Raku Lingua::NumericWordForms

🇦🇿 🇧🇬 🇨🇿 🇬🇧 🇫🇷 🇬🇷 🇯🇵 🇮🇷 🇵🇱 🇵🇹 🇷🇺 🇪🇸 🇺🇦

Introduction

This repository provides a Raku package with functions for the generation, parsing, and interpretation of numeric word forms in different languages.

The initial versions of the code in this repository can be found in the GitHub repository [AAr1].

The Raku package Lingua::Number, [BL1], provides word forms (cardinal, ordinal, etc.) generation in many languages. (But at least for one language the produced forms are incorrect.)

The Raku package Lingua::EN::Numbers, [SS1], also provides word forms (cardinal, ordinal, etc.) generation in English.

The parsers and interpreters of this package can be seen as complementary to the functions in [BL1, SS1].

Remark: Maybe a more complete version of this package should be merged with Lingua::Number, [BL1].

Remark: I can judge the quality of the results only of the languages: Bulgarian, English, and Russian. The numeric word form interpreters for the rest of the languages pass testing, but they might have potentially many deficiencies. (That are easily detected by people who have mastered those languages.)

Remark: The package also "understands" (i.e. parses and translates to) Koremutake.

Installation

Package installations from both sources use zef installer (which should be bundled with the "standard" Rakudo installation file.)

To install the package via Zef's ecosystem use the shell command:

zef install Lingua::NumericWordForms

To install the package from the GitHub repository use the shell command:

zef install https://github.com/antononcube/Raku-Lingua-NumericWordForms.git

Examples

Generation

The generation of numeric word forms is a secondary goal of this package. Currently only generation of Bulgarian, English, and Koremutake numeric word forms are implemented. (I plan to implement Russian word forms generation soon...) Here are examples:

use Lingua::NumericWordForms;
say to-numeric-word-form(8093);
say to-numeric-word-form(8093, 'Bulgarian');
say to-numeric-word-form(8093, 'Koremutake');

# eight thousand, ninety three
# осем хиляди и деветдесет и три
# pohy

The first argument of to-numeric-word-form can be:

An integer
A string that can be parsed into an integer
A string of numbers separated by ";"
A list of numbers or strings

Here are examples of the latter two:

to-numeric-word-form('123; 232; 898_934').join('; ');

# one hundred twenty three; two hundred thirty two; eight hundred ninety eight thousand, nine hundred thirty four

to-numeric-word-form([321, '992', 100_904]).join('; ');

# three hundred twenty one; nine hundred ninety two; one hundred thousand, nine hundred four

Interpretation

Interpretation of numeric word forms is the primary goal of this package. Multiple language are supported. Here are examples:

use Lingua::NumericWordForms;
say from-numeric-word-form('one thousand and twenty three');
say from-numeric-word-form('хиляда двадесет и три', 'Bulgarian');
say from-numeric-word-form('tysiąc dwadzieścia trzy', 'Polish');
say from-numeric-word-form('одна тысяча двадцать три', lang => 'Russian');
say from-numeric-word-form('mil veintitrés', lang => 'Spanish');

The function from-numeric-word-form can take as a first argument:

A string that is a numeric word form
A string comprised of numeric word forms separated by ";"
A list or an array of strings

The language can be specified as a second positional argument or with the named argument "lang". In addition to the names of the supported languages the value of the language argument can be also Whatever or "Automatic".

Here are corresponding examples:

from-numeric-word-form('twenty six');

# 26

from-numeric-word-form(['mil veintitrés', 'dos mil setenta y dos']);

# (1023 2072)

from-numeric-word-form('two hundred and five; триста четиридесет и две; 二十万六十五'):p;

# (english => 205 bulgarian => 342 japanese => 200065)

For more examples see the file NumericWordForms-examples.raku.

Here we retrieve a list of all supported languages:

from-numeric-word-form('languages').sort

# (azerbaijani azərbaycan bulgarian czech english español français french greek japanese korean koremutake persian polish polski portuguese português russian spanish ukrainian český ελληνικά български руский український 日本語 한국어)

Remark: In the list above some languages appear twice, with both their English and native names.

Type of the result

The returned result can be an Int object or a Str object -- that is controlled with the adverb number (which by default is True.) Here is an example:

my $res = from-numeric-word-form('one thousand and twenty three'); 
say $res, ' ', $res.WHAT;

# 1023 (Int)

$res = from-numeric-word-form('one thousand and twenty three', :!number); 
say $res, ' ', $res.WHAT;

# 1023 (Str)

Automatic language detection

Automatic language detection is invoked if the second argument is Whatever or "Automatic":

say from-numeric-word-form('tysiąc dwadzieścia trzy', Whatever):p;
say from-numeric-word-form('триста двадесет и три', lang => 'Automatic'):p;

# polish => 1023
# bulgarian => 323

say from-numeric-word-form(['tysiąc dwadzieścia trzy', 'twenty three']):p;

# (polish => 1023 english => 23)

The adverb :pairs (:p) specifies whether the result should be a Pair object or a List of Pair objects with the detected languages as keys.

