📄 Some Observations on the Construct Marker -a in Classical Ethiopic

Summary

In Classical Ethiopic (Gǝʿǝz), the ending -a marks both the accusative and the head noun of a construct (genitive) phrase, whatever role that head plays in the clause. The traditional account says the accusative ending spread by analogy into the construct state. Tropper (2000) challenged this. He argued that -a is a relic of an early Semitic ergative–absolutive system, so the construct never had a -u form.

Siegismund tests Tropper's arguments against the cited sources, the comparative Semitic evidence and Ethiopic-internal data. He finds misreadings of Hartmann, Gragg and Lipiński. He draws on Waltisberg's and Hasselbach's rejections of the ergative hypothesis. He also points to evidence that Proto-Semitic construct nouns carried full case marking.

His conclusion supports the traditional view. Once nominative -u and genitive -i weakened to ǝ and disappeared, -a was the only case vowel left. It was then recruited as a general construct marker, plausibly for euphonic reasons, and later spread to plurals. The paper matters because it rests a long-range typological claim on simpler, attested sound changes.

Key points

• Ethiopic -a continues the Proto-Semitic accusative *-a. Nominative *-u and genitive *-i were reduced to ǝ and then lost, which left a contrast of -a versus zero. In the construct state this contrast is neutralized.
• Calling the classical construct "the same principle" as in Amharic is misleading. Amharic puts the possessor first with a prefixed yä-, and the classical pattern survives in modern languages only as loans, religious terms and compounds.
• Waltisberg (2002) argues Proto-Semitic was accusative in structure. Hasselbach (2013) proposes an archaic marked-nominative stage instead of an ergative one. Footnote evidence from Danish (and partly English) oblique pronouns shows that "unexpected" accusative uses can arise without any ergative past.
• Ethiopic-internal evidence: nouns before pronominal suffixes still distinguish ǝ (non-accusative) from a (accusative). Words like "father" keep -u before suffixes, which looks like a relic of case-marked construct nouns.
• In the proposed derivation, when a head X precedes a genitive Y-i, a head ending in -u or -i loses its vowel along with Y's ending (giving XØ YØ), whereas a head in -a keeps its vowel and only Y's ending drops (giving Xa YØ). Since only -a survived, it became the construct marker, and internal plurals helped it spread to the plural.

Interesting ideas

• I1. One surface ending can serve two unrelated grammatical jobs, so a marker alone does not reveal a noun's clause role; the structure around it does.
• I2. Laying out every input configuration with its outcome (X-u/a/i before Y-i) makes it visible that only one marker could survive, which turns a vague claim into a checkable table.
• I3. The author re-reads the sources an opponent cites and finds that one page actually says the opposite of the claim built on it.
• I4. Fixed environments, such as positions before suffixes or religious vocabulary, keep older patterns that the productive grammar has lost, so they should be analysed as lexicalized rather than as live rules.
• I5. The paper separates describing a parallel (two constructions paraphrased the same way) from explaining why the forms coincide.

For Mova

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

• I1 (tree): Treat English look-alikes (possessive 's vs. contraction, object-case pronouns used as subjects in "me and him went") as structurally resolved, never mapped straight from form to relation.
• I2 (learn): Self-induced repair rules could be justified by an exhaustive table of input cases and outcomes, so a reviewer can see exactly which configurations a rule covers.
• I3 (learn): When an LLM teacher tier backs a label with a cited rule or example, check that citation against the original before accepting the label as silver data.
• I4 (tokens): Flag frozen expressions and archaic compounds as lexical units, so the morphology and tagger do not generalize from them to productive English.
• I5 (answer): When readers justify an answer, label whether a step is an observed correlation or a causal account, which keeps explanations honest.