<s>
Spelling	B-Application
suggestion	I-Application
is	O
a	O
feature	O
of	O
many	O
computer	B-Application
software	I-Application
applications	O
used	O
to	O
suggest	O
plausible	O
replacements	O
for	O
words	O
that	O
are	O
likely	O
to	O
have	O
been	O
misspelled	O
.	O
</s>
<s>
Spelling	B-Application
suggestion	I-Application
features	O
are	O
commonly	O
included	O
in	O
Internet	B-Application
search	I-Application
engines	I-Application
,	O
word	B-General_Concept
processors	I-General_Concept
,	O
spell	B-Application
checkers	I-Application
,	O
medical	O
transcription	O
,	O
automatic	B-Algorithm
query	I-Algorithm
reformulation	I-Algorithm
,	O
and	O
frequency-log	O
statistics	O
reporting	O
.	O
</s>
<s>
Any	O
spell	B-Application
checker	I-Application
must	O
have	O
some	O
data	O
about	O
the	O
words	O
in	O
the	O
target	O
language	O
,	O
either	O
in	O
general	O
usage	O
or	O
with	O
specialized	O
knowledge	O
(	O
like	O
medical	O
vocabulary	O
)	O
.	O
</s>
<s>
A	O
dictionary	B-Operating_System
of	O
all	O
known	O
words	O
.	O
</s>
<s>
Logs	O
of	O
human	O
text	O
input	O
,	O
such	O
as	O
from	O
a	O
popular	O
search	B-Application
engine	I-Application
.	O
</s>
<s>
Data	O
might	O
be	O
included	O
about	O
when	O
people	O
click	O
on	O
a	O
spelling	B-Application
suggestion	I-Application
or	O
make	O
a	O
second	O
,	O
very	O
similar	O
query	O
;	O
this	O
creates	O
a	O
crowdsourced	O
mapping	O
of	O
misspelled	O
words	O
to	O
reliable	O
corrections	O
.	O
</s>
<s>
To	O
make	O
use	O
of	O
a	O
dictionary	B-Operating_System
without	O
a	O
pre-existing	O
mapping	O
from	O
misspellings	O
to	O
corrections	O
,	O
the	O
typical	O
technique	O
is	O
to	O
calculate	O
the	O
edit	O
distance	O
between	O
an	O
input	O
word	O
and	O
any	O
given	O
word	O
in	O
the	O
dictionary	B-Operating_System
.	O
</s>
<s>
Dictionary	B-Operating_System
words	O
that	O
are	O
an	O
edit	O
distance	O
of	O
1	O
away	O
from	O
the	O
input	O
word	O
are	O
considered	O
highly	O
likely	O
as	O
corrections	O
,	O
edit	O
distance	O
2	O
less	O
likely	O
,	O
and	O
edit	O
distance	O
3	O
sometimes	O
included	O
in	O
suggestions	O
and	O
sometimes	O
ignored	O
.	O
</s>
<s>
A	O
text	O
corpus	O
can	O
be	O
summed	O
up	O
as	O
a	O
dictionary	B-Operating_System
of	O
known	O
words	O
,	O
with	O
a	O
frequency	O
of	O
appearance	O
for	O
each	O
word	O
.	O
</s>
<s>
This	O
can	O
be	O
used	O
to	O
sort	O
the	O
spelling	B-Application
suggestions	I-Application
.	O
</s>
<s>
Because	O
a	O
dictionary	B-Operating_System
of	O
known	O
words	O
is	O
very	O
large	O
,	O
calculating	O
the	O
edit	O
distance	O
between	O
an	O
input	O
word	O
and	O
every	O
word	O
in	O
the	O
dictionary	B-Operating_System
is	O
computationally	O
intensive	O
and	O
thus	O
relatively	O
slow	O
.	O
</s>
<s>
Various	O
data	B-General_Concept
structures	I-General_Concept
can	O
be	O
utilized	O
to	O
speed	O
up	O
storage	O
lookups	O
,	O
such	O
as	O
BK-trees	B-Data_Structure
.	O
</s>
<s>
A	O
faster	O
approach	O
adopted	O
by	O
Peter	O
Norvig	O
generates	O
all	O
the	O
permutations	B-Algorithm
from	O
an	O
input	O
word	O
of	O
all	O
possible	O
edits	O
.	O
</s>
<s>
Using	O
only	O
the	O
26	O
letters	O
in	O
the	O
English	O
alphabet	O
,	O
this	O
would	O
produce	O
only	O
54*n+25	O
dictionary	B-Operating_System
lookups	O
,	O
minus	O
any	O
duplicates	O
(	O
which	O
depends	O
on	O
the	O
specific	O
letters	O
in	O
the	O
word	O
)	O
.	O
</s>
<s>
This	O
is	O
relatively	O
small	O
when	O
compared	O
to	O
a	O
dictionary	B-Operating_System
of	O
hundreds	O
of	O
thousands	O
of	O
words	O
.	O
</s>
<s>
A	O
further	O
innovation	O
adopted	O
by	O
Wolf	O
Garbe	O
,	O
known	O
as	O
SymSpell	O
(	O
"	O
sym	O
"	O
as	O
in	O
"	O
symmetry	O
"	O
)	O
speeds	O
the	O
input-time	O
calculation	O
by	O
utilizing	O
the	O
fact	O
that	O
only	O
permutations	B-Algorithm
involving	O
deletions	O
need	O
to	O
be	O
generated	O
for	O
input	O
words	O
if	O
the	O
same	O
deletion	O
permutations	B-Algorithm
are	O
per-calculated	O
on	O
the	O
dictionary	B-Operating_System
.	O
</s>
<s>
The	O
algorithms	O
described	O
so	O
far	O
do	O
not	O
deal	O
well	O
with	O
correct	O
words	O
that	O
are	O
not	O
in	O
the	O
dictionary	B-Operating_System
.	O
</s>
<s>
These	O
can	O
be	O
accommodated	O
algorithmically	O
,	O
especially	O
if	O
the	O
dictionary	B-Operating_System
contains	O
the	O
part	O
of	O
speech	O
.	O
</s>
<s>
Spelling	B-Application
suggestions	I-Application
can	O
also	O
be	O
made	O
more	O
accurate	O
by	O
taking	O
into	O
account	O
more	O
than	O
one	O
word	O
at	O
a	O
time	O
.	O
</s>
<s>
Multi-word	O
sequences	O
are	O
known	O
as	O
n-grams	B-Language
(	O
where	O
n	O
is	O
the	O
number	O
of	O
words	O
in	O
the	O
sequence	O
)	O
.	O
</s>
<s>
A	O
very	O
large	O
database	O
of	O
n-grams	B-Language
up	O
to	O
5	O
words	O
in	O
length	O
is	O
available	O
from	O
Google	O
for	O
this	O
and	O
other	O
purposes	O
.	O
</s>
<s>
Others	O
have	O
experimented	O
with	O
using	O
large	O
amounts	O
of	O
data	O
and	O
deep	B-Algorithm
learning	I-Algorithm
techniques	O
(	O
a	O
form	O
of	O
machine	O
learning	O
to	O
train	O
neural	B-Architecture
networks	I-Architecture
to	O
perform	O
spelling	O
correction	O
.	O
</s>
