📄 A fragment from a Commentary on the Andalusi Books of Hebrew Verbs (T-S Ar 31.7)

Summary

This is a philological study, not an NLP paper. It offers an edition, English translation and analysis of two leaves from Cambridge's Genizah collection (T-S Ar 31.7). The leaves belong to an anonymous Judeo-Arabic treatise on Hebrew verbs and cover the section on verbs whose second radical is weak. The author places the text within the Andalusi genre of "books of Hebrew verbs" (kutub al-afʿāl). That genre began with Ḥayyūj in tenth-century Cordoba. Its works used morphology and analogy to assign each Biblical word to a root.

The study concludes that the fragment is a commentary rather than a true book of verbs. It works as a condensed monograph: it glosses Ḥayyūj and also gathers material from Ibn Janāḥ's Kitāb al-mustalḥaq and Kitāb al-uṣūl. That borrowing dates it after the Kitāb al-uṣūl.

The work matters for the history of Hebrew linguistics. It shows how the root catalogue started by Ḥayyūj kept being extended, compared and digested in later compilations.

Key points

• The manuscript is T-S Ar 31.7 in Cambridge: two fragments of an anonymous treatise, edited, translated and studied here. Aharon Maman independently produced a transcription for the Friedberg Genizah Project.
• Ḥayyūj's two books, on weak and on geminative verbs, catalogue up to 454 roots in four blocks: weak first radical, weak second radical, weak third radical, and geminated.
• Ibn Janāḥ's Kitāb al-mustalḥaq, written in Zaragoza after 1013:
  • proposes 63 new roots;
  • adds forms or raises questions for 128 roots already listed;
  • introduces a class of "ambiguous verbs" that includes 12 pilpel roots.
• The paper's test for telling true books of verbs from commentaries is whether the work is arranged by roots. Following Téné, the genre unfolded over about thirty years.
• The edited entries include חוד, דוך, דום, דון and דוץ. Each entry gives:
  • stem groups, such as the light verb versus hifʿil;
  • paradigm forms;
  • Biblical attestations;
  • competing views from Abū Zakariyā' (Ḥayyūj) and Abū l-Walīd (Ibn Janāḥ);
  • supporting evidence from the Targum and rabbinic sayings.

Interesting ideas

• I1. An unfamiliar form is assigned a root by matching it to a known paradigm ("according to the form of X and its analogues"), so analogy does the work of explicit inference.
• I2. Each lexicon entry is arranged as root → stem groups → paradigm → attested quotations → separate senses, so one root can hold several distinct meanings.
• I3. Rival analyses are stored with their authors named, and the commentator notes when following one authority makes another rule weaker.
• I4. The genre grows by supplements: new roots are added, and existing entries are questioned or extended, rather than the base catalogue being rewritten.
• I5. A translation into a related language (the Aramaic Targum) serves as evidence for an obscure meaning, and the text separates a noun reading from an adjectival one.

For Mova

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

• I1 (tokens): unknown English word forms could be lemmatized by aligning them to a known inflection pattern, with the analogue form recorded as the explanation.
• I2 (world): the compiled verb lexicon could keep senses under a lemma and attach attested example sentences to each sense, which keeps sense choice traceable.
• I3 (learn): when witness parsers disagree, keep each analysis tagged with its source and log which rules depend on which, rather than discarding the losing vote.
• I4 (learn): self-induced repair rules could be issued as auditable supplements to a fixed base, with counts of what each one adds and what it challenges.
• I5 (back): parallel translations through the meta-world could act as extra witnesses that settle a word's sense and part of speech.