The Porter stemming algorithm

Links to resources

Here is a case study on how to code up a stemming algorithm in Snowball. First, the definition of the Porter stemmer, as it appeared in Program, Vol 14 no. 3 pp 130-137, July 1980.

THE ALGORITHM

A consonant in a word is a letter other than A, E, I, O or U, and other than Y preceded by a consonant. (The fact that the term ‘consonant’ is defined to some extent in terms of itself does not make it ambiguous.) So in TOY the consonants are T and Y, and in SYZYGY they are S, Z and G. If a letter is not a consonant it is a vowel.

A consonant will be denoted by c, a vowel by v. A list ccc... of length greater than 0 will be denoted by C, and a list vvv... of length greater than 0 will be denoted by V. Any word, or part of a word, therefore has one of the four forms:

CVCV ... C
CVCV ... V
VCVC ... C
VCVC ... V

These may all be represented by the single form

[C]VCVC ... [V]

where the square brackets denote arbitrary presence of their contents. Using (VC)^m to denote VC repeated m times, this may again be written as

[C](VC)^m[V].

m will be called the measure of any word or word part when represented in this form. The case m = 0 covers the null word. Here are some examples:

m=0		TR, EE, TREE, Y, BY.
m=1		TROUBLE, OATS, TREES, IVY.
m=2		TROUBLES, PRIVATE, OATEN, ORRERY.

The rules for removing a suffix will be given in the form

(condition) S1 -> S2

This means that if a word ends with the suffix S1, and the stem before S1 satisfies the given condition, S1 is replaced by S2. The condition is usually given in terms of m, e.g.

(m > 1) EMENT ->

Here S1 is ‘EMENT’ and S2 is null. This would map REPLACEMENT to REPLAC, since REPLAC is a word part for which m = 2.

The ‘condition’ part may also contain the following:

*S	-	the stem ends with S (and similarly for the other letters).
v	-	the stem contains a vowel.
*d	-	the stem ends with a double consonant (e.g. -TT, -SS).
*o	-	the stem ends cvc, where the second c is not W, X or Y (e.g. -WIL, -HOP).

And the condition part may also contain expressions with and, or and not, so that

(m>1 and (*S or *T))

tests for a stem with m>1 ending in S or T, while

(*d and not (*L or *S or *Z))

tests for a stem ending with a double consonant other than L, S or Z. Elaborate conditions like this are required only rarely.

In a set of rules written beneath each other, only one is obeyed, and this will be the one with the longest matching S1 for the given word. For example, with

SSES	`->`	SS
IES	`->`	I
SS	`->`	SS
S	`->`

(here the conditions are all null) CARESSES maps to CARESS since SSES is the longest match for S1. Equally CARESS maps to CARESS (S1=‘SS’) and CARES to CARE (S1=‘S’).

In the rules below, examples of their application, successful or otherwise, are given on the right in lower case. The algorithm now follows:

Step 1a

SSES -> SS caresses -> caress

IES -> I ponies -> poni

ties -> ti

SS -> SS caress -> caress

S -> cats -> cat

Step 1b

(m>0) EED -> EE feed -> feed

agreed -> agree

(*v*) ED -> plastered -> plaster

bled -> bled

(*v*) ING -> motoring -> motor

sing -> sing

If the second or third of the rules in Step 1b is successful, the following is done:

AT -> ATE conflat(ed) -> conflate

BL -> BLE troubl(ed) -> trouble

IZ -> IZE siz(ed) -> size

(*d and not (*L or *S or *Z)) -> single letter hopp(ing) -> hop

tann(ed) -> tan

fall(ing) -> fall

hiss(ing) -> hiss

fizz(ed) -> fizz

(m=1 and *o) -> E fail(ing) -> fail

fil(ing) -> file

The rule to map to a single letter causes the removal of one of the double letter pair. The -E is put back on -AT, -BL and -IZ, so that the suffixes -ATE, -BLE and -IZE can be recognised later. This E may be removed in step 4.

Step 1c

(*v*) Y -> I happy -> happi

sky -> sky

Step 1 deals with plurals and past participles. The subsequent steps are much more straightforward.

