Skip to content

Latest commit

 

History

History
211 lines (186 loc) · 10.7 KB

README.md

File metadata and controls

211 lines (186 loc) · 10.7 KB
Raw

AsoSoft Library (Python)

AsoSoft Library offers the following natural language processing (NLP) algorithms for the Kurdish Language (ckb: Central branch of Kurdish):

  • Grapheme-to-Phoneme (G2P) converter and Transliterator: converts Kurdish text into syllabified phoneme string. Also transliterates Kurdish texts from Arabic script into Latin script and vice versa.
  • Normalizer: normalizes the Kurdish text and punctuation marks, unifies numerals, replaces Html Entities, extracts and replaces URLs and emails, and more.
  • Numeral Converter: converts any type of numbers into Kurdish words.
  • Sort: Sorts a list in correct Kurdish alphabet order.
  • Poem Meter Classifier: Classifies the meter of the input Kurdish poem

AsoSoft Library is originally written in C# by Aso Mahmudi and this library is its Python port.

How to use?

  • Requierements: Python 3.8+
  • Install the package using pip: pip install asosoft
  • Import the package in your Python file:
import asosoft

Grapheme-to-Phoneme (G2P) converter and Transliteration

This function is based on the study "Automated Grapheme-to-Phoneme Conversion for Central Kurdish based on Optimality Theory".

Kurdish G2P converter

Converts Central Kurdish text in standard Arabic script into syllabified phonemic Latin script (i.e. graphemes to phonems)

General format:

asosoft.KurdishG2P(text, convertNumbersToWord = False, backMergeConjunction = True, singleOutputPerWord = True)

An example:

>>> print(asosoft.KurdishG2P("شەو و ڕۆژ بووین بە گرفت. درێژیی دیوارەکەی گرتن"))
ˈşeˈwû ˈřoj ˈbûyn ˈbe ˈgiˈrift. ˈdiˈrêˈjîy ˈdîˈwaˈreˈkey ˈgirˈtin

Transliteration

Arabic script into Hawar Latin script (ح‌غ‌ڕڵ→ḧẍřł):

>>> print(asosoft.Ar2La("گیرۆدەی خاڵی ڕەشتە؛ گوێت لە نەغمەی تویوورە؟"))
gîrodey xałî řeşte; gwêt le neẍmey tuyûre?

Arabic script into the Latin script suggested by Dr. Feryad Fazil Omar:

>>> print(asosoft.Ar2LaFeryad("گیرۆدەی خاڵی ڕەشتە؛ گوێت لە نەغمەی تویوورە؟"))
gîrodey xaḻî ṟeşte; gwêt le nex̱mey tuyûre?

Arabic script into simplified (ح‌غ‌ڕڵ→hxrl) Hawar Latin script:

>>> print(asosoft.Ar2LaSimple("گیرۆدەی خاڵی ڕەشتە؛ گوێت لە نەغمەی تویوورە؟"))
gîrodey xalî reşte; gwêt le nexmey tuyûre?

Latin script (Hawar) into Arabic script:

>>> print(asosoft.La2Ar("Gelî keç û xortên kurdan, hûn hemû bi xêr biçin"))
گەلی کەچ و خۆرتێن کوردان، هوون هەموو ب خێر بچن

Arabic script into IPA:

>>> print(asosoft.Phonemes2IPA(asosoft.KurdishG2P("شەو و ڕۆژ بووین بە گرفت. درێژیی دیوارەکە گرتن")))
ʃa·wu ro̞ʒ bujn ba gɪ·ɾɪft. dɪ·ɾɛ·ʒij di·wä·ɾa·ka gɪɾ·tɪn

Kurdish Text Normalizer

Several functions needed for Central Kurdish text normalization:

Normalize Kurdish

Two character replacement lists are provided as the resources of the library:

