📄 David Appleyard: Colloquial Amharic: A Complete Language Course

Summary

This is not a research paper. It is a short review by Verena Böll and Getie Gelaye of David Appleyard's textbook Colloquial Amharic (Routledge, 1995). The PDF header gives Aethiopica 1 (1998), which differs from the 2013 date in the library metadata. The reviewers assess how well the book teaches modern spoken Amharic to English speakers. The course has fourteen lessons, a key to exercises, an English–Amharic glossary, and two cassettes.

The reviewers praise several things:

• Script, transliteration and translation are presented side by side.
• Gemination is marked only in the transliteration, so the Ethiopic script stays unaltered.
• The content reflects present-day Ethiopia.
• Cultural notes explain greetings and holidays.

They also list weaknesses:

• Some dialogue phrasings are calqued from English instead of idiomatic.
• There are small errors in greetings and in the answer key.
• Grammar is split into fragments with poor cross-referencing.
• Exercises lack model sentences.
• Some recordings are too fast.

Overall they judge it a good, more up-to-date entry point than Leslau or Richter, best used alongside a reference grammar. Its relevance to NLP is indirect: it shows how a learner resource for a low-resource Semitic language is designed and where it fails.

Key points

• The book has vi + 374 pp.: an introduction (pp. 1–15), 14 lessons (pp. 16–295), a key to exercises (pp. 296–323), an English–Amharic glossary (pp. 324–371), a two-page grammatical index and a one-page Amharic index.
• Two ~60-minute cassettes were recorded by Appleyard with four native speakers (two women, two men), supervised by Yalew Kebede of SOAS.
• Transliteration covers seven vowels, and ejectives use IPA-style apostrophes (k', t', p', s'). Gemination appears only in the Latin transliteration, unlike Leslau's 1967 superscript dots.
• Verbs follow the anglophone classification: triliteral vs. biliteral, with biliterals split into Type A and Type B.
• Concrete faults are noted:
  • unidiomatic renderings of "long distance call", "Who is speaking?" and "I'm in a hurry";
  • a wrong evening greeting;
  • an overly colloquial plural form;
  • a key error (exercise 3, no. 2, p. 24: should be 1st person plural, not singular).

Interesting ideas

• I1. The original script is left unmodified, and pronunciation details such as consonant doubling go in a separate parallel transliteration layer.
• I2. A sentence can be grammatical yet unnatural, because it calques English instead of using the phrasing natives actually say.
• I3. Which greeting is correct depends on situational context, such as time of day, not only on form.
• I4. An answer key is valuable for self-checking, yet the reviewers found errors in the key itself.
• I5. The reviewers fault grammar scattered across lessons whose index points only to the first mention.

For Mova

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

• I1 (tokens): Keep the raw surface string immutable and store normalizations, phonetic or orthographic hints as separate annotation layers. Every token can then be traced back to the exact input.
• I2 (back): Generation and translation checks could compare output against attested idiomatic phrasings for a given situation, not just grammaticality, and flag calques.
• I3 (frames): Greeting and other speech-act frames could carry a situational slot (time, setting), so the choice and interpretation of a formula is explained by context.
• I4 (learn): Treat gold answers and teacher labels as fallible. Periodically audit them with independent witnesses rather than trusting them as ground truth.
• I5 (answer): When explaining an answer, link each grammatical rule or fact to every place it is used, so the explanation trail is complete, not just the first occurrence.