<s>
In	O
linguistic	O
morphology	O
and	O
information	B-Library
retrieval	I-Library
,	O
stemming	B-General_Concept
is	O
the	O
process	O
of	O
reducing	O
inflected	O
(	O
or	O
sometimes	O
derived	O
)	O
words	O
to	O
their	O
word	O
stem	O
,	O
base	O
or	O
root	O
form	O
—	O
generally	O
a	O
written	O
word	O
form	O
.	O
</s>
<s>
Algorithms	O
for	O
stemming	B-General_Concept
have	O
been	O
studied	O
in	O
computer	B-General_Concept
science	I-General_Concept
since	O
the	O
1960s	O
.	O
</s>
<s>
Many	O
search	B-Application
engines	I-Application
treat	O
words	O
with	O
the	O
same	O
stem	O
as	O
synonyms	B-Application
as	O
a	O
kind	O
of	O
query	B-Algorithm
expansion	I-Algorithm
,	O
a	O
process	O
called	O
conflation	O
.	O
</s>
<s>
A	O
computer	B-Application
program	I-Application
or	O
subroutine	O
that	O
stems	O
word	O
may	O
be	O
called	O
a	O
stemming	B-General_Concept
program	O
,	O
stemming	B-General_Concept
algorithm	I-General_Concept
,	O
or	O
stemmer	O
.	O
</s>
<s>
A	O
stemming	B-General_Concept
algorithm	I-General_Concept
might	O
also	O
reduce	O
the	O
words	O
fishing	O
,	O
fished	O
,	O
and	O
fisher	O
to	O
the	O
stem	O
fish	O
.	O
</s>
<s>
Her	O
paper	O
refers	O
to	O
three	O
earlier	O
major	O
attempts	O
at	O
stemming	B-General_Concept
algorithms	I-General_Concept
,	O
by	O
Professor	O
John	O
W	O
.	O
Tukey	O
of	O
Princeton	O
University	O
,	O
the	O
algorithm	O
developed	O
at	O
Harvard	O
University	O
by	O
Michael	O
Lesk	O
,	O
under	O
the	O
direction	O
of	O
Professor	O
Gerard	O
Salton	O
,	O
and	O
a	O
third	O
algorithm	O
developed	O
by	O
James	O
L	O
.	O
Dolby	O
of	O
R	O
and	O
D	O
Consultants	O
,	O
Los	O
Altos	O
,	O
California	O
.	O
</s>
<s>
This	O
stemmer	O
was	O
very	O
widely	O
used	O
and	O
became	O
the	O
de	O
facto	O
standard	O
algorithm	O
used	O
for	O
English	O
stemming	B-General_Concept
.	O
</s>
<s>
Dr.	O
Porter	O
received	O
the	O
Tony	O
Kent	O
Strix	O
award	O
in	O
2000	O
for	O
his	O
work	O
on	O
stemming	B-General_Concept
and	O
information	B-Library
retrieval	I-Library
.	O
</s>
<s>
Many	O
implementations	O
of	O
the	O
Porter	O
stemming	B-General_Concept
algorithm	I-General_Concept
were	O
written	O
and	O
freely	O
distributed	O
;	O
however	O
,	O
many	O
of	O
these	O
implementations	O
contained	O
subtle	O
flaws	O
.	O
</s>
<s>
To	O
eliminate	O
this	O
source	O
of	O
error	O
,	O
Martin	O
Porter	O
released	O
an	O
official	O
free	B-Application
software	I-Application
(	O
mostly	O
BSD-licensed	O
)	O
implementation	O
of	O
the	O
algorithm	O
around	O
the	O
year	O
2000	O
.	O
</s>
<s>
He	O
extended	O
this	O
work	O
over	O
the	O
next	O
few	O
years	O
by	O
building	O
Snowball	B-Language
,	O
a	O
framework	O
for	O
writing	O
stemming	B-General_Concept
algorithms	I-General_Concept
,	O
and	O
implemented	O
an	O
improved	O
English	O
