Text::Fuzzy - Partial string matching using edit distances


    use Text::Fuzzy;
    my $tf = Text::Fuzzy->new ('boboon');
    print "Distance is ", $tf->distance ('babboon'), "\n";
    my @words = qw/the quick brown fox jumped over the lazy dog/;
    my $nearest = $tf->nearestv (\@words);
    print "Nearest array entry is $nearest\n";

produces output

    Distance is 2
    Nearest array entry is brown

(This example is included as in the distribution.)


This documents version 0.29 of Text::Fuzzy corresponding to git commit 0e5db1fbccbc7d7518ec39a95d43a5b20166a727 released on Thu Dec 10 14:38:31 2020 +0900.


This module calculates edit distances between words, and searches arrays and files to find the nearest entry by edit distance. It handles both byte strings and character strings (strings containing Unicode), treating each Unicode character as a single entity.

    use Text::Fuzzy;
    use utf8;
    my $tf = Text::Fuzzy->new ('あいうえお☺');
    print $tf->distance ('うえお☺'), "\n";

produces output


(This example is included as in the distribution.)

The default edit distance is the Levenshtein one, which counts each addition (cat -> cart), substitution (cat -> cut), and deletion (carp -> cap) as one unit. The Damerau-Levenshtein edit distance, which also allows transpositions (salt -> slat) may also be selected with the "transpositions_ok" method or the "trans" option.

This module is particularly suited to searching for the nearest match to a term over a list of words, using the "nearestv" or "nearest" methods. It studies the target string to be matched (the first argument to "new") to build information to rapidly reject mismatches in a list. Since computing the Levenshtein and Damerau-Levenshtein edit distances with the Wagner-Fischer algorithm is computationally expensive, the module offers a boost in performance for searching for a string in a list of words.



    my $tf = Text::Fuzzy->new ('bibbety bobbety boo');

Create a new Text::Fuzzy object from the supplied word.

The following parameters may be supplied to new:

    my $tf = Text::Fuzzy->new ('Cinderella', max => 3);

This option affects the behaviour of "nearestv" and "nearest" methods. When searching over an array, this sets the maximum edit distance allowed for a word to be considered a "near match". For example, with

    my $tf = Text::Fuzzy->new ('Cinderella');
    $tf->set_max_distance (3);

when using "nearest", 'Cinder' will not be considered a match, but 'derella' will.

To switch off the maximum distance, and allow all words to be considered, you can set max to be a negative value:

    my $tf = Text::Fuzzy->new ('Cinderella', max => -1);

Note that this is the default, so there is hardly any point specifying it, except if you want to make self-documenting code, or you're worried that the module's default behaviour may suddenly change.

Setting max to zero makes $tf only match exactly.

The method "set_max_distance" does the same thing as this parameter.

    my $tf = Text::Fuzzy->new ('slipper', no_exact => 1);

This parameter switches on rejection of exact matches, in the same way as the method "no_exact":

    my $tf = Text::Fuzzy->new ('slipper');
    $tf->no_exact (1);

This is useful for the case of scanning an array which contains the search term itself, when we are interested in near matches only. For example, if we have a dictionary of words and we need to find near matches for a word which is in the dictionary.

    my $tf = Text::Fuzzy->new ('glass', trans => 1);

This switches on transpositions, in other words it uses the Damerau-Levenshtein edit distance rather than the Levenshtein edit distance. The method "transpositions_ok" has the same effect as this.


    my $dist = $tf->distance ($word);

This method's return value is the edit distance to $word from the word used to create the object in "new".

    use Text::Fuzzy;
    my $cat = Text::Fuzzy->new ('cat');
    print $cat->distance ('cut'), "\n";
    print $cat->distance ('cart'), "\n";
    print $cat->distance ('catamaran'), "\n";
    use utf8;
    print $cat->distance ('γάτος'), "\n";

produces output


(This example is included as in the distribution.)

To know which edits are used to convert the words, use "distance_edits".


    my $nearest_word = $tf->nearestv (\@words);
    my @nearest_words = $tf->nearestv (\@words);

Returns the value in @words which has the nearest distance to the value given to $tf in "new". In array context, it returns a list of the nearest values.

    use Text::Fuzzy;
    my @words = (qw/who where what when why/);
    my $tf = Text::Fuzzy->new ('whammo');
    my @nearest = $tf->nearestv (\@words);
    print "@nearest\n";

produces output

    who what

(This example is included as in the distribution.)

