id	sid	tid	token	lemma	pos
lesk	1	1	Fragility	fragility	NN
lesk	1	2	and	and	CC
lesk	1	3	Intelligibility	Intelligibility	NNP
lesk	1	4	of	of	IN
lesk	1	5	Deep	Deep	NNP
lesk	1	6	Learning	Learning	NNP
lesk	1	7	for	for	IN
lesk	1	8	Libraries	library	NNS
lesk	1	9	Michael	Michael	NNP
lesk	1	10	Lesk	Lesk	NNP
lesk	1	11	Rutgers	Rutgers	NNP
lesk	1	12	University	University	NNP
lesk	1	13	Abstract	Abstract	NNP
lesk	1	14	:	:	:
lesk	1	15	Comment	comment	NN
lesk	1	16	by	by	IN
lesk	1	17	Eric	Eric	NNP
lesk	1	18	Lease	Lease	NNP
lesk	1	19	Morgan	Morgan	NNP
lesk	1	20	:	:	:
lesk	1	21	After	after	IN
lesk	1	22	reading	read	VBG
lesk	1	23	this	this	DT
lesk	1	24	second	second	JJ
lesk	1	25	draft	draft	NN
lesk	1	26	,	,	,
lesk	1	27	I	-PRON-	PRP
lesk	1	28	have	have	VBP
lesk	1	29	nothing	nothing	NN
lesk	1	30	to	to	TO
lesk	1	31	add	add	VB
lesk	1	32	;	;	:
lesk	1	33	in	in	IN
lesk	1	34	my	-PRON-	PRP$
lesk	1	35	mind	mind	NN
lesk	1	36	the	the	DT
lesk	1	37	article	article	NN
lesk	1	38	is	be	VBZ
lesk	1	39	sound	sound	JJ
lesk	1	40	.	.	.
lesk	2	1	More	more	RBR
lesk	2	2	specifically	specifically	RB
lesk	2	3	,	,	,
lesk	2	4	the	the	DT
lesk	2	5	article	article	NN
lesk	2	6	describes	describe	VBZ
lesk	2	7	why	why	WRB
lesk	2	8	one	one	NN
lesk	2	9	ought	ought	MD
lesk	2	10	to	to	TO
lesk	2	11	be	be	VB
lesk	2	12	a	a	DT
lesk	2	13	bit	bit	NN
lesk	2	14	apprehensive	apprehensive	JJ
lesk	2	15	when	when	WRB
lesk	2	16	it	-PRON-	PRP
lesk	2	17	comes	come	VBZ
lesk	2	18	to	to	IN
lesk	2	19	the	the	DT
lesk	2	20	conclusions	conclusion	NNS
lesk	2	21	of	of	IN
lesk	2	22	"	"	``
lesk	2	23	deep	deep	JJ
lesk	2	24	learning	learning	NN
lesk	2	25	"	"	''
lesk	2	26	.	.	.
lesk	3	1	It	-PRON-	PRP
lesk	3	2	does	do	VBZ
lesk	3	3	so	so	RB
lesk	3	4	with	with	IN
lesk	3	5	numerous	numerous	JJ
lesk	3	6	examples	example	NNS
lesk	3	7	and	and	CC
lesk	3	8	in	in	IN
lesk	3	9	a	a	DT
lesk	3	10	scholarly	scholarly	JJ
lesk	3	11	fashion	fashion	NN
lesk	3	12	.	.	.
lesk	4	1	Yet	yet	RB
lesk	4	2	,	,	,
lesk	4	3	the	the	DT
lesk	4	4	article	article	NN
lesk	4	5	does	do	VBZ
lesk	4	6	not	not	RB
lesk	4	7	despair	despair	VB
lesk	4	8	.	.	.
lesk	5	1	Instead	instead	RB
lesk	5	2	,	,	,
lesk	5	3	in	in	IN
lesk	5	4	the	the	DT
lesk	5	5	end	end	NN
lesk	5	6	,	,	,
lesk	5	7	it	-PRON-	PRP
lesk	5	8	offers	offer	VBZ
lesk	5	9	recommendations	recommendation	NNS
lesk	5	10	and	and	CC
lesk	5	11	suggestions	suggestion	NNS
lesk	5	12	for	for	IN
lesk	5	13	overcoming	overcome	VBG
lesk	5	14	the	the	DT
lesk	5	15	apprehension	apprehension	NN
lesk	5	16	.	.	.
lesk	6	1	We	-PRON-	PRP
lesk	6	2	are	be	VBP
lesk	6	3	reading	read	VBG
lesk	6	4	more	more	JJR
lesk	6	5	and	and	CC
lesk	6	6	more	more	JJR
lesk	6	7	about	about	IN
lesk	6	8	practical	practical	JJ
lesk	6	9	use	use	NN
lesk	6	10	of	of	IN
lesk	6	11	artificial	artificial	JJ
lesk	6	12	intelligence	intelligence	NN
lesk	6	13	,	,	,
lesk	6	14	ranging	range	VBG
lesk	6	15	from	from	IN
lesk	6	16	self	self	NN
lesk	6	17	-	-	HYPH
lesk	6	18	driving	drive	VBG
lesk	6	19	cars	car	NNS
lesk	6	20	to	to	IN
lesk	6	21	bank	bank	NN
lesk	6	22	loan	loan	NN
lesk	6	23	approvals	approval	NNS
lesk	6	24	.	.	.
lesk	7	1	Increasingly	increasingly	RB
lesk	7	2	,	,	,
lesk	7	3	we	-PRON-	PRP
lesk	7	4	are	be	VBP
lesk	7	5	asked	ask	VBN
lesk	7	6	to	to	TO
lesk	7	7	trust	trust	VB
lesk	7	8	systems	system	NNS
lesk	7	9	that	that	IN
lesk	7	10	we	-PRON-	PRP
lesk	7	11	can	can	MD
lesk	7	12	not	not	RB
lesk	7	13	understand	understand	VB
lesk	7	14	.	.	.
lesk	8	1	Can	Can	MD
lesk	8	2	we	-PRON-	PRP
lesk	8	3	rely	rely	VB
lesk	8	4	on	on	IN
lesk	8	5	such	such	JJ
lesk	8	6	software	software	NN
lesk	8	7	?	?	.
lesk	9	1	This	this	DT
lesk	9	2	article	article	NN
lesk	9	3	talks	talk	VBZ
lesk	9	4	about	about	IN
lesk	9	5	the	the	DT
lesk	9	6	risks	risk	NNS
lesk	9	7	that	that	WDT
lesk	9	8	fragile	fragile	JJ
lesk	9	9	AI	AI	NNP
lesk	9	10	software	software	NN
lesk	9	11	will	will	MD
lesk	9	12	be	be	VB
lesk	9	13	unreliable	unreliable	JJ
lesk	9	14	and	and	CC
lesk	9	15	that	that	IN
lesk	9	16	we	-PRON-	PRP
lesk	9	17	wo	will	MD
lesk	9	18	n’t	not	RB
lesk	9	19	be	be	VB
lesk	9	20	able	able	JJ
lesk	9	21	to	to	TO
lesk	9	22	know	know	VB
lesk	9	23	when	when	WRB
lesk	9	24	or	or	CC
lesk	9	25	why	why	WRB
lesk	9	26	.	.	.
lesk	10	1	Traditional	traditional	JJ
lesk	10	2	machine	machine	NN
lesk	10	3	learning	learning	NN
lesk	10	4	involved	involve	VBN
lesk	10	5	feature	feature	NN
lesk	10	6	detection	detection	NN
lesk	10	7	and	and	CC
lesk	10	8	then	then	RB
lesk	10	9	some	some	DT
lesk	10	10	variety	variety	NN
lesk	10	11	of	of	IN
lesk	10	12	clustering	clustering	NN
lesk	10	13	or	or	CC
lesk	10	14	discrimination	discrimination	NN
lesk	10	15	rules	rule	NNS
lesk	10	16	.	.	.
lesk	11	1	It	-PRON-	PRP
lesk	11	2	was	be	VBD
lesk	11	3	generally	generally	RB
lesk	11	4	possible	possible	JJ
lesk	11	5	to	to	TO
lesk	11	6	explain	explain	VB
lesk	11	7	what	what	WP
lesk	11	8	features	feature	NNS
lesk	11	9	were	be	VBD
lesk	11	10	being	be	VBG
lesk	11	11	used	use	VBN
lesk	11	12	and	and	CC
lesk	11	13	how	how	WRB
lesk	11	14	they	-PRON-	PRP
lesk	11	15	were	be	VBD
lesk	11	16	related	relate	VBN
lesk	11	17	to	to	IN
lesk	11	18	the	the	DT
lesk	11	19	task	task	NN
lesk	11	20	.	.	.
lesk	12	1	These	these	DT
lesk	12	2	methods	method	NNS
lesk	12	3	are	be	VBP
lesk	12	4	now	now	RB
lesk	12	5	being	be	VBG
lesk	12	6	supplanted	supplant	VBN
lesk	12	7	by	by	IN
lesk	12	8	"	"	``
lesk	12	9	deep	deep	JJ
lesk	12	10	learning	learning	NN
lesk	12	11	.	.	.
lesk	12	12	"	"	''
lesk	13	1	Enormous	enormous	JJ
lesk	13	2	quantities	quantity	NNS
lesk	13	3	of	of	IN
lesk	13	4	data	datum	NNS
lesk	13	5	are	be	VBP
lesk	13	6	sometimes	sometimes	RB
lesk	13	7	available	available	JJ
lesk	13	8	,	,	,
lesk	13	9	which	which	WDT
lesk	13	10	seems	seem	VBZ
lesk	13	11	to	to	TO
lesk	13	12	substitute	substitute	VB
lesk	13	13	for	for	IN
lesk	13	14	analyzing	analyze	VBG
lesk	13	15	the	the	DT
lesk	13	16	problem	problem	NN
lesk	13	17	.	.	.
lesk	14	1	In	in	IN
lesk	14	2	a	a	DT
lesk	14	3	famous	famous	JJ
lesk	14	4	paper	paper	NN
lesk	14	5	,	,	,
lesk	14	6	Peter	Peter	NNP
lesk	14	7	Norvig	Norvig	NNP
lesk	14	8	said	say	VBD
lesk	14	9	that	that	IN
lesk	14	10	more	more	JJR
lesk	14	11	data	data	NN
lesk	14	12	beats	beat	VBZ
lesk	14	13	better	well	JJR
lesk	14	14	algorithms	algorithm	NNS
lesk	14	15	.	.	.
lesk	15	1	[	[	-LRB-
lesk	15	2	footnoteRef:0	footnoteref:0	NN
lesk	15	3	]	]	-RRB-
lesk	15	4	Comment	Comment	NNP
lesk	15	5	by	by	IN
lesk	15	6	Daniel	Daniel	NNP
lesk	15	7	Johnson	Johnson	NNP
lesk	15	8	:	:	:
lesk	15	9	(	(	-LRB-
lesk	15	10	converted	convert	VBN
lesk	15	11	bibliographic	bibliographic	NNP
lesk	15	12	note	note	NN
lesk	15	13	to	to	TO
lesk	15	14	footnote	footnote	VB
lesk	15	15	,	,	,
lesk	15	16	per	per	IN
lesk	15	17	our	-PRON-	PRP$
lesk	15	18	use	use	NN
lesk	15	19	of	of	IN
lesk	15	20	Chicago	Chicago	NNP
lesk	15	21	)	)	-RRB-
lesk	15	22	[	[	-LRB-
lesk	15	23	0	0	CD
lesk	15	24	:	:	:
lesk	15	25	See	see	VB
lesk	15	26	Halevy	Halevy	NNP
lesk	15	27	,	,	,
lesk	15	28	Norvig	Norvig	NNP
lesk	15	29	,	,	,
lesk	15	30	and	and	CC
lesk	15	31	Pereira	Pereira	NNP
lesk	15	32	2018	2018	CD
lesk	15	33	,	,	,
lesk	15	34	but	but	CC
lesk	15	35	the	the	DT
lesk	15	36	exact	exact	JJ
lesk	15	37	quote	quote	NN
lesk	15	38	is	be	VBZ
lesk	15	39	in	in	IN
lesk	15	40	a	a	DT
lesk	15	41	talk	talk	NN
lesk	15	42	,	,	,
lesk	15	43	not	not	RB
lesk	15	44	this	this	DT
lesk	15	45	article	article	NN
lesk	15	46	.	.	.
lesk	15	47	]	]	-RRB-
lesk	16	1	But	but	CC
lesk	16	2	if	if	IN
lesk	16	3	software	software	NN
lesk	16	4	has	have	VBZ
lesk	16	5	been	be	VBN
lesk	16	6	trained	train	VBN
lesk	16	7	on	on	IN
lesk	16	8	a	a	DT
lesk	16	9	small	small	JJ
lesk	16	10	and	and	CC
lesk	16	11	specific	specific	JJ
lesk	16	12	subset	subset	NN
lesk	16	13	,	,	,
lesk	16	14	will	will	MD
lesk	16	15	it	-PRON-	PRP
lesk	16	16	generalize	generalize	VB
lesk	16	17	?	?	.
lesk	17	1	Will	Will	MD
lesk	17	2	face	face	VB
lesk	17	3	recognition	recognition	NN
lesk	17	4	software	software	NN
lesk	17	5	trained	train	VBN
lesk	17	6	on	on	IN
lesk	17	7	adult	adult	NN
lesk	17	8	white	white	JJ
lesk	17	9	male	male	JJ
lesk	17	10	faces	face	NNS
lesk	17	11	work	work	VB
lesk	17	12	with	with	IN
lesk	17	13	other	other	JJ
lesk	17	14	races	race	NNS
lesk	17	15	,	,	,
lesk	17	16	with	with	IN
lesk	17	17	women	woman	NNS
lesk	17	18	,	,	,
lesk	17	19	and	and	CC
lesk	17	20	with	with	IN
lesk	17	21	children	child	NNS
lesk	17	22	?	?	.
lesk	18	1	Even	even	RB
lesk	18	2	without	without	IN
lesk	18	3	generalizing	generalize	VBG
lesk	18	4	,	,	,
lesk	18	5	if	if	IN
lesk	18	6	we	-PRON-	PRP
lesk	18	7	do	do	VBP
lesk	18	8	n't	not	RB
lesk	18	9	know	know	VB
lesk	18	10	what	what	WP
lesk	18	11	is	be	VBZ
lesk	18	12	driving	drive	VBG
lesk	18	13	the	the	DT
lesk	18	14	decision	decision	NN
lesk	18	15	algorithm	algorithm	NN
lesk	18	16	,	,	,
lesk	18	17	we	-PRON-	PRP
lesk	18	18	may	may	MD
lesk	18	19	find	find	VB
lesk	18	20	that	that	IN
lesk	18	21	the	the	DT
lesk	18	22	conclusions	conclusion	NNS
lesk	18	23	are	be	VBP
lesk	18	24	very	very	RB
lesk	18	25	fragile	fragile	JJ
lesk	18	26	.	.	.
lesk	19	1	If	if	IN
lesk	19	2	image	image	NN
lesk	19	3	recognition	recognition	NN
lesk	19	4	systems	system	NNS
lesk	19	5	can	can	MD
lesk	19	6	be	be	VB
lesk	19	7	fooled	fool	VBN
lesk	19	8	by	by	IN
lesk	19	9	slight	slight	JJ
lesk	19	10	changes	change	NNS
lesk	19	11	in	in	IN
lesk	19	12	an	an	DT
lesk	19	13	image	image	NN
lesk	19	14	,	,	,
lesk	19	15	what	what	WP
lesk	19	16	happens	happen	VBZ
lesk	19	17	if	if	IN
lesk	19	18	self	self	NN
lesk	19	19	-	-	HYPH
lesk	19	20	driving	drive	VBG
lesk	19	21	cars	car	NNS
lesk	19	22	misread	misread	JJ
lesk	19	23	road	road	NN
lesk	19	24	signs	sign	NNS
lesk	19	25	?	?	.
lesk	20	1	1	1	LS
lesk	20	2	.	.	.
lesk	21	1	Introduction	introduction	NN
lesk	21	2	.	.	.
lesk	22	1	On	on	IN
lesk	22	2	February	February	NNP
lesk	22	3	7	7	CD
lesk	22	4	,	,	,
lesk	22	5	2018	2018	CD
lesk	22	6	,	,	,
lesk	22	7	Mounir	Mounir	NNP
lesk	22	8	Mahjoubi	Mahjoubi	NNP
lesk	22	9	,	,	,
lesk	22	10	then	then	RB
lesk	22	11	the	the	DT
lesk	22	12	“	"	``
lesk	22	13	digital	digital	NNP
lesk	22	14	minister	minister	NN
lesk	22	15	”	"	''
lesk	22	16	of	of	IN
lesk	22	17	France	France	NNP
lesk	22	18	(	(	-LRB-
lesk	22	19	le	le	NNP
lesk	22	20	secrétariat	secrétariat	NNP
lesk	22	21	d’État	d’État	NNP
lesk	22	22	chargé	chargé	NNS
lesk	22	23	du	du	NNP
lesk	22	24	Numérique	Numérique	NNP
lesk	22	25	)	)	-RRB-
lesk	22	26	,	,	,
lesk	22	27	told	tell	VBD
lesk	22	28	the	the	DT
lesk	22	29	civil	civil	JJ
lesk	22	30	service	service	NN
lesk	22	31	to	to	TO
lesk	22	32	use	use	VB
lesk	22	33	only	only	JJ
lesk	22	34	computer	computer	NN
lesk	22	35	methods	method	NNS
lesk	22	36	that	that	WDT
lesk	22	37	could	could	MD
lesk	22	38	be	be	VB
lesk	22	39	understood	understand	VBN
lesk	22	40	(	(	-LRB-
lesk	22	41	Mahjoubi	Mahjoubi	NNP
lesk	22	42	2018	2018	CD
lesk	22	43	)	)	-RRB-
lesk	22	44	.	.	.
lesk	23	1	To	to	TO
lesk	23	2	be	be	VB
lesk	23	3	precise	precise	JJ
lesk	23	4	,	,	,
lesk	23	5	what	what	WP
lesk	23	6	he	-PRON-	PRP
lesk	23	7	actually	actually	RB
lesk	23	8	said	say	VBD
lesk	23	9	to	to	IN
lesk	23	10	l'Assemblée	l'Assemblée	NNP
lesk	23	11	Nationale	Nationale	NNP
lesk	23	12	was	be	VBD
lesk	23	13	:	:	:
lesk	23	14	Aucun	Aucun	NNP
lesk	23	15	algorithme	algorithme	FW
lesk	23	16	non	non	AFX
lesk	23	17	explicable	explicable	JJ
lesk	23	18	ne	ne	NNP
lesk	23	19	pourra	pourra	NNP
lesk	23	20	être	être	NNP
lesk	23	21	utilisé	utilisé	NNP
lesk	23	22	.	.	.
lesk	24	1	I	-PRON-	PRP
lesk	24	2	gave	give	VBD
lesk	24	3	this	this	DT
lesk	24	4	to	to	IN
lesk	24	5	Google	Google	NNP
lesk	24	6	Translate	Translate	NNP
lesk	24	7	and	and	CC
lesk	24	8	asked	ask	VBD
lesk	24	9	for	for	IN
lesk	24	10	it	-PRON-	PRP
lesk	24	11	in	in	IN
lesk	24	12	English	English	NNP
lesk	24	13	.	.	.
lesk	25	1	What	what	WP
lesk	25	2	I	-PRON-	PRP
lesk	25	3	got	get	VBD
lesk	25	4	(	(	-LRB-
lesk	25	5	on	on	IN
lesk	25	6	October	October	NNP
lesk	25	7	13	13	CD
lesk	25	8	,	,	,
lesk	25	9	2019	2019	CD
lesk	25	10	)	)	-RRB-
lesk	25	11	was	be	VBD
lesk	25	12	:	:	:
lesk	25	13	No	no	UH
lesk	25	14	algorithm	algorithm	NN
lesk	25	15	that	that	WDT
lesk	25	16	can	can	MD
lesk	25	17	not	not	RB
lesk	25	18	be	be	VB
lesk	25	19	explained	explain	VBN
lesk	25	20	can	can	MD
lesk	25	21	not	not	RB
lesk	25	22	be	be	VB
lesk	25	23	used	use	VBN
lesk	25	24	.	.	.
lesk	26	1	That	that	DT
lesk	26	2	’s	’	VBZ
lesk	26	3	a	a	DT
lesk	26	4	long	long	JJ
lesk	26	5	way	way	NN
lesk	26	6	from	from	IN
lesk	26	7	fluent	fluent	JJ
lesk	26	8	English	English	NNP
lesk	26	9	.	.	.
lesk	27	1	As	as	IN
lesk	27	2	I	-PRON-	PRP
lesk	27	3	count	count	VBP
lesk	27	4	the	the	DT
lesk	27	5	“	"	``
lesk	27	6	not	not	RB
lesk	27	7	”	"	''
lesk	27	8	words	word	NNS
lesk	27	9	,	,	,
lesk	27	10	it	-PRON-	PRP
lesk	27	11	’s	’	VBZ
lesk	27	12	actually	actually	RB
lesk	27	13	reversed	reverse	VBN
lesk	27	14	in	in	IN
lesk	27	15	meaning	meaning	NN
lesk	27	16	.	.	.
lesk	28	1	But	but	CC
lesk	28	2	,	,	,
lesk	28	3	what	what	WP
lesk	28	4	if	if	IN
lesk	28	5	I	-PRON-	PRP
lesk	28	6	leave	leave	VBP
lesk	28	7	off	off	RP
lesk	28	8	the	the	DT
lesk	28	9	final	final	JJ
lesk	28	10	period	period	NN
lesk	28	11	when	when	WRB
lesk	28	12	I	-PRON-	PRP
lesk	28	13	enter	enter	VBP
lesk	28	14	it	-PRON-	PRP
lesk	28	15	in	in	IN
lesk	28	16	Google	Google	NNP
lesk	28	17	Translate	Translate	NNP
lesk	28	18	?	?	.
lesk	29	1	Then	then	RB
lesk	29	2	I	-PRON-	PRP
lesk	29	3	get	get	VBP
lesk	29	4	:	:	:
lesk	29	5	No	no	DT
lesk	29	6	non	non	JJ
lesk	29	7	-	-	JJ
lesk	29	8	explainable	explainable	JJ
lesk	29	9	algorithm	algorithm	NN
lesk	29	10	can	can	MD
lesk	29	11	be	be	VB
lesk	29	12	used	use	VBN
lesk	29	13	Quite	quite	RB
lesk	29	14	different	different	JJ
lesk	29	15	,	,	,
lesk	29	16	and	and	CC
lesk	29	17	although	although	IN
lesk	29	18	only	only	RB
lesk	29	19	barely	barely	RB
lesk	29	20	fluent	fluent	JJ
lesk	29	21	,	,	,
lesk	29	22	now	now	RB
lesk	29	23	the	the	DT
lesk	29	24	meaning	meaning	NN
lesk	29	25	is	be	VBZ
lesk	29	26	right	right	JJ
lesk	29	27	.	.	.
lesk	30	1	The	the	DT
lesk	30	2	difference	difference	NN
lesk	30	3	was	be	VBD
lesk	30	4	only	only	RB
lesk	30	5	the	the	DT
lesk	30	6	final	final	JJ
lesk	30	7	punctuation	punctuation	NN
lesk	30	8	on	on	IN
lesk	30	9	the	the	DT
lesk	30	10	sentence	sentence	NN
lesk	30	11	.	.	.
lesk	31	1	This	this	DT
lesk	31	2	is	be	VBZ
lesk	31	3	an	an	DT
lesk	31	4	example	example	NN
lesk	31	5	of	of	IN
lesk	31	6	the	the	DT
lesk	31	7	fragility	fragility	NN
lesk	31	8	of	of	IN
lesk	31	9	an	an	DT
lesk	31	10	AI	AI	NNP
lesk	31	11	algorithm	algorithm	NN
lesk	31	12	.	.	.
lesk	32	1	The	the	DT
lesk	32	2	point	point	NN
lesk	32	3	is	be	VBZ
lesk	32	4	not	not	RB
lesk	32	5	that	that	IN
lesk	32	6	both	both	DT
lesk	32	7	translations	translation	NNS
lesk	32	8	are	be	VBP
lesk	32	9	of	of	IN
lesk	32	10	doubtful	doubtful	JJ
lesk	32	11	quality	quality	NN
lesk	32	12	.	.	.
lesk	33	1	The	the	DT
lesk	33	2	point	point	NN
lesk	33	3	is	be	VBZ
lesk	33	4	that	that	IN
lesk	33	5	a	a	DT
lesk	33	6	seemingly	seemingly	RB
lesk	33	7	insignificant	insignificant	JJ
lesk	33	8	change	change	NN
lesk	33	9	in	in	IN
lesk	33	10	the	the	DT
lesk	33	11	input	input	NN
lesk	33	12	produced	produce	VBD
lesk	33	13	such	such	PDT
lesk	33	14	a	a	DT
lesk	33	15	difference	difference	NN
lesk	33	16	in	in	IN
lesk	33	17	the	the	DT
lesk	33	18	output	output	NN
lesk	33	19	.	.	.
lesk	34	1	In	in	IN
lesk	34	2	this	this	DT
lesk	34	3	case	case	NN
lesk	34	4	,	,	,
lesk	34	5	the	the	DT
lesk	34	6	fragility	fragility	NN
lesk	34	7	was	be	VBD
lesk	34	8	detected	detect	VBN
lesk	34	9	by	by	IN
lesk	34	10	accident	accident	NN
lesk	34	11	.	.	.
lesk	35	1	Machine	machine	NN
lesk	35	2	learning	learning	NN
lesk	35	3	systems	system	NNS
lesk	35	4	have	have	VBP
lesk	35	5	a	a	DT
lesk	35	6	set	set	NN
lesk	35	7	of	of	IN
lesk	35	8	data	datum	NNS
lesk	35	9	for	for	IN
lesk	35	10	training	training	NN
lesk	35	11	.	.	.
lesk	36	1	For	for	IN
lesk	36	2	example	example	NN
lesk	36	3	,	,	,
lesk	36	4	if	if	IN
lesk	36	5	you	-PRON-	PRP
lesk	36	6	are	be	VBP
lesk	36	7	interested	interested	JJ
lesk	36	8	in	in	IN
lesk	36	9	translation	translation	NN
lesk	36	10	,	,	,
lesk	36	11	and	and	CC
lesk	36	12	you	-PRON-	PRP
lesk	36	13	have	have	VBP
lesk	36	14	a	a	DT
lesk	36	15	large	large	JJ
lesk	36	16	collection	collection	NN
lesk	36	17	of	of	IN
lesk	36	18	text	text	NN
lesk	36	19	in	in	IN
lesk	36	20	both	both	CC
lesk	36	21	French	French	NNP
lesk	36	22	and	and	CC
lesk	36	23	English	English	NNP
lesk	36	24	,	,	,
lesk	36	25	you	-PRON-	PRP
lesk	36	26	might	may	MD
lesk	36	27	notice	notice	VB
lesk	36	28	that	that	IN
lesk	36	29	the	the	DT
lesk	36	30	word	word	NN
lesk	36	31	truck	truck	NN
lesk	36	32	in	in	IN
lesk	36	33	English	English	NNP
lesk	36	34	appears	appear	VBZ
lesk	36	35	where	where	WRB
lesk	36	36	the	the	DT
lesk	36	37	word	word	NN
lesk	36	38	camion	camion	NN
lesk	36	39	appears	appear	VBZ
lesk	36	40	in	in	IN
lesk	36	41	French	French	NNP
lesk	36	42	.	.	.
lesk	37	1	And	and	CC
lesk	37	2	the	the	DT
lesk	37	3	system	system	NN
lesk	37	4	might	may	MD
lesk	37	5	“	"	``
lesk	37	6	learn	learn	VB
lesk	37	7	”	"	''
lesk	37	8	this	this	DT
lesk	37	9	translation	translation	NN
lesk	37	10	.	.	.
lesk	38	1	It	-PRON-	PRP
lesk	38	2	would	would	MD
lesk	38	3	then	then	RB
lesk	38	4	apply	apply	VB
lesk	38	5	this	this	DT
lesk	38	6	in	in	IN
lesk	38	7	other	other	JJ
lesk	38	8	examples	example	NNS
lesk	38	9	;	;	:
lesk	38	10	this	this	DT
lesk	38	11	is	be	VBZ
lesk	38	12	called	call	VBN
lesk	38	13	generalization	generalization	NN
lesk	38	14	.	.	.
lesk	39	1	Of	of	RB
lesk	39	2	course	course	RB
lesk	39	3	if	if	IN
lesk	39	4	you	-PRON-	PRP
lesk	39	5	wish	wish	VBP
lesk	39	6	to	to	TO
lesk	39	7	translate	translate	VB
lesk	39	8	French	French	NNP
lesk	39	9	into	into	IN
lesk	39	10	British	British	NNP
lesk	39	11	English	English	NNP
lesk	39	12	,	,	,
lesk	39	13	a	a	DT
lesk	39	14	preferred	preferred	JJ
lesk	39	15	translation	translation	NN
lesk	39	16	of	of	IN
lesk	39	17	camion	camion	NN
lesk	39	18	is	be	VBZ
lesk	39	19	lorry	lorry	NN
lesk	39	20	.	.	.
lesk	40	1	And	and	CC
lesk	40	2	if	if	IN
lesk	40	3	the	the	DT
lesk	40	4	context	context	NN
lesk	40	5	of	of	IN
lesk	40	6	your	-PRON-	PRP$
lesk	40	7	English	english	JJ
lesk	40	8	truck	truck	NN
lesk	40	9	is	be	VBZ
lesk	40	10	a	a	DT
lesk	40	11	US	US	NNP
lesk	40	12	discussion	discussion	NN
lesk	40	13	of	of	IN
lesk	40	14	the	the	DT
lesk	40	15	wheels	wheel	NNS
lesk	40	16	and	and	CC
lesk	40	17	axles	axle	NNS
lesk	40	18	underneath	underneath	IN
lesk	40	19	railway	railway	NN
lesk	40	20	vehicles	vehicle	NNS
lesk	40	21	,	,	,
lesk	40	22	the	the	DT
lesk	40	23	better	well	JJR
lesk	40	24	French	french	JJ
lesk	40	25	word	word	NN
lesk	40	26	is	be	VBZ
lesk	40	27	le	le	NNP
lesk	40	28	bogie	bogie	NN
lesk	40	29	.	.	.
