📄 Shimelis Mazengia, Nominalization via Verbal Derivation: Amharic, Tigrinya and Oromo

Summary

This is a review by Magdalena Krzyżanowska of Shimelis Mazengia's 2015 doctoral monograph. The book compares how verbs become nouns in three Ethiopian languages: Amharic, Tigrinya as spoken in Tigray, and Harärge Oromo. It works at three levels: words (derivation and compounding), phrases (action nominal constructions) and clauses (object complements and headless relatives). The framework is mainly Dixon's Basic Linguistic Theory. Comparing three languages at once is rare in Ethiopian studies. The main finding is that Amharic and Tigrinya resemble each other far more than either resembles Oromo in nominalization. This weakens the common view that Amharic is heavily Cushitic-influenced and sits between Semitic Tigrinya and Cushitic Oromo.

The reviewer praises the clear organization and the rich data. She also corrects several claims with her own Amharic examples. Periphrastic reciprocals only reinforce reciprocal verbs and cannot replace them. Epistemic verbs always take ɨndǝ. Sentence stress does interact with the choice of nominalization type. Finite subject clauses are possible with some predicates. The book matters as a typological reference for derivational morphology in under-described languages.

Key points

• The book appeared in Abhandlungen für die Kunde des Morgenlandes 99 (Harrassowitz, 2015, 284 pp.) and is the author's PhD thesis. The author is a native speaker of Amharic and Oromo.
• Its structure:
  • lexical nominalization in Chapters 3–5;
  • phrasal nominalization (the action nominal construction, ANC) in Chapter 6;
  • clausal nominalization in Chapter 7;
  • each chapter closes with a contrastive summary.
• Argument nominals fall into four principal types: Agent, Instrument, Result and Manner. Nine secondary types (Patient, Causer, Experiencer, Goal and others) are built on the Agent pattern.
• Compounds are mostly endocentric. Amharic compounds are right-headed, while Tigrinya and Oromo tend toward left-headed ones. The reviewer notes that only about 6 of the compounding chapter's 21 pages address the core topic.
• Of Noonan's fourteen complement-taking predicate classes, three are studied: utterance, desiderative and epistemic.
  • With desideratives, the Amharic complementizer depends on subject coreference: lɨ for the same subject, ɨndǝ for a different one.
  • The reviewer adds that Amharic and Tigrinya also allow a bare or lǝ-marked verbal noun when the subjects are the same.

Interesting ideas

• I1. Instead of a binary noun/verb label, measure how noun-like or verb-like a nominal is by checking which verbal markers (tense, aspect) and nominal markers (definiteness) it accepts. A graded profile describes mixed categories better.
• I2. Deverbal nouns can be grouped by which argument of the source verb they denote (agent, instrument, result, manner). Rarer roles then reuse the agent pattern, which gives a compact map from derived noun to verb frame.
• I3. The choice of complementizer signals whether the embedded subject is the same as the matrix subject. Grammar thus hands coreference information to the reader.
• I4. The reviewer overturns a generalization by swapping in one different predicate ("it raises doubts" instead of "it is surprising"). This shows that conclusions drawn from a single test item are fragile.
• I5. A finite "that"-clause and a nominalized complement can share truth conditions yet differ in stress and theme–rheme structure. The choice of form therefore carries information structure.

For Mova

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

• I1 (tokens): English gerunds and -ing forms could carry a feature profile (takes a determiner? takes an object? takes adverbs?) rather than a forced NOUN/VERB tag, which would make the tagger's decision explainable.
• I2 (world): words like "teacher", "cutter" or "building" could be linked to the verb's frame with an explicit role slot, so "the baker" introduces an entity filling the Agent of "bake".
• I3 (frames): English "want to sell" versus "want him to sell" could be treated the same way, resolving the empty embedded subject to the matrix subject or the object when filling who-does-what.
• I4 (learn): when inducing repair rules or grammar claims from silver data, test each rule across several lexical items, and keep counterexample controls before accepting it.
• I5 (back): when generating from the meaning graph, choose between "that you prepared well" and "your being well prepared" according to which part is meant to be new or focal.