The ancient model and its adaptations
The most immediate reaction to Hockett’s mea culpa was the ‘defence’ of WP offered in Robins (1959). Prompted by the fact that “neither Hockett nor anyone else seems as yet to have taken up the suggestion” that WP should be given the same consideration as IA and IP (pp. ii6f.), Robins set out some of the empirical considerations that had led earlier proponents of the WP model to regard it as a general theoretical model, rather than as a practical framework for language description. Robins stresses particularly the distinctive role played by words and collections of words in a classical WP approach:
The main distinctive characteristics of a formalized WP model of grammatical description would then be the following: the word is taken as the basic unit of both syntax and morphology, and variable words are grouped into paradigms for the statement of their morphological forms and the listing of their various syntactic functions. (Robins 1959:127)
Significantly, when Robins turns to the advantages of WP analyses, he, like Hockett, defines the WP model largely in opposition to IA and IP models. Whereas classical models had no unit intervening between words and sounds (or letters), Robins (1959:127) acknowledges that “the morpheme must be recognized as the minimal element of grammatical structure”. He then clarifies that these ‘morphemes’ are essentially units of form (corresponding to what Hockett (1947) had termed ‘morphs’) and that meaning and grammatical functions are primarily associated with words in WP approaches:
Some entities that are clearly to be assigned morphemic status may be seen in several languages to bear conflicting and even contradictory grammatical functions when considered in isolation... On the other hand words anchored, as it were, in the paradigms of which they form a part usually bear a consistent, relatively simple and statable grammatical function. The word is a more stable and solid focus of grammatical relations than the component morpheme by itself. Put another way, grammatical statements are abstractions, but they are more profitably abstracted from words as wholes than from individual morphemes. (Robins 1959:128)
In the course of identifying “the characteristics of a formalized WP model of grammatical description”, Robins offers a modern reinterpretation of the classical model that introduces a number of substantive assumptions. By distinguishing the minimal ‘units of form’ in a morphological system from the minimal units that can be assigned a stable meaning or function, he arrives at a theoretical hybrid that combines what he sees as the descriptive strengths of WP and IA/IP models. The resulting hybrid grafts an IA view of morphotactics, which recognizes sub-word units, onto a WP view of morphosyntax/morphosemantics, (1965), Carstairs (1983), or the axiomatic approach to morphology in Integrational Linguistics (Lieb 1976,1980,1983,1992, 2005, 2013), represent independent developments of the classical model.
which associates grammatical properties and meanings with whole words. Other aspects of Robins’s reinterpretation are more subtle, though their effects on the development of the WP tradition are just as far reaching. By focussing on the role of words in a classical model, rather than on the analogical relations that bind words to paradigms, Robins reinforces the impression that the primary contrast between WP and IA/IP concerns units of analysis. The idea that models are distinguished by the units they recognize more than by the relations they establish between units had been fostered by Hockett’s terminology and would frame the terms of subsequent debates between advocates of ‘word-based’ and ‘morpheme-based’ models. By treating “ ‘derivation’ and ‘inflection’... as covering approximately the comparable situations” (p. 136), Robins further de-emphasizes the implicational properties that had placed paradigms at the centre of the classical WP model. What is mainly carried forward from the classical WP model in this formalization is the assumption that words are the primary locus of grammatical meaning.
The rehabilitation of the WP model continued in the work of Matthews (1965, 1972,1991), which also provided a bridge between theoretical linguistics and classical and philological traditions, in which the WP model had never entirely fallen out of favour. Although Robins (1959) had sketched out the architecture of a WP model and described a range of patterns that seemed particularly amenable to analysis in WP terms, he stopped short of formalizing these patterns explicitly. Matthews (1965) takes up this challenge by showing how the inflectional component of a grammar can be described in terms of sets of morphosyntactic properties and rules that ‘realize’ them:
On the positive side, the word-and-paradigm model appears to have some specific advantages. A prima facie case, in the field of inflectional and ‘derivational’ morphology, has been established by Robins (1959): the patterning of overt ‘morphemic segments’, within the word, may often be described in a way which is quite at variance with the patterning of the relevant ‘morphemes’. But it is, of course, no more than a prima facie case. The discussion can be carried no further until the word-and-paradigm approach has been characterized at least as clearly as current versions of morphemics.
