DENOTE
ITALICIS

DENOTE
SUPERSCRIPT

AN
ALGORITHM
FOR
SUFFIX
STRIP
MFPORTER
1980
ORIGINALLY
PUBLISH
IN
PROGRAM
NO
3
PP
130137
JULY
1980
1
INTRODUCT
REMOV
SUFFIXE
BY
AUTOMAT
MEAN
I
AN
OPER
WHICH
I
ESPECIALLY
USE
IN
THE
FIELD
OF
INFORM
RETRIEV
IN
A
TYPIC
IR
ENVIRON
ONE
HA
A
COLLECT
OF
DOCUM
EACH
DESCRIB
BY
THE
WORD
IN
THE
DOCUM
TITLE
AND
POSSIBLY
BY
WORD
IN
THE
DOCUM
ABSTRACT
IGNOR
THE
ISSUE
OF
PRECISELY
WHERE
THE
WORD
ORIGIN
WE
CAN
SAY
THAT
A
DOCUM
I
REPRES
BY
A
VETOR
OF
WORD
OR
ERM
TERM
WITH
A
COMMON
STEM
WILL
USUALLY
HAVE
SIMILAR
MEAN
FOR
EXAMPLE
CONNECT
CONNECT
CONNECT
CONNECT
CONNECT
FREQUENTLY
THE
PERFORM
OF
AN
IR
SYSTEM
WILL
BE
IMPROV
IF
TERM
GROUP
SUCH
A
THI
ARE
CONFLATE
INTO
A
SINGLE
TERM
THI
MAY
BE
DONE
BY
REMOV
OF
THE
VARIOU
SUFFIXE
ED
ING
ION
ION
TO
LEAVE
THE
SINGLE
TERM
CONNECT
IN
ADDIT
THE
SUFFIX
STRIP
PROCESS
WILL
REDUCE
THE
TOTAL
NUMBER
OF
TERM
IN
THE
IR
SYSTEM
AND
HENCE
REDUCE
THE
SIZE
AND
COMPLEXITY
OF
THE
DATA
IN
THE
SYSTEM
WHICH
I
ALWAY
ADVANTAGE
MANY
STRATEGI
FOR
SUFFIX
STRIP
HAVE
BEEN
REPORT
IN
THE
LITERATUREEG
16
THE
NATURE
OF
THE
TASK
WILL
VARY
CONSIDERABLY
DEPEND
ON
WHETHER
A
STEM
DICTIONARY
I
BE
US
WHETHER
A
SUFFIX
LIST
I
BE
US
AND
OF
COURSE
ON
THE
PURPOSE
FOR
WHICH
THE
SUFFIX
STRIP
I
BE
DONE
ASSUM
THAT
ONE
I
NOT
MAKE
USE
OF
A
STEM
DICTIONARY
AND
THAT
THE
PURPOSE
OF
THE
TASK
I
TO
IMPROVE
IR
PERFORM
THE
SUFFIX
STRIP
PROGRAM
WILL
USUALLY
BE
GIVEN
AN
EXPLICIT
LIST
OF
SIFFIXE
AND
WITH
EACH
SUFFIX
THE
CRITERION
UNDER
WHICH
IT
MAY
BE
REMOV
FROM
A
WORD
TO
LEAVE
A
VALID
STEM
THI
I
THE
APPROACH
ADOPT
HERE
THE
MAIN
MERIT
OF
THE
PRESENT
PROGRAM
ARE
THAT
IT
I
SMALL
LESS
THAN
400
LINE
OF
BCPL
FAST
IT
WILL
PROCESS
A
VOCABULARY
OF
10000
DIFFER
WORD
IN
ABOUT
81
SECOND
ON
THE
IBM
370165
AT
CAMBRIDGE
UNIVERSITY
AND
REASONABLY
SIMPLE
AT
ANY
RATE
IT
I
SIMPLE
ENOUGH
TO
BE
DESCRIB
IN
FULL
A
AN
ALGORITHM
IN
THI
PAPER
THE
PRESENT
VERSION
IN
BCPL
I
FREELY
AVAIL
FROM
THE
AUTHOR
BCPL
I
ITSELF
AVAIL
ON
A
WIDE
RANGE
OF
DIFFER
COMPUT
BUT
ANYONE
WISH
TO
USE
THE
PROGRAM
SHOULD
HAVE
LITTLE
DIFFICULTY
IN
CODE
IT
UP
IN
OTHER
PROGRAM
LANGUAGE
GIVEN
THE
SPEED
OF
THE
PROGRAM
IT
WOULD
BE
QUITE
REALIST
TO
APPLY
IT
TO
EVERY
WORD
IN
A
LARGE
FILE
OF
CONTINU
TEXT
ALTHOUGH
FOR
HISTOR
REASON
WE
HAVE
FOUND
IT
CONVENI
TO
APPLY
IT
ONLY
TO
RELATIVELY
SMALL
VOCABULARY
LIST
DERIV
FROM
CONTINU
TEXT
FILE
IN
ANY
SUFFIX
STRIP
PROGRAM
FOR
IR
WORK
TWO
POINT
MUST
BE
BORNE
IN
MIND
FIRSTLY
THE
SUFFIXE
ARE
BE
REMOV
SIMPLY
TO
IMPROVE
IR
PERFORM
AND
NOT
A
A
LINGUIST
EXERCISE
THI
MEAN
THAT
IT
WOULD
NOT
BE
AT
ALL
OBVIOU
UNDER
WHAT
CIRCUMST
A
SUFFIX
SHOULD
BE
REMOV
EVEN
IF
WE
COULD
EXACTLY
DETERMINE
THE
SUFFIXE
OF
A
WORD
BY
AUTOMAT
MEAN
PERHAP
THE
BEST
CRITERION
FOR
REMOV
SUFFIXE
FROM
TWO
WORD
W1
AND
W2
TO
PRODUCE
A
SINGLE
STEM

