📄 The Job Targum from Qumran Cave 11 (11QtgJob)

Summary

This is a philological article, not an NLP paper. It presents 11QtgJob, an Aramaic rendering of Job recovered from Qumran Cave 11, and asks how this translation relates to the Hebrew original. The author first describes how the manuscript was acquired and published, and what survives of it. He then dates the copy by its handwriting. He dates the translation itself separately, using grammatical markers that place its Aramaic between two corpora of known relative age. He also discusses unusual vocabulary that can only be explained through Syriac, Arabic, Akkadian or Persian. The main claim is that 11QtgJob is the oldest targum that has come down to us. Its copy dates to about the mid‑first century C.E., but the translation probably goes back to the late second or early first century B.C.E. Comparing passages with the Masoretic text and the Septuagint shows how the targumist sometimes expanded or reinterpreted difficult Hebrew. This makes the scroll an early witness to how Jewish translators read Job.

Key points

• The Palestine Archaeological Museum bought the scroll from Bedouins, who had found the cave in 1956. The Dutch Royal Academy obtained the editing rights. The edition by van der Ploeg and van der Woude, with Jongeling, came out in French in 1971 (Brill, Leiden). It includes a 10‑page word list and facsimiles.
• The surviving material has two parts:
  • a continuous 109 cm scroll section covering Job xxxvii 10 – xlii 11, which lacks 6–7 lines at the foot of each column;
  • 27 fragments spanning Job xvii 14 – xxxvi 33.
• Dating evidence works on two levels:
  • The script points to roughly the middle of the first century C.E., and the copy must predate 70 C.E.
  • The language is younger than Biblical Aramaic but older than the Genesis Apocryphon, which places the translation around the late 2nd or early 1st century B.C.E.
• The diagnostic grammatical features are:
  • the relative particle dy versus d-;
  • the causative prefix h- versus '-;
  • the words for "if" (hn/'n) and "there" (tmh/tmn);
  • the demonstratives dn/dnh.
• The vocabulary includes items new to Aramaic:
  • the Persian word dyšt "steppe", previously known only from Syriac;
  • gs' "belch forth";
  • a word for a horse's neighing, explained through Arabic;
  • a verb for the clasping of the crocodile's scales, cognate with Arabic and Akkadian forms.

Interesting ideas

• I1. The translation is dated relatively, by checking a bundle of small grammatical markers against one older and one younger reference corpus. This avoids relying on any single feature. — For Mova (tree): A similar marker bundle could detect which UD dialect or treebank convention an input follows before conversion. Placing it between two known conventions is more reliable than a single cue.
• I2. One candidate reading is ruled out because a feminine noun cannot be the subject of a verb with a masculine prefix. Agreement serves as a hard filter on syntactic analyses. — For Mova (tree): The parser could use number and gender agreement to prune impossible subject attachments. This would give an explainable reason for each rejection.
• I3. The author argues that the translator rendered one ambiguous Hebrew form twice, once under each possible analysis. Two readings were kept side by side rather than forcing an early choice. — For Mova (frames): Ambiguous tokens could keep both role analyses alive as parallel frames. The world model would later pick between them, or keep both with a stated reason.
• I4. Lacunae and odd renderings are explained by lining up several independent versions (Masoretic text, Septuagint, targum). Where witnesses agree, the reconstruction is accepted. — For Mova (learn): This mirrors silver data agreed by several witness parsers. Disagreements between parsers would mark exactly the spots that need a teacher or a repair rule.
• I5. A doubtful word sense ("disease" rather than "earth") gains support because the same root keeps appearing with the same theme, the plagues of Egypt, in other texts. — For Mova (world): Word senses could be chosen by checking which sense the knowledge graph already links to the entities in the story. This makes the choice traceable to stored facts.