📄 Bi-Consonantal Reduplication in Amharic and Ethio-Semitic

Summary

This is a one-page abstract of Peter Unseth's PhD dissertation (University of Texas at Arlington, 2002), printed in Aethiopica 6. The printed issue is dated 2003, not 2013. It studies an Amharic word-formation pattern that copies the final two consonants of a root, which the author names Bi-Consonantal Reduplication (BCR). Earlier accounts used inconsistent labels, and their definitions either let in other reduplication types or missed real cases. The dissertation replaces them with a criterion meant to capture exactly the genuine instances. Under this sharper definition, several word classes once credited to an unrelated reduplication process turn out to be BCR outputs. Root type determines how a BCR form behaves. Some forms take verbal inflection. Others are limited to functioning as nouns, as adjectives, or as non-inflecting stems inside compound verbs. Some root classes cannot undergo BCR at all. A semantic survey finds recurring meanings, notably impaired walking and showy dressing. The work also argues that BCR is very old in Semitic, points to it in languages where it had gone unnoticed, and suggests why other languages lost it.

Key points

• New term: Bi-Consonantal Reduplication (BCR), meaning repetition of a root's last two consonants, described as never before studied systematically in any Semitic language.
• Some form classes previously thought to come from a separate reduplication type are reassigned to BCR.
• Inflectability depends on root type: some BCR derivatives work as verbs, while others appear only as nouns, adjectives, or uninflected stems that need a host verb to form a compound verb.
• Identified root classes that block BCR. Frequent semantic fields are gait impairment and dressing up fancily.
• Diachronic claims: BCR goes back to early Semitic, it is found in languages where it was overlooked, and a hypothesis explains its loss elsewhere. The abstract reports no corpus size or counts.

Interesting ideas

• I1. Define a phenomenon by a criterion checked against both over-inclusion and under-inclusion, rather than inheriting fuzzy labels from earlier work.
• I2. Forms assumed to come from different processes can turn out to share one derivation, which shrinks the rule inventory.
• I3. A derived word may carry verbal meaning yet lack verbal inflection, so it relies on a host verb in a compound.
• I4. Documenting which inputs a process can never apply to is as informative as listing where it applies.
• I5. A formal pattern can cluster around a few semantic fields, so form gives a hint about meaning.

For Mova

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

• I1 (tokens): when writing morphology rules in crate en, test each rule against the forms it wrongly catches as well as the forms it misses, and keep both lists as regression fixtures.
• I2 (learn): when self-induced repair rules multiply, periodically check whether two rule families reduce to one shared pattern, so rules stay fewer and easier to explain.
• I3 (frames): treat English light-verb constructions in the same way, letting the frame and its roles come from the nominal base while the host verb supplies only tense and agreement.
• I4 (learn): build negative-control sets from inputs that must never trigger a given rule, and use them to catch overgeneralizing silver data.
• I5 (world): let a recognized derivational pattern serve as a weak, overridable prior when assigning ontology classes to unknown words, and keep that prior visible in the explanation trace.