootleg-2020	45	34	on	on	IN
orr-bootleg-2020	45	35	the	the	DT
orr-bootleg-2020	45	36	tail	tail	NN
orr-bootleg-2020	45	37	of	of	IN
orr-bootleg-2020	45	38	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	45	39	.	.	.
orr-bootleg-2020	46	1	•	•	VB
orr-bootleg-2020	46	2	Generalizing	generalize	VBG
orr-bootleg-2020	46	3	Learning	learn	VBG
orr-bootleg-2020	46	4	to	to	IN
orr-bootleg-2020	46	5	the	the	DT
orr-bootleg-2020	46	6	Tail	tail	NN
orr-bootleg-2020	46	7	:	:	:
orr-bootleg-2020	46	8	Our	-PRON-	PRP$
orr-bootleg-2020	46	9	key	key	JJ
orr-bootleg-2020	46	10	insight	insight	NN
orr-bootleg-2020	46	11	is	be	VBZ
orr-bootleg-2020	46	12	that	that	IN
orr-bootleg-2020	46	13	there	there	EX
orr-bootleg-2020	46	14	are	be	VBP
orr-bootleg-2020	46	15	distinct	distinct	JJ
orr-bootleg-2020	46	16	entity-	entity-	NN
orr-bootleg-2020	46	17	,	,	,
orr-bootleg-2020	46	18	type-	type-	NNP
orr-bootleg-2020	46	19	,	,	,
orr-bootleg-2020	46	20	and	and	CC
orr-bootleg-2020	46	21	relation-	relation-	NN
orr-bootleg-2020	46	22	tails	tail	NNS
orr-bootleg-2020	46	23	.	.	.
orr-bootleg-2020	47	1	Over	over	IN
orr-bootleg-2020	47	2	tail	tail	NN
orr-bootleg-2020	47	3	entities	entity	NNS
orr-bootleg-2020	47	4	(	(	-LRB-
orr-bootleg-2020	47	5	based	base	VBN
orr-bootleg-2020	47	6	on	on	IN
orr-bootleg-2020	47	7	entity	entity	NN
orr-bootleg-2020	47	8	count	count	NN
orr-bootleg-2020	47	9	in	in	IN
orr-bootleg-2020	47	10	the	the	DT
orr-bootleg-2020	47	11	training	training	NN
orr-bootleg-2020	47	12	data	datum	NNS
orr-bootleg-2020	47	13	)	)	-RRB-
orr-bootleg-2020	47	14	,	,	,
orr-bootleg-2020	47	15	88	88	CD
orr-bootleg-2020	47	16	%	%	NN
orr-bootleg-2020	47	17	have	have	VBP
orr-bootleg-2020	47	18	non	non	JJ
orr-bootleg-2020	47	19	-	-	JJ
orr-bootleg-2020	47	20	tail	tail	JJ
orr-bootleg-2020	47	21	types	type	NNS
orr-bootleg-2020	47	22	and	and	CC
orr-bootleg-2020	47	23	90	90	CD
orr-bootleg-2020	47	24	%	%	NN
orr-bootleg-2020	47	25	have	have	VBP
orr-bootleg-2020	47	26	non	non	JJ
orr-bootleg-2020	47	27	-	-	JJ
orr-bootleg-2020	47	28	tail	tail	JJ
orr-bootleg-2020	47	29	relations	relation	NNS
orr-bootleg-2020	47	30	.	.	.
orr-bootleg-2020	48	1	The	the	DT
orr-bootleg-2020	48	2	model	model	NN
orr-bootleg-2020	48	3	should	should	MD
orr-bootleg-2020	48	4	balance	balance	VB
orr-bootleg-2020	48	5	these	these	DT
orr-bootleg-2020	48	6	signals	signal	NNS
orr-bootleg-2020	48	7	differently	differently	RB
orr-bootleg-2020	48	8	depending	depend	VBG
orr-bootleg-2020	48	9	on	on	IN
orr-bootleg-2020	48	10	the	the	DT
orr-bootleg-2020	48	11	particular	particular	JJ
orr-bootleg-2020	48	12	entity	entity	NN
orr-bootleg-2020	48	13	being	be	VBG
orr-bootleg-2020	48	14	disambiguated	disambiguate	VBN
orr-bootleg-2020	48	15	.	.	.
orr-bootleg-2020	49	1	We	-PRON-	PRP
orr-bootleg-2020	49	2	thus	thus	RB
orr-bootleg-2020	49	3	contribute	contribute	VBP
orr-bootleg-2020	49	4	a	a	DT
orr-bootleg-2020	49	5	new	new	JJ
orr-bootleg-2020	49	6	2D	2d	JJ
orr-bootleg-2020	49	7	regularization	regularization	NN
orr-bootleg-2020	49	8	scheme	scheme	NN
orr-bootleg-2020	49	9	to	to	TO
orr-bootleg-2020	49	10	combine	combine	VB
orr-bootleg-2020	49	11	the	the	DT
orr-bootleg-2020	49	12	entity	entity	NN
orr-bootleg-2020	49	13	,	,	,
orr-bootleg-2020	49	14	tail	tail	NN
orr-bootleg-2020	49	15	,	,	,
orr-bootleg-2020	49	16	and	and	CC
orr-bootleg-2020	49	17	relation	relation	NN
orr-bootleg-2020	49	18	signals	signal	NNS
orr-bootleg-2020	49	19	and	and	CC
orr-bootleg-2020	49	20	achieve	achieve	VB
orr-bootleg-2020	49	21	a	a	DT
orr-bootleg-2020	49	22	lift	lift	NN
orr-bootleg-2020	49	23	of	of	IN
orr-bootleg-2020	49	24	13.6	13.6	CD
orr-bootleg-2020	49	25	F1	f1	NN
orr-bootleg-2020	49	26	points	point	NNS
orr-bootleg-2020	49	27	on	on	IN
orr-bootleg-2020	49	28	unseen	unseen	JJ
orr-bootleg-2020	49	29	entities	entity	NNS
orr-bootleg-2020	49	30	compared	compare	VBN
orr-bootleg-2020	49	31	to	to	IN
orr-bootleg-2020	49	32	the	the	DT
orr-bootleg-2020	49	33	model	model	NN
orr-bootleg-2020	49	34	using	use	VBG
orr-bootleg-2020	49	35	standard	standard	JJ
orr-bootleg-2020	49	36	regularization	regularization	NN
orr-bootleg-2020	49	37	techniques	technique	NNS
orr-bootleg-2020	49	38	.	.	.
orr-bootleg-2020	50	1	We	-PRON-	PRP
orr-bootleg-2020	50	2	conduct	conduct	VBP
orr-bootleg-2020	50	3	extensive	extensive	JJ
orr-bootleg-2020	50	4	ablation	ablation	NN
orr-bootleg-2020	50	5	studies	study	NNS
orr-bootleg-2020	50	6	to	to	TO
orr-bootleg-2020	50	7	verify	verify	VB
orr-bootleg-2020	50	8	the	the	DT
orr-bootleg-2020	50	9	effectiveness	effectiveness	NN
orr-bootleg-2020	50	10	of	of	IN
orr-bootleg-2020	50	11	our	-PRON-	PRP$
orr-bootleg-2020	50	12	approach	approach	NN
orr-bootleg-2020	50	13	.	.	.
orr-bootleg-2020	51	1	•	•	VB
orr-bootleg-2020	51	2	Weak	weak	JJ
orr-bootleg-2020	51	3	Labelling	labelling	NN
orr-bootleg-2020	51	4	of	of	IN
orr-bootleg-2020	51	5	Data	Data	NNPS
orr-bootleg-2020	51	6	:	:	:
orr-bootleg-2020	51	7	Our	-PRON-	PRP$
orr-bootleg-2020	51	8	insight	insight	NN
orr-bootleg-2020	51	9	is	be	VBZ
orr-bootleg-2020	51	10	that	that	IN
orr-bootleg-2020	51	11	because	because	IN
orr-bootleg-2020	51	12	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	51	13	is	be	VBZ
orr-bootleg-2020	51	14	highly	highly	RB
orr-bootleg-2020	51	15	structured	structured	JJ
orr-bootleg-2020	51	16	—	—	:
orr-bootleg-2020	51	17	most	most	JJS
orr-bootleg-2020	51	18	sentences	sentence	NNS
orr-bootleg-2020	51	19	on	on	IN
orr-bootleg-2020	51	20	an	an	DT
orr-bootleg-2020	51	21	entity	entity	NN
orr-bootleg-2020	51	22	’s	’s	POS
orr-bootleg-2020	51	23	Wikipedia	wikipedia	NN
orr-bootleg-2020	51	24	page	page	NN
orr-bootleg-2020	51	25	refer	refer	NN
orr-bootleg-2020	51	26	to	to	IN
orr-bootleg-2020	51	27	that	that	DT
orr-bootleg-2020	51	28	entity	entity	NN
orr-bootleg-2020	51	29	via	via	IN
orr-bootleg-2020	51	30	pronouns	pronoun	NNS
orr-bootleg-2020	51	31	or	or	CC
orr-bootleg-2020	51	32	alternative	alternative	JJ
orr-bootleg-2020	51	33	names	name	NNS
orr-bootleg-2020	51	34	—	—	:
orr-bootleg-2020	51	35	we	-PRON-	PRP
orr-bootleg-2020	51	36	can	can	MD
orr-bootleg-2020	51	37	weakly	weakly	RB
orr-bootleg-2020	51	38	label	label	VB
orr-bootleg-2020	51	39	our	-PRON-	PRP$
orr-bootleg-2020	51	40	training	training	NN
orr-bootleg-2020	51	41	data	datum	NNS
orr-bootleg-2020	51	42	to	to	IN
orr-bootleg-2020	51	43	label	label	NN
orr-bootleg-2020	51	44	mentions	mention	NNS
orr-bootleg-2020	51	45	.	.	.
orr-bootleg-2020	52	1	Through	through	IN
orr-bootleg-2020	52	2	weak	weak	JJ
orr-bootleg-2020	52	3	labeling	labeling	NN
orr-bootleg-2020	52	4	,	,	,
orr-bootleg-2020	52	5	we	-PRON-	PRP
orr-bootleg-2020	52	6	increase	increase	VBP
orr-bootleg-2020	52	7	the	the	DT
orr-bootleg-2020	52	8	number	number	NN
orr-bootleg-2020	52	9	of	of	IN
orr-bootleg-2020	52	10	labeled	label	VBN
orr-bootleg-2020	52	11	mentions	mention	NNS
orr-bootleg-2020	52	12	in	in	IN
orr-bootleg-2020	52	13	the	the	DT
orr-bootleg-2020	52	14	training	training	NN
orr-bootleg-2020	52	15	data	datum	NNS
orr-bootleg-2020	52	16	by	by	IN
orr-bootleg-2020	52	17	1.7x	1.7x	CD
orr-bootleg-2020	52	18	,	,	,
orr-bootleg-2020	52	19	and	and	CC
orr-bootleg-2020	52	20	find	find	VB
orr-bootleg-2020	52	21	this	this	DT
orr-bootleg-2020	52	22	provides	provide	VBZ
orr-bootleg-2020	52	23	a	a	DT
orr-bootleg-2020	52	24	2.6	2.6	CD
orr-bootleg-2020	52	25	F1	f1	NN
orr-bootleg-2020	52	26	point	point	NN
orr-bootleg-2020	52	27	lift	lift	NN
orr-bootleg-2020	52	28	on	on	IN
orr-bootleg-2020	52	29	unseen	unseen	JJ
orr-bootleg-2020	52	30	entities	entity	NNS
orr-bootleg-2020	52	31	.	.	.
orr-bootleg-2020	53	1	With	with	IN
orr-bootleg-2020	53	2	these	these	DT
orr-bootleg-2020	53	3	three	three	CD
orr-bootleg-2020	53	4	contributions	contribution	NNS
orr-bootleg-2020	53	5	,	,	,
orr-bootleg-2020	53	6	Bootleg	Bootleg	NNP
orr-bootleg-2020	53	7	achieves	achieve	VBZ
orr-bootleg-2020	53	8	SotA	SotA	NNP
orr-bootleg-2020	53	9	on	on	IN
orr-bootleg-2020	53	10	three	three	CD
orr-bootleg-2020	53	11	NED	NED	NNP
orr-bootleg-2020	53	12	benchmarks	benchmark	NNS
orr-bootleg-2020	53	13	.	.	.
orr-bootleg-2020	54	1	We	-PRON-	PRP
orr-bootleg-2020	54	2	further	further	RB
orr-bootleg-2020	54	3	show	show	VBP
orr-bootleg-2020	54	4	that	that	IN
orr-bootleg-2020	54	5	embeddings	embedding	NNS
orr-bootleg-2020	54	6	from	from	IN
orr-bootleg-2020	54	7	Bootleg	Bootleg	NNP
orr-bootleg-2020	54	8	are	be	VBP
orr-bootleg-2020	54	9	useful	useful	JJ
orr-bootleg-2020	54	10	for	for	IN
orr-bootleg-2020	54	11	downstream	downstream	JJ
orr-bootleg-2020	54	12	applications	application	NNS
orr-bootleg-2020	54	13	that	that	WDT
orr-bootleg-2020	54	14	require	require	VBP
orr-bootleg-2020	54	15	the	the	DT
orr-bootleg-2020	54	16	knowledge	knowledge	NN
orr-bootleg-2020	54	17	of	of	IN
orr-bootleg-2020	54	18	entities	entity	NNS
orr-bootleg-2020	54	19	.	.	.
orr-bootleg-2020	55	1	We	-PRON-	PRP
orr-bootleg-2020	55	2	show	show	VBP
orr-bootleg-2020	55	3	the	the	DT
orr-bootleg-2020	55	4	reasoning	reasoning	NN
orr-bootleg-2020	55	5	patterns	pattern	NNS
orr-bootleg-2020	55	6	learned	learn	VBN
orr-bootleg-2020	55	7	in	in	IN
orr-bootleg-2020	55	8	Bootleg	Bootleg	NNP
orr-bootleg-2020	55	9	transfer	transfer	VBP
orr-bootleg-2020	55	10	to	to	IN
orr-bootleg-2020	55	11	tasks	task	NNS
orr-bootleg-2020	55	12	beyond	beyond	IN
orr-bootleg-2020	55	13	NED	NED	NNP
orr-bootleg-2020	55	14	by	by	IN
orr-bootleg-2020	55	15	extracting	extract	VBG
orr-bootleg-2020	55	16	Bootleg	Bootleg	NNP
orr-bootleg-2020	55	17	’s	’s	POS
orr-bootleg-2020	55	18	learned	learn	VBD
orr-bootleg-2020	55	19	embeddings	embedding	NNS
orr-bootleg-2020	55	20	and	and	CC
orr-bootleg-2020	55	21	using	use	VBG
orr-bootleg-2020	55	22	them	-PRON-	PRP
orr-bootleg-2020	55	23	to	to	TO
orr-bootleg-2020	55	24	set	set	VB
orr-bootleg-2020	55	25	a	a	DT
orr-bootleg-2020	55	26	new	new	JJ
orr-bootleg-2020	55	27	SotA	SotA	NNP
orr-bootleg-2020	55	28	by	by	IN
orr-bootleg-2020	55	29	1.0	1.0	CD
orr-bootleg-2020	55	30	F1	f1	NN
orr-bootleg-2020	55	31	points	point	NNS
orr-bootleg-2020	55	32	on	on	IN
orr-bootleg-2020	55	33	the	the	DT
orr-bootleg-2020	55	34	TACRED	TACRED	NNP
orr-bootleg-2020	55	35	relation	relation	NN
orr-bootleg-2020	55	36	extraction	extraction	NN
orr-bootleg-2020	55	37	task	task	NN
orr-bootleg-2020	55	38	[	[	-LRB-
orr-bootleg-2020	55	39	2	2	CD
orr-bootleg-2020	55	40	,	,	,
orr-bootleg-2020	55	41	53	53	CD
orr-bootleg-2020	55	42	]	]	-RRB-
orr-bootleg-2020	55	43	,	,	,
orr-bootleg-2020	55	44	where	where	WRB
orr-bootleg-2020	55	45	the	the	DT
orr-bootleg-2020	55	46	prior	prior	JJ
orr-bootleg-2020	55	47	SotA	SotA	NNP
orr-bootleg-2020	55	48	model	model	NN
orr-bootleg-2020	55	49	also	also	RB
orr-bootleg-2020	55	50	uses	use	VBZ
orr-bootleg-2020	55	51	entity	entity	NN
orr-bootleg-2020	55	52	-	-	HYPH
orr-bootleg-2020	55	53	based	base	VBN
orr-bootleg-2020	55	54	knowledge	knowledge	NN
orr-bootleg-2020	55	55	[	[	-LRB-
orr-bootleg-2020	55	56	38	38	CD
orr-bootleg-2020	55	57	]	]	-RRB-
orr-bootleg-2020	55	58	.	.	.
orr-bootleg-2020	56	1	Bootleg	bootleg	NN
orr-bootleg-2020	56	2	representations	representation	NNS
orr-bootleg-2020	56	3	further	further	RB
orr-bootleg-2020	56	4	provide	provide	VBP
orr-bootleg-2020	56	5	an	an	DT
orr-bootleg-2020	56	6	8	8	CD
orr-bootleg-2020	56	7	%	%	NN
orr-bootleg-2020	56	8	performance	performance	NN
orr-bootleg-2020	56	9	lift	lift	NN
orr-bootleg-2020	56	10	on	on	IN
orr-bootleg-2020	56	11	highly	highly	RB
orr-bootleg-2020	56	12	optimized	optimize	VBN
orr-bootleg-2020	56	13	industrial	industrial	JJ
orr-bootleg-2020	56	14	search	search	NN
orr-bootleg-2020	56	15	and	and	CC
orr-bootleg-2020	56	16	assistant	assistant	NN
orr-bootleg-2020	56	17	tasks	task	NNS
orr-bootleg-2020	56	18	at	at	IN
orr-bootleg-2020	56	19	a	a	DT
orr-bootleg-2020	56	20	major	major	JJ
orr-bootleg-2020	56	21	technology	technology	NN
orr-bootleg-2020	56	22	company	company	NN
orr-bootleg-2020	56	23	.	.	.
orr-bootleg-2020	57	1	For	for	IN
orr-bootleg-2020	57	2	Bootleg	Bootleg	NNP
orr-bootleg-2020	57	3	’s	’s	POS
orr-bootleg-2020	57	4	embeddings	embedding	NNS
orr-bootleg-2020	57	5	to	to	TO
orr-bootleg-2020	57	6	be	be	VB
orr-bootleg-2020	57	7	viable	viable	JJ
orr-bootleg-2020	57	8	for	for	IN
orr-bootleg-2020	57	9	production	production	NN
orr-bootleg-2020	57	10	,	,	,
orr-bootleg-2020	57	11	it	-PRON-	PRP
orr-bootleg-2020	57	12	is	be	VBZ
orr-bootleg-2020	57	13	critical	critical	JJ
orr-bootleg-2020	57	14	that	that	IN
orr-bootleg-2020	57	15	these	these	DT
orr-bootleg-2020	57	16	models	model	NNS
orr-bootleg-2020	57	17	are	be	VBP
orr-bootleg-2020	57	18	space	space	NN
orr-bootleg-2020	57	19	-	-	HYPH
orr-bootleg-2020	57	20	efficient	efficient	JJ
orr-bootleg-2020	57	21	:	:	:
orr-bootleg-2020	57	22	the	the	DT
orr-bootleg-2020	57	23	models	model	NNS
orr-bootleg-2020	57	24	using	use	VBG
orr-bootleg-2020	57	25	only	only	RB
orr-bootleg-2020	57	26	Bootleg	Bootleg	NNP
orr-bootleg-2020	57	27	relation	relation	NN
orr-bootleg-2020	57	28	and	and	CC
orr-bootleg-2020	57	29	type	type	NN
orr-bootleg-2020	57	30	embeddings	embedding	NNS
orr-bootleg-2020	57	31	each	each	DT
orr-bootleg-2020	57	32	achieve	achieve	VBP
orr-bootleg-2020	57	33	3.3x	3.3x	CD
orr-bootleg-2020	57	34	the	the	DT
orr-bootleg-2020	57	35	performance	performance	NN
orr-bootleg-2020	57	36	of	of	IN
orr-bootleg-2020	57	37	the	the	DT
orr-bootleg-2020	57	38	prior	prior	JJ
orr-bootleg-2020	57	39	SotA	SotA	NNP
orr-bootleg-2020	57	40	baseline	baseline	NNP
orr-bootleg-2020	57	41	over	over	IN
orr-bootleg-2020	57	42	unseen	unseen	JJ
orr-bootleg-2020	57	43	entities	entity	NNS
orr-bootleg-2020	57	44	using	use	VBG
orr-bootleg-2020	57	45	1	1	CD
orr-bootleg-2020	57	46	%	%	NN
orr-bootleg-2020	57	47	of	of	IN
orr-bootleg-2020	57	48	the	the	DT
orr-bootleg-2020	57	49	space	space	NN
orr-bootleg-2020	57	50	.	.	.
orr-bootleg-2020	58	1	2	2	CD
orr-bootleg-2020	58	2	NED	NED	NNP
orr-bootleg-2020	58	3	Overview	Overview	NNP
orr-bootleg-2020	58	4	and	and	CC
orr-bootleg-2020	58	5	Reasoning	Reasoning	NNP
orr-bootleg-2020	58	6	Patterns	Patterns	NNPS
orr-bootleg-2020	58	7	We	-PRON-	PRP
orr-bootleg-2020	58	8	now	now	RB
orr-bootleg-2020	58	9	define	define	VBP
orr-bootleg-2020	58	10	the	the	DT
orr-bootleg-2020	58	11	task	task	NN
orr-bootleg-2020	58	12	of	of	IN
orr-bootleg-2020	58	13	named	name	VBN
orr-bootleg-2020	58	14	entity	entity	NN
orr-bootleg-2020	58	15	disambiguation	disambiguation	NN
orr-bootleg-2020	58	16	(	(	-LRB-
orr-bootleg-2020	58	17	NED	NED	NNP
orr-bootleg-2020	58	18	)	)	-RRB-
orr-bootleg-2020	58	19	,	,	,
orr-bootleg-2020	58	20	the	the	DT
orr-bootleg-2020	58	21	four	four	CD
orr-bootleg-2020	58	22	core	core	NN
orr-bootleg-2020	58	23	reasoning	reason	VBG
orr-bootleg-2020	58	24	patterns	pattern	NNS
orr-bootleg-2020	58	25	,	,	,
orr-bootleg-2020	58	26	and	and	CC
orr-bootleg-2020	58	27	the	the	DT
orr-bootleg-2020	58	28	structural	structural	JJ
orr-bootleg-2020	58	29	resources	resource	NNS
orr-bootleg-2020	58	30	required	require	VBN
orr-bootleg-2020	58	31	for	for	IN
orr-bootleg-2020	58	32	learning	learn	VBG
orr-bootleg-2020	58	33	the	the	DT
orr-bootleg-2020	58	34	patterns	pattern	NNS
orr-bootleg-2020	58	35	.	.	.
orr-bootleg-2020	59	1	Task	task	NN
orr-bootleg-2020	59	2	Definition	definition	NN
orr-bootleg-2020	59	3	Given	give	VBD
orr-bootleg-2020	59	4	a	a	DT
orr-bootleg-2020	59	5	knowledge	knowledge	NN
orr-bootleg-2020	59	6	base	base	NN
orr-bootleg-2020	59	7	of	of	IN
orr-bootleg-2020	59	8	entities	entity	NNS
orr-bootleg-2020	59	9	E	e	NN
orr-bootleg-2020	59	10	and	and	CC
orr-bootleg-2020	59	11	an	an	DT
orr-bootleg-2020	59	12	input	input	NN
orr-bootleg-2020	59	13	sentence	sentence	NN
orr-bootleg-2020	59	14	,	,	,
orr-bootleg-2020	59	15	the	the	DT
orr-bootleg-2020	59	16	goal	goal	NN
orr-bootleg-2020	59	17	of	of	IN
orr-bootleg-2020	59	18	named	name	VBN
orr-bootleg-2020	59	19	entity	entity	NN
orr-bootleg-2020	59	20	disambiguation	disambiguation	NN
orr-bootleg-2020	59	21	is	be	VBZ
orr-bootleg-2020	59	22	to	to	TO
orr-bootleg-2020	59	23	determine	determine	VB
orr-bootleg-2020	59	24	the	the	DT
orr-bootleg-2020	59	25	entities	entity	NNS
orr-bootleg-2020	59	26	e	e	LS
orr-bootleg-2020	59	27	∈	∈	NNP
orr-bootleg-2020	59	28	E	e	NN
orr-bootleg-2020	59	29	referenced	reference	VBN
orr-bootleg-2020	59	30	in	in	IN
orr-bootleg-2020	59	31	each	each	DT
orr-bootleg-2020	59	32	sentence	sentence	NN
orr-bootleg-2020	59	33	.	.	.
orr-bootleg-2020	60	1	Specifically	specifically	RB
orr-bootleg-2020	60	2	,	,	,
orr-bootleg-2020	60	3	the	the	DT
orr-bootleg-2020	60	4	input	input	NN
orr-bootleg-2020	60	5	is	be	VBZ
orr-bootleg-2020	60	6	a	a	DT
orr-bootleg-2020	60	7	sequence	sequence	NN
orr-bootleg-2020	60	8	of	of	IN
orr-bootleg-2020	60	9	N	n	NN
orr-bootleg-2020	60	10	tokens	token	NNS
orr-bootleg-2020	60	11	W	w	NN
orr-bootleg-2020	60	12	=	=	NFP
orr-bootleg-2020	60	13	{	{	-LRB-
orr-bootleg-2020	60	14	w1	w1	NNP
orr-bootleg-2020	60	15	,	,	,
orr-bootleg-2020	60	16	.	.	.
orr-bootleg-2020	61	1	.	.	.
orr-bootleg-2020	62	1	.	.	.
orr-bootleg-2020	63	1	,	,	,
orr-bootleg-2020	63	2	wN	wN	NNP
orr-bootleg-2020	63	3	}	}	-RRB-
orr-bootleg-2020	63	4	and	and	CC
orr-bootleg-2020	63	5	a	a	DT
orr-bootleg-2020	63	6	set	set	NN
orr-bootleg-2020	63	7	of	of	IN
orr-bootleg-2020	63	8	M	M	NNP
orr-bootleg-2020	63	9	non	non	JJ
orr-bootleg-2020	63	10	-	-	JJ
orr-bootleg-2020	63	11	overlapping	overlapping	JJ
orr-bootleg-2020	63	12	spans	span	NNS
orr-bootleg-2020	63	13	in	in	IN
orr-bootleg-2020	63	14	the	the	DT
orr-bootleg-2020	63	15	sequence	sequence	NN
orr-bootleg-2020	63	16	W	W	NNP
orr-bootleg-2020	63	17	,	,	,
orr-bootleg-2020	63	18	termed	term	VBN
orr-bootleg-2020	63	19	mentions	mention	NNS
orr-bootleg-2020	63	20	,	,	,
orr-bootleg-2020	63	21	to	to	TO
orr-bootleg-2020	63	22	be	be	VB
orr-bootleg-2020	63	23	disambiguated	disambiguate	VBN
orr-bootleg-2020	63	24	M	m	NN
orr-bootleg-2020	63	25	=	=	NFP
orr-bootleg-2020	63	26	{	{	-LRB-
orr-bootleg-2020	63	27	m1	m1	NNP
orr-bootleg-2020	63	28	,	,	,
orr-bootleg-2020	63	29	.	.	.
orr-bootleg-2020	64	1	.	.	.
orr-bootleg-2020	65	1	.	.	.
orr-bootleg-2020	66	1	,	,	,
orr-bootleg-2020	66	2	mM	mM	NNP
orr-bootleg-2020	66	3	}	}	-RRB-
orr-bootleg-2020	66	4	.	.	.
orr-bootleg-2020	67	1	The	the	DT
orr-bootleg-2020	67	2	output	output	NN
orr-bootleg-2020	67	3	is	be	VBZ
orr-bootleg-2020	67	4	the	the	DT
orr-bootleg-2020	67	5	most	most	RBS
orr-bootleg-2020	67	6	likely	likely	JJ
orr-bootleg-2020	67	7	entity	entity	NN
orr-bootleg-2020	67	8	for	for	IN
orr-bootleg-2020	67	9	each	each	DT
orr-bootleg-2020	67	10	mention	mention	NN
orr-bootleg-2020	67	11	.	.	.
orr-bootleg-2020	68	1	The	the	DT
orr-bootleg-2020	68	2	Tail	Tail	NNP
orr-bootleg-2020	68	3	of	of	IN
orr-bootleg-2020	68	4	NED	NED	NNP
orr-bootleg-2020	68	5	We	-PRON-	PRP
orr-bootleg-2020	68	6	define	define	VBP
orr-bootleg-2020	68	7	the	the	DT
orr-bootleg-2020	68	8	tail	tail	NN
orr-bootleg-2020	68	9	,	,	,
orr-bootleg-2020	68	10	torso	torso	VB
orr-bootleg-2020	68	11	,	,	,
orr-bootleg-2020	68	12	and	and	CC
orr-bootleg-2020	68	13	head	head	NN
orr-bootleg-2020	68	14	of	of	IN
orr-bootleg-2020	68	15	NED	NED	NNP
orr-bootleg-2020	68	16	as	as	IN
orr-bootleg-2020	68	17	entities	entity	NNS
orr-bootleg-2020	68	18	occurring	occur	VBG
orr-bootleg-2020	68	19	less	less	JJR
orr-bootleg-2020	68	20	than	than	IN
orr-bootleg-2020	68	21	11	11	CD
orr-bootleg-2020	68	22	times	time	NNS
orr-bootleg-2020	68	23	,	,	,
orr-bootleg-2020	68	24	between	between	IN
orr-bootleg-2020	68	25	11	11	CD
orr-bootleg-2020	68	26	and	and	CC
orr-bootleg-2020	68	27	1,000	1,000	CD
orr-bootleg-2020	68	28	,	,	,
orr-bootleg-2020	68	29	and	and	CC
orr-bootleg-2020	68	30	more	more	JJR
orr-bootleg-2020	68	31	than	than	IN
orr-bootleg-2020	68	32	1,000	1,000	CD
orr-bootleg-2020	68	33	times	time	NNS
orr-bootleg-2020	68	34	in	in	IN
orr-bootleg-2020	68	35	training	training	NN
orr-bootleg-2020	68	36	,	,	,
orr-bootleg-2020	68	37	respectively	respectively	RB
orr-bootleg-2020	68	38	.	.	.
orr-bootleg-2020	69	1	Following	follow	VBG
orr-bootleg-2020	69	2	Figure	figure	NN
orr-bootleg-2020	69	3	1	1	CD
orr-bootleg-2020	69	4	(	(	-LRB-
orr-bootleg-2020	69	5	right	right	NN
orr-bootleg-2020	69	6	)	)	-RRB-
orr-bootleg-2020	69	7	,	,	,
orr-bootleg-2020	69	8	the	the	DT
orr-bootleg-2020	69	9	head	head	NN
orr-bootleg-2020	69	10	represents	represent	VBZ
orr-bootleg-2020	69	11	those	those	DT
orr-bootleg-2020	69	12	entities	entity	NNS
orr-bootleg-2020	69	13	a	a	DT
orr-bootleg-2020	69	14	simple	simple	JJ
orr-bootleg-2020	69	15	language	language	NN
orr-bootleg-2020	69	16	-	-	HYPH
orr-bootleg-2020	69	17	based	base	VBN
orr-bootleg-2020	69	18	baseline	baseline	NN
orr-bootleg-2020	69	19	model	model	NN
orr-bootleg-2020	69	20	can	can	MD
orr-bootleg-2020	69	21	easily	easily	RB
orr-bootleg-2020	69	22	resolve	resolve	VB
orr-bootleg-2020	69	23	,	,	,
orr-bootleg-2020	69	24	as	as	IN
orr-bootleg-2020	69	25	shown	show	VBN
orr-bootleg-2020	69	26	by	by	IN
orr-bootleg-2020	69	27	a	a	DT
orr-bootleg-2020	69	28	baseline	baseline	JJ
orr-bootleg-2020	69	29	SotA	SotA	NNP
orr-bootleg-2020	69	30	model	model	NN
orr-bootleg-2020	69	31	from	from	IN
orr-bootleg-2020	69	32	[	[	-LRB-
orr-bootleg-2020	69	33	16	16	CD
orr-bootleg-2020	69	34	]	]	-RRB-
orr-bootleg-2020	69	35	achieving	achieve	VBG
orr-bootleg-2020	69	36	86	86	CD
orr-bootleg-2020	69	37	F1	f1	NN
orr-bootleg-2020	69	38	over	over	IN
orr-bootleg-2020	69	39	all	all	DT
orr-bootleg-2020	69	40	entities	entity	NNS
orr-bootleg-2020	69	41	.	.	.
orr-bootleg-2020	70	1	These	these	DT
orr-bootleg-2020	70	2	entities	entity	NNS
orr-bootleg-2020	70	3	were	be	VBD
orr-bootleg-2020	70	4	seen	see	VBN
orr-bootleg-2020	70	5	enough	enough	JJ
orr-bootleg-2020	70	6	times	time	NNS
orr-bootleg-2020	70	7	during	during	IN
orr-bootleg-2020	70	8	training	training	NN
orr-bootleg-2020	70	9	to	to	TO
orr-bootleg-2020	70	10	memorize	memorize	VB
orr-bootleg-2020	70	11	distinguishing	distinguish	VBG
orr-bootleg-2020	70	12	contextual	contextual	JJ
orr-bootleg-2020	70	13	cues	cue	NNS
orr-bootleg-2020	70	14	.	.	.
orr-bootleg-2020	71	1	The	the	DT
orr-bootleg-2020	71	2	tail	tail	NN
orr-bootleg-2020	71	3	represents	represent	VBZ
orr-bootleg-2020	71	4	the	the	DT
orr-bootleg-2020	71	5	entities	entity	NNS
orr-bootleg-2020	71	6	these	these	DT
orr-bootleg-2020	71	7	models	model	NNS
orr-bootleg-2020	71	8	struggle	struggle	VBP
orr-bootleg-2020	71	9	to	to	TO
orr-bootleg-2020	71	10	resolve	resolve	VB
orr-bootleg-2020	71	11	due	due	IN
orr-bootleg-2020	71	12	to	to	IN
orr-bootleg-2020	71	13	their	-PRON-	PRP$
orr-bootleg-2020	71	14	rarity	rarity	NN
orr-bootleg-2020	71	15	in	in	IN
orr-bootleg-2020	71	16	training	training	NN
orr-bootleg-2020	71	17	data	datum	NNS
orr-bootleg-2020	71	18	,	,	,
orr-bootleg-2020	71	19	as	as	IN
orr-bootleg-2020	71	20	shown	show	VBN
orr-bootleg-2020	71	21	by	by	IN
orr-bootleg-2020	71	22	the	the	DT
orr-bootleg-2020	71	23	same	same	JJ
orr-bootleg-2020	71	24	baseline	baseline	JJ
orr-bootleg-2020	71	25	model	model	NN
orr-bootleg-2020	71	26	achieving	achieve	VBG
orr-bootleg-2020	71	27	less	less	JJR
orr-bootleg-2020	71	28	than	than	IN
orr-bootleg-2020	71	29	28	28	CD
orr-bootleg-2020	71	30	F1	f1	NN
orr-bootleg-2020	71	31	on	on	IN
orr-bootleg-2020	71	32	the	the	DT
orr-bootleg-2020	71	33	tail	tail	NN
orr-bootleg-2020	71	34	.	.	.
orr-bootleg-2020	72	1	4We	4we	CD
orr-bootleg-2020	72	2	computed	compute	VBD
orr-bootleg-2020	72	3	this	this	DT
orr-bootleg-2020	72	4	statistic	statistic	NN
orr-bootleg-2020	72	5	by	by	IN
orr-bootleg-2020	72	6	computing	compute	VBG
orr-bootleg-2020	72	7	the	the	DT
orr-bootleg-2020	72	8	number	number	NN
orr-bootleg-2020	72	9	of	of	IN
orr-bootleg-2020	72	10	proper	proper	JJ
orr-bootleg-2020	72	11	nouns	noun	NNS
orr-bootleg-2020	72	12	and	and	CC
orr-bootleg-2020	72	13	the	the	DT
orr-bootleg-2020	72	14	number	number	NN
orr-bootleg-2020	72	15	of	of	IN
orr-bootleg-2020	72	16	pronouns	pronoun	NNS
orr-bootleg-2020	72	17	/	/	SYM
orr-bootleg-2020	72	18	known	know	VBN
orr-bootleg-2020	72	19	aliases	alias	NNS
orr-bootleg-2020	72	20	for	for	IN
orr-bootleg-2020	72	21	an	an	DT
orr-bootleg-2020	72	22	entity	entity	NN
orr-bootleg-2020	72	23	on	on	IN
orr-bootleg-2020	72	24	that	that	DT
orr-bootleg-2020	72	25	entity	entity	NN
orr-bootleg-2020	72	26	’s	’s	POS
orr-bootleg-2020	72	27	page	page	NN
orr-bootleg-2020	72	28	that	that	WDT
orr-bootleg-2020	72	29	were	be	VBD
orr-bootleg-2020	72	30	not	not	RB
orr-bootleg-2020	72	31	already	already	RB
orr-bootleg-2020	72	32	linked	link	VBN
orr-bootleg-2020	72	33	.	.	.
orr-bootleg-2020	73	1	3	3	CD
orr-bootleg-2020	73	2	2.1	2.1	CD
orr-bootleg-2020	73	3	Four	four	CD
orr-bootleg-2020	73	4	Reasoning	Reasoning	NNP
orr-bootleg-2020	73	5	Patterns	Patterns	NNPS
orr-bootleg-2020	73	6	When	when	WRB
orr-bootleg-2020	73	7	humans	human	NNS
orr-bootleg-2020	73	8	disambiguate	disambiguate	VBP
orr-bootleg-2020	73	9	entities	entity	NNS
orr-bootleg-2020	73	10	in	in	IN
orr-bootleg-2020	73	11	text	text	NN
orr-bootleg-2020	73	12	,	,	,
orr-bootleg-2020	73	13	they	-PRON-	PRP
orr-bootleg-2020	73	14	conceptually	conceptually	RB
orr-bootleg-2020	73	15	leverage	leverage	VBP
orr-bootleg-2020	73	16	signals	signal	NNS
orr-bootleg-2020	73	17	over	over	IN
orr-bootleg-2020	73	18	entities	entity	NNS
orr-bootleg-2020	73	19	,	,	,
orr-bootleg-2020	73	20	relationships	relationship	NNS
orr-bootleg-2020	73	21	,	,	,
orr-bootleg-2020	73	22	and	and	CC
orr-bootleg-2020	73	23	types	type	NNS
orr-bootleg-2020	73	24	.	.	.
orr-bootleg-2020	74	1	Our	-PRON-	PRP$
orr-bootleg-2020	74	2	empirical	empirical	JJ
orr-bootleg-2020	74	3	analysis	analysis	NN
orr-bootleg-2020	74	4	reveals	reveal	VBZ
orr-bootleg-2020	74	5	a	a	DT
orr-bootleg-2020	74	6	set	set	NN
orr-bootleg-2020	74	7	of	of	IN
orr-bootleg-2020	74	8	desirable	desirable	JJ
orr-bootleg-2020	74	9	reasoning	reasoning	NN
orr-bootleg-2020	74	10	patterns	pattern	NNS
orr-bootleg-2020	74	11	for	for	IN
orr-bootleg-2020	74	12	NED	NED	NNP
orr-bootleg-2020	74	13	.	.	.
orr-bootleg-2020	75	1	The	the	DT
orr-bootleg-2020	75	2	patterns	pattern	NNS
orr-bootleg-2020	75	3	operate	operate	VBP
orr-bootleg-2020	75	4	at	at	IN
orr-bootleg-2020	75	5	different	different	JJ
orr-bootleg-2020	75	6	levels	level	NNS
orr-bootleg-2020	75	7	of	of	IN
orr-bootleg-2020	75	8	granularity	granularity	NN
orr-bootleg-2020	75	9	(	(	-LRB-
orr-bootleg-2020	75	10	see	see	VB
orr-bootleg-2020	75	11	Figure	Figure	NNP
orr-bootleg-2020	75	12	1	1	CD
orr-bootleg-2020	75	13	(	(	-LRB-
orr-bootleg-2020	75	14	left))—from	left))—from	JJ
orr-bootleg-2020	75	15	patterns	pattern	NNS
orr-bootleg-2020	75	16	which	which	WDT
orr-bootleg-2020	75	17	are	be	VBP
orr-bootleg-2020	75	18	highly	highly	RB
orr-bootleg-2020	75	19	specific	specific	JJ
orr-bootleg-2020	75	20	to	to	IN
orr-bootleg-2020	75	21	an	an	DT
orr-bootleg-2020	75	22	entity	entity	NN
orr-bootleg-2020	75	23	,	,	,
orr-bootleg-2020	75	24	to	to	IN
orr-bootleg-2020	75	25	patterns	pattern	NNS
orr-bootleg-2020	75	26	which	which	WDT
orr-bootleg-2020	75	27	apply	apply	VBP
orr-bootleg-2020	75	28	to	to	IN
orr-bootleg-2020	75	29	categories	category	NNS
orr-bootleg-2020	75	30	of	of	IN
orr-bootleg-2020	75	31	entities	entity	NNS
orr-bootleg-2020	75	32	—	—	:
orr-bootleg-2020	75	33	and	and	CC
orr-bootleg-2020	75	34	are	be	VBP
orr-bootleg-2020	75	35	defined	define	VBN
orr-bootleg-2020	75	36	as	as	IN
orr-bootleg-2020	75	37	follows	follow	VBZ
orr-bootleg-2020	75	38	.	.	.
orr-bootleg-2020	76	1	•	•	NNP
orr-bootleg-2020	76	2	Entity	entity	NN
orr-bootleg-2020	76	3	Memorization	memorization	NN
orr-bootleg-2020	76	4	:	:	:
orr-bootleg-2020	76	5	We	-PRON-	PRP
orr-bootleg-2020	76	6	define	define	VBP
orr-bootleg-2020	76	7	entity	entity	NN
orr-bootleg-2020	76	8	memorization	memorization	NN
orr-bootleg-2020	76	9	as	as	IN
orr-bootleg-2020	76	10	the	the	DT
orr-bootleg-2020	76	11	factual	factual	JJ
orr-bootleg-2020	76	12	knowledge	knowledge	NN
orr-bootleg-2020	76	13	associated	associate	VBN
orr-bootleg-2020	76	14	with	with	IN
orr-bootleg-2020	76	15	a	a	DT
orr-bootleg-2020	76	16	specific	specific	JJ
orr-bootleg-2020	76	17	entity	entity	NN
orr-bootleg-2020	76	18	.	.	.
orr-bootleg-2020	77	1	Disambiguating	disambiguate	VBG
orr-bootleg-2020	77	2	“	"	``
orr-bootleg-2020	77	3	Lincoln	Lincoln	NNP
orr-bootleg-2020	77	4	”	"	''
orr-bootleg-2020	77	5	in	in	IN
orr-bootleg-2020	77	6	the	the	DT
orr-bootleg-2020	77	7	text	text	NN
orr-bootleg-2020	77	8	“	"	``
orr-bootleg-2020	77	9	Where	where	WRB
orr-bootleg-2020	77	10	is	be	VBZ
orr-bootleg-2020	77	11	Lincoln	Lincoln	NNP
orr-bootleg-2020	77	12	,	,	,
orr-bootleg-2020	77	13	Nebraska	Nebraska	NNP
orr-bootleg-2020	77	14	?	?	.
orr-bootleg-2020	77	15	”	"	''
orr-bootleg-2020	77	16	requires	require	VBZ
orr-bootleg-2020	77	17	memorizing	memorize	VBG
orr-bootleg-2020	77	18	that	that	IN
orr-bootleg-2020	77	19	“	"	``
orr-bootleg-2020	77	20	Lincoln	Lincoln	NNP
orr-bootleg-2020	77	21	,	,	,
orr-bootleg-2020	77	22	Nebraska	Nebraska	NNP
orr-bootleg-2020	77	23	”	"	''
orr-bootleg-2020	77	24	,	,	,
orr-bootleg-2020	77	25	not	not	RB
orr-bootleg-2020	77	26	“	"	``
orr-bootleg-2020	77	27	Abraham	Abraham	NNP
orr-bootleg-2020	77	28	Lincoln	Lincoln	NNP
orr-bootleg-2020	77	29	”	"	''
orr-bootleg-2020	77	30	frequently	frequently	RB
orr-bootleg-2020	77	31	occurs	occur	VBZ
orr-bootleg-2020	77	32	with	with	IN
orr-bootleg-2020	77	33	the	the	DT
orr-bootleg-2020	77	34	text	text	NN
orr-bootleg-2020	77	35	“	"	``
orr-bootleg-2020	77	36	Nebraska	Nebraska	NNP
orr-bootleg-2020	77	37	”	"	''
orr-bootleg-2020	77	38	(	(	-LRB-
orr-bootleg-2020	77	39	Figure	figure	NN
orr-bootleg-2020	77	40	1	1	CD
orr-bootleg-2020	77	41	(	(	-LRB-
orr-bootleg-2020	77	42	left	left	RB
orr-bootleg-2020	77	43	)	)	-RRB-
orr-bootleg-2020	77	44	)	)	-RRB-
orr-bootleg-2020	77	45	.	.	.
orr-bootleg-2020	78	1	This	this	DT
orr-bootleg-2020	78	2	pattern	pattern	NN
orr-bootleg-2020	78	3	is	be	VBZ
orr-bootleg-2020	78	4	easily	easily	RB
orr-bootleg-2020	78	5	learned	learn	VBN
orr-bootleg-2020	78	6	by	by	IN
orr-bootleg-2020	78	7	now	now	RB
orr-bootleg-2020	78	8	-	-	HYPH
orr-bootleg-2020	78	9	standard	standard	JJ
orr-bootleg-2020	78	10	Transformer	transformer	NN
orr-bootleg-2020	78	11	-	-	HYPH
orr-bootleg-2020	78	12	based	base	VBN
orr-bootleg-2020	78	13	language	language	NN
orr-bootleg-2020	78	14	models	model	NNS
orr-bootleg-2020	78	15	.	.	.
orr-bootleg-2020	79	1	As	as	IN
orr-bootleg-2020	79	2	this	this	DT
orr-bootleg-2020	79	3	pattern	pattern	NN
orr-bootleg-2020	79	4	is	be	VBZ
orr-bootleg-2020	79	5	at	at	IN
orr-bootleg-2020	79	6	the	the	DT
orr-bootleg-2020	79	7	entity	entity	NN
orr-bootleg-2020	79	8	-	-	HYPH
orr-bootleg-2020	79	9	level	level	NN
orr-bootleg-2020	79	10	,	,	,
orr-bootleg-2020	79	11	it	-PRON-	PRP
orr-bootleg-2020	79	12	is	be	VBZ
orr-bootleg-2020	79	13	the	the	DT
orr-bootleg-2020	79	14	least	least	JJS
orr-bootleg-2020	79	15	general	general	JJ
orr-bootleg-2020	79	16	pattern	pattern	NN
orr-bootleg-2020	79	17	.	.	.
orr-bootleg-2020	80	1	•	•	NNP
orr-bootleg-2020	80	2	Type	Type	NNP
orr-bootleg-2020	80	3	Consistency	consistency	NN
orr-bootleg-2020	80	4	:	:	:
orr-bootleg-2020	80	5	Type	type	NN
orr-bootleg-2020	80	6	consistency	consistency	NN
orr-bootleg-2020	80	7	is	be	VBZ
orr-bootleg-2020	80	8	the	the	DT
orr-bootleg-2020	80	9	pattern	pattern	NN
orr-bootleg-2020	80	10	that	that	IN
orr-bootleg-2020	80	11	certain	certain	JJ
orr-bootleg-2020	80	12	textual	textual	JJ
orr-bootleg-2020	80	13	signals	signal	NNS
orr-bootleg-2020	80	14	in	in	IN
orr-bootleg-2020	80	15	text	text	NN
orr-bootleg-2020	80	16	indicate	indicate	VBP
orr-bootleg-2020	80	17	that	that	IN
orr-bootleg-2020	80	18	the	the	DT
orr-bootleg-2020	80	19	types	type	NNS
orr-bootleg-2020	80	20	of	of	IN
orr-bootleg-2020	80	21	entities	entity	NNS
orr-bootleg-2020	80	22	in	in	IN
orr-bootleg-2020	80	23	a	a	DT
orr-bootleg-2020	80	24	collection	collection	NN
orr-bootleg-2020	80	25	are	be	VBP
orr-bootleg-2020	80	26	likely	likely	RB
orr-bootleg-2020	80	27	similar	similar	JJ
orr-bootleg-2020	80	28	.	.	.
orr-bootleg-2020	81	1	For	for	IN
orr-bootleg-2020	81	2	example	example	NN
orr-bootleg-2020	81	3	,	,	,
orr-bootleg-2020	81	4	when	when	WRB
orr-bootleg-2020	81	5	disambiguating	disambiguate	VBG
orr-bootleg-2020	81	6	“	"	``
orr-bootleg-2020	81	7	Lincoln	Lincoln	NNP
orr-bootleg-2020	81	8	”	"	''
orr-bootleg-2020	81	9	in	in	IN
orr-bootleg-2020	81	10	the	the	DT
orr-bootleg-2020	81	11	text	text	NN
orr-bootleg-2020	81	12	“	"	``
orr-bootleg-2020	81	13	Is	be	VBZ
orr-bootleg-2020	81	14	a	a	DT
orr-bootleg-2020	81	15	Lincoln	Lincoln	NNP
orr-bootleg-2020	81	16	or	or	CC
orr-bootleg-2020	81	17	Ford	Ford	NNP
orr-bootleg-2020	81	18	more	more	RBR
orr-bootleg-2020	81	19	expensive	expensive	JJ
orr-bootleg-2020	81	20	?	?	.
orr-bootleg-2020	81	21	”	"	''
orr-bootleg-2020	81	22	,	,	,
orr-bootleg-2020	81	23	the	the	DT
orr-bootleg-2020	81	24	keyword	keyword	NNP
orr-bootleg-2020	81	25	“	"	``
orr-bootleg-2020	81	26	or	or	CC
orr-bootleg-2020	81	27	”	"	``
orr-bootleg-2020	81	28	indicates	indicate	VBZ
orr-bootleg-2020	81	29	that	that	IN
orr-bootleg-2020	81	30	the	the	DT
orr-bootleg-2020	81	31	entities	entity	NNS
orr-bootleg-2020	81	32	in	in	IN
orr-bootleg-2020	81	33	the	the	DT
orr-bootleg-2020	81	34	pair	pair	NN
orr-bootleg-2020	81	35	(	(	-LRB-
orr-bootleg-2020	81	36	or	or	CC
orr-bootleg-2020	81	37	sequence	sequence	NN
orr-bootleg-2020	81	38	)	)	-RRB-
orr-bootleg-2020	81	39	are	be	VBP
orr-bootleg-2020	81	40	likely	likely	JJ
orr-bootleg-2020	81	41	of	of	IN
orr-bootleg-2020	81	42	the	the	DT
orr-bootleg-2020	81	43	same	same	JJ
orr-bootleg-2020	81	44	Wikidata	Wikidata	NNP
orr-bootleg-2020	81	45	type	type	NN
orr-bootleg-2020	81	46	,	,	,
orr-bootleg-2020	81	47	“	"	``
orr-bootleg-2020	81	48	car	car	NN
orr-bootleg-2020	81	49	company	company	NN
orr-bootleg-2020	81	50	”	"	''
orr-bootleg-2020	81	51	.	.	.
orr-bootleg-2020	82	1	Type	type	NN
orr-bootleg-2020	82	2	consistency	consistency	NN
orr-bootleg-2020	82	3	is	be	VBZ
orr-bootleg-2020	82	4	a	a	DT
orr-bootleg-2020	82	5	more	more	RBR
orr-bootleg-2020	82	6	general	general	JJ
orr-bootleg-2020	82	7	pattern	pattern	NN
orr-bootleg-2020	82	8	than	than	IN
orr-bootleg-2020	82	9	entity	entity	NN
orr-bootleg-2020	82	10	memorization	memorization	NN
orr-bootleg-2020	82	11	,	,	,
orr-bootleg-2020	82	12	covering	cover	VBG
orr-bootleg-2020	82	13	12	12	CD
orr-bootleg-2020	82	14	%	%	NN
orr-bootleg-2020	82	15	of	of	IN
orr-bootleg-2020	82	16	the	the	DT
orr-bootleg-2020	82	17	tail	tail	NN
orr-bootleg-2020	82	18	examples	example	NNS
orr-bootleg-2020	82	19	in	in	IN
orr-bootleg-2020	82	20	a	a	DT
orr-bootleg-2020	82	21	sample	sample	NN
orr-bootleg-2020	82	22	of	of	IN
orr-bootleg-2020	82	23	Wikipedia.5	wikipedia.5	WP
orr-bootleg-2020	82	24	•	•	NNP
orr-bootleg-2020	82	25	KG	kg	NN
orr-bootleg-2020	82	26	Relations	relation	NNS
orr-bootleg-2020	82	27	:	:	:
orr-bootleg-2020	82	28	We	-PRON-	PRP
orr-bootleg-2020	82	29	define	define	VBP
orr-bootleg-2020	82	30	the	the	DT
orr-bootleg-2020	82	31	knowledge	knowledge	NN
orr-bootleg-2020	82	32	graph	graph	NN
orr-bootleg-2020	82	33	(	(	-LRB-
orr-bootleg-2020	82	34	KG	KG	NNP
orr-bootleg-2020	82	35	)	)	-RRB-
orr-bootleg-2020	82	36	relation	relation	NN
orr-bootleg-2020	82	37	pattern	pattern	NN
orr-bootleg-2020	82	38	as	as	IN
orr-bootleg-2020	82	39	when	when	WRB
orr-bootleg-2020	82	40	two	two	CD
orr-bootleg-2020	82	41	candidates	candidate	NNS
orr-bootleg-2020	82	42	have	have	VBP
orr-bootleg-2020	82	43	a	a	DT
orr-bootleg-2020	82	44	known	know	VBN
orr-bootleg-2020	82	45	KG	kg	NN
orr-bootleg-2020	82	46	relationship	relationship	NN
orr-bootleg-2020	82	47	and	and	CC
orr-bootleg-2020	82	48	textual	textual	JJ
orr-bootleg-2020	82	49	signals	signal	NNS
orr-bootleg-2020	82	50	indicate	indicate	VBP
orr-bootleg-2020	82	51	that	that	IN
orr-bootleg-2020	82	52	the	the	DT
orr-bootleg-2020	82	53	relation	relation	NN
orr-bootleg-2020	82	54	is	be	VBZ
orr-bootleg-2020	82	55	discussed	discuss	VBN
orr-bootleg-2020	82	56	in	in	IN
orr-bootleg-2020	82	57	the	the	DT
orr-bootleg-2020	82	58	sentence	sentence	NN
orr-bootleg-2020	82	59	.	.	.
orr-bootleg-2020	83	1	For	for	IN
orr-bootleg-2020	83	2	example	example	NN
orr-bootleg-2020	83	3	,	,	,
orr-bootleg-2020	83	4	when	when	WRB
orr-bootleg-2020	83	5	disambiguating	disambiguate	VBG
orr-bootleg-2020	83	6	“	"	``
orr-bootleg-2020	83	7	Lincoln	Lincoln	NNP
orr-bootleg-2020	83	8	”	"	''
orr-bootleg-2020	83	9	in	in	IN
orr-bootleg-2020	83	10	the	the	DT
orr-bootleg-2020	83	11	sentence	sentence	NN
orr-bootleg-2020	83	12	“	"	``
orr-bootleg-2020	83	13	Where	where	WRB
orr-bootleg-2020	83	14	is	be	VBZ
orr-bootleg-2020	83	15	Lincoln	Lincoln	NNP
orr-bootleg-2020	83	16	in	in	IN
orr-bootleg-2020	83	17	Logan	Logan	NNP
orr-bootleg-2020	83	18	County	County	NNP
orr-bootleg-2020	83	19	?	?	.
orr-bootleg-2020	83	20	”	"	''
orr-bootleg-2020	83	21	,	,	,
orr-bootleg-2020	83	22	“	"	``
orr-bootleg-2020	83	23	Lincoln	Lincoln	NNP
orr-bootleg-2020	83	24	,	,	,
orr-bootleg-2020	83	25	IL	IL	NNP
orr-bootleg-2020	83	26	”	"	''
orr-bootleg-2020	83	27	has	have	VBZ
orr-bootleg-2020	83	28	the	the	DT
orr-bootleg-2020	83	29	KG	kg	NN
orr-bootleg-2020	83	30	relationship	relationship	NN
orr-bootleg-2020	83	31	“	"	``
orr-bootleg-2020	83	32	capital	capital	NN
orr-bootleg-2020	83	33	of	of	IN
orr-bootleg-2020	83	34	”	"	''
orr-bootleg-2020	83	35	with	with	IN
orr-bootleg-2020	83	36	Logan	Logan	NNP
orr-bootleg-2020	83	37	County	County	NNP
orr-bootleg-2020	83	38	while	while	IN
orr-bootleg-2020	83	39	Lincoln	Lincoln	NNP
orr-bootleg-2020	83	40	,	,	,
orr-bootleg-2020	83	41	NE	NE	NNP
orr-bootleg-2020	83	42	does	do	VBZ
orr-bootleg-2020	83	43	not	not	RB
orr-bootleg-2020	83	44	.	.	.
orr-bootleg-2020	84	1	The	the	DT
orr-bootleg-2020	84	2	keyword	keyword	NN
orr-bootleg-2020	84	3	“	"	``
orr-bootleg-2020	84	4	in	in	IN
orr-bootleg-2020	84	5	”	"	''
orr-bootleg-2020	84	6	is	be	VBZ
orr-bootleg-2020	84	7	associated	associate	VBN
orr-bootleg-2020	84	8	with	with	IN
orr-bootleg-2020	84	9	the	the	DT
orr-bootleg-2020	84	10	relation	relation	NN
orr-bootleg-2020	84	11	“	"	``
orr-bootleg-2020	84	12	capital	capital	NN
orr-bootleg-2020	84	13	of	of	IN
orr-bootleg-2020	84	14	”	"	''
orr-bootleg-2020	84	15	between	between	IN
orr-bootleg-2020	84	16	two	two	CD
orr-bootleg-2020	84	17	location	location	NN
orr-bootleg-2020	84	18	entities	entity	NNS
orr-bootleg-2020	84	19	,	,	,
orr-bootleg-2020	84	20	indicating	indicate	VBG
orr-bootleg-2020	84	21	that	that	IN
orr-bootleg-2020	84	22	“	"	``
orr-bootleg-2020	84	23	Lincoln	Lincoln	NNP
orr-bootleg-2020	84	24	,	,	,
orr-bootleg-2020	84	25	IL	IL	NNP
orr-bootleg-2020	84	26	”	"	''
orr-bootleg-2020	84	27	is	be	VBZ
orr-bootleg-2020	84	28	correct	correct	JJ
orr-bootleg-2020	84	29	,	,	,
orr-bootleg-2020	84	30	despite	despite	IN
orr-bootleg-2020	84	31	being	be	VBG
orr-bootleg-2020	84	32	the	the	DT
orr-bootleg-2020	84	33	less	less	RBR
orr-bootleg-2020	84	34	popular	popular	JJ
orr-bootleg-2020	84	35	candidate	candidate	NN
orr-bootleg-2020	84	36	entity	entity	NN
orr-bootleg-2020	84	37	associated	associate	VBN
orr-bootleg-2020	84	38	with	with	IN
orr-bootleg-2020	84	39	“	"	``
orr-bootleg-2020	84	40	Lincoln	Lincoln	NNP
orr-bootleg-2020	84	41	”	"	''
orr-bootleg-2020	84	42	.	.	.
orr-bootleg-2020	85	1	As	as	IN
orr-bootleg-2020	85	2	patterns	pattern	NNS
orr-bootleg-2020	85	3	over	over	IN
orr-bootleg-2020	85	4	pairs	pair	NNS
orr-bootleg-2020	85	5	of	of	IN
orr-bootleg-2020	85	6	entities	entity	NNS
orr-bootleg-2020	85	7	with	with	IN
orr-bootleg-2020	85	8	KG	KG	NNP
orr-bootleg-2020	85	9	relations	relation	NNS
orr-bootleg-2020	85	10	cover	cover	VBP
orr-bootleg-2020	85	11	23	23	CD
orr-bootleg-2020	85	12	%	%	NN
orr-bootleg-2020	85	13	of	of	IN
orr-bootleg-2020	85	14	the	the	DT
orr-bootleg-2020	85	15	tail	tail	NN
orr-bootleg-2020	85	16	examples	example	NNS
orr-bootleg-2020	85	17	,	,	,
orr-bootleg-2020	85	18	this	this	DT
orr-bootleg-2020	85	19	is	be	VBZ
orr-bootleg-2020	85	20	a	a	DT
orr-bootleg-2020	85	21	more	more	RBR
orr-bootleg-2020	85	22	general	general	JJ
orr-bootleg-2020	85	23	reasoning	reasoning	NN
orr-bootleg-2020	85	24	pattern	pattern	NN
orr-bootleg-2020	85	25	than	than	IN
orr-bootleg-2020	85	26	consistency	consistency	NN
orr-bootleg-2020	85	27	.	.	.
orr-bootleg-2020	86	1	•	•	VB
orr-bootleg-2020	86	2	Type	type	NN
orr-bootleg-2020	86	3	Affordance	affordance	NN
orr-bootleg-2020	86	4	:	:	:
orr-bootleg-2020	86	5	We	-PRON-	PRP
orr-bootleg-2020	86	6	define	define	VBP
orr-bootleg-2020	86	7	type	type	NN
orr-bootleg-2020	86	8	affordance	affordance	NN
orr-bootleg-2020	86	9	as	as	IN
orr-bootleg-2020	86	10	the	the	DT
orr-bootleg-2020	86	11	textual	textual	JJ
orr-bootleg-2020	86	12	signals	signal	NNS
orr-bootleg-2020	86	13	associated	associate	VBN
orr-bootleg-2020	86	14	with	with	IN
orr-bootleg-2020	86	15	a	a	DT
orr-bootleg-2020	86	16	specific	specific	JJ
orr-bootleg-2020	86	17	entity-	entity-	JJ
orr-bootleg-2020	86	18	type	type	NN
orr-bootleg-2020	86	19	in	in	IN
orr-bootleg-2020	86	20	natural	natural	JJ
orr-bootleg-2020	86	21	language	language	NN
orr-bootleg-2020	86	22	.	.	.
orr-bootleg-2020	87	1	For	for	IN
orr-bootleg-2020	87	2	example	example	NN
orr-bootleg-2020	87	3	,	,	,
orr-bootleg-2020	87	4	“	"	``
orr-bootleg-2020	87	5	Manhattan	Manhattan	NNP
orr-bootleg-2020	87	6	”	"	''
orr-bootleg-2020	87	7	is	be	VBZ
orr-bootleg-2020	87	8	likely	likely	RB
orr-bootleg-2020	87	9	resolved	resolve	VBN
orr-bootleg-2020	87	10	to	to	IN
orr-bootleg-2020	87	11	the	the	DT
orr-bootleg-2020	87	12	cocktail	cocktail	NN
orr-bootleg-2020	87	13	rather	rather	RB
orr-bootleg-2020	87	14	than	than	IN
orr-bootleg-2020	87	15	the	the	DT
orr-bootleg-2020	87	16	burrough	burrough	NN
orr-bootleg-2020	87	17	in	in	IN
orr-bootleg-2020	87	18	the	the	DT
orr-bootleg-2020	87	19	sentence	sentence	NN
orr-bootleg-2020	87	20	“	"	``
orr-bootleg-2020	87	21	He	-PRON-	PRP
orr-bootleg-2020	87	22	ordered	order	VBD
orr-bootleg-2020	87	23	a	a	DT
orr-bootleg-2020	87	24	Manhattan	Manhattan	NNP
orr-bootleg-2020	87	25	.	.	.
orr-bootleg-2020	87	26	”	"	''
orr-bootleg-2020	87	27	due	due	IN
orr-bootleg-2020	87	28	to	to	IN
orr-bootleg-2020	87	29	the	the	DT
orr-bootleg-2020	87	30	affordance	affordance	NN
orr-bootleg-2020	87	31	that	that	WDT
orr-bootleg-2020	87	32	drinks	drink	VBZ
orr-bootleg-2020	87	33	,	,	,
orr-bootleg-2020	87	34	not	not	RB
orr-bootleg-2020	87	35	locations	location	NNS
orr-bootleg-2020	87	36	,	,	,
orr-bootleg-2020	87	37	are	be	VBP
orr-bootleg-2020	87	38	“	"	``
orr-bootleg-2020	87	39	ordered	order	VBN
orr-bootleg-2020	87	40	”	"	''
orr-bootleg-2020	87	41	.	.	.
orr-bootleg-2020	88	1	As	as	IN
orr-bootleg-2020	88	2	affordance	affordance	NN
orr-bootleg-2020	88	3	signals	signal	NNS
orr-bootleg-2020	88	4	cover	cover	VBP
orr-bootleg-2020	88	5	76	76	CD
orr-bootleg-2020	88	6	%	%	NN
orr-bootleg-2020	88	7	of	of	IN
orr-bootleg-2020	88	8	the	the	DT
orr-bootleg-2020	88	9	tail	tail	NN
orr-bootleg-2020	88	10	examples	example	NNS
orr-bootleg-2020	88	11	,	,	,
orr-bootleg-2020	88	12	it	-PRON-	PRP
orr-bootleg-2020	88	13	is	be	VBZ
orr-bootleg-2020	88	14	the	the	DT
orr-bootleg-2020	88	15	most	most	RBS
orr-bootleg-2020	88	16	general	general	JJ
orr-bootleg-2020	88	17	reasoning	reasoning	NN
orr-bootleg-2020	88	18	pattern	pattern	NN
orr-bootleg-2020	88	19	.	.	.
orr-bootleg-2020	89	1	Required	Required	NNP
orr-bootleg-2020	89	2	Structural	Structural	NNP
orr-bootleg-2020	89	3	Resources	Resources	NNPS
orr-bootleg-2020	89	4	An	an	DT
orr-bootleg-2020	89	5	NED	NED	NNP
orr-bootleg-2020	89	6	system	system	NN
orr-bootleg-2020	89	7	requires	require	VBZ
orr-bootleg-2020	89	8	entity	entity	NN
orr-bootleg-2020	89	9	,	,	,
orr-bootleg-2020	89	10	relation	relation	NN
orr-bootleg-2020	89	11	,	,	,
orr-bootleg-2020	89	12	and	and	CC
orr-bootleg-2020	89	13	type	type	NN
orr-bootleg-2020	89	14	knowledge	knowledge	NN
orr-bootleg-2020	89	15	signals	signal	NNS
orr-bootleg-2020	89	16	to	to	TO
orr-bootleg-2020	89	17	learn	learn	VB
orr-bootleg-2020	89	18	these	these	DT
orr-bootleg-2020	89	19	reasoning	reasoning	NN
orr-bootleg-2020	89	20	patterns	pattern	NNS
orr-bootleg-2020	89	21	.	.	.
orr-bootleg-2020	90	1	Entity	entity	NN
orr-bootleg-2020	90	2	knowledge	knowledge	NN
orr-bootleg-2020	90	3	is	be	VBZ
orr-bootleg-2020	90	4	captured	capture	VBN
orr-bootleg-2020	90	5	in	in	IN
orr-bootleg-2020	90	6	unstructured	unstructured	JJ
orr-bootleg-2020	90	7	text	text	NN
orr-bootleg-2020	90	8	,	,	,
orr-bootleg-2020	90	9	while	while	IN
orr-bootleg-2020	90	10	relation	relation	NN
orr-bootleg-2020	90	11	signals	signal	NNS
orr-bootleg-2020	90	12	and	and	CC
orr-bootleg-2020	90	13	type	type	NN
orr-bootleg-2020	90	14	signals	signal	NNS
orr-bootleg-2020	90	15	are	be	VBP
orr-bootleg-2020	90	16	readily	readily	RB
orr-bootleg-2020	90	17	available	available	JJ
orr-bootleg-2020	90	18	in	in	IN
orr-bootleg-2020	90	19	structured	structured	JJ
orr-bootleg-2020	90	20	knowledge	knowledge	NN
orr-bootleg-2020	90	21	bases	basis	NNS
orr-bootleg-2020	90	22	such	such	JJ
orr-bootleg-2020	90	23	as	as	IN
orr-bootleg-2020	90	24	Wikidata	wikidata	NN
orr-bootleg-2020	90	25	:	:	:
orr-bootleg-2020	90	26	from	from	IN
orr-bootleg-2020	90	27	a	a	DT
orr-bootleg-2020	90	28	sample	sample	NN
orr-bootleg-2020	90	29	of	of	IN
orr-bootleg-2020	90	30	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	90	31	,	,	,
orr-bootleg-2020	90	32	27	27	CD
orr-bootleg-2020	90	33	%	%	NN
orr-bootleg-2020	90	34	of	of	IN
orr-bootleg-2020	90	35	all	all	DT
orr-bootleg-2020	90	36	mentions	mention	NNS
orr-bootleg-2020	90	37	and	and	CC
orr-bootleg-2020	90	38	23	23	CD
orr-bootleg-2020	90	39	%	%	NN
orr-bootleg-2020	90	40	of	of	IN
orr-bootleg-2020	90	41	tail	tail	NN
orr-bootleg-2020	90	42	mentions	mention	NNS
orr-bootleg-2020	90	43	participate	participate	VBP
orr-bootleg-2020	90	44	in	in	IN
orr-bootleg-2020	90	45	a	a	DT
orr-bootleg-2020	90	46	relation	relation	NN
orr-bootleg-2020	90	47	,	,	,
orr-bootleg-2020	90	48	and	and	CC
orr-bootleg-2020	90	49	97	97	CD
orr-bootleg-2020	90	50	%	%	NN
orr-bootleg-2020	90	51	of	of	IN
orr-bootleg-2020	90	52	all	all	DT
orr-bootleg-2020	90	53	mentions	mention	NNS
orr-bootleg-2020	90	54	and	and	CC
orr-bootleg-2020	90	55	92	92	CD
orr-bootleg-2020	90	56	%	%	NN
orr-bootleg-2020	90	57	of	of	IN
orr-bootleg-2020	90	58	tail	tail	NN
orr-bootleg-2020	90	59	mentions	mention	NNS
orr-bootleg-2020	90	60	are	be	VBP
orr-bootleg-2020	90	61	assigned	assign	VBN
orr-bootleg-2020	90	62	some	some	DT
orr-bootleg-2020	90	63	type	type	NN
orr-bootleg-2020	90	64	in	in	IN
orr-bootleg-2020	90	65	Wikidata	Wikidata	NNP
orr-bootleg-2020	90	66	.	.	.
orr-bootleg-2020	91	1	As	as	IN
orr-bootleg-2020	91	2	these	these	DT
orr-bootleg-2020	91	3	structural	structural	JJ
orr-bootleg-2020	91	4	resources	resource	NNS
orr-bootleg-2020	91	5	are	be	VBP
orr-bootleg-2020	91	6	readily	readily	RB
orr-bootleg-2020	91	7	available	available	JJ
orr-bootleg-2020	91	8	for	for	IN
orr-bootleg-2020	91	9	all	all	DT
orr-bootleg-2020	91	10	entities	entity	NNS
orr-bootleg-2020	91	11	,	,	,
orr-bootleg-2020	91	12	they	-PRON-	PRP
orr-bootleg-2020	91	13	are	be	VBP
orr-bootleg-2020	91	14	useful	useful	JJ
orr-bootleg-2020	91	15	for	for	IN
orr-bootleg-2020	91	16	generalizing	generalize	VBG
orr-bootleg-2020	91	17	to	to	IN
orr-bootleg-2020	91	18	the	the	DT
orr-bootleg-2020	91	19	tail	tail	NN
orr-bootleg-2020	91	20	.	.	.
orr-bootleg-2020	92	1	A	a	DT
orr-bootleg-2020	92	2	rare	rare	JJ
orr-bootleg-2020	92	3	entity	entity	NN
orr-bootleg-2020	92	4	with	with	IN
orr-bootleg-2020	92	5	a	a	DT
orr-bootleg-2020	92	6	particular	particular	JJ
orr-bootleg-2020	92	7	type	type	NN
orr-bootleg-2020	92	8	or	or	CC
orr-bootleg-2020	92	9	relation	relation	NN
orr-bootleg-2020	92	10	can	can	MD
orr-bootleg-2020	92	11	leverage	leverage	VB
orr-bootleg-2020	92	12	textual	textual	JJ
orr-bootleg-2020	92	13	patterns	pattern	NNS
orr-bootleg-2020	92	14	learned	learn	VBN
orr-bootleg-2020	92	15	from	from	IN
orr-bootleg-2020	92	16	every	every	DT
orr-bootleg-2020	92	17	other	other	JJ
orr-bootleg-2020	92	18	entity	entity	NN
orr-bootleg-2020	92	19	with	with	IN
orr-bootleg-2020	92	20	that	that	DT
orr-bootleg-2020	92	21	type	type	NN
orr-bootleg-2020	92	22	or	or	CC
orr-bootleg-2020	92	23	relation	relation	NN
orr-bootleg-2020	92	24	.	.	.
orr-bootleg-2020	93	1	Given	give	VBN
orr-bootleg-2020	93	2	the	the	DT
orr-bootleg-2020	93	3	input	input	NN
orr-bootleg-2020	93	4	signals	signal	NNS
orr-bootleg-2020	93	5	and	and	CC
orr-bootleg-2020	93	6	reasoning	reasoning	NN
orr-bootleg-2020	93	7	patterns	pattern	NNS
orr-bootleg-2020	93	8	,	,	,
orr-bootleg-2020	93	9	the	the	DT
orr-bootleg-2020	93	10	next	next	JJ
orr-bootleg-2020	93	11	key	key	JJ
orr-bootleg-2020	93	12	challenge	challenge	NN
orr-bootleg-2020	93	13	is	be	VBZ
orr-bootleg-2020	93	14	ensuring	ensure	VBG
orr-bootleg-2020	93	15	that	that	IN
orr-bootleg-2020	93	16	the	the	DT
orr-bootleg-2020	93	17	model	model	NN
orr-bootleg-2020	93	18	combines	combine	VBZ
orr-bootleg-2020	93	19	the	the	DT
orr-bootleg-2020	93	20	discriminative	discriminative	JJ
orr-bootleg-2020	93	21	entity	entity	NN
orr-bootleg-2020	93	22	and	and	CC
orr-bootleg-2020	93	23	more	more	JJR
orr-bootleg-2020	93	24	general	general	JJ
orr-bootleg-2020	93	25	relation	relation	NN
orr-bootleg-2020	93	26	and	and	CC
orr-bootleg-2020	93	27	type	type	NN
orr-bootleg-2020	93	28	signals	signal	NNS
orr-bootleg-2020	93	29	that	that	WDT
orr-bootleg-2020	93	30	are	be	VBP
orr-bootleg-2020	93	31	useful	useful	JJ
orr-bootleg-2020	93	32	for	for	IN
orr-bootleg-2020	93	33	disambiguation	disambiguation	NN
orr-bootleg-2020	93	34	.	.	.
orr-bootleg-2020	94	1	3	3	CD
orr-bootleg-2020	94	2	Bootleg	Bootleg	NNP
orr-bootleg-2020	94	3	Architecture	Architecture	NNP
orr-bootleg-2020	94	4	for	for	IN
orr-bootleg-2020	94	5	Tail	Tail	NNP
orr-bootleg-2020	94	6	Disambiguation	Disambiguation	NNP
orr-bootleg-2020	94	7	We	-PRON-	PRP
orr-bootleg-2020	94	8	now	now	RB
orr-bootleg-2020	94	9	describe	describe	VBP
orr-bootleg-2020	94	10	our	-PRON-	PRP$
orr-bootleg-2020	94	11	approach	approach	NN
orr-bootleg-2020	94	12	to	to	TO
orr-bootleg-2020	94	13	leverage	leverage	VB
orr-bootleg-2020	94	14	the	the	DT
orr-bootleg-2020	94	15	reasoning	reasoning	NN
orr-bootleg-2020	94	16	patterns	pattern	NNS
orr-bootleg-2020	94	17	based	base	VBN
orr-bootleg-2020	94	18	on	on	IN
orr-bootleg-2020	94	19	entity	entity	NN
orr-bootleg-2020	94	20	,	,	,
orr-bootleg-2020	94	21	relation	relation	NN
orr-bootleg-2020	94	22	,	,	,
orr-bootleg-2020	94	23	and	and	CC
orr-bootleg-2020	94	24	type	type	NN
orr-bootleg-2020	94	25	signals	signal	NNS
orr-bootleg-2020	94	26	.	.	.
orr-bootleg-2020	95	1	We	-PRON-	PRP
orr-bootleg-2020	95	2	then	then	RB
orr-bootleg-2020	95	3	present	present	VBP
orr-bootleg-2020	95	4	our	-PRON-	PRP$
orr-bootleg-2020	95	5	new	new	JJ
orr-bootleg-2020	95	6	regularization	regularization	NN
orr-bootleg-2020	95	7	scheme	scheme	NN
orr-bootleg-2020	95	8	to	to	TO
orr-bootleg-2020	95	9	inject	inject	VB
orr-bootleg-2020	95	10	inductive	inductive	JJ
orr-bootleg-2020	95	11	bias	bias	NN
orr-bootleg-2020	95	12	of	of	IN
orr-bootleg-2020	95	13	when	when	WRB
orr-bootleg-2020	95	14	to	to	TO
orr-bootleg-2020	95	15	use	use	VB
orr-bootleg-2020	95	16	general	general	JJ
orr-bootleg-2020	95	17	versus	versus	IN
orr-bootleg-2020	95	18	discriminative	discriminative	JJ
orr-bootleg-2020	95	19	reasoning	reasoning	NN
orr-bootleg-2020	95	20	patterns	pattern	NNS
orr-bootleg-2020	95	21	and	and	CC
orr-bootleg-2020	95	22	our	-PRON-	PRP$
orr-bootleg-2020	95	23	weak	weak	JJ
orr-bootleg-2020	95	24	labeling	labeling	NN
orr-bootleg-2020	95	25	technique	technique	NN
orr-bootleg-2020	95	26	to	to	TO
orr-bootleg-2020	95	27	extract	extract	VB
orr-bootleg-2020	95	28	more	more	JJR
orr-bootleg-2020	95	29	signal	signal	NN
orr-bootleg-2020	95	30	from	from	IN
orr-bootleg-2020	95	31	the	the	DT
orr-bootleg-2020	95	32	self	self	NN
orr-bootleg-2020	95	33	-	-	HYPH
orr-bootleg-2020	95	34	supervision	supervision	NN
orr-bootleg-2020	95	35	training	training	NN
orr-bootleg-2020	95	36	data	datum	NNS
orr-bootleg-2020	95	37	.	.	.
orr-bootleg-2020	96	1	5Coverage	5coverage	JJ
orr-bootleg-2020	96	2	numbers	number	NNS
orr-bootleg-2020	96	3	are	be	VBP
orr-bootleg-2020	96	4	calculated	calculate	VBN
orr-bootleg-2020	96	5	from	from	IN
orr-bootleg-2020	96	6	representative	representative	JJ
orr-bootleg-2020	96	7	slices	slice	NNS
orr-bootleg-2020	96	8	of	of	IN
orr-bootleg-2020	96	9	Wikidata	Wikidata	NNP
orr-bootleg-2020	96	10	that	that	WDT
orr-bootleg-2020	96	11	require	require	VBP
orr-bootleg-2020	96	12	each	each	DT
orr-bootleg-2020	96	13	reasoning	reasoning	NN
orr-bootleg-2020	96	14	pattern	pattern	NN
orr-bootleg-2020	96	15	.	.	.
orr-bootleg-2020	97	1	Additional	additional	JJ
orr-bootleg-2020	97	2	details	detail	NNS
orr-bootleg-2020	97	3	in	in	IN
orr-bootleg-2020	97	4	Section	section	NN
orr-bootleg-2020	97	5	5	5	CD
orr-bootleg-2020	97	6	.	.	.
orr-bootleg-2020	98	1	4	4	LS
orr-bootleg-2020	98	2	“	"	``
orr-bootleg-2020	98	3	Where	where	WRB
orr-bootleg-2020	98	4	is	be	VBZ
orr-bootleg-2020	98	5	Lincoln	Lincoln	NNP
orr-bootleg-2020	98	6	in	in	IN
orr-bootleg-2020	98	7	Logan	Logan	NNP
orr-bootleg-2020	98	8	County	County	NNP
orr-bootleg-2020	98	9	?	?	.
orr-bootleg-2020	98	10	”	"	''
orr-bootleg-2020	98	11	Lincoln	Lincoln	NNP
orr-bootleg-2020	98	12	,	,	,
orr-bootleg-2020	98	13	ILLincoln	ILLincoln	NNP
orr-bootleg-2020	98	14	,	,	,
orr-bootleg-2020	98	15	NEAbraham	NEAbraham	NNP
orr-bootleg-2020	98	16	Lincoln	Lincoln	NNP
orr-bootleg-2020	98	17	AddAttn	AddAttn	NNP
orr-bootleg-2020	98	18	type	type	NN
orr-bootleg-2020	98	19	embsentity	embsentity	NN
orr-bootleg-2020	98	20	emb	emb	VBD
orr-bootleg-2020	98	21	AddAttn	AddAttn	NNP
orr-bootleg-2020	98	22	relation	relation	NN
orr-bootleg-2020	98	23	embs	emb	VBZ
orr-bootleg-2020	98	24	Cat	Cat	NNP
orr-bootleg-2020	98	25	+	+	CC
orr-bootleg-2020	98	26	Proj	Proj	NNP
orr-bootleg-2020	98	27	Ent2Ent	Ent2Ent	NNP
orr-bootleg-2020	98	28	Phrase2Ent	phrase2ent	NN
orr-bootleg-2020	98	29	Softmax	Softmax	NNP
orr-bootleg-2020	98	30	+	+	SYM
orr-bootleg-2020	98	31	Lincoln	Lincoln	NNP
orr-bootleg-2020	98	32	,	,	,
orr-bootleg-2020	98	33	IL	IL	NNP
orr-bootleg-2020	98	34	Logan	Logan	NNP
orr-bootleg-2020	98	35	County	County	NNP
orr-bootleg-2020	98	36	,	,	,
orr-bootleg-2020	98	37	IL	IL	NNP
orr-bootleg-2020	98	38	KG2Ent	kg2ent	NN
orr-bootleg-2020	98	39	single	single	JJ
orr-bootleg-2020	98	40	layer	layer	NN
orr-bootleg-2020	98	41	Logan	Logan	NNP
orr-bootleg-2020	98	42	Country	Country	NNP
orr-bootleg-2020	98	43	,	,	,
orr-bootleg-2020	98	44	OHLogan	OHLogan	NNP
orr-bootleg-2020	98	45	County	County	NNP
orr-bootleg-2020	98	46	,	,	,
orr-bootleg-2020	98	47	OKLogan	OKLogan	NNP
orr-bootleg-2020	98	48	Country	Country	NNP
orr-bootleg-2020	98	49	,	,	,
orr-bootleg-2020	98	50	IL	IL	NNP
orr-bootleg-2020	98	51	ue	ue	NNP
orr-bootleg-2020	98	52	re	re	NN
orr-bootleg-2020	98	53	te	te	NNP
orr-bootleg-2020	98	54	E	E	NNP
orr-bootleg-2020	98	55	W	w	NN
orr-bootleg-2020	98	56	E	e	NN
orr-bootleg-2020	98	57	W	w	NN
orr-bootleg-2020	98	58	BERT	bert	NN
orr-bootleg-2020	98	59	Figure	figure	NN
orr-bootleg-2020	98	60	2	2	CD
orr-bootleg-2020	98	61	:	:	:
orr-bootleg-2020	98	62	Bootleg	Bootleg	NNP
orr-bootleg-2020	98	63	’s	’s	POS
orr-bootleg-2020	98	64	neural	neural	JJ
orr-bootleg-2020	98	65	model	model	NN
orr-bootleg-2020	98	66	.	.	.
orr-bootleg-2020	99	1	The	the	DT
orr-bootleg-2020	99	2	entity	entity	NN
orr-bootleg-2020	99	3	,	,	,
orr-bootleg-2020	99	4	type	type	NN
orr-bootleg-2020	99	5	,	,	,
orr-bootleg-2020	99	6	and	and	CC
orr-bootleg-2020	99	7	relation	relation	NN
orr-bootleg-2020	99	8	embeddings	embedding	NNS
orr-bootleg-2020	99	9	are	be	VBP
orr-bootleg-2020	99	10	generated	generate	VBN
orr-bootleg-2020	99	11	for	for	IN
orr-bootleg-2020	99	12	each	each	DT
orr-bootleg-2020	99	13	candidate	candidate	NN
orr-bootleg-2020	99	14	and	and	CC
orr-bootleg-2020	99	15	concatenated	concatenate	VBD
orr-bootleg-2020	99	16	to	to	TO
orr-bootleg-2020	99	17	form	form	VB
orr-bootleg-2020	99	18	our	-PRON-	PRP$
orr-bootleg-2020	99	19	entity	entity	NN
orr-bootleg-2020	99	20	representation	representation	NN
orr-bootleg-2020	99	21	matrix	matrix	NN
orr-bootleg-2020	99	22	E.	E.	NNP
orr-bootleg-2020	99	23	This	This	NNP
orr-bootleg-2020	99	24	,	,	,
orr-bootleg-2020	99	25	together	together	RB
orr-bootleg-2020	99	26	with	with	IN
orr-bootleg-2020	99	27	our	-PRON-	PRP$
orr-bootleg-2020	99	28	word	word	NN
orr-bootleg-2020	99	29	embedding	embed	VBG
orr-bootleg-2020	99	30	matrix	matrix	NN
orr-bootleg-2020	99	31	W	W	NNP
orr-bootleg-2020	99	32	,	,	,
orr-bootleg-2020	99	33	are	be	VBP
orr-bootleg-2020	99	34	inputs	input	NNS
orr-bootleg-2020	99	35	to	to	IN
orr-bootleg-2020	99	36	Bootleg	Bootleg	NNP
orr-bootleg-2020	99	37	’s	’s	POS
orr-bootleg-2020	99	38	Ent2Ent	Ent2Ent	NNP
orr-bootleg-2020	99	39	,	,	,
orr-bootleg-2020	99	40	Phrase2Ent	phrase2ent	NN
orr-bootleg-2020	99	41	,	,	,
orr-bootleg-2020	99	42	and	and	CC
orr-bootleg-2020	99	43	KG2Ent	kg2ent	NN
orr-bootleg-2020	99	44	modules	module	NNS
orr-bootleg-2020	99	45	which	which	WDT
orr-bootleg-2020	99	46	aim	aim	VBP
orr-bootleg-2020	99	47	to	to	TO
orr-bootleg-2020	99	48	encode	encode	VB
orr-bootleg-2020	99	49	the	the	DT
orr-bootleg-2020	99	50	four	four	CD
orr-bootleg-2020	99	51	reasoning	reasoning	NN
orr-bootleg-2020	99	52	patterns	pattern	NNS
orr-bootleg-2020	99	53	.	.	.
orr-bootleg-2020	100	1	The	the	DT
orr-bootleg-2020	100	2	most	most	RBS
orr-bootleg-2020	100	3	likely	likely	JJ
orr-bootleg-2020	100	4	candidate	candidate	NN
orr-bootleg-2020	100	5	for	for	IN
orr-bootleg-2020	100	6	each	each	DT
orr-bootleg-2020	100	7	mention	mention	NN
orr-bootleg-2020	100	8	is	be	VBZ
orr-bootleg-2020	100	9	returned	return	VBN
orr-bootleg-2020	100	10	.	.	.
orr-bootleg-2020	101	1	3.1	3.1	CD
orr-bootleg-2020	101	2	Encoding	encode	VBG
orr-bootleg-2020	101	3	the	the	DT
orr-bootleg-2020	101	4	Signals	signal	NNS
orr-bootleg-2020	101	5	We	-PRON-	PRP
orr-bootleg-2020	101	6	first	first	RB
orr-bootleg-2020	101	7	encode	encode	VBP
orr-bootleg-2020	101	8	the	the	DT
orr-bootleg-2020	101	9	structural	structural	JJ
orr-bootleg-2020	101	10	signals	signal	NNS
orr-bootleg-2020	101	11	—	—	:
orr-bootleg-2020	101	12	entities	entity	NNS
orr-bootleg-2020	101	13	,	,	,
orr-bootleg-2020	101	14	KG	kg	NN
orr-bootleg-2020	101	15	relations	relation	NNS
orr-bootleg-2020	101	16	and	and	CC
orr-bootleg-2020	101	17	types	type	NNS
orr-bootleg-2020	101	18	—	—	:
orr-bootleg-2020	101	19	by	by	IN
orr-bootleg-2020	101	20	mapping	map	VBG
orr-bootleg-2020	101	21	each	each	DT
orr-bootleg-2020	101	22	to	to	IN
orr-bootleg-2020	101	23	a	a	DT
orr-bootleg-2020	101	24	set	set	NN
orr-bootleg-2020	101	25	of	of	IN
orr-bootleg-2020	101	26	embeddings	embedding	NNS
orr-bootleg-2020	101	27	.	.	.
orr-bootleg-2020	102	1	•	•	VB
orr-bootleg-2020	102	2	Entity	entity	NN
orr-bootleg-2020	102	3	Embedding	embedding	NN
orr-bootleg-2020	102	4	:	:	:
orr-bootleg-2020	102	5	Each	each	DT
orr-bootleg-2020	102	6	entity	entity	NN
orr-bootleg-2020	102	7	e	e	NN
orr-bootleg-2020	102	8	is	be	VBZ
orr-bootleg-2020	102	9	represented	represent	VBN
orr-bootleg-2020	102	10	by	by	IN
orr-bootleg-2020	102	11	a	a	DT
orr-bootleg-2020	102	12	unique	unique	JJ
orr-bootleg-2020	102	13	embedding	embedding	NN
orr-bootleg-2020	102	14	ue	ue	JJ
orr-bootleg-2020	102	15	.	.	.
orr-bootleg-2020	103	1	•	•	NN
orr-bootleg-2020	103	2	Type	type	NN
orr-bootleg-2020	103	3	Embedding	embedding	NN
orr-bootleg-2020	103	4	:	:	:
orr-bootleg-2020	103	5	Let	let	VB
orr-bootleg-2020	103	6	T	t	PRP
orr-bootleg-2020	103	7	be	be	VB
orr-bootleg-2020	103	8	the	the	DT
orr-bootleg-2020	103	9	set	set	NN
orr-bootleg-2020	103	10	of	of	IN
orr-bootleg-2020	103	11	possible	possible	JJ
orr-bootleg-2020	103	12	entity	entity	NN
orr-bootleg-2020	103	13	types	type	NNS
orr-bootleg-2020	103	14	.	.	.
orr-bootleg-2020	104	1	Given	give	VBN
orr-bootleg-2020	104	2	a	a	DT
orr-bootleg-2020	104	3	known	know	VBN
orr-bootleg-2020	104	4	mapping	mapping	NN
orr-bootleg-2020	104	5	from	from	IN
orr-bootleg-2020	104	6	an	an	DT
orr-bootleg-2020	104	7	entity	entity	NN
orr-bootleg-2020	104	8	e	e	NN
orr-bootleg-2020	104	9	to	to	IN
orr-bootleg-2020	104	10	its	-PRON-	PRP$
orr-bootleg-2020	104	11	set	set	NN
orr-bootleg-2020	104	12	{	{	-LRB-
orr-bootleg-2020	104	13	te,1	te,1	NNP
orr-bootleg-2020	104	14	,	,	,
orr-bootleg-2020	104	15	.	.	.
orr-bootleg-2020	105	1	.	.	.
orr-bootleg-2020	106	1	.	.	.
orr-bootleg-2020	107	1	,	,	,
orr-bootleg-2020	107	2	te	te	NNP
orr-bootleg-2020	107	3	,	,	,
orr-bootleg-2020	107	4	T	T	NNP
orr-bootleg-2020	107	5	|te	|te	NNP
orr-bootleg-2020	107	6	,	,	,
orr-bootleg-2020	107	7	i	i	PRP
orr-bootleg-2020	107	8	∈	∈	VBP
orr-bootleg-2020	107	9	T	T	NNP
orr-bootleg-2020	107	10	}	}	-RRB-
orr-bootleg-2020	107	11	of	of	IN
orr-bootleg-2020	107	12	T	t	NN
orr-bootleg-2020	107	13	possible	possible	JJ
orr-bootleg-2020	107	14	types	type	NNS
orr-bootleg-2020	107	15	,	,	,
orr-bootleg-2020	107	16	Bootleg	Bootleg	NNP
orr-bootleg-2020	107	17	assigns	assign	VBZ
orr-bootleg-2020	107	18	an	an	DT
orr-bootleg-2020	107	19	embedding	embedding	NN
orr-bootleg-2020	107	20	te	te	NNP
orr-bootleg-2020	107	21	,	,	,
orr-bootleg-2020	107	22	i	i	PRP
orr-bootleg-2020	107	23	to	to	IN
orr-bootleg-2020	107	24	each	each	DT
orr-bootleg-2020	107	25	type	type	NN
orr-bootleg-2020	107	26	.	.	.
orr-bootleg-2020	108	1	Because	because	IN
orr-bootleg-2020	108	2	an	an	DT
orr-bootleg-2020	108	3	entity	entity	NN
orr-bootleg-2020	108	4	can	can	MD
orr-bootleg-2020	108	5	have	have	VB
orr-bootleg-2020	108	6	multiple	multiple	JJ
orr-bootleg-2020	108	7	types	type	NNS
orr-bootleg-2020	108	8	,	,	,
orr-bootleg-2020	108	9	we	-PRON-	PRP
orr-bootleg-2020	108	10	use	use	VBP
orr-bootleg-2020	108	11	an	an	DT
orr-bootleg-2020	108	12	additive	additive	JJ
orr-bootleg-2020	108	13	attention	attention	NN
orr-bootleg-2020	108	14	[	[	-LRB-
orr-bootleg-2020	108	15	3	3	CD
orr-bootleg-2020	108	16	]	]	-RRB-
orr-bootleg-2020	108	17	,	,	,
orr-bootleg-2020	108	18	AddAttn	AddAttn	NNP
orr-bootleg-2020	108	19	,	,	,
orr-bootleg-2020	108	20	to	to	TO
orr-bootleg-2020	108	21	create	create	VB
orr-bootleg-2020	108	22	a	a	DT
orr-bootleg-2020	108	23	single	single	JJ
orr-bootleg-2020	108	24	type	type	NN
orr-bootleg-2020	108	25	embedding	embedding	NN
orr-bootleg-2020	108	26	te	te	NNP
orr-bootleg-2020	108	27	=	=	-RRB-
orr-bootleg-2020	108	28	AddAttn([te,1	AddAttn([te,1	NNP
orr-bootleg-2020	108	29	,	,	,
orr-bootleg-2020	108	30	.	.	.
orr-bootleg-2020	109	1	.	.	.
orr-bootleg-2020	110	1	.	.	.
orr-bootleg-2020	111	1	,	,	,
orr-bootleg-2020	111	2	te	te	NNP
orr-bootleg-2020	111	3	,	,	,
orr-bootleg-2020	111	4	T	T	NNP
orr-bootleg-2020	111	5	]	]	-RRB-
orr-bootleg-2020	111	6	)	)	-RRB-
orr-bootleg-2020	111	7	.	.	.
orr-bootleg-2020	112	1	We	-PRON-	PRP
orr-bootleg-2020	112	2	further	further	RB
orr-bootleg-2020	112	3	allow	allow	VBP
orr-bootleg-2020	112	4	the	the	DT
orr-bootleg-2020	112	5	model	model	NN
orr-bootleg-2020	112	6	to	to	TO
orr-bootleg-2020	112	7	leverage	leverage	NN
orr-bootleg-2020	112	8	coarse	coarse	RB
orr-bootleg-2020	112	9	named	name	VBN
orr-bootleg-2020	112	10	entity	entity	NN
orr-bootleg-2020	112	11	recognition	recognition	NN
orr-bootleg-2020	112	12	types	type	NNS
orr-bootleg-2020	112	13	through	through	IN
orr-bootleg-2020	112	14	a	a	DT
orr-bootleg-2020	112	15	mention	mention	JJ
orr-bootleg-2020	112	16	-	-	HYPH
orr-bootleg-2020	112	17	type	type	NN
orr-bootleg-2020	112	18	prediction	prediction	NN
orr-bootleg-2020	112	19	module	module	NN
orr-bootleg-2020	112	20	(	(	-LRB-
orr-bootleg-2020	112	21	see	see	VB
orr-bootleg-2020	112	22	Appendix	Appendix	NNP
orr-bootleg-2020	112	23	A	A	NNP
orr-bootleg-2020	112	24	for	for	IN
orr-bootleg-2020	112	25	details	detail	NNS
orr-bootleg-2020	112	26	)	)	-RRB-
orr-bootleg-2020	112	27	.	.	.
orr-bootleg-2020	113	1	This	this	DT
orr-bootleg-2020	113	2	coarse	coarse	RB
orr-bootleg-2020	113	3	predicted	predict	VBD
orr-bootleg-2020	113	4	type	type	NN
orr-bootleg-2020	113	5	is	be	VBZ
orr-bootleg-2020	113	6	concatenated	concatenate	VBN
orr-bootleg-2020	113	7	with	with	IN
orr-bootleg-2020	113	8	the	the	DT
orr-bootleg-2020	113	9	assigned	assign	VBN
orr-bootleg-2020	113	10	type	type	NN
orr-bootleg-2020	113	11	to	to	TO
orr-bootleg-2020	113	12	form	form	VB
orr-bootleg-2020	113	13	te	te	NNP
orr-bootleg-2020	113	14	.	.	.
orr-bootleg-2020	114	1	•	•	VB
orr-bootleg-2020	114	2	Relation	relation	NN
orr-bootleg-2020	114	3	Embedding	embedding	NN
orr-bootleg-2020	114	4	:	:	:
orr-bootleg-2020	114	5	Let	let	VB
orr-bootleg-2020	114	6	R	r	PRP
orr-bootleg-2020	114	7	represent	represent	VB
orr-bootleg-2020	114	8	the	the	DT
orr-bootleg-2020	114	9	set	set	NN
orr-bootleg-2020	114	10	of	of	IN
orr-bootleg-2020	114	11	possible	possible	JJ
orr-bootleg-2020	114	12	relationships	relationship	NNS
orr-bootleg-2020	114	13	any	any	DT
orr-bootleg-2020	114	14	entity	entity	NN
orr-bootleg-2020	114	15	can	can	MD
orr-bootleg-2020	114	16	participate	participate	VB
orr-bootleg-2020	114	17	in	in	IN
orr-bootleg-2020	114	18	.	.	.
orr-bootleg-2020	115	1	Similar	similar	JJ
orr-bootleg-2020	115	2	to	to	IN
orr-bootleg-2020	115	3	types	type	NNS
orr-bootleg-2020	115	4	,	,	,
orr-bootleg-2020	115	5	given	give	VBN
orr-bootleg-2020	115	6	a	a	DT
orr-bootleg-2020	115	7	mapping	mapping	NN
orr-bootleg-2020	115	8	from	from	IN
orr-bootleg-2020	115	9	an	an	DT
orr-bootleg-2020	115	10	entity	entity	NN
orr-bootleg-2020	115	11	e	e	NN
orr-bootleg-2020	115	12	to	to	IN
orr-bootleg-2020	115	13	its	-PRON-	PRP$
orr-bootleg-2020	115	14	set	set	NN
orr-bootleg-2020	115	15	{	{	-LRB-
orr-bootleg-2020	115	16	re,1	re,1	NNP
orr-bootleg-2020	115	17	,	,	,
orr-bootleg-2020	115	18	.	.	.
orr-bootleg-2020	116	1	.	.	.
orr-bootleg-2020	117	1	.	.	.
orr-bootleg-2020	118	1	,	,	,
orr-bootleg-2020	118	2	re	re	IN
orr-bootleg-2020	118	3	,	,	,
orr-bootleg-2020	118	4	R|re	R|re	NNP
orr-bootleg-2020	118	5	,	,	,
orr-bootleg-2020	118	6	i	i	PRP
orr-bootleg-2020	118	7	∈R	∈r	UH
orr-bootleg-2020	118	8	}	}	-RRB-
orr-bootleg-2020	118	9	of	of	IN
orr-bootleg-2020	118	10	R	r	NN
orr-bootleg-2020	118	11	relationships	relationship	NNS
orr-bootleg-2020	118	12	,	,	,
orr-bootleg-2020	118	13	Bootleg	Bootleg	NNP
orr-bootleg-2020	118	14	assigns	assign	VBZ
orr-bootleg-2020	118	15	an	an	DT
orr-bootleg-2020	118	16	embedding	embedding	NN
orr-bootleg-2020	118	17	re	re	NN
orr-bootleg-2020	118	18	,	,	,
orr-bootleg-2020	118	19	i	i	PRP
orr-bootleg-2020	118	20	to	to	IN
orr-bootleg-2020	118	21	each	each	DT
orr-bootleg-2020	118	22	relation	relation	NN
orr-bootleg-2020	118	23	.	.	.
orr-bootleg-2020	119	1	Because	because	IN
orr-bootleg-2020	119	2	an	an	DT
orr-bootleg-2020	119	3	entity	entity	NN
orr-bootleg-2020	119	4	can	can	MD
orr-bootleg-2020	119	5	participate	participate	VB
orr-bootleg-2020	119	6	in	in	IN
orr-bootleg-2020	119	7	multiple	multiple	JJ
orr-bootleg-2020	119	8	relations	relation	NNS
orr-bootleg-2020	119	9	,	,	,
orr-bootleg-2020	119	10	we	-PRON-	PRP
orr-bootleg-2020	119	11	use	use	VBP
orr-bootleg-2020	119	12	the	the	DT
orr-bootleg-2020	119	13	additive	additive	JJ
orr-bootleg-2020	119	14	attention	attention	NN
orr-bootleg-2020	119	15	to	to	IN
orr-bootleg-2020	119	16	compute	compute	NN
orr-bootleg-2020	119	17	re	re	IN
orr-bootleg-2020	119	18	=	=	SYM
orr-bootleg-2020	119	19	AddAttn([re,1	AddAttn([re,1	NNP
orr-bootleg-2020	119	20	,	,	,
orr-bootleg-2020	119	21	.	.	.
orr-bootleg-2020	120	1	.	.	.
orr-bootleg-2020	121	1	.	.	.
orr-bootleg-2020	122	1	,	,	,
orr-bootleg-2020	122	2	re	re	VB
orr-bootleg-2020	122	3	,	,	,
orr-bootleg-2020	122	4	R	r	NN
orr-bootleg-2020	122	5	]	]	-RRB-
orr-bootleg-2020	122	6	)	)	-RRB-
orr-bootleg-2020	122	7	.	.	.
orr-bootleg-2020	123	1	As	as	IN
orr-bootleg-2020	123	2	in	in	IN
orr-bootleg-2020	123	3	existing	exist	VBG
orr-bootleg-2020	123	4	work	work	NN
orr-bootleg-2020	123	5	[	[	-LRB-
orr-bootleg-2020	123	6	16	16	CD
orr-bootleg-2020	123	7	,	,	,
orr-bootleg-2020	123	8	40	40	CD
orr-bootleg-2020	123	9	]	]	-RRB-
orr-bootleg-2020	123	10	,	,	,
orr-bootleg-2020	123	11	given	give	VBN
orr-bootleg-2020	123	12	the	the	DT
orr-bootleg-2020	123	13	input	input	JJ
orr-bootleg-2020	123	14	sentence	sentence	NN
orr-bootleg-2020	123	15	of	of	IN
orr-bootleg-2020	123	16	length	length	NN
orr-bootleg-2020	123	17	N	n	NN
orr-bootleg-2020	123	18	and	and	CC
orr-bootleg-2020	123	19	set	set	NN
orr-bootleg-2020	123	20	of	of	IN
orr-bootleg-2020	123	21	M	m	NN
orr-bootleg-2020	123	22	mentions	mention	NNS
orr-bootleg-2020	123	23	,	,	,
orr-bootleg-2020	123	24	Bootleg	Bootleg	NNP
orr-bootleg-2020	123	25	generates	generate	VBZ
orr-bootleg-2020	123	26	for	for	IN
orr-bootleg-2020	123	27	each	each	DT
orr-bootleg-2020	123	28	mention	mention	NN
orr-bootleg-2020	123	29	mi	mi	IN
orr-bootleg-2020	123	30	a	a	DT
orr-bootleg-2020	123	31	set	set	NN
orr-bootleg-2020	123	32	Γ(mi	γ(mi	NN
orr-bootleg-2020	123	33	)	)	-RRB-
orr-bootleg-2020	123	34	=	=	NFP
orr-bootleg-2020	123	35	{	{	-LRB-
orr-bootleg-2020	123	36	e1i	e1i	NNP
orr-bootleg-2020	123	37	,	,	,
orr-bootleg-2020	123	38	.	.	.
orr-bootleg-2020	124	1	.	.	.
orr-bootleg-2020	125	1	.	.	.
orr-bootleg-2020	126	1	,	,	,
orr-bootleg-2020	126	2	e	e	NNP
orr-bootleg-2020	126	3	K	K	NNP
orr-bootleg-2020	126	4	i	i	NNP
orr-bootleg-2020	126	5	}	}	-RRB-
orr-bootleg-2020	126	6	of	of	IN
orr-bootleg-2020	126	7	K	k	NN
orr-bootleg-2020	126	8	possible	possible	JJ
orr-bootleg-2020	126	9	entity	entity	NN
orr-bootleg-2020	126	10	candidates	candidate	NNS
orr-bootleg-2020	126	11	that	that	WDT
orr-bootleg-2020	126	12	could	could	MD
orr-bootleg-2020	126	13	be	be	VB
orr-bootleg-2020	126	14	referred	refer	VBN
orr-bootleg-2020	126	15	to	to	IN
orr-bootleg-2020	126	16	by	by	IN
orr-bootleg-2020	126	17	mi	mi	NNP
orr-bootleg-2020	126	18	.	.	.
orr-bootleg-2020	127	1	For	for	IN
orr-bootleg-2020	127	2	each	each	DT
orr-bootleg-2020	127	3	candidate	candidate	NN
orr-bootleg-2020	127	4	and	and	CC
orr-bootleg-2020	127	5	its	-PRON-	PRP$
orr-bootleg-2020	127	6	associated	associated	JJ
orr-bootleg-2020	127	7	types	type	NNS
orr-bootleg-2020	127	8	and	and	CC
orr-bootleg-2020	127	9	relations	relation	NNS
orr-bootleg-2020	127	10	,	,	,
orr-bootleg-2020	127	11	Bootleg	Bootleg	NNP
orr-bootleg-2020	127	12	uses	use	VBZ
orr-bootleg-2020	127	13	a	a	DT
orr-bootleg-2020	127	14	multi	multi	JJ
orr-bootleg-2020	127	15	-	-	JJ
orr-bootleg-2020	127	16	layer	layer	NN
orr-bootleg-2020	127	17	perceptron	perceptron	NNP
orr-bootleg-2020	127	18	e	e	NNP
orr-bootleg-2020	127	19	=	=	-RRB-
orr-bootleg-2020	127	20	MLP([ue	MLP([ue	NNP
orr-bootleg-2020	127	21	,	,	,
orr-bootleg-2020	127	22	te	te	NNP
orr-bootleg-2020	127	23	,	,	,
orr-bootleg-2020	127	24	re	re	NN
orr-bootleg-2020	127	25	]	]	-RRB-
orr-bootleg-2020	127	26	)	)	-RRB-
orr-bootleg-2020	127	27	to	to	TO
orr-bootleg-2020	127	28	generate	generate	VB
orr-bootleg-2020	127	29	a	a	DT
orr-bootleg-2020	127	30	vector	vector	NN
orr-bootleg-2020	127	31	representation	representation	NN
orr-bootleg-2020	127	32	for	for	IN
orr-bootleg-2020	127	33	each	each	DT
orr-bootleg-2020	127	34	candidate	candidate	NN
orr-bootleg-2020	127	35	entity	entity	NN
orr-bootleg-2020	127	36	,	,	,
orr-bootleg-2020	127	37	for	for	IN
orr-bootleg-2020	127	38	each	each	DT
orr-bootleg-2020	127	39	mention	mention	NN
orr-bootleg-2020	127	40	.	.	.
orr-bootleg-2020	128	1	We	-PRON-	PRP
orr-bootleg-2020	128	2	denote	denote	VBP
orr-bootleg-2020	128	3	this	this	DT
orr-bootleg-2020	128	4	entity	entity	NN
orr-bootleg-2020	128	5	matrix	matrix	NN
orr-bootleg-2020	128	6	as	as	IN
orr-bootleg-2020	128	7	E	E	NNP
orr-bootleg-2020	128	8	∈	∈	NNP
orr-bootleg-2020	128	9	RM×K×H	RM×K×H	NNP
orr-bootleg-2020	128	10	,	,	,
orr-bootleg-2020	128	11	where	where	WRB
orr-bootleg-2020	128	12	H	h	NN
orr-bootleg-2020	128	13	is	be	VBZ
orr-bootleg-2020	128	14	the	the	DT
orr-bootleg-2020	128	15	hidden	hidden	JJ
orr-bootleg-2020	128	16	dimension	dimension	NN
orr-bootleg-2020	128	17	.	.	.
orr-bootleg-2020	129	1	We	-PRON-	PRP
orr-bootleg-2020	129	2	use	use	VBP
orr-bootleg-2020	129	3	BERT	BERT	NNP
orr-bootleg-2020	129	4	to	to	TO
orr-bootleg-2020	129	5	generate	generate	VB
orr-bootleg-2020	129	6	contextual	contextual	JJ
orr-bootleg-2020	129	7	embeddings	embedding	NNS
orr-bootleg-2020	129	8	for	for	IN
orr-bootleg-2020	129	9	each	each	DT
orr-bootleg-2020	129	10	token	token	VBN
orr-bootleg-2020	129	11	in	in	IN
orr-bootleg-2020	129	12	the	the	DT
orr-bootleg-2020	129	13	input	input	NN
orr-bootleg-2020	129	14	sentence	sentence	NN
orr-bootleg-2020	129	15	.	.	.
orr-bootleg-2020	130	1	We	-PRON-	PRP
orr-bootleg-2020	130	2	denote	denote	VBP
orr-bootleg-2020	130	3	this	this	DT
orr-bootleg-2020	130	4	sentence	sentence	NN
orr-bootleg-2020	130	5	embedding	embed	VBG
orr-bootleg-2020	130	6	as	as	IN
orr-bootleg-2020	130	7	W	W	NNP
orr-bootleg-2020	130	8	∈	∈	NNP
orr-bootleg-2020	130	9	RN×H.	RN×H.	NNP
orr-bootleg-2020	131	1	W	w	NN
orr-bootleg-2020	131	2	and	and	CC
orr-bootleg-2020	131	3	E	e	NN
orr-bootleg-2020	131	4	are	be	VBP
orr-bootleg-2020	131	5	passed	pass	VBN
orr-bootleg-2020	131	6	to	to	IN
orr-bootleg-2020	131	7	Bootleg	Bootleg	NNP
orr-bootleg-2020	131	8	’s	’s	POS
orr-bootleg-2020	131	9	model	model	NN
orr-bootleg-2020	131	10	architecture	architecture	NN
orr-bootleg-2020	131	11	,	,	,
orr-bootleg-2020	131	12	described	describe	VBN
orr-bootleg-2020	131	13	next	next	RB
orr-bootleg-2020	131	14	.	.	.
orr-bootleg-2020	132	1	3.2	3.2	CD
orr-bootleg-2020	132	2	Bootleg	Bootleg	NNP
orr-bootleg-2020	132	3	Model	Model	NNP
orr-bootleg-2020	132	4	Architecture	Architecture	NNP
orr-bootleg-2020	132	5	The	the	DT
orr-bootleg-2020	132	6	design	design	NN
orr-bootleg-2020	132	7	goal	goal	NN
orr-bootleg-2020	132	8	of	of	IN
orr-bootleg-2020	132	9	Bootleg	Bootleg	NNP
orr-bootleg-2020	132	10	is	be	VBZ
orr-bootleg-2020	132	11	to	to	TO
orr-bootleg-2020	132	12	capture	capture	VB
orr-bootleg-2020	132	13	the	the	DT
orr-bootleg-2020	132	14	reasoning	reasoning	NN
orr-bootleg-2020	132	15	patterns	pattern	NNS
orr-bootleg-2020	132	16	by	by	IN
orr-bootleg-2020	132	17	modeling	model	VBG
orr-bootleg-2020	132	18	textual	textual	JJ
orr-bootleg-2020	132	19	signals	signal	NNS
orr-bootleg-2020	132	20	associated	associate	VBN
orr-bootleg-2020	132	21	with	with	IN
orr-bootleg-2020	132	22	entities	entity	NNS
orr-bootleg-2020	132	23	(	(	-LRB-
orr-bootleg-2020	132	24	for	for	IN
orr-bootleg-2020	132	25	entity	entity	NN
orr-bootleg-2020	132	26	memorization	memorization	NN
orr-bootleg-2020	132	27	)	)	-RRB-
orr-bootleg-2020	132	28	,	,	,
orr-bootleg-2020	132	29	co	co	NNS
orr-bootleg-2020	132	30	-	-	NNS
orr-bootleg-2020	132	31	occurrences	occurrence	NNS
orr-bootleg-2020	132	32	between	between	IN
orr-bootleg-2020	132	33	entity	entity	NN
orr-bootleg-2020	132	34	types	type	NNS
orr-bootleg-2020	132	35	(	(	-LRB-
orr-bootleg-2020	132	36	for	for	IN
orr-bootleg-2020	132	37	type	type	NN
orr-bootleg-2020	132	38	consistency	consistency	NN
orr-bootleg-2020	132	39	)	)	-RRB-
orr-bootleg-2020	132	40	,	,	,
orr-bootleg-2020	132	41	textual	textual	JJ
orr-bootleg-2020	132	42	signals	signal	NNS
orr-bootleg-2020	132	43	associated	associate	VBN
orr-bootleg-2020	132	44	with	with	IN
orr-bootleg-2020	132	45	relations	relation	NNS
orr-bootleg-2020	132	46	along	along	IN
orr-bootleg-2020	132	47	with	with	IN
orr-bootleg-2020	132	48	which	which	WDT
orr-bootleg-2020	132	49	entities	entity	NNS
orr-bootleg-2020	132	50	are	be	VBP
orr-bootleg-2020	132	51	explicitly	explicitly	RB
orr-bootleg-2020	132	52	linked	link	VBN
orr-bootleg-2020	132	53	in	in	IN
orr-bootleg-2020	132	54	the	the	DT
orr-bootleg-2020	132	55	KG	kg	NN
orr-bootleg-2020	132	56	(	(	-LRB-
orr-bootleg-2020	132	57	for	for	IN
orr-bootleg-2020	132	58	KG	kg	NN
orr-bootleg-2020	132	59	relations	relation	NNS
orr-bootleg-2020	132	60	)	)	-RRB-
orr-bootleg-2020	132	61	,	,	,
orr-bootleg-2020	132	62	5	5	CD
orr-bootleg-2020	132	63	and	and	CC
orr-bootleg-2020	132	64	textual	textual	JJ
orr-bootleg-2020	132	65	signals	signal	NNS
orr-bootleg-2020	132	66	associated	associate	VBN
orr-bootleg-2020	132	67	with	with	IN
orr-bootleg-2020	132	68	types	type	NNS
orr-bootleg-2020	132	69	(	(	-LRB-
orr-bootleg-2020	132	70	for	for	IN
orr-bootleg-2020	132	71	type	type	NN
orr-bootleg-2020	132	72	affordance	affordance	NN
orr-bootleg-2020	132	73	)	)	-RRB-
orr-bootleg-2020	132	74	.	.	.
