Abstract analysis and historical change: Tera

One such argument for the psychological reality of abstract analysis comes from Tera. Newman 1968 provides a synchronic and diachronic argument for abstract phonology, where similar surface forms have different underlying forms.

The synchronic argument. Data in (65) illustrate a basic alternation. Some nouns ending in [i] in their citation forms lack that vowel in phrase medial contexts:

Not all words ending in [i] prepausally engage in this alternation, as the data in (66) demonstrate:

Given a vowel ~ Ø alternation plus a set of stems which are invariantly i-final in (66), we might be led to surmise that the stems in (65) are C-final, and take an epenthetic vowel [i] phrase-finally. This can be ruled out given (67), where the stem ends in a consonant both phrase-medially and phrase-finally.

A completely surface-oriented account where the underlying form must be one of the surface variants is untenable: the nouns in (65) have a variant with the vowel [i], but selecting /i/ for the underlying form fails to distinguish (65) from (66) which always have [i]; and the nouns of (65) also have a variant with no final vowel, but the nouns in (67) always lack a final vowel.

Other roots of the variable-final type give evidence that the problematic stems in (65) underlyingly end in schwa. The data in (68) provide monosyllabic words which have the shape Ci prepausally and Cə phrase medially.

These words contrast with ones that have invariant [i] in both contexts.

For the stems in (68), an obvious nonabstract solution is available: the stems end with /ə/, and there is a rule turning schwa into [i] prepausally:

This applies in dala wa ɗi ‘Dala went’ from dala wa ɗə, but final schwa is unaffected in dala wa ɗə goma ‘Dala went to the market.’ The stems in (69) do not alternate since they end in the vowel /i/. This solution is nonabstract since the underlying form, /ɗə/, is one of the observed surface variants.

There are other stems with final [i] prepausally and [ə] phrase medially.

These stems either have the shape [CVCCə] phrase-medially, or else [CVZə] where Z is a voiced consonant.

This gives the following groups of stems with an underlying final schwa:

For most of these stems, postulating underlying schwa is quite concrete, since schwa actually surfaces in phrase-medial context. However, in polysyllabic stems such as deɓi ~ deɓ with a single voiceless consonant before final schwa, the analysis is abstract because schwa is never phonetically manifested in the morpheme. The decision that the vowel in question is schwa is based on analogy with a known behavior of schwa: it becomes [i] prepausally.

Our analysis requires a rule that deletes word-final phrase-medial schwa providing the stem is polysyllabic and ends only in a single voiceless consonant.

More evidence supports abstract schwa in certain words. The examples in (74a) show that when a vowel -a marking definite nouns is suffixed to a stem such as /pərsə/ which ends in schwa, schwa deletes, whereas underlying /i/ is not deleted. The data in (74b) show the same thing with the imperative suffix /u/:

This motivates a rule of prevocalic schwa deletion, which provides another diagnostic that differentiates schwa from /i/.

Although ‘throw’ only has the surface variants [mbuki] ~ [mbuk], it behaves exactly like stems such as /kədə/ where schwa is phonetically realized, and acts unlike /vi/, in losing its final vowel before another vowel. Finally, there is an allomorphic variation in the form of the adjective suffix -kandi, which shows up as -kandi when the stem ends in a vowel (saɓir taɗa-kandi ‘heavy stick’) and as -ndi when the stem ends in a consonant (saɓir teɓer-ndi ‘straight stick’). The stem of the word for ‘long’ ends in abstract schwa, since it alternates between final [i] (saɓira kəri ‘the stick is long’) and medial Ø (saɓira kər ɓa ‘the stick is not long’). Furthermore, the stem selects the postvocalic variant of the adjective suffix (saɓir kər-kandi ‘long stick’), even though on the surface the stem ends with a consonant and not a vowel. This anomaly is explained by the hypothesis that the stem does in fact end in a vowel, namely schwa. Thus multiple lines of argument establish the presence of an abstract vowel schwa in a number of words in the synchronic grammar of Tera.

The diachronic argument. A recent sound change in Tera provides a grammar-external test of the abstract hypothesis. In one dialect of Tera, spoken in the town of Zambuk, a rule was added which palatalized t, d and ɗ to t ʃ , dʒ and d’ ʒ before i. The dialect of Tera, spoken in Wuyo, is representative of the rest of Tera, in retaining the original alveolars. Thus we find Wuyo da, Zambuk da ‘one’ with no palatalization, but Wuyo di, Zambuk dʒ i ‘to get up’ where d palatalizes. There are synchronic alternations which further motivate this palatalization process in the contemporary grammar of the Zambuk dialect, so where the Wuyo dialect has xat-a ‘my brother,’ xat-in ‘his brother,’ the Zambuk dialect has xat-a, xatʃ -in. In Wuyo one finds wuɗi ‘milk’ and in Zambuk one finds wud’ ʒ i, deriving from /wuɗi/ – that the final vowel is /i/ and not /ə/ is shown by the phrase medial form wuɗi.

While palatalization is active in the Zambuk dialect, it does not affect all surface sequences of alveolar plus [i], in particular it does not affect [i] which derives from schwa. In the Wuyo dialect ‘to pull’ is kədi before pause, kədə medially (cf. dala wa kədə koro ‘Dala pulled a donkey’), and therefore we know that the stem is /kədə/. In the Zambuk dialect, the medial form is also kədə, showing that the stem ends in schwa in that dialect, and the prepausal form is kədi. Thus palatalization does not apply to the output of final schwa-fronting: the failure of palatalization to apply to this derived [di] sequence provides another diagnostic of the distinction between /i/ and [i] derived from /ə/.

Further confirming our hypothesis about abstract schwa, the stem /wuɗə/ ‘to point’ which appears in the Wuyo dialect as wuɗi prepausally and as wuɗ medially (dala wa wuɗ koro ‘Dala pointed at a donkey’) appears as wuɗi in the Zambuk dialect, without palatalization, as is regularly the case with the vowel [i] derived from /ə/. The fact that the innovative sound change of palatalization found in the Zambuk dialect is sensitive to the sometimes abstract distinction between underlying /i/ versus ones derived from schwas, especially when the schwa never surfaces, supports the claim that abstract underlying forms can be psychologically real.