Step 2

(m>0) ATIONAL -> ATE relational -> relate

(m>0) TIONAL -> TION conditional -> condition

rational -> rational

(m>0) ENCI -> ENCE valenci -> valence

(m>0) ANCI -> ANCE hesitanci -> hesitance

(m>0) IZER -> IZE digitizer -> digitize

(m>0) ABLI -> ABLE conformabli -> conformable

(m>0) ALLI -> AL radicalli -> radical

(m>0) ENTLI -> ENT differentli -> different

(m>0) ELI -> E vileli -> vile

(m>0) OUSLI -> OUS analogousli -> analogous

(m>0) IZATION -> IZE vietnamization -> vietnamize

(m>0) ATION -> ATE predication -> predicate

(m>0) ATOR -> ATE operator -> operate

(m>0) ALISM -> AL feudalism -> feudal

(m>0) IVENESS -> IVE decisiveness -> decisive

(m>0) FULNESS -> FUL hopefulness -> hopeful

(m>0) OUSNESS -> OUS callousness -> callous

(m>0) ALITI -> AL formaliti -> formal

(m>0) IVITI -> IVE sensitiviti -> sensitive

(m>0) BILITI -> BLE sensibiliti -> sensible

The test for the string S1 can be made fast by doing a program switch on the penultimate letter of the word being tested. This gives a fairly even breakdown of the possible values of the string S1. It will be seen in fact that the S1-strings in step 2 are presented here in the alphabetical order of their penultimate letter. Similar techniques may be applied in the other steps.

Step 3

(m>0) ICATE -> IC triplicate -> triplic

(m>0) ATIVE -> formative -> form

(m>0) ALIZE -> AL formalize -> formal

(m>0) ICITI -> IC electriciti -> electric

(m>0) ICAL -> IC electrical -> electric

(m>0) FUL -> hopeful -> hope

(m>0) NESS -> goodness -> good

Step 4

(m>1) AL -> revival -> reviv

(m>1) ANCE -> allowance -> allow

(m>1) ENCE -> inference -> infer

(m>1) ER -> airliner -> airlin

(m>1) IC -> gyroscopic -> gyroscop

(m>1) ABLE -> adjustable -> adjust

(m>1) IBLE -> defensible -> defens

(m>1) ANT -> irritant -> irrit

(m>1) EMENT -> replacement -> replac

(m>1) MENT -> adjustment -> adjust

(m>1) ENT -> dependent -> depend

(m>1 and (*S or *T)) ION -> adoption -> adopt

(m>1) OU -> homologou -> homolog

(m>1) ISM -> communism -> commun

(m>1) ATE -> activate -> activ

(m>1) ITI -> angulariti -> angular

(m>1) OUS -> homologous -> homolog

(m>1) IVE -> effective -> effect

(m>1) IZE -> bowdlerize -> bowdler

The suffixes are now removed. All that remains is a little tidying up.

Step 5a

(m>1) E -> probate -> probat

rate -> rate

(m=1 and not *o) E -> cease -> ceas

Step 5b

(m > 1 and *d and *L) -> single letter controll -> control

roll -> roll

Now, turning it into Snowball.

The Porter stemmer makes a use of a measure, m, of the length of a word or word part. If C is a sequence of one or more consonants, and V a sequence of one or more vowels, any word part has the form

[C](VC)^m[V],

which is to be read as an optional C, followed by m repetitions of VC, followed by an optional V. This defines m. So for crepuscular the measure would be 4.

    c r e p u s c u l a r
       |   |     |   |   |
    [C] V C  V C  V C V C
         1    2    3   4

Most of the rules for suffix removal involve leaving behind a stem whose measure exceeds some value, for example,

(m > 0) eed -> ee

means ‘replace eed with ee if the stem before eed has measure m > 0’. Implementations of the Porter stemmer usually have a routine that computes m each time there is a possible candidate for removal.

In fact the only tests on m in the Porter stemmer are m > 0, m > 1, and, at two interesting points, m = 1. This suggests that there are two critical positions in a word: the point at which, going from left to right, m > 0 becomes true, and then the point at which m > 1 becomes true. It turns out that m > 0 becomes true at the point after the first consonant following a vowel, and m > 1 becomes true at the point after the first consonant following a vowel following a consonant following a vowel. Calling these positions p1 and p2, we can determine them quite simply in Snowball:

    define v 'aeiouy'

    ....

    do(
        gopast v  gopast non-v  setmark p1
        gopast v  gopast non-v  setmark p2
    )

The region to the right of p1 will be denoted by R1, the region to the right of p2 by R2:

    c r e p u s c u l a r
           |   |
           p1  p2
           <---  R1  --->
               <-- R2 -->

We can test for being in these regions with calls to R1 and R2, defined by,

    define R1 as $p1 <= cursor
    define R2 as $p2 <= cursor

and using these tests instead of computing m is acceptable, so long as the stemming process never alters the p1 and p2 positions, which is indeed true in the Porter stemmer.

