The inflectional component of a WP grammar
In the half century since the initial formalizations, a range of approaches have been developed, which, in different ways and to varying degrees, develop the insights of a classical WP model. The most familiar group of models consists of what are usually termed ‘realizational’ approaches. The realization rules that give these approaches their name are essentially interpretive counterparts of the processes of an ‘Item and Process’ model. As described in Chapter 2.3, an IP process P can be thought of applying to an ‘input’ pair (B, X), consisting of a property bundle B and a form X. The ‘output’ of the process is a pair (f (B), o(X)), in which f (B) is the result of applying the feature-changing function f to B and o(X) is the result of applying the operation o to X. In contrast, when a realizational rule of exponence R applies to (B,X), it preserves the properties in B, and just applies an operation to X. The difference between these rule types is illustrated by the contrast between the ‘feature-changing’ process in (6.1a) and the ‘interpretive’ exponence rule in (6.1b).
1 Liebs model again falls entirely outside the realizational tradition, as reflected in the use of non- realizational processes to cover both word formation and inflection in Lieb (2013), supplanting the ‘paradigm bases’ previously used to arrive at paradigms.
Word and Paradigm Morphology. First edition. James P. Blevins © James P. Blevins 2016. First published 2016 by Oxford University Press
(6.1) Processes vs exponence rules
a. P(fB,X» = f(B), o(X)}
b. R(fB,X» = o(X)
The structure of the rule in (6.1a) isolates the components that Anderson (1992) references in his description of‘A-Morphous Morphology’ (AMM):
Within the general typology of Hockett 1954, which continues to be referred to in the literature, this is an ‘Item and Process’ model of word structure, though, especially in its treatment of inflection, it could also be called a “Word and Paradigm” view (Anderson 1992:72)
The affinity with classical WP models reflects the fact that the property bundle B typically represents the features that define a paradigm cell. It is the use of the operation o(X) in (6.1b) that expresses an IP conception of word structure, in which form variation is described in terms of phonological operations.
Anderson’s description of AMM in the quotation above also summarizes the properties that distinguish realizational approaches as a class from the IA and IP models reviewed in Chapter 2. By retaining an operational treatment of word structure, realizational approaches preserve the advantages that IP models enjoy over IA accounts in the analysis of non-affixal patterns. Moreover, by embedding operations within interpretive rules that do not modify property bundles, realiza- tional models permit multiple rules to apply to the same bundle. Different rules may then ‘spell out’ overlapping features by distinct operations in the realization of a form. This flexibility allows realizational rules to describe the variable relations between ‘units of meaning’ and ‘units of form’ that challenge the underlying morphemic conception of an IP or IA model.
Yet as implictly acknowledged in Anderson’s description, the combination of an ‘IP model of word structure’ with a ‘WP view of inflection’ conveys advantages for the analysis of specific types of phenomena. From the inception of the realizational tradition, it has been understood that interpretive rules are particularly well adapted to the description of inflectional systems. The descriptive success of these rules derives largely from the fact that inflectional systems can be idealized in terms of an essentially closed and uniform feature space. This space can thus be modelled by sets of property bundles whose composition is determined by the distinctive features of a language. Applied to a set of independently-specified property bundles, interpretive rules can then define the feature combinations that have distinctive formal realization in a language.
The extension of an interpretive approach to derivational patterns is far from straightforward, given that the derivational family of an item is much more variable and much less predictable. Even within the domain of inflection, the advantages of a flexible spell-out relation depend on assumptions about the granularity and function of units of form and units of meaning. Realizational models relax a morphemic form-meaning correspondence to a weaker, many-many relation. But they retain the assumption that the global relation between systems of contrasts at the level of meaning and the level form are mediated by relations between individual meaning contrasts and form contrasts.
To clarify this point, consider the example of two expressions, x andy, that differ minimally in form and in meaning. The expression x is associated with a property Fx and form Фх that contrast with the property Fy and form Фу associated with y. A morphemic model will establish morphemes that link the associated units of form and meaning, Fx with Фх and Fy with Фу. A realizational model will likewise treat the units of form Фx and Фу as the ‘spell-outs’ of corresponding properties Fx and Fy. As noted in Section 6.3.2, the innocuous-seeming linkage defined by a realizational model has a range of implications, including characteristic patterns of ‘overgeneralization’. These effects do not arise in models, such as the learning-based discriminative models outlined in Chapter 8.3, which do not invariably reduce the relation between systems of contrasts at the level of meaning and the level of form to relations between individual contrasts at the level of meaning and form.
A wide range of strategies and devices have been proposed within realizational approaches to accommodate derivational formations, to ‘block’ overgeneralization, and to describe other classes of patterns that resist description in terms of interpretive rules. The fragmentation created by these often incompatible extensions is compounded by other factors, including the lack of a common rule format for exponence rules, chronic disagreement about the status of rules of ‘referral, etc. Variation in the degree of formalization introduces a further dimension of variation. The models outlined in Matthews (1991) and Aronoff (1994) are among the purest expositions of realization-based analyses, though the simplicity of these analyses is often achieved by leaving execution details open. At the other extreme, analyses formulated within models of Paradigm Function Morphology (PFM) of Stump (2001) are highly formalized. However, this degree of formalization often comes at the cost of obscuring the intuitive content of an analysis. Models of Network Morphology (NM; Corbett and Fraser 1993; Brown and Hippisley 2012), have likewise been specified in sufficient detail to permit computational implementation. Yet this formalization carries a commitment to an inheritance-based view of generalizations and a particular knowledge- representation system.
The fragmentation of realizational approaches has also inhibited the development of a secondary literature of the kind that facilitated collaboration within the Post-Bloomfieldian tradition. In part, this fragmentation is due to the fact that realizational approaches—unlike morphemic models—were not consciously developed by a single cohesive community. To the extent that the various real- izational approaches are defined less by shared assumptions than by a shared morphemic adversary, it may even be misleading to treat them as instantiations of a common framework in the sense of Chapter 1.3.2.
These factors present challenges for a summary of the realizational tradition, and it is unlikely that any summary will satisfy proponents of individual approaches. The limited goal of the present chapter is just to provide an overview of the real- izational tradition that highlights the principal benefits of basing morphological analyses on realization relations. This objective is in many ways orthogonal to the aims of the individual approaches, each of which explores a different idiosyncratic path through the space of extensions to a basic realizational conception. To avoid potential misunderstandings and allay concerns about misrepresentation, it is perhaps worth emphasizing that the chapter will, insofar as possible, attempt to isolate the core intuitions underlying realizational approaches and focus primarily on maximally general notions of ‘exponence rules’, ‘referral rules’, ‘rule ordering, ‘rule blocks’, etc. Section 6.4 rounds out this overview with a concise summary of distinctive aspects of some currently influential realizational models. Otherwise, formalism-specific variants are discussed only where this serves some larger expositional goal.
-  As were the analyses in the initial monograph-length study of Matthews (Г972).