orr-bootleg-2020	133	1	We	-PRON-	PRP
orr-bootleg-2020	133	2	design	design	VBP
orr-bootleg-2020	133	3	three	three	CD
orr-bootleg-2020	133	4	modules	module	NNS
orr-bootleg-2020	133	5	to	to	TO
orr-bootleg-2020	133	6	capture	capture	VB
orr-bootleg-2020	133	7	these	these	DT
orr-bootleg-2020	133	8	design	design	NN
orr-bootleg-2020	133	9	goals	goal	NNS
orr-bootleg-2020	133	10	:	:	:
orr-bootleg-2020	133	11	a	a	DT
orr-bootleg-2020	133	12	phrase	phrase	NN
orr-bootleg-2020	133	13	memorization	memorization	NN
orr-bootleg-2020	133	14	module	module	NN
orr-bootleg-2020	133	15	,	,	,
orr-bootleg-2020	133	16	a	a	DT
orr-bootleg-2020	133	17	co	co	NN
orr-bootleg-2020	133	18	-	-	JJ
orr-bootleg-2020	133	19	occurrence	occurrence	JJ
orr-bootleg-2020	133	20	memorization	memorization	NN
orr-bootleg-2020	133	21	module	module	NNP
orr-bootleg-2020	133	22	,	,	,
orr-bootleg-2020	133	23	and	and	CC
orr-bootleg-2020	133	24	a	a	DT
orr-bootleg-2020	133	25	knowledge	knowledge	NN
orr-bootleg-2020	133	26	graph	graph	NN
orr-bootleg-2020	133	27	connection	connection	NN
orr-bootleg-2020	133	28	module	module	NN
orr-bootleg-2020	133	29	.	.	.
orr-bootleg-2020	134	1	The	the	DT
orr-bootleg-2020	134	2	model	model	NN
orr-bootleg-2020	134	3	architecture	architecture	NN
orr-bootleg-2020	134	4	is	be	VBZ
orr-bootleg-2020	134	5	shown	show	VBN
orr-bootleg-2020	134	6	in	in	IN
orr-bootleg-2020	134	7	Figure	Figure	NNP
orr-bootleg-2020	134	8	2	2	CD
orr-bootleg-2020	134	9	.	.	.
orr-bootleg-2020	135	1	We	-PRON-	PRP
orr-bootleg-2020	135	2	describe	describe	VBP
orr-bootleg-2020	135	3	each	each	DT
orr-bootleg-2020	135	4	module	module	NN
orr-bootleg-2020	135	5	next	next	RB
orr-bootleg-2020	135	6	.	.	.
orr-bootleg-2020	136	1	Phrase	Phrase	NNP
orr-bootleg-2020	136	2	Memorization	Memorization	NNP
orr-bootleg-2020	136	3	Module	Module	NNP
orr-bootleg-2020	136	4	We	-PRON-	PRP
orr-bootleg-2020	136	5	design	design	VBP
orr-bootleg-2020	136	6	the	the	DT
orr-bootleg-2020	136	7	phrase	phrase	NN
orr-bootleg-2020	136	8	memorization	memorization	NN
orr-bootleg-2020	136	9	module	module	NN
orr-bootleg-2020	136	10	,	,	,
orr-bootleg-2020	136	11	Phrase2Ent	Phrase2Ent	NNP
orr-bootleg-2020	136	12	,	,	,
orr-bootleg-2020	136	13	to	to	TO
orr-bootleg-2020	136	14	encode	encode	VB
orr-bootleg-2020	136	15	the	the	DT
orr-bootleg-2020	136	16	dependencies	dependency	NNS
orr-bootleg-2020	136	17	between	between	IN
orr-bootleg-2020	136	18	the	the	DT
orr-bootleg-2020	136	19	input	input	NN
orr-bootleg-2020	136	20	text	text	NN
orr-bootleg-2020	136	21	and	and	CC
orr-bootleg-2020	136	22	the	the	DT
orr-bootleg-2020	136	23	entity	entity	NN
orr-bootleg-2020	136	24	,	,	,
orr-bootleg-2020	136	25	relation	relation	NN
orr-bootleg-2020	136	26	,	,	,
orr-bootleg-2020	136	27	and	and	CC
orr-bootleg-2020	136	28	type	type	NN
orr-bootleg-2020	136	29	embeddings	embedding	NNS
orr-bootleg-2020	136	30	.	.	.
orr-bootleg-2020	137	1	The	the	DT
orr-bootleg-2020	137	2	purpose	purpose	NN
orr-bootleg-2020	137	3	of	of	IN
orr-bootleg-2020	137	4	this	this	DT
orr-bootleg-2020	137	5	module	module	NN
orr-bootleg-2020	137	6	is	be	VBZ
orr-bootleg-2020	137	7	to	to	TO
orr-bootleg-2020	137	8	learn	learn	VB
orr-bootleg-2020	137	9	textual	textual	JJ
orr-bootleg-2020	137	10	cues	cue	NNS
orr-bootleg-2020	137	11	for	for	IN
orr-bootleg-2020	137	12	the	the	DT
orr-bootleg-2020	137	13	entity	entity	NN
orr-bootleg-2020	137	14	memorization	memorization	NN
orr-bootleg-2020	137	15	and	and	CC
orr-bootleg-2020	137	16	type	type	NN
orr-bootleg-2020	137	17	affordance	affordance	NN
orr-bootleg-2020	137	18	patterns	pattern	NNS
orr-bootleg-2020	137	19	.	.	.
orr-bootleg-2020	138	1	It	-PRON-	PRP
orr-bootleg-2020	138	2	should	should	MD
orr-bootleg-2020	138	3	also	also	RB
orr-bootleg-2020	138	4	learn	learn	VB
orr-bootleg-2020	138	5	relation	relation	NN
orr-bootleg-2020	138	6	context	context	NN
orr-bootleg-2020	138	7	for	for	IN
orr-bootleg-2020	138	8	the	the	DT
orr-bootleg-2020	138	9	KG	KG	NNP
orr-bootleg-2020	138	10	relation	relation	NN
orr-bootleg-2020	138	11	pattern	pattern	NN
orr-bootleg-2020	138	12	.	.	.
orr-bootleg-2020	139	1	It	-PRON-	PRP
orr-bootleg-2020	139	2	will	will	MD
orr-bootleg-2020	139	3	,	,	,
orr-bootleg-2020	139	4	for	for	IN
orr-bootleg-2020	139	5	example	example	NN
orr-bootleg-2020	139	6	,	,	,
orr-bootleg-2020	139	7	allow	allow	VB
orr-bootleg-2020	139	8	the	the	DT
orr-bootleg-2020	139	9	person	person	NN
orr-bootleg-2020	139	10	type	type	NN
orr-bootleg-2020	139	11	embedding	embed	VBG
orr-bootleg-2020	139	12	to	to	TO
orr-bootleg-2020	139	13	encode	encode	VB
orr-bootleg-2020	139	14	the	the	DT
orr-bootleg-2020	139	15	association	association	NN
orr-bootleg-2020	139	16	with	with	IN
orr-bootleg-2020	139	17	the	the	DT
orr-bootleg-2020	139	18	keyword	keyword	NNP
orr-bootleg-2020	139	19	“	"	``
orr-bootleg-2020	139	20	height	height	NN
orr-bootleg-2020	139	21	”	"	''
orr-bootleg-2020	139	22	.	.	.
orr-bootleg-2020	140	1	The	the	DT
orr-bootleg-2020	140	2	module	module	NN
orr-bootleg-2020	140	3	accepts	accept	VBZ
orr-bootleg-2020	140	4	as	as	IN
orr-bootleg-2020	140	5	input	input	NN
orr-bootleg-2020	140	6	E	e	NN
orr-bootleg-2020	140	7	and	and	CC
orr-bootleg-2020	140	8	W	w	NN
orr-bootleg-2020	140	9	and	and	CC
orr-bootleg-2020	140	10	outputs	output	NNS
orr-bootleg-2020	140	11	Ep	Ep	NNP
orr-bootleg-2020	140	12	=	=	SYM
orr-bootleg-2020	140	13	MHA(E	MHA(E	NNP
orr-bootleg-2020	140	14	,	,	,
orr-bootleg-2020	140	15	W	w	NN
orr-bootleg-2020	140	16	)	)	-RRB-
orr-bootleg-2020	140	17	,	,	,
orr-bootleg-2020	140	18	where	where	WRB
orr-bootleg-2020	140	19	MHA	MHA	NNP
orr-bootleg-2020	140	20	is	be	VBZ
orr-bootleg-2020	140	21	the	the	DT
orr-bootleg-2020	140	22	standard	standard	JJ
orr-bootleg-2020	140	23	multi	multi	JJ
orr-bootleg-2020	140	24	-	-	JJ
orr-bootleg-2020	140	25	headed	headed	JJ
orr-bootleg-2020	140	26	attention	attention	NN
orr-bootleg-2020	140	27	with	with	IN
orr-bootleg-2020	140	28	a	a	DT
orr-bootleg-2020	140	29	feed	feed	NN
orr-bootleg-2020	140	30	-	-	HYPH
orr-bootleg-2020	140	31	forward	forward	NN
orr-bootleg-2020	140	32	layer	layer	NN
orr-bootleg-2020	140	33	and	and	CC
orr-bootleg-2020	140	34	skip	skip	NN
orr-bootleg-2020	140	35	connections	connection	NNS
orr-bootleg-2020	140	36	[	[	-LRB-
orr-bootleg-2020	140	37	48	48	CD
orr-bootleg-2020	140	38	]	]	-RRB-
orr-bootleg-2020	140	39	.	.	.
orr-bootleg-2020	141	1	Co	co	NN
orr-bootleg-2020	141	2	-	-	NN
orr-bootleg-2020	141	3	occurrence	occurrence	NN
orr-bootleg-2020	141	4	Memorization	Memorization	NNP
orr-bootleg-2020	141	5	Module	Module	NNP
orr-bootleg-2020	141	6	We	-PRON-	PRP
orr-bootleg-2020	141	7	design	design	VBP
orr-bootleg-2020	141	8	the	the	DT
orr-bootleg-2020	141	9	co	co	NN
orr-bootleg-2020	141	10	-	-	NN
orr-bootleg-2020	141	11	occurrence	occurrence	JJ
orr-bootleg-2020	141	12	memorization	memorization	NN
orr-bootleg-2020	141	13	module	module	NN
orr-bootleg-2020	141	14	,	,	,
orr-bootleg-2020	141	15	Ent2Ent	Ent2Ent	NNP
orr-bootleg-2020	141	16	,	,	,
orr-bootleg-2020	141	17	to	to	TO
orr-bootleg-2020	141	18	encode	encode	VB
orr-bootleg-2020	141	19	the	the	DT
orr-bootleg-2020	141	20	dependencies	dependency	NNS
orr-bootleg-2020	141	21	between	between	IN
orr-bootleg-2020	141	22	entities	entity	NNS
orr-bootleg-2020	141	23	.	.	.
orr-bootleg-2020	142	1	The	the	DT
orr-bootleg-2020	142	2	purpose	purpose	NN
orr-bootleg-2020	142	3	of	of	IN
orr-bootleg-2020	142	4	the	the	DT
orr-bootleg-2020	142	5	Ent2Ent	Ent2Ent	NNP
orr-bootleg-2020	142	6	module	module	NN
orr-bootleg-2020	142	7	is	be	VBZ
orr-bootleg-2020	142	8	to	to	TO
orr-bootleg-2020	142	9	learn	learn	VB
orr-bootleg-2020	142	10	textual	textual	JJ
orr-bootleg-2020	142	11	cues	cue	NNS
orr-bootleg-2020	142	12	for	for	IN
orr-bootleg-2020	142	13	the	the	DT
orr-bootleg-2020	142	14	type	type	NN
orr-bootleg-2020	142	15	consistency	consistency	NN
orr-bootleg-2020	142	16	pattern	pattern	NN
orr-bootleg-2020	142	17	.	.	.
orr-bootleg-2020	143	1	The	the	DT
orr-bootleg-2020	143	2	module	module	NN
orr-bootleg-2020	143	3	accepts	accept	VBZ
orr-bootleg-2020	143	4	E	e	NN
orr-bootleg-2020	143	5	and	and	CC
orr-bootleg-2020	143	6	computes	compute	NNS
orr-bootleg-2020	143	7	Ec	Ec	NNP
orr-bootleg-2020	143	8	=	=	SYM
orr-bootleg-2020	143	9	MHA(E	MHA(E	NNP
orr-bootleg-2020	143	10	)	)	-RRB-
orr-bootleg-2020	143	11	using	use	VBG
orr-bootleg-2020	143	12	self	self	NN
orr-bootleg-2020	143	13	-	-	HYPH
orr-bootleg-2020	143	14	attention	attention	NN
orr-bootleg-2020	143	15	.	.	.
orr-bootleg-2020	144	1	Knowledge	Knowledge	NNP
orr-bootleg-2020	144	2	Graph	Graph	NNP
orr-bootleg-2020	144	3	(	(	-LRB-
orr-bootleg-2020	144	4	KG	KG	NNP
orr-bootleg-2020	144	5	)	)	-RRB-
orr-bootleg-2020	144	6	Connection	connection	NN
orr-bootleg-2020	144	7	Module	Module	NNP
orr-bootleg-2020	144	8	We	-PRON-	PRP
orr-bootleg-2020	144	9	design	design	VBP
orr-bootleg-2020	144	10	the	the	DT
orr-bootleg-2020	144	11	KG	kg	NN
orr-bootleg-2020	144	12	module	module	NN
orr-bootleg-2020	144	13	,	,	,
orr-bootleg-2020	144	14	KG2Ent	kg2ent	NN
orr-bootleg-2020	144	15	,	,	,
orr-bootleg-2020	144	16	to	to	TO
orr-bootleg-2020	144	17	collectively	collectively	RB
orr-bootleg-2020	144	18	resolve	resolve	VB
orr-bootleg-2020	144	19	entities	entity	NNS
orr-bootleg-2020	144	20	based	base	VBN
orr-bootleg-2020	144	21	on	on	IN
orr-bootleg-2020	144	22	pairwise	pairwise	NN
orr-bootleg-2020	144	23	connectivity	connectivity	NN
orr-bootleg-2020	144	24	features	feature	NNS
orr-bootleg-2020	144	25	.	.	.
orr-bootleg-2020	145	1	Let	let	VB
orr-bootleg-2020	145	2	K	K	NNP
orr-bootleg-2020	145	3	represent	represent	VB
orr-bootleg-2020	145	4	the	the	DT
orr-bootleg-2020	145	5	adjacency	adjacency	NN
orr-bootleg-2020	145	6	matrix	matrix	NN
orr-bootleg-2020	145	7	of	of	IN
orr-bootleg-2020	145	8	a	a	DT
orr-bootleg-2020	145	9	(	(	-LRB-
orr-bootleg-2020	145	10	possibly	possibly	RB
orr-bootleg-2020	145	11	weighted	weight	VBN
orr-bootleg-2020	145	12	)	)	-RRB-
orr-bootleg-2020	145	13	graph	graph	NN
orr-bootleg-2020	145	14	where	where	WRB
orr-bootleg-2020	145	15	the	the	DT
orr-bootleg-2020	145	16	nodes	node	NNS
orr-bootleg-2020	145	17	are	be	VBP
orr-bootleg-2020	145	18	entities	entity	NNS
orr-bootleg-2020	145	19	and	and	CC
orr-bootleg-2020	145	20	an	an	DT
orr-bootleg-2020	145	21	edge	edge	NN
orr-bootleg-2020	145	22	between	between	IN
orr-bootleg-2020	145	23	ei	ei	NNP
orr-bootleg-2020	145	24	and	and	CC
orr-bootleg-2020	145	25	ej	ej	NNP
orr-bootleg-2020	145	26	signifies	signify	VBZ
orr-bootleg-2020	145	27	that	that	IN
orr-bootleg-2020	145	28	the	the	DT
orr-bootleg-2020	145	29	two	two	CD
orr-bootleg-2020	145	30	entities	entity	NNS
orr-bootleg-2020	145	31	share	share	VBP
orr-bootleg-2020	145	32	some	some	DT
orr-bootleg-2020	145	33	pairwise	pairwise	NN
orr-bootleg-2020	145	34	feature	feature	NN
orr-bootleg-2020	145	35	.	.	.
orr-bootleg-2020	146	1	Given	give	VBN
orr-bootleg-2020	146	2	E	e	NN
orr-bootleg-2020	146	3	,	,	,
orr-bootleg-2020	146	4	KG2Ent	kg2ent	NN
orr-bootleg-2020	146	5	computes	compute	VBZ
orr-bootleg-2020	146	6	Ek	Ek	NNP
orr-bootleg-2020	146	7	=	=	SYM
orr-bootleg-2020	146	8	softmax(K	softmax(k	XX
orr-bootleg-2020	146	9	+	+	SYM
orr-bootleg-2020	146	10	wI)E	wI)E	NNS
orr-bootleg-2020	146	11	+	+	SYM
orr-bootleg-2020	146	12	E	e	NN
orr-bootleg-2020	146	13	where	where	WRB
orr-bootleg-2020	146	14	I	-PRON-	PRP
orr-bootleg-2020	146	15	is	be	VBZ
orr-bootleg-2020	146	16	the	the	DT
orr-bootleg-2020	146	17	identity	identity	NN
orr-bootleg-2020	146	18	and	and	CC
orr-bootleg-2020	146	19	w	w	NNP
orr-bootleg-2020	146	20	is	be	VBZ
orr-bootleg-2020	146	21	a	a	DT
orr-bootleg-2020	146	22	learned	learn	VBN
orr-bootleg-2020	146	23	scalar	scalar	JJ
orr-bootleg-2020	146	24	weight	weight	NN
orr-bootleg-2020	146	25	that	that	WDT
orr-bootleg-2020	146	26	allows	allow	VBZ
orr-bootleg-2020	146	27	Bootleg	Bootleg	NNP
orr-bootleg-2020	146	28	to	to	TO
orr-bootleg-2020	146	29	learn	learn	VB
orr-bootleg-2020	146	30	to	to	TO
orr-bootleg-2020	146	31	balance	balance	VB
orr-bootleg-2020	146	32	the	the	DT
orr-bootleg-2020	146	33	original	original	JJ
orr-bootleg-2020	146	34	entity	entity	NN
orr-bootleg-2020	146	35	and	and	CC
orr-bootleg-2020	146	36	its	-PRON-	PRP$
orr-bootleg-2020	146	37	connections	connection	NNS
orr-bootleg-2020	146	38	.	.	.
orr-bootleg-2020	147	1	This	this	DT
orr-bootleg-2020	147	2	module	module	NN
orr-bootleg-2020	147	3	allows	allow	VBZ
orr-bootleg-2020	147	4	for	for	IN
orr-bootleg-2020	147	5	representation	representation	NN
orr-bootleg-2020	147	6	transfer	transfer	NN
orr-bootleg-2020	147	7	between	between	IN
orr-bootleg-2020	147	8	two	two	CD
orr-bootleg-2020	147	9	related	related	JJ
orr-bootleg-2020	147	10	entities	entity	NNS
orr-bootleg-2020	147	11	,	,	,
orr-bootleg-2020	147	12	meaning	mean	VBG
orr-bootleg-2020	147	13	entities	entity	NNS
orr-bootleg-2020	147	14	with	with	IN
orr-bootleg-2020	147	15	a	a	DT
orr-bootleg-2020	147	16	high	high	RB
orr-bootleg-2020	147	17	-	-	HYPH
orr-bootleg-2020	147	18	scoring	score	VBG
orr-bootleg-2020	147	19	representation	representation	NN
orr-bootleg-2020	147	20	will	will	MD
orr-bootleg-2020	147	21	boost	boost	VB
orr-bootleg-2020	147	22	the	the	DT
orr-bootleg-2020	147	23	score	score	NN
orr-bootleg-2020	147	24	of	of	IN
orr-bootleg-2020	147	25	related	related	JJ
orr-bootleg-2020	147	26	entities	entity	NNS
orr-bootleg-2020	147	27	.	.	.
orr-bootleg-2020	148	1	The	the	DT
orr-bootleg-2020	148	2	second	second	JJ
orr-bootleg-2020	148	3	computation	computation	NN
orr-bootleg-2020	148	4	acts	act	VBZ
orr-bootleg-2020	148	5	as	as	IN
orr-bootleg-2020	148	6	a	a	DT
orr-bootleg-2020	148	7	skip	skip	NN
orr-bootleg-2020	148	8	connection	connection	NN
orr-bootleg-2020	148	9	between	between	IN
orr-bootleg-2020	148	10	the	the	DT
orr-bootleg-2020	148	11	input	input	NN
orr-bootleg-2020	148	12	and	and	CC
orr-bootleg-2020	148	13	output	output	NN
orr-bootleg-2020	148	14	.	.	.
orr-bootleg-2020	149	1	In	in	IN
orr-bootleg-2020	149	2	Bootleg	Bootleg	NNP
orr-bootleg-2020	149	3	,	,	,
orr-bootleg-2020	149	4	we	-PRON-	PRP
orr-bootleg-2020	149	5	allow	allow	VBP
orr-bootleg-2020	149	6	the	the	DT
orr-bootleg-2020	149	7	user	user	NN
orr-bootleg-2020	149	8	to	to	TO
orr-bootleg-2020	149	9	specify	specify	VB
orr-bootleg-2020	149	10	multiple	multiple	JJ
orr-bootleg-2020	149	11	KG2Ent	kg2ent	NN
orr-bootleg-2020	149	12	modules	module	NNS
orr-bootleg-2020	149	13	:	:	:
orr-bootleg-2020	149	14	one	one	CD
orr-bootleg-2020	149	15	for	for	IN
orr-bootleg-2020	149	16	each	each	DT
orr-bootleg-2020	149	17	adjacency	adjacency	NN
orr-bootleg-2020	149	18	matrix	matrix	NN
orr-bootleg-2020	149	19	.	.	.
orr-bootleg-2020	150	1	The	the	DT
orr-bootleg-2020	150	2	purpose	purpose	NN
orr-bootleg-2020	150	3	of	of	IN
orr-bootleg-2020	150	4	KG2Ent	kg2ent	NN
orr-bootleg-2020	150	5	along	along	IN
orr-bootleg-2020	150	6	with	with	IN
orr-bootleg-2020	150	7	Phrase2Ent	phrase2ent	NN
orr-bootleg-2020	150	8	is	be	VBZ
orr-bootleg-2020	150	9	to	to	TO
orr-bootleg-2020	150	10	learn	learn	VB
orr-bootleg-2020	150	11	the	the	DT
orr-bootleg-2020	150	12	KG	kg	NN
orr-bootleg-2020	150	13	relation	relation	NN
orr-bootleg-2020	150	14	pattern	pattern	NN
orr-bootleg-2020	150	15	.	.	.
orr-bootleg-2020	151	1	End	end	NN
orr-bootleg-2020	151	2	-	-	HYPH
orr-bootleg-2020	151	3	to	to	IN
orr-bootleg-2020	151	4	-	-	HYPH
orr-bootleg-2020	151	5	End	end	NN
orr-bootleg-2020	151	6	The	the	DT
orr-bootleg-2020	151	7	computations	computation	NNS
orr-bootleg-2020	151	8	for	for	IN
orr-bootleg-2020	151	9	one	one	CD
orr-bootleg-2020	151	10	layer	layer	NN
orr-bootleg-2020	151	11	of	of	IN
orr-bootleg-2020	151	12	Bootleg	Bootleg	NNP
orr-bootleg-2020	151	13	includes	include	VBZ
orr-bootleg-2020	151	14	:	:	:
orr-bootleg-2020	151	15	E′	e′	ADD
orr-bootleg-2020	151	16	=	=	SYM
orr-bootleg-2020	151	17	MHA(E	MHA(E	NNP
orr-bootleg-2020	151	18	,	,	,
orr-bootleg-2020	151	19	W	w	NN
orr-bootleg-2020	151	20	)	)	-RRB-
orr-bootleg-2020	151	21	+	+	SYM
orr-bootleg-2020	151	22	MHA(E	MHA(E	NNP
orr-bootleg-2020	151	23	)	)	-RRB-
orr-bootleg-2020	151	24	Ek	Ek	NNP
orr-bootleg-2020	151	25	=	=	NFP
orr-bootleg-2020	151	26	softmax(K	softmax(K	.
orr-bootleg-2020	151	27	+	+	CC
orr-bootleg-2020	151	28	wI)E′	wI)E′	NNS
orr-bootleg-2020	151	29	+	+	CC
orr-bootleg-2020	151	30	E′	e′	XX
orr-bootleg-2020	151	31	where	where	WRB
orr-bootleg-2020	151	32	Ek	Ek	NNP
orr-bootleg-2020	151	33	is	be	VBZ
orr-bootleg-2020	151	34	passed	pass	VBN
orr-bootleg-2020	151	35	as	as	IN
orr-bootleg-2020	151	36	the	the	DT
orr-bootleg-2020	151	37	entity	entity	NN
orr-bootleg-2020	151	38	matrix	matrix	NN
orr-bootleg-2020	151	39	to	to	IN
orr-bootleg-2020	151	40	the	the	DT
orr-bootleg-2020	151	41	next	next	JJ
orr-bootleg-2020	151	42	layer	layer	NN
orr-bootleg-2020	151	43	.	.	.
orr-bootleg-2020	152	1	After	after	IN
orr-bootleg-2020	152	2	the	the	DT
orr-bootleg-2020	152	3	final	final	JJ
orr-bootleg-2020	152	4	layer	layer	NN
orr-bootleg-2020	152	5	,	,	,
orr-bootleg-2020	152	6	Bootleg	Bootleg	NNP
orr-bootleg-2020	152	7	scores	score	VBZ
orr-bootleg-2020	152	8	each	each	DT
orr-bootleg-2020	152	9	entity	entity	NN
orr-bootleg-2020	152	10	by	by	IN
orr-bootleg-2020	152	11	computing	compute	VBG
orr-bootleg-2020	152	12	Sdis	Sdis	NNP
orr-bootleg-2020	152	13	=	=	SYM
orr-bootleg-2020	152	14	max(EkvT	max(ekvt	NN
orr-bootleg-2020	152	15	,	,	,
orr-bootleg-2020	152	16	E′vT	E′vT	NNP
orr-bootleg-2020	152	17	)	)	-RRB-
orr-bootleg-2020	152	18	with	with	IN
orr-bootleg-2020	152	19	Sdis	Sdis	NNP
orr-bootleg-2020	152	20	∈	∈	NNP
orr-bootleg-2020	152	21	RM×K	RM×K	,
orr-bootleg-2020	152	22	and	and	CC
orr-bootleg-2020	152	23	learned	learn	VBD
orr-bootleg-2020	152	24	scoring	score	VBG
orr-bootleg-2020	152	25	vector	vector	NN
orr-bootleg-2020	152	26	v	v	NNP
orr-bootleg-2020	152	27	∈	∈	NNP
orr-bootleg-2020	152	28	RH	RH	NNP
orr-bootleg-2020	152	29	.	.	.
orr-bootleg-2020	153	1	Bootleg	Bootleg	NNP
orr-bootleg-2020	153	2	then	then	RB
orr-bootleg-2020	153	3	outputs	output	VBZ
orr-bootleg-2020	153	4	the	the	DT
orr-bootleg-2020	153	5	highest	high	JJS
orr-bootleg-2020	153	6	scoring	score	VBG
orr-bootleg-2020	153	7	candidate	candidate	NN
orr-bootleg-2020	153	8	for	for	IN
orr-bootleg-2020	153	9	each	each	DT
orr-bootleg-2020	153	10	mention	mention	NN
orr-bootleg-2020	153	11	.	.	.
orr-bootleg-2020	154	1	This	this	DT
orr-bootleg-2020	154	2	scoring	score	VBG
orr-bootleg-2020	154	3	treats	treat	VBZ
orr-bootleg-2020	154	4	Ek	Ek	NNP
orr-bootleg-2020	154	5	and	and	CC
orr-bootleg-2020	154	6	E′	e′	ADD
orr-bootleg-2020	154	7	as	as	IN
orr-bootleg-2020	154	8	two	two	CD
orr-bootleg-2020	154	9	separate	separate	JJ
orr-bootleg-2020	154	10	predictions	prediction	NNS
orr-bootleg-2020	154	11	in	in	IN
orr-bootleg-2020	154	12	an	an	DT
orr-bootleg-2020	154	13	ensemble	ensemble	JJ
orr-bootleg-2020	154	14	method	method	NN
orr-bootleg-2020	154	15	,	,	,
orr-bootleg-2020	154	16	allowing	allow	VBG
orr-bootleg-2020	154	17	the	the	DT
orr-bootleg-2020	154	18	model	model	NN
orr-bootleg-2020	154	19	to	to	TO
orr-bootleg-2020	154	20	use	use	VB
orr-bootleg-2020	154	21	collective	collective	JJ
orr-bootleg-2020	154	22	reasoning	reasoning	NN
orr-bootleg-2020	154	23	from	from	IN
orr-bootleg-2020	154	24	Ek	Ek	NNP
orr-bootleg-2020	154	25	when	when	WRB
orr-bootleg-2020	154	26	it	-PRON-	PRP
orr-bootleg-2020	154	27	achieves	achieve	VBZ
orr-bootleg-2020	154	28	the	the	DT
orr-bootleg-2020	154	29	highest	high	JJS
orr-bootleg-2020	154	30	scoring	score	VBG
orr-bootleg-2020	154	31	representation	representation	NN
orr-bootleg-2020	154	32	.	.	.
orr-bootleg-2020	155	1	If	if	IN
orr-bootleg-2020	155	2	there	there	EX
orr-bootleg-2020	155	3	are	be	VBP
orr-bootleg-2020	155	4	multiple	multiple	JJ
orr-bootleg-2020	155	5	KG2Ent	kg2ent	NN
orr-bootleg-2020	155	6	modules	module	NNS
orr-bootleg-2020	155	7	,	,	,
orr-bootleg-2020	155	8	we	-PRON-	PRP
orr-bootleg-2020	155	9	use	use	VBP
orr-bootleg-2020	155	10	the	the	DT
orr-bootleg-2020	155	11	average	average	NN
orr-bootleg-2020	155	12	of	of	IN
orr-bootleg-2020	155	13	their	-PRON-	PRP$
orr-bootleg-2020	155	14	outputs	output	NNS
orr-bootleg-2020	155	15	as	as	IN
orr-bootleg-2020	155	16	input	input	NN
orr-bootleg-2020	155	17	to	to	IN
orr-bootleg-2020	155	18	the	the	DT
orr-bootleg-2020	155	19	next	next	JJ
orr-bootleg-2020	155	20	layer	layer	NN
orr-bootleg-2020	155	21	and	and	CC
orr-bootleg-2020	155	22	,	,	,
orr-bootleg-2020	155	23	for	for	IN
orr-bootleg-2020	155	24	scoring	score	VBG
orr-bootleg-2020	155	25	,	,	,
orr-bootleg-2020	155	26	take	take	VB
orr-bootleg-2020	155	27	the	the	DT
orr-bootleg-2020	155	28	maximum	maximum	JJ
orr-bootleg-2020	155	29	score	score	NN
orr-bootleg-2020	155	30	across	across	IN
orr-bootleg-2020	155	31	all	all	DT
orr-bootleg-2020	155	32	outputs	output	NNS
orr-bootleg-2020	155	33	.	.	.
orr-bootleg-2020	156	1	For	for	IN
orr-bootleg-2020	156	2	training	training	NN
orr-bootleg-2020	156	3	,	,	,
orr-bootleg-2020	156	4	we	-PRON-	PRP
orr-bootleg-2020	156	5	use	use	VBP
orr-bootleg-2020	156	6	the	the	DT
orr-bootleg-2020	156	7	cross	cross	JJ
orr-bootleg-2020	156	8	-	-	JJ
orr-bootleg-2020	156	9	entropy	entropy	JJ
orr-bootleg-2020	156	10	loss	loss	NN
orr-bootleg-2020	156	11	of	of	IN
orr-bootleg-2020	156	12	S	S	NNP
orr-bootleg-2020	156	13	to	to	TO
orr-bootleg-2020	156	14	compute	compute	VB
orr-bootleg-2020	156	15	the	the	DT
orr-bootleg-2020	156	16	disambiguation	disambiguation	NN
orr-bootleg-2020	156	17	loss	loss	NN
orr-bootleg-2020	156	18	Ldis	Ldis	NNP
orr-bootleg-2020	156	19	.	.	.
orr-bootleg-2020	157	1	3.3	3.3	CD
orr-bootleg-2020	157	2	Improving	Improving	NNP
orr-bootleg-2020	157	3	Tail	Tail	NNP
orr-bootleg-2020	157	4	Generalization	Generalization	NNP
orr-bootleg-2020	157	5	Regularization	Regularization	NNP
orr-bootleg-2020	157	6	is	be	VBZ
orr-bootleg-2020	157	7	the	the	DT
orr-bootleg-2020	157	8	standard	standard	JJ
orr-bootleg-2020	157	9	technique	technique	NN
orr-bootleg-2020	157	10	to	to	TO
orr-bootleg-2020	157	11	encourage	encourage	VB
orr-bootleg-2020	157	12	models	model	NNS
orr-bootleg-2020	157	13	to	to	TO
orr-bootleg-2020	157	14	generalize	generalize	VB
orr-bootleg-2020	157	15	,	,	,
orr-bootleg-2020	157	16	as	as	IN
orr-bootleg-2020	157	17	models	model	NNS
orr-bootleg-2020	157	18	will	will	MD
orr-bootleg-2020	157	19	naturally	naturally	RB
orr-bootleg-2020	157	20	fit	fit	VB
orr-bootleg-2020	157	21	to	to	IN
orr-bootleg-2020	157	22	discriminative	discriminative	JJ
orr-bootleg-2020	157	23	features	feature	NNS
orr-bootleg-2020	157	24	.	.	.
orr-bootleg-2020	158	1	However	however	RB
orr-bootleg-2020	158	2	,	,	,
orr-bootleg-2020	158	3	we	-PRON-	PRP
orr-bootleg-2020	158	4	demonstrate	demonstrate	VBP
orr-bootleg-2020	158	5	that	that	IN
orr-bootleg-2020	158	6	standard	standard	JJ
orr-bootleg-2020	158	7	regularization	regularization	NN
orr-bootleg-2020	158	8	is	be	VBZ
orr-bootleg-2020	158	9	not	not	RB
orr-bootleg-2020	158	10	effective	effective	JJ
orr-bootleg-2020	158	11	when	when	WRB
orr-bootleg-2020	158	12	we	-PRON-	PRP
orr-bootleg-2020	158	13	want	want	VBP
orr-bootleg-2020	158	14	to	to	TO
orr-bootleg-2020	158	15	leverage	leverage	VB
orr-bootleg-2020	158	16	a	a	DT
orr-bootleg-2020	158	17	combination	combination	NN
orr-bootleg-2020	158	18	of	of	IN
orr-bootleg-2020	158	19	general	general	JJ
orr-bootleg-2020	158	20	and	and	CC
orr-bootleg-2020	158	21	discriminative	discriminative	JJ
orr-bootleg-2020	158	22	signals	signal	NNS
orr-bootleg-2020	158	23	.	.	.
orr-bootleg-2020	159	1	We	-PRON-	PRP
orr-bootleg-2020	159	2	then	then	RB
orr-bootleg-2020	159	3	present	present	VBP
orr-bootleg-2020	159	4	two	two	CD
orr-bootleg-2020	159	5	techniques	technique	NNS
orr-bootleg-2020	159	6	,	,	,
orr-bootleg-2020	159	7	regularization	regularization	NN
orr-bootleg-2020	159	8	and	and	CC
orr-bootleg-2020	159	9	weak	weak	JJ
orr-bootleg-2020	159	10	labeling	labeling	NN
orr-bootleg-2020	159	11	,	,	,
orr-bootleg-2020	159	12	to	to	TO
orr-bootleg-2020	159	13	encourage	encourage	VB
orr-bootleg-2020	159	14	Bootleg	Bootleg	NNP
orr-bootleg-2020	159	15	to	to	TO
orr-bootleg-2020	159	16	incorporate	incorporate	VB
orr-bootleg-2020	159	17	general	general	JJ
orr-bootleg-2020	159	18	structural	structural	JJ
orr-bootleg-2020	159	19	signals	signal	NNS
orr-bootleg-2020	159	20	and	and	CC
orr-bootleg-2020	159	21	learn	learn	VB
orr-bootleg-2020	159	22	general	general	JJ
orr-bootleg-2020	159	23	reasoning	reasoning	NN
orr-bootleg-2020	159	24	patterns	pattern	NNS
orr-bootleg-2020	159	25	.	.	.
orr-bootleg-2020	160	1	3.3.1	3.3.1	CD
orr-bootleg-2020	160	2	Regularization	regularization	NN
orr-bootleg-2020	160	3	We	-PRON-	PRP
orr-bootleg-2020	160	4	hypothesize	hypothesize	VBP
orr-bootleg-2020	160	5	that	that	IN
orr-bootleg-2020	160	6	Bootleg	Bootleg	NNP
orr-bootleg-2020	160	7	will	will	MD
orr-bootleg-2020	160	8	over	over	RB
orr-bootleg-2020	160	9	-	-	HYPH
orr-bootleg-2020	160	10	rely	rely	VB
orr-bootleg-2020	160	11	on	on	IN
orr-bootleg-2020	160	12	the	the	DT
orr-bootleg-2020	160	13	more	more	RBR
orr-bootleg-2020	160	14	discriminative	discriminative	JJ
orr-bootleg-2020	160	15	entity	entity	NN
orr-bootleg-2020	160	16	features	feature	NNS
orr-bootleg-2020	160	17	compared	compare	VBN
orr-bootleg-2020	160	18	to	to	IN
orr-bootleg-2020	160	19	the	the	DT
orr-bootleg-2020	160	20	more	more	RBR
orr-bootleg-2020	160	21	general	general	JJ
orr-bootleg-2020	160	22	type	type	NN
orr-bootleg-2020	160	23	and	and	CC
orr-bootleg-2020	160	24	relation	relation	NN
orr-bootleg-2020	160	25	features	feature	NNS
orr-bootleg-2020	160	26	to	to	TO
orr-bootleg-2020	160	27	lower	lower	VB
orr-bootleg-2020	160	28	training	training	NN
orr-bootleg-2020	160	29	loss	loss	NN
orr-bootleg-2020	160	30	.	.	.
orr-bootleg-2020	161	1	However	however	RB
orr-bootleg-2020	161	2	,	,	,
orr-bootleg-2020	161	3	tail	tail	NN
orr-bootleg-2020	161	4	disambiguation	disambiguation	NN
orr-bootleg-2020	161	5	requires	require	VBZ
orr-bootleg-2020	161	6	Bootleg	Bootleg	NNP
orr-bootleg-2020	161	7	to	to	TO
orr-bootleg-2020	161	8	leverage	leverage	VB
orr-bootleg-2020	161	9	the	the	DT
orr-bootleg-2020	161	10	general	general	JJ
orr-bootleg-2020	161	11	features	feature	NNS
orr-bootleg-2020	161	12	.	.	.
orr-bootleg-2020	162	1	Using	use	VBG
orr-bootleg-2020	162	2	standard	standard	JJ
orr-bootleg-2020	162	3	regularization	regularization	NN
orr-bootleg-2020	162	4	techniques	technique	NNS
orr-bootleg-2020	162	5	,	,	,
orr-bootleg-2020	162	6	we	-PRON-	PRP
orr-bootleg-2020	162	7	evaluate	evaluate	VBP
orr-bootleg-2020	162	8	three	three	CD
orr-bootleg-2020	162	9	models	model	NNS
orr-bootleg-2020	162	10	which	which	WDT
orr-bootleg-2020	162	11	respectively	respectively	RB
orr-bootleg-2020	162	12	use	use	VBP
orr-bootleg-2020	162	13	only	only	JJ
orr-bootleg-2020	162	14	type	type	NN
orr-bootleg-2020	162	15	embeddings	embedding	NNS
orr-bootleg-2020	162	16	,	,	,
orr-bootleg-2020	162	17	only	only	RB
orr-bootleg-2020	162	18	relation	relation	NN
orr-bootleg-2020	162	19	embeddings	embedding	NNS
orr-bootleg-2020	162	20	,	,	,
orr-bootleg-2020	162	21	and	and	CC
orr-bootleg-2020	162	22	a	a	DT
orr-bootleg-2020	162	23	combination	combination	NN
orr-bootleg-2020	162	24	of	of	IN
orr-bootleg-2020	162	25	type	type	NN
orr-bootleg-2020	162	26	,	,	,
orr-bootleg-2020	162	27	relation	relation	NN
orr-bootleg-2020	162	28	,	,	,
orr-bootleg-2020	162	29	and	and	CC
orr-bootleg-2020	162	30	6	6	CD
orr-bootleg-2020	162	31	entity	entity	NN
orr-bootleg-2020	162	32	embeddings	embedding	NNS
orr-bootleg-2020	162	33	.	.	.
orr-bootleg-2020	163	1	Bootleg	Bootleg	NNP
orr-bootleg-2020	163	2	’s	’s	POS
orr-bootleg-2020	163	3	performance	performance	NN
orr-bootleg-2020	163	4	on	on	IN
orr-bootleg-2020	163	5	unseen	unseen	JJ
orr-bootleg-2020	163	6	entities	entity	NNS
orr-bootleg-2020	163	7	is	be	VBZ
orr-bootleg-2020	163	8	10	10	CD
orr-bootleg-2020	163	9	F1	f1	NN
orr-bootleg-2020	163	10	points	point	NNS
orr-bootleg-2020	163	11	worse	bad	JJR
orr-bootleg-2020	163	12	on	on	IN
orr-bootleg-2020	163	13	the	the	DT
orr-bootleg-2020	163	14	latter	latter	JJ
orr-bootleg-2020	163	15	than	than	IN
orr-bootleg-2020	163	16	each	each	DT
orr-bootleg-2020	163	17	of	of	IN
orr-bootleg-2020	163	18	the	the	DT
orr-bootleg-2020	163	19	former	former	JJ
orr-bootleg-2020	163	20	two	two	CD
orr-bootleg-2020	163	21	,	,	,
orr-bootleg-2020	163	22	suggesting	suggest	VBG
orr-bootleg-2020	163	23	that	that	IN
orr-bootleg-2020	163	24	standard	standard	JJ
orr-bootleg-2020	163	25	regularization	regularization	NN
orr-bootleg-2020	163	26	is	be	VBZ
orr-bootleg-2020	163	27	not	not	RB
orr-bootleg-2020	163	28	sufficient	sufficient	JJ
orr-bootleg-2020	163	29	when	when	WRB
orr-bootleg-2020	163	30	the	the	DT
orr-bootleg-2020	163	31	signals	signal	NNS
orr-bootleg-2020	163	32	operate	operate	VBP
orr-bootleg-2020	163	33	at	at	IN
orr-bootleg-2020	163	34	different	different	JJ
orr-bootleg-2020	163	35	granularities	granularity	NNS
orr-bootleg-2020	163	36	(	(	-LRB-
orr-bootleg-2020	163	37	details	detail	NNS
orr-bootleg-2020	163	38	Table	table	NN
orr-bootleg-2020	163	39	9	9	CD
orr-bootleg-2020	163	40	in	in	IN
orr-bootleg-2020	163	41	Appendix	Appendix	NNP
orr-bootleg-2020	163	42	B	B	NNP
orr-bootleg-2020	163	43	)	)	-RRB-
orr-bootleg-2020	163	44	.	.	.
orr-bootleg-2020	164	1	We	-PRON-	PRP
orr-bootleg-2020	164	2	can	can	MD
orr-bootleg-2020	164	3	improve	improve	VB
orr-bootleg-2020	164	4	tail	tail	NN
orr-bootleg-2020	164	5	performance	performance	NN
orr-bootleg-2020	164	6	if	if	IN
orr-bootleg-2020	164	7	Bootleg	Bootleg	NNP
orr-bootleg-2020	164	8	leverages	leverage	NNS
orr-bootleg-2020	164	9	memorized	memorize	VBD
orr-bootleg-2020	164	10	discriminative	discriminative	JJ
orr-bootleg-2020	164	11	features	feature	NNS
orr-bootleg-2020	164	12	for	for	IN
orr-bootleg-2020	164	13	popular	popular	JJ
orr-bootleg-2020	164	14	entities	entity	NNS
orr-bootleg-2020	164	15	and	and	CC
orr-bootleg-2020	164	16	general	general	JJ
orr-bootleg-2020	164	17	features	feature	NNS
orr-bootleg-2020	164	18	for	for	IN
orr-bootleg-2020	164	19	rare	rare	JJ
orr-bootleg-2020	164	20	entities	entity	NNS
orr-bootleg-2020	164	21	.	.	.
orr-bootleg-2020	165	1	We	-PRON-	PRP
orr-bootleg-2020	165	2	achieve	achieve	VBP
orr-bootleg-2020	165	3	this	this	DT
orr-bootleg-2020	165	4	by	by	IN
orr-bootleg-2020	165	5	designing	design	VBG
orr-bootleg-2020	165	6	a	a	DT
orr-bootleg-2020	165	7	new	new	JJ
orr-bootleg-2020	165	8	regularization	regularization	NN
orr-bootleg-2020	165	9	scheme	scheme	NN
orr-bootleg-2020	165	10	for	for	IN
orr-bootleg-2020	165	11	the	the	DT
orr-bootleg-2020	165	12	entity	entity	NN
orr-bootleg-2020	165	13	-	-	HYPH
orr-bootleg-2020	165	14	specific	specific	NN
orr-bootleg-2020	165	15	embedding	embed	VBG
orr-bootleg-2020	165	16	u	u	NN
orr-bootleg-2020	165	17	,	,	,
orr-bootleg-2020	165	18	which	which	WDT
orr-bootleg-2020	165	19	has	have	VBZ
orr-bootleg-2020	165	20	two	two	CD
orr-bootleg-2020	165	21	key	key	JJ
orr-bootleg-2020	165	22	properties	property	NNS
orr-bootleg-2020	165	23	:	:	:
orr-bootleg-2020	165	24	it	-PRON-	PRP
orr-bootleg-2020	165	25	is	be	VBZ
orr-bootleg-2020	165	26	2-dimensional	2-dimensional	CD
orr-bootleg-2020	165	27	and	and	CC
orr-bootleg-2020	165	28	more	more	RBR
orr-bootleg-2020	165	29	popular	popular	JJ
orr-bootleg-2020	165	30	entities	entity	NNS
orr-bootleg-2020	165	31	are	be	VBP
orr-bootleg-2020	165	32	regularized	regularize	VBN
orr-bootleg-2020	165	33	less	less	RBR
orr-bootleg-2020	165	34	than	than	IN
orr-bootleg-2020	165	35	less	less	RBR
orr-bootleg-2020	165	36	popular	popular	JJ
orr-bootleg-2020	165	37	ones	one	NNS
orr-bootleg-2020	165	38	.	.	.
orr-bootleg-2020	166	1	•	•	NNP
orr-bootleg-2020	166	2	2-dimensional	2-dimensional	CD
orr-bootleg-2020	166	3	:	:	:
orr-bootleg-2020	166	4	In	in	IN
orr-bootleg-2020	166	5	contrast	contrast	NN
orr-bootleg-2020	166	6	to	to	IN
orr-bootleg-2020	166	7	1-dimensional	1-dimensional	CD
orr-bootleg-2020	166	8	dropout	dropout	NN
orr-bootleg-2020	166	9	,	,	,
orr-bootleg-2020	166	10	2-dimensional	2-dimensional	CD
orr-bootleg-2020	166	11	regularization	regularization	NN
orr-bootleg-2020	166	12	involves	involve	VBZ
orr-bootleg-2020	166	13	masking	mask	VBG
orr-bootleg-2020	166	14	the	the	DT
orr-bootleg-2020	166	15	full	full	JJ
orr-bootleg-2020	166	16	embedding	embedding	NN
orr-bootleg-2020	166	17	.	.	.
orr-bootleg-2020	167	1	With	with	IN
orr-bootleg-2020	167	2	probability	probability	NN
orr-bootleg-2020	167	3	p(e	p(e	NNP
orr-bootleg-2020	167	4	)	)	-RRB-
orr-bootleg-2020	167	5	,	,	,
orr-bootleg-2020	167	6	we	-PRON-	PRP
orr-bootleg-2020	167	7	set	set	VBD
orr-bootleg-2020	167	8	u	u	NNP
orr-bootleg-2020	167	9	=	=	SYM
orr-bootleg-2020	167	10	0	0	CD
orr-bootleg-2020	167	11	before	before	IN
orr-bootleg-2020	167	12	the	the	DT
orr-bootleg-2020	167	13	MLP	MLP	NNP
orr-bootleg-2020	167	14	layer	layer	NN
orr-bootleg-2020	167	15	;	;	:
orr-bootleg-2020	167	16	i.e.	i.e.	FW
orr-bootleg-2020	167	17	,	,	,
orr-bootleg-2020	167	18	e	e	LS
orr-bootleg-2020	167	19	=	=	SYM
orr-bootleg-2020	167	20	MLP([0	MLP([0	NNP
orr-bootleg-2020	167	21	,	,	,
orr-bootleg-2020	167	22	te	te	NNP
orr-bootleg-2020	167	23	,	,	,
orr-bootleg-2020	167	24	re	re	NN
orr-bootleg-2020	167	25	]	]	-RRB-
orr-bootleg-2020	167	26	)	)	-RRB-
orr-bootleg-2020	167	27	.	.	.
orr-bootleg-2020	168	1	Entirely	entirely	RB
orr-bootleg-2020	168	2	masking	mask	VBG
orr-bootleg-2020	168	3	the	the	DT
orr-bootleg-2020	168	4	entity	entity	NN
orr-bootleg-2020	168	5	embedding	embedding	NN
orr-bootleg-2020	168	6	in	in	IN
orr-bootleg-2020	168	7	these	these	DT
orr-bootleg-2020	168	8	cases	case	NNS
orr-bootleg-2020	168	9	,	,	,
orr-bootleg-2020	168	10	the	the	DT
orr-bootleg-2020	168	11	model	model	NN
orr-bootleg-2020	168	12	learns	learn	VBZ
orr-bootleg-2020	168	13	to	to	TO
orr-bootleg-2020	168	14	disambiguate	disambiguate	VB
orr-bootleg-2020	168	15	using	use	VBG
orr-bootleg-2020	168	16	the	the	DT
orr-bootleg-2020	168	17	type	type	NN
orr-bootleg-2020	168	18	and	and	CC
orr-bootleg-2020	168	19	relation	relation	NN
orr-bootleg-2020	168	20	patterns	pattern	NNS
orr-bootleg-2020	168	21	,	,	,
orr-bootleg-2020	168	22	without	without	IN
orr-bootleg-2020	168	23	entity	entity	NN
orr-bootleg-2020	168	24	knowledge	knowledge	NN
orr-bootleg-2020	168	25	.	.	.
orr-bootleg-2020	169	1	•	•	VB
orr-bootleg-2020	169	2	Inverse	Inverse	NNP
orr-bootleg-2020	169	3	Popularity	popularity	NN
orr-bootleg-2020	169	4	:	:	:
orr-bootleg-2020	169	5	We	-PRON-	PRP
orr-bootleg-2020	169	6	find	find	VBP
orr-bootleg-2020	169	7	in	in	IN
orr-bootleg-2020	169	8	ablations	ablation	NNS
orr-bootleg-2020	169	9	(	(	-LRB-
orr-bootleg-2020	169	10	Appendix	Appendix	NNP
orr-bootleg-2020	169	11	B	B	NNP
orr-bootleg-2020	169	12	)	)	-RRB-
orr-bootleg-2020	169	13	that	that	IN
orr-bootleg-2020	169	14	setting	set	VBG
orr-bootleg-2020	169	15	p(e	p(e	NNP
orr-bootleg-2020	169	16	)	)	-RRB-
orr-bootleg-2020	169	17	proportional	proportional	VBD
orr-bootleg-2020	169	18	to	to	IN
orr-bootleg-2020	169	19	the	the	DT
orr-bootleg-2020	169	20	power	power	NN
orr-bootleg-2020	169	21	of	of	IN
orr-bootleg-2020	169	22	the	the	DT
orr-bootleg-2020	169	23	inverse	inverse	NN
orr-bootleg-2020	169	24	of	of	IN
orr-bootleg-2020	169	25	the	the	DT
orr-bootleg-2020	169	26	entity	entity	NN
orr-bootleg-2020	169	27	e	e	NN
orr-bootleg-2020	169	28	’s	’s	POS
orr-bootleg-2020	169	29	popularity	popularity	NN
orr-bootleg-2020	169	30	in	in	IN
orr-bootleg-2020	169	31	the	the	DT
orr-bootleg-2020	169	32	training	training	NN
orr-bootleg-2020	169	33	data	datum	NNS
orr-bootleg-2020	169	34	(	(	-LRB-
orr-bootleg-2020	169	35	i.e.	i.e.	FW
orr-bootleg-2020	169	36	,	,	,
orr-bootleg-2020	169	37	the	the	DT
orr-bootleg-2020	169	38	more	more	RBR
orr-bootleg-2020	169	39	popular	popular	JJ
orr-bootleg-2020	169	40	the	the	DT
orr-bootleg-2020	169	41	less	less	RBR
orr-bootleg-2020	169	42	regularized	regularized	JJ
orr-bootleg-2020	169	43	)	)	-RRB-
orr-bootleg-2020	169	44	,	,	,
orr-bootleg-2020	169	45	gives	give	VBZ
orr-bootleg-2020	169	46	us	-PRON-	PRP
orr-bootleg-2020	169	47	the	the	DT
orr-bootleg-2020	169	48	best	good	JJS
orr-bootleg-2020	169	49	performance	performance	NN
orr-bootleg-2020	169	50	and	and	CC
orr-bootleg-2020	169	51	improves	improve	VBZ
orr-bootleg-2020	169	52	by	by	IN
orr-bootleg-2020	169	53	13.6	13.6	CD
orr-bootleg-2020	169	54	F1	f1	NN
orr-bootleg-2020	169	55	on	on	IN
orr-bootleg-2020	169	56	unseen	unseen	JJ
orr-bootleg-2020	169	57	entities	entity	NNS
orr-bootleg-2020	169	58	over	over	IN
orr-bootleg-2020	169	59	standard	standard	JJ
orr-bootleg-2020	169	60	regularization	regularization	NN
orr-bootleg-2020	169	61	.	.	.
orr-bootleg-2020	170	1	In	in	IN
orr-bootleg-2020	170	2	contrast	contrast	NN
orr-bootleg-2020	170	3	,	,	,
orr-bootleg-2020	170	4	fixing	fix	VBG
orr-bootleg-2020	170	5	p(e	p(e	NNP
orr-bootleg-2020	170	6	)	)	-RRB-
orr-bootleg-2020	170	7	at	at	IN
orr-bootleg-2020	170	8	80	80	CD
orr-bootleg-2020	170	9	%	%	NN
orr-bootleg-2020	170	10	improves	improve	VBZ
orr-bootleg-2020	170	11	performance	performance	NN
orr-bootleg-2020	170	12	by	by	IN
orr-bootleg-2020	170	13	over	over	IN
orr-bootleg-2020	170	14	11.3	11.3	CD
orr-bootleg-2020	170	15	F1	f1	NN
orr-bootleg-2020	170	16	over	over	IN
orr-bootleg-2020	170	17	standard	standard	JJ
orr-bootleg-2020	170	18	regularization	regularization	NN
orr-bootleg-2020	170	19	,	,	,
orr-bootleg-2020	170	20	and	and	CC
orr-bootleg-2020	170	21	regularizing	regularize	VBG
orr-bootleg-2020	170	22	proportional	proportional	JJ
orr-bootleg-2020	170	23	to	to	IN
orr-bootleg-2020	170	24	the	the	DT
orr-bootleg-2020	170	25	power	power	NN
orr-bootleg-2020	170	26	of	of	IN
orr-bootleg-2020	170	27	popularity	popularity	NN
orr-bootleg-2020	170	28	only	only	RB
orr-bootleg-2020	170	29	improves	improve	VBZ
orr-bootleg-2020	170	30	performance	performance	NN
orr-bootleg-2020	170	31	by	by	IN
orr-bootleg-2020	170	32	3.8	3.8	CD
orr-bootleg-2020	170	33	F1	f1	NN
orr-bootleg-2020	170	34	(	(	-LRB-
orr-bootleg-2020	170	35	details	detail	NNS
orr-bootleg-2020	170	36	in	in	IN
orr-bootleg-2020	170	37	Section	section	NN
orr-bootleg-2020	170	38	4	4	CD
orr-bootleg-2020	170	39	)	)	-RRB-
orr-bootleg-2020	170	40	.	.	.
orr-bootleg-2020	171	1	The	the	DT
orr-bootleg-2020	171	2	regularization	regularization	NN
orr-bootleg-2020	171	3	scheme	scheme	NN
orr-bootleg-2020	171	4	encourages	encourage	VBZ
orr-bootleg-2020	171	5	Bootleg	Bootleg	NNP
orr-bootleg-2020	171	6	to	to	TO
orr-bootleg-2020	171	7	use	use	VB
orr-bootleg-2020	171	8	entity	entity	NN
orr-bootleg-2020	171	9	-	-	HYPH
orr-bootleg-2020	171	10	specific	specific	JJ
orr-bootleg-2020	171	11	knowledge	knowledge	NN
orr-bootleg-2020	171	12	when	when	WRB
orr-bootleg-2020	171	13	the	the	DT
orr-bootleg-2020	171	14	entity	entity	NN
orr-bootleg-2020	171	15	is	be	VBZ
orr-bootleg-2020	171	16	seen	see	VBN
orr-bootleg-2020	171	17	enough	enough	JJ
orr-bootleg-2020	171	18	times	time	NNS
orr-bootleg-2020	171	19	to	to	TO
orr-bootleg-2020	171	20	memorize	memorize	VB
orr-bootleg-2020	171	21	entity	entity	NN
orr-bootleg-2020	171	22	patterns	pattern	NNS
orr-bootleg-2020	171	23	and	and	CC
orr-bootleg-2020	171	24	encourages	encourage	VBZ
orr-bootleg-2020	171	25	the	the	DT
orr-bootleg-2020	171	26	use	use	NN
orr-bootleg-2020	171	27	of	of	IN
orr-bootleg-2020	171	28	generalizable	generalizable	JJ
orr-bootleg-2020	171	29	patterns	pattern	NNS
orr-bootleg-2020	171	30	over	over	IN
orr-bootleg-2020	171	31	the	the	DT
orr-bootleg-2020	171	32	rare	rare	JJ
orr-bootleg-2020	171	33	,	,	,
orr-bootleg-2020	171	34	highly	highly	RB
orr-bootleg-2020	171	35	masked	masked	JJ
orr-bootleg-2020	171	36	,	,	,
orr-bootleg-2020	171	37	entities	entity	NNS
orr-bootleg-2020	171	38	.	.	.
orr-bootleg-2020	172	1	3.3.2	3.3.2	CD
orr-bootleg-2020	172	2	Weakly	Weakly	NNP
orr-bootleg-2020	172	3	Supervised	supervise	VBD
orr-bootleg-2020	172	4	Data	Data	NNPS
orr-bootleg-2020	172	5	Labeling	labeling	NN
orr-bootleg-2020	172	6	We	-PRON-	PRP
orr-bootleg-2020	172	7	use	use	VBP
orr-bootleg-2020	172	8	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	172	9	to	to	TO
orr-bootleg-2020	172	10	train	train	VB
orr-bootleg-2020	172	11	Bootleg	Bootleg	NNP
orr-bootleg-2020	172	12	:	:	:
orr-bootleg-2020	172	13	we	-PRON-	PRP
orr-bootleg-2020	172	14	define	define	VBP
orr-bootleg-2020	172	15	a	a	DT
orr-bootleg-2020	172	16	self	self	NN
orr-bootleg-2020	172	17	-	-	HYPH
orr-bootleg-2020	172	18	supervision	supervision	NN
orr-bootleg-2020	172	19	task	task	NN
orr-bootleg-2020	172	20	in	in	IN
orr-bootleg-2020	172	21	which	which	WDT
orr-bootleg-2020	172	22	the	the	DT
orr-bootleg-2020	172	23	internal	internal	JJ
orr-bootleg-2020	172	24	links	link	NNS
orr-bootleg-2020	172	25	in	in	IN
orr-bootleg-2020	172	26	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	172	27	are	be	VBP
orr-bootleg-2020	172	28	the	the	DT
orr-bootleg-2020	172	29	gold	gold	JJ
orr-bootleg-2020	172	30	entity	entity	NN
orr-bootleg-2020	172	31	labels	label	NNS
orr-bootleg-2020	172	32	for	for	IN
orr-bootleg-2020	172	33	mentions	mention	NNS
orr-bootleg-2020	172	34	during	during	IN
orr-bootleg-2020	172	35	training	training	NN
orr-bootleg-2020	172	36	.	.	.
orr-bootleg-2020	173	1	Although	although	IN
orr-bootleg-2020	173	2	this	this	DT
orr-bootleg-2020	173	3	dataset	dataset	NN
orr-bootleg-2020	173	4	is	be	VBZ
orr-bootleg-2020	173	5	large	large	JJ
orr-bootleg-2020	173	6	and	and	CC
orr-bootleg-2020	173	7	widely	widely	RB
orr-bootleg-2020	173	8	used	use	VBN
orr-bootleg-2020	173	9	,	,	,
orr-bootleg-2020	173	10	it	-PRON-	PRP
orr-bootleg-2020	173	11	is	be	VBZ
orr-bootleg-2020	173	12	often	often	RB
orr-bootleg-2020	173	13	incomplete	incomplete	JJ
orr-bootleg-2020	173	14	with	with	IN
orr-bootleg-2020	173	15	an	an	DT
orr-bootleg-2020	173	16	estimated	estimate	VBN
orr-bootleg-2020	173	17	68	68	CD
orr-bootleg-2020	173	18	%	%	NN
orr-bootleg-2020	173	19	of	of	IN
orr-bootleg-2020	173	20	named	name	VBN
orr-bootleg-2020	173	21	entities	entity	NNS
orr-bootleg-2020	173	22	being	be	VBG
orr-bootleg-2020	173	23	unlabeled	unlabele	VBN
orr-bootleg-2020	173	24	.	.	.
orr-bootleg-2020	174	1	Given	give	VBN
orr-bootleg-2020	174	2	the	the	DT
orr-bootleg-2020	174	3	scale	scale	NN
orr-bootleg-2020	174	4	and	and	CC
orr-bootleg-2020	174	5	the	the	DT
orr-bootleg-2020	174	6	requirement	requirement	NN
orr-bootleg-2020	174	7	that	that	IN
orr-bootleg-2020	174	8	Bootleg	Bootleg	NNP
orr-bootleg-2020	174	9	be	be	VB
orr-bootleg-2020	174	10	self	self	NN
orr-bootleg-2020	174	11	-	-	HYPH
orr-bootleg-2020	174	12	supervised	supervise	VBN
orr-bootleg-2020	174	13	,	,	,
orr-bootleg-2020	174	14	it	-PRON-	PRP
orr-bootleg-2020	174	15	is	be	VBZ
orr-bootleg-2020	174	16	not	not	RB
orr-bootleg-2020	174	17	feasible	feasible	JJ
orr-bootleg-2020	174	18	to	to	IN
orr-bootleg-2020	174	19	hand	hand	NN
orr-bootleg-2020	174	20	-	-	HYPH
orr-bootleg-2020	174	21	label	label	VB
orr-bootleg-2020	174	22	the	the	DT
orr-bootleg-2020	174	23	data	datum	NNS
orr-bootleg-2020	174	24	.	.	.
orr-bootleg-2020	175	1	Our	-PRON-	PRP$
orr-bootleg-2020	175	2	insight	insight	NN
orr-bootleg-2020	175	3	is	be	VBZ
orr-bootleg-2020	175	4	that	that	IN
orr-bootleg-2020	175	5	because	because	IN
orr-bootleg-2020	175	6	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	175	7	is	be	VBZ
orr-bootleg-2020	175	8	highly	highly	RB
orr-bootleg-2020	175	9	structured	structured	JJ
orr-bootleg-2020	175	10	—	—	:
orr-bootleg-2020	175	11	most	most	JJS
orr-bootleg-2020	175	12	sentences	sentence	NNS
orr-bootleg-2020	175	13	on	on	IN
orr-bootleg-2020	175	14	an	an	DT
orr-bootleg-2020	175	15	entity	entity	NN
orr-bootleg-2020	175	16	’s	’s	POS
orr-bootleg-2020	175	17	Wikipedia	wikipedia	NN
orr-bootleg-2020	175	18	page	page	NN
orr-bootleg-2020	175	19	refer	refer	NN
orr-bootleg-2020	175	20	to	to	IN
orr-bootleg-2020	175	21	that	that	DT
orr-bootleg-2020	175	22	entity	entity	NN
orr-bootleg-2020	175	23	via	via	IN
orr-bootleg-2020	175	24	pronouns	pronoun	NNS
orr-bootleg-2020	175	25	or	or	CC
orr-bootleg-2020	175	26	alternative	alternative	JJ
orr-bootleg-2020	175	27	names	name	NNS
orr-bootleg-2020	175	28	—	—	:
orr-bootleg-2020	175	29	we	-PRON-	PRP
orr-bootleg-2020	175	30	can	can	MD
orr-bootleg-2020	175	31	weakly	weakly	RB
orr-bootleg-2020	175	32	label	label	VB
orr-bootleg-2020	175	33	our	-PRON-	PRP$
orr-bootleg-2020	175	34	training	training	NN
orr-bootleg-2020	175	35	data	datum	NNS
orr-bootleg-2020	175	36	[	[	-LRB-
orr-bootleg-2020	175	37	44	44	CD
orr-bootleg-2020	175	38	]	]	-RRB-
orr-bootleg-2020	175	39	to	to	IN
orr-bootleg-2020	175	40	label	label	NNP
orr-bootleg-2020	175	41	mentions	mention	NNS
orr-bootleg-2020	175	42	.	.	.
orr-bootleg-2020	176	1	We	-PRON-	PRP
orr-bootleg-2020	176	2	use	use	VBP
orr-bootleg-2020	176	3	two	two	CD
orr-bootleg-2020	176	4	heuristics	heuristic	NNS
orr-bootleg-2020	176	5	for	for	IN
orr-bootleg-2020	176	6	weak	weak	JJ
orr-bootleg-2020	176	7	labeling	labeling	NN
orr-bootleg-2020	176	8	:	:	:
orr-bootleg-2020	176	9	the	the	DT
orr-bootleg-2020	176	10	first	first	JJ
orr-bootleg-2020	176	11	labels	label	NNS
orr-bootleg-2020	176	12	pronouns	pronoun	NNS
orr-bootleg-2020	176	13	that	that	WDT
orr-bootleg-2020	176	14	match	match	VBP
orr-bootleg-2020	176	15	the	the	DT
orr-bootleg-2020	176	16	gender	gender	NN
orr-bootleg-2020	176	17	of	of	IN
orr-bootleg-2020	176	18	a	a	DT
orr-bootleg-2020	176	19	person	person	NN
orr-bootleg-2020	176	20	’s	’s	POS
orr-bootleg-2020	176	21	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	176	22	page	page	NN
orr-bootleg-2020	176	23	as	as	IN
orr-bootleg-2020	176	24	references	reference	NNS
orr-bootleg-2020	176	25	to	to	IN
orr-bootleg-2020	176	26	that	that	DT
orr-bootleg-2020	176	27	person	person	NN
orr-bootleg-2020	176	28	,	,	,
orr-bootleg-2020	176	29	and	and	CC
orr-bootleg-2020	176	30	the	the	DT
orr-bootleg-2020	176	31	second	second	JJ
orr-bootleg-2020	176	32	labels	label	NNS
orr-bootleg-2020	176	33	known	know	VBD
orr-bootleg-2020	176	34	alternative	alternative	JJ
orr-bootleg-2020	176	35	names	name	NNS
orr-bootleg-2020	176	36	for	for	IN
orr-bootleg-2020	176	37	an	an	DT
orr-bootleg-2020	176	38	entity	entity	NN
orr-bootleg-2020	176	39	if	if	IN
orr-bootleg-2020	176	40	the	the	DT
orr-bootleg-2020	176	41	alternative	alternative	JJ
orr-bootleg-2020	176	42	name	name	NN
orr-bootleg-2020	176	43	appears	appear	VBZ
orr-bootleg-2020	176	44	in	in	IN
orr-bootleg-2020	176	45	sentences	sentence	NNS
orr-bootleg-2020	176	46	on	on	IN
orr-bootleg-2020	176	47	the	the	DT
orr-bootleg-2020	176	48	entity	entity	NN
orr-bootleg-2020	176	49	’s	’s	POS
orr-bootleg-2020	176	50	Wikipedia	wikipedia	NN
orr-bootleg-2020	176	51	page	page	NN
orr-bootleg-2020	176	52	.	.	.
orr-bootleg-2020	177	1	Through	through	IN
orr-bootleg-2020	177	2	weak	weak	JJ
orr-bootleg-2020	177	3	labeling	labeling	NN
orr-bootleg-2020	177	4	,	,	,
orr-bootleg-2020	177	5	we	-PRON-	PRP
orr-bootleg-2020	177	6	increase	increase	VBP
orr-bootleg-2020	177	7	the	the	DT
orr-bootleg-2020	177	8	number	number	NN
orr-bootleg-2020	177	9	of	of	IN
orr-bootleg-2020	177	10	labeled	label	VBN
orr-bootleg-2020	177	11	mentions	mention	NNS
orr-bootleg-2020	177	12	in	in	IN
orr-bootleg-2020	177	13	the	the	DT
orr-bootleg-2020	177	14	training	training	NN
orr-bootleg-2020	177	15	data	datum	NNS
orr-bootleg-2020	177	16	by	by	IN
orr-bootleg-2020	177	17	1.7x	1.7x	CD
orr-bootleg-2020	177	18	across	across	IN
orr-bootleg-2020	177	19	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	177	20	,	,	,
orr-bootleg-2020	177	21	and	and	CC
orr-bootleg-2020	177	22	find	find	VB
orr-bootleg-2020	177	23	this	this	DT
orr-bootleg-2020	177	24	provides	provide	VBZ
orr-bootleg-2020	177	25	a	a	DT
orr-bootleg-2020	177	26	2.6	2.6	CD
orr-bootleg-2020	177	27	F1	f1	NN
orr-bootleg-2020	177	28	lift	lift	NN
orr-bootleg-2020	177	29	on	on	IN
orr-bootleg-2020	177	30	unseen	unseen	JJ
orr-bootleg-2020	177	31	entities	entity	NNS
orr-bootleg-2020	177	32	(	(	-LRB-
orr-bootleg-2020	177	33	full	full	JJ
orr-bootleg-2020	177	34	results	result	NNS
orr-bootleg-2020	177	35	in	in	IN
orr-bootleg-2020	177	36	Appendix	Appendix	NNP
orr-bootleg-2020	177	37	B	B	NNP
orr-bootleg-2020	177	38	Table	Table	NNP
orr-bootleg-2020	177	39	11	11	CD
orr-bootleg-2020	177	40	)	)	-RRB-
orr-bootleg-2020	177	41	.	.	.
orr-bootleg-2020	178	1	4	4	LS
orr-bootleg-2020	178	2	Experiments	experiment	NNS
orr-bootleg-2020	178	3	We	-PRON-	PRP
orr-bootleg-2020	178	4	demonstrate	demonstrate	VBP
orr-bootleg-2020	178	5	that	that	IN
orr-bootleg-2020	178	6	Bootleg	Bootleg	NNP
orr-bootleg-2020	178	7	(	(	-LRB-
orr-bootleg-2020	178	8	1	1	LS
orr-bootleg-2020	178	9	)	)	-RRB-
orr-bootleg-2020	178	10	nearly	nearly	RB
orr-bootleg-2020	178	11	matches	match	VBZ
orr-bootleg-2020	178	12	or	or	CC
orr-bootleg-2020	178	13	exceeds	exceed	VBZ
orr-bootleg-2020	178	14	state	state	NN
orr-bootleg-2020	178	15	-	-	HYPH
orr-bootleg-2020	178	16	of	of	IN
orr-bootleg-2020	178	17	-	-	HYPH
orr-bootleg-2020	178	18	the	the	DT
orr-bootleg-2020	178	19	-	-	HYPH
orr-bootleg-2020	178	20	art	art	NN
orr-bootleg-2020	178	21	performance	performance	NN
orr-bootleg-2020	178	22	on	on	IN
orr-bootleg-2020	178	23	three	three	CD
orr-bootleg-2020	178	24	standard	standard	JJ
orr-bootleg-2020	178	25	NED	NED	NNP
orr-bootleg-2020	178	26	benchmarks	benchmark	NNS
orr-bootleg-2020	178	27	and	and	CC
orr-bootleg-2020	178	28	(	(	-LRB-
orr-bootleg-2020	178	29	2	2	LS
orr-bootleg-2020	178	30	)	)	-RRB-
orr-bootleg-2020	178	31	outperforms	outperform	VBZ
orr-bootleg-2020	178	32	a	a	DT
orr-bootleg-2020	178	33	BERT	BERT	NNP
orr-bootleg-2020	178	34	-	-	HYPH
orr-bootleg-2020	178	35	based	base	VBN
orr-bootleg-2020	178	36	NED	NED	NNP
orr-bootleg-2020	178	37	baseline	baseline	NN
orr-bootleg-2020	178	38	on	on	IN
orr-bootleg-2020	178	39	the	the	DT
orr-bootleg-2020	178	40	tail	tail	NN
orr-bootleg-2020	178	41	.	.	.
orr-bootleg-2020	179	1	As	as	IN
orr-bootleg-2020	179	2	NED	NED	NNP
orr-bootleg-2020	179	3	is	be	VBZ
orr-bootleg-2020	179	4	critical	critical	JJ
orr-bootleg-2020	179	5	for	for	IN
orr-bootleg-2020	179	6	downstream	downstream	JJ
orr-bootleg-2020	179	7	tasks	task	NNS
orr-bootleg-2020	179	8	that	that	WDT
orr-bootleg-2020	179	9	require	require	VBP
orr-bootleg-2020	179	10	the	the	DT
orr-bootleg-2020	179	11	knowledge	knowledge	NN
orr-bootleg-2020	179	12	of	of	IN
orr-bootleg-2020	179	13	entities	entity	NNS
orr-bootleg-2020	179	14	,	,	,
orr-bootleg-2020	179	15	we	-PRON-	PRP
orr-bootleg-2020	179	16	(	(	-LRB-
orr-bootleg-2020	179	17	3	3	LS
orr-bootleg-2020	179	18	)	)	-RRB-
orr-bootleg-2020	179	19	verify	verify	VB
orr-bootleg-2020	179	20	Bootleg	Bootleg	NNP
orr-bootleg-2020	179	21	’s	’s	POS
orr-bootleg-2020	179	22	learned	learn	VBD
orr-bootleg-2020	179	23	reasoning	reasoning	NN
orr-bootleg-2020	179	24	patterns	pattern	NNS
orr-bootleg-2020	179	25	can	can	MD
orr-bootleg-2020	179	26	transfer	transfer	VB
orr-bootleg-2020	179	27	by	by	IN
orr-bootleg-2020	179	28	using	use	VBG
orr-bootleg-2020	179	29	them	-PRON-	PRP
orr-bootleg-2020	179	30	for	for	IN
orr-bootleg-2020	179	31	a	a	DT
orr-bootleg-2020	179	32	downstream	downstream	JJ
orr-bootleg-2020	179	33	task	task	NN
orr-bootleg-2020	179	34	:	:	:
orr-bootleg-2020	179	35	using	use	VBG
orr-bootleg-2020	179	36	Bootleg	Bootleg	NNP
orr-bootleg-2020	179	37	’s	’s	POS
orr-bootleg-2020	179	38	learned	learn	VBD
orr-bootleg-2020	179	39	representations	representation	NNS
orr-bootleg-2020	179	40	,	,	,
orr-bootleg-2020	179	41	we	-PRON-	PRP
orr-bootleg-2020	179	42	achieve	achieve	VBP
orr-bootleg-2020	179	43	a	a	DT
orr-bootleg-2020	179	44	new	new	JJ
orr-bootleg-2020	179	45	SotA	SotA	NNP
orr-bootleg-2020	179	46	on	on	IN
orr-bootleg-2020	179	47	the	the	DT
orr-bootleg-2020	179	48	TACRED	TACRED	NNP
orr-bootleg-2020	179	49	relation	relation	NN
orr-bootleg-2020	179	50	extraction	extraction	NN
orr-bootleg-2020	179	51	task	task	NN
orr-bootleg-2020	179	52	and	and	CC
orr-bootleg-2020	179	53	improve	improve	VB
orr-bootleg-2020	179	54	performance	performance	NN
orr-bootleg-2020	179	55	on	on	IN
orr-bootleg-2020	179	56	a	a	DT
orr-bootleg-2020	179	57	production	production	NN
orr-bootleg-2020	179	58	task	task	NN
orr-bootleg-2020	179	59	at	at	IN
orr-bootleg-2020	179	60	a	a	DT
orr-bootleg-2020	179	61	major	major	JJ
orr-bootleg-2020	179	62	technology	technology	NN
orr-bootleg-2020	179	63	company	company	NN
orr-bootleg-2020	179	64	by	by	IN
orr-bootleg-2020	179	65	8	8	CD
orr-bootleg-2020	179	66	%	%	NN
orr-bootleg-2020	179	67	.	.	.
orr-bootleg-2020	180	1	Finally	finally	RB
orr-bootleg-2020	180	2	,	,	,
orr-bootleg-2020	180	3	we	-PRON-	PRP
orr-bootleg-2020	180	4	(	(	-LRB-
orr-bootleg-2020	180	5	4	4	LS
orr-bootleg-2020	180	6	)	)	-RRB-
orr-bootleg-2020	180	7	demonstrate	demonstrate	VBP
orr-bootleg-2020	180	8	that	that	IN
orr-bootleg-2020	180	9	Bootleg	Bootleg	NNP
orr-bootleg-2020	180	10	can	can	MD
orr-bootleg-2020	180	11	be	be	VB
orr-bootleg-2020	180	12	sample	sample	NN
orr-bootleg-2020	180	13	-	-	HYPH
orr-bootleg-2020	180	14	efficient	efficient	JJ
orr-bootleg-2020	180	15	by	by	IN
orr-bootleg-2020	180	16	using	use	VBG
orr-bootleg-2020	180	17	only	only	RB
orr-bootleg-2020	180	18	a	a	DT
orr-bootleg-2020	180	19	fraction	fraction	NN
orr-bootleg-2020	180	20	of	of	IN
orr-bootleg-2020	180	21	its	-PRON-	PRP$
orr-bootleg-2020	180	22	learned	learned	JJ
orr-bootleg-2020	180	23	entity	entity	NN
orr-bootleg-2020	180	24	embeddings	embedding	NNS
orr-bootleg-2020	180	25	without	without	IN
orr-bootleg-2020	180	26	sacrificing	sacrifice	VBG
orr-bootleg-2020	180	27	performance	performance	NN
orr-bootleg-2020	180	28	.	.	.
orr-bootleg-2020	181	1	We	-PRON-	PRP
orr-bootleg-2020	181	2	(	(	-LRB-
orr-bootleg-2020	181	3	5	5	LS
orr-bootleg-2020	181	4	)	)	-RRB-
orr-bootleg-2020	181	5	ablate	ablate	NN
orr-bootleg-2020	181	6	Bootleg	Bootleg	NNP
orr-bootleg-2020	181	7	to	to	TO
orr-bootleg-2020	181	8	understand	understand	VB
orr-bootleg-2020	181	9	the	the	DT
orr-bootleg-2020	181	10	impact	impact	NN
orr-bootleg-2020	181	11	of	of	IN
orr-bootleg-2020	181	12	the	the	DT
orr-bootleg-2020	181	13	structural	structural	JJ
orr-bootleg-2020	181	14	signals	signal	NNS
orr-bootleg-2020	181	15	and	and	CC
orr-bootleg-2020	181	16	the	the	DT
orr-bootleg-2020	181	17	regularization	regularization	NN
orr-bootleg-2020	181	18	scheme	scheme	NN
orr-bootleg-2020	181	19	on	on	IN
orr-bootleg-2020	181	20	improved	improved	JJ
orr-bootleg-2020	181	21	tail	tail	NN
orr-bootleg-2020	181	22	performance	performance	NN
orr-bootleg-2020	181	23	.	.	.
orr-bootleg-2020	182	1	4.1	4.1	CD
orr-bootleg-2020	182	2	Experimental	Experimental	NNP
orr-bootleg-2020	182	3	Setup	Setup	NNP
orr-bootleg-2020	182	4	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	182	5	Data	Data	NNP
orr-bootleg-2020	182	6	We	-PRON-	PRP
orr-bootleg-2020	182	7	define	define	VBP
orr-bootleg-2020	182	8	our	-PRON-	PRP$
orr-bootleg-2020	182	9	knowledge	knowledge	NN
orr-bootleg-2020	182	10	base	base	NN
orr-bootleg-2020	182	11	as	as	IN
orr-bootleg-2020	182	12	the	the	DT
orr-bootleg-2020	182	13	set	set	NN
orr-bootleg-2020	182	14	of	of	IN
orr-bootleg-2020	182	15	entities	entity	NNS
orr-bootleg-2020	182	16	with	with	IN
orr-bootleg-2020	182	17	mentions	mention	NNS
orr-bootleg-2020	182	18	in	in	IN
orr-bootleg-2020	182	19	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	182	20	(	(	-LRB-
orr-bootleg-2020	182	21	for	for	IN
orr-bootleg-2020	182	22	a	a	DT
orr-bootleg-2020	182	23	total	total	NN
orr-bootleg-2020	182	24	of	of	IN
orr-bootleg-2020	182	25	5.3	5.3	CD
orr-bootleg-2020	182	26	M	M	NNP
orr-bootleg-2020	182	27	entities	entity	NNS
orr-bootleg-2020	182	28	)	)	-RRB-
orr-bootleg-2020	182	29	.	.	.
orr-bootleg-2020	183	1	We	-PRON-	PRP
orr-bootleg-2020	183	2	allow	allow	VBP
orr-bootleg-2020	183	3	each	each	DT
orr-bootleg-2020	183	4	mention	mention	NN
orr-bootleg-2020	183	5	to	to	TO
orr-bootleg-2020	183	6	have	have	VB
orr-bootleg-2020	183	7	up	up	IN
orr-bootleg-2020	183	8	to	to	IN
orr-bootleg-2020	183	9	K	k	NN
orr-bootleg-2020	183	10	=	=	SYM
orr-bootleg-2020	183	11	30	30	CD
orr-bootleg-2020	183	12	possible	possible	JJ
orr-bootleg-2020	183	13	candidates	candidate	NNS
orr-bootleg-2020	183	14	.	.	.
orr-bootleg-2020	184	1	As	as	IN
orr-bootleg-2020	184	2	Bootleg	Bootleg	NNP
orr-bootleg-2020	184	3	is	be	VBZ
orr-bootleg-2020	184	4	a	a	DT
orr-bootleg-2020	184	5	sentence	sentence	NN
orr-bootleg-2020	184	6	disambiguation	disambiguation	NN
orr-bootleg-2020	184	7	system	system	NN
orr-bootleg-2020	184	8	,	,	,
orr-bootleg-2020	184	9	we	-PRON-	PRP
orr-bootleg-2020	184	10	train	train	VBP
orr-bootleg-2020	184	11	on	on	IN
orr-bootleg-2020	184	12	individual	individual	JJ
orr-bootleg-2020	184	13	sentences	sentence	NNS
orr-bootleg-2020	184	14	from	from	IN
orr-bootleg-2020	184	15	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	184	16	,	,	,
orr-bootleg-2020	184	17	where	where	WRB
orr-bootleg-2020	184	18	the	the	DT
orr-bootleg-2020	184	19	anchor	anchor	NN
orr-bootleg-2020	184	20	links	link	NNS
orr-bootleg-2020	184	21	and	and	CC
orr-bootleg-2020	184	22	our	-PRON-	PRP$
orr-bootleg-2020	184	23	weak	weak	JJ
orr-bootleg-2020	184	24	labeling	labeling	NN
orr-bootleg-2020	184	25	(	(	-LRB-
orr-bootleg-2020	184	26	Section	section	NN
orr-bootleg-2020	184	27	3.3	3.3	CD
orr-bootleg-2020	184	28	)	)	-RRB-
orr-bootleg-2020	184	29	serve	serve	VBP
orr-bootleg-2020	184	30	as	as	IN
orr-bootleg-2020	184	31	mention	mention	NN
orr-bootleg-2020	184	32	labels	label	NNS
orr-bootleg-2020	184	33	.	.	.
orr-bootleg-2020	185	1	7	7	CD
orr-bootleg-2020	185	2	Table	table	NN
orr-bootleg-2020	185	3	1	1	CD
orr-bootleg-2020	185	4	:	:	:
orr-bootleg-2020	185	5	We	-PRON-	PRP
orr-bootleg-2020	185	6	compare	compare	VBP
orr-bootleg-2020	185	7	Bootleg	Bootleg	NNP
orr-bootleg-2020	185	8	to	to	IN
orr-bootleg-2020	185	9	the	the	DT
orr-bootleg-2020	185	10	best	well	RBS
orr-bootleg-2020	185	11	published	publish	VBN
orr-bootleg-2020	185	12	numbers	number	NNS
orr-bootleg-2020	185	13	on	on	IN
orr-bootleg-2020	185	14	three	three	CD
orr-bootleg-2020	185	15	NED	NED	NNP
orr-bootleg-2020	185	16	benchmarks	benchmark	NNS
orr-bootleg-2020	185	17	.	.	.
orr-bootleg-2020	186	1	“	"	``
orr-bootleg-2020	186	2	-	-	:
orr-bootleg-2020	186	3	”	"	''
orr-bootleg-2020	186	4	indicates	indicate	VBZ
orr-bootleg-2020	186	5	that	that	IN
orr-bootleg-2020	186	6	the	the	DT
orr-bootleg-2020	186	7	metric	metric	NN
orr-bootleg-2020	186	8	was	be	VBD
orr-bootleg-2020	186	9	not	not	RB
orr-bootleg-2020	186	10	reported	report	VBN
orr-bootleg-2020	186	11	.	.	.
orr-bootleg-2020	187	1	Bolded	bolded	JJ
orr-bootleg-2020	187	2	numbers	number	NNS
orr-bootleg-2020	187	3	indicate	indicate	VBP
orr-bootleg-2020	187	4	the	the	DT
orr-bootleg-2020	187	5	best	good	JJS
orr-bootleg-2020	187	6	value	value	NN
orr-bootleg-2020	187	7	for	for	IN
orr-bootleg-2020	187	8	each	each	DT
orr-bootleg-2020	187	9	metric	metric	JJ
orr-bootleg-2020	187	10	on	on	IN
orr-bootleg-2020	187	11	each	each	DT
orr-bootleg-2020	187	12	benchmark	benchmark	NN
orr-bootleg-2020	187	13	.	.	.
orr-bootleg-2020	188	1	Benchmark	Benchmark	NNP
orr-bootleg-2020	188	2	Model	Model	NNP
orr-bootleg-2020	188	3	Precision	Precision	NNP
orr-bootleg-2020	188	4	Recall	Recall	NNP
orr-bootleg-2020	188	5	F1	f1	NN
orr-bootleg-2020	188	6	KORE50	KORE50	NNP
orr-bootleg-2020	188	7	Hu	Hu	NNP
orr-bootleg-2020	188	8	et	et	FW
orr-bootleg-2020	188	9	al	al	NNP
orr-bootleg-2020	188	10	.	.	.
orr-bootleg-2020	189	1	[	[	-LRB-
orr-bootleg-2020	189	2	24]7	24]7	CD
orr-bootleg-2020	189	3	80.0	80.0	CD
orr-bootleg-2020	189	4	79.8	79.8	CD
orr-bootleg-2020	189	5	79.9	79.9	CD
orr-bootleg-2020	189	6	Bootleg	Bootleg	NNP
orr-bootleg-2020	189	7	86.0	86.0	CD
orr-bootleg-2020	189	8	85.4	85.4	CD
orr-bootleg-2020	189	9	85.7	85.7	CD
orr-bootleg-2020	189	10	RSS500	RSS500	NNP
orr-bootleg-2020	189	11	Phan	Phan	NNP
orr-bootleg-2020	189	12	et	et	NNP
orr-bootleg-2020	189	13	al	al	NNP
orr-bootleg-2020	189	14	.	.	.
orr-bootleg-2020	190	1	[	[	-LRB-
orr-bootleg-2020	190	2	40	40	CD
orr-bootleg-2020	190	3	]	]	-RRB-
orr-bootleg-2020	190	4	82.3	82.3	CD
orr-bootleg-2020	190	5	82.3	82.3	CD
orr-bootleg-2020	190	6	82.3	82.3	CD
orr-bootleg-2020	190	7	Bootleg	Bootleg	NNP
orr-bootleg-2020	190	8	82.5	82.5	CD
orr-bootleg-2020	190	9	82.5	82.5	CD
orr-bootleg-2020	190	10	82.5	82.5	CD
orr-bootleg-2020	190	11	AIDA	AIDA	NNP
orr-bootleg-2020	190	12	Févry	Févry	NNP
orr-bootleg-2020	190	13	et	et	NNP
orr-bootleg-2020	190	14	al	al	NNP
orr-bootleg-2020	190	15	.	.	.
orr-bootleg-2020	191	1	[	[	-LRB-
orr-bootleg-2020	191	2	16	16	CD
orr-bootleg-2020	191	3	]	]	-RRB-
orr-bootleg-2020	191	4	-	-	HYPH
orr-bootleg-2020	191	5	96.7	96.7	CD
orr-bootleg-2020	191	6	-	-	HYPH
orr-bootleg-2020	191	7	Bootleg	bootleg	NN
orr-bootleg-2020	191	8	96.9	96.9	CD
orr-bootleg-2020	191	9	96.7	96.7	CD
orr-bootleg-2020	191	10	96.8	96.8	CD
orr-bootleg-2020	191	11	Our	-PRON-	PRP$
orr-bootleg-2020	191	12	candidate	candidate	NN
orr-bootleg-2020	191	13	lists	list	NNS
orr-bootleg-2020	191	14	Γ	Γ	NNP
orr-bootleg-2020	191	15	are	be	VBP
orr-bootleg-2020	191	16	mined	mine	VBN
orr-bootleg-2020	191	17	from	from	IN
orr-bootleg-2020	191	18	Wikipedia	wikipedia	JJ
orr-bootleg-2020	191	19	anchor	anchor	NN
orr-bootleg-2020	191	20	links	link	NNS
orr-bootleg-2020	191	21	and	and	CC
orr-bootleg-2020	191	22	the	the	DT
orr-bootleg-2020	191	23	“	"	``
orr-bootleg-2020	191	24	also	also	RB
orr-bootleg-2020	191	25	known	know	VBN
orr-bootleg-2020	191	26	as	as	IN
orr-bootleg-2020	191	27	”	"	``
orr-bootleg-2020	191	28	field	field	NN
orr-bootleg-2020	191	29	in	in	IN
orr-bootleg-2020	191	30	Wikidata	Wikidata	NNP
orr-bootleg-2020	191	31	.	.	.
orr-bootleg-2020	192	1	For	for	IN
orr-bootleg-2020	192	2	each	each	DT
orr-bootleg-2020	192	3	person	person	NN
orr-bootleg-2020	192	4	,	,	,
orr-bootleg-2020	192	5	we	-PRON-	PRP
orr-bootleg-2020	192	6	further	further	RB
orr-bootleg-2020	192	7	add	add	VBP
orr-bootleg-2020	192	8	their	-PRON-	PRP$
orr-bootleg-2020	192	9	first	first	JJ
orr-bootleg-2020	192	10	and	and	CC
orr-bootleg-2020	192	11	last	last	JJ
orr-bootleg-2020	192	12	name	name	NN
orr-bootleg-2020	192	13	as	as	IN
orr-bootleg-2020	192	14	aliases	alias	NNS
orr-bootleg-2020	192	15	linking	link	VBG
orr-bootleg-2020	192	16	to	to	IN
orr-bootleg-2020	192	17	that	that	DT
orr-bootleg-2020	192	18	person	person	NN
orr-bootleg-2020	192	19	.	.	.
orr-bootleg-2020	193	1	We	-PRON-	PRP
orr-bootleg-2020	193	2	use	use	VBP
orr-bootleg-2020	193	3	the	the	DT
orr-bootleg-2020	193	4	mention	mention	NN
orr-bootleg-2020	193	5	boundaries	boundary	NNS
orr-bootleg-2020	193	6	provided	provide	VBN
orr-bootleg-2020	193	7	in	in	IN
orr-bootleg-2020	193	8	the	the	DT
orr-bootleg-2020	193	9	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	193	10	data	datum	NNS
orr-bootleg-2020	193	11	and	and	CC
orr-bootleg-2020	193	12	generate	generate	VB
orr-bootleg-2020	193	13	candidates	candidate	NNS
orr-bootleg-2020	193	14	by	by	IN
orr-bootleg-2020	193	15	performing	perform	VBG
orr-bootleg-2020	193	16	a	a	DT
orr-bootleg-2020	193	17	direct	direct	JJ
orr-bootleg-2020	193	18	lookup	lookup	NN
orr-bootleg-2020	193	19	in	in	IN
orr-bootleg-2020	193	20	Γ.	Γ.	NNP
orr-bootleg-2020	194	1	We	-PRON-	PRP
orr-bootleg-2020	194	2	use	use	VBP
orr-bootleg-2020	194	3	Wikidata	wikidata	NN
orr-bootleg-2020	194	4	and	and	CC
orr-bootleg-2020	194	5	YAGO	yago	JJ
orr-bootleg-2020	194	6	knowledge	knowledge	NN
orr-bootleg-2020	194	7	graphs	graphs	NN
orr-bootleg-2020	194	8	and	and	CC
orr-bootleg-2020	194	9	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	194	10	to	to	TO
orr-bootleg-2020	194	11	extract	extract	VB
orr-bootleg-2020	194	12	structural	structural	JJ
orr-bootleg-2020	194	13	data	datum	NNS
orr-bootleg-2020	194	14	about	about	IN
orr-bootleg-2020	194	15	entity	entity	NN
orr-bootleg-2020	194	16	types	type	NNS
orr-bootleg-2020	194	17	and	and	CC
orr-bootleg-2020	194	18	relations	relation	NNS
orr-bootleg-2020	194	19	as	as	IN
orr-bootleg-2020	194	20	input	input	NN
orr-bootleg-2020	194	21	for	for	IN
orr-bootleg-2020	194	22	Bootleg	Bootleg	NNP
orr-bootleg-2020	194	23	.	.	.
orr-bootleg-2020	195	1	Further	further	JJ
orr-bootleg-2020	195	2	details	detail	NNS
orr-bootleg-2020	195	3	about	about	IN
orr-bootleg-2020	195	4	data	datum	NNS
orr-bootleg-2020	195	5	are	be	VBP
orr-bootleg-2020	195	6	in	in	IN
orr-bootleg-2020	195	7	Appendix	Appendix	NNP
orr-bootleg-2020	195	8	B.	B.	NNP
orr-bootleg-2020	195	9	Metrics	Metrics	NNP
orr-bootleg-2020	195	10	We	-PRON-	PRP
orr-bootleg-2020	195	11	report	report	VBP
orr-bootleg-2020	195	12	micro	micro	JJ
orr-bootleg-2020	195	13	-	-	JJ
orr-bootleg-2020	195	14	average	average	JJ
orr-bootleg-2020	195	15	F1	f1	NN
orr-bootleg-2020	195	16	scores	score	NNS
orr-bootleg-2020	195	17	for	for	IN
orr-bootleg-2020	195	18	all	all	DT
orr-bootleg-2020	195	19	metrics	metric	NNS
orr-bootleg-2020	195	20	over	over	IN
orr-bootleg-2020	195	21	true	true	JJ
orr-bootleg-2020	195	22	anchor	anchor	NN
orr-bootleg-2020	195	23	links	link	NNS
orr-bootleg-2020	195	24	in	in	IN
orr-bootleg-2020	195	25	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	195	26	(	(	-LRB-
orr-bootleg-2020	195	27	not	not	RB
orr-bootleg-2020	195	28	weak	weak	JJ
orr-bootleg-2020	195	29	labels	label	NNS
orr-bootleg-2020	195	30	)	)	-RRB-
orr-bootleg-2020	195	31	.	.	.
orr-bootleg-2020	196	1	We	-PRON-	PRP
orr-bootleg-2020	196	2	measure	measure	VBP
orr-bootleg-2020	196	3	the	the	DT
orr-bootleg-2020	196	4	torso	torso	NN
orr-bootleg-2020	196	5	and	and	CC
orr-bootleg-2020	196	6	tail	tail	NN
orr-bootleg-2020	196	7	sets	set	NNS
orr-bootleg-2020	196	8	based	base	VBN
orr-bootleg-2020	196	9	on	on	IN
orr-bootleg-2020	196	10	the	the	DT
orr-bootleg-2020	196	11	number	number	NN
orr-bootleg-2020	196	12	of	of	IN
orr-bootleg-2020	196	13	times	time	NNS
orr-bootleg-2020	196	14	that	that	IN
orr-bootleg-2020	196	15	an	an	DT
orr-bootleg-2020	196	16	entity	entity	NN
orr-bootleg-2020	196	17	is	be	VBZ
orr-bootleg-2020	196	18	the	the	DT
orr-bootleg-2020	196	19	gold	gold	JJ
orr-bootleg-2020	196	20	entity	entity	NN
orr-bootleg-2020	196	21	across	across	IN
orr-bootleg-2020	196	22	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	196	23	anchors	anchor	NNS
orr-bootleg-2020	196	24	and	and	CC
orr-bootleg-2020	196	25	weak	weak	JJ
orr-bootleg-2020	196	26	labels	label	NNS
orr-bootleg-2020	196	27	,	,	,
orr-bootleg-2020	196	28	as	as	IN
orr-bootleg-2020	196	29	this	this	DT
orr-bootleg-2020	196	30	represents	represent	VBZ
orr-bootleg-2020	196	31	the	the	DT
orr-bootleg-2020	196	32	number	number	NN
orr-bootleg-2020	196	33	of	of	IN
orr-bootleg-2020	196	34	times	time	NNS
orr-bootleg-2020	196	35	an	an	DT
orr-bootleg-2020	196	36	entity	entity	NN
orr-bootleg-2020	196	37	is	be	VBZ
orr-bootleg-2020	196	38	seen	see	VBN
orr-bootleg-2020	196	39	by	by	IN
orr-bootleg-2020	196	40	Bootleg	Bootleg	NNP
orr-bootleg-2020	196	41	.	.	.
orr-bootleg-2020	197	1	For	for	IN
orr-bootleg-2020	197	2	benchmarks	benchmark	NNS
orr-bootleg-2020	197	3	,	,	,
orr-bootleg-2020	197	4	we	-PRON-	PRP
orr-bootleg-2020	197	5	also	also	RB
orr-bootleg-2020	197	6	report	report	VBP
orr-bootleg-2020	197	7	precision	precision	NN
orr-bootleg-2020	197	8	and	and	CC
orr-bootleg-2020	197	9	recall	recall	NN
orr-bootleg-2020	197	10	using	use	VBG
orr-bootleg-2020	197	11	the	the	DT
orr-bootleg-2020	197	12	number	number	NN
orr-bootleg-2020	197	13	of	of	IN
orr-bootleg-2020	197	14	mentions	mention	NNS
orr-bootleg-2020	197	15	extracted	extract	VBN
orr-bootleg-2020	197	16	by	by	IN
orr-bootleg-2020	197	17	Bootleg	Bootleg	NNP
orr-bootleg-2020	197	18	and	and	CC
orr-bootleg-2020	197	19	the	the	DT
orr-bootleg-2020	197	20	number	number	NN
orr-bootleg-2020	197	21	of	of	IN
orr-bootleg-2020	197	22	mentions	mention	NNS
orr-bootleg-2020	197	23	defined	define	VBN
orr-bootleg-2020	197	24	in	in	IN
orr-bootleg-2020	197	25	the	the	DT
orr-bootleg-2020	197	26	data	datum	NNS
orr-bootleg-2020	197	27	as	as	IN
orr-bootleg-2020	197	28	denominators	denominator	NNS
orr-bootleg-2020	197	29	,	,	,
orr-bootleg-2020	197	30	respectively	respectively	RB
orr-bootleg-2020	197	31	.	.	.
orr-bootleg-2020	198	1	The	the	DT
orr-bootleg-2020	198	2	numerator	numerator	NN
orr-bootleg-2020	198	3	is	be	VBZ
orr-bootleg-2020	198	4	the	the	DT
orr-bootleg-2020	198	5	number	number	NN
orr-bootleg-2020	198	6	of	of	IN
orr-bootleg-2020	198	7	correctly	correctly	RB
orr-bootleg-2020	198	8	disambiguated	disambiguate	VBN
orr-bootleg-2020	198	9	mentions	mention	NNS
orr-bootleg-2020	198	10	.	.	.
orr-bootleg-2020	199	1	For	for	IN
orr-bootleg-2020	199	2	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	199	3	data	datum	NNS
orr-bootleg-2020	199	4	experiments	experiment	NNS
orr-bootleg-2020	199	5	,	,	,
orr-bootleg-2020	199	6	we	-PRON-	PRP
orr-bootleg-2020	199	7	filter	filter	VBP
orr-bootleg-2020	199	8	mentions	mention	NNS
orr-bootleg-2020	199	9	such	such	JJ
orr-bootleg-2020	199	10	that	that	IN
orr-bootleg-2020	199	11	(	(	-LRB-
orr-bootleg-2020	199	12	a	a	LS
orr-bootleg-2020	199	13	)	)	-RRB-
orr-bootleg-2020	199	14	the	the	DT
orr-bootleg-2020	199	15	gold	gold	NN
orr-bootleg-2020	199	16	entity	entity	NN
orr-bootleg-2020	199	17	is	be	VBZ
orr-bootleg-2020	199	18	in	in	IN
orr-bootleg-2020	199	19	the	the	DT
orr-bootleg-2020	199	20	candidate	candidate	NN
orr-bootleg-2020	199	21	set	set	VBD
orr-bootleg-2020	199	22	and	and	CC
orr-bootleg-2020	199	23	(	(	-LRB-
orr-bootleg-2020	199	24	b	b	NN
orr-bootleg-2020	199	25	)	)	-RRB-
orr-bootleg-2020	199	26	they	-PRON-	PRP
orr-bootleg-2020	199	27	have	have	VBP
orr-bootleg-2020	199	28	more	more	JJR
orr-bootleg-2020	199	29	than	than	IN
orr-bootleg-2020	199	30	one	one	CD
orr-bootleg-2020	199	31	possible	possible	JJ
orr-bootleg-2020	199	32	candidate	candidate	NN
orr-bootleg-2020	199	33	.	.	.
orr-bootleg-2020	200	1	The	the	DT
orr-bootleg-2020	200	2	former	former	JJ
orr-bootleg-2020	200	3	is	be	VBZ
orr-bootleg-2020	200	4	to	to	TO
orr-bootleg-2020	200	5	decouple	decouple	VB
orr-bootleg-2020	200	6	candidate	candidate	NN
orr-bootleg-2020	200	7	generation	generation	NN
orr-bootleg-2020	200	8	from	from	IN
orr-bootleg-2020	200	9	model	model	NN
orr-bootleg-2020	200	10	performance	performance	NN
orr-bootleg-2020	200	11	for	for	IN
orr-bootleg-2020	200	12	ablations.6	ablations.6	NFP
orr-bootleg-2020	200	13	The	the	DT
orr-bootleg-2020	200	14	latter	latter	NN
orr-bootleg-2020	200	15	is	be	VBZ
orr-bootleg-2020	200	16	to	to	TO
orr-bootleg-2020	200	17	not	not	RB
orr-bootleg-2020	200	18	inflate	inflate	VB
orr-bootleg-2020	200	19	a	a	DT
orr-bootleg-2020	200	20	model	model	NN
orr-bootleg-2020	200	21	’s	’s	POS
orr-bootleg-2020	200	22	performance	performance	NN
orr-bootleg-2020	200	23	,	,	,
orr-bootleg-2020	200	24	as	as	IN
orr-bootleg-2020	200	25	all	all	DT
orr-bootleg-2020	200	26	models	model	NNS
orr-bootleg-2020	200	27	are	be	VBP
orr-bootleg-2020	200	28	trivially	trivially	RB
orr-bootleg-2020	200	29	correct	correct	JJ
orr-bootleg-2020	200	30	when	when	WRB
orr-bootleg-2020	200	31	there	there	EX
orr-bootleg-2020	200	32	is	be	VBZ
orr-bootleg-2020	200	33	a	a	DT
orr-bootleg-2020	200	34	single	single	JJ
orr-bootleg-2020	200	35	candidate	candidate	NN
orr-bootleg-2020	200	36	.	.	.
orr-bootleg-2020	201	1	Training	training	NN
orr-bootleg-2020	201	2	For	for	IN
orr-bootleg-2020	201	3	our	-PRON-	PRP$
orr-bootleg-2020	201	4	main	main	JJ
orr-bootleg-2020	201	5	Bootleg	Bootleg	NNP
orr-bootleg-2020	201	6	model	model	NN
orr-bootleg-2020	201	7	,	,	,
orr-bootleg-2020	201	8	we	-PRON-	PRP
orr-bootleg-2020	201	9	train	train	VBP
orr-bootleg-2020	201	10	for	for	IN
orr-bootleg-2020	201	11	two	two	CD
orr-bootleg-2020	201	12	epochs	epoch	NNS
orr-bootleg-2020	201	13	on	on	IN
orr-bootleg-2020	201	14	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	201	15	sentences	sentence	NNS
orr-bootleg-2020	201	16	with	with	IN
orr-bootleg-2020	201	17	a	a	DT
orr-bootleg-2020	201	18	maximum	maximum	JJ
orr-bootleg-2020	201	19	sentence	sentence	NN
orr-bootleg-2020	201	20	length	length	NN
orr-bootleg-2020	201	21	of	of	IN
orr-bootleg-2020	201	22	100	100	CD
orr-bootleg-2020	201	23	.	.	.
orr-bootleg-2020	202	1	For	for	IN
orr-bootleg-2020	202	2	our	-PRON-	PRP$
orr-bootleg-2020	202	3	benchmark	benchmark	NN
orr-bootleg-2020	202	4	model	model	NN
orr-bootleg-2020	202	5	,	,	,
orr-bootleg-2020	202	6	we	-PRON-	PRP
orr-bootleg-2020	202	7	train	train	VBP
orr-bootleg-2020	202	8	for	for	IN
orr-bootleg-2020	202	9	one	one	CD
orr-bootleg-2020	202	10	epoch	epoch	NN
orr-bootleg-2020	202	11	and	and	CC
orr-bootleg-2020	202	12	additionally	additionally	RB
orr-bootleg-2020	202	13	add	add	VB
orr-bootleg-2020	202	14	a	a	DT
orr-bootleg-2020	202	15	title	title	NN
orr-bootleg-2020	202	16	embedding	embed	VBG
orr-bootleg-2020	202	17	feature	feature	NN
orr-bootleg-2020	202	18	,	,	,
orr-bootleg-2020	202	19	a	a	DT
orr-bootleg-2020	202	20	sentence	sentence	NN
orr-bootleg-2020	202	21	co	co	NN
orr-bootleg-2020	202	22	-	-	JJ
orr-bootleg-2020	202	23	occurrence	occurrence	JJ
orr-bootleg-2020	202	24	KG	kg	NN
orr-bootleg-2020	202	25	matrix	matrix	NN
orr-bootleg-2020	202	26	as	as	IN
orr-bootleg-2020	202	27	another	another	DT
orr-bootleg-2020	202	28	KG	KG	NNP
orr-bootleg-2020	202	29	module	module	NN
orr-bootleg-2020	202	30	,	,	,
orr-bootleg-2020	202	31	and	and	CC
orr-bootleg-2020	202	32	a	a	DT
orr-bootleg-2020	202	33	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	202	34	page	page	NN
orr-bootleg-2020	202	35	co	co	NN
orr-bootleg-2020	202	36	-	-	JJ
orr-bootleg-2020	202	37	occurrence	occurrence	JJ
orr-bootleg-2020	202	38	statistical	statistical	JJ
orr-bootleg-2020	202	39	feature	feature	NN
orr-bootleg-2020	202	40	.	.	.
orr-bootleg-2020	203	1	Additional	additional	JJ
orr-bootleg-2020	203	2	details	detail	NNS
orr-bootleg-2020	203	3	about	about	IN
orr-bootleg-2020	203	4	the	the	DT
orr-bootleg-2020	203	5	models	model	NNS
orr-bootleg-2020	203	6	and	and	CC
orr-bootleg-2020	203	7	training	training	NN
orr-bootleg-2020	203	8	procedure	procedure	NN
orr-bootleg-2020	203	9	are	be	VBP
orr-bootleg-2020	203	10	in	in	IN
orr-bootleg-2020	203	11	Appendix	Appendix	NNP
orr-bootleg-2020	203	12	B.	B.	NNP
orr-bootleg-2020	204	1	4.2	4.2	CD
orr-bootleg-2020	204	2	Bootleg	Bootleg	NNP
orr-bootleg-2020	204	3	Performance	Performance	NNP
orr-bootleg-2020	204	4	Benchmark	Benchmark	NNP
orr-bootleg-2020	204	5	Performance	Performance	NNP
orr-bootleg-2020	204	6	To	to	TO
orr-bootleg-2020	204	7	understand	understand	VB
orr-bootleg-2020	204	8	the	the	DT
orr-bootleg-2020	204	9	overall	overall	JJ
orr-bootleg-2020	204	10	performance	performance	NN
orr-bootleg-2020	204	11	of	of	IN
orr-bootleg-2020	204	12	Bootleg	Bootleg	NNP
orr-bootleg-2020	204	13	,	,	,
orr-bootleg-2020	204	14	we	-PRON-	PRP
orr-bootleg-2020	204	15	compare	compare	VBP
orr-bootleg-2020	204	16	against	against	IN
orr-bootleg-2020	204	17	reported	report	VBN
orr-bootleg-2020	204	18	state	state	NN
orr-bootleg-2020	204	19	-	-	HYPH
orr-bootleg-2020	204	20	of	of	IN
orr-bootleg-2020	204	21	-	-	HYPH
orr-bootleg-2020	204	22	the	the	DT
orr-bootleg-2020	204	23	-	-	HYPH
orr-bootleg-2020	204	24	art	art	NN
orr-bootleg-2020	204	25	numbers	number	NNS
orr-bootleg-2020	204	26	of	of	IN
orr-bootleg-2020	204	27	two	two	CD
orr-bootleg-2020	204	28	standard	standard	JJ
orr-bootleg-2020	204	29	sentence	sentence	NN
orr-bootleg-2020	204	30	benchmarks	benchmark	NNS
orr-bootleg-2020	204	31	(	(	-LRB-
orr-bootleg-2020	204	32	KORE50	KORE50	NNP
orr-bootleg-2020	204	33	,	,	,
orr-bootleg-2020	204	34	RSS500	RSS500	NNP
orr-bootleg-2020	204	35	)	)	-RRB-
orr-bootleg-2020	204	36	and	and	CC
orr-bootleg-2020	204	37	the	the	DT
orr-bootleg-2020	204	38	standard	standard	JJ
orr-bootleg-2020	204	39	document	document	NN
orr-bootleg-2020	204	40	benchmark	benchmark	NN
orr-bootleg-2020	204	41	(	(	-LRB-
orr-bootleg-2020	204	42	AIDA	AIDA	NNP
orr-bootleg-2020	204	43	CoNLL	CoNLL	NNP
orr-bootleg-2020	204	44	-	-	HYPH
orr-bootleg-2020	204	45	YAGO	YAGO	NNP
orr-bootleg-2020	204	46	)	)	-RRB-
orr-bootleg-2020	204	47	.	.	.