The present paper is intended to supply a part of this formulation. It is restricted to inflectional problems alone: to be more precise, it deals with that subsection of the grammar (we will call it the inflectional component) which assigns a realization, or various alternative realizations, to each grammatical word. (Matthews 1965:142)
The model that Matthews proposes is strikingly simple in its basic conception. The paradigm of an item is represented by sets of properties (what in other traditions are termed ‘features’), each corresponding to a cell of the paradigm. The lexical entry of the item specifies a root or stem form on which the forms of the paradigm are based. The realization rules of a grammar ‘interpret’ properties by applying operations to a form. A set of such rules realize or ‘spell out’ the inflected surface form that is associated with a paradigm cell of an item by interpreting the properties of the cell and successively modifying the base form of the item.
This approach is elaborated in detail in Matthews (1972), where it is applied to the analysis of aspects of Latin conjugation.
To a large extent, the precise formalization developed in Matthews (1965,1972) served more as a ‘proof of concept’ than as an explicit model for subsequent analyses. Nevertheless, the basic conception of morphological analysis was both appealing and influential. Because property sets were specified independently of forms, these sets could be interpreted by property-preserving rules that defined surface forms. Correspondences between properties and forms could be stated where they obtained, but dissociations could also be accommodated. This conception of morphological analysis-as-interpretation is what unites the class of contemporary realizational models. In some ways the most direct descendant of initial realizational accounts is the model of Autonomous Morphology (Aronoff 1994), which expands the role of ‘stem indices’ and exploits the distinction between morphological rules and the operations they apply but does not add any new rule types. Network Morphology (Corbett and Fraser 1993; Brown and Hippisley
2012) also represents a conservative extension of a basic realizational model in which generalizations are expressed by means of inheritance hierarchies. The Extended WP model (Anderson 1982)—subsequently A-Morphous Morphology (Anderson 1992)—shows the greatest influence of the generative tradition, as it assembles the property bundles that define paradigm cells at the preterminal nodes of syntactic representations, and invokes a rule of lexical insertion to introduce lexical stems. The use of paradigm functions and other devices in Paradigm Function Morphology (Stump 2001) reduce the role of realization rules to the point that the model can perhaps only residually be characterized as realizational.
A larger group of approaches exploit the descriptive potential of dissociating ‘units of content’ and ‘units of form’, which Beard (1995) terms the Separation Hypothesis. These include ‘lexeme-based’ models (Zwicky 1985; Beard 1995), along with approaches that mediate realization via ‘stem spaces’ (Bonami and Boye 2006, 2007) or other types of articulated lexical structures (Sagot and Walther
2013) . A version of the Separation Hypothesis is also adopted by morphemic models such as Distributed Morphology, when they “endorse the separation of the terminal elements involved in the syntax from the phonological realization of these elements” (Halle and Marantz 1993:111).
Although initial formalizations of WP models principally explored realizational strategies, the exemplar-based perspective of classical WP models found a more direct resonance in the approach to derivational and syntactic constructions in Construction Grammar and Morphology (Booij 2010). Some of the leading ideas of classical WP models were also developed in different ways in other theoretical traditions. The notion of an inflectional paradigm is central to the Paradigm Economy Principle of Carstairs (1983) and to the descendant No Blur Principle of CarstairsMcCarthy (1994). The economy effects measured in these studies suggest the relevance of paradigmatic organization to affix-based models. Integrational models (Lieb 1976, 2013) likewise extend the classical notion of inflectional paradigm by organizing word forms into syntactic word paradigms, and sub-word units into morphological stem and affix paradigms. This extension has the effect of projecting the Separation Hypothesis onto parallel syntactic and morphological paradigm spaces, which are much like the linked form and content paradigms proposed in Stump (2006).
-  For the sake of clarity, the following descriptions adapt Matthews’ conventions. Grammaticalattributes such as case are designated as ‘features, specifications such as genitive as ‘values, and theterm ‘properties’ is reserved for feature-value pairs.