Distribution · language

Lingua::NumericWordForms

Works

Numbers spelled out as words and read back again, across nineteen languages — with automatic detection of which language a spelled-out number is in.

Version: 0.6.2 zef:antononcube
Depends: nothing outside the core
License: Artistic-2.0
Its own test suite: 25 files, green
Checked: 2026-09-15 against Raku++ 3.28.0 and Rakudo 2026.08
Where it lives: raku.land · source

Install it #

$ rakupp install Lingua::NumericWordForms

zef install Lingua::NumericWordForms writes the same store; either installer leaves the module usable by both engines.

What it is for #

Turning 2026 into "two thousand twenty six" is a cheque printer, an accessibility layer, a text-to-speech front end. Turning "две тысячи двадцать шесть" back into 2026 is the harder and rarer direction, and it is what a natural-language interface needs when a person types a quantity as words. This distribution does both, and across a genuinely wide set of languages — nineteen grammars, from Armenian and Azerbaijani through Japanese and Korean to Ukrainian, plus Koremutake, a syllable encoding of integers that is not a language at all.

It can also tell you which language a spelled-out number is in, by trying the grammars.

Reading and writing #

File

use Lingua::NumericWordForms;

for 0, 42, 1234, 2026, 1_000_000 -> $n {
    my $words = to-numeric-word-form($n);
    say $n, ' -> ', $words, ' -> ', from-numeric-word-form($words);
}

say to-numeric-word-form([1, 2, 3], 'English');
say from-numeric-word-form(['one', 'two', 'three']);

Output

0 -> zero -> 0
42 -> forty two -> 42
1234 -> one thousand, two hundred thirty four -> 1234
2026 -> two thousand, twenty six -> 2026
1000000 -> one million -> 1000000
(one two three)
(1 2 3)

Nineteen grammars, and guessing between them #

File

use Lingua::NumericWordForms;

my %samples =
    English  => 'two thousand twenty six',
    Russian  => 'две тысячи двадцать шесть',
    Spanish  => 'dos mil veintiséis',
    German   => 'zweitausendsechsundzwanzig',
    French   => 'deux mille vingt-six',
    Japanese => '二千二十六';

for %samples.keys.sort -> $lang {
    say $lang, ': told=', from-numeric-word-form(%samples{$lang}, $lang),
        '  guessed=', from-numeric-word-form(%samples{$lang}, :p);
}

say translate-numeric-word-form('две тысячи двадцать шесть', :from<Russian>, :to<English>);
say translate-numeric-word-form('two thousand twenty six', :from<English>, :to<Japanese>);

Output

English: told=2026  guessed=english => 2026
French: told=2026  guessed=french => 2026
German: told=2026  guessed=german => 2026
Japanese: told=2026  guessed=japanese => 2026
Russian: told=2026  guessed=russian => 2026
Spanish: told=2026  guessed=spanish => 2026
two thousand, twenty six
二千二十六

Passing :p returns a pair of the detected language and the number, which is the shape to use when the input could be in any of them. translate is the two directions composed: read in one language, write in another.

The one thing to know #

It parses nineteen languages and can only write five of them. Ask for any of the other fourteen and you get English back, with a note on standard error and a perfectly ordinary string as the return value:

File

use Lingua::NumericWordForms;

my @langs = <Bulgarian English Japanese Russian Koremutake
             Spanish German French Polish Korean>;
my $english = to-numeric-word-form(42, 'English');
for @langs -> $lang {
    my $out = to-numeric-word-form(42, $lang);
    say sprintf('%-11s %-18s %s', $lang, $out,
                ($out eq $english && $lang ne 'English') ?? 'silently English' !! '');
}
say to-numeric-word-form(42, 'Klingon');
say to-numeric-word-form(42, 'Klingon').defined;

Output

Bulgarian   четиридесет и две  
English     forty two          
Japanese    四十二                
Russian     сорок два          
Koremutake  la                 
Spanish     forty two          silently English
German      forty two          silently English
French      forty two          silently English
Polish      forty two          silently English
Korean      forty two          silently English
forty two
True

Bulgarian, English, Japanese, Russian and Koremutake have generator classes; the rest have only grammars, and the generator falls back rather than refusing. An invented language behaves the same way, so there is no way to tell a supported language from an unsupported one except by comparing the output against English and guessing. The same fallback quietly defeats translate-numeric-word-form with a :to in the other fourteen.

Watch the warning channel, too: the fallback notice goes to standard error, so a program capturing only standard output never sees it. And negative numbers throw in every language except Russian.