📄 Towards a Comprehensive Edition of the Arabic-Ethiopic Glossary of al-Malik al-Afḍal Part I: New Readings from the First Sheet

Summary

The paper revisits a remarkable late-14th-century Yemeni manuscript. It is a glossary of 475 Arabic words, each given equivalents in several South Ethiopian (Ethio-Semitic) varieties written in Arabic script. Franz-Christoph Muth first identified and published it (Muth 2009–2010). Bulakh and Kogan had been preparing a full monograph edition on their own since 2009, so they could check their readings against his. In this first preliminary note they take 34 entries from the glossary's first sheet where Muth either offered no identification or one they find weak. For each entry they propose a reading of the Arabic headword and the Ethiopic gloss, reconstruct a proto-form, and support it with cognates from Geʿez, Tigrinya, Amharic, Harari, Argobba, Gafat and many Gurage varieties. They rely mainly on Leslau's and Kane's dictionaries. The arguments combine handwriting evidence, regular sound changes and plausible meaning shifts. The new readings fill gaps, correct several earlier identifications, and give early evidence on the historical phonology and dialect spread of Ethiopian Semitic.

Key points

• The glossary has 475 Arabic lexemes and belongs to a codex of al-Malik al-Afḍal. Daniel M. Varisco found the codex in a private Yemeni collection, and in 1998 he and G. Rex Smith brought out an edition of it.
• The note covers 34 entries from sheet 1 (p. 217 of the edition), including margin glosses that Muth skipped. Examples are two Ethiopic words for "woman" and wänd "young man".
• Entries are cited with a positional scheme: six double columns A–F plus the row number (e.g. 217 A 15), with alternative glosses labelled Gloss A–D.
• Several of Muth's readings are rejected on palaeographic grounds. In one case the Arabic entry was read as "palms of the hands", but the authors re-identify it as "nape of the neck" because both of its Ethiopic glosses mean that.
• The authors note features the attested forms lack, such as spirantized *k, palatalization and extra final -t. They also openly admit where their identifications stay uncertain. The project has been funded by RFH (No. 12-04-00092a) since 2012.

Interesting ideas

• I1. Two teams deciphered the same document independently and then compared results, which shows where a single interpretation is fragile and where readings converge.
• I2. When a headword is illegible, the meanings of its several translations constrain what it must mean, as with the "nape of the neck" entry.
• I3. A reading is accepted only when three kinds of evidence (letter shapes, sound correspondences, meaning) agree, and any leftover mismatch is stated rather than hidden.
• I4. Variant word forms are linked through known regular processes, such as spirantization or palatalization near y, instead of loose similarity.
• I5. Documented semantic shifts (tail to penis, quail to cock) are used to license a gloss whose meaning differs from the expected cognate.

For Mova

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

• I1 (learn): Treat disagreements between independent witness parsers as the priority list for review, as the authors did with Muth's readings, rather than trusting either source alone.
• I2 (world): When one mention is ambiguous, intersect the senses of its paraphrases or co-referring expressions to pin down the entity or concept.
• I3 (learn): Score silver annotations on several independent checks and record the specific residual conflict, so each accepted label carries an explainable confidence.
• I4 (tokens): Normalize spelling and dialect variants to lemmas through explicit, named rewrite rules, so every lemma choice can be traced to a rule.
• I5 (world): Keep a catalogue of attested meaning-shift patterns in the ontology, so near-miss senses can be linked with a stated reason rather than rejected or silently merged.