  • Deep Unicode Corrections:
    • replacing deprecated Arabic Presentation Forms (FB50–FDFF and FE70–FEFF) with corresponding standard characters.
    • replacing different types of dashes and spaces
    • removing Unicode control character
  • Additional Unicode Corrections
    • replacing special Arabic math signs with corresponding Latin characters
    • replacing similar, but different letters with standard characters (e.g. ڪ,ے,ٶ with ک,ی,ؤ)

The normalization task in this function:

  • for all Arabic scripts (including Kurdish, Arabic, and Persian):
    • Character-based replacement:
      • Above mentioned replacement lists
      • Private Use Area (U+E000 to U+F8FF) with White Square character
  • Standardizing and removing duplicated or unnecessary Zero-Width characters
  • removing unnecessary Tatweels (U+0640)
  • only for Central Kurdish:
    • standardizing Kurdish characters: ە, هـ, ی, and ک
    • correcting miss-converted characters from non-Unicode fonts
    • replacing word-initial ر with ڕ

the simple overloading:

>>> print(asosoft.Normalize("دەقے شیَعري خـــۆش. ره‌نگه‌كاني خاك"))
دەقی شێعری خۆش. ڕەنگەکانی خاک

or the complete overloading:

>>> asosoft.Normalize(text, isOnlyKurdish=True, changeInitialR=True, deepUnicodeCorrectios=True, additionalUnicodeCorrections=True, usersReplaceList)

AliK to Unicode

AliK2Unicode converts Kurdish text written in AliK fonts (developed by Abas Majid in 1997) into Unicode standard. Ali-K fonts: Alwand, Azzam, Hasan, Jiddah, kanaqen, Khalid, Sahifa, Sahifa Bold, Samik, Sayid, Sharif, Shrif Bold, Sulaimania, Traditional

>>> print(asosoft.AliK2Unicode("ئاشناكردنى خويَندكار بة طوَرِانكاريية كوَمةلاَيةتييةكان"))
ئاشناکردنی خوێندکار بە گۆڕانکارییە کۆمەڵایەتییەکان

AliWeb to Unicode

AliWeb2Unicode converts Kurdish text written in AliK fonts into Unicode standard. Ali-Web fonts: Malper, Malper Bold, Samik, Traditional, Traditional Bold

>>> print(asosoft.AliWeb2Unicode("هةر جةرةيانصکي مصذووُيي کة أوو دةدا"))
ھەر جەرەیانێکی مێژوویی کە ڕوو دەدا

Dylan to Unicode

Dylan2Unicode converts Kurdish text written in Dylan fonts (developed by Dylan Saleh at KurdSoft in 2001) into Unicode standard.

>>> print(asosoft.Dylan2Unicode("لثكؤلثنةران بؤيان دةركةوتووة كة دةتوانث بؤ لةش بةكةصك بث"))
لێکۆلێنەران بۆیان دەرکەوتووە کە دەتوانێ بۆ لەش بەکەڵک بێ

Zarnegar to Unicode

Zarnegar2Unicode converts Kurdish text written in Zarnegar word processor (developed by SinaSoft with RDF converter by NoorSoft) and into Unicode standard.

>>> print(asosoft.Zarnegar2Unicode("بلٌيٌين و بگه‌رٍيٌين بوٌ هه‌لاٌلٌه‌ى سىٌيه‌مى فه‌لسه‌فه"))
بڵێین و بگەڕێین بۆ هەڵاڵەی سێیەمی فەلسەفە

NormalizePunctuations

NormalizePunctuations corrects spaces before and after of the punctuations. When seprateAllPunctuations is true,

>>> print(asosoft.NormalizePunctuations("دەقی«کوردی » و ڕێنووس ،((خاڵبەندی )) چۆنە ؟", false))
دەقی «کوردی» و ڕێنووس، «خاڵبەندی» چۆنە؟

Trim Line

Trim starting and ending white spaces (including zero width spaces) of line, TrimLine

>>> print(TrimLine("   دەق\u200c  "))
دەق

Replace Html Entities

ReplaceHtmlEntity replaces HTML Entities with single Unicode characters (e.g. "é" with "é"). It is useful in web crawled corpora.