I
TO
SAY
THAT
WE
DO
SO
IF
THERE
APPEAR
TO
BE
NO
DIFFER
BETWEEN
THE
TWO
STATEM
A
DOCUM
I
ABOUT
W1
AND
A
DOCUM
I
ABOUT
W2
SO
IF
W1CONNECT
AND
W2CONNECT
IT
SEEM
VERY
REASON
TO
CONFLATE
THEM
TO
A
SINGLE
STEM
BUT
IF
W1RELATE
AND
W2RELATIVITY
IT
SEEM
PERHAP
UNREASON
ESPECIALLY
IF
THE
DOCUM
COLLECT
I
CONCERN
WITH
THEORET
PHYSIC
IT
SHOULD
PERHAP
BE
AD
THAT
RELATE
AND
RELATIVITY
RECONFL
TOGETH
IN
THE
ALGORITHM
DESCRIB
HERE
BETWEEN
THESE
TWO
EXTREME
THERE
I
A
CONTINUUM
OF
DIFFER
CASE
AND
GIVEN
TWO
TERM
W1
AND
W2
THERE
WILL
BE
SOME
VARIATE
IN
OPINION
A
TO
WHETHER
THEY
SHOULD
BE
CONFLATE
JUST
A
THERE
I
WITH
DECID
THE
RELEV
OF
SOME
DOCUM
TO
A
QUERY
THE
EVALU
OF
THE
WORTH
OF
A
SUFFIX
STRIP
SYSTEM
I
CORRESPONDINGLY
DIFFICULT
THE
SECOND
POINT
I
THAT
WITH
THE
APPROACH
ADOPT
HERE
IE
THE
USE
OF
A
SUFFIX
LIST
WITH
VARIOU
RULE
THE
SUCCESS
RATE
FOR
THE
SUFFIX
STRIP
WILL
BE
SIGNIFICANTLY
LESS
THAN
100
IRRESPECT
OF
HOW
THE
PROCESS
I
EVALU
FOR
EXAMPLE
IF
SAND
AND
SANDER
GET
CONFLATE
SO
MOST
PROBABLY
WILL
WAND
AND
WANDER
THE
ERROR
HERE
I
THAT
THE
ER
OF
WANDER
HA
BEEN
TREATE
A
A
SUFFIX
WHEN
IN
FACT
IT
I
PART
OF
THE
STEM
EQUALLY
A
SUFFIX
MAY
COMPLETELY
ALTER
THE
MEAN
OF
A
WORD
IN
WHICH
CASE
IT
REMOV
I
UNHELP
PROBE
AND
PROBATE
FOR
EXAMPLE
HAVE
QUITE
DISTINCT
MEAN
IN
MODERN
ENGLISH
IN
FACT
THESE
WOULD
NOT
BE
CONFLATE
IN
OUR
PRESENT
ALGORITHM
THERE
COME
A
STAGE
IN
THE
DEVELOP
OF
A
SUFFIX
STRIP
PROGRAM
WHERE
THE
ADDIT
OF
MORE
RULE
TO
INCREASE
THE
PERFORM
IN
ONE
AREA
OF
THE
VOCABULARY
CAUSE
AN
EQUAL
DEGRAD
OF
PERFORM
ELSEWHERE
UNLESS
THI
PHENOMENON
I
NOTIC
IN
TIME
IT
I
VERY
EASY
FOR
THE
PROGRAM
TO
BECOME
MUCH
MORE
COMPLEX
THAN
I
REALLY
NECESSARY
IT
I
ALSO
EASY
TO
GIVE
UNDUE
EMPHASI
TO
CASE
WHICH
APPEAR
TO
BE
IMPORT
BUT
WHICH
TURN
UT
TO
BE
RATHER
RARE
FOR
EXAMPLE
CASE
IN
WHICH
THE
ROOT
OF
A
WORD
CHANGE
WITH
THE
ADDIT
OF
A
SUFFIX
A
IN
DECEIVEDECEPT
RESUMERESUMPT
INDEXINDICE
OCCUR
MUCH
MORE
RARELY
IN
REAL
VOCABULARI
THAN
ONE
MIGHT
AT
FIRST
SUPPOSE
IN
VIEW
OF
THE
ERROR
RATE
THAT
MUST
IN
ANY
CASE
BE
EXPECT
IT
DID
NOT
SEEM
WORTHWHILE
TO
TRY
AND
COPE
WITH
THESE
CASE
IT
I
NOT
OBVIOU
THAT
THE
SIMPLICITY
OF
THE
PRESENT
PROGRAM
I
ANY
DEMERIT
IN
A
TEST
ON
THE
WELLKNOWN
CRANFIELD
200
COLLECTION7
IT
GAVE
AN
IMPROV
IN
RETRIEV
PERFORM
WHEN
COMPAR
WITH
A
VERY
MUCH
MORE
ELABOR
PROGRAM
WHICH
HA
BEEN
IN
USE
IN
IR
RESEARCH
IN
CAMBRIDGE
SINCE
197126
THE
TEST
WA
DONE
A
FOLLOW
THE
WORD
OF
THE
TITLE
AND
ABSTRACT
IN
THE
DOCUM
WERE
PASS
THROUGH
THE
EARLIER
SUFFIX
STRIP
SYSTEM
AND
THE
RESULTI
STEM
WERE
US
TO
INDEX
THE
DOCUM
THE
WORD
OF
THE
QUERI
WERE
REDUC
TO
STEM
IN
THE
SAME
WAY
AND
THE
DOCUM
WERE
RANK
FOR
EACH
QUERY
US
TERM
COORDIN
MATCH
OF
QUERY
AGAINST
DOCUM
FROM
THESE
RANK
RECAL
AND
PRECIS
VALUE
WERE
OBTAIN
US
THE
STANDARD
RECAL
CUTOFF
METHOD
THE
ENTIRE
PROCESS
WA
THEN
REPEATE
US
THE
SUFFIX
STRIP
SYSTEM
DESCRIB
IN
THI
PAPER
AND
THE
RESULT
WERE
A
FOLLOW
EARLIER
SYSTEM
PRESENT
SYSTEM