orr-bootleg-2020	205	1	Benchmark	benchmark	JJ
orr-bootleg-2020	205	2	details	detail	NNS
orr-bootleg-2020	205	3	are	be	VBP
orr-bootleg-2020	205	4	in	in	IN
orr-bootleg-2020	205	5	Appendix	Appendix	NNP
orr-bootleg-2020	205	6	B.	B.	NNP
orr-bootleg-2020	206	1	For	for	IN
orr-bootleg-2020	206	2	AIDA	AIDA	NNP
orr-bootleg-2020	206	3	,	,	,
orr-bootleg-2020	206	4	we	-PRON-	PRP
orr-bootleg-2020	206	5	first	first	RB
orr-bootleg-2020	206	6	convert	convert	VBP
orr-bootleg-2020	206	7	each	each	DT
orr-bootleg-2020	206	8	document	document	NN
orr-bootleg-2020	206	9	into	into	IN
orr-bootleg-2020	206	10	a	a	DT
orr-bootleg-2020	206	11	set	set	NN
orr-bootleg-2020	206	12	of	of	IN
orr-bootleg-2020	206	13	sentences	sentence	NNS
orr-bootleg-2020	206	14	where	where	WRB
orr-bootleg-2020	206	15	a	a	DT
orr-bootleg-2020	206	16	sentence	sentence	NN
orr-bootleg-2020	206	17	is	be	VBZ
orr-bootleg-2020	206	18	the	the	DT
orr-bootleg-2020	206	19	document	document	NN
orr-bootleg-2020	206	20	title	title	NN
orr-bootleg-2020	206	21	,	,	,
orr-bootleg-2020	206	22	a	a	DT
orr-bootleg-2020	206	23	BERT	BERT	NNP
orr-bootleg-2020	206	24	SEP	SEP	NNP
orr-bootleg-2020	206	25	token	token	NN
orr-bootleg-2020	206	26	,	,	,
orr-bootleg-2020	206	27	and	and	CC
orr-bootleg-2020	206	28	the	the	DT
orr-bootleg-2020	206	29	sentence	sentence	NN
orr-bootleg-2020	206	30	.	.	.
orr-bootleg-2020	207	1	We	-PRON-	PRP
orr-bootleg-2020	207	2	find	find	VBP
orr-bootleg-2020	207	3	this	this	DT
orr-bootleg-2020	207	4	is	be	VBZ
orr-bootleg-2020	207	5	sufficient	sufficient	JJ
orr-bootleg-2020	207	6	to	to	TO
orr-bootleg-2020	207	7	encode	encode	VB
orr-bootleg-2020	207	8	document	document	NN
orr-bootleg-2020	207	9	context	context	NN
orr-bootleg-2020	207	10	into	into	IN
orr-bootleg-2020	207	11	Bootleg	Bootleg	NNP
orr-bootleg-2020	207	12	.	.	.
orr-bootleg-2020	208	1	We	-PRON-	PRP
orr-bootleg-2020	208	2	fine	fine	JJ
orr-bootleg-2020	208	3	-	-	HYPH
orr-bootleg-2020	208	4	tune	tune	NN
orr-bootleg-2020	208	5	the	the	DT
orr-bootleg-2020	208	6	pretrained	pretrained	JJ
orr-bootleg-2020	208	7	Bootleg	Bootleg	NNP
orr-bootleg-2020	208	8	model	model	NN
orr-bootleg-2020	208	9	on	on	IN
orr-bootleg-2020	208	10	the	the	DT
orr-bootleg-2020	208	11	AIDA	AIDA	NNP
orr-bootleg-2020	208	12	training	training	NN
orr-bootleg-2020	208	13	set	set	VBN
orr-bootleg-2020	208	14	with	with	IN
orr-bootleg-2020	208	15	learning	learning	NN
orr-bootleg-2020	208	16	rate	rate	NN
orr-bootleg-2020	208	17	of	of	IN
orr-bootleg-2020	208	18	0.00007	0.00007	CD
orr-bootleg-2020	208	19	,	,	,
orr-bootleg-2020	208	20	2	2	CD
orr-bootleg-2020	208	21	epochs	epoch	NNS
orr-bootleg-2020	208	22	,	,	,
orr-bootleg-2020	208	23	batch	batch	NN
orr-bootleg-2020	208	24	size	size	NN
orr-bootleg-2020	208	25	of	of	IN
orr-bootleg-2020	208	26	16	16	CD
orr-bootleg-2020	208	27	,	,	,
orr-bootleg-2020	208	28	and	and	CC
orr-bootleg-2020	208	29	evaluating	evaluate	VBG
orr-bootleg-2020	208	30	every	every	DT
orr-bootleg-2020	208	31	25	25	CD
orr-bootleg-2020	208	32	steps	step	NNS
orr-bootleg-2020	208	33	.	.	.
orr-bootleg-2020	209	1	We	-PRON-	PRP
orr-bootleg-2020	209	2	choose	choose	VBP
orr-bootleg-2020	209	3	the	the	DT
orr-bootleg-2020	209	4	test	test	NN
orr-bootleg-2020	209	5	score	score	NN
orr-bootleg-2020	209	6	associated	associate	VBN
orr-bootleg-2020	209	7	with	with	IN
orr-bootleg-2020	209	8	the	the	DT
orr-bootleg-2020	209	9	best	good	JJS
orr-bootleg-2020	209	10	validation	validation	NN
orr-bootleg-2020	209	11	score.8	score.8	NN
orr-bootleg-2020	209	12	In	in	IN
orr-bootleg-2020	209	13	Table	table	NN
orr-bootleg-2020	209	14	1	1	CD
orr-bootleg-2020	209	15	,	,	,
orr-bootleg-2020	209	16	we	-PRON-	PRP
orr-bootleg-2020	209	17	show	show	VBP
orr-bootleg-2020	209	18	that	that	IN
orr-bootleg-2020	209	19	Bootleg	Bootleg	NNP
orr-bootleg-2020	209	20	achieves	achieve	VBZ
orr-bootleg-2020	209	21	up	up	IN
orr-bootleg-2020	209	22	to	to	TO
orr-bootleg-2020	209	23	5.8	5.8	CD
orr-bootleg-2020	209	24	F1	f1	NN
orr-bootleg-2020	209	25	points	point	NNS
orr-bootleg-2020	209	26	higher	high	JJR
orr-bootleg-2020	209	27	than	than	IN
orr-bootleg-2020	209	28	prior	prior	RB
orr-bootleg-2020	209	29	reported	report	VBN
orr-bootleg-2020	209	30	numbers	number	NNS
orr-bootleg-2020	209	31	on	on	IN
orr-bootleg-2020	209	32	benchmarks	benchmark	NNS
orr-bootleg-2020	209	33	.	.	.
orr-bootleg-2020	210	1	Tail	tail	VB
orr-bootleg-2020	210	2	Performance	Performance	NNP
orr-bootleg-2020	210	3	To	to	TO
orr-bootleg-2020	210	4	validate	validate	VB
orr-bootleg-2020	210	5	that	that	IN
orr-bootleg-2020	210	6	Bootleg	Bootleg	NNP
orr-bootleg-2020	210	7	improves	improve	VBZ
orr-bootleg-2020	210	8	tail	tail	NN
orr-bootleg-2020	210	9	disambiguation	disambiguation	NN
orr-bootleg-2020	210	10	,	,	,
orr-bootleg-2020	210	11	we	-PRON-	PRP
orr-bootleg-2020	210	12	compare	compare	VBP
orr-bootleg-2020	210	13	against	against	IN
orr-bootleg-2020	210	14	a	a	DT
orr-bootleg-2020	210	15	baseline	baseline	JJ
orr-bootleg-2020	210	16	model	model	NN
orr-bootleg-2020	210	17	from	from	IN
orr-bootleg-2020	210	18	Févry	Févry	NNP
orr-bootleg-2020	210	19	et	et	NNP
orr-bootleg-2020	210	20	al	al	NNP
orr-bootleg-2020	210	21	.	.	.
orr-bootleg-2020	211	1	[	[	-LRB-
orr-bootleg-2020	211	2	16	16	CD
orr-bootleg-2020	211	3	]	]	-RRB-
orr-bootleg-2020	211	4	,	,	,
orr-bootleg-2020	211	5	which	which	WDT
orr-bootleg-2020	211	6	we	-PRON-	PRP
orr-bootleg-2020	211	7	refer	refer	VBP
orr-bootleg-2020	211	8	to	to	IN
orr-bootleg-2020	211	9	as	as	IN
orr-bootleg-2020	211	10	NED	NED	NNP
orr-bootleg-2020	211	11	-	-	HYPH
orr-bootleg-2020	211	12	Base.9	Base.9	NNP
orr-bootleg-2020	211	13	NED	NED	NNP
orr-bootleg-2020	211	14	-	-	HYPH
orr-bootleg-2020	211	15	Base	Base	NNP
orr-bootleg-2020	211	16	learns	learn	VBZ
orr-bootleg-2020	211	17	entity	entity	NN
orr-bootleg-2020	211	18	embeddings	embedding	NNS
orr-bootleg-2020	211	19	by	by	IN
orr-bootleg-2020	211	20	6We	6we	CD
orr-bootleg-2020	211	21	drop	drop	NN
orr-bootleg-2020	211	22	only	only	RB
orr-bootleg-2020	211	23	1	1	CD
orr-bootleg-2020	211	24	%	%	NN
orr-bootleg-2020	211	25	of	of	IN
orr-bootleg-2020	211	26	mentions	mention	NNS
orr-bootleg-2020	211	27	from	from	IN
orr-bootleg-2020	211	28	this	this	DT
orr-bootleg-2020	211	29	filter	filter	NN
orr-bootleg-2020	211	30	.	.	.
orr-bootleg-2020	212	1	8We	8We	NNS
orr-bootleg-2020	212	2	use	use	VBP
orr-bootleg-2020	212	3	the	the	DT
orr-bootleg-2020	212	4	standard	standard	JJ
orr-bootleg-2020	212	5	candidate	candidate	NN
orr-bootleg-2020	212	6	list	list	NN
orr-bootleg-2020	212	7	from	from	IN
orr-bootleg-2020	212	8	Pershina	Pershina	NNP
orr-bootleg-2020	212	9	et	et	NNP
orr-bootleg-2020	212	10	al	al	NNP
orr-bootleg-2020	212	11	.	.	.
orr-bootleg-2020	213	1	[	[	-LRB-
orr-bootleg-2020	213	2	36	36	CD
orr-bootleg-2020	213	3	]	]	-RRB-
orr-bootleg-2020	213	4	when	when	WRB
orr-bootleg-2020	213	5	comparing	compare	VBG
orr-bootleg-2020	213	6	to	to	IN
orr-bootleg-2020	213	7	existing	exist	VBG
orr-bootleg-2020	213	8	systems	system	NNS
orr-bootleg-2020	213	9	for	for	IN
orr-bootleg-2020	213	10	fine	fine	RB
orr-bootleg-2020	213	11	-	-	HYPH
orr-bootleg-2020	213	12	tuning	tuning	NN
orr-bootleg-2020	213	13	and	and	CC
orr-bootleg-2020	213	14	inference	inference	NN
orr-bootleg-2020	213	15	for	for	IN
orr-bootleg-2020	213	16	AIDA	AIDA	NNP
orr-bootleg-2020	213	17	CoNLL	CoNLL	NNP
orr-bootleg-2020	213	18	-	-	HYPH
orr-bootleg-2020	213	19	YAGO	YAGO	NNP
orr-bootleg-2020	213	20	.	.	.
orr-bootleg-2020	214	1	9As	9as	CD
orr-bootleg-2020	214	2	code	code	NN
orr-bootleg-2020	214	3	for	for	IN
orr-bootleg-2020	214	4	the	the	DT
orr-bootleg-2020	214	5	model	model	NN
orr-bootleg-2020	214	6	from	from	IN
orr-bootleg-2020	214	7	Févry	Févry	NNP
orr-bootleg-2020	214	8	et	et	NNP
orr-bootleg-2020	214	9	al	al	NNP
orr-bootleg-2020	214	10	.	.	.
orr-bootleg-2020	215	1	[	[	-LRB-
orr-bootleg-2020	215	2	16	16	CD
orr-bootleg-2020	215	3	]	]	-RRB-
orr-bootleg-2020	215	4	is	be	VBZ
orr-bootleg-2020	215	5	not	not	RB
orr-bootleg-2020	215	6	publicly	publicly	RB
orr-bootleg-2020	215	7	available	available	JJ
orr-bootleg-2020	215	8	,	,	,
orr-bootleg-2020	215	9	we	-PRON-	PRP
orr-bootleg-2020	215	10	re	re	VBP
orr-bootleg-2020	215	11	-	-	VBD
orr-bootleg-2020	215	12	implemented	implement	VBD
orr-bootleg-2020	215	13	the	the	DT
orr-bootleg-2020	215	14	model	model	NN
orr-bootleg-2020	215	15	.	.	.
orr-bootleg-2020	216	1	We	-PRON-	PRP
orr-bootleg-2020	216	2	used	use	VBD
orr-bootleg-2020	216	3	our	-PRON-	PRP$
orr-bootleg-2020	216	4	candidate	candidate	NN
orr-bootleg-2020	216	5	8	8	CD
orr-bootleg-2020	216	6	Table	table	NN
orr-bootleg-2020	216	7	2	2	CD
orr-bootleg-2020	216	8	:	:	:
orr-bootleg-2020	216	9	(	(	-LRB-
orr-bootleg-2020	216	10	top	top	NN
orr-bootleg-2020	216	11	)	)	-RRB-
orr-bootleg-2020	216	12	We	-PRON-	PRP
orr-bootleg-2020	216	13	compare	compare	VBP
orr-bootleg-2020	216	14	Bootleg	Bootleg	NNP
orr-bootleg-2020	216	15	to	to	IN
orr-bootleg-2020	216	16	a	a	DT
orr-bootleg-2020	216	17	BERT	BERT	NNP
orr-bootleg-2020	216	18	-	-	HYPH
orr-bootleg-2020	216	19	based	base	VBN
orr-bootleg-2020	216	20	NED	NED	NNP
orr-bootleg-2020	216	21	baseline	baseline	NN
orr-bootleg-2020	216	22	(	(	-LRB-
orr-bootleg-2020	216	23	NED	NED	NNP
orr-bootleg-2020	216	24	-	-	HYPH
orr-bootleg-2020	216	25	Base	Base	NNP
orr-bootleg-2020	216	26	)	)	-RRB-
orr-bootleg-2020	216	27	on	on	IN
orr-bootleg-2020	216	28	validation	validation	NN
orr-bootleg-2020	216	29	sets	set	NNS
orr-bootleg-2020	216	30	of	of	IN
orr-bootleg-2020	216	31	a	a	DT
orr-bootleg-2020	216	32	Wikipedia	wikipedia	JJ
orr-bootleg-2020	216	33	dataset	dataset	NN
orr-bootleg-2020	216	34	.	.	.
orr-bootleg-2020	217	1	We	-PRON-	PRP
orr-bootleg-2020	217	2	report	report	VBP
orr-bootleg-2020	217	3	micro	micro	JJ
orr-bootleg-2020	217	4	-	-	JJ
orr-bootleg-2020	217	5	average	average	JJ
orr-bootleg-2020	217	6	F1	f1	NN
orr-bootleg-2020	217	7	scores	score	NNS
orr-bootleg-2020	217	8	.	.	.
orr-bootleg-2020	218	1	All	all	DT
orr-bootleg-2020	218	2	torso	torso	NN
orr-bootleg-2020	218	3	,	,	,
orr-bootleg-2020	218	4	tail	tail	NN
orr-bootleg-2020	218	5	,	,	,
orr-bootleg-2020	218	6	and	and	CC
orr-bootleg-2020	218	7	unseen	unseen	JJ
orr-bootleg-2020	218	8	validation	validation	NN
orr-bootleg-2020	218	9	sets	set	NNS
orr-bootleg-2020	218	10	are	be	VBP
orr-bootleg-2020	218	11	filtered	filter	VBN
orr-bootleg-2020	218	12	by	by	IN
orr-bootleg-2020	218	13	the	the	DT
orr-bootleg-2020	218	14	number	number	NN
orr-bootleg-2020	218	15	of	of	IN
orr-bootleg-2020	218	16	entity	entity	NN
orr-bootleg-2020	218	17	occurrences	occurrence	NNS
orr-bootleg-2020	218	18	in	in	IN
orr-bootleg-2020	218	19	the	the	DT
orr-bootleg-2020	218	20	training	training	NN
orr-bootleg-2020	218	21	data	datum	NNS
orr-bootleg-2020	218	22	and	and	CC
orr-bootleg-2020	218	23	such	such	JJ
orr-bootleg-2020	218	24	that	that	IN
orr-bootleg-2020	218	25	the	the	DT
orr-bootleg-2020	218	26	mention	mention	NN
orr-bootleg-2020	218	27	has	have	VBZ
orr-bootleg-2020	218	28	more	more	JJR
orr-bootleg-2020	218	29	than	than	IN
orr-bootleg-2020	218	30	one	one	CD
orr-bootleg-2020	218	31	candidate	candidate	NN
orr-bootleg-2020	218	32	.	.	.
orr-bootleg-2020	219	1	Model	Model	NNP
orr-bootleg-2020	219	2	All	all	DT
orr-bootleg-2020	219	3	Entities	entity	NNS
orr-bootleg-2020	219	4	Torso	Torso	NNP
orr-bootleg-2020	219	5	Entities	Entities	NNPS
orr-bootleg-2020	219	6	Tail	Tail	NNP
orr-bootleg-2020	219	7	Entities	Entities	NNP
orr-bootleg-2020	219	8	Unseen	Unseen	NNP
orr-bootleg-2020	219	9	Entities	Entities	NNP
orr-bootleg-2020	219	10	NED	NED	NNP
orr-bootleg-2020	219	11	-	-	HYPH
orr-bootleg-2020	219	12	Base	Base	NNP
orr-bootleg-2020	219	13	85.9	85.9	CD
orr-bootleg-2020	219	14	79.3	79.3	CD
orr-bootleg-2020	219	15	27.8	27.8	CD
orr-bootleg-2020	219	16	18.5	18.5	CD
orr-bootleg-2020	219	17	Bootleg	Bootleg	NNP
orr-bootleg-2020	219	18	91.3	91.3	CD
orr-bootleg-2020	219	19	87.3	87.3	CD
orr-bootleg-2020	219	20	69.0	69.0	CD
orr-bootleg-2020	219	21	68.5	68.5	CD
orr-bootleg-2020	219	22	Bootleg	Bootleg	NNP
orr-bootleg-2020	219	23	(	(	-LRB-
orr-bootleg-2020	219	24	Ent	ent	NN
orr-bootleg-2020	219	25	-	-	,
orr-bootleg-2020	219	26	only	only	RB
orr-bootleg-2020	219	27	)	)	-RRB-
orr-bootleg-2020	219	28	85.8	85.8	CD
orr-bootleg-2020	219	29	79.0	79.0	CD
orr-bootleg-2020	219	30	37.9	37.9	CD
orr-bootleg-2020	219	31	14.9	14.9	CD
orr-bootleg-2020	219	32	Bootleg	Bootleg	NNP
orr-bootleg-2020	219	33	(	(	-LRB-
orr-bootleg-2020	219	34	Type	type	NN
orr-bootleg-2020	219	35	-	-	HYPH
orr-bootleg-2020	219	36	only	only	RB
orr-bootleg-2020	219	37	)	)	-RRB-
orr-bootleg-2020	219	38	88.0	88.0	CD
orr-bootleg-2020	219	39	81.6	81.6	CD
orr-bootleg-2020	219	40	62.9	62.9	CD
orr-bootleg-2020	219	41	61.6	61.6	CD
orr-bootleg-2020	219	42	Bootleg	Bootleg	NNP
orr-bootleg-2020	219	43	(	(	-LRB-
orr-bootleg-2020	219	44	KG	KG	NNP
orr-bootleg-2020	219	45	-	-	HYPH
orr-bootleg-2020	219	46	only	only	RB
orr-bootleg-2020	219	47	)	)	-RRB-
orr-bootleg-2020	219	48	87.1	87.1	CD
orr-bootleg-2020	219	49	79.4	79.4	CD
orr-bootleg-2020	219	50	64.0	64.0	CD
orr-bootleg-2020	219	51	64.7	64.7	CD
orr-bootleg-2020	219	52	#	#	NNP
orr-bootleg-2020	219	53	Mentions	Mentions	NNPS
orr-bootleg-2020	219	54	4,065,778	4,065,778	CD
orr-bootleg-2020	219	55	1,911,590	1,911,590	CD
orr-bootleg-2020	219	56	162,761	162,761	CD
orr-bootleg-2020	219	57	9,626	9,626	CD
orr-bootleg-2020	219	58	maximizing	maximize	VBG
orr-bootleg-2020	219	59	the	the	DT
orr-bootleg-2020	219	60	dot	dot	NN
orr-bootleg-2020	219	61	product	product	NN
orr-bootleg-2020	219	62	between	between	IN
orr-bootleg-2020	219	63	the	the	DT
orr-bootleg-2020	219	64	entity	entity	NN
orr-bootleg-2020	219	65	candidates	candidate	NNS
orr-bootleg-2020	219	66	and	and	CC
orr-bootleg-2020	219	67	fine	fine	RB
orr-bootleg-2020	219	68	-	-	HYPH
orr-bootleg-2020	219	69	tuned	tune	VBN
orr-bootleg-2020	219	70	BERT	BERT	NNP
orr-bootleg-2020	219	71	-	-	HYPH
orr-bootleg-2020	219	72	contextual	contextual	JJ
orr-bootleg-2020	219	73	representations	representation	NNS
orr-bootleg-2020	219	74	of	of	IN
orr-bootleg-2020	219	75	the	the	DT
orr-bootleg-2020	219	76	mention	mention	NN
orr-bootleg-2020	219	77	.	.	.
orr-bootleg-2020	220	1	NED	NED	NNP
orr-bootleg-2020	220	2	-	-	HYPH
orr-bootleg-2020	220	3	Base	Base	NNP
orr-bootleg-2020	220	4	is	be	VBZ
orr-bootleg-2020	220	5	successful	successful	JJ
orr-bootleg-2020	220	6	overall	overall	RB
orr-bootleg-2020	220	7	on	on	IN
orr-bootleg-2020	220	8	the	the	DT
orr-bootleg-2020	220	9	validation	validation	NN
orr-bootleg-2020	220	10	achieving	achieve	VBG
orr-bootleg-2020	220	11	85.9	85.9	CD
orr-bootleg-2020	220	12	F1	F1	NNP
orr-bootleg-2020	220	13	points	point	NNS
orr-bootleg-2020	220	14	,	,	,
orr-bootleg-2020	220	15	which	which	WDT
orr-bootleg-2020	220	16	is	be	VBZ
orr-bootleg-2020	220	17	within	within	IN
orr-bootleg-2020	220	18	5.4	5.4	CD
orr-bootleg-2020	220	19	F1	f1	NN
orr-bootleg-2020	220	20	points	point	NNS
orr-bootleg-2020	220	21	of	of	IN
orr-bootleg-2020	220	22	Bootleg	Bootleg	NNP
orr-bootleg-2020	220	23	(	(	-LRB-
orr-bootleg-2020	220	24	Table	table	NN
orr-bootleg-2020	220	25	2	2	CD
orr-bootleg-2020	220	26	)	)	-RRB-
orr-bootleg-2020	220	27	.	.	.
orr-bootleg-2020	221	1	However	however	RB
orr-bootleg-2020	221	2	,	,	,
orr-bootleg-2020	221	3	when	when	WRB
orr-bootleg-2020	221	4	we	-PRON-	PRP
orr-bootleg-2020	221	5	examine	examine	VBP
orr-bootleg-2020	221	6	performance	performance	NN
orr-bootleg-2020	221	7	over	over	IN
orr-bootleg-2020	221	8	the	the	DT
orr-bootleg-2020	221	9	torso	torso	NN
orr-bootleg-2020	221	10	and	and	CC
orr-bootleg-2020	221	11	tail	tail	NN
orr-bootleg-2020	221	12	,	,	,
orr-bootleg-2020	221	13	we	-PRON-	PRP
orr-bootleg-2020	221	14	see	see	VBP
orr-bootleg-2020	221	15	that	that	IN
orr-bootleg-2020	221	16	Bootleg	Bootleg	NNP
orr-bootleg-2020	221	17	outperforms	outperform	VBZ
orr-bootleg-2020	221	18	NED	NED	NNP
orr-bootleg-2020	221	19	-	-	HYPH
orr-bootleg-2020	221	20	Base	Base	NNP
orr-bootleg-2020	221	21	by	by	IN
orr-bootleg-2020	221	22	8	8	CD
orr-bootleg-2020	221	23	and	and	CC
orr-bootleg-2020	221	24	41.2	41.2	CD
orr-bootleg-2020	221	25	F1	F1	NNP
orr-bootleg-2020	221	26	points	point	NNS
orr-bootleg-2020	221	27	,	,	,
orr-bootleg-2020	221	28	respectively	respectively	RB
orr-bootleg-2020	221	29	.	.	.
orr-bootleg-2020	222	1	Finally	finally	RB
orr-bootleg-2020	222	2	,	,	,
orr-bootleg-2020	222	3	on	on	IN
orr-bootleg-2020	222	4	unseen	unseen	JJ
orr-bootleg-2020	222	5	entities	entity	NNS
orr-bootleg-2020	222	6	,	,	,
orr-bootleg-2020	222	7	Bootleg	Bootleg	NNP
orr-bootleg-2020	222	8	outperforms	outperform	VBZ
orr-bootleg-2020	222	9	NED	NED	NNP
orr-bootleg-2020	222	10	-	-	HYPH
orr-bootleg-2020	222	11	Base	Base	NNP
orr-bootleg-2020	222	12	by	by	IN
orr-bootleg-2020	222	13	50	50	CD
orr-bootleg-2020	222	14	F1	f1	NN
orr-bootleg-2020	222	15	points	point	NNS
orr-bootleg-2020	222	16	.	.	.
orr-bootleg-2020	223	1	Note	note	VB
orr-bootleg-2020	223	2	that	that	IN
orr-bootleg-2020	223	3	NED	NED	NNP
orr-bootleg-2020	223	4	-	-	HYPH
orr-bootleg-2020	223	5	Base	Base	NNP
orr-bootleg-2020	223	6	only	only	RB
orr-bootleg-2020	223	7	has	have	VBZ
orr-bootleg-2020	223	8	access	access	NN
orr-bootleg-2020	223	9	to	to	IN
orr-bootleg-2020	223	10	textual	textual	JJ
orr-bootleg-2020	223	11	data	datum	NNS
orr-bootleg-2020	223	12	,	,	,
orr-bootleg-2020	223	13	indicating	indicate	VBG
orr-bootleg-2020	223	14	that	that	IN
orr-bootleg-2020	223	15	text	text	NN
orr-bootleg-2020	223	16	is	be	VBZ
orr-bootleg-2020	223	17	often	often	RB
orr-bootleg-2020	223	18	sufficient	sufficient	JJ
orr-bootleg-2020	223	19	for	for	IN
orr-bootleg-2020	223	20	popular	popular	JJ
orr-bootleg-2020	223	21	entities	entity	NNS
orr-bootleg-2020	223	22	,	,	,
orr-bootleg-2020	223	23	but	but	CC
orr-bootleg-2020	223	24	not	not	RB
orr-bootleg-2020	223	25	for	for	IN
orr-bootleg-2020	223	26	rare	rare	JJ
orr-bootleg-2020	223	27	entities	entity	NNS
orr-bootleg-2020	223	28	.	.	.
orr-bootleg-2020	224	1	4.3	4.3	CD
orr-bootleg-2020	224	2	Downstream	Downstream	NNP
orr-bootleg-2020	224	3	Evaluation	Evaluation	NNP
orr-bootleg-2020	224	4	Relation	Relation	NNP
orr-bootleg-2020	224	5	Extraction	Extraction	NNP
orr-bootleg-2020	224	6	Using	use	VBG
orr-bootleg-2020	224	7	the	the	DT
orr-bootleg-2020	224	8	learned	learn	VBN
orr-bootleg-2020	224	9	representations	representation	NNS
orr-bootleg-2020	224	10	from	from	IN
orr-bootleg-2020	224	11	Bootleg	Bootleg	NNP
orr-bootleg-2020	224	12	,	,	,
orr-bootleg-2020	224	13	we	-PRON-	PRP
orr-bootleg-2020	224	14	achieve	achieve	VBP
orr-bootleg-2020	224	15	the	the	DT
orr-bootleg-2020	224	16	new	new	JJ
orr-bootleg-2020	224	17	state	state	NN
orr-bootleg-2020	224	18	-	-	HYPH
orr-bootleg-2020	224	19	of	of	IN
orr-bootleg-2020	224	20	-	-	HYPH
orr-bootleg-2020	224	21	the	the	DT
orr-bootleg-2020	224	22	-	-	HYPH
orr-bootleg-2020	224	23	art	art	NN
orr-bootleg-2020	224	24	on	on	IN
orr-bootleg-2020	224	25	TACRED	TACRED	NNP
orr-bootleg-2020	224	26	,	,	,
orr-bootleg-2020	224	27	a	a	DT
orr-bootleg-2020	224	28	standard	standard	JJ
orr-bootleg-2020	224	29	relation	relation	NN
orr-bootleg-2020	224	30	extraction	extraction	NN
orr-bootleg-2020	224	31	benchmark	benchmark	NN
orr-bootleg-2020	224	32	.	.	.
orr-bootleg-2020	225	1	TACRED	TACRED	NNP
orr-bootleg-2020	225	2	involves	involve	VBZ
orr-bootleg-2020	225	3	identifying	identify	VBG
orr-bootleg-2020	225	4	the	the	DT
orr-bootleg-2020	225	5	relationship	relationship	NN
orr-bootleg-2020	225	6	between	between	IN
orr-bootleg-2020	225	7	a	a	DT
orr-bootleg-2020	225	8	specified	specify	VBN
orr-bootleg-2020	225	9	subject	subject	NN
orr-bootleg-2020	225	10	and	and	CC
orr-bootleg-2020	225	11	object	object	NN
orr-bootleg-2020	225	12	in	in	IN
orr-bootleg-2020	225	13	an	an	DT
orr-bootleg-2020	225	14	example	example	NN
orr-bootleg-2020	225	15	sentence	sentence	NN
orr-bootleg-2020	225	16	as	as	IN
orr-bootleg-2020	225	17	one	one	CD
orr-bootleg-2020	225	18	of	of	IN
orr-bootleg-2020	225	19	41	41	CD
orr-bootleg-2020	225	20	relation	relation	NN
orr-bootleg-2020	225	21	types	type	NNS
orr-bootleg-2020	225	22	(	(	-LRB-
orr-bootleg-2020	225	23	e.g.	e.g.	RB
orr-bootleg-2020	225	24	,	,	,
orr-bootleg-2020	225	25	spouse	spouse	NN
orr-bootleg-2020	225	26	)	)	-RRB-
orr-bootleg-2020	225	27	or	or	CC
orr-bootleg-2020	225	28	no	no	DT
orr-bootleg-2020	225	29	relation	relation	NN
orr-bootleg-2020	225	30	.	.	.
orr-bootleg-2020	226	1	Relation	relation	NN
orr-bootleg-2020	226	2	extraction	extraction	NN
orr-bootleg-2020	226	3	is	be	VBZ
orr-bootleg-2020	226	4	a	a	DT
orr-bootleg-2020	226	5	well	well	RB
orr-bootleg-2020	226	6	-	-	HYPH
orr-bootleg-2020	226	7	suited	suit	VBN
orr-bootleg-2020	226	8	for	for	IN
orr-bootleg-2020	226	9	evaluating	evaluate	VBG
orr-bootleg-2020	226	10	Bootleg	Bootleg	NNP
orr-bootleg-2020	226	11	because	because	IN
orr-bootleg-2020	226	12	the	the	DT
orr-bootleg-2020	226	13	substrings	substring	NNS
orr-bootleg-2020	226	14	in	in	IN
orr-bootleg-2020	226	15	the	the	DT
orr-bootleg-2020	226	16	text	text	NN
orr-bootleg-2020	226	17	can	can	MD
orr-bootleg-2020	226	18	refer	refer	VB
orr-bootleg-2020	226	19	to	to	IN
orr-bootleg-2020	226	20	many	many	JJ
orr-bootleg-2020	226	21	different	different	JJ
orr-bootleg-2020	226	22	entities	entity	NNS
orr-bootleg-2020	226	23	,	,	,
orr-bootleg-2020	226	24	and	and	CC
orr-bootleg-2020	226	25	the	the	DT
orr-bootleg-2020	226	26	disambiguated	disambiguate	VBN
orr-bootleg-2020	226	27	entities	entity	NNS
orr-bootleg-2020	226	28	impact	impact	VBP
orr-bootleg-2020	226	29	the	the	DT
orr-bootleg-2020	226	30	set	set	NN
orr-bootleg-2020	226	31	of	of	IN
orr-bootleg-2020	226	32	likely	likely	JJ
orr-bootleg-2020	226	33	relations	relation	NNS
orr-bootleg-2020	226	34	.	.	.
orr-bootleg-2020	227	1	Given	give	VBN
orr-bootleg-2020	227	2	an	an	DT
orr-bootleg-2020	227	3	example	example	NN
orr-bootleg-2020	227	4	,	,	,
orr-bootleg-2020	227	5	we	-PRON-	PRP
orr-bootleg-2020	227	6	run	run	VBP
orr-bootleg-2020	227	7	inference	inference	NN
orr-bootleg-2020	227	8	with	with	IN
orr-bootleg-2020	227	9	the	the	DT
orr-bootleg-2020	227	10	Bootleg	Bootleg	NNP
orr-bootleg-2020	227	11	model	model	NN
orr-bootleg-2020	227	12	to	to	TO
orr-bootleg-2020	227	13	disambiguate	disambiguate	VB
orr-bootleg-2020	227	14	named	name	VBN
orr-bootleg-2020	227	15	entities	entity	NNS
orr-bootleg-2020	227	16	and	and	CC
orr-bootleg-2020	227	17	generate	generate	VB
orr-bootleg-2020	227	18	the	the	DT
orr-bootleg-2020	227	19	contextual	contextual	JJ
orr-bootleg-2020	227	20	Bootleg	Bootleg	NNP
orr-bootleg-2020	227	21	entity	entity	NN
orr-bootleg-2020	227	22	embedding	embed	VBG
orr-bootleg-2020	227	23	matrix	matrix	NN
orr-bootleg-2020	227	24	,	,	,
orr-bootleg-2020	227	25	which	which	WDT
orr-bootleg-2020	227	26	we	-PRON-	PRP
orr-bootleg-2020	227	27	feed	feed	VBP
orr-bootleg-2020	227	28	to	to	IN
orr-bootleg-2020	227	29	a	a	DT
orr-bootleg-2020	227	30	simple	simple	JJ
orr-bootleg-2020	227	31	Transformer	transformer	NN
orr-bootleg-2020	227	32	architecture	architecture	NN
orr-bootleg-2020	227	33	that	that	WDT
orr-bootleg-2020	227	34	uses	use	VBZ
orr-bootleg-2020	227	35	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	227	36	[	[	-LRB-
orr-bootleg-2020	227	37	27	27	CD
orr-bootleg-2020	227	38	]	]	-RRB-
orr-bootleg-2020	227	39	(	(	-LRB-
orr-bootleg-2020	227	40	details	detail	NNS
orr-bootleg-2020	227	41	in	in	IN
orr-bootleg-2020	227	42	Appendix	Appendix	NNP
orr-bootleg-2020	227	43	C	C	NNP
orr-bootleg-2020	227	44	)	)	-RRB-
orr-bootleg-2020	227	45	.	.	.
orr-bootleg-2020	228	1	We	-PRON-	PRP
orr-bootleg-2020	228	2	achieve	achieve	VBP
orr-bootleg-2020	228	3	a	a	DT
orr-bootleg-2020	228	4	micro	micro	JJ
orr-bootleg-2020	228	5	-	-	JJ
orr-bootleg-2020	228	6	average	average	JJ
orr-bootleg-2020	228	7	test	test	NN
orr-bootleg-2020	228	8	F1	F1	NNP
orr-bootleg-2020	228	9	score	score	NN
orr-bootleg-2020	228	10	of	of	IN
orr-bootleg-2020	228	11	80.3	80.3	CD
orr-bootleg-2020	228	12	,	,	,
orr-bootleg-2020	228	13	which	which	WDT
orr-bootleg-2020	228	14	improves	improve	VBZ
orr-bootleg-2020	228	15	upon	upon	IN
orr-bootleg-2020	228	16	the	the	DT
orr-bootleg-2020	228	17	prior	prior	JJ
orr-bootleg-2020	228	18	state	state	NN
orr-bootleg-2020	228	19	of	of	IN
orr-bootleg-2020	228	20	the	the	DT
orr-bootleg-2020	228	21	art	art	NN
orr-bootleg-2020	228	22	—	—	:
orr-bootleg-2020	228	23	KnowBERT	KnowBERT	NNP
orr-bootleg-2020	228	24	[	[	-LRB-
orr-bootleg-2020	228	25	38	38	CD
orr-bootleg-2020	228	26	]	]	-RRB-
orr-bootleg-2020	228	27	,	,	,
orr-bootleg-2020	228	28	which	which	WDT
orr-bootleg-2020	228	29	also	also	RB
orr-bootleg-2020	228	30	uses	use	VBZ
orr-bootleg-2020	228	31	entity	entity	NN
orr-bootleg-2020	228	32	-	-	HYPH
orr-bootleg-2020	228	33	based	base	VBN
orr-bootleg-2020	228	34	knowledge	knowledge	NN
orr-bootleg-2020	228	35	—	—	:
orr-bootleg-2020	228	36	by	by	IN
orr-bootleg-2020	228	37	1.0	1.0	CD
orr-bootleg-2020	228	38	F1	f1	NN
orr-bootleg-2020	228	39	points	point	NNS
orr-bootleg-2020	228	40	and	and	CC
orr-bootleg-2020	228	41	the	the	DT
orr-bootleg-2020	228	42	baseline	baseline	NN
orr-bootleg-2020	228	43	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	228	44	model	model	NN
orr-bootleg-2020	228	45	by	by	IN
orr-bootleg-2020	228	46	2.3	2.3	CD
orr-bootleg-2020	228	47	F1	f1	NN
orr-bootleg-2020	228	48	points	point	NNS
orr-bootleg-2020	228	49	on	on	IN
orr-bootleg-2020	228	50	TACRED	TACRED	NNP
orr-bootleg-2020	228	51	-	-	HYPH
orr-bootleg-2020	228	52	Revisited	revisit	VBN
orr-bootleg-2020	228	53	data	datum	NNS
orr-bootleg-2020	228	54	(	(	-LRB-
orr-bootleg-2020	228	55	Table	table	NN
orr-bootleg-2020	228	56	3	3	CD
orr-bootleg-2020	228	57	)	)	-RRB-
orr-bootleg-2020	228	58	(	(	-LRB-
orr-bootleg-2020	228	59	[	[	-LRB-
orr-bootleg-2020	228	60	53	53	CD
orr-bootleg-2020	228	61	]	]	-RRB-
orr-bootleg-2020	228	62	,	,	,
orr-bootleg-2020	228	63	Alt	Alt	NNP
orr-bootleg-2020	228	64	et	et	NNP
orr-bootleg-2020	228	65	al	al	NNP
orr-bootleg-2020	228	66	.	.	.
orr-bootleg-2020	229	1	[	[	-LRB-
orr-bootleg-2020	229	2	2	2	CD
orr-bootleg-2020	229	3	]	]	-RRB-
orr-bootleg-2020	229	4	)	)	-RRB-
orr-bootleg-2020	229	5	.	.	.
orr-bootleg-2020	230	1	We	-PRON-	PRP
orr-bootleg-2020	230	2	find	find	VBP
orr-bootleg-2020	230	3	that	that	IN
orr-bootleg-2020	230	4	the	the	DT
orr-bootleg-2020	230	5	Bootleg	Bootleg	NNP
orr-bootleg-2020	230	6	downstream	downstream	JJ
orr-bootleg-2020	230	7	model	model	NN
orr-bootleg-2020	230	8	corrects	correct	VBZ
orr-bootleg-2020	230	9	errors	error	NNS
orr-bootleg-2020	230	10	made	make	VBN
orr-bootleg-2020	230	11	by	by	IN
orr-bootleg-2020	230	12	the	the	DT
orr-bootleg-2020	230	13	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	230	14	baseline	baseline	NN
orr-bootleg-2020	230	15	,	,	,
orr-bootleg-2020	230	16	for	for	IN
orr-bootleg-2020	230	17	example	example	NN
orr-bootleg-2020	230	18	by	by	IN
orr-bootleg-2020	230	19	leveraging	leverage	VBG
orr-bootleg-2020	230	20	entity	entity	NN
orr-bootleg-2020	230	21	,	,	,
orr-bootleg-2020	230	22	type	type	NN
orr-bootleg-2020	230	23	,	,	,
orr-bootleg-2020	230	24	and	and	CC
orr-bootleg-2020	230	25	relation	relation	NN
orr-bootleg-2020	230	26	information	information	NN
orr-bootleg-2020	230	27	or	or	CC
orr-bootleg-2020	230	28	recognizing	recognize	VBG
orr-bootleg-2020	230	29	that	that	DT
orr-bootleg-2020	230	30	different	different	JJ
orr-bootleg-2020	230	31	textual	textual	JJ
orr-bootleg-2020	230	32	aliases	alias	NNS
orr-bootleg-2020	230	33	refer	refer	VBP
orr-bootleg-2020	230	34	to	to	IN
orr-bootleg-2020	230	35	the	the	DT
orr-bootleg-2020	230	36	same	same	JJ
orr-bootleg-2020	230	37	entity	entity	NN
orr-bootleg-2020	230	38	(	(	-LRB-
orr-bootleg-2020	230	39	see	see	VB
orr-bootleg-2020	230	40	Table	table	NN
orr-bootleg-2020	230	41	4	4	CD
orr-bootleg-2020	230	42	)	)	-RRB-
orr-bootleg-2020	230	43	.	.	.
orr-bootleg-2020	231	1	generators	generator	NNS
orr-bootleg-2020	231	2	and	and	CC
orr-bootleg-2020	231	3	fine	fine	RB
orr-bootleg-2020	231	4	-	-	HYPH
orr-bootleg-2020	231	5	tuned	tune	VBN
orr-bootleg-2020	231	6	a	a	DT
orr-bootleg-2020	231	7	pretrained	pretrained	JJ
orr-bootleg-2020	231	8	BERT	BERT	NNP
orr-bootleg-2020	231	9	encoder	encoder	NN
orr-bootleg-2020	231	10	rather	rather	RB
orr-bootleg-2020	231	11	than	than	IN
orr-bootleg-2020	231	12	training	train	VBG
orr-bootleg-2020	231	13	a	a	DT
orr-bootleg-2020	231	14	BERT	BERT	NNP
orr-bootleg-2020	231	15	encoder	encoder	NN
orr-bootleg-2020	231	16	from	from	IN
orr-bootleg-2020	231	17	scratch	scratch	NN
orr-bootleg-2020	231	18	,	,	,
orr-bootleg-2020	231	19	as	as	IN
orr-bootleg-2020	231	20	is	be	VBZ
orr-bootleg-2020	231	21	done	do	VBN
orr-bootleg-2020	231	22	in	in	IN
orr-bootleg-2020	231	23	Févry	Févry	NNP
orr-bootleg-2020	231	24	et	et	NNP
orr-bootleg-2020	231	25	al	al	NNP
orr-bootleg-2020	231	26	.	.	.
orr-bootleg-2020	232	1	[	[	-LRB-
orr-bootleg-2020	232	2	16	16	CD
orr-bootleg-2020	232	3	]	]	-RRB-
orr-bootleg-2020	232	4	.	.	.
orr-bootleg-2020	233	1	We	-PRON-	PRP
orr-bootleg-2020	233	2	trained	train	VBD
orr-bootleg-2020	233	3	NED	NED	NNP
orr-bootleg-2020	233	4	-	-	HYPH
orr-bootleg-2020	233	5	Base	Base	NNP
orr-bootleg-2020	233	6	on	on	IN
orr-bootleg-2020	233	7	the	the	DT
orr-bootleg-2020	233	8	same	same	JJ
orr-bootleg-2020	233	9	weak	weak	JJ
orr-bootleg-2020	233	10	labelled	label	VBN
orr-bootleg-2020	233	11	data	datum	NNS
orr-bootleg-2020	233	12	as	as	IN
orr-bootleg-2020	233	13	Bootleg	Bootleg	NNP
orr-bootleg-2020	233	14	for	for	IN
orr-bootleg-2020	233	15	2	2	CD
orr-bootleg-2020	233	16	epochs	epoch	NNS
orr-bootleg-2020	233	17	.	.	.
orr-bootleg-2020	234	1	Table	table	NN
orr-bootleg-2020	234	2	3	3	CD
orr-bootleg-2020	234	3	:	:	:
orr-bootleg-2020	234	4	Test	Test	NNP
orr-bootleg-2020	234	5	micro	micro	JJ
orr-bootleg-2020	234	6	-	-	JJ
orr-bootleg-2020	234	7	average	average	JJ
orr-bootleg-2020	234	8	F1	f1	NN
orr-bootleg-2020	234	9	score	score	NN
orr-bootleg-2020	234	10	on	on	IN
orr-bootleg-2020	234	11	revised	revised	JJ
orr-bootleg-2020	234	12	TACRED	TACRED	NNP
orr-bootleg-2020	234	13	dataset	dataset	NN
orr-bootleg-2020	234	14	.	.	.
orr-bootleg-2020	235	1	Validation	validation	NN
orr-bootleg-2020	235	2	Set	Set	NNP
orr-bootleg-2020	235	3	F1	F1	NNP
orr-bootleg-2020	235	4	Bootleg	Bootleg	NNP
orr-bootleg-2020	235	5	Model	Model	NNP
orr-bootleg-2020	235	6	80.3	80.3	CD
orr-bootleg-2020	235	7	KnowBERT	KnowBERT	NNP
orr-bootleg-2020	235	8	79.3	79.3	CD
orr-bootleg-2020	235	9	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	235	10	78.0	78.0	CD
orr-bootleg-2020	235	11	9	9	CD
orr-bootleg-2020	235	12	Table	table	NN
orr-bootleg-2020	235	13	4	4	CD
orr-bootleg-2020	235	14	:	:	:
orr-bootleg-2020	235	15	The	the	DT
orr-bootleg-2020	235	16	following	follow	VBG
orr-bootleg-2020	235	17	are	be	VBP
orr-bootleg-2020	235	18	examples	example	NNS
orr-bootleg-2020	235	19	of	of	IN
orr-bootleg-2020	235	20	how	how	WRB
orr-bootleg-2020	235	21	the	the	DT
orr-bootleg-2020	235	22	contextual	contextual	JJ
orr-bootleg-2020	235	23	entity	entity	NN
orr-bootleg-2020	235	24	representation	representation	NN
orr-bootleg-2020	235	25	from	from	IN
orr-bootleg-2020	235	26	Bootleg	Bootleg	NNP
orr-bootleg-2020	235	27	,	,	,
orr-bootleg-2020	235	28	generated	generate	VBN
orr-bootleg-2020	235	29	from	from	IN
orr-bootleg-2020	235	30	entity	entity	NN
orr-bootleg-2020	235	31	,	,	,
orr-bootleg-2020	235	32	relation	relation	NN
orr-bootleg-2020	235	33	,	,	,
orr-bootleg-2020	235	34	and	and	CC
orr-bootleg-2020	235	35	type	type	NN
orr-bootleg-2020	235	36	signals	signal	NNS
orr-bootleg-2020	235	37	,	,	,
orr-bootleg-2020	235	38	can	can	MD
orr-bootleg-2020	235	39	help	help	VB
orr-bootleg-2020	235	40	our	-PRON-	PRP$
orr-bootleg-2020	235	41	downstream	downstream	JJ
orr-bootleg-2020	235	42	model	model	NN
orr-bootleg-2020	235	43	.	.	.
orr-bootleg-2020	236	1	We	-PRON-	PRP
orr-bootleg-2020	236	2	provide	provide	VBP
orr-bootleg-2020	236	3	the	the	DT
orr-bootleg-2020	236	4	TACRED	tacred	JJ
orr-bootleg-2020	236	5	example	example	NN
orr-bootleg-2020	236	6	,	,	,
orr-bootleg-2020	236	7	signals	signal	NNS
orr-bootleg-2020	236	8	provided	provide	VBN
orr-bootleg-2020	236	9	by	by	IN
orr-bootleg-2020	236	10	Bootleg	Bootleg	NNP
orr-bootleg-2020	236	11	,	,	,
orr-bootleg-2020	236	12	as	as	IN
orr-bootleg-2020	236	13	well	well	RB
orr-bootleg-2020	236	14	our	-PRON-	PRP$
orr-bootleg-2020	236	15	model	model	NN
orr-bootleg-2020	236	16	and	and	CC
orr-bootleg-2020	236	17	the	the	DT
orr-bootleg-2020	236	18	baseline	baseline	NN
orr-bootleg-2020	236	19	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	236	20	models	model	NNS
orr-bootleg-2020	236	21	’	’	POS
orr-bootleg-2020	236	22	predictions	prediction	NNS
orr-bootleg-2020	236	23	.	.	.
orr-bootleg-2020	237	1	TACRED	TACRED	NNP
orr-bootleg-2020	237	2	Example	Example	NNP
orr-bootleg-2020	237	3	Bootleg	Bootleg	NNP
orr-bootleg-2020	237	4	Signals	signal	VBZ
orr-bootleg-2020	237	5	Our	-PRON-	PRP$
orr-bootleg-2020	237	6	Prediction	Prediction	NNP
orr-bootleg-2020	237	7	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	237	8	Prediction	Prediction	NNP
orr-bootleg-2020	237	9	Vincent	Vincent	NNP
orr-bootleg-2020	237	10	Astor	Astor	NNP
orr-bootleg-2020	237	11	,	,	,
orr-bootleg-2020	237	12	like	like	IN
orr-bootleg-2020	237	13	Marshall	Marshall	NNP
orr-bootleg-2020	237	14	(	(	-LRB-
orr-bootleg-2020	237	15	subj	subj	NNP
orr-bootleg-2020	237	16	)	)	-RRB-
orr-bootleg-2020	237	17	,	,	,
orr-bootleg-2020	237	18	died	die	VBD
orr-bootleg-2020	237	19	unexpectedly	unexpectedly	RB
orr-bootleg-2020	237	20	of	of	IN
orr-bootleg-2020	237	21	a	a	DT
orr-bootleg-2020	237	22	heart	heart	NN
orr-bootleg-2020	237	23	attack	attack	NN
orr-bootleg-2020	237	24	(	(	-LRB-
orr-bootleg-2020	237	25	obj	obj	UH
orr-bootleg-2020	237	26	)	)	-RRB-
orr-bootleg-2020	237	27	in	in	IN
orr-bootleg-2020	237	28	1959	1959	CD
orr-bootleg-2020	237	29	..	..	NFP
orr-bootleg-2020	237	30	.	.	.
orr-bootleg-2020	238	1	Gold	gold	NN
orr-bootleg-2020	238	2	relation	relation	NN
orr-bootleg-2020	238	3	:	:	:
orr-bootleg-2020	238	4	Cause	cause	NN
orr-bootleg-2020	238	5	of	of	IN
orr-bootleg-2020	238	6	Death	Death	NNP
orr-bootleg-2020	238	7	Disambiguates	Disambiguates	NNPS
orr-bootleg-2020	238	8	“	"	``
orr-bootleg-2020	238	9	Marshall	Marshall	NNP
orr-bootleg-2020	238	10	”	"	''
orr-bootleg-2020	238	11	to	to	IN
orr-bootleg-2020	238	12	Thomas	Thomas	NNP
orr-bootleg-2020	238	13	Riley	Riley	NNP
orr-bootleg-2020	238	14	Marshall	Marshall	NNP
orr-bootleg-2020	238	15	and	and	CC
orr-bootleg-2020	238	16	“	"	``
orr-bootleg-2020	238	17	heart	heart	NN
orr-bootleg-2020	238	18	attack	attack	NN
orr-bootleg-2020	238	19	”	"	''
orr-bootleg-2020	238	20	to	to	IN
orr-bootleg-2020	238	21	myocardial	myocardial	JJ
orr-bootleg-2020	238	22	infarction	infarction	NN
orr-bootleg-2020	238	23	,	,	,
orr-bootleg-2020	238	24	which	which	WDT
orr-bootleg-2020	238	25	have	have	VBP
orr-bootleg-2020	238	26	the	the	DT
orr-bootleg-2020	238	27	Wikidata	Wikidata	NNP
orr-bootleg-2020	238	28	relation	relation	NN
orr-bootleg-2020	238	29	“	"	``
orr-bootleg-2020	238	30	cause	cause	NN
orr-bootleg-2020	238	31	of	of	IN
orr-bootleg-2020	238	32	death	death	NN
orr-bootleg-2020	238	33	”	"	''
orr-bootleg-2020	238	34	Cause	cause	NN
orr-bootleg-2020	238	35	of	of	IN
orr-bootleg-2020	238	36	Death	death	NN
orr-bootleg-2020	238	37	No	no	UH
orr-bootleg-2020	238	38	Relation	Relation	NNP
orr-bootleg-2020	238	39	The	the	DT
orr-bootleg-2020	238	40	International	International	NNP
orr-bootleg-2020	238	41	Water	Water	NNP
orr-bootleg-2020	238	42	Management	Management	NNP
orr-bootleg-2020	238	43	(	(	-LRB-
orr-bootleg-2020	238	44	obj	obj	DT
orr-bootleg-2020	238	45	)	)	-RRB-
orr-bootleg-2020	238	46	Institute	Institute	NNP
orr-bootleg-2020	238	47	or	or	CC
orr-bootleg-2020	238	48	IWMI	IWMI	NNP
orr-bootleg-2020	238	49	(	(	-LRB-
orr-bootleg-2020	238	50	subj	subj	NNP
orr-bootleg-2020	238	51	)	)	-RRB-
orr-bootleg-2020	238	52	study	study	NN
orr-bootleg-2020	238	53	said	say	VBD
orr-bootleg-2020	238	54	both	both	DT
orr-bootleg-2020	238	55	.	.	.
orr-bootleg-2020	239	1	...	...	:
orr-bootleg-2020	239	2	Gold	gold	NN
orr-bootleg-2020	239	3	relation	relation	NN
orr-bootleg-2020	239	4	:	:	:
orr-bootleg-2020	239	5	Alternate	Alternate	NNP
orr-bootleg-2020	239	6	Names	Names	NNPS
orr-bootleg-2020	239	7	Disambiguates	Disambiguates	NNPS
orr-bootleg-2020	239	8	alias	alia	NNS
orr-bootleg-2020	239	9	“	"	``
orr-bootleg-2020	239	10	International	International	NNP
orr-bootleg-2020	239	11	Water	Water	NNP
orr-bootleg-2020	239	12	Management	Management	NNP
orr-bootleg-2020	239	13	Institute	Institute	NNP
orr-bootleg-2020	239	14	”	"	''
orr-bootleg-2020	239	15	and	and	CC
orr-bootleg-2020	239	16	its	-PRON-	PRP$
orr-bootleg-2020	239	17	acronym	acronym	NN
orr-bootleg-2020	239	18	,	,	,
orr-bootleg-2020	239	19	the	the	DT
orr-bootleg-2020	239	20	alias	alia	NNS
orr-bootleg-2020	239	21	“	"	``
orr-bootleg-2020	239	22	IWMI	IWMI	NNP
orr-bootleg-2020	239	23	”	"	''
orr-bootleg-2020	239	24	,	,	,
orr-bootleg-2020	239	25	to	to	IN
orr-bootleg-2020	239	26	the	the	DT
orr-bootleg-2020	239	27	same	same	JJ
orr-bootleg-2020	239	28	Wikidata	Wikidata	NNP
orr-bootleg-2020	239	29	entity	entity	NN
orr-bootleg-2020	239	30	Alternate	Alternate	NNP
orr-bootleg-2020	239	31	Names	name	VBZ
orr-bootleg-2020	239	32	No	no	UH
orr-bootleg-2020	239	33	Relation	relation	NN
orr-bootleg-2020	239	34	In	in	IN
orr-bootleg-2020	239	35	studying	study	VBG
orr-bootleg-2020	239	36	the	the	DT
orr-bootleg-2020	239	37	slices	slice	NNS
orr-bootleg-2020	239	38	for	for	IN
orr-bootleg-2020	239	39	which	which	WDT
orr-bootleg-2020	239	40	the	the	DT
orr-bootleg-2020	239	41	Bootleg	Bootleg	NNP
orr-bootleg-2020	239	42	downstream	downstream	JJ
orr-bootleg-2020	239	43	model	model	NN
orr-bootleg-2020	239	44	improves	improve	VBZ
orr-bootleg-2020	239	45	upon	upon	IN
orr-bootleg-2020	239	46	the	the	DT
orr-bootleg-2020	239	47	baseline	baseline	JJ
orr-bootleg-2020	239	48	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	239	49	model	model	NN
orr-bootleg-2020	239	50	,	,	,
orr-bootleg-2020	239	51	we	-PRON-	PRP
orr-bootleg-2020	239	52	rank	rank	VBP
orr-bootleg-2020	239	53	TACRED	TACRED	NNP
orr-bootleg-2020	239	54	examples	example	NNS
orr-bootleg-2020	239	55	in	in	IN
orr-bootleg-2020	239	56	three	three	CD
orr-bootleg-2020	239	57	ways	way	NNS
orr-bootleg-2020	239	58	:	:	:
orr-bootleg-2020	239	59	by	by	IN
orr-bootleg-2020	239	60	the	the	DT
orr-bootleg-2020	239	61	proportion	proportion	NN
orr-bootleg-2020	239	62	of	of	IN
orr-bootleg-2020	239	63	words	word	NNS
orr-bootleg-2020	239	64	where	where	WRB
orr-bootleg-2020	239	65	Bootleg	Bootleg	NNP
orr-bootleg-2020	239	66	disambiguates	disambiguate	VBZ
orr-bootleg-2020	239	67	it	-PRON-	PRP
orr-bootleg-2020	239	68	as	as	IN
orr-bootleg-2020	239	69	an	an	DT
orr-bootleg-2020	239	70	entity	entity	NN
orr-bootleg-2020	239	71	,	,	,
orr-bootleg-2020	239	72	leverages	leverage	VBZ
orr-bootleg-2020	239	73	Wikidata	Wikidata	NNP
orr-bootleg-2020	239	74	relations	relation	NNS
orr-bootleg-2020	239	75	for	for	IN
orr-bootleg-2020	239	76	the	the	DT
orr-bootleg-2020	239	77	embedding	embedding	NN
orr-bootleg-2020	239	78	,	,	,
orr-bootleg-2020	239	79	and	and	CC
orr-bootleg-2020	239	80	leverages	leverage	VBZ
orr-bootleg-2020	239	81	Wikidata	Wikidata	NNP
orr-bootleg-2020	239	82	types	type	NNS
orr-bootleg-2020	239	83	for	for	IN
orr-bootleg-2020	239	84	the	the	DT
orr-bootleg-2020	239	85	embedding	embedding	NN
orr-bootleg-2020	239	86	.	.	.
orr-bootleg-2020	240	1	For	for	IN
orr-bootleg-2020	240	2	each	each	DT
orr-bootleg-2020	240	3	of	of	IN
orr-bootleg-2020	240	4	these	these	DT
orr-bootleg-2020	240	5	three	three	CD
orr-bootleg-2020	240	6	,	,	,
orr-bootleg-2020	240	7	we	-PRON-	PRP
orr-bootleg-2020	240	8	report	report	VBP
orr-bootleg-2020	240	9	the	the	DT
orr-bootleg-2020	240	10	gap	gap	NN
orr-bootleg-2020	240	11	between	between	IN
orr-bootleg-2020	240	12	the	the	DT
orr-bootleg-2020	240	13	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	240	14	model	model	NN
orr-bootleg-2020	240	15	and	and	CC
orr-bootleg-2020	240	16	Bootleg	Bootleg	NNP
orr-bootleg-2020	240	17	model	model	NN
orr-bootleg-2020	240	18	’s	’s	POS
orr-bootleg-2020	240	19	error	error	NN
orr-bootleg-2020	240	20	rates	rate	NNS
orr-bootleg-2020	240	21	on	on	IN
orr-bootleg-2020	240	22	the	the	DT
orr-bootleg-2020	240	23	examples	example	NNS
orr-bootleg-2020	240	24	with	with	IN
orr-bootleg-2020	240	25	above	above	JJ
orr-bootleg-2020	240	26	-	-	HYPH
orr-bootleg-2020	240	27	median	median	JJ
orr-bootleg-2020	240	28	proportion	proportion	NN
orr-bootleg-2020	240	29	(	(	-LRB-
orr-bootleg-2020	240	30	more	more	JJR
orr-bootleg-2020	240	31	Bootleg	Bootleg	NNP
orr-bootleg-2020	240	32	signal	signal	NN
orr-bootleg-2020	240	33	)	)	-RRB-
orr-bootleg-2020	240	34	relative	relative	JJ
orr-bootleg-2020	240	35	to	to	IN
orr-bootleg-2020	240	36	the	the	DT
orr-bootleg-2020	240	37	below	below	IN
orr-bootleg-2020	240	38	-	-	HYPH
orr-bootleg-2020	240	39	median	median	JJ
orr-bootleg-2020	240	40	proportion	proportion	NN
orr-bootleg-2020	240	41	(	(	-LRB-
orr-bootleg-2020	240	42	less	less	JJR
orr-bootleg-2020	240	43	Bootleg	Bootleg	NNP
orr-bootleg-2020	240	44	signal	signal	NN
orr-bootleg-2020	240	45	)	)	-RRB-
orr-bootleg-2020	240	46	.	.	.
orr-bootleg-2020	241	1	We	-PRON-	PRP
orr-bootleg-2020	241	2	find	find	VBP
orr-bootleg-2020	241	3	that	that	IN
orr-bootleg-2020	241	4	the	the	DT
orr-bootleg-2020	241	5	relative	relative	JJ
orr-bootleg-2020	241	6	gap	gap	NN
orr-bootleg-2020	241	7	between	between	IN
orr-bootleg-2020	241	8	the	the	DT
orr-bootleg-2020	241	9	baseline	baseline	NN
orr-bootleg-2020	241	10	and	and	CC
orr-bootleg-2020	241	11	Bootleg	Bootleg	NNP
orr-bootleg-2020	241	12	error	error	NN
orr-bootleg-2020	241	13	rates	rate	NNS
orr-bootleg-2020	241	14	is	be	VBZ
orr-bootleg-2020	241	15	larger	large	JJR
orr-bootleg-2020	241	16	on	on	IN
orr-bootleg-2020	241	17	the	the	DT
orr-bootleg-2020	241	18	slice	slice	NN
orr-bootleg-2020	241	19	above	above	RB
orr-bootleg-2020	241	20	(	(	-LRB-
orr-bootleg-2020	241	21	with	with	IN
orr-bootleg-2020	241	22	more	more	JJR
orr-bootleg-2020	241	23	Bootleg	Bootleg	NNP
orr-bootleg-2020	241	24	information	information	NN
orr-bootleg-2020	241	25	)	)	-RRB-
orr-bootleg-2020	241	26	than	than	IN
orr-bootleg-2020	241	27	below	below	IN
orr-bootleg-2020	241	28	the	the	DT
orr-bootleg-2020	241	29	median	median	NN
orr-bootleg-2020	241	30	by	by	IN
orr-bootleg-2020	241	31	1.10x	1.10x	CD
orr-bootleg-2020	241	32	,	,	,
orr-bootleg-2020	241	33	4.67x	4.67x	CD
orr-bootleg-2020	241	34	,	,	,
orr-bootleg-2020	241	35	and	and	CC
orr-bootleg-2020	241	36	1.35x	1.35x	CD
orr-bootleg-2020	241	37	respectively	respectively	RB
orr-bootleg-2020	241	38	:	:	:
orr-bootleg-2020	241	39	with	with	IN
orr-bootleg-2020	241	40	more	more	JJR
orr-bootleg-2020	241	41	Bootleg	Bootleg	NNP
orr-bootleg-2020	241	42	information	information	NN
orr-bootleg-2020	241	43	,	,	,
orr-bootleg-2020	241	44	the	the	DT
orr-bootleg-2020	241	45	improvement	improvement	NN
orr-bootleg-2020	241	46	our	-PRON-	PRP$
orr-bootleg-2020	241	47	SotA	SotA	NNP
orr-bootleg-2020	241	48	model	model	NN
orr-bootleg-2020	241	49	provides	provide	VBZ
orr-bootleg-2020	241	50	over	over	IN
orr-bootleg-2020	241	51	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	241	52	increases	increase	NNS
orr-bootleg-2020	241	53	(	(	-LRB-
orr-bootleg-2020	241	54	more	more	JJR
orr-bootleg-2020	241	55	details	detail	NNS
orr-bootleg-2020	241	56	in	in	IN
orr-bootleg-2020	241	57	Appendix	Appendix	NNP
orr-bootleg-2020	241	58	C	C	NNP
orr-bootleg-2020	241	59	)	)	-RRB-
orr-bootleg-2020	241	60	.	.	.
orr-bootleg-2020	242	1	Industry	industry	NN
orr-bootleg-2020	242	2	Use	Use	NNP
orr-bootleg-2020	242	3	Case	Case	NNP
orr-bootleg-2020	242	4	We	-PRON-	PRP
orr-bootleg-2020	242	5	additionally	additionally	RB
orr-bootleg-2020	242	6	demonstrate	demonstrate	VBP
orr-bootleg-2020	242	7	how	how	WRB
orr-bootleg-2020	242	8	the	the	DT
orr-bootleg-2020	242	9	learned	learn	VBN
orr-bootleg-2020	242	10	entity	entity	NN
orr-bootleg-2020	242	11	embeddings	embedding	NNS
orr-bootleg-2020	242	12	from	from	IN
orr-bootleg-2020	242	13	Bootleg	Bootleg	NNP
orr-bootleg-2020	242	14	provide	provide	VBP
orr-bootleg-2020	242	15	useful	useful	JJ
orr-bootleg-2020	242	16	information	information	NN
orr-bootleg-2020	242	17	to	to	IN
orr-bootleg-2020	242	18	a	a	DT
orr-bootleg-2020	242	19	system	system	NN
orr-bootleg-2020	242	20	at	at	IN
orr-bootleg-2020	242	21	a	a	DT
orr-bootleg-2020	242	22	large	large	JJ
orr-bootleg-2020	242	23	technology	technology	NN
orr-bootleg-2020	242	24	company	company	NN
orr-bootleg-2020	242	25	that	that	WDT
orr-bootleg-2020	242	26	answers	answer	VBZ
orr-bootleg-2020	242	27	factoid	factoid	JJ
orr-bootleg-2020	242	28	queries	query	NNS
orr-bootleg-2020	242	29	such	such	JJ
orr-bootleg-2020	242	30	as	as	IN
orr-bootleg-2020	242	31	“	"	``
orr-bootleg-2020	242	32	How	how	WRB
orr-bootleg-2020	242	33	tall	tall	JJ
orr-bootleg-2020	242	34	is	be	VBZ
orr-bootleg-2020	242	35	the	the	DT
orr-bootleg-2020	242	36	president	president	NN
orr-bootleg-2020	242	37	of	of	IN
orr-bootleg-2020	242	38	the	the	DT
orr-bootleg-2020	242	39	United	United	NNP
orr-bootleg-2020	242	40	States	States	NNP
orr-bootleg-2020	242	41	?	?	.
orr-bootleg-2020	242	42	"	"	''
orr-bootleg-2020	242	43	.	.	.
orr-bootleg-2020	243	1	We	-PRON-	PRP
orr-bootleg-2020	243	2	use	use	VBP
orr-bootleg-2020	243	3	Bootleg	Bootleg	NNP
orr-bootleg-2020	243	4	’s	’s	POS
orr-bootleg-2020	243	5	embeddings	embedding	NNS
orr-bootleg-2020	243	6	in	in	IN
orr-bootleg-2020	243	7	the	the	DT
orr-bootleg-2020	243	8	Overton	Overton	NNP
orr-bootleg-2020	243	9	[	[	-LRB-
orr-bootleg-2020	243	10	45	45	CD
orr-bootleg-2020	243	11	]	]	-RRB-
orr-bootleg-2020	243	12	system	system	NN
orr-bootleg-2020	243	13	and	and	CC
orr-bootleg-2020	243	14	compare	compare	VB
orr-bootleg-2020	243	15	to	to	IN
orr-bootleg-2020	243	16	the	the	DT
orr-bootleg-2020	243	17	same	same	JJ
orr-bootleg-2020	243	18	system	system	NN
orr-bootleg-2020	243	19	without	without	IN
orr-bootleg-2020	243	20	Bootleg	Bootleg	NNP
orr-bootleg-2020	243	21	embeddings	embedding	NNS
orr-bootleg-2020	243	22	as	as	IN
orr-bootleg-2020	243	23	the	the	DT
orr-bootleg-2020	243	24	baseline	baseline	NN
orr-bootleg-2020	243	25	.	.	.
orr-bootleg-2020	244	1	We	-PRON-	PRP
orr-bootleg-2020	244	2	measure	measure	VBP
orr-bootleg-2020	244	3	the	the	DT
orr-bootleg-2020	244	4	overall	overall	JJ
orr-bootleg-2020	244	5	test	test	NN
orr-bootleg-2020	244	6	quality	quality	NN
orr-bootleg-2020	244	7	(	(	-LRB-
orr-bootleg-2020	244	8	F1	F1	NNP
orr-bootleg-2020	244	9	)	)	-RRB-
orr-bootleg-2020	244	10	on	on	IN
orr-bootleg-2020	244	11	an	an	DT
orr-bootleg-2020	244	12	in	in	IN
orr-bootleg-2020	244	13	-	-	HYPH
orr-bootleg-2020	244	14	house	house	NN
orr-bootleg-2020	244	15	entity	entity	NN
orr-bootleg-2020	244	16	disambiguation	disambiguation	NN
orr-bootleg-2020	244	17	task	task	NN
orr-bootleg-2020	244	18	as	as	RB
orr-bootleg-2020	244	19	well	well	RB
orr-bootleg-2020	244	20	as	as	IN
orr-bootleg-2020	244	21	the	the	DT
orr-bootleg-2020	244	22	quality	quality	NN
orr-bootleg-2020	244	23	over	over	IN
orr-bootleg-2020	244	24	the	the	DT
orr-bootleg-2020	244	25	tail	tail	NN
orr-bootleg-2020	244	26	slices	slice	NNS
orr-bootleg-2020	244	27	which	which	WDT
orr-bootleg-2020	244	28	include	include	VBP
orr-bootleg-2020	244	29	unseen	unseen	JJ
orr-bootleg-2020	244	30	entities	entity	NNS
orr-bootleg-2020	244	31	.	.	.
orr-bootleg-2020	245	1	Per	per	IN
orr-bootleg-2020	245	2	company	company	NN
orr-bootleg-2020	245	3	policy	policy	NN
orr-bootleg-2020	245	4	,	,	,
orr-bootleg-2020	245	5	we	-PRON-	PRP
orr-bootleg-2020	245	6	report	report	VBP
orr-bootleg-2020	245	7	relative	relative	JJ
orr-bootleg-2020	245	8	to	to	IN
orr-bootleg-2020	245	9	the	the	DT
orr-bootleg-2020	245	10	baseline	baseline	NN
orr-bootleg-2020	245	11	rather	rather	RB
orr-bootleg-2020	245	12	than	than	IN
orr-bootleg-2020	245	13	raw	raw	NNP
orr-bootleg-2020	245	14	F1	F1	NNP
orr-bootleg-2020	245	15	score	score	NN
orr-bootleg-2020	245	16	;	;	:
orr-bootleg-2020	245	17	for	for	IN
orr-bootleg-2020	245	18	example	example	NN
orr-bootleg-2020	245	19	,	,	,
orr-bootleg-2020	245	20	if	if	IN
orr-bootleg-2020	245	21	the	the	DT
orr-bootleg-2020	245	22	baseline	baseline	NNP
orr-bootleg-2020	245	23	F1	F1	NNP
orr-bootleg-2020	245	24	score	score	NN
orr-bootleg-2020	245	25	is	be	VBZ
orr-bootleg-2020	245	26	80.0	80.0	CD
orr-bootleg-2020	245	27	and	and	CC
orr-bootleg-2020	245	28	the	the	DT
orr-bootleg-2020	245	29	subject	subject	JJ
orr-bootleg-2020	245	30	F1	F1	NNP
orr-bootleg-2020	245	31	is	be	VBZ
orr-bootleg-2020	245	32	88.0	88.0	CD
orr-bootleg-2020	245	33	,	,	,
orr-bootleg-2020	245	34	the	the	DT
orr-bootleg-2020	245	35	relative	relative	JJ
orr-bootleg-2020	245	36	quality	quality	NN
orr-bootleg-2020	245	37	is	be	VBZ
orr-bootleg-2020	245	38	88.0/80.0	88.0/80.0	NNP
orr-bootleg-2020	245	39	=	=	SYM
orr-bootleg-2020	245	40	1.1	1.1	CD
orr-bootleg-2020	245	41	.	.	.
orr-bootleg-2020	246	1	Table	table	NN
orr-bootleg-2020	246	2	5	5	CD
orr-bootleg-2020	246	3	shows	show	VBZ
orr-bootleg-2020	246	4	that	that	IN
orr-bootleg-2020	246	5	the	the	DT
orr-bootleg-2020	246	6	use	use	NN
orr-bootleg-2020	246	7	of	of	IN
orr-bootleg-2020	246	8	Bootleg	Bootleg	NNP
orr-bootleg-2020	246	9	’s	’s	POS
orr-bootleg-2020	246	10	embeddings	embedding	NNS
orr-bootleg-2020	246	11	consistently	consistently	RB
orr-bootleg-2020	246	12	results	result	VBZ
orr-bootleg-2020	246	13	in	in	IN
orr-bootleg-2020	246	14	a	a	DT
orr-bootleg-2020	246	15	positive	positive	JJ
orr-bootleg-2020	246	16	relative	relative	JJ
orr-bootleg-2020	246	17	quality	quality	NN
orr-bootleg-2020	246	18	,	,	,
orr-bootleg-2020	246	19	even	even	RB
orr-bootleg-2020	246	20	over	over	IN
orr-bootleg-2020	246	21	Spanish	spanish	JJ
orr-bootleg-2020	246	22	,	,	,
orr-bootleg-2020	246	23	French	French	NNP
orr-bootleg-2020	246	24	,	,	,
orr-bootleg-2020	246	25	and	and	CC
orr-bootleg-2020	246	26	German	German	NNP
orr-bootleg-2020	246	27	,	,	,
orr-bootleg-2020	246	28	where	where	WRB
orr-bootleg-2020	246	29	improvements	improvement	NNS
orr-bootleg-2020	246	30	are	be	VBP
orr-bootleg-2020	246	31	most	most	RBS
orr-bootleg-2020	246	32	visible	visible	JJ
orr-bootleg-2020	246	33	over	over	IN
orr-bootleg-2020	246	34	tail	tail	NN
orr-bootleg-2020	246	35	entities	entity	NNS
orr-bootleg-2020	246	36	.	.	.
orr-bootleg-2020	247	1	4.4	4.4	CD
orr-bootleg-2020	247	2	Memory	Memory	NNP
orr-bootleg-2020	247	3	Usage	Usage	NNP
orr-bootleg-2020	247	4	We	-PRON-	PRP
orr-bootleg-2020	247	5	explore	explore	VBP
orr-bootleg-2020	247	6	the	the	DT
orr-bootleg-2020	247	7	memory	memory	NN
orr-bootleg-2020	247	8	usage	usage	NN
orr-bootleg-2020	247	9	of	of	IN
orr-bootleg-2020	247	10	Bootleg	Bootleg	NNP
orr-bootleg-2020	247	11	during	during	IN
orr-bootleg-2020	247	12	inference	inference	NN
orr-bootleg-2020	247	13	and	and	CC
orr-bootleg-2020	247	14	demonstrate	demonstrate	VBP
orr-bootleg-2020	247	15	that	that	IN
orr-bootleg-2020	247	16	by	by	IN
orr-bootleg-2020	247	17	only	only	RB
orr-bootleg-2020	247	18	using	use	VBG
orr-bootleg-2020	247	19	the	the	DT
orr-bootleg-2020	247	20	entity	entity	NN
orr-bootleg-2020	247	21	embeddings	embedding	NNS
orr-bootleg-2020	247	22	for	for	IN
orr-bootleg-2020	247	23	the	the	DT
orr-bootleg-2020	247	24	top	top	JJ
orr-bootleg-2020	247	25	5	5	CD
orr-bootleg-2020	247	26	%	%	NN
orr-bootleg-2020	247	27	of	of	IN
orr-bootleg-2020	247	28	entities	entity	NNS
orr-bootleg-2020	247	29	,	,	,
orr-bootleg-2020	247	30	ranked	rank	VBN
orr-bootleg-2020	247	31	by	by	IN
orr-bootleg-2020	247	32	popularity	popularity	NN
orr-bootleg-2020	247	33	in	in	IN
orr-bootleg-2020	247	34	the	the	DT
orr-bootleg-2020	247	35	training	training	NN
orr-bootleg-2020	247	36	data	datum	NNS
orr-bootleg-2020	247	37	,	,	,
orr-bootleg-2020	247	38	Bootleg	Bootleg	NNP
orr-bootleg-2020	247	39	reduces	reduce	VBZ
orr-bootleg-2020	247	40	its	-PRON-	PRP$
orr-bootleg-2020	247	41	Table	table	NN
orr-bootleg-2020	247	42	5	5	CD
orr-bootleg-2020	247	43	:	:	:
orr-bootleg-2020	247	44	Relative	relative	JJ
orr-bootleg-2020	247	45	F1	f1	NN
orr-bootleg-2020	247	46	quality	quality	NN
orr-bootleg-2020	247	47	of	of	IN
orr-bootleg-2020	247	48	an	an	DT
orr-bootleg-2020	247	49	Overton[45	overton[45	ADD
orr-bootleg-2020	247	50	]	]	-RRB-
orr-bootleg-2020	247	51	model	model	NN
orr-bootleg-2020	247	52	with	with	IN
orr-bootleg-2020	247	53	Bootleg	Bootleg	NNP
orr-bootleg-2020	247	54	embeddings	embedding	NNS
orr-bootleg-2020	247	55	over	over	IN
orr-bootleg-2020	247	56	one	one	CD
orr-bootleg-2020	247	57	without	without	IN
orr-bootleg-2020	247	58	in	in	IN
orr-bootleg-2020	247	59	four	four	CD
orr-bootleg-2020	247	60	languages	language	NNS
orr-bootleg-2020	247	61	.	.	.
orr-bootleg-2020	248	1	Validation	validation	NN
orr-bootleg-2020	248	2	Set	Set	NNP
orr-bootleg-2020	248	3	English	English	NNP
orr-bootleg-2020	248	4	Spanish	Spanish	NNP
orr-bootleg-2020	248	5	French	french	JJ
orr-bootleg-2020	248	6	German	German	NNP
orr-bootleg-2020	248	7	All	all	DT
orr-bootleg-2020	248	8	Entities	entity	NNS
orr-bootleg-2020	248	9	1.08	1.08	CD
orr-bootleg-2020	248	10	1.03	1.03	CD
orr-bootleg-2020	248	11	1.02	1.02	CD
orr-bootleg-2020	248	12	1.00	1.00	CD
orr-bootleg-2020	248	13	Tail	Tail	NNP
orr-bootleg-2020	248	14	Entities	Entities	NNP
orr-bootleg-2020	248	15	1.08	1.08	CD
orr-bootleg-2020	248	16	1.17	1.17	CD
orr-bootleg-2020	248	17	1.05	1.05	CD
orr-bootleg-2020	248	18	1.03	1.03	CD
orr-bootleg-2020	248	19	10	10	CD
orr-bootleg-2020	248	20	All	all	DT
orr-bootleg-2020	248	21	Torso	Torso	NNP
orr-bootleg-2020	248	22	Tail	Tail	NNP
orr-bootleg-2020	248	23	Unseen	Unseen	NNP
orr-bootleg-2020	248	24	F1	F1	NNP
orr-bootleg-2020	248	25	0.6	0.6	CD
orr-bootleg-2020	248	26	0.7	0.7	CD
orr-bootleg-2020	248	27	0.8	0.8	CD
orr-bootleg-2020	248	28	0.9	0.9	CD
orr-bootleg-2020	248	29	1.0	1.0	CD
orr-bootleg-2020	248	30	Compression	Compression	NNP
orr-bootleg-2020	248	31	ratio	ratio	NN
orr-bootleg-2020	248	32	0	0	CD
orr-bootleg-2020	248	33	20	20	CD
orr-bootleg-2020	248	34	40	40	CD
orr-bootleg-2020	248	35	60	60	CD
orr-bootleg-2020	248	36	80	80	CD
orr-bootleg-2020	248	37	100	100	CD
orr-bootleg-2020	248	38	Figure	figure	NN
orr-bootleg-2020	248	39	3	3	CD
orr-bootleg-2020	248	40	:	:	:
orr-bootleg-2020	248	41	We	-PRON-	PRP
orr-bootleg-2020	248	42	show	show	VBP
orr-bootleg-2020	248	43	the	the	DT
orr-bootleg-2020	248	44	error	error	NN
orr-bootleg-2020	248	45	across	across	IN
orr-bootleg-2020	248	46	all	all	DT
orr-bootleg-2020	248	47	entities	entity	NNS
orr-bootleg-2020	248	48	,	,	,
orr-bootleg-2020	248	49	torso	torso	VB
orr-bootleg-2020	248	50	entities	entity	NNS
orr-bootleg-2020	248	51	,	,	,
orr-bootleg-2020	248	52	tail	tail	NN
orr-bootleg-2020	248	53	entities	entity	NNS
orr-bootleg-2020	248	54	,	,	,
orr-bootleg-2020	248	55	and	and	CC
orr-bootleg-2020	248	56	unseen	unseen	JJ
orr-bootleg-2020	248	57	entities	entity	NNS
orr-bootleg-2020	248	58	as	as	IN
orr-bootleg-2020	248	59	we	-PRON-	PRP
orr-bootleg-2020	248	60	decrease	decrease	VBP
orr-bootleg-2020	248	61	the	the	DT
orr-bootleg-2020	248	62	number	number	NN
orr-bootleg-2020	248	63	embeddings	embedding	NNS
orr-bootleg-2020	248	64	we	-PRON-	PRP
orr-bootleg-2020	248	65	use	use	VBP
orr-bootleg-2020	248	66	during	during	IN
orr-bootleg-2020	248	67	inference	inference	NN
orr-bootleg-2020	248	68	,	,	,
orr-bootleg-2020	248	69	assigning	assign	VBG
orr-bootleg-2020	248	70	the	the	DT
orr-bootleg-2020	248	71	non	non	JJ
orr-bootleg-2020	248	72	-	-	JJ
orr-bootleg-2020	248	73	popular	popular	JJ
orr-bootleg-2020	248	74	entities	entity	NNS
orr-bootleg-2020	248	75	to	to	IN
orr-bootleg-2020	248	76	a	a	DT
orr-bootleg-2020	248	77	fixed	fix	VBN
orr-bootleg-2020	248	78	unseen	unseen	JJ
orr-bootleg-2020	248	79	entity	entity	NN
orr-bootleg-2020	248	80	embedding	embedding	NN
orr-bootleg-2020	248	81	.	.	.
orr-bootleg-2020	249	1	For	for	IN
orr-bootleg-2020	249	2	example	example	NN
orr-bootleg-2020	249	3	,	,	,
orr-bootleg-2020	249	4	a	a	DT
orr-bootleg-2020	249	5	compression	compression	NN
orr-bootleg-2020	249	6	ratio	ratio	NN
orr-bootleg-2020	249	7	of	of	IN
orr-bootleg-2020	249	8	80	80	CD
orr-bootleg-2020	249	9	means	mean	NNS
orr-bootleg-2020	249	10	only	only	RB
orr-bootleg-2020	249	11	the	the	DT
orr-bootleg-2020	249	12	top	top	JJ
orr-bootleg-2020	249	13	20	20	CD
orr-bootleg-2020	249	14	%	%	NN
orr-bootleg-2020	249	15	of	of	IN
orr-bootleg-2020	249	16	entity	entity	NN
orr-bootleg-2020	249	17	embeddings	embedding	NNS
orr-bootleg-2020	249	18	are	be	VBP
orr-bootleg-2020	249	19	used	use	VBN
orr-bootleg-2020	249	20	,	,	,
orr-bootleg-2020	249	21	ranked	rank	VBN
orr-bootleg-2020	249	22	by	by	IN
orr-bootleg-2020	249	23	entity	entity	NN
orr-bootleg-2020	249	24	popularity	popularity	NN
orr-bootleg-2020	249	25	.	.	.
orr-bootleg-2020	250	1	embedding	embed	VBG
orr-bootleg-2020	250	2	memory	memory	NN
orr-bootleg-2020	250	3	consumption	consumption	NN
orr-bootleg-2020	250	4	by	by	IN
orr-bootleg-2020	250	5	95	95	CD
orr-bootleg-2020	250	6	%	%	NN
orr-bootleg-2020	250	7	,	,	,
orr-bootleg-2020	250	8	while	while	IN
orr-bootleg-2020	250	9	sacrificing	sacrifice	VBG
orr-bootleg-2020	250	10	only	only	RB
orr-bootleg-2020	250	11	0.8	0.8	CD
orr-bootleg-2020	250	12	F1	f1	NN
orr-bootleg-2020	250	13	points	point	NNS
orr-bootleg-2020	250	14	over	over	IN
orr-bootleg-2020	250	15	all	all	DT
orr-bootleg-2020	250	16	entities	entity	NNS
orr-bootleg-2020	250	17	.	.	.
orr-bootleg-2020	251	1	We	-PRON-	PRP
orr-bootleg-2020	251	2	find	find	VBP
orr-bootleg-2020	251	3	that	that	IN
orr-bootleg-2020	251	4	the	the	DT
orr-bootleg-2020	251	5	5.3	5.3	CD
orr-bootleg-2020	251	6	M	M	NNP
orr-bootleg-2020	251	7	entity	entity	NN
orr-bootleg-2020	251	8	embeddings	embedding	NNS
orr-bootleg-2020	251	9	used	use	VBN
orr-bootleg-2020	251	10	in	in	IN
orr-bootleg-2020	251	11	Bootleg	Bootleg	NNP
orr-bootleg-2020	251	12	consume	consume	VBP
orr-bootleg-2020	251	13	the	the	DT
orr-bootleg-2020	251	14	most	most	JJS
orr-bootleg-2020	251	15	memory	memory	NN
orr-bootleg-2020	251	16	,	,	,
orr-bootleg-2020	251	17	taking	take	VBG
orr-bootleg-2020	251	18	5.2	5.2	CD
orr-bootleg-2020	251	19	GB	GB	NNP
orr-bootleg-2020	251	20	of	of	IN
orr-bootleg-2020	251	21	space	space	NN
orr-bootleg-2020	251	22	while	while	IN
orr-bootleg-2020	251	23	the	the	DT
orr-bootleg-2020	251	24	attention	attention	NN
orr-bootleg-2020	251	25	network	network	NN
orr-bootleg-2020	251	26	only	only	RB
orr-bootleg-2020	251	27	consumes	consume	VBZ
orr-bootleg-2020	251	28	39	39	CD
orr-bootleg-2020	251	29	MB	mb	NN
orr-bootleg-2020	251	30	(	(	-LRB-
orr-bootleg-2020	251	31	1.37B	1.37B	VBN
orr-bootleg-2020	251	32	updated	update	VBN
orr-bootleg-2020	251	33	model	model	NN
orr-bootleg-2020	251	34	parameters	parameter	NNS
orr-bootleg-2020	251	35	in	in	IN
orr-bootleg-2020	251	36	total	total	NN
orr-bootleg-2020	251	37	,	,	,
orr-bootleg-2020	251	38	1.36B	1.36b	CD
orr-bootleg-2020	251	39	from	from	IN
orr-bootleg-2020	251	40	embeddings	embedding	NNS
orr-bootleg-2020	251	41	)	)	-RRB-
orr-bootleg-2020	251	42	.	.	.
orr-bootleg-2020	252	1	As	as	IN
orr-bootleg-2020	252	2	Bootleg	Bootleg	NNP
orr-bootleg-2020	252	3	’s	’s	POS
orr-bootleg-2020	252	4	representations	representation	NNS
orr-bootleg-2020	252	5	must	must	MD
orr-bootleg-2020	252	6	be	be	VB
orr-bootleg-2020	252	7	used	use	VBN
orr-bootleg-2020	252	8	in	in	IN
orr-bootleg-2020	252	9	a	a	DT
orr-bootleg-2020	252	10	variety	variety	NN
orr-bootleg-2020	252	11	of	of	IN
orr-bootleg-2020	252	12	downstream	downstream	JJ
orr-bootleg-2020	252	13	tasks	task	NNS
orr-bootleg-2020	252	14	,	,	,
orr-bootleg-2020	252	15	the	the	DT
orr-bootleg-2020	252	16	representations	representation	NNS
orr-bootleg-2020	252	17	must	must	MD
orr-bootleg-2020	252	18	be	be	VB
orr-bootleg-2020	252	19	memory	memory	NN
orr-bootleg-2020	252	20	-	-	HYPH
orr-bootleg-2020	252	21	efficient	efficient	JJ
orr-bootleg-2020	252	22	:	:	:
orr-bootleg-2020	252	23	we	-PRON-	PRP
orr-bootleg-2020	252	24	thus	thus	RB
orr-bootleg-2020	252	25	study	study	VBP
orr-bootleg-2020	252	26	the	the	DT
orr-bootleg-2020	252	27	effect	effect	NN
orr-bootleg-2020	252	28	of	of	IN
orr-bootleg-2020	252	29	reducing	reduce	VBG
orr-bootleg-2020	252	30	Bootleg	Bootleg	NNP
orr-bootleg-2020	252	31	’s	’s	POS
orr-bootleg-2020	252	32	memory	memory	NN
orr-bootleg-2020	252	33	footprint	footprint	NN
orr-bootleg-2020	252	34	by	by	IN
orr-bootleg-2020	252	35	only	only	RB
orr-bootleg-2020	252	36	using	use	VBG
orr-bootleg-2020	252	37	the	the	DT
orr-bootleg-2020	252	38	most	most	RBS
orr-bootleg-2020	252	39	popular	popular	JJ
orr-bootleg-2020	252	40	entity	entity	NN
orr-bootleg-2020	252	41	embeddings	embedding	NNS
orr-bootleg-2020	252	42	.	.	.
orr-bootleg-2020	253	1	Specifically	specifically	RB
orr-bootleg-2020	253	2	,	,	,
orr-bootleg-2020	253	3	for	for	IN
orr-bootleg-2020	253	4	the	the	DT
orr-bootleg-2020	253	5	top	top	JJ
orr-bootleg-2020	253	6	k%	k%	NN
orr-bootleg-2020	253	7	of	of	IN
orr-bootleg-2020	253	8	entities	entity	NNS
orr-bootleg-2020	253	9	ranked	rank	VBN
orr-bootleg-2020	253	10	by	by	IN
orr-bootleg-2020	253	11	the	the	DT
orr-bootleg-2020	253	12	number	number	NN
orr-bootleg-2020	253	13	of	of	IN
orr-bootleg-2020	253	14	occurrences	occurrence	NNS
orr-bootleg-2020	253	15	in	in	IN
orr-bootleg-2020	253	16	training	training	NN
orr-bootleg-2020	253	17	data	datum	NNS
orr-bootleg-2020	253	18	,	,	,
orr-bootleg-2020	253	19	we	-PRON-	PRP
orr-bootleg-2020	253	20	keep	keep	VBP
orr-bootleg-2020	253	21	the	the	DT
orr-bootleg-2020	253	22	learned	learn	VBN
orr-bootleg-2020	253	23	entity	entity	NN
orr-bootleg-2020	253	24	embedding	embed	VBG
orr-bootleg-2020	253	25	intact	intact	JJ
orr-bootleg-2020	253	26	.	.	.
orr-bootleg-2020	254	1	For	for	IN
orr-bootleg-2020	254	2	the	the	DT
orr-bootleg-2020	254	3	remaining	remain	VBG
orr-bootleg-2020	254	4	entities	entity	NNS
orr-bootleg-2020	254	5	,	,	,
orr-bootleg-2020	254	6	we	-PRON-	PRP
orr-bootleg-2020	254	7	choose	choose	VBP
orr-bootleg-2020	254	8	a	a	DT
orr-bootleg-2020	254	9	random	random	JJ
orr-bootleg-2020	254	10	entity	entity	NN
orr-bootleg-2020	254	11	embedding	embedding	NN
orr-bootleg-2020	254	12	for	for	IN
orr-bootleg-2020	254	13	an	an	DT
orr-bootleg-2020	254	14	unseen	unseen	JJ
orr-bootleg-2020	254	15	entity	entity	NN
orr-bootleg-2020	254	16	to	to	TO
orr-bootleg-2020	254	17	use	use	VB
orr-bootleg-2020	254	18	instead	instead	RB
orr-bootleg-2020	254	19	.	.	.
orr-bootleg-2020	255	1	Instead	instead	RB
orr-bootleg-2020	255	2	of	of	IN
orr-bootleg-2020	255	3	storing	store	VBG
orr-bootleg-2020	255	4	5.3	5.3	CD
orr-bootleg-2020	255	5	M	M	NNP
orr-bootleg-2020	255	6	entity	entity	NN
orr-bootleg-2020	255	7	embeddings	embedding	NNS
orr-bootleg-2020	255	8	,	,	,
orr-bootleg-2020	255	9	we	-PRON-	PRP
orr-bootleg-2020	255	10	thus	thus	RB
orr-bootleg-2020	255	11	store	store	VBP
orr-bootleg-2020	255	12	(	(	-LRB-
orr-bootleg-2020	255	13	(	(	-LRB-
orr-bootleg-2020	255	14	100−k)/100)∗5.3	100−k)/100)∗5.3	CD
orr-bootleg-2020	255	15	M	M	NNP
orr-bootleg-2020	255	16	,	,	,
orr-bootleg-2020	255	17	which	which	WDT
orr-bootleg-2020	255	18	gives	give	VBZ
orr-bootleg-2020	255	19	a	a	DT
orr-bootleg-2020	255	20	compression	compression	NN
orr-bootleg-2020	255	21	ratio	ratio	NN
orr-bootleg-2020	255	22	of	of	IN
orr-bootleg-2020	255	23	(	(	-LRB-
orr-bootleg-2020	255	24	100	100	CD
orr-bootleg-2020	255	25	−k	−k	CD
orr-bootleg-2020	255	26	)	)	-RRB-
orr-bootleg-2020	255	27	.	.	.
orr-bootleg-2020	256	1	Figure	figure	NN
orr-bootleg-2020	256	2	3	3	CD
orr-bootleg-2020	256	3	shows	show	VBZ
orr-bootleg-2020	256	4	performance	performance	NN
orr-bootleg-2020	256	5	for	for	IN
orr-bootleg-2020	256	6	k	k	NNP
orr-bootleg-2020	256	7	of	of	IN
orr-bootleg-2020	256	8	100	100	CD
orr-bootleg-2020	256	9	,	,	,
orr-bootleg-2020	256	10	50	50	CD
orr-bootleg-2020	256	11	,	,	,
orr-bootleg-2020	256	12	20	20	CD
orr-bootleg-2020	256	13	,	,	,
orr-bootleg-2020	256	14	10	10	CD
orr-bootleg-2020	256	15	,	,	,
orr-bootleg-2020	256	16	5	5	CD
orr-bootleg-2020	256	17	,	,	,
orr-bootleg-2020	256	18	1	1	CD
orr-bootleg-2020	256	19	,	,	,
orr-bootleg-2020	256	20	and	and	CC
orr-bootleg-2020	256	21	0.1	0.1	CD
orr-bootleg-2020	256	22	.	.	.
orr-bootleg-2020	257	1	We	-PRON-	PRP
orr-bootleg-2020	257	2	see	see	VBP
orr-bootleg-2020	257	3	that	that	IN
orr-bootleg-2020	257	4	when	when	WRB
orr-bootleg-2020	257	5	just	just	RB
orr-bootleg-2020	257	6	the	the	DT
orr-bootleg-2020	257	7	top	top	JJ
orr-bootleg-2020	257	8	5	5	CD
orr-bootleg-2020	257	9	%	%	NN
orr-bootleg-2020	257	10	of	of	IN
orr-bootleg-2020	257	11	entity	entity	NN
orr-bootleg-2020	257	12	embeddings	embedding	NNS
orr-bootleg-2020	257	13	are	be	VBP
orr-bootleg-2020	257	14	used	use	VBN
orr-bootleg-2020	257	15	,	,	,
orr-bootleg-2020	257	16	we	-PRON-	PRP
orr-bootleg-2020	257	17	only	only	RB
orr-bootleg-2020	257	18	sacrifice	sacrifice	VBP
orr-bootleg-2020	257	19	0.8	0.8	CD
orr-bootleg-2020	257	20	F1	f1	NN
orr-bootleg-2020	257	21	points	point	NNS
orr-bootleg-2020	257	22	overall	overall	RB
orr-bootleg-2020	257	23	and	and	CC
orr-bootleg-2020	257	24	in	in	IN
orr-bootleg-2020	257	25	fact	fact	NN
orr-bootleg-2020	257	26	score	score	NN
orr-bootleg-2020	257	27	2	2	CD
orr-bootleg-2020	257	28	F1	f1	NN
orr-bootleg-2020	257	29	points	point	NNS
orr-bootleg-2020	257	30	higher	high	JJR
orr-bootleg-2020	257	31	over	over	IN
orr-bootleg-2020	257	32	the	the	DT
orr-bootleg-2020	257	33	tail	tail	NN
orr-bootleg-2020	257	34	.	.	.
orr-bootleg-2020	258	1	We	-PRON-	PRP
orr-bootleg-2020	258	2	hypothesize	hypothesize	VBP
orr-bootleg-2020	258	3	that	that	IN
orr-bootleg-2020	258	4	the	the	DT
orr-bootleg-2020	258	5	increase	increase	NN
orr-bootleg-2020	258	6	in	in	IN
orr-bootleg-2020	258	7	tail	tail	NN
orr-bootleg-2020	258	8	performance	performance	NN
orr-bootleg-2020	258	9	is	be	VBZ
orr-bootleg-2020	258	10	due	due	JJ
orr-bootleg-2020	258	11	to	to	IN
orr-bootleg-2020	258	12	the	the	DT
orr-bootleg-2020	258	13	fact	fact	NN
orr-bootleg-2020	258	14	that	that	IN
orr-bootleg-2020	258	15	the	the	DT
orr-bootleg-2020	258	16	majority	majority	NN
orr-bootleg-2020	258	17	of	of	IN
orr-bootleg-2020	258	18	mention	mention	NN
orr-bootleg-2020	258	19	candidates	candidate	NNS
orr-bootleg-2020	258	20	all	all	DT
orr-bootleg-2020	258	21	have	have	VBP
orr-bootleg-2020	258	22	the	the	DT
orr-bootleg-2020	258	23	same	same	JJ
orr-bootleg-2020	258	24	learned	learned	JJ
orr-bootleg-2020	258	25	embedding	embedding	NN
orr-bootleg-2020	258	26	,	,	,
orr-bootleg-2020	258	27	decreasing	decrease	VBG
orr-bootleg-2020	258	28	the	the	DT
orr-bootleg-2020	258	29	amount	amount	NN
orr-bootleg-2020	258	30	of	of	IN
orr-bootleg-2020	258	31	conflict	conflict	NN
orr-bootleg-2020	258	32	among	among	IN
orr-bootleg-2020	258	33	candidates	candidate	NNS
orr-bootleg-2020	258	34	from	from	IN
orr-bootleg-2020	258	35	textual	textual	JJ
orr-bootleg-2020	258	36	patterns	pattern	NNS
orr-bootleg-2020	258	37	.	.	.
orr-bootleg-2020	259	1	4.5	4.5	CD
orr-bootleg-2020	259	2	Ablation	Ablation	NNP
orr-bootleg-2020	259	3	Study	Study	NNP
orr-bootleg-2020	259	4	Bootleg	Bootleg	NNP
orr-bootleg-2020	259	5	To	to	TO
orr-bootleg-2020	259	6	better	well	RBR
orr-bootleg-2020	259	7	understand	understand	VB
orr-bootleg-2020	259	8	the	the	DT
orr-bootleg-2020	259	9	performance	performance	NN
orr-bootleg-2020	259	10	gains	gain	NNS
orr-bootleg-2020	259	11	of	of	IN
orr-bootleg-2020	259	12	Bootleg	Bootleg	NNP
orr-bootleg-2020	259	13	,	,	,
orr-bootleg-2020	259	14	we	-PRON-	PRP
orr-bootleg-2020	259	15	perform	perform	VBP
orr-bootleg-2020	259	16	an	an	DT
orr-bootleg-2020	259	17	ablation	ablation	NN
orr-bootleg-2020	259	18	study	study	NN
orr-bootleg-2020	259	19	over	over	IN
orr-bootleg-2020	259	20	a	a	DT
orr-bootleg-2020	259	21	subset	subset	NN
orr-bootleg-2020	259	22	of	of	IN
orr-bootleg-2020	259	23	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	259	24	(	(	-LRB-
orr-bootleg-2020	259	25	data	datum	NNS
orr-bootleg-2020	259	26	details	detail	NNS
orr-bootleg-2020	259	27	explained	explain	VBD
orr-bootleg-2020	259	28	in	in	IN
orr-bootleg-2020	259	29	Appendix	Appendix	NNP
orr-bootleg-2020	259	30	B	B	NNP
orr-bootleg-2020	259	31	)	)	-RRB-
orr-bootleg-2020	259	32	.	.	.
orr-bootleg-2020	260	1	We	-PRON-	PRP
orr-bootleg-2020	260	2	train	train	VBP
orr-bootleg-2020	260	3	Bootleg	Bootleg	NNP
orr-bootleg-2020	260	4	with	with	IN
orr-bootleg-2020	260	5	:	:	:
orr-bootleg-2020	260	6	(	(	-LRB-
orr-bootleg-2020	260	7	1	1	LS
orr-bootleg-2020	260	8	)	)	-RRB-
orr-bootleg-2020	260	9	only	only	RB
orr-bootleg-2020	260	10	learned	learn	VBD
orr-bootleg-2020	260	11	entity	entity	NN
orr-bootleg-2020	260	12	embeddings	embedding	NNS
orr-bootleg-2020	260	13	(	(	-LRB-
orr-bootleg-2020	260	14	Ent	ent	NN
orr-bootleg-2020	260	15	-	-	,
orr-bootleg-2020	260	16	only	only	RB
orr-bootleg-2020	260	17	)	)	-RRB-
orr-bootleg-2020	260	18	,	,	,
orr-bootleg-2020	260	19	(	(	-LRB-
orr-bootleg-2020	260	20	2	2	LS
orr-bootleg-2020	260	21	)	)	-RRB-
orr-bootleg-2020	260	22	only	only	JJ
orr-bootleg-2020	260	23	type	type	NN
orr-bootleg-2020	260	24	information	information	NN
orr-bootleg-2020	260	25	from	from	IN
orr-bootleg-2020	260	26	type	type	NN
orr-bootleg-2020	260	27	embeddings	embedding	NNS
orr-bootleg-2020	260	28	(	(	-LRB-
orr-bootleg-2020	260	29	Type	type	NN
orr-bootleg-2020	260	30	-	-	HYPH
orr-bootleg-2020	260	31	only	only	RB
orr-bootleg-2020	260	32	)	)	-RRB-
orr-bootleg-2020	260	33	,	,	,
orr-bootleg-2020	260	34	and	and	CC
orr-bootleg-2020	260	35	(	(	-LRB-
orr-bootleg-2020	260	36	3	3	LS
orr-bootleg-2020	260	37	)	)	-RRB-
orr-bootleg-2020	260	38	only	only	JJ
orr-bootleg-2020	260	39	knowledge	knowledge	NN
orr-bootleg-2020	260	40	graph	graph	NN
orr-bootleg-2020	260	41	information	information	NN
orr-bootleg-2020	260	42	from	from	IN
orr-bootleg-2020	260	43	relation	relation	NN
orr-bootleg-2020	260	44	embeddings	embedding	NNS
orr-bootleg-2020	260	45	and	and	CC
orr-bootleg-2020	260	46	knowledge	knowledge	NN
orr-bootleg-2020	260	47	graph	graph	NN
orr-bootleg-2020	260	48	connections	connection	NNS
orr-bootleg-2020	260	49	(	(	-LRB-
orr-bootleg-2020	260	50	KG	kg	NN
orr-bootleg-2020	260	51	-	-	HYPH
orr-bootleg-2020	260	52	only	only	RB
orr-bootleg-2020	260	53	)	)	-RRB-
orr-bootleg-2020	260	54	.	.	.
orr-bootleg-2020	261	1	All	all	DT
orr-bootleg-2020	261	2	model	model	NN
orr-bootleg-2020	261	3	sizes	size	NNS
orr-bootleg-2020	261	4	are	be	VBP
orr-bootleg-2020	261	5	reported	report	VBN
orr-bootleg-2020	261	6	in	in	IN
orr-bootleg-2020	261	7	Appendix	Appendix	NNP
orr-bootleg-2020	261	8	B	B	NNP
orr-bootleg-2020	261	9	Table	Table	NNP
orr-bootleg-2020	261	10	10	10	CD
orr-bootleg-2020	261	11	.	.	.
orr-bootleg-2020	262	1	In	in	IN
orr-bootleg-2020	262	2	Table	table	NN
orr-bootleg-2020	262	3	2	2	CD
orr-bootleg-2020	262	4	,	,	,
orr-bootleg-2020	262	5	we	-PRON-	PRP
orr-bootleg-2020	262	6	see	see	VBP
orr-bootleg-2020	262	7	that	that	IN
orr-bootleg-2020	262	8	just	just	RB
orr-bootleg-2020	262	9	using	use	VBG
orr-bootleg-2020	262	10	type	type	NN
orr-bootleg-2020	262	11	or	or	CC
orr-bootleg-2020	262	12	knowledge	knowledge	NN
orr-bootleg-2020	262	13	graph	graph	NN
orr-bootleg-2020	262	14	information	information	NN
orr-bootleg-2020	262	15	leads	lead	VBZ
orr-bootleg-2020	262	16	to	to	IN
orr-bootleg-2020	262	17	improvements	improvement	NNS
orr-bootleg-2020	262	18	on	on	IN
orr-bootleg-2020	262	19	the	the	DT
orr-bootleg-2020	262	20	tail	tail	NN
orr-bootleg-2020	262	21	of	of	IN
orr-bootleg-2020	262	22	over	over	IN
orr-bootleg-2020	262	23	25	25	CD
orr-bootleg-2020	262	24	F1	f1	NN
orr-bootleg-2020	262	25	points	point	NNS
orr-bootleg-2020	262	26	and	and	CC
orr-bootleg-2020	262	27	on	on	IN
orr-bootleg-2020	262	28	the	the	DT
orr-bootleg-2020	262	29	unseen	unseen	JJ
orr-bootleg-2020	262	30	entities	entity	NNS
orr-bootleg-2020	262	31	of	of	IN
orr-bootleg-2020	262	32	over	over	IN
orr-bootleg-2020	262	33	46	46	CD
orr-bootleg-2020	262	34	F1	f1	NN
orr-bootleg-2020	262	35	points	point	NNS
orr-bootleg-2020	262	36	compared	compare	VBN
orr-bootleg-2020	262	37	to	to	IN
orr-bootleg-2020	262	38	the	the	DT
orr-bootleg-2020	262	39	Ent	ent	NN
orr-bootleg-2020	262	40	-	-	HYPH
orr-bootleg-2020	262	41	only	only	JJ
orr-bootleg-2020	262	42	model	model	NN
orr-bootleg-2020	262	43	.	.	.
orr-bootleg-2020	263	1	However	however	RB
orr-bootleg-2020	263	2	,	,	,
orr-bootleg-2020	263	3	neither	neither	CC
orr-bootleg-2020	263	4	the	the	DT
orr-bootleg-2020	263	5	Type	type	NN
orr-bootleg-2020	263	6	-	-	HYPH
orr-bootleg-2020	263	7	only	only	NN
orr-bootleg-2020	263	8	nor	nor	CC
orr-bootleg-2020	263	9	KG	kg	NN
orr-bootleg-2020	263	10	-	-	HYPH
orr-bootleg-2020	263	11	only	only	RB
orr-bootleg-2020	263	12	model	model	NN
orr-bootleg-2020	263	13	performs	perform	VBZ
orr-bootleg-2020	263	14	as	as	RB
orr-bootleg-2020	263	15	well	well	RB
orr-bootleg-2020	263	16	on	on	IN
orr-bootleg-2020	263	17	any	any	DT
orr-bootleg-2020	263	18	of	of	IN
orr-bootleg-2020	263	19	the	the	DT
orr-bootleg-2020	263	20	validation	validation	NN
orr-bootleg-2020	263	21	sets	set	NNS
orr-bootleg-2020	263	22	as	as	IN
orr-bootleg-2020	263	23	the	the	DT
orr-bootleg-2020	263	24	full	full	JJ
orr-bootleg-2020	263	25	Bootleg	Bootleg	NNP
orr-bootleg-2020	263	26	model	model	NN
orr-bootleg-2020	263	27	.	.	.
orr-bootleg-2020	264	1	An	an	DT
orr-bootleg-2020	264	2	interesting	interesting	JJ
orr-bootleg-2020	264	3	comparison	comparison	NN
orr-bootleg-2020	264	4	is	be	VBZ
orr-bootleg-2020	264	5	between	between	IN
orr-bootleg-2020	264	6	Ent	ent	NN
orr-bootleg-2020	264	7	-	-	,
orr-bootleg-2020	264	8	only	only	RB
orr-bootleg-2020	264	9	and	and	CC
orr-bootleg-2020	264	10	NED	NED	NNP
orr-bootleg-2020	264	11	-	-	HYPH
orr-bootleg-2020	264	12	Base	Base	NNP
orr-bootleg-2020	264	13	.	.	.
orr-bootleg-2020	265	1	NED	NED	NNP
orr-bootleg-2020	265	2	-	-	HYPH
orr-bootleg-2020	265	3	Base	base	JJ
orr-bootleg-2020	265	4	overall	overall	RB
orr-bootleg-2020	265	5	outperforms	outperform	VBZ
orr-bootleg-2020	265	6	Ent	ent	NN
orr-bootleg-2020	265	7	-	-	,
orr-bootleg-2020	265	8	only	only	RB
orr-bootleg-2020	265	9	due	due	IN
orr-bootleg-2020	265	10	to	to	IN
orr-bootleg-2020	265	11	the	the	DT
orr-bootleg-2020	265	12	fine	fine	NN
orr-bootleg-2020	265	13	-	-	HYPH
orr-bootleg-2020	265	14	tuning	tuning	NN
orr-bootleg-2020	265	15	of	of	IN
orr-bootleg-2020	265	16	BERT	BERT	NNP
orr-bootleg-2020	265	17	word	word	NN
orr-bootleg-2020	265	18	embeddings	embedding	NNS
orr-bootleg-2020	265	19	.	.	.
orr-bootleg-2020	266	1	We	-PRON-	PRP
orr-bootleg-2020	266	2	attribute	attribute	VBP
orr-bootleg-2020	266	3	the	the	DT
orr-bootleg-2020	266	4	high	high	JJ
orr-bootleg-2020	266	5	performance	performance	NN
orr-bootleg-2020	266	6	of	of	IN
orr-bootleg-2020	266	7	Ent	ent	NN
orr-bootleg-2020	266	8	-	-	,
orr-bootleg-2020	266	9	only	only	RB
orr-bootleg-2020	266	10	on	on	IN
orr-bootleg-2020	266	11	the	the	DT
orr-bootleg-2020	266	12	tail	tail	NN
orr-bootleg-2020	266	13	compared	compare	VBN
orr-bootleg-2020	266	14	to	to	IN
orr-bootleg-2020	266	15	NED	NED	NNP
orr-bootleg-2020	266	16	-	-	HYPH
orr-bootleg-2020	266	17	Base	Base	NNP
orr-bootleg-2020	266	18	to	to	IN
orr-bootleg-2020	266	19	our	-PRON-	PRP$
orr-bootleg-2020	266	20	Ent2Ent	Ent2Ent	NNP
orr-bootleg-2020	266	21	module	module	NN
orr-bootleg-2020	266	22	which	which	WDT
orr-bootleg-2020	266	23	allows	allow	VBZ
orr-bootleg-2020	266	24	for	for	IN
orr-bootleg-2020	266	25	memorizing	memorize	VBG
orr-bootleg-2020	266	26	co	co	JJ
orr-bootleg-2020	266	27	-	-	JJ
orr-bootleg-2020	266	28	occurrence	occurrence	JJ
orr-bootleg-2020	266	29	patterns	pattern	NNS
orr-bootleg-2020	266	30	over	over	IN
orr-bootleg-2020	266	31	entities	entity	NNS
orr-bootleg-2020	266	32	.	.	.
orr-bootleg-2020	267	1	Regularization	regularization	NN
orr-bootleg-2020	267	2	To	to	TO
orr-bootleg-2020	267	3	understand	understand	VB
orr-bootleg-2020	267	4	the	the	DT
orr-bootleg-2020	267	5	impact	impact	NN
orr-bootleg-2020	267	6	of	of	IN
orr-bootleg-2020	267	7	our	-PRON-	PRP$
orr-bootleg-2020	267	8	entity	entity	NN
orr-bootleg-2020	267	9	regularization	regularization	NN
orr-bootleg-2020	267	10	function	function	NN
orr-bootleg-2020	267	11	p(e	p(e	NNP
orr-bootleg-2020	267	12	)	)	-RRB-
orr-bootleg-2020	267	13	on	on	IN
orr-bootleg-2020	267	14	overall	overall	JJ
orr-bootleg-2020	267	15	performance	performance	NN
orr-bootleg-2020	267	16	,	,	,
orr-bootleg-2020	267	17	we	-PRON-	PRP
orr-bootleg-2020	267	18	perform	perform	VBP
orr-bootleg-2020	267	19	an	an	DT
orr-bootleg-2020	267	20	ablation	ablation	NN
orr-bootleg-2020	267	21	study	study	NN
orr-bootleg-2020	267	22	on	on	IN
orr-bootleg-2020	267	23	a	a	DT
orr-bootleg-2020	267	24	sample	sample	NN
orr-bootleg-2020	267	25	of	of	IN
orr-bootleg-2020	267	26	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	267	27	(	(	-LRB-
orr-bootleg-2020	267	28	explained	explain	VBN
orr-bootleg-2020	267	29	in	in	IN
orr-bootleg-2020	267	30	Appendix	Appendix	NNP
orr-bootleg-2020	267	31	B	B	NNP
orr-bootleg-2020	267	32	)	)	-RRB-
orr-bootleg-2020	267	33	.	.	.