>>> print(asosoft.ReplaceHtmlEntity("ئێوە "دەق" بە زمانی <کوردی> دەنووسن"))
ئێوە "دەق" بە زمانی <کوردی> دەنووسن

Replace URLs and emails

ReplaceUrlEmail replaces URLs and emails with a certain word. It improves language models.

Unify Numerals

UnifyNumerals unifies numeral characters into desired numeral type from en (0123456789) or ar (٠١٢٣٤٥٦٧٨٩)

>>> print(asosoft.UnifyNumerals("ژمارەکانی ٤٥٦ و ۴۵۶ و 456", "en"))
ژمارەکانی 456 و 456 و 456

Seperate Digits from words

SeperateDigits add a space between joined numerals and words (e.g. replacing "12کەس" with "12 کەس"). It improves language models.

>>> print(asosoft.SeperateDigits("لە ساڵی1950دا1000دۆلاریان بە 5کەس دا"))
لە ساڵی 1950 دا 1000 دۆلاریان بە 5 کەس دا

Word to Word Replacment

Word2WordReplacement applies a "string to string" replacement dictionary on the text. It replaces the full-matched words not a part of them.

>>> print(asosoft.Word2WordReplacement("مال، نووری مالیکی", {"مال": "ماڵ", "سلاو": "سڵاو"}))
ماڵ، نووری مالیکی

Character to Character Replacment

Char2CharReplacment applies a "char to char" replacement dictionary on the text. It uses as the final step needed for some non-Unicode systems.

Kurdish Numeral converter

It converts numerals into Central Kurdish words. It is useful in text-to-speech tools.

  • integers (1100 => )
  • floats (10.11)
  • negatives (-10.11)
  • percent (100% or %100)
  • querency marks ($100, £100, and €100)
>>> print(asosoft.Number2Word("لە ساڵی 1999دا بڕی 40% لە پارەکەیان واتە $102.1 یان وەرگرت"))
لە ساڵی هەزار و نۆسەد و نەوەد و نۆدا بڕی چل لە سەد لە پارەکەیان واتە سەد و دوو پۆینت یەک دۆلاریان وەرگرت

Kurdish Sort

Sorting a string list in correct order of Kurdish alphabet ("ئءاآأإبپتثجچحخدڎذرڕزژسشصضطظعغفڤقكکگلڵمنوۆۊۉهھەیێ")

>>> myList = ["یەک", "ڕەنگ", "ئەو", "ئاو", "ڤەژین", "فڵان"]
>>> print(asosoft.KurdishSort(myList))
["ئاو", "ئەو", "ڕەنگ", "فڵان", "ڤەژین", "یەک"]

or using your custom order:

>>> input_list = ["یەک", "ڕەنگ", "ئەو", "ئاو", "ڤەژین", "فڵان"]
>>> input_order = list("ئءاآأإبپتثجچحخدڎڊذرڕزژسشصضطظعغفڤقكکگڴلڵمنوۆۊۉۋهھەیێ")
>>> print(asosoft.CustomSort(input_list, input_order))
["ئاو", "ئەو", "ڕەنگ", "فڵان", "ڤەژین", "یەک"]

Poem Meter Classifier

It classifies the meter of the input Kurdish poem typed in Arabic script. The lines of the poem should be seprated by new line char ('\n'). You can find Kurdish poems in https://books.vejin.net/.

>>> poem = f"گەرچی تووشی ڕەنجەڕۆیی و حەسرەت و دەردم ئەمن\nقەت لەدەس ئەم چەرخە سپڵە نابەزم مەردم ئەمن\nمن لە زنجیر و تەناف و دار و بەند باکم نییە\nلەت لەتم کەن، بمکوژن، هێشتا دەڵێم کوردم ئەمن"
>>> classified = asosoft.ClassifyKurdishPoem(poem)
>>> print("Poem Type= " + classified.overalMeterType)
Quantitative/عەرووزی
>>> print("Poem Meter= " + classified.overalPattern)
فاعلاتن فاعلاتن فاعلاتن فاعلن