📄 Michael A. Knibb: Translating the Bible: The Ethiopic Version of the Old Testament

Summary

This item is not an NLP paper. It is Steven Kaplan's short book review in Aethiopica 4 (the PDF header dates the issue 2001). The work under review is Michael A. Knibb's Translating the Bible, subtitled The Ethiopic Version of the Old Testament, which grew out of his 1995 Schweich Lectures and was issued by Oxford University Press in 1999. The book asks how the Ge'ez Old Testament came into being: which source texts it was made from and revised against, and how faithfully its translators worked. Knibb draws on a much larger pool of newly catalogued manuscripts and on his own critical edition of Ethiopic Ezekiel. He argues for a Greek original, a later revision against an Arabic text descended from Syriac, and a third revision against the Hebrew in the 15th or 16th century. He also finds that the translation is mostly literal but contains many free renderings, outright errors and inconsistent word choices. Kaplan praises the scholarship and the clarity of the historical chapter. He objects to Knibb's quick dismissal of the view that the translators were Ethiopians, and to the book's frequent typos.

Key points

• The book has 145 pages in three chapters: manuscripts and textual history (pp. 1–54), translation technique (pp. 55–87) and inconsistency in rendering individual words (pp. 87–112).
• It revisits themes from Edward Ullendorff's Ethiopia and the Bible (Schweich Lectures 1967), to whom the book is dedicated. Against Ullendorff, Knibb argues for a late revision based on the Hebrew.
• Manuscript cataloguing by Getatchew Haile, William Macomber, Veronika Six and Ernst Hammerschmidt widened the evidence base. That evidence mostly supports a Greek Vorlage followed by a revision from a Syriac-derived Arabic text.
• Knibb challenges the common idea that Syriac-speaking monks did the translating. Kaplan argues that the misreadings of the Greek point to local Ethiopian translators, as Paolo Marrassini and Getatchew Haile had suggested.
• Kaplan counts typographical errors at up to two per page.

Interesting ideas

• I1. Knibb grades a translation along a spectrum from literal to free to mistaken, using concrete aligned examples, instead of giving a single overall verdict.
• I2. He checks whether the same source word is rendered the same way every time it appears, and treats inconsistency as evidence in its own right.
• I3. Knibb reconstructs the text's history in layers: an original source plus later revisions against other sources, each detected from distinct traces in the witnesses.
• I4. Kaplan uses the kind of errors made, misunderstanding the source language rather than the target, to infer who produced the text.

For Mova

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

• I1 (back): Our translation through the meta-world could tag each rendered clause as literal, free or erroneous against the source graph, which gives an explainable fidelity profile in place of one score.
• I2 (learn): A lexical-consistency audit across silver data could flag lemmas or frames that the pipeline maps differently in identical contexts, as candidates for repair rules.
• I3 (learn): When several witness parsers disagree, the disagreements could be attributed to the specific parser or training layer that introduced them, so silver data keeps its provenance.
• I4 (answer): The type of each error made by a reader or student module (misparse versus missing world knowledge) can diagnose which pipeline stage failed, which is more useful than raw accuracy.