orr-bootleg-2020	268	1	We	-PRON-	PRP
orr-bootleg-2020	268	2	apply	apply	VBP
orr-bootleg-2020	268	3	(	(	-LRB-
orr-bootleg-2020	268	4	1	1	LS
orr-bootleg-2020	268	5	)	)	-RRB-
orr-bootleg-2020	268	6	a	a	DT
orr-bootleg-2020	268	7	fixed	fix	VBN
orr-bootleg-2020	268	8	regularization	regularization	NN
orr-bootleg-2020	268	9	set	set	VBN
orr-bootleg-2020	268	10	to	to	IN
orr-bootleg-2020	268	11	a	a	DT
orr-bootleg-2020	268	12	constant	constant	JJ
orr-bootleg-2020	268	13	percent	percent	NN
orr-bootleg-2020	268	14	of	of	IN
orr-bootleg-2020	268	15	0	0	CD
orr-bootleg-2020	268	16	,	,	,
orr-bootleg-2020	268	17	20	20	CD
orr-bootleg-2020	268	18	,	,	,
orr-bootleg-2020	268	19	50	50	CD
orr-bootleg-2020	268	20	and	and	CC
orr-bootleg-2020	268	21	80	80	CD
orr-bootleg-2020	268	22	,	,	,
orr-bootleg-2020	268	23	(	(	-LRB-
orr-bootleg-2020	268	24	2	2	LS
orr-bootleg-2020	268	25	)	)	-RRB-
orr-bootleg-2020	268	26	a	a	DT
orr-bootleg-2020	268	27	regularization	regularization	NN
orr-bootleg-2020	268	28	function	function	NN
orr-bootleg-2020	268	29	proportional	proportional	JJ
orr-bootleg-2020	268	30	to	to	IN
orr-bootleg-2020	268	31	the	the	DT
orr-bootleg-2020	268	32	power	power	NN
orr-bootleg-2020	268	33	of	of	IN
orr-bootleg-2020	268	34	the	the	DT
orr-bootleg-2020	268	35	inverse	inverse	NN
orr-bootleg-2020	268	36	popularity	popularity	NN
orr-bootleg-2020	268	37	,	,	,
orr-bootleg-2020	268	38	and	and	CC
orr-bootleg-2020	268	39	(	(	-LRB-
orr-bootleg-2020	268	40	3	3	LS
orr-bootleg-2020	268	41	)	)	-RRB-
orr-bootleg-2020	268	42	the	the	DT
orr-bootleg-2020	268	43	inverse	inverse	NN
orr-bootleg-2020	268	44	of	of	IN
orr-bootleg-2020	268	45	(	(	-LRB-
orr-bootleg-2020	268	46	2	2	CD
orr-bootleg-2020	268	47	)	)	-RRB-
orr-bootleg-2020	268	48	.	.	.
orr-bootleg-2020	269	1	Table	table	NN
orr-bootleg-2020	269	2	6	6	CD
orr-bootleg-2020	269	3	shows	show	VBZ
orr-bootleg-2020	269	4	results	result	NNS
orr-bootleg-2020	269	5	over	over	IN
orr-bootleg-2020	269	6	unseen	unseen	JJ
orr-bootleg-2020	269	7	entities	entity	NNS
orr-bootleg-2020	269	8	(	(	-LRB-
orr-bootleg-2020	269	9	full	full	JJ
orr-bootleg-2020	269	10	results	result	NNS
orr-bootleg-2020	269	11	and	and	CC
orr-bootleg-2020	269	12	details	detail	NNS
orr-bootleg-2020	269	13	in	in	IN
orr-bootleg-2020	269	14	Appendix	Appendix	NNP
orr-bootleg-2020	269	15	B	B	NNP
orr-bootleg-2020	269	16	)	)	-RRB-
orr-bootleg-2020	269	17	.	.	.
orr-bootleg-2020	270	1	We	-PRON-	PRP
orr-bootleg-2020	270	2	see	see	VBP
orr-bootleg-2020	270	3	that	that	IN
orr-bootleg-2020	270	4	the	the	DT
orr-bootleg-2020	270	5	fixed	fix	VBN
orr-bootleg-2020	270	6	regularization	regularization	NN
orr-bootleg-2020	270	7	of	of	IN
orr-bootleg-2020	270	8	80	80	CD
orr-bootleg-2020	270	9	%	%	NN
orr-bootleg-2020	270	10	achieves	achieve	VBZ
orr-bootleg-2020	270	11	the	the	DT
orr-bootleg-2020	270	12	highest	high	JJS
orr-bootleg-2020	270	13	F1	f1	NN
orr-bootleg-2020	270	14	over	over	IN
orr-bootleg-2020	270	15	the	the	DT
orr-bootleg-2020	270	16	fixed	fix	VBN
orr-bootleg-2020	270	17	regularizations	regularization	NNS
orr-bootleg-2020	270	18	of	of	IN
orr-bootleg-2020	270	19	(	(	-LRB-
orr-bootleg-2020	270	20	1	1	CD
orr-bootleg-2020	270	21	)	)	-RRB-
orr-bootleg-2020	270	22	.	.	.
orr-bootleg-2020	271	1	The	the	DT
orr-bootleg-2020	271	2	method	method	NN
orr-bootleg-2020	271	3	that	that	WDT
orr-bootleg-2020	271	4	regularizes	regularize	VBZ
orr-bootleg-2020	271	5	by	by	IN
orr-bootleg-2020	271	6	inverse	inverse	NN
orr-bootleg-2020	271	7	popularity	popularity	NN
orr-bootleg-2020	271	8	achieves	achieve	VBZ
orr-bootleg-2020	271	9	the	the	DT
orr-bootleg-2020	271	10	highest	high	JJS
orr-bootleg-2020	271	11	11	11	CD
orr-bootleg-2020	271	12	Table	table	NN
orr-bootleg-2020	271	13	6	6	CD
orr-bootleg-2020	271	14	:	:	:
orr-bootleg-2020	271	15	We	-PRON-	PRP
orr-bootleg-2020	271	16	show	show	VBP
orr-bootleg-2020	271	17	the	the	DT
orr-bootleg-2020	271	18	micro	micro	NNP
orr-bootleg-2020	271	19	F1	F1	NNP
orr-bootleg-2020	271	20	score	score	NN
orr-bootleg-2020	271	21	over	over	IN
orr-bootleg-2020	271	22	unseen	unseen	JJ
orr-bootleg-2020	271	23	entities	entity	NNS
orr-bootleg-2020	271	24	for	for	IN
orr-bootleg-2020	271	25	a	a	DT
orr-bootleg-2020	271	26	Wikipedia	wikipedia	JJ
orr-bootleg-2020	271	27	sample	sample	NN
orr-bootleg-2020	271	28	as	as	IN
orr-bootleg-2020	271	29	we	-PRON-	PRP
orr-bootleg-2020	271	30	vary	vary	VBP
orr-bootleg-2020	271	31	the	the	DT
orr-bootleg-2020	271	32	entity	entity	NN
orr-bootleg-2020	271	33	regularization	regularization	NN
orr-bootleg-2020	271	34	scheme	scheme	NN
orr-bootleg-2020	271	35	p(e	p(e	NNP
orr-bootleg-2020	271	36	)	)	-RRB-
orr-bootleg-2020	271	37	.	.	.
orr-bootleg-2020	272	1	A	a	DT
orr-bootleg-2020	272	2	scalar	scalar	JJ
orr-bootleg-2020	272	3	percent	percent	NN
orr-bootleg-2020	272	4	means	mean	VBZ
orr-bootleg-2020	272	5	a	a	DT
orr-bootleg-2020	272	6	fixed	fix	VBN
orr-bootleg-2020	272	7	regularization	regularization	NN
orr-bootleg-2020	272	8	.	.	.
orr-bootleg-2020	273	1	InvPop	InvPop	NNP
orr-bootleg-2020	273	2	(	(	-LRB-
orr-bootleg-2020	273	3	inverse	inverse	NNP
orr-bootleg-2020	273	4	poularity	poularity	NN
orr-bootleg-2020	273	5	scheme	scheme	NN
orr-bootleg-2020	273	6	)	)	-RRB-
orr-bootleg-2020	273	7	applies	apply	VBZ
orr-bootleg-2020	273	8	less	less	JJR
orr-bootleg-2020	273	9	regularization	regularization	NN
orr-bootleg-2020	273	10	for	for	IN
orr-bootleg-2020	273	11	more	more	RBR
orr-bootleg-2020	273	12	popular	popular	JJ
orr-bootleg-2020	273	13	entities	entity	NNS
orr-bootleg-2020	273	14	and	and	CC
orr-bootleg-2020	273	15	Pop	Pop	NNP
orr-bootleg-2020	273	16	applies	apply	VBZ
orr-bootleg-2020	273	17	more	more	JJR
orr-bootleg-2020	273	18	regularization	regularization	NN
orr-bootleg-2020	273	19	for	for	IN
orr-bootleg-2020	273	20	more	more	RBR
orr-bootleg-2020	273	21	popular	popular	JJ
orr-bootleg-2020	273	22	entities	entity	NNS
orr-bootleg-2020	273	23	.	.	.
orr-bootleg-2020	274	1	p(e	p(e	NNP
orr-bootleg-2020	274	2	)	)	-RRB-
orr-bootleg-2020	274	3	0	0	CD
orr-bootleg-2020	274	4	%	%	NN
orr-bootleg-2020	274	5	20	20	CD
orr-bootleg-2020	274	6	%	%	NN
orr-bootleg-2020	274	7	50	50	CD
orr-bootleg-2020	274	8	%	%	NN
orr-bootleg-2020	274	9	80	80	CD
orr-bootleg-2020	274	10	%	%	NN
orr-bootleg-2020	274	11	Pop	pop	NN
orr-bootleg-2020	274	12	InvPop	InvPop	NNP
orr-bootleg-2020	274	13	Unseen	Unseen	NNP
orr-bootleg-2020	274	14	Entities	Entities	NNPS
orr-bootleg-2020	274	15	48.6	48.6	CD
orr-bootleg-2020	274	16	52.5	52.5	CD
orr-bootleg-2020	274	17	57.7	57.7	CD
orr-bootleg-2020	274	18	59.9	59.9	CD
orr-bootleg-2020	274	19	52.4	52.4	CD
orr-bootleg-2020	274	20	62.2	62.2	CD
orr-bootleg-2020	274	21	overall	overall	JJ
orr-bootleg-2020	274	22	F1	F1	NNP
orr-bootleg-2020	274	23	.	.	.
orr-bootleg-2020	275	1	We	-PRON-	PRP
orr-bootleg-2020	275	2	further	further	RB
orr-bootleg-2020	275	3	see	see	VBP
orr-bootleg-2020	275	4	that	that	IN
orr-bootleg-2020	275	5	the	the	DT
orr-bootleg-2020	275	6	scheme	scheme	NN
orr-bootleg-2020	275	7	where	where	WRB
orr-bootleg-2020	275	8	popular	popular	JJ
orr-bootleg-2020	275	9	entities	entity	NNS
orr-bootleg-2020	275	10	are	be	VBP
orr-bootleg-2020	275	11	more	more	RBR
orr-bootleg-2020	275	12	regularized	regularize	VBN
orr-bootleg-2020	275	13	sees	see	VBZ
orr-bootleg-2020	275	14	a	a	DT
orr-bootleg-2020	275	15	drop	drop	NN
orr-bootleg-2020	275	16	of	of	IN
orr-bootleg-2020	275	17	9.8	9.8	CD
orr-bootleg-2020	275	18	F1	f1	NN
orr-bootleg-2020	275	19	points	point	NNS
orr-bootleg-2020	275	20	in	in	IN
orr-bootleg-2020	275	21	performance	performance	NN
orr-bootleg-2020	275	22	compared	compare	VBN
orr-bootleg-2020	275	23	to	to	IN
orr-bootleg-2020	275	24	the	the	DT
orr-bootleg-2020	275	25	inverse	inverse	NN
orr-bootleg-2020	275	26	popularity	popularity	NN
orr-bootleg-2020	275	27	scheme	scheme	NN
orr-bootleg-2020	275	28	.	.	.
orr-bootleg-2020	276	1	5	5	CD
orr-bootleg-2020	276	2	Analysis	analysis	NN
orr-bootleg-2020	276	3	We	-PRON-	PRP
orr-bootleg-2020	276	4	have	have	VBP
orr-bootleg-2020	276	5	shown	show	VBN
orr-bootleg-2020	276	6	that	that	IN
orr-bootleg-2020	276	7	Bootleg	Bootleg	NNP
orr-bootleg-2020	276	8	excels	excel	VBZ
orr-bootleg-2020	276	9	on	on	IN
orr-bootleg-2020	276	10	benchmark	benchmark	NN
orr-bootleg-2020	276	11	tasks	task	NNS
orr-bootleg-2020	276	12	and	and	CC
orr-bootleg-2020	276	13	that	that	IN
orr-bootleg-2020	276	14	Bootleg	Bootleg	NNP
orr-bootleg-2020	276	15	’s	’s	POS
orr-bootleg-2020	276	16	learned	learn	VBD
orr-bootleg-2020	276	17	patterns	pattern	NNS
orr-bootleg-2020	276	18	can	can	MD
orr-bootleg-2020	276	19	transfer	transfer	VB
orr-bootleg-2020	276	20	to	to	IN
orr-bootleg-2020	276	21	non	non	JJ
orr-bootleg-2020	276	22	-	-	JJ
orr-bootleg-2020	276	23	NED	ned	JJ
orr-bootleg-2020	276	24	tasks	task	NNS
orr-bootleg-2020	276	25	.	.	.
orr-bootleg-2020	277	1	We	-PRON-	PRP
orr-bootleg-2020	277	2	now	now	RB
orr-bootleg-2020	277	3	verify	verify	VBP
orr-bootleg-2020	277	4	whether	whether	IN
orr-bootleg-2020	277	5	the	the	DT
orr-bootleg-2020	277	6	defined	define	VBN
orr-bootleg-2020	277	7	entity	entity	NN
orr-bootleg-2020	277	8	,	,	,
orr-bootleg-2020	277	9	type	type	NN
orr-bootleg-2020	277	10	consistency	consistency	NN
orr-bootleg-2020	277	11	,	,	,
orr-bootleg-2020	277	12	KG	kg	NN
orr-bootleg-2020	277	13	relation	relation	NN
orr-bootleg-2020	277	14	,	,	,
orr-bootleg-2020	277	15	and	and	CC
orr-bootleg-2020	277	16	affordance	affordance	NN
orr-bootleg-2020	277	17	reasoning	reasoning	NN
orr-bootleg-2020	277	18	patterns	pattern	NNS
orr-bootleg-2020	277	19	are	be	VBP
orr-bootleg-2020	277	20	responsible	responsible	JJ
orr-bootleg-2020	277	21	for	for	IN
orr-bootleg-2020	277	22	these	these	DT
orr-bootleg-2020	277	23	results	result	NNS
orr-bootleg-2020	277	24	.	.	.
orr-bootleg-2020	278	1	We	-PRON-	PRP
orr-bootleg-2020	278	2	evaluate	evaluate	VBP
orr-bootleg-2020	278	3	each	each	DT
orr-bootleg-2020	278	4	over	over	IN
orr-bootleg-2020	278	5	a	a	DT
orr-bootleg-2020	278	6	representative	representative	JJ
orr-bootleg-2020	278	7	slice	slice	NN
orr-bootleg-2020	278	8	of	of	IN
orr-bootleg-2020	278	9	the	the	DT
orr-bootleg-2020	278	10	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	278	11	validation	validation	NN
orr-bootleg-2020	278	12	set	set	VBD
orr-bootleg-2020	278	13	that	that	WDT
orr-bootleg-2020	278	14	exemplifies	exemplify	VBZ
orr-bootleg-2020	278	15	one	one	CD
orr-bootleg-2020	278	16	of	of	IN
orr-bootleg-2020	278	17	the	the	DT
orr-bootleg-2020	278	18	reasoning	reasoning	NN
orr-bootleg-2020	278	19	patterns	pattern	NNS
orr-bootleg-2020	278	20	and	and	CC
orr-bootleg-2020	278	21	present	present	VB
orr-bootleg-2020	278	22	the	the	DT
orr-bootleg-2020	278	23	results	result	NNS
orr-bootleg-2020	278	24	from	from	IN
orr-bootleg-2020	278	25	each	each	DT
orr-bootleg-2020	278	26	ablated	ablate	VBN
orr-bootleg-2020	278	27	model	model	NN
orr-bootleg-2020	278	28	(	(	-LRB-
orr-bootleg-2020	278	29	Table	table	NN
orr-bootleg-2020	278	30	7	7	CD
orr-bootleg-2020	278	31	)	)	-RRB-
orr-bootleg-2020	278	32	.	.	.
orr-bootleg-2020	279	1	•	•	VB
orr-bootleg-2020	279	2	Entity	entity	NN
orr-bootleg-2020	279	3	To	to	TO
orr-bootleg-2020	279	4	evaluate	evaluate	VB
orr-bootleg-2020	279	5	whether	whether	IN
orr-bootleg-2020	279	6	Bootleg	Bootleg	NNP
orr-bootleg-2020	279	7	captures	capture	VBZ
orr-bootleg-2020	279	8	factual	factual	JJ
orr-bootleg-2020	279	9	knowledge	knowledge	NN
orr-bootleg-2020	279	10	about	about	IN
orr-bootleg-2020	279	11	entities	entity	NNS
orr-bootleg-2020	279	12	in	in	IN
orr-bootleg-2020	279	13	the	the	DT
orr-bootleg-2020	279	14	form	form	NN
orr-bootleg-2020	279	15	of	of	IN
orr-bootleg-2020	279	16	textual	textual	JJ
orr-bootleg-2020	279	17	entity	entity	NN
orr-bootleg-2020	279	18	cues	cue	NNS
orr-bootleg-2020	279	19	,	,	,
orr-bootleg-2020	279	20	we	-PRON-	PRP
orr-bootleg-2020	279	21	consider	consider	VBP
orr-bootleg-2020	279	22	the	the	DT
orr-bootleg-2020	279	23	slice	slice	NN
orr-bootleg-2020	279	24	of	of	IN
orr-bootleg-2020	279	25	28	28	CD
orr-bootleg-2020	279	26	K	K	NNP
orr-bootleg-2020	279	27	overall	overall	RB
orr-bootleg-2020	279	28	,	,	,
orr-bootleg-2020	279	29	5	5	CD
orr-bootleg-2020	279	30	K	k	NN
orr-bootleg-2020	279	31	tail	tail	NN
orr-bootleg-2020	279	32	examples	example	NNS
orr-bootleg-2020	279	33	where	where	WRB
orr-bootleg-2020	279	34	the	the	DT
orr-bootleg-2020	279	35	gold	gold	JJ
orr-bootleg-2020	279	36	entity	entity	NN
orr-bootleg-2020	279	37	has	have	VBZ
orr-bootleg-2020	279	38	no	no	DT
orr-bootleg-2020	279	39	relation	relation	NN
orr-bootleg-2020	279	40	or	or	CC
orr-bootleg-2020	279	41	type	type	NN
orr-bootleg-2020	279	42	signals	signal	NNS
orr-bootleg-2020	279	43	available	available	JJ
orr-bootleg-2020	279	44	.	.	.
orr-bootleg-2020	280	1	•	•	NN
orr-bootleg-2020	280	2	Type	type	NN
orr-bootleg-2020	280	3	Consistency	consistency	NN
orr-bootleg-2020	280	4	To	to	TO
orr-bootleg-2020	280	5	evaluate	evaluate	VB
orr-bootleg-2020	280	6	whether	whether	IN
orr-bootleg-2020	280	7	Bootleg	Bootleg	NNP
orr-bootleg-2020	280	8	captures	capture	VBZ
orr-bootleg-2020	280	9	consistency	consistency	NN
orr-bootleg-2020	280	10	patterns	pattern	NNS
orr-bootleg-2020	280	11	,	,	,
orr-bootleg-2020	280	12	we	-PRON-	PRP
orr-bootleg-2020	280	13	consider	consider	VBP
orr-bootleg-2020	280	14	the	the	DT
orr-bootleg-2020	280	15	slice	slice	NN
orr-bootleg-2020	280	16	of	of	IN
orr-bootleg-2020	280	17	312	312	CD
orr-bootleg-2020	280	18	K	k	NN
orr-bootleg-2020	280	19	overall	overall	RB
orr-bootleg-2020	280	20	,	,	,
orr-bootleg-2020	280	21	19	19	CD
orr-bootleg-2020	280	22	K	k	NN
orr-bootleg-2020	280	23	tail	tail	NN
orr-bootleg-2020	280	24	examples	example	NNS
orr-bootleg-2020	280	25	that	that	WDT
orr-bootleg-2020	280	26	contain	contain	VBP
orr-bootleg-2020	280	27	a	a	DT
orr-bootleg-2020	280	28	list	list	NN
orr-bootleg-2020	280	29	of	of	IN
orr-bootleg-2020	280	30	three	three	CD
orr-bootleg-2020	280	31	or	or	CC
orr-bootleg-2020	280	32	more	more	RBR
orr-bootleg-2020	280	33	sequential	sequential	JJ
orr-bootleg-2020	280	34	distinct	distinct	JJ
orr-bootleg-2020	280	35	gold	gold	NN
orr-bootleg-2020	280	36	entities	entity	NNS
orr-bootleg-2020	280	37	,	,	,
orr-bootleg-2020	280	38	where	where	WRB
orr-bootleg-2020	280	39	all	all	DT
orr-bootleg-2020	280	40	items	item	NNS
orr-bootleg-2020	280	41	in	in	IN
orr-bootleg-2020	280	42	the	the	DT
orr-bootleg-2020	280	43	list	list	NN
orr-bootleg-2020	280	44	share	share	NN
orr-bootleg-2020	280	45	at	at	RB
orr-bootleg-2020	280	46	least	least	RBS
orr-bootleg-2020	280	47	one	one	CD
orr-bootleg-2020	280	48	type	type	NN
orr-bootleg-2020	280	49	.	.	.
orr-bootleg-2020	281	1	•	•	VB
orr-bootleg-2020	281	2	KG	kg	NN
orr-bootleg-2020	281	3	Relation	relation	NN
orr-bootleg-2020	281	4	To	to	TO
orr-bootleg-2020	281	5	evaluate	evaluate	VB
orr-bootleg-2020	281	6	whether	whether	IN
orr-bootleg-2020	281	7	Bootleg	Bootleg	NNP
orr-bootleg-2020	281	8	captures	capture	VBZ
orr-bootleg-2020	281	9	KG	kg	NN
orr-bootleg-2020	281	10	relation	relation	NN
orr-bootleg-2020	281	11	patterns	pattern	NNS
orr-bootleg-2020	281	12	,	,	,
orr-bootleg-2020	281	13	we	-PRON-	PRP
orr-bootleg-2020	281	14	consider	consider	VBP
orr-bootleg-2020	281	15	the	the	DT
orr-bootleg-2020	281	16	slice	slice	NN
orr-bootleg-2020	281	17	of	of	IN
orr-bootleg-2020	281	18	1.1	1.1	CD
orr-bootleg-2020	281	19	M	M	NNP
orr-bootleg-2020	281	20	overall	overall	NN
orr-bootleg-2020	281	21	,	,	,
orr-bootleg-2020	281	22	37	37	CD
orr-bootleg-2020	281	23	K	k	NN
orr-bootleg-2020	281	24	tail	tail	NN
orr-bootleg-2020	281	25	examples	example	NNS
orr-bootleg-2020	281	26	for	for	IN
orr-bootleg-2020	281	27	which	which	WDT
orr-bootleg-2020	281	28	the	the	DT
orr-bootleg-2020	281	29	gold	gold	NN
orr-bootleg-2020	281	30	entities	entity	NNS
orr-bootleg-2020	281	31	are	be	VBP
orr-bootleg-2020	281	32	connected	connect	VBN
orr-bootleg-2020	281	33	by	by	IN
orr-bootleg-2020	281	34	a	a	DT
orr-bootleg-2020	281	35	known	know	VBN
orr-bootleg-2020	281	36	relation	relation	NN
orr-bootleg-2020	281	37	in	in	IN
orr-bootleg-2020	281	38	the	the	DT
orr-bootleg-2020	281	39	Wikidata	Wikidata	NNP
orr-bootleg-2020	281	40	knowledge	knowledge	NN
orr-bootleg-2020	281	41	graph	graph	NN
orr-bootleg-2020	281	42	.	.	.
orr-bootleg-2020	282	1	•	•	VB
orr-bootleg-2020	282	2	Type	type	NN
orr-bootleg-2020	282	3	Affordance	affordance	NN
orr-bootleg-2020	282	4	To	to	TO
orr-bootleg-2020	282	5	evaluate	evaluate	VB
orr-bootleg-2020	282	6	whether	whether	IN
orr-bootleg-2020	282	7	Bootleg	Bootleg	NNP
orr-bootleg-2020	282	8	captures	capture	VBZ
orr-bootleg-2020	282	9	affordance	affordance	NN
orr-bootleg-2020	282	10	patterns	pattern	NNS
orr-bootleg-2020	282	11	,	,	,
orr-bootleg-2020	282	12	we	-PRON-	PRP
orr-bootleg-2020	282	13	consider	consider	VBP
orr-bootleg-2020	282	14	a	a	DT
orr-bootleg-2020	282	15	slice	slice	NN
orr-bootleg-2020	282	16	where	where	WRB
orr-bootleg-2020	282	17	the	the	DT
orr-bootleg-2020	282	18	sentence	sentence	NN
orr-bootleg-2020	282	19	contains	contain	VBZ
orr-bootleg-2020	282	20	keywords	keyword	NNS
orr-bootleg-2020	282	21	that	that	WDT
orr-bootleg-2020	282	22	are	be	VBP
orr-bootleg-2020	282	23	afforded	afford	VBN
orr-bootleg-2020	282	24	by	by	IN
orr-bootleg-2020	282	25	the	the	DT
orr-bootleg-2020	282	26	type	type	NN
orr-bootleg-2020	282	27	of	of	IN
orr-bootleg-2020	282	28	the	the	DT
orr-bootleg-2020	282	29	gold	gold	JJ
orr-bootleg-2020	282	30	entity	entity	NN
orr-bootleg-2020	282	31	.	.	.
orr-bootleg-2020	283	1	We	-PRON-	PRP
orr-bootleg-2020	283	2	mine	mine	VBP
orr-bootleg-2020	283	3	the	the	DT
orr-bootleg-2020	283	4	keywords	keyword	NNS
orr-bootleg-2020	283	5	afforded	afford	VBN
orr-bootleg-2020	283	6	by	by	IN
orr-bootleg-2020	283	7	a	a	DT
orr-bootleg-2020	283	8	type	type	NN
orr-bootleg-2020	283	9	by	by	IN
orr-bootleg-2020	283	10	taking	take	VBG
orr-bootleg-2020	283	11	the	the	DT
orr-bootleg-2020	283	12	15	15	CD
orr-bootleg-2020	283	13	keywords	keyword	NNS
orr-bootleg-2020	283	14	that	that	WDT
orr-bootleg-2020	283	15	receive	receive	VBP
orr-bootleg-2020	283	16	the	the	DT
orr-bootleg-2020	283	17	highest	high	JJS
orr-bootleg-2020	283	18	TF	TF	NNP
orr-bootleg-2020	283	19	-	-	HYPH
orr-bootleg-2020	283	20	IDF	IDF	NNP
orr-bootleg-2020	283	21	scores	score	NNS
orr-bootleg-2020	283	22	over	over	IN
orr-bootleg-2020	283	23	training	training	NN
orr-bootleg-2020	283	24	examples	example	NNS
orr-bootleg-2020	283	25	with	with	IN
orr-bootleg-2020	283	26	that	that	DT
orr-bootleg-2020	283	27	type	type	NN
orr-bootleg-2020	283	28	.	.	.
orr-bootleg-2020	284	1	This	this	DT
orr-bootleg-2020	284	2	slice	slice	NN
orr-bootleg-2020	284	3	has	have	VBZ
orr-bootleg-2020	284	4	3.4	3.4	CD
orr-bootleg-2020	284	5	M	M	NNP
orr-bootleg-2020	284	6	overall	overall	NN
orr-bootleg-2020	284	7	,	,	,
orr-bootleg-2020	284	8	124	124	CD
orr-bootleg-2020	284	9	K	k	NN
orr-bootleg-2020	284	10	tail	tail	NN
orr-bootleg-2020	284	11	examples	example	NNS
orr-bootleg-2020	284	12	.	.	.
orr-bootleg-2020	285	1	Pattern	Pattern	NNP
orr-bootleg-2020	285	2	Analysis	Analysis	NNP
orr-bootleg-2020	285	3	For	for	IN
orr-bootleg-2020	285	4	the	the	DT
orr-bootleg-2020	285	5	slice	slice	NN
orr-bootleg-2020	285	6	representing	represent	VBG
orr-bootleg-2020	285	7	each	each	DT
orr-bootleg-2020	285	8	reasoning	reasoning	NN
orr-bootleg-2020	285	9	pattern	pattern	NN
orr-bootleg-2020	285	10	,	,	,
orr-bootleg-2020	285	11	we	-PRON-	PRP
orr-bootleg-2020	285	12	find	find	VBP
orr-bootleg-2020	285	13	that	that	IN
orr-bootleg-2020	285	14	Bootleg	Bootleg	NNP
orr-bootleg-2020	285	15	provides	provide	VBZ
orr-bootleg-2020	285	16	a	a	DT
orr-bootleg-2020	285	17	lift	lift	NN
orr-bootleg-2020	285	18	over	over	IN
orr-bootleg-2020	285	19	the	the	DT
orr-bootleg-2020	285	20	Entity	entity	NN
orr-bootleg-2020	285	21	-	-	,
orr-bootleg-2020	285	22	only	only	RB
orr-bootleg-2020	285	23	and	and	CC
orr-bootleg-2020	285	24	NED	NED	NNP
orr-bootleg-2020	285	25	-	-	HYPH
orr-bootleg-2020	285	26	Base	Base	NNP
orr-bootleg-2020	285	27	models	model	NNS
orr-bootleg-2020	285	28	,	,	,
orr-bootleg-2020	285	29	especially	especially	RB
orr-bootleg-2020	285	30	over	over	IN
orr-bootleg-2020	285	31	the	the	DT
orr-bootleg-2020	285	32	tail	tail	NN
orr-bootleg-2020	285	33	.	.	.
orr-bootleg-2020	286	1	We	-PRON-	PRP
orr-bootleg-2020	286	2	find	find	VBP
orr-bootleg-2020	286	3	that	that	IN
orr-bootleg-2020	286	4	Bootleg	Bootleg	NNP
orr-bootleg-2020	286	5	generally	generally	RB
orr-bootleg-2020	286	6	combines	combine	VBZ
orr-bootleg-2020	286	7	the	the	DT
orr-bootleg-2020	286	8	entity	entity	NN
orr-bootleg-2020	286	9	,	,	,
orr-bootleg-2020	286	10	relation	relation	NN
orr-bootleg-2020	286	11	,	,	,
orr-bootleg-2020	286	12	and	and	CC
orr-bootleg-2020	286	13	type	type	NN
orr-bootleg-2020	286	14	signals	signal	NNS
orr-bootleg-2020	286	15	effectively	effectively	RB
orr-bootleg-2020	286	16	,	,	,
orr-bootleg-2020	286	17	performing	perform	VBG
orr-bootleg-2020	286	18	better	well	RBR
orr-bootleg-2020	286	19	than	than	IN
orr-bootleg-2020	286	20	the	the	DT
orr-bootleg-2020	286	21	individual	individual	JJ
orr-bootleg-2020	286	22	Entity	entity	NN
orr-bootleg-2020	286	23	-	-	,
orr-bootleg-2020	286	24	only	only	RB
orr-bootleg-2020	286	25	,	,	,
orr-bootleg-2020	286	26	KG	kg	NN
orr-bootleg-2020	286	27	-	-	,
orr-bootleg-2020	286	28	only	only	RB
orr-bootleg-2020	286	29	,	,	,
orr-bootleg-2020	286	30	and	and	CC
orr-bootleg-2020	286	31	Type	type	NN
orr-bootleg-2020	286	32	-	-	HYPH
orr-bootleg-2020	286	33	only	only	RB
orr-bootleg-2020	286	34	models	model	NNS
orr-bootleg-2020	286	35	,	,	,
orr-bootleg-2020	286	36	although	although	IN
orr-bootleg-2020	286	37	the	the	DT
orr-bootleg-2020	286	38	KG	kg	NN
orr-bootleg-2020	286	39	-	-	HYPH
orr-bootleg-2020	286	40	only	only	RB
orr-bootleg-2020	286	41	model	model	NN
orr-bootleg-2020	286	42	performs	perform	VBZ
orr-bootleg-2020	286	43	well	well	RB
orr-bootleg-2020	286	44	on	on	IN
orr-bootleg-2020	286	45	the	the	DT
orr-bootleg-2020	286	46	KG	KG	NNP
orr-bootleg-2020	286	47	relation	relation	NN
orr-bootleg-2020	286	48	slice	slice	NN
orr-bootleg-2020	286	49	.	.	.
orr-bootleg-2020	287	1	The	the	DT
orr-bootleg-2020	287	2	lift	lift	NN
orr-bootleg-2020	287	3	from	from	IN
orr-bootleg-2020	287	4	Bootleg	Bootleg	NNP
orr-bootleg-2020	287	5	across	across	IN
orr-bootleg-2020	287	6	slices	slice	NNS
orr-bootleg-2020	287	7	indicates	indicate	VBZ
orr-bootleg-2020	287	8	the	the	DT
orr-bootleg-2020	287	9	model	model	NN
orr-bootleg-2020	287	10	’s	’s	POS
orr-bootleg-2020	287	11	ability	ability	NN
orr-bootleg-2020	287	12	to	to	TO
orr-bootleg-2020	287	13	capture	capture	VB
orr-bootleg-2020	287	14	the	the	DT
orr-bootleg-2020	287	15	reasoning	reasoning	NN
orr-bootleg-2020	287	16	required	require	VBN
orr-bootleg-2020	287	17	for	for	IN
orr-bootleg-2020	287	18	the	the	DT
orr-bootleg-2020	287	19	slice	slice	NN
orr-bootleg-2020	287	20	.	.	.
orr-bootleg-2020	288	1	We	-PRON-	PRP
orr-bootleg-2020	288	2	provide	provide	VBP
orr-bootleg-2020	288	3	additional	additional	JJ
orr-bootleg-2020	288	4	details	detail	NNS
orr-bootleg-2020	288	5	in	in	IN
orr-bootleg-2020	288	6	Appendix	Appendix	NNP
orr-bootleg-2020	288	7	D.	D.	NNP
orr-bootleg-2020	288	8	Error	Error	NNP
orr-bootleg-2020	288	9	Analysis	Analysis	NNP
orr-bootleg-2020	288	10	We	-PRON-	PRP
orr-bootleg-2020	288	11	next	next	JJ
orr-bootleg-2020	288	12	study	study	NN
orr-bootleg-2020	288	13	the	the	DT
orr-bootleg-2020	288	14	errors	error	NNS
orr-bootleg-2020	288	15	made	make	VBN
orr-bootleg-2020	288	16	by	by	IN
orr-bootleg-2020	288	17	Bootleg	Bootleg	NNP
orr-bootleg-2020	288	18	and	and	CC
orr-bootleg-2020	288	19	find	find	VB
orr-bootleg-2020	288	20	four	four	CD
orr-bootleg-2020	288	21	key	key	JJ
orr-bootleg-2020	288	22	error	error	NN
orr-bootleg-2020	288	23	buckets	bucket	NNS
orr-bootleg-2020	288	24	.	.	.
orr-bootleg-2020	289	1	•	•	NNP
orr-bootleg-2020	289	2	Granularity	Granularity	NNP
orr-bootleg-2020	289	3	Bootleg	Bootleg	NNP
orr-bootleg-2020	289	4	struggles	struggle	VBZ
orr-bootleg-2020	289	5	with	with	IN
orr-bootleg-2020	289	6	granularity	granularity	NN
orr-bootleg-2020	289	7	,	,	,
orr-bootleg-2020	289	8	predicting	predict	VBG
orr-bootleg-2020	289	9	an	an	DT
orr-bootleg-2020	289	10	entity	entity	NN
orr-bootleg-2020	289	11	that	that	WDT
orr-bootleg-2020	289	12	is	be	VBZ
orr-bootleg-2020	289	13	too	too	RB
orr-bootleg-2020	289	14	general	general	JJ
orr-bootleg-2020	289	15	or	or	CC
orr-bootleg-2020	289	16	too	too	RB
orr-bootleg-2020	289	17	specific	specific	JJ
orr-bootleg-2020	289	18	compared	compare	VBN
orr-bootleg-2020	289	19	to	to	IN
orr-bootleg-2020	289	20	the	the	DT
orr-bootleg-2020	289	21	gold	gold	JJ
orr-bootleg-2020	289	22	entity	entity	NN
orr-bootleg-2020	289	23	(	(	-LRB-
orr-bootleg-2020	289	24	example	example	NN
orr-bootleg-2020	289	25	in	in	IN
orr-bootleg-2020	289	26	Table	table	NN
orr-bootleg-2020	289	27	8)	8)	CD
orr-bootleg-2020	289	28	.	.	.
orr-bootleg-2020	290	1	Considering	consider	VBG
orr-bootleg-2020	290	2	the	the	DT
orr-bootleg-2020	290	3	set	set	NN
orr-bootleg-2020	290	4	of	of	IN
orr-bootleg-2020	290	5	examples	example	NNS
orr-bootleg-2020	290	6	where	where	WRB
orr-bootleg-2020	290	7	the	the	DT
orr-bootleg-2020	290	8	predicted	predict	VBN
orr-bootleg-2020	290	9	entity	entity	NN
orr-bootleg-2020	290	10	is	be	VBZ
orr-bootleg-2020	290	11	a	a	DT
orr-bootleg-2020	290	12	Wikidata	Wikidata	NNP
orr-bootleg-2020	290	13	subclass	subclass	NN
orr-bootleg-2020	290	14	of	of	IN
orr-bootleg-2020	290	15	the	the	DT
orr-bootleg-2020	290	16	gold	gold	JJ
orr-bootleg-2020	290	17	entity	entity	NN
orr-bootleg-2020	290	18	or	or	CC
orr-bootleg-2020	290	19	vice	vice	NN
orr-bootleg-2020	290	20	versa	versa	RB
orr-bootleg-2020	290	21	,	,	,
orr-bootleg-2020	290	22	Bootleg	Bootleg	NNP
orr-bootleg-2020	290	23	predicts	predict	VBZ
orr-bootleg-2020	290	24	a	a	DT
orr-bootleg-2020	290	25	too	too	RB
orr-bootleg-2020	290	26	general	general	JJ
orr-bootleg-2020	290	27	or	or	CC
orr-bootleg-2020	290	28	specific	specific	JJ
orr-bootleg-2020	290	29	entity	entity	NN
orr-bootleg-2020	290	30	in	in	IN
orr-bootleg-2020	290	31	12	12	CD
orr-bootleg-2020	290	32	%	%	NN
orr-bootleg-2020	290	33	of	of	IN
orr-bootleg-2020	290	34	overall	overall	JJ
orr-bootleg-2020	290	35	and	and	CC
orr-bootleg-2020	290	36	7	7	CD
orr-bootleg-2020	290	37	%	%	NN
orr-bootleg-2020	290	38	of	of	IN
orr-bootleg-2020	290	39	tail	tail	NN
orr-bootleg-2020	290	40	errors	error	NNS
orr-bootleg-2020	290	41	.	.	.
orr-bootleg-2020	291	1	•	•	NNP
orr-bootleg-2020	291	2	Numerical	Numerical	NNP
orr-bootleg-2020	291	3	Bootleg	Bootleg	NNP
orr-bootleg-2020	291	4	struggles	struggle	VBZ
orr-bootleg-2020	291	5	with	with	IN
orr-bootleg-2020	291	6	entities	entity	NNS
orr-bootleg-2020	291	7	containing	contain	VBG
orr-bootleg-2020	291	8	numerical	numerical	JJ
orr-bootleg-2020	291	9	tokens	token	NNS
orr-bootleg-2020	291	10	,	,	,
orr-bootleg-2020	291	11	which	which	WDT
orr-bootleg-2020	291	12	may	may	MD
orr-bootleg-2020	291	13	be	be	VB
orr-bootleg-2020	291	14	due	due	IN
orr-bootleg-2020	291	15	to	to	IN
orr-bootleg-2020	291	16	the	the	DT
orr-bootleg-2020	291	17	fact	fact	NN
orr-bootleg-2020	291	18	that	that	IN
orr-bootleg-2020	291	19	the	the	DT
orr-bootleg-2020	291	20	BERT	BERT	NNP
orr-bootleg-2020	291	21	model	model	NN
orr-bootleg-2020	291	22	represents	represent	VBZ
orr-bootleg-2020	291	23	some	some	DT
orr-bootleg-2020	291	24	numbers	number	NNS
orr-bootleg-2020	291	25	with	with	IN
orr-bootleg-2020	291	26	sub	sub	JJ
orr-bootleg-2020	291	27	-	-	HYPH
orr-bootleg-2020	291	28	word	word	NN
orr-bootleg-2020	291	29	tokens	token	NNS
orr-bootleg-2020	291	30	and	and	CC
orr-bootleg-2020	291	31	is	be	VBZ
orr-bootleg-2020	291	32	known	know	VBN
orr-bootleg-2020	291	33	to	to	TO
orr-bootleg-2020	291	34	not	not	RB
orr-bootleg-2020	291	35	perform	perform	VB
orr-bootleg-2020	291	36	as	as	RB
orr-bootleg-2020	291	37	well	well	RB
orr-bootleg-2020	291	38	for	for	IN
orr-bootleg-2020	291	39	numbers	number	NNS
orr-bootleg-2020	291	40	as	as	IN
orr-bootleg-2020	291	41	other	other	JJ
orr-bootleg-2020	291	42	language	language	NN
orr-bootleg-2020	291	43	models	model	NNS
orr-bootleg-2020	291	44	[	[	-LRB-
orr-bootleg-2020	291	45	49	49	CD
orr-bootleg-2020	291	46	]	]	-RRB-
orr-bootleg-2020	291	47	(	(	-LRB-
orr-bootleg-2020	291	48	example	example	NN
orr-bootleg-2020	291	49	in	in	IN
orr-bootleg-2020	291	50	Table	Table	NNP
orr-bootleg-2020	291	51	8)	8)	CD
orr-bootleg-2020	291	52	.	.	.
orr-bootleg-2020	292	1	To	to	TO
orr-bootleg-2020	292	2	evaluate	evaluate	VB
orr-bootleg-2020	292	3	examples	example	NNS
orr-bootleg-2020	292	4	requiring	require	VBG
orr-bootleg-2020	292	5	12	12	CD
orr-bootleg-2020	292	6	Table	table	NN
orr-bootleg-2020	292	7	7	7	CD
orr-bootleg-2020	292	8	:	:	:
orr-bootleg-2020	292	9	We	-PRON-	PRP
orr-bootleg-2020	292	10	report	report	VBP
orr-bootleg-2020	292	11	the	the	DT
orr-bootleg-2020	292	12	Overall	Overall	NNP
orr-bootleg-2020	292	13	/	/	SYM
orr-bootleg-2020	292	14	Tail	Tail	NNP
orr-bootleg-2020	292	15	F1	F1	NNP
orr-bootleg-2020	292	16	scores	score	NNS
orr-bootleg-2020	292	17	across	across	IN
orr-bootleg-2020	292	18	each	each	DT
orr-bootleg-2020	292	19	ablation	ablation	NN
orr-bootleg-2020	292	20	model	model	NN
orr-bootleg-2020	292	21	for	for	IN
orr-bootleg-2020	292	22	a	a	DT
orr-bootleg-2020	292	23	slice	slice	NN
orr-bootleg-2020	292	24	of	of	IN
orr-bootleg-2020	292	25	data	datum	NNS
orr-bootleg-2020	292	26	that	that	WDT
orr-bootleg-2020	292	27	exemplifies	exemplify	VBZ
orr-bootleg-2020	292	28	a	a	DT
orr-bootleg-2020	292	29	reasoning	reasoning	NN
orr-bootleg-2020	292	30	pattern	pattern	NN
orr-bootleg-2020	292	31	.	.	.
orr-bootleg-2020	293	1	Each	each	DT
orr-bootleg-2020	293	2	slice	slice	NN
orr-bootleg-2020	293	3	is	be	VBZ
orr-bootleg-2020	293	4	representative	representative	JJ
orr-bootleg-2020	293	5	but	but	CC
orr-bootleg-2020	293	6	may	may	MD
orr-bootleg-2020	293	7	not	not	RB
orr-bootleg-2020	293	8	cover	cover	VB
orr-bootleg-2020	293	9	every	every	DT
orr-bootleg-2020	293	10	example	example	NN
orr-bootleg-2020	293	11	that	that	WDT
orr-bootleg-2020	293	12	contains	contain	VBZ
orr-bootleg-2020	293	13	the	the	DT
orr-bootleg-2020	293	14	reasoning	reasoning	NN
orr-bootleg-2020	293	15	pattern	pattern	NN
orr-bootleg-2020	293	16	.	.	.
orr-bootleg-2020	294	1	Model	Model	NNP
orr-bootleg-2020	294	2	Entity	Entity	NNP
orr-bootleg-2020	294	3	Type	Type	NNP
orr-bootleg-2020	294	4	Consistency	Consistency	NNP
orr-bootleg-2020	294	5	KG	KG	NNP
orr-bootleg-2020	294	6	Relation	Relation	NNP
orr-bootleg-2020	294	7	Type	Type	NNP
orr-bootleg-2020	294	8	Affordance	Affordance	NNP
orr-bootleg-2020	294	9	NED	NED	NNP
orr-bootleg-2020	294	10	-	-	HYPH
orr-bootleg-2020	294	11	Base	Base	NNP
orr-bootleg-2020	294	12	59/29	59/29	CD
orr-bootleg-2020	294	13	84/29	84/29	CD
orr-bootleg-2020	294	14	91/30	91/30	CD
orr-bootleg-2020	294	15	87/28	87/28	CD
orr-bootleg-2020	294	16	Bootleg	bootleg	NN
orr-bootleg-2020	294	17	66/47	66/47	CD
orr-bootleg-2020	294	18	95/85	95/85	CD
orr-bootleg-2020	294	19	98/92	98/92	CD
orr-bootleg-2020	294	20	93/73	93/73	CD
orr-bootleg-2020	294	21	Bootleg	Bootleg	NNP
orr-bootleg-2020	294	22	(	(	-LRB-
orr-bootleg-2020	294	23	Ent	ent	NN
orr-bootleg-2020	294	24	-	-	,
orr-bootleg-2020	294	25	only	only	RB
orr-bootleg-2020	294	26	)	)	-RRB-
orr-bootleg-2020	294	27	59/31	59/31	CD
orr-bootleg-2020	294	28	87/45	87/45	CD
orr-bootleg-2020	294	29	90/42	90/42	CD
orr-bootleg-2020	294	30	87/39	87/39	CD
orr-bootleg-2020	294	31	Bootleg	Bootleg	NNP
orr-bootleg-2020	294	32	(	(	-LRB-
orr-bootleg-2020	294	33	Type	type	NN
orr-bootleg-2020	294	34	-	-	HYPH
orr-bootleg-2020	294	35	only	only	RB
orr-bootleg-2020	294	36	)	)	-RRB-
orr-bootleg-2020	294	37	53/44	53/44	CD
orr-bootleg-2020	294	38	93/80	93/80	CD
orr-bootleg-2020	294	39	93/69	93/69	CD
orr-bootleg-2020	294	40	90/66	90/66	CD
orr-bootleg-2020	294	41	Bootleg	Bootleg	NNP
orr-bootleg-2020	294	42	(	(	-LRB-
orr-bootleg-2020	294	43	KG	KG	NNP
orr-bootleg-2020	294	44	-	-	HYPH
orr-bootleg-2020	294	45	only	only	RB
orr-bootleg-2020	294	46	)	)	-RRB-
orr-bootleg-2020	294	47	40/29	40/29	CD
orr-bootleg-2020	294	48	92/79	92/79	CD
orr-bootleg-2020	294	49	97/93	97/93	CD
orr-bootleg-2020	294	50	89/68	89/68	CD
orr-bootleg-2020	294	51	%	%	NN
orr-bootleg-2020	294	52	Coverage	coverage	NN
orr-bootleg-2020	294	53	0.7%/3.3	0.7%/3.3	CD
orr-bootleg-2020	294	54	%	%	NN
orr-bootleg-2020	294	55	8%/12	8%/12	CD
orr-bootleg-2020	294	56	%	%	NN
orr-bootleg-2020	294	57	27%/23	27%/23	CD
orr-bootleg-2020	294	58	%	%	NN
orr-bootleg-2020	294	59	84%/76	84%/76	CD
orr-bootleg-2020	294	60	%	%	NN
orr-bootleg-2020	294	61	reasoning	reason	VBG
orr-bootleg-2020	294	62	over	over	IN
orr-bootleg-2020	294	63	numbers	number	NNS
orr-bootleg-2020	294	64	,	,	,
orr-bootleg-2020	294	65	we	-PRON-	PRP
orr-bootleg-2020	294	66	consider	consider	VBP
orr-bootleg-2020	294	67	the	the	DT
orr-bootleg-2020	294	68	slice	slice	NN
orr-bootleg-2020	294	69	of	of	IN
orr-bootleg-2020	294	70	data	datum	NNS
orr-bootleg-2020	294	71	where	where	WRB
orr-bootleg-2020	294	72	the	the	DT
orr-bootleg-2020	294	73	entity	entity	NN
orr-bootleg-2020	294	74	title	title	NN
orr-bootleg-2020	294	75	contains	contain	VBZ
orr-bootleg-2020	294	76	a	a	DT
orr-bootleg-2020	294	77	year	year	NN
orr-bootleg-2020	294	78	,	,	,
orr-bootleg-2020	294	79	as	as	IN
orr-bootleg-2020	294	80	this	this	DT
orr-bootleg-2020	294	81	is	be	VBZ
orr-bootleg-2020	294	82	the	the	DT
orr-bootleg-2020	294	83	most	most	RBS
orr-bootleg-2020	294	84	common	common	JJ
orr-bootleg-2020	294	85	numerical	numerical	JJ
orr-bootleg-2020	294	86	feature	feature	NN
orr-bootleg-2020	294	87	in	in	IN
orr-bootleg-2020	294	88	a	a	DT
orr-bootleg-2020	294	89	title	title	NN
orr-bootleg-2020	294	90	.	.	.
orr-bootleg-2020	295	1	This	this	DT
orr-bootleg-2020	295	2	slice	slice	NN
orr-bootleg-2020	295	3	covers	cover	VBZ
orr-bootleg-2020	295	4	14	14	CD
orr-bootleg-2020	295	5	%	%	NN
orr-bootleg-2020	295	6	of	of	IN
orr-bootleg-2020	295	7	overall	overall	NN
orr-bootleg-2020	295	8	and	and	CC
orr-bootleg-2020	295	9	25	25	CD
orr-bootleg-2020	295	10	%	%	NN
orr-bootleg-2020	295	11	of	of	IN
orr-bootleg-2020	295	12	tail	tail	NN
orr-bootleg-2020	295	13	errors	error	NNS
orr-bootleg-2020	295	14	.	.	.
orr-bootleg-2020	296	1	•	•	NNP
orr-bootleg-2020	296	2	Multi	Multi	NNP
orr-bootleg-2020	296	3	-	-	HYPH
orr-bootleg-2020	296	4	Hop	Hop	NNP
orr-bootleg-2020	296	5	There	there	EX
orr-bootleg-2020	296	6	is	be	VBZ
orr-bootleg-2020	296	7	room	room	NN
orr-bootleg-2020	296	8	for	for	IN
orr-bootleg-2020	296	9	improvement	improvement	NN
orr-bootleg-2020	296	10	in	in	IN
orr-bootleg-2020	296	11	multi	multi	JJ
orr-bootleg-2020	296	12	-	-	JJ
orr-bootleg-2020	296	13	hop	hop	JJ
orr-bootleg-2020	296	14	reasoning	reasoning	NN
orr-bootleg-2020	296	15	.	.	.
orr-bootleg-2020	297	1	In	in	IN
orr-bootleg-2020	297	2	the	the	DT
orr-bootleg-2020	297	3	example	example	NN
orr-bootleg-2020	297	4	shown	show	VBN
orr-bootleg-2020	297	5	Table	table	NN
orr-bootleg-2020	297	6	8	8	CD
orr-bootleg-2020	297	7	,	,	,
orr-bootleg-2020	297	8	none	none	NN
orr-bootleg-2020	297	9	of	of	IN
orr-bootleg-2020	297	10	the	the	DT
orr-bootleg-2020	297	11	present	present	JJ
orr-bootleg-2020	297	12	gold	gold	NN
orr-bootleg-2020	297	13	entities	entity	NNS
orr-bootleg-2020	297	14	—	—	:
orr-bootleg-2020	297	15	Stillwater	Stillwater	NNP
orr-bootleg-2020	297	16	Santa	Santa	NNP
orr-bootleg-2020	297	17	Fe	Fe	NNP
orr-bootleg-2020	297	18	Depot	Depot	NNP
orr-bootleg-2020	297	19	,	,	,
orr-bootleg-2020	297	20	Citizens	Citizens	NNPS
orr-bootleg-2020	297	21	Bank	Bank	NNP
orr-bootleg-2020	297	22	Building	Building	NNP
orr-bootleg-2020	297	23	(	(	-LRB-
orr-bootleg-2020	297	24	Stillwater	Stillwater	NNP
orr-bootleg-2020	297	25	,	,	,
orr-bootleg-2020	297	26	Oklahoma	Oklahoma	NNP
orr-bootleg-2020	297	27	)	)	-RRB-
orr-bootleg-2020	297	28	,	,	,
orr-bootleg-2020	297	29	Hoke	Hoke	NNP
orr-bootleg-2020	297	30	Building	Building	NNP
orr-bootleg-2020	297	31	(	(	-LRB-
orr-bootleg-2020	297	32	Stillwater	Stillwater	NNP
orr-bootleg-2020	297	33	,	,	,
orr-bootleg-2020	297	34	Oklahoma	Oklahoma	NNP
orr-bootleg-2020	297	35	)	)	-RRB-
orr-bootleg-2020	297	36	,	,	,
orr-bootleg-2020	297	37	or	or	CC
orr-bootleg-2020	297	38	Walker	Walker	NNP
orr-bootleg-2020	297	39	Building	Building	NNP
orr-bootleg-2020	297	40	(	(	-LRB-
orr-bootleg-2020	297	41	Stillwater	Stillwater	NNP
orr-bootleg-2020	297	42	,	,	,
orr-bootleg-2020	297	43	Oklahoma)—are	oklahoma)—are	CD
orr-bootleg-2020	297	44	directly	directly	RB
orr-bootleg-2020	297	45	connected	connect	VBN
orr-bootleg-2020	297	46	in	in	IN
orr-bootleg-2020	297	47	Wikidata	Wikidata	NNP
orr-bootleg-2020	297	48	;	;	:
orr-bootleg-2020	297	49	however	however	RB
orr-bootleg-2020	297	50	,	,	,
orr-bootleg-2020	297	51	they	-PRON-	PRP
orr-bootleg-2020	297	52	share	share	VBP
orr-bootleg-2020	297	53	connections	connection	NNS
orr-bootleg-2020	297	54	to	to	IN
orr-bootleg-2020	297	55	the	the	DT
orr-bootleg-2020	297	56	entity	entity	NN
orr-bootleg-2020	297	57	“	"	``
orr-bootleg-2020	297	58	Oklahoma	Oklahoma	NNP
orr-bootleg-2020	297	59	”	"	''
orr-bootleg-2020	297	60	.	.	.
orr-bootleg-2020	298	1	This	this	DT
orr-bootleg-2020	298	2	indicates	indicate	VBZ
orr-bootleg-2020	298	3	that	that	IN
orr-bootleg-2020	298	4	the	the	DT
orr-bootleg-2020	298	5	correct	correct	JJ
orr-bootleg-2020	298	6	disambiguation	disambiguation	NN
orr-bootleg-2020	298	7	is	be	VBZ
orr-bootleg-2020	298	8	Citizens	Citizens	NNPS
orr-bootleg-2020	298	9	Bank	Bank	NNP
orr-bootleg-2020	298	10	Building	Building	NNP
orr-bootleg-2020	298	11	(	(	-LRB-
orr-bootleg-2020	298	12	Stillwater	Stillwater	NNP
orr-bootleg-2020	298	13	,	,	,
orr-bootleg-2020	298	14	Oklahoma	Oklahoma	NNP
orr-bootleg-2020	298	15	)	)	-RRB-
orr-bootleg-2020	298	16	,	,	,
orr-bootleg-2020	298	17	not	not	RB
orr-bootleg-2020	298	18	Citizens	Citizens	NNPS
orr-bootleg-2020	298	19	Bank	Bank	NNP
orr-bootleg-2020	298	20	Building	Building	NNP
orr-bootleg-2020	298	21	(	(	-LRB-
orr-bootleg-2020	298	22	Burnsville	Burnsville	NNP
orr-bootleg-2020	298	23	,	,	,
orr-bootleg-2020	298	24	North	North	NNP
orr-bootleg-2020	298	25	Carolina	Carolina	NNP
orr-bootleg-2020	298	26	)	)	-RRB-
orr-bootleg-2020	298	27	.	.	.
orr-bootleg-2020	299	1	To	to	TO
orr-bootleg-2020	299	2	evaluate	evaluate	VB
orr-bootleg-2020	299	3	examples	example	NNS
orr-bootleg-2020	299	4	requiring	require	VBG
orr-bootleg-2020	299	5	2-hop	2-hop	CD
orr-bootleg-2020	299	6	reasoning	reasoning	NN
orr-bootleg-2020	299	7	,	,	,
orr-bootleg-2020	299	8	we	-PRON-	PRP
orr-bootleg-2020	299	9	consider	consider	VBP
orr-bootleg-2020	299	10	examples	example	NNS
orr-bootleg-2020	299	11	where	where	WRB
orr-bootleg-2020	299	12	none	none	NN
orr-bootleg-2020	299	13	of	of	IN
orr-bootleg-2020	299	14	the	the	DT
orr-bootleg-2020	299	15	present	present	JJ
orr-bootleg-2020	299	16	entities	entity	NNS
orr-bootleg-2020	299	17	are	be	VBP
orr-bootleg-2020	299	18	directly	directly	RB
orr-bootleg-2020	299	19	linked	link	VBN
orr-bootleg-2020	299	20	in	in	IN
orr-bootleg-2020	299	21	the	the	DT
orr-bootleg-2020	299	22	KG	kg	NN
orr-bootleg-2020	299	23	,	,	,
orr-bootleg-2020	299	24	but	but	CC
orr-bootleg-2020	299	25	a	a	DT
orr-bootleg-2020	299	26	present	present	JJ
orr-bootleg-2020	299	27	pair	pair	NN
orr-bootleg-2020	299	28	connects	connect	NNS
orr-bootleg-2020	299	29	to	to	IN
orr-bootleg-2020	299	30	a	a	DT
orr-bootleg-2020	299	31	different	different	JJ
orr-bootleg-2020	299	32	entity	entity	NN
orr-bootleg-2020	299	33	that	that	WDT
orr-bootleg-2020	299	34	is	be	VBZ
orr-bootleg-2020	299	35	not	not	RB
orr-bootleg-2020	299	36	present	present	JJ
orr-bootleg-2020	299	37	in	in	IN
orr-bootleg-2020	299	38	the	the	DT
orr-bootleg-2020	299	39	sentence	sentence	NN
orr-bootleg-2020	299	40	.	.	.
orr-bootleg-2020	300	1	We	-PRON-	PRP
orr-bootleg-2020	300	2	find	find	VBP
orr-bootleg-2020	300	3	this	this	DT
orr-bootleg-2020	300	4	occurs	occur	VBZ
orr-bootleg-2020	300	5	in	in	IN
orr-bootleg-2020	300	6	6	6	CD
orr-bootleg-2020	300	7	%	%	NN
orr-bootleg-2020	300	8	of	of	IN
orr-bootleg-2020	300	9	overall	overall	JJ
orr-bootleg-2020	300	10	and	and	CC
orr-bootleg-2020	300	11	7	7	CD
orr-bootleg-2020	300	12	%	%	NN
orr-bootleg-2020	300	13	of	of	IN
orr-bootleg-2020	300	14	tail	tail	NN
orr-bootleg-2020	300	15	errors	error	NNS
orr-bootleg-2020	300	16	.	.	.
orr-bootleg-2020	301	1	This	this	DT
orr-bootleg-2020	301	2	type	type	NN
orr-bootleg-2020	301	3	of	of	IN
orr-bootleg-2020	301	4	error	error	NN
orr-bootleg-2020	301	5	represents	represent	VBZ
orr-bootleg-2020	301	6	a	a	DT
orr-bootleg-2020	301	7	fundamental	fundamental	JJ
orr-bootleg-2020	301	8	limitation	limitation	NN
orr-bootleg-2020	301	9	of	of	IN
orr-bootleg-2020	301	10	Bootleg	Bootleg	NNP
orr-bootleg-2020	301	11	as	as	IN
orr-bootleg-2020	301	12	we	-PRON-	PRP
orr-bootleg-2020	301	13	do	do	VBP
orr-bootleg-2020	301	14	not	not	RB
orr-bootleg-2020	301	15	encode	encode	VB
orr-bootleg-2020	301	16	any	any	DT
orr-bootleg-2020	301	17	form	form	NN
orr-bootleg-2020	301	18	of	of	IN
orr-bootleg-2020	301	19	multi	multi	JJ
orr-bootleg-2020	301	20	-	-	JJ
orr-bootleg-2020	301	21	hop	hop	JJ
orr-bootleg-2020	301	22	reasoning	reasoning	NN
orr-bootleg-2020	301	23	over	over	IN
orr-bootleg-2020	301	24	a	a	DT
orr-bootleg-2020	301	25	KG	kg	NN
orr-bootleg-2020	301	26	in	in	IN
orr-bootleg-2020	301	27	Bootleg	Bootleg	NNP
orr-bootleg-2020	301	28	.	.	.
orr-bootleg-2020	302	1	Our	-PRON-	PRP$
orr-bootleg-2020	302	2	KG	kg	NN
orr-bootleg-2020	302	3	information	information	NN
orr-bootleg-2020	302	4	only	only	RB
orr-bootleg-2020	302	5	encodes	encode	VBZ
orr-bootleg-2020	302	6	single	single	JJ
orr-bootleg-2020	302	7	-	-	HYPH
orr-bootleg-2020	302	8	hop	hop	NN
orr-bootleg-2020	302	9	patterns	pattern	NNS
orr-bootleg-2020	302	10	(	(	-LRB-
orr-bootleg-2020	302	11	i.e.	i.e.	FW
orr-bootleg-2020	302	12	,	,	,
orr-bootleg-2020	302	13	direct	direct	JJ
orr-bootleg-2020	302	14	connections	connection	NNS
orr-bootleg-2020	302	15	)	)	-RRB-
orr-bootleg-2020	302	16	.	.	.
orr-bootleg-2020	303	1	•	•	VB
orr-bootleg-2020	303	2	Exact	exact	JJ
orr-bootleg-2020	303	3	Match	Match	NNP
orr-bootleg-2020	303	4	Bootleg	Bootleg	NNP
orr-bootleg-2020	303	5	struggles	struggle	VBZ
orr-bootleg-2020	303	6	on	on	IN
orr-bootleg-2020	303	7	several	several	JJ
orr-bootleg-2020	303	8	examples	example	NNS
orr-bootleg-2020	303	9	in	in	IN
orr-bootleg-2020	303	10	which	which	WDT
orr-bootleg-2020	303	11	the	the	DT
orr-bootleg-2020	303	12	exact	exact	JJ
orr-bootleg-2020	303	13	entity	entity	NN
orr-bootleg-2020	303	14	title	title	NN
orr-bootleg-2020	303	15	is	be	VBZ
orr-bootleg-2020	303	16	present	present	JJ
orr-bootleg-2020	303	17	in	in	IN
orr-bootleg-2020	303	18	the	the	DT
orr-bootleg-2020	303	19	text	text	NN
orr-bootleg-2020	303	20	.	.	.
orr-bootleg-2020	304	1	Considering	consider	VBG
orr-bootleg-2020	304	2	examples	example	NNS
orr-bootleg-2020	304	3	where	where	WRB
orr-bootleg-2020	304	4	the	the	DT
orr-bootleg-2020	304	5	BERT	BERT	NNP
orr-bootleg-2020	304	6	Baseline	Baseline	NNP
orr-bootleg-2020	304	7	is	be	VBZ
orr-bootleg-2020	304	8	correct	correct	JJ
orr-bootleg-2020	304	9	but	but	CC
orr-bootleg-2020	304	10	Bootleg	Bootleg	NNP
orr-bootleg-2020	304	11	is	be	VBZ
orr-bootleg-2020	304	12	incorrect	incorrect	JJ
orr-bootleg-2020	304	13	,	,	,
orr-bootleg-2020	304	14	in	in	IN
orr-bootleg-2020	304	15	28	28	CD
orr-bootleg-2020	304	16	%	%	NN
orr-bootleg-2020	304	17	of	of	IN
orr-bootleg-2020	304	18	the	the	DT
orr-bootleg-2020	304	19	examples	example	NNS
orr-bootleg-2020	304	20	,	,	,
orr-bootleg-2020	304	21	the	the	DT
orr-bootleg-2020	304	22	textual	textual	JJ
orr-bootleg-2020	304	23	mention	mention	NN
orr-bootleg-2020	304	24	is	be	VBZ
orr-bootleg-2020	304	25	an	an	DT
orr-bootleg-2020	304	26	exact	exact	JJ
orr-bootleg-2020	304	27	match	match	NN
orr-bootleg-2020	304	28	of	of	IN
orr-bootleg-2020	304	29	the	the	DT
orr-bootleg-2020	304	30	entity	entity	NN
orr-bootleg-2020	304	31	title	title	NN
orr-bootleg-2020	304	32	.	.	.
orr-bootleg-2020	305	1	Further	further	RB
orr-bootleg-2020	305	2	,	,	,
orr-bootleg-2020	305	3	32	32	CD
orr-bootleg-2020	305	4	%	%	NN
orr-bootleg-2020	305	5	of	of	IN
orr-bootleg-2020	305	6	the	the	DT
orr-bootleg-2020	305	7	examples	example	NNS
orr-bootleg-2020	305	8	contain	contain	VBP
orr-bootleg-2020	305	9	a	a	DT
orr-bootleg-2020	305	10	keyword	keyword	NN
orr-bootleg-2020	305	11	from	from	IN
orr-bootleg-2020	305	12	the	the	DT
orr-bootleg-2020	305	13	entity	entity	NN
orr-bootleg-2020	305	14	title	title	NN
orr-bootleg-2020	305	15	that	that	WDT
orr-bootleg-2020	305	16	Bootleg	Bootleg	NNP
orr-bootleg-2020	305	17	misses	miss	VBZ
orr-bootleg-2020	305	18	(	(	-LRB-
orr-bootleg-2020	305	19	example	example	NN
orr-bootleg-2020	305	20	in	in	IN
orr-bootleg-2020	305	21	Table	table	NN
orr-bootleg-2020	305	22	8)	8)	CD
orr-bootleg-2020	305	23	.	.	.
orr-bootleg-2020	306	1	We	-PRON-	PRP
orr-bootleg-2020	306	2	attribute	attribute	VBP
orr-bootleg-2020	306	3	this	this	DT
orr-bootleg-2020	306	4	decrease	decrease	NN
orr-bootleg-2020	306	5	in	in	IN
orr-bootleg-2020	306	6	performance	performance	NN
orr-bootleg-2020	306	7	to	to	IN
orr-bootleg-2020	306	8	Bootleg	Bootleg	NNP
orr-bootleg-2020	306	9	’s	’s	POS
orr-bootleg-2020	306	10	regularization	regularization	NN
orr-bootleg-2020	306	11	.	.	.
orr-bootleg-2020	307	1	This	this	DT
orr-bootleg-2020	307	2	mention	mention	NN
orr-bootleg-2020	307	3	-	-	HYPH
orr-bootleg-2020	307	4	to	to	IN
orr-bootleg-2020	307	5	-	-	HYPH
orr-bootleg-2020	307	6	entity	entity	NN
orr-bootleg-2020	307	7	similarity	similarity	NN
orr-bootleg-2020	307	8	would	would	MD
orr-bootleg-2020	307	9	need	need	VB
orr-bootleg-2020	307	10	to	to	TO
orr-bootleg-2020	307	11	be	be	VB
orr-bootleg-2020	307	12	encoded	encode	VBN
orr-bootleg-2020	307	13	in	in	IN
orr-bootleg-2020	307	14	Bootleg	Bootleg	NNP
orr-bootleg-2020	307	15	’s	’s	POS
orr-bootleg-2020	307	16	entity	entity	NN
orr-bootleg-2020	307	17	embedding	embedding	NN
orr-bootleg-2020	307	18	,	,	,
orr-bootleg-2020	307	19	but	but	CC
orr-bootleg-2020	307	20	the	the	DT
orr-bootleg-2020	307	21	regularization	regularization	NN
orr-bootleg-2020	307	22	encourages	encourage	VBZ
orr-bootleg-2020	307	23	Bootleg	Bootleg	NNP
orr-bootleg-2020	307	24	to	to	TO
orr-bootleg-2020	307	25	not	not	RB
orr-bootleg-2020	307	26	use	use	VB
orr-bootleg-2020	307	27	entity	entity	NN
orr-bootleg-2020	307	28	-	-	HYPH
orr-bootleg-2020	307	29	level	level	NN
orr-bootleg-2020	307	30	information	information	NN
orr-bootleg-2020	307	31	.	.	.
orr-bootleg-2020	308	1	6	6	CD
orr-bootleg-2020	308	2	Related	Related	NNP
orr-bootleg-2020	308	3	Work	work	NN
orr-bootleg-2020	308	4	We	-PRON-	PRP
orr-bootleg-2020	308	5	discuss	discuss	VBP
orr-bootleg-2020	308	6	related	related	JJ
orr-bootleg-2020	308	7	work	work	NN
orr-bootleg-2020	308	8	in	in	IN
orr-bootleg-2020	308	9	terms	term	NNS
orr-bootleg-2020	308	10	of	of	IN
orr-bootleg-2020	308	11	both	both	DT
orr-bootleg-2020	308	12	NED	NED	NNP
orr-bootleg-2020	308	13	and	and	CC
orr-bootleg-2020	308	14	the	the	DT
orr-bootleg-2020	308	15	broader	broad	JJR
orr-bootleg-2020	308	16	picture	picture	NN
orr-bootleg-2020	308	17	of	of	IN
orr-bootleg-2020	308	18	self	self	NN
orr-bootleg-2020	308	19	-	-	HYPH
orr-bootleg-2020	308	20	supervised	supervise	VBN
orr-bootleg-2020	308	21	models	model	NNS
orr-bootleg-2020	308	22	and	and	CC
orr-bootleg-2020	308	23	tail	tail	NN
orr-bootleg-2020	308	24	data	datum	NNS
orr-bootleg-2020	308	25	.	.	.
orr-bootleg-2020	309	1	Standard	standard	JJ
orr-bootleg-2020	309	2	,	,	,
orr-bootleg-2020	309	3	pre	pre	JJ
orr-bootleg-2020	309	4	-	-	JJ
orr-bootleg-2020	309	5	deep	deep	JJ
orr-bootleg-2020	309	6	-	-	HYPH
orr-bootleg-2020	309	7	learning	learn	VBG
orr-bootleg-2020	309	8	approaches	approach	NNS
orr-bootleg-2020	309	9	to	to	IN
orr-bootleg-2020	309	10	NED	NED	NNP
orr-bootleg-2020	309	11	have	have	VBP
orr-bootleg-2020	309	12	been	be	VBN
orr-bootleg-2020	309	13	rule	rule	NN
orr-bootleg-2020	309	14	-	-	HYPH
orr-bootleg-2020	309	15	based	base	VBN
orr-bootleg-2020	309	16	[	[	-LRB-
orr-bootleg-2020	309	17	1	1	CD
orr-bootleg-2020	309	18	]	]	-RRB-
orr-bootleg-2020	309	19	or	or	CC
orr-bootleg-2020	309	20	leverage	leverage	VB
orr-bootleg-2020	309	21	statistical	statistical	JJ
orr-bootleg-2020	309	22	techniques	technique	NNS
orr-bootleg-2020	309	23	and	and	CC
orr-bootleg-2020	309	24	manual	manual	JJ
orr-bootleg-2020	309	25	feature	feature	NN
orr-bootleg-2020	309	26	engineering	engineer	VBG
orr-bootleg-2020	309	27	to	to	TO
orr-bootleg-2020	309	28	filter	filter	VB
orr-bootleg-2020	309	29	and	and	CC
orr-bootleg-2020	309	30	rank	rank	NN
orr-bootleg-2020	309	31	candidates	candidate	NNS
orr-bootleg-2020	309	32	[	[	-LRB-
orr-bootleg-2020	309	33	50	50	CD
orr-bootleg-2020	309	34	]	]	-RRB-
orr-bootleg-2020	309	35	.	.	.
orr-bootleg-2020	310	1	For	for	IN
orr-bootleg-2020	310	2	example	example	NN
orr-bootleg-2020	310	3	,	,	,
orr-bootleg-2020	310	4	link	link	NN
orr-bootleg-2020	310	5	counts	count	NNS
orr-bootleg-2020	310	6	and	and	CC
orr-bootleg-2020	310	7	similarity	similarity	NN
orr-bootleg-2020	310	8	scores	score	NNS
orr-bootleg-2020	310	9	between	between	IN
orr-bootleg-2020	310	10	entity	entity	NN
orr-bootleg-2020	310	11	titles	title	NNS
orr-bootleg-2020	310	12	and	and	CC
orr-bootleg-2020	310	13	mention	mention	NN
orr-bootleg-2020	310	14	are	be	VBP
orr-bootleg-2020	310	15	two	two	CD
orr-bootleg-2020	310	16	such	such	JJ
orr-bootleg-2020	310	17	features	feature	NNS
orr-bootleg-2020	310	18	[	[	-LRB-
orr-bootleg-2020	310	19	12	12	CD
orr-bootleg-2020	310	20	]	]	-RRB-
orr-bootleg-2020	310	21	.	.	.
orr-bootleg-2020	311	1	These	these	DT
orr-bootleg-2020	311	2	systems	system	NNS
orr-bootleg-2020	311	3	tend	tend	VBP
orr-bootleg-2020	311	4	to	to	TO
orr-bootleg-2020	311	5	be	be	VB
orr-bootleg-2020	311	6	hard	hard	JJ
orr-bootleg-2020	311	7	to	to	TO
orr-bootleg-2020	311	8	maintain	maintain	VB
orr-bootleg-2020	311	9	over	over	IN
orr-bootleg-2020	311	10	time	time	NN
orr-bootleg-2020	311	11	,	,	,
orr-bootleg-2020	311	12	with	with	IN
orr-bootleg-2020	311	13	the	the	DT
orr-bootleg-2020	311	14	work	work	NN
orr-bootleg-2020	311	15	of	of	IN
orr-bootleg-2020	311	16	Petasis	Petasis	NNP
orr-bootleg-2020	311	17	et	et	NNP
orr-bootleg-2020	311	18	al	al	NNP
orr-bootleg-2020	311	19	.	.	.
orr-bootleg-2020	312	1	[	[	-LRB-
orr-bootleg-2020	312	2	37	37	CD
orr-bootleg-2020	312	3	]	]	-RRB-
orr-bootleg-2020	312	4	building	build	VBG
orr-bootleg-2020	312	5	a	a	DT
orr-bootleg-2020	312	6	model	model	NN
orr-bootleg-2020	312	7	to	to	TO
orr-bootleg-2020	312	8	detect	detect	VB
orr-bootleg-2020	312	9	when	when	WRB
orr-bootleg-2020	312	10	a	a	DT
orr-bootleg-2020	312	11	rule	rule	NN
orr-bootleg-2020	312	12	-	-	HYPH
orr-bootleg-2020	312	13	based	base	VBN
orr-bootleg-2020	312	14	NED	NED	NNP
orr-bootleg-2020	312	15	system	system	NN
orr-bootleg-2020	312	16	needs	need	VBZ
orr-bootleg-2020	312	17	to	to	TO
orr-bootleg-2020	312	18	be	be	VB
orr-bootleg-2020	312	19	retrained	retrain	VBN
orr-bootleg-2020	312	20	and	and	CC
orr-bootleg-2020	312	21	updated	update	VBN
orr-bootleg-2020	312	22	.	.	.
orr-bootleg-2020	313	1	In	in	IN
orr-bootleg-2020	313	2	recent	recent	JJ
orr-bootleg-2020	313	3	years	year	NNS
orr-bootleg-2020	313	4	,	,	,
orr-bootleg-2020	313	5	deep	deep	JJ
orr-bootleg-2020	313	6	learning	learning	NN
orr-bootleg-2020	313	7	systems	system	NNS
orr-bootleg-2020	313	8	have	have	VBP
orr-bootleg-2020	313	9	become	become	VBN
orr-bootleg-2020	313	10	the	the	DT
orr-bootleg-2020	313	11	new	new	JJ
orr-bootleg-2020	313	12	standard	standard	NN
orr-bootleg-2020	313	13	(	(	-LRB-
orr-bootleg-2020	313	14	see	see	VB
orr-bootleg-2020	313	15	Mudgal	Mudgal	NNP
orr-bootleg-2020	313	16	et	et	NNP
orr-bootleg-2020	313	17	al	al	NNP
orr-bootleg-2020	313	18	.	.	.
orr-bootleg-2020	314	1	[	[	-LRB-
orr-bootleg-2020	314	2	32	32	CD
orr-bootleg-2020	314	3	]	]	-RRB-
orr-bootleg-2020	314	4	for	for	IN
orr-bootleg-2020	314	5	a	a	DT
orr-bootleg-2020	314	6	high	high	JJ
orr-bootleg-2020	314	7	-	-	HYPH
orr-bootleg-2020	314	8	level	level	NN
orr-bootleg-2020	314	9	overview	overview	NN
orr-bootleg-2020	314	10	of	of	IN
orr-bootleg-2020	314	11	deep	deep	JJ
orr-bootleg-2020	314	12	learning	learning	NN
orr-bootleg-2020	314	13	approaches	approach	NNS
orr-bootleg-2020	314	14	to	to	IN
orr-bootleg-2020	314	15	entity	entity	NN
orr-bootleg-2020	314	16	disambiguation	disambiguation	NN
orr-bootleg-2020	314	17	and	and	CC
orr-bootleg-2020	314	18	entity	entity	NN
orr-bootleg-2020	314	19	matching	matching	NN
orr-bootleg-2020	314	20	problems	problem	NNS
orr-bootleg-2020	314	21	)	)	-RRB-
orr-bootleg-2020	314	22	.	.	.
orr-bootleg-2020	315	1	The	the	DT
orr-bootleg-2020	315	2	most	most	RBS
orr-bootleg-2020	315	3	recent	recent	JJ
orr-bootleg-2020	315	4	state	state	NN
orr-bootleg-2020	315	5	-	-	HYPH
orr-bootleg-2020	315	6	of	of	IN
orr-bootleg-2020	315	7	-	-	HYPH
orr-bootleg-2020	315	8	the	the	DT
orr-bootleg-2020	315	9	-	-	HYPH
orr-bootleg-2020	315	10	art	art	NN
orr-bootleg-2020	315	11	models	model	NNS
orr-bootleg-2020	315	12	generally	generally	RB
orr-bootleg-2020	315	13	rely	rely	VBP
orr-bootleg-2020	315	14	on	on	IN
orr-bootleg-2020	315	15	deep	deep	JJ
orr-bootleg-2020	315	16	contextual	contextual	JJ
orr-bootleg-2020	315	17	word	word	NN
orr-bootleg-2020	315	18	embeddings	embedding	NNS
orr-bootleg-2020	315	19	with	with	IN
orr-bootleg-2020	315	20	entity	entity	NN
orr-bootleg-2020	315	21	embeddings	embedding	NNS
orr-bootleg-2020	315	22	[	[	-LRB-
orr-bootleg-2020	315	23	16	16	CD
orr-bootleg-2020	315	24	,	,	,
orr-bootleg-2020	315	25	46	46	CD
orr-bootleg-2020	315	26	,	,	,
orr-bootleg-2020	315	27	51	51	CD
orr-bootleg-2020	315	28	]	]	-RRB-
orr-bootleg-2020	315	29	.	.	.
orr-bootleg-2020	316	1	As	as	IN
orr-bootleg-2020	316	2	we	-PRON-	PRP
orr-bootleg-2020	316	3	showed	show	VBD
orr-bootleg-2020	316	4	in	in	IN
orr-bootleg-2020	316	5	Table	table	NN
orr-bootleg-2020	316	6	2	2	CD
orr-bootleg-2020	316	7	,	,	,
orr-bootleg-2020	316	8	these	these	DT
orr-bootleg-2020	316	9	models	model	NNS
orr-bootleg-2020	316	10	perform	perform	VBP
orr-bootleg-2020	316	11	well	well	RB
orr-bootleg-2020	316	12	over	over	IN
orr-bootleg-2020	316	13	popular	popular	JJ
orr-bootleg-2020	316	14	entities	entity	NNS
orr-bootleg-2020	316	15	,	,	,
orr-bootleg-2020	316	16	but	but	CC
orr-bootleg-2020	316	17	struggle	struggle	NN
orr-bootleg-2020	316	18	to	to	TO
orr-bootleg-2020	316	19	resolve	resolve	VB
orr-bootleg-2020	316	20	the	the	DT
orr-bootleg-2020	316	21	tail	tail	NN
orr-bootleg-2020	316	22	.	.	.
orr-bootleg-2020	317	1	Jin	Jin	NNP
orr-bootleg-2020	317	2	et	et	NNP
orr-bootleg-2020	317	3	al	al	NNP
orr-bootleg-2020	317	4	.	.	.
orr-bootleg-2020	318	1	[	[	-LRB-
orr-bootleg-2020	318	2	26	26	CD
orr-bootleg-2020	318	3	]	]	-RRB-
orr-bootleg-2020	318	4	and	and	CC
orr-bootleg-2020	318	5	Hoffart	Hoffart	NNP
orr-bootleg-2020	318	6	et	et	NNP
orr-bootleg-2020	318	7	al	al	NNP
orr-bootleg-2020	318	8	.	.	.
orr-bootleg-2020	319	1	[	[	-LRB-
orr-bootleg-2020	319	2	23	23	CD
orr-bootleg-2020	319	3	]	]	-RRB-
orr-bootleg-2020	319	4	study	study	NN
orr-bootleg-2020	319	5	disambiguation	disambiguation	NN
orr-bootleg-2020	319	6	at	at	IN
orr-bootleg-2020	319	7	the	the	DT
orr-bootleg-2020	319	8	tail	tail	NN
orr-bootleg-2020	319	9	,	,	,
orr-bootleg-2020	319	10	and	and	CC
orr-bootleg-2020	319	11	both	both	DT
orr-bootleg-2020	319	12	rely	rely	VBP
orr-bootleg-2020	319	13	on	on	IN
orr-bootleg-2020	319	14	phrase	phrase	NN
orr-bootleg-2020	319	15	-	-	HYPH
orr-bootleg-2020	319	16	based	base	VBN
orr-bootleg-2020	319	17	language	language	NN
orr-bootleg-2020	319	18	models	model	NNS
orr-bootleg-2020	319	19	for	for	IN
orr-bootleg-2020	319	20	feature	feature	NN
orr-bootleg-2020	319	21	extraction	extraction	NN
orr-bootleg-2020	319	22	.	.	.
orr-bootleg-2020	320	1	Unlike	unlike	IN
orr-bootleg-2020	320	2	our	-PRON-	PRP$
orr-bootleg-2020	320	3	work	work	NN
orr-bootleg-2020	320	4	,	,	,
orr-bootleg-2020	320	5	they	-PRON-	PRP
orr-bootleg-2020	320	6	do	do	VBP
orr-bootleg-2020	320	7	not	not	RB
orr-bootleg-2020	320	8	fuse	fuse	VB
orr-bootleg-2020	320	9	type	type	NN
orr-bootleg-2020	320	10	or	or	CC
orr-bootleg-2020	320	11	knowledge	knowledge	NN
orr-bootleg-2020	320	12	graph	graph	NN
orr-bootleg-2020	320	13	information	information	NN
orr-bootleg-2020	320	14	for	for	IN
orr-bootleg-2020	320	15	disambiguation	disambiguation	NN
orr-bootleg-2020	320	16	.	.	.
orr-bootleg-2020	321	1	13	13	CD
orr-bootleg-2020	321	2	Table	table	NN
orr-bootleg-2020	321	3	8	8	CD
orr-bootleg-2020	321	4	:	:	:
orr-bootleg-2020	321	5	We	-PRON-	PRP
orr-bootleg-2020	321	6	identify	identify	VBP
orr-bootleg-2020	321	7	four	four	CD
orr-bootleg-2020	321	8	key	key	JJ
orr-bootleg-2020	321	9	error	error	NN
orr-bootleg-2020	321	10	buckets	bucket	NNS
orr-bootleg-2020	321	11	for	for	IN
orr-bootleg-2020	321	12	Bootleg	Bootleg	NNP
orr-bootleg-2020	321	13	:	:	:
orr-bootleg-2020	321	14	granularity	granularity	NN
orr-bootleg-2020	321	15	,	,	,
orr-bootleg-2020	321	16	numerical	numerical	JJ
orr-bootleg-2020	321	17	errors	error	NNS
orr-bootleg-2020	321	18	,	,	,
orr-bootleg-2020	321	19	multi	multi	JJ
orr-bootleg-2020	321	20	-	-	JJ
orr-bootleg-2020	321	21	hop	hop	JJ
orr-bootleg-2020	321	22	reasoning	reasoning	NN
orr-bootleg-2020	321	23	,	,	,
orr-bootleg-2020	321	24	and	and	CC
orr-bootleg-2020	321	25	missed	miss	VBN
orr-bootleg-2020	321	26	exact	exact	JJ
orr-bootleg-2020	321	27	matches	match	NNS
orr-bootleg-2020	321	28	.	.	.
orr-bootleg-2020	322	1	We	-PRON-	PRP
orr-bootleg-2020	322	2	provide	provide	VBP
orr-bootleg-2020	322	3	a	a	DT
orr-bootleg-2020	322	4	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	322	5	example	example	NN
orr-bootleg-2020	322	6	,	,	,
orr-bootleg-2020	322	7	the	the	DT
orr-bootleg-2020	322	8	gold	gold	JJ
orr-bootleg-2020	322	9	entity	entity	NN
orr-bootleg-2020	322	10	,	,	,
orr-bootleg-2020	322	11	and	and	CC
orr-bootleg-2020	322	12	Bootleg	Bootleg	NNP
orr-bootleg-2020	322	13	’s	’s	POS
orr-bootleg-2020	322	14	predicted	predict	VBD
orr-bootleg-2020	322	15	entity	entity	NN
orr-bootleg-2020	322	16	for	for	IN
orr-bootleg-2020	322	17	each	each	DT
orr-bootleg-2020	322	18	example	example	NN
orr-bootleg-2020	322	19	.	.	.
orr-bootleg-2020	323	1	Error	error	NN
orr-bootleg-2020	323	2	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	323	3	Example	Example	NNP
orr-bootleg-2020	323	4	Bootleg	Bootleg	NNP
orr-bootleg-2020	323	5	Prediction	Prediction	NNP
orr-bootleg-2020	323	6	Gold	Gold	NNP
orr-bootleg-2020	323	7	Entity	Entity	NNP
orr-bootleg-2020	323	8	Granularity	Granularity	NNP
orr-bootleg-2020	323	9	Posey	Posey	NNP
orr-bootleg-2020	323	10	is	be	VBZ
orr-bootleg-2020	323	11	the	the	DT
orr-bootleg-2020	323	12	recipient	recipient	NN
orr-bootleg-2020	323	13	of	of	IN
orr-bootleg-2020	323	14	a	a	DT
orr-bootleg-2020	323	15	Golden	Golden	NNP
orr-bootleg-2020	323	16	Globe	Globe	NNP
orr-bootleg-2020	323	17	Award	Award	NNP
orr-bootleg-2020	323	18	nomination	nomination	NN
orr-bootleg-2020	323	19	,	,	,
orr-bootleg-2020	323	20	a	a	DT
orr-bootleg-2020	323	21	Satel-	Satel-	NNP
orr-bootleg-2020	323	22	lite	lite	NN
orr-bootleg-2020	323	23	Award	Award	NNP
orr-bootleg-2020	323	24	nomination	nomination	NN
orr-bootleg-2020	323	25	and	and	CC
orr-bootleg-2020	323	26	two	two	CD
orr-bootleg-2020	323	27	In-	in-	JJ
orr-bootleg-2020	323	28	dependent	dependent	JJ
orr-bootleg-2020	323	29	Spirit	Spirit	NNP
orr-bootleg-2020	323	30	Award	Award	NNP
orr-bootleg-2020	323	31	nominations	nomination	NNS
orr-bootleg-2020	323	32	.	.	.
orr-bootleg-2020	324	1	Satellite	Satellite	NNP
orr-bootleg-2020	324	2	Awards	Awards	NNPS
orr-bootleg-2020	324	3	Satellite	Satellite	NNP
orr-bootleg-2020	324	4	Award	Award	NNP
orr-bootleg-2020	324	5	for	for	IN
orr-bootleg-2020	324	6	Best	Best	NNP
orr-bootleg-2020	324	7	Actress	Actress	NNP
orr-bootleg-2020	324	8	–	–	:
orr-bootleg-2020	324	9	Motion	Motion	NNP
orr-bootleg-2020	324	10	Picture	Picture	NNP
orr-bootleg-2020	324	11	Numerical	Numerical	NNP
orr-bootleg-2020	324	12	He	-PRON-	PRP
orr-bootleg-2020	324	13	competed	compete	VBD
orr-bootleg-2020	324	14	in	in	IN
orr-bootleg-2020	324	15	the	the	DT
orr-bootleg-2020	324	16	individual	individual	JJ
orr-bootleg-2020	324	17	road	road	NN
orr-bootleg-2020	324	18	race	race	NN
orr-bootleg-2020	324	19	and	and	CC
orr-bootleg-2020	324	20	team	team	NN
orr-bootleg-2020	324	21	time	time	NN
orr-bootleg-2020	324	22	trial	trial	NN
orr-bootleg-2020	324	23	events	event	NNS
orr-bootleg-2020	324	24	at	at	IN
orr-bootleg-2020	324	25	the	the	DT
orr-bootleg-2020	324	26	1976	1976	CD
orr-bootleg-2020	324	27	Summer	Summer	NNP
orr-bootleg-2020	324	28	Olympics	Olympics	NNPS
orr-bootleg-2020	324	29	.	.	.
orr-bootleg-2020	325	1	Cycling	cycle	VBG
orr-bootleg-2020	325	2	at	at	IN
orr-bootleg-2020	325	3	the	the	DT
orr-bootleg-2020	325	4	1960	1960	CD
orr-bootleg-2020	325	5	Summer	Summer	NNP
orr-bootleg-2020	325	6	Olympics	Olympics	NNPS
orr-bootleg-2020	325	7	–	–	:
orr-bootleg-2020	325	8	1960	1960	CD
orr-bootleg-2020	325	9	Men	Men	NNPS
orr-bootleg-2020	325	10	’s	’s	POS
orr-bootleg-2020	325	11	Road	Road	NNP
orr-bootleg-2020	325	12	Race	Race	NNP
orr-bootleg-2020	325	13	Cycling	Cycling	NNP
orr-bootleg-2020	325	14	at	at	IN
orr-bootleg-2020	325	15	the	the	DT
orr-bootleg-2020	325	16	1976	1976	CD
orr-bootleg-2020	325	17	Sum-	Sum-	NNP
orr-bootleg-2020	325	18	mer	mer	NNP
orr-bootleg-2020	325	19	Olympics	Olympics	NNP
orr-bootleg-2020	325	20	–	–	:
orr-bootleg-2020	325	21	1976	1976	CD
orr-bootleg-2020	325	22	Men	Men	NNPS
orr-bootleg-2020	325	23	’s	’s	POS
orr-bootleg-2020	325	24	Road	Road	NNP
orr-bootleg-2020	325	25	Race	Race	NNP
orr-bootleg-2020	325	26	Multi	Multi	NNP
orr-bootleg-2020	325	27	-	-	JJ
orr-bootleg-2020	325	28	hop	hop	JJ
orr-bootleg-2020	325	29	Other	other	JJ
orr-bootleg-2020	325	30	nearby	nearby	JJ
orr-bootleg-2020	325	31	historic	historic	JJ
orr-bootleg-2020	325	32	buildings	building	NNS
orr-bootleg-2020	325	33	in-	in-	VBG
orr-bootleg-2020	325	34	clude	clude	VB
orr-bootleg-2020	325	35	the	the	DT
orr-bootleg-2020	325	36	Santa	Santa	NNP
orr-bootleg-2020	325	37	Fe	Fe	NNP
orr-bootleg-2020	325	38	Depot	Depot	NNP
orr-bootleg-2020	325	39	,	,	,
orr-bootleg-2020	325	40	the	the	DT
orr-bootleg-2020	325	41	Cit-	Cit-	NNP
orr-bootleg-2020	325	42	izens	izen	VBZ
orr-bootleg-2020	325	43	Bank	Bank	NNP
orr-bootleg-2020	325	44	Building	Building	NNP
orr-bootleg-2020	325	45	,	,	,
orr-bootleg-2020	325	46	the	the	DT
orr-bootleg-2020	325	47	Hoke	Hoke	NNP
orr-bootleg-2020	325	48	Building	Building	NNP
orr-bootleg-2020	325	49	,	,	,
orr-bootleg-2020	325	50	the	the	DT
orr-bootleg-2020	325	51	Walker	Walker	NNP
orr-bootleg-2020	325	52	Building	Building	NNP
orr-bootleg-2020	325	53	,	,	,
orr-bootleg-2020	325	54	and	and	CC
orr-bootleg-2020	325	55	the	the	DT
orr-bootleg-2020	325	56	Courthouse	Courthouse	NNP
orr-bootleg-2020	325	57	Citizens	Citizens	NNPS
orr-bootleg-2020	325	58	Bank	Bank	NNP
orr-bootleg-2020	325	59	Build-	Build-	NNP
orr-bootleg-2020	325	60	ing	ing	NNP
orr-bootleg-2020	325	61	(	(	-LRB-
orr-bootleg-2020	325	62	Burnsville	Burnsville	NNP
orr-bootleg-2020	325	63	,	,	,
orr-bootleg-2020	325	64	North	North	NNP
orr-bootleg-2020	325	65	Carolina	Carolina	NNP
orr-bootleg-2020	325	66	)	)	-RRB-
orr-bootleg-2020	325	67	Citizens	Citizens	NNPS
orr-bootleg-2020	325	68	Bank	Bank	NNP
orr-bootleg-2020	325	69	Building	Building	NNP
orr-bootleg-2020	325	70	(	(	-LRB-
orr-bootleg-2020	325	71	Stillwater	Stillwater	NNP
orr-bootleg-2020	325	72	,	,	,
orr-bootleg-2020	325	73	Oklahoma	Oklahoma	NNP
orr-bootleg-2020	325	74	)	)	-RRB-
orr-bootleg-2020	325	75	Exact	Exact	NNP
orr-bootleg-2020	325	76	Match	Match	NNP
orr-bootleg-2020	325	77	According	accord	VBG
orr-bootleg-2020	325	78	to	to	IN
orr-bootleg-2020	325	79	the	the	DT
orr-bootleg-2020	325	80	Nielsen	Nielsen	NNP
orr-bootleg-2020	325	81	Media	Media	NNP
orr-bootleg-2020	325	82	Research	Research	NNP
orr-bootleg-2020	325	83	,	,	,
orr-bootleg-2020	325	84	the	the	DT
orr-bootleg-2020	325	85	episode	episode	NN
orr-bootleg-2020	325	86	was	be	VBD
orr-bootleg-2020	325	87	watched	watch	VBN
orr-bootleg-2020	325	88	by	by	IN
orr-bootleg-2020	325	89	469	469	CD
orr-bootleg-2020	325	90	million	million	CD
orr-bootleg-2020	325	91	viewers	viewer	NNS
orr-bootleg-2020	325	92	...	...	:
orr-bootleg-2020	325	93	Nielsen	Nielsen	NNP
orr-bootleg-2020	325	94	ratings	rating	NNS
orr-bootleg-2020	325	95	Nielsen	Nielsen	NNP
orr-bootleg-2020	325	96	Media	Media	NNP
orr-bootleg-2020	325	97	Research	Research	NNP
orr-bootleg-2020	325	98	Disambiguation	Disambiguation	NNP
orr-bootleg-2020	325	99	with	with	IN
orr-bootleg-2020	325	100	Types	Types	NNP
orr-bootleg-2020	325	101	Similar	Similar	NNP
orr-bootleg-2020	325	102	to	to	IN
orr-bootleg-2020	325	103	our	-PRON-	PRP$
orr-bootleg-2020	325	104	work	work	NN
orr-bootleg-2020	325	105	,	,	,
orr-bootleg-2020	325	106	recent	recent	JJ
orr-bootleg-2020	325	107	approaches	approach	NNS
orr-bootleg-2020	325	108	have	have	VBP
orr-bootleg-2020	325	109	found	find	VBN
orr-bootleg-2020	325	110	that	that	DT
orr-bootleg-2020	325	111	type	type	NN
orr-bootleg-2020	325	112	information	information	NN
orr-bootleg-2020	325	113	can	can	MD
orr-bootleg-2020	325	114	be	be	VB
orr-bootleg-2020	325	115	useful	useful	JJ
orr-bootleg-2020	325	116	for	for	IN
orr-bootleg-2020	325	117	entity	entity	NN
orr-bootleg-2020	325	118	disambiguation	disambiguation	NN
orr-bootleg-2020	325	119	[	[	-LRB-
orr-bootleg-2020	325	120	9	9	CD
orr-bootleg-2020	325	121	,	,	,
orr-bootleg-2020	325	122	14	14	CD
orr-bootleg-2020	325	123	,	,	,
orr-bootleg-2020	325	124	21	21	CD
orr-bootleg-2020	325	125	,	,	,
orr-bootleg-2020	325	126	31	31	CD
orr-bootleg-2020	325	127	,	,	,
orr-bootleg-2020	325	128	43	43	CD
orr-bootleg-2020	325	129	,	,	,
orr-bootleg-2020	325	130	55	55	CD
orr-bootleg-2020	325	131	]	]	-RRB-
orr-bootleg-2020	325	132	.	.	.
orr-bootleg-2020	326	1	Dredze	Dredze	NNP
orr-bootleg-2020	326	2	et	et	NNP
orr-bootleg-2020	326	3	al	al	NNP
orr-bootleg-2020	326	4	.	.	.
orr-bootleg-2020	327	1	[	[	-LRB-
orr-bootleg-2020	327	2	14	14	CD
orr-bootleg-2020	327	3	]	]	-RRB-
orr-bootleg-2020	327	4	use	use	NN
orr-bootleg-2020	327	5	predicted	predict	VBN
orr-bootleg-2020	327	6	coarse	coarse	RB
orr-bootleg-2020	327	7	-	-	HYPH
orr-bootleg-2020	327	8	grained	grain	VBN
orr-bootleg-2020	327	9	types	type	NNS
orr-bootleg-2020	327	10	as	as	IN
orr-bootleg-2020	327	11	entity	entity	NN
orr-bootleg-2020	327	12	features	feature	VBZ
orr-bootleg-2020	327	13	into	into	IN
orr-bootleg-2020	327	14	a	a	DT
orr-bootleg-2020	327	15	SVM	SVM	NNP
orr-bootleg-2020	327	16	classifier	classifier	NN
orr-bootleg-2020	327	17	.	.	.
orr-bootleg-2020	328	1	Chen	Chen	NNP
orr-bootleg-2020	328	2	et	et	FW
orr-bootleg-2020	328	3	al	al	NNP
orr-bootleg-2020	328	4	.	.	.
orr-bootleg-2020	329	1	[	[	-LRB-
orr-bootleg-2020	329	2	9	9	CD
orr-bootleg-2020	329	3	]	]	-RRB-
orr-bootleg-2020	329	4	models	model	NNS
orr-bootleg-2020	329	5	type	type	VB
orr-bootleg-2020	329	6	information	information	NN
orr-bootleg-2020	329	7	as	as	IN
orr-bootleg-2020	329	8	local	local	JJ
orr-bootleg-2020	329	9	context	context	NN
orr-bootleg-2020	329	10	and	and	CC
orr-bootleg-2020	329	11	integrates	integrate	VBZ
orr-bootleg-2020	329	12	a	a	DT
orr-bootleg-2020	329	13	BERT	BERT	NNP
orr-bootleg-2020	329	14	contextual	contextual	JJ
orr-bootleg-2020	329	15	embedding	embedding	NN
orr-bootleg-2020	329	16	into	into	IN
orr-bootleg-2020	329	17	the	the	DT
orr-bootleg-2020	329	18	model	model	NN
orr-bootleg-2020	329	19	from	from	IN
orr-bootleg-2020	329	20	[	[	-LRB-
orr-bootleg-2020	329	21	17	17	CD
orr-bootleg-2020	329	22	]	]	-RRB-
orr-bootleg-2020	329	23	.	.	.
orr-bootleg-2020	330	1	Raiman	Raiman	NNP
orr-bootleg-2020	330	2	and	and	CC
orr-bootleg-2020	330	3	Raiman	Raiman	NNP
orr-bootleg-2020	330	4	[	[	-LRB-
orr-bootleg-2020	330	5	43	43	CD
orr-bootleg-2020	330	6	]	]	-RRB-
orr-bootleg-2020	330	7	learns	learn	VBZ
orr-bootleg-2020	330	8	its	-PRON-	PRP$
orr-bootleg-2020	330	9	own	own	JJ
orr-bootleg-2020	330	10	type	type	NN
orr-bootleg-2020	330	11	systems	system	NNS
orr-bootleg-2020	330	12	and	and	CC
orr-bootleg-2020	330	13	performs	perform	VBZ
orr-bootleg-2020	330	14	disambiguation	disambiguation	NNP
orr-bootleg-2020	330	15	through	through	IN
orr-bootleg-2020	330	16	type	type	NN
orr-bootleg-2020	330	17	prediction	prediction	NN
orr-bootleg-2020	330	18	alone	alone	RB
orr-bootleg-2020	330	19	(	(	-LRB-
orr-bootleg-2020	330	20	essentially	essentially	RB
orr-bootleg-2020	330	21	capturing	capture	VBG
orr-bootleg-2020	330	22	the	the	DT
orr-bootleg-2020	330	23	type	type	NN
orr-bootleg-2020	330	24	affordance	affordance	NN
orr-bootleg-2020	330	25	pattern	pattern	NN
orr-bootleg-2020	330	26	)	)	-RRB-
orr-bootleg-2020	330	27	.	.	.
orr-bootleg-2020	331	1	Ling	Ling	NNP
orr-bootleg-2020	331	2	et	et	FW
orr-bootleg-2020	331	3	al	al	NNP
orr-bootleg-2020	331	4	.	.	.
orr-bootleg-2020	332	1	[	[	-LRB-
orr-bootleg-2020	332	2	31	31	CD
orr-bootleg-2020	332	3	]	]	-RRB-
orr-bootleg-2020	332	4	demonstrate	demonstrate	VBP
orr-bootleg-2020	332	5	that	that	IN
orr-bootleg-2020	332	6	the	the	DT
orr-bootleg-2020	332	7	112-type	112-type	CD
orr-bootleg-2020	332	8	FIGER	FIGER	NNP
orr-bootleg-2020	332	9	type	type	NN
orr-bootleg-2020	332	10	ontology	ontology	NN
orr-bootleg-2020	332	11	could	could	MD
orr-bootleg-2020	332	12	improve	improve	VB
orr-bootleg-2020	332	13	entity	entity	NN
orr-bootleg-2020	332	14	disambiguation	disambiguation	NN
orr-bootleg-2020	332	15	,	,	,
orr-bootleg-2020	332	16	and	and	CC
orr-bootleg-2020	332	17	the	the	DT
orr-bootleg-2020	332	18	LATTE	LATTE	NNP
orr-bootleg-2020	332	19	framework	framework	NN
orr-bootleg-2020	332	20	[	[	-LRB-
orr-bootleg-2020	332	21	55	55	CD
orr-bootleg-2020	332	22	]	]	-RRB-
orr-bootleg-2020	332	23	uses	use	VBZ
orr-bootleg-2020	332	24	multi	multi	JJ
orr-bootleg-2020	332	25	-	-	JJ
orr-bootleg-2020	332	26	task	task	JJ
orr-bootleg-2020	332	27	learning	learn	VBG
orr-bootleg-2020	332	28	to	to	TO
orr-bootleg-2020	332	29	jointly	jointly	RB
orr-bootleg-2020	332	30	perform	perform	VB
orr-bootleg-2020	332	31	type	type	NN
orr-bootleg-2020	332	32	classification	classification	NN
orr-bootleg-2020	332	33	and	and	CC
orr-bootleg-2020	332	34	entity	entity	NN
orr-bootleg-2020	332	35	disambiguation	disambiguation	NN
orr-bootleg-2020	332	36	on	on	IN
orr-bootleg-2020	332	37	biomedical	biomedical	JJ
orr-bootleg-2020	332	38	data	datum	NNS
orr-bootleg-2020	332	39	.	.	.
orr-bootleg-2020	333	1	Gupta	Gupta	NNP
orr-bootleg-2020	333	2	et	et	FW
orr-bootleg-2020	333	3	al	al	NNP
orr-bootleg-2020	333	4	.	.	.
orr-bootleg-2020	334	1	[	[	-LRB-
orr-bootleg-2020	334	2	21	21	CD
orr-bootleg-2020	334	3	]	]	-RRB-
orr-bootleg-2020	334	4	adds	add	VBZ
orr-bootleg-2020	334	5	both	both	CC
orr-bootleg-2020	334	6	an	an	DT
orr-bootleg-2020	334	7	entity	entity	NN
orr-bootleg-2020	334	8	-	-	HYPH
orr-bootleg-2020	334	9	level	level	NN
orr-bootleg-2020	334	10	and	and	CC
orr-bootleg-2020	334	11	mention	mention	NN
orr-bootleg-2020	334	12	-	-	HYPH
orr-bootleg-2020	334	13	level	level	NN
orr-bootleg-2020	334	14	type	type	NN
orr-bootleg-2020	334	15	objective	objective	NN
orr-bootleg-2020	334	16	using	use	VBG
orr-bootleg-2020	334	17	type	type	NN
orr-bootleg-2020	334	18	embeddings	embedding	NNS
orr-bootleg-2020	334	19	embeddings	embedding	NNS
orr-bootleg-2020	334	20	.	.	.
orr-bootleg-2020	335	1	We	-PRON-	PRP
orr-bootleg-2020	335	2	build	build	VBP
orr-bootleg-2020	335	3	on	on	IN
orr-bootleg-2020	335	4	these	these	DT
orr-bootleg-2020	335	5	works	work	NNS
orr-bootleg-2020	335	6	using	use	VBG
orr-bootleg-2020	335	7	fine	fine	JJ
orr-bootleg-2020	335	8	and	and	CC
orr-bootleg-2020	335	9	coarse	coarse	JJ
orr-bootleg-2020	335	10	-	-	HYPH
orr-bootleg-2020	335	11	grained	grain	VBN
orr-bootleg-2020	335	12	entity	entity	NN
orr-bootleg-2020	335	13	-	-	HYPH
orr-bootleg-2020	335	14	level	level	NN
orr-bootleg-2020	335	15	type	type	NN
orr-bootleg-2020	335	16	embeddings	embedding	NNS
orr-bootleg-2020	335	17	and	and	CC
orr-bootleg-2020	335	18	a	a	DT
orr-bootleg-2020	335	19	mention	mention	JJ
orr-bootleg-2020	335	20	-	-	HYPH
orr-bootleg-2020	335	21	level	level	NN
orr-bootleg-2020	335	22	type	type	NN
orr-bootleg-2020	335	23	prediction	prediction	NN
orr-bootleg-2020	335	24	task	task	NN
orr-bootleg-2020	335	25	.	.	.
orr-bootleg-2020	336	1	Disambiguation	disambiguation	NN
orr-bootleg-2020	336	2	with	with	IN
orr-bootleg-2020	336	3	Knowledge	Knowledge	NNP
orr-bootleg-2020	336	4	Graphs	Graphs	NNP
orr-bootleg-2020	336	5	Several	several	JJ
orr-bootleg-2020	336	6	recent	recent	JJ
orr-bootleg-2020	336	7	works	work	NNS
orr-bootleg-2020	336	8	have	have	VBP
orr-bootleg-2020	336	9	also	also	RB
orr-bootleg-2020	336	10	incorporated	incorporate	VBN
orr-bootleg-2020	336	11	(	(	-LRB-
orr-bootleg-2020	336	12	knowledge	knowledge	NN
orr-bootleg-2020	336	13	)	)	-RRB-
orr-bootleg-2020	336	14	graph	graph	NN
orr-bootleg-2020	336	15	information	information	NN
orr-bootleg-2020	336	16	through	through	IN
orr-bootleg-2020	336	17	graph	graph	NN
orr-bootleg-2020	336	18	embeddings	embedding	NNS
orr-bootleg-2020	336	19	[	[	-LRB-
orr-bootleg-2020	336	20	35	35	CD
orr-bootleg-2020	336	21	]	]	-RRB-
orr-bootleg-2020	336	22	,	,	,
orr-bootleg-2020	336	23	co	co	NNS
orr-bootleg-2020	336	24	-	-	NNS
orr-bootleg-2020	336	25	occurrences	occurrence	NNS
orr-bootleg-2020	336	26	in	in	IN
orr-bootleg-2020	336	27	the	the	DT
orr-bootleg-2020	336	28	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	336	29	hyperlink	hyperlink	NN
orr-bootleg-2020	336	30	graph	graph	NN
orr-bootleg-2020	336	31	[	[	-LRB-
orr-bootleg-2020	336	32	42	42	CD
orr-bootleg-2020	336	33	]	]	-RRB-
orr-bootleg-2020	336	34	,	,	,
orr-bootleg-2020	336	35	and	and	CC
orr-bootleg-2020	336	36	the	the	DT
orr-bootleg-2020	336	37	incorporation	incorporation	NN
orr-bootleg-2020	336	38	of	of	IN
orr-bootleg-2020	336	39	latent	latent	JJ
orr-bootleg-2020	336	40	relation	relation	NN
orr-bootleg-2020	336	41	variables	variable	NNS
orr-bootleg-2020	336	42	[	[	-LRB-
orr-bootleg-2020	336	43	30	30	CD
orr-bootleg-2020	336	44	]	]	-RRB-
orr-bootleg-2020	336	45	to	to	TO
orr-bootleg-2020	336	46	aid	aid	VB
orr-bootleg-2020	336	47	disambiguation	disambiguation	NN
orr-bootleg-2020	336	48	.	.	.
orr-bootleg-2020	337	1	Cetoli	Cetoli	NNP
orr-bootleg-2020	337	2	et	et	FW
orr-bootleg-2020	337	3	al	al	NNP
orr-bootleg-2020	337	4	.	.	.
orr-bootleg-2020	338	1	[	[	-LRB-
orr-bootleg-2020	338	2	8	8	CD
orr-bootleg-2020	338	3	]	]	-RRB-
orr-bootleg-2020	338	4	and	and	CC
orr-bootleg-2020	338	5	Mulang	Mulang	NNP
orr-bootleg-2020	338	6	et	et	FW
orr-bootleg-2020	338	7	al	al	NNP
orr-bootleg-2020	338	8	.	.	.
orr-bootleg-2020	339	1	[	[	-LRB-
orr-bootleg-2020	339	2	33	33	CD
orr-bootleg-2020	339	3	]	]	-RRB-
orr-bootleg-2020	339	4	incorporate	incorporate	VB
orr-bootleg-2020	339	5	Wikidata	wikidata	NN
orr-bootleg-2020	339	6	triples	triple	NNS
orr-bootleg-2020	339	7	as	as	IN
orr-bootleg-2020	339	8	context	context	NN
orr-bootleg-2020	339	9	into	into	IN
orr-bootleg-2020	339	10	entity	entity	NN
orr-bootleg-2020	339	11	disambiguation	disambiguation	NN
orr-bootleg-2020	339	12	by	by	IN
orr-bootleg-2020	339	13	encoding	encode	VBG
orr-bootleg-2020	339	14	triples	triple	NNS
orr-bootleg-2020	339	15	as	as	IN
orr-bootleg-2020	339	16	textual	textual	JJ
orr-bootleg-2020	339	17	phrases	phrase	NNS
orr-bootleg-2020	339	18	(	(	-LRB-
orr-bootleg-2020	339	19	e.g.	e.g.	RB
orr-bootleg-2020	339	20	,	,	,
orr-bootleg-2020	339	21	“	"	``
orr-bootleg-2020	339	22	<	<	XX
orr-bootleg-2020	339	23	subject	subject	JJ
orr-bootleg-2020	339	24	>	>	XX
orr-bootleg-2020	339	25	<	<	XX
orr-bootleg-2020	339	26	predict	predict	VBP
orr-bootleg-2020	339	27	>	>	XX
orr-bootleg-2020	339	28	<	<	XX
orr-bootleg-2020	339	29	object	object	VBP
orr-bootleg-2020	339	30	>	>	XX
orr-bootleg-2020	339	31	”	"	''
orr-bootleg-2020	339	32	)	)	-RRB-
orr-bootleg-2020	339	33	to	to	TO
orr-bootleg-2020	339	34	use	use	VB
orr-bootleg-2020	339	35	as	as	IN
orr-bootleg-2020	339	36	additional	additional	JJ
orr-bootleg-2020	339	37	inputs	input	NNS
orr-bootleg-2020	339	38	,	,	,
orr-bootleg-2020	339	39	along	along	IN
orr-bootleg-2020	339	40	with	with	IN
orr-bootleg-2020	339	41	the	the	DT
orr-bootleg-2020	339	42	original	original	JJ
orr-bootleg-2020	339	43	text	text	NN
orr-bootleg-2020	339	44	to	to	TO
orr-bootleg-2020	339	45	disambiguate	disambiguate	VB
orr-bootleg-2020	339	46	,	,	,
orr-bootleg-2020	339	47	into	into	IN
orr-bootleg-2020	339	48	a	a	DT
orr-bootleg-2020	339	49	language	language	NN
orr-bootleg-2020	339	50	model	model	NN
orr-bootleg-2020	339	51	.	.	.