The behaviour of the match can be controlled with "no_exact" and "set_max_distance" in exactly the same way as "nearest".

This is a convenient wrapper around the "nearest" function. "nearest" is annoying to use, because it only returns array offsets, and also error-prone due to having to check to distinguish the first element of the array from an undefined value using defined.

This method was added in version 0.18 of Text::Fuzzy.


    my $index = $tf->nearest (\@words);
    my $nearest_word = $words[$index];

Given an array reference, this returns a number, the index of the nearest element in the array @words to the argument to "new". Having found the nearest match you then need to look up the value in the array, as in $nearest_word above.

It is possible to set a maximum edit distance, beyond which entries are rejected, using "set_max_distance" or the max parameter to "new". In this case, if none of the elements of @words are less than the maximum distance away from the word, $index is the undefined value, so when setting a maximum distance, it is necessary to check the return value of index using defined.

    use Text::Fuzzy;
    my $tf = Text::Fuzzy->new ('calamari', max => 1);
    my @words = qw/Have you ever kissed in the moonlight
                   In the grand and glorious
                   Gay notorious
                   South American Way?/;
    my $index = $tf->nearest (\@words);
    if (defined $index) {
        printf "Found at $index, distance was %d.\n",
        $tf->last_distance ();
    else {
        print "Not found anywhere.\n";

produces output

    Not found anywhere.

(This example is included as in the distribution.)

If there is more than one word with the same edit distance in @words, this returns the last one found, unless it is an exact match, in which case it returns the first one found. To get all matches, call it in array context:

    my @nearest = $tf->nearest (\@words);

In array context, if there are no matches within the minimum distance, nearest returns an empty list. If there is one or more match, it returns the array offset of it or them, not the value itself.

    use Text::Fuzzy;
    my @funky_words = qw/nice funky rice gibbon lice graeme garden/;
    my $tf = Text::Fuzzy->new ('dice');
    my @nearest = $tf->nearest (\@funky_words);
    print "The nearest words are ";
    print join ", ", (map {$funky_words[$_]} @nearest);
    printf ", distance %d.\n", $tf->last_distance ();

produces output

    The nearest words are nice, rice, lice, distance 1.

(This example is included as in the distribution.)


    my $last_distance = $tf->last_distance ();

The distance from the previous match's closest match. This is used in conjunction with "nearest" or "nearestv" to find the edit distance to the previous match.

    use Text::Fuzzy;
    my @words = (qw/who where what when why/);
    my $tf = Text::Fuzzy->new ('whammo');
    my @nearest = $tf->nearestv (\@words);
    print "@nearest\n";
    print $tf->last_distance (), "\n";
    # Prints 3, the number of edits needed to turn "whammo" into "who"
    # (delete a, m, m) or into "what" (replace m with t, delete m, delete
    # o).

produces output

    who what

(This example is included as in the distribution.)


    # Set the max distance.
    $tf->set_max_distance (3);

Set the maximum edit distance of $tf. Set the maximum distance to a low value to improve the speed of searches over lists with "nearest", or to reject unlikely matches. When searching for a near match, anything with an edit distance of a value over the maximum is rejected without computing the exact distance. To compute exact distances, call this method without an argument:

    $tf->set_max_distance ();

The maximum edit distance is switched off, and whatever the nearest match is is accepted. A negative value also switches it off:

    $tf->set_max_distance (-1);

The object created by "new" has no maximum distance unless specified by the user.

    use Text::Fuzzy;
    my $tf = Text::Fuzzy->new ('nopqrstuvwxyz');
    # Prints 13, the correct value.
    print $tf->distance ('abcdefghijklm'), "\n";
    $tf->set_max_distance (10);
    # Prints 11, one more than the maximum distance, because the search
    # stopped when the distance was exceeded.
    print $tf->distance ('abcdefghijklm'), "\n";

produces output


(This example is included as in the distribution.)

