The Porter Stemming Algorithm was developed by Martin Porter for reducing English words to their word stems. For example, the word “connections” would be reduced to its stem form “connect.”
<?php
/*************************************************************************
* *
* class.stemmer.inc *
* *
*************************************************************************
* *
* Implementation of the Porter Stemming Alorithm *
* *
* Copyright (c) 2003-2007 Jon Abernathy <jon@chuggnutt.com> *
* All rights reserved. *
* *
* This script is free software; you can redistribute it and/or modify *
* it under the terms of the GNU General Public License as published by *
* the Free Software Foundation; either version 2 of the License, or *
* (at your option) any later version. *
* *
* The GNU General Public License can be found at *
* http://www.gnu.org/copyleft/gpl.html. *
* *
* This script is distributed in the hope that it will be useful, *
* but WITHOUT ANY WARRANTY; without even the implied warranty of *
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the *
* GNU General Public License for more details. *
* *
* Author(s): Jon Abernathy <jon@chuggnutt.com> *
* *
* Last modified: 08/08/07 *
* *
*************************************************************************/
/**
* Takes a word, or list of words, and reduces them to their English stems.
*
* This is a fairly faithful implementation of the Porter stemming algorithm that
* reduces English words to their stems, originally adapted from the ANSI-C code found
* on the official Porter Stemming Algorithm website, located at
* http://www.tartarus.org/~martin/PorterStemmer and later changed to conform
* more accurately to the algorithm itself.
*
* There is a deviation in the way compound words are stemmed, such as
* hyphenated words and words starting with certain prefixes. For instance,
* "international" should be reduced to "internation" and not "intern," but
* an unmodified version of the alorithm will do just that. Currently, only
* hyphenated words are accounted for.
*
* Thanks to Mike Boone (http://www.boonedocks.net/) for finding a fatal
* error in the is_consonant() function dealing with short word stems beginning
* with "Y".
*
* Additional thanks to Mark Plumbley for finding an additional problem with
* short words beginning with "Y"--the word "yves" for example. I fixed the
* _o() and is_consonant() functions to appropriately sanity check the values
* being passed around. Updated 3/12/04.
*
* Thanks to Andrew Jeffries (http://www.nextgendevelopment.co.uk/) for
* discovering a bug for words beginning with "yy"--this would cause the
* is_consonant() method checking either of these first "y"s to fall into
* a recursive infinite loop and crash the program. Updated 9/23/05.
*
* 11/09/05, big update. Prompted by an email from Richard Shelquist, I went
* back over the class and fixed some errors in the algorithm; in particular
* I made sure to conform EXACTLY to the written algorithm found at
* the Stemmer website. This class now takes the test vocabulary file found at
* http://tartarus.org/~martin/PorterStemmer/voc.txt and stems every single
* word exactly as shown in the output file found at
* http://tartarus.org/~martin/PorterStemmer/output.txt, with two exceptions:
* "ycleped" and "ycliped", which I believe my version stems correctly, due
* to assuming the "Y" at the beginning of a word followed by a consonant--
* as in "Yvette"--is to be treated as a vowel and NOT a consonant. Yeah,
* that's arrogant; allow me some, okay?
* Of course, should someone find an exception after boasting of my arrogance,
* please let me know. I'm only human, after all.
*
* Thanks to Damon Sauve (http://www.shopping.com/) for suggesting a better
* fix to the handling of hyphenated words (in his case, multi-hyphenated
* words). His fix used a regular expression to extract the final part of the
* hyphenated word, while mine does a substr() split instead. Also, his version
* allows dots and apostrophes in words, such as URLs and contractions, and
* I realize this is a real-world scenario that I didn't account for, so it's
* been incorporated.
*
* @author Jon Abernathy <jon@chuggnutt.com>
* @version 2.1
*/
class Stemmer
{
/**
* Takes a word and returns it reduced to its stem.
*
* Non-alphanumerics and hyphens are removed, except for dots and
* apostrophes, and if the word is less than three characters in
* length, it will be stemmed according to the five-step
* Porter stemming algorithm.
*
* Note special cases here: hyphenated words (such as half-life) will
* only have the base after the last hyphen stemmed (so half-life would
* only have "life" subject to stemming). Handles multi-hyphenated
* words, too.
*
* @param string $word Word to reduce
* @access public
* @return string Stemmed word
*/
function stem( $word )
{
if ( empty($word) ) {
return false;
}
$result = '';
$word = strtolower($word);
// Strip punctuation, etc. Keep ' and . for URLs and contractions.
if ( substr($word, -2) == "'s" ) {
$word = substr($word, 0, -2);
}
$word = preg_replace("/[^a-z0-9'.-]/", '', $word);
$first = '';
if ( strpos($word, '-') !== false ) {
//list($first, $word) = explode('-', $word);
//$first .= '-';
$first = substr($word, 0, strrpos($word, '-') + 1); // Grabs hyphen too
$word = substr($word, strrpos($word, '-') + 1);
}
if ( strlen($word) > 2 ) {
$word = $this->_step_1($word);
$word = $this->_step_2($word);
$word = $this->_step_3($word);
$word = $this->_step_4($word);
$word = $this->_step_5($word);
}
$result = $first . $word;
return $result;
}
/**
* Takes a list of words and returns them reduced to their stems.