lesk	41	1	Deep	deep	JJ
lesk	41	2	learning	learning	NN
lesk	41	3	enthusiasts	enthusiast	NNS
lesk	41	4	believe	believe	VBP
lesk	41	5	that	that	IN
lesk	41	6	with	with	IN
lesk	41	7	enough	enough	JJ
lesk	41	8	examples	example	NNS
lesk	41	9	,	,	,
lesk	41	10	machine	machine	NN
lesk	41	11	learning	learning	NN
lesk	41	12	systems	system	NNS
lesk	41	13	will	will	MD
lesk	41	14	be	be	VB
lesk	41	15	able	able	JJ
lesk	41	16	to	to	TO
lesk	41	17	generalize	generalize	VB
lesk	41	18	correctly	correctly	RB
lesk	41	19	.	.	.
lesk	42	1	There	there	EX
lesk	42	2	can	can	MD
lesk	42	3	be	be	VB
lesk	42	4	various	various	JJ
lesk	42	5	kinds	kind	NNS
lesk	42	6	of	of	IN
lesk	42	7	failures	failure	NNS
lesk	42	8	:	:	:
lesk	42	9	we	-PRON-	PRP
lesk	42	10	can	can	MD
lesk	42	11	discuss	discuss	VB
lesk	42	12	both	both	DT
lesk	42	13	(	(	-LRB-
lesk	42	14	a	a	DT
lesk	42	15	)	)	-RRB-
lesk	42	16	problems	problem	NNS
lesk	42	17	in	in	IN
lesk	42	18	the	the	DT
lesk	42	19	scope	scope	NN
lesk	42	20	of	of	IN
lesk	42	21	the	the	DT
lesk	42	22	training	training	NN
lesk	42	23	data	datum	NNS
lesk	42	24	and	and	CC
lesk	42	25	(	(	-LRB-
lesk	42	26	b	b	NN
lesk	42	27	)	)	-RRB-
lesk	42	28	problems	problem	NNS
lesk	42	29	in	in	IN
lesk	42	30	the	the	DT
lesk	42	31	kind	kind	NN
lesk	42	32	of	of	IN
lesk	42	33	modeling	modeling	NN
lesk	42	34	done	do	VBN
lesk	42	35	.	.	.
lesk	43	1	If	if	IN
lesk	43	2	the	the	DT
lesk	43	3	system	system	NN
lesk	43	4	has	have	VBZ
lesk	43	5	sufficiently	sufficiently	RB
lesk	43	6	general	general	JJ
lesk	43	7	input	input	NN
lesk	43	8	data	datum	NNS
lesk	43	9	so	so	IN
lesk	43	10	that	that	IN
lesk	43	11	it	-PRON-	PRP
lesk	43	12	learns	learn	VBZ
lesk	43	13	well	well	RB
lesk	43	14	enough	enough	RB
lesk	43	15	to	to	TO
lesk	43	16	produce	produce	VB
lesk	43	17	reliably	reliably	RB
lesk	43	18	correct	correct	JJ
lesk	43	19	results	result	NNS
lesk	43	20	on	on	IN
lesk	43	21	examples	example	NNS
lesk	43	22	it	-PRON-	PRP
lesk	43	23	has	have	VBZ
lesk	43	24	not	not	RB
lesk	43	25	seen	see	VBN
lesk	43	26	,	,	,
lesk	43	27	we	-PRON-	PRP
lesk	43	28	call	call	VBP
lesk	43	29	it	-PRON-	PRP
lesk	43	30	robust	robust	RB
lesk	43	31	;	;	:
lesk	43	32	robustness	robustness	NNP
lesk	43	33	is	be	VBZ
lesk	43	34	the	the	DT
lesk	43	35	opposite	opposite	NN
lesk	43	36	of	of	IN
lesk	43	37	fragility	fragility	NN
lesk	43	38	.	.	.
lesk	44	1	Fragility	fragility	NN
lesk	44	2	errors	error	NNS
lesk	44	3	here	here	RB
lesk	44	4	can	can	MD
lesk	44	5	arise	arise	VB
lesk	44	6	from	from	IN
lesk	44	7	many	many	JJ
lesk	44	8	sources	source	NNS
lesk	44	9	-	-	:
lesk	44	10	for	for	IN
lesk	44	11	example	example	NN
lesk	44	12	,	,	,
lesk	44	13	the	the	DT
lesk	44	14	training	training	NN
lesk	44	15	data	datum	NNS
lesk	44	16	may	may	MD
lesk	44	17	not	not	RB
lesk	44	18	be	be	VB
lesk	44	19	representative	representative	JJ
lesk	44	20	of	of	IN
lesk	44	21	the	the	DT
lesk	44	22	real	real	JJ
lesk	44	23	problem	problem	NN
lesk	44	24	(	(	-LRB-
lesk	44	25	if	if	IN
lesk	44	26	you	-PRON-	PRP
lesk	44	27	train	train	VBP
lesk	44	28	a	a	DT
lesk	44	29	machine	machine	NN
lesk	44	30	translation	translation	NN
lesk	44	31	program	program	NN
lesk	44	32	solely	solely	RB
lesk	44	33	on	on	IN
lesk	44	34	engineering	engineering	NN
lesk	44	35	documents	document	NNS
lesk	44	36	,	,	,
lesk	44	37	do	do	VBP
lesk	44	38	not	not	RB
lesk	44	39	expect	expect	VB
lesk	44	40	it	-PRON-	PRP
lesk	44	41	to	to	TO
lesk	44	42	do	do	VB
lesk	44	43	well	well	RB
lesk	44	44	on	on	IN
lesk	44	45	theater	theater	NN
lesk	44	46	reviews	review	NNS
lesk	44	47	)	)	-RRB-
lesk	44	48	.	.	.
lesk	45	1	Or	or	CC
lesk	45	2	,	,	,
lesk	45	3	the	the	DT
lesk	45	4	data	datum	NNS
lesk	45	5	may	may	MD
lesk	45	6	not	not	RB
lesk	45	7	have	have	VB
lesk	45	8	the	the	DT
lesk	45	9	scope	scope	NN
lesk	45	10	of	of	IN
lesk	45	11	the	the	DT
lesk	45	12	real	real	JJ
lesk	45	13	problem	problem	NN
lesk	45	14	:	:	:
lesk	45	15	if	if	IN
lesk	45	16	you	-PRON-	PRP
lesk	45	17	train	train	VBP
lesk	45	18	for	for	IN
lesk	45	19	“	"	``
lesk	45	20	boat	boat	NN
lesk	45	21	”	"	''
lesk	45	22	based	base	VBN
lesk	45	23	on	on	IN
lesk	45	24	ocean	ocean	NN
lesk	45	25	liners	liner	NNS
lesk	45	26	,	,	,
lesk	45	27	do	do	VB
lesk	45	28	n’t	not	RB
lesk	45	29	be	be	VB
lesk	45	30	surprised	surprised	JJ
lesk	45	31	if	if	IN
lesk	45	32	the	the	DT
lesk	45	33	program	program	NN
lesk	45	34	fails	fail	VBZ
lesk	45	35	on	on	IN
lesk	45	36	canoes	canoe	NNS
lesk	45	37	.	.	.
lesk	46	1	In	in	IN
lesk	46	2	addition	addition	NN
lesk	46	3	,	,	,
lesk	46	4	there	there	EX
lesk	46	5	are	be	VBP
lesk	46	6	also	also	RB
lesk	46	7	modeling	model	VBG
lesk	46	8	issues	issue	NNS
lesk	46	9	.	.	.
lesk	47	1	Suppose	suppose	VB
lesk	47	2	you	-PRON-	PRP
lesk	47	3	use	use	VBP
lesk	47	4	a	a	DT
lesk	47	5	very	very	RB
lesk	47	6	simple	simple	JJ
lesk	47	7	model	model	NN
lesk	47	8	,	,	,
lesk	47	9	such	such	JJ
lesk	47	10	as	as	IN
lesk	47	11	a	a	DT
lesk	47	12	linear	linear	JJ
lesk	47	13	model	model	NN
lesk	47	14	,	,	,
lesk	47	15	for	for	IN
lesk	47	16	data	datum	NNS
lesk	47	17	that	that	WDT
lesk	47	18	is	be	VBZ
lesk	47	19	actually	actually	RB
lesk	47	20	perhaps	perhaps	RB
lesk	47	21	quadratic	quadratic	JJ
lesk	47	22	or	or	CC
lesk	47	23	exponential	exponential	JJ
lesk	47	24	.	.	.
lesk	48	1	This	this	DT
lesk	48	2	is	be	VBZ
lesk	48	3	called	call	VBN
lesk	48	4	“	"	``
lesk	48	5	underfitting	underfitte	VBG
lesk	48	6	”	"	''
lesk	48	7	and	and	CC
lesk	48	8	may	may	MD
lesk	48	9	often	often	RB
lesk	48	10	arise	arise	VB
lesk	48	11	when	when	WRB
lesk	48	12	there	there	EX
lesk	48	13	is	be	VBZ
lesk	48	14	not	not	RB
lesk	48	15	enough	enough	JJ
lesk	48	16	training	training	NN
lesk	48	17	data	datum	NNS
lesk	48	18	.	.	.
lesk	49	1	The	the	DT
lesk	49	2	reverse	reverse	NN
lesk	49	3	is	be	VBZ
lesk	49	4	also	also	RB
lesk	49	5	possible	possible	JJ
lesk	49	6	:	:	:
lesk	49	7	there	there	EX
lesk	49	8	may	may	MD
lesk	49	9	be	be	VB
lesk	49	10	a	a	DT
lesk	49	11	lot	lot	NN
lesk	49	12	of	of	IN
lesk	49	13	training	training	NN
lesk	49	14	data	datum	NNS
lesk	49	15	,	,	,
lesk	49	16	including	include	VBG
lesk	49	17	many	many	JJ
lesk	49	18	noisy	noisy	JJ
lesk	49	19	points	point	NNS
lesk	49	20	,	,	,
lesk	49	21	and	and	CC
lesk	49	22	the	the	DT
lesk	49	23	program	program	NN
lesk	49	24	may	may	MD
lesk	49	25	decide	decide	VB
lesk	49	26	on	on	IN
lesk	49	27	a	a	DT
lesk	49	28	very	very	RB
lesk	49	29	complex	complex	JJ
lesk	49	30	model	model	NN
lesk	49	31	to	to	TO
lesk	49	32	cover	cover	VB
lesk	49	33	all	all	PDT
lesk	49	34	the	the	DT
lesk	49	35	noise	noise	NN
lesk	49	36	in	in	IN
lesk	49	37	the	the	DT
lesk	49	38	training	training	NN
lesk	49	39	data	datum	NNS
lesk	49	40	.	.	.
lesk	50	1	This	this	DT
lesk	50	2	is	be	VBZ
lesk	50	3	called	call	VBN
lesk	50	4	“	"	``
lesk	50	5	overfitting	overfitting	NN
lesk	50	6	”	"	''
lesk	50	7	and	and	CC
lesk	50	8	gives	give	VBZ
lesk	50	9	you	-PRON-	PRP
lesk	50	10	an	an	DT
lesk	50	11	answer	answer	NN
lesk	50	12	too	too	RB
lesk	50	13	dependent	dependent	JJ
lesk	50	14	on	on	IN
lesk	50	15	noise	noise	NN
lesk	50	16	and	and	CC
lesk	50	17	outliers	outlier	NNS
lesk	50	18	in	in	IN
lesk	50	19	your	-PRON-	PRP$
lesk	50	20	data	datum	NNS
lesk	50	21	.	.	.
lesk	51	1	For	for	IN
lesk	51	2	example	example	NN
lesk	51	3	,	,	,
lesk	51	4	1998	1998	CD
lesk	51	5	was	be	VBD
lesk	51	6	an	an	DT
lesk	51	7	unusually	unusually	RB
lesk	51	8	warm	warm	JJ
lesk	51	9	year	year	NN
lesk	51	10	,	,	,
lesk	51	11	but	but	CC
lesk	51	12	the	the	DT
lesk	51	13	decline	decline	NN
lesk	51	14	in	in	IN
lesk	51	15	world	world	NN
lesk	51	16	temperature	temperature	NN
lesk	51	17	for	for	IN
lesk	51	18	the	the	DT
lesk	51	19	next	next	JJ
lesk	51	20	few	few	JJ
lesk	51	21	years	year	NNS
lesk	51	22	suggests	suggest	VBZ
lesk	51	23	it	-PRON-	PRP
lesk	51	24	was	be	VBD
lesk	51	25	noise	noise	NN
lesk	51	26	in	in	IN
lesk	51	27	the	the	DT
lesk	51	28	data	datum	NNS
lesk	51	29	,	,	,
lesk	51	30	not	not	RB
lesk	51	31	a	a	DT
lesk	51	32	change	change	NN
lesk	51	33	in	in	IN
lesk	51	34	the	the	DT
lesk	51	35	development	development	NN
lesk	51	36	of	of	IN
lesk	51	37	climate	climate	NN
lesk	51	38	.	.	.
lesk	52	1	Fragility	fragility	NN
lesk	52	2	is	be	VBZ
lesk	52	3	also	also	RB
lesk	52	4	a	a	DT
lesk	52	5	problem	problem	NN
lesk	52	6	in	in	IN
lesk	52	7	image	image	NN
lesk	52	8	recognition	recognition	NN
lesk	52	9	(	(	-LRB-
lesk	52	10	“	"	``
lesk	52	11	AI	AI	NNP
lesk	52	12	Recognition	Recognition	NNP
lesk	52	13	”	"	''
lesk	52	14	2017	2017	CD
lesk	52	15	)	)	-RRB-
lesk	52	16	.	.	.
lesk	53	1	Currently	currently	RB
lesk	53	2	the	the	DT
lesk	53	3	most	most	RBS
lesk	53	4	common	common	JJ
lesk	53	5	technique	technique	NN
lesk	53	6	for	for	IN
lesk	53	7	image	image	NN
lesk	53	8	recognition	recognition	NN
lesk	53	9	research	research	NN
lesk	53	10	projects	project	NNS
lesk	53	11	is	be	VBZ
lesk	53	12	the	the	DT
lesk	53	13	use	use	NN
lesk	53	14	of	of	IN
lesk	53	15	convolutional	convolutional	JJ
lesk	53	16	neural	neural	JJ
lesk	53	17	nets	net	NNS
lesk	53	18	.	.	.
lesk	54	1	Recently	recently	RB
lesk	54	2	,	,	,
lesk	54	3	several	several	JJ
lesk	54	4	papers	paper	NNS
lesk	54	5	have	have	VBP
lesk	54	6	looked	look	VBN
lesk	54	7	at	at	IN
lesk	54	8	how	how	WRB
lesk	54	9	trivial	trivial	JJ
lesk	54	10	modifications	modification	NNS
lesk	54	11	to	to	IN
lesk	54	12	images	image	NNS
lesk	54	13	may	may	MD
lesk	54	14	impact	impact	VB
lesk	54	15	image	image	NN
lesk	54	16	classification	classification	NN
lesk	54	17	.	.	.
lesk	55	1	Here	here	RB
lesk	55	2	is	be	VBZ
lesk	55	3	an	an	DT
lesk	55	4	image	image	NN
lesk	55	5	taken	take	VBN
lesk	55	6	from	from	IN
lesk	55	7	(	(	-LRB-
lesk	55	8	Su	Su	NNP
lesk	55	9	,	,	,
lesk	55	10	Vargas	Vargas	NNP
lesk	55	11	,	,	,
lesk	55	12	and	and	CC
lesk	55	13	Sakurai	Sakurai	NNP
lesk	55	14	2019	2019	CD
lesk	55	15	)	)	-RRB-
lesk	55	16	.	.	.
lesk	56	1	The	the	DT
lesk	56	2	original	original	JJ
lesk	56	3	image	image	NN
lesk	56	4	class	class	NN
lesk	56	5	is	be	VBZ
lesk	56	6	in	in	IN
lesk	56	7	black	black	JJ
lesk	56	8	and	and	CC
lesk	56	9	the	the	DT
lesk	56	10	classifier	classifier	NN
lesk	56	11	choice	choice	NN
lesk	56	12	(	(	-LRB-
lesk	56	13	and	and	CC
lesk	56	14	confidence	confidence	NN
lesk	56	15	)	)	-RRB-
lesk	56	16	after	after	IN
lesk	56	17	adding	add	VBG
lesk	56	18	a	a	DT
lesk	56	19	single	single	JJ
lesk	56	20	unusual	unusual	JJ
lesk	56	21	pixel	pixel	NN
lesk	56	22	are	be	VBP
lesk	56	23	shown	show	VBN
lesk	56	24	in	in	IN
lesk	56	25	blue	blue	NNP
lesk	56	26	,	,	,
lesk	56	27	with	with	IN
lesk	56	28	the	the	DT
lesk	56	29	extraneous	extraneous	JJ
lesk	56	30	pixel	pixel	NN
lesk	56	31	in	in	IN
lesk	56	32	white	white	NNP
lesk	56	33	.	.	.
lesk	57	1	The	the	DT
lesk	57	2	images	image	NNS
lesk	57	3	were	be	VBD
lesk	57	4	deliberately	deliberately	RB
lesk	57	5	processed	process	VBN
lesk	57	6	at	at	IN
lesk	57	7	low	low	JJ
lesk	57	8	resolution	resolution	NN
lesk	57	9	-	-	:
lesk	57	10	hence	hence	RB
lesk	57	11	the	the	DT
lesk	57	12	pixellation	pixellation	NN
lesk	57	13	-	-	:
lesk	57	14	to	to	TO
lesk	57	15	match	match	VB
lesk	57	16	the	the	DT
lesk	57	17	input	input	NN
lesk	57	18	requirement	requirement	NN
lesk	57	19	of	of	IN
lesk	57	20	a	a	DT
lesk	57	21	popular	popular	JJ
lesk	57	22	image	image	NN
lesk	57	23	classification	classification	NN
lesk	57	24	program	program	NN
lesk	57	25	.	.	.
lesk	58	1	The	the	DT
lesk	58	2	authors	author	NNS
lesk	58	3	experimented	experiment	VBD
lesk	58	4	with	with	IN
lesk	58	5	algorithms	algorithm	NNS
lesk	58	6	to	to	TO
lesk	58	7	find	find	VB
lesk	58	8	the	the	DT
lesk	58	9	quickest	quick	JJS
lesk	58	10	single	single	JJ
lesk	58	11	-	-	HYPH
lesk	58	12	pixel	pixel	NN
lesk	58	13	change	change	NN
lesk	58	14	that	that	WDT
lesk	58	15	would	would	MD
lesk	58	16	deceive	deceive	VB
lesk	58	17	an	an	DT
lesk	58	18	image	image	NN
lesk	58	19	classifier	classifier	NN
lesk	58	20	.	.	.
lesk	59	1	They	-PRON-	PRP
lesk	59	2	were	be	VBD
lesk	59	3	routinely	routinely	RB
lesk	59	4	able	able	JJ
lesk	59	5	to	to	TO
lesk	59	6	fool	fool	VB
lesk	59	7	the	the	DT
lesk	59	8	recognition	recognition	NN
lesk	59	9	software	software	NN
lesk	59	10	.	.	.
lesk	60	1	In	in	IN
lesk	60	2	this	this	DT
lesk	60	3	example	example	NN
lesk	60	4	,	,	,
lesk	60	5	the	the	DT
lesk	60	6	deception	deception	NN
lesk	60	7	was	be	VBD
lesk	60	8	deliberate	deliberate	JJ
lesk	60	9	;	;	:
lesk	60	10	the	the	DT
lesk	60	11	researchers	researcher	NNS
lesk	60	12	searched	search	VBD
lesk	60	13	for	for	IN
lesk	60	14	the	the	DT
lesk	60	15	best	good	JJS
lesk	60	16	place	place	NN
lesk	60	17	to	to	TO
lesk	60	18	change	change	VB
lesk	60	19	the	the	DT
lesk	60	20	image	image	NN
lesk	60	21	.	.	.
lesk	61	1	2	2	LS
lesk	61	2	.	.	.
lesk	62	1	Bias	bias	NN
lesk	62	2	and	and	CC
lesk	62	3	mistakes	mistake	NNS
lesk	62	4	We	-PRON-	PRP
lesk	62	5	have	have	VBP
lesk	62	6	seen	see	VBN
lesk	62	7	a	a	DT
lesk	62	8	major	major	JJ
lesk	62	9	change	change	NN
lesk	62	10	in	in	IN
lesk	62	11	the	the	DT
lesk	62	12	way	way	NN
lesk	62	13	we	-PRON-	PRP
lesk	62	14	do	do	VBP
lesk	62	15	machine	machine	NN
lesk	62	16	learning	learning	NN
lesk	62	17	,	,	,
lesk	62	18	and	and	CC
lesk	62	19	there	there	EX
lesk	62	20	are	be	VBP
lesk	62	21	real	real	JJ
lesk	62	22	dangers	danger	NNS
lesk	62	23	involved	involve	VBN
lesk	62	24	.	.	.
lesk	63	1	The	the	DT
lesk	63	2	current	current	JJ
lesk	63	3	enthusiasm	enthusiasm	NN
lesk	63	4	for	for	IN
lesk	63	5	neural	neural	JJ
lesk	63	6	nets	net	NNS
lesk	63	7	risks	risk	VBZ
lesk	63	8	the	the	DT
lesk	63	9	use	use	NN
lesk	63	10	of	of	IN
lesk	63	11	processes	process	NNS
lesk	63	12	which	which	WDT
lesk	63	13	can	can	MD
lesk	63	14	not	not	RB
lesk	63	15	be	be	VB
lesk	63	16	understood	understand	VBN
lesk	63	17	,	,	,
lesk	63	18	as	as	IN
lesk	63	19	Mahjoubi	Mahjoubi	NNP
lesk	63	20	warned	warn	VBD
lesk	63	21	,	,	,
lesk	63	22	and	and	CC
lesk	63	23	which	which	WDT
lesk	63	24	can	can	MD
lesk	63	25	thus	thus	RB
lesk	63	26	conceal	conceal	VB
lesk	63	27	methods	method	NNS
lesk	63	28	we	-PRON-	PRP
lesk	63	29	would	would	MD
lesk	63	30	not	not	RB
lesk	63	31	approve	approve	VB
lesk	63	32	of	of	IN
lesk	63	33	,	,	,
lesk	63	34	such	such	JJ
lesk	63	35	as	as	IN
lesk	63	36	discrimination	discrimination	NN
lesk	63	37	in	in	IN
lesk	63	38	lending	lending	NN
lesk	63	39	or	or	CC
lesk	63	40	hiring	hiring	NN
lesk	63	41	.	.	.
lesk	64	1	Cathy	Cathy	NNP
lesk	64	2	O'Neil	O'Neil	NNP
lesk	64	3	has	have	VBZ
lesk	64	4	described	describe	VBN
lesk	64	5	this	this	DT
lesk	64	6	in	in	IN
lesk	64	7	her	-PRON-	PRP$
lesk	64	8	book	book	NN
lesk	64	9	Weapons	Weapons	NNPS
lesk	64	10	of	of	IN
lesk	64	11	Math	Math	NNP
lesk	64	12	Destruction	Destruction	NNP
lesk	64	13	(	(	-LRB-
lesk	64	14	2016	2016	CD
lesk	64	15	)	)	-RRB-
lesk	64	16	.	.	.
lesk	65	1	There	there	EX
lesk	65	2	is	be	VBZ
lesk	65	3	much	much	JJ
lesk	65	4	research	research	NN
lesk	65	5	today	today	NN
lesk	65	6	that	that	WDT
lesk	65	7	seeks	seek	VBZ
lesk	65	8	methods	method	NNS
lesk	65	9	to	to	TO
lesk	65	10	explain	explain	VB
lesk	65	11	what	what	WP
lesk	65	12	neural	neural	JJ
lesk	65	13	nets	net	NNS
lesk	65	14	are	be	VBP
lesk	65	15	doing	do	VBG
lesk	65	16	.	.	.
lesk	66	1	See	see	VB
lesk	66	2	Guidiotti	Guidiotti	NNP
lesk	66	3	et	et	NNP
lesk	66	4	al	al	NNP
lesk	66	5	.	.	.
lesk	67	1	(	(	-LRB-
lesk	67	2	2017	2017	CD
lesk	67	3	)	)	-RRB-
lesk	67	4	for	for	IN
lesk	67	5	a	a	DT
lesk	67	6	survey	survey	NN
lesk	67	7	.	.	.
lesk	68	1	There	there	EX
lesk	68	2	is	be	VBZ
lesk	68	3	also	also	RB
lesk	68	4	a	a	DT
lesk	68	5	2018	2018	CD
lesk	68	6	DARPA	DARPA	NNP
lesk	68	7	program	program	NN
lesk	68	8	on	on	IN
lesk	68	9	“	"	``
lesk	68	10	Explainable	explainable	JJ
lesk	68	11	AI	ai	NN
lesk	68	12	.	.	.
lesk	68	13	”	"	''
lesk	68	14	Techniques	technique	NNS
lesk	68	15	used	use	VBN
lesk	68	16	can	can	MD
lesk	68	17	include	include	VB
lesk	68	18	looking	look	VBG
lesk	68	19	at	at	IN
lesk	68	20	the	the	DT
lesk	68	21	results	result	NNS
lesk	68	22	over	over	IN
lesk	68	23	a	a	DT
lesk	68	24	range	range	NN
lesk	68	25	of	of	IN
lesk	68	26	input	input	NN
lesk	68	27	data	datum	NNS
lesk	68	28	and	and	CC
lesk	68	29	seeing	see	VBG
lesk	68	30	if	if	IN
lesk	68	31	the	the	DT
lesk	68	32	neural	neural	JJ
lesk	68	33	net	net	NN
lesk	68	34	can	can	MD
lesk	68	35	be	be	VB
lesk	68	36	modeled	model	VBN
lesk	68	37	by	by	IN
lesk	68	38	a	a	DT
lesk	68	39	decision	decision	NN
lesk	68	40	tree	tree	NN
lesk	68	41	,	,	,
lesk	68	42	or	or	CC
lesk	68	43	modifying	modify	VBG
lesk	68	44	the	the	DT
lesk	68	45	input	input	NN
lesk	68	46	data	datum	NNS
lesk	68	47	to	to	TO
lesk	68	48	see	see	VB
lesk	68	49	which	which	WDT
lesk	68	50	input	input	NN
lesk	68	51	elements	element	NNS
lesk	68	52	have	have	VBP
lesk	68	53	the	the	DT
lesk	68	54	greatest	great	JJS
lesk	68	55	effect	effect	NN
lesk	68	56	on	on	IN
lesk	68	57	the	the	DT
lesk	68	58	results	result	NNS
lesk	68	59	,	,	,
lesk	68	60	and	and	CC
lesk	68	61	then	then	RB
lesk	68	62	showing	show	VBG
lesk	68	63	that	that	DT
lesk	68	64	to	to	IN
lesk	68	65	the	the	DT
lesk	68	66	user	user	NN
lesk	68	67	.	.	.
lesk	69	1	For	for	IN
lesk	69	2	example	example	NN
lesk	69	3	,	,	,
lesk	69	4	Mariusz	Mariusz	NNP
lesk	69	5	Bojarski	Bojarski	NNP
lesk	69	6	et	et	NNP
lesk	69	7	al	al	NNP
lesk	69	8	.	.	.
lesk	70	1	describe	describe	VB
lesk	70	2	a	a	DT
lesk	70	3	self	self	NN
lesk	70	4	-	-	HYPH
lesk	70	5	driving	drive	VBG
lesk	70	6	system	system	NN
lesk	70	7	that	that	WDT
lesk	70	8	highlights	highlight	VBZ
lesk	70	9	what	what	WP
lesk	70	10	it	-PRON-	PRP
lesk	70	11	thinks	think	VBZ
lesk	70	12	is	be	VBZ
lesk	70	13	important	important	JJ
lesk	70	14	in	in	IN
lesk	70	15	what	what	WP
lesk	70	16	it	-PRON-	PRP
lesk	70	17	is	be	VBZ
lesk	70	18	seeing	see	VBG
lesk	70	19	(	(	-LRB-
lesk	70	20	2017	2017	CD
lesk	70	21	)	)	-RRB-
lesk	70	22	.	.	.
lesk	71	1	However	however	RB
lesk	71	2	,	,	,
lesk	71	3	this	this	DT
lesk	71	4	is	be	VBZ
lesk	71	5	generally	generally	RB
lesk	71	6	research	research	NN
lesk	71	7	in	in	IN
lesk	71	8	progress	progress	NN
lesk	71	9	,	,	,
lesk	71	10	and	and	CC
lesk	71	11	it	-PRON-	PRP
lesk	71	12	raises	raise	VBZ
lesk	71	13	the	the	DT
lesk	71	14	question	question	NN
lesk	71	15	of	of	IN
lesk	71	16	whether	whether	IN
lesk	71	17	we	-PRON-	PRP
lesk	71	18	can	can	MD
lesk	71	19	trust	trust	VB
lesk	71	20	the	the	DT
lesk	71	21	explanation	explanation	NN
lesk	71	22	generator	generator	NN
lesk	71	23	.	.	.
lesk	72	1	Many	many	JJ
lesk	72	2	popular	popular	JJ
lesk	72	3	magazines	magazine	NNS
lesk	72	4	have	have	VBP
lesk	72	5	discussed	discuss	VBN
lesk	72	6	this	this	DT
lesk	72	7	problem	problem	NN
lesk	72	8	;	;	:
lesk	72	9	Forbes	Forbes	NNP
lesk	72	10	,	,	,
lesk	72	11	for	for	IN
lesk	72	12	example	example	NN
lesk	72	13	,	,	,
lesk	72	14	had	have	VBD
lesk	72	15	an	an	DT
lesk	72	16	explanation	explanation	NN
lesk	72	17	of	of	IN
lesk	72	18	how	how	WRB
lesk	72	19	the	the	DT
lesk	72	20	choice	choice	NN
lesk	72	21	of	of	IN
lesk	72	22	datasets	dataset	NNS
lesk	72	23	can	can	MD
lesk	72	24	produce	produce	VB
lesk	72	25	a	a	DT
lesk	72	26	biased	biased	JJ
lesk	72	27	result	result	NN
lesk	72	28	without	without	IN
lesk	72	29	any	any	DT
lesk	72	30	deliberate	deliberate	JJ
lesk	72	31	attempt	attempt	NN
lesk	72	32	to	to	TO
lesk	72	33	do	do	VB
lesk	72	34	so	so	RB
lesk	72	35	(	(	-LRB-
lesk	72	36	Taulli	Taulli	NNP
lesk	72	37	2019	2019	CD
lesk	72	38	)	)	-RRB-
lesk	72	39	.	.	.