Setting the maximum distance is a way to make a search more rapid. For example if you are searching over a dictionary of 100,000 or a million words, and only need close matches, you can more rapidly reject unwanted matches by setting the maximum distance to a lower value. Calculating Levenshtein distance is an O(n^2) algorithm in the lengths of the words, so even a small increase in the maximum permitted distance means a much larger amount of work for the computer to do. With the maximum distance set, the computer can give up calculating more quickly with bad matches.


    $tf->transpositions_ok (1);

A true value in the argument changes the type of edit distance used to allow transpositions, such as clam and calm. Initially transpositions are not allowed, giving the Levenshtein edit distance. If transpositions are used, the edit distance becomes the Damerau-Levenshtein edit distance. A false value disallows transpositions:

    $tf->transpositions_ok (0);


    $tf->no_exact (1);

This is a flag to "nearest" which makes it ignore exact matches. For example,

    use Text::Fuzzy;
    my @words = qw/bibbity bobbity boo/;
    for my $word (@words) {
        my $tf = Text::Fuzzy->new ($word);
        $tf->no_exact (0);
        my $nearest1 = $tf->nearest (\@words);
        print "With exact, nearest to $word is $words[$nearest1]\n";
        # Make "$word" not match itself.
        $tf->no_exact (1);
        my $nearest2 = $tf->nearest (\@words);
        print "Without exact, nearest to $word is $words[$nearest2]\n";

produces output

    With exact, nearest to bibbity is bibbity
    Without exact, nearest to bibbity is bobbity
    With exact, nearest to bobbity is bobbity
    Without exact, nearest to bobbity is bibbity
    With exact, nearest to boo is boo
    Without exact, nearest to boo is bobbity

(This example is included as in the distribution.)

This is for the case of searching over an array which contains the searched-for item itself.


    my $nearest = $tf->scan_file ('/usr/share/dict/words');

Scan a file to find the nearest match to the word used in "new". This assumes that the file contains lines of text separated by newlines, and finds the closest match in the file. Its return value is a string rather than a line number. It cannot return an array of values. It does not currently support Unicode-encoded files.


These functions do not require a Text::Fuzzy object.


    my ($distance, $edits) = distance_edits ('before', 'after');

This returns the edit distance between the two arguments, and the edits necessary to transform the first one into the second one. $Edits is a string containing the four letters k, r, d, and i, for "keep", "replace", "delete", and "insert" respectively. For example, for "piece" and "peace", $edits contains "krrkk" for "keep, replace, replace, keep, keep".

    use Text::Fuzzy 'distance_edits';
    my @words = (qw/who where what when why/);
    my $tf = Text::Fuzzy->new ('whammo');
    my @nearest = $tf->nearestv (\@words);
    print "@nearest\n";
    # Prints "who what"
    print $tf->last_distance (), "\n";
    # Prints 3, the number of edits needed to turn "whammo" into "who"
    # (delete a, m, m) or into "what" (replace m with t, delete m, delete
    # o).
    my ($distance, $edits) = distance_edits ('whammo', 'who');
    print "$edits\n";
    # Prints kkdddk, keep w, keep h, delete a, delete m, delete m, keep o.

produces output

    who what

(This example is included as in the distribution.)

This does not handle transpositions. Unlike the rest of the module, this is pure Perl rather than XS, and not optimized for speed. The edit distance search within "nearest" is optimized for speed, and hence discards its record of edits used to get the result.


    my ($offset, $edits, $distance) = fuzzy_index ($needle, $haystack);

Searches for $needle in $haystack using fuzzy matching.

Return value is the offest of the closest match found, the edits necessary on $needle to make it into the matching text, and the Levenshtein edit distance between the matching part of $haystack and $needle.

For the algorithm used, see

This is implemented in Perl not C, and it's slow due to lots of debugging code. Please expect the interface and internals to change.


This section gives extended examples of the use of the module to solve practical problems.


