<s>
The	O
IOB	O
format	O
(	O
short	O
for	O
inside	O
,	O
outside	O
,	O
beginning	O
)	O
,	O
also	O
commonly	O
referred	O
to	O
as	O
the	O
BIO	O
format	O
,	O
is	O
a	O
common	O
tagging	O
format	O
for	O
tagging	O
tokens	B-Application
in	O
a	O
chunking	B-General_Concept
task	O
in	O
computational	O
linguistics	O
(	O
ex	O
.	O
</s>
<s>
named-entity	B-General_Concept
recognition	I-General_Concept
)	O
.	O
</s>
<s>
It	O
was	O
presented	O
by	O
Ramshaw	O
and	O
Marcus	O
in	O
their	O
paper	O
"	O
Text	O
Chunking	B-General_Concept
using	O
Transformation-Based	O
Learning	O
"	O
,	O
1995	O
The	O
I	O
-	O
prefix	O
before	O
a	O
tag	O
indicates	O
that	O
the	O
tag	O
is	O
inside	O
a	O
chunk	B-General_Concept
.	O
</s>
<s>
An	O
O	O
tag	O
indicates	O
that	O
a	O
token	O
belongs	O
to	O
no	O
chunk	B-General_Concept
.	O
</s>
<s>
The	O
B	O
-	O
prefix	O
before	O
a	O
tag	O
indicates	O
that	O
the	O
tag	O
is	O
the	O
beginning	O
of	O
a	O
chunk	B-General_Concept
that	O
immediately	O
follows	O
another	O
chunk	B-General_Concept
without	O
O	O
tags	O
between	O
them	O
.	O
</s>
<s>
It	O
is	O
used	O
only	O
in	O
that	O
case	O
:	O
when	O
a	O
chunk	B-General_Concept
comes	O
after	O
an	O
O	O
tag	O
,	O
the	O
first	O
token	O
of	O
the	O
chunk	B-General_Concept
takes	O
the	O
I	O
-	O
prefix	O
.	O
</s>
<s>
Another	O
similar	O
format	O
which	O
is	O
widely	O
used	O
is	O
IOB2	O
format	O
,	O
which	O
is	O
the	O
same	O
as	O
the	O
IOB	O
format	O
except	O
that	O
the	O
B	O
-	O
tag	O
is	O
used	O
in	O
the	O
beginning	O
of	O
every	O
chunk	B-General_Concept
(	O
i.e.	O
</s>
<s>
all	O
chunks	B-General_Concept
start	O
with	O
the	O
B	O
-	O
tag	O
)	O
.	O
</s>
<s>
Notice	O
how	O
"	O
Alex	O
"	O
,	O
"	O
Los	O
"	O
and	O
"	O
California	O
"	O
,	O
although	O
first	O
tokens	B-Application
of	O
their	O
chunk	B-General_Concept
,	O
have	O
the	O
"	O
I-	O
"	O
prefix	O
.	O
</s>
<s>
Notice	O
how	O
"	O
California	O
"	O
now	O
has	O
the	O
"	O
B-	O
"	O
prefix	O
,	O
because	O
it	O
immediately	O
follows	O
another	O
LOC	O
chunk	B-General_Concept
.	O
</s>
<s>
Related	O
tagging	O
schemes	O
sometimes	O
include	O
"	O
START/END	O
:	O
This	O
consists	O
of	O
the	O
tags	O
B	O
,	O
E	O
,	O
I	O
,	O
S	O
or	O
O	O
where	O
S	O
is	O
used	O
to	O
represent	O
a	O
chunk	B-General_Concept
containing	O
a	O
single	O
token	O
.	O
</s>
<s>
Chunks	B-General_Concept
of	O
length	O
greater	O
than	O
or	O
equal	O
to	O
two	O
always	O
start	O
with	O
the	O
B	O
tag	O
and	O
end	O
with	O
the	O
E	O
tag.	O
"	O
</s>
<s>
IOB	O
syntax	O
does	O
not	O
permit	O
any	O
nesting	O
,	O
so	O
cannot	O
(	O
unless	O
