📄 Internal Grammar in Amharic Place-Names

Summary

Research on the grammar of proper names usually looks at how a name behaves as a whole unit in a sentence: no plural, no article, and so on. Little looks instead at the grammar inside names that have more than one part. Her data are just over 50 two-part Ethiopian place-names, mostly Amharic. They were collected from government documents and maps, European travel and missionary accounts, and a Peace Corps booklet. She sorts them by structure: generic noun before or after the specific one, generic present or left out, and adjective, verb or preposition combined with a noun. Three findings stand out. Generic-before-specific names look older, because they keep archaic words and an unproductive construct-state suffix. Names have almost no bound morphology. Above all, the modern genitive/relative prefix yä- never appears inside them, so its presence could mark a phrase as a common noun rather than a name. English shows a similar split between "the University of Texas" and "Texas University." Little argues that place-names keep traces of older syntax and calls for cross-linguistic comparison of how place-names are built.

Key points

• The sample has over 50 two-constituent names, mostly referring to towns, districts and mountains, with a few lakes and rivers. Most are noun-noun, and fewer than half contain the generic noun.
• The construct-state suffix -ä (historically an accusative ending, no longer productive) is common in generic-first names. däbrä ("mountain") now survives only inside names, while tärara serves as the everyday noun.
• Names lack the plural -očč, the definite -u, and the genitive/relative yä-. A waterfall near t'is ïsat is described with yä-, which marks that phrase as a common noun.
• The same area is called durä Jala on one side of the mountain and Jala durä on the other. The second form reorders the words to the modern pattern but keeps the old marker, which betrays syntactic change.
• Several names contain Teferi, reflecting Haile Selassie's popularity. Some places have both an Amharic and an older Cushitic name, such as Debre Zeyt and Bishoftu.

Interesting ideas

• I1. The absence of a productive genitive or definite marker inside a multiword expression is a cue that it is a name, not a descriptive phrase.
• I2. Some words survive only inside fixed names, as däbrä does, so their appearance signals a frozen proper name.
• I3. Many names leave out their true head (town, mountain), so the kind of entity must come from context rather than from the name's own words.
• I4. A word order that clashes with the morphology it carries is evidence that the form was reanalysed after a grammar change.
• I5. One place can have several names, from different languages or with different word orders, depending on who is speaking.

For Mova

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

• I1 (tokens): The tagger could use "of"/article patterns to separate PROPN sequences like "Texas University" from the common-noun phrase "the university of Texas."
• I2 (tokens): The lexicon could flag such name-only forms so that lemmatization does not treat them as ordinary content words.
• I3 (world): When creating entities from place-names, the world builder should infer a suppressed type such as town or river instead of reading it off the last noun.
• I4 (tree): When gparse sees names whose internal order conflicts with their markers, it could treat them as flat units instead of forcing a compound-head analysis.
• I5 (world): The knowledge graph should store alias sets, such as Debre Zeyt/Bishoftu, so that coreference links variant names to a single entity.