<s>
Never-Ending	B-Algorithm
Language	I-Algorithm
Learning	I-Algorithm
system	O
(	O
NELL	B-Algorithm
)	O
is	O
a	O
semantic	B-Application
machine	O
learning	O
system	O
developed	O
by	O
a	O
research	O
team	O
at	O
Carnegie	O
Mellon	O
University	O
,	O
and	O
supported	O
by	O
grants	O
from	O
DARPA	O
,	O
Google	B-Application
,	O
NSF	O
,	O
and	O
CNPq	O
with	O
portions	O
of	O
the	O
system	O
running	O
on	O
a	O
supercomputing	B-Architecture
cluster	B-Architecture
provided	O
by	O
Yahoo	B-Application
!	I-Application
.	O
</s>
<s>
NELL	B-Algorithm
was	O
programmed	O
by	O
its	O
developers	O
to	O
be	O
able	O
to	O
identify	O
a	O
basic	O
set	O
of	O
fundamental	O
semantic	B-Application
relationships	O
between	O
a	O
few	O
hundred	O
predefined	O
categories	O
of	O
data	O
,	O
such	O
as	O
cities	O
,	O
companies	O
,	O
emotions	O
and	O
sports	O
teams	O
.	O
</s>
<s>
Since	O
the	O
beginning	O
of	O
2010	O
,	O
the	O
Carnegie	O
Mellon	O
research	O
team	O
has	O
been	O
running	O
NELL	B-Algorithm
around	O
the	O
clock	O
,	O
sifting	O
through	O
hundreds	O
of	O
millions	O
of	O
web	O
pages	O
looking	O
for	O
connections	O
between	O
the	O
information	O
it	O
already	O
knows	O
and	O
what	O
it	O
finds	O
through	O
its	O
search	O
process	O
to	O
make	O
new	O
connections	O
in	O
a	O
manner	O
that	O
is	O
intended	O
to	O
mimic	O
the	O
way	O
humans	O
learn	O
new	O
information	O
.	O
</s>
<s>
For	O
example	O
,	O
in	O
encountering	O
the	O
word	O
pair	O
"	O
Pikes	O
Peak	O
"	O
,	O
NELL	B-Algorithm
would	O
notice	O
that	O
both	O
words	O
are	O
capitalized	O
and	O
deduce	O
from	O
the	O
second	O
word	O
that	O
it	O
was	O
the	O
name	O
of	O
a	O
mountain	O
,	O
and	O
then	O
build	O
on	O
the	O
relationship	O
of	O
words	O
surrounding	O
those	O
two	O
words	O
to	O
deduce	O
other	O
connections	O
.	O
</s>
<s>
The	O
goal	O
of	O
NELL	B-Algorithm
and	O
other	O
semantic	B-Application
learning	O
systems	O
,	O
such	O
as	O
IBM	O
's	O
Watson	B-Application
system	O
,	O
is	O
to	O
be	O
able	O
to	O
develop	O
means	O
of	O
answering	B-Algorithm
questions	I-Algorithm
posed	O
by	O
users	O
in	O
natural	O
language	O
with	O
no	O
human	O
intervention	O
in	O
the	O
process	O
.	O
</s>
<s>
Oren	O
Etzioni	O
of	O
the	O
University	O
of	O
Washington	O
lauded	O
the	O
system	O
's	O
"	O
continuous	O
learning	O
,	O
as	O
if	O
NELL	B-Algorithm
is	O
exercising	O
curiosity	O
on	O
its	O
own	O
,	O
with	O
little	O
human	O
help	O
"	O
.	O
</s>
<s>
By	O
October	O
2010	O
,	O
NELL	B-Algorithm
has	O
doubled	O
the	O
number	O
of	O
relationships	O
it	O
has	O
available	O
in	O
its	O
knowledge	O
base	O
and	O
has	O
learned	O
440,000	O
new	O
facts	O
,	O
with	O
an	O
accuracy	O
of	O
87%	O
.	O
</s>
<s>
Team	O
leader	O
Tom	O
M	O
.	O
Mitchell	O
,	O
chairman	O
of	O
the	O
machine	O
learning	O
department	O
at	O
Carnegie	O
Mellon	O
described	O
how	O
NELL	B-Algorithm
"	O
self-corrects	O
when	O
it	O
has	O
more	O
information	O
,	O
as	O
it	O
learns	O
more	O
"	O
,	O
though	O
it	O
does	O
sometimes	O
arrive	O
at	O
incorrect	O
conclusions	O
.	O
</s>
<s>
Accumulated	O
errors	O
,	O
such	O
as	O
the	O
deduction	O
that	O
Internet	B-Application
cookies	I-Application
were	O
a	O
kind	O
of	O
baked	O
good	O
,	O
led	O
NELL	B-Algorithm
to	O
deduce	O
from	O
the	O
phrases	O
"	O
I	O
deleted	O
my	O
Internet	B-Application
cookies	I-Application
"	O
and	O
"	O
I	O
deleted	O
my	O
files	O
"	O
that	O
"	O
computer	B-Operating_System
files	I-Operating_System
"	O
also	O
belonged	O
in	O
the	O
baked	O
goods	O
category	O
.	O
</s>