extended	O
)	O
also	O
represent	O
even	O
very	O
simple	O
phenomena	O
such	O
as	O
sentence	O
boundaries	O
(	O
which	O
are	O
not	O
trivial	O
to	O
locate	O
reliably	O
)	O
,	O
the	O
scope	O
of	O
parenthetical	O
expressions	O
in	O
sentences	O
,	O
grammatical	O
structures	O
,	O
nested	O
Named	B-General_Concept
Entities	I-General_Concept
such	O
as	O
"	O
University	O
of	O
Wisconsin	O
Dept	O
.	O
</s>
<s>
It	O
also	O
leaves	O
no	O
place	O
for	O
metadata	O
such	O
as	O
an	O
identifier	O
for	O
the	O
particular	O
sample	O
,	O
the	O
confidence	O
level	O
of	O
the	O
NER	O
assignment	O
,	O
and	O
so	O
on	O
,	O
which	O
are	O
commonplace	O
in	O
NLP	B-Language
systems	O
.	O
</s>
<s>
The	O
space	O
and	O
"	O
O	O
"	O
(	O
meaning	O
"	O
not	O
in	O
any	O
chunk	B-General_Concept
"	O
)	O
convey	O
no	O
information	O
and	O
could	O
simply	O
be	O
omitted	O
.	O
</s>
<s>
IOB	O
's	O
"	O
one	O
token	O
per	O
line	O
"	O
depends	O
on	O
the	O
tokenization	O
used	O
,	O
even	O
though	O
tokenization	O
is	O
not	O
standardized	O
in	O
NLP	B-Language
,	O
and	O
details	O
of	O
tokenization	O
do	O
not	O
have	O
to	O
be	O
entangled	O
with	O
the	O
representations	O
of	O
NERs	O
.	O
</s>
<s>
"	O
11/31/2019	O
"	O
could	O
be	O
anywhere	O
from	O
one	O
to	O
five	O
tokens	B-Application
in	O
different	O
systems	O
,	O
but	O
the	O
NER	O
is	O
the	O
same	O
.	O
</s>
<s>
Some	O
systems	O
even	O
permit	O
whitespace	O
within	O
tokens	B-Application
,	O
and	O
space	O
as	O
a	O
delimiter	O
collides	O
with	O
this	O
,	O
narrowing	O
the	O
applicability	O
of	O
IOB	O
and	O
motivating	O
more	O
extensions	O
.	O
</s>
<s>
More	O
powerful	O
formats	O
(	O
most	O
obviously	O
XML	B-Protocol
and	O
JSON	B-General_Concept
)	O
can	O
handle	O
far	O
more	O
diverse	O
annotations	O
,	O
have	O
less	O
variation	O
between	O
implementations	O
,	O
and	O
are	O
often	O
shorter	O
and	O
more	O
readable	O
as	O
well	O
.	O
</s>
<s>
XML	B-Protocol
takes	O
80	O
bytes	O
to	O
do	O
the	O
same	O
things	O
as	O
the	O
91	O
byte	O
BIOES	O
version	O
shown	O
above	O
,	O
or	O
the	O
79	O
byte	O
IOB	O
version	O
.	O
</s>
<s>
However	O
,	O
it	O
can	O
easily	O
also	O
support	O
sentence	O
boundaries	O
,	O
part-of-speech	O
annotations	O
,	O
and	O
other	O
features	O
commonly	O
needed	O
in	O
NLP	B-Language
systems	O
.	O
</s>
<s>
Breaking	O
all	O
tokens	B-Application
in	O
particular	O
places	O
is	O
not	O
strictly	O
part	O
of	O
the	O
NER	O
task	O
;	O
but	O
even	O
if	O
every	O
token	O
were	O
tagged	O
(	O
like	O
"	O
is	O
"	O
)	O
the	O
total	O
would	O
grow	O
only	O
to	O
139	O
bytes	O
:	O
</s>
