<s>
In	O
natural	B-Language
language	I-Language
processing	I-Language
,	O
language	B-General_Concept
identification	I-General_Concept
or	O
language	B-General_Concept
guessing	I-General_Concept
is	O
the	O
problem	O
of	O
determining	O
which	O
natural	O
language	O
given	O
content	O
is	O
in	O
.	O
</s>
<s>
Computational	O
approaches	O
to	O
this	O
problem	O
view	O
it	O
as	O
a	O
special	O
case	O
of	O
text	B-Algorithm
categorization	I-Algorithm
,	O
solved	O
with	O
various	O
statistical	O
methods	O
.	O
</s>
<s>
There	O
are	O
several	O
statistical	O
approaches	O
to	O
language	B-General_Concept
identification	I-General_Concept
using	O
different	O
techniques	O
to	O
classify	O
the	O
data	O
.	O
</s>
<s>
Another	O
technique	O
,	O
as	O
described	O
by	O
Cavnar	O
and	O
Trenkle	O
(	O
1994	O
)	O
and	O
Dunning	O
(	O
1994	O
)	O
is	O
to	O
create	O
a	O
language	O
n-gram	B-Language
model	O
from	O
a	O
"	O
training	O
text	O
"	O
for	O
each	O
of	O
the	O
languages	O
.	O
</s>
<s>
These	O
models	O
can	O
be	O
based	O
on	O
characters	O
(	O
Cavnar	O
and	O
Trenkle	O
)	O
or	O
encoded	O
bytes	O
(	O
Dunning	O
)	O
;	O
in	O
the	O
latter	O
,	O
language	B-General_Concept
identification	I-General_Concept
and	O
character	O
encoding	O
detection	O
are	O
integrated	O
.	O
</s>
<s>
This	O
method	O
can	O
detect	O
multiple	O
languages	O
in	O
an	O
unstructured	O
piece	O
of	O
text	O
and	O
works	O
robustly	O
on	O
short	O
texts	O
of	O
only	O
a	O
few	O
words	O
:	O
something	O
that	O
the	O
n-gram	B-Language
approaches	O
struggle	O
with	O
.	O
</s>
<s>
A	O
common	O
non-statistical	O
intuitive	B-Application
approach	O
(	O
though	O
highly	O
uncertain	O
)	O
is	O
to	O
look	O
for	O
common	O
letter	O
combinations	O
,	O
or	O
distinctive	O
diacritics	O
or	O
punctuation	O
.	O
</s>
<s>
One	O
of	O
the	O
great	O
bottlenecks	O
of	O
language	B-General_Concept
identification	I-General_Concept
systems	O
is	O
to	O
distinguish	O
between	O
closely	O
related	O
languages	O
.	O
</s>