orr-bootleg-2020	340	1	In	in	IN
orr-bootleg-2020	340	2	Bootleg	Bootleg	NNP
orr-bootleg-2020	340	3	,	,	,
orr-bootleg-2020	340	4	the	the	DT
orr-bootleg-2020	340	5	Wikidata	Wikidata	NNP
orr-bootleg-2020	340	6	connections	connection	NNS
orr-bootleg-2020	340	7	through	through	IN
orr-bootleg-2020	340	8	the	the	DT
orr-bootleg-2020	340	9	KG2Ent	kg2ent	NN
orr-bootleg-2020	340	10	module	module	NNP
orr-bootleg-2020	340	11	allow	allow	VB
orr-bootleg-2020	340	12	for	for	IN
orr-bootleg-2020	340	13	collective	collective	JJ
orr-bootleg-2020	340	14	resolution	resolution	NN
orr-bootleg-2020	340	15	and	and	CC
orr-bootleg-2020	340	16	are	be	VBP
orr-bootleg-2020	340	17	not	not	RB
orr-bootleg-2020	340	18	just	just	RB
orr-bootleg-2020	340	19	additional	additional	JJ
orr-bootleg-2020	340	20	features	feature	NNS
orr-bootleg-2020	340	21	.	.	.
orr-bootleg-2020	341	1	Entity	entity	NN
orr-bootleg-2020	341	2	Knowledge	knowledge	NN
orr-bootleg-2020	341	3	in	in	IN
orr-bootleg-2020	341	4	Downstream	Downstream	NNP
orr-bootleg-2020	341	5	Tasks	task	NNS
orr-bootleg-2020	341	6	The	the	DT
orr-bootleg-2020	341	7	works	work	NNS
orr-bootleg-2020	341	8	of	of	IN
orr-bootleg-2020	341	9	Broscheit	Broscheit	NNP
orr-bootleg-2020	341	10	[	[	-LRB-
orr-bootleg-2020	341	11	6	6	CD
orr-bootleg-2020	341	12	]	]	-RRB-
orr-bootleg-2020	341	13	,	,	,
orr-bootleg-2020	341	14	Peters	Peters	NNP
orr-bootleg-2020	341	15	et	et	FW
orr-bootleg-2020	341	16	al	al	NNP
orr-bootleg-2020	341	17	.	.	.
orr-bootleg-2020	342	1	[	[	-LRB-
orr-bootleg-2020	342	2	38	38	CD
orr-bootleg-2020	342	3	]	]	-RRB-
orr-bootleg-2020	342	4	,	,	,
orr-bootleg-2020	342	5	Poerner	Poerner	NNP
orr-bootleg-2020	342	6	et	et	NNP
orr-bootleg-2020	342	7	al	al	NNP
orr-bootleg-2020	342	8	.	.	.
orr-bootleg-2020	343	1	[	[	-LRB-
orr-bootleg-2020	343	2	41	41	CD
orr-bootleg-2020	343	3	]	]	-RRB-
orr-bootleg-2020	343	4	,	,	,
orr-bootleg-2020	343	5	Zhang	Zhang	NNP
orr-bootleg-2020	343	6	et	et	NNP
orr-bootleg-2020	343	7	al	al	NNP
orr-bootleg-2020	343	8	.	.	.
orr-bootleg-2020	344	1	[	[	-LRB-
orr-bootleg-2020	344	2	54	54	CD
orr-bootleg-2020	344	3	]	]	-RRB-
orr-bootleg-2020	344	4	all	all	DT
orr-bootleg-2020	344	5	try	try	VBP
orr-bootleg-2020	344	6	to	to	TO
orr-bootleg-2020	344	7	add	add	VB
orr-bootleg-2020	344	8	entity	entity	NN
orr-bootleg-2020	344	9	knowledge	knowledge	NN
orr-bootleg-2020	344	10	into	into	IN
orr-bootleg-2020	344	11	a	a	DT
orr-bootleg-2020	344	12	deep	deep	JJ
orr-bootleg-2020	344	13	language	language	NN
orr-bootleg-2020	344	14	model	model	NN
orr-bootleg-2020	344	15	to	to	TO
orr-bootleg-2020	344	16	improve	improve	VB
orr-bootleg-2020	344	17	downstream	downstream	JJ
orr-bootleg-2020	344	18	natural	natural	JJ
orr-bootleg-2020	344	19	language	language	NN
orr-bootleg-2020	344	20	task	task	NN
orr-bootleg-2020	344	21	performance	performance	NN
orr-bootleg-2020	344	22	.	.	.
orr-bootleg-2020	345	1	Peters	Peters	NNP
orr-bootleg-2020	345	2	et	et	FW
orr-bootleg-2020	345	3	al	al	NNP
orr-bootleg-2020	345	4	.	.	.
orr-bootleg-2020	346	1	[	[	-LRB-
orr-bootleg-2020	346	2	38	38	CD
orr-bootleg-2020	346	3	]	]	-RRB-
orr-bootleg-2020	346	4	,	,	,
orr-bootleg-2020	346	5	Poerner	Poerner	NNP
orr-bootleg-2020	346	6	et	et	NNP
orr-bootleg-2020	346	7	al	al	NNP
orr-bootleg-2020	346	8	.	.	.
orr-bootleg-2020	347	1	[	[	-LRB-
orr-bootleg-2020	347	2	41	41	CD
orr-bootleg-2020	347	3	]	]	-RRB-
orr-bootleg-2020	347	4	,	,	,
orr-bootleg-2020	347	5	Zhang	Zhang	NNP
orr-bootleg-2020	347	6	et	et	NNP
orr-bootleg-2020	347	7	al	al	NNP
orr-bootleg-2020	347	8	.	.	.
orr-bootleg-2020	348	1	[	[	-LRB-
orr-bootleg-2020	348	2	54	54	CD
orr-bootleg-2020	348	3	]	]	-RRB-
orr-bootleg-2020	348	4	incorporate	incorporate	VB
orr-bootleg-2020	348	5	pretrained	pretrained	JJ
orr-bootleg-2020	348	6	entity	entity	NN
orr-bootleg-2020	348	7	embeddings	embedding	NNS
orr-bootleg-2020	348	8	and	and	CC
orr-bootleg-2020	348	9	finetune	finetune	VB
orr-bootleg-2020	348	10	either	either	CC
orr-bootleg-2020	348	11	on	on	IN
orr-bootleg-2020	348	12	a	a	DT
orr-bootleg-2020	348	13	the	the	DT
orr-bootleg-2020	348	14	standard	standard	JJ
orr-bootleg-2020	348	15	masked	masked	JJ
orr-bootleg-2020	348	16	sequence	sequence	NN
orr-bootleg-2020	348	17	-	-	HYPH
orr-bootleg-2020	348	18	to	to	IN
orr-bootleg-2020	348	19	-	-	HYPH
orr-bootleg-2020	348	20	sequence	sequence	NN
orr-bootleg-2020	348	21	prediction	prediction	NN
orr-bootleg-2020	348	22	task	task	NN
orr-bootleg-2020	348	23	or	or	CC
orr-bootleg-2020	348	24	combined	combine	VBN
orr-bootleg-2020	348	25	with	with	IN
orr-bootleg-2020	348	26	an	an	DT
orr-bootleg-2020	348	27	entity	entity	NN
orr-bootleg-2020	348	28	disambiguation	disambiguation	NN
orr-bootleg-2020	348	29	/	/	,
orr-bootleg-2020	348	30	linking	link	VBG
orr-bootleg-2020	348	31	task.10	task.10	NNP
orr-bootleg-2020	348	32	On	on	IN
orr-bootleg-2020	348	33	the	the	DT
orr-bootleg-2020	348	34	other	other	JJ
orr-bootleg-2020	348	35	hand	hand	NN
orr-bootleg-2020	348	36	,	,	,
orr-bootleg-2020	348	37	Broscheit	Broscheit	NNP
orr-bootleg-2020	348	38	[	[	-LRB-
orr-bootleg-2020	348	39	6	6	CD
orr-bootleg-2020	348	40	]	]	-RRB-
orr-bootleg-2020	348	41	trains	train	VBZ
orr-bootleg-2020	348	42	its	-PRON-	PRP$
orr-bootleg-2020	348	43	own	own	JJ
orr-bootleg-2020	348	44	entity	entity	NN
orr-bootleg-2020	348	45	embeddings	embedding	NNS
orr-bootleg-2020	348	46	.	.	.
orr-bootleg-2020	349	1	Most	Most	JJS
orr-bootleg-2020	349	2	works	work	VBZ
orr-bootleg-2020	349	3	,	,	,
orr-bootleg-2020	349	4	like	like	IN
orr-bootleg-2020	349	5	Bootleg	Bootleg	NNP
orr-bootleg-2020	349	6	,	,	,
orr-bootleg-2020	349	7	see	see	VB
orr-bootleg-2020	349	8	lift	lift	NN
orr-bootleg-2020	349	9	from	from	IN
orr-bootleg-2020	349	10	incorporating	incorporate	VBG
orr-bootleg-2020	349	11	entity	entity	NN
orr-bootleg-2020	349	12	representations	representation	NNS
orr-bootleg-2020	349	13	in	in	IN
orr-bootleg-2020	349	14	the	the	DT
orr-bootleg-2020	349	15	downstream	downstream	JJ
orr-bootleg-2020	349	16	tasks	task	NNS
orr-bootleg-2020	349	17	.	.	.
orr-bootleg-2020	350	1	10Entity	10entity	CD
orr-bootleg-2020	350	2	disambiguation	disambiguation	NN
orr-bootleg-2020	350	3	refers	refer	VBZ
orr-bootleg-2020	350	4	to	to	IN
orr-bootleg-2020	350	5	when	when	WRB
orr-bootleg-2020	350	6	the	the	DT
orr-bootleg-2020	350	7	mentions	mention	NNS
orr-bootleg-2020	350	8	are	be	VBP
orr-bootleg-2020	350	9	pre	pre	JJ
orr-bootleg-2020	350	10	-	-	VBN
orr-bootleg-2020	350	11	detected	detect	VBN
orr-bootleg-2020	350	12	in	in	IN
orr-bootleg-2020	350	13	text	text	NN
orr-bootleg-2020	350	14	.	.	.
orr-bootleg-2020	351	1	Entity	entity	NN
orr-bootleg-2020	351	2	linking	linking	NN
orr-bootleg-2020	351	3	includes	include	VBZ
orr-bootleg-2020	351	4	the	the	DT
orr-bootleg-2020	351	5	mention	mention	NN
orr-bootleg-2020	351	6	detection	detection	NN
orr-bootleg-2020	351	7	phase	phase	NN
orr-bootleg-2020	351	8	.	.	.
orr-bootleg-2020	352	1	In	in	IN
orr-bootleg-2020	352	2	Bootleg	Bootleg	NNP
orr-bootleg-2020	352	3	,	,	,
orr-bootleg-2020	352	4	we	-PRON-	PRP
orr-bootleg-2020	352	5	focus	focus	VBP
orr-bootleg-2020	352	6	on	on	IN
orr-bootleg-2020	352	7	the	the	DT
orr-bootleg-2020	352	8	entity	entity	NN
orr-bootleg-2020	352	9	disambiguation	disambiguation	NN
orr-bootleg-2020	352	10	task	task	NN
orr-bootleg-2020	352	11	.	.	.
orr-bootleg-2020	353	1	14	14	CD
orr-bootleg-2020	353	2	Wikipedia	wikipedia	NN
orr-bootleg-2020	353	3	Weak	weak	JJ
orr-bootleg-2020	353	4	Labelling	labelling	NN
orr-bootleg-2020	353	5	Although	although	IN
orr-bootleg-2020	353	6	uncommon	uncommon	JJ
orr-bootleg-2020	353	7	,	,	,
orr-bootleg-2020	353	8	Broscheit	Broscheit	NNP
orr-bootleg-2020	353	9	[	[	-LRB-
orr-bootleg-2020	353	10	6	6	CD
orr-bootleg-2020	353	11	]	]	-RRB-
orr-bootleg-2020	353	12	,	,	,
orr-bootleg-2020	353	13	De	De	NNP
orr-bootleg-2020	353	14	Cao	Cao	NNP
orr-bootleg-2020	353	15	et	et	NNP
orr-bootleg-2020	353	16	al	al	NNP
orr-bootleg-2020	353	17	.	.	.
orr-bootleg-2020	354	1	[	[	-LRB-
orr-bootleg-2020	354	2	13	13	CD
orr-bootleg-2020	354	3	]	]	-RRB-
orr-bootleg-2020	354	4	,	,	,
orr-bootleg-2020	354	5	Ghaddar	Ghaddar	NNP
orr-bootleg-2020	354	6	and	and	CC
orr-bootleg-2020	354	7	Langlais	Langlais	NNP
orr-bootleg-2020	354	8	[	[	-LRB-
orr-bootleg-2020	354	9	19	19	CD
orr-bootleg-2020	354	10	]	]	-RRB-
orr-bootleg-2020	354	11	,	,	,
orr-bootleg-2020	354	12	Nothman	Nothman	NNP
orr-bootleg-2020	354	13	et	et	NNP
orr-bootleg-2020	354	14	al	al	NNP
orr-bootleg-2020	354	15	.	.	.
orr-bootleg-2020	355	1	[	[	-LRB-
orr-bootleg-2020	355	2	34	34	CD
orr-bootleg-2020	355	3	]	]	-RRB-
orr-bootleg-2020	355	4	all	all	RB
orr-bootleg-2020	355	5	apply	apply	VBP
orr-bootleg-2020	355	6	some	some	DT
orr-bootleg-2020	355	7	heuristic	heuristic	JJ
orr-bootleg-2020	355	8	weak	weak	JJ
orr-bootleg-2020	355	9	labelling	labelling	NN
orr-bootleg-2020	355	10	techniques	technique	NNS
orr-bootleg-2020	355	11	to	to	TO
orr-bootleg-2020	355	12	increase	increase	VB
orr-bootleg-2020	355	13	link	link	NN
orr-bootleg-2020	355	14	coverage	coverage	NN
orr-bootleg-2020	355	15	in	in	IN
orr-bootleg-2020	355	16	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	355	17	for	for	IN
orr-bootleg-2020	355	18	either	either	DT
orr-bootleg-2020	355	19	entity	entity	NN
orr-bootleg-2020	355	20	disambiguation	disambiguation	NN
orr-bootleg-2020	355	21	or	or	CC
orr-bootleg-2020	355	22	named	name	VBN
orr-bootleg-2020	355	23	entity	entity	NN
orr-bootleg-2020	355	24	recognition	recognition	NN
orr-bootleg-2020	355	25	.	.	.
orr-bootleg-2020	356	1	All	all	DT
orr-bootleg-2020	356	2	methods	method	NNS
orr-bootleg-2020	356	3	generally	generally	RB
orr-bootleg-2020	356	4	rely	rely	VBP
orr-bootleg-2020	356	5	on	on	IN
orr-bootleg-2020	356	6	finding	find	VBG
orr-bootleg-2020	356	7	known	know	VBN
orr-bootleg-2020	356	8	surface	surface	NN
orr-bootleg-2020	356	9	forms	form	NNS
orr-bootleg-2020	356	10	for	for	IN
orr-bootleg-2020	356	11	entities	entity	NNS
orr-bootleg-2020	356	12	and	and	CC
orr-bootleg-2020	356	13	labelling	label	VBG
orr-bootleg-2020	356	14	those	those	DT
orr-bootleg-2020	356	15	in	in	IN
orr-bootleg-2020	356	16	the	the	DT
orr-bootleg-2020	356	17	text	text	NN
orr-bootleg-2020	356	18	.	.	.
orr-bootleg-2020	357	1	Bootleg	Bootleg	NNP
orr-bootleg-2020	357	2	is	be	VBZ
orr-bootleg-2020	357	3	the	the	DT
orr-bootleg-2020	357	4	first	first	JJ
orr-bootleg-2020	357	5	to	to	TO
orr-bootleg-2020	357	6	investigate	investigate	VB
orr-bootleg-2020	357	7	the	the	DT
orr-bootleg-2020	357	8	lift	lift	NN
orr-bootleg-2020	357	9	from	from	IN
orr-bootleg-2020	357	10	incorporating	incorporate	VBG
orr-bootleg-2020	357	11	weakly	weakly	RB
orr-bootleg-2020	357	12	labelled	label	VBN
orr-bootleg-2020	357	13	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	357	14	data	datum	NNS
orr-bootleg-2020	357	15	over	over	IN
orr-bootleg-2020	357	16	the	the	DT
orr-bootleg-2020	357	17	tail	tail	NN
orr-bootleg-2020	357	18	.	.	.
orr-bootleg-2020	358	1	Self	Self	NNP
orr-bootleg-2020	358	2	-	-	HYPH
orr-bootleg-2020	358	3	Supervision	Supervision	NNP
orr-bootleg-2020	358	4	and	and	CC
orr-bootleg-2020	358	5	the	the	DT
orr-bootleg-2020	358	6	Tail	Tail	NNP
orr-bootleg-2020	358	7	The	the	DT
orr-bootleg-2020	358	8	works	work	NNS
orr-bootleg-2020	358	9	of	of	IN
orr-bootleg-2020	358	10	Tata	Tata	NNP
orr-bootleg-2020	358	11	et	et	NNP
orr-bootleg-2020	358	12	al	al	NNP
orr-bootleg-2020	358	13	.	.	.
orr-bootleg-2020	359	1	[	[	-LRB-
orr-bootleg-2020	359	2	47	47	CD
orr-bootleg-2020	359	3	]	]	-RRB-
orr-bootleg-2020	359	4	,	,	,
orr-bootleg-2020	359	5	Chung	Chung	NNP
orr-bootleg-2020	359	6	et	et	NNP
orr-bootleg-2020	359	7	al	al	NNP
orr-bootleg-2020	359	8	.	.	.
orr-bootleg-2020	360	1	[	[	-LRB-
orr-bootleg-2020	360	2	11	11	CD
orr-bootleg-2020	360	3	]	]	-RRB-
orr-bootleg-2020	360	4	,	,	,
orr-bootleg-2020	360	5	Ilievski	Ilievski	NNP
orr-bootleg-2020	360	6	et	et	NNP
orr-bootleg-2020	360	7	al	al	NNP
orr-bootleg-2020	360	8	.	.	.
orr-bootleg-2020	361	1	[	[	-LRB-
orr-bootleg-2020	361	2	25	25	CD
orr-bootleg-2020	361	3	]	]	-RRB-
orr-bootleg-2020	361	4	,	,	,
orr-bootleg-2020	361	5	and	and	CC
orr-bootleg-2020	361	6	Chung	Chung	NNP
orr-bootleg-2020	361	7	et	et	NNP
orr-bootleg-2020	361	8	al	al	NNP
orr-bootleg-2020	361	9	.	.	.
orr-bootleg-2020	362	1	[	[	-LRB-
orr-bootleg-2020	362	2	10	10	CD
orr-bootleg-2020	362	3	]	]	-RRB-
orr-bootleg-2020	362	4	all	all	DT
orr-bootleg-2020	362	5	focus	focus	VBP
orr-bootleg-2020	362	6	on	on	IN
orr-bootleg-2020	362	7	the	the	DT
orr-bootleg-2020	362	8	importance	importance	NN
orr-bootleg-2020	362	9	of	of	IN
orr-bootleg-2020	362	10	the	the	DT
orr-bootleg-2020	362	11	tail	tail	NN
orr-bootleg-2020	362	12	during	during	IN
orr-bootleg-2020	362	13	inference	inference	NN
orr-bootleg-2020	362	14	and	and	CC
orr-bootleg-2020	362	15	the	the	DT
orr-bootleg-2020	362	16	challenges	challenge	NNS
orr-bootleg-2020	362	17	of	of	IN
orr-bootleg-2020	362	18	capturing	capture	VBG
orr-bootleg-2020	362	19	it	-PRON-	PRP
orr-bootleg-2020	362	20	during	during	IN
orr-bootleg-2020	362	21	training	training	NN
orr-bootleg-2020	362	22	.	.	.
orr-bootleg-2020	363	1	They	-PRON-	PRP
orr-bootleg-2020	363	2	all	all	DT
orr-bootleg-2020	363	3	highlight	highlight	VBP
orr-bootleg-2020	363	4	the	the	DT
orr-bootleg-2020	363	5	data	data	NN
orr-bootleg-2020	363	6	management	management	NN
orr-bootleg-2020	363	7	challenges	challenge	NNS
orr-bootleg-2020	363	8	of	of	IN
orr-bootleg-2020	363	9	monitoring	monitor	VBG
orr-bootleg-2020	363	10	the	the	DT
orr-bootleg-2020	363	11	tail	tail	NN
orr-bootleg-2020	363	12	(	(	-LRB-
orr-bootleg-2020	363	13	and	and	CC
orr-bootleg-2020	363	14	other	other	JJ
orr-bootleg-2020	363	15	missed	miss	VBN
orr-bootleg-2020	363	16	slices	slice	NNS
orr-bootleg-2020	363	17	of	of	IN
orr-bootleg-2020	363	18	data	datum	NNS
orr-bootleg-2020	363	19	)	)	-RRB-
orr-bootleg-2020	363	20	and	and	CC
orr-bootleg-2020	363	21	improving	improve	VBG
orr-bootleg-2020	363	22	generalizability	generalizability	NN
orr-bootleg-2020	363	23	.	.	.
orr-bootleg-2020	364	1	In	in	IN
orr-bootleg-2020	364	2	particular	particular	JJ
orr-bootleg-2020	364	3	,	,	,
orr-bootleg-2020	364	4	Ilievski	Ilievski	NNP
orr-bootleg-2020	364	5	et	et	NNP
orr-bootleg-2020	364	6	al	al	NNP
orr-bootleg-2020	364	7	.	.	.
orr-bootleg-2020	365	1	[	[	-LRB-
orr-bootleg-2020	365	2	25	25	CD
orr-bootleg-2020	365	3	]	]	-RRB-
orr-bootleg-2020	365	4	studies	study	NNS
orr-bootleg-2020	365	5	the	the	DT
orr-bootleg-2020	365	6	tail	tail	NN
orr-bootleg-2020	365	7	in	in	IN
orr-bootleg-2020	365	8	NED	NED	NNP
orr-bootleg-2020	365	9	and	and	CC
orr-bootleg-2020	365	10	encourages	encourage	VBZ
orr-bootleg-2020	365	11	the	the	DT
orr-bootleg-2020	365	12	use	use	NN
orr-bootleg-2020	365	13	of	of	IN
orr-bootleg-2020	365	14	separate	separate	JJ
orr-bootleg-2020	365	15	head	head	NN
orr-bootleg-2020	365	16	and	and	CC
orr-bootleg-2020	365	17	tail	tail	NN
orr-bootleg-2020	365	18	subsets	subset	NNS
orr-bootleg-2020	365	19	of	of	IN
orr-bootleg-2020	365	20	data	datum	NNS
orr-bootleg-2020	365	21	.	.	.
orr-bootleg-2020	366	1	From	from	IN
orr-bootleg-2020	366	2	a	a	DT
orr-bootleg-2020	366	3	broader	broad	JJR
orr-bootleg-2020	366	4	perspective	perspective	NN
orr-bootleg-2020	366	5	of	of	IN
orr-bootleg-2020	366	6	natural	natural	JJ
orr-bootleg-2020	366	7	language	language	NN
orr-bootleg-2020	366	8	systems	system	NNS
orr-bootleg-2020	366	9	and	and	CC
orr-bootleg-2020	366	10	generalizability	generalizability	NN
orr-bootleg-2020	366	11	,	,	,
orr-bootleg-2020	366	12	Ettinger	Ettinger	NNP
orr-bootleg-2020	366	13	et	et	FW
orr-bootleg-2020	366	14	al	al	NNP
orr-bootleg-2020	366	15	.	.	.
orr-bootleg-2020	367	1	[	[	-LRB-
orr-bootleg-2020	367	2	15	15	CD
orr-bootleg-2020	367	3	]	]	-RRB-
orr-bootleg-2020	367	4	highlights	highlight	NNS
orr-bootleg-2020	367	5	that	that	IN
orr-bootleg-2020	367	6	many	many	JJ
orr-bootleg-2020	367	7	NLP	NLP	NNP
orr-bootleg-2020	367	8	systems	system	NNS
orr-bootleg-2020	367	9	are	be	VBP
orr-bootleg-2020	367	10	brittle	brittle	JJ
orr-bootleg-2020	367	11	in	in	IN
orr-bootleg-2020	367	12	the	the	DT
orr-bootleg-2020	367	13	face	face	NN
orr-bootleg-2020	367	14	of	of	IN
orr-bootleg-2020	367	15	tail	tail	JJ
orr-bootleg-2020	367	16	linguistic	linguistic	JJ
orr-bootleg-2020	367	17	patterns	pattern	NNS
orr-bootleg-2020	367	18	.	.	.
orr-bootleg-2020	368	1	Bootleg	bootleg	NN
orr-bootleg-2020	368	2	builds	build	VBZ
orr-bootleg-2020	368	3	off	off	RP
orr-bootleg-2020	368	4	this	this	DT
orr-bootleg-2020	368	5	work	work	NN
orr-bootleg-2020	368	6	,	,	,
orr-bootleg-2020	368	7	investigating	investigate	VBG
orr-bootleg-2020	368	8	the	the	DT
orr-bootleg-2020	368	9	tail	tail	NN
orr-bootleg-2020	368	10	with	with	IN
orr-bootleg-2020	368	11	respect	respect	NN
orr-bootleg-2020	368	12	to	to	IN
orr-bootleg-2020	368	13	NED	NED	NNP
orr-bootleg-2020	368	14	and	and	CC
orr-bootleg-2020	368	15	demonstrating	demonstrate	VBG
orr-bootleg-2020	368	16	the	the	DT
orr-bootleg-2020	368	17	generalizable	generalizable	JJ
orr-bootleg-2020	368	18	reasoning	reasoning	NN
orr-bootleg-2020	368	19	patterns	pattern	NNS
orr-bootleg-2020	368	20	over	over	IN
orr-bootleg-2020	368	21	structural	structural	JJ
orr-bootleg-2020	368	22	resources	resource	NNS
orr-bootleg-2020	368	23	can	can	MD
orr-bootleg-2020	368	24	aid	aid	VB
orr-bootleg-2020	368	25	tail	tail	NN
orr-bootleg-2020	368	26	disambiguation	disambiguation	NN
orr-bootleg-2020	368	27	.	.	.
orr-bootleg-2020	369	1	7	7	LS
orr-bootleg-2020	369	2	Conclusion	Conclusion	NNP
orr-bootleg-2020	369	3	We	-PRON-	PRP
orr-bootleg-2020	369	4	present	present	VBP
orr-bootleg-2020	369	5	Bootleg	Bootleg	NNP
orr-bootleg-2020	369	6	,	,	,
orr-bootleg-2020	369	7	a	a	DT
orr-bootleg-2020	369	8	state	state	NN
orr-bootleg-2020	369	9	-	-	HYPH
orr-bootleg-2020	369	10	of	of	IN
orr-bootleg-2020	369	11	-	-	HYPH
orr-bootleg-2020	369	12	the	the	DT
orr-bootleg-2020	369	13	-	-	HYPH
orr-bootleg-2020	369	14	art	art	NN
orr-bootleg-2020	369	15	NED	NED	NNP
orr-bootleg-2020	369	16	system	system	NN
orr-bootleg-2020	369	17	that	that	WDT
orr-bootleg-2020	369	18	is	be	VBZ
orr-bootleg-2020	369	19	explicitly	explicitly	RB
orr-bootleg-2020	369	20	grounded	ground	VBN
orr-bootleg-2020	369	21	in	in	IN
orr-bootleg-2020	369	22	a	a	DT
orr-bootleg-2020	369	23	principled	principle	VBN
orr-bootleg-2020	369	24	set	set	NN
orr-bootleg-2020	369	25	of	of	IN
orr-bootleg-2020	369	26	reasoning	reason	VBG
orr-bootleg-2020	369	27	patterns	pattern	NNS
orr-bootleg-2020	369	28	for	for	IN
orr-bootleg-2020	369	29	disambiguation	disambiguation	NN
orr-bootleg-2020	369	30	,	,	,
orr-bootleg-2020	369	31	defined	define	VBN
orr-bootleg-2020	369	32	over	over	IN
orr-bootleg-2020	369	33	entities	entity	NNS
orr-bootleg-2020	369	34	,	,	,
orr-bootleg-2020	369	35	types	type	NNS
orr-bootleg-2020	369	36	,	,	,
orr-bootleg-2020	369	37	and	and	CC
orr-bootleg-2020	369	38	knowledge	knowledge	NN
orr-bootleg-2020	369	39	graph	graph	NN
orr-bootleg-2020	369	40	relations	relation	NNS
orr-bootleg-2020	369	41	.	.	.
orr-bootleg-2020	370	1	The	the	DT
orr-bootleg-2020	370	2	contributions	contribution	NNS
orr-bootleg-2020	370	3	of	of	IN
orr-bootleg-2020	370	4	this	this	DT
orr-bootleg-2020	370	5	work	work	NN
orr-bootleg-2020	370	6	include	include	VBP
orr-bootleg-2020	370	7	the	the	DT
orr-bootleg-2020	370	8	characterization	characterization	NN
orr-bootleg-2020	370	9	and	and	CC
orr-bootleg-2020	370	10	evaluation	evaluation	NN
orr-bootleg-2020	370	11	of	of	IN
orr-bootleg-2020	370	12	core	core	NN
orr-bootleg-2020	370	13	reasoning	reasoning	NN
orr-bootleg-2020	370	14	patterns	pattern	NNS
orr-bootleg-2020	370	15	for	for	IN
orr-bootleg-2020	370	16	disambiguation	disambiguation	NN
orr-bootleg-2020	370	17	,	,	,
orr-bootleg-2020	370	18	a	a	DT
orr-bootleg-2020	370	19	new	new	JJ
orr-bootleg-2020	370	20	learning	learning	NN
orr-bootleg-2020	370	21	procedure	procedure	NN
orr-bootleg-2020	370	22	to	to	TO
orr-bootleg-2020	370	23	encourage	encourage	VB
orr-bootleg-2020	370	24	the	the	DT
orr-bootleg-2020	370	25	model	model	NN
orr-bootleg-2020	370	26	to	to	TO
orr-bootleg-2020	370	27	learn	learn	VB
orr-bootleg-2020	370	28	the	the	DT
orr-bootleg-2020	370	29	patterns	pattern	NNS
orr-bootleg-2020	370	30	,	,	,
orr-bootleg-2020	370	31	and	and	CC
orr-bootleg-2020	370	32	a	a	DT
orr-bootleg-2020	370	33	weak	weak	JJ
orr-bootleg-2020	370	34	supervision	supervision	NN
orr-bootleg-2020	370	35	technique	technique	NN
orr-bootleg-2020	370	36	to	to	TO
orr-bootleg-2020	370	37	increase	increase	VB
orr-bootleg-2020	370	38	utilization	utilization	NN
orr-bootleg-2020	370	39	of	of	IN
orr-bootleg-2020	370	40	the	the	DT
orr-bootleg-2020	370	41	training	training	NN
orr-bootleg-2020	370	42	data	datum	NNS
orr-bootleg-2020	370	43	.	.	.
orr-bootleg-2020	371	1	We	-PRON-	PRP
orr-bootleg-2020	371	2	find	find	VBP
orr-bootleg-2020	371	3	that	that	IN
orr-bootleg-2020	371	4	Bootleg	Bootleg	NNP
orr-bootleg-2020	371	5	improves	improve	VBZ
orr-bootleg-2020	371	6	over	over	IN
orr-bootleg-2020	371	7	the	the	DT
orr-bootleg-2020	371	8	baseline	baseline	JJ
orr-bootleg-2020	371	9	SotA	SotA	NNP
orr-bootleg-2020	371	10	model	model	NN
orr-bootleg-2020	371	11	by	by	IN
orr-bootleg-2020	371	12	over	over	IN
orr-bootleg-2020	371	13	40	40	CD
orr-bootleg-2020	371	14	F1	f1	NN
orr-bootleg-2020	371	15	points	point	NNS
orr-bootleg-2020	371	16	on	on	IN
orr-bootleg-2020	371	17	the	the	DT
orr-bootleg-2020	371	18	tail	tail	NN
orr-bootleg-2020	371	19	of	of	IN
orr-bootleg-2020	371	20	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	371	21	.	.	.
orr-bootleg-2020	372	1	Using	use	VBG
orr-bootleg-2020	372	2	Bootleg	Bootleg	NNP
orr-bootleg-2020	372	3	’s	’s	POS
orr-bootleg-2020	372	4	entity	entity	NN
orr-bootleg-2020	372	5	embeddings	embedding	NNS
orr-bootleg-2020	372	6	for	for	IN
orr-bootleg-2020	372	7	a	a	DT
orr-bootleg-2020	372	8	downstream	downstream	JJ
orr-bootleg-2020	372	9	relation	relation	NN
orr-bootleg-2020	372	10	extraction	extraction	NN
orr-bootleg-2020	372	11	task	task	NN
orr-bootleg-2020	372	12	improves	improve	VBZ
orr-bootleg-2020	372	13	performance	performance	NN
orr-bootleg-2020	372	14	by	by	IN
orr-bootleg-2020	372	15	1.0	1.0	CD
orr-bootleg-2020	372	16	F1	f1	NN
orr-bootleg-2020	372	17	points	point	NNS
orr-bootleg-2020	372	18	,	,	,
orr-bootleg-2020	372	19	and	and	CC
orr-bootleg-2020	372	20	Bootleg	Bootleg	NNP
orr-bootleg-2020	372	21	’s	’s	POS
orr-bootleg-2020	372	22	representations	representation	NNS
orr-bootleg-2020	372	23	lead	lead	VBP
orr-bootleg-2020	372	24	to	to	IN
orr-bootleg-2020	372	25	an	an	DT
orr-bootleg-2020	372	26	8	8	CD
orr-bootleg-2020	372	27	%	%	NN
orr-bootleg-2020	372	28	lift	lift	NN
orr-bootleg-2020	372	29	on	on	IN
orr-bootleg-2020	372	30	highly	highly	RB
orr-bootleg-2020	372	31	optimized	optimized	JJ
orr-bootleg-2020	372	32	production	production	NN
orr-bootleg-2020	372	33	tasks	task	NNS
orr-bootleg-2020	372	34	at	at	IN
orr-bootleg-2020	372	35	a	a	DT
orr-bootleg-2020	372	36	major	major	JJ
orr-bootleg-2020	372	37	technology	technology	NN
orr-bootleg-2020	372	38	company	company	NN
orr-bootleg-2020	372	39	.	.	.
orr-bootleg-2020	373	1	We	-PRON-	PRP
orr-bootleg-2020	373	2	hope	hope	VBP
orr-bootleg-2020	373	3	this	this	DT
orr-bootleg-2020	373	4	work	work	NN
orr-bootleg-2020	373	5	inspires	inspire	VBZ
orr-bootleg-2020	373	6	future	future	JJ
orr-bootleg-2020	373	7	research	research	NN
orr-bootleg-2020	373	8	on	on	IN
orr-bootleg-2020	373	9	improving	improve	VBG
orr-bootleg-2020	373	10	tail	tail	NN
orr-bootleg-2020	373	11	performance	performance	NN
orr-bootleg-2020	373	12	by	by	IN
orr-bootleg-2020	373	13	incorporating	incorporate	VBG
orr-bootleg-2020	373	14	outside	outside	JJ
orr-bootleg-2020	373	15	knowledge	knowledge	NN
orr-bootleg-2020	373	16	in	in	IN
orr-bootleg-2020	373	17	deep	deep	JJ
orr-bootleg-2020	373	18	models	model	NNS
orr-bootleg-2020	373	19	.	.	.
orr-bootleg-2020	374	1	Acknowledgements	acknowledgement	NNS
orr-bootleg-2020	374	2	:	:	:
orr-bootleg-2020	374	3	We	-PRON-	PRP
orr-bootleg-2020	374	4	thank	thank	VBP
orr-bootleg-2020	374	5	Jared	Jared	NNP
orr-bootleg-2020	374	6	Dunnmon	Dunnmon	NNP
orr-bootleg-2020	374	7	,	,	,
orr-bootleg-2020	374	8	Dan	Dan	NNP
orr-bootleg-2020	374	9	Fu	Fu	NNP
orr-bootleg-2020	374	10	,	,	,
orr-bootleg-2020	374	11	Karan	Karan	NNP
orr-bootleg-2020	374	12	Goel	Goel	NNP
orr-bootleg-2020	374	13	,	,	,
orr-bootleg-2020	374	14	Sarah	Sarah	NNP
orr-bootleg-2020	374	15	Hooper	Hooper	NNP
orr-bootleg-2020	374	16	,	,	,
orr-bootleg-2020	374	17	Monica	Monica	NNP
orr-bootleg-2020	374	18	Lam	Lam	NNP
orr-bootleg-2020	374	19	,	,	,
orr-bootleg-2020	374	20	Fred	Fred	NNP
orr-bootleg-2020	374	21	Sala	Sala	NNP
orr-bootleg-2020	374	22	,	,	,
orr-bootleg-2020	374	23	Nimit	Nimit	NNP
orr-bootleg-2020	374	24	Sohoni	Sohoni	NNP
orr-bootleg-2020	374	25	,	,	,
orr-bootleg-2020	374	26	and	and	CC
orr-bootleg-2020	374	27	Silei	Silei	NNP
orr-bootleg-2020	374	28	Xu	Xu	NNP
orr-bootleg-2020	374	29	for	for	IN
orr-bootleg-2020	374	30	their	-PRON-	PRP$
orr-bootleg-2020	374	31	valuable	valuable	JJ
orr-bootleg-2020	374	32	feedback	feedback	NN
orr-bootleg-2020	374	33	and	and	CC
orr-bootleg-2020	374	34	Pallavi	pallavi	JJ
orr-bootleg-2020	374	35	Gudipati	Gudipati	NNP
orr-bootleg-2020	374	36	for	for	IN
orr-bootleg-2020	374	37	help	help	NN
orr-bootleg-2020	374	38	with	with	IN
orr-bootleg-2020	374	39	experiments	experiment	NNS
orr-bootleg-2020	374	40	.	.	.
orr-bootleg-2020	375	1	We	-PRON-	PRP
orr-bootleg-2020	375	2	gratefully	gratefully	RB
orr-bootleg-2020	375	3	acknowledge	acknowledge	VBP
orr-bootleg-2020	375	4	the	the	DT
orr-bootleg-2020	375	5	support	support	NN
orr-bootleg-2020	375	6	of	of	IN
orr-bootleg-2020	375	7	DARPA	DARPA	NNP
orr-bootleg-2020	375	8	under	under	IN
orr-bootleg-2020	375	9	Nos	Nos	NNP
orr-bootleg-2020	375	10	.	.	.
orr-bootleg-2020	376	1	FA86501827865	FA86501827865	NNP
orr-bootleg-2020	376	2	(	(	-LRB-
orr-bootleg-2020	376	3	SDH	SDH	NNP
orr-bootleg-2020	376	4	)	)	-RRB-
orr-bootleg-2020	376	5	and	and	CC
orr-bootleg-2020	376	6	FA86501827882	FA86501827882	NNP
orr-bootleg-2020	376	7	(	(	-LRB-
orr-bootleg-2020	376	8	ASED	ased	NN
orr-bootleg-2020	376	9	)	)	-RRB-
orr-bootleg-2020	376	10	;	;	:
orr-bootleg-2020	376	11	NIH	NIH	NNP
orr-bootleg-2020	376	12	under	under	IN
orr-bootleg-2020	376	13	No	no	NN
orr-bootleg-2020	376	14	.	.	.
orr-bootleg-2020	377	1	U54EB020405	U54EB020405	NNP
orr-bootleg-2020	377	2	(	(	-LRB-
orr-bootleg-2020	377	3	Mobilize	Mobilize	NNP
orr-bootleg-2020	377	4	)	)	-RRB-
orr-bootleg-2020	377	5	,	,	,
orr-bootleg-2020	377	6	NSF	NSF	NNP
orr-bootleg-2020	377	7	under	under	IN
orr-bootleg-2020	377	8	Nos	Nos	NNP
orr-bootleg-2020	377	9	.	.	.
orr-bootleg-2020	378	1	CCF1763315	CCF1763315	NNP
orr-bootleg-2020	378	2	(	(	-LRB-
orr-bootleg-2020	378	3	Beyond	beyond	IN
orr-bootleg-2020	378	4	Sparsity	Sparsity	NNP
orr-bootleg-2020	378	5	)	)	-RRB-
orr-bootleg-2020	378	6	,	,	,
orr-bootleg-2020	378	7	CCF1563078	CCF1563078	NNP
orr-bootleg-2020	378	8	(	(	-LRB-
orr-bootleg-2020	378	9	Volume	volume	NN
orr-bootleg-2020	378	10	to	to	IN
orr-bootleg-2020	378	11	Velocity	Velocity	NNP
orr-bootleg-2020	378	12	)	)	-RRB-
orr-bootleg-2020	378	13	,	,	,
orr-bootleg-2020	378	14	and	and	CC
orr-bootleg-2020	378	15	1937301	1937301	CD
orr-bootleg-2020	378	16	(	(	-LRB-
orr-bootleg-2020	378	17	RTML	RTML	NNP
orr-bootleg-2020	378	18	)	)	-RRB-
orr-bootleg-2020	378	19	;	;	:
orr-bootleg-2020	378	20	ONR	ONR	NNP
orr-bootleg-2020	378	21	under	under	IN
orr-bootleg-2020	378	22	No	no	NN
orr-bootleg-2020	378	23	.	.	.
orr-bootleg-2020	379	1	N000141712266	N000141712266	NNP
orr-bootleg-2020	379	2	(	(	-LRB-
orr-bootleg-2020	379	3	Unifying	unify	VBG
orr-bootleg-2020	379	4	Weak	weak	JJ
orr-bootleg-2020	379	5	Supervi-	Supervi-	NNP
orr-bootleg-2020	379	6	sion	sion	NN
orr-bootleg-2020	379	7	)	)	-RRB-
orr-bootleg-2020	379	8	;	;	:
orr-bootleg-2020	379	9	the	the	DT
orr-bootleg-2020	379	10	Moore	Moore	NNP
orr-bootleg-2020	379	11	Foundation	Foundation	NNP
orr-bootleg-2020	379	12	,	,	,
orr-bootleg-2020	379	13	NXP	NXP	NNP
orr-bootleg-2020	379	14	,	,	,
orr-bootleg-2020	379	15	Xilinx	Xilinx	NNP
orr-bootleg-2020	379	16	,	,	,
orr-bootleg-2020	379	17	LETI	LETI	NNP
orr-bootleg-2020	379	18	-	-	HYPH
orr-bootleg-2020	379	19	CEA	CEA	NNP
orr-bootleg-2020	379	20	,	,	,
orr-bootleg-2020	379	21	Intel	Intel	NNP
orr-bootleg-2020	379	22	,	,	,
orr-bootleg-2020	379	23	IBM	IBM	NNP
orr-bootleg-2020	379	24	,	,	,
orr-bootleg-2020	379	25	Microsoft	Microsoft	NNP
orr-bootleg-2020	379	26	,	,	,
orr-bootleg-2020	379	27	NEC	NEC	NNP
orr-bootleg-2020	379	28	,	,	,
orr-bootleg-2020	379	29	Toshiba	Toshiba	NNP
orr-bootleg-2020	379	30	,	,	,
orr-bootleg-2020	379	31	TSMC	TSMC	NNP
orr-bootleg-2020	379	32	,	,	,
orr-bootleg-2020	379	33	ARM	ARM	NNP
orr-bootleg-2020	379	34	,	,	,
orr-bootleg-2020	379	35	Hitachi	Hitachi	NNP
orr-bootleg-2020	379	36	,	,	,
orr-bootleg-2020	379	37	BASF	BASF	NNP
orr-bootleg-2020	379	38	,	,	,
orr-bootleg-2020	379	39	Accenture	Accenture	NNP
orr-bootleg-2020	379	40	,	,	,
orr-bootleg-2020	379	41	Ericsson	Ericsson	NNP
orr-bootleg-2020	379	42	,	,	,
orr-bootleg-2020	379	43	Qualcomm	Qualcomm	NNP
orr-bootleg-2020	379	44	,	,	,
orr-bootleg-2020	379	45	Analog	Analog	NNP
orr-bootleg-2020	379	46	Devices	Devices	NNPS
orr-bootleg-2020	379	47	,	,	,
orr-bootleg-2020	379	48	the	the	DT
orr-bootleg-2020	379	49	Okawa	Okawa	NNP
orr-bootleg-2020	379	50	Foundation	Foundation	NNP
orr-bootleg-2020	379	51	,	,	,
orr-bootleg-2020	379	52	American	American	NNP
orr-bootleg-2020	379	53	Family	Family	NNP
orr-bootleg-2020	379	54	Insurance	Insurance	NNP
orr-bootleg-2020	379	55	,	,	,
orr-bootleg-2020	379	56	Google	Google	NNP
orr-bootleg-2020	379	57	Cloud	Cloud	NNP
orr-bootleg-2020	379	58	,	,	,
orr-bootleg-2020	379	59	Swiss	Swiss	NNP
orr-bootleg-2020	379	60	Re	Re	NNP
orr-bootleg-2020	379	61	,	,	,
orr-bootleg-2020	379	62	the	the	DT
orr-bootleg-2020	379	63	HAI	HAI	NNP
orr-bootleg-2020	379	64	-	-	HYPH
orr-bootleg-2020	379	65	AWS	AWS	NNP
orr-bootleg-2020	379	66	Cloud	Cloud	NNP
orr-bootleg-2020	379	67	Credits	Credits	NNPS
orr-bootleg-2020	379	68	for	for	IN
orr-bootleg-2020	379	69	Research	research	NN
orr-bootleg-2020	379	70	program	program	NN
orr-bootleg-2020	379	71	,	,	,
orr-bootleg-2020	379	72	and	and	CC
orr-bootleg-2020	379	73	members	member	NNS
orr-bootleg-2020	379	74	of	of	IN
orr-bootleg-2020	379	75	the	the	DT
orr-bootleg-2020	379	76	Stanford	Stanford	NNP
orr-bootleg-2020	379	77	DAWN	DAWN	NNP
orr-bootleg-2020	379	78	project	project	NN
orr-bootleg-2020	379	79	:	:	:
orr-bootleg-2020	379	80	Teradata	Teradata	NNS
orr-bootleg-2020	379	81	,	,	,
orr-bootleg-2020	379	82	Facebook	Facebook	NNP
orr-bootleg-2020	379	83	,	,	,
orr-bootleg-2020	379	84	Google	Google	NNP
orr-bootleg-2020	379	85	,	,	,
orr-bootleg-2020	379	86	Ant	Ant	NNP
orr-bootleg-2020	379	87	Financial	Financial	NNP
orr-bootleg-2020	379	88	,	,	,
orr-bootleg-2020	379	89	NEC	NEC	NNP
orr-bootleg-2020	379	90	,	,	,
orr-bootleg-2020	379	91	VMWare	VMWare	NNP
orr-bootleg-2020	379	92	,	,	,
orr-bootleg-2020	379	93	and	and	CC
orr-bootleg-2020	379	94	Infosys	Infosys	NNP
orr-bootleg-2020	379	95	.	.	.
orr-bootleg-2020	380	1	The	the	DT
orr-bootleg-2020	380	2	U.S.	U.S.	NNP
orr-bootleg-2020	380	3	Government	Government	NNP
orr-bootleg-2020	380	4	is	be	VBZ
orr-bootleg-2020	380	5	authorized	authorize	VBN
orr-bootleg-2020	380	6	to	to	TO
orr-bootleg-2020	380	7	reproduce	reproduce	VB
orr-bootleg-2020	380	8	and	and	CC
orr-bootleg-2020	380	9	distribute	distribute	VB
orr-bootleg-2020	380	10	reprints	reprint	NNS
orr-bootleg-2020	380	11	for	for	IN
orr-bootleg-2020	380	12	Governmental	governmental	JJ
orr-bootleg-2020	380	13	purposes	purpose	NNS
orr-bootleg-2020	380	14	notwith-	notwith-	JJ
orr-bootleg-2020	380	15	standing	stand	VBG
orr-bootleg-2020	380	16	any	any	DT
orr-bootleg-2020	380	17	copyright	copyright	NN
orr-bootleg-2020	380	18	notation	notation	NN
orr-bootleg-2020	380	19	thereon	thereon	NN
orr-bootleg-2020	380	20	.	.	.
orr-bootleg-2020	381	1	Any	any	DT
orr-bootleg-2020	381	2	opinions	opinion	NNS
orr-bootleg-2020	381	3	,	,	,
orr-bootleg-2020	381	4	findings	finding	NNS
orr-bootleg-2020	381	5	,	,	,
orr-bootleg-2020	381	6	and	and	CC
orr-bootleg-2020	381	7	conclusions	conclusion	NNS
orr-bootleg-2020	381	8	or	or	CC
orr-bootleg-2020	381	9	recommendations	recommendation	NNS
orr-bootleg-2020	381	10	expressed	express	VBN
orr-bootleg-2020	381	11	in	in	IN
orr-bootleg-2020	381	12	this	this	DT
orr-bootleg-2020	381	13	material	material	NN
orr-bootleg-2020	381	14	are	be	VBP
orr-bootleg-2020	381	15	those	those	DT
orr-bootleg-2020	381	16	of	of	IN
orr-bootleg-2020	381	17	the	the	DT
orr-bootleg-2020	381	18	authors	author	NNS
orr-bootleg-2020	381	19	and	and	CC
orr-bootleg-2020	381	20	do	do	VBP
orr-bootleg-2020	381	21	not	not	RB
orr-bootleg-2020	381	22	necessarily	necessarily	RB
orr-bootleg-2020	381	23	reflect	reflect	VB
orr-bootleg-2020	381	24	the	the	DT
orr-bootleg-2020	381	25	views	view	NNS
orr-bootleg-2020	381	26	,	,	,
orr-bootleg-2020	381	27	policies	policy	NNS
orr-bootleg-2020	381	28	,	,	,
orr-bootleg-2020	381	29	or	or	CC
orr-bootleg-2020	381	30	endorsements	endorsement	NNS
orr-bootleg-2020	381	31	,	,	,
orr-bootleg-2020	381	32	either	either	CC
orr-bootleg-2020	381	33	expressed	express	VBN
orr-bootleg-2020	381	34	or	or	CC
orr-bootleg-2020	381	35	implied	imply	VBN
orr-bootleg-2020	381	36	,	,	,
orr-bootleg-2020	381	37	of	of	IN
orr-bootleg-2020	381	38	DARPA	DARPA	NNP
orr-bootleg-2020	381	39	,	,	,
orr-bootleg-2020	381	40	NIH	NIH	NNP
orr-bootleg-2020	381	41	,	,	,
orr-bootleg-2020	381	42	ONR	ONR	NNP
orr-bootleg-2020	381	43	,	,	,
orr-bootleg-2020	381	44	or	or	CC
orr-bootleg-2020	381	45	the	the	DT
orr-bootleg-2020	381	46	U.S.	U.S.	NNP
orr-bootleg-2020	381	47	Government	Government	NNP
orr-bootleg-2020	381	48	.	.	.
orr-bootleg-2020	382	1	References	reference	NNS
orr-bootleg-2020	382	2	[	[	-LRB-
orr-bootleg-2020	382	3	1	1	CD
orr-bootleg-2020	382	4	]	]	-RRB-
orr-bootleg-2020	382	5	John	John	NNP
orr-bootleg-2020	382	6	Aberdeen	Aberdeen	NNP
orr-bootleg-2020	382	7	,	,	,
orr-bootleg-2020	382	8	John	John	NNP
orr-bootleg-2020	382	9	D	D	NNP
orr-bootleg-2020	382	10	Burger	Burger	NNP
orr-bootleg-2020	382	11	,	,	,
orr-bootleg-2020	382	12	David	David	NNP
orr-bootleg-2020	382	13	Day	Day	NNP
orr-bootleg-2020	382	14	,	,	,
orr-bootleg-2020	382	15	Lynette	Lynette	NNP
orr-bootleg-2020	382	16	Hirschman	Hirschman	NNP
orr-bootleg-2020	382	17	,	,	,
orr-bootleg-2020	382	18	David	David	NNP
orr-bootleg-2020	382	19	D	D	NNP
orr-bootleg-2020	382	20	Palmer	Palmer	NNP
orr-bootleg-2020	382	21	,	,	,
orr-bootleg-2020	382	22	Patricia	Patricia	NNP
orr-bootleg-2020	382	23	Robinson	Robinson	NNP
orr-bootleg-2020	382	24	,	,	,
orr-bootleg-2020	382	25	and	and	CC
orr-bootleg-2020	382	26	Marc	Marc	NNP
orr-bootleg-2020	382	27	Vilain	Vilain	NNP
orr-bootleg-2020	382	28	.	.	.
orr-bootleg-2020	383	1	Mitre	Mitre	NNP
orr-bootleg-2020	383	2	:	:	:
orr-bootleg-2020	383	3	Description	description	NN
orr-bootleg-2020	383	4	of	of	IN
orr-bootleg-2020	383	5	the	the	DT
orr-bootleg-2020	383	6	alembic	alembic	JJ
orr-bootleg-2020	383	7	system	system	NN
orr-bootleg-2020	383	8	as	as	IN
orr-bootleg-2020	383	9	used	use	VBN
orr-bootleg-2020	383	10	in	in	IN
orr-bootleg-2020	383	11	met	meet	VBN
orr-bootleg-2020	383	12	.	.	.
orr-bootleg-2020	384	1	In	in	IN
orr-bootleg-2020	384	2	TIPSTER	tipster	NN
orr-bootleg-2020	384	3	TEXT	text	NN
orr-bootleg-2020	384	4	PROGRAM	PROGRAM	NNS
orr-bootleg-2020	384	5	PHASE	phase	VBP
orr-bootleg-2020	384	6	II	ii	CD
orr-bootleg-2020	384	7	:	:	:
orr-bootleg-2020	384	8	Proceedings	proceeding	NNS
orr-bootleg-2020	384	9	of	of	IN
orr-bootleg-2020	384	10	a	a	DT
orr-bootleg-2020	384	11	Workshop	Workshop	NNP
orr-bootleg-2020	384	12	held	hold	VBN
orr-bootleg-2020	384	13	at	at	IN
orr-bootleg-2020	384	14	Vienna	Vienna	NNP
orr-bootleg-2020	384	15	,	,	,
orr-bootleg-2020	384	16	Virginia	Virginia	NNP
orr-bootleg-2020	384	17	,	,	,
orr-bootleg-2020	384	18	May	May	NNP
orr-bootleg-2020	384	19	6	6	CD
orr-bootleg-2020	384	20	-	-	SYM
orr-bootleg-2020	384	21	8	8	CD
orr-bootleg-2020	384	22	,	,	,
orr-bootleg-2020	384	23	1996	1996	CD
orr-bootleg-2020	384	24	,	,	,
orr-bootleg-2020	384	25	pages	page	NNS
orr-bootleg-2020	384	26	461–462	461–462	CD
orr-bootleg-2020	384	27	,	,	,
orr-bootleg-2020	384	28	1996	1996	CD
orr-bootleg-2020	384	29	.	.	.
orr-bootleg-2020	385	1	[	[	-LRB-
orr-bootleg-2020	385	2	2	2	LS
orr-bootleg-2020	385	3	]	]	-RRB-
orr-bootleg-2020	385	4	Christoph	Christoph	NNP
orr-bootleg-2020	385	5	Alt	Alt	NNP
orr-bootleg-2020	385	6	,	,	,
orr-bootleg-2020	385	7	Aleksandra	Aleksandra	NNP
orr-bootleg-2020	385	8	Gabryszak	Gabryszak	NNP
orr-bootleg-2020	385	9	,	,	,
orr-bootleg-2020	385	10	and	and	CC
orr-bootleg-2020	385	11	Leonhard	Leonhard	NNP
orr-bootleg-2020	385	12	Hennig	Hennig	NNP
orr-bootleg-2020	385	13	.	.	.
orr-bootleg-2020	386	1	Tacred	Tacred	NNP
orr-bootleg-2020	386	2	revisited	revisit	VBD
orr-bootleg-2020	386	3	:	:	:
orr-bootleg-2020	386	4	A	a	DT
orr-bootleg-2020	386	5	thorough	thorough	JJ
orr-bootleg-2020	386	6	evaluation	evaluation	NN
orr-bootleg-2020	386	7	of	of	IN
orr-bootleg-2020	386	8	the	the	DT
orr-bootleg-2020	386	9	tacred	tacre	VBN
orr-bootleg-2020	386	10	relation	relation	NN
orr-bootleg-2020	386	11	extraction	extraction	NN
orr-bootleg-2020	386	12	task	task	NN
orr-bootleg-2020	386	13	.	.	.
orr-bootleg-2020	387	1	In	in	IN
orr-bootleg-2020	387	2	ACL	ACL	NNP
orr-bootleg-2020	387	3	,	,	,
orr-bootleg-2020	387	4	2020	2020	CD
orr-bootleg-2020	387	5	.	.	.
orr-bootleg-2020	388	1	[	[	-LRB-
orr-bootleg-2020	388	2	3	3	LS
orr-bootleg-2020	388	3	]	]	-RRB-
orr-bootleg-2020	388	4	Dzmitry	Dzmitry	NNP
orr-bootleg-2020	388	5	Bahdanau	Bahdanau	NNP
orr-bootleg-2020	388	6	,	,	,
orr-bootleg-2020	388	7	Kyunghyun	Kyunghyun	NNP
orr-bootleg-2020	388	8	Cho	Cho	NNP
orr-bootleg-2020	388	9	,	,	,
orr-bootleg-2020	388	10	and	and	CC
orr-bootleg-2020	388	11	Yoshua	Yoshua	NNP
orr-bootleg-2020	388	12	Bengio	Bengio	NNP
orr-bootleg-2020	388	13	.	.	.
orr-bootleg-2020	389	1	Neural	neural	JJ
orr-bootleg-2020	389	2	machine	machine	NN
orr-bootleg-2020	389	3	translation	translation	NN
orr-bootleg-2020	389	4	by	by	IN
orr-bootleg-2020	389	5	jointly	jointly	RB
orr-bootleg-2020	389	6	learning	learn	VBG
orr-bootleg-2020	389	7	to	to	TO
orr-bootleg-2020	389	8	align	align	VB
orr-bootleg-2020	389	9	and	and	CC
orr-bootleg-2020	389	10	translate	translate	VB
orr-bootleg-2020	389	11	.	.	.
orr-bootleg-2020	390	1	arXiv	arXiv	NNP
orr-bootleg-2020	390	2	preprint	preprint	NN
orr-bootleg-2020	390	3	arXiv:1409.0473	arXiv:1409.0473	NNP
orr-bootleg-2020	390	4	,	,	,
orr-bootleg-2020	390	5	2014	2014	CD
orr-bootleg-2020	390	6	.	.	.
orr-bootleg-2020	391	1	[	[	-LRB-
orr-bootleg-2020	391	2	4	4	CD
orr-bootleg-2020	391	3	]	]	-RRB-
orr-bootleg-2020	391	4	Michael	Michael	NNP
orr-bootleg-2020	391	5	S	S	NNP
orr-bootleg-2020	391	6	Bernstein	Bernstein	NNP
orr-bootleg-2020	391	7	,	,	,
orr-bootleg-2020	391	8	Jaime	Jaime	NNP
orr-bootleg-2020	391	9	Teevan	Teevan	NNP
orr-bootleg-2020	391	10	,	,	,
orr-bootleg-2020	391	11	Susan	Susan	NNP
orr-bootleg-2020	391	12	Dumais	Dumais	NNP
orr-bootleg-2020	391	13	,	,	,
orr-bootleg-2020	391	14	Daniel	Daniel	NNP
orr-bootleg-2020	391	15	Liebling	Liebling	NNP
orr-bootleg-2020	391	16	,	,	,
orr-bootleg-2020	391	17	and	and	CC
orr-bootleg-2020	391	18	Eric	Eric	NNP
orr-bootleg-2020	391	19	Horvitz	Horvitz	NNP
orr-bootleg-2020	391	20	.	.	.
orr-bootleg-2020	392	1	Direct	direct	JJ
orr-bootleg-2020	392	2	answers	answer	NNS
orr-bootleg-2020	392	3	for	for	IN
orr-bootleg-2020	392	4	search	search	NN
orr-bootleg-2020	392	5	queries	query	NNS
orr-bootleg-2020	392	6	in	in	IN
orr-bootleg-2020	392	7	the	the	DT
orr-bootleg-2020	392	8	long	long	JJ
orr-bootleg-2020	392	9	tail	tail	NN
orr-bootleg-2020	392	10	.	.	.
orr-bootleg-2020	393	1	In	in	IN
orr-bootleg-2020	393	2	SIGCHI	SIGCHI	NNP
orr-bootleg-2020	393	3	,	,	,
orr-bootleg-2020	393	4	2012	2012	CD
orr-bootleg-2020	393	5	.	.	.
orr-bootleg-2020	394	1	[	[	-LRB-
orr-bootleg-2020	394	2	5	5	CD
orr-bootleg-2020	394	3	]	]	-RRB-
orr-bootleg-2020	394	4	Terra	Terra	NNP
orr-bootleg-2020	394	5	Blevins	Blevins	NNPS
orr-bootleg-2020	394	6	and	and	CC
orr-bootleg-2020	394	7	Luke	Luke	NNP
orr-bootleg-2020	394	8	Zettlemoyer	Zettlemoyer	NNP
orr-bootleg-2020	394	9	.	.	.
orr-bootleg-2020	395	1	Moving	move	VBG
orr-bootleg-2020	395	2	down	down	IN
orr-bootleg-2020	395	3	the	the	DT
orr-bootleg-2020	395	4	long	long	JJ
orr-bootleg-2020	395	5	tail	tail	NN
orr-bootleg-2020	395	6	of	of	IN
orr-bootleg-2020	395	7	word	word	NN
orr-bootleg-2020	395	8	sense	sense	NN
orr-bootleg-2020	395	9	disambiguation	disambiguation	NN
orr-bootleg-2020	395	10	with	with	IN
orr-bootleg-2020	395	11	gloss	gloss	NN
orr-bootleg-2020	395	12	-	-	HYPH
orr-bootleg-2020	395	13	informed	inform	VBN
orr-bootleg-2020	395	14	biencoders	biencoder	NNS
orr-bootleg-2020	395	15	.	.	.
orr-bootleg-2020	396	1	arXiv	arXiv	NNP
orr-bootleg-2020	396	2	preprint	preprint	NNP
orr-bootleg-2020	396	3	arXiv:2005.02590	arXiv:2005.02590	NNP
orr-bootleg-2020	396	4	,	,	,
orr-bootleg-2020	396	5	2020	2020	CD
orr-bootleg-2020	396	6	.	.	.
orr-bootleg-2020	397	1	15	15	CD
orr-bootleg-2020	397	2	[	[	-LRB-
orr-bootleg-2020	397	3	6	6	CD
orr-bootleg-2020	397	4	]	]	-RRB-
orr-bootleg-2020	397	5	Samuel	Samuel	NNP
orr-bootleg-2020	397	6	Broscheit	Broscheit	NNP
orr-bootleg-2020	397	7	.	.	.
orr-bootleg-2020	398	1	Investigating	investigate	VBG
orr-bootleg-2020	398	2	entity	entity	NN
orr-bootleg-2020	398	3	knowledge	knowledge	NN
orr-bootleg-2020	398	4	in	in	IN
orr-bootleg-2020	398	5	bert	bert	NNP
orr-bootleg-2020	398	6	with	with	IN
orr-bootleg-2020	398	7	simple	simple	JJ
orr-bootleg-2020	398	8	neural	neural	JJ
orr-bootleg-2020	398	9	end	end	NN
orr-bootleg-2020	398	10	-	-	HYPH
orr-bootleg-2020	398	11	to	to	IN
orr-bootleg-2020	398	12	-	-	HYPH
orr-bootleg-2020	398	13	end	end	NN
orr-bootleg-2020	398	14	entity	entity	NN
orr-bootleg-2020	398	15	linking	linking	NN
orr-bootleg-2020	398	16	.	.	.
orr-bootleg-2020	399	1	arXiv	arXiv	NNP
orr-bootleg-2020	399	2	preprint	preprint	NN
orr-bootleg-2020	399	3	arXiv:2003.05473	arXiv:2003.05473	NNP
orr-bootleg-2020	399	4	,	,	,
orr-bootleg-2020	399	5	2020	2020	CD
orr-bootleg-2020	399	6	.	.	.
orr-bootleg-2020	400	1	[	[	-LRB-
orr-bootleg-2020	400	2	7	7	CD
orr-bootleg-2020	400	3	]	]	-RRB-
orr-bootleg-2020	400	4	Tom	Tom	NNP
orr-bootleg-2020	400	5	B	B	NNP
orr-bootleg-2020	400	6	Brown	Brown	NNP
orr-bootleg-2020	400	7	,	,	,
orr-bootleg-2020	400	8	Benjamin	Benjamin	NNP
orr-bootleg-2020	400	9	Mann	Mann	NNP
orr-bootleg-2020	400	10	,	,	,
orr-bootleg-2020	400	11	Nick	Nick	NNP
orr-bootleg-2020	400	12	Ryder	Ryder	NNP
orr-bootleg-2020	400	13	,	,	,
orr-bootleg-2020	400	14	Melanie	Melanie	NNP
orr-bootleg-2020	400	15	Subbiah	Subbiah	NNP
orr-bootleg-2020	400	16	,	,	,
orr-bootleg-2020	400	17	Jared	Jared	NNP
orr-bootleg-2020	400	18	Kaplan	Kaplan	NNP
orr-bootleg-2020	400	19	,	,	,
orr-bootleg-2020	400	20	Prafulla	Prafulla	NNP
orr-bootleg-2020	400	21	Dhariwal	Dhariwal	NNP
orr-bootleg-2020	400	22	,	,	,
orr-bootleg-2020	400	23	Arvind	Arvind	NNP
orr-bootleg-2020	400	24	Neelakantan	Neelakantan	NNP
orr-bootleg-2020	400	25	,	,	,
orr-bootleg-2020	400	26	Pranav	Pranav	NNP
orr-bootleg-2020	400	27	Shyam	Shyam	NNP
orr-bootleg-2020	400	28	,	,	,
orr-bootleg-2020	400	29	Girish	Girish	NNP
orr-bootleg-2020	400	30	Sastry	Sastry	NNP
orr-bootleg-2020	400	31	,	,	,
orr-bootleg-2020	400	32	Amanda	Amanda	NNP
orr-bootleg-2020	400	33	Askell	Askell	NNP
orr-bootleg-2020	400	34	,	,	,
orr-bootleg-2020	400	35	et	et	NNP
orr-bootleg-2020	400	36	al	al	NNP
orr-bootleg-2020	400	37	.	.	.
orr-bootleg-2020	401	1	Language	language	NN
orr-bootleg-2020	401	2	models	model	NNS
orr-bootleg-2020	401	3	are	be	VBP
orr-bootleg-2020	401	4	few	few	JJ
orr-bootleg-2020	401	5	-	-	HYPH
orr-bootleg-2020	401	6	shot	shoot	VBN
orr-bootleg-2020	401	7	learners	learner	NNS
orr-bootleg-2020	401	8	.	.	.
orr-bootleg-2020	402	1	arXiv	arXiv	NNP
orr-bootleg-2020	402	2	preprint	preprint	NN
orr-bootleg-2020	402	3	arXiv:2005.14165	arXiv:2005.14165	NNP
orr-bootleg-2020	402	4	,	,	,
orr-bootleg-2020	402	5	2020	2020	CD
orr-bootleg-2020	402	6	.	.	.
orr-bootleg-2020	403	1	[	[	-LRB-
orr-bootleg-2020	403	2	8	8	CD
orr-bootleg-2020	403	3	]	]	-RRB-
orr-bootleg-2020	403	4	Alberto	Alberto	NNP
orr-bootleg-2020	403	5	Cetoli	Cetoli	NNP
orr-bootleg-2020	403	6	,	,	,
orr-bootleg-2020	403	7	Mohammad	Mohammad	NNP
orr-bootleg-2020	403	8	Akbari	Akbari	NNP
orr-bootleg-2020	403	9	,	,	,
orr-bootleg-2020	403	10	Stefano	Stefano	NNP
orr-bootleg-2020	403	11	Bragaglia	Bragaglia	NNP
orr-bootleg-2020	403	12	,	,	,
orr-bootleg-2020	403	13	Andrew	Andrew	NNP
orr-bootleg-2020	403	14	D	D	NNP
orr-bootleg-2020	403	15	O’Harney	O’Harney	NNP
orr-bootleg-2020	403	16	,	,	,
orr-bootleg-2020	403	17	and	and	CC
orr-bootleg-2020	403	18	Marc	Marc	NNP
orr-bootleg-2020	403	19	Sloan	Sloan	NNP
orr-bootleg-2020	403	20	.	.	.
orr-bootleg-2020	404	1	Named	name	VBN
orr-bootleg-2020	404	2	entity	entity	NN
orr-bootleg-2020	404	3	disambiguation	disambiguation	NN
orr-bootleg-2020	404	4	using	use	VBG
orr-bootleg-2020	404	5	deep	deep	JJ
orr-bootleg-2020	404	6	learning	learning	NN
orr-bootleg-2020	404	7	on	on	IN
orr-bootleg-2020	404	8	graphs	graphs	NN
orr-bootleg-2020	404	9	.	.	.
orr-bootleg-2020	405	1	arXiv	arXiv	NNP
orr-bootleg-2020	405	2	preprint	preprint	NNP
orr-bootleg-2020	405	3	arXiv:1810.09164	arXiv:1810.09164	NNP
orr-bootleg-2020	405	4	,	,	,
orr-bootleg-2020	405	5	2018	2018	CD
orr-bootleg-2020	405	6	.	.	.
orr-bootleg-2020	406	1	[	[	-LRB-
orr-bootleg-2020	406	2	9	9	CD
orr-bootleg-2020	406	3	]	]	-RRB-
orr-bootleg-2020	406	4	Shuang	Shuang	NNP
orr-bootleg-2020	406	5	Chen	Chen	NNP
orr-bootleg-2020	406	6	,	,	,
orr-bootleg-2020	406	7	Jinpeng	Jinpeng	NNP
orr-bootleg-2020	406	8	Wang	Wang	NNP
orr-bootleg-2020	406	9	,	,	,
orr-bootleg-2020	406	10	Feng	Feng	NNP
orr-bootleg-2020	406	11	Jiang	Jiang	NNP
orr-bootleg-2020	406	12	,	,	,
orr-bootleg-2020	406	13	and	and	CC
orr-bootleg-2020	406	14	Chin	Chin	NNP
orr-bootleg-2020	406	15	-	-	HYPH
orr-bootleg-2020	406	16	Yew	Yew	NNP
orr-bootleg-2020	406	17	Lin	Lin	NNP
orr-bootleg-2020	406	18	.	.	.
orr-bootleg-2020	407	1	Improving	improve	VBG
orr-bootleg-2020	407	2	entity	entity	NN
orr-bootleg-2020	407	3	linking	link	VBG
orr-bootleg-2020	407	4	by	by	IN
orr-bootleg-2020	407	5	modeling	model	VBG
orr-bootleg-2020	407	6	latent	latent	NN
orr-bootleg-2020	407	7	entity	entity	NN
orr-bootleg-2020	407	8	type	type	NN
orr-bootleg-2020	407	9	information	information	NN
orr-bootleg-2020	407	10	.	.	.
orr-bootleg-2020	408	1	arXiv	arXiv	NNP
orr-bootleg-2020	408	2	preprint	preprint	NN
orr-bootleg-2020	408	3	arXiv:2001.01447	arXiv:2001.01447	NNP
orr-bootleg-2020	408	4	,	,	,
orr-bootleg-2020	408	5	2020	2020	CD
orr-bootleg-2020	408	6	.	.	.
orr-bootleg-2020	409	1	[	[	-LRB-
orr-bootleg-2020	409	2	10	10	CD
orr-bootleg-2020	409	3	]	]	-RRB-
orr-bootleg-2020	409	4	Yeounoh	Yeounoh	NNP
orr-bootleg-2020	409	5	Chung	Chung	NNP
orr-bootleg-2020	409	6	,	,	,
orr-bootleg-2020	409	7	Peter	Peter	NNP
orr-bootleg-2020	409	8	J	J	NNP
orr-bootleg-2020	409	9	Haas	Haas	NNP
orr-bootleg-2020	409	10	,	,	,
orr-bootleg-2020	409	11	Eli	Eli	NNP
orr-bootleg-2020	409	12	Upfal	Upfal	NNP
orr-bootleg-2020	409	13	,	,	,
orr-bootleg-2020	409	14	and	and	CC
orr-bootleg-2020	409	15	Tim	Tim	NNP
orr-bootleg-2020	409	16	Kraska	Kraska	NNP
orr-bootleg-2020	409	17	.	.	.
orr-bootleg-2020	410	1	Unknown	unknown	JJ
orr-bootleg-2020	410	2	examples	example	NNS
orr-bootleg-2020	410	3	&	&	CC
orr-bootleg-2020	410	4	machine	machine	NN
orr-bootleg-2020	410	5	learning	learning	NN
orr-bootleg-2020	410	6	model	model	NN
orr-bootleg-2020	410	7	generalization	generalization	NN
orr-bootleg-2020	410	8	.	.	.
orr-bootleg-2020	411	1	arXiv	arXiv	NNP
orr-bootleg-2020	411	2	preprint	preprint	NN
orr-bootleg-2020	411	3	arXiv:1808.08294	arXiv:1808.08294	NNP
orr-bootleg-2020	411	4	,	,	,
orr-bootleg-2020	411	5	2018	2018	CD
orr-bootleg-2020	411	6	.	.	.
orr-bootleg-2020	412	1	[	[	-LRB-
orr-bootleg-2020	412	2	11	11	CD
orr-bootleg-2020	412	3	]	]	-RRB-
orr-bootleg-2020	412	4	Yeounoh	Yeounoh	NNP
orr-bootleg-2020	412	5	Chung	Chung	NNP
orr-bootleg-2020	412	6	,	,	,
orr-bootleg-2020	412	7	Neoklis	Neoklis	NNP
orr-bootleg-2020	412	8	Polyzotis	Polyzotis	NNP
orr-bootleg-2020	412	9	,	,	,
orr-bootleg-2020	412	10	Kihyun	Kihyun	NNP
orr-bootleg-2020	412	11	Tae	Tae	NNP
orr-bootleg-2020	412	12	,	,	,
orr-bootleg-2020	412	13	Steven	Steven	NNP
orr-bootleg-2020	412	14	Euijong	Euijong	NNP
orr-bootleg-2020	412	15	Whang	Whang	NNP
orr-bootleg-2020	412	16	,	,	,
orr-bootleg-2020	412	17	et	et	NNP
orr-bootleg-2020	412	18	al	al	NNP
orr-bootleg-2020	412	19	.	.	.
orr-bootleg-2020	413	1	Automated	automate	VBN
orr-bootleg-2020	413	2	data	datum	NNS
orr-bootleg-2020	413	3	slicing	slice	VBG
orr-bootleg-2020	413	4	for	for	IN
orr-bootleg-2020	413	5	model	model	NN
orr-bootleg-2020	413	6	validation	validation	NN
orr-bootleg-2020	413	7	:	:	:
orr-bootleg-2020	413	8	A	a	DT
orr-bootleg-2020	413	9	big	big	JJ
orr-bootleg-2020	413	10	data	datum	NNS
orr-bootleg-2020	413	11	-	-	HYPH
orr-bootleg-2020	413	12	ai	ai	NN
orr-bootleg-2020	413	13	integration	integration	NN
orr-bootleg-2020	413	14	approach	approach	NN
orr-bootleg-2020	413	15	.	.	.
orr-bootleg-2020	414	1	TKDE	TKDE	NNP
orr-bootleg-2020	414	2	,	,	,
orr-bootleg-2020	414	3	2019	2019	CD
orr-bootleg-2020	414	4	.	.	.
orr-bootleg-2020	415	1	[	[	-LRB-
orr-bootleg-2020	415	2	12	12	CD
orr-bootleg-2020	415	3	]	]	-RRB-
orr-bootleg-2020	415	4	Silviu	Silviu	NNP
orr-bootleg-2020	415	5	Cucerzan	Cucerzan	NNP
orr-bootleg-2020	415	6	.	.	.
orr-bootleg-2020	416	1	Large	large	JJ
orr-bootleg-2020	416	2	-	-	HYPH
orr-bootleg-2020	416	3	scale	scale	NN
orr-bootleg-2020	416	4	named	name	VBN
orr-bootleg-2020	416	5	entity	entity	NN
orr-bootleg-2020	416	6	disambiguation	disambiguation	NN
orr-bootleg-2020	416	7	based	base	VBN
orr-bootleg-2020	416	8	on	on	IN
orr-bootleg-2020	416	9	wikipedia	wikipedia	NNP
orr-bootleg-2020	416	10	data	datum	NNS
orr-bootleg-2020	416	11	.	.	.
orr-bootleg-2020	417	1	In	in	IN
orr-bootleg-2020	417	2	Proceedings	proceeding	NNS
orr-bootleg-2020	417	3	of	of	IN
orr-bootleg-2020	417	4	the	the	DT
orr-bootleg-2020	417	5	2007	2007	CD
orr-bootleg-2020	417	6	Joint	Joint	NNP
orr-bootleg-2020	417	7	Conference	Conference	NNP
orr-bootleg-2020	417	8	on	on	IN
orr-bootleg-2020	417	9	Empirical	Empirical	NNP
orr-bootleg-2020	417	10	Methods	Methods	NNPS
orr-bootleg-2020	417	11	in	in	IN
orr-bootleg-2020	417	12	Natural	Natural	NNP
orr-bootleg-2020	417	13	Language	Language	NNP
orr-bootleg-2020	417	14	Processing	Processing	NNP
orr-bootleg-2020	417	15	and	and	CC
orr-bootleg-2020	417	16	Computational	Computational	NNP
orr-bootleg-2020	417	17	Natural	Natural	NNP
orr-bootleg-2020	417	18	Language	Language	NNP
orr-bootleg-2020	417	19	Learning	Learning	NNP
orr-bootleg-2020	417	20	(	(	-LRB-
orr-bootleg-2020	417	21	EMNLP	EMNLP	NNP
orr-bootleg-2020	417	22	-	-	HYPH
orr-bootleg-2020	417	23	CoNLL	CoNLL	NNP
orr-bootleg-2020	417	24	)	)	-RRB-
orr-bootleg-2020	417	25	,	,	,
orr-bootleg-2020	417	26	pages	page	VBZ
orr-bootleg-2020	417	27	708–716	708–716	CD
orr-bootleg-2020	417	28	,	,	,
orr-bootleg-2020	417	29	2007	2007	CD
orr-bootleg-2020	417	30	.	.	.
orr-bootleg-2020	418	1	[	[	-LRB-
orr-bootleg-2020	418	2	13	13	CD
orr-bootleg-2020	418	3	]	]	-RRB-
orr-bootleg-2020	418	4	Nicola	Nicola	NNP
orr-bootleg-2020	418	5	De	De	NNP
orr-bootleg-2020	418	6	Cao	Cao	NNP
orr-bootleg-2020	418	7	,	,	,
orr-bootleg-2020	418	8	Gautier	Gautier	NNP
orr-bootleg-2020	418	9	Izacard	Izacard	NNP
orr-bootleg-2020	418	10	,	,	,
orr-bootleg-2020	418	11	Sebastian	Sebastian	NNP
orr-bootleg-2020	418	12	Riedel	Riedel	NNP
orr-bootleg-2020	418	13	,	,	,
orr-bootleg-2020	418	14	and	and	CC
orr-bootleg-2020	418	15	Fabio	Fabio	NNP
orr-bootleg-2020	418	16	Petroni	Petroni	NNP
orr-bootleg-2020	418	17	.	.	.
orr-bootleg-2020	419	1	Autoregressive	autoregressive	JJ
orr-bootleg-2020	419	2	entity	entity	NN
orr-bootleg-2020	419	3	retrieval	retrieval	NN
orr-bootleg-2020	419	4	.	.	.
orr-bootleg-2020	420	1	arXiv	arXiv	NNP
orr-bootleg-2020	420	2	preprint	preprint	NN
orr-bootleg-2020	420	3	arXiv:2010.00904	arXiv:2010.00904	NNP
orr-bootleg-2020	420	4	,	,	,
orr-bootleg-2020	420	5	2020	2020	CD
orr-bootleg-2020	420	6	.	.	.
orr-bootleg-2020	421	1	[	[	-LRB-
orr-bootleg-2020	421	2	14	14	CD
orr-bootleg-2020	421	3	]	]	-RRB-
orr-bootleg-2020	421	4	Mark	Mark	NNP
orr-bootleg-2020	421	5	Dredze	Dredze	NNP
orr-bootleg-2020	421	6	,	,	,
orr-bootleg-2020	421	7	Paul	Paul	NNP
orr-bootleg-2020	421	8	McNamee	McNamee	NNP
orr-bootleg-2020	421	9	,	,	,
orr-bootleg-2020	421	10	Delip	Delip	NNP
orr-bootleg-2020	421	11	Rao	Rao	NNP
orr-bootleg-2020	421	12	,	,	,
orr-bootleg-2020	421	13	Adam	Adam	NNP
orr-bootleg-2020	421	14	Gerber	Gerber	NNP
orr-bootleg-2020	421	15	,	,	,
orr-bootleg-2020	421	16	and	and	CC
orr-bootleg-2020	421	17	Tim	Tim	NNP
orr-bootleg-2020	421	18	Finin	Finin	NNP
orr-bootleg-2020	421	19	.	.	.
orr-bootleg-2020	422	1	Entity	entity	NN
orr-bootleg-2020	422	2	disambiguation	disambiguation	NN
orr-bootleg-2020	422	3	for	for	IN
orr-bootleg-2020	422	4	knowledge	knowledge	NN
orr-bootleg-2020	422	5	base	base	NN
orr-bootleg-2020	422	6	population	population	NN
orr-bootleg-2020	422	7	.	.	.
orr-bootleg-2020	423	1	In	in	IN
orr-bootleg-2020	423	2	Proceedings	proceeding	NNS
orr-bootleg-2020	423	3	of	of	IN
orr-bootleg-2020	423	4	the	the	DT
orr-bootleg-2020	423	5	23rd	23rd	JJ
orr-bootleg-2020	423	6	International	International	NNP
orr-bootleg-2020	423	7	Conference	Conference	NNP
orr-bootleg-2020	423	8	on	on	IN
orr-bootleg-2020	423	9	Computational	Computational	NNP
orr-bootleg-2020	423	10	Linguistics	Linguistics	NNP
orr-bootleg-2020	423	11	(	(	-LRB-
orr-bootleg-2020	423	12	Coling	Coling	NNP
orr-bootleg-2020	423	13	2010	2010	CD
orr-bootleg-2020	423	14	)	)	-RRB-
orr-bootleg-2020	423	15	,	,	,
orr-bootleg-2020	423	16	pages	page	VBZ
orr-bootleg-2020	423	17	277–285	277–285	CD
orr-bootleg-2020	423	18	,	,	,
orr-bootleg-2020	423	19	2010	2010	CD
orr-bootleg-2020	423	20	.	.	.
orr-bootleg-2020	424	1	[	[	-LRB-
orr-bootleg-2020	424	2	15	15	CD
orr-bootleg-2020	424	3	]	]	-RRB-
orr-bootleg-2020	424	4	Allyson	Allyson	NNP
orr-bootleg-2020	424	5	Ettinger	Ettinger	NNP
orr-bootleg-2020	424	6	,	,	,
orr-bootleg-2020	424	7	Sudha	Sudha	NNP
orr-bootleg-2020	424	8	Rao	Rao	NNP
orr-bootleg-2020	424	9	,	,	,
orr-bootleg-2020	424	10	Hal	Hal	NNP
orr-bootleg-2020	424	11	Daumé	Daumé	NNP
orr-bootleg-2020	424	12	III	III	NNP
orr-bootleg-2020	424	13	,	,	,
orr-bootleg-2020	424	14	and	and	CC
orr-bootleg-2020	424	15	Emily	Emily	NNP
orr-bootleg-2020	424	16	M	M	NNP
orr-bootleg-2020	424	17	Bender	Bender	NNP
orr-bootleg-2020	424	18	.	.	.
orr-bootleg-2020	425	1	Towards	towards	IN
orr-bootleg-2020	425	2	linguistically	linguistically	RB
orr-bootleg-2020	425	3	generalizable	generalizable	JJ
orr-bootleg-2020	425	4	nlp	nlp	NN
orr-bootleg-2020	425	5	systems	system	NNS
orr-bootleg-2020	425	6	:	:	:
orr-bootleg-2020	425	7	A	a	DT
orr-bootleg-2020	425	8	workshop	workshop	NN
orr-bootleg-2020	425	9	and	and	CC
orr-bootleg-2020	425	10	shared	share	VBD
orr-bootleg-2020	425	11	task	task	NN
orr-bootleg-2020	425	12	.	.	.
orr-bootleg-2020	426	1	arXiv	arXiv	NNP
orr-bootleg-2020	426	2	preprint	preprint	NNP
orr-bootleg-2020	426	3	arXiv:1711.01505	arXiv:1711.01505	NNP
orr-bootleg-2020	426	4	,	,	,
orr-bootleg-2020	426	5	2017	2017	CD
orr-bootleg-2020	426	6	.	.	.
orr-bootleg-2020	427	1	[	[	-LRB-
orr-bootleg-2020	427	2	16	16	CD
orr-bootleg-2020	427	3	]	]	-RRB-
orr-bootleg-2020	427	4	Thibault	Thibault	NNP
orr-bootleg-2020	427	5	Févry	Févry	NNP
orr-bootleg-2020	427	6	,	,	,
orr-bootleg-2020	427	7	Nicholas	Nicholas	NNP
orr-bootleg-2020	427	8	FitzGerald	FitzGerald	NNP
orr-bootleg-2020	427	9	,	,	,
orr-bootleg-2020	427	10	Livio	Livio	NNP
orr-bootleg-2020	427	11	Baldini	Baldini	NNP
orr-bootleg-2020	427	12	Soares	Soares	NNP
orr-bootleg-2020	427	13	,	,	,
orr-bootleg-2020	427	14	and	and	CC
orr-bootleg-2020	427	15	Tom	Tom	NNP
orr-bootleg-2020	427	16	Kwiatkowski	Kwiatkowski	NNP
orr-bootleg-2020	427	17	.	.	.
orr-bootleg-2020	428	1	Empirical	empirical	JJ
orr-bootleg-2020	428	2	evaluation	evaluation	NN
orr-bootleg-2020	428	3	of	of	IN
orr-bootleg-2020	428	4	pretraining	pretraine	VBG
orr-bootleg-2020	428	5	strategies	strategy	NNS
orr-bootleg-2020	428	6	for	for	IN
orr-bootleg-2020	428	7	supervised	supervised	JJ
orr-bootleg-2020	428	8	entity	entity	NN
orr-bootleg-2020	428	9	linking	linking	NN
orr-bootleg-2020	428	10	.	.	.
orr-bootleg-2020	429	1	In	in	IN
orr-bootleg-2020	429	2	AKBC	AKBC	NNP
orr-bootleg-2020	429	3	,	,	,
orr-bootleg-2020	429	4	2020	2020	CD
orr-bootleg-2020	429	5	.	.	.
orr-bootleg-2020	430	1	[	[	-LRB-
orr-bootleg-2020	430	2	17	17	CD
orr-bootleg-2020	430	3	]	]	-RRB-
orr-bootleg-2020	430	4	Octavian	Octavian	NNP
orr-bootleg-2020	430	5	-	-	HYPH
orr-bootleg-2020	430	6	Eugen	Eugen	NNP
orr-bootleg-2020	430	7	Ganea	Ganea	NNP
orr-bootleg-2020	430	8	and	and	CC
orr-bootleg-2020	430	9	Thomas	Thomas	NNP
orr-bootleg-2020	430	10	Hofmann	Hofmann	NNP
orr-bootleg-2020	430	11	.	.	.
orr-bootleg-2020	431	1	Deep	deep	JJ
orr-bootleg-2020	431	2	joint	joint	JJ
orr-bootleg-2020	431	3	entity	entity	NN
orr-bootleg-2020	431	4	disambiguation	disambiguation	NN
orr-bootleg-2020	431	5	with	with	IN
orr-bootleg-2020	431	6	local	local	JJ
orr-bootleg-2020	431	7	neural	neural	JJ
orr-bootleg-2020	431	8	attention	attention	NN
orr-bootleg-2020	431	9	.	.	.
orr-bootleg-2020	432	1	arXiv	arXiv	NNP
orr-bootleg-2020	432	2	preprint	preprint	NN
orr-bootleg-2020	432	3	arXiv:1704.04920	arXiv:1704.04920	NNP
orr-bootleg-2020	432	4	,	,	,
orr-bootleg-2020	432	5	2017	2017	CD
orr-bootleg-2020	432	6	.	.	.
orr-bootleg-2020	433	1	[	[	-LRB-
orr-bootleg-2020	433	2	18	18	CD
orr-bootleg-2020	433	3	]	]	-RRB-
orr-bootleg-2020	433	4	Daniel	Daniel	NNP
orr-bootleg-2020	433	5	Gerber	Gerber	NNP
orr-bootleg-2020	433	6	,	,	,
orr-bootleg-2020	433	7	Sebastian	Sebastian	NNP
orr-bootleg-2020	433	8	Hellmann	Hellmann	NNP
orr-bootleg-2020	433	9	,	,	,
orr-bootleg-2020	433	10	Lorenz	Lorenz	NNP
orr-bootleg-2020	433	11	Bühmann	Bühmann	NNP
orr-bootleg-2020	433	12	,	,	,
orr-bootleg-2020	433	13	Tommaso	Tommaso	NNP
orr-bootleg-2020	433	14	Soru	Soru	NNP
orr-bootleg-2020	433	15	,	,	,
orr-bootleg-2020	433	16	Ricardo	Ricardo	NNP
orr-bootleg-2020	433	17	Usbeck	Usbeck	NNP
orr-bootleg-2020	433	18	,	,	,
orr-bootleg-2020	433	19	and	and	CC
orr-bootleg-2020	433	20	Axel-	Axel-	NNP
orr-bootleg-2020	433	21	Cyrille	Cyrille	NNP
orr-bootleg-2020	433	22	Ngonga	Ngonga	NNP
orr-bootleg-2020	433	23	Ngomo	Ngomo	NNP
orr-bootleg-2020	433	24	.	.	.
orr-bootleg-2020	434	1	Real	real	JJ
orr-bootleg-2020	434	2	-	-	HYPH
orr-bootleg-2020	434	3	time	time	NN
orr-bootleg-2020	434	4	rdf	rdf	NN
orr-bootleg-2020	434	5	extraction	extraction	NN
orr-bootleg-2020	434	6	from	from	IN
orr-bootleg-2020	434	7	unstructured	unstructured	JJ
orr-bootleg-2020	434	8	data	datum	NNS
orr-bootleg-2020	434	9	streams	stream	NNS
orr-bootleg-2020	434	10	.	.	.
orr-bootleg-2020	435	1	In	in	IN
orr-bootleg-2020	435	2	ISWC	ISWC	NNP
orr-bootleg-2020	435	3	,	,	,
orr-bootleg-2020	435	4	2013	2013	CD
orr-bootleg-2020	435	5	.	.	.
orr-bootleg-2020	436	1	[	[	-LRB-
orr-bootleg-2020	436	2	19	19	CD
orr-bootleg-2020	436	3	]	]	-RRB-
orr-bootleg-2020	436	4	Abbas	Abbas	NNP
orr-bootleg-2020	436	5	Ghaddar	Ghaddar	NNP
orr-bootleg-2020	436	6	and	and	CC
orr-bootleg-2020	436	7	Philippe	Philippe	NNP
orr-bootleg-2020	436	8	Langlais	Langlais	NNP
orr-bootleg-2020	436	9	.	.	.
orr-bootleg-2020	437	1	Winer	winer	NN
orr-bootleg-2020	437	2	:	:	:
orr-bootleg-2020	437	3	A	a	DT
orr-bootleg-2020	437	4	wikipedia	wikipedia	NN
orr-bootleg-2020	437	5	annotated	annotate	VBD
orr-bootleg-2020	437	6	corpus	corpus	NNP
orr-bootleg-2020	437	7	for	for	IN
orr-bootleg-2020	437	8	named	name	VBN
orr-bootleg-2020	437	9	entity	entity	NN
orr-bootleg-2020	437	10	recognition	recognition	NN
orr-bootleg-2020	437	11	.	.	.
orr-bootleg-2020	438	1	In	in	IN
orr-bootleg-2020	438	2	Proceedings	Proceedings	NNP
orr-bootleg-2020	438	3	of	of	IN
orr-bootleg-2020	438	4	the	the	DT
orr-bootleg-2020	438	5	Eighth	Eighth	NNP
orr-bootleg-2020	438	6	International	International	NNP
orr-bootleg-2020	438	7	Joint	Joint	NNP
orr-bootleg-2020	438	8	Conference	Conference	NNP
orr-bootleg-2020	438	9	on	on	IN
orr-bootleg-2020	438	10	Natural	Natural	NNP
orr-bootleg-2020	438	11	Language	Language	NNP
orr-bootleg-2020	438	12	Processing	Processing	NNP
orr-bootleg-2020	438	13	(	(	-LRB-
orr-bootleg-2020	438	14	Volume	volume	NN
orr-bootleg-2020	438	15	1	1	CD
orr-bootleg-2020	438	16	:	:	:
orr-bootleg-2020	438	17	Long	Long	NNP
orr-bootleg-2020	438	18	Papers	Papers	NNPS
orr-bootleg-2020	438	19	)	)	-RRB-
orr-bootleg-2020	438	20	,	,	,
orr-bootleg-2020	438	21	pages	page	VBZ
orr-bootleg-2020	438	22	413–422	413–422	CD
orr-bootleg-2020	438	23	,	,	,
orr-bootleg-2020	438	24	2017	2017	CD
orr-bootleg-2020	438	25	.	.	.
orr-bootleg-2020	439	1	[	[	-LRB-
orr-bootleg-2020	439	2	20	20	CD
orr-bootleg-2020	439	3	]	]	-RRB-
orr-bootleg-2020	439	4	Ben	Ben	NNP
orr-bootleg-2020	439	5	Gomes	Gomes	NNP
orr-bootleg-2020	439	6	.	.	.
orr-bootleg-2020	440	1	Our	-PRON-	PRP$
orr-bootleg-2020	440	2	latest	late	JJS
orr-bootleg-2020	440	3	quality	quality	NN
orr-bootleg-2020	440	4	improvements	improvement	NNS
orr-bootleg-2020	440	5	for	for	IN
orr-bootleg-2020	440	6	search	search	NN
orr-bootleg-2020	440	7	.	.	.
orr-bootleg-2020	441	1	https://blog.google/products/search/	https://blog.google/products/search/	VB
orr-bootleg-2020	441	2	our	-PRON-	PRP$
orr-bootleg-2020	441	3	-	-	HYPH
orr-bootleg-2020	441	4	latest	late	JJS
orr-bootleg-2020	441	5	-	-	HYPH
orr-bootleg-2020	441	6	quality	quality	NN
orr-bootleg-2020	441	7	-	-	HYPH
orr-bootleg-2020	441	8	improvements	improvement	NNS
orr-bootleg-2020	441	9	-	-	HYPH
orr-bootleg-2020	441	10	search/	search/	NNS
orr-bootleg-2020	441	11	,	,	,
orr-bootleg-2020	441	12	2017	2017	CD
orr-bootleg-2020	441	13	.	.	.
orr-bootleg-2020	442	1	[	[	-LRB-
orr-bootleg-2020	442	2	21	21	CD
orr-bootleg-2020	442	3	]	]	-RRB-
orr-bootleg-2020	442	4	Nitish	Nitish	NNP
orr-bootleg-2020	442	5	Gupta	Gupta	NNP
orr-bootleg-2020	442	6	,	,	,
orr-bootleg-2020	442	7	Sameer	Sameer	NNP
orr-bootleg-2020	442	8	Singh	Singh	NNP
orr-bootleg-2020	442	9	,	,	,
orr-bootleg-2020	442	10	and	and	CC
orr-bootleg-2020	442	11	Dan	Dan	NNP
orr-bootleg-2020	442	12	Roth	Roth	NNP
orr-bootleg-2020	442	13	.	.	.
orr-bootleg-2020	443	1	Entity	entity	NN
orr-bootleg-2020	443	2	linking	link	VBG
orr-bootleg-2020	443	3	via	via	IN
orr-bootleg-2020	443	4	joint	joint	JJ
orr-bootleg-2020	443	5	encoding	encoding	NN
orr-bootleg-2020	443	6	of	of	IN
orr-bootleg-2020	443	7	types	type	NNS
orr-bootleg-2020	443	8	,	,	,
orr-bootleg-2020	443	9	descriptions	description	NNS
orr-bootleg-2020	443	10	,	,	,
orr-bootleg-2020	443	11	and	and	CC
orr-bootleg-2020	443	12	context	context	NN
orr-bootleg-2020	443	13	.	.	.
orr-bootleg-2020	444	1	In	in	IN
orr-bootleg-2020	444	2	Proceedings	proceeding	NNS
orr-bootleg-2020	444	3	of	of	IN
orr-bootleg-2020	444	4	the	the	DT
orr-bootleg-2020	444	5	2017	2017	CD
orr-bootleg-2020	444	6	Conference	Conference	NNP
orr-bootleg-2020	444	7	on	on	IN
orr-bootleg-2020	444	8	Empirical	Empirical	NNP
orr-bootleg-2020	444	9	Methods	Methods	NNPS
orr-bootleg-2020	444	10	in	in	IN
orr-bootleg-2020	444	11	Natural	Natural	NNP
orr-bootleg-2020	444	12	Language	Language	NNP
orr-bootleg-2020	444	13	Processing	Processing	NNP
orr-bootleg-2020	444	14	,	,	,
orr-bootleg-2020	444	15	pages	page	NNS
orr-bootleg-2020	444	16	2681–2690	2681–2690	CD
orr-bootleg-2020	444	17	,	,	,
orr-bootleg-2020	444	18	2017	2017	CD
orr-bootleg-2020	444	19	.	.	.
orr-bootleg-2020	445	1	[	[	-LRB-
orr-bootleg-2020	445	2	22	22	CD
orr-bootleg-2020	445	3	]	]	-RRB-
orr-bootleg-2020	445	4	Johannes	Johannes	NNP
orr-bootleg-2020	445	5	Hoffart	Hoffart	NNP
orr-bootleg-2020	445	6	,	,	,
orr-bootleg-2020	445	7	Mohamed	Mohamed	NNP
orr-bootleg-2020	445	8	Amir	Amir	NNP
orr-bootleg-2020	445	9	Yosef	Yosef	NNP
orr-bootleg-2020	445	10	,	,	,
orr-bootleg-2020	445	11	Ilaria	Ilaria	NNP
orr-bootleg-2020	445	12	Bordino	Bordino	NNP
orr-bootleg-2020	445	13	,	,	,
orr-bootleg-2020	445	14	Hagen	Hagen	NNP
orr-bootleg-2020	445	15	Fürstenau	Fürstenau	NNP
orr-bootleg-2020	445	16	,	,	,
orr-bootleg-2020	445	17	Manfred	Manfred	NNP
orr-bootleg-2020	445	18	Pinkal	Pinkal	NNP
orr-bootleg-2020	445	19	,	,	,
orr-bootleg-2020	445	20	Marc	Marc	NNP
orr-bootleg-2020	445	21	Spaniol	Spaniol	NNP
orr-bootleg-2020	445	22	,	,	,
orr-bootleg-2020	445	23	Bilyana	Bilyana	NNP
orr-bootleg-2020	445	24	Taneva	Taneva	NNP
orr-bootleg-2020	445	25	,	,	,
orr-bootleg-2020	445	26	Stefan	Stefan	NNP
orr-bootleg-2020	445	27	Thater	Thater	NNP
orr-bootleg-2020	445	28	,	,	,
orr-bootleg-2020	445	29	and	and	CC
orr-bootleg-2020	445	30	Gerhard	Gerhard	NNP
orr-bootleg-2020	445	31	Weikum	Weikum	NNP
orr-bootleg-2020	445	32	.	.	.
orr-bootleg-2020	446	1	Robust	robust	JJ
orr-bootleg-2020	446	2	disambiguation	disambiguation	NN
orr-bootleg-2020	446	3	of	of	IN
orr-bootleg-2020	446	4	named	name	VBN
orr-bootleg-2020	446	5	entities	entity	NNS
orr-bootleg-2020	446	6	in	in	IN
orr-bootleg-2020	446	7	text	text	NN
orr-bootleg-2020	446	8	.	.	.
orr-bootleg-2020	447	1	In	in	IN
orr-bootleg-2020	447	2	Proceedings	Proceedings	NNP
orr-bootleg-2020	447	3	of	of	IN
orr-bootleg-2020	447	4	the	the	DT
orr-bootleg-2020	447	5	Conference	Conference	NNP
orr-bootleg-2020	447	6	on	on	IN
orr-bootleg-2020	447	7	Empirical	Empirical	NNP
orr-bootleg-2020	447	8	Methods	Methods	NNPS
orr-bootleg-2020	447	9	in	in	IN
orr-bootleg-2020	447	10	Natural	Natural	NNP
orr-bootleg-2020	447	11	Language	Language	NNP
orr-bootleg-2020	447	12	Processing	Processing	NNP
orr-bootleg-2020	447	13	,	,	,
orr-bootleg-2020	447	14	pages	page	VBZ
orr-bootleg-2020	447	15	782–792	782–792	CD
orr-bootleg-2020	447	16	.	.	.
orr-bootleg-2020	448	1	Association	Association	NNP
orr-bootleg-2020	448	2	for	for	IN
orr-bootleg-2020	448	3	Computational	Computational	NNP
orr-bootleg-2020	448	4	Linguistics	Linguistics	NNP
orr-bootleg-2020	448	5	,	,	,
orr-bootleg-2020	448	6	2011	2011	CD
orr-bootleg-2020	448	7	.	.	.
orr-bootleg-2020	449	1	[	[	-LRB-
orr-bootleg-2020	449	2	23	23	CD
orr-bootleg-2020	449	3	]	]	-RRB-
orr-bootleg-2020	449	4	Johannes	Johannes	NNP
orr-bootleg-2020	449	5	Hoffart	Hoffart	NNP
orr-bootleg-2020	449	6	,	,	,
orr-bootleg-2020	449	7	Stephan	Stephan	NNP
orr-bootleg-2020	449	8	Seufert	Seufert	NNP
orr-bootleg-2020	449	9	,	,	,
orr-bootleg-2020	449	10	Dat	Dat	NNP
orr-bootleg-2020	449	11	Ba	Ba	NNP
orr-bootleg-2020	449	12	Nguyen	Nguyen	NNP
orr-bootleg-2020	449	13	,	,	,
orr-bootleg-2020	449	14	Martin	Martin	NNP
orr-bootleg-2020	449	15	Theobald	Theobald	NNP
orr-bootleg-2020	449	16	,	,	,
orr-bootleg-2020	449	17	and	and	CC
orr-bootleg-2020	449	18	Gerhard	Gerhard	NNP
orr-bootleg-2020	449	19	Weikum	Weikum	NNP
orr-bootleg-2020	449	20	.	.	.
orr-bootleg-2020	450	1	Kore	Kore	NNP
orr-bootleg-2020	450	2	:	:	:
orr-bootleg-2020	450	3	keyphrase	keyphrase	NNP
orr-bootleg-2020	450	4	overlap	overlap	NNP
orr-bootleg-2020	450	5	relatedness	relatedness	NNP
orr-bootleg-2020	450	6	for	for	IN
orr-bootleg-2020	450	7	entity	entity	NN
orr-bootleg-2020	450	8	disambiguation	disambiguation	NN
orr-bootleg-2020	450	9	.	.	.
orr-bootleg-2020	451	1	In	in	IN
orr-bootleg-2020	451	2	CIKM	CIKM	NNP
orr-bootleg-2020	451	3	,	,	,
orr-bootleg-2020	451	4	2012	2012	CD
orr-bootleg-2020	451	5	.	.	.
orr-bootleg-2020	452	1	16	16	CD
orr-bootleg-2020	452	2	https://blog.google/products/search/our-latest-quality-improvements-search/	https://blog.google/products/search/our-latest-quality-improvements-search/	JJ
orr-bootleg-2020	452	3	https://blog.google/products/search/our-latest-quality-improvements-search/	https://blog.google/products/search/our-latest-quality-improvements-search/	JJ
orr-bootleg-2020	452	4	[	[	-LRB-
orr-bootleg-2020	452	5	24	24	CD
orr-bootleg-2020	452	6	]	]	-RRB-
orr-bootleg-2020	452	7	Shengze	Shengze	NNP
orr-bootleg-2020	452	8	Hu	Hu	NNP
orr-bootleg-2020	452	9	,	,	,
orr-bootleg-2020	452	10	Zhen	Zhen	NNP
orr-bootleg-2020	452	11	Tan	Tan	NNP
orr-bootleg-2020	452	12	,	,	,
orr-bootleg-2020	452	13	Weixin	Weixin	NNP
orr-bootleg-2020	452	14	Zeng	Zeng	NNP
orr-bootleg-2020	452	15	,	,	,
orr-bootleg-2020	452	16	Bin	Bin	NNP
orr-bootleg-2020	452	17	Ge	Ge	NNP
orr-bootleg-2020	452	18	,	,	,
orr-bootleg-2020	452	19	and	and	CC
orr-bootleg-2020	452	20	Weidong	Weidong	NNP
orr-bootleg-2020	452	21	Xiao	Xiao	NNP
orr-bootleg-2020	452	22	.	.	.
orr-bootleg-2020	453	1	Entity	entity	NN
orr-bootleg-2020	453	2	linking	link	VBG
orr-bootleg-2020	453	3	via	via	IN
orr-bootleg-2020	453	4	symmetrical	symmetrical	JJ
orr-bootleg-2020	453	5	attention	attention	NN
orr-bootleg-2020	453	6	-	-	HYPH
orr-bootleg-2020	453	7	based	base	VBN
orr-bootleg-2020	453	8	neural	neural	JJ
orr-bootleg-2020	453	9	network	network	NN
orr-bootleg-2020	453	10	and	and	CC
orr-bootleg-2020	453	11	entity	entity	NN
orr-bootleg-2020	453	12	structural	structural	JJ
orr-bootleg-2020	453	13	features	feature	NNS
orr-bootleg-2020	453	14	.	.	.
orr-bootleg-2020	454	1	Symmetry	Symmetry	NNP
orr-bootleg-2020	454	2	,	,	,
orr-bootleg-2020	454	3	2019	2019	CD
orr-bootleg-2020	454	4	.	.	.
orr-bootleg-2020	455	1	[	[	-LRB-
orr-bootleg-2020	455	2	25	25	CD
orr-bootleg-2020	455	3	]	]	-RRB-
orr-bootleg-2020	455	4	Filip	Filip	NNP
orr-bootleg-2020	455	5	Ilievski	Ilievski	NNP
orr-bootleg-2020	455	6	,	,	,
orr-bootleg-2020	455	7	Piek	Piek	NNP
orr-bootleg-2020	455	8	Vossen	Vossen	NNP
orr-bootleg-2020	455	9	,	,	,
orr-bootleg-2020	455	10	and	and	CC
orr-bootleg-2020	455	11	Stefan	Stefan	NNP
orr-bootleg-2020	455	12	Schlobach	Schlobach	NNP
orr-bootleg-2020	455	13	.	.	.
orr-bootleg-2020	456	1	Systematic	systematic	JJ
orr-bootleg-2020	456	2	study	study	NN
orr-bootleg-2020	456	3	of	of	IN
orr-bootleg-2020	456	4	long	long	JJ
orr-bootleg-2020	456	5	tail	tail	NN
orr-bootleg-2020	456	6	phenomena	phenomenon	NNS
orr-bootleg-2020	456	7	in	in	IN
orr-bootleg-2020	456	8	entity	entity	NN
orr-bootleg-2020	456	9	linking	linking	NN
orr-bootleg-2020	456	10	.	.	.
orr-bootleg-2020	457	1	In	in	IN
orr-bootleg-2020	457	2	Proceedings	proceeding	NNS
orr-bootleg-2020	457	3	of	of	IN
orr-bootleg-2020	457	4	the	the	DT
orr-bootleg-2020	457	5	27th	27th	JJ
orr-bootleg-2020	457	6	International	International	NNP
orr-bootleg-2020	457	7	Conference	Conference	NNP
orr-bootleg-2020	457	8	on	on	IN
orr-bootleg-2020	457	9	Computational	Computational	NNP
orr-bootleg-2020	457	10	Linguistics	Linguistics	NNP
orr-bootleg-2020	457	11	,	,	,
orr-bootleg-2020	457	12	pages	page	VBZ
orr-bootleg-2020	457	13	664–674	664–674	CD
orr-bootleg-2020	457	14	,	,	,
orr-bootleg-2020	457	15	2018	2018	CD
orr-bootleg-2020	457	16	.	.	.
orr-bootleg-2020	458	1	[	[	-LRB-
orr-bootleg-2020	458	2	26	26	CD
orr-bootleg-2020	458	3	]	]	-RRB-
orr-bootleg-2020	458	4	Yuzhe	Yuzhe	NNP
orr-bootleg-2020	458	5	Jin	Jin	NNP
orr-bootleg-2020	458	6	,	,	,
orr-bootleg-2020	458	7	Emre	Emre	NNP
orr-bootleg-2020	458	8	Kiciman	Kiciman	NNP
orr-bootleg-2020	458	9	,	,	,
orr-bootleg-2020	458	10	Kuansan	Kuansan	NNP
orr-bootleg-2020	458	11	Wang	Wang	NNP
orr-bootleg-2020	458	12	,	,	,
orr-bootleg-2020	458	13	and	and	CC
orr-bootleg-2020	458	14	Ricky	Ricky	NNP
orr-bootleg-2020	458	15	Loynd	Loynd	NNP
orr-bootleg-2020	458	16	.	.	.
orr-bootleg-2020	459	1	Entity	entity	NN
orr-bootleg-2020	459	2	linking	link	VBG
orr-bootleg-2020	459	3	at	at	IN
orr-bootleg-2020	459	4	the	the	DT
orr-bootleg-2020	459	5	tail	tail	NN
orr-bootleg-2020	459	6	:	:	:
orr-bootleg-2020	459	7	Sparse	sparse	NN
orr-bootleg-2020	459	8	signals	signal	NNS
orr-bootleg-2020	459	9	,	,	,
orr-bootleg-2020	459	10	unknown	unknown	JJ
orr-bootleg-2020	459	11	entities	entity	NNS
orr-bootleg-2020	459	12	and	and	CC
orr-bootleg-2020	459	13	phrase	phrase	NN
orr-bootleg-2020	459	14	models	model	NNS
orr-bootleg-2020	459	15	.	.	.
orr-bootleg-2020	460	1	In	in	IN
orr-bootleg-2020	460	2	WSDM	WSDM	NNP
orr-bootleg-2020	460	3	’	'	''
orr-bootleg-2020	460	4	14	14	CD
orr-bootleg-2020	460	5	Proceedings	proceeding	NNS
orr-bootleg-2020	460	6	of	of	IN
orr-bootleg-2020	460	7	the	the	DT
orr-bootleg-2020	460	8	7th	7th	JJ
orr-bootleg-2020	460	9	ACM	ACM	NNP
orr-bootleg-2020	460	10	international	international	JJ
orr-bootleg-2020	460	11	conference	conference	NN
orr-bootleg-2020	460	12	on	on	IN
orr-bootleg-2020	460	13	Web	web	NN
orr-bootleg-2020	460	14	search	search	NN
orr-bootleg-2020	460	15	and	and	CC
orr-bootleg-2020	460	16	data	datum	NNS
orr-bootleg-2020	460	17	mining	mining	NN
orr-bootleg-2020	460	18	,	,	,
orr-bootleg-2020	460	19	pages	page	VBZ
orr-bootleg-2020	460	20	453–462	453–462	CD
orr-bootleg-2020	460	21	.	.	.
orr-bootleg-2020	461	1	ACM	ACM	NNP
orr-bootleg-2020	461	2	,	,	,
orr-bootleg-2020	461	3	February	February	NNP
orr-bootleg-2020	461	4	2014	2014	CD
orr-bootleg-2020	461	5	.	.	.
orr-bootleg-2020	462	1	[	[	-LRB-
orr-bootleg-2020	462	2	27	27	CD
orr-bootleg-2020	462	3	]	]	-RRB-
orr-bootleg-2020	462	4	Mandar	Mandar	NNP
orr-bootleg-2020	462	5	Joshi	Joshi	NNP
orr-bootleg-2020	462	6	,	,	,
orr-bootleg-2020	462	7	Danqi	Danqi	NNP
orr-bootleg-2020	462	8	Chen	Chen	NNP
orr-bootleg-2020	462	9	,	,	,
orr-bootleg-2020	462	10	Yinhan	Yinhan	NNP
orr-bootleg-2020	462	11	Liu	Liu	NNP
orr-bootleg-2020	462	12	,	,	,
orr-bootleg-2020	462	13	Daniel	Daniel	NNP
orr-bootleg-2020	462	14	S.	S.	NNP
orr-bootleg-2020	462	15	Weld	Weld	NNP
orr-bootleg-2020	462	16	,	,	,
orr-bootleg-2020	462	17	Luke	Luke	NNP
orr-bootleg-2020	462	18	Zettlemoyer	Zettlemoyer	NNP
orr-bootleg-2020	462	19	,	,	,
orr-bootleg-2020	462	20	and	and	CC
orr-bootleg-2020	462	21	Omer	Omer	NNP
orr-bootleg-2020	462	22	Levy	Levy	NNP
orr-bootleg-2020	462	23	.	.	.