lesk	73	1	Similarly	similarly	RB
lesk	73	2	,	,	,
lesk	73	3	the	the	DT
lesk	73	4	New	New	NNP
lesk	73	5	York	York	NNP
lesk	73	6	Times	Times	NNP
lesk	73	7	discussed	discuss	VBD
lesk	73	8	the	the	DT
lesk	73	9	way	way	NN
lesk	73	10	groups	group	NNS
lesk	73	11	of	of	IN
lesk	73	12	primarily	primarily	RB
lesk	73	13	young	young	JJ
lesk	73	14	white	white	JJ
lesk	73	15	men	man	NNS
lesk	73	16	will	will	MD
lesk	73	17	build	build	VB
lesk	73	18	systems	system	NNS
lesk	73	19	that	that	WDT
lesk	73	20	focus	focus	VBP
lesk	73	21	on	on	IN
lesk	73	22	their	-PRON-	PRP$
lesk	73	23	data	datum	NNS
lesk	73	24	,	,	,
lesk	73	25	and	and	CC
lesk	73	26	give	give	VB
lesk	73	27	wrong	wrong	JJ
lesk	73	28	or	or	CC
lesk	73	29	discriminatory	discriminatory	JJ
lesk	73	30	answers	answer	NNS
lesk	73	31	in	in	IN
lesk	73	32	more	more	JJR
lesk	73	33	general	general	JJ
lesk	73	34	situations	situation	NNS
lesk	73	35	(	(	-LRB-
lesk	73	36	Tugend	Tugend	NNP
lesk	73	37	2019	2019	CD
lesk	73	38	)	)	-RRB-
lesk	73	39	.	.	.
lesk	74	1	The	the	DT
lesk	74	2	MIT	MIT	NNP
lesk	74	3	Media	Media	NNPS
lesk	74	4	Lab	Lab	NNP
lesk	74	5	hosts	host	VBZ
lesk	74	6	the	the	DT
lesk	74	7	Algorithmic	Algorithmic	NNP
lesk	74	8	Justice	Justice	NNP
lesk	74	9	League	League	NNP
lesk	74	10	,	,	,
lesk	74	11	trying	try	VBG
lesk	74	12	to	to	TO
lesk	74	13	stop	stop	VB
lesk	74	14	organizations	organization	NNS
lesk	74	15	from	from	IN
lesk	74	16	building	build	VBG
lesk	74	17	socially	socially	RB
lesk	74	18	slanted	slant	VBN
lesk	74	19	systems	system	NNS
lesk	74	20	.	.	.
lesk	75	1	Similar	similar	JJ
lesk	75	2	thoughts	thought	NNS
lesk	75	3	come	come	VBP
lesk	75	4	from	from	IN
lesk	75	5	groups	group	NNS
lesk	75	6	like	like	IN
lesk	75	7	the	the	DT
lesk	75	8	Data	Data	NNP
lesk	75	9	and	and	CC
lesk	75	10	Society	Society	NNP
lesk	75	11	Research	Research	NNP
lesk	75	12	Institute	Institute	NNP
lesk	75	13	or	or	CC
lesk	75	14	the	the	DT
lesk	75	15	AI	AI	NNP
lesk	75	16	Now	Now	NNP
lesk	75	17	Institute	Institute	NNP
lesk	75	18	.	.	.
lesk	76	1	Again	again	RB
lesk	76	2	,	,	,
lesk	76	3	the	the	DT
lesk	76	4	problems	problem	NNS
lesk	76	5	may	may	MD
lesk	76	6	be	be	VB
lesk	76	7	accidental	accidental	JJ
lesk	76	8	or	or	CC
lesk	76	9	deliberate	deliberate	JJ
lesk	76	10	.	.	.
lesk	77	1	The	the	DT
lesk	77	2	phrase	phrase	NN
lesk	77	3	“	"	``
lesk	77	4	data	data	NN
lesk	77	5	poisoning	poisoning	NN
lesk	77	6	”	"	''
lesk	77	7	has	have	VBZ
lesk	77	8	been	be	VBN
lesk	77	9	used	use	VBN
lesk	77	10	to	to	TO
lesk	77	11	suggest	suggest	VB
lesk	77	12	malicious	malicious	JJ
lesk	77	13	creation	creation	NN
lesk	77	14	of	of	IN
lesk	77	15	training	training	NN
lesk	77	16	data	datum	NNS
lesk	77	17	or	or	CC
lesk	77	18	examples	example	NNS
lesk	77	19	of	of	IN
lesk	77	20	data	datum	NNS
lesk	77	21	designed	design	VBN
lesk	77	22	to	to	TO
lesk	77	23	deceive	deceive	VB
lesk	77	24	machine	machine	NN
lesk	77	25	learning	learning	NN
lesk	77	26	systems	system	NNS
lesk	77	27	.	.	.
lesk	78	1	There	there	EX
lesk	78	2	is	be	VBZ
lesk	78	3	now	now	RB
lesk	78	4	a	a	DT
lesk	78	5	DARPA	DARPA	NNP
lesk	78	6	research	research	NN
lesk	78	7	program	program	NN
lesk	78	8	,	,	,
lesk	78	9	“	"	``
lesk	78	10	Guaranteeing	guarantee	VBG
lesk	78	11	AI	AI	NNP
lesk	78	12	Robustness	Robustness	NNP
lesk	78	13	against	against	IN
lesk	78	14	Deception	Deception	NNP
lesk	78	15	(	(	-LRB-
lesk	78	16	GARD	GARD	NNP
lesk	78	17	)	)	-RRB-
lesk	78	18	,	,	,
lesk	78	19	”	"	''
lesk	78	20	supporting	support	VBG
lesk	78	21	research	research	NN
lesk	78	22	to	to	TO
lesk	78	23	learn	learn	VB
lesk	78	24	how	how	WRB
lesk	78	25	to	to	TO
lesk	78	26	stop	stop	VB
lesk	78	27	trickery	trickery	NN
lesk	78	28	such	such	JJ
lesk	78	29	as	as	IN
lesk	78	30	a	a	DT
lesk	78	31	demonstration	demonstration	NN
lesk	78	32	of	of	IN
lesk	78	33	converting	convert	VBG
lesk	78	34	a	a	DT
lesk	78	35	traffic	traffic	NN
lesk	78	36	stop	stop	NN
lesk	78	37	sign	sign	NN
lesk	78	38	to	to	IN
lesk	78	39	a	a	DT
lesk	78	40	45	45	CD
lesk	78	41	mph	mph	NN
lesk	78	42	speed	speed	NN
lesk	78	43	limit	limit	NN
lesk	78	44	with	with	IN
lesk	78	45	a	a	DT
lesk	78	46	few	few	JJ
lesk	78	47	stickers	sticker	NNS
lesk	78	48	(	(	-LRB-
lesk	78	49	Eykholt	Eykholt	NNP
lesk	78	50	et	et	FW
lesk	78	51	al	al	NNP
lesk	78	52	.	.	.
lesk	79	1	2018	2018	CD
lesk	79	2	)	)	-RRB-
lesk	79	3	.	.	.
lesk	80	1	More	more	RBR
lesk	80	2	generally	generally	RB
lesk	80	3	,	,	,
lesk	80	4	bias	bias	NN
lesk	80	5	in	in	IN
lesk	80	6	systems	system	NNS
lesk	80	7	deciding	decide	VBG
lesk	80	8	whether	whether	IN
lesk	80	9	to	to	TO
lesk	80	10	grant	grant	VB
lesk	80	11	loans	loan	NNS
lesk	80	12	may	may	MD
lesk	80	13	be	be	VB
lesk	80	14	discriminatory	discriminatory	JJ
lesk	80	15	but	but	CC
lesk	80	16	nevertheless	nevertheless	RB
lesk	80	17	profitable	profitable	JJ
lesk	80	18	.	.	.
lesk	81	1	Even	even	RB
lesk	81	2	if	if	IN
lesk	81	3	you	-PRON-	PRP
lesk	81	4	want	want	VBP
lesk	81	5	to	to	TO
lesk	81	6	detect	detect	VB
lesk	81	7	AI	AI	NNP
lesk	81	8	mistakes	mistake	NNS
lesk	81	9	,	,	,
lesk	81	10	recognizing	recognize	VBG
lesk	81	11	such	such	JJ
lesk	81	12	problems	problem	NNS
lesk	81	13	is	be	VBZ
lesk	81	14	difficult	difficult	JJ
lesk	81	15	.	.	.
lesk	82	1	Often	often	RB
lesk	82	2	things	thing	NNS
lesk	82	3	will	will	MD
lesk	82	4	be	be	VB
lesk	82	5	wrong	wrong	JJ
lesk	82	6	and	and	CC
lesk	82	7	we	-PRON-	PRP
lesk	82	8	wo	will	MD
lesk	82	9	n’t	not	RB
lesk	82	10	know	know	VB
lesk	82	11	why	why	WRB
lesk	82	12	.	.	.
lesk	83	1	And	and	CC
lesk	83	2	even	even	RB
lesk	83	3	hypothetical	hypothetical	JJ
lesk	83	4	(	(	-LRB-
lesk	83	5	but	but	CC
lesk	83	6	perhaps	perhaps	RB
lesk	83	7	erroneous	erroneous	JJ
lesk	83	8	)	)	-RRB-
lesk	83	9	explanations	explanation	NNS
lesk	83	10	can	can	MD
lesk	83	11	be	be	VB
lesk	83	12	very	very	RB
lesk	83	13	convincing	convincing	JJ
lesk	83	14	;	;	:
lesk	83	15	people	people	NNS
lesk	83	16	easily	easily	RB
lesk	83	17	believe	believe	VBP
lesk	83	18	plausible	plausible	JJ
lesk	83	19	stories	story	NNS
lesk	83	20	.	.	.
lesk	84	1	I	-PRON-	PRP
lesk	84	2	routinely	routinely	RB
lesk	84	3	give	give	VBP
lesk	84	4	my	-PRON-	PRP$
lesk	84	5	students	student	NNS
lesk	84	6	a	a	DT
lesk	84	7	paper	paper	NN
lesk	84	8	that	that	WDT
lesk	84	9	concludes	conclude	VBZ
lesk	84	10	that	that	IN
lesk	84	11	prior	prior	JJ
lesk	84	12	ownership	ownership	NN
lesk	84	13	of	of	IN
lesk	84	14	a	a	DT
lesk	84	15	cat	cat	NN
lesk	84	16	prevents	prevent	VBZ
lesk	84	17	fatal	fatal	JJ
lesk	84	18	myocardial	myocardial	JJ
lesk	84	19	infarctions	infarction	NNS
lesk	84	20	;	;	:
lesk	84	21	its	-PRON-	PRP$
lesk	84	22	result	result	NN
lesk	84	23	implies	imply	VBZ
lesk	84	24	that	that	IN
lesk	84	25	cats	cat	NNS
lesk	84	26	are	be	VBP
lesk	84	27	more	more	RBR
lesk	84	28	protective	protective	JJ
lesk	84	29	than	than	IN
lesk	84	30	statin	statin	NN
lesk	84	31	drugs	drug	NNS
lesk	84	32	(	(	-LRB-
lesk	84	33	Qureshi	Qureshi	NNP
lesk	84	34	et	et	NNP
lesk	84	35	al	al	NNP
lesk	84	36	.	.	.
lesk	85	1	2009	2009	CD
lesk	85	2	)	)	-RRB-
lesk	85	3	.	.	.
lesk	86	1	The	the	DT
lesk	86	2	students	student	NNS
lesk	86	3	are	be	VBP
lesk	86	4	very	very	RB
lesk	86	5	quick	quick	JJ
lesk	86	6	to	to	TO
lesk	86	7	come	come	VB
lesk	86	8	up	up	RP
lesk	86	9	with	with	IN
lesk	86	10	possibilities	possibility	NNS
lesk	86	11	like	like	IN
lesk	86	12	“	"	``
lesk	86	13	petting	pet	VBG
lesk	86	14	a	a	DT
lesk	86	15	cat	cat	NN
lesk	86	16	is	be	VBZ
lesk	86	17	relaxing	relaxing	JJ
lesk	86	18	,	,	,
lesk	86	19	relaxation	relaxation	NN
lesk	86	20	reduces	reduce	VBZ
lesk	86	21	your	-PRON-	PRP$
lesk	86	22	blood	blood	NN
lesk	86	23	pressure	pressure	NN
lesk	86	24	,	,	,
lesk	86	25	and	and	CC
lesk	86	26	lower	low	JJR
lesk	86	27	blood	blood	NN
lesk	86	28	pressure	pressure	NN
lesk	86	29	decreases	decrease	VBZ
lesk	86	30	the	the	DT
lesk	86	31	risk	risk	NN
lesk	86	32	of	of	IN
lesk	86	33	heart	heart	NN
lesk	86	34	attacks	attack	NNS
lesk	86	35	.	.	.
lesk	86	36	”	"	''
lesk	86	37	Then	then	RB
lesk	86	38	I	-PRON-	PRP
lesk	86	39	have	have	VBP
lesk	86	40	to	to	TO
lesk	86	41	explain	explain	VB
lesk	86	42	that	that	IN
lesk	86	43	the	the	DT
lesk	86	44	paper	paper	NN
lesk	86	45	evaluates	evaluate	VBZ
lesk	86	46	32	32	CD
lesk	86	47	possibilities	possibility	NNS
lesk	86	48	(	(	-LRB-
lesk	86	49	prior	prior	RB
lesk	86	50	/	/	SYM
lesk	86	51	current	current	JJ
lesk	86	52	ownership	ownership	NN
lesk	86	53	X	x	NN
lesk	86	54	cats	cat	NNS
lesk	86	55	/	/	SYM
lesk	86	56	dogs	dog	NNS
lesk	86	57	X	X	NNP
lesk	86	58	4	4	CD
lesk	86	59	medical	medical	JJ
lesk	86	60	conditions	condition	NNS
lesk	86	61	X	x	VBP
lesk	86	62	fatal	fatal	JJ
lesk	86	63	/	/	SYM
lesk	86	64	nonfatal	nonfatal	JJ
lesk	86	65	)	)	-RRB-
lesk	86	66	and	and	CC
lesk	86	67	you	-PRON-	PRP
lesk	86	68	should	should	MD
lesk	86	69	n’t	not	RB
lesk	86	70	be	be	VB
lesk	86	71	surprised	surprised	JJ
lesk	86	72	if	if	IN
lesk	86	73	you	-PRON-	PRP
lesk	86	74	evaluate	evaluate	VBP
lesk	86	75	32	32	CD
lesk	86	76	chances	chance	NNS
lesk	86	77	and	and	CC
lesk	86	78	one	one	CD
lesk	86	79	is	be	VBZ
lesk	86	80	significant	significant	JJ
lesk	86	81	at	at	IN
lesk	86	82	the	the	DT
lesk	86	83	0.05	0.05	CD
lesk	86	84	level	level	NN
lesk	86	85	,	,	,
lesk	86	86	which	which	WDT
lesk	86	87	is	be	VBZ
lesk	86	88	only	only	RB
lesk	86	89	1	1	CD
lesk	86	90	in	in	IN
lesk	86	91	20	20	CD
lesk	86	92	.	.	.
lesk	87	1	In	in	IN
lesk	87	2	this	this	DT
lesk	87	3	example	example	NN
lesk	87	4	,	,	,
lesk	87	5	there	there	EX
lesk	87	6	is	be	VBZ
lesk	87	7	also	also	RB
lesk	87	8	the	the	DT
lesk	87	9	question	question	NN
lesk	87	10	of	of	IN
lesk	87	11	reverse	reverse	JJ
lesk	87	12	causality	causality	NN
lesk	87	13	:	:	:
lesk	87	14	perhaps	perhaps	RB
lesk	87	15	someone	someone	NN
lesk	87	16	who	who	WP
lesk	87	17	is	be	VBZ
lesk	87	18	in	in	IN
lesk	87	19	ill	ill	JJ
lesk	87	20	health	health	NN
lesk	87	21	will	will	MD
lesk	87	22	decide	decide	VB
lesk	87	23	he	-PRON-	PRP
lesk	87	24	is	be	VBZ
lesk	87	25	too	too	RB
lesk	87	26	sick	sick	JJ
lesk	87	27	to	to	TO
lesk	87	28	take	take	VB
lesk	87	29	care	care	NN
lesk	87	30	of	of	IN
lesk	87	31	a	a	DT
lesk	87	32	pet	pet	NN
lesk	87	33	,	,	,
lesk	87	34	so	so	IN
lesk	87	35	that	that	IN
lesk	87	36	the	the	DT
lesk	87	37	poor	poor	JJ
lesk	87	38	health	health	NN
lesk	87	39	is	be	VBZ
lesk	87	40	not	not	RB
lesk	87	41	caused	cause	VBN
lesk	87	42	by	by	IN
lesk	87	43	the	the	DT
lesk	87	44	lack	lack	NN
lesk	87	45	of	of	IN
lesk	87	46	a	a	DT
lesk	87	47	cat	cat	NN
lesk	87	48	,	,	,
lesk	87	49	but	but	CC
lesk	87	50	rather	rather	RB
lesk	87	51	the	the	DT
lesk	87	52	poor	poor	JJ
lesk	87	53	health	health	NN
lesk	87	54	causes	cause	VBZ
lesk	87	55	the	the	DT
lesk	87	56	absence	absence	NN
lesk	87	57	of	of	IN
lesk	87	58	a	a	DT
lesk	87	59	cat	cat	NN
lesk	87	60	.	.	.
lesk	88	1	Sometimes	sometimes	RB
lesk	88	2	explanations	explanation	NNS
lesk	88	3	can	can	MD
lesk	88	4	help	help	VB
lesk	88	5	,	,	,
lesk	88	6	as	as	IN
lesk	88	7	in	in	IN
lesk	88	8	a	a	DT
lesk	88	9	machine	machine	NN
lesk	88	10	learning	learn	VBG
lesk	88	11	program	program	NN
lesk	88	12	that	that	WDT
lesk	88	13	was	be	VBD
lesk	88	14	deliberately	deliberately	RB
lesk	88	15	trained	train	VBN
lesk	88	16	to	to	TO
lesk	88	17	distinguish	distinguish	VB
lesk	88	18	images	image	NNS
lesk	88	19	of	of	IN
lesk	88	20	wolves	wolf	NNS
lesk	88	21	and	and	CC
lesk	88	22	dogs	dog	NNS
lesk	88	23	but	but	CC
lesk	88	24	was	be	VBD
lesk	88	25	trained	train	VBN
lesk	88	26	using	use	VBG
lesk	88	27	pictures	picture	NNS
lesk	88	28	of	of	IN
lesk	88	29	wolves	wolf	NNS
lesk	88	30	that	that	WDT
lesk	88	31	always	always	RB
lesk	88	32	contained	contain	VBD
lesk	88	33	snow	snow	NN
lesk	88	34	and	and	CC
lesk	88	35	pictures	picture	NNS
lesk	88	36	of	of	IN
lesk	88	37	dogs	dog	NNS
lesk	88	38	that	that	WDT
lesk	88	39	never	never	RB
lesk	88	40	did	do	VBD
lesk	88	41	(	(	-LRB-
lesk	88	42	Ribeiro	Ribeiro	NNP
lesk	88	43	,	,	,
lesk	88	44	Singh	Singh	NNP
lesk	88	45	,	,	,
lesk	88	46	and	and	CC
lesk	88	47	Guestrin	Guestrin	NNP
lesk	88	48	2016	2016	CD
lesk	88	49	)	)	-RRB-
lesk	88	50	.	.	.
lesk	89	1	Without	without	IN
lesk	89	2	explaining	explain	VBG
lesk	89	3	that	that	IN
lesk	89	4	,	,	,
lesk	89	5	10	10	CD
lesk	89	6	of	of	IN
lesk	89	7	27	27	CD
lesk	89	8	subjects	subject	NNS
lesk	89	9	thought	think	VBD
lesk	89	10	the	the	DT
lesk	89	11	classifier	classifier	NN
lesk	89	12	was	be	VBD
lesk	89	13	trustworthy	trustworthy	JJ
lesk	89	14	;	;	:
lesk	89	15	after	after	IN
lesk	89	16	pointing	point	VBG
lesk	89	17	out	out	RP
lesk	89	18	the	the	DT
lesk	89	19	snow	snow	NN
lesk	89	20	only	only	RB
lesk	89	21	3	3	CD
lesk	89	22	of	of	IN
lesk	89	23	27	27	CD
lesk	89	24	subjects	subject	NNS
lesk	89	25	believed	believe	VBD
lesk	89	26	the	the	DT
lesk	89	27	system	system	NN
lesk	89	28	.	.	.
lesk	90	1	Usually	usually	RB
lesk	90	2	you	-PRON-	PRP
lesk	90	3	do	do	VBP
lesk	90	4	n’t	not	RB
lesk	90	5	get	get	VB
lesk	90	6	such	such	PDT
lesk	90	7	a	a	DT
lesk	90	8	clear	clear	JJ
lesk	90	9	presentation	presentation	NN
lesk	90	10	of	of	IN
lesk	90	11	a	a	DT
lesk	90	12	mis	mis	NN
lesk	90	13	-	-	HYPH
lesk	90	14	trained	train	VBN
lesk	90	15	system	system	NN
lesk	90	16	.	.	.
lesk	91	1	3	3	LS
lesk	91	2	.	.	.
lesk	92	1	Recognition	recognition	NN
lesk	92	2	of	of	IN
lesk	92	3	problems	problem	NNS
lesk	92	4	Can	Can	MD
lesk	92	5	we	-PRON-	PRP
lesk	92	6	tell	tell	VB
lesk	92	7	when	when	WRB
lesk	92	8	something	something	NN
lesk	92	9	is	be	VBZ
lesk	92	10	wrong	wrong	JJ
lesk	92	11	?	?	.
lesk	93	1	Here	here	RB
lesk	93	2	’s	’	VBZ
lesk	93	3	the	the	DT
lesk	93	4	result	result	NN
lesk	93	5	of	of	IN
lesk	93	6	a	a	DT
lesk	93	7	Google	Google	NNP
lesk	93	8	Photo	Photo	NNP
lesk	93	9	merge	merge	NN
lesk	93	10	of	of	IN
lesk	93	11	three	three	CD
lesk	93	12	other	other	JJ
lesk	93	13	photos	photo	NNS
lesk	93	14	;	;	:
lesk	93	15	two	two	CD
lesk	93	16	landscapes	landscape	NNS
lesk	93	17	and	and	CC
lesk	93	18	a	a	DT
lesk	93	19	picture	picture	NN
lesk	93	20	of	of	IN
lesk	93	21	somebody	somebody	NN
lesk	93	22	’s	’s	POS
lesk	93	23	friend	friend	NN
lesk	93	24	.	.	.
lesk	94	1	The	the	DT
lesk	94	2	software	software	NN
lesk	94	3	was	be	VBD
lesk	94	4	told	tell	VBN
lesk	94	5	to	to	TO
lesk	94	6	make	make	VB
lesk	94	7	a	a	DT
lesk	94	8	panorama	panorama	NN
lesk	94	9	and	and	CC
lesk	94	10	stitched	stitch	VBD
lesk	94	11	the	the	DT
lesk	94	12	images	image	NNS
lesk	94	13	together	together	RB
lesk	94	14	(	(	-LRB-
lesk	94	15	Peng	Peng	NNP
lesk	94	16	2018	2018	CD
lesk	94	17	)	)	-RRB-
lesk	94	18	.	.	.
lesk	95	1	It	-PRON-	PRP
lesk	95	2	looks	look	VBZ
lesk	95	3	like	like	IN
lesk	95	4	a	a	DT
lesk	95	5	joke	joke	NN
lesk	95	6	,	,	,
lesk	95	7	and	and	CC
lesk	95	8	even	even	RB
lesk	95	9	made	make	VBD
lesk	95	10	it	-PRON-	PRP
lesk	95	11	into	into	IN
lesk	95	12	a	a	DT
lesk	95	13	list	list	NN
lesk	95	14	of	of	IN
lesk	95	15	top	top	JJ
lesk	95	16	jokes	joke	NNS
lesk	95	17	on	on	IN
lesk	95	18	reddit	reddit	NN
lesk	95	19	.	.	.
lesk	96	1	The	the	DT
lesk	96	2	author	author	NN
lesk	96	3	’s	’s	POS
lesk	96	4	point	point	NN
lesk	96	5	was	be	VBD
lesk	96	6	that	that	IN
lesk	96	7	the	the	DT
lesk	96	8	panorama	panorama	NNP
lesk	96	9	system	system	NN
lesk	96	10	did	do	VBD
lesk	96	11	n’t	not	RB
lesk	96	12	understand	understand	VB
lesk	96	13	basic	basic	JJ
lesk	96	14	composition	composition	NN
lesk	96	15	:	:	:
lesk	96	16	people	people	NNS
lesk	96	17	are	be	VBP
lesk	96	18	not	not	RB
lesk	96	19	the	the	DT
lesk	96	20	same	same	JJ
lesk	96	21	scale	scale	NN
lesk	96	22	as	as	IN
lesk	96	23	mountains	mountain	NNS
lesk	96	24	.	.	.
lesk	97	1	Often	often	RB
lesk	97	2	,	,	,
lesk	97	3	machine	machine	NN
lesk	97	4	learning	learning	NN
lesk	97	5	results	result	NNS
lesk	97	6	are	be	VBP
lesk	97	7	overstated	overstate	VBN
lesk	97	8	.	.	.
lesk	98	1	Google	Google	NNP
lesk	98	2	Flu	Flu	NNP
lesk	98	3	Trends	Trends	NNPS
lesk	98	4	was	be	VBD
lesk	98	5	acclaimed	acclaim	VBN
lesk	98	6	for	for	IN
lesk	98	7	several	several	JJ
lesk	98	8	years	year	NNS
lesk	98	9	and	and	CC
lesk	98	10	then	then	RB
lesk	98	11	turned	turn	VBD
lesk	98	12	out	out	RP
lesk	98	13	to	to	TO
lesk	98	14	be	be	VB
lesk	98	15	undependable	undependable	JJ
lesk	98	16	(	(	-LRB-
lesk	98	17	Lazer	Lazer	NNP
lesk	98	18	et	et	NNP
lesk	98	19	al	al	NNP
lesk	98	20	.	.	.
lesk	99	1	2014	2014	CD
lesk	99	2	)	)	-RRB-
lesk	99	3	.	.	.
lesk	100	1	A	a	DT
lesk	100	2	study	study	NN
lesk	100	3	that	that	WDT
lesk	100	4	attempted	attempt	VBD
lesk	100	5	to	to	TO
lesk	100	6	compare	compare	VB
lesk	100	7	the	the	DT
lesk	100	8	performance	performance	NN
lesk	100	9	of	of	IN
lesk	100	10	machine	machine	NN
lesk	100	11	learning	learning	NN
lesk	100	12	systems	system	NNS
lesk	100	13	for	for	IN
lesk	100	14	medical	medical	JJ
lesk	100	15	diagnosis	diagnosis	NN
lesk	100	16	with	with	IN
lesk	100	17	actual	actual	JJ
lesk	100	18	doctors	doctor	NNS
lesk	100	19	found	find	VBD
lesk	100	20	that	that	DT
lesk	100	21	of	of	IN
lesk	100	22	over	over	IN
lesk	100	23	20,000	20,000	CD
lesk	100	24	papers	paper	NNS
lesk	100	25	analyzed	analyze	VBN
lesk	100	26	,	,	,
lesk	100	27	only	only	RB
lesk	100	28	a	a	DT
lesk	100	29	few	few	JJ
lesk	100	30	dozen	dozen	NN
lesk	100	31	had	have	VBD
lesk	100	32	data	datum	NNS
lesk	100	33	suitable	suitable	JJ
lesk	100	34	for	for	IN
lesk	100	35	an	an	DT
lesk	100	36	evaluation	evaluation	NN
lesk	100	37	(	(	-LRB-
lesk	100	38	Liu	Liu	NNP
lesk	100	39	et	et	NNP
lesk	100	40	al	al	NNP
lesk	100	41	.	.	.
lesk	101	1	2019	2019	CD
lesk	101	2	)	)	-RRB-
lesk	101	3	.	.	.
lesk	102	1	The	the	DT
lesk	102	2	results	result	NNS
lesk	102	3	claimed	claim	VBD
lesk	102	4	comparable	comparable	JJ
lesk	102	5	accuracy	accuracy	NN
lesk	102	6	,	,	,
lesk	102	7	but	but	CC
lesk	102	8	virtually	virtually	RB
lesk	102	9	none	none	NN
lesk	102	10	of	of	IN
lesk	102	11	the	the	DT
lesk	102	12	papers	paper	NNS
lesk	102	13	presented	present	VBD
lesk	102	14	adequate	adequate	JJ
lesk	102	15	data	datum	NNS
lesk	102	16	to	to	TO
lesk	102	17	support	support	VB
lesk	102	18	that	that	DT
lesk	102	19	conclusion	conclusion	NN
lesk	102	20	.	.	.
lesk	103	1	Unusually	unusually	RB
lesk	103	2	promising	promising	JJ
lesk	103	3	results	result	NNS
lesk	103	4	are	be	VBP
lesk	103	5	sometimes	sometimes	RB
lesk	103	6	the	the	DT
lesk	103	7	result	result	NN
lesk	103	8	of	of	IN
lesk	103	9	overfitting	overfitting	NN
lesk	103	10	(	(	-LRB-
lesk	103	11	Brownlee	Brownlee	NNP
lesk	103	12	2018	2018	CD
lesk	103	13	)	)	-RRB-
lesk	103	14	;	;	:
lesk	103	15	this	this	DT
lesk	103	16	is	be	VBZ
lesk	103	17	what	what	WP
lesk	103	18	was	be	VBD
lesk	103	19	wrong	wrong	JJ
lesk	103	20	with	with	IN
lesk	103	21	Google	Google	NNP
lesk	103	22	Flu	Flu	NNP
lesk	103	23	Trends	Trends	NNPS
lesk	103	24	.	.	.
