📄 The Effect of Ottoman Calligraphy on Arabic Grammar

Summary

This Arabic-language philological study examines how the Uthmanic orthography (al-rasm al-ʿUthmānī, rendered in the English abstract as "Ottoman calligraphy") relates to Arabic grammar. The Uthmanic orthography is the spelling convention of the Quran codex compiled under Caliph Uthman ibn Affan. Using a descriptive method, the authors trace:

• how the Quran was recorded in the Prophet's time and collected under Abu Bakr and Uthman;
• the origins of Arabic script in Nabataean and Syriac Aramaic hands;
• the rules of the Uthmanic spelling, such as letter deletion and retention.

They also discuss how this spelling relates to the variant Quranic readings, to grammatical error (laḥn), and to scribal distortion. They conclude that:

• the codex spelling originally matched ordinary writing of its era and was refined later;
• its spelling is dominant but not fully consistent;
• it reflects the Basran–Kufan grammatical disputes.

Scholars disagree on rewriting the Quran in modern spelling: some forbid it, some permit it, and some require it. The study matters because the script deliberately preserves several valid readings and records traces of old Arabic dialects and original letter forms.

Key points

• The paper distinguishes three kinds of writing:
  • standard (qiyāsī) spelling;
  • prosodic spelling, which follows actual pronunciation, e.g. writing nunation and dropping the connective hamza;
  • codex spelling, which follows the Companions' precedent.
• Following al-Qastallani, codex spelling is split into regular (matches pronunciation) and conventional (iṣṭilāḥī, deviates from it). Rules include deletion and retention, e.g. three kinds of alif deletion: indicative, abbreviating and restricting.
• Example: in Al-Kahf 18:17 the alif is dropped in tazāwaru, so the written form also accommodates another reading, with sukūn on the zay and doubled ra'. The paper attributes that reading to Ibn Amir and Yaʿqub.
• Uthman's copying committee consisted of Zayd ibn Thabit, Abdullah ibn al-Zubayr, Saʿid ibn al-As and Abd al-Rahman ibn al-Harith ibn Hisham. They were told to resolve disagreements in favour of the Quraysh dialect. Copies were sent to the provinces and divergent codices were burned.
• Under Abu Bakr, Zayd accepted a written fragment only when two witnesses confirmed it had been written before the Prophet. He also checked it against what people had memorised.

Interesting ideas

• I1. A single written form can be kept underspecified on purpose so that it encodes several legitimate readings at once, instead of committing early to one.
• I2. Spellings are sorted by whether they follow a rule (qiyāsī), follow pronunciation, or follow inherited convention, and every deviation is named and typed.
• I3. When the committee disagreed, it used a fixed reference variety (the Quraysh dialect) as the tie-breaker.
• I4. A fragment entered the canonical collection only if two independent witnesses confirmed it and it also matched a second source channel (memory).
• I5. The paper openly records that the same word is spelled differently in different places, and treats this as evidence rather than noise.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

• I1 (tokens): The tagger and lemmatizer could keep a packed set of analyses for an ambiguous token and let later stages resolve it, rather than forcing a single choice.
• I2 (tokens): Tokenizer normalisation rules could carry an explicit type (rule-based, phonetic, conventional), making each rewrite explainable.
• I3 (tree): When converting between UD dialects or merging treebanks, a declared reference convention can settle annotation conflicts deterministically and transparently.
• I4 (learn): Silver trees could be admitted only when several witness parsers agree and an independent check, such as a rule or an LLM-teacher tier, also confirms them.
• I5 (learn): Evaluation could log inconsistent analyses of the same construction across a corpus as diagnostic signals for self-induced repair rules.