*
* $words can be either a string or an array. If it is a string, it will
* be split into separate words on whitespace, commas, or semicolons. If
* an array, it assumes one word per element.
*
* @param mixed $words String or array of word(s) to reduce
* @access public
* @return array List of word stems
*/
function stem_list( $words )
{
if ( empty($words) ) {
return false;
}
$results = array();
if ( !is_array($words) ) {
$words = split("[ ,;\n\r\t]+", trim($words));
}
foreach ( $words as $word ) {
if ( $result = $this->stem($word) ) {
$results[] = $result;
}
}
return $results;
}
/**
* Performs the functions of steps 1a and 1b of the Porter Stemming Algorithm.
*
* First, if the word is in plural form, it is reduced to singular form.
* Then, any -ed or -ing endings are removed as appropriate, and finally,
* words ending in "y" with a vowel in the stem have the "y" changed to "i".
*
* @param string $word Word to reduce
* @access private
* @return string Reduced word
*/
function _step_1( $word )
{
// Step 1a
if ( substr($word, -1) == 's' ) {
if ( substr($word, -4) == 'sses' ) {
$word = substr($word, 0, -2);
} elseif ( substr($word, -3) == 'ies' ) {
$word = substr($word, 0, -2);
} elseif ( substr($word, -2, 1) != 's' ) {
// If second-to-last character is not "s"
$word = substr($word, 0, -1);
}
}
// Step 1b
if ( substr($word, -3) == 'eed' ) {
if ($this->count_vc(substr($word, 0, -3)) > 0 ) {
// Convert '-eed' to '-ee'
$word = substr($word, 0, -1);
}
} else {
if ( preg_match('/([aeiou]|[^aeiou]y).*(ed|ing)$/', $word) ) { // vowel in stem
// Strip '-ed' or '-ing'
if ( substr($word, -2) == 'ed' ) {
$word = substr($word, 0, -2);
} else {
$word = substr($word, 0, -3);
}
if ( substr($word, -2) == 'at' || substr($word, -2) == 'bl' ||
substr($word, -2) == 'iz' ) {
$word .= 'e';
} else {
$last_char = substr($word, -1, 1);
$next_to_last = substr($word, -2, 1);
// Strip ending double consonants to single, unless "l", "s" or "z"
if ( $this->is_consonant($word, -1) &&
$last_char == $next_to_last &&
$last_char != 'l' && $last_char != 's' && $last_char != 'z' ) {
$word = substr($word, 0, -1);
} else {
// If VC, and cvc (but not w,x,y at end)
if ( $this->count_vc($word) == 1 && $this->_o($word) ) {
$word .= 'e';
}
}
}
}
}
// Step 1c
// Turn y into i when another vowel in stem
if ( preg_match('/([aeiou]|[^aeiou]y).*y$/', $word) ) { // vowel in stem
$word = substr($word, 0, -1) . 'i';
}
return $word;
}
/**
* Performs the function of step 2 of the Porter Stemming Algorithm.
*
* Step 2 maps double suffixes to single ones when the second-to-last character
* matches the given letters. So "-ization" (which is "-ize" plus "-ation"
* becomes "-ize". Mapping to a single character occurence speeds up the script
* by reducing the number of possible string searches.
*
* Note: for this step (and steps 3 and 4), the algorithm requires that if
* a suffix match is found (checks longest first), then the step ends, regardless
* if a replacement occurred. Some (or many) implementations simply keep
* searching though a list of suffixes, even if one is found.
*
* @param string $word Word to reduce
* @access private
* @return string Reduced word
*/
function _step_2( $word )
{
switch ( substr($word, -2, 1) ) {
case 'a':
if ( $this->_replace($word, 'ational', 'ate', 0) ) {
return $word;
}
if ( $this->_replace($word, 'tional', 'tion', 0) ) {
return $word;
}
break;
case 'c':
if ( $this->_replace($word, 'enci', 'ence', 0) ) {
return $word;
}
if ( $this->_replace($word, 'anci', 'ance', 0) ) {
return $word;
}
break;
case 'e':
if ( $this->_replace($word, 'izer', 'ize', 0) ) {
return $word;
}
break;
case 'l':
// This condition is a departure from the original algorithm;
// I adapted it from the departure in the ANSI-C version.
