As in classical WP grammars (and WP models), the grammatical word remains the primary locus of meaning. Yet realizational models are also theoretical hybrids, reflecting the influence of the Bloomfieldian tradition in their treatment of roots or stems as basic units of lexical form. This conception departs from classical approaches, in which the word is the basic unit of lexical form, and exponents, or inflectional exponents at any rate, are regarded as abstractions over whole words. Although the implicational approaches in Chapter 7 provide the most faithful models of the classical WP conception, realizational models can also be interpreted in a more ‘abstractive’ way (Blevins 2006b).
One way of reinterpreting realizational models is to construe them as automata rather than as grammars. Formal language theory defines a correspondence between classes of rewrite grammars, which ‘output’ symbols or sequences of symbols, and classes of automata, which ‘input’ symbols or sequences. Like most morphological models, the realizational formalisms summarized in this chapter are conventionally represented as grammars. On this interpretation, a realizationbased model assembles forms by applying operations to lexical roots or stems that ‘spell out’ specified features, as discussed above. The same class of approaches can also be construed as automata. On this interpretation, a realization-based model accepts (or rejects) forms, depending on whether they satisfy the feature- form associations expressed by spell-out rules. As with formal grammars and automata, the language (the set of surface feature-form pairs) ‘generated’ by a real- izational grammar will be the same as the language ‘accepted’ by the corresponding automaton.
Despite their extensional equivalence, the two interpretations differ in ways that bear on a range of formal and empirical issues. The most salient difference concerns the composition of the lexicon. Whereas realizational grammars assume a model of the lexicon consisting mainly of root or stem entries, realizational automata are compatible with a classical WP view on which the lexicon consists primarily of full word entries. On this interpretation, realization rules function as ‘word admissibility conditions’ that constrain the compatibility of particular features and form variations in the feature-form pairing associated with a word entry. Hence interpreting realization-based models as automata suggests a rapprochement with the implicational models in Chapter 7, in which words are basic morphotactic as well as morphosyntactic units. In addition, applying realization rules a to word-based lexicon permits a form-based treatment of referral rules, which can be interpreted as directly relating all or part of the form of multiple paradigm cells.
An automaton-based interpretation of realizational models is also compatible with the model-theoretic perspective of constraint-based approaches to syntax. The feature specifications that determine the applicability of a realization rule can be interpreted formally as a partial description of the feature bundles associated with word entries. The operation performed by a rule is then interpreted as a partial description of a class of forms. An operation of the form /X ^ Xrn/ is not interpreted as ‘constructing’ an output form /Xrn/ by concatenating the string rn to a base X, but as ‘admitting’ a form that is analyzable as a sequence Xrn. A word form is admissible if there is an assignment of forms to variables (like ‘X’) on which it satisfies all of the applicable realization rules. This shift in orientation offers a slightly different perspective on morphophonemic processes that often remain implicit in realization-based models. To express the standard claim that the forms of the regular past participle ending in English [t], [d] and [id] “are variants of one formative, not separate formatives”, Matthews (1991: i45f.) posits “a single morphological process” that introduces a “basic form” or “basic allomorph”. In the models of Anderson (1992) or Aronoff (1994), the ‘basic form’ is represented by the corresponding phoneme, here /d/. Yet the morphological analyses defined by these models are either abstract or incomplete, as they fail to specify the ‘back-end’ morphophonemic processes that define surface forms. On a word admissibility interpretation, the phonemes in realization rules can be taken to be descriptions that are satisfied by surface forms in a word-based lexicon. Thistreatment of morphophonemicssubsumesthe typesofmorphological metageneralizations discussed by Stump (2001: §6.2) within the kind of model- theoretic approach to phonology set out in Bach and Wheeler (1983).
To those accustomed to thinking of formal grammars as admissibility conditions, the main effect of adopting an automaton-based interpretation might be that the objects admitted by these conditions are the words of a word-based lexicon, though this shift in perspective has other potential consequences. Applying admissibility conditions to a word-based lexicon should reduce the role of ‘indexical’ features, since it is no longer necessary to reassemble words from parts that have been dispersed across stem and rule inventories. It is of course to be expected that some problems that arise on a grammar interpretation of a realizational model will recur in a different form on an automaton interpretation. In particular, any challenges that derive from the realizational focus on individual forms rather than systems will have counterparts on grammar and automaton interpretations. To clarify how these kinds of challenges can be met by approaches that incorporate a complex system perspective, the following chapter now turns to the family of implicational WP models.