orr-bootleg-2020	463	1	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	463	2	:	:	:
orr-bootleg-2020	463	3	Improving	improve	VBG
orr-bootleg-2020	463	4	pre	pre	JJ
orr-bootleg-2020	463	5	-	-	JJ
orr-bootleg-2020	463	6	training	training	NN
orr-bootleg-2020	463	7	by	by	IN
orr-bootleg-2020	463	8	representing	represent	VBG
orr-bootleg-2020	463	9	and	and	CC
orr-bootleg-2020	463	10	predicting	predict	VBG
orr-bootleg-2020	463	11	spans	span	NNS
orr-bootleg-2020	463	12	.	.	.
orr-bootleg-2020	464	1	arXiv	arXiv	NNP
orr-bootleg-2020	464	2	preprint	preprint	NNP
orr-bootleg-2020	464	3	arXiv:1907.10529	arXiv:1907.10529	NNP
orr-bootleg-2020	464	4	,	,	,
orr-bootleg-2020	464	5	2019	2019	CD
orr-bootleg-2020	464	6	.	.	.
orr-bootleg-2020	465	1	[	[	-LRB-
orr-bootleg-2020	465	2	28	28	CD
orr-bootleg-2020	465	3	]	]	-RRB-
orr-bootleg-2020	465	4	Diederik	Diederik	NNP
orr-bootleg-2020	465	5	P.	P.	NNP
orr-bootleg-2020	465	6	Kingma	Kingma	NNP
orr-bootleg-2020	465	7	and	and	CC
orr-bootleg-2020	465	8	Jimmy	Jimmy	NNP
orr-bootleg-2020	465	9	Ba	Ba	NNP
orr-bootleg-2020	465	10	.	.	.
orr-bootleg-2020	466	1	Adam	Adam	NNP
orr-bootleg-2020	466	2	:	:	:
orr-bootleg-2020	466	3	A	a	DT
orr-bootleg-2020	466	4	method	method	NN
orr-bootleg-2020	466	5	for	for	IN
orr-bootleg-2020	466	6	stochastic	stochastic	JJ
orr-bootleg-2020	466	7	optimization	optimization	NN
orr-bootleg-2020	466	8	.	.	.
orr-bootleg-2020	467	1	In	in	IN
orr-bootleg-2020	467	2	Yoshua	Yoshua	NNP
orr-bootleg-2020	467	3	Bengio	Bengio	NNP
orr-bootleg-2020	467	4	and	and	CC
orr-bootleg-2020	467	5	Yann	Yann	NNP
orr-bootleg-2020	467	6	LeCun	LeCun	NNP
orr-bootleg-2020	467	7	,	,	,
orr-bootleg-2020	467	8	editors	editor	NNS
orr-bootleg-2020	467	9	,	,	,
orr-bootleg-2020	467	10	3rd	3rd	NNP
orr-bootleg-2020	467	11	International	International	NNP
orr-bootleg-2020	467	12	Conference	Conference	NNP
orr-bootleg-2020	467	13	on	on	IN
orr-bootleg-2020	467	14	Learning	Learning	NNP
orr-bootleg-2020	467	15	Representations	Representations	NNPS
orr-bootleg-2020	467	16	,	,	,
orr-bootleg-2020	467	17	ICLR	ICLR	NNP
orr-bootleg-2020	467	18	2015	2015	CD
orr-bootleg-2020	467	19	,	,	,
orr-bootleg-2020	467	20	San	San	NNP
orr-bootleg-2020	467	21	Diego	Diego	NNP
orr-bootleg-2020	467	22	,	,	,
orr-bootleg-2020	467	23	CA	CA	NNP
orr-bootleg-2020	467	24	,	,	,
orr-bootleg-2020	467	25	USA	USA	NNP
orr-bootleg-2020	467	26	,	,	,
orr-bootleg-2020	467	27	May	May	NNP
orr-bootleg-2020	467	28	7	7	CD
orr-bootleg-2020	467	29	-	-	SYM
orr-bootleg-2020	467	30	9	9	CD
orr-bootleg-2020	467	31	,	,	,
orr-bootleg-2020	467	32	2015	2015	CD
orr-bootleg-2020	467	33	,	,	,
orr-bootleg-2020	467	34	Conference	Conference	NNP
orr-bootleg-2020	467	35	Track	Track	NNP
orr-bootleg-2020	467	36	Proceedings	Proceedings	NNP
orr-bootleg-2020	467	37	,	,	,
orr-bootleg-2020	467	38	2015	2015	CD
orr-bootleg-2020	467	39	.	.	.
orr-bootleg-2020	468	1	[	[	-LRB-
orr-bootleg-2020	468	2	29	29	CD
orr-bootleg-2020	468	3	]	]	-RRB-
orr-bootleg-2020	468	4	Tom	Tom	NNP
orr-bootleg-2020	468	5	Kwiatkowski	Kwiatkowski	NNP
orr-bootleg-2020	468	6	,	,	,
orr-bootleg-2020	468	7	Jennimaria	Jennimaria	NNP
orr-bootleg-2020	468	8	Palomaki	Palomaki	NNP
orr-bootleg-2020	468	9	,	,	,
orr-bootleg-2020	468	10	Olivia	Olivia	NNP
orr-bootleg-2020	468	11	Redfield	Redfield	NNP
orr-bootleg-2020	468	12	,	,	,
orr-bootleg-2020	468	13	Michael	Michael	NNP
orr-bootleg-2020	468	14	Collins	Collins	NNP
orr-bootleg-2020	468	15	,	,	,
orr-bootleg-2020	468	16	Ankur	Ankur	NNP
orr-bootleg-2020	468	17	Parikh	Parikh	NNP
orr-bootleg-2020	468	18	,	,	,
orr-bootleg-2020	468	19	Chris	Chris	NNP
orr-bootleg-2020	468	20	Alberti	Alberti	NNP
orr-bootleg-2020	468	21	,	,	,
orr-bootleg-2020	468	22	Danielle	Danielle	NNP
orr-bootleg-2020	468	23	Epstein	Epstein	NNP
orr-bootleg-2020	468	24	,	,	,
orr-bootleg-2020	468	25	Illia	Illia	NNP
orr-bootleg-2020	468	26	Polosukhin	Polosukhin	NNP
orr-bootleg-2020	468	27	,	,	,
orr-bootleg-2020	468	28	Jacob	Jacob	NNP
orr-bootleg-2020	468	29	Devlin	Devlin	NNP
orr-bootleg-2020	468	30	,	,	,
orr-bootleg-2020	468	31	Kenton	Kenton	NNP
orr-bootleg-2020	468	32	Lee	Lee	NNP
orr-bootleg-2020	468	33	,	,	,
orr-bootleg-2020	468	34	et	et	NNP
orr-bootleg-2020	468	35	al	al	NNP
orr-bootleg-2020	468	36	.	.	.
orr-bootleg-2020	469	1	Natural	natural	JJ
orr-bootleg-2020	469	2	questions	question	NNS
orr-bootleg-2020	469	3	:	:	:
orr-bootleg-2020	469	4	a	a	DT
orr-bootleg-2020	469	5	benchmark	benchmark	NN
orr-bootleg-2020	469	6	for	for	IN
orr-bootleg-2020	469	7	question	question	NN
orr-bootleg-2020	469	8	answering	answer	VBG
orr-bootleg-2020	469	9	research	research	NN
orr-bootleg-2020	469	10	.	.	.
orr-bootleg-2020	470	1	Transactions	transaction	NNS
orr-bootleg-2020	470	2	of	of	IN
orr-bootleg-2020	470	3	the	the	DT
orr-bootleg-2020	470	4	Association	Association	NNP
orr-bootleg-2020	470	5	for	for	IN
orr-bootleg-2020	470	6	Computational	Computational	NNP
orr-bootleg-2020	470	7	Linguistics	Linguistics	NNP
orr-bootleg-2020	470	8	,	,	,
orr-bootleg-2020	470	9	7:453–466	7:453–466	NNP
orr-bootleg-2020	470	10	,	,	,
orr-bootleg-2020	470	11	2019	2019	CD
orr-bootleg-2020	470	12	.	.	.
orr-bootleg-2020	471	1	[	[	-LRB-
orr-bootleg-2020	471	2	30	30	CD
orr-bootleg-2020	471	3	]	]	-RRB-
orr-bootleg-2020	471	4	Phong	Phong	NNP
orr-bootleg-2020	471	5	Le	Le	NNP
orr-bootleg-2020	471	6	and	and	CC
orr-bootleg-2020	471	7	Ivan	Ivan	NNP
orr-bootleg-2020	471	8	Titov	Titov	NNP
orr-bootleg-2020	471	9	.	.	.
orr-bootleg-2020	472	1	Improving	improve	VBG
orr-bootleg-2020	472	2	entity	entity	NN
orr-bootleg-2020	472	3	linking	link	VBG
orr-bootleg-2020	472	4	by	by	IN
orr-bootleg-2020	472	5	modeling	model	VBG
orr-bootleg-2020	472	6	latent	latent	NN
orr-bootleg-2020	472	7	relations	relation	NNS
orr-bootleg-2020	472	8	between	between	IN
orr-bootleg-2020	472	9	mentions	mention	NNS
orr-bootleg-2020	472	10	.	.	.
orr-bootleg-2020	473	1	arXiv	arXiv	NNP
orr-bootleg-2020	473	2	preprint	preprint	NN
orr-bootleg-2020	473	3	arXiv:1804.10637	arXiv:1804.10637	NNP
orr-bootleg-2020	473	4	,	,	,
orr-bootleg-2020	473	5	2018	2018	CD
orr-bootleg-2020	473	6	.	.	.
orr-bootleg-2020	474	1	[	[	-LRB-
orr-bootleg-2020	474	2	31	31	CD
orr-bootleg-2020	474	3	]	]	-RRB-
orr-bootleg-2020	474	4	Xiao	Xiao	NNP
orr-bootleg-2020	474	5	Ling	Ling	NNP
orr-bootleg-2020	474	6	,	,	,
orr-bootleg-2020	474	7	Sameer	Sameer	NNP
orr-bootleg-2020	474	8	Singh	Singh	NNP
orr-bootleg-2020	474	9	,	,	,
orr-bootleg-2020	474	10	and	and	CC
orr-bootleg-2020	474	11	Daniel	Daniel	NNP
orr-bootleg-2020	474	12	S	S	NNP
orr-bootleg-2020	474	13	Weld	Weld	NNP
orr-bootleg-2020	474	14	.	.	.
orr-bootleg-2020	475	1	Design	design	NN
orr-bootleg-2020	475	2	challenges	challenge	NNS
orr-bootleg-2020	475	3	for	for	IN
orr-bootleg-2020	475	4	entity	entity	NN
orr-bootleg-2020	475	5	linking	linking	NN
orr-bootleg-2020	475	6	.	.	.
orr-bootleg-2020	476	1	Transactions	transaction	NNS
orr-bootleg-2020	476	2	of	of	IN
orr-bootleg-2020	476	3	the	the	DT
orr-bootleg-2020	476	4	Association	Association	NNP
orr-bootleg-2020	476	5	for	for	IN
orr-bootleg-2020	476	6	Computational	Computational	NNP
orr-bootleg-2020	476	7	Linguistics	Linguistics	NNP
orr-bootleg-2020	476	8	,	,	,
orr-bootleg-2020	476	9	3:315–328	3:315–328	CD
orr-bootleg-2020	476	10	,	,	,
orr-bootleg-2020	476	11	2015	2015	CD
orr-bootleg-2020	476	12	.	.	.
orr-bootleg-2020	477	1	[	[	-LRB-
orr-bootleg-2020	477	2	32	32	CD
orr-bootleg-2020	477	3	]	]	-RRB-
orr-bootleg-2020	477	4	Sidharth	Sidharth	NNP
orr-bootleg-2020	477	5	Mudgal	Mudgal	NNP
orr-bootleg-2020	477	6	,	,	,
orr-bootleg-2020	477	7	Han	Han	NNP
orr-bootleg-2020	477	8	Li	Li	NNP
orr-bootleg-2020	477	9	,	,	,
orr-bootleg-2020	477	10	Theodoros	Theodoros	NNP
orr-bootleg-2020	477	11	Rekatsinas	Rekatsinas	NNP
orr-bootleg-2020	477	12	,	,	,
orr-bootleg-2020	477	13	AnHai	AnHai	NNP
orr-bootleg-2020	477	14	Doan	Doan	NNP
orr-bootleg-2020	477	15	,	,	,
orr-bootleg-2020	477	16	Youngchoon	Youngchoon	NNP
orr-bootleg-2020	477	17	Park	Park	NNP
orr-bootleg-2020	477	18	,	,	,
orr-bootleg-2020	477	19	Ganesh	Ganesh	NNP
orr-bootleg-2020	477	20	Krishnan	Krishnan	NNP
orr-bootleg-2020	477	21	,	,	,
orr-bootleg-2020	477	22	Rohit	Rohit	NNP
orr-bootleg-2020	477	23	Deep	Deep	NNP
orr-bootleg-2020	477	24	,	,	,
orr-bootleg-2020	477	25	Esteban	Esteban	NNP
orr-bootleg-2020	477	26	Arcaute	Arcaute	NNP
orr-bootleg-2020	477	27	,	,	,
orr-bootleg-2020	477	28	and	and	CC
orr-bootleg-2020	477	29	Vijay	Vijay	NNP
orr-bootleg-2020	477	30	Raghavendra	Raghavendra	NNP
orr-bootleg-2020	477	31	.	.	.
orr-bootleg-2020	478	1	Deep	deep	JJ
orr-bootleg-2020	478	2	learning	learning	NN
orr-bootleg-2020	478	3	for	for	IN
orr-bootleg-2020	478	4	entity	entity	NN
orr-bootleg-2020	478	5	matching	matching	NN
orr-bootleg-2020	478	6	:	:	:
orr-bootleg-2020	478	7	A	a	DT
orr-bootleg-2020	478	8	design	design	NN
orr-bootleg-2020	478	9	space	space	NN
orr-bootleg-2020	478	10	exploration	exploration	NN
orr-bootleg-2020	478	11	.	.	.
orr-bootleg-2020	479	1	In	in	IN
orr-bootleg-2020	479	2	SIGMOD	SIGMOD	NNP
orr-bootleg-2020	479	3	,	,	,
orr-bootleg-2020	479	4	2018	2018	CD
orr-bootleg-2020	479	5	.	.	.
orr-bootleg-2020	480	1	[	[	-LRB-
orr-bootleg-2020	480	2	33	33	CD
orr-bootleg-2020	480	3	]	]	-RRB-
orr-bootleg-2020	480	4	Isaiah	Isaiah	NNP
orr-bootleg-2020	480	5	Onando	Onando	NNP
orr-bootleg-2020	480	6	Mulang	Mulang	NNP
orr-bootleg-2020	480	7	,	,	,
orr-bootleg-2020	480	8	Kuldeep	Kuldeep	NNP
orr-bootleg-2020	480	9	Singh	Singh	NNP
orr-bootleg-2020	480	10	,	,	,
orr-bootleg-2020	480	11	Chaitali	Chaitali	NNP
orr-bootleg-2020	480	12	Prabhu	Prabhu	NNP
orr-bootleg-2020	480	13	,	,	,
orr-bootleg-2020	480	14	Abhishek	Abhishek	NNP
orr-bootleg-2020	480	15	Nadgeri	Nadgeri	NNP
orr-bootleg-2020	480	16	,	,	,
orr-bootleg-2020	480	17	Johannes	Johannes	NNP
orr-bootleg-2020	480	18	Hoffart	Hoffart	NNP
orr-bootleg-2020	480	19	,	,	,
orr-bootleg-2020	480	20	and	and	CC
orr-bootleg-2020	480	21	Jens	Jens	NNP
orr-bootleg-2020	480	22	Lehmann	Lehmann	NNP
orr-bootleg-2020	480	23	.	.	.
orr-bootleg-2020	481	1	Evaluating	evaluate	VBG
orr-bootleg-2020	481	2	the	the	DT
orr-bootleg-2020	481	3	impact	impact	NN
orr-bootleg-2020	481	4	of	of	IN
orr-bootleg-2020	481	5	knowledge	knowledge	NN
orr-bootleg-2020	481	6	graph	graph	NNP
orr-bootleg-2020	481	7	contexton	contexton	NNP
orr-bootleg-2020	481	8	entity	entity	NNP
orr-bootleg-2020	481	9	disambiguation	disambiguation	NNP
orr-bootleg-2020	481	10	models	model	NNS
orr-bootleg-2020	481	11	.	.	.
orr-bootleg-2020	482	1	arXiv	arXiv	NNP
orr-bootleg-2020	482	2	preprint	preprint	NN
orr-bootleg-2020	482	3	arXiv:2008.05190	arXiv:2008.05190	NNP
orr-bootleg-2020	482	4	,	,	,
orr-bootleg-2020	482	5	2020	2020	CD
orr-bootleg-2020	482	6	.	.	.
orr-bootleg-2020	483	1	[	[	-LRB-
orr-bootleg-2020	483	2	34	34	CD
orr-bootleg-2020	483	3	]	]	-RRB-
orr-bootleg-2020	483	4	Joel	Joel	NNP
orr-bootleg-2020	483	5	Nothman	Nothman	NNP
orr-bootleg-2020	483	6	,	,	,
orr-bootleg-2020	483	7	James	James	NNP
orr-bootleg-2020	483	8	R	R	NNP
orr-bootleg-2020	483	9	Curran	Curran	NNP
orr-bootleg-2020	483	10	,	,	,
orr-bootleg-2020	483	11	and	and	CC
orr-bootleg-2020	483	12	Tara	Tara	NNP
orr-bootleg-2020	483	13	Murphy	Murphy	NNP
orr-bootleg-2020	483	14	.	.	.
orr-bootleg-2020	484	1	Transforming	transform	VBG
orr-bootleg-2020	484	2	wikipedia	wikipedia	NN
orr-bootleg-2020	484	3	into	into	IN
orr-bootleg-2020	484	4	named	name	VBN
orr-bootleg-2020	484	5	entity	entity	NN
orr-bootleg-2020	484	6	training	training	NN
orr-bootleg-2020	484	7	data	datum	NNS
orr-bootleg-2020	484	8	.	.	.
orr-bootleg-2020	485	1	In	in	IN
orr-bootleg-2020	485	2	Proceedings	Proceedings	NNP
orr-bootleg-2020	485	3	of	of	IN
orr-bootleg-2020	485	4	the	the	DT
orr-bootleg-2020	485	5	Australasian	Australasian	NNP
orr-bootleg-2020	485	6	Language	Language	NNP
orr-bootleg-2020	485	7	Technology	Technology	NNP
orr-bootleg-2020	485	8	Association	Association	NNP
orr-bootleg-2020	485	9	Workshop	Workshop	NNP
orr-bootleg-2020	485	10	2008	2008	CD
orr-bootleg-2020	485	11	,	,	,
orr-bootleg-2020	485	12	pages	page	NNS
orr-bootleg-2020	485	13	124–132	124–132	CD
orr-bootleg-2020	485	14	,	,	,
orr-bootleg-2020	485	15	2008	2008	CD
orr-bootleg-2020	485	16	.	.	.
orr-bootleg-2020	486	1	[	[	-LRB-
orr-bootleg-2020	486	2	35	35	CD
orr-bootleg-2020	486	3	]	]	-RRB-
orr-bootleg-2020	486	4	Alberto	Alberto	NNP
orr-bootleg-2020	486	5	Parravicini	Parravicini	NNP
orr-bootleg-2020	486	6	,	,	,
orr-bootleg-2020	486	7	Rhicheek	Rhicheek	NNP
orr-bootleg-2020	486	8	Patra	Patra	NNP
orr-bootleg-2020	486	9	,	,	,
orr-bootleg-2020	486	10	Davide	Davide	NNP
orr-bootleg-2020	486	11	B	B	NNP
orr-bootleg-2020	486	12	Bartolini	Bartolini	NNP
orr-bootleg-2020	486	13	,	,	,
orr-bootleg-2020	486	14	and	and	CC
orr-bootleg-2020	486	15	Marco	Marco	NNP
orr-bootleg-2020	486	16	D	D	NNP
orr-bootleg-2020	486	17	Santambrogio	Santambrogio	NNP
orr-bootleg-2020	486	18	.	.	.
orr-bootleg-2020	487	1	Fast	fast	JJ
orr-bootleg-2020	487	2	and	and	CC
orr-bootleg-2020	487	3	accurate	accurate	JJ
orr-bootleg-2020	487	4	entity	entity	NN
orr-bootleg-2020	487	5	linking	link	VBG
orr-bootleg-2020	487	6	via	via	IN
orr-bootleg-2020	487	7	graph	graph	NN
orr-bootleg-2020	487	8	embedding	embedding	NN
orr-bootleg-2020	487	9	.	.	.
orr-bootleg-2020	488	1	In	in	IN
orr-bootleg-2020	488	2	Proceedings	Proceedings	NNP
orr-bootleg-2020	488	3	of	of	IN
orr-bootleg-2020	488	4	the	the	DT
orr-bootleg-2020	488	5	2nd	2nd	JJ
orr-bootleg-2020	488	6	Joint	Joint	NNP
orr-bootleg-2020	488	7	International	International	NNP
orr-bootleg-2020	488	8	Workshop	Workshop	NNP
orr-bootleg-2020	488	9	on	on	IN
orr-bootleg-2020	488	10	Graph	Graph	NNP
orr-bootleg-2020	488	11	Data	Data	NNP
orr-bootleg-2020	488	12	Management	Management	NNP
orr-bootleg-2020	488	13	Experiences	Experiences	NNPS
orr-bootleg-2020	488	14	&	&	CC
orr-bootleg-2020	488	15	Systems	Systems	NNPS
orr-bootleg-2020	488	16	(	(	-LRB-
orr-bootleg-2020	488	17	GRADES	GRADES	NNP
orr-bootleg-2020	488	18	)	)	-RRB-
orr-bootleg-2020	488	19	and	and	CC
orr-bootleg-2020	488	20	Network	Network	NNP
orr-bootleg-2020	488	21	Data	Data	NNP
orr-bootleg-2020	488	22	Analytics	Analytics	NNP
orr-bootleg-2020	488	23	(	(	-LRB-
orr-bootleg-2020	488	24	NDA	NDA	NNP
orr-bootleg-2020	488	25	)	)	-RRB-
orr-bootleg-2020	488	26	,	,	,
orr-bootleg-2020	488	27	pages	page	NNS
orr-bootleg-2020	488	28	1–9	1–9	CD
orr-bootleg-2020	488	29	,	,	,
orr-bootleg-2020	488	30	2019	2019	CD
orr-bootleg-2020	488	31	.	.	.
orr-bootleg-2020	489	1	[	[	-LRB-
orr-bootleg-2020	489	2	36	36	CD
orr-bootleg-2020	489	3	]	]	-RRB-
orr-bootleg-2020	489	4	Maria	Maria	NNP
orr-bootleg-2020	489	5	Pershina	Pershina	NNP
orr-bootleg-2020	489	6	,	,	,
orr-bootleg-2020	489	7	Yifan	Yifan	NNP
orr-bootleg-2020	489	8	He	He	NNP
orr-bootleg-2020	489	9	,	,	,
orr-bootleg-2020	489	10	and	and	CC
orr-bootleg-2020	489	11	Ralph	Ralph	NNP
orr-bootleg-2020	489	12	Grishman	Grishman	NNP
orr-bootleg-2020	489	13	.	.	.
orr-bootleg-2020	490	1	Personalized	personalized	JJ
orr-bootleg-2020	490	2	page	page	NN
orr-bootleg-2020	490	3	rank	rank	NN
orr-bootleg-2020	490	4	for	for	IN
orr-bootleg-2020	490	5	named	name	VBN
orr-bootleg-2020	490	6	entity	entity	NN
orr-bootleg-2020	490	7	disambiguation	disambiguation	NN
orr-bootleg-2020	490	8	.	.	.
orr-bootleg-2020	491	1	In	in	IN
orr-bootleg-2020	491	2	NAACL	naacl	NN
orr-bootleg-2020	491	3	,	,	,
orr-bootleg-2020	491	4	2015	2015	CD
orr-bootleg-2020	491	5	.	.	.
orr-bootleg-2020	492	1	[	[	-LRB-
orr-bootleg-2020	492	2	37	37	CD
orr-bootleg-2020	492	3	]	]	-RRB-
orr-bootleg-2020	492	4	Georgios	Georgios	NNP
orr-bootleg-2020	492	5	Petasis	Petasis	NNP
orr-bootleg-2020	492	6	,	,	,
orr-bootleg-2020	492	7	Frantz	Frantz	NNP
orr-bootleg-2020	492	8	Vichot	Vichot	NNP
orr-bootleg-2020	492	9	,	,	,
orr-bootleg-2020	492	10	Francis	Francis	NNP
orr-bootleg-2020	492	11	Wolinski	Wolinski	NNP
orr-bootleg-2020	492	12	,	,	,
orr-bootleg-2020	492	13	Georgios	Georgios	NNP
orr-bootleg-2020	492	14	Paliouras	Paliouras	NNP
orr-bootleg-2020	492	15	,	,	,
orr-bootleg-2020	492	16	Vangelis	Vangelis	NNP
orr-bootleg-2020	492	17	Karkaletsis	Karkaletsis	NNP
orr-bootleg-2020	492	18	,	,	,
orr-bootleg-2020	492	19	and	and	CC
orr-bootleg-2020	492	20	Con-	Con-	NNP
orr-bootleg-2020	492	21	stantine	stantine	VBP
orr-bootleg-2020	492	22	D	D	NNP
orr-bootleg-2020	492	23	Spyropoulos	Spyropoulos	NNP
orr-bootleg-2020	492	24	.	.	.
orr-bootleg-2020	493	1	Using	use	VBG
orr-bootleg-2020	493	2	machine	machine	NN
orr-bootleg-2020	493	3	learning	learn	VBG
orr-bootleg-2020	493	4	to	to	TO
orr-bootleg-2020	493	5	maintain	maintain	VB
orr-bootleg-2020	493	6	rule	rule	NN
orr-bootleg-2020	493	7	-	-	HYPH
orr-bootleg-2020	493	8	based	base	VBN
orr-bootleg-2020	493	9	named	name	VBN
orr-bootleg-2020	493	10	-	-	HYPH
orr-bootleg-2020	493	11	entity	entity	NN
orr-bootleg-2020	493	12	recognition	recognition	NN
orr-bootleg-2020	493	13	and	and	CC
orr-bootleg-2020	493	14	classification	classification	NN
orr-bootleg-2020	493	15	systems	system	NNS
orr-bootleg-2020	493	16	.	.	.
orr-bootleg-2020	494	1	In	in	IN
orr-bootleg-2020	494	2	Proceedings	proceeding	NNS
orr-bootleg-2020	494	3	of	of	IN
orr-bootleg-2020	494	4	the	the	DT
orr-bootleg-2020	494	5	39th	39th	JJ
orr-bootleg-2020	494	6	Annual	Annual	NNP
orr-bootleg-2020	494	7	Meeting	Meeting	NNP
orr-bootleg-2020	494	8	of	of	IN
orr-bootleg-2020	494	9	the	the	DT
orr-bootleg-2020	494	10	Association	Association	NNP
orr-bootleg-2020	494	11	for	for	IN
orr-bootleg-2020	494	12	Computational	Computational	NNP
orr-bootleg-2020	494	13	Linguistics	Linguistics	NNP
orr-bootleg-2020	494	14	,	,	,
orr-bootleg-2020	494	15	pages	page	VBZ
orr-bootleg-2020	494	16	426–433	426–433	CD
orr-bootleg-2020	494	17	,	,	,
orr-bootleg-2020	494	18	2001	2001	CD
orr-bootleg-2020	494	19	.	.	.
orr-bootleg-2020	495	1	[	[	-LRB-
orr-bootleg-2020	495	2	38	38	CD
orr-bootleg-2020	495	3	]	]	-RRB-
orr-bootleg-2020	495	4	Matthew	Matthew	NNP
orr-bootleg-2020	495	5	E.	E.	NNP
orr-bootleg-2020	495	6	Peters	Peters	NNP
orr-bootleg-2020	495	7	,	,	,
orr-bootleg-2020	495	8	Mark	Mark	NNP
orr-bootleg-2020	495	9	Neumann	Neumann	NNP
orr-bootleg-2020	495	10	,	,	,
orr-bootleg-2020	495	11	Robert	Robert	NNP
orr-bootleg-2020	495	12	Logan	Logan	NNP
orr-bootleg-2020	495	13	,	,	,
orr-bootleg-2020	495	14	Roy	Roy	NNP
orr-bootleg-2020	495	15	Schwartz	Schwartz	NNP
orr-bootleg-2020	495	16	,	,	,
orr-bootleg-2020	495	17	Vidur	Vidur	NNP
orr-bootleg-2020	495	18	Joshi	Joshi	NNP
orr-bootleg-2020	495	19	,	,	,
orr-bootleg-2020	495	20	Sameer	Sameer	NNP
orr-bootleg-2020	495	21	Singh	Singh	NNP
orr-bootleg-2020	495	22	,	,	,
orr-bootleg-2020	495	23	and	and	CC
orr-bootleg-2020	495	24	Noah	Noah	NNP
orr-bootleg-2020	495	25	A.	A.	NNP
orr-bootleg-2020	495	26	Smith	Smith	NNP
orr-bootleg-2020	495	27	.	.	.
orr-bootleg-2020	496	1	Knowledge	knowledge	NN
orr-bootleg-2020	496	2	enhanced	enhance	VBD
orr-bootleg-2020	496	3	contextual	contextual	JJ
orr-bootleg-2020	496	4	word	word	NN
orr-bootleg-2020	496	5	representations	representation	NNS
orr-bootleg-2020	496	6	.	.	.
orr-bootleg-2020	497	1	In	in	IN
orr-bootleg-2020	497	2	Proceedings	Proceedings	NNP
orr-bootleg-2020	497	3	of	of	IN
orr-bootleg-2020	497	4	the	the	DT
orr-bootleg-2020	497	5	2019	2019	CD
orr-bootleg-2020	497	6	Conference	Conference	NNP
orr-bootleg-2020	497	7	on	on	IN
orr-bootleg-2020	497	8	Empirical	Empirical	NNP
orr-bootleg-2020	497	9	Methods	Methods	NNPS
orr-bootleg-2020	497	10	in	in	IN
orr-bootleg-2020	497	11	Natural	Natural	NNP
orr-bootleg-2020	497	12	Language	Language	NNP
orr-bootleg-2020	497	13	Processing	Processing	NNP
orr-bootleg-2020	497	14	and	and	CC
orr-bootleg-2020	497	15	the	the	DT
orr-bootleg-2020	497	16	9th	9th	JJ
orr-bootleg-2020	497	17	International	International	NNP
orr-bootleg-2020	497	18	Joint	Joint	NNP
orr-bootleg-2020	497	19	Conference	Conference	NNP
orr-bootleg-2020	497	20	on	on	IN
orr-bootleg-2020	497	21	Natural	Natural	NNP
orr-bootleg-2020	497	22	Language	Language	NNP
orr-bootleg-2020	497	23	Processing	Processing	NNP
orr-bootleg-2020	497	24	(	(	-LRB-
orr-bootleg-2020	497	25	EMNLP	EMNLP	NNP
orr-bootleg-2020	497	26	-	-	HYPH
orr-bootleg-2020	497	27	IJCNLP	IJCNLP	NNP
orr-bootleg-2020	497	28	)	)	-RRB-
orr-bootleg-2020	497	29	,	,	,
orr-bootleg-2020	497	30	pages	page	NNS
orr-bootleg-2020	497	31	43–54	43–54	CD
orr-bootleg-2020	497	32	,	,	,
orr-bootleg-2020	497	33	Hong	Hong	NNP
orr-bootleg-2020	497	34	Kong	Kong	NNP
orr-bootleg-2020	497	35	,	,	,
orr-bootleg-2020	497	36	China	China	NNP
orr-bootleg-2020	497	37	,	,	,
orr-bootleg-2020	497	38	November	November	NNP
orr-bootleg-2020	497	39	2019	2019	CD
orr-bootleg-2020	497	40	.	.	.
orr-bootleg-2020	498	1	Association	Association	NNP
orr-bootleg-2020	498	2	for	for	IN
orr-bootleg-2020	498	3	Computational	Computational	NNP
orr-bootleg-2020	498	4	Linguistics	Linguistics	NNP
orr-bootleg-2020	498	5	.	.	.
orr-bootleg-2020	499	1	doi	doi	XX
orr-bootleg-2020	499	2	:	:	:
orr-bootleg-2020	499	3	10.18653	10.18653	CD
orr-bootleg-2020	499	4	/	/	SYM
orr-bootleg-2020	499	5	v1	v1	NN
orr-bootleg-2020	499	6	/	/	SYM
orr-bootleg-2020	499	7	D19	D19	NNP
orr-bootleg-2020	499	8	-	-	HYPH
orr-bootleg-2020	499	9	1005	1005	CD
orr-bootleg-2020	499	10	.	.	.
orr-bootleg-2020	500	1	17	17	CD
orr-bootleg-2020	500	2	[	[	-LRB-
orr-bootleg-2020	500	3	39	39	CD
orr-bootleg-2020	500	4	]	]	-RRB-
orr-bootleg-2020	500	5	Fabio	Fabio	NNP
orr-bootleg-2020	500	6	Petroni	Petroni	NNP
orr-bootleg-2020	500	7	,	,	,
orr-bootleg-2020	500	8	Aleksandra	Aleksandra	NNP
orr-bootleg-2020	500	9	Piktus	Piktus	NNP
orr-bootleg-2020	500	10	,	,	,
orr-bootleg-2020	500	11	Angela	Angela	NNP
orr-bootleg-2020	500	12	Fan	Fan	NNP
orr-bootleg-2020	500	13	,	,	,
orr-bootleg-2020	500	14	Patrick	Patrick	NNP
orr-bootleg-2020	500	15	Lewis	Lewis	NNP
orr-bootleg-2020	500	16	,	,	,
orr-bootleg-2020	500	17	Majid	Majid	NNP
orr-bootleg-2020	500	18	Yazdani	Yazdani	NNP
orr-bootleg-2020	500	19	,	,	,
orr-bootleg-2020	500	20	Nicola	Nicola	NNP
orr-bootleg-2020	500	21	De	De	NNP
orr-bootleg-2020	500	22	Cao	Cao	NNP
orr-bootleg-2020	500	23	,	,	,
orr-bootleg-2020	500	24	James	James	NNP
orr-bootleg-2020	500	25	Thorne	Thorne	NNP
orr-bootleg-2020	500	26	,	,	,
orr-bootleg-2020	500	27	Yacine	Yacine	NNP
orr-bootleg-2020	500	28	Jernite	Jernite	NNP
orr-bootleg-2020	500	29	,	,	,
orr-bootleg-2020	500	30	Vassilis	Vassilis	NNP
orr-bootleg-2020	500	31	Plachouras	Plachouras	NNP
orr-bootleg-2020	500	32	,	,	,
orr-bootleg-2020	500	33	Tim	Tim	NNP
orr-bootleg-2020	500	34	Rocktäschel	Rocktäschel	NNP
orr-bootleg-2020	500	35	,	,	,
orr-bootleg-2020	500	36	et	et	NNP
orr-bootleg-2020	500	37	al	al	NNP
orr-bootleg-2020	500	38	.	.	.
orr-bootleg-2020	501	1	Kilt	kilt	VB
orr-bootleg-2020	501	2	:	:	:
orr-bootleg-2020	501	3	a	a	DT
orr-bootleg-2020	501	4	benchmark	benchmark	NN
orr-bootleg-2020	501	5	for	for	IN
orr-bootleg-2020	501	6	knowledge	knowledge	NN
orr-bootleg-2020	501	7	intensive	intensive	JJ
orr-bootleg-2020	501	8	language	language	NN
orr-bootleg-2020	501	9	tasks	task	NNS
orr-bootleg-2020	501	10	.	.	.
orr-bootleg-2020	502	1	arXiv	arXiv	NNP
orr-bootleg-2020	502	2	preprint	preprint	NN
orr-bootleg-2020	502	3	arXiv:2009.02252	arXiv:2009.02252	NNP
orr-bootleg-2020	502	4	,	,	,
orr-bootleg-2020	502	5	2020	2020	CD
orr-bootleg-2020	502	6	.	.	.
orr-bootleg-2020	503	1	[	[	-LRB-
orr-bootleg-2020	503	2	40	40	CD
orr-bootleg-2020	503	3	]	]	-RRB-
orr-bootleg-2020	503	4	Minh	Minh	NNP
orr-bootleg-2020	503	5	C.	C.	NNP
orr-bootleg-2020	503	6	Phan	Phan	NNP
orr-bootleg-2020	503	7	,	,	,
orr-bootleg-2020	503	8	Aixin	Aixin	NNP
orr-bootleg-2020	503	9	Sun	Sun	NNP
orr-bootleg-2020	503	10	,	,	,
orr-bootleg-2020	503	11	Yi	Yi	NNP
orr-bootleg-2020	503	12	Tay	Tay	NNP
orr-bootleg-2020	503	13	,	,	,
orr-bootleg-2020	503	14	Jialong	Jialong	NNP
orr-bootleg-2020	503	15	Han	Han	NNP
orr-bootleg-2020	503	16	,	,	,
orr-bootleg-2020	503	17	and	and	CC
orr-bootleg-2020	503	18	Chenliang	Chenliang	NNP
orr-bootleg-2020	503	19	Li	Li	NNP
orr-bootleg-2020	503	20	.	.	.
orr-bootleg-2020	504	1	Pair	pair	NN
orr-bootleg-2020	504	2	-	-	HYPH
orr-bootleg-2020	504	3	linking	link	VBG
orr-bootleg-2020	504	4	for	for	IN
orr-bootleg-2020	504	5	collective	collective	JJ
orr-bootleg-2020	504	6	entity	entity	NN
orr-bootleg-2020	504	7	disambiguation	disambiguation	NN
orr-bootleg-2020	504	8	:	:	:
orr-bootleg-2020	504	9	Two	two	CD
orr-bootleg-2020	504	10	could	could	MD
orr-bootleg-2020	504	11	be	be	VB
orr-bootleg-2020	504	12	better	well	JJR
orr-bootleg-2020	504	13	than	than	IN
orr-bootleg-2020	504	14	all	all	DT
orr-bootleg-2020	504	15	.	.	.
orr-bootleg-2020	505	1	TKDE	TKDE	NNP
orr-bootleg-2020	505	2	,	,	,
orr-bootleg-2020	505	3	2019	2019	CD
orr-bootleg-2020	505	4	.	.	.
orr-bootleg-2020	506	1	[	[	-LRB-
orr-bootleg-2020	506	2	41	41	CD
orr-bootleg-2020	506	3	]	]	-RRB-
orr-bootleg-2020	506	4	N	N	NNP
orr-bootleg-2020	506	5	Poerner	Poerner	NNP
orr-bootleg-2020	506	6	,	,	,
orr-bootleg-2020	506	7	U	U	NNP
orr-bootleg-2020	506	8	Waltinger	Waltinger	NNP
orr-bootleg-2020	506	9	,	,	,
orr-bootleg-2020	506	10	and	and	CC
orr-bootleg-2020	506	11	H	H	NNP
orr-bootleg-2020	506	12	Schütze	Schütze	NNP
orr-bootleg-2020	506	13	.	.	.
orr-bootleg-2020	507	1	E	E	NNP
orr-bootleg-2020	507	2	-	-	NN
orr-bootleg-2020	507	3	bert	bert	NN
orr-bootleg-2020	507	4	:	:	:
orr-bootleg-2020	507	5	Efficient	Efficient	NNP
orr-bootleg-2020	507	6	-	-	HYPH
orr-bootleg-2020	507	7	yet	yet	RB
orr-bootleg-2020	507	8	-	-	HYPH
orr-bootleg-2020	507	9	effective	effective	JJ
orr-bootleg-2020	507	10	entity	entity	NN
orr-bootleg-2020	507	11	embeddings	embedding	NNS
orr-bootleg-2020	507	12	for	for	IN
orr-bootleg-2020	507	13	bert	bert	NNP
orr-bootleg-2020	507	14	.	.	.
orr-bootleg-2020	508	1	arXiv	arXiv	NNP
orr-bootleg-2020	508	2	preprint	preprint	NN
orr-bootleg-2020	508	3	arXiv:1911.03681	arXiv:1911.03681	NNP
orr-bootleg-2020	508	4	,	,	,
orr-bootleg-2020	508	5	2019	2019	CD
orr-bootleg-2020	508	6	.	.	.
orr-bootleg-2020	509	1	[	[	-LRB-
orr-bootleg-2020	509	2	42	42	CD
orr-bootleg-2020	509	3	]	]	-RRB-
orr-bootleg-2020	509	4	Priya	Priya	NNP
orr-bootleg-2020	509	5	Radhakrishnan	Radhakrishnan	NNP
orr-bootleg-2020	509	6	,	,	,
orr-bootleg-2020	509	7	Partha	Partha	NNP
orr-bootleg-2020	509	8	Talukdar	Talukdar	NNP
orr-bootleg-2020	509	9	,	,	,
orr-bootleg-2020	509	10	and	and	CC
orr-bootleg-2020	509	11	Vasudeva	Vasudeva	NNP
orr-bootleg-2020	509	12	Varma	Varma	NNP
orr-bootleg-2020	509	13	.	.	.
orr-bootleg-2020	510	1	Elden	Elden	NNP
orr-bootleg-2020	510	2	:	:	:
orr-bootleg-2020	510	3	Improved	improved	JJ
orr-bootleg-2020	510	4	entity	entity	NN
orr-bootleg-2020	510	5	linking	linking	NN
orr-bootleg-2020	510	6	using	use	VBG
orr-bootleg-2020	510	7	densified	densifie	VBN
orr-bootleg-2020	510	8	knowledge	knowledge	NN
orr-bootleg-2020	510	9	graphs	graphs	NN
orr-bootleg-2020	510	10	.	.	.
orr-bootleg-2020	511	1	In	in	IN
orr-bootleg-2020	511	2	Proceedings	proceeding	NNS
orr-bootleg-2020	511	3	of	of	IN
orr-bootleg-2020	511	4	the	the	DT
orr-bootleg-2020	511	5	2018	2018	CD
orr-bootleg-2020	511	6	Conference	Conference	NNP
orr-bootleg-2020	511	7	of	of	IN
orr-bootleg-2020	511	8	the	the	DT
orr-bootleg-2020	511	9	North	north	JJ
orr-bootleg-2020	511	10	American	American	NNP
orr-bootleg-2020	511	11	Chapter	Chapter	NNP
orr-bootleg-2020	511	12	of	of	IN
orr-bootleg-2020	511	13	the	the	DT
orr-bootleg-2020	511	14	Association	Association	NNP
orr-bootleg-2020	511	15	for	for	IN
orr-bootleg-2020	511	16	Computational	Computational	NNP
orr-bootleg-2020	511	17	Linguistics	Linguistics	NNPS
orr-bootleg-2020	511	18	:	:	:
orr-bootleg-2020	511	19	Human	Human	NNP
orr-bootleg-2020	511	20	Language	Language	NNP
orr-bootleg-2020	511	21	Technologies	Technologies	NNPS
orr-bootleg-2020	511	22	,	,	,
orr-bootleg-2020	511	23	Volume	volume	NN
orr-bootleg-2020	511	24	1	1	CD
orr-bootleg-2020	511	25	(	(	-LRB-
orr-bootleg-2020	511	26	Long	Long	NNP
orr-bootleg-2020	511	27	Papers	Papers	NNPS
orr-bootleg-2020	511	28	)	)	-RRB-
orr-bootleg-2020	511	29	,	,	,
orr-bootleg-2020	511	30	pages	page	NNS
orr-bootleg-2020	511	31	1844–1853	1844–1853	CD
orr-bootleg-2020	511	32	,	,	,
orr-bootleg-2020	511	33	2018	2018	CD
orr-bootleg-2020	511	34	.	.	.
orr-bootleg-2020	512	1	[	[	-LRB-
orr-bootleg-2020	512	2	43	43	CD
orr-bootleg-2020	512	3	]	]	-RRB-
orr-bootleg-2020	512	4	Jonathan	Jonathan	NNP
orr-bootleg-2020	512	5	Raphael	Raphael	NNP
orr-bootleg-2020	512	6	Raiman	Raiman	NNP
orr-bootleg-2020	512	7	and	and	CC
orr-bootleg-2020	512	8	Olivier	Olivier	NNP
orr-bootleg-2020	512	9	Michel	Michel	NNP
orr-bootleg-2020	512	10	Raiman	Raiman	NNP
orr-bootleg-2020	512	11	.	.	.
orr-bootleg-2020	513	1	Deeptype	deeptype	NN
orr-bootleg-2020	513	2	:	:	:
orr-bootleg-2020	513	3	multilingual	multilingual	JJ
orr-bootleg-2020	513	4	entity	entity	NN
orr-bootleg-2020	513	5	linking	link	VBG
orr-bootleg-2020	513	6	by	by	IN
orr-bootleg-2020	513	7	neural	neural	JJ
orr-bootleg-2020	513	8	type	type	NN
orr-bootleg-2020	513	9	system	system	NN
orr-bootleg-2020	513	10	evolution	evolution	NN
orr-bootleg-2020	513	11	.	.	.
orr-bootleg-2020	514	1	In	in	IN
orr-bootleg-2020	514	2	Thirty	thirty	CD
orr-bootleg-2020	514	3	-	-	HYPH
orr-bootleg-2020	514	4	Second	Second	NNP
orr-bootleg-2020	514	5	AAAI	AAAI	NNP
orr-bootleg-2020	514	6	Conference	Conference	NNP
orr-bootleg-2020	514	7	on	on	IN
orr-bootleg-2020	514	8	Artificial	Artificial	NNP
orr-bootleg-2020	514	9	Intelligence	Intelligence	NNP
orr-bootleg-2020	514	10	,	,	,
orr-bootleg-2020	514	11	2018	2018	CD
orr-bootleg-2020	514	12	.	.	.
orr-bootleg-2020	515	1	[	[	-LRB-
orr-bootleg-2020	515	2	44	44	CD
orr-bootleg-2020	515	3	]	]	-RRB-
orr-bootleg-2020	515	4	Alexander	Alexander	NNP
orr-bootleg-2020	515	5	Ratner	Ratner	NNP
orr-bootleg-2020	515	6	,	,	,
orr-bootleg-2020	515	7	Stephen	Stephen	NNP
orr-bootleg-2020	515	8	H	H	NNP
orr-bootleg-2020	515	9	Bach	Bach	NNP
orr-bootleg-2020	515	10	,	,	,
orr-bootleg-2020	515	11	Henry	Henry	NNP
orr-bootleg-2020	515	12	Ehrenberg	Ehrenberg	NNP
orr-bootleg-2020	515	13	,	,	,
orr-bootleg-2020	515	14	Jason	Jason	NNP
orr-bootleg-2020	515	15	Fries	Fries	NNP
orr-bootleg-2020	515	16	,	,	,
orr-bootleg-2020	515	17	Sen	Sen	NNP
orr-bootleg-2020	515	18	Wu	Wu	NNP
orr-bootleg-2020	515	19	,	,	,
orr-bootleg-2020	515	20	and	and	CC
orr-bootleg-2020	515	21	Christopher	Christopher	NNP
orr-bootleg-2020	515	22	Ré	Ré	NNP
orr-bootleg-2020	515	23	.	.	.
orr-bootleg-2020	516	1	Snorkel	snorkel	NN
orr-bootleg-2020	516	2	:	:	:
orr-bootleg-2020	516	3	Rapid	rapid	JJ
orr-bootleg-2020	516	4	training	training	NN
orr-bootleg-2020	516	5	data	datum	NNS
orr-bootleg-2020	516	6	creation	creation	NN
orr-bootleg-2020	516	7	with	with	IN
orr-bootleg-2020	516	8	weak	weak	JJ
orr-bootleg-2020	516	9	supervision	supervision	NN
orr-bootleg-2020	516	10	.	.	.
orr-bootleg-2020	517	1	In	in	IN
orr-bootleg-2020	517	2	VLDB	VLDB	NNP
orr-bootleg-2020	517	3	,	,	,
orr-bootleg-2020	517	4	2017	2017	CD
orr-bootleg-2020	517	5	.	.	.
orr-bootleg-2020	518	1	[	[	-LRB-
orr-bootleg-2020	518	2	45	45	CD
orr-bootleg-2020	518	3	]	]	-RRB-
orr-bootleg-2020	518	4	Christopher	Christopher	NNP
orr-bootleg-2020	518	5	Ré	Ré	NNP
orr-bootleg-2020	518	6	,	,	,
orr-bootleg-2020	518	7	Feng	Feng	NNP
orr-bootleg-2020	518	8	Niu	Niu	NNP
orr-bootleg-2020	518	9	,	,	,
orr-bootleg-2020	518	10	Pallavi	Pallavi	NNP
orr-bootleg-2020	518	11	Gudipati	Gudipati	NNP
orr-bootleg-2020	518	12	,	,	,
orr-bootleg-2020	518	13	and	and	CC
orr-bootleg-2020	518	14	Charles	Charles	NNP
orr-bootleg-2020	518	15	Srisuwananukorn	Srisuwananukorn	NNP
orr-bootleg-2020	518	16	.	.	.
orr-bootleg-2020	519	1	Overton	Overton	NNP
orr-bootleg-2020	519	2	:	:	:
orr-bootleg-2020	519	3	A	a	DT
orr-bootleg-2020	519	4	data	data	NN
orr-bootleg-2020	519	5	system	system	NN
orr-bootleg-2020	519	6	for	for	IN
orr-bootleg-2020	519	7	monitoring	monitor	VBG
orr-bootleg-2020	519	8	and	and	CC
orr-bootleg-2020	519	9	improving	improve	VBG
orr-bootleg-2020	519	10	machine	machine	NN
orr-bootleg-2020	519	11	-	-	HYPH
orr-bootleg-2020	519	12	learned	learn	VBN
orr-bootleg-2020	519	13	products	product	NNS
orr-bootleg-2020	519	14	.	.	.
orr-bootleg-2020	520	1	In	in	IN
orr-bootleg-2020	520	2	CIDR	CIDR	NNP
orr-bootleg-2020	520	3	,	,	,
orr-bootleg-2020	520	4	2020	2020	CD
orr-bootleg-2020	520	5	.	.	.
orr-bootleg-2020	521	1	[	[	-LRB-
orr-bootleg-2020	521	2	46	46	CD
orr-bootleg-2020	521	3	]	]	-RRB-
orr-bootleg-2020	521	4	Hamed	hamed	JJ
orr-bootleg-2020	521	5	Shahbazi	Shahbazi	NNP
orr-bootleg-2020	521	6	,	,	,
orr-bootleg-2020	521	7	Xiaoli	Xiaoli	NNP
orr-bootleg-2020	521	8	Z	Z	NNP
orr-bootleg-2020	521	9	Fern	Fern	NNP
orr-bootleg-2020	521	10	,	,	,
orr-bootleg-2020	521	11	Reza	Reza	NNP
orr-bootleg-2020	521	12	Ghaeini	Ghaeini	NNP
orr-bootleg-2020	521	13	,	,	,
orr-bootleg-2020	521	14	Rasha	Rasha	NNP
orr-bootleg-2020	521	15	Obeidat	Obeidat	NNP
orr-bootleg-2020	521	16	,	,	,
orr-bootleg-2020	521	17	and	and	CC
orr-bootleg-2020	521	18	Prasad	Prasad	NNP
orr-bootleg-2020	521	19	Tadepalli	Tadepalli	NNP
orr-bootleg-2020	521	20	.	.	.
orr-bootleg-2020	522	1	Entity	entity	NN
orr-bootleg-2020	522	2	-	-	HYPH
orr-bootleg-2020	522	3	aware	aware	JJ
orr-bootleg-2020	522	4	elmo	elmo	RB
orr-bootleg-2020	522	5	:	:	:
orr-bootleg-2020	522	6	Learning	learn	VBG
orr-bootleg-2020	522	7	contextual	contextual	JJ
orr-bootleg-2020	522	8	entity	entity	NN
orr-bootleg-2020	522	9	representation	representation	NN
orr-bootleg-2020	522	10	for	for	IN
orr-bootleg-2020	522	11	entity	entity	NN
orr-bootleg-2020	522	12	disambiguation	disambiguation	NN
orr-bootleg-2020	522	13	.	.	.
orr-bootleg-2020	523	1	arXiv	arXiv	NNP
orr-bootleg-2020	523	2	preprint	preprint	NN
orr-bootleg-2020	523	3	arXiv:1908.05762	arXiv:1908.05762	NNP
orr-bootleg-2020	523	4	,	,	,
orr-bootleg-2020	523	5	2019	2019	CD
orr-bootleg-2020	523	6	.	.	.
orr-bootleg-2020	524	1	[	[	-LRB-
orr-bootleg-2020	524	2	47	47	CD
orr-bootleg-2020	524	3	]	]	-RRB-
orr-bootleg-2020	524	4	Sandeep	Sandeep	NNP
orr-bootleg-2020	524	5	Tata	Tata	NNP
orr-bootleg-2020	524	6	,	,	,
orr-bootleg-2020	524	7	Vlad	Vlad	NNP
orr-bootleg-2020	524	8	Panait	Panait	NNP
orr-bootleg-2020	524	9	,	,	,
orr-bootleg-2020	524	10	Suming	Suming	NNP
orr-bootleg-2020	524	11	Jeremiah	Jeremiah	NNP
orr-bootleg-2020	524	12	Chen	Chen	NNP
orr-bootleg-2020	524	13	,	,	,
orr-bootleg-2020	524	14	and	and	CC
orr-bootleg-2020	524	15	Mike	Mike	NNP
orr-bootleg-2020	524	16	Colagrosso	Colagrosso	NNP
orr-bootleg-2020	524	17	.	.	.
orr-bootleg-2020	525	1	Itemsuggest	itemsuggest	JJ
orr-bootleg-2020	525	2	:	:	:
orr-bootleg-2020	525	3	A	a	DT
orr-bootleg-2020	525	4	data	datum	NNS
orr-bootleg-2020	525	5	management	management	NN
orr-bootleg-2020	525	6	platform	platform	NN
orr-bootleg-2020	525	7	for	for	IN
orr-bootleg-2020	525	8	machine	machine	NN
orr-bootleg-2020	525	9	learned	learn	VBD
orr-bootleg-2020	525	10	ranking	ranking	NN
orr-bootleg-2020	525	11	services	service	NNS
orr-bootleg-2020	525	12	.	.	.
orr-bootleg-2020	526	1	In	in	IN
orr-bootleg-2020	526	2	CIDR	CIDR	NNP
orr-bootleg-2020	526	3	,	,	,
orr-bootleg-2020	526	4	2019	2019	CD
orr-bootleg-2020	526	5	.	.	.
orr-bootleg-2020	527	1	[	[	-LRB-
orr-bootleg-2020	527	2	48	48	CD
orr-bootleg-2020	527	3	]	]	-RRB-
orr-bootleg-2020	527	4	Ashish	Ashish	NNP
orr-bootleg-2020	527	5	Vaswani	Vaswani	NNP
orr-bootleg-2020	527	6	,	,	,
orr-bootleg-2020	527	7	Noam	Noam	NNP
orr-bootleg-2020	527	8	Shazeer	Shazeer	NNP
orr-bootleg-2020	527	9	,	,	,
orr-bootleg-2020	527	10	Niki	Niki	NNP
orr-bootleg-2020	527	11	Parmar	Parmar	NNP
orr-bootleg-2020	527	12	,	,	,
orr-bootleg-2020	527	13	Jakob	Jakob	NNP
orr-bootleg-2020	527	14	Uszkoreit	Uszkoreit	NNP
orr-bootleg-2020	527	15	,	,	,
orr-bootleg-2020	527	16	Llion	Llion	NNP
orr-bootleg-2020	527	17	Jones	Jones	NNP
orr-bootleg-2020	527	18	,	,	,
orr-bootleg-2020	527	19	Aidan	Aidan	NNP
orr-bootleg-2020	527	20	N	N	NNP
orr-bootleg-2020	527	21	Gomez	Gomez	NNP
orr-bootleg-2020	527	22	,	,	,
orr-bootleg-2020	527	23	Łukasz	Łukasz	NNP
orr-bootleg-2020	527	24	Kaiser	Kaiser	NNP
orr-bootleg-2020	527	25	,	,	,
orr-bootleg-2020	527	26	and	and	CC
orr-bootleg-2020	527	27	Illia	Illia	NNP
orr-bootleg-2020	527	28	Polosukhin	Polosukhin	NNP
orr-bootleg-2020	527	29	.	.	.
orr-bootleg-2020	528	1	Attention	attention	NN
orr-bootleg-2020	528	2	is	be	VBZ
orr-bootleg-2020	528	3	all	all	DT
orr-bootleg-2020	528	4	you	-PRON-	PRP
orr-bootleg-2020	528	5	need	need	VBP
orr-bootleg-2020	528	6	.	.	.
orr-bootleg-2020	529	1	In	in	IN
orr-bootleg-2020	529	2	Neurips	Neurips	NNP
orr-bootleg-2020	529	3	,	,	,
orr-bootleg-2020	529	4	2017	2017	CD
orr-bootleg-2020	529	5	.	.	.
orr-bootleg-2020	530	1	[	[	-LRB-
orr-bootleg-2020	530	2	49	49	CD
orr-bootleg-2020	530	3	]	]	-RRB-
orr-bootleg-2020	530	4	Eric	Eric	NNP
orr-bootleg-2020	530	5	Wallace	Wallace	NNP
orr-bootleg-2020	530	6	,	,	,
orr-bootleg-2020	530	7	Yizhong	Yizhong	NNP
orr-bootleg-2020	530	8	Wang	Wang	NNP
orr-bootleg-2020	530	9	,	,	,
orr-bootleg-2020	530	10	Sujian	Sujian	NNP
orr-bootleg-2020	530	11	Li	Li	NNP
orr-bootleg-2020	530	12	,	,	,
orr-bootleg-2020	530	13	Sameer	Sameer	NNP
orr-bootleg-2020	530	14	Singh	Singh	NNP
orr-bootleg-2020	530	15	,	,	,
orr-bootleg-2020	530	16	and	and	CC
orr-bootleg-2020	530	17	Matt	Matt	NNP
orr-bootleg-2020	530	18	Gardner	Gardner	NNP
orr-bootleg-2020	530	19	.	.	.
orr-bootleg-2020	531	1	Do	do	VBP
orr-bootleg-2020	531	2	nlp	nlp	NN
orr-bootleg-2020	531	3	models	model	NNS
orr-bootleg-2020	531	4	know	know	VB
orr-bootleg-2020	531	5	numbers	number	NNS
orr-bootleg-2020	531	6	?	?	.
orr-bootleg-2020	532	1	probing	probe	VBG
orr-bootleg-2020	532	2	numeracy	numeracy	NN
orr-bootleg-2020	532	3	in	in	IN
orr-bootleg-2020	532	4	embeddings	embedding	NNS
orr-bootleg-2020	532	5	.	.	.
orr-bootleg-2020	533	1	arXiv	arXiv	NNP
orr-bootleg-2020	533	2	preprint	preprint	NN
orr-bootleg-2020	533	3	arXiv:1909.07940	arXiv:1909.07940	NNP
orr-bootleg-2020	533	4	,	,	,
orr-bootleg-2020	533	5	2019	2019	CD
orr-bootleg-2020	533	6	.	.	.
orr-bootleg-2020	534	1	[	[	-LRB-
orr-bootleg-2020	534	2	50	50	CD
orr-bootleg-2020	534	3	]	]	-RRB-
orr-bootleg-2020	534	4	Vikas	Vikas	NNP
orr-bootleg-2020	534	5	Yadav	Yadav	NNP
orr-bootleg-2020	534	6	and	and	CC
orr-bootleg-2020	534	7	Steven	Steven	NNP
orr-bootleg-2020	534	8	Bethard	Bethard	NNP
orr-bootleg-2020	534	9	.	.	.
orr-bootleg-2020	535	1	A	a	DT
orr-bootleg-2020	535	2	survey	survey	NN
orr-bootleg-2020	535	3	on	on	IN
orr-bootleg-2020	535	4	recent	recent	JJ
orr-bootleg-2020	535	5	advances	advance	NNS
orr-bootleg-2020	535	6	in	in	IN
orr-bootleg-2020	535	7	named	name	VBN
orr-bootleg-2020	535	8	entity	entity	NN
orr-bootleg-2020	535	9	recognition	recognition	NN
orr-bootleg-2020	535	10	from	from	IN
orr-bootleg-2020	535	11	deep	deep	JJ
orr-bootleg-2020	535	12	learning	learning	NN
orr-bootleg-2020	535	13	models	model	NNS
orr-bootleg-2020	535	14	.	.	.
orr-bootleg-2020	536	1	arXiv	arXiv	NNP
orr-bootleg-2020	536	2	preprint	preprint	NN
orr-bootleg-2020	536	3	arXiv:1910.11470	arXiv:1910.11470	NNP
orr-bootleg-2020	536	4	,	,	,
orr-bootleg-2020	536	5	2019	2019	CD
orr-bootleg-2020	536	6	.	.	.
orr-bootleg-2020	537	1	[	[	-LRB-
orr-bootleg-2020	537	2	51	51	CD
orr-bootleg-2020	537	3	]	]	-RRB-
orr-bootleg-2020	537	4	Ikuya	Ikuya	NNP
orr-bootleg-2020	537	5	Yamada	Yamada	NNP
orr-bootleg-2020	537	6	and	and	CC
orr-bootleg-2020	537	7	Hiroyuki	Hiroyuki	NNP
orr-bootleg-2020	537	8	Shindo	Shindo	NNP
orr-bootleg-2020	537	9	.	.	.
orr-bootleg-2020	538	1	Pre	pre	JJ
orr-bootleg-2020	538	2	-	-	JJ
orr-bootleg-2020	538	3	training	training	NN
orr-bootleg-2020	538	4	of	of	IN
orr-bootleg-2020	538	5	deep	deep	JJ
orr-bootleg-2020	538	6	contextualized	contextualized	JJ
orr-bootleg-2020	538	7	embeddings	embedding	NNS
orr-bootleg-2020	538	8	of	of	IN
orr-bootleg-2020	538	9	words	word	NNS
orr-bootleg-2020	538	10	and	and	CC
orr-bootleg-2020	538	11	entities	entity	NNS
orr-bootleg-2020	538	12	for	for	IN
orr-bootleg-2020	538	13	named	name	VBN
orr-bootleg-2020	538	14	entity	entity	NN
orr-bootleg-2020	538	15	disambiguation	disambiguation	NN
orr-bootleg-2020	538	16	.	.	.
orr-bootleg-2020	539	1	arXiv	arXiv	NNP
orr-bootleg-2020	539	2	preprint	preprint	NNP
orr-bootleg-2020	539	3	arXiv:1909.00426	arXiv:1909.00426	NNP
orr-bootleg-2020	539	4	,	,	,
orr-bootleg-2020	539	5	2019	2019	CD
orr-bootleg-2020	539	6	.	.	.
orr-bootleg-2020	540	1	[	[	-LRB-
orr-bootleg-2020	540	2	52	52	CD
orr-bootleg-2020	540	3	]	]	-RRB-
orr-bootleg-2020	540	4	Mohamed	Mohamed	NNP
orr-bootleg-2020	540	5	Amir	Amir	NNP
orr-bootleg-2020	540	6	Yosef	Yosef	NNP
orr-bootleg-2020	540	7	,	,	,
orr-bootleg-2020	540	8	Sandro	Sandro	NNP
orr-bootleg-2020	540	9	Bauer	Bauer	NNP
orr-bootleg-2020	540	10	,	,	,
orr-bootleg-2020	540	11	Johannes	Johannes	NNP
orr-bootleg-2020	540	12	Hoffart	Hoffart	NNP
orr-bootleg-2020	540	13	,	,	,
orr-bootleg-2020	540	14	Marc	Marc	NNP
orr-bootleg-2020	540	15	Spaniol	Spaniol	NNP
orr-bootleg-2020	540	16	,	,	,
orr-bootleg-2020	540	17	and	and	CC
orr-bootleg-2020	540	18	Gerhard	Gerhard	NNP
orr-bootleg-2020	540	19	Weikum	Weikum	NNP
orr-bootleg-2020	540	20	.	.	.
orr-bootleg-2020	541	1	HYENA	hyena	NN
orr-bootleg-2020	541	2	:	:	:
orr-bootleg-2020	541	3	Hierarchical	hierarchical	JJ
orr-bootleg-2020	541	4	type	type	NN
orr-bootleg-2020	541	5	classification	classification	NN
orr-bootleg-2020	541	6	for	for	IN
orr-bootleg-2020	541	7	entity	entity	NN
orr-bootleg-2020	541	8	names	name	NNS
orr-bootleg-2020	541	9	.	.	.
orr-bootleg-2020	542	1	In	in	IN
orr-bootleg-2020	542	2	Proceedings	Proceedings	NNP
orr-bootleg-2020	542	3	of	of	IN
orr-bootleg-2020	542	4	COLING	COLING	NNP
orr-bootleg-2020	542	5	2012	2012	CD
orr-bootleg-2020	542	6	:	:	:
orr-bootleg-2020	542	7	Posters	poster	NNS
orr-bootleg-2020	542	8	,	,	,
orr-bootleg-2020	542	9	pages	page	NNS
orr-bootleg-2020	542	10	1361–1370	1361–1370	CD
orr-bootleg-2020	542	11	,	,	,
orr-bootleg-2020	542	12	Mumbai	Mumbai	NNP
orr-bootleg-2020	542	13	,	,	,
orr-bootleg-2020	542	14	India	India	NNP
orr-bootleg-2020	542	15	,	,	,
orr-bootleg-2020	542	16	December	December	NNP
orr-bootleg-2020	542	17	2012	2012	CD
orr-bootleg-2020	542	18	.	.	.
orr-bootleg-2020	543	1	The	the	DT
orr-bootleg-2020	543	2	COLING	coling	NN
orr-bootleg-2020	543	3	2012	2012	CD
orr-bootleg-2020	543	4	Organizing	Organizing	NNP
orr-bootleg-2020	543	5	Committee	Committee	NNP
orr-bootleg-2020	543	6	.	.	.
orr-bootleg-2020	544	1	[	[	-LRB-
orr-bootleg-2020	544	2	53	53	CD
orr-bootleg-2020	544	3	]	]	-RRB-
orr-bootleg-2020	544	4	Yuhao	Yuhao	NNP
orr-bootleg-2020	544	5	Zhang	Zhang	NNP
orr-bootleg-2020	544	6	,	,	,
orr-bootleg-2020	544	7	Victor	Victor	NNP
orr-bootleg-2020	544	8	Zhong	Zhong	NNP
orr-bootleg-2020	544	9	,	,	,
orr-bootleg-2020	544	10	Danqi	Danqi	NNP
orr-bootleg-2020	544	11	Chen	Chen	NNP
orr-bootleg-2020	544	12	,	,	,
orr-bootleg-2020	544	13	Gabor	Gabor	NNP
orr-bootleg-2020	544	14	Angeli	Angeli	NNP
orr-bootleg-2020	544	15	,	,	,
orr-bootleg-2020	544	16	and	and	CC
orr-bootleg-2020	544	17	Christopher	Christopher	NNP
orr-bootleg-2020	544	18	D.	D.	NNP
orr-bootleg-2020	544	19	Manning	Manning	NNP
orr-bootleg-2020	544	20	.	.	.
orr-bootleg-2020	545	1	Position	position	NN
orr-bootleg-2020	545	2	-	-	HYPH
orr-bootleg-2020	545	3	aware	aware	JJ
orr-bootleg-2020	545	4	attention	attention	NN
orr-bootleg-2020	545	5	and	and	CC
orr-bootleg-2020	545	6	supervised	supervise	VBN
orr-bootleg-2020	545	7	data	datum	NNS
orr-bootleg-2020	545	8	improve	improve	VBP
orr-bootleg-2020	545	9	slot	slot	NN
orr-bootleg-2020	545	10	filling	filling	NN
orr-bootleg-2020	545	11	.	.	.
orr-bootleg-2020	546	1	In	in	IN
orr-bootleg-2020	546	2	Proceedings	proceeding	NNS
orr-bootleg-2020	546	3	of	of	IN
orr-bootleg-2020	546	4	the	the	DT
orr-bootleg-2020	546	5	2017	2017	CD
orr-bootleg-2020	546	6	Conference	Conference	NNP
orr-bootleg-2020	546	7	on	on	IN
orr-bootleg-2020	546	8	Empirical	Empirical	NNP
orr-bootleg-2020	546	9	Methods	Methods	NNPS
orr-bootleg-2020	546	10	in	in	IN
orr-bootleg-2020	546	11	Natural	Natural	NNP
orr-bootleg-2020	546	12	Language	Language	NNP
orr-bootleg-2020	546	13	Processing	Processing	NNP
orr-bootleg-2020	546	14	(	(	-LRB-
orr-bootleg-2020	546	15	EMNLP	EMNLP	NNP
orr-bootleg-2020	546	16	2017	2017	CD
orr-bootleg-2020	546	17	)	)	-RRB-
orr-bootleg-2020	546	18	,	,	,
orr-bootleg-2020	546	19	pages	page	VBZ
orr-bootleg-2020	546	20	35–45	35–45	CD
orr-bootleg-2020	546	21	,	,	,
orr-bootleg-2020	546	22	2017	2017	CD
orr-bootleg-2020	546	23	.	.	.
orr-bootleg-2020	547	1	[	[	-LRB-
orr-bootleg-2020	547	2	54	54	CD
orr-bootleg-2020	547	3	]	]	-RRB-
orr-bootleg-2020	547	4	Zhengyan	Zhengyan	NNP
orr-bootleg-2020	547	5	Zhang	Zhang	NNP
orr-bootleg-2020	547	6	,	,	,
orr-bootleg-2020	547	7	Xu	Xu	NNP
orr-bootleg-2020	547	8	Han	Han	NNP
orr-bootleg-2020	547	9	,	,	,
orr-bootleg-2020	547	10	Zhiyuan	Zhiyuan	NNP
orr-bootleg-2020	547	11	Liu	Liu	NNP
orr-bootleg-2020	547	12	,	,	,
orr-bootleg-2020	547	13	Xin	Xin	NNP
orr-bootleg-2020	547	14	Jiang	Jiang	NNP
orr-bootleg-2020	547	15	,	,	,
orr-bootleg-2020	547	16	Maosong	Maosong	NNP
orr-bootleg-2020	547	17	Sun	Sun	NNP
orr-bootleg-2020	547	18	,	,	,
orr-bootleg-2020	547	19	and	and	CC
orr-bootleg-2020	547	20	Qun	Qun	NNP
orr-bootleg-2020	547	21	Liu	Liu	NNP
orr-bootleg-2020	547	22	.	.	.
orr-bootleg-2020	548	1	Ernie	Ernie	NNP
orr-bootleg-2020	548	2	:	:	:
orr-bootleg-2020	548	3	Enhanced	enhanced	JJ
orr-bootleg-2020	548	4	language	language	NN
orr-bootleg-2020	548	5	representation	representation	NN
orr-bootleg-2020	548	6	with	with	IN
orr-bootleg-2020	548	7	informative	informative	JJ
orr-bootleg-2020	548	8	entities	entity	NNS
orr-bootleg-2020	548	9	.	.	.
orr-bootleg-2020	549	1	arXiv	arXiv	NNP
orr-bootleg-2020	549	2	preprint	preprint	NNP
orr-bootleg-2020	549	3	arXiv:1905.07129	arXiv:1905.07129	NNP
orr-bootleg-2020	549	4	,	,	,
orr-bootleg-2020	549	5	2019	2019	CD
orr-bootleg-2020	549	6	.	.	.
orr-bootleg-2020	550	1	[	[	-LRB-
orr-bootleg-2020	550	2	55	55	CD
orr-bootleg-2020	550	3	]	]	-RRB-
orr-bootleg-2020	550	4	Ming	Ming	NNP
orr-bootleg-2020	550	5	Zhu	Zhu	NNP
orr-bootleg-2020	550	6	,	,	,
orr-bootleg-2020	550	7	Busra	Busra	NNP
orr-bootleg-2020	550	8	Celikkaya	Celikkaya	NNP
orr-bootleg-2020	550	9	,	,	,
orr-bootleg-2020	550	10	Parminder	Parminder	NNP
orr-bootleg-2020	550	11	Bhatia	Bhatia	NNP
orr-bootleg-2020	550	12	,	,	,
orr-bootleg-2020	550	13	and	and	CC
orr-bootleg-2020	550	14	Chandan	Chandan	NNP
orr-bootleg-2020	550	15	K.	K.	NNP
orr-bootleg-2020	550	16	Reddy	Reddy	NNP
orr-bootleg-2020	550	17	.	.	.
orr-bootleg-2020	551	1	Latte	latte	NN
orr-bootleg-2020	551	2	:	:	:
orr-bootleg-2020	551	3	Latent	latent	NN
orr-bootleg-2020	551	4	type	type	NN
orr-bootleg-2020	551	5	modeling	modeling	NN
orr-bootleg-2020	551	6	for	for	IN
orr-bootleg-2020	551	7	biomedical	biomedical	JJ
orr-bootleg-2020	551	8	entity	entity	NN
orr-bootleg-2020	551	9	linking	linking	NN
orr-bootleg-2020	551	10	.	.	.
orr-bootleg-2020	552	1	In	in	IN
orr-bootleg-2020	552	2	AAAI	AAAI	NNP
orr-bootleg-2020	552	3	,	,	,
orr-bootleg-2020	552	4	2020	2020	CD
orr-bootleg-2020	552	5	.	.	.
orr-bootleg-2020	553	1	[	[	-LRB-
orr-bootleg-2020	553	2	56	56	CD
orr-bootleg-2020	553	3	]	]	-RRB-
orr-bootleg-2020	553	4	Yuke	Yuke	NNP
orr-bootleg-2020	553	5	Zhu	Zhu	NNP
orr-bootleg-2020	553	6	,	,	,
orr-bootleg-2020	553	7	Joseph	Joseph	NNP
orr-bootleg-2020	553	8	J	J	NNP
orr-bootleg-2020	553	9	Lim	Lim	NNP
orr-bootleg-2020	553	10	,	,	,
orr-bootleg-2020	553	11	and	and	CC
orr-bootleg-2020	553	12	Li	Li	NNP
orr-bootleg-2020	553	13	Fei	Fei	NNP
orr-bootleg-2020	553	14	-	-	HYPH
orr-bootleg-2020	553	15	Fei	Fei	NNP
orr-bootleg-2020	553	16	.	.	.
orr-bootleg-2020	554	1	Knowledge	knowledge	NN
orr-bootleg-2020	554	2	acquisition	acquisition	NN
orr-bootleg-2020	554	3	for	for	IN
orr-bootleg-2020	554	4	visual	visual	JJ
orr-bootleg-2020	554	5	question	question	NN
orr-bootleg-2020	554	6	answering	answer	VBG
orr-bootleg-2020	554	7	via	via	IN
orr-bootleg-2020	554	8	iterative	iterative	JJ
orr-bootleg-2020	554	9	querying	querying	NN
orr-bootleg-2020	554	10	.	.	.
orr-bootleg-2020	555	1	In	in	IN
orr-bootleg-2020	555	2	Proceedings	proceeding	NNS
orr-bootleg-2020	555	3	of	of	IN
orr-bootleg-2020	555	4	the	the	DT
orr-bootleg-2020	555	5	IEEE	IEEE	NNP
orr-bootleg-2020	555	6	Conference	Conference	NNP
orr-bootleg-2020	555	7	on	on	IN
orr-bootleg-2020	555	8	Computer	Computer	NNP
orr-bootleg-2020	555	9	Vision	Vision	NNP
orr-bootleg-2020	555	10	and	and	CC
orr-bootleg-2020	555	11	Pattern	Pattern	NNP
orr-bootleg-2020	555	12	Recognition	Recognition	NNP
orr-bootleg-2020	555	13	,	,	,
orr-bootleg-2020	555	14	pages	page	NNS
orr-bootleg-2020	555	15	1154–1163	1154–1163	CD
orr-bootleg-2020	555	16	,	,	,
orr-bootleg-2020	555	17	2017	2017	CD
orr-bootleg-2020	555	18	.	.	.
orr-bootleg-2020	556	1	18	18	CD
orr-bootleg-2020	556	2	A	a	DT
orr-bootleg-2020	556	3	Extended	Extended	NNP
orr-bootleg-2020	556	4	Model	Model	NNP
orr-bootleg-2020	556	5	Details	detail	NNS
orr-bootleg-2020	556	6	We	-PRON-	PRP
orr-bootleg-2020	556	7	now	now	RB
orr-bootleg-2020	556	8	provide	provide	VBP
orr-bootleg-2020	556	9	additional	additional	JJ
orr-bootleg-2020	556	10	details	detail	NNS
orr-bootleg-2020	556	11	about	about	IN
orr-bootleg-2020	556	12	the	the	DT
orr-bootleg-2020	556	13	model	model	NN
orr-bootleg-2020	556	14	introduced	introduce	VBN
orr-bootleg-2020	556	15	in	in	IN
orr-bootleg-2020	556	16	Section	section	NN
orr-bootleg-2020	556	17	3	3	CD
orr-bootleg-2020	556	18	.	.	.
orr-bootleg-2020	557	1	We	-PRON-	PRP
orr-bootleg-2020	557	2	first	first	RB
orr-bootleg-2020	557	3	describe	describe	VBP
orr-bootleg-2020	557	4	our	-PRON-	PRP$
orr-bootleg-2020	557	5	type	type	NN
orr-bootleg-2020	557	6	prediction	prediction	NN
orr-bootleg-2020	557	7	module	module	NN
orr-bootleg-2020	557	8	and	and	CC
orr-bootleg-2020	557	9	then	then	RB
orr-bootleg-2020	557	10	describe	describe	VB
orr-bootleg-2020	557	11	the	the	DT
orr-bootleg-2020	557	12	added	add	VBN
orr-bootleg-2020	557	13	entity	entity	NN
orr-bootleg-2020	557	14	positional	positional	JJ
orr-bootleg-2020	557	15	encoding	encoding	NN
orr-bootleg-2020	557	16	.	.	.
orr-bootleg-2020	558	1	Type	Type	NNP
orr-bootleg-2020	558	2	Prediction	Prediction	NNP
orr-bootleg-2020	558	3	To	to	TO
orr-bootleg-2020	558	4	allow	allow	VB
orr-bootleg-2020	558	5	the	the	DT
orr-bootleg-2020	558	6	model	model	NN
orr-bootleg-2020	558	7	to	to	TO
orr-bootleg-2020	558	8	further	further	RB
orr-bootleg-2020	558	9	infer	infer	VB
orr-bootleg-2020	558	10	the	the	DT
orr-bootleg-2020	558	11	correct	correct	JJ
orr-bootleg-2020	558	12	types	type	NNS
orr-bootleg-2020	558	13	for	for	IN
orr-bootleg-2020	558	14	an	an	DT
orr-bootleg-2020	558	15	entity	entity	NN
orr-bootleg-2020	558	16	,	,	,
orr-bootleg-2020	558	17	especially	especially	RB
orr-bootleg-2020	558	18	when	when	WRB
orr-bootleg-2020	558	19	the	the	DT
orr-bootleg-2020	558	20	entity	entity	NN
orr-bootleg-2020	558	21	does	do	VBZ
orr-bootleg-2020	558	22	not	not	RB
orr-bootleg-2020	558	23	have	have	VB
orr-bootleg-2020	558	24	a	a	DT
orr-bootleg-2020	558	25	preassigned	preassigne	VBN
orr-bootleg-2020	558	26	type	type	NN
orr-bootleg-2020	558	27	,	,	,
orr-bootleg-2020	558	28	we	-PRON-	PRP
orr-bootleg-2020	558	29	add	add	VBP
orr-bootleg-2020	558	30	a	a	DT
orr-bootleg-2020	558	31	coarse	coarse	JJ
orr-bootleg-2020	558	32	mention	mention	NN
orr-bootleg-2020	558	33	type	type	NN
orr-bootleg-2020	558	34	prediction	prediction	NN
orr-bootleg-2020	558	35	task	task	NN
orr-bootleg-2020	558	36	given	give	VBN
orr-bootleg-2020	558	37	the	the	DT
orr-bootleg-2020	558	38	mention	mention	NN
orr-bootleg-2020	558	39	embedding	embed	VBG
orr-bootleg-2020	558	40	.	.	.
orr-bootleg-2020	559	1	Given	give	VBN
orr-bootleg-2020	559	2	a	a	DT
orr-bootleg-2020	559	3	mention	mention	NN
orr-bootleg-2020	559	4	m	m	NN
orr-bootleg-2020	559	5	and	and	CC
orr-bootleg-2020	559	6	a	a	DT
orr-bootleg-2020	559	7	coarse	coarse	JJ
orr-bootleg-2020	559	8	type	type	NN
orr-bootleg-2020	559	9	embedding	embed	VBG
orr-bootleg-2020	559	10	matrix	matrix	NN
orr-bootleg-2020	559	11	T	T	NNP
orr-bootleg-2020	559	12	,	,	,
orr-bootleg-2020	559	13	the	the	DT
orr-bootleg-2020	559	14	task	task	NN
orr-bootleg-2020	559	15	is	be	VBZ
orr-bootleg-2020	559	16	to	to	TO
orr-bootleg-2020	559	17	assign	assign	VB
orr-bootleg-2020	559	18	a	a	DT
orr-bootleg-2020	559	19	coarse	coarse	JJ
orr-bootleg-2020	559	20	type	type	NN
orr-bootleg-2020	559	21	embedding	embedding	NN
orr-bootleg-2020	559	22	for	for	IN
orr-bootleg-2020	559	23	the	the	DT
orr-bootleg-2020	559	24	mention	mention	NN
orr-bootleg-2020	559	25	m	m	NNP
orr-bootleg-2020	559	26	;	;	:
orr-bootleg-2020	559	27	i.e.	i.e.	FW
orr-bootleg-2020	559	28	,	,	,
orr-bootleg-2020	559	29	determine	determine	VB
orr-bootleg-2020	559	30	tm	tm	NNP
orr-bootleg-2020	559	31	.	.	.
orr-bootleg-2020	560	1	We	-PRON-	PRP
orr-bootleg-2020	560	2	do	do	VBP
orr-bootleg-2020	560	3	so	so	RB
orr-bootleg-2020	560	4	by	by	IN
orr-bootleg-2020	560	5	adding	add	VBG
orr-bootleg-2020	560	6	the	the	DT
orr-bootleg-2020	560	7	first	first	JJ
orr-bootleg-2020	560	8	and	and	CC
orr-bootleg-2020	560	9	last	last	JJ
orr-bootleg-2020	560	10	token	token	NN
orr-bootleg-2020	560	11	of	of	IN
orr-bootleg-2020	560	12	the	the	DT
orr-bootleg-2020	560	13	mention	mention	NN
orr-bootleg-2020	560	14	from	from	IN
orr-bootleg-2020	560	15	W	W	NNP
orr-bootleg-2020	560	16	to	to	TO
orr-bootleg-2020	560	17	generate	generate	VB
orr-bootleg-2020	560	18	a	a	DT
orr-bootleg-2020	560	19	contextualized	contextualized	JJ
orr-bootleg-2020	560	20	mention	mention	NN
orr-bootleg-2020	560	21	embedding	embed	VBG
orr-bootleg-2020	560	22	m.	m.	NN
orr-bootleg-2020	560	23	We	-PRON-	PRP
orr-bootleg-2020	560	24	predict	predict	VBP
orr-bootleg-2020	560	25	the	the	DT
orr-bootleg-2020	560	26	coarse	coarse	JJ
orr-bootleg-2020	560	27	type	type	NN
orr-bootleg-2020	560	28	of	of	IN
orr-bootleg-2020	560	29	the	the	DT
orr-bootleg-2020	560	30	mention	mention	NN
orr-bootleg-2020	560	31	t̂m	t̂m	NN
orr-bootleg-2020	560	32	by	by	IN
orr-bootleg-2020	560	33	computing	compute	VBG
orr-bootleg-2020	560	34	Stype	stype	NN
orr-bootleg-2020	560	35	=	=	SYM
orr-bootleg-2020	560	36	softmax(MLP(m	softmax(mlp(m	CD
orr-bootleg-2020	560	37	)	)	-RRB-
orr-bootleg-2020	560	38	)	)	-RRB-
orr-bootleg-2020	560	39	t̂m	t̂m	NNP
orr-bootleg-2020	560	40	=	=	SYM
orr-bootleg-2020	560	41	StypeT	StypeT	NNS
orr-bootleg-2020	560	42	where	where	WRB
orr-bootleg-2020	560	43	Stype	Stype	NNP
orr-bootleg-2020	560	44	generates	generate	VBZ
orr-bootleg-2020	560	45	a	a	DT
orr-bootleg-2020	560	46	distribution	distribution	NN
orr-bootleg-2020	560	47	over	over	IN
orr-bootleg-2020	560	48	coarse	coarse	JJ
orr-bootleg-2020	560	49	types	type	NNS
orr-bootleg-2020	560	50	.	.	.
orr-bootleg-2020	561	1	For	for	IN
orr-bootleg-2020	561	2	each	each	DT
orr-bootleg-2020	561	3	entity	entity	NN
orr-bootleg-2020	561	4	candidate	candidate	NN
orr-bootleg-2020	561	5	of	of	IN
orr-bootleg-2020	561	6	m	m	NN
orr-bootleg-2020	561	7	,	,	,
orr-bootleg-2020	561	8	t̂m	t̂m	NNP
orr-bootleg-2020	561	9	gets	get	VBZ
orr-bootleg-2020	561	10	concatenated	concatenate	VBN
orr-bootleg-2020	561	11	to	to	IN
orr-bootleg-2020	561	12	the	the	DT
orr-bootleg-2020	561	13	other	other	JJ
orr-bootleg-2020	561	14	type	type	NN
orr-bootleg-2020	561	15	embedding	embed	VBG
orr-bootleg-2020	561	16	te	te	NNP
orr-bootleg-2020	561	17	before	before	IN
orr-bootleg-2020	561	18	the	the	DT
orr-bootleg-2020	561	19	MLP	MLP	NNP
orr-bootleg-2020	561	20	.	.	.
orr-bootleg-2020	562	1	This	this	DT
orr-bootleg-2020	562	2	is	be	VBZ
orr-bootleg-2020	562	3	supervised	supervise	VBN
orr-bootleg-2020	562	4	by	by	IN
orr-bootleg-2020	562	5	minimizing	minimize	VBG
orr-bootleg-2020	562	6	the	the	DT
orr-bootleg-2020	562	7	cross	cross	NN
orr-bootleg-2020	562	8	entropy	entropy	JJ
orr-bootleg-2020	562	9	between	between	IN
orr-bootleg-2020	562	10	Stype	Stype	NNP
orr-bootleg-2020	562	11	and	and	CC
orr-bootleg-2020	562	12	the	the	DT
orr-bootleg-2020	562	13	true	true	JJ
orr-bootleg-2020	562	14	coarse	coarse	JJ
orr-bootleg-2020	562	15	type	type	NN
orr-bootleg-2020	562	16	for	for	IN
orr-bootleg-2020	562	17	the	the	DT
orr-bootleg-2020	562	18	gold	gold	JJ
orr-bootleg-2020	562	19	entity	entity	NN
orr-bootleg-2020	562	20	,	,	,
orr-bootleg-2020	562	21	generating	generate	VBG
orr-bootleg-2020	562	22	a	a	DT
orr-bootleg-2020	562	23	type	type	NN
orr-bootleg-2020	562	24	prediction	prediction	NN
orr-bootleg-2020	562	25	loss	loss	NN
orr-bootleg-2020	562	26	Ltype	Ltype	NNP
orr-bootleg-2020	562	27	.	.	.
orr-bootleg-2020	563	1	When	when	WRB
orr-bootleg-2020	563	2	performing	perform	VBG
orr-bootleg-2020	563	3	type	type	NN
orr-bootleg-2020	563	4	prediction	prediction	NN
orr-bootleg-2020	563	5	,	,	,
orr-bootleg-2020	563	6	our	-PRON-	PRP$
orr-bootleg-2020	563	7	overall	overall	JJ
orr-bootleg-2020	563	8	loss	loss	NN
orr-bootleg-2020	563	9	is	be	VBZ
orr-bootleg-2020	563	10	Ldis	Ldis	NNP
orr-bootleg-2020	563	11	+	+	SYM
orr-bootleg-2020	563	12	Ltype	Ltype	NNP
orr-bootleg-2020	563	13	.	.	.
orr-bootleg-2020	564	1	Position	position	NN
orr-bootleg-2020	564	2	Encoding	encode	VBG
orr-bootleg-2020	564	3	We	-PRON-	PRP
orr-bootleg-2020	564	4	need	need	VBP
orr-bootleg-2020	564	5	Bootleg	Bootleg	NNP
orr-bootleg-2020	564	6	to	to	TO
orr-bootleg-2020	564	7	be	be	VB
orr-bootleg-2020	564	8	able	able	JJ
orr-bootleg-2020	564	9	to	to	TO
orr-bootleg-2020	564	10	reason	reason	VB
orr-bootleg-2020	564	11	over	over	RP
orr-bootleg-2020	564	12	absolute	absolute	JJ
orr-bootleg-2020	564	13	and	and	CC
orr-bootleg-2020	564	14	relative	relative	JJ
orr-bootleg-2020	564	15	positioning	positioning	NN
orr-bootleg-2020	564	16	of	of	IN
orr-bootleg-2020	564	17	the	the	DT
orr-bootleg-2020	564	18	words	word	NNS
orr-bootleg-2020	564	19	in	in	IN
orr-bootleg-2020	564	20	the	the	DT
orr-bootleg-2020	564	21	sentence	sentence	NN
orr-bootleg-2020	564	22	and	and	CC
orr-bootleg-2020	564	23	the	the	DT
orr-bootleg-2020	564	24	mentions	mention	NNS
orr-bootleg-2020	564	25	.	.	.
orr-bootleg-2020	565	1	For	for	IN
orr-bootleg-2020	565	2	example	example	NN
orr-bootleg-2020	565	3	,	,	,
orr-bootleg-2020	565	4	in	in	IN
orr-bootleg-2020	565	5	the	the	DT
orr-bootleg-2020	565	6	sentence	sentence	NN
orr-bootleg-2020	565	7	“	"	``
orr-bootleg-2020	565	8	Where	where	WRB
orr-bootleg-2020	565	9	is	be	VBZ
orr-bootleg-2020	565	10	America	America	NNP
orr-bootleg-2020	565	11	in	in	IN
orr-bootleg-2020	565	12	Indiana	Indiana	NNP
orr-bootleg-2020	565	13	?	?	.
orr-bootleg-2020	565	14	”	"	''
orr-bootleg-2020	565	15	,	,	,
orr-bootleg-2020	565	16	“	"	``
orr-bootleg-2020	565	17	America	America	NNP
orr-bootleg-2020	565	18	”	"	''
orr-bootleg-2020	565	19	refers	refer	VBZ
orr-bootleg-2020	565	20	to	to	IN
orr-bootleg-2020	565	21	the	the	DT
orr-bootleg-2020	565	22	city	city	NN
orr-bootleg-2020	565	23	in	in	IN
orr-bootleg-2020	565	24	Indiana	Indiana	NNP
orr-bootleg-2020	565	25	,	,	,
orr-bootleg-2020	565	26	not	not	RB
orr-bootleg-2020	565	27	the	the	DT
orr-bootleg-2020	565	28	United	United	NNP
orr-bootleg-2020	565	29	States	States	NNP
orr-bootleg-2020	565	30	.	.	.
orr-bootleg-2020	566	1	In	in	IN
orr-bootleg-2020	566	2	the	the	DT
orr-bootleg-2020	566	3	sentence	sentence	NN
orr-bootleg-2020	566	4	“	"	``
orr-bootleg-2020	566	5	Where	where	WRB
orr-bootleg-2020	566	6	is	be	VBZ
orr-bootleg-2020	566	7	Indiana	Indiana	NNP
orr-bootleg-2020	566	8	in	in	IN
orr-bootleg-2020	566	9	America	America	NNP
orr-bootleg-2020	566	10	?	?	.
orr-bootleg-2020	566	11	”	"	''
orr-bootleg-2020	566	12	,	,	,
orr-bootleg-2020	566	13	“	"	``
orr-bootleg-2020	566	14	America	America	NNP
orr-bootleg-2020	566	15	”	"	''
orr-bootleg-2020	566	16	refers	refer	VBZ
orr-bootleg-2020	566	17	to	to	IN
orr-bootleg-2020	566	18	the	the	DT
orr-bootleg-2020	566	19	United	United	NNP
orr-bootleg-2020	566	20	States	States	NNP
orr-bootleg-2020	566	21	.	.	.
orr-bootleg-2020	567	1	The	the	DT
orr-bootleg-2020	567	2	relative	relative	JJ
orr-bootleg-2020	567	3	position	position	NN
orr-bootleg-2020	567	4	of	of	IN
orr-bootleg-2020	567	5	“	"	``
orr-bootleg-2020	567	6	Indiana	Indiana	NNP
orr-bootleg-2020	567	7	”	"	''
orr-bootleg-2020	567	8	,	,	,
orr-bootleg-2020	567	9	“	"	``
orr-bootleg-2020	567	10	in	in	IN
orr-bootleg-2020	567	11	”	"	''
orr-bootleg-2020	567	12	,	,	,
orr-bootleg-2020	567	13	and	and	CC
orr-bootleg-2020	567	14	“	"	``
orr-bootleg-2020	567	15	America	America	NNP
orr-bootleg-2020	567	16	”	"	''
orr-bootleg-2020	567	17	signals	signal	VBZ
orr-bootleg-2020	567	18	the	the	DT
orr-bootleg-2020	567	19	correct	correct	JJ
orr-bootleg-2020	567	20	answer	answer	NN
orr-bootleg-2020	567	21	.	.	.
orr-bootleg-2020	568	1	To	to	TO
orr-bootleg-2020	568	2	achieve	achieve	VB
orr-bootleg-2020	568	3	this	this	DT
orr-bootleg-2020	568	4	signaling	signaling	NN
orr-bootleg-2020	568	5	,	,	,
orr-bootleg-2020	568	6	we	-PRON-	PRP
orr-bootleg-2020	568	7	add	add	VBP
orr-bootleg-2020	568	8	the	the	DT
orr-bootleg-2020	568	9	sin	sin	NN
orr-bootleg-2020	568	10	positional	positional	JJ
orr-bootleg-2020	568	11	encoding	encoding	NN
orr-bootleg-2020	568	12	from	from	IN
orr-bootleg-2020	568	13	Vaswani	Vaswani	NNP
orr-bootleg-2020	568	14	et	et	FW
orr-bootleg-2020	568	15	al	al	NNP
orr-bootleg-2020	568	16	.	.	.
orr-bootleg-2020	569	1	[	[	-LRB-
orr-bootleg-2020	569	2	48	48	CD
orr-bootleg-2020	569	3	]	]	-RRB-
orr-bootleg-2020	569	4	to	to	IN
orr-bootleg-2020	569	5	E	e	NN
orr-bootleg-2020	569	6	before	before	IN
orr-bootleg-2020	569	7	it	-PRON-	PRP
orr-bootleg-2020	569	8	is	be	VBZ
orr-bootleg-2020	569	9	passed	pass	VBN
orr-bootleg-2020	569	10	to	to	IN
orr-bootleg-2020	569	11	our	-PRON-	PRP$
orr-bootleg-2020	569	12	neural	neural	JJ
orr-bootleg-2020	569	13	model	model	NN
orr-bootleg-2020	569	14	.	.	.
orr-bootleg-2020	570	1	Specifically	specifically	RB
orr-bootleg-2020	570	2	,	,	,
orr-bootleg-2020	570	3	for	for	IN
orr-bootleg-2020	570	4	mention	mention	NN
orr-bootleg-2020	570	5	m	m	NNP
orr-bootleg-2020	570	6	,	,	,
orr-bootleg-2020	570	7	we	-PRON-	PRP
orr-bootleg-2020	570	8	concatenate	concatenate	VBP
orr-bootleg-2020	570	9	of	of	IN
orr-bootleg-2020	570	10	the	the	DT
orr-bootleg-2020	570	11	positional	positional	JJ
orr-bootleg-2020	570	12	encoding	encoding	NN
orr-bootleg-2020	570	13	of	of	IN
orr-bootleg-2020	570	14	the	the	DT
orr-bootleg-2020	570	15	first	first	JJ
orr-bootleg-2020	570	16	and	and	CC
orr-bootleg-2020	570	17	last	last	JJ
orr-bootleg-2020	570	18	token	token	NN
orr-bootleg-2020	570	19	of	of	IN
orr-bootleg-2020	570	20	m	m	NN
orr-bootleg-2020	570	21	,	,	,
orr-bootleg-2020	570	22	project	project	VB
orr-bootleg-2020	570	23	the	the	DT
orr-bootleg-2020	570	24	concatenation	concatenation	NN
orr-bootleg-2020	570	25	to	to	IN
orr-bootleg-2020	570	26	dimension	dimension	NN
orr-bootleg-2020	570	27	H	h	NN
orr-bootleg-2020	570	28	,	,	,
orr-bootleg-2020	570	29	and	and	CC
orr-bootleg-2020	570	30	add	add	VB
orr-bootleg-2020	570	31	it	-PRON-	PRP
orr-bootleg-2020	570	32	to	to	IN
orr-bootleg-2020	570	33	each	each	DT
orr-bootleg-2020	570	34	of	of	IN
orr-bootleg-2020	570	35	m	m	NN
orr-bootleg-2020	570	36	’s	’s	NN
orr-bootleg-2020	570	37	K	K	NNP
orr-bootleg-2020	570	38	candidates	candidate	NNS
orr-bootleg-2020	570	39	in	in	IN
orr-bootleg-2020	570	40	E.	E.	NNP
orr-bootleg-2020	570	41	As	as	IN
orr-bootleg-2020	570	42	we	-PRON-	PRP
orr-bootleg-2020	570	43	use	use	VBP
orr-bootleg-2020	570	44	BERT	BERT	NNP
orr-bootleg-2020	570	45	word	word	NN
orr-bootleg-2020	570	46	embeddings	embedding	NNS
orr-bootleg-2020	570	47	for	for	IN
orr-bootleg-2020	570	48	W	W	NNP
orr-bootleg-2020	570	49	,	,	,
orr-bootleg-2020	570	50	the	the	DT
orr-bootleg-2020	570	51	positional	positional	JJ
orr-bootleg-2020	570	52	encoding	encoding	NN
orr-bootleg-2020	570	53	is	be	VBZ
orr-bootleg-2020	570	54	already	already	RB
orr-bootleg-2020	570	55	added	add	VBN
orr-bootleg-2020	570	56	to	to	IN
orr-bootleg-2020	570	57	words	word	NNS
orr-bootleg-2020	570	58	in	in	IN
orr-bootleg-2020	570	59	the	the	DT
orr-bootleg-2020	570	60	sentence	sentence	NN
orr-bootleg-2020	570	61	.	.	.
orr-bootleg-2020	571	1	B	B	NNP
orr-bootleg-2020	571	2	Extended	extended	JJ
orr-bootleg-2020	571	3	Results	result	NNS
orr-bootleg-2020	571	4	We	-PRON-	PRP
orr-bootleg-2020	571	5	now	now	RB
orr-bootleg-2020	571	6	give	give	VBP
orr-bootleg-2020	571	7	the	the	DT
orr-bootleg-2020	571	8	details	detail	NNS
orr-bootleg-2020	571	9	of	of	IN
orr-bootleg-2020	571	10	our	-PRON-	PRP$
orr-bootleg-2020	571	11	experimental	experimental	JJ
orr-bootleg-2020	571	12	setup	setup	NN
orr-bootleg-2020	571	13	and	and	CC
orr-bootleg-2020	571	14	training	training	NN
orr-bootleg-2020	571	15	.	.	.
orr-bootleg-2020	572	1	We	-PRON-	PRP
orr-bootleg-2020	572	2	then	then	RB
orr-bootleg-2020	572	3	give	give	VBP
orr-bootleg-2020	572	4	extended	extended	JJ
orr-bootleg-2020	572	5	results	result	NNS
orr-bootleg-2020	572	6	over	over	IN
orr-bootleg-2020	572	7	the	the	DT
orr-bootleg-2020	572	8	regularization	regularization	NN
orr-bootleg-2020	572	9	scheme	scheme	NN
orr-bootleg-2020	572	10	and	and	CC
orr-bootleg-2020	572	11	model	model	NN
orr-bootleg-2020	572	12	ablations	ablation	NNS
orr-bootleg-2020	572	13	.	.	.
orr-bootleg-2020	573	1	Lastly	lastly	RB
orr-bootleg-2020	573	2	,	,	,
orr-bootleg-2020	573	3	we	-PRON-	PRP
orr-bootleg-2020	573	4	extend	extend	VBP
orr-bootleg-2020	573	5	our	-PRON-	PRP$
orr-bootleg-2020	573	6	error	error	NN
orr-bootleg-2020	573	7	analysis	analysis	NN
orr-bootleg-2020	573	8	to	to	TO
orr-bootleg-2020	573	9	validate	validate	VB
orr-bootleg-2020	573	10	Bootleg	Bootleg	NNP
orr-bootleg-2020	573	11	’s	’s	POS
orr-bootleg-2020	573	12	ability	ability	NN
orr-bootleg-2020	573	13	to	to	TO
orr-bootleg-2020	573	14	reason	reason	VB
orr-bootleg-2020	573	15	over	over	IN
orr-bootleg-2020	573	16	the	the	DT
orr-bootleg-2020	573	17	four	four	CD
orr-bootleg-2020	573	18	patterns	pattern	NNS
orr-bootleg-2020	573	19	.	.	.
orr-bootleg-2020	574	1	B.1	B.1	NNP
orr-bootleg-2020	574	2	Evaluation	Evaluation	NNP
orr-bootleg-2020	574	3	Data	Data	NNP
orr-bootleg-2020	574	4	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	574	5	Datasets	Datasets	NNPS
orr-bootleg-2020	574	6	We	-PRON-	PRP
orr-bootleg-2020	574	7	use	use	VBP
orr-bootleg-2020	574	8	two	two	CD
orr-bootleg-2020	574	9	main	main	JJ
orr-bootleg-2020	574	10	datasets	dataset	NNS
orr-bootleg-2020	574	11	to	to	TO
orr-bootleg-2020	574	12	evaluate	evaluate	VB
orr-bootleg-2020	574	13	Bootleg	Bootleg	NNP
orr-bootleg-2020	574	14	.	.	.
orr-bootleg-2020	575	1	•	•	NNP
orr-bootleg-2020	575	2	Wikipedia	wikipedia	NN
orr-bootleg-2020	575	3	:	:	:
orr-bootleg-2020	575	4	we	-PRON-	PRP
orr-bootleg-2020	575	5	use	use	VBP
orr-bootleg-2020	575	6	the	the	DT
orr-bootleg-2020	575	7	November	November	NNP
orr-bootleg-2020	575	8	2019	2019	CD
orr-bootleg-2020	575	9	dump	dump	NN
orr-bootleg-2020	575	10	of	of	IN
orr-bootleg-2020	575	11	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	575	12	to	to	TO
orr-bootleg-2020	575	13	train	train	VB
orr-bootleg-2020	575	14	Bootleg	Bootleg	NNP
orr-bootleg-2020	575	15	.	.	.
orr-bootleg-2020	576	1	We	-PRON-	PRP
orr-bootleg-2020	576	2	use	use	VBP
orr-bootleg-2020	576	3	the	the	DT
orr-bootleg-2020	576	4	set	set	NN
orr-bootleg-2020	576	5	of	of	IN
orr-bootleg-2020	576	6	entities	entity	NNS
orr-bootleg-2020	576	7	that	that	WDT
orr-bootleg-2020	576	8	are	be	VBP
orr-bootleg-2020	576	9	linked	link	VBN
orr-bootleg-2020	576	10	to	to	IN
orr-bootleg-2020	576	11	in	in	IN
orr-bootleg-2020	576	12	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	576	13	for	for	IN
orr-bootleg-2020	576	14	a	a	DT
orr-bootleg-2020	576	15	total	total	NN
orr-bootleg-2020	576	16	of	of	IN
orr-bootleg-2020	576	17	3.3	3.3	CD
orr-bootleg-2020	576	18	M	M	NNP
orr-bootleg-2020	576	19	entities	entity	NNS
orr-bootleg-2020	576	20	.	.	.
orr-bootleg-2020	577	1	After	after	IN
orr-bootleg-2020	577	2	weak	weak	JJ
orr-bootleg-2020	577	3	labelling	labelling	NN
orr-bootleg-2020	577	4	,	,	,
orr-bootleg-2020	577	5	we	-PRON-	PRP
orr-bootleg-2020	577	6	have	have	VBP
orr-bootleg-2020	577	7	a	a	DT
orr-bootleg-2020	577	8	total	total	NN
orr-bootleg-2020	577	9	of	of	IN
orr-bootleg-2020	577	10	5.7	5.7	CD
orr-bootleg-2020	577	11	M	M	NNP
orr-bootleg-2020	577	12	sentences	sentence	NNS
orr-bootleg-2020	577	13	.	.	.
orr-bootleg-2020	578	1	•	•	NNP
orr-bootleg-2020	578	2	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	578	3	Subset	Subset	NNP
orr-bootleg-2020	578	4	:	:	:
orr-bootleg-2020	578	5	we	-PRON-	PRP
orr-bootleg-2020	578	6	use	use	VBP
orr-bootleg-2020	578	7	a	a	DT
orr-bootleg-2020	578	8	subset	subset	NN
orr-bootleg-2020	578	9	of	of	IN
orr-bootleg-2020	578	10	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	578	11	for	for	IN
orr-bootleg-2020	578	12	our	-PRON-	PRP$
orr-bootleg-2020	578	13	micro	micro	JJ
orr-bootleg-2020	578	14	ablation	ablation	NN
orr-bootleg-2020	578	15	experiments	experiment	NNS
orr-bootleg-2020	578	16	over	over	IN
orr-bootleg-2020	578	17	regularization	regularization	NN
orr-bootleg-2020	578	18	parameters	parameter	NNS
orr-bootleg-2020	578	19	.	.	.
orr-bootleg-2020	579	1	We	-PRON-	PRP
orr-bootleg-2020	579	2	generate	generate	VBP
orr-bootleg-2020	579	3	this	this	DT
orr-bootleg-2020	579	4	subset	subset	NN
orr-bootleg-2020	579	5	by	by	IN
orr-bootleg-2020	579	6	taking	take	VBG
orr-bootleg-2020	579	7	all	all	DT
orr-bootleg-2020	579	8	sentences	sentence	NNS
orr-bootleg-2020	579	9	where	where	WRB
orr-bootleg-2020	579	10	at	at	RB
orr-bootleg-2020	579	11	least	least	RBS
orr-bootleg-2020	579	12	one	one	CD
orr-bootleg-2020	579	13	mention	mention	NN
orr-bootleg-2020	579	14	is	be	VBZ
orr-bootleg-2020	579	15	a	a	DT
orr-bootleg-2020	579	16	mention	mention	NN
orr-bootleg-2020	579	17	from	from	IN
orr-bootleg-2020	579	18	the	the	DT
orr-bootleg-2020	579	19	KORE50	KORE50	NNP
orr-bootleg-2020	579	20	disambiguation	disambiguation	NNP
orr-bootleg-2020	579	21	benchmark	benchmark	NN
orr-bootleg-2020	579	22	.	.	.
orr-bootleg-2020	580	1	Our	-PRON-	PRP$
orr-bootleg-2020	580	2	set	set	NN
orr-bootleg-2020	580	3	of	of	IN
orr-bootleg-2020	580	4	entities	entity	NNS
orr-bootleg-2020	580	5	is	be	VBZ
orr-bootleg-2020	580	6	all	all	DT
orr-bootleg-2020	580	7	entities	entity	NNS
orr-bootleg-2020	580	8	and	and	CC
orr-bootleg-2020	580	9	entity	entity	NN
orr-bootleg-2020	580	10	candidates	candidate	NNS
orr-bootleg-2020	580	11	referred	refer	VBN
orr-bootleg-2020	580	12	to	to	IN
orr-bootleg-2020	580	13	by	by	IN
orr-bootleg-2020	580	14	mentions	mention	NNS
orr-bootleg-2020	580	15	in	in	IN
orr-bootleg-2020	580	16	this	this	DT
orr-bootleg-2020	580	17	subset	subset	NN
orr-bootleg-2020	580	18	of	of	IN
orr-bootleg-2020	580	19	sentences	sentence	NNS
orr-bootleg-2020	580	20	.	.	.
orr-bootleg-2020	581	1	We	-PRON-	PRP
orr-bootleg-2020	581	2	have	have	VBP
orr-bootleg-2020	581	3	a	a	DT
orr-bootleg-2020	581	4	total	total	NN
orr-bootleg-2020	581	5	of	of	IN
orr-bootleg-2020	581	6	370,000	370,000	CD
orr-bootleg-2020	581	7	entities	entity	NNS
orr-bootleg-2020	581	8	and	and	CC
orr-bootleg-2020	581	9	520,000	520,000	CD
orr-bootleg-2020	581	10	sentences	sentence	NNS
orr-bootleg-2020	581	11	.	.	.
orr-bootleg-2020	582	1	For	for	IN
orr-bootleg-2020	582	2	our	-PRON-	PRP$
orr-bootleg-2020	582	3	Wikipedia	wikipedia	JJ
orr-bootleg-2020	582	4	experiments	experiment	NNS
orr-bootleg-2020	582	5	,	,	,
orr-bootleg-2020	582	6	we	-PRON-	PRP
orr-bootleg-2020	582	7	use	use	VBP
orr-bootleg-2020	582	8	a	a	DT
orr-bootleg-2020	582	9	80/10/10	80/10/10	CD
orr-bootleg-2020	582	10	train	train	NN
orr-bootleg-2020	582	11	/	/	SYM
orr-bootleg-2020	582	12	test	test	NN
orr-bootleg-2020	582	13	/	/	SYM
orr-bootleg-2020	582	14	dev	dev	NNS
orr-bootleg-2020	582	15	split	split	VBN
orr-bootleg-2020	582	16	by	by	IN
orr-bootleg-2020	582	17	Wikipedia	wikipedia	JJ
orr-bootleg-2020	582	18	pages	page	NNS
orr-bootleg-2020	582	19	,	,	,
orr-bootleg-2020	582	20	meaning	mean	VBG
orr-bootleg-2020	582	21	all	all	DT
orr-bootleg-2020	582	22	sentences	sentence	NNS
orr-bootleg-2020	582	23	for	for	IN
orr-bootleg-2020	582	24	a	a	DT
orr-bootleg-2020	582	25	single	single	JJ
orr-bootleg-2020	582	26	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	582	27	page	page	NN
orr-bootleg-2020	582	28	get	get	VB
orr-bootleg-2020	582	29	placed	place	VBN
orr-bootleg-2020	582	30	into	into	IN
orr-bootleg-2020	582	31	one	one	CD
orr-bootleg-2020	582	32	of	of	IN
orr-bootleg-2020	582	33	the	the	DT
orr-bootleg-2020	582	34	splits	split	NNS
orr-bootleg-2020	582	35	.	.	.
orr-bootleg-2020	583	1	For	for	IN
orr-bootleg-2020	583	2	our	-PRON-	PRP$
orr-bootleg-2020	583	3	benchmark	benchmark	NN
orr-bootleg-2020	583	4	model	model	NN
orr-bootleg-2020	583	5	,	,	,
orr-bootleg-2020	583	6	we	-PRON-	PRP
orr-bootleg-2020	583	7	use	use	VBP
orr-bootleg-2020	583	8	a	a	DT
orr-bootleg-2020	583	9	96/2/2	96/2/2	CD
orr-bootleg-2020	583	10	train	train	NN
orr-bootleg-2020	583	11	/	/	SYM
orr-bootleg-2020	583	12	test	test	NN
orr-bootleg-2020	583	13	/	/	SYM
orr-bootleg-2020	583	14	dev	dev	NNP
orr-bootleg-2020	583	15	split	split	VBP
orr-bootleg-2020	583	16	over	over	IN
orr-bootleg-2020	583	17	sentences	sentence	NNS
orr-bootleg-2020	583	18	to	to	TO
orr-bootleg-2020	583	19	allow	allow	VB
orr-bootleg-2020	583	20	our	-PRON-	PRP$
orr-bootleg-2020	583	21	model	model	NN
orr-bootleg-2020	583	22	to	to	TO
orr-bootleg-2020	583	23	learn	learn	VB
orr-bootleg-2020	583	24	as	as	RB
orr-bootleg-2020	583	25	much	much	JJ
orr-bootleg-2020	583	26	as	as	IN
orr-bootleg-2020	583	27	possible	possible	JJ
orr-bootleg-2020	583	28	from	from	IN
orr-bootleg-2020	583	29	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	583	30	for	for	IN
orr-bootleg-2020	583	31	our	-PRON-	PRP$
orr-bootleg-2020	583	32	benchmark	benchmark	JJ
orr-bootleg-2020	583	33	tasks	task	NNS
orr-bootleg-2020	583	34	.	.	.
orr-bootleg-2020	584	1	19	19	CD
orr-bootleg-2020	584	2	Benchmark	benchmark	JJ
orr-bootleg-2020	584	3	Datasets	dataset	NNS
orr-bootleg-2020	584	4	We	-PRON-	PRP
orr-bootleg-2020	584	5	use	use	VBP
orr-bootleg-2020	584	6	three	three	CD
orr-bootleg-2020	584	7	benchmark	benchmark	JJ
orr-bootleg-2020	584	8	NED	NED	NNP
orr-bootleg-2020	584	9	datasets	dataset	NNS
orr-bootleg-2020	584	10	.	.	.