Translation

Translation from one language to another:

translate-numeric-word-form('хиляда двадесет и три', 'Bulgarian' => 'English');

# one thousand, twenty three

translate-numeric-word-form('two hundred thousand and five', 'English' => 'Bulgarian');

# двеста хиляди и пет

Remark: Currently that function translates to Bulgarian, English, and Koremutake only.

Here is are Russian to Koremutake example:

my $numForm = "три тысячи восемьсот девяносто";
my $trRes = translate-numeric-word-form($numForm, 'Automatic' => 'Koremutake');
say "Given           : $numForm";
say "To Koremutake   : $trRes";
say "From Koremutake : {from-numeric-word-form($trRes)}";

# Given           : три тысячи восемьсот девяносто
# To Koremutake   : jami
# From Koremutake : 3890

The named arguments "from" and "to" can be also used:

translate-numeric-word-form($numForm, from => Whatever, to => 'English');

# three thousand, eight hundred ninety

Roles

This package provides (exports) roles that can be used in grammars or roles in other packages, applications, etc.

For example, see the roles:

Lingua::NumericWordForms::Roles::Bulgarian::WordedNumberSpec
Lingua::NumericWordForms::Roles::English::WordedNumberSpec

A grammar or role that does the roles above should use the rule:

<numeric-word-form>

For code examples see the file Parsing-examples.raku.

Remark: The role Lingua::NumericWordForms::Roles::WordedNumberSpec and the corresponding actions class Lingua::NumericWordForms::Actions::WordedNumberSpec are "abstract". They were introduced in order to have simpler roles and actions code (and non-duplicated implementations.) Hence, that role and class should not be used in grammars and roles outside of this package.

TODO

The following TODO items are ordered by priority, the most important are on top.

TODO Expand parsing beyond trillions
DONE Automatic determination of the language
TODO Word form generation:
- DONE Bulgarian
- DONE English
- DONE Koremutake
- TODO Russian
- TODO General algorithm
TODO Documentation of the general programming approach.
- TODO What are the main challenges?
- TODO How the chosen software architecture decisions address them?
- TODO Concrete implementations walk-through.
- TODO How to implement / include a new language?
- TODO How the random numbers test files were made?
- TODO Profiling, limitations, alternatives.
TODO Full, consistent Persian numbers parsing.
- Currently, Persian number parsing works only for numbers less than 101.
DONE General strategy for parsing and interpretation of numeric word forms of East Asia languages
- Those languages use groupings based on 10^4 instead of 10^3.
- DONE Implementation for Japanese.
TODO Implement parsing of ordinal numeric word forms
- DONE English, French, Greek, and Spanish
- DONE Bulgarian
- DONE Czech, Russian, Ukrainian, Polish
- DONE Japanese
- DONE Koremutake
- DONE Portuguese
- DONE Azerbaijani
- TODO Korean
  - Implemented to a point.
- TODO Persian
  - Implemented to a point.
- TODO Sanskrit
TODO Implement parsing of year "shortcut" word forms, like "twenty o three"
TODO Implement parsing of numeric word forms for rationals, like "five twelfths"
DONE Translation function (from one language to another)

Collaboration notes

The main rule is that the main branch should always be installable and pass all of its tests.
From the main rule it follows that new features are developed in separate branches or forks.
The easiest way to collaborate is to create and commit new test files or corrections to existing test files.
- Then I would change the corresponding grammars rules and actions in order the package to pass the tests.
Please use Conventional Commits (CC).
- Here is the CC short form stencil (in Raku): <type> ['(' <scope> ')']? ':' <description>.
- See the recent commits in this repository for examples.
- Here are additional examples of CC messages (each line is a separate message):

feat:Implemented the parsing of Danish numeric word forms.
docs:Added documentation of right-to-left word forms parsing.
fix(Persian):Corrected tests for numbers larger that 1000.
test:Added new corner cases tests.
test(Ukrainian):Added new tests.

Acknowledgements

Thanks to spyrettas for:
- Riding "shotgun" during the initial implementation of the Greek role, actions, and tests
- Proofreading and correcting Greek tests and role
Thanks to Denis for:
- Proofreading the Czech language unit tests and suggesting corrections.

References

[AAr1] Anton Antonov, Raku::DSL::Shared.

[BL1] Brent "Labster" Laabs, Lingua::Number.

[SS1] Larry Wall, Steve Schulze, Lingua::EN::Numbers.

Anton Antonov
Florida, USA
April-May, 2021