PRECIS
RECAL
PRECIS
RECAL
0
5724
0
5860
10
5685
10
5813
20
5285
20
5392
30
4261
30
4351
40
4220
40
3939
50
3906
50
3885
60
3286
60
3318
70
3164
70
3119
80
2715
80
2752
90
2459
90
2585
100
2459
100
2585
CLEARY
THE
PERFORM
I
NOT
VERY
DIFFER
THE
IMPORT
POINT
I
THAT
THE
EARLIER
MORE
ELABOR
SYSTEM
CERTAINLY
PERFORM
NO
BETTER
THAN
THE
PRESENT
SIMPLE
SYSTEM
THI
TEST
WA
DONE
BY
PROF
CJ
VAN
RIJSBERGEN
2
THE
ALGORITHM
TO
PRESENT
THE
SUFFIX
STRIP
ALGORITHM
IN
IT
ENTIRETY
WE
WILL
NEED
A
FEW
DIFINIT
A
CONSONANTIN
A
WORD
I
A
LETTER
OTHER
THAN
A
E
I
O
OR
U
AND
OTHER
THAN
Y
PRECED
BY
A
CONSON
THE
FACT
THAT
THE
TERM
CONSON
I
DEFIN
TO
SOME
EXTENT
IN
TERM
OF
ITSELF
DOE
NOT
MAKE
IT
AMBIGU
SO
IN
TOY
THE
CONSON
ARE
T
AND
Y
AND
IN
SYZYGY
THEY
ARE

