📄 Reliance on Predicative Units as a Method of Analysing and Translating Amharic Written Texts

Summary

Fridman tackles a recurring classroom problem: advanced learners find it hard to turn the strictly left-branching, verb-final syntax of written Amharic into the right-branching structure of European languages. A common approach is to parse "from the end", starting at the sentence-final verb. In the author's teaching experience this fails, because students lose track of the earlier material by the time they reach the start. He proposes a four-step procedure instead.

1. Find every verb form that can act as a predicate.
2. Cut the sentence into clauses, each closed on the right by its predicate.
3. Gloss each clause roughly, left to right, using only the verb's bare lexical sense.
4. Join the glosses using knowledge of morphosyntax.

The paper first sets out a working, function-based notion of finiteness and a typology of non-finite forms. It then lists constructions that make the method harder to apply and demonstrates it on a long biographical sentence. That sentence is first rendered as a chain of clause glosses and then as several idiomatic English sentences. The result is a practical, teachable way to analyse long Amharic periods.

Key points

• Finite forms are defined by function: a form is finite if it can head a syntactically independent clause. Non-finite forms head only dependent clauses.
• Amharic tends strongly towards deranking. Combined with SOV order, which always puts the main predicate last, this makes finite and non-finite predicates easy to tell apart.
• Non-finite forms fall into three groups:
  • the converb, i.e. the suffixal converb or "gerund" (nägro);
  • conjunctional forms, including relative forms;
  • infinitival forms.
• When adjectives (noun-like or relative forms of stative verbs) are used predicatively, a copula is obligatory. The author argues that copular predicates are always overtly present in Amharic, unlike in Russian or Arabic.
• The paper catalogues at least nine difficult cases for the method. These include:
  • coordinated predicates;
  • cleft sentences;
  • relative predicates with a head noun, sometimes masked by a preposition;
  • analytic verbal constructions, such as ǝyyä + perfect + hedä for gradual development;
  • a bare infinitive or an infinitive with lä-;
  • a finite verb followed by a verb of speaking.

Interesting ideas

• I1. Split the sentence into predicate-headed chunks before working out how they relate, since in a head-final language each chunk ends at its predicate. — For Mova (tree): An English clause chunker anchored on predicate verbs could run before full attachment, both to narrow the parser's search and as an explainable check on clause boundaries.
• I2. In a first pass, gloss each clause with the verb stripped to its bare lexical meaning, and add tense, mood and subordination only in a second "binding" pass. — For Mova (frames): Frames could first be built per clause from verb lemmas and arguments, with tense, aspect and inter-clause links (cause, condition, time) attached afterwards, so each layer stays separately inspectable.
• I3. Before applying a general procedure, catalogue the specific constructions that break it, so they can be handled deliberately rather than by accident. — For Mova (learn): A curated suite of hard constructions (clefts, coordinated predicates, analytic verb chains, verbs of speaking) could serve as a targeted evaluation and negative-control set for parser and frame builder.
• I4. Treat stative or adjectival predication as a full predicate with its own copula, even when another language leaves it unexpressed. — For Mova (frames): Copular and small-clause predications (e.g. "found him tired") could get explicit state frames, so the world model records states as reliably as actions.
• I5. When translating, break a long embedded period into several short sentences, add explicit connectors such as "however" to make the links clear, and then prune connectors that have become redundant. — For Mova (back): Generation from the meaning graph could split complex event chains into short sentences, emit discourse connectors for each relation, and then remove unneeded ones in a stylistic pass.