lesk	104	1	A	a	DT
lesk	104	2	machine	machine	NN
lesk	104	3	learning	learning	NN
lesk	104	4	program	program	NN
lesk	104	5	can	can	MD
lesk	104	6	learn	learn	VB
lesk	104	7	a	a	DT
lesk	104	8	large	large	JJ
lesk	104	9	number	number	NN
lesk	104	10	of	of	IN
lesk	104	11	special	special	JJ
lesk	104	12	cases	case	NNS
lesk	104	13	and	and	CC
lesk	104	14	then	then	RB
lesk	104	15	find	find	VB
lesk	104	16	that	that	IN
lesk	104	17	the	the	DT
lesk	104	18	results	result	NNS
lesk	104	19	do	do	VBP
lesk	104	20	not	not	RB
lesk	104	21	generalize	generalize	VB
lesk	104	22	.	.	.
lesk	105	1	In	in	IN
lesk	105	2	other	other	JJ
lesk	105	3	cases	case	NNS
lesk	105	4	problems	problem	NNS
lesk	105	5	can	can	MD
lesk	105	6	result	result	VB
lesk	105	7	when	when	WRB
lesk	105	8	using	use	VBG
lesk	105	9	“	"	``
lesk	105	10	clean	clean	JJ
lesk	105	11	”	"	''
lesk	105	12	data	datum	NNS
lesk	105	13	for	for	IN
lesk	105	14	training	training	NN
lesk	105	15	,	,	,
lesk	105	16	and	and	CC
lesk	105	17	then	then	RB
lesk	105	18	encountering	encounter	VBG
lesk	105	19	messier	messy	JJR
lesk	105	20	data	datum	NNS
lesk	105	21	in	in	IN
lesk	105	22	applications	application	NNS
lesk	105	23	.	.	.
lesk	106	1	Ideally	ideally	RB
lesk	106	2	,	,	,
lesk	106	3	training	training	NN
lesk	106	4	and	and	CC
lesk	106	5	testing	testing	NN
lesk	106	6	data	datum	NNS
lesk	106	7	should	should	MD
lesk	106	8	be	be	VB
lesk	106	9	from	from	IN
lesk	106	10	the	the	DT
lesk	106	11	same	same	JJ
lesk	106	12	dataset	dataset	NN
lesk	106	13	and	and	CC
lesk	106	14	divided	divide	VBD
lesk	106	15	at	at	IN
lesk	106	16	random	random	JJ
lesk	106	17	,	,	,
lesk	106	18	but	but	CC
lesk	106	19	it	-PRON-	PRP
lesk	106	20	can	can	MD
lesk	106	21	be	be	VB
lesk	106	22	tempting	tempting	JJ
lesk	106	23	to	to	TO
lesk	106	24	start	start	VB
lesk	106	25	off	off	RP
lesk	106	26	with	with	IN
lesk	106	27	examples	example	NNS
lesk	106	28	that	that	WDT
lesk	106	29	are	be	VBP
lesk	106	30	the	the	DT
lesk	106	31	result	result	NN
lesk	106	32	of	of	IN
lesk	106	33	initial	initial	JJ
lesk	106	34	and	and	CC
lesk	106	35	higher	high	JJR
lesk	106	36	quality	quality	NN
lesk	106	37	data	datum	NNS
lesk	106	38	collection	collection	NN
lesk	106	39	.	.	.
lesk	107	1	Sometimes	sometimes	RB
lesk	107	2	in	in	IN
lesk	107	3	the	the	DT
lesk	107	4	past	past	NN
lesk	107	5	we	-PRON-	PRP
lesk	107	6	had	have	VBD
lesk	107	7	a	a	DT
lesk	107	8	choice	choice	NN
lesk	107	9	between	between	IN
lesk	107	10	modeling	modeling	NN
lesk	107	11	and	and	CC
lesk	107	12	data	datum	NNS
lesk	107	13	for	for	IN
lesk	107	14	predictions	prediction	NNS
lesk	107	15	.	.	.
lesk	108	1	Consider	consider	VB
lesk	108	2	,	,	,
lesk	108	3	for	for	IN
lesk	108	4	example	example	NN
lesk	108	5	,	,	,
lesk	108	6	the	the	DT
lesk	108	7	problem	problem	NN
lesk	108	8	of	of	IN
lesk	108	9	guessing	guess	VBG
lesk	108	10	what	what	WP
lesk	108	11	the	the	DT
lesk	108	12	weather	weather	NN
lesk	108	13	will	will	MD
lesk	108	14	be	be	VB
lesk	108	15	tomorrow	tomorrow	NN
lesk	108	16	.	.	.
lesk	109	1	We	-PRON-	PRP
lesk	109	2	now	now	RB
lesk	109	3	do	do	VBP
lesk	109	4	this	this	DT
lesk	109	5	based	base	VBN
lesk	109	6	on	on	IN
lesk	109	7	a	a	DT
lesk	109	8	model	model	NN
lesk	109	9	of	of	IN
lesk	109	10	the	the	DT
lesk	109	11	atmosphere	atmosphere	NN
lesk	109	12	that	that	WDT
lesk	109	13	uses	use	VBZ
lesk	109	14	the	the	DT
lesk	109	15	Navier	Navier	NNP
lesk	109	16	-	-	HYPH
lesk	109	17	Stokes	Stokes	NNP
lesk	109	18	equations	equation	NNS
lesk	109	19	;	;	:
lesk	109	20	we	-PRON-	PRP
lesk	109	21	use	use	VBP
lesk	109	22	supercomputers	supercomputer	NNS
lesk	109	23	and	and	CC
lesk	109	24	derive	derive	VB
lesk	109	25	tomorrow	tomorrow	NN
lesk	109	26	’s	’s	POS
lesk	109	27	atmosphere	atmosphere	NN
lesk	109	28	from	from	IN
lesk	109	29	today	today	NN
lesk	109	30	’s	’s	,
lesk	109	31	(	(	-LRB-
lesk	109	32	Christensen	Christensen	NNP
lesk	109	33	2015	2015	CD
lesk	109	34	)	)	-RRB-
lesk	109	35	.	.	.
lesk	110	1	What	what	WP
lesk	110	2	did	do	VBD
lesk	110	3	we	-PRON-	PRP
lesk	110	4	do	do	VB
lesk	110	5	before	before	IN
lesk	110	6	we	-PRON-	PRP
lesk	110	7	had	have	VBD
lesk	110	8	supercomputers	supercomputer	NNS
lesk	110	9	?	?	.
lesk	111	1	Solving	solve	VBG
lesk	111	2	those	those	DT
lesk	111	3	equations	equation	NNS
lesk	111	4	by	by	IN
lesk	111	5	hand	hand	NN
lesk	111	6	is	be	VBZ
lesk	111	7	impractical	impractical	JJ
lesk	111	8	.	.	.
lesk	112	1	One	one	CD
lesk	112	2	of	of	IN
lesk	112	3	the	the	DT
lesk	112	4	methods	method	NNS
lesk	112	5	was	be	VBD
lesk	112	6	“	"	``
lesk	112	7	prediction	prediction	NN
lesk	112	8	by	by	IN
lesk	112	9	analogy	analogy	NN
lesk	112	10	”	"	''
lesk	112	11	:	:	:
lesk	112	12	find	find	VB
lesk	112	13	some	some	DT
lesk	112	14	day	day	NN
lesk	112	15	in	in	IN
lesk	112	16	the	the	DT
lesk	112	17	past	past	NN
lesk	112	18	whose	whose	WP$
lesk	112	19	weather	weather	NN
lesk	112	20	was	be	VBD
lesk	112	21	most	most	RBS
lesk	112	22	similar	similar	JJ
lesk	112	23	to	to	IN
lesk	112	24	today	today	NN
lesk	112	25	.	.	.
lesk	113	1	Suppose	suppose	VB
lesk	113	2	that	that	DT
lesk	113	3	day	day	NN
lesk	113	4	is	be	VBZ
lesk	113	5	Oct.	October	NNP
lesk	113	6	20	20	CD
lesk	113	7	,	,	,
lesk	113	8	1970	1970	CD
lesk	113	9	.	.	.
lesk	114	1	Then	then	RB
lesk	114	2	use	use	VB
lesk	114	3	October	October	NNP
lesk	114	4	21	21	CD
lesk	114	5	,	,	,
lesk	114	6	1970	1970	CD
lesk	114	7	as	as	IN
lesk	114	8	tomorrow	tomorrow	NN
lesk	114	9	’s	’s	POS
lesk	114	10	prediction	prediction	NN
lesk	114	11	.	.	.
lesk	115	1	Prediction	prediction	NN
lesk	115	2	by	by	IN
lesk	115	3	analogy	analogy	NN
lesk	115	4	does	do	VBZ
lesk	115	5	n’t	not	RB
lesk	115	6	require	require	VB
lesk	115	7	you	-PRON-	PRP
lesk	115	8	to	to	TO
lesk	115	9	have	have	VB
lesk	115	10	a	a	DT
lesk	115	11	model	model	NN
lesk	115	12	or	or	CC
lesk	115	13	use	use	VB
lesk	115	14	advanced	advanced	JJ
lesk	115	15	mathematics	mathematic	NNS
lesk	115	16	.	.	.
lesk	116	1	In	in	IN
lesk	116	2	this	this	DT
lesk	116	3	case	case	NN
lesk	116	4	,	,	,
lesk	116	5	however	however	RB
lesk	116	6	,	,	,
lesk	116	7	it	-PRON-	PRP
lesk	116	8	does	do	VBZ
lesk	116	9	n’t	not	RB
lesk	116	10	work	work	VB
lesk	116	11	as	as	RB
lesk	116	12	well	well	RB
lesk	116	13	-	-	HYPH
lesk	116	14	partly	partly	RB
lesk	116	15	because	because	IN
lesk	116	16	we	-PRON-	PRP
lesk	116	17	do	do	VBP
lesk	116	18	n’t	not	RB
lesk	116	19	have	have	VB
lesk	116	20	enough	enough	JJ
lesk	116	21	past	past	JJ
lesk	116	22	days	day	NNS
lesk	116	23	to	to	TO
lesk	116	24	choose	choose	VB
lesk	116	25	from	from	IN
lesk	116	26	,	,	,
lesk	116	27	and	and	CC
lesk	116	28	we	-PRON-	PRP
lesk	116	29	only	only	RB
lesk	116	30	get	get	VBP
lesk	116	31	new	new	JJ
lesk	116	32	days	day	NNS
lesk	116	33	at	at	IN
lesk	116	34	the	the	DT
lesk	116	35	rate	rate	NN
lesk	116	36	of	of	IN
lesk	116	37	one	one	CD
lesk	116	38	per	per	IN
lesk	116	39	day	day	NN
lesk	116	40	.	.	.
lesk	117	1	In	in	IN
lesk	117	2	fact	fact	NN
lesk	117	3	,	,	,
lesk	117	4	Huug	Huug	NNP
lesk	117	5	van	van	NNP
lesk	117	6	den	den	NNP
lesk	117	7	Dool	Dool	NNP
lesk	117	8	estimated	estimate	VBD
lesk	117	9	the	the	DT
lesk	117	10	number	number	NN
lesk	117	11	of	of	IN
lesk	117	12	days	day	NNS
lesk	117	13	of	of	IN
lesk	117	14	data	datum	NNS
lesk	117	15	needed	need	VBN
lesk	117	16	to	to	TO
lesk	117	17	make	make	VB
lesk	117	18	accurate	accurate	JJ
lesk	117	19	predictions	prediction	NNS
lesk	117	20	as	as	IN
lesk	117	21	1030	1030	CD
lesk	117	22	years	year	NNS
lesk	117	23	,	,	,
lesk	117	24	which	which	WDT
lesk	117	25	is	be	VBZ
lesk	117	26	far	far	RB
lesk	117	27	more	more	JJR
lesk	117	28	than	than	IN
lesk	117	29	the	the	DT
lesk	117	30	age	age	NN
lesk	117	31	of	of	IN
lesk	117	32	the	the	DT
lesk	117	33	universe	universe	NN
lesk	117	34	(	(	-LRB-
lesk	117	35	Wilks	Wilks	NNP
lesk	117	36	2008	2008	CD
lesk	117	37	)	)	-RRB-
lesk	117	38	.	.	.
lesk	118	1	The	the	DT
lesk	118	2	underlying	underlie	VBG
lesk	118	3	problem	problem	NN
lesk	118	4	is	be	VBZ
lesk	118	5	that	that	IN
lesk	118	6	the	the	DT
lesk	118	7	weather	weather	NN
lesk	118	8	is	be	VBZ
lesk	118	9	very	very	RB
lesk	118	10	random	random	JJ
lesk	118	11	.	.	.
lesk	119	1	If	if	IN
lesk	119	2	your	-PRON-	PRP$
lesk	119	3	state	state	NN
lesk	119	4	lottery	lottery	NN
lesk	119	5	is	be	VBZ
lesk	119	6	properly	properly	RB
lesk	119	7	run	run	VBN
lesk	119	8	,	,	,
lesk	119	9	it	-PRON-	PRP
lesk	119	10	should	should	MD
lesk	119	11	be	be	VB
lesk	119	12	completely	completely	RB
lesk	119	13	pointless	pointless	JJ
lesk	119	14	to	to	TO
lesk	119	15	look	look	VB
lesk	119	16	at	at	IN
lesk	119	17	past	past	JJ
lesk	119	18	winning	win	VBG
lesk	119	19	numbers	number	NNS
lesk	119	20	and	and	CC
lesk	119	21	try	try	VBP
lesk	119	22	to	to	TO
lesk	119	23	guess	guess	VB
lesk	119	24	the	the	DT
lesk	119	25	next	next	JJ
lesk	119	26	one	one	CD
lesk	119	27	.	.	.
lesk	120	1	The	the	DT
lesk	120	2	weather	weather	NN
lesk	120	3	is	be	VBZ
lesk	120	4	not	not	RB
lesk	120	5	that	that	RB
lesk	120	6	random	random	JJ
lesk	120	7	but	but	CC
lesk	120	8	it	-PRON-	PRP
lesk	120	9	has	have	VBZ
lesk	120	10	too	too	RB
lesk	120	11	much	much	JJ
lesk	120	12	variation	variation	NN
lesk	120	13	to	to	TO
lesk	120	14	be	be	VB
lesk	120	15	solved	solve	VBN
lesk	120	16	easily	easily	RB
lesk	120	17	by	by	IN
lesk	120	18	analogy	analogy	NN
lesk	120	19	.	.	.
lesk	121	1	If	if	IN
lesk	121	2	your	-PRON-	PRP$
lesk	121	3	problem	problem	NN
lesk	121	4	is	be	VBZ
lesk	121	5	very	very	RB
lesk	121	6	simple	simple	JJ
lesk	121	7	(	(	-LRB-
lesk	121	8	tic	tic	JJ
lesk	121	9	-	-	HYPH
lesk	121	10	tac	tac	JJ
lesk	121	11	-	-	HYPH
lesk	121	12	toe	toe	NN
lesk	121	13	)	)	-RRB-
lesk	121	14	you	-PRON-	PRP
lesk	121	15	could	could	MD
lesk	121	16	indeed	indeed	RB
lesk	121	17	write	write	VB
lesk	121	18	down	down	RP
lesk	121	19	each	each	DT
lesk	121	20	position	position	NN
lesk	121	21	and	and	CC
lesk	121	22	what	what	WP
lesk	121	23	the	the	DT
lesk	121	24	best	well	RBS
lesk	121	25	next	next	JJ
lesk	121	26	move	move	NN
lesk	121	27	is	be	VBZ
lesk	121	28	;	;	:
lesk	121	29	there	there	EX
lesk	121	30	are	be	VBP
lesk	121	31	only	only	RB
lesk	121	32	about	about	IN
lesk	121	33	255,000	255,000	CD
lesk	121	34	games	game	NNS
lesk	121	35	.	.	.
lesk	122	1	To	to	TO
lesk	122	2	deal	deal	VB
lesk	122	3	with	with	IN
lesk	122	4	more	more	JJR
lesk	122	5	realistic	realistic	JJ
lesk	122	6	problems	problem	NNS
lesk	122	7	,	,	,
lesk	122	8	much	much	JJ
lesk	122	9	of	of	IN
lesk	122	10	machine	machine	NN
lesk	122	11	learning	learning	NN
lesk	122	12	research	research	NN
lesk	122	13	is	be	VBZ
lesk	122	14	now	now	RB
lesk	122	15	focused	focus	VBN
lesk	122	16	on	on	IN
lesk	122	17	obtaining	obtain	VBG
lesk	122	18	larger	large	JJR
lesk	122	19	training	training	NN
lesk	122	20	sets	set	NNS
lesk	122	21	.	.	.
lesk	123	1	Instead	instead	RB
lesk	123	2	of	of	IN
lesk	123	3	trying	try	VBG
lesk	123	4	to	to	TO
lesk	123	5	learn	learn	VB
lesk	123	6	more	more	JJR
lesk	123	7	about	about	IN
lesk	123	8	the	the	DT
lesk	123	9	characteristics	characteristic	NNS
lesk	123	10	of	of	IN
lesk	123	11	a	a	DT
lesk	123	12	system	system	NN
lesk	123	13	that	that	WDT
lesk	123	14	is	be	VBZ
lesk	123	15	being	be	VBG
lesk	123	16	modeled	model	VBN
lesk	123	17	,	,	,
lesk	123	18	the	the	DT
lesk	123	19	effort	effort	NN
lesk	123	20	is	be	VBZ
lesk	123	21	driven	drive	VBN
lesk	123	22	by	by	IN
lesk	123	23	the	the	DT
lesk	123	24	dictum	dictum	NN
lesk	123	25	,	,	,
lesk	123	26	“	"	``
lesk	123	27	more	more	JJR
lesk	123	28	data	data	NN
lesk	123	29	beats	beat	VBZ
lesk	123	30	better	well	JJR
lesk	123	31	algorithms	algorithm	NNS
lesk	123	32	.	.	.
lesk	123	33	”	"	''
lesk	123	34	In	in	IN
lesk	123	35	a	a	DT
lesk	123	36	review	review	NN
lesk	123	37	of	of	IN
lesk	123	38	the	the	DT
lesk	123	39	history	history	NN
lesk	123	40	of	of	IN
lesk	123	41	speech	speech	NN
lesk	123	42	recognition	recognition	NN
lesk	123	43	,	,	,
lesk	123	44	Xuedong	Xuedong	NNP
lesk	123	45	Huang	Huang	NNP
lesk	123	46	,	,	,
lesk	123	47	James	James	NNP
lesk	123	48	Baker	Baker	NNP
lesk	123	49	,	,	,
lesk	123	50	and	and	CC
lesk	123	51	Raj	Raj	NNP
lesk	123	52	Reddy	Reddy	NNP
lesk	123	53	write	write	VBP
lesk	123	54	,	,	,
lesk	123	55	“	"	``
lesk	123	56	The	the	DT
lesk	123	57	power	power	NN
lesk	123	58	of	of	IN
lesk	123	59	these	these	DT
lesk	123	60	systems	system	NNS
lesk	123	61	arises	arise	VBZ
lesk	123	62	mainly	mainly	RB
lesk	123	63	from	from	IN
lesk	123	64	their	-PRON-	PRP$
lesk	123	65	ability	ability	NN
lesk	123	66	to	to	TO
lesk	123	67	collect	collect	VB
lesk	123	68	,	,	,
lesk	123	69	process	process	NN
lesk	123	70	,	,	,
lesk	123	71	and	and	CC
lesk	123	72	learn	learn	VB
lesk	123	73	from	from	IN
lesk	123	74	very	very	RB
lesk	123	75	large	large	JJ
lesk	123	76	datasets	dataset	NNS
lesk	123	77	.	.	.
lesk	124	1	The	the	DT
lesk	124	2	basic	basic	JJ
lesk	124	3	learning	learning	NN
lesk	124	4	and	and	CC
lesk	124	5	decoding	decode	VBG
lesk	124	6	algorithms	algorithm	NNS
lesk	124	7	have	have	VBP
lesk	124	8	not	not	RB
lesk	124	9	changed	change	VBN
lesk	124	10	substantially	substantially	RB
lesk	124	11	in	in	IN
lesk	124	12	40	40	CD
lesk	124	13	years	year	NNS
lesk	124	14	”	"	''
lesk	124	15	(	(	-LRB-
lesk	124	16	2014	2014	CD
lesk	124	17	)	)	-RRB-
lesk	124	18	.	.	.
lesk	125	1	Nevertheless	nevertheless	RB
lesk	125	2	,	,	,
lesk	125	3	speech	speech	NN
lesk	125	4	recognition	recognition	NN
lesk	125	5	has	have	VBZ
lesk	125	6	gone	go	VBN
lesk	125	7	from	from	IN
lesk	125	8	frustration	frustration	NN
lesk	125	9	to	to	IN
lesk	125	10	useful	useful	JJ
lesk	125	11	products	product	NNS
lesk	125	12	such	such	JJ
lesk	125	13	as	as	IN
lesk	125	14	dictation	dictation	NN
lesk	125	15	software	software	NN
lesk	125	16	or	or	CC
lesk	125	17	home	home	NN
lesk	125	18	appliances	appliance	NNS
lesk	125	19	.	.	.
lesk	126	1	Lacking	lack	VBG
lesk	126	2	a	a	DT
lesk	126	3	model	model	NN
lesk	126	4	,	,	,
lesk	126	5	however	however	RB
lesk	126	6	,	,	,
lesk	126	7	means	mean	VBZ
lesk	126	8	that	that	IN
lesk	126	9	we	-PRON-	PRP
lesk	126	10	wo	will	MD
lesk	126	11	n’t	not	RB
lesk	126	12	know	know	VB
lesk	126	13	the	the	DT
lesk	126	14	limits	limit	NNS
lesk	126	15	of	of	IN
lesk	126	16	the	the	DT
lesk	126	17	calculations	calculation	NNS
lesk	126	18	being	be	VBG
lesk	126	19	done	do	VBN
lesk	126	20	.	.	.
lesk	127	1	For	for	IN
lesk	127	2	example	example	NN
lesk	127	3	,	,	,
lesk	127	4	if	if	IN
lesk	127	5	you	-PRON-	PRP
lesk	127	6	have	have	VBP
lesk	127	7	some	some	DT
lesk	127	8	data	datum	NNS
lesk	127	9	that	that	WDT
lesk	127	10	looks	look	VBZ
lesk	127	11	quadratic	quadratic	JJ
lesk	127	12	,	,	,
lesk	127	13	but	but	CC
lesk	127	14	you	-PRON-	PRP
lesk	127	15	fit	fit	VBP
lesk	127	16	a	a	DT
lesk	127	17	linear	linear	JJ
lesk	127	18	model	model	NN
lesk	127	19	,	,	,
lesk	127	20	any	any	DT
lesk	127	21	attempt	attempt	NN
lesk	127	22	at	at	IN
lesk	127	23	extrapolation	extrapolation	NN
lesk	127	24	is	be	VBZ
lesk	127	25	fraught	fraught	JJ
lesk	127	26	with	with	IN
lesk	127	27	error	error	NN
lesk	127	28	.	.	.
lesk	128	1	If	if	IN
lesk	128	2	you	-PRON-	PRP
lesk	128	3	are	be	VBP
lesk	128	4	using	use	VBG
lesk	128	5	a	a	DT
lesk	128	6	“	"	``
lesk	128	7	black	black	JJ
lesk	128	8	box	box	NN
lesk	128	9	”	"	''
lesk	128	10	system	system	NN
lesk	128	11	,	,	,
lesk	128	12	you	-PRON-	PRP
lesk	128	13	do	do	VBP
lesk	128	14	n’t	not	RB
lesk	128	15	know	know	VB
lesk	128	16	when	when	WRB
lesk	128	17	this	this	DT
lesk	128	18	is	be	VBZ
lesk	128	19	happening	happen	VBG
lesk	128	20	.	.	.
lesk	129	1	And	and	CC
lesk	129	2	,	,	,
lesk	129	3	regrettably	regrettably	RB
lesk	129	4	,	,	,
lesk	129	5	many	many	JJ
lesk	129	6	of	of	IN
lesk	129	7	the	the	DT
lesk	129	8	AI	AI	NNP
lesk	129	9	software	software	NN
lesk	129	10	systems	system	NNS
lesk	129	11	are	be	VBP
lesk	129	12	sold	sell	VBN
lesk	129	13	as	as	IN
lesk	129	14	black	black	JJ
lesk	129	15	boxes	box	NNS
lesk	129	16	where	where	WRB
lesk	129	17	the	the	DT
lesk	129	18	purchasers	purchaser	NNS
lesk	129	19	and	and	CC
lesk	129	20	users	user	NNS
lesk	129	21	do	do	VBP
lesk	129	22	not	not	RB
lesk	129	23	have	have	VB
lesk	129	24	access	access	NN
lesk	129	25	to	to	IN
lesk	129	26	the	the	DT
lesk	129	27	process	process	NN
lesk	129	28	,	,	,
lesk	129	29	even	even	RB
lesk	129	30	if	if	IN
lesk	129	31	they	-PRON-	PRP
lesk	129	32	are	be	VBP
lesk	129	33	imagined	imagine	VBN
lesk	129	34	to	to	TO
lesk	129	35	be	be	VB
lesk	129	36	able	able	JJ
lesk	129	37	to	to	TO
lesk	129	38	understand	understand	VB
lesk	129	39	it	-PRON-	PRP
lesk	129	40	.	.	.
lesk	130	1	4	4	LS
lesk	130	2	.	.	.
lesk	131	1	What	what	WP
lesk	131	2	’s	’	VBZ
lesk	131	3	changing	change	VBG
lesk	131	4	Many	many	JJ
lesk	131	5	AI	ai	NN
lesk	131	6	researchers	researcher	NNS
lesk	131	7	are	be	VBP
lesk	131	8	sensitive	sensitive	JJ
lesk	131	9	to	to	IN
lesk	131	10	the	the	DT
lesk	131	11	risks	risk	NNS
lesk	131	12	,	,	,
lesk	131	13	especially	especially	RB
lesk	131	14	given	give	VBN
lesk	131	15	the	the	DT
lesk	131	16	publicity	publicity	NN
lesk	131	17	over	over	IN
lesk	131	18	self	self	NN
lesk	131	19	-	-	HYPH
lesk	131	20	driving	drive	VBG
lesk	131	21	cars	car	NNS
lesk	131	22	.	.	.
lesk	132	1	As	as	IN
lesk	132	2	the	the	DT
lesk	132	3	hype	hype	NN
lesk	132	4	over	over	IN
lesk	132	5	“	"	``
lesk	132	6	deep	deep	JJ
lesk	132	7	learning	learning	NN
lesk	132	8	”	"	''
lesk	132	9	built	build	VBN
lesk	132	10	up	up	RP
lesk	132	11	,	,	,
lesk	132	12	writers	writer	NNS
lesk	132	13	discussed	discuss	VBD
lesk	132	14	examples	example	NNS
lesk	132	15	such	such	JJ
lesk	132	16	as	as	IN
lesk	132	17	a	a	DT
lesk	132	18	Pittsburgh	Pittsburgh	NNP
lesk	132	19	medical	medical	JJ
lesk	132	20	system	system	NN
lesk	132	21	that	that	WDT
lesk	132	22	proposed	propose	VBD
lesk	132	23	to	to	TO
lesk	132	24	send	send	VB
lesk	132	25	patients	patient	NNS
lesk	132	26	with	with	IN
lesk	132	27	both	both	DT
lesk	132	28	pneumonia	pneumonia	NN
lesk	132	29	and	and	CC
lesk	132	30	asthma	asthma	NN
lesk	132	31	home	home	RB
lesk	132	32	,	,	,
lesk	132	33	because	because	IN
lesk	132	34	the	the	DT
lesk	132	35	computer	computer	NN
lesk	132	36	had	have	VBD
lesk	132	37	not	not	RB
lesk	132	38	understood	understand	VBN
lesk	132	39	that	that	IN
lesk	132	40	patients	patient	NNS
lesk	132	41	with	with	IN
lesk	132	42	both	both	DT
lesk	132	43	problems	problem	NNS
lesk	132	44	were	be	VBD
lesk	132	45	actually	actually	RB
lesk	132	46	being	be	VBG
lesk	132	47	sent	send	VBN
lesk	132	48	to	to	IN
lesk	132	49	the	the	DT
lesk	132	50	ICU	ICU	NNP
lesk	132	51	(	(	-LRB-
lesk	132	52	Bornstein	Bornstein	NNP
lesk	132	53	2016	2016	CD
lesk	132	54	;	;	:
lesk	132	55	Caruana	Caruana	NNP
lesk	132	56	et	et	NNP
lesk	132	57	al	al	NNP
lesk	132	58	.	.	.
lesk	133	1	2015	2015	CD
lesk	133	2	)	)	-RRB-
lesk	133	3	.	.	.
lesk	134	1	Many	many	JJ
lesk	134	2	people	people	NNS
lesk	134	3	work	work	VBP
lesk	134	4	on	on	IN
lesk	134	5	ways	way	NNS
lesk	134	6	of	of	IN
lesk	134	7	explaining	explain	VBG
lesk	134	8	or	or	CC
lesk	134	9	presenting	present	VBG
lesk	134	10	neural	neural	JJ
lesk	134	11	net	net	JJ
lesk	134	12	software	software	NN
lesk	134	13	(	(	-LRB-
lesk	134	14	Harley	Harley	NNP
lesk	134	15	2015	2015	CD
lesk	134	16	)	)	-RRB-
lesk	134	17	.	.	.
lesk	135	1	Most	most	RBS
lesk	135	2	important	important	JJ
lesk	135	3	,	,	,
lesk	135	4	perhaps	perhaps	RB
lesk	135	5	,	,	,
lesk	135	6	are	be	VBP
lesk	135	7	new	new	JJ
lesk	135	8	EU	EU	NNP
lesk	135	9	regulations	regulation	NNS
lesk	135	10	that	that	WDT
lesk	135	11	prohibit	prohibit	VBP
lesk	135	12	automated	automate	VBD
lesk	135	13	decision	decision	NN
lesk	135	14	making	make	VBG
lesk	135	15	that	that	WDT
lesk	135	16	affects	affect	VBZ
lesk	135	17	EU	EU	NNP
lesk	135	18	citizens	citizen	NNS
lesk	135	19	,	,	,
lesk	135	20	and	and	CC
lesk	135	21	provides	provide	VBZ
lesk	135	22	a	a	DT
lesk	135	23	“	"	``
lesk	135	24	right	right	NN
lesk	135	25	of	of	IN
lesk	135	26	explanation	explanation	NN
lesk	135	27	”	"	''
lesk	135	28	(	(	-LRB-
lesk	135	29	Metz	Metz	NNP
lesk	135	30	2016	2016	CD
lesk	135	31	)	)	-RRB-
lesk	135	32	.	.	.