A particularly interesting feature of the stemmers presented here is the common use they make of the positions p1 and p2. The details of marking p1 and p2 vary between the languages because the definitions of vowel and consonant vary. For example, French i preceded and followed by vowel should be treated as a consonant (inquiétude); Portuguese (ã and õ should be treated as a vowel-consonant pair (São João). A third important position is pV, which tries to mark the position of the shortest acceptable verb stem. Its definition varies somewhat between languages. The Porter stemmer does not use a pV explicitly, but the idea appears when the verb endings ing and ed are removed only when preceded by a vowel. In English therefore pV would be defined as the position after the first vowel.

The Porter stemmer is divided into five steps, step 1 is divided further into steps 1a, 1b and 1c, and step 5 into steps 5a and 5b. Step 1 removes the i-suffixes, and steps 2 to 4 the d-suffixes (*). Composite d-suffixes are reduced to single d-suffixes one at a time. So for example if a word ends icational, step 2 reduces it to icate and step 3 to ic. Three steps are sufficient for this process in English. Step 5 does some tidying up.

One can see how easily the stemming rules translate into Snowball by comparing the definition of Step 1a from the 1980 paper,

    Step 1a:
        SSES -> SS                         caresses  ->  caress
        IES  -> I                          ponies    ->  poni
                                           ties      ->  ti
        SS   -> SS                         caress    ->  caress
        S    ->                            cats      ->  cat

with its Snowball equivalent,

    define Step_1a as (
        [substring] among (
            'sses' (<-'ss')
            'ies'  (<-'i')
            'ss'   ()
            's'    (delete)
        )
    )

The word to be stemmed is being scanned right to left from the end. The longest of 'sses', 'ies', 'ss' or 's' is searched for and defined as the slice. (If none are found, Step_1a signals f.) If 'sses' is found, it is replaced by 'ss', and so on. Of course, replacing 'ss' by 'ss' is a dummy action, so we can write

            'ss'   ()

instead of

            'ss'   (<-'ss')

Remember that delete just means <- ''.

The really tricky part of the whole algorithm is step 1b, which may be worth looking at in detail. Here it is, without the example words on the far right,

    Step 1b:
        (m > 0) EED -> EE
        (*v*)   ED  ->
        (*v*)   ING ->

    If the second or third of the rules in Step 1b is successful, the
    following is done:

        AT -> ATE
        BL -> BLE
        IZ -> IZE
        (*d and not (*L or *S or *Z)) -> single letter
        (m = 1 and *o) -> E

The first part of the rule means that eed maps to ee if eed is in R1 (which is equivalent to m > 0), or ed and ing are removed if they are preceded by a vowel. In Snowball this is simply,

    define Step_1b as (
        [substring] among (
            'eed'  (R1 <-'ee')
            'ed'
            'ing'  (test gopast v  delete)
        )
    )

But this must be modified by the second part of the rule. *d indicates a test for double letter consonant — bb, dd etc. *L, *S, *Z are tests for l, s, z. *o is a short vowel test — it is matched by consonant-vowel-consonant, where the consonant on the right is not w, x or y. If the short vowel test is satisfied, m = 1 is equivalent to the cursor being at p1. So the second part of the rule means, map at, bl, iz to ate, ble, ize; map certain double letters to single letters; and add e after a short vowel in words of one syllable.

We first need two extra groupings,

    define v        'aeiouy'
    define v_WXY    v + 'wxY'   // v with 'w', 'x' and 'y'-consonant
    define v_LSZ    v + 'lsz'   // v with 'l', 's', 'z'

and a test for a short vowel,

    define shortv as ( non-v_WXY v non-v )

(The v_WXY test comes first because we are scanning backwards, from right to left.)

The double to single letter map can be done as follows: first define the slice as the next non-v_LSZ and copy it to a string, ch, as a single character,

    strings ( ch )

    ....

