📄 From Glosses to the Linguistic Nature of Canaano-Akkadian

Summary

The paper asks what kind of linguistic system Canaano-Akkadian was. This is the hybrid Akkadian used by Canaanite scribes in the 14th-century BCE Amarna letters, and scholars disagree whether it was a real dialect, an interlanguage, a mixed language, a social-scribal code, or simply Canaanite written with Akkadian cuneiform. The authors avoid the problem of telling individual habits apart from dialect features by narrowing the evidence to one subset: Canaanite glosses that the same scribe wrote more than once. Most of these repeated glosses vary in spelling, while a few stay fixed. From this the authors infer that scribes treated their native Canaanite as a low variety beside prestigious Akkadian, which points to a diglossic setting. They then test the code against the defining traits of pidgins, creoles, koinés, mixed languages and jargons. No single category fits, so they propose a dynamic, partly fuzzy, multi-category definition. They argue this definition reconciles most earlier hypotheses as complementary views of different aspects of the language.

Key points

• The study draws on about thirty repeated same-scribe glosses, out of roughly 110 glosses in total (counts based on Izre'el 1998b).
• Some glosses are spelled consistently. A Gezer scribe wrote la-bi-tu for SIG4 'brick' in both EA 266:20 and EA 296:17, and Byblos scribes always wrote ki-lu-bi 'cage'. Unstable spellings are more frequent, for example a pharyngeal or laryngeal sometimes written and sometimes omitted, or a different syllabogram chosen.
• Items whose origin is uncertain, such as pû 'mouth', are excluded, so that only genuinely Canaanite elements are analysed.
• The final classification describes the code as:
  • a professional high-status jargon, written but also spoken among scribes;
  • a two-source mixed language;
  • a "soft" koiné built on related but mutually unintelligible systems;
  • a factor in a triglossic setting at scribal centres;
  • a system showing pidgin-like tertiary hybridization and post-pidgin continuum tendencies.
• The proposal brings together the views of Rainey, Kossmann, Gianto, Izre'el and Sanders. Sheng in Kenya is cited as a comparable case of a language that resists any single category.

Interesting ideas

• I1. Using only repetitions by the same writer controls for the individual, so the remaining variation says something about that writer's attitude rather than about dialect differences. — For Mova (learn): When evaluating parsers or LLM teachers, compare outputs on repeated items from a single source, so that source-specific noise is not mistaken for genuine linguistic variation.
• I2. How consistently a writer spells a word can reveal how much prestige that writer gives the word's language or register. — For Mova (tokens): Variant-spelling statistics could serve as a register feature in the tokenizer, flagging informal or low-status spans that need lenient normalization before lemmatization.
• I3. A single object can be described as belonging partly to several categories at once, instead of being forced into one rigid class. — For Mova (world): The ontology layer could let entities carry several class memberships with graded strength, so that reasoning stays honest when WordNet or Wikidata types conflict.
• I4. Competing theories are treated as complementary views of different aspects of the same thing, each fitting part of the data. — For Mova (learn): When witness parsers disagree, the system could record which data subsets each one explains rather than taking a flat majority vote, and use that record to route silver labels.
• I5. The authors propose as future work that every individual trait be linked to the category it supports, making the overall classification auditable. — For Mova (frames): Each frame or speech-act label could be stored with the specific cues that triggered it, keeping the classification explainable trait by trait.