if ( $this->_replace($word, 'bli', 'ble', 0) ) {
return $word;
}
if ( $this->_replace($word, 'alli', 'al', 0) ) {
return $word;
}
if ( $this->_replace($word, 'entli', 'ent', 0) ) {
return $word;
}
if ( $this->_replace($word, 'eli', 'e', 0) ) {
return $word;
}
if ( $this->_replace($word, 'ousli', 'ous', 0) ) {
return $word;
}
break;
case 'o':
if ( $this->_replace($word, 'ization', 'ize', 0) ) {
return $word;
}
if ( $this->_replace($word, 'isation', 'ize', 0) ) {
return $word;
}
if ( $this->_replace($word, 'ation', 'ate', 0) ) {
return $word;
}
if ( $this->_replace($word, 'ator', 'ate', 0) ) {
return $word;
}
break;
case 's':
if ( $this->_replace($word, 'alism', 'al', 0) ) {
return $word;
}
if ( $this->_replace($word, 'iveness', 'ive', 0) ) {
return $word;
}
if ( $this->_replace($word, 'fulness', 'ful', 0) ) {
return $word;
}
if ( $this->_replace($word, 'ousness', 'ous', 0) ) {
return $word;
}
break;
case 't':
if ( $this->_replace($word, 'aliti', 'al', 0) ) {
return $word;
}
if ( $this->_replace($word, 'iviti', 'ive', 0) ) {
return $word;
}
if ( $this->_replace($word, 'biliti', 'ble', 0) ) {
return $word;
}
break;
case 'g':
// This condition is a departure from the original algorithm;
// I adapted it from the departure in the ANSI-C version.
if ( $this->_replace($word, 'logi', 'log', 0) ) { //*****
return $word;
}
break;
}
return $word;
}
/**
* Performs the function of step 3 of the Porter Stemming Algorithm.
*
* Step 3 works in a similar stragegy to step 2, though checking the
* last character.
*
* @param string $word Word to reduce
* @access private
* @return string Reduced word
*/
function _step_3( $word )
{
switch ( substr($word, -1) ) {
case 'e':
if ( $this->_replace($word, 'icate', 'ic', 0) ) {
return $word;
}
if ( $this->_replace($word, 'ative', '', 0) ) {
return $word;
}
if ( $this->_replace($word, 'alize', 'al', 0) ) {
return $word;
}
break;
case 'i':
if ( $this->_replace($word, 'iciti', 'ic', 0) ) {
return $word;
}
break;
case 'l':
if ( $this->_replace($word, 'ical', 'ic', 0) ) {
return $word;
}
if ( $this->_replace($word, 'ful', '', 0) ) {
return $word;
}
break;
case 's':
if ( $this->_replace($word, 'ness', '', 0) ) {
return $word;
}
break;
}
return $word;
}
/**
* Performs the function of step 4 of the Porter Stemming Algorithm.
*
* Step 4 works similarly to steps 3 and 2, above, though it removes
* the endings in the context of VCVC (vowel-consonant-vowel-consonant
* combinations).
*
* @param string $word Word to reduce
* @access private
* @return string Reduced word
*/
function _step_4( $word )
{
switch ( substr($word, -2, 1) ) {
case 'a':
if ( $this->_replace($word, 'al', '', 1) ) {
return $word;
}
break;
case 'c':
if ( $this->_replace($word, 'ance', '', 1) ) {
return $word;
}
if ( $this->_replace($word, 'ence', '', 1) ) {
return $word;
}
break;
case 'e':
if ( $this->_replace($word, 'er', '', 1) ) {
return $word;
}
break;
case 'i':
if ( $this->_replace($word, 'ic', '', 1) ) {
return $word;
}
break;
case 'l':
if ( $this->_replace($word, 'able', '', 1) ) {
return $word;
}
if ( $this->_replace($word, 'ible', '', 1) ) {
return $word;
}
break;
case 'n':
if ( $this->_replace($word, 'ant', '', 1) ) {
return $word;
}
if ( $this->_replace($word, 'ement', '', 1) ) {
return $word;
}
if ( $this->_replace($word, 'ment', '', 1) ) {
return $word;
}
if ( $this->_replace($word, 'ent', '', 1) ) {
return $word;
}
break;
case 'o':
// special cases
if ( substr($word, -4) == 'sion' || substr($word, -4) == 'tion' ) {
if ( $this->_replace($word, 'ion', '', 1) ) {
return $word;
}
}
if ( $this->_replace($word, 'ou', '', 1) ) {
return $word;
}
break;
case 's':
if ( $this->_replace($word, 'ism', '', 1) ) {
return $word;
}
break;
case 't':
if ( $this->_replace($word, 'ate', '', 1) ) {
return $word;
}
if ( $this->_replace($word, 'iti', '', 1) ) {
return $word;
}
break;
case 'u':
if ( $this->_replace($word, 'ous', '', 1) ) {
return $word;
}
break;
case 'v':
if ( $this->_replace($word, 'ive', '', 1) ) {
return $word;
}
break;
case 'z':
if ( $this->_replace($word, 'ize', '', 1) ) {
return $word;
}
break;
}
return $word;
}
/**
* Performs the function of step 5 of the Porter Stemming Algorithm.