lesk	136	1	We	-PRON-	PRP
lesk	136	2	recognize	recognize	VBP
lesk	136	3	that	that	IN
lesk	136	4	systems	system	NNS
lesk	136	5	which	which	WDT
lesk	136	6	do	do	VBP
lesk	136	7	n’t	not	RB
lesk	136	8	rely	rely	VB
lesk	136	9	on	on	IN
lesk	136	10	a	a	DT
lesk	136	11	mathematical	mathematical	JJ
lesk	136	12	model	model	NN
lesk	136	13	may	may	MD
lesk	136	14	be	be	VB
lesk	136	15	cheaper	cheap	JJR
lesk	136	16	to	to	TO
lesk	136	17	build	build	VB
lesk	136	18	than	than	IN
lesk	136	19	one	one	CD
lesk	136	20	where	where	WRB
lesk	136	21	the	the	DT
lesk	136	22	coders	coder	NNS
lesk	136	23	understand	understand	VBP
lesk	136	24	what	what	WP
lesk	136	25	is	be	VBZ
lesk	136	26	going	go	VBG
lesk	136	27	on	on	RP
lesk	136	28	.	.	.
lesk	137	1	More	more	RBR
lesk	137	2	serious	serious	JJ
lesk	137	3	is	be	VBZ
lesk	137	4	that	that	IN
lesk	137	5	they	-PRON-	PRP
lesk	137	6	may	may	MD
lesk	137	7	be	be	VB
lesk	137	8	more	more	RBR
lesk	137	9	accurate	accurate	JJ
lesk	137	10	.	.	.
lesk	138	1	This	this	DT
lesk	138	2	image	image	NN
lesk	138	3	is	be	VBZ
lesk	138	4	from	from	IN
lesk	138	5	the	the	DT
lesk	138	6	same	same	JJ
lesk	138	7	article	article	NN
lesk	138	8	on	on	IN
lesk	138	9	understandability	understandability	NN
lesk	138	10	(	(	-LRB-
lesk	138	11	Bornstein	Bornstein	NNP
lesk	138	12	2016	2016	CD
lesk	138	13	)	)	-RRB-
lesk	138	14	.	.	.
lesk	139	1	If	if	IN
lesk	139	2	there	there	EX
lesk	139	3	really	really	RB
lesk	139	4	is	be	VBZ
lesk	139	5	a	a	DT
lesk	139	6	tradeoff	tradeoff	NN
lesk	139	7	between	between	IN
lesk	139	8	what	what	WP
lesk	139	9	will	will	MD
lesk	139	10	solve	solve	VB
lesk	139	11	the	the	DT
lesk	139	12	problem	problem	NN
lesk	139	13	and	and	CC
lesk	139	14	what	what	WP
lesk	139	15	can	can	MD
lesk	139	16	be	be	VB
lesk	139	17	explained	explain	VBN
lesk	139	18	,	,	,
lesk	139	19	we	-PRON-	PRP
lesk	139	20	know	know	VBP
lesk	139	21	that	that	IN
lesk	139	22	many	many	JJ
lesk	139	23	system	system	NN
lesk	139	24	builders	builder	NNS
lesk	139	25	will	will	MD
lesk	139	26	choose	choose	VB
lesk	139	27	to	to	TO
lesk	139	28	solve	solve	VB
lesk	139	29	the	the	DT
lesk	139	30	problem	problem	NN
lesk	139	31	.	.	.
lesk	140	1	And	and	CC
lesk	140	2	yet	yet	RB
lesk	140	3	even	even	RB
lesk	140	4	having	have	VBG
lesk	140	5	explanations	explanation	NNS
lesk	140	6	may	may	MD
lesk	140	7	not	not	RB
lesk	140	8	be	be	VB
lesk	140	9	an	an	DT
lesk	140	10	answer	answer	NN
lesk	140	11	;	;	:
lesk	140	12	a	a	DT
lesk	140	13	key	key	JJ
lesk	140	14	paper	paper	NN
lesk	140	15	on	on	IN
lesk	140	16	interpretability	interpretability	NN
lesk	140	17	discusses	discuss	VBZ
lesk	140	18	the	the	DT
lesk	140	19	complexities	complexity	NNS
lesk	140	20	of	of	IN
lesk	140	21	meaning	meaning	NN
lesk	140	22	related	relate	VBN
lesk	140	23	to	to	IN
lesk	140	24	explanation	explanation	NN
lesk	140	25	,	,	,
lesk	140	26	causality	causality	NN
lesk	140	27	,	,	,
lesk	140	28	and	and	CC
lesk	140	29	modeling	model	VBG
lesk	140	30	(	(	-LRB-
lesk	140	31	Lipton	Lipton	NNP
lesk	140	32	2018	2018	CD
lesk	140	33	)	)	-RRB-
lesk	140	34	.	.	.
lesk	141	1	Arend	Arend	NNP
lesk	141	2	Hintze	Hintze	NNP
lesk	141	3	has	have	VBZ
lesk	141	4	noted	note	VBN
lesk	141	5	that	that	IN
lesk	141	6	we	-PRON-	PRP
lesk	141	7	do	do	VBP
lesk	141	8	not	not	RB
lesk	141	9	always	always	RB
lesk	141	10	impose	impose	VB
lesk	141	11	a	a	DT
lesk	141	12	demand	demand	NN
lesk	141	13	for	for	IN
lesk	141	14	explanation	explanation	NN
lesk	141	15	on	on	IN
lesk	141	16	people	people	NNS
lesk	141	17	.	.	.
lesk	142	1	I	-PRON-	PRP
lesk	142	2	can	can	MD
lesk	142	3	write	write	VB
lesk	142	4	that	that	IN
lesk	142	5	the	the	DT
lesk	142	6	New	New	NNP
lesk	142	7	York	York	NNP
lesk	142	8	Public	Public	NNP
lesk	142	9	Library	Library	NNP
lesk	142	10	main	main	JJ
lesk	142	11	building	building	NN
lesk	142	12	is	be	VBZ
lesk	142	13	well	well	RB
lesk	142	14	proportioned	proportioned	JJ
lesk	142	15	and	and	CC
lesk	142	16	attractive	attractive	JJ
lesk	142	17	without	without	IN
lesk	142	18	anyone	anyone	NN
lesk	142	19	expecting	expect	VBG
lesk	142	20	that	that	IN
lesk	142	21	I	-PRON-	PRP
lesk	142	22	will	will	MD
lesk	142	23	recite	recite	VB
lesk	142	24	its	-PRON-	PRP$
lesk	142	25	dimensions	dimension	NNS
lesk	142	26	or	or	CC
lesk	142	27	the	the	DT
lesk	142	28	source	source	NN
lesk	142	29	of	of	IN
lesk	142	30	the	the	DT
lesk	142	31	marble	marble	NN
lesk	142	32	used	use	VBN
lesk	142	33	to	to	TO
lesk	142	34	construct	construct	VB
lesk	142	35	it	-PRON-	PRP
lesk	142	36	.	.	.
lesk	143	1	And	and	CC
lesk	143	2	for	for	IN
lesk	143	3	some	some	DT
lesk	143	4	problems	problem	NNS
lesk	143	5	that	that	WDT
lesk	143	6	’s	’	VBZ
lesk	143	7	fine	fine	JJ
lesk	143	8	:	:	:
lesk	143	9	I	-PRON-	PRP
lesk	143	10	do	do	VBP
lesk	143	11	n’t	not	RB
lesk	143	12	care	care	VB
lesk	143	13	how	how	WRB
lesk	143	14	my	-PRON-	PRP$
lesk	143	15	camera	camera	NN
lesk	143	16	decides	decide	VBZ
lesk	143	17	on	on	IN
lesk	143	18	the	the	DT
lesk	143	19	focus	focus	JJ
lesk	143	20	distance	distance	NN
lesk	143	21	to	to	IN
lesk	143	22	the	the	DT
lesk	143	23	subject	subject	NN
lesk	143	24	.	.	.
lesk	144	1	Where	where	WRB
lesk	144	2	it	-PRON-	PRP
lesk	144	3	matters	matter	VBZ
lesk	144	4	,	,	,
lesk	144	5	however	however	RB
lesk	144	6	,	,	,
lesk	144	7	we	-PRON-	PRP
lesk	144	8	often	often	RB
lesk	144	9	want	want	VBP
lesk	144	10	explanations	explanation	NNS
lesk	144	11	;	;	:
lesk	144	12	the	the	DT
lesk	144	13	hard	hard	JJ
lesk	144	14	ethical	ethical	JJ
lesk	144	15	problem	problem	NN
lesk	144	16	,	,	,
lesk	144	17	as	as	IN
lesk	144	18	noted	note	VBN
lesk	144	19	before	before	RB
lesk	144	20	,	,	,
lesk	144	21	is	be	VBZ
lesk	144	22	if	if	IN
lesk	144	23	better	well	JJR
lesk	144	24	performance	performance	NN
lesk	144	25	can	can	MD
lesk	144	26	be	be	VB
lesk	144	27	achieved	achieve	VBN
lesk	144	28	in	in	IN
lesk	144	29	an	an	DT
lesk	144	30	inexplicable	inexplicable	JJ
lesk	144	31	way	way	NN
lesk	144	32	.	.	.
lesk	145	1	5	5	CD
lesk	145	2	.	.	.
lesk	146	1	Recommendations	recommendation	NNS
lesk	146	2	2017	2017	CD
lesk	146	3	saw	see	VBD
lesk	146	4	the	the	DT
lesk	146	5	publication	publication	NN
lesk	146	6	of	of	IN
lesk	146	7	the	the	DT
lesk	146	8	“	"	``
lesk	146	9	Asilomar	Asilomar	NNP
lesk	146	10	AI	AI	NNP
lesk	146	11	principles	principle	NNS
lesk	146	12	”	"	''
lesk	146	13	(	(	-LRB-
lesk	146	14	2017	2017	CD
lesk	146	15	)	)	-RRB-
lesk	146	16	.	.	.
lesk	147	1	Two	two	CD
lesk	147	2	of	of	IN
lesk	147	3	these	these	DT
lesk	147	4	principles	principle	NNS
lesk	147	5	are	be	VBP
lesk	147	6	:	:	:
lesk	147	7	*	*	NFP
lesk	147	8	Safety	safety	NN
lesk	147	9	:	:	:
lesk	147	10	AI	AI	NNP
lesk	147	11	systems	system	NNS
lesk	147	12	should	should	MD
lesk	147	13	be	be	VB
lesk	147	14	safe	safe	JJ
lesk	147	15	and	and	CC
lesk	147	16	secure	secure	JJ
lesk	147	17	throughout	throughout	IN
lesk	147	18	their	-PRON-	PRP$
lesk	147	19	operational	operational	JJ
lesk	147	20	lifetime	lifetime	NN
lesk	147	21	,	,	,
lesk	147	22	and	and	CC
lesk	147	23	verifiably	verifiably	RB
lesk	147	24	so	so	RB
lesk	147	25	where	where	WRB
lesk	147	26	applicable	applicable	JJ
lesk	147	27	and	and	CC
lesk	147	28	feasible	feasible	JJ
lesk	147	29	.	.	.
lesk	148	1	*	*	NFP
lesk	148	2	Failure	failure	NN
lesk	148	3	Transparency	transparency	NN
lesk	148	4	:	:	:
lesk	148	5	If	if	IN
lesk	148	6	an	an	DT
lesk	148	7	AI	AI	NNP
lesk	148	8	system	system	NN
lesk	148	9	causes	cause	VBZ
lesk	148	10	harm	harm	NN
lesk	148	11	,	,	,
lesk	148	12	it	-PRON-	PRP
lesk	148	13	should	should	MD
lesk	148	14	be	be	VB
lesk	148	15	possible	possible	JJ
lesk	148	16	to	to	TO
lesk	148	17	ascertain	ascertain	VB
lesk	148	18	why	why	WRB
lesk	148	19	.	.	.
lesk	149	1	The	the	DT
lesk	149	2	problem	problem	NN
lesk	149	3	is	be	VBZ
lesk	149	4	that	that	IN
lesk	149	5	the	the	DT
lesk	149	6	technology	technology	NN
lesk	149	7	used	use	VBN
lesk	149	8	to	to	TO
lesk	149	9	build	build	VB
lesk	149	10	many	many	JJ
lesk	149	11	systems	system	NNS
lesk	149	12	does	do	VBZ
lesk	149	13	not	not	RB
lesk	149	14	enable	enable	VB
lesk	149	15	verifiability	verifiability	NN
lesk	149	16	and	and	CC
lesk	149	17	explanation	explanation	NN
lesk	149	18	.	.	.
lesk	150	1	Similarly	similarly	RB
lesk	150	2	the	the	DT
lesk	150	3	World	World	NNP
lesk	150	4	Economic	Economic	NNP
lesk	150	5	Forum	Forum	NNP
lesk	150	6	calls	call	VBZ
lesk	150	7	for	for	IN
lesk	150	8	protection	protection	NN
lesk	150	9	against	against	IN
lesk	150	10	discrimination	discrimination	NN
lesk	150	11	but	but	CC
lesk	150	12	notes	note	VBZ
lesk	150	13	many	many	JJ
lesk	150	14	ways	way	NNS
lesk	150	15	in	in	IN
lesk	150	16	which	which	WDT
lesk	150	17	technology	technology	NN
lesk	150	18	can	can	MD
lesk	150	19	have	have	VB
lesk	150	20	unanticipated	unanticipated	JJ
lesk	150	21	and	and	CC
lesk	150	22	undesirable	undesirable	JJ
lesk	150	23	effects	effect	NNS
lesk	150	24	as	as	IN
lesk	150	25	a	a	DT
lesk	150	26	result	result	NN
lesk	150	27	of	of	IN
lesk	150	28	machine	machine	NN
lesk	150	29	learning	learning	NN
lesk	150	30	(	(	-LRB-
lesk	150	31	“	"	``
lesk	150	32	How	how	WRB
lesk	150	33	”	"	''
lesk	150	34	2018	2018	CD
lesk	150	35	)	)	-RRB-
lesk	150	36	.	.	.
lesk	151	1	Historically	historically	RB
lesk	151	2	there	there	EX
lesk	151	3	has	have	VBZ
lesk	151	4	been	be	VBN
lesk	151	5	and	and	CC
lesk	151	6	continues	continue	VBZ
lesk	151	7	to	to	TO
lesk	151	8	be	be	VB
lesk	151	9	too	too	RB
lesk	151	10	much	much	JJ
lesk	151	11	hype	hype	NN
lesk	151	12	.	.	.
lesk	152	1	An	an	DT
lesk	152	2	important	important	JJ
lesk	152	3	image	image	NN
lesk	152	4	recognition	recognition	NN
lesk	152	5	task	task	NN
lesk	152	6	is	be	VBZ
lesk	152	7	distinguishing	distinguish	VBG
lesk	152	8	malignant	malignant	JJ
lesk	152	9	and	and	CC
lesk	152	10	benign	benign	JJ
lesk	152	11	spots	spot	NNS
lesk	152	12	on	on	IN
lesk	152	13	mammograms	mammogram	NNS
lesk	152	14	.	.	.
lesk	153	1	There	there	EX
lesk	153	2	have	have	VBP
lesk	153	3	been	be	VBN
lesk	153	4	promises	promise	NNS
lesk	153	5	for	for	IN
lesk	153	6	decades	decade	NNS
lesk	153	7	that	that	IN
lesk	153	8	computers	computer	NNS
lesk	153	9	would	would	MD
lesk	153	10	do	do	VB
lesk	153	11	this	this	DT
lesk	153	12	better	well	JJR
lesk	153	13	than	than	IN
lesk	153	14	radiologists	radiologist	NNS
lesk	153	15	.	.	.
lesk	154	1	Here	here	RB
lesk	154	2	are	be	VBP
lesk	154	3	examples	example	NNS
lesk	154	4	from	from	IN
lesk	154	5	1995	1995	CD
lesk	154	6	(	(	-LRB-
lesk	154	7	“	"	``
lesk	154	8	computer	computer	NN
lesk	154	9	-	-	HYPH
lesk	154	10	aided	aid	VBN
lesk	154	11	diagnosis	diagnosis	NN
lesk	154	12	can	can	MD
lesk	154	13	improve	improve	VB
lesk	154	14	radiologists	radiologist	NNS
lesk	154	15	’	’	POS
lesk	154	16	observational	observational	JJ
lesk	154	17	performance	performance	NN
lesk	154	18	”	"	''
lesk	154	19	)	)	-RRB-
lesk	154	20	(	(	-LRB-
lesk	154	21	Schmidt	Schmidt	NNP
lesk	154	22	and	and	CC
lesk	154	23	Nishikawa	Nishikawa	NNP
lesk	154	24	)	)	-RRB-
lesk	154	25	and	and	CC
lesk	154	26	2009	2009	CD
lesk	154	27	(	(	-LRB-
lesk	154	28	“	"	``
lesk	154	29	The	the	DT
lesk	154	30	Bayesian	bayesian	JJ
lesk	154	31	network	network	NN
lesk	154	32	significantly	significantly	RB
lesk	154	33	exceeded	exceed	VBD
lesk	154	34	the	the	DT
lesk	154	35	performance	performance	NN
lesk	154	36	of	of	IN
lesk	154	37	interpreting	interpret	VBG
lesk	154	38	radiologists	radiologist	NNS
lesk	154	39	”	"	''
lesk	154	40	)	)	-RRB-
lesk	154	41	(	(	-LRB-
lesk	154	42	Burnside	Burnside	NNP
lesk	154	43	et	et	FW
lesk	154	44	al	al	NNP
lesk	154	45	.	.	.
lesk	154	46	)	)	-RRB-
lesk	154	47	.	.	.
lesk	155	1	A	a	DT
lesk	155	2	typical	typical	JJ
lesk	155	3	recent	recent	JJ
lesk	155	4	AI	AI	NNP
lesk	155	5	paper	paper	NN
lesk	155	6	to	to	TO
lesk	155	7	do	do	VB
lesk	155	8	this	this	DT
lesk	155	9	with	with	IN
lesk	155	10	convolutional	convolutional	JJ
lesk	155	11	neural	neural	JJ
lesk	155	12	nets	net	NNS
lesk	155	13	reports	report	VBZ
lesk	155	14	90	90	CD
lesk	155	15	%	%	NN
lesk	155	16	accuracy	accuracy	NN
lesk	155	17	(	(	-LRB-
lesk	155	18	Singh	Singh	NNP
lesk	155	19	et	et	NNP
lesk	155	20	al	al	NNP
lesk	155	21	.	.	.
lesk	156	1	2020	2020	CD
lesk	156	2	)	)	-RRB-
lesk	156	3	.	.	.
lesk	157	1	To	to	TO
lesk	157	2	put	put	VB
lesk	157	3	this	this	DT
lesk	157	4	in	in	IN
lesk	157	5	perspective	perspective	NN
lesk	157	6	,	,	,
lesk	157	7	the	the	DT
lesk	157	8	problem	problem	NN
lesk	157	9	is	be	VBZ
lesk	157	10	complex	complex	JJ
lesk	157	11	,	,	,
lesk	157	12	but	but	CC
lesk	157	13	some	some	DT
lesk	157	14	examples	example	NNS
lesk	157	15	are	be	VBP
lesk	157	16	more	more	RBR
lesk	157	17	straightforward	straightforward	JJ
lesk	157	18	,	,	,
lesk	157	19	and	and	CC
lesk	157	20	even	even	RB
lesk	157	21	pigeons	pigeon	NNS
lesk	157	22	can	can	MD
lesk	157	23	reach	reach	VB
lesk	157	24	85	85	CD
lesk	157	25	%	%	NN
lesk	157	26	(	(	-LRB-
lesk	157	27	Levenson	Levenson	NNP
lesk	157	28	et	et	FW
lesk	157	29	al	al	NNP
lesk	157	30	.	.	.
lesk	158	1	2015	2015	CD
lesk	158	2	)	)	-RRB-
lesk	158	3	.	.	.
lesk	159	1	A	a	DT
lesk	159	2	serious	serious	JJ
lesk	159	3	recent	recent	JJ
lesk	159	4	review	review	NN
lesk	159	5	is	be	VBZ
lesk	159	6	“	"	``
lesk	159	7	Diagnostic	Diagnostic	NNP
lesk	159	8	Accuracy	Accuracy	NNP
lesk	159	9	of	of	IN
lesk	159	10	Digital	Digital	NNP
lesk	159	11	Screening	Screening	NNP
lesk	159	12	Mammography	Mammography	NNP
lesk	159	13	With	with	IN
lesk	159	14	and	and	CC
lesk	159	15	Without	without	IN
lesk	159	16	Computer	Computer	NNP
lesk	159	17	-	-	HYPH
lesk	159	18	Aided	aid	VBN
lesk	159	19	Detection	detection	NN
lesk	159	20	”	"	''
lesk	159	21	(	(	-LRB-
lesk	159	22	Lehman	Lehman	NNP
lesk	159	23	et	et	NNP
lesk	159	24	al	al	NNP
lesk	159	25	.	.	.
lesk	160	1	2015	2015	CD
lesk	160	2	)	)	-RRB-
lesk	160	3	.	.	.
lesk	161	1	Very	very	RB
lesk	161	2	recently	recently	RB
lesk	161	3	there	there	EX
lesk	161	4	was	be	VBD
lesk	161	5	another	another	DT
lesk	161	6	claim	claim	NN
lesk	161	7	that	that	IN
lesk	161	8	computers	computer	NNS
lesk	161	9	have	have	VBP
lesk	161	10	surpassed	surpass	VBN
lesk	161	11	radiologists	radiologist	NNS
lesk	161	12	(	(	-LRB-
lesk	161	13	Walsh	Walsh	NNP
lesk	161	14	2020	2020	CD
lesk	161	15	)	)	-RRB-
lesk	161	16	;	;	:
lesk	161	17	we	-PRON-	PRP
lesk	161	18	will	will	MD
lesk	161	19	have	have	VB
lesk	161	20	to	to	TO
lesk	161	21	await	await	VB
lesk	161	22	evaluation	evaluation	NN
lesk	161	23	.	.	.
lesk	162	1	As	as	IN
lesk	162	2	with	with	IN
lesk	162	3	many	many	JJ
lesk	162	4	claims	claim	NNS
lesk	162	5	of	of	IN
lesk	162	6	medical	medical	JJ
lesk	162	7	progress	progress	NN
lesk	162	8	,	,	,
lesk	162	9	replicability	replicability	NN
lesk	162	10	and	and	CC
lesk	162	11	evaluation	evaluation	NN
lesk	162	12	are	be	VBP
lesk	162	13	needed	need	VBN
lesk	162	14	before	before	IN
lesk	162	15	doctors	doctor	NNS
lesk	162	16	will	will	MD
lesk	162	17	be	be	VB
lesk	162	18	willing	willing	JJ
lesk	162	19	to	to	TO
lesk	162	20	believe	believe	VB
lesk	162	21	them	-PRON-	PRP
lesk	162	22	.	.	.
lesk	163	1	What	what	WP
lesk	163	2	should	should	MD
lesk	163	3	we	-PRON-	PRP
lesk	163	4	do	do	VB
lesk	163	5	?	?	.
lesk	164	1	Software	software	NN
lesk	164	2	testing	testing	NN
lesk	164	3	generally	generally	RB
lesk	164	4	is	be	VBZ
lesk	164	5	a	a	DT
lesk	164	6	decades	decade	NNS
lesk	164	7	-	-	HYPH
lesk	164	8	old	old	JJ
lesk	164	9	discipline	discipline	NN
lesk	164	10	,	,	,
lesk	164	11	and	and	CC
lesk	164	12	many	many	JJ
lesk	164	13	basic	basic	JJ
lesk	164	14	principles	principle	NNS
lesk	164	15	of	of	IN
lesk	164	16	regression	regression	NN
lesk	164	17	testing	testing	NN
lesk	164	18	apply	apply	NN
lesk	164	19	here	here	RB
lesk	164	20	also	also	RB
lesk	164	21	:	:	:
lesk	164	22	·	·	NFP
lesk	164	23	Test	test	NN
lesk	164	24	data	datum	NNS
lesk	164	25	should	should	MD
lesk	164	26	cover	cover	VB
lesk	164	27	the	the	DT
lesk	164	28	full	full	JJ
lesk	164	29	range	range	NN
lesk	164	30	of	of	IN
lesk	164	31	expected	expect	VBN
lesk	164	32	input	input	NN
lesk	164	33	.	.	.
lesk	165	1	·	·	NFP
lesk	165	2	Test	test	NN
lesk	165	3	data	datum	NNS
lesk	165	4	should	should	MD
lesk	165	5	also	also	RB
lesk	165	6	cover	cover	VB
lesk	165	7	unexpected	unexpected	JJ
lesk	165	8	and	and	CC
lesk	165	9	even	even	RB
lesk	165	10	illegal	illegal	JJ
lesk	165	11	input	input	NN
lesk	165	12	.	.	.
lesk	166	1	·	·	NFP
lesk	166	2	Test	test	NN
lesk	166	3	data	datum	NNS
lesk	166	4	should	should	MD
lesk	166	5	include	include	VB
lesk	166	6	known	know	VBN
lesk	166	7	past	past	JJ
lesk	166	8	failures	failure	NNS
lesk	166	9	believed	believe	VBN
lesk	166	10	cleared	clear	VBN
lesk	166	11	up	up	RP
lesk	166	12	.	.	.
lesk	167	1	·	·	NFP
lesk	167	2	Test	test	NN
lesk	167	3	data	datum	NNS
lesk	167	4	should	should	MD
lesk	167	5	exercise	exercise	VB
lesk	167	6	all	all	DT
lesk	167	7	parts	part	NNS
lesk	167	8	of	of	IN
lesk	167	9	the	the	DT
lesk	167	10	program	program	NN
lesk	167	11	,	,	,
lesk	167	12	and	and	CC
lesk	167	13	all	all	DT
lesk	167	14	important	important	JJ
lesk	167	15	paths	path	NNS
lesk	167	16	(	(	-LRB-
lesk	167	17	coverage	coverage	NN
lesk	167	18	)	)	-RRB-
lesk	167	19	.	.	.
lesk	168	1	·	·	NFP
lesk	168	2	Test	test	NN
lesk	168	3	data	datum	NNS
lesk	168	4	should	should	MD
lesk	168	5	include	include	VB
lesk	168	6	a	a	DT
lesk	168	7	set	set	NN
lesk	168	8	of	of	IN
lesk	168	9	data	datum	NNS
lesk	168	10	which	which	WDT
lesk	168	11	is	be	VBZ
lesk	168	12	representative	representative	JJ
lesk	168	13	of	of	IN
lesk	168	14	the	the	DT
lesk	168	15	distribution	distribution	NN
lesk	168	16	of	of	IN
lesk	168	17	actual	actual	JJ
lesk	168	18	data	datum	NNS
lesk	168	19	,	,	,
lesk	168	20	to	to	TO
lesk	168	21	be	be	VB
lesk	168	22	used	use	VBN
lesk	168	23	for	for	IN
lesk	168	24	timing	timing	NN
lesk	168	25	purposes	purpose	NNS
lesk	168	26	.	.	.
lesk	169	1	It	-PRON-	PRP
lesk	169	2	is	be	VBZ
lesk	169	3	difficult	difficult	JJ
lesk	169	4	to	to	TO
lesk	169	5	apply	apply	VB
lesk	169	6	these	these	DT
lesk	169	7	ideas	idea	NNS
lesk	169	8	in	in	IN
lesk	169	9	parts	part	NNS
lesk	169	10	of	of	IN
lesk	169	11	the	the	DT
lesk	169	12	AI	AI	NNP
lesk	169	13	world	world	NN
lesk	169	14	.	.	.
lesk	170	1	If	if	IN
lesk	170	2	the	the	DT
lesk	170	3	allowed	allow	VBN
lesk	170	4	input	input	NN
lesk	170	5	is	be	VBZ
lesk	170	6	speech	speech	NN
lesk	170	7	,	,	,
lesk	170	8	there	there	EX
lesk	170	9	is	be	VBZ
lesk	170	10	no	no	DT
lesk	170	11	exhaustive	exhaustive	JJ
lesk	170	12	list	list	NN
lesk	170	13	of	of	IN
lesk	170	14	utterances	utterance	NNS
lesk	170	15	which	which	WDT
lesk	170	16	can	can	MD
lesk	170	17	be	be	VB
lesk	170	18	sampled	sample	VBN
lesk	170	19	.	.	.
lesk	171	1	If	if	IN
lesk	171	2	a	a	DT
lesk	171	3	black	black	JJ
lesk	171	4	-	-	HYPH
lesk	171	5	box	box	NN
lesk	171	6	commercial	commercial	JJ
lesk	171	7	machine	machine	NN
lesk	171	8	learning	learning	NN
lesk	171	9	package	package	NN
lesk	171	10	is	be	VBZ
lesk	171	11	being	be	VBG
lesk	171	12	used	use	VBN
lesk	171	13	,	,	,
lesk	171	14	there	there	EX
lesk	171	15	is	be	VBZ
lesk	171	16	no	no	DT
lesk	171	17	way	way	NN
lesk	171	18	to	to	TO
lesk	171	19	ask	ask	VB
lesk	171	20	about	about	IN
lesk	171	21	coverage	coverage	NN
lesk	171	22	of	of	IN
lesk	171	23	any	any	DT
lesk	171	24	number	number	NN
lesk	171	25	of	of	IN
lesk	171	26	test	test	NN
lesk	171	27	cases	case	NNS
lesk	171	28	.	.	.
