Allomorphy in Optimality Theory: Polish Iotation

This article discusses iotation, a process that has been analyzed in generative phonology as a palatalization rule. We argue that optimality theory predicts the treatment of this process in terms of allomorphy, which in fact is desirable for a synchronic analysis. The consequence is that, with regard to iotation effects, the task of phonology is to account for the distribution of allomorphs rather than to derive them from a single underlying representation. While, as a result of diachronic changes, the allomorphs are arbitrary, their distribution is not. It follows from the interaction of universal phonological and morphological constraints, and from the considerations of segment markedness.*

All phonological theories address the problem of the surface realization of morphemes. This is often straightforward because many morphemes have one invariable realization; for example, Polish dom ‘house’ is always represented as [dom]. Central to phonology are instances that are realized differently in different contexts. Traditionally, it is said that such instances show allomorphy because allomorphy is understood as a situation in which two or more different morphs share the same grammatical or semantic function and hence are allomorphs of one morpheme. The task of phonology is then to explain the distribution and the phonological shape of allomorphs. The most straightforward way to achieve this goal is to posit a single underlying representation and account for surface allomorphs in terms of phonological generalizations.

A clear example of this situation is an analysis of the surface allomorphs [hut] and [hud] of hoed ‘hat’ in Dutch. The former occurs in the singular while the latter is found in the plural, before the plural suffix en: [hut] versus [hud + ən]. The solution is to posit hud/ as the underlying representation and derive [hut] phonologically by final devoicing (Booij 1995). The success of this analysis is due to two facts: first, the allomorphs differ from each other in a minimal way (devoicing) and second, the distribution of the allomorphs is governed by a clear phonological generalization (final devoicing). However, there is also another type of allomorphy, one that is dramatically different from the case just described.

One example is the Dutch suffixes -iteit (English -ity) and -heid (English -hood). The former is a non-native suffix, and it attaches to non-native stems. The latter is a native suffix, and it attaches to both native and non-native stems. The situation here is different from that described in connection with [hut] and [hud]. First, -iteit and -heid do not show any phonological similarity and, consequently, are not reducible to a single underlying representation. Second, the distribution of the suffixes is governed by a purely morphological generalization: the lexical subcategorization of the stems and the suffixes with regard to their native versus non-native status. The distribution of -iteit and -heid falls within the realm of morphology rather than phonology.

To summarize, allomorphy is found at both of the two extreme points along the dimension of the phonology-morphology interface: some allomorphy is purely phonological, some is purely morphological. In the former case, the typical analysis is to posit a single underlying representation and appeal to phonological generalizations in order to account for the surface differences between allomorphs. In the latter case, we have two or more underlying representations and the distribution of the allomorphs is governed morphologically. The difficulty for all analyses is that languages also exhibit allomorphy that cannot be placed at either of the two extreme points: phonological or morphological. In other words, there is a continuum between these two points, with allomorphy being phonological or morphological to a varying degree. We illustrate this problem below.

Carstairs-McCarthy (1988) notes that the distribution of the Hungarian suffixes for the second person singular indefinite present indicative -ol and -(a)sz is governed phonologically. The former occurs after sibilants and the latter after affricates. Yet, there is no reasonable analysis that would reduce these suffixes to a single underlying representation. Aronoff (1976) observes that the English nominee is derived from the verbal base nominate. The problem is that the ate of nominate does not occur in nominee. The deletion of ate cannot be ascribed...