orr-bootleg-2020	585	1	Following	follow	VBG
orr-bootleg-2020	585	2	standard	standard	JJ
orr-bootleg-2020	585	3	procedure	procedure	NN
orr-bootleg-2020	585	4	[	[	-LRB-
orr-bootleg-2020	585	5	17	17	CD
orr-bootleg-2020	585	6	]	]	-RRB-
orr-bootleg-2020	585	7	,	,	,
orr-bootleg-2020	585	8	we	-PRON-	PRP
orr-bootleg-2020	585	9	only	only	RB
orr-bootleg-2020	585	10	consider	consider	VBP
orr-bootleg-2020	585	11	mentions	mention	NNS
orr-bootleg-2020	585	12	whose	whose	WP$
orr-bootleg-2020	585	13	linked	link	VBN
orr-bootleg-2020	585	14	entities	entity	NNS
orr-bootleg-2020	585	15	appear	appear	VBP
orr-bootleg-2020	585	16	in	in	IN
orr-bootleg-2020	585	17	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	585	18	.	.	.
orr-bootleg-2020	586	1	The	the	DT
orr-bootleg-2020	586	2	datasets	dataset	NNS
orr-bootleg-2020	586	3	are	be	VBP
orr-bootleg-2020	586	4	summarized	summarize	VBN
orr-bootleg-2020	586	5	as	as	IN
orr-bootleg-2020	586	6	follows	follow	VBZ
orr-bootleg-2020	586	7	:	:	:
orr-bootleg-2020	586	8	•	•	NNP
orr-bootleg-2020	586	9	KORE50	KORE50	NNP
orr-bootleg-2020	586	10	:	:	:
orr-bootleg-2020	586	11	KORE50	KORE50	NNP
orr-bootleg-2020	586	12	[	[	-LRB-
orr-bootleg-2020	586	13	23	23	CD
orr-bootleg-2020	586	14	]	]	-RRB-
orr-bootleg-2020	586	15	represents	represent	VBZ
orr-bootleg-2020	586	16	difficult	difficult	JJ
orr-bootleg-2020	586	17	-	-	HYPH
orr-bootleg-2020	586	18	to	to	IN
orr-bootleg-2020	586	19	-	-	HYPH
orr-bootleg-2020	586	20	disambiguate	disambiguate	NN
orr-bootleg-2020	586	21	sentences	sentence	NNS
orr-bootleg-2020	586	22	and	and	CC
orr-bootleg-2020	586	23	contains	contain	VBZ
orr-bootleg-2020	586	24	144	144	CD
orr-bootleg-2020	586	25	mentions	mention	NNS
orr-bootleg-2020	586	26	to	to	TO
orr-bootleg-2020	586	27	disambiguate	disambiguate	VB
orr-bootleg-2020	586	28	.	.	.
orr-bootleg-2020	587	1	Note	note	VB
orr-bootleg-2020	587	2	,	,	,
orr-bootleg-2020	587	3	as	as	IN
orr-bootleg-2020	587	4	of	of	IN
orr-bootleg-2020	587	5	the	the	DT
orr-bootleg-2020	587	6	Nov	Nov	NNP
orr-bootleg-2020	587	7	2019	2019	CD
orr-bootleg-2020	587	8	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	587	9	dump	dump	NN
orr-bootleg-2020	587	10	,	,	,
orr-bootleg-2020	587	11	one	one	CD
orr-bootleg-2020	587	12	mention	mention	NN
orr-bootleg-2020	587	13	in	in	IN
orr-bootleg-2020	587	14	the	the	DT
orr-bootleg-2020	587	15	144	144	CD
orr-bootleg-2020	587	16	does	do	VBZ
orr-bootleg-2020	587	17	not	not	RB
orr-bootleg-2020	587	18	have	have	VB
orr-bootleg-2020	587	19	a	a	DT
orr-bootleg-2020	587	20	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	587	21	page	page	NN
orr-bootleg-2020	587	22	.	.	.
orr-bootleg-2020	588	1	Although	although	IN
orr-bootleg-2020	588	2	it	-PRON-	PRP
orr-bootleg-2020	588	3	is	be	VBZ
orr-bootleg-2020	588	4	standard	standard	JJ
orr-bootleg-2020	588	5	to	to	TO
orr-bootleg-2020	588	6	remove	remove	VB
orr-bootleg-2020	588	7	mentions	mention	NNS
orr-bootleg-2020	588	8	that	that	WDT
orr-bootleg-2020	588	9	do	do	VBP
orr-bootleg-2020	588	10	not	not	RB
orr-bootleg-2020	588	11	link	link	VB
orr-bootleg-2020	588	12	to	to	IN
orr-bootleg-2020	588	13	an	an	DT
orr-bootleg-2020	588	14	entity	entity	NN
orr-bootleg-2020	588	15	in	in	IN
orr-bootleg-2020	588	16	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	588	17	,	,	,
orr-bootleg-2020	588	18	to	to	TO
orr-bootleg-2020	588	19	be	be	VB
orr-bootleg-2020	588	20	comparable	comparable	JJ
orr-bootleg-2020	588	21	to	to	IN
orr-bootleg-2020	588	22	other	other	JJ
orr-bootleg-2020	588	23	methods	method	NNS
orr-bootleg-2020	588	24	,	,	,
orr-bootleg-2020	588	25	we	-PRON-	PRP
orr-bootleg-2020	588	26	measure	measure	VBP
orr-bootleg-2020	588	27	with	with	IN
orr-bootleg-2020	588	28	144	144	CD
orr-bootleg-2020	588	29	mentions	mention	NNS
orr-bootleg-2020	588	30	,	,	,
orr-bootleg-2020	588	31	not	not	RB
orr-bootleg-2020	588	32	143	143	CD
orr-bootleg-2020	588	33	.	.	.
orr-bootleg-2020	589	1	•	•	NNP
orr-bootleg-2020	589	2	RSS500	RSS500	NNP
orr-bootleg-2020	589	3	:	:	:
orr-bootleg-2020	589	4	RSS500	RSS500	NNP
orr-bootleg-2020	589	5	[	[	-LRB-
orr-bootleg-2020	589	6	18	18	CD
orr-bootleg-2020	589	7	]	]	-RRB-
orr-bootleg-2020	589	8	is	be	VBZ
orr-bootleg-2020	589	9	a	a	DT
orr-bootleg-2020	589	10	dataset	dataset	NN
orr-bootleg-2020	589	11	of	of	IN
orr-bootleg-2020	589	12	news	news	NN
orr-bootleg-2020	589	13	sentences	sentence	NNS
orr-bootleg-2020	589	14	and	and	CC
orr-bootleg-2020	589	15	contains	contain	VBZ
orr-bootleg-2020	589	16	520	520	CD
orr-bootleg-2020	589	17	mentions	mention	NNS
orr-bootleg-2020	589	18	(	(	-LRB-
orr-bootleg-2020	589	19	4	4	CD
orr-bootleg-2020	589	20	of	of	IN
orr-bootleg-2020	589	21	the	the	DT
orr-bootleg-2020	589	22	mentions	mention	NNS
orr-bootleg-2020	589	23	did	do	VBD
orr-bootleg-2020	589	24	not	not	RB
orr-bootleg-2020	589	25	have	have	VB
orr-bootleg-2020	589	26	entities	entity	NNS
orr-bootleg-2020	589	27	in	in	IN
orr-bootleg-2020	589	28	E	e	NN
orr-bootleg-2020	589	29	)	)	-RRB-
orr-bootleg-2020	589	30	.	.	.
orr-bootleg-2020	590	1	•	•	NNP
orr-bootleg-2020	590	2	AIDA	AIDA	NNP
orr-bootleg-2020	590	3	CoNLL	CoNLL	NNP
orr-bootleg-2020	590	4	-	-	HYPH
orr-bootleg-2020	590	5	YAGO	YAGO	NNP
orr-bootleg-2020	590	6	:	:	:
orr-bootleg-2020	590	7	AIDA	AIDA	NNP
orr-bootleg-2020	590	8	CoNLL	CoNLL	NNP
orr-bootleg-2020	590	9	-	-	HYPH
orr-bootleg-2020	590	10	YAGO	YAGO	NNP
orr-bootleg-2020	590	11	[	[	-LRB-
orr-bootleg-2020	590	12	22	22	CD
orr-bootleg-2020	590	13	]	]	-RRB-
orr-bootleg-2020	590	14	is	be	VBZ
orr-bootleg-2020	590	15	a	a	DT
orr-bootleg-2020	590	16	document	document	NN
orr-bootleg-2020	590	17	-	-	HYPH
orr-bootleg-2020	590	18	based	base	VBN
orr-bootleg-2020	590	19	news	news	NN
orr-bootleg-2020	590	20	dataset	dataset	NN
orr-bootleg-2020	590	21	containing	contain	VBG
orr-bootleg-2020	590	22	4	4	CD
orr-bootleg-2020	590	23	,	,	,
orr-bootleg-2020	590	24	485	485	CD
orr-bootleg-2020	590	25	test	test	NN
orr-bootleg-2020	590	26	mentions	mention	NNS
orr-bootleg-2020	590	27	,	,	,
orr-bootleg-2020	590	28	4	4	CD
orr-bootleg-2020	590	29	,	,	,
orr-bootleg-2020	590	30	791	791	CD
orr-bootleg-2020	590	31	validation	validation	NN
orr-bootleg-2020	590	32	set	set	NN
orr-bootleg-2020	590	33	mentions	mention	NNS
orr-bootleg-2020	590	34	,	,	,
orr-bootleg-2020	590	35	and	and	CC
orr-bootleg-2020	590	36	18	18	CD
orr-bootleg-2020	590	37	,	,	,
orr-bootleg-2020	590	38	541	541	CD
orr-bootleg-2020	590	39	training	training	NN
orr-bootleg-2020	590	40	mentions	mention	NNS
orr-bootleg-2020	590	41	.	.	.
orr-bootleg-2020	591	1	As	as	IN
orr-bootleg-2020	591	2	Bootleg	Bootleg	NNP
orr-bootleg-2020	591	3	is	be	VBZ
orr-bootleg-2020	591	4	a	a	DT
orr-bootleg-2020	591	5	sentence	sentence	NN
orr-bootleg-2020	591	6	-	-	HYPH
orr-bootleg-2020	591	7	level	level	NN
orr-bootleg-2020	591	8	NED	NED	NNP
orr-bootleg-2020	591	9	system	system	NN
orr-bootleg-2020	591	10	,	,	,
orr-bootleg-2020	591	11	we	-PRON-	PRP
orr-bootleg-2020	591	12	create	create	VBP
orr-bootleg-2020	591	13	sentences	sentence	NNS
orr-bootleg-2020	591	14	from	from	IN
orr-bootleg-2020	591	15	documents	document	NNS
orr-bootleg-2020	591	16	following	follow	VBG
orr-bootleg-2020	591	17	the	the	DT
orr-bootleg-2020	591	18	technique	technique	NN
orr-bootleg-2020	591	19	from	from	IN
orr-bootleg-2020	591	20	Févry	Févry	NNP
orr-bootleg-2020	591	21	et	et	NNP
orr-bootleg-2020	591	22	al	al	NNP
orr-bootleg-2020	591	23	.	.	.
orr-bootleg-2020	592	1	[	[	-LRB-
orr-bootleg-2020	592	2	16	16	CD
orr-bootleg-2020	592	3	]	]	-RRB-
orr-bootleg-2020	592	4	where	where	WRB
orr-bootleg-2020	592	5	we	-PRON-	PRP
orr-bootleg-2020	592	6	concatenate	concatenate	VBP
orr-bootleg-2020	592	7	the	the	DT
orr-bootleg-2020	592	8	title	title	NN
orr-bootleg-2020	592	9	of	of	IN
orr-bootleg-2020	592	10	the	the	DT
orr-bootleg-2020	592	11	document	document	NN
orr-bootleg-2020	592	12	to	to	IN
orr-bootleg-2020	592	13	the	the	DT
orr-bootleg-2020	592	14	beginning	beginning	NN
orr-bootleg-2020	592	15	of	of	IN
orr-bootleg-2020	592	16	each	each	DT
orr-bootleg-2020	592	17	sentence	sentence	NN
orr-bootleg-2020	592	18	.	.	.
orr-bootleg-2020	593	1	To	to	TO
orr-bootleg-2020	593	2	improve	improve	VB
orr-bootleg-2020	593	3	the	the	DT
orr-bootleg-2020	593	4	quality	quality	NN
orr-bootleg-2020	593	5	of	of	IN
orr-bootleg-2020	593	6	annotated	annotate	VBN
orr-bootleg-2020	593	7	mention	mention	NN
orr-bootleg-2020	593	8	boundaries	boundary	NNS
orr-bootleg-2020	593	9	in	in	IN
orr-bootleg-2020	593	10	the	the	DT
orr-bootleg-2020	593	11	benchmarks	benchmark	NNS
orr-bootleg-2020	593	12	,	,	,
orr-bootleg-2020	593	13	we	-PRON-	PRP
orr-bootleg-2020	593	14	follow	follow	VBP
orr-bootleg-2020	593	15	the	the	DT
orr-bootleg-2020	593	16	technique	technique	NN
orr-bootleg-2020	593	17	of	of	IN
orr-bootleg-2020	593	18	Phan	Phan	NNP
orr-bootleg-2020	593	19	et	et	NNP
orr-bootleg-2020	593	20	al	al	NNP
orr-bootleg-2020	593	21	.	.	.
orr-bootleg-2020	594	1	[	[	-LRB-
orr-bootleg-2020	594	2	40	40	CD
orr-bootleg-2020	594	3	]	]	-RRB-
orr-bootleg-2020	594	4	and	and	CC
orr-bootleg-2020	594	5	allow	allow	VB
orr-bootleg-2020	594	6	for	for	IN
orr-bootleg-2020	594	7	mention	mention	NN
orr-bootleg-2020	594	8	boundary	boundary	JJ
orr-bootleg-2020	594	9	expansion	expansion	NN
orr-bootleg-2020	594	10	using	use	VBG
orr-bootleg-2020	594	11	a	a	DT
orr-bootleg-2020	594	12	standard	standard	JJ
orr-bootleg-2020	594	13	off	off	IN
orr-bootleg-2020	594	14	-	-	HYPH
orr-bootleg-2020	594	15	the	the	DT
orr-bootleg-2020	594	16	-	-	HYPH
orr-bootleg-2020	594	17	shelf	shelf	NN
orr-bootleg-2020	594	18	NER	NER	NNP
orr-bootleg-2020	594	19	tagger.11	tagger.11	NNP
orr-bootleg-2020	594	20	For	for	IN
orr-bootleg-2020	594	21	candidate	candidate	NN
orr-bootleg-2020	594	22	generation	generation	NN
orr-bootleg-2020	594	23	,	,	,
orr-bootleg-2020	594	24	as	as	IN
orr-bootleg-2020	594	25	aliases	alias	NNS
orr-bootleg-2020	594	26	may	may	MD
orr-bootleg-2020	594	27	not	not	RB
orr-bootleg-2020	594	28	exist	exist	VB
orr-bootleg-2020	594	29	in	in	IN
orr-bootleg-2020	594	30	Γ	Γ	NNP
orr-bootleg-2020	594	31	,	,	,
orr-bootleg-2020	594	32	we	-PRON-	PRP
orr-bootleg-2020	594	33	gather	gather	VBP
orr-bootleg-2020	594	34	possible	possible	JJ
orr-bootleg-2020	594	35	candidates	candidate	NNS
orr-bootleg-2020	594	36	by	by	IN
orr-bootleg-2020	594	37	looking	look	VBG
orr-bootleg-2020	594	38	at	at	IN
orr-bootleg-2020	594	39	n	n	NN
orr-bootleg-2020	594	40	-	-	HYPH
orr-bootleg-2020	594	41	grams	gram	NNS
orr-bootleg-2020	594	42	in	in	IN
orr-bootleg-2020	594	43	descending	descend	VBG
orr-bootleg-2020	594	44	order	order	NN
orr-bootleg-2020	594	45	of	of	IN
orr-bootleg-2020	594	46	length	length	NN
orr-bootleg-2020	594	47	and	and	CC
orr-bootleg-2020	594	48	determine	determine	VB
orr-bootleg-2020	594	49	the	the	DT
orr-bootleg-2020	594	50	top	top	JJ
orr-bootleg-2020	594	51	30	30	CD
orr-bootleg-2020	594	52	by	by	IN
orr-bootleg-2020	594	53	measuring	measure	VBG
orr-bootleg-2020	594	54	the	the	DT
orr-bootleg-2020	594	55	similarity	similarity	NN
orr-bootleg-2020	594	56	of	of	IN
orr-bootleg-2020	594	57	the	the	DT
orr-bootleg-2020	594	58	proper	proper	JJ
orr-bootleg-2020	594	59	nouns	noun	NNS
orr-bootleg-2020	594	60	in	in	IN
orr-bootleg-2020	594	61	the	the	DT
orr-bootleg-2020	594	62	example	example	NN
orr-bootleg-2020	594	63	sentence	sentence	NN
orr-bootleg-2020	594	64	to	to	IN
orr-bootleg-2020	594	65	each	each	DT
orr-bootleg-2020	594	66	candidate	candidate	NN
orr-bootleg-2020	594	67	’s	’s	POS
orr-bootleg-2020	594	68	Wikipedia	wikipedia	NN
orr-bootleg-2020	594	69	page	page	NN
orr-bootleg-2020	594	70	text	text	NN
orr-bootleg-2020	594	71	.	.	.
orr-bootleg-2020	595	1	Structural	Structural	NNP
orr-bootleg-2020	595	2	Resources	resource	NNS
orr-bootleg-2020	595	3	The	the	DT
orr-bootleg-2020	595	4	last	last	JJ
orr-bootleg-2020	595	5	source	source	NN
orr-bootleg-2020	595	6	of	of	IN
orr-bootleg-2020	595	7	input	input	NN
orr-bootleg-2020	595	8	data	datum	NNS
orr-bootleg-2020	595	9	to	to	IN
orr-bootleg-2020	595	10	Bootleg	Bootleg	NNP
orr-bootleg-2020	595	11	is	be	VBZ
orr-bootleg-2020	595	12	the	the	DT
orr-bootleg-2020	595	13	structural	structural	JJ
orr-bootleg-2020	595	14	resources	resource	NNS
orr-bootleg-2020	595	15	of	of	IN
orr-bootleg-2020	595	16	types	type	NNS
orr-bootleg-2020	595	17	and	and	CC
orr-bootleg-2020	595	18	knowledge	knowledge	NN
orr-bootleg-2020	595	19	graph	graph	NN
orr-bootleg-2020	595	20	relations	relation	NNS
orr-bootleg-2020	595	21	.	.	.
orr-bootleg-2020	596	1	We	-PRON-	PRP
orr-bootleg-2020	596	2	extract	extract	VBP
orr-bootleg-2020	596	3	relations	relation	NNS
orr-bootleg-2020	596	4	from	from	IN
orr-bootleg-2020	596	5	Wikidata	wikidata	JJ
orr-bootleg-2020	596	6	knowledge	knowledge	NN
orr-bootleg-2020	596	7	graph	graph	NN
orr-bootleg-2020	596	8	triples	triple	NNS
orr-bootleg-2020	596	9	.	.	.
orr-bootleg-2020	597	1	For	for	IN
orr-bootleg-2020	597	2	our	-PRON-	PRP$
orr-bootleg-2020	597	3	pairwise	pairwise	NN
orr-bootleg-2020	597	4	KG	kg	NN
orr-bootleg-2020	597	5	adjacency	adjacency	NN
orr-bootleg-2020	597	6	matrix	matrix	NN
orr-bootleg-2020	597	7	used	use	VBN
orr-bootleg-2020	597	8	in	in	IN
orr-bootleg-2020	597	9	KG2Ent	kg2ent	NN
orr-bootleg-2020	597	10	,	,	,
orr-bootleg-2020	597	11	we	-PRON-	PRP
orr-bootleg-2020	597	12	require	require	VBP
orr-bootleg-2020	597	13	the	the	DT
orr-bootleg-2020	597	14	subject	subject	NN
orr-bootleg-2020	597	15	and	and	CC
orr-bootleg-2020	597	16	object	object	NN
orr-bootleg-2020	597	17	to	to	TO
orr-bootleg-2020	597	18	be	be	VB
orr-bootleg-2020	597	19	in	in	IN
orr-bootleg-2020	597	20	E.	E.	NNP
orr-bootleg-2020	597	21	For	for	IN
orr-bootleg-2020	597	22	our	-PRON-	PRP$
orr-bootleg-2020	597	23	relation	relation	NN
orr-bootleg-2020	597	24	embeddings	embedding	NNS
orr-bootleg-2020	597	25	,	,	,
orr-bootleg-2020	597	26	we	-PRON-	PRP
orr-bootleg-2020	597	27	only	only	RB
orr-bootleg-2020	597	28	require	require	VBP
orr-bootleg-2020	597	29	the	the	DT
orr-bootleg-2020	597	30	subject	subject	NN
orr-bootleg-2020	597	31	be	be	VB
orr-bootleg-2020	597	32	in	in	IN
orr-bootleg-2020	597	33	E	e	NN
orr-bootleg-2020	597	34	as	as	IN
orr-bootleg-2020	597	35	our	-PRON-	PRP$
orr-bootleg-2020	597	36	goal	goal	NN
orr-bootleg-2020	597	37	is	be	VBZ
orr-bootleg-2020	597	38	to	to	TO
orr-bootleg-2020	597	39	extract	extract	VB
orr-bootleg-2020	597	40	all	all	DT
orr-bootleg-2020	597	41	relations	relation	NNS
orr-bootleg-2020	597	42	an	an	DT
orr-bootleg-2020	597	43	entity	entity	NN
orr-bootleg-2020	597	44	participates	participate	VBZ
orr-bootleg-2020	597	45	in	in	IN
orr-bootleg-2020	597	46	independent	independent	JJ
orr-bootleg-2020	597	47	of	of	IN
orr-bootleg-2020	597	48	the	the	DT
orr-bootleg-2020	597	49	other	other	JJ
orr-bootleg-2020	597	50	entities	entity	NNS
orr-bootleg-2020	597	51	in	in	IN
orr-bootleg-2020	597	52	the	the	DT
orr-bootleg-2020	597	53	sentence	sentence	NN
orr-bootleg-2020	597	54	.	.	.
orr-bootleg-2020	598	1	We	-PRON-	PRP
orr-bootleg-2020	598	2	have	have	VBP
orr-bootleg-2020	598	3	a	a	DT
orr-bootleg-2020	598	4	total	total	NN
orr-bootleg-2020	598	5	of	of	IN
orr-bootleg-2020	598	6	1,197	1,197	CD
orr-bootleg-2020	598	7	relations	relation	NNS
orr-bootleg-2020	598	8	.	.	.
orr-bootleg-2020	599	1	We	-PRON-	PRP
orr-bootleg-2020	599	2	use	use	VBP
orr-bootleg-2020	599	3	two	two	CD
orr-bootleg-2020	599	4	different	different	JJ
orr-bootleg-2020	599	5	type	type	NN
orr-bootleg-2020	599	6	sources	source	NNS
orr-bootleg-2020	599	7	to	to	IN
orr-bootleg-2020	599	8	assign	assign	VB
orr-bootleg-2020	599	9	types	type	NNS
orr-bootleg-2020	599	10	to	to	IN
orr-bootleg-2020	599	11	entities	entity	NNS
orr-bootleg-2020	599	12	—	—	:
orr-bootleg-2020	599	13	Wikidata	wikidata	NN
orr-bootleg-2020	599	14	types	type	NNS
orr-bootleg-2020	599	15	and	and	CC
orr-bootleg-2020	599	16	HYENA	hyena	NN
orr-bootleg-2020	599	17	types	type	NNS
orr-bootleg-2020	599	18	[	[	-LRB-
orr-bootleg-2020	599	19	52]—and	52]—and	CD
orr-bootleg-2020	599	20	use	use	NN
orr-bootleg-2020	599	21	coarse	coarse	JJ
orr-bootleg-2020	599	22	HYENA	hyena	JJ
orr-bootleg-2020	599	23	types	type	NNS
orr-bootleg-2020	599	24	for	for	IN
orr-bootleg-2020	599	25	type	type	NN
orr-bootleg-2020	599	26	prediction	prediction	NN
orr-bootleg-2020	599	27	.	.	.
orr-bootleg-2020	600	1	The	the	DT
orr-bootleg-2020	600	2	Wikidata	Wikidata	NNP
orr-bootleg-2020	600	3	types	type	NNS
orr-bootleg-2020	600	4	are	be	VBP
orr-bootleg-2020	600	5	generated	generate	VBN
orr-bootleg-2020	600	6	from	from	IN
orr-bootleg-2020	600	7	Wikidata	Wikidata	NNP
orr-bootleg-2020	600	8	’s	’s	''
orr-bootleg-2020	600	9	“	"	``
orr-bootleg-2020	600	10	instance	instance	NN
orr-bootleg-2020	600	11	of	of	IN
orr-bootleg-2020	600	12	”	"	''
orr-bootleg-2020	600	13	,	,	,
orr-bootleg-2020	600	14	“	"	``
orr-bootleg-2020	600	15	subclass	subclass	NN
orr-bootleg-2020	600	16	of	of	IN
orr-bootleg-2020	600	17	”	"	''
orr-bootleg-2020	600	18	,	,	,
orr-bootleg-2020	600	19	and	and	CC
orr-bootleg-2020	600	20	“	"	``
orr-bootleg-2020	600	21	occupation	occupation	NN
orr-bootleg-2020	600	22	”	"	''
orr-bootleg-2020	600	23	relationships	relationship	NNS
orr-bootleg-2020	600	24	.	.	.
orr-bootleg-2020	601	1	The	the	DT
orr-bootleg-2020	601	2	“	"	``
orr-bootleg-2020	601	3	occupation	occupation	NN
orr-bootleg-2020	601	4	"	"	''
orr-bootleg-2020	601	5	types	type	NNS
orr-bootleg-2020	601	6	are	be	VBP
orr-bootleg-2020	601	7	used	use	VBN
orr-bootleg-2020	601	8	to	to	TO
orr-bootleg-2020	601	9	improve	improve	VB
orr-bootleg-2020	601	10	disambiguation	disambiguation	NN
orr-bootleg-2020	601	11	of	of	IN
orr-bootleg-2020	601	12	people	people	NNS
orr-bootleg-2020	601	13	,	,	,
orr-bootleg-2020	601	14	which	which	WDT
orr-bootleg-2020	601	15	otherwise	otherwise	RB
orr-bootleg-2020	601	16	only	only	RB
orr-bootleg-2020	601	17	receive	receive	VBP
orr-bootleg-2020	601	18	“	"	``
orr-bootleg-2020	601	19	human	human	JJ
orr-bootleg-2020	601	20	"	"	''
orr-bootleg-2020	601	21	types	type	NNS
orr-bootleg-2020	601	22	in	in	IN
orr-bootleg-2020	601	23	Wikidata	Wikidata	NNP
orr-bootleg-2020	601	24	.	.	.
orr-bootleg-2020	602	1	We	-PRON-	PRP
orr-bootleg-2020	602	2	filter	filter	VBP
orr-bootleg-2020	602	3	the	the	DT
orr-bootleg-2020	602	4	set	set	NN
orr-bootleg-2020	602	5	of	of	IN
orr-bootleg-2020	602	6	Wikidata	Wikidata	NNP
orr-bootleg-2020	602	7	types	type	NNS
orr-bootleg-2020	602	8	to	to	TO
orr-bootleg-2020	602	9	be	be	VB
orr-bootleg-2020	602	10	only	only	RB
orr-bootleg-2020	602	11	those	those	DT
orr-bootleg-2020	602	12	occurring	occur	VBG
orr-bootleg-2020	602	13	100	100	CD
orr-bootleg-2020	602	14	or	or	CC
orr-bootleg-2020	602	15	more	more	JJR
orr-bootleg-2020	602	16	times	time	NNS
orr-bootleg-2020	602	17	in	in	IN
orr-bootleg-2020	602	18	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	602	19	,	,	,
orr-bootleg-2020	602	20	leaving	leave	VBG
orr-bootleg-2020	602	21	27	27	CD
orr-bootleg-2020	602	22	K	k	NN
orr-bootleg-2020	602	23	Wikidata	Wikidata	NNP
orr-bootleg-2020	602	24	types	type	NNS
orr-bootleg-2020	602	25	in	in	IN
orr-bootleg-2020	602	26	total	total	NN
orr-bootleg-2020	602	27	.	.	.
orr-bootleg-2020	603	1	The	the	DT
orr-bootleg-2020	603	2	HYENA	HYENA	NNP
orr-bootleg-2020	603	3	type	type	NN
orr-bootleg-2020	603	4	hierarchy	hierarchy	NN
orr-bootleg-2020	603	5	has	have	VBZ
orr-bootleg-2020	603	6	505	505	CD
orr-bootleg-2020	603	7	types	type	NNS
orr-bootleg-2020	603	8	derived	derive	VBN
orr-bootleg-2020	603	9	from	from	IN
orr-bootleg-2020	603	10	the	the	DT
orr-bootleg-2020	603	11	YAGO	YAGO	NNP
orr-bootleg-2020	603	12	type	type	NN
orr-bootleg-2020	603	13	hierarchy	hierarchy	NN
orr-bootleg-2020	603	14	.	.	.
orr-bootleg-2020	604	1	We	-PRON-	PRP
orr-bootleg-2020	604	2	also	also	RB
orr-bootleg-2020	604	3	use	use	VBP
orr-bootleg-2020	604	4	the	the	DT
orr-bootleg-2020	604	5	coarsest	coarsest	NN
orr-bootleg-2020	604	6	HYENA	hyena	NN
orr-bootleg-2020	604	7	type	type	NN
orr-bootleg-2020	604	8	for	for	IN
orr-bootleg-2020	604	9	an	an	DT
orr-bootleg-2020	604	10	entity	entity	NN
orr-bootleg-2020	604	11	as	as	IN
orr-bootleg-2020	604	12	the	the	DT
orr-bootleg-2020	604	13	gold	gold	JJ
orr-bootleg-2020	604	14	type	type	NN
orr-bootleg-2020	604	15	for	for	IN
orr-bootleg-2020	604	16	type	type	NN
orr-bootleg-2020	604	17	prediction	prediction	NN
orr-bootleg-2020	604	18	.	.	.
orr-bootleg-2020	605	1	There	there	EX
orr-bootleg-2020	605	2	are	be	VBP
orr-bootleg-2020	605	3	5	5	CD
orr-bootleg-2020	605	4	coarse	coarse	JJ
orr-bootleg-2020	605	5	HYENA	hyena	JJ
orr-bootleg-2020	605	6	types	type	NNS
orr-bootleg-2020	605	7	of	of	IN
orr-bootleg-2020	605	8	person	person	NN
orr-bootleg-2020	605	9	,	,	,
orr-bootleg-2020	605	10	location	location	NN
orr-bootleg-2020	605	11	,	,	,
orr-bootleg-2020	605	12	organization	organization	NN
orr-bootleg-2020	605	13	,	,	,
orr-bootleg-2020	605	14	artifact	artifact	NN
orr-bootleg-2020	605	15	,	,	,
orr-bootleg-2020	605	16	event	event	NN
orr-bootleg-2020	605	17	,	,	,
orr-bootleg-2020	605	18	and	and	CC
orr-bootleg-2020	605	19	miscellaneous	miscellaneous	JJ
orr-bootleg-2020	605	20	.	.	.
orr-bootleg-2020	606	1	B.2	B.2	NNP
orr-bootleg-2020	606	2	Training	Training	NNP
orr-bootleg-2020	606	3	Details	Details	NNP
orr-bootleg-2020	606	4	Model	Model	NNP
orr-bootleg-2020	606	5	Parameters	Parameters	NNPS
orr-bootleg-2020	606	6	We	-PRON-	PRP
orr-bootleg-2020	606	7	run	run	VBP
orr-bootleg-2020	606	8	three	three	CD
orr-bootleg-2020	606	9	separate	separate	JJ
orr-bootleg-2020	606	10	models	model	NNS
orr-bootleg-2020	606	11	of	of	IN
orr-bootleg-2020	606	12	Bootleg	Bootleg	NNP
orr-bootleg-2020	606	13	:	:	:
orr-bootleg-2020	606	14	two	two	CD
orr-bootleg-2020	606	15	on	on	IN
orr-bootleg-2020	606	16	our	-PRON-	PRP$
orr-bootleg-2020	606	17	full	full	JJ
orr-bootleg-2020	606	18	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	606	19	data	datum	NNS
orr-bootleg-2020	606	20	(	(	-LRB-
orr-bootleg-2020	606	21	one	one	CD
orr-bootleg-2020	606	22	for	for	IN
orr-bootleg-2020	606	23	the	the	DT
orr-bootleg-2020	606	24	ablation	ablation	NN
orr-bootleg-2020	606	25	and	and	CC
orr-bootleg-2020	606	26	one	one	CD
orr-bootleg-2020	606	27	for	for	IN
orr-bootleg-2020	606	28	the	the	DT
orr-bootleg-2020	606	29	benchmarks	benchmark	NNS
orr-bootleg-2020	606	30	)	)	-RRB-
orr-bootleg-2020	606	31	and	and	CC
orr-bootleg-2020	606	32	one	one	CD
orr-bootleg-2020	606	33	on	on	IN
orr-bootleg-2020	606	34	our	-PRON-	PRP$
orr-bootleg-2020	606	35	micro	micro	JJ
orr-bootleg-2020	606	36	data	datum	NNS
orr-bootleg-2020	606	37	.	.	.
orr-bootleg-2020	607	1	For	for	IN
orr-bootleg-2020	607	2	all	all	DT
orr-bootleg-2020	607	3	models	model	NNS
orr-bootleg-2020	607	4	we	-PRON-	PRP
orr-bootleg-2020	607	5	use	use	VBP
orr-bootleg-2020	607	6	30	30	CD
orr-bootleg-2020	607	7	candidates	candidate	NNS
orr-bootleg-2020	607	8	for	for	IN
orr-bootleg-2020	607	9	each	each	DT
orr-bootleg-2020	607	10	mention	mention	NN
orr-bootleg-2020	607	11	and	and	CC
orr-bootleg-2020	607	12	incorporate	incorporate	VB
orr-bootleg-2020	607	13	the	the	DT
orr-bootleg-2020	607	14	structural	structural	JJ
orr-bootleg-2020	607	15	resources	resource	NNS
orr-bootleg-2020	607	16	discussed	discuss	VBN
orr-bootleg-2020	607	17	above	above	RB
orr-bootleg-2020	607	18	.	.	.
orr-bootleg-2020	608	1	We	-PRON-	PRP
orr-bootleg-2020	608	2	set	set	VBP
orr-bootleg-2020	608	3	T	t	NN
orr-bootleg-2020	608	4	=	=	SYM
orr-bootleg-2020	608	5	3	3	CD
orr-bootleg-2020	608	6	and	and	CC
orr-bootleg-2020	608	7	R	r	NN
orr-bootleg-2020	608	8	=	=	SYM
orr-bootleg-2020	608	9	50	50	CD
orr-bootleg-2020	608	10	for	for	IN
orr-bootleg-2020	608	11	the	the	DT
orr-bootleg-2020	608	12	number	number	NN
orr-bootleg-2020	608	13	of	of	IN
orr-bootleg-2020	608	14	types	type	NNS
orr-bootleg-2020	608	15	and	and	CC
orr-bootleg-2020	608	16	relations	relation	NNS
orr-bootleg-2020	608	17	assigned	assign	VBN
orr-bootleg-2020	608	18	to	to	IN
orr-bootleg-2020	608	19	each	each	DT
orr-bootleg-2020	608	20	entity	entity	NN
orr-bootleg-2020	608	21	.	.	.
orr-bootleg-2020	609	1	For	for	IN
orr-bootleg-2020	609	2	the	the	DT
orr-bootleg-2020	609	3	models	model	NNS
orr-bootleg-2020	609	4	trained	train	VBN
orr-bootleg-2020	609	5	on	on	IN
orr-bootleg-2020	609	6	our	-PRON-	PRP$
orr-bootleg-2020	609	7	full	full	JJ
orr-bootleg-2020	609	8	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	609	9	data	datum	NNS
orr-bootleg-2020	609	10	,	,	,
orr-bootleg-2020	609	11	we	-PRON-	PRP
orr-bootleg-2020	609	12	set	set	VBD
orr-bootleg-2020	609	13	the	the	DT
orr-bootleg-2020	609	14	hidden	hide	VBN
orr-bootleg-2020	609	15	dimension	dimension	NN
orr-bootleg-2020	609	16	to	to	IN
orr-bootleg-2020	609	17	512	512	CD
orr-bootleg-2020	609	18	,	,	,
orr-bootleg-2020	609	19	the	the	DT
orr-bootleg-2020	609	20	dimension	dimension	NN
orr-bootleg-2020	609	21	of	of	IN
orr-bootleg-2020	609	22	u	u	NNP
orr-bootleg-2020	609	23	to	to	IN
orr-bootleg-2020	609	24	256	256	CD
orr-bootleg-2020	609	25	,	,	,
orr-bootleg-2020	609	26	and	and	CC
orr-bootleg-2020	609	27	the	the	DT
orr-bootleg-2020	609	28	dimension	dimension	NN
orr-bootleg-2020	609	29	of	of	IN
orr-bootleg-2020	609	30	all	all	DT
orr-bootleg-2020	609	31	other	other	JJ
orr-bootleg-2020	609	32	type	type	NN
orr-bootleg-2020	609	33	and	and	CC
orr-bootleg-2020	609	34	relation	relation	NN
orr-bootleg-2020	609	35	embeddings	embedding	NNS
orr-bootleg-2020	609	36	to	to	IN
orr-bootleg-2020	609	37	128	128	CD
orr-bootleg-2020	609	38	.	.	.
orr-bootleg-2020	610	1	For	for	IN
orr-bootleg-2020	610	2	our	-PRON-	PRP$
orr-bootleg-2020	610	3	models	model	NNS
orr-bootleg-2020	610	4	trained	train	VBN
orr-bootleg-2020	610	5	on	on	IN
orr-bootleg-2020	610	6	our	-PRON-	PRP$
orr-bootleg-2020	610	7	micro	micro	JJ
orr-bootleg-2020	610	8	dataset	dataset	NN
orr-bootleg-2020	610	9	,	,	,
orr-bootleg-2020	610	10	we	-PRON-	PRP
orr-bootleg-2020	610	11	set	set	VBD
orr-bootleg-2020	610	12	the	the	DT
orr-bootleg-2020	610	13	hidden	hide	VBN
orr-bootleg-2020	610	14	dimension	dimension	NN
orr-bootleg-2020	610	15	to	to	IN
orr-bootleg-2020	610	16	256	256	CD
orr-bootleg-2020	610	17	,	,	,
orr-bootleg-2020	610	18	the	the	DT
orr-bootleg-2020	610	19	dimension	dimension	NN
orr-bootleg-2020	610	20	of	of	IN
orr-bootleg-2020	610	21	u	u	NNP
orr-bootleg-2020	610	22	to	to	IN
orr-bootleg-2020	610	23	256	256	CD
orr-bootleg-2020	610	24	,	,	,
orr-bootleg-2020	610	25	and	and	CC
orr-bootleg-2020	610	26	the	the	DT
orr-bootleg-2020	610	27	dimension	dimension	NN
orr-bootleg-2020	610	28	of	of	IN
orr-bootleg-2020	610	29	all	all	DT
orr-bootleg-2020	610	30	other	other	JJ
orr-bootleg-2020	610	31	type	type	NN
orr-bootleg-2020	610	32	and	and	CC
orr-bootleg-2020	610	33	relation	relation	NN
orr-bootleg-2020	610	34	embeddings	embedding	NNS
orr-bootleg-2020	610	35	to	to	IN
orr-bootleg-2020	610	36	128	128	CD
orr-bootleg-2020	610	37	.	.	.
orr-bootleg-2020	611	1	The	the	DT
orr-bootleg-2020	611	2	final	final	JJ
orr-bootleg-2020	611	3	differences	difference	NNS
orr-bootleg-2020	611	4	to	to	TO
orr-bootleg-2020	611	5	discuss	discuss	VB
orr-bootleg-2020	611	6	are	be	VBP
orr-bootleg-2020	611	7	between	between	IN
orr-bootleg-2020	611	8	the	the	DT
orr-bootleg-2020	611	9	benchmark	benchmark	JJ
orr-bootleg-2020	611	10	model	model	NN
orr-bootleg-2020	611	11	and	and	CC
orr-bootleg-2020	611	12	ablation	ablation	NN
orr-bootleg-2020	611	13	model	model	NN
orr-bootleg-2020	611	14	over	over	IN
orr-bootleg-2020	611	15	all	all	DT
orr-bootleg-2020	611	16	of	of	IN
orr-bootleg-2020	611	17	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	611	18	.	.	.
orr-bootleg-2020	612	1	To	to	TO
orr-bootleg-2020	612	2	make	make	VB
orr-bootleg-2020	612	3	the	the	DT
orr-bootleg-2020	612	4	best	good	JJS
orr-bootleg-2020	612	5	performing	perform	VBG
orr-bootleg-2020	612	6	model	model	NN
orr-bootleg-2020	612	7	for	for	IN
orr-bootleg-2020	612	8	benchmarks	benchmark	NNS
orr-bootleg-2020	612	9	,	,	,
orr-bootleg-2020	612	10	we	-PRON-	PRP
orr-bootleg-2020	612	11	add	add	VBP
orr-bootleg-2020	612	12	two	two	CD
orr-bootleg-2020	612	13	additional	additional	JJ
orr-bootleg-2020	612	14	additional	additional	JJ
orr-bootleg-2020	612	15	features	feature	NNS
orr-bootleg-2020	612	16	we	-PRON-	PRP
orr-bootleg-2020	612	17	found	find	VBD
orr-bootleg-2020	612	18	improved	improve	VBN
orr-bootleg-2020	612	19	performance	performance	NN
orr-bootleg-2020	612	20	:	:	:
orr-bootleg-2020	612	21	•	•	VB
orr-bootleg-2020	612	22	We	-PRON-	PRP
orr-bootleg-2020	612	23	use	use	VBP
orr-bootleg-2020	612	24	an	an	DT
orr-bootleg-2020	612	25	additional	additional	JJ
orr-bootleg-2020	612	26	KG2Ent	kg2ent	NN
orr-bootleg-2020	612	27	module	module	NN
orr-bootleg-2020	612	28	in	in	IN
orr-bootleg-2020	612	29	addition	addition	NN
orr-bootleg-2020	612	30	to	to	IN
orr-bootleg-2020	612	31	an	an	DT
orr-bootleg-2020	612	32	adjacency	adjacency	NN
orr-bootleg-2020	612	33	matrix	matrix	NN
orr-bootleg-2020	612	34	indicating	indicate	VBG
orr-bootleg-2020	612	35	if	if	IN
orr-bootleg-2020	612	36	two	two	CD
orr-bootleg-2020	612	37	entities	entity	NNS
orr-bootleg-2020	612	38	are	be	VBP
orr-bootleg-2020	612	39	connected	connect	VBN
orr-bootleg-2020	612	40	in	in	IN
orr-bootleg-2020	612	41	Wikidata	Wikidata	NNP
orr-bootleg-2020	612	42	.	.	.
orr-bootleg-2020	613	1	We	-PRON-	PRP
orr-bootleg-2020	613	2	add	add	VBP
orr-bootleg-2020	613	3	a	a	DT
orr-bootleg-2020	613	4	matrix	matrix	NN
orr-bootleg-2020	613	5	containing	contain	VBG
orr-bootleg-2020	613	6	the	the	DT
orr-bootleg-2020	613	7	log	log	NN
orr-bootleg-2020	613	8	of	of	IN
orr-bootleg-2020	613	9	the	the	DT
orr-bootleg-2020	613	10	number	number	NN
orr-bootleg-2020	613	11	of	of	IN
orr-bootleg-2020	613	12	times	time	NNS
orr-bootleg-2020	613	13	two	two	CD
orr-bootleg-2020	613	14	entities	entity	NNS
orr-bootleg-2020	613	15	occur	occur	VBP
orr-bootleg-2020	613	16	in	in	IN
orr-bootleg-2020	613	17	a	a	DT
orr-bootleg-2020	613	18	sentence	sentence	NN
orr-bootleg-2020	613	19	together	together	RB
orr-bootleg-2020	613	20	in	in	IN
orr-bootleg-2020	613	21	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	613	22	.	.	.
orr-bootleg-2020	614	1	If	if	IN
orr-bootleg-2020	614	2	they	-PRON-	PRP
orr-bootleg-2020	614	3	co	co	VBP
orr-bootleg-2020	614	4	-	-	VBP
orr-bootleg-2020	614	5	occur	occur	VBP
orr-bootleg-2020	614	6	less	less	JJR
orr-bootleg-2020	614	7	than	than	IN
orr-bootleg-2020	614	8	10	10	CD
orr-bootleg-2020	614	9	times	time	NNS
orr-bootleg-2020	614	10	,	,	,
orr-bootleg-2020	614	11	the	the	DT
orr-bootleg-2020	614	12	weight	weight	NN
orr-bootleg-2020	614	13	is	be	VBZ
orr-bootleg-2020	614	14	0	0	CD
orr-bootleg-2020	614	15	.	.	.
orr-bootleg-2020	615	1	We	-PRON-	PRP
orr-bootleg-2020	615	2	found	find	VBD
orr-bootleg-2020	615	3	this	this	DT
orr-bootleg-2020	615	4	helped	help	VBD
orr-bootleg-2020	615	5	the	the	DT
orr-bootleg-2020	615	6	model	model	NN
orr-bootleg-2020	615	7	better	well	RBR
orr-bootleg-2020	615	8	learn	learn	VB
orr-bootleg-2020	615	9	entity	entity	NN
orr-bootleg-2020	615	10	co	co	NNS
orr-bootleg-2020	615	11	-	-	NNS
orr-bootleg-2020	615	12	occurrences	occurrence	NNS
orr-bootleg-2020	615	13	from	from	IN
orr-bootleg-2020	615	14	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	615	15	.	.	.
orr-bootleg-2020	616	1	•	•	UH
orr-bootleg-2020	616	2	We	-PRON-	PRP
orr-bootleg-2020	616	3	allow	allow	VBP
orr-bootleg-2020	616	4	our	-PRON-	PRP$
orr-bootleg-2020	616	5	model	model	NN
orr-bootleg-2020	616	6	to	to	TO
orr-bootleg-2020	616	7	use	use	VB
orr-bootleg-2020	616	8	additional	additional	JJ
orr-bootleg-2020	616	9	entity	entity	NN
orr-bootleg-2020	616	10	-	-	HYPH
orr-bootleg-2020	616	11	based	base	VBN
orr-bootleg-2020	616	12	features	feature	NNS
orr-bootleg-2020	616	13	to	to	TO
orr-bootleg-2020	616	14	be	be	VB
orr-bootleg-2020	616	15	concatenated	concatenate	VBN
orr-bootleg-2020	616	16	into	into	IN
orr-bootleg-2020	616	17	our	-PRON-	PRP$
orr-bootleg-2020	616	18	final	final	JJ
orr-bootleg-2020	616	19	E	e	NN
orr-bootleg-2020	616	20	matrix	matrix	NN
orr-bootleg-2020	616	21	.	.	.
orr-bootleg-2020	617	1	We	-PRON-	PRP
orr-bootleg-2020	617	2	add	add	VBP
orr-bootleg-2020	617	3	two	two	CD
orr-bootleg-2020	617	4	features	feature	NNS
orr-bootleg-2020	617	5	.	.	.
orr-bootleg-2020	618	1	The	the	DT
orr-bootleg-2020	618	2	first	first	JJ
orr-bootleg-2020	618	3	is	be	VBZ
orr-bootleg-2020	618	4	the	the	DT
orr-bootleg-2020	618	5	average	average	JJ
orr-bootleg-2020	618	6	BERT	BERT	NNP
orr-bootleg-2020	618	7	WordPiece	WordPiece	NNP
orr-bootleg-2020	618	8	embeddings	embedding	NNS
orr-bootleg-2020	618	9	of	of	IN
orr-bootleg-2020	618	10	the	the	DT
orr-bootleg-2020	618	11	title	title	NN
orr-bootleg-2020	618	12	of	of	IN
orr-bootleg-2020	618	13	an	an	DT
orr-bootleg-2020	618	14	entity	entity	NN
orr-bootleg-2020	618	15	.	.	.
orr-bootleg-2020	619	1	This	this	DT
orr-bootleg-2020	619	2	is	be	VBZ
orr-bootleg-2020	619	3	11We	11We	NNP
orr-bootleg-2020	619	4	use	use	VB
orr-bootleg-2020	619	5	the	the	DT
orr-bootleg-2020	619	6	spaCy	spacy	ADD
orr-bootleg-2020	619	7	NER	ner	NN
orr-bootleg-2020	619	8	tagger	tagger	NN
orr-bootleg-2020	619	9	from	from	IN
orr-bootleg-2020	619	10	https://spacy.io/	https://spacy.io/	NNP
orr-bootleg-2020	619	11	20	20	CD
orr-bootleg-2020	619	12	https://spacy.io/	https://spacy.io/	NNP
orr-bootleg-2020	619	13	Table	table	NN
orr-bootleg-2020	619	14	9	9	CD
orr-bootleg-2020	619	15	:	:	:
orr-bootleg-2020	619	16	(	(	-LRB-
orr-bootleg-2020	619	17	top	top	NN
orr-bootleg-2020	619	18	)	)	-RRB-
orr-bootleg-2020	619	19	We	-PRON-	PRP
orr-bootleg-2020	619	20	compare	compare	VBP
orr-bootleg-2020	619	21	Bootleg	Bootleg	NNP
orr-bootleg-2020	619	22	to	to	IN
orr-bootleg-2020	619	23	a	a	DT
orr-bootleg-2020	619	24	BERT	BERT	NNP
orr-bootleg-2020	619	25	-	-	HYPH
orr-bootleg-2020	619	26	based	base	VBN
orr-bootleg-2020	619	27	NED	NED	NNP
orr-bootleg-2020	619	28	baseline	baseline	NN
orr-bootleg-2020	619	29	(	(	-LRB-
orr-bootleg-2020	619	30	NED	NED	NNP
orr-bootleg-2020	619	31	-	-	HYPH
orr-bootleg-2020	619	32	Base	Base	NNP
orr-bootleg-2020	619	33	)	)	-RRB-
orr-bootleg-2020	619	34	on	on	IN
orr-bootleg-2020	619	35	validation	validation	NN
orr-bootleg-2020	619	36	sets	set	NNS
orr-bootleg-2020	619	37	of	of	IN
orr-bootleg-2020	619	38	our	-PRON-	PRP$
orr-bootleg-2020	619	39	micro	micro	JJ
orr-bootleg-2020	619	40	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	619	41	dataset	dataset	NN
orr-bootleg-2020	619	42	and	and	CC
orr-bootleg-2020	619	43	ablate	ablate	VB
orr-bootleg-2020	619	44	Bootleg	Bootleg	NNP
orr-bootleg-2020	619	45	by	by	IN
orr-bootleg-2020	619	46	only	only	JJ
orr-bootleg-2020	619	47	training	training	NN
orr-bootleg-2020	619	48	with	with	IN
orr-bootleg-2020	619	49	entity	entity	NN
orr-bootleg-2020	619	50	,	,	,
orr-bootleg-2020	619	51	type	type	NN
orr-bootleg-2020	619	52	,	,	,
orr-bootleg-2020	619	53	or	or	CC
orr-bootleg-2020	619	54	knowledge	knowledge	NN
orr-bootleg-2020	619	55	graph	graph	NN
orr-bootleg-2020	619	56	data	datum	NNS
orr-bootleg-2020	619	57	.	.	.
orr-bootleg-2020	620	1	We	-PRON-	PRP
orr-bootleg-2020	620	2	further	further	RB
orr-bootleg-2020	620	3	ablate	ablate	NN
orr-bootleg-2020	620	4	(	(	-LRB-
orr-bootleg-2020	620	5	bottom	bottom	JJ
orr-bootleg-2020	620	6	8	8	CD
orr-bootleg-2020	620	7	rows	row	NNS
orr-bootleg-2020	620	8	)	)	-RRB-
orr-bootleg-2020	620	9	the	the	DT
orr-bootleg-2020	620	10	regularization	regularization	NN
orr-bootleg-2020	620	11	schemes	scheme	NNS
orr-bootleg-2020	620	12	for	for	IN
orr-bootleg-2020	620	13	the	the	DT
orr-bootleg-2020	620	14	entity	entity	NN
orr-bootleg-2020	620	15	embeddings	embedding	NNS
orr-bootleg-2020	620	16	for	for	IN
orr-bootleg-2020	620	17	Bootleg	Bootleg	NNP
orr-bootleg-2020	620	18	.	.	.
orr-bootleg-2020	621	1	Model	Model	NNP
orr-bootleg-2020	621	2	All	all	DT
orr-bootleg-2020	621	3	Entities	entity	NNS
orr-bootleg-2020	621	4	Torso	Torso	NNP
orr-bootleg-2020	621	5	Entities	Entities	NNPS
orr-bootleg-2020	621	6	Tail	Tail	NNP
orr-bootleg-2020	621	7	Entities	Entities	NNP
orr-bootleg-2020	621	8	Unseen	Unseen	NNP
orr-bootleg-2020	621	9	Entities	Entities	NNP
orr-bootleg-2020	621	10	NED	NED	NNP
orr-bootleg-2020	621	11	-	-	HYPH
orr-bootleg-2020	621	12	Base	Base	NNP
orr-bootleg-2020	621	13	90.2	90.2	CD
orr-bootleg-2020	621	14	91.6	91.6	CD
orr-bootleg-2020	621	15	50.5	50.5	CD
orr-bootleg-2020	621	16	21.5	21.5	CD
orr-bootleg-2020	621	17	Bootleg	Bootleg	NNP
orr-bootleg-2020	621	18	(	(	-LRB-
orr-bootleg-2020	621	19	Ent	ent	NN
orr-bootleg-2020	621	20	-	-	,
orr-bootleg-2020	621	21	only	only	RB
orr-bootleg-2020	621	22	)	)	-RRB-
orr-bootleg-2020	621	23	89.1	89.1	CD
orr-bootleg-2020	621	24	89.0	89.0	CD
orr-bootleg-2020	621	25	48.3	48.3	CD
orr-bootleg-2020	621	26	15.5	15.5	CD
orr-bootleg-2020	621	27	Bootleg	Bootleg	NNP
orr-bootleg-2020	621	28	(	(	-LRB-
orr-bootleg-2020	621	29	Type	Type	NNP
orr-bootleg-2020	621	30	-	-	HYPH
orr-bootleg-2020	621	31	only	only	RB
orr-bootleg-2020	621	32	)	)	-RRB-
orr-bootleg-2020	621	33	91.6	91.6	CD
orr-bootleg-2020	621	34	90.4	90.4	CD
orr-bootleg-2020	621	35	65.9	65.9	CD
orr-bootleg-2020	621	36	56.8	56.8	CD
orr-bootleg-2020	621	37	Bootleg	Bootleg	NNP
orr-bootleg-2020	621	38	(	(	-LRB-
orr-bootleg-2020	621	39	KG	KG	NNP
orr-bootleg-2020	621	40	-	-	HYPH
orr-bootleg-2020	621	41	only	only	RB
orr-bootleg-2020	621	42	)	)	-RRB-
orr-bootleg-2020	621	43	91.8	91.8	CD
orr-bootleg-2020	621	44	90.8	90.8	CD
orr-bootleg-2020	621	45	65.3	65.3	CD
orr-bootleg-2020	621	46	58.6	58.6	CD
orr-bootleg-2020	621	47	Bootleg	Bootleg	NNP
orr-bootleg-2020	621	48	(	(	-LRB-
orr-bootleg-2020	621	49	p(e	p(e	NNP
orr-bootleg-2020	621	50	)	)	-RRB-
orr-bootleg-2020	621	51	=	=	NFP
orr-bootleg-2020	621	52	0	0	CD
orr-bootleg-2020	621	53	%	%	NN
orr-bootleg-2020	621	54	)	)	-RRB-
orr-bootleg-2020	621	55	92.5	92.5	CD
orr-bootleg-2020	621	56	92.3	92.3	CD
orr-bootleg-2020	621	57	67.7	67.7	CD
orr-bootleg-2020	621	58	48.6	48.6	CD
orr-bootleg-2020	621	59	Bootleg	Bootleg	NNP
orr-bootleg-2020	621	60	(	(	-LRB-
orr-bootleg-2020	621	61	p(e	p(e	NNP
orr-bootleg-2020	621	62	)	)	-RRB-
orr-bootleg-2020	621	63	=	=	SYM
orr-bootleg-2020	621	64	20	20	CD
orr-bootleg-2020	621	65	%	%	NN
orr-bootleg-2020	621	66	)	)	-RRB-
orr-bootleg-2020	621	67	92.8	92.8	CD
orr-bootleg-2020	621	68	92.5	92.5	CD
orr-bootleg-2020	621	69	68.9	68.9	CD
orr-bootleg-2020	621	70	52.5	52.5	CD
orr-bootleg-2020	621	71	Bootleg	Bootleg	NNP
orr-bootleg-2020	621	72	(	(	-LRB-
orr-bootleg-2020	621	73	p(e	p(e	NNP
orr-bootleg-2020	621	74	)	)	-RRB-
orr-bootleg-2020	621	75	=	=	NFP
orr-bootleg-2020	621	76	50	50	CD
orr-bootleg-2020	621	77	%	%	NN
orr-bootleg-2020	621	78	)	)	-RRB-
orr-bootleg-2020	621	79	92.9	92.9	CD
orr-bootleg-2020	621	80	92.7	92.7	CD
orr-bootleg-2020	621	81	70.1	70.1	CD
orr-bootleg-2020	621	82	57.7	57.7	CD
orr-bootleg-2020	621	83	Bootleg	Bootleg	NNP
orr-bootleg-2020	621	84	(	(	-LRB-
orr-bootleg-2020	621	85	p(e	p(e	NNP
orr-bootleg-2020	621	86	)	)	-RRB-
orr-bootleg-2020	621	87	=	=	NFP
orr-bootleg-2020	621	88	80	80	CD
orr-bootleg-2020	621	89	%	%	NN
orr-bootleg-2020	621	90	)	)	-RRB-
orr-bootleg-2020	621	91	92.8	92.8	CD
orr-bootleg-2020	621	92	92.2	92.2	CD
orr-bootleg-2020	621	93	69.5	69.5	CD
orr-bootleg-2020	621	94	59.9	59.9	CD
orr-bootleg-2020	621	95	Bootleg	Bootleg	NNP
orr-bootleg-2020	621	96	(	(	-LRB-
orr-bootleg-2020	621	97	InvPopLog	InvPopLog	NNP
orr-bootleg-2020	621	98	)	)	-RRB-
orr-bootleg-2020	621	99	92.7	92.7	CD
orr-bootleg-2020	621	100	91.9	91.9	CD
orr-bootleg-2020	621	101	69.7	69.7	CD
orr-bootleg-2020	621	102	61.1	61.1	CD
orr-bootleg-2020	621	103	Bootleg	Bootleg	NNP
orr-bootleg-2020	621	104	(	(	-LRB-
orr-bootleg-2020	621	105	InvPopPow	InvPopPow	NNP
orr-bootleg-2020	621	106	)	)	-RRB-
orr-bootleg-2020	621	107	92.8	92.8	CD
orr-bootleg-2020	621	108	92.3	92.3	CD
orr-bootleg-2020	621	109	70.5	70.5	CD
orr-bootleg-2020	621	110	62.2	62.2	CD
orr-bootleg-2020	621	111	Bootleg	Bootleg	NNP
orr-bootleg-2020	621	112	(	(	-LRB-
orr-bootleg-2020	621	113	InvPopLin	InvPopLin	NNP
orr-bootleg-2020	621	114	)	)	-RRB-
orr-bootleg-2020	621	115	92.6	92.6	CD
orr-bootleg-2020	621	116	91.8	91.8	CD
orr-bootleg-2020	621	117	69.7	69.7	CD
orr-bootleg-2020	621	118	61.0	61.0	CD
orr-bootleg-2020	621	119	Bootleg	Bootleg	NNP
orr-bootleg-2020	621	120	(	(	-LRB-
orr-bootleg-2020	621	121	PopPow	PopPow	NNP
orr-bootleg-2020	621	122	)	)	-RRB-
orr-bootleg-2020	621	123	92.9	92.9	CD
orr-bootleg-2020	621	124	92.5	92.5	CD
orr-bootleg-2020	621	125	68.9	68.9	CD
orr-bootleg-2020	621	126	52.4	52.4	CD
orr-bootleg-2020	621	127	#	#	$
orr-bootleg-2020	621	128	Mentions	Mentions	NNPS
orr-bootleg-2020	621	129	96,237	96,237	CD
orr-bootleg-2020	621	130	37,077	37,077	CD
orr-bootleg-2020	621	131	11,087	11,087	CD
orr-bootleg-2020	621	132	2,810	2,810	CD
orr-bootleg-2020	621	133	similar	similar	JJ
orr-bootleg-2020	621	134	to	to	IN
orr-bootleg-2020	621	135	improving	improve	VBG
orr-bootleg-2020	621	136	tail	tail	NN
orr-bootleg-2020	621	137	generalization	generalization	NN
orr-bootleg-2020	621	138	by	by	IN
orr-bootleg-2020	621	139	embedding	embed	VBG
orr-bootleg-2020	621	140	a	a	DT
orr-bootleg-2020	621	141	word	word	NN
orr-bootleg-2020	621	142	definition	definition	NN
orr-bootleg-2020	621	143	in	in	IN
orr-bootleg-2020	621	144	word	word	NN
orr-bootleg-2020	621	145	sense	sense	NN
orr-bootleg-2020	621	146	disambiguation	disambiguation	NNP
orr-bootleg-2020	621	147	[	[	-LRB-
orr-bootleg-2020	621	148	5	5	CD
orr-bootleg-2020	621	149	]	]	-RRB-
orr-bootleg-2020	621	150	.	.	.
orr-bootleg-2020	622	1	This	this	DT
orr-bootleg-2020	622	2	allows	allow	VBZ
orr-bootleg-2020	622	3	the	the	DT
orr-bootleg-2020	622	4	model	model	NN
orr-bootleg-2020	622	5	to	to	TO
orr-bootleg-2020	622	6	better	well	RBR
orr-bootleg-2020	622	7	capture	capture	VB
orr-bootleg-2020	622	8	textual	textual	JJ
orr-bootleg-2020	622	9	cues	cue	NNS
orr-bootleg-2020	622	10	indicating	indicate	VBG
orr-bootleg-2020	622	11	the	the	DT
orr-bootleg-2020	622	12	correct	correct	JJ
orr-bootleg-2020	622	13	entity	entity	NN
orr-bootleg-2020	622	14	.	.	.
orr-bootleg-2020	623	1	We	-PRON-	PRP
orr-bootleg-2020	623	2	also	also	RB
orr-bootleg-2020	623	3	append	append	VBP
orr-bootleg-2020	623	4	a	a	DT
orr-bootleg-2020	623	5	1-dimensional	1-dimensional	CD
orr-bootleg-2020	623	6	feature	feature	NN
orr-bootleg-2020	623	7	of	of	IN
orr-bootleg-2020	623	8	how	how	WRB
orr-bootleg-2020	623	9	many	many	JJ
orr-bootleg-2020	623	10	other	other	JJ
orr-bootleg-2020	623	11	entities	entity	NNS
orr-bootleg-2020	623	12	in	in	IN
orr-bootleg-2020	623	13	the	the	DT
orr-bootleg-2020	623	14	sentence	sentence	NN
orr-bootleg-2020	623	15	appear	appear	VBP
orr-bootleg-2020	623	16	on	on	IN
orr-bootleg-2020	623	17	the	the	DT
orr-bootleg-2020	623	18	entity	entity	NN
orr-bootleg-2020	623	19	’s	’s	POS
orr-bootleg-2020	623	20	Wikipedia	wikipedia	NN
orr-bootleg-2020	623	21	page	page	NN
orr-bootleg-2020	623	22	.	.	.
orr-bootleg-2020	624	1	This	this	DT
orr-bootleg-2020	624	2	increases	increase	VBZ
orr-bootleg-2020	624	3	the	the	DT
orr-bootleg-2020	624	4	likelihood	likelihood	NN
orr-bootleg-2020	624	5	of	of	IN
orr-bootleg-2020	624	6	an	an	DT
orr-bootleg-2020	624	7	entity	entity	NN
orr-bootleg-2020	624	8	that	that	WDT
orr-bootleg-2020	624	9	has	have	VBZ
orr-bootleg-2020	624	10	more	more	JJR
orr-bootleg-2020	624	11	connection	connection	NN
orr-bootleg-2020	624	12	to	to	IN
orr-bootleg-2020	624	13	other	other	JJ
orr-bootleg-2020	624	14	candidates	candidate	NNS
orr-bootleg-2020	624	15	in	in	IN
orr-bootleg-2020	624	16	the	the	DT
orr-bootleg-2020	624	17	sentence	sentence	NN
orr-bootleg-2020	624	18	.	.	.
orr-bootleg-2020	625	1	We	-PRON-	PRP
orr-bootleg-2020	625	2	empirically	empirically	RB
orr-bootleg-2020	625	3	find	find	VBP
orr-bootleg-2020	625	4	that	that	IN
orr-bootleg-2020	625	5	using	use	VBG
orr-bootleg-2020	625	6	the	the	DT
orr-bootleg-2020	625	7	page	page	NN
orr-bootleg-2020	625	8	co	co	NNS
orr-bootleg-2020	625	9	-	-	NNS
orr-bootleg-2020	625	10	occurrences	occurrence	NNS
orr-bootleg-2020	625	11	as	as	IN
orr-bootleg-2020	625	12	an	an	DT
orr-bootleg-2020	625	13	entity	entity	NN
orr-bootleg-2020	625	14	feature	feature	NN
orr-bootleg-2020	625	15	rather	rather	RB
orr-bootleg-2020	625	16	than	than	IN
orr-bootleg-2020	625	17	as	as	IN
orr-bootleg-2020	625	18	a	a	DT
orr-bootleg-2020	625	19	KG2Ent	kg2ent	NN
orr-bootleg-2020	625	20	module	module	NN
orr-bootleg-2020	625	21	performs	perform	VBZ
orr-bootleg-2020	625	22	similarly	similarly	RB
orr-bootleg-2020	625	23	and	and	CC
orr-bootleg-2020	625	24	reduces	reduce	VBZ
orr-bootleg-2020	625	25	the	the	DT
orr-bootleg-2020	625	26	runtime	runtime	NN
orr-bootleg-2020	625	27	.	.	.
orr-bootleg-2020	626	1	Further	further	RB
orr-bootleg-2020	626	2	,	,	,
orr-bootleg-2020	626	3	our	-PRON-	PRP$
orr-bootleg-2020	626	4	benchmark	benchmark	JJ
orr-bootleg-2020	626	5	model	model	NN
orr-bootleg-2020	626	6	uses	use	VBZ
orr-bootleg-2020	626	7	a	a	DT
orr-bootleg-2020	626	8	fixed	fix	VBN
orr-bootleg-2020	626	9	regularization	regularization	NN
orr-bootleg-2020	626	10	scheme	scheme	NN
orr-bootleg-2020	626	11	of	of	IN
orr-bootleg-2020	626	12	80	80	CD
orr-bootleg-2020	626	13	%	%	NN
orr-bootleg-2020	626	14	which	which	WDT
orr-bootleg-2020	626	15	did	do	VBD
orr-bootleg-2020	626	16	not	not	RB
orr-bootleg-2020	626	17	hurt	hurt	VB
orr-bootleg-2020	626	18	benchmark	benchmark	JJ
orr-bootleg-2020	626	19	performance	performance	NN
orr-bootleg-2020	626	20	and	and	CC
orr-bootleg-2020	626	21	training	training	NN
orr-bootleg-2020	626	22	was	be	VBD
orr-bootleg-2020	626	23	marginally	marginally	RB
orr-bootleg-2020	626	24	faster	fast	JJR
orr-bootleg-2020	626	25	than	than	IN
orr-bootleg-2020	626	26	the	the	DT
orr-bootleg-2020	626	27	inverse	inverse	NN
orr-bootleg-2020	626	28	popularity	popularity	NN
orr-bootleg-2020	626	29	scheme	scheme	NN
orr-bootleg-2020	626	30	.	.	.
orr-bootleg-2020	627	1	We	-PRON-	PRP
orr-bootleg-2020	627	2	did	do	VBD
orr-bootleg-2020	627	3	not	not	RB
orr-bootleg-2020	627	4	use	use	VB
orr-bootleg-2020	627	5	these	these	DT
orr-bootleg-2020	627	6	features	feature	NNS
orr-bootleg-2020	627	7	for	for	IN
orr-bootleg-2020	627	8	ablations	ablation	NNS
orr-bootleg-2020	627	9	as	as	IN
orr-bootleg-2020	627	10	we	-PRON-	PRP
orr-bootleg-2020	627	11	wanted	want	VBD
orr-bootleg-2020	627	12	a	a	DT
orr-bootleg-2020	627	13	clean	clean	JJ
orr-bootleg-2020	627	14	study	study	NN
orr-bootleg-2020	627	15	of	of	IN
orr-bootleg-2020	627	16	the	the	DT
orr-bootleg-2020	627	17	model	model	NN
orr-bootleg-2020	627	18	components	component	NNS
orr-bootleg-2020	627	19	as	as	IN
orr-bootleg-2020	627	20	described	describe	VBN
orr-bootleg-2020	627	21	in	in	IN
orr-bootleg-2020	627	22	Section	section	NN
orr-bootleg-2020	627	23	3	3	CD
orr-bootleg-2020	627	24	with	with	IN
orr-bootleg-2020	627	25	respect	respect	NN
orr-bootleg-2020	627	26	to	to	IN
orr-bootleg-2020	627	27	the	the	DT
orr-bootleg-2020	627	28	reasoning	reasoning	NN
orr-bootleg-2020	627	29	patterns	pattern	NNS
orr-bootleg-2020	627	30	.	.	.
orr-bootleg-2020	628	1	Training	training	NN
orr-bootleg-2020	628	2	We	-PRON-	PRP
orr-bootleg-2020	628	3	initialize	initialize	VBP
orr-bootleg-2020	628	4	all	all	DT
orr-bootleg-2020	628	5	entity	entity	NN
orr-bootleg-2020	628	6	embeddings	embedding	NNS
orr-bootleg-2020	628	7	to	to	IN
orr-bootleg-2020	628	8	the	the	DT
orr-bootleg-2020	628	9	same	same	JJ
orr-bootleg-2020	628	10	vector	vector	NN
orr-bootleg-2020	628	11	to	to	TO
orr-bootleg-2020	628	12	reduce	reduce	VB
orr-bootleg-2020	628	13	the	the	DT
orr-bootleg-2020	628	14	impact	impact	NN
orr-bootleg-2020	628	15	of	of	IN
orr-bootleg-2020	628	16	noise	noise	NN
orr-bootleg-2020	628	17	from	from	IN
orr-bootleg-2020	628	18	unseen	unseen	JJ
orr-bootleg-2020	628	19	entities	entity	NNS
orr-bootleg-2020	628	20	receiving	receive	VBG
orr-bootleg-2020	628	21	different	different	JJ
orr-bootleg-2020	628	22	random	random	JJ
orr-bootleg-2020	628	23	embeddings	embedding	NNS
orr-bootleg-2020	628	24	.	.	.
orr-bootleg-2020	629	1	We	-PRON-	PRP
orr-bootleg-2020	629	2	use	use	VBP
orr-bootleg-2020	629	3	the	the	DT
orr-bootleg-2020	629	4	Adam	Adam	NNP
orr-bootleg-2020	629	5	optimizer	optimizer	NN
orr-bootleg-2020	629	6	[	[	-LRB-
orr-bootleg-2020	629	7	28	28	CD
orr-bootleg-2020	629	8	]	]	-RRB-
orr-bootleg-2020	629	9	with	with	IN
orr-bootleg-2020	629	10	a	a	DT
orr-bootleg-2020	629	11	learning	learning	NN
orr-bootleg-2020	629	12	rate	rate	NN
orr-bootleg-2020	629	13	of	of	IN
orr-bootleg-2020	629	14	0.0001	0.0001	CD
orr-bootleg-2020	629	15	and	and	CC
orr-bootleg-2020	629	16	a	a	DT
orr-bootleg-2020	629	17	dropout	dropout	NN
orr-bootleg-2020	629	18	of	of	IN
orr-bootleg-2020	629	19	0.1	0.1	CD
orr-bootleg-2020	629	20	in	in	IN
orr-bootleg-2020	629	21	all	all	DT
orr-bootleg-2020	629	22	feedforward	feedforward	NN
orr-bootleg-2020	629	23	layers	layer	NNS
orr-bootleg-2020	629	24	,	,	,
orr-bootleg-2020	629	25	16	16	CD
orr-bootleg-2020	629	26	heads	head	NNS
orr-bootleg-2020	629	27	in	in	IN
orr-bootleg-2020	629	28	our	-PRON-	PRP$
orr-bootleg-2020	629	29	attention	attention	NN
orr-bootleg-2020	629	30	modules	module	NNS
orr-bootleg-2020	629	31	,	,	,
orr-bootleg-2020	629	32	and	and	CC
orr-bootleg-2020	629	33	we	-PRON-	PRP
orr-bootleg-2020	629	34	freeze	freeze	VBP
orr-bootleg-2020	629	35	the	the	DT
orr-bootleg-2020	629	36	BERT	BERT	NNP
orr-bootleg-2020	629	37	encoder	encoder	NN
orr-bootleg-2020	629	38	stack	stack	NN
orr-bootleg-2020	629	39	.	.	.
orr-bootleg-2020	630	1	Note	note	VB
orr-bootleg-2020	630	2	for	for	IN
orr-bootleg-2020	630	3	the	the	DT
orr-bootleg-2020	630	4	NED	NED	NNP
orr-bootleg-2020	630	5	-	-	HYPH
orr-bootleg-2020	630	6	Base	Base	NNP
orr-bootleg-2020	630	7	model	model	NN
orr-bootleg-2020	630	8	,	,	,
orr-bootleg-2020	630	9	we	-PRON-	PRP
orr-bootleg-2020	630	10	do	do	VBP
orr-bootleg-2020	630	11	not	not	RB
orr-bootleg-2020	630	12	freeze	freeze	VB
orr-bootleg-2020	630	13	the	the	DT
orr-bootleg-2020	630	14	encoder	encoder	NN
orr-bootleg-2020	630	15	stack	stack	NN
orr-bootleg-2020	630	16	to	to	TO
orr-bootleg-2020	630	17	be	be	VB
orr-bootleg-2020	630	18	consistent	consistent	JJ
orr-bootleg-2020	630	19	with	with	IN
orr-bootleg-2020	630	20	Févry	Févry	NNP
orr-bootleg-2020	630	21	et	et	NNP
orr-bootleg-2020	630	22	al	al	NNP
orr-bootleg-2020	630	23	.	.	.
orr-bootleg-2020	631	1	[	[	-LRB-
orr-bootleg-2020	631	2	16	16	CD
orr-bootleg-2020	631	3	]	]	-RRB-
orr-bootleg-2020	631	4	.	.	.
orr-bootleg-2020	632	1	For	for	IN
orr-bootleg-2020	632	2	the	the	DT
orr-bootleg-2020	632	3	models	model	NNS
orr-bootleg-2020	632	4	trained	train	VBN
orr-bootleg-2020	632	5	on	on	IN
orr-bootleg-2020	632	6	all	all	DT
orr-bootleg-2020	632	7	of	of	IN
orr-bootleg-2020	632	8	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	632	9	,	,	,
orr-bootleg-2020	632	10	we	-PRON-	PRP
orr-bootleg-2020	632	11	use	use	VBP
orr-bootleg-2020	632	12	a	a	DT
orr-bootleg-2020	632	13	batch	batch	NN
orr-bootleg-2020	632	14	size	size	NN
orr-bootleg-2020	632	15	of	of	IN
orr-bootleg-2020	632	16	512	512	CD
orr-bootleg-2020	632	17	and	and	CC
orr-bootleg-2020	632	18	train	train	VB
orr-bootleg-2020	632	19	for	for	IN
orr-bootleg-2020	632	20	1	1	CD
orr-bootleg-2020	632	21	epoch	epoch	NN
orr-bootleg-2020	632	22	for	for	IN
orr-bootleg-2020	632	23	the	the	DT
orr-bootleg-2020	632	24	benchmark	benchmark	JJ
orr-bootleg-2020	632	25	model	model	NN
orr-bootleg-2020	632	26	and	and	CC
orr-bootleg-2020	632	27	2	2	CD
orr-bootleg-2020	632	28	epochs	epoch	NNS
orr-bootleg-2020	632	29	for	for	IN
orr-bootleg-2020	632	30	the	the	DT
orr-bootleg-2020	632	31	ablation	ablation	NN
orr-bootleg-2020	632	32	models	model	NNS
orr-bootleg-2020	632	33	on	on	IN
orr-bootleg-2020	632	34	8	8	CD
orr-bootleg-2020	632	35	NVIDIA	NVIDIA	NNP
orr-bootleg-2020	632	36	V100	V100	NNP
orr-bootleg-2020	632	37	GPUs	gpu	NNS
orr-bootleg-2020	632	38	.	.	.
orr-bootleg-2020	633	1	For	for	IN
orr-bootleg-2020	633	2	our	-PRON-	PRP$
orr-bootleg-2020	633	3	micro	micro	JJ
orr-bootleg-2020	633	4	data	data	NNP
orr-bootleg-2020	633	5	model	model	NN
orr-bootleg-2020	633	6	,	,	,
orr-bootleg-2020	633	7	we	-PRON-	PRP
orr-bootleg-2020	633	8	use	use	VBP
orr-bootleg-2020	633	9	a	a	DT
orr-bootleg-2020	633	10	batch	batch	NN
orr-bootleg-2020	633	11	size	size	NN
orr-bootleg-2020	633	12	of	of	IN
orr-bootleg-2020	633	13	96	96	CD
orr-bootleg-2020	633	14	and	and	CC
orr-bootleg-2020	633	15	train	train	NN
orr-bootleg-2020	633	16	for	for	IN
orr-bootleg-2020	633	17	8	8	CD
orr-bootleg-2020	633	18	epochs	epoch	NNS
orr-bootleg-2020	633	19	on	on	IN
orr-bootleg-2020	633	20	a	a	DT
orr-bootleg-2020	633	21	NVIDIA	nvidia	NN
orr-bootleg-2020	633	22	P100	P100	NNP
orr-bootleg-2020	633	23	GPU	GPU	NNP
orr-bootleg-2020	633	24	.	.	.
orr-bootleg-2020	634	1	B.3	B.3	NNP
orr-bootleg-2020	634	2	Extended	Extended	NNP
orr-bootleg-2020	634	3	Ablation	ablation	NN
orr-bootleg-2020	634	4	Results	Results	NNPS
orr-bootleg-2020	634	5	Ablation	Ablation	NNP
orr-bootleg-2020	634	6	Model	Model	NNP
orr-bootleg-2020	634	7	Size	Size	NNP
orr-bootleg-2020	634	8	Table	table	NN
orr-bootleg-2020	634	9	10	10	CD
orr-bootleg-2020	634	10	reports	report	VBZ
orr-bootleg-2020	634	11	the	the	DT
orr-bootleg-2020	634	12	model	model	NN
orr-bootleg-2020	634	13	sizes	size	NNS
orr-bootleg-2020	634	14	of	of	IN
orr-bootleg-2020	634	15	each	each	DT
orr-bootleg-2020	634	16	of	of	IN
orr-bootleg-2020	634	17	the	the	DT
orr-bootleg-2020	634	18	five	five	CD
orr-bootleg-2020	634	19	ablation	ablation	NN
orr-bootleg-2020	634	20	models	model	NNS
orr-bootleg-2020	634	21	from	from	IN
orr-bootleg-2020	634	22	Table	table	NN
orr-bootleg-2020	634	23	2	2	CD
orr-bootleg-2020	634	24	.	.	.
orr-bootleg-2020	635	1	As	as	IN
orr-bootleg-2020	635	2	we	-PRON-	PRP
orr-bootleg-2020	635	3	finetuned	finetune	VBD
orr-bootleg-2020	635	4	the	the	DT
orr-bootleg-2020	635	5	BERT	BERT	NNP
orr-bootleg-2020	635	6	language	language	NN
orr-bootleg-2020	635	7	model	model	NN
orr-bootleg-2020	635	8	in	in	IN
orr-bootleg-2020	635	9	NED	NED	NNP
orr-bootleg-2020	635	10	-	-	HYPH
orr-bootleg-2020	635	11	Base	Base	NNP
orr-bootleg-2020	635	12	(	(	-LRB-
orr-bootleg-2020	635	13	to	to	TO
orr-bootleg-2020	635	14	be	be	VB
orr-bootleg-2020	635	15	consistent	consistent	JJ
orr-bootleg-2020	635	16	with	with	IN
orr-bootleg-2020	635	17	Févry	Févry	NNP
orr-bootleg-2020	635	18	et	et	NNP
orr-bootleg-2020	635	19	al	al	NNP
orr-bootleg-2020	635	20	.	.	.
orr-bootleg-2020	636	1	[	[	-LRB-
orr-bootleg-2020	636	2	16	16	CD
orr-bootleg-2020	636	3	]	]	-RRB-
orr-bootleg-2020	636	4	)	)	-RRB-
orr-bootleg-2020	636	5	but	but	CC
orr-bootleg-2020	636	6	do	do	VBP
orr-bootleg-2020	636	7	not	not	RB
orr-bootleg-2020	636	8	do	do	VB
orr-bootleg-2020	636	9	so	so	RB
orr-bootleg-2020	636	10	in	in	IN
orr-bootleg-2020	636	11	Bootleg	Bootleg	NNP
orr-bootleg-2020	636	12	,	,	,
orr-bootleg-2020	636	13	we	-PRON-	PRP
orr-bootleg-2020	636	14	do	do	VBP
orr-bootleg-2020	636	15	not	not	RB
orr-bootleg-2020	636	16	count	count	VB
orr-bootleg-2020	636	17	the	the	DT
orr-bootleg-2020	636	18	BERT	BERT	NNP
orr-bootleg-2020	636	19	parameters	parameter	NNS
orr-bootleg-2020	636	20	in	in	IN
orr-bootleg-2020	636	21	our	-PRON-	PRP$
orr-bootleg-2020	636	22	reported	report	VBN
orr-bootleg-2020	636	23	sizes	size	NNS
orr-bootleg-2020	636	24	to	to	TO
orr-bootleg-2020	636	25	be	be	VB
orr-bootleg-2020	636	26	comparable	comparable	JJ
orr-bootleg-2020	636	27	.	.	.
orr-bootleg-2020	637	1	Regularization	regularization	NN
orr-bootleg-2020	637	2	We	-PRON-	PRP
orr-bootleg-2020	637	3	now	now	RB
orr-bootleg-2020	637	4	present	present	VBP
orr-bootleg-2020	637	5	the	the	DT
orr-bootleg-2020	637	6	extended	extended	JJ
orr-bootleg-2020	637	7	results	result	NNS
orr-bootleg-2020	637	8	of	of	IN
orr-bootleg-2020	637	9	our	-PRON-	PRP$
orr-bootleg-2020	637	10	regularization	regularization	NN
orr-bootleg-2020	637	11	and	and	CC
orr-bootleg-2020	637	12	weak	weak	JJ
orr-bootleg-2020	637	13	labelling	labelling	NN
orr-bootleg-2020	637	14	ablations	ablation	NNS
orr-bootleg-2020	637	15	over	over	IN
orr-bootleg-2020	637	16	our	-PRON-	PRP$
orr-bootleg-2020	637	17	representative	representative	JJ
orr-bootleg-2020	637	18	micro	micro	NNP
orr-bootleg-2020	637	19	dataset	dataset	NNP
orr-bootleg-2020	637	20	.	.	.
orr-bootleg-2020	638	1	Table	table	NN
orr-bootleg-2020	638	2	9	9	CD
orr-bootleg-2020	638	3	gives	give	VBZ
orr-bootleg-2020	638	4	full	full	JJ
orr-bootleg-2020	638	5	ablations	ablation	NNS
orr-bootleg-2020	638	6	over	over	IN
orr-bootleg-2020	638	7	a	a	DT
orr-bootleg-2020	638	8	variety	variety	NN
orr-bootleg-2020	638	9	of	of	IN
orr-bootleg-2020	638	10	regularization	regularization	NN
orr-bootleg-2020	638	11	techniques	technique	NNS
orr-bootleg-2020	638	12	.	.	.
orr-bootleg-2020	639	1	As	as	IN
orr-bootleg-2020	639	2	in	in	IN
orr-bootleg-2020	639	3	Table	table	NN
orr-bootleg-2020	639	4	2	2	CD
orr-bootleg-2020	639	5	,	,	,
orr-bootleg-2020	639	6	we	-PRON-	PRP
orr-bootleg-2020	639	7	include	include	VBP
orr-bootleg-2020	639	8	results	result	NNS
orr-bootleg-2020	639	9	from	from	IN
orr-bootleg-2020	639	10	models	model	NNS
orr-bootleg-2020	639	11	using	use	VBG
orr-bootleg-2020	639	12	only	only	RB
orr-bootleg-2020	639	13	the	the	DT
orr-bootleg-2020	639	14	entity	entity	NN
orr-bootleg-2020	639	15	,	,	,
orr-bootleg-2020	639	16	type	type	NN
orr-bootleg-2020	639	17	,	,	,
orr-bootleg-2020	639	18	or	or	CC
orr-bootleg-2020	639	19	relation	relation	NN
orr-bootleg-2020	639	20	information	information	NN
orr-bootleg-2020	639	21	,	,	,
orr-bootleg-2020	639	22	in	in	IN
orr-bootleg-2020	639	23	addition	addition	NN
orr-bootleg-2020	639	24	to	to	IN
orr-bootleg-2020	639	25	the	the	DT
orr-bootleg-2020	639	26	BERT	BERT	NNP
orr-bootleg-2020	639	27	and	and	CC
orr-bootleg-2020	639	28	Bootleg	Bootleg	NNP
orr-bootleg-2020	639	29	models	model	NNS
orr-bootleg-2020	639	30	.	.	.
orr-bootleg-2020	640	1	21	21	CD
orr-bootleg-2020	640	2	Table	table	NN
orr-bootleg-2020	640	3	10	10	CD
orr-bootleg-2020	640	4	:	:	:
orr-bootleg-2020	640	5	We	-PRON-	PRP
orr-bootleg-2020	640	6	report	report	VBP
orr-bootleg-2020	640	7	the	the	DT
orr-bootleg-2020	640	8	model	model	NN
orr-bootleg-2020	640	9	sizes	size	NNS
orr-bootleg-2020	640	10	in	in	IN
orr-bootleg-2020	640	11	MB	mb	NN
orr-bootleg-2020	640	12	of	of	IN
orr-bootleg-2020	640	13	each	each	DT
orr-bootleg-2020	640	14	of	of	IN
orr-bootleg-2020	640	15	the	the	DT
orr-bootleg-2020	640	16	five	five	CD
orr-bootleg-2020	640	17	ablation	ablation	NN
orr-bootleg-2020	640	18	models	model	NNS
orr-bootleg-2020	640	19	:	:	:
orr-bootleg-2020	640	20	NED	NED	NNP
orr-bootleg-2020	640	21	-	-	HYPH
orr-bootleg-2020	640	22	Base	Base	NNP
orr-bootleg-2020	640	23	,	,	,
orr-bootleg-2020	640	24	Bootleg	Bootleg	NNP
orr-bootleg-2020	640	25	,	,	,
orr-bootleg-2020	640	26	Bootleg	Bootleg	NNP
orr-bootleg-2020	640	27	(	(	-LRB-
orr-bootleg-2020	640	28	Ent	Ent	NNP
orr-bootleg-2020	640	29	-	-	HYPH
orr-bootleg-2020	640	30	Only	only	RB
orr-bootleg-2020	640	31	)	)	-RRB-
orr-bootleg-2020	640	32	,	,	,
orr-bootleg-2020	640	33	Bootleg	Bootleg	NNP
orr-bootleg-2020	640	34	(	(	-LRB-
orr-bootleg-2020	640	35	KG	KG	NNP
orr-bootleg-2020	640	36	-	-	HYPH
orr-bootleg-2020	640	37	Only	only	RB
orr-bootleg-2020	640	38	)	)	-RRB-
orr-bootleg-2020	640	39	,	,	,
orr-bootleg-2020	640	40	and	and	CC
orr-bootleg-2020	640	41	Bootleg	Bootleg	NNP
orr-bootleg-2020	640	42	(	(	-LRB-
orr-bootleg-2020	640	43	Type	Type	NNP
orr-bootleg-2020	640	44	-	-	HYPH
orr-bootleg-2020	640	45	Only	only	RB
orr-bootleg-2020	640	46	)	)	-RRB-
orr-bootleg-2020	640	47	.	.	.
orr-bootleg-2020	641	1	Model	Model	NNP
orr-bootleg-2020	641	2	NED	NED	NNP
orr-bootleg-2020	641	3	-	-	HYPH
orr-bootleg-2020	641	4	Base	Base	NNP
orr-bootleg-2020	641	5	Bootleg	Bootleg	NNP
orr-bootleg-2020	641	6	Ent	Ent	NNP
orr-bootleg-2020	641	7	-	-	HYPH
orr-bootleg-2020	641	8	Only	Only	NNP
orr-bootleg-2020	641	9	Type	type	NN
orr-bootleg-2020	641	10	-	-	HYPH
orr-bootleg-2020	641	11	Only	only	RB
orr-bootleg-2020	641	12	KG	kg	NN
orr-bootleg-2020	641	13	-	-	HYPH
orr-bootleg-2020	641	14	Only	only	NN
orr-bootleg-2020	641	15	Embedding	embedding	NN
orr-bootleg-2020	641	16	Size	size	NN
orr-bootleg-2020	641	17	(	(	-LRB-
orr-bootleg-2020	641	18	MB	MB	NNP
orr-bootleg-2020	641	19	)	)	-RRB-
orr-bootleg-2020	641	20	5,186	5,186	CD
orr-bootleg-2020	641	21	5,201	5,201	CD
orr-bootleg-2020	641	22	5,186	5,186	CD
orr-bootleg-2020	641	23	13	13	CD
orr-bootleg-2020	641	24	1	1	CD
orr-bootleg-2020	641	25	Network	Network	NNP
orr-bootleg-2020	641	26	Size	Size	NNP
orr-bootleg-2020	641	27	(	(	-LRB-
orr-bootleg-2020	641	28	MB	MB	NNP
orr-bootleg-2020	641	29	)	)	-RRB-
orr-bootleg-2020	641	30	4	4	CD
orr-bootleg-2020	641	31	39	39	CD
orr-bootleg-2020	641	32	35	35	CD
orr-bootleg-2020	641	33	38	38	CD
orr-bootleg-2020	641	34	34	34	CD
orr-bootleg-2020	641	35	Total	Total	NNP
orr-bootleg-2020	641	36	Size	Size	NNP
orr-bootleg-2020	641	37	(	(	-LRB-
orr-bootleg-2020	641	38	MB	MB	NNP
orr-bootleg-2020	641	39	)	)	-RRB-
orr-bootleg-2020	641	40	5,190	5,190	CD
orr-bootleg-2020	641	41	5,240	5,240	CD
orr-bootleg-2020	641	42	5,221	5,221	CD
orr-bootleg-2020	641	43	51	51	CD
orr-bootleg-2020	641	44	35	35	CD
orr-bootleg-2020	641	45	Table	table	NN
orr-bootleg-2020	641	46	11	11	CD
orr-bootleg-2020	641	47	:	:	:
orr-bootleg-2020	641	48	We	-PRON-	PRP
orr-bootleg-2020	641	49	report	report	VBP
orr-bootleg-2020	641	50	Bootleg	Bootleg	NNP
orr-bootleg-2020	641	51	trained	train	VBN
orr-bootleg-2020	641	52	with	with	IN
orr-bootleg-2020	641	53	versus	versus	NN
orr-bootleg-2020	641	54	without	without	IN
orr-bootleg-2020	641	55	weak	weak	JJ
orr-bootleg-2020	641	56	labelling	labelling	NN
orr-bootleg-2020	641	57	on	on	IN
orr-bootleg-2020	641	58	our	-PRON-	PRP$
orr-bootleg-2020	641	59	micro	micro	JJ
orr-bootleg-2020	641	60	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	641	61	dataset	dataset	NN
orr-bootleg-2020	641	62	.	.	.
orr-bootleg-2020	642	1	The	the	DT
orr-bootleg-2020	642	2	slices	slice	NNS
orr-bootleg-2020	642	3	defined	define	VBN
orr-bootleg-2020	642	4	by	by	IN
orr-bootleg-2020	642	5	gold	gold	NN
orr-bootleg-2020	642	6	anchor	anchor	NN
orr-bootleg-2020	642	7	counts	count	NNS
orr-bootleg-2020	642	8	(	(	-LRB-
orr-bootleg-2020	642	9	pre	pre	JJ
orr-bootleg-2020	642	10	-	-	JJ
orr-bootleg-2020	642	11	weak	weak	JJ
orr-bootleg-2020	642	12	labelling	labelling	NN
orr-bootleg-2020	642	13	)	)	-RRB-
orr-bootleg-2020	642	14	.	.	.
orr-bootleg-2020	643	1	We	-PRON-	PRP
orr-bootleg-2020	643	2	use	use	VBP
orr-bootleg-2020	643	3	the	the	DT
orr-bootleg-2020	643	4	InvPopPow	invpoppow	NN
orr-bootleg-2020	643	5	regularization	regularization	NN
orr-bootleg-2020	643	6	for	for	IN
orr-bootleg-2020	643	7	both	both	DT
orr-bootleg-2020	643	8	.	.	.
orr-bootleg-2020	644	1	Model	Model	NNP
orr-bootleg-2020	644	2	All	all	DT
orr-bootleg-2020	644	3	Entities	entity	NNS
orr-bootleg-2020	644	4	Torso	Torso	NNP
orr-bootleg-2020	644	5	Entities	Entities	NNPS
orr-bootleg-2020	644	6	Tail	Tail	NNP
orr-bootleg-2020	644	7	Entities	Entities	NNP
orr-bootleg-2020	644	8	Unseen	Unseen	NNP
orr-bootleg-2020	644	9	Entities	Entities	NNP
orr-bootleg-2020	644	10	Bootleg	Bootleg	NNP
orr-bootleg-2020	644	11	92.8	92.8	CD
orr-bootleg-2020	644	12	92.6	92.6	CD
orr-bootleg-2020	644	13	70.5	70.5	CD
orr-bootleg-2020	644	14	63.3	63.3	CD
orr-bootleg-2020	644	15	Bootleg	Bootleg	NNP
orr-bootleg-2020	644	16	(	(	-LRB-
orr-bootleg-2020	644	17	No	no	DT
orr-bootleg-2020	644	18	WL	WL	NNP
orr-bootleg-2020	644	19	)	)	-RRB-
orr-bootleg-2020	644	20	93.3	93.3	CD
orr-bootleg-2020	644	21	93.1	93.1	CD
orr-bootleg-2020	644	22	70.2	70.2	CD
orr-bootleg-2020	644	23	60.7	60.7	CD
orr-bootleg-2020	644	24	#	#	$
orr-bootleg-2020	644	25	Mentions	Mentions	NNPS
orr-bootleg-2020	644	26	96,237	96,237	CD
orr-bootleg-2020	644	27	36,904	36,904	CD
orr-bootleg-2020	644	28	11,541	11,541	CD
orr-bootleg-2020	644	29	3,146	3,146	CD
orr-bootleg-2020	644	30	We	-PRON-	PRP
orr-bootleg-2020	644	31	report	report	VBP
orr-bootleg-2020	644	32	the	the	DT
orr-bootleg-2020	644	33	results	result	NNS
orr-bootleg-2020	644	34	of	of	IN
orr-bootleg-2020	644	35	inverse	inverse	NN
orr-bootleg-2020	644	36	popularity	popularity	NN
orr-bootleg-2020	644	37	regularization	regularization	NN
orr-bootleg-2020	644	38	based	base	VBN
orr-bootleg-2020	644	39	on	on	IN
orr-bootleg-2020	644	40	three	three	CD
orr-bootleg-2020	644	41	different	different	JJ
orr-bootleg-2020	644	42	functions	function	NNS
orr-bootleg-2020	644	43	that	that	WDT
orr-bootleg-2020	644	44	map	map	VBP
orr-bootleg-2020	644	45	the	the	DT
orr-bootleg-2020	644	46	the	the	DT
orr-bootleg-2020	644	47	curve	curve	NN
orr-bootleg-2020	644	48	of	of	IN
orr-bootleg-2020	644	49	entity	entity	NN
orr-bootleg-2020	644	50	counts	count	NNS
orr-bootleg-2020	644	51	in	in	IN
orr-bootleg-2020	644	52	training	training	NN
orr-bootleg-2020	644	53	to	to	IN
orr-bootleg-2020	644	54	the	the	DT
orr-bootleg-2020	644	55	regularization	regularization	NN
orr-bootleg-2020	644	56	value	value	NN
orr-bootleg-2020	644	57	.	.	.
orr-bootleg-2020	645	1	For	for	IN
orr-bootleg-2020	645	2	each	each	DT
orr-bootleg-2020	645	3	function	function	NN
orr-bootleg-2020	645	4	,	,	,
orr-bootleg-2020	645	5	we	-PRON-	PRP
orr-bootleg-2020	645	6	fix	fix	VBP
orr-bootleg-2020	645	7	that	that	IN
orr-bootleg-2020	645	8	entities	entitie	VBZ
orr-bootleg-2020	645	9	with	with	IN
orr-bootleg-2020	645	10	a	a	DT
orr-bootleg-2020	645	11	frequency	frequency	NN
orr-bootleg-2020	645	12	1	1	CD
orr-bootleg-2020	645	13	receive	receive	VBP
orr-bootleg-2020	645	14	a	a	DT
orr-bootleg-2020	645	15	regularization	regularization	NN
orr-bootleg-2020	645	16	value	value	NN
orr-bootleg-2020	645	17	of	of	IN
orr-bootleg-2020	645	18	0.95	0.95	CD
orr-bootleg-2020	645	19	while	while	IN
orr-bootleg-2020	645	20	entities	entity	NNS
orr-bootleg-2020	645	21	with	with	IN
orr-bootleg-2020	645	22	a	a	DT
orr-bootleg-2020	645	23	frequency	frequency	NN
orr-bootleg-2020	645	24	of	of	IN
orr-bootleg-2020	645	25	10,000	10,000	CD
orr-bootleg-2020	645	26	receive	receive	VBP
orr-bootleg-2020	645	27	a	a	DT
orr-bootleg-2020	645	28	value	value	NN
orr-bootleg-2020	645	29	of	of	IN
orr-bootleg-2020	645	30	0.05	0.05	CD
orr-bootleg-2020	645	31	and	and	CC
orr-bootleg-2020	645	32	assign	assign	NN
orr-bootleg-2020	645	33	intermediate	intermediate	JJ
orr-bootleg-2020	645	34	values	value	NNS
orr-bootleg-2020	645	35	to	to	TO
orr-bootleg-2020	645	36	generate	generate	VB
orr-bootleg-2020	645	37	a	a	DT
orr-bootleg-2020	645	38	linear	linear	JJ
orr-bootleg-2020	645	39	,	,	,
orr-bootleg-2020	645	40	logarithmic	logarithmic	JJ
orr-bootleg-2020	645	41	,	,	,
orr-bootleg-2020	645	42	and	and	CC
orr-bootleg-2020	645	43	power	power	NN
orr-bootleg-2020	645	44	curve	curve	NN
orr-bootleg-2020	645	45	that	that	WDT
orr-bootleg-2020	645	46	applies	apply	VBZ
orr-bootleg-2020	645	47	less	less	JJR
orr-bootleg-2020	645	48	regularization	regularization	NN
orr-bootleg-2020	645	49	for	for	IN
orr-bootleg-2020	645	50	more	more	RBR
orr-bootleg-2020	645	51	popular	popular	JJ
orr-bootleg-2020	645	52	entities	entity	NNS
orr-bootleg-2020	645	53	.	.	.
orr-bootleg-2020	646	1	The	the	DT
orr-bootleg-2020	646	2	regularization	regularization	NN
orr-bootleg-2020	646	3	reported	report	VBN
orr-bootleg-2020	646	4	in	in	IN
orr-bootleg-2020	646	5	Table	table	NN
orr-bootleg-2020	646	6	6	6	CD
orr-bootleg-2020	646	7	uses	use	VBZ
orr-bootleg-2020	646	8	a	a	DT
orr-bootleg-2020	646	9	power	power	NN
orr-bootleg-2020	646	10	law	law	NN
orr-bootleg-2020	646	11	function	function	NN
orr-bootleg-2020	646	12	(	(	-LRB-
orr-bootleg-2020	646	13	f(x	f(x	NNP
orr-bootleg-2020	646	14	)	)	-RRB-
orr-bootleg-2020	646	15	=	=	NFP
orr-bootleg-2020	646	16	0.95(x−0.32	0.95(x−0.32	XX
orr-bootleg-2020	646	17	)	)	-RRB-
orr-bootleg-2020	646	18	)	)	-RRB-
orr-bootleg-2020	646	19	.	.	.
orr-bootleg-2020	647	1	We	-PRON-	PRP
orr-bootleg-2020	647	2	also	also	RB
orr-bootleg-2020	647	3	report	report	VBP
orr-bootleg-2020	647	4	in	in	IN
orr-bootleg-2020	647	5	Table	table	NN
orr-bootleg-2020	647	6	9	9	CD
orr-bootleg-2020	647	7	a	a	DT
orr-bootleg-2020	647	8	linear	linear	JJ
orr-bootleg-2020	647	9	function	function	NN
orr-bootleg-2020	647	10	(	(	-LRB-
orr-bootleg-2020	647	11	f(x	f(x	NNP
orr-bootleg-2020	647	12	)	)	-RRB-
orr-bootleg-2020	647	13	=	=	NFP
orr-bootleg-2020	647	14	−0.00009x	−0.00009x	CD
orr-bootleg-2020	647	15	+	+	SYM
orr-bootleg-2020	647	16	0.9501	0.9501	CD
orr-bootleg-2020	647	17	)	)	-RRB-
orr-bootleg-2020	647	18	and	and	CC
orr-bootleg-2020	647	19	a	a	DT
orr-bootleg-2020	647	20	logarithmic	logarithmic	JJ
orr-bootleg-2020	647	21	function	function	NN
orr-bootleg-2020	647	22	(	(	-LRB-
orr-bootleg-2020	647	23	f(x	f(x	NNP
orr-bootleg-2020	647	24	)	)	-RRB-
orr-bootleg-2020	647	25	=	=	NFP
orr-bootleg-2020	647	26	−0.097	−0.097	ADD
orr-bootleg-2020	647	27	log(x	log(x	NN
orr-bootleg-2020	647	28	)	)	-RRB-
orr-bootleg-2020	647	29	+	+	CC
orr-bootleg-2020	647	30	0.96	0.96	CD
orr-bootleg-2020	647	31	)	)	-RRB-
orr-bootleg-2020	647	32	.	.	.
orr-bootleg-2020	648	1	Each	each	DT
orr-bootleg-2020	648	2	regularization	regularization	NN
orr-bootleg-2020	648	3	function	function	NN
orr-bootleg-2020	648	4	is	be	VBZ
orr-bootleg-2020	648	5	set	set	VBN
orr-bootleg-2020	648	6	to	to	IN
orr-bootleg-2020	648	7	a	a	DT
orr-bootleg-2020	648	8	range	range	NN
orr-bootleg-2020	648	9	of	of	IN
orr-bootleg-2020	648	10	0.05	0.05	CD
orr-bootleg-2020	648	11	to	to	IN
orr-bootleg-2020	648	12	0.95	0.95	CD
orr-bootleg-2020	648	13	.	.	.
orr-bootleg-2020	649	1	We	-PRON-	PRP
orr-bootleg-2020	649	2	leave	leave	VBP
orr-bootleg-2020	649	3	it	-PRON-	PRP
orr-bootleg-2020	649	4	as	as	IN
orr-bootleg-2020	649	5	future	future	JJ
orr-bootleg-2020	649	6	work	work	NN
orr-bootleg-2020	649	7	to	to	TO
orr-bootleg-2020	649	8	explore	explore	VB
orr-bootleg-2020	649	9	other	other	JJ
orr-bootleg-2020	649	10	varied	varied	JJ
orr-bootleg-2020	649	11	regularization	regularization	NN
orr-bootleg-2020	649	12	functions	function	NNS
orr-bootleg-2020	649	13	.	.	.
orr-bootleg-2020	650	1	Table	table	NN
orr-bootleg-2020	650	2	9	9	CD
orr-bootleg-2020	650	3	shows	show	VBZ
orr-bootleg-2020	650	4	similar	similar	JJ
orr-bootleg-2020	650	5	trends	trend	NNS
orr-bootleg-2020	650	6	as	as	IN
orr-bootleg-2020	650	7	reported	report	VBN
orr-bootleg-2020	650	8	in	in	IN
orr-bootleg-2020	650	9	Section	section	NN
orr-bootleg-2020	650	10	4	4	CD
orr-bootleg-2020	650	11	that	that	IN
orr-bootleg-2020	650	12	Bootleg	Bootleg	NNP
orr-bootleg-2020	650	13	with	with	IN
orr-bootleg-2020	650	14	all	all	DT
orr-bootleg-2020	650	15	sources	source	NNS
orr-bootleg-2020	650	16	of	of	IN
orr-bootleg-2020	650	17	information	information	NN
orr-bootleg-2020	650	18	and	and	CC
orr-bootleg-2020	650	19	the	the	DT
orr-bootleg-2020	650	20	power	power	NN
orr-bootleg-2020	650	21	law	law	NN
orr-bootleg-2020	650	22	inverse	inverse	NN
orr-bootleg-2020	650	23	regularization	regularization	NN
orr-bootleg-2020	650	24	performs	perform	VBZ
orr-bootleg-2020	650	25	best	well	RBS
orr-bootleg-2020	650	26	over	over	IN
orr-bootleg-2020	650	27	the	the	DT
orr-bootleg-2020	650	28	tail	tail	NN
orr-bootleg-2020	650	29	and	and	CC
orr-bootleg-2020	650	30	the	the	DT
orr-bootleg-2020	650	31	unseen	unseen	JJ
orr-bootleg-2020	650	32	entities	entity	NNS
orr-bootleg-2020	650	33	.	.	.
orr-bootleg-2020	651	1	We	-PRON-	PRP
orr-bootleg-2020	651	2	do	do	VBP
orr-bootleg-2020	651	3	see	see	VB
orr-bootleg-2020	651	4	that	that	IN
orr-bootleg-2020	651	5	the	the	DT
orr-bootleg-2020	651	6	model	model	NN
orr-bootleg-2020	651	7	trained	train	VBN
orr-bootleg-2020	651	8	with	with	IN
orr-bootleg-2020	651	9	a	a	DT
orr-bootleg-2020	651	10	fixed	fix	VBN
orr-bootleg-2020	651	11	regularization	regularization	NN
orr-bootleg-2020	651	12	of	of	IN
orr-bootleg-2020	651	13	0.5	0.5	CD
orr-bootleg-2020	651	14	performs	perform	VBZ
orr-bootleg-2020	651	15	marginally	marginally	RB
orr-bootleg-2020	651	16	better	well	RBR
orr-bootleg-2020	651	17	on	on	IN
orr-bootleg-2020	651	18	the	the	DT
orr-bootleg-2020	651	19	torso	torso	NN
orr-bootleg-2020	651	20	and	and	CC
orr-bootleg-2020	651	21	over	over	IN
orr-bootleg-2020	651	22	all	all	DT
orr-bootleg-2020	651	23	entities	entity	NNS
orr-bootleg-2020	651	24	,	,	,
orr-bootleg-2020	651	25	likely	likely	RB
orr-bootleg-2020	651	26	because	because	IN
orr-bootleg-2020	651	27	this	this	DT
orr-bootleg-2020	651	28	involves	involve	VBZ
orr-bootleg-2020	651	29	less	less	JJR
orr-bootleg-2020	651	30	regularization	regularization	NN
orr-bootleg-2020	651	31	over	over	IN
orr-bootleg-2020	651	32	those	those	DT
orr-bootleg-2020	651	33	entity	entity	NN
orr-bootleg-2020	651	34	embeddings	embedding	NNS
orr-bootleg-2020	651	35	,	,	,
orr-bootleg-2020	651	36	allowing	allow	VBG
orr-bootleg-2020	651	37	it	-PRON-	PRP
orr-bootleg-2020	651	38	to	to	TO
orr-bootleg-2020	651	39	better	well	RBR
orr-bootleg-2020	651	40	leverage	leverage	VB
orr-bootleg-2020	651	41	the	the	DT
orr-bootleg-2020	651	42	memorized	memorized	JJ
orr-bootleg-2020	651	43	entity	entity	NN
orr-bootleg-2020	651	44	patterns	pattern	NNS
orr-bootleg-2020	651	45	,	,	,
orr-bootleg-2020	651	46	while	while	IN
orr-bootleg-2020	651	47	also	also	RB
orr-bootleg-2020	651	48	leveraging	leverage	VBG
orr-bootleg-2020	651	49	some	some	DT
orr-bootleg-2020	651	50	type	type	NN
orr-bootleg-2020	651	51	and	and	CC
orr-bootleg-2020	651	52	relational	relational	JJ
orr-bootleg-2020	651	53	information	information	NN
orr-bootleg-2020	651	54	(	(	-LRB-
orr-bootleg-2020	651	55	as	as	IN
orr-bootleg-2020	651	56	shown	show	VBN
orr-bootleg-2020	651	57	by	by	IN
orr-bootleg-2020	651	58	its	-PRON-	PRP$
orr-bootleg-2020	651	59	improved	improved	JJ
orr-bootleg-2020	651	60	performance	performance	NN
orr-bootleg-2020	651	61	over	over	IN
orr-bootleg-2020	651	62	a	a	DT
orr-bootleg-2020	651	63	lower	low	JJR
orr-bootleg-2020	651	64	fixed	fix	VBN
orr-bootleg-2020	651	65	regularization	regularization	NN
orr-bootleg-2020	651	66	)	)	-RRB-
orr-bootleg-2020	651	67	.	.	.
orr-bootleg-2020	652	1	This	this	DT
orr-bootleg-2020	652	2	model	model	NN
orr-bootleg-2020	652	3	,	,	,
orr-bootleg-2020	652	4	however	however	RB
orr-bootleg-2020	652	5	,	,	,
orr-bootleg-2020	652	6	performs	perform	VBZ
orr-bootleg-2020	652	7	4.5	4.5	CD
orr-bootleg-2020	652	8	F1	f1	NN
orr-bootleg-2020	652	9	points	point	NNS
orr-bootleg-2020	652	10	worse	bad	JJR
orr-bootleg-2020	652	11	over	over	IN
orr-bootleg-2020	652	12	unseen	unseen	JJ
orr-bootleg-2020	652	13	entities	entity	NNS
orr-bootleg-2020	652	14	than	than	IN
orr-bootleg-2020	652	15	the	the	DT
orr-bootleg-2020	652	16	best	good	JJS
orr-bootleg-2020	652	17	model	model	NN
orr-bootleg-2020	652	18	.	.	.
orr-bootleg-2020	653	1	Weak	weak	JJ
orr-bootleg-2020	653	2	Labeling	labeling	NN
orr-bootleg-2020	653	3	Table	table	NN
orr-bootleg-2020	653	4	11	11	CD
orr-bootleg-2020	653	5	shows	show	VBZ
orr-bootleg-2020	653	6	Bootleg	Bootleg	NNP
orr-bootleg-2020	653	7	’s	’s	POS
orr-bootleg-2020	653	8	results	result	VBZ
orr-bootleg-2020	653	9	with	with	IN
orr-bootleg-2020	653	10	the	the	DT
orr-bootleg-2020	653	11	inverse	inverse	NNP
orr-bootleg-2020	653	12	power	power	NN
orr-bootleg-2020	653	13	law	law	NN
orr-bootleg-2020	653	14	regularization	regularization	NN
orr-bootleg-2020	653	15	with	with	IN
orr-bootleg-2020	653	16	and	and	CC
orr-bootleg-2020	653	17	without	without	IN
orr-bootleg-2020	653	18	weak	weak	JJ
orr-bootleg-2020	653	19	labelling	labelling	NN
orr-bootleg-2020	653	20	.	.	.
orr-bootleg-2020	654	1	For	for	IN
orr-bootleg-2020	654	2	this	this	DT
orr-bootleg-2020	654	3	ablation	ablation	NN
orr-bootleg-2020	654	4	,	,	,
orr-bootleg-2020	654	5	we	-PRON-	PRP
orr-bootleg-2020	654	6	define	define	VBP
orr-bootleg-2020	654	7	our	-PRON-	PRP$
orr-bootleg-2020	654	8	set	set	NN
orr-bootleg-2020	654	9	of	of	IN
orr-bootleg-2020	654	10	torso	torso	NN
orr-bootleg-2020	654	11	,	,	,
orr-bootleg-2020	654	12	tail	tail	NN
orr-bootleg-2020	654	13	,	,	,
orr-bootleg-2020	654	14	and	and	CC
orr-bootleg-2020	654	15	unseen	unseen	JJ
orr-bootleg-2020	654	16	entities	entity	NNS
orr-bootleg-2020	654	17	by	by	IN
orr-bootleg-2020	654	18	counting	count	VBG
orr-bootleg-2020	654	19	entity	entity	NN
orr-bootleg-2020	654	20	occurrence	occurrence	NN
orr-bootleg-2020	654	21	before	before	IN
orr-bootleg-2020	654	22	weak	weak	JJ
orr-bootleg-2020	654	23	labelling	labelling	NN
orr-bootleg-2020	654	24	to	to	TO
orr-bootleg-2020	654	25	have	have	VB
orr-bootleg-2020	654	26	a	a	DT
orr-bootleg-2020	654	27	better	well	JJR
orr-bootleg-2020	654	28	understanding	understanding	NN
orr-bootleg-2020	654	29	as	as	IN
orr-bootleg-2020	654	30	to	to	IN
orr-bootleg-2020	654	31	the	the	DT
orr-bootleg-2020	654	32	lift	lift	NN
orr-bootleg-2020	654	33	from	from	IN
orr-bootleg-2020	654	34	adding	add	VBG
orr-bootleg-2020	654	35	weak	weak	JJ
orr-bootleg-2020	654	36	labelling	labelling	NN
orr-bootleg-2020	654	37	(	(	-LRB-
orr-bootleg-2020	654	38	rather	rather	RB
orr-bootleg-2020	654	39	than	than	IN
orr-bootleg-2020	654	40	the	the	DT
orr-bootleg-2020	654	41	drop	drop	NN
orr-bootleg-2020	654	42	without	without	IN
orr-bootleg-2020	654	43	it	-PRON-	PRP
orr-bootleg-2020	654	44	)	)	-RRB-
orr-bootleg-2020	654	45	.	.	.