lesk	172	1	If	if	IN
lesk	172	2	a	a	DT
lesk	172	3	program	program	NN
lesk	172	4	is	be	VBZ
lesk	172	5	constantly	constantly	RB
lesk	172	6	learning	learn	VBG
lesk	172	7	from	from	IN
lesk	172	8	new	new	JJ
lesk	172	9	data	datum	NNS
lesk	172	10	,	,	,
lesk	172	11	there	there	EX
lesk	172	12	is	be	VBZ
lesk	172	13	no	no	DT
lesk	172	14	list	list	NN
lesk	172	15	of	of	IN
lesk	172	16	previously	previously	RB
lesk	172	17	fixed	fix	VBN
lesk	172	18	failures	failure	NNS
lesk	172	19	to	to	TO
lesk	172	20	be	be	VB
lesk	172	21	collected	collect	VBN
lesk	172	22	that	that	DT
lesk	172	23	reflects	reflect	VBZ
lesk	172	24	the	the	DT
lesk	172	25	constantly	constantly	RB
lesk	172	26	changing	change	VBG
lesk	172	27	program	program	NN
lesk	172	28	.	.	.
lesk	173	1	And	and	CC
lesk	173	2	obviously	obviously	RB
lesk	173	3	the	the	DT
lesk	173	4	circumstances	circumstance	NNS
lesk	173	5	of	of	IN
lesk	173	6	use	use	NN
lesk	173	7	matter	matter	NN
lesk	173	8	.	.	.
lesk	174	1	We	-PRON-	PRP
lesk	174	2	may	may	MD
lesk	174	3	well	well	RB
lesk	174	4	,	,	,
lesk	174	5	as	as	IN
lesk	174	6	a	a	DT
lesk	174	7	society	society	NN
lesk	174	8	,	,	,
lesk	174	9	decide	decide	VB
lesk	174	10	that	that	IN
lesk	174	11	forcing	force	VBG
lesk	174	12	banks	bank	NNS
lesk	174	13	evaluating	evaluate	VBG
lesk	174	14	loan	loan	NN
lesk	174	15	applications	application	NNS
lesk	174	16	to	to	TO
lesk	174	17	use	use	VB
lesk	174	18	decision	decision	NN
lesk	174	19	trees	tree	NNS
lesk	174	20	instead	instead	RB
lesk	174	21	of	of	IN
lesk	174	22	deep	deep	JJ
lesk	174	23	learning	learning	NN
lesk	174	24	is	be	VBZ
lesk	174	25	appropriate	appropriate	JJ
lesk	174	26	,	,	,
lesk	174	27	so	so	IN
lesk	174	28	that	that	IN
lesk	174	29	we	-PRON-	PRP
lesk	174	30	know	know	VBP
lesk	174	31	whether	whether	IN
lesk	174	32	illegal	illegal	JJ
lesk	174	33	discrimination	discrimination	NN
lesk	174	34	is	be	VBZ
lesk	174	35	going	go	VBG
lesk	174	36	on	on	RP
lesk	174	37	,	,	,
lesk	174	38	even	even	RB
lesk	174	39	if	if	IN
lesk	174	40	this	this	DT
lesk	174	41	raises	raise	VBZ
lesk	174	42	the	the	DT
lesk	174	43	costs	cost	NNS
lesk	174	44	to	to	IN
lesk	174	45	the	the	DT
lesk	174	46	banks	bank	NNS
lesk	174	47	.	.	.
lesk	175	1	We	-PRON-	PRP
lesk	175	2	might	may	MD
lesk	175	3	also	also	RB
lesk	175	4	believe	believe	VB
lesk	175	5	that	that	IN
lesk	175	6	the	the	DT
lesk	175	7	safest	safe	JJS
lesk	175	8	possible	possible	JJ
lesk	175	9	railway	railway	NN
lesk	175	10	operation	operation	NN
lesk	175	11	is	be	VBZ
lesk	175	12	important	important	JJ
lesk	175	13	,	,	,
lesk	175	14	even	even	RB
lesk	175	15	if	if	IN
lesk	175	16	the	the	DT
lesk	175	17	automated	automate	VBN
lesk	175	18	train	train	NN
lesk	175	19	does	do	VBZ
lesk	175	20	n’t	not	RB
lesk	175	21	routinely	routinely	RB
lesk	175	22	explain	explain	VB
lesk	175	23	how	how	WRB
lesk	175	24	it	-PRON-	PRP
lesk	175	25	balanced	balance	VBD
lesk	175	26	its	-PRON-	PRP$
lesk	175	27	choices	choice	NNS
lesk	175	28	of	of	IN
lesk	175	29	acceleration	acceleration	NN
lesk	175	30	to	to	TO
lesk	175	31	achieve	achieve	VB
lesk	175	32	high	high	JJ
lesk	175	33	punctuality	punctuality	NN
lesk	175	34	and	and	CC
lesk	175	35	low	low	JJ
lesk	175	36	risk	risk	NN
lesk	175	37	.	.	.
lesk	176	1	What	what	WP
lesk	176	2	would	would	MD
lesk	176	3	I	-PRON-	PRP
lesk	176	4	suggest	suggest	VB
lesk	176	5	?	?	.
lesk	177	1	Organizationally	organizationally	RB
lesk	177	2	:	:	:
lesk	177	3	·	·	NFP
lesk	177	4	Have	have	VBP
lesk	177	5	teams	team	NNS
lesk	177	6	including	include	VBG
lesk	177	7	both	both	CC
lesk	177	8	the	the	DT
lesk	177	9	computer	computer	NN
lesk	177	10	scientists	scientist	NNS
lesk	177	11	and	and	CC
lesk	177	12	the	the	DT
lesk	177	13	users	user	NNS
lesk	177	14	.	.	.
lesk	178	1	·	·	NFP
lesk	178	2	Collaborate	collaborate	VB
lesk	178	3	with	with	IN
lesk	178	4	a	a	DT
lesk	178	5	statistician	statistician	NN
lesk	178	6	:	:	:
lesk	178	7	they	-PRON-	PRP
lesk	178	8	’ve	have	VB
lesk	178	9	seen	see	VBN
lesk	178	10	a	a	DT
lesk	178	11	lot	lot	NN
lesk	178	12	of	of	IN
lesk	178	13	these	these	DT
lesk	178	14	problems	problem	NNS
lesk	178	15	before	before	RB
lesk	178	16	.	.	.
lesk	179	1	·	·	NFP
lesk	179	2	Work	work	VB
lesk	179	3	on	on	IN
lesk	179	4	easier	easy	JJR
lesk	179	5	problems	problem	NNS
lesk	179	6	.	.	.
lesk	180	1	As	as	IN
lesk	180	2	examples	example	NNS
lesk	180	3	,	,	,
lesk	180	4	I	-PRON-	PRP
lesk	180	5	watched	watch	VBD
lesk	180	6	a	a	DT
lesk	180	7	group	group	NN
lesk	180	8	of	of	IN
lesk	180	9	zoologists	zoologists	NNPS
lesk	180	10	with	with	IN
lesk	180	11	a	a	DT
lesk	180	12	group	group	NN
lesk	180	13	of	of	IN
lesk	180	14	computer	computer	NN
lesk	180	15	scientists	scientist	NNS
lesk	180	16	discussing	discuss	VBG
lesk	180	17	how	how	WRB
lesk	180	18	to	to	TO
lesk	180	19	improve	improve	VB
lesk	180	20	accuracy	accuracy	NN
lesk	180	21	at	at	IN
lesk	180	22	identifying	identify	VBG
lesk	180	23	animals	animal	NNS
lesk	180	24	in	in	IN
lesk	180	25	photographs	photograph	NNS
lesk	180	26	.	.	.
lesk	181	1	The	the	DT
lesk	181	2	discussion	discussion	NN
lesk	181	3	indicated	indicate	VBD
lesk	181	4	that	that	IN
lesk	181	5	you	-PRON-	PRP
lesk	181	6	needed	need	VBD
lesk	181	7	hundreds	hundred	NNS
lesk	181	8	of	of	IN
lesk	181	9	training	training	NN
lesk	181	10	examples	example	NNS
lesk	181	11	at	at	IN
lesk	181	12	a	a	DT
lesk	181	13	minimum	minimum	NN
lesk	181	14	,	,	,
lesk	181	15	if	if	IN
lesk	181	16	not	not	RB
lesk	181	17	thousands	thousand	NNS
lesk	181	18	,	,	,
lesk	181	19	since	since	IN
lesk	181	20	the	the	DT
lesk	181	21	animals	animal	NNS
lesk	181	22	do	do	VBP
lesk	181	23	not	not	RB
lesk	181	24	typically	typically	RB
lesk	181	25	walk	walk	VB
lesk	181	26	up	up	IN
lesk	181	27	to	to	IN
lesk	181	28	the	the	DT
lesk	181	29	camera	camera	NN
lesk	181	30	and	and	CC
lesk	181	31	pose	pose	VB
lesk	181	32	for	for	IN
lesk	181	33	a	a	DT
lesk	181	34	full	full	JJ
lesk	181	35	-	-	HYPH
lesk	181	36	frame	frame	NN
lesk	181	37	shot	shot	NN
lesk	181	38	.	.	.
lesk	182	1	It	-PRON-	PRP
lesk	182	2	was	be	VBD
lesk	182	3	important	important	JJ
lesk	182	4	to	to	TO
lesk	182	5	have	have	VB
lesk	182	6	both	both	CC
lesk	182	7	the	the	DT
lesk	182	8	people	people	NNS
lesk	182	9	who	who	WP
lesk	182	10	understood	understand	VBD
lesk	182	11	the	the	DT
lesk	182	12	learning	learning	NN
lesk	182	13	systems	system	NNS
lesk	182	14	and	and	CC
lesk	182	15	the	the	DT
lesk	182	16	people	people	NNS
lesk	182	17	who	who	WP
lesk	182	18	knew	know	VBD
lesk	182	19	what	what	WP
lesk	182	20	the	the	DT
lesk	182	21	pictures	picture	NNS
lesk	182	22	were	be	VBD
lesk	182	23	realistically	realistically	RB
lesk	182	24	like	like	IN
lesk	182	25	.	.	.
lesk	183	1	The	the	DT
lesk	183	2	most	most	RBS
lesk	183	3	amusing	amusing	JJ
lesk	183	4	contribution	contribution	NN
lesk	183	5	by	by	IN
lesk	183	6	a	a	DT
lesk	183	7	statistician	statistician	NN
lesk	183	8	happened	happen	VBD
lesk	183	9	when	when	WRB
lesk	183	10	a	a	DT
lesk	183	11	computer	computer	NN
lesk	183	12	scientist	scientist	NN
lesk	183	13	offered	offer	VBD
lesk	183	14	a	a	DT
lesk	183	15	program	program	NN
lesk	183	16	that	that	WDT
lesk	183	17	tried	try	VBD
lesk	183	18	to	to	TO
lesk	183	19	recognize	recognize	VB
lesk	183	20	individual	individual	JJ
lesk	183	21	giraffes	giraffe	NNS
lesk	183	22	,	,	,
lesk	183	23	and	and	CC
lesk	183	24	a	a	DT
lesk	183	25	zoologist	zoologist	NN
lesk	183	26	complained	complain	VBD
lesk	183	27	that	that	IN
lesk	183	28	it	-PRON-	PRP
lesk	183	29	only	only	RB
lesk	183	30	worked	work	VBD
lesk	183	31	if	if	IN
lesk	183	32	you	-PRON-	PRP
lesk	183	33	had	have	VBD
lesk	183	34	a	a	DT
lesk	183	35	view	view	NN
lesk	183	36	of	of	IN
lesk	183	37	the	the	DT
lesk	183	38	right	right	JJ
lesk	183	39	-	-	HYPH
lesk	183	40	hand	hand	NN
lesk	183	41	side	side	NN
lesk	183	42	of	of	IN
lesk	183	43	the	the	DT
lesk	183	44	giraffe	giraffe	NN
lesk	183	45	.	.	.
lesk	184	1	Somebody	somebody	NN
lesk	184	2	who	who	WP
lesk	184	3	knew	know	VBD
lesk	184	4	statistics	statistic	NNS
lesk	184	5	perked	perk	VBD
lesk	184	6	up	up	RP
lesk	184	7	and	and	CC
lesk	184	8	said	say	VBD
lesk	184	9	“	"	``
lesk	184	10	it	-PRON-	PRP
lesk	184	11	’s	’	VBZ
lesk	184	12	a	a	DT
lesk	184	13	50	50	CD
lesk	184	14	%	%	NN
lesk	184	15	chance	chance	NN
lesk	184	16	of	of	IN
lesk	184	17	recognizing	recognize	VBG
lesk	184	18	the	the	DT
lesk	184	19	animal	animal	NN
lesk	184	20	?	?	.
lesk	185	1	I	-PRON-	PRP
lesk	185	2	can	can	MD
lesk	185	3	do	do	VB
lesk	185	4	the	the	DT
lesk	185	5	math	math	NN
lesk	185	6	for	for	IN
lesk	185	7	that	that	DT
lesk	185	8	.	.	.
lesk	185	9	”	"	''
lesk	185	10	And	and	CC
lesk	185	11	it	-PRON-	PRP
lesk	185	12	is	be	VBZ
lesk	185	13	simpler	simple	JJR
lesk	185	14	to	to	TO
lesk	185	15	do	do	VB
lesk	185	16	“	"	``
lesk	185	17	is	be	VBZ
lesk	185	18	there	there	EX
lesk	185	19	any	any	DT
lesk	185	20	animal	animal	NN
lesk	185	21	in	in	IN
lesk	185	22	the	the	DT
lesk	185	23	picture	picture	NN
lesk	185	24	?	?	.
lesk	185	25	”	"	''
lesk	185	26	before	before	IN
lesk	185	27	asking	ask	VBG
lesk	185	28	“	"	``
lesk	185	29	which	which	WDT
lesk	185	30	animal	animal	NN
lesk	185	31	is	be	VBZ
lesk	185	32	it	-PRON-	PRP
lesk	185	33	?	?	.
lesk	185	34	”	"	''
lesk	185	35	and	and	CC
lesk	185	36	create	create	VB
lesk	185	37	two	two	CD
lesk	185	38	easier	easy	JJR
lesk	185	39	problems	problem	NNS
lesk	185	40	.	.	.
lesk	186	1	Technically	technically	RB
lesk	186	2	:	:	:
lesk	186	3	·	·	NFP
lesk	186	4	Try	try	VB
lesk	186	5	to	to	TO
lesk	186	6	interpolate	interpolate	VB
lesk	186	7	rather	rather	RB
lesk	186	8	than	than	IN
lesk	186	9	extrapolate	extrapolate	VB
lesk	186	10	:	:	:
lesk	186	11	use	use	VB
lesk	186	12	the	the	DT
lesk	186	13	algorithm	algorithm	NN
lesk	186	14	on	on	IN
lesk	186	15	points	point	NNS
lesk	186	16	“	"	``
lesk	186	17	inside	inside	RB
lesk	186	18	”	"	''
lesk	186	19	the	the	DT
lesk	186	20	training	training	NN
lesk	186	21	set	set	NN
lesk	186	22	(	(	-LRB-
lesk	186	23	thinking	think	VBG
lesk	186	24	in	in	IN
lesk	186	25	multiple	multiple	JJ
lesk	186	26	dimensions	dimension	NNS
lesk	186	27	)	)	-RRB-
lesk	186	28	.	.	.
lesk	187	1	·	·	NFP
lesk	187	2	Lean	lean	VB
lesk	187	3	towards	towards	IN
lesk	187	4	feature	feature	NN
lesk	187	5	detection	detection	NN
lesk	187	6	and	and	CC
lesk	187	7	modeling	model	VBG
lesk	187	8	rather	rather	RB
lesk	187	9	than	than	IN
lesk	187	10	completely	completely	RB
lesk	187	11	unsupervised	unsupervise	VBN
lesk	187	12	learning	learning	NN
lesk	187	13	.	.	.
lesk	188	1	·	·	NFP
lesk	188	2	Emphasize	emphasize	VB
lesk	188	3	continuous	continuous	JJ
lesk	188	4	rather	rather	RB
lesk	188	5	than	than	IN
lesk	188	6	discrete	discrete	JJ
lesk	188	7	variables	variable	NNS
lesk	188	8	.	.	.
lesk	189	1	I	-PRON-	PRP
lesk	189	2	suggest	suggest	VBP
lesk	189	3	using	use	VBG
lesk	189	4	methods	method	NNS
lesk	189	5	that	that	WDT
lesk	189	6	involve	involve	VBP
lesk	189	7	feature	feature	NN
lesk	189	8	detection	detection	NN
lesk	189	9	,	,	,
lesk	189	10	since	since	IN
lesk	189	11	that	that	DT
lesk	189	12	tells	tell	VBZ
lesk	189	13	you	-PRON-	PRP
lesk	189	14	what	what	WP
lesk	189	15	the	the	DT
lesk	189	16	algorithm	algorithm	NNP
lesk	189	17	is	be	VBZ
lesk	189	18	relying	rely	VBG
lesk	189	19	on	on	IN
lesk	189	20	.	.	.
lesk	190	1	For	for	IN
lesk	190	2	example	example	NN
lesk	190	3	,	,	,
lesk	190	4	consider	consider	VB
lesk	190	5	the	the	DT
lesk	190	6	Google	Google	NNP
lesk	190	7	Flu	Flu	NNP
lesk	190	8	Trends	Trends	NNP
lesk	190	9	failure	failure	NN
lesk	190	10	;	;	:
lesk	190	11	the	the	DT
lesk	190	12	public	public	NN
lesk	190	13	was	be	VBD
lesk	190	14	not	not	RB
lesk	190	15	told	tell	VBN
lesk	190	16	what	what	WP
lesk	190	17	terms	term	NNS
lesk	190	18	were	be	VBD
lesk	190	19	used	use	VBN
lesk	190	20	.	.	.
lesk	191	1	As	as	IN
lesk	191	2	David	David	NNP
lesk	191	3	Lazer	Lazer	NNP
lesk	191	4	noted	note	VBD
lesk	191	5	,	,	,
lesk	191	6	some	some	DT
lesk	191	7	of	of	IN
lesk	191	8	them	-PRON-	PRP
lesk	191	9	were	be	VBD
lesk	191	10	just	just	RB
lesk	191	11	“	"	``
lesk	191	12	winter	winter	NN
lesk	191	13	”	"	''
lesk	191	14	terms	term	NNS
lesk	191	15	(	(	-LRB-
lesk	191	16	like	like	UH
lesk	191	17	‘	'	``
lesk	191	18	basketball	basketball	NN
lesk	191	19	’	'	''
lesk	191	20	)	)	-RRB-
lesk	191	21	.	.	.
lesk	192	1	If	if	IN
lesk	192	2	you	-PRON-	PRP
lesk	192	3	know	know	VBP
lesk	192	4	that	that	IN
lesk	192	5	,	,	,
lesk	192	6	you	-PRON-	PRP
lesk	192	7	might	may	MD
lesk	192	8	be	be	VB
lesk	192	9	skeptical	skeptical	JJ
lesk	192	10	.	.	.
lesk	193	1	More	more	RBR
lesk	193	2	significant	significant	JJ
lesk	193	3	are	be	VBP
lesk	193	4	decisions	decision	NNS
lesk	193	5	like	like	IN
lesk	193	6	jail	jail	NN
lesk	193	7	sentences	sentence	NNS
lesk	193	8	or	or	CC
lesk	193	9	college	college	NN
lesk	193	10	admissions	admission	NNS
lesk	193	11	;	;	,
lesk	193	12	knowing	know	VBG
lesk	193	13	that	that	DT
lesk	193	14	racial	racial	JJ
lesk	193	15	or	or	CC
lesk	193	16	religious	religious	JJ
lesk	193	17	discrimination	discrimination	NN
lesk	193	18	are	be	VBP
lesk	193	19	not	not	RB
lesk	193	20	relevant	relevant	JJ
lesk	193	21	can	can	MD
lesk	193	22	be	be	VB
lesk	193	23	verified	verify	VBN
lesk	193	24	by	by	IN
lesk	193	25	knowing	know	VBG
lesk	193	26	that	that	IN
lesk	193	27	the	the	DT
lesk	193	28	program	program	NN
lesk	193	29	did	do	VBD
lesk	193	30	not	not	RB
lesk	193	31	use	use	VB
lesk	193	32	them	-PRON-	PRP
lesk	193	33	.	.	.
lesk	194	1	Knowing	know	VBG
lesk	194	2	what	what	WP
lesk	194	3	features	feature	NNS
lesk	194	4	were	be	VBD
lesk	194	5	used	use	VBN
lesk	194	6	can	can	MD
lesk	194	7	sometimes	sometimes	RB
lesk	194	8	help	help	VB
lesk	194	9	the	the	DT
lesk	194	10	user	user	NN
lesk	194	11	:	:	:
lesk	194	12	if	if	IN
lesk	194	13	you	-PRON-	PRP
lesk	194	14	know	know	VBP
lesk	194	15	that	that	IN
lesk	194	16	your	-PRON-	PRP$
lesk	194	17	loan	loan	NN
lesk	194	18	application	application	NN
lesk	194	19	was	be	VBD
lesk	194	20	downrated	downrate	VBN
lesk	194	21	because	because	IN
lesk	194	22	of	of	IN
lesk	194	23	your	-PRON-	PRP$
lesk	194	24	credit	credit	NN
lesk	194	25	score	score	NN
lesk	194	26	,	,	,
lesk	194	27	it	-PRON-	PRP
lesk	194	28	may	may	MD
lesk	194	29	be	be	VB
lesk	194	30	possible	possible	JJ
lesk	194	31	for	for	IN
lesk	194	32	you	-PRON-	PRP
lesk	194	33	to	to	TO
lesk	194	34	pay	pay	VB
lesk	194	35	off	off	RP
lesk	194	36	some	some	DT
lesk	194	37	bill	bill	NN
lesk	194	38	to	to	TO
lesk	194	39	raise	raise	VB
lesk	194	40	the	the	DT
lesk	194	41	score	score	NN
lesk	194	42	.	.	.
lesk	195	1	Sometimes	sometimes	RB
lesk	195	2	you	-PRON-	PRP
lesk	195	3	have	have	VBP
lesk	195	4	to	to	TO
lesk	195	5	use	use	VB
lesk	195	6	categorical	categorical	JJ
lesk	195	7	variables	variable	NNS
lesk	195	8	(	(	-LRB-
lesk	195	9	what	what	WP
lesk	195	10	county	county	NN
lesk	195	11	do	do	VBP
lesk	195	12	you	-PRON-	PRP
lesk	195	13	live	live	VB
lesk	195	14	in	in	IN
lesk	195	15	?	?	.
lesk	195	16	)	)	-RRB-
lesk	196	1	but	but	CC
lesk	196	2	if	if	IN
lesk	196	3	you	-PRON-	PRP
lesk	196	4	have	have	VBP
lesk	196	5	a	a	DT
lesk	196	6	choice	choice	NN
lesk	196	7	of	of	IN
lesk	196	8	how	how	WRB
lesk	196	9	you	-PRON-	PRP
lesk	196	10	phrase	phrase	VBP
lesk	196	11	a	a	DT
lesk	196	12	variable	variable	NN
lesk	196	13	,	,	,
lesk	196	14	asking	ask	VBG
lesk	196	15	something	something	NN
lesk	196	16	like	like	IN
lesk	196	17	“	"	``
lesk	196	18	how	how	WRB
lesk	196	19	many	many	JJ
lesk	196	20	minutes	minute	NNS
lesk	196	21	a	a	DT
lesk	196	22	day	day	NN
lesk	196	23	do	do	VBP
lesk	196	24	you	-PRON-	PRP
lesk	196	25	spend	spend	VB
lesk	196	26	reading	read	VBG
lesk	196	27	?	?	.
lesk	196	28	”	"	''
lesk	196	29	is	be	VBZ
lesk	196	30	likely	likely	JJ
lesk	196	31	to	to	TO
lesk	196	32	produce	produce	VB
lesk	196	33	a	a	DT
lesk	196	34	better	well	JJR
lesk	196	35	fit	fit	NN
lesk	196	36	than	than	IN
lesk	196	37	asking	ask	VBG
lesk	196	38	people	people	NNS
lesk	196	39	to	to	TO
lesk	196	40	choose	choose	VB
lesk	196	41	“	"	``
lesk	196	42	how	how	WRB
lesk	196	43	much	much	JJ
lesk	196	44	do	do	VBP
lesk	196	45	you	-PRON-	PRP
lesk	196	46	read	read	VB
lesk	196	47	:	:	:
lesk	196	48	never	never	RB
lesk	196	49	,	,	,
lesk	196	50	sometimes	sometimes	RB
lesk	196	51	,	,	,
lesk	196	52	a	a	DT
lesk	196	53	lot	lot	NN
lesk	196	54	?	?	.
lesk	196	55	”	"	''
lesk	196	56	A	a	DT
lesk	196	57	machine	machine	NN
lesk	196	58	learning	learn	VBG
lesk	196	59	algorithm	algorithm	NN
lesk	196	60	may	may	MD
lesk	196	61	tell	tell	VB
lesk	196	62	you	-PRON-	PRP
lesk	196	63	how	how	WRB
lesk	196	64	much	much	JJ
lesk	196	65	of	of	IN
lesk	196	66	the	the	DT
lesk	196	67	variance	variance	NN
lesk	196	68	each	each	DT
lesk	196	69	input	input	NN
lesk	196	70	variable	variable	NN
lesk	196	71	explains	explain	VBZ
lesk	196	72	;	;	:
lesk	196	73	you	-PRON-	PRP
lesk	196	74	can	can	MD
lesk	196	75	use	use	VB
lesk	196	76	that	that	DT
lesk	196	77	information	information	NN
lesk	196	78	to	to	TO
lesk	196	79	focus	focus	VB
lesk	196	80	on	on	IN
lesk	196	81	the	the	DT
lesk	196	82	variables	variable	NNS
lesk	196	83	that	that	WDT
lesk	196	84	are	be	VBP
lesk	196	85	most	most	RBS
lesk	196	86	important	important	JJ
lesk	196	87	to	to	IN
lesk	196	88	your	-PRON-	PRP$
lesk	196	89	problem	problem	NN
lesk	196	90	,	,	,
lesk	196	91	and	and	CC
lesk	196	92	decide	decide	VB
lesk	196	93	whether	whether	IN
lesk	196	94	you	-PRON-	PRP
lesk	196	95	think	think	VBP
lesk	196	96	you	-PRON-	PRP
lesk	196	97	are	be	VBP
lesk	196	98	measuring	measure	VBG
lesk	196	99	them	-PRON-	PRP
lesk	196	100	well	well	RB
lesk	196	101	enough	enough	RB
lesk	196	102	.	.	.
lesk	197	1	Why	why	WRB
lesk	197	2	not	not	RB
lesk	197	3	extrapolate	extrapolate	JJ
lesk	197	4	?	?	.
lesk	198	1	Sadly	sadly	RB
lesk	198	2	,	,	,
lesk	198	3	as	as	IN
lesk	198	4	I	-PRON-	PRP
lesk	198	5	write	write	VBP
lesk	198	6	this	this	DT
lesk	198	7	in	in	IN
lesk	198	8	early	early	JJ
lesk	198	9	April	April	NNP
lesk	198	10	2020	2020	CD
lesk	198	11	,	,	,
lesk	198	12	we	-PRON-	PRP
lesk	198	13	are	be	VBP
lesk	198	14	seeing	see	VBG
lesk	198	15	all	all	DT
lesk	198	16	sorts	sort	NNS
lesk	198	17	of	of	IN
lesk	198	18	extrapolations	extrapolation	NNS
lesk	198	19	of	of	IN
lesk	198	20	the	the	DT
lesk	198	21	COVID-19	covid-19	NN
lesk	198	22	epidemic	epidemic	NN
lesk	198	23	,	,	,
lesk	198	24	with	with	IN
lesk	198	25	expected	expect	VBN
lesk	198	26	US	US	NNP
lesk	198	27	deaths	death	NNS
lesk	198	28	ranging	range	VBG
lesk	198	29	from	from	IN
lesk	198	30	30,000	30,000	CD
lesk	198	31	to	to	IN
lesk	198	32	2	2	CD
lesk	198	33	million	million	CD
lesk	198	34	,	,	,
lesk	198	35	as	as	IN
lesk	198	36	people	people	NNS
lesk	198	37	try	try	VBP
lesk	198	38	to	to	TO
lesk	198	39	fit	fit	VB
lesk	198	40	various	various	JJ
lesk	198	41	functions	function	NNS
lesk	198	42	(	(	-LRB-
lesk	198	43	Gaussians	Gaussians	NNP
lesk	198	44	,	,	,
lesk	198	45	logistic	logistic	JJ
lesk	198	46	regression	regression	NN
lesk	198	47	,	,	,
lesk	198	48	or	or	CC
lesk	198	49	whatever	whatever	WDT
lesk	198	50	)	)	-RRB-
lesk	198	51	with	with	IN
lesk	198	52	inadequately	inadequately	RB
lesk	198	53	precise	precise	JJ
lesk	198	54	data	datum	NNS
lesk	198	55	and	and	CC
lesk	198	56	uncertain	uncertain	JJ
lesk	198	57	models	model	NNS
lesk	198	58	.	.	.
lesk	199	1	A	a	DT
lesk	199	2	simpler	simple	JJR
lesk	199	3	example	example	NN
lesk	199	4	is	be	VBZ
lesk	199	5	Mark	Mark	NNP
lesk	199	6	Twain	Twain	NNP
lesk	199	7	’s	’s	''
lesk	199	8	:	:	:
lesk	199	9	“	"	``
lesk	199	10	In	in	IN
lesk	199	11	the	the	DT
lesk	199	12	space	space	NN
lesk	199	13	of	of	IN
lesk	199	14	one	one	CD
lesk	199	15	hundred	hundred	CD
lesk	199	16	and	and	CC
lesk	199	17	seventy	seventy	CD
lesk	199	18	-	-	HYPH
lesk	199	19	six	six	CD
lesk	199	20	years	year	NNS
lesk	199	21	the	the	DT
lesk	199	22	Lower	low	JJR
lesk	199	23	Mississippi	Mississippi	NNP
lesk	199	24	has	have	VBZ
lesk	199	25	shortened	shorten	VBN
lesk	199	26	itself	-PRON-	PRP
lesk	199	27	two	two	CD
lesk	199	28	hundred	hundred	CD
lesk	199	29	and	and	CC
lesk	199	30	forty	forty	CD
lesk	199	31	-	-	HYPH
lesk	199	32	two	two	CD
lesk	199	33	miles	mile	NNS
lesk	199	34	.	.	.
