abdelhaqueidali/Quran-kabyle-ayt-mensour

Dataset Card: Quran Kabyle Translation (Ramdane At Mensour)

Dataset Summary

This dataset contains the Kabyle (Taqbaylit / Amazigh) translation of the Holy Quran titled "LEQWṚAN S TMAZIƔT", translated by Ramdane At Mensour (Remḍan At Menṣuṛ). It provides verse-by-verse alignments across three script representations: legacy custom-encoded ASCII, standardized INALCO Latin, and IRCAM Tifinagh.

Previously, digital distributions of this translation across mobile apps, websites, and databases suffered from severe mojibake. The original text relied on legacy 8-bit ASCII fonts where standard characters were repurposed to display specific Amazigh glyphs, whether it was the Tifinagh version or Latin depend on the font applied to get the two different versions. When rendered without those dedicated fonts, the text displayed incorrect characters (e.g., Öebbi instead of Ṛebbi, â instead of ɛ). This dataset fixes the legacy encoding programmatically and derives a standardized, orthographically aligned parallel corpus.

The original content relis on the original ASCII version. INALCO-Latin and IRCAM-Tifinagh are an automated conversion from ASCII-Latin.

A copy of the version used here can be found at: https://surahquran.com/Surah-translation/Quran-language-17.html , the ascii issue is in it. Also note that it is using the audios of another version, the Hadj Mohand Tayeb version which no digital version is found yet, scanned pdf pages can be found in internet archive though.

The version at https://tanzil.net/trans/ uses a different ASCII-Latin encoding form. (seems letter A plus another character, like Ä- corresponds to Ö which is Ṛ.)

This version seems to follows the Kufan / Makkan system (most notably the standard Hafs ‘an ‘Asim counting).

• —Verse 1 is the Basmala ("S yisem n Ṛebbi, Aḥnin Itḥunun").
• —Verse 7 combines the entire final passage ("Abrid n wid i Tɣateḍ; ur n wid i K isserfayen..." / Ṣirāṭ allaḏhīna...) into a single verse.

A copy is found at Internet Archive but with burnt-in text: https://archive.org/details/LEQRAN.pdfAneruUeremSurRemanAtMansur

Details of the Original physical copy (For any research needed about the Author and his translation)

• —Title: LEQWṚAN S TMAZIƔT (القرآن بالأمازيغية)
• —Translator: Ramdane At Mensour (رمضان آث منصور)
• —Language & Variety: Amazigh / Berber (Kabyle / Taqbaylit / kab)
• —Publisher: Dar Zyriab (دار النشر زرياب), Algiers (2006)
• —Institutional Support: Published with the support of the Algerian Ministry of Culture, featuring a concluding endorsement letter from the Minister of Religious Affairs and Endowments.
• —Original Physical Format: 491 pages (15.5 × 23 cm), featuring facing-page parallel scripts (Latin script on the left page, Tifinagh on the right page), a one-page French introduction, a four-page Surah index, and accompanying audio CD distributions.
• —Translator's Script Philosophy: The translator explicitly noted in the original work that Tifinagh serves as the authentic indigenous writing system of Tamazight, considering Latin script usage a transitional phase.

Dataset Structure

Data Fields

• —`surah_number` (integer): Surah (chapter) number (1–114).
• —`ayah_number` (integer): Ayah (verse) number within the Surah.
• —`origin_ascii` (string): The uncorrected legacy text relying on vintage custom font mappings.
• —`latin` (string): Standard Amazigh Latin orthography.
• —`tifinagh` (string): Standard Neo-Tifinagh orthography.

Example Row

surah_number,ayah_number,origin_ascii,latin,tifinagh
1,1,"S yisem n Öebbi, Aênin Ipêunun,","S yisem n Ṛebbi, Aḥnin Itḥunun,","ⵙ ⵢⵉⵙⵎ ⵏ ⵕⴱⴱⵉ, ⴰⵃⵏⵉⵏ ⵉⵜⵃⵓⵏⵓⵏ,"

Conversion & Normalization Pipeline

1. Legacy ASCII to Standard INALCO Latin

The raw text was mapped to standard Unicode Amazigh Latin characters following the INALCO standard:

2. Standard Latin to IRCAM Tifinagh & Internal e Elision

• —Transliteration: Each Latin grapheme is mapped to its official Unicode Tifinagh equivalent (e.g., ɛ -> ⵄ, ɣ -> ⵖ, ṛ -> ⵕ).
• —Orthographic Differences:
• —Latin (INALCO standard): Retains explicit neutral vowels / schwas (e), representing phonetic epenthesis (e.g., yisem). The segmentation of the original text relies only on spaces, so no INALCO-Hyphenation.
• —Tifinagh (IRCAM standard): Internal schwas (ⴻ) occurring between consonants within words are omitted per IRCAM orthographic conventions, and starting shwa as well (e.g., ⵢⵉⵙⴻⵎ -> ⵢⵉⵙⵎ, ⴻⴷ => ⴷ). Word-boundary and essential standalone occurrences are preserved as the original. ---

Applications & Use Cases

• —Machine Translation (MT): Parallel corpus pairing with Arabic, English, or French Quranic corpora.
• —Low-Resource NLP: Tokenization, language modeling, and POS-tagging for Northern Amazigh and Kabyle dialects.