📄 Lines on the classification of Ethopian-Semitic

Summary

Fellman addresses a stalled debate: how the Ethiopian-Semitic languages should be grouped internally. The last full proposal, Hetzron (1972), was heavily criticized by Goldenberg (1977), and no accepted scheme replaced it. Instead of building a new theory, the note offers a deliberately minimal baseline. It starts from Marcel Cohen's 1931 division into Northern and Southern Ethiopic, both traced to a Proto-Ethiopic koine close to Ge'ez. It then adds a few well-supported later findings by Polotsky, Leslau, Goldenberg and Hetzron. These concern where Soddo, Gafat, Argobba and Gogot belong, judged by shared features and by how well speakers understand each other. The result is a single family tree. Its main structural claim is that "Gurage" is not one linguistic unit: its eastern varieties group with Harari, and its western varieties form a separate branch together with Gafat. The value of the paper is a modest, agreed starting point that later classifications can test and refine.

Key points

• The family's members: Ethiopian-Semitic is made up of Ge'ez along with Amharic, Tigrinya, Tigre, Harari, Argobba and Gafat, plus the cluster of Gurage varieties.
• Cohen's 1931 scheme:
  • Northern Ethiopic contains Tigre and Tigrinya.
  • Southern Ethiopic contains Amharic, Harari and Gurage.
  • Gurage splits into mutually unintelligible Eastern and Western clusters.
• Eastern Gurage: Wolane, Selti-Ulbarag and, tentatively, Zway. This group is closest to Harari.
• Western Gurage: two subgroups.
  • Chaha, Ezha, Ennemor (Inor), Gumar, Gyeto and Endegefi.
  • Muher, Gogot and Masqan.
• Revisions to Cohen:
  • Polotsky (1949) places Soddo (Aymallel) in Western Gurage; Cohen had tentatively called it "North-Eastern Gurage."
  • Leslau (1960) treats Argobba as a conservative dialect of Amharic and links Gafat to Soddo.
  • Goldenberg (1968) and Hetzron (1972) report that Gogot is a Soddo variety that Soddo speakers largely understand.

Interesting ideas

• I1. Instead of inventing a new full theory after a failed one, the author builds a minimal baseline from an older consensus plus a few well-attested corrections, so future work has a stable reference point.
• I2. A familiar umbrella label ("Gurage") is shown not to name a real unit, and its parts are reassigned to different branches.
• I3. Uncertain members (Zway, Endegefi, Soddo in Cohen) are placed "in the present state of our knowledge" or left unclassified, rather than forced into a slot.
• I4. Mutual intelligibility between speakers is used as direct evidence that one variety (Gogot) is a dialect of another (Soddo).
• I5. Argobba is treated as a conservative dialect of Amharic, separating archaic retention from true distinctness.

For Mova

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

• I1 (learn): when silver data is contested, start from the subset on which all witness parsers agree, and add corrections only when backed by independent evidence.
• I2 (world): ontology categories imported from WordNet or Wikidata should be allowed to split when their members behave differently, rather than being kept as one node for convenience.
• I3 (world): store tentative facts with an explicit confidence or provenance flag, so the reasoner can explain when an answer rests on a provisional link.
• I4 (tree): when converting between UD dialects or treebank conventions, measure how much one annotation scheme can be mapped losslessly onto another to decide whether it is a variant or a distinct scheme.
• I5 (tokens): archaic or dialectal English forms could be mapped to their modern lemma as variants, rather than becoming separate lexical entries in the morphology.