lesk	200	1	That	that	DT
lesk	200	2	is	be	VBZ
lesk	200	3	an	an	DT
lesk	200	4	average	average	NN
lesk	200	5	of	of	IN
lesk	200	6	a	a	DT
lesk	200	7	trifle	trifle	NN
lesk	200	8	over	over	IN
lesk	200	9	one	one	CD
lesk	200	10	mile	mile	NN
lesk	200	11	and	and	CC
lesk	200	12	a	a	DT
lesk	200	13	third	third	NN
lesk	200	14	per	per	IN
lesk	200	15	year	year	NN
lesk	200	16	.	.	.
lesk	201	1	Therefore	therefore	RB
lesk	201	2	,	,	,
lesk	201	3	any	any	DT
lesk	201	4	calm	calm	JJ
lesk	201	5	person	person	NN
lesk	201	6	,	,	,
lesk	201	7	who	who	WP
lesk	201	8	is	be	VBZ
lesk	201	9	not	not	RB
lesk	201	10	blind	blind	JJ
lesk	201	11	or	or	CC
lesk	201	12	idiotic	idiotic	JJ
lesk	201	13	,	,	,
lesk	201	14	can	can	MD
lesk	201	15	see	see	VB
lesk	201	16	that	that	IN
lesk	201	17	in	in	IN
lesk	201	18	the	the	DT
lesk	201	19	‘	'	``
lesk	201	20	Old	Old	NNP
lesk	201	21	Oolitic	oolitic	JJ
lesk	201	22	Silurian	silurian	JJ
lesk	201	23	Period	Period	NNP
lesk	201	24	,	,	,
lesk	201	25	’	'	''
lesk	201	26	just	just	RB
lesk	201	27	a	a	DT
lesk	201	28	million	million	CD
lesk	201	29	years	year	NNS
lesk	201	30	ago	ago	RB
lesk	201	31	next	next	JJ
lesk	201	32	November	November	NNP
lesk	201	33	,	,	,
lesk	201	34	the	the	DT
lesk	201	35	Lower	Lower	NNP
lesk	201	36	Mississippi	Mississippi	NNP
lesk	201	37	River	River	NNP
lesk	201	38	was	be	VBD
lesk	201	39	upwards	upwards	JJ
lesk	201	40	of	of	IN
lesk	201	41	one	one	CD
lesk	201	42	million	million	CD
lesk	201	43	three	three	CD
lesk	201	44	hundred	hundred	CD
lesk	201	45	thousand	thousand	CD
lesk	201	46	miles	mile	NNS
lesk	201	47	long	long	JJ
lesk	201	48	,	,	,
lesk	201	49	and	and	CC
lesk	201	50	stuck	stick	VBD
lesk	201	51	out	out	RB
lesk	201	52	over	over	IN
lesk	201	53	the	the	DT
lesk	201	54	Gulf	Gulf	NNP
lesk	201	55	of	of	IN
lesk	201	56	Mexico	Mexico	NNP
lesk	201	57	like	like	IN
lesk	201	58	a	a	DT
lesk	201	59	fishing	fishing	NN
lesk	201	60	-	-	HYPH
lesk	201	61	rod	rod	NN
lesk	201	62	.	.	.
lesk	202	1	And	and	CC
lesk	202	2	by	by	IN
lesk	202	3	the	the	DT
lesk	202	4	same	same	JJ
lesk	202	5	token	token	NN
lesk	202	6	any	any	DT
lesk	202	7	person	person	NN
lesk	202	8	can	can	MD
lesk	202	9	see	see	VB
lesk	202	10	that	that	IN
lesk	202	11	seven	seven	CD
lesk	202	12	hundred	hundred	CD
lesk	202	13	and	and	CC
lesk	202	14	forty	forty	CD
lesk	202	15	-	-	HYPH
lesk	202	16	two	two	CD
lesk	202	17	years	year	NNS
lesk	202	18	from	from	IN
lesk	202	19	now	now	RB
lesk	202	20	the	the	DT
lesk	202	21	Lower	Lower	NNP
lesk	202	22	Mississippi	Mississippi	NNP
lesk	202	23	will	will	MD
lesk	202	24	be	be	VB
lesk	202	25	only	only	RB
lesk	202	26	a	a	DT
lesk	202	27	mile	mile	NN
lesk	202	28	and	and	CC
lesk	202	29	three	three	CD
lesk	202	30	-	-	HYPH
lesk	202	31	quarters	quarter	NNS
lesk	202	32	long	long	JJ
lesk	202	33	,	,	,
lesk	202	34	and	and	CC
lesk	202	35	Cairo	Cairo	NNP
lesk	202	36	and	and	CC
lesk	202	37	New	New	NNP
lesk	202	38	Orleans	Orleans	NNP
lesk	202	39	will	will	MD
lesk	202	40	have	have	VB
lesk	202	41	joined	join	VBN
lesk	202	42	their	-PRON-	PRP$
lesk	202	43	streets	street	NNS
lesk	202	44	together	together	RB
lesk	202	45	,	,	,
lesk	202	46	and	and	CC
lesk	202	47	be	be	VB
lesk	202	48	plodding	plod	VBG
lesk	202	49	comfortably	comfortably	RB
lesk	202	50	along	along	IN
lesk	202	51	under	under	IN
lesk	202	52	a	a	DT
lesk	202	53	single	single	JJ
lesk	202	54	mayor	mayor	NN
lesk	202	55	and	and	CC
lesk	202	56	a	a	DT
lesk	202	57	mutual	mutual	JJ
lesk	202	58	board	board	NN
lesk	202	59	of	of	IN
lesk	202	60	aldermen	aldermen	NNP
lesk	202	61	”	"	''
lesk	202	62	(	(	-LRB-
lesk	202	63	1883	1883	CD
lesk	202	64	)	)	-RRB-
lesk	202	65	.	.	.
lesk	203	1	Finally	finally	RB
lesk	203	2	,	,	,
lesk	203	3	note	note	VB
lesk	203	4	the	the	DT
lesk	203	5	advice	advice	NN
lesk	203	6	of	of	IN
lesk	203	7	Edgar	Edgar	NNP
lesk	203	8	Allan	Allan	NNP
lesk	203	9	Poe	Poe	NNP
lesk	203	10	:	:	:
lesk	203	11	“	"	``
lesk	203	12	Believe	believe	VB
lesk	203	13	nothing	nothing	NN
lesk	203	14	you	-PRON-	PRP
lesk	203	15	hear	hear	VBP
lesk	203	16	,	,	,
lesk	203	17	and	and	CC
lesk	203	18	only	only	RB
lesk	203	19	one	one	CD
lesk	203	20	half	half	NN
lesk	203	21	that	that	WDT
lesk	203	22	you	-PRON-	PRP
lesk	203	23	see	see	VBP
lesk	203	24	.	.	.
lesk	203	25	”	"	''
lesk	203	26	References	reference	NNS
lesk	203	27	“	"	``
lesk	203	28	AI	AI	NNP
lesk	203	29	Recognition	recognition	NN
lesk	203	30	Fooled	fool	VBN
lesk	203	31	by	by	IN
lesk	203	32	Single	Single	NNP
lesk	203	33	Pixel	Pixel	NNP
lesk	203	34	Change	Change	NNP
lesk	203	35	.	.	.
lesk	203	36	”	"	''
lesk	203	37	BBC	BBC	NNP
lesk	203	38	News	News	NNP
lesk	203	39	,	,	,
lesk	203	40	November	November	NNP
lesk	203	41	3	3	CD
lesk	203	42	,	,	,
lesk	203	43	2017	2017	CD
lesk	203	44	.	.	.
lesk	203	45	https://www.bbc.com/news/technology-41845878	https://www.bbc.com/news/technology-41845878	NNP
lesk	203	46	.	.	.
lesk	204	1	“	"	``
lesk	204	2	Asilomar	Asilomar	NNP
lesk	204	3	AI	AI	NNP
lesk	204	4	Principles	Principles	NNPS
lesk	204	5	.	.	.
lesk	204	6	”	"	''
lesk	204	7	2017	2017	CD
lesk	204	8	.	.	.
lesk	204	9	https://futureoflife.org/ai-principles/.	https://futureoflife.org/ai-principles/.	ADD
lesk	205	1	Bojarski	Bojarski	NNP
lesk	205	2	,	,	,
lesk	205	3	Mariusz	Mariusz	NNP
lesk	205	4	,	,	,
lesk	205	5	Larry	Larry	NNP
lesk	205	6	Jackel	Jackel	NNP
lesk	205	7	,	,	,
lesk	205	8	Ben	Ben	NNP
lesk	205	9	Firner	Firner	NNP
lesk	205	10	,	,	,
lesk	205	11	and	and	CC
lesk	205	12	Urs	Urs	NNP
lesk	205	13	Muller	Muller	NNP
lesk	205	14	.	.	.
lesk	206	1	2017	2017	CD
lesk	206	2	.	.	.
lesk	207	1	“	"	``
lesk	207	2	Explaining	explain	VBG
lesk	207	3	How	how	WRB
lesk	207	4	End	end	NN
lesk	207	5	-	-	HYPH
lesk	207	6	to	to	IN
lesk	207	7	-	-	HYPH
lesk	207	8	End	end	NN
lesk	207	9	Deep	Deep	NNP
lesk	207	10	Learning	learning	NN
lesk	207	11	Steers	steer	VBZ
lesk	207	12	a	a	DT
lesk	207	13	Self	self	NN
lesk	207	14	-	-	HYPH
lesk	207	15	Driving	drive	VBG
lesk	207	16	Car	car	NN
lesk	207	17	.	.	.
lesk	207	18	”	"	''
lesk	207	19	NVIDIA	NVIDIA	NNP
lesk	207	20	Developer	Developer	NNP
lesk	207	21	Blog	Blog	NNP
lesk	207	22	.	.	.
lesk	208	1	https://devblogs.nvidia.com/explaining-deep-learning-self-driving-car/.	https://devblogs.nvidia.com/explaining-deep-learning-self-driving-car/.	PRP
lesk	209	1	Bornstein	Bornstein	NNP
lesk	209	2	,	,	,
lesk	209	3	Aaron	Aaron	NNP
lesk	209	4	.	.	.
lesk	210	1	2016	2016	CD
lesk	210	2	.	.	.
lesk	211	1	“	"	``
lesk	211	2	Is	be	VBZ
lesk	211	3	Artificial	Artificial	NNP
lesk	211	4	Intelligence	intelligence	NN
lesk	211	5	Permanently	permanently	RB
lesk	211	6	Inscrutable	inscrutable	JJ
lesk	211	7	?	?	.
lesk	211	8	”	"	''
lesk	211	9	Nautilus	nautilus	JJ
lesk	211	10	40	40	CD
lesk	211	11	(	(	-LRB-
lesk	211	12	1	1	CD
lesk	211	13	)	)	-RRB-
lesk	211	14	.	.	.
lesk	212	1	http://nautil.us/issue/40/learning/is-artificial-intelligence-permanently-inscrutable	http://nautil.us/issue/40/learning/is-artificial-intelligence-permanently-inscrutable	JJ
lesk	212	2	.	.	.
lesk	213	1	Brownlee	Brownlee	NNP
lesk	213	2	,	,	,
lesk	213	3	Jason	Jason	NNP
lesk	213	4	.	.	.
lesk	214	1	2018	2018	CD
lesk	214	2	“	"	``
lesk	214	3	The	the	DT
lesk	214	4	Model	Model	NNP
lesk	214	5	Performance	Performance	NNP
lesk	214	6	Mismatch	Mismatch	NNP
lesk	214	7	Problem	Problem	NNP
lesk	214	8	(	(	-LRB-
lesk	214	9	and	and	CC
lesk	214	10	What	what	WP
lesk	214	11	to	to	TO
lesk	214	12	Do	do	VB
lesk	214	13	about	about	IN
lesk	214	14	It	-PRON-	PRP
lesk	214	15	)	)	-RRB-
lesk	214	16	.	.	.
lesk	214	17	”	"	''
lesk	214	18	Machine	Machine	NNP
lesk	214	19	Learning	Learning	NNP
lesk	214	20	Mastery	Mastery	NNP
lesk	214	21	.	.	.
lesk	215	1	https://machinelearningmastery.com/the-model-performance-mismatch-problem/.	https://machinelearningmastery.com/the-model-performance-mismatch-problem/.	NNP
lesk	216	1	Burnside	Burnside	NNP
lesk	216	2	,	,	,
lesk	216	3	Elizabeth	Elizabeth	NNP
lesk	216	4	S.	S.	NNP
lesk	216	5	,	,	,
lesk	216	6	Jessie	Jessie	NNP
lesk	216	7	Davis	Davis	NNP
lesk	216	8	,	,	,
lesk	216	9	Jagpreet	Jagpreet	NNP
lesk	216	10	Chhatwal	Chhatwal	NNP
lesk	216	11	,	,	,
lesk	216	12	Oguzhan	Oguzhan	NNP
lesk	216	13	Alagoz	Alagoz	NNP
lesk	216	14	,	,	,
lesk	216	15	Mary	Mary	NNP
lesk	216	16	J.	J.	NNP
lesk	216	17	Lindstrom	Lindstrom	NNP
lesk	216	18	,	,	,
lesk	216	19	Berta	Berta	NNP
lesk	216	20	M.	M.	NNP
lesk	216	21	Geller	Geller	NNP
lesk	216	22	,	,	,
lesk	216	23	Benjamin	Benjamin	NNP
lesk	216	24	Littenberg	Littenberg	NNP
lesk	216	25	,	,	,
lesk	216	26	Katherine	Katherine	NNP
lesk	216	27	A.	A.	NNP
lesk	216	28	Shaffer	Shaffer	NNP
lesk	216	29	,	,	,
lesk	216	30	Charles	Charles	NNP
lesk	216	31	E.	E.	NNP
lesk	216	32	Kahn	Kahn	NNP
lesk	216	33	,	,	,
lesk	216	34	and	and	CC
lesk	216	35	C.	C.	NNP
lesk	216	36	David	David	NNP
lesk	216	37	Page	Page	NNP
lesk	216	38	.	.	.
lesk	217	1	2009	2009	CD
lesk	217	2	.	.	.
lesk	218	1	“	"	``
lesk	218	2	Probabilistic	Probabilistic	NNP
lesk	218	3	Computer	Computer	NNP
lesk	218	4	Model	Model	NNP
lesk	218	5	Developed	develop	VBN
lesk	218	6	from	from	IN
lesk	218	7	Clinical	Clinical	NNP
lesk	218	8	Data	Data	NNP
lesk	218	9	in	in	IN
lesk	218	10	National	National	NNP
lesk	218	11	Mammography	Mammography	NNP
lesk	218	12	Database	Database	NNP
lesk	218	13	Format	Format	NNP
lesk	218	14	to	to	TO
lesk	218	15	Classify	classify	VB
lesk	218	16	Mammographic	Mammographic	NNP
lesk	218	17	Findings	finding	NNS
lesk	218	18	.	.	.
lesk	218	19	”	"	''
lesk	218	20	Radiology	radiology	NN
lesk	218	21	251	251	CD
lesk	218	22	(	(	-LRB-
lesk	218	23	3	3	CD
lesk	218	24	)	)	-RRB-
lesk	218	25	:	:	:
lesk	218	26	663	663	CD
lesk	218	27	-	-	SYM
lesk	218	28	72	72	CD
lesk	218	29	.	.	.
lesk	219	1	Caruana	Caruana	NNP
lesk	219	2	,	,	,
lesk	219	3	Rich	Rich	NNP
lesk	219	4	,	,	,
lesk	219	5	Yin	Yin	NNP
lesk	219	6	Lou	Lou	NNP
lesk	219	7	,	,	,
lesk	219	8	Johannes	Johannes	NNP
lesk	219	9	Gehrke	Gehrke	NNP
lesk	219	10	,	,	,
lesk	219	11	Paul	Paul	NNP
lesk	219	12	Koch	Koch	NNP
lesk	219	13	,	,	,
lesk	219	14	Marc	Marc	NNP
lesk	219	15	Sturm	Sturm	NNP
lesk	219	16	,	,	,
lesk	219	17	and	and	CC
lesk	219	18	Noemie	Noemie	NNP
lesk	219	19	Elhadad	Elhadad	NNP
lesk	219	20	.	.	.
lesk	220	1	2015	2015	CD
lesk	220	2	.	.	.
lesk	221	1	“	"	``
lesk	221	2	Intelligible	Intelligible	NNP
lesk	221	3	Models	Models	NNP
lesk	221	4	for	for	IN
lesk	221	5	HealthCare	HealthCare	NNP
lesk	221	6	:	:	:
lesk	221	7	Predicting	Predicting	NNP
lesk	221	8	Pneumonia	Pneumonia	NNP
lesk	221	9	Risk	Risk	NNP
lesk	221	10	and	and	CC
lesk	221	11	Hospital	Hospital	NNP
lesk	221	12	30-day	30-day	CD
lesk	221	13	Readmission	Readmission	NNP
lesk	221	14	.	.	.
lesk	221	15	”	"	''
lesk	221	16	In	in	IN
lesk	221	17	Proceedings	Proceedings	NNP
lesk	221	18	of	of	IN
lesk	221	19	the	the	DT
lesk	221	20	21th	21th	JJ
lesk	221	21	ACM	ACM	NNP
lesk	221	22	SIGKDD	SIGKDD	NNP
lesk	221	23	International	International	NNP
lesk	221	24	Conference	Conference	NNP
lesk	221	25	on	on	IN
lesk	221	26	Knowledge	Knowledge	NNP
lesk	221	27	Discovery	Discovery	NNP
lesk	221	28	and	and	CC
lesk	221	29	Data	Data	NNPS
lesk	221	30	Mining	Mining	NNP
lesk	221	31	(	(	-LRB-
lesk	221	32	KDD	KDD	NNP
lesk	221	33	'	'	POS
lesk	221	34	15	15	CD
lesk	221	35	)	)	-RRB-
lesk	221	36	,	,	,
lesk	221	37	1721	1721	CD
lesk	221	38	-	-	SYM
lesk	221	39	30	30	CD
lesk	221	40	.	.	.
lesk	222	1	New	New	NNP
lesk	222	2	York	York	NNP
lesk	222	3	:	:	:
lesk	222	4	ACM	ACM	NNP
lesk	222	5	Press	Press	NNP
lesk	222	6	.	.	.
lesk	223	1	https://doi.org/10.1145/2783258.2788613	https://doi.org/10.1145/2783258.2788613	ADD
lesk	223	2	.	.	.
lesk	224	1	Christensen	Christensen	NNP
lesk	224	2	,	,	,
lesk	224	3	Hannah	Hannah	NNP
lesk	224	4	.	.	.
lesk	225	1	2015	2015	CD
lesk	225	2	.	.	.
lesk	226	1	“	"	``
lesk	226	2	Banking	bank	VBG
lesk	226	3	on	on	IN
lesk	226	4	better	well	JJR
lesk	226	5	forecasts	forecast	NNS
lesk	226	6	:	:	:
lesk	226	7	the	the	DT
lesk	226	8	new	new	JJ
lesk	226	9	maths	math	NNS
lesk	226	10	of	of	IN
lesk	226	11	weather	weather	NN
lesk	226	12	prediction	prediction	NN
lesk	226	13	.	.	.
lesk	226	14	”	"	''
lesk	226	15	The	the	DT
lesk	226	16	Guardian	Guardian	NNP
lesk	226	17	,	,	,
lesk	226	18	8	8	CD
lesk	226	19	Jan	Jan	NNP
lesk	226	20	2015	2015	CD
lesk	226	21	.	.	.
lesk	226	22	https://www.theguardian.com/science/alexs-adventures-in-numberland/2015/jan/08/banking-forecasts-maths-weather-prediction-stochastic-processes	https://www.theguardian.com/science/alexs-adventures-in-numberland/2015/jan/08/banking-forecasts-maths-weather-prediction-stochastic-processes	NNP
lesk	226	23	.	.	.
lesk	227	1	Eykholt	eykholt	NN
lesk	227	2	,	,	,
lesk	227	3	Kevin	Kevin	NNP
lesk	227	4	,	,	,
lesk	227	5	Ivan	Ivan	NNP
lesk	227	6	Evtimov	Evtimov	NNP
lesk	227	7	,	,	,
lesk	227	8	Earlence	Earlence	NNP
lesk	227	9	Fernandes	Fernandes	NNP
lesk	227	10	,	,	,
lesk	227	11	Bo	Bo	NNP
lesk	227	12	Li	Li	NNP
lesk	227	13	,	,	,
lesk	227	14	Amir	Amir	NNP
lesk	227	15	Rahmati	Rahmati	NNP
lesk	227	16	,	,	,
lesk	227	17	Florian	Florian	NNP
lesk	227	18	Tramèr	Tramèr	NNP
lesk	227	19	,	,	,
lesk	227	20	Atul	Atul	NNP
lesk	227	21	Prakash	Prakash	NNP
lesk	227	22	,	,	,
lesk	227	23	Tadayoshi	Tadayoshi	NNP
lesk	227	24	Kohno	Kohno	NNP
lesk	227	25	,	,	,
lesk	227	26	and	and	CC
lesk	227	27	Dawn	Dawn	NNP
lesk	227	28	Song	Song	NNP
lesk	227	29	.	.	.
lesk	228	1	2018	2018	CD
lesk	228	2	.	.	.
lesk	229	1	“	"	``
lesk	229	2	Physical	Physical	NNP
lesk	229	3	Adversarial	Adversarial	NNP
lesk	229	4	Examples	Examples	NNPS
lesk	229	5	for	for	IN
lesk	229	6	Object	Object	NNP
lesk	229	7	Detectors	detector	NNS
lesk	229	8	.	.	.
lesk	229	9	”	"	''
lesk	229	10	12th	12th	JJ
lesk	229	11	USENIX	USENIX	NNP
lesk	229	12	Workshop	Workshop	NNP
lesk	229	13	on	on	IN
lesk	229	14	Offensive	Offensive	NNP
lesk	229	15	Technologies	Technologies	NNPS
lesk	229	16	(	(	-LRB-
lesk	229	17	WOOT	WOOT	NNP
lesk	229	18	18	18	CD
lesk	229	19	)	)	-RRB-
lesk	229	20	.	.	.
lesk	230	1	Guidiotti	Guidiotti	NNP
lesk	230	2	,	,	,
lesk	230	3	Riccardo	Riccardo	NNP
lesk	230	4	,	,	,
lesk	230	5	Anna	Anna	NNP
lesk	230	6	Monreale	Monreale	NNP
lesk	230	7	,	,	,
lesk	230	8	Salvatore	Salvatore	NNP
lesk	230	9	Ruggieri	Ruggieri	NNP
lesk	230	10	,	,	,
lesk	230	11	Franco	Franco	NNP
lesk	230	12	Turini	Turini	NNP
lesk	230	13	,	,	,
lesk	230	14	Giannotti	Giannotti	NNP
lesk	230	15	Fosca	Fosca	NNP
lesk	230	16	,	,	,
lesk	230	17	and	and	CC
lesk	230	18	Dino	Dino	NNP
lesk	230	19	Pedreschi	Pedreschi	NNP
lesk	230	20	.	.	.
lesk	231	1	2018	2018	CD
lesk	231	2	.	.	.
lesk	232	1	“	"	``
lesk	232	2	A	a	DT
lesk	232	3	Survey	Survey	NNP
lesk	232	4	of	of	IN
lesk	232	5	Methods	Methods	NNPS
lesk	232	6	for	for	IN
lesk	232	7	Explaining	explain	VBG
lesk	232	8	Black	Black	NNP
lesk	232	9	Box	Box	NNP
lesk	232	10	Models	Models	NNP
lesk	232	11	.	.	.
lesk	232	12	”	"	''
lesk	232	13	ACM	ACM	NNP
lesk	232	14	Computing	Computing	NNP
lesk	232	15	Surveys	Surveys	NNP
lesk	232	16	51	51	CD
lesk	232	17	(	(	-LRB-
lesk	232	18	5	5	CD
lesk	232	19	)	)	-RRB-
lesk	232	20	:	:	:
lesk	232	21	1	1	CD
lesk	232	22	-	-	SYM
lesk	232	23	42	42	CD
lesk	232	24	.	.	.
lesk	233	1	Halevy	halevy	UH
lesk	233	2	,	,	,
lesk	233	3	Alon	Alon	NNP
lesk	233	4	,	,	,
lesk	233	5	Peter	Peter	NNP
lesk	233	6	Norvig	Norvig	NNP
lesk	233	7	,	,	,
lesk	233	8	and	and	CC
lesk	233	9	Fernando	Fernando	NNP
lesk	233	10	Pereira	Pereira	NNP
lesk	233	11	.	.	.
lesk	234	1	2009	2009	CD
lesk	234	2	.	.	.
lesk	235	1	“	"	``
lesk	235	2	The	the	DT
lesk	235	3	Unreasonable	Unreasonable	NNP
lesk	235	4	Effectiveness	Effectiveness	NNP
lesk	235	5	of	of	IN
lesk	235	6	Data	Data	NNP
lesk	235	7	.	.	.
lesk	235	8	”	"	''
lesk	235	9	IEEE	IEEE	NNP
lesk	235	10	Intelligent	Intelligent	NNP
lesk	235	11	Systems	system	NNS
lesk	235	12	24	24	CD
lesk	235	13	(	(	-LRB-
lesk	235	14	2	2	CD
lesk	235	15	)	)	-RRB-
lesk	235	16	.	.	.
lesk	236	1	Harley	Harley	NNP
lesk	236	2	,	,	,
lesk	236	3	Adam	Adam	NNP
lesk	236	4	W.	W.	NNP
lesk	236	5	2015	2015	CD
lesk	236	6	.	.	.
lesk	237	1	“	"	``
lesk	237	2	An	an	DT
lesk	237	3	Interactive	Interactive	NNP
lesk	237	4	Node	Node	NNP
lesk	237	5	-	-	HYPH
lesk	237	6	Link	Link	NNP
lesk	237	7	Visualization	visualization	NN
lesk	237	8	of	of	IN
lesk	237	9	Convolutional	Convolutional	NNP
lesk	237	10	Neural	Neural	NNP
lesk	237	11	Networks	Networks	NNP
lesk	237	12	.	.	.
lesk	237	13	”	"	''
lesk	237	14	In	in	IN
lesk	237	15	Advances	advance	NNS
lesk	237	16	in	in	IN
lesk	237	17	Visual	Visual	NNP
lesk	237	18	Computing	Computing	NNP
lesk	237	19	,	,	,
lesk	237	20	edited	edit	VBN
lesk	237	21	by	by	IN
lesk	237	22	George	George	NNP
lesk	237	23	Bebis	Bebis	NNP
lesk	237	24	et	et	NNP
lesk	237	25	al	al	NNP
lesk	237	26	.	.	NNP
lesk	237	27	,	,	,
lesk	237	28	867	867	CD
lesk	237	29	-	-	SYM
lesk	237	30	77	77	CD
lesk	237	31	.	.	.
lesk	238	1	Lecture	Lecture	NNP
lesk	238	2	Notes	Notes	NNPS
lesk	238	3	in	in	IN
lesk	238	4	Computer	Computer	NNP
lesk	238	5	Science	Science	NNP
lesk	238	6	.	.	.
lesk	239	1	Cham	Cham	NNP
lesk	239	2	:	:	:
lesk	239	3	Springer	Springer	NNP
lesk	239	4	International	International	NNP
lesk	239	5	Publishing	Publishing	NNP
lesk	239	6	.	.	.
lesk	240	1	“	"	``
lesk	240	2	How	how	WRB
lesk	240	3	to	to	TO
lesk	240	4	Prevent	prevent	VB
lesk	240	5	Discriminatory	Discriminatory	NNP
lesk	240	6	Outcomes	Outcomes	NNP
lesk	240	7	in	in	IN
lesk	240	8	Machine	Machine	NNP
lesk	240	9	Learning	Learning	NNP
lesk	240	10	.	.	.
lesk	240	11	”	"	''
lesk	240	12	2018	2018	CD
lesk	240	13	.	.	.
lesk	241	1	White	White	NNP
lesk	241	2	Paper	Paper	NNP
lesk	241	3	from	from	IN
lesk	241	4	the	the	DT
lesk	241	5	Global	Global	NNP
lesk	241	6	Future	Future	NNP
lesk	241	7	Council	Council	NNP
lesk	241	8	on	on	IN
lesk	241	9	Human	Human	NNP
lesk	241	10	Rights	Rights	NNPS
lesk	241	11	2016	2016	CD
lesk	241	12	-	-	SYM
lesk	241	13	2018	2018	CD
lesk	241	14	,	,	,
lesk	241	15	World	World	NNP
lesk	241	16	Economic	Economic	NNP
lesk	241	17	Forum	Forum	NNP
lesk	241	18	.	.	.
lesk	242	1	https://www.weforum.org/whitepapers/how-to-prevent-discriminatory-outcomes-in-machine-learning	https://www.weforum.org/whitepapers/how-to-prevent-discriminatory-outcomes-in-machine-learning	NN
lesk	242	2	.	.	.
lesk	243	1	Huang	Huang	NNP
lesk	243	2	,	,	,
lesk	243	3	Xuedong	Xuedong	NNP
lesk	243	4	,	,	,
lesk	243	5	James	James	NNP
lesk	243	6	Baker	Baker	NNP
lesk	243	7	,	,	,
lesk	243	8	and	and	CC
lesk	243	9	Raj	Raj	NNP
lesk	243	10	Reddy	Reddy	NNP
lesk	243	11	.	.	.
lesk	244	1	2014	2014	CD
lesk	244	2	.	.	.
lesk	245	1	“	"	``
lesk	245	2	A	a	DT
lesk	245	3	Historical	Historical	NNP
lesk	245	4	Perspective	Perspective	NNP
lesk	245	5	of	of	IN
lesk	245	6	Speech	Speech	NNP
lesk	245	7	Recognition	Recognition	NNP
lesk	245	8	.	.	.
lesk	245	9	”	"	''
lesk	245	10	Communications	communication	NNS
lesk	245	11	of	of	IN
lesk	245	12	the	the	DT
lesk	245	13	ACM	ACM	NNP
lesk	245	14	57	57	CD
lesk	245	15	(	(	-LRB-
lesk	245	16	1	1	CD
lesk	245	17	)	)	-RRB-
lesk	245	18	:	:	:
lesk	245	19	94	94	CD
lesk	245	20	-	-	SYM
lesk	245	21	103	103	CD
lesk	245	22	.	.	.