orr-bootleg-2020	655	1	We	-PRON-	PRP
orr-bootleg-2020	655	2	see	see	VBP
orr-bootleg-2020	655	3	that	that	IN
orr-bootleg-2020	655	4	weak	weak	JJ
orr-bootleg-2020	655	5	labelling	labelling	NN
orr-bootleg-2020	655	6	provides	provide	VBZ
orr-bootleg-2020	655	7	a	a	DT
orr-bootleg-2020	655	8	lift	lift	NN
orr-bootleg-2020	655	9	of	of	IN
orr-bootleg-2020	655	10	2.6	2.6	CD
orr-bootleg-2020	655	11	F1	F1	NNP
orr-bootleg-2020	655	12	points	point	NNS
orr-bootleg-2020	655	13	over	over	IN
orr-bootleg-2020	655	14	unseen	unseen	JJ
orr-bootleg-2020	655	15	entities	entity	NNS
orr-bootleg-2020	655	16	and	and	CC
orr-bootleg-2020	655	17	0.3	0.3	CD
orr-bootleg-2020	655	18	F1	F1	NNP
orr-bootleg-2020	655	19	points	point	NNS
orr-bootleg-2020	655	20	over	over	IN
orr-bootleg-2020	655	21	tail	tail	NN
orr-bootleg-2020	655	22	entities	entity	NNS
orr-bootleg-2020	655	23	.	.	.
orr-bootleg-2020	656	1	Surprisingly	surprisingly	RB
orr-bootleg-2020	656	2	,	,	,
orr-bootleg-2020	656	3	without	without	IN
orr-bootleg-2020	656	4	weak	weak	JJ
orr-bootleg-2020	656	5	labeling	labeling	NN
orr-bootleg-2020	656	6	,	,	,
orr-bootleg-2020	656	7	Bootleg	Bootleg	NNP
orr-bootleg-2020	656	8	performs	perform	VBZ
orr-bootleg-2020	656	9	0.5	0.5	CD
orr-bootleg-2020	656	10	F1	f1	NN
orr-bootleg-2020	656	11	points	point	NNS
orr-bootleg-2020	656	12	better	well	RBR
orr-bootleg-2020	656	13	on	on	IN
orr-bootleg-2020	656	14	torso	torso	NNP
orr-bootleg-2020	656	15	entities	entity	NNS
orr-bootleg-2020	656	16	.	.	.
orr-bootleg-2020	657	1	We	-PRON-	PRP
orr-bootleg-2020	657	2	hypothesize	hypothesize	VBP
orr-bootleg-2020	657	3	this	this	DT
orr-bootleg-2020	657	4	occurs	occur	VBZ
orr-bootleg-2020	657	5	because	because	IN
orr-bootleg-2020	657	6	the	the	DT
orr-bootleg-2020	657	7	noisy	noisy	JJ
orr-bootleg-2020	657	8	weakly	weakly	NN
orr-bootleg-2020	657	9	labels	label	NNS
orr-bootleg-2020	657	10	increase	increase	VBP
orr-bootleg-2020	657	11	the	the	DT
orr-bootleg-2020	657	12	amount	amount	NN
orr-bootleg-2020	657	13	available	available	JJ
orr-bootleg-2020	657	14	signals	signal	NNS
orr-bootleg-2020	657	15	for	for	IN
orr-bootleg-2020	657	16	Bootleg	Bootleg	NNP
orr-bootleg-2020	657	17	to	to	TO
orr-bootleg-2020	657	18	learn	learn	VB
orr-bootleg-2020	657	19	consistency	consistency	NN
orr-bootleg-2020	657	20	patterns	pattern	NNS
orr-bootleg-2020	657	21	for	for	IN
orr-bootleg-2020	657	22	rarer	rare	JJR
orr-bootleg-2020	657	23	entities	entity	NNS
orr-bootleg-2020	657	24	—	—	:
orr-bootleg-2020	657	25	noisy	noisy	JJ
orr-bootleg-2020	657	26	signals	signal	NNS
orr-bootleg-2020	657	27	are	be	VBP
orr-bootleg-2020	657	28	better	well	JJR
orr-bootleg-2020	657	29	than	than	IN
orr-bootleg-2020	657	30	no	no	DT
orr-bootleg-2020	657	31	signals	signal	NNS
orr-bootleg-2020	657	32	—	—	:
orr-bootleg-2020	657	33	however	however	RB
orr-bootleg-2020	657	34	,	,	,
orr-bootleg-2020	657	35	popular	popular	JJ
orr-bootleg-2020	657	36	entities	entity	NNS
orr-bootleg-2020	657	37	have	have	VBP
orr-bootleg-2020	657	38	enough	enough	JJ
orr-bootleg-2020	657	39	less	less	JJR
orr-bootleg-2020	657	40	-	-	HYPH
orr-bootleg-2020	657	41	noisy	noisy	JJ
orr-bootleg-2020	657	42	gold	gold	NN
orr-bootleg-2020	657	43	labels	label	NNS
orr-bootleg-2020	657	44	in	in	IN
orr-bootleg-2020	657	45	the	the	DT
orr-bootleg-2020	657	46	training	training	NN
orr-bootleg-2020	657	47	data	datum	NNS
orr-bootleg-2020	657	48	,	,	,
orr-bootleg-2020	657	49	so	so	CC
orr-bootleg-2020	657	50	the	the	DT
orr-bootleg-2020	657	51	noise	noise	NN
orr-bootleg-2020	657	52	from	from	IN
orr-bootleg-2020	657	53	weak	weak	JJ
orr-bootleg-2020	657	54	labels	label	NNS
orr-bootleg-2020	657	55	may	may	MD
orr-bootleg-2020	657	56	create	create	VB
orr-bootleg-2020	657	57	conflicting	conflicting	JJ
orr-bootleg-2020	657	58	signals	signal	NNS
orr-bootleg-2020	657	59	that	that	WDT
orr-bootleg-2020	657	60	hurt	hurt	VBD
orr-bootleg-2020	657	61	performance	performance	NN
orr-bootleg-2020	657	62	.	.	.
orr-bootleg-2020	658	1	To	to	TO
orr-bootleg-2020	658	2	validate	validate	VB
orr-bootleg-2020	658	3	this	this	DT
orr-bootleg-2020	658	4	hypothesis	hypothesis	NN
orr-bootleg-2020	658	5	,	,	,
orr-bootleg-2020	658	6	we	-PRON-	PRP
orr-bootleg-2020	658	7	see	see	VBP
orr-bootleg-2020	658	8	that	that	DT
orr-bootleg-2020	658	9	overall	overall	RB
orr-bootleg-2020	658	10	,	,	,
orr-bootleg-2020	658	11	counting	count	VBG
orr-bootleg-2020	658	12	both	both	CC
orr-bootleg-2020	658	13	true	true	JJ
orr-bootleg-2020	658	14	and	and	CC
orr-bootleg-2020	658	15	weakly	weakly	RB
orr-bootleg-2020	658	16	labelled	label	VBN
orr-bootleg-2020	658	17	mentions	mention	NNS
orr-bootleg-2020	658	18	,	,	,
orr-bootleg-2020	658	19	4	4	CD
orr-bootleg-2020	658	20	%	%	NN
orr-bootleg-2020	658	21	of	of	IN
orr-bootleg-2020	658	22	mentions	mention	NNS
orr-bootleg-2020	658	23	without	without	IN
orr-bootleg-2020	658	24	weak	weak	JJ
orr-bootleg-2020	658	25	labeling	labeling	NN
orr-bootleg-2020	658	26	share	share	NN
orr-bootleg-2020	658	27	the	the	DT
orr-bootleg-2020	658	28	same	same	JJ
orr-bootleg-2020	658	29	types	type	NNS
orr-bootleg-2020	658	30	as	as	IN
orr-bootleg-2020	658	31	at	at	RB
orr-bootleg-2020	658	32	least	least	RBS
orr-bootleg-2020	658	33	one	one	CD
orr-bootleg-2020	658	34	other	other	JJ
orr-bootleg-2020	658	35	mention	mention	NN
orr-bootleg-2020	658	36	in	in	IN
orr-bootleg-2020	658	37	the	the	DT
orr-bootleg-2020	658	38	sentence	sentence	NN
orr-bootleg-2020	658	39	while	while	IN
orr-bootleg-2020	658	40	14	14	CD
orr-bootleg-2020	658	41	%	%	NN
orr-bootleg-2020	658	42	of	of	IN
orr-bootleg-2020	658	43	mentions	mention	NNS
orr-bootleg-2020	658	44	with	with	IN
orr-bootleg-2020	658	45	weak	weak	JJ
orr-bootleg-2020	658	46	labelling	labelling	NN
orr-bootleg-2020	658	47	do	do	NN
orr-bootleg-2020	658	48	.	.	.
orr-bootleg-2020	659	1	Our	-PRON-	PRP$
orr-bootleg-2020	659	2	model	model	NN
orr-bootleg-2020	659	3	predicts	predict	VBZ
orr-bootleg-2020	659	4	a	a	DT
orr-bootleg-2020	659	5	consistent	consistent	JJ
orr-bootleg-2020	659	6	answer	answer	NN
orr-bootleg-2020	659	7	only	only	RB
orr-bootleg-2020	659	8	4	4	CD
orr-bootleg-2020	659	9	%	%	NN
orr-bootleg-2020	659	10	of	of	IN
orr-bootleg-2020	659	11	the	the	DT
orr-bootleg-2020	659	12	time	time	NN
orr-bootleg-2020	659	13	without	without	IN
orr-bootleg-2020	659	14	weak	weak	JJ
orr-bootleg-2020	659	15	labeling	labeling	NN
orr-bootleg-2020	659	16	compared	compare	VBN
orr-bootleg-2020	659	17	to	to	IN
orr-bootleg-2020	659	18	13	13	CD
orr-bootleg-2020	659	19	%	%	NN
orr-bootleg-2020	659	20	of	of	IN
orr-bootleg-2020	659	21	the	the	DT
orr-bootleg-2020	659	22	time	time	NN
orr-bootleg-2020	659	23	with	with	IN
orr-bootleg-2020	659	24	weak	weak	JJ
orr-bootleg-2020	659	25	labeling	labeling	NN
orr-bootleg-2020	659	26	.	.	.
orr-bootleg-2020	660	1	Note	note	VB
orr-bootleg-2020	660	2	this	this	DT
orr-bootleg-2020	660	3	is	be	VBZ
orr-bootleg-2020	660	4	a	a	DT
orr-bootleg-2020	660	5	slightly	slightly	RB
orr-bootleg-2020	660	6	higher	high	JJR
orr-bootleg-2020	660	7	coverage	coverage	NN
orr-bootleg-2020	660	8	numbers	number	NNS
orr-bootleg-2020	660	9	than	than	IN
orr-bootleg-2020	660	10	reported	report	VBN
orr-bootleg-2020	660	11	in	in	IN
orr-bootleg-2020	660	12	Section	section	NN
orr-bootleg-2020	660	13	5	5	CD
orr-bootleg-2020	660	14	as	as	IN
orr-bootleg-2020	660	15	we	-PRON-	PRP
orr-bootleg-2020	660	16	are	be	VBP
orr-bootleg-2020	660	17	using	use	VBG
orr-bootleg-2020	660	18	a	a	DT
orr-bootleg-2020	660	19	weaker	weak	JJR
orr-bootleg-2020	660	20	form	form	NN
orr-bootleg-2020	660	21	of	of	IN
orr-bootleg-2020	660	22	consistency	consistency	NN
orr-bootleg-2020	660	23	—	—	:
orr-bootleg-2020	660	24	two	two	CD
orr-bootleg-2020	660	25	mentions	mention	NNS
orr-bootleg-2020	660	26	in	in	IN
orr-bootleg-2020	660	27	the	the	DT
orr-bootleg-2020	660	28	sentence	sentence	NN
orr-bootleg-2020	660	29	share	share	VB
orr-bootleg-2020	660	30	the	the	DT
orr-bootleg-2020	660	31	same	same	JJ
orr-bootleg-2020	660	32	type	type	NN
orr-bootleg-2020	660	33	independent	independent	JJ
orr-bootleg-2020	660	34	of	of	IN
orr-bootleg-2020	660	35	position	position	NN
orr-bootleg-2020	660	36	and	and	CC
orr-bootleg-2020	660	37	ordering	ordering	NN
orr-bootleg-2020	660	38	—	—	:
orr-bootleg-2020	660	39	and	and	CC
orr-bootleg-2020	660	40	are	be	VBP
orr-bootleg-2020	660	41	including	include	VBG
orr-bootleg-2020	660	42	weakly	weakly	RB
orr-bootleg-2020	660	43	labelled	label	VBN
orr-bootleg-2020	660	44	mentions	mention	NNS
orr-bootleg-2020	660	45	.	.	.
orr-bootleg-2020	661	1	This	this	DT
orr-bootleg-2020	661	2	22	22	CD
orr-bootleg-2020	661	3	indicates	indicate	VBZ
orr-bootleg-2020	661	4	consistency	consistency	NN
orr-bootleg-2020	661	5	is	be	VBZ
orr-bootleg-2020	661	6	a	a	DT
orr-bootleg-2020	661	7	significantly	significantly	RB
orr-bootleg-2020	661	8	more	more	RBR
orr-bootleg-2020	661	9	useful	useful	JJ
orr-bootleg-2020	661	10	pattern	pattern	NN
orr-bootleg-2020	661	11	with	with	IN
orr-bootleg-2020	661	12	weak	weak	JJ
orr-bootleg-2020	661	13	labelling	labelling	NN
orr-bootleg-2020	661	14	,	,	,
orr-bootleg-2020	661	15	and	and	CC
orr-bootleg-2020	661	16	our	-PRON-	PRP$
orr-bootleg-2020	661	17	model	model	NN
orr-bootleg-2020	661	18	predicts	predict	VBZ
orr-bootleg-2020	661	19	more	more	RBR
orr-bootleg-2020	661	20	consistent	consistent	JJ
orr-bootleg-2020	661	21	answers	answer	NNS
orr-bootleg-2020	661	22	with	with	IN
orr-bootleg-2020	661	23	weak	weak	JJ
orr-bootleg-2020	661	24	labelling	labelling	NN
orr-bootleg-2020	661	25	than	than	IN
orr-bootleg-2020	661	26	without	without	IN
orr-bootleg-2020	661	27	.	.	.
orr-bootleg-2020	662	1	Over	over	IN
orr-bootleg-2020	662	2	the	the	DT
orr-bootleg-2020	662	3	torso	torso	NN
orr-bootleg-2020	662	4	with	with	IN
orr-bootleg-2020	662	5	weak	weak	JJ
orr-bootleg-2020	662	6	labelling	labelling	NN
orr-bootleg-2020	662	7	,	,	,
orr-bootleg-2020	662	8	we	-PRON-	PRP
orr-bootleg-2020	662	9	find	find	VBP
orr-bootleg-2020	662	10	that	that	IN
orr-bootleg-2020	662	11	14	14	CD
orr-bootleg-2020	662	12	%	%	NN
orr-bootleg-2020	662	13	of	of	IN
orr-bootleg-2020	662	14	the	the	DT
orr-bootleg-2020	662	15	errors	error	NNS
orr-bootleg-2020	662	16	across	across	IN
orr-bootleg-2020	662	17	all	all	DT
orr-bootleg-2020	662	18	mentions	mention	NNS
orr-bootleg-2020	662	19	(	(	-LRB-
orr-bootleg-2020	662	20	weak	weak	JJ
orr-bootleg-2020	662	21	labelled	labelled	NN
orr-bootleg-2020	662	22	and	and	CC
orr-bootleg-2020	662	23	anchor	anchor	NN
orr-bootleg-2020	662	24	)	)	-RRB-
orr-bootleg-2020	662	25	are	be	VBP
orr-bootleg-2020	662	26	when	when	WRB
orr-bootleg-2020	662	27	Bootleg	Bootleg	NNP
orr-bootleg-2020	662	28	uses	use	VBZ
orr-bootleg-2020	662	29	consistency	consistency	NN
orr-bootleg-2020	662	30	reasoning	reasoning	NN
orr-bootleg-2020	662	31	,	,	,
orr-bootleg-2020	662	32	but	but	CC
orr-bootleg-2020	662	33	the	the	DT
orr-bootleg-2020	662	34	correct	correct	JJ
orr-bootleg-2020	662	35	answer	answer	NN
orr-bootleg-2020	662	36	does	do	VBZ
orr-bootleg-2020	662	37	not	not	RB
orr-bootleg-2020	662	38	follow	follow	VB
orr-bootleg-2020	662	39	the	the	DT
orr-bootleg-2020	662	40	consistency	consistency	NN
orr-bootleg-2020	662	41	pattern	pattern	NN
orr-bootleg-2020	662	42	.	.	.
orr-bootleg-2020	663	1	Without	without	IN
orr-bootleg-2020	663	2	weak	weak	JJ
orr-bootleg-2020	663	3	labelling	labelling	NN
orr-bootleg-2020	663	4	,	,	,
orr-bootleg-2020	663	5	only	only	RB
orr-bootleg-2020	663	6	5	5	CD
orr-bootleg-2020	663	7	%	%	NN
orr-bootleg-2020	663	8	of	of	IN
orr-bootleg-2020	663	9	the	the	DT
orr-bootleg-2020	663	10	errors	error	NNS
orr-bootleg-2020	663	11	are	be	VBP
orr-bootleg-2020	663	12	due	due	JJ
orr-bootleg-2020	663	13	to	to	IN
orr-bootleg-2020	663	14	consistency	consistency	NN
orr-bootleg-2020	663	15	.	.	.
orr-bootleg-2020	664	1	C	C	NNP
orr-bootleg-2020	664	2	Extended	Extended	NNP
orr-bootleg-2020	664	3	Downstream	Downstream	NNP
orr-bootleg-2020	664	4	Details	detail	NNS
orr-bootleg-2020	664	5	We	-PRON-	PRP
orr-bootleg-2020	664	6	now	now	RB
orr-bootleg-2020	664	7	provide	provide	VBP
orr-bootleg-2020	664	8	additional	additional	JJ
orr-bootleg-2020	664	9	details	detail	NNS
orr-bootleg-2020	664	10	of	of	IN
orr-bootleg-2020	664	11	our	-PRON-	PRP$
orr-bootleg-2020	664	12	SotA	SotA	NNP
orr-bootleg-2020	664	13	TACRED	TACRED	NNP
orr-bootleg-2020	664	14	model	model	NN
orr-bootleg-2020	664	15	,	,	,
orr-bootleg-2020	664	16	which	which	WDT
orr-bootleg-2020	664	17	uses	use	VBZ
orr-bootleg-2020	664	18	Bootleg	Bootleg	NNP
orr-bootleg-2020	664	19	embeddings	embedding	NNS
orr-bootleg-2020	664	20	.	.	.
orr-bootleg-2020	665	1	Input	input	NN
orr-bootleg-2020	665	2	We	-PRON-	PRP
orr-bootleg-2020	665	3	use	use	VBP
orr-bootleg-2020	665	4	the	the	DT
orr-bootleg-2020	665	5	revisited	revisit	VBN
orr-bootleg-2020	665	6	TACRED	tacre	VBN
orr-bootleg-2020	665	7	dataset	dataset	NN
orr-bootleg-2020	665	8	[	[	-LRB-
orr-bootleg-2020	665	9	2	2	CD
orr-bootleg-2020	665	10	]	]	-RRB-
orr-bootleg-2020	665	11	:	:	:
orr-bootleg-2020	665	12	each	each	DT
orr-bootleg-2020	665	13	example	example	NN
orr-bootleg-2020	665	14	includes	include	VBZ
orr-bootleg-2020	665	15	text	text	NN
orr-bootleg-2020	665	16	and	and	CC
orr-bootleg-2020	665	17	subject	subject	NN
orr-bootleg-2020	665	18	and	and	CC
orr-bootleg-2020	665	19	object	object	NN
orr-bootleg-2020	665	20	positions	position	NNS
orr-bootleg-2020	665	21	in	in	IN
orr-bootleg-2020	665	22	the	the	DT
orr-bootleg-2020	665	23	text	text	NN
orr-bootleg-2020	665	24	.	.	.
orr-bootleg-2020	666	1	The	the	DT
orr-bootleg-2020	666	2	task	task	NN
orr-bootleg-2020	666	3	involves	involve	VBZ
orr-bootleg-2020	666	4	extracting	extract	VBG
orr-bootleg-2020	666	5	the	the	DT
orr-bootleg-2020	666	6	relation	relation	NN
orr-bootleg-2020	666	7	between	between	IN
orr-bootleg-2020	666	8	the	the	DT
orr-bootleg-2020	666	9	subject	subject	NN
orr-bootleg-2020	666	10	and	and	CC
orr-bootleg-2020	666	11	object	object	NN
orr-bootleg-2020	666	12	.	.	.
orr-bootleg-2020	667	1	There	there	EX
orr-bootleg-2020	667	2	are	be	VBP
orr-bootleg-2020	667	3	41	41	CD
orr-bootleg-2020	667	4	potential	potential	JJ
orr-bootleg-2020	667	5	relations	relation	NNS
orr-bootleg-2020	667	6	as	as	RB
orr-bootleg-2020	667	7	well	well	RB
orr-bootleg-2020	667	8	as	as	IN
orr-bootleg-2020	667	9	a	a	DT
orr-bootleg-2020	667	10	“	"	``
orr-bootleg-2020	667	11	no	no	DT
orr-bootleg-2020	667	12	relation	relation	NN
orr-bootleg-2020	667	13	”	"	''
orr-bootleg-2020	667	14	option	option	NN
orr-bootleg-2020	667	15	.	.	.
orr-bootleg-2020	668	1	The	the	DT
orr-bootleg-2020	668	2	other	other	JJ
orr-bootleg-2020	668	3	features	feature	NNS
orr-bootleg-2020	668	4	we	-PRON-	PRP
orr-bootleg-2020	668	5	use	use	VBP
orr-bootleg-2020	668	6	as	as	IN
orr-bootleg-2020	668	7	inputs	input	NNS
orr-bootleg-2020	668	8	are	be	VBP
orr-bootleg-2020	668	9	NER	NER	NNP
orr-bootleg-2020	668	10	,	,	,
orr-bootleg-2020	668	11	POS	POS	NNP
orr-bootleg-2020	668	12	tags	tag	NNS
orr-bootleg-2020	668	13	,	,	,
orr-bootleg-2020	668	14	and	and	CC
orr-bootleg-2020	668	15	contextual	contextual	JJ
orr-bootleg-2020	668	16	Bootleg	Bootleg	NNP
orr-bootleg-2020	668	17	embeddings	embedding	NNS
orr-bootleg-2020	668	18	for	for	IN
orr-bootleg-2020	668	19	entities	entity	NNS
orr-bootleg-2020	668	20	that	that	WDT
orr-bootleg-2020	668	21	Bootleg	Bootleg	NNP
orr-bootleg-2020	668	22	disambiguates	disambiguate	NNS
orr-bootleg-2020	668	23	in	in	IN
orr-bootleg-2020	668	24	the	the	DT
orr-bootleg-2020	668	25	sentence	sentence	NN
orr-bootleg-2020	668	26	.	.	.
orr-bootleg-2020	669	1	Bootleg	Bootleg	NNP
orr-bootleg-2020	669	2	Model	Model	NNP
orr-bootleg-2020	669	3	As	as	IN
orr-bootleg-2020	669	4	TACRED	TACRED	NNP
orr-bootleg-2020	669	5	does	do	VBZ
orr-bootleg-2020	669	6	not	not	RB
orr-bootleg-2020	669	7	come	come	VB
orr-bootleg-2020	669	8	with	with	IN
orr-bootleg-2020	669	9	existing	exist	VBG
orr-bootleg-2020	669	10	mention	mention	NN
orr-bootleg-2020	669	11	boundaries	boundary	NNS
orr-bootleg-2020	669	12	,	,	,
orr-bootleg-2020	669	13	we	-PRON-	PRP
orr-bootleg-2020	669	14	perform	perform	VBP
orr-bootleg-2020	669	15	mention	mention	VBP
orr-bootleg-2020	669	16	extraction	extraction	NN
orr-bootleg-2020	669	17	by	by	IN
orr-bootleg-2020	669	18	searching	search	VBG
orr-bootleg-2020	669	19	over	over	IN
orr-bootleg-2020	669	20	n	n	JJ
orr-bootleg-2020	669	21	-	-	HYPH
orr-bootleg-2020	669	22	grams	gram	NNS
orr-bootleg-2020	669	23	,	,	,
orr-bootleg-2020	669	24	from	from	IN
orr-bootleg-2020	669	25	longest	long	JJS
orr-bootleg-2020	669	26	to	to	IN
orr-bootleg-2020	669	27	shortest	short	JJS
orr-bootleg-2020	669	28	,	,	,
orr-bootleg-2020	669	29	in	in	IN
orr-bootleg-2020	669	30	the	the	DT
orr-bootleg-2020	669	31	sentence	sentence	NN
orr-bootleg-2020	669	32	and	and	CC
orr-bootleg-2020	669	33	extract	extract	VB
orr-bootleg-2020	669	34	those	those	DT
orr-bootleg-2020	669	35	that	that	WDT
orr-bootleg-2020	669	36	are	be	VBP
orr-bootleg-2020	669	37	known	know	VBN
orr-bootleg-2020	669	38	mentions	mention	NNS
orr-bootleg-2020	669	39	in	in	IN
orr-bootleg-2020	669	40	Bootleg	Bootleg	NNP
orr-bootleg-2020	669	41	’s	’s	POS
orr-bootleg-2020	669	42	candidate	candidate	NN
orr-bootleg-2020	669	43	maps	map	NNS
orr-bootleg-2020	669	44	.	.	.
orr-bootleg-2020	670	1	We	-PRON-	PRP
orr-bootleg-2020	670	2	use	use	VBP
orr-bootleg-2020	670	3	the	the	DT
orr-bootleg-2020	670	4	same	same	JJ
orr-bootleg-2020	670	5	Bootleg	Bootleg	NNP
orr-bootleg-2020	670	6	model	model	NN
orr-bootleg-2020	670	7	from	from	IN
orr-bootleg-2020	670	8	our	-PRON-	PRP$
orr-bootleg-2020	670	9	ablations	ablation	NNS
orr-bootleg-2020	670	10	with	with	IN
orr-bootleg-2020	670	11	entity	entity	NN
orr-bootleg-2020	670	12	,	,	,
orr-bootleg-2020	670	13	KG	kg	NN
orr-bootleg-2020	670	14	,	,	,
orr-bootleg-2020	670	15	and	and	CC
orr-bootleg-2020	670	16	type	type	NN
orr-bootleg-2020	670	17	information	information	NN
orr-bootleg-2020	670	18	except	except	IN
orr-bootleg-2020	670	19	with	with	IN
orr-bootleg-2020	670	20	the	the	DT
orr-bootleg-2020	670	21	addition	addition	NN
orr-bootleg-2020	670	22	of	of	IN
orr-bootleg-2020	670	23	fine	fine	RB
orr-bootleg-2020	670	24	-	-	HYPH
orr-bootleg-2020	670	25	tuned	tune	VBN
orr-bootleg-2020	670	26	BERT	BERT	NNP
orr-bootleg-2020	670	27	word	word	NN
orr-bootleg-2020	670	28	embeddings	embedding	NNS
orr-bootleg-2020	670	29	.	.	.
orr-bootleg-2020	671	1	For	for	IN
orr-bootleg-2020	671	2	efficiency	efficiency	NN
orr-bootleg-2020	671	3	,	,	,
orr-bootleg-2020	671	4	we	-PRON-	PRP
orr-bootleg-2020	671	5	train	train	VBP
orr-bootleg-2020	671	6	on	on	IN
orr-bootleg-2020	671	7	a	a	DT
orr-bootleg-2020	671	8	subset	subset	NN
orr-bootleg-2020	671	9	of	of	IN
orr-bootleg-2020	671	10	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	671	11	training	training	NN
orr-bootleg-2020	671	12	data	datum	NNS
orr-bootleg-2020	671	13	relevant	relevant	JJ
orr-bootleg-2020	671	14	to	to	IN
orr-bootleg-2020	671	15	TACRED	TACRED	NNP
orr-bootleg-2020	671	16	.	.	.
orr-bootleg-2020	672	1	To	to	TO
orr-bootleg-2020	672	2	obtain	obtain	VB
orr-bootleg-2020	672	3	the	the	DT
orr-bootleg-2020	672	4	relevant	relevant	JJ
orr-bootleg-2020	672	5	subset	subset	NN
orr-bootleg-2020	672	6	,	,	,
orr-bootleg-2020	672	7	we	-PRON-	PRP
orr-bootleg-2020	672	8	take	take	VBP
orr-bootleg-2020	672	9	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	672	10	sentences	sentence	NNS
orr-bootleg-2020	672	11	containing	contain	VBG
orr-bootleg-2020	672	12	entities	entity	NNS
orr-bootleg-2020	672	13	extracted	extract	VBN
orr-bootleg-2020	672	14	during	during	IN
orr-bootleg-2020	672	15	candidate	candidate	NN
orr-bootleg-2020	672	16	generation	generation	NN
orr-bootleg-2020	672	17	from	from	IN
orr-bootleg-2020	672	18	a	a	DT
orr-bootleg-2020	672	19	uniform	uniform	JJ
orr-bootleg-2020	672	20	sample	sample	NN
orr-bootleg-2020	672	21	of	of	IN
orr-bootleg-2020	672	22	TACRED	TACRED	NNP
orr-bootleg-2020	672	23	data	datum	NNS
orr-bootleg-2020	672	24	;	;	:
orr-bootleg-2020	672	25	i.e.	i.e.	FW
orr-bootleg-2020	672	26	,	,	,
orr-bootleg-2020	672	27	entities	entity	NNS
orr-bootleg-2020	672	28	in	in	IN
orr-bootleg-2020	672	29	the	the	DT
orr-bootleg-2020	672	30	candidate	candidate	NN
orr-bootleg-2020	672	31	lists	list	NNS
orr-bootleg-2020	672	32	of	of	IN
orr-bootleg-2020	672	33	detected	detect	VBN
orr-bootleg-2020	672	34	mentions	mention	NNS
orr-bootleg-2020	672	35	from	from	IN
orr-bootleg-2020	672	36	a	a	DT
orr-bootleg-2020	672	37	uniform	uniform	JJ
orr-bootleg-2020	672	38	TACRED	TACRED	NNP
orr-bootleg-2020	672	39	sample	sample	NN
orr-bootleg-2020	672	40	.	.	.
orr-bootleg-2020	673	1	The	the	DT
orr-bootleg-2020	673	2	contextualized	contextualized	JJ
orr-bootleg-2020	673	3	entity	entity	NN
orr-bootleg-2020	673	4	embeddings	embedding	NNS
orr-bootleg-2020	673	5	from	from	IN
orr-bootleg-2020	673	6	Bootleg	Bootleg	NNP
orr-bootleg-2020	673	7	over	over	IN
orr-bootleg-2020	673	8	all	all	DT
orr-bootleg-2020	673	9	TACRED	tacre	VBN
orr-bootleg-2020	673	10	are	be	VBP
orr-bootleg-2020	673	11	fed	feed	VBN
orr-bootleg-2020	673	12	to	to	IN
orr-bootleg-2020	673	13	the	the	DT
orr-bootleg-2020	673	14	downstream	downstream	JJ
orr-bootleg-2020	673	15	model	model	NN
orr-bootleg-2020	673	16	.	.	.
orr-bootleg-2020	674	1	Downstream	Downstream	NNP
orr-bootleg-2020	674	2	Model	Model	NNP
orr-bootleg-2020	674	3	We	-PRON-	PRP
orr-bootleg-2020	674	4	first	first	RB
orr-bootleg-2020	674	5	use	use	VBP
orr-bootleg-2020	674	6	standard	standard	JJ
orr-bootleg-2020	674	7	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	674	8	-	-	HYPH
orr-bootleg-2020	674	9	Large	large	JJ
orr-bootleg-2020	674	10	embeddings	embedding	NNS
orr-bootleg-2020	674	11	to	to	TO
orr-bootleg-2020	674	12	encode	encode	VB
orr-bootleg-2020	674	13	the	the	DT
orr-bootleg-2020	674	14	input	input	NN
orr-bootleg-2020	674	15	text	text	NN
orr-bootleg-2020	674	16	,	,	,
orr-bootleg-2020	674	17	concatenate	concatenate	VB
orr-bootleg-2020	674	18	the	the	DT
orr-bootleg-2020	674	19	contextual	contextual	JJ
orr-bootleg-2020	674	20	Bootleg	Bootleg	NNP
orr-bootleg-2020	674	21	embeddings	embedding	NNS
orr-bootleg-2020	674	22	with	with	IN
orr-bootleg-2020	674	23	the	the	DT
orr-bootleg-2020	674	24	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	674	25	embeddings	embedding	NNS
orr-bootleg-2020	674	26	,	,	,
orr-bootleg-2020	674	27	and	and	CC
orr-bootleg-2020	674	28	then	then	RB
orr-bootleg-2020	674	29	pass	pass	VB
orr-bootleg-2020	674	30	this	this	DT
orr-bootleg-2020	674	31	through	through	IN
orr-bootleg-2020	674	32	four	four	CD
orr-bootleg-2020	674	33	transformer	transformer	NN
orr-bootleg-2020	674	34	layers	layer	NNS
orr-bootleg-2020	674	35	.	.	.
orr-bootleg-2020	675	1	We	-PRON-	PRP
orr-bootleg-2020	675	2	then	then	RB
orr-bootleg-2020	675	3	calculate	calculate	VBP
orr-bootleg-2020	675	4	the	the	DT
orr-bootleg-2020	675	5	cross	cross	JJ
orr-bootleg-2020	675	6	-	-	JJ
orr-bootleg-2020	675	7	entropy	entropy	JJ
orr-bootleg-2020	675	8	loss	loss	NN
orr-bootleg-2020	675	9	and	and	CC
orr-bootleg-2020	675	10	apply	apply	VB
orr-bootleg-2020	675	11	a	a	DT
orr-bootleg-2020	675	12	softmax	softmax	NN
orr-bootleg-2020	675	13	for	for	IN
orr-bootleg-2020	675	14	scoring	score	VBG
orr-bootleg-2020	675	15	.	.	.
orr-bootleg-2020	676	1	We	-PRON-	PRP
orr-bootleg-2020	676	2	freeze	freeze	VBP
orr-bootleg-2020	676	3	the	the	DT
orr-bootleg-2020	676	4	Bootleg	Bootleg	NNP
orr-bootleg-2020	676	5	embeddings	embedding	NNS
orr-bootleg-2020	676	6	and	and	CC
orr-bootleg-2020	676	7	fine	fine	JJ
orr-bootleg-2020	676	8	-	-	HYPH
orr-bootleg-2020	676	9	tune	tune	NN
orr-bootleg-2020	676	10	the	the	DT
orr-bootleg-2020	676	11	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	676	12	embeddings	embedding	NNS
orr-bootleg-2020	676	13	.	.	.
orr-bootleg-2020	677	1	We	-PRON-	PRP
orr-bootleg-2020	677	2	use	use	VBP
orr-bootleg-2020	677	3	the	the	DT
orr-bootleg-2020	677	4	following	follow	VBG
orr-bootleg-2020	677	5	hyperparameters	hyperparameter	NNS
orr-bootleg-2020	677	6	:	:	:
orr-bootleg-2020	677	7	the	the	DT
orr-bootleg-2020	677	8	learning	learning	NN
orr-bootleg-2020	677	9	rate	rate	NN
orr-bootleg-2020	677	10	is	be	VBZ
orr-bootleg-2020	677	11	0.00002	0.00002	CD
orr-bootleg-2020	677	12	,	,	,
orr-bootleg-2020	677	13	batch	batch	NN
orr-bootleg-2020	677	14	size	size	NN
orr-bootleg-2020	677	15	is	be	VBZ
orr-bootleg-2020	677	16	8	8	CD
orr-bootleg-2020	677	17	,	,	,
orr-bootleg-2020	677	18	gradient	gradient	JJ
orr-bootleg-2020	677	19	accummulation	accummulation	NN
orr-bootleg-2020	677	20	is	be	VBZ
orr-bootleg-2020	677	21	6	6	CD
orr-bootleg-2020	677	22	,	,	,
orr-bootleg-2020	677	23	number	number	NN
orr-bootleg-2020	677	24	of	of	IN
orr-bootleg-2020	677	25	epochs	epoch	NNS
orr-bootleg-2020	677	26	is	be	VBZ
orr-bootleg-2020	677	27	10	10	CD
orr-bootleg-2020	677	28	,	,	,
orr-bootleg-2020	677	29	L2	l2	JJ
orr-bootleg-2020	677	30	regularization	regularization	NN
orr-bootleg-2020	677	31	is	be	VBZ
orr-bootleg-2020	677	32	0.008	0.008	CD
orr-bootleg-2020	677	33	,	,	,
orr-bootleg-2020	677	34	warm	warm	JJ
orr-bootleg-2020	677	35	-	-	HYPH
orr-bootleg-2020	677	36	up	up	RP
orr-bootleg-2020	677	37	percentage	percentage	NN
orr-bootleg-2020	677	38	is	be	VBZ
orr-bootleg-2020	677	39	0.1	0.1	CD
orr-bootleg-2020	677	40	,	,	,
orr-bootleg-2020	677	41	and	and	CC
orr-bootleg-2020	677	42	maximum	maximum	JJ
orr-bootleg-2020	677	43	sequence	sequence	NN
orr-bootleg-2020	677	44	length	length	NN
orr-bootleg-2020	677	45	is	be	VBZ
orr-bootleg-2020	677	46	128	128	CD
orr-bootleg-2020	677	47	.	.	.
orr-bootleg-2020	678	1	We	-PRON-	PRP
orr-bootleg-2020	678	2	train	train	VBP
orr-bootleg-2020	678	3	with	with	IN
orr-bootleg-2020	678	4	a	a	DT
orr-bootleg-2020	678	5	NVIDIA	NVIDIA	NNP
orr-bootleg-2020	678	6	Tesla	Tesla	NNP
orr-bootleg-2020	678	7	V100	V100	NNP
orr-bootleg-2020	678	8	GPU	GPU	NNP
orr-bootleg-2020	678	9	.	.	.
orr-bootleg-2020	679	1	Extended	extended	JJ
orr-bootleg-2020	679	2	Results	result	NNS
orr-bootleg-2020	679	3	We	-PRON-	PRP
orr-bootleg-2020	679	4	study	study	VBP
orr-bootleg-2020	679	5	the	the	DT
orr-bootleg-2020	679	6	model	model	NN
orr-bootleg-2020	679	7	performance	performance	NN
orr-bootleg-2020	679	8	as	as	IN
orr-bootleg-2020	679	9	a	a	DT
orr-bootleg-2020	679	10	function	function	NN
orr-bootleg-2020	679	11	of	of	IN
orr-bootleg-2020	679	12	the	the	DT
orr-bootleg-2020	679	13	signals	signal	NNS
orr-bootleg-2020	679	14	provided	provide	VBN
orr-bootleg-2020	679	15	by	by	IN
orr-bootleg-2020	679	16	Bootleg	Bootleg	NNP
orr-bootleg-2020	679	17	.	.	.
orr-bootleg-2020	680	1	In	in	IN
orr-bootleg-2020	680	2	Table	table	NN
orr-bootleg-2020	680	3	12	12	CD
orr-bootleg-2020	680	4	,	,	,
orr-bootleg-2020	680	5	we	-PRON-	PRP
orr-bootleg-2020	680	6	show	show	VBP
orr-bootleg-2020	680	7	that	that	IN
orr-bootleg-2020	680	8	on	on	IN
orr-bootleg-2020	680	9	slices	slice	NNS
orr-bootleg-2020	680	10	with	with	IN
orr-bootleg-2020	680	11	above	above	JJ
orr-bootleg-2020	680	12	-	-	HYPH
orr-bootleg-2020	680	13	median	median	JJ
orr-bootleg-2020	680	14	numbers	number	NNS
orr-bootleg-2020	680	15	of	of	IN
orr-bootleg-2020	680	16	Bootleg	Bootleg	NNP
orr-bootleg-2020	680	17	entity	entity	NN
orr-bootleg-2020	680	18	,	,	,
orr-bootleg-2020	680	19	relation	relation	NN
orr-bootleg-2020	680	20	,	,	,
orr-bootleg-2020	680	21	and	and	CC
orr-bootleg-2020	680	22	type	type	NN
orr-bootleg-2020	680	23	signal	signal	NN
orr-bootleg-2020	680	24	counts	count	NNS
orr-bootleg-2020	680	25	identified	identify	VBN
orr-bootleg-2020	680	26	in	in	IN
orr-bootleg-2020	680	27	the	the	DT
orr-bootleg-2020	680	28	TACRED	TACRED	NNP
orr-bootleg-2020	680	29	example	example	NN
orr-bootleg-2020	680	30	,	,	,
orr-bootleg-2020	680	31	the	the	DT
orr-bootleg-2020	680	32	relative	relative	JJ
orr-bootleg-2020	680	33	gap	gap	NN
orr-bootleg-2020	680	34	between	between	IN
orr-bootleg-2020	680	35	BERT	BERT	NNP
orr-bootleg-2020	680	36	and	and	CC
orr-bootleg-2020	680	37	Bootleg	Bootleg	NNP
orr-bootleg-2020	680	38	errors	error	NNS
orr-bootleg-2020	680	39	is	be	VBZ
orr-bootleg-2020	680	40	larger	large	JJR
orr-bootleg-2020	680	41	on	on	IN
orr-bootleg-2020	680	42	the	the	DT
orr-bootleg-2020	680	43	slice	slice	NN
orr-bootleg-2020	680	44	above	above	RB
orr-bootleg-2020	680	45	,	,	,
orr-bootleg-2020	680	46	than	than	IN
orr-bootleg-2020	680	47	below	below	RB
orr-bootleg-2020	680	48	,	,	,
orr-bootleg-2020	680	49	the	the	DT
orr-bootleg-2020	680	50	median	median	NN
orr-bootleg-2020	680	51	by	by	IN
orr-bootleg-2020	680	52	1.10x	1.10x	CD
orr-bootleg-2020	680	53	,	,	,
orr-bootleg-2020	680	54	4.67x	4.67x	CD
orr-bootleg-2020	680	55	,	,	,
orr-bootleg-2020	680	56	and	and	CC
orr-bootleg-2020	680	57	1.35x	1.35x	CD
orr-bootleg-2020	680	58	respectively	respectively	RB
orr-bootleg-2020	680	59	.	.	.
orr-bootleg-2020	681	1	In	in	IN
orr-bootleg-2020	681	2	Table	table	NN
orr-bootleg-2020	681	3	13	13	CD
orr-bootleg-2020	681	4	we	-PRON-	PRP
orr-bootleg-2020	681	5	show	show	VBP
orr-bootleg-2020	681	6	the	the	DT
orr-bootleg-2020	681	7	relative	relative	JJ
orr-bootleg-2020	681	8	error	error	NN
orr-bootleg-2020	681	9	rates	rate	NNS
orr-bootleg-2020	681	10	from	from	IN
orr-bootleg-2020	681	11	the	the	DT
orr-bootleg-2020	681	12	Bootleg	Bootleg	NNP
orr-bootleg-2020	681	13	and	and	CC
orr-bootleg-2020	681	14	baseline	baseline	NN
orr-bootleg-2020	681	15	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	681	16	models	model	NNS
orr-bootleg-2020	681	17	for	for	IN
orr-bootleg-2020	681	18	the	the	DT
orr-bootleg-2020	681	19	slices	slice	NNS
orr-bootleg-2020	681	20	where	where	WRB
orr-bootleg-2020	681	21	Bootleg	Bootleg	NNP
orr-bootleg-2020	681	22	provides	provide	VBZ
orr-bootleg-2020	681	23	an	an	DT
orr-bootleg-2020	681	24	entity	entity	NN
orr-bootleg-2020	681	25	,	,	,
orr-bootleg-2020	681	26	relation	relation	NN
orr-bootleg-2020	681	27	,	,	,
orr-bootleg-2020	681	28	or	or	CC
orr-bootleg-2020	681	29	type	type	NN
orr-bootleg-2020	681	30	signal	signal	NN
orr-bootleg-2020	681	31	for	for	IN
orr-bootleg-2020	681	32	the	the	DT
orr-bootleg-2020	681	33	TACRED	TACRED	NNP
orr-bootleg-2020	681	34	example	example	NN
orr-bootleg-2020	681	35	’s	’s	POS
orr-bootleg-2020	681	36	subject	subject	NN
orr-bootleg-2020	681	37	or	or	CC
orr-bootleg-2020	681	38	object	object	NN
orr-bootleg-2020	681	39	.	.	.
orr-bootleg-2020	682	1	On	on	IN
orr-bootleg-2020	682	2	the	the	DT
orr-bootleg-2020	682	3	slice	slice	NN
orr-bootleg-2020	682	4	of	of	IN
orr-bootleg-2020	682	5	these	these	DT
orr-bootleg-2020	682	6	signals	signal	NNS
orr-bootleg-2020	682	7	are	be	VBP
orr-bootleg-2020	682	8	respectively	respectively	RB
orr-bootleg-2020	682	9	present	present	JJ
orr-bootleg-2020	682	10	,	,	,
orr-bootleg-2020	682	11	the	the	DT
orr-bootleg-2020	682	12	baseline	baseline	NN
orr-bootleg-2020	682	13	model	model	NN
orr-bootleg-2020	682	14	performs	perform	VBZ
orr-bootleg-2020	682	15	1.20x	1.20x	CD
orr-bootleg-2020	682	16	,	,	,
orr-bootleg-2020	682	17	1.18x	1.18x	CD
orr-bootleg-2020	682	18	,	,	,
orr-bootleg-2020	682	19	and	and	CC
orr-bootleg-2020	682	20	1.20x	1.20x	CD
orr-bootleg-2020	682	21	worse	bad	JJR
orr-bootleg-2020	682	22	than	than	IN
orr-bootleg-2020	682	23	the	the	DT
orr-bootleg-2020	682	24	Bootleg	Bootleg	NNP
orr-bootleg-2020	682	25	TACRED	tacre	VBD
orr-bootleg-2020	682	26	model	model	NN
orr-bootleg-2020	682	27	.	.	.
orr-bootleg-2020	683	1	These	these	DT
orr-bootleg-2020	683	2	results	result	NNS
orr-bootleg-2020	683	3	indicate	indicate	VBP
orr-bootleg-2020	683	4	that	that	IN
orr-bootleg-2020	683	5	the	the	DT
orr-bootleg-2020	683	6	knowledge	knowledge	NN
orr-bootleg-2020	683	7	representations	representation	NNS
orr-bootleg-2020	683	8	from	from	IN
orr-bootleg-2020	683	9	Bootleg	Bootleg	NNP
orr-bootleg-2020	683	10	successfully	successfully	RB
orr-bootleg-2020	683	11	transfer	transfer	VBP
orr-bootleg-2020	683	12	useful	useful	JJ
orr-bootleg-2020	683	13	information	information	NN
orr-bootleg-2020	683	14	to	to	IN
orr-bootleg-2020	683	15	the	the	DT
orr-bootleg-2020	683	16	downstream	downstream	JJ
orr-bootleg-2020	683	17	model	model	NN
orr-bootleg-2020	683	18	.	.	.
orr-bootleg-2020	684	1	D	D	NNP
orr-bootleg-2020	684	2	Extended	Extended	NNP
orr-bootleg-2020	684	3	Error	Error	NNP
orr-bootleg-2020	684	4	Analysis	Analysis	NNP
orr-bootleg-2020	684	5	D.1	D.1	NNP
orr-bootleg-2020	684	6	Reasoning	Reasoning	NNP
orr-bootleg-2020	684	7	Patterns	Patterns	NNPS
orr-bootleg-2020	684	8	Here	here	RB
orr-bootleg-2020	684	9	we	-PRON-	PRP
orr-bootleg-2020	684	10	provide	provide	VBP
orr-bootleg-2020	684	11	additional	additional	JJ
orr-bootleg-2020	684	12	details	detail	NNS
orr-bootleg-2020	684	13	about	about	IN
orr-bootleg-2020	684	14	the	the	DT
orr-bootleg-2020	684	15	distributions	distribution	NNS
orr-bootleg-2020	684	16	of	of	IN
orr-bootleg-2020	684	17	types	type	NNS
orr-bootleg-2020	684	18	and	and	CC
orr-bootleg-2020	684	19	relations	relation	NNS
orr-bootleg-2020	684	20	in	in	IN
orr-bootleg-2020	684	21	the	the	DT
orr-bootleg-2020	684	22	data	datum	NNS
orr-bootleg-2020	684	23	.	.	.
orr-bootleg-2020	685	1	Distinct	distinct	JJ
orr-bootleg-2020	685	2	Tails	tail	NNS
orr-bootleg-2020	685	3	Like	like	IN
orr-bootleg-2020	685	4	entities	entity	NNS
orr-bootleg-2020	685	5	,	,	,
orr-bootleg-2020	685	6	types	type	NNS
orr-bootleg-2020	685	7	and	and	CC
orr-bootleg-2020	685	8	relations	relation	NNS
orr-bootleg-2020	685	9	also	also	RB
orr-bootleg-2020	685	10	have	have	VBP
orr-bootleg-2020	685	11	tail	tail	NN
orr-bootleg-2020	685	12	distributions	distribution	NNS
orr-bootleg-2020	685	13	.	.	.
orr-bootleg-2020	686	1	For	for	IN
orr-bootleg-2020	686	2	example	example	NN
orr-bootleg-2020	686	3	,	,	,
orr-bootleg-2020	686	4	types	type	NNS
orr-bootleg-2020	686	5	such	such	JJ
orr-bootleg-2020	686	6	as	as	IN
orr-bootleg-2020	686	7	“	"	``
orr-bootleg-2020	686	8	country	country	NN
orr-bootleg-2020	686	9	”	"	''
orr-bootleg-2020	686	10	and	and	CC
orr-bootleg-2020	686	11	“	"	``
orr-bootleg-2020	686	12	film	film	NN
orr-bootleg-2020	686	13	”	"	''
orr-bootleg-2020	686	14	appear	appear	VBP
orr-bootleg-2020	686	15	2.7	2.7	CD
orr-bootleg-2020	686	16	M	M	NNP
orr-bootleg-2020	686	17	and	and	CC
orr-bootleg-2020	686	18	800k	800k	NNPS
orr-bootleg-2020	686	19	times	time	NNS
orr-bootleg-2020	686	20	respectively	respectively	RB
orr-bootleg-2020	686	21	,	,	,
orr-bootleg-2020	686	22	while	while	IN
orr-bootleg-2020	686	23	types	type	NNS
orr-bootleg-2020	686	24	such	such	JJ
orr-bootleg-2020	686	25	as	as	IN
orr-bootleg-2020	686	26	“	"	``
orr-bootleg-2020	686	27	quasiregular	quasiregular	JJ
orr-bootleg-2020	686	28	polyhedron	polyhedron	NN
orr-bootleg-2020	686	29	”	"	''
orr-bootleg-2020	686	30	and	and	CC
orr-bootleg-2020	686	31	“	"	``
orr-bootleg-2020	686	32	hospital	hospital	NN
orr-bootleg-2020	686	33	-	-	HYPH
orr-bootleg-2020	686	34	acquired	acquire	VBN
orr-bootleg-2020	686	35	infection	infection	NN
orr-bootleg-2020	686	36	”	"	''
orr-bootleg-2020	686	37	appear	appear	VBP
orr-bootleg-2020	686	38	once	once	IN
orr-bootleg-2020	686	39	each	each	DT
orr-bootleg-2020	686	40	in	in	IN
orr-bootleg-2020	686	41	our	-PRON-	PRP$
orr-bootleg-2020	686	42	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	686	43	training	training	NN
orr-bootleg-2020	686	44	data	datum	NNS
orr-bootleg-2020	686	45	.	.	.
orr-bootleg-2020	687	1	Meanwhile	meanwhile	RB
orr-bootleg-2020	687	2	,	,	,
orr-bootleg-2020	687	3	relations	relation	NNS
orr-bootleg-2020	687	4	such	such	JJ
orr-bootleg-2020	687	5	as	as	IN
orr-bootleg-2020	687	6	“	"	``
orr-bootleg-2020	687	7	occupation	occupation	NN
orr-bootleg-2020	687	8	”	"	''
orr-bootleg-2020	687	9	and	and	CC
orr-bootleg-2020	687	10	“	"	``
orr-bootleg-2020	687	11	educated	educate	VBN
orr-bootleg-2020	687	12	at	at	IN
orr-bootleg-2020	687	13	”	"	''
orr-bootleg-2020	687	14	appear	appear	VB
orr-bootleg-2020	687	15	35	35	CD
orr-bootleg-2020	687	16	M	m	NN
orr-bootleg-2020	687	17	and	and	CC
orr-bootleg-2020	687	18	16	16	CD
orr-bootleg-2020	687	19	M	M	NNP
orr-bootleg-2020	687	20	times	time	NNS
orr-bootleg-2020	687	21	respectively	respectively	RB
orr-bootleg-2020	687	22	,	,	,
orr-bootleg-2020	687	23	while	while	IN
orr-bootleg-2020	687	24	relations	relation	NNS
orr-bootleg-2020	687	25	such	such	JJ
orr-bootleg-2020	687	26	as	as	IN
orr-bootleg-2020	687	27	23	23	CD
orr-bootleg-2020	687	28	Table	table	NN
orr-bootleg-2020	687	29	12	12	CD
orr-bootleg-2020	687	30	:	:	:
orr-bootleg-2020	687	31	We	-PRON-	PRP
orr-bootleg-2020	687	32	rank	rank	VBP
orr-bootleg-2020	687	33	TACRED	tacre	VBN
orr-bootleg-2020	687	34	examples	example	NNS
orr-bootleg-2020	687	35	by	by	IN
orr-bootleg-2020	687	36	the	the	DT
orr-bootleg-2020	687	37	proportion	proportion	NN
orr-bootleg-2020	687	38	of	of	IN
orr-bootleg-2020	687	39	words	word	NNS
orr-bootleg-2020	687	40	that	that	WDT
orr-bootleg-2020	687	41	receive	receive	VBP
orr-bootleg-2020	687	42	Bootleg	Bootleg	NNP
orr-bootleg-2020	687	43	embedding	embed	VBG
orr-bootleg-2020	687	44	features	feature	NNS
orr-bootleg-2020	687	45	where	where	WRB
orr-bootleg-2020	687	46	:	:	:
orr-bootleg-2020	687	47	Bootleg	Bootleg	NNP
orr-bootleg-2020	687	48	disambiguates	disambiguate	VBZ
orr-bootleg-2020	687	49	an	an	DT
orr-bootleg-2020	687	50	entity	entity	NN
orr-bootleg-2020	687	51	,	,	,
orr-bootleg-2020	687	52	leverages	leverage	VBZ
orr-bootleg-2020	687	53	Wikidata	Wikidata	NNP
orr-bootleg-2020	687	54	relations	relation	NNS
orr-bootleg-2020	687	55	for	for	IN
orr-bootleg-2020	687	56	the	the	DT
orr-bootleg-2020	687	57	embedding	embedding	NN
orr-bootleg-2020	687	58	,	,	,
orr-bootleg-2020	687	59	and	and	CC
orr-bootleg-2020	687	60	leverages	leverage	VBZ
orr-bootleg-2020	687	61	Wikidata	Wikidata	NNP
orr-bootleg-2020	687	62	types	type	NNS
orr-bootleg-2020	687	63	for	for	IN
orr-bootleg-2020	687	64	the	the	DT
orr-bootleg-2020	687	65	embedding	embedding	NN
orr-bootleg-2020	687	66	.	.	.
orr-bootleg-2020	688	1	We	-PRON-	PRP
orr-bootleg-2020	688	2	take	take	VBP
orr-bootleg-2020	688	3	examples	example	NNS
orr-bootleg-2020	688	4	where	where	WRB
orr-bootleg-2020	688	5	the	the	DT
orr-bootleg-2020	688	6	proportion	proportion	NN
orr-bootleg-2020	688	7	is	be	VBZ
orr-bootleg-2020	688	8	greater	great	JJR
orr-bootleg-2020	688	9	than	than	IN
orr-bootleg-2020	688	10	0	0	CD
orr-bootleg-2020	688	11	.	.	.
orr-bootleg-2020	689	1	For	for	IN
orr-bootleg-2020	689	2	each	each	DT
orr-bootleg-2020	689	3	of	of	IN
orr-bootleg-2020	689	4	these	these	DT
orr-bootleg-2020	689	5	three	three	CD
orr-bootleg-2020	689	6	slices	slice	NNS
orr-bootleg-2020	689	7	,	,	,
orr-bootleg-2020	689	8	we	-PRON-	PRP
orr-bootleg-2020	689	9	report	report	VBP
orr-bootleg-2020	689	10	the	the	DT
orr-bootleg-2020	689	11	gap	gap	NN
orr-bootleg-2020	689	12	between	between	IN
orr-bootleg-2020	689	13	the	the	DT
orr-bootleg-2020	689	14	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	689	15	model	model	NN
orr-bootleg-2020	689	16	and	and	CC
orr-bootleg-2020	689	17	Bootleg	Bootleg	NNP
orr-bootleg-2020	689	18	model	model	NN
orr-bootleg-2020	689	19	’s	’s	POS
orr-bootleg-2020	689	20	error	error	NN
orr-bootleg-2020	689	21	rates	rate	NNS
orr-bootleg-2020	689	22	on	on	IN
orr-bootleg-2020	689	23	the	the	DT
orr-bootleg-2020	689	24	examples	example	NNS
orr-bootleg-2020	689	25	with	with	IN
orr-bootleg-2020	689	26	above	above	JJ
orr-bootleg-2020	689	27	-	-	HYPH
orr-bootleg-2020	689	28	median	median	JJ
orr-bootleg-2020	689	29	proportion	proportion	NN
orr-bootleg-2020	689	30	(	(	-LRB-
orr-bootleg-2020	689	31	more	more	JJR
orr-bootleg-2020	689	32	Bootleg	Bootleg	NNP
orr-bootleg-2020	689	33	signal	signal	NN
orr-bootleg-2020	689	34	)	)	-RRB-
orr-bootleg-2020	689	35	relative	relative	JJ
orr-bootleg-2020	689	36	to	to	IN
orr-bootleg-2020	689	37	the	the	DT
orr-bootleg-2020	689	38	below	below	IN
orr-bootleg-2020	689	39	-	-	HYPH
orr-bootleg-2020	689	40	median	median	JJ
orr-bootleg-2020	689	41	proportion	proportion	NN
orr-bootleg-2020	689	42	(	(	-LRB-
orr-bootleg-2020	689	43	less	less	JJR
orr-bootleg-2020	689	44	Bootleg	Bootleg	NNP
orr-bootleg-2020	689	45	signal	signal	NN
orr-bootleg-2020	689	46	)	)	-RRB-
orr-bootleg-2020	689	47	.	.	.
orr-bootleg-2020	690	1	With	with	IN
orr-bootleg-2020	690	2	more	more	JJR
orr-bootleg-2020	690	3	Bootleg	Bootleg	NNP
orr-bootleg-2020	690	4	information	information	NN
orr-bootleg-2020	690	5	,	,	,
orr-bootleg-2020	690	6	we	-PRON-	PRP
orr-bootleg-2020	690	7	see	see	VBP
orr-bootleg-2020	690	8	the	the	DT
orr-bootleg-2020	690	9	improvement	improvement	NN
orr-bootleg-2020	690	10	our	-PRON-	PRP$
orr-bootleg-2020	690	11	SotA	SotA	NNP
orr-bootleg-2020	690	12	model	model	NN
orr-bootleg-2020	690	13	provides	provide	VBZ
orr-bootleg-2020	690	14	over	over	IN
orr-bootleg-2020	690	15	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	690	16	increases	increase	NNS
orr-bootleg-2020	690	17	.	.	.
orr-bootleg-2020	691	1	Bootleg	Bootleg	NNP
orr-bootleg-2020	691	2	Signal	Signal	NNP
orr-bootleg-2020	691	3	#	#	$
orr-bootleg-2020	691	4	Examples	Examples	NNPS
orr-bootleg-2020	691	5	with	with	IN
orr-bootleg-2020	691	6	the	the	DT
orr-bootleg-2020	691	7	Signal	Signal	NNP
orr-bootleg-2020	691	8	Gap	Gap	NNP
orr-bootleg-2020	691	9	Above	Above	NNP
orr-bootleg-2020	691	10	/	/	SYM
orr-bootleg-2020	691	11	Below	Below	NNP
orr-bootleg-2020	691	12	Median	Median	NNP
orr-bootleg-2020	691	13	Entity	Entity	NNP
orr-bootleg-2020	691	14	15323	15323	CD
orr-bootleg-2020	691	15	1.10	1.10	CD
orr-bootleg-2020	691	16	Relation	Relation	NNP
orr-bootleg-2020	691	17	5400	5400	CD
orr-bootleg-2020	691	18	4.67	4.67	CD
orr-bootleg-2020	691	19	Type	type	NN
orr-bootleg-2020	691	20	15509	15509	CD
orr-bootleg-2020	691	21	1.35	1.35	CD
orr-bootleg-2020	691	22	Table	table	NN
orr-bootleg-2020	691	23	13	13	CD
orr-bootleg-2020	691	24	:	:	:
orr-bootleg-2020	691	25	We	-PRON-	PRP
orr-bootleg-2020	691	26	compute	compute	VBP
orr-bootleg-2020	691	27	the	the	DT
orr-bootleg-2020	691	28	error	error	NN
orr-bootleg-2020	691	29	rate	rate	NN
orr-bootleg-2020	691	30	of	of	IN
orr-bootleg-2020	691	31	SpanBERT	SpanBERT	NNP
orr-bootleg-2020	691	32	relative	relative	JJ
orr-bootleg-2020	691	33	to	to	IN
orr-bootleg-2020	691	34	our	-PRON-	PRP$
orr-bootleg-2020	691	35	Bootleg	Bootleg	NNP
orr-bootleg-2020	691	36	downstream	downstream	JJ
orr-bootleg-2020	691	37	model	model	NN
orr-bootleg-2020	691	38	for	for	IN
orr-bootleg-2020	691	39	three	three	CD
orr-bootleg-2020	691	40	slices	slice	NNS
orr-bootleg-2020	691	41	of	of	IN
orr-bootleg-2020	691	42	TACRED	tacred	JJ
orr-bootleg-2020	691	43	data	datum	NNS
orr-bootleg-2020	691	44	where	where	WRB
orr-bootleg-2020	691	45	respectively	respectively	RB
orr-bootleg-2020	691	46	Bootleg	Bootleg	NNP
orr-bootleg-2020	691	47	disambiguates	disambiguate	VBZ
orr-bootleg-2020	691	48	the	the	DT
orr-bootleg-2020	691	49	subject	subject	NN
orr-bootleg-2020	691	50	and/or	and/or	CC
orr-bootleg-2020	691	51	object	object	NN
orr-bootleg-2020	691	52	,	,	,
orr-bootleg-2020	691	53	Bootleg	Bootleg	NNP
orr-bootleg-2020	691	54	leverages	leverage	VBZ
orr-bootleg-2020	691	55	Wikidata	wikidata	NN
orr-bootleg-2020	691	56	relations	relation	NNS
orr-bootleg-2020	691	57	for	for	IN
orr-bootleg-2020	691	58	the	the	DT
orr-bootleg-2020	691	59	embedding	embedding	NN
orr-bootleg-2020	691	60	of	of	IN
orr-bootleg-2020	691	61	the	the	DT
orr-bootleg-2020	691	62	subject	subject	NN
orr-bootleg-2020	691	63	and	and	CC
orr-bootleg-2020	691	64	object	object	NN
orr-bootleg-2020	691	65	pair	pair	NN
orr-bootleg-2020	691	66	,	,	,
orr-bootleg-2020	691	67	and	and	CC
orr-bootleg-2020	691	68	Bootleg	Bootleg	NNP
orr-bootleg-2020	691	69	leverages	leverage	VBZ
orr-bootleg-2020	691	70	Wikidata	Wikidata	NNP
orr-bootleg-2020	691	71	types	type	NNS
orr-bootleg-2020	691	72	for	for	IN
orr-bootleg-2020	691	73	the	the	DT
orr-bootleg-2020	691	74	embedding	embedding	NN
orr-bootleg-2020	691	75	of	of	IN
orr-bootleg-2020	691	76	the	the	DT
orr-bootleg-2020	691	77	subject	subject	NN
orr-bootleg-2020	691	78	and/or	and/or	CC
orr-bootleg-2020	691	79	object	object	NN
orr-bootleg-2020	691	80	in	in	IN
orr-bootleg-2020	691	81	the	the	DT
orr-bootleg-2020	691	82	example	example	NN
orr-bootleg-2020	691	83	.	.	.
orr-bootleg-2020	692	1	Subject	subject	JJ
orr-bootleg-2020	692	2	-	-	HYPH
orr-bootleg-2020	692	3	Object	object	NN
orr-bootleg-2020	692	4	Signal	signal	NN
orr-bootleg-2020	692	5	#	#	NN
orr-bootleg-2020	692	6	Examples	Examples	NNPS
orr-bootleg-2020	692	7	BERT	BERT	NNP
orr-bootleg-2020	692	8	/	/	SYM
orr-bootleg-2020	692	9	Bootleg	Bootleg	NNP
orr-bootleg-2020	692	10	Error	Error	NNP
orr-bootleg-2020	692	11	Rate	rate	NN
orr-bootleg-2020	692	12	Entity	Entity	NNP
orr-bootleg-2020	692	13	12621	12621	CD
orr-bootleg-2020	692	14	1.20	1.20	CD
orr-bootleg-2020	692	15	Relation	relation	NN
orr-bootleg-2020	692	16	542	542	CD
orr-bootleg-2020	692	17	1.18	1.18	CD
orr-bootleg-2020	692	18	Obj	Obj	NNP
orr-bootleg-2020	692	19	Type	type	NN
orr-bootleg-2020	692	20	12044	12044	CD
orr-bootleg-2020	692	21	1.20	1.20	CD
orr-bootleg-2020	692	22	“	"	``
orr-bootleg-2020	692	23	positive	positive	JJ
orr-bootleg-2020	692	24	diagnostic	diagnostic	JJ
orr-bootleg-2020	692	25	predictor	predictor	NN
orr-bootleg-2020	692	26	”	"	''
orr-bootleg-2020	692	27	and	and	CC
orr-bootleg-2020	692	28	“	"	``
orr-bootleg-2020	692	29	author	author	NN
orr-bootleg-2020	692	30	of	of	IN
orr-bootleg-2020	692	31	afterword	afterword	IN
orr-bootleg-2020	692	32	”	"	''
orr-bootleg-2020	692	33	appear	appear	VBP
orr-bootleg-2020	692	34	7	7	CD
orr-bootleg-2020	692	35	times	time	NNS
orr-bootleg-2020	692	36	respectively	respectively	RB
orr-bootleg-2020	692	37	in	in	IN
orr-bootleg-2020	692	38	the	the	DT
orr-bootleg-2020	692	39	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	692	40	training	training	NN
orr-bootleg-2020	692	41	data	datum	NNS
orr-bootleg-2020	692	42	.	.	.
orr-bootleg-2020	693	1	However	however	RB
orr-bootleg-2020	693	2	we	-PRON-	PRP
orr-bootleg-2020	693	3	find	find	VBP
orr-bootleg-2020	693	4	that	that	IN
orr-bootleg-2020	693	5	the	the	DT
orr-bootleg-2020	693	6	entity-	entity-	NN
orr-bootleg-2020	693	7	,	,	,
orr-bootleg-2020	693	8	relation-	relation-	NNP
orr-bootleg-2020	693	9	,	,	,
orr-bootleg-2020	693	10	and	and	CC
orr-bootleg-2020	693	11	type	type	NN
orr-bootleg-2020	693	12	-	-	HYPH
orr-bootleg-2020	693	13	tails	tail	NNS
orr-bootleg-2020	693	14	are	be	VBP
orr-bootleg-2020	693	15	distinct	distinct	JJ
orr-bootleg-2020	693	16	:	:	:
orr-bootleg-2020	693	17	88	88	CD
orr-bootleg-2020	693	18	%	%	NN
orr-bootleg-2020	693	19	of	of	IN
orr-bootleg-2020	693	20	the	the	DT
orr-bootleg-2020	693	21	tail	tail	NN
orr-bootleg-2020	693	22	-	-	HYPH
orr-bootleg-2020	693	23	entities	entity	NNS
orr-bootleg-2020	693	24	by	by	IN
orr-bootleg-2020	693	25	entity	entity	NN
orr-bootleg-2020	693	26	-	-	HYPH
orr-bootleg-2020	693	27	count	count	NN
orr-bootleg-2020	693	28	have	have	VBP
orr-bootleg-2020	693	29	Wikidata	Wikidata	NNP
orr-bootleg-2020	693	30	types	type	NNS
orr-bootleg-2020	693	31	that	that	WDT
orr-bootleg-2020	693	32	are	be	VBP
orr-bootleg-2020	693	33	non	non	JJ
orr-bootleg-2020	693	34	-	-	JJ
orr-bootleg-2020	693	35	tail	tail	JJ
orr-bootleg-2020	693	36	types	type	NNS
orr-bootleg-2020	693	37	and	and	CC
orr-bootleg-2020	693	38	90	90	CD
orr-bootleg-2020	693	39	%	%	NN
orr-bootleg-2020	693	40	of	of	IN
orr-bootleg-2020	693	41	the	the	DT
orr-bootleg-2020	693	42	tail	tail	NN
orr-bootleg-2020	693	43	-	-	HYPH
orr-bootleg-2020	693	44	entities	entity	NNS
orr-bootleg-2020	693	45	by	by	IN
orr-bootleg-2020	693	46	entity	entity	NN
orr-bootleg-2020	693	47	-	-	HYPH
orr-bootleg-2020	693	48	count	count	NN
orr-bootleg-2020	693	49	have	have	VBP
orr-bootleg-2020	693	50	non	non	JJ
orr-bootleg-2020	693	51	-	-	JJ
orr-bootleg-2020	693	52	tail	tail	JJ
orr-bootleg-2020	693	53	relations.12	relations.12	NN
orr-bootleg-2020	693	54	For	for	IN
orr-bootleg-2020	693	55	example	example	NN
orr-bootleg-2020	693	56	,	,	,
orr-bootleg-2020	693	57	the	the	DT
orr-bootleg-2020	693	58	head	head	NN
orr-bootleg-2020	693	59	Wikidata	Wikidata	NNP
orr-bootleg-2020	693	60	type	type	NN
orr-bootleg-2020	693	61	“	"	``
orr-bootleg-2020	693	62	country	country	NN
orr-bootleg-2020	693	63	”	"	''
orr-bootleg-2020	693	64	contains	contain	VBZ
orr-bootleg-2020	693	65	rare	rare	JJ
orr-bootleg-2020	693	66	entities	entity	NNS
orr-bootleg-2020	693	67	“	"	``
orr-bootleg-2020	693	68	Palaú	Palaú	NNP
orr-bootleg-2020	693	69	”	"	''
orr-bootleg-2020	693	70	and	and	CC
orr-bootleg-2020	693	71	“	"	``
orr-bootleg-2020	693	72	Belgium	Belgium	NNP
orr-bootleg-2020	693	73	–	–	:
orr-bootleg-2020	693	74	France	France	NNP
orr-bootleg-2020	693	75	border	border	NN
orr-bootleg-2020	693	76	”	"	''
orr-bootleg-2020	693	77	.	.	.
orr-bootleg-2020	694	1	We	-PRON-	PRP
orr-bootleg-2020	694	2	observe	observe	VBP
orr-bootleg-2020	694	3	that	that	IN
orr-bootleg-2020	694	4	Bootleg	Bootleg	NNP
orr-bootleg-2020	694	5	significantly	significantly	RB
orr-bootleg-2020	694	6	improves	improve	VBZ
orr-bootleg-2020	694	7	tail	tail	NN
orr-bootleg-2020	694	8	performance	performance	NN
orr-bootleg-2020	694	9	over	over	IN
orr-bootleg-2020	694	10	each	each	DT
orr-bootleg-2020	694	11	of	of	IN
orr-bootleg-2020	694	12	the	the	DT
orr-bootleg-2020	694	13	tails	tail	NNS
orr-bootleg-2020	694	14	.	.	.
orr-bootleg-2020	695	1	We	-PRON-	PRP
orr-bootleg-2020	695	2	rank	rank	VBP
orr-bootleg-2020	695	3	the	the	DT
orr-bootleg-2020	695	4	Wikidata	Wikidata	NNP
orr-bootleg-2020	695	5	types	type	NNS
orr-bootleg-2020	695	6	and	and	CC
orr-bootleg-2020	695	7	relations	relation	NNS
orr-bootleg-2020	695	8	by	by	IN
orr-bootleg-2020	695	9	the	the	DT
orr-bootleg-2020	695	10	number	number	NN
orr-bootleg-2020	695	11	of	of	IN
orr-bootleg-2020	695	12	occurrences	occurrence	NNS
orr-bootleg-2020	695	13	in	in	IN
orr-bootleg-2020	695	14	the	the	DT
orr-bootleg-2020	695	15	training	training	NN
orr-bootleg-2020	695	16	data	datum	NNS
orr-bootleg-2020	695	17	and	and	CC
orr-bootleg-2020	695	18	study	study	VB
orr-bootleg-2020	695	19	the	the	DT
orr-bootleg-2020	695	20	lift	lift	NN
orr-bootleg-2020	695	21	from	from	IN
orr-bootleg-2020	695	22	Bootleg	Bootleg	NNP
orr-bootleg-2020	695	23	as	as	IN
orr-bootleg-2020	695	24	a	a	DT
orr-bootleg-2020	695	25	function	function	NN
orr-bootleg-2020	695	26	of	of	IN
orr-bootleg-2020	695	27	the	the	DT
orr-bootleg-2020	695	28	number	number	NN
orr-bootleg-2020	695	29	of	of	IN
orr-bootleg-2020	695	30	times	time	NNS
orr-bootleg-2020	695	31	the	the	DT
orr-bootleg-2020	695	32	signal	signal	NN
orr-bootleg-2020	695	33	appears	appear	VBZ
orr-bootleg-2020	695	34	during	during	IN
orr-bootleg-2020	695	35	training	training	NN
orr-bootleg-2020	695	36	.	.	.
orr-bootleg-2020	696	1	Bootleg	Bootleg	NNP
orr-bootleg-2020	696	2	performs	perform	VBZ
orr-bootleg-2020	696	3	an	an	DT
orr-bootleg-2020	696	4	9.4	9.4	CD
orr-bootleg-2020	696	5	F1	f1	NN
orr-bootleg-2020	696	6	and	and	CC
orr-bootleg-2020	696	7	20.3	20.3	CD
orr-bootleg-2020	696	8	F1	F1	NNP
orr-bootleg-2020	696	9	points	point	VBZ
orr-bootleg-2020	696	10	better	well	RBR
orr-bootleg-2020	696	11	than	than	IN
orr-bootleg-2020	696	12	the	the	DT
orr-bootleg-2020	696	13	NED	NED	NNP
orr-bootleg-2020	696	14	-	-	HYPH
orr-bootleg-2020	696	15	Base	Base	NNP
orr-bootleg-2020	696	16	baseline	baseline	NN
orr-bootleg-2020	696	17	for	for	IN
orr-bootleg-2020	696	18	examples	example	NNS
orr-bootleg-2020	696	19	with	with	IN
orr-bootleg-2020	696	20	gold	gold	JJ
orr-bootleg-2020	696	21	types	type	NNS
orr-bootleg-2020	696	22	appearing	appear	VBG
orr-bootleg-2020	696	23	more	more	RBR
orr-bootleg-2020	696	24	and	and	CC
orr-bootleg-2020	696	25	less	less	JJR
orr-bootleg-2020	696	26	than	than	IN
orr-bootleg-2020	696	27	the	the	DT
orr-bootleg-2020	696	28	median	median	JJ
orr-bootleg-2020	696	29	number	number	NN
orr-bootleg-2020	696	30	of	of	IN
orr-bootleg-2020	696	31	times	time	NNS
orr-bootleg-2020	696	32	during	during	IN
orr-bootleg-2020	696	33	training	training	NN
orr-bootleg-2020	696	34	respectively	respectively	RB
orr-bootleg-2020	696	35	.	.	.
orr-bootleg-2020	697	1	Bootleg	Bootleg	NNP
orr-bootleg-2020	697	2	provides	provide	VBZ
orr-bootleg-2020	697	3	a	a	DT
orr-bootleg-2020	697	4	a	a	DT
orr-bootleg-2020	697	5	7.8	7.8	CD
orr-bootleg-2020	697	6	F1	f1	NN
orr-bootleg-2020	697	7	points	point	NNS
orr-bootleg-2020	697	8	and	and	CC
orr-bootleg-2020	697	9	13.7	13.7	CD
orr-bootleg-2020	697	10	F1	F1	NNP
orr-bootleg-2020	697	11	points	point	VBZ
orr-bootleg-2020	697	12	better	well	RBR
orr-bootleg-2020	697	13	than	than	IN
orr-bootleg-2020	697	14	the	the	DT
orr-bootleg-2020	697	15	baseline	baseline	NN
orr-bootleg-2020	697	16	for	for	IN
orr-bootleg-2020	697	17	examples	example	NNS
orr-bootleg-2020	697	18	with	with	IN
orr-bootleg-2020	697	19	gold	gold	NN
orr-bootleg-2020	697	20	relations	relation	NNS
orr-bootleg-2020	697	21	appearing	appear	VBG
orr-bootleg-2020	697	22	more	more	RBR
orr-bootleg-2020	697	23	and	and	CC
orr-bootleg-2020	697	24	less	less	JJR
orr-bootleg-2020	697	25	than	than	IN
orr-bootleg-2020	697	26	the	the	DT
orr-bootleg-2020	697	27	median	median	JJ
orr-bootleg-2020	697	28	number	number	NN
orr-bootleg-2020	697	29	of	of	IN
orr-bootleg-2020	697	30	times	time	NNS
orr-bootleg-2020	697	31	during	during	IN
orr-bootleg-2020	697	32	training	training	NN
orr-bootleg-2020	697	33	respectively	respectively	RB
orr-bootleg-2020	697	34	.	.	.
orr-bootleg-2020	698	1	These	these	DT
orr-bootleg-2020	698	2	results	result	NNS
orr-bootleg-2020	698	3	indicate	indicate	VBP
orr-bootleg-2020	698	4	that	that	IN
orr-bootleg-2020	698	5	Bootleg	Bootleg	NNP
orr-bootleg-2020	698	6	excels	excel	VBZ
orr-bootleg-2020	698	7	on	on	IN
orr-bootleg-2020	698	8	the	the	DT
orr-bootleg-2020	698	9	tails	tail	NNS
orr-bootleg-2020	698	10	of	of	IN
orr-bootleg-2020	698	11	types	type	NNS
orr-bootleg-2020	698	12	and	and	CC
orr-bootleg-2020	698	13	relations	relation	NNS
orr-bootleg-2020	698	14	as	as	RB
orr-bootleg-2020	698	15	well	well	RB
orr-bootleg-2020	698	16	.	.	.
orr-bootleg-2020	699	1	Next	next	RB
orr-bootleg-2020	699	2	,	,	,
orr-bootleg-2020	699	3	ranking	rank	VBG
orr-bootleg-2020	699	4	the	the	DT
orr-bootleg-2020	699	5	Wikidata	Wikidata	NNP
orr-bootleg-2020	699	6	types	type	NNS
orr-bootleg-2020	699	7	and	and	CC
orr-bootleg-2020	699	8	relations	relation	NNS
orr-bootleg-2020	699	9	by	by	IN
orr-bootleg-2020	699	10	the	the	DT
orr-bootleg-2020	699	11	proportion	proportion	NN
orr-bootleg-2020	699	12	of	of	IN
orr-bootleg-2020	699	13	comprised	comprise	VBN
orr-bootleg-2020	699	14	rare	rare	JJ
orr-bootleg-2020	699	15	(	(	-LRB-
orr-bootleg-2020	699	16	tail	tail	NN
orr-bootleg-2020	699	17	and	and	CC
orr-bootleg-2020	699	18	unseen	unseen	JJ
orr-bootleg-2020	699	19	)	)	-RRB-
orr-bootleg-2020	699	20	entities	entity	NNS
orr-bootleg-2020	699	21	,	,	,
orr-bootleg-2020	699	22	we	-PRON-	PRP
orr-bootleg-2020	699	23	further	further	RB
orr-bootleg-2020	699	24	find	find	VBP
orr-bootleg-2020	699	25	that	that	IN
orr-bootleg-2020	699	26	Bootleg	Bootleg	NNP
orr-bootleg-2020	699	27	provides	provide	VBZ
orr-bootleg-2020	699	28	the	the	DT
orr-bootleg-2020	699	29	lowest	low	JJS
orr-bootleg-2020	699	30	error	error	NN
orr-bootleg-2020	699	31	rates	rate	NNS
orr-bootleg-2020	699	32	across	across	IN
orr-bootleg-2020	699	33	types	type	NNS
orr-bootleg-2020	699	34	and	and	CC
orr-bootleg-2020	699	35	relations	relation	NNS
orr-bootleg-2020	699	36	,	,	,
orr-bootleg-2020	699	37	regardless	regardless	RB
orr-bootleg-2020	699	38	of	of	IN
orr-bootleg-2020	699	39	the	the	DT
orr-bootleg-2020	699	40	proportion	proportion	NN
orr-bootleg-2020	699	41	of	of	IN
orr-bootleg-2020	699	42	rare	rare	JJ
orr-bootleg-2020	699	43	entities	entity	NNS
orr-bootleg-2020	699	44	,	,	,
orr-bootleg-2020	699	45	while	while	IN
orr-bootleg-2020	699	46	the	the	DT
orr-bootleg-2020	699	47	baseline	baseline	NN
orr-bootleg-2020	699	48	and	and	CC
orr-bootleg-2020	699	49	Entity	Entity	NNP
orr-bootleg-2020	699	50	-	-	HYPH
orr-bootleg-2020	699	51	Only	only	RB
orr-bootleg-2020	699	52	models	model	NNS
orr-bootleg-2020	699	53	give	give	VBP
orr-bootleg-2020	699	54	relatively	relatively	RB
orr-bootleg-2020	699	55	larger	large	JJR
orr-bootleg-2020	699	56	error	error	NN
orr-bootleg-2020	699	57	rates	rate	NNS
orr-bootleg-2020	699	58	as	as	IN
orr-bootleg-2020	699	59	the	the	DT
orr-bootleg-2020	699	60	proportion	proportion	NN
orr-bootleg-2020	699	61	of	of	IN
orr-bootleg-2020	699	62	rare	rare	JJ
orr-bootleg-2020	699	63	entities	entity	NNS
orr-bootleg-2020	699	64	increases	increase	NNS
orr-bootleg-2020	699	65	(	(	-LRB-
orr-bootleg-2020	699	66	Figure	figure	NN
orr-bootleg-2020	699	67	4	4	CD
orr-bootleg-2020	699	68	)	)	-RRB-
orr-bootleg-2020	699	69	.	.	.
orr-bootleg-2020	700	1	The	the	DT
orr-bootleg-2020	700	2	trend	trend	NN
orr-bootleg-2020	700	3	for	for	IN
orr-bootleg-2020	700	4	types	type	NNS
orr-bootleg-2020	700	5	is	be	VBZ
orr-bootleg-2020	700	6	flat	flat	JJ
orr-bootleg-2020	700	7	as	as	IN
orr-bootleg-2020	700	8	the	the	DT
orr-bootleg-2020	700	9	proportion	proportion	NN
orr-bootleg-2020	700	10	of	of	IN
orr-bootleg-2020	700	11	rare	rare	JJ
orr-bootleg-2020	700	12	entities	entity	NNS
orr-bootleg-2020	700	13	increases	increase	NNS
orr-bootleg-2020	700	14	,	,	,
orr-bootleg-2020	700	15	while	while	IN
orr-bootleg-2020	700	16	the	the	DT
orr-bootleg-2020	700	17	trend	trend	NN
orr-bootleg-2020	700	18	for	for	IN
orr-bootleg-2020	700	19	relations	relation	NNS
orr-bootleg-2020	700	20	is	be	VBZ
orr-bootleg-2020	700	21	upwards	upwards	JJ
orr-bootleg-2020	700	22	sloping	sloping	NN
orr-bootleg-2020	700	23	.	.	.
orr-bootleg-2020	701	1	These	these	DT
orr-bootleg-2020	701	2	results	result	NNS
orr-bootleg-2020	701	3	indicate	indicate	VBP
orr-bootleg-2020	701	4	that	that	IN
orr-bootleg-2020	701	5	Bootleg	Bootleg	NNP
orr-bootleg-2020	701	6	is	be	VBZ
orr-bootleg-2020	701	7	better	well	RBR
orr-bootleg-2020	701	8	able	able	JJ
orr-bootleg-2020	701	9	to	to	TO
orr-bootleg-2020	701	10	transfer	transfer	VB
orr-bootleg-2020	701	11	the	the	DT
orr-bootleg-2020	701	12	patterns	pattern	NNS
orr-bootleg-2020	701	13	learned	learn	VBN
orr-bootleg-2020	701	14	from	from	IN
orr-bootleg-2020	701	15	one	one	CD
orr-bootleg-2020	701	16	entity	entity	NN
orr-bootleg-2020	701	17	to	to	IN
orr-bootleg-2020	701	18	other	other	JJ
orr-bootleg-2020	701	19	entities	entity	NNS
orr-bootleg-2020	701	20	that	that	WDT
orr-bootleg-2020	701	21	share	share	VBP
orr-bootleg-2020	701	22	its	-PRON-	PRP$
orr-bootleg-2020	701	23	same	same	JJ
orr-bootleg-2020	701	24	types	type	NNS
orr-bootleg-2020	701	25	and	and	CC
orr-bootleg-2020	701	26	relations	relation	NNS
orr-bootleg-2020	701	27	.	.	.
orr-bootleg-2020	702	1	The	the	DT
orr-bootleg-2020	702	2	improvement	improvement	NN
orr-bootleg-2020	702	3	from	from	IN
orr-bootleg-2020	702	4	Bootleg	Bootleg	NNP
orr-bootleg-2020	702	5	over	over	IN
orr-bootleg-2020	702	6	the	the	DT
orr-bootleg-2020	702	7	baseline	baseline	NN
orr-bootleg-2020	702	8	increases	increase	NNS
orr-bootleg-2020	702	9	as	as	IN
orr-bootleg-2020	702	10	the	the	DT
orr-bootleg-2020	702	11	rare	rare	JJ
orr-bootleg-2020	702	12	-	-	HYPH
orr-bootleg-2020	702	13	proportion	proportion	NN
orr-bootleg-2020	702	14	increases	increase	NNS
orr-bootleg-2020	702	15	,	,	,
orr-bootleg-2020	702	16	indicating	indicate	VBG
orr-bootleg-2020	702	17	that	that	IN
orr-bootleg-2020	702	18	Bootleg	Bootleg	NNP
orr-bootleg-2020	702	19	is	be	VBZ
orr-bootleg-2020	702	20	able	able	JJ
orr-bootleg-2020	702	21	to	to	TO
orr-bootleg-2020	702	22	efficiently	efficiently	RB
orr-bootleg-2020	702	23	transfer	transfer	VB
orr-bootleg-2020	702	24	knowledge	knowledge	NN
orr-bootleg-2020	702	25	even	even	RB
orr-bootleg-2020	702	26	when	when	WRB
orr-bootleg-2020	702	27	the	the	DT
orr-bootleg-2020	702	28	type	type	NN
orr-bootleg-2020	702	29	or	or	CC
orr-bootleg-2020	702	30	relation	relation	NN
orr-bootleg-2020	702	31	category	category	NN
orr-bootleg-2020	702	32	contains	contain	VBZ
orr-bootleg-2020	702	33	none	none	NN
orr-bootleg-2020	702	34	or	or	CC
orr-bootleg-2020	702	35	few	few	JJ
orr-bootleg-2020	702	36	popular	popular	JJ
orr-bootleg-2020	702	37	entities	entity	NNS
orr-bootleg-2020	702	38	.	.	.
orr-bootleg-2020	703	1	Type	Type	NNP
orr-bootleg-2020	703	2	Affordance	affordance	NN
orr-bootleg-2020	703	3	For	for	IN
orr-bootleg-2020	703	4	the	the	DT
orr-bootleg-2020	703	5	type	type	NN
orr-bootleg-2020	703	6	affordance	affordance	NN
orr-bootleg-2020	703	7	pattern	pattern	NN
orr-bootleg-2020	703	8	,	,	,
orr-bootleg-2020	703	9	we	-PRON-	PRP
orr-bootleg-2020	703	10	find	find	VBP
orr-bootleg-2020	703	11	that	that	IN
orr-bootleg-2020	703	12	the	the	DT
orr-bootleg-2020	703	13	TF	TF	NNP
orr-bootleg-2020	703	14	-	-	HYPH
orr-bootleg-2020	703	15	IDF	IDF	NNP
orr-bootleg-2020	703	16	keywords	keyword	NNS
orr-bootleg-2020	703	17	provide	provide	VBP
orr-bootleg-2020	703	18	high	high	JJ
orr-bootleg-2020	703	19	coverage	coverage	NN
orr-bootleg-2020	703	20	over	over	IN
orr-bootleg-2020	703	21	the	the	DT
orr-bootleg-2020	703	22	examples	example	NNS
orr-bootleg-2020	703	23	containing	contain	VBG
orr-bootleg-2020	703	24	the	the	DT
orr-bootleg-2020	703	25	gold	gold	NN
orr-bootleg-2020	703	26	type	type	NN
orr-bootleg-2020	703	27	:	:	:
orr-bootleg-2020	703	28	88	88	CD
orr-bootleg-2020	703	29	%	%	NN
orr-bootleg-2020	703	30	of	of	IN
orr-bootleg-2020	703	31	examples	example	NNS
orr-bootleg-2020	703	32	where	where	WRB
orr-bootleg-2020	703	33	the	the	DT
orr-bootleg-2020	703	34	gold	gold	JJ
orr-bootleg-2020	703	35	entity	entity	NN
orr-bootleg-2020	703	36	has	have	VBZ
orr-bootleg-2020	703	37	a	a	DT
orr-bootleg-2020	703	38	particular	particular	JJ
orr-bootleg-2020	703	39	type	type	NN
orr-bootleg-2020	703	40	contain	contain	NN
orr-bootleg-2020	703	41	an	an	DT
orr-bootleg-2020	703	42	affordance	affordance	NN
orr-bootleg-2020	703	43	keyword	keyword	NN
orr-bootleg-2020	703	44	for	for	IN
orr-bootleg-2020	703	45	that	that	DT
orr-bootleg-2020	703	46	type	type	NN
orr-bootleg-2020	703	47	.	.	.
orr-bootleg-2020	704	1	An	an	DT
orr-bootleg-2020	704	2	example	example	NN
orr-bootleg-2020	704	3	of	of	IN
orr-bootleg-2020	704	4	a	a	DT
orr-bootleg-2020	704	5	type	type	NN
orr-bootleg-2020	704	6	with	with	IN
orr-bootleg-2020	704	7	full	full	JJ
orr-bootleg-2020	704	8	coverage	coverage	NN
orr-bootleg-2020	704	9	by	by	IN
orr-bootleg-2020	704	10	the	the	DT
orr-bootleg-2020	704	11	affordance	affordance	NN
orr-bootleg-2020	704	12	12Similar	12similar	CD
orr-bootleg-2020	704	13	to	to	TO
orr-bootleg-2020	704	14	tail	tail	NN
orr-bootleg-2020	704	15	-	-	HYPH
orr-bootleg-2020	704	16	entities	entity	NNS
orr-bootleg-2020	704	17	,	,	,
orr-bootleg-2020	704	18	tail	tail	NN
orr-bootleg-2020	704	19	-	-	HYPH
orr-bootleg-2020	704	20	types	type	NNS
orr-bootleg-2020	704	21	and	and	CC
orr-bootleg-2020	704	22	tail	tail	NN
orr-bootleg-2020	704	23	-	-	HYPH
orr-bootleg-2020	704	24	relations	relation	NNS
orr-bootleg-2020	704	25	are	be	VBP
orr-bootleg-2020	704	26	defined	define	VBN
orr-bootleg-2020	704	27	as	as	IN
orr-bootleg-2020	704	28	those	those	DT
orr-bootleg-2020	704	29	appearing	appear	VBG
orr-bootleg-2020	704	30	1	1	CD
orr-bootleg-2020	704	31	-	-	SYM
orr-bootleg-2020	704	32	10	10	CD
orr-bootleg-2020	704	33	times	time	NNS
orr-bootleg-2020	704	34	in	in	IN
orr-bootleg-2020	704	35	the	the	DT
orr-bootleg-2020	704	36	training	training	NN
orr-bootleg-2020	704	37	data	datum	NNS
orr-bootleg-2020	704	38	.	.	.
orr-bootleg-2020	705	1	24	24	CD
orr-bootleg-2020	705	2	0.0	0.0	CD
orr-bootleg-2020	705	3	0.2	0.2	CD
orr-bootleg-2020	705	4	0.4	0.4	CD
orr-bootleg-2020	705	5	0.6	0.6	CD
orr-bootleg-2020	705	6	0.8	0.8	CD
orr-bootleg-2020	705	7	1.0	1.0	CD
orr-bootleg-2020	705	8	Rare	Rare	NNP
orr-bootleg-2020	705	9	-	-	HYPH
orr-bootleg-2020	705	10	Entities	entities	NN
orr-bootleg-2020	705	11	Proportion	proportion	NN
orr-bootleg-2020	705	12	of	of	IN
orr-bootleg-2020	705	13	Relation	Relation	NNP
orr-bootleg-2020	705	14	0.0	0.0	CD
orr-bootleg-2020	705	15	0.2	0.2	CD
orr-bootleg-2020	705	16	0.4	0.4	CD
orr-bootleg-2020	705	17	0.6	0.6	CD
orr-bootleg-2020	705	18	0.8	0.8	CD
orr-bootleg-2020	705	19	1.0	1.0	CD
orr-bootleg-2020	705	20	Er	er	UH
orr-bootleg-2020	705	21	ro	ro	NN
orr-bootleg-2020	705	22	r	r	NN
orr-bootleg-2020	705	23	R	r	NN
orr-bootleg-2020	705	24	at	at	IN
orr-bootleg-2020	705	25	e	e	DT
orr-bootleg-2020	705	26	BOOTLEG	bootleg	JJ
orr-bootleg-2020	705	27	BERT	BERT	NNP
orr-bootleg-2020	705	28	KG	kg	NN
orr-bootleg-2020	705	29	-	-	HYPH
orr-bootleg-2020	705	30	Only	Only	NNP
orr-bootleg-2020	705	31	Type	type	NN
orr-bootleg-2020	705	32	-	-	HYPH
orr-bootleg-2020	705	33	Only	Only	NNP
orr-bootleg-2020	705	34	Entity	entity	NN
orr-bootleg-2020	705	35	-	-	,
orr-bootleg-2020	705	36	Only	only	RB
orr-bootleg-2020	705	37	0.0	0.0	CD
orr-bootleg-2020	705	38	0.2	0.2	CD
orr-bootleg-2020	705	39	0.4	0.4	CD
orr-bootleg-2020	705	40	0.6	0.6	CD
orr-bootleg-2020	705	41	0.8	0.8	CD
orr-bootleg-2020	705	42	1.0	1.0	CD
orr-bootleg-2020	705	43	Rare	Rare	NNP
orr-bootleg-2020	705	44	-	-	HYPH
orr-bootleg-2020	705	45	Entities	entity	NNS
orr-bootleg-2020	705	46	Proportion	proportion	NN
orr-bootleg-2020	705	47	of	of	IN
orr-bootleg-2020	705	48	Type	Type	NNP
orr-bootleg-2020	705	49	0.0	0.0	CD
orr-bootleg-2020	705	50	0.2	0.2	CD
orr-bootleg-2020	705	51	0.4	0.4	CD
orr-bootleg-2020	705	52	0.6	0.6	CD
orr-bootleg-2020	705	53	0.8	0.8	CD
orr-bootleg-2020	705	54	1.0	1.0	CD
orr-bootleg-2020	705	55	Er	er	UH
orr-bootleg-2020	705	56	ro	ro	NN
orr-bootleg-2020	705	57	r	r	NN
orr-bootleg-2020	705	58	R	r	NN
orr-bootleg-2020	705	59	at	at	IN
orr-bootleg-2020	705	60	e	e	DT
orr-bootleg-2020	705	61	BOOTLEG	bootleg	JJ
orr-bootleg-2020	705	62	BERT	BERT	NNP
orr-bootleg-2020	705	63	KG	kg	NN
orr-bootleg-2020	705	64	-	-	HYPH
orr-bootleg-2020	705	65	Only	Only	NNP
orr-bootleg-2020	705	66	Type	type	NN
orr-bootleg-2020	705	67	-	-	HYPH
orr-bootleg-2020	705	68	Only	Only	NNP
orr-bootleg-2020	705	69	Entity	entity	NN
orr-bootleg-2020	705	70	-	-	HYPH
orr-bootleg-2020	705	71	Only	Only	NNP
orr-bootleg-2020	705	72	Figure	figure	NN
orr-bootleg-2020	705	73	4	4	CD
orr-bootleg-2020	705	74	:	:	:
orr-bootleg-2020	705	75	For	for	IN
orr-bootleg-2020	705	76	all	all	PDT
orr-bootleg-2020	705	77	the	the	DT
orr-bootleg-2020	705	78	entities	entity	NNS
orr-bootleg-2020	705	79	of	of	IN
orr-bootleg-2020	705	80	a	a	DT
orr-bootleg-2020	705	81	particular	particular	JJ
orr-bootleg-2020	705	82	type	type	NN
orr-bootleg-2020	705	83	or	or	CC
orr-bootleg-2020	705	84	relation	relation	NN
orr-bootleg-2020	705	85	,	,	,
orr-bootleg-2020	705	86	we	-PRON-	PRP
orr-bootleg-2020	705	87	calculate	calculate	VBP
orr-bootleg-2020	705	88	the	the	DT
orr-bootleg-2020	705	89	percentage	percentage	NN
orr-bootleg-2020	705	90	of	of	IN
orr-bootleg-2020	705	91	rare	rare	JJ
orr-bootleg-2020	705	92	entities	entity	NNS
orr-bootleg-2020	705	93	(	(	-LRB-
orr-bootleg-2020	705	94	tails	tail	NNS
orr-bootleg-2020	705	95	and	and	CC
orr-bootleg-2020	705	96	toes	toe	VBZ
orr-bootleg-2020	705	97	entities	entity	NNS
orr-bootleg-2020	705	98	)	)	-RRB-
orr-bootleg-2020	705	99	.	.	.
orr-bootleg-2020	706	1	We	-PRON-	PRP
orr-bootleg-2020	706	2	show	show	VBP
orr-bootleg-2020	706	3	the	the	DT
orr-bootleg-2020	706	4	error	error	NN
orr-bootleg-2020	706	5	rate	rate	NN
orr-bootleg-2020	706	6	on	on	IN
orr-bootleg-2020	706	7	the	the	DT
orr-bootleg-2020	706	8	Wikipedia	Wikipedia	NNP
orr-bootleg-2020	706	9	validation	validation	NN
orr-bootleg-2020	706	10	set	set	VBD
orr-bootleg-2020	706	11	as	as	IN
orr-bootleg-2020	706	12	a	a	DT
orr-bootleg-2020	706	13	function	function	NN
orr-bootleg-2020	706	14	of	of	IN
orr-bootleg-2020	706	15	the	the	DT
orr-bootleg-2020	706	16	rare	rare	JJ
orr-bootleg-2020	706	17	-	-	HYPH
orr-bootleg-2020	706	18	proportion	proportion	NN
orr-bootleg-2020	706	19	of	of	IN
orr-bootleg-2020	706	20	entities	entity	NNS
orr-bootleg-2020	706	21	of	of	IN
orr-bootleg-2020	706	22	a	a	DT
orr-bootleg-2020	706	23	given	give	VBN
orr-bootleg-2020	706	24	(	(	-LRB-
orr-bootleg-2020	706	25	Left	left	NN
orr-bootleg-2020	706	26	)	)	-RRB-
orr-bootleg-2020	706	27	relation	relation	NN
orr-bootleg-2020	706	28	or	or	CC
orr-bootleg-2020	706	29	(	(	-LRB-
orr-bootleg-2020	706	30	Right	right	UH
orr-bootleg-2020	706	31	)	)	-RRB-
orr-bootleg-2020	706	32	type	type	NN
orr-bootleg-2020	706	33	appearing	appear	VBG
orr-bootleg-2020	706	34	in	in	IN
orr-bootleg-2020	706	35	the	the	DT
orr-bootleg-2020	706	36	validation	validation	NN
orr-bootleg-2020	706	37	set	set	VBN
orr-bootleg-2020	706	38	.	.	.
orr-bootleg-2020	707	1	keywords	keywords	NNPS
orr-bootleg-2020	707	2	is	be	VBZ
orr-bootleg-2020	707	3	“	"	``
orr-bootleg-2020	707	4	café	café	NN
orr-bootleg-2020	707	5	”	"	''
orr-bootleg-2020	707	6	,	,	,
orr-bootleg-2020	707	7	with	with	IN
orr-bootleg-2020	707	8	keywords	keyword	NNS
orr-bootleg-2020	707	9	such	such	JJ
orr-bootleg-2020	707	10	as	as	IN
orr-bootleg-2020	707	11	“	"	``
orr-bootleg-2020	707	12	coffee	coffee	NN
orr-bootleg-2020	707	13	”	"	''
orr-bootleg-2020	707	14	,	,	,
orr-bootleg-2020	707	15	“	"	``
orr-bootleg-2020	707	16	Starbucks	Starbucks	NNP
orr-bootleg-2020	707	17	”	"	''
orr-bootleg-2020	707	18	,	,	,
orr-bootleg-2020	707	19	and	and	CC
orr-bootleg-2020	707	20	“	"	``
orr-bootleg-2020	707	21	Internet	internet	NN
orr-bootleg-2020	707	22	”	"	''
orr-bootleg-2020	707	23	;	;	:
orr-bootleg-2020	707	24	in	in	IN
orr-bootleg-2020	707	25	each	each	DT
orr-bootleg-2020	707	26	of	of	IN
orr-bootleg-2020	707	27	the	the	DT
orr-bootleg-2020	707	28	77	77	CD
orr-bootleg-2020	707	29	times	time	NNS
orr-bootleg-2020	707	30	an	an	DT
orr-bootleg-2020	707	31	entity	entity	NN
orr-bootleg-2020	707	32	of	of	IN
orr-bootleg-2020	707	33	the	the	DT
orr-bootleg-2020	707	34	type	type	NN
orr-bootleg-2020	707	35	“	"	``
orr-bootleg-2020	707	36	cafe	cafe	NN
orr-bootleg-2020	707	37	”	"	''
orr-bootleg-2020	707	38	appears	appear	VBZ
orr-bootleg-2020	707	39	in	in	IN
orr-bootleg-2020	707	40	the	the	DT
orr-bootleg-2020	707	41	validation	validation	NN
orr-bootleg-2020	707	42	set	set	NN
orr-bootleg-2020	707	43	,	,	,
orr-bootleg-2020	707	44	an	an	DT
orr-bootleg-2020	707	45	affordance	affordance	NN
orr-bootleg-2020	707	46	keyword	keyword	NN
orr-bootleg-2020	707	47	is	be	VBZ
orr-bootleg-2020	707	48	present	present	JJ
orr-bootleg-2020	707	49	.	.	.
orr-bootleg-2020	708	1	Types	type	NNS
orr-bootleg-2020	708	2	with	with	IN
orr-bootleg-2020	708	3	low	low	JJ
orr-bootleg-2020	708	4	coverage	coverage	NN
orr-bootleg-2020	708	5	in	in	IN
orr-bootleg-2020	708	6	the	the	DT
orr-bootleg-2020	708	7	validation	validation	NN
orr-bootleg-2020	708	8	set	set	VBN
orr-bootleg-2020	708	9	by	by	IN
orr-bootleg-2020	708	10	affordance	affordance	NN
orr-bootleg-2020	708	11	keywords	keyword	NNS
orr-bootleg-2020	708	12	tend	tend	VBP
orr-bootleg-2020	708	13	to	to	TO
orr-bootleg-2020	708	14	be	be	VB
orr-bootleg-2020	708	15	the	the	DT
orr-bootleg-2020	708	16	rare	rare	JJ
orr-bootleg-2020	708	17	types	type	NNS
orr-bootleg-2020	708	18	:	:	:
orr-bootleg-2020	708	19	for	for	IN
orr-bootleg-2020	708	20	the	the	DT
orr-bootleg-2020	708	21	types	type	NNS
orr-bootleg-2020	708	22	with	with	IN
orr-bootleg-2020	708	23	coverage	coverage	NN
orr-bootleg-2020	708	24	less	less	JJR
orr-bootleg-2020	708	25	than	than	IN
orr-bootleg-2020	708	26	50	50	CD
orr-bootleg-2020	708	27	%	%	NN
orr-bootleg-2020	708	28	,	,	,
orr-bootleg-2020	708	29	such	such	JJ
orr-bootleg-2020	708	30	as	as	IN
orr-bootleg-2020	708	31	“	"	``
orr-bootleg-2020	708	32	dietician	dietician	JJ
orr-bootleg-2020	708	33	”	"	''
orr-bootleg-2020	708	34	or	or	CC
orr-bootleg-2020	708	35	“	"	``
orr-bootleg-2020	708	36	chess	chess	NNP
orr-bootleg-2020	708	37	official	official	NN
orr-bootleg-2020	708	38	”	"	''
orr-bootleg-2020	708	39	,	,	,
orr-bootleg-2020	708	40	the	the	DT
orr-bootleg-2020	708	41	median	median	JJ
orr-bootleg-2020	708	42	number	number	NN
orr-bootleg-2020	708	43	of	of	IN
orr-bootleg-2020	708	44	occurrences	occurrence	NNS
orr-bootleg-2020	708	45	in	in	IN
orr-bootleg-2020	708	46	the	the	DT
orr-bootleg-2020	708	47	validation	validation	NN
orr-bootleg-2020	708	48	set	set	VBN
orr-bootleg-2020	708	49	is	be	VBZ
orr-bootleg-2020	708	50	1	1	CD
orr-bootleg-2020	708	51	.	.	.
orr-bootleg-2020	709	1	This	this	DT
orr-bootleg-2020	709	2	supports	support	VBZ
orr-bootleg-2020	709	3	the	the	DT
orr-bootleg-2020	709	4	need	need	NN
orr-bootleg-2020	709	5	for	for	IN
orr-bootleg-2020	709	6	knowledge	knowledge	NN
orr-bootleg-2020	709	7	signals	signal	NNS
orr-bootleg-2020	709	8	with	with	IN
orr-bootleg-2020	709	9	distinct	distinct	JJ
orr-bootleg-2020	709	10	tails	tail	NNS
orr-bootleg-2020	709	11	,	,	,
orr-bootleg-2020	709	12	which	which	WDT
orr-bootleg-2020	709	13	can	can	MD
orr-bootleg-2020	709	14	be	be	VB
orr-bootleg-2020	709	15	assembled	assemble	VBN
orr-bootleg-2020	709	16	to	to	TO
orr-bootleg-2020	709	17	together	together	RB
orr-bootleg-2020	709	18	address	address	VB
orr-bootleg-2020	709	19	the	the	DT
orr-bootleg-2020	709	20	rare	rare	JJ
orr-bootleg-2020	709	21	examples	example	NNS
orr-bootleg-2020	709	22	.	.	.
orr-bootleg-2020	710	1	25	25	CD
orr-bootleg-2020	710	2	1	1	CD
orr-bootleg-2020	710	3	Introduction	introduction	NN
orr-bootleg-2020	710	4	2	2	CD
orr-bootleg-2020	710	5	NED	NED	NNP
orr-bootleg-2020	710	6	Overview	Overview	NNP
orr-bootleg-2020	710	7	and	and	CC
orr-bootleg-2020	710	8	Reasoning	Reasoning	NNP
orr-bootleg-2020	710	9	Patterns	Patterns	NNPS
orr-bootleg-2020	710	10	2.1	2.1	CD
orr-bootleg-2020	710	11	Four	four	CD
orr-bootleg-2020	710	12	Reasoning	Reasoning	NNP
orr-bootleg-2020	710	13	Patterns	Patterns	NNPS
orr-bootleg-2020	710	14	3	3	CD
orr-bootleg-2020	710	15	Bootleg	Bootleg	NNP
orr-bootleg-2020	710	16	Architecture	Architecture	NNP
orr-bootleg-2020	710	17	for	for	IN
orr-bootleg-2020	710	18	Tail	Tail	NNP
orr-bootleg-2020	710	19	Disambiguation	Disambiguation	NNP
orr-bootleg-2020	710	20	3.1	3.1	CD
orr-bootleg-2020	710	21	Encoding	encode	VBG
orr-bootleg-2020	710	22	the	the	DT
orr-bootleg-2020	710	23	Signals	signal	NNS
orr-bootleg-2020	710	24	3.2	3.2	CD
orr-bootleg-2020	710	25	Bootleg	Bootleg	NNP
orr-bootleg-2020	710	26	Model	Model	NNP
orr-bootleg-2020	710	27	Architecture	Architecture	NNP
orr-bootleg-2020	710	28	3.3	3.3	CD
orr-bootleg-2020	710	29	Improving	improve	VBG
orr-bootleg-2020	710	30	Tail	Tail	NNP
orr-bootleg-2020	710	31	Generalization	Generalization	NNP
orr-bootleg-2020	710	32	3.3.1	3.3.1	CD
orr-bootleg-2020	710	33	Regularization	Regularization	NNP
orr-bootleg-2020	710	34	3.3.2	3.3.2	NNP
orr-bootleg-2020	710	35	Weakly	Weakly	NNP
orr-bootleg-2020	710	36	Supervised	supervise	VBD
orr-bootleg-2020	710	37	Data	Data	NNP
orr-bootleg-2020	710	38	Labeling	Labeling	NNP
orr-bootleg-2020	710	39	4	4	CD
orr-bootleg-2020	710	40	Experiments	experiment	NNS
orr-bootleg-2020	710	41	4.1	4.1	CD
orr-bootleg-2020	710	42	Experimental	experimental	JJ
orr-bootleg-2020	710	43	Setup	setup	NN
orr-bootleg-2020	710	44	4.2	4.2	CD
orr-bootleg-2020	710	45	Bootleg	Bootleg	NNP
orr-bootleg-2020	710	46	Performance	Performance	NNP
orr-bootleg-2020	710	47	4.3	4.3	CD
orr-bootleg-2020	710	48	Downstream	Downstream	NNP
orr-bootleg-2020	710	49	Evaluation	Evaluation	NNP
orr-bootleg-2020	710	50	4.4	4.4	CD
orr-bootleg-2020	710	51	Memory	Memory	NNP
orr-bootleg-2020	710	52	Usage	usage	NN
orr-bootleg-2020	710	53	4.5	4.5	CD
orr-bootleg-2020	710	54	Ablation	Ablation	NNP
orr-bootleg-2020	710	55	Study	Study	NNP
orr-bootleg-2020	710	56	5	5	CD
orr-bootleg-2020	710	57	Analysis	Analysis	NNP
orr-bootleg-2020	710	58	6	6	CD
orr-bootleg-2020	710	59	Related	Related	NNP
orr-bootleg-2020	710	60	Work	work	NN
orr-bootleg-2020	710	61	7	7	CD
orr-bootleg-2020	710	62	Conclusion	conclusion	NN
orr-bootleg-2020	710	63	A	a	DT
orr-bootleg-2020	710	64	Extended	Extended	NNP
orr-bootleg-2020	710	65	Model	Model	NNP
orr-bootleg-2020	710	66	Details	detail	NNS
orr-bootleg-2020	710	67	B	B	NNP
orr-bootleg-2020	710	68	Extended	extend	VBN
orr-bootleg-2020	710	69	Results	Results	NNP
orr-bootleg-2020	710	70	B.1	b.1	NN
orr-bootleg-2020	710	71	Evaluation	evaluation	NN
orr-bootleg-2020	710	72	Data	datum	NNS
orr-bootleg-2020	710	73	B.2	b.2	NN
orr-bootleg-2020	710	74	Training	train	VBG
orr-bootleg-2020	710	75	Details	detail	NNS
orr-bootleg-2020	710	76	B.3	b.3	IN
orr-bootleg-2020	710	77	Extended	Extended	NNP
orr-bootleg-2020	710	78	Ablation	ablation	NN
orr-bootleg-2020	710	79	Results	result	NNS
orr-bootleg-2020	710	80	C	c	NN
orr-bootleg-2020	710	81	Extended	extend	VBN
orr-bootleg-2020	710	82	Downstream	Downstream	NNP
orr-bootleg-2020	710	83	Details	detail	NNS
orr-bootleg-2020	710	84	D	D	NNP
orr-bootleg-2020	710	85	Extended	Extended	NNP
orr-bootleg-2020	710	86	Error	Error	NNP
orr-bootleg-2020	710	87	Analysis	Analysis	NNP
orr-bootleg-2020	710	88	D.1	D.1	NNP
orr-bootleg-2020	710	89	Reasoning	Reasoning	NNP
orr-bootleg-2020	710	90	Patterns	Patterns	NNPS