Z
AND
G
IF
A
LETTER
I
NOT
A
CONSON
IT
I
A
OWEL
A
CONSON
WILL
BE
DENOT
BY
C
A
VOWEL
BY
V
A
LIST
CCC
OF
LENGTH
GREATER
THAN
0
WILL
BE
DENOT
BY
C
AND
A
LIST
VVV
OF
LENGTH
GREATER
THAN
0
WILL
BE
DENOT
BY
V
ANY
WORD
OR
PART
OF
A
WORD
THEREFORE
HA
ONE
OF
THE
FOUR
FORM
CVCV

C
CVCV

V
VCVC

C
VCVC

V
THESE
MAY
ALL
BE
REPRES
BY
THE
SINGLE
FORM
CVCVC

V
WHERE
THE
SQUARE
BRACKET
DENOTE
ARBITRARY
PRESENCE
OF
THEIR
CONTENT
US
VCM
TO
DENOTE
VC
REPEATE
M
TIME
THI
MAY
AGAIN
BE
WRITTEN
A
CVCMV
M
WILL
BE
CALL
THE
MEASUREOF
ANY
WORD
OR
WORD
PART
WHEN
REPRES
IN
THI
FORM
THE
CASE
M

0
COVER
THE
NULL
WORD
HERE
ARE
SOME
EXAMPLE
M0
TR
EE
TREE
Y
BY
M1
TROUBLE
OAT
TREE
IVY
M2
TROUBLE
PRIVATE
OATEN
ORRERY
THE
ULESFOR
REMOV
A
SUFFIX
WILL
BE
GIVEN
IN
THE
FORM
CONDIT
S1

S2
THI
MEAN
THAT
IF
A
WORD
END
WITH
THE
SUFFIX
S1
AND
THE
STEM
BEFORE
S1
SATISFI
THE
GIVEN
CONDIT
S1
I
REPLAC
BY
S2
THE
CONDIT
I
USUALLY
GIVEN
IN
TERM
OF
M
EG
M

1
EMENT

HERE
S1
I
EMENT
AND
S2
I
NULL
THI
WOULD
MAP
REPLAC
TO
REPLAC
SINCE
REPLAC
I
A
WORD
PART
FOR
WHICH
M

2
THE
CONDIT
PART
MAY
ALSO
CONTAIN
THE
FOLLOW


THE
STEM
END
WITH

AND
SIMILARLY
FOR
THE
OTHER
LETTER
V

THE
STEM
CONTAIN
A
VOWEL
D

THE
STEM
END
WITH
A
DOUBLE
CONSON
EG
TT
SS
O

THE
STEM
END
CVC
WHERE
THE
SECOND
C
I
NOT
W
X
OR
Y
EG
WIL
HOP
AND
THE
CONDIT
PART
MAY
ALSO
CONTAIN
EXPRESS
WITH
ND
ORAND
OT
SO
THAT
M1
AND