*
* Step 5 removes a final "-e" and changes "-ll" to "-l" in the context
* of VCVC (vowel-consonant-vowel-consonant combinations).
*
* @param string $word Word to reduce
* @access private
* @return string Reduced word
*/
function _step_5( $word )
{
if ( substr($word, -1) == 'e' ) {
$short = substr($word, 0, -1);
// Only remove in vcvc context...
if ( $this->count_vc($short) > 1 ) {
$word = $short;
} elseif ( $this->count_vc($short) == 1 && !$this->_o($short) ) {
$word = $short;
}
}
if ( substr($word, -2) == 'll' ) {
// Only remove in vcvc context...
if ( $this->count_vc($word) > 1 ) {
$word = substr($word, 0, -1);
}
}
return $word;
}
/**
* Checks that the specified letter (position) in the word is a consonant.
*
* Handy check adapted from the ANSI C program. Regular vowels always return
* FALSE, while "y" is a special case: if the prececing character is a vowel,
* "y" is a consonant, otherwise it's a vowel.
*
* And, if checking "y" in the first position and the word starts with "yy",
* return true even though it's not a legitimate word (it crashes otherwise).
*
* @param string $word Word to check
* @param integer $pos Position in the string to check
* @access public
* @return boolean
*/
function is_consonant( $word, $pos )
{
// Sanity checking $pos
if ( abs($pos) > strlen($word) ) {
if ( $pos < 0 ) {
// Points "too far back" in the string. Set it to beginning.
$pos = 0;
} else {
// Points "too far forward." Set it to end.
$pos = -1;
}
}
$char = substr($word, $pos, 1);
switch ( $char ) {
case 'a':
case 'e':
case 'i':
case 'o':
case 'u':
return false;
case 'y':
if ( $pos == 0 || strlen($word) == -$pos ) {
// Check second letter of word.
// If word starts with "yy", return true.
if ( substr($word, 1, 1) == 'y' ) {
return true;
}
return !($this->is_consonant($word, 1));
} else {
return !($this->is_consonant($word, $pos - 1));
}
default:
return true;
}
}
/**
* Counts (measures) the number of vowel-consonant occurences.
*
* Based on the algorithm; this handy function counts the number of
* occurences of vowels (1 or more) followed by consonants (1 or more),
* ignoring any beginning consonants or trailing vowels. A legitimate
* VC combination counts as 1 (ie. VCVC = 2, VCVCVC = 3, etc.).
*
* @param string $word Word to measure
* @access public
* @return integer
*/
function count_vc( $word )
{
$m = 0;
$length = strlen($word);
$prev_c = false;
for ( $i = 0; $i < $length; $i++ ) {
$is_c = $this->is_consonant($word, $i);
if ( $is_c ) {
if ( $m > 0 && !$prev_c ) {
$m += 0.5;
}
} else {
if ( $prev_c || $m == 0 ) {
$m += 0.5;
}
}
$prev_c = $is_c;
}
$m = floor($m);
return $m;
}
/**
* Checks for a specific consonant-vowel-consonant condition.
*
* This function is named directly from the original algorithm. It
* looks the last three characters of the word ending as
* consonant-vowel-consonant, with the final consonant NOT being one
* of "w", "x" or "y".
*
* @param string $word Word to check
* @access private
* @return boolean
*/
function _o( $word )
{
if ( strlen($word) >= 3 ) {
if ( $this->is_consonant($word, -1) && !$this->is_consonant($word, -2) &&
$this->is_consonant($word, -3) ) {
$last_char = substr($word, -1);
if ( $last_char == 'w' || $last_char == 'x' || $last_char == 'y' ) {
return false;
}
return true;
}
}
return false;
}
/**
* Replaces suffix, if found and word measure is a minimum count.
*
* @param string $word Word to check and modify
* @param string $suffix Suffix to look for
* @param string $replace Suffix replacement
* @param integer $m Word measure value that the word must be greater
* than to replace
* @access private
* @return boolean
*/
function _replace( &$word, $suffix, $replace, $m = 0 )
{
$sl = strlen($suffix);
if ( substr($word, -$sl) == $suffix ) {
$short = substr_replace($word, '', -$sl);
if ( $this->count_vc($short) > $m ) {
$word = $short . $replace;
}
// Found this suffix, doesn't matter if replacement succeeded
return true;
}
return false;
}
}
?>