lesk	246	1	Lazer	Lazer	NNP
lesk	246	2	,	,	,
lesk	246	3	David	David	NNP
lesk	246	4	,	,	,
lesk	246	5	Ryan	Ryan	NNP
lesk	246	6	Kennedy	Kennedy	NNP
lesk	246	7	,	,	,
lesk	246	8	Gary	Gary	NNP
lesk	246	9	King	King	NNP
lesk	246	10	,	,	,
lesk	246	11	and	and	CC
lesk	246	12	Alessandro	Alessandro	NNP
lesk	246	13	Vespignani	Vespignani	NNP
lesk	246	14	.	.	.
lesk	247	1	2014	2014	CD
lesk	247	2	.	.	.
lesk	248	1	“	"	``
lesk	248	2	The	the	DT
lesk	248	3	Parable	Parable	NNP
lesk	248	4	of	of	IN
lesk	248	5	Google	Google	NNP
lesk	248	6	Flu	Flu	NNP
lesk	248	7	:	:	:
lesk	248	8	Traps	trap	NNS
lesk	248	9	in	in	IN
lesk	248	10	Big	Big	NNP
lesk	248	11	Data	Data	NNP
lesk	248	12	Analysis	Analysis	NNP
lesk	248	13	.	.	.
lesk	248	14	”	"	''
lesk	248	15	Science	science	NN
lesk	248	16	343	343	CD
lesk	248	17	(	(	-LRB-
lesk	248	18	6176	6176	CD
lesk	248	19	)	)	-RRB-
lesk	248	20	:	:	:
lesk	248	21	1203–1205	1203–1205	ADD
lesk	248	22	.	.	.
lesk	249	1	Lehman	Lehman	NNP
lesk	249	2	,	,	,
lesk	249	3	Constance	Constance	NNP
lesk	249	4	,	,	,
lesk	249	5	Robert	Robert	NNP
lesk	249	6	Wellman	Wellman	NNP
lesk	249	7	,	,	,
lesk	249	8	Diana	Diana	NNP
lesk	249	9	Buist	Buist	NNP
lesk	249	10	,	,	,
lesk	249	11	Karl	Karl	NNP
lesk	249	12	Kerlikowske	Kerlikowske	NNP
lesk	249	13	,	,	,
lesk	249	14	Anna	Anna	NNP
lesk	249	15	Tosteson	Tosteson	NNP
lesk	249	16	,	,	,
lesk	249	17	and	and	CC
lesk	249	18	Diana	Diana	NNP
lesk	249	19	Miglioretti	Miglioretti	NNP
lesk	249	20	.	.	.
lesk	250	1	2015	2015	CD
lesk	250	2	.	.	.
lesk	251	1	“	"	``
lesk	251	2	Diagnostic	Diagnostic	NNP
lesk	251	3	Accuracy	Accuracy	NNP
lesk	251	4	of	of	IN
lesk	251	5	Digital	Digital	NNP
lesk	251	6	Screening	Screening	NNP
lesk	251	7	Mammography	Mammography	NNP
lesk	251	8	with	with	IN
lesk	251	9	and	and	CC
lesk	251	10	without	without	IN
lesk	251	11	Computer	Computer	NNP
lesk	251	12	-	-	HYPH
lesk	251	13	Aided	aid	VBN
lesk	251	14	Detection	detection	NN
lesk	251	15	.	.	.
lesk	251	16	”	"	''
lesk	251	17	JAMA	JAMA	NNP
lesk	251	18	Intern	Intern	NNP
lesk	251	19	Med	Med	NNP
lesk	251	20	175	175	CD
lesk	251	21	(	(	-LRB-
lesk	251	22	11	11	CD
lesk	251	23	)	)	-RRB-
lesk	251	24	:	:	:
lesk	251	25	1828	1828	CD
lesk	251	26	-	-	SYM
lesk	251	27	1837	1837	CD
lesk	251	28	.	.	.
lesk	252	1	Levenson	Levenson	NNP
lesk	252	2	,	,	,
lesk	252	3	Richard	Richard	NNP
lesk	252	4	M.	M.	NNP
lesk	252	5	,	,	,
lesk	252	6	Elizabeth	Elizabeth	NNP
lesk	252	7	A.	A.	NNP
lesk	252	8	Krupinski	Krupinski	NNP
lesk	252	9	,	,	,
lesk	252	10	Victor	Victor	NNP
lesk	252	11	M.	M.	NNP
lesk	252	12	Navarro	Navarro	NNP
lesk	252	13	,	,	,
lesk	252	14	and	and	CC
lesk	252	15	Edward	Edward	NNP
lesk	252	16	A.	A.	NNP
lesk	252	17	Wasserman	Wasserman	NNP
lesk	252	18	.	.	.
lesk	253	1	2015	2015	CD
lesk	253	2	.	.	.
lesk	254	1	“	"	``
lesk	254	2	Pigeons	pigeon	NNS
lesk	254	3	(	(	-LRB-
lesk	254	4	Columba	Columba	NNP
lesk	254	5	livia	livia	NN
lesk	254	6	)	)	-RRB-
lesk	254	7	as	as	IN
lesk	254	8	Trainable	Trainable	NNP
lesk	254	9	Observers	Observers	NNPS
lesk	254	10	of	of	IN
lesk	254	11	Pathology	Pathology	NNP
lesk	254	12	and	and	CC
lesk	254	13	Radiology	Radiology	NNP
lesk	254	14	Breast	Breast	NNP
lesk	254	15	Cancer	Cancer	NNP
lesk	254	16	Images	Images	NNPS
lesk	254	17	.	.	.
lesk	254	18	”	"	''
lesk	254	19	PLoS	PLoS	NNP
lesk	254	20	One	one	CD
lesk	254	21	,	,	,
lesk	254	22	November	November	NNP
lesk	254	23	18	18	CD
lesk	254	24	,	,	,
lesk	254	25	2015	2015	CD
lesk	254	26	.	.	.
lesk	254	27	https://doi.org/10.1371/journal.pone.0141357	https://doi.org/10.1371/journal.pone.0141357	NNP
lesk	254	28	.	.	.
lesk	255	1	Lipton	Lipton	NNP
lesk	255	2	,	,	,
lesk	255	3	Zachary	Zachary	NNP
lesk	255	4	.	.	.
lesk	256	1	2018	2018	CD
lesk	256	2	.	.	.
lesk	257	1	“	"	``
lesk	257	2	The	the	DT
lesk	257	3	Mythos	Mythos	NNP
lesk	257	4	of	of	IN
lesk	257	5	Model	Model	NNP
lesk	257	6	Interpretability	Interpretability	NNP
lesk	257	7	.	.	.
lesk	257	8	”	"	''
lesk	257	9	ACM	ACM	NNP
lesk	257	10	Queue	Queue	NNP
lesk	257	11	61	61	CD
lesk	257	12	(	(	-LRB-
lesk	257	13	10	10	CD
lesk	257	14	)	)	-RRB-
lesk	257	15	:	:	:
lesk	257	16	36	36	CD
lesk	257	17	-	-	SYM
lesk	257	18	43	43	CD
lesk	257	19	.	.	.
lesk	258	1	Liu	Liu	NNP
lesk	258	2	,	,	,
lesk	258	3	Xiaoxuan	Xiaoxuan	NNP
lesk	258	4	et	et	FW
lesk	258	5	al	al	NNP
lesk	258	6	.	.	.
lesk	259	1	2019	2019	CD
lesk	259	2	.	.	.
lesk	260	1	“	"	``
lesk	260	2	A	a	DT
lesk	260	3	Comparison	Comparison	NNP
lesk	260	4	of	of	IN
lesk	260	5	Deep	Deep	NNP
lesk	260	6	Learning	Learning	NNP
lesk	260	7	Performance	Performance	NNP
lesk	260	8	against	against	IN
lesk	260	9	Health	Health	NNP
lesk	260	10	-	-	HYPH
lesk	260	11	Care	Care	NNP
lesk	260	12	Professionals	professional	NNS
lesk	260	13	in	in	IN
lesk	260	14	Detecting	detect	VBG
lesk	260	15	Diseases	Diseases	NNPS
lesk	260	16	from	from	IN
lesk	260	17	Medical	Medical	NNP
lesk	260	18	Imaging	Imaging	NNP
lesk	260	19	:	:	:
lesk	260	20	a	a	DT
lesk	260	21	Systematic	Systematic	NNP
lesk	260	22	Review	Review	NNP
lesk	260	23	and	and	CC
lesk	260	24	Meta	Meta	NNP
lesk	260	25	-	-	HYPH
lesk	260	26	Analysis	Analysis	NNP
lesk	260	27	.	.	.
lesk	260	28	”	"	''
lesk	260	29	Lancet	Lancet	NNP
lesk	260	30	Digital	Digital	NNP
lesk	260	31	Health	Health	NNP
lesk	260	32	1	1	CD
lesk	260	33	(	(	-LRB-
lesk	260	34	6	6	CD
lesk	260	35	)	)	-RRB-
lesk	260	36	:	:	:
lesk	260	37	e271	e271	LS
lesk	260	38	-	-	SYM
lesk	260	39	97	97	CD
lesk	260	40	.	.	.
lesk	261	1	https://www.sciencedirect.com/science/article/pii/S2589750019301232	https://www.sciencedirect.com/science/article/pii/S2589750019301232	NNP
lesk	261	2	Mahjoubi	Mahjoubi	NNP
lesk	261	3	,	,	,
lesk	261	4	Mounir	Mounir	NNP
lesk	261	5	.	.	.
lesk	262	1	2018	2018	CD
lesk	262	2	.	.	.
lesk	263	1	“	"	``
lesk	263	2	Assemblée	assemblée	JJ
lesk	263	3	nationale	nationale	NN
lesk	263	4	,	,	,
lesk	263	5	XVe	xve	NN
lesk	263	6	législature	législature	NN
lesk	263	7	.	.	.
lesk	264	1	Session	session	NN
lesk	264	2	ordinaire	ordinaire	NN
lesk	264	3	de	de	FW
lesk	264	4	2017	2017	CD
lesk	264	5	-	-	SYM
lesk	264	6	2018	2018	CD
lesk	264	7	.	.	.
lesk	264	8	”	"	''
lesk	264	9	Compte	Compte	NNP
lesk	264	10	rendu	rendu	VBP
lesk	264	11	intégral	intégral	NN
lesk	264	12	,	,	,
lesk	264	13	Deuxième	Deuxième	NNP
lesk	264	14	séance	séance	NN
lesk	264	15	du	du	NNP
lesk	264	16	mercredi	mercredi	NNP
lesk	264	17	07	07	CD
lesk	264	18	février	févri	JJR
lesk	264	19	2018	2018	CD
lesk	264	20	.	.	.
lesk	264	21	http://www.assemblee-nationale.fr/15/cri/2017-2018/20180137.asp	http://www.assemblee-nationale.fr/15/cri/2017-2018/20180137.asp	NNP
lesk	264	22	.	.	.
lesk	265	1	Metz	Metz	NNP
lesk	265	2	,	,	,
lesk	265	3	Cade	Cade	NNP
lesk	265	4	.	.	.
lesk	266	1	2016	2016	CD
lesk	266	2	.	.	.
lesk	267	1	“	"	``
lesk	267	2	Artificial	Artificial	NNP
lesk	267	3	Intelligence	Intelligence	NNP
lesk	267	4	Is	be	VBZ
lesk	267	5	Setting	set	VBG
lesk	267	6	Up	up	RP
lesk	267	7	the	the	DT
lesk	267	8	Internet	internet	NN
lesk	267	9	for	for	IN
lesk	267	10	a	a	DT
lesk	267	11	Huge	Huge	NNP
lesk	267	12	Clash	Clash	NNP
lesk	267	13	with	with	IN
lesk	267	14	Europe	Europe	NNP
lesk	267	15	.	.	.
lesk	267	16	”	"	''
lesk	267	17	Wired	wire	VBN
lesk	267	18	,	,	,
lesk	267	19	July	July	NNP
lesk	267	20	11	11	CD
lesk	267	21	,	,	,
lesk	267	22	2016	2016	CD
lesk	267	23	.	.	.
lesk	267	24	https://www.wired.com/2016/07/artificial-intelligence-setting-internet-huge-clash-europe/.	https://www.wired.com/2016/07/artificial-intelligence-setting-internet-huge-clash-europe/.	ADD
lesk	268	1	O’Neil	O’Neil	NNP
lesk	268	2	,	,	,
lesk	268	3	Cathy	Cathy	NNP
lesk	268	4	.	.	.
lesk	269	1	2016	2016	CD
lesk	269	2	.	.	.
lesk	270	1	Weapons	Weapons	NNP
lesk	270	2	of	of	IN
lesk	270	3	Math	Math	NNP
lesk	270	4	Destruction	Destruction	NNP
lesk	270	5	.	.	.
lesk	271	1	New	New	NNP
lesk	271	2	York	York	NNP
lesk	271	3	:	:	:
lesk	271	4	Crown	Crown	NNP
lesk	271	5	.	.	.
lesk	272	1	Peng	Peng	NNP
lesk	272	2	,	,	,
lesk	272	3	Tony	Tony	NNP
lesk	272	4	.	.	.
lesk	273	1	2018	2018	CD
lesk	273	2	.	.	.
lesk	274	1	“	"	``
lesk	274	2	2018	2018	CD
lesk	274	3	in	in	IN
lesk	274	4	review	review	NN
lesk	274	5	:	:	:
lesk	274	6	10	10	CD
lesk	274	7	AI	ai	NN
lesk	274	8	failures	failure	NNS
lesk	274	9	.	.	.
lesk	274	10	”	"	''
lesk	274	11	Medium	Medium	NNP
lesk	274	12	,	,	,
lesk	274	13	December	December	NNP
lesk	274	14	10	10	CD
lesk	274	15	,	,	,
lesk	274	16	2018	2018	CD
lesk	274	17	.	.	.
lesk	274	18	https://medium.com/syncedreview/2018-in-review-10-ai-failures-c18faadf5983	https://medium.com/syncedreview/2018-in-review-10-ai-failures-c18faadf5983	ADD
lesk	274	19	.	.	.
lesk	275	1	Qureshi	Qureshi	NNP
lesk	275	2	,	,	,
lesk	275	3	A.	a.	NN
lesk	275	4	I.	I.	NNP
lesk	275	5	,	,	,
lesk	275	6	M.	M.	NNP
lesk	275	7	Z.	Z.	NNP
lesk	275	8	Memon	Memon	NNP
lesk	275	9	,	,	,
lesk	275	10	G.	G.	NNP
lesk	275	11	Vazquez	Vazquez	NNP
lesk	275	12	,	,	,
lesk	275	13	and	and	CC
lesk	275	14	M.	M.	NNP
lesk	275	15	F.	F.	NNP
lesk	275	16	Suri	Suri	NNP
lesk	275	17	.	.	.
lesk	276	1	2009	2009	CD
lesk	276	2	.	.	.
lesk	277	1	“	"	``
lesk	277	2	Cat	cat	NN
lesk	277	3	ownership	ownership	NN
lesk	277	4	and	and	CC
lesk	277	5	the	the	DT
lesk	277	6	Risk	Risk	NNP
lesk	277	7	of	of	IN
lesk	277	8	Fatal	Fatal	NNP
lesk	277	9	Cardiovascular	Cardiovascular	NNP
lesk	277	10	Diseases	Diseases	NNPS
lesk	277	11	.	.	.
lesk	278	1	Results	result	NNS
lesk	278	2	from	from	IN
lesk	278	3	the	the	DT
lesk	278	4	Second	Second	NNP
lesk	278	5	National	National	NNP
lesk	278	6	Health	Health	NNP
lesk	278	7	and	and	CC
lesk	278	8	Nutrition	Nutrition	NNP
lesk	278	9	Examination	Examination	NNP
lesk	278	10	Study	Study	NNP
lesk	278	11	Mortality	Mortality	NNP
lesk	278	12	Follow	Follow	NNP
lesk	278	13	-	-	HYPH
lesk	278	14	up	up	RP
lesk	278	15	Study	Study	NNP
lesk	278	16	.	.	.
lesk	278	17	”	"	''
lesk	278	18	Journal	Journal	NNP
lesk	278	19	of	of	IN
lesk	278	20	Vascular	Vascular	NNP
lesk	278	21	and	and	CC
lesk	278	22	Interventional	Interventional	NNP
lesk	278	23	Neurology	Neurology	NNP
lesk	278	24	2	2	CD
lesk	278	25	(	(	-LRB-
lesk	278	26	1	1	CD
lesk	278	27	)	)	-RRB-
lesk	278	28	:	:	:
lesk	278	29	132	132	CD
lesk	278	30	-	-	SYM
lesk	278	31	5	5	CD
lesk	278	32	.	.	.
lesk	278	33	https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3317329	https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3317329	NNP
lesk	278	34	.	.	.
lesk	279	1	Ribeiro	Ribeiro	NNP
lesk	279	2	,	,	,
lesk	279	3	Marco	Marco	NNP
lesk	279	4	Tulio	Tulio	NNP
lesk	279	5	,	,	,
lesk	279	6	Sameer	Sameer	NNP
lesk	279	7	Singh	Singh	NNP
lesk	279	8	,	,	,
lesk	279	9	and	and	CC
lesk	279	10	Carlos	Carlos	NNP
lesk	279	11	Guestrin	Guestrin	NNP
lesk	279	12	.	.	.
lesk	280	1	2016	2016	CD
lesk	280	2	.	.	.
lesk	281	1	“	"	``
lesk	281	2	‘	'	``
lesk	281	3	Why	why	WRB
lesk	281	4	Should	Should	MD
lesk	281	5	I	-PRON-	PRP
lesk	281	6	Trust	trust	VB
lesk	281	7	You	-PRON-	PRP
lesk	281	8	?	?	.
lesk	281	9	’	'	''
lesk	281	10	:	:	:
lesk	281	11	Explaining	explain	VBG
lesk	281	12	the	the	DT
lesk	281	13	Predictions	prediction	NNS
lesk	281	14	of	of	IN
lesk	281	15	Any	any	DT
lesk	281	16	Classifier	Classifier	NNP
lesk	281	17	.	.	.
lesk	281	18	”	"	''
lesk	281	19	In	in	IN
lesk	281	20	Proceedings	Proceedings	NNP
lesk	281	21	of	of	IN
lesk	281	22	the	the	DT
lesk	281	23	22nd	22nd	JJ
lesk	281	24	ACM	ACM	NNP
lesk	281	25	SIGKDD	SIGKDD	NNP
lesk	281	26	International	International	NNP
lesk	281	27	Conference	Conference	NNP
lesk	281	28	on	on	IN
lesk	281	29	Knowledge	Knowledge	NNP
lesk	281	30	Discovery	Discovery	NNP
lesk	281	31	and	and	CC
lesk	281	32	Data	Data	NNPS
lesk	281	33	Mining	Mining	NNP
lesk	281	34	(	(	-LRB-
lesk	281	35	KDD	KDD	NNP
lesk	281	36	'	'	POS
lesk	281	37	16	16	CD
lesk	281	38	)	)	-RRB-
lesk	281	39	,	,	,
lesk	281	40	1135	1135	CD
lesk	281	41	-	-	SYM
lesk	281	42	1144	1144	CD
lesk	281	43	.	.	.
lesk	282	1	New	New	NNP
lesk	282	2	York	York	NNP
lesk	282	3	:	:	:
lesk	282	4	ACM	ACM	NNP
lesk	282	5	Press	Press	NNP
lesk	282	6	.	.	.
lesk	283	1	Schmidt	Schmidt	NNP
lesk	283	2	,	,	,
lesk	283	3	R.	R.	NNP
lesk	283	4	A.	A.	NNP
lesk	283	5	and	and	CC
lesk	283	6	R.	R.	NNP
lesk	283	7	M.	M.	NNP
lesk	283	8	Nishikawa	Nishikawa	NNP
lesk	283	9	.	.	.
lesk	284	1	1995	1995	CD
lesk	284	2	.	.	.
lesk	285	1	“	"	``
lesk	285	2	Clinical	clinical	JJ
lesk	285	3	Use	Use	NNP
lesk	285	4	of	of	IN
lesk	285	5	Digital	Digital	NNP
lesk	285	6	Mammography	Mammography	NNP
lesk	285	7	:	:	:
lesk	285	8	the	the	DT
lesk	285	9	Present	present	NN
lesk	285	10	and	and	CC
lesk	285	11	the	the	DT
lesk	285	12	Prospects	Prospects	NNPS
lesk	285	13	.	.	.
lesk	285	14	”	"	''
lesk	285	15	Journal	Journal	NNP
lesk	285	16	of	of	IN
lesk	285	17	Digital	Digital	NNP
lesk	285	18	Imaging	Imaging	NNP
lesk	285	19	8	8	CD
lesk	285	20	(	(	-LRB-
lesk	285	21	1	1	CD
lesk	285	22	Suppl	Suppl	NNP
lesk	285	23	1	1	CD
lesk	285	24	)	)	-RRB-
lesk	285	25	:	:	:
lesk	285	26	74	74	CD
lesk	285	27	-	-	SYM
lesk	285	28	9	9	CD
lesk	285	29	.	.	.
lesk	286	1	Singh	Singh	NNP
lesk	286	2	,	,	,
lesk	286	3	Vivek	Vivek	NNP
lesk	286	4	Kumar	Kumar	NNP
lesk	286	5	et	et	NNP
lesk	286	6	al	al	NNP
lesk	286	7	.	.	.
lesk	287	1	2020	2020	CD
lesk	287	2	.	.	.
lesk	288	1	“	"	``
lesk	288	2	Breast	Breast	NNP
lesk	288	3	Tumor	Tumor	NNP
lesk	288	4	Segmentation	segmentation	NN
lesk	288	5	and	and	CC
lesk	288	6	Shape	Shape	NNP
lesk	288	7	Classification	Classification	NNP
lesk	288	8	in	in	IN
lesk	288	9	Mammograms	Mammograms	NNP
lesk	288	10	Using	use	VBG
lesk	288	11	Generative	Generative	NNP
lesk	288	12	Adversarial	Adversarial	NNP
lesk	288	13	and	and	CC
lesk	288	14	Convolutional	Convolutional	NNP
lesk	288	15	Neural	Neural	NNP
lesk	288	16	Network	Network	NNP
lesk	288	17	.	.	.
lesk	288	18	”	"	''
lesk	288	19	Expert	expert	NN
lesk	288	20	Systems	system	NNS
lesk	288	21	with	with	IN
lesk	288	22	Applications	Applications	NNPS
lesk	288	23	139	139	CD
lesk	288	24	.	.	.
lesk	289	1	Su	Su	NNP
lesk	289	2	,	,	,
lesk	289	3	Jiawei	Jiawei	NNP
lesk	289	4	,	,	,
lesk	289	5	Danilo	Danilo	NNP
lesk	289	6	Vasconcellos	Vasconcellos	NNP
lesk	289	7	Vargas	Vargas	NNP
lesk	289	8	,	,	,
lesk	289	9	and	and	CC
lesk	289	10	Kouichi	Kouichi	NNP
lesk	289	11	Sakurai	Sakurai	NNP
lesk	289	12	.	.	.
lesk	290	1	2019	2019	CD
lesk	290	2	.	.	.
lesk	291	1	“	"	``
lesk	291	2	One	one	CD
lesk	291	3	Pixel	Pixel	NNP
lesk	291	4	Attack	Attack	NNP
lesk	291	5	for	for	IN
lesk	291	6	Fooling	Fooling	NNP
lesk	291	7	Deep	Deep	NNP
lesk	291	8	Neural	Neural	NNP
lesk	291	9	Networks	Networks	NNPS
lesk	291	10	.	.	.
lesk	291	11	”	"	''
lesk	291	12	IEEE	IEEE	NNP
lesk	291	13	Transactions	Transactions	NNPS
lesk	291	14	on	on	IN
lesk	291	15	Evolutionary	Evolutionary	NNP
lesk	291	16	Computation	Computation	NNP
lesk	291	17	23	23	CD
lesk	291	18	(	(	-LRB-
lesk	291	19	5	5	CD
lesk	291	20	)	)	-RRB-
lesk	291	21	:	:	:
lesk	291	22	828	828	CD
lesk	291	23	-	-	SYM
lesk	291	24	841	841	CD
lesk	291	25	.	.	.
lesk	292	1	Taulli	Taulli	NNP
lesk	292	2	,	,	,
lesk	292	3	Tom	Tom	NNP
lesk	292	4	.	.	.
lesk	293	1	2019	2019	CD
lesk	293	2	.	.	.
lesk	294	1	“	"	``
lesk	294	2	How	how	WRB
lesk	294	3	Bias	Bias	NNP
lesk	294	4	Distorts	Distorts	NNPS
lesk	294	5	AI	AI	NNP
lesk	294	6	(	(	-LRB-
lesk	294	7	Artificial	Artificial	NNP
lesk	294	8	Intelligence	Intelligence	NNP
lesk	294	9	)	)	-RRB-
lesk	294	10	.	.	.
lesk	294	11	”	"	''
lesk	294	12	Forbes	Forbes	NNP
lesk	294	13	,	,	,
lesk	294	14	August	August	NNP
lesk	294	15	4	4	CD
lesk	294	16	,	,	,
lesk	294	17	2019	2019	CD
lesk	294	18	.	.	.
lesk	294	19	https://www.forbes.com/sites/tomtaulli/2019/08/04/bias-the-silent-killer-of-ai-artificial-intelligence/#1cc6f35d7d87	https://www.forbes.com/sites/tomtaulli/2019/08/04/bias-the-silent-killer-of-ai-artificial-intelligence/#1cc6f35d7d87	ADD
lesk	294	20	.	.	.
lesk	295	1	Twain	Twain	NNP
lesk	295	2	,	,	,
lesk	295	3	Mark	Mark	NNP
lesk	295	4	.	.	.
lesk	296	1	1883	1883	CD
lesk	296	2	.	.	.
lesk	297	1	Life	life	NN
lesk	297	2	on	on	IN
lesk	297	3	the	the	DT
lesk	297	4	Mississippi	Mississippi	NNP
lesk	297	5	.	.	.
lesk	298	1	Boston	Boston	NNP
lesk	298	2	:	:	:
lesk	298	3	J.	J.	NNP
lesk	298	4	R.	R.	NNP
lesk	298	5	Osgood	Osgood	NNP
lesk	298	6	&	&	CC
lesk	298	7	Co.	Co.	NNP
lesk	298	8	Tugend	Tugend	NNP
lesk	298	9	,	,	,
lesk	298	10	Alina	Alina	NNP
lesk	298	11	.	.	.
lesk	299	1	2019	2019	CD
lesk	299	2	.	.	.
lesk	300	1	“	"	``
lesk	300	2	The	the	DT
lesk	300	3	Bias	Bias	NNP
lesk	300	4	Embedded	embed	VBN
lesk	300	5	in	in	IN
lesk	300	6	Tech	Tech	NNP
lesk	300	7	.	.	.
lesk	300	8	”	"	''
lesk	300	9	The	the	DT
lesk	300	10	New	New	NNP
lesk	300	11	York	York	NNP
lesk	300	12	Times	Times	NNP
lesk	300	13	,	,	,
lesk	300	14	June	June	NNP
lesk	300	15	17	17	CD
lesk	300	16	,	,	,
lesk	300	17	2019	2019	CD
lesk	300	18	,	,	,
lesk	300	19	section	section	NN
lesk	300	20	F	F	NNP
lesk	300	21	,	,	,
lesk	300	22	10	10	CD
lesk	300	23	.	.	.
lesk	301	1	Walsh	Walsh	NNP
lesk	301	2	,	,	,
lesk	301	3	Fergus	Fergus	NNP
lesk	301	4	.	.	.
lesk	302	1	2020	2020	CD
lesk	302	2	.	.	.
lesk	303	1	“	"	``
lesk	303	2	AI	AI	NNP
lesk	303	3	‘	'	``
lesk	303	4	outperforms	outperform	NNS
lesk	303	5	’	'	''
lesk	303	6	doctors	doctor	NNS
lesk	303	7	diagnosing	diagnose	VBG
lesk	303	8	breast	breast	NN
lesk	303	9	cancer	cancer	NN
lesk	303	10	.	.	.
lesk	303	11	”	"	''
lesk	303	12	BBC	BBC	NNP
lesk	303	13	News	News	NNP
lesk	303	14	,	,	,
lesk	303	15	January	January	NNP
lesk	303	16	2	2	CD
lesk	303	17	,	,	,
lesk	303	18	2020	2020	CD
lesk	303	19	.	.	.
lesk	303	20	https://www.bbc.com/news/health-50857759	https://www.bbc.com/news/health-50857759	NNP
lesk	303	21	.	.	.
lesk	304	1	Wilks	Wilks	NNP
lesk	304	2	,	,	,
lesk	304	3	Daniel	Daniel	NNP
lesk	304	4	S.	S.	NNP
lesk	304	5	2008	2008	CD
lesk	304	6	.	.	.
lesk	304	7	Review	Review	NNP
lesk	304	8	of	of	IN
lesk	304	9	Empirical	Empirical	NNP
lesk	304	10	Methods	Methods	NNPS
lesk	304	11	in	in	IN
lesk	304	12	Short	Short	NNP
lesk	304	13	-	-	HYPH
lesk	304	14	Term	Term	NNP
lesk	304	15	Climate	Climate	NNP
lesk	304	16	Prediction	Prediction	NNP
lesk	304	17	,	,	,
lesk	304	18	by	by	IN
lesk	304	19	Huug	Huug	NNP
lesk	304	20	van	van	NNP
lesk	304	21	den	den	NNP
lesk	304	22	Dool	Dool	NNP
lesk	304	23	.	.	.
lesk	305	1	Bulletin	bulletin	NN
lesk	305	2	of	of	IN
lesk	305	3	the	the	DT
lesk	305	4	American	American	NNP
lesk	305	5	Meteorological	Meteorological	NNP
lesk	305	6	Society	Society	NNP
lesk	305	7	89	89	CD
lesk	305	8	(	(	-LRB-
lesk	305	9	6	6	CD
lesk	305	10	)	)	-RRB-
lesk	305	11	:	:	:
lesk	305	12	887	887	CD
lesk	305	13	-	-	SYM
lesk	305	14	88	88	CD
lesk	305	15	.	.	.