OR
T
TEST
FOR
A
STEM
WITH
M1
END
IN

OR
T
WHILE
D
AND
NOT
L
OR

OR
Z
TEST
FOR
A
STEM
END
WITHA
DOUBLE
CONSON
OTHER
THAN
L

OR
Z
ELABOR
CONDIT
LIKE
THI
ARE
REQUIR
ONLY
RARELY
IN
A
SET
OF
RULE
WRITTEN
BENEATH
EACH
OTHER
ONLY
ONE
I
OBEY
AND
THI
WILL
BE
THE
ONE
WITH
THE
LONGEST
MATCH
S1
FOR
THE
GIVEN
WORD
FOR
EXAMPLE
WITH
SS

SS
I

I
SS

SS


HERE
THE
CONDIT
ARE
ALL
NULL
CARESS
MAP
TO
CARESS
SINCE
SS
I
THE
LONGEST
MATCH
FOR
S1
EQUALLY
CARESS
MAP
TO
CARESS
S1SS
AND
CARE
TO
CARE
S1
IN
THE
RULE
BELOW
EXAMPLE
OF
THEIR
APPLIC
SUCCESS
OR
OTHERWISE
ARE
GIVEN
ON
THE
RIGHT
IN
LOWER
CASE
THE
ALGORITHM
NOW
FOLLOW
STEP
1A
SS

SS
CARESS

CARESS
I

I
PONI

PONI
TI

TI
SS

SS
CARESS

CARESS


CAT

CAT
STEP
1B
M0
EED

E
FEED

FEED
AGRE

AGREE
V
ED

PLASTER

PLASTER
BLED

BLED
V
ING

MOTOR

MOTOR
SING

SING
IF
THE
SECOND
OR
THIRD
OF
THE
RULE
IN
STEP
1B
I
SUCCESS
THE
FOLLOW
I
DONE
AT

ATE
CONFLATE

CONFLATE
BL

BLE
TROUBLE

TROUBLE
IZ

IZE
SIZE

SIZE
D
AND
NOT
L
OR

OR
Z

SINGLE
LETTER
HOP

HOP
TAN

TAN
FALL

FALL
HISS

HISS
FIZZ

FIZZ
M1
AND
O

E
FAIL

FAIL
FILE

FILE
THE
RULE
TO
MAP
TO
A
SINGLE
LETTER
CAUSE
THE
REMOV
OF
ONE
OF
THE
DOUBLE
LETTER
PAIR
THE
E
I
PUT
BACK
ON
AT
BL
AND
IZ
SO
THAT
THE
SUFFIXE
ATE
BLE
AND
IZE
CAN
BE
RECOGNISE
LATER
THI
E
MAY
BE
REMOV
IN
STEP
4
STEP
1C
V
Y