The file examples/misspelt-web-page.cgi is an example of a CGI script which does something similar to the Apache mod_speling module, offering spelling corrections for mistyped URLs and sending the user to a correct page.

    use Text::Fuzzy;
    # The directory of files served by the web server.
    my $web_root = '/usr/local/www/data';
    # If the query is "", $path_info is
    # "abc/xyz.html".
    my $path_info = $ENV{REQUEST_URI};
    if (! defined $path_info) {
        fail ("No path info");
    if ($0 =~ /$path_info/) {
        fail ("Don't redirect to self");
    # This is the list of files under the main page.
    my @allfiles = get_all_files ($web_root, '');
    # This is our spelling search engine.
    my $tf = Text::Fuzzy->new ($path_info);
    my $nearest = $tf->nearest (\@allfiles, max => 5);
    if (defined $nearest) {
        redirect ($allfiles[$nearest]);
    else {
        fail ("Nothing like $path_info was found");
    # Read all the files under "$root/$dir". This is recursive. The return
    # value is an array containing all files found.
    sub get_all_files
        my ($root, $dir) = @_;
        my @allfiles;
        my $full_dir = "$root/$dir";
        if (! -d $full_dir) {
            fail ("$full_dir is not a directory");
        opendir DIR, $full_dir or fail ("Can't open directory $full_dir: $!");
        my @files = grep !/^\./, readdir DIR;
        closedir DIR or fail ("Can't close $full_dir: $!");
        for my $file (@files) {
            my $dir_file = "$dir/$file";
            my $full_file = "$root/$dir_file";
            if (-d $full_file) {
                push @allfiles, get_all_files ($root, $dir_file);
            else {
                push @allfiles, $dir_file;
        return @allfiles;
    # Print a "permanent redirect" to the respelt name, then exit.
    sub redirect
        my ($url) = @_;
        print <<EOF;
    Status: 301
    Location: $url
    # Print an error message for the sake of the requester, and print a
    # message to the error log, then exit.
    sub fail
        my ($error) = @_;
        print <<EOF;
    Content-Type: text/plain
        # Add the name of the program and the time to the error message,
        # otherwise the error log will get awfully confusing-looking.
        my $time = scalar gmtime ();
        print STDERR "$0: $time: $error\n";

See also for how to set up .htaccess to use the script.

The file examples/ is a spell checker. It uses a dictionary of words specified by a command-line option "-d": -d /usr/dict/words file1.txt file2.txt

It prints out any words which look like spelling mistakes, using the dictionary.

    use Getopt::Long;
    use Text::Fuzzy;
    use Lingua::EN::PluralToSingular 'to_singular';
    # The location of the Unix dictionary.
    my $dict = '/usr/share/dict/words';
    # Default maximum edit distance. Five is quite a big number for a
    # spelling mistake.
    my $max = 5;
    GetOptions (
        "dict=s" => \$dict,
        "max=i" => \$max,
    my @words;
    my %words;
    my $min_length = 4;
    read_dictionary ($dict, \@words, \%words);
    # Known mistakes, don't repeat.
    my %known;
    # Spell-check each file on the command line.
    for my $file (@ARGV) {
        open my $input, "<", $file or die "Can't open $file: $!";
        while (<$input>) {
            my @line = split /[^a-z']+/i, $_;
            for my $word (@line) {
                # Remove leading/trailing apostrophes.
                $word =~ s/^'|'$//g;
                my $clean_word = to_singular (lc $word);
                $clean_word =~ s/'s$//;
                if ($words{$clean_word} || $words{$word}) {
                    # It is in the dictionary.
                if (length $word < $min_length) {
                    # Very short words are ignored.
                if ($word eq uc $word) {
                    # Acronym like BBC, IRA, etc.
                if ($known{$clean_word}) {
                    # This word was already given to the user.
                if ($clean_word =~ /(.*)ed$/ || $clean_word =~ /(.*)ing/) {
                    my $stem = $1;
                    if ($words{$stem} || $words{"${stem}e"}) {
                        # Past/gerund of $stem/${stem}e
                    # Test for doubled end consonants,
                    # e.g. "submitted"/"submit".
                    if ($stem =~ /([bcdfghjklmnpqrstvwxz])\1/) {
                        $stem =~ s/$1$//;
                        if ($words{$stem}) {
                            # Past/gerund of $stem/${stem}e
                my $tf = Text::Fuzzy->new ($clean_word, max => $max);
                my $nearest = $tf->nearest (\@words);
                # We have set a maximum distance to search for, so we need
                # to check whether $nearest is defined.
                if (defined $nearest) {
                    my $correction = $words[$nearest];
                    print "$file:$.: '$word' may be '$correction'.\n";
                    $known{$clean_word} = $correction;
                else {
                    print "$file:$.: $word may be a spelling mistake.\n";
                    $known{$clean_word} = 1;
        close $input or die $!;
    sub read_dictionary
        my ($dict, $words_array, $words_hash) = @_;    
        open my $din, "<", $dict or die "Can't open dictionary $dict: $!";
        my @words;
        while (<$din>) {
            push @words, $_;
        close $din or die $!;
        # Apostrophe words
        my @apo = qw/
                        let's I'll you'll he'll she'll they'll we'll I'm
                        you're he's she's it's we're they're I've they've
                        you've we've one's isn't aren't doesn't don't
                        won't wouldn't I'd you'd he'd we'd they'd
                        shouldn't couldn't didn't can't
        # Irregular past participles.
        my @pp = qw/became/;
        push @words, @apo, @pp;
        for (@words) {
            push @$words_array, lc $_;
            $words_hash->{$_} = 1;
            $words_hash->{lc $_} = 1;