stemmer	O
together	O
with	O
stemmers	O
for	O
several	O
other	O
languages	O
.	O
</s>
<s>
The	O
Paice-Husk	O
Stemmer	O
was	O
developed	O
by	O
Chris	O
D	O
Paice	O
at	O
Lancaster	O
University	O
in	O
the	O
late	O
1980s	O
,	O
it	O
is	O
an	O
iterative	O
stemmer	O
and	O
features	O
an	O
externally	O
stored	O
set	O
of	O
stemming	B-General_Concept
rules	O
.	O
</s>
<s>
Paice	O
also	O
developed	O
a	O
direct	O
measurement	O
for	O
comparing	O
stemmers	O
based	O
on	O
counting	O
the	O
over-stemming	O
and	O
under-stemming	O
errors	O
.	O
</s>
<s>
There	O
are	O
several	O
types	O
of	O
stemming	B-General_Concept
algorithms	I-General_Concept
which	O
differ	O
in	O
respect	O
to	O
performance	O
and	O
accuracy	O
and	O
how	O
certain	O
stemming	B-General_Concept
obstacles	O
are	O
overcome	O
.	O
</s>
<s>
A	O
simple	O
stemmer	O
looks	O
up	O
the	O
inflected	O
form	O
in	O
a	O
lookup	B-Data_Structure
table	I-Data_Structure
.	O
</s>
<s>
A	O
lookup	B-Data_Structure
approach	O
may	O
use	O
preliminary	O
part-of-speech	O
tagging	O
to	O
avoid	O
overstemming	O
.	O
</s>
<s>
The	O
lookup	B-Data_Structure
table	I-Data_Structure
used	O
by	O
a	O
stemmer	O
is	O
generally	O
produced	O
semi-automatically	O
.	O
</s>
<s>
Suffix	O
stripping	O
algorithms	O
do	O
not	O
rely	O
on	O
a	O
lookup	B-Data_Structure
table	I-Data_Structure
that	O
consists	O
of	O
inflected	O
forms	O
and	O
root	O
form	O
relations	O
.	O
</s>
<s>
Lemmatisation	B-General_Concept
attempts	O
to	O
improve	O
upon	O
this	O
challenge	O
.	O
</s>
<s>
Diving	O
further	O
into	O
the	O
details	O
,	O
a	O
common	O
technique	O
is	O
to	O
apply	O
rules	O
in	O
a	O
cyclical	O
fashion	O
(	O
recursively	O
,	O
as	O
computer	B-General_Concept
scientists	I-General_Concept
would	O
say	O
)	O
.	O
</s>
<s>
Chances	O
are	O
that	O
the	O
brute	O
force	O
approach	O
would	O
be	O
slower	O
,	O
as	O
lookup	B-Data_Structure
algorithms	O
have	O
a	O
direct	O
access	O
to	O
the	O
solution	O
,	O
while	O
rule-based	O
should	O
try	O
several	O
options	O
,	O
and	O
combinations	O
of	O
them	O
,	O
and	O
then	O
choose	O
which	O
result	O
seems	O
to	O
be	O
the	O
best	O
.	O
</s>
<s>
A	O
more	O
complex	O
approach	O
to	O
the	O
problem	O
of	O
determining	O
a	O
stem	O
of	O
a	O
word	O
is	O
lemmatisation	B-General_Concept
.	O
</s>
<s>
The	O
part	O
of	O
speech	O
is	O
first	O
detected	O
prior	O
to	O
attempting	O
to	O
find	O
the	O
root	O
since	O
for	O
some	O
languages	O
,	O
the	O
stemming	B-General_Concept
rules	O
change	O
depending	O
on	O
a	O
word	O
's	O
part	O
of	O
speech	O
.	O
</s>
<s>
This	O
model	O
is	O
typically	O
expressed	O
in	O
the	O
form	O
of	O
complex	O
linguistic	O
rules	O
,	O
similar	O
in	O
nature	O
to	O
those	O
in	O
suffix	O
stripping	O
or	O
lemmatisation	B-General_Concept
.	O
</s>
<s>
Stemming	B-General_Concept
