Integration of phonology and morphology

One vital choice for LP is whether or not the model should be interactionist, with morphological and phonological operations interspersed. This interaction was one of the major motivations for the development of LP, and remains for many phonologists an attractive feature of the model; but it is not without its problems.

Some of these difficulties seem relatively minor. For instance, certain affixes appear to display properties of both Class I and Class II; thus,-ism is stress-shifting in Cathólicism from Cátholic, but stress-neutral in Prótestantism from Prótestant. Other morphological concerns are less tractable; thus, the existence of so-called bracketing paradoxes (like the famous ungrammaticality; see Badecker 1991) has led to Aronoff and Sridhar's (1983) contention that the Affix Ordering Generalization is invalid, and morphological level-ordering untenable. Further critiques of the same sort are included in Sproat (1985) and Szpyra (1989). Halle and Vergnaud (1987), for example, consequently adopt a non-interactionist model, with a separate morphological module which precedes all phonology, and contact between the two components limited to the fact that `morphology ... creates the objects on which the rules of phonology operate' (1987: 78).

We can respond to these developments in two ways. First, we might agree with Badecker (1991: 131) that `there is substantial content to the role of morphology in Lexical Phonology even when Level Ordering is subtracted out'; and indeed, Halle and Vergnaud (1987) still find it necessary to account for the behavior of stress-neutral versus stress sensitive suffixes, for instance. On the other hand, we might wish to maintain an integrationist approach, with level ordering retained and respected for both morphology and phonology; this stronger version of LP is more in keeping with the origins of the model, and is the approach I adopt here. Hargus (1993), in a defence of interactionism, demonstrates that phonology must precede morphology in some cases, since morph ology may necessarily refer to a derived phonological property, often stress. Furthermore, the domain of phonological rules may exclude material reflecting a morphological process: thus, spirantization in Luiseño fails to apply to reduplicative structures, while nasal harmony in Sundanese must precede and follow plural infixation. Hargus argues that, although some cases previously seen as supporting interaction have been reanalyzed, not all can be. Giegerich (in press) also argues strongly for interaction, albeit in a model of base-driven stratification rather different from standard LP. Giegerich highlights failures of the Affix Ordering Generalization, and the large number of affixes with at least potentially dual membership of Levels 1 and 2, but claims that these are only problematic when the stratal distinction is driven by affix behavior.

If we assume instead that properties of the base are predominantly at issue, with Level 1 being the domain of roots and Level 2 of words, we can derive stratification while allowing dual membership as the norm for derivational suffixes in English, for instance. There are many consequences of this change in perspective. Others, for instance Giegerich's argument that morphology on Level 1 will effectively involve listing, with each root being stored along with the list of Level 1 affixes it can potentially attract (from which follows the unproductive and semantically idiosyncratic nature of Level 1 morphology), cannot be fully developed here. Nonetheless, morphological developments of this kind, as well as the arguments given earlier, may justify retaining an integrated model.