Because the usual Unix dictionary doesn't have plurals, it uses Lingua::EN::PluralToSingular, to convert nouns into singular forms. Unfortunately it still misses past participles and past tenses of verbs.

The file examples/ extracts the kana entries from "edict", a freely-available Japanese to English electronic dictionary, and does some fuzzy searches on them. It requires a local copy of the file to run. This script demonstrates the use of Unicode searches with Text::Fuzzy.

    use Lingua::JA::Moji ':all';
    use Text::Fuzzy;
    use utf8;
    my $infile = '/home/ben/data/edrdg/edict';
    open my $in, "<:encoding(EUC-JP)", $infile or die $!;
    my @kana;
    while (<$in>) {
        my $kana;
        if (/\[(\p{InKana}+)\]/) {
            $kana = $1;
        elsif (/^(\p{InKana}+)/) {
            $kana = $1;
        if ($kana) {
            $kana = kana2katakana ($kana);
            push @kana, $kana;
    printf "Starting fuzzy searches over %d lines.\n", scalar @kana;
    search ('ウオソウコ');
    search ('アイウエオカキクケコバビブベボハヒフヘホ');
    search ('アルベルトアインシュタイン');
    search ('バババブ');
    search ('バババブアルベルト');
    sub search
        my ($silly) = @_;
        my $max = 10;
        my $search = Text::Fuzzy->new ($silly, max => $max);
        my $n = $search->nearest (\@kana);
        if (defined $n) {
            printf "$silly nearest is $kana[$n] (distance %d)\n",
                $search->last_distance ();
        else {
            printf "Nothing like '$silly' was found within $max edits.\n";


The module Lingua::JA::Gairaigo::Fuzzy tries to determine whether two Japanese loanwords are the same word or not.


The module CPAN::Nearest offers a search over the titles of CPAN modules using a fuzzy search to get the nearest match.


This module has no dependencies on other modules.


Reporting a bug

There is a bug tracker for the module at


The CPAN tester results are at The ActiveState tester results are at


The following methods are for benchmarking the module and checking its correctness.


    $tf->no_alphabet (1);

This turns off alphabetizing of the string. Alphabetizing is a filter where the intersection of all the characters in the two strings is computed, and if the alphabetical difference of the two strings is greater than the maximum distance, the match is rejected without applying the dynamic programming algorithm. This increases speed, because the dynamic programming algorithm is slow.

The alphabetizing should not ever reject anything which is a legitimate match, and it should make the program run faster in almost every case. The only envisaged uses of switching this off are checking that the algorithm is working correctly, and benchmarking performance.


    my $trans_ok = $tf->get_trans ();

This returns the value set by "transpositions_ok".


    my $length = $tf->unicode_length ();

This returns the length in characters (not bytes) of the string used in "new". If the string is not marked as Unicode, it returns the undefined value. In the following, $l1 should be equal to $l2.

    use utf8;
    my $word = 'ⅅⅆⅇⅈⅉ';
    my $l1 = length $word;
    my $tf = Text::Fuzzy->new ($word);
    my $l2 = $tf->unicode_length ();


    my $rejected = $tf->ualphabet_rejections ();

After running "nearest" over an array, this returns the number of entries of the array which were rejected using only the Unicode alphabet. Its value is reset to zero each time "nearest" is called.


    my $rejected = $tf->alphabet_rejections ();

After running "nearest" over an array, this returns the number of entries of the array which were rejected using only the non-Unicode alphabet. Its value is reset to zero each time "nearest" is called.


    my $rejected = $tf->length_rejections ();

After running "nearest" over an array, this returns the number of entries of the array which were rejected because the length difference between them and the target string was larger than the maximum distance allowed.


