📄 The Syro-Aramaic Reading of the Koran

Summary

This is a book review rather than an NLP paper. Langermann assesses the English edition of Christoph Luxenberg's (a pseudonym) contested study of the Qur'an. He separates two claims. The broad one says the Qur'an began as a lectionary of biblical material, written in Karshuni (Syriac in Arabic letters) with an early script that lacked diacritical dots, so later readers who assumed Arabic misread it. The narrower one is a large set of word-level reinterpretations built on Syro-Aramaic roots.

The reviewer finds the broad thesis weak. It leaves unclear who fixed the text and reduces Muhammad to someone who could not explain his own verses. Many of the individual readings, however, are judged worth taking seriously. His main methodological criticism is that Luxenberg's sound and spelling changes follow no explicit, consistently applied rules, which makes the conclusions look arbitrary. Luxenberg also ignores standard comparative Semitic grammars. The review closes with notes on flaws in the translation, which matter for anyone weighing the book's reception in English.

Key points

• Reviewed book: The Syro-Aramaic Reading of the Koran, Hans Schiler, Berlin 2007, 349 pp. The English text follows mostly the first German edition, although a third edition already exists.
• Core mechanism of the theory: a Karshuni text written without diacritical dots, later "pointed" by poorly qualified "originators" (p. 37). The misreading of tawra is blamed on Jewish informants (p. 92), and that of Ibrahim on a misread kursi stroke (p. 93).
• Case study al-raqim (18:9, pp. 80–85):
  • Luxenberg reconstructs al-raqid from the root r-q-d and argues that readers changed it to an existing Arabic word.
  • The reviewer counters that keeping raqid would have been more reasonable. The form is morphologically valid, works as an intensive, and means "deep sleeper."
• A strong point, by the reviewer's own account: final alif in mathalan (11:24) and al-hawaya (6:146) is explained better by Aramaic forms than by the strained accusative reading (pp. 44–45).
• Omitted references: the comparative grammars of William Wright, De Lacy O'Leary and Edward Lipinski.
• Translation slips:
  • "arabisch" left untranslated twice (p. 90);
  • German "j" kept where English needs "y";
  • fremde rendered as "strange" instead of "foreign."

Interesting ideas

• I1. The review says a reconstruction is only convincing when every transformation comes from an explicit rule set applied the same way each time; one-off derivations look arbitrary however clever they are. — For Mova (learn): Self-induced repair rules should be logged as a named, reusable inventory and applied uniformly, so each fix can be audited and tested against negative controls.
• I2. Script that drops distinguishing marks, like undotted letters, creates ambiguity that readers resolve using their assumptions about which language they are reading. — For Mova (tokens): When input has lost signals such as casing, punctuation or diacritics, the tagger should keep several candidate readings with stated assumptions instead of quietly committing to one.
• I3. The raqid argument separates "the word exists in the lexicon" from "the form is morphologically legitimate and fits the context." Snapping to a known word can be worse than accepting a well-formed unseen one. — For Mova (tokens): For out-of-vocabulary items, prefer a productive morphological analysis that fits the context over a forced correction to the nearest dictionary word.
• I4. The review checks a theory partly by how it describes the agents involved: it asks who produced or understood the text and whether their assigned roles fit together. — For Mova (frames): When building who-did-what frames, check that roles given to the same entity across a text stay consistent, and flag conflicts as candidate errors.
• I5. The translation flaws found here (untranslated leftovers, pronunciation-specific spellings carried over, false friends like fremde/strange) form a ready checklist of failure types. — For Mova (back): Translation through the meta-world could be evaluated with explicit probes for leftover source words, transliteration conventions and false-friend lexical choices.