Realization and structure
Viewed as a class, realizational approaches represent a more faithful model of the morphological descriptions assigned by classical WP grammars than of the morphological systems described by these grammars. This interpretation of real- izational models is reinforced particularly by the way that they reify traditional descriptive vocabulary and associate morphotactically unstructured analyses with complex forms. These points are clarified briefly below.
The nomenclature of classes and series in a classical grammar serves a primarily descriptive function, providing a frame of reference for the classification of forms and systems. These classes and series are not located in languages per se, but merely define the descriptive lens through which the language is viewed. It is these aspects of the classical WP tradition that realizational approaches model when they reify schemes of classification by means of indexical features. Inflection class features appear initially in the treatment of German declensions in Chomsky (1965:171) and in the analysis of Latin conjungations in Matthews (1965:150). Once established, these features came to be assumed in nearly all realizational approaches. A strategy of reencoding classifications by means of indices was subsequently extended to encompass other aspects of a traditional description. The analysis of Georgian case marking proposed in Anderson (1992: 150) expresses conjugational series as ‘series indices’ of the form [±Series II]. The covert ‘lexeme indices’ discussed in connection with PFM in Section 6.4.3 are also assumed in many realizational models. At the sub-word level, ‘stem indices’ are widely adopted in analyses of stem syncretisms (Aronoff 1994; Stump 2001; Baerman et al. 2005).
In many of these cases, indexical features drive a realizational analysis by determining the choice of an element from a set of alternatives, triggering the application of a rule, or in some other way explicitly guiding choices that are underdetermined by non-indexical features. In languages where inflection classes are not predictable from grammatical or phonological properties, encoding class membership as a separate morphological ‘feature’ permits the selection of class- specific exponents. A ‘stem indexing’ analysis allows rules to select recurrent units of form that do not realize a consistent set of grammatical properties in the different constructions in which they occur. More generally, by cross-referencing cells, entries, rules and other elements, it is possible to select, arrange and manipulate units whose form and distribution is not otherwise predictable from a decomposed representation.
In effect, indexical features serve as the ‘glue’ of a realizational model, providing a general mechanism for reassembling the words and larger units from the parts distributed across the stem and rule inventories of the model. This use of indexical features fundamentally changes the nature of a realizational model. It is unclear in what meaningful sense exponence rules remain ‘interpretive’ if they can ‘spell out’ indexical features that function as assembly instructions. Reassembly is typically the only observable function of these features, given that they rarely if ever seem to participate in grammatical dependencies such as agreement, concord or government.
Hence an approach that ‘spells out’ indexical features can, as suggested above, be interpreted as providing an explicit model of a system of language description. However, as a model of language structure, it runs the risk of incorporating the error, which Hockett (1967) states below, of confounding its own descriptive apparatus with the intended object of description:
One of the most dangerous traps in any of the more complex branches of science... is that of confusing one’s machinery of analysis with one’s object of analysis. One version of this is pandemic in linguistic theory today: almost all theorists take morphophonemes to be things in a language rather than merely part of our equipment for the analysis and description of the language. (Hockett 1967:221)
The empirical scope of realizational models also reflects the descriptive strengths and weaknesses of traditional grammars. Rules that spell out antecedently- specified features are well equipped to exploit the closed, relatively uniform feature space of an inflectional system. Within this closed space, the features of paradigm cells can be defined independently of the forms that realize them. From the class of an open-class item, it is usually possible to determine the features that are distinctive for that class and predict the number of cells in the paradigm of the item. Apart from irregular items, paradigms are broadly comparable in size and structure within a word or inflection class.
However the fact that realizational models are so finely tuned to the structure of inflectional systems creates difficulties in extending them to other types of patterns. In particular, realizational models are less applicable to the variable structure exhibited by ‘families’ of derivational forms. Processes that relate items with variable word class, valence or other intrinsic properties usually do not define a finite set of forms within a uniform feature space. From just the word or inflection class of an item, one cannot in general predict the number and type of derivational formations in which it occurs. Given a list of derivational processes active in a language, it is of course possible to assign a uniform family ofpotential’ forms to all of the members of a word class. Yet the uniformity achieved is deceptive, because it collapses a critical distinction between those forms that are established in a language andthose that aremerelypossibleinprinciple. Thepoint maybeclearer in connection with compounds. Of the many possible noun compounds in a language such as English, only a comparatively small number are established, and a speaker cannot predict the established compounds containing an item from the item itself.
Even in cases where a derivational item is predictable from an item of a different word class, it is far from obvious what useful contribution is made by expressing the correspondence as the spell-out of antecedently specified features. Consider just the simple patterns exhibited by deverbal nouns in English. Nearly all verbs that allow an agentive interpretation have corresponding agentive nominals consisting of the verb stem and an invariant -er ending. Many verbs also have counterpart agent nominals, marked by the productive ending -ing and/or by a variety of other endings (Marchand i960). In the case of agentive nominals, the meaning expressed is essentially relational: a nominal of the form ‘Xer is interpreted as ‘one who Xs’. Yet what ‘features’ in the bundle associated with an agentive nominal are ‘spelled out’ as -er? In the case of event nominals, the choice of ending is neither uniform, nor determined by otherwise motivated morphological classes. Instead, the choice is a lexical property of individual items, as argued at length in debates about transformational and lexicalist analyses of derived nominals (Chomsky 1970).
To accommodate derivational patterns, realizational accounts such as AMM and PFM incorporate a class of derivational ‘word formation rules’. As specified in the following passage, these rules, like the lexical rules of an IP model, map one feature- form pair onto a different feature-form pair:
Our view is that of Stump (1993Я, 1995, 2001): that a morphological expression is headed if and only if it arises through the application of a category-preserving rule of word formation. A category-preserving rule of derivation or compounding is one which allows one or more morphosyntactic properties of a base to persist as properties of its derivative. (Stewart and Stump 2007:407)
Whatever the merits of this treatment of derivational patterns, the use of entry- to-entry mappings implicitly acknowledges the limits of realizational strategies. As in the initial formulation in Matthews (1965), realization rules in models such as AMM and PFM operate within ‘the inflectional component of a WP grammar’. Other types of rules, with a greater or lesser family resemblance to realization rules, apply in other components. Hence, as Anderson (1992) emphasizes in the continuation of the quotation on p. 120 above, models like AMM are defined more by their opposition to procedures of morphemic analysis than by their consistent use of purely realizational strategies:
The principal opposition is to “Item and Arrangement” models; and since the dominant, classical picture of word structure based on the structuralist morphemes is firmly of this sort, it is this distinction that is most important to pursue. (Anderson 1992:72)
The opposition to the IA “picture of word structure” brings out another parallel between realizational approaches and classical WP descriptions. Strikingly, neither incorporates any model of word structure. It is a familiar criticism of the classical model that it developed no counterpart of morphemes—not even roots (Law 1998:112)—and that it failed to demarcate any unit between sounds/letters and words. Realizational models exhibit a remarkably similar perspective. As in the passage on p. 122 above, realization rules are “interpretive phonological rules” that “operate on the phonological matrix of the lexical entry, giving, finally, a phonetic matrix” (Chomsky 1965:172). The rule formats summarized in preceding sections map an input form onto an output form. The application of a rule is not conventionally interpreted as introducing ‘boundary’ elements or as marking morphotactic structure in any way. Instead, as in an IP model, morphological structure is represented in the ‘derivational history’ defined by the rules that apply to define a surface form. As a consequence, realizational models not only reject morphemic analysis but also dispense with morphs and with morphotactic structure in general.
-  Though see Spencer (2012) for a morph-based realizational perspective.