    # Get the maximum edit distance.
    print "The max distance is ", $tf->get_max_distance (), "\n";

Get the maximum edit distance of $tf. The maximum distance may be set with "set_max_distance".


Other CPAN modules

Similar modules on CPAN include the following.


Approximate matching (fuzzy matching) using the Levenshtein edit distance. As a bonus, if you don't have a headache, you can get one easily trying to make head or tail out of this module's documentation.


Determine the edit transcript between two strings. This is similar to what you get from "distance_edits" in this module.


This is Nick Logan's Pure Perl version of this module.


Levenshtein-Damerau edit distance.


XS Levenshtein distance calculation with bounds and adjustable costs (so the cost of deletion can be more than the cost of addition, etc.) See also Text::WagnerFischer for a pure-Perl module which also allows altered costs.


An XS implementation of the Levenshtein edit distance. It claims to be a drop-in replacement for Text::LevenshteinXS which does Unicode correctly.


An XS implementation of the Levenshtein edit distance. Does not do Unicode very well. See


A wrapper around the edlib library that computes Levenshtein edit distance and optimal alignment path for a pair of strings.


Structure for efficient fuzzy matching.


An implementation of the Brew edit distance.


Implements the Wagner-Fischer algorithm to calculate edit distances. This is generalised version of the Levenshtein edit distance. See also Text::Levenshtein::Flexible for an XS version.


Some benchmarks of various modules including this one.


Another text similarity measure.

About the algorithms

This section contains some blog posts which I found useful in understanding the algorithms.

Fuzzy substring matching with Levenshtein distance in Python by Ryan Ginstrom explains the Levenshtein algorithm and its use in substring matching.

Damerau-Levenshtein Edit Distance Explained by James M. Jensen II explains the Damerau-Levenshtein edit distance (the algorithm used with "transpositions_ok").

I recommend steering fairly clear of the Wikipedia articles on these things, which are very poorly written indeed.


Here are the original research papers by the algorithms' discoverers.


Damerau, Fred J. (March 1964), "A technique for computer detection and correction of spelling errors", Communications of the ACM, ACM, 7 (3): 171–176, doi:10.1145/363958.363994


Levenshtein, Vladimir I. (February 1966), "Binary codes capable of correcting deletions, insertions, and reversals", Soviet Physics Doklady, 10 (8): 707–710

Wagner and Fischer

R. Wagner and M. Fischer (1974), "The string to string correction problem", Journal of the ACM, 21:168-178, doi:10.1145/321796.321811


  • A bug was fixed where an input string may be overwritten in code like

        my $tf = Text::Fuzzy->new ($x);
        $tf->distance ($y);

    if $x is a plain ASCII string and $y is a Unicode string.

  • Links were added to the examples, and the outputs of the examples were added as part of the documentation.

  • The transposition code (the implementation of the Damerau-Levenshtein distance) was completely rewritten to make it more efficient. The back-indexing of strings to find transpositions was changed so that the index of the object's string (the first argument to "new") is preserved from one query to the next. A useless indexing of characters in the other string was removed. The structure used to hold the characters was changed from an unsorted allocated linked list to a sorted array in the case of Unicode strings, and a 256 character array in the case of non-Unicode strings. The oddly-named variables were renamed to more meaningful names.

  • The transposition code now allows for a maximum distance to be set, beyond which no further matches will be allowed.


The edit distance including transpositions was contributed by Nick Logan (UGEXE). (This code was largely rewritten in version "0.28", so Nick Logan can no longer be held responsible for the Text::Fuzzy module's failings.) Some of the tests in t/trans.t are taken from the Text::Levenshtein::Damerau::XS module. Nils Boeffel reported a bug where strings may be overwritten in version 0.25.


Ben Bullock, <>


This package and associated files are copyright (C) 2012-2020 Ben Bullock.

You can use, copy, modify and redistribute this package and associated files under the Perl Artistic Licence or the GNU General Public Licence.