is	O
performed	O
by	O
inputting	O
an	O
inflected	O
form	O
to	O
the	O
trained	O
model	O
and	O
having	O
the	O
model	O
produce	O
the	O
root	O
form	O
according	O
to	O
its	O
internal	O
ruleset	O
,	O
which	O
again	O
is	O
similar	O
to	O
suffix	O
stripping	O
and	O
lemmatisation	B-General_Concept
,	O
except	O
that	O
the	O
decisions	O
involved	O
in	O
applying	O
the	O
most	O
appropriate	O
rule	O
,	O
or	O
whether	O
or	O
not	O
to	O
stem	O
the	O
word	O
and	O
just	O
return	O
the	O
same	O
word	O
,	O
or	O
whether	O
to	O
apply	O
two	O
different	O
rules	O
sequentially	O
,	O
are	O
applied	O
on	O
the	O
grounds	O
that	O
the	O
output	O
word	O
will	O
have	O
the	O
highest	O
probability	O
of	O
being	O
correct	O
(	O
which	O
is	O
to	O
say	O
,	O
the	O
smallest	O
probability	O
of	O
being	O
incorrect	O
,	O
which	O
is	O
how	O
it	O
is	O
typically	O
measured	O
)	O
.	O
</s>
<s>
Some	O
lemmatisation	B-General_Concept
algorithms	O
are	O
stochastic	O
in	O
that	O
,	O
given	O
a	O
word	O
which	O
may	O
belong	O
to	O
multiple	O
parts	O
of	O
speech	O
,	O
a	O
probability	O
is	O
assigned	O
to	O
each	O
possible	O
part	O
.	O
</s>
<s>
Some	O
stemming	B-General_Concept
techniques	O
use	O
the	O
n-gram	B-Language
context	O
of	O
a	O
word	O
to	O
choose	O
the	O
correct	O
stem	O
for	O
a	O
word	O
.	O
</s>
<s>
A	O
simple	O
example	O
is	O
a	O
suffix	O
tree	O
algorithm	O
which	O
first	O
consults	O
a	O
lookup	B-Data_Structure
table	I-Data_Structure
using	O
brute	O
force	O
.	O
</s>
<s>
However	O
,	O
instead	O
of	O
trying	O
to	O
store	O
the	O
entire	O
set	O
of	O
relations	O
between	O
words	O
in	O
a	O
given	O
language	O
,	O
the	O
lookup	B-Data_Structure
table	I-Data_Structure
is	O
kept	O
small	O
and	O
is	O
only	O
used	O
to	O
store	O
a	O
minute	O
amount	O
of	O
"	O
frequent	O
exceptions	O
"	O
like	O
"	O
ran	O
=>	O
run	O
"	O
.	O
</s>
<s>
If	O
the	O
word	O
is	O
not	O
in	O
the	O
exception	O
list	O
,	O
apply	O
suffix	O
stripping	O
or	O
lemmatisation	B-General_Concept
and	O
output	O
the	O
result	O
.	O
</s>
<s>
A	O
study	O
of	O
affix	O
stemming	B-General_Concept
for	O
several	O
European	O
languages	O
can	O
be	O
found	O
here.Jongejan	O
,	O
B.	O
;	O
and	O
Dalianis	O
,	O
H.	O
;	O
Automatic	O
Training	O
of	O
Lemmatization	B-General_Concept
Rules	O
that	O
Handle	O
Morphological	O
Changes	O
in	O
pre-	O
,	O
in	O
-	O
and	O
Suffixes	O
Alike	O
,	O
in	O
the	O
Proceedings	O
of	O
the	O
ACL-2009	O
,	O
Joint	O
conference	O
of	O
the	O
47th	O
Annual	O
Meeting	O
of	O
the	O
Association	O
for	O
Computational	O
Linguistics	O
and	O
the	O
4th	O
International	O
Joint	O
Conference	O
on	O
Natural	B-Language
Language	I-Language
Processing	I-Language
of	O
the	O
Asian	O
Federation	O
of	O
Natural	B-Language
Language	I-Language
Processing	I-Language
,	O
Singapore	O
,	O
August	O
2	O
–	O
7	O
,	O
2009	O
,	O
pp	O
.	O
</s>
<s>
While	O
much	O
of	O
the	O
early	O
academic	O
work	O
in	O
this	O
area	O
was	O
focused	O
on	O
the	O
English	O
language	O
(	O
with	O
significant	O
use	O
of	O
the	O
Porter	B-General_Concept
Stemmer	I-General_Concept
algorithm	I-General_Concept
)	O
,	O
many	O
other	O
languages	O
have	O
been	O
investigated.Dolamic	O
,	O
Ljiljana	O
;	O
and	O
Savoy	O
,	O
Jacques	O
;	O
Stemming	B-General_Concept
Approaches	O
for	O
East	O
European	O
Languages	O
(	O
CLEF	O
2007	O
)	O
Savoy	O
,	O
Jacques	O
;	O
Light	O
Stemming	B-General_Concept
Approaches	O
for	O
the	O
French	O
,	O
Portuguese	O
,	O
German	O
and	O
Hungarian	O
Languages	O
,	O
ACM	O
Symposium	O
on	O
Applied	O
Computing	O
,	O
SAC	O
2006	O
,	O
Popovič	O
,	O
Mirko	O
;	O
and	O
Willett	O
,	O
Peter	O
(	O
1992	O
)	O
;	O
The	O
Effectiveness	O
of	O
Stemming	B-General_Concept
for	O
Natural-Language	O
Access	O
to	O
Slovene	O
Textual	O
Data	O
,	O
Journal	O
of	O
the	O
American	O
Society	O
for	O
Information	O
Science	O
,	O
Volume	O
43	O
,	O
Issue	O
5	O
(	O
June	O
)	O
,	O
pp	O
.	O
</s>
<s>
Hebrew	O
and	O
Arabic	O
are	O
still	O
considered	O
difficult	O
research	O
languages	O
for	O
stemming	B-General_Concept
.	O
</s>
<s>
Multilingual	O
stemming	B-General_Concept
applies	O
morphological	O
rules	O
of	O
two	O
or	O
more	O
languages	O
simultaneously	O
instead	O
of	O
rules	O
for	O
only	O
a	O
single	O
language	O
when	O
interpreting	O
a	O
search	O
query	B-Library
.	O
</s>
<s>
Commercial	O
systems	O
using	O
multilingual	O
stemming	B-General_Concept
exist	O
.	O
</s>
<s>
There	O
are	O
two	O
error	O
measurements	O
in	O
stemming	B-General_Concept
algorithms	I-General_Concept
,	O
overstemming	O
and	O
understemming	O
.	O
</s>
<s>
Stemming	B-General_Concept
algorithms	I-General_Concept
attempt	O
to	O
minimize	O
each	O
type	O
of	O
error	O
,	O
although	O
reducing	O
one	O
type	O
can	O
lead	O
to	O
increasing	O
the	O
other	O
.	O
</s>
<s>
For	O
example	O
,	O
the	O
widely	O
used	O
Porter	B-General_Concept
stemmer	I-General_Concept
stems	O
"	O
universal	O
"	O
,	O
"	O
university	O
"	O
,	O
and	O
"	O
universe	O
"	O
to	O
"	O
univers	O
"	O
.	O
</s>
<s>
This	O
is	O
a	O
case	O
of	O
overstemming	O
:	O
though	O
these	O
three	O
words	O
are	O
etymologically	O
related	O
,	O
their	O
modern	O
meanings	O
are	O
in	O
widely	O
different	O
domains	O
,	O
so	O
treating	O
them	O
as	O
synonyms	B-Application
in	O
a	O
search	B-Application
engine	I-Application
will	O
likely	O
reduce	O
the	O
relevance	O
of	O
the	O
search	B-Application
results	I-Application
.	O
</s>
<s>
An	O
example	O
of	O
understemming	O
in	O
the	O
Porter	B-General_Concept
stemmer	I-General_Concept
is	O
"	O
alumnus	O
"	O
→	O
"	O
alumnu	O
"	O
,	O
"	O
alumni	O
"	O
→	O
"	O
alumni	O
"	O
,	O
"	O
alumna	O
"	O
/	O
"	O
alumnae	O
"	O
→	O
"	O
alumna	O
"	O
.	O
</s>
<s>
This	O
English	O
word	O
keeps	O
Latin	O
morphology	O
,	O
and	O
so	O
these	O
near-synonyms	O
are	O
not	O
conflated	O
.	O
</s>
<s>
Stemming	B-General_Concept
is	O
used	O
as	O
an	O
approximate	O
method	O
for	O
grouping	O
words	O
with	O
a	O
similar	O
basic	O
meaning	O
together	O
.	O
</s>
<s>
Stemmers	O
are	O
common	O
elements	O
in	O
query	B-Library
systems	I-Library
such	O
as	O
Web	B-Application
search	I-Application
engines	I-Application
.	O
</s>
<s>
The	O
effectiveness	O
of	O
stemming	B-General_Concept
for	O
English	O
query	B-Library
systems	I-Library
were	O
soon	O
found	O
to	O
be	O
rather	O
limited	O
,	O
however	O
,	O
and	O
this	O
has	O
led	O
early	O
information	B-Library
retrieval	I-Library
researchers	O
to	O
deem	O
stemming	B-General_Concept
irrelevant	O
in	O
general.Baeza-Yates	O
,	O
Ricardo	O
;	O
and	O
Ribeiro-Neto	O
,	O
Berthier	O
(	O
1999	O
)	O
;	O
Modern	O
Information	B-Library
Retrieval	I-Library
,	O
ACM	O
Press/Addison	O
Wesley	O
An	O
alternative	O
approach	O
,	O
based	O
on	O
searching	O
for	O
n-grams	B-Language
rather	O
than	O
stems	O
,	O
may	O
be	O
used	O
instead	O
.	O
</s>
<s>
Stemming	B-General_Concept
is	O
used	O
to	O
determine	O
domain	O
vocabularies	O
in	O
domain	O
analysis.Frakes	O
,	O
W.	O
;	O
Prieto-Diaz	O
,	O
R.	O
;	O
&	O
Fox	O
,	O
C	O
.	O
(	O
1998	O
)	O
.	O
</s>
<s>
"	O
DARE	O
:	O
Domain	B-General_Concept
Analysis	I-General_Concept
and	O
Reuse	O
Environment	O
"	O
,	O
Annals	O
of	O
Software	O
Engineering	O
(	O
5	O
)	O
,	O
pp	O
.	O
</s>
<s>
Google	B-Application
Search	I-Application
adopted	O
word	B-General_Concept
stemming	I-General_Concept
in	O
2003.The	O
Essentials	O
of	O
Google	B-Application
Search	I-Application
,	O
Web	B-Application
Search	I-Application
Help	O
Center	O
,	O
Google	B-Application
Inc	I-Application
.	I-Application
</s>
<s>
Other	O
software	O
search	O
algorithms	O
vary	O
in	O
their	O
use	O
of	O
word	B-General_Concept
stemming	I-General_Concept
.	O
</s>
<s>
Stemming	B-General_Concept
is	O
used	O
as	O
a	O
task	O
in	O
pre-processing	O
texts	O
before	O
performing	O
text	B-Algorithm
mining	I-Algorithm
analyses	O
on	O
it	O
.	O
</s>
<s>
(	O
1992	O
)	O
;	O
Stemming	B-General_Concept
algorithms	I-General_Concept
,	O
Information	B-Library
retrieval	I-Library
:	O
data	O
structures	O
and	O
algorithms	O
,	O
Upper	O
Saddle	O
River	O
,	O
NJ	O
:	O
Prentice-Hall	O
,	O
Inc	O
.	O
</s>
<s>
Popovič	O
,	O
Mirko	O
;	O
and	O
Willett	O
,	O
Peter	O
(	O
1992	O
)	O
;	O
The	O
Effectiveness	O
of	O
Stemming	B-General_Concept
for	O
Natural-Language	O
Access	O
to	O
Slovene	O
Textual	O
Data	O
,	O
Journal	O
of	O
the	O
American	O
Society	O
for	O
Information	O
Science	O
,	O
Volume	O
43	O
,	O
Issue	O
5	O
(	O
June	O
)	O
,	O
pp	O
.	O
</s>