I
HAPPY

HAPPI
SKY

SKY
STEP
1
DEAL
WITH
PLURAL
AND
PAST
PARTICIPLE
THE
SUBSEQU
STEP
ARE
MUCH
MORE
STRAIGHTFORWARD
STEP
2
M0
ATION

ATE
RELATE

RELATE
M0
TIONAL

TION
CONDIT

CONDIT
RATION

RATION
M0
ENCI

ENCE
VALENCE

VALENCE
M0
ANCI

ANCE
HESIT

HESIT
M0
IZER

IZE
DIGIT

DIGIT
M0
ABLI

ABLE
CONFORM

CONFORM
M0
ALLI

AL
RADIC

RADIC
M0
ENTLI

ENT
DIFFER

DIFFER
M0
ELI

E
VILE


VILE
M0
OUSLI

OU
ANALOG

ANALOG
M0
IZATE

IZE
VIETNAM

VIETNAM
M0
ATION

ATE
PREDIC

PREDIC
M0
ATOR

ATE
OPER

OPER
M0
ALISM

AL
FEUDAL

FEUDAL
M0
IVE

IVE
DECIS

DECIS
M0
FUL

FUL
HOPE

HOPE
M0
OUS

OU
CALLOUS

CALLOU
M0
ALITI

AL
FORMAL

FORMAL
M0
IVITI

IVE
SENSIT

SENSIT
M0
BILITI

BLE
SENSIBLE

SENSIBLE
THE
TEST
FOR
THE
STRING
S1
CAN
BE
MADE
FAST
BY
DO
A
PROGRAM
SWITCH
ON
THE
PENULTIM
LETTER
OF
THE
WORD
BE
TEST
THI
GIVE
A
FAIRLY
EVEN
BREAKDOWN
OF
THE
POSSIBLE
VALUE
OF
THE
STRING
S1
IT
WILL
BE
SEEN
IN
FACT
THAT
THE
S1STRING
IN
STEP
2
ARE
PRESENT
HERE
IN
THE
ALPHABET
ORDER
OF
THEIR
PENULTIM
LETTER
SIMILAR
TECHNIQUE
MAY
BE
APPLI
IN
THE
OTHER
STEP
STEP
3
M0
ICATE

IC
TRIPLIC

TRIPLIC
M0
ATIVE

FORM

FORM
M0
ALIZE

AL
FORMAL

FORMAL
M0
ICITI

IC
ELECTR

ELECTR
M0
ICAL

IC
ELECTR

ELECTR
M0
FUL

HOPE

HOPE
M0
NESS

GOOD

GOOD
STEP
4
M1
AL

REVIV

REVIV
M1
ANCE

ALLOW

ALLOW
M1
ENCE

INFER

INFER
M1
ER

AIRLIN

AIRLIN
M1
IC

GYROSCOP

GYROSCOP
M1
ABLE

ADJUST

ADJUST
M1
IBLE

DEFENS

DEFEN
M1
ANT

IRRIT

IRRIT
M1
EMENT

REPLAC

REPLAC
M1
MENT

ADJUST

ADJUST
M1
ENT

DEPEND

DEPEND
M1
AND

OR
T
ION

ADOPT

ADOPT
M1
OU

HOMOLOG

HOMOLOG
M1
ISM

COMMUN

COMMUN
M1
ATE

ACTIV

ACTIV
M1
ITI

ANGULAR

ANGULAR
M1
OU

HOMOLOG

HOMOLOG
M1
IVE

EFFECT

EFFECT
M1
IZE

BOWDLER

BOWDLER
THE
SUFFIXE
ARE
NOW
REMOV
ALL
THAT
REMAIN
I
A
LITTLE
TIDY
UP
STEP
5A
M1
E

PROBATE

PROBAT
RATE

RATE
M1
AND
NOT
O
E

CEASE

CEA
STEP
5B
M

1
AND
D
AND
L

SINGLE
LETTER
CONTROL

CONTROL
ROLL

ROLL
THE
ALGORITHM
I
CARE
NOT
TO
REMOVE
A
SUFFIX
WHEN
THE
STEM
I
TOO
SHORT
THE
LENGTH
OF
THE
STEM
BE
GIVEN
BY
IT
MEASURE
M
THERE
I
NO
LINGUIST
BASI
FOR
THI
APPROACH
IT
WA
MERELY
OBSERV
THAT
M
COULD
BE
US
QUITE
EFFECTIVELY
TO
HELP
DECIDE
WHETHER
OR
NOT
IT
WA
WISE
TO
TAKE
OFF
A
SUFFIX
FOR
EXAMPLE
IN
THE
FOLLOW
TWO
LIST
LIST
A
LIST
B


RELATE
DERIV
PROBATE
ACTIV
CONFLATE
DEMONSTR
PIRATE
NECESSIT
PRELATE
RENOV
ATE
I
REMOV
FROM
THE
LIST
B
WORD
BUT
NOT
FROM
THE
LIST
A
WORD
THI
MEAN
THAT
THE
PAIR
DERIVATEDER
ACTIVATEACT
DEMONSTRATEDEMON
TRABLE
NECESSITATENECESSIT
WILL
CONFLATE
TOGETH
THE
FACT
THAT
NO
ATTEMPT
I
MADE
TO
IDENTIFY
PREFIXE
CAN
MAKE
THE
RESULT
LOOK
RATHER
INCONSIST
THU
PRELATE
DOE
NOT
LOSE
THE
ATE
BUT
ARCHPREL
BECOME
ARCHPREL
IN
PRACTICE
THI
DOE
NOT
MATTER
TOO
MUCH
BECAUSE
THE
PRESENCE
OF
THE
PREFIX
DECREASE
THE
PROBABILITY
OF
AN
ERRONE
CONFLATE
COMPLEX
SUFFIXE
ARE
REMOV
BIT
BY
BIT
IN
THE
DIFFER
STEP
THU
GENER
I
STRIP
TO
GENER
STEP
1
THEN
TO
GENER
STEP
2
THEN
TO
GENER
STEP
3
AND
THEN
TO
GENER
STEP
4
OSCIL
I
STRIP
TO
OSCIL
STEP
1
THEN
TO
OSCIL
STEP
2
THEN
TO
OSCIL
STEP
4
AND
THEN
TO
OSCIL
STEP
5
IN
A
VOCABULARY
OF
10000
WORD
THE
REDUCT
IN
SIZE
OF
THE
STEM
WA
DISTRIBUT
AMONG
THE
STEP
A
FOLLOW
SUFFIX
STRIP
OF
A
VOCABULARY
OF
10000
WORD

NUMBER
OF
WORD
REDUC
IN
STEP
1
3597
2766
3
327
42424
5
1373
NUMBER
OF
WORD
NOT
REDUC
3650
THE
RESULT
VOCABULARY
OF
STEM
CONTAIN
6370
DISTINCT
ENTRI
THU
THE
SUFFIX
STRIP
PROCESS
REDUC
THE
SIZE
OF
THE
VOCABULARY
BY
ABOUT
ONE
THIRD
REFER
1
LOVIN
JB
DEVELOP
OF
A
STEM
ALGORITHM
MECHAN
TRANSLATE
AND
COMPUT
LINGUIST
1
MARCH
1968
PP
2331
2
ANDREW
K
THE
DEVELOP
OF
A
FAST
CONFLATE
ALGORITHM
FOR
ENGLISH
DISSERT
FOR
THE
DIPLOMA
IN
COMPUT
SCIENCE
COMPUT
LABORATORY
UNIVERSITY
OF
CAMBRIDGE
1971
3
PETRARCA
AE
AND
LAY
WM
USE
OF
AN
AUTOMATICALLY
GENER
AUTHORITY
LIST
TO
ELIMIN
SCATTER
CAUS
BY
SOME
SINGULAR
AND
PLURAL
MAIN
INDEX
TERM
PROCEED
OF
THE
AMERICAN
SOCIETY
FOR
INFORM
SCIENCE
1969
PP
277282
4
DATTOLA
ROBERT
T
FIRST
FLEXIBLE
INFORM
RETRIEV
SYSTEM
FOR
TEXT
WEBSTER
NY
XEROX
CORPOR
12
DEC
1975
5
COLOMBO
D
AND
NIEHOFF
RT
FINAL
REPORT
ON
IMPROV
ACCESS
TO
SCIENTIF
AND
TECHNIC
INFORM
THROUGH
AUTOM
VOCABULARY
SWITCHINGNSF
GRANT
NO
SIS7512924
TO
THE
NATION
SCIENCE
FOUNDATE
6
DAWSON
JL
SUFFIX
REMOV
AND
WORD
CONFLATE
ALLC
BULLETIN
MICHAELMA
1974
P3346
7
CLEVERDON
CW
MILL
J
AND
KEEN
M
FACTOR
DETERMIN
THE
PERFORM
OF
INDEX
SYSTEMS2
VOL
COLLEGE
OF
AERONAUT
CRANFIELD
1966