    [non-v_LSZ] ->ch

A further test, ch, tests that the next letter of the string is the same as the one in ch, and if this gives signal t, delete deletes the slice,

    [non-v_LSZ] ->ch  ch  delete

Step_1b can then be written like this,

    define Step_1b as (
        [substring] among (
            'eed'  (R1 <-'ee')
            'ed'
            'ing' (
                test gopast v  delete
                (test among('at' 'bl' 'iz')  <+ 'e')
                or
                ([non-v_LSZ]->ch  ch  delete)
                or
                (atmark p1  test shortv  <+ 'e')
            )
        )
    )

But we can improve the appearance, and speed, of this by turning the second part of the rule into another among command, noting that the only letters that need undoubling are b, d, f, g, m, n, p, r and t,

    define Step_1b as (
        [substring] among (
            'eed'  (R1 <-'ee')
            'ed'
            'ing' (
                test gopast v  delete
                test substring among(
                    'at' 'bl' 'iz'
                         (<+ 'e')
                    'bb' 'dd' 'ff' 'gg' 'mm' 'nn' 'pp' 'rr' 'tt'
                    // ignoring double c, h, j, k, q, v, w, and x
                         ([next]  delete)
                    ''   (atmark p1  test shortv  <+ 'e')
                )
            )
        )
    )

Note the null string in the second among, which acts as a default case.

The Porter stemmer in Snowball is given below. This is an exact implementation of the algorithm described in the 1980 paper, unlike the other implementations distributed by the author, which have, and have always had, three small points of difference (clearly indicated) from the original algorithm. Since all other implementations of the algorithm seen by the author are in some degree inexact, this may well be the first ever correct implementation.

The full algorithm in Snowball


integers ( p1 p2 )
booleans ( Y_found )

routines (
   shortv
   R1 R2
   Step_1a Step_1b Step_1c Step_2 Step_3 Step_4 Step_5a Step_5b
)

externals ( stem )

groupings ( v v_WXY )

define v        'aeiouy'
define v_WXY    v + 'wxY'

backwardmode (

    define shortv as ( non-v_WXY v non-v )

    define R1 as $p1 <= cursor
    define R2 as $p2 <= cursor

    define Step_1a as (
        [substring] among (
            'sses' (<-'ss')
            'ies'  (<-'i')
            'ss'   ()
            's'    (delete)
        )
    )

    define Step_1b as (
        [substring] among (
            'eed'  (R1 <-'ee')
            'ed'
            'ing' (
                test gopast v  delete
                test substring among(
                    'at' 'bl' 'iz'
                         (<+ 'e')
                    'bb' 'dd' 'ff' 'gg' 'mm' 'nn' 'pp' 'rr' 'tt'
                    // ignoring double c, h, j, k, q, v, w, and x
                         ([next]  delete)
                    ''   (atmark p1  test shortv  <+ 'e')
                )
            )
        )
    )

    define Step_1c as (
        ['y' or 'Y']
        gopast v
        <-'i'
    )

    define Step_2 as (
        [substring] R1 among (
            'tional'  (<-'tion')
            'enci'    (<-'ence')
            'anci'    (<-'ance')
            'abli'    (<-'able')
            'entli'   (<-'ent')
            'eli'     (<-'e')
            'izer' 'ization'
                      (<-'ize')
            'ational' 'ation' 'ator'
                      (<-'ate')
            'alli'    (<-'al')
            'alism' 'aliti'
                      (<-'al')
            'fulness' (<-'ful')
            'ousli' 'ousness'
                      (<-'ous')
            'iveness' 'iviti'
                      (<-'ive')
            'biliti'  (<-'ble')
        )
    )

    define Step_3 as (
        [substring] R1 among (
            'alize'   (<-'al')
            'icate' 'iciti' 'ical'
                      (<-'ic')
            'ative' 'ful' 'ness'
                      (delete)
        )
    )

    define Step_4 as (
        [substring] R2 among (
            'al' 'ance' 'ence' 'er' 'ic' 'able' 'ible' 'ant' 'ement'
            'ment' 'ent' 'ou' 'ism' 'ate' 'iti' 'ous' 'ive' 'ize'
                      (delete)
            'ion'     ('s' or 't' delete)
        )
    )

    define Step_5a as (
        ['e']
        R2 or (R1 not shortv)
        delete
    )

    define Step_5b as (
        ['l']
        R2 'l'
        delete
    )
)

define stem as (

    unset Y_found
    do ( ['y'] <-'Y' set Y_found)
    do repeat(goto (v ['y']) <-'Y' set Y_found)

    $p1 = limit
    $p2 = limit
    do(
        gopast v  gopast non-v  setmark p1
        gopast v  gopast non-v  setmark p2
    )

    backwards (
        do Step_1a
        do Step_1b
        do Step_1c
        do Step_2
        do Step_3
        do Step_4
        do Step_5a
        do Step_5b
    )

    do(Y_found  repeat(goto (['Y']) <-'y'))

)