<s>
Kneser	B-General_Concept
–	I-General_Concept
Ney	I-General_Concept
smoothing	I-General_Concept
,	O
also	O
known	O
as	O
Kneser-Essen-Ney	O
smoothing	B-Application
,	O
is	O
a	O
method	O
primarily	O
used	O
to	O
calculate	O
the	O
probability	O
distribution	O
of	O
n-grams	B-Language
in	O
a	O
document	O
based	O
on	O
their	O
histories	O
.	O
</s>
<s>
It	O
is	O
widely	O
considered	O
the	O
most	O
effective	O
method	O
of	O
smoothing	B-Application
due	O
to	O
its	O
use	O
of	O
absolute	O
discounting	O
by	O
subtracting	O
a	O
fixed	O
value	O
from	O
the	O
probability	O
's	O
lower	O
order	O
terms	O
to	O
omit	O
n-grams	B-Language
with	O
lower	O
frequencies	O
.	O
</s>
<s>
This	O
approach	O
has	O
been	O
considered	O
equally	O
effective	O
for	O
both	O
higher	O
and	O
lower	O
order	O
n-grams	B-Language
.	O
</s>
<s>
If	O
it	O
appears	O
several	O
times	O
in	O
a	O
training	O
corpus	O
,	O
the	O
frequency	O
of	O
the	O
unigram	B-Language
"	O
Francisco	O
"	O
will	O
also	O
be	O
high	O
.	O
</s>
<s>
Relying	O
on	O
only	O
the	O
unigram	B-Language
frequency	O
to	O
predict	O
the	O
frequencies	O
of	O
n-grams	B-Language
leads	O
to	O
skewed	O
results	O
;	O
however	O
,	O
Kneser	B-General_Concept
–	I-General_Concept
Ney	I-General_Concept
smoothing	I-General_Concept
corrects	O
this	O
by	O
considering	O
the	O
frequency	O
of	O
the	O
unigram	B-Language
in	O
relation	O
to	O
possible	O
words	O
preceding	O
it	O
.	O
</s>
<s>
Where	O
the	O
unigram	B-Language
probability	O
depends	O
on	O
how	O
likely	O
it	O
is	O
to	O
see	O
the	O
word	O
in	O
an	O
unfamiliar	O
context	O
,	O
which	O
is	O
estimated	O
as	O
the	O
number	O
of	O
times	O
it	O
appears	O
after	O
any	O
other	O
word	O
divided	O
by	O
the	O
number	O
of	O
distinct	O
pairs	O
of	O
consecutive	O
words	O
in	O
the	O
corpus	O
:	O
</s>
<s>
The	O
parameter	O
is	O
a	O
constant	O
which	O
denotes	O
the	O
discount	O
value	O
subtracted	O
from	O
the	O
count	O
of	O
each	O
n-gram	B-Language
,	O
usually	O
between	O
0	O
and	O
1	O
.	O
</s>
<s>
This	O
equation	O
can	O
be	O
extended	O
to	O
n-grams	B-Language
.	O
</s>
<s>
The	O
addition	O
of	O
the	O
term	O
for	O
lower	O
order	O
n-grams	B-Language
adds	O
more	O
weight	O
to	O
the	O
overall	O
probability	O
when	O
the	O
count	O
for	O
the	O
higher	O
order	O
n-grams	B-Language
is	O
zero	O
.	O
</s>
<s>
Similarly	O
,	O
the	O
weight	O
of	O
the	O
lower	O
order	O
model	O
decreases	O
when	O
the	O
count	O
of	O
the	O
n-gram	B-Language
is	O
non	O
zero	O
.	O
</s>
<s>
This	O
approach	O
is	O
once	O
used	O
for	O
Google	B-Application
Translate	I-Application
under	O
a	O
MapReduce	B-Operating_System
implementation	O
.	O
</s>
