Diagnosing harmony and assimilation errors in a case-suffix drill
Eight-item word-formation drill from a Gagauz grammar workbook. Candidates must put each stem into the case or number shown in brackets: ev (locative), kitap (ablative), sokak (locative), göz (ablative), yol (dative), dil (plural), süt (ablative), okul (locative). One candidate's script reads: evda, kitapdan, sokakda, gözden, yola, diller, sütden, okulda. The teacher's key gives: evdä, kitaptan, sokakta, gözden, yola, diller, sütten, okulda. The candidate scores 4/8 and says the task is unfair because 'the same suffix is spelled two ways for no reason'. The class has already covered vowel harmony but has not formally revised consonant assimilation. The teacher has 15 minutes of the next lesson and wants to correct the underlying rule, not the eight forms one by one.
1. Are the candidate's four errors one problem or two? Name the rule behind each error and say what the pattern reveals about the candidate's working rule.
Read the worked answer
Two rules, four errors, distributed asymmetrically.
- evda should be evdä: vowel harmony. The stem has front vowels (e), so the locative must be -dä, not -da.
- sokakda should be sokakta: assimilation. Stem-final k is voiceless, so the suffix-initial d becomes t.
- kitapdan should be kitaptan: assimilation after voiceless p.
- sütden should be sütten: assimilation after voiceless t.
So one harmony error and three assimilation errors. The profile is diagnostic: the candidate has stored a default shape -da/-dan and applies it unless harmony forces a change. The correct answers gözden and okulda are not evidence of assimilation knowledge, because the default d happens to be right there; only kitapdan, sokakda and sütden expose the missing rule.
A 15-minute plan follows from this: one minute on the voiceless set (p, t, k, s, ş, ç, f, h), then rapid oral alternation on new stems, then the three follow-up items below. Do not re-teach harmony: it is already largely in place.
2. The candidate asks why gözden keeps d although the stem ends in a consonant. Answer the question and say whether the alternation is phonetic or orthographic.
Read the worked answer
The trigger is not 'ends in a consonant' but 'ends in a voiceless consonant'. Voiceless obstruents (p, t, k, s, ş, ç, f, h, and the affricate written ţ) force -ta/-tan/-te/-ten; voiced segments, including all vowels and voiced consonants such as z, l, m, n, r, y, ğ, b, c, d, g, j, v, leave the suffix as -da/-dan/-de/-den. Since göz ends in voiced z, the form is gözden, exactly parallel to kitaptan with voiceless p.
Phonetic or orthographic? The conditioning is phonological (regressive voicing assimilation), but the result is fixed in the written norm: Gagauz spelling writes the assimilated consonant, so kitaptan and gözden are both prescribed forms, not free variants. That matters pedagogically: the candidate cannot write 'kitapdan' and defend it as a pronunciation. It also explains the error type: learners who learn the suffix visually over-apply d, while learners who learn it only by ear may over-apply t.
3. Design a three-item follow-up that separates the two rules, and predict the likely wrong answer to each.
Read the worked answer
Pick stems so that each item varies only one feature.
- küçük (ablative) → küçükten. Front vowels and voiceless k. The harmony is already automatic for this candidate, so any error must be assimilation: predicted error 'küçükden'.
- dağ (ablative) → dağdan. Back vowel and voiced ğ. Harmony is also automatic, so any error must be over-application of assimilation: predicted error 'dağtan'.
- köprü (locative) → köprüde. Front rounded vowels, vowel-final stem. Neither harmony nor assimilation is in play, so this item is a control: an error here signals a problem outside both rules.
Why stems rather than the original eight: repeating drilled forms measures memory of the key. Fresh stems measure the rule. The pattern of errors across the three items tells the teacher which rule is unstable: errors on 1 only means assimilation is missing; errors on 2 only means the candidate has over-generalised t; errors on 3